]>
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; | |
2a23aa1d JB |
23 | }; |
24 | ||
1da177e4 LT |
25 | static LIST_HEAD(_log_types); |
26 | static DEFINE_SPINLOCK(_lock); | |
27 | ||
2a23aa1d | 28 | static struct dm_dirty_log_internal *__find_dirty_log_type(const char *name) |
1da177e4 | 29 | { |
2a23aa1d JB |
30 | struct dm_dirty_log_internal *log_type; |
31 | ||
32 | list_for_each_entry(log_type, &_log_types, list) | |
33 | if (!strcmp(name, log_type->type->name)) | |
34 | return log_type; | |
35 | ||
36 | return NULL; | |
37 | } | |
38 | ||
39 | static struct dm_dirty_log_internal *_get_dirty_log_type(const char *name) | |
40 | { | |
41 | struct dm_dirty_log_internal *log_type; | |
1da177e4 LT |
42 | |
43 | spin_lock(&_lock); | |
2a23aa1d JB |
44 | |
45 | log_type = __find_dirty_log_type(name); | |
84e67c93 MS |
46 | if (log_type && !try_module_get(log_type->type->module)) |
47 | log_type = NULL; | |
1da177e4 LT |
48 | |
49 | spin_unlock(&_lock); | |
2a23aa1d JB |
50 | |
51 | return log_type; | |
1da177e4 LT |
52 | } |
53 | ||
fb8b2848 JB |
54 | /* |
55 | * get_type | |
56 | * @type_name | |
57 | * | |
2a23aa1d | 58 | * Attempt to retrieve the dm_dirty_log_type by name. If not already |
fb8b2848 JB |
59 | * available, attempt to load the appropriate module. |
60 | * | |
61 | * Log modules are named "dm-log-" followed by the 'type_name'. | |
62 | * Modules may contain multiple types. | |
63 | * This function will first try the module "dm-log-<type_name>", | |
64 | * then truncate 'type_name' on the last '-' and try again. | |
65 | * | |
66 | * For example, if type_name was "clustered-disk", it would search | |
67 | * 'dm-log-clustered-disk' then 'dm-log-clustered'. | |
68 | * | |
69 | * Returns: dirty_log_type* on success, NULL on failure | |
70 | */ | |
416cd17b | 71 | static struct dm_dirty_log_type *get_type(const char *type_name) |
fb8b2848 JB |
72 | { |
73 | char *p, *type_name_dup; | |
2a23aa1d | 74 | struct dm_dirty_log_internal *log_type; |
fb8b2848 | 75 | |
2a23aa1d JB |
76 | if (!type_name) |
77 | return NULL; | |
78 | ||
79 | log_type = _get_dirty_log_type(type_name); | |
80 | if (log_type) | |
81 | return log_type->type; | |
fb8b2848 JB |
82 | |
83 | type_name_dup = kstrdup(type_name, GFP_KERNEL); | |
84 | if (!type_name_dup) { | |
85 | DMWARN("No memory left to attempt log module load for \"%s\"", | |
86 | type_name); | |
87 | return NULL; | |
88 | } | |
89 | ||
90 | while (request_module("dm-log-%s", type_name_dup) || | |
2a23aa1d | 91 | !(log_type = _get_dirty_log_type(type_name))) { |
fb8b2848 JB |
92 | p = strrchr(type_name_dup, '-'); |
93 | if (!p) | |
94 | break; | |
95 | p[0] = '\0'; | |
96 | } | |
97 | ||
2a23aa1d | 98 | if (!log_type) |
fb8b2848 JB |
99 | DMWARN("Module for logging type \"%s\" not found.", type_name); |
100 | ||
101 | kfree(type_name_dup); | |
102 | ||
2a23aa1d | 103 | return log_type ? log_type->type : NULL; |
fb8b2848 JB |
104 | } |
105 | ||
416cd17b | 106 | static void put_type(struct dm_dirty_log_type *type) |
1da177e4 | 107 | { |
2a23aa1d JB |
108 | struct dm_dirty_log_internal *log_type; |
109 | ||
110 | if (!type) | |
111 | return; | |
112 | ||
1da177e4 | 113 | spin_lock(&_lock); |
2a23aa1d JB |
114 | log_type = __find_dirty_log_type(type->name); |
115 | if (!log_type) | |
116 | goto out; | |
117 | ||
84e67c93 | 118 | module_put(type->module); |
2a23aa1d JB |
119 | |
120 | out: | |
1da177e4 LT |
121 | spin_unlock(&_lock); |
122 | } | |
123 | ||
2a23aa1d JB |
124 | static struct dm_dirty_log_internal *_alloc_dirty_log_type(struct dm_dirty_log_type *type) |
125 | { | |
126 | struct dm_dirty_log_internal *log_type = kzalloc(sizeof(*log_type), | |
127 | GFP_KERNEL); | |
128 | ||
129 | if (log_type) | |
130 | log_type->type = type; | |
131 | ||
132 | return log_type; | |
133 | } | |
134 | ||
b8206bc3 AK |
135 | int dm_dirty_log_type_register(struct dm_dirty_log_type *type) |
136 | { | |
2a23aa1d JB |
137 | struct dm_dirty_log_internal *log_type = _alloc_dirty_log_type(type); |
138 | int r = 0; | |
139 | ||
140 | if (!log_type) | |
141 | return -ENOMEM; | |
142 | ||
b8206bc3 | 143 | spin_lock(&_lock); |
2a23aa1d JB |
144 | if (!__find_dirty_log_type(type->name)) |
145 | list_add(&log_type->list, &_log_types); | |
146 | else { | |
147 | kfree(log_type); | |
148 | r = -EEXIST; | |
149 | } | |
b8206bc3 AK |
150 | spin_unlock(&_lock); |
151 | ||
2a23aa1d | 152 | return r; |
b8206bc3 AK |
153 | } |
154 | EXPORT_SYMBOL(dm_dirty_log_type_register); | |
155 | ||
156 | int dm_dirty_log_type_unregister(struct dm_dirty_log_type *type) | |
157 | { | |
2a23aa1d JB |
158 | struct dm_dirty_log_internal *log_type; |
159 | ||
b8206bc3 AK |
160 | spin_lock(&_lock); |
161 | ||
2a23aa1d JB |
162 | log_type = __find_dirty_log_type(type->name); |
163 | if (!log_type) { | |
164 | spin_unlock(&_lock); | |
165 | return -EINVAL; | |
166 | } | |
167 | ||
2a23aa1d | 168 | list_del(&log_type->list); |
b8206bc3 AK |
169 | |
170 | spin_unlock(&_lock); | |
2a23aa1d | 171 | kfree(log_type); |
b8206bc3 AK |
172 | |
173 | return 0; | |
174 | } | |
175 | EXPORT_SYMBOL(dm_dirty_log_type_unregister); | |
176 | ||
416cd17b HM |
177 | struct dm_dirty_log *dm_dirty_log_create(const char *type_name, |
178 | struct dm_target *ti, | |
179 | unsigned int argc, char **argv) | |
1da177e4 | 180 | { |
416cd17b HM |
181 | struct dm_dirty_log_type *type; |
182 | struct dm_dirty_log *log; | |
1da177e4 LT |
183 | |
184 | log = kmalloc(sizeof(*log), GFP_KERNEL); | |
185 | if (!log) | |
186 | return NULL; | |
187 | ||
188 | type = get_type(type_name); | |
189 | if (!type) { | |
190 | kfree(log); | |
191 | return NULL; | |
192 | } | |
193 | ||
194 | log->type = type; | |
195 | if (type->ctr(log, ti, argc, argv)) { | |
196 | kfree(log); | |
197 | put_type(type); | |
198 | return NULL; | |
199 | } | |
200 | ||
201 | return log; | |
202 | } | |
416cd17b | 203 | EXPORT_SYMBOL(dm_dirty_log_create); |
1da177e4 | 204 | |
416cd17b | 205 | void dm_dirty_log_destroy(struct dm_dirty_log *log) |
1da177e4 LT |
206 | { |
207 | log->type->dtr(log); | |
208 | put_type(log->type); | |
209 | kfree(log); | |
210 | } | |
416cd17b | 211 | EXPORT_SYMBOL(dm_dirty_log_destroy); |
1da177e4 LT |
212 | |
213 | /*----------------------------------------------------------------- | |
214 | * Persistent and core logs share a lot of their implementation. | |
215 | * FIXME: need a reload method to be called from a resume | |
216 | *---------------------------------------------------------------*/ | |
217 | /* | |
218 | * Magic for persistent mirrors: "MiRr" | |
219 | */ | |
220 | #define MIRROR_MAGIC 0x4D695272 | |
221 | ||
222 | /* | |
223 | * The on-disk version of the metadata. | |
224 | */ | |
a4fc4717 | 225 | #define MIRROR_DISK_VERSION 2 |
1da177e4 LT |
226 | #define LOG_OFFSET 2 |
227 | ||
228 | struct log_header { | |
229 | uint32_t magic; | |
230 | ||
231 | /* | |
232 | * Simple, incrementing version. no backward | |
233 | * compatibility. | |
234 | */ | |
235 | uint32_t version; | |
236 | sector_t nr_regions; | |
237 | }; | |
238 | ||
239 | struct log_c { | |
240 | struct dm_target *ti; | |
241 | int touched; | |
242 | uint32_t region_size; | |
243 | unsigned int region_count; | |
244 | region_t sync_count; | |
245 | ||
246 | unsigned bitset_uint32_count; | |
247 | uint32_t *clean_bits; | |
248 | uint32_t *sync_bits; | |
249 | uint32_t *recovering_bits; /* FIXME: this seems excessive */ | |
250 | ||
251 | int sync_search; | |
252 | ||
253 | /* Resync flag */ | |
254 | enum sync { | |
255 | DEFAULTSYNC, /* Synchronize if necessary */ | |
256 | NOSYNC, /* Devices known to be already in sync */ | |
257 | FORCESYNC, /* Force a sync to happen */ | |
258 | } sync; | |
259 | ||
5d234d1e HM |
260 | struct dm_io_request io_req; |
261 | ||
1da177e4 LT |
262 | /* |
263 | * Disk log fields | |
264 | */ | |
01d03a66 | 265 | int log_dev_failed; |
1da177e4 LT |
266 | struct dm_dev *log_dev; |
267 | struct log_header header; | |
268 | ||
22a1ceb1 | 269 | struct dm_io_region header_location; |
1da177e4 | 270 | struct log_header *disk_header; |
1da177e4 LT |
271 | }; |
272 | ||
273 | /* | |
274 | * The touched member needs to be updated every time we access | |
275 | * one of the bitsets. | |
276 | */ | |
416cd17b | 277 | static inline int log_test_bit(uint32_t *bs, unsigned bit) |
1da177e4 | 278 | { |
a4fc4717 | 279 | return ext2_test_bit(bit, (unsigned long *) bs) ? 1 : 0; |
1da177e4 LT |
280 | } |
281 | ||
282 | static inline void log_set_bit(struct log_c *l, | |
283 | uint32_t *bs, unsigned bit) | |
284 | { | |
a4fc4717 | 285 | ext2_set_bit(bit, (unsigned long *) bs); |
1da177e4 LT |
286 | l->touched = 1; |
287 | } | |
288 | ||
289 | static inline void log_clear_bit(struct log_c *l, | |
290 | uint32_t *bs, unsigned bit) | |
291 | { | |
a4fc4717 | 292 | ext2_clear_bit(bit, (unsigned long *) bs); |
1da177e4 LT |
293 | l->touched = 1; |
294 | } | |
295 | ||
296 | /*---------------------------------------------------------------- | |
297 | * Header IO | |
298 | *--------------------------------------------------------------*/ | |
299 | static void header_to_disk(struct log_header *core, struct log_header *disk) | |
300 | { | |
301 | disk->magic = cpu_to_le32(core->magic); | |
302 | disk->version = cpu_to_le32(core->version); | |
303 | disk->nr_regions = cpu_to_le64(core->nr_regions); | |
304 | } | |
305 | ||
306 | static void header_from_disk(struct log_header *core, struct log_header *disk) | |
307 | { | |
308 | core->magic = le32_to_cpu(disk->magic); | |
309 | core->version = le32_to_cpu(disk->version); | |
310 | core->nr_regions = le64_to_cpu(disk->nr_regions); | |
311 | } | |
312 | ||
5d234d1e HM |
313 | static int rw_header(struct log_c *lc, int rw) |
314 | { | |
315 | lc->io_req.bi_rw = rw; | |
5d234d1e HM |
316 | |
317 | return dm_io(&lc->io_req, 1, &lc->header_location, NULL); | |
318 | } | |
319 | ||
1da177e4 LT |
320 | static int read_header(struct log_c *log) |
321 | { | |
322 | int r; | |
1da177e4 | 323 | |
5d234d1e | 324 | r = rw_header(log, READ); |
1da177e4 LT |
325 | if (r) |
326 | return r; | |
327 | ||
328 | header_from_disk(&log->header, log->disk_header); | |
329 | ||
330 | /* New log required? */ | |
331 | if (log->sync != DEFAULTSYNC || log->header.magic != MIRROR_MAGIC) { | |
332 | log->header.magic = MIRROR_MAGIC; | |
333 | log->header.version = MIRROR_DISK_VERSION; | |
334 | log->header.nr_regions = 0; | |
335 | } | |
336 | ||
a4fc4717 PC |
337 | #ifdef __LITTLE_ENDIAN |
338 | if (log->header.version == 1) | |
339 | log->header.version = 2; | |
340 | #endif | |
341 | ||
1da177e4 LT |
342 | if (log->header.version != MIRROR_DISK_VERSION) { |
343 | DMWARN("incompatible disk log version"); | |
344 | return -EINVAL; | |
345 | } | |
346 | ||
347 | return 0; | |
348 | } | |
349 | ||
2045e88e MB |
350 | static int _check_region_size(struct dm_target *ti, uint32_t region_size) |
351 | { | |
352 | if (region_size < 2 || region_size > ti->len) | |
353 | return 0; | |
354 | ||
355 | if (!is_power_of_2(region_size)) | |
356 | return 0; | |
357 | ||
358 | return 1; | |
359 | } | |
360 | ||
1da177e4 LT |
361 | /*---------------------------------------------------------------- |
362 | * core log constructor/destructor | |
363 | * | |
364 | * argv contains region_size followed optionally by [no]sync | |
365 | *--------------------------------------------------------------*/ | |
366 | #define BYTE_SHIFT 3 | |
416cd17b | 367 | static int create_log_context(struct dm_dirty_log *log, struct dm_target *ti, |
b7cca195 AK |
368 | unsigned int argc, char **argv, |
369 | struct dm_dev *dev) | |
1da177e4 LT |
370 | { |
371 | enum sync sync = DEFAULTSYNC; | |
372 | ||
373 | struct log_c *lc; | |
374 | uint32_t region_size; | |
375 | unsigned int region_count; | |
b7cca195 | 376 | size_t bitset_size, buf_size; |
5d234d1e | 377 | int r; |
1da177e4 LT |
378 | |
379 | if (argc < 1 || argc > 2) { | |
b7fd54a7 | 380 | DMWARN("wrong number of arguments to dirty region log"); |
1da177e4 LT |
381 | return -EINVAL; |
382 | } | |
383 | ||
384 | if (argc > 1) { | |
385 | if (!strcmp(argv[1], "sync")) | |
386 | sync = FORCESYNC; | |
387 | else if (!strcmp(argv[1], "nosync")) | |
388 | sync = NOSYNC; | |
389 | else { | |
b7fd54a7 HM |
390 | DMWARN("unrecognised sync argument to " |
391 | "dirty region log: %s", argv[1]); | |
1da177e4 LT |
392 | return -EINVAL; |
393 | } | |
394 | } | |
395 | ||
2045e88e MB |
396 | if (sscanf(argv[0], "%u", ®ion_size) != 1 || |
397 | !_check_region_size(ti, region_size)) { | |
398 | DMWARN("invalid region size %s", argv[0]); | |
1da177e4 LT |
399 | return -EINVAL; |
400 | } | |
401 | ||
402 | region_count = dm_sector_div_up(ti->len, region_size); | |
403 | ||
404 | lc = kmalloc(sizeof(*lc), GFP_KERNEL); | |
405 | if (!lc) { | |
406 | DMWARN("couldn't allocate core log"); | |
407 | return -ENOMEM; | |
408 | } | |
409 | ||
410 | lc->ti = ti; | |
411 | lc->touched = 0; | |
412 | lc->region_size = region_size; | |
413 | lc->region_count = region_count; | |
414 | lc->sync = sync; | |
415 | ||
416 | /* | |
0e56822d | 417 | * Work out how many "unsigned long"s we need to hold the bitset. |
1da177e4 LT |
418 | */ |
419 | bitset_size = dm_round_up(region_count, | |
29121bd0 | 420 | sizeof(*lc->clean_bits) << BYTE_SHIFT); |
1da177e4 LT |
421 | bitset_size >>= BYTE_SHIFT; |
422 | ||
29121bd0 | 423 | lc->bitset_uint32_count = bitset_size / sizeof(*lc->clean_bits); |
b7cca195 AK |
424 | |
425 | /* | |
426 | * Disk log? | |
427 | */ | |
428 | if (!dev) { | |
429 | lc->clean_bits = vmalloc(bitset_size); | |
430 | if (!lc->clean_bits) { | |
431 | DMWARN("couldn't allocate clean bitset"); | |
432 | kfree(lc); | |
433 | return -ENOMEM; | |
434 | } | |
435 | lc->disk_header = NULL; | |
436 | } else { | |
437 | lc->log_dev = dev; | |
01d03a66 | 438 | lc->log_dev_failed = 0; |
b7cca195 AK |
439 | lc->header_location.bdev = lc->log_dev->bdev; |
440 | lc->header_location.sector = 0; | |
441 | ||
442 | /* | |
443 | * Buffer holds both header and bitset. | |
444 | */ | |
445 | buf_size = dm_round_up((LOG_OFFSET << SECTOR_SHIFT) + | |
446 | bitset_size, ti->limits.hardsect_size); | |
ac1f0ac2 MB |
447 | |
448 | if (buf_size > dev->bdev->bd_inode->i_size) { | |
449 | DMWARN("log device %s too small: need %llu bytes", | |
450 | dev->name, (unsigned long long)buf_size); | |
451 | kfree(lc); | |
452 | return -EINVAL; | |
453 | } | |
454 | ||
b7cca195 | 455 | lc->header_location.count = buf_size >> SECTOR_SHIFT; |
6f3af01c | 456 | |
5d234d1e | 457 | lc->io_req.mem.type = DM_IO_VMA; |
6f3af01c | 458 | lc->io_req.notify.fn = NULL; |
5d234d1e HM |
459 | lc->io_req.client = dm_io_client_create(dm_div_up(buf_size, |
460 | PAGE_SIZE)); | |
461 | if (IS_ERR(lc->io_req.client)) { | |
462 | r = PTR_ERR(lc->io_req.client); | |
463 | DMWARN("couldn't allocate disk io client"); | |
464 | kfree(lc); | |
465 | return -ENOMEM; | |
466 | } | |
b7cca195 AK |
467 | |
468 | lc->disk_header = vmalloc(buf_size); | |
469 | if (!lc->disk_header) { | |
470 | DMWARN("couldn't allocate disk log buffer"); | |
c7a2bd19 | 471 | dm_io_client_destroy(lc->io_req.client); |
b7cca195 AK |
472 | kfree(lc); |
473 | return -ENOMEM; | |
474 | } | |
475 | ||
6f3af01c | 476 | lc->io_req.mem.ptr.vma = lc->disk_header; |
b7cca195 AK |
477 | lc->clean_bits = (void *)lc->disk_header + |
478 | (LOG_OFFSET << SECTOR_SHIFT); | |
1da177e4 | 479 | } |
b7cca195 | 480 | |
1da177e4 LT |
481 | memset(lc->clean_bits, -1, bitset_size); |
482 | ||
483 | lc->sync_bits = vmalloc(bitset_size); | |
484 | if (!lc->sync_bits) { | |
485 | DMWARN("couldn't allocate sync bitset"); | |
b7cca195 AK |
486 | if (!dev) |
487 | vfree(lc->clean_bits); | |
c7a2bd19 TY |
488 | else |
489 | dm_io_client_destroy(lc->io_req.client); | |
b7cca195 | 490 | vfree(lc->disk_header); |
1da177e4 LT |
491 | kfree(lc); |
492 | return -ENOMEM; | |
493 | } | |
494 | memset(lc->sync_bits, (sync == NOSYNC) ? -1 : 0, bitset_size); | |
495 | lc->sync_count = (sync == NOSYNC) ? region_count : 0; | |
496 | ||
497 | lc->recovering_bits = vmalloc(bitset_size); | |
498 | if (!lc->recovering_bits) { | |
499 | DMWARN("couldn't allocate sync bitset"); | |
500 | vfree(lc->sync_bits); | |
b7cca195 AK |
501 | if (!dev) |
502 | vfree(lc->clean_bits); | |
c7a2bd19 TY |
503 | else |
504 | dm_io_client_destroy(lc->io_req.client); | |
b7cca195 | 505 | vfree(lc->disk_header); |
1da177e4 LT |
506 | kfree(lc); |
507 | return -ENOMEM; | |
508 | } | |
509 | memset(lc->recovering_bits, 0, bitset_size); | |
510 | lc->sync_search = 0; | |
511 | log->context = lc; | |
b7cca195 | 512 | |
1da177e4 LT |
513 | return 0; |
514 | } | |
515 | ||
416cd17b | 516 | static int core_ctr(struct dm_dirty_log *log, struct dm_target *ti, |
b7cca195 AK |
517 | unsigned int argc, char **argv) |
518 | { | |
519 | return create_log_context(log, ti, argc, argv, NULL); | |
520 | } | |
521 | ||
522 | static void destroy_log_context(struct log_c *lc) | |
1da177e4 | 523 | { |
1da177e4 LT |
524 | vfree(lc->sync_bits); |
525 | vfree(lc->recovering_bits); | |
526 | kfree(lc); | |
527 | } | |
528 | ||
416cd17b | 529 | static void core_dtr(struct dm_dirty_log *log) |
b7cca195 AK |
530 | { |
531 | struct log_c *lc = (struct log_c *) log->context; | |
532 | ||
533 | vfree(lc->clean_bits); | |
534 | destroy_log_context(lc); | |
535 | } | |
536 | ||
1da177e4 LT |
537 | /*---------------------------------------------------------------- |
538 | * disk log constructor/destructor | |
539 | * | |
540 | * argv contains log_device region_size followed optionally by [no]sync | |
541 | *--------------------------------------------------------------*/ | |
416cd17b | 542 | static int disk_ctr(struct dm_dirty_log *log, struct dm_target *ti, |
1da177e4 LT |
543 | unsigned int argc, char **argv) |
544 | { | |
545 | int r; | |
1da177e4 LT |
546 | struct dm_dev *dev; |
547 | ||
548 | if (argc < 2 || argc > 3) { | |
b7fd54a7 | 549 | DMWARN("wrong number of arguments to disk dirty region log"); |
1da177e4 LT |
550 | return -EINVAL; |
551 | } | |
552 | ||
553 | r = dm_get_device(ti, argv[0], 0, 0 /* FIXME */, | |
554 | FMODE_READ | FMODE_WRITE, &dev); | |
555 | if (r) | |
556 | return r; | |
557 | ||
b7cca195 | 558 | r = create_log_context(log, ti, argc - 1, argv + 1, dev); |
1da177e4 LT |
559 | if (r) { |
560 | dm_put_device(ti, dev); | |
561 | return r; | |
562 | } | |
563 | ||
1da177e4 | 564 | return 0; |
1da177e4 LT |
565 | } |
566 | ||
416cd17b | 567 | static void disk_dtr(struct dm_dirty_log *log) |
1da177e4 LT |
568 | { |
569 | struct log_c *lc = (struct log_c *) log->context; | |
b7cca195 | 570 | |
1da177e4 LT |
571 | dm_put_device(lc->ti, lc->log_dev); |
572 | vfree(lc->disk_header); | |
5d234d1e | 573 | dm_io_client_destroy(lc->io_req.client); |
b7cca195 | 574 | destroy_log_context(lc); |
1da177e4 LT |
575 | } |
576 | ||
577 | static int count_bits32(uint32_t *addr, unsigned size) | |
578 | { | |
579 | int count = 0, i; | |
580 | ||
581 | for (i = 0; i < size; i++) { | |
582 | count += hweight32(*(addr+i)); | |
583 | } | |
584 | return count; | |
585 | } | |
586 | ||
01d03a66 JB |
587 | static void fail_log_device(struct log_c *lc) |
588 | { | |
589 | if (lc->log_dev_failed) | |
590 | return; | |
591 | ||
592 | lc->log_dev_failed = 1; | |
593 | dm_table_event(lc->ti->table); | |
594 | } | |
595 | ||
416cd17b | 596 | static int disk_resume(struct dm_dirty_log *log) |
1da177e4 LT |
597 | { |
598 | int r; | |
599 | unsigned i; | |
600 | struct log_c *lc = (struct log_c *) log->context; | |
601 | size_t size = lc->bitset_uint32_count * sizeof(uint32_t); | |
602 | ||
603 | /* read the disk header */ | |
604 | r = read_header(lc); | |
01d03a66 | 605 | if (r) { |
b7fd54a7 | 606 | DMWARN("%s: Failed to read header on dirty region log device", |
01d03a66 JB |
607 | lc->log_dev->name); |
608 | fail_log_device(lc); | |
ba8b45ce JB |
609 | /* |
610 | * If the log device cannot be read, we must assume | |
611 | * all regions are out-of-sync. If we simply return | |
612 | * here, the state will be uninitialized and could | |
613 | * lead us to return 'in-sync' status for regions | |
614 | * that are actually 'out-of-sync'. | |
615 | */ | |
616 | lc->header.nr_regions = 0; | |
01d03a66 | 617 | } |
1da177e4 | 618 | |
8a835f11 | 619 | /* set or clear any new bits -- device has grown */ |
1da177e4 LT |
620 | if (lc->sync == NOSYNC) |
621 | for (i = lc->header.nr_regions; i < lc->region_count; i++) | |
622 | /* FIXME: amazingly inefficient */ | |
623 | log_set_bit(lc, lc->clean_bits, i); | |
624 | else | |
625 | for (i = lc->header.nr_regions; i < lc->region_count; i++) | |
626 | /* FIXME: amazingly inefficient */ | |
627 | log_clear_bit(lc, lc->clean_bits, i); | |
628 | ||
8a835f11 AK |
629 | /* clear any old bits -- device has shrunk */ |
630 | for (i = lc->region_count; i % (sizeof(*lc->clean_bits) << BYTE_SHIFT); i++) | |
631 | log_clear_bit(lc, lc->clean_bits, i); | |
632 | ||
1da177e4 LT |
633 | /* copy clean across to sync */ |
634 | memcpy(lc->sync_bits, lc->clean_bits, size); | |
635 | lc->sync_count = count_bits32(lc->clean_bits, lc->bitset_uint32_count); | |
88b20a1a | 636 | lc->sync_search = 0; |
1da177e4 | 637 | |
1da177e4 LT |
638 | /* set the correct number of regions in the header */ |
639 | lc->header.nr_regions = lc->region_count; | |
640 | ||
6f3af01c TY |
641 | header_to_disk(&lc->header, lc->disk_header); |
642 | ||
1da177e4 | 643 | /* write the new header */ |
6f3af01c | 644 | r = rw_header(lc, WRITE); |
01d03a66 | 645 | if (r) { |
b7fd54a7 | 646 | DMWARN("%s: Failed to write header on dirty region log device", |
01d03a66 JB |
647 | lc->log_dev->name); |
648 | fail_log_device(lc); | |
649 | } | |
650 | ||
651 | return r; | |
1da177e4 LT |
652 | } |
653 | ||
416cd17b | 654 | static uint32_t core_get_region_size(struct dm_dirty_log *log) |
1da177e4 LT |
655 | { |
656 | struct log_c *lc = (struct log_c *) log->context; | |
657 | return lc->region_size; | |
658 | } | |
659 | ||
416cd17b | 660 | static int core_resume(struct dm_dirty_log *log) |
88b20a1a JB |
661 | { |
662 | struct log_c *lc = (struct log_c *) log->context; | |
663 | lc->sync_search = 0; | |
664 | return 0; | |
665 | } | |
666 | ||
416cd17b | 667 | static int core_is_clean(struct dm_dirty_log *log, region_t region) |
1da177e4 LT |
668 | { |
669 | struct log_c *lc = (struct log_c *) log->context; | |
670 | return log_test_bit(lc->clean_bits, region); | |
671 | } | |
672 | ||
416cd17b | 673 | static int core_in_sync(struct dm_dirty_log *log, region_t region, int block) |
1da177e4 LT |
674 | { |
675 | struct log_c *lc = (struct log_c *) log->context; | |
676 | return log_test_bit(lc->sync_bits, region); | |
677 | } | |
678 | ||
416cd17b | 679 | static int core_flush(struct dm_dirty_log *log) |
1da177e4 LT |
680 | { |
681 | /* no op */ | |
682 | return 0; | |
683 | } | |
684 | ||
416cd17b | 685 | static int disk_flush(struct dm_dirty_log *log) |
1da177e4 LT |
686 | { |
687 | int r; | |
688 | struct log_c *lc = (struct log_c *) log->context; | |
689 | ||
690 | /* only write if the log has changed */ | |
691 | if (!lc->touched) | |
692 | return 0; | |
693 | ||
6f3af01c | 694 | r = rw_header(lc, WRITE); |
01d03a66 JB |
695 | if (r) |
696 | fail_log_device(lc); | |
697 | else | |
1da177e4 LT |
698 | lc->touched = 0; |
699 | ||
700 | return r; | |
701 | } | |
702 | ||
416cd17b | 703 | static void core_mark_region(struct dm_dirty_log *log, region_t region) |
1da177e4 LT |
704 | { |
705 | struct log_c *lc = (struct log_c *) log->context; | |
706 | log_clear_bit(lc, lc->clean_bits, region); | |
707 | } | |
708 | ||
416cd17b | 709 | static void core_clear_region(struct dm_dirty_log *log, region_t region) |
1da177e4 LT |
710 | { |
711 | struct log_c *lc = (struct log_c *) log->context; | |
712 | log_set_bit(lc, lc->clean_bits, region); | |
713 | } | |
714 | ||
416cd17b | 715 | static int core_get_resync_work(struct dm_dirty_log *log, region_t *region) |
1da177e4 LT |
716 | { |
717 | struct log_c *lc = (struct log_c *) log->context; | |
718 | ||
719 | if (lc->sync_search >= lc->region_count) | |
720 | return 0; | |
721 | ||
722 | do { | |
1113a7e9 SB |
723 | *region = ext2_find_next_zero_bit( |
724 | (unsigned long *) lc->sync_bits, | |
1da177e4 LT |
725 | lc->region_count, |
726 | lc->sync_search); | |
727 | lc->sync_search = *region + 1; | |
728 | ||
ac81b2ee | 729 | if (*region >= lc->region_count) |
1da177e4 LT |
730 | return 0; |
731 | ||
732 | } while (log_test_bit(lc->recovering_bits, *region)); | |
733 | ||
734 | log_set_bit(lc, lc->recovering_bits, *region); | |
735 | return 1; | |
736 | } | |
737 | ||
416cd17b | 738 | static void core_set_region_sync(struct dm_dirty_log *log, region_t region, |
f3ee6b2f | 739 | int in_sync) |
1da177e4 LT |
740 | { |
741 | struct log_c *lc = (struct log_c *) log->context; | |
742 | ||
743 | log_clear_bit(lc, lc->recovering_bits, region); | |
f3ee6b2f | 744 | if (in_sync) { |
1da177e4 LT |
745 | log_set_bit(lc, lc->sync_bits, region); |
746 | lc->sync_count++; | |
f3ee6b2f JB |
747 | } else if (log_test_bit(lc->sync_bits, region)) { |
748 | lc->sync_count--; | |
749 | log_clear_bit(lc, lc->sync_bits, region); | |
750 | } | |
1da177e4 LT |
751 | } |
752 | ||
416cd17b | 753 | static region_t core_get_sync_count(struct dm_dirty_log *log) |
1da177e4 LT |
754 | { |
755 | struct log_c *lc = (struct log_c *) log->context; | |
756 | ||
757 | return lc->sync_count; | |
758 | } | |
759 | ||
760 | #define DMEMIT_SYNC \ | |
761 | if (lc->sync != DEFAULTSYNC) \ | |
762 | DMEMIT("%ssync ", lc->sync == NOSYNC ? "no" : "") | |
763 | ||
416cd17b | 764 | static int core_status(struct dm_dirty_log *log, status_type_t status, |
1da177e4 LT |
765 | char *result, unsigned int maxlen) |
766 | { | |
767 | int sz = 0; | |
768 | struct log_c *lc = log->context; | |
769 | ||
770 | switch(status) { | |
771 | case STATUSTYPE_INFO: | |
315dcc22 | 772 | DMEMIT("1 %s", log->type->name); |
1da177e4 LT |
773 | break; |
774 | ||
775 | case STATUSTYPE_TABLE: | |
776 | DMEMIT("%s %u %u ", log->type->name, | |
777 | lc->sync == DEFAULTSYNC ? 1 : 2, lc->region_size); | |
778 | DMEMIT_SYNC; | |
779 | } | |
780 | ||
781 | return sz; | |
782 | } | |
783 | ||
416cd17b | 784 | static int disk_status(struct dm_dirty_log *log, status_type_t status, |
1da177e4 LT |
785 | char *result, unsigned int maxlen) |
786 | { | |
787 | int sz = 0; | |
1da177e4 LT |
788 | struct log_c *lc = log->context; |
789 | ||
790 | switch(status) { | |
791 | case STATUSTYPE_INFO: | |
315dcc22 JB |
792 | DMEMIT("3 %s %s %c", log->type->name, lc->log_dev->name, |
793 | lc->log_dev_failed ? 'D' : 'A'); | |
1da177e4 LT |
794 | break; |
795 | ||
796 | case STATUSTYPE_TABLE: | |
1da177e4 | 797 | DMEMIT("%s %u %s %u ", log->type->name, |
315dcc22 | 798 | lc->sync == DEFAULTSYNC ? 2 : 3, lc->log_dev->name, |
1da177e4 LT |
799 | lc->region_size); |
800 | DMEMIT_SYNC; | |
801 | } | |
802 | ||
803 | return sz; | |
804 | } | |
805 | ||
416cd17b | 806 | static struct dm_dirty_log_type _core_type = { |
1da177e4 LT |
807 | .name = "core", |
808 | .module = THIS_MODULE, | |
809 | .ctr = core_ctr, | |
810 | .dtr = core_dtr, | |
88b20a1a | 811 | .resume = core_resume, |
1da177e4 LT |
812 | .get_region_size = core_get_region_size, |
813 | .is_clean = core_is_clean, | |
814 | .in_sync = core_in_sync, | |
815 | .flush = core_flush, | |
816 | .mark_region = core_mark_region, | |
817 | .clear_region = core_clear_region, | |
818 | .get_resync_work = core_get_resync_work, | |
f3ee6b2f | 819 | .set_region_sync = core_set_region_sync, |
1da177e4 LT |
820 | .get_sync_count = core_get_sync_count, |
821 | .status = core_status, | |
822 | }; | |
823 | ||
416cd17b | 824 | static struct dm_dirty_log_type _disk_type = { |
1da177e4 LT |
825 | .name = "disk", |
826 | .module = THIS_MODULE, | |
827 | .ctr = disk_ctr, | |
828 | .dtr = disk_dtr, | |
6b3df0d7 | 829 | .postsuspend = disk_flush, |
1da177e4 LT |
830 | .resume = disk_resume, |
831 | .get_region_size = core_get_region_size, | |
832 | .is_clean = core_is_clean, | |
833 | .in_sync = core_in_sync, | |
834 | .flush = disk_flush, | |
835 | .mark_region = core_mark_region, | |
836 | .clear_region = core_clear_region, | |
837 | .get_resync_work = core_get_resync_work, | |
f3ee6b2f | 838 | .set_region_sync = core_set_region_sync, |
1da177e4 LT |
839 | .get_sync_count = core_get_sync_count, |
840 | .status = disk_status, | |
841 | }; | |
842 | ||
c8da2f8d | 843 | static int __init dm_dirty_log_init(void) |
1da177e4 LT |
844 | { |
845 | int r; | |
846 | ||
416cd17b | 847 | r = dm_dirty_log_type_register(&_core_type); |
1da177e4 LT |
848 | if (r) |
849 | DMWARN("couldn't register core log"); | |
850 | ||
416cd17b | 851 | r = dm_dirty_log_type_register(&_disk_type); |
1da177e4 LT |
852 | if (r) { |
853 | DMWARN("couldn't register disk type"); | |
416cd17b | 854 | dm_dirty_log_type_unregister(&_core_type); |
1da177e4 LT |
855 | } |
856 | ||
857 | return r; | |
858 | } | |
859 | ||
c8da2f8d | 860 | static void __exit dm_dirty_log_exit(void) |
1da177e4 | 861 | { |
416cd17b HM |
862 | dm_dirty_log_type_unregister(&_disk_type); |
863 | dm_dirty_log_type_unregister(&_core_type); | |
1da177e4 LT |
864 | } |
865 | ||
769aef30 HM |
866 | module_init(dm_dirty_log_init); |
867 | module_exit(dm_dirty_log_exit); | |
868 | ||
869 | MODULE_DESCRIPTION(DM_NAME " dirty region log"); | |
870 | MODULE_AUTHOR("Joe Thornber, Heinz Mauelshagen <dm-devel@redhat.com>"); | |
871 | MODULE_LICENSE("GPL"); |