]>
Commit | Line | Data |
---|---|---|
4db6bfe0 AK |
1 | /* |
2 | * Copyright (C) 2001-2002 Sistina Software (UK) Limited. | |
3 | * Copyright (C) 2006-2008 Red Hat GmbH | |
4 | * | |
5 | * This file is released under the GPL. | |
6 | */ | |
7 | ||
8 | #include "dm-exception-store.h" | |
4db6bfe0 AK |
9 | |
10 | #include <linux/mm.h> | |
11 | #include <linux/pagemap.h> | |
12 | #include <linux/vmalloc.h> | |
13 | #include <linux/slab.h> | |
14 | #include <linux/dm-io.h> | |
15 | ||
16 | #define DM_MSG_PREFIX "persistent snapshot" | |
17 | #define DM_CHUNK_SIZE_DEFAULT_SECTORS 32 /* 16KB */ | |
18 | ||
19 | /*----------------------------------------------------------------- | |
20 | * Persistent snapshots, by persistent we mean that the snapshot | |
21 | * will survive a reboot. | |
22 | *---------------------------------------------------------------*/ | |
23 | ||
24 | /* | |
25 | * We need to store a record of which parts of the origin have | |
26 | * been copied to the snapshot device. The snapshot code | |
27 | * requires that we copy exception chunks to chunk aligned areas | |
28 | * of the COW store. It makes sense therefore, to store the | |
29 | * metadata in chunk size blocks. | |
30 | * | |
31 | * There is no backward or forward compatibility implemented, | |
32 | * snapshots with different disk versions than the kernel will | |
33 | * not be usable. It is expected that "lvcreate" will blank out | |
34 | * the start of a fresh COW device before calling the snapshot | |
35 | * constructor. | |
36 | * | |
37 | * The first chunk of the COW device just contains the header. | |
38 | * After this there is a chunk filled with exception metadata, | |
39 | * followed by as many exception chunks as can fit in the | |
40 | * metadata areas. | |
41 | * | |
42 | * All on disk structures are in little-endian format. The end | |
43 | * of the exceptions info is indicated by an exception with a | |
44 | * new_chunk of 0, which is invalid since it would point to the | |
45 | * header chunk. | |
46 | */ | |
47 | ||
48 | /* | |
49 | * Magic for persistent snapshots: "SnAp" - Feeble isn't it. | |
50 | */ | |
51 | #define SNAP_MAGIC 0x70416e53 | |
52 | ||
53 | /* | |
54 | * The on-disk version of the metadata. | |
55 | */ | |
56 | #define SNAPSHOT_DISK_VERSION 1 | |
57 | ||
4454a621 MP |
58 | #define NUM_SNAPSHOT_HDR_CHUNKS 1 |
59 | ||
4db6bfe0 | 60 | struct disk_header { |
283a8328 | 61 | __le32 magic; |
4db6bfe0 AK |
62 | |
63 | /* | |
64 | * Is this snapshot valid. There is no way of recovering | |
65 | * an invalid snapshot. | |
66 | */ | |
283a8328 | 67 | __le32 valid; |
4db6bfe0 AK |
68 | |
69 | /* | |
70 | * Simple, incrementing version. no backward | |
71 | * compatibility. | |
72 | */ | |
283a8328 | 73 | __le32 version; |
4db6bfe0 AK |
74 | |
75 | /* In sectors */ | |
283a8328 AK |
76 | __le32 chunk_size; |
77 | } __packed; | |
4db6bfe0 AK |
78 | |
79 | struct disk_exception { | |
283a8328 AK |
80 | __le64 old_chunk; |
81 | __le64 new_chunk; | |
82 | } __packed; | |
83 | ||
84 | struct core_exception { | |
4db6bfe0 AK |
85 | uint64_t old_chunk; |
86 | uint64_t new_chunk; | |
87 | }; | |
88 | ||
89 | struct commit_callback { | |
90 | void (*callback)(void *, int success); | |
91 | void *context; | |
92 | }; | |
93 | ||
94 | /* | |
95 | * The top level structure for a persistent exception store. | |
96 | */ | |
97 | struct pstore { | |
71fab00a | 98 | struct dm_exception_store *store; |
4db6bfe0 AK |
99 | int version; |
100 | int valid; | |
101 | uint32_t exceptions_per_area; | |
102 | ||
103 | /* | |
104 | * Now that we have an asynchronous kcopyd there is no | |
105 | * need for large chunk sizes, so it wont hurt to have a | |
106 | * whole chunks worth of metadata in memory at once. | |
107 | */ | |
108 | void *area; | |
109 | ||
110 | /* | |
111 | * An area of zeros used to clear the next area. | |
112 | */ | |
113 | void *zero_area; | |
114 | ||
61578dcd MP |
115 | /* |
116 | * An area used for header. The header can be written | |
117 | * concurrently with metadata (when invalidating the snapshot), | |
118 | * so it needs a separate buffer. | |
119 | */ | |
120 | void *header_area; | |
121 | ||
4db6bfe0 AK |
122 | /* |
123 | * Used to keep track of which metadata area the data in | |
124 | * 'chunk' refers to. | |
125 | */ | |
126 | chunk_t current_area; | |
127 | ||
128 | /* | |
129 | * The next free chunk for an exception. | |
4454a621 MP |
130 | * |
131 | * When creating exceptions, all the chunks here and above are | |
132 | * free. It holds the next chunk to be allocated. On rare | |
133 | * occasions (e.g. after a system crash) holes can be left in | |
134 | * the exception store because chunks can be committed out of | |
135 | * order. | |
136 | * | |
137 | * When merging exceptions, it does not necessarily mean all the | |
138 | * chunks here and above are free. It holds the value it would | |
139 | * have held if all chunks had been committed in order of | |
140 | * allocation. Consequently the value may occasionally be | |
141 | * slightly too low, but since it's only used for 'status' and | |
142 | * it can never reach its minimum value too early this doesn't | |
143 | * matter. | |
4db6bfe0 | 144 | */ |
4454a621 | 145 | |
4db6bfe0 AK |
146 | chunk_t next_free; |
147 | ||
148 | /* | |
149 | * The index of next free exception in the current | |
150 | * metadata area. | |
151 | */ | |
152 | uint32_t current_committed; | |
153 | ||
154 | atomic_t pending_count; | |
155 | uint32_t callback_count; | |
156 | struct commit_callback *callbacks; | |
157 | struct dm_io_client *io_client; | |
158 | ||
159 | struct workqueue_struct *metadata_wq; | |
160 | }; | |
161 | ||
4db6bfe0 AK |
162 | static int alloc_area(struct pstore *ps) |
163 | { | |
164 | int r = -ENOMEM; | |
165 | size_t len; | |
166 | ||
71fab00a | 167 | len = ps->store->chunk_size << SECTOR_SHIFT; |
4db6bfe0 AK |
168 | |
169 | /* | |
170 | * Allocate the chunk_size block of memory that will hold | |
171 | * a single metadata area. | |
172 | */ | |
173 | ps->area = vmalloc(len); | |
174 | if (!ps->area) | |
61578dcd | 175 | goto err_area; |
4db6bfe0 | 176 | |
e29e65aa | 177 | ps->zero_area = vzalloc(len); |
61578dcd MP |
178 | if (!ps->zero_area) |
179 | goto err_zero_area; | |
4db6bfe0 | 180 | |
61578dcd MP |
181 | ps->header_area = vmalloc(len); |
182 | if (!ps->header_area) | |
183 | goto err_header_area; | |
184 | ||
4db6bfe0 | 185 | return 0; |
61578dcd MP |
186 | |
187 | err_header_area: | |
188 | vfree(ps->zero_area); | |
189 | ||
190 | err_zero_area: | |
191 | vfree(ps->area); | |
192 | ||
193 | err_area: | |
194 | return r; | |
4db6bfe0 AK |
195 | } |
196 | ||
197 | static void free_area(struct pstore *ps) | |
198 | { | |
a32079ce JB |
199 | if (ps->area) |
200 | vfree(ps->area); | |
4db6bfe0 | 201 | ps->area = NULL; |
a32079ce JB |
202 | |
203 | if (ps->zero_area) | |
204 | vfree(ps->zero_area); | |
4db6bfe0 | 205 | ps->zero_area = NULL; |
61578dcd MP |
206 | |
207 | if (ps->header_area) | |
208 | vfree(ps->header_area); | |
209 | ps->header_area = NULL; | |
4db6bfe0 AK |
210 | } |
211 | ||
212 | struct mdata_req { | |
213 | struct dm_io_region *where; | |
214 | struct dm_io_request *io_req; | |
215 | struct work_struct work; | |
216 | int result; | |
217 | }; | |
218 | ||
219 | static void do_metadata(struct work_struct *work) | |
220 | { | |
221 | struct mdata_req *req = container_of(work, struct mdata_req, work); | |
222 | ||
223 | req->result = dm_io(req->io_req, 1, req->where, NULL); | |
224 | } | |
225 | ||
226 | /* | |
227 | * Read or write a chunk aligned and sized block of data from a device. | |
228 | */ | |
02d2fd31 MP |
229 | static int chunk_io(struct pstore *ps, void *area, chunk_t chunk, int rw, |
230 | int metadata) | |
4db6bfe0 AK |
231 | { |
232 | struct dm_io_region where = { | |
fc56f6fb | 233 | .bdev = dm_snap_cow(ps->store->snap)->bdev, |
71fab00a JB |
234 | .sector = ps->store->chunk_size * chunk, |
235 | .count = ps->store->chunk_size, | |
4db6bfe0 AK |
236 | }; |
237 | struct dm_io_request io_req = { | |
238 | .bi_rw = rw, | |
239 | .mem.type = DM_IO_VMA, | |
02d2fd31 | 240 | .mem.ptr.vma = area, |
4db6bfe0 AK |
241 | .client = ps->io_client, |
242 | .notify.fn = NULL, | |
243 | }; | |
244 | struct mdata_req req; | |
245 | ||
246 | if (!metadata) | |
247 | return dm_io(&io_req, 1, &where, NULL); | |
248 | ||
249 | req.where = &where; | |
250 | req.io_req = &io_req; | |
251 | ||
252 | /* | |
253 | * Issue the synchronous I/O from a different thread | |
254 | * to avoid generic_make_request recursion. | |
255 | */ | |
ca1cab37 | 256 | INIT_WORK_ONSTACK(&req.work, do_metadata); |
4db6bfe0 | 257 | queue_work(ps->metadata_wq, &req.work); |
239c8dd5 | 258 | flush_work(&req.work); |
4db6bfe0 AK |
259 | |
260 | return req.result; | |
261 | } | |
262 | ||
263 | /* | |
264 | * Convert a metadata area index to a chunk index. | |
265 | */ | |
266 | static chunk_t area_location(struct pstore *ps, chunk_t area) | |
267 | { | |
87c961cb | 268 | return NUM_SNAPSHOT_HDR_CHUNKS + ((ps->exceptions_per_area + 1) * area); |
4db6bfe0 AK |
269 | } |
270 | ||
271 | /* | |
272 | * Read or write a metadata area. Remembering to skip the first | |
273 | * chunk which holds the header. | |
274 | */ | |
275 | static int area_io(struct pstore *ps, int rw) | |
276 | { | |
277 | int r; | |
278 | chunk_t chunk; | |
279 | ||
280 | chunk = area_location(ps, ps->current_area); | |
281 | ||
02d2fd31 | 282 | r = chunk_io(ps, ps->area, chunk, rw, 0); |
4db6bfe0 AK |
283 | if (r) |
284 | return r; | |
285 | ||
286 | return 0; | |
287 | } | |
288 | ||
289 | static void zero_memory_area(struct pstore *ps) | |
290 | { | |
71fab00a | 291 | memset(ps->area, 0, ps->store->chunk_size << SECTOR_SHIFT); |
4db6bfe0 AK |
292 | } |
293 | ||
294 | static int zero_disk_area(struct pstore *ps, chunk_t area) | |
295 | { | |
02d2fd31 | 296 | return chunk_io(ps, ps->zero_area, area_location(ps, area), WRITE, 0); |
4db6bfe0 AK |
297 | } |
298 | ||
299 | static int read_header(struct pstore *ps, int *new_snapshot) | |
300 | { | |
301 | int r; | |
302 | struct disk_header *dh; | |
df96eee6 | 303 | unsigned chunk_size; |
4db6bfe0 | 304 | int chunk_size_supplied = 1; |
ae0b7448 | 305 | char *chunk_err; |
4db6bfe0 AK |
306 | |
307 | /* | |
df96eee6 MP |
308 | * Use default chunk size (or logical_block_size, if larger) |
309 | * if none supplied | |
4db6bfe0 | 310 | */ |
71fab00a JB |
311 | if (!ps->store->chunk_size) { |
312 | ps->store->chunk_size = max(DM_CHUNK_SIZE_DEFAULT_SECTORS, | |
fc56f6fb MS |
313 | bdev_logical_block_size(dm_snap_cow(ps->store->snap)-> |
314 | bdev) >> 9); | |
71fab00a JB |
315 | ps->store->chunk_mask = ps->store->chunk_size - 1; |
316 | ps->store->chunk_shift = ffs(ps->store->chunk_size) - 1; | |
4db6bfe0 AK |
317 | chunk_size_supplied = 0; |
318 | } | |
319 | ||
bda8efec | 320 | ps->io_client = dm_io_client_create(); |
4db6bfe0 AK |
321 | if (IS_ERR(ps->io_client)) |
322 | return PTR_ERR(ps->io_client); | |
323 | ||
324 | r = alloc_area(ps); | |
325 | if (r) | |
326 | return r; | |
327 | ||
61578dcd | 328 | r = chunk_io(ps, ps->header_area, 0, READ, 1); |
4db6bfe0 AK |
329 | if (r) |
330 | goto bad; | |
331 | ||
61578dcd | 332 | dh = ps->header_area; |
4db6bfe0 AK |
333 | |
334 | if (le32_to_cpu(dh->magic) == 0) { | |
335 | *new_snapshot = 1; | |
336 | return 0; | |
337 | } | |
338 | ||
339 | if (le32_to_cpu(dh->magic) != SNAP_MAGIC) { | |
340 | DMWARN("Invalid or corrupt snapshot"); | |
341 | r = -ENXIO; | |
342 | goto bad; | |
343 | } | |
344 | ||
345 | *new_snapshot = 0; | |
346 | ps->valid = le32_to_cpu(dh->valid); | |
347 | ps->version = le32_to_cpu(dh->version); | |
348 | chunk_size = le32_to_cpu(dh->chunk_size); | |
349 | ||
ae0b7448 | 350 | if (ps->store->chunk_size == chunk_size) |
4db6bfe0 AK |
351 | return 0; |
352 | ||
ae0b7448 | 353 | if (chunk_size_supplied) |
df96eee6 MP |
354 | DMWARN("chunk size %u in device metadata overrides " |
355 | "table chunk size of %u.", | |
356 | chunk_size, ps->store->chunk_size); | |
4db6bfe0 AK |
357 | |
358 | /* We had a bogus chunk_size. Fix stuff up. */ | |
359 | free_area(ps); | |
360 | ||
ae0b7448 MP |
361 | r = dm_exception_store_set_chunk_size(ps->store, chunk_size, |
362 | &chunk_err); | |
363 | if (r) { | |
df96eee6 MP |
364 | DMERR("invalid on-disk chunk size %u: %s.", |
365 | chunk_size, chunk_err); | |
ae0b7448 MP |
366 | return r; |
367 | } | |
4db6bfe0 | 368 | |
4db6bfe0 AK |
369 | r = alloc_area(ps); |
370 | return r; | |
371 | ||
372 | bad: | |
373 | free_area(ps); | |
374 | return r; | |
375 | } | |
376 | ||
377 | static int write_header(struct pstore *ps) | |
378 | { | |
379 | struct disk_header *dh; | |
380 | ||
61578dcd | 381 | memset(ps->header_area, 0, ps->store->chunk_size << SECTOR_SHIFT); |
4db6bfe0 | 382 | |
61578dcd | 383 | dh = ps->header_area; |
4db6bfe0 AK |
384 | dh->magic = cpu_to_le32(SNAP_MAGIC); |
385 | dh->valid = cpu_to_le32(ps->valid); | |
386 | dh->version = cpu_to_le32(ps->version); | |
71fab00a | 387 | dh->chunk_size = cpu_to_le32(ps->store->chunk_size); |
4db6bfe0 | 388 | |
61578dcd | 389 | return chunk_io(ps, ps->header_area, 0, WRITE, 1); |
4db6bfe0 AK |
390 | } |
391 | ||
392 | /* | |
393 | * Access functions for the disk exceptions, these do the endian conversions. | |
394 | */ | |
395 | static struct disk_exception *get_exception(struct pstore *ps, uint32_t index) | |
396 | { | |
397 | BUG_ON(index >= ps->exceptions_per_area); | |
398 | ||
399 | return ((struct disk_exception *) ps->area) + index; | |
400 | } | |
401 | ||
402 | static void read_exception(struct pstore *ps, | |
283a8328 | 403 | uint32_t index, struct core_exception *result) |
4db6bfe0 | 404 | { |
283a8328 | 405 | struct disk_exception *de = get_exception(ps, index); |
4db6bfe0 AK |
406 | |
407 | /* copy it */ | |
283a8328 AK |
408 | result->old_chunk = le64_to_cpu(de->old_chunk); |
409 | result->new_chunk = le64_to_cpu(de->new_chunk); | |
4db6bfe0 AK |
410 | } |
411 | ||
412 | static void write_exception(struct pstore *ps, | |
283a8328 | 413 | uint32_t index, struct core_exception *e) |
4db6bfe0 | 414 | { |
283a8328 | 415 | struct disk_exception *de = get_exception(ps, index); |
4db6bfe0 AK |
416 | |
417 | /* copy it */ | |
283a8328 AK |
418 | de->old_chunk = cpu_to_le64(e->old_chunk); |
419 | de->new_chunk = cpu_to_le64(e->new_chunk); | |
4db6bfe0 AK |
420 | } |
421 | ||
4454a621 MP |
422 | static void clear_exception(struct pstore *ps, uint32_t index) |
423 | { | |
283a8328 | 424 | struct disk_exception *de = get_exception(ps, index); |
4454a621 MP |
425 | |
426 | /* clear it */ | |
283a8328 AK |
427 | de->old_chunk = 0; |
428 | de->new_chunk = 0; | |
4454a621 MP |
429 | } |
430 | ||
4db6bfe0 AK |
431 | /* |
432 | * Registers the exceptions that are present in the current area. | |
433 | * 'full' is filled in to indicate if the area has been | |
434 | * filled. | |
435 | */ | |
a159c1ac JB |
436 | static int insert_exceptions(struct pstore *ps, |
437 | int (*callback)(void *callback_context, | |
438 | chunk_t old, chunk_t new), | |
439 | void *callback_context, | |
440 | int *full) | |
4db6bfe0 AK |
441 | { |
442 | int r; | |
443 | unsigned int i; | |
283a8328 | 444 | struct core_exception e; |
4db6bfe0 AK |
445 | |
446 | /* presume the area is full */ | |
447 | *full = 1; | |
448 | ||
449 | for (i = 0; i < ps->exceptions_per_area; i++) { | |
283a8328 | 450 | read_exception(ps, i, &e); |
4db6bfe0 AK |
451 | |
452 | /* | |
453 | * If the new_chunk is pointing at the start of | |
454 | * the COW device, where the first metadata area | |
455 | * is we know that we've hit the end of the | |
456 | * exceptions. Therefore the area is not full. | |
457 | */ | |
283a8328 | 458 | if (e.new_chunk == 0LL) { |
4db6bfe0 AK |
459 | ps->current_committed = i; |
460 | *full = 0; | |
461 | break; | |
462 | } | |
463 | ||
464 | /* | |
465 | * Keep track of the start of the free chunks. | |
466 | */ | |
283a8328 AK |
467 | if (ps->next_free <= e.new_chunk) |
468 | ps->next_free = e.new_chunk + 1; | |
4db6bfe0 AK |
469 | |
470 | /* | |
471 | * Otherwise we add the exception to the snapshot. | |
472 | */ | |
283a8328 | 473 | r = callback(callback_context, e.old_chunk, e.new_chunk); |
4db6bfe0 AK |
474 | if (r) |
475 | return r; | |
476 | } | |
477 | ||
478 | return 0; | |
479 | } | |
480 | ||
a159c1ac JB |
481 | static int read_exceptions(struct pstore *ps, |
482 | int (*callback)(void *callback_context, chunk_t old, | |
483 | chunk_t new), | |
484 | void *callback_context) | |
4db6bfe0 AK |
485 | { |
486 | int r, full = 1; | |
487 | ||
488 | /* | |
489 | * Keeping reading chunks and inserting exceptions until | |
490 | * we find a partially full area. | |
491 | */ | |
492 | for (ps->current_area = 0; full; ps->current_area++) { | |
493 | r = area_io(ps, READ); | |
494 | if (r) | |
495 | return r; | |
496 | ||
a159c1ac | 497 | r = insert_exceptions(ps, callback, callback_context, &full); |
4db6bfe0 AK |
498 | if (r) |
499 | return r; | |
500 | } | |
501 | ||
502 | ps->current_area--; | |
503 | ||
504 | return 0; | |
505 | } | |
506 | ||
507 | static struct pstore *get_info(struct dm_exception_store *store) | |
508 | { | |
509 | return (struct pstore *) store->context; | |
510 | } | |
511 | ||
985903bb MS |
512 | static void persistent_usage(struct dm_exception_store *store, |
513 | sector_t *total_sectors, | |
514 | sector_t *sectors_allocated, | |
515 | sector_t *metadata_sectors) | |
4db6bfe0 | 516 | { |
985903bb MS |
517 | struct pstore *ps = get_info(store); |
518 | ||
519 | *sectors_allocated = ps->next_free * store->chunk_size; | |
fc56f6fb | 520 | *total_sectors = get_dev_size(dm_snap_cow(store->snap)->bdev); |
985903bb MS |
521 | |
522 | /* | |
523 | * First chunk is the fixed header. | |
524 | * Then there are (ps->current_area + 1) metadata chunks, each one | |
525 | * separated from the next by ps->exceptions_per_area data chunks. | |
526 | */ | |
4454a621 MP |
527 | *metadata_sectors = (ps->current_area + 1 + NUM_SNAPSHOT_HDR_CHUNKS) * |
528 | store->chunk_size; | |
4db6bfe0 AK |
529 | } |
530 | ||
493df71c | 531 | static void persistent_dtr(struct dm_exception_store *store) |
4db6bfe0 AK |
532 | { |
533 | struct pstore *ps = get_info(store); | |
534 | ||
535 | destroy_workqueue(ps->metadata_wq); | |
a32079ce JB |
536 | |
537 | /* Created in read_header */ | |
538 | if (ps->io_client) | |
539 | dm_io_client_destroy(ps->io_client); | |
4db6bfe0 | 540 | free_area(ps); |
a32079ce JB |
541 | |
542 | /* Allocated in persistent_read_metadata */ | |
543 | if (ps->callbacks) | |
544 | vfree(ps->callbacks); | |
545 | ||
4db6bfe0 AK |
546 | kfree(ps); |
547 | } | |
548 | ||
a159c1ac JB |
549 | static int persistent_read_metadata(struct dm_exception_store *store, |
550 | int (*callback)(void *callback_context, | |
551 | chunk_t old, chunk_t new), | |
552 | void *callback_context) | |
4db6bfe0 AK |
553 | { |
554 | int r, uninitialized_var(new_snapshot); | |
555 | struct pstore *ps = get_info(store); | |
556 | ||
557 | /* | |
558 | * Read the snapshot header. | |
559 | */ | |
560 | r = read_header(ps, &new_snapshot); | |
561 | if (r) | |
562 | return r; | |
563 | ||
564 | /* | |
565 | * Now we know correct chunk_size, complete the initialisation. | |
566 | */ | |
71fab00a JB |
567 | ps->exceptions_per_area = (ps->store->chunk_size << SECTOR_SHIFT) / |
568 | sizeof(struct disk_exception); | |
4db6bfe0 AK |
569 | ps->callbacks = dm_vcalloc(ps->exceptions_per_area, |
570 | sizeof(*ps->callbacks)); | |
571 | if (!ps->callbacks) | |
572 | return -ENOMEM; | |
573 | ||
574 | /* | |
575 | * Do we need to setup a new snapshot ? | |
576 | */ | |
577 | if (new_snapshot) { | |
578 | r = write_header(ps); | |
579 | if (r) { | |
580 | DMWARN("write_header failed"); | |
581 | return r; | |
582 | } | |
583 | ||
584 | ps->current_area = 0; | |
585 | zero_memory_area(ps); | |
586 | r = zero_disk_area(ps, 0); | |
f5acc834 | 587 | if (r) |
4db6bfe0 | 588 | DMWARN("zero_disk_area(0) failed"); |
f5acc834 JB |
589 | return r; |
590 | } | |
591 | /* | |
592 | * Sanity checks. | |
593 | */ | |
594 | if (ps->version != SNAPSHOT_DISK_VERSION) { | |
595 | DMWARN("unable to handle snapshot disk version %d", | |
596 | ps->version); | |
597 | return -EINVAL; | |
598 | } | |
4db6bfe0 | 599 | |
f5acc834 JB |
600 | /* |
601 | * Metadata are valid, but snapshot is invalidated | |
602 | */ | |
603 | if (!ps->valid) | |
604 | return 1; | |
4db6bfe0 | 605 | |
f5acc834 JB |
606 | /* |
607 | * Read the metadata. | |
608 | */ | |
609 | r = read_exceptions(ps, callback, callback_context); | |
4db6bfe0 | 610 | |
f5acc834 | 611 | return r; |
4db6bfe0 AK |
612 | } |
613 | ||
a159c1ac | 614 | static int persistent_prepare_exception(struct dm_exception_store *store, |
1d4989c8 | 615 | struct dm_exception *e) |
4db6bfe0 AK |
616 | { |
617 | struct pstore *ps = get_info(store); | |
618 | uint32_t stride; | |
619 | chunk_t next_free; | |
fc56f6fb | 620 | sector_t size = get_dev_size(dm_snap_cow(store->snap)->bdev); |
4db6bfe0 AK |
621 | |
622 | /* Is there enough room ? */ | |
d0216849 | 623 | if (size < ((ps->next_free + 1) * store->chunk_size)) |
4db6bfe0 AK |
624 | return -ENOSPC; |
625 | ||
626 | e->new_chunk = ps->next_free; | |
627 | ||
628 | /* | |
629 | * Move onto the next free pending, making sure to take | |
630 | * into account the location of the metadata chunks. | |
631 | */ | |
632 | stride = (ps->exceptions_per_area + 1); | |
633 | next_free = ++ps->next_free; | |
634 | if (sector_div(next_free, stride) == 1) | |
635 | ps->next_free++; | |
636 | ||
637 | atomic_inc(&ps->pending_count); | |
638 | return 0; | |
639 | } | |
640 | ||
a159c1ac | 641 | static void persistent_commit_exception(struct dm_exception_store *store, |
1d4989c8 | 642 | struct dm_exception *e, |
a159c1ac JB |
643 | void (*callback) (void *, int success), |
644 | void *callback_context) | |
4db6bfe0 AK |
645 | { |
646 | unsigned int i; | |
647 | struct pstore *ps = get_info(store); | |
283a8328 | 648 | struct core_exception ce; |
4db6bfe0 AK |
649 | struct commit_callback *cb; |
650 | ||
283a8328 AK |
651 | ce.old_chunk = e->old_chunk; |
652 | ce.new_chunk = e->new_chunk; | |
653 | write_exception(ps, ps->current_committed++, &ce); | |
4db6bfe0 AK |
654 | |
655 | /* | |
656 | * Add the callback to the back of the array. This code | |
657 | * is the only place where the callback array is | |
658 | * manipulated, and we know that it will never be called | |
659 | * multiple times concurrently. | |
660 | */ | |
661 | cb = ps->callbacks + ps->callback_count++; | |
662 | cb->callback = callback; | |
663 | cb->context = callback_context; | |
664 | ||
665 | /* | |
666 | * If there are exceptions in flight and we have not yet | |
667 | * filled this metadata area there's nothing more to do. | |
668 | */ | |
669 | if (!atomic_dec_and_test(&ps->pending_count) && | |
670 | (ps->current_committed != ps->exceptions_per_area)) | |
671 | return; | |
672 | ||
673 | /* | |
674 | * If we completely filled the current area, then wipe the next one. | |
675 | */ | |
676 | if ((ps->current_committed == ps->exceptions_per_area) && | |
677 | zero_disk_area(ps, ps->current_area + 1)) | |
678 | ps->valid = 0; | |
679 | ||
680 | /* | |
681 | * Commit exceptions to disk. | |
682 | */ | |
d87f4c14 | 683 | if (ps->valid && area_io(ps, WRITE_FLUSH_FUA)) |
4db6bfe0 AK |
684 | ps->valid = 0; |
685 | ||
686 | /* | |
687 | * Advance to the next area if this one is full. | |
688 | */ | |
689 | if (ps->current_committed == ps->exceptions_per_area) { | |
690 | ps->current_committed = 0; | |
691 | ps->current_area++; | |
692 | zero_memory_area(ps); | |
693 | } | |
694 | ||
695 | for (i = 0; i < ps->callback_count; i++) { | |
696 | cb = ps->callbacks + i; | |
697 | cb->callback(cb->context, ps->valid); | |
698 | } | |
699 | ||
700 | ps->callback_count = 0; | |
701 | } | |
702 | ||
4454a621 MP |
703 | static int persistent_prepare_merge(struct dm_exception_store *store, |
704 | chunk_t *last_old_chunk, | |
705 | chunk_t *last_new_chunk) | |
706 | { | |
707 | struct pstore *ps = get_info(store); | |
283a8328 | 708 | struct core_exception ce; |
4454a621 MP |
709 | int nr_consecutive; |
710 | int r; | |
711 | ||
712 | /* | |
713 | * When current area is empty, move back to preceding area. | |
714 | */ | |
715 | if (!ps->current_committed) { | |
716 | /* | |
717 | * Have we finished? | |
718 | */ | |
719 | if (!ps->current_area) | |
720 | return 0; | |
721 | ||
722 | ps->current_area--; | |
723 | r = area_io(ps, READ); | |
724 | if (r < 0) | |
725 | return r; | |
726 | ps->current_committed = ps->exceptions_per_area; | |
727 | } | |
728 | ||
283a8328 AK |
729 | read_exception(ps, ps->current_committed - 1, &ce); |
730 | *last_old_chunk = ce.old_chunk; | |
731 | *last_new_chunk = ce.new_chunk; | |
4454a621 MP |
732 | |
733 | /* | |
734 | * Find number of consecutive chunks within the current area, | |
735 | * working backwards. | |
736 | */ | |
737 | for (nr_consecutive = 1; nr_consecutive < ps->current_committed; | |
738 | nr_consecutive++) { | |
739 | read_exception(ps, ps->current_committed - 1 - nr_consecutive, | |
283a8328 AK |
740 | &ce); |
741 | if (ce.old_chunk != *last_old_chunk - nr_consecutive || | |
742 | ce.new_chunk != *last_new_chunk - nr_consecutive) | |
4454a621 MP |
743 | break; |
744 | } | |
745 | ||
746 | return nr_consecutive; | |
747 | } | |
748 | ||
749 | static int persistent_commit_merge(struct dm_exception_store *store, | |
750 | int nr_merged) | |
751 | { | |
752 | int r, i; | |
753 | struct pstore *ps = get_info(store); | |
754 | ||
755 | BUG_ON(nr_merged > ps->current_committed); | |
756 | ||
757 | for (i = 0; i < nr_merged; i++) | |
758 | clear_exception(ps, ps->current_committed - 1 - i); | |
759 | ||
762a80d9 | 760 | r = area_io(ps, WRITE_FLUSH_FUA); |
4454a621 MP |
761 | if (r < 0) |
762 | return r; | |
763 | ||
764 | ps->current_committed -= nr_merged; | |
765 | ||
766 | /* | |
767 | * At this stage, only persistent_usage() uses ps->next_free, so | |
768 | * we make no attempt to keep ps->next_free strictly accurate | |
769 | * as exceptions may have been committed out-of-order originally. | |
770 | * Once a snapshot has become merging, we set it to the value it | |
771 | * would have held had all the exceptions been committed in order. | |
772 | * | |
773 | * ps->current_area does not get reduced by prepare_merge() until | |
774 | * after commit_merge() has removed the nr_merged previous exceptions. | |
775 | */ | |
87c961cb TK |
776 | ps->next_free = area_location(ps, ps->current_area) + |
777 | ps->current_committed + 1; | |
4454a621 MP |
778 | |
779 | return 0; | |
780 | } | |
781 | ||
a159c1ac | 782 | static void persistent_drop_snapshot(struct dm_exception_store *store) |
4db6bfe0 AK |
783 | { |
784 | struct pstore *ps = get_info(store); | |
785 | ||
786 | ps->valid = 0; | |
787 | if (write_header(ps)) | |
788 | DMWARN("write header failed"); | |
789 | } | |
790 | ||
493df71c JB |
791 | static int persistent_ctr(struct dm_exception_store *store, |
792 | unsigned argc, char **argv) | |
4db6bfe0 AK |
793 | { |
794 | struct pstore *ps; | |
795 | ||
796 | /* allocate the pstore */ | |
a32079ce | 797 | ps = kzalloc(sizeof(*ps), GFP_KERNEL); |
4db6bfe0 AK |
798 | if (!ps) |
799 | return -ENOMEM; | |
800 | ||
71fab00a | 801 | ps->store = store; |
4db6bfe0 AK |
802 | ps->valid = 1; |
803 | ps->version = SNAPSHOT_DISK_VERSION; | |
804 | ps->area = NULL; | |
61578dcd MP |
805 | ps->zero_area = NULL; |
806 | ps->header_area = NULL; | |
4454a621 | 807 | ps->next_free = NUM_SNAPSHOT_HDR_CHUNKS + 1; /* header and 1st area */ |
4db6bfe0 AK |
808 | ps->current_committed = 0; |
809 | ||
810 | ps->callback_count = 0; | |
811 | atomic_set(&ps->pending_count, 0); | |
812 | ps->callbacks = NULL; | |
813 | ||
239c8dd5 | 814 | ps->metadata_wq = alloc_workqueue("ksnaphd", WQ_MEM_RECLAIM, 0); |
4db6bfe0 AK |
815 | if (!ps->metadata_wq) { |
816 | kfree(ps); | |
817 | DMERR("couldn't start header metadata update thread"); | |
818 | return -ENOMEM; | |
819 | } | |
820 | ||
4db6bfe0 AK |
821 | store->context = ps; |
822 | ||
823 | return 0; | |
824 | } | |
825 | ||
1e302a92 JB |
826 | static unsigned persistent_status(struct dm_exception_store *store, |
827 | status_type_t status, char *result, | |
828 | unsigned maxlen) | |
493df71c | 829 | { |
1e302a92 JB |
830 | unsigned sz = 0; |
831 | ||
832 | switch (status) { | |
833 | case STATUSTYPE_INFO: | |
834 | break; | |
835 | case STATUSTYPE_TABLE: | |
fc56f6fb | 836 | DMEMIT(" P %llu", (unsigned long long)store->chunk_size); |
1e302a92 | 837 | } |
493df71c JB |
838 | |
839 | return sz; | |
840 | } | |
841 | ||
842 | static struct dm_exception_store_type _persistent_type = { | |
843 | .name = "persistent", | |
844 | .module = THIS_MODULE, | |
845 | .ctr = persistent_ctr, | |
846 | .dtr = persistent_dtr, | |
847 | .read_metadata = persistent_read_metadata, | |
848 | .prepare_exception = persistent_prepare_exception, | |
849 | .commit_exception = persistent_commit_exception, | |
4454a621 MP |
850 | .prepare_merge = persistent_prepare_merge, |
851 | .commit_merge = persistent_commit_merge, | |
493df71c | 852 | .drop_snapshot = persistent_drop_snapshot, |
985903bb | 853 | .usage = persistent_usage, |
493df71c JB |
854 | .status = persistent_status, |
855 | }; | |
856 | ||
857 | static struct dm_exception_store_type _persistent_compat_type = { | |
858 | .name = "P", | |
859 | .module = THIS_MODULE, | |
860 | .ctr = persistent_ctr, | |
861 | .dtr = persistent_dtr, | |
862 | .read_metadata = persistent_read_metadata, | |
863 | .prepare_exception = persistent_prepare_exception, | |
864 | .commit_exception = persistent_commit_exception, | |
4454a621 MP |
865 | .prepare_merge = persistent_prepare_merge, |
866 | .commit_merge = persistent_commit_merge, | |
493df71c | 867 | .drop_snapshot = persistent_drop_snapshot, |
985903bb | 868 | .usage = persistent_usage, |
493df71c JB |
869 | .status = persistent_status, |
870 | }; | |
871 | ||
4db6bfe0 AK |
872 | int dm_persistent_snapshot_init(void) |
873 | { | |
493df71c JB |
874 | int r; |
875 | ||
876 | r = dm_exception_store_type_register(&_persistent_type); | |
877 | if (r) { | |
878 | DMERR("Unable to register persistent exception store type"); | |
879 | return r; | |
880 | } | |
881 | ||
882 | r = dm_exception_store_type_register(&_persistent_compat_type); | |
883 | if (r) { | |
884 | DMERR("Unable to register old-style persistent exception " | |
885 | "store type"); | |
886 | dm_exception_store_type_unregister(&_persistent_type); | |
887 | return r; | |
888 | } | |
889 | ||
890 | return r; | |
4db6bfe0 AK |
891 | } |
892 | ||
893 | void dm_persistent_snapshot_exit(void) | |
894 | { | |
493df71c JB |
895 | dm_exception_store_type_unregister(&_persistent_type); |
896 | dm_exception_store_type_unregister(&_persistent_compat_type); | |
4db6bfe0 | 897 | } |