blob: 6b11a266299f14a724afd178604c9d87a31232a1 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002 * Copyright (C) 2001-2002 Sistina Software (UK) Limited.
3 *
4 * This file is released under the GPL.
5 */
6
7#include <linux/blkdev.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -07008#include <linux/device-mapper.h>
Mikulas Patocka90fa1522009-01-06 03:04:54 +00009#include <linux/delay.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070010#include <linux/fs.h>
11#include <linux/init.h>
12#include <linux/kdev_t.h>
13#include <linux/list.h>
Nikos Tsironisf79ae412019-03-17 14:22:57 +020014#include <linux/list_bl.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070015#include <linux/mempool.h>
16#include <linux/module.h>
17#include <linux/slab.h>
18#include <linux/vmalloc.h>
vignesh babu6f3c3f02007-10-19 22:38:44 +010019#include <linux/log2.h>
Alasdair G Kergona765e202008-04-24 22:02:01 +010020#include <linux/dm-kcopyd.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070021
Mikulas Patockab735fed2015-02-26 11:40:35 -050022#include "dm.h"
23
Jonathan Brassowaea53d92009-01-06 03:05:15 +000024#include "dm-exception-store.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070025
Alasdair G Kergon72d94862006-06-26 00:27:35 -070026#define DM_MSG_PREFIX "snapshots"
27
Mikulas Patockad698aa42009-12-10 23:52:30 +000028static const char dm_snapshot_merge_target_name[] = "snapshot-merge";
29
30#define dm_target_is_snapshot_merge(ti) \
31 ((ti)->type->name == dm_snapshot_merge_target_name)
32
Linus Torvalds1da177e2005-04-16 15:20:36 -070033/*
Mikulas Patockacd45daf2008-07-21 12:00:32 +010034 * The size of the mempool used to track chunks in use.
35 */
36#define MIN_IOS 256
37
Jonathan Brassowccc45ea2009-04-02 19:55:34 +010038#define DM_TRACKED_CHUNK_HASH_SIZE 16
39#define DM_TRACKED_CHUNK_HASH(x) ((unsigned long)(x) & \
40 (DM_TRACKED_CHUNK_HASH_SIZE - 1))
41
Jon Brassow191437a2009-12-10 23:52:10 +000042struct dm_exception_table {
Jonathan Brassowccc45ea2009-04-02 19:55:34 +010043 uint32_t hash_mask;
44 unsigned hash_shift;
Nikos Tsironisf79ae412019-03-17 14:22:57 +020045 struct hlist_bl_head *table;
Jonathan Brassowccc45ea2009-04-02 19:55:34 +010046};
47
48struct dm_snapshot {
Nikos Tsironis4ad8d882019-03-17 14:22:56 +020049 struct rw_semaphore lock;
Jonathan Brassowccc45ea2009-04-02 19:55:34 +010050
51 struct dm_dev *origin;
Mike Snitzerfc56f6f2009-12-10 23:52:12 +000052 struct dm_dev *cow;
53
54 struct dm_target *ti;
Jonathan Brassowccc45ea2009-04-02 19:55:34 +010055
56 /* List of snapshots per Origin */
57 struct list_head list;
58
Mike Snitzerd8ddb1c2009-12-10 23:52:35 +000059 /*
60 * You can't use a snapshot if this is 0 (e.g. if full).
61 * A snapshot-merge target never clears this.
62 */
Jonathan Brassowccc45ea2009-04-02 19:55:34 +010063 int valid;
64
Mikulas Patocka76c44f62015-06-21 16:31:33 -040065 /*
66 * The snapshot overflowed because of a write to the snapshot device.
67 * We don't have to invalidate the snapshot in this case, but we need
68 * to prevent further writes.
69 */
70 int snapshot_overflowed;
71
Jonathan Brassowccc45ea2009-04-02 19:55:34 +010072 /* Origin writes don't trigger exceptions until this is set */
73 int active;
74
Jonathan Brassowccc45ea2009-04-02 19:55:34 +010075 atomic_t pending_exceptions_count;
76
Nikos Tsironis3f1637f2019-03-17 14:22:58 +020077 spinlock_t pe_allocation_lock;
78
79 /* Protected by "pe_allocation_lock" */
Mikulas Patocka230c83a2013-11-29 18:13:37 -050080 sector_t exception_start_sequence;
81
82 /* Protected by kcopyd single-threaded callback */
83 sector_t exception_complete_sequence;
84
85 /*
86 * A list of pending exceptions that completed out of order.
87 * Protected by kcopyd single-threaded callback.
88 */
David Jeffery3db27762018-08-07 16:56:00 -040089 struct rb_root out_of_order_tree;
Mikulas Patocka230c83a2013-11-29 18:13:37 -050090
Kent Overstreet6f1c8192018-05-20 18:25:53 -040091 mempool_t pending_pool;
Mike Snitzer924e6002010-03-06 02:32:33 +000092
Jon Brassow191437a2009-12-10 23:52:10 +000093 struct dm_exception_table pending;
94 struct dm_exception_table complete;
Jonathan Brassowccc45ea2009-04-02 19:55:34 +010095
96 /*
97 * pe_lock protects all pending_exception operations and access
98 * as well as the snapshot_bios list.
99 */
100 spinlock_t pe_lock;
101
Mike Snitzer924e6002010-03-06 02:32:33 +0000102 /* Chunks with outstanding reads */
103 spinlock_t tracked_chunk_lock;
Mike Snitzer924e6002010-03-06 02:32:33 +0000104 struct hlist_head tracked_chunk_hash[DM_TRACKED_CHUNK_HASH_SIZE];
105
Jonathan Brassowccc45ea2009-04-02 19:55:34 +0100106 /* The on disk metadata handler */
107 struct dm_exception_store *store;
108
Mikulas Patockab2155572019-10-02 06:15:53 -0400109 unsigned in_progress;
110 struct wait_queue_head in_progress_wait;
Nikos Tsironis721b1d92018-10-31 17:53:08 -0400111
Jonathan Brassowccc45ea2009-04-02 19:55:34 +0100112 struct dm_kcopyd_client *kcopyd_client;
113
Mike Snitzer924e6002010-03-06 02:32:33 +0000114 /* Wait for events based on state_bits */
115 unsigned long state_bits;
116
117 /* Range of chunks currently being merged. */
118 chunk_t first_merging_chunk;
119 int num_merging_chunks;
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000120
Mike Snitzerd8ddb1c2009-12-10 23:52:35 +0000121 /*
122 * The merge operation failed if this flag is set.
123 * Failure modes are handled as follows:
124 * - I/O error reading the header
125 * => don't load the target; abort.
126 * - Header does not have "valid" flag set
127 * => use the origin; forget about the snapshot.
128 * - I/O error when reading exceptions
129 * => don't load the target; abort.
130 * (We can't use the intermediate origin state.)
131 * - I/O error while merging
132 * => stop merging; set merge_failed; process I/O normally.
133 */
Mike Snitzer2e602382019-06-19 17:05:54 -0400134 bool merge_failed:1;
135
136 bool discard_zeroes_cow:1;
137 bool discard_passdown_origin:1;
Mike Snitzerd8ddb1c2009-12-10 23:52:35 +0000138
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000139 /*
140 * Incoming bios that overlap with chunks being merged must wait
141 * for them to be committed.
142 */
143 struct bio_list bios_queued_during_merge;
Jonathan Brassowccc45ea2009-04-02 19:55:34 +0100144};
145
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000146/*
147 * state_bits:
148 * RUNNING_MERGE - Merge operation is in progress.
149 * SHUTDOWN_MERGE - Set to signal that merge needs to be stopped;
150 * cleared afterwards.
151 */
152#define RUNNING_MERGE 0
153#define SHUTDOWN_MERGE 1
154
Nikos Tsironis721b1d92018-10-31 17:53:08 -0400155/*
156 * Maximum number of chunks being copied on write.
157 *
158 * The value was decided experimentally as a trade-off between memory
159 * consumption, stalling the kernel's workqueues and maintaining a high enough
160 * throughput.
161 */
162#define DEFAULT_COW_THRESHOLD 2048
163
Mikulas Patockab2155572019-10-02 06:15:53 -0400164static unsigned cow_threshold = DEFAULT_COW_THRESHOLD;
165module_param_named(snapshot_cow_threshold, cow_threshold, uint, 0644);
Nikos Tsironis721b1d92018-10-31 17:53:08 -0400166MODULE_PARM_DESC(snapshot_cow_threshold, "Maximum number of chunks being copied on write");
167
Mikulas Patockadf5d2e92013-03-01 22:45:49 +0000168DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(snapshot_copy_throttle,
169 "A percentage of time allocated for copy on write");
170
Mikulas Patockac2411042010-08-12 04:13:51 +0100171struct dm_dev *dm_snap_origin(struct dm_snapshot *s)
172{
173 return s->origin;
174}
175EXPORT_SYMBOL(dm_snap_origin);
176
Mike Snitzerfc56f6f2009-12-10 23:52:12 +0000177struct dm_dev *dm_snap_cow(struct dm_snapshot *s)
178{
179 return s->cow;
180}
181EXPORT_SYMBOL(dm_snap_cow);
182
Jonathan Brassowccc45ea2009-04-02 19:55:34 +0100183static sector_t chunk_to_sector(struct dm_exception_store *store,
184 chunk_t chunk)
185{
186 return chunk << store->chunk_shift;
187}
188
189static int bdev_equal(struct block_device *lhs, struct block_device *rhs)
190{
191 /*
192 * There is only ever one instance of a particular block
193 * device so we can compare pointers safely.
194 */
195 return lhs == rhs;
196}
197
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100198struct dm_snap_pending_exception {
Jon Brassow1d4989c2009-12-10 23:52:10 +0000199 struct dm_exception e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200
201 /*
202 * Origin buffers waiting for this to complete are held
203 * in a bio list
204 */
205 struct bio_list origin_bios;
206 struct bio_list snapshot_bios;
207
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208 /* Pointer back to snapshot context */
209 struct dm_snapshot *snap;
210
211 /*
212 * 1 indicates the exception has already been sent to
213 * kcopyd.
214 */
215 int started;
Mikulas Patockaa6e50b42011-08-02 12:32:04 +0100216
Mikulas Patocka230c83a2013-11-29 18:13:37 -0500217 /* There was copying error. */
218 int copy_error;
219
220 /* A sequence number, it is used for in-order completion. */
221 sector_t exception_sequence;
222
David Jeffery3db27762018-08-07 16:56:00 -0400223 struct rb_node out_of_order_node;
Mikulas Patocka230c83a2013-11-29 18:13:37 -0500224
Mikulas Patockaa6e50b42011-08-02 12:32:04 +0100225 /*
226 * For writing a complete chunk, bypassing the copy.
227 */
228 struct bio *full_bio;
229 bio_end_io_t *full_bio_end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230};
231
232/*
233 * Hash table mapping origin volumes to lists of snapshots and
234 * a lock to protect it
235 */
Christoph Lametere18b8902006-12-06 20:33:20 -0800236static struct kmem_cache *exception_cache;
237static struct kmem_cache *pending_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238
Mikulas Patockacd45daf2008-07-21 12:00:32 +0100239struct dm_snap_tracked_chunk {
240 struct hlist_node node;
241 chunk_t chunk;
242};
243
Mikulas Patockaee180262012-12-21 20:23:41 +0000244static void init_tracked_chunk(struct bio *bio)
245{
246 struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk));
247 INIT_HLIST_NODE(&c->node);
248}
249
250static bool is_bio_tracked(struct bio *bio)
251{
252 struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk));
253 return !hlist_unhashed(&c->node);
254}
255
256static void track_chunk(struct dm_snapshot *s, struct bio *bio, chunk_t chunk)
Mikulas Patockacd45daf2008-07-21 12:00:32 +0100257{
Mikulas Patocka42bc9542012-12-21 20:23:38 +0000258 struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk));
Mikulas Patockacd45daf2008-07-21 12:00:32 +0100259
260 c->chunk = chunk;
261
Mikulas Patocka9aa0c0e2012-12-21 20:23:33 +0000262 spin_lock_irq(&s->tracked_chunk_lock);
Mikulas Patockacd45daf2008-07-21 12:00:32 +0100263 hlist_add_head(&c->node,
264 &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)]);
Mikulas Patocka9aa0c0e2012-12-21 20:23:33 +0000265 spin_unlock_irq(&s->tracked_chunk_lock);
Mikulas Patockacd45daf2008-07-21 12:00:32 +0100266}
267
Mikulas Patockaee180262012-12-21 20:23:41 +0000268static void stop_tracking_chunk(struct dm_snapshot *s, struct bio *bio)
Mikulas Patockacd45daf2008-07-21 12:00:32 +0100269{
Mikulas Patockaee180262012-12-21 20:23:41 +0000270 struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk));
Mikulas Patockacd45daf2008-07-21 12:00:32 +0100271 unsigned long flags;
272
273 spin_lock_irqsave(&s->tracked_chunk_lock, flags);
274 hlist_del(&c->node);
275 spin_unlock_irqrestore(&s->tracked_chunk_lock, flags);
Mikulas Patockacd45daf2008-07-21 12:00:32 +0100276}
277
Mikulas Patockaa8d41b52008-07-21 12:00:34 +0100278static int __chunk_is_tracked(struct dm_snapshot *s, chunk_t chunk)
279{
280 struct dm_snap_tracked_chunk *c;
Mikulas Patockaa8d41b52008-07-21 12:00:34 +0100281 int found = 0;
282
283 spin_lock_irq(&s->tracked_chunk_lock);
284
Sasha Levinb67bfe02013-02-27 17:06:00 -0800285 hlist_for_each_entry(c,
Mikulas Patockaa8d41b52008-07-21 12:00:34 +0100286 &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)], node) {
287 if (c->chunk == chunk) {
288 found = 1;
289 break;
290 }
291 }
292
293 spin_unlock_irq(&s->tracked_chunk_lock);
294
295 return found;
296}
297
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298/*
Mike Snitzer615d1eb2009-12-10 23:52:29 +0000299 * This conflicting I/O is extremely improbable in the caller,
300 * so msleep(1) is sufficient and there is no need for a wait queue.
301 */
302static void __check_for_conflicting_io(struct dm_snapshot *s, chunk_t chunk)
303{
304 while (__chunk_is_tracked(s, chunk))
305 msleep(1);
306}
307
308/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309 * One of these per registered origin, held in the snapshot_origins hash
310 */
311struct origin {
312 /* The origin device */
313 struct block_device *bdev;
314
315 struct list_head hash_list;
316
317 /* List of snapshots for this origin */
318 struct list_head snapshots;
319};
320
321/*
Mikulas Patockab735fed2015-02-26 11:40:35 -0500322 * This structure is allocated for each origin target
323 */
324struct dm_origin {
325 struct dm_dev *dev;
326 struct dm_target *ti;
327 unsigned split_boundary;
328 struct list_head hash_list;
329};
330
331/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 * Size of the hash table for origin volumes. If we make this
333 * the size of the minors list then it should be nearly perfect
334 */
335#define ORIGIN_HASH_SIZE 256
336#define ORIGIN_MASK 0xFF
337static struct list_head *_origins;
Mikulas Patockab735fed2015-02-26 11:40:35 -0500338static struct list_head *_dm_origins;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339static struct rw_semaphore _origins_lock;
340
Mikulas Patocka73dfd072009-12-10 23:52:34 +0000341static DECLARE_WAIT_QUEUE_HEAD(_pending_exceptions_done);
342static DEFINE_SPINLOCK(_pending_exceptions_done_spinlock);
343static uint64_t _pending_exceptions_done_count;
344
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345static int init_origin_hash(void)
346{
347 int i;
348
Kees Cook6da2ec52018-06-12 13:55:00 -0700349 _origins = kmalloc_array(ORIGIN_HASH_SIZE, sizeof(struct list_head),
350 GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351 if (!_origins) {
Mikulas Patockab735fed2015-02-26 11:40:35 -0500352 DMERR("unable to allocate memory for _origins");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353 return -ENOMEM;
354 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355 for (i = 0; i < ORIGIN_HASH_SIZE; i++)
356 INIT_LIST_HEAD(_origins + i);
Mikulas Patockab735fed2015-02-26 11:40:35 -0500357
Kees Cook6da2ec52018-06-12 13:55:00 -0700358 _dm_origins = kmalloc_array(ORIGIN_HASH_SIZE,
359 sizeof(struct list_head),
360 GFP_KERNEL);
Mikulas Patockab735fed2015-02-26 11:40:35 -0500361 if (!_dm_origins) {
362 DMERR("unable to allocate memory for _dm_origins");
363 kfree(_origins);
364 return -ENOMEM;
365 }
366 for (i = 0; i < ORIGIN_HASH_SIZE; i++)
367 INIT_LIST_HEAD(_dm_origins + i);
368
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 init_rwsem(&_origins_lock);
370
371 return 0;
372}
373
374static void exit_origin_hash(void)
375{
376 kfree(_origins);
Mikulas Patockab735fed2015-02-26 11:40:35 -0500377 kfree(_dm_origins);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378}
379
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100380static unsigned origin_hash(struct block_device *bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381{
382 return bdev->bd_dev & ORIGIN_MASK;
383}
384
385static struct origin *__lookup_origin(struct block_device *origin)
386{
387 struct list_head *ol;
388 struct origin *o;
389
390 ol = &_origins[origin_hash(origin)];
391 list_for_each_entry (o, ol, hash_list)
392 if (bdev_equal(o->bdev, origin))
393 return o;
394
395 return NULL;
396}
397
398static void __insert_origin(struct origin *o)
399{
400 struct list_head *sl = &_origins[origin_hash(o->bdev)];
401 list_add_tail(&o->hash_list, sl);
402}
403
Mikulas Patockab735fed2015-02-26 11:40:35 -0500404static struct dm_origin *__lookup_dm_origin(struct block_device *origin)
405{
406 struct list_head *ol;
407 struct dm_origin *o;
408
409 ol = &_dm_origins[origin_hash(origin)];
410 list_for_each_entry (o, ol, hash_list)
411 if (bdev_equal(o->dev->bdev, origin))
412 return o;
413
414 return NULL;
415}
416
417static void __insert_dm_origin(struct dm_origin *o)
418{
419 struct list_head *sl = &_dm_origins[origin_hash(o->dev->bdev)];
420 list_add_tail(&o->hash_list, sl);
421}
422
423static void __remove_dm_origin(struct dm_origin *o)
424{
425 list_del(&o->hash_list);
426}
427
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428/*
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000429 * _origins_lock must be held when calling this function.
430 * Returns number of snapshots registered using the supplied cow device, plus:
431 * snap_src - a snapshot suitable for use as a source of exception handover
432 * snap_dest - a snapshot capable of receiving exception handover.
Mikulas Patocka9d3b15c2009-12-10 23:52:32 +0000433 * snap_merge - an existing snapshot-merge target linked to the same origin.
434 * There can be at most one snapshot-merge target. The parameter is optional.
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000435 *
Mikulas Patocka9d3b15c2009-12-10 23:52:32 +0000436 * Possible return values and states of snap_src and snap_dest.
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000437 * 0: NULL, NULL - first new snapshot
438 * 1: snap_src, NULL - normal snapshot
439 * 2: snap_src, snap_dest - waiting for handover
440 * 2: snap_src, NULL - handed over, waiting for old to be deleted
441 * 1: NULL, snap_dest - source got destroyed without handover
442 */
443static int __find_snapshots_sharing_cow(struct dm_snapshot *snap,
444 struct dm_snapshot **snap_src,
Mikulas Patocka9d3b15c2009-12-10 23:52:32 +0000445 struct dm_snapshot **snap_dest,
446 struct dm_snapshot **snap_merge)
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000447{
448 struct dm_snapshot *s;
449 struct origin *o;
450 int count = 0;
451 int active;
452
453 o = __lookup_origin(snap->origin->bdev);
454 if (!o)
455 goto out;
456
457 list_for_each_entry(s, &o->snapshots, list) {
Mikulas Patocka9d3b15c2009-12-10 23:52:32 +0000458 if (dm_target_is_snapshot_merge(s->ti) && snap_merge)
459 *snap_merge = s;
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000460 if (!bdev_equal(s->cow->bdev, snap->cow->bdev))
461 continue;
462
Nikos Tsironis4ad8d882019-03-17 14:22:56 +0200463 down_read(&s->lock);
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000464 active = s->active;
Nikos Tsironis4ad8d882019-03-17 14:22:56 +0200465 up_read(&s->lock);
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000466
467 if (active) {
468 if (snap_src)
469 *snap_src = s;
470 } else if (snap_dest)
471 *snap_dest = s;
472
473 count++;
474 }
475
476out:
477 return count;
478}
479
480/*
481 * On success, returns 1 if this snapshot is a handover destination,
482 * otherwise returns 0.
483 */
484static int __validate_exception_handover(struct dm_snapshot *snap)
485{
486 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
Mikulas Patocka9d3b15c2009-12-10 23:52:32 +0000487 struct dm_snapshot *snap_merge = NULL;
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000488
489 /* Does snapshot need exceptions handed over to it? */
Mikulas Patocka9d3b15c2009-12-10 23:52:32 +0000490 if ((__find_snapshots_sharing_cow(snap, &snap_src, &snap_dest,
491 &snap_merge) == 2) ||
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000492 snap_dest) {
493 snap->ti->error = "Snapshot cow pairing for exception "
494 "table handover failed";
495 return -EINVAL;
496 }
497
498 /*
499 * If no snap_src was found, snap cannot become a handover
500 * destination.
501 */
502 if (!snap_src)
503 return 0;
504
Mikulas Patocka9d3b15c2009-12-10 23:52:32 +0000505 /*
506 * Non-snapshot-merge handover?
507 */
508 if (!dm_target_is_snapshot_merge(snap->ti))
509 return 1;
510
511 /*
512 * Do not allow more than one merging snapshot.
513 */
514 if (snap_merge) {
515 snap->ti->error = "A snapshot is already merging.";
516 return -EINVAL;
517 }
518
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000519 if (!snap_src->store->type->prepare_merge ||
520 !snap_src->store->type->commit_merge) {
521 snap->ti->error = "Snapshot exception store does not "
522 "support snapshot-merge.";
523 return -EINVAL;
524 }
525
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000526 return 1;
527}
528
529static void __insert_snapshot(struct origin *o, struct dm_snapshot *s)
530{
531 struct dm_snapshot *l;
532
533 /* Sort the list according to chunk size, largest-first smallest-last */
534 list_for_each_entry(l, &o->snapshots, list)
535 if (l->store->chunk_size < s->store->chunk_size)
536 break;
537 list_add_tail(&s->list, &l->list);
538}
539
540/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541 * Make a note of the snapshot and its origin so we can look it
542 * up when the origin has a write on it.
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000543 *
544 * Also validate snapshot exception store handovers.
545 * On success, returns 1 if this registration is a handover destination,
546 * otherwise returns 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547 */
548static int register_snapshot(struct dm_snapshot *snap)
549{
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000550 struct origin *o, *new_o = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551 struct block_device *bdev = snap->origin->bdev;
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000552 int r = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553
Mikulas Patocka60c856c82008-10-30 13:33:12 +0000554 new_o = kmalloc(sizeof(*new_o), GFP_KERNEL);
555 if (!new_o)
556 return -ENOMEM;
557
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558 down_write(&_origins_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000560 r = __validate_exception_handover(snap);
561 if (r < 0) {
562 kfree(new_o);
563 goto out;
564 }
565
566 o = __lookup_origin(bdev);
Mikulas Patocka60c856c82008-10-30 13:33:12 +0000567 if (o)
568 kfree(new_o);
569 else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 /* New origin */
Mikulas Patocka60c856c82008-10-30 13:33:12 +0000571 o = new_o;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572
573 /* Initialise the struct */
574 INIT_LIST_HEAD(&o->snapshots);
575 o->bdev = bdev;
576
577 __insert_origin(o);
578 }
579
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000580 __insert_snapshot(o, snap);
581
582out:
583 up_write(&_origins_lock);
584
585 return r;
586}
587
588/*
589 * Move snapshot to correct place in list according to chunk size.
590 */
591static void reregister_snapshot(struct dm_snapshot *s)
592{
593 struct block_device *bdev = s->origin->bdev;
594
595 down_write(&_origins_lock);
596
597 list_del(&s->list);
598 __insert_snapshot(__lookup_origin(bdev), s);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599
600 up_write(&_origins_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601}
602
603static void unregister_snapshot(struct dm_snapshot *s)
604{
605 struct origin *o;
606
607 down_write(&_origins_lock);
608 o = __lookup_origin(s->origin->bdev);
609
610 list_del(&s->list);
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000611 if (o && list_empty(&o->snapshots)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 list_del(&o->hash_list);
613 kfree(o);
614 }
615
616 up_write(&_origins_lock);
617}
618
619/*
620 * Implementation of the exception hash tables.
Milan Brozd74f81f2008-02-08 02:11:27 +0000621 * The lowest hash_shift bits of the chunk number are ignored, allowing
622 * some consecutive chunks to be grouped together.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 */
Nikos Tsironisf79ae412019-03-17 14:22:57 +0200624static uint32_t exception_hash(struct dm_exception_table *et, chunk_t chunk);
625
626/* Lock to protect access to the completed and pending exception hash tables. */
627struct dm_exception_table_lock {
628 struct hlist_bl_head *complete_slot;
629 struct hlist_bl_head *pending_slot;
630};
631
632static void dm_exception_table_lock_init(struct dm_snapshot *s, chunk_t chunk,
633 struct dm_exception_table_lock *lock)
634{
635 struct dm_exception_table *complete = &s->complete;
636 struct dm_exception_table *pending = &s->pending;
637
638 lock->complete_slot = &complete->table[exception_hash(complete, chunk)];
639 lock->pending_slot = &pending->table[exception_hash(pending, chunk)];
640}
641
642static void dm_exception_table_lock(struct dm_exception_table_lock *lock)
643{
644 hlist_bl_lock(lock->complete_slot);
645 hlist_bl_lock(lock->pending_slot);
646}
647
648static void dm_exception_table_unlock(struct dm_exception_table_lock *lock)
649{
650 hlist_bl_unlock(lock->pending_slot);
651 hlist_bl_unlock(lock->complete_slot);
652}
653
Jon Brassow3510cb92009-12-10 23:52:11 +0000654static int dm_exception_table_init(struct dm_exception_table *et,
655 uint32_t size, unsigned hash_shift)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656{
657 unsigned int i;
658
Milan Brozd74f81f2008-02-08 02:11:27 +0000659 et->hash_shift = hash_shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660 et->hash_mask = size - 1;
Nikos Tsironisf79ae412019-03-17 14:22:57 +0200661 et->table = dm_vcalloc(size, sizeof(struct hlist_bl_head));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662 if (!et->table)
663 return -ENOMEM;
664
665 for (i = 0; i < size; i++)
Nikos Tsironisf79ae412019-03-17 14:22:57 +0200666 INIT_HLIST_BL_HEAD(et->table + i);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667
668 return 0;
669}
670
Jon Brassow3510cb92009-12-10 23:52:11 +0000671static void dm_exception_table_exit(struct dm_exception_table *et,
672 struct kmem_cache *mem)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673{
Nikos Tsironisf79ae412019-03-17 14:22:57 +0200674 struct hlist_bl_head *slot;
675 struct dm_exception *ex;
676 struct hlist_bl_node *pos, *n;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677 int i, size;
678
679 size = et->hash_mask + 1;
680 for (i = 0; i < size; i++) {
681 slot = et->table + i;
682
Nikos Tsironisf79ae412019-03-17 14:22:57 +0200683 hlist_bl_for_each_entry_safe(ex, pos, n, slot, hash_list)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684 kmem_cache_free(mem, ex);
685 }
686
687 vfree(et->table);
688}
689
Jon Brassow191437a2009-12-10 23:52:10 +0000690static uint32_t exception_hash(struct dm_exception_table *et, chunk_t chunk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691{
Milan Brozd74f81f2008-02-08 02:11:27 +0000692 return (chunk >> et->hash_shift) & et->hash_mask;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693}
694
Jon Brassow3510cb92009-12-10 23:52:11 +0000695static void dm_remove_exception(struct dm_exception *e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696{
Nikos Tsironisf79ae412019-03-17 14:22:57 +0200697 hlist_bl_del(&e->hash_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698}
699
700/*
701 * Return the exception data for a sector, or NULL if not
702 * remapped.
703 */
Jon Brassow3510cb92009-12-10 23:52:11 +0000704static struct dm_exception *dm_lookup_exception(struct dm_exception_table *et,
705 chunk_t chunk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706{
Nikos Tsironisf79ae412019-03-17 14:22:57 +0200707 struct hlist_bl_head *slot;
708 struct hlist_bl_node *pos;
Jon Brassow1d4989c2009-12-10 23:52:10 +0000709 struct dm_exception *e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710
711 slot = &et->table[exception_hash(et, chunk)];
Nikos Tsironisf79ae412019-03-17 14:22:57 +0200712 hlist_bl_for_each_entry(e, pos, slot, hash_list)
Milan Brozd74f81f2008-02-08 02:11:27 +0000713 if (chunk >= e->old_chunk &&
714 chunk <= e->old_chunk + dm_consecutive_chunk_count(e))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715 return e;
716
717 return NULL;
718}
719
Mikulas Patocka119bc542014-01-13 19:13:36 -0500720static struct dm_exception *alloc_completed_exception(gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721{
Jon Brassow1d4989c2009-12-10 23:52:10 +0000722 struct dm_exception *e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723
Mikulas Patocka119bc542014-01-13 19:13:36 -0500724 e = kmem_cache_alloc(exception_cache, gfp);
725 if (!e && gfp == GFP_NOIO)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 e = kmem_cache_alloc(exception_cache, GFP_ATOMIC);
727
728 return e;
729}
730
Jon Brassow3510cb92009-12-10 23:52:11 +0000731static void free_completed_exception(struct dm_exception *e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732{
733 kmem_cache_free(exception_cache, e);
734}
735
Mikulas Patocka92e86812008-07-21 12:00:35 +0100736static struct dm_snap_pending_exception *alloc_pending_exception(struct dm_snapshot *s)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737{
Kent Overstreet6f1c8192018-05-20 18:25:53 -0400738 struct dm_snap_pending_exception *pe = mempool_alloc(&s->pending_pool,
Mikulas Patocka92e86812008-07-21 12:00:35 +0100739 GFP_NOIO);
740
Mikulas Patocka879129d22008-10-30 13:33:16 +0000741 atomic_inc(&s->pending_exceptions_count);
Mikulas Patocka92e86812008-07-21 12:00:35 +0100742 pe->snap = s;
743
744 return pe;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745}
746
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100747static void free_pending_exception(struct dm_snap_pending_exception *pe)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748{
Mikulas Patocka879129d22008-10-30 13:33:16 +0000749 struct dm_snapshot *s = pe->snap;
750
Kent Overstreet6f1c8192018-05-20 18:25:53 -0400751 mempool_free(pe, &s->pending_pool);
Peter Zijlstra4e857c52014-03-17 18:06:10 +0100752 smp_mb__before_atomic();
Mikulas Patocka879129d22008-10-30 13:33:16 +0000753 atomic_dec(&s->pending_exceptions_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754}
755
Jon Brassow3510cb92009-12-10 23:52:11 +0000756static void dm_insert_exception(struct dm_exception_table *eh,
757 struct dm_exception *new_e)
Milan Brozd74f81f2008-02-08 02:11:27 +0000758{
Nikos Tsironisf79ae412019-03-17 14:22:57 +0200759 struct hlist_bl_head *l;
760 struct hlist_bl_node *pos;
Jon Brassow1d4989c2009-12-10 23:52:10 +0000761 struct dm_exception *e = NULL;
Milan Brozd74f81f2008-02-08 02:11:27 +0000762
763 l = &eh->table[exception_hash(eh, new_e->old_chunk)];
764
765 /* Add immediately if this table doesn't support consecutive chunks */
766 if (!eh->hash_shift)
767 goto out;
768
769 /* List is ordered by old_chunk */
Nikos Tsironisf79ae412019-03-17 14:22:57 +0200770 hlist_bl_for_each_entry(e, pos, l, hash_list) {
Milan Brozd74f81f2008-02-08 02:11:27 +0000771 /* Insert after an existing chunk? */
772 if (new_e->old_chunk == (e->old_chunk +
773 dm_consecutive_chunk_count(e) + 1) &&
774 new_e->new_chunk == (dm_chunk_number(e->new_chunk) +
775 dm_consecutive_chunk_count(e) + 1)) {
776 dm_consecutive_chunk_count_inc(e);
Jon Brassow3510cb92009-12-10 23:52:11 +0000777 free_completed_exception(new_e);
Milan Brozd74f81f2008-02-08 02:11:27 +0000778 return;
779 }
780
781 /* Insert before an existing chunk? */
782 if (new_e->old_chunk == (e->old_chunk - 1) &&
783 new_e->new_chunk == (dm_chunk_number(e->new_chunk) - 1)) {
784 dm_consecutive_chunk_count_inc(e);
785 e->old_chunk--;
786 e->new_chunk--;
Jon Brassow3510cb92009-12-10 23:52:11 +0000787 free_completed_exception(new_e);
Milan Brozd74f81f2008-02-08 02:11:27 +0000788 return;
789 }
790
Nikos Tsironisf79ae412019-03-17 14:22:57 +0200791 if (new_e->old_chunk < e->old_chunk)
Milan Brozd74f81f2008-02-08 02:11:27 +0000792 break;
793 }
794
795out:
Nikos Tsironisf79ae412019-03-17 14:22:57 +0200796 if (!e) {
797 /*
798 * Either the table doesn't support consecutive chunks or slot
799 * l is empty.
800 */
801 hlist_bl_add_head(&new_e->hash_list, l);
802 } else if (new_e->old_chunk < e->old_chunk) {
803 /* Add before an existing exception */
804 hlist_bl_add_before(&new_e->hash_list, &e->hash_list);
805 } else {
806 /* Add to l's tail: e is the last exception in this slot */
807 hlist_bl_add_behind(&new_e->hash_list, &e->hash_list);
808 }
Milan Brozd74f81f2008-02-08 02:11:27 +0000809}
810
Jonathan Brassowa159c1a2009-01-06 03:05:19 +0000811/*
812 * Callback used by the exception stores to load exceptions when
813 * initialising.
814 */
815static int dm_add_exception(void *context, chunk_t old, chunk_t new)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816{
Nikos Tsironisf79ae412019-03-17 14:22:57 +0200817 struct dm_exception_table_lock lock;
Jonathan Brassowa159c1a2009-01-06 03:05:19 +0000818 struct dm_snapshot *s = context;
Jon Brassow1d4989c2009-12-10 23:52:10 +0000819 struct dm_exception *e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820
Mikulas Patocka119bc542014-01-13 19:13:36 -0500821 e = alloc_completed_exception(GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822 if (!e)
823 return -ENOMEM;
824
825 e->old_chunk = old;
Milan Brozd74f81f2008-02-08 02:11:27 +0000826
827 /* Consecutive_count is implicitly initialised to zero */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828 e->new_chunk = new;
Milan Brozd74f81f2008-02-08 02:11:27 +0000829
Nikos Tsironisf79ae412019-03-17 14:22:57 +0200830 /*
831 * Although there is no need to lock access to the exception tables
832 * here, if we don't then hlist_bl_add_head(), called by
833 * dm_insert_exception(), will complain about accessing the
834 * corresponding list without locking it first.
835 */
836 dm_exception_table_lock_init(s, old, &lock);
837
838 dm_exception_table_lock(&lock);
Jon Brassow3510cb92009-12-10 23:52:11 +0000839 dm_insert_exception(&s->complete, e);
Nikos Tsironisf79ae412019-03-17 14:22:57 +0200840 dm_exception_table_unlock(&lock);
Milan Brozd74f81f2008-02-08 02:11:27 +0000841
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842 return 0;
843}
844
Mikulas Patocka7e201b32009-12-10 23:52:08 +0000845/*
846 * Return a minimum chunk size of all snapshots that have the specified origin.
847 * Return zero if the origin has no snapshots.
848 */
Mike Snitzer542f9032012-07-27 15:08:00 +0100849static uint32_t __minimum_chunk_size(struct origin *o)
Mikulas Patocka7e201b32009-12-10 23:52:08 +0000850{
851 struct dm_snapshot *snap;
852 unsigned chunk_size = 0;
853
854 if (o)
855 list_for_each_entry(snap, &o->snapshots, list)
856 chunk_size = min_not_zero(chunk_size,
857 snap->store->chunk_size);
858
Mike Snitzer542f9032012-07-27 15:08:00 +0100859 return (uint32_t) chunk_size;
Mikulas Patocka7e201b32009-12-10 23:52:08 +0000860}
861
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862/*
863 * Hard coded magic.
864 */
865static int calc_max_buckets(void)
866{
867 /* use a fixed size of 2MB */
868 unsigned long mem = 2 * 1024 * 1024;
Nikos Tsironisf79ae412019-03-17 14:22:57 +0200869 mem /= sizeof(struct hlist_bl_head);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870
871 return mem;
872}
873
874/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875 * Allocate room for a suitable hash table.
876 */
Jonathan Brassowfee19982009-04-02 19:55:34 +0100877static int init_hash_tables(struct dm_snapshot *s)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878{
Mikulas Patocka60e356f2013-09-18 19:40:42 -0400879 sector_t hash_size, cow_dev_size, max_buckets;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880
881 /*
882 * Calculate based on the size of the original volume or
883 * the COW volume...
884 */
Mike Snitzerfc56f6f2009-12-10 23:52:12 +0000885 cow_dev_size = get_dev_size(s->cow->bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886 max_buckets = calc_max_buckets();
887
Mikulas Patocka60e356f2013-09-18 19:40:42 -0400888 hash_size = cow_dev_size >> s->store->chunk_shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889 hash_size = min(hash_size, max_buckets);
890
Mikulas Patocka8e87b9b2009-12-10 23:51:54 +0000891 if (hash_size < 64)
892 hash_size = 64;
Robert P. J. Day8defd832008-02-08 02:10:06 +0000893 hash_size = rounddown_pow_of_two(hash_size);
Jon Brassow3510cb92009-12-10 23:52:11 +0000894 if (dm_exception_table_init(&s->complete, hash_size,
895 DM_CHUNK_CONSECUTIVE_BITS))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896 return -ENOMEM;
897
898 /*
899 * Allocate hash table for in-flight exceptions
900 * Make this smaller than the real hash table
901 */
902 hash_size >>= 3;
903 if (hash_size < 64)
904 hash_size = 64;
905
Jon Brassow3510cb92009-12-10 23:52:11 +0000906 if (dm_exception_table_init(&s->pending, hash_size, 0)) {
907 dm_exception_table_exit(&s->complete, exception_cache);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908 return -ENOMEM;
909 }
910
911 return 0;
912}
913
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000914static void merge_shutdown(struct dm_snapshot *s)
915{
916 clear_bit_unlock(RUNNING_MERGE, &s->state_bits);
Peter Zijlstra4e857c52014-03-17 18:06:10 +0100917 smp_mb__after_atomic();
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000918 wake_up_bit(&s->state_bits, RUNNING_MERGE);
919}
920
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000921static struct bio *__release_queued_bios_after_merge(struct dm_snapshot *s)
922{
923 s->first_merging_chunk = 0;
924 s->num_merging_chunks = 0;
925
926 return bio_list_get(&s->bios_queued_during_merge);
927}
928
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000929/*
930 * Remove one chunk from the index of completed exceptions.
931 */
932static int __remove_single_exception_chunk(struct dm_snapshot *s,
933 chunk_t old_chunk)
934{
935 struct dm_exception *e;
936
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000937 e = dm_lookup_exception(&s->complete, old_chunk);
938 if (!e) {
939 DMERR("Corruption detected: exception for block %llu is "
940 "on disk but not in memory",
941 (unsigned long long)old_chunk);
942 return -EINVAL;
943 }
944
945 /*
946 * If this is the only chunk using this exception, remove exception.
947 */
948 if (!dm_consecutive_chunk_count(e)) {
949 dm_remove_exception(e);
950 free_completed_exception(e);
951 return 0;
952 }
953
954 /*
955 * The chunk may be either at the beginning or the end of a
956 * group of consecutive chunks - never in the middle. We are
957 * removing chunks in the opposite order to that in which they
958 * were added, so this should always be true.
959 * Decrement the consecutive chunk counter and adjust the
960 * starting point if necessary.
961 */
962 if (old_chunk == e->old_chunk) {
963 e->old_chunk++;
964 e->new_chunk++;
965 } else if (old_chunk != e->old_chunk +
966 dm_consecutive_chunk_count(e)) {
967 DMERR("Attempt to merge block %llu from the "
968 "middle of a chunk range [%llu - %llu]",
969 (unsigned long long)old_chunk,
970 (unsigned long long)e->old_chunk,
971 (unsigned long long)
972 e->old_chunk + dm_consecutive_chunk_count(e));
973 return -EINVAL;
974 }
975
976 dm_consecutive_chunk_count_dec(e);
977
978 return 0;
979}
980
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000981static void flush_bios(struct bio *bio);
982
983static int remove_single_exception_chunk(struct dm_snapshot *s)
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000984{
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000985 struct bio *b = NULL;
986 int r;
987 chunk_t old_chunk = s->first_merging_chunk + s->num_merging_chunks - 1;
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000988
Nikos Tsironis4ad8d882019-03-17 14:22:56 +0200989 down_write(&s->lock);
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000990
991 /*
992 * Process chunks (and associated exceptions) in reverse order
993 * so that dm_consecutive_chunk_count_dec() accounting works.
994 */
995 do {
996 r = __remove_single_exception_chunk(s, old_chunk);
997 if (r)
998 goto out;
999 } while (old_chunk-- > s->first_merging_chunk);
1000
1001 b = __release_queued_bios_after_merge(s);
1002
1003out:
Nikos Tsironis4ad8d882019-03-17 14:22:56 +02001004 up_write(&s->lock);
Mikulas Patocka9fe862542009-12-10 23:52:33 +00001005 if (b)
1006 flush_bios(b);
Mikulas Patocka1e03f972009-12-10 23:52:32 +00001007
1008 return r;
1009}
1010
Mikulas Patocka73dfd072009-12-10 23:52:34 +00001011static int origin_write_extent(struct dm_snapshot *merging_snap,
1012 sector_t sector, unsigned chunk_size);
1013
Mikulas Patocka1e03f972009-12-10 23:52:32 +00001014static void merge_callback(int read_err, unsigned long write_err,
1015 void *context);
1016
Mikulas Patocka73dfd072009-12-10 23:52:34 +00001017static uint64_t read_pending_exceptions_done_count(void)
1018{
1019 uint64_t pending_exceptions_done;
1020
1021 spin_lock(&_pending_exceptions_done_spinlock);
1022 pending_exceptions_done = _pending_exceptions_done_count;
1023 spin_unlock(&_pending_exceptions_done_spinlock);
1024
1025 return pending_exceptions_done;
1026}
1027
1028static void increment_pending_exceptions_done_count(void)
1029{
1030 spin_lock(&_pending_exceptions_done_spinlock);
1031 _pending_exceptions_done_count++;
1032 spin_unlock(&_pending_exceptions_done_spinlock);
1033
1034 wake_up_all(&_pending_exceptions_done);
1035}
1036
Mikulas Patocka1e03f972009-12-10 23:52:32 +00001037static void snapshot_merge_next_chunks(struct dm_snapshot *s)
1038{
Mike Snitzer8a2d5282009-12-10 23:52:34 +00001039 int i, linear_chunks;
Mikulas Patocka1e03f972009-12-10 23:52:32 +00001040 chunk_t old_chunk, new_chunk;
1041 struct dm_io_region src, dest;
Mike Snitzer8a2d5282009-12-10 23:52:34 +00001042 sector_t io_size;
Mikulas Patocka73dfd072009-12-10 23:52:34 +00001043 uint64_t previous_count;
Mikulas Patocka1e03f972009-12-10 23:52:32 +00001044
1045 BUG_ON(!test_bit(RUNNING_MERGE, &s->state_bits));
1046 if (unlikely(test_bit(SHUTDOWN_MERGE, &s->state_bits)))
1047 goto shut;
1048
1049 /*
1050 * valid flag never changes during merge, so no lock required.
1051 */
1052 if (!s->valid) {
1053 DMERR("Snapshot is invalid: can't merge");
1054 goto shut;
1055 }
1056
Mike Snitzer8a2d5282009-12-10 23:52:34 +00001057 linear_chunks = s->store->type->prepare_merge(s->store, &old_chunk,
1058 &new_chunk);
1059 if (linear_chunks <= 0) {
Mike Snitzerd8ddb1c2009-12-10 23:52:35 +00001060 if (linear_chunks < 0) {
Mikulas Patocka1e03f972009-12-10 23:52:32 +00001061 DMERR("Read error in exception store: "
1062 "shutting down merge");
Nikos Tsironis4ad8d882019-03-17 14:22:56 +02001063 down_write(&s->lock);
zhengbin1d1dda82019-12-24 14:38:02 +08001064 s->merge_failed = true;
Nikos Tsironis4ad8d882019-03-17 14:22:56 +02001065 up_write(&s->lock);
Mike Snitzerd8ddb1c2009-12-10 23:52:35 +00001066 }
Mikulas Patocka1e03f972009-12-10 23:52:32 +00001067 goto shut;
1068 }
1069
Mike Snitzer8a2d5282009-12-10 23:52:34 +00001070 /* Adjust old_chunk and new_chunk to reflect start of linear region */
1071 old_chunk = old_chunk + 1 - linear_chunks;
1072 new_chunk = new_chunk + 1 - linear_chunks;
1073
1074 /*
1075 * Use one (potentially large) I/O to copy all 'linear_chunks'
1076 * from the exception store to the origin
1077 */
1078 io_size = linear_chunks * s->store->chunk_size;
Mikulas Patocka1e03f972009-12-10 23:52:32 +00001079
Mikulas Patocka1e03f972009-12-10 23:52:32 +00001080 dest.bdev = s->origin->bdev;
1081 dest.sector = chunk_to_sector(s->store, old_chunk);
Mike Snitzer8a2d5282009-12-10 23:52:34 +00001082 dest.count = min(io_size, get_dev_size(dest.bdev) - dest.sector);
Mikulas Patocka1e03f972009-12-10 23:52:32 +00001083
1084 src.bdev = s->cow->bdev;
1085 src.sector = chunk_to_sector(s->store, new_chunk);
1086 src.count = dest.count;
1087
Mikulas Patocka73dfd072009-12-10 23:52:34 +00001088 /*
1089 * Reallocate any exceptions needed in other snapshots then
1090 * wait for the pending exceptions to complete.
1091 * Each time any pending exception (globally on the system)
1092 * completes we are woken and repeat the process to find out
1093 * if we can proceed. While this may not seem a particularly
1094 * efficient algorithm, it is not expected to have any
1095 * significant impact on performance.
1096 */
1097 previous_count = read_pending_exceptions_done_count();
Mike Snitzer8a2d5282009-12-10 23:52:34 +00001098 while (origin_write_extent(s, dest.sector, io_size)) {
Mikulas Patocka73dfd072009-12-10 23:52:34 +00001099 wait_event(_pending_exceptions_done,
1100 (read_pending_exceptions_done_count() !=
1101 previous_count));
1102 /* Retry after the wait, until all exceptions are done. */
1103 previous_count = read_pending_exceptions_done_count();
1104 }
1105
Nikos Tsironis4ad8d882019-03-17 14:22:56 +02001106 down_write(&s->lock);
Mikulas Patocka9fe862542009-12-10 23:52:33 +00001107 s->first_merging_chunk = old_chunk;
Mike Snitzer8a2d5282009-12-10 23:52:34 +00001108 s->num_merging_chunks = linear_chunks;
Nikos Tsironis4ad8d882019-03-17 14:22:56 +02001109 up_write(&s->lock);
Mikulas Patocka9fe862542009-12-10 23:52:33 +00001110
Mike Snitzer8a2d5282009-12-10 23:52:34 +00001111 /* Wait until writes to all 'linear_chunks' drain */
1112 for (i = 0; i < linear_chunks; i++)
1113 __check_for_conflicting_io(s, old_chunk + i);
Mikulas Patocka9fe862542009-12-10 23:52:33 +00001114
Mikulas Patocka1e03f972009-12-10 23:52:32 +00001115 dm_kcopyd_copy(s->kcopyd_client, &src, 1, &dest, 0, merge_callback, s);
1116 return;
1117
1118shut:
1119 merge_shutdown(s);
1120}
1121
Mikulas Patocka9fe862542009-12-10 23:52:33 +00001122static void error_bios(struct bio *bio);
1123
Mikulas Patocka1e03f972009-12-10 23:52:32 +00001124static void merge_callback(int read_err, unsigned long write_err, void *context)
1125{
1126 struct dm_snapshot *s = context;
Mikulas Patocka9fe862542009-12-10 23:52:33 +00001127 struct bio *b = NULL;
Mikulas Patocka1e03f972009-12-10 23:52:32 +00001128
1129 if (read_err || write_err) {
1130 if (read_err)
1131 DMERR("Read error: shutting down merge.");
1132 else
1133 DMERR("Write error: shutting down merge.");
1134 goto shut;
1135 }
1136
Mikulas Patocka9fe862542009-12-10 23:52:33 +00001137 if (s->store->type->commit_merge(s->store,
1138 s->num_merging_chunks) < 0) {
Mikulas Patocka1e03f972009-12-10 23:52:32 +00001139 DMERR("Write error in exception store: shutting down merge");
1140 goto shut;
1141 }
1142
Mikulas Patocka9fe862542009-12-10 23:52:33 +00001143 if (remove_single_exception_chunk(s) < 0)
1144 goto shut;
1145
Mikulas Patocka1e03f972009-12-10 23:52:32 +00001146 snapshot_merge_next_chunks(s);
1147
1148 return;
1149
1150shut:
Nikos Tsironis4ad8d882019-03-17 14:22:56 +02001151 down_write(&s->lock);
zhengbin1d1dda82019-12-24 14:38:02 +08001152 s->merge_failed = true;
Mikulas Patocka9fe862542009-12-10 23:52:33 +00001153 b = __release_queued_bios_after_merge(s);
Nikos Tsironis4ad8d882019-03-17 14:22:56 +02001154 up_write(&s->lock);
Mikulas Patocka9fe862542009-12-10 23:52:33 +00001155 error_bios(b);
1156
Mikulas Patocka1e03f972009-12-10 23:52:32 +00001157 merge_shutdown(s);
1158}
1159
1160static void start_merge(struct dm_snapshot *s)
1161{
1162 if (!test_and_set_bit(RUNNING_MERGE, &s->state_bits))
1163 snapshot_merge_next_chunks(s);
1164}
1165
Mikulas Patocka1e03f972009-12-10 23:52:32 +00001166/*
1167 * Stop the merging process and wait until it finishes.
1168 */
1169static void stop_merge(struct dm_snapshot *s)
1170{
1171 set_bit(SHUTDOWN_MERGE, &s->state_bits);
NeilBrown74316202014-07-07 15:16:04 +10001172 wait_on_bit(&s->state_bits, RUNNING_MERGE, TASK_UNINTERRUPTIBLE);
Mikulas Patocka1e03f972009-12-10 23:52:32 +00001173 clear_bit(SHUTDOWN_MERGE, &s->state_bits);
1174}
1175
Mike Snitzer2e602382019-06-19 17:05:54 -04001176static int parse_snapshot_features(struct dm_arg_set *as, struct dm_snapshot *s,
1177 struct dm_target *ti)
1178{
1179 int r;
1180 unsigned argc;
1181 const char *arg_name;
1182
1183 static const struct dm_arg _args[] = {
1184 {0, 2, "Invalid number of feature arguments"},
1185 };
1186
1187 /*
1188 * No feature arguments supplied.
1189 */
1190 if (!as->argc)
1191 return 0;
1192
1193 r = dm_read_arg_group(_args, as, &argc, &ti->error);
1194 if (r)
1195 return -EINVAL;
1196
1197 while (argc && !r) {
1198 arg_name = dm_shift_arg(as);
1199 argc--;
1200
1201 if (!strcasecmp(arg_name, "discard_zeroes_cow"))
1202 s->discard_zeroes_cow = true;
1203
1204 else if (!strcasecmp(arg_name, "discard_passdown_origin"))
1205 s->discard_passdown_origin = true;
1206
1207 else {
1208 ti->error = "Unrecognised feature requested";
1209 r = -EINVAL;
1210 break;
1211 }
1212 }
1213
1214 if (!s->discard_zeroes_cow && s->discard_passdown_origin) {
1215 /*
1216 * TODO: really these are disjoint.. but ti->num_discard_bios
1217 * and dm_bio_get_target_bio_nr() require rigid constraints.
1218 */
1219 ti->error = "discard_passdown_origin feature depends on discard_zeroes_cow";
1220 r = -EINVAL;
1221 }
1222
1223 return r;
1224}
1225
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226/*
Mike Snitzer2e602382019-06-19 17:05:54 -04001227 * Construct a snapshot mapping:
1228 * <origin_dev> <COW-dev> <p|po|n> <chunk-size> [<# feature args> [<arg>]*]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229 */
1230static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv)
1231{
1232 struct dm_snapshot *s;
Mike Snitzer2e602382019-06-19 17:05:54 -04001233 struct dm_arg_set as;
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001234 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235 int r = -EINVAL;
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001236 char *origin_path, *cow_path;
DingXiang4df2bf42016-02-02 12:29:18 +08001237 dev_t origin_dev, cow_dev;
Alasdair G Kergon55a62ee2013-03-01 22:45:47 +00001238 unsigned args_used, num_flush_bios = 1;
Mike Snitzer10b81062009-12-10 23:52:31 +00001239 fmode_t origin_mode = FMODE_READ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240
Mike Snitzer2e602382019-06-19 17:05:54 -04001241 if (argc < 4) {
1242 ti->error = "requires 4 or more arguments";
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243 r = -EINVAL;
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001244 goto bad;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245 }
1246
Mike Snitzer10b81062009-12-10 23:52:31 +00001247 if (dm_target_is_snapshot_merge(ti)) {
Alasdair G Kergon55a62ee2013-03-01 22:45:47 +00001248 num_flush_bios = 2;
Mike Snitzer10b81062009-12-10 23:52:31 +00001249 origin_mode = FMODE_WRITE;
1250 }
1251
Kent Overstreetd3775352018-06-05 05:26:33 -04001252 s = kzalloc(sizeof(*s), GFP_KERNEL);
Jonathan Brassowfee19982009-04-02 19:55:34 +01001253 if (!s) {
Jonathan Brassowa2d2b032011-08-02 12:32:03 +01001254 ti->error = "Cannot allocate private snapshot structure";
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255 r = -ENOMEM;
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001256 goto bad;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257 }
1258
Mike Snitzer2e602382019-06-19 17:05:54 -04001259 as.argc = argc;
1260 as.argv = argv;
1261 dm_consume_args(&as, 4);
1262 r = parse_snapshot_features(&as, s, ti);
1263 if (r)
1264 goto bad_features;
1265
Mikulas Patockac2411042010-08-12 04:13:51 +01001266 origin_path = argv[0];
1267 argv++;
1268 argc--;
1269
1270 r = dm_get_device(ti, origin_path, origin_mode, &s->origin);
1271 if (r) {
1272 ti->error = "Cannot get origin device";
1273 goto bad_origin;
1274 }
DingXiang4df2bf42016-02-02 12:29:18 +08001275 origin_dev = s->origin->bdev->bd_dev;
Mikulas Patockac2411042010-08-12 04:13:51 +01001276
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001277 cow_path = argv[0];
1278 argv++;
1279 argc--;
1280
DingXiang4df2bf42016-02-02 12:29:18 +08001281 cow_dev = dm_get_dev_t(cow_path);
1282 if (cow_dev && cow_dev == origin_dev) {
1283 ti->error = "COW device cannot be the same as origin device";
1284 r = -EINVAL;
1285 goto bad_cow;
1286 }
1287
Milan Broz024d37e2011-03-24 13:52:14 +00001288 r = dm_get_device(ti, cow_path, dm_table_get_mode(ti->table), &s->cow);
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001289 if (r) {
1290 ti->error = "Cannot get COW device";
1291 goto bad_cow;
1292 }
1293
1294 r = dm_exception_store_create(ti, argc, argv, s, &args_used, &s->store);
1295 if (r) {
1296 ti->error = "Couldn't create exception store";
1297 r = -EINVAL;
1298 goto bad_store;
1299 }
1300
1301 argv += args_used;
1302 argc -= args_used;
1303
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001304 s->ti = ti;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305 s->valid = 1;
Mikulas Patocka76c44f62015-06-21 16:31:33 -04001306 s->snapshot_overflowed = 0;
Alasdair G Kergonaa14ede2006-02-01 03:04:50 -08001307 s->active = 0;
Mikulas Patocka879129d22008-10-30 13:33:16 +00001308 atomic_set(&s->pending_exceptions_count, 0);
Nikos Tsironis3f1637f2019-03-17 14:22:58 +02001309 spin_lock_init(&s->pe_allocation_lock);
Mikulas Patocka230c83a2013-11-29 18:13:37 -05001310 s->exception_start_sequence = 0;
1311 s->exception_complete_sequence = 0;
David Jeffery3db27762018-08-07 16:56:00 -04001312 s->out_of_order_tree = RB_ROOT;
Nikos Tsironis4ad8d882019-03-17 14:22:56 +02001313 init_rwsem(&s->lock);
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001314 INIT_LIST_HEAD(&s->list);
Alasdair G Kergonca3a9312006-10-03 01:15:30 -07001315 spin_lock_init(&s->pe_lock);
Mikulas Patocka1e03f972009-12-10 23:52:32 +00001316 s->state_bits = 0;
zhengbin1d1dda82019-12-24 14:38:02 +08001317 s->merge_failed = false;
Mikulas Patocka9fe862542009-12-10 23:52:33 +00001318 s->first_merging_chunk = 0;
1319 s->num_merging_chunks = 0;
1320 bio_list_init(&s->bios_queued_during_merge);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321
1322 /* Allocate hash table for COW data */
Jonathan Brassowfee19982009-04-02 19:55:34 +01001323 if (init_hash_tables(s)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324 ti->error = "Unable to allocate hash table space";
1325 r = -ENOMEM;
Jonathan Brassowfee19982009-04-02 19:55:34 +01001326 goto bad_hash_tables;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327 }
1328
Mikulas Patockab2155572019-10-02 06:15:53 -04001329 init_waitqueue_head(&s->in_progress_wait);
Nikos Tsironis721b1d92018-10-31 17:53:08 -04001330
Mikulas Patockadf5d2e92013-03-01 22:45:49 +00001331 s->kcopyd_client = dm_kcopyd_client_create(&dm_kcopyd_throttle);
Mikulas Patockafa34ce72011-05-29 13:03:13 +01001332 if (IS_ERR(s->kcopyd_client)) {
1333 r = PTR_ERR(s->kcopyd_client);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334 ti->error = "Could not create kcopyd client";
Jonathan Brassowfee19982009-04-02 19:55:34 +01001335 goto bad_kcopyd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336 }
1337
Kent Overstreet6f1c8192018-05-20 18:25:53 -04001338 r = mempool_init_slab_pool(&s->pending_pool, MIN_IOS, pending_cache);
1339 if (r) {
Mikulas Patocka92e86812008-07-21 12:00:35 +01001340 ti->error = "Could not allocate mempool for pending exceptions";
Jonathan Brassowfee19982009-04-02 19:55:34 +01001341 goto bad_pending_pool;
Mikulas Patocka92e86812008-07-21 12:00:35 +01001342 }
1343
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001344 for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++)
1345 INIT_HLIST_HEAD(&s->tracked_chunk_hash[i]);
1346
1347 spin_lock_init(&s->tracked_chunk_lock);
1348
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001349 ti->private = s;
Alasdair G Kergon55a62ee2013-03-01 22:45:47 +00001350 ti->num_flush_bios = num_flush_bios;
Mike Snitzer2e602382019-06-19 17:05:54 -04001351 if (s->discard_zeroes_cow)
1352 ti->num_discard_bios = (s->discard_passdown_origin ? 2 : 1);
Mike Snitzer30187e12016-01-31 13:28:26 -05001353 ti->per_io_data_size = sizeof(struct dm_snap_tracked_chunk);
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001354
1355 /* Add snapshot to the list of snapshots for this origin */
1356 /* Exceptions aren't triggered till snapshot_resume() is called */
1357 r = register_snapshot(s);
1358 if (r == -ENOMEM) {
1359 ti->error = "Snapshot origin struct allocation failed";
1360 goto bad_load_and_register;
1361 } else if (r < 0) {
1362 /* invalid handover, register_snapshot has set ti->error */
1363 goto bad_load_and_register;
1364 }
1365
1366 /*
1367 * Metadata must only be loaded into one table at once, so skip this
1368 * if metadata will be handed over during resume.
1369 * Chunk size will be set during the handover - set it to zero to
1370 * ensure it's ignored.
1371 */
1372 if (r > 0) {
1373 s->store->chunk_size = 0;
1374 return 0;
1375 }
1376
Jonathan Brassow493df712009-04-02 19:55:31 +01001377 r = s->store->type->read_metadata(s->store, dm_add_exception,
1378 (void *)s);
Milan Broz07641472007-07-12 17:28:13 +01001379 if (r < 0) {
Mark McLoughlinf9cea4f2006-10-03 01:15:25 -07001380 ti->error = "Failed to read snapshot metadata";
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001381 goto bad_read_metadata;
Milan Broz07641472007-07-12 17:28:13 +01001382 } else if (r > 0) {
1383 s->valid = 0;
1384 DMWARN("Snapshot is marked invalid.");
Mark McLoughlinf9cea4f2006-10-03 01:15:25 -07001385 }
Alasdair G Kergonaa14ede2006-02-01 03:04:50 -08001386
Mikulas Patocka3f2412d2009-10-16 23:18:16 +01001387 if (!s->store->chunk_size) {
1388 ti->error = "Chunk size not set";
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001389 goto bad_read_metadata;
Mikulas Patocka3f2412d2009-10-16 23:18:16 +01001390 }
Mike Snitzer542f9032012-07-27 15:08:00 +01001391
1392 r = dm_set_target_max_io_len(ti, s->store->chunk_size);
1393 if (r)
1394 goto bad_read_metadata;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395
1396 return 0;
1397
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001398bad_read_metadata:
1399 unregister_snapshot(s);
Jonathan Brassowfee19982009-04-02 19:55:34 +01001400bad_load_and_register:
Kent Overstreet6f1c8192018-05-20 18:25:53 -04001401 mempool_exit(&s->pending_pool);
Jonathan Brassowfee19982009-04-02 19:55:34 +01001402bad_pending_pool:
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +01001403 dm_kcopyd_client_destroy(s->kcopyd_client);
Jonathan Brassowfee19982009-04-02 19:55:34 +01001404bad_kcopyd:
Jon Brassow3510cb92009-12-10 23:52:11 +00001405 dm_exception_table_exit(&s->pending, pending_cache);
1406 dm_exception_table_exit(&s->complete, exception_cache);
Jonathan Brassowfee19982009-04-02 19:55:34 +01001407bad_hash_tables:
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001408 dm_exception_store_destroy(s->store);
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001409bad_store:
1410 dm_put_device(ti, s->cow);
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001411bad_cow:
Mikulas Patockac2411042010-08-12 04:13:51 +01001412 dm_put_device(ti, s->origin);
Mikulas Patockac2411042010-08-12 04:13:51 +01001413bad_origin:
Mike Snitzer2e602382019-06-19 17:05:54 -04001414bad_features:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415 kfree(s);
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001416bad:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417 return r;
1418}
1419
Milan Broz31c93a0c2006-12-08 02:41:11 -08001420static void __free_exceptions(struct dm_snapshot *s)
1421{
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +01001422 dm_kcopyd_client_destroy(s->kcopyd_client);
Milan Broz31c93a0c2006-12-08 02:41:11 -08001423 s->kcopyd_client = NULL;
1424
Jon Brassow3510cb92009-12-10 23:52:11 +00001425 dm_exception_table_exit(&s->pending, pending_cache);
1426 dm_exception_table_exit(&s->complete, exception_cache);
Milan Broz31c93a0c2006-12-08 02:41:11 -08001427}
1428
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001429static void __handover_exceptions(struct dm_snapshot *snap_src,
1430 struct dm_snapshot *snap_dest)
1431{
1432 union {
1433 struct dm_exception_table table_swap;
1434 struct dm_exception_store *store_swap;
1435 } u;
1436
1437 /*
1438 * Swap all snapshot context information between the two instances.
1439 */
1440 u.table_swap = snap_dest->complete;
1441 snap_dest->complete = snap_src->complete;
1442 snap_src->complete = u.table_swap;
1443
1444 u.store_swap = snap_dest->store;
1445 snap_dest->store = snap_src->store;
Mike Snitzerb0d3cc02015-10-08 18:05:41 -04001446 snap_dest->store->userspace_supports_overflow = u.store_swap->userspace_supports_overflow;
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001447 snap_src->store = u.store_swap;
1448
1449 snap_dest->store->snap = snap_dest;
1450 snap_src->store->snap = snap_src;
1451
Mike Snitzer542f9032012-07-27 15:08:00 +01001452 snap_dest->ti->max_io_len = snap_dest->store->chunk_size;
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001453 snap_dest->valid = snap_src->valid;
Mikulas Patocka76c44f62015-06-21 16:31:33 -04001454 snap_dest->snapshot_overflowed = snap_src->snapshot_overflowed;
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001455
1456 /*
1457 * Set source invalid to ensure it receives no further I/O.
1458 */
1459 snap_src->valid = 0;
1460}
1461
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462static void snapshot_dtr(struct dm_target *ti)
1463{
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001464#ifdef CONFIG_DM_DEBUG
1465 int i;
1466#endif
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001467 struct dm_snapshot *s = ti->private;
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001468 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001470 down_read(&_origins_lock);
1471 /* Check whether exception handover must be cancelled */
Mikulas Patocka9d3b15c2009-12-10 23:52:32 +00001472 (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL);
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001473 if (snap_src && snap_dest && (s == snap_src)) {
Nikos Tsironis4ad8d882019-03-17 14:22:56 +02001474 down_write(&snap_dest->lock);
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001475 snap_dest->valid = 0;
Nikos Tsironis4ad8d882019-03-17 14:22:56 +02001476 up_write(&snap_dest->lock);
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001477 DMERR("Cancelling snapshot handover.");
1478 }
1479 up_read(&_origins_lock);
1480
Mikulas Patocka1e03f972009-12-10 23:52:32 +00001481 if (dm_target_is_snapshot_merge(ti))
1482 stop_merge(s);
1483
Alasdair G Kergon138728dc2006-03-27 01:17:50 -08001484 /* Prevent further origin writes from using this snapshot. */
1485 /* After this returns there can be no new kcopyd jobs. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 unregister_snapshot(s);
1487
Mikulas Patocka879129d22008-10-30 13:33:16 +00001488 while (atomic_read(&s->pending_exceptions_count))
Mikulas Patocka90fa1522009-01-06 03:04:54 +00001489 msleep(1);
Mikulas Patocka879129d22008-10-30 13:33:16 +00001490 /*
Kent Overstreet6f1c8192018-05-20 18:25:53 -04001491 * Ensure instructions in mempool_exit aren't reordered
Mikulas Patocka879129d22008-10-30 13:33:16 +00001492 * before atomic_read.
1493 */
1494 smp_mb();
1495
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001496#ifdef CONFIG_DM_DEBUG
1497 for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++)
1498 BUG_ON(!hlist_empty(&s->tracked_chunk_hash[i]));
1499#endif
1500
Milan Broz31c93a0c2006-12-08 02:41:11 -08001501 __free_exceptions(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502
Kent Overstreet6f1c8192018-05-20 18:25:53 -04001503 mempool_exit(&s->pending_pool);
Mikulas Patocka92e86812008-07-21 12:00:35 +01001504
Jonathan Brassowfee19982009-04-02 19:55:34 +01001505 dm_exception_store_destroy(s->store);
Alasdair G Kergon138728dc2006-03-27 01:17:50 -08001506
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001507 dm_put_device(ti, s->cow);
1508
Mikulas Patockac2411042010-08-12 04:13:51 +01001509 dm_put_device(ti, s->origin);
1510
Mikulas Patockab2155572019-10-02 06:15:53 -04001511 WARN_ON(s->in_progress);
1512
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513 kfree(s);
1514}
1515
Mikulas Patockaa2f83e82019-10-02 06:14:17 -04001516static void account_start_copy(struct dm_snapshot *s)
1517{
Mikulas Patockab2155572019-10-02 06:15:53 -04001518 spin_lock(&s->in_progress_wait.lock);
1519 s->in_progress++;
1520 spin_unlock(&s->in_progress_wait.lock);
Mikulas Patockaa2f83e82019-10-02 06:14:17 -04001521}
1522
1523static void account_end_copy(struct dm_snapshot *s)
1524{
Mikulas Patockab2155572019-10-02 06:15:53 -04001525 spin_lock(&s->in_progress_wait.lock);
1526 BUG_ON(!s->in_progress);
1527 s->in_progress--;
1528 if (likely(s->in_progress <= cow_threshold) &&
1529 unlikely(waitqueue_active(&s->in_progress_wait)))
1530 wake_up_locked(&s->in_progress_wait);
1531 spin_unlock(&s->in_progress_wait.lock);
1532}
1533
1534static bool wait_for_in_progress(struct dm_snapshot *s, bool unlock_origins)
1535{
1536 if (unlikely(s->in_progress > cow_threshold)) {
1537 spin_lock(&s->in_progress_wait.lock);
1538 if (likely(s->in_progress > cow_threshold)) {
1539 /*
1540 * NOTE: this throttle doesn't account for whether
1541 * the caller is servicing an IO that will trigger a COW
1542 * so excess throttling may result for chunks not required
1543 * to be COW'd. But if cow_threshold was reached, extra
1544 * throttling is unlikely to negatively impact performance.
1545 */
1546 DECLARE_WAITQUEUE(wait, current);
1547 __add_wait_queue(&s->in_progress_wait, &wait);
1548 __set_current_state(TASK_UNINTERRUPTIBLE);
1549 spin_unlock(&s->in_progress_wait.lock);
1550 if (unlock_origins)
1551 up_read(&_origins_lock);
1552 io_schedule();
1553 remove_wait_queue(&s->in_progress_wait, &wait);
1554 return false;
1555 }
1556 spin_unlock(&s->in_progress_wait.lock);
1557 }
1558 return true;
Mikulas Patockaa2f83e82019-10-02 06:14:17 -04001559}
1560
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561/*
1562 * Flush a list of buffers.
1563 */
1564static void flush_bios(struct bio *bio)
1565{
1566 struct bio *n;
1567
1568 while (bio) {
1569 n = bio->bi_next;
1570 bio->bi_next = NULL;
1571 generic_make_request(bio);
1572 bio = n;
1573 }
1574}
1575
Mikulas Patockab2155572019-10-02 06:15:53 -04001576static int do_origin(struct dm_dev *origin, struct bio *bio, bool limit);
Mikulas Patocka515ad662009-12-10 23:52:30 +00001577
1578/*
1579 * Flush a list of buffers.
1580 */
1581static void retry_origin_bios(struct dm_snapshot *s, struct bio *bio)
1582{
1583 struct bio *n;
1584 int r;
1585
1586 while (bio) {
1587 n = bio->bi_next;
1588 bio->bi_next = NULL;
Mikulas Patockab2155572019-10-02 06:15:53 -04001589 r = do_origin(s->origin, bio, false);
Mikulas Patocka515ad662009-12-10 23:52:30 +00001590 if (r == DM_MAPIO_REMAPPED)
1591 generic_make_request(bio);
1592 bio = n;
1593 }
1594}
1595
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596/*
1597 * Error a list of buffers.
1598 */
1599static void error_bios(struct bio *bio)
1600{
1601 struct bio *n;
1602
1603 while (bio) {
1604 n = bio->bi_next;
1605 bio->bi_next = NULL;
NeilBrown6712ecf2007-09-27 12:47:43 +02001606 bio_io_error(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607 bio = n;
1608 }
1609}
1610
Alasdair G Kergon695368a2006-10-03 01:15:31 -07001611static void __invalidate_snapshot(struct dm_snapshot *s, int err)
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001612{
1613 if (!s->valid)
1614 return;
1615
1616 if (err == -EIO)
1617 DMERR("Invalidating snapshot: Error reading/writing.");
1618 else if (err == -ENOMEM)
1619 DMERR("Invalidating snapshot: Unable to allocate exception.");
1620
Jonathan Brassow493df712009-04-02 19:55:31 +01001621 if (s->store->type->drop_snapshot)
1622 s->store->type->drop_snapshot(s->store);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001623
1624 s->valid = 0;
1625
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001626 dm_table_event(s->ti->table);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001627}
1628
Nikos Tsironis3f1637f2019-03-17 14:22:58 +02001629static void invalidate_snapshot(struct dm_snapshot *s, int err)
1630{
1631 down_write(&s->lock);
1632 __invalidate_snapshot(s, err);
1633 up_write(&s->lock);
1634}
1635
Mikulas Patocka385277b2016-01-08 19:07:55 -05001636static void pending_complete(void *context, int success)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637{
Mikulas Patocka385277b2016-01-08 19:07:55 -05001638 struct dm_snap_pending_exception *pe = context;
Jon Brassow1d4989c2009-12-10 23:52:10 +00001639 struct dm_exception *e;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640 struct dm_snapshot *s = pe->snap;
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001641 struct bio *origin_bios = NULL;
1642 struct bio *snapshot_bios = NULL;
Mikulas Patockaa6e50b42011-08-02 12:32:04 +01001643 struct bio *full_bio = NULL;
Nikos Tsironisf79ae412019-03-17 14:22:57 +02001644 struct dm_exception_table_lock lock;
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001645 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646
Nikos Tsironisf79ae412019-03-17 14:22:57 +02001647 dm_exception_table_lock_init(s, pe->e.old_chunk, &lock);
1648
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001649 if (!success) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650 /* Read/write error - snapshot is unusable */
Nikos Tsironis3f1637f2019-03-17 14:22:58 +02001651 invalidate_snapshot(s, -EIO);
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001652 error = 1;
Nikos Tsironisf79ae412019-03-17 14:22:57 +02001653
1654 dm_exception_table_lock(&lock);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001655 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656 }
1657
Mikulas Patocka119bc542014-01-13 19:13:36 -05001658 e = alloc_completed_exception(GFP_NOIO);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001659 if (!e) {
Nikos Tsironis3f1637f2019-03-17 14:22:58 +02001660 invalidate_snapshot(s, -ENOMEM);
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001661 error = 1;
Nikos Tsironisf79ae412019-03-17 14:22:57 +02001662
1663 dm_exception_table_lock(&lock);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001664 goto out;
1665 }
1666 *e = pe->e;
1667
Nikos Tsironis3f1637f2019-03-17 14:22:58 +02001668 down_read(&s->lock);
Nikos Tsironisf79ae412019-03-17 14:22:57 +02001669 dm_exception_table_lock(&lock);
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001670 if (!s->valid) {
Nikos Tsironis3f1637f2019-03-17 14:22:58 +02001671 up_read(&s->lock);
Jon Brassow3510cb92009-12-10 23:52:11 +00001672 free_completed_exception(e);
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001673 error = 1;
Nikos Tsironis3f1637f2019-03-17 14:22:58 +02001674
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001675 goto out;
1676 }
1677
Mikulas Patockaa8d41b52008-07-21 12:00:34 +01001678 /*
Nikos Tsironis65fc7c32019-03-17 14:22:55 +02001679 * Add a proper exception. After inserting the completed exception all
1680 * subsequent snapshot reads to this chunk will be redirected to the
1681 * COW device. This ensures that we do not starve. Moreover, as long
1682 * as the pending exception exists, neither origin writes nor snapshot
1683 * merging can overwrite the chunk in origin.
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001684 */
Jon Brassow3510cb92009-12-10 23:52:11 +00001685 dm_insert_exception(&s->complete, e);
Nikos Tsironis3f1637f2019-03-17 14:22:58 +02001686 up_read(&s->lock);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001687
Nikos Tsironis65fc7c32019-03-17 14:22:55 +02001688 /* Wait for conflicting reads to drain */
1689 if (__chunk_is_tracked(s, pe->e.old_chunk)) {
Nikos Tsironisf79ae412019-03-17 14:22:57 +02001690 dm_exception_table_unlock(&lock);
Nikos Tsironis65fc7c32019-03-17 14:22:55 +02001691 __check_for_conflicting_io(s, pe->e.old_chunk);
Nikos Tsironisf79ae412019-03-17 14:22:57 +02001692 dm_exception_table_lock(&lock);
Nikos Tsironis65fc7c32019-03-17 14:22:55 +02001693 }
1694
Jonathan Brassowa2d2b032011-08-02 12:32:03 +01001695out:
Nikos Tsironis65fc7c32019-03-17 14:22:55 +02001696 /* Remove the in-flight exception from the list */
Jon Brassow3510cb92009-12-10 23:52:11 +00001697 dm_remove_exception(&pe->e);
Nikos Tsironisf79ae412019-03-17 14:22:57 +02001698
1699 dm_exception_table_unlock(&lock);
1700
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001701 snapshot_bios = bio_list_get(&pe->snapshot_bios);
Mikulas Patocka515ad662009-12-10 23:52:30 +00001702 origin_bios = bio_list_get(&pe->origin_bios);
Mikulas Patockaa6e50b42011-08-02 12:32:04 +01001703 full_bio = pe->full_bio;
Mikulas Patockafe3265b2015-11-25 16:03:31 -05001704 if (full_bio)
Mikulas Patockaa6e50b42011-08-02 12:32:04 +01001705 full_bio->bi_end_io = pe->full_bio_end_io;
Mikulas Patocka73dfd072009-12-10 23:52:34 +00001706 increment_pending_exceptions_done_count();
1707
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001708 /* Submit any pending write bios */
Mikulas Patockaa6e50b42011-08-02 12:32:04 +01001709 if (error) {
1710 if (full_bio)
1711 bio_io_error(full_bio);
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001712 error_bios(snapshot_bios);
Mikulas Patockaa6e50b42011-08-02 12:32:04 +01001713 } else {
1714 if (full_bio)
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001715 bio_endio(full_bio);
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001716 flush_bios(snapshot_bios);
Mikulas Patockaa6e50b42011-08-02 12:32:04 +01001717 }
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001718
Mikulas Patocka515ad662009-12-10 23:52:30 +00001719 retry_origin_bios(s, origin_bios);
Mikulas Patocka22aa66a2015-02-17 14:34:00 -05001720
1721 free_pending_exception(pe);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001722}
1723
Mikulas Patocka230c83a2013-11-29 18:13:37 -05001724static void complete_exception(struct dm_snap_pending_exception *pe)
1725{
1726 struct dm_snapshot *s = pe->snap;
1727
Mikulas Patocka385277b2016-01-08 19:07:55 -05001728 /* Update the metadata if we are persistent */
1729 s->store->type->commit_exception(s->store, &pe->e, !pe->copy_error,
1730 pending_complete, pe);
Mikulas Patocka230c83a2013-11-29 18:13:37 -05001731}
1732
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733/*
1734 * Called when the copy I/O has finished. kcopyd actually runs
1735 * this code so don't block.
1736 */
Alasdair G Kergon4cdc1d12008-03-28 14:16:10 -07001737static void copy_callback(int read_err, unsigned long write_err, void *context)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001738{
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001739 struct dm_snap_pending_exception *pe = context;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001740 struct dm_snapshot *s = pe->snap;
1741
Mikulas Patocka230c83a2013-11-29 18:13:37 -05001742 pe->copy_error = read_err || write_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743
Mikulas Patocka230c83a2013-11-29 18:13:37 -05001744 if (pe->exception_sequence == s->exception_complete_sequence) {
David Jeffery3db27762018-08-07 16:56:00 -04001745 struct rb_node *next;
1746
Mikulas Patocka230c83a2013-11-29 18:13:37 -05001747 s->exception_complete_sequence++;
1748 complete_exception(pe);
1749
David Jeffery3db27762018-08-07 16:56:00 -04001750 next = rb_first(&s->out_of_order_tree);
1751 while (next) {
1752 pe = rb_entry(next, struct dm_snap_pending_exception,
1753 out_of_order_node);
Mikulas Patocka230c83a2013-11-29 18:13:37 -05001754 if (pe->exception_sequence != s->exception_complete_sequence)
1755 break;
David Jeffery3db27762018-08-07 16:56:00 -04001756 next = rb_next(next);
Mikulas Patocka230c83a2013-11-29 18:13:37 -05001757 s->exception_complete_sequence++;
David Jeffery3db27762018-08-07 16:56:00 -04001758 rb_erase(&pe->out_of_order_node, &s->out_of_order_tree);
Mikulas Patocka230c83a2013-11-29 18:13:37 -05001759 complete_exception(pe);
David Jeffery3db27762018-08-07 16:56:00 -04001760 cond_resched();
Mikulas Patocka230c83a2013-11-29 18:13:37 -05001761 }
1762 } else {
David Jeffery3db27762018-08-07 16:56:00 -04001763 struct rb_node *parent = NULL;
1764 struct rb_node **p = &s->out_of_order_tree.rb_node;
Mikulas Patocka230c83a2013-11-29 18:13:37 -05001765 struct dm_snap_pending_exception *pe2;
1766
David Jeffery3db27762018-08-07 16:56:00 -04001767 while (*p) {
1768 pe2 = rb_entry(*p, struct dm_snap_pending_exception, out_of_order_node);
1769 parent = *p;
1770
1771 BUG_ON(pe->exception_sequence == pe2->exception_sequence);
1772 if (pe->exception_sequence < pe2->exception_sequence)
1773 p = &((*p)->rb_left);
1774 else
1775 p = &((*p)->rb_right);
Mikulas Patocka230c83a2013-11-29 18:13:37 -05001776 }
David Jeffery3db27762018-08-07 16:56:00 -04001777
1778 rb_link_node(&pe->out_of_order_node, parent, p);
1779 rb_insert_color(&pe->out_of_order_node, &s->out_of_order_tree);
Mikulas Patocka230c83a2013-11-29 18:13:37 -05001780 }
Mikulas Patockaa2f83e82019-10-02 06:14:17 -04001781 account_end_copy(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001782}
1783
1784/*
1785 * Dispatches the copy operation to kcopyd.
1786 */
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001787static void start_copy(struct dm_snap_pending_exception *pe)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788{
1789 struct dm_snapshot *s = pe->snap;
Heinz Mauelshagen22a1ceb2008-04-24 21:43:17 +01001790 struct dm_io_region src, dest;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791 struct block_device *bdev = s->origin->bdev;
1792 sector_t dev_size;
1793
1794 dev_size = get_dev_size(bdev);
1795
1796 src.bdev = bdev;
Jonathan Brassow71fab002009-04-02 19:55:33 +01001797 src.sector = chunk_to_sector(s->store, pe->e.old_chunk);
Mikulas Patockadf96eee2009-10-16 23:18:17 +01001798 src.count = min((sector_t)s->store->chunk_size, dev_size - src.sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001800 dest.bdev = s->cow->bdev;
Jonathan Brassow71fab002009-04-02 19:55:33 +01001801 dest.sector = chunk_to_sector(s->store, pe->e.new_chunk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802 dest.count = src.count;
1803
1804 /* Hand over to kcopyd */
Mikulas Patockaa2f83e82019-10-02 06:14:17 -04001805 account_start_copy(s);
Jonathan Brassowa2d2b032011-08-02 12:32:03 +01001806 dm_kcopyd_copy(s->kcopyd_client, &src, 1, &dest, 0, copy_callback, pe);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001807}
1808
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001809static void full_bio_end_io(struct bio *bio)
Mikulas Patockaa6e50b42011-08-02 12:32:04 +01001810{
1811 void *callback_data = bio->bi_private;
1812
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +02001813 dm_kcopyd_do_callback(callback_data, 0, bio->bi_status ? 1 : 0);
Mikulas Patockaa6e50b42011-08-02 12:32:04 +01001814}
1815
1816static void start_full_bio(struct dm_snap_pending_exception *pe,
1817 struct bio *bio)
1818{
1819 struct dm_snapshot *s = pe->snap;
1820 void *callback_data;
1821
1822 pe->full_bio = bio;
1823 pe->full_bio_end_io = bio->bi_end_io;
Mikulas Patockaa6e50b42011-08-02 12:32:04 +01001824
Mikulas Patockaa2f83e82019-10-02 06:14:17 -04001825 account_start_copy(s);
Mikulas Patockaa6e50b42011-08-02 12:32:04 +01001826 callback_data = dm_kcopyd_prepare_callback(s->kcopyd_client,
1827 copy_callback, pe);
1828
1829 bio->bi_end_io = full_bio_end_io;
1830 bio->bi_private = callback_data;
1831
1832 generic_make_request(bio);
1833}
1834
Mikulas Patocka29138082009-04-02 19:55:25 +01001835static struct dm_snap_pending_exception *
1836__lookup_pending_exception(struct dm_snapshot *s, chunk_t chunk)
1837{
Jon Brassow3510cb92009-12-10 23:52:11 +00001838 struct dm_exception *e = dm_lookup_exception(&s->pending, chunk);
Mikulas Patocka29138082009-04-02 19:55:25 +01001839
1840 if (!e)
1841 return NULL;
1842
1843 return container_of(e, struct dm_snap_pending_exception, e);
1844}
1845
Linus Torvalds1da177e2005-04-16 15:20:36 -07001846/*
Nikos Tsironis65fc7c32019-03-17 14:22:55 +02001847 * Inserts a pending exception into the pending table.
1848 *
Nikos Tsironis3f1637f2019-03-17 14:22:58 +02001849 * NOTE: a write lock must be held on the chunk's pending exception table slot
1850 * before calling this.
Nikos Tsironis65fc7c32019-03-17 14:22:55 +02001851 */
1852static struct dm_snap_pending_exception *
1853__insert_pending_exception(struct dm_snapshot *s,
1854 struct dm_snap_pending_exception *pe, chunk_t chunk)
1855{
1856 pe->e.old_chunk = chunk;
1857 bio_list_init(&pe->origin_bios);
1858 bio_list_init(&pe->snapshot_bios);
1859 pe->started = 0;
1860 pe->full_bio = NULL;
1861
Nikos Tsironis3f1637f2019-03-17 14:22:58 +02001862 spin_lock(&s->pe_allocation_lock);
Nikos Tsironis65fc7c32019-03-17 14:22:55 +02001863 if (s->store->type->prepare_exception(s->store, &pe->e)) {
Nikos Tsironis3f1637f2019-03-17 14:22:58 +02001864 spin_unlock(&s->pe_allocation_lock);
Nikos Tsironis65fc7c32019-03-17 14:22:55 +02001865 free_pending_exception(pe);
1866 return NULL;
1867 }
1868
1869 pe->exception_sequence = s->exception_start_sequence++;
Nikos Tsironis3f1637f2019-03-17 14:22:58 +02001870 spin_unlock(&s->pe_allocation_lock);
Nikos Tsironis65fc7c32019-03-17 14:22:55 +02001871
1872 dm_insert_exception(&s->pending, &pe->e);
1873
1874 return pe;
1875}
1876
1877/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878 * Looks to see if this snapshot already has a pending exception
1879 * for this chunk, otherwise it allocates a new one and inserts
1880 * it into the pending table.
1881 *
Nikos Tsironis3f1637f2019-03-17 14:22:58 +02001882 * NOTE: a write lock must be held on the chunk's pending exception table slot
1883 * before calling this.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001884 */
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001885static struct dm_snap_pending_exception *
Mikulas Patockac6621392009-04-02 19:55:25 +01001886__find_pending_exception(struct dm_snapshot *s,
1887 struct dm_snap_pending_exception *pe, chunk_t chunk)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001888{
Mikulas Patockac6621392009-04-02 19:55:25 +01001889 struct dm_snap_pending_exception *pe2;
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001890
Mikulas Patocka29138082009-04-02 19:55:25 +01001891 pe2 = __lookup_pending_exception(s, chunk);
1892 if (pe2) {
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001893 free_pending_exception(pe);
Mikulas Patocka29138082009-04-02 19:55:25 +01001894 return pe2;
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001895 }
1896
Nikos Tsironis65fc7c32019-03-17 14:22:55 +02001897 return __insert_pending_exception(s, pe, chunk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001898}
1899
Jon Brassow1d4989c2009-12-10 23:52:10 +00001900static void remap_exception(struct dm_snapshot *s, struct dm_exception *e,
Milan Brozd74f81f2008-02-08 02:11:27 +00001901 struct bio *bio, chunk_t chunk)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001902{
Christoph Hellwig74d46992017-08-23 19:10:32 +02001903 bio_set_dev(bio, s->cow->bdev);
Kent Overstreet4f024f32013-10-11 15:44:27 -07001904 bio->bi_iter.bi_sector =
1905 chunk_to_sector(s->store, dm_chunk_number(e->new_chunk) +
1906 (chunk - e->old_chunk)) +
1907 (bio->bi_iter.bi_sector & s->store->chunk_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001908}
1909
Mike Snitzer2e602382019-06-19 17:05:54 -04001910static void zero_callback(int read_err, unsigned long write_err, void *context)
1911{
1912 struct bio *bio = context;
1913 struct dm_snapshot *s = bio->bi_private;
1914
Mikulas Patockaa2f83e82019-10-02 06:14:17 -04001915 account_end_copy(s);
Mike Snitzer2e602382019-06-19 17:05:54 -04001916 bio->bi_status = write_err ? BLK_STS_IOERR : 0;
1917 bio_endio(bio);
1918}
1919
1920static void zero_exception(struct dm_snapshot *s, struct dm_exception *e,
1921 struct bio *bio, chunk_t chunk)
1922{
1923 struct dm_io_region dest;
1924
1925 dest.bdev = s->cow->bdev;
1926 dest.sector = bio->bi_iter.bi_sector;
1927 dest.count = s->store->chunk_size;
1928
Mikulas Patockaa2f83e82019-10-02 06:14:17 -04001929 account_start_copy(s);
Mike Snitzer2e602382019-06-19 17:05:54 -04001930 WARN_ON_ONCE(bio->bi_private);
1931 bio->bi_private = s;
1932 dm_kcopyd_zero(s->kcopyd_client, 1, &dest, 0, zero_callback, bio);
1933}
1934
1935static bool io_overlaps_chunk(struct dm_snapshot *s, struct bio *bio)
1936{
1937 return bio->bi_iter.bi_size ==
1938 (s->store->chunk_size << SECTOR_SHIFT);
1939}
1940
Mikulas Patocka7de3ee52012-12-21 20:23:41 +00001941static int snapshot_map(struct dm_target *ti, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001942{
Jon Brassow1d4989c2009-12-10 23:52:10 +00001943 struct dm_exception *e;
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001944 struct dm_snapshot *s = ti->private;
Kiyoshi Uedad2a7ad22006-12-08 02:41:06 -08001945 int r = DM_MAPIO_REMAPPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946 chunk_t chunk;
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001947 struct dm_snap_pending_exception *pe = NULL;
Nikos Tsironisf79ae412019-03-17 14:22:57 +02001948 struct dm_exception_table_lock lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949
Mikulas Patockaee180262012-12-21 20:23:41 +00001950 init_tracked_chunk(bio);
1951
Jens Axboe1eff9d32016-08-05 15:35:16 -06001952 if (bio->bi_opf & REQ_PREFLUSH) {
Christoph Hellwig74d46992017-08-23 19:10:32 +02001953 bio_set_dev(bio, s->cow->bdev);
Mikulas Patocka494b3ee2009-06-22 10:12:25 +01001954 return DM_MAPIO_REMAPPED;
1955 }
1956
Kent Overstreet4f024f32013-10-11 15:44:27 -07001957 chunk = sector_to_chunk(s->store, bio->bi_iter.bi_sector);
Nikos Tsironisf79ae412019-03-17 14:22:57 +02001958 dm_exception_table_lock_init(s, chunk, &lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001959
1960 /* Full snapshots are not usable */
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001961 /* To get here the table must be live so s->active is always set. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962 if (!s->valid)
Christoph Hellwig846785e2017-06-03 09:38:02 +02001963 return DM_MAPIO_KILL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001964
Mikulas Patockab2155572019-10-02 06:15:53 -04001965 if (bio_data_dir(bio) == WRITE) {
1966 while (unlikely(!wait_for_in_progress(s, false)))
1967 ; /* wait_for_in_progress() has slept */
1968 }
1969
Nikos Tsironis3f1637f2019-03-17 14:22:58 +02001970 down_read(&s->lock);
Nikos Tsironisf79ae412019-03-17 14:22:57 +02001971 dm_exception_table_lock(&lock);
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07001972
Christoph Hellwig70246282016-07-19 11:28:41 +02001973 if (!s->valid || (unlikely(s->snapshot_overflowed) &&
1974 bio_data_dir(bio) == WRITE)) {
Christoph Hellwig846785e2017-06-03 09:38:02 +02001975 r = DM_MAPIO_KILL;
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07001976 goto out_unlock;
1977 }
1978
Mike Snitzer2e602382019-06-19 17:05:54 -04001979 if (unlikely(bio_op(bio) == REQ_OP_DISCARD)) {
1980 if (s->discard_passdown_origin && dm_bio_get_target_bio_nr(bio)) {
1981 /*
1982 * passdown discard to origin (without triggering
1983 * snapshot exceptions via do_origin; doing so would
1984 * defeat the goal of freeing space in origin that is
1985 * implied by the "discard_passdown_origin" feature)
1986 */
1987 bio_set_dev(bio, s->origin->bdev);
1988 track_chunk(s, bio, chunk);
1989 goto out_unlock;
1990 }
1991 /* discard to snapshot (target_bio_nr == 0) zeroes exceptions */
1992 }
1993
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07001994 /* If the block is already remapped - use that, else remap it */
Jon Brassow3510cb92009-12-10 23:52:11 +00001995 e = dm_lookup_exception(&s->complete, chunk);
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07001996 if (e) {
Milan Brozd74f81f2008-02-08 02:11:27 +00001997 remap_exception(s, e, bio, chunk);
Mike Snitzer2e602382019-06-19 17:05:54 -04001998 if (unlikely(bio_op(bio) == REQ_OP_DISCARD) &&
1999 io_overlaps_chunk(s, bio)) {
2000 dm_exception_table_unlock(&lock);
2001 up_read(&s->lock);
2002 zero_exception(s, e, bio, chunk);
2003 r = DM_MAPIO_SUBMITTED; /* discard is not issued */
2004 goto out;
2005 }
2006 goto out_unlock;
2007 }
2008
2009 if (unlikely(bio_op(bio) == REQ_OP_DISCARD)) {
2010 /*
2011 * If no exception exists, complete discard immediately
2012 * otherwise it'll trigger copy-out.
2013 */
2014 bio_endio(bio);
2015 r = DM_MAPIO_SUBMITTED;
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07002016 goto out_unlock;
2017 }
2018
Linus Torvalds1da177e2005-04-16 15:20:36 -07002019 /*
2020 * Write to snapshot - higher level takes care of RW/RO
2021 * flags so we should only get this if we are
2022 * writeable.
2023 */
Christoph Hellwig70246282016-07-19 11:28:41 +02002024 if (bio_data_dir(bio) == WRITE) {
Mikulas Patocka29138082009-04-02 19:55:25 +01002025 pe = __lookup_pending_exception(s, chunk);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08002026 if (!pe) {
Nikos Tsironisf79ae412019-03-17 14:22:57 +02002027 dm_exception_table_unlock(&lock);
Mikulas Patockac6621392009-04-02 19:55:25 +01002028 pe = alloc_pending_exception(s);
Nikos Tsironisf79ae412019-03-17 14:22:57 +02002029 dm_exception_table_lock(&lock);
Mikulas Patockac6621392009-04-02 19:55:25 +01002030
Jon Brassow3510cb92009-12-10 23:52:11 +00002031 e = dm_lookup_exception(&s->complete, chunk);
Mikulas Patocka35bf6592009-04-02 19:55:26 +01002032 if (e) {
2033 free_pending_exception(pe);
2034 remap_exception(s, e, bio, chunk);
2035 goto out_unlock;
2036 }
2037
Mikulas Patockac6621392009-04-02 19:55:25 +01002038 pe = __find_pending_exception(s, pe, chunk);
Mikulas Patocka29138082009-04-02 19:55:25 +01002039 if (!pe) {
Nikos Tsironisf79ae412019-03-17 14:22:57 +02002040 dm_exception_table_unlock(&lock);
Nikos Tsironis3f1637f2019-03-17 14:22:58 +02002041 up_read(&s->lock);
2042
2043 down_write(&s->lock);
Nikos Tsironisf79ae412019-03-17 14:22:57 +02002044
Mike Snitzerb0d3cc02015-10-08 18:05:41 -04002045 if (s->store->userspace_supports_overflow) {
Nikos Tsironis3f1637f2019-03-17 14:22:58 +02002046 if (s->valid && !s->snapshot_overflowed) {
2047 s->snapshot_overflowed = 1;
2048 DMERR("Snapshot overflowed: Unable to allocate exception.");
2049 }
Mike Snitzerb0d3cc02015-10-08 18:05:41 -04002050 } else
2051 __invalidate_snapshot(s, -ENOMEM);
Nikos Tsironisf79ae412019-03-17 14:22:57 +02002052 up_write(&s->lock);
2053
Christoph Hellwig846785e2017-06-03 09:38:02 +02002054 r = DM_MAPIO_KILL;
Nikos Tsironisf79ae412019-03-17 14:22:57 +02002055 goto out;
Mikulas Patocka29138082009-04-02 19:55:25 +01002056 }
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08002057 }
2058
Milan Brozd74f81f2008-02-08 02:11:27 +00002059 remap_exception(s, &pe->e, bio, chunk);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08002060
Kiyoshi Uedad2a7ad22006-12-08 02:41:06 -08002061 r = DM_MAPIO_SUBMITTED;
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07002062
Mike Snitzer2e602382019-06-19 17:05:54 -04002063 if (!pe->started && io_overlaps_chunk(s, bio)) {
Mikulas Patockaa6e50b42011-08-02 12:32:04 +01002064 pe->started = 1;
Nikos Tsironis3f1637f2019-03-17 14:22:58 +02002065
Nikos Tsironisf79ae412019-03-17 14:22:57 +02002066 dm_exception_table_unlock(&lock);
Nikos Tsironis3f1637f2019-03-17 14:22:58 +02002067 up_read(&s->lock);
2068
Mikulas Patockaa6e50b42011-08-02 12:32:04 +01002069 start_full_bio(pe, bio);
2070 goto out;
2071 }
2072
2073 bio_list_add(&pe->snapshot_bios, bio);
2074
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08002075 if (!pe->started) {
Nikos Tsironis3f1637f2019-03-17 14:22:58 +02002076 /* this is protected by the exception table lock */
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08002077 pe->started = 1;
Nikos Tsironis3f1637f2019-03-17 14:22:58 +02002078
Nikos Tsironisf79ae412019-03-17 14:22:57 +02002079 dm_exception_table_unlock(&lock);
Nikos Tsironis3f1637f2019-03-17 14:22:58 +02002080 up_read(&s->lock);
2081
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08002082 start_copy(pe);
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07002083 goto out;
2084 }
Mikulas Patockacd45daf2008-07-21 12:00:32 +01002085 } else {
Christoph Hellwig74d46992017-08-23 19:10:32 +02002086 bio_set_dev(bio, s->origin->bdev);
Mikulas Patockaee180262012-12-21 20:23:41 +00002087 track_chunk(s, bio, chunk);
Mikulas Patockacd45daf2008-07-21 12:00:32 +01002088 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002089
Jonathan Brassowa2d2b032011-08-02 12:32:03 +01002090out_unlock:
Nikos Tsironisf79ae412019-03-17 14:22:57 +02002091 dm_exception_table_unlock(&lock);
Nikos Tsironis3f1637f2019-03-17 14:22:58 +02002092 up_read(&s->lock);
Jonathan Brassowa2d2b032011-08-02 12:32:03 +01002093out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002094 return r;
2095}
2096
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00002097/*
2098 * A snapshot-merge target behaves like a combination of a snapshot
2099 * target and a snapshot-origin target. It only generates new
2100 * exceptions in other snapshots and not in the one that is being
2101 * merged.
2102 *
2103 * For each chunk, if there is an existing exception, it is used to
2104 * redirect I/O to the cow device. Otherwise I/O is sent to the origin,
2105 * which in turn might generate exceptions in other snapshots.
Mikulas Patocka9fe862542009-12-10 23:52:33 +00002106 * If merging is currently taking place on the chunk in question, the
2107 * I/O is deferred by adding it to s->bios_queued_during_merge.
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00002108 */
Mikulas Patocka7de3ee52012-12-21 20:23:41 +00002109static int snapshot_merge_map(struct dm_target *ti, struct bio *bio)
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00002110{
2111 struct dm_exception *e;
2112 struct dm_snapshot *s = ti->private;
2113 int r = DM_MAPIO_REMAPPED;
2114 chunk_t chunk;
2115
Mikulas Patockaee180262012-12-21 20:23:41 +00002116 init_tracked_chunk(bio);
2117
Jens Axboe1eff9d32016-08-05 15:35:16 -06002118 if (bio->bi_opf & REQ_PREFLUSH) {
Alasdair G Kergon55a62ee2013-03-01 22:45:47 +00002119 if (!dm_bio_get_target_bio_nr(bio))
Christoph Hellwig74d46992017-08-23 19:10:32 +02002120 bio_set_dev(bio, s->origin->bdev);
Mike Snitzer10b81062009-12-10 23:52:31 +00002121 else
Christoph Hellwig74d46992017-08-23 19:10:32 +02002122 bio_set_dev(bio, s->cow->bdev);
Mike Snitzer10b81062009-12-10 23:52:31 +00002123 return DM_MAPIO_REMAPPED;
2124 }
2125
Mike Snitzer3ee25482019-07-17 11:12:30 -04002126 if (unlikely(bio_op(bio) == REQ_OP_DISCARD)) {
2127 /* Once merging, discards no longer effect change */
2128 bio_endio(bio);
2129 return DM_MAPIO_SUBMITTED;
2130 }
2131
Kent Overstreet4f024f32013-10-11 15:44:27 -07002132 chunk = sector_to_chunk(s->store, bio->bi_iter.bi_sector);
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00002133
Nikos Tsironis4ad8d882019-03-17 14:22:56 +02002134 down_write(&s->lock);
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00002135
Mikulas Patockad2fdb772009-12-10 23:52:36 +00002136 /* Full merging snapshots are redirected to the origin */
2137 if (!s->valid)
2138 goto redirect_to_origin;
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00002139
2140 /* If the block is already remapped - use that */
2141 e = dm_lookup_exception(&s->complete, chunk);
2142 if (e) {
Mikulas Patocka9fe862542009-12-10 23:52:33 +00002143 /* Queue writes overlapping with chunks being merged */
Christoph Hellwig70246282016-07-19 11:28:41 +02002144 if (bio_data_dir(bio) == WRITE &&
Mikulas Patocka9fe862542009-12-10 23:52:33 +00002145 chunk >= s->first_merging_chunk &&
2146 chunk < (s->first_merging_chunk +
2147 s->num_merging_chunks)) {
Christoph Hellwig74d46992017-08-23 19:10:32 +02002148 bio_set_dev(bio, s->origin->bdev);
Mikulas Patocka9fe862542009-12-10 23:52:33 +00002149 bio_list_add(&s->bios_queued_during_merge, bio);
2150 r = DM_MAPIO_SUBMITTED;
2151 goto out_unlock;
2152 }
Mikulas Patocka17aa0332009-12-10 23:52:33 +00002153
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00002154 remap_exception(s, e, bio, chunk);
Mikulas Patocka17aa0332009-12-10 23:52:33 +00002155
Christoph Hellwig70246282016-07-19 11:28:41 +02002156 if (bio_data_dir(bio) == WRITE)
Mikulas Patockaee180262012-12-21 20:23:41 +00002157 track_chunk(s, bio, chunk);
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00002158 goto out_unlock;
2159 }
2160
Mikulas Patockad2fdb772009-12-10 23:52:36 +00002161redirect_to_origin:
Christoph Hellwig74d46992017-08-23 19:10:32 +02002162 bio_set_dev(bio, s->origin->bdev);
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00002163
Christoph Hellwig70246282016-07-19 11:28:41 +02002164 if (bio_data_dir(bio) == WRITE) {
Nikos Tsironis4ad8d882019-03-17 14:22:56 +02002165 up_write(&s->lock);
Mikulas Patockab2155572019-10-02 06:15:53 -04002166 return do_origin(s->origin, bio, false);
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00002167 }
2168
2169out_unlock:
Nikos Tsironis4ad8d882019-03-17 14:22:56 +02002170 up_write(&s->lock);
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00002171
2172 return r;
2173}
2174
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +02002175static int snapshot_end_io(struct dm_target *ti, struct bio *bio,
2176 blk_status_t *error)
Mikulas Patockacd45daf2008-07-21 12:00:32 +01002177{
2178 struct dm_snapshot *s = ti->private;
Mikulas Patockacd45daf2008-07-21 12:00:32 +01002179
Mikulas Patockaee180262012-12-21 20:23:41 +00002180 if (is_bio_tracked(bio))
2181 stop_tracking_chunk(s, bio);
Mikulas Patockacd45daf2008-07-21 12:00:32 +01002182
Christoph Hellwig1be56902017-06-03 09:38:03 +02002183 return DM_ENDIO_DONE;
Mikulas Patockacd45daf2008-07-21 12:00:32 +01002184}
2185
Mikulas Patocka1e03f972009-12-10 23:52:32 +00002186static void snapshot_merge_presuspend(struct dm_target *ti)
2187{
2188 struct dm_snapshot *s = ti->private;
2189
2190 stop_merge(s);
2191}
2192
Mike Snitzerc1f0c182009-12-10 23:52:24 +00002193static int snapshot_preresume(struct dm_target *ti)
2194{
2195 int r = 0;
2196 struct dm_snapshot *s = ti->private;
2197 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
2198
2199 down_read(&_origins_lock);
Mikulas Patocka9d3b15c2009-12-10 23:52:32 +00002200 (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL);
Mike Snitzerc1f0c182009-12-10 23:52:24 +00002201 if (snap_src && snap_dest) {
Nikos Tsironis4ad8d882019-03-17 14:22:56 +02002202 down_read(&snap_src->lock);
Mike Snitzerc1f0c182009-12-10 23:52:24 +00002203 if (s == snap_src) {
2204 DMERR("Unable to resume snapshot source until "
2205 "handover completes.");
2206 r = -EINVAL;
Mike Snitzerb83b2f22011-01-13 19:59:59 +00002207 } else if (!dm_suspended(snap_src->ti)) {
Mike Snitzerc1f0c182009-12-10 23:52:24 +00002208 DMERR("Unable to perform snapshot handover until "
2209 "source is suspended.");
2210 r = -EINVAL;
2211 }
Nikos Tsironis4ad8d882019-03-17 14:22:56 +02002212 up_read(&snap_src->lock);
Mike Snitzerc1f0c182009-12-10 23:52:24 +00002213 }
2214 up_read(&_origins_lock);
2215
2216 return r;
2217}
2218
Linus Torvalds1da177e2005-04-16 15:20:36 -07002219static void snapshot_resume(struct dm_target *ti)
2220{
Alasdair G Kergon028867a2007-07-12 17:26:32 +01002221 struct dm_snapshot *s = ti->private;
Mikulas Patocka09ee96b2015-02-26 11:41:28 -05002222 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL, *snap_merging = NULL;
Mikulas Patockab735fed2015-02-26 11:40:35 -05002223 struct dm_origin *o;
2224 struct mapped_device *origin_md = NULL;
Mikulas Patocka09ee96b2015-02-26 11:41:28 -05002225 bool must_restart_merging = false;
Mike Snitzerc1f0c182009-12-10 23:52:24 +00002226
2227 down_read(&_origins_lock);
Mikulas Patockab735fed2015-02-26 11:40:35 -05002228
2229 o = __lookup_dm_origin(s->origin->bdev);
2230 if (o)
2231 origin_md = dm_table_get_md(o->ti->table);
Mikulas Patocka09ee96b2015-02-26 11:41:28 -05002232 if (!origin_md) {
2233 (void) __find_snapshots_sharing_cow(s, NULL, NULL, &snap_merging);
2234 if (snap_merging)
2235 origin_md = dm_table_get_md(snap_merging->ti->table);
2236 }
Mikulas Patockab735fed2015-02-26 11:40:35 -05002237 if (origin_md == dm_table_get_md(ti->table))
2238 origin_md = NULL;
Mikulas Patocka09ee96b2015-02-26 11:41:28 -05002239 if (origin_md) {
2240 if (dm_hold(origin_md))
2241 origin_md = NULL;
2242 }
Mikulas Patockab735fed2015-02-26 11:40:35 -05002243
Mikulas Patocka09ee96b2015-02-26 11:41:28 -05002244 up_read(&_origins_lock);
2245
2246 if (origin_md) {
Mikulas Patockab735fed2015-02-26 11:40:35 -05002247 dm_internal_suspend_fast(origin_md);
Mikulas Patocka09ee96b2015-02-26 11:41:28 -05002248 if (snap_merging && test_bit(RUNNING_MERGE, &snap_merging->state_bits)) {
2249 must_restart_merging = true;
2250 stop_merge(snap_merging);
2251 }
2252 }
2253
2254 down_read(&_origins_lock);
Mikulas Patockab735fed2015-02-26 11:40:35 -05002255
Mikulas Patocka9d3b15c2009-12-10 23:52:32 +00002256 (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL);
Mike Snitzerc1f0c182009-12-10 23:52:24 +00002257 if (snap_src && snap_dest) {
Nikos Tsironis4ad8d882019-03-17 14:22:56 +02002258 down_write(&snap_src->lock);
2259 down_write_nested(&snap_dest->lock, SINGLE_DEPTH_NESTING);
Mike Snitzerc1f0c182009-12-10 23:52:24 +00002260 __handover_exceptions(snap_src, snap_dest);
Nikos Tsironis4ad8d882019-03-17 14:22:56 +02002261 up_write(&snap_dest->lock);
2262 up_write(&snap_src->lock);
Mike Snitzerc1f0c182009-12-10 23:52:24 +00002263 }
Mikulas Patockab735fed2015-02-26 11:40:35 -05002264
Mike Snitzerc1f0c182009-12-10 23:52:24 +00002265 up_read(&_origins_lock);
2266
Mikulas Patocka09ee96b2015-02-26 11:41:28 -05002267 if (origin_md) {
2268 if (must_restart_merging)
2269 start_merge(snap_merging);
2270 dm_internal_resume_fast(origin_md);
2271 dm_put(origin_md);
2272 }
2273
Mike Snitzerc1f0c182009-12-10 23:52:24 +00002274 /* Now we have correct chunk size, reregister */
2275 reregister_snapshot(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002276
Nikos Tsironis4ad8d882019-03-17 14:22:56 +02002277 down_write(&s->lock);
Alasdair G Kergonaa14ede2006-02-01 03:04:50 -08002278 s->active = 1;
Nikos Tsironis4ad8d882019-03-17 14:22:56 +02002279 up_write(&s->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002280}
2281
Mike Snitzer542f9032012-07-27 15:08:00 +01002282static uint32_t get_origin_minimum_chunksize(struct block_device *bdev)
Mikulas Patocka1e03f972009-12-10 23:52:32 +00002283{
Mike Snitzer542f9032012-07-27 15:08:00 +01002284 uint32_t min_chunksize;
Mikulas Patocka1e03f972009-12-10 23:52:32 +00002285
2286 down_read(&_origins_lock);
2287 min_chunksize = __minimum_chunk_size(__lookup_origin(bdev));
2288 up_read(&_origins_lock);
2289
2290 return min_chunksize;
2291}
2292
2293static void snapshot_merge_resume(struct dm_target *ti)
2294{
2295 struct dm_snapshot *s = ti->private;
2296
2297 /*
2298 * Handover exceptions from existing snapshot.
2299 */
2300 snapshot_resume(ti);
2301
2302 /*
Mike Snitzer542f9032012-07-27 15:08:00 +01002303 * snapshot-merge acts as an origin, so set ti->max_io_len
Mikulas Patocka1e03f972009-12-10 23:52:32 +00002304 */
Mike Snitzer542f9032012-07-27 15:08:00 +01002305 ti->max_io_len = get_origin_minimum_chunksize(s->origin->bdev);
Mikulas Patocka1e03f972009-12-10 23:52:32 +00002306
2307 start_merge(s);
2308}
2309
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002310static void snapshot_status(struct dm_target *ti, status_type_t type,
2311 unsigned status_flags, char *result, unsigned maxlen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002312{
Jonathan Brassow2e4a31d2009-04-02 19:55:34 +01002313 unsigned sz = 0;
Alasdair G Kergon028867a2007-07-12 17:26:32 +01002314 struct dm_snapshot *snap = ti->private;
Mike Snitzer2e602382019-06-19 17:05:54 -04002315 unsigned num_features;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002316
2317 switch (type) {
2318 case STATUSTYPE_INFO:
Mikulas Patocka94e765722009-12-10 23:51:53 +00002319
Nikos Tsironis4ad8d882019-03-17 14:22:56 +02002320 down_write(&snap->lock);
Mikulas Patocka94e765722009-12-10 23:51:53 +00002321
Linus Torvalds1da177e2005-04-16 15:20:36 -07002322 if (!snap->valid)
Jonathan Brassow2e4a31d2009-04-02 19:55:34 +01002323 DMEMIT("Invalid");
Mike Snitzerd8ddb1c2009-12-10 23:52:35 +00002324 else if (snap->merge_failed)
2325 DMEMIT("Merge failed");
Mikulas Patocka76c44f62015-06-21 16:31:33 -04002326 else if (snap->snapshot_overflowed)
2327 DMEMIT("Overflow");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002328 else {
Mike Snitzer985903b2009-12-10 23:52:11 +00002329 if (snap->store->type->usage) {
2330 sector_t total_sectors, sectors_allocated,
2331 metadata_sectors;
2332 snap->store->type->usage(snap->store,
2333 &total_sectors,
2334 &sectors_allocated,
2335 &metadata_sectors);
2336 DMEMIT("%llu/%llu %llu",
2337 (unsigned long long)sectors_allocated,
2338 (unsigned long long)total_sectors,
2339 (unsigned long long)metadata_sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002340 }
2341 else
Jonathan Brassow2e4a31d2009-04-02 19:55:34 +01002342 DMEMIT("Unknown");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002343 }
Mikulas Patocka94e765722009-12-10 23:51:53 +00002344
Nikos Tsironis4ad8d882019-03-17 14:22:56 +02002345 up_write(&snap->lock);
Mikulas Patocka94e765722009-12-10 23:51:53 +00002346
Linus Torvalds1da177e2005-04-16 15:20:36 -07002347 break;
2348
2349 case STATUSTYPE_TABLE:
2350 /*
2351 * kdevname returns a static pointer so we need
2352 * to make private copies if the output is to
2353 * make sense.
2354 */
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00002355 DMEMIT("%s %s", snap->origin->name, snap->cow->name);
Mike Snitzer2e602382019-06-19 17:05:54 -04002356 sz += snap->store->type->status(snap->store, type, result + sz,
2357 maxlen - sz);
2358 num_features = snap->discard_zeroes_cow + snap->discard_passdown_origin;
2359 if (num_features) {
2360 DMEMIT(" %u", num_features);
2361 if (snap->discard_zeroes_cow)
2362 DMEMIT(" discard_zeroes_cow");
2363 if (snap->discard_passdown_origin)
2364 DMEMIT(" discard_passdown_origin");
2365 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002366 break;
2367 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002368}
2369
Mike Snitzer8811f462009-09-04 20:40:19 +01002370static int snapshot_iterate_devices(struct dm_target *ti,
2371 iterate_devices_callout_fn fn, void *data)
2372{
2373 struct dm_snapshot *snap = ti->private;
Mikulas Patocka1e5554c2010-08-12 04:13:50 +01002374 int r;
Mike Snitzer8811f462009-09-04 20:40:19 +01002375
Mikulas Patocka1e5554c2010-08-12 04:13:50 +01002376 r = fn(ti, snap->origin, 0, ti->len, data);
2377
2378 if (!r)
2379 r = fn(ti, snap->cow, 0, get_dev_size(snap->cow->bdev), data);
2380
2381 return r;
Mike Snitzer8811f462009-09-04 20:40:19 +01002382}
2383
Mike Snitzer2e602382019-06-19 17:05:54 -04002384static void snapshot_io_hints(struct dm_target *ti, struct queue_limits *limits)
2385{
2386 struct dm_snapshot *snap = ti->private;
2387
2388 if (snap->discard_zeroes_cow) {
2389 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
2390
Mike Snitzer3ee25482019-07-17 11:12:30 -04002391 down_read(&_origins_lock);
2392
Mike Snitzer2e602382019-06-19 17:05:54 -04002393 (void) __find_snapshots_sharing_cow(snap, &snap_src, &snap_dest, NULL);
2394 if (snap_src && snap_dest)
2395 snap = snap_src;
2396
2397 /* All discards are split on chunk_size boundary */
2398 limits->discard_granularity = snap->store->chunk_size;
2399 limits->max_discard_sectors = snap->store->chunk_size;
Mike Snitzer3ee25482019-07-17 11:12:30 -04002400
2401 up_read(&_origins_lock);
Mike Snitzer2e602382019-06-19 17:05:54 -04002402 }
2403}
Mike Snitzer8811f462009-09-04 20:40:19 +01002404
Linus Torvalds1da177e2005-04-16 15:20:36 -07002405/*-----------------------------------------------------------------
2406 * Origin methods
2407 *---------------------------------------------------------------*/
Mikulas Patocka9eaae8f2009-12-10 23:52:28 +00002408
2409/*
2410 * If no exceptions need creating, DM_MAPIO_REMAPPED is returned and any
2411 * supplied bio was ignored. The caller may submit it immediately.
2412 * (No remapping actually occurs as the origin is always a direct linear
2413 * map.)
2414 *
2415 * If further exceptions are required, DM_MAPIO_SUBMITTED is returned
2416 * and any supplied bio is added to a list to be submitted once all
2417 * the necessary exceptions exist.
2418 */
2419static int __origin_write(struct list_head *snapshots, sector_t sector,
2420 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002421{
Mikulas Patocka515ad662009-12-10 23:52:30 +00002422 int r = DM_MAPIO_REMAPPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002423 struct dm_snapshot *snap;
Jon Brassow1d4989c2009-12-10 23:52:10 +00002424 struct dm_exception *e;
Nikos Tsironis65fc7c32019-03-17 14:22:55 +02002425 struct dm_snap_pending_exception *pe, *pe2;
Mikulas Patocka515ad662009-12-10 23:52:30 +00002426 struct dm_snap_pending_exception *pe_to_start_now = NULL;
2427 struct dm_snap_pending_exception *pe_to_start_last = NULL;
Nikos Tsironisf79ae412019-03-17 14:22:57 +02002428 struct dm_exception_table_lock lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002429 chunk_t chunk;
2430
2431 /* Do all the snapshots on this origin */
2432 list_for_each_entry (snap, snapshots, list) {
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00002433 /*
2434 * Don't make new exceptions in a merging snapshot
2435 * because it has effectively been deleted
2436 */
2437 if (dm_target_is_snapshot_merge(snap->ti))
2438 continue;
2439
Alasdair G Kergond5e404c2005-07-12 15:53:05 -07002440 /* Nothing to do if writing beyond end of snapshot */
Mikulas Patocka9eaae8f2009-12-10 23:52:28 +00002441 if (sector >= dm_table_get_size(snap->ti->table))
Nikos Tsironisf79ae412019-03-17 14:22:57 +02002442 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002443
2444 /*
2445 * Remember, different snapshots can have
2446 * different chunk sizes.
2447 */
Mikulas Patocka9eaae8f2009-12-10 23:52:28 +00002448 chunk = sector_to_chunk(snap->store, sector);
Nikos Tsironisf79ae412019-03-17 14:22:57 +02002449 dm_exception_table_lock_init(snap, chunk, &lock);
2450
Nikos Tsironis3f1637f2019-03-17 14:22:58 +02002451 down_read(&snap->lock);
Nikos Tsironisf79ae412019-03-17 14:22:57 +02002452 dm_exception_table_lock(&lock);
2453
2454 /* Only deal with valid and active snapshots */
2455 if (!snap->valid || !snap->active)
2456 goto next_snapshot;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002457
Mikulas Patocka29138082009-04-02 19:55:25 +01002458 pe = __lookup_pending_exception(snap, chunk);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08002459 if (!pe) {
Nikos Tsironis65fc7c32019-03-17 14:22:55 +02002460 /*
2461 * Check exception table to see if block is already
2462 * remapped in this snapshot and trigger an exception
2463 * if not.
2464 */
2465 e = dm_lookup_exception(&snap->complete, chunk);
2466 if (e)
2467 goto next_snapshot;
2468
Nikos Tsironisf79ae412019-03-17 14:22:57 +02002469 dm_exception_table_unlock(&lock);
Mikulas Patockac6621392009-04-02 19:55:25 +01002470 pe = alloc_pending_exception(snap);
Nikos Tsironisf79ae412019-03-17 14:22:57 +02002471 dm_exception_table_lock(&lock);
Mikulas Patockac6621392009-04-02 19:55:25 +01002472
Nikos Tsironis65fc7c32019-03-17 14:22:55 +02002473 pe2 = __lookup_pending_exception(snap, chunk);
Mikulas Patocka35bf6592009-04-02 19:55:26 +01002474
Nikos Tsironis65fc7c32019-03-17 14:22:55 +02002475 if (!pe2) {
2476 e = dm_lookup_exception(&snap->complete, chunk);
2477 if (e) {
2478 free_pending_exception(pe);
2479 goto next_snapshot;
2480 }
2481
2482 pe = __insert_pending_exception(snap, pe, chunk);
2483 if (!pe) {
Nikos Tsironisf79ae412019-03-17 14:22:57 +02002484 dm_exception_table_unlock(&lock);
Nikos Tsironis3f1637f2019-03-17 14:22:58 +02002485 up_read(&snap->lock);
Nikos Tsironisf79ae412019-03-17 14:22:57 +02002486
Nikos Tsironis3f1637f2019-03-17 14:22:58 +02002487 invalidate_snapshot(snap, -ENOMEM);
Nikos Tsironisf79ae412019-03-17 14:22:57 +02002488 continue;
Nikos Tsironis65fc7c32019-03-17 14:22:55 +02002489 }
2490 } else {
2491 free_pending_exception(pe);
2492 pe = pe2;
Mikulas Patocka29138082009-04-02 19:55:25 +01002493 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002494 }
2495
Mikulas Patocka515ad662009-12-10 23:52:30 +00002496 r = DM_MAPIO_SUBMITTED;
2497
2498 /*
2499 * If an origin bio was supplied, queue it to wait for the
2500 * completion of this exception, and start this one last,
2501 * at the end of the function.
2502 */
2503 if (bio) {
2504 bio_list_add(&pe->origin_bios, bio);
2505 bio = NULL;
2506
2507 if (!pe->started) {
2508 pe->started = 1;
2509 pe_to_start_last = pe;
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08002510 }
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08002511 }
2512
2513 if (!pe->started) {
2514 pe->started = 1;
Mikulas Patocka515ad662009-12-10 23:52:30 +00002515 pe_to_start_now = pe;
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08002516 }
2517
Jonathan Brassowa2d2b032011-08-02 12:32:03 +01002518next_snapshot:
Nikos Tsironisf79ae412019-03-17 14:22:57 +02002519 dm_exception_table_unlock(&lock);
Nikos Tsironis3f1637f2019-03-17 14:22:58 +02002520 up_read(&snap->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002521
Mikulas Patocka515ad662009-12-10 23:52:30 +00002522 if (pe_to_start_now) {
2523 start_copy(pe_to_start_now);
2524 pe_to_start_now = NULL;
2525 }
Alasdair G Kergonb4b610f2006-03-27 01:17:44 -08002526 }
2527
Linus Torvalds1da177e2005-04-16 15:20:36 -07002528 /*
Mikulas Patocka515ad662009-12-10 23:52:30 +00002529 * Submit the exception against which the bio is queued last,
2530 * to give the other exceptions a head start.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002531 */
Mikulas Patocka515ad662009-12-10 23:52:30 +00002532 if (pe_to_start_last)
2533 start_copy(pe_to_start_last);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002534
2535 return r;
2536}
2537
2538/*
2539 * Called on a write from the origin driver.
2540 */
Mikulas Patockab2155572019-10-02 06:15:53 -04002541static int do_origin(struct dm_dev *origin, struct bio *bio, bool limit)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002542{
2543 struct origin *o;
Kiyoshi Uedad2a7ad22006-12-08 02:41:06 -08002544 int r = DM_MAPIO_REMAPPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002545
Mikulas Patockab2155572019-10-02 06:15:53 -04002546again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002547 down_read(&_origins_lock);
2548 o = __lookup_origin(origin->bdev);
Mikulas Patockab2155572019-10-02 06:15:53 -04002549 if (o) {
2550 if (limit) {
2551 struct dm_snapshot *s;
2552 list_for_each_entry(s, &o->snapshots, list)
2553 if (unlikely(!wait_for_in_progress(s, true)))
2554 goto again;
2555 }
2556
Kent Overstreet4f024f32013-10-11 15:44:27 -07002557 r = __origin_write(&o->snapshots, bio->bi_iter.bi_sector, bio);
Mikulas Patockab2155572019-10-02 06:15:53 -04002558 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002559 up_read(&_origins_lock);
2560
2561 return r;
2562}
2563
2564/*
Mikulas Patocka73dfd072009-12-10 23:52:34 +00002565 * Trigger exceptions in all non-merging snapshots.
2566 *
2567 * The chunk size of the merging snapshot may be larger than the chunk
2568 * size of some other snapshot so we may need to reallocate multiple
2569 * chunks in other snapshots.
2570 *
2571 * We scan all the overlapping exceptions in the other snapshots.
2572 * Returns 1 if anything was reallocated and must be waited for,
2573 * otherwise returns 0.
2574 *
2575 * size must be a multiple of merging_snap's chunk_size.
2576 */
2577static int origin_write_extent(struct dm_snapshot *merging_snap,
2578 sector_t sector, unsigned size)
2579{
2580 int must_wait = 0;
2581 sector_t n;
2582 struct origin *o;
2583
2584 /*
Mike Snitzer542f9032012-07-27 15:08:00 +01002585 * The origin's __minimum_chunk_size() got stored in max_io_len
Mikulas Patocka73dfd072009-12-10 23:52:34 +00002586 * by snapshot_merge_resume().
2587 */
2588 down_read(&_origins_lock);
2589 o = __lookup_origin(merging_snap->origin->bdev);
Mike Snitzer542f9032012-07-27 15:08:00 +01002590 for (n = 0; n < size; n += merging_snap->ti->max_io_len)
Mikulas Patocka73dfd072009-12-10 23:52:34 +00002591 if (__origin_write(&o->snapshots, sector + n, NULL) ==
2592 DM_MAPIO_SUBMITTED)
2593 must_wait = 1;
2594 up_read(&_origins_lock);
2595
2596 return must_wait;
2597}
2598
2599/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002600 * Origin: maps a linear range of a device, with hooks for snapshotting.
2601 */
2602
2603/*
2604 * Construct an origin mapping: <dev_path>
2605 * The context for an origin is merely a 'struct dm_dev *'
2606 * pointing to the real device.
2607 */
2608static int origin_ctr(struct dm_target *ti, unsigned int argc, char **argv)
2609{
2610 int r;
Mikulas Patocka599cdf32014-03-14 18:42:12 -04002611 struct dm_origin *o;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002612
2613 if (argc != 1) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -07002614 ti->error = "origin: incorrect number of arguments";
Linus Torvalds1da177e2005-04-16 15:20:36 -07002615 return -EINVAL;
2616 }
2617
Mikulas Patocka599cdf32014-03-14 18:42:12 -04002618 o = kmalloc(sizeof(struct dm_origin), GFP_KERNEL);
2619 if (!o) {
2620 ti->error = "Cannot allocate private origin structure";
2621 r = -ENOMEM;
2622 goto bad_alloc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002623 }
2624
Mikulas Patocka599cdf32014-03-14 18:42:12 -04002625 r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &o->dev);
2626 if (r) {
2627 ti->error = "Cannot get target device";
2628 goto bad_open;
2629 }
2630
Mikulas Patockab735fed2015-02-26 11:40:35 -05002631 o->ti = ti;
Mikulas Patocka599cdf32014-03-14 18:42:12 -04002632 ti->private = o;
Alasdair G Kergon55a62ee2013-03-01 22:45:47 +00002633 ti->num_flush_bios = 1;
Mikulas Patocka494b3ee2009-06-22 10:12:25 +01002634
Linus Torvalds1da177e2005-04-16 15:20:36 -07002635 return 0;
Mikulas Patocka599cdf32014-03-14 18:42:12 -04002636
2637bad_open:
2638 kfree(o);
2639bad_alloc:
2640 return r;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002641}
2642
2643static void origin_dtr(struct dm_target *ti)
2644{
Mikulas Patocka599cdf32014-03-14 18:42:12 -04002645 struct dm_origin *o = ti->private;
Mikulas Patockab735fed2015-02-26 11:40:35 -05002646
Mikulas Patocka599cdf32014-03-14 18:42:12 -04002647 dm_put_device(ti, o->dev);
2648 kfree(o);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002649}
2650
Mikulas Patocka7de3ee52012-12-21 20:23:41 +00002651static int origin_map(struct dm_target *ti, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002652{
Mikulas Patocka599cdf32014-03-14 18:42:12 -04002653 struct dm_origin *o = ti->private;
Mikulas Patocka298eaa82014-03-14 18:43:07 -04002654 unsigned available_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002655
Christoph Hellwig74d46992017-08-23 19:10:32 +02002656 bio_set_dev(bio, o->dev->bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002657
Jens Axboe1eff9d32016-08-05 15:35:16 -06002658 if (unlikely(bio->bi_opf & REQ_PREFLUSH))
Mikulas Patocka494b3ee2009-06-22 10:12:25 +01002659 return DM_MAPIO_REMAPPED;
2660
Christoph Hellwig70246282016-07-19 11:28:41 +02002661 if (bio_data_dir(bio) != WRITE)
Mikulas Patocka298eaa82014-03-14 18:43:07 -04002662 return DM_MAPIO_REMAPPED;
2663
2664 available_sectors = o->split_boundary -
2665 ((unsigned)bio->bi_iter.bi_sector & (o->split_boundary - 1));
2666
2667 if (bio_sectors(bio) > available_sectors)
2668 dm_accept_partial_bio(bio, available_sectors);
2669
Linus Torvalds1da177e2005-04-16 15:20:36 -07002670 /* Only tell snapshots if this is a write */
Mikulas Patockab2155572019-10-02 06:15:53 -04002671 return do_origin(o->dev, bio, true);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002672}
2673
Linus Torvalds1da177e2005-04-16 15:20:36 -07002674/*
Mike Snitzer542f9032012-07-27 15:08:00 +01002675 * Set the target "max_io_len" field to the minimum of all the snapshots'
Linus Torvalds1da177e2005-04-16 15:20:36 -07002676 * chunk sizes.
2677 */
2678static void origin_resume(struct dm_target *ti)
2679{
Mikulas Patocka599cdf32014-03-14 18:42:12 -04002680 struct dm_origin *o = ti->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002681
Mikulas Patocka298eaa82014-03-14 18:43:07 -04002682 o->split_boundary = get_origin_minimum_chunksize(o->dev->bdev);
Mikulas Patockab735fed2015-02-26 11:40:35 -05002683
2684 down_write(&_origins_lock);
2685 __insert_dm_origin(o);
2686 up_write(&_origins_lock);
2687}
2688
2689static void origin_postsuspend(struct dm_target *ti)
2690{
2691 struct dm_origin *o = ti->private;
2692
2693 down_write(&_origins_lock);
2694 __remove_dm_origin(o);
2695 up_write(&_origins_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002696}
2697
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002698static void origin_status(struct dm_target *ti, status_type_t type,
2699 unsigned status_flags, char *result, unsigned maxlen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002700{
Mikulas Patocka599cdf32014-03-14 18:42:12 -04002701 struct dm_origin *o = ti->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002702
2703 switch (type) {
2704 case STATUSTYPE_INFO:
2705 result[0] = '\0';
2706 break;
2707
2708 case STATUSTYPE_TABLE:
Mikulas Patocka599cdf32014-03-14 18:42:12 -04002709 snprintf(result, maxlen, "%s", o->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002710 break;
2711 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002712}
2713
Mike Snitzer8811f462009-09-04 20:40:19 +01002714static int origin_iterate_devices(struct dm_target *ti,
2715 iterate_devices_callout_fn fn, void *data)
2716{
Mikulas Patocka599cdf32014-03-14 18:42:12 -04002717 struct dm_origin *o = ti->private;
Mike Snitzer8811f462009-09-04 20:40:19 +01002718
Mikulas Patocka599cdf32014-03-14 18:42:12 -04002719 return fn(ti, o->dev, 0, ti->len, data);
Mike Snitzer8811f462009-09-04 20:40:19 +01002720}
2721
Linus Torvalds1da177e2005-04-16 15:20:36 -07002722static struct target_type origin_target = {
2723 .name = "snapshot-origin",
Mikulas Patockab735fed2015-02-26 11:40:35 -05002724 .version = {1, 9, 0},
Linus Torvalds1da177e2005-04-16 15:20:36 -07002725 .module = THIS_MODULE,
2726 .ctr = origin_ctr,
2727 .dtr = origin_dtr,
2728 .map = origin_map,
2729 .resume = origin_resume,
Mikulas Patockab735fed2015-02-26 11:40:35 -05002730 .postsuspend = origin_postsuspend,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002731 .status = origin_status,
Mike Snitzer8811f462009-09-04 20:40:19 +01002732 .iterate_devices = origin_iterate_devices,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002733};
2734
2735static struct target_type snapshot_target = {
2736 .name = "snapshot",
Mike Snitzer2e602382019-06-19 17:05:54 -04002737 .version = {1, 16, 0},
Linus Torvalds1da177e2005-04-16 15:20:36 -07002738 .module = THIS_MODULE,
2739 .ctr = snapshot_ctr,
2740 .dtr = snapshot_dtr,
2741 .map = snapshot_map,
Mikulas Patockacd45daf2008-07-21 12:00:32 +01002742 .end_io = snapshot_end_io,
Mike Snitzerc1f0c182009-12-10 23:52:24 +00002743 .preresume = snapshot_preresume,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002744 .resume = snapshot_resume,
2745 .status = snapshot_status,
Mike Snitzer8811f462009-09-04 20:40:19 +01002746 .iterate_devices = snapshot_iterate_devices,
Mike Snitzer2e602382019-06-19 17:05:54 -04002747 .io_hints = snapshot_io_hints,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002748};
2749
Mikulas Patockad698aa42009-12-10 23:52:30 +00002750static struct target_type merge_target = {
2751 .name = dm_snapshot_merge_target_name,
Mike Snitzer2e602382019-06-19 17:05:54 -04002752 .version = {1, 5, 0},
Mikulas Patockad698aa42009-12-10 23:52:30 +00002753 .module = THIS_MODULE,
2754 .ctr = snapshot_ctr,
2755 .dtr = snapshot_dtr,
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00002756 .map = snapshot_merge_map,
Mikulas Patockad698aa42009-12-10 23:52:30 +00002757 .end_io = snapshot_end_io,
Mikulas Patocka1e03f972009-12-10 23:52:32 +00002758 .presuspend = snapshot_merge_presuspend,
Mikulas Patockad698aa42009-12-10 23:52:30 +00002759 .preresume = snapshot_preresume,
Mikulas Patocka1e03f972009-12-10 23:52:32 +00002760 .resume = snapshot_merge_resume,
Mikulas Patockad698aa42009-12-10 23:52:30 +00002761 .status = snapshot_status,
2762 .iterate_devices = snapshot_iterate_devices,
Mike Snitzer2e602382019-06-19 17:05:54 -04002763 .io_hints = snapshot_io_hints,
Mikulas Patockad698aa42009-12-10 23:52:30 +00002764};
2765
Linus Torvalds1da177e2005-04-16 15:20:36 -07002766static int __init dm_snapshot_init(void)
2767{
2768 int r;
2769
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +00002770 r = dm_exception_store_init();
2771 if (r) {
2772 DMERR("Failed to initialize exception stores");
2773 return r;
2774 }
2775
Linus Torvalds1da177e2005-04-16 15:20:36 -07002776 r = init_origin_hash();
2777 if (r) {
2778 DMERR("init_origin_hash failed.");
Mikulas Patockad698aa42009-12-10 23:52:30 +00002779 goto bad_origin_hash;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002780 }
2781
Jon Brassow1d4989c2009-12-10 23:52:10 +00002782 exception_cache = KMEM_CACHE(dm_exception, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002783 if (!exception_cache) {
2784 DMERR("Couldn't create exception cache.");
2785 r = -ENOMEM;
Mikulas Patockad698aa42009-12-10 23:52:30 +00002786 goto bad_exception_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002787 }
2788
Alasdair G Kergon028867a2007-07-12 17:26:32 +01002789 pending_cache = KMEM_CACHE(dm_snap_pending_exception, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002790 if (!pending_cache) {
2791 DMERR("Couldn't create pending cache.");
2792 r = -ENOMEM;
Mikulas Patockad698aa42009-12-10 23:52:30 +00002793 goto bad_pending_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002794 }
2795
monty_pavel@sina.com7e6358d2017-11-25 01:43:50 +08002796 r = dm_register_target(&snapshot_target);
2797 if (r < 0) {
2798 DMERR("snapshot target register failed %d", r);
2799 goto bad_register_snapshot_target;
2800 }
2801
2802 r = dm_register_target(&origin_target);
2803 if (r < 0) {
2804 DMERR("Origin target register failed %d", r);
2805 goto bad_register_origin_target;
2806 }
2807
2808 r = dm_register_target(&merge_target);
2809 if (r < 0) {
2810 DMERR("Merge target register failed %d", r);
2811 goto bad_register_merge_target;
2812 }
2813
Linus Torvalds1da177e2005-04-16 15:20:36 -07002814 return 0;
2815
Mikulas Patockad698aa42009-12-10 23:52:30 +00002816bad_register_merge_target:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002817 dm_unregister_target(&origin_target);
Mikulas Patockad698aa42009-12-10 23:52:30 +00002818bad_register_origin_target:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002819 dm_unregister_target(&snapshot_target);
Jonathan Brassow034a1862009-10-16 23:18:14 +01002820bad_register_snapshot_target:
monty_pavel@sina.com7e6358d2017-11-25 01:43:50 +08002821 kmem_cache_destroy(pending_cache);
2822bad_pending_cache:
2823 kmem_cache_destroy(exception_cache);
2824bad_exception_cache:
2825 exit_origin_hash();
2826bad_origin_hash:
Jonathan Brassow034a1862009-10-16 23:18:14 +01002827 dm_exception_store_exit();
Mikulas Patockad698aa42009-12-10 23:52:30 +00002828
Linus Torvalds1da177e2005-04-16 15:20:36 -07002829 return r;
2830}
2831
2832static void __exit dm_snapshot_exit(void)
2833{
Mikulas Patocka10d3bd02009-01-06 03:04:58 +00002834 dm_unregister_target(&snapshot_target);
2835 dm_unregister_target(&origin_target);
Mikulas Patockad698aa42009-12-10 23:52:30 +00002836 dm_unregister_target(&merge_target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002837
2838 exit_origin_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002839 kmem_cache_destroy(pending_cache);
2840 kmem_cache_destroy(exception_cache);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +00002841
2842 dm_exception_store_exit();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002843}
2844
2845/* Module hooks */
2846module_init(dm_snapshot_init);
2847module_exit(dm_snapshot_exit);
2848
2849MODULE_DESCRIPTION(DM_NAME " snapshot target");
2850MODULE_AUTHOR("Joe Thornber");
2851MODULE_LICENSE("GPL");
Mikulas Patocka23cb2102013-03-01 22:45:47 +00002852MODULE_ALIAS("dm-snapshot-origin");
2853MODULE_ALIAS("dm-snapshot-merge");