]>
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 | { | |
60c856c8 | 232 | struct origin *o, *new_o; |
1da177e4 LT |
233 | struct block_device *bdev = snap->origin->bdev; |
234 | ||
60c856c8 MP |
235 | new_o = kmalloc(sizeof(*new_o), GFP_KERNEL); |
236 | if (!new_o) | |
237 | return -ENOMEM; | |
238 | ||
1da177e4 LT |
239 | down_write(&_origins_lock); |
240 | o = __lookup_origin(bdev); | |
241 | ||
60c856c8 MP |
242 | if (o) |
243 | kfree(new_o); | |
244 | else { | |
1da177e4 | 245 | /* New origin */ |
60c856c8 | 246 | o = new_o; |
1da177e4 LT |
247 | |
248 | /* Initialise the struct */ | |
249 | INIT_LIST_HEAD(&o->snapshots); | |
250 | o->bdev = bdev; | |
251 | ||
252 | __insert_origin(o); | |
253 | } | |
254 | ||
255 | list_add_tail(&snap->list, &o->snapshots); | |
256 | ||
257 | up_write(&_origins_lock); | |
258 | return 0; | |
259 | } | |
260 | ||
261 | static void unregister_snapshot(struct dm_snapshot *s) | |
262 | { | |
263 | struct origin *o; | |
264 | ||
265 | down_write(&_origins_lock); | |
266 | o = __lookup_origin(s->origin->bdev); | |
267 | ||
268 | list_del(&s->list); | |
269 | if (list_empty(&o->snapshots)) { | |
270 | list_del(&o->hash_list); | |
271 | kfree(o); | |
272 | } | |
273 | ||
274 | up_write(&_origins_lock); | |
275 | } | |
276 | ||
277 | /* | |
278 | * Implementation of the exception hash tables. | |
d74f81f8 MB |
279 | * The lowest hash_shift bits of the chunk number are ignored, allowing |
280 | * some consecutive chunks to be grouped together. | |
1da177e4 | 281 | */ |
d74f81f8 MB |
282 | static int init_exception_table(struct exception_table *et, uint32_t size, |
283 | unsigned hash_shift) | |
1da177e4 LT |
284 | { |
285 | unsigned int i; | |
286 | ||
d74f81f8 | 287 | et->hash_shift = hash_shift; |
1da177e4 LT |
288 | et->hash_mask = size - 1; |
289 | et->table = dm_vcalloc(size, sizeof(struct list_head)); | |
290 | if (!et->table) | |
291 | return -ENOMEM; | |
292 | ||
293 | for (i = 0; i < size; i++) | |
294 | INIT_LIST_HEAD(et->table + i); | |
295 | ||
296 | return 0; | |
297 | } | |
298 | ||
e18b890b | 299 | static void exit_exception_table(struct exception_table *et, struct kmem_cache *mem) |
1da177e4 LT |
300 | { |
301 | struct list_head *slot; | |
028867ac | 302 | struct dm_snap_exception *ex, *next; |
1da177e4 LT |
303 | int i, size; |
304 | ||
305 | size = et->hash_mask + 1; | |
306 | for (i = 0; i < size; i++) { | |
307 | slot = et->table + i; | |
308 | ||
309 | list_for_each_entry_safe (ex, next, slot, hash_list) | |
310 | kmem_cache_free(mem, ex); | |
311 | } | |
312 | ||
313 | vfree(et->table); | |
314 | } | |
315 | ||
028867ac | 316 | static uint32_t exception_hash(struct exception_table *et, chunk_t chunk) |
1da177e4 | 317 | { |
d74f81f8 | 318 | return (chunk >> et->hash_shift) & et->hash_mask; |
1da177e4 LT |
319 | } |
320 | ||
028867ac AK |
321 | static void insert_exception(struct exception_table *eh, |
322 | struct dm_snap_exception *e) | |
1da177e4 LT |
323 | { |
324 | struct list_head *l = &eh->table[exception_hash(eh, e->old_chunk)]; | |
325 | list_add(&e->hash_list, l); | |
326 | } | |
327 | ||
028867ac | 328 | static void remove_exception(struct dm_snap_exception *e) |
1da177e4 LT |
329 | { |
330 | list_del(&e->hash_list); | |
331 | } | |
332 | ||
333 | /* | |
334 | * Return the exception data for a sector, or NULL if not | |
335 | * remapped. | |
336 | */ | |
028867ac AK |
337 | static struct dm_snap_exception *lookup_exception(struct exception_table *et, |
338 | chunk_t chunk) | |
1da177e4 LT |
339 | { |
340 | struct list_head *slot; | |
028867ac | 341 | struct dm_snap_exception *e; |
1da177e4 LT |
342 | |
343 | slot = &et->table[exception_hash(et, chunk)]; | |
344 | list_for_each_entry (e, slot, hash_list) | |
d74f81f8 MB |
345 | if (chunk >= e->old_chunk && |
346 | chunk <= e->old_chunk + dm_consecutive_chunk_count(e)) | |
1da177e4 LT |
347 | return e; |
348 | ||
349 | return NULL; | |
350 | } | |
351 | ||
028867ac | 352 | static struct dm_snap_exception *alloc_exception(void) |
1da177e4 | 353 | { |
028867ac | 354 | struct dm_snap_exception *e; |
1da177e4 LT |
355 | |
356 | e = kmem_cache_alloc(exception_cache, GFP_NOIO); | |
357 | if (!e) | |
358 | e = kmem_cache_alloc(exception_cache, GFP_ATOMIC); | |
359 | ||
360 | return e; | |
361 | } | |
362 | ||
028867ac | 363 | static void free_exception(struct dm_snap_exception *e) |
1da177e4 LT |
364 | { |
365 | kmem_cache_free(exception_cache, e); | |
366 | } | |
367 | ||
92e86812 | 368 | static struct dm_snap_pending_exception *alloc_pending_exception(struct dm_snapshot *s) |
1da177e4 | 369 | { |
92e86812 MP |
370 | struct dm_snap_pending_exception *pe = mempool_alloc(s->pending_pool, |
371 | GFP_NOIO); | |
372 | ||
879129d2 | 373 | atomic_inc(&s->pending_exceptions_count); |
92e86812 MP |
374 | pe->snap = s; |
375 | ||
376 | return pe; | |
1da177e4 LT |
377 | } |
378 | ||
028867ac | 379 | static void free_pending_exception(struct dm_snap_pending_exception *pe) |
1da177e4 | 380 | { |
879129d2 MP |
381 | struct dm_snapshot *s = pe->snap; |
382 | ||
383 | mempool_free(pe, s->pending_pool); | |
384 | smp_mb__before_atomic_dec(); | |
385 | atomic_dec(&s->pending_exceptions_count); | |
1da177e4 LT |
386 | } |
387 | ||
d74f81f8 MB |
388 | static void insert_completed_exception(struct dm_snapshot *s, |
389 | struct dm_snap_exception *new_e) | |
390 | { | |
391 | struct exception_table *eh = &s->complete; | |
392 | struct list_head *l; | |
393 | struct dm_snap_exception *e = NULL; | |
394 | ||
395 | l = &eh->table[exception_hash(eh, new_e->old_chunk)]; | |
396 | ||
397 | /* Add immediately if this table doesn't support consecutive chunks */ | |
398 | if (!eh->hash_shift) | |
399 | goto out; | |
400 | ||
401 | /* List is ordered by old_chunk */ | |
402 | list_for_each_entry_reverse(e, l, hash_list) { | |
403 | /* Insert after an existing chunk? */ | |
404 | if (new_e->old_chunk == (e->old_chunk + | |
405 | dm_consecutive_chunk_count(e) + 1) && | |
406 | new_e->new_chunk == (dm_chunk_number(e->new_chunk) + | |
407 | dm_consecutive_chunk_count(e) + 1)) { | |
408 | dm_consecutive_chunk_count_inc(e); | |
409 | free_exception(new_e); | |
410 | return; | |
411 | } | |
412 | ||
413 | /* Insert before an existing chunk? */ | |
414 | if (new_e->old_chunk == (e->old_chunk - 1) && | |
415 | new_e->new_chunk == (dm_chunk_number(e->new_chunk) - 1)) { | |
416 | dm_consecutive_chunk_count_inc(e); | |
417 | e->old_chunk--; | |
418 | e->new_chunk--; | |
419 | free_exception(new_e); | |
420 | return; | |
421 | } | |
422 | ||
423 | if (new_e->old_chunk > e->old_chunk) | |
424 | break; | |
425 | } | |
426 | ||
427 | out: | |
428 | list_add(&new_e->hash_list, e ? &e->hash_list : l); | |
429 | } | |
430 | ||
1da177e4 LT |
431 | int dm_add_exception(struct dm_snapshot *s, chunk_t old, chunk_t new) |
432 | { | |
028867ac | 433 | struct dm_snap_exception *e; |
1da177e4 LT |
434 | |
435 | e = alloc_exception(); | |
436 | if (!e) | |
437 | return -ENOMEM; | |
438 | ||
439 | e->old_chunk = old; | |
d74f81f8 MB |
440 | |
441 | /* Consecutive_count is implicitly initialised to zero */ | |
1da177e4 | 442 | e->new_chunk = new; |
d74f81f8 MB |
443 | |
444 | insert_completed_exception(s, e); | |
445 | ||
1da177e4 LT |
446 | return 0; |
447 | } | |
448 | ||
449 | /* | |
450 | * Hard coded magic. | |
451 | */ | |
452 | static int calc_max_buckets(void) | |
453 | { | |
454 | /* use a fixed size of 2MB */ | |
455 | unsigned long mem = 2 * 1024 * 1024; | |
456 | mem /= sizeof(struct list_head); | |
457 | ||
458 | return mem; | |
459 | } | |
460 | ||
1da177e4 LT |
461 | /* |
462 | * Allocate room for a suitable hash table. | |
463 | */ | |
464 | static int init_hash_tables(struct dm_snapshot *s) | |
465 | { | |
466 | sector_t hash_size, cow_dev_size, origin_dev_size, max_buckets; | |
467 | ||
468 | /* | |
469 | * Calculate based on the size of the original volume or | |
470 | * the COW volume... | |
471 | */ | |
472 | cow_dev_size = get_dev_size(s->cow->bdev); | |
473 | origin_dev_size = get_dev_size(s->origin->bdev); | |
474 | max_buckets = calc_max_buckets(); | |
475 | ||
476 | hash_size = min(origin_dev_size, cow_dev_size) >> s->chunk_shift; | |
477 | hash_size = min(hash_size, max_buckets); | |
478 | ||
8defd830 | 479 | hash_size = rounddown_pow_of_two(hash_size); |
d74f81f8 MB |
480 | if (init_exception_table(&s->complete, hash_size, |
481 | DM_CHUNK_CONSECUTIVE_BITS)) | |
1da177e4 LT |
482 | return -ENOMEM; |
483 | ||
484 | /* | |
485 | * Allocate hash table for in-flight exceptions | |
486 | * Make this smaller than the real hash table | |
487 | */ | |
488 | hash_size >>= 3; | |
489 | if (hash_size < 64) | |
490 | hash_size = 64; | |
491 | ||
d74f81f8 | 492 | if (init_exception_table(&s->pending, hash_size, 0)) { |
1da177e4 LT |
493 | exit_exception_table(&s->complete, exception_cache); |
494 | return -ENOMEM; | |
495 | } | |
496 | ||
497 | return 0; | |
498 | } | |
499 | ||
500 | /* | |
501 | * Round a number up to the nearest 'size' boundary. size must | |
502 | * be a power of 2. | |
503 | */ | |
028867ac | 504 | static ulong round_up(ulong n, ulong size) |
1da177e4 LT |
505 | { |
506 | size--; | |
507 | return (n + size) & ~size; | |
508 | } | |
509 | ||
4c7e3bf4 MM |
510 | static int set_chunk_size(struct dm_snapshot *s, const char *chunk_size_arg, |
511 | char **error) | |
512 | { | |
513 | unsigned long chunk_size; | |
514 | char *value; | |
515 | ||
516 | chunk_size = simple_strtoul(chunk_size_arg, &value, 10); | |
517 | if (*chunk_size_arg == '\0' || *value != '\0') { | |
518 | *error = "Invalid chunk size"; | |
519 | return -EINVAL; | |
520 | } | |
521 | ||
522 | if (!chunk_size) { | |
523 | s->chunk_size = s->chunk_mask = s->chunk_shift = 0; | |
524 | return 0; | |
525 | } | |
526 | ||
527 | /* | |
528 | * Chunk size must be multiple of page size. Silently | |
529 | * round up if it's not. | |
530 | */ | |
531 | chunk_size = round_up(chunk_size, PAGE_SIZE >> 9); | |
532 | ||
533 | /* Check chunk_size is a power of 2 */ | |
6f3c3f0a | 534 | if (!is_power_of_2(chunk_size)) { |
4c7e3bf4 MM |
535 | *error = "Chunk size is not a power of 2"; |
536 | return -EINVAL; | |
537 | } | |
538 | ||
539 | /* Validate the chunk size against the device block size */ | |
540 | if (chunk_size % (bdev_hardsect_size(s->cow->bdev) >> 9)) { | |
541 | *error = "Chunk size is not a multiple of device blocksize"; | |
542 | return -EINVAL; | |
543 | } | |
544 | ||
545 | s->chunk_size = chunk_size; | |
546 | s->chunk_mask = chunk_size - 1; | |
547 | s->chunk_shift = ffs(chunk_size) - 1; | |
548 | ||
549 | return 0; | |
550 | } | |
551 | ||
1da177e4 LT |
552 | /* |
553 | * Construct a snapshot mapping: <origin_dev> <COW-dev> <p/n> <chunk-size> | |
554 | */ | |
555 | static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv) | |
556 | { | |
557 | struct dm_snapshot *s; | |
cd45daff | 558 | int i; |
1da177e4 LT |
559 | int r = -EINVAL; |
560 | char persistent; | |
561 | char *origin_path; | |
562 | char *cow_path; | |
1da177e4 | 563 | |
4c7e3bf4 | 564 | if (argc != 4) { |
72d94861 | 565 | ti->error = "requires exactly 4 arguments"; |
1da177e4 LT |
566 | r = -EINVAL; |
567 | goto bad1; | |
568 | } | |
569 | ||
570 | origin_path = argv[0]; | |
571 | cow_path = argv[1]; | |
572 | persistent = toupper(*argv[2]); | |
573 | ||
574 | if (persistent != 'P' && persistent != 'N') { | |
575 | ti->error = "Persistent flag is not P or N"; | |
576 | r = -EINVAL; | |
577 | goto bad1; | |
578 | } | |
579 | ||
1da177e4 LT |
580 | s = kmalloc(sizeof(*s), GFP_KERNEL); |
581 | if (s == NULL) { | |
582 | ti->error = "Cannot allocate snapshot context private " | |
583 | "structure"; | |
584 | r = -ENOMEM; | |
585 | goto bad1; | |
586 | } | |
587 | ||
588 | r = dm_get_device(ti, origin_path, 0, ti->len, FMODE_READ, &s->origin); | |
589 | if (r) { | |
590 | ti->error = "Cannot get origin device"; | |
591 | goto bad2; | |
592 | } | |
593 | ||
594 | r = dm_get_device(ti, cow_path, 0, 0, | |
595 | FMODE_READ | FMODE_WRITE, &s->cow); | |
596 | if (r) { | |
597 | dm_put_device(ti, s->origin); | |
598 | ti->error = "Cannot get COW device"; | |
599 | goto bad2; | |
600 | } | |
601 | ||
4c7e3bf4 MM |
602 | r = set_chunk_size(s, argv[3], &ti->error); |
603 | if (r) | |
1da177e4 | 604 | goto bad3; |
1da177e4 | 605 | |
1da177e4 | 606 | s->type = persistent; |
1da177e4 LT |
607 | |
608 | s->valid = 1; | |
aa14edeb | 609 | s->active = 0; |
879129d2 | 610 | atomic_set(&s->pending_exceptions_count, 0); |
1da177e4 | 611 | init_rwsem(&s->lock); |
ca3a931f | 612 | spin_lock_init(&s->pe_lock); |
72727bad | 613 | s->ti = ti; |
1da177e4 LT |
614 | |
615 | /* Allocate hash table for COW data */ | |
616 | if (init_hash_tables(s)) { | |
617 | ti->error = "Unable to allocate hash table space"; | |
618 | r = -ENOMEM; | |
619 | goto bad3; | |
620 | } | |
621 | ||
1da177e4 LT |
622 | s->store.snap = s; |
623 | ||
624 | if (persistent == 'P') | |
4c7e3bf4 | 625 | r = dm_create_persistent(&s->store); |
1da177e4 | 626 | else |
4c7e3bf4 | 627 | r = dm_create_transient(&s->store); |
1da177e4 LT |
628 | |
629 | if (r) { | |
630 | ti->error = "Couldn't create exception store"; | |
631 | r = -EINVAL; | |
632 | goto bad4; | |
633 | } | |
634 | ||
eb69aca5 | 635 | r = dm_kcopyd_client_create(SNAPSHOT_PAGES, &s->kcopyd_client); |
1da177e4 LT |
636 | if (r) { |
637 | ti->error = "Could not create kcopyd client"; | |
638 | goto bad5; | |
639 | } | |
640 | ||
92e86812 MP |
641 | s->pending_pool = mempool_create_slab_pool(MIN_IOS, pending_cache); |
642 | if (!s->pending_pool) { | |
643 | ti->error = "Could not allocate mempool for pending exceptions"; | |
644 | goto bad6; | |
645 | } | |
646 | ||
cd45daff MP |
647 | s->tracked_chunk_pool = mempool_create_slab_pool(MIN_IOS, |
648 | tracked_chunk_cache); | |
649 | if (!s->tracked_chunk_pool) { | |
650 | ti->error = "Could not allocate tracked_chunk mempool for " | |
651 | "tracking reads"; | |
92e86812 | 652 | goto bad_tracked_chunk_pool; |
cd45daff MP |
653 | } |
654 | ||
655 | for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++) | |
656 | INIT_HLIST_HEAD(&s->tracked_chunk_hash[i]); | |
657 | ||
658 | spin_lock_init(&s->tracked_chunk_lock); | |
659 | ||
aa14edeb | 660 | /* Metadata must only be loaded into one table at once */ |
f9cea4f7 | 661 | r = s->store.read_metadata(&s->store); |
0764147b | 662 | if (r < 0) { |
f9cea4f7 | 663 | ti->error = "Failed to read snapshot metadata"; |
cd45daff | 664 | goto bad_load_and_register; |
0764147b MB |
665 | } else if (r > 0) { |
666 | s->valid = 0; | |
667 | DMWARN("Snapshot is marked invalid."); | |
f9cea4f7 | 668 | } |
aa14edeb | 669 | |
ca3a931f | 670 | bio_list_init(&s->queued_bios); |
c4028958 | 671 | INIT_WORK(&s->queued_bios_work, flush_queued_bios); |
ca3a931f | 672 | |
1da177e4 | 673 | /* Add snapshot to the list of snapshots for this origin */ |
aa14edeb | 674 | /* Exceptions aren't triggered till snapshot_resume() is called */ |
1da177e4 LT |
675 | if (register_snapshot(s)) { |
676 | r = -EINVAL; | |
677 | ti->error = "Cannot register snapshot origin"; | |
cd45daff | 678 | goto bad_load_and_register; |
1da177e4 LT |
679 | } |
680 | ||
681 | ti->private = s; | |
c51c2752 | 682 | ti->split_io = s->chunk_size; |
1da177e4 LT |
683 | |
684 | return 0; | |
685 | ||
cd45daff MP |
686 | bad_load_and_register: |
687 | mempool_destroy(s->tracked_chunk_pool); | |
688 | ||
92e86812 MP |
689 | bad_tracked_chunk_pool: |
690 | mempool_destroy(s->pending_pool); | |
691 | ||
1da177e4 | 692 | bad6: |
eb69aca5 | 693 | dm_kcopyd_client_destroy(s->kcopyd_client); |
1da177e4 LT |
694 | |
695 | bad5: | |
696 | s->store.destroy(&s->store); | |
697 | ||
698 | bad4: | |
699 | exit_exception_table(&s->pending, pending_cache); | |
700 | exit_exception_table(&s->complete, exception_cache); | |
701 | ||
702 | bad3: | |
703 | dm_put_device(ti, s->cow); | |
704 | dm_put_device(ti, s->origin); | |
705 | ||
706 | bad2: | |
707 | kfree(s); | |
708 | ||
709 | bad1: | |
710 | return r; | |
711 | } | |
712 | ||
31c93a0c MB |
713 | static void __free_exceptions(struct dm_snapshot *s) |
714 | { | |
eb69aca5 | 715 | dm_kcopyd_client_destroy(s->kcopyd_client); |
31c93a0c MB |
716 | s->kcopyd_client = NULL; |
717 | ||
718 | exit_exception_table(&s->pending, pending_cache); | |
719 | exit_exception_table(&s->complete, exception_cache); | |
720 | ||
721 | s->store.destroy(&s->store); | |
722 | } | |
723 | ||
1da177e4 LT |
724 | static void snapshot_dtr(struct dm_target *ti) |
725 | { | |
cd45daff MP |
726 | #ifdef CONFIG_DM_DEBUG |
727 | int i; | |
728 | #endif | |
028867ac | 729 | struct dm_snapshot *s = ti->private; |
1da177e4 | 730 | |
ca3a931f AK |
731 | flush_workqueue(ksnapd); |
732 | ||
138728dc AK |
733 | /* Prevent further origin writes from using this snapshot. */ |
734 | /* After this returns there can be no new kcopyd jobs. */ | |
1da177e4 LT |
735 | unregister_snapshot(s); |
736 | ||
879129d2 MP |
737 | while (atomic_read(&s->pending_exceptions_count)) |
738 | yield(); | |
739 | /* | |
740 | * Ensure instructions in mempool_destroy aren't reordered | |
741 | * before atomic_read. | |
742 | */ | |
743 | smp_mb(); | |
744 | ||
cd45daff MP |
745 | #ifdef CONFIG_DM_DEBUG |
746 | for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++) | |
747 | BUG_ON(!hlist_empty(&s->tracked_chunk_hash[i])); | |
748 | #endif | |
749 | ||
750 | mempool_destroy(s->tracked_chunk_pool); | |
751 | ||
31c93a0c | 752 | __free_exceptions(s); |
1da177e4 | 753 | |
92e86812 MP |
754 | mempool_destroy(s->pending_pool); |
755 | ||
1da177e4 LT |
756 | dm_put_device(ti, s->origin); |
757 | dm_put_device(ti, s->cow); | |
138728dc | 758 | |
1da177e4 LT |
759 | kfree(s); |
760 | } | |
761 | ||
762 | /* | |
763 | * Flush a list of buffers. | |
764 | */ | |
765 | static void flush_bios(struct bio *bio) | |
766 | { | |
767 | struct bio *n; | |
768 | ||
769 | while (bio) { | |
770 | n = bio->bi_next; | |
771 | bio->bi_next = NULL; | |
772 | generic_make_request(bio); | |
773 | bio = n; | |
774 | } | |
775 | } | |
776 | ||
c4028958 | 777 | static void flush_queued_bios(struct work_struct *work) |
ca3a931f | 778 | { |
c4028958 DH |
779 | struct dm_snapshot *s = |
780 | container_of(work, struct dm_snapshot, queued_bios_work); | |
ca3a931f AK |
781 | struct bio *queued_bios; |
782 | unsigned long flags; | |
783 | ||
784 | spin_lock_irqsave(&s->pe_lock, flags); | |
785 | queued_bios = bio_list_get(&s->queued_bios); | |
786 | spin_unlock_irqrestore(&s->pe_lock, flags); | |
787 | ||
788 | flush_bios(queued_bios); | |
789 | } | |
790 | ||
1da177e4 LT |
791 | /* |
792 | * Error a list of buffers. | |
793 | */ | |
794 | static void error_bios(struct bio *bio) | |
795 | { | |
796 | struct bio *n; | |
797 | ||
798 | while (bio) { | |
799 | n = bio->bi_next; | |
800 | bio->bi_next = NULL; | |
6712ecf8 | 801 | bio_io_error(bio); |
1da177e4 LT |
802 | bio = n; |
803 | } | |
804 | } | |
805 | ||
695368ac | 806 | static void __invalidate_snapshot(struct dm_snapshot *s, int err) |
76df1c65 AK |
807 | { |
808 | if (!s->valid) | |
809 | return; | |
810 | ||
811 | if (err == -EIO) | |
812 | DMERR("Invalidating snapshot: Error reading/writing."); | |
813 | else if (err == -ENOMEM) | |
814 | DMERR("Invalidating snapshot: Unable to allocate exception."); | |
815 | ||
76df1c65 AK |
816 | if (s->store.drop_snapshot) |
817 | s->store.drop_snapshot(&s->store); | |
818 | ||
819 | s->valid = 0; | |
820 | ||
72727bad | 821 | dm_table_event(s->ti->table); |
76df1c65 AK |
822 | } |
823 | ||
028867ac | 824 | static void get_pending_exception(struct dm_snap_pending_exception *pe) |
4b832e8d AK |
825 | { |
826 | atomic_inc(&pe->ref_count); | |
827 | } | |
828 | ||
028867ac | 829 | static struct bio *put_pending_exception(struct dm_snap_pending_exception *pe) |
4b832e8d | 830 | { |
028867ac | 831 | struct dm_snap_pending_exception *primary_pe; |
4b832e8d AK |
832 | struct bio *origin_bios = NULL; |
833 | ||
834 | primary_pe = pe->primary_pe; | |
835 | ||
836 | /* | |
837 | * If this pe is involved in a write to the origin and | |
838 | * it is the last sibling to complete then release | |
839 | * the bios for the original write to the origin. | |
840 | */ | |
841 | if (primary_pe && | |
7c5f78b9 | 842 | atomic_dec_and_test(&primary_pe->ref_count)) { |
4b832e8d | 843 | origin_bios = bio_list_get(&primary_pe->origin_bios); |
7c5f78b9 MP |
844 | free_pending_exception(primary_pe); |
845 | } | |
4b832e8d AK |
846 | |
847 | /* | |
848 | * Free the pe if it's not linked to an origin write or if | |
849 | * it's not itself a primary pe. | |
850 | */ | |
851 | if (!primary_pe || primary_pe != pe) | |
852 | free_pending_exception(pe); | |
853 | ||
4b832e8d AK |
854 | return origin_bios; |
855 | } | |
856 | ||
028867ac | 857 | static void pending_complete(struct dm_snap_pending_exception *pe, int success) |
1da177e4 | 858 | { |
028867ac | 859 | struct dm_snap_exception *e; |
1da177e4 | 860 | struct dm_snapshot *s = pe->snap; |
9d493fa8 AK |
861 | struct bio *origin_bios = NULL; |
862 | struct bio *snapshot_bios = NULL; | |
863 | int error = 0; | |
1da177e4 | 864 | |
76df1c65 AK |
865 | if (!success) { |
866 | /* Read/write error - snapshot is unusable */ | |
1da177e4 | 867 | down_write(&s->lock); |
695368ac | 868 | __invalidate_snapshot(s, -EIO); |
9d493fa8 | 869 | error = 1; |
76df1c65 AK |
870 | goto out; |
871 | } | |
872 | ||
873 | e = alloc_exception(); | |
874 | if (!e) { | |
1da177e4 | 875 | down_write(&s->lock); |
695368ac | 876 | __invalidate_snapshot(s, -ENOMEM); |
9d493fa8 | 877 | error = 1; |
76df1c65 AK |
878 | goto out; |
879 | } | |
880 | *e = pe->e; | |
1da177e4 | 881 | |
76df1c65 AK |
882 | down_write(&s->lock); |
883 | if (!s->valid) { | |
76df1c65 | 884 | free_exception(e); |
9d493fa8 | 885 | error = 1; |
76df1c65 | 886 | goto out; |
1da177e4 LT |
887 | } |
888 | ||
a8d41b59 MP |
889 | /* |
890 | * Check for conflicting reads. This is extremely improbable, | |
891 | * so yield() is sufficient and there is no need for a wait queue. | |
892 | */ | |
893 | while (__chunk_is_tracked(s, pe->e.old_chunk)) | |
894 | yield(); | |
895 | ||
9d493fa8 AK |
896 | /* |
897 | * Add a proper exception, and remove the | |
898 | * in-flight exception from the list. | |
899 | */ | |
d74f81f8 | 900 | insert_completed_exception(s, e); |
76df1c65 | 901 | |
1da177e4 | 902 | out: |
695368ac | 903 | remove_exception(&pe->e); |
9d493fa8 | 904 | snapshot_bios = bio_list_get(&pe->snapshot_bios); |
4b832e8d | 905 | origin_bios = put_pending_exception(pe); |
1da177e4 | 906 | |
9d493fa8 AK |
907 | up_write(&s->lock); |
908 | ||
909 | /* Submit any pending write bios */ | |
910 | if (error) | |
911 | error_bios(snapshot_bios); | |
912 | else | |
913 | flush_bios(snapshot_bios); | |
914 | ||
915 | flush_bios(origin_bios); | |
1da177e4 LT |
916 | } |
917 | ||
918 | static void commit_callback(void *context, int success) | |
919 | { | |
028867ac AK |
920 | struct dm_snap_pending_exception *pe = context; |
921 | ||
1da177e4 LT |
922 | pending_complete(pe, success); |
923 | } | |
924 | ||
925 | /* | |
926 | * Called when the copy I/O has finished. kcopyd actually runs | |
927 | * this code so don't block. | |
928 | */ | |
4cdc1d1f | 929 | static void copy_callback(int read_err, unsigned long write_err, void *context) |
1da177e4 | 930 | { |
028867ac | 931 | struct dm_snap_pending_exception *pe = context; |
1da177e4 LT |
932 | struct dm_snapshot *s = pe->snap; |
933 | ||
934 | if (read_err || write_err) | |
935 | pending_complete(pe, 0); | |
936 | ||
937 | else | |
938 | /* Update the metadata if we are persistent */ | |
939 | s->store.commit_exception(&s->store, &pe->e, commit_callback, | |
940 | pe); | |
941 | } | |
942 | ||
943 | /* | |
944 | * Dispatches the copy operation to kcopyd. | |
945 | */ | |
028867ac | 946 | static void start_copy(struct dm_snap_pending_exception *pe) |
1da177e4 LT |
947 | { |
948 | struct dm_snapshot *s = pe->snap; | |
22a1ceb1 | 949 | struct dm_io_region src, dest; |
1da177e4 LT |
950 | struct block_device *bdev = s->origin->bdev; |
951 | sector_t dev_size; | |
952 | ||
953 | dev_size = get_dev_size(bdev); | |
954 | ||
955 | src.bdev = bdev; | |
956 | src.sector = chunk_to_sector(s, pe->e.old_chunk); | |
957 | src.count = min(s->chunk_size, dev_size - src.sector); | |
958 | ||
959 | dest.bdev = s->cow->bdev; | |
960 | dest.sector = chunk_to_sector(s, pe->e.new_chunk); | |
961 | dest.count = src.count; | |
962 | ||
963 | /* Hand over to kcopyd */ | |
eb69aca5 | 964 | dm_kcopyd_copy(s->kcopyd_client, |
1da177e4 LT |
965 | &src, 1, &dest, 0, copy_callback, pe); |
966 | } | |
967 | ||
968 | /* | |
969 | * Looks to see if this snapshot already has a pending exception | |
970 | * for this chunk, otherwise it allocates a new one and inserts | |
971 | * it into the pending table. | |
972 | * | |
973 | * NOTE: a write lock must be held on snap->lock before calling | |
974 | * this. | |
975 | */ | |
028867ac | 976 | static struct dm_snap_pending_exception * |
1da177e4 LT |
977 | __find_pending_exception(struct dm_snapshot *s, struct bio *bio) |
978 | { | |
028867ac AK |
979 | struct dm_snap_exception *e; |
980 | struct dm_snap_pending_exception *pe; | |
1da177e4 LT |
981 | chunk_t chunk = sector_to_chunk(s, bio->bi_sector); |
982 | ||
983 | /* | |
984 | * Is there a pending exception for this already ? | |
985 | */ | |
986 | e = lookup_exception(&s->pending, chunk); | |
987 | if (e) { | |
988 | /* cast the exception to a pending exception */ | |
028867ac | 989 | pe = container_of(e, struct dm_snap_pending_exception, e); |
76df1c65 AK |
990 | goto out; |
991 | } | |
1da177e4 | 992 | |
76df1c65 AK |
993 | /* |
994 | * Create a new pending exception, we don't want | |
995 | * to hold the lock while we do this. | |
996 | */ | |
997 | up_write(&s->lock); | |
92e86812 | 998 | pe = alloc_pending_exception(s); |
76df1c65 | 999 | down_write(&s->lock); |
1da177e4 | 1000 | |
76df1c65 AK |
1001 | if (!s->valid) { |
1002 | free_pending_exception(pe); | |
1003 | return NULL; | |
1004 | } | |
1da177e4 | 1005 | |
76df1c65 AK |
1006 | e = lookup_exception(&s->pending, chunk); |
1007 | if (e) { | |
1008 | free_pending_exception(pe); | |
028867ac | 1009 | pe = container_of(e, struct dm_snap_pending_exception, e); |
76df1c65 | 1010 | goto out; |
1da177e4 LT |
1011 | } |
1012 | ||
76df1c65 AK |
1013 | pe->e.old_chunk = chunk; |
1014 | bio_list_init(&pe->origin_bios); | |
1015 | bio_list_init(&pe->snapshot_bios); | |
1016 | pe->primary_pe = NULL; | |
4b832e8d | 1017 | atomic_set(&pe->ref_count, 0); |
76df1c65 AK |
1018 | pe->started = 0; |
1019 | ||
1020 | if (s->store.prepare_exception(&s->store, &pe->e)) { | |
1021 | free_pending_exception(pe); | |
1022 | return NULL; | |
1023 | } | |
1024 | ||
4b832e8d | 1025 | get_pending_exception(pe); |
76df1c65 AK |
1026 | insert_exception(&s->pending, &pe->e); |
1027 | ||
1028 | out: | |
1da177e4 LT |
1029 | return pe; |
1030 | } | |
1031 | ||
028867ac | 1032 | static void remap_exception(struct dm_snapshot *s, struct dm_snap_exception *e, |
d74f81f8 | 1033 | struct bio *bio, chunk_t chunk) |
1da177e4 LT |
1034 | { |
1035 | bio->bi_bdev = s->cow->bdev; | |
d74f81f8 MB |
1036 | bio->bi_sector = chunk_to_sector(s, dm_chunk_number(e->new_chunk) + |
1037 | (chunk - e->old_chunk)) + | |
1038 | (bio->bi_sector & s->chunk_mask); | |
1da177e4 LT |
1039 | } |
1040 | ||
1041 | static int snapshot_map(struct dm_target *ti, struct bio *bio, | |
1042 | union map_info *map_context) | |
1043 | { | |
028867ac AK |
1044 | struct dm_snap_exception *e; |
1045 | struct dm_snapshot *s = ti->private; | |
d2a7ad29 | 1046 | int r = DM_MAPIO_REMAPPED; |
1da177e4 | 1047 | chunk_t chunk; |
028867ac | 1048 | struct dm_snap_pending_exception *pe = NULL; |
1da177e4 LT |
1049 | |
1050 | chunk = sector_to_chunk(s, bio->bi_sector); | |
1051 | ||
1052 | /* Full snapshots are not usable */ | |
76df1c65 | 1053 | /* To get here the table must be live so s->active is always set. */ |
1da177e4 | 1054 | if (!s->valid) |
f6a80ea8 | 1055 | return -EIO; |
1da177e4 | 1056 | |
ba40a2aa AK |
1057 | /* FIXME: should only take write lock if we need |
1058 | * to copy an exception */ | |
1059 | down_write(&s->lock); | |
1060 | ||
1061 | if (!s->valid) { | |
1062 | r = -EIO; | |
1063 | goto out_unlock; | |
1064 | } | |
1065 | ||
1066 | /* If the block is already remapped - use that, else remap it */ | |
1067 | e = lookup_exception(&s->complete, chunk); | |
1068 | if (e) { | |
d74f81f8 | 1069 | remap_exception(s, e, bio, chunk); |
ba40a2aa AK |
1070 | goto out_unlock; |
1071 | } | |
1072 | ||
1da177e4 LT |
1073 | /* |
1074 | * Write to snapshot - higher level takes care of RW/RO | |
1075 | * flags so we should only get this if we are | |
1076 | * writeable. | |
1077 | */ | |
1078 | if (bio_rw(bio) == WRITE) { | |
76df1c65 AK |
1079 | pe = __find_pending_exception(s, bio); |
1080 | if (!pe) { | |
695368ac | 1081 | __invalidate_snapshot(s, -ENOMEM); |
76df1c65 AK |
1082 | r = -EIO; |
1083 | goto out_unlock; | |
1da177e4 LT |
1084 | } |
1085 | ||
d74f81f8 | 1086 | remap_exception(s, &pe->e, bio, chunk); |
76df1c65 AK |
1087 | bio_list_add(&pe->snapshot_bios, bio); |
1088 | ||
d2a7ad29 | 1089 | r = DM_MAPIO_SUBMITTED; |
ba40a2aa | 1090 | |
76df1c65 AK |
1091 | if (!pe->started) { |
1092 | /* this is protected by snap->lock */ | |
1093 | pe->started = 1; | |
ba40a2aa | 1094 | up_write(&s->lock); |
76df1c65 | 1095 | start_copy(pe); |
ba40a2aa AK |
1096 | goto out; |
1097 | } | |
cd45daff | 1098 | } else { |
ba40a2aa | 1099 | bio->bi_bdev = s->origin->bdev; |
cd45daff MP |
1100 | map_context->ptr = track_chunk(s, chunk); |
1101 | } | |
1da177e4 | 1102 | |
ba40a2aa AK |
1103 | out_unlock: |
1104 | up_write(&s->lock); | |
1105 | out: | |
1da177e4 LT |
1106 | return r; |
1107 | } | |
1108 | ||
cd45daff MP |
1109 | static int snapshot_end_io(struct dm_target *ti, struct bio *bio, |
1110 | int error, union map_info *map_context) | |
1111 | { | |
1112 | struct dm_snapshot *s = ti->private; | |
1113 | struct dm_snap_tracked_chunk *c = map_context->ptr; | |
1114 | ||
1115 | if (c) | |
1116 | stop_tracking_chunk(s, c); | |
1117 | ||
1118 | return 0; | |
1119 | } | |
1120 | ||
1da177e4 LT |
1121 | static void snapshot_resume(struct dm_target *ti) |
1122 | { | |
028867ac | 1123 | struct dm_snapshot *s = ti->private; |
1da177e4 | 1124 | |
aa14edeb AK |
1125 | down_write(&s->lock); |
1126 | s->active = 1; | |
1127 | up_write(&s->lock); | |
1da177e4 LT |
1128 | } |
1129 | ||
1130 | static int snapshot_status(struct dm_target *ti, status_type_t type, | |
1131 | char *result, unsigned int maxlen) | |
1132 | { | |
028867ac | 1133 | struct dm_snapshot *snap = ti->private; |
1da177e4 LT |
1134 | |
1135 | switch (type) { | |
1136 | case STATUSTYPE_INFO: | |
1137 | if (!snap->valid) | |
1138 | snprintf(result, maxlen, "Invalid"); | |
1139 | else { | |
1140 | if (snap->store.fraction_full) { | |
1141 | sector_t numerator, denominator; | |
1142 | snap->store.fraction_full(&snap->store, | |
1143 | &numerator, | |
1144 | &denominator); | |
4ee218cd AM |
1145 | snprintf(result, maxlen, "%llu/%llu", |
1146 | (unsigned long long)numerator, | |
1147 | (unsigned long long)denominator); | |
1da177e4 LT |
1148 | } |
1149 | else | |
1150 | snprintf(result, maxlen, "Unknown"); | |
1151 | } | |
1152 | break; | |
1153 | ||
1154 | case STATUSTYPE_TABLE: | |
1155 | /* | |
1156 | * kdevname returns a static pointer so we need | |
1157 | * to make private copies if the output is to | |
1158 | * make sense. | |
1159 | */ | |
4ee218cd | 1160 | snprintf(result, maxlen, "%s %s %c %llu", |
1da177e4 | 1161 | snap->origin->name, snap->cow->name, |
4ee218cd AM |
1162 | snap->type, |
1163 | (unsigned long long)snap->chunk_size); | |
1da177e4 LT |
1164 | break; |
1165 | } | |
1166 | ||
1167 | return 0; | |
1168 | } | |
1169 | ||
1170 | /*----------------------------------------------------------------- | |
1171 | * Origin methods | |
1172 | *---------------------------------------------------------------*/ | |
1da177e4 LT |
1173 | static int __origin_write(struct list_head *snapshots, struct bio *bio) |
1174 | { | |
d2a7ad29 | 1175 | int r = DM_MAPIO_REMAPPED, first = 0; |
1da177e4 | 1176 | struct dm_snapshot *snap; |
028867ac AK |
1177 | struct dm_snap_exception *e; |
1178 | struct dm_snap_pending_exception *pe, *next_pe, *primary_pe = NULL; | |
1da177e4 | 1179 | chunk_t chunk; |
eccf0817 | 1180 | LIST_HEAD(pe_queue); |
1da177e4 LT |
1181 | |
1182 | /* Do all the snapshots on this origin */ | |
1183 | list_for_each_entry (snap, snapshots, list) { | |
1184 | ||
76df1c65 AK |
1185 | down_write(&snap->lock); |
1186 | ||
aa14edeb AK |
1187 | /* Only deal with valid and active snapshots */ |
1188 | if (!snap->valid || !snap->active) | |
76df1c65 | 1189 | goto next_snapshot; |
1da177e4 | 1190 | |
d5e404c1 | 1191 | /* Nothing to do if writing beyond end of snapshot */ |
72727bad | 1192 | if (bio->bi_sector >= dm_table_get_size(snap->ti->table)) |
76df1c65 | 1193 | goto next_snapshot; |
1da177e4 LT |
1194 | |
1195 | /* | |
1196 | * Remember, different snapshots can have | |
1197 | * different chunk sizes. | |
1198 | */ | |
1199 | chunk = sector_to_chunk(snap, bio->bi_sector); | |
1200 | ||
1201 | /* | |
1202 | * Check exception table to see if block | |
1203 | * is already remapped in this snapshot | |
1204 | * and trigger an exception if not. | |
b4b610f6 | 1205 | * |
4b832e8d | 1206 | * ref_count is initialised to 1 so pending_complete() |
b4b610f6 | 1207 | * won't destroy the primary_pe while we're inside this loop. |
1da177e4 LT |
1208 | */ |
1209 | e = lookup_exception(&snap->complete, chunk); | |
76df1c65 AK |
1210 | if (e) |
1211 | goto next_snapshot; | |
1212 | ||
1213 | pe = __find_pending_exception(snap, bio); | |
1214 | if (!pe) { | |
695368ac | 1215 | __invalidate_snapshot(snap, -ENOMEM); |
76df1c65 AK |
1216 | goto next_snapshot; |
1217 | } | |
1218 | ||
1219 | if (!primary_pe) { | |
1220 | /* | |
1221 | * Either every pe here has same | |
1222 | * primary_pe or none has one yet. | |
1223 | */ | |
1224 | if (pe->primary_pe) | |
1225 | primary_pe = pe->primary_pe; | |
1226 | else { | |
1227 | primary_pe = pe; | |
1228 | first = 1; | |
1da177e4 | 1229 | } |
76df1c65 AK |
1230 | |
1231 | bio_list_add(&primary_pe->origin_bios, bio); | |
1232 | ||
d2a7ad29 | 1233 | r = DM_MAPIO_SUBMITTED; |
76df1c65 AK |
1234 | } |
1235 | ||
1236 | if (!pe->primary_pe) { | |
76df1c65 | 1237 | pe->primary_pe = primary_pe; |
4b832e8d | 1238 | get_pending_exception(primary_pe); |
76df1c65 AK |
1239 | } |
1240 | ||
1241 | if (!pe->started) { | |
1242 | pe->started = 1; | |
1243 | list_add_tail(&pe->list, &pe_queue); | |
1da177e4 LT |
1244 | } |
1245 | ||
76df1c65 | 1246 | next_snapshot: |
1da177e4 LT |
1247 | up_write(&snap->lock); |
1248 | } | |
1249 | ||
b4b610f6 | 1250 | if (!primary_pe) |
4b832e8d | 1251 | return r; |
b4b610f6 AK |
1252 | |
1253 | /* | |
1254 | * If this is the first time we're processing this chunk and | |
4b832e8d | 1255 | * ref_count is now 1 it means all the pending exceptions |
b4b610f6 AK |
1256 | * got completed while we were in the loop above, so it falls to |
1257 | * us here to remove the primary_pe and submit any origin_bios. | |
1258 | */ | |
1259 | ||
4b832e8d | 1260 | if (first && atomic_dec_and_test(&primary_pe->ref_count)) { |
b4b610f6 AK |
1261 | flush_bios(bio_list_get(&primary_pe->origin_bios)); |
1262 | free_pending_exception(primary_pe); | |
1263 | /* If we got here, pe_queue is necessarily empty. */ | |
4b832e8d | 1264 | return r; |
b4b610f6 AK |
1265 | } |
1266 | ||
1da177e4 LT |
1267 | /* |
1268 | * Now that we have a complete pe list we can start the copying. | |
1269 | */ | |
eccf0817 AK |
1270 | list_for_each_entry_safe(pe, next_pe, &pe_queue, list) |
1271 | start_copy(pe); | |
1da177e4 LT |
1272 | |
1273 | return r; | |
1274 | } | |
1275 | ||
1276 | /* | |
1277 | * Called on a write from the origin driver. | |
1278 | */ | |
1279 | static int do_origin(struct dm_dev *origin, struct bio *bio) | |
1280 | { | |
1281 | struct origin *o; | |
d2a7ad29 | 1282 | int r = DM_MAPIO_REMAPPED; |
1da177e4 LT |
1283 | |
1284 | down_read(&_origins_lock); | |
1285 | o = __lookup_origin(origin->bdev); | |
1286 | if (o) | |
1287 | r = __origin_write(&o->snapshots, bio); | |
1288 | up_read(&_origins_lock); | |
1289 | ||
1290 | return r; | |
1291 | } | |
1292 | ||
1293 | /* | |
1294 | * Origin: maps a linear range of a device, with hooks for snapshotting. | |
1295 | */ | |
1296 | ||
1297 | /* | |
1298 | * Construct an origin mapping: <dev_path> | |
1299 | * The context for an origin is merely a 'struct dm_dev *' | |
1300 | * pointing to the real device. | |
1301 | */ | |
1302 | static int origin_ctr(struct dm_target *ti, unsigned int argc, char **argv) | |
1303 | { | |
1304 | int r; | |
1305 | struct dm_dev *dev; | |
1306 | ||
1307 | if (argc != 1) { | |
72d94861 | 1308 | ti->error = "origin: incorrect number of arguments"; |
1da177e4 LT |
1309 | return -EINVAL; |
1310 | } | |
1311 | ||
1312 | r = dm_get_device(ti, argv[0], 0, ti->len, | |
1313 | dm_table_get_mode(ti->table), &dev); | |
1314 | if (r) { | |
1315 | ti->error = "Cannot get target device"; | |
1316 | return r; | |
1317 | } | |
1318 | ||
1319 | ti->private = dev; | |
1320 | return 0; | |
1321 | } | |
1322 | ||
1323 | static void origin_dtr(struct dm_target *ti) | |
1324 | { | |
028867ac | 1325 | struct dm_dev *dev = ti->private; |
1da177e4 LT |
1326 | dm_put_device(ti, dev); |
1327 | } | |
1328 | ||
1329 | static int origin_map(struct dm_target *ti, struct bio *bio, | |
1330 | union map_info *map_context) | |
1331 | { | |
028867ac | 1332 | struct dm_dev *dev = ti->private; |
1da177e4 LT |
1333 | bio->bi_bdev = dev->bdev; |
1334 | ||
1335 | /* Only tell snapshots if this is a write */ | |
d2a7ad29 | 1336 | return (bio_rw(bio) == WRITE) ? do_origin(dev, bio) : DM_MAPIO_REMAPPED; |
1da177e4 LT |
1337 | } |
1338 | ||
1339 | #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r)) | |
1340 | ||
1341 | /* | |
1342 | * Set the target "split_io" field to the minimum of all the snapshots' | |
1343 | * chunk sizes. | |
1344 | */ | |
1345 | static void origin_resume(struct dm_target *ti) | |
1346 | { | |
028867ac | 1347 | struct dm_dev *dev = ti->private; |
1da177e4 LT |
1348 | struct dm_snapshot *snap; |
1349 | struct origin *o; | |
1350 | chunk_t chunk_size = 0; | |
1351 | ||
1352 | down_read(&_origins_lock); | |
1353 | o = __lookup_origin(dev->bdev); | |
1354 | if (o) | |
1355 | list_for_each_entry (snap, &o->snapshots, list) | |
1356 | chunk_size = min_not_zero(chunk_size, snap->chunk_size); | |
1357 | up_read(&_origins_lock); | |
1358 | ||
1359 | ti->split_io = chunk_size; | |
1360 | } | |
1361 | ||
1362 | static int origin_status(struct dm_target *ti, status_type_t type, char *result, | |
1363 | unsigned int maxlen) | |
1364 | { | |
028867ac | 1365 | struct dm_dev *dev = ti->private; |
1da177e4 LT |
1366 | |
1367 | switch (type) { | |
1368 | case STATUSTYPE_INFO: | |
1369 | result[0] = '\0'; | |
1370 | break; | |
1371 | ||
1372 | case STATUSTYPE_TABLE: | |
1373 | snprintf(result, maxlen, "%s", dev->name); | |
1374 | break; | |
1375 | } | |
1376 | ||
1377 | return 0; | |
1378 | } | |
1379 | ||
1380 | static struct target_type origin_target = { | |
1381 | .name = "snapshot-origin", | |
d74f81f8 | 1382 | .version = {1, 6, 0}, |
1da177e4 LT |
1383 | .module = THIS_MODULE, |
1384 | .ctr = origin_ctr, | |
1385 | .dtr = origin_dtr, | |
1386 | .map = origin_map, | |
1387 | .resume = origin_resume, | |
1388 | .status = origin_status, | |
1389 | }; | |
1390 | ||
1391 | static struct target_type snapshot_target = { | |
1392 | .name = "snapshot", | |
d74f81f8 | 1393 | .version = {1, 6, 0}, |
1da177e4 LT |
1394 | .module = THIS_MODULE, |
1395 | .ctr = snapshot_ctr, | |
1396 | .dtr = snapshot_dtr, | |
1397 | .map = snapshot_map, | |
cd45daff | 1398 | .end_io = snapshot_end_io, |
1da177e4 LT |
1399 | .resume = snapshot_resume, |
1400 | .status = snapshot_status, | |
1401 | }; | |
1402 | ||
1403 | static int __init dm_snapshot_init(void) | |
1404 | { | |
1405 | int r; | |
1406 | ||
1407 | r = dm_register_target(&snapshot_target); | |
1408 | if (r) { | |
1409 | DMERR("snapshot target register failed %d", r); | |
1410 | return r; | |
1411 | } | |
1412 | ||
1413 | r = dm_register_target(&origin_target); | |
1414 | if (r < 0) { | |
72d94861 | 1415 | DMERR("Origin target register failed %d", r); |
1da177e4 LT |
1416 | goto bad1; |
1417 | } | |
1418 | ||
1419 | r = init_origin_hash(); | |
1420 | if (r) { | |
1421 | DMERR("init_origin_hash failed."); | |
1422 | goto bad2; | |
1423 | } | |
1424 | ||
028867ac | 1425 | exception_cache = KMEM_CACHE(dm_snap_exception, 0); |
1da177e4 LT |
1426 | if (!exception_cache) { |
1427 | DMERR("Couldn't create exception cache."); | |
1428 | r = -ENOMEM; | |
1429 | goto bad3; | |
1430 | } | |
1431 | ||
028867ac | 1432 | pending_cache = KMEM_CACHE(dm_snap_pending_exception, 0); |
1da177e4 LT |
1433 | if (!pending_cache) { |
1434 | DMERR("Couldn't create pending cache."); | |
1435 | r = -ENOMEM; | |
1436 | goto bad4; | |
1437 | } | |
1438 | ||
cd45daff MP |
1439 | tracked_chunk_cache = KMEM_CACHE(dm_snap_tracked_chunk, 0); |
1440 | if (!tracked_chunk_cache) { | |
1441 | DMERR("Couldn't create cache to track chunks in use."); | |
1442 | r = -ENOMEM; | |
1443 | goto bad5; | |
1444 | } | |
1445 | ||
ca3a931f AK |
1446 | ksnapd = create_singlethread_workqueue("ksnapd"); |
1447 | if (!ksnapd) { | |
1448 | DMERR("Failed to create ksnapd workqueue."); | |
1449 | r = -ENOMEM; | |
92e86812 | 1450 | goto bad_pending_pool; |
ca3a931f AK |
1451 | } |
1452 | ||
1da177e4 LT |
1453 | return 0; |
1454 | ||
cd45daff MP |
1455 | bad_pending_pool: |
1456 | kmem_cache_destroy(tracked_chunk_cache); | |
1da177e4 LT |
1457 | bad5: |
1458 | kmem_cache_destroy(pending_cache); | |
1459 | bad4: | |
1460 | kmem_cache_destroy(exception_cache); | |
1461 | bad3: | |
1462 | exit_origin_hash(); | |
1463 | bad2: | |
1464 | dm_unregister_target(&origin_target); | |
1465 | bad1: | |
1466 | dm_unregister_target(&snapshot_target); | |
1467 | return r; | |
1468 | } | |
1469 | ||
1470 | static void __exit dm_snapshot_exit(void) | |
1471 | { | |
1472 | int r; | |
1473 | ||
ca3a931f AK |
1474 | destroy_workqueue(ksnapd); |
1475 | ||
1da177e4 LT |
1476 | r = dm_unregister_target(&snapshot_target); |
1477 | if (r) | |
1478 | DMERR("snapshot unregister failed %d", r); | |
1479 | ||
1480 | r = dm_unregister_target(&origin_target); | |
1481 | if (r) | |
1482 | DMERR("origin unregister failed %d", r); | |
1483 | ||
1484 | exit_origin_hash(); | |
1da177e4 LT |
1485 | kmem_cache_destroy(pending_cache); |
1486 | kmem_cache_destroy(exception_cache); | |
cd45daff | 1487 | kmem_cache_destroy(tracked_chunk_cache); |
1da177e4 LT |
1488 | } |
1489 | ||
1490 | /* Module hooks */ | |
1491 | module_init(dm_snapshot_init); | |
1492 | module_exit(dm_snapshot_exit); | |
1493 | ||
1494 | MODULE_DESCRIPTION(DM_NAME " snapshot target"); | |
1495 | MODULE_AUTHOR("Joe Thornber"); | |
1496 | MODULE_LICENSE("GPL"); |