Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
6cca1e7a | 2 | * dm-exception-store.c |
1da177e4 LT |
3 | * |
4 | * Copyright (C) 2001-2002 Sistina Software (UK) Limited. | |
6cca1e7a | 5 | * Copyright (C) 2006 Red Hat GmbH |
1da177e4 LT |
6 | * |
7 | * This file is released under the GPL. | |
8 | */ | |
9 | ||
10 | #include "dm.h" | |
11 | #include "dm-snap.h" | |
1da177e4 LT |
12 | |
13 | #include <linux/mm.h> | |
14 | #include <linux/pagemap.h> | |
15 | #include <linux/vmalloc.h> | |
16 | #include <linux/slab.h> | |
a765e20e AK |
17 | #include <linux/dm-io.h> |
18 | #include <linux/dm-kcopyd.h> | |
1da177e4 | 19 | |
72d94861 | 20 | #define DM_MSG_PREFIX "snapshots" |
4c7e3bf4 | 21 | #define DM_CHUNK_SIZE_DEFAULT_SECTORS 32 /* 16KB */ |
72d94861 | 22 | |
1da177e4 LT |
23 | /*----------------------------------------------------------------- |
24 | * Persistent snapshots, by persistent we mean that the snapshot | |
25 | * will survive a reboot. | |
26 | *---------------------------------------------------------------*/ | |
27 | ||
28 | /* | |
29 | * We need to store a record of which parts of the origin have | |
30 | * been copied to the snapshot device. The snapshot code | |
31 | * requires that we copy exception chunks to chunk aligned areas | |
32 | * of the COW store. It makes sense therefore, to store the | |
33 | * metadata in chunk size blocks. | |
34 | * | |
35 | * There is no backward or forward compatibility implemented, | |
36 | * snapshots with different disk versions than the kernel will | |
37 | * not be usable. It is expected that "lvcreate" will blank out | |
38 | * the start of a fresh COW device before calling the snapshot | |
39 | * constructor. | |
40 | * | |
41 | * The first chunk of the COW device just contains the header. | |
42 | * After this there is a chunk filled with exception metadata, | |
43 | * followed by as many exception chunks as can fit in the | |
44 | * metadata areas. | |
45 | * | |
46 | * All on disk structures are in little-endian format. The end | |
47 | * of the exceptions info is indicated by an exception with a | |
48 | * new_chunk of 0, which is invalid since it would point to the | |
49 | * header chunk. | |
50 | */ | |
51 | ||
52 | /* | |
53 | * Magic for persistent snapshots: "SnAp" - Feeble isn't it. | |
54 | */ | |
55 | #define SNAP_MAGIC 0x70416e53 | |
56 | ||
57 | /* | |
58 | * The on-disk version of the metadata. | |
59 | */ | |
60 | #define SNAPSHOT_DISK_VERSION 1 | |
61 | ||
62 | struct disk_header { | |
63 | uint32_t magic; | |
64 | ||
65 | /* | |
66 | * Is this snapshot valid. There is no way of recovering | |
67 | * an invalid snapshot. | |
68 | */ | |
69 | uint32_t valid; | |
70 | ||
71 | /* | |
72 | * Simple, incrementing version. no backward | |
73 | * compatibility. | |
74 | */ | |
75 | uint32_t version; | |
76 | ||
77 | /* In sectors */ | |
78 | uint32_t chunk_size; | |
79 | }; | |
80 | ||
81 | struct disk_exception { | |
82 | uint64_t old_chunk; | |
83 | uint64_t new_chunk; | |
84 | }; | |
85 | ||
86 | struct commit_callback { | |
87 | void (*callback)(void *, int success); | |
88 | void *context; | |
89 | }; | |
90 | ||
91 | /* | |
92 | * The top level structure for a persistent exception store. | |
93 | */ | |
94 | struct pstore { | |
95 | struct dm_snapshot *snap; /* up pointer to my snapshot */ | |
96 | int version; | |
97 | int valid; | |
1da177e4 LT |
98 | uint32_t exceptions_per_area; |
99 | ||
100 | /* | |
101 | * Now that we have an asynchronous kcopyd there is no | |
102 | * need for large chunk sizes, so it wont hurt to have a | |
103 | * whole chunks worth of metadata in memory at once. | |
104 | */ | |
105 | void *area; | |
106 | ||
107 | /* | |
108 | * Used to keep track of which metadata area the data in | |
109 | * 'chunk' refers to. | |
110 | */ | |
fd14acf6 | 111 | chunk_t current_area; |
1da177e4 LT |
112 | |
113 | /* | |
114 | * The next free chunk for an exception. | |
115 | */ | |
fd14acf6 | 116 | chunk_t next_free; |
1da177e4 LT |
117 | |
118 | /* | |
119 | * The index of next free exception in the current | |
120 | * metadata area. | |
121 | */ | |
122 | uint32_t current_committed; | |
123 | ||
124 | atomic_t pending_count; | |
125 | uint32_t callback_count; | |
126 | struct commit_callback *callbacks; | |
6cca1e7a | 127 | struct dm_io_client *io_client; |
fcac03ab MB |
128 | |
129 | struct workqueue_struct *metadata_wq; | |
1da177e4 LT |
130 | }; |
131 | ||
028867ac | 132 | static unsigned sectors_to_pages(unsigned sectors) |
1da177e4 | 133 | { |
92436262 | 134 | return DIV_ROUND_UP(sectors, PAGE_SIZE >> 9); |
1da177e4 LT |
135 | } |
136 | ||
137 | static int alloc_area(struct pstore *ps) | |
138 | { | |
139 | int r = -ENOMEM; | |
140 | size_t len; | |
141 | ||
c51c2752 | 142 | len = ps->snap->chunk_size << SECTOR_SHIFT; |
1da177e4 LT |
143 | |
144 | /* | |
145 | * Allocate the chunk_size block of memory that will hold | |
146 | * a single metadata area. | |
147 | */ | |
148 | ps->area = vmalloc(len); | |
149 | if (!ps->area) | |
150 | return r; | |
151 | ||
152 | return 0; | |
153 | } | |
154 | ||
155 | static void free_area(struct pstore *ps) | |
156 | { | |
157 | vfree(ps->area); | |
4c7e3bf4 | 158 | ps->area = NULL; |
1da177e4 LT |
159 | } |
160 | ||
fcac03ab | 161 | struct mdata_req { |
22a1ceb1 | 162 | struct dm_io_region *where; |
fcac03ab MB |
163 | struct dm_io_request *io_req; |
164 | struct work_struct work; | |
165 | int result; | |
166 | }; | |
167 | ||
168 | static void do_metadata(struct work_struct *work) | |
169 | { | |
170 | struct mdata_req *req = container_of(work, struct mdata_req, work); | |
171 | ||
172 | req->result = dm_io(req->io_req, 1, req->where, NULL); | |
173 | } | |
174 | ||
1da177e4 LT |
175 | /* |
176 | * Read or write a chunk aligned and sized block of data from a device. | |
177 | */ | |
fd14acf6 | 178 | static int chunk_io(struct pstore *ps, chunk_t chunk, int rw, int metadata) |
1da177e4 | 179 | { |
22a1ceb1 | 180 | struct dm_io_region where = { |
6cca1e7a HM |
181 | .bdev = ps->snap->cow->bdev, |
182 | .sector = ps->snap->chunk_size * chunk, | |
183 | .count = ps->snap->chunk_size, | |
184 | }; | |
185 | struct dm_io_request io_req = { | |
186 | .bi_rw = rw, | |
187 | .mem.type = DM_IO_VMA, | |
188 | .mem.ptr.vma = ps->area, | |
189 | .client = ps->io_client, | |
190 | .notify.fn = NULL, | |
191 | }; | |
fcac03ab MB |
192 | struct mdata_req req; |
193 | ||
194 | if (!metadata) | |
195 | return dm_io(&io_req, 1, &where, NULL); | |
6cca1e7a | 196 | |
fcac03ab MB |
197 | req.where = &where; |
198 | req.io_req = &io_req; | |
199 | ||
200 | /* | |
201 | * Issue the synchronous I/O from a different thread | |
202 | * to avoid generic_make_request recursion. | |
203 | */ | |
204 | INIT_WORK(&req.work, do_metadata); | |
205 | queue_work(ps->metadata_wq, &req.work); | |
206 | flush_workqueue(ps->metadata_wq); | |
207 | ||
208 | return req.result; | |
1da177e4 LT |
209 | } |
210 | ||
a481db78 MP |
211 | /* |
212 | * Convert a metadata area index to a chunk index. | |
213 | */ | |
214 | static chunk_t area_location(struct pstore *ps, chunk_t area) | |
215 | { | |
216 | return 1 + ((ps->exceptions_per_area + 1) * area); | |
217 | } | |
218 | ||
1da177e4 LT |
219 | /* |
220 | * Read or write a metadata area. Remembering to skip the first | |
221 | * chunk which holds the header. | |
222 | */ | |
fd14acf6 | 223 | static int area_io(struct pstore *ps, chunk_t area, int rw) |
1da177e4 LT |
224 | { |
225 | int r; | |
fd14acf6 | 226 | chunk_t chunk; |
1da177e4 | 227 | |
a481db78 | 228 | chunk = area_location(ps, area); |
1da177e4 | 229 | |
fcac03ab | 230 | r = chunk_io(ps, chunk, rw, 0); |
1da177e4 LT |
231 | if (r) |
232 | return r; | |
233 | ||
234 | ps->current_area = area; | |
235 | return 0; | |
236 | } | |
237 | ||
fd14acf6 | 238 | static int zero_area(struct pstore *ps, chunk_t area) |
1da177e4 | 239 | { |
c51c2752 | 240 | memset(ps->area, 0, ps->snap->chunk_size << SECTOR_SHIFT); |
1da177e4 LT |
241 | return area_io(ps, area, WRITE); |
242 | } | |
243 | ||
244 | static int read_header(struct pstore *ps, int *new_snapshot) | |
245 | { | |
246 | int r; | |
247 | struct disk_header *dh; | |
c51c2752 | 248 | chunk_t chunk_size; |
4c7e3bf4 | 249 | int chunk_size_supplied = 1; |
1da177e4 | 250 | |
4c7e3bf4 MM |
251 | /* |
252 | * Use default chunk size (or hardsect_size, if larger) if none supplied | |
253 | */ | |
254 | if (!ps->snap->chunk_size) { | |
255 | ps->snap->chunk_size = max(DM_CHUNK_SIZE_DEFAULT_SECTORS, | |
256 | bdev_hardsect_size(ps->snap->cow->bdev) >> 9); | |
257 | ps->snap->chunk_mask = ps->snap->chunk_size - 1; | |
258 | ps->snap->chunk_shift = ffs(ps->snap->chunk_size) - 1; | |
259 | chunk_size_supplied = 0; | |
260 | } | |
261 | ||
6cca1e7a HM |
262 | ps->io_client = dm_io_client_create(sectors_to_pages(ps->snap-> |
263 | chunk_size)); | |
264 | if (IS_ERR(ps->io_client)) | |
265 | return PTR_ERR(ps->io_client); | |
1da177e4 | 266 | |
4c7e3bf4 MM |
267 | r = alloc_area(ps); |
268 | if (r) | |
6cca1e7a | 269 | return r; |
4c7e3bf4 | 270 | |
fcac03ab | 271 | r = chunk_io(ps, 0, READ, 1); |
4c7e3bf4 | 272 | if (r) |
6cca1e7a | 273 | goto bad; |
4c7e3bf4 | 274 | |
1da177e4 LT |
275 | dh = (struct disk_header *) ps->area; |
276 | ||
277 | if (le32_to_cpu(dh->magic) == 0) { | |
278 | *new_snapshot = 1; | |
4c7e3bf4 MM |
279 | return 0; |
280 | } | |
1da177e4 | 281 | |
4c7e3bf4 MM |
282 | if (le32_to_cpu(dh->magic) != SNAP_MAGIC) { |
283 | DMWARN("Invalid or corrupt snapshot"); | |
1da177e4 | 284 | r = -ENXIO; |
6cca1e7a | 285 | goto bad; |
1da177e4 LT |
286 | } |
287 | ||
4c7e3bf4 MM |
288 | *new_snapshot = 0; |
289 | ps->valid = le32_to_cpu(dh->valid); | |
290 | ps->version = le32_to_cpu(dh->version); | |
291 | chunk_size = le32_to_cpu(dh->chunk_size); | |
292 | ||
293 | if (!chunk_size_supplied || ps->snap->chunk_size == chunk_size) | |
294 | return 0; | |
295 | ||
296 | DMWARN("chunk size %llu in device metadata overrides " | |
297 | "table chunk size of %llu.", | |
298 | (unsigned long long)chunk_size, | |
299 | (unsigned long long)ps->snap->chunk_size); | |
300 | ||
301 | /* We had a bogus chunk_size. Fix stuff up. */ | |
4c7e3bf4 MM |
302 | free_area(ps); |
303 | ||
304 | ps->snap->chunk_size = chunk_size; | |
305 | ps->snap->chunk_mask = chunk_size - 1; | |
306 | ps->snap->chunk_shift = ffs(chunk_size) - 1; | |
307 | ||
6cca1e7a HM |
308 | r = dm_io_client_resize(sectors_to_pages(ps->snap->chunk_size), |
309 | ps->io_client); | |
4c7e3bf4 MM |
310 | if (r) |
311 | return r; | |
312 | ||
313 | r = alloc_area(ps); | |
6cca1e7a | 314 | return r; |
4c7e3bf4 | 315 | |
6cca1e7a | 316 | bad: |
4c7e3bf4 | 317 | free_area(ps); |
1da177e4 LT |
318 | return r; |
319 | } | |
320 | ||
321 | static int write_header(struct pstore *ps) | |
322 | { | |
323 | struct disk_header *dh; | |
324 | ||
c51c2752 | 325 | memset(ps->area, 0, ps->snap->chunk_size << SECTOR_SHIFT); |
1da177e4 LT |
326 | |
327 | dh = (struct disk_header *) ps->area; | |
328 | dh->magic = cpu_to_le32(SNAP_MAGIC); | |
329 | dh->valid = cpu_to_le32(ps->valid); | |
330 | dh->version = cpu_to_le32(ps->version); | |
c51c2752 | 331 | dh->chunk_size = cpu_to_le32(ps->snap->chunk_size); |
1da177e4 | 332 | |
fcac03ab | 333 | return chunk_io(ps, 0, WRITE, 1); |
1da177e4 LT |
334 | } |
335 | ||
336 | /* | |
337 | * Access functions for the disk exceptions, these do the endian conversions. | |
338 | */ | |
339 | static struct disk_exception *get_exception(struct pstore *ps, uint32_t index) | |
340 | { | |
e4ff496d | 341 | BUG_ON(index >= ps->exceptions_per_area); |
1da177e4 LT |
342 | |
343 | return ((struct disk_exception *) ps->area) + index; | |
344 | } | |
345 | ||
e4ff496d MM |
346 | static void read_exception(struct pstore *ps, |
347 | uint32_t index, struct disk_exception *result) | |
1da177e4 | 348 | { |
e4ff496d | 349 | struct disk_exception *e = get_exception(ps, index); |
1da177e4 LT |
350 | |
351 | /* copy it */ | |
352 | result->old_chunk = le64_to_cpu(e->old_chunk); | |
353 | result->new_chunk = le64_to_cpu(e->new_chunk); | |
1da177e4 LT |
354 | } |
355 | ||
e4ff496d MM |
356 | static void write_exception(struct pstore *ps, |
357 | uint32_t index, struct disk_exception *de) | |
1da177e4 | 358 | { |
e4ff496d | 359 | struct disk_exception *e = get_exception(ps, index); |
1da177e4 LT |
360 | |
361 | /* copy it */ | |
362 | e->old_chunk = cpu_to_le64(de->old_chunk); | |
363 | e->new_chunk = cpu_to_le64(de->new_chunk); | |
1da177e4 LT |
364 | } |
365 | ||
366 | /* | |
367 | * Registers the exceptions that are present in the current area. | |
368 | * 'full' is filled in to indicate if the area has been | |
369 | * filled. | |
370 | */ | |
371 | static int insert_exceptions(struct pstore *ps, int *full) | |
372 | { | |
373 | int r; | |
374 | unsigned int i; | |
375 | struct disk_exception de; | |
376 | ||
377 | /* presume the area is full */ | |
378 | *full = 1; | |
379 | ||
380 | for (i = 0; i < ps->exceptions_per_area; i++) { | |
e4ff496d | 381 | read_exception(ps, i, &de); |
1da177e4 LT |
382 | |
383 | /* | |
384 | * If the new_chunk is pointing at the start of | |
385 | * the COW device, where the first metadata area | |
386 | * is we know that we've hit the end of the | |
387 | * exceptions. Therefore the area is not full. | |
388 | */ | |
389 | if (de.new_chunk == 0LL) { | |
390 | ps->current_committed = i; | |
391 | *full = 0; | |
392 | break; | |
393 | } | |
394 | ||
395 | /* | |
396 | * Keep track of the start of the free chunks. | |
397 | */ | |
398 | if (ps->next_free <= de.new_chunk) | |
399 | ps->next_free = de.new_chunk + 1; | |
400 | ||
401 | /* | |
402 | * Otherwise we add the exception to the snapshot. | |
403 | */ | |
404 | r = dm_add_exception(ps->snap, de.old_chunk, de.new_chunk); | |
405 | if (r) | |
406 | return r; | |
407 | } | |
408 | ||
409 | return 0; | |
410 | } | |
411 | ||
412 | static int read_exceptions(struct pstore *ps) | |
413 | { | |
fd14acf6 | 414 | chunk_t area; |
1da177e4 LT |
415 | int r, full = 1; |
416 | ||
417 | /* | |
418 | * Keeping reading chunks and inserting exceptions until | |
419 | * we find a partially full area. | |
420 | */ | |
421 | for (area = 0; full; area++) { | |
422 | r = area_io(ps, area, READ); | |
423 | if (r) | |
424 | return r; | |
425 | ||
426 | r = insert_exceptions(ps, &full); | |
427 | if (r) | |
428 | return r; | |
429 | } | |
430 | ||
431 | return 0; | |
432 | } | |
433 | ||
028867ac | 434 | static struct pstore *get_info(struct exception_store *store) |
1da177e4 LT |
435 | { |
436 | return (struct pstore *) store->context; | |
437 | } | |
438 | ||
439 | static void persistent_fraction_full(struct exception_store *store, | |
440 | sector_t *numerator, sector_t *denominator) | |
441 | { | |
442 | *numerator = get_info(store)->next_free * store->snap->chunk_size; | |
443 | *denominator = get_dev_size(store->snap->cow->bdev); | |
444 | } | |
445 | ||
446 | static void persistent_destroy(struct exception_store *store) | |
447 | { | |
448 | struct pstore *ps = get_info(store); | |
449 | ||
fcac03ab | 450 | destroy_workqueue(ps->metadata_wq); |
6cca1e7a | 451 | dm_io_client_destroy(ps->io_client); |
1da177e4 LT |
452 | vfree(ps->callbacks); |
453 | free_area(ps); | |
454 | kfree(ps); | |
455 | } | |
456 | ||
457 | static int persistent_read_metadata(struct exception_store *store) | |
458 | { | |
e48b9db2 | 459 | int r, uninitialized_var(new_snapshot); |
1da177e4 LT |
460 | struct pstore *ps = get_info(store); |
461 | ||
462 | /* | |
463 | * Read the snapshot header. | |
464 | */ | |
465 | r = read_header(ps, &new_snapshot); | |
466 | if (r) | |
467 | return r; | |
468 | ||
c51c2752 AK |
469 | /* |
470 | * Now we know correct chunk_size, complete the initialisation. | |
471 | */ | |
472 | ps->exceptions_per_area = (ps->snap->chunk_size << SECTOR_SHIFT) / | |
473 | sizeof(struct disk_exception); | |
474 | ps->callbacks = dm_vcalloc(ps->exceptions_per_area, | |
475 | sizeof(*ps->callbacks)); | |
476 | if (!ps->callbacks) | |
477 | return -ENOMEM; | |
478 | ||
1da177e4 LT |
479 | /* |
480 | * Do we need to setup a new snapshot ? | |
481 | */ | |
482 | if (new_snapshot) { | |
483 | r = write_header(ps); | |
484 | if (r) { | |
485 | DMWARN("write_header failed"); | |
486 | return r; | |
487 | } | |
488 | ||
489 | r = zero_area(ps, 0); | |
490 | if (r) { | |
491 | DMWARN("zero_area(0) failed"); | |
492 | return r; | |
493 | } | |
494 | ||
495 | } else { | |
496 | /* | |
497 | * Sanity checks. | |
498 | */ | |
1da177e4 LT |
499 | if (ps->version != SNAPSHOT_DISK_VERSION) { |
500 | DMWARN("unable to handle snapshot disk version %d", | |
501 | ps->version); | |
502 | return -EINVAL; | |
503 | } | |
504 | ||
0764147b MB |
505 | /* |
506 | * Metadata are valid, but snapshot is invalidated | |
507 | */ | |
508 | if (!ps->valid) | |
509 | return 1; | |
510 | ||
1da177e4 LT |
511 | /* |
512 | * Read the metadata. | |
513 | */ | |
514 | r = read_exceptions(ps); | |
515 | if (r) | |
516 | return r; | |
517 | } | |
518 | ||
519 | return 0; | |
520 | } | |
521 | ||
522 | static int persistent_prepare(struct exception_store *store, | |
028867ac | 523 | struct dm_snap_exception *e) |
1da177e4 LT |
524 | { |
525 | struct pstore *ps = get_info(store); | |
526 | uint32_t stride; | |
fd14acf6 | 527 | chunk_t next_free; |
1da177e4 LT |
528 | sector_t size = get_dev_size(store->snap->cow->bdev); |
529 | ||
530 | /* Is there enough room ? */ | |
531 | if (size < ((ps->next_free + 1) * store->snap->chunk_size)) | |
532 | return -ENOSPC; | |
533 | ||
534 | e->new_chunk = ps->next_free; | |
535 | ||
536 | /* | |
537 | * Move onto the next free pending, making sure to take | |
538 | * into account the location of the metadata chunks. | |
539 | */ | |
540 | stride = (ps->exceptions_per_area + 1); | |
fd14acf6 MP |
541 | next_free = ++ps->next_free; |
542 | if (sector_div(next_free, stride) == 1) | |
1da177e4 LT |
543 | ps->next_free++; |
544 | ||
545 | atomic_inc(&ps->pending_count); | |
546 | return 0; | |
547 | } | |
548 | ||
549 | static void persistent_commit(struct exception_store *store, | |
028867ac | 550 | struct dm_snap_exception *e, |
1da177e4 LT |
551 | void (*callback) (void *, int success), |
552 | void *callback_context) | |
553 | { | |
554 | int r; | |
555 | unsigned int i; | |
556 | struct pstore *ps = get_info(store); | |
557 | struct disk_exception de; | |
558 | struct commit_callback *cb; | |
559 | ||
560 | de.old_chunk = e->old_chunk; | |
561 | de.new_chunk = e->new_chunk; | |
562 | write_exception(ps, ps->current_committed++, &de); | |
563 | ||
564 | /* | |
565 | * Add the callback to the back of the array. This code | |
566 | * is the only place where the callback array is | |
567 | * manipulated, and we know that it will never be called | |
568 | * multiple times concurrently. | |
569 | */ | |
570 | cb = ps->callbacks + ps->callback_count++; | |
571 | cb->callback = callback; | |
572 | cb->context = callback_context; | |
573 | ||
574 | /* | |
575 | * If there are no more exceptions in flight, or we have | |
576 | * filled this metadata area we commit the exceptions to | |
577 | * disk. | |
578 | */ | |
579 | if (atomic_dec_and_test(&ps->pending_count) || | |
580 | (ps->current_committed == ps->exceptions_per_area)) { | |
581 | r = area_io(ps, ps->current_area, WRITE); | |
582 | if (r) | |
583 | ps->valid = 0; | |
584 | ||
927ffe7c MM |
585 | /* |
586 | * Have we completely filled the current area ? | |
587 | */ | |
588 | if (ps->current_committed == ps->exceptions_per_area) { | |
589 | ps->current_committed = 0; | |
590 | r = zero_area(ps, ps->current_area + 1); | |
591 | if (r) | |
592 | ps->valid = 0; | |
593 | } | |
594 | ||
1da177e4 LT |
595 | for (i = 0; i < ps->callback_count; i++) { |
596 | cb = ps->callbacks + i; | |
597 | cb->callback(cb->context, r == 0 ? 1 : 0); | |
598 | } | |
599 | ||
600 | ps->callback_count = 0; | |
601 | } | |
1da177e4 LT |
602 | } |
603 | ||
604 | static void persistent_drop(struct exception_store *store) | |
605 | { | |
606 | struct pstore *ps = get_info(store); | |
607 | ||
608 | ps->valid = 0; | |
609 | if (write_header(ps)) | |
610 | DMWARN("write header failed"); | |
611 | } | |
612 | ||
4c7e3bf4 | 613 | int dm_create_persistent(struct exception_store *store) |
1da177e4 | 614 | { |
1da177e4 LT |
615 | struct pstore *ps; |
616 | ||
1da177e4 LT |
617 | /* allocate the pstore */ |
618 | ps = kmalloc(sizeof(*ps), GFP_KERNEL); | |
4c7e3bf4 MM |
619 | if (!ps) |
620 | return -ENOMEM; | |
1da177e4 LT |
621 | |
622 | ps->snap = store->snap; | |
623 | ps->valid = 1; | |
624 | ps->version = SNAPSHOT_DISK_VERSION; | |
4c7e3bf4 | 625 | ps->area = NULL; |
1da177e4 LT |
626 | ps->next_free = 2; /* skipping the header and first area */ |
627 | ps->current_committed = 0; | |
628 | ||
1da177e4 LT |
629 | ps->callback_count = 0; |
630 | atomic_set(&ps->pending_count, 0); | |
c51c2752 | 631 | ps->callbacks = NULL; |
1da177e4 | 632 | |
fcac03ab MB |
633 | ps->metadata_wq = create_singlethread_workqueue("ksnaphd"); |
634 | if (!ps->metadata_wq) { | |
851a8a7f | 635 | kfree(ps); |
fcac03ab MB |
636 | DMERR("couldn't start header metadata update thread"); |
637 | return -ENOMEM; | |
638 | } | |
639 | ||
1da177e4 LT |
640 | store->destroy = persistent_destroy; |
641 | store->read_metadata = persistent_read_metadata; | |
642 | store->prepare_exception = persistent_prepare; | |
643 | store->commit_exception = persistent_commit; | |
644 | store->drop_snapshot = persistent_drop; | |
645 | store->fraction_full = persistent_fraction_full; | |
646 | store->context = ps; | |
647 | ||
648 | return 0; | |
1da177e4 LT |
649 | } |
650 | ||
651 | /*----------------------------------------------------------------- | |
652 | * Implementation of the store for non-persistent snapshots. | |
653 | *---------------------------------------------------------------*/ | |
654 | struct transient_c { | |
655 | sector_t next_free; | |
656 | }; | |
657 | ||
658 | static void transient_destroy(struct exception_store *store) | |
659 | { | |
660 | kfree(store->context); | |
661 | } | |
662 | ||
663 | static int transient_read_metadata(struct exception_store *store) | |
664 | { | |
665 | return 0; | |
666 | } | |
667 | ||
028867ac AK |
668 | static int transient_prepare(struct exception_store *store, |
669 | struct dm_snap_exception *e) | |
1da177e4 LT |
670 | { |
671 | struct transient_c *tc = (struct transient_c *) store->context; | |
672 | sector_t size = get_dev_size(store->snap->cow->bdev); | |
673 | ||
674 | if (size < (tc->next_free + store->snap->chunk_size)) | |
675 | return -1; | |
676 | ||
677 | e->new_chunk = sector_to_chunk(store->snap, tc->next_free); | |
678 | tc->next_free += store->snap->chunk_size; | |
679 | ||
680 | return 0; | |
681 | } | |
682 | ||
683 | static void transient_commit(struct exception_store *store, | |
028867ac AK |
684 | struct dm_snap_exception *e, |
685 | void (*callback) (void *, int success), | |
686 | void *callback_context) | |
1da177e4 LT |
687 | { |
688 | /* Just succeed */ | |
689 | callback(callback_context, 1); | |
690 | } | |
691 | ||
692 | static void transient_fraction_full(struct exception_store *store, | |
693 | sector_t *numerator, sector_t *denominator) | |
694 | { | |
695 | *numerator = ((struct transient_c *) store->context)->next_free; | |
696 | *denominator = get_dev_size(store->snap->cow->bdev); | |
697 | } | |
698 | ||
4c7e3bf4 | 699 | int dm_create_transient(struct exception_store *store) |
1da177e4 LT |
700 | { |
701 | struct transient_c *tc; | |
702 | ||
1da177e4 LT |
703 | store->destroy = transient_destroy; |
704 | store->read_metadata = transient_read_metadata; | |
705 | store->prepare_exception = transient_prepare; | |
706 | store->commit_exception = transient_commit; | |
4c7e3bf4 | 707 | store->drop_snapshot = NULL; |
1da177e4 | 708 | store->fraction_full = transient_fraction_full; |
1da177e4 LT |
709 | |
710 | tc = kmalloc(sizeof(struct transient_c), GFP_KERNEL); | |
711 | if (!tc) | |
712 | return -ENOMEM; | |
713 | ||
714 | tc->next_free = 0; | |
715 | store->context = tc; | |
716 | ||
717 | return 0; | |
718 | } |