]>
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> | |
10 | #include <linux/config.h> | |
11 | #include <linux/ctype.h> | |
12 | #include <linux/device-mapper.h> | |
13 | #include <linux/fs.h> | |
14 | #include <linux/init.h> | |
15 | #include <linux/kdev_t.h> | |
16 | #include <linux/list.h> | |
17 | #include <linux/mempool.h> | |
18 | #include <linux/module.h> | |
19 | #include <linux/slab.h> | |
20 | #include <linux/vmalloc.h> | |
21 | ||
22 | #include "dm-snap.h" | |
23 | #include "dm-bio-list.h" | |
24 | #include "kcopyd.h" | |
25 | ||
26 | /* | |
27 | * The percentage increment we will wake up users at | |
28 | */ | |
29 | #define WAKE_UP_PERCENT 5 | |
30 | ||
31 | /* | |
32 | * kcopyd priority of snapshot operations | |
33 | */ | |
34 | #define SNAPSHOT_COPY_PRIORITY 2 | |
35 | ||
36 | /* | |
37 | * Each snapshot reserves this many pages for io | |
38 | */ | |
39 | #define SNAPSHOT_PAGES 256 | |
40 | ||
41 | struct pending_exception { | |
42 | struct exception e; | |
43 | ||
44 | /* | |
45 | * Origin buffers waiting for this to complete are held | |
46 | * in a bio list | |
47 | */ | |
48 | struct bio_list origin_bios; | |
49 | struct bio_list snapshot_bios; | |
50 | ||
51 | /* | |
52 | * Other pending_exceptions that are processing this | |
53 | * chunk. When this list is empty, we know we can | |
54 | * complete the origins. | |
55 | */ | |
56 | struct list_head siblings; | |
57 | ||
58 | /* Pointer back to snapshot context */ | |
59 | struct dm_snapshot *snap; | |
60 | ||
61 | /* | |
62 | * 1 indicates the exception has already been sent to | |
63 | * kcopyd. | |
64 | */ | |
65 | int started; | |
66 | }; | |
67 | ||
68 | /* | |
69 | * Hash table mapping origin volumes to lists of snapshots and | |
70 | * a lock to protect it | |
71 | */ | |
72 | static kmem_cache_t *exception_cache; | |
73 | static kmem_cache_t *pending_cache; | |
74 | static mempool_t *pending_pool; | |
75 | ||
76 | /* | |
77 | * One of these per registered origin, held in the snapshot_origins hash | |
78 | */ | |
79 | struct origin { | |
80 | /* The origin device */ | |
81 | struct block_device *bdev; | |
82 | ||
83 | struct list_head hash_list; | |
84 | ||
85 | /* List of snapshots for this origin */ | |
86 | struct list_head snapshots; | |
87 | }; | |
88 | ||
89 | /* | |
90 | * Size of the hash table for origin volumes. If we make this | |
91 | * the size of the minors list then it should be nearly perfect | |
92 | */ | |
93 | #define ORIGIN_HASH_SIZE 256 | |
94 | #define ORIGIN_MASK 0xFF | |
95 | static struct list_head *_origins; | |
96 | static struct rw_semaphore _origins_lock; | |
97 | ||
98 | static int init_origin_hash(void) | |
99 | { | |
100 | int i; | |
101 | ||
102 | _origins = kmalloc(ORIGIN_HASH_SIZE * sizeof(struct list_head), | |
103 | GFP_KERNEL); | |
104 | if (!_origins) { | |
105 | DMERR("Device mapper: Snapshot: unable to allocate memory"); | |
106 | return -ENOMEM; | |
107 | } | |
108 | ||
109 | for (i = 0; i < ORIGIN_HASH_SIZE; i++) | |
110 | INIT_LIST_HEAD(_origins + i); | |
111 | init_rwsem(&_origins_lock); | |
112 | ||
113 | return 0; | |
114 | } | |
115 | ||
116 | static void exit_origin_hash(void) | |
117 | { | |
118 | kfree(_origins); | |
119 | } | |
120 | ||
121 | static inline unsigned int origin_hash(struct block_device *bdev) | |
122 | { | |
123 | return bdev->bd_dev & ORIGIN_MASK; | |
124 | } | |
125 | ||
126 | static struct origin *__lookup_origin(struct block_device *origin) | |
127 | { | |
128 | struct list_head *ol; | |
129 | struct origin *o; | |
130 | ||
131 | ol = &_origins[origin_hash(origin)]; | |
132 | list_for_each_entry (o, ol, hash_list) | |
133 | if (bdev_equal(o->bdev, origin)) | |
134 | return o; | |
135 | ||
136 | return NULL; | |
137 | } | |
138 | ||
139 | static void __insert_origin(struct origin *o) | |
140 | { | |
141 | struct list_head *sl = &_origins[origin_hash(o->bdev)]; | |
142 | list_add_tail(&o->hash_list, sl); | |
143 | } | |
144 | ||
145 | /* | |
146 | * Make a note of the snapshot and its origin so we can look it | |
147 | * up when the origin has a write on it. | |
148 | */ | |
149 | static int register_snapshot(struct dm_snapshot *snap) | |
150 | { | |
151 | struct origin *o; | |
152 | struct block_device *bdev = snap->origin->bdev; | |
153 | ||
154 | down_write(&_origins_lock); | |
155 | o = __lookup_origin(bdev); | |
156 | ||
157 | if (!o) { | |
158 | /* New origin */ | |
159 | o = kmalloc(sizeof(*o), GFP_KERNEL); | |
160 | if (!o) { | |
161 | up_write(&_origins_lock); | |
162 | return -ENOMEM; | |
163 | } | |
164 | ||
165 | /* Initialise the struct */ | |
166 | INIT_LIST_HEAD(&o->snapshots); | |
167 | o->bdev = bdev; | |
168 | ||
169 | __insert_origin(o); | |
170 | } | |
171 | ||
172 | list_add_tail(&snap->list, &o->snapshots); | |
173 | ||
174 | up_write(&_origins_lock); | |
175 | return 0; | |
176 | } | |
177 | ||
178 | static void unregister_snapshot(struct dm_snapshot *s) | |
179 | { | |
180 | struct origin *o; | |
181 | ||
182 | down_write(&_origins_lock); | |
183 | o = __lookup_origin(s->origin->bdev); | |
184 | ||
185 | list_del(&s->list); | |
186 | if (list_empty(&o->snapshots)) { | |
187 | list_del(&o->hash_list); | |
188 | kfree(o); | |
189 | } | |
190 | ||
191 | up_write(&_origins_lock); | |
192 | } | |
193 | ||
194 | /* | |
195 | * Implementation of the exception hash tables. | |
196 | */ | |
197 | static int init_exception_table(struct exception_table *et, uint32_t size) | |
198 | { | |
199 | unsigned int i; | |
200 | ||
201 | et->hash_mask = size - 1; | |
202 | et->table = dm_vcalloc(size, sizeof(struct list_head)); | |
203 | if (!et->table) | |
204 | return -ENOMEM; | |
205 | ||
206 | for (i = 0; i < size; i++) | |
207 | INIT_LIST_HEAD(et->table + i); | |
208 | ||
209 | return 0; | |
210 | } | |
211 | ||
212 | static void exit_exception_table(struct exception_table *et, kmem_cache_t *mem) | |
213 | { | |
214 | struct list_head *slot; | |
215 | struct exception *ex, *next; | |
216 | int i, size; | |
217 | ||
218 | size = et->hash_mask + 1; | |
219 | for (i = 0; i < size; i++) { | |
220 | slot = et->table + i; | |
221 | ||
222 | list_for_each_entry_safe (ex, next, slot, hash_list) | |
223 | kmem_cache_free(mem, ex); | |
224 | } | |
225 | ||
226 | vfree(et->table); | |
227 | } | |
228 | ||
229 | static inline uint32_t exception_hash(struct exception_table *et, chunk_t chunk) | |
230 | { | |
231 | return chunk & et->hash_mask; | |
232 | } | |
233 | ||
234 | static void insert_exception(struct exception_table *eh, struct exception *e) | |
235 | { | |
236 | struct list_head *l = &eh->table[exception_hash(eh, e->old_chunk)]; | |
237 | list_add(&e->hash_list, l); | |
238 | } | |
239 | ||
240 | static inline void remove_exception(struct exception *e) | |
241 | { | |
242 | list_del(&e->hash_list); | |
243 | } | |
244 | ||
245 | /* | |
246 | * Return the exception data for a sector, or NULL if not | |
247 | * remapped. | |
248 | */ | |
249 | static struct exception *lookup_exception(struct exception_table *et, | |
250 | chunk_t chunk) | |
251 | { | |
252 | struct list_head *slot; | |
253 | struct exception *e; | |
254 | ||
255 | slot = &et->table[exception_hash(et, chunk)]; | |
256 | list_for_each_entry (e, slot, hash_list) | |
257 | if (e->old_chunk == chunk) | |
258 | return e; | |
259 | ||
260 | return NULL; | |
261 | } | |
262 | ||
263 | static inline struct exception *alloc_exception(void) | |
264 | { | |
265 | struct exception *e; | |
266 | ||
267 | e = kmem_cache_alloc(exception_cache, GFP_NOIO); | |
268 | if (!e) | |
269 | e = kmem_cache_alloc(exception_cache, GFP_ATOMIC); | |
270 | ||
271 | return e; | |
272 | } | |
273 | ||
274 | static inline void free_exception(struct exception *e) | |
275 | { | |
276 | kmem_cache_free(exception_cache, e); | |
277 | } | |
278 | ||
279 | static inline struct pending_exception *alloc_pending_exception(void) | |
280 | { | |
281 | return mempool_alloc(pending_pool, GFP_NOIO); | |
282 | } | |
283 | ||
284 | static inline void free_pending_exception(struct pending_exception *pe) | |
285 | { | |
286 | mempool_free(pe, pending_pool); | |
287 | } | |
288 | ||
289 | int dm_add_exception(struct dm_snapshot *s, chunk_t old, chunk_t new) | |
290 | { | |
291 | struct exception *e; | |
292 | ||
293 | e = alloc_exception(); | |
294 | if (!e) | |
295 | return -ENOMEM; | |
296 | ||
297 | e->old_chunk = old; | |
298 | e->new_chunk = new; | |
299 | insert_exception(&s->complete, e); | |
300 | return 0; | |
301 | } | |
302 | ||
303 | /* | |
304 | * Hard coded magic. | |
305 | */ | |
306 | static int calc_max_buckets(void) | |
307 | { | |
308 | /* use a fixed size of 2MB */ | |
309 | unsigned long mem = 2 * 1024 * 1024; | |
310 | mem /= sizeof(struct list_head); | |
311 | ||
312 | return mem; | |
313 | } | |
314 | ||
315 | /* | |
316 | * Rounds a number down to a power of 2. | |
317 | */ | |
318 | static inline uint32_t round_down(uint32_t n) | |
319 | { | |
320 | while (n & (n - 1)) | |
321 | n &= (n - 1); | |
322 | return n; | |
323 | } | |
324 | ||
325 | /* | |
326 | * Allocate room for a suitable hash table. | |
327 | */ | |
328 | static int init_hash_tables(struct dm_snapshot *s) | |
329 | { | |
330 | sector_t hash_size, cow_dev_size, origin_dev_size, max_buckets; | |
331 | ||
332 | /* | |
333 | * Calculate based on the size of the original volume or | |
334 | * the COW volume... | |
335 | */ | |
336 | cow_dev_size = get_dev_size(s->cow->bdev); | |
337 | origin_dev_size = get_dev_size(s->origin->bdev); | |
338 | max_buckets = calc_max_buckets(); | |
339 | ||
340 | hash_size = min(origin_dev_size, cow_dev_size) >> s->chunk_shift; | |
341 | hash_size = min(hash_size, max_buckets); | |
342 | ||
343 | /* Round it down to a power of 2 */ | |
344 | hash_size = round_down(hash_size); | |
345 | if (init_exception_table(&s->complete, hash_size)) | |
346 | return -ENOMEM; | |
347 | ||
348 | /* | |
349 | * Allocate hash table for in-flight exceptions | |
350 | * Make this smaller than the real hash table | |
351 | */ | |
352 | hash_size >>= 3; | |
353 | if (hash_size < 64) | |
354 | hash_size = 64; | |
355 | ||
356 | if (init_exception_table(&s->pending, hash_size)) { | |
357 | exit_exception_table(&s->complete, exception_cache); | |
358 | return -ENOMEM; | |
359 | } | |
360 | ||
361 | return 0; | |
362 | } | |
363 | ||
364 | /* | |
365 | * Round a number up to the nearest 'size' boundary. size must | |
366 | * be a power of 2. | |
367 | */ | |
368 | static inline ulong round_up(ulong n, ulong size) | |
369 | { | |
370 | size--; | |
371 | return (n + size) & ~size; | |
372 | } | |
373 | ||
2d38fe20 AK |
374 | static void read_snapshot_metadata(struct dm_snapshot *s) |
375 | { | |
2d38fe20 AK |
376 | if (s->store.read_metadata(&s->store)) { |
377 | down_write(&s->lock); | |
378 | s->valid = 0; | |
379 | up_write(&s->lock); | |
380 | } | |
2d38fe20 AK |
381 | } |
382 | ||
1da177e4 LT |
383 | /* |
384 | * Construct a snapshot mapping: <origin_dev> <COW-dev> <p/n> <chunk-size> | |
385 | */ | |
386 | static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv) | |
387 | { | |
388 | struct dm_snapshot *s; | |
389 | unsigned long chunk_size; | |
390 | int r = -EINVAL; | |
391 | char persistent; | |
392 | char *origin_path; | |
393 | char *cow_path; | |
394 | char *value; | |
395 | int blocksize; | |
396 | ||
397 | if (argc < 4) { | |
398 | ti->error = "dm-snapshot: requires exactly 4 arguments"; | |
399 | r = -EINVAL; | |
400 | goto bad1; | |
401 | } | |
402 | ||
403 | origin_path = argv[0]; | |
404 | cow_path = argv[1]; | |
405 | persistent = toupper(*argv[2]); | |
406 | ||
407 | if (persistent != 'P' && persistent != 'N') { | |
408 | ti->error = "Persistent flag is not P or N"; | |
409 | r = -EINVAL; | |
410 | goto bad1; | |
411 | } | |
412 | ||
413 | chunk_size = simple_strtoul(argv[3], &value, 10); | |
414 | if (chunk_size == 0 || value == NULL) { | |
415 | ti->error = "Invalid chunk size"; | |
416 | r = -EINVAL; | |
417 | goto bad1; | |
418 | } | |
419 | ||
420 | s = kmalloc(sizeof(*s), GFP_KERNEL); | |
421 | if (s == NULL) { | |
422 | ti->error = "Cannot allocate snapshot context private " | |
423 | "structure"; | |
424 | r = -ENOMEM; | |
425 | goto bad1; | |
426 | } | |
427 | ||
428 | r = dm_get_device(ti, origin_path, 0, ti->len, FMODE_READ, &s->origin); | |
429 | if (r) { | |
430 | ti->error = "Cannot get origin device"; | |
431 | goto bad2; | |
432 | } | |
433 | ||
434 | r = dm_get_device(ti, cow_path, 0, 0, | |
435 | FMODE_READ | FMODE_WRITE, &s->cow); | |
436 | if (r) { | |
437 | dm_put_device(ti, s->origin); | |
438 | ti->error = "Cannot get COW device"; | |
439 | goto bad2; | |
440 | } | |
441 | ||
442 | /* | |
443 | * Chunk size must be multiple of page size. Silently | |
444 | * round up if it's not. | |
445 | */ | |
446 | chunk_size = round_up(chunk_size, PAGE_SIZE >> 9); | |
447 | ||
448 | /* Validate the chunk size against the device block size */ | |
449 | blocksize = s->cow->bdev->bd_disk->queue->hardsect_size; | |
450 | if (chunk_size % (blocksize >> 9)) { | |
451 | ti->error = "Chunk size is not a multiple of device blocksize"; | |
452 | r = -EINVAL; | |
453 | goto bad3; | |
454 | } | |
455 | ||
456 | /* Check chunk_size is a power of 2 */ | |
457 | if (chunk_size & (chunk_size - 1)) { | |
458 | ti->error = "Chunk size is not a power of 2"; | |
459 | r = -EINVAL; | |
460 | goto bad3; | |
461 | } | |
462 | ||
463 | s->chunk_size = chunk_size; | |
464 | s->chunk_mask = chunk_size - 1; | |
465 | s->type = persistent; | |
466 | s->chunk_shift = ffs(chunk_size) - 1; | |
467 | ||
468 | s->valid = 1; | |
aa14edeb | 469 | s->active = 0; |
1da177e4 LT |
470 | s->last_percent = 0; |
471 | init_rwsem(&s->lock); | |
472 | s->table = ti->table; | |
473 | ||
474 | /* Allocate hash table for COW data */ | |
475 | if (init_hash_tables(s)) { | |
476 | ti->error = "Unable to allocate hash table space"; | |
477 | r = -ENOMEM; | |
478 | goto bad3; | |
479 | } | |
480 | ||
481 | /* | |
482 | * Check the persistent flag - done here because we need the iobuf | |
483 | * to check the LV header | |
484 | */ | |
485 | s->store.snap = s; | |
486 | ||
487 | if (persistent == 'P') | |
488 | r = dm_create_persistent(&s->store, chunk_size); | |
489 | else | |
490 | r = dm_create_transient(&s->store, s, blocksize); | |
491 | ||
492 | if (r) { | |
493 | ti->error = "Couldn't create exception store"; | |
494 | r = -EINVAL; | |
495 | goto bad4; | |
496 | } | |
497 | ||
498 | r = kcopyd_client_create(SNAPSHOT_PAGES, &s->kcopyd_client); | |
499 | if (r) { | |
500 | ti->error = "Could not create kcopyd client"; | |
501 | goto bad5; | |
502 | } | |
503 | ||
aa14edeb AK |
504 | /* Metadata must only be loaded into one table at once */ |
505 | read_snapshot_metadata(s); | |
506 | ||
1da177e4 | 507 | /* Add snapshot to the list of snapshots for this origin */ |
aa14edeb | 508 | /* Exceptions aren't triggered till snapshot_resume() is called */ |
1da177e4 LT |
509 | if (register_snapshot(s)) { |
510 | r = -EINVAL; | |
511 | ti->error = "Cannot register snapshot origin"; | |
512 | goto bad6; | |
513 | } | |
514 | ||
515 | ti->private = s; | |
516 | ti->split_io = chunk_size; | |
517 | ||
518 | return 0; | |
519 | ||
520 | bad6: | |
521 | kcopyd_client_destroy(s->kcopyd_client); | |
522 | ||
523 | bad5: | |
524 | s->store.destroy(&s->store); | |
525 | ||
526 | bad4: | |
527 | exit_exception_table(&s->pending, pending_cache); | |
528 | exit_exception_table(&s->complete, exception_cache); | |
529 | ||
530 | bad3: | |
531 | dm_put_device(ti, s->cow); | |
532 | dm_put_device(ti, s->origin); | |
533 | ||
534 | bad2: | |
535 | kfree(s); | |
536 | ||
537 | bad1: | |
538 | return r; | |
539 | } | |
540 | ||
541 | static void snapshot_dtr(struct dm_target *ti) | |
542 | { | |
543 | struct dm_snapshot *s = (struct dm_snapshot *) ti->private; | |
544 | ||
545 | unregister_snapshot(s); | |
546 | ||
547 | exit_exception_table(&s->pending, pending_cache); | |
548 | exit_exception_table(&s->complete, exception_cache); | |
549 | ||
550 | /* Deallocate memory used */ | |
551 | s->store.destroy(&s->store); | |
552 | ||
553 | dm_put_device(ti, s->origin); | |
554 | dm_put_device(ti, s->cow); | |
555 | kcopyd_client_destroy(s->kcopyd_client); | |
556 | kfree(s); | |
557 | } | |
558 | ||
559 | /* | |
560 | * Flush a list of buffers. | |
561 | */ | |
562 | static void flush_bios(struct bio *bio) | |
563 | { | |
564 | struct bio *n; | |
565 | ||
566 | while (bio) { | |
567 | n = bio->bi_next; | |
568 | bio->bi_next = NULL; | |
569 | generic_make_request(bio); | |
570 | bio = n; | |
571 | } | |
572 | } | |
573 | ||
574 | /* | |
575 | * Error a list of buffers. | |
576 | */ | |
577 | static void error_bios(struct bio *bio) | |
578 | { | |
579 | struct bio *n; | |
580 | ||
581 | while (bio) { | |
582 | n = bio->bi_next; | |
583 | bio->bi_next = NULL; | |
584 | bio_io_error(bio, bio->bi_size); | |
585 | bio = n; | |
586 | } | |
587 | } | |
588 | ||
589 | static struct bio *__flush_bios(struct pending_exception *pe) | |
590 | { | |
591 | struct pending_exception *sibling; | |
592 | ||
593 | if (list_empty(&pe->siblings)) | |
594 | return bio_list_get(&pe->origin_bios); | |
595 | ||
596 | sibling = list_entry(pe->siblings.next, | |
597 | struct pending_exception, siblings); | |
598 | ||
599 | list_del(&pe->siblings); | |
600 | ||
601 | /* This is fine as long as kcopyd is single-threaded. If kcopyd | |
602 | * becomes multi-threaded, we'll need some locking here. | |
603 | */ | |
604 | bio_list_merge(&sibling->origin_bios, &pe->origin_bios); | |
605 | ||
606 | return NULL; | |
607 | } | |
608 | ||
609 | static void pending_complete(struct pending_exception *pe, int success) | |
610 | { | |
611 | struct exception *e; | |
612 | struct dm_snapshot *s = pe->snap; | |
613 | struct bio *flush = NULL; | |
614 | ||
615 | if (success) { | |
616 | e = alloc_exception(); | |
617 | if (!e) { | |
618 | DMWARN("Unable to allocate exception."); | |
619 | down_write(&s->lock); | |
620 | s->store.drop_snapshot(&s->store); | |
621 | s->valid = 0; | |
622 | flush = __flush_bios(pe); | |
623 | up_write(&s->lock); | |
624 | ||
625 | error_bios(bio_list_get(&pe->snapshot_bios)); | |
626 | goto out; | |
627 | } | |
628 | *e = pe->e; | |
629 | ||
630 | /* | |
631 | * Add a proper exception, and remove the | |
632 | * in-flight exception from the list. | |
633 | */ | |
634 | down_write(&s->lock); | |
635 | insert_exception(&s->complete, e); | |
636 | remove_exception(&pe->e); | |
637 | flush = __flush_bios(pe); | |
638 | ||
639 | /* Submit any pending write bios */ | |
640 | up_write(&s->lock); | |
641 | ||
642 | flush_bios(bio_list_get(&pe->snapshot_bios)); | |
643 | } else { | |
644 | /* Read/write error - snapshot is unusable */ | |
645 | down_write(&s->lock); | |
646 | if (s->valid) | |
647 | DMERR("Error reading/writing snapshot"); | |
648 | s->store.drop_snapshot(&s->store); | |
649 | s->valid = 0; | |
650 | remove_exception(&pe->e); | |
651 | flush = __flush_bios(pe); | |
652 | up_write(&s->lock); | |
653 | ||
654 | error_bios(bio_list_get(&pe->snapshot_bios)); | |
655 | ||
656 | dm_table_event(s->table); | |
657 | } | |
658 | ||
659 | out: | |
660 | free_pending_exception(pe); | |
661 | ||
662 | if (flush) | |
663 | flush_bios(flush); | |
664 | } | |
665 | ||
666 | static void commit_callback(void *context, int success) | |
667 | { | |
668 | struct pending_exception *pe = (struct pending_exception *) context; | |
669 | pending_complete(pe, success); | |
670 | } | |
671 | ||
672 | /* | |
673 | * Called when the copy I/O has finished. kcopyd actually runs | |
674 | * this code so don't block. | |
675 | */ | |
676 | static void copy_callback(int read_err, unsigned int write_err, void *context) | |
677 | { | |
678 | struct pending_exception *pe = (struct pending_exception *) context; | |
679 | struct dm_snapshot *s = pe->snap; | |
680 | ||
681 | if (read_err || write_err) | |
682 | pending_complete(pe, 0); | |
683 | ||
684 | else | |
685 | /* Update the metadata if we are persistent */ | |
686 | s->store.commit_exception(&s->store, &pe->e, commit_callback, | |
687 | pe); | |
688 | } | |
689 | ||
690 | /* | |
691 | * Dispatches the copy operation to kcopyd. | |
692 | */ | |
858119e1 | 693 | static void start_copy(struct pending_exception *pe) |
1da177e4 LT |
694 | { |
695 | struct dm_snapshot *s = pe->snap; | |
696 | struct io_region src, dest; | |
697 | struct block_device *bdev = s->origin->bdev; | |
698 | sector_t dev_size; | |
699 | ||
700 | dev_size = get_dev_size(bdev); | |
701 | ||
702 | src.bdev = bdev; | |
703 | src.sector = chunk_to_sector(s, pe->e.old_chunk); | |
704 | src.count = min(s->chunk_size, dev_size - src.sector); | |
705 | ||
706 | dest.bdev = s->cow->bdev; | |
707 | dest.sector = chunk_to_sector(s, pe->e.new_chunk); | |
708 | dest.count = src.count; | |
709 | ||
710 | /* Hand over to kcopyd */ | |
711 | kcopyd_copy(s->kcopyd_client, | |
712 | &src, 1, &dest, 0, copy_callback, pe); | |
713 | } | |
714 | ||
715 | /* | |
716 | * Looks to see if this snapshot already has a pending exception | |
717 | * for this chunk, otherwise it allocates a new one and inserts | |
718 | * it into the pending table. | |
719 | * | |
720 | * NOTE: a write lock must be held on snap->lock before calling | |
721 | * this. | |
722 | */ | |
723 | static struct pending_exception * | |
724 | __find_pending_exception(struct dm_snapshot *s, struct bio *bio) | |
725 | { | |
726 | struct exception *e; | |
727 | struct pending_exception *pe; | |
728 | chunk_t chunk = sector_to_chunk(s, bio->bi_sector); | |
729 | ||
730 | /* | |
731 | * Is there a pending exception for this already ? | |
732 | */ | |
733 | e = lookup_exception(&s->pending, chunk); | |
734 | if (e) { | |
735 | /* cast the exception to a pending exception */ | |
736 | pe = container_of(e, struct pending_exception, e); | |
737 | ||
738 | } else { | |
739 | /* | |
740 | * Create a new pending exception, we don't want | |
741 | * to hold the lock while we do this. | |
742 | */ | |
743 | up_write(&s->lock); | |
744 | pe = alloc_pending_exception(); | |
745 | down_write(&s->lock); | |
746 | ||
747 | e = lookup_exception(&s->pending, chunk); | |
748 | if (e) { | |
749 | free_pending_exception(pe); | |
750 | pe = container_of(e, struct pending_exception, e); | |
751 | } else { | |
752 | pe->e.old_chunk = chunk; | |
753 | bio_list_init(&pe->origin_bios); | |
754 | bio_list_init(&pe->snapshot_bios); | |
755 | INIT_LIST_HEAD(&pe->siblings); | |
756 | pe->snap = s; | |
757 | pe->started = 0; | |
758 | ||
759 | if (s->store.prepare_exception(&s->store, &pe->e)) { | |
760 | free_pending_exception(pe); | |
761 | s->valid = 0; | |
762 | return NULL; | |
763 | } | |
764 | ||
765 | insert_exception(&s->pending, &pe->e); | |
766 | } | |
767 | } | |
768 | ||
769 | return pe; | |
770 | } | |
771 | ||
772 | static inline void remap_exception(struct dm_snapshot *s, struct exception *e, | |
773 | struct bio *bio) | |
774 | { | |
775 | bio->bi_bdev = s->cow->bdev; | |
776 | bio->bi_sector = chunk_to_sector(s, e->new_chunk) + | |
777 | (bio->bi_sector & s->chunk_mask); | |
778 | } | |
779 | ||
780 | static int snapshot_map(struct dm_target *ti, struct bio *bio, | |
781 | union map_info *map_context) | |
782 | { | |
783 | struct exception *e; | |
784 | struct dm_snapshot *s = (struct dm_snapshot *) ti->private; | |
785 | int r = 1; | |
786 | chunk_t chunk; | |
787 | struct pending_exception *pe; | |
788 | ||
789 | chunk = sector_to_chunk(s, bio->bi_sector); | |
790 | ||
791 | /* Full snapshots are not usable */ | |
792 | if (!s->valid) | |
f6a80ea8 | 793 | return -EIO; |
1da177e4 | 794 | |
4aac0a63 AK |
795 | if (unlikely(bio_barrier(bio))) |
796 | return -EOPNOTSUPP; | |
797 | ||
1da177e4 LT |
798 | /* |
799 | * Write to snapshot - higher level takes care of RW/RO | |
800 | * flags so we should only get this if we are | |
801 | * writeable. | |
802 | */ | |
803 | if (bio_rw(bio) == WRITE) { | |
804 | ||
805 | /* FIXME: should only take write lock if we need | |
806 | * to copy an exception */ | |
807 | down_write(&s->lock); | |
808 | ||
809 | /* If the block is already remapped - use that, else remap it */ | |
810 | e = lookup_exception(&s->complete, chunk); | |
811 | if (e) { | |
812 | remap_exception(s, e, bio); | |
813 | up_write(&s->lock); | |
814 | ||
815 | } else { | |
816 | pe = __find_pending_exception(s, bio); | |
817 | ||
818 | if (!pe) { | |
819 | if (s->store.drop_snapshot) | |
820 | s->store.drop_snapshot(&s->store); | |
821 | s->valid = 0; | |
822 | r = -EIO; | |
823 | up_write(&s->lock); | |
824 | } else { | |
825 | remap_exception(s, &pe->e, bio); | |
826 | bio_list_add(&pe->snapshot_bios, bio); | |
827 | ||
828 | if (!pe->started) { | |
829 | /* this is protected by snap->lock */ | |
830 | pe->started = 1; | |
831 | up_write(&s->lock); | |
832 | start_copy(pe); | |
833 | } else | |
834 | up_write(&s->lock); | |
835 | r = 0; | |
836 | } | |
837 | } | |
838 | ||
839 | } else { | |
840 | /* | |
841 | * FIXME: this read path scares me because we | |
842 | * always use the origin when we have a pending | |
843 | * exception. However I can't think of a | |
844 | * situation where this is wrong - ejt. | |
845 | */ | |
846 | ||
847 | /* Do reads */ | |
848 | down_read(&s->lock); | |
849 | ||
850 | /* See if it it has been remapped */ | |
851 | e = lookup_exception(&s->complete, chunk); | |
852 | if (e) | |
853 | remap_exception(s, e, bio); | |
854 | else | |
855 | bio->bi_bdev = s->origin->bdev; | |
856 | ||
857 | up_read(&s->lock); | |
858 | } | |
859 | ||
860 | return r; | |
861 | } | |
862 | ||
863 | static void snapshot_resume(struct dm_target *ti) | |
864 | { | |
865 | struct dm_snapshot *s = (struct dm_snapshot *) ti->private; | |
866 | ||
aa14edeb AK |
867 | down_write(&s->lock); |
868 | s->active = 1; | |
869 | up_write(&s->lock); | |
1da177e4 LT |
870 | } |
871 | ||
872 | static int snapshot_status(struct dm_target *ti, status_type_t type, | |
873 | char *result, unsigned int maxlen) | |
874 | { | |
875 | struct dm_snapshot *snap = (struct dm_snapshot *) ti->private; | |
876 | ||
877 | switch (type) { | |
878 | case STATUSTYPE_INFO: | |
879 | if (!snap->valid) | |
880 | snprintf(result, maxlen, "Invalid"); | |
881 | else { | |
882 | if (snap->store.fraction_full) { | |
883 | sector_t numerator, denominator; | |
884 | snap->store.fraction_full(&snap->store, | |
885 | &numerator, | |
886 | &denominator); | |
887 | snprintf(result, maxlen, | |
888 | SECTOR_FORMAT "/" SECTOR_FORMAT, | |
889 | numerator, denominator); | |
890 | } | |
891 | else | |
892 | snprintf(result, maxlen, "Unknown"); | |
893 | } | |
894 | break; | |
895 | ||
896 | case STATUSTYPE_TABLE: | |
897 | /* | |
898 | * kdevname returns a static pointer so we need | |
899 | * to make private copies if the output is to | |
900 | * make sense. | |
901 | */ | |
902 | snprintf(result, maxlen, "%s %s %c " SECTOR_FORMAT, | |
903 | snap->origin->name, snap->cow->name, | |
904 | snap->type, snap->chunk_size); | |
905 | break; | |
906 | } | |
907 | ||
908 | return 0; | |
909 | } | |
910 | ||
911 | /*----------------------------------------------------------------- | |
912 | * Origin methods | |
913 | *---------------------------------------------------------------*/ | |
914 | static void list_merge(struct list_head *l1, struct list_head *l2) | |
915 | { | |
916 | struct list_head *l1_n, *l2_p; | |
917 | ||
918 | l1_n = l1->next; | |
919 | l2_p = l2->prev; | |
920 | ||
921 | l1->next = l2; | |
922 | l2->prev = l1; | |
923 | ||
924 | l2_p->next = l1_n; | |
925 | l1_n->prev = l2_p; | |
926 | } | |
927 | ||
928 | static int __origin_write(struct list_head *snapshots, struct bio *bio) | |
929 | { | |
930 | int r = 1, first = 1; | |
931 | struct dm_snapshot *snap; | |
932 | struct exception *e; | |
933 | struct pending_exception *pe, *last = NULL; | |
934 | chunk_t chunk; | |
935 | ||
936 | /* Do all the snapshots on this origin */ | |
937 | list_for_each_entry (snap, snapshots, list) { | |
938 | ||
aa14edeb AK |
939 | /* Only deal with valid and active snapshots */ |
940 | if (!snap->valid || !snap->active) | |
1da177e4 LT |
941 | continue; |
942 | ||
d5e404c1 AK |
943 | /* Nothing to do if writing beyond end of snapshot */ |
944 | if (bio->bi_sector >= dm_table_get_size(snap->table)) | |
945 | continue; | |
946 | ||
1da177e4 LT |
947 | down_write(&snap->lock); |
948 | ||
949 | /* | |
950 | * Remember, different snapshots can have | |
951 | * different chunk sizes. | |
952 | */ | |
953 | chunk = sector_to_chunk(snap, bio->bi_sector); | |
954 | ||
955 | /* | |
956 | * Check exception table to see if block | |
957 | * is already remapped in this snapshot | |
958 | * and trigger an exception if not. | |
959 | */ | |
960 | e = lookup_exception(&snap->complete, chunk); | |
961 | if (!e) { | |
962 | pe = __find_pending_exception(snap, bio); | |
963 | if (!pe) { | |
964 | snap->store.drop_snapshot(&snap->store); | |
965 | snap->valid = 0; | |
966 | ||
967 | } else { | |
968 | if (last) | |
969 | list_merge(&pe->siblings, | |
970 | &last->siblings); | |
971 | ||
972 | last = pe; | |
973 | r = 0; | |
974 | } | |
975 | } | |
976 | ||
977 | up_write(&snap->lock); | |
978 | } | |
979 | ||
980 | /* | |
981 | * Now that we have a complete pe list we can start the copying. | |
982 | */ | |
983 | if (last) { | |
984 | pe = last; | |
985 | do { | |
986 | down_write(&pe->snap->lock); | |
987 | if (first) | |
988 | bio_list_add(&pe->origin_bios, bio); | |
989 | if (!pe->started) { | |
990 | pe->started = 1; | |
991 | up_write(&pe->snap->lock); | |
992 | start_copy(pe); | |
993 | } else | |
994 | up_write(&pe->snap->lock); | |
995 | first = 0; | |
996 | pe = list_entry(pe->siblings.next, | |
997 | struct pending_exception, siblings); | |
998 | ||
999 | } while (pe != last); | |
1000 | } | |
1001 | ||
1002 | return r; | |
1003 | } | |
1004 | ||
1005 | /* | |
1006 | * Called on a write from the origin driver. | |
1007 | */ | |
1008 | static int do_origin(struct dm_dev *origin, struct bio *bio) | |
1009 | { | |
1010 | struct origin *o; | |
1011 | int r = 1; | |
1012 | ||
1013 | down_read(&_origins_lock); | |
1014 | o = __lookup_origin(origin->bdev); | |
1015 | if (o) | |
1016 | r = __origin_write(&o->snapshots, bio); | |
1017 | up_read(&_origins_lock); | |
1018 | ||
1019 | return r; | |
1020 | } | |
1021 | ||
1022 | /* | |
1023 | * Origin: maps a linear range of a device, with hooks for snapshotting. | |
1024 | */ | |
1025 | ||
1026 | /* | |
1027 | * Construct an origin mapping: <dev_path> | |
1028 | * The context for an origin is merely a 'struct dm_dev *' | |
1029 | * pointing to the real device. | |
1030 | */ | |
1031 | static int origin_ctr(struct dm_target *ti, unsigned int argc, char **argv) | |
1032 | { | |
1033 | int r; | |
1034 | struct dm_dev *dev; | |
1035 | ||
1036 | if (argc != 1) { | |
1037 | ti->error = "dm-origin: incorrect number of arguments"; | |
1038 | return -EINVAL; | |
1039 | } | |
1040 | ||
1041 | r = dm_get_device(ti, argv[0], 0, ti->len, | |
1042 | dm_table_get_mode(ti->table), &dev); | |
1043 | if (r) { | |
1044 | ti->error = "Cannot get target device"; | |
1045 | return r; | |
1046 | } | |
1047 | ||
1048 | ti->private = dev; | |
1049 | return 0; | |
1050 | } | |
1051 | ||
1052 | static void origin_dtr(struct dm_target *ti) | |
1053 | { | |
1054 | struct dm_dev *dev = (struct dm_dev *) ti->private; | |
1055 | dm_put_device(ti, dev); | |
1056 | } | |
1057 | ||
1058 | static int origin_map(struct dm_target *ti, struct bio *bio, | |
1059 | union map_info *map_context) | |
1060 | { | |
1061 | struct dm_dev *dev = (struct dm_dev *) ti->private; | |
1062 | bio->bi_bdev = dev->bdev; | |
1063 | ||
4aac0a63 AK |
1064 | if (unlikely(bio_barrier(bio))) |
1065 | return -EOPNOTSUPP; | |
1066 | ||
1da177e4 LT |
1067 | /* Only tell snapshots if this is a write */ |
1068 | return (bio_rw(bio) == WRITE) ? do_origin(dev, bio) : 1; | |
1069 | } | |
1070 | ||
1071 | #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r)) | |
1072 | ||
1073 | /* | |
1074 | * Set the target "split_io" field to the minimum of all the snapshots' | |
1075 | * chunk sizes. | |
1076 | */ | |
1077 | static void origin_resume(struct dm_target *ti) | |
1078 | { | |
1079 | struct dm_dev *dev = (struct dm_dev *) ti->private; | |
1080 | struct dm_snapshot *snap; | |
1081 | struct origin *o; | |
1082 | chunk_t chunk_size = 0; | |
1083 | ||
1084 | down_read(&_origins_lock); | |
1085 | o = __lookup_origin(dev->bdev); | |
1086 | if (o) | |
1087 | list_for_each_entry (snap, &o->snapshots, list) | |
1088 | chunk_size = min_not_zero(chunk_size, snap->chunk_size); | |
1089 | up_read(&_origins_lock); | |
1090 | ||
1091 | ti->split_io = chunk_size; | |
1092 | } | |
1093 | ||
1094 | static int origin_status(struct dm_target *ti, status_type_t type, char *result, | |
1095 | unsigned int maxlen) | |
1096 | { | |
1097 | struct dm_dev *dev = (struct dm_dev *) ti->private; | |
1098 | ||
1099 | switch (type) { | |
1100 | case STATUSTYPE_INFO: | |
1101 | result[0] = '\0'; | |
1102 | break; | |
1103 | ||
1104 | case STATUSTYPE_TABLE: | |
1105 | snprintf(result, maxlen, "%s", dev->name); | |
1106 | break; | |
1107 | } | |
1108 | ||
1109 | return 0; | |
1110 | } | |
1111 | ||
1112 | static struct target_type origin_target = { | |
1113 | .name = "snapshot-origin", | |
aa14edeb | 1114 | .version = {1, 1, 0}, |
1da177e4 LT |
1115 | .module = THIS_MODULE, |
1116 | .ctr = origin_ctr, | |
1117 | .dtr = origin_dtr, | |
1118 | .map = origin_map, | |
1119 | .resume = origin_resume, | |
1120 | .status = origin_status, | |
1121 | }; | |
1122 | ||
1123 | static struct target_type snapshot_target = { | |
1124 | .name = "snapshot", | |
aa14edeb | 1125 | .version = {1, 1, 0}, |
1da177e4 LT |
1126 | .module = THIS_MODULE, |
1127 | .ctr = snapshot_ctr, | |
1128 | .dtr = snapshot_dtr, | |
1129 | .map = snapshot_map, | |
1130 | .resume = snapshot_resume, | |
1131 | .status = snapshot_status, | |
1132 | }; | |
1133 | ||
1134 | static int __init dm_snapshot_init(void) | |
1135 | { | |
1136 | int r; | |
1137 | ||
1138 | r = dm_register_target(&snapshot_target); | |
1139 | if (r) { | |
1140 | DMERR("snapshot target register failed %d", r); | |
1141 | return r; | |
1142 | } | |
1143 | ||
1144 | r = dm_register_target(&origin_target); | |
1145 | if (r < 0) { | |
1146 | DMERR("Device mapper: Origin: register failed %d\n", r); | |
1147 | goto bad1; | |
1148 | } | |
1149 | ||
1150 | r = init_origin_hash(); | |
1151 | if (r) { | |
1152 | DMERR("init_origin_hash failed."); | |
1153 | goto bad2; | |
1154 | } | |
1155 | ||
1156 | exception_cache = kmem_cache_create("dm-snapshot-ex", | |
1157 | sizeof(struct exception), | |
1158 | __alignof__(struct exception), | |
1159 | 0, NULL, NULL); | |
1160 | if (!exception_cache) { | |
1161 | DMERR("Couldn't create exception cache."); | |
1162 | r = -ENOMEM; | |
1163 | goto bad3; | |
1164 | } | |
1165 | ||
1166 | pending_cache = | |
1167 | kmem_cache_create("dm-snapshot-in", | |
1168 | sizeof(struct pending_exception), | |
1169 | __alignof__(struct pending_exception), | |
1170 | 0, NULL, NULL); | |
1171 | if (!pending_cache) { | |
1172 | DMERR("Couldn't create pending cache."); | |
1173 | r = -ENOMEM; | |
1174 | goto bad4; | |
1175 | } | |
1176 | ||
1177 | pending_pool = mempool_create(128, mempool_alloc_slab, | |
1178 | mempool_free_slab, pending_cache); | |
1179 | if (!pending_pool) { | |
1180 | DMERR("Couldn't create pending pool."); | |
1181 | r = -ENOMEM; | |
1182 | goto bad5; | |
1183 | } | |
1184 | ||
1185 | return 0; | |
1186 | ||
1187 | bad5: | |
1188 | kmem_cache_destroy(pending_cache); | |
1189 | bad4: | |
1190 | kmem_cache_destroy(exception_cache); | |
1191 | bad3: | |
1192 | exit_origin_hash(); | |
1193 | bad2: | |
1194 | dm_unregister_target(&origin_target); | |
1195 | bad1: | |
1196 | dm_unregister_target(&snapshot_target); | |
1197 | return r; | |
1198 | } | |
1199 | ||
1200 | static void __exit dm_snapshot_exit(void) | |
1201 | { | |
1202 | int r; | |
1203 | ||
1204 | r = dm_unregister_target(&snapshot_target); | |
1205 | if (r) | |
1206 | DMERR("snapshot unregister failed %d", r); | |
1207 | ||
1208 | r = dm_unregister_target(&origin_target); | |
1209 | if (r) | |
1210 | DMERR("origin unregister failed %d", r); | |
1211 | ||
1212 | exit_origin_hash(); | |
1213 | mempool_destroy(pending_pool); | |
1214 | kmem_cache_destroy(pending_cache); | |
1215 | kmem_cache_destroy(exception_cache); | |
1216 | } | |
1217 | ||
1218 | /* Module hooks */ | |
1219 | module_init(dm_snapshot_init); | |
1220 | module_exit(dm_snapshot_exit); | |
1221 | ||
1222 | MODULE_DESCRIPTION(DM_NAME " snapshot target"); | |
1223 | MODULE_AUTHOR("Joe Thornber"); | |
1224 | MODULE_LICENSE("GPL"); |