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> | |
90fa1527 | 12 | #include <linux/delay.h> |
1da177e4 LT |
13 | #include <linux/fs.h> |
14 | #include <linux/init.h> | |
15 | #include <linux/kdev_t.h> | |
16 | #include <linux/list.h> | |
17 | #include <linux/mempool.h> | |
18 | #include <linux/module.h> | |
19 | #include <linux/slab.h> | |
20 | #include <linux/vmalloc.h> | |
6f3c3f0a | 21 | #include <linux/log2.h> |
a765e20e | 22 | #include <linux/dm-kcopyd.h> |
1da177e4 | 23 | |
aea53d92 | 24 | #include "dm-exception-store.h" |
1da177e4 LT |
25 | #include "dm-snap.h" |
26 | #include "dm-bio-list.h" | |
1da177e4 | 27 | |
72d94861 AK |
28 | #define DM_MSG_PREFIX "snapshots" |
29 | ||
1da177e4 LT |
30 | /* |
31 | * The percentage increment we will wake up users at | |
32 | */ | |
33 | #define WAKE_UP_PERCENT 5 | |
34 | ||
35 | /* | |
36 | * kcopyd priority of snapshot operations | |
37 | */ | |
38 | #define SNAPSHOT_COPY_PRIORITY 2 | |
39 | ||
40 | /* | |
8ee2767a | 41 | * Reserve 1MB for each snapshot initially (with minimum of 1 page). |
1da177e4 | 42 | */ |
8ee2767a | 43 | #define SNAPSHOT_PAGES (((1UL << 20) >> PAGE_SHIFT) ? : 1) |
1da177e4 | 44 | |
cd45daff MP |
45 | /* |
46 | * The size of the mempool used to track chunks in use. | |
47 | */ | |
48 | #define MIN_IOS 256 | |
49 | ||
c642f9e0 | 50 | static struct workqueue_struct *ksnapd; |
c4028958 | 51 | static void flush_queued_bios(struct work_struct *work); |
ca3a931f | 52 | |
028867ac AK |
53 | struct dm_snap_pending_exception { |
54 | struct dm_snap_exception e; | |
1da177e4 LT |
55 | |
56 | /* | |
57 | * Origin buffers waiting for this to complete are held | |
58 | * in a bio list | |
59 | */ | |
60 | struct bio_list origin_bios; | |
61 | struct bio_list snapshot_bios; | |
62 | ||
eccf0817 AK |
63 | /* |
64 | * Short-term queue of pending exceptions prior to submission. | |
65 | */ | |
66 | struct list_head list; | |
67 | ||
1da177e4 | 68 | /* |
b4b610f6 | 69 | * The primary pending_exception is the one that holds |
4b832e8d | 70 | * the ref_count and the list of origin_bios for a |
b4b610f6 AK |
71 | * group of pending_exceptions. It is always last to get freed. |
72 | * These fields get set up when writing to the origin. | |
1da177e4 | 73 | */ |
028867ac | 74 | struct dm_snap_pending_exception *primary_pe; |
b4b610f6 AK |
75 | |
76 | /* | |
77 | * Number of pending_exceptions processing this chunk. | |
78 | * When this drops to zero we must complete the origin bios. | |
79 | * If incrementing or decrementing this, hold pe->snap->lock for | |
80 | * the sibling concerned and not pe->primary_pe->snap->lock unless | |
81 | * they are the same. | |
82 | */ | |
4b832e8d | 83 | atomic_t ref_count; |
1da177e4 LT |
84 | |
85 | /* Pointer back to snapshot context */ | |
86 | struct dm_snapshot *snap; | |
87 | ||
88 | /* | |
89 | * 1 indicates the exception has already been sent to | |
90 | * kcopyd. | |
91 | */ | |
92 | int started; | |
93 | }; | |
94 | ||
95 | /* | |
96 | * Hash table mapping origin volumes to lists of snapshots and | |
97 | * a lock to protect it | |
98 | */ | |
e18b890b CL |
99 | static struct kmem_cache *exception_cache; |
100 | static struct kmem_cache *pending_cache; | |
1da177e4 | 101 | |
cd45daff MP |
102 | struct dm_snap_tracked_chunk { |
103 | struct hlist_node node; | |
104 | chunk_t chunk; | |
105 | }; | |
106 | ||
107 | static struct kmem_cache *tracked_chunk_cache; | |
108 | ||
109 | static struct dm_snap_tracked_chunk *track_chunk(struct dm_snapshot *s, | |
110 | chunk_t chunk) | |
111 | { | |
112 | struct dm_snap_tracked_chunk *c = mempool_alloc(s->tracked_chunk_pool, | |
113 | GFP_NOIO); | |
114 | unsigned long flags; | |
115 | ||
116 | c->chunk = chunk; | |
117 | ||
118 | spin_lock_irqsave(&s->tracked_chunk_lock, flags); | |
119 | hlist_add_head(&c->node, | |
120 | &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)]); | |
121 | spin_unlock_irqrestore(&s->tracked_chunk_lock, flags); | |
122 | ||
123 | return c; | |
124 | } | |
125 | ||
126 | static void stop_tracking_chunk(struct dm_snapshot *s, | |
127 | struct dm_snap_tracked_chunk *c) | |
128 | { | |
129 | unsigned long flags; | |
130 | ||
131 | spin_lock_irqsave(&s->tracked_chunk_lock, flags); | |
132 | hlist_del(&c->node); | |
133 | spin_unlock_irqrestore(&s->tracked_chunk_lock, flags); | |
134 | ||
135 | mempool_free(c, s->tracked_chunk_pool); | |
136 | } | |
137 | ||
a8d41b59 MP |
138 | static int __chunk_is_tracked(struct dm_snapshot *s, chunk_t chunk) |
139 | { | |
140 | struct dm_snap_tracked_chunk *c; | |
141 | struct hlist_node *hn; | |
142 | int found = 0; | |
143 | ||
144 | spin_lock_irq(&s->tracked_chunk_lock); | |
145 | ||
146 | hlist_for_each_entry(c, hn, | |
147 | &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)], node) { | |
148 | if (c->chunk == chunk) { | |
149 | found = 1; | |
150 | break; | |
151 | } | |
152 | } | |
153 | ||
154 | spin_unlock_irq(&s->tracked_chunk_lock); | |
155 | ||
156 | return found; | |
157 | } | |
158 | ||
1da177e4 LT |
159 | /* |
160 | * One of these per registered origin, held in the snapshot_origins hash | |
161 | */ | |
162 | struct origin { | |
163 | /* The origin device */ | |
164 | struct block_device *bdev; | |
165 | ||
166 | struct list_head hash_list; | |
167 | ||
168 | /* List of snapshots for this origin */ | |
169 | struct list_head snapshots; | |
170 | }; | |
171 | ||
172 | /* | |
173 | * Size of the hash table for origin volumes. If we make this | |
174 | * the size of the minors list then it should be nearly perfect | |
175 | */ | |
176 | #define ORIGIN_HASH_SIZE 256 | |
177 | #define ORIGIN_MASK 0xFF | |
178 | static struct list_head *_origins; | |
179 | static struct rw_semaphore _origins_lock; | |
180 | ||
181 | static int init_origin_hash(void) | |
182 | { | |
183 | int i; | |
184 | ||
185 | _origins = kmalloc(ORIGIN_HASH_SIZE * sizeof(struct list_head), | |
186 | GFP_KERNEL); | |
187 | if (!_origins) { | |
72d94861 | 188 | DMERR("unable to allocate memory"); |
1da177e4 LT |
189 | return -ENOMEM; |
190 | } | |
191 | ||
192 | for (i = 0; i < ORIGIN_HASH_SIZE; i++) | |
193 | INIT_LIST_HEAD(_origins + i); | |
194 | init_rwsem(&_origins_lock); | |
195 | ||
196 | return 0; | |
197 | } | |
198 | ||
199 | static void exit_origin_hash(void) | |
200 | { | |
201 | kfree(_origins); | |
202 | } | |
203 | ||
028867ac | 204 | static unsigned origin_hash(struct block_device *bdev) |
1da177e4 LT |
205 | { |
206 | return bdev->bd_dev & ORIGIN_MASK; | |
207 | } | |
208 | ||
209 | static struct origin *__lookup_origin(struct block_device *origin) | |
210 | { | |
211 | struct list_head *ol; | |
212 | struct origin *o; | |
213 | ||
214 | ol = &_origins[origin_hash(origin)]; | |
215 | list_for_each_entry (o, ol, hash_list) | |
216 | if (bdev_equal(o->bdev, origin)) | |
217 | return o; | |
218 | ||
219 | return NULL; | |
220 | } | |
221 | ||
222 | static void __insert_origin(struct origin *o) | |
223 | { | |
224 | struct list_head *sl = &_origins[origin_hash(o->bdev)]; | |
225 | list_add_tail(&o->hash_list, sl); | |
226 | } | |
227 | ||
228 | /* | |
229 | * Make a note of the snapshot and its origin so we can look it | |
230 | * up when the origin has a write on it. | |
231 | */ | |
232 | static int register_snapshot(struct dm_snapshot *snap) | |
233 | { | |
60c856c8 | 234 | struct origin *o, *new_o; |
1da177e4 LT |
235 | struct block_device *bdev = snap->origin->bdev; |
236 | ||
60c856c8 MP |
237 | new_o = kmalloc(sizeof(*new_o), GFP_KERNEL); |
238 | if (!new_o) | |
239 | return -ENOMEM; | |
240 | ||
1da177e4 LT |
241 | down_write(&_origins_lock); |
242 | o = __lookup_origin(bdev); | |
243 | ||
60c856c8 MP |
244 | if (o) |
245 | kfree(new_o); | |
246 | else { | |
1da177e4 | 247 | /* New origin */ |
60c856c8 | 248 | o = new_o; |
1da177e4 LT |
249 | |
250 | /* Initialise the struct */ | |
251 | INIT_LIST_HEAD(&o->snapshots); | |
252 | o->bdev = bdev; | |
253 | ||
254 | __insert_origin(o); | |
255 | } | |
256 | ||
257 | list_add_tail(&snap->list, &o->snapshots); | |
258 | ||
259 | up_write(&_origins_lock); | |
260 | return 0; | |
261 | } | |
262 | ||
263 | static void unregister_snapshot(struct dm_snapshot *s) | |
264 | { | |
265 | struct origin *o; | |
266 | ||
267 | down_write(&_origins_lock); | |
268 | o = __lookup_origin(s->origin->bdev); | |
269 | ||
270 | list_del(&s->list); | |
271 | if (list_empty(&o->snapshots)) { | |
272 | list_del(&o->hash_list); | |
273 | kfree(o); | |
274 | } | |
275 | ||
276 | up_write(&_origins_lock); | |
277 | } | |
278 | ||
279 | /* | |
280 | * Implementation of the exception hash tables. | |
d74f81f8 MB |
281 | * The lowest hash_shift bits of the chunk number are ignored, allowing |
282 | * some consecutive chunks to be grouped together. | |
1da177e4 | 283 | */ |
d74f81f8 MB |
284 | static int init_exception_table(struct exception_table *et, uint32_t size, |
285 | unsigned hash_shift) | |
1da177e4 LT |
286 | { |
287 | unsigned int i; | |
288 | ||
d74f81f8 | 289 | et->hash_shift = hash_shift; |
1da177e4 LT |
290 | et->hash_mask = size - 1; |
291 | et->table = dm_vcalloc(size, sizeof(struct list_head)); | |
292 | if (!et->table) | |
293 | return -ENOMEM; | |
294 | ||
295 | for (i = 0; i < size; i++) | |
296 | INIT_LIST_HEAD(et->table + i); | |
297 | ||
298 | return 0; | |
299 | } | |
300 | ||
e18b890b | 301 | static void exit_exception_table(struct exception_table *et, struct kmem_cache *mem) |
1da177e4 LT |
302 | { |
303 | struct list_head *slot; | |
028867ac | 304 | struct dm_snap_exception *ex, *next; |
1da177e4 LT |
305 | int i, size; |
306 | ||
307 | size = et->hash_mask + 1; | |
308 | for (i = 0; i < size; i++) { | |
309 | slot = et->table + i; | |
310 | ||
311 | list_for_each_entry_safe (ex, next, slot, hash_list) | |
312 | kmem_cache_free(mem, ex); | |
313 | } | |
314 | ||
315 | vfree(et->table); | |
316 | } | |
317 | ||
028867ac | 318 | static uint32_t exception_hash(struct exception_table *et, chunk_t chunk) |
1da177e4 | 319 | { |
d74f81f8 | 320 | return (chunk >> et->hash_shift) & et->hash_mask; |
1da177e4 LT |
321 | } |
322 | ||
028867ac AK |
323 | static void insert_exception(struct exception_table *eh, |
324 | struct dm_snap_exception *e) | |
1da177e4 LT |
325 | { |
326 | struct list_head *l = &eh->table[exception_hash(eh, e->old_chunk)]; | |
327 | list_add(&e->hash_list, l); | |
328 | } | |
329 | ||
028867ac | 330 | static void remove_exception(struct dm_snap_exception *e) |
1da177e4 LT |
331 | { |
332 | list_del(&e->hash_list); | |
333 | } | |
334 | ||
335 | /* | |
336 | * Return the exception data for a sector, or NULL if not | |
337 | * remapped. | |
338 | */ | |
028867ac AK |
339 | static struct dm_snap_exception *lookup_exception(struct exception_table *et, |
340 | chunk_t chunk) | |
1da177e4 LT |
341 | { |
342 | struct list_head *slot; | |
028867ac | 343 | struct dm_snap_exception *e; |
1da177e4 LT |
344 | |
345 | slot = &et->table[exception_hash(et, chunk)]; | |
346 | list_for_each_entry (e, slot, hash_list) | |
d74f81f8 MB |
347 | if (chunk >= e->old_chunk && |
348 | chunk <= e->old_chunk + dm_consecutive_chunk_count(e)) | |
1da177e4 LT |
349 | return e; |
350 | ||
351 | return NULL; | |
352 | } | |
353 | ||
028867ac | 354 | static struct dm_snap_exception *alloc_exception(void) |
1da177e4 | 355 | { |
028867ac | 356 | struct dm_snap_exception *e; |
1da177e4 LT |
357 | |
358 | e = kmem_cache_alloc(exception_cache, GFP_NOIO); | |
359 | if (!e) | |
360 | e = kmem_cache_alloc(exception_cache, GFP_ATOMIC); | |
361 | ||
362 | return e; | |
363 | } | |
364 | ||
028867ac | 365 | static void free_exception(struct dm_snap_exception *e) |
1da177e4 LT |
366 | { |
367 | kmem_cache_free(exception_cache, e); | |
368 | } | |
369 | ||
92e86812 | 370 | static struct dm_snap_pending_exception *alloc_pending_exception(struct dm_snapshot *s) |
1da177e4 | 371 | { |
92e86812 MP |
372 | struct dm_snap_pending_exception *pe = mempool_alloc(s->pending_pool, |
373 | GFP_NOIO); | |
374 | ||
879129d2 | 375 | atomic_inc(&s->pending_exceptions_count); |
92e86812 MP |
376 | pe->snap = s; |
377 | ||
378 | return pe; | |
1da177e4 LT |
379 | } |
380 | ||
028867ac | 381 | static void free_pending_exception(struct dm_snap_pending_exception *pe) |
1da177e4 | 382 | { |
879129d2 MP |
383 | struct dm_snapshot *s = pe->snap; |
384 | ||
385 | mempool_free(pe, s->pending_pool); | |
386 | smp_mb__before_atomic_dec(); | |
387 | atomic_dec(&s->pending_exceptions_count); | |
1da177e4 LT |
388 | } |
389 | ||
d74f81f8 MB |
390 | static void insert_completed_exception(struct dm_snapshot *s, |
391 | struct dm_snap_exception *new_e) | |
392 | { | |
393 | struct exception_table *eh = &s->complete; | |
394 | struct list_head *l; | |
395 | struct dm_snap_exception *e = NULL; | |
396 | ||
397 | l = &eh->table[exception_hash(eh, new_e->old_chunk)]; | |
398 | ||
399 | /* Add immediately if this table doesn't support consecutive chunks */ | |
400 | if (!eh->hash_shift) | |
401 | goto out; | |
402 | ||
403 | /* List is ordered by old_chunk */ | |
404 | list_for_each_entry_reverse(e, l, hash_list) { | |
405 | /* Insert after an existing chunk? */ | |
406 | if (new_e->old_chunk == (e->old_chunk + | |
407 | dm_consecutive_chunk_count(e) + 1) && | |
408 | new_e->new_chunk == (dm_chunk_number(e->new_chunk) + | |
409 | dm_consecutive_chunk_count(e) + 1)) { | |
410 | dm_consecutive_chunk_count_inc(e); | |
411 | free_exception(new_e); | |
412 | return; | |
413 | } | |
414 | ||
415 | /* Insert before an existing chunk? */ | |
416 | if (new_e->old_chunk == (e->old_chunk - 1) && | |
417 | new_e->new_chunk == (dm_chunk_number(e->new_chunk) - 1)) { | |
418 | dm_consecutive_chunk_count_inc(e); | |
419 | e->old_chunk--; | |
420 | e->new_chunk--; | |
421 | free_exception(new_e); | |
422 | return; | |
423 | } | |
424 | ||
425 | if (new_e->old_chunk > e->old_chunk) | |
426 | break; | |
427 | } | |
428 | ||
429 | out: | |
430 | list_add(&new_e->hash_list, e ? &e->hash_list : l); | |
431 | } | |
432 | ||
a159c1ac JB |
433 | /* |
434 | * Callback used by the exception stores to load exceptions when | |
435 | * initialising. | |
436 | */ | |
437 | static int dm_add_exception(void *context, chunk_t old, chunk_t new) | |
1da177e4 | 438 | { |
a159c1ac | 439 | struct dm_snapshot *s = context; |
028867ac | 440 | struct dm_snap_exception *e; |
1da177e4 LT |
441 | |
442 | e = alloc_exception(); | |
443 | if (!e) | |
444 | return -ENOMEM; | |
445 | ||
446 | e->old_chunk = old; | |
d74f81f8 MB |
447 | |
448 | /* Consecutive_count is implicitly initialised to zero */ | |
1da177e4 | 449 | e->new_chunk = new; |
d74f81f8 MB |
450 | |
451 | insert_completed_exception(s, e); | |
452 | ||
1da177e4 LT |
453 | return 0; |
454 | } | |
455 | ||
456 | /* | |
457 | * Hard coded magic. | |
458 | */ | |
459 | static int calc_max_buckets(void) | |
460 | { | |
461 | /* use a fixed size of 2MB */ | |
462 | unsigned long mem = 2 * 1024 * 1024; | |
463 | mem /= sizeof(struct list_head); | |
464 | ||
465 | return mem; | |
466 | } | |
467 | ||
1da177e4 LT |
468 | /* |
469 | * Allocate room for a suitable hash table. | |
470 | */ | |
471 | static int init_hash_tables(struct dm_snapshot *s) | |
472 | { | |
473 | sector_t hash_size, cow_dev_size, origin_dev_size, max_buckets; | |
474 | ||
475 | /* | |
476 | * Calculate based on the size of the original volume or | |
477 | * the COW volume... | |
478 | */ | |
479 | cow_dev_size = get_dev_size(s->cow->bdev); | |
480 | origin_dev_size = get_dev_size(s->origin->bdev); | |
481 | max_buckets = calc_max_buckets(); | |
482 | ||
483 | hash_size = min(origin_dev_size, cow_dev_size) >> s->chunk_shift; | |
484 | hash_size = min(hash_size, max_buckets); | |
485 | ||
8defd830 | 486 | hash_size = rounddown_pow_of_two(hash_size); |
d74f81f8 MB |
487 | if (init_exception_table(&s->complete, hash_size, |
488 | DM_CHUNK_CONSECUTIVE_BITS)) | |
1da177e4 LT |
489 | return -ENOMEM; |
490 | ||
491 | /* | |
492 | * Allocate hash table for in-flight exceptions | |
493 | * Make this smaller than the real hash table | |
494 | */ | |
495 | hash_size >>= 3; | |
496 | if (hash_size < 64) | |
497 | hash_size = 64; | |
498 | ||
d74f81f8 | 499 | if (init_exception_table(&s->pending, hash_size, 0)) { |
1da177e4 LT |
500 | exit_exception_table(&s->complete, exception_cache); |
501 | return -ENOMEM; | |
502 | } | |
503 | ||
504 | return 0; | |
505 | } | |
506 | ||
507 | /* | |
508 | * Round a number up to the nearest 'size' boundary. size must | |
509 | * be a power of 2. | |
510 | */ | |
028867ac | 511 | static ulong round_up(ulong n, ulong size) |
1da177e4 LT |
512 | { |
513 | size--; | |
514 | return (n + size) & ~size; | |
515 | } | |
516 | ||
4c7e3bf4 MM |
517 | static int set_chunk_size(struct dm_snapshot *s, const char *chunk_size_arg, |
518 | char **error) | |
519 | { | |
520 | unsigned long chunk_size; | |
521 | char *value; | |
522 | ||
523 | chunk_size = simple_strtoul(chunk_size_arg, &value, 10); | |
524 | if (*chunk_size_arg == '\0' || *value != '\0') { | |
525 | *error = "Invalid chunk size"; | |
526 | return -EINVAL; | |
527 | } | |
528 | ||
529 | if (!chunk_size) { | |
530 | s->chunk_size = s->chunk_mask = s->chunk_shift = 0; | |
531 | return 0; | |
532 | } | |
533 | ||
534 | /* | |
535 | * Chunk size must be multiple of page size. Silently | |
536 | * round up if it's not. | |
537 | */ | |
538 | chunk_size = round_up(chunk_size, PAGE_SIZE >> 9); | |
539 | ||
540 | /* Check chunk_size is a power of 2 */ | |
6f3c3f0a | 541 | if (!is_power_of_2(chunk_size)) { |
4c7e3bf4 MM |
542 | *error = "Chunk size is not a power of 2"; |
543 | return -EINVAL; | |
544 | } | |
545 | ||
546 | /* Validate the chunk size against the device block size */ | |
547 | if (chunk_size % (bdev_hardsect_size(s->cow->bdev) >> 9)) { | |
548 | *error = "Chunk size is not a multiple of device blocksize"; | |
549 | return -EINVAL; | |
550 | } | |
551 | ||
552 | s->chunk_size = chunk_size; | |
553 | s->chunk_mask = chunk_size - 1; | |
554 | s->chunk_shift = ffs(chunk_size) - 1; | |
555 | ||
556 | return 0; | |
557 | } | |
558 | ||
1da177e4 LT |
559 | /* |
560 | * Construct a snapshot mapping: <origin_dev> <COW-dev> <p/n> <chunk-size> | |
561 | */ | |
562 | static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv) | |
563 | { | |
564 | struct dm_snapshot *s; | |
cd45daff | 565 | int i; |
1da177e4 LT |
566 | int r = -EINVAL; |
567 | char persistent; | |
568 | char *origin_path; | |
569 | char *cow_path; | |
1da177e4 | 570 | |
4c7e3bf4 | 571 | if (argc != 4) { |
72d94861 | 572 | ti->error = "requires exactly 4 arguments"; |
1da177e4 LT |
573 | r = -EINVAL; |
574 | goto bad1; | |
575 | } | |
576 | ||
577 | origin_path = argv[0]; | |
578 | cow_path = argv[1]; | |
579 | persistent = toupper(*argv[2]); | |
580 | ||
581 | if (persistent != 'P' && persistent != 'N') { | |
582 | ti->error = "Persistent flag is not P or N"; | |
583 | r = -EINVAL; | |
584 | goto bad1; | |
585 | } | |
586 | ||
1da177e4 LT |
587 | s = kmalloc(sizeof(*s), GFP_KERNEL); |
588 | if (s == NULL) { | |
589 | ti->error = "Cannot allocate snapshot context private " | |
590 | "structure"; | |
591 | r = -ENOMEM; | |
592 | goto bad1; | |
593 | } | |
594 | ||
595 | r = dm_get_device(ti, origin_path, 0, ti->len, FMODE_READ, &s->origin); | |
596 | if (r) { | |
597 | ti->error = "Cannot get origin device"; | |
598 | goto bad2; | |
599 | } | |
600 | ||
601 | r = dm_get_device(ti, cow_path, 0, 0, | |
602 | FMODE_READ | FMODE_WRITE, &s->cow); | |
603 | if (r) { | |
604 | dm_put_device(ti, s->origin); | |
605 | ti->error = "Cannot get COW device"; | |
606 | goto bad2; | |
607 | } | |
608 | ||
4c7e3bf4 MM |
609 | r = set_chunk_size(s, argv[3], &ti->error); |
610 | if (r) | |
1da177e4 | 611 | goto bad3; |
1da177e4 | 612 | |
1da177e4 | 613 | s->valid = 1; |
aa14edeb | 614 | s->active = 0; |
879129d2 | 615 | atomic_set(&s->pending_exceptions_count, 0); |
1da177e4 | 616 | init_rwsem(&s->lock); |
ca3a931f | 617 | spin_lock_init(&s->pe_lock); |
1da177e4 LT |
618 | |
619 | /* Allocate hash table for COW data */ | |
620 | if (init_hash_tables(s)) { | |
621 | ti->error = "Unable to allocate hash table space"; | |
622 | r = -ENOMEM; | |
623 | goto bad3; | |
624 | } | |
625 | ||
0cea9c78 | 626 | r = dm_exception_store_create(argv[2], ti, &s->store); |
1da177e4 LT |
627 | if (r) { |
628 | ti->error = "Couldn't create exception store"; | |
629 | r = -EINVAL; | |
630 | goto bad4; | |
631 | } | |
632 | ||
493df71c JB |
633 | s->store->snap = s; |
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 */ |
493df71c JB |
661 | r = s->store->type->read_metadata(s->store, dm_add_exception, |
662 | (void *)s); | |
0764147b | 663 | if (r < 0) { |
f9cea4f7 | 664 | ti->error = "Failed to read snapshot metadata"; |
cd45daff | 665 | goto bad_load_and_register; |
0764147b MB |
666 | } else if (r > 0) { |
667 | s->valid = 0; | |
668 | DMWARN("Snapshot is marked invalid."); | |
f9cea4f7 | 669 | } |
aa14edeb | 670 | |
ca3a931f | 671 | bio_list_init(&s->queued_bios); |
c4028958 | 672 | INIT_WORK(&s->queued_bios_work, flush_queued_bios); |
ca3a931f | 673 | |
1da177e4 | 674 | /* Add snapshot to the list of snapshots for this origin */ |
aa14edeb | 675 | /* Exceptions aren't triggered till snapshot_resume() is called */ |
1da177e4 LT |
676 | if (register_snapshot(s)) { |
677 | r = -EINVAL; | |
678 | ti->error = "Cannot register snapshot origin"; | |
cd45daff | 679 | goto bad_load_and_register; |
1da177e4 LT |
680 | } |
681 | ||
682 | ti->private = s; | |
c51c2752 | 683 | ti->split_io = s->chunk_size; |
1da177e4 LT |
684 | |
685 | return 0; | |
686 | ||
cd45daff MP |
687 | bad_load_and_register: |
688 | mempool_destroy(s->tracked_chunk_pool); | |
689 | ||
92e86812 MP |
690 | bad_tracked_chunk_pool: |
691 | mempool_destroy(s->pending_pool); | |
692 | ||
1da177e4 | 693 | bad6: |
eb69aca5 | 694 | dm_kcopyd_client_destroy(s->kcopyd_client); |
1da177e4 LT |
695 | |
696 | bad5: | |
493df71c | 697 | s->store->type->dtr(s->store); |
1da177e4 LT |
698 | |
699 | bad4: | |
700 | exit_exception_table(&s->pending, pending_cache); | |
701 | exit_exception_table(&s->complete, exception_cache); | |
702 | ||
703 | bad3: | |
704 | dm_put_device(ti, s->cow); | |
705 | dm_put_device(ti, s->origin); | |
706 | ||
707 | bad2: | |
708 | kfree(s); | |
709 | ||
710 | bad1: | |
711 | return r; | |
712 | } | |
713 | ||
31c93a0c MB |
714 | static void __free_exceptions(struct dm_snapshot *s) |
715 | { | |
eb69aca5 | 716 | dm_kcopyd_client_destroy(s->kcopyd_client); |
31c93a0c MB |
717 | s->kcopyd_client = NULL; |
718 | ||
719 | exit_exception_table(&s->pending, pending_cache); | |
720 | exit_exception_table(&s->complete, exception_cache); | |
721 | ||
493df71c | 722 | s->store->type->dtr(s->store); |
31c93a0c MB |
723 | } |
724 | ||
1da177e4 LT |
725 | static void snapshot_dtr(struct dm_target *ti) |
726 | { | |
cd45daff MP |
727 | #ifdef CONFIG_DM_DEBUG |
728 | int i; | |
729 | #endif | |
028867ac | 730 | struct dm_snapshot *s = ti->private; |
1da177e4 | 731 | |
ca3a931f AK |
732 | flush_workqueue(ksnapd); |
733 | ||
138728dc AK |
734 | /* Prevent further origin writes from using this snapshot. */ |
735 | /* After this returns there can be no new kcopyd jobs. */ | |
1da177e4 LT |
736 | unregister_snapshot(s); |
737 | ||
879129d2 | 738 | while (atomic_read(&s->pending_exceptions_count)) |
90fa1527 | 739 | msleep(1); |
879129d2 MP |
740 | /* |
741 | * Ensure instructions in mempool_destroy aren't reordered | |
742 | * before atomic_read. | |
743 | */ | |
744 | smp_mb(); | |
745 | ||
cd45daff MP |
746 | #ifdef CONFIG_DM_DEBUG |
747 | for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++) | |
748 | BUG_ON(!hlist_empty(&s->tracked_chunk_hash[i])); | |
749 | #endif | |
750 | ||
751 | mempool_destroy(s->tracked_chunk_pool); | |
752 | ||
31c93a0c | 753 | __free_exceptions(s); |
1da177e4 | 754 | |
92e86812 MP |
755 | mempool_destroy(s->pending_pool); |
756 | ||
1da177e4 LT |
757 | dm_put_device(ti, s->origin); |
758 | dm_put_device(ti, s->cow); | |
138728dc | 759 | |
1da177e4 LT |
760 | kfree(s); |
761 | } | |
762 | ||
763 | /* | |
764 | * Flush a list of buffers. | |
765 | */ | |
766 | static void flush_bios(struct bio *bio) | |
767 | { | |
768 | struct bio *n; | |
769 | ||
770 | while (bio) { | |
771 | n = bio->bi_next; | |
772 | bio->bi_next = NULL; | |
773 | generic_make_request(bio); | |
774 | bio = n; | |
775 | } | |
776 | } | |
777 | ||
c4028958 | 778 | static void flush_queued_bios(struct work_struct *work) |
ca3a931f | 779 | { |
c4028958 DH |
780 | struct dm_snapshot *s = |
781 | container_of(work, struct dm_snapshot, queued_bios_work); | |
ca3a931f AK |
782 | struct bio *queued_bios; |
783 | unsigned long flags; | |
784 | ||
785 | spin_lock_irqsave(&s->pe_lock, flags); | |
786 | queued_bios = bio_list_get(&s->queued_bios); | |
787 | spin_unlock_irqrestore(&s->pe_lock, flags); | |
788 | ||
789 | flush_bios(queued_bios); | |
790 | } | |
791 | ||
1da177e4 LT |
792 | /* |
793 | * Error a list of buffers. | |
794 | */ | |
795 | static void error_bios(struct bio *bio) | |
796 | { | |
797 | struct bio *n; | |
798 | ||
799 | while (bio) { | |
800 | n = bio->bi_next; | |
801 | bio->bi_next = NULL; | |
6712ecf8 | 802 | bio_io_error(bio); |
1da177e4 LT |
803 | bio = n; |
804 | } | |
805 | } | |
806 | ||
695368ac | 807 | static void __invalidate_snapshot(struct dm_snapshot *s, int err) |
76df1c65 AK |
808 | { |
809 | if (!s->valid) | |
810 | return; | |
811 | ||
812 | if (err == -EIO) | |
813 | DMERR("Invalidating snapshot: Error reading/writing."); | |
814 | else if (err == -ENOMEM) | |
815 | DMERR("Invalidating snapshot: Unable to allocate exception."); | |
816 | ||
493df71c JB |
817 | if (s->store->type->drop_snapshot) |
818 | s->store->type->drop_snapshot(s->store); | |
76df1c65 AK |
819 | |
820 | s->valid = 0; | |
821 | ||
0cea9c78 | 822 | dm_table_event(s->store->ti->table); |
76df1c65 AK |
823 | } |
824 | ||
028867ac | 825 | static void get_pending_exception(struct dm_snap_pending_exception *pe) |
4b832e8d AK |
826 | { |
827 | atomic_inc(&pe->ref_count); | |
828 | } | |
829 | ||
028867ac | 830 | static struct bio *put_pending_exception(struct dm_snap_pending_exception *pe) |
4b832e8d | 831 | { |
028867ac | 832 | struct dm_snap_pending_exception *primary_pe; |
4b832e8d AK |
833 | struct bio *origin_bios = NULL; |
834 | ||
835 | primary_pe = pe->primary_pe; | |
836 | ||
837 | /* | |
838 | * If this pe is involved in a write to the origin and | |
839 | * it is the last sibling to complete then release | |
840 | * the bios for the original write to the origin. | |
841 | */ | |
842 | if (primary_pe && | |
7c5f78b9 | 843 | atomic_dec_and_test(&primary_pe->ref_count)) { |
4b832e8d | 844 | origin_bios = bio_list_get(&primary_pe->origin_bios); |
7c5f78b9 MP |
845 | free_pending_exception(primary_pe); |
846 | } | |
4b832e8d AK |
847 | |
848 | /* | |
849 | * Free the pe if it's not linked to an origin write or if | |
850 | * it's not itself a primary pe. | |
851 | */ | |
852 | if (!primary_pe || primary_pe != pe) | |
853 | free_pending_exception(pe); | |
854 | ||
4b832e8d AK |
855 | return origin_bios; |
856 | } | |
857 | ||
028867ac | 858 | static void pending_complete(struct dm_snap_pending_exception *pe, int success) |
1da177e4 | 859 | { |
028867ac | 860 | struct dm_snap_exception *e; |
1da177e4 | 861 | struct dm_snapshot *s = pe->snap; |
9d493fa8 AK |
862 | struct bio *origin_bios = NULL; |
863 | struct bio *snapshot_bios = NULL; | |
864 | int error = 0; | |
1da177e4 | 865 | |
76df1c65 AK |
866 | if (!success) { |
867 | /* Read/write error - snapshot is unusable */ | |
1da177e4 | 868 | down_write(&s->lock); |
695368ac | 869 | __invalidate_snapshot(s, -EIO); |
9d493fa8 | 870 | error = 1; |
76df1c65 AK |
871 | goto out; |
872 | } | |
873 | ||
874 | e = alloc_exception(); | |
875 | if (!e) { | |
1da177e4 | 876 | down_write(&s->lock); |
695368ac | 877 | __invalidate_snapshot(s, -ENOMEM); |
9d493fa8 | 878 | error = 1; |
76df1c65 AK |
879 | goto out; |
880 | } | |
881 | *e = pe->e; | |
1da177e4 | 882 | |
76df1c65 AK |
883 | down_write(&s->lock); |
884 | if (!s->valid) { | |
76df1c65 | 885 | free_exception(e); |
9d493fa8 | 886 | error = 1; |
76df1c65 | 887 | goto out; |
1da177e4 LT |
888 | } |
889 | ||
a8d41b59 MP |
890 | /* |
891 | * Check for conflicting reads. This is extremely improbable, | |
90fa1527 | 892 | * so msleep(1) is sufficient and there is no need for a wait queue. |
a8d41b59 MP |
893 | */ |
894 | while (__chunk_is_tracked(s, pe->e.old_chunk)) | |
90fa1527 | 895 | msleep(1); |
a8d41b59 | 896 | |
9d493fa8 AK |
897 | /* |
898 | * Add a proper exception, and remove the | |
899 | * in-flight exception from the list. | |
900 | */ | |
d74f81f8 | 901 | insert_completed_exception(s, e); |
76df1c65 | 902 | |
1da177e4 | 903 | out: |
695368ac | 904 | remove_exception(&pe->e); |
9d493fa8 | 905 | snapshot_bios = bio_list_get(&pe->snapshot_bios); |
4b832e8d | 906 | origin_bios = put_pending_exception(pe); |
1da177e4 | 907 | |
9d493fa8 AK |
908 | up_write(&s->lock); |
909 | ||
910 | /* Submit any pending write bios */ | |
911 | if (error) | |
912 | error_bios(snapshot_bios); | |
913 | else | |
914 | flush_bios(snapshot_bios); | |
915 | ||
916 | flush_bios(origin_bios); | |
1da177e4 LT |
917 | } |
918 | ||
919 | static void commit_callback(void *context, int success) | |
920 | { | |
028867ac AK |
921 | struct dm_snap_pending_exception *pe = context; |
922 | ||
1da177e4 LT |
923 | pending_complete(pe, success); |
924 | } | |
925 | ||
926 | /* | |
927 | * Called when the copy I/O has finished. kcopyd actually runs | |
928 | * this code so don't block. | |
929 | */ | |
4cdc1d1f | 930 | static void copy_callback(int read_err, unsigned long write_err, void *context) |
1da177e4 | 931 | { |
028867ac | 932 | struct dm_snap_pending_exception *pe = context; |
1da177e4 LT |
933 | struct dm_snapshot *s = pe->snap; |
934 | ||
935 | if (read_err || write_err) | |
936 | pending_complete(pe, 0); | |
937 | ||
938 | else | |
939 | /* Update the metadata if we are persistent */ | |
493df71c JB |
940 | s->store->type->commit_exception(s->store, &pe->e, |
941 | commit_callback, pe); | |
1da177e4 LT |
942 | } |
943 | ||
944 | /* | |
945 | * Dispatches the copy operation to kcopyd. | |
946 | */ | |
028867ac | 947 | static void start_copy(struct dm_snap_pending_exception *pe) |
1da177e4 LT |
948 | { |
949 | struct dm_snapshot *s = pe->snap; | |
22a1ceb1 | 950 | struct dm_io_region src, dest; |
1da177e4 LT |
951 | struct block_device *bdev = s->origin->bdev; |
952 | sector_t dev_size; | |
953 | ||
954 | dev_size = get_dev_size(bdev); | |
955 | ||
956 | src.bdev = bdev; | |
957 | src.sector = chunk_to_sector(s, pe->e.old_chunk); | |
958 | src.count = min(s->chunk_size, dev_size - src.sector); | |
959 | ||
960 | dest.bdev = s->cow->bdev; | |
961 | dest.sector = chunk_to_sector(s, pe->e.new_chunk); | |
962 | dest.count = src.count; | |
963 | ||
964 | /* Hand over to kcopyd */ | |
eb69aca5 | 965 | dm_kcopyd_copy(s->kcopyd_client, |
1da177e4 LT |
966 | &src, 1, &dest, 0, copy_callback, pe); |
967 | } | |
968 | ||
2913808e MP |
969 | static struct dm_snap_pending_exception * |
970 | __lookup_pending_exception(struct dm_snapshot *s, chunk_t chunk) | |
971 | { | |
972 | struct dm_snap_exception *e = lookup_exception(&s->pending, chunk); | |
973 | ||
974 | if (!e) | |
975 | return NULL; | |
976 | ||
977 | return container_of(e, struct dm_snap_pending_exception, e); | |
978 | } | |
979 | ||
1da177e4 LT |
980 | /* |
981 | * Looks to see if this snapshot already has a pending exception | |
982 | * for this chunk, otherwise it allocates a new one and inserts | |
983 | * it into the pending table. | |
984 | * | |
985 | * NOTE: a write lock must be held on snap->lock before calling | |
986 | * this. | |
987 | */ | |
028867ac | 988 | static struct dm_snap_pending_exception * |
c6621392 MP |
989 | __find_pending_exception(struct dm_snapshot *s, |
990 | struct dm_snap_pending_exception *pe, chunk_t chunk) | |
1da177e4 | 991 | { |
c6621392 | 992 | struct dm_snap_pending_exception *pe2; |
1da177e4 | 993 | |
2913808e MP |
994 | pe2 = __lookup_pending_exception(s, chunk); |
995 | if (pe2) { | |
76df1c65 | 996 | free_pending_exception(pe); |
2913808e | 997 | return pe2; |
1da177e4 LT |
998 | } |
999 | ||
76df1c65 AK |
1000 | pe->e.old_chunk = chunk; |
1001 | bio_list_init(&pe->origin_bios); | |
1002 | bio_list_init(&pe->snapshot_bios); | |
1003 | pe->primary_pe = NULL; | |
4b832e8d | 1004 | atomic_set(&pe->ref_count, 0); |
76df1c65 AK |
1005 | pe->started = 0; |
1006 | ||
493df71c | 1007 | if (s->store->type->prepare_exception(s->store, &pe->e)) { |
76df1c65 AK |
1008 | free_pending_exception(pe); |
1009 | return NULL; | |
1010 | } | |
1011 | ||
4b832e8d | 1012 | get_pending_exception(pe); |
76df1c65 AK |
1013 | insert_exception(&s->pending, &pe->e); |
1014 | ||
1da177e4 LT |
1015 | return pe; |
1016 | } | |
1017 | ||
028867ac | 1018 | static void remap_exception(struct dm_snapshot *s, struct dm_snap_exception *e, |
d74f81f8 | 1019 | struct bio *bio, chunk_t chunk) |
1da177e4 LT |
1020 | { |
1021 | bio->bi_bdev = s->cow->bdev; | |
d74f81f8 MB |
1022 | bio->bi_sector = chunk_to_sector(s, dm_chunk_number(e->new_chunk) + |
1023 | (chunk - e->old_chunk)) + | |
1024 | (bio->bi_sector & s->chunk_mask); | |
1da177e4 LT |
1025 | } |
1026 | ||
1027 | static int snapshot_map(struct dm_target *ti, struct bio *bio, | |
1028 | union map_info *map_context) | |
1029 | { | |
028867ac AK |
1030 | struct dm_snap_exception *e; |
1031 | struct dm_snapshot *s = ti->private; | |
d2a7ad29 | 1032 | int r = DM_MAPIO_REMAPPED; |
1da177e4 | 1033 | chunk_t chunk; |
028867ac | 1034 | struct dm_snap_pending_exception *pe = NULL; |
1da177e4 LT |
1035 | |
1036 | chunk = sector_to_chunk(s, bio->bi_sector); | |
1037 | ||
1038 | /* Full snapshots are not usable */ | |
76df1c65 | 1039 | /* To get here the table must be live so s->active is always set. */ |
1da177e4 | 1040 | if (!s->valid) |
f6a80ea8 | 1041 | return -EIO; |
1da177e4 | 1042 | |
ba40a2aa AK |
1043 | /* FIXME: should only take write lock if we need |
1044 | * to copy an exception */ | |
1045 | down_write(&s->lock); | |
1046 | ||
1047 | if (!s->valid) { | |
1048 | r = -EIO; | |
1049 | goto out_unlock; | |
1050 | } | |
1051 | ||
1052 | /* If the block is already remapped - use that, else remap it */ | |
1053 | e = lookup_exception(&s->complete, chunk); | |
1054 | if (e) { | |
d74f81f8 | 1055 | remap_exception(s, e, bio, chunk); |
ba40a2aa AK |
1056 | goto out_unlock; |
1057 | } | |
1058 | ||
1da177e4 LT |
1059 | /* |
1060 | * Write to snapshot - higher level takes care of RW/RO | |
1061 | * flags so we should only get this if we are | |
1062 | * writeable. | |
1063 | */ | |
1064 | if (bio_rw(bio) == WRITE) { | |
2913808e | 1065 | pe = __lookup_pending_exception(s, chunk); |
76df1c65 | 1066 | if (!pe) { |
c6621392 MP |
1067 | up_write(&s->lock); |
1068 | pe = alloc_pending_exception(s); | |
1069 | down_write(&s->lock); | |
1070 | ||
1071 | if (!s->valid) { | |
1072 | free_pending_exception(pe); | |
1073 | r = -EIO; | |
1074 | goto out_unlock; | |
1075 | } | |
1076 | ||
35bf659b MP |
1077 | e = lookup_exception(&s->complete, chunk); |
1078 | if (e) { | |
1079 | free_pending_exception(pe); | |
1080 | remap_exception(s, e, bio, chunk); | |
1081 | goto out_unlock; | |
1082 | } | |
1083 | ||
c6621392 | 1084 | pe = __find_pending_exception(s, pe, chunk); |
2913808e MP |
1085 | if (!pe) { |
1086 | __invalidate_snapshot(s, -ENOMEM); | |
1087 | r = -EIO; | |
1088 | goto out_unlock; | |
1089 | } | |
1da177e4 LT |
1090 | } |
1091 | ||
d74f81f8 | 1092 | remap_exception(s, &pe->e, bio, chunk); |
76df1c65 AK |
1093 | bio_list_add(&pe->snapshot_bios, bio); |
1094 | ||
d2a7ad29 | 1095 | r = DM_MAPIO_SUBMITTED; |
ba40a2aa | 1096 | |
76df1c65 AK |
1097 | if (!pe->started) { |
1098 | /* this is protected by snap->lock */ | |
1099 | pe->started = 1; | |
ba40a2aa | 1100 | up_write(&s->lock); |
76df1c65 | 1101 | start_copy(pe); |
ba40a2aa AK |
1102 | goto out; |
1103 | } | |
cd45daff | 1104 | } else { |
ba40a2aa | 1105 | bio->bi_bdev = s->origin->bdev; |
cd45daff MP |
1106 | map_context->ptr = track_chunk(s, chunk); |
1107 | } | |
1da177e4 | 1108 | |
ba40a2aa AK |
1109 | out_unlock: |
1110 | up_write(&s->lock); | |
1111 | out: | |
1da177e4 LT |
1112 | return r; |
1113 | } | |
1114 | ||
cd45daff MP |
1115 | static int snapshot_end_io(struct dm_target *ti, struct bio *bio, |
1116 | int error, union map_info *map_context) | |
1117 | { | |
1118 | struct dm_snapshot *s = ti->private; | |
1119 | struct dm_snap_tracked_chunk *c = map_context->ptr; | |
1120 | ||
1121 | if (c) | |
1122 | stop_tracking_chunk(s, c); | |
1123 | ||
1124 | return 0; | |
1125 | } | |
1126 | ||
1da177e4 LT |
1127 | static void snapshot_resume(struct dm_target *ti) |
1128 | { | |
028867ac | 1129 | struct dm_snapshot *s = ti->private; |
1da177e4 | 1130 | |
aa14edeb AK |
1131 | down_write(&s->lock); |
1132 | s->active = 1; | |
1133 | up_write(&s->lock); | |
1da177e4 LT |
1134 | } |
1135 | ||
1136 | static int snapshot_status(struct dm_target *ti, status_type_t type, | |
1137 | char *result, unsigned int maxlen) | |
1138 | { | |
028867ac | 1139 | struct dm_snapshot *snap = ti->private; |
1da177e4 LT |
1140 | |
1141 | switch (type) { | |
1142 | case STATUSTYPE_INFO: | |
1143 | if (!snap->valid) | |
1144 | snprintf(result, maxlen, "Invalid"); | |
1145 | else { | |
493df71c | 1146 | if (snap->store->type->fraction_full) { |
1da177e4 | 1147 | sector_t numerator, denominator; |
493df71c JB |
1148 | snap->store->type->fraction_full(snap->store, |
1149 | &numerator, | |
1150 | &denominator); | |
4ee218cd AM |
1151 | snprintf(result, maxlen, "%llu/%llu", |
1152 | (unsigned long long)numerator, | |
1153 | (unsigned long long)denominator); | |
1da177e4 LT |
1154 | } |
1155 | else | |
1156 | snprintf(result, maxlen, "Unknown"); | |
1157 | } | |
1158 | break; | |
1159 | ||
1160 | case STATUSTYPE_TABLE: | |
1161 | /* | |
1162 | * kdevname returns a static pointer so we need | |
1163 | * to make private copies if the output is to | |
1164 | * make sense. | |
1165 | */ | |
493df71c | 1166 | snprintf(result, maxlen, "%s %s %s %llu", |
1da177e4 | 1167 | snap->origin->name, snap->cow->name, |
493df71c | 1168 | snap->store->type->name, |
4ee218cd | 1169 | (unsigned long long)snap->chunk_size); |
1da177e4 LT |
1170 | break; |
1171 | } | |
1172 | ||
1173 | return 0; | |
1174 | } | |
1175 | ||
1176 | /*----------------------------------------------------------------- | |
1177 | * Origin methods | |
1178 | *---------------------------------------------------------------*/ | |
1da177e4 LT |
1179 | static int __origin_write(struct list_head *snapshots, struct bio *bio) |
1180 | { | |
d2a7ad29 | 1181 | int r = DM_MAPIO_REMAPPED, first = 0; |
1da177e4 | 1182 | struct dm_snapshot *snap; |
028867ac AK |
1183 | struct dm_snap_exception *e; |
1184 | struct dm_snap_pending_exception *pe, *next_pe, *primary_pe = NULL; | |
1da177e4 | 1185 | chunk_t chunk; |
eccf0817 | 1186 | LIST_HEAD(pe_queue); |
1da177e4 LT |
1187 | |
1188 | /* Do all the snapshots on this origin */ | |
1189 | list_for_each_entry (snap, snapshots, list) { | |
1190 | ||
76df1c65 AK |
1191 | down_write(&snap->lock); |
1192 | ||
aa14edeb AK |
1193 | /* Only deal with valid and active snapshots */ |
1194 | if (!snap->valid || !snap->active) | |
76df1c65 | 1195 | goto next_snapshot; |
1da177e4 | 1196 | |
d5e404c1 | 1197 | /* Nothing to do if writing beyond end of snapshot */ |
0cea9c78 | 1198 | if (bio->bi_sector >= dm_table_get_size(snap->store->ti->table)) |
76df1c65 | 1199 | goto next_snapshot; |
1da177e4 LT |
1200 | |
1201 | /* | |
1202 | * Remember, different snapshots can have | |
1203 | * different chunk sizes. | |
1204 | */ | |
1205 | chunk = sector_to_chunk(snap, bio->bi_sector); | |
1206 | ||
1207 | /* | |
1208 | * Check exception table to see if block | |
1209 | * is already remapped in this snapshot | |
1210 | * and trigger an exception if not. | |
b4b610f6 | 1211 | * |
4b832e8d | 1212 | * ref_count is initialised to 1 so pending_complete() |
b4b610f6 | 1213 | * won't destroy the primary_pe while we're inside this loop. |
1da177e4 LT |
1214 | */ |
1215 | e = lookup_exception(&snap->complete, chunk); | |
76df1c65 AK |
1216 | if (e) |
1217 | goto next_snapshot; | |
1218 | ||
2913808e | 1219 | pe = __lookup_pending_exception(snap, chunk); |
76df1c65 | 1220 | if (!pe) { |
c6621392 MP |
1221 | up_write(&snap->lock); |
1222 | pe = alloc_pending_exception(snap); | |
1223 | down_write(&snap->lock); | |
1224 | ||
1225 | if (!snap->valid) { | |
1226 | free_pending_exception(pe); | |
1227 | goto next_snapshot; | |
1228 | } | |
1229 | ||
35bf659b MP |
1230 | e = lookup_exception(&snap->complete, chunk); |
1231 | if (e) { | |
1232 | free_pending_exception(pe); | |
1233 | goto next_snapshot; | |
1234 | } | |
1235 | ||
c6621392 | 1236 | pe = __find_pending_exception(snap, pe, chunk); |
2913808e MP |
1237 | if (!pe) { |
1238 | __invalidate_snapshot(snap, -ENOMEM); | |
1239 | goto next_snapshot; | |
1240 | } | |
76df1c65 AK |
1241 | } |
1242 | ||
1243 | if (!primary_pe) { | |
1244 | /* | |
1245 | * Either every pe here has same | |
1246 | * primary_pe or none has one yet. | |
1247 | */ | |
1248 | if (pe->primary_pe) | |
1249 | primary_pe = pe->primary_pe; | |
1250 | else { | |
1251 | primary_pe = pe; | |
1252 | first = 1; | |
1da177e4 | 1253 | } |
76df1c65 AK |
1254 | |
1255 | bio_list_add(&primary_pe->origin_bios, bio); | |
1256 | ||
d2a7ad29 | 1257 | r = DM_MAPIO_SUBMITTED; |
76df1c65 AK |
1258 | } |
1259 | ||
1260 | if (!pe->primary_pe) { | |
76df1c65 | 1261 | pe->primary_pe = primary_pe; |
4b832e8d | 1262 | get_pending_exception(primary_pe); |
76df1c65 AK |
1263 | } |
1264 | ||
1265 | if (!pe->started) { | |
1266 | pe->started = 1; | |
1267 | list_add_tail(&pe->list, &pe_queue); | |
1da177e4 LT |
1268 | } |
1269 | ||
76df1c65 | 1270 | next_snapshot: |
1da177e4 LT |
1271 | up_write(&snap->lock); |
1272 | } | |
1273 | ||
b4b610f6 | 1274 | if (!primary_pe) |
4b832e8d | 1275 | return r; |
b4b610f6 AK |
1276 | |
1277 | /* | |
1278 | * If this is the first time we're processing this chunk and | |
4b832e8d | 1279 | * ref_count is now 1 it means all the pending exceptions |
b4b610f6 AK |
1280 | * got completed while we were in the loop above, so it falls to |
1281 | * us here to remove the primary_pe and submit any origin_bios. | |
1282 | */ | |
1283 | ||
4b832e8d | 1284 | if (first && atomic_dec_and_test(&primary_pe->ref_count)) { |
b4b610f6 AK |
1285 | flush_bios(bio_list_get(&primary_pe->origin_bios)); |
1286 | free_pending_exception(primary_pe); | |
1287 | /* If we got here, pe_queue is necessarily empty. */ | |
4b832e8d | 1288 | return r; |
b4b610f6 AK |
1289 | } |
1290 | ||
1da177e4 LT |
1291 | /* |
1292 | * Now that we have a complete pe list we can start the copying. | |
1293 | */ | |
eccf0817 AK |
1294 | list_for_each_entry_safe(pe, next_pe, &pe_queue, list) |
1295 | start_copy(pe); | |
1da177e4 LT |
1296 | |
1297 | return r; | |
1298 | } | |
1299 | ||
1300 | /* | |
1301 | * Called on a write from the origin driver. | |
1302 | */ | |
1303 | static int do_origin(struct dm_dev *origin, struct bio *bio) | |
1304 | { | |
1305 | struct origin *o; | |
d2a7ad29 | 1306 | int r = DM_MAPIO_REMAPPED; |
1da177e4 LT |
1307 | |
1308 | down_read(&_origins_lock); | |
1309 | o = __lookup_origin(origin->bdev); | |
1310 | if (o) | |
1311 | r = __origin_write(&o->snapshots, bio); | |
1312 | up_read(&_origins_lock); | |
1313 | ||
1314 | return r; | |
1315 | } | |
1316 | ||
1317 | /* | |
1318 | * Origin: maps a linear range of a device, with hooks for snapshotting. | |
1319 | */ | |
1320 | ||
1321 | /* | |
1322 | * Construct an origin mapping: <dev_path> | |
1323 | * The context for an origin is merely a 'struct dm_dev *' | |
1324 | * pointing to the real device. | |
1325 | */ | |
1326 | static int origin_ctr(struct dm_target *ti, unsigned int argc, char **argv) | |
1327 | { | |
1328 | int r; | |
1329 | struct dm_dev *dev; | |
1330 | ||
1331 | if (argc != 1) { | |
72d94861 | 1332 | ti->error = "origin: incorrect number of arguments"; |
1da177e4 LT |
1333 | return -EINVAL; |
1334 | } | |
1335 | ||
1336 | r = dm_get_device(ti, argv[0], 0, ti->len, | |
1337 | dm_table_get_mode(ti->table), &dev); | |
1338 | if (r) { | |
1339 | ti->error = "Cannot get target device"; | |
1340 | return r; | |
1341 | } | |
1342 | ||
1343 | ti->private = dev; | |
1344 | return 0; | |
1345 | } | |
1346 | ||
1347 | static void origin_dtr(struct dm_target *ti) | |
1348 | { | |
028867ac | 1349 | struct dm_dev *dev = ti->private; |
1da177e4 LT |
1350 | dm_put_device(ti, dev); |
1351 | } | |
1352 | ||
1353 | static int origin_map(struct dm_target *ti, struct bio *bio, | |
1354 | union map_info *map_context) | |
1355 | { | |
028867ac | 1356 | struct dm_dev *dev = ti->private; |
1da177e4 LT |
1357 | bio->bi_bdev = dev->bdev; |
1358 | ||
1359 | /* Only tell snapshots if this is a write */ | |
d2a7ad29 | 1360 | return (bio_rw(bio) == WRITE) ? do_origin(dev, bio) : DM_MAPIO_REMAPPED; |
1da177e4 LT |
1361 | } |
1362 | ||
1363 | #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r)) | |
1364 | ||
1365 | /* | |
1366 | * Set the target "split_io" field to the minimum of all the snapshots' | |
1367 | * chunk sizes. | |
1368 | */ | |
1369 | static void origin_resume(struct dm_target *ti) | |
1370 | { | |
028867ac | 1371 | struct dm_dev *dev = ti->private; |
1da177e4 LT |
1372 | struct dm_snapshot *snap; |
1373 | struct origin *o; | |
1374 | chunk_t chunk_size = 0; | |
1375 | ||
1376 | down_read(&_origins_lock); | |
1377 | o = __lookup_origin(dev->bdev); | |
1378 | if (o) | |
1379 | list_for_each_entry (snap, &o->snapshots, list) | |
1380 | chunk_size = min_not_zero(chunk_size, snap->chunk_size); | |
1381 | up_read(&_origins_lock); | |
1382 | ||
1383 | ti->split_io = chunk_size; | |
1384 | } | |
1385 | ||
1386 | static int origin_status(struct dm_target *ti, status_type_t type, char *result, | |
1387 | unsigned int maxlen) | |
1388 | { | |
028867ac | 1389 | struct dm_dev *dev = ti->private; |
1da177e4 LT |
1390 | |
1391 | switch (type) { | |
1392 | case STATUSTYPE_INFO: | |
1393 | result[0] = '\0'; | |
1394 | break; | |
1395 | ||
1396 | case STATUSTYPE_TABLE: | |
1397 | snprintf(result, maxlen, "%s", dev->name); | |
1398 | break; | |
1399 | } | |
1400 | ||
1401 | return 0; | |
1402 | } | |
1403 | ||
1404 | static struct target_type origin_target = { | |
1405 | .name = "snapshot-origin", | |
d74f81f8 | 1406 | .version = {1, 6, 0}, |
1da177e4 LT |
1407 | .module = THIS_MODULE, |
1408 | .ctr = origin_ctr, | |
1409 | .dtr = origin_dtr, | |
1410 | .map = origin_map, | |
1411 | .resume = origin_resume, | |
1412 | .status = origin_status, | |
1413 | }; | |
1414 | ||
1415 | static struct target_type snapshot_target = { | |
1416 | .name = "snapshot", | |
d74f81f8 | 1417 | .version = {1, 6, 0}, |
1da177e4 LT |
1418 | .module = THIS_MODULE, |
1419 | .ctr = snapshot_ctr, | |
1420 | .dtr = snapshot_dtr, | |
1421 | .map = snapshot_map, | |
cd45daff | 1422 | .end_io = snapshot_end_io, |
1da177e4 LT |
1423 | .resume = snapshot_resume, |
1424 | .status = snapshot_status, | |
1425 | }; | |
1426 | ||
1427 | static int __init dm_snapshot_init(void) | |
1428 | { | |
1429 | int r; | |
1430 | ||
4db6bfe0 AK |
1431 | r = dm_exception_store_init(); |
1432 | if (r) { | |
1433 | DMERR("Failed to initialize exception stores"); | |
1434 | return r; | |
1435 | } | |
1436 | ||
1da177e4 LT |
1437 | r = dm_register_target(&snapshot_target); |
1438 | if (r) { | |
1439 | DMERR("snapshot target register failed %d", r); | |
1440 | return r; | |
1441 | } | |
1442 | ||
1443 | r = dm_register_target(&origin_target); | |
1444 | if (r < 0) { | |
72d94861 | 1445 | DMERR("Origin target register failed %d", r); |
1da177e4 LT |
1446 | goto bad1; |
1447 | } | |
1448 | ||
1449 | r = init_origin_hash(); | |
1450 | if (r) { | |
1451 | DMERR("init_origin_hash failed."); | |
1452 | goto bad2; | |
1453 | } | |
1454 | ||
028867ac | 1455 | exception_cache = KMEM_CACHE(dm_snap_exception, 0); |
1da177e4 LT |
1456 | if (!exception_cache) { |
1457 | DMERR("Couldn't create exception cache."); | |
1458 | r = -ENOMEM; | |
1459 | goto bad3; | |
1460 | } | |
1461 | ||
028867ac | 1462 | pending_cache = KMEM_CACHE(dm_snap_pending_exception, 0); |
1da177e4 LT |
1463 | if (!pending_cache) { |
1464 | DMERR("Couldn't create pending cache."); | |
1465 | r = -ENOMEM; | |
1466 | goto bad4; | |
1467 | } | |
1468 | ||
cd45daff MP |
1469 | tracked_chunk_cache = KMEM_CACHE(dm_snap_tracked_chunk, 0); |
1470 | if (!tracked_chunk_cache) { | |
1471 | DMERR("Couldn't create cache to track chunks in use."); | |
1472 | r = -ENOMEM; | |
1473 | goto bad5; | |
1474 | } | |
1475 | ||
ca3a931f AK |
1476 | ksnapd = create_singlethread_workqueue("ksnapd"); |
1477 | if (!ksnapd) { | |
1478 | DMERR("Failed to create ksnapd workqueue."); | |
1479 | r = -ENOMEM; | |
92e86812 | 1480 | goto bad_pending_pool; |
ca3a931f AK |
1481 | } |
1482 | ||
1da177e4 LT |
1483 | return 0; |
1484 | ||
4db6bfe0 | 1485 | bad_pending_pool: |
cd45daff | 1486 | kmem_cache_destroy(tracked_chunk_cache); |
4db6bfe0 | 1487 | bad5: |
1da177e4 | 1488 | kmem_cache_destroy(pending_cache); |
4db6bfe0 | 1489 | bad4: |
1da177e4 | 1490 | kmem_cache_destroy(exception_cache); |
4db6bfe0 | 1491 | bad3: |
1da177e4 | 1492 | exit_origin_hash(); |
4db6bfe0 | 1493 | bad2: |
1da177e4 | 1494 | dm_unregister_target(&origin_target); |
4db6bfe0 | 1495 | bad1: |
1da177e4 LT |
1496 | dm_unregister_target(&snapshot_target); |
1497 | return r; | |
1498 | } | |
1499 | ||
1500 | static void __exit dm_snapshot_exit(void) | |
1501 | { | |
ca3a931f AK |
1502 | destroy_workqueue(ksnapd); |
1503 | ||
10d3bd09 MP |
1504 | dm_unregister_target(&snapshot_target); |
1505 | dm_unregister_target(&origin_target); | |
1da177e4 LT |
1506 | |
1507 | exit_origin_hash(); | |
1da177e4 LT |
1508 | kmem_cache_destroy(pending_cache); |
1509 | kmem_cache_destroy(exception_cache); | |
cd45daff | 1510 | kmem_cache_destroy(tracked_chunk_cache); |
4db6bfe0 AK |
1511 | |
1512 | dm_exception_store_exit(); | |
1da177e4 LT |
1513 | } |
1514 | ||
1515 | /* Module hooks */ | |
1516 | module_init(dm_snapshot_init); | |
1517 | module_exit(dm_snapshot_exit); | |
1518 | ||
1519 | MODULE_DESCRIPTION(DM_NAME " snapshot target"); | |
1520 | MODULE_AUTHOR("Joe Thornber"); | |
1521 | MODULE_LICENSE("GPL"); |