blob: 7e84baccf0ad6c90f7a3f62ec084007003319b21 [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;
Mike Snitzer787a996c2013-12-06 16:21:43 -0500147 bool error_if_no_space:1;
Joe Thornber67e2e2b2012-03-28 18:41:29 +0100148};
149
Joe Thornbere49e5822012-07-27 15:08:16 +0100150struct thin_c;
151typedef void (*process_bio_fn)(struct thin_c *tc, struct bio *bio);
152typedef void (*process_mapping_fn)(struct dm_thin_new_mapping *m);
153
Joe Thornber991d9fa2011-10-31 20:21:18 +0000154struct pool {
155 struct list_head list;
156 struct dm_target *ti; /* Only set if a pool target is bound */
157
158 struct mapped_device *pool_md;
159 struct block_device *md_dev;
160 struct dm_pool_metadata *pmd;
161
Joe Thornber991d9fa2011-10-31 20:21:18 +0000162 dm_block_t low_water_blocks;
Mike Snitzer55f2b8b2012-07-27 15:08:02 +0100163 uint32_t sectors_per_block;
Mikulas Patockaf9a8e0c2012-07-27 15:08:03 +0100164 int sectors_per_block_shift;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000165
Joe Thornber67e2e2b2012-03-28 18:41:29 +0100166 struct pool_features pf;
Joe Thornber88a66212013-12-04 20:16:12 -0500167 bool low_water_triggered:1; /* A dm event has been sent */
Joe Thornber991d9fa2011-10-31 20:21:18 +0000168
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);
Mike Snitzer399cadd2013-12-05 16:03:33 -0500201static void out_of_data_space(struct pool *pool);
Joe Thornberb5330652013-12-04 19:51:33 -0500202static void metadata_operation_failed(struct pool *pool, const char *op, int r);
Joe Thornbere49e5822012-07-27 15:08:16 +0100203
Joe Thornber991d9fa2011-10-31 20:21:18 +0000204/*
205 * Target context for a pool.
206 */
207struct pool_c {
208 struct dm_target *ti;
209 struct pool *pool;
210 struct dm_dev *data_dev;
211 struct dm_dev *metadata_dev;
212 struct dm_target_callbacks callbacks;
213
214 dm_block_t low_water_blocks;
Mike Snitzer0424caa2012-09-26 23:45:47 +0100215 struct pool_features requested_pf; /* Features requested during table load */
216 struct pool_features adjusted_pf; /* Features used after adjusting for constituent devices */
Joe Thornber991d9fa2011-10-31 20:21:18 +0000217};
218
219/*
220 * Target context for a thin.
221 */
222struct thin_c {
223 struct dm_dev *pool_dev;
Joe Thornber2dd9c252012-03-28 18:41:28 +0100224 struct dm_dev *origin_dev;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000225 dm_thin_id dev_id;
226
227 struct pool *pool;
228 struct dm_thin_device *td;
229};
230
231/*----------------------------------------------------------------*/
232
Joe Thornber025b9682013-03-01 22:45:50 +0000233/*
234 * wake_worker() is used when new work is queued and when pool_resume is
235 * ready to continue deferred IO processing.
236 */
237static void wake_worker(struct pool *pool)
238{
239 queue_work(pool->wq, &pool->worker);
240}
241
242/*----------------------------------------------------------------*/
243
Joe Thornber6beca5e2013-03-01 22:45:50 +0000244static int bio_detain(struct pool *pool, struct dm_cell_key *key, struct bio *bio,
245 struct dm_bio_prison_cell **cell_result)
246{
247 int r;
248 struct dm_bio_prison_cell *cell_prealloc;
249
250 /*
251 * Allocate a cell from the prison's mempool.
252 * This might block but it can't fail.
253 */
254 cell_prealloc = dm_bio_prison_alloc_cell(pool->prison, GFP_NOIO);
255
256 r = dm_bio_detain(pool->prison, key, bio, cell_prealloc, cell_result);
257 if (r)
258 /*
259 * We reused an old cell; we can get rid of
260 * the new one.
261 */
262 dm_bio_prison_free_cell(pool->prison, cell_prealloc);
263
264 return r;
265}
266
267static void cell_release(struct pool *pool,
268 struct dm_bio_prison_cell *cell,
269 struct bio_list *bios)
270{
271 dm_cell_release(pool->prison, cell, bios);
272 dm_bio_prison_free_cell(pool->prison, cell);
273}
274
275static void cell_release_no_holder(struct pool *pool,
276 struct dm_bio_prison_cell *cell,
277 struct bio_list *bios)
278{
279 dm_cell_release_no_holder(pool->prison, cell, bios);
280 dm_bio_prison_free_cell(pool->prison, cell);
281}
282
Joe Thornber025b9682013-03-01 22:45:50 +0000283static void cell_defer_no_holder_no_free(struct thin_c *tc,
284 struct dm_bio_prison_cell *cell)
285{
286 struct pool *pool = tc->pool;
287 unsigned long flags;
288
289 spin_lock_irqsave(&pool->lock, flags);
290 dm_cell_release_no_holder(pool->prison, cell, &pool->deferred_bios);
291 spin_unlock_irqrestore(&pool->lock, flags);
292
293 wake_worker(pool);
294}
295
Joe Thornber6beca5e2013-03-01 22:45:50 +0000296static void cell_error(struct pool *pool,
297 struct dm_bio_prison_cell *cell)
298{
299 dm_cell_error(pool->prison, cell);
300 dm_bio_prison_free_cell(pool->prison, cell);
301}
302
303/*----------------------------------------------------------------*/
304
Joe Thornber991d9fa2011-10-31 20:21:18 +0000305/*
306 * A global list of pools that uses a struct mapped_device as a key.
307 */
308static struct dm_thin_pool_table {
309 struct mutex mutex;
310 struct list_head pools;
311} dm_thin_pool_table;
312
313static void pool_table_init(void)
314{
315 mutex_init(&dm_thin_pool_table.mutex);
316 INIT_LIST_HEAD(&dm_thin_pool_table.pools);
317}
318
319static void __pool_table_insert(struct pool *pool)
320{
321 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
322 list_add(&pool->list, &dm_thin_pool_table.pools);
323}
324
325static void __pool_table_remove(struct pool *pool)
326{
327 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
328 list_del(&pool->list);
329}
330
331static struct pool *__pool_table_lookup(struct mapped_device *md)
332{
333 struct pool *pool = NULL, *tmp;
334
335 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
336
337 list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) {
338 if (tmp->pool_md == md) {
339 pool = tmp;
340 break;
341 }
342 }
343
344 return pool;
345}
346
347static struct pool *__pool_table_lookup_metadata_dev(struct block_device *md_dev)
348{
349 struct pool *pool = NULL, *tmp;
350
351 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
352
353 list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) {
354 if (tmp->md_dev == md_dev) {
355 pool = tmp;
356 break;
357 }
358 }
359
360 return pool;
361}
362
363/*----------------------------------------------------------------*/
364
Mike Snitzera24c2562012-06-03 00:30:00 +0100365struct dm_thin_endio_hook {
Joe Thornbereb2aa482012-03-28 18:41:28 +0100366 struct thin_c *tc;
Mike Snitzer44feb382012-10-12 21:02:10 +0100367 struct dm_deferred_entry *shared_read_entry;
368 struct dm_deferred_entry *all_io_entry;
Mike Snitzera24c2562012-06-03 00:30:00 +0100369 struct dm_thin_new_mapping *overwrite_mapping;
Joe Thornbereb2aa482012-03-28 18:41:28 +0100370};
371
Joe Thornber991d9fa2011-10-31 20:21:18 +0000372static void __requeue_bio_list(struct thin_c *tc, struct bio_list *master)
373{
374 struct bio *bio;
375 struct bio_list bios;
376
377 bio_list_init(&bios);
378 bio_list_merge(&bios, master);
379 bio_list_init(master);
380
381 while ((bio = bio_list_pop(&bios))) {
Mikulas Patocka59c3d2c2012-12-21 20:23:40 +0000382 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
Mike Snitzera24c2562012-06-03 00:30:00 +0100383
Joe Thornbereb2aa482012-03-28 18:41:28 +0100384 if (h->tc == tc)
Joe Thornber991d9fa2011-10-31 20:21:18 +0000385 bio_endio(bio, DM_ENDIO_REQUEUE);
386 else
387 bio_list_add(master, bio);
388 }
389}
390
391static void requeue_io(struct thin_c *tc)
392{
393 struct pool *pool = tc->pool;
394 unsigned long flags;
395
396 spin_lock_irqsave(&pool->lock, flags);
397 __requeue_bio_list(tc, &pool->deferred_bios);
398 __requeue_bio_list(tc, &pool->retry_on_resume_list);
399 spin_unlock_irqrestore(&pool->lock, flags);
400}
401
402/*
403 * This section of code contains the logic for processing a thin device's IO.
404 * Much of the code depends on pool object resources (lists, workqueues, etc)
405 * but most is exclusively called from the thin target rather than the thin-pool
406 * target.
407 */
408
Mike Snitzer58f77a22013-03-01 22:45:45 +0000409static bool block_size_is_power_of_two(struct pool *pool)
410{
411 return pool->sectors_per_block_shift >= 0;
412}
413
Joe Thornber991d9fa2011-10-31 20:21:18 +0000414static dm_block_t get_bio_block(struct thin_c *tc, struct bio *bio)
415{
Mike Snitzer58f77a22013-03-01 22:45:45 +0000416 struct pool *pool = tc->pool;
Kent Overstreet4f024f32013-10-11 15:44:27 -0700417 sector_t block_nr = bio->bi_iter.bi_sector;
Mike Snitzer55f2b8b2012-07-27 15:08:02 +0100418
Mike Snitzer58f77a22013-03-01 22:45:45 +0000419 if (block_size_is_power_of_two(pool))
420 block_nr >>= pool->sectors_per_block_shift;
Mikulas Patockaf9a8e0c2012-07-27 15:08:03 +0100421 else
Mike Snitzer58f77a22013-03-01 22:45:45 +0000422 (void) sector_div(block_nr, pool->sectors_per_block);
Mike Snitzer55f2b8b2012-07-27 15:08:02 +0100423
424 return block_nr;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000425}
426
427static void remap(struct thin_c *tc, struct bio *bio, dm_block_t block)
428{
429 struct pool *pool = tc->pool;
Kent Overstreet4f024f32013-10-11 15:44:27 -0700430 sector_t bi_sector = bio->bi_iter.bi_sector;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000431
432 bio->bi_bdev = tc->pool_dev->bdev;
Mike Snitzer58f77a22013-03-01 22:45:45 +0000433 if (block_size_is_power_of_two(pool))
Kent Overstreet4f024f32013-10-11 15:44:27 -0700434 bio->bi_iter.bi_sector =
435 (block << pool->sectors_per_block_shift) |
436 (bi_sector & (pool->sectors_per_block - 1));
Mike Snitzer58f77a22013-03-01 22:45:45 +0000437 else
Kent Overstreet4f024f32013-10-11 15:44:27 -0700438 bio->bi_iter.bi_sector = (block * pool->sectors_per_block) +
Mike Snitzer58f77a22013-03-01 22:45:45 +0000439 sector_div(bi_sector, pool->sectors_per_block);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000440}
441
Joe Thornber2dd9c252012-03-28 18:41:28 +0100442static void remap_to_origin(struct thin_c *tc, struct bio *bio)
443{
444 bio->bi_bdev = tc->origin_dev->bdev;
445}
446
Joe Thornber4afdd682012-07-27 15:08:14 +0100447static int bio_triggers_commit(struct thin_c *tc, struct bio *bio)
448{
449 return (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) &&
450 dm_thin_changed_this_transaction(tc->td);
451}
452
Joe Thornbere8088072012-12-21 20:23:31 +0000453static void inc_all_io_entry(struct pool *pool, struct bio *bio)
454{
455 struct dm_thin_endio_hook *h;
456
457 if (bio->bi_rw & REQ_DISCARD)
458 return;
459
Mikulas Patocka59c3d2c2012-12-21 20:23:40 +0000460 h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
Joe Thornbere8088072012-12-21 20:23:31 +0000461 h->all_io_entry = dm_deferred_entry_inc(pool->all_io_ds);
462}
463
Joe Thornber2dd9c252012-03-28 18:41:28 +0100464static void issue(struct thin_c *tc, struct bio *bio)
Joe Thornber991d9fa2011-10-31 20:21:18 +0000465{
466 struct pool *pool = tc->pool;
467 unsigned long flags;
468
Joe Thornbere49e5822012-07-27 15:08:16 +0100469 if (!bio_triggers_commit(tc, bio)) {
Joe Thornber991d9fa2011-10-31 20:21:18 +0000470 generic_make_request(bio);
Joe Thornbere49e5822012-07-27 15:08:16 +0100471 return;
472 }
473
474 /*
475 * Complete bio with an error if earlier I/O caused changes to
476 * the metadata that can't be committed e.g, due to I/O errors
477 * on the metadata device.
478 */
479 if (dm_thin_aborted_changes(tc->td)) {
480 bio_io_error(bio);
481 return;
482 }
483
484 /*
485 * Batch together any bios that trigger commits and then issue a
486 * single commit for them in process_deferred_bios().
487 */
488 spin_lock_irqsave(&pool->lock, flags);
489 bio_list_add(&pool->deferred_flush_bios, bio);
490 spin_unlock_irqrestore(&pool->lock, flags);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000491}
492
Joe Thornber2dd9c252012-03-28 18:41:28 +0100493static void remap_to_origin_and_issue(struct thin_c *tc, struct bio *bio)
494{
495 remap_to_origin(tc, bio);
496 issue(tc, bio);
497}
498
499static void remap_and_issue(struct thin_c *tc, struct bio *bio,
500 dm_block_t block)
501{
502 remap(tc, bio, block);
503 issue(tc, bio);
504}
505
Joe Thornber991d9fa2011-10-31 20:21:18 +0000506/*----------------------------------------------------------------*/
507
508/*
509 * Bio endio functions.
510 */
Mike Snitzera24c2562012-06-03 00:30:00 +0100511struct dm_thin_new_mapping {
Joe Thornber991d9fa2011-10-31 20:21:18 +0000512 struct list_head list;
513
Mike Snitzer7f214662013-12-17 13:43:31 -0500514 bool quiesced:1;
515 bool prepared:1;
516 bool pass_discard:1;
517 bool definitely_not_shared:1;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000518
Mike Snitzer7f214662013-12-17 13:43:31 -0500519 int err;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000520 struct thin_c *tc;
521 dm_block_t virt_block;
522 dm_block_t data_block;
Mike Snitzera24c2562012-06-03 00:30:00 +0100523 struct dm_bio_prison_cell *cell, *cell2;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000524
525 /*
526 * If the bio covers the whole area of a block then we can avoid
527 * zeroing or copying. Instead this bio is hooked. The bio will
528 * still be in the cell, so care has to be taken to avoid issuing
529 * the bio twice.
530 */
531 struct bio *bio;
532 bio_end_io_t *saved_bi_end_io;
533};
534
Mike Snitzera24c2562012-06-03 00:30:00 +0100535static void __maybe_add_mapping(struct dm_thin_new_mapping *m)
Joe Thornber991d9fa2011-10-31 20:21:18 +0000536{
537 struct pool *pool = m->tc->pool;
538
Joe Thornbereb2aa482012-03-28 18:41:28 +0100539 if (m->quiesced && m->prepared) {
Mike Snitzerdaec3382013-12-11 14:01:20 -0500540 list_add_tail(&m->list, &pool->prepared_mappings);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000541 wake_worker(pool);
542 }
543}
544
545static void copy_complete(int read_err, unsigned long write_err, void *context)
546{
547 unsigned long flags;
Mike Snitzera24c2562012-06-03 00:30:00 +0100548 struct dm_thin_new_mapping *m = context;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000549 struct pool *pool = m->tc->pool;
550
551 m->err = read_err || write_err ? -EIO : 0;
552
553 spin_lock_irqsave(&pool->lock, flags);
Mike Snitzer7f214662013-12-17 13:43:31 -0500554 m->prepared = true;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000555 __maybe_add_mapping(m);
556 spin_unlock_irqrestore(&pool->lock, flags);
557}
558
559static void overwrite_endio(struct bio *bio, int err)
560{
561 unsigned long flags;
Mikulas Patocka59c3d2c2012-12-21 20:23:40 +0000562 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
Mike Snitzera24c2562012-06-03 00:30:00 +0100563 struct dm_thin_new_mapping *m = h->overwrite_mapping;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000564 struct pool *pool = m->tc->pool;
565
566 m->err = err;
567
568 spin_lock_irqsave(&pool->lock, flags);
Mike Snitzer7f214662013-12-17 13:43:31 -0500569 m->prepared = true;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000570 __maybe_add_mapping(m);
571 spin_unlock_irqrestore(&pool->lock, flags);
572}
573
Joe Thornber991d9fa2011-10-31 20:21:18 +0000574/*----------------------------------------------------------------*/
575
576/*
577 * Workqueue.
578 */
579
580/*
581 * Prepared mapping jobs.
582 */
583
584/*
585 * This sends the bios in the cell back to the deferred_bios list.
586 */
Joe Thornber2aab3852012-12-21 20:23:33 +0000587static void cell_defer(struct thin_c *tc, struct dm_bio_prison_cell *cell)
Joe Thornber991d9fa2011-10-31 20:21:18 +0000588{
589 struct pool *pool = tc->pool;
590 unsigned long flags;
591
592 spin_lock_irqsave(&pool->lock, flags);
Joe Thornber6beca5e2013-03-01 22:45:50 +0000593 cell_release(pool, cell, &pool->deferred_bios);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000594 spin_unlock_irqrestore(&tc->pool->lock, flags);
595
596 wake_worker(pool);
597}
598
599/*
Joe Thornber6beca5e2013-03-01 22:45:50 +0000600 * Same as cell_defer above, except it omits the original holder of the cell.
Joe Thornber991d9fa2011-10-31 20:21:18 +0000601 */
Joe Thornberf286ba02012-12-21 20:23:33 +0000602static void cell_defer_no_holder(struct thin_c *tc, struct dm_bio_prison_cell *cell)
Joe Thornber991d9fa2011-10-31 20:21:18 +0000603{
Joe Thornber991d9fa2011-10-31 20:21:18 +0000604 struct pool *pool = tc->pool;
605 unsigned long flags;
606
Joe Thornber991d9fa2011-10-31 20:21:18 +0000607 spin_lock_irqsave(&pool->lock, flags);
Joe Thornber6beca5e2013-03-01 22:45:50 +0000608 cell_release_no_holder(pool, cell, &pool->deferred_bios);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000609 spin_unlock_irqrestore(&pool->lock, flags);
610
611 wake_worker(pool);
612}
613
Joe Thornbere49e5822012-07-27 15:08:16 +0100614static void process_prepared_mapping_fail(struct dm_thin_new_mapping *m)
615{
Kent Overstreet196d38b2013-11-23 18:34:15 -0800616 if (m->bio) {
Joe Thornbere49e5822012-07-27 15:08:16 +0100617 m->bio->bi_end_io = m->saved_bi_end_io;
Kent Overstreet196d38b2013-11-23 18:34:15 -0800618 atomic_inc(&m->bio->bi_remaining);
619 }
Joe Thornber6beca5e2013-03-01 22:45:50 +0000620 cell_error(m->tc->pool, m->cell);
Joe Thornbere49e5822012-07-27 15:08:16 +0100621 list_del(&m->list);
622 mempool_free(m, m->tc->pool->mapping_pool);
623}
Joe Thornber025b9682013-03-01 22:45:50 +0000624
Mike Snitzera24c2562012-06-03 00:30:00 +0100625static void process_prepared_mapping(struct dm_thin_new_mapping *m)
Joe Thornber991d9fa2011-10-31 20:21:18 +0000626{
627 struct thin_c *tc = m->tc;
Joe Thornber6beca5e2013-03-01 22:45:50 +0000628 struct pool *pool = tc->pool;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000629 struct bio *bio;
630 int r;
631
632 bio = m->bio;
Kent Overstreet196d38b2013-11-23 18:34:15 -0800633 if (bio) {
Joe Thornber991d9fa2011-10-31 20:21:18 +0000634 bio->bi_end_io = m->saved_bi_end_io;
Kent Overstreet196d38b2013-11-23 18:34:15 -0800635 atomic_inc(&bio->bi_remaining);
636 }
Joe Thornber991d9fa2011-10-31 20:21:18 +0000637
638 if (m->err) {
Joe Thornber6beca5e2013-03-01 22:45:50 +0000639 cell_error(pool, m->cell);
Joe Thornber905386f2012-07-27 15:08:05 +0100640 goto out;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000641 }
642
643 /*
644 * Commit the prepared block into the mapping btree.
645 * Any I/O for this block arriving after this point will get
646 * remapped to it directly.
647 */
648 r = dm_thin_insert_block(tc->td, m->virt_block, m->data_block);
649 if (r) {
Joe Thornberb5330652013-12-04 19:51:33 -0500650 metadata_operation_failed(pool, "dm_thin_insert_block", r);
Joe Thornber6beca5e2013-03-01 22:45:50 +0000651 cell_error(pool, m->cell);
Joe Thornber905386f2012-07-27 15:08:05 +0100652 goto out;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000653 }
654
655 /*
656 * Release any bios held while the block was being provisioned.
657 * If we are processing a write bio that completely covers the block,
658 * we already processed it so can ignore it now when processing
659 * the bios in the cell.
660 */
661 if (bio) {
Joe Thornberf286ba02012-12-21 20:23:33 +0000662 cell_defer_no_holder(tc, m->cell);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000663 bio_endio(bio, 0);
664 } else
Joe Thornber2aab3852012-12-21 20:23:33 +0000665 cell_defer(tc, m->cell);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000666
Joe Thornber905386f2012-07-27 15:08:05 +0100667out:
Joe Thornber991d9fa2011-10-31 20:21:18 +0000668 list_del(&m->list);
Joe Thornber6beca5e2013-03-01 22:45:50 +0000669 mempool_free(m, pool->mapping_pool);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000670}
671
Joe Thornbere49e5822012-07-27 15:08:16 +0100672static void process_prepared_discard_fail(struct dm_thin_new_mapping *m)
Joe Thornber104655f2012-03-28 18:41:28 +0100673{
Joe Thornber104655f2012-03-28 18:41:28 +0100674 struct thin_c *tc = m->tc;
675
Joe Thornbere49e5822012-07-27 15:08:16 +0100676 bio_io_error(m->bio);
Joe Thornberf286ba02012-12-21 20:23:33 +0000677 cell_defer_no_holder(tc, m->cell);
678 cell_defer_no_holder(tc, m->cell2);
Joe Thornbere49e5822012-07-27 15:08:16 +0100679 mempool_free(m, tc->pool->mapping_pool);
680}
Joe Thornber104655f2012-03-28 18:41:28 +0100681
Joe Thornbere49e5822012-07-27 15:08:16 +0100682static void process_prepared_discard_passdown(struct dm_thin_new_mapping *m)
683{
684 struct thin_c *tc = m->tc;
685
Joe Thornbere8088072012-12-21 20:23:31 +0000686 inc_all_io_entry(tc->pool, m->bio);
Joe Thornberf286ba02012-12-21 20:23:33 +0000687 cell_defer_no_holder(tc, m->cell);
688 cell_defer_no_holder(tc, m->cell2);
Joe Thornbere8088072012-12-21 20:23:31 +0000689
Joe Thornber104655f2012-03-28 18:41:28 +0100690 if (m->pass_discard)
Joe Thornber19fa1a62013-12-17 12:09:40 -0500691 if (m->definitely_not_shared)
692 remap_and_issue(tc, m->bio, m->data_block);
693 else {
694 bool used = false;
695 if (dm_pool_block_is_used(tc->pool->pmd, m->data_block, &used) || used)
696 bio_endio(m->bio, 0);
697 else
698 remap_and_issue(tc, m->bio, m->data_block);
699 }
Joe Thornber104655f2012-03-28 18:41:28 +0100700 else
701 bio_endio(m->bio, 0);
702
Joe Thornber104655f2012-03-28 18:41:28 +0100703 mempool_free(m, tc->pool->mapping_pool);
704}
705
Joe Thornbere49e5822012-07-27 15:08:16 +0100706static void process_prepared_discard(struct dm_thin_new_mapping *m)
707{
708 int r;
709 struct thin_c *tc = m->tc;
710
711 r = dm_thin_remove_block(tc->td, m->virt_block);
712 if (r)
Mike Snitzerc3977412012-12-21 20:23:34 +0000713 DMERR_LIMIT("dm_thin_remove_block() failed");
Joe Thornbere49e5822012-07-27 15:08:16 +0100714
715 process_prepared_discard_passdown(m);
716}
717
Joe Thornber104655f2012-03-28 18:41:28 +0100718static void process_prepared(struct pool *pool, struct list_head *head,
Joe Thornbere49e5822012-07-27 15:08:16 +0100719 process_mapping_fn *fn)
Joe Thornber991d9fa2011-10-31 20:21:18 +0000720{
721 unsigned long flags;
722 struct list_head maps;
Mike Snitzera24c2562012-06-03 00:30:00 +0100723 struct dm_thin_new_mapping *m, *tmp;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000724
725 INIT_LIST_HEAD(&maps);
726 spin_lock_irqsave(&pool->lock, flags);
Joe Thornber104655f2012-03-28 18:41:28 +0100727 list_splice_init(head, &maps);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000728 spin_unlock_irqrestore(&pool->lock, flags);
729
730 list_for_each_entry_safe(m, tmp, &maps, list)
Joe Thornbere49e5822012-07-27 15:08:16 +0100731 (*fn)(m);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000732}
733
734/*
735 * Deferred bio jobs.
736 */
Joe Thornber104655f2012-03-28 18:41:28 +0100737static int io_overlaps_block(struct pool *pool, struct bio *bio)
738{
Kent Overstreet4f024f32013-10-11 15:44:27 -0700739 return bio->bi_iter.bi_size ==
740 (pool->sectors_per_block << SECTOR_SHIFT);
Joe Thornber104655f2012-03-28 18:41:28 +0100741}
742
Joe Thornber991d9fa2011-10-31 20:21:18 +0000743static int io_overwrites_block(struct pool *pool, struct bio *bio)
744{
Joe Thornber104655f2012-03-28 18:41:28 +0100745 return (bio_data_dir(bio) == WRITE) &&
746 io_overlaps_block(pool, bio);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000747}
748
749static void save_and_set_endio(struct bio *bio, bio_end_io_t **save,
750 bio_end_io_t *fn)
751{
752 *save = bio->bi_end_io;
753 bio->bi_end_io = fn;
754}
755
756static int ensure_next_mapping(struct pool *pool)
757{
758 if (pool->next_mapping)
759 return 0;
760
761 pool->next_mapping = mempool_alloc(pool->mapping_pool, GFP_ATOMIC);
762
763 return pool->next_mapping ? 0 : -ENOMEM;
764}
765
Mike Snitzera24c2562012-06-03 00:30:00 +0100766static struct dm_thin_new_mapping *get_next_mapping(struct pool *pool)
Joe Thornber991d9fa2011-10-31 20:21:18 +0000767{
Mike Snitzer16961b02013-12-17 13:19:11 -0500768 struct dm_thin_new_mapping *m = pool->next_mapping;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000769
770 BUG_ON(!pool->next_mapping);
771
Mike Snitzer16961b02013-12-17 13:19:11 -0500772 memset(m, 0, sizeof(struct dm_thin_new_mapping));
773 INIT_LIST_HEAD(&m->list);
774 m->bio = NULL;
775
Joe Thornber991d9fa2011-10-31 20:21:18 +0000776 pool->next_mapping = NULL;
777
Mike Snitzer16961b02013-12-17 13:19:11 -0500778 return m;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000779}
780
781static void schedule_copy(struct thin_c *tc, dm_block_t virt_block,
Joe Thornber2dd9c252012-03-28 18:41:28 +0100782 struct dm_dev *origin, dm_block_t data_origin,
783 dm_block_t data_dest,
Mike Snitzera24c2562012-06-03 00:30:00 +0100784 struct dm_bio_prison_cell *cell, struct bio *bio)
Joe Thornber991d9fa2011-10-31 20:21:18 +0000785{
786 int r;
787 struct pool *pool = tc->pool;
Mike Snitzera24c2562012-06-03 00:30:00 +0100788 struct dm_thin_new_mapping *m = get_next_mapping(pool);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000789
Joe Thornber991d9fa2011-10-31 20:21:18 +0000790 m->tc = tc;
791 m->virt_block = virt_block;
792 m->data_block = data_dest;
793 m->cell = cell;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000794
Mike Snitzer44feb382012-10-12 21:02:10 +0100795 if (!dm_deferred_set_add_work(pool->shared_read_ds, &m->list))
Mike Snitzer7f214662013-12-17 13:43:31 -0500796 m->quiesced = true;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000797
798 /*
799 * IO to pool_dev remaps to the pool target's data_dev.
800 *
801 * If the whole block of data is being overwritten, we can issue the
802 * bio immediately. Otherwise we use kcopyd to clone the data first.
803 */
804 if (io_overwrites_block(pool, bio)) {
Mikulas Patocka59c3d2c2012-12-21 20:23:40 +0000805 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
Mike Snitzera24c2562012-06-03 00:30:00 +0100806
Joe Thornbereb2aa482012-03-28 18:41:28 +0100807 h->overwrite_mapping = m;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000808 m->bio = bio;
809 save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio);
Joe Thornbere8088072012-12-21 20:23:31 +0000810 inc_all_io_entry(pool, bio);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000811 remap_and_issue(tc, bio, data_dest);
812 } else {
813 struct dm_io_region from, to;
814
Joe Thornber2dd9c252012-03-28 18:41:28 +0100815 from.bdev = origin->bdev;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000816 from.sector = data_origin * pool->sectors_per_block;
817 from.count = pool->sectors_per_block;
818
819 to.bdev = tc->pool_dev->bdev;
820 to.sector = data_dest * pool->sectors_per_block;
821 to.count = pool->sectors_per_block;
822
823 r = dm_kcopyd_copy(pool->copier, &from, 1, &to,
824 0, copy_complete, m);
825 if (r < 0) {
826 mempool_free(m, pool->mapping_pool);
Mike Snitzerc3977412012-12-21 20:23:34 +0000827 DMERR_LIMIT("dm_kcopyd_copy() failed");
Joe Thornber6beca5e2013-03-01 22:45:50 +0000828 cell_error(pool, cell);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000829 }
830 }
831}
832
Joe Thornber2dd9c252012-03-28 18:41:28 +0100833static void schedule_internal_copy(struct thin_c *tc, dm_block_t virt_block,
834 dm_block_t data_origin, dm_block_t data_dest,
Mike Snitzera24c2562012-06-03 00:30:00 +0100835 struct dm_bio_prison_cell *cell, struct bio *bio)
Joe Thornber2dd9c252012-03-28 18:41:28 +0100836{
837 schedule_copy(tc, virt_block, tc->pool_dev,
838 data_origin, data_dest, cell, bio);
839}
840
841static void schedule_external_copy(struct thin_c *tc, dm_block_t virt_block,
842 dm_block_t data_dest,
Mike Snitzera24c2562012-06-03 00:30:00 +0100843 struct dm_bio_prison_cell *cell, struct bio *bio)
Joe Thornber2dd9c252012-03-28 18:41:28 +0100844{
845 schedule_copy(tc, virt_block, tc->origin_dev,
846 virt_block, data_dest, cell, bio);
847}
848
Joe Thornber991d9fa2011-10-31 20:21:18 +0000849static void schedule_zero(struct thin_c *tc, dm_block_t virt_block,
Mike Snitzera24c2562012-06-03 00:30:00 +0100850 dm_block_t data_block, struct dm_bio_prison_cell *cell,
Joe Thornber991d9fa2011-10-31 20:21:18 +0000851 struct bio *bio)
852{
853 struct pool *pool = tc->pool;
Mike Snitzera24c2562012-06-03 00:30:00 +0100854 struct dm_thin_new_mapping *m = get_next_mapping(pool);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000855
Mike Snitzer7f214662013-12-17 13:43:31 -0500856 m->quiesced = true;
857 m->prepared = false;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000858 m->tc = tc;
859 m->virt_block = virt_block;
860 m->data_block = data_block;
861 m->cell = cell;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000862
863 /*
864 * If the whole block of data is being overwritten or we are not
865 * zeroing pre-existing data, we can issue the bio immediately.
866 * Otherwise we use kcopyd to zero the data first.
867 */
Joe Thornber67e2e2b2012-03-28 18:41:29 +0100868 if (!pool->pf.zero_new_blocks)
Joe Thornber991d9fa2011-10-31 20:21:18 +0000869 process_prepared_mapping(m);
870
871 else if (io_overwrites_block(pool, bio)) {
Mikulas Patocka59c3d2c2012-12-21 20:23:40 +0000872 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
Mike Snitzera24c2562012-06-03 00:30:00 +0100873
Joe Thornbereb2aa482012-03-28 18:41:28 +0100874 h->overwrite_mapping = m;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000875 m->bio = bio;
876 save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio);
Joe Thornbere8088072012-12-21 20:23:31 +0000877 inc_all_io_entry(pool, bio);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000878 remap_and_issue(tc, bio, data_block);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000879 } else {
880 int r;
881 struct dm_io_region to;
882
883 to.bdev = tc->pool_dev->bdev;
884 to.sector = data_block * pool->sectors_per_block;
885 to.count = pool->sectors_per_block;
886
887 r = dm_kcopyd_zero(pool->copier, 1, &to, 0, copy_complete, m);
888 if (r < 0) {
889 mempool_free(m, pool->mapping_pool);
Mike Snitzerc3977412012-12-21 20:23:34 +0000890 DMERR_LIMIT("dm_kcopyd_zero() failed");
Joe Thornber6beca5e2013-03-01 22:45:50 +0000891 cell_error(pool, cell);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000892 }
893 }
894}
895
Joe Thornbere49e5822012-07-27 15:08:16 +0100896/*
897 * A non-zero return indicates read_only or fail_io mode.
898 * Many callers don't care about the return value.
899 */
Joe Thornber020cc3b2013-12-04 15:05:36 -0500900static int commit(struct pool *pool)
Joe Thornbere49e5822012-07-27 15:08:16 +0100901{
902 int r;
903
904 if (get_pool_mode(pool) != PM_WRITE)
905 return -EINVAL;
906
Joe Thornber020cc3b2013-12-04 15:05:36 -0500907 r = dm_pool_commit_metadata(pool->pmd);
Joe Thornberb5330652013-12-04 19:51:33 -0500908 if (r)
909 metadata_operation_failed(pool, "dm_pool_commit_metadata", r);
Joe Thornbere49e5822012-07-27 15:08:16 +0100910
911 return r;
912}
913
Joe Thornber88a66212013-12-04 20:16:12 -0500914static void check_low_water_mark(struct pool *pool, dm_block_t free_blocks)
915{
916 unsigned long flags;
917
918 if (free_blocks <= pool->low_water_blocks && !pool->low_water_triggered) {
919 DMWARN("%s: reached low water mark for data device: sending event.",
920 dm_device_name(pool->pool_md));
921 spin_lock_irqsave(&pool->lock, flags);
922 pool->low_water_triggered = true;
923 spin_unlock_irqrestore(&pool->lock, flags);
924 dm_table_event(pool->ti->table);
925 }
926}
927
Joe Thornber991d9fa2011-10-31 20:21:18 +0000928static int alloc_data_block(struct thin_c *tc, dm_block_t *result)
929{
930 int r;
931 dm_block_t free_blocks;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000932 struct pool *pool = tc->pool;
933
Joe Thornber8d30abf2013-12-04 19:16:11 -0500934 if (get_pool_mode(pool) != PM_WRITE)
935 return -EINVAL;
936
Joe Thornber991d9fa2011-10-31 20:21:18 +0000937 r = dm_pool_get_free_block_count(pool->pmd, &free_blocks);
Joe Thornberb5330652013-12-04 19:51:33 -0500938 if (r) {
939 metadata_operation_failed(pool, "dm_pool_get_free_block_count", r);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000940 return r;
Joe Thornberb5330652013-12-04 19:51:33 -0500941 }
Joe Thornber991d9fa2011-10-31 20:21:18 +0000942
Joe Thornber88a66212013-12-04 20:16:12 -0500943 check_low_water_mark(pool, free_blocks);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000944
945 if (!free_blocks) {
Mike Snitzer94563ba2013-08-22 09:56:18 -0400946 /*
947 * Try to commit to see if that will free up some
948 * more space.
949 */
Joe Thornber020cc3b2013-12-04 15:05:36 -0500950 r = commit(pool);
951 if (r)
952 return r;
Mike Snitzer94563ba2013-08-22 09:56:18 -0400953
954 r = dm_pool_get_free_block_count(pool->pmd, &free_blocks);
Joe Thornberb5330652013-12-04 19:51:33 -0500955 if (r) {
956 metadata_operation_failed(pool, "dm_pool_get_free_block_count", r);
Mike Snitzer94563ba2013-08-22 09:56:18 -0400957 return r;
Joe Thornberb5330652013-12-04 19:51:33 -0500958 }
Mike Snitzer94563ba2013-08-22 09:56:18 -0400959
Mike Snitzer94563ba2013-08-22 09:56:18 -0400960 if (!free_blocks) {
Mike Snitzer399cadd2013-12-05 16:03:33 -0500961 out_of_data_space(pool);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000962 return -ENOSPC;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000963 }
964 }
965
966 r = dm_pool_alloc_data_block(pool->pmd, result);
Mike Snitzer4a02b342013-12-03 12:20:57 -0500967 if (r) {
Joe Thornberb5330652013-12-04 19:51:33 -0500968 metadata_operation_failed(pool, "dm_pool_alloc_data_block", r);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000969 return r;
Mike Snitzer4a02b342013-12-03 12:20:57 -0500970 }
Joe Thornber991d9fa2011-10-31 20:21:18 +0000971
972 return 0;
973}
974
975/*
976 * If we have run out of space, queue bios until the device is
977 * resumed, presumably after having been reloaded with more space.
978 */
979static void retry_on_resume(struct bio *bio)
980{
Mikulas Patocka59c3d2c2012-12-21 20:23:40 +0000981 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
Joe Thornbereb2aa482012-03-28 18:41:28 +0100982 struct thin_c *tc = h->tc;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000983 struct pool *pool = tc->pool;
984 unsigned long flags;
985
986 spin_lock_irqsave(&pool->lock, flags);
987 bio_list_add(&pool->retry_on_resume_list, bio);
988 spin_unlock_irqrestore(&pool->lock, flags);
989}
990
Mike Snitzer8c0f0e82013-12-05 15:47:24 -0500991static void handle_unserviceable_bio(struct pool *pool, struct bio *bio)
992{
993 /*
994 * When pool is read-only, no cell locking is needed because
995 * nothing is changing.
996 */
997 WARN_ON_ONCE(get_pool_mode(pool) != PM_READ_ONLY);
998
Mike Snitzer6d162022013-12-20 18:09:02 -0500999 if (pool->pf.error_if_no_space)
Mike Snitzer8c0f0e82013-12-05 15:47:24 -05001000 bio_io_error(bio);
Mike Snitzer6d162022013-12-20 18:09:02 -05001001 else
1002 retry_on_resume(bio);
Mike Snitzer8c0f0e82013-12-05 15:47:24 -05001003}
1004
Mike Snitzer399cadd2013-12-05 16:03:33 -05001005static void retry_bios_on_resume(struct pool *pool, struct dm_bio_prison_cell *cell)
Joe Thornber991d9fa2011-10-31 20:21:18 +00001006{
1007 struct bio *bio;
1008 struct bio_list bios;
1009
1010 bio_list_init(&bios);
Joe Thornber6beca5e2013-03-01 22:45:50 +00001011 cell_release(pool, cell, &bios);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001012
1013 while ((bio = bio_list_pop(&bios)))
Mike Snitzer8c0f0e82013-12-05 15:47:24 -05001014 handle_unserviceable_bio(pool, bio);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001015}
1016
Joe Thornber104655f2012-03-28 18:41:28 +01001017static void process_discard(struct thin_c *tc, struct bio *bio)
1018{
1019 int r;
Mike Snitzerc3a0ce22012-05-12 01:43:16 +01001020 unsigned long flags;
Joe Thornber104655f2012-03-28 18:41:28 +01001021 struct pool *pool = tc->pool;
Mike Snitzera24c2562012-06-03 00:30:00 +01001022 struct dm_bio_prison_cell *cell, *cell2;
Mike Snitzer44feb382012-10-12 21:02:10 +01001023 struct dm_cell_key key, key2;
Joe Thornber104655f2012-03-28 18:41:28 +01001024 dm_block_t block = get_bio_block(tc, bio);
1025 struct dm_thin_lookup_result lookup_result;
Mike Snitzera24c2562012-06-03 00:30:00 +01001026 struct dm_thin_new_mapping *m;
Joe Thornber104655f2012-03-28 18:41:28 +01001027
1028 build_virtual_key(tc->td, block, &key);
Joe Thornber6beca5e2013-03-01 22:45:50 +00001029 if (bio_detain(tc->pool, &key, bio, &cell))
Joe Thornber104655f2012-03-28 18:41:28 +01001030 return;
1031
1032 r = dm_thin_find_block(tc->td, block, 1, &lookup_result);
1033 switch (r) {
1034 case 0:
1035 /*
1036 * Check nobody is fiddling with this pool block. This can
1037 * happen if someone's in the process of breaking sharing
1038 * on this block.
1039 */
1040 build_data_key(tc->td, lookup_result.block, &key2);
Joe Thornber6beca5e2013-03-01 22:45:50 +00001041 if (bio_detain(tc->pool, &key2, bio, &cell2)) {
Joe Thornberf286ba02012-12-21 20:23:33 +00001042 cell_defer_no_holder(tc, cell);
Joe Thornber104655f2012-03-28 18:41:28 +01001043 break;
1044 }
1045
1046 if (io_overlaps_block(pool, bio)) {
1047 /*
1048 * IO may still be going to the destination block. We must
1049 * quiesce before we can do the removal.
1050 */
1051 m = get_next_mapping(pool);
1052 m->tc = tc;
Joe Thornber19fa1a62013-12-17 12:09:40 -05001053 m->pass_discard = pool->pf.discard_passdown;
1054 m->definitely_not_shared = !lookup_result.shared;
Joe Thornber104655f2012-03-28 18:41:28 +01001055 m->virt_block = block;
1056 m->data_block = lookup_result.block;
1057 m->cell = cell;
1058 m->cell2 = cell2;
Joe Thornber104655f2012-03-28 18:41:28 +01001059 m->bio = bio;
1060
Mike Snitzer44feb382012-10-12 21:02:10 +01001061 if (!dm_deferred_set_add_work(pool->all_io_ds, &m->list)) {
Mike Snitzerc3a0ce22012-05-12 01:43:16 +01001062 spin_lock_irqsave(&pool->lock, flags);
Mike Snitzerdaec3382013-12-11 14:01:20 -05001063 list_add_tail(&m->list, &pool->prepared_discards);
Mike Snitzerc3a0ce22012-05-12 01:43:16 +01001064 spin_unlock_irqrestore(&pool->lock, flags);
Joe Thornber104655f2012-03-28 18:41:28 +01001065 wake_worker(pool);
1066 }
1067 } else {
Joe Thornbere8088072012-12-21 20:23:31 +00001068 inc_all_io_entry(pool, bio);
Joe Thornberf286ba02012-12-21 20:23:33 +00001069 cell_defer_no_holder(tc, cell);
1070 cell_defer_no_holder(tc, cell2);
Joe Thornbere8088072012-12-21 20:23:31 +00001071
Joe Thornber104655f2012-03-28 18:41:28 +01001072 /*
Mikulas Patocka49296302012-07-27 15:08:03 +01001073 * The DM core makes sure that the discard doesn't span
1074 * a block boundary. So we submit the discard of a
1075 * partial block appropriately.
Joe Thornber104655f2012-03-28 18:41:28 +01001076 */
Mikulas Patocka650d2a02012-07-20 14:25:05 +01001077 if ((!lookup_result.shared) && pool->pf.discard_passdown)
1078 remap_and_issue(tc, bio, lookup_result.block);
1079 else
1080 bio_endio(bio, 0);
Joe Thornber104655f2012-03-28 18:41:28 +01001081 }
1082 break;
1083
1084 case -ENODATA:
1085 /*
1086 * It isn't provisioned, just forget it.
1087 */
Joe Thornberf286ba02012-12-21 20:23:33 +00001088 cell_defer_no_holder(tc, cell);
Joe Thornber104655f2012-03-28 18:41:28 +01001089 bio_endio(bio, 0);
1090 break;
1091
1092 default:
Mike Snitzerc3977412012-12-21 20:23:34 +00001093 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1094 __func__, r);
Joe Thornberf286ba02012-12-21 20:23:33 +00001095 cell_defer_no_holder(tc, cell);
Joe Thornber104655f2012-03-28 18:41:28 +01001096 bio_io_error(bio);
1097 break;
1098 }
1099}
1100
Joe Thornber991d9fa2011-10-31 20:21:18 +00001101static void break_sharing(struct thin_c *tc, struct bio *bio, dm_block_t block,
Mike Snitzer44feb382012-10-12 21:02:10 +01001102 struct dm_cell_key *key,
Joe Thornber991d9fa2011-10-31 20:21:18 +00001103 struct dm_thin_lookup_result *lookup_result,
Mike Snitzera24c2562012-06-03 00:30:00 +01001104 struct dm_bio_prison_cell *cell)
Joe Thornber991d9fa2011-10-31 20:21:18 +00001105{
1106 int r;
1107 dm_block_t data_block;
Mike Snitzerd6fc2042013-08-21 17:40:11 -04001108 struct pool *pool = tc->pool;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001109
1110 r = alloc_data_block(tc, &data_block);
1111 switch (r) {
1112 case 0:
Joe Thornber2dd9c252012-03-28 18:41:28 +01001113 schedule_internal_copy(tc, block, lookup_result->block,
1114 data_block, cell, bio);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001115 break;
1116
1117 case -ENOSPC:
Mike Snitzer399cadd2013-12-05 16:03:33 -05001118 retry_bios_on_resume(pool, cell);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001119 break;
1120
1121 default:
Mike Snitzerc3977412012-12-21 20:23:34 +00001122 DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1123 __func__, r);
Mike Snitzerd6fc2042013-08-21 17:40:11 -04001124 cell_error(pool, cell);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001125 break;
1126 }
1127}
1128
1129static void process_shared_bio(struct thin_c *tc, struct bio *bio,
1130 dm_block_t block,
1131 struct dm_thin_lookup_result *lookup_result)
1132{
Mike Snitzera24c2562012-06-03 00:30:00 +01001133 struct dm_bio_prison_cell *cell;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001134 struct pool *pool = tc->pool;
Mike Snitzer44feb382012-10-12 21:02:10 +01001135 struct dm_cell_key key;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001136
1137 /*
1138 * If cell is already occupied, then sharing is already in the process
1139 * of being broken so we have nothing further to do here.
1140 */
1141 build_data_key(tc->td, lookup_result->block, &key);
Joe Thornber6beca5e2013-03-01 22:45:50 +00001142 if (bio_detain(pool, &key, bio, &cell))
Joe Thornber991d9fa2011-10-31 20:21:18 +00001143 return;
1144
Kent Overstreet4f024f32013-10-11 15:44:27 -07001145 if (bio_data_dir(bio) == WRITE && bio->bi_iter.bi_size)
Joe Thornber991d9fa2011-10-31 20:21:18 +00001146 break_sharing(tc, bio, block, &key, lookup_result, cell);
1147 else {
Mikulas Patocka59c3d2c2012-12-21 20:23:40 +00001148 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
Joe Thornber991d9fa2011-10-31 20:21:18 +00001149
Mike Snitzer44feb382012-10-12 21:02:10 +01001150 h->shared_read_entry = dm_deferred_entry_inc(pool->shared_read_ds);
Joe Thornbere8088072012-12-21 20:23:31 +00001151 inc_all_io_entry(pool, bio);
Joe Thornberf286ba02012-12-21 20:23:33 +00001152 cell_defer_no_holder(tc, cell);
Joe Thornbere8088072012-12-21 20:23:31 +00001153
Joe Thornber991d9fa2011-10-31 20:21:18 +00001154 remap_and_issue(tc, bio, lookup_result->block);
1155 }
1156}
1157
1158static void provision_block(struct thin_c *tc, struct bio *bio, dm_block_t block,
Mike Snitzera24c2562012-06-03 00:30:00 +01001159 struct dm_bio_prison_cell *cell)
Joe Thornber991d9fa2011-10-31 20:21:18 +00001160{
1161 int r;
1162 dm_block_t data_block;
Joe Thornber6beca5e2013-03-01 22:45:50 +00001163 struct pool *pool = tc->pool;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001164
1165 /*
1166 * Remap empty bios (flushes) immediately, without provisioning.
1167 */
Kent Overstreet4f024f32013-10-11 15:44:27 -07001168 if (!bio->bi_iter.bi_size) {
Joe Thornber6beca5e2013-03-01 22:45:50 +00001169 inc_all_io_entry(pool, bio);
Joe Thornberf286ba02012-12-21 20:23:33 +00001170 cell_defer_no_holder(tc, cell);
Joe Thornbere8088072012-12-21 20:23:31 +00001171
Joe Thornber991d9fa2011-10-31 20:21:18 +00001172 remap_and_issue(tc, bio, 0);
1173 return;
1174 }
1175
1176 /*
1177 * Fill read bios with zeroes and complete them immediately.
1178 */
1179 if (bio_data_dir(bio) == READ) {
1180 zero_fill_bio(bio);
Joe Thornberf286ba02012-12-21 20:23:33 +00001181 cell_defer_no_holder(tc, cell);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001182 bio_endio(bio, 0);
1183 return;
1184 }
1185
1186 r = alloc_data_block(tc, &data_block);
1187 switch (r) {
1188 case 0:
Joe Thornber2dd9c252012-03-28 18:41:28 +01001189 if (tc->origin_dev)
1190 schedule_external_copy(tc, block, data_block, cell, bio);
1191 else
1192 schedule_zero(tc, block, data_block, cell, bio);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001193 break;
1194
1195 case -ENOSPC:
Mike Snitzer399cadd2013-12-05 16:03:33 -05001196 retry_bios_on_resume(pool, cell);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001197 break;
1198
1199 default:
Mike Snitzerc3977412012-12-21 20:23:34 +00001200 DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1201 __func__, r);
Joe Thornber6beca5e2013-03-01 22:45:50 +00001202 cell_error(pool, cell);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001203 break;
1204 }
1205}
1206
1207static void process_bio(struct thin_c *tc, struct bio *bio)
1208{
1209 int r;
Joe Thornber6beca5e2013-03-01 22:45:50 +00001210 struct pool *pool = tc->pool;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001211 dm_block_t block = get_bio_block(tc, bio);
Mike Snitzera24c2562012-06-03 00:30:00 +01001212 struct dm_bio_prison_cell *cell;
Mike Snitzer44feb382012-10-12 21:02:10 +01001213 struct dm_cell_key key;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001214 struct dm_thin_lookup_result lookup_result;
1215
1216 /*
1217 * If cell is already occupied, then the block is already
1218 * being provisioned so we have nothing further to do here.
1219 */
1220 build_virtual_key(tc->td, block, &key);
Joe Thornber6beca5e2013-03-01 22:45:50 +00001221 if (bio_detain(pool, &key, bio, &cell))
Joe Thornber991d9fa2011-10-31 20:21:18 +00001222 return;
1223
1224 r = dm_thin_find_block(tc->td, block, 1, &lookup_result);
1225 switch (r) {
1226 case 0:
Joe Thornbere8088072012-12-21 20:23:31 +00001227 if (lookup_result.shared) {
Joe Thornber991d9fa2011-10-31 20:21:18 +00001228 process_shared_bio(tc, bio, block, &lookup_result);
Joe Thornber6beca5e2013-03-01 22:45:50 +00001229 cell_defer_no_holder(tc, cell); /* FIXME: pass this cell into process_shared? */
Joe Thornbere8088072012-12-21 20:23:31 +00001230 } else {
Joe Thornber6beca5e2013-03-01 22:45:50 +00001231 inc_all_io_entry(pool, bio);
Joe Thornberf286ba02012-12-21 20:23:33 +00001232 cell_defer_no_holder(tc, cell);
Joe Thornbere8088072012-12-21 20:23:31 +00001233
Joe Thornber991d9fa2011-10-31 20:21:18 +00001234 remap_and_issue(tc, bio, lookup_result.block);
Joe Thornbere8088072012-12-21 20:23:31 +00001235 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00001236 break;
1237
1238 case -ENODATA:
Joe Thornber2dd9c252012-03-28 18:41:28 +01001239 if (bio_data_dir(bio) == READ && tc->origin_dev) {
Joe Thornber6beca5e2013-03-01 22:45:50 +00001240 inc_all_io_entry(pool, bio);
Joe Thornberf286ba02012-12-21 20:23:33 +00001241 cell_defer_no_holder(tc, cell);
Joe Thornbere8088072012-12-21 20:23:31 +00001242
Joe Thornber2dd9c252012-03-28 18:41:28 +01001243 remap_to_origin_and_issue(tc, bio);
1244 } else
1245 provision_block(tc, bio, block, cell);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001246 break;
1247
1248 default:
Mike Snitzerc3977412012-12-21 20:23:34 +00001249 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1250 __func__, r);
Joe Thornberf286ba02012-12-21 20:23:33 +00001251 cell_defer_no_holder(tc, cell);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001252 bio_io_error(bio);
1253 break;
1254 }
1255}
1256
Joe Thornbere49e5822012-07-27 15:08:16 +01001257static void process_bio_read_only(struct thin_c *tc, struct bio *bio)
1258{
1259 int r;
1260 int rw = bio_data_dir(bio);
1261 dm_block_t block = get_bio_block(tc, bio);
1262 struct dm_thin_lookup_result lookup_result;
1263
1264 r = dm_thin_find_block(tc->td, block, 1, &lookup_result);
1265 switch (r) {
1266 case 0:
Kent Overstreet4f024f32013-10-11 15:44:27 -07001267 if (lookup_result.shared && (rw == WRITE) && bio->bi_iter.bi_size)
Mike Snitzer8c0f0e82013-12-05 15:47:24 -05001268 handle_unserviceable_bio(tc->pool, bio);
Joe Thornbere8088072012-12-21 20:23:31 +00001269 else {
1270 inc_all_io_entry(tc->pool, bio);
Joe Thornbere49e5822012-07-27 15:08:16 +01001271 remap_and_issue(tc, bio, lookup_result.block);
Joe Thornbere8088072012-12-21 20:23:31 +00001272 }
Joe Thornbere49e5822012-07-27 15:08:16 +01001273 break;
1274
1275 case -ENODATA:
1276 if (rw != READ) {
Mike Snitzer8c0f0e82013-12-05 15:47:24 -05001277 handle_unserviceable_bio(tc->pool, bio);
Joe Thornbere49e5822012-07-27 15:08:16 +01001278 break;
1279 }
1280
1281 if (tc->origin_dev) {
Joe Thornbere8088072012-12-21 20:23:31 +00001282 inc_all_io_entry(tc->pool, bio);
Joe Thornbere49e5822012-07-27 15:08:16 +01001283 remap_to_origin_and_issue(tc, bio);
1284 break;
1285 }
1286
1287 zero_fill_bio(bio);
1288 bio_endio(bio, 0);
1289 break;
1290
1291 default:
Mike Snitzerc3977412012-12-21 20:23:34 +00001292 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1293 __func__, r);
Joe Thornbere49e5822012-07-27 15:08:16 +01001294 bio_io_error(bio);
1295 break;
1296 }
1297}
1298
1299static void process_bio_fail(struct thin_c *tc, struct bio *bio)
1300{
1301 bio_io_error(bio);
1302}
1303
Joe Thornberac8c3f32013-05-10 14:37:21 +01001304/*
1305 * FIXME: should we also commit due to size of transaction, measured in
1306 * metadata blocks?
1307 */
Joe Thornber905e51b2012-03-28 18:41:27 +01001308static int need_commit_due_to_time(struct pool *pool)
1309{
1310 return jiffies < pool->last_commit_jiffies ||
1311 jiffies > pool->last_commit_jiffies + COMMIT_PERIOD;
1312}
1313
Joe Thornber991d9fa2011-10-31 20:21:18 +00001314static void process_deferred_bios(struct pool *pool)
1315{
1316 unsigned long flags;
1317 struct bio *bio;
1318 struct bio_list bios;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001319
1320 bio_list_init(&bios);
1321
1322 spin_lock_irqsave(&pool->lock, flags);
1323 bio_list_merge(&bios, &pool->deferred_bios);
1324 bio_list_init(&pool->deferred_bios);
1325 spin_unlock_irqrestore(&pool->lock, flags);
1326
1327 while ((bio = bio_list_pop(&bios))) {
Mikulas Patocka59c3d2c2012-12-21 20:23:40 +00001328 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
Joe Thornbereb2aa482012-03-28 18:41:28 +01001329 struct thin_c *tc = h->tc;
1330
Joe Thornber991d9fa2011-10-31 20:21:18 +00001331 /*
1332 * If we've got no free new_mapping structs, and processing
1333 * this bio might require one, we pause until there are some
1334 * prepared mappings to process.
1335 */
1336 if (ensure_next_mapping(pool)) {
1337 spin_lock_irqsave(&pool->lock, flags);
1338 bio_list_merge(&pool->deferred_bios, &bios);
1339 spin_unlock_irqrestore(&pool->lock, flags);
1340
1341 break;
1342 }
Joe Thornber104655f2012-03-28 18:41:28 +01001343
1344 if (bio->bi_rw & REQ_DISCARD)
Joe Thornbere49e5822012-07-27 15:08:16 +01001345 pool->process_discard(tc, bio);
Joe Thornber104655f2012-03-28 18:41:28 +01001346 else
Joe Thornbere49e5822012-07-27 15:08:16 +01001347 pool->process_bio(tc, bio);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001348 }
1349
1350 /*
1351 * If there are any deferred flush bios, we must commit
1352 * the metadata before issuing them.
1353 */
1354 bio_list_init(&bios);
1355 spin_lock_irqsave(&pool->lock, flags);
1356 bio_list_merge(&bios, &pool->deferred_flush_bios);
1357 bio_list_init(&pool->deferred_flush_bios);
1358 spin_unlock_irqrestore(&pool->lock, flags);
1359
Mike Snitzer4d1662a2014-02-06 06:08:56 -05001360 if (bio_list_empty(&bios) &&
1361 !(dm_pool_changed_this_transaction(pool->pmd) && need_commit_due_to_time(pool)))
Joe Thornber991d9fa2011-10-31 20:21:18 +00001362 return;
1363
Joe Thornber020cc3b2013-12-04 15:05:36 -05001364 if (commit(pool)) {
Joe Thornber991d9fa2011-10-31 20:21:18 +00001365 while ((bio = bio_list_pop(&bios)))
1366 bio_io_error(bio);
1367 return;
1368 }
Joe Thornber905e51b2012-03-28 18:41:27 +01001369 pool->last_commit_jiffies = jiffies;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001370
1371 while ((bio = bio_list_pop(&bios)))
1372 generic_make_request(bio);
1373}
1374
1375static void do_worker(struct work_struct *ws)
1376{
1377 struct pool *pool = container_of(ws, struct pool, worker);
1378
Joe Thornbere49e5822012-07-27 15:08:16 +01001379 process_prepared(pool, &pool->prepared_mappings, &pool->process_prepared_mapping);
1380 process_prepared(pool, &pool->prepared_discards, &pool->process_prepared_discard);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001381 process_deferred_bios(pool);
1382}
1383
Joe Thornber905e51b2012-03-28 18:41:27 +01001384/*
1385 * We want to commit periodically so that not too much
1386 * unwritten data builds up.
1387 */
1388static void do_waker(struct work_struct *ws)
1389{
1390 struct pool *pool = container_of(to_delayed_work(ws), struct pool, waker);
1391 wake_worker(pool);
1392 queue_delayed_work(pool->wq, &pool->waker, COMMIT_PERIOD);
1393}
1394
Joe Thornber991d9fa2011-10-31 20:21:18 +00001395/*----------------------------------------------------------------*/
1396
Joe Thornbere49e5822012-07-27 15:08:16 +01001397static enum pool_mode get_pool_mode(struct pool *pool)
1398{
1399 return pool->pf.mode;
1400}
1401
Mike Snitzer8b64e882013-12-20 14:27:28 -05001402static void set_pool_mode(struct pool *pool, enum pool_mode new_mode)
Joe Thornbere49e5822012-07-27 15:08:16 +01001403{
1404 int r;
Mike Snitzer8b64e882013-12-20 14:27:28 -05001405 enum pool_mode old_mode = pool->pf.mode;
Joe Thornbere49e5822012-07-27 15:08:16 +01001406
Mike Snitzer8b64e882013-12-20 14:27:28 -05001407 switch (new_mode) {
Joe Thornbere49e5822012-07-27 15:08:16 +01001408 case PM_FAIL:
Mike Snitzer8b64e882013-12-20 14:27:28 -05001409 if (old_mode != new_mode)
1410 DMERR("%s: switching pool to failure mode",
1411 dm_device_name(pool->pool_md));
Joe Thornber5383ef32013-12-04 16:30:01 -05001412 dm_pool_metadata_read_only(pool->pmd);
Joe Thornbere49e5822012-07-27 15:08:16 +01001413 pool->process_bio = process_bio_fail;
1414 pool->process_discard = process_bio_fail;
1415 pool->process_prepared_mapping = process_prepared_mapping_fail;
1416 pool->process_prepared_discard = process_prepared_discard_fail;
1417 break;
1418
1419 case PM_READ_ONLY:
Mike Snitzer8b64e882013-12-20 14:27:28 -05001420 if (old_mode != new_mode)
1421 DMERR("%s: switching pool to read-only mode",
1422 dm_device_name(pool->pool_md));
Joe Thornbere49e5822012-07-27 15:08:16 +01001423 r = dm_pool_abort_metadata(pool->pmd);
1424 if (r) {
Mike Snitzer4fa59712013-08-21 17:30:40 -04001425 DMERR("%s: aborting transaction failed",
1426 dm_device_name(pool->pool_md));
Mike Snitzer8b64e882013-12-20 14:27:28 -05001427 new_mode = PM_FAIL;
1428 set_pool_mode(pool, new_mode);
Joe Thornbere49e5822012-07-27 15:08:16 +01001429 } else {
1430 dm_pool_metadata_read_only(pool->pmd);
1431 pool->process_bio = process_bio_read_only;
1432 pool->process_discard = process_discard;
1433 pool->process_prepared_mapping = process_prepared_mapping_fail;
1434 pool->process_prepared_discard = process_prepared_discard_passdown;
1435 }
1436 break;
1437
1438 case PM_WRITE:
Mike Snitzer8b64e882013-12-20 14:27:28 -05001439 if (old_mode != new_mode)
1440 DMINFO("%s: switching pool to write mode",
1441 dm_device_name(pool->pool_md));
Joe Thornber9b7aaa62013-12-04 16:58:19 -05001442 dm_pool_metadata_read_write(pool->pmd);
Joe Thornbere49e5822012-07-27 15:08:16 +01001443 pool->process_bio = process_bio;
1444 pool->process_discard = process_discard;
1445 pool->process_prepared_mapping = process_prepared_mapping;
1446 pool->process_prepared_discard = process_prepared_discard;
1447 break;
1448 }
Mike Snitzer8b64e882013-12-20 14:27:28 -05001449
1450 pool->pf.mode = new_mode;
Joe Thornbere49e5822012-07-27 15:08:16 +01001451}
1452
Joe Thornberb5330652013-12-04 19:51:33 -05001453/*
1454 * Rather than calling set_pool_mode directly, use these which describe the
1455 * reason for mode degradation.
1456 */
Mike Snitzer399cadd2013-12-05 16:03:33 -05001457static void out_of_data_space(struct pool *pool)
1458{
1459 DMERR_LIMIT("%s: no free data space available.",
1460 dm_device_name(pool->pool_md));
Mike Snitzer399cadd2013-12-05 16:03:33 -05001461 set_pool_mode(pool, PM_READ_ONLY);
1462}
1463
Joe Thornberb5330652013-12-04 19:51:33 -05001464static void metadata_operation_failed(struct pool *pool, const char *op, int r)
1465{
Mike Snitzer399cadd2013-12-05 16:03:33 -05001466 dm_block_t free_blocks;
1467
Joe Thornberb5330652013-12-04 19:51:33 -05001468 DMERR_LIMIT("%s: metadata operation '%s' failed: error = %d",
1469 dm_device_name(pool->pool_md), op, r);
1470
Mike Snitzer399cadd2013-12-05 16:03:33 -05001471 if (r == -ENOSPC &&
1472 !dm_pool_get_free_metadata_block_count(pool->pmd, &free_blocks) &&
Mike Snitzer6d162022013-12-20 18:09:02 -05001473 !free_blocks)
Mike Snitzer399cadd2013-12-05 16:03:33 -05001474 DMERR_LIMIT("%s: no free metadata space available.",
1475 dm_device_name(pool->pool_md));
Mike Snitzer399cadd2013-12-05 16:03:33 -05001476
Joe Thornberb5330652013-12-04 19:51:33 -05001477 set_pool_mode(pool, PM_READ_ONLY);
1478}
1479
Joe Thornbere49e5822012-07-27 15:08:16 +01001480/*----------------------------------------------------------------*/
1481
Joe Thornber991d9fa2011-10-31 20:21:18 +00001482/*
1483 * Mapping functions.
1484 */
1485
1486/*
1487 * Called only while mapping a thin bio to hand it over to the workqueue.
1488 */
1489static void thin_defer_bio(struct thin_c *tc, struct bio *bio)
1490{
1491 unsigned long flags;
1492 struct pool *pool = tc->pool;
1493
1494 spin_lock_irqsave(&pool->lock, flags);
1495 bio_list_add(&pool->deferred_bios, bio);
1496 spin_unlock_irqrestore(&pool->lock, flags);
1497
1498 wake_worker(pool);
1499}
1500
Mikulas Patocka59c3d2c2012-12-21 20:23:40 +00001501static void thin_hook_bio(struct thin_c *tc, struct bio *bio)
Joe Thornbereb2aa482012-03-28 18:41:28 +01001502{
Mikulas Patocka59c3d2c2012-12-21 20:23:40 +00001503 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
Joe Thornbereb2aa482012-03-28 18:41:28 +01001504
1505 h->tc = tc;
1506 h->shared_read_entry = NULL;
Joe Thornbere8088072012-12-21 20:23:31 +00001507 h->all_io_entry = NULL;
Joe Thornbereb2aa482012-03-28 18:41:28 +01001508 h->overwrite_mapping = NULL;
Joe Thornbereb2aa482012-03-28 18:41:28 +01001509}
1510
Joe Thornber991d9fa2011-10-31 20:21:18 +00001511/*
1512 * Non-blocking function called from the thin target's map function.
1513 */
Mikulas Patocka7de3ee52012-12-21 20:23:41 +00001514static int thin_bio_map(struct dm_target *ti, struct bio *bio)
Joe Thornber991d9fa2011-10-31 20:21:18 +00001515{
1516 int r;
1517 struct thin_c *tc = ti->private;
1518 dm_block_t block = get_bio_block(tc, bio);
1519 struct dm_thin_device *td = tc->td;
1520 struct dm_thin_lookup_result result;
Joe Thornber025b9682013-03-01 22:45:50 +00001521 struct dm_bio_prison_cell cell1, cell2;
1522 struct dm_bio_prison_cell *cell_result;
Joe Thornbere8088072012-12-21 20:23:31 +00001523 struct dm_cell_key key;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001524
Mikulas Patocka59c3d2c2012-12-21 20:23:40 +00001525 thin_hook_bio(tc, bio);
Joe Thornbere49e5822012-07-27 15:08:16 +01001526
1527 if (get_pool_mode(tc->pool) == PM_FAIL) {
1528 bio_io_error(bio);
1529 return DM_MAPIO_SUBMITTED;
1530 }
1531
Joe Thornber104655f2012-03-28 18:41:28 +01001532 if (bio->bi_rw & (REQ_DISCARD | REQ_FLUSH | REQ_FUA)) {
Joe Thornber991d9fa2011-10-31 20:21:18 +00001533 thin_defer_bio(tc, bio);
1534 return DM_MAPIO_SUBMITTED;
1535 }
1536
1537 r = dm_thin_find_block(td, block, 0, &result);
1538
1539 /*
1540 * Note that we defer readahead too.
1541 */
1542 switch (r) {
1543 case 0:
1544 if (unlikely(result.shared)) {
1545 /*
1546 * We have a race condition here between the
1547 * result.shared value returned by the lookup and
1548 * snapshot creation, which may cause new
1549 * sharing.
1550 *
1551 * To avoid this always quiesce the origin before
1552 * taking the snap. You want to do this anyway to
1553 * ensure a consistent application view
1554 * (i.e. lockfs).
1555 *
1556 * More distant ancestors are irrelevant. The
1557 * shared flag will be set in their case.
1558 */
1559 thin_defer_bio(tc, bio);
Joe Thornbere8088072012-12-21 20:23:31 +00001560 return DM_MAPIO_SUBMITTED;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001561 }
Joe Thornbere8088072012-12-21 20:23:31 +00001562
1563 build_virtual_key(tc->td, block, &key);
Joe Thornber025b9682013-03-01 22:45:50 +00001564 if (dm_bio_detain(tc->pool->prison, &key, bio, &cell1, &cell_result))
Joe Thornbere8088072012-12-21 20:23:31 +00001565 return DM_MAPIO_SUBMITTED;
1566
1567 build_data_key(tc->td, result.block, &key);
Joe Thornber025b9682013-03-01 22:45:50 +00001568 if (dm_bio_detain(tc->pool->prison, &key, bio, &cell2, &cell_result)) {
1569 cell_defer_no_holder_no_free(tc, &cell1);
Joe Thornbere8088072012-12-21 20:23:31 +00001570 return DM_MAPIO_SUBMITTED;
1571 }
1572
1573 inc_all_io_entry(tc->pool, bio);
Joe Thornber025b9682013-03-01 22:45:50 +00001574 cell_defer_no_holder_no_free(tc, &cell2);
1575 cell_defer_no_holder_no_free(tc, &cell1);
Joe Thornbere8088072012-12-21 20:23:31 +00001576
1577 remap(tc, bio, result.block);
1578 return DM_MAPIO_REMAPPED;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001579
1580 case -ENODATA:
Joe Thornbere49e5822012-07-27 15:08:16 +01001581 if (get_pool_mode(tc->pool) == PM_READ_ONLY) {
1582 /*
1583 * This block isn't provisioned, and we have no way
Mike Snitzer8c0f0e82013-12-05 15:47:24 -05001584 * of doing so.
Joe Thornbere49e5822012-07-27 15:08:16 +01001585 */
Mike Snitzer8c0f0e82013-12-05 15:47:24 -05001586 handle_unserviceable_bio(tc->pool, bio);
Joe Thornber2aab3852012-12-21 20:23:33 +00001587 return DM_MAPIO_SUBMITTED;
Joe Thornbere49e5822012-07-27 15:08:16 +01001588 }
1589 /* fall through */
1590
1591 case -EWOULDBLOCK:
Joe Thornber991d9fa2011-10-31 20:21:18 +00001592 /*
1593 * In future, the failed dm_thin_find_block above could
1594 * provide the hint to load the metadata into cache.
1595 */
Joe Thornber991d9fa2011-10-31 20:21:18 +00001596 thin_defer_bio(tc, bio);
Joe Thornber2aab3852012-12-21 20:23:33 +00001597 return DM_MAPIO_SUBMITTED;
Joe Thornbere49e5822012-07-27 15:08:16 +01001598
1599 default:
1600 /*
1601 * Must always call bio_io_error on failure.
1602 * dm_thin_find_block can fail with -EINVAL if the
1603 * pool is switched to fail-io mode.
1604 */
1605 bio_io_error(bio);
Joe Thornber2aab3852012-12-21 20:23:33 +00001606 return DM_MAPIO_SUBMITTED;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001607 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00001608}
1609
1610static int pool_is_congested(struct dm_target_callbacks *cb, int bdi_bits)
1611{
1612 int r;
1613 unsigned long flags;
1614 struct pool_c *pt = container_of(cb, struct pool_c, callbacks);
1615
1616 spin_lock_irqsave(&pt->pool->lock, flags);
1617 r = !bio_list_empty(&pt->pool->retry_on_resume_list);
1618 spin_unlock_irqrestore(&pt->pool->lock, flags);
1619
1620 if (!r) {
1621 struct request_queue *q = bdev_get_queue(pt->data_dev->bdev);
1622 r = bdi_congested(&q->backing_dev_info, bdi_bits);
1623 }
1624
1625 return r;
1626}
1627
1628static void __requeue_bios(struct pool *pool)
1629{
1630 bio_list_merge(&pool->deferred_bios, &pool->retry_on_resume_list);
1631 bio_list_init(&pool->retry_on_resume_list);
1632}
1633
1634/*----------------------------------------------------------------
1635 * Binding of control targets to a pool object
1636 *--------------------------------------------------------------*/
Mike Snitzer9bc142d2012-09-26 23:45:46 +01001637static bool data_dev_supports_discard(struct pool_c *pt)
1638{
1639 struct request_queue *q = bdev_get_queue(pt->data_dev->bdev);
1640
1641 return q && blk_queue_discard(q);
1642}
1643
Joe Thornber58051b92013-03-20 17:21:25 +00001644static bool is_factor(sector_t block_size, uint32_t n)
1645{
1646 return !sector_div(block_size, n);
1647}
1648
Mike Snitzer9bc142d2012-09-26 23:45:46 +01001649/*
1650 * If discard_passdown was enabled verify that the data device
Mike Snitzer0424caa2012-09-26 23:45:47 +01001651 * supports discards. Disable discard_passdown if not.
Mike Snitzer9bc142d2012-09-26 23:45:46 +01001652 */
Mike Snitzer0424caa2012-09-26 23:45:47 +01001653static void disable_passdown_if_not_supported(struct pool_c *pt)
Mike Snitzer9bc142d2012-09-26 23:45:46 +01001654{
Mike Snitzer0424caa2012-09-26 23:45:47 +01001655 struct pool *pool = pt->pool;
1656 struct block_device *data_bdev = pt->data_dev->bdev;
1657 struct queue_limits *data_limits = &bdev_get_queue(data_bdev)->limits;
1658 sector_t block_size = pool->sectors_per_block << SECTOR_SHIFT;
1659 const char *reason = NULL;
Mike Snitzer9bc142d2012-09-26 23:45:46 +01001660 char buf[BDEVNAME_SIZE];
1661
Mike Snitzer0424caa2012-09-26 23:45:47 +01001662 if (!pt->adjusted_pf.discard_passdown)
Mike Snitzer9bc142d2012-09-26 23:45:46 +01001663 return;
1664
Mike Snitzer0424caa2012-09-26 23:45:47 +01001665 if (!data_dev_supports_discard(pt))
1666 reason = "discard unsupported";
Mike Snitzer9bc142d2012-09-26 23:45:46 +01001667
Mike Snitzer0424caa2012-09-26 23:45:47 +01001668 else if (data_limits->max_discard_sectors < pool->sectors_per_block)
1669 reason = "max discard sectors smaller than a block";
1670
1671 else if (data_limits->discard_granularity > block_size)
1672 reason = "discard granularity larger than a block";
1673
Joe Thornber58051b92013-03-20 17:21:25 +00001674 else if (!is_factor(block_size, data_limits->discard_granularity))
Mike Snitzer0424caa2012-09-26 23:45:47 +01001675 reason = "discard granularity not a factor of block size";
1676
1677 if (reason) {
1678 DMWARN("Data device (%s) %s: Disabling discard passdown.", bdevname(data_bdev, buf), reason);
1679 pt->adjusted_pf.discard_passdown = false;
1680 }
Mike Snitzer9bc142d2012-09-26 23:45:46 +01001681}
1682
Joe Thornber991d9fa2011-10-31 20:21:18 +00001683static int bind_control_target(struct pool *pool, struct dm_target *ti)
1684{
1685 struct pool_c *pt = ti->private;
1686
Joe Thornbere49e5822012-07-27 15:08:16 +01001687 /*
Joe Thornber9b7aaa62013-12-04 16:58:19 -05001688 * We want to make sure that a pool in PM_FAIL mode is never upgraded.
Joe Thornbere49e5822012-07-27 15:08:16 +01001689 */
1690 enum pool_mode old_mode = pool->pf.mode;
Mike Snitzer0424caa2012-09-26 23:45:47 +01001691 enum pool_mode new_mode = pt->adjusted_pf.mode;
Joe Thornbere49e5822012-07-27 15:08:16 +01001692
Joe Thornber9b7aaa62013-12-04 16:58:19 -05001693 /*
Mike Snitzer8b64e882013-12-20 14:27:28 -05001694 * Don't change the pool's mode until set_pool_mode() below.
1695 * Otherwise the pool's process_* function pointers may
1696 * not match the desired pool mode.
1697 */
1698 pt->adjusted_pf.mode = old_mode;
1699
1700 pool->ti = ti;
1701 pool->pf = pt->adjusted_pf;
1702 pool->low_water_blocks = pt->low_water_blocks;
1703
1704 /*
Joe Thornber9b7aaa62013-12-04 16:58:19 -05001705 * If we were in PM_FAIL mode, rollback of metadata failed. We're
1706 * not going to recover without a thin_repair. So we never let the
1707 * pool move out of the old mode. On the other hand a PM_READ_ONLY
1708 * may have been due to a lack of metadata or data space, and may
1709 * now work (ie. if the underlying devices have been resized).
1710 */
1711 if (old_mode == PM_FAIL)
Joe Thornbere49e5822012-07-27 15:08:16 +01001712 new_mode = old_mode;
1713
Mike Snitzer9bc142d2012-09-26 23:45:46 +01001714 set_pool_mode(pool, new_mode);
Mike Snitzerf4026932012-05-19 01:01:01 +01001715
Joe Thornber991d9fa2011-10-31 20:21:18 +00001716 return 0;
1717}
1718
1719static void unbind_control_target(struct pool *pool, struct dm_target *ti)
1720{
1721 if (pool->ti == ti)
1722 pool->ti = NULL;
1723}
1724
1725/*----------------------------------------------------------------
1726 * Pool creation
1727 *--------------------------------------------------------------*/
Joe Thornber67e2e2b2012-03-28 18:41:29 +01001728/* Initialize pool features. */
1729static void pool_features_init(struct pool_features *pf)
1730{
Joe Thornbere49e5822012-07-27 15:08:16 +01001731 pf->mode = PM_WRITE;
Mike Snitzer9bc142d2012-09-26 23:45:46 +01001732 pf->zero_new_blocks = true;
1733 pf->discard_enabled = true;
1734 pf->discard_passdown = true;
Mike Snitzer787a996c2013-12-06 16:21:43 -05001735 pf->error_if_no_space = false;
Joe Thornber67e2e2b2012-03-28 18:41:29 +01001736}
1737
Joe Thornber991d9fa2011-10-31 20:21:18 +00001738static void __pool_destroy(struct pool *pool)
1739{
1740 __pool_table_remove(pool);
1741
1742 if (dm_pool_metadata_close(pool->pmd) < 0)
1743 DMWARN("%s: dm_pool_metadata_close() failed.", __func__);
1744
Mike Snitzer44feb382012-10-12 21:02:10 +01001745 dm_bio_prison_destroy(pool->prison);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001746 dm_kcopyd_client_destroy(pool->copier);
1747
1748 if (pool->wq)
1749 destroy_workqueue(pool->wq);
1750
1751 if (pool->next_mapping)
1752 mempool_free(pool->next_mapping, pool->mapping_pool);
1753 mempool_destroy(pool->mapping_pool);
Mike Snitzer44feb382012-10-12 21:02:10 +01001754 dm_deferred_set_destroy(pool->shared_read_ds);
1755 dm_deferred_set_destroy(pool->all_io_ds);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001756 kfree(pool);
1757}
1758
Mike Snitzera24c2562012-06-03 00:30:00 +01001759static struct kmem_cache *_new_mapping_cache;
Mike Snitzera24c2562012-06-03 00:30:00 +01001760
Joe Thornber991d9fa2011-10-31 20:21:18 +00001761static struct pool *pool_create(struct mapped_device *pool_md,
1762 struct block_device *metadata_dev,
Joe Thornbere49e5822012-07-27 15:08:16 +01001763 unsigned long block_size,
1764 int read_only, char **error)
Joe Thornber991d9fa2011-10-31 20:21:18 +00001765{
1766 int r;
1767 void *err_p;
1768 struct pool *pool;
1769 struct dm_pool_metadata *pmd;
Joe Thornbere49e5822012-07-27 15:08:16 +01001770 bool format_device = read_only ? false : true;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001771
Joe Thornbere49e5822012-07-27 15:08:16 +01001772 pmd = dm_pool_metadata_open(metadata_dev, block_size, format_device);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001773 if (IS_ERR(pmd)) {
1774 *error = "Error creating metadata object";
1775 return (struct pool *)pmd;
1776 }
1777
1778 pool = kmalloc(sizeof(*pool), GFP_KERNEL);
1779 if (!pool) {
1780 *error = "Error allocating memory for pool";
1781 err_p = ERR_PTR(-ENOMEM);
1782 goto bad_pool;
1783 }
1784
1785 pool->pmd = pmd;
1786 pool->sectors_per_block = block_size;
Mikulas Patockaf9a8e0c2012-07-27 15:08:03 +01001787 if (block_size & (block_size - 1))
1788 pool->sectors_per_block_shift = -1;
1789 else
1790 pool->sectors_per_block_shift = __ffs(block_size);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001791 pool->low_water_blocks = 0;
Joe Thornber67e2e2b2012-03-28 18:41:29 +01001792 pool_features_init(&pool->pf);
Mike Snitzer44feb382012-10-12 21:02:10 +01001793 pool->prison = dm_bio_prison_create(PRISON_CELLS);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001794 if (!pool->prison) {
1795 *error = "Error creating pool's bio prison";
1796 err_p = ERR_PTR(-ENOMEM);
1797 goto bad_prison;
1798 }
1799
Mikulas Patockadf5d2e92013-03-01 22:45:49 +00001800 pool->copier = dm_kcopyd_client_create(&dm_kcopyd_throttle);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001801 if (IS_ERR(pool->copier)) {
1802 r = PTR_ERR(pool->copier);
1803 *error = "Error creating pool's kcopyd client";
1804 err_p = ERR_PTR(r);
1805 goto bad_kcopyd_client;
1806 }
1807
1808 /*
1809 * Create singlethreaded workqueue that will service all devices
1810 * that use this metadata.
1811 */
1812 pool->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM);
1813 if (!pool->wq) {
1814 *error = "Error creating pool's workqueue";
1815 err_p = ERR_PTR(-ENOMEM);
1816 goto bad_wq;
1817 }
1818
1819 INIT_WORK(&pool->worker, do_worker);
Joe Thornber905e51b2012-03-28 18:41:27 +01001820 INIT_DELAYED_WORK(&pool->waker, do_waker);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001821 spin_lock_init(&pool->lock);
1822 bio_list_init(&pool->deferred_bios);
1823 bio_list_init(&pool->deferred_flush_bios);
1824 INIT_LIST_HEAD(&pool->prepared_mappings);
Joe Thornber104655f2012-03-28 18:41:28 +01001825 INIT_LIST_HEAD(&pool->prepared_discards);
Joe Thornber88a66212013-12-04 20:16:12 -05001826 pool->low_water_triggered = false;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001827 bio_list_init(&pool->retry_on_resume_list);
Mike Snitzer44feb382012-10-12 21:02:10 +01001828
1829 pool->shared_read_ds = dm_deferred_set_create();
1830 if (!pool->shared_read_ds) {
1831 *error = "Error creating pool's shared read deferred set";
1832 err_p = ERR_PTR(-ENOMEM);
1833 goto bad_shared_read_ds;
1834 }
1835
1836 pool->all_io_ds = dm_deferred_set_create();
1837 if (!pool->all_io_ds) {
1838 *error = "Error creating pool's all io deferred set";
1839 err_p = ERR_PTR(-ENOMEM);
1840 goto bad_all_io_ds;
1841 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00001842
1843 pool->next_mapping = NULL;
Mike Snitzera24c2562012-06-03 00:30:00 +01001844 pool->mapping_pool = mempool_create_slab_pool(MAPPING_POOL_SIZE,
1845 _new_mapping_cache);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001846 if (!pool->mapping_pool) {
1847 *error = "Error creating pool's mapping mempool";
1848 err_p = ERR_PTR(-ENOMEM);
1849 goto bad_mapping_pool;
1850 }
1851
Joe Thornber991d9fa2011-10-31 20:21:18 +00001852 pool->ref_count = 1;
Joe Thornber905e51b2012-03-28 18:41:27 +01001853 pool->last_commit_jiffies = jiffies;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001854 pool->pool_md = pool_md;
1855 pool->md_dev = metadata_dev;
1856 __pool_table_insert(pool);
1857
1858 return pool;
1859
Joe Thornber991d9fa2011-10-31 20:21:18 +00001860bad_mapping_pool:
Mike Snitzer44feb382012-10-12 21:02:10 +01001861 dm_deferred_set_destroy(pool->all_io_ds);
1862bad_all_io_ds:
1863 dm_deferred_set_destroy(pool->shared_read_ds);
1864bad_shared_read_ds:
Joe Thornber991d9fa2011-10-31 20:21:18 +00001865 destroy_workqueue(pool->wq);
1866bad_wq:
1867 dm_kcopyd_client_destroy(pool->copier);
1868bad_kcopyd_client:
Mike Snitzer44feb382012-10-12 21:02:10 +01001869 dm_bio_prison_destroy(pool->prison);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001870bad_prison:
1871 kfree(pool);
1872bad_pool:
1873 if (dm_pool_metadata_close(pmd))
1874 DMWARN("%s: dm_pool_metadata_close() failed.", __func__);
1875
1876 return err_p;
1877}
1878
1879static void __pool_inc(struct pool *pool)
1880{
1881 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
1882 pool->ref_count++;
1883}
1884
1885static void __pool_dec(struct pool *pool)
1886{
1887 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
1888 BUG_ON(!pool->ref_count);
1889 if (!--pool->ref_count)
1890 __pool_destroy(pool);
1891}
1892
1893static struct pool *__pool_find(struct mapped_device *pool_md,
1894 struct block_device *metadata_dev,
Joe Thornbere49e5822012-07-27 15:08:16 +01001895 unsigned long block_size, int read_only,
1896 char **error, int *created)
Joe Thornber991d9fa2011-10-31 20:21:18 +00001897{
1898 struct pool *pool = __pool_table_lookup_metadata_dev(metadata_dev);
1899
1900 if (pool) {
Mike Snitzerf09996c2012-07-27 15:07:59 +01001901 if (pool->pool_md != pool_md) {
1902 *error = "metadata device already in use by a pool";
Joe Thornber991d9fa2011-10-31 20:21:18 +00001903 return ERR_PTR(-EBUSY);
Mike Snitzerf09996c2012-07-27 15:07:59 +01001904 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00001905 __pool_inc(pool);
1906
1907 } else {
1908 pool = __pool_table_lookup(pool_md);
1909 if (pool) {
Mike Snitzerf09996c2012-07-27 15:07:59 +01001910 if (pool->md_dev != metadata_dev) {
1911 *error = "different pool cannot replace a pool";
Joe Thornber991d9fa2011-10-31 20:21:18 +00001912 return ERR_PTR(-EINVAL);
Mike Snitzerf09996c2012-07-27 15:07:59 +01001913 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00001914 __pool_inc(pool);
1915
Joe Thornber67e2e2b2012-03-28 18:41:29 +01001916 } else {
Joe Thornbere49e5822012-07-27 15:08:16 +01001917 pool = pool_create(pool_md, metadata_dev, block_size, read_only, error);
Joe Thornber67e2e2b2012-03-28 18:41:29 +01001918 *created = 1;
1919 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00001920 }
1921
1922 return pool;
1923}
1924
1925/*----------------------------------------------------------------
1926 * Pool target methods
1927 *--------------------------------------------------------------*/
1928static void pool_dtr(struct dm_target *ti)
1929{
1930 struct pool_c *pt = ti->private;
1931
1932 mutex_lock(&dm_thin_pool_table.mutex);
1933
1934 unbind_control_target(pt->pool, ti);
1935 __pool_dec(pt->pool);
1936 dm_put_device(ti, pt->metadata_dev);
1937 dm_put_device(ti, pt->data_dev);
1938 kfree(pt);
1939
1940 mutex_unlock(&dm_thin_pool_table.mutex);
1941}
1942
Joe Thornber991d9fa2011-10-31 20:21:18 +00001943static int parse_pool_features(struct dm_arg_set *as, struct pool_features *pf,
1944 struct dm_target *ti)
1945{
1946 int r;
1947 unsigned argc;
1948 const char *arg_name;
1949
1950 static struct dm_arg _args[] = {
Mike Snitzer74aa45c2014-01-15 19:07:58 -05001951 {0, 4, "Invalid number of pool feature arguments"},
Joe Thornber991d9fa2011-10-31 20:21:18 +00001952 };
1953
1954 /*
1955 * No feature arguments supplied.
1956 */
1957 if (!as->argc)
1958 return 0;
1959
1960 r = dm_read_arg_group(_args, as, &argc, &ti->error);
1961 if (r)
1962 return -EINVAL;
1963
1964 while (argc && !r) {
1965 arg_name = dm_shift_arg(as);
1966 argc--;
1967
Joe Thornbere49e5822012-07-27 15:08:16 +01001968 if (!strcasecmp(arg_name, "skip_block_zeroing"))
Mike Snitzer9bc142d2012-09-26 23:45:46 +01001969 pf->zero_new_blocks = false;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001970
Joe Thornbere49e5822012-07-27 15:08:16 +01001971 else if (!strcasecmp(arg_name, "ignore_discard"))
Mike Snitzer9bc142d2012-09-26 23:45:46 +01001972 pf->discard_enabled = false;
Joe Thornbere49e5822012-07-27 15:08:16 +01001973
1974 else if (!strcasecmp(arg_name, "no_discard_passdown"))
Mike Snitzer9bc142d2012-09-26 23:45:46 +01001975 pf->discard_passdown = false;
Joe Thornbere49e5822012-07-27 15:08:16 +01001976
1977 else if (!strcasecmp(arg_name, "read_only"))
1978 pf->mode = PM_READ_ONLY;
1979
Mike Snitzer787a996c2013-12-06 16:21:43 -05001980 else if (!strcasecmp(arg_name, "error_if_no_space"))
1981 pf->error_if_no_space = true;
1982
Joe Thornbere49e5822012-07-27 15:08:16 +01001983 else {
1984 ti->error = "Unrecognised pool feature requested";
1985 r = -EINVAL;
1986 break;
1987 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00001988 }
1989
1990 return r;
1991}
1992
Joe Thornberac8c3f32013-05-10 14:37:21 +01001993static void metadata_low_callback(void *context)
1994{
1995 struct pool *pool = context;
1996
1997 DMWARN("%s: reached low water mark for metadata device: sending event.",
1998 dm_device_name(pool->pool_md));
1999
2000 dm_table_event(pool->ti->table);
2001}
2002
Mike Snitzer7d489352014-02-12 23:58:15 -05002003static sector_t get_dev_size(struct block_device *bdev)
Joe Thornberb17446d2013-05-10 14:37:18 +01002004{
Mike Snitzer7d489352014-02-12 23:58:15 -05002005 return i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
2006}
2007
2008static void warn_if_metadata_device_too_big(struct block_device *bdev)
2009{
2010 sector_t metadata_dev_size = get_dev_size(bdev);
Joe Thornberb17446d2013-05-10 14:37:18 +01002011 char buffer[BDEVNAME_SIZE];
2012
Mike Snitzer7d489352014-02-12 23:58:15 -05002013 if (metadata_dev_size > THIN_METADATA_MAX_SECTORS_WARNING)
Joe Thornberb17446d2013-05-10 14:37:18 +01002014 DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.",
2015 bdevname(bdev, buffer), THIN_METADATA_MAX_SECTORS);
Mike Snitzer7d489352014-02-12 23:58:15 -05002016}
2017
2018static sector_t get_metadata_dev_size(struct block_device *bdev)
2019{
2020 sector_t metadata_dev_size = get_dev_size(bdev);
2021
2022 if (metadata_dev_size > THIN_METADATA_MAX_SECTORS)
2023 metadata_dev_size = THIN_METADATA_MAX_SECTORS;
Joe Thornberb17446d2013-05-10 14:37:18 +01002024
2025 return metadata_dev_size;
2026}
2027
Joe Thornber24347e92013-05-10 14:37:19 +01002028static dm_block_t get_metadata_dev_size_in_blocks(struct block_device *bdev)
2029{
2030 sector_t metadata_dev_size = get_metadata_dev_size(bdev);
2031
Mike Snitzer7d489352014-02-12 23:58:15 -05002032 sector_div(metadata_dev_size, THIN_METADATA_BLOCK_SIZE);
Joe Thornber24347e92013-05-10 14:37:19 +01002033
2034 return metadata_dev_size;
2035}
2036
Joe Thornber991d9fa2011-10-31 20:21:18 +00002037/*
Joe Thornberac8c3f32013-05-10 14:37:21 +01002038 * When a metadata threshold is crossed a dm event is triggered, and
2039 * userland should respond by growing the metadata device. We could let
2040 * userland set the threshold, like we do with the data threshold, but I'm
2041 * not sure they know enough to do this well.
2042 */
2043static dm_block_t calc_metadata_threshold(struct pool_c *pt)
2044{
2045 /*
2046 * 4M is ample for all ops with the possible exception of thin
2047 * device deletion which is harmless if it fails (just retry the
2048 * delete after you've grown the device).
2049 */
2050 dm_block_t quarter = get_metadata_dev_size_in_blocks(pt->metadata_dev->bdev) / 4;
2051 return min((dm_block_t)1024ULL /* 4M */, quarter);
2052}
2053
2054/*
Joe Thornber991d9fa2011-10-31 20:21:18 +00002055 * thin-pool <metadata dev> <data dev>
2056 * <data block size (sectors)>
2057 * <low water mark (blocks)>
2058 * [<#feature args> [<arg>]*]
2059 *
2060 * Optional feature arguments are:
2061 * skip_block_zeroing: skips the zeroing of newly-provisioned blocks.
Joe Thornber67e2e2b2012-03-28 18:41:29 +01002062 * ignore_discard: disable discard
2063 * no_discard_passdown: don't pass discards down to the data device
Mike Snitzer787a996c2013-12-06 16:21:43 -05002064 * read_only: Don't allow any changes to be made to the pool metadata.
2065 * error_if_no_space: error IOs, instead of queueing, if no space.
Joe Thornber991d9fa2011-10-31 20:21:18 +00002066 */
2067static int pool_ctr(struct dm_target *ti, unsigned argc, char **argv)
2068{
Joe Thornber67e2e2b2012-03-28 18:41:29 +01002069 int r, pool_created = 0;
Joe Thornber991d9fa2011-10-31 20:21:18 +00002070 struct pool_c *pt;
2071 struct pool *pool;
2072 struct pool_features pf;
2073 struct dm_arg_set as;
2074 struct dm_dev *data_dev;
2075 unsigned long block_size;
2076 dm_block_t low_water_blocks;
2077 struct dm_dev *metadata_dev;
Joe Thornber5d0db962013-05-10 14:37:19 +01002078 fmode_t metadata_mode;
Joe Thornber991d9fa2011-10-31 20:21:18 +00002079
2080 /*
2081 * FIXME Remove validation from scope of lock.
2082 */
2083 mutex_lock(&dm_thin_pool_table.mutex);
2084
2085 if (argc < 4) {
2086 ti->error = "Invalid argument count";
2087 r = -EINVAL;
2088 goto out_unlock;
2089 }
Joe Thornber5d0db962013-05-10 14:37:19 +01002090
Joe Thornber991d9fa2011-10-31 20:21:18 +00002091 as.argc = argc;
2092 as.argv = argv;
2093
Joe Thornber5d0db962013-05-10 14:37:19 +01002094 /*
2095 * Set default pool features.
2096 */
2097 pool_features_init(&pf);
2098
2099 dm_consume_args(&as, 4);
2100 r = parse_pool_features(&as, &pf, ti);
2101 if (r)
2102 goto out_unlock;
2103
2104 metadata_mode = FMODE_READ | ((pf.mode == PM_READ_ONLY) ? 0 : FMODE_WRITE);
2105 r = dm_get_device(ti, argv[0], metadata_mode, &metadata_dev);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002106 if (r) {
2107 ti->error = "Error opening metadata block device";
2108 goto out_unlock;
2109 }
Mike Snitzer7d489352014-02-12 23:58:15 -05002110 warn_if_metadata_device_too_big(metadata_dev->bdev);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002111
2112 r = dm_get_device(ti, argv[1], FMODE_READ | FMODE_WRITE, &data_dev);
2113 if (r) {
2114 ti->error = "Error getting data device";
2115 goto out_metadata;
2116 }
2117
2118 if (kstrtoul(argv[2], 10, &block_size) || !block_size ||
2119 block_size < DATA_DEV_BLOCK_SIZE_MIN_SECTORS ||
2120 block_size > DATA_DEV_BLOCK_SIZE_MAX_SECTORS ||
Mike Snitzer55f2b8b2012-07-27 15:08:02 +01002121 block_size & (DATA_DEV_BLOCK_SIZE_MIN_SECTORS - 1)) {
Joe Thornber991d9fa2011-10-31 20:21:18 +00002122 ti->error = "Invalid block size";
2123 r = -EINVAL;
2124 goto out;
2125 }
2126
2127 if (kstrtoull(argv[3], 10, (unsigned long long *)&low_water_blocks)) {
2128 ti->error = "Invalid low water mark";
2129 r = -EINVAL;
2130 goto out;
2131 }
2132
Joe Thornber991d9fa2011-10-31 20:21:18 +00002133 pt = kzalloc(sizeof(*pt), GFP_KERNEL);
2134 if (!pt) {
2135 r = -ENOMEM;
2136 goto out;
2137 }
2138
2139 pool = __pool_find(dm_table_get_md(ti->table), metadata_dev->bdev,
Joe Thornbere49e5822012-07-27 15:08:16 +01002140 block_size, pf.mode == PM_READ_ONLY, &ti->error, &pool_created);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002141 if (IS_ERR(pool)) {
2142 r = PTR_ERR(pool);
2143 goto out_free_pt;
2144 }
2145
Joe Thornber67e2e2b2012-03-28 18:41:29 +01002146 /*
2147 * 'pool_created' reflects whether this is the first table load.
2148 * Top level discard support is not allowed to be changed after
2149 * initial load. This would require a pool reload to trigger thin
2150 * device changes.
2151 */
2152 if (!pool_created && pf.discard_enabled != pool->pf.discard_enabled) {
2153 ti->error = "Discard support cannot be disabled once enabled";
2154 r = -EINVAL;
2155 goto out_flags_changed;
2156 }
2157
Joe Thornber991d9fa2011-10-31 20:21:18 +00002158 pt->pool = pool;
2159 pt->ti = ti;
2160 pt->metadata_dev = metadata_dev;
2161 pt->data_dev = data_dev;
2162 pt->low_water_blocks = low_water_blocks;
Mike Snitzer0424caa2012-09-26 23:45:47 +01002163 pt->adjusted_pf = pt->requested_pf = pf;
Alasdair G Kergon55a62ee2013-03-01 22:45:47 +00002164 ti->num_flush_bios = 1;
Mike Snitzer9bc142d2012-09-26 23:45:46 +01002165
Joe Thornber67e2e2b2012-03-28 18:41:29 +01002166 /*
2167 * Only need to enable discards if the pool should pass
2168 * them down to the data device. The thin device's discard
2169 * processing will cause mappings to be removed from the btree.
2170 */
Mike Snitzerb60ab992013-09-19 18:49:11 -04002171 ti->discard_zeroes_data_unsupported = true;
Joe Thornber67e2e2b2012-03-28 18:41:29 +01002172 if (pf.discard_enabled && pf.discard_passdown) {
Alasdair G Kergon55a62ee2013-03-01 22:45:47 +00002173 ti->num_discard_bios = 1;
Mike Snitzer9bc142d2012-09-26 23:45:46 +01002174
Joe Thornber67e2e2b2012-03-28 18:41:29 +01002175 /*
2176 * Setting 'discards_supported' circumvents the normal
2177 * stacking of discard limits (this keeps the pool and
2178 * thin devices' discard limits consistent).
2179 */
Alasdair G Kergon0ac55482012-07-27 15:08:08 +01002180 ti->discards_supported = true;
Joe Thornber67e2e2b2012-03-28 18:41:29 +01002181 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00002182 ti->private = pt;
2183
Joe Thornberac8c3f32013-05-10 14:37:21 +01002184 r = dm_pool_register_metadata_threshold(pt->pool->pmd,
2185 calc_metadata_threshold(pt),
2186 metadata_low_callback,
2187 pool);
2188 if (r)
2189 goto out_free_pt;
2190
Joe Thornber991d9fa2011-10-31 20:21:18 +00002191 pt->callbacks.congested_fn = pool_is_congested;
2192 dm_table_add_target_callbacks(ti->table, &pt->callbacks);
2193
2194 mutex_unlock(&dm_thin_pool_table.mutex);
2195
2196 return 0;
2197
Joe Thornber67e2e2b2012-03-28 18:41:29 +01002198out_flags_changed:
2199 __pool_dec(pool);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002200out_free_pt:
2201 kfree(pt);
2202out:
2203 dm_put_device(ti, data_dev);
2204out_metadata:
2205 dm_put_device(ti, metadata_dev);
2206out_unlock:
2207 mutex_unlock(&dm_thin_pool_table.mutex);
2208
2209 return r;
2210}
2211
Mikulas Patocka7de3ee52012-12-21 20:23:41 +00002212static int pool_map(struct dm_target *ti, struct bio *bio)
Joe Thornber991d9fa2011-10-31 20:21:18 +00002213{
2214 int r;
2215 struct pool_c *pt = ti->private;
2216 struct pool *pool = pt->pool;
2217 unsigned long flags;
2218
2219 /*
2220 * As this is a singleton target, ti->begin is always zero.
2221 */
2222 spin_lock_irqsave(&pool->lock, flags);
2223 bio->bi_bdev = pt->data_dev->bdev;
2224 r = DM_MAPIO_REMAPPED;
2225 spin_unlock_irqrestore(&pool->lock, flags);
2226
2227 return r;
2228}
2229
Joe Thornberb17446d2013-05-10 14:37:18 +01002230static int maybe_resize_data_dev(struct dm_target *ti, bool *need_commit)
2231{
2232 int r;
2233 struct pool_c *pt = ti->private;
2234 struct pool *pool = pt->pool;
2235 sector_t data_size = ti->len;
2236 dm_block_t sb_data_size;
2237
2238 *need_commit = false;
2239
2240 (void) sector_div(data_size, pool->sectors_per_block);
2241
2242 r = dm_pool_get_data_dev_size(pool->pmd, &sb_data_size);
2243 if (r) {
Mike Snitzer4fa59712013-08-21 17:30:40 -04002244 DMERR("%s: failed to retrieve data device size",
2245 dm_device_name(pool->pool_md));
Joe Thornberb17446d2013-05-10 14:37:18 +01002246 return r;
2247 }
2248
2249 if (data_size < sb_data_size) {
Mike Snitzer4fa59712013-08-21 17:30:40 -04002250 DMERR("%s: pool target (%llu blocks) too small: expected %llu",
2251 dm_device_name(pool->pool_md),
Joe Thornberb17446d2013-05-10 14:37:18 +01002252 (unsigned long long)data_size, sb_data_size);
2253 return -EINVAL;
2254
2255 } else if (data_size > sb_data_size) {
Mike Snitzer6f7f51d2013-12-04 10:25:53 -05002256 if (sb_data_size)
2257 DMINFO("%s: growing the data device from %llu to %llu blocks",
2258 dm_device_name(pool->pool_md),
2259 sb_data_size, (unsigned long long)data_size);
Joe Thornberb17446d2013-05-10 14:37:18 +01002260 r = dm_pool_resize_data_dev(pool->pmd, data_size);
2261 if (r) {
Joe Thornberb5330652013-12-04 19:51:33 -05002262 metadata_operation_failed(pool, "dm_pool_resize_data_dev", r);
Joe Thornberb17446d2013-05-10 14:37:18 +01002263 return r;
2264 }
2265
2266 *need_commit = true;
2267 }
2268
2269 return 0;
2270}
2271
Joe Thornber24347e92013-05-10 14:37:19 +01002272static int maybe_resize_metadata_dev(struct dm_target *ti, bool *need_commit)
2273{
2274 int r;
2275 struct pool_c *pt = ti->private;
2276 struct pool *pool = pt->pool;
2277 dm_block_t metadata_dev_size, sb_metadata_dev_size;
2278
2279 *need_commit = false;
2280
Alasdair G Kergon610bba82013-05-19 18:57:50 +01002281 metadata_dev_size = get_metadata_dev_size_in_blocks(pool->md_dev);
Joe Thornber24347e92013-05-10 14:37:19 +01002282
2283 r = dm_pool_get_metadata_dev_size(pool->pmd, &sb_metadata_dev_size);
2284 if (r) {
Mike Snitzer4fa59712013-08-21 17:30:40 -04002285 DMERR("%s: failed to retrieve metadata device size",
2286 dm_device_name(pool->pool_md));
Joe Thornber24347e92013-05-10 14:37:19 +01002287 return r;
2288 }
2289
2290 if (metadata_dev_size < sb_metadata_dev_size) {
Mike Snitzer4fa59712013-08-21 17:30:40 -04002291 DMERR("%s: metadata device (%llu blocks) too small: expected %llu",
2292 dm_device_name(pool->pool_md),
Joe Thornber24347e92013-05-10 14:37:19 +01002293 metadata_dev_size, sb_metadata_dev_size);
2294 return -EINVAL;
2295
2296 } else if (metadata_dev_size > sb_metadata_dev_size) {
Mike Snitzer7d489352014-02-12 23:58:15 -05002297 warn_if_metadata_device_too_big(pool->md_dev);
Mike Snitzer6f7f51d2013-12-04 10:25:53 -05002298 DMINFO("%s: growing the metadata device from %llu to %llu blocks",
2299 dm_device_name(pool->pool_md),
2300 sb_metadata_dev_size, metadata_dev_size);
Joe Thornber24347e92013-05-10 14:37:19 +01002301 r = dm_pool_resize_metadata_dev(pool->pmd, metadata_dev_size);
2302 if (r) {
Joe Thornberb5330652013-12-04 19:51:33 -05002303 metadata_operation_failed(pool, "dm_pool_resize_metadata_dev", r);
Joe Thornber24347e92013-05-10 14:37:19 +01002304 return r;
2305 }
2306
2307 *need_commit = true;
2308 }
2309
2310 return 0;
2311}
2312
Joe Thornber991d9fa2011-10-31 20:21:18 +00002313/*
2314 * Retrieves the number of blocks of the data device from
2315 * the superblock and compares it to the actual device size,
2316 * thus resizing the data device in case it has grown.
2317 *
2318 * This both copes with opening preallocated data devices in the ctr
2319 * being followed by a resume
2320 * -and-
2321 * calling the resume method individually after userspace has
2322 * grown the data device in reaction to a table event.
2323 */
2324static int pool_preresume(struct dm_target *ti)
2325{
2326 int r;
Joe Thornber24347e92013-05-10 14:37:19 +01002327 bool need_commit1, need_commit2;
Joe Thornber991d9fa2011-10-31 20:21:18 +00002328 struct pool_c *pt = ti->private;
2329 struct pool *pool = pt->pool;
Joe Thornber991d9fa2011-10-31 20:21:18 +00002330
2331 /*
2332 * Take control of the pool object.
2333 */
2334 r = bind_control_target(pool, ti);
2335 if (r)
2336 return r;
2337
Joe Thornberb17446d2013-05-10 14:37:18 +01002338 r = maybe_resize_data_dev(ti, &need_commit1);
2339 if (r)
Joe Thornber991d9fa2011-10-31 20:21:18 +00002340 return r;
Joe Thornber991d9fa2011-10-31 20:21:18 +00002341
Joe Thornber24347e92013-05-10 14:37:19 +01002342 r = maybe_resize_metadata_dev(ti, &need_commit2);
2343 if (r)
2344 return r;
2345
2346 if (need_commit1 || need_commit2)
Joe Thornber020cc3b2013-12-04 15:05:36 -05002347 (void) commit(pool);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002348
2349 return 0;
2350}
2351
2352static void pool_resume(struct dm_target *ti)
2353{
2354 struct pool_c *pt = ti->private;
2355 struct pool *pool = pt->pool;
2356 unsigned long flags;
2357
2358 spin_lock_irqsave(&pool->lock, flags);
Joe Thornber88a66212013-12-04 20:16:12 -05002359 pool->low_water_triggered = false;
Joe Thornber991d9fa2011-10-31 20:21:18 +00002360 __requeue_bios(pool);
2361 spin_unlock_irqrestore(&pool->lock, flags);
2362
Joe Thornber905e51b2012-03-28 18:41:27 +01002363 do_waker(&pool->waker.work);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002364}
2365
2366static void pool_postsuspend(struct dm_target *ti)
2367{
Joe Thornber991d9fa2011-10-31 20:21:18 +00002368 struct pool_c *pt = ti->private;
2369 struct pool *pool = pt->pool;
2370
Joe Thornber905e51b2012-03-28 18:41:27 +01002371 cancel_delayed_work(&pool->waker);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002372 flush_workqueue(pool->wq);
Joe Thornber020cc3b2013-12-04 15:05:36 -05002373 (void) commit(pool);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002374}
2375
2376static int check_arg_count(unsigned argc, unsigned args_required)
2377{
2378 if (argc != args_required) {
2379 DMWARN("Message received with %u arguments instead of %u.",
2380 argc, args_required);
2381 return -EINVAL;
2382 }
2383
2384 return 0;
2385}
2386
2387static int read_dev_id(char *arg, dm_thin_id *dev_id, int warning)
2388{
2389 if (!kstrtoull(arg, 10, (unsigned long long *)dev_id) &&
2390 *dev_id <= MAX_DEV_ID)
2391 return 0;
2392
2393 if (warning)
2394 DMWARN("Message received with invalid device id: %s", arg);
2395
2396 return -EINVAL;
2397}
2398
2399static int process_create_thin_mesg(unsigned argc, char **argv, struct pool *pool)
2400{
2401 dm_thin_id dev_id;
2402 int r;
2403
2404 r = check_arg_count(argc, 2);
2405 if (r)
2406 return r;
2407
2408 r = read_dev_id(argv[1], &dev_id, 1);
2409 if (r)
2410 return r;
2411
2412 r = dm_pool_create_thin(pool->pmd, dev_id);
2413 if (r) {
2414 DMWARN("Creation of new thinly-provisioned device with id %s failed.",
2415 argv[1]);
2416 return r;
2417 }
2418
2419 return 0;
2420}
2421
2422static int process_create_snap_mesg(unsigned argc, char **argv, struct pool *pool)
2423{
2424 dm_thin_id dev_id;
2425 dm_thin_id origin_dev_id;
2426 int r;
2427
2428 r = check_arg_count(argc, 3);
2429 if (r)
2430 return r;
2431
2432 r = read_dev_id(argv[1], &dev_id, 1);
2433 if (r)
2434 return r;
2435
2436 r = read_dev_id(argv[2], &origin_dev_id, 1);
2437 if (r)
2438 return r;
2439
2440 r = dm_pool_create_snap(pool->pmd, dev_id, origin_dev_id);
2441 if (r) {
2442 DMWARN("Creation of new snapshot %s of device %s failed.",
2443 argv[1], argv[2]);
2444 return r;
2445 }
2446
2447 return 0;
2448}
2449
2450static int process_delete_mesg(unsigned argc, char **argv, struct pool *pool)
2451{
2452 dm_thin_id dev_id;
2453 int r;
2454
2455 r = check_arg_count(argc, 2);
2456 if (r)
2457 return r;
2458
2459 r = read_dev_id(argv[1], &dev_id, 1);
2460 if (r)
2461 return r;
2462
2463 r = dm_pool_delete_thin_device(pool->pmd, dev_id);
2464 if (r)
2465 DMWARN("Deletion of thin device %s failed.", argv[1]);
2466
2467 return r;
2468}
2469
2470static int process_set_transaction_id_mesg(unsigned argc, char **argv, struct pool *pool)
2471{
2472 dm_thin_id old_id, new_id;
2473 int r;
2474
2475 r = check_arg_count(argc, 3);
2476 if (r)
2477 return r;
2478
2479 if (kstrtoull(argv[1], 10, (unsigned long long *)&old_id)) {
2480 DMWARN("set_transaction_id message: Unrecognised id %s.", argv[1]);
2481 return -EINVAL;
2482 }
2483
2484 if (kstrtoull(argv[2], 10, (unsigned long long *)&new_id)) {
2485 DMWARN("set_transaction_id message: Unrecognised new id %s.", argv[2]);
2486 return -EINVAL;
2487 }
2488
2489 r = dm_pool_set_metadata_transaction_id(pool->pmd, old_id, new_id);
2490 if (r) {
2491 DMWARN("Failed to change transaction id from %s to %s.",
2492 argv[1], argv[2]);
2493 return r;
2494 }
2495
2496 return 0;
2497}
2498
Joe Thornbercc8394d2012-06-03 00:30:01 +01002499static int process_reserve_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool)
2500{
2501 int r;
2502
2503 r = check_arg_count(argc, 1);
2504 if (r)
2505 return r;
2506
Joe Thornber020cc3b2013-12-04 15:05:36 -05002507 (void) commit(pool);
Joe Thornber0d200ae2012-07-03 12:55:31 +01002508
Joe Thornbercc8394d2012-06-03 00:30:01 +01002509 r = dm_pool_reserve_metadata_snap(pool->pmd);
2510 if (r)
2511 DMWARN("reserve_metadata_snap message failed.");
2512
2513 return r;
2514}
2515
2516static int process_release_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool)
2517{
2518 int r;
2519
2520 r = check_arg_count(argc, 1);
2521 if (r)
2522 return r;
2523
2524 r = dm_pool_release_metadata_snap(pool->pmd);
2525 if (r)
2526 DMWARN("release_metadata_snap message failed.");
2527
2528 return r;
2529}
2530
Joe Thornber991d9fa2011-10-31 20:21:18 +00002531/*
2532 * Messages supported:
2533 * create_thin <dev_id>
2534 * create_snap <dev_id> <origin_id>
2535 * delete <dev_id>
2536 * trim <dev_id> <new_size_in_sectors>
2537 * set_transaction_id <current_trans_id> <new_trans_id>
Joe Thornbercc8394d2012-06-03 00:30:01 +01002538 * reserve_metadata_snap
2539 * release_metadata_snap
Joe Thornber991d9fa2011-10-31 20:21:18 +00002540 */
2541static int pool_message(struct dm_target *ti, unsigned argc, char **argv)
2542{
2543 int r = -EINVAL;
2544 struct pool_c *pt = ti->private;
2545 struct pool *pool = pt->pool;
2546
2547 if (!strcasecmp(argv[0], "create_thin"))
2548 r = process_create_thin_mesg(argc, argv, pool);
2549
2550 else if (!strcasecmp(argv[0], "create_snap"))
2551 r = process_create_snap_mesg(argc, argv, pool);
2552
2553 else if (!strcasecmp(argv[0], "delete"))
2554 r = process_delete_mesg(argc, argv, pool);
2555
2556 else if (!strcasecmp(argv[0], "set_transaction_id"))
2557 r = process_set_transaction_id_mesg(argc, argv, pool);
2558
Joe Thornbercc8394d2012-06-03 00:30:01 +01002559 else if (!strcasecmp(argv[0], "reserve_metadata_snap"))
2560 r = process_reserve_metadata_snap_mesg(argc, argv, pool);
2561
2562 else if (!strcasecmp(argv[0], "release_metadata_snap"))
2563 r = process_release_metadata_snap_mesg(argc, argv, pool);
2564
Joe Thornber991d9fa2011-10-31 20:21:18 +00002565 else
2566 DMWARN("Unrecognised thin pool target message received: %s", argv[0]);
2567
Joe Thornbere49e5822012-07-27 15:08:16 +01002568 if (!r)
Joe Thornber020cc3b2013-12-04 15:05:36 -05002569 (void) commit(pool);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002570
2571 return r;
2572}
2573
Joe Thornbere49e5822012-07-27 15:08:16 +01002574static void emit_flags(struct pool_features *pf, char *result,
2575 unsigned sz, unsigned maxlen)
2576{
2577 unsigned count = !pf->zero_new_blocks + !pf->discard_enabled +
Mike Snitzer787a996c2013-12-06 16:21:43 -05002578 !pf->discard_passdown + (pf->mode == PM_READ_ONLY) +
2579 pf->error_if_no_space;
Joe Thornbere49e5822012-07-27 15:08:16 +01002580 DMEMIT("%u ", count);
2581
2582 if (!pf->zero_new_blocks)
2583 DMEMIT("skip_block_zeroing ");
2584
2585 if (!pf->discard_enabled)
2586 DMEMIT("ignore_discard ");
2587
2588 if (!pf->discard_passdown)
2589 DMEMIT("no_discard_passdown ");
2590
2591 if (pf->mode == PM_READ_ONLY)
2592 DMEMIT("read_only ");
Mike Snitzer787a996c2013-12-06 16:21:43 -05002593
2594 if (pf->error_if_no_space)
2595 DMEMIT("error_if_no_space ");
Joe Thornbere49e5822012-07-27 15:08:16 +01002596}
2597
Joe Thornber991d9fa2011-10-31 20:21:18 +00002598/*
2599 * Status line is:
2600 * <transaction id> <used metadata sectors>/<total metadata sectors>
2601 * <used data sectors>/<total data sectors> <held metadata root>
2602 */
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002603static void pool_status(struct dm_target *ti, status_type_t type,
2604 unsigned status_flags, char *result, unsigned maxlen)
Joe Thornber991d9fa2011-10-31 20:21:18 +00002605{
Joe Thornbere49e5822012-07-27 15:08:16 +01002606 int r;
Joe Thornber991d9fa2011-10-31 20:21:18 +00002607 unsigned sz = 0;
2608 uint64_t transaction_id;
2609 dm_block_t nr_free_blocks_data;
2610 dm_block_t nr_free_blocks_metadata;
2611 dm_block_t nr_blocks_data;
2612 dm_block_t nr_blocks_metadata;
2613 dm_block_t held_root;
2614 char buf[BDEVNAME_SIZE];
2615 char buf2[BDEVNAME_SIZE];
2616 struct pool_c *pt = ti->private;
2617 struct pool *pool = pt->pool;
2618
2619 switch (type) {
2620 case STATUSTYPE_INFO:
Joe Thornbere49e5822012-07-27 15:08:16 +01002621 if (get_pool_mode(pool) == PM_FAIL) {
2622 DMEMIT("Fail");
2623 break;
2624 }
2625
Alasdair G Kergon1f4e0ff2012-07-27 15:08:16 +01002626 /* Commit to ensure statistics aren't out-of-date */
2627 if (!(status_flags & DM_STATUS_NOFLUSH_FLAG) && !dm_suspended(ti))
Joe Thornber020cc3b2013-12-04 15:05:36 -05002628 (void) commit(pool);
Alasdair G Kergon1f4e0ff2012-07-27 15:08:16 +01002629
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002630 r = dm_pool_get_metadata_transaction_id(pool->pmd, &transaction_id);
2631 if (r) {
Mike Snitzer4fa59712013-08-21 17:30:40 -04002632 DMERR("%s: dm_pool_get_metadata_transaction_id returned %d",
2633 dm_device_name(pool->pool_md), r);
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002634 goto err;
2635 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00002636
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002637 r = dm_pool_get_free_metadata_block_count(pool->pmd, &nr_free_blocks_metadata);
2638 if (r) {
Mike Snitzer4fa59712013-08-21 17:30:40 -04002639 DMERR("%s: dm_pool_get_free_metadata_block_count returned %d",
2640 dm_device_name(pool->pool_md), r);
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002641 goto err;
2642 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00002643
2644 r = dm_pool_get_metadata_dev_size(pool->pmd, &nr_blocks_metadata);
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002645 if (r) {
Mike Snitzer4fa59712013-08-21 17:30:40 -04002646 DMERR("%s: dm_pool_get_metadata_dev_size returned %d",
2647 dm_device_name(pool->pool_md), r);
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002648 goto err;
2649 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00002650
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002651 r = dm_pool_get_free_block_count(pool->pmd, &nr_free_blocks_data);
2652 if (r) {
Mike Snitzer4fa59712013-08-21 17:30:40 -04002653 DMERR("%s: dm_pool_get_free_block_count returned %d",
2654 dm_device_name(pool->pool_md), r);
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002655 goto err;
2656 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00002657
2658 r = dm_pool_get_data_dev_size(pool->pmd, &nr_blocks_data);
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002659 if (r) {
Mike Snitzer4fa59712013-08-21 17:30:40 -04002660 DMERR("%s: dm_pool_get_data_dev_size returned %d",
2661 dm_device_name(pool->pool_md), r);
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002662 goto err;
2663 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00002664
Joe Thornbercc8394d2012-06-03 00:30:01 +01002665 r = dm_pool_get_metadata_snap(pool->pmd, &held_root);
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002666 if (r) {
Mike Snitzer4fa59712013-08-21 17:30:40 -04002667 DMERR("%s: dm_pool_get_metadata_snap returned %d",
2668 dm_device_name(pool->pool_md), r);
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002669 goto err;
2670 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00002671
2672 DMEMIT("%llu %llu/%llu %llu/%llu ",
2673 (unsigned long long)transaction_id,
2674 (unsigned long long)(nr_blocks_metadata - nr_free_blocks_metadata),
2675 (unsigned long long)nr_blocks_metadata,
2676 (unsigned long long)(nr_blocks_data - nr_free_blocks_data),
2677 (unsigned long long)nr_blocks_data);
2678
2679 if (held_root)
Joe Thornbere49e5822012-07-27 15:08:16 +01002680 DMEMIT("%llu ", held_root);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002681 else
Joe Thornbere49e5822012-07-27 15:08:16 +01002682 DMEMIT("- ");
2683
2684 if (pool->pf.mode == PM_READ_ONLY)
2685 DMEMIT("ro ");
2686 else
2687 DMEMIT("rw ");
2688
Mike Snitzer018debe2012-12-21 20:23:32 +00002689 if (!pool->pf.discard_enabled)
Mike Snitzer787a996c2013-12-06 16:21:43 -05002690 DMEMIT("ignore_discard ");
Mike Snitzer018debe2012-12-21 20:23:32 +00002691 else if (pool->pf.discard_passdown)
Mike Snitzer787a996c2013-12-06 16:21:43 -05002692 DMEMIT("discard_passdown ");
Joe Thornbere49e5822012-07-27 15:08:16 +01002693 else
Mike Snitzer787a996c2013-12-06 16:21:43 -05002694 DMEMIT("no_discard_passdown ");
2695
2696 if (pool->pf.error_if_no_space)
2697 DMEMIT("error_if_no_space ");
2698 else
2699 DMEMIT("queue_if_no_space ");
Joe Thornber991d9fa2011-10-31 20:21:18 +00002700
2701 break;
2702
2703 case STATUSTYPE_TABLE:
2704 DMEMIT("%s %s %lu %llu ",
2705 format_dev_t(buf, pt->metadata_dev->bdev->bd_dev),
2706 format_dev_t(buf2, pt->data_dev->bdev->bd_dev),
2707 (unsigned long)pool->sectors_per_block,
2708 (unsigned long long)pt->low_water_blocks);
Mike Snitzer0424caa2012-09-26 23:45:47 +01002709 emit_flags(&pt->requested_pf, result, sz, maxlen);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002710 break;
2711 }
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002712 return;
Joe Thornber991d9fa2011-10-31 20:21:18 +00002713
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002714err:
2715 DMEMIT("Error");
Joe Thornber991d9fa2011-10-31 20:21:18 +00002716}
2717
2718static int pool_iterate_devices(struct dm_target *ti,
2719 iterate_devices_callout_fn fn, void *data)
2720{
2721 struct pool_c *pt = ti->private;
2722
2723 return fn(ti, pt->data_dev, 0, ti->len, data);
2724}
2725
2726static int pool_merge(struct dm_target *ti, struct bvec_merge_data *bvm,
2727 struct bio_vec *biovec, int max_size)
2728{
2729 struct pool_c *pt = ti->private;
2730 struct request_queue *q = bdev_get_queue(pt->data_dev->bdev);
2731
2732 if (!q->merge_bvec_fn)
2733 return max_size;
2734
2735 bvm->bi_bdev = pt->data_dev->bdev;
2736
2737 return min(max_size, q->merge_bvec_fn(q, bvm, biovec));
2738}
2739
Mike Snitzer0424caa2012-09-26 23:45:47 +01002740static void set_discard_limits(struct pool_c *pt, struct queue_limits *limits)
Joe Thornber104655f2012-03-28 18:41:28 +01002741{
Mike Snitzer0424caa2012-09-26 23:45:47 +01002742 struct pool *pool = pt->pool;
2743 struct queue_limits *data_limits;
2744
Joe Thornber104655f2012-03-28 18:41:28 +01002745 limits->max_discard_sectors = pool->sectors_per_block;
2746
2747 /*
Mike Snitzer0424caa2012-09-26 23:45:47 +01002748 * discard_granularity is just a hint, and not enforced.
Joe Thornber104655f2012-03-28 18:41:28 +01002749 */
Mike Snitzer0424caa2012-09-26 23:45:47 +01002750 if (pt->adjusted_pf.discard_passdown) {
2751 data_limits = &bdev_get_queue(pt->data_dev->bdev)->limits;
2752 limits->discard_granularity = data_limits->discard_granularity;
Mike Snitzerf13945d2013-03-01 22:45:44 +00002753 } else
Mike Snitzer0424caa2012-09-26 23:45:47 +01002754 limits->discard_granularity = pool->sectors_per_block << SECTOR_SHIFT;
Joe Thornber104655f2012-03-28 18:41:28 +01002755}
2756
Joe Thornber991d9fa2011-10-31 20:21:18 +00002757static void pool_io_hints(struct dm_target *ti, struct queue_limits *limits)
2758{
2759 struct pool_c *pt = ti->private;
2760 struct pool *pool = pt->pool;
Mike Snitzer0cc67cd2013-08-20 15:02:41 -04002761 uint64_t io_opt_sectors = limits->io_opt >> SECTOR_SHIFT;
Joe Thornber991d9fa2011-10-31 20:21:18 +00002762
Mike Snitzer0cc67cd2013-08-20 15:02:41 -04002763 /*
2764 * If the system-determined stacked limits are compatible with the
2765 * pool's blocksize (io_opt is a factor) do not override them.
2766 */
2767 if (io_opt_sectors < pool->sectors_per_block ||
2768 do_div(io_opt_sectors, pool->sectors_per_block)) {
2769 blk_limits_io_min(limits, 0);
2770 blk_limits_io_opt(limits, pool->sectors_per_block << SECTOR_SHIFT);
2771 }
Mike Snitzer0424caa2012-09-26 23:45:47 +01002772
2773 /*
2774 * pt->adjusted_pf is a staging area for the actual features to use.
2775 * They get transferred to the live pool in bind_control_target()
2776 * called from pool_preresume().
2777 */
Mike Snitzerb60ab992013-09-19 18:49:11 -04002778 if (!pt->adjusted_pf.discard_enabled) {
2779 /*
2780 * Must explicitly disallow stacking discard limits otherwise the
2781 * block layer will stack them if pool's data device has support.
2782 * QUEUE_FLAG_DISCARD wouldn't be set but there is no way for the
2783 * user to see that, so make sure to set all discard limits to 0.
2784 */
2785 limits->discard_granularity = 0;
Mike Snitzer0424caa2012-09-26 23:45:47 +01002786 return;
Mike Snitzerb60ab992013-09-19 18:49:11 -04002787 }
Mike Snitzer0424caa2012-09-26 23:45:47 +01002788
2789 disable_passdown_if_not_supported(pt);
2790
2791 set_discard_limits(pt, limits);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002792}
2793
2794static struct target_type pool_target = {
2795 .name = "thin-pool",
2796 .features = DM_TARGET_SINGLETON | DM_TARGET_ALWAYS_WRITEABLE |
2797 DM_TARGET_IMMUTABLE,
Mike Snitzer787a996c2013-12-06 16:21:43 -05002798 .version = {1, 10, 0},
Joe Thornber991d9fa2011-10-31 20:21:18 +00002799 .module = THIS_MODULE,
2800 .ctr = pool_ctr,
2801 .dtr = pool_dtr,
2802 .map = pool_map,
2803 .postsuspend = pool_postsuspend,
2804 .preresume = pool_preresume,
2805 .resume = pool_resume,
2806 .message = pool_message,
2807 .status = pool_status,
2808 .merge = pool_merge,
2809 .iterate_devices = pool_iterate_devices,
2810 .io_hints = pool_io_hints,
2811};
2812
2813/*----------------------------------------------------------------
2814 * Thin target methods
2815 *--------------------------------------------------------------*/
2816static void thin_dtr(struct dm_target *ti)
2817{
2818 struct thin_c *tc = ti->private;
2819
2820 mutex_lock(&dm_thin_pool_table.mutex);
2821
2822 __pool_dec(tc->pool);
2823 dm_pool_close_thin_device(tc->td);
2824 dm_put_device(ti, tc->pool_dev);
Joe Thornber2dd9c252012-03-28 18:41:28 +01002825 if (tc->origin_dev)
2826 dm_put_device(ti, tc->origin_dev);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002827 kfree(tc);
2828
2829 mutex_unlock(&dm_thin_pool_table.mutex);
2830}
2831
2832/*
2833 * Thin target parameters:
2834 *
Joe Thornber2dd9c252012-03-28 18:41:28 +01002835 * <pool_dev> <dev_id> [origin_dev]
Joe Thornber991d9fa2011-10-31 20:21:18 +00002836 *
2837 * pool_dev: the path to the pool (eg, /dev/mapper/my_pool)
2838 * dev_id: the internal device identifier
Joe Thornber2dd9c252012-03-28 18:41:28 +01002839 * origin_dev: a device external to the pool that should act as the origin
Joe Thornber67e2e2b2012-03-28 18:41:29 +01002840 *
2841 * If the pool device has discards disabled, they get disabled for the thin
2842 * device as well.
Joe Thornber991d9fa2011-10-31 20:21:18 +00002843 */
2844static int thin_ctr(struct dm_target *ti, unsigned argc, char **argv)
2845{
2846 int r;
2847 struct thin_c *tc;
Joe Thornber2dd9c252012-03-28 18:41:28 +01002848 struct dm_dev *pool_dev, *origin_dev;
Joe Thornber991d9fa2011-10-31 20:21:18 +00002849 struct mapped_device *pool_md;
2850
2851 mutex_lock(&dm_thin_pool_table.mutex);
2852
Joe Thornber2dd9c252012-03-28 18:41:28 +01002853 if (argc != 2 && argc != 3) {
Joe Thornber991d9fa2011-10-31 20:21:18 +00002854 ti->error = "Invalid argument count";
2855 r = -EINVAL;
2856 goto out_unlock;
2857 }
2858
2859 tc = ti->private = kzalloc(sizeof(*tc), GFP_KERNEL);
2860 if (!tc) {
2861 ti->error = "Out of memory";
2862 r = -ENOMEM;
2863 goto out_unlock;
2864 }
2865
Joe Thornber2dd9c252012-03-28 18:41:28 +01002866 if (argc == 3) {
2867 r = dm_get_device(ti, argv[2], FMODE_READ, &origin_dev);
2868 if (r) {
2869 ti->error = "Error opening origin device";
2870 goto bad_origin_dev;
2871 }
2872 tc->origin_dev = origin_dev;
2873 }
2874
Joe Thornber991d9fa2011-10-31 20:21:18 +00002875 r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &pool_dev);
2876 if (r) {
2877 ti->error = "Error opening pool device";
2878 goto bad_pool_dev;
2879 }
2880 tc->pool_dev = pool_dev;
2881
2882 if (read_dev_id(argv[1], (unsigned long long *)&tc->dev_id, 0)) {
2883 ti->error = "Invalid device id";
2884 r = -EINVAL;
2885 goto bad_common;
2886 }
2887
2888 pool_md = dm_get_md(tc->pool_dev->bdev->bd_dev);
2889 if (!pool_md) {
2890 ti->error = "Couldn't get pool mapped device";
2891 r = -EINVAL;
2892 goto bad_common;
2893 }
2894
2895 tc->pool = __pool_table_lookup(pool_md);
2896 if (!tc->pool) {
2897 ti->error = "Couldn't find pool object";
2898 r = -EINVAL;
2899 goto bad_pool_lookup;
2900 }
2901 __pool_inc(tc->pool);
2902
Joe Thornbere49e5822012-07-27 15:08:16 +01002903 if (get_pool_mode(tc->pool) == PM_FAIL) {
2904 ti->error = "Couldn't open thin device, Pool is in fail mode";
Mike Snitzer1acacc02014-02-19 20:32:33 -05002905 r = -EINVAL;
Joe Thornbere49e5822012-07-27 15:08:16 +01002906 goto bad_thin_open;
2907 }
2908
Joe Thornber991d9fa2011-10-31 20:21:18 +00002909 r = dm_pool_open_thin_device(tc->pool->pmd, tc->dev_id, &tc->td);
2910 if (r) {
2911 ti->error = "Couldn't open thin internal device";
2912 goto bad_thin_open;
2913 }
2914
Mike Snitzer542f9032012-07-27 15:08:00 +01002915 r = dm_set_target_max_io_len(ti, tc->pool->sectors_per_block);
2916 if (r)
Mike Snitzer1acacc02014-02-19 20:32:33 -05002917 goto bad_target_max_io_len;
Mike Snitzer542f9032012-07-27 15:08:00 +01002918
Alasdair G Kergon55a62ee2013-03-01 22:45:47 +00002919 ti->num_flush_bios = 1;
Joe Thornber16ad3d12012-07-27 15:08:07 +01002920 ti->flush_supported = true;
Mikulas Patocka59c3d2c2012-12-21 20:23:40 +00002921 ti->per_bio_data_size = sizeof(struct dm_thin_endio_hook);
Joe Thornber67e2e2b2012-03-28 18:41:29 +01002922
2923 /* In case the pool supports discards, pass them on. */
Mike Snitzerb60ab992013-09-19 18:49:11 -04002924 ti->discard_zeroes_data_unsupported = true;
Joe Thornber67e2e2b2012-03-28 18:41:29 +01002925 if (tc->pool->pf.discard_enabled) {
Alasdair G Kergon0ac55482012-07-27 15:08:08 +01002926 ti->discards_supported = true;
Alasdair G Kergon55a62ee2013-03-01 22:45:47 +00002927 ti->num_discard_bios = 1;
Alasdair G Kergon55a62ee2013-03-01 22:45:47 +00002928 /* Discard bios must be split on a block boundary */
2929 ti->split_discard_bios = true;
Joe Thornber67e2e2b2012-03-28 18:41:29 +01002930 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00002931
2932 dm_put(pool_md);
2933
2934 mutex_unlock(&dm_thin_pool_table.mutex);
2935
2936 return 0;
2937
Mike Snitzer1acacc02014-02-19 20:32:33 -05002938bad_target_max_io_len:
2939 dm_pool_close_thin_device(tc->td);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002940bad_thin_open:
2941 __pool_dec(tc->pool);
2942bad_pool_lookup:
2943 dm_put(pool_md);
2944bad_common:
2945 dm_put_device(ti, tc->pool_dev);
2946bad_pool_dev:
Joe Thornber2dd9c252012-03-28 18:41:28 +01002947 if (tc->origin_dev)
2948 dm_put_device(ti, tc->origin_dev);
2949bad_origin_dev:
Joe Thornber991d9fa2011-10-31 20:21:18 +00002950 kfree(tc);
2951out_unlock:
2952 mutex_unlock(&dm_thin_pool_table.mutex);
2953
2954 return r;
2955}
2956
Mikulas Patocka7de3ee52012-12-21 20:23:41 +00002957static int thin_map(struct dm_target *ti, struct bio *bio)
Joe Thornber991d9fa2011-10-31 20:21:18 +00002958{
Kent Overstreet4f024f32013-10-11 15:44:27 -07002959 bio->bi_iter.bi_sector = dm_target_offset(ti, bio->bi_iter.bi_sector);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002960
Mikulas Patocka7de3ee52012-12-21 20:23:41 +00002961 return thin_bio_map(ti, bio);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002962}
2963
Mikulas Patocka7de3ee52012-12-21 20:23:41 +00002964static int thin_endio(struct dm_target *ti, struct bio *bio, int err)
Joe Thornbereb2aa482012-03-28 18:41:28 +01002965{
2966 unsigned long flags;
Mikulas Patocka59c3d2c2012-12-21 20:23:40 +00002967 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
Joe Thornbereb2aa482012-03-28 18:41:28 +01002968 struct list_head work;
Mike Snitzera24c2562012-06-03 00:30:00 +01002969 struct dm_thin_new_mapping *m, *tmp;
Joe Thornbereb2aa482012-03-28 18:41:28 +01002970 struct pool *pool = h->tc->pool;
2971
2972 if (h->shared_read_entry) {
2973 INIT_LIST_HEAD(&work);
Mike Snitzer44feb382012-10-12 21:02:10 +01002974 dm_deferred_entry_dec(h->shared_read_entry, &work);
Joe Thornbereb2aa482012-03-28 18:41:28 +01002975
2976 spin_lock_irqsave(&pool->lock, flags);
2977 list_for_each_entry_safe(m, tmp, &work, list) {
2978 list_del(&m->list);
Mike Snitzer7f214662013-12-17 13:43:31 -05002979 m->quiesced = true;
Joe Thornbereb2aa482012-03-28 18:41:28 +01002980 __maybe_add_mapping(m);
2981 }
2982 spin_unlock_irqrestore(&pool->lock, flags);
2983 }
2984
Joe Thornber104655f2012-03-28 18:41:28 +01002985 if (h->all_io_entry) {
2986 INIT_LIST_HEAD(&work);
Mike Snitzer44feb382012-10-12 21:02:10 +01002987 dm_deferred_entry_dec(h->all_io_entry, &work);
Joe Thornber563af182012-12-21 20:23:31 +00002988 if (!list_empty(&work)) {
2989 spin_lock_irqsave(&pool->lock, flags);
2990 list_for_each_entry_safe(m, tmp, &work, list)
Mike Snitzerdaec3382013-12-11 14:01:20 -05002991 list_add_tail(&m->list, &pool->prepared_discards);
Joe Thornber563af182012-12-21 20:23:31 +00002992 spin_unlock_irqrestore(&pool->lock, flags);
2993 wake_worker(pool);
2994 }
Joe Thornber104655f2012-03-28 18:41:28 +01002995 }
2996
Joe Thornbereb2aa482012-03-28 18:41:28 +01002997 return 0;
2998}
2999
Joe Thornber991d9fa2011-10-31 20:21:18 +00003000static void thin_postsuspend(struct dm_target *ti)
3001{
3002 if (dm_noflush_suspending(ti))
3003 requeue_io((struct thin_c *)ti->private);
3004}
3005
3006/*
3007 * <nr mapped sectors> <highest mapped sector>
3008 */
Mikulas Patockafd7c0922013-03-01 22:45:44 +00003009static void thin_status(struct dm_target *ti, status_type_t type,
3010 unsigned status_flags, char *result, unsigned maxlen)
Joe Thornber991d9fa2011-10-31 20:21:18 +00003011{
3012 int r;
3013 ssize_t sz = 0;
3014 dm_block_t mapped, highest;
3015 char buf[BDEVNAME_SIZE];
3016 struct thin_c *tc = ti->private;
3017
Joe Thornbere49e5822012-07-27 15:08:16 +01003018 if (get_pool_mode(tc->pool) == PM_FAIL) {
3019 DMEMIT("Fail");
Mikulas Patockafd7c0922013-03-01 22:45:44 +00003020 return;
Joe Thornbere49e5822012-07-27 15:08:16 +01003021 }
3022
Joe Thornber991d9fa2011-10-31 20:21:18 +00003023 if (!tc->td)
3024 DMEMIT("-");
3025 else {
3026 switch (type) {
3027 case STATUSTYPE_INFO:
3028 r = dm_thin_get_mapped_count(tc->td, &mapped);
Mikulas Patockafd7c0922013-03-01 22:45:44 +00003029 if (r) {
3030 DMERR("dm_thin_get_mapped_count returned %d", r);
3031 goto err;
3032 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00003033
3034 r = dm_thin_get_highest_mapped_block(tc->td, &highest);
Mikulas Patockafd7c0922013-03-01 22:45:44 +00003035 if (r < 0) {
3036 DMERR("dm_thin_get_highest_mapped_block returned %d", r);
3037 goto err;
3038 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00003039
3040 DMEMIT("%llu ", mapped * tc->pool->sectors_per_block);
3041 if (r)
3042 DMEMIT("%llu", ((highest + 1) *
3043 tc->pool->sectors_per_block) - 1);
3044 else
3045 DMEMIT("-");
3046 break;
3047
3048 case STATUSTYPE_TABLE:
3049 DMEMIT("%s %lu",
3050 format_dev_t(buf, tc->pool_dev->bdev->bd_dev),
3051 (unsigned long) tc->dev_id);
Joe Thornber2dd9c252012-03-28 18:41:28 +01003052 if (tc->origin_dev)
3053 DMEMIT(" %s", format_dev_t(buf, tc->origin_dev->bdev->bd_dev));
Joe Thornber991d9fa2011-10-31 20:21:18 +00003054 break;
3055 }
3056 }
3057
Mikulas Patockafd7c0922013-03-01 22:45:44 +00003058 return;
3059
3060err:
3061 DMEMIT("Error");
Joe Thornber991d9fa2011-10-31 20:21:18 +00003062}
3063
3064static int thin_iterate_devices(struct dm_target *ti,
3065 iterate_devices_callout_fn fn, void *data)
3066{
Mike Snitzer55f2b8b2012-07-27 15:08:02 +01003067 sector_t blocks;
Joe Thornber991d9fa2011-10-31 20:21:18 +00003068 struct thin_c *tc = ti->private;
Mike Snitzer55f2b8b2012-07-27 15:08:02 +01003069 struct pool *pool = tc->pool;
Joe Thornber991d9fa2011-10-31 20:21:18 +00003070
3071 /*
3072 * We can't call dm_pool_get_data_dev_size() since that blocks. So
3073 * we follow a more convoluted path through to the pool's target.
3074 */
Mike Snitzer55f2b8b2012-07-27 15:08:02 +01003075 if (!pool->ti)
Joe Thornber991d9fa2011-10-31 20:21:18 +00003076 return 0; /* nothing is bound */
3077
Mike Snitzer55f2b8b2012-07-27 15:08:02 +01003078 blocks = pool->ti->len;
3079 (void) sector_div(blocks, pool->sectors_per_block);
Joe Thornber991d9fa2011-10-31 20:21:18 +00003080 if (blocks)
Mike Snitzer55f2b8b2012-07-27 15:08:02 +01003081 return fn(ti, tc->pool_dev, 0, pool->sectors_per_block * blocks, data);
Joe Thornber991d9fa2011-10-31 20:21:18 +00003082
3083 return 0;
3084}
3085
Joe Thornber991d9fa2011-10-31 20:21:18 +00003086static struct target_type thin_target = {
3087 .name = "thin",
Mike Snitzer787a996c2013-12-06 16:21:43 -05003088 .version = {1, 10, 0},
Joe Thornber991d9fa2011-10-31 20:21:18 +00003089 .module = THIS_MODULE,
3090 .ctr = thin_ctr,
3091 .dtr = thin_dtr,
3092 .map = thin_map,
Joe Thornbereb2aa482012-03-28 18:41:28 +01003093 .end_io = thin_endio,
Joe Thornber991d9fa2011-10-31 20:21:18 +00003094 .postsuspend = thin_postsuspend,
3095 .status = thin_status,
3096 .iterate_devices = thin_iterate_devices,
Joe Thornber991d9fa2011-10-31 20:21:18 +00003097};
3098
3099/*----------------------------------------------------------------*/
3100
3101static int __init dm_thin_init(void)
3102{
3103 int r;
3104
3105 pool_table_init();
3106
3107 r = dm_register_target(&thin_target);
3108 if (r)
3109 return r;
3110
3111 r = dm_register_target(&pool_target);
3112 if (r)
Mike Snitzera24c2562012-06-03 00:30:00 +01003113 goto bad_pool_target;
3114
3115 r = -ENOMEM;
3116
Mike Snitzera24c2562012-06-03 00:30:00 +01003117 _new_mapping_cache = KMEM_CACHE(dm_thin_new_mapping, 0);
3118 if (!_new_mapping_cache)
3119 goto bad_new_mapping_cache;
3120
Mike Snitzera24c2562012-06-03 00:30:00 +01003121 return 0;
3122
Mike Snitzera24c2562012-06-03 00:30:00 +01003123bad_new_mapping_cache:
Mike Snitzera24c2562012-06-03 00:30:00 +01003124 dm_unregister_target(&pool_target);
3125bad_pool_target:
3126 dm_unregister_target(&thin_target);
Joe Thornber991d9fa2011-10-31 20:21:18 +00003127
3128 return r;
3129}
3130
3131static void dm_thin_exit(void)
3132{
3133 dm_unregister_target(&thin_target);
3134 dm_unregister_target(&pool_target);
Mike Snitzera24c2562012-06-03 00:30:00 +01003135
Mike Snitzera24c2562012-06-03 00:30:00 +01003136 kmem_cache_destroy(_new_mapping_cache);
Joe Thornber991d9fa2011-10-31 20:21:18 +00003137}
3138
3139module_init(dm_thin_init);
3140module_exit(dm_thin_exit);
3141
Alasdair G Kergon7cab8bf2012-05-12 01:43:19 +01003142MODULE_DESCRIPTION(DM_NAME " thin provisioning target");
Joe Thornber991d9fa2011-10-31 20:21:18 +00003143MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3144MODULE_LICENSE("GPL");