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 LT |
23 | |
24 | #include "dm-snap.h" | |
25 | #include "dm-bio-list.h" | |
1da177e4 | 26 | |
72d94861 AK |
27 | #define DM_MSG_PREFIX "snapshots" |
28 | ||
1da177e4 LT |
29 | /* |
30 | * The percentage increment we will wake up users at | |
31 | */ | |
32 | #define WAKE_UP_PERCENT 5 | |
33 | ||
34 | /* | |
35 | * kcopyd priority of snapshot operations | |
36 | */ | |
37 | #define SNAPSHOT_COPY_PRIORITY 2 | |
38 | ||
39 | /* | |
8ee2767a | 40 | * Reserve 1MB for each snapshot initially (with minimum of 1 page). |
1da177e4 | 41 | */ |
8ee2767a | 42 | #define SNAPSHOT_PAGES (((1UL << 20) >> PAGE_SHIFT) ? : 1) |
1da177e4 | 43 | |
cd45daff MP |
44 | /* |
45 | * The size of the mempool used to track chunks in use. | |
46 | */ | |
47 | #define MIN_IOS 256 | |
48 | ||
c642f9e0 | 49 | static struct workqueue_struct *ksnapd; |
c4028958 | 50 | static void flush_queued_bios(struct work_struct *work); |
ca3a931f | 51 | |
028867ac AK |
52 | struct dm_snap_pending_exception { |
53 | struct dm_snap_exception e; | |
1da177e4 LT |
54 | |
55 | /* | |
56 | * Origin buffers waiting for this to complete are held | |
57 | * in a bio list | |
58 | */ | |
59 | struct bio_list origin_bios; | |
60 | struct bio_list snapshot_bios; | |
61 | ||
eccf0817 AK |
62 | /* |
63 | * Short-term queue of pending exceptions prior to submission. | |
64 | */ | |
65 | struct list_head list; | |
66 | ||
1da177e4 | 67 | /* |
b4b610f6 | 68 | * The primary pending_exception is the one that holds |
4b832e8d | 69 | * the ref_count and the list of origin_bios for a |
b4b610f6 AK |
70 | * group of pending_exceptions. It is always last to get freed. |
71 | * These fields get set up when writing to the origin. | |
1da177e4 | 72 | */ |
028867ac | 73 | struct dm_snap_pending_exception *primary_pe; |
b4b610f6 AK |
74 | |
75 | /* | |
76 | * Number of pending_exceptions processing this chunk. | |
77 | * When this drops to zero we must complete the origin bios. | |
78 | * If incrementing or decrementing this, hold pe->snap->lock for | |
79 | * the sibling concerned and not pe->primary_pe->snap->lock unless | |
80 | * they are the same. | |
81 | */ | |
4b832e8d | 82 | atomic_t ref_count; |
1da177e4 LT |
83 | |
84 | /* Pointer back to snapshot context */ | |
85 | struct dm_snapshot *snap; | |
86 | ||
87 | /* | |
88 | * 1 indicates the exception has already been sent to | |
89 | * kcopyd. | |
90 | */ | |
91 | int started; | |
92 | }; | |
93 | ||
94 | /* | |
95 | * Hash table mapping origin volumes to lists of snapshots and | |
96 | * a lock to protect it | |
97 | */ | |
e18b890b CL |
98 | static struct kmem_cache *exception_cache; |
99 | static struct kmem_cache *pending_cache; | |
1da177e4 | 100 | |
cd45daff MP |
101 | struct dm_snap_tracked_chunk { |
102 | struct hlist_node node; | |
103 | chunk_t chunk; | |
104 | }; | |
105 | ||
106 | static struct kmem_cache *tracked_chunk_cache; | |
107 | ||
108 | static struct dm_snap_tracked_chunk *track_chunk(struct dm_snapshot *s, | |
109 | chunk_t chunk) | |
110 | { | |
111 | struct dm_snap_tracked_chunk *c = mempool_alloc(s->tracked_chunk_pool, | |
112 | GFP_NOIO); | |
113 | unsigned long flags; | |
114 | ||
115 | c->chunk = chunk; | |
116 | ||
117 | spin_lock_irqsave(&s->tracked_chunk_lock, flags); | |
118 | hlist_add_head(&c->node, | |
119 | &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)]); | |
120 | spin_unlock_irqrestore(&s->tracked_chunk_lock, flags); | |
121 | ||
122 | return c; | |
123 | } | |
124 | ||
125 | static void stop_tracking_chunk(struct dm_snapshot *s, | |
126 | struct dm_snap_tracked_chunk *c) | |
127 | { | |
128 | unsigned long flags; | |
129 | ||
130 | spin_lock_irqsave(&s->tracked_chunk_lock, flags); | |
131 | hlist_del(&c->node); | |
132 | spin_unlock_irqrestore(&s->tracked_chunk_lock, flags); | |
133 | ||
134 | mempool_free(c, s->tracked_chunk_pool); | |
135 | } | |
136 | ||
a8d41b59 MP |
137 | static int __chunk_is_tracked(struct dm_snapshot *s, chunk_t chunk) |
138 | { | |
139 | struct dm_snap_tracked_chunk *c; | |
140 | struct hlist_node *hn; | |
141 | int found = 0; | |
142 | ||
143 | spin_lock_irq(&s->tracked_chunk_lock); | |
144 | ||
145 | hlist_for_each_entry(c, hn, | |
146 | &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)], node) { | |
147 | if (c->chunk == chunk) { | |
148 | found = 1; | |
149 | break; | |
150 | } | |
151 | } | |
152 | ||
153 | spin_unlock_irq(&s->tracked_chunk_lock); | |
154 | ||
155 | return found; | |
156 | } | |
157 | ||
1da177e4 LT |
158 | /* |
159 | * One of these per registered origin, held in the snapshot_origins hash | |
160 | */ | |
161 | struct origin { | |
162 | /* The origin device */ | |
163 | struct block_device *bdev; | |
164 | ||
165 | struct list_head hash_list; | |
166 | ||
167 | /* List of snapshots for this origin */ | |
168 | struct list_head snapshots; | |
169 | }; | |
170 | ||
171 | /* | |
172 | * Size of the hash table for origin volumes. If we make this | |
173 | * the size of the minors list then it should be nearly perfect | |
174 | */ | |
175 | #define ORIGIN_HASH_SIZE 256 | |
176 | #define ORIGIN_MASK 0xFF | |
177 | static struct list_head *_origins; | |
178 | static struct rw_semaphore _origins_lock; | |
179 | ||
180 | static int init_origin_hash(void) | |
181 | { | |
182 | int i; | |
183 | ||
184 | _origins = kmalloc(ORIGIN_HASH_SIZE * sizeof(struct list_head), | |
185 | GFP_KERNEL); | |
186 | if (!_origins) { | |
72d94861 | 187 | DMERR("unable to allocate memory"); |
1da177e4 LT |
188 | return -ENOMEM; |
189 | } | |
190 | ||
191 | for (i = 0; i < ORIGIN_HASH_SIZE; i++) | |
192 | INIT_LIST_HEAD(_origins + i); | |
193 | init_rwsem(&_origins_lock); | |
194 | ||
195 | return 0; | |
196 | } | |
197 | ||
198 | static void exit_origin_hash(void) | |
199 | { | |
200 | kfree(_origins); | |
201 | } | |
202 | ||
028867ac | 203 | static unsigned origin_hash(struct block_device *bdev) |
1da177e4 LT |
204 | { |
205 | return bdev->bd_dev & ORIGIN_MASK; | |
206 | } | |
207 | ||
208 | static struct origin *__lookup_origin(struct block_device *origin) | |
209 | { | |
210 | struct list_head *ol; | |
211 | struct origin *o; | |
212 | ||
213 | ol = &_origins[origin_hash(origin)]; | |
214 | list_for_each_entry (o, ol, hash_list) | |
215 | if (bdev_equal(o->bdev, origin)) | |
216 | return o; | |
217 | ||
218 | return NULL; | |
219 | } | |
220 | ||
221 | static void __insert_origin(struct origin *o) | |
222 | { | |
223 | struct list_head *sl = &_origins[origin_hash(o->bdev)]; | |
224 | list_add_tail(&o->hash_list, sl); | |
225 | } | |
226 | ||
227 | /* | |
228 | * Make a note of the snapshot and its origin so we can look it | |
229 | * up when the origin has a write on it. | |
230 | */ | |
231 | static int register_snapshot(struct dm_snapshot *snap) | |
232 | { | |
60c856c8 | 233 | struct origin *o, *new_o; |
1da177e4 LT |
234 | struct block_device *bdev = snap->origin->bdev; |
235 | ||
60c856c8 MP |
236 | new_o = kmalloc(sizeof(*new_o), GFP_KERNEL); |
237 | if (!new_o) | |
238 | return -ENOMEM; | |
239 | ||
1da177e4 LT |
240 | down_write(&_origins_lock); |
241 | o = __lookup_origin(bdev); | |
242 | ||
60c856c8 MP |
243 | if (o) |
244 | kfree(new_o); | |
245 | else { | |
1da177e4 | 246 | /* New origin */ |
60c856c8 | 247 | o = new_o; |
1da177e4 LT |
248 | |
249 | /* Initialise the struct */ | |
250 | INIT_LIST_HEAD(&o->snapshots); | |
251 | o->bdev = bdev; | |
252 | ||
253 | __insert_origin(o); | |
254 | } | |
255 | ||
256 | list_add_tail(&snap->list, &o->snapshots); | |
257 | ||
258 | up_write(&_origins_lock); | |
259 | return 0; | |
260 | } | |
261 | ||
262 | static void unregister_snapshot(struct dm_snapshot *s) | |
263 | { | |
264 | struct origin *o; | |
265 | ||
266 | down_write(&_origins_lock); | |
267 | o = __lookup_origin(s->origin->bdev); | |
268 | ||
269 | list_del(&s->list); | |
270 | if (list_empty(&o->snapshots)) { | |
271 | list_del(&o->hash_list); | |
272 | kfree(o); | |
273 | } | |
274 | ||
275 | up_write(&_origins_lock); | |
276 | } | |
277 | ||
278 | /* | |
279 | * Implementation of the exception hash tables. | |
d74f81f8 MB |
280 | * The lowest hash_shift bits of the chunk number are ignored, allowing |
281 | * some consecutive chunks to be grouped together. | |
1da177e4 | 282 | */ |
d74f81f8 MB |
283 | static int init_exception_table(struct exception_table *et, uint32_t size, |
284 | unsigned hash_shift) | |
1da177e4 LT |
285 | { |
286 | unsigned int i; | |
287 | ||
d74f81f8 | 288 | et->hash_shift = hash_shift; |
1da177e4 LT |
289 | et->hash_mask = size - 1; |
290 | et->table = dm_vcalloc(size, sizeof(struct list_head)); | |
291 | if (!et->table) | |
292 | return -ENOMEM; | |
293 | ||
294 | for (i = 0; i < size; i++) | |
295 | INIT_LIST_HEAD(et->table + i); | |
296 | ||
297 | return 0; | |
298 | } | |
299 | ||
e18b890b | 300 | static void exit_exception_table(struct exception_table *et, struct kmem_cache *mem) |
1da177e4 LT |
301 | { |
302 | struct list_head *slot; | |
028867ac | 303 | struct dm_snap_exception *ex, *next; |
1da177e4 LT |
304 | int i, size; |
305 | ||
306 | size = et->hash_mask + 1; | |
307 | for (i = 0; i < size; i++) { | |
308 | slot = et->table + i; | |
309 | ||
310 | list_for_each_entry_safe (ex, next, slot, hash_list) | |
311 | kmem_cache_free(mem, ex); | |
312 | } | |
313 | ||
314 | vfree(et->table); | |
315 | } | |
316 | ||
028867ac | 317 | static uint32_t exception_hash(struct exception_table *et, chunk_t chunk) |
1da177e4 | 318 | { |
d74f81f8 | 319 | return (chunk >> et->hash_shift) & et->hash_mask; |
1da177e4 LT |
320 | } |
321 | ||
028867ac AK |
322 | static void insert_exception(struct exception_table *eh, |
323 | struct dm_snap_exception *e) | |
1da177e4 LT |
324 | { |
325 | struct list_head *l = &eh->table[exception_hash(eh, e->old_chunk)]; | |
326 | list_add(&e->hash_list, l); | |
327 | } | |
328 | ||
028867ac | 329 | static void remove_exception(struct dm_snap_exception *e) |
1da177e4 LT |
330 | { |
331 | list_del(&e->hash_list); | |
332 | } | |
333 | ||
334 | /* | |
335 | * Return the exception data for a sector, or NULL if not | |
336 | * remapped. | |
337 | */ | |
028867ac AK |
338 | static struct dm_snap_exception *lookup_exception(struct exception_table *et, |
339 | chunk_t chunk) | |
1da177e4 LT |
340 | { |
341 | struct list_head *slot; | |
028867ac | 342 | struct dm_snap_exception *e; |
1da177e4 LT |
343 | |
344 | slot = &et->table[exception_hash(et, chunk)]; | |
345 | list_for_each_entry (e, slot, hash_list) | |
d74f81f8 MB |
346 | if (chunk >= e->old_chunk && |
347 | chunk <= e->old_chunk + dm_consecutive_chunk_count(e)) | |
1da177e4 LT |
348 | return e; |
349 | ||
350 | return NULL; | |
351 | } | |
352 | ||
028867ac | 353 | static struct dm_snap_exception *alloc_exception(void) |
1da177e4 | 354 | { |
028867ac | 355 | struct dm_snap_exception *e; |
1da177e4 LT |
356 | |
357 | e = kmem_cache_alloc(exception_cache, GFP_NOIO); | |
358 | if (!e) | |
359 | e = kmem_cache_alloc(exception_cache, GFP_ATOMIC); | |
360 | ||
361 | return e; | |
362 | } | |
363 | ||
028867ac | 364 | static void free_exception(struct dm_snap_exception *e) |
1da177e4 LT |
365 | { |
366 | kmem_cache_free(exception_cache, e); | |
367 | } | |
368 | ||
92e86812 | 369 | static struct dm_snap_pending_exception *alloc_pending_exception(struct dm_snapshot *s) |
1da177e4 | 370 | { |
92e86812 MP |
371 | struct dm_snap_pending_exception *pe = mempool_alloc(s->pending_pool, |
372 | GFP_NOIO); | |
373 | ||
879129d2 | 374 | atomic_inc(&s->pending_exceptions_count); |
92e86812 MP |
375 | pe->snap = s; |
376 | ||
377 | return pe; | |
1da177e4 LT |
378 | } |
379 | ||
028867ac | 380 | static void free_pending_exception(struct dm_snap_pending_exception *pe) |
1da177e4 | 381 | { |
879129d2 MP |
382 | struct dm_snapshot *s = pe->snap; |
383 | ||
384 | mempool_free(pe, s->pending_pool); | |
385 | smp_mb__before_atomic_dec(); | |
386 | atomic_dec(&s->pending_exceptions_count); | |
1da177e4 LT |
387 | } |
388 | ||
d74f81f8 MB |
389 | static void insert_completed_exception(struct dm_snapshot *s, |
390 | struct dm_snap_exception *new_e) | |
391 | { | |
392 | struct exception_table *eh = &s->complete; | |
393 | struct list_head *l; | |
394 | struct dm_snap_exception *e = NULL; | |
395 | ||
396 | l = &eh->table[exception_hash(eh, new_e->old_chunk)]; | |
397 | ||
398 | /* Add immediately if this table doesn't support consecutive chunks */ | |
399 | if (!eh->hash_shift) | |
400 | goto out; | |
401 | ||
402 | /* List is ordered by old_chunk */ | |
403 | list_for_each_entry_reverse(e, l, hash_list) { | |
404 | /* Insert after an existing chunk? */ | |
405 | if (new_e->old_chunk == (e->old_chunk + | |
406 | dm_consecutive_chunk_count(e) + 1) && | |
407 | new_e->new_chunk == (dm_chunk_number(e->new_chunk) + | |
408 | dm_consecutive_chunk_count(e) + 1)) { | |
409 | dm_consecutive_chunk_count_inc(e); | |
410 | free_exception(new_e); | |
411 | return; | |
412 | } | |
413 | ||
414 | /* Insert before an existing chunk? */ | |
415 | if (new_e->old_chunk == (e->old_chunk - 1) && | |
416 | new_e->new_chunk == (dm_chunk_number(e->new_chunk) - 1)) { | |
417 | dm_consecutive_chunk_count_inc(e); | |
418 | e->old_chunk--; | |
419 | e->new_chunk--; | |
420 | free_exception(new_e); | |
421 | return; | |
422 | } | |
423 | ||
424 | if (new_e->old_chunk > e->old_chunk) | |
425 | break; | |
426 | } | |
427 | ||
428 | out: | |
429 | list_add(&new_e->hash_list, e ? &e->hash_list : l); | |
430 | } | |
431 | ||
1da177e4 LT |
432 | int dm_add_exception(struct dm_snapshot *s, chunk_t old, chunk_t new) |
433 | { | |
028867ac | 434 | struct dm_snap_exception *e; |
1da177e4 LT |
435 | |
436 | e = alloc_exception(); | |
437 | if (!e) | |
438 | return -ENOMEM; | |
439 | ||
440 | e->old_chunk = old; | |
d74f81f8 MB |
441 | |
442 | /* Consecutive_count is implicitly initialised to zero */ | |
1da177e4 | 443 | e->new_chunk = new; |
d74f81f8 MB |
444 | |
445 | insert_completed_exception(s, e); | |
446 | ||
1da177e4 LT |
447 | return 0; |
448 | } | |
449 | ||
450 | /* | |
451 | * Hard coded magic. | |
452 | */ | |
453 | static int calc_max_buckets(void) | |
454 | { | |
455 | /* use a fixed size of 2MB */ | |
456 | unsigned long mem = 2 * 1024 * 1024; | |
457 | mem /= sizeof(struct list_head); | |
458 | ||
459 | return mem; | |
460 | } | |
461 | ||
1da177e4 LT |
462 | /* |
463 | * Allocate room for a suitable hash table. | |
464 | */ | |
465 | static int init_hash_tables(struct dm_snapshot *s) | |
466 | { | |
467 | sector_t hash_size, cow_dev_size, origin_dev_size, max_buckets; | |
468 | ||
469 | /* | |
470 | * Calculate based on the size of the original volume or | |
471 | * the COW volume... | |
472 | */ | |
473 | cow_dev_size = get_dev_size(s->cow->bdev); | |
474 | origin_dev_size = get_dev_size(s->origin->bdev); | |
475 | max_buckets = calc_max_buckets(); | |
476 | ||
477 | hash_size = min(origin_dev_size, cow_dev_size) >> s->chunk_shift; | |
478 | hash_size = min(hash_size, max_buckets); | |
479 | ||
8defd830 | 480 | hash_size = rounddown_pow_of_two(hash_size); |
d74f81f8 MB |
481 | if (init_exception_table(&s->complete, hash_size, |
482 | DM_CHUNK_CONSECUTIVE_BITS)) | |
1da177e4 LT |
483 | return -ENOMEM; |
484 | ||
485 | /* | |
486 | * Allocate hash table for in-flight exceptions | |
487 | * Make this smaller than the real hash table | |
488 | */ | |
489 | hash_size >>= 3; | |
490 | if (hash_size < 64) | |
491 | hash_size = 64; | |
492 | ||
d74f81f8 | 493 | if (init_exception_table(&s->pending, hash_size, 0)) { |
1da177e4 LT |
494 | exit_exception_table(&s->complete, exception_cache); |
495 | return -ENOMEM; | |
496 | } | |
497 | ||
498 | return 0; | |
499 | } | |
500 | ||
501 | /* | |
502 | * Round a number up to the nearest 'size' boundary. size must | |
503 | * be a power of 2. | |
504 | */ | |
028867ac | 505 | static ulong round_up(ulong n, ulong size) |
1da177e4 LT |
506 | { |
507 | size--; | |
508 | return (n + size) & ~size; | |
509 | } | |
510 | ||
4c7e3bf4 MM |
511 | static int set_chunk_size(struct dm_snapshot *s, const char *chunk_size_arg, |
512 | char **error) | |
513 | { | |
514 | unsigned long chunk_size; | |
515 | char *value; | |
516 | ||
517 | chunk_size = simple_strtoul(chunk_size_arg, &value, 10); | |
518 | if (*chunk_size_arg == '\0' || *value != '\0') { | |
519 | *error = "Invalid chunk size"; | |
520 | return -EINVAL; | |
521 | } | |
522 | ||
523 | if (!chunk_size) { | |
524 | s->chunk_size = s->chunk_mask = s->chunk_shift = 0; | |
525 | return 0; | |
526 | } | |
527 | ||
528 | /* | |
529 | * Chunk size must be multiple of page size. Silently | |
530 | * round up if it's not. | |
531 | */ | |
532 | chunk_size = round_up(chunk_size, PAGE_SIZE >> 9); | |
533 | ||
534 | /* Check chunk_size is a power of 2 */ | |
6f3c3f0a | 535 | if (!is_power_of_2(chunk_size)) { |
4c7e3bf4 MM |
536 | *error = "Chunk size is not a power of 2"; |
537 | return -EINVAL; | |
538 | } | |
539 | ||
540 | /* Validate the chunk size against the device block size */ | |
541 | if (chunk_size % (bdev_hardsect_size(s->cow->bdev) >> 9)) { | |
542 | *error = "Chunk size is not a multiple of device blocksize"; | |
543 | return -EINVAL; | |
544 | } | |
545 | ||
546 | s->chunk_size = chunk_size; | |
547 | s->chunk_mask = chunk_size - 1; | |
548 | s->chunk_shift = ffs(chunk_size) - 1; | |
549 | ||
550 | return 0; | |
551 | } | |
552 | ||
1da177e4 LT |
553 | /* |
554 | * Construct a snapshot mapping: <origin_dev> <COW-dev> <p/n> <chunk-size> | |
555 | */ | |
556 | static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv) | |
557 | { | |
558 | struct dm_snapshot *s; | |
cd45daff | 559 | int i; |
1da177e4 LT |
560 | int r = -EINVAL; |
561 | char persistent; | |
562 | char *origin_path; | |
563 | char *cow_path; | |
1da177e4 | 564 | |
4c7e3bf4 | 565 | if (argc != 4) { |
72d94861 | 566 | ti->error = "requires exactly 4 arguments"; |
1da177e4 LT |
567 | r = -EINVAL; |
568 | goto bad1; | |
569 | } | |
570 | ||
571 | origin_path = argv[0]; | |
572 | cow_path = argv[1]; | |
573 | persistent = toupper(*argv[2]); | |
574 | ||
575 | if (persistent != 'P' && persistent != 'N') { | |
576 | ti->error = "Persistent flag is not P or N"; | |
577 | r = -EINVAL; | |
578 | goto bad1; | |
579 | } | |
580 | ||
1da177e4 LT |
581 | s = kmalloc(sizeof(*s), GFP_KERNEL); |
582 | if (s == NULL) { | |
583 | ti->error = "Cannot allocate snapshot context private " | |
584 | "structure"; | |
585 | r = -ENOMEM; | |
586 | goto bad1; | |
587 | } | |
588 | ||
589 | r = dm_get_device(ti, origin_path, 0, ti->len, FMODE_READ, &s->origin); | |
590 | if (r) { | |
591 | ti->error = "Cannot get origin device"; | |
592 | goto bad2; | |
593 | } | |
594 | ||
595 | r = dm_get_device(ti, cow_path, 0, 0, | |
596 | FMODE_READ | FMODE_WRITE, &s->cow); | |
597 | if (r) { | |
598 | dm_put_device(ti, s->origin); | |
599 | ti->error = "Cannot get COW device"; | |
600 | goto bad2; | |
601 | } | |
602 | ||
4c7e3bf4 MM |
603 | r = set_chunk_size(s, argv[3], &ti->error); |
604 | if (r) | |
1da177e4 | 605 | goto bad3; |
1da177e4 | 606 | |
1da177e4 | 607 | s->type = persistent; |
1da177e4 LT |
608 | |
609 | s->valid = 1; | |
aa14edeb | 610 | s->active = 0; |
879129d2 | 611 | atomic_set(&s->pending_exceptions_count, 0); |
1da177e4 | 612 | init_rwsem(&s->lock); |
ca3a931f | 613 | spin_lock_init(&s->pe_lock); |
72727bad | 614 | s->ti = ti; |
1da177e4 LT |
615 | |
616 | /* Allocate hash table for COW data */ | |
617 | if (init_hash_tables(s)) { | |
618 | ti->error = "Unable to allocate hash table space"; | |
619 | r = -ENOMEM; | |
620 | goto bad3; | |
621 | } | |
622 | ||
1da177e4 LT |
623 | s->store.snap = s; |
624 | ||
625 | if (persistent == 'P') | |
4c7e3bf4 | 626 | r = dm_create_persistent(&s->store); |
1da177e4 | 627 | else |
4c7e3bf4 | 628 | r = dm_create_transient(&s->store); |
1da177e4 LT |
629 | |
630 | if (r) { | |
631 | ti->error = "Couldn't create exception store"; | |
632 | r = -EINVAL; | |
633 | goto bad4; | |
634 | } | |
635 | ||
eb69aca5 | 636 | r = dm_kcopyd_client_create(SNAPSHOT_PAGES, &s->kcopyd_client); |
1da177e4 LT |
637 | if (r) { |
638 | ti->error = "Could not create kcopyd client"; | |
639 | goto bad5; | |
640 | } | |
641 | ||
92e86812 MP |
642 | s->pending_pool = mempool_create_slab_pool(MIN_IOS, pending_cache); |
643 | if (!s->pending_pool) { | |
644 | ti->error = "Could not allocate mempool for pending exceptions"; | |
645 | goto bad6; | |
646 | } | |
647 | ||
cd45daff MP |
648 | s->tracked_chunk_pool = mempool_create_slab_pool(MIN_IOS, |
649 | tracked_chunk_cache); | |
650 | if (!s->tracked_chunk_pool) { | |
651 | ti->error = "Could not allocate tracked_chunk mempool for " | |
652 | "tracking reads"; | |
92e86812 | 653 | goto bad_tracked_chunk_pool; |
cd45daff MP |
654 | } |
655 | ||
656 | for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++) | |
657 | INIT_HLIST_HEAD(&s->tracked_chunk_hash[i]); | |
658 | ||
659 | spin_lock_init(&s->tracked_chunk_lock); | |
660 | ||
aa14edeb | 661 | /* Metadata must only be loaded into one table at once */ |
f9cea4f7 | 662 | r = s->store.read_metadata(&s->store); |
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: | |
697 | s->store.destroy(&s->store); | |
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 | ||
722 | s->store.destroy(&s->store); | |
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 | ||
76df1c65 AK |
817 | if (s->store.drop_snapshot) |
818 | s->store.drop_snapshot(&s->store); | |
819 | ||
820 | s->valid = 0; | |
821 | ||
72727bad | 822 | dm_table_event(s->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 */ | |
940 | s->store.commit_exception(&s->store, &pe->e, commit_callback, | |
941 | pe); | |
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 | ||
969 | /* | |
970 | * Looks to see if this snapshot already has a pending exception | |
971 | * for this chunk, otherwise it allocates a new one and inserts | |
972 | * it into the pending table. | |
973 | * | |
974 | * NOTE: a write lock must be held on snap->lock before calling | |
975 | * this. | |
976 | */ | |
028867ac | 977 | static struct dm_snap_pending_exception * |
1da177e4 LT |
978 | __find_pending_exception(struct dm_snapshot *s, struct bio *bio) |
979 | { | |
028867ac AK |
980 | struct dm_snap_exception *e; |
981 | struct dm_snap_pending_exception *pe; | |
1da177e4 LT |
982 | chunk_t chunk = sector_to_chunk(s, bio->bi_sector); |
983 | ||
984 | /* | |
985 | * Is there a pending exception for this already ? | |
986 | */ | |
987 | e = lookup_exception(&s->pending, chunk); | |
988 | if (e) { | |
989 | /* cast the exception to a pending exception */ | |
028867ac | 990 | pe = container_of(e, struct dm_snap_pending_exception, e); |
76df1c65 AK |
991 | goto out; |
992 | } | |
1da177e4 | 993 | |
76df1c65 AK |
994 | /* |
995 | * Create a new pending exception, we don't want | |
996 | * to hold the lock while we do this. | |
997 | */ | |
998 | up_write(&s->lock); | |
92e86812 | 999 | pe = alloc_pending_exception(s); |
76df1c65 | 1000 | down_write(&s->lock); |
1da177e4 | 1001 | |
76df1c65 AK |
1002 | if (!s->valid) { |
1003 | free_pending_exception(pe); | |
1004 | return NULL; | |
1005 | } | |
1da177e4 | 1006 | |
76df1c65 AK |
1007 | e = lookup_exception(&s->pending, chunk); |
1008 | if (e) { | |
1009 | free_pending_exception(pe); | |
028867ac | 1010 | pe = container_of(e, struct dm_snap_pending_exception, e); |
76df1c65 | 1011 | goto out; |
1da177e4 LT |
1012 | } |
1013 | ||
76df1c65 AK |
1014 | pe->e.old_chunk = chunk; |
1015 | bio_list_init(&pe->origin_bios); | |
1016 | bio_list_init(&pe->snapshot_bios); | |
1017 | pe->primary_pe = NULL; | |
4b832e8d | 1018 | atomic_set(&pe->ref_count, 0); |
76df1c65 AK |
1019 | pe->started = 0; |
1020 | ||
1021 | if (s->store.prepare_exception(&s->store, &pe->e)) { | |
1022 | free_pending_exception(pe); | |
1023 | return NULL; | |
1024 | } | |
1025 | ||
4b832e8d | 1026 | get_pending_exception(pe); |
76df1c65 AK |
1027 | insert_exception(&s->pending, &pe->e); |
1028 | ||
1029 | out: | |
1da177e4 LT |
1030 | return pe; |
1031 | } | |
1032 | ||
028867ac | 1033 | static void remap_exception(struct dm_snapshot *s, struct dm_snap_exception *e, |
d74f81f8 | 1034 | struct bio *bio, chunk_t chunk) |
1da177e4 LT |
1035 | { |
1036 | bio->bi_bdev = s->cow->bdev; | |
d74f81f8 MB |
1037 | bio->bi_sector = chunk_to_sector(s, dm_chunk_number(e->new_chunk) + |
1038 | (chunk - e->old_chunk)) + | |
1039 | (bio->bi_sector & s->chunk_mask); | |
1da177e4 LT |
1040 | } |
1041 | ||
1042 | static int snapshot_map(struct dm_target *ti, struct bio *bio, | |
1043 | union map_info *map_context) | |
1044 | { | |
028867ac AK |
1045 | struct dm_snap_exception *e; |
1046 | struct dm_snapshot *s = ti->private; | |
d2a7ad29 | 1047 | int r = DM_MAPIO_REMAPPED; |
1da177e4 | 1048 | chunk_t chunk; |
028867ac | 1049 | struct dm_snap_pending_exception *pe = NULL; |
1da177e4 LT |
1050 | |
1051 | chunk = sector_to_chunk(s, bio->bi_sector); | |
1052 | ||
1053 | /* Full snapshots are not usable */ | |
76df1c65 | 1054 | /* To get here the table must be live so s->active is always set. */ |
1da177e4 | 1055 | if (!s->valid) |
f6a80ea8 | 1056 | return -EIO; |
1da177e4 | 1057 | |
ba40a2aa AK |
1058 | /* FIXME: should only take write lock if we need |
1059 | * to copy an exception */ | |
1060 | down_write(&s->lock); | |
1061 | ||
1062 | if (!s->valid) { | |
1063 | r = -EIO; | |
1064 | goto out_unlock; | |
1065 | } | |
1066 | ||
1067 | /* If the block is already remapped - use that, else remap it */ | |
1068 | e = lookup_exception(&s->complete, chunk); | |
1069 | if (e) { | |
d74f81f8 | 1070 | remap_exception(s, e, bio, chunk); |
ba40a2aa AK |
1071 | goto out_unlock; |
1072 | } | |
1073 | ||
1da177e4 LT |
1074 | /* |
1075 | * Write to snapshot - higher level takes care of RW/RO | |
1076 | * flags so we should only get this if we are | |
1077 | * writeable. | |
1078 | */ | |
1079 | if (bio_rw(bio) == WRITE) { | |
76df1c65 AK |
1080 | pe = __find_pending_exception(s, bio); |
1081 | if (!pe) { | |
695368ac | 1082 | __invalidate_snapshot(s, -ENOMEM); |
76df1c65 AK |
1083 | r = -EIO; |
1084 | goto out_unlock; | |
1da177e4 LT |
1085 | } |
1086 | ||
d74f81f8 | 1087 | remap_exception(s, &pe->e, bio, chunk); |
76df1c65 AK |
1088 | bio_list_add(&pe->snapshot_bios, bio); |
1089 | ||
d2a7ad29 | 1090 | r = DM_MAPIO_SUBMITTED; |
ba40a2aa | 1091 | |
76df1c65 AK |
1092 | if (!pe->started) { |
1093 | /* this is protected by snap->lock */ | |
1094 | pe->started = 1; | |
ba40a2aa | 1095 | up_write(&s->lock); |
76df1c65 | 1096 | start_copy(pe); |
ba40a2aa AK |
1097 | goto out; |
1098 | } | |
cd45daff | 1099 | } else { |
ba40a2aa | 1100 | bio->bi_bdev = s->origin->bdev; |
cd45daff MP |
1101 | map_context->ptr = track_chunk(s, chunk); |
1102 | } | |
1da177e4 | 1103 | |
ba40a2aa AK |
1104 | out_unlock: |
1105 | up_write(&s->lock); | |
1106 | out: | |
1da177e4 LT |
1107 | return r; |
1108 | } | |
1109 | ||
cd45daff MP |
1110 | static int snapshot_end_io(struct dm_target *ti, struct bio *bio, |
1111 | int error, union map_info *map_context) | |
1112 | { | |
1113 | struct dm_snapshot *s = ti->private; | |
1114 | struct dm_snap_tracked_chunk *c = map_context->ptr; | |
1115 | ||
1116 | if (c) | |
1117 | stop_tracking_chunk(s, c); | |
1118 | ||
1119 | return 0; | |
1120 | } | |
1121 | ||
1da177e4 LT |
1122 | static void snapshot_resume(struct dm_target *ti) |
1123 | { | |
028867ac | 1124 | struct dm_snapshot *s = ti->private; |
1da177e4 | 1125 | |
aa14edeb AK |
1126 | down_write(&s->lock); |
1127 | s->active = 1; | |
1128 | up_write(&s->lock); | |
1da177e4 LT |
1129 | } |
1130 | ||
1131 | static int snapshot_status(struct dm_target *ti, status_type_t type, | |
1132 | char *result, unsigned int maxlen) | |
1133 | { | |
028867ac | 1134 | struct dm_snapshot *snap = ti->private; |
1da177e4 LT |
1135 | |
1136 | switch (type) { | |
1137 | case STATUSTYPE_INFO: | |
1138 | if (!snap->valid) | |
1139 | snprintf(result, maxlen, "Invalid"); | |
1140 | else { | |
1141 | if (snap->store.fraction_full) { | |
1142 | sector_t numerator, denominator; | |
1143 | snap->store.fraction_full(&snap->store, | |
1144 | &numerator, | |
1145 | &denominator); | |
4ee218cd AM |
1146 | snprintf(result, maxlen, "%llu/%llu", |
1147 | (unsigned long long)numerator, | |
1148 | (unsigned long long)denominator); | |
1da177e4 LT |
1149 | } |
1150 | else | |
1151 | snprintf(result, maxlen, "Unknown"); | |
1152 | } | |
1153 | break; | |
1154 | ||
1155 | case STATUSTYPE_TABLE: | |
1156 | /* | |
1157 | * kdevname returns a static pointer so we need | |
1158 | * to make private copies if the output is to | |
1159 | * make sense. | |
1160 | */ | |
4ee218cd | 1161 | snprintf(result, maxlen, "%s %s %c %llu", |
1da177e4 | 1162 | snap->origin->name, snap->cow->name, |
4ee218cd AM |
1163 | snap->type, |
1164 | (unsigned long long)snap->chunk_size); | |
1da177e4 LT |
1165 | break; |
1166 | } | |
1167 | ||
1168 | return 0; | |
1169 | } | |
1170 | ||
1171 | /*----------------------------------------------------------------- | |
1172 | * Origin methods | |
1173 | *---------------------------------------------------------------*/ | |
1da177e4 LT |
1174 | static int __origin_write(struct list_head *snapshots, struct bio *bio) |
1175 | { | |
d2a7ad29 | 1176 | int r = DM_MAPIO_REMAPPED, first = 0; |
1da177e4 | 1177 | struct dm_snapshot *snap; |
028867ac AK |
1178 | struct dm_snap_exception *e; |
1179 | struct dm_snap_pending_exception *pe, *next_pe, *primary_pe = NULL; | |
1da177e4 | 1180 | chunk_t chunk; |
eccf0817 | 1181 | LIST_HEAD(pe_queue); |
1da177e4 LT |
1182 | |
1183 | /* Do all the snapshots on this origin */ | |
1184 | list_for_each_entry (snap, snapshots, list) { | |
1185 | ||
76df1c65 AK |
1186 | down_write(&snap->lock); |
1187 | ||
aa14edeb AK |
1188 | /* Only deal with valid and active snapshots */ |
1189 | if (!snap->valid || !snap->active) | |
76df1c65 | 1190 | goto next_snapshot; |
1da177e4 | 1191 | |
d5e404c1 | 1192 | /* Nothing to do if writing beyond end of snapshot */ |
72727bad | 1193 | if (bio->bi_sector >= dm_table_get_size(snap->ti->table)) |
76df1c65 | 1194 | goto next_snapshot; |
1da177e4 LT |
1195 | |
1196 | /* | |
1197 | * Remember, different snapshots can have | |
1198 | * different chunk sizes. | |
1199 | */ | |
1200 | chunk = sector_to_chunk(snap, bio->bi_sector); | |
1201 | ||
1202 | /* | |
1203 | * Check exception table to see if block | |
1204 | * is already remapped in this snapshot | |
1205 | * and trigger an exception if not. | |
b4b610f6 | 1206 | * |
4b832e8d | 1207 | * ref_count is initialised to 1 so pending_complete() |
b4b610f6 | 1208 | * won't destroy the primary_pe while we're inside this loop. |
1da177e4 LT |
1209 | */ |
1210 | e = lookup_exception(&snap->complete, chunk); | |
76df1c65 AK |
1211 | if (e) |
1212 | goto next_snapshot; | |
1213 | ||
1214 | pe = __find_pending_exception(snap, bio); | |
1215 | if (!pe) { | |
695368ac | 1216 | __invalidate_snapshot(snap, -ENOMEM); |
76df1c65 AK |
1217 | goto next_snapshot; |
1218 | } | |
1219 | ||
1220 | if (!primary_pe) { | |
1221 | /* | |
1222 | * Either every pe here has same | |
1223 | * primary_pe or none has one yet. | |
1224 | */ | |
1225 | if (pe->primary_pe) | |
1226 | primary_pe = pe->primary_pe; | |
1227 | else { | |
1228 | primary_pe = pe; | |
1229 | first = 1; | |
1da177e4 | 1230 | } |
76df1c65 AK |
1231 | |
1232 | bio_list_add(&primary_pe->origin_bios, bio); | |
1233 | ||
d2a7ad29 | 1234 | r = DM_MAPIO_SUBMITTED; |
76df1c65 AK |
1235 | } |
1236 | ||
1237 | if (!pe->primary_pe) { | |
76df1c65 | 1238 | pe->primary_pe = primary_pe; |
4b832e8d | 1239 | get_pending_exception(primary_pe); |
76df1c65 AK |
1240 | } |
1241 | ||
1242 | if (!pe->started) { | |
1243 | pe->started = 1; | |
1244 | list_add_tail(&pe->list, &pe_queue); | |
1da177e4 LT |
1245 | } |
1246 | ||
76df1c65 | 1247 | next_snapshot: |
1da177e4 LT |
1248 | up_write(&snap->lock); |
1249 | } | |
1250 | ||
b4b610f6 | 1251 | if (!primary_pe) |
4b832e8d | 1252 | return r; |
b4b610f6 AK |
1253 | |
1254 | /* | |
1255 | * If this is the first time we're processing this chunk and | |
4b832e8d | 1256 | * ref_count is now 1 it means all the pending exceptions |
b4b610f6 AK |
1257 | * got completed while we were in the loop above, so it falls to |
1258 | * us here to remove the primary_pe and submit any origin_bios. | |
1259 | */ | |
1260 | ||
4b832e8d | 1261 | if (first && atomic_dec_and_test(&primary_pe->ref_count)) { |
b4b610f6 AK |
1262 | flush_bios(bio_list_get(&primary_pe->origin_bios)); |
1263 | free_pending_exception(primary_pe); | |
1264 | /* If we got here, pe_queue is necessarily empty. */ | |
4b832e8d | 1265 | return r; |
b4b610f6 AK |
1266 | } |
1267 | ||
1da177e4 LT |
1268 | /* |
1269 | * Now that we have a complete pe list we can start the copying. | |
1270 | */ | |
eccf0817 AK |
1271 | list_for_each_entry_safe(pe, next_pe, &pe_queue, list) |
1272 | start_copy(pe); | |
1da177e4 LT |
1273 | |
1274 | return r; | |
1275 | } | |
1276 | ||
1277 | /* | |
1278 | * Called on a write from the origin driver. | |
1279 | */ | |
1280 | static int do_origin(struct dm_dev *origin, struct bio *bio) | |
1281 | { | |
1282 | struct origin *o; | |
d2a7ad29 | 1283 | int r = DM_MAPIO_REMAPPED; |
1da177e4 LT |
1284 | |
1285 | down_read(&_origins_lock); | |
1286 | o = __lookup_origin(origin->bdev); | |
1287 | if (o) | |
1288 | r = __origin_write(&o->snapshots, bio); | |
1289 | up_read(&_origins_lock); | |
1290 | ||
1291 | return r; | |
1292 | } | |
1293 | ||
1294 | /* | |
1295 | * Origin: maps a linear range of a device, with hooks for snapshotting. | |
1296 | */ | |
1297 | ||
1298 | /* | |
1299 | * Construct an origin mapping: <dev_path> | |
1300 | * The context for an origin is merely a 'struct dm_dev *' | |
1301 | * pointing to the real device. | |
1302 | */ | |
1303 | static int origin_ctr(struct dm_target *ti, unsigned int argc, char **argv) | |
1304 | { | |
1305 | int r; | |
1306 | struct dm_dev *dev; | |
1307 | ||
1308 | if (argc != 1) { | |
72d94861 | 1309 | ti->error = "origin: incorrect number of arguments"; |
1da177e4 LT |
1310 | return -EINVAL; |
1311 | } | |
1312 | ||
1313 | r = dm_get_device(ti, argv[0], 0, ti->len, | |
1314 | dm_table_get_mode(ti->table), &dev); | |
1315 | if (r) { | |
1316 | ti->error = "Cannot get target device"; | |
1317 | return r; | |
1318 | } | |
1319 | ||
1320 | ti->private = dev; | |
1321 | return 0; | |
1322 | } | |
1323 | ||
1324 | static void origin_dtr(struct dm_target *ti) | |
1325 | { | |
028867ac | 1326 | struct dm_dev *dev = ti->private; |
1da177e4 LT |
1327 | dm_put_device(ti, dev); |
1328 | } | |
1329 | ||
1330 | static int origin_map(struct dm_target *ti, struct bio *bio, | |
1331 | union map_info *map_context) | |
1332 | { | |
028867ac | 1333 | struct dm_dev *dev = ti->private; |
1da177e4 LT |
1334 | bio->bi_bdev = dev->bdev; |
1335 | ||
1336 | /* Only tell snapshots if this is a write */ | |
d2a7ad29 | 1337 | return (bio_rw(bio) == WRITE) ? do_origin(dev, bio) : DM_MAPIO_REMAPPED; |
1da177e4 LT |
1338 | } |
1339 | ||
1340 | #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r)) | |
1341 | ||
1342 | /* | |
1343 | * Set the target "split_io" field to the minimum of all the snapshots' | |
1344 | * chunk sizes. | |
1345 | */ | |
1346 | static void origin_resume(struct dm_target *ti) | |
1347 | { | |
028867ac | 1348 | struct dm_dev *dev = ti->private; |
1da177e4 LT |
1349 | struct dm_snapshot *snap; |
1350 | struct origin *o; | |
1351 | chunk_t chunk_size = 0; | |
1352 | ||
1353 | down_read(&_origins_lock); | |
1354 | o = __lookup_origin(dev->bdev); | |
1355 | if (o) | |
1356 | list_for_each_entry (snap, &o->snapshots, list) | |
1357 | chunk_size = min_not_zero(chunk_size, snap->chunk_size); | |
1358 | up_read(&_origins_lock); | |
1359 | ||
1360 | ti->split_io = chunk_size; | |
1361 | } | |
1362 | ||
1363 | static int origin_status(struct dm_target *ti, status_type_t type, char *result, | |
1364 | unsigned int maxlen) | |
1365 | { | |
028867ac | 1366 | struct dm_dev *dev = ti->private; |
1da177e4 LT |
1367 | |
1368 | switch (type) { | |
1369 | case STATUSTYPE_INFO: | |
1370 | result[0] = '\0'; | |
1371 | break; | |
1372 | ||
1373 | case STATUSTYPE_TABLE: | |
1374 | snprintf(result, maxlen, "%s", dev->name); | |
1375 | break; | |
1376 | } | |
1377 | ||
1378 | return 0; | |
1379 | } | |
1380 | ||
1381 | static struct target_type origin_target = { | |
1382 | .name = "snapshot-origin", | |
d74f81f8 | 1383 | .version = {1, 6, 0}, |
1da177e4 LT |
1384 | .module = THIS_MODULE, |
1385 | .ctr = origin_ctr, | |
1386 | .dtr = origin_dtr, | |
1387 | .map = origin_map, | |
1388 | .resume = origin_resume, | |
1389 | .status = origin_status, | |
1390 | }; | |
1391 | ||
1392 | static struct target_type snapshot_target = { | |
1393 | .name = "snapshot", | |
d74f81f8 | 1394 | .version = {1, 6, 0}, |
1da177e4 LT |
1395 | .module = THIS_MODULE, |
1396 | .ctr = snapshot_ctr, | |
1397 | .dtr = snapshot_dtr, | |
1398 | .map = snapshot_map, | |
cd45daff | 1399 | .end_io = snapshot_end_io, |
1da177e4 LT |
1400 | .resume = snapshot_resume, |
1401 | .status = snapshot_status, | |
1402 | }; | |
1403 | ||
1404 | static int __init dm_snapshot_init(void) | |
1405 | { | |
1406 | int r; | |
1407 | ||
1408 | r = dm_register_target(&snapshot_target); | |
1409 | if (r) { | |
1410 | DMERR("snapshot target register failed %d", r); | |
1411 | return r; | |
1412 | } | |
1413 | ||
1414 | r = dm_register_target(&origin_target); | |
1415 | if (r < 0) { | |
72d94861 | 1416 | DMERR("Origin target register failed %d", r); |
1da177e4 LT |
1417 | goto bad1; |
1418 | } | |
1419 | ||
1420 | r = init_origin_hash(); | |
1421 | if (r) { | |
1422 | DMERR("init_origin_hash failed."); | |
1423 | goto bad2; | |
1424 | } | |
1425 | ||
028867ac | 1426 | exception_cache = KMEM_CACHE(dm_snap_exception, 0); |
1da177e4 LT |
1427 | if (!exception_cache) { |
1428 | DMERR("Couldn't create exception cache."); | |
1429 | r = -ENOMEM; | |
1430 | goto bad3; | |
1431 | } | |
1432 | ||
028867ac | 1433 | pending_cache = KMEM_CACHE(dm_snap_pending_exception, 0); |
1da177e4 LT |
1434 | if (!pending_cache) { |
1435 | DMERR("Couldn't create pending cache."); | |
1436 | r = -ENOMEM; | |
1437 | goto bad4; | |
1438 | } | |
1439 | ||
cd45daff MP |
1440 | tracked_chunk_cache = KMEM_CACHE(dm_snap_tracked_chunk, 0); |
1441 | if (!tracked_chunk_cache) { | |
1442 | DMERR("Couldn't create cache to track chunks in use."); | |
1443 | r = -ENOMEM; | |
1444 | goto bad5; | |
1445 | } | |
1446 | ||
ca3a931f AK |
1447 | ksnapd = create_singlethread_workqueue("ksnapd"); |
1448 | if (!ksnapd) { | |
1449 | DMERR("Failed to create ksnapd workqueue."); | |
1450 | r = -ENOMEM; | |
92e86812 | 1451 | goto bad_pending_pool; |
ca3a931f AK |
1452 | } |
1453 | ||
1da177e4 LT |
1454 | return 0; |
1455 | ||
cd45daff MP |
1456 | bad_pending_pool: |
1457 | kmem_cache_destroy(tracked_chunk_cache); | |
1da177e4 LT |
1458 | bad5: |
1459 | kmem_cache_destroy(pending_cache); | |
1460 | bad4: | |
1461 | kmem_cache_destroy(exception_cache); | |
1462 | bad3: | |
1463 | exit_origin_hash(); | |
1464 | bad2: | |
1465 | dm_unregister_target(&origin_target); | |
1466 | bad1: | |
1467 | dm_unregister_target(&snapshot_target); | |
1468 | return r; | |
1469 | } | |
1470 | ||
1471 | static void __exit dm_snapshot_exit(void) | |
1472 | { | |
1473 | int r; | |
1474 | ||
ca3a931f AK |
1475 | destroy_workqueue(ksnapd); |
1476 | ||
1da177e4 LT |
1477 | r = dm_unregister_target(&snapshot_target); |
1478 | if (r) | |
1479 | DMERR("snapshot unregister failed %d", r); | |
1480 | ||
1481 | r = dm_unregister_target(&origin_target); | |
1482 | if (r) | |
1483 | DMERR("origin unregister failed %d", r); | |
1484 | ||
1485 | exit_origin_hash(); | |
1da177e4 LT |
1486 | kmem_cache_destroy(pending_cache); |
1487 | kmem_cache_destroy(exception_cache); | |
cd45daff | 1488 | kmem_cache_destroy(tracked_chunk_cache); |
1da177e4 LT |
1489 | } |
1490 | ||
1491 | /* Module hooks */ | |
1492 | module_init(dm_snapshot_init); | |
1493 | module_exit(dm_snapshot_exit); | |
1494 | ||
1495 | MODULE_DESCRIPTION(DM_NAME " snapshot target"); | |
1496 | MODULE_AUTHOR("Joe Thornber"); | |
1497 | MODULE_LICENSE("GPL"); |