]>
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 LT |
99 | static mempool_t *pending_pool; |
100 | ||
cd45daff MP |
101 | struct dm_snap_tracked_chunk { |
102 | struct hlist_node node; | |
103 | chunk_t chunk; | |
104 | }; | |
105 | ||
106 | static struct kmem_cache *tracked_chunk_cache; | |
107 | ||
108 | static struct dm_snap_tracked_chunk *track_chunk(struct dm_snapshot *s, | |
109 | chunk_t chunk) | |
110 | { | |
111 | struct dm_snap_tracked_chunk *c = mempool_alloc(s->tracked_chunk_pool, | |
112 | GFP_NOIO); | |
113 | unsigned long flags; | |
114 | ||
115 | c->chunk = chunk; | |
116 | ||
117 | spin_lock_irqsave(&s->tracked_chunk_lock, flags); | |
118 | hlist_add_head(&c->node, | |
119 | &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)]); | |
120 | spin_unlock_irqrestore(&s->tracked_chunk_lock, flags); | |
121 | ||
122 | return c; | |
123 | } | |
124 | ||
125 | static void stop_tracking_chunk(struct dm_snapshot *s, | |
126 | struct dm_snap_tracked_chunk *c) | |
127 | { | |
128 | unsigned long flags; | |
129 | ||
130 | spin_lock_irqsave(&s->tracked_chunk_lock, flags); | |
131 | hlist_del(&c->node); | |
132 | spin_unlock_irqrestore(&s->tracked_chunk_lock, flags); | |
133 | ||
134 | mempool_free(c, s->tracked_chunk_pool); | |
135 | } | |
136 | ||
a8d41b59 MP |
137 | static int __chunk_is_tracked(struct dm_snapshot *s, chunk_t chunk) |
138 | { | |
139 | struct dm_snap_tracked_chunk *c; | |
140 | struct hlist_node *hn; | |
141 | int found = 0; | |
142 | ||
143 | spin_lock_irq(&s->tracked_chunk_lock); | |
144 | ||
145 | hlist_for_each_entry(c, hn, | |
146 | &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)], node) { | |
147 | if (c->chunk == chunk) { | |
148 | found = 1; | |
149 | break; | |
150 | } | |
151 | } | |
152 | ||
153 | spin_unlock_irq(&s->tracked_chunk_lock); | |
154 | ||
155 | return found; | |
156 | } | |
157 | ||
1da177e4 LT |
158 | /* |
159 | * One of these per registered origin, held in the snapshot_origins hash | |
160 | */ | |
161 | struct origin { | |
162 | /* The origin device */ | |
163 | struct block_device *bdev; | |
164 | ||
165 | struct list_head hash_list; | |
166 | ||
167 | /* List of snapshots for this origin */ | |
168 | struct list_head snapshots; | |
169 | }; | |
170 | ||
171 | /* | |
172 | * Size of the hash table for origin volumes. If we make this | |
173 | * the size of the minors list then it should be nearly perfect | |
174 | */ | |
175 | #define ORIGIN_HASH_SIZE 256 | |
176 | #define ORIGIN_MASK 0xFF | |
177 | static struct list_head *_origins; | |
178 | static struct rw_semaphore _origins_lock; | |
179 | ||
180 | static int init_origin_hash(void) | |
181 | { | |
182 | int i; | |
183 | ||
184 | _origins = kmalloc(ORIGIN_HASH_SIZE * sizeof(struct list_head), | |
185 | GFP_KERNEL); | |
186 | if (!_origins) { | |
72d94861 | 187 | DMERR("unable to allocate memory"); |
1da177e4 LT |
188 | return -ENOMEM; |
189 | } | |
190 | ||
191 | for (i = 0; i < ORIGIN_HASH_SIZE; i++) | |
192 | INIT_LIST_HEAD(_origins + i); | |
193 | init_rwsem(&_origins_lock); | |
194 | ||
195 | return 0; | |
196 | } | |
197 | ||
198 | static void exit_origin_hash(void) | |
199 | { | |
200 | kfree(_origins); | |
201 | } | |
202 | ||
028867ac | 203 | static unsigned origin_hash(struct block_device *bdev) |
1da177e4 LT |
204 | { |
205 | return bdev->bd_dev & ORIGIN_MASK; | |
206 | } | |
207 | ||
208 | static struct origin *__lookup_origin(struct block_device *origin) | |
209 | { | |
210 | struct list_head *ol; | |
211 | struct origin *o; | |
212 | ||
213 | ol = &_origins[origin_hash(origin)]; | |
214 | list_for_each_entry (o, ol, hash_list) | |
215 | if (bdev_equal(o->bdev, origin)) | |
216 | return o; | |
217 | ||
218 | return NULL; | |
219 | } | |
220 | ||
221 | static void __insert_origin(struct origin *o) | |
222 | { | |
223 | struct list_head *sl = &_origins[origin_hash(o->bdev)]; | |
224 | list_add_tail(&o->hash_list, sl); | |
225 | } | |
226 | ||
227 | /* | |
228 | * Make a note of the snapshot and its origin so we can look it | |
229 | * up when the origin has a write on it. | |
230 | */ | |
231 | static int register_snapshot(struct dm_snapshot *snap) | |
232 | { | |
233 | struct origin *o; | |
234 | struct block_device *bdev = snap->origin->bdev; | |
235 | ||
236 | down_write(&_origins_lock); | |
237 | o = __lookup_origin(bdev); | |
238 | ||
239 | if (!o) { | |
240 | /* New origin */ | |
241 | o = kmalloc(sizeof(*o), GFP_KERNEL); | |
242 | if (!o) { | |
243 | up_write(&_origins_lock); | |
244 | return -ENOMEM; | |
245 | } | |
246 | ||
247 | /* Initialise the struct */ | |
248 | INIT_LIST_HEAD(&o->snapshots); | |
249 | o->bdev = bdev; | |
250 | ||
251 | __insert_origin(o); | |
252 | } | |
253 | ||
254 | list_add_tail(&snap->list, &o->snapshots); | |
255 | ||
256 | up_write(&_origins_lock); | |
257 | return 0; | |
258 | } | |
259 | ||
260 | static void unregister_snapshot(struct dm_snapshot *s) | |
261 | { | |
262 | struct origin *o; | |
263 | ||
264 | down_write(&_origins_lock); | |
265 | o = __lookup_origin(s->origin->bdev); | |
266 | ||
267 | list_del(&s->list); | |
268 | if (list_empty(&o->snapshots)) { | |
269 | list_del(&o->hash_list); | |
270 | kfree(o); | |
271 | } | |
272 | ||
273 | up_write(&_origins_lock); | |
274 | } | |
275 | ||
276 | /* | |
277 | * Implementation of the exception hash tables. | |
d74f81f8 MB |
278 | * The lowest hash_shift bits of the chunk number are ignored, allowing |
279 | * some consecutive chunks to be grouped together. | |
1da177e4 | 280 | */ |
d74f81f8 MB |
281 | static int init_exception_table(struct exception_table *et, uint32_t size, |
282 | unsigned hash_shift) | |
1da177e4 LT |
283 | { |
284 | unsigned int i; | |
285 | ||
d74f81f8 | 286 | et->hash_shift = hash_shift; |
1da177e4 LT |
287 | et->hash_mask = size - 1; |
288 | et->table = dm_vcalloc(size, sizeof(struct list_head)); | |
289 | if (!et->table) | |
290 | return -ENOMEM; | |
291 | ||
292 | for (i = 0; i < size; i++) | |
293 | INIT_LIST_HEAD(et->table + i); | |
294 | ||
295 | return 0; | |
296 | } | |
297 | ||
e18b890b | 298 | static void exit_exception_table(struct exception_table *et, struct kmem_cache *mem) |
1da177e4 LT |
299 | { |
300 | struct list_head *slot; | |
028867ac | 301 | struct dm_snap_exception *ex, *next; |
1da177e4 LT |
302 | int i, size; |
303 | ||
304 | size = et->hash_mask + 1; | |
305 | for (i = 0; i < size; i++) { | |
306 | slot = et->table + i; | |
307 | ||
308 | list_for_each_entry_safe (ex, next, slot, hash_list) | |
309 | kmem_cache_free(mem, ex); | |
310 | } | |
311 | ||
312 | vfree(et->table); | |
313 | } | |
314 | ||
028867ac | 315 | static uint32_t exception_hash(struct exception_table *et, chunk_t chunk) |
1da177e4 | 316 | { |
d74f81f8 | 317 | return (chunk >> et->hash_shift) & et->hash_mask; |
1da177e4 LT |
318 | } |
319 | ||
028867ac AK |
320 | static void insert_exception(struct exception_table *eh, |
321 | struct dm_snap_exception *e) | |
1da177e4 LT |
322 | { |
323 | struct list_head *l = &eh->table[exception_hash(eh, e->old_chunk)]; | |
324 | list_add(&e->hash_list, l); | |
325 | } | |
326 | ||
028867ac | 327 | static void remove_exception(struct dm_snap_exception *e) |
1da177e4 LT |
328 | { |
329 | list_del(&e->hash_list); | |
330 | } | |
331 | ||
332 | /* | |
333 | * Return the exception data for a sector, or NULL if not | |
334 | * remapped. | |
335 | */ | |
028867ac AK |
336 | static struct dm_snap_exception *lookup_exception(struct exception_table *et, |
337 | chunk_t chunk) | |
1da177e4 LT |
338 | { |
339 | struct list_head *slot; | |
028867ac | 340 | struct dm_snap_exception *e; |
1da177e4 LT |
341 | |
342 | slot = &et->table[exception_hash(et, chunk)]; | |
343 | list_for_each_entry (e, slot, hash_list) | |
d74f81f8 MB |
344 | if (chunk >= e->old_chunk && |
345 | chunk <= e->old_chunk + dm_consecutive_chunk_count(e)) | |
1da177e4 LT |
346 | return e; |
347 | ||
348 | return NULL; | |
349 | } | |
350 | ||
028867ac | 351 | static struct dm_snap_exception *alloc_exception(void) |
1da177e4 | 352 | { |
028867ac | 353 | struct dm_snap_exception *e; |
1da177e4 LT |
354 | |
355 | e = kmem_cache_alloc(exception_cache, GFP_NOIO); | |
356 | if (!e) | |
357 | e = kmem_cache_alloc(exception_cache, GFP_ATOMIC); | |
358 | ||
359 | return e; | |
360 | } | |
361 | ||
028867ac | 362 | static void free_exception(struct dm_snap_exception *e) |
1da177e4 LT |
363 | { |
364 | kmem_cache_free(exception_cache, e); | |
365 | } | |
366 | ||
028867ac | 367 | static struct dm_snap_pending_exception *alloc_pending_exception(void) |
1da177e4 LT |
368 | { |
369 | return mempool_alloc(pending_pool, GFP_NOIO); | |
370 | } | |
371 | ||
028867ac | 372 | static void free_pending_exception(struct dm_snap_pending_exception *pe) |
1da177e4 LT |
373 | { |
374 | mempool_free(pe, pending_pool); | |
375 | } | |
376 | ||
d74f81f8 MB |
377 | static void insert_completed_exception(struct dm_snapshot *s, |
378 | struct dm_snap_exception *new_e) | |
379 | { | |
380 | struct exception_table *eh = &s->complete; | |
381 | struct list_head *l; | |
382 | struct dm_snap_exception *e = NULL; | |
383 | ||
384 | l = &eh->table[exception_hash(eh, new_e->old_chunk)]; | |
385 | ||
386 | /* Add immediately if this table doesn't support consecutive chunks */ | |
387 | if (!eh->hash_shift) | |
388 | goto out; | |
389 | ||
390 | /* List is ordered by old_chunk */ | |
391 | list_for_each_entry_reverse(e, l, hash_list) { | |
392 | /* Insert after an existing chunk? */ | |
393 | if (new_e->old_chunk == (e->old_chunk + | |
394 | dm_consecutive_chunk_count(e) + 1) && | |
395 | new_e->new_chunk == (dm_chunk_number(e->new_chunk) + | |
396 | dm_consecutive_chunk_count(e) + 1)) { | |
397 | dm_consecutive_chunk_count_inc(e); | |
398 | free_exception(new_e); | |
399 | return; | |
400 | } | |
401 | ||
402 | /* Insert before an existing chunk? */ | |
403 | if (new_e->old_chunk == (e->old_chunk - 1) && | |
404 | new_e->new_chunk == (dm_chunk_number(e->new_chunk) - 1)) { | |
405 | dm_consecutive_chunk_count_inc(e); | |
406 | e->old_chunk--; | |
407 | e->new_chunk--; | |
408 | free_exception(new_e); | |
409 | return; | |
410 | } | |
411 | ||
412 | if (new_e->old_chunk > e->old_chunk) | |
413 | break; | |
414 | } | |
415 | ||
416 | out: | |
417 | list_add(&new_e->hash_list, e ? &e->hash_list : l); | |
418 | } | |
419 | ||
1da177e4 LT |
420 | int dm_add_exception(struct dm_snapshot *s, chunk_t old, chunk_t new) |
421 | { | |
028867ac | 422 | struct dm_snap_exception *e; |
1da177e4 LT |
423 | |
424 | e = alloc_exception(); | |
425 | if (!e) | |
426 | return -ENOMEM; | |
427 | ||
428 | e->old_chunk = old; | |
d74f81f8 MB |
429 | |
430 | /* Consecutive_count is implicitly initialised to zero */ | |
1da177e4 | 431 | e->new_chunk = new; |
d74f81f8 MB |
432 | |
433 | insert_completed_exception(s, e); | |
434 | ||
1da177e4 LT |
435 | return 0; |
436 | } | |
437 | ||
438 | /* | |
439 | * Hard coded magic. | |
440 | */ | |
441 | static int calc_max_buckets(void) | |
442 | { | |
443 | /* use a fixed size of 2MB */ | |
444 | unsigned long mem = 2 * 1024 * 1024; | |
445 | mem /= sizeof(struct list_head); | |
446 | ||
447 | return mem; | |
448 | } | |
449 | ||
1da177e4 LT |
450 | /* |
451 | * Allocate room for a suitable hash table. | |
452 | */ | |
453 | static int init_hash_tables(struct dm_snapshot *s) | |
454 | { | |
455 | sector_t hash_size, cow_dev_size, origin_dev_size, max_buckets; | |
456 | ||
457 | /* | |
458 | * Calculate based on the size of the original volume or | |
459 | * the COW volume... | |
460 | */ | |
461 | cow_dev_size = get_dev_size(s->cow->bdev); | |
462 | origin_dev_size = get_dev_size(s->origin->bdev); | |
463 | max_buckets = calc_max_buckets(); | |
464 | ||
465 | hash_size = min(origin_dev_size, cow_dev_size) >> s->chunk_shift; | |
466 | hash_size = min(hash_size, max_buckets); | |
467 | ||
8defd830 | 468 | hash_size = rounddown_pow_of_two(hash_size); |
d74f81f8 MB |
469 | if (init_exception_table(&s->complete, hash_size, |
470 | DM_CHUNK_CONSECUTIVE_BITS)) | |
1da177e4 LT |
471 | return -ENOMEM; |
472 | ||
473 | /* | |
474 | * Allocate hash table for in-flight exceptions | |
475 | * Make this smaller than the real hash table | |
476 | */ | |
477 | hash_size >>= 3; | |
478 | if (hash_size < 64) | |
479 | hash_size = 64; | |
480 | ||
d74f81f8 | 481 | if (init_exception_table(&s->pending, hash_size, 0)) { |
1da177e4 LT |
482 | exit_exception_table(&s->complete, exception_cache); |
483 | return -ENOMEM; | |
484 | } | |
485 | ||
486 | return 0; | |
487 | } | |
488 | ||
489 | /* | |
490 | * Round a number up to the nearest 'size' boundary. size must | |
491 | * be a power of 2. | |
492 | */ | |
028867ac | 493 | static ulong round_up(ulong n, ulong size) |
1da177e4 LT |
494 | { |
495 | size--; | |
496 | return (n + size) & ~size; | |
497 | } | |
498 | ||
4c7e3bf4 MM |
499 | static int set_chunk_size(struct dm_snapshot *s, const char *chunk_size_arg, |
500 | char **error) | |
501 | { | |
502 | unsigned long chunk_size; | |
503 | char *value; | |
504 | ||
505 | chunk_size = simple_strtoul(chunk_size_arg, &value, 10); | |
506 | if (*chunk_size_arg == '\0' || *value != '\0') { | |
507 | *error = "Invalid chunk size"; | |
508 | return -EINVAL; | |
509 | } | |
510 | ||
511 | if (!chunk_size) { | |
512 | s->chunk_size = s->chunk_mask = s->chunk_shift = 0; | |
513 | return 0; | |
514 | } | |
515 | ||
516 | /* | |
517 | * Chunk size must be multiple of page size. Silently | |
518 | * round up if it's not. | |
519 | */ | |
520 | chunk_size = round_up(chunk_size, PAGE_SIZE >> 9); | |
521 | ||
522 | /* Check chunk_size is a power of 2 */ | |
6f3c3f0a | 523 | if (!is_power_of_2(chunk_size)) { |
4c7e3bf4 MM |
524 | *error = "Chunk size is not a power of 2"; |
525 | return -EINVAL; | |
526 | } | |
527 | ||
528 | /* Validate the chunk size against the device block size */ | |
529 | if (chunk_size % (bdev_hardsect_size(s->cow->bdev) >> 9)) { | |
530 | *error = "Chunk size is not a multiple of device blocksize"; | |
531 | return -EINVAL; | |
532 | } | |
533 | ||
534 | s->chunk_size = chunk_size; | |
535 | s->chunk_mask = chunk_size - 1; | |
536 | s->chunk_shift = ffs(chunk_size) - 1; | |
537 | ||
538 | return 0; | |
539 | } | |
540 | ||
1da177e4 LT |
541 | /* |
542 | * Construct a snapshot mapping: <origin_dev> <COW-dev> <p/n> <chunk-size> | |
543 | */ | |
544 | static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv) | |
545 | { | |
546 | struct dm_snapshot *s; | |
cd45daff | 547 | int i; |
1da177e4 LT |
548 | int r = -EINVAL; |
549 | char persistent; | |
550 | char *origin_path; | |
551 | char *cow_path; | |
1da177e4 | 552 | |
4c7e3bf4 | 553 | if (argc != 4) { |
72d94861 | 554 | ti->error = "requires exactly 4 arguments"; |
1da177e4 LT |
555 | r = -EINVAL; |
556 | goto bad1; | |
557 | } | |
558 | ||
559 | origin_path = argv[0]; | |
560 | cow_path = argv[1]; | |
561 | persistent = toupper(*argv[2]); | |
562 | ||
563 | if (persistent != 'P' && persistent != 'N') { | |
564 | ti->error = "Persistent flag is not P or N"; | |
565 | r = -EINVAL; | |
566 | goto bad1; | |
567 | } | |
568 | ||
1da177e4 LT |
569 | s = kmalloc(sizeof(*s), GFP_KERNEL); |
570 | if (s == NULL) { | |
571 | ti->error = "Cannot allocate snapshot context private " | |
572 | "structure"; | |
573 | r = -ENOMEM; | |
574 | goto bad1; | |
575 | } | |
576 | ||
577 | r = dm_get_device(ti, origin_path, 0, ti->len, FMODE_READ, &s->origin); | |
578 | if (r) { | |
579 | ti->error = "Cannot get origin device"; | |
580 | goto bad2; | |
581 | } | |
582 | ||
583 | r = dm_get_device(ti, cow_path, 0, 0, | |
584 | FMODE_READ | FMODE_WRITE, &s->cow); | |
585 | if (r) { | |
586 | dm_put_device(ti, s->origin); | |
587 | ti->error = "Cannot get COW device"; | |
588 | goto bad2; | |
589 | } | |
590 | ||
4c7e3bf4 MM |
591 | r = set_chunk_size(s, argv[3], &ti->error); |
592 | if (r) | |
1da177e4 | 593 | goto bad3; |
1da177e4 | 594 | |
1da177e4 | 595 | s->type = persistent; |
1da177e4 LT |
596 | |
597 | s->valid = 1; | |
aa14edeb | 598 | s->active = 0; |
1da177e4 LT |
599 | s->last_percent = 0; |
600 | init_rwsem(&s->lock); | |
ca3a931f | 601 | spin_lock_init(&s->pe_lock); |
72727bad | 602 | s->ti = ti; |
1da177e4 LT |
603 | |
604 | /* Allocate hash table for COW data */ | |
605 | if (init_hash_tables(s)) { | |
606 | ti->error = "Unable to allocate hash table space"; | |
607 | r = -ENOMEM; | |
608 | goto bad3; | |
609 | } | |
610 | ||
1da177e4 LT |
611 | s->store.snap = s; |
612 | ||
613 | if (persistent == 'P') | |
4c7e3bf4 | 614 | r = dm_create_persistent(&s->store); |
1da177e4 | 615 | else |
4c7e3bf4 | 616 | r = dm_create_transient(&s->store); |
1da177e4 LT |
617 | |
618 | if (r) { | |
619 | ti->error = "Couldn't create exception store"; | |
620 | r = -EINVAL; | |
621 | goto bad4; | |
622 | } | |
623 | ||
eb69aca5 | 624 | r = dm_kcopyd_client_create(SNAPSHOT_PAGES, &s->kcopyd_client); |
1da177e4 LT |
625 | if (r) { |
626 | ti->error = "Could not create kcopyd client"; | |
627 | goto bad5; | |
628 | } | |
629 | ||
cd45daff MP |
630 | s->tracked_chunk_pool = mempool_create_slab_pool(MIN_IOS, |
631 | tracked_chunk_cache); | |
632 | if (!s->tracked_chunk_pool) { | |
633 | ti->error = "Could not allocate tracked_chunk mempool for " | |
634 | "tracking reads"; | |
635 | goto bad6; | |
636 | } | |
637 | ||
638 | for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++) | |
639 | INIT_HLIST_HEAD(&s->tracked_chunk_hash[i]); | |
640 | ||
641 | spin_lock_init(&s->tracked_chunk_lock); | |
642 | ||
aa14edeb | 643 | /* Metadata must only be loaded into one table at once */ |
f9cea4f7 | 644 | r = s->store.read_metadata(&s->store); |
0764147b | 645 | if (r < 0) { |
f9cea4f7 | 646 | ti->error = "Failed to read snapshot metadata"; |
cd45daff | 647 | goto bad_load_and_register; |
0764147b MB |
648 | } else if (r > 0) { |
649 | s->valid = 0; | |
650 | DMWARN("Snapshot is marked invalid."); | |
f9cea4f7 | 651 | } |
aa14edeb | 652 | |
ca3a931f | 653 | bio_list_init(&s->queued_bios); |
c4028958 | 654 | INIT_WORK(&s->queued_bios_work, flush_queued_bios); |
ca3a931f | 655 | |
1da177e4 | 656 | /* Add snapshot to the list of snapshots for this origin */ |
aa14edeb | 657 | /* Exceptions aren't triggered till snapshot_resume() is called */ |
1da177e4 LT |
658 | if (register_snapshot(s)) { |
659 | r = -EINVAL; | |
660 | ti->error = "Cannot register snapshot origin"; | |
cd45daff | 661 | goto bad_load_and_register; |
1da177e4 LT |
662 | } |
663 | ||
664 | ti->private = s; | |
c51c2752 | 665 | ti->split_io = s->chunk_size; |
1da177e4 LT |
666 | |
667 | return 0; | |
668 | ||
cd45daff MP |
669 | bad_load_and_register: |
670 | mempool_destroy(s->tracked_chunk_pool); | |
671 | ||
1da177e4 | 672 | bad6: |
eb69aca5 | 673 | dm_kcopyd_client_destroy(s->kcopyd_client); |
1da177e4 LT |
674 | |
675 | bad5: | |
676 | s->store.destroy(&s->store); | |
677 | ||
678 | bad4: | |
679 | exit_exception_table(&s->pending, pending_cache); | |
680 | exit_exception_table(&s->complete, exception_cache); | |
681 | ||
682 | bad3: | |
683 | dm_put_device(ti, s->cow); | |
684 | dm_put_device(ti, s->origin); | |
685 | ||
686 | bad2: | |
687 | kfree(s); | |
688 | ||
689 | bad1: | |
690 | return r; | |
691 | } | |
692 | ||
31c93a0c MB |
693 | static void __free_exceptions(struct dm_snapshot *s) |
694 | { | |
eb69aca5 | 695 | dm_kcopyd_client_destroy(s->kcopyd_client); |
31c93a0c MB |
696 | s->kcopyd_client = NULL; |
697 | ||
698 | exit_exception_table(&s->pending, pending_cache); | |
699 | exit_exception_table(&s->complete, exception_cache); | |
700 | ||
701 | s->store.destroy(&s->store); | |
702 | } | |
703 | ||
1da177e4 LT |
704 | static void snapshot_dtr(struct dm_target *ti) |
705 | { | |
cd45daff MP |
706 | #ifdef CONFIG_DM_DEBUG |
707 | int i; | |
708 | #endif | |
028867ac | 709 | struct dm_snapshot *s = ti->private; |
1da177e4 | 710 | |
ca3a931f AK |
711 | flush_workqueue(ksnapd); |
712 | ||
138728dc AK |
713 | /* Prevent further origin writes from using this snapshot. */ |
714 | /* After this returns there can be no new kcopyd jobs. */ | |
1da177e4 LT |
715 | unregister_snapshot(s); |
716 | ||
cd45daff MP |
717 | #ifdef CONFIG_DM_DEBUG |
718 | for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++) | |
719 | BUG_ON(!hlist_empty(&s->tracked_chunk_hash[i])); | |
720 | #endif | |
721 | ||
722 | mempool_destroy(s->tracked_chunk_pool); | |
723 | ||
31c93a0c | 724 | __free_exceptions(s); |
1da177e4 LT |
725 | |
726 | dm_put_device(ti, s->origin); | |
727 | dm_put_device(ti, s->cow); | |
138728dc | 728 | |
1da177e4 LT |
729 | kfree(s); |
730 | } | |
731 | ||
732 | /* | |
733 | * Flush a list of buffers. | |
734 | */ | |
735 | static void flush_bios(struct bio *bio) | |
736 | { | |
737 | struct bio *n; | |
738 | ||
739 | while (bio) { | |
740 | n = bio->bi_next; | |
741 | bio->bi_next = NULL; | |
742 | generic_make_request(bio); | |
743 | bio = n; | |
744 | } | |
745 | } | |
746 | ||
c4028958 | 747 | static void flush_queued_bios(struct work_struct *work) |
ca3a931f | 748 | { |
c4028958 DH |
749 | struct dm_snapshot *s = |
750 | container_of(work, struct dm_snapshot, queued_bios_work); | |
ca3a931f AK |
751 | struct bio *queued_bios; |
752 | unsigned long flags; | |
753 | ||
754 | spin_lock_irqsave(&s->pe_lock, flags); | |
755 | queued_bios = bio_list_get(&s->queued_bios); | |
756 | spin_unlock_irqrestore(&s->pe_lock, flags); | |
757 | ||
758 | flush_bios(queued_bios); | |
759 | } | |
760 | ||
1da177e4 LT |
761 | /* |
762 | * Error a list of buffers. | |
763 | */ | |
764 | static void error_bios(struct bio *bio) | |
765 | { | |
766 | struct bio *n; | |
767 | ||
768 | while (bio) { | |
769 | n = bio->bi_next; | |
770 | bio->bi_next = NULL; | |
6712ecf8 | 771 | bio_io_error(bio); |
1da177e4 LT |
772 | bio = n; |
773 | } | |
774 | } | |
775 | ||
695368ac | 776 | static void __invalidate_snapshot(struct dm_snapshot *s, int err) |
76df1c65 AK |
777 | { |
778 | if (!s->valid) | |
779 | return; | |
780 | ||
781 | if (err == -EIO) | |
782 | DMERR("Invalidating snapshot: Error reading/writing."); | |
783 | else if (err == -ENOMEM) | |
784 | DMERR("Invalidating snapshot: Unable to allocate exception."); | |
785 | ||
76df1c65 AK |
786 | if (s->store.drop_snapshot) |
787 | s->store.drop_snapshot(&s->store); | |
788 | ||
789 | s->valid = 0; | |
790 | ||
72727bad | 791 | dm_table_event(s->ti->table); |
76df1c65 AK |
792 | } |
793 | ||
028867ac | 794 | static void get_pending_exception(struct dm_snap_pending_exception *pe) |
4b832e8d AK |
795 | { |
796 | atomic_inc(&pe->ref_count); | |
797 | } | |
798 | ||
028867ac | 799 | static struct bio *put_pending_exception(struct dm_snap_pending_exception *pe) |
4b832e8d | 800 | { |
028867ac | 801 | struct dm_snap_pending_exception *primary_pe; |
4b832e8d AK |
802 | struct bio *origin_bios = NULL; |
803 | ||
804 | primary_pe = pe->primary_pe; | |
805 | ||
806 | /* | |
807 | * If this pe is involved in a write to the origin and | |
808 | * it is the last sibling to complete then release | |
809 | * the bios for the original write to the origin. | |
810 | */ | |
811 | if (primary_pe && | |
812 | atomic_dec_and_test(&primary_pe->ref_count)) | |
813 | origin_bios = bio_list_get(&primary_pe->origin_bios); | |
814 | ||
815 | /* | |
816 | * Free the pe if it's not linked to an origin write or if | |
817 | * it's not itself a primary pe. | |
818 | */ | |
819 | if (!primary_pe || primary_pe != pe) | |
820 | free_pending_exception(pe); | |
821 | ||
822 | /* | |
823 | * Free the primary pe if nothing references it. | |
824 | */ | |
825 | if (primary_pe && !atomic_read(&primary_pe->ref_count)) | |
826 | free_pending_exception(primary_pe); | |
827 | ||
828 | return origin_bios; | |
829 | } | |
830 | ||
028867ac | 831 | static void pending_complete(struct dm_snap_pending_exception *pe, int success) |
1da177e4 | 832 | { |
028867ac | 833 | struct dm_snap_exception *e; |
1da177e4 | 834 | struct dm_snapshot *s = pe->snap; |
9d493fa8 AK |
835 | struct bio *origin_bios = NULL; |
836 | struct bio *snapshot_bios = NULL; | |
837 | int error = 0; | |
1da177e4 | 838 | |
76df1c65 AK |
839 | if (!success) { |
840 | /* Read/write error - snapshot is unusable */ | |
1da177e4 | 841 | down_write(&s->lock); |
695368ac | 842 | __invalidate_snapshot(s, -EIO); |
9d493fa8 | 843 | error = 1; |
76df1c65 AK |
844 | goto out; |
845 | } | |
846 | ||
847 | e = alloc_exception(); | |
848 | if (!e) { | |
1da177e4 | 849 | down_write(&s->lock); |
695368ac | 850 | __invalidate_snapshot(s, -ENOMEM); |
9d493fa8 | 851 | error = 1; |
76df1c65 AK |
852 | goto out; |
853 | } | |
854 | *e = pe->e; | |
1da177e4 | 855 | |
76df1c65 AK |
856 | down_write(&s->lock); |
857 | if (!s->valid) { | |
76df1c65 | 858 | free_exception(e); |
9d493fa8 | 859 | error = 1; |
76df1c65 | 860 | goto out; |
1da177e4 LT |
861 | } |
862 | ||
a8d41b59 MP |
863 | /* |
864 | * Check for conflicting reads. This is extremely improbable, | |
865 | * so yield() is sufficient and there is no need for a wait queue. | |
866 | */ | |
867 | while (__chunk_is_tracked(s, pe->e.old_chunk)) | |
868 | yield(); | |
869 | ||
9d493fa8 AK |
870 | /* |
871 | * Add a proper exception, and remove the | |
872 | * in-flight exception from the list. | |
873 | */ | |
d74f81f8 | 874 | insert_completed_exception(s, e); |
76df1c65 | 875 | |
1da177e4 | 876 | out: |
695368ac | 877 | remove_exception(&pe->e); |
9d493fa8 | 878 | snapshot_bios = bio_list_get(&pe->snapshot_bios); |
4b832e8d | 879 | origin_bios = put_pending_exception(pe); |
1da177e4 | 880 | |
9d493fa8 AK |
881 | up_write(&s->lock); |
882 | ||
883 | /* Submit any pending write bios */ | |
884 | if (error) | |
885 | error_bios(snapshot_bios); | |
886 | else | |
887 | flush_bios(snapshot_bios); | |
888 | ||
889 | flush_bios(origin_bios); | |
1da177e4 LT |
890 | } |
891 | ||
892 | static void commit_callback(void *context, int success) | |
893 | { | |
028867ac AK |
894 | struct dm_snap_pending_exception *pe = context; |
895 | ||
1da177e4 LT |
896 | pending_complete(pe, success); |
897 | } | |
898 | ||
899 | /* | |
900 | * Called when the copy I/O has finished. kcopyd actually runs | |
901 | * this code so don't block. | |
902 | */ | |
4cdc1d1f | 903 | static void copy_callback(int read_err, unsigned long write_err, void *context) |
1da177e4 | 904 | { |
028867ac | 905 | struct dm_snap_pending_exception *pe = context; |
1da177e4 LT |
906 | struct dm_snapshot *s = pe->snap; |
907 | ||
908 | if (read_err || write_err) | |
909 | pending_complete(pe, 0); | |
910 | ||
911 | else | |
912 | /* Update the metadata if we are persistent */ | |
913 | s->store.commit_exception(&s->store, &pe->e, commit_callback, | |
914 | pe); | |
915 | } | |
916 | ||
917 | /* | |
918 | * Dispatches the copy operation to kcopyd. | |
919 | */ | |
028867ac | 920 | static void start_copy(struct dm_snap_pending_exception *pe) |
1da177e4 LT |
921 | { |
922 | struct dm_snapshot *s = pe->snap; | |
22a1ceb1 | 923 | struct dm_io_region src, dest; |
1da177e4 LT |
924 | struct block_device *bdev = s->origin->bdev; |
925 | sector_t dev_size; | |
926 | ||
927 | dev_size = get_dev_size(bdev); | |
928 | ||
929 | src.bdev = bdev; | |
930 | src.sector = chunk_to_sector(s, pe->e.old_chunk); | |
931 | src.count = min(s->chunk_size, dev_size - src.sector); | |
932 | ||
933 | dest.bdev = s->cow->bdev; | |
934 | dest.sector = chunk_to_sector(s, pe->e.new_chunk); | |
935 | dest.count = src.count; | |
936 | ||
937 | /* Hand over to kcopyd */ | |
eb69aca5 | 938 | dm_kcopyd_copy(s->kcopyd_client, |
1da177e4 LT |
939 | &src, 1, &dest, 0, copy_callback, pe); |
940 | } | |
941 | ||
942 | /* | |
943 | * Looks to see if this snapshot already has a pending exception | |
944 | * for this chunk, otherwise it allocates a new one and inserts | |
945 | * it into the pending table. | |
946 | * | |
947 | * NOTE: a write lock must be held on snap->lock before calling | |
948 | * this. | |
949 | */ | |
028867ac | 950 | static struct dm_snap_pending_exception * |
1da177e4 LT |
951 | __find_pending_exception(struct dm_snapshot *s, struct bio *bio) |
952 | { | |
028867ac AK |
953 | struct dm_snap_exception *e; |
954 | struct dm_snap_pending_exception *pe; | |
1da177e4 LT |
955 | chunk_t chunk = sector_to_chunk(s, bio->bi_sector); |
956 | ||
957 | /* | |
958 | * Is there a pending exception for this already ? | |
959 | */ | |
960 | e = lookup_exception(&s->pending, chunk); | |
961 | if (e) { | |
962 | /* cast the exception to a pending exception */ | |
028867ac | 963 | pe = container_of(e, struct dm_snap_pending_exception, e); |
76df1c65 AK |
964 | goto out; |
965 | } | |
1da177e4 | 966 | |
76df1c65 AK |
967 | /* |
968 | * Create a new pending exception, we don't want | |
969 | * to hold the lock while we do this. | |
970 | */ | |
971 | up_write(&s->lock); | |
972 | pe = alloc_pending_exception(); | |
973 | down_write(&s->lock); | |
1da177e4 | 974 | |
76df1c65 AK |
975 | if (!s->valid) { |
976 | free_pending_exception(pe); | |
977 | return NULL; | |
978 | } | |
1da177e4 | 979 | |
76df1c65 AK |
980 | e = lookup_exception(&s->pending, chunk); |
981 | if (e) { | |
982 | free_pending_exception(pe); | |
028867ac | 983 | pe = container_of(e, struct dm_snap_pending_exception, e); |
76df1c65 | 984 | goto out; |
1da177e4 LT |
985 | } |
986 | ||
76df1c65 AK |
987 | pe->e.old_chunk = chunk; |
988 | bio_list_init(&pe->origin_bios); | |
989 | bio_list_init(&pe->snapshot_bios); | |
990 | pe->primary_pe = NULL; | |
4b832e8d | 991 | atomic_set(&pe->ref_count, 0); |
76df1c65 AK |
992 | pe->snap = s; |
993 | pe->started = 0; | |
994 | ||
995 | if (s->store.prepare_exception(&s->store, &pe->e)) { | |
996 | free_pending_exception(pe); | |
997 | return NULL; | |
998 | } | |
999 | ||
4b832e8d | 1000 | get_pending_exception(pe); |
76df1c65 AK |
1001 | insert_exception(&s->pending, &pe->e); |
1002 | ||
1003 | out: | |
1da177e4 LT |
1004 | return pe; |
1005 | } | |
1006 | ||
028867ac | 1007 | static void remap_exception(struct dm_snapshot *s, struct dm_snap_exception *e, |
d74f81f8 | 1008 | struct bio *bio, chunk_t chunk) |
1da177e4 LT |
1009 | { |
1010 | bio->bi_bdev = s->cow->bdev; | |
d74f81f8 MB |
1011 | bio->bi_sector = chunk_to_sector(s, dm_chunk_number(e->new_chunk) + |
1012 | (chunk - e->old_chunk)) + | |
1013 | (bio->bi_sector & s->chunk_mask); | |
1da177e4 LT |
1014 | } |
1015 | ||
1016 | static int snapshot_map(struct dm_target *ti, struct bio *bio, | |
1017 | union map_info *map_context) | |
1018 | { | |
028867ac AK |
1019 | struct dm_snap_exception *e; |
1020 | struct dm_snapshot *s = ti->private; | |
d2a7ad29 | 1021 | int r = DM_MAPIO_REMAPPED; |
1da177e4 | 1022 | chunk_t chunk; |
028867ac | 1023 | struct dm_snap_pending_exception *pe = NULL; |
1da177e4 LT |
1024 | |
1025 | chunk = sector_to_chunk(s, bio->bi_sector); | |
1026 | ||
1027 | /* Full snapshots are not usable */ | |
76df1c65 | 1028 | /* To get here the table must be live so s->active is always set. */ |
1da177e4 | 1029 | if (!s->valid) |
f6a80ea8 | 1030 | return -EIO; |
1da177e4 | 1031 | |
ba40a2aa AK |
1032 | /* FIXME: should only take write lock if we need |
1033 | * to copy an exception */ | |
1034 | down_write(&s->lock); | |
1035 | ||
1036 | if (!s->valid) { | |
1037 | r = -EIO; | |
1038 | goto out_unlock; | |
1039 | } | |
1040 | ||
1041 | /* If the block is already remapped - use that, else remap it */ | |
1042 | e = lookup_exception(&s->complete, chunk); | |
1043 | if (e) { | |
d74f81f8 | 1044 | remap_exception(s, e, bio, chunk); |
ba40a2aa AK |
1045 | goto out_unlock; |
1046 | } | |
1047 | ||
1da177e4 LT |
1048 | /* |
1049 | * Write to snapshot - higher level takes care of RW/RO | |
1050 | * flags so we should only get this if we are | |
1051 | * writeable. | |
1052 | */ | |
1053 | if (bio_rw(bio) == WRITE) { | |
76df1c65 AK |
1054 | pe = __find_pending_exception(s, bio); |
1055 | if (!pe) { | |
695368ac | 1056 | __invalidate_snapshot(s, -ENOMEM); |
76df1c65 AK |
1057 | r = -EIO; |
1058 | goto out_unlock; | |
1da177e4 LT |
1059 | } |
1060 | ||
d74f81f8 | 1061 | remap_exception(s, &pe->e, bio, chunk); |
76df1c65 AK |
1062 | bio_list_add(&pe->snapshot_bios, bio); |
1063 | ||
d2a7ad29 | 1064 | r = DM_MAPIO_SUBMITTED; |
ba40a2aa | 1065 | |
76df1c65 AK |
1066 | if (!pe->started) { |
1067 | /* this is protected by snap->lock */ | |
1068 | pe->started = 1; | |
ba40a2aa | 1069 | up_write(&s->lock); |
76df1c65 | 1070 | start_copy(pe); |
ba40a2aa AK |
1071 | goto out; |
1072 | } | |
cd45daff | 1073 | } else { |
ba40a2aa | 1074 | bio->bi_bdev = s->origin->bdev; |
cd45daff MP |
1075 | map_context->ptr = track_chunk(s, chunk); |
1076 | } | |
1da177e4 | 1077 | |
ba40a2aa AK |
1078 | out_unlock: |
1079 | up_write(&s->lock); | |
1080 | out: | |
1da177e4 LT |
1081 | return r; |
1082 | } | |
1083 | ||
cd45daff MP |
1084 | static int snapshot_end_io(struct dm_target *ti, struct bio *bio, |
1085 | int error, union map_info *map_context) | |
1086 | { | |
1087 | struct dm_snapshot *s = ti->private; | |
1088 | struct dm_snap_tracked_chunk *c = map_context->ptr; | |
1089 | ||
1090 | if (c) | |
1091 | stop_tracking_chunk(s, c); | |
1092 | ||
1093 | return 0; | |
1094 | } | |
1095 | ||
1da177e4 LT |
1096 | static void snapshot_resume(struct dm_target *ti) |
1097 | { | |
028867ac | 1098 | struct dm_snapshot *s = ti->private; |
1da177e4 | 1099 | |
aa14edeb AK |
1100 | down_write(&s->lock); |
1101 | s->active = 1; | |
1102 | up_write(&s->lock); | |
1da177e4 LT |
1103 | } |
1104 | ||
1105 | static int snapshot_status(struct dm_target *ti, status_type_t type, | |
1106 | char *result, unsigned int maxlen) | |
1107 | { | |
028867ac | 1108 | struct dm_snapshot *snap = ti->private; |
1da177e4 LT |
1109 | |
1110 | switch (type) { | |
1111 | case STATUSTYPE_INFO: | |
1112 | if (!snap->valid) | |
1113 | snprintf(result, maxlen, "Invalid"); | |
1114 | else { | |
1115 | if (snap->store.fraction_full) { | |
1116 | sector_t numerator, denominator; | |
1117 | snap->store.fraction_full(&snap->store, | |
1118 | &numerator, | |
1119 | &denominator); | |
4ee218cd AM |
1120 | snprintf(result, maxlen, "%llu/%llu", |
1121 | (unsigned long long)numerator, | |
1122 | (unsigned long long)denominator); | |
1da177e4 LT |
1123 | } |
1124 | else | |
1125 | snprintf(result, maxlen, "Unknown"); | |
1126 | } | |
1127 | break; | |
1128 | ||
1129 | case STATUSTYPE_TABLE: | |
1130 | /* | |
1131 | * kdevname returns a static pointer so we need | |
1132 | * to make private copies if the output is to | |
1133 | * make sense. | |
1134 | */ | |
4ee218cd | 1135 | snprintf(result, maxlen, "%s %s %c %llu", |
1da177e4 | 1136 | snap->origin->name, snap->cow->name, |
4ee218cd AM |
1137 | snap->type, |
1138 | (unsigned long long)snap->chunk_size); | |
1da177e4 LT |
1139 | break; |
1140 | } | |
1141 | ||
1142 | return 0; | |
1143 | } | |
1144 | ||
1145 | /*----------------------------------------------------------------- | |
1146 | * Origin methods | |
1147 | *---------------------------------------------------------------*/ | |
1da177e4 LT |
1148 | static int __origin_write(struct list_head *snapshots, struct bio *bio) |
1149 | { | |
d2a7ad29 | 1150 | int r = DM_MAPIO_REMAPPED, first = 0; |
1da177e4 | 1151 | struct dm_snapshot *snap; |
028867ac AK |
1152 | struct dm_snap_exception *e; |
1153 | struct dm_snap_pending_exception *pe, *next_pe, *primary_pe = NULL; | |
1da177e4 | 1154 | chunk_t chunk; |
eccf0817 | 1155 | LIST_HEAD(pe_queue); |
1da177e4 LT |
1156 | |
1157 | /* Do all the snapshots on this origin */ | |
1158 | list_for_each_entry (snap, snapshots, list) { | |
1159 | ||
76df1c65 AK |
1160 | down_write(&snap->lock); |
1161 | ||
aa14edeb AK |
1162 | /* Only deal with valid and active snapshots */ |
1163 | if (!snap->valid || !snap->active) | |
76df1c65 | 1164 | goto next_snapshot; |
1da177e4 | 1165 | |
d5e404c1 | 1166 | /* Nothing to do if writing beyond end of snapshot */ |
72727bad | 1167 | if (bio->bi_sector >= dm_table_get_size(snap->ti->table)) |
76df1c65 | 1168 | goto next_snapshot; |
1da177e4 LT |
1169 | |
1170 | /* | |
1171 | * Remember, different snapshots can have | |
1172 | * different chunk sizes. | |
1173 | */ | |
1174 | chunk = sector_to_chunk(snap, bio->bi_sector); | |
1175 | ||
1176 | /* | |
1177 | * Check exception table to see if block | |
1178 | * is already remapped in this snapshot | |
1179 | * and trigger an exception if not. | |
b4b610f6 | 1180 | * |
4b832e8d | 1181 | * ref_count is initialised to 1 so pending_complete() |
b4b610f6 | 1182 | * won't destroy the primary_pe while we're inside this loop. |
1da177e4 LT |
1183 | */ |
1184 | e = lookup_exception(&snap->complete, chunk); | |
76df1c65 AK |
1185 | if (e) |
1186 | goto next_snapshot; | |
1187 | ||
1188 | pe = __find_pending_exception(snap, bio); | |
1189 | if (!pe) { | |
695368ac | 1190 | __invalidate_snapshot(snap, -ENOMEM); |
76df1c65 AK |
1191 | goto next_snapshot; |
1192 | } | |
1193 | ||
1194 | if (!primary_pe) { | |
1195 | /* | |
1196 | * Either every pe here has same | |
1197 | * primary_pe or none has one yet. | |
1198 | */ | |
1199 | if (pe->primary_pe) | |
1200 | primary_pe = pe->primary_pe; | |
1201 | else { | |
1202 | primary_pe = pe; | |
1203 | first = 1; | |
1da177e4 | 1204 | } |
76df1c65 AK |
1205 | |
1206 | bio_list_add(&primary_pe->origin_bios, bio); | |
1207 | ||
d2a7ad29 | 1208 | r = DM_MAPIO_SUBMITTED; |
76df1c65 AK |
1209 | } |
1210 | ||
1211 | if (!pe->primary_pe) { | |
76df1c65 | 1212 | pe->primary_pe = primary_pe; |
4b832e8d | 1213 | get_pending_exception(primary_pe); |
76df1c65 AK |
1214 | } |
1215 | ||
1216 | if (!pe->started) { | |
1217 | pe->started = 1; | |
1218 | list_add_tail(&pe->list, &pe_queue); | |
1da177e4 LT |
1219 | } |
1220 | ||
76df1c65 | 1221 | next_snapshot: |
1da177e4 LT |
1222 | up_write(&snap->lock); |
1223 | } | |
1224 | ||
b4b610f6 | 1225 | if (!primary_pe) |
4b832e8d | 1226 | return r; |
b4b610f6 AK |
1227 | |
1228 | /* | |
1229 | * If this is the first time we're processing this chunk and | |
4b832e8d | 1230 | * ref_count is now 1 it means all the pending exceptions |
b4b610f6 AK |
1231 | * got completed while we were in the loop above, so it falls to |
1232 | * us here to remove the primary_pe and submit any origin_bios. | |
1233 | */ | |
1234 | ||
4b832e8d | 1235 | if (first && atomic_dec_and_test(&primary_pe->ref_count)) { |
b4b610f6 AK |
1236 | flush_bios(bio_list_get(&primary_pe->origin_bios)); |
1237 | free_pending_exception(primary_pe); | |
1238 | /* If we got here, pe_queue is necessarily empty. */ | |
4b832e8d | 1239 | return r; |
b4b610f6 AK |
1240 | } |
1241 | ||
1da177e4 LT |
1242 | /* |
1243 | * Now that we have a complete pe list we can start the copying. | |
1244 | */ | |
eccf0817 AK |
1245 | list_for_each_entry_safe(pe, next_pe, &pe_queue, list) |
1246 | start_copy(pe); | |
1da177e4 LT |
1247 | |
1248 | return r; | |
1249 | } | |
1250 | ||
1251 | /* | |
1252 | * Called on a write from the origin driver. | |
1253 | */ | |
1254 | static int do_origin(struct dm_dev *origin, struct bio *bio) | |
1255 | { | |
1256 | struct origin *o; | |
d2a7ad29 | 1257 | int r = DM_MAPIO_REMAPPED; |
1da177e4 LT |
1258 | |
1259 | down_read(&_origins_lock); | |
1260 | o = __lookup_origin(origin->bdev); | |
1261 | if (o) | |
1262 | r = __origin_write(&o->snapshots, bio); | |
1263 | up_read(&_origins_lock); | |
1264 | ||
1265 | return r; | |
1266 | } | |
1267 | ||
1268 | /* | |
1269 | * Origin: maps a linear range of a device, with hooks for snapshotting. | |
1270 | */ | |
1271 | ||
1272 | /* | |
1273 | * Construct an origin mapping: <dev_path> | |
1274 | * The context for an origin is merely a 'struct dm_dev *' | |
1275 | * pointing to the real device. | |
1276 | */ | |
1277 | static int origin_ctr(struct dm_target *ti, unsigned int argc, char **argv) | |
1278 | { | |
1279 | int r; | |
1280 | struct dm_dev *dev; | |
1281 | ||
1282 | if (argc != 1) { | |
72d94861 | 1283 | ti->error = "origin: incorrect number of arguments"; |
1da177e4 LT |
1284 | return -EINVAL; |
1285 | } | |
1286 | ||
1287 | r = dm_get_device(ti, argv[0], 0, ti->len, | |
1288 | dm_table_get_mode(ti->table), &dev); | |
1289 | if (r) { | |
1290 | ti->error = "Cannot get target device"; | |
1291 | return r; | |
1292 | } | |
1293 | ||
1294 | ti->private = dev; | |
1295 | return 0; | |
1296 | } | |
1297 | ||
1298 | static void origin_dtr(struct dm_target *ti) | |
1299 | { | |
028867ac | 1300 | struct dm_dev *dev = ti->private; |
1da177e4 LT |
1301 | dm_put_device(ti, dev); |
1302 | } | |
1303 | ||
1304 | static int origin_map(struct dm_target *ti, struct bio *bio, | |
1305 | union map_info *map_context) | |
1306 | { | |
028867ac | 1307 | struct dm_dev *dev = ti->private; |
1da177e4 LT |
1308 | bio->bi_bdev = dev->bdev; |
1309 | ||
1310 | /* Only tell snapshots if this is a write */ | |
d2a7ad29 | 1311 | return (bio_rw(bio) == WRITE) ? do_origin(dev, bio) : DM_MAPIO_REMAPPED; |
1da177e4 LT |
1312 | } |
1313 | ||
1314 | #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r)) | |
1315 | ||
1316 | /* | |
1317 | * Set the target "split_io" field to the minimum of all the snapshots' | |
1318 | * chunk sizes. | |
1319 | */ | |
1320 | static void origin_resume(struct dm_target *ti) | |
1321 | { | |
028867ac | 1322 | struct dm_dev *dev = ti->private; |
1da177e4 LT |
1323 | struct dm_snapshot *snap; |
1324 | struct origin *o; | |
1325 | chunk_t chunk_size = 0; | |
1326 | ||
1327 | down_read(&_origins_lock); | |
1328 | o = __lookup_origin(dev->bdev); | |
1329 | if (o) | |
1330 | list_for_each_entry (snap, &o->snapshots, list) | |
1331 | chunk_size = min_not_zero(chunk_size, snap->chunk_size); | |
1332 | up_read(&_origins_lock); | |
1333 | ||
1334 | ti->split_io = chunk_size; | |
1335 | } | |
1336 | ||
1337 | static int origin_status(struct dm_target *ti, status_type_t type, char *result, | |
1338 | unsigned int maxlen) | |
1339 | { | |
028867ac | 1340 | struct dm_dev *dev = ti->private; |
1da177e4 LT |
1341 | |
1342 | switch (type) { | |
1343 | case STATUSTYPE_INFO: | |
1344 | result[0] = '\0'; | |
1345 | break; | |
1346 | ||
1347 | case STATUSTYPE_TABLE: | |
1348 | snprintf(result, maxlen, "%s", dev->name); | |
1349 | break; | |
1350 | } | |
1351 | ||
1352 | return 0; | |
1353 | } | |
1354 | ||
1355 | static struct target_type origin_target = { | |
1356 | .name = "snapshot-origin", | |
d74f81f8 | 1357 | .version = {1, 6, 0}, |
1da177e4 LT |
1358 | .module = THIS_MODULE, |
1359 | .ctr = origin_ctr, | |
1360 | .dtr = origin_dtr, | |
1361 | .map = origin_map, | |
1362 | .resume = origin_resume, | |
1363 | .status = origin_status, | |
1364 | }; | |
1365 | ||
1366 | static struct target_type snapshot_target = { | |
1367 | .name = "snapshot", | |
d74f81f8 | 1368 | .version = {1, 6, 0}, |
1da177e4 LT |
1369 | .module = THIS_MODULE, |
1370 | .ctr = snapshot_ctr, | |
1371 | .dtr = snapshot_dtr, | |
1372 | .map = snapshot_map, | |
cd45daff | 1373 | .end_io = snapshot_end_io, |
1da177e4 LT |
1374 | .resume = snapshot_resume, |
1375 | .status = snapshot_status, | |
1376 | }; | |
1377 | ||
1378 | static int __init dm_snapshot_init(void) | |
1379 | { | |
1380 | int r; | |
1381 | ||
1382 | r = dm_register_target(&snapshot_target); | |
1383 | if (r) { | |
1384 | DMERR("snapshot target register failed %d", r); | |
1385 | return r; | |
1386 | } | |
1387 | ||
1388 | r = dm_register_target(&origin_target); | |
1389 | if (r < 0) { | |
72d94861 | 1390 | DMERR("Origin target register failed %d", r); |
1da177e4 LT |
1391 | goto bad1; |
1392 | } | |
1393 | ||
1394 | r = init_origin_hash(); | |
1395 | if (r) { | |
1396 | DMERR("init_origin_hash failed."); | |
1397 | goto bad2; | |
1398 | } | |
1399 | ||
028867ac | 1400 | exception_cache = KMEM_CACHE(dm_snap_exception, 0); |
1da177e4 LT |
1401 | if (!exception_cache) { |
1402 | DMERR("Couldn't create exception cache."); | |
1403 | r = -ENOMEM; | |
1404 | goto bad3; | |
1405 | } | |
1406 | ||
028867ac | 1407 | pending_cache = KMEM_CACHE(dm_snap_pending_exception, 0); |
1da177e4 LT |
1408 | if (!pending_cache) { |
1409 | DMERR("Couldn't create pending cache."); | |
1410 | r = -ENOMEM; | |
1411 | goto bad4; | |
1412 | } | |
1413 | ||
cd45daff MP |
1414 | tracked_chunk_cache = KMEM_CACHE(dm_snap_tracked_chunk, 0); |
1415 | if (!tracked_chunk_cache) { | |
1416 | DMERR("Couldn't create cache to track chunks in use."); | |
1417 | r = -ENOMEM; | |
1418 | goto bad5; | |
1419 | } | |
1420 | ||
93d2341c | 1421 | pending_pool = mempool_create_slab_pool(128, pending_cache); |
1da177e4 LT |
1422 | if (!pending_pool) { |
1423 | DMERR("Couldn't create pending pool."); | |
1424 | r = -ENOMEM; | |
cd45daff | 1425 | goto bad_pending_pool; |
1da177e4 LT |
1426 | } |
1427 | ||
ca3a931f AK |
1428 | ksnapd = create_singlethread_workqueue("ksnapd"); |
1429 | if (!ksnapd) { | |
1430 | DMERR("Failed to create ksnapd workqueue."); | |
1431 | r = -ENOMEM; | |
1432 | goto bad6; | |
1433 | } | |
1434 | ||
1da177e4 LT |
1435 | return 0; |
1436 | ||
ca3a931f AK |
1437 | bad6: |
1438 | mempool_destroy(pending_pool); | |
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(); | |
1469 | mempool_destroy(pending_pool); | |
1470 | kmem_cache_destroy(pending_cache); | |
1471 | kmem_cache_destroy(exception_cache); | |
cd45daff | 1472 | kmem_cache_destroy(tracked_chunk_cache); |
1da177e4 LT |
1473 | } |
1474 | ||
1475 | /* Module hooks */ | |
1476 | module_init(dm_snapshot_init); | |
1477 | module_exit(dm_snapshot_exit); | |
1478 | ||
1479 | MODULE_DESCRIPTION(DM_NAME " snapshot target"); | |
1480 | MODULE_AUTHOR("Joe Thornber"); | |
1481 | MODULE_LICENSE("GPL"); |