Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * dm-snapshot.c | |
3 | * | |
4 | * Copyright (C) 2001-2002 Sistina Software (UK) Limited. | |
5 | * | |
6 | * This file is released under the GPL. | |
7 | */ | |
8 | ||
9 | #include <linux/blkdev.h> | |
1da177e4 LT |
10 | #include <linux/ctype.h> |
11 | #include <linux/device-mapper.h> | |
12 | #include <linux/fs.h> | |
13 | #include <linux/init.h> | |
14 | #include <linux/kdev_t.h> | |
15 | #include <linux/list.h> | |
16 | #include <linux/mempool.h> | |
17 | #include <linux/module.h> | |
18 | #include <linux/slab.h> | |
19 | #include <linux/vmalloc.h> | |
6f3c3f0a | 20 | #include <linux/log2.h> |
a765e20e | 21 | #include <linux/dm-kcopyd.h> |
1da177e4 LT |
22 | |
23 | #include "dm-snap.h" | |
24 | #include "dm-bio-list.h" | |
1da177e4 | 25 | |
72d94861 AK |
26 | #define DM_MSG_PREFIX "snapshots" |
27 | ||
1da177e4 LT |
28 | /* |
29 | * The percentage increment we will wake up users at | |
30 | */ | |
31 | #define WAKE_UP_PERCENT 5 | |
32 | ||
33 | /* | |
34 | * kcopyd priority of snapshot operations | |
35 | */ | |
36 | #define SNAPSHOT_COPY_PRIORITY 2 | |
37 | ||
38 | /* | |
8ee2767a | 39 | * Reserve 1MB for each snapshot initially (with minimum of 1 page). |
1da177e4 | 40 | */ |
8ee2767a | 41 | #define SNAPSHOT_PAGES (((1UL << 20) >> PAGE_SHIFT) ? : 1) |
1da177e4 | 42 | |
cd45daff MP |
43 | /* |
44 | * The size of the mempool used to track chunks in use. | |
45 | */ | |
46 | #define MIN_IOS 256 | |
47 | ||
c642f9e0 | 48 | static struct workqueue_struct *ksnapd; |
c4028958 | 49 | static void flush_queued_bios(struct work_struct *work); |
ca3a931f | 50 | |
028867ac AK |
51 | struct dm_snap_pending_exception { |
52 | struct dm_snap_exception e; | |
1da177e4 LT |
53 | |
54 | /* | |
55 | * Origin buffers waiting for this to complete are held | |
56 | * in a bio list | |
57 | */ | |
58 | struct bio_list origin_bios; | |
59 | struct bio_list snapshot_bios; | |
60 | ||
eccf0817 AK |
61 | /* |
62 | * Short-term queue of pending exceptions prior to submission. | |
63 | */ | |
64 | struct list_head list; | |
65 | ||
1da177e4 | 66 | /* |
b4b610f6 | 67 | * The primary pending_exception is the one that holds |
4b832e8d | 68 | * the ref_count and the list of origin_bios for a |
b4b610f6 AK |
69 | * group of pending_exceptions. It is always last to get freed. |
70 | * These fields get set up when writing to the origin. | |
1da177e4 | 71 | */ |
028867ac | 72 | struct dm_snap_pending_exception *primary_pe; |
b4b610f6 AK |
73 | |
74 | /* | |
75 | * Number of pending_exceptions processing this chunk. | |
76 | * When this drops to zero we must complete the origin bios. | |
77 | * If incrementing or decrementing this, hold pe->snap->lock for | |
78 | * the sibling concerned and not pe->primary_pe->snap->lock unless | |
79 | * they are the same. | |
80 | */ | |
4b832e8d | 81 | atomic_t ref_count; |
1da177e4 LT |
82 | |
83 | /* Pointer back to snapshot context */ | |
84 | struct dm_snapshot *snap; | |
85 | ||
86 | /* | |
87 | * 1 indicates the exception has already been sent to | |
88 | * kcopyd. | |
89 | */ | |
90 | int started; | |
91 | }; | |
92 | ||
93 | /* | |
94 | * Hash table mapping origin volumes to lists of snapshots and | |
95 | * a lock to protect it | |
96 | */ | |
e18b890b CL |
97 | static struct kmem_cache *exception_cache; |
98 | static struct kmem_cache *pending_cache; | |
1da177e4 | 99 | |
cd45daff MP |
100 | struct dm_snap_tracked_chunk { |
101 | struct hlist_node node; | |
102 | chunk_t chunk; | |
103 | }; | |
104 | ||
105 | static struct kmem_cache *tracked_chunk_cache; | |
106 | ||
107 | static struct dm_snap_tracked_chunk *track_chunk(struct dm_snapshot *s, | |
108 | chunk_t chunk) | |
109 | { | |
110 | struct dm_snap_tracked_chunk *c = mempool_alloc(s->tracked_chunk_pool, | |
111 | GFP_NOIO); | |
112 | unsigned long flags; | |
113 | ||
114 | c->chunk = chunk; | |
115 | ||
116 | spin_lock_irqsave(&s->tracked_chunk_lock, flags); | |
117 | hlist_add_head(&c->node, | |
118 | &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)]); | |
119 | spin_unlock_irqrestore(&s->tracked_chunk_lock, flags); | |
120 | ||
121 | return c; | |
122 | } | |
123 | ||
124 | static void stop_tracking_chunk(struct dm_snapshot *s, | |
125 | struct dm_snap_tracked_chunk *c) | |
126 | { | |
127 | unsigned long flags; | |
128 | ||
129 | spin_lock_irqsave(&s->tracked_chunk_lock, flags); | |
130 | hlist_del(&c->node); | |
131 | spin_unlock_irqrestore(&s->tracked_chunk_lock, flags); | |
132 | ||
133 | mempool_free(c, s->tracked_chunk_pool); | |
134 | } | |
135 | ||
a8d41b59 MP |
136 | static int __chunk_is_tracked(struct dm_snapshot *s, chunk_t chunk) |
137 | { | |
138 | struct dm_snap_tracked_chunk *c; | |
139 | struct hlist_node *hn; | |
140 | int found = 0; | |
141 | ||
142 | spin_lock_irq(&s->tracked_chunk_lock); | |
143 | ||
144 | hlist_for_each_entry(c, hn, | |
145 | &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)], node) { | |
146 | if (c->chunk == chunk) { | |
147 | found = 1; | |
148 | break; | |
149 | } | |
150 | } | |
151 | ||
152 | spin_unlock_irq(&s->tracked_chunk_lock); | |
153 | ||
154 | return found; | |
155 | } | |
156 | ||
1da177e4 LT |
157 | /* |
158 | * One of these per registered origin, held in the snapshot_origins hash | |
159 | */ | |
160 | struct origin { | |
161 | /* The origin device */ | |
162 | struct block_device *bdev; | |
163 | ||
164 | struct list_head hash_list; | |
165 | ||
166 | /* List of snapshots for this origin */ | |
167 | struct list_head snapshots; | |
168 | }; | |
169 | ||
170 | /* | |
171 | * Size of the hash table for origin volumes. If we make this | |
172 | * the size of the minors list then it should be nearly perfect | |
173 | */ | |
174 | #define ORIGIN_HASH_SIZE 256 | |
175 | #define ORIGIN_MASK 0xFF | |
176 | static struct list_head *_origins; | |
177 | static struct rw_semaphore _origins_lock; | |
178 | ||
179 | static int init_origin_hash(void) | |
180 | { | |
181 | int i; | |
182 | ||
183 | _origins = kmalloc(ORIGIN_HASH_SIZE * sizeof(struct list_head), | |
184 | GFP_KERNEL); | |
185 | if (!_origins) { | |
72d94861 | 186 | DMERR("unable to allocate memory"); |
1da177e4 LT |
187 | return -ENOMEM; |
188 | } | |
189 | ||
190 | for (i = 0; i < ORIGIN_HASH_SIZE; i++) | |
191 | INIT_LIST_HEAD(_origins + i); | |
192 | init_rwsem(&_origins_lock); | |
193 | ||
194 | return 0; | |
195 | } | |
196 | ||
197 | static void exit_origin_hash(void) | |
198 | { | |
199 | kfree(_origins); | |
200 | } | |
201 | ||
028867ac | 202 | static unsigned origin_hash(struct block_device *bdev) |
1da177e4 LT |
203 | { |
204 | return bdev->bd_dev & ORIGIN_MASK; | |
205 | } | |
206 | ||
207 | static struct origin *__lookup_origin(struct block_device *origin) | |
208 | { | |
209 | struct list_head *ol; | |
210 | struct origin *o; | |
211 | ||
212 | ol = &_origins[origin_hash(origin)]; | |
213 | list_for_each_entry (o, ol, hash_list) | |
214 | if (bdev_equal(o->bdev, origin)) | |
215 | return o; | |
216 | ||
217 | return NULL; | |
218 | } | |
219 | ||
220 | static void __insert_origin(struct origin *o) | |
221 | { | |
222 | struct list_head *sl = &_origins[origin_hash(o->bdev)]; | |
223 | list_add_tail(&o->hash_list, sl); | |
224 | } | |
225 | ||
226 | /* | |
227 | * Make a note of the snapshot and its origin so we can look it | |
228 | * up when the origin has a write on it. | |
229 | */ | |
230 | static int register_snapshot(struct dm_snapshot *snap) | |
231 | { | |
232 | struct origin *o; | |
233 | struct block_device *bdev = snap->origin->bdev; | |
234 | ||
235 | down_write(&_origins_lock); | |
236 | o = __lookup_origin(bdev); | |
237 | ||
238 | if (!o) { | |
239 | /* New origin */ | |
240 | o = kmalloc(sizeof(*o), GFP_KERNEL); | |
241 | if (!o) { | |
242 | up_write(&_origins_lock); | |
243 | return -ENOMEM; | |
244 | } | |
245 | ||
246 | /* Initialise the struct */ | |
247 | INIT_LIST_HEAD(&o->snapshots); | |
248 | o->bdev = bdev; | |
249 | ||
250 | __insert_origin(o); | |
251 | } | |
252 | ||
253 | list_add_tail(&snap->list, &o->snapshots); | |
254 | ||
255 | up_write(&_origins_lock); | |
256 | return 0; | |
257 | } | |
258 | ||
259 | static void unregister_snapshot(struct dm_snapshot *s) | |
260 | { | |
261 | struct origin *o; | |
262 | ||
263 | down_write(&_origins_lock); | |
264 | o = __lookup_origin(s->origin->bdev); | |
265 | ||
266 | list_del(&s->list); | |
267 | if (list_empty(&o->snapshots)) { | |
268 | list_del(&o->hash_list); | |
269 | kfree(o); | |
270 | } | |
271 | ||
272 | up_write(&_origins_lock); | |
273 | } | |
274 | ||
275 | /* | |
276 | * Implementation of the exception hash tables. | |
d74f81f8 MB |
277 | * The lowest hash_shift bits of the chunk number are ignored, allowing |
278 | * some consecutive chunks to be grouped together. | |
1da177e4 | 279 | */ |
d74f81f8 MB |
280 | static int init_exception_table(struct exception_table *et, uint32_t size, |
281 | unsigned hash_shift) | |
1da177e4 LT |
282 | { |
283 | unsigned int i; | |
284 | ||
d74f81f8 | 285 | et->hash_shift = hash_shift; |
1da177e4 LT |
286 | et->hash_mask = size - 1; |
287 | et->table = dm_vcalloc(size, sizeof(struct list_head)); | |
288 | if (!et->table) | |
289 | return -ENOMEM; | |
290 | ||
291 | for (i = 0; i < size; i++) | |
292 | INIT_LIST_HEAD(et->table + i); | |
293 | ||
294 | return 0; | |
295 | } | |
296 | ||
e18b890b | 297 | static void exit_exception_table(struct exception_table *et, struct kmem_cache *mem) |
1da177e4 LT |
298 | { |
299 | struct list_head *slot; | |
028867ac | 300 | struct dm_snap_exception *ex, *next; |
1da177e4 LT |
301 | int i, size; |
302 | ||
303 | size = et->hash_mask + 1; | |
304 | for (i = 0; i < size; i++) { | |
305 | slot = et->table + i; | |
306 | ||
307 | list_for_each_entry_safe (ex, next, slot, hash_list) | |
308 | kmem_cache_free(mem, ex); | |
309 | } | |
310 | ||
311 | vfree(et->table); | |
312 | } | |
313 | ||
028867ac | 314 | static uint32_t exception_hash(struct exception_table *et, chunk_t chunk) |
1da177e4 | 315 | { |
d74f81f8 | 316 | return (chunk >> et->hash_shift) & et->hash_mask; |
1da177e4 LT |
317 | } |
318 | ||
028867ac AK |
319 | static void insert_exception(struct exception_table *eh, |
320 | struct dm_snap_exception *e) | |
1da177e4 LT |
321 | { |
322 | struct list_head *l = &eh->table[exception_hash(eh, e->old_chunk)]; | |
323 | list_add(&e->hash_list, l); | |
324 | } | |
325 | ||
028867ac | 326 | static void remove_exception(struct dm_snap_exception *e) |
1da177e4 LT |
327 | { |
328 | list_del(&e->hash_list); | |
329 | } | |
330 | ||
331 | /* | |
332 | * Return the exception data for a sector, or NULL if not | |
333 | * remapped. | |
334 | */ | |
028867ac AK |
335 | static struct dm_snap_exception *lookup_exception(struct exception_table *et, |
336 | chunk_t chunk) | |
1da177e4 LT |
337 | { |
338 | struct list_head *slot; | |
028867ac | 339 | struct dm_snap_exception *e; |
1da177e4 LT |
340 | |
341 | slot = &et->table[exception_hash(et, chunk)]; | |
342 | list_for_each_entry (e, slot, hash_list) | |
d74f81f8 MB |
343 | if (chunk >= e->old_chunk && |
344 | chunk <= e->old_chunk + dm_consecutive_chunk_count(e)) | |
1da177e4 LT |
345 | return e; |
346 | ||
347 | return NULL; | |
348 | } | |
349 | ||
028867ac | 350 | static struct dm_snap_exception *alloc_exception(void) |
1da177e4 | 351 | { |
028867ac | 352 | struct dm_snap_exception *e; |
1da177e4 LT |
353 | |
354 | e = kmem_cache_alloc(exception_cache, GFP_NOIO); | |
355 | if (!e) | |
356 | e = kmem_cache_alloc(exception_cache, GFP_ATOMIC); | |
357 | ||
358 | return e; | |
359 | } | |
360 | ||
028867ac | 361 | static void free_exception(struct dm_snap_exception *e) |
1da177e4 LT |
362 | { |
363 | kmem_cache_free(exception_cache, e); | |
364 | } | |
365 | ||
92e86812 | 366 | static struct dm_snap_pending_exception *alloc_pending_exception(struct dm_snapshot *s) |
1da177e4 | 367 | { |
92e86812 MP |
368 | struct dm_snap_pending_exception *pe = mempool_alloc(s->pending_pool, |
369 | GFP_NOIO); | |
370 | ||
371 | pe->snap = s; | |
372 | ||
373 | return pe; | |
1da177e4 LT |
374 | } |
375 | ||
028867ac | 376 | static void free_pending_exception(struct dm_snap_pending_exception *pe) |
1da177e4 | 377 | { |
92e86812 | 378 | mempool_free(pe, pe->snap->pending_pool); |
1da177e4 LT |
379 | } |
380 | ||
d74f81f8 MB |
381 | static void insert_completed_exception(struct dm_snapshot *s, |
382 | struct dm_snap_exception *new_e) | |
383 | { | |
384 | struct exception_table *eh = &s->complete; | |
385 | struct list_head *l; | |
386 | struct dm_snap_exception *e = NULL; | |
387 | ||
388 | l = &eh->table[exception_hash(eh, new_e->old_chunk)]; | |
389 | ||
390 | /* Add immediately if this table doesn't support consecutive chunks */ | |
391 | if (!eh->hash_shift) | |
392 | goto out; | |
393 | ||
394 | /* List is ordered by old_chunk */ | |
395 | list_for_each_entry_reverse(e, l, hash_list) { | |
396 | /* Insert after an existing chunk? */ | |
397 | if (new_e->old_chunk == (e->old_chunk + | |
398 | dm_consecutive_chunk_count(e) + 1) && | |
399 | new_e->new_chunk == (dm_chunk_number(e->new_chunk) + | |
400 | dm_consecutive_chunk_count(e) + 1)) { | |
401 | dm_consecutive_chunk_count_inc(e); | |
402 | free_exception(new_e); | |
403 | return; | |
404 | } | |
405 | ||
406 | /* Insert before an existing chunk? */ | |
407 | if (new_e->old_chunk == (e->old_chunk - 1) && | |
408 | new_e->new_chunk == (dm_chunk_number(e->new_chunk) - 1)) { | |
409 | dm_consecutive_chunk_count_inc(e); | |
410 | e->old_chunk--; | |
411 | e->new_chunk--; | |
412 | free_exception(new_e); | |
413 | return; | |
414 | } | |
415 | ||
416 | if (new_e->old_chunk > e->old_chunk) | |
417 | break; | |
418 | } | |
419 | ||
420 | out: | |
421 | list_add(&new_e->hash_list, e ? &e->hash_list : l); | |
422 | } | |
423 | ||
1da177e4 LT |
424 | int dm_add_exception(struct dm_snapshot *s, chunk_t old, chunk_t new) |
425 | { | |
028867ac | 426 | struct dm_snap_exception *e; |
1da177e4 LT |
427 | |
428 | e = alloc_exception(); | |
429 | if (!e) | |
430 | return -ENOMEM; | |
431 | ||
432 | e->old_chunk = old; | |
d74f81f8 MB |
433 | |
434 | /* Consecutive_count is implicitly initialised to zero */ | |
1da177e4 | 435 | e->new_chunk = new; |
d74f81f8 MB |
436 | |
437 | insert_completed_exception(s, e); | |
438 | ||
1da177e4 LT |
439 | return 0; |
440 | } | |
441 | ||
442 | /* | |
443 | * Hard coded magic. | |
444 | */ | |
445 | static int calc_max_buckets(void) | |
446 | { | |
447 | /* use a fixed size of 2MB */ | |
448 | unsigned long mem = 2 * 1024 * 1024; | |
449 | mem /= sizeof(struct list_head); | |
450 | ||
451 | return mem; | |
452 | } | |
453 | ||
1da177e4 LT |
454 | /* |
455 | * Allocate room for a suitable hash table. | |
456 | */ | |
457 | static int init_hash_tables(struct dm_snapshot *s) | |
458 | { | |
459 | sector_t hash_size, cow_dev_size, origin_dev_size, max_buckets; | |
460 | ||
461 | /* | |
462 | * Calculate based on the size of the original volume or | |
463 | * the COW volume... | |
464 | */ | |
465 | cow_dev_size = get_dev_size(s->cow->bdev); | |
466 | origin_dev_size = get_dev_size(s->origin->bdev); | |
467 | max_buckets = calc_max_buckets(); | |
468 | ||
469 | hash_size = min(origin_dev_size, cow_dev_size) >> s->chunk_shift; | |
470 | hash_size = min(hash_size, max_buckets); | |
471 | ||
8defd830 | 472 | hash_size = rounddown_pow_of_two(hash_size); |
d74f81f8 MB |
473 | if (init_exception_table(&s->complete, hash_size, |
474 | DM_CHUNK_CONSECUTIVE_BITS)) | |
1da177e4 LT |
475 | return -ENOMEM; |
476 | ||
477 | /* | |
478 | * Allocate hash table for in-flight exceptions | |
479 | * Make this smaller than the real hash table | |
480 | */ | |
481 | hash_size >>= 3; | |
482 | if (hash_size < 64) | |
483 | hash_size = 64; | |
484 | ||
d74f81f8 | 485 | if (init_exception_table(&s->pending, hash_size, 0)) { |
1da177e4 LT |
486 | exit_exception_table(&s->complete, exception_cache); |
487 | return -ENOMEM; | |
488 | } | |
489 | ||
490 | return 0; | |
491 | } | |
492 | ||
493 | /* | |
494 | * Round a number up to the nearest 'size' boundary. size must | |
495 | * be a power of 2. | |
496 | */ | |
028867ac | 497 | static ulong round_up(ulong n, ulong size) |
1da177e4 LT |
498 | { |
499 | size--; | |
500 | return (n + size) & ~size; | |
501 | } | |
502 | ||
4c7e3bf4 MM |
503 | static int set_chunk_size(struct dm_snapshot *s, const char *chunk_size_arg, |
504 | char **error) | |
505 | { | |
506 | unsigned long chunk_size; | |
507 | char *value; | |
508 | ||
509 | chunk_size = simple_strtoul(chunk_size_arg, &value, 10); | |
510 | if (*chunk_size_arg == '\0' || *value != '\0') { | |
511 | *error = "Invalid chunk size"; | |
512 | return -EINVAL; | |
513 | } | |
514 | ||
515 | if (!chunk_size) { | |
516 | s->chunk_size = s->chunk_mask = s->chunk_shift = 0; | |
517 | return 0; | |
518 | } | |
519 | ||
520 | /* | |
521 | * Chunk size must be multiple of page size. Silently | |
522 | * round up if it's not. | |
523 | */ | |
524 | chunk_size = round_up(chunk_size, PAGE_SIZE >> 9); | |
525 | ||
526 | /* Check chunk_size is a power of 2 */ | |
6f3c3f0a | 527 | if (!is_power_of_2(chunk_size)) { |
4c7e3bf4 MM |
528 | *error = "Chunk size is not a power of 2"; |
529 | return -EINVAL; | |
530 | } | |
531 | ||
532 | /* Validate the chunk size against the device block size */ | |
533 | if (chunk_size % (bdev_hardsect_size(s->cow->bdev) >> 9)) { | |
534 | *error = "Chunk size is not a multiple of device blocksize"; | |
535 | return -EINVAL; | |
536 | } | |
537 | ||
538 | s->chunk_size = chunk_size; | |
539 | s->chunk_mask = chunk_size - 1; | |
540 | s->chunk_shift = ffs(chunk_size) - 1; | |
541 | ||
542 | return 0; | |
543 | } | |
544 | ||
1da177e4 LT |
545 | /* |
546 | * Construct a snapshot mapping: <origin_dev> <COW-dev> <p/n> <chunk-size> | |
547 | */ | |
548 | static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv) | |
549 | { | |
550 | struct dm_snapshot *s; | |
cd45daff | 551 | int i; |
1da177e4 LT |
552 | int r = -EINVAL; |
553 | char persistent; | |
554 | char *origin_path; | |
555 | char *cow_path; | |
1da177e4 | 556 | |
4c7e3bf4 | 557 | if (argc != 4) { |
72d94861 | 558 | ti->error = "requires exactly 4 arguments"; |
1da177e4 LT |
559 | r = -EINVAL; |
560 | goto bad1; | |
561 | } | |
562 | ||
563 | origin_path = argv[0]; | |
564 | cow_path = argv[1]; | |
565 | persistent = toupper(*argv[2]); | |
566 | ||
567 | if (persistent != 'P' && persistent != 'N') { | |
568 | ti->error = "Persistent flag is not P or N"; | |
569 | r = -EINVAL; | |
570 | goto bad1; | |
571 | } | |
572 | ||
1da177e4 LT |
573 | s = kmalloc(sizeof(*s), GFP_KERNEL); |
574 | if (s == NULL) { | |
575 | ti->error = "Cannot allocate snapshot context private " | |
576 | "structure"; | |
577 | r = -ENOMEM; | |
578 | goto bad1; | |
579 | } | |
580 | ||
581 | r = dm_get_device(ti, origin_path, 0, ti->len, FMODE_READ, &s->origin); | |
582 | if (r) { | |
583 | ti->error = "Cannot get origin device"; | |
584 | goto bad2; | |
585 | } | |
586 | ||
587 | r = dm_get_device(ti, cow_path, 0, 0, | |
588 | FMODE_READ | FMODE_WRITE, &s->cow); | |
589 | if (r) { | |
590 | dm_put_device(ti, s->origin); | |
591 | ti->error = "Cannot get COW device"; | |
592 | goto bad2; | |
593 | } | |
594 | ||
4c7e3bf4 MM |
595 | r = set_chunk_size(s, argv[3], &ti->error); |
596 | if (r) | |
1da177e4 | 597 | goto bad3; |
1da177e4 | 598 | |
1da177e4 | 599 | s->type = persistent; |
1da177e4 LT |
600 | |
601 | s->valid = 1; | |
aa14edeb | 602 | s->active = 0; |
1da177e4 | 603 | init_rwsem(&s->lock); |
ca3a931f | 604 | spin_lock_init(&s->pe_lock); |
72727bad | 605 | s->ti = ti; |
1da177e4 LT |
606 | |
607 | /* Allocate hash table for COW data */ | |
608 | if (init_hash_tables(s)) { | |
609 | ti->error = "Unable to allocate hash table space"; | |
610 | r = -ENOMEM; | |
611 | goto bad3; | |
612 | } | |
613 | ||
1da177e4 LT |
614 | s->store.snap = s; |
615 | ||
616 | if (persistent == 'P') | |
4c7e3bf4 | 617 | r = dm_create_persistent(&s->store); |
1da177e4 | 618 | else |
4c7e3bf4 | 619 | r = dm_create_transient(&s->store); |
1da177e4 LT |
620 | |
621 | if (r) { | |
622 | ti->error = "Couldn't create exception store"; | |
623 | r = -EINVAL; | |
624 | goto bad4; | |
625 | } | |
626 | ||
eb69aca5 | 627 | r = dm_kcopyd_client_create(SNAPSHOT_PAGES, &s->kcopyd_client); |
1da177e4 LT |
628 | if (r) { |
629 | ti->error = "Could not create kcopyd client"; | |
630 | goto bad5; | |
631 | } | |
632 | ||
92e86812 MP |
633 | s->pending_pool = mempool_create_slab_pool(MIN_IOS, pending_cache); |
634 | if (!s->pending_pool) { | |
635 | ti->error = "Could not allocate mempool for pending exceptions"; | |
636 | goto bad6; | |
637 | } | |
638 | ||
cd45daff MP |
639 | s->tracked_chunk_pool = mempool_create_slab_pool(MIN_IOS, |
640 | tracked_chunk_cache); | |
641 | if (!s->tracked_chunk_pool) { | |
642 | ti->error = "Could not allocate tracked_chunk mempool for " | |
643 | "tracking reads"; | |
92e86812 | 644 | goto bad_tracked_chunk_pool; |
cd45daff MP |
645 | } |
646 | ||
647 | for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++) | |
648 | INIT_HLIST_HEAD(&s->tracked_chunk_hash[i]); | |
649 | ||
650 | spin_lock_init(&s->tracked_chunk_lock); | |
651 | ||
aa14edeb | 652 | /* Metadata must only be loaded into one table at once */ |
f9cea4f7 | 653 | r = s->store.read_metadata(&s->store); |
0764147b | 654 | if (r < 0) { |
f9cea4f7 | 655 | ti->error = "Failed to read snapshot metadata"; |
cd45daff | 656 | goto bad_load_and_register; |
0764147b MB |
657 | } else if (r > 0) { |
658 | s->valid = 0; | |
659 | DMWARN("Snapshot is marked invalid."); | |
f9cea4f7 | 660 | } |
aa14edeb | 661 | |
ca3a931f | 662 | bio_list_init(&s->queued_bios); |
c4028958 | 663 | INIT_WORK(&s->queued_bios_work, flush_queued_bios); |
ca3a931f | 664 | |
1da177e4 | 665 | /* Add snapshot to the list of snapshots for this origin */ |
aa14edeb | 666 | /* Exceptions aren't triggered till snapshot_resume() is called */ |
1da177e4 LT |
667 | if (register_snapshot(s)) { |
668 | r = -EINVAL; | |
669 | ti->error = "Cannot register snapshot origin"; | |
cd45daff | 670 | goto bad_load_and_register; |
1da177e4 LT |
671 | } |
672 | ||
673 | ti->private = s; | |
c51c2752 | 674 | ti->split_io = s->chunk_size; |
1da177e4 LT |
675 | |
676 | return 0; | |
677 | ||
cd45daff MP |
678 | bad_load_and_register: |
679 | mempool_destroy(s->tracked_chunk_pool); | |
680 | ||
92e86812 MP |
681 | bad_tracked_chunk_pool: |
682 | mempool_destroy(s->pending_pool); | |
683 | ||
1da177e4 | 684 | bad6: |
eb69aca5 | 685 | dm_kcopyd_client_destroy(s->kcopyd_client); |
1da177e4 LT |
686 | |
687 | bad5: | |
688 | s->store.destroy(&s->store); | |
689 | ||
690 | bad4: | |
691 | exit_exception_table(&s->pending, pending_cache); | |
692 | exit_exception_table(&s->complete, exception_cache); | |
693 | ||
694 | bad3: | |
695 | dm_put_device(ti, s->cow); | |
696 | dm_put_device(ti, s->origin); | |
697 | ||
698 | bad2: | |
699 | kfree(s); | |
700 | ||
701 | bad1: | |
702 | return r; | |
703 | } | |
704 | ||
31c93a0c MB |
705 | static void __free_exceptions(struct dm_snapshot *s) |
706 | { | |
eb69aca5 | 707 | dm_kcopyd_client_destroy(s->kcopyd_client); |
31c93a0c MB |
708 | s->kcopyd_client = NULL; |
709 | ||
710 | exit_exception_table(&s->pending, pending_cache); | |
711 | exit_exception_table(&s->complete, exception_cache); | |
712 | ||
713 | s->store.destroy(&s->store); | |
714 | } | |
715 | ||
1da177e4 LT |
716 | static void snapshot_dtr(struct dm_target *ti) |
717 | { | |
cd45daff MP |
718 | #ifdef CONFIG_DM_DEBUG |
719 | int i; | |
720 | #endif | |
028867ac | 721 | struct dm_snapshot *s = ti->private; |
1da177e4 | 722 | |
ca3a931f AK |
723 | flush_workqueue(ksnapd); |
724 | ||
138728dc AK |
725 | /* Prevent further origin writes from using this snapshot. */ |
726 | /* After this returns there can be no new kcopyd jobs. */ | |
1da177e4 LT |
727 | unregister_snapshot(s); |
728 | ||
cd45daff MP |
729 | #ifdef CONFIG_DM_DEBUG |
730 | for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++) | |
731 | BUG_ON(!hlist_empty(&s->tracked_chunk_hash[i])); | |
732 | #endif | |
733 | ||
734 | mempool_destroy(s->tracked_chunk_pool); | |
735 | ||
31c93a0c | 736 | __free_exceptions(s); |
1da177e4 | 737 | |
92e86812 MP |
738 | mempool_destroy(s->pending_pool); |
739 | ||
1da177e4 LT |
740 | dm_put_device(ti, s->origin); |
741 | dm_put_device(ti, s->cow); | |
138728dc | 742 | |
1da177e4 LT |
743 | kfree(s); |
744 | } | |
745 | ||
746 | /* | |
747 | * Flush a list of buffers. | |
748 | */ | |
749 | static void flush_bios(struct bio *bio) | |
750 | { | |
751 | struct bio *n; | |
752 | ||
753 | while (bio) { | |
754 | n = bio->bi_next; | |
755 | bio->bi_next = NULL; | |
756 | generic_make_request(bio); | |
757 | bio = n; | |
758 | } | |
759 | } | |
760 | ||
c4028958 | 761 | static void flush_queued_bios(struct work_struct *work) |
ca3a931f | 762 | { |
c4028958 DH |
763 | struct dm_snapshot *s = |
764 | container_of(work, struct dm_snapshot, queued_bios_work); | |
ca3a931f AK |
765 | struct bio *queued_bios; |
766 | unsigned long flags; | |
767 | ||
768 | spin_lock_irqsave(&s->pe_lock, flags); | |
769 | queued_bios = bio_list_get(&s->queued_bios); | |
770 | spin_unlock_irqrestore(&s->pe_lock, flags); | |
771 | ||
772 | flush_bios(queued_bios); | |
773 | } | |
774 | ||
1da177e4 LT |
775 | /* |
776 | * Error a list of buffers. | |
777 | */ | |
778 | static void error_bios(struct bio *bio) | |
779 | { | |
780 | struct bio *n; | |
781 | ||
782 | while (bio) { | |
783 | n = bio->bi_next; | |
784 | bio->bi_next = NULL; | |
6712ecf8 | 785 | bio_io_error(bio); |
1da177e4 LT |
786 | bio = n; |
787 | } | |
788 | } | |
789 | ||
695368ac | 790 | static void __invalidate_snapshot(struct dm_snapshot *s, int err) |
76df1c65 AK |
791 | { |
792 | if (!s->valid) | |
793 | return; | |
794 | ||
795 | if (err == -EIO) | |
796 | DMERR("Invalidating snapshot: Error reading/writing."); | |
797 | else if (err == -ENOMEM) | |
798 | DMERR("Invalidating snapshot: Unable to allocate exception."); | |
799 | ||
76df1c65 AK |
800 | if (s->store.drop_snapshot) |
801 | s->store.drop_snapshot(&s->store); | |
802 | ||
803 | s->valid = 0; | |
804 | ||
72727bad | 805 | dm_table_event(s->ti->table); |
76df1c65 AK |
806 | } |
807 | ||
028867ac | 808 | static void get_pending_exception(struct dm_snap_pending_exception *pe) |
4b832e8d AK |
809 | { |
810 | atomic_inc(&pe->ref_count); | |
811 | } | |
812 | ||
028867ac | 813 | static struct bio *put_pending_exception(struct dm_snap_pending_exception *pe) |
4b832e8d | 814 | { |
028867ac | 815 | struct dm_snap_pending_exception *primary_pe; |
4b832e8d AK |
816 | struct bio *origin_bios = NULL; |
817 | ||
818 | primary_pe = pe->primary_pe; | |
819 | ||
820 | /* | |
821 | * If this pe is involved in a write to the origin and | |
822 | * it is the last sibling to complete then release | |
823 | * the bios for the original write to the origin. | |
824 | */ | |
825 | if (primary_pe && | |
7c5f78b9 | 826 | atomic_dec_and_test(&primary_pe->ref_count)) { |
4b832e8d | 827 | origin_bios = bio_list_get(&primary_pe->origin_bios); |
7c5f78b9 MP |
828 | free_pending_exception(primary_pe); |
829 | } | |
4b832e8d AK |
830 | |
831 | /* | |
832 | * Free the pe if it's not linked to an origin write or if | |
833 | * it's not itself a primary pe. | |
834 | */ | |
835 | if (!primary_pe || primary_pe != pe) | |
836 | free_pending_exception(pe); | |
837 | ||
4b832e8d AK |
838 | return origin_bios; |
839 | } | |
840 | ||
028867ac | 841 | static void pending_complete(struct dm_snap_pending_exception *pe, int success) |
1da177e4 | 842 | { |
028867ac | 843 | struct dm_snap_exception *e; |
1da177e4 | 844 | struct dm_snapshot *s = pe->snap; |
9d493fa8 AK |
845 | struct bio *origin_bios = NULL; |
846 | struct bio *snapshot_bios = NULL; | |
847 | int error = 0; | |
1da177e4 | 848 | |
76df1c65 AK |
849 | if (!success) { |
850 | /* Read/write error - snapshot is unusable */ | |
1da177e4 | 851 | down_write(&s->lock); |
695368ac | 852 | __invalidate_snapshot(s, -EIO); |
9d493fa8 | 853 | error = 1; |
76df1c65 AK |
854 | goto out; |
855 | } | |
856 | ||
857 | e = alloc_exception(); | |
858 | if (!e) { | |
1da177e4 | 859 | down_write(&s->lock); |
695368ac | 860 | __invalidate_snapshot(s, -ENOMEM); |
9d493fa8 | 861 | error = 1; |
76df1c65 AK |
862 | goto out; |
863 | } | |
864 | *e = pe->e; | |
1da177e4 | 865 | |
76df1c65 AK |
866 | down_write(&s->lock); |
867 | if (!s->valid) { | |
76df1c65 | 868 | free_exception(e); |
9d493fa8 | 869 | error = 1; |
76df1c65 | 870 | goto out; |
1da177e4 LT |
871 | } |
872 | ||
a8d41b59 MP |
873 | /* |
874 | * Check for conflicting reads. This is extremely improbable, | |
875 | * so yield() is sufficient and there is no need for a wait queue. | |
876 | */ | |
877 | while (__chunk_is_tracked(s, pe->e.old_chunk)) | |
878 | yield(); | |
879 | ||
9d493fa8 AK |
880 | /* |
881 | * Add a proper exception, and remove the | |
882 | * in-flight exception from the list. | |
883 | */ | |
d74f81f8 | 884 | insert_completed_exception(s, e); |
76df1c65 | 885 | |
1da177e4 | 886 | out: |
695368ac | 887 | remove_exception(&pe->e); |
9d493fa8 | 888 | snapshot_bios = bio_list_get(&pe->snapshot_bios); |
4b832e8d | 889 | origin_bios = put_pending_exception(pe); |
1da177e4 | 890 | |
9d493fa8 AK |
891 | up_write(&s->lock); |
892 | ||
893 | /* Submit any pending write bios */ | |
894 | if (error) | |
895 | error_bios(snapshot_bios); | |
896 | else | |
897 | flush_bios(snapshot_bios); | |
898 | ||
899 | flush_bios(origin_bios); | |
1da177e4 LT |
900 | } |
901 | ||
902 | static void commit_callback(void *context, int success) | |
903 | { | |
028867ac AK |
904 | struct dm_snap_pending_exception *pe = context; |
905 | ||
1da177e4 LT |
906 | pending_complete(pe, success); |
907 | } | |
908 | ||
909 | /* | |
910 | * Called when the copy I/O has finished. kcopyd actually runs | |
911 | * this code so don't block. | |
912 | */ | |
4cdc1d1f | 913 | static void copy_callback(int read_err, unsigned long write_err, void *context) |
1da177e4 | 914 | { |
028867ac | 915 | struct dm_snap_pending_exception *pe = context; |
1da177e4 LT |
916 | struct dm_snapshot *s = pe->snap; |
917 | ||
918 | if (read_err || write_err) | |
919 | pending_complete(pe, 0); | |
920 | ||
921 | else | |
922 | /* Update the metadata if we are persistent */ | |
923 | s->store.commit_exception(&s->store, &pe->e, commit_callback, | |
924 | pe); | |
925 | } | |
926 | ||
927 | /* | |
928 | * Dispatches the copy operation to kcopyd. | |
929 | */ | |
028867ac | 930 | static void start_copy(struct dm_snap_pending_exception *pe) |
1da177e4 LT |
931 | { |
932 | struct dm_snapshot *s = pe->snap; | |
22a1ceb1 | 933 | struct dm_io_region src, dest; |
1da177e4 LT |
934 | struct block_device *bdev = s->origin->bdev; |
935 | sector_t dev_size; | |
936 | ||
937 | dev_size = get_dev_size(bdev); | |
938 | ||
939 | src.bdev = bdev; | |
940 | src.sector = chunk_to_sector(s, pe->e.old_chunk); | |
941 | src.count = min(s->chunk_size, dev_size - src.sector); | |
942 | ||
943 | dest.bdev = s->cow->bdev; | |
944 | dest.sector = chunk_to_sector(s, pe->e.new_chunk); | |
945 | dest.count = src.count; | |
946 | ||
947 | /* Hand over to kcopyd */ | |
eb69aca5 | 948 | dm_kcopyd_copy(s->kcopyd_client, |
1da177e4 LT |
949 | &src, 1, &dest, 0, copy_callback, pe); |
950 | } | |
951 | ||
952 | /* | |
953 | * Looks to see if this snapshot already has a pending exception | |
954 | * for this chunk, otherwise it allocates a new one and inserts | |
955 | * it into the pending table. | |
956 | * | |
957 | * NOTE: a write lock must be held on snap->lock before calling | |
958 | * this. | |
959 | */ | |
028867ac | 960 | static struct dm_snap_pending_exception * |
1da177e4 LT |
961 | __find_pending_exception(struct dm_snapshot *s, struct bio *bio) |
962 | { | |
028867ac AK |
963 | struct dm_snap_exception *e; |
964 | struct dm_snap_pending_exception *pe; | |
1da177e4 LT |
965 | chunk_t chunk = sector_to_chunk(s, bio->bi_sector); |
966 | ||
967 | /* | |
968 | * Is there a pending exception for this already ? | |
969 | */ | |
970 | e = lookup_exception(&s->pending, chunk); | |
971 | if (e) { | |
972 | /* cast the exception to a pending exception */ | |
028867ac | 973 | pe = container_of(e, struct dm_snap_pending_exception, e); |
76df1c65 AK |
974 | goto out; |
975 | } | |
1da177e4 | 976 | |
76df1c65 AK |
977 | /* |
978 | * Create a new pending exception, we don't want | |
979 | * to hold the lock while we do this. | |
980 | */ | |
981 | up_write(&s->lock); | |
92e86812 | 982 | pe = alloc_pending_exception(s); |
76df1c65 | 983 | down_write(&s->lock); |
1da177e4 | 984 | |
76df1c65 AK |
985 | if (!s->valid) { |
986 | free_pending_exception(pe); | |
987 | return NULL; | |
988 | } | |
1da177e4 | 989 | |
76df1c65 AK |
990 | e = lookup_exception(&s->pending, chunk); |
991 | if (e) { | |
992 | free_pending_exception(pe); | |
028867ac | 993 | pe = container_of(e, struct dm_snap_pending_exception, e); |
76df1c65 | 994 | goto out; |
1da177e4 LT |
995 | } |
996 | ||
76df1c65 AK |
997 | pe->e.old_chunk = chunk; |
998 | bio_list_init(&pe->origin_bios); | |
999 | bio_list_init(&pe->snapshot_bios); | |
1000 | pe->primary_pe = NULL; | |
4b832e8d | 1001 | atomic_set(&pe->ref_count, 0); |
76df1c65 AK |
1002 | pe->started = 0; |
1003 | ||
1004 | if (s->store.prepare_exception(&s->store, &pe->e)) { | |
1005 | free_pending_exception(pe); | |
1006 | return NULL; | |
1007 | } | |
1008 | ||
4b832e8d | 1009 | get_pending_exception(pe); |
76df1c65 AK |
1010 | insert_exception(&s->pending, &pe->e); |
1011 | ||
1012 | out: | |
1da177e4 LT |
1013 | return pe; |
1014 | } | |
1015 | ||
028867ac | 1016 | static void remap_exception(struct dm_snapshot *s, struct dm_snap_exception *e, |
d74f81f8 | 1017 | struct bio *bio, chunk_t chunk) |
1da177e4 LT |
1018 | { |
1019 | bio->bi_bdev = s->cow->bdev; | |
d74f81f8 MB |
1020 | bio->bi_sector = chunk_to_sector(s, dm_chunk_number(e->new_chunk) + |
1021 | (chunk - e->old_chunk)) + | |
1022 | (bio->bi_sector & s->chunk_mask); | |
1da177e4 LT |
1023 | } |
1024 | ||
1025 | static int snapshot_map(struct dm_target *ti, struct bio *bio, | |
1026 | union map_info *map_context) | |
1027 | { | |
028867ac AK |
1028 | struct dm_snap_exception *e; |
1029 | struct dm_snapshot *s = ti->private; | |
d2a7ad29 | 1030 | int r = DM_MAPIO_REMAPPED; |
1da177e4 | 1031 | chunk_t chunk; |
028867ac | 1032 | struct dm_snap_pending_exception *pe = NULL; |
1da177e4 LT |
1033 | |
1034 | chunk = sector_to_chunk(s, bio->bi_sector); | |
1035 | ||
1036 | /* Full snapshots are not usable */ | |
76df1c65 | 1037 | /* To get here the table must be live so s->active is always set. */ |
1da177e4 | 1038 | if (!s->valid) |
f6a80ea8 | 1039 | return -EIO; |
1da177e4 | 1040 | |
ba40a2aa AK |
1041 | /* FIXME: should only take write lock if we need |
1042 | * to copy an exception */ | |
1043 | down_write(&s->lock); | |
1044 | ||
1045 | if (!s->valid) { | |
1046 | r = -EIO; | |
1047 | goto out_unlock; | |
1048 | } | |
1049 | ||
1050 | /* If the block is already remapped - use that, else remap it */ | |
1051 | e = lookup_exception(&s->complete, chunk); | |
1052 | if (e) { | |
d74f81f8 | 1053 | remap_exception(s, e, bio, chunk); |
ba40a2aa AK |
1054 | goto out_unlock; |
1055 | } | |
1056 | ||
1da177e4 LT |
1057 | /* |
1058 | * Write to snapshot - higher level takes care of RW/RO | |
1059 | * flags so we should only get this if we are | |
1060 | * writeable. | |
1061 | */ | |
1062 | if (bio_rw(bio) == WRITE) { | |
76df1c65 AK |
1063 | pe = __find_pending_exception(s, bio); |
1064 | if (!pe) { | |
695368ac | 1065 | __invalidate_snapshot(s, -ENOMEM); |
76df1c65 AK |
1066 | r = -EIO; |
1067 | goto out_unlock; | |
1da177e4 LT |
1068 | } |
1069 | ||
d74f81f8 | 1070 | remap_exception(s, &pe->e, bio, chunk); |
76df1c65 AK |
1071 | bio_list_add(&pe->snapshot_bios, bio); |
1072 | ||
d2a7ad29 | 1073 | r = DM_MAPIO_SUBMITTED; |
ba40a2aa | 1074 | |
76df1c65 AK |
1075 | if (!pe->started) { |
1076 | /* this is protected by snap->lock */ | |
1077 | pe->started = 1; | |
ba40a2aa | 1078 | up_write(&s->lock); |
76df1c65 | 1079 | start_copy(pe); |
ba40a2aa AK |
1080 | goto out; |
1081 | } | |
cd45daff | 1082 | } else { |
ba40a2aa | 1083 | bio->bi_bdev = s->origin->bdev; |
cd45daff MP |
1084 | map_context->ptr = track_chunk(s, chunk); |
1085 | } | |
1da177e4 | 1086 | |
ba40a2aa AK |
1087 | out_unlock: |
1088 | up_write(&s->lock); | |
1089 | out: | |
1da177e4 LT |
1090 | return r; |
1091 | } | |
1092 | ||
cd45daff MP |
1093 | static int snapshot_end_io(struct dm_target *ti, struct bio *bio, |
1094 | int error, union map_info *map_context) | |
1095 | { | |
1096 | struct dm_snapshot *s = ti->private; | |
1097 | struct dm_snap_tracked_chunk *c = map_context->ptr; | |
1098 | ||
1099 | if (c) | |
1100 | stop_tracking_chunk(s, c); | |
1101 | ||
1102 | return 0; | |
1103 | } | |
1104 | ||
1da177e4 LT |
1105 | static void snapshot_resume(struct dm_target *ti) |
1106 | { | |
028867ac | 1107 | struct dm_snapshot *s = ti->private; |
1da177e4 | 1108 | |
aa14edeb AK |
1109 | down_write(&s->lock); |
1110 | s->active = 1; | |
1111 | up_write(&s->lock); | |
1da177e4 LT |
1112 | } |
1113 | ||
1114 | static int snapshot_status(struct dm_target *ti, status_type_t type, | |
1115 | char *result, unsigned int maxlen) | |
1116 | { | |
028867ac | 1117 | struct dm_snapshot *snap = ti->private; |
1da177e4 LT |
1118 | |
1119 | switch (type) { | |
1120 | case STATUSTYPE_INFO: | |
1121 | if (!snap->valid) | |
1122 | snprintf(result, maxlen, "Invalid"); | |
1123 | else { | |
1124 | if (snap->store.fraction_full) { | |
1125 | sector_t numerator, denominator; | |
1126 | snap->store.fraction_full(&snap->store, | |
1127 | &numerator, | |
1128 | &denominator); | |
4ee218cd AM |
1129 | snprintf(result, maxlen, "%llu/%llu", |
1130 | (unsigned long long)numerator, | |
1131 | (unsigned long long)denominator); | |
1da177e4 LT |
1132 | } |
1133 | else | |
1134 | snprintf(result, maxlen, "Unknown"); | |
1135 | } | |
1136 | break; | |
1137 | ||
1138 | case STATUSTYPE_TABLE: | |
1139 | /* | |
1140 | * kdevname returns a static pointer so we need | |
1141 | * to make private copies if the output is to | |
1142 | * make sense. | |
1143 | */ | |
4ee218cd | 1144 | snprintf(result, maxlen, "%s %s %c %llu", |
1da177e4 | 1145 | snap->origin->name, snap->cow->name, |
4ee218cd AM |
1146 | snap->type, |
1147 | (unsigned long long)snap->chunk_size); | |
1da177e4 LT |
1148 | break; |
1149 | } | |
1150 | ||
1151 | return 0; | |
1152 | } | |
1153 | ||
1154 | /*----------------------------------------------------------------- | |
1155 | * Origin methods | |
1156 | *---------------------------------------------------------------*/ | |
1da177e4 LT |
1157 | static int __origin_write(struct list_head *snapshots, struct bio *bio) |
1158 | { | |
d2a7ad29 | 1159 | int r = DM_MAPIO_REMAPPED, first = 0; |
1da177e4 | 1160 | struct dm_snapshot *snap; |
028867ac AK |
1161 | struct dm_snap_exception *e; |
1162 | struct dm_snap_pending_exception *pe, *next_pe, *primary_pe = NULL; | |
1da177e4 | 1163 | chunk_t chunk; |
eccf0817 | 1164 | LIST_HEAD(pe_queue); |
1da177e4 LT |
1165 | |
1166 | /* Do all the snapshots on this origin */ | |
1167 | list_for_each_entry (snap, snapshots, list) { | |
1168 | ||
76df1c65 AK |
1169 | down_write(&snap->lock); |
1170 | ||
aa14edeb AK |
1171 | /* Only deal with valid and active snapshots */ |
1172 | if (!snap->valid || !snap->active) | |
76df1c65 | 1173 | goto next_snapshot; |
1da177e4 | 1174 | |
d5e404c1 | 1175 | /* Nothing to do if writing beyond end of snapshot */ |
72727bad | 1176 | if (bio->bi_sector >= dm_table_get_size(snap->ti->table)) |
76df1c65 | 1177 | goto next_snapshot; |
1da177e4 LT |
1178 | |
1179 | /* | |
1180 | * Remember, different snapshots can have | |
1181 | * different chunk sizes. | |
1182 | */ | |
1183 | chunk = sector_to_chunk(snap, bio->bi_sector); | |
1184 | ||
1185 | /* | |
1186 | * Check exception table to see if block | |
1187 | * is already remapped in this snapshot | |
1188 | * and trigger an exception if not. | |
b4b610f6 | 1189 | * |
4b832e8d | 1190 | * ref_count is initialised to 1 so pending_complete() |
b4b610f6 | 1191 | * won't destroy the primary_pe while we're inside this loop. |
1da177e4 LT |
1192 | */ |
1193 | e = lookup_exception(&snap->complete, chunk); | |
76df1c65 AK |
1194 | if (e) |
1195 | goto next_snapshot; | |
1196 | ||
1197 | pe = __find_pending_exception(snap, bio); | |
1198 | if (!pe) { | |
695368ac | 1199 | __invalidate_snapshot(snap, -ENOMEM); |
76df1c65 AK |
1200 | goto next_snapshot; |
1201 | } | |
1202 | ||
1203 | if (!primary_pe) { | |
1204 | /* | |
1205 | * Either every pe here has same | |
1206 | * primary_pe or none has one yet. | |
1207 | */ | |
1208 | if (pe->primary_pe) | |
1209 | primary_pe = pe->primary_pe; | |
1210 | else { | |
1211 | primary_pe = pe; | |
1212 | first = 1; | |
1da177e4 | 1213 | } |
76df1c65 AK |
1214 | |
1215 | bio_list_add(&primary_pe->origin_bios, bio); | |
1216 | ||
d2a7ad29 | 1217 | r = DM_MAPIO_SUBMITTED; |
76df1c65 AK |
1218 | } |
1219 | ||
1220 | if (!pe->primary_pe) { | |
76df1c65 | 1221 | pe->primary_pe = primary_pe; |
4b832e8d | 1222 | get_pending_exception(primary_pe); |
76df1c65 AK |
1223 | } |
1224 | ||
1225 | if (!pe->started) { | |
1226 | pe->started = 1; | |
1227 | list_add_tail(&pe->list, &pe_queue); | |
1da177e4 LT |
1228 | } |
1229 | ||
76df1c65 | 1230 | next_snapshot: |
1da177e4 LT |
1231 | up_write(&snap->lock); |
1232 | } | |
1233 | ||
b4b610f6 | 1234 | if (!primary_pe) |
4b832e8d | 1235 | return r; |
b4b610f6 AK |
1236 | |
1237 | /* | |
1238 | * If this is the first time we're processing this chunk and | |
4b832e8d | 1239 | * ref_count is now 1 it means all the pending exceptions |
b4b610f6 AK |
1240 | * got completed while we were in the loop above, so it falls to |
1241 | * us here to remove the primary_pe and submit any origin_bios. | |
1242 | */ | |
1243 | ||
4b832e8d | 1244 | if (first && atomic_dec_and_test(&primary_pe->ref_count)) { |
b4b610f6 AK |
1245 | flush_bios(bio_list_get(&primary_pe->origin_bios)); |
1246 | free_pending_exception(primary_pe); | |
1247 | /* If we got here, pe_queue is necessarily empty. */ | |
4b832e8d | 1248 | return r; |
b4b610f6 AK |
1249 | } |
1250 | ||
1da177e4 LT |
1251 | /* |
1252 | * Now that we have a complete pe list we can start the copying. | |
1253 | */ | |
eccf0817 AK |
1254 | list_for_each_entry_safe(pe, next_pe, &pe_queue, list) |
1255 | start_copy(pe); | |
1da177e4 LT |
1256 | |
1257 | return r; | |
1258 | } | |
1259 | ||
1260 | /* | |
1261 | * Called on a write from the origin driver. | |
1262 | */ | |
1263 | static int do_origin(struct dm_dev *origin, struct bio *bio) | |
1264 | { | |
1265 | struct origin *o; | |
d2a7ad29 | 1266 | int r = DM_MAPIO_REMAPPED; |
1da177e4 LT |
1267 | |
1268 | down_read(&_origins_lock); | |
1269 | o = __lookup_origin(origin->bdev); | |
1270 | if (o) | |
1271 | r = __origin_write(&o->snapshots, bio); | |
1272 | up_read(&_origins_lock); | |
1273 | ||
1274 | return r; | |
1275 | } | |
1276 | ||
1277 | /* | |
1278 | * Origin: maps a linear range of a device, with hooks for snapshotting. | |
1279 | */ | |
1280 | ||
1281 | /* | |
1282 | * Construct an origin mapping: <dev_path> | |
1283 | * The context for an origin is merely a 'struct dm_dev *' | |
1284 | * pointing to the real device. | |
1285 | */ | |
1286 | static int origin_ctr(struct dm_target *ti, unsigned int argc, char **argv) | |
1287 | { | |
1288 | int r; | |
1289 | struct dm_dev *dev; | |
1290 | ||
1291 | if (argc != 1) { | |
72d94861 | 1292 | ti->error = "origin: incorrect number of arguments"; |
1da177e4 LT |
1293 | return -EINVAL; |
1294 | } | |
1295 | ||
1296 | r = dm_get_device(ti, argv[0], 0, ti->len, | |
1297 | dm_table_get_mode(ti->table), &dev); | |
1298 | if (r) { | |
1299 | ti->error = "Cannot get target device"; | |
1300 | return r; | |
1301 | } | |
1302 | ||
1303 | ti->private = dev; | |
1304 | return 0; | |
1305 | } | |
1306 | ||
1307 | static void origin_dtr(struct dm_target *ti) | |
1308 | { | |
028867ac | 1309 | struct dm_dev *dev = ti->private; |
1da177e4 LT |
1310 | dm_put_device(ti, dev); |
1311 | } | |
1312 | ||
1313 | static int origin_map(struct dm_target *ti, struct bio *bio, | |
1314 | union map_info *map_context) | |
1315 | { | |
028867ac | 1316 | struct dm_dev *dev = ti->private; |
1da177e4 LT |
1317 | bio->bi_bdev = dev->bdev; |
1318 | ||
1319 | /* Only tell snapshots if this is a write */ | |
d2a7ad29 | 1320 | return (bio_rw(bio) == WRITE) ? do_origin(dev, bio) : DM_MAPIO_REMAPPED; |
1da177e4 LT |
1321 | } |
1322 | ||
1323 | #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r)) | |
1324 | ||
1325 | /* | |
1326 | * Set the target "split_io" field to the minimum of all the snapshots' | |
1327 | * chunk sizes. | |
1328 | */ | |
1329 | static void origin_resume(struct dm_target *ti) | |
1330 | { | |
028867ac | 1331 | struct dm_dev *dev = ti->private; |
1da177e4 LT |
1332 | struct dm_snapshot *snap; |
1333 | struct origin *o; | |
1334 | chunk_t chunk_size = 0; | |
1335 | ||
1336 | down_read(&_origins_lock); | |
1337 | o = __lookup_origin(dev->bdev); | |
1338 | if (o) | |
1339 | list_for_each_entry (snap, &o->snapshots, list) | |
1340 | chunk_size = min_not_zero(chunk_size, snap->chunk_size); | |
1341 | up_read(&_origins_lock); | |
1342 | ||
1343 | ti->split_io = chunk_size; | |
1344 | } | |
1345 | ||
1346 | static int origin_status(struct dm_target *ti, status_type_t type, char *result, | |
1347 | unsigned int maxlen) | |
1348 | { | |
028867ac | 1349 | struct dm_dev *dev = ti->private; |
1da177e4 LT |
1350 | |
1351 | switch (type) { | |
1352 | case STATUSTYPE_INFO: | |
1353 | result[0] = '\0'; | |
1354 | break; | |
1355 | ||
1356 | case STATUSTYPE_TABLE: | |
1357 | snprintf(result, maxlen, "%s", dev->name); | |
1358 | break; | |
1359 | } | |
1360 | ||
1361 | return 0; | |
1362 | } | |
1363 | ||
1364 | static struct target_type origin_target = { | |
1365 | .name = "snapshot-origin", | |
d74f81f8 | 1366 | .version = {1, 6, 0}, |
1da177e4 LT |
1367 | .module = THIS_MODULE, |
1368 | .ctr = origin_ctr, | |
1369 | .dtr = origin_dtr, | |
1370 | .map = origin_map, | |
1371 | .resume = origin_resume, | |
1372 | .status = origin_status, | |
1373 | }; | |
1374 | ||
1375 | static struct target_type snapshot_target = { | |
1376 | .name = "snapshot", | |
d74f81f8 | 1377 | .version = {1, 6, 0}, |
1da177e4 LT |
1378 | .module = THIS_MODULE, |
1379 | .ctr = snapshot_ctr, | |
1380 | .dtr = snapshot_dtr, | |
1381 | .map = snapshot_map, | |
cd45daff | 1382 | .end_io = snapshot_end_io, |
1da177e4 LT |
1383 | .resume = snapshot_resume, |
1384 | .status = snapshot_status, | |
1385 | }; | |
1386 | ||
1387 | static int __init dm_snapshot_init(void) | |
1388 | { | |
1389 | int r; | |
1390 | ||
1391 | r = dm_register_target(&snapshot_target); | |
1392 | if (r) { | |
1393 | DMERR("snapshot target register failed %d", r); | |
1394 | return r; | |
1395 | } | |
1396 | ||
1397 | r = dm_register_target(&origin_target); | |
1398 | if (r < 0) { | |
72d94861 | 1399 | DMERR("Origin target register failed %d", r); |
1da177e4 LT |
1400 | goto bad1; |
1401 | } | |
1402 | ||
1403 | r = init_origin_hash(); | |
1404 | if (r) { | |
1405 | DMERR("init_origin_hash failed."); | |
1406 | goto bad2; | |
1407 | } | |
1408 | ||
028867ac | 1409 | exception_cache = KMEM_CACHE(dm_snap_exception, 0); |
1da177e4 LT |
1410 | if (!exception_cache) { |
1411 | DMERR("Couldn't create exception cache."); | |
1412 | r = -ENOMEM; | |
1413 | goto bad3; | |
1414 | } | |
1415 | ||
028867ac | 1416 | pending_cache = KMEM_CACHE(dm_snap_pending_exception, 0); |
1da177e4 LT |
1417 | if (!pending_cache) { |
1418 | DMERR("Couldn't create pending cache."); | |
1419 | r = -ENOMEM; | |
1420 | goto bad4; | |
1421 | } | |
1422 | ||
cd45daff MP |
1423 | tracked_chunk_cache = KMEM_CACHE(dm_snap_tracked_chunk, 0); |
1424 | if (!tracked_chunk_cache) { | |
1425 | DMERR("Couldn't create cache to track chunks in use."); | |
1426 | r = -ENOMEM; | |
1427 | goto bad5; | |
1428 | } | |
1429 | ||
ca3a931f AK |
1430 | ksnapd = create_singlethread_workqueue("ksnapd"); |
1431 | if (!ksnapd) { | |
1432 | DMERR("Failed to create ksnapd workqueue."); | |
1433 | r = -ENOMEM; | |
92e86812 | 1434 | goto bad_pending_pool; |
ca3a931f AK |
1435 | } |
1436 | ||
1da177e4 LT |
1437 | return 0; |
1438 | ||
cd45daff MP |
1439 | bad_pending_pool: |
1440 | kmem_cache_destroy(tracked_chunk_cache); | |
1da177e4 LT |
1441 | bad5: |
1442 | kmem_cache_destroy(pending_cache); | |
1443 | bad4: | |
1444 | kmem_cache_destroy(exception_cache); | |
1445 | bad3: | |
1446 | exit_origin_hash(); | |
1447 | bad2: | |
1448 | dm_unregister_target(&origin_target); | |
1449 | bad1: | |
1450 | dm_unregister_target(&snapshot_target); | |
1451 | return r; | |
1452 | } | |
1453 | ||
1454 | static void __exit dm_snapshot_exit(void) | |
1455 | { | |
1456 | int r; | |
1457 | ||
ca3a931f AK |
1458 | destroy_workqueue(ksnapd); |
1459 | ||
1da177e4 LT |
1460 | r = dm_unregister_target(&snapshot_target); |
1461 | if (r) | |
1462 | DMERR("snapshot unregister failed %d", r); | |
1463 | ||
1464 | r = dm_unregister_target(&origin_target); | |
1465 | if (r) | |
1466 | DMERR("origin unregister failed %d", r); | |
1467 | ||
1468 | exit_origin_hash(); | |
1da177e4 LT |
1469 | kmem_cache_destroy(pending_cache); |
1470 | kmem_cache_destroy(exception_cache); | |
cd45daff | 1471 | kmem_cache_destroy(tracked_chunk_cache); |
1da177e4 LT |
1472 | } |
1473 | ||
1474 | /* Module hooks */ | |
1475 | module_init(dm_snapshot_init); | |
1476 | module_exit(dm_snapshot_exit); | |
1477 | ||
1478 | MODULE_DESCRIPTION(DM_NAME " snapshot target"); | |
1479 | MODULE_AUTHOR("Joe Thornber"); | |
1480 | MODULE_LICENSE("GPL"); |