]>
Commit | Line | Data |
---|---|---|
eec40579 JT |
1 | #include "dm.h" |
2 | #include "persistent-data/dm-transaction-manager.h" | |
3 | #include "persistent-data/dm-bitset.h" | |
4 | #include "persistent-data/dm-space-map.h" | |
5 | ||
6 | #include <linux/dm-io.h> | |
7 | #include <linux/dm-kcopyd.h> | |
8 | #include <linux/init.h> | |
9 | #include <linux/mempool.h> | |
10 | #include <linux/module.h> | |
11 | #include <linux/slab.h> | |
12 | #include <linux/vmalloc.h> | |
13 | ||
14 | #define DM_MSG_PREFIX "era" | |
15 | ||
16 | #define SUPERBLOCK_LOCATION 0 | |
17 | #define SUPERBLOCK_MAGIC 2126579579 | |
18 | #define SUPERBLOCK_CSUM_XOR 146538381 | |
19 | #define MIN_ERA_VERSION 1 | |
20 | #define MAX_ERA_VERSION 1 | |
21 | #define INVALID_WRITESET_ROOT SUPERBLOCK_LOCATION | |
22 | #define MIN_BLOCK_SIZE 8 | |
23 | ||
24 | /*---------------------------------------------------------------- | |
25 | * Writeset | |
26 | *--------------------------------------------------------------*/ | |
27 | struct writeset_metadata { | |
28 | uint32_t nr_bits; | |
29 | dm_block_t root; | |
30 | }; | |
31 | ||
32 | struct writeset { | |
33 | struct writeset_metadata md; | |
34 | ||
35 | /* | |
36 | * An in core copy of the bits to save constantly doing look ups on | |
37 | * disk. | |
38 | */ | |
39 | unsigned long *bits; | |
40 | }; | |
41 | ||
42 | /* | |
43 | * This does not free off the on disk bitset as this will normally be done | |
44 | * after digesting into the era array. | |
45 | */ | |
46 | static void writeset_free(struct writeset *ws) | |
47 | { | |
48 | vfree(ws->bits); | |
49 | } | |
50 | ||
51 | static int setup_on_disk_bitset(struct dm_disk_bitset *info, | |
52 | unsigned nr_bits, dm_block_t *root) | |
53 | { | |
54 | int r; | |
55 | ||
56 | r = dm_bitset_empty(info, root); | |
57 | if (r) | |
58 | return r; | |
59 | ||
60 | return dm_bitset_resize(info, *root, 0, nr_bits, false, root); | |
61 | } | |
62 | ||
63 | static size_t bitset_size(unsigned nr_bits) | |
64 | { | |
65 | return sizeof(unsigned long) * dm_div_up(nr_bits, BITS_PER_LONG); | |
66 | } | |
67 | ||
68 | /* | |
69 | * Allocates memory for the in core bitset. | |
70 | */ | |
71 | static int writeset_alloc(struct writeset *ws, dm_block_t nr_blocks) | |
72 | { | |
73 | ws->md.nr_bits = nr_blocks; | |
74 | ws->md.root = INVALID_WRITESET_ROOT; | |
75 | ws->bits = vzalloc(bitset_size(nr_blocks)); | |
76 | if (!ws->bits) { | |
77 | DMERR("%s: couldn't allocate in memory bitset", __func__); | |
78 | return -ENOMEM; | |
79 | } | |
80 | ||
81 | return 0; | |
82 | } | |
83 | ||
84 | /* | |
85 | * Wipes the in-core bitset, and creates a new on disk bitset. | |
86 | */ | |
87 | static int writeset_init(struct dm_disk_bitset *info, struct writeset *ws) | |
88 | { | |
89 | int r; | |
90 | ||
91 | memset(ws->bits, 0, bitset_size(ws->md.nr_bits)); | |
92 | ||
93 | r = setup_on_disk_bitset(info, ws->md.nr_bits, &ws->md.root); | |
94 | if (r) { | |
95 | DMERR("%s: setup_on_disk_bitset failed", __func__); | |
96 | return r; | |
97 | } | |
98 | ||
99 | return 0; | |
100 | } | |
101 | ||
102 | static bool writeset_marked(struct writeset *ws, dm_block_t block) | |
103 | { | |
104 | return test_bit(block, ws->bits); | |
105 | } | |
106 | ||
107 | static int writeset_marked_on_disk(struct dm_disk_bitset *info, | |
108 | struct writeset_metadata *m, dm_block_t block, | |
109 | bool *result) | |
110 | { | |
111 | dm_block_t old = m->root; | |
112 | ||
113 | /* | |
114 | * The bitset was flushed when it was archived, so we know there'll | |
115 | * be no change to the root. | |
116 | */ | |
117 | int r = dm_bitset_test_bit(info, m->root, block, &m->root, result); | |
118 | if (r) { | |
119 | DMERR("%s: dm_bitset_test_bit failed", __func__); | |
120 | return r; | |
121 | } | |
122 | ||
123 | BUG_ON(m->root != old); | |
124 | ||
125 | return r; | |
126 | } | |
127 | ||
128 | /* | |
129 | * Returns < 0 on error, 0 if the bit wasn't previously set, 1 if it was. | |
130 | */ | |
131 | static int writeset_test_and_set(struct dm_disk_bitset *info, | |
132 | struct writeset *ws, uint32_t block) | |
133 | { | |
134 | int r; | |
135 | ||
136 | if (!test_and_set_bit(block, ws->bits)) { | |
137 | r = dm_bitset_set_bit(info, ws->md.root, block, &ws->md.root); | |
138 | if (r) { | |
139 | /* FIXME: fail mode */ | |
140 | return r; | |
141 | } | |
142 | ||
143 | return 0; | |
144 | } | |
145 | ||
146 | return 1; | |
147 | } | |
148 | ||
149 | /*---------------------------------------------------------------- | |
150 | * On disk metadata layout | |
151 | *--------------------------------------------------------------*/ | |
152 | #define SPACE_MAP_ROOT_SIZE 128 | |
153 | #define UUID_LEN 16 | |
154 | ||
155 | struct writeset_disk { | |
156 | __le32 nr_bits; | |
157 | __le64 root; | |
158 | } __packed; | |
159 | ||
160 | struct superblock_disk { | |
161 | __le32 csum; | |
162 | __le32 flags; | |
163 | __le64 blocknr; | |
164 | ||
165 | __u8 uuid[UUID_LEN]; | |
166 | __le64 magic; | |
167 | __le32 version; | |
168 | ||
169 | __u8 metadata_space_map_root[SPACE_MAP_ROOT_SIZE]; | |
170 | ||
171 | __le32 data_block_size; | |
172 | __le32 metadata_block_size; | |
173 | __le32 nr_blocks; | |
174 | ||
175 | __le32 current_era; | |
176 | struct writeset_disk current_writeset; | |
177 | ||
178 | /* | |
179 | * Only these two fields are valid within the metadata snapshot. | |
180 | */ | |
181 | __le64 writeset_tree_root; | |
182 | __le64 era_array_root; | |
183 | ||
184 | __le64 metadata_snap; | |
185 | } __packed; | |
186 | ||
187 | /*---------------------------------------------------------------- | |
188 | * Superblock validation | |
189 | *--------------------------------------------------------------*/ | |
190 | static void sb_prepare_for_write(struct dm_block_validator *v, | |
191 | struct dm_block *b, | |
192 | size_t sb_block_size) | |
193 | { | |
194 | struct superblock_disk *disk = dm_block_data(b); | |
195 | ||
196 | disk->blocknr = cpu_to_le64(dm_block_location(b)); | |
197 | disk->csum = cpu_to_le32(dm_bm_checksum(&disk->flags, | |
198 | sb_block_size - sizeof(__le32), | |
199 | SUPERBLOCK_CSUM_XOR)); | |
200 | } | |
201 | ||
202 | static int check_metadata_version(struct superblock_disk *disk) | |
203 | { | |
204 | uint32_t metadata_version = le32_to_cpu(disk->version); | |
205 | if (metadata_version < MIN_ERA_VERSION || metadata_version > MAX_ERA_VERSION) { | |
206 | DMERR("Era metadata version %u found, but only versions between %u and %u supported.", | |
207 | metadata_version, MIN_ERA_VERSION, MAX_ERA_VERSION); | |
208 | return -EINVAL; | |
209 | } | |
210 | ||
211 | return 0; | |
212 | } | |
213 | ||
214 | static int sb_check(struct dm_block_validator *v, | |
215 | struct dm_block *b, | |
216 | size_t sb_block_size) | |
217 | { | |
218 | struct superblock_disk *disk = dm_block_data(b); | |
219 | __le32 csum_le; | |
220 | ||
221 | if (dm_block_location(b) != le64_to_cpu(disk->blocknr)) { | |
222 | DMERR("sb_check failed: blocknr %llu: wanted %llu", | |
223 | le64_to_cpu(disk->blocknr), | |
224 | (unsigned long long)dm_block_location(b)); | |
225 | return -ENOTBLK; | |
226 | } | |
227 | ||
228 | if (le64_to_cpu(disk->magic) != SUPERBLOCK_MAGIC) { | |
229 | DMERR("sb_check failed: magic %llu: wanted %llu", | |
230 | le64_to_cpu(disk->magic), | |
231 | (unsigned long long) SUPERBLOCK_MAGIC); | |
232 | return -EILSEQ; | |
233 | } | |
234 | ||
235 | csum_le = cpu_to_le32(dm_bm_checksum(&disk->flags, | |
236 | sb_block_size - sizeof(__le32), | |
237 | SUPERBLOCK_CSUM_XOR)); | |
238 | if (csum_le != disk->csum) { | |
239 | DMERR("sb_check failed: csum %u: wanted %u", | |
240 | le32_to_cpu(csum_le), le32_to_cpu(disk->csum)); | |
241 | return -EILSEQ; | |
242 | } | |
243 | ||
244 | return check_metadata_version(disk); | |
245 | } | |
246 | ||
247 | static struct dm_block_validator sb_validator = { | |
248 | .name = "superblock", | |
249 | .prepare_for_write = sb_prepare_for_write, | |
250 | .check = sb_check | |
251 | }; | |
252 | ||
253 | /*---------------------------------------------------------------- | |
254 | * Low level metadata handling | |
255 | *--------------------------------------------------------------*/ | |
256 | #define DM_ERA_METADATA_BLOCK_SIZE 4096 | |
257 | #define DM_ERA_METADATA_CACHE_SIZE 64 | |
258 | #define ERA_MAX_CONCURRENT_LOCKS 5 | |
259 | ||
260 | struct era_metadata { | |
261 | struct block_device *bdev; | |
262 | struct dm_block_manager *bm; | |
263 | struct dm_space_map *sm; | |
264 | struct dm_transaction_manager *tm; | |
265 | ||
266 | dm_block_t block_size; | |
267 | uint32_t nr_blocks; | |
268 | ||
269 | uint32_t current_era; | |
270 | ||
271 | /* | |
272 | * We preallocate 2 writesets. When an era rolls over we | |
273 | * switch between them. This means the allocation is done at | |
274 | * preresume time, rather than on the io path. | |
275 | */ | |
276 | struct writeset writesets[2]; | |
277 | struct writeset *current_writeset; | |
278 | ||
279 | dm_block_t writeset_tree_root; | |
280 | dm_block_t era_array_root; | |
281 | ||
282 | struct dm_disk_bitset bitset_info; | |
283 | struct dm_btree_info writeset_tree_info; | |
284 | struct dm_array_info era_array_info; | |
285 | ||
286 | dm_block_t metadata_snap; | |
287 | ||
288 | /* | |
289 | * A flag that is set whenever a writeset has been archived. | |
290 | */ | |
291 | bool archived_writesets; | |
292 | }; | |
293 | ||
294 | static int superblock_read_lock(struct era_metadata *md, | |
295 | struct dm_block **sblock) | |
296 | { | |
297 | return dm_bm_read_lock(md->bm, SUPERBLOCK_LOCATION, | |
298 | &sb_validator, sblock); | |
299 | } | |
300 | ||
301 | static int superblock_lock_zero(struct era_metadata *md, | |
302 | struct dm_block **sblock) | |
303 | { | |
304 | return dm_bm_write_lock_zero(md->bm, SUPERBLOCK_LOCATION, | |
305 | &sb_validator, sblock); | |
306 | } | |
307 | ||
308 | static int superblock_lock(struct era_metadata *md, | |
309 | struct dm_block **sblock) | |
310 | { | |
311 | return dm_bm_write_lock(md->bm, SUPERBLOCK_LOCATION, | |
312 | &sb_validator, sblock); | |
313 | } | |
314 | ||
315 | /* FIXME: duplication with cache and thin */ | |
316 | static int superblock_all_zeroes(struct dm_block_manager *bm, bool *result) | |
317 | { | |
318 | int r; | |
319 | unsigned i; | |
320 | struct dm_block *b; | |
321 | __le64 *data_le, zero = cpu_to_le64(0); | |
322 | unsigned sb_block_size = dm_bm_block_size(bm) / sizeof(__le64); | |
323 | ||
324 | /* | |
325 | * We can't use a validator here - it may be all zeroes. | |
326 | */ | |
327 | r = dm_bm_read_lock(bm, SUPERBLOCK_LOCATION, NULL, &b); | |
328 | if (r) | |
329 | return r; | |
330 | ||
331 | data_le = dm_block_data(b); | |
332 | *result = true; | |
333 | for (i = 0; i < sb_block_size; i++) { | |
334 | if (data_le[i] != zero) { | |
335 | *result = false; | |
336 | break; | |
337 | } | |
338 | } | |
339 | ||
340 | return dm_bm_unlock(b); | |
341 | } | |
342 | ||
343 | /*----------------------------------------------------------------*/ | |
344 | ||
345 | static void ws_pack(const struct writeset_metadata *core, struct writeset_disk *disk) | |
346 | { | |
347 | disk->nr_bits = cpu_to_le32(core->nr_bits); | |
348 | disk->root = cpu_to_le64(core->root); | |
349 | } | |
350 | ||
351 | static void ws_unpack(const struct writeset_disk *disk, struct writeset_metadata *core) | |
352 | { | |
353 | core->nr_bits = le32_to_cpu(disk->nr_bits); | |
354 | core->root = le64_to_cpu(disk->root); | |
355 | } | |
356 | ||
357 | static void ws_inc(void *context, const void *value) | |
358 | { | |
359 | struct era_metadata *md = context; | |
360 | struct writeset_disk ws_d; | |
361 | dm_block_t b; | |
362 | ||
363 | memcpy(&ws_d, value, sizeof(ws_d)); | |
364 | b = le64_to_cpu(ws_d.root); | |
365 | ||
366 | dm_tm_inc(md->tm, b); | |
367 | } | |
368 | ||
369 | static void ws_dec(void *context, const void *value) | |
370 | { | |
371 | struct era_metadata *md = context; | |
372 | struct writeset_disk ws_d; | |
373 | dm_block_t b; | |
374 | ||
375 | memcpy(&ws_d, value, sizeof(ws_d)); | |
376 | b = le64_to_cpu(ws_d.root); | |
377 | ||
378 | dm_bitset_del(&md->bitset_info, b); | |
379 | } | |
380 | ||
381 | static int ws_eq(void *context, const void *value1, const void *value2) | |
382 | { | |
383 | return !memcmp(value1, value2, sizeof(struct writeset_metadata)); | |
384 | } | |
385 | ||
386 | /*----------------------------------------------------------------*/ | |
387 | ||
388 | static void setup_writeset_tree_info(struct era_metadata *md) | |
389 | { | |
390 | struct dm_btree_value_type *vt = &md->writeset_tree_info.value_type; | |
391 | md->writeset_tree_info.tm = md->tm; | |
392 | md->writeset_tree_info.levels = 1; | |
393 | vt->context = md; | |
394 | vt->size = sizeof(struct writeset_disk); | |
395 | vt->inc = ws_inc; | |
396 | vt->dec = ws_dec; | |
397 | vt->equal = ws_eq; | |
398 | } | |
399 | ||
400 | static void setup_era_array_info(struct era_metadata *md) | |
401 | ||
402 | { | |
403 | struct dm_btree_value_type vt; | |
404 | vt.context = NULL; | |
405 | vt.size = sizeof(__le32); | |
406 | vt.inc = NULL; | |
407 | vt.dec = NULL; | |
408 | vt.equal = NULL; | |
409 | ||
410 | dm_array_info_init(&md->era_array_info, md->tm, &vt); | |
411 | } | |
412 | ||
413 | static void setup_infos(struct era_metadata *md) | |
414 | { | |
415 | dm_disk_bitset_init(md->tm, &md->bitset_info); | |
416 | setup_writeset_tree_info(md); | |
417 | setup_era_array_info(md); | |
418 | } | |
419 | ||
420 | /*----------------------------------------------------------------*/ | |
421 | ||
422 | static int create_fresh_metadata(struct era_metadata *md) | |
423 | { | |
424 | int r; | |
425 | ||
426 | r = dm_tm_create_with_sm(md->bm, SUPERBLOCK_LOCATION, | |
427 | &md->tm, &md->sm); | |
428 | if (r < 0) { | |
429 | DMERR("dm_tm_create_with_sm failed"); | |
430 | return r; | |
431 | } | |
432 | ||
433 | setup_infos(md); | |
434 | ||
435 | r = dm_btree_empty(&md->writeset_tree_info, &md->writeset_tree_root); | |
436 | if (r) { | |
437 | DMERR("couldn't create new writeset tree"); | |
438 | goto bad; | |
439 | } | |
440 | ||
441 | r = dm_array_empty(&md->era_array_info, &md->era_array_root); | |
442 | if (r) { | |
443 | DMERR("couldn't create era array"); | |
444 | goto bad; | |
445 | } | |
446 | ||
447 | return 0; | |
448 | ||
449 | bad: | |
450 | dm_sm_destroy(md->sm); | |
451 | dm_tm_destroy(md->tm); | |
452 | ||
453 | return r; | |
454 | } | |
455 | ||
456 | /* | |
457 | * Writes a superblock, including the static fields that don't get updated | |
458 | * with every commit (possible optimisation here). 'md' should be fully | |
459 | * constructed when this is called. | |
460 | */ | |
461 | static int prepare_superblock(struct era_metadata *md, struct superblock_disk *disk) | |
462 | { | |
463 | int r; | |
464 | size_t metadata_len; | |
465 | ||
466 | disk->magic = cpu_to_le64(SUPERBLOCK_MAGIC); | |
467 | disk->flags = cpu_to_le32(0ul); | |
468 | ||
469 | /* FIXME: can't keep blanking the uuid (uuid is currently unused though) */ | |
470 | memset(disk->uuid, 0, sizeof(disk->uuid)); | |
471 | disk->version = cpu_to_le32(MAX_ERA_VERSION); | |
472 | ||
473 | r = dm_sm_root_size(md->sm, &metadata_len); | |
474 | if (r < 0) | |
475 | return r; | |
476 | ||
477 | r = dm_sm_copy_root(md->sm, &disk->metadata_space_map_root, | |
478 | metadata_len); | |
479 | if (r < 0) | |
480 | return r; | |
481 | ||
482 | disk->data_block_size = cpu_to_le32(md->block_size); | |
483 | disk->metadata_block_size = cpu_to_le32(DM_ERA_METADATA_BLOCK_SIZE >> SECTOR_SHIFT); | |
484 | disk->nr_blocks = cpu_to_le32(md->nr_blocks); | |
485 | disk->current_era = cpu_to_le32(md->current_era); | |
486 | ||
487 | ws_pack(&md->current_writeset->md, &disk->current_writeset); | |
488 | disk->writeset_tree_root = cpu_to_le64(md->writeset_tree_root); | |
489 | disk->era_array_root = cpu_to_le64(md->era_array_root); | |
490 | disk->metadata_snap = cpu_to_le64(md->metadata_snap); | |
491 | ||
492 | return 0; | |
493 | } | |
494 | ||
495 | static int write_superblock(struct era_metadata *md) | |
496 | { | |
497 | int r; | |
498 | struct dm_block *sblock; | |
499 | struct superblock_disk *disk; | |
500 | ||
501 | r = superblock_lock_zero(md, &sblock); | |
502 | if (r) | |
503 | return r; | |
504 | ||
505 | disk = dm_block_data(sblock); | |
506 | r = prepare_superblock(md, disk); | |
507 | if (r) { | |
508 | DMERR("%s: prepare_superblock failed", __func__); | |
509 | dm_bm_unlock(sblock); /* FIXME: does this commit? */ | |
510 | return r; | |
511 | } | |
512 | ||
513 | return dm_tm_commit(md->tm, sblock); | |
514 | } | |
515 | ||
516 | /* | |
517 | * Assumes block_size and the infos are set. | |
518 | */ | |
519 | static int format_metadata(struct era_metadata *md) | |
520 | { | |
521 | int r; | |
522 | ||
523 | r = create_fresh_metadata(md); | |
524 | if (r) | |
525 | return r; | |
526 | ||
527 | r = write_superblock(md); | |
528 | if (r) { | |
529 | dm_sm_destroy(md->sm); | |
530 | dm_tm_destroy(md->tm); | |
531 | return r; | |
532 | } | |
533 | ||
534 | return 0; | |
535 | } | |
536 | ||
537 | static int open_metadata(struct era_metadata *md) | |
538 | { | |
539 | int r; | |
540 | struct dm_block *sblock; | |
541 | struct superblock_disk *disk; | |
542 | ||
543 | r = superblock_read_lock(md, &sblock); | |
544 | if (r) { | |
545 | DMERR("couldn't read_lock superblock"); | |
546 | return r; | |
547 | } | |
548 | ||
549 | disk = dm_block_data(sblock); | |
550 | r = dm_tm_open_with_sm(md->bm, SUPERBLOCK_LOCATION, | |
551 | disk->metadata_space_map_root, | |
552 | sizeof(disk->metadata_space_map_root), | |
553 | &md->tm, &md->sm); | |
554 | if (r) { | |
555 | DMERR("dm_tm_open_with_sm failed"); | |
556 | goto bad; | |
557 | } | |
558 | ||
559 | setup_infos(md); | |
560 | ||
561 | md->block_size = le32_to_cpu(disk->data_block_size); | |
562 | md->nr_blocks = le32_to_cpu(disk->nr_blocks); | |
563 | md->current_era = le32_to_cpu(disk->current_era); | |
564 | ||
565 | md->writeset_tree_root = le64_to_cpu(disk->writeset_tree_root); | |
566 | md->era_array_root = le64_to_cpu(disk->era_array_root); | |
567 | md->metadata_snap = le64_to_cpu(disk->metadata_snap); | |
568 | md->archived_writesets = true; | |
569 | ||
570 | return dm_bm_unlock(sblock); | |
571 | ||
572 | bad: | |
573 | dm_bm_unlock(sblock); | |
574 | return r; | |
575 | } | |
576 | ||
577 | static int open_or_format_metadata(struct era_metadata *md, | |
578 | bool may_format) | |
579 | { | |
580 | int r; | |
581 | bool unformatted = false; | |
582 | ||
583 | r = superblock_all_zeroes(md->bm, &unformatted); | |
584 | if (r) | |
585 | return r; | |
586 | ||
587 | if (unformatted) | |
588 | return may_format ? format_metadata(md) : -EPERM; | |
589 | ||
590 | return open_metadata(md); | |
591 | } | |
592 | ||
593 | static int create_persistent_data_objects(struct era_metadata *md, | |
594 | bool may_format) | |
595 | { | |
596 | int r; | |
597 | ||
598 | md->bm = dm_block_manager_create(md->bdev, DM_ERA_METADATA_BLOCK_SIZE, | |
599 | DM_ERA_METADATA_CACHE_SIZE, | |
600 | ERA_MAX_CONCURRENT_LOCKS); | |
601 | if (IS_ERR(md->bm)) { | |
602 | DMERR("could not create block manager"); | |
603 | return PTR_ERR(md->bm); | |
604 | } | |
605 | ||
606 | r = open_or_format_metadata(md, may_format); | |
607 | if (r) | |
608 | dm_block_manager_destroy(md->bm); | |
609 | ||
610 | return r; | |
611 | } | |
612 | ||
613 | static void destroy_persistent_data_objects(struct era_metadata *md) | |
614 | { | |
615 | dm_sm_destroy(md->sm); | |
616 | dm_tm_destroy(md->tm); | |
617 | dm_block_manager_destroy(md->bm); | |
618 | } | |
619 | ||
620 | /* | |
621 | * This waits until all era_map threads have picked up the new filter. | |
622 | */ | |
623 | static void swap_writeset(struct era_metadata *md, struct writeset *new_writeset) | |
624 | { | |
625 | rcu_assign_pointer(md->current_writeset, new_writeset); | |
626 | synchronize_rcu(); | |
627 | } | |
628 | ||
629 | /*---------------------------------------------------------------- | |
630 | * Writesets get 'digested' into the main era array. | |
631 | * | |
632 | * We're using a coroutine here so the worker thread can do the digestion, | |
633 | * thus avoiding synchronisation of the metadata. Digesting a whole | |
634 | * writeset in one go would cause too much latency. | |
635 | *--------------------------------------------------------------*/ | |
636 | struct digest { | |
637 | uint32_t era; | |
638 | unsigned nr_bits, current_bit; | |
639 | struct writeset_metadata writeset; | |
640 | __le32 value; | |
641 | struct dm_disk_bitset info; | |
642 | ||
643 | int (*step)(struct era_metadata *, struct digest *); | |
644 | }; | |
645 | ||
646 | static int metadata_digest_lookup_writeset(struct era_metadata *md, | |
647 | struct digest *d); | |
648 | ||
649 | static int metadata_digest_remove_writeset(struct era_metadata *md, | |
650 | struct digest *d) | |
651 | { | |
652 | int r; | |
653 | uint64_t key = d->era; | |
654 | ||
655 | r = dm_btree_remove(&md->writeset_tree_info, md->writeset_tree_root, | |
656 | &key, &md->writeset_tree_root); | |
657 | if (r) { | |
658 | DMERR("%s: dm_btree_remove failed", __func__); | |
659 | return r; | |
660 | } | |
661 | ||
662 | d->step = metadata_digest_lookup_writeset; | |
663 | return 0; | |
664 | } | |
665 | ||
666 | #define INSERTS_PER_STEP 100 | |
667 | ||
668 | static int metadata_digest_transcribe_writeset(struct era_metadata *md, | |
669 | struct digest *d) | |
670 | { | |
671 | int r; | |
672 | bool marked; | |
673 | unsigned b, e = min(d->current_bit + INSERTS_PER_STEP, d->nr_bits); | |
674 | ||
675 | for (b = d->current_bit; b < e; b++) { | |
676 | r = writeset_marked_on_disk(&d->info, &d->writeset, b, &marked); | |
677 | if (r) { | |
678 | DMERR("%s: writeset_marked_on_disk failed", __func__); | |
679 | return r; | |
680 | } | |
681 | ||
682 | if (!marked) | |
683 | continue; | |
684 | ||
685 | __dm_bless_for_disk(&d->value); | |
686 | r = dm_array_set_value(&md->era_array_info, md->era_array_root, | |
687 | b, &d->value, &md->era_array_root); | |
688 | if (r) { | |
689 | DMERR("%s: dm_array_set_value failed", __func__); | |
690 | return r; | |
691 | } | |
692 | } | |
693 | ||
694 | if (b == d->nr_bits) | |
695 | d->step = metadata_digest_remove_writeset; | |
696 | else | |
697 | d->current_bit = b; | |
698 | ||
699 | return 0; | |
700 | } | |
701 | ||
702 | static int metadata_digest_lookup_writeset(struct era_metadata *md, | |
703 | struct digest *d) | |
704 | { | |
705 | int r; | |
706 | uint64_t key; | |
707 | struct writeset_disk disk; | |
708 | ||
709 | r = dm_btree_find_lowest_key(&md->writeset_tree_info, | |
710 | md->writeset_tree_root, &key); | |
711 | if (r < 0) | |
712 | return r; | |
713 | ||
714 | d->era = key; | |
715 | ||
716 | r = dm_btree_lookup(&md->writeset_tree_info, | |
717 | md->writeset_tree_root, &key, &disk); | |
718 | if (r) { | |
719 | if (r == -ENODATA) { | |
720 | d->step = NULL; | |
721 | return 0; | |
722 | } | |
723 | ||
724 | DMERR("%s: dm_btree_lookup failed", __func__); | |
725 | return r; | |
726 | } | |
727 | ||
728 | ws_unpack(&disk, &d->writeset); | |
729 | d->value = cpu_to_le32(key); | |
730 | ||
731 | d->nr_bits = min(d->writeset.nr_bits, md->nr_blocks); | |
732 | d->current_bit = 0; | |
733 | d->step = metadata_digest_transcribe_writeset; | |
734 | ||
735 | return 0; | |
736 | } | |
737 | ||
738 | static int metadata_digest_start(struct era_metadata *md, struct digest *d) | |
739 | { | |
740 | if (d->step) | |
741 | return 0; | |
742 | ||
743 | memset(d, 0, sizeof(*d)); | |
744 | ||
745 | /* | |
746 | * We initialise another bitset info to avoid any caching side | |
747 | * effects with the previous one. | |
748 | */ | |
749 | dm_disk_bitset_init(md->tm, &d->info); | |
750 | d->step = metadata_digest_lookup_writeset; | |
751 | ||
752 | return 0; | |
753 | } | |
754 | ||
755 | /*---------------------------------------------------------------- | |
756 | * High level metadata interface. Target methods should use these, and not | |
757 | * the lower level ones. | |
758 | *--------------------------------------------------------------*/ | |
759 | static struct era_metadata *metadata_open(struct block_device *bdev, | |
760 | sector_t block_size, | |
761 | bool may_format) | |
762 | { | |
763 | int r; | |
764 | struct era_metadata *md = kzalloc(sizeof(*md), GFP_KERNEL); | |
765 | ||
766 | if (!md) | |
767 | return NULL; | |
768 | ||
769 | md->bdev = bdev; | |
770 | md->block_size = block_size; | |
771 | ||
772 | md->writesets[0].md.root = INVALID_WRITESET_ROOT; | |
773 | md->writesets[1].md.root = INVALID_WRITESET_ROOT; | |
774 | md->current_writeset = &md->writesets[0]; | |
775 | ||
776 | r = create_persistent_data_objects(md, may_format); | |
777 | if (r) { | |
778 | kfree(md); | |
779 | return ERR_PTR(r); | |
780 | } | |
781 | ||
782 | return md; | |
783 | } | |
784 | ||
785 | static void metadata_close(struct era_metadata *md) | |
786 | { | |
787 | destroy_persistent_data_objects(md); | |
788 | kfree(md); | |
789 | } | |
790 | ||
791 | static bool valid_nr_blocks(dm_block_t n) | |
792 | { | |
793 | /* | |
794 | * dm_bitset restricts us to 2^32. test_bit & co. restrict us | |
795 | * further to 2^31 - 1 | |
796 | */ | |
797 | return n < (1ull << 31); | |
798 | } | |
799 | ||
800 | static int metadata_resize(struct era_metadata *md, void *arg) | |
801 | { | |
802 | int r; | |
803 | dm_block_t *new_size = arg; | |
804 | __le32 value; | |
805 | ||
806 | if (!valid_nr_blocks(*new_size)) { | |
807 | DMERR("Invalid number of origin blocks %llu", | |
808 | (unsigned long long) *new_size); | |
809 | return -EINVAL; | |
810 | } | |
811 | ||
812 | writeset_free(&md->writesets[0]); | |
813 | writeset_free(&md->writesets[1]); | |
814 | ||
815 | r = writeset_alloc(&md->writesets[0], *new_size); | |
816 | if (r) { | |
817 | DMERR("%s: writeset_alloc failed for writeset 0", __func__); | |
818 | return r; | |
819 | } | |
820 | ||
821 | r = writeset_alloc(&md->writesets[1], *new_size); | |
822 | if (r) { | |
823 | DMERR("%s: writeset_alloc failed for writeset 1", __func__); | |
824 | return r; | |
825 | } | |
826 | ||
827 | value = cpu_to_le32(0u); | |
828 | __dm_bless_for_disk(&value); | |
829 | r = dm_array_resize(&md->era_array_info, md->era_array_root, | |
830 | md->nr_blocks, *new_size, | |
831 | &value, &md->era_array_root); | |
832 | if (r) { | |
833 | DMERR("%s: dm_array_resize failed", __func__); | |
834 | return r; | |
835 | } | |
836 | ||
837 | md->nr_blocks = *new_size; | |
838 | return 0; | |
839 | } | |
840 | ||
841 | static int metadata_era_archive(struct era_metadata *md) | |
842 | { | |
843 | int r; | |
844 | uint64_t keys[1]; | |
845 | struct writeset_disk value; | |
846 | ||
847 | r = dm_bitset_flush(&md->bitset_info, md->current_writeset->md.root, | |
848 | &md->current_writeset->md.root); | |
849 | if (r) { | |
850 | DMERR("%s: dm_bitset_flush failed", __func__); | |
851 | return r; | |
852 | } | |
853 | ||
854 | ws_pack(&md->current_writeset->md, &value); | |
855 | md->current_writeset->md.root = INVALID_WRITESET_ROOT; | |
856 | ||
857 | keys[0] = md->current_era; | |
858 | __dm_bless_for_disk(&value); | |
859 | r = dm_btree_insert(&md->writeset_tree_info, md->writeset_tree_root, | |
860 | keys, &value, &md->writeset_tree_root); | |
861 | if (r) { | |
862 | DMERR("%s: couldn't insert writeset into btree", __func__); | |
863 | /* FIXME: fail mode */ | |
864 | return r; | |
865 | } | |
866 | ||
867 | md->archived_writesets = true; | |
868 | ||
869 | return 0; | |
870 | } | |
871 | ||
872 | static struct writeset *next_writeset(struct era_metadata *md) | |
873 | { | |
874 | return (md->current_writeset == &md->writesets[0]) ? | |
875 | &md->writesets[1] : &md->writesets[0]; | |
876 | } | |
877 | ||
878 | static int metadata_new_era(struct era_metadata *md) | |
879 | { | |
880 | int r; | |
881 | struct writeset *new_writeset = next_writeset(md); | |
882 | ||
883 | r = writeset_init(&md->bitset_info, new_writeset); | |
884 | if (r) { | |
885 | DMERR("%s: writeset_init failed", __func__); | |
886 | return r; | |
887 | } | |
888 | ||
889 | swap_writeset(md, new_writeset); | |
890 | md->current_era++; | |
891 | ||
892 | return 0; | |
893 | } | |
894 | ||
895 | static int metadata_era_rollover(struct era_metadata *md) | |
896 | { | |
897 | int r; | |
898 | ||
899 | if (md->current_writeset->md.root != INVALID_WRITESET_ROOT) { | |
900 | r = metadata_era_archive(md); | |
901 | if (r) { | |
902 | DMERR("%s: metadata_archive_era failed", __func__); | |
903 | /* FIXME: fail mode? */ | |
904 | return r; | |
905 | } | |
906 | } | |
907 | ||
908 | r = metadata_new_era(md); | |
909 | if (r) { | |
910 | DMERR("%s: new era failed", __func__); | |
911 | /* FIXME: fail mode */ | |
912 | return r; | |
913 | } | |
914 | ||
915 | return 0; | |
916 | } | |
917 | ||
918 | static bool metadata_current_marked(struct era_metadata *md, dm_block_t block) | |
919 | { | |
920 | bool r; | |
921 | struct writeset *ws; | |
922 | ||
923 | rcu_read_lock(); | |
924 | ws = rcu_dereference(md->current_writeset); | |
925 | r = writeset_marked(ws, block); | |
926 | rcu_read_unlock(); | |
927 | ||
928 | return r; | |
929 | } | |
930 | ||
931 | static int metadata_commit(struct era_metadata *md) | |
932 | { | |
933 | int r; | |
934 | struct dm_block *sblock; | |
935 | ||
936 | if (md->current_writeset->md.root != SUPERBLOCK_LOCATION) { | |
937 | r = dm_bitset_flush(&md->bitset_info, md->current_writeset->md.root, | |
938 | &md->current_writeset->md.root); | |
939 | if (r) { | |
940 | DMERR("%s: bitset flush failed", __func__); | |
941 | return r; | |
942 | } | |
943 | } | |
944 | ||
945 | r = dm_tm_pre_commit(md->tm); | |
946 | if (r) { | |
947 | DMERR("%s: pre commit failed", __func__); | |
948 | return r; | |
949 | } | |
950 | ||
951 | r = superblock_lock(md, &sblock); | |
952 | if (r) { | |
953 | DMERR("%s: superblock lock failed", __func__); | |
954 | return r; | |
955 | } | |
956 | ||
957 | r = prepare_superblock(md, dm_block_data(sblock)); | |
958 | if (r) { | |
959 | DMERR("%s: prepare_superblock failed", __func__); | |
960 | dm_bm_unlock(sblock); /* FIXME: does this commit? */ | |
961 | return r; | |
962 | } | |
963 | ||
964 | return dm_tm_commit(md->tm, sblock); | |
965 | } | |
966 | ||
967 | static int metadata_checkpoint(struct era_metadata *md) | |
968 | { | |
969 | /* | |
970 | * For now we just rollover, but later I want to put a check in to | |
971 | * avoid this if the filter is still pretty fresh. | |
972 | */ | |
973 | return metadata_era_rollover(md); | |
974 | } | |
975 | ||
976 | /* | |
977 | * Metadata snapshots allow userland to access era data. | |
978 | */ | |
979 | static int metadata_take_snap(struct era_metadata *md) | |
980 | { | |
981 | int r, inc; | |
982 | struct dm_block *clone; | |
983 | ||
984 | if (md->metadata_snap != SUPERBLOCK_LOCATION) { | |
985 | DMERR("%s: metadata snapshot already exists", __func__); | |
986 | return -EINVAL; | |
987 | } | |
988 | ||
989 | r = metadata_era_rollover(md); | |
990 | if (r) { | |
991 | DMERR("%s: era rollover failed", __func__); | |
992 | return r; | |
993 | } | |
994 | ||
995 | r = metadata_commit(md); | |
996 | if (r) { | |
997 | DMERR("%s: pre commit failed", __func__); | |
998 | return r; | |
999 | } | |
1000 | ||
1001 | r = dm_sm_inc_block(md->sm, SUPERBLOCK_LOCATION); | |
1002 | if (r) { | |
1003 | DMERR("%s: couldn't increment superblock", __func__); | |
1004 | return r; | |
1005 | } | |
1006 | ||
1007 | r = dm_tm_shadow_block(md->tm, SUPERBLOCK_LOCATION, | |
1008 | &sb_validator, &clone, &inc); | |
1009 | if (r) { | |
1010 | DMERR("%s: couldn't shadow superblock", __func__); | |
1011 | dm_sm_dec_block(md->sm, SUPERBLOCK_LOCATION); | |
1012 | return r; | |
1013 | } | |
1014 | BUG_ON(!inc); | |
1015 | ||
1016 | r = dm_sm_inc_block(md->sm, md->writeset_tree_root); | |
1017 | if (r) { | |
1018 | DMERR("%s: couldn't inc writeset tree root", __func__); | |
1019 | dm_tm_unlock(md->tm, clone); | |
1020 | return r; | |
1021 | } | |
1022 | ||
1023 | r = dm_sm_inc_block(md->sm, md->era_array_root); | |
1024 | if (r) { | |
1025 | DMERR("%s: couldn't inc era tree root", __func__); | |
1026 | dm_sm_dec_block(md->sm, md->writeset_tree_root); | |
1027 | dm_tm_unlock(md->tm, clone); | |
1028 | return r; | |
1029 | } | |
1030 | ||
1031 | md->metadata_snap = dm_block_location(clone); | |
1032 | ||
1033 | r = dm_tm_unlock(md->tm, clone); | |
1034 | if (r) { | |
1035 | DMERR("%s: couldn't unlock clone", __func__); | |
1036 | md->metadata_snap = SUPERBLOCK_LOCATION; | |
1037 | return r; | |
1038 | } | |
1039 | ||
1040 | return 0; | |
1041 | } | |
1042 | ||
1043 | static int metadata_drop_snap(struct era_metadata *md) | |
1044 | { | |
1045 | int r; | |
1046 | dm_block_t location; | |
1047 | struct dm_block *clone; | |
1048 | struct superblock_disk *disk; | |
1049 | ||
1050 | if (md->metadata_snap == SUPERBLOCK_LOCATION) { | |
1051 | DMERR("%s: no snap to drop", __func__); | |
1052 | return -EINVAL; | |
1053 | } | |
1054 | ||
1055 | r = dm_tm_read_lock(md->tm, md->metadata_snap, &sb_validator, &clone); | |
1056 | if (r) { | |
1057 | DMERR("%s: couldn't read lock superblock clone", __func__); | |
1058 | return r; | |
1059 | } | |
1060 | ||
1061 | /* | |
1062 | * Whatever happens now we'll commit with no record of the metadata | |
1063 | * snap. | |
1064 | */ | |
1065 | md->metadata_snap = SUPERBLOCK_LOCATION; | |
1066 | ||
1067 | disk = dm_block_data(clone); | |
1068 | r = dm_btree_del(&md->writeset_tree_info, | |
1069 | le64_to_cpu(disk->writeset_tree_root)); | |
1070 | if (r) { | |
1071 | DMERR("%s: error deleting writeset tree clone", __func__); | |
1072 | dm_tm_unlock(md->tm, clone); | |
1073 | return r; | |
1074 | } | |
1075 | ||
1076 | r = dm_array_del(&md->era_array_info, le64_to_cpu(disk->era_array_root)); | |
1077 | if (r) { | |
1078 | DMERR("%s: error deleting era array clone", __func__); | |
1079 | dm_tm_unlock(md->tm, clone); | |
1080 | return r; | |
1081 | } | |
1082 | ||
1083 | location = dm_block_location(clone); | |
1084 | dm_tm_unlock(md->tm, clone); | |
1085 | ||
1086 | return dm_sm_dec_block(md->sm, location); | |
1087 | } | |
1088 | ||
1089 | struct metadata_stats { | |
1090 | dm_block_t used; | |
1091 | dm_block_t total; | |
1092 | dm_block_t snap; | |
1093 | uint32_t era; | |
1094 | }; | |
1095 | ||
1096 | static int metadata_get_stats(struct era_metadata *md, void *ptr) | |
1097 | { | |
1098 | int r; | |
1099 | struct metadata_stats *s = ptr; | |
1100 | dm_block_t nr_free, nr_total; | |
1101 | ||
1102 | r = dm_sm_get_nr_free(md->sm, &nr_free); | |
1103 | if (r) { | |
1104 | DMERR("dm_sm_get_nr_free returned %d", r); | |
1105 | return r; | |
1106 | } | |
1107 | ||
1108 | r = dm_sm_get_nr_blocks(md->sm, &nr_total); | |
1109 | if (r) { | |
1110 | DMERR("dm_pool_get_metadata_dev_size returned %d", r); | |
1111 | return r; | |
1112 | } | |
1113 | ||
1114 | s->used = nr_total - nr_free; | |
1115 | s->total = nr_total; | |
1116 | s->snap = md->metadata_snap; | |
1117 | s->era = md->current_era; | |
1118 | ||
1119 | return 0; | |
1120 | } | |
1121 | ||
1122 | /*----------------------------------------------------------------*/ | |
1123 | ||
1124 | struct era { | |
1125 | struct dm_target *ti; | |
1126 | struct dm_target_callbacks callbacks; | |
1127 | ||
1128 | struct dm_dev *metadata_dev; | |
1129 | struct dm_dev *origin_dev; | |
1130 | ||
1131 | dm_block_t nr_blocks; | |
1132 | uint32_t sectors_per_block; | |
1133 | int sectors_per_block_shift; | |
1134 | struct era_metadata *md; | |
1135 | ||
1136 | struct workqueue_struct *wq; | |
1137 | struct work_struct worker; | |
1138 | ||
1139 | spinlock_t deferred_lock; | |
1140 | struct bio_list deferred_bios; | |
1141 | ||
1142 | spinlock_t rpc_lock; | |
1143 | struct list_head rpc_calls; | |
1144 | ||
1145 | struct digest digest; | |
1146 | atomic_t suspended; | |
1147 | }; | |
1148 | ||
1149 | struct rpc { | |
1150 | struct list_head list; | |
1151 | ||
1152 | int (*fn0)(struct era_metadata *); | |
1153 | int (*fn1)(struct era_metadata *, void *); | |
1154 | void *arg; | |
1155 | int result; | |
1156 | ||
1157 | struct completion complete; | |
1158 | }; | |
1159 | ||
1160 | /*---------------------------------------------------------------- | |
1161 | * Remapping. | |
1162 | *---------------------------------------------------------------*/ | |
1163 | static bool block_size_is_power_of_two(struct era *era) | |
1164 | { | |
1165 | return era->sectors_per_block_shift >= 0; | |
1166 | } | |
1167 | ||
1168 | static dm_block_t get_block(struct era *era, struct bio *bio) | |
1169 | { | |
1170 | sector_t block_nr = bio->bi_iter.bi_sector; | |
1171 | ||
1172 | if (!block_size_is_power_of_two(era)) | |
1173 | (void) sector_div(block_nr, era->sectors_per_block); | |
1174 | else | |
1175 | block_nr >>= era->sectors_per_block_shift; | |
1176 | ||
1177 | return block_nr; | |
1178 | } | |
1179 | ||
1180 | static void remap_to_origin(struct era *era, struct bio *bio) | |
1181 | { | |
1182 | bio->bi_bdev = era->origin_dev->bdev; | |
1183 | } | |
1184 | ||
1185 | /*---------------------------------------------------------------- | |
1186 | * Worker thread | |
1187 | *--------------------------------------------------------------*/ | |
1188 | static void wake_worker(struct era *era) | |
1189 | { | |
1190 | if (!atomic_read(&era->suspended)) | |
1191 | queue_work(era->wq, &era->worker); | |
1192 | } | |
1193 | ||
1194 | static void process_old_eras(struct era *era) | |
1195 | { | |
1196 | int r; | |
1197 | ||
1198 | if (!era->digest.step) | |
1199 | return; | |
1200 | ||
1201 | r = era->digest.step(era->md, &era->digest); | |
1202 | if (r < 0) { | |
1203 | DMERR("%s: digest step failed, stopping digestion", __func__); | |
1204 | era->digest.step = NULL; | |
1205 | ||
1206 | } else if (era->digest.step) | |
1207 | wake_worker(era); | |
1208 | } | |
1209 | ||
1210 | static void process_deferred_bios(struct era *era) | |
1211 | { | |
1212 | int r; | |
1213 | struct bio_list deferred_bios, marked_bios; | |
1214 | struct bio *bio; | |
1215 | bool commit_needed = false; | |
1216 | bool failed = false; | |
1217 | ||
1218 | bio_list_init(&deferred_bios); | |
1219 | bio_list_init(&marked_bios); | |
1220 | ||
1221 | spin_lock(&era->deferred_lock); | |
1222 | bio_list_merge(&deferred_bios, &era->deferred_bios); | |
1223 | bio_list_init(&era->deferred_bios); | |
1224 | spin_unlock(&era->deferred_lock); | |
1225 | ||
1226 | while ((bio = bio_list_pop(&deferred_bios))) { | |
1227 | r = writeset_test_and_set(&era->md->bitset_info, | |
1228 | era->md->current_writeset, | |
1229 | get_block(era, bio)); | |
1230 | if (r < 0) { | |
1231 | /* | |
1232 | * This is bad news, we need to rollback. | |
1233 | * FIXME: finish. | |
1234 | */ | |
1235 | failed = true; | |
1236 | ||
1237 | } else if (r == 0) | |
1238 | commit_needed = true; | |
1239 | ||
1240 | bio_list_add(&marked_bios, bio); | |
1241 | } | |
1242 | ||
1243 | if (commit_needed) { | |
1244 | r = metadata_commit(era->md); | |
1245 | if (r) | |
1246 | failed = true; | |
1247 | } | |
1248 | ||
1249 | if (failed) | |
1250 | while ((bio = bio_list_pop(&marked_bios))) | |
1251 | bio_io_error(bio); | |
1252 | else | |
1253 | while ((bio = bio_list_pop(&marked_bios))) | |
1254 | generic_make_request(bio); | |
1255 | } | |
1256 | ||
1257 | static void process_rpc_calls(struct era *era) | |
1258 | { | |
1259 | int r; | |
1260 | bool need_commit = false; | |
1261 | struct list_head calls; | |
1262 | struct rpc *rpc, *tmp; | |
1263 | ||
1264 | INIT_LIST_HEAD(&calls); | |
1265 | spin_lock(&era->rpc_lock); | |
1266 | list_splice_init(&era->rpc_calls, &calls); | |
1267 | spin_unlock(&era->rpc_lock); | |
1268 | ||
1269 | list_for_each_entry_safe(rpc, tmp, &calls, list) { | |
1270 | rpc->result = rpc->fn0 ? rpc->fn0(era->md) : rpc->fn1(era->md, rpc->arg); | |
1271 | need_commit = true; | |
1272 | } | |
1273 | ||
1274 | if (need_commit) { | |
1275 | r = metadata_commit(era->md); | |
1276 | if (r) | |
1277 | list_for_each_entry_safe(rpc, tmp, &calls, list) | |
1278 | rpc->result = r; | |
1279 | } | |
1280 | ||
1281 | list_for_each_entry_safe(rpc, tmp, &calls, list) | |
1282 | complete(&rpc->complete); | |
1283 | } | |
1284 | ||
1285 | static void kick_off_digest(struct era *era) | |
1286 | { | |
1287 | if (era->md->archived_writesets) { | |
1288 | era->md->archived_writesets = false; | |
1289 | metadata_digest_start(era->md, &era->digest); | |
1290 | } | |
1291 | } | |
1292 | ||
1293 | static void do_work(struct work_struct *ws) | |
1294 | { | |
1295 | struct era *era = container_of(ws, struct era, worker); | |
1296 | ||
1297 | kick_off_digest(era); | |
1298 | process_old_eras(era); | |
1299 | process_deferred_bios(era); | |
1300 | process_rpc_calls(era); | |
1301 | } | |
1302 | ||
1303 | static void defer_bio(struct era *era, struct bio *bio) | |
1304 | { | |
1305 | spin_lock(&era->deferred_lock); | |
1306 | bio_list_add(&era->deferred_bios, bio); | |
1307 | spin_unlock(&era->deferred_lock); | |
1308 | ||
1309 | wake_worker(era); | |
1310 | } | |
1311 | ||
1312 | /* | |
1313 | * Make an rpc call to the worker to change the metadata. | |
1314 | */ | |
1315 | static int perform_rpc(struct era *era, struct rpc *rpc) | |
1316 | { | |
1317 | rpc->result = 0; | |
1318 | init_completion(&rpc->complete); | |
1319 | ||
1320 | spin_lock(&era->rpc_lock); | |
1321 | list_add(&rpc->list, &era->rpc_calls); | |
1322 | spin_unlock(&era->rpc_lock); | |
1323 | ||
1324 | wake_worker(era); | |
1325 | wait_for_completion(&rpc->complete); | |
1326 | ||
1327 | return rpc->result; | |
1328 | } | |
1329 | ||
1330 | static int in_worker0(struct era *era, int (*fn)(struct era_metadata *)) | |
1331 | { | |
1332 | struct rpc rpc; | |
1333 | rpc.fn0 = fn; | |
1334 | rpc.fn1 = NULL; | |
1335 | ||
1336 | return perform_rpc(era, &rpc); | |
1337 | } | |
1338 | ||
1339 | static int in_worker1(struct era *era, | |
1340 | int (*fn)(struct era_metadata *, void *), void *arg) | |
1341 | { | |
1342 | struct rpc rpc; | |
1343 | rpc.fn0 = NULL; | |
1344 | rpc.fn1 = fn; | |
1345 | rpc.arg = arg; | |
1346 | ||
1347 | return perform_rpc(era, &rpc); | |
1348 | } | |
1349 | ||
1350 | static void start_worker(struct era *era) | |
1351 | { | |
1352 | atomic_set(&era->suspended, 0); | |
1353 | } | |
1354 | ||
1355 | static void stop_worker(struct era *era) | |
1356 | { | |
1357 | atomic_set(&era->suspended, 1); | |
1358 | flush_workqueue(era->wq); | |
1359 | } | |
1360 | ||
1361 | /*---------------------------------------------------------------- | |
1362 | * Target methods | |
1363 | *--------------------------------------------------------------*/ | |
1364 | static int dev_is_congested(struct dm_dev *dev, int bdi_bits) | |
1365 | { | |
1366 | struct request_queue *q = bdev_get_queue(dev->bdev); | |
1367 | return bdi_congested(&q->backing_dev_info, bdi_bits); | |
1368 | } | |
1369 | ||
1370 | static int era_is_congested(struct dm_target_callbacks *cb, int bdi_bits) | |
1371 | { | |
1372 | struct era *era = container_of(cb, struct era, callbacks); | |
1373 | return dev_is_congested(era->origin_dev, bdi_bits); | |
1374 | } | |
1375 | ||
1376 | static void era_destroy(struct era *era) | |
1377 | { | |
1378 | metadata_close(era->md); | |
1379 | ||
1380 | if (era->wq) | |
1381 | destroy_workqueue(era->wq); | |
1382 | ||
1383 | if (era->origin_dev) | |
1384 | dm_put_device(era->ti, era->origin_dev); | |
1385 | ||
1386 | if (era->metadata_dev) | |
1387 | dm_put_device(era->ti, era->metadata_dev); | |
1388 | ||
1389 | kfree(era); | |
1390 | } | |
1391 | ||
1392 | static dm_block_t calc_nr_blocks(struct era *era) | |
1393 | { | |
1394 | return dm_sector_div_up(era->ti->len, era->sectors_per_block); | |
1395 | } | |
1396 | ||
1397 | static bool valid_block_size(dm_block_t block_size) | |
1398 | { | |
1399 | bool greater_than_zero = block_size > 0; | |
1400 | bool multiple_of_min_block_size = (block_size & (MIN_BLOCK_SIZE - 1)) == 0; | |
1401 | ||
1402 | return greater_than_zero && multiple_of_min_block_size; | |
1403 | } | |
1404 | ||
1405 | /* | |
1406 | * <metadata dev> <data dev> <data block size (sectors)> | |
1407 | */ | |
1408 | static int era_ctr(struct dm_target *ti, unsigned argc, char **argv) | |
1409 | { | |
1410 | int r; | |
1411 | char dummy; | |
1412 | struct era *era; | |
1413 | struct era_metadata *md; | |
1414 | ||
1415 | if (argc != 3) { | |
1416 | ti->error = "Invalid argument count"; | |
1417 | return -EINVAL; | |
1418 | } | |
1419 | ||
1420 | era = kzalloc(sizeof(*era), GFP_KERNEL); | |
1421 | if (!era) { | |
1422 | ti->error = "Error allocating era structure"; | |
1423 | return -ENOMEM; | |
1424 | } | |
1425 | ||
1426 | era->ti = ti; | |
1427 | ||
1428 | r = dm_get_device(ti, argv[0], FMODE_READ | FMODE_WRITE, &era->metadata_dev); | |
1429 | if (r) { | |
1430 | ti->error = "Error opening metadata device"; | |
1431 | era_destroy(era); | |
1432 | return -EINVAL; | |
1433 | } | |
1434 | ||
1435 | r = dm_get_device(ti, argv[1], FMODE_READ | FMODE_WRITE, &era->origin_dev); | |
1436 | if (r) { | |
1437 | ti->error = "Error opening data device"; | |
1438 | era_destroy(era); | |
1439 | return -EINVAL; | |
1440 | } | |
1441 | ||
1442 | r = sscanf(argv[2], "%u%c", &era->sectors_per_block, &dummy); | |
1443 | if (r != 1) { | |
1444 | ti->error = "Error parsing block size"; | |
1445 | era_destroy(era); | |
1446 | return -EINVAL; | |
1447 | } | |
1448 | ||
1449 | r = dm_set_target_max_io_len(ti, era->sectors_per_block); | |
1450 | if (r) { | |
1451 | ti->error = "could not set max io len"; | |
1452 | era_destroy(era); | |
1453 | return -EINVAL; | |
1454 | } | |
1455 | ||
1456 | if (!valid_block_size(era->sectors_per_block)) { | |
1457 | ti->error = "Invalid block size"; | |
1458 | era_destroy(era); | |
1459 | return -EINVAL; | |
1460 | } | |
1461 | if (era->sectors_per_block & (era->sectors_per_block - 1)) | |
1462 | era->sectors_per_block_shift = -1; | |
1463 | else | |
1464 | era->sectors_per_block_shift = __ffs(era->sectors_per_block); | |
1465 | ||
1466 | md = metadata_open(era->metadata_dev->bdev, era->sectors_per_block, true); | |
1467 | if (IS_ERR(md)) { | |
1468 | ti->error = "Error reading metadata"; | |
1469 | era_destroy(era); | |
1470 | return PTR_ERR(md); | |
1471 | } | |
1472 | era->md = md; | |
1473 | ||
1474 | era->nr_blocks = calc_nr_blocks(era); | |
1475 | ||
1476 | r = metadata_resize(era->md, &era->nr_blocks); | |
1477 | if (r) { | |
1478 | ti->error = "couldn't resize metadata"; | |
1479 | era_destroy(era); | |
1480 | return -ENOMEM; | |
1481 | } | |
1482 | ||
1483 | era->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM); | |
1484 | if (!era->wq) { | |
1485 | ti->error = "could not create workqueue for metadata object"; | |
1486 | era_destroy(era); | |
1487 | return -ENOMEM; | |
1488 | } | |
1489 | INIT_WORK(&era->worker, do_work); | |
1490 | ||
1491 | spin_lock_init(&era->deferred_lock); | |
1492 | bio_list_init(&era->deferred_bios); | |
1493 | ||
1494 | spin_lock_init(&era->rpc_lock); | |
1495 | INIT_LIST_HEAD(&era->rpc_calls); | |
1496 | ||
1497 | ti->private = era; | |
1498 | ti->num_flush_bios = 1; | |
1499 | ti->flush_supported = true; | |
1500 | ||
1501 | ti->num_discard_bios = 1; | |
1502 | ti->discards_supported = true; | |
1503 | era->callbacks.congested_fn = era_is_congested; | |
1504 | dm_table_add_target_callbacks(ti->table, &era->callbacks); | |
1505 | ||
1506 | return 0; | |
1507 | } | |
1508 | ||
1509 | static void era_dtr(struct dm_target *ti) | |
1510 | { | |
1511 | era_destroy(ti->private); | |
1512 | } | |
1513 | ||
1514 | static int era_map(struct dm_target *ti, struct bio *bio) | |
1515 | { | |
1516 | struct era *era = ti->private; | |
1517 | dm_block_t block = get_block(era, bio); | |
1518 | ||
1519 | /* | |
1520 | * All bios get remapped to the origin device. We do this now, but | |
1521 | * it may not get issued until later. Depending on whether the | |
1522 | * block is marked in this era. | |
1523 | */ | |
1524 | remap_to_origin(era, bio); | |
1525 | ||
1526 | /* | |
1527 | * REQ_FLUSH bios carry no data, so we're not interested in them. | |
1528 | */ | |
1529 | if (!(bio->bi_rw & REQ_FLUSH) && | |
1530 | (bio_data_dir(bio) == WRITE) && | |
1531 | !metadata_current_marked(era->md, block)) { | |
1532 | defer_bio(era, bio); | |
1533 | return DM_MAPIO_SUBMITTED; | |
1534 | } | |
1535 | ||
1536 | return DM_MAPIO_REMAPPED; | |
1537 | } | |
1538 | ||
1539 | static void era_postsuspend(struct dm_target *ti) | |
1540 | { | |
1541 | int r; | |
1542 | struct era *era = ti->private; | |
1543 | ||
1544 | r = in_worker0(era, metadata_era_archive); | |
1545 | if (r) { | |
1546 | DMERR("%s: couldn't archive current era", __func__); | |
1547 | /* FIXME: fail mode */ | |
1548 | } | |
1549 | ||
1550 | stop_worker(era); | |
1551 | } | |
1552 | ||
1553 | static int era_preresume(struct dm_target *ti) | |
1554 | { | |
1555 | int r; | |
1556 | struct era *era = ti->private; | |
1557 | dm_block_t new_size = calc_nr_blocks(era); | |
1558 | ||
1559 | if (era->nr_blocks != new_size) { | |
1560 | r = in_worker1(era, metadata_resize, &new_size); | |
1561 | if (r) | |
1562 | return r; | |
1563 | ||
1564 | era->nr_blocks = new_size; | |
1565 | } | |
1566 | ||
1567 | start_worker(era); | |
1568 | ||
1569 | r = in_worker0(era, metadata_new_era); | |
1570 | if (r) { | |
1571 | DMERR("%s: metadata_era_rollover failed", __func__); | |
1572 | return r; | |
1573 | } | |
1574 | ||
1575 | return 0; | |
1576 | } | |
1577 | ||
1578 | /* | |
1579 | * Status format: | |
1580 | * | |
1581 | * <metadata block size> <#used metadata blocks>/<#total metadata blocks> | |
1582 | * <current era> <held metadata root | '-'> | |
1583 | */ | |
1584 | static void era_status(struct dm_target *ti, status_type_t type, | |
1585 | unsigned status_flags, char *result, unsigned maxlen) | |
1586 | { | |
1587 | int r; | |
1588 | struct era *era = ti->private; | |
1589 | ssize_t sz = 0; | |
1590 | struct metadata_stats stats; | |
1591 | char buf[BDEVNAME_SIZE]; | |
1592 | ||
1593 | switch (type) { | |
1594 | case STATUSTYPE_INFO: | |
1595 | r = in_worker1(era, metadata_get_stats, &stats); | |
1596 | if (r) | |
1597 | goto err; | |
1598 | ||
1599 | DMEMIT("%u %llu/%llu %u", | |
1600 | (unsigned) (DM_ERA_METADATA_BLOCK_SIZE >> SECTOR_SHIFT), | |
1601 | (unsigned long long) stats.used, | |
1602 | (unsigned long long) stats.total, | |
1603 | (unsigned) stats.era); | |
1604 | ||
1605 | if (stats.snap != SUPERBLOCK_LOCATION) | |
1606 | DMEMIT(" %llu", stats.snap); | |
1607 | else | |
1608 | DMEMIT(" -"); | |
1609 | break; | |
1610 | ||
1611 | case STATUSTYPE_TABLE: | |
1612 | format_dev_t(buf, era->metadata_dev->bdev->bd_dev); | |
1613 | DMEMIT("%s ", buf); | |
1614 | format_dev_t(buf, era->origin_dev->bdev->bd_dev); | |
1615 | DMEMIT("%s %u", buf, era->sectors_per_block); | |
1616 | break; | |
1617 | } | |
1618 | ||
1619 | return; | |
1620 | ||
1621 | err: | |
1622 | DMEMIT("Error"); | |
1623 | } | |
1624 | ||
1625 | static int era_message(struct dm_target *ti, unsigned argc, char **argv) | |
1626 | { | |
1627 | struct era *era = ti->private; | |
1628 | ||
1629 | if (argc != 1) { | |
1630 | DMERR("incorrect number of message arguments"); | |
1631 | return -EINVAL; | |
1632 | } | |
1633 | ||
1634 | if (!strcasecmp(argv[0], "checkpoint")) | |
1635 | return in_worker0(era, metadata_checkpoint); | |
1636 | ||
1637 | if (!strcasecmp(argv[0], "take_metadata_snap")) | |
1638 | return in_worker0(era, metadata_take_snap); | |
1639 | ||
1640 | if (!strcasecmp(argv[0], "drop_metadata_snap")) | |
1641 | return in_worker0(era, metadata_drop_snap); | |
1642 | ||
1643 | DMERR("unsupported message '%s'", argv[0]); | |
1644 | return -EINVAL; | |
1645 | } | |
1646 | ||
1647 | static sector_t get_dev_size(struct dm_dev *dev) | |
1648 | { | |
1649 | return i_size_read(dev->bdev->bd_inode) >> SECTOR_SHIFT; | |
1650 | } | |
1651 | ||
1652 | static int era_iterate_devices(struct dm_target *ti, | |
1653 | iterate_devices_callout_fn fn, void *data) | |
1654 | { | |
1655 | struct era *era = ti->private; | |
1656 | return fn(ti, era->origin_dev, 0, get_dev_size(era->origin_dev), data); | |
1657 | } | |
1658 | ||
1659 | static int era_merge(struct dm_target *ti, struct bvec_merge_data *bvm, | |
1660 | struct bio_vec *biovec, int max_size) | |
1661 | { | |
1662 | struct era *era = ti->private; | |
1663 | struct request_queue *q = bdev_get_queue(era->origin_dev->bdev); | |
1664 | ||
1665 | if (!q->merge_bvec_fn) | |
1666 | return max_size; | |
1667 | ||
1668 | bvm->bi_bdev = era->origin_dev->bdev; | |
1669 | ||
1670 | return min(max_size, q->merge_bvec_fn(q, bvm, biovec)); | |
1671 | } | |
1672 | ||
1673 | static void era_io_hints(struct dm_target *ti, struct queue_limits *limits) | |
1674 | { | |
1675 | struct era *era = ti->private; | |
1676 | uint64_t io_opt_sectors = limits->io_opt >> SECTOR_SHIFT; | |
1677 | ||
1678 | /* | |
1679 | * If the system-determined stacked limits are compatible with the | |
1680 | * era device's blocksize (io_opt is a factor) do not override them. | |
1681 | */ | |
1682 | if (io_opt_sectors < era->sectors_per_block || | |
1683 | do_div(io_opt_sectors, era->sectors_per_block)) { | |
1684 | blk_limits_io_min(limits, 0); | |
1685 | blk_limits_io_opt(limits, era->sectors_per_block << SECTOR_SHIFT); | |
1686 | } | |
1687 | } | |
1688 | ||
1689 | /*----------------------------------------------------------------*/ | |
1690 | ||
1691 | static struct target_type era_target = { | |
1692 | .name = "era", | |
1693 | .version = {1, 0, 0}, | |
1694 | .module = THIS_MODULE, | |
1695 | .ctr = era_ctr, | |
1696 | .dtr = era_dtr, | |
1697 | .map = era_map, | |
1698 | .postsuspend = era_postsuspend, | |
1699 | .preresume = era_preresume, | |
1700 | .status = era_status, | |
1701 | .message = era_message, | |
1702 | .iterate_devices = era_iterate_devices, | |
1703 | .merge = era_merge, | |
1704 | .io_hints = era_io_hints | |
1705 | }; | |
1706 | ||
1707 | static int __init dm_era_init(void) | |
1708 | { | |
1709 | int r; | |
1710 | ||
1711 | r = dm_register_target(&era_target); | |
1712 | if (r) { | |
1713 | DMERR("era target registration failed: %d", r); | |
1714 | return r; | |
1715 | } | |
1716 | ||
1717 | return 0; | |
1718 | } | |
1719 | ||
1720 | static void __exit dm_era_exit(void) | |
1721 | { | |
1722 | dm_unregister_target(&era_target); | |
1723 | } | |
1724 | ||
1725 | module_init(dm_era_init); | |
1726 | module_exit(dm_era_exit); | |
1727 | ||
1728 | MODULE_DESCRIPTION(DM_NAME " era target"); | |
1729 | MODULE_AUTHOR("Joe Thornber <ejt@redhat.com>"); | |
1730 | MODULE_LICENSE("GPL"); |