blob: fc803d50f9f026e45932f8b131764ad479831237 [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>
Manuel Schölling0f30af92014-05-22 22:42:37 +020014#include <linux/jiffies.h>
Mike Snitzer604ea902014-10-09 18:43:25 -040015#include <linux/log2.h>
Joe Thornber991d9fa2011-10-31 20:21:18 +000016#include <linux/list.h>
Mike Snitzerc140e1c2014-03-20 21:17:14 -040017#include <linux/rculist.h>
Joe Thornber991d9fa2011-10-31 20:21:18 +000018#include <linux/init.h>
19#include <linux/module.h>
20#include <linux/slab.h>
Joe Thornbera822c832015-07-03 10:22:42 +010021#include <linux/vmalloc.h>
Joe Thornberac4c3f32014-10-10 16:42:10 +010022#include <linux/sort.h>
Mike Snitzer67324ea2014-03-21 18:33:41 -040023#include <linux/rbtree.h>
Joe Thornber991d9fa2011-10-31 20:21:18 +000024
25#define DM_MSG_PREFIX "thin"
26
27/*
28 * Tunable constants
29 */
Alasdair G Kergon7768ed32012-07-27 15:07:57 +010030#define ENDIO_HOOK_POOL_SIZE 1024
Joe Thornber991d9fa2011-10-31 20:21:18 +000031#define MAPPING_POOL_SIZE 1024
Joe Thornber905e51b2012-03-28 18:41:27 +010032#define COMMIT_PERIOD HZ
Mike Snitzer80c57892014-05-20 13:38:33 -040033#define NO_SPACE_TIMEOUT_SECS 60
34
35static unsigned no_space_timeout_secs = NO_SPACE_TIMEOUT_SECS;
Joe Thornber991d9fa2011-10-31 20:21:18 +000036
Mikulas Patockadf5d2e92013-03-01 22:45:49 +000037DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(snapshot_copy_throttle,
38 "A percentage of time allocated for copy on write");
39
Joe Thornber991d9fa2011-10-31 20:21:18 +000040/*
41 * The block size of the device holding pool data must be
42 * between 64KB and 1GB.
43 */
44#define DATA_DEV_BLOCK_SIZE_MIN_SECTORS (64 * 1024 >> SECTOR_SHIFT)
45#define DATA_DEV_BLOCK_SIZE_MAX_SECTORS (1024 * 1024 * 1024 >> SECTOR_SHIFT)
46
47/*
Joe Thornber991d9fa2011-10-31 20:21:18 +000048 * Device id is restricted to 24 bits.
49 */
50#define MAX_DEV_ID ((1 << 24) - 1)
51
52/*
53 * How do we handle breaking sharing of data blocks?
54 * =================================================
55 *
56 * We use a standard copy-on-write btree to store the mappings for the
57 * devices (note I'm talking about copy-on-write of the metadata here, not
58 * the data). When you take an internal snapshot you clone the root node
59 * of the origin btree. After this there is no concept of an origin or a
60 * snapshot. They are just two device trees that happen to point to the
61 * same data blocks.
62 *
63 * When we get a write in we decide if it's to a shared data block using
64 * some timestamp magic. If it is, we have to break sharing.
65 *
66 * Let's say we write to a shared block in what was the origin. The
67 * steps are:
68 *
69 * i) plug io further to this physical block. (see bio_prison code).
70 *
71 * ii) quiesce any read io to that shared data block. Obviously
Mike Snitzer44feb382012-10-12 21:02:10 +010072 * including all devices that share this block. (see dm_deferred_set code)
Joe Thornber991d9fa2011-10-31 20:21:18 +000073 *
74 * iii) copy the data block to a newly allocate block. This step can be
75 * missed out if the io covers the block. (schedule_copy).
76 *
77 * iv) insert the new mapping into the origin's btree
Joe Thornberfe878f32012-03-28 18:41:24 +010078 * (process_prepared_mapping). This act of inserting breaks some
Joe Thornber991d9fa2011-10-31 20:21:18 +000079 * sharing of btree nodes between the two devices. Breaking sharing only
80 * effects the btree of that specific device. Btrees for the other
81 * devices that share the block never change. The btree for the origin
82 * device as it was after the last commit is untouched, ie. we're using
83 * persistent data structures in the functional programming sense.
84 *
85 * v) unplug io to this physical block, including the io that triggered
86 * the breaking of sharing.
87 *
88 * Steps (ii) and (iii) occur in parallel.
89 *
90 * The metadata _doesn't_ need to be committed before the io continues. We
91 * get away with this because the io is always written to a _new_ block.
92 * If there's a crash, then:
93 *
94 * - The origin mapping will point to the old origin block (the shared
95 * one). This will contain the data as it was before the io that triggered
96 * the breaking of sharing came in.
97 *
98 * - The snap mapping still points to the old block. As it would after
99 * the commit.
100 *
101 * The downside of this scheme is the timestamp magic isn't perfect, and
102 * will continue to think that data block in the snapshot device is shared
103 * even after the write to the origin has broken sharing. I suspect data
104 * blocks will typically be shared by many different devices, so we're
105 * breaking sharing n + 1 times, rather than n, where n is the number of
106 * devices that reference this data block. At the moment I think the
107 * benefits far, far outweigh the disadvantages.
108 */
109
110/*----------------------------------------------------------------*/
111
112/*
Joe Thornber991d9fa2011-10-31 20:21:18 +0000113 * Key building.
114 */
Joe Thornber34fbcf62015-04-16 12:58:35 +0100115enum lock_space {
116 VIRTUAL,
117 PHYSICAL
118};
119
120static void build_key(struct dm_thin_device *td, enum lock_space ls,
121 dm_block_t b, dm_block_t e, struct dm_cell_key *key)
Joe Thornber991d9fa2011-10-31 20:21:18 +0000122{
Joe Thornber34fbcf62015-04-16 12:58:35 +0100123 key->virtual = (ls == VIRTUAL);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000124 key->dev = dm_thin_dev_id(td);
Joe Thornber5f274d82014-09-17 10:17:39 +0100125 key->block_begin = b;
Joe Thornber34fbcf62015-04-16 12:58:35 +0100126 key->block_end = e;
127}
128
129static void build_data_key(struct dm_thin_device *td, dm_block_t b,
130 struct dm_cell_key *key)
131{
132 build_key(td, PHYSICAL, b, b + 1llu, key);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000133}
134
135static void build_virtual_key(struct dm_thin_device *td, dm_block_t b,
Mike Snitzer44feb382012-10-12 21:02:10 +0100136 struct dm_cell_key *key)
Joe Thornber991d9fa2011-10-31 20:21:18 +0000137{
Joe Thornber34fbcf62015-04-16 12:58:35 +0100138 build_key(td, VIRTUAL, b, b + 1llu, key);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000139}
140
141/*----------------------------------------------------------------*/
142
Joe Thornber7d327fe2014-10-06 15:45:59 +0100143#define THROTTLE_THRESHOLD (1 * HZ)
144
145struct throttle {
146 struct rw_semaphore lock;
147 unsigned long threshold;
148 bool throttle_applied;
149};
150
151static void throttle_init(struct throttle *t)
152{
153 init_rwsem(&t->lock);
154 t->throttle_applied = false;
155}
156
157static void throttle_work_start(struct throttle *t)
158{
159 t->threshold = jiffies + THROTTLE_THRESHOLD;
160}
161
162static void throttle_work_update(struct throttle *t)
163{
164 if (!t->throttle_applied && jiffies > t->threshold) {
165 down_write(&t->lock);
166 t->throttle_applied = true;
167 }
168}
169
170static void throttle_work_complete(struct throttle *t)
171{
172 if (t->throttle_applied) {
173 t->throttle_applied = false;
174 up_write(&t->lock);
175 }
176}
177
178static void throttle_lock(struct throttle *t)
179{
180 down_read(&t->lock);
181}
182
183static void throttle_unlock(struct throttle *t)
184{
185 up_read(&t->lock);
186}
187
188/*----------------------------------------------------------------*/
189
Joe Thornber991d9fa2011-10-31 20:21:18 +0000190/*
191 * A pool device ties together a metadata device and a data device. It
192 * also provides the interface for creating and destroying internal
193 * devices.
194 */
Mike Snitzera24c2562012-06-03 00:30:00 +0100195struct dm_thin_new_mapping;
Joe Thornber67e2e2b2012-03-28 18:41:29 +0100196
Joe Thornbere49e5822012-07-27 15:08:16 +0100197/*
Joe Thornber3e1a0692014-03-03 16:03:26 +0000198 * The pool runs in 4 modes. Ordered in degraded order for comparisons.
Joe Thornbere49e5822012-07-27 15:08:16 +0100199 */
200enum pool_mode {
201 PM_WRITE, /* metadata may be changed */
Joe Thornber3e1a0692014-03-03 16:03:26 +0000202 PM_OUT_OF_DATA_SPACE, /* metadata may be changed, though data may not be allocated */
Joe Thornbere49e5822012-07-27 15:08:16 +0100203 PM_READ_ONLY, /* metadata may not be changed */
204 PM_FAIL, /* all I/O fails */
205};
206
Joe Thornber67e2e2b2012-03-28 18:41:29 +0100207struct pool_features {
Joe Thornbere49e5822012-07-27 15:08:16 +0100208 enum pool_mode mode;
209
Mike Snitzer9bc142d2012-09-26 23:45:46 +0100210 bool zero_new_blocks:1;
211 bool discard_enabled:1;
212 bool discard_passdown:1;
Mike Snitzer787a996c2013-12-06 16:21:43 -0500213 bool error_if_no_space:1;
Joe Thornber67e2e2b2012-03-28 18:41:29 +0100214};
215
Joe Thornbere49e5822012-07-27 15:08:16 +0100216struct thin_c;
217typedef void (*process_bio_fn)(struct thin_c *tc, struct bio *bio);
Joe Thornbera374bb22014-10-10 13:43:14 +0100218typedef void (*process_cell_fn)(struct thin_c *tc, struct dm_bio_prison_cell *cell);
Joe Thornbere49e5822012-07-27 15:08:16 +0100219typedef void (*process_mapping_fn)(struct dm_thin_new_mapping *m);
220
Joe Thornberac4c3f32014-10-10 16:42:10 +0100221#define CELL_SORT_ARRAY_SIZE 8192
222
Joe Thornber991d9fa2011-10-31 20:21:18 +0000223struct pool {
224 struct list_head list;
225 struct dm_target *ti; /* Only set if a pool target is bound */
226
227 struct mapped_device *pool_md;
228 struct block_device *md_dev;
229 struct dm_pool_metadata *pmd;
230
Joe Thornber991d9fa2011-10-31 20:21:18 +0000231 dm_block_t low_water_blocks;
Mike Snitzer55f2b8b2012-07-27 15:08:02 +0100232 uint32_t sectors_per_block;
Mikulas Patockaf9a8e0c2012-07-27 15:08:03 +0100233 int sectors_per_block_shift;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000234
Joe Thornber67e2e2b2012-03-28 18:41:29 +0100235 struct pool_features pf;
Joe Thornber88a66212013-12-04 20:16:12 -0500236 bool low_water_triggered:1; /* A dm event has been sent */
Mike Snitzer80e96c52014-11-07 15:09:46 -0500237 bool suspended:1;
Mike Snitzerc3667cc2016-03-10 11:31:35 -0500238 bool out_of_data_space:1;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000239
Mike Snitzer44feb382012-10-12 21:02:10 +0100240 struct dm_bio_prison *prison;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000241 struct dm_kcopyd_client *copier;
242
243 struct workqueue_struct *wq;
Joe Thornber7d327fe2014-10-06 15:45:59 +0100244 struct throttle throttle;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000245 struct work_struct worker;
Joe Thornber905e51b2012-03-28 18:41:27 +0100246 struct delayed_work waker;
Joe Thornber85ad643b2014-05-09 15:59:38 +0100247 struct delayed_work no_space_timeout;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000248
Joe Thornber905e51b2012-03-28 18:41:27 +0100249 unsigned long last_commit_jiffies;
Mike Snitzer55f2b8b2012-07-27 15:08:02 +0100250 unsigned ref_count;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000251
252 spinlock_t lock;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000253 struct bio_list deferred_flush_bios;
254 struct list_head prepared_mappings;
Joe Thornber104655f2012-03-28 18:41:28 +0100255 struct list_head prepared_discards;
Mike Snitzerc140e1c2014-03-20 21:17:14 -0400256 struct list_head active_thins;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000257
Mike Snitzer44feb382012-10-12 21:02:10 +0100258 struct dm_deferred_set *shared_read_ds;
259 struct dm_deferred_set *all_io_ds;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000260
Mike Snitzera24c2562012-06-03 00:30:00 +0100261 struct dm_thin_new_mapping *next_mapping;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000262 mempool_t *mapping_pool;
Joe Thornbere49e5822012-07-27 15:08:16 +0100263
264 process_bio_fn process_bio;
265 process_bio_fn process_discard;
266
Joe Thornbera374bb22014-10-10 13:43:14 +0100267 process_cell_fn process_cell;
268 process_cell_fn process_discard_cell;
269
Joe Thornbere49e5822012-07-27 15:08:16 +0100270 process_mapping_fn process_prepared_mapping;
271 process_mapping_fn process_prepared_discard;
Joe Thornberac4c3f32014-10-10 16:42:10 +0100272
Joe Thornbera822c832015-07-03 10:22:42 +0100273 struct dm_bio_prison_cell **cell_sort_array;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000274};
275
Joe Thornbere49e5822012-07-27 15:08:16 +0100276static enum pool_mode get_pool_mode(struct pool *pool);
Joe Thornberb5330652013-12-04 19:51:33 -0500277static void metadata_operation_failed(struct pool *pool, const char *op, int r);
Joe Thornbere49e5822012-07-27 15:08:16 +0100278
Joe Thornber991d9fa2011-10-31 20:21:18 +0000279/*
280 * Target context for a pool.
281 */
282struct pool_c {
283 struct dm_target *ti;
284 struct pool *pool;
285 struct dm_dev *data_dev;
286 struct dm_dev *metadata_dev;
287 struct dm_target_callbacks callbacks;
288
289 dm_block_t low_water_blocks;
Mike Snitzer0424caa2012-09-26 23:45:47 +0100290 struct pool_features requested_pf; /* Features requested during table load */
291 struct pool_features adjusted_pf; /* Features used after adjusting for constituent devices */
Joe Thornber991d9fa2011-10-31 20:21:18 +0000292};
293
294/*
295 * Target context for a thin.
296 */
297struct thin_c {
Mike Snitzerc140e1c2014-03-20 21:17:14 -0400298 struct list_head list;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000299 struct dm_dev *pool_dev;
Joe Thornber2dd9c252012-03-28 18:41:28 +0100300 struct dm_dev *origin_dev;
Joe Thornbere5aea7b2014-06-13 14:47:24 +0100301 sector_t origin_size;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000302 dm_thin_id dev_id;
303
304 struct pool *pool;
305 struct dm_thin_device *td;
Mike Snitzer583024d2014-10-28 20:58:45 -0400306 struct mapped_device *thin_md;
307
Joe Thornber738211f2014-03-03 15:52:28 +0000308 bool requeue_mode:1;
Mike Snitzerc140e1c2014-03-20 21:17:14 -0400309 spinlock_t lock;
Joe Thornbera374bb22014-10-10 13:43:14 +0100310 struct list_head deferred_cells;
Mike Snitzerc140e1c2014-03-20 21:17:14 -0400311 struct bio_list deferred_bio_list;
312 struct bio_list retry_on_resume_list;
Mike Snitzer67324ea2014-03-21 18:33:41 -0400313 struct rb_root sort_bio_list; /* sorted list of deferred bios */
Joe Thornberb10ebd32014-04-08 11:29:01 +0100314
315 /*
316 * Ensures the thin is not destroyed until the worker has finished
317 * iterating the active_thins list.
318 */
319 atomic_t refcount;
320 struct completion can_destroy;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000321};
322
323/*----------------------------------------------------------------*/
324
Joe Thornber34fbcf62015-04-16 12:58:35 +0100325static bool block_size_is_power_of_two(struct pool *pool)
326{
327 return pool->sectors_per_block_shift >= 0;
328}
329
330static sector_t block_to_sectors(struct pool *pool, dm_block_t b)
331{
332 return block_size_is_power_of_two(pool) ?
333 (b << pool->sectors_per_block_shift) :
334 (b * pool->sectors_per_block);
335}
336
Joe Thornber202bae52016-05-04 14:12:42 -0400337/*----------------------------------------------------------------*/
338
339struct discard_op {
340 struct thin_c *tc;
341 struct blk_plug plug;
342 struct bio *parent_bio;
343 struct bio *bio;
344};
345
346static void begin_discard(struct discard_op *op, struct thin_c *tc, struct bio *parent)
Joe Thornber34fbcf62015-04-16 12:58:35 +0100347{
Joe Thornber202bae52016-05-04 14:12:42 -0400348 BUG_ON(!parent);
349
350 op->tc = tc;
351 blk_start_plug(&op->plug);
352 op->parent_bio = parent;
353 op->bio = NULL;
354}
355
356static int issue_discard(struct discard_op *op, dm_block_t data_b, dm_block_t data_e)
357{
358 struct thin_c *tc = op->tc;
Joe Thornber34fbcf62015-04-16 12:58:35 +0100359 sector_t s = block_to_sectors(tc->pool, data_b);
360 sector_t len = block_to_sectors(tc->pool, data_e - data_b);
361
Joe Thornber202bae52016-05-04 14:12:42 -0400362 return __blkdev_issue_discard(tc->pool_dev->bdev, s, len,
363 GFP_NOWAIT, REQ_WRITE | REQ_DISCARD, &op->bio);
364}
365
366static void end_discard(struct discard_op *op, int r)
367{
368 if (op->bio) {
369 /*
370 * Even if one of the calls to issue_discard failed, we
371 * need to wait for the chain to complete.
372 */
373 bio_chain(op->bio, op->parent_bio);
374 submit_bio(REQ_WRITE | REQ_DISCARD, op->bio);
Mike Snitzer3dba53a92016-05-02 20:16:21 -0400375 }
Mike Snitzer3dba53a92016-05-02 20:16:21 -0400376
Joe Thornber202bae52016-05-04 14:12:42 -0400377 blk_finish_plug(&op->plug);
378
379 /*
380 * Even if r is set, there could be sub discards in flight that we
381 * need to wait for.
382 */
383 if (r && !op->parent_bio->bi_error)
384 op->parent_bio->bi_error = r;
385 bio_endio(op->parent_bio);
Joe Thornber34fbcf62015-04-16 12:58:35 +0100386}
387
388/*----------------------------------------------------------------*/
389
Joe Thornber025b9682013-03-01 22:45:50 +0000390/*
391 * wake_worker() is used when new work is queued and when pool_resume is
392 * ready to continue deferred IO processing.
393 */
394static void wake_worker(struct pool *pool)
395{
396 queue_work(pool->wq, &pool->worker);
397}
398
399/*----------------------------------------------------------------*/
400
Joe Thornber6beca5e2013-03-01 22:45:50 +0000401static int bio_detain(struct pool *pool, struct dm_cell_key *key, struct bio *bio,
402 struct dm_bio_prison_cell **cell_result)
403{
404 int r;
405 struct dm_bio_prison_cell *cell_prealloc;
406
407 /*
408 * Allocate a cell from the prison's mempool.
409 * This might block but it can't fail.
410 */
411 cell_prealloc = dm_bio_prison_alloc_cell(pool->prison, GFP_NOIO);
412
413 r = dm_bio_detain(pool->prison, key, bio, cell_prealloc, cell_result);
414 if (r)
415 /*
416 * We reused an old cell; we can get rid of
417 * the new one.
418 */
419 dm_bio_prison_free_cell(pool->prison, cell_prealloc);
420
421 return r;
422}
423
424static void cell_release(struct pool *pool,
425 struct dm_bio_prison_cell *cell,
426 struct bio_list *bios)
427{
428 dm_cell_release(pool->prison, cell, bios);
429 dm_bio_prison_free_cell(pool->prison, cell);
430}
431
Joe Thornber2d759a42014-10-10 15:27:16 +0100432static void cell_visit_release(struct pool *pool,
433 void (*fn)(void *, struct dm_bio_prison_cell *),
434 void *context,
435 struct dm_bio_prison_cell *cell)
436{
437 dm_cell_visit_release(pool->prison, fn, context, cell);
438 dm_bio_prison_free_cell(pool->prison, cell);
439}
440
Joe Thornber6beca5e2013-03-01 22:45:50 +0000441static void cell_release_no_holder(struct pool *pool,
442 struct dm_bio_prison_cell *cell,
443 struct bio_list *bios)
444{
445 dm_cell_release_no_holder(pool->prison, cell, bios);
446 dm_bio_prison_free_cell(pool->prison, cell);
447}
448
Mike Snitzeraf918052014-05-22 14:32:51 -0400449static void cell_error_with_code(struct pool *pool,
450 struct dm_bio_prison_cell *cell, int error_code)
Joe Thornber6beca5e2013-03-01 22:45:50 +0000451{
Mike Snitzeraf918052014-05-22 14:32:51 -0400452 dm_cell_error(pool->prison, cell, error_code);
Joe Thornber6beca5e2013-03-01 22:45:50 +0000453 dm_bio_prison_free_cell(pool->prison, cell);
454}
455
Mike Snitzerc3667cc2016-03-10 11:31:35 -0500456static int get_pool_io_error_code(struct pool *pool)
457{
458 return pool->out_of_data_space ? -ENOSPC : -EIO;
459}
460
Mike Snitzeraf918052014-05-22 14:32:51 -0400461static void cell_error(struct pool *pool, struct dm_bio_prison_cell *cell)
462{
Mike Snitzerc3667cc2016-03-10 11:31:35 -0500463 int error = get_pool_io_error_code(pool);
464
465 cell_error_with_code(pool, cell, error);
Mike Snitzeraf918052014-05-22 14:32:51 -0400466}
467
Joe Thornbera374bb22014-10-10 13:43:14 +0100468static void cell_success(struct pool *pool, struct dm_bio_prison_cell *cell)
469{
470 cell_error_with_code(pool, cell, 0);
471}
472
473static void cell_requeue(struct pool *pool, struct dm_bio_prison_cell *cell)
474{
475 cell_error_with_code(pool, cell, DM_ENDIO_REQUEUE);
476}
477
Joe Thornber6beca5e2013-03-01 22:45:50 +0000478/*----------------------------------------------------------------*/
479
Joe Thornber991d9fa2011-10-31 20:21:18 +0000480/*
481 * A global list of pools that uses a struct mapped_device as a key.
482 */
483static struct dm_thin_pool_table {
484 struct mutex mutex;
485 struct list_head pools;
486} dm_thin_pool_table;
487
488static void pool_table_init(void)
489{
490 mutex_init(&dm_thin_pool_table.mutex);
491 INIT_LIST_HEAD(&dm_thin_pool_table.pools);
492}
493
494static void __pool_table_insert(struct pool *pool)
495{
496 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
497 list_add(&pool->list, &dm_thin_pool_table.pools);
498}
499
500static void __pool_table_remove(struct pool *pool)
501{
502 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
503 list_del(&pool->list);
504}
505
506static struct pool *__pool_table_lookup(struct mapped_device *md)
507{
508 struct pool *pool = NULL, *tmp;
509
510 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
511
512 list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) {
513 if (tmp->pool_md == md) {
514 pool = tmp;
515 break;
516 }
517 }
518
519 return pool;
520}
521
522static struct pool *__pool_table_lookup_metadata_dev(struct block_device *md_dev)
523{
524 struct pool *pool = NULL, *tmp;
525
526 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
527
528 list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) {
529 if (tmp->md_dev == md_dev) {
530 pool = tmp;
531 break;
532 }
533 }
534
535 return pool;
536}
537
538/*----------------------------------------------------------------*/
539
Mike Snitzera24c2562012-06-03 00:30:00 +0100540struct dm_thin_endio_hook {
Joe Thornbereb2aa482012-03-28 18:41:28 +0100541 struct thin_c *tc;
Mike Snitzer44feb382012-10-12 21:02:10 +0100542 struct dm_deferred_entry *shared_read_entry;
543 struct dm_deferred_entry *all_io_entry;
Mike Snitzera24c2562012-06-03 00:30:00 +0100544 struct dm_thin_new_mapping *overwrite_mapping;
Mike Snitzer67324ea2014-03-21 18:33:41 -0400545 struct rb_node rb_node;
Joe Thornber34fbcf62015-04-16 12:58:35 +0100546 struct dm_bio_prison_cell *cell;
Joe Thornbereb2aa482012-03-28 18:41:28 +0100547};
548
Mike Snitzer42d6a8c2014-10-19 07:52:44 -0400549static void __merge_bio_list(struct bio_list *bios, struct bio_list *master)
550{
551 bio_list_merge(bios, master);
552 bio_list_init(master);
553}
554
555static void error_bio_list(struct bio_list *bios, int error)
Joe Thornber991d9fa2011-10-31 20:21:18 +0000556{
557 struct bio *bio;
Mike Snitzer42d6a8c2014-10-19 07:52:44 -0400558
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200559 while ((bio = bio_list_pop(bios))) {
560 bio->bi_error = error;
561 bio_endio(bio);
562 }
Mike Snitzer42d6a8c2014-10-19 07:52:44 -0400563}
564
565static void error_thin_bio_list(struct thin_c *tc, struct bio_list *master, int error)
566{
Joe Thornber991d9fa2011-10-31 20:21:18 +0000567 struct bio_list bios;
Joe Thornber18adc572014-03-03 15:46:42 +0000568 unsigned long flags;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000569
570 bio_list_init(&bios);
Joe Thornber18adc572014-03-03 15:46:42 +0000571
Mike Snitzerc140e1c2014-03-20 21:17:14 -0400572 spin_lock_irqsave(&tc->lock, flags);
Mike Snitzer42d6a8c2014-10-19 07:52:44 -0400573 __merge_bio_list(&bios, master);
Mike Snitzerc140e1c2014-03-20 21:17:14 -0400574 spin_unlock_irqrestore(&tc->lock, flags);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000575
Mike Snitzer42d6a8c2014-10-19 07:52:44 -0400576 error_bio_list(&bios, error);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000577}
578
Joe Thornbera374bb22014-10-10 13:43:14 +0100579static void requeue_deferred_cells(struct thin_c *tc)
580{
581 struct pool *pool = tc->pool;
582 unsigned long flags;
583 struct list_head cells;
584 struct dm_bio_prison_cell *cell, *tmp;
585
586 INIT_LIST_HEAD(&cells);
587
588 spin_lock_irqsave(&tc->lock, flags);
589 list_splice_init(&tc->deferred_cells, &cells);
590 spin_unlock_irqrestore(&tc->lock, flags);
591
592 list_for_each_entry_safe(cell, tmp, &cells, user_list)
593 cell_requeue(pool, cell);
594}
595
Joe Thornber991d9fa2011-10-31 20:21:18 +0000596static void requeue_io(struct thin_c *tc)
597{
Joe Thornber3e1a0692014-03-03 16:03:26 +0000598 struct bio_list bios;
Mike Snitzer42d6a8c2014-10-19 07:52:44 -0400599 unsigned long flags;
Joe Thornber3e1a0692014-03-03 16:03:26 +0000600
601 bio_list_init(&bios);
602
Mike Snitzerc140e1c2014-03-20 21:17:14 -0400603 spin_lock_irqsave(&tc->lock, flags);
Mike Snitzer42d6a8c2014-10-19 07:52:44 -0400604 __merge_bio_list(&bios, &tc->deferred_bio_list);
605 __merge_bio_list(&bios, &tc->retry_on_resume_list);
Mike Snitzerc140e1c2014-03-20 21:17:14 -0400606 spin_unlock_irqrestore(&tc->lock, flags);
Joe Thornber3e1a0692014-03-03 16:03:26 +0000607
Mike Snitzer42d6a8c2014-10-19 07:52:44 -0400608 error_bio_list(&bios, DM_ENDIO_REQUEUE);
609 requeue_deferred_cells(tc);
Joe Thornber3e1a0692014-03-03 16:03:26 +0000610}
611
Mike Snitzer0a927c22015-07-21 13:20:46 -0400612static void error_retry_list_with_code(struct pool *pool, int error)
Mike Snitzerc140e1c2014-03-20 21:17:14 -0400613{
614 struct thin_c *tc;
615
616 rcu_read_lock();
617 list_for_each_entry_rcu(tc, &pool->active_thins, list)
Mike Snitzer0a927c22015-07-21 13:20:46 -0400618 error_thin_bio_list(tc, &tc->retry_on_resume_list, error);
Mike Snitzerc140e1c2014-03-20 21:17:14 -0400619 rcu_read_unlock();
620}
621
Mike Snitzer0a927c22015-07-21 13:20:46 -0400622static void error_retry_list(struct pool *pool)
623{
Mike Snitzerc3667cc2016-03-10 11:31:35 -0500624 int error = get_pool_io_error_code(pool);
625
Amitoj Kaur Chawla813923b2016-04-11 11:37:08 +0530626 error_retry_list_with_code(pool, error);
Mike Snitzer0a927c22015-07-21 13:20:46 -0400627}
628
Joe Thornber991d9fa2011-10-31 20:21:18 +0000629/*
630 * This section of code contains the logic for processing a thin device's IO.
631 * Much of the code depends on pool object resources (lists, workqueues, etc)
632 * but most is exclusively called from the thin target rather than the thin-pool
633 * target.
634 */
635
636static dm_block_t get_bio_block(struct thin_c *tc, struct bio *bio)
637{
Mike Snitzer58f77a22013-03-01 22:45:45 +0000638 struct pool *pool = tc->pool;
Kent Overstreet4f024f32013-10-11 15:44:27 -0700639 sector_t block_nr = bio->bi_iter.bi_sector;
Mike Snitzer55f2b8b2012-07-27 15:08:02 +0100640
Mike Snitzer58f77a22013-03-01 22:45:45 +0000641 if (block_size_is_power_of_two(pool))
642 block_nr >>= pool->sectors_per_block_shift;
Mikulas Patockaf9a8e0c2012-07-27 15:08:03 +0100643 else
Mike Snitzer58f77a22013-03-01 22:45:45 +0000644 (void) sector_div(block_nr, pool->sectors_per_block);
Mike Snitzer55f2b8b2012-07-27 15:08:02 +0100645
646 return block_nr;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000647}
648
Joe Thornber34fbcf62015-04-16 12:58:35 +0100649/*
650 * Returns the _complete_ blocks that this bio covers.
651 */
652static void get_bio_block_range(struct thin_c *tc, struct bio *bio,
653 dm_block_t *begin, dm_block_t *end)
654{
655 struct pool *pool = tc->pool;
656 sector_t b = bio->bi_iter.bi_sector;
657 sector_t e = b + (bio->bi_iter.bi_size >> SECTOR_SHIFT);
658
659 b += pool->sectors_per_block - 1ull; /* so we round up */
660
661 if (block_size_is_power_of_two(pool)) {
662 b >>= pool->sectors_per_block_shift;
663 e >>= pool->sectors_per_block_shift;
664 } else {
665 (void) sector_div(b, pool->sectors_per_block);
666 (void) sector_div(e, pool->sectors_per_block);
667 }
668
669 if (e < b)
670 /* Can happen if the bio is within a single block. */
671 e = b;
672
673 *begin = b;
674 *end = e;
675}
676
Joe Thornber991d9fa2011-10-31 20:21:18 +0000677static void remap(struct thin_c *tc, struct bio *bio, dm_block_t block)
678{
679 struct pool *pool = tc->pool;
Kent Overstreet4f024f32013-10-11 15:44:27 -0700680 sector_t bi_sector = bio->bi_iter.bi_sector;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000681
682 bio->bi_bdev = tc->pool_dev->bdev;
Mike Snitzer58f77a22013-03-01 22:45:45 +0000683 if (block_size_is_power_of_two(pool))
Kent Overstreet4f024f32013-10-11 15:44:27 -0700684 bio->bi_iter.bi_sector =
685 (block << pool->sectors_per_block_shift) |
686 (bi_sector & (pool->sectors_per_block - 1));
Mike Snitzer58f77a22013-03-01 22:45:45 +0000687 else
Kent Overstreet4f024f32013-10-11 15:44:27 -0700688 bio->bi_iter.bi_sector = (block * pool->sectors_per_block) +
Mike Snitzer58f77a22013-03-01 22:45:45 +0000689 sector_div(bi_sector, pool->sectors_per_block);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000690}
691
Joe Thornber2dd9c252012-03-28 18:41:28 +0100692static void remap_to_origin(struct thin_c *tc, struct bio *bio)
693{
694 bio->bi_bdev = tc->origin_dev->bdev;
695}
696
Joe Thornber4afdd682012-07-27 15:08:14 +0100697static int bio_triggers_commit(struct thin_c *tc, struct bio *bio)
698{
699 return (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) &&
700 dm_thin_changed_this_transaction(tc->td);
701}
702
Joe Thornbere8088072012-12-21 20:23:31 +0000703static void inc_all_io_entry(struct pool *pool, struct bio *bio)
704{
705 struct dm_thin_endio_hook *h;
706
707 if (bio->bi_rw & REQ_DISCARD)
708 return;
709
Mikulas Patocka59c3d2c2012-12-21 20:23:40 +0000710 h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
Joe Thornbere8088072012-12-21 20:23:31 +0000711 h->all_io_entry = dm_deferred_entry_inc(pool->all_io_ds);
712}
713
Joe Thornber2dd9c252012-03-28 18:41:28 +0100714static void issue(struct thin_c *tc, struct bio *bio)
Joe Thornber991d9fa2011-10-31 20:21:18 +0000715{
716 struct pool *pool = tc->pool;
717 unsigned long flags;
718
Joe Thornbere49e5822012-07-27 15:08:16 +0100719 if (!bio_triggers_commit(tc, bio)) {
Joe Thornber991d9fa2011-10-31 20:21:18 +0000720 generic_make_request(bio);
Joe Thornbere49e5822012-07-27 15:08:16 +0100721 return;
722 }
723
724 /*
725 * Complete bio with an error if earlier I/O caused changes to
726 * the metadata that can't be committed e.g, due to I/O errors
727 * on the metadata device.
728 */
729 if (dm_thin_aborted_changes(tc->td)) {
730 bio_io_error(bio);
731 return;
732 }
733
734 /*
735 * Batch together any bios that trigger commits and then issue a
736 * single commit for them in process_deferred_bios().
737 */
738 spin_lock_irqsave(&pool->lock, flags);
739 bio_list_add(&pool->deferred_flush_bios, bio);
740 spin_unlock_irqrestore(&pool->lock, flags);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000741}
742
Joe Thornber2dd9c252012-03-28 18:41:28 +0100743static void remap_to_origin_and_issue(struct thin_c *tc, struct bio *bio)
744{
745 remap_to_origin(tc, bio);
746 issue(tc, bio);
747}
748
749static void remap_and_issue(struct thin_c *tc, struct bio *bio,
750 dm_block_t block)
751{
752 remap(tc, bio, block);
753 issue(tc, bio);
754}
755
Joe Thornber991d9fa2011-10-31 20:21:18 +0000756/*----------------------------------------------------------------*/
757
758/*
759 * Bio endio functions.
760 */
Mike Snitzera24c2562012-06-03 00:30:00 +0100761struct dm_thin_new_mapping {
Joe Thornber991d9fa2011-10-31 20:21:18 +0000762 struct list_head list;
763
Mike Snitzer7f214662013-12-17 13:43:31 -0500764 bool pass_discard:1;
Joe Thornber34fbcf62015-04-16 12:58:35 +0100765 bool maybe_shared:1;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000766
Joe Thornber50f3c3e2014-06-13 13:57:09 +0100767 /*
768 * Track quiescing, copying and zeroing preparation actions. When this
769 * counter hits zero the block is prepared and can be inserted into the
770 * btree.
771 */
772 atomic_t prepare_actions;
773
Mike Snitzer7f214662013-12-17 13:43:31 -0500774 int err;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000775 struct thin_c *tc;
Joe Thornber34fbcf62015-04-16 12:58:35 +0100776 dm_block_t virt_begin, virt_end;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000777 dm_block_t data_block;
Joe Thornber34fbcf62015-04-16 12:58:35 +0100778 struct dm_bio_prison_cell *cell;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000779
780 /*
781 * If the bio covers the whole area of a block then we can avoid
782 * zeroing or copying. Instead this bio is hooked. The bio will
783 * still be in the cell, so care has to be taken to avoid issuing
784 * the bio twice.
785 */
786 struct bio *bio;
787 bio_end_io_t *saved_bi_end_io;
788};
789
Joe Thornber50f3c3e2014-06-13 13:57:09 +0100790static void __complete_mapping_preparation(struct dm_thin_new_mapping *m)
Joe Thornber991d9fa2011-10-31 20:21:18 +0000791{
792 struct pool *pool = m->tc->pool;
793
Joe Thornber50f3c3e2014-06-13 13:57:09 +0100794 if (atomic_dec_and_test(&m->prepare_actions)) {
Mike Snitzerdaec3382013-12-11 14:01:20 -0500795 list_add_tail(&m->list, &pool->prepared_mappings);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000796 wake_worker(pool);
797 }
798}
799
Joe Thornbere5aea7b2014-06-13 14:47:24 +0100800static void complete_mapping_preparation(struct dm_thin_new_mapping *m)
Joe Thornber991d9fa2011-10-31 20:21:18 +0000801{
802 unsigned long flags;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000803 struct pool *pool = m->tc->pool;
804
Joe Thornber991d9fa2011-10-31 20:21:18 +0000805 spin_lock_irqsave(&pool->lock, flags);
Joe Thornber50f3c3e2014-06-13 13:57:09 +0100806 __complete_mapping_preparation(m);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000807 spin_unlock_irqrestore(&pool->lock, flags);
808}
809
Joe Thornbere5aea7b2014-06-13 14:47:24 +0100810static void copy_complete(int read_err, unsigned long write_err, void *context)
811{
812 struct dm_thin_new_mapping *m = context;
813
814 m->err = read_err || write_err ? -EIO : 0;
815 complete_mapping_preparation(m);
816}
817
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200818static void overwrite_endio(struct bio *bio)
Joe Thornber991d9fa2011-10-31 20:21:18 +0000819{
Mikulas Patocka59c3d2c2012-12-21 20:23:40 +0000820 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
Mike Snitzera24c2562012-06-03 00:30:00 +0100821 struct dm_thin_new_mapping *m = h->overwrite_mapping;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000822
Mike Snitzer8b908f82015-05-13 17:53:13 -0400823 bio->bi_end_io = m->saved_bi_end_io;
824
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200825 m->err = bio->bi_error;
Joe Thornbere5aea7b2014-06-13 14:47:24 +0100826 complete_mapping_preparation(m);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000827}
828
Joe Thornber991d9fa2011-10-31 20:21:18 +0000829/*----------------------------------------------------------------*/
830
831/*
832 * Workqueue.
833 */
834
835/*
836 * Prepared mapping jobs.
837 */
838
839/*
Joe Thornber2d759a42014-10-10 15:27:16 +0100840 * This sends the bios in the cell, except the original holder, back
841 * to the deferred_bios list.
Joe Thornber991d9fa2011-10-31 20:21:18 +0000842 */
Joe Thornberf286ba02012-12-21 20:23:33 +0000843static void cell_defer_no_holder(struct thin_c *tc, struct dm_bio_prison_cell *cell)
Joe Thornber991d9fa2011-10-31 20:21:18 +0000844{
Joe Thornber991d9fa2011-10-31 20:21:18 +0000845 struct pool *pool = tc->pool;
846 unsigned long flags;
847
Mike Snitzerc140e1c2014-03-20 21:17:14 -0400848 spin_lock_irqsave(&tc->lock, flags);
849 cell_release_no_holder(pool, cell, &tc->deferred_bio_list);
850 spin_unlock_irqrestore(&tc->lock, flags);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000851
852 wake_worker(pool);
853}
854
Joe Thornbera374bb22014-10-10 13:43:14 +0100855static void thin_defer_bio(struct thin_c *tc, struct bio *bio);
856
Joe Thornber2d759a42014-10-10 15:27:16 +0100857struct remap_info {
858 struct thin_c *tc;
859 struct bio_list defer_bios;
860 struct bio_list issue_bios;
861};
862
863static void __inc_remap_and_issue_cell(void *context,
864 struct dm_bio_prison_cell *cell)
865{
866 struct remap_info *info = context;
867 struct bio *bio;
868
869 while ((bio = bio_list_pop(&cell->bios))) {
870 if (bio->bi_rw & (REQ_DISCARD | REQ_FLUSH | REQ_FUA))
871 bio_list_add(&info->defer_bios, bio);
872 else {
873 inc_all_io_entry(info->tc->pool, bio);
874
875 /*
876 * We can't issue the bios with the bio prison lock
877 * held, so we add them to a list to issue on
878 * return from this function.
879 */
880 bio_list_add(&info->issue_bios, bio);
881 }
882 }
883}
884
Joe Thornbera374bb22014-10-10 13:43:14 +0100885static void inc_remap_and_issue_cell(struct thin_c *tc,
886 struct dm_bio_prison_cell *cell,
887 dm_block_t block)
888{
889 struct bio *bio;
Joe Thornber2d759a42014-10-10 15:27:16 +0100890 struct remap_info info;
Joe Thornbera374bb22014-10-10 13:43:14 +0100891
Joe Thornber2d759a42014-10-10 15:27:16 +0100892 info.tc = tc;
893 bio_list_init(&info.defer_bios);
894 bio_list_init(&info.issue_bios);
Joe Thornbera374bb22014-10-10 13:43:14 +0100895
Joe Thornber2d759a42014-10-10 15:27:16 +0100896 /*
897 * We have to be careful to inc any bios we're about to issue
898 * before the cell is released, and avoid a race with new bios
899 * being added to the cell.
900 */
901 cell_visit_release(tc->pool, __inc_remap_and_issue_cell,
902 &info, cell);
903
904 while ((bio = bio_list_pop(&info.defer_bios)))
905 thin_defer_bio(tc, bio);
906
907 while ((bio = bio_list_pop(&info.issue_bios)))
908 remap_and_issue(info.tc, bio, block);
Joe Thornbera374bb22014-10-10 13:43:14 +0100909}
910
Joe Thornbere49e5822012-07-27 15:08:16 +0100911static void process_prepared_mapping_fail(struct dm_thin_new_mapping *m)
912{
Joe Thornber6beca5e2013-03-01 22:45:50 +0000913 cell_error(m->tc->pool, m->cell);
Joe Thornbere49e5822012-07-27 15:08:16 +0100914 list_del(&m->list);
915 mempool_free(m, m->tc->pool->mapping_pool);
916}
Joe Thornber025b9682013-03-01 22:45:50 +0000917
Mike Snitzera24c2562012-06-03 00:30:00 +0100918static void process_prepared_mapping(struct dm_thin_new_mapping *m)
Joe Thornber991d9fa2011-10-31 20:21:18 +0000919{
920 struct thin_c *tc = m->tc;
Joe Thornber6beca5e2013-03-01 22:45:50 +0000921 struct pool *pool = tc->pool;
Mike Snitzer8b908f82015-05-13 17:53:13 -0400922 struct bio *bio = m->bio;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000923 int r;
924
Joe Thornber991d9fa2011-10-31 20:21:18 +0000925 if (m->err) {
Joe Thornber6beca5e2013-03-01 22:45:50 +0000926 cell_error(pool, m->cell);
Joe Thornber905386f2012-07-27 15:08:05 +0100927 goto out;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000928 }
929
930 /*
931 * Commit the prepared block into the mapping btree.
932 * Any I/O for this block arriving after this point will get
933 * remapped to it directly.
934 */
Joe Thornber34fbcf62015-04-16 12:58:35 +0100935 r = dm_thin_insert_block(tc->td, m->virt_begin, m->data_block);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000936 if (r) {
Joe Thornberb5330652013-12-04 19:51:33 -0500937 metadata_operation_failed(pool, "dm_thin_insert_block", r);
Joe Thornber6beca5e2013-03-01 22:45:50 +0000938 cell_error(pool, m->cell);
Joe Thornber905386f2012-07-27 15:08:05 +0100939 goto out;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000940 }
941
942 /*
943 * Release any bios held while the block was being provisioned.
944 * If we are processing a write bio that completely covers the block,
945 * we already processed it so can ignore it now when processing
946 * the bios in the cell.
947 */
948 if (bio) {
Joe Thornber2d759a42014-10-10 15:27:16 +0100949 inc_remap_and_issue_cell(tc, m->cell, m->data_block);
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200950 bio_endio(bio);
Joe Thornber2d759a42014-10-10 15:27:16 +0100951 } else {
952 inc_all_io_entry(tc->pool, m->cell->holder);
953 remap_and_issue(tc, m->cell->holder, m->data_block);
954 inc_remap_and_issue_cell(tc, m->cell, m->data_block);
955 }
Joe Thornber991d9fa2011-10-31 20:21:18 +0000956
Joe Thornber905386f2012-07-27 15:08:05 +0100957out:
Joe Thornber991d9fa2011-10-31 20:21:18 +0000958 list_del(&m->list);
Joe Thornber6beca5e2013-03-01 22:45:50 +0000959 mempool_free(m, pool->mapping_pool);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000960}
961
Joe Thornber34fbcf62015-04-16 12:58:35 +0100962/*----------------------------------------------------------------*/
963
964static void free_discard_mapping(struct dm_thin_new_mapping *m)
965{
966 struct thin_c *tc = m->tc;
967 if (m->cell)
968 cell_defer_no_holder(tc, m->cell);
969 mempool_free(m, tc->pool->mapping_pool);
970}
971
Joe Thornbere49e5822012-07-27 15:08:16 +0100972static void process_prepared_discard_fail(struct dm_thin_new_mapping *m)
Joe Thornber104655f2012-03-28 18:41:28 +0100973{
Joe Thornbere49e5822012-07-27 15:08:16 +0100974 bio_io_error(m->bio);
Joe Thornber34fbcf62015-04-16 12:58:35 +0100975 free_discard_mapping(m);
Joe Thornbere49e5822012-07-27 15:08:16 +0100976}
Joe Thornber104655f2012-03-28 18:41:28 +0100977
Joe Thornber34fbcf62015-04-16 12:58:35 +0100978static void process_prepared_discard_success(struct dm_thin_new_mapping *m)
Joe Thornbere49e5822012-07-27 15:08:16 +0100979{
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200980 bio_endio(m->bio);
Joe Thornber34fbcf62015-04-16 12:58:35 +0100981 free_discard_mapping(m);
Joe Thornber104655f2012-03-28 18:41:28 +0100982}
983
Joe Thornber34fbcf62015-04-16 12:58:35 +0100984static void process_prepared_discard_no_passdown(struct dm_thin_new_mapping *m)
Joe Thornbere49e5822012-07-27 15:08:16 +0100985{
986 int r;
987 struct thin_c *tc = m->tc;
988
Joe Thornber34fbcf62015-04-16 12:58:35 +0100989 r = dm_thin_remove_range(tc->td, m->cell->key.block_begin, m->cell->key.block_end);
990 if (r) {
991 metadata_operation_failed(tc->pool, "dm_thin_remove_range", r);
992 bio_io_error(m->bio);
993 } else
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200994 bio_endio(m->bio);
Joe Thornbere49e5822012-07-27 15:08:16 +0100995
Joe Thornber34fbcf62015-04-16 12:58:35 +0100996 cell_defer_no_holder(tc, m->cell);
997 mempool_free(m, tc->pool->mapping_pool);
998}
999
Joe Thornber202bae52016-05-04 14:12:42 -04001000/*----------------------------------------------------------------*/
1001
1002static void passdown_double_checking_shared_status(struct dm_thin_new_mapping *m)
Joe Thornber34fbcf62015-04-16 12:58:35 +01001003{
1004 /*
1005 * We've already unmapped this range of blocks, but before we
1006 * passdown we have to check that these blocks are now unused.
1007 */
Joe Thornber202bae52016-05-04 14:12:42 -04001008 int r = 0;
Joe Thornber34fbcf62015-04-16 12:58:35 +01001009 bool used = true;
1010 struct thin_c *tc = m->tc;
1011 struct pool *pool = tc->pool;
1012 dm_block_t b = m->data_block, e, end = m->data_block + m->virt_end - m->virt_begin;
Joe Thornber202bae52016-05-04 14:12:42 -04001013 struct discard_op op;
Joe Thornber34fbcf62015-04-16 12:58:35 +01001014
Joe Thornber202bae52016-05-04 14:12:42 -04001015 begin_discard(&op, tc, m->bio);
Joe Thornber34fbcf62015-04-16 12:58:35 +01001016 while (b != end) {
1017 /* find start of unmapped run */
1018 for (; b < end; b++) {
1019 r = dm_pool_block_is_used(pool->pmd, b, &used);
1020 if (r)
Joe Thornber202bae52016-05-04 14:12:42 -04001021 goto out;
Joe Thornber34fbcf62015-04-16 12:58:35 +01001022
1023 if (!used)
1024 break;
1025 }
1026
1027 if (b == end)
1028 break;
1029
1030 /* find end of run */
1031 for (e = b + 1; e != end; e++) {
1032 r = dm_pool_block_is_used(pool->pmd, e, &used);
1033 if (r)
Joe Thornber202bae52016-05-04 14:12:42 -04001034 goto out;
Joe Thornber34fbcf62015-04-16 12:58:35 +01001035
1036 if (used)
1037 break;
1038 }
1039
Joe Thornber202bae52016-05-04 14:12:42 -04001040 r = issue_discard(&op, b, e);
Joe Thornber34fbcf62015-04-16 12:58:35 +01001041 if (r)
Joe Thornber202bae52016-05-04 14:12:42 -04001042 goto out;
Joe Thornber34fbcf62015-04-16 12:58:35 +01001043
1044 b = e;
1045 }
Joe Thornber202bae52016-05-04 14:12:42 -04001046out:
1047 end_discard(&op, r);
Joe Thornber34fbcf62015-04-16 12:58:35 +01001048}
1049
1050static void process_prepared_discard_passdown(struct dm_thin_new_mapping *m)
1051{
1052 int r;
1053 struct thin_c *tc = m->tc;
1054 struct pool *pool = tc->pool;
1055
1056 r = dm_thin_remove_range(tc->td, m->virt_begin, m->virt_end);
Joe Thornber202bae52016-05-04 14:12:42 -04001057 if (r) {
Joe Thornber34fbcf62015-04-16 12:58:35 +01001058 metadata_operation_failed(pool, "dm_thin_remove_range", r);
Joe Thornber202bae52016-05-04 14:12:42 -04001059 bio_io_error(m->bio);
Joe Thornber34fbcf62015-04-16 12:58:35 +01001060
Joe Thornber202bae52016-05-04 14:12:42 -04001061 } else if (m->maybe_shared) {
1062 passdown_double_checking_shared_status(m);
Joe Thornber34fbcf62015-04-16 12:58:35 +01001063
Joe Thornber202bae52016-05-04 14:12:42 -04001064 } else {
1065 struct discard_op op;
1066 begin_discard(&op, tc, m->bio);
1067 r = issue_discard(&op, m->data_block,
1068 m->data_block + (m->virt_end - m->virt_begin));
1069 end_discard(&op, r);
1070 }
1071
Joe Thornber34fbcf62015-04-16 12:58:35 +01001072 cell_defer_no_holder(tc, m->cell);
1073 mempool_free(m, pool->mapping_pool);
Joe Thornbere49e5822012-07-27 15:08:16 +01001074}
1075
Joe Thornber104655f2012-03-28 18:41:28 +01001076static void process_prepared(struct pool *pool, struct list_head *head,
Joe Thornbere49e5822012-07-27 15:08:16 +01001077 process_mapping_fn *fn)
Joe Thornber991d9fa2011-10-31 20:21:18 +00001078{
1079 unsigned long flags;
1080 struct list_head maps;
Mike Snitzera24c2562012-06-03 00:30:00 +01001081 struct dm_thin_new_mapping *m, *tmp;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001082
1083 INIT_LIST_HEAD(&maps);
1084 spin_lock_irqsave(&pool->lock, flags);
Joe Thornber104655f2012-03-28 18:41:28 +01001085 list_splice_init(head, &maps);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001086 spin_unlock_irqrestore(&pool->lock, flags);
1087
1088 list_for_each_entry_safe(m, tmp, &maps, list)
Joe Thornbere49e5822012-07-27 15:08:16 +01001089 (*fn)(m);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001090}
1091
1092/*
1093 * Deferred bio jobs.
1094 */
Joe Thornber104655f2012-03-28 18:41:28 +01001095static int io_overlaps_block(struct pool *pool, struct bio *bio)
1096{
Kent Overstreet4f024f32013-10-11 15:44:27 -07001097 return bio->bi_iter.bi_size ==
1098 (pool->sectors_per_block << SECTOR_SHIFT);
Joe Thornber104655f2012-03-28 18:41:28 +01001099}
1100
Joe Thornber991d9fa2011-10-31 20:21:18 +00001101static int io_overwrites_block(struct pool *pool, struct bio *bio)
1102{
Joe Thornber104655f2012-03-28 18:41:28 +01001103 return (bio_data_dir(bio) == WRITE) &&
1104 io_overlaps_block(pool, bio);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001105}
1106
1107static void save_and_set_endio(struct bio *bio, bio_end_io_t **save,
1108 bio_end_io_t *fn)
1109{
1110 *save = bio->bi_end_io;
1111 bio->bi_end_io = fn;
1112}
1113
1114static int ensure_next_mapping(struct pool *pool)
1115{
1116 if (pool->next_mapping)
1117 return 0;
1118
1119 pool->next_mapping = mempool_alloc(pool->mapping_pool, GFP_ATOMIC);
1120
1121 return pool->next_mapping ? 0 : -ENOMEM;
1122}
1123
Mike Snitzera24c2562012-06-03 00:30:00 +01001124static struct dm_thin_new_mapping *get_next_mapping(struct pool *pool)
Joe Thornber991d9fa2011-10-31 20:21:18 +00001125{
Mike Snitzer16961b02013-12-17 13:19:11 -05001126 struct dm_thin_new_mapping *m = pool->next_mapping;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001127
1128 BUG_ON(!pool->next_mapping);
1129
Mike Snitzer16961b02013-12-17 13:19:11 -05001130 memset(m, 0, sizeof(struct dm_thin_new_mapping));
1131 INIT_LIST_HEAD(&m->list);
1132 m->bio = NULL;
1133
Joe Thornber991d9fa2011-10-31 20:21:18 +00001134 pool->next_mapping = NULL;
1135
Mike Snitzer16961b02013-12-17 13:19:11 -05001136 return m;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001137}
1138
Joe Thornbere5aea7b2014-06-13 14:47:24 +01001139static void ll_zero(struct thin_c *tc, struct dm_thin_new_mapping *m,
1140 sector_t begin, sector_t end)
1141{
1142 int r;
1143 struct dm_io_region to;
1144
1145 to.bdev = tc->pool_dev->bdev;
1146 to.sector = begin;
1147 to.count = end - begin;
1148
1149 r = dm_kcopyd_zero(tc->pool->copier, 1, &to, 0, copy_complete, m);
1150 if (r < 0) {
1151 DMERR_LIMIT("dm_kcopyd_zero() failed");
1152 copy_complete(1, 1, m);
1153 }
1154}
1155
Mike Snitzer452d7a62014-10-09 19:20:21 -04001156static void remap_and_issue_overwrite(struct thin_c *tc, struct bio *bio,
Joe Thornber34fbcf62015-04-16 12:58:35 +01001157 dm_block_t data_begin,
Mike Snitzer452d7a62014-10-09 19:20:21 -04001158 struct dm_thin_new_mapping *m)
1159{
1160 struct pool *pool = tc->pool;
1161 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1162
1163 h->overwrite_mapping = m;
1164 m->bio = bio;
1165 save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio);
1166 inc_all_io_entry(pool, bio);
Joe Thornber34fbcf62015-04-16 12:58:35 +01001167 remap_and_issue(tc, bio, data_begin);
Mike Snitzer452d7a62014-10-09 19:20:21 -04001168}
1169
Joe Thornbere5aea7b2014-06-13 14:47:24 +01001170/*
1171 * A partial copy also needs to zero the uncopied region.
1172 */
Joe Thornber991d9fa2011-10-31 20:21:18 +00001173static void schedule_copy(struct thin_c *tc, dm_block_t virt_block,
Joe Thornber2dd9c252012-03-28 18:41:28 +01001174 struct dm_dev *origin, dm_block_t data_origin,
1175 dm_block_t data_dest,
Joe Thornbere5aea7b2014-06-13 14:47:24 +01001176 struct dm_bio_prison_cell *cell, struct bio *bio,
1177 sector_t len)
Joe Thornber991d9fa2011-10-31 20:21:18 +00001178{
1179 int r;
1180 struct pool *pool = tc->pool;
Mike Snitzera24c2562012-06-03 00:30:00 +01001181 struct dm_thin_new_mapping *m = get_next_mapping(pool);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001182
Joe Thornber991d9fa2011-10-31 20:21:18 +00001183 m->tc = tc;
Joe Thornber34fbcf62015-04-16 12:58:35 +01001184 m->virt_begin = virt_block;
1185 m->virt_end = virt_block + 1u;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001186 m->data_block = data_dest;
1187 m->cell = cell;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001188
Joe Thornbere5aea7b2014-06-13 14:47:24 +01001189 /*
1190 * quiesce action + copy action + an extra reference held for the
1191 * duration of this function (we may need to inc later for a
1192 * partial zero).
1193 */
1194 atomic_set(&m->prepare_actions, 3);
1195
Mike Snitzer44feb382012-10-12 21:02:10 +01001196 if (!dm_deferred_set_add_work(pool->shared_read_ds, &m->list))
Joe Thornbere5aea7b2014-06-13 14:47:24 +01001197 complete_mapping_preparation(m); /* already quiesced */
Joe Thornber991d9fa2011-10-31 20:21:18 +00001198
1199 /*
1200 * IO to pool_dev remaps to the pool target's data_dev.
1201 *
1202 * If the whole block of data is being overwritten, we can issue the
1203 * bio immediately. Otherwise we use kcopyd to clone the data first.
1204 */
Mike Snitzer452d7a62014-10-09 19:20:21 -04001205 if (io_overwrites_block(pool, bio))
1206 remap_and_issue_overwrite(tc, bio, data_dest, m);
1207 else {
Joe Thornber991d9fa2011-10-31 20:21:18 +00001208 struct dm_io_region from, to;
1209
Joe Thornber2dd9c252012-03-28 18:41:28 +01001210 from.bdev = origin->bdev;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001211 from.sector = data_origin * pool->sectors_per_block;
Joe Thornbere5aea7b2014-06-13 14:47:24 +01001212 from.count = len;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001213
1214 to.bdev = tc->pool_dev->bdev;
1215 to.sector = data_dest * pool->sectors_per_block;
Joe Thornbere5aea7b2014-06-13 14:47:24 +01001216 to.count = len;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001217
1218 r = dm_kcopyd_copy(pool->copier, &from, 1, &to,
1219 0, copy_complete, m);
1220 if (r < 0) {
Mike Snitzerc3977412012-12-21 20:23:34 +00001221 DMERR_LIMIT("dm_kcopyd_copy() failed");
Joe Thornbere5aea7b2014-06-13 14:47:24 +01001222 copy_complete(1, 1, m);
1223
1224 /*
1225 * We allow the zero to be issued, to simplify the
1226 * error path. Otherwise we'd need to start
1227 * worrying about decrementing the prepare_actions
1228 * counter.
1229 */
1230 }
1231
1232 /*
1233 * Do we need to zero a tail region?
1234 */
1235 if (len < pool->sectors_per_block && pool->pf.zero_new_blocks) {
1236 atomic_inc(&m->prepare_actions);
1237 ll_zero(tc, m,
1238 data_dest * pool->sectors_per_block + len,
1239 (data_dest + 1) * pool->sectors_per_block);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001240 }
1241 }
Joe Thornbere5aea7b2014-06-13 14:47:24 +01001242
1243 complete_mapping_preparation(m); /* drop our ref */
Joe Thornber991d9fa2011-10-31 20:21:18 +00001244}
1245
Joe Thornber2dd9c252012-03-28 18:41:28 +01001246static void schedule_internal_copy(struct thin_c *tc, dm_block_t virt_block,
1247 dm_block_t data_origin, dm_block_t data_dest,
Mike Snitzera24c2562012-06-03 00:30:00 +01001248 struct dm_bio_prison_cell *cell, struct bio *bio)
Joe Thornber2dd9c252012-03-28 18:41:28 +01001249{
1250 schedule_copy(tc, virt_block, tc->pool_dev,
Joe Thornbere5aea7b2014-06-13 14:47:24 +01001251 data_origin, data_dest, cell, bio,
1252 tc->pool->sectors_per_block);
Joe Thornber2dd9c252012-03-28 18:41:28 +01001253}
1254
Joe Thornber991d9fa2011-10-31 20:21:18 +00001255static void schedule_zero(struct thin_c *tc, dm_block_t virt_block,
Mike Snitzera24c2562012-06-03 00:30:00 +01001256 dm_block_t data_block, struct dm_bio_prison_cell *cell,
Joe Thornber991d9fa2011-10-31 20:21:18 +00001257 struct bio *bio)
1258{
1259 struct pool *pool = tc->pool;
Mike Snitzera24c2562012-06-03 00:30:00 +01001260 struct dm_thin_new_mapping *m = get_next_mapping(pool);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001261
Joe Thornber50f3c3e2014-06-13 13:57:09 +01001262 atomic_set(&m->prepare_actions, 1); /* no need to quiesce */
Joe Thornber991d9fa2011-10-31 20:21:18 +00001263 m->tc = tc;
Joe Thornber34fbcf62015-04-16 12:58:35 +01001264 m->virt_begin = virt_block;
1265 m->virt_end = virt_block + 1u;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001266 m->data_block = data_block;
1267 m->cell = cell;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001268
1269 /*
1270 * If the whole block of data is being overwritten or we are not
1271 * zeroing pre-existing data, we can issue the bio immediately.
1272 * Otherwise we use kcopyd to zero the data first.
1273 */
Mike Snitzerf8ae7522015-05-14 11:28:37 -04001274 if (pool->pf.zero_new_blocks) {
1275 if (io_overwrites_block(pool, bio))
1276 remap_and_issue_overwrite(tc, bio, data_block, m);
1277 else
1278 ll_zero(tc, m, data_block * pool->sectors_per_block,
1279 (data_block + 1) * pool->sectors_per_block);
1280 } else
Joe Thornber991d9fa2011-10-31 20:21:18 +00001281 process_prepared_mapping(m);
Joe Thornbere5aea7b2014-06-13 14:47:24 +01001282}
Joe Thornber991d9fa2011-10-31 20:21:18 +00001283
Joe Thornbere5aea7b2014-06-13 14:47:24 +01001284static void schedule_external_copy(struct thin_c *tc, dm_block_t virt_block,
1285 dm_block_t data_dest,
1286 struct dm_bio_prison_cell *cell, struct bio *bio)
1287{
1288 struct pool *pool = tc->pool;
1289 sector_t virt_block_begin = virt_block * pool->sectors_per_block;
1290 sector_t virt_block_end = (virt_block + 1) * pool->sectors_per_block;
1291
1292 if (virt_block_end <= tc->origin_size)
1293 schedule_copy(tc, virt_block, tc->origin_dev,
1294 virt_block, data_dest, cell, bio,
1295 pool->sectors_per_block);
1296
1297 else if (virt_block_begin < tc->origin_size)
1298 schedule_copy(tc, virt_block, tc->origin_dev,
1299 virt_block, data_dest, cell, bio,
1300 tc->origin_size - virt_block_begin);
1301
1302 else
1303 schedule_zero(tc, virt_block, data_dest, cell, bio);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001304}
1305
Joe Thornber2c43fd22014-12-11 11:12:19 +00001306static void set_pool_mode(struct pool *pool, enum pool_mode new_mode);
1307
1308static void check_for_space(struct pool *pool)
1309{
1310 int r;
1311 dm_block_t nr_free;
1312
1313 if (get_pool_mode(pool) != PM_OUT_OF_DATA_SPACE)
1314 return;
1315
1316 r = dm_pool_get_free_block_count(pool->pmd, &nr_free);
1317 if (r)
1318 return;
1319
1320 if (nr_free)
1321 set_pool_mode(pool, PM_WRITE);
1322}
1323
Joe Thornbere49e5822012-07-27 15:08:16 +01001324/*
1325 * A non-zero return indicates read_only or fail_io mode.
1326 * Many callers don't care about the return value.
1327 */
Joe Thornber020cc3b2013-12-04 15:05:36 -05001328static int commit(struct pool *pool)
Joe Thornbere49e5822012-07-27 15:08:16 +01001329{
1330 int r;
1331
Joe Thornber8d07e8a2014-05-06 16:28:14 +01001332 if (get_pool_mode(pool) >= PM_READ_ONLY)
Joe Thornbere49e5822012-07-27 15:08:16 +01001333 return -EINVAL;
1334
Joe Thornber020cc3b2013-12-04 15:05:36 -05001335 r = dm_pool_commit_metadata(pool->pmd);
Joe Thornberb5330652013-12-04 19:51:33 -05001336 if (r)
1337 metadata_operation_failed(pool, "dm_pool_commit_metadata", r);
Joe Thornber2c43fd22014-12-11 11:12:19 +00001338 else
1339 check_for_space(pool);
Joe Thornbere49e5822012-07-27 15:08:16 +01001340
1341 return r;
1342}
1343
Joe Thornber88a66212013-12-04 20:16:12 -05001344static void check_low_water_mark(struct pool *pool, dm_block_t free_blocks)
1345{
1346 unsigned long flags;
1347
1348 if (free_blocks <= pool->low_water_blocks && !pool->low_water_triggered) {
1349 DMWARN("%s: reached low water mark for data device: sending event.",
1350 dm_device_name(pool->pool_md));
1351 spin_lock_irqsave(&pool->lock, flags);
1352 pool->low_water_triggered = true;
1353 spin_unlock_irqrestore(&pool->lock, flags);
1354 dm_table_event(pool->ti->table);
1355 }
1356}
1357
Joe Thornber991d9fa2011-10-31 20:21:18 +00001358static int alloc_data_block(struct thin_c *tc, dm_block_t *result)
1359{
1360 int r;
1361 dm_block_t free_blocks;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001362 struct pool *pool = tc->pool;
1363
Joe Thornber3e1a0692014-03-03 16:03:26 +00001364 if (WARN_ON(get_pool_mode(pool) != PM_WRITE))
Joe Thornber8d30abf2013-12-04 19:16:11 -05001365 return -EINVAL;
1366
Joe Thornber991d9fa2011-10-31 20:21:18 +00001367 r = dm_pool_get_free_block_count(pool->pmd, &free_blocks);
Joe Thornberb5330652013-12-04 19:51:33 -05001368 if (r) {
1369 metadata_operation_failed(pool, "dm_pool_get_free_block_count", r);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001370 return r;
Joe Thornberb5330652013-12-04 19:51:33 -05001371 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00001372
Joe Thornber88a66212013-12-04 20:16:12 -05001373 check_low_water_mark(pool, free_blocks);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001374
1375 if (!free_blocks) {
Mike Snitzer94563ba2013-08-22 09:56:18 -04001376 /*
1377 * Try to commit to see if that will free up some
1378 * more space.
1379 */
Joe Thornber020cc3b2013-12-04 15:05:36 -05001380 r = commit(pool);
1381 if (r)
1382 return r;
Mike Snitzer94563ba2013-08-22 09:56:18 -04001383
1384 r = dm_pool_get_free_block_count(pool->pmd, &free_blocks);
Joe Thornberb5330652013-12-04 19:51:33 -05001385 if (r) {
1386 metadata_operation_failed(pool, "dm_pool_get_free_block_count", r);
Mike Snitzer94563ba2013-08-22 09:56:18 -04001387 return r;
Joe Thornberb5330652013-12-04 19:51:33 -05001388 }
Mike Snitzer94563ba2013-08-22 09:56:18 -04001389
Mike Snitzer94563ba2013-08-22 09:56:18 -04001390 if (!free_blocks) {
Joe Thornber3e1a0692014-03-03 16:03:26 +00001391 set_pool_mode(pool, PM_OUT_OF_DATA_SPACE);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001392 return -ENOSPC;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001393 }
1394 }
1395
1396 r = dm_pool_alloc_data_block(pool->pmd, result);
Mike Snitzer4a02b342013-12-03 12:20:57 -05001397 if (r) {
Joe Thornberb5330652013-12-04 19:51:33 -05001398 metadata_operation_failed(pool, "dm_pool_alloc_data_block", r);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001399 return r;
Mike Snitzer4a02b342013-12-03 12:20:57 -05001400 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00001401
1402 return 0;
1403}
1404
1405/*
1406 * If we have run out of space, queue bios until the device is
1407 * resumed, presumably after having been reloaded with more space.
1408 */
1409static void retry_on_resume(struct bio *bio)
1410{
Mikulas Patocka59c3d2c2012-12-21 20:23:40 +00001411 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
Joe Thornbereb2aa482012-03-28 18:41:28 +01001412 struct thin_c *tc = h->tc;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001413 unsigned long flags;
1414
Mike Snitzerc140e1c2014-03-20 21:17:14 -04001415 spin_lock_irqsave(&tc->lock, flags);
1416 bio_list_add(&tc->retry_on_resume_list, bio);
1417 spin_unlock_irqrestore(&tc->lock, flags);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001418}
1419
Mike Snitzeraf918052014-05-22 14:32:51 -04001420static int should_error_unserviceable_bio(struct pool *pool)
Joe Thornber3e1a0692014-03-03 16:03:26 +00001421{
1422 enum pool_mode m = get_pool_mode(pool);
1423
1424 switch (m) {
1425 case PM_WRITE:
1426 /* Shouldn't get here */
1427 DMERR_LIMIT("bio unserviceable, yet pool is in PM_WRITE mode");
Mike Snitzeraf918052014-05-22 14:32:51 -04001428 return -EIO;
Joe Thornber3e1a0692014-03-03 16:03:26 +00001429
1430 case PM_OUT_OF_DATA_SPACE:
Mike Snitzeraf918052014-05-22 14:32:51 -04001431 return pool->pf.error_if_no_space ? -ENOSPC : 0;
Joe Thornber3e1a0692014-03-03 16:03:26 +00001432
1433 case PM_READ_ONLY:
1434 case PM_FAIL:
Mike Snitzeraf918052014-05-22 14:32:51 -04001435 return -EIO;
Joe Thornber3e1a0692014-03-03 16:03:26 +00001436 default:
1437 /* Shouldn't get here */
1438 DMERR_LIMIT("bio unserviceable, yet pool has an unknown mode");
Mike Snitzeraf918052014-05-22 14:32:51 -04001439 return -EIO;
Joe Thornber3e1a0692014-03-03 16:03:26 +00001440 }
1441}
1442
Mike Snitzer8c0f0e82013-12-05 15:47:24 -05001443static void handle_unserviceable_bio(struct pool *pool, struct bio *bio)
1444{
Mike Snitzeraf918052014-05-22 14:32:51 -04001445 int error = should_error_unserviceable_bio(pool);
1446
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001447 if (error) {
1448 bio->bi_error = error;
1449 bio_endio(bio);
1450 } else
Mike Snitzer6d162022013-12-20 18:09:02 -05001451 retry_on_resume(bio);
Mike Snitzer8c0f0e82013-12-05 15:47:24 -05001452}
1453
Mike Snitzer399cadd2013-12-05 16:03:33 -05001454static void retry_bios_on_resume(struct pool *pool, struct dm_bio_prison_cell *cell)
Joe Thornber991d9fa2011-10-31 20:21:18 +00001455{
1456 struct bio *bio;
1457 struct bio_list bios;
Mike Snitzeraf918052014-05-22 14:32:51 -04001458 int error;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001459
Mike Snitzeraf918052014-05-22 14:32:51 -04001460 error = should_error_unserviceable_bio(pool);
1461 if (error) {
1462 cell_error_with_code(pool, cell, error);
Joe Thornber3e1a0692014-03-03 16:03:26 +00001463 return;
1464 }
1465
Joe Thornber991d9fa2011-10-31 20:21:18 +00001466 bio_list_init(&bios);
Joe Thornber6beca5e2013-03-01 22:45:50 +00001467 cell_release(pool, cell, &bios);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001468
Mike Snitzer9d094ee2014-10-19 08:23:09 -04001469 while ((bio = bio_list_pop(&bios)))
1470 retry_on_resume(bio);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001471}
1472
Joe Thornber34fbcf62015-04-16 12:58:35 +01001473static void process_discard_cell_no_passdown(struct thin_c *tc,
1474 struct dm_bio_prison_cell *virt_cell)
Joe Thornber104655f2012-03-28 18:41:28 +01001475{
Joe Thornber104655f2012-03-28 18:41:28 +01001476 struct pool *pool = tc->pool;
Joe Thornber34fbcf62015-04-16 12:58:35 +01001477 struct dm_thin_new_mapping *m = get_next_mapping(pool);
1478
1479 /*
1480 * We don't need to lock the data blocks, since there's no
1481 * passdown. We only lock data blocks for allocation and breaking sharing.
1482 */
1483 m->tc = tc;
1484 m->virt_begin = virt_cell->key.block_begin;
1485 m->virt_end = virt_cell->key.block_end;
1486 m->cell = virt_cell;
1487 m->bio = virt_cell->holder;
1488
1489 if (!dm_deferred_set_add_work(pool->all_io_ds, &m->list))
1490 pool->process_prepared_discard(m);
1491}
1492
Joe Thornber34fbcf62015-04-16 12:58:35 +01001493static void break_up_discard_bio(struct thin_c *tc, dm_block_t begin, dm_block_t end,
1494 struct bio *bio)
1495{
1496 struct pool *pool = tc->pool;
1497
1498 int r;
1499 bool maybe_shared;
1500 struct dm_cell_key data_key;
1501 struct dm_bio_prison_cell *data_cell;
Mike Snitzera24c2562012-06-03 00:30:00 +01001502 struct dm_thin_new_mapping *m;
Joe Thornber34fbcf62015-04-16 12:58:35 +01001503 dm_block_t virt_begin, virt_end, data_begin;
Joe Thornber104655f2012-03-28 18:41:28 +01001504
Joe Thornber34fbcf62015-04-16 12:58:35 +01001505 while (begin != end) {
1506 r = ensure_next_mapping(pool);
1507 if (r)
1508 /* we did our best */
1509 return;
Joe Thornber104655f2012-03-28 18:41:28 +01001510
Joe Thornber34fbcf62015-04-16 12:58:35 +01001511 r = dm_thin_find_mapped_range(tc->td, begin, end, &virt_begin, &virt_end,
1512 &data_begin, &maybe_shared);
1513 if (r)
1514 /*
1515 * Silently fail, letting any mappings we've
1516 * created complete.
1517 */
Joe Thornber104655f2012-03-28 18:41:28 +01001518 break;
Joe Thornber34fbcf62015-04-16 12:58:35 +01001519
1520 build_key(tc->td, PHYSICAL, data_begin, data_begin + (virt_end - virt_begin), &data_key);
1521 if (bio_detain(tc->pool, &data_key, NULL, &data_cell)) {
1522 /* contention, we'll give up with this range */
1523 begin = virt_end;
1524 continue;
Joe Thornber104655f2012-03-28 18:41:28 +01001525 }
1526
Joe Thornber104655f2012-03-28 18:41:28 +01001527 /*
Joe Thornber34fbcf62015-04-16 12:58:35 +01001528 * IO may still be going to the destination block. We must
1529 * quiesce before we can do the removal.
Joe Thornber104655f2012-03-28 18:41:28 +01001530 */
Joe Thornber34fbcf62015-04-16 12:58:35 +01001531 m = get_next_mapping(pool);
1532 m->tc = tc;
1533 m->maybe_shared = maybe_shared;
1534 m->virt_begin = virt_begin;
1535 m->virt_end = virt_end;
1536 m->data_block = data_begin;
1537 m->cell = data_cell;
1538 m->bio = bio;
Joe Thornber104655f2012-03-28 18:41:28 +01001539
Joe Thornber34fbcf62015-04-16 12:58:35 +01001540 /*
1541 * The parent bio must not complete before sub discard bios are
Joe Thornber202bae52016-05-04 14:12:42 -04001542 * chained to it (see end_discard's bio_chain)!
Joe Thornber34fbcf62015-04-16 12:58:35 +01001543 *
1544 * This per-mapping bi_remaining increment is paired with
1545 * the implicit decrement that occurs via bio_endio() in
Joe Thornber202bae52016-05-04 14:12:42 -04001546 * end_discard().
Joe Thornber34fbcf62015-04-16 12:58:35 +01001547 */
Mike Snitzer13e4f8a2016-05-04 15:05:44 -04001548 bio_inc_remaining(bio);
Joe Thornber34fbcf62015-04-16 12:58:35 +01001549 if (!dm_deferred_set_add_work(pool->all_io_ds, &m->list))
1550 pool->process_prepared_discard(m);
1551
1552 begin = virt_end;
Joe Thornber104655f2012-03-28 18:41:28 +01001553 }
1554}
1555
Joe Thornber34fbcf62015-04-16 12:58:35 +01001556static void process_discard_cell_passdown(struct thin_c *tc, struct dm_bio_prison_cell *virt_cell)
1557{
1558 struct bio *bio = virt_cell->holder;
1559 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1560
1561 /*
1562 * The virt_cell will only get freed once the origin bio completes.
1563 * This means it will remain locked while all the individual
1564 * passdown bios are in flight.
1565 */
1566 h->cell = virt_cell;
1567 break_up_discard_bio(tc, virt_cell->key.block_begin, virt_cell->key.block_end, bio);
1568
1569 /*
1570 * We complete the bio now, knowing that the bi_remaining field
1571 * will prevent completion until the sub range discards have
1572 * completed.
1573 */
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001574 bio_endio(bio);
Joe Thornber34fbcf62015-04-16 12:58:35 +01001575}
1576
Joe Thornbera374bb22014-10-10 13:43:14 +01001577static void process_discard_bio(struct thin_c *tc, struct bio *bio)
1578{
Joe Thornber34fbcf62015-04-16 12:58:35 +01001579 dm_block_t begin, end;
1580 struct dm_cell_key virt_key;
1581 struct dm_bio_prison_cell *virt_cell;
Joe Thornbera374bb22014-10-10 13:43:14 +01001582
Joe Thornber34fbcf62015-04-16 12:58:35 +01001583 get_bio_block_range(tc, bio, &begin, &end);
1584 if (begin == end) {
1585 /*
1586 * The discard covers less than a block.
1587 */
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001588 bio_endio(bio);
Joe Thornber34fbcf62015-04-16 12:58:35 +01001589 return;
1590 }
1591
1592 build_key(tc->td, VIRTUAL, begin, end, &virt_key);
1593 if (bio_detain(tc->pool, &virt_key, bio, &virt_cell))
1594 /*
1595 * Potential starvation issue: We're relying on the
1596 * fs/application being well behaved, and not trying to
1597 * send IO to a region at the same time as discarding it.
1598 * If they do this persistently then it's possible this
1599 * cell will never be granted.
1600 */
Joe Thornbera374bb22014-10-10 13:43:14 +01001601 return;
1602
Joe Thornber34fbcf62015-04-16 12:58:35 +01001603 tc->pool->process_discard_cell(tc, virt_cell);
Joe Thornbera374bb22014-10-10 13:43:14 +01001604}
1605
Joe Thornber991d9fa2011-10-31 20:21:18 +00001606static void break_sharing(struct thin_c *tc, struct bio *bio, dm_block_t block,
Mike Snitzer44feb382012-10-12 21:02:10 +01001607 struct dm_cell_key *key,
Joe Thornber991d9fa2011-10-31 20:21:18 +00001608 struct dm_thin_lookup_result *lookup_result,
Mike Snitzera24c2562012-06-03 00:30:00 +01001609 struct dm_bio_prison_cell *cell)
Joe Thornber991d9fa2011-10-31 20:21:18 +00001610{
1611 int r;
1612 dm_block_t data_block;
Mike Snitzerd6fc2042013-08-21 17:40:11 -04001613 struct pool *pool = tc->pool;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001614
1615 r = alloc_data_block(tc, &data_block);
1616 switch (r) {
1617 case 0:
Joe Thornber2dd9c252012-03-28 18:41:28 +01001618 schedule_internal_copy(tc, block, lookup_result->block,
1619 data_block, cell, bio);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001620 break;
1621
1622 case -ENOSPC:
Mike Snitzer399cadd2013-12-05 16:03:33 -05001623 retry_bios_on_resume(pool, cell);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001624 break;
1625
1626 default:
Mike Snitzerc3977412012-12-21 20:23:34 +00001627 DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1628 __func__, r);
Mike Snitzerd6fc2042013-08-21 17:40:11 -04001629 cell_error(pool, cell);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001630 break;
1631 }
1632}
1633
Joe Thornber23ca2bb2014-10-15 14:46:58 +01001634static void __remap_and_issue_shared_cell(void *context,
1635 struct dm_bio_prison_cell *cell)
1636{
1637 struct remap_info *info = context;
1638 struct bio *bio;
1639
1640 while ((bio = bio_list_pop(&cell->bios))) {
1641 if ((bio_data_dir(bio) == WRITE) ||
1642 (bio->bi_rw & (REQ_DISCARD | REQ_FLUSH | REQ_FUA)))
1643 bio_list_add(&info->defer_bios, bio);
1644 else {
1645 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));;
1646
1647 h->shared_read_entry = dm_deferred_entry_inc(info->tc->pool->shared_read_ds);
1648 inc_all_io_entry(info->tc->pool, bio);
1649 bio_list_add(&info->issue_bios, bio);
1650 }
1651 }
1652}
1653
1654static void remap_and_issue_shared_cell(struct thin_c *tc,
1655 struct dm_bio_prison_cell *cell,
1656 dm_block_t block)
1657{
1658 struct bio *bio;
1659 struct remap_info info;
1660
1661 info.tc = tc;
1662 bio_list_init(&info.defer_bios);
1663 bio_list_init(&info.issue_bios);
1664
1665 cell_visit_release(tc->pool, __remap_and_issue_shared_cell,
1666 &info, cell);
1667
1668 while ((bio = bio_list_pop(&info.defer_bios)))
1669 thin_defer_bio(tc, bio);
1670
1671 while ((bio = bio_list_pop(&info.issue_bios)))
1672 remap_and_issue(tc, bio, block);
1673}
1674
Joe Thornber991d9fa2011-10-31 20:21:18 +00001675static void process_shared_bio(struct thin_c *tc, struct bio *bio,
1676 dm_block_t block,
Joe Thornber23ca2bb2014-10-15 14:46:58 +01001677 struct dm_thin_lookup_result *lookup_result,
1678 struct dm_bio_prison_cell *virt_cell)
Joe Thornber991d9fa2011-10-31 20:21:18 +00001679{
Joe Thornber23ca2bb2014-10-15 14:46:58 +01001680 struct dm_bio_prison_cell *data_cell;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001681 struct pool *pool = tc->pool;
Mike Snitzer44feb382012-10-12 21:02:10 +01001682 struct dm_cell_key key;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001683
1684 /*
1685 * If cell is already occupied, then sharing is already in the process
1686 * of being broken so we have nothing further to do here.
1687 */
1688 build_data_key(tc->td, lookup_result->block, &key);
Joe Thornber23ca2bb2014-10-15 14:46:58 +01001689 if (bio_detain(pool, &key, bio, &data_cell)) {
1690 cell_defer_no_holder(tc, virt_cell);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001691 return;
Joe Thornber23ca2bb2014-10-15 14:46:58 +01001692 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00001693
Joe Thornber23ca2bb2014-10-15 14:46:58 +01001694 if (bio_data_dir(bio) == WRITE && bio->bi_iter.bi_size) {
1695 break_sharing(tc, bio, block, &key, lookup_result, data_cell);
1696 cell_defer_no_holder(tc, virt_cell);
1697 } else {
Mikulas Patocka59c3d2c2012-12-21 20:23:40 +00001698 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
Joe Thornber991d9fa2011-10-31 20:21:18 +00001699
Mike Snitzer44feb382012-10-12 21:02:10 +01001700 h->shared_read_entry = dm_deferred_entry_inc(pool->shared_read_ds);
Joe Thornbere8088072012-12-21 20:23:31 +00001701 inc_all_io_entry(pool, bio);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001702 remap_and_issue(tc, bio, lookup_result->block);
Joe Thornber23ca2bb2014-10-15 14:46:58 +01001703
1704 remap_and_issue_shared_cell(tc, data_cell, lookup_result->block);
1705 remap_and_issue_shared_cell(tc, virt_cell, lookup_result->block);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001706 }
1707}
1708
1709static void provision_block(struct thin_c *tc, struct bio *bio, dm_block_t block,
Mike Snitzera24c2562012-06-03 00:30:00 +01001710 struct dm_bio_prison_cell *cell)
Joe Thornber991d9fa2011-10-31 20:21:18 +00001711{
1712 int r;
1713 dm_block_t data_block;
Joe Thornber6beca5e2013-03-01 22:45:50 +00001714 struct pool *pool = tc->pool;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001715
1716 /*
1717 * Remap empty bios (flushes) immediately, without provisioning.
1718 */
Kent Overstreet4f024f32013-10-11 15:44:27 -07001719 if (!bio->bi_iter.bi_size) {
Joe Thornber6beca5e2013-03-01 22:45:50 +00001720 inc_all_io_entry(pool, bio);
Joe Thornberf286ba02012-12-21 20:23:33 +00001721 cell_defer_no_holder(tc, cell);
Joe Thornbere8088072012-12-21 20:23:31 +00001722
Joe Thornber991d9fa2011-10-31 20:21:18 +00001723 remap_and_issue(tc, bio, 0);
1724 return;
1725 }
1726
1727 /*
1728 * Fill read bios with zeroes and complete them immediately.
1729 */
1730 if (bio_data_dir(bio) == READ) {
1731 zero_fill_bio(bio);
Joe Thornberf286ba02012-12-21 20:23:33 +00001732 cell_defer_no_holder(tc, cell);
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001733 bio_endio(bio);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001734 return;
1735 }
1736
1737 r = alloc_data_block(tc, &data_block);
1738 switch (r) {
1739 case 0:
Joe Thornber2dd9c252012-03-28 18:41:28 +01001740 if (tc->origin_dev)
1741 schedule_external_copy(tc, block, data_block, cell, bio);
1742 else
1743 schedule_zero(tc, block, data_block, cell, bio);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001744 break;
1745
1746 case -ENOSPC:
Mike Snitzer399cadd2013-12-05 16:03:33 -05001747 retry_bios_on_resume(pool, cell);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001748 break;
1749
1750 default:
Mike Snitzerc3977412012-12-21 20:23:34 +00001751 DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1752 __func__, r);
Joe Thornber6beca5e2013-03-01 22:45:50 +00001753 cell_error(pool, cell);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001754 break;
1755 }
1756}
1757
Joe Thornbera374bb22014-10-10 13:43:14 +01001758static void process_cell(struct thin_c *tc, struct dm_bio_prison_cell *cell)
Joe Thornber991d9fa2011-10-31 20:21:18 +00001759{
1760 int r;
Joe Thornber6beca5e2013-03-01 22:45:50 +00001761 struct pool *pool = tc->pool;
Joe Thornbera374bb22014-10-10 13:43:14 +01001762 struct bio *bio = cell->holder;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001763 dm_block_t block = get_bio_block(tc, bio);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001764 struct dm_thin_lookup_result lookup_result;
1765
Joe Thornbera374bb22014-10-10 13:43:14 +01001766 if (tc->requeue_mode) {
1767 cell_requeue(pool, cell);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001768 return;
Joe Thornbera374bb22014-10-10 13:43:14 +01001769 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00001770
1771 r = dm_thin_find_block(tc->td, block, 1, &lookup_result);
1772 switch (r) {
1773 case 0:
Joe Thornber23ca2bb2014-10-15 14:46:58 +01001774 if (lookup_result.shared)
1775 process_shared_bio(tc, bio, block, &lookup_result, cell);
1776 else {
Joe Thornber6beca5e2013-03-01 22:45:50 +00001777 inc_all_io_entry(pool, bio);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001778 remap_and_issue(tc, bio, lookup_result.block);
Joe Thornbera374bb22014-10-10 13:43:14 +01001779 inc_remap_and_issue_cell(tc, cell, lookup_result.block);
Joe Thornbere8088072012-12-21 20:23:31 +00001780 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00001781 break;
1782
1783 case -ENODATA:
Joe Thornber2dd9c252012-03-28 18:41:28 +01001784 if (bio_data_dir(bio) == READ && tc->origin_dev) {
Joe Thornber6beca5e2013-03-01 22:45:50 +00001785 inc_all_io_entry(pool, bio);
Joe Thornberf286ba02012-12-21 20:23:33 +00001786 cell_defer_no_holder(tc, cell);
Joe Thornbere8088072012-12-21 20:23:31 +00001787
Joe Thornbere5aea7b2014-06-13 14:47:24 +01001788 if (bio_end_sector(bio) <= tc->origin_size)
1789 remap_to_origin_and_issue(tc, bio);
1790
1791 else if (bio->bi_iter.bi_sector < tc->origin_size) {
1792 zero_fill_bio(bio);
1793 bio->bi_iter.bi_size = (tc->origin_size - bio->bi_iter.bi_sector) << SECTOR_SHIFT;
1794 remap_to_origin_and_issue(tc, bio);
1795
1796 } else {
1797 zero_fill_bio(bio);
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001798 bio_endio(bio);
Joe Thornbere5aea7b2014-06-13 14:47:24 +01001799 }
Joe Thornber2dd9c252012-03-28 18:41:28 +01001800 } else
1801 provision_block(tc, bio, block, cell);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001802 break;
1803
1804 default:
Mike Snitzerc3977412012-12-21 20:23:34 +00001805 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1806 __func__, r);
Joe Thornberf286ba02012-12-21 20:23:33 +00001807 cell_defer_no_holder(tc, cell);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001808 bio_io_error(bio);
1809 break;
1810 }
1811}
1812
Joe Thornbera374bb22014-10-10 13:43:14 +01001813static void process_bio(struct thin_c *tc, struct bio *bio)
1814{
1815 struct pool *pool = tc->pool;
1816 dm_block_t block = get_bio_block(tc, bio);
1817 struct dm_bio_prison_cell *cell;
1818 struct dm_cell_key key;
1819
1820 /*
1821 * If cell is already occupied, then the block is already
1822 * being provisioned so we have nothing further to do here.
1823 */
1824 build_virtual_key(tc->td, block, &key);
1825 if (bio_detain(pool, &key, bio, &cell))
1826 return;
1827
1828 process_cell(tc, cell);
1829}
1830
1831static void __process_bio_read_only(struct thin_c *tc, struct bio *bio,
1832 struct dm_bio_prison_cell *cell)
Joe Thornbere49e5822012-07-27 15:08:16 +01001833{
1834 int r;
1835 int rw = bio_data_dir(bio);
1836 dm_block_t block = get_bio_block(tc, bio);
1837 struct dm_thin_lookup_result lookup_result;
1838
1839 r = dm_thin_find_block(tc->td, block, 1, &lookup_result);
1840 switch (r) {
1841 case 0:
Joe Thornbera374bb22014-10-10 13:43:14 +01001842 if (lookup_result.shared && (rw == WRITE) && bio->bi_iter.bi_size) {
Mike Snitzer8c0f0e82013-12-05 15:47:24 -05001843 handle_unserviceable_bio(tc->pool, bio);
Joe Thornbera374bb22014-10-10 13:43:14 +01001844 if (cell)
1845 cell_defer_no_holder(tc, cell);
1846 } else {
Joe Thornbere8088072012-12-21 20:23:31 +00001847 inc_all_io_entry(tc->pool, bio);
Joe Thornbere49e5822012-07-27 15:08:16 +01001848 remap_and_issue(tc, bio, lookup_result.block);
Joe Thornbera374bb22014-10-10 13:43:14 +01001849 if (cell)
1850 inc_remap_and_issue_cell(tc, cell, lookup_result.block);
Joe Thornbere8088072012-12-21 20:23:31 +00001851 }
Joe Thornbere49e5822012-07-27 15:08:16 +01001852 break;
1853
1854 case -ENODATA:
Joe Thornbera374bb22014-10-10 13:43:14 +01001855 if (cell)
1856 cell_defer_no_holder(tc, cell);
Joe Thornbere49e5822012-07-27 15:08:16 +01001857 if (rw != READ) {
Mike Snitzer8c0f0e82013-12-05 15:47:24 -05001858 handle_unserviceable_bio(tc->pool, bio);
Joe Thornbere49e5822012-07-27 15:08:16 +01001859 break;
1860 }
1861
1862 if (tc->origin_dev) {
Joe Thornbere8088072012-12-21 20:23:31 +00001863 inc_all_io_entry(tc->pool, bio);
Joe Thornbere49e5822012-07-27 15:08:16 +01001864 remap_to_origin_and_issue(tc, bio);
1865 break;
1866 }
1867
1868 zero_fill_bio(bio);
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001869 bio_endio(bio);
Joe Thornbere49e5822012-07-27 15:08:16 +01001870 break;
1871
1872 default:
Mike Snitzerc3977412012-12-21 20:23:34 +00001873 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1874 __func__, r);
Joe Thornbera374bb22014-10-10 13:43:14 +01001875 if (cell)
1876 cell_defer_no_holder(tc, cell);
Joe Thornbere49e5822012-07-27 15:08:16 +01001877 bio_io_error(bio);
1878 break;
1879 }
1880}
1881
Joe Thornbera374bb22014-10-10 13:43:14 +01001882static void process_bio_read_only(struct thin_c *tc, struct bio *bio)
1883{
1884 __process_bio_read_only(tc, bio, NULL);
1885}
1886
1887static void process_cell_read_only(struct thin_c *tc, struct dm_bio_prison_cell *cell)
1888{
1889 __process_bio_read_only(tc, cell->holder, cell);
1890}
1891
Joe Thornber3e1a0692014-03-03 16:03:26 +00001892static void process_bio_success(struct thin_c *tc, struct bio *bio)
1893{
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001894 bio_endio(bio);
Joe Thornber3e1a0692014-03-03 16:03:26 +00001895}
1896
Joe Thornbere49e5822012-07-27 15:08:16 +01001897static void process_bio_fail(struct thin_c *tc, struct bio *bio)
1898{
1899 bio_io_error(bio);
1900}
1901
Joe Thornbera374bb22014-10-10 13:43:14 +01001902static void process_cell_success(struct thin_c *tc, struct dm_bio_prison_cell *cell)
1903{
1904 cell_success(tc->pool, cell);
1905}
1906
1907static void process_cell_fail(struct thin_c *tc, struct dm_bio_prison_cell *cell)
1908{
1909 cell_error(tc->pool, cell);
1910}
1911
Joe Thornberac8c3f32013-05-10 14:37:21 +01001912/*
1913 * FIXME: should we also commit due to size of transaction, measured in
1914 * metadata blocks?
1915 */
Joe Thornber905e51b2012-03-28 18:41:27 +01001916static int need_commit_due_to_time(struct pool *pool)
1917{
Manuel Schölling0f30af92014-05-22 22:42:37 +02001918 return !time_in_range(jiffies, pool->last_commit_jiffies,
1919 pool->last_commit_jiffies + COMMIT_PERIOD);
Joe Thornber905e51b2012-03-28 18:41:27 +01001920}
1921
Mike Snitzer67324ea2014-03-21 18:33:41 -04001922#define thin_pbd(node) rb_entry((node), struct dm_thin_endio_hook, rb_node)
1923#define thin_bio(pbd) dm_bio_from_per_bio_data((pbd), sizeof(struct dm_thin_endio_hook))
1924
1925static void __thin_bio_rb_add(struct thin_c *tc, struct bio *bio)
1926{
1927 struct rb_node **rbp, *parent;
1928 struct dm_thin_endio_hook *pbd;
1929 sector_t bi_sector = bio->bi_iter.bi_sector;
1930
1931 rbp = &tc->sort_bio_list.rb_node;
1932 parent = NULL;
1933 while (*rbp) {
1934 parent = *rbp;
1935 pbd = thin_pbd(parent);
1936
1937 if (bi_sector < thin_bio(pbd)->bi_iter.bi_sector)
1938 rbp = &(*rbp)->rb_left;
1939 else
1940 rbp = &(*rbp)->rb_right;
1941 }
1942
1943 pbd = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1944 rb_link_node(&pbd->rb_node, parent, rbp);
1945 rb_insert_color(&pbd->rb_node, &tc->sort_bio_list);
1946}
1947
1948static void __extract_sorted_bios(struct thin_c *tc)
1949{
1950 struct rb_node *node;
1951 struct dm_thin_endio_hook *pbd;
1952 struct bio *bio;
1953
1954 for (node = rb_first(&tc->sort_bio_list); node; node = rb_next(node)) {
1955 pbd = thin_pbd(node);
1956 bio = thin_bio(pbd);
1957
1958 bio_list_add(&tc->deferred_bio_list, bio);
1959 rb_erase(&pbd->rb_node, &tc->sort_bio_list);
1960 }
1961
1962 WARN_ON(!RB_EMPTY_ROOT(&tc->sort_bio_list));
1963}
1964
1965static void __sort_thin_deferred_bios(struct thin_c *tc)
1966{
1967 struct bio *bio;
1968 struct bio_list bios;
1969
1970 bio_list_init(&bios);
1971 bio_list_merge(&bios, &tc->deferred_bio_list);
1972 bio_list_init(&tc->deferred_bio_list);
1973
1974 /* Sort deferred_bio_list using rb-tree */
1975 while ((bio = bio_list_pop(&bios)))
1976 __thin_bio_rb_add(tc, bio);
1977
1978 /*
1979 * Transfer the sorted bios in sort_bio_list back to
1980 * deferred_bio_list to allow lockless submission of
1981 * all bios.
1982 */
1983 __extract_sorted_bios(tc);
1984}
1985
Mike Snitzerc140e1c2014-03-20 21:17:14 -04001986static void process_thin_deferred_bios(struct thin_c *tc)
Joe Thornber991d9fa2011-10-31 20:21:18 +00001987{
Mike Snitzerc140e1c2014-03-20 21:17:14 -04001988 struct pool *pool = tc->pool;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001989 unsigned long flags;
1990 struct bio *bio;
1991 struct bio_list bios;
Mike Snitzer67324ea2014-03-21 18:33:41 -04001992 struct blk_plug plug;
Joe Thornber8a01a6a2014-10-06 15:28:30 +01001993 unsigned count = 0;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001994
Mike Snitzerc140e1c2014-03-20 21:17:14 -04001995 if (tc->requeue_mode) {
Mike Snitzer42d6a8c2014-10-19 07:52:44 -04001996 error_thin_bio_list(tc, &tc->deferred_bio_list, DM_ENDIO_REQUEUE);
Mike Snitzerc140e1c2014-03-20 21:17:14 -04001997 return;
1998 }
1999
Joe Thornber991d9fa2011-10-31 20:21:18 +00002000 bio_list_init(&bios);
2001
Mike Snitzerc140e1c2014-03-20 21:17:14 -04002002 spin_lock_irqsave(&tc->lock, flags);
Mike Snitzer67324ea2014-03-21 18:33:41 -04002003
2004 if (bio_list_empty(&tc->deferred_bio_list)) {
2005 spin_unlock_irqrestore(&tc->lock, flags);
2006 return;
2007 }
2008
2009 __sort_thin_deferred_bios(tc);
2010
Mike Snitzerc140e1c2014-03-20 21:17:14 -04002011 bio_list_merge(&bios, &tc->deferred_bio_list);
2012 bio_list_init(&tc->deferred_bio_list);
Mike Snitzer67324ea2014-03-21 18:33:41 -04002013
Mike Snitzerc140e1c2014-03-20 21:17:14 -04002014 spin_unlock_irqrestore(&tc->lock, flags);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002015
Mike Snitzer67324ea2014-03-21 18:33:41 -04002016 blk_start_plug(&plug);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002017 while ((bio = bio_list_pop(&bios))) {
Joe Thornber991d9fa2011-10-31 20:21:18 +00002018 /*
2019 * If we've got no free new_mapping structs, and processing
2020 * this bio might require one, we pause until there are some
2021 * prepared mappings to process.
2022 */
2023 if (ensure_next_mapping(pool)) {
Mike Snitzerc140e1c2014-03-20 21:17:14 -04002024 spin_lock_irqsave(&tc->lock, flags);
2025 bio_list_add(&tc->deferred_bio_list, bio);
2026 bio_list_merge(&tc->deferred_bio_list, &bios);
2027 spin_unlock_irqrestore(&tc->lock, flags);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002028 break;
2029 }
Joe Thornber104655f2012-03-28 18:41:28 +01002030
2031 if (bio->bi_rw & REQ_DISCARD)
Joe Thornbere49e5822012-07-27 15:08:16 +01002032 pool->process_discard(tc, bio);
Joe Thornber104655f2012-03-28 18:41:28 +01002033 else
Joe Thornbere49e5822012-07-27 15:08:16 +01002034 pool->process_bio(tc, bio);
Joe Thornber8a01a6a2014-10-06 15:28:30 +01002035
2036 if ((count++ & 127) == 0) {
Joe Thornber7d327fe2014-10-06 15:45:59 +01002037 throttle_work_update(&pool->throttle);
Joe Thornber8a01a6a2014-10-06 15:28:30 +01002038 dm_pool_issue_prefetches(pool->pmd);
2039 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00002040 }
Mike Snitzer67324ea2014-03-21 18:33:41 -04002041 blk_finish_plug(&plug);
Mike Snitzerc140e1c2014-03-20 21:17:14 -04002042}
2043
Joe Thornberac4c3f32014-10-10 16:42:10 +01002044static int cmp_cells(const void *lhs, const void *rhs)
2045{
2046 struct dm_bio_prison_cell *lhs_cell = *((struct dm_bio_prison_cell **) lhs);
2047 struct dm_bio_prison_cell *rhs_cell = *((struct dm_bio_prison_cell **) rhs);
2048
2049 BUG_ON(!lhs_cell->holder);
2050 BUG_ON(!rhs_cell->holder);
2051
2052 if (lhs_cell->holder->bi_iter.bi_sector < rhs_cell->holder->bi_iter.bi_sector)
2053 return -1;
2054
2055 if (lhs_cell->holder->bi_iter.bi_sector > rhs_cell->holder->bi_iter.bi_sector)
2056 return 1;
2057
2058 return 0;
2059}
2060
2061static unsigned sort_cells(struct pool *pool, struct list_head *cells)
2062{
2063 unsigned count = 0;
2064 struct dm_bio_prison_cell *cell, *tmp;
2065
2066 list_for_each_entry_safe(cell, tmp, cells, user_list) {
2067 if (count >= CELL_SORT_ARRAY_SIZE)
2068 break;
2069
2070 pool->cell_sort_array[count++] = cell;
2071 list_del(&cell->user_list);
2072 }
2073
2074 sort(pool->cell_sort_array, count, sizeof(cell), cmp_cells, NULL);
2075
2076 return count;
2077}
2078
Joe Thornbera374bb22014-10-10 13:43:14 +01002079static void process_thin_deferred_cells(struct thin_c *tc)
2080{
2081 struct pool *pool = tc->pool;
2082 unsigned long flags;
2083 struct list_head cells;
Joe Thornberac4c3f32014-10-10 16:42:10 +01002084 struct dm_bio_prison_cell *cell;
2085 unsigned i, j, count;
Joe Thornbera374bb22014-10-10 13:43:14 +01002086
2087 INIT_LIST_HEAD(&cells);
2088
2089 spin_lock_irqsave(&tc->lock, flags);
2090 list_splice_init(&tc->deferred_cells, &cells);
2091 spin_unlock_irqrestore(&tc->lock, flags);
2092
2093 if (list_empty(&cells))
2094 return;
2095
Joe Thornberac4c3f32014-10-10 16:42:10 +01002096 do {
2097 count = sort_cells(tc->pool, &cells);
Joe Thornbera374bb22014-10-10 13:43:14 +01002098
Joe Thornberac4c3f32014-10-10 16:42:10 +01002099 for (i = 0; i < count; i++) {
2100 cell = pool->cell_sort_array[i];
2101 BUG_ON(!cell->holder);
2102
2103 /*
2104 * If we've got no free new_mapping structs, and processing
2105 * this bio might require one, we pause until there are some
2106 * prepared mappings to process.
2107 */
2108 if (ensure_next_mapping(pool)) {
2109 for (j = i; j < count; j++)
2110 list_add(&pool->cell_sort_array[j]->user_list, &cells);
2111
2112 spin_lock_irqsave(&tc->lock, flags);
2113 list_splice(&cells, &tc->deferred_cells);
2114 spin_unlock_irqrestore(&tc->lock, flags);
2115 return;
2116 }
2117
2118 if (cell->holder->bi_rw & REQ_DISCARD)
2119 pool->process_discard_cell(tc, cell);
2120 else
2121 pool->process_cell(tc, cell);
Joe Thornbera374bb22014-10-10 13:43:14 +01002122 }
Joe Thornberac4c3f32014-10-10 16:42:10 +01002123 } while (!list_empty(&cells));
Joe Thornbera374bb22014-10-10 13:43:14 +01002124}
2125
Joe Thornberb10ebd32014-04-08 11:29:01 +01002126static void thin_get(struct thin_c *tc);
2127static void thin_put(struct thin_c *tc);
2128
2129/*
2130 * We can't hold rcu_read_lock() around code that can block. So we
2131 * find a thin with the rcu lock held; bump a refcount; then drop
2132 * the lock.
2133 */
2134static struct thin_c *get_first_thin(struct pool *pool)
2135{
2136 struct thin_c *tc = NULL;
2137
2138 rcu_read_lock();
2139 if (!list_empty(&pool->active_thins)) {
2140 tc = list_entry_rcu(pool->active_thins.next, struct thin_c, list);
2141 thin_get(tc);
2142 }
2143 rcu_read_unlock();
2144
2145 return tc;
2146}
2147
2148static struct thin_c *get_next_thin(struct pool *pool, struct thin_c *tc)
2149{
2150 struct thin_c *old_tc = tc;
2151
2152 rcu_read_lock();
2153 list_for_each_entry_continue_rcu(tc, &pool->active_thins, list) {
2154 thin_get(tc);
2155 thin_put(old_tc);
2156 rcu_read_unlock();
2157 return tc;
2158 }
2159 thin_put(old_tc);
2160 rcu_read_unlock();
2161
2162 return NULL;
2163}
2164
Mike Snitzerc140e1c2014-03-20 21:17:14 -04002165static void process_deferred_bios(struct pool *pool)
2166{
2167 unsigned long flags;
2168 struct bio *bio;
2169 struct bio_list bios;
2170 struct thin_c *tc;
2171
Joe Thornberb10ebd32014-04-08 11:29:01 +01002172 tc = get_first_thin(pool);
2173 while (tc) {
Joe Thornbera374bb22014-10-10 13:43:14 +01002174 process_thin_deferred_cells(tc);
Mike Snitzerc140e1c2014-03-20 21:17:14 -04002175 process_thin_deferred_bios(tc);
Joe Thornberb10ebd32014-04-08 11:29:01 +01002176 tc = get_next_thin(pool, tc);
2177 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00002178
2179 /*
2180 * If there are any deferred flush bios, we must commit
2181 * the metadata before issuing them.
2182 */
2183 bio_list_init(&bios);
2184 spin_lock_irqsave(&pool->lock, flags);
2185 bio_list_merge(&bios, &pool->deferred_flush_bios);
2186 bio_list_init(&pool->deferred_flush_bios);
2187 spin_unlock_irqrestore(&pool->lock, flags);
2188
Mike Snitzer4d1662a2014-02-06 06:08:56 -05002189 if (bio_list_empty(&bios) &&
2190 !(dm_pool_changed_this_transaction(pool->pmd) && need_commit_due_to_time(pool)))
Joe Thornber991d9fa2011-10-31 20:21:18 +00002191 return;
2192
Joe Thornber020cc3b2013-12-04 15:05:36 -05002193 if (commit(pool)) {
Joe Thornber991d9fa2011-10-31 20:21:18 +00002194 while ((bio = bio_list_pop(&bios)))
2195 bio_io_error(bio);
2196 return;
2197 }
Joe Thornber905e51b2012-03-28 18:41:27 +01002198 pool->last_commit_jiffies = jiffies;
Joe Thornber991d9fa2011-10-31 20:21:18 +00002199
2200 while ((bio = bio_list_pop(&bios)))
2201 generic_make_request(bio);
2202}
2203
2204static void do_worker(struct work_struct *ws)
2205{
2206 struct pool *pool = container_of(ws, struct pool, worker);
2207
Joe Thornber7d327fe2014-10-06 15:45:59 +01002208 throttle_work_start(&pool->throttle);
Joe Thornber8a01a6a2014-10-06 15:28:30 +01002209 dm_pool_issue_prefetches(pool->pmd);
Joe Thornber7d327fe2014-10-06 15:45:59 +01002210 throttle_work_update(&pool->throttle);
Joe Thornbere49e5822012-07-27 15:08:16 +01002211 process_prepared(pool, &pool->prepared_mappings, &pool->process_prepared_mapping);
Joe Thornber7d327fe2014-10-06 15:45:59 +01002212 throttle_work_update(&pool->throttle);
Joe Thornbere49e5822012-07-27 15:08:16 +01002213 process_prepared(pool, &pool->prepared_discards, &pool->process_prepared_discard);
Joe Thornber7d327fe2014-10-06 15:45:59 +01002214 throttle_work_update(&pool->throttle);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002215 process_deferred_bios(pool);
Joe Thornber7d327fe2014-10-06 15:45:59 +01002216 throttle_work_complete(&pool->throttle);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002217}
2218
Joe Thornber905e51b2012-03-28 18:41:27 +01002219/*
2220 * We want to commit periodically so that not too much
2221 * unwritten data builds up.
2222 */
2223static void do_waker(struct work_struct *ws)
2224{
2225 struct pool *pool = container_of(to_delayed_work(ws), struct pool, waker);
2226 wake_worker(pool);
2227 queue_delayed_work(pool->wq, &pool->waker, COMMIT_PERIOD);
2228}
2229
Mike Snitzerbcc696f2015-07-15 16:52:04 -04002230static void notify_of_pool_mode_change_to_oods(struct pool *pool);
2231
Joe Thornber85ad643b2014-05-09 15:59:38 +01002232/*
2233 * We're holding onto IO to allow userland time to react. After the
2234 * timeout either the pool will have been resized (and thus back in
Mike Snitzerbcc696f2015-07-15 16:52:04 -04002235 * PM_WRITE mode), or we degrade to PM_OUT_OF_DATA_SPACE w/ error_if_no_space.
Joe Thornber85ad643b2014-05-09 15:59:38 +01002236 */
2237static void do_no_space_timeout(struct work_struct *ws)
2238{
2239 struct pool *pool = container_of(to_delayed_work(ws), struct pool,
2240 no_space_timeout);
2241
Mike Snitzerbcc696f2015-07-15 16:52:04 -04002242 if (get_pool_mode(pool) == PM_OUT_OF_DATA_SPACE && !pool->pf.error_if_no_space) {
2243 pool->pf.error_if_no_space = true;
2244 notify_of_pool_mode_change_to_oods(pool);
Mike Snitzer0a927c22015-07-21 13:20:46 -04002245 error_retry_list_with_code(pool, -ENOSPC);
Mike Snitzerbcc696f2015-07-15 16:52:04 -04002246 }
Joe Thornber85ad643b2014-05-09 15:59:38 +01002247}
2248
Joe Thornber991d9fa2011-10-31 20:21:18 +00002249/*----------------------------------------------------------------*/
2250
Joe Thornbere7a3e872014-05-13 16:14:14 -04002251struct pool_work {
Joe Thornber738211f2014-03-03 15:52:28 +00002252 struct work_struct worker;
Joe Thornbere7a3e872014-05-13 16:14:14 -04002253 struct completion complete;
Joe Thornber738211f2014-03-03 15:52:28 +00002254};
2255
Joe Thornbere7a3e872014-05-13 16:14:14 -04002256static struct pool_work *to_pool_work(struct work_struct *ws)
Joe Thornber738211f2014-03-03 15:52:28 +00002257{
Joe Thornbere7a3e872014-05-13 16:14:14 -04002258 return container_of(ws, struct pool_work, worker);
2259}
2260
2261static void pool_work_complete(struct pool_work *pw)
2262{
2263 complete(&pw->complete);
2264}
2265
2266static void pool_work_wait(struct pool_work *pw, struct pool *pool,
2267 void (*fn)(struct work_struct *))
2268{
2269 INIT_WORK_ONSTACK(&pw->worker, fn);
2270 init_completion(&pw->complete);
2271 queue_work(pool->wq, &pw->worker);
2272 wait_for_completion(&pw->complete);
2273}
2274
2275/*----------------------------------------------------------------*/
2276
2277struct noflush_work {
2278 struct pool_work pw;
2279 struct thin_c *tc;
2280};
2281
2282static struct noflush_work *to_noflush(struct work_struct *ws)
2283{
2284 return container_of(to_pool_work(ws), struct noflush_work, pw);
Joe Thornber738211f2014-03-03 15:52:28 +00002285}
2286
2287static void do_noflush_start(struct work_struct *ws)
2288{
Joe Thornbere7a3e872014-05-13 16:14:14 -04002289 struct noflush_work *w = to_noflush(ws);
Joe Thornber738211f2014-03-03 15:52:28 +00002290 w->tc->requeue_mode = true;
2291 requeue_io(w->tc);
Joe Thornbere7a3e872014-05-13 16:14:14 -04002292 pool_work_complete(&w->pw);
Joe Thornber738211f2014-03-03 15:52:28 +00002293}
2294
2295static void do_noflush_stop(struct work_struct *ws)
2296{
Joe Thornbere7a3e872014-05-13 16:14:14 -04002297 struct noflush_work *w = to_noflush(ws);
Joe Thornber738211f2014-03-03 15:52:28 +00002298 w->tc->requeue_mode = false;
Joe Thornbere7a3e872014-05-13 16:14:14 -04002299 pool_work_complete(&w->pw);
Joe Thornber738211f2014-03-03 15:52:28 +00002300}
2301
2302static void noflush_work(struct thin_c *tc, void (*fn)(struct work_struct *))
2303{
2304 struct noflush_work w;
2305
Joe Thornber738211f2014-03-03 15:52:28 +00002306 w.tc = tc;
Joe Thornbere7a3e872014-05-13 16:14:14 -04002307 pool_work_wait(&w.pw, tc->pool, fn);
Joe Thornber738211f2014-03-03 15:52:28 +00002308}
2309
2310/*----------------------------------------------------------------*/
2311
Joe Thornbere49e5822012-07-27 15:08:16 +01002312static enum pool_mode get_pool_mode(struct pool *pool)
2313{
2314 return pool->pf.mode;
2315}
2316
Joe Thornber3e1a0692014-03-03 16:03:26 +00002317static void notify_of_pool_mode_change(struct pool *pool, const char *new_mode)
2318{
2319 dm_table_event(pool->ti->table);
2320 DMINFO("%s: switching pool to %s mode",
2321 dm_device_name(pool->pool_md), new_mode);
2322}
2323
Mike Snitzerbcc696f2015-07-15 16:52:04 -04002324static void notify_of_pool_mode_change_to_oods(struct pool *pool)
2325{
2326 if (!pool->pf.error_if_no_space)
2327 notify_of_pool_mode_change(pool, "out-of-data-space (queue IO)");
2328 else
2329 notify_of_pool_mode_change(pool, "out-of-data-space (error IO)");
2330}
2331
Joe Thornber34fbcf62015-04-16 12:58:35 +01002332static bool passdown_enabled(struct pool_c *pt)
2333{
2334 return pt->adjusted_pf.discard_passdown;
2335}
2336
2337static void set_discard_callbacks(struct pool *pool)
2338{
2339 struct pool_c *pt = pool->ti->private;
2340
2341 if (passdown_enabled(pt)) {
2342 pool->process_discard_cell = process_discard_cell_passdown;
2343 pool->process_prepared_discard = process_prepared_discard_passdown;
2344 } else {
2345 pool->process_discard_cell = process_discard_cell_no_passdown;
2346 pool->process_prepared_discard = process_prepared_discard_no_passdown;
2347 }
2348}
2349
Mike Snitzer8b64e882013-12-20 14:27:28 -05002350static void set_pool_mode(struct pool *pool, enum pool_mode new_mode)
Joe Thornbere49e5822012-07-27 15:08:16 +01002351{
Mike Snitzercdc2b412014-02-14 18:10:55 -05002352 struct pool_c *pt = pool->ti->private;
Mike Snitzer07f2b6e2014-02-14 11:58:41 -05002353 bool needs_check = dm_pool_metadata_needs_check(pool->pmd);
2354 enum pool_mode old_mode = get_pool_mode(pool);
Mike Snitzer80c57892014-05-20 13:38:33 -04002355 unsigned long no_space_timeout = ACCESS_ONCE(no_space_timeout_secs) * HZ;
Mike Snitzer07f2b6e2014-02-14 11:58:41 -05002356
2357 /*
2358 * Never allow the pool to transition to PM_WRITE mode if user
2359 * intervention is required to verify metadata and data consistency.
2360 */
2361 if (new_mode == PM_WRITE && needs_check) {
2362 DMERR("%s: unable to switch pool to write mode until repaired.",
2363 dm_device_name(pool->pool_md));
2364 if (old_mode != new_mode)
2365 new_mode = old_mode;
2366 else
2367 new_mode = PM_READ_ONLY;
2368 }
2369 /*
2370 * If we were in PM_FAIL mode, rollback of metadata failed. We're
2371 * not going to recover without a thin_repair. So we never let the
2372 * pool move out of the old mode.
2373 */
2374 if (old_mode == PM_FAIL)
2375 new_mode = old_mode;
Joe Thornbere49e5822012-07-27 15:08:16 +01002376
Mike Snitzer8b64e882013-12-20 14:27:28 -05002377 switch (new_mode) {
Joe Thornbere49e5822012-07-27 15:08:16 +01002378 case PM_FAIL:
Mike Snitzer8b64e882013-12-20 14:27:28 -05002379 if (old_mode != new_mode)
Joe Thornber3e1a0692014-03-03 16:03:26 +00002380 notify_of_pool_mode_change(pool, "failure");
Joe Thornber5383ef32013-12-04 16:30:01 -05002381 dm_pool_metadata_read_only(pool->pmd);
Joe Thornbere49e5822012-07-27 15:08:16 +01002382 pool->process_bio = process_bio_fail;
2383 pool->process_discard = process_bio_fail;
Joe Thornbera374bb22014-10-10 13:43:14 +01002384 pool->process_cell = process_cell_fail;
2385 pool->process_discard_cell = process_cell_fail;
Joe Thornbere49e5822012-07-27 15:08:16 +01002386 pool->process_prepared_mapping = process_prepared_mapping_fail;
2387 pool->process_prepared_discard = process_prepared_discard_fail;
Joe Thornber3e1a0692014-03-03 16:03:26 +00002388
2389 error_retry_list(pool);
Joe Thornbere49e5822012-07-27 15:08:16 +01002390 break;
2391
2392 case PM_READ_ONLY:
Mike Snitzer8b64e882013-12-20 14:27:28 -05002393 if (old_mode != new_mode)
Joe Thornber3e1a0692014-03-03 16:03:26 +00002394 notify_of_pool_mode_change(pool, "read-only");
2395 dm_pool_metadata_read_only(pool->pmd);
2396 pool->process_bio = process_bio_read_only;
2397 pool->process_discard = process_bio_success;
Joe Thornbera374bb22014-10-10 13:43:14 +01002398 pool->process_cell = process_cell_read_only;
2399 pool->process_discard_cell = process_cell_success;
Joe Thornber3e1a0692014-03-03 16:03:26 +00002400 pool->process_prepared_mapping = process_prepared_mapping_fail;
Joe Thornber34fbcf62015-04-16 12:58:35 +01002401 pool->process_prepared_discard = process_prepared_discard_success;
Joe Thornber3e1a0692014-03-03 16:03:26 +00002402
2403 error_retry_list(pool);
2404 break;
2405
2406 case PM_OUT_OF_DATA_SPACE:
2407 /*
2408 * Ideally we'd never hit this state; the low water mark
2409 * would trigger userland to extend the pool before we
2410 * completely run out of data space. However, many small
2411 * IOs to unprovisioned space can consume data space at an
2412 * alarming rate. Adjust your low water mark if you're
2413 * frequently seeing this mode.
2414 */
2415 if (old_mode != new_mode)
Mike Snitzerbcc696f2015-07-15 16:52:04 -04002416 notify_of_pool_mode_change_to_oods(pool);
Mike Snitzerc3667cc2016-03-10 11:31:35 -05002417 pool->out_of_data_space = true;
Joe Thornber3e1a0692014-03-03 16:03:26 +00002418 pool->process_bio = process_bio_read_only;
Joe Thornbera374bb22014-10-10 13:43:14 +01002419 pool->process_discard = process_discard_bio;
2420 pool->process_cell = process_cell_read_only;
Joe Thornber3e1a0692014-03-03 16:03:26 +00002421 pool->process_prepared_mapping = process_prepared_mapping;
Joe Thornber34fbcf62015-04-16 12:58:35 +01002422 set_discard_callbacks(pool);
Joe Thornber85ad643b2014-05-09 15:59:38 +01002423
Mike Snitzer80c57892014-05-20 13:38:33 -04002424 if (!pool->pf.error_if_no_space && no_space_timeout)
2425 queue_delayed_work(pool->wq, &pool->no_space_timeout, no_space_timeout);
Joe Thornbere49e5822012-07-27 15:08:16 +01002426 break;
2427
2428 case PM_WRITE:
Mike Snitzer8b64e882013-12-20 14:27:28 -05002429 if (old_mode != new_mode)
Joe Thornber3e1a0692014-03-03 16:03:26 +00002430 notify_of_pool_mode_change(pool, "write");
Mike Snitzerc3667cc2016-03-10 11:31:35 -05002431 pool->out_of_data_space = false;
Mike Snitzer172c2382015-11-06 10:53:01 -05002432 pool->pf.error_if_no_space = pt->requested_pf.error_if_no_space;
Joe Thornber9b7aaa62013-12-04 16:58:19 -05002433 dm_pool_metadata_read_write(pool->pmd);
Joe Thornbere49e5822012-07-27 15:08:16 +01002434 pool->process_bio = process_bio;
Joe Thornbera374bb22014-10-10 13:43:14 +01002435 pool->process_discard = process_discard_bio;
2436 pool->process_cell = process_cell;
Joe Thornbere49e5822012-07-27 15:08:16 +01002437 pool->process_prepared_mapping = process_prepared_mapping;
Joe Thornber34fbcf62015-04-16 12:58:35 +01002438 set_discard_callbacks(pool);
Joe Thornbere49e5822012-07-27 15:08:16 +01002439 break;
2440 }
Mike Snitzer8b64e882013-12-20 14:27:28 -05002441
2442 pool->pf.mode = new_mode;
Mike Snitzercdc2b412014-02-14 18:10:55 -05002443 /*
2444 * The pool mode may have changed, sync it so bind_control_target()
2445 * doesn't cause an unexpected mode transition on resume.
2446 */
2447 pt->adjusted_pf.mode = new_mode;
Joe Thornbere49e5822012-07-27 15:08:16 +01002448}
2449
Mike Snitzer07f2b6e2014-02-14 11:58:41 -05002450static void abort_transaction(struct pool *pool)
2451{
2452 const char *dev_name = dm_device_name(pool->pool_md);
2453
2454 DMERR_LIMIT("%s: aborting current metadata transaction", dev_name);
2455 if (dm_pool_abort_metadata(pool->pmd)) {
2456 DMERR("%s: failed to abort metadata transaction", dev_name);
2457 set_pool_mode(pool, PM_FAIL);
2458 }
2459
2460 if (dm_pool_metadata_set_needs_check(pool->pmd)) {
2461 DMERR("%s: failed to set 'needs_check' flag in metadata", dev_name);
2462 set_pool_mode(pool, PM_FAIL);
2463 }
2464}
2465
Joe Thornberb5330652013-12-04 19:51:33 -05002466static void metadata_operation_failed(struct pool *pool, const char *op, int r)
2467{
2468 DMERR_LIMIT("%s: metadata operation '%s' failed: error = %d",
2469 dm_device_name(pool->pool_md), op, r);
2470
Mike Snitzer07f2b6e2014-02-14 11:58:41 -05002471 abort_transaction(pool);
Joe Thornberb5330652013-12-04 19:51:33 -05002472 set_pool_mode(pool, PM_READ_ONLY);
2473}
2474
Joe Thornbere49e5822012-07-27 15:08:16 +01002475/*----------------------------------------------------------------*/
2476
Joe Thornber991d9fa2011-10-31 20:21:18 +00002477/*
2478 * Mapping functions.
2479 */
2480
2481/*
2482 * Called only while mapping a thin bio to hand it over to the workqueue.
2483 */
2484static void thin_defer_bio(struct thin_c *tc, struct bio *bio)
2485{
2486 unsigned long flags;
2487 struct pool *pool = tc->pool;
2488
Mike Snitzerc140e1c2014-03-20 21:17:14 -04002489 spin_lock_irqsave(&tc->lock, flags);
2490 bio_list_add(&tc->deferred_bio_list, bio);
2491 spin_unlock_irqrestore(&tc->lock, flags);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002492
2493 wake_worker(pool);
2494}
2495
Joe Thornber7d327fe2014-10-06 15:45:59 +01002496static void thin_defer_bio_with_throttle(struct thin_c *tc, struct bio *bio)
2497{
2498 struct pool *pool = tc->pool;
2499
2500 throttle_lock(&pool->throttle);
2501 thin_defer_bio(tc, bio);
2502 throttle_unlock(&pool->throttle);
2503}
2504
Joe Thornbera374bb22014-10-10 13:43:14 +01002505static void thin_defer_cell(struct thin_c *tc, struct dm_bio_prison_cell *cell)
2506{
2507 unsigned long flags;
2508 struct pool *pool = tc->pool;
2509
2510 throttle_lock(&pool->throttle);
2511 spin_lock_irqsave(&tc->lock, flags);
2512 list_add_tail(&cell->user_list, &tc->deferred_cells);
2513 spin_unlock_irqrestore(&tc->lock, flags);
2514 throttle_unlock(&pool->throttle);
2515
2516 wake_worker(pool);
2517}
2518
Mikulas Patocka59c3d2c2012-12-21 20:23:40 +00002519static void thin_hook_bio(struct thin_c *tc, struct bio *bio)
Joe Thornbereb2aa482012-03-28 18:41:28 +01002520{
Mikulas Patocka59c3d2c2012-12-21 20:23:40 +00002521 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
Joe Thornbereb2aa482012-03-28 18:41:28 +01002522
2523 h->tc = tc;
2524 h->shared_read_entry = NULL;
Joe Thornbere8088072012-12-21 20:23:31 +00002525 h->all_io_entry = NULL;
Joe Thornbereb2aa482012-03-28 18:41:28 +01002526 h->overwrite_mapping = NULL;
Joe Thornber34fbcf62015-04-16 12:58:35 +01002527 h->cell = NULL;
Joe Thornbereb2aa482012-03-28 18:41:28 +01002528}
2529
Joe Thornber991d9fa2011-10-31 20:21:18 +00002530/*
2531 * Non-blocking function called from the thin target's map function.
2532 */
Mikulas Patocka7de3ee52012-12-21 20:23:41 +00002533static int thin_bio_map(struct dm_target *ti, struct bio *bio)
Joe Thornber991d9fa2011-10-31 20:21:18 +00002534{
2535 int r;
2536 struct thin_c *tc = ti->private;
2537 dm_block_t block = get_bio_block(tc, bio);
2538 struct dm_thin_device *td = tc->td;
2539 struct dm_thin_lookup_result result;
Joe Thornbera374bb22014-10-10 13:43:14 +01002540 struct dm_bio_prison_cell *virt_cell, *data_cell;
Joe Thornbere8088072012-12-21 20:23:31 +00002541 struct dm_cell_key key;
Joe Thornber991d9fa2011-10-31 20:21:18 +00002542
Mikulas Patocka59c3d2c2012-12-21 20:23:40 +00002543 thin_hook_bio(tc, bio);
Joe Thornbere49e5822012-07-27 15:08:16 +01002544
Joe Thornber738211f2014-03-03 15:52:28 +00002545 if (tc->requeue_mode) {
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02002546 bio->bi_error = DM_ENDIO_REQUEUE;
2547 bio_endio(bio);
Joe Thornber738211f2014-03-03 15:52:28 +00002548 return DM_MAPIO_SUBMITTED;
2549 }
2550
Joe Thornbere49e5822012-07-27 15:08:16 +01002551 if (get_pool_mode(tc->pool) == PM_FAIL) {
2552 bio_io_error(bio);
2553 return DM_MAPIO_SUBMITTED;
2554 }
2555
Joe Thornber104655f2012-03-28 18:41:28 +01002556 if (bio->bi_rw & (REQ_DISCARD | REQ_FLUSH | REQ_FUA)) {
Joe Thornber7d327fe2014-10-06 15:45:59 +01002557 thin_defer_bio_with_throttle(tc, bio);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002558 return DM_MAPIO_SUBMITTED;
2559 }
2560
Joe Thornberc822ed92014-10-10 09:41:09 +01002561 /*
2562 * We must hold the virtual cell before doing the lookup, otherwise
2563 * there's a race with discard.
2564 */
2565 build_virtual_key(tc->td, block, &key);
Joe Thornbera374bb22014-10-10 13:43:14 +01002566 if (bio_detain(tc->pool, &key, bio, &virt_cell))
Joe Thornberc822ed92014-10-10 09:41:09 +01002567 return DM_MAPIO_SUBMITTED;
2568
Joe Thornber991d9fa2011-10-31 20:21:18 +00002569 r = dm_thin_find_block(td, block, 0, &result);
2570
2571 /*
2572 * Note that we defer readahead too.
2573 */
2574 switch (r) {
2575 case 0:
2576 if (unlikely(result.shared)) {
2577 /*
2578 * We have a race condition here between the
2579 * result.shared value returned by the lookup and
2580 * snapshot creation, which may cause new
2581 * sharing.
2582 *
2583 * To avoid this always quiesce the origin before
2584 * taking the snap. You want to do this anyway to
2585 * ensure a consistent application view
2586 * (i.e. lockfs).
2587 *
2588 * More distant ancestors are irrelevant. The
2589 * shared flag will be set in their case.
2590 */
Joe Thornbera374bb22014-10-10 13:43:14 +01002591 thin_defer_cell(tc, virt_cell);
Joe Thornbere8088072012-12-21 20:23:31 +00002592 return DM_MAPIO_SUBMITTED;
Joe Thornber991d9fa2011-10-31 20:21:18 +00002593 }
Joe Thornbere8088072012-12-21 20:23:31 +00002594
Joe Thornbere8088072012-12-21 20:23:31 +00002595 build_data_key(tc->td, result.block, &key);
Joe Thornbera374bb22014-10-10 13:43:14 +01002596 if (bio_detain(tc->pool, &key, bio, &data_cell)) {
2597 cell_defer_no_holder(tc, virt_cell);
Joe Thornbere8088072012-12-21 20:23:31 +00002598 return DM_MAPIO_SUBMITTED;
2599 }
2600
2601 inc_all_io_entry(tc->pool, bio);
Joe Thornbera374bb22014-10-10 13:43:14 +01002602 cell_defer_no_holder(tc, data_cell);
2603 cell_defer_no_holder(tc, virt_cell);
Joe Thornbere8088072012-12-21 20:23:31 +00002604
2605 remap(tc, bio, result.block);
2606 return DM_MAPIO_REMAPPED;
Joe Thornber991d9fa2011-10-31 20:21:18 +00002607
2608 case -ENODATA:
Joe Thornbere49e5822012-07-27 15:08:16 +01002609 case -EWOULDBLOCK:
Joe Thornbera374bb22014-10-10 13:43:14 +01002610 thin_defer_cell(tc, virt_cell);
Joe Thornber2aab3852012-12-21 20:23:33 +00002611 return DM_MAPIO_SUBMITTED;
Joe Thornbere49e5822012-07-27 15:08:16 +01002612
2613 default:
2614 /*
2615 * Must always call bio_io_error on failure.
2616 * dm_thin_find_block can fail with -EINVAL if the
2617 * pool is switched to fail-io mode.
2618 */
2619 bio_io_error(bio);
Joe Thornbera374bb22014-10-10 13:43:14 +01002620 cell_defer_no_holder(tc, virt_cell);
Joe Thornber2aab3852012-12-21 20:23:33 +00002621 return DM_MAPIO_SUBMITTED;
Joe Thornber991d9fa2011-10-31 20:21:18 +00002622 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00002623}
2624
2625static int pool_is_congested(struct dm_target_callbacks *cb, int bdi_bits)
2626{
Joe Thornber991d9fa2011-10-31 20:21:18 +00002627 struct pool_c *pt = container_of(cb, struct pool_c, callbacks);
Mike Snitzer760fe672014-03-20 08:36:47 -04002628 struct request_queue *q;
Joe Thornber991d9fa2011-10-31 20:21:18 +00002629
Mike Snitzer760fe672014-03-20 08:36:47 -04002630 if (get_pool_mode(pt->pool) == PM_OUT_OF_DATA_SPACE)
2631 return 1;
Joe Thornber991d9fa2011-10-31 20:21:18 +00002632
Mike Snitzer760fe672014-03-20 08:36:47 -04002633 q = bdev_get_queue(pt->data_dev->bdev);
2634 return bdi_congested(&q->backing_dev_info, bdi_bits);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002635}
2636
Mike Snitzerc140e1c2014-03-20 21:17:14 -04002637static void requeue_bios(struct pool *pool)
Joe Thornber991d9fa2011-10-31 20:21:18 +00002638{
Mike Snitzerc140e1c2014-03-20 21:17:14 -04002639 unsigned long flags;
2640 struct thin_c *tc;
2641
2642 rcu_read_lock();
2643 list_for_each_entry_rcu(tc, &pool->active_thins, list) {
2644 spin_lock_irqsave(&tc->lock, flags);
2645 bio_list_merge(&tc->deferred_bio_list, &tc->retry_on_resume_list);
2646 bio_list_init(&tc->retry_on_resume_list);
2647 spin_unlock_irqrestore(&tc->lock, flags);
2648 }
2649 rcu_read_unlock();
Joe Thornber991d9fa2011-10-31 20:21:18 +00002650}
2651
2652/*----------------------------------------------------------------
2653 * Binding of control targets to a pool object
2654 *--------------------------------------------------------------*/
Mike Snitzer9bc142d2012-09-26 23:45:46 +01002655static bool data_dev_supports_discard(struct pool_c *pt)
2656{
2657 struct request_queue *q = bdev_get_queue(pt->data_dev->bdev);
2658
2659 return q && blk_queue_discard(q);
2660}
2661
Joe Thornber58051b92013-03-20 17:21:25 +00002662static bool is_factor(sector_t block_size, uint32_t n)
2663{
2664 return !sector_div(block_size, n);
2665}
2666
Mike Snitzer9bc142d2012-09-26 23:45:46 +01002667/*
2668 * If discard_passdown was enabled verify that the data device
Mike Snitzer0424caa2012-09-26 23:45:47 +01002669 * supports discards. Disable discard_passdown if not.
Mike Snitzer9bc142d2012-09-26 23:45:46 +01002670 */
Mike Snitzer0424caa2012-09-26 23:45:47 +01002671static void disable_passdown_if_not_supported(struct pool_c *pt)
Mike Snitzer9bc142d2012-09-26 23:45:46 +01002672{
Mike Snitzer0424caa2012-09-26 23:45:47 +01002673 struct pool *pool = pt->pool;
2674 struct block_device *data_bdev = pt->data_dev->bdev;
2675 struct queue_limits *data_limits = &bdev_get_queue(data_bdev)->limits;
Mike Snitzer0424caa2012-09-26 23:45:47 +01002676 const char *reason = NULL;
Mike Snitzer9bc142d2012-09-26 23:45:46 +01002677 char buf[BDEVNAME_SIZE];
2678
Mike Snitzer0424caa2012-09-26 23:45:47 +01002679 if (!pt->adjusted_pf.discard_passdown)
Mike Snitzer9bc142d2012-09-26 23:45:46 +01002680 return;
2681
Mike Snitzer0424caa2012-09-26 23:45:47 +01002682 if (!data_dev_supports_discard(pt))
2683 reason = "discard unsupported";
Mike Snitzer9bc142d2012-09-26 23:45:46 +01002684
Mike Snitzer0424caa2012-09-26 23:45:47 +01002685 else if (data_limits->max_discard_sectors < pool->sectors_per_block)
2686 reason = "max discard sectors smaller than a block";
2687
Mike Snitzer0424caa2012-09-26 23:45:47 +01002688 if (reason) {
2689 DMWARN("Data device (%s) %s: Disabling discard passdown.", bdevname(data_bdev, buf), reason);
2690 pt->adjusted_pf.discard_passdown = false;
2691 }
Mike Snitzer9bc142d2012-09-26 23:45:46 +01002692}
2693
Joe Thornber991d9fa2011-10-31 20:21:18 +00002694static int bind_control_target(struct pool *pool, struct dm_target *ti)
2695{
2696 struct pool_c *pt = ti->private;
2697
Joe Thornbere49e5822012-07-27 15:08:16 +01002698 /*
Joe Thornber9b7aaa62013-12-04 16:58:19 -05002699 * We want to make sure that a pool in PM_FAIL mode is never upgraded.
Joe Thornbere49e5822012-07-27 15:08:16 +01002700 */
Mike Snitzer07f2b6e2014-02-14 11:58:41 -05002701 enum pool_mode old_mode = get_pool_mode(pool);
Mike Snitzer0424caa2012-09-26 23:45:47 +01002702 enum pool_mode new_mode = pt->adjusted_pf.mode;
Joe Thornbere49e5822012-07-27 15:08:16 +01002703
Joe Thornber9b7aaa62013-12-04 16:58:19 -05002704 /*
Mike Snitzer8b64e882013-12-20 14:27:28 -05002705 * Don't change the pool's mode until set_pool_mode() below.
2706 * Otherwise the pool's process_* function pointers may
2707 * not match the desired pool mode.
2708 */
2709 pt->adjusted_pf.mode = old_mode;
2710
2711 pool->ti = ti;
2712 pool->pf = pt->adjusted_pf;
2713 pool->low_water_blocks = pt->low_water_blocks;
2714
Mike Snitzer9bc142d2012-09-26 23:45:46 +01002715 set_pool_mode(pool, new_mode);
Mike Snitzerf4026932012-05-19 01:01:01 +01002716
Joe Thornber991d9fa2011-10-31 20:21:18 +00002717 return 0;
2718}
2719
2720static void unbind_control_target(struct pool *pool, struct dm_target *ti)
2721{
2722 if (pool->ti == ti)
2723 pool->ti = NULL;
2724}
2725
2726/*----------------------------------------------------------------
2727 * Pool creation
2728 *--------------------------------------------------------------*/
Joe Thornber67e2e2b2012-03-28 18:41:29 +01002729/* Initialize pool features. */
2730static void pool_features_init(struct pool_features *pf)
2731{
Joe Thornbere49e5822012-07-27 15:08:16 +01002732 pf->mode = PM_WRITE;
Mike Snitzer9bc142d2012-09-26 23:45:46 +01002733 pf->zero_new_blocks = true;
2734 pf->discard_enabled = true;
2735 pf->discard_passdown = true;
Mike Snitzer787a996c2013-12-06 16:21:43 -05002736 pf->error_if_no_space = false;
Joe Thornber67e2e2b2012-03-28 18:41:29 +01002737}
2738
Joe Thornber991d9fa2011-10-31 20:21:18 +00002739static void __pool_destroy(struct pool *pool)
2740{
2741 __pool_table_remove(pool);
2742
Joe Thornbera822c832015-07-03 10:22:42 +01002743 vfree(pool->cell_sort_array);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002744 if (dm_pool_metadata_close(pool->pmd) < 0)
2745 DMWARN("%s: dm_pool_metadata_close() failed.", __func__);
2746
Mike Snitzer44feb382012-10-12 21:02:10 +01002747 dm_bio_prison_destroy(pool->prison);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002748 dm_kcopyd_client_destroy(pool->copier);
2749
2750 if (pool->wq)
2751 destroy_workqueue(pool->wq);
2752
2753 if (pool->next_mapping)
2754 mempool_free(pool->next_mapping, pool->mapping_pool);
2755 mempool_destroy(pool->mapping_pool);
Mike Snitzer44feb382012-10-12 21:02:10 +01002756 dm_deferred_set_destroy(pool->shared_read_ds);
2757 dm_deferred_set_destroy(pool->all_io_ds);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002758 kfree(pool);
2759}
2760
Mike Snitzera24c2562012-06-03 00:30:00 +01002761static struct kmem_cache *_new_mapping_cache;
Mike Snitzera24c2562012-06-03 00:30:00 +01002762
Joe Thornber991d9fa2011-10-31 20:21:18 +00002763static struct pool *pool_create(struct mapped_device *pool_md,
2764 struct block_device *metadata_dev,
Joe Thornbere49e5822012-07-27 15:08:16 +01002765 unsigned long block_size,
2766 int read_only, char **error)
Joe Thornber991d9fa2011-10-31 20:21:18 +00002767{
2768 int r;
2769 void *err_p;
2770 struct pool *pool;
2771 struct dm_pool_metadata *pmd;
Joe Thornbere49e5822012-07-27 15:08:16 +01002772 bool format_device = read_only ? false : true;
Joe Thornber991d9fa2011-10-31 20:21:18 +00002773
Joe Thornbere49e5822012-07-27 15:08:16 +01002774 pmd = dm_pool_metadata_open(metadata_dev, block_size, format_device);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002775 if (IS_ERR(pmd)) {
2776 *error = "Error creating metadata object";
2777 return (struct pool *)pmd;
2778 }
2779
2780 pool = kmalloc(sizeof(*pool), GFP_KERNEL);
2781 if (!pool) {
2782 *error = "Error allocating memory for pool";
2783 err_p = ERR_PTR(-ENOMEM);
2784 goto bad_pool;
2785 }
2786
2787 pool->pmd = pmd;
2788 pool->sectors_per_block = block_size;
Mikulas Patockaf9a8e0c2012-07-27 15:08:03 +01002789 if (block_size & (block_size - 1))
2790 pool->sectors_per_block_shift = -1;
2791 else
2792 pool->sectors_per_block_shift = __ffs(block_size);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002793 pool->low_water_blocks = 0;
Joe Thornber67e2e2b2012-03-28 18:41:29 +01002794 pool_features_init(&pool->pf);
Joe Thornbera195db22014-10-06 16:30:06 -04002795 pool->prison = dm_bio_prison_create();
Joe Thornber991d9fa2011-10-31 20:21:18 +00002796 if (!pool->prison) {
2797 *error = "Error creating pool's bio prison";
2798 err_p = ERR_PTR(-ENOMEM);
2799 goto bad_prison;
2800 }
2801
Mikulas Patockadf5d2e92013-03-01 22:45:49 +00002802 pool->copier = dm_kcopyd_client_create(&dm_kcopyd_throttle);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002803 if (IS_ERR(pool->copier)) {
2804 r = PTR_ERR(pool->copier);
2805 *error = "Error creating pool's kcopyd client";
2806 err_p = ERR_PTR(r);
2807 goto bad_kcopyd_client;
2808 }
2809
2810 /*
2811 * Create singlethreaded workqueue that will service all devices
2812 * that use this metadata.
2813 */
2814 pool->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM);
2815 if (!pool->wq) {
2816 *error = "Error creating pool's workqueue";
2817 err_p = ERR_PTR(-ENOMEM);
2818 goto bad_wq;
2819 }
2820
Joe Thornber7d327fe2014-10-06 15:45:59 +01002821 throttle_init(&pool->throttle);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002822 INIT_WORK(&pool->worker, do_worker);
Joe Thornber905e51b2012-03-28 18:41:27 +01002823 INIT_DELAYED_WORK(&pool->waker, do_waker);
Joe Thornber85ad643b2014-05-09 15:59:38 +01002824 INIT_DELAYED_WORK(&pool->no_space_timeout, do_no_space_timeout);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002825 spin_lock_init(&pool->lock);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002826 bio_list_init(&pool->deferred_flush_bios);
2827 INIT_LIST_HEAD(&pool->prepared_mappings);
Joe Thornber104655f2012-03-28 18:41:28 +01002828 INIT_LIST_HEAD(&pool->prepared_discards);
Mike Snitzerc140e1c2014-03-20 21:17:14 -04002829 INIT_LIST_HEAD(&pool->active_thins);
Joe Thornber88a66212013-12-04 20:16:12 -05002830 pool->low_water_triggered = false;
Mike Snitzer80e96c52014-11-07 15:09:46 -05002831 pool->suspended = true;
Mike Snitzerc3667cc2016-03-10 11:31:35 -05002832 pool->out_of_data_space = false;
Mike Snitzer44feb382012-10-12 21:02:10 +01002833
2834 pool->shared_read_ds = dm_deferred_set_create();
2835 if (!pool->shared_read_ds) {
2836 *error = "Error creating pool's shared read deferred set";
2837 err_p = ERR_PTR(-ENOMEM);
2838 goto bad_shared_read_ds;
2839 }
2840
2841 pool->all_io_ds = dm_deferred_set_create();
2842 if (!pool->all_io_ds) {
2843 *error = "Error creating pool's all io deferred set";
2844 err_p = ERR_PTR(-ENOMEM);
2845 goto bad_all_io_ds;
2846 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00002847
2848 pool->next_mapping = NULL;
Mike Snitzera24c2562012-06-03 00:30:00 +01002849 pool->mapping_pool = mempool_create_slab_pool(MAPPING_POOL_SIZE,
2850 _new_mapping_cache);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002851 if (!pool->mapping_pool) {
2852 *error = "Error creating pool's mapping mempool";
2853 err_p = ERR_PTR(-ENOMEM);
2854 goto bad_mapping_pool;
2855 }
2856
Joe Thornbera822c832015-07-03 10:22:42 +01002857 pool->cell_sort_array = vmalloc(sizeof(*pool->cell_sort_array) * CELL_SORT_ARRAY_SIZE);
2858 if (!pool->cell_sort_array) {
2859 *error = "Error allocating cell sort array";
2860 err_p = ERR_PTR(-ENOMEM);
2861 goto bad_sort_array;
2862 }
2863
Joe Thornber991d9fa2011-10-31 20:21:18 +00002864 pool->ref_count = 1;
Joe Thornber905e51b2012-03-28 18:41:27 +01002865 pool->last_commit_jiffies = jiffies;
Joe Thornber991d9fa2011-10-31 20:21:18 +00002866 pool->pool_md = pool_md;
2867 pool->md_dev = metadata_dev;
2868 __pool_table_insert(pool);
2869
2870 return pool;
2871
Joe Thornbera822c832015-07-03 10:22:42 +01002872bad_sort_array:
2873 mempool_destroy(pool->mapping_pool);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002874bad_mapping_pool:
Mike Snitzer44feb382012-10-12 21:02:10 +01002875 dm_deferred_set_destroy(pool->all_io_ds);
2876bad_all_io_ds:
2877 dm_deferred_set_destroy(pool->shared_read_ds);
2878bad_shared_read_ds:
Joe Thornber991d9fa2011-10-31 20:21:18 +00002879 destroy_workqueue(pool->wq);
2880bad_wq:
2881 dm_kcopyd_client_destroy(pool->copier);
2882bad_kcopyd_client:
Mike Snitzer44feb382012-10-12 21:02:10 +01002883 dm_bio_prison_destroy(pool->prison);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002884bad_prison:
2885 kfree(pool);
2886bad_pool:
2887 if (dm_pool_metadata_close(pmd))
2888 DMWARN("%s: dm_pool_metadata_close() failed.", __func__);
2889
2890 return err_p;
2891}
2892
2893static void __pool_inc(struct pool *pool)
2894{
2895 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
2896 pool->ref_count++;
2897}
2898
2899static void __pool_dec(struct pool *pool)
2900{
2901 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
2902 BUG_ON(!pool->ref_count);
2903 if (!--pool->ref_count)
2904 __pool_destroy(pool);
2905}
2906
2907static struct pool *__pool_find(struct mapped_device *pool_md,
2908 struct block_device *metadata_dev,
Joe Thornbere49e5822012-07-27 15:08:16 +01002909 unsigned long block_size, int read_only,
2910 char **error, int *created)
Joe Thornber991d9fa2011-10-31 20:21:18 +00002911{
2912 struct pool *pool = __pool_table_lookup_metadata_dev(metadata_dev);
2913
2914 if (pool) {
Mike Snitzerf09996c2012-07-27 15:07:59 +01002915 if (pool->pool_md != pool_md) {
2916 *error = "metadata device already in use by a pool";
Joe Thornber991d9fa2011-10-31 20:21:18 +00002917 return ERR_PTR(-EBUSY);
Mike Snitzerf09996c2012-07-27 15:07:59 +01002918 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00002919 __pool_inc(pool);
2920
2921 } else {
2922 pool = __pool_table_lookup(pool_md);
2923 if (pool) {
Mike Snitzerf09996c2012-07-27 15:07:59 +01002924 if (pool->md_dev != metadata_dev) {
2925 *error = "different pool cannot replace a pool";
Joe Thornber991d9fa2011-10-31 20:21:18 +00002926 return ERR_PTR(-EINVAL);
Mike Snitzerf09996c2012-07-27 15:07:59 +01002927 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00002928 __pool_inc(pool);
2929
Joe Thornber67e2e2b2012-03-28 18:41:29 +01002930 } else {
Joe Thornbere49e5822012-07-27 15:08:16 +01002931 pool = pool_create(pool_md, metadata_dev, block_size, read_only, error);
Joe Thornber67e2e2b2012-03-28 18:41:29 +01002932 *created = 1;
2933 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00002934 }
2935
2936 return pool;
2937}
2938
2939/*----------------------------------------------------------------
2940 * Pool target methods
2941 *--------------------------------------------------------------*/
2942static void pool_dtr(struct dm_target *ti)
2943{
2944 struct pool_c *pt = ti->private;
2945
2946 mutex_lock(&dm_thin_pool_table.mutex);
2947
2948 unbind_control_target(pt->pool, ti);
2949 __pool_dec(pt->pool);
2950 dm_put_device(ti, pt->metadata_dev);
2951 dm_put_device(ti, pt->data_dev);
2952 kfree(pt);
2953
2954 mutex_unlock(&dm_thin_pool_table.mutex);
2955}
2956
Joe Thornber991d9fa2011-10-31 20:21:18 +00002957static int parse_pool_features(struct dm_arg_set *as, struct pool_features *pf,
2958 struct dm_target *ti)
2959{
2960 int r;
2961 unsigned argc;
2962 const char *arg_name;
2963
2964 static struct dm_arg _args[] = {
Mike Snitzer74aa45c2014-01-15 19:07:58 -05002965 {0, 4, "Invalid number of pool feature arguments"},
Joe Thornber991d9fa2011-10-31 20:21:18 +00002966 };
2967
2968 /*
2969 * No feature arguments supplied.
2970 */
2971 if (!as->argc)
2972 return 0;
2973
2974 r = dm_read_arg_group(_args, as, &argc, &ti->error);
2975 if (r)
2976 return -EINVAL;
2977
2978 while (argc && !r) {
2979 arg_name = dm_shift_arg(as);
2980 argc--;
2981
Joe Thornbere49e5822012-07-27 15:08:16 +01002982 if (!strcasecmp(arg_name, "skip_block_zeroing"))
Mike Snitzer9bc142d2012-09-26 23:45:46 +01002983 pf->zero_new_blocks = false;
Joe Thornber991d9fa2011-10-31 20:21:18 +00002984
Joe Thornbere49e5822012-07-27 15:08:16 +01002985 else if (!strcasecmp(arg_name, "ignore_discard"))
Mike Snitzer9bc142d2012-09-26 23:45:46 +01002986 pf->discard_enabled = false;
Joe Thornbere49e5822012-07-27 15:08:16 +01002987
2988 else if (!strcasecmp(arg_name, "no_discard_passdown"))
Mike Snitzer9bc142d2012-09-26 23:45:46 +01002989 pf->discard_passdown = false;
Joe Thornbere49e5822012-07-27 15:08:16 +01002990
2991 else if (!strcasecmp(arg_name, "read_only"))
2992 pf->mode = PM_READ_ONLY;
2993
Mike Snitzer787a996c2013-12-06 16:21:43 -05002994 else if (!strcasecmp(arg_name, "error_if_no_space"))
2995 pf->error_if_no_space = true;
2996
Joe Thornbere49e5822012-07-27 15:08:16 +01002997 else {
2998 ti->error = "Unrecognised pool feature requested";
2999 r = -EINVAL;
3000 break;
3001 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00003002 }
3003
3004 return r;
3005}
3006
Joe Thornberac8c3f32013-05-10 14:37:21 +01003007static void metadata_low_callback(void *context)
3008{
3009 struct pool *pool = context;
3010
3011 DMWARN("%s: reached low water mark for metadata device: sending event.",
3012 dm_device_name(pool->pool_md));
3013
3014 dm_table_event(pool->ti->table);
3015}
3016
Mike Snitzer7d489352014-02-12 23:58:15 -05003017static sector_t get_dev_size(struct block_device *bdev)
Joe Thornberb17446d2013-05-10 14:37:18 +01003018{
Mike Snitzer7d489352014-02-12 23:58:15 -05003019 return i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
3020}
3021
3022static void warn_if_metadata_device_too_big(struct block_device *bdev)
3023{
3024 sector_t metadata_dev_size = get_dev_size(bdev);
Joe Thornberb17446d2013-05-10 14:37:18 +01003025 char buffer[BDEVNAME_SIZE];
3026
Mike Snitzer7d489352014-02-12 23:58:15 -05003027 if (metadata_dev_size > THIN_METADATA_MAX_SECTORS_WARNING)
Joe Thornberb17446d2013-05-10 14:37:18 +01003028 DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.",
3029 bdevname(bdev, buffer), THIN_METADATA_MAX_SECTORS);
Mike Snitzer7d489352014-02-12 23:58:15 -05003030}
3031
3032static sector_t get_metadata_dev_size(struct block_device *bdev)
3033{
3034 sector_t metadata_dev_size = get_dev_size(bdev);
3035
3036 if (metadata_dev_size > THIN_METADATA_MAX_SECTORS)
3037 metadata_dev_size = THIN_METADATA_MAX_SECTORS;
Joe Thornberb17446d2013-05-10 14:37:18 +01003038
3039 return metadata_dev_size;
3040}
3041
Joe Thornber24347e92013-05-10 14:37:19 +01003042static dm_block_t get_metadata_dev_size_in_blocks(struct block_device *bdev)
3043{
3044 sector_t metadata_dev_size = get_metadata_dev_size(bdev);
3045
Mike Snitzer7d489352014-02-12 23:58:15 -05003046 sector_div(metadata_dev_size, THIN_METADATA_BLOCK_SIZE);
Joe Thornber24347e92013-05-10 14:37:19 +01003047
3048 return metadata_dev_size;
3049}
3050
Joe Thornber991d9fa2011-10-31 20:21:18 +00003051/*
Joe Thornberac8c3f32013-05-10 14:37:21 +01003052 * When a metadata threshold is crossed a dm event is triggered, and
3053 * userland should respond by growing the metadata device. We could let
3054 * userland set the threshold, like we do with the data threshold, but I'm
3055 * not sure they know enough to do this well.
3056 */
3057static dm_block_t calc_metadata_threshold(struct pool_c *pt)
3058{
3059 /*
3060 * 4M is ample for all ops with the possible exception of thin
3061 * device deletion which is harmless if it fails (just retry the
3062 * delete after you've grown the device).
3063 */
3064 dm_block_t quarter = get_metadata_dev_size_in_blocks(pt->metadata_dev->bdev) / 4;
3065 return min((dm_block_t)1024ULL /* 4M */, quarter);
3066}
3067
3068/*
Joe Thornber991d9fa2011-10-31 20:21:18 +00003069 * thin-pool <metadata dev> <data dev>
3070 * <data block size (sectors)>
3071 * <low water mark (blocks)>
3072 * [<#feature args> [<arg>]*]
3073 *
3074 * Optional feature arguments are:
3075 * skip_block_zeroing: skips the zeroing of newly-provisioned blocks.
Joe Thornber67e2e2b2012-03-28 18:41:29 +01003076 * ignore_discard: disable discard
3077 * no_discard_passdown: don't pass discards down to the data device
Mike Snitzer787a996c2013-12-06 16:21:43 -05003078 * read_only: Don't allow any changes to be made to the pool metadata.
3079 * error_if_no_space: error IOs, instead of queueing, if no space.
Joe Thornber991d9fa2011-10-31 20:21:18 +00003080 */
3081static int pool_ctr(struct dm_target *ti, unsigned argc, char **argv)
3082{
Joe Thornber67e2e2b2012-03-28 18:41:29 +01003083 int r, pool_created = 0;
Joe Thornber991d9fa2011-10-31 20:21:18 +00003084 struct pool_c *pt;
3085 struct pool *pool;
3086 struct pool_features pf;
3087 struct dm_arg_set as;
3088 struct dm_dev *data_dev;
3089 unsigned long block_size;
3090 dm_block_t low_water_blocks;
3091 struct dm_dev *metadata_dev;
Joe Thornber5d0db962013-05-10 14:37:19 +01003092 fmode_t metadata_mode;
Joe Thornber991d9fa2011-10-31 20:21:18 +00003093
3094 /*
3095 * FIXME Remove validation from scope of lock.
3096 */
3097 mutex_lock(&dm_thin_pool_table.mutex);
3098
3099 if (argc < 4) {
3100 ti->error = "Invalid argument count";
3101 r = -EINVAL;
3102 goto out_unlock;
3103 }
Joe Thornber5d0db962013-05-10 14:37:19 +01003104
Joe Thornber991d9fa2011-10-31 20:21:18 +00003105 as.argc = argc;
3106 as.argv = argv;
3107
Joe Thornber5d0db962013-05-10 14:37:19 +01003108 /*
3109 * Set default pool features.
3110 */
3111 pool_features_init(&pf);
3112
3113 dm_consume_args(&as, 4);
3114 r = parse_pool_features(&as, &pf, ti);
3115 if (r)
3116 goto out_unlock;
3117
3118 metadata_mode = FMODE_READ | ((pf.mode == PM_READ_ONLY) ? 0 : FMODE_WRITE);
3119 r = dm_get_device(ti, argv[0], metadata_mode, &metadata_dev);
Joe Thornber991d9fa2011-10-31 20:21:18 +00003120 if (r) {
3121 ti->error = "Error opening metadata block device";
3122 goto out_unlock;
3123 }
Mike Snitzer7d489352014-02-12 23:58:15 -05003124 warn_if_metadata_device_too_big(metadata_dev->bdev);
Joe Thornber991d9fa2011-10-31 20:21:18 +00003125
3126 r = dm_get_device(ti, argv[1], FMODE_READ | FMODE_WRITE, &data_dev);
3127 if (r) {
3128 ti->error = "Error getting data device";
3129 goto out_metadata;
3130 }
3131
3132 if (kstrtoul(argv[2], 10, &block_size) || !block_size ||
3133 block_size < DATA_DEV_BLOCK_SIZE_MIN_SECTORS ||
3134 block_size > DATA_DEV_BLOCK_SIZE_MAX_SECTORS ||
Mike Snitzer55f2b8b2012-07-27 15:08:02 +01003135 block_size & (DATA_DEV_BLOCK_SIZE_MIN_SECTORS - 1)) {
Joe Thornber991d9fa2011-10-31 20:21:18 +00003136 ti->error = "Invalid block size";
3137 r = -EINVAL;
3138 goto out;
3139 }
3140
3141 if (kstrtoull(argv[3], 10, (unsigned long long *)&low_water_blocks)) {
3142 ti->error = "Invalid low water mark";
3143 r = -EINVAL;
3144 goto out;
3145 }
3146
Joe Thornber991d9fa2011-10-31 20:21:18 +00003147 pt = kzalloc(sizeof(*pt), GFP_KERNEL);
3148 if (!pt) {
3149 r = -ENOMEM;
3150 goto out;
3151 }
3152
3153 pool = __pool_find(dm_table_get_md(ti->table), metadata_dev->bdev,
Joe Thornbere49e5822012-07-27 15:08:16 +01003154 block_size, pf.mode == PM_READ_ONLY, &ti->error, &pool_created);
Joe Thornber991d9fa2011-10-31 20:21:18 +00003155 if (IS_ERR(pool)) {
3156 r = PTR_ERR(pool);
3157 goto out_free_pt;
3158 }
3159
Joe Thornber67e2e2b2012-03-28 18:41:29 +01003160 /*
3161 * 'pool_created' reflects whether this is the first table load.
3162 * Top level discard support is not allowed to be changed after
3163 * initial load. This would require a pool reload to trigger thin
3164 * device changes.
3165 */
3166 if (!pool_created && pf.discard_enabled != pool->pf.discard_enabled) {
3167 ti->error = "Discard support cannot be disabled once enabled";
3168 r = -EINVAL;
3169 goto out_flags_changed;
3170 }
3171
Joe Thornber991d9fa2011-10-31 20:21:18 +00003172 pt->pool = pool;
3173 pt->ti = ti;
3174 pt->metadata_dev = metadata_dev;
3175 pt->data_dev = data_dev;
3176 pt->low_water_blocks = low_water_blocks;
Mike Snitzer0424caa2012-09-26 23:45:47 +01003177 pt->adjusted_pf = pt->requested_pf = pf;
Alasdair G Kergon55a62ee2013-03-01 22:45:47 +00003178 ti->num_flush_bios = 1;
Mike Snitzer9bc142d2012-09-26 23:45:46 +01003179
Joe Thornber67e2e2b2012-03-28 18:41:29 +01003180 /*
3181 * Only need to enable discards if the pool should pass
3182 * them down to the data device. The thin device's discard
3183 * processing will cause mappings to be removed from the btree.
3184 */
Mike Snitzerb60ab992013-09-19 18:49:11 -04003185 ti->discard_zeroes_data_unsupported = true;
Joe Thornber67e2e2b2012-03-28 18:41:29 +01003186 if (pf.discard_enabled && pf.discard_passdown) {
Alasdair G Kergon55a62ee2013-03-01 22:45:47 +00003187 ti->num_discard_bios = 1;
Mike Snitzer9bc142d2012-09-26 23:45:46 +01003188
Joe Thornber67e2e2b2012-03-28 18:41:29 +01003189 /*
3190 * Setting 'discards_supported' circumvents the normal
3191 * stacking of discard limits (this keeps the pool and
3192 * thin devices' discard limits consistent).
3193 */
Alasdair G Kergon0ac55482012-07-27 15:08:08 +01003194 ti->discards_supported = true;
Joe Thornber67e2e2b2012-03-28 18:41:29 +01003195 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00003196 ti->private = pt;
3197
Joe Thornberac8c3f32013-05-10 14:37:21 +01003198 r = dm_pool_register_metadata_threshold(pt->pool->pmd,
3199 calc_metadata_threshold(pt),
3200 metadata_low_callback,
3201 pool);
3202 if (r)
Mike Snitzerba306702015-10-13 12:04:28 -04003203 goto out_flags_changed;
Joe Thornberac8c3f32013-05-10 14:37:21 +01003204
Joe Thornber991d9fa2011-10-31 20:21:18 +00003205 pt->callbacks.congested_fn = pool_is_congested;
3206 dm_table_add_target_callbacks(ti->table, &pt->callbacks);
3207
3208 mutex_unlock(&dm_thin_pool_table.mutex);
3209
3210 return 0;
3211
Joe Thornber67e2e2b2012-03-28 18:41:29 +01003212out_flags_changed:
3213 __pool_dec(pool);
Joe Thornber991d9fa2011-10-31 20:21:18 +00003214out_free_pt:
3215 kfree(pt);
3216out:
3217 dm_put_device(ti, data_dev);
3218out_metadata:
3219 dm_put_device(ti, metadata_dev);
3220out_unlock:
3221 mutex_unlock(&dm_thin_pool_table.mutex);
3222
3223 return r;
3224}
3225
Mikulas Patocka7de3ee52012-12-21 20:23:41 +00003226static int pool_map(struct dm_target *ti, struct bio *bio)
Joe Thornber991d9fa2011-10-31 20:21:18 +00003227{
3228 int r;
3229 struct pool_c *pt = ti->private;
3230 struct pool *pool = pt->pool;
3231 unsigned long flags;
3232
3233 /*
3234 * As this is a singleton target, ti->begin is always zero.
3235 */
3236 spin_lock_irqsave(&pool->lock, flags);
3237 bio->bi_bdev = pt->data_dev->bdev;
3238 r = DM_MAPIO_REMAPPED;
3239 spin_unlock_irqrestore(&pool->lock, flags);
3240
3241 return r;
3242}
3243
Joe Thornberb17446d2013-05-10 14:37:18 +01003244static int maybe_resize_data_dev(struct dm_target *ti, bool *need_commit)
3245{
3246 int r;
3247 struct pool_c *pt = ti->private;
3248 struct pool *pool = pt->pool;
3249 sector_t data_size = ti->len;
3250 dm_block_t sb_data_size;
3251
3252 *need_commit = false;
3253
3254 (void) sector_div(data_size, pool->sectors_per_block);
3255
3256 r = dm_pool_get_data_dev_size(pool->pmd, &sb_data_size);
3257 if (r) {
Mike Snitzer4fa59712013-08-21 17:30:40 -04003258 DMERR("%s: failed to retrieve data device size",
3259 dm_device_name(pool->pool_md));
Joe Thornberb17446d2013-05-10 14:37:18 +01003260 return r;
3261 }
3262
3263 if (data_size < sb_data_size) {
Mike Snitzer4fa59712013-08-21 17:30:40 -04003264 DMERR("%s: pool target (%llu blocks) too small: expected %llu",
3265 dm_device_name(pool->pool_md),
Joe Thornberb17446d2013-05-10 14:37:18 +01003266 (unsigned long long)data_size, sb_data_size);
3267 return -EINVAL;
3268
3269 } else if (data_size > sb_data_size) {
Mike Snitzer07f2b6e2014-02-14 11:58:41 -05003270 if (dm_pool_metadata_needs_check(pool->pmd)) {
3271 DMERR("%s: unable to grow the data device until repaired.",
3272 dm_device_name(pool->pool_md));
3273 return 0;
3274 }
3275
Mike Snitzer6f7f51d2013-12-04 10:25:53 -05003276 if (sb_data_size)
3277 DMINFO("%s: growing the data device from %llu to %llu blocks",
3278 dm_device_name(pool->pool_md),
3279 sb_data_size, (unsigned long long)data_size);
Joe Thornberb17446d2013-05-10 14:37:18 +01003280 r = dm_pool_resize_data_dev(pool->pmd, data_size);
3281 if (r) {
Joe Thornberb5330652013-12-04 19:51:33 -05003282 metadata_operation_failed(pool, "dm_pool_resize_data_dev", r);
Joe Thornberb17446d2013-05-10 14:37:18 +01003283 return r;
3284 }
3285
3286 *need_commit = true;
3287 }
3288
3289 return 0;
3290}
3291
Joe Thornber24347e92013-05-10 14:37:19 +01003292static int maybe_resize_metadata_dev(struct dm_target *ti, bool *need_commit)
3293{
3294 int r;
3295 struct pool_c *pt = ti->private;
3296 struct pool *pool = pt->pool;
3297 dm_block_t metadata_dev_size, sb_metadata_dev_size;
3298
3299 *need_commit = false;
3300
Alasdair G Kergon610bba82013-05-19 18:57:50 +01003301 metadata_dev_size = get_metadata_dev_size_in_blocks(pool->md_dev);
Joe Thornber24347e92013-05-10 14:37:19 +01003302
3303 r = dm_pool_get_metadata_dev_size(pool->pmd, &sb_metadata_dev_size);
3304 if (r) {
Mike Snitzer4fa59712013-08-21 17:30:40 -04003305 DMERR("%s: failed to retrieve metadata device size",
3306 dm_device_name(pool->pool_md));
Joe Thornber24347e92013-05-10 14:37:19 +01003307 return r;
3308 }
3309
3310 if (metadata_dev_size < sb_metadata_dev_size) {
Mike Snitzer4fa59712013-08-21 17:30:40 -04003311 DMERR("%s: metadata device (%llu blocks) too small: expected %llu",
3312 dm_device_name(pool->pool_md),
Joe Thornber24347e92013-05-10 14:37:19 +01003313 metadata_dev_size, sb_metadata_dev_size);
3314 return -EINVAL;
3315
3316 } else if (metadata_dev_size > sb_metadata_dev_size) {
Mike Snitzer07f2b6e2014-02-14 11:58:41 -05003317 if (dm_pool_metadata_needs_check(pool->pmd)) {
3318 DMERR("%s: unable to grow the metadata device until repaired.",
3319 dm_device_name(pool->pool_md));
3320 return 0;
3321 }
3322
Mike Snitzer7d489352014-02-12 23:58:15 -05003323 warn_if_metadata_device_too_big(pool->md_dev);
Mike Snitzer6f7f51d2013-12-04 10:25:53 -05003324 DMINFO("%s: growing the metadata device from %llu to %llu blocks",
3325 dm_device_name(pool->pool_md),
3326 sb_metadata_dev_size, metadata_dev_size);
Joe Thornber24347e92013-05-10 14:37:19 +01003327 r = dm_pool_resize_metadata_dev(pool->pmd, metadata_dev_size);
3328 if (r) {
Joe Thornberb5330652013-12-04 19:51:33 -05003329 metadata_operation_failed(pool, "dm_pool_resize_metadata_dev", r);
Joe Thornber24347e92013-05-10 14:37:19 +01003330 return r;
3331 }
3332
3333 *need_commit = true;
3334 }
3335
3336 return 0;
3337}
3338
Joe Thornber991d9fa2011-10-31 20:21:18 +00003339/*
3340 * Retrieves the number of blocks of the data device from
3341 * the superblock and compares it to the actual device size,
3342 * thus resizing the data device in case it has grown.
3343 *
3344 * This both copes with opening preallocated data devices in the ctr
3345 * being followed by a resume
3346 * -and-
3347 * calling the resume method individually after userspace has
3348 * grown the data device in reaction to a table event.
3349 */
3350static int pool_preresume(struct dm_target *ti)
3351{
3352 int r;
Joe Thornber24347e92013-05-10 14:37:19 +01003353 bool need_commit1, need_commit2;
Joe Thornber991d9fa2011-10-31 20:21:18 +00003354 struct pool_c *pt = ti->private;
3355 struct pool *pool = pt->pool;
Joe Thornber991d9fa2011-10-31 20:21:18 +00003356
3357 /*
3358 * Take control of the pool object.
3359 */
3360 r = bind_control_target(pool, ti);
3361 if (r)
3362 return r;
3363
Joe Thornberb17446d2013-05-10 14:37:18 +01003364 r = maybe_resize_data_dev(ti, &need_commit1);
3365 if (r)
Joe Thornber991d9fa2011-10-31 20:21:18 +00003366 return r;
Joe Thornber991d9fa2011-10-31 20:21:18 +00003367
Joe Thornber24347e92013-05-10 14:37:19 +01003368 r = maybe_resize_metadata_dev(ti, &need_commit2);
3369 if (r)
3370 return r;
3371
3372 if (need_commit1 || need_commit2)
Joe Thornber020cc3b2013-12-04 15:05:36 -05003373 (void) commit(pool);
Joe Thornber991d9fa2011-10-31 20:21:18 +00003374
3375 return 0;
3376}
3377
Mike Snitzer583024d2014-10-28 20:58:45 -04003378static void pool_suspend_active_thins(struct pool *pool)
3379{
3380 struct thin_c *tc;
3381
3382 /* Suspend all active thin devices */
3383 tc = get_first_thin(pool);
3384 while (tc) {
3385 dm_internal_suspend_noflush(tc->thin_md);
3386 tc = get_next_thin(pool, tc);
3387 }
3388}
3389
3390static void pool_resume_active_thins(struct pool *pool)
3391{
3392 struct thin_c *tc;
3393
3394 /* Resume all active thin devices */
3395 tc = get_first_thin(pool);
3396 while (tc) {
3397 dm_internal_resume(tc->thin_md);
3398 tc = get_next_thin(pool, tc);
3399 }
3400}
3401
Joe Thornber991d9fa2011-10-31 20:21:18 +00003402static void pool_resume(struct dm_target *ti)
3403{
3404 struct pool_c *pt = ti->private;
3405 struct pool *pool = pt->pool;
3406 unsigned long flags;
3407
Mike Snitzer583024d2014-10-28 20:58:45 -04003408 /*
3409 * Must requeue active_thins' bios and then resume
3410 * active_thins _before_ clearing 'suspend' flag.
3411 */
3412 requeue_bios(pool);
3413 pool_resume_active_thins(pool);
3414
Joe Thornber991d9fa2011-10-31 20:21:18 +00003415 spin_lock_irqsave(&pool->lock, flags);
Joe Thornber88a66212013-12-04 20:16:12 -05003416 pool->low_water_triggered = false;
Mike Snitzer80e96c52014-11-07 15:09:46 -05003417 pool->suspended = false;
Joe Thornber991d9fa2011-10-31 20:21:18 +00003418 spin_unlock_irqrestore(&pool->lock, flags);
Mike Snitzer80e96c52014-11-07 15:09:46 -05003419
Joe Thornber905e51b2012-03-28 18:41:27 +01003420 do_waker(&pool->waker.work);
Joe Thornber991d9fa2011-10-31 20:21:18 +00003421}
3422
Mike Snitzer80e96c52014-11-07 15:09:46 -05003423static void pool_presuspend(struct dm_target *ti)
3424{
3425 struct pool_c *pt = ti->private;
3426 struct pool *pool = pt->pool;
3427 unsigned long flags;
3428
3429 spin_lock_irqsave(&pool->lock, flags);
3430 pool->suspended = true;
3431 spin_unlock_irqrestore(&pool->lock, flags);
Mike Snitzer583024d2014-10-28 20:58:45 -04003432
3433 pool_suspend_active_thins(pool);
Mike Snitzer80e96c52014-11-07 15:09:46 -05003434}
3435
3436static void pool_presuspend_undo(struct dm_target *ti)
3437{
3438 struct pool_c *pt = ti->private;
3439 struct pool *pool = pt->pool;
3440 unsigned long flags;
3441
Mike Snitzer583024d2014-10-28 20:58:45 -04003442 pool_resume_active_thins(pool);
3443
Mike Snitzer80e96c52014-11-07 15:09:46 -05003444 spin_lock_irqsave(&pool->lock, flags);
3445 pool->suspended = false;
3446 spin_unlock_irqrestore(&pool->lock, flags);
3447}
3448
Joe Thornber991d9fa2011-10-31 20:21:18 +00003449static void pool_postsuspend(struct dm_target *ti)
3450{
Joe Thornber991d9fa2011-10-31 20:21:18 +00003451 struct pool_c *pt = ti->private;
3452 struct pool *pool = pt->pool;
3453
Nikolay Borisov18d03e82015-12-17 18:03:35 +02003454 cancel_delayed_work_sync(&pool->waker);
3455 cancel_delayed_work_sync(&pool->no_space_timeout);
Joe Thornber991d9fa2011-10-31 20:21:18 +00003456 flush_workqueue(pool->wq);
Joe Thornber020cc3b2013-12-04 15:05:36 -05003457 (void) commit(pool);
Joe Thornber991d9fa2011-10-31 20:21:18 +00003458}
3459
3460static int check_arg_count(unsigned argc, unsigned args_required)
3461{
3462 if (argc != args_required) {
3463 DMWARN("Message received with %u arguments instead of %u.",
3464 argc, args_required);
3465 return -EINVAL;
3466 }
3467
3468 return 0;
3469}
3470
3471static int read_dev_id(char *arg, dm_thin_id *dev_id, int warning)
3472{
3473 if (!kstrtoull(arg, 10, (unsigned long long *)dev_id) &&
3474 *dev_id <= MAX_DEV_ID)
3475 return 0;
3476
3477 if (warning)
3478 DMWARN("Message received with invalid device id: %s", arg);
3479
3480 return -EINVAL;
3481}
3482
3483static int process_create_thin_mesg(unsigned argc, char **argv, struct pool *pool)
3484{
3485 dm_thin_id dev_id;
3486 int r;
3487
3488 r = check_arg_count(argc, 2);
3489 if (r)
3490 return r;
3491
3492 r = read_dev_id(argv[1], &dev_id, 1);
3493 if (r)
3494 return r;
3495
3496 r = dm_pool_create_thin(pool->pmd, dev_id);
3497 if (r) {
3498 DMWARN("Creation of new thinly-provisioned device with id %s failed.",
3499 argv[1]);
3500 return r;
3501 }
3502
3503 return 0;
3504}
3505
3506static int process_create_snap_mesg(unsigned argc, char **argv, struct pool *pool)
3507{
3508 dm_thin_id dev_id;
3509 dm_thin_id origin_dev_id;
3510 int r;
3511
3512 r = check_arg_count(argc, 3);
3513 if (r)
3514 return r;
3515
3516 r = read_dev_id(argv[1], &dev_id, 1);
3517 if (r)
3518 return r;
3519
3520 r = read_dev_id(argv[2], &origin_dev_id, 1);
3521 if (r)
3522 return r;
3523
3524 r = dm_pool_create_snap(pool->pmd, dev_id, origin_dev_id);
3525 if (r) {
3526 DMWARN("Creation of new snapshot %s of device %s failed.",
3527 argv[1], argv[2]);
3528 return r;
3529 }
3530
3531 return 0;
3532}
3533
3534static int process_delete_mesg(unsigned argc, char **argv, struct pool *pool)
3535{
3536 dm_thin_id dev_id;
3537 int r;
3538
3539 r = check_arg_count(argc, 2);
3540 if (r)
3541 return r;
3542
3543 r = read_dev_id(argv[1], &dev_id, 1);
3544 if (r)
3545 return r;
3546
3547 r = dm_pool_delete_thin_device(pool->pmd, dev_id);
3548 if (r)
3549 DMWARN("Deletion of thin device %s failed.", argv[1]);
3550
3551 return r;
3552}
3553
3554static int process_set_transaction_id_mesg(unsigned argc, char **argv, struct pool *pool)
3555{
3556 dm_thin_id old_id, new_id;
3557 int r;
3558
3559 r = check_arg_count(argc, 3);
3560 if (r)
3561 return r;
3562
3563 if (kstrtoull(argv[1], 10, (unsigned long long *)&old_id)) {
3564 DMWARN("set_transaction_id message: Unrecognised id %s.", argv[1]);
3565 return -EINVAL;
3566 }
3567
3568 if (kstrtoull(argv[2], 10, (unsigned long long *)&new_id)) {
3569 DMWARN("set_transaction_id message: Unrecognised new id %s.", argv[2]);
3570 return -EINVAL;
3571 }
3572
3573 r = dm_pool_set_metadata_transaction_id(pool->pmd, old_id, new_id);
3574 if (r) {
3575 DMWARN("Failed to change transaction id from %s to %s.",
3576 argv[1], argv[2]);
3577 return r;
3578 }
3579
3580 return 0;
3581}
3582
Joe Thornbercc8394d2012-06-03 00:30:01 +01003583static int process_reserve_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool)
3584{
3585 int r;
3586
3587 r = check_arg_count(argc, 1);
3588 if (r)
3589 return r;
3590
Joe Thornber020cc3b2013-12-04 15:05:36 -05003591 (void) commit(pool);
Joe Thornber0d200ae2012-07-03 12:55:31 +01003592
Joe Thornbercc8394d2012-06-03 00:30:01 +01003593 r = dm_pool_reserve_metadata_snap(pool->pmd);
3594 if (r)
3595 DMWARN("reserve_metadata_snap message failed.");
3596
3597 return r;
3598}
3599
3600static int process_release_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool)
3601{
3602 int r;
3603
3604 r = check_arg_count(argc, 1);
3605 if (r)
3606 return r;
3607
3608 r = dm_pool_release_metadata_snap(pool->pmd);
3609 if (r)
3610 DMWARN("release_metadata_snap message failed.");
3611
3612 return r;
3613}
3614
Joe Thornber991d9fa2011-10-31 20:21:18 +00003615/*
3616 * Messages supported:
3617 * create_thin <dev_id>
3618 * create_snap <dev_id> <origin_id>
3619 * delete <dev_id>
Joe Thornber991d9fa2011-10-31 20:21:18 +00003620 * set_transaction_id <current_trans_id> <new_trans_id>
Joe Thornbercc8394d2012-06-03 00:30:01 +01003621 * reserve_metadata_snap
3622 * release_metadata_snap
Joe Thornber991d9fa2011-10-31 20:21:18 +00003623 */
3624static int pool_message(struct dm_target *ti, unsigned argc, char **argv)
3625{
3626 int r = -EINVAL;
3627 struct pool_c *pt = ti->private;
3628 struct pool *pool = pt->pool;
3629
Joe Thornber2a7eaea2015-01-26 11:38:21 +00003630 if (get_pool_mode(pool) >= PM_READ_ONLY) {
3631 DMERR("%s: unable to service pool target messages in READ_ONLY or FAIL mode",
3632 dm_device_name(pool->pool_md));
Mike Snitzerfd467692015-06-09 12:31:26 -04003633 return -EOPNOTSUPP;
Joe Thornber2a7eaea2015-01-26 11:38:21 +00003634 }
3635
Joe Thornber991d9fa2011-10-31 20:21:18 +00003636 if (!strcasecmp(argv[0], "create_thin"))
3637 r = process_create_thin_mesg(argc, argv, pool);
3638
3639 else if (!strcasecmp(argv[0], "create_snap"))
3640 r = process_create_snap_mesg(argc, argv, pool);
3641
3642 else if (!strcasecmp(argv[0], "delete"))
3643 r = process_delete_mesg(argc, argv, pool);
3644
3645 else if (!strcasecmp(argv[0], "set_transaction_id"))
3646 r = process_set_transaction_id_mesg(argc, argv, pool);
3647
Joe Thornbercc8394d2012-06-03 00:30:01 +01003648 else if (!strcasecmp(argv[0], "reserve_metadata_snap"))
3649 r = process_reserve_metadata_snap_mesg(argc, argv, pool);
3650
3651 else if (!strcasecmp(argv[0], "release_metadata_snap"))
3652 r = process_release_metadata_snap_mesg(argc, argv, pool);
3653
Joe Thornber991d9fa2011-10-31 20:21:18 +00003654 else
3655 DMWARN("Unrecognised thin pool target message received: %s", argv[0]);
3656
Joe Thornbere49e5822012-07-27 15:08:16 +01003657 if (!r)
Joe Thornber020cc3b2013-12-04 15:05:36 -05003658 (void) commit(pool);
Joe Thornber991d9fa2011-10-31 20:21:18 +00003659
3660 return r;
3661}
3662
Joe Thornbere49e5822012-07-27 15:08:16 +01003663static void emit_flags(struct pool_features *pf, char *result,
3664 unsigned sz, unsigned maxlen)
3665{
3666 unsigned count = !pf->zero_new_blocks + !pf->discard_enabled +
Mike Snitzer787a996c2013-12-06 16:21:43 -05003667 !pf->discard_passdown + (pf->mode == PM_READ_ONLY) +
3668 pf->error_if_no_space;
Joe Thornbere49e5822012-07-27 15:08:16 +01003669 DMEMIT("%u ", count);
3670
3671 if (!pf->zero_new_blocks)
3672 DMEMIT("skip_block_zeroing ");
3673
3674 if (!pf->discard_enabled)
3675 DMEMIT("ignore_discard ");
3676
3677 if (!pf->discard_passdown)
3678 DMEMIT("no_discard_passdown ");
3679
3680 if (pf->mode == PM_READ_ONLY)
3681 DMEMIT("read_only ");
Mike Snitzer787a996c2013-12-06 16:21:43 -05003682
3683 if (pf->error_if_no_space)
3684 DMEMIT("error_if_no_space ");
Joe Thornbere49e5822012-07-27 15:08:16 +01003685}
3686
Joe Thornber991d9fa2011-10-31 20:21:18 +00003687/*
3688 * Status line is:
3689 * <transaction id> <used metadata sectors>/<total metadata sectors>
3690 * <used data sectors>/<total data sectors> <held metadata root>
Mike Snitzere4c78e22015-07-15 11:40:24 -04003691 * <pool mode> <discard config> <no space config> <needs_check>
Joe Thornber991d9fa2011-10-31 20:21:18 +00003692 */
Mikulas Patockafd7c0922013-03-01 22:45:44 +00003693static void pool_status(struct dm_target *ti, status_type_t type,
3694 unsigned status_flags, char *result, unsigned maxlen)
Joe Thornber991d9fa2011-10-31 20:21:18 +00003695{
Joe Thornbere49e5822012-07-27 15:08:16 +01003696 int r;
Joe Thornber991d9fa2011-10-31 20:21:18 +00003697 unsigned sz = 0;
3698 uint64_t transaction_id;
3699 dm_block_t nr_free_blocks_data;
3700 dm_block_t nr_free_blocks_metadata;
3701 dm_block_t nr_blocks_data;
3702 dm_block_t nr_blocks_metadata;
3703 dm_block_t held_root;
3704 char buf[BDEVNAME_SIZE];
3705 char buf2[BDEVNAME_SIZE];
3706 struct pool_c *pt = ti->private;
3707 struct pool *pool = pt->pool;
3708
3709 switch (type) {
3710 case STATUSTYPE_INFO:
Joe Thornbere49e5822012-07-27 15:08:16 +01003711 if (get_pool_mode(pool) == PM_FAIL) {
3712 DMEMIT("Fail");
3713 break;
3714 }
3715
Alasdair G Kergon1f4e0ff2012-07-27 15:08:16 +01003716 /* Commit to ensure statistics aren't out-of-date */
3717 if (!(status_flags & DM_STATUS_NOFLUSH_FLAG) && !dm_suspended(ti))
Joe Thornber020cc3b2013-12-04 15:05:36 -05003718 (void) commit(pool);
Alasdair G Kergon1f4e0ff2012-07-27 15:08:16 +01003719
Mikulas Patockafd7c0922013-03-01 22:45:44 +00003720 r = dm_pool_get_metadata_transaction_id(pool->pmd, &transaction_id);
3721 if (r) {
Mike Snitzer4fa59712013-08-21 17:30:40 -04003722 DMERR("%s: dm_pool_get_metadata_transaction_id returned %d",
3723 dm_device_name(pool->pool_md), r);
Mikulas Patockafd7c0922013-03-01 22:45:44 +00003724 goto err;
3725 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00003726
Mikulas Patockafd7c0922013-03-01 22:45:44 +00003727 r = dm_pool_get_free_metadata_block_count(pool->pmd, &nr_free_blocks_metadata);
3728 if (r) {
Mike Snitzer4fa59712013-08-21 17:30:40 -04003729 DMERR("%s: dm_pool_get_free_metadata_block_count returned %d",
3730 dm_device_name(pool->pool_md), r);
Mikulas Patockafd7c0922013-03-01 22:45:44 +00003731 goto err;
3732 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00003733
3734 r = dm_pool_get_metadata_dev_size(pool->pmd, &nr_blocks_metadata);
Mikulas Patockafd7c0922013-03-01 22:45:44 +00003735 if (r) {
Mike Snitzer4fa59712013-08-21 17:30:40 -04003736 DMERR("%s: dm_pool_get_metadata_dev_size returned %d",
3737 dm_device_name(pool->pool_md), r);
Mikulas Patockafd7c0922013-03-01 22:45:44 +00003738 goto err;
3739 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00003740
Mikulas Patockafd7c0922013-03-01 22:45:44 +00003741 r = dm_pool_get_free_block_count(pool->pmd, &nr_free_blocks_data);
3742 if (r) {
Mike Snitzer4fa59712013-08-21 17:30:40 -04003743 DMERR("%s: dm_pool_get_free_block_count returned %d",
3744 dm_device_name(pool->pool_md), r);
Mikulas Patockafd7c0922013-03-01 22:45:44 +00003745 goto err;
3746 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00003747
3748 r = dm_pool_get_data_dev_size(pool->pmd, &nr_blocks_data);
Mikulas Patockafd7c0922013-03-01 22:45:44 +00003749 if (r) {
Mike Snitzer4fa59712013-08-21 17:30:40 -04003750 DMERR("%s: dm_pool_get_data_dev_size returned %d",
3751 dm_device_name(pool->pool_md), r);
Mikulas Patockafd7c0922013-03-01 22:45:44 +00003752 goto err;
3753 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00003754
Joe Thornbercc8394d2012-06-03 00:30:01 +01003755 r = dm_pool_get_metadata_snap(pool->pmd, &held_root);
Mikulas Patockafd7c0922013-03-01 22:45:44 +00003756 if (r) {
Mike Snitzer4fa59712013-08-21 17:30:40 -04003757 DMERR("%s: dm_pool_get_metadata_snap returned %d",
3758 dm_device_name(pool->pool_md), r);
Mikulas Patockafd7c0922013-03-01 22:45:44 +00003759 goto err;
3760 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00003761
3762 DMEMIT("%llu %llu/%llu %llu/%llu ",
3763 (unsigned long long)transaction_id,
3764 (unsigned long long)(nr_blocks_metadata - nr_free_blocks_metadata),
3765 (unsigned long long)nr_blocks_metadata,
3766 (unsigned long long)(nr_blocks_data - nr_free_blocks_data),
3767 (unsigned long long)nr_blocks_data);
3768
3769 if (held_root)
Joe Thornbere49e5822012-07-27 15:08:16 +01003770 DMEMIT("%llu ", held_root);
Joe Thornber991d9fa2011-10-31 20:21:18 +00003771 else
Joe Thornbere49e5822012-07-27 15:08:16 +01003772 DMEMIT("- ");
3773
Joe Thornber3e1a0692014-03-03 16:03:26 +00003774 if (pool->pf.mode == PM_OUT_OF_DATA_SPACE)
3775 DMEMIT("out_of_data_space ");
3776 else if (pool->pf.mode == PM_READ_ONLY)
Joe Thornbere49e5822012-07-27 15:08:16 +01003777 DMEMIT("ro ");
3778 else
3779 DMEMIT("rw ");
3780
Mike Snitzer018debe2012-12-21 20:23:32 +00003781 if (!pool->pf.discard_enabled)
Mike Snitzer787a996c2013-12-06 16:21:43 -05003782 DMEMIT("ignore_discard ");
Mike Snitzer018debe2012-12-21 20:23:32 +00003783 else if (pool->pf.discard_passdown)
Mike Snitzer787a996c2013-12-06 16:21:43 -05003784 DMEMIT("discard_passdown ");
Joe Thornbere49e5822012-07-27 15:08:16 +01003785 else
Mike Snitzer787a996c2013-12-06 16:21:43 -05003786 DMEMIT("no_discard_passdown ");
3787
3788 if (pool->pf.error_if_no_space)
3789 DMEMIT("error_if_no_space ");
3790 else
3791 DMEMIT("queue_if_no_space ");
Joe Thornber991d9fa2011-10-31 20:21:18 +00003792
Mike Snitzere4c78e22015-07-15 11:40:24 -04003793 if (dm_pool_metadata_needs_check(pool->pmd))
3794 DMEMIT("needs_check ");
3795 else
3796 DMEMIT("- ");
3797
Joe Thornber991d9fa2011-10-31 20:21:18 +00003798 break;
3799
3800 case STATUSTYPE_TABLE:
3801 DMEMIT("%s %s %lu %llu ",
3802 format_dev_t(buf, pt->metadata_dev->bdev->bd_dev),
3803 format_dev_t(buf2, pt->data_dev->bdev->bd_dev),
3804 (unsigned long)pool->sectors_per_block,
3805 (unsigned long long)pt->low_water_blocks);
Mike Snitzer0424caa2012-09-26 23:45:47 +01003806 emit_flags(&pt->requested_pf, result, sz, maxlen);
Joe Thornber991d9fa2011-10-31 20:21:18 +00003807 break;
3808 }
Mikulas Patockafd7c0922013-03-01 22:45:44 +00003809 return;
Joe Thornber991d9fa2011-10-31 20:21:18 +00003810
Mikulas Patockafd7c0922013-03-01 22:45:44 +00003811err:
3812 DMEMIT("Error");
Joe Thornber991d9fa2011-10-31 20:21:18 +00003813}
3814
3815static int pool_iterate_devices(struct dm_target *ti,
3816 iterate_devices_callout_fn fn, void *data)
3817{
3818 struct pool_c *pt = ti->private;
3819
3820 return fn(ti, pt->data_dev, 0, ti->len, data);
3821}
3822
Joe Thornber991d9fa2011-10-31 20:21:18 +00003823static void pool_io_hints(struct dm_target *ti, struct queue_limits *limits)
3824{
3825 struct pool_c *pt = ti->private;
3826 struct pool *pool = pt->pool;
Mike Snitzer604ea902014-10-09 18:43:25 -04003827 sector_t io_opt_sectors = limits->io_opt >> SECTOR_SHIFT;
3828
3829 /*
Mike Snitzerd200c302014-11-20 18:07:43 -05003830 * If max_sectors is smaller than pool->sectors_per_block adjust it
3831 * to the highest possible power-of-2 factor of pool->sectors_per_block.
3832 * This is especially beneficial when the pool's data device is a RAID
3833 * device that has a full stripe width that matches pool->sectors_per_block
3834 * -- because even though partial RAID stripe-sized IOs will be issued to a
3835 * single RAID stripe; when aggregated they will end on a full RAID stripe
3836 * boundary.. which avoids additional partial RAID stripe writes cascading
Mike Snitzer604ea902014-10-09 18:43:25 -04003837 */
Mike Snitzer604ea902014-10-09 18:43:25 -04003838 if (limits->max_sectors < pool->sectors_per_block) {
3839 while (!is_factor(pool->sectors_per_block, limits->max_sectors)) {
3840 if ((limits->max_sectors & (limits->max_sectors - 1)) == 0)
3841 limits->max_sectors--;
3842 limits->max_sectors = rounddown_pow_of_two(limits->max_sectors);
3843 }
Mike Snitzer604ea902014-10-09 18:43:25 -04003844 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00003845
Mike Snitzer0cc67cd2013-08-20 15:02:41 -04003846 /*
3847 * If the system-determined stacked limits are compatible with the
3848 * pool's blocksize (io_opt is a factor) do not override them.
3849 */
3850 if (io_opt_sectors < pool->sectors_per_block ||
Mike Snitzer604ea902014-10-09 18:43:25 -04003851 !is_factor(io_opt_sectors, pool->sectors_per_block)) {
3852 if (is_factor(pool->sectors_per_block, limits->max_sectors))
3853 blk_limits_io_min(limits, limits->max_sectors << SECTOR_SHIFT);
3854 else
3855 blk_limits_io_min(limits, pool->sectors_per_block << SECTOR_SHIFT);
Mike Snitzer0cc67cd2013-08-20 15:02:41 -04003856 blk_limits_io_opt(limits, pool->sectors_per_block << SECTOR_SHIFT);
3857 }
Mike Snitzer0424caa2012-09-26 23:45:47 +01003858
3859 /*
3860 * pt->adjusted_pf is a staging area for the actual features to use.
3861 * They get transferred to the live pool in bind_control_target()
3862 * called from pool_preresume().
3863 */
Mike Snitzerb60ab992013-09-19 18:49:11 -04003864 if (!pt->adjusted_pf.discard_enabled) {
3865 /*
3866 * Must explicitly disallow stacking discard limits otherwise the
3867 * block layer will stack them if pool's data device has support.
3868 * QUEUE_FLAG_DISCARD wouldn't be set but there is no way for the
3869 * user to see that, so make sure to set all discard limits to 0.
3870 */
3871 limits->discard_granularity = 0;
Mike Snitzer0424caa2012-09-26 23:45:47 +01003872 return;
Mike Snitzerb60ab992013-09-19 18:49:11 -04003873 }
Mike Snitzer0424caa2012-09-26 23:45:47 +01003874
3875 disable_passdown_if_not_supported(pt);
3876
Joe Thornber34fbcf62015-04-16 12:58:35 +01003877 /*
3878 * The pool uses the same discard limits as the underlying data
3879 * device. DM core has already set this up.
3880 */
Joe Thornber991d9fa2011-10-31 20:21:18 +00003881}
3882
3883static struct target_type pool_target = {
3884 .name = "thin-pool",
3885 .features = DM_TARGET_SINGLETON | DM_TARGET_ALWAYS_WRITEABLE |
3886 DM_TARGET_IMMUTABLE,
Joe Thornber202bae52016-05-04 14:12:42 -04003887 .version = {1, 19, 0},
Joe Thornber991d9fa2011-10-31 20:21:18 +00003888 .module = THIS_MODULE,
3889 .ctr = pool_ctr,
3890 .dtr = pool_dtr,
3891 .map = pool_map,
Mike Snitzer80e96c52014-11-07 15:09:46 -05003892 .presuspend = pool_presuspend,
3893 .presuspend_undo = pool_presuspend_undo,
Joe Thornber991d9fa2011-10-31 20:21:18 +00003894 .postsuspend = pool_postsuspend,
3895 .preresume = pool_preresume,
3896 .resume = pool_resume,
3897 .message = pool_message,
3898 .status = pool_status,
Joe Thornber991d9fa2011-10-31 20:21:18 +00003899 .iterate_devices = pool_iterate_devices,
3900 .io_hints = pool_io_hints,
3901};
3902
3903/*----------------------------------------------------------------
3904 * Thin target methods
3905 *--------------------------------------------------------------*/
Joe Thornberb10ebd32014-04-08 11:29:01 +01003906static void thin_get(struct thin_c *tc)
3907{
3908 atomic_inc(&tc->refcount);
3909}
3910
3911static void thin_put(struct thin_c *tc)
3912{
3913 if (atomic_dec_and_test(&tc->refcount))
3914 complete(&tc->can_destroy);
3915}
3916
Joe Thornber991d9fa2011-10-31 20:21:18 +00003917static void thin_dtr(struct dm_target *ti)
3918{
3919 struct thin_c *tc = ti->private;
Mike Snitzerc140e1c2014-03-20 21:17:14 -04003920 unsigned long flags;
3921
3922 spin_lock_irqsave(&tc->pool->lock, flags);
3923 list_del_rcu(&tc->list);
3924 spin_unlock_irqrestore(&tc->pool->lock, flags);
3925 synchronize_rcu();
Joe Thornber991d9fa2011-10-31 20:21:18 +00003926
Mikulas Patocka17181fb2014-11-05 17:00:13 -05003927 thin_put(tc);
3928 wait_for_completion(&tc->can_destroy);
3929
Joe Thornber991d9fa2011-10-31 20:21:18 +00003930 mutex_lock(&dm_thin_pool_table.mutex);
3931
3932 __pool_dec(tc->pool);
3933 dm_pool_close_thin_device(tc->td);
3934 dm_put_device(ti, tc->pool_dev);
Joe Thornber2dd9c252012-03-28 18:41:28 +01003935 if (tc->origin_dev)
3936 dm_put_device(ti, tc->origin_dev);
Joe Thornber991d9fa2011-10-31 20:21:18 +00003937 kfree(tc);
3938
3939 mutex_unlock(&dm_thin_pool_table.mutex);
3940}
3941
3942/*
3943 * Thin target parameters:
3944 *
Joe Thornber2dd9c252012-03-28 18:41:28 +01003945 * <pool_dev> <dev_id> [origin_dev]
Joe Thornber991d9fa2011-10-31 20:21:18 +00003946 *
3947 * pool_dev: the path to the pool (eg, /dev/mapper/my_pool)
3948 * dev_id: the internal device identifier
Joe Thornber2dd9c252012-03-28 18:41:28 +01003949 * origin_dev: a device external to the pool that should act as the origin
Joe Thornber67e2e2b2012-03-28 18:41:29 +01003950 *
3951 * If the pool device has discards disabled, they get disabled for the thin
3952 * device as well.
Joe Thornber991d9fa2011-10-31 20:21:18 +00003953 */
3954static int thin_ctr(struct dm_target *ti, unsigned argc, char **argv)
3955{
3956 int r;
3957 struct thin_c *tc;
Joe Thornber2dd9c252012-03-28 18:41:28 +01003958 struct dm_dev *pool_dev, *origin_dev;
Joe Thornber991d9fa2011-10-31 20:21:18 +00003959 struct mapped_device *pool_md;
Joe Thornber5e3283e2014-04-08 11:08:41 +01003960 unsigned long flags;
Joe Thornber991d9fa2011-10-31 20:21:18 +00003961
3962 mutex_lock(&dm_thin_pool_table.mutex);
3963
Joe Thornber2dd9c252012-03-28 18:41:28 +01003964 if (argc != 2 && argc != 3) {
Joe Thornber991d9fa2011-10-31 20:21:18 +00003965 ti->error = "Invalid argument count";
3966 r = -EINVAL;
3967 goto out_unlock;
3968 }
3969
3970 tc = ti->private = kzalloc(sizeof(*tc), GFP_KERNEL);
3971 if (!tc) {
3972 ti->error = "Out of memory";
3973 r = -ENOMEM;
3974 goto out_unlock;
3975 }
Mike Snitzer583024d2014-10-28 20:58:45 -04003976 tc->thin_md = dm_table_get_md(ti->table);
Mike Snitzerc140e1c2014-03-20 21:17:14 -04003977 spin_lock_init(&tc->lock);
Joe Thornbera374bb22014-10-10 13:43:14 +01003978 INIT_LIST_HEAD(&tc->deferred_cells);
Mike Snitzerc140e1c2014-03-20 21:17:14 -04003979 bio_list_init(&tc->deferred_bio_list);
3980 bio_list_init(&tc->retry_on_resume_list);
Mike Snitzer67324ea2014-03-21 18:33:41 -04003981 tc->sort_bio_list = RB_ROOT;
Joe Thornber991d9fa2011-10-31 20:21:18 +00003982
Joe Thornber2dd9c252012-03-28 18:41:28 +01003983 if (argc == 3) {
3984 r = dm_get_device(ti, argv[2], FMODE_READ, &origin_dev);
3985 if (r) {
3986 ti->error = "Error opening origin device";
3987 goto bad_origin_dev;
3988 }
3989 tc->origin_dev = origin_dev;
3990 }
3991
Joe Thornber991d9fa2011-10-31 20:21:18 +00003992 r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &pool_dev);
3993 if (r) {
3994 ti->error = "Error opening pool device";
3995 goto bad_pool_dev;
3996 }
3997 tc->pool_dev = pool_dev;
3998
3999 if (read_dev_id(argv[1], (unsigned long long *)&tc->dev_id, 0)) {
4000 ti->error = "Invalid device id";
4001 r = -EINVAL;
4002 goto bad_common;
4003 }
4004
4005 pool_md = dm_get_md(tc->pool_dev->bdev->bd_dev);
4006 if (!pool_md) {
4007 ti->error = "Couldn't get pool mapped device";
4008 r = -EINVAL;
4009 goto bad_common;
4010 }
4011
4012 tc->pool = __pool_table_lookup(pool_md);
4013 if (!tc->pool) {
4014 ti->error = "Couldn't find pool object";
4015 r = -EINVAL;
4016 goto bad_pool_lookup;
4017 }
4018 __pool_inc(tc->pool);
4019
Joe Thornbere49e5822012-07-27 15:08:16 +01004020 if (get_pool_mode(tc->pool) == PM_FAIL) {
4021 ti->error = "Couldn't open thin device, Pool is in fail mode";
Mike Snitzer1acacc02014-02-19 20:32:33 -05004022 r = -EINVAL;
Mike Snitzer80e96c52014-11-07 15:09:46 -05004023 goto bad_pool;
Joe Thornbere49e5822012-07-27 15:08:16 +01004024 }
4025
Joe Thornber991d9fa2011-10-31 20:21:18 +00004026 r = dm_pool_open_thin_device(tc->pool->pmd, tc->dev_id, &tc->td);
4027 if (r) {
4028 ti->error = "Couldn't open thin internal device";
Mike Snitzer80e96c52014-11-07 15:09:46 -05004029 goto bad_pool;
Joe Thornber991d9fa2011-10-31 20:21:18 +00004030 }
4031
Mike Snitzer542f9032012-07-27 15:08:00 +01004032 r = dm_set_target_max_io_len(ti, tc->pool->sectors_per_block);
4033 if (r)
Mike Snitzer80e96c52014-11-07 15:09:46 -05004034 goto bad;
Mike Snitzer542f9032012-07-27 15:08:00 +01004035
Alasdair G Kergon55a62ee2013-03-01 22:45:47 +00004036 ti->num_flush_bios = 1;
Joe Thornber16ad3d12012-07-27 15:08:07 +01004037 ti->flush_supported = true;
Mike Snitzer30187e12016-01-31 13:28:26 -05004038 ti->per_io_data_size = sizeof(struct dm_thin_endio_hook);
Joe Thornber67e2e2b2012-03-28 18:41:29 +01004039
4040 /* In case the pool supports discards, pass them on. */
Mike Snitzerb60ab992013-09-19 18:49:11 -04004041 ti->discard_zeroes_data_unsupported = true;
Joe Thornber67e2e2b2012-03-28 18:41:29 +01004042 if (tc->pool->pf.discard_enabled) {
Alasdair G Kergon0ac55482012-07-27 15:08:08 +01004043 ti->discards_supported = true;
Alasdair G Kergon55a62ee2013-03-01 22:45:47 +00004044 ti->num_discard_bios = 1;
Joe Thornber34fbcf62015-04-16 12:58:35 +01004045 ti->split_discard_bios = false;
Joe Thornber67e2e2b2012-03-28 18:41:29 +01004046 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00004047
Joe Thornber991d9fa2011-10-31 20:21:18 +00004048 mutex_unlock(&dm_thin_pool_table.mutex);
4049
Joe Thornber5e3283e2014-04-08 11:08:41 +01004050 spin_lock_irqsave(&tc->pool->lock, flags);
Mike Snitzer80e96c52014-11-07 15:09:46 -05004051 if (tc->pool->suspended) {
4052 spin_unlock_irqrestore(&tc->pool->lock, flags);
4053 mutex_lock(&dm_thin_pool_table.mutex); /* reacquire for __pool_dec */
4054 ti->error = "Unable to activate thin device while pool is suspended";
4055 r = -EINVAL;
4056 goto bad;
4057 }
Marc Dionne2b94e892014-12-17 07:59:59 -05004058 atomic_set(&tc->refcount, 1);
4059 init_completion(&tc->can_destroy);
Mike Snitzerc140e1c2014-03-20 21:17:14 -04004060 list_add_tail_rcu(&tc->list, &tc->pool->active_thins);
Joe Thornber5e3283e2014-04-08 11:08:41 +01004061 spin_unlock_irqrestore(&tc->pool->lock, flags);
Mike Snitzerc140e1c2014-03-20 21:17:14 -04004062 /*
4063 * This synchronize_rcu() call is needed here otherwise we risk a
4064 * wake_worker() call finding no bios to process (because the newly
4065 * added tc isn't yet visible). So this reduces latency since we
4066 * aren't then dependent on the periodic commit to wake_worker().
4067 */
4068 synchronize_rcu();
4069
Mike Snitzer80e96c52014-11-07 15:09:46 -05004070 dm_put(pool_md);
4071
Joe Thornber991d9fa2011-10-31 20:21:18 +00004072 return 0;
4073
Mike Snitzer80e96c52014-11-07 15:09:46 -05004074bad:
Mike Snitzer1acacc02014-02-19 20:32:33 -05004075 dm_pool_close_thin_device(tc->td);
Mike Snitzer80e96c52014-11-07 15:09:46 -05004076bad_pool:
Joe Thornber991d9fa2011-10-31 20:21:18 +00004077 __pool_dec(tc->pool);
4078bad_pool_lookup:
4079 dm_put(pool_md);
4080bad_common:
4081 dm_put_device(ti, tc->pool_dev);
4082bad_pool_dev:
Joe Thornber2dd9c252012-03-28 18:41:28 +01004083 if (tc->origin_dev)
4084 dm_put_device(ti, tc->origin_dev);
4085bad_origin_dev:
Joe Thornber991d9fa2011-10-31 20:21:18 +00004086 kfree(tc);
4087out_unlock:
4088 mutex_unlock(&dm_thin_pool_table.mutex);
4089
4090 return r;
4091}
4092
Mikulas Patocka7de3ee52012-12-21 20:23:41 +00004093static int thin_map(struct dm_target *ti, struct bio *bio)
Joe Thornber991d9fa2011-10-31 20:21:18 +00004094{
Kent Overstreet4f024f32013-10-11 15:44:27 -07004095 bio->bi_iter.bi_sector = dm_target_offset(ti, bio->bi_iter.bi_sector);
Joe Thornber991d9fa2011-10-31 20:21:18 +00004096
Mikulas Patocka7de3ee52012-12-21 20:23:41 +00004097 return thin_bio_map(ti, bio);
Joe Thornber991d9fa2011-10-31 20:21:18 +00004098}
4099
Mikulas Patocka7de3ee52012-12-21 20:23:41 +00004100static int thin_endio(struct dm_target *ti, struct bio *bio, int err)
Joe Thornbereb2aa482012-03-28 18:41:28 +01004101{
4102 unsigned long flags;
Mikulas Patocka59c3d2c2012-12-21 20:23:40 +00004103 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
Joe Thornbereb2aa482012-03-28 18:41:28 +01004104 struct list_head work;
Mike Snitzera24c2562012-06-03 00:30:00 +01004105 struct dm_thin_new_mapping *m, *tmp;
Joe Thornbereb2aa482012-03-28 18:41:28 +01004106 struct pool *pool = h->tc->pool;
4107
4108 if (h->shared_read_entry) {
4109 INIT_LIST_HEAD(&work);
Mike Snitzer44feb382012-10-12 21:02:10 +01004110 dm_deferred_entry_dec(h->shared_read_entry, &work);
Joe Thornbereb2aa482012-03-28 18:41:28 +01004111
4112 spin_lock_irqsave(&pool->lock, flags);
4113 list_for_each_entry_safe(m, tmp, &work, list) {
4114 list_del(&m->list);
Joe Thornber50f3c3e2014-06-13 13:57:09 +01004115 __complete_mapping_preparation(m);
Joe Thornbereb2aa482012-03-28 18:41:28 +01004116 }
4117 spin_unlock_irqrestore(&pool->lock, flags);
4118 }
4119
Joe Thornber104655f2012-03-28 18:41:28 +01004120 if (h->all_io_entry) {
4121 INIT_LIST_HEAD(&work);
Mike Snitzer44feb382012-10-12 21:02:10 +01004122 dm_deferred_entry_dec(h->all_io_entry, &work);
Joe Thornber563af182012-12-21 20:23:31 +00004123 if (!list_empty(&work)) {
4124 spin_lock_irqsave(&pool->lock, flags);
4125 list_for_each_entry_safe(m, tmp, &work, list)
Mike Snitzerdaec3382013-12-11 14:01:20 -05004126 list_add_tail(&m->list, &pool->prepared_discards);
Joe Thornber563af182012-12-21 20:23:31 +00004127 spin_unlock_irqrestore(&pool->lock, flags);
4128 wake_worker(pool);
4129 }
Joe Thornber104655f2012-03-28 18:41:28 +01004130 }
4131
Joe Thornber34fbcf62015-04-16 12:58:35 +01004132 if (h->cell)
4133 cell_defer_no_holder(h->tc, h->cell);
4134
Joe Thornbereb2aa482012-03-28 18:41:28 +01004135 return 0;
4136}
4137
Joe Thornber738211f2014-03-03 15:52:28 +00004138static void thin_presuspend(struct dm_target *ti)
4139{
4140 struct thin_c *tc = ti->private;
4141
4142 if (dm_noflush_suspending(ti))
4143 noflush_work(tc, do_noflush_start);
4144}
4145
Joe Thornber991d9fa2011-10-31 20:21:18 +00004146static void thin_postsuspend(struct dm_target *ti)
4147{
Joe Thornber738211f2014-03-03 15:52:28 +00004148 struct thin_c *tc = ti->private;
4149
4150 /*
4151 * The dm_noflush_suspending flag has been cleared by now, so
4152 * unfortunately we must always run this.
4153 */
4154 noflush_work(tc, do_noflush_stop);
Joe Thornber991d9fa2011-10-31 20:21:18 +00004155}
4156
Joe Thornbere5aea7b2014-06-13 14:47:24 +01004157static int thin_preresume(struct dm_target *ti)
4158{
4159 struct thin_c *tc = ti->private;
4160
4161 if (tc->origin_dev)
4162 tc->origin_size = get_dev_size(tc->origin_dev->bdev);
4163
4164 return 0;
4165}
4166
Joe Thornber991d9fa2011-10-31 20:21:18 +00004167/*
4168 * <nr mapped sectors> <highest mapped sector>
4169 */
Mikulas Patockafd7c0922013-03-01 22:45:44 +00004170static void thin_status(struct dm_target *ti, status_type_t type,
4171 unsigned status_flags, char *result, unsigned maxlen)
Joe Thornber991d9fa2011-10-31 20:21:18 +00004172{
4173 int r;
4174 ssize_t sz = 0;
4175 dm_block_t mapped, highest;
4176 char buf[BDEVNAME_SIZE];
4177 struct thin_c *tc = ti->private;
4178
Joe Thornbere49e5822012-07-27 15:08:16 +01004179 if (get_pool_mode(tc->pool) == PM_FAIL) {
4180 DMEMIT("Fail");
Mikulas Patockafd7c0922013-03-01 22:45:44 +00004181 return;
Joe Thornbere49e5822012-07-27 15:08:16 +01004182 }
4183
Joe Thornber991d9fa2011-10-31 20:21:18 +00004184 if (!tc->td)
4185 DMEMIT("-");
4186 else {
4187 switch (type) {
4188 case STATUSTYPE_INFO:
4189 r = dm_thin_get_mapped_count(tc->td, &mapped);
Mikulas Patockafd7c0922013-03-01 22:45:44 +00004190 if (r) {
4191 DMERR("dm_thin_get_mapped_count returned %d", r);
4192 goto err;
4193 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00004194
4195 r = dm_thin_get_highest_mapped_block(tc->td, &highest);
Mikulas Patockafd7c0922013-03-01 22:45:44 +00004196 if (r < 0) {
4197 DMERR("dm_thin_get_highest_mapped_block returned %d", r);
4198 goto err;
4199 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00004200
4201 DMEMIT("%llu ", mapped * tc->pool->sectors_per_block);
4202 if (r)
4203 DMEMIT("%llu", ((highest + 1) *
4204 tc->pool->sectors_per_block) - 1);
4205 else
4206 DMEMIT("-");
4207 break;
4208
4209 case STATUSTYPE_TABLE:
4210 DMEMIT("%s %lu",
4211 format_dev_t(buf, tc->pool_dev->bdev->bd_dev),
4212 (unsigned long) tc->dev_id);
Joe Thornber2dd9c252012-03-28 18:41:28 +01004213 if (tc->origin_dev)
4214 DMEMIT(" %s", format_dev_t(buf, tc->origin_dev->bdev->bd_dev));
Joe Thornber991d9fa2011-10-31 20:21:18 +00004215 break;
4216 }
4217 }
4218
Mikulas Patockafd7c0922013-03-01 22:45:44 +00004219 return;
4220
4221err:
4222 DMEMIT("Error");
Joe Thornber991d9fa2011-10-31 20:21:18 +00004223}
4224
4225static int thin_iterate_devices(struct dm_target *ti,
4226 iterate_devices_callout_fn fn, void *data)
4227{
Mike Snitzer55f2b8b2012-07-27 15:08:02 +01004228 sector_t blocks;
Joe Thornber991d9fa2011-10-31 20:21:18 +00004229 struct thin_c *tc = ti->private;
Mike Snitzer55f2b8b2012-07-27 15:08:02 +01004230 struct pool *pool = tc->pool;
Joe Thornber991d9fa2011-10-31 20:21:18 +00004231
4232 /*
4233 * We can't call dm_pool_get_data_dev_size() since that blocks. So
4234 * we follow a more convoluted path through to the pool's target.
4235 */
Mike Snitzer55f2b8b2012-07-27 15:08:02 +01004236 if (!pool->ti)
Joe Thornber991d9fa2011-10-31 20:21:18 +00004237 return 0; /* nothing is bound */
4238
Mike Snitzer55f2b8b2012-07-27 15:08:02 +01004239 blocks = pool->ti->len;
4240 (void) sector_div(blocks, pool->sectors_per_block);
Joe Thornber991d9fa2011-10-31 20:21:18 +00004241 if (blocks)
Mike Snitzer55f2b8b2012-07-27 15:08:02 +01004242 return fn(ti, tc->pool_dev, 0, pool->sectors_per_block * blocks, data);
Joe Thornber991d9fa2011-10-31 20:21:18 +00004243
4244 return 0;
4245}
4246
Joe Thornber34fbcf62015-04-16 12:58:35 +01004247static void thin_io_hints(struct dm_target *ti, struct queue_limits *limits)
4248{
4249 struct thin_c *tc = ti->private;
4250 struct pool *pool = tc->pool;
Mike Snitzer21607672015-09-08 08:56:13 -04004251
Mike Snitzer0fcb04d2015-11-23 13:44:38 -05004252 if (!pool->pf.discard_enabled)
4253 return;
Joe Thornber34fbcf62015-04-16 12:58:35 +01004254
4255 limits->discard_granularity = pool->sectors_per_block << SECTOR_SHIFT;
4256 limits->max_discard_sectors = 2048 * 1024 * 16; /* 16G */
4257}
4258
Joe Thornber991d9fa2011-10-31 20:21:18 +00004259static struct target_type thin_target = {
4260 .name = "thin",
Joe Thornber202bae52016-05-04 14:12:42 -04004261 .version = {1, 19, 0},
Joe Thornber991d9fa2011-10-31 20:21:18 +00004262 .module = THIS_MODULE,
4263 .ctr = thin_ctr,
4264 .dtr = thin_dtr,
4265 .map = thin_map,
Joe Thornbereb2aa482012-03-28 18:41:28 +01004266 .end_io = thin_endio,
Joe Thornbere5aea7b2014-06-13 14:47:24 +01004267 .preresume = thin_preresume,
Joe Thornber738211f2014-03-03 15:52:28 +00004268 .presuspend = thin_presuspend,
Joe Thornber991d9fa2011-10-31 20:21:18 +00004269 .postsuspend = thin_postsuspend,
4270 .status = thin_status,
4271 .iterate_devices = thin_iterate_devices,
Joe Thornber34fbcf62015-04-16 12:58:35 +01004272 .io_hints = thin_io_hints,
Joe Thornber991d9fa2011-10-31 20:21:18 +00004273};
4274
4275/*----------------------------------------------------------------*/
4276
4277static int __init dm_thin_init(void)
4278{
4279 int r;
4280
4281 pool_table_init();
4282
4283 r = dm_register_target(&thin_target);
4284 if (r)
4285 return r;
4286
4287 r = dm_register_target(&pool_target);
4288 if (r)
Mike Snitzera24c2562012-06-03 00:30:00 +01004289 goto bad_pool_target;
4290
4291 r = -ENOMEM;
4292
Mike Snitzera24c2562012-06-03 00:30:00 +01004293 _new_mapping_cache = KMEM_CACHE(dm_thin_new_mapping, 0);
4294 if (!_new_mapping_cache)
4295 goto bad_new_mapping_cache;
4296
Mike Snitzera24c2562012-06-03 00:30:00 +01004297 return 0;
4298
Mike Snitzera24c2562012-06-03 00:30:00 +01004299bad_new_mapping_cache:
Mike Snitzera24c2562012-06-03 00:30:00 +01004300 dm_unregister_target(&pool_target);
4301bad_pool_target:
4302 dm_unregister_target(&thin_target);
Joe Thornber991d9fa2011-10-31 20:21:18 +00004303
4304 return r;
4305}
4306
4307static void dm_thin_exit(void)
4308{
4309 dm_unregister_target(&thin_target);
4310 dm_unregister_target(&pool_target);
Mike Snitzera24c2562012-06-03 00:30:00 +01004311
Mike Snitzera24c2562012-06-03 00:30:00 +01004312 kmem_cache_destroy(_new_mapping_cache);
Joe Thornber991d9fa2011-10-31 20:21:18 +00004313}
4314
4315module_init(dm_thin_init);
4316module_exit(dm_thin_exit);
4317
Mike Snitzer80c57892014-05-20 13:38:33 -04004318module_param_named(no_space_timeout, no_space_timeout_secs, uint, S_IRUGO | S_IWUSR);
4319MODULE_PARM_DESC(no_space_timeout, "Out of data space queue IO timeout in seconds");
4320
Alasdair G Kergon7cab8bf2012-05-12 01:43:19 +01004321MODULE_DESCRIPTION(DM_NAME " thin provisioning target");
Joe Thornber991d9fa2011-10-31 20:21:18 +00004322MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
4323MODULE_LICENSE("GPL");