Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Copyright (C) 2003 Sistina Software | |
b7fd54a7 | 3 | * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved. |
1da177e4 LT |
4 | * |
5 | * This file is released under the LGPL. | |
6 | */ | |
7 | ||
8 | #include <linux/init.h> | |
9 | #include <linux/slab.h> | |
10 | #include <linux/module.h> | |
11 | #include <linux/vmalloc.h> | |
a765e20e AK |
12 | #include <linux/dm-io.h> |
13 | #include <linux/dm-dirty-log.h> | |
1da177e4 | 14 | |
586e80e6 | 15 | #include <linux/device-mapper.h> |
1da177e4 | 16 | |
b7fd54a7 | 17 | #define DM_MSG_PREFIX "dirty region log" |
72d94861 | 18 | |
2a23aa1d JB |
19 | struct dm_dirty_log_internal { |
20 | struct dm_dirty_log_type *type; | |
21 | ||
22 | struct list_head list; | |
23 | long use; | |
24 | }; | |
25 | ||
1da177e4 LT |
26 | static LIST_HEAD(_log_types); |
27 | static DEFINE_SPINLOCK(_lock); | |
28 | ||
2a23aa1d | 29 | static struct dm_dirty_log_internal *__find_dirty_log_type(const char *name) |
1da177e4 | 30 | { |
2a23aa1d JB |
31 | struct dm_dirty_log_internal *log_type; |
32 | ||
33 | list_for_each_entry(log_type, &_log_types, list) | |
34 | if (!strcmp(name, log_type->type->name)) | |
35 | return log_type; | |
36 | ||
37 | return NULL; | |
38 | } | |
39 | ||
40 | static struct dm_dirty_log_internal *_get_dirty_log_type(const char *name) | |
41 | { | |
42 | struct dm_dirty_log_internal *log_type; | |
1da177e4 LT |
43 | |
44 | spin_lock(&_lock); | |
2a23aa1d JB |
45 | |
46 | log_type = __find_dirty_log_type(name); | |
47 | if (log_type) { | |
48 | if (!log_type->use && !try_module_get(log_type->type->module)) | |
49 | log_type = NULL; | |
50 | else | |
51 | log_type->use++; | |
52 | } | |
1da177e4 LT |
53 | |
54 | spin_unlock(&_lock); | |
2a23aa1d JB |
55 | |
56 | return log_type; | |
1da177e4 LT |
57 | } |
58 | ||
fb8b2848 JB |
59 | /* |
60 | * get_type | |
61 | * @type_name | |
62 | * | |
2a23aa1d | 63 | * Attempt to retrieve the dm_dirty_log_type by name. If not already |
fb8b2848 JB |
64 | * available, attempt to load the appropriate module. |
65 | * | |
66 | * Log modules are named "dm-log-" followed by the 'type_name'. | |
67 | * Modules may contain multiple types. | |
68 | * This function will first try the module "dm-log-<type_name>", | |
69 | * then truncate 'type_name' on the last '-' and try again. | |
70 | * | |
71 | * For example, if type_name was "clustered-disk", it would search | |
72 | * 'dm-log-clustered-disk' then 'dm-log-clustered'. | |
73 | * | |
74 | * Returns: dirty_log_type* on success, NULL on failure | |
75 | */ | |
416cd17b | 76 | static struct dm_dirty_log_type *get_type(const char *type_name) |
fb8b2848 JB |
77 | { |
78 | char *p, *type_name_dup; | |
2a23aa1d | 79 | struct dm_dirty_log_internal *log_type; |
fb8b2848 | 80 | |
2a23aa1d JB |
81 | if (!type_name) |
82 | return NULL; | |
83 | ||
84 | log_type = _get_dirty_log_type(type_name); | |
85 | if (log_type) | |
86 | return log_type->type; | |
fb8b2848 JB |
87 | |
88 | type_name_dup = kstrdup(type_name, GFP_KERNEL); | |
89 | if (!type_name_dup) { | |
90 | DMWARN("No memory left to attempt log module load for \"%s\"", | |
91 | type_name); | |
92 | return NULL; | |
93 | } | |
94 | ||
95 | while (request_module("dm-log-%s", type_name_dup) || | |
2a23aa1d | 96 | !(log_type = _get_dirty_log_type(type_name))) { |
fb8b2848 JB |
97 | p = strrchr(type_name_dup, '-'); |
98 | if (!p) | |
99 | break; | |
100 | p[0] = '\0'; | |
101 | } | |
102 | ||
2a23aa1d | 103 | if (!log_type) |
fb8b2848 JB |
104 | DMWARN("Module for logging type \"%s\" not found.", type_name); |
105 | ||
106 | kfree(type_name_dup); | |
107 | ||
2a23aa1d | 108 | return log_type ? log_type->type : NULL; |
fb8b2848 JB |
109 | } |
110 | ||
416cd17b | 111 | static void put_type(struct dm_dirty_log_type *type) |
1da177e4 | 112 | { |
2a23aa1d JB |
113 | struct dm_dirty_log_internal *log_type; |
114 | ||
115 | if (!type) | |
116 | return; | |
117 | ||
1da177e4 | 118 | spin_lock(&_lock); |
2a23aa1d JB |
119 | log_type = __find_dirty_log_type(type->name); |
120 | if (!log_type) | |
121 | goto out; | |
122 | ||
123 | if (!--log_type->use) | |
1da177e4 | 124 | module_put(type->module); |
2a23aa1d JB |
125 | |
126 | BUG_ON(log_type->use < 0); | |
127 | ||
128 | out: | |
1da177e4 LT |
129 | spin_unlock(&_lock); |
130 | } | |
131 | ||
2a23aa1d JB |
132 | static struct dm_dirty_log_internal *_alloc_dirty_log_type(struct dm_dirty_log_type *type) |
133 | { | |
134 | struct dm_dirty_log_internal *log_type = kzalloc(sizeof(*log_type), | |
135 | GFP_KERNEL); | |
136 | ||
137 | if (log_type) | |
138 | log_type->type = type; | |
139 | ||
140 | return log_type; | |
141 | } | |
142 | ||
b8206bc3 AK |
143 | int dm_dirty_log_type_register(struct dm_dirty_log_type *type) |
144 | { | |
2a23aa1d JB |
145 | struct dm_dirty_log_internal *log_type = _alloc_dirty_log_type(type); |
146 | int r = 0; | |
147 | ||
148 | if (!log_type) | |
149 | return -ENOMEM; | |
150 | ||
b8206bc3 | 151 | spin_lock(&_lock); |
2a23aa1d JB |
152 | if (!__find_dirty_log_type(type->name)) |
153 | list_add(&log_type->list, &_log_types); | |
154 | else { | |
155 | kfree(log_type); | |
156 | r = -EEXIST; | |
157 | } | |
b8206bc3 AK |
158 | spin_unlock(&_lock); |
159 | ||
2a23aa1d | 160 | return r; |
b8206bc3 AK |
161 | } |
162 | EXPORT_SYMBOL(dm_dirty_log_type_register); | |
163 | ||
164 | int dm_dirty_log_type_unregister(struct dm_dirty_log_type *type) | |
165 | { | |
2a23aa1d JB |
166 | struct dm_dirty_log_internal *log_type; |
167 | ||
b8206bc3 AK |
168 | spin_lock(&_lock); |
169 | ||
2a23aa1d JB |
170 | log_type = __find_dirty_log_type(type->name); |
171 | if (!log_type) { | |
172 | spin_unlock(&_lock); | |
173 | return -EINVAL; | |
174 | } | |
175 | ||
176 | if (log_type->use) { | |
177 | spin_unlock(&_lock); | |
178 | return -ETXTBSY; | |
179 | } | |
180 | ||
181 | list_del(&log_type->list); | |
b8206bc3 AK |
182 | |
183 | spin_unlock(&_lock); | |
2a23aa1d | 184 | kfree(log_type); |
b8206bc3 AK |
185 | |
186 | return 0; | |
187 | } | |
188 | EXPORT_SYMBOL(dm_dirty_log_type_unregister); | |
189 | ||
416cd17b HM |
190 | struct dm_dirty_log *dm_dirty_log_create(const char *type_name, |
191 | struct dm_target *ti, | |
192 | unsigned int argc, char **argv) | |
1da177e4 | 193 | { |
416cd17b HM |
194 | struct dm_dirty_log_type *type; |
195 | struct dm_dirty_log *log; | |
1da177e4 LT |
196 | |
197 | log = kmalloc(sizeof(*log), GFP_KERNEL); | |
198 | if (!log) | |
199 | return NULL; | |
200 | ||
201 | type = get_type(type_name); | |
202 | if (!type) { | |
203 | kfree(log); | |
204 | return NULL; | |
205 | } | |
206 | ||
207 | log->type = type; | |
208 | if (type->ctr(log, ti, argc, argv)) { | |
209 | kfree(log); | |
210 | put_type(type); | |
211 | return NULL; | |
212 | } | |
213 | ||
214 | return log; | |
215 | } | |
416cd17b | 216 | EXPORT_SYMBOL(dm_dirty_log_create); |
1da177e4 | 217 | |
416cd17b | 218 | void dm_dirty_log_destroy(struct dm_dirty_log *log) |
1da177e4 LT |
219 | { |
220 | log->type->dtr(log); | |
221 | put_type(log->type); | |
222 | kfree(log); | |
223 | } | |
416cd17b | 224 | EXPORT_SYMBOL(dm_dirty_log_destroy); |
1da177e4 LT |
225 | |
226 | /*----------------------------------------------------------------- | |
227 | * Persistent and core logs share a lot of their implementation. | |
228 | * FIXME: need a reload method to be called from a resume | |
229 | *---------------------------------------------------------------*/ | |
230 | /* | |
231 | * Magic for persistent mirrors: "MiRr" | |
232 | */ | |
233 | #define MIRROR_MAGIC 0x4D695272 | |
234 | ||
235 | /* | |
236 | * The on-disk version of the metadata. | |
237 | */ | |
a4fc4717 | 238 | #define MIRROR_DISK_VERSION 2 |
1da177e4 LT |
239 | #define LOG_OFFSET 2 |
240 | ||
241 | struct log_header { | |
242 | uint32_t magic; | |
243 | ||
244 | /* | |
245 | * Simple, incrementing version. no backward | |
246 | * compatibility. | |
247 | */ | |
248 | uint32_t version; | |
249 | sector_t nr_regions; | |
250 | }; | |
251 | ||
252 | struct log_c { | |
253 | struct dm_target *ti; | |
254 | int touched; | |
255 | uint32_t region_size; | |
256 | unsigned int region_count; | |
257 | region_t sync_count; | |
258 | ||
259 | unsigned bitset_uint32_count; | |
260 | uint32_t *clean_bits; | |
261 | uint32_t *sync_bits; | |
262 | uint32_t *recovering_bits; /* FIXME: this seems excessive */ | |
263 | ||
264 | int sync_search; | |
265 | ||
266 | /* Resync flag */ | |
267 | enum sync { | |
268 | DEFAULTSYNC, /* Synchronize if necessary */ | |
269 | NOSYNC, /* Devices known to be already in sync */ | |
270 | FORCESYNC, /* Force a sync to happen */ | |
271 | } sync; | |
272 | ||
5d234d1e HM |
273 | struct dm_io_request io_req; |
274 | ||
1da177e4 LT |
275 | /* |
276 | * Disk log fields | |
277 | */ | |
01d03a66 | 278 | int log_dev_failed; |
1da177e4 LT |
279 | struct dm_dev *log_dev; |
280 | struct log_header header; | |
281 | ||
22a1ceb1 | 282 | struct dm_io_region header_location; |
1da177e4 | 283 | struct log_header *disk_header; |
1da177e4 LT |
284 | }; |
285 | ||
286 | /* | |
287 | * The touched member needs to be updated every time we access | |
288 | * one of the bitsets. | |
289 | */ | |
416cd17b | 290 | static inline int log_test_bit(uint32_t *bs, unsigned bit) |
1da177e4 | 291 | { |
a4fc4717 | 292 | return ext2_test_bit(bit, (unsigned long *) bs) ? 1 : 0; |
1da177e4 LT |
293 | } |
294 | ||
295 | static inline void log_set_bit(struct log_c *l, | |
296 | uint32_t *bs, unsigned bit) | |
297 | { | |
a4fc4717 | 298 | ext2_set_bit(bit, (unsigned long *) bs); |
1da177e4 LT |
299 | l->touched = 1; |
300 | } | |
301 | ||
302 | static inline void log_clear_bit(struct log_c *l, | |
303 | uint32_t *bs, unsigned bit) | |
304 | { | |
a4fc4717 | 305 | ext2_clear_bit(bit, (unsigned long *) bs); |
1da177e4 LT |
306 | l->touched = 1; |
307 | } | |
308 | ||
309 | /*---------------------------------------------------------------- | |
310 | * Header IO | |
311 | *--------------------------------------------------------------*/ | |
312 | static void header_to_disk(struct log_header *core, struct log_header *disk) | |
313 | { | |
314 | disk->magic = cpu_to_le32(core->magic); | |
315 | disk->version = cpu_to_le32(core->version); | |
316 | disk->nr_regions = cpu_to_le64(core->nr_regions); | |
317 | } | |
318 | ||
319 | static void header_from_disk(struct log_header *core, struct log_header *disk) | |
320 | { | |
321 | core->magic = le32_to_cpu(disk->magic); | |
322 | core->version = le32_to_cpu(disk->version); | |
323 | core->nr_regions = le64_to_cpu(disk->nr_regions); | |
324 | } | |
325 | ||
5d234d1e HM |
326 | static int rw_header(struct log_c *lc, int rw) |
327 | { | |
328 | lc->io_req.bi_rw = rw; | |
329 | lc->io_req.mem.ptr.vma = lc->disk_header; | |
330 | lc->io_req.notify.fn = NULL; | |
331 | ||
332 | return dm_io(&lc->io_req, 1, &lc->header_location, NULL); | |
333 | } | |
334 | ||
1da177e4 LT |
335 | static int read_header(struct log_c *log) |
336 | { | |
337 | int r; | |
1da177e4 | 338 | |
5d234d1e | 339 | r = rw_header(log, READ); |
1da177e4 LT |
340 | if (r) |
341 | return r; | |
342 | ||
343 | header_from_disk(&log->header, log->disk_header); | |
344 | ||
345 | /* New log required? */ | |
346 | if (log->sync != DEFAULTSYNC || log->header.magic != MIRROR_MAGIC) { | |
347 | log->header.magic = MIRROR_MAGIC; | |
348 | log->header.version = MIRROR_DISK_VERSION; | |
349 | log->header.nr_regions = 0; | |
350 | } | |
351 | ||
a4fc4717 PC |
352 | #ifdef __LITTLE_ENDIAN |
353 | if (log->header.version == 1) | |
354 | log->header.version = 2; | |
355 | #endif | |
356 | ||
1da177e4 LT |
357 | if (log->header.version != MIRROR_DISK_VERSION) { |
358 | DMWARN("incompatible disk log version"); | |
359 | return -EINVAL; | |
360 | } | |
361 | ||
362 | return 0; | |
363 | } | |
364 | ||
365 | static inline int write_header(struct log_c *log) | |
366 | { | |
1da177e4 | 367 | header_to_disk(&log->header, log->disk_header); |
5d234d1e | 368 | return rw_header(log, WRITE); |
1da177e4 LT |
369 | } |
370 | ||
1da177e4 LT |
371 | /*---------------------------------------------------------------- |
372 | * core log constructor/destructor | |
373 | * | |
374 | * argv contains region_size followed optionally by [no]sync | |
375 | *--------------------------------------------------------------*/ | |
376 | #define BYTE_SHIFT 3 | |
416cd17b | 377 | static int create_log_context(struct dm_dirty_log *log, struct dm_target *ti, |
b7cca195 AK |
378 | unsigned int argc, char **argv, |
379 | struct dm_dev *dev) | |
1da177e4 LT |
380 | { |
381 | enum sync sync = DEFAULTSYNC; | |
382 | ||
383 | struct log_c *lc; | |
384 | uint32_t region_size; | |
385 | unsigned int region_count; | |
b7cca195 | 386 | size_t bitset_size, buf_size; |
5d234d1e | 387 | int r; |
1da177e4 LT |
388 | |
389 | if (argc < 1 || argc > 2) { | |
b7fd54a7 | 390 | DMWARN("wrong number of arguments to dirty region log"); |
1da177e4 LT |
391 | return -EINVAL; |
392 | } | |
393 | ||
394 | if (argc > 1) { | |
395 | if (!strcmp(argv[1], "sync")) | |
396 | sync = FORCESYNC; | |
397 | else if (!strcmp(argv[1], "nosync")) | |
398 | sync = NOSYNC; | |
399 | else { | |
b7fd54a7 HM |
400 | DMWARN("unrecognised sync argument to " |
401 | "dirty region log: %s", argv[1]); | |
1da177e4 LT |
402 | return -EINVAL; |
403 | } | |
404 | } | |
405 | ||
406 | if (sscanf(argv[0], "%u", ®ion_size) != 1) { | |
407 | DMWARN("invalid region size string"); | |
408 | return -EINVAL; | |
409 | } | |
410 | ||
411 | region_count = dm_sector_div_up(ti->len, region_size); | |
412 | ||
413 | lc = kmalloc(sizeof(*lc), GFP_KERNEL); | |
414 | if (!lc) { | |
415 | DMWARN("couldn't allocate core log"); | |
416 | return -ENOMEM; | |
417 | } | |
418 | ||
419 | lc->ti = ti; | |
420 | lc->touched = 0; | |
421 | lc->region_size = region_size; | |
422 | lc->region_count = region_count; | |
423 | lc->sync = sync; | |
424 | ||
425 | /* | |
0e56822d | 426 | * Work out how many "unsigned long"s we need to hold the bitset. |
1da177e4 LT |
427 | */ |
428 | bitset_size = dm_round_up(region_count, | |
29121bd0 | 429 | sizeof(*lc->clean_bits) << BYTE_SHIFT); |
1da177e4 LT |
430 | bitset_size >>= BYTE_SHIFT; |
431 | ||
29121bd0 | 432 | lc->bitset_uint32_count = bitset_size / sizeof(*lc->clean_bits); |
b7cca195 AK |
433 | |
434 | /* | |
435 | * Disk log? | |
436 | */ | |
437 | if (!dev) { | |
438 | lc->clean_bits = vmalloc(bitset_size); | |
439 | if (!lc->clean_bits) { | |
440 | DMWARN("couldn't allocate clean bitset"); | |
441 | kfree(lc); | |
442 | return -ENOMEM; | |
443 | } | |
444 | lc->disk_header = NULL; | |
445 | } else { | |
446 | lc->log_dev = dev; | |
01d03a66 | 447 | lc->log_dev_failed = 0; |
b7cca195 AK |
448 | lc->header_location.bdev = lc->log_dev->bdev; |
449 | lc->header_location.sector = 0; | |
450 | ||
451 | /* | |
452 | * Buffer holds both header and bitset. | |
453 | */ | |
454 | buf_size = dm_round_up((LOG_OFFSET << SECTOR_SHIFT) + | |
455 | bitset_size, ti->limits.hardsect_size); | |
456 | lc->header_location.count = buf_size >> SECTOR_SHIFT; | |
5d234d1e HM |
457 | lc->io_req.mem.type = DM_IO_VMA; |
458 | lc->io_req.client = dm_io_client_create(dm_div_up(buf_size, | |
459 | PAGE_SIZE)); | |
460 | if (IS_ERR(lc->io_req.client)) { | |
461 | r = PTR_ERR(lc->io_req.client); | |
462 | DMWARN("couldn't allocate disk io client"); | |
463 | kfree(lc); | |
464 | return -ENOMEM; | |
465 | } | |
b7cca195 AK |
466 | |
467 | lc->disk_header = vmalloc(buf_size); | |
468 | if (!lc->disk_header) { | |
469 | DMWARN("couldn't allocate disk log buffer"); | |
c7a2bd19 | 470 | dm_io_client_destroy(lc->io_req.client); |
b7cca195 AK |
471 | kfree(lc); |
472 | return -ENOMEM; | |
473 | } | |
474 | ||
475 | lc->clean_bits = (void *)lc->disk_header + | |
476 | (LOG_OFFSET << SECTOR_SHIFT); | |
1da177e4 | 477 | } |
b7cca195 | 478 | |
1da177e4 LT |
479 | memset(lc->clean_bits, -1, bitset_size); |
480 | ||
481 | lc->sync_bits = vmalloc(bitset_size); | |
482 | if (!lc->sync_bits) { | |
483 | DMWARN("couldn't allocate sync bitset"); | |
b7cca195 AK |
484 | if (!dev) |
485 | vfree(lc->clean_bits); | |
c7a2bd19 TY |
486 | else |
487 | dm_io_client_destroy(lc->io_req.client); | |
b7cca195 | 488 | vfree(lc->disk_header); |
1da177e4 LT |
489 | kfree(lc); |
490 | return -ENOMEM; | |
491 | } | |
492 | memset(lc->sync_bits, (sync == NOSYNC) ? -1 : 0, bitset_size); | |
493 | lc->sync_count = (sync == NOSYNC) ? region_count : 0; | |
494 | ||
495 | lc->recovering_bits = vmalloc(bitset_size); | |
496 | if (!lc->recovering_bits) { | |
497 | DMWARN("couldn't allocate sync bitset"); | |
498 | vfree(lc->sync_bits); | |
b7cca195 AK |
499 | if (!dev) |
500 | vfree(lc->clean_bits); | |
c7a2bd19 TY |
501 | else |
502 | dm_io_client_destroy(lc->io_req.client); | |
b7cca195 | 503 | vfree(lc->disk_header); |
1da177e4 LT |
504 | kfree(lc); |
505 | return -ENOMEM; | |
506 | } | |
507 | memset(lc->recovering_bits, 0, bitset_size); | |
508 | lc->sync_search = 0; | |
509 | log->context = lc; | |
b7cca195 | 510 | |
1da177e4 LT |
511 | return 0; |
512 | } | |
513 | ||
416cd17b | 514 | static int core_ctr(struct dm_dirty_log *log, struct dm_target *ti, |
b7cca195 AK |
515 | unsigned int argc, char **argv) |
516 | { | |
517 | return create_log_context(log, ti, argc, argv, NULL); | |
518 | } | |
519 | ||
520 | static void destroy_log_context(struct log_c *lc) | |
1da177e4 | 521 | { |
1da177e4 LT |
522 | vfree(lc->sync_bits); |
523 | vfree(lc->recovering_bits); | |
524 | kfree(lc); | |
525 | } | |
526 | ||
416cd17b | 527 | static void core_dtr(struct dm_dirty_log *log) |
b7cca195 AK |
528 | { |
529 | struct log_c *lc = (struct log_c *) log->context; | |
530 | ||
531 | vfree(lc->clean_bits); | |
532 | destroy_log_context(lc); | |
533 | } | |
534 | ||
1da177e4 LT |
535 | /*---------------------------------------------------------------- |
536 | * disk log constructor/destructor | |
537 | * | |
538 | * argv contains log_device region_size followed optionally by [no]sync | |
539 | *--------------------------------------------------------------*/ | |
416cd17b | 540 | static int disk_ctr(struct dm_dirty_log *log, struct dm_target *ti, |
1da177e4 LT |
541 | unsigned int argc, char **argv) |
542 | { | |
543 | int r; | |
1da177e4 LT |
544 | struct dm_dev *dev; |
545 | ||
546 | if (argc < 2 || argc > 3) { | |
b7fd54a7 | 547 | DMWARN("wrong number of arguments to disk dirty region log"); |
1da177e4 LT |
548 | return -EINVAL; |
549 | } | |
550 | ||
551 | r = dm_get_device(ti, argv[0], 0, 0 /* FIXME */, | |
552 | FMODE_READ | FMODE_WRITE, &dev); | |
553 | if (r) | |
554 | return r; | |
555 | ||
b7cca195 | 556 | r = create_log_context(log, ti, argc - 1, argv + 1, dev); |
1da177e4 LT |
557 | if (r) { |
558 | dm_put_device(ti, dev); | |
559 | return r; | |
560 | } | |
561 | ||
1da177e4 | 562 | return 0; |
1da177e4 LT |
563 | } |
564 | ||
416cd17b | 565 | static void disk_dtr(struct dm_dirty_log *log) |
1da177e4 LT |
566 | { |
567 | struct log_c *lc = (struct log_c *) log->context; | |
b7cca195 | 568 | |
1da177e4 LT |
569 | dm_put_device(lc->ti, lc->log_dev); |
570 | vfree(lc->disk_header); | |
5d234d1e | 571 | dm_io_client_destroy(lc->io_req.client); |
b7cca195 | 572 | destroy_log_context(lc); |
1da177e4 LT |
573 | } |
574 | ||
575 | static int count_bits32(uint32_t *addr, unsigned size) | |
576 | { | |
577 | int count = 0, i; | |
578 | ||
579 | for (i = 0; i < size; i++) { | |
580 | count += hweight32(*(addr+i)); | |
581 | } | |
582 | return count; | |
583 | } | |
584 | ||
01d03a66 JB |
585 | static void fail_log_device(struct log_c *lc) |
586 | { | |
587 | if (lc->log_dev_failed) | |
588 | return; | |
589 | ||
590 | lc->log_dev_failed = 1; | |
591 | dm_table_event(lc->ti->table); | |
592 | } | |
593 | ||
416cd17b | 594 | static int disk_resume(struct dm_dirty_log *log) |
1da177e4 LT |
595 | { |
596 | int r; | |
597 | unsigned i; | |
598 | struct log_c *lc = (struct log_c *) log->context; | |
599 | size_t size = lc->bitset_uint32_count * sizeof(uint32_t); | |
600 | ||
601 | /* read the disk header */ | |
602 | r = read_header(lc); | |
01d03a66 | 603 | if (r) { |
b7fd54a7 | 604 | DMWARN("%s: Failed to read header on dirty region log device", |
01d03a66 JB |
605 | lc->log_dev->name); |
606 | fail_log_device(lc); | |
ba8b45ce JB |
607 | /* |
608 | * If the log device cannot be read, we must assume | |
609 | * all regions are out-of-sync. If we simply return | |
610 | * here, the state will be uninitialized and could | |
611 | * lead us to return 'in-sync' status for regions | |
612 | * that are actually 'out-of-sync'. | |
613 | */ | |
614 | lc->header.nr_regions = 0; | |
01d03a66 | 615 | } |
1da177e4 | 616 | |
8a835f11 | 617 | /* set or clear any new bits -- device has grown */ |
1da177e4 LT |
618 | if (lc->sync == NOSYNC) |
619 | for (i = lc->header.nr_regions; i < lc->region_count; i++) | |
620 | /* FIXME: amazingly inefficient */ | |
621 | log_set_bit(lc, lc->clean_bits, i); | |
622 | else | |
623 | for (i = lc->header.nr_regions; i < lc->region_count; i++) | |
624 | /* FIXME: amazingly inefficient */ | |
625 | log_clear_bit(lc, lc->clean_bits, i); | |
626 | ||
8a835f11 AK |
627 | /* clear any old bits -- device has shrunk */ |
628 | for (i = lc->region_count; i % (sizeof(*lc->clean_bits) << BYTE_SHIFT); i++) | |
629 | log_clear_bit(lc, lc->clean_bits, i); | |
630 | ||
1da177e4 LT |
631 | /* copy clean across to sync */ |
632 | memcpy(lc->sync_bits, lc->clean_bits, size); | |
633 | lc->sync_count = count_bits32(lc->clean_bits, lc->bitset_uint32_count); | |
88b20a1a | 634 | lc->sync_search = 0; |
1da177e4 | 635 | |
1da177e4 LT |
636 | /* set the correct number of regions in the header */ |
637 | lc->header.nr_regions = lc->region_count; | |
638 | ||
639 | /* write the new header */ | |
01d03a66 JB |
640 | r = write_header(lc); |
641 | if (r) { | |
b7fd54a7 | 642 | DMWARN("%s: Failed to write header on dirty region log device", |
01d03a66 JB |
643 | lc->log_dev->name); |
644 | fail_log_device(lc); | |
645 | } | |
646 | ||
647 | return r; | |
1da177e4 LT |
648 | } |
649 | ||
416cd17b | 650 | static uint32_t core_get_region_size(struct dm_dirty_log *log) |
1da177e4 LT |
651 | { |
652 | struct log_c *lc = (struct log_c *) log->context; | |
653 | return lc->region_size; | |
654 | } | |
655 | ||
416cd17b | 656 | static int core_resume(struct dm_dirty_log *log) |
88b20a1a JB |
657 | { |
658 | struct log_c *lc = (struct log_c *) log->context; | |
659 | lc->sync_search = 0; | |
660 | return 0; | |
661 | } | |
662 | ||
416cd17b | 663 | static int core_is_clean(struct dm_dirty_log *log, region_t region) |
1da177e4 LT |
664 | { |
665 | struct log_c *lc = (struct log_c *) log->context; | |
666 | return log_test_bit(lc->clean_bits, region); | |
667 | } | |
668 | ||
416cd17b | 669 | static int core_in_sync(struct dm_dirty_log *log, region_t region, int block) |
1da177e4 LT |
670 | { |
671 | struct log_c *lc = (struct log_c *) log->context; | |
672 | return log_test_bit(lc->sync_bits, region); | |
673 | } | |
674 | ||
416cd17b | 675 | static int core_flush(struct dm_dirty_log *log) |
1da177e4 LT |
676 | { |
677 | /* no op */ | |
678 | return 0; | |
679 | } | |
680 | ||
416cd17b | 681 | static int disk_flush(struct dm_dirty_log *log) |
1da177e4 LT |
682 | { |
683 | int r; | |
684 | struct log_c *lc = (struct log_c *) log->context; | |
685 | ||
686 | /* only write if the log has changed */ | |
687 | if (!lc->touched) | |
688 | return 0; | |
689 | ||
702ca6f0 | 690 | r = write_header(lc); |
01d03a66 JB |
691 | if (r) |
692 | fail_log_device(lc); | |
693 | else | |
1da177e4 LT |
694 | lc->touched = 0; |
695 | ||
696 | return r; | |
697 | } | |
698 | ||
416cd17b | 699 | static void core_mark_region(struct dm_dirty_log *log, region_t region) |
1da177e4 LT |
700 | { |
701 | struct log_c *lc = (struct log_c *) log->context; | |
702 | log_clear_bit(lc, lc->clean_bits, region); | |
703 | } | |
704 | ||
416cd17b | 705 | static void core_clear_region(struct dm_dirty_log *log, region_t region) |
1da177e4 LT |
706 | { |
707 | struct log_c *lc = (struct log_c *) log->context; | |
708 | log_set_bit(lc, lc->clean_bits, region); | |
709 | } | |
710 | ||
416cd17b | 711 | static int core_get_resync_work(struct dm_dirty_log *log, region_t *region) |
1da177e4 LT |
712 | { |
713 | struct log_c *lc = (struct log_c *) log->context; | |
714 | ||
715 | if (lc->sync_search >= lc->region_count) | |
716 | return 0; | |
717 | ||
718 | do { | |
1113a7e9 SB |
719 | *region = ext2_find_next_zero_bit( |
720 | (unsigned long *) lc->sync_bits, | |
1da177e4 LT |
721 | lc->region_count, |
722 | lc->sync_search); | |
723 | lc->sync_search = *region + 1; | |
724 | ||
ac81b2ee | 725 | if (*region >= lc->region_count) |
1da177e4 LT |
726 | return 0; |
727 | ||
728 | } while (log_test_bit(lc->recovering_bits, *region)); | |
729 | ||
730 | log_set_bit(lc, lc->recovering_bits, *region); | |
731 | return 1; | |
732 | } | |
733 | ||
416cd17b | 734 | static void core_set_region_sync(struct dm_dirty_log *log, region_t region, |
f3ee6b2f | 735 | int in_sync) |
1da177e4 LT |
736 | { |
737 | struct log_c *lc = (struct log_c *) log->context; | |
738 | ||
739 | log_clear_bit(lc, lc->recovering_bits, region); | |
f3ee6b2f | 740 | if (in_sync) { |
1da177e4 LT |
741 | log_set_bit(lc, lc->sync_bits, region); |
742 | lc->sync_count++; | |
f3ee6b2f JB |
743 | } else if (log_test_bit(lc->sync_bits, region)) { |
744 | lc->sync_count--; | |
745 | log_clear_bit(lc, lc->sync_bits, region); | |
746 | } | |
1da177e4 LT |
747 | } |
748 | ||
416cd17b | 749 | static region_t core_get_sync_count(struct dm_dirty_log *log) |
1da177e4 LT |
750 | { |
751 | struct log_c *lc = (struct log_c *) log->context; | |
752 | ||
753 | return lc->sync_count; | |
754 | } | |
755 | ||
756 | #define DMEMIT_SYNC \ | |
757 | if (lc->sync != DEFAULTSYNC) \ | |
758 | DMEMIT("%ssync ", lc->sync == NOSYNC ? "no" : "") | |
759 | ||
416cd17b | 760 | static int core_status(struct dm_dirty_log *log, status_type_t status, |
1da177e4 LT |
761 | char *result, unsigned int maxlen) |
762 | { | |
763 | int sz = 0; | |
764 | struct log_c *lc = log->context; | |
765 | ||
766 | switch(status) { | |
767 | case STATUSTYPE_INFO: | |
315dcc22 | 768 | DMEMIT("1 %s", log->type->name); |
1da177e4 LT |
769 | break; |
770 | ||
771 | case STATUSTYPE_TABLE: | |
772 | DMEMIT("%s %u %u ", log->type->name, | |
773 | lc->sync == DEFAULTSYNC ? 1 : 2, lc->region_size); | |
774 | DMEMIT_SYNC; | |
775 | } | |
776 | ||
777 | return sz; | |
778 | } | |
779 | ||
416cd17b | 780 | static int disk_status(struct dm_dirty_log *log, status_type_t status, |
1da177e4 LT |
781 | char *result, unsigned int maxlen) |
782 | { | |
783 | int sz = 0; | |
1da177e4 LT |
784 | struct log_c *lc = log->context; |
785 | ||
786 | switch(status) { | |
787 | case STATUSTYPE_INFO: | |
315dcc22 JB |
788 | DMEMIT("3 %s %s %c", log->type->name, lc->log_dev->name, |
789 | lc->log_dev_failed ? 'D' : 'A'); | |
1da177e4 LT |
790 | break; |
791 | ||
792 | case STATUSTYPE_TABLE: | |
1da177e4 | 793 | DMEMIT("%s %u %s %u ", log->type->name, |
315dcc22 | 794 | lc->sync == DEFAULTSYNC ? 2 : 3, lc->log_dev->name, |
1da177e4 LT |
795 | lc->region_size); |
796 | DMEMIT_SYNC; | |
797 | } | |
798 | ||
799 | return sz; | |
800 | } | |
801 | ||
416cd17b | 802 | static struct dm_dirty_log_type _core_type = { |
1da177e4 LT |
803 | .name = "core", |
804 | .module = THIS_MODULE, | |
805 | .ctr = core_ctr, | |
806 | .dtr = core_dtr, | |
88b20a1a | 807 | .resume = core_resume, |
1da177e4 LT |
808 | .get_region_size = core_get_region_size, |
809 | .is_clean = core_is_clean, | |
810 | .in_sync = core_in_sync, | |
811 | .flush = core_flush, | |
812 | .mark_region = core_mark_region, | |
813 | .clear_region = core_clear_region, | |
814 | .get_resync_work = core_get_resync_work, | |
f3ee6b2f | 815 | .set_region_sync = core_set_region_sync, |
1da177e4 LT |
816 | .get_sync_count = core_get_sync_count, |
817 | .status = core_status, | |
818 | }; | |
819 | ||
416cd17b | 820 | static struct dm_dirty_log_type _disk_type = { |
1da177e4 LT |
821 | .name = "disk", |
822 | .module = THIS_MODULE, | |
823 | .ctr = disk_ctr, | |
824 | .dtr = disk_dtr, | |
6b3df0d7 | 825 | .postsuspend = disk_flush, |
1da177e4 LT |
826 | .resume = disk_resume, |
827 | .get_region_size = core_get_region_size, | |
828 | .is_clean = core_is_clean, | |
829 | .in_sync = core_in_sync, | |
830 | .flush = disk_flush, | |
831 | .mark_region = core_mark_region, | |
832 | .clear_region = core_clear_region, | |
833 | .get_resync_work = core_get_resync_work, | |
f3ee6b2f | 834 | .set_region_sync = core_set_region_sync, |
1da177e4 LT |
835 | .get_sync_count = core_get_sync_count, |
836 | .status = disk_status, | |
837 | }; | |
838 | ||
c8da2f8d | 839 | static int __init dm_dirty_log_init(void) |
1da177e4 LT |
840 | { |
841 | int r; | |
842 | ||
416cd17b | 843 | r = dm_dirty_log_type_register(&_core_type); |
1da177e4 LT |
844 | if (r) |
845 | DMWARN("couldn't register core log"); | |
846 | ||
416cd17b | 847 | r = dm_dirty_log_type_register(&_disk_type); |
1da177e4 LT |
848 | if (r) { |
849 | DMWARN("couldn't register disk type"); | |
416cd17b | 850 | dm_dirty_log_type_unregister(&_core_type); |
1da177e4 LT |
851 | } |
852 | ||
853 | return r; | |
854 | } | |
855 | ||
c8da2f8d | 856 | static void __exit dm_dirty_log_exit(void) |
1da177e4 | 857 | { |
416cd17b HM |
858 | dm_dirty_log_type_unregister(&_disk_type); |
859 | dm_dirty_log_type_unregister(&_core_type); | |
1da177e4 LT |
860 | } |
861 | ||
769aef30 HM |
862 | module_init(dm_dirty_log_init); |
863 | module_exit(dm_dirty_log_exit); | |
864 | ||
865 | MODULE_DESCRIPTION(DM_NAME " dirty region log"); | |
866 | MODULE_AUTHOR("Joe Thornber, Heinz Mauelshagen <dm-devel@redhat.com>"); | |
867 | MODULE_LICENSE("GPL"); |