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 "dm.h" | |
10 | #include "dm-snap.h" | |
11 | #include "dm-io.h" | |
12 | #include "kcopyd.h" | |
13 | ||
14 | #include <linux/mm.h> | |
15 | #include <linux/pagemap.h> | |
16 | #include <linux/vmalloc.h> | |
17 | #include <linux/slab.h> | |
18 | ||
72d94861 AK |
19 | #define DM_MSG_PREFIX "snapshots" |
20 | ||
1da177e4 LT |
21 | /*----------------------------------------------------------------- |
22 | * Persistent snapshots, by persistent we mean that the snapshot | |
23 | * will survive a reboot. | |
24 | *---------------------------------------------------------------*/ | |
25 | ||
26 | /* | |
27 | * We need to store a record of which parts of the origin have | |
28 | * been copied to the snapshot device. The snapshot code | |
29 | * requires that we copy exception chunks to chunk aligned areas | |
30 | * of the COW store. It makes sense therefore, to store the | |
31 | * metadata in chunk size blocks. | |
32 | * | |
33 | * There is no backward or forward compatibility implemented, | |
34 | * snapshots with different disk versions than the kernel will | |
35 | * not be usable. It is expected that "lvcreate" will blank out | |
36 | * the start of a fresh COW device before calling the snapshot | |
37 | * constructor. | |
38 | * | |
39 | * The first chunk of the COW device just contains the header. | |
40 | * After this there is a chunk filled with exception metadata, | |
41 | * followed by as many exception chunks as can fit in the | |
42 | * metadata areas. | |
43 | * | |
44 | * All on disk structures are in little-endian format. The end | |
45 | * of the exceptions info is indicated by an exception with a | |
46 | * new_chunk of 0, which is invalid since it would point to the | |
47 | * header chunk. | |
48 | */ | |
49 | ||
50 | /* | |
51 | * Magic for persistent snapshots: "SnAp" - Feeble isn't it. | |
52 | */ | |
53 | #define SNAP_MAGIC 0x70416e53 | |
54 | ||
55 | /* | |
56 | * The on-disk version of the metadata. | |
57 | */ | |
58 | #define SNAPSHOT_DISK_VERSION 1 | |
59 | ||
60 | struct disk_header { | |
61 | uint32_t magic; | |
62 | ||
63 | /* | |
64 | * Is this snapshot valid. There is no way of recovering | |
65 | * an invalid snapshot. | |
66 | */ | |
67 | uint32_t valid; | |
68 | ||
69 | /* | |
70 | * Simple, incrementing version. no backward | |
71 | * compatibility. | |
72 | */ | |
73 | uint32_t version; | |
74 | ||
75 | /* In sectors */ | |
76 | uint32_t chunk_size; | |
77 | }; | |
78 | ||
79 | struct disk_exception { | |
80 | uint64_t old_chunk; | |
81 | uint64_t new_chunk; | |
82 | }; | |
83 | ||
84 | struct commit_callback { | |
85 | void (*callback)(void *, int success); | |
86 | void *context; | |
87 | }; | |
88 | ||
89 | /* | |
90 | * The top level structure for a persistent exception store. | |
91 | */ | |
92 | struct pstore { | |
93 | struct dm_snapshot *snap; /* up pointer to my snapshot */ | |
94 | int version; | |
95 | int valid; | |
1da177e4 LT |
96 | uint32_t exceptions_per_area; |
97 | ||
98 | /* | |
99 | * Now that we have an asynchronous kcopyd there is no | |
100 | * need for large chunk sizes, so it wont hurt to have a | |
101 | * whole chunks worth of metadata in memory at once. | |
102 | */ | |
103 | void *area; | |
104 | ||
105 | /* | |
106 | * Used to keep track of which metadata area the data in | |
107 | * 'chunk' refers to. | |
108 | */ | |
109 | uint32_t current_area; | |
110 | ||
111 | /* | |
112 | * The next free chunk for an exception. | |
113 | */ | |
114 | uint32_t next_free; | |
115 | ||
116 | /* | |
117 | * The index of next free exception in the current | |
118 | * metadata area. | |
119 | */ | |
120 | uint32_t current_committed; | |
121 | ||
122 | atomic_t pending_count; | |
123 | uint32_t callback_count; | |
124 | struct commit_callback *callbacks; | |
125 | }; | |
126 | ||
127 | static inline unsigned int sectors_to_pages(unsigned int sectors) | |
128 | { | |
129 | return sectors / (PAGE_SIZE >> 9); | |
130 | } | |
131 | ||
132 | static int alloc_area(struct pstore *ps) | |
133 | { | |
134 | int r = -ENOMEM; | |
135 | size_t len; | |
136 | ||
c51c2752 | 137 | len = ps->snap->chunk_size << SECTOR_SHIFT; |
1da177e4 LT |
138 | |
139 | /* | |
140 | * Allocate the chunk_size block of memory that will hold | |
141 | * a single metadata area. | |
142 | */ | |
143 | ps->area = vmalloc(len); | |
144 | if (!ps->area) | |
145 | return r; | |
146 | ||
147 | return 0; | |
148 | } | |
149 | ||
150 | static void free_area(struct pstore *ps) | |
151 | { | |
152 | vfree(ps->area); | |
153 | } | |
154 | ||
155 | /* | |
156 | * Read or write a chunk aligned and sized block of data from a device. | |
157 | */ | |
158 | static int chunk_io(struct pstore *ps, uint32_t chunk, int rw) | |
159 | { | |
160 | struct io_region where; | |
161 | unsigned long bits; | |
162 | ||
163 | where.bdev = ps->snap->cow->bdev; | |
c51c2752 AK |
164 | where.sector = ps->snap->chunk_size * chunk; |
165 | where.count = ps->snap->chunk_size; | |
1da177e4 LT |
166 | |
167 | return dm_io_sync_vm(1, &where, rw, ps->area, &bits); | |
168 | } | |
169 | ||
170 | /* | |
171 | * Read or write a metadata area. Remembering to skip the first | |
172 | * chunk which holds the header. | |
173 | */ | |
174 | static int area_io(struct pstore *ps, uint32_t area, int rw) | |
175 | { | |
176 | int r; | |
177 | uint32_t chunk; | |
178 | ||
179 | /* convert a metadata area index to a chunk index */ | |
180 | chunk = 1 + ((ps->exceptions_per_area + 1) * area); | |
181 | ||
182 | r = chunk_io(ps, chunk, rw); | |
183 | if (r) | |
184 | return r; | |
185 | ||
186 | ps->current_area = area; | |
187 | return 0; | |
188 | } | |
189 | ||
190 | static int zero_area(struct pstore *ps, uint32_t area) | |
191 | { | |
c51c2752 | 192 | memset(ps->area, 0, ps->snap->chunk_size << SECTOR_SHIFT); |
1da177e4 LT |
193 | return area_io(ps, area, WRITE); |
194 | } | |
195 | ||
196 | static int read_header(struct pstore *ps, int *new_snapshot) | |
197 | { | |
198 | int r; | |
199 | struct disk_header *dh; | |
c51c2752 | 200 | chunk_t chunk_size; |
1da177e4 LT |
201 | |
202 | r = chunk_io(ps, 0, READ); | |
203 | if (r) | |
204 | return r; | |
205 | ||
206 | dh = (struct disk_header *) ps->area; | |
207 | ||
208 | if (le32_to_cpu(dh->magic) == 0) { | |
209 | *new_snapshot = 1; | |
210 | ||
211 | } else if (le32_to_cpu(dh->magic) == SNAP_MAGIC) { | |
212 | *new_snapshot = 0; | |
213 | ps->valid = le32_to_cpu(dh->valid); | |
214 | ps->version = le32_to_cpu(dh->version); | |
c51c2752 AK |
215 | chunk_size = le32_to_cpu(dh->chunk_size); |
216 | if (ps->snap->chunk_size != chunk_size) { | |
217 | DMWARN("chunk size %llu in device metadata overrides " | |
218 | "table chunk size of %llu.", | |
219 | (unsigned long long)chunk_size, | |
220 | (unsigned long long)ps->snap->chunk_size); | |
221 | ||
222 | /* We had a bogus chunk_size. Fix stuff up. */ | |
223 | dm_io_put(sectors_to_pages(ps->snap->chunk_size)); | |
224 | free_area(ps); | |
225 | ||
226 | ps->snap->chunk_size = chunk_size; | |
227 | ps->snap->chunk_mask = chunk_size - 1; | |
228 | ps->snap->chunk_shift = ffs(chunk_size) - 1; | |
229 | ||
230 | r = alloc_area(ps); | |
231 | if (r) | |
232 | return r; | |
233 | ||
234 | r = dm_io_get(sectors_to_pages(chunk_size)); | |
235 | if (r) | |
236 | return r; | |
237 | } | |
1da177e4 LT |
238 | } else { |
239 | DMWARN("Invalid/corrupt snapshot"); | |
240 | r = -ENXIO; | |
241 | } | |
242 | ||
243 | return r; | |
244 | } | |
245 | ||
246 | static int write_header(struct pstore *ps) | |
247 | { | |
248 | struct disk_header *dh; | |
249 | ||
c51c2752 | 250 | memset(ps->area, 0, ps->snap->chunk_size << SECTOR_SHIFT); |
1da177e4 LT |
251 | |
252 | dh = (struct disk_header *) ps->area; | |
253 | dh->magic = cpu_to_le32(SNAP_MAGIC); | |
254 | dh->valid = cpu_to_le32(ps->valid); | |
255 | dh->version = cpu_to_le32(ps->version); | |
c51c2752 | 256 | dh->chunk_size = cpu_to_le32(ps->snap->chunk_size); |
1da177e4 LT |
257 | |
258 | return chunk_io(ps, 0, WRITE); | |
259 | } | |
260 | ||
261 | /* | |
262 | * Access functions for the disk exceptions, these do the endian conversions. | |
263 | */ | |
264 | static struct disk_exception *get_exception(struct pstore *ps, uint32_t index) | |
265 | { | |
266 | if (index >= ps->exceptions_per_area) | |
267 | return NULL; | |
268 | ||
269 | return ((struct disk_exception *) ps->area) + index; | |
270 | } | |
271 | ||
272 | static int read_exception(struct pstore *ps, | |
273 | uint32_t index, struct disk_exception *result) | |
274 | { | |
275 | struct disk_exception *e; | |
276 | ||
277 | e = get_exception(ps, index); | |
278 | if (!e) | |
279 | return -EINVAL; | |
280 | ||
281 | /* copy it */ | |
282 | result->old_chunk = le64_to_cpu(e->old_chunk); | |
283 | result->new_chunk = le64_to_cpu(e->new_chunk); | |
284 | ||
285 | return 0; | |
286 | } | |
287 | ||
288 | static int write_exception(struct pstore *ps, | |
289 | uint32_t index, struct disk_exception *de) | |
290 | { | |
291 | struct disk_exception *e; | |
292 | ||
293 | e = get_exception(ps, index); | |
294 | if (!e) | |
295 | return -EINVAL; | |
296 | ||
297 | /* copy it */ | |
298 | e->old_chunk = cpu_to_le64(de->old_chunk); | |
299 | e->new_chunk = cpu_to_le64(de->new_chunk); | |
300 | ||
301 | return 0; | |
302 | } | |
303 | ||
304 | /* | |
305 | * Registers the exceptions that are present in the current area. | |
306 | * 'full' is filled in to indicate if the area has been | |
307 | * filled. | |
308 | */ | |
309 | static int insert_exceptions(struct pstore *ps, int *full) | |
310 | { | |
311 | int r; | |
312 | unsigned int i; | |
313 | struct disk_exception de; | |
314 | ||
315 | /* presume the area is full */ | |
316 | *full = 1; | |
317 | ||
318 | for (i = 0; i < ps->exceptions_per_area; i++) { | |
319 | r = read_exception(ps, i, &de); | |
320 | ||
321 | if (r) | |
322 | return r; | |
323 | ||
324 | /* | |
325 | * If the new_chunk is pointing at the start of | |
326 | * the COW device, where the first metadata area | |
327 | * is we know that we've hit the end of the | |
328 | * exceptions. Therefore the area is not full. | |
329 | */ | |
330 | if (de.new_chunk == 0LL) { | |
331 | ps->current_committed = i; | |
332 | *full = 0; | |
333 | break; | |
334 | } | |
335 | ||
336 | /* | |
337 | * Keep track of the start of the free chunks. | |
338 | */ | |
339 | if (ps->next_free <= de.new_chunk) | |
340 | ps->next_free = de.new_chunk + 1; | |
341 | ||
342 | /* | |
343 | * Otherwise we add the exception to the snapshot. | |
344 | */ | |
345 | r = dm_add_exception(ps->snap, de.old_chunk, de.new_chunk); | |
346 | if (r) | |
347 | return r; | |
348 | } | |
349 | ||
350 | return 0; | |
351 | } | |
352 | ||
353 | static int read_exceptions(struct pstore *ps) | |
354 | { | |
355 | uint32_t area; | |
356 | int r, full = 1; | |
357 | ||
358 | /* | |
359 | * Keeping reading chunks and inserting exceptions until | |
360 | * we find a partially full area. | |
361 | */ | |
362 | for (area = 0; full; area++) { | |
363 | r = area_io(ps, area, READ); | |
364 | if (r) | |
365 | return r; | |
366 | ||
367 | r = insert_exceptions(ps, &full); | |
368 | if (r) | |
369 | return r; | |
370 | } | |
371 | ||
372 | return 0; | |
373 | } | |
374 | ||
375 | static inline struct pstore *get_info(struct exception_store *store) | |
376 | { | |
377 | return (struct pstore *) store->context; | |
378 | } | |
379 | ||
380 | static void persistent_fraction_full(struct exception_store *store, | |
381 | sector_t *numerator, sector_t *denominator) | |
382 | { | |
383 | *numerator = get_info(store)->next_free * store->snap->chunk_size; | |
384 | *denominator = get_dev_size(store->snap->cow->bdev); | |
385 | } | |
386 | ||
387 | static void persistent_destroy(struct exception_store *store) | |
388 | { | |
389 | struct pstore *ps = get_info(store); | |
390 | ||
c51c2752 | 391 | dm_io_put(sectors_to_pages(ps->snap->chunk_size)); |
1da177e4 LT |
392 | vfree(ps->callbacks); |
393 | free_area(ps); | |
394 | kfree(ps); | |
395 | } | |
396 | ||
397 | static int persistent_read_metadata(struct exception_store *store) | |
398 | { | |
399 | int r, new_snapshot; | |
400 | struct pstore *ps = get_info(store); | |
401 | ||
402 | /* | |
403 | * Read the snapshot header. | |
404 | */ | |
405 | r = read_header(ps, &new_snapshot); | |
406 | if (r) | |
407 | return r; | |
408 | ||
c51c2752 AK |
409 | /* |
410 | * Now we know correct chunk_size, complete the initialisation. | |
411 | */ | |
412 | ps->exceptions_per_area = (ps->snap->chunk_size << SECTOR_SHIFT) / | |
413 | sizeof(struct disk_exception); | |
414 | ps->callbacks = dm_vcalloc(ps->exceptions_per_area, | |
415 | sizeof(*ps->callbacks)); | |
416 | if (!ps->callbacks) | |
417 | return -ENOMEM; | |
418 | ||
1da177e4 LT |
419 | /* |
420 | * Do we need to setup a new snapshot ? | |
421 | */ | |
422 | if (new_snapshot) { | |
423 | r = write_header(ps); | |
424 | if (r) { | |
425 | DMWARN("write_header failed"); | |
426 | return r; | |
427 | } | |
428 | ||
429 | r = zero_area(ps, 0); | |
430 | if (r) { | |
431 | DMWARN("zero_area(0) failed"); | |
432 | return r; | |
433 | } | |
434 | ||
435 | } else { | |
436 | /* | |
437 | * Sanity checks. | |
438 | */ | |
439 | if (!ps->valid) { | |
440 | DMWARN("snapshot is marked invalid"); | |
441 | return -EINVAL; | |
442 | } | |
443 | ||
444 | if (ps->version != SNAPSHOT_DISK_VERSION) { | |
445 | DMWARN("unable to handle snapshot disk version %d", | |
446 | ps->version); | |
447 | return -EINVAL; | |
448 | } | |
449 | ||
450 | /* | |
451 | * Read the metadata. | |
452 | */ | |
453 | r = read_exceptions(ps); | |
454 | if (r) | |
455 | return r; | |
456 | } | |
457 | ||
458 | return 0; | |
459 | } | |
460 | ||
461 | static int persistent_prepare(struct exception_store *store, | |
462 | struct exception *e) | |
463 | { | |
464 | struct pstore *ps = get_info(store); | |
465 | uint32_t stride; | |
466 | sector_t size = get_dev_size(store->snap->cow->bdev); | |
467 | ||
468 | /* Is there enough room ? */ | |
469 | if (size < ((ps->next_free + 1) * store->snap->chunk_size)) | |
470 | return -ENOSPC; | |
471 | ||
472 | e->new_chunk = ps->next_free; | |
473 | ||
474 | /* | |
475 | * Move onto the next free pending, making sure to take | |
476 | * into account the location of the metadata chunks. | |
477 | */ | |
478 | stride = (ps->exceptions_per_area + 1); | |
479 | if ((++ps->next_free % stride) == 1) | |
480 | ps->next_free++; | |
481 | ||
482 | atomic_inc(&ps->pending_count); | |
483 | return 0; | |
484 | } | |
485 | ||
486 | static void persistent_commit(struct exception_store *store, | |
487 | struct exception *e, | |
488 | void (*callback) (void *, int success), | |
489 | void *callback_context) | |
490 | { | |
491 | int r; | |
492 | unsigned int i; | |
493 | struct pstore *ps = get_info(store); | |
494 | struct disk_exception de; | |
495 | struct commit_callback *cb; | |
496 | ||
497 | de.old_chunk = e->old_chunk; | |
498 | de.new_chunk = e->new_chunk; | |
499 | write_exception(ps, ps->current_committed++, &de); | |
500 | ||
501 | /* | |
502 | * Add the callback to the back of the array. This code | |
503 | * is the only place where the callback array is | |
504 | * manipulated, and we know that it will never be called | |
505 | * multiple times concurrently. | |
506 | */ | |
507 | cb = ps->callbacks + ps->callback_count++; | |
508 | cb->callback = callback; | |
509 | cb->context = callback_context; | |
510 | ||
511 | /* | |
512 | * If there are no more exceptions in flight, or we have | |
513 | * filled this metadata area we commit the exceptions to | |
514 | * disk. | |
515 | */ | |
516 | if (atomic_dec_and_test(&ps->pending_count) || | |
517 | (ps->current_committed == ps->exceptions_per_area)) { | |
518 | r = area_io(ps, ps->current_area, WRITE); | |
519 | if (r) | |
520 | ps->valid = 0; | |
521 | ||
522 | for (i = 0; i < ps->callback_count; i++) { | |
523 | cb = ps->callbacks + i; | |
524 | cb->callback(cb->context, r == 0 ? 1 : 0); | |
525 | } | |
526 | ||
527 | ps->callback_count = 0; | |
528 | } | |
529 | ||
530 | /* | |
531 | * Have we completely filled the current area ? | |
532 | */ | |
533 | if (ps->current_committed == ps->exceptions_per_area) { | |
534 | ps->current_committed = 0; | |
535 | r = zero_area(ps, ps->current_area + 1); | |
536 | if (r) | |
537 | ps->valid = 0; | |
538 | } | |
539 | } | |
540 | ||
541 | static void persistent_drop(struct exception_store *store) | |
542 | { | |
543 | struct pstore *ps = get_info(store); | |
544 | ||
545 | ps->valid = 0; | |
546 | if (write_header(ps)) | |
547 | DMWARN("write header failed"); | |
548 | } | |
549 | ||
550 | int dm_create_persistent(struct exception_store *store, uint32_t chunk_size) | |
551 | { | |
552 | int r; | |
553 | struct pstore *ps; | |
554 | ||
555 | r = dm_io_get(sectors_to_pages(chunk_size)); | |
556 | if (r) | |
557 | return r; | |
558 | ||
559 | /* allocate the pstore */ | |
560 | ps = kmalloc(sizeof(*ps), GFP_KERNEL); | |
561 | if (!ps) { | |
562 | r = -ENOMEM; | |
563 | goto bad; | |
564 | } | |
565 | ||
566 | ps->snap = store->snap; | |
567 | ps->valid = 1; | |
568 | ps->version = SNAPSHOT_DISK_VERSION; | |
1da177e4 LT |
569 | ps->next_free = 2; /* skipping the header and first area */ |
570 | ps->current_committed = 0; | |
571 | ||
572 | r = alloc_area(ps); | |
573 | if (r) | |
574 | goto bad; | |
575 | ||
1da177e4 LT |
576 | ps->callback_count = 0; |
577 | atomic_set(&ps->pending_count, 0); | |
c51c2752 | 578 | ps->callbacks = NULL; |
1da177e4 LT |
579 | |
580 | store->destroy = persistent_destroy; | |
581 | store->read_metadata = persistent_read_metadata; | |
582 | store->prepare_exception = persistent_prepare; | |
583 | store->commit_exception = persistent_commit; | |
584 | store->drop_snapshot = persistent_drop; | |
585 | store->fraction_full = persistent_fraction_full; | |
586 | store->context = ps; | |
587 | ||
588 | return 0; | |
589 | ||
590 | bad: | |
591 | dm_io_put(sectors_to_pages(chunk_size)); | |
f9101210 JJ |
592 | if (ps && ps->area) |
593 | free_area(ps); | |
594 | kfree(ps); | |
1da177e4 LT |
595 | return r; |
596 | } | |
597 | ||
598 | /*----------------------------------------------------------------- | |
599 | * Implementation of the store for non-persistent snapshots. | |
600 | *---------------------------------------------------------------*/ | |
601 | struct transient_c { | |
602 | sector_t next_free; | |
603 | }; | |
604 | ||
605 | static void transient_destroy(struct exception_store *store) | |
606 | { | |
607 | kfree(store->context); | |
608 | } | |
609 | ||
610 | static int transient_read_metadata(struct exception_store *store) | |
611 | { | |
612 | return 0; | |
613 | } | |
614 | ||
615 | static int transient_prepare(struct exception_store *store, struct exception *e) | |
616 | { | |
617 | struct transient_c *tc = (struct transient_c *) store->context; | |
618 | sector_t size = get_dev_size(store->snap->cow->bdev); | |
619 | ||
620 | if (size < (tc->next_free + store->snap->chunk_size)) | |
621 | return -1; | |
622 | ||
623 | e->new_chunk = sector_to_chunk(store->snap, tc->next_free); | |
624 | tc->next_free += store->snap->chunk_size; | |
625 | ||
626 | return 0; | |
627 | } | |
628 | ||
629 | static void transient_commit(struct exception_store *store, | |
630 | struct exception *e, | |
631 | void (*callback) (void *, int success), | |
632 | void *callback_context) | |
633 | { | |
634 | /* Just succeed */ | |
635 | callback(callback_context, 1); | |
636 | } | |
637 | ||
638 | static void transient_fraction_full(struct exception_store *store, | |
639 | sector_t *numerator, sector_t *denominator) | |
640 | { | |
641 | *numerator = ((struct transient_c *) store->context)->next_free; | |
642 | *denominator = get_dev_size(store->snap->cow->bdev); | |
643 | } | |
644 | ||
645 | int dm_create_transient(struct exception_store *store, | |
646 | struct dm_snapshot *s, int blocksize) | |
647 | { | |
648 | struct transient_c *tc; | |
649 | ||
650 | memset(store, 0, sizeof(*store)); | |
651 | store->destroy = transient_destroy; | |
652 | store->read_metadata = transient_read_metadata; | |
653 | store->prepare_exception = transient_prepare; | |
654 | store->commit_exception = transient_commit; | |
655 | store->fraction_full = transient_fraction_full; | |
656 | store->snap = s; | |
657 | ||
658 | tc = kmalloc(sizeof(struct transient_c), GFP_KERNEL); | |
659 | if (!tc) | |
660 | return -ENOMEM; | |
661 | ||
662 | tc->next_free = 0; | |
663 | store->context = tc; | |
664 | ||
665 | return 0; | |
666 | } |