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> | |
daaa5f7c | 13 | #include <linux/export.h> |
4db6bfe0 AK |
14 | #include <linux/slab.h> |
15 | #include <linux/dm-io.h> | |
16 | ||
17 | #define DM_MSG_PREFIX "persistent snapshot" | |
18 | #define DM_CHUNK_SIZE_DEFAULT_SECTORS 32 /* 16KB */ | |
19 | ||
20 | /*----------------------------------------------------------------- | |
21 | * Persistent snapshots, by persistent we mean that the snapshot | |
22 | * will survive a reboot. | |
23 | *---------------------------------------------------------------*/ | |
24 | ||
25 | /* | |
26 | * We need to store a record of which parts of the origin have | |
27 | * been copied to the snapshot device. The snapshot code | |
28 | * requires that we copy exception chunks to chunk aligned areas | |
29 | * of the COW store. It makes sense therefore, to store the | |
30 | * metadata in chunk size blocks. | |
31 | * | |
32 | * There is no backward or forward compatibility implemented, | |
33 | * snapshots with different disk versions than the kernel will | |
34 | * not be usable. It is expected that "lvcreate" will blank out | |
35 | * the start of a fresh COW device before calling the snapshot | |
36 | * constructor. | |
37 | * | |
38 | * The first chunk of the COW device just contains the header. | |
39 | * After this there is a chunk filled with exception metadata, | |
40 | * followed by as many exception chunks as can fit in the | |
41 | * metadata areas. | |
42 | * | |
43 | * All on disk structures are in little-endian format. The end | |
44 | * of the exceptions info is indicated by an exception with a | |
45 | * new_chunk of 0, which is invalid since it would point to the | |
46 | * header chunk. | |
47 | */ | |
48 | ||
49 | /* | |
50 | * Magic for persistent snapshots: "SnAp" - Feeble isn't it. | |
51 | */ | |
52 | #define SNAP_MAGIC 0x70416e53 | |
53 | ||
54 | /* | |
55 | * The on-disk version of the metadata. | |
56 | */ | |
57 | #define SNAPSHOT_DISK_VERSION 1 | |
58 | ||
4454a621 MP |
59 | #define NUM_SNAPSHOT_HDR_CHUNKS 1 |
60 | ||
4db6bfe0 | 61 | struct disk_header { |
283a8328 | 62 | __le32 magic; |
4db6bfe0 AK |
63 | |
64 | /* | |
65 | * Is this snapshot valid. There is no way of recovering | |
66 | * an invalid snapshot. | |
67 | */ | |
283a8328 | 68 | __le32 valid; |
4db6bfe0 AK |
69 | |
70 | /* | |
71 | * Simple, incrementing version. no backward | |
72 | * compatibility. | |
73 | */ | |
283a8328 | 74 | __le32 version; |
4db6bfe0 AK |
75 | |
76 | /* In sectors */ | |
283a8328 AK |
77 | __le32 chunk_size; |
78 | } __packed; | |
4db6bfe0 AK |
79 | |
80 | struct disk_exception { | |
283a8328 AK |
81 | __le64 old_chunk; |
82 | __le64 new_chunk; | |
83 | } __packed; | |
84 | ||
85 | struct core_exception { | |
4db6bfe0 AK |
86 | uint64_t old_chunk; |
87 | uint64_t new_chunk; | |
88 | }; | |
89 | ||
90 | struct commit_callback { | |
91 | void (*callback)(void *, int success); | |
92 | void *context; | |
93 | }; | |
94 | ||
95 | /* | |
96 | * The top level structure for a persistent exception store. | |
97 | */ | |
98 | struct pstore { | |
71fab00a | 99 | struct dm_exception_store *store; |
4db6bfe0 AK |
100 | int version; |
101 | int valid; | |
102 | uint32_t exceptions_per_area; | |
103 | ||
104 | /* | |
105 | * Now that we have an asynchronous kcopyd there is no | |
106 | * need for large chunk sizes, so it wont hurt to have a | |
107 | * whole chunks worth of metadata in memory at once. | |
108 | */ | |
109 | void *area; | |
110 | ||
111 | /* | |
112 | * An area of zeros used to clear the next area. | |
113 | */ | |
114 | void *zero_area; | |
115 | ||
61578dcd MP |
116 | /* |
117 | * An area used for header. The header can be written | |
118 | * concurrently with metadata (when invalidating the snapshot), | |
119 | * so it needs a separate buffer. | |
120 | */ | |
121 | void *header_area; | |
122 | ||
4db6bfe0 AK |
123 | /* |
124 | * Used to keep track of which metadata area the data in | |
125 | * 'chunk' refers to. | |
126 | */ | |
127 | chunk_t current_area; | |
128 | ||
129 | /* | |
130 | * The next free chunk for an exception. | |
4454a621 MP |
131 | * |
132 | * When creating exceptions, all the chunks here and above are | |
133 | * free. It holds the next chunk to be allocated. On rare | |
134 | * occasions (e.g. after a system crash) holes can be left in | |
135 | * the exception store because chunks can be committed out of | |
136 | * order. | |
137 | * | |
138 | * When merging exceptions, it does not necessarily mean all the | |
139 | * chunks here and above are free. It holds the value it would | |
140 | * have held if all chunks had been committed in order of | |
141 | * allocation. Consequently the value may occasionally be | |
142 | * slightly too low, but since it's only used for 'status' and | |
143 | * it can never reach its minimum value too early this doesn't | |
144 | * matter. | |
4db6bfe0 | 145 | */ |
4454a621 | 146 | |
4db6bfe0 AK |
147 | chunk_t next_free; |
148 | ||
149 | /* | |
150 | * The index of next free exception in the current | |
151 | * metadata area. | |
152 | */ | |
153 | uint32_t current_committed; | |
154 | ||
155 | atomic_t pending_count; | |
156 | uint32_t callback_count; | |
157 | struct commit_callback *callbacks; | |
158 | struct dm_io_client *io_client; | |
159 | ||
160 | struct workqueue_struct *metadata_wq; | |
161 | }; | |
162 | ||
4db6bfe0 AK |
163 | static int alloc_area(struct pstore *ps) |
164 | { | |
165 | int r = -ENOMEM; | |
166 | size_t len; | |
167 | ||
71fab00a | 168 | len = ps->store->chunk_size << SECTOR_SHIFT; |
4db6bfe0 AK |
169 | |
170 | /* | |
171 | * Allocate the chunk_size block of memory that will hold | |
172 | * a single metadata area. | |
173 | */ | |
174 | ps->area = vmalloc(len); | |
175 | if (!ps->area) | |
61578dcd | 176 | goto err_area; |
4db6bfe0 | 177 | |
e29e65aa | 178 | ps->zero_area = vzalloc(len); |
61578dcd MP |
179 | if (!ps->zero_area) |
180 | goto err_zero_area; | |
4db6bfe0 | 181 | |
61578dcd MP |
182 | ps->header_area = vmalloc(len); |
183 | if (!ps->header_area) | |
184 | goto err_header_area; | |
185 | ||
4db6bfe0 | 186 | return 0; |
61578dcd MP |
187 | |
188 | err_header_area: | |
189 | vfree(ps->zero_area); | |
190 | ||
191 | err_zero_area: | |
192 | vfree(ps->area); | |
193 | ||
194 | err_area: | |
195 | return r; | |
4db6bfe0 AK |
196 | } |
197 | ||
198 | static void free_area(struct pstore *ps) | |
199 | { | |
a32079ce JB |
200 | if (ps->area) |
201 | vfree(ps->area); | |
4db6bfe0 | 202 | ps->area = NULL; |
a32079ce JB |
203 | |
204 | if (ps->zero_area) | |
205 | vfree(ps->zero_area); | |
4db6bfe0 | 206 | ps->zero_area = NULL; |
61578dcd MP |
207 | |
208 | if (ps->header_area) | |
209 | vfree(ps->header_area); | |
210 | ps->header_area = NULL; | |
4db6bfe0 AK |
211 | } |
212 | ||
213 | struct mdata_req { | |
214 | struct dm_io_region *where; | |
215 | struct dm_io_request *io_req; | |
216 | struct work_struct work; | |
217 | int result; | |
218 | }; | |
219 | ||
220 | static void do_metadata(struct work_struct *work) | |
221 | { | |
222 | struct mdata_req *req = container_of(work, struct mdata_req, work); | |
223 | ||
224 | req->result = dm_io(req->io_req, 1, req->where, NULL); | |
225 | } | |
226 | ||
227 | /* | |
228 | * Read or write a chunk aligned and sized block of data from a device. | |
229 | */ | |
02d2fd31 MP |
230 | static int chunk_io(struct pstore *ps, void *area, chunk_t chunk, int rw, |
231 | int metadata) | |
4db6bfe0 AK |
232 | { |
233 | struct dm_io_region where = { | |
fc56f6fb | 234 | .bdev = dm_snap_cow(ps->store->snap)->bdev, |
71fab00a JB |
235 | .sector = ps->store->chunk_size * chunk, |
236 | .count = ps->store->chunk_size, | |
4db6bfe0 AK |
237 | }; |
238 | struct dm_io_request io_req = { | |
239 | .bi_rw = rw, | |
240 | .mem.type = DM_IO_VMA, | |
02d2fd31 | 241 | .mem.ptr.vma = area, |
4db6bfe0 AK |
242 | .client = ps->io_client, |
243 | .notify.fn = NULL, | |
244 | }; | |
245 | struct mdata_req req; | |
246 | ||
247 | if (!metadata) | |
248 | return dm_io(&io_req, 1, &where, NULL); | |
249 | ||
250 | req.where = &where; | |
251 | req.io_req = &io_req; | |
252 | ||
253 | /* | |
254 | * Issue the synchronous I/O from a different thread | |
255 | * to avoid generic_make_request recursion. | |
256 | */ | |
ca1cab37 | 257 | INIT_WORK_ONSTACK(&req.work, do_metadata); |
4db6bfe0 | 258 | queue_work(ps->metadata_wq, &req.work); |
4541f4e3 | 259 | flush_workqueue(ps->metadata_wq); |
4db6bfe0 AK |
260 | |
261 | return req.result; | |
262 | } | |
263 | ||
264 | /* | |
265 | * Convert a metadata area index to a chunk index. | |
266 | */ | |
267 | static chunk_t area_location(struct pstore *ps, chunk_t area) | |
268 | { | |
87c961cb | 269 | return NUM_SNAPSHOT_HDR_CHUNKS + ((ps->exceptions_per_area + 1) * area); |
4db6bfe0 AK |
270 | } |
271 | ||
2d99b6dd MP |
272 | static void skip_metadata(struct pstore *ps) |
273 | { | |
274 | uint32_t stride = ps->exceptions_per_area + 1; | |
275 | chunk_t next_free = ps->next_free; | |
276 | if (sector_div(next_free, stride) == NUM_SNAPSHOT_HDR_CHUNKS) | |
277 | ps->next_free++; | |
278 | } | |
279 | ||
4db6bfe0 AK |
280 | /* |
281 | * Read or write a metadata area. Remembering to skip the first | |
282 | * chunk which holds the header. | |
283 | */ | |
284 | static int area_io(struct pstore *ps, int rw) | |
285 | { | |
286 | int r; | |
287 | chunk_t chunk; | |
288 | ||
289 | chunk = area_location(ps, ps->current_area); | |
290 | ||
02d2fd31 | 291 | r = chunk_io(ps, ps->area, chunk, rw, 0); |
4db6bfe0 AK |
292 | if (r) |
293 | return r; | |
294 | ||
295 | return 0; | |
296 | } | |
297 | ||
298 | static void zero_memory_area(struct pstore *ps) | |
299 | { | |
71fab00a | 300 | memset(ps->area, 0, ps->store->chunk_size << SECTOR_SHIFT); |
4db6bfe0 AK |
301 | } |
302 | ||
303 | static int zero_disk_area(struct pstore *ps, chunk_t area) | |
304 | { | |
02d2fd31 | 305 | return chunk_io(ps, ps->zero_area, area_location(ps, area), WRITE, 0); |
4db6bfe0 AK |
306 | } |
307 | ||
308 | static int read_header(struct pstore *ps, int *new_snapshot) | |
309 | { | |
310 | int r; | |
311 | struct disk_header *dh; | |
df96eee6 | 312 | unsigned chunk_size; |
4db6bfe0 | 313 | int chunk_size_supplied = 1; |
ae0b7448 | 314 | char *chunk_err; |
4db6bfe0 AK |
315 | |
316 | /* | |
df96eee6 MP |
317 | * Use default chunk size (or logical_block_size, if larger) |
318 | * if none supplied | |
4db6bfe0 | 319 | */ |
71fab00a JB |
320 | if (!ps->store->chunk_size) { |
321 | ps->store->chunk_size = max(DM_CHUNK_SIZE_DEFAULT_SECTORS, | |
fc56f6fb MS |
322 | bdev_logical_block_size(dm_snap_cow(ps->store->snap)-> |
323 | bdev) >> 9); | |
71fab00a JB |
324 | ps->store->chunk_mask = ps->store->chunk_size - 1; |
325 | ps->store->chunk_shift = ffs(ps->store->chunk_size) - 1; | |
4db6bfe0 AK |
326 | chunk_size_supplied = 0; |
327 | } | |
328 | ||
bda8efec | 329 | ps->io_client = dm_io_client_create(); |
4db6bfe0 AK |
330 | if (IS_ERR(ps->io_client)) |
331 | return PTR_ERR(ps->io_client); | |
332 | ||
333 | r = alloc_area(ps); | |
334 | if (r) | |
335 | return r; | |
336 | ||
61578dcd | 337 | r = chunk_io(ps, ps->header_area, 0, READ, 1); |
4db6bfe0 AK |
338 | if (r) |
339 | goto bad; | |
340 | ||
61578dcd | 341 | dh = ps->header_area; |
4db6bfe0 AK |
342 | |
343 | if (le32_to_cpu(dh->magic) == 0) { | |
344 | *new_snapshot = 1; | |
345 | return 0; | |
346 | } | |
347 | ||
348 | if (le32_to_cpu(dh->magic) != SNAP_MAGIC) { | |
349 | DMWARN("Invalid or corrupt snapshot"); | |
350 | r = -ENXIO; | |
351 | goto bad; | |
352 | } | |
353 | ||
354 | *new_snapshot = 0; | |
355 | ps->valid = le32_to_cpu(dh->valid); | |
356 | ps->version = le32_to_cpu(dh->version); | |
357 | chunk_size = le32_to_cpu(dh->chunk_size); | |
358 | ||
ae0b7448 | 359 | if (ps->store->chunk_size == chunk_size) |
4db6bfe0 AK |
360 | return 0; |
361 | ||
ae0b7448 | 362 | if (chunk_size_supplied) |
df96eee6 MP |
363 | DMWARN("chunk size %u in device metadata overrides " |
364 | "table chunk size of %u.", | |
365 | chunk_size, ps->store->chunk_size); | |
4db6bfe0 AK |
366 | |
367 | /* We had a bogus chunk_size. Fix stuff up. */ | |
368 | free_area(ps); | |
369 | ||
ae0b7448 MP |
370 | r = dm_exception_store_set_chunk_size(ps->store, chunk_size, |
371 | &chunk_err); | |
372 | if (r) { | |
df96eee6 MP |
373 | DMERR("invalid on-disk chunk size %u: %s.", |
374 | chunk_size, chunk_err); | |
ae0b7448 MP |
375 | return r; |
376 | } | |
4db6bfe0 | 377 | |
4db6bfe0 AK |
378 | r = alloc_area(ps); |
379 | return r; | |
380 | ||
381 | bad: | |
382 | free_area(ps); | |
383 | return r; | |
384 | } | |
385 | ||
386 | static int write_header(struct pstore *ps) | |
387 | { | |
388 | struct disk_header *dh; | |
389 | ||
61578dcd | 390 | memset(ps->header_area, 0, ps->store->chunk_size << SECTOR_SHIFT); |
4db6bfe0 | 391 | |
61578dcd | 392 | dh = ps->header_area; |
4db6bfe0 AK |
393 | dh->magic = cpu_to_le32(SNAP_MAGIC); |
394 | dh->valid = cpu_to_le32(ps->valid); | |
395 | dh->version = cpu_to_le32(ps->version); | |
71fab00a | 396 | dh->chunk_size = cpu_to_le32(ps->store->chunk_size); |
4db6bfe0 | 397 | |
61578dcd | 398 | return chunk_io(ps, ps->header_area, 0, WRITE, 1); |
4db6bfe0 AK |
399 | } |
400 | ||
401 | /* | |
402 | * Access functions for the disk exceptions, these do the endian conversions. | |
403 | */ | |
404 | static struct disk_exception *get_exception(struct pstore *ps, uint32_t index) | |
405 | { | |
406 | BUG_ON(index >= ps->exceptions_per_area); | |
407 | ||
408 | return ((struct disk_exception *) ps->area) + index; | |
409 | } | |
410 | ||
411 | static void read_exception(struct pstore *ps, | |
283a8328 | 412 | uint32_t index, struct core_exception *result) |
4db6bfe0 | 413 | { |
283a8328 | 414 | struct disk_exception *de = get_exception(ps, index); |
4db6bfe0 AK |
415 | |
416 | /* copy it */ | |
283a8328 AK |
417 | result->old_chunk = le64_to_cpu(de->old_chunk); |
418 | result->new_chunk = le64_to_cpu(de->new_chunk); | |
4db6bfe0 AK |
419 | } |
420 | ||
421 | static void write_exception(struct pstore *ps, | |
283a8328 | 422 | uint32_t index, struct core_exception *e) |
4db6bfe0 | 423 | { |
283a8328 | 424 | struct disk_exception *de = get_exception(ps, index); |
4db6bfe0 AK |
425 | |
426 | /* copy it */ | |
283a8328 AK |
427 | de->old_chunk = cpu_to_le64(e->old_chunk); |
428 | de->new_chunk = cpu_to_le64(e->new_chunk); | |
4db6bfe0 AK |
429 | } |
430 | ||
4454a621 MP |
431 | static void clear_exception(struct pstore *ps, uint32_t index) |
432 | { | |
283a8328 | 433 | struct disk_exception *de = get_exception(ps, index); |
4454a621 MP |
434 | |
435 | /* clear it */ | |
283a8328 AK |
436 | de->old_chunk = 0; |
437 | de->new_chunk = 0; | |
4454a621 MP |
438 | } |
439 | ||
4db6bfe0 AK |
440 | /* |
441 | * Registers the exceptions that are present in the current area. | |
442 | * 'full' is filled in to indicate if the area has been | |
443 | * filled. | |
444 | */ | |
a159c1ac JB |
445 | static int insert_exceptions(struct pstore *ps, |
446 | int (*callback)(void *callback_context, | |
447 | chunk_t old, chunk_t new), | |
448 | void *callback_context, | |
449 | int *full) | |
4db6bfe0 AK |
450 | { |
451 | int r; | |
452 | unsigned int i; | |
283a8328 | 453 | struct core_exception e; |
4db6bfe0 AK |
454 | |
455 | /* presume the area is full */ | |
456 | *full = 1; | |
457 | ||
458 | for (i = 0; i < ps->exceptions_per_area; i++) { | |
283a8328 | 459 | read_exception(ps, i, &e); |
4db6bfe0 AK |
460 | |
461 | /* | |
462 | * If the new_chunk is pointing at the start of | |
463 | * the COW device, where the first metadata area | |
464 | * is we know that we've hit the end of the | |
465 | * exceptions. Therefore the area is not full. | |
466 | */ | |
283a8328 | 467 | if (e.new_chunk == 0LL) { |
4db6bfe0 AK |
468 | ps->current_committed = i; |
469 | *full = 0; | |
470 | break; | |
471 | } | |
472 | ||
473 | /* | |
474 | * Keep track of the start of the free chunks. | |
475 | */ | |
283a8328 AK |
476 | if (ps->next_free <= e.new_chunk) |
477 | ps->next_free = e.new_chunk + 1; | |
4db6bfe0 AK |
478 | |
479 | /* | |
480 | * Otherwise we add the exception to the snapshot. | |
481 | */ | |
283a8328 | 482 | r = callback(callback_context, e.old_chunk, e.new_chunk); |
4db6bfe0 AK |
483 | if (r) |
484 | return r; | |
485 | } | |
486 | ||
487 | return 0; | |
488 | } | |
489 | ||
a159c1ac JB |
490 | static int read_exceptions(struct pstore *ps, |
491 | int (*callback)(void *callback_context, chunk_t old, | |
492 | chunk_t new), | |
493 | void *callback_context) | |
4db6bfe0 AK |
494 | { |
495 | int r, full = 1; | |
496 | ||
497 | /* | |
498 | * Keeping reading chunks and inserting exceptions until | |
499 | * we find a partially full area. | |
500 | */ | |
501 | for (ps->current_area = 0; full; ps->current_area++) { | |
502 | r = area_io(ps, READ); | |
503 | if (r) | |
504 | return r; | |
505 | ||
a159c1ac | 506 | r = insert_exceptions(ps, callback, callback_context, &full); |
4db6bfe0 AK |
507 | if (r) |
508 | return r; | |
509 | } | |
510 | ||
511 | ps->current_area--; | |
512 | ||
2d99b6dd MP |
513 | skip_metadata(ps); |
514 | ||
4db6bfe0 AK |
515 | return 0; |
516 | } | |
517 | ||
518 | static struct pstore *get_info(struct dm_exception_store *store) | |
519 | { | |
520 | return (struct pstore *) store->context; | |
521 | } | |
522 | ||
985903bb MS |
523 | static void persistent_usage(struct dm_exception_store *store, |
524 | sector_t *total_sectors, | |
525 | sector_t *sectors_allocated, | |
526 | sector_t *metadata_sectors) | |
4db6bfe0 | 527 | { |
985903bb MS |
528 | struct pstore *ps = get_info(store); |
529 | ||
530 | *sectors_allocated = ps->next_free * store->chunk_size; | |
fc56f6fb | 531 | *total_sectors = get_dev_size(dm_snap_cow(store->snap)->bdev); |
985903bb MS |
532 | |
533 | /* | |
534 | * First chunk is the fixed header. | |
535 | * Then there are (ps->current_area + 1) metadata chunks, each one | |
536 | * separated from the next by ps->exceptions_per_area data chunks. | |
537 | */ | |
4454a621 MP |
538 | *metadata_sectors = (ps->current_area + 1 + NUM_SNAPSHOT_HDR_CHUNKS) * |
539 | store->chunk_size; | |
4db6bfe0 AK |
540 | } |
541 | ||
493df71c | 542 | static void persistent_dtr(struct dm_exception_store *store) |
4db6bfe0 AK |
543 | { |
544 | struct pstore *ps = get_info(store); | |
545 | ||
546 | destroy_workqueue(ps->metadata_wq); | |
a32079ce JB |
547 | |
548 | /* Created in read_header */ | |
549 | if (ps->io_client) | |
550 | dm_io_client_destroy(ps->io_client); | |
4db6bfe0 | 551 | free_area(ps); |
a32079ce JB |
552 | |
553 | /* Allocated in persistent_read_metadata */ | |
554 | if (ps->callbacks) | |
555 | vfree(ps->callbacks); | |
556 | ||
4db6bfe0 AK |
557 | kfree(ps); |
558 | } | |
559 | ||
a159c1ac JB |
560 | static int persistent_read_metadata(struct dm_exception_store *store, |
561 | int (*callback)(void *callback_context, | |
562 | chunk_t old, chunk_t new), | |
563 | void *callback_context) | |
4db6bfe0 AK |
564 | { |
565 | int r, uninitialized_var(new_snapshot); | |
566 | struct pstore *ps = get_info(store); | |
567 | ||
568 | /* | |
569 | * Read the snapshot header. | |
570 | */ | |
571 | r = read_header(ps, &new_snapshot); | |
572 | if (r) | |
573 | return r; | |
574 | ||
575 | /* | |
576 | * Now we know correct chunk_size, complete the initialisation. | |
577 | */ | |
71fab00a JB |
578 | ps->exceptions_per_area = (ps->store->chunk_size << SECTOR_SHIFT) / |
579 | sizeof(struct disk_exception); | |
4db6bfe0 | 580 | ps->callbacks = dm_vcalloc(ps->exceptions_per_area, |
a2d2b034 | 581 | sizeof(*ps->callbacks)); |
4db6bfe0 AK |
582 | if (!ps->callbacks) |
583 | return -ENOMEM; | |
584 | ||
585 | /* | |
586 | * Do we need to setup a new snapshot ? | |
587 | */ | |
588 | if (new_snapshot) { | |
589 | r = write_header(ps); | |
590 | if (r) { | |
591 | DMWARN("write_header failed"); | |
592 | return r; | |
593 | } | |
594 | ||
595 | ps->current_area = 0; | |
596 | zero_memory_area(ps); | |
597 | r = zero_disk_area(ps, 0); | |
f5acc834 | 598 | if (r) |
4db6bfe0 | 599 | DMWARN("zero_disk_area(0) failed"); |
f5acc834 JB |
600 | return r; |
601 | } | |
602 | /* | |
603 | * Sanity checks. | |
604 | */ | |
605 | if (ps->version != SNAPSHOT_DISK_VERSION) { | |
606 | DMWARN("unable to handle snapshot disk version %d", | |
607 | ps->version); | |
608 | return -EINVAL; | |
609 | } | |
4db6bfe0 | 610 | |
f5acc834 JB |
611 | /* |
612 | * Metadata are valid, but snapshot is invalidated | |
613 | */ | |
614 | if (!ps->valid) | |
615 | return 1; | |
4db6bfe0 | 616 | |
f5acc834 JB |
617 | /* |
618 | * Read the metadata. | |
619 | */ | |
620 | r = read_exceptions(ps, callback, callback_context); | |
4db6bfe0 | 621 | |
f5acc834 | 622 | return r; |
4db6bfe0 AK |
623 | } |
624 | ||
a159c1ac | 625 | static int persistent_prepare_exception(struct dm_exception_store *store, |
1d4989c8 | 626 | struct dm_exception *e) |
4db6bfe0 AK |
627 | { |
628 | struct pstore *ps = get_info(store); | |
fc56f6fb | 629 | sector_t size = get_dev_size(dm_snap_cow(store->snap)->bdev); |
4db6bfe0 AK |
630 | |
631 | /* Is there enough room ? */ | |
d0216849 | 632 | if (size < ((ps->next_free + 1) * store->chunk_size)) |
4db6bfe0 AK |
633 | return -ENOSPC; |
634 | ||
635 | e->new_chunk = ps->next_free; | |
636 | ||
637 | /* | |
638 | * Move onto the next free pending, making sure to take | |
639 | * into account the location of the metadata chunks. | |
640 | */ | |
2d99b6dd MP |
641 | ps->next_free++; |
642 | skip_metadata(ps); | |
4db6bfe0 AK |
643 | |
644 | atomic_inc(&ps->pending_count); | |
645 | return 0; | |
646 | } | |
647 | ||
a159c1ac | 648 | static void persistent_commit_exception(struct dm_exception_store *store, |
eeb7b0e0 | 649 | struct dm_exception *e, int valid, |
a159c1ac JB |
650 | void (*callback) (void *, int success), |
651 | void *callback_context) | |
4db6bfe0 AK |
652 | { |
653 | unsigned int i; | |
654 | struct pstore *ps = get_info(store); | |
283a8328 | 655 | struct core_exception ce; |
4db6bfe0 AK |
656 | struct commit_callback *cb; |
657 | ||
eeb7b0e0 MP |
658 | if (!valid) |
659 | ps->valid = 0; | |
660 | ||
283a8328 AK |
661 | ce.old_chunk = e->old_chunk; |
662 | ce.new_chunk = e->new_chunk; | |
663 | write_exception(ps, ps->current_committed++, &ce); | |
4db6bfe0 AK |
664 | |
665 | /* | |
666 | * Add the callback to the back of the array. This code | |
667 | * is the only place where the callback array is | |
668 | * manipulated, and we know that it will never be called | |
669 | * multiple times concurrently. | |
670 | */ | |
671 | cb = ps->callbacks + ps->callback_count++; | |
672 | cb->callback = callback; | |
673 | cb->context = callback_context; | |
674 | ||
675 | /* | |
676 | * If there are exceptions in flight and we have not yet | |
677 | * filled this metadata area there's nothing more to do. | |
678 | */ | |
679 | if (!atomic_dec_and_test(&ps->pending_count) && | |
680 | (ps->current_committed != ps->exceptions_per_area)) | |
681 | return; | |
682 | ||
683 | /* | |
684 | * If we completely filled the current area, then wipe the next one. | |
685 | */ | |
686 | if ((ps->current_committed == ps->exceptions_per_area) && | |
a2d2b034 | 687 | zero_disk_area(ps, ps->current_area + 1)) |
4db6bfe0 AK |
688 | ps->valid = 0; |
689 | ||
690 | /* | |
691 | * Commit exceptions to disk. | |
692 | */ | |
d87f4c14 | 693 | if (ps->valid && area_io(ps, WRITE_FLUSH_FUA)) |
4db6bfe0 AK |
694 | ps->valid = 0; |
695 | ||
696 | /* | |
697 | * Advance to the next area if this one is full. | |
698 | */ | |
699 | if (ps->current_committed == ps->exceptions_per_area) { | |
700 | ps->current_committed = 0; | |
701 | ps->current_area++; | |
702 | zero_memory_area(ps); | |
703 | } | |
704 | ||
705 | for (i = 0; i < ps->callback_count; i++) { | |
706 | cb = ps->callbacks + i; | |
707 | cb->callback(cb->context, ps->valid); | |
708 | } | |
709 | ||
710 | ps->callback_count = 0; | |
711 | } | |
712 | ||
4454a621 MP |
713 | static int persistent_prepare_merge(struct dm_exception_store *store, |
714 | chunk_t *last_old_chunk, | |
715 | chunk_t *last_new_chunk) | |
716 | { | |
717 | struct pstore *ps = get_info(store); | |
283a8328 | 718 | struct core_exception ce; |
4454a621 MP |
719 | int nr_consecutive; |
720 | int r; | |
721 | ||
722 | /* | |
723 | * When current area is empty, move back to preceding area. | |
724 | */ | |
725 | if (!ps->current_committed) { | |
726 | /* | |
727 | * Have we finished? | |
728 | */ | |
729 | if (!ps->current_area) | |
730 | return 0; | |
731 | ||
732 | ps->current_area--; | |
733 | r = area_io(ps, READ); | |
734 | if (r < 0) | |
735 | return r; | |
736 | ps->current_committed = ps->exceptions_per_area; | |
737 | } | |
738 | ||
283a8328 AK |
739 | read_exception(ps, ps->current_committed - 1, &ce); |
740 | *last_old_chunk = ce.old_chunk; | |
741 | *last_new_chunk = ce.new_chunk; | |
4454a621 MP |
742 | |
743 | /* | |
744 | * Find number of consecutive chunks within the current area, | |
745 | * working backwards. | |
746 | */ | |
747 | for (nr_consecutive = 1; nr_consecutive < ps->current_committed; | |
748 | nr_consecutive++) { | |
749 | read_exception(ps, ps->current_committed - 1 - nr_consecutive, | |
283a8328 AK |
750 | &ce); |
751 | if (ce.old_chunk != *last_old_chunk - nr_consecutive || | |
752 | ce.new_chunk != *last_new_chunk - nr_consecutive) | |
4454a621 MP |
753 | break; |
754 | } | |
755 | ||
756 | return nr_consecutive; | |
757 | } | |
758 | ||
759 | static int persistent_commit_merge(struct dm_exception_store *store, | |
760 | int nr_merged) | |
761 | { | |
762 | int r, i; | |
763 | struct pstore *ps = get_info(store); | |
764 | ||
765 | BUG_ON(nr_merged > ps->current_committed); | |
766 | ||
767 | for (i = 0; i < nr_merged; i++) | |
768 | clear_exception(ps, ps->current_committed - 1 - i); | |
769 | ||
762a80d9 | 770 | r = area_io(ps, WRITE_FLUSH_FUA); |
4454a621 MP |
771 | if (r < 0) |
772 | return r; | |
773 | ||
774 | ps->current_committed -= nr_merged; | |
775 | ||
776 | /* | |
777 | * At this stage, only persistent_usage() uses ps->next_free, so | |
778 | * we make no attempt to keep ps->next_free strictly accurate | |
779 | * as exceptions may have been committed out-of-order originally. | |
780 | * Once a snapshot has become merging, we set it to the value it | |
781 | * would have held had all the exceptions been committed in order. | |
782 | * | |
783 | * ps->current_area does not get reduced by prepare_merge() until | |
784 | * after commit_merge() has removed the nr_merged previous exceptions. | |
785 | */ | |
87c961cb TK |
786 | ps->next_free = area_location(ps, ps->current_area) + |
787 | ps->current_committed + 1; | |
4454a621 MP |
788 | |
789 | return 0; | |
790 | } | |
791 | ||
a159c1ac | 792 | static void persistent_drop_snapshot(struct dm_exception_store *store) |
4db6bfe0 AK |
793 | { |
794 | struct pstore *ps = get_info(store); | |
795 | ||
796 | ps->valid = 0; | |
797 | if (write_header(ps)) | |
798 | DMWARN("write header failed"); | |
799 | } | |
800 | ||
493df71c JB |
801 | static int persistent_ctr(struct dm_exception_store *store, |
802 | unsigned argc, char **argv) | |
4db6bfe0 AK |
803 | { |
804 | struct pstore *ps; | |
805 | ||
806 | /* allocate the pstore */ | |
a32079ce | 807 | ps = kzalloc(sizeof(*ps), GFP_KERNEL); |
4db6bfe0 AK |
808 | if (!ps) |
809 | return -ENOMEM; | |
810 | ||
71fab00a | 811 | ps->store = store; |
4db6bfe0 AK |
812 | ps->valid = 1; |
813 | ps->version = SNAPSHOT_DISK_VERSION; | |
814 | ps->area = NULL; | |
61578dcd MP |
815 | ps->zero_area = NULL; |
816 | ps->header_area = NULL; | |
4454a621 | 817 | ps->next_free = NUM_SNAPSHOT_HDR_CHUNKS + 1; /* header and 1st area */ |
4db6bfe0 AK |
818 | ps->current_committed = 0; |
819 | ||
820 | ps->callback_count = 0; | |
821 | atomic_set(&ps->pending_count, 0); | |
822 | ps->callbacks = NULL; | |
823 | ||
239c8dd5 | 824 | ps->metadata_wq = alloc_workqueue("ksnaphd", WQ_MEM_RECLAIM, 0); |
4db6bfe0 AK |
825 | if (!ps->metadata_wq) { |
826 | kfree(ps); | |
827 | DMERR("couldn't start header metadata update thread"); | |
828 | return -ENOMEM; | |
829 | } | |
830 | ||
4db6bfe0 AK |
831 | store->context = ps; |
832 | ||
833 | return 0; | |
834 | } | |
835 | ||
1e302a92 JB |
836 | static unsigned persistent_status(struct dm_exception_store *store, |
837 | status_type_t status, char *result, | |
838 | unsigned maxlen) | |
493df71c | 839 | { |
1e302a92 JB |
840 | unsigned sz = 0; |
841 | ||
842 | switch (status) { | |
843 | case STATUSTYPE_INFO: | |
844 | break; | |
845 | case STATUSTYPE_TABLE: | |
fc56f6fb | 846 | DMEMIT(" P %llu", (unsigned long long)store->chunk_size); |
1e302a92 | 847 | } |
493df71c JB |
848 | |
849 | return sz; | |
850 | } | |
851 | ||
852 | static struct dm_exception_store_type _persistent_type = { | |
853 | .name = "persistent", | |
854 | .module = THIS_MODULE, | |
855 | .ctr = persistent_ctr, | |
856 | .dtr = persistent_dtr, | |
857 | .read_metadata = persistent_read_metadata, | |
858 | .prepare_exception = persistent_prepare_exception, | |
859 | .commit_exception = persistent_commit_exception, | |
4454a621 MP |
860 | .prepare_merge = persistent_prepare_merge, |
861 | .commit_merge = persistent_commit_merge, | |
493df71c | 862 | .drop_snapshot = persistent_drop_snapshot, |
985903bb | 863 | .usage = persistent_usage, |
493df71c JB |
864 | .status = persistent_status, |
865 | }; | |
866 | ||
867 | static struct dm_exception_store_type _persistent_compat_type = { | |
868 | .name = "P", | |
869 | .module = THIS_MODULE, | |
870 | .ctr = persistent_ctr, | |
871 | .dtr = persistent_dtr, | |
872 | .read_metadata = persistent_read_metadata, | |
873 | .prepare_exception = persistent_prepare_exception, | |
874 | .commit_exception = persistent_commit_exception, | |
4454a621 MP |
875 | .prepare_merge = persistent_prepare_merge, |
876 | .commit_merge = persistent_commit_merge, | |
493df71c | 877 | .drop_snapshot = persistent_drop_snapshot, |
985903bb | 878 | .usage = persistent_usage, |
493df71c JB |
879 | .status = persistent_status, |
880 | }; | |
881 | ||
4db6bfe0 AK |
882 | int dm_persistent_snapshot_init(void) |
883 | { | |
493df71c JB |
884 | int r; |
885 | ||
886 | r = dm_exception_store_type_register(&_persistent_type); | |
887 | if (r) { | |
888 | DMERR("Unable to register persistent exception store type"); | |
889 | return r; | |
890 | } | |
891 | ||
892 | r = dm_exception_store_type_register(&_persistent_compat_type); | |
893 | if (r) { | |
894 | DMERR("Unable to register old-style persistent exception " | |
895 | "store type"); | |
896 | dm_exception_store_type_unregister(&_persistent_type); | |
897 | return r; | |
898 | } | |
899 | ||
900 | return r; | |
4db6bfe0 AK |
901 | } |
902 | ||
903 | void dm_persistent_snapshot_exit(void) | |
904 | { | |
493df71c JB |
905 | dm_exception_store_type_unregister(&_persistent_type); |
906 | dm_exception_store_type_unregister(&_persistent_compat_type); | |
4db6bfe0 | 907 | } |