| 1 | /* |
| 2 | * Copyright (C) 2003 Sistina Software |
| 3 | * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved. |
| 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> |
| 12 | #include <linux/dm-io.h> |
| 13 | #include <linux/dm-dirty-log.h> |
| 14 | |
| 15 | #include <linux/device-mapper.h> |
| 16 | |
| 17 | #define DM_MSG_PREFIX "dirty region log" |
| 18 | |
| 19 | static LIST_HEAD(_log_types); |
| 20 | static DEFINE_SPINLOCK(_lock); |
| 21 | |
| 22 | static struct dm_dirty_log_type *__find_dirty_log_type(const char *name) |
| 23 | { |
| 24 | struct dm_dirty_log_type *log_type; |
| 25 | |
| 26 | list_for_each_entry(log_type, &_log_types, list) |
| 27 | if (!strcmp(name, log_type->name)) |
| 28 | return log_type; |
| 29 | |
| 30 | return NULL; |
| 31 | } |
| 32 | |
| 33 | static struct dm_dirty_log_type *_get_dirty_log_type(const char *name) |
| 34 | { |
| 35 | struct dm_dirty_log_type *log_type; |
| 36 | |
| 37 | spin_lock(&_lock); |
| 38 | |
| 39 | log_type = __find_dirty_log_type(name); |
| 40 | if (log_type && !try_module_get(log_type->module)) |
| 41 | log_type = NULL; |
| 42 | |
| 43 | spin_unlock(&_lock); |
| 44 | |
| 45 | return log_type; |
| 46 | } |
| 47 | |
| 48 | /* |
| 49 | * get_type |
| 50 | * @type_name |
| 51 | * |
| 52 | * Attempt to retrieve the dm_dirty_log_type by name. If not already |
| 53 | * available, attempt to load the appropriate module. |
| 54 | * |
| 55 | * Log modules are named "dm-log-" followed by the 'type_name'. |
| 56 | * Modules may contain multiple types. |
| 57 | * This function will first try the module "dm-log-<type_name>", |
| 58 | * then truncate 'type_name' on the last '-' and try again. |
| 59 | * |
| 60 | * For example, if type_name was "clustered-disk", it would search |
| 61 | * 'dm-log-clustered-disk' then 'dm-log-clustered'. |
| 62 | * |
| 63 | * Returns: dirty_log_type* on success, NULL on failure |
| 64 | */ |
| 65 | static struct dm_dirty_log_type *get_type(const char *type_name) |
| 66 | { |
| 67 | char *p, *type_name_dup; |
| 68 | struct dm_dirty_log_type *log_type; |
| 69 | |
| 70 | if (!type_name) |
| 71 | return NULL; |
| 72 | |
| 73 | log_type = _get_dirty_log_type(type_name); |
| 74 | if (log_type) |
| 75 | return log_type; |
| 76 | |
| 77 | type_name_dup = kstrdup(type_name, GFP_KERNEL); |
| 78 | if (!type_name_dup) { |
| 79 | DMWARN("No memory left to attempt log module load for \"%s\"", |
| 80 | type_name); |
| 81 | return NULL; |
| 82 | } |
| 83 | |
| 84 | while (request_module("dm-log-%s", type_name_dup) || |
| 85 | !(log_type = _get_dirty_log_type(type_name))) { |
| 86 | p = strrchr(type_name_dup, '-'); |
| 87 | if (!p) |
| 88 | break; |
| 89 | p[0] = '\0'; |
| 90 | } |
| 91 | |
| 92 | if (!log_type) |
| 93 | DMWARN("Module for logging type \"%s\" not found.", type_name); |
| 94 | |
| 95 | kfree(type_name_dup); |
| 96 | |
| 97 | return log_type; |
| 98 | } |
| 99 | |
| 100 | static void put_type(struct dm_dirty_log_type *type) |
| 101 | { |
| 102 | if (!type) |
| 103 | return; |
| 104 | |
| 105 | spin_lock(&_lock); |
| 106 | if (!__find_dirty_log_type(type->name)) |
| 107 | goto out; |
| 108 | |
| 109 | module_put(type->module); |
| 110 | |
| 111 | out: |
| 112 | spin_unlock(&_lock); |
| 113 | } |
| 114 | |
| 115 | int dm_dirty_log_type_register(struct dm_dirty_log_type *type) |
| 116 | { |
| 117 | int r = 0; |
| 118 | |
| 119 | spin_lock(&_lock); |
| 120 | if (!__find_dirty_log_type(type->name)) |
| 121 | list_add(&type->list, &_log_types); |
| 122 | else |
| 123 | r = -EEXIST; |
| 124 | spin_unlock(&_lock); |
| 125 | |
| 126 | return r; |
| 127 | } |
| 128 | EXPORT_SYMBOL(dm_dirty_log_type_register); |
| 129 | |
| 130 | int dm_dirty_log_type_unregister(struct dm_dirty_log_type *type) |
| 131 | { |
| 132 | spin_lock(&_lock); |
| 133 | |
| 134 | if (!__find_dirty_log_type(type->name)) { |
| 135 | spin_unlock(&_lock); |
| 136 | return -EINVAL; |
| 137 | } |
| 138 | |
| 139 | list_del(&type->list); |
| 140 | |
| 141 | spin_unlock(&_lock); |
| 142 | |
| 143 | return 0; |
| 144 | } |
| 145 | EXPORT_SYMBOL(dm_dirty_log_type_unregister); |
| 146 | |
| 147 | struct dm_dirty_log *dm_dirty_log_create(const char *type_name, |
| 148 | struct dm_target *ti, |
| 149 | int (*flush_callback_fn)(struct dm_target *ti), |
| 150 | unsigned int argc, char **argv) |
| 151 | { |
| 152 | struct dm_dirty_log_type *type; |
| 153 | struct dm_dirty_log *log; |
| 154 | |
| 155 | log = kmalloc(sizeof(*log), GFP_KERNEL); |
| 156 | if (!log) |
| 157 | return NULL; |
| 158 | |
| 159 | type = get_type(type_name); |
| 160 | if (!type) { |
| 161 | kfree(log); |
| 162 | return NULL; |
| 163 | } |
| 164 | |
| 165 | log->flush_callback_fn = flush_callback_fn; |
| 166 | log->type = type; |
| 167 | if (type->ctr(log, ti, argc, argv)) { |
| 168 | kfree(log); |
| 169 | put_type(type); |
| 170 | return NULL; |
| 171 | } |
| 172 | |
| 173 | return log; |
| 174 | } |
| 175 | EXPORT_SYMBOL(dm_dirty_log_create); |
| 176 | |
| 177 | void dm_dirty_log_destroy(struct dm_dirty_log *log) |
| 178 | { |
| 179 | log->type->dtr(log); |
| 180 | put_type(log->type); |
| 181 | kfree(log); |
| 182 | } |
| 183 | EXPORT_SYMBOL(dm_dirty_log_destroy); |
| 184 | |
| 185 | /*----------------------------------------------------------------- |
| 186 | * Persistent and core logs share a lot of their implementation. |
| 187 | * FIXME: need a reload method to be called from a resume |
| 188 | *---------------------------------------------------------------*/ |
| 189 | /* |
| 190 | * Magic for persistent mirrors: "MiRr" |
| 191 | */ |
| 192 | #define MIRROR_MAGIC 0x4D695272 |
| 193 | |
| 194 | /* |
| 195 | * The on-disk version of the metadata. |
| 196 | */ |
| 197 | #define MIRROR_DISK_VERSION 2 |
| 198 | #define LOG_OFFSET 2 |
| 199 | |
| 200 | struct log_header { |
| 201 | uint32_t magic; |
| 202 | |
| 203 | /* |
| 204 | * Simple, incrementing version. no backward |
| 205 | * compatibility. |
| 206 | */ |
| 207 | uint32_t version; |
| 208 | sector_t nr_regions; |
| 209 | }; |
| 210 | |
| 211 | struct log_c { |
| 212 | struct dm_target *ti; |
| 213 | int touched_dirtied; |
| 214 | int touched_cleaned; |
| 215 | int flush_failed; |
| 216 | uint32_t region_size; |
| 217 | unsigned int region_count; |
| 218 | region_t sync_count; |
| 219 | |
| 220 | unsigned bitset_uint32_count; |
| 221 | uint32_t *clean_bits; |
| 222 | uint32_t *sync_bits; |
| 223 | uint32_t *recovering_bits; /* FIXME: this seems excessive */ |
| 224 | |
| 225 | int sync_search; |
| 226 | |
| 227 | /* Resync flag */ |
| 228 | enum sync { |
| 229 | DEFAULTSYNC, /* Synchronize if necessary */ |
| 230 | NOSYNC, /* Devices known to be already in sync */ |
| 231 | FORCESYNC, /* Force a sync to happen */ |
| 232 | } sync; |
| 233 | |
| 234 | struct dm_io_request io_req; |
| 235 | |
| 236 | /* |
| 237 | * Disk log fields |
| 238 | */ |
| 239 | int log_dev_failed; |
| 240 | int log_dev_flush_failed; |
| 241 | struct dm_dev *log_dev; |
| 242 | struct log_header header; |
| 243 | |
| 244 | struct dm_io_region header_location; |
| 245 | struct log_header *disk_header; |
| 246 | }; |
| 247 | |
| 248 | /* |
| 249 | * The touched member needs to be updated every time we access |
| 250 | * one of the bitsets. |
| 251 | */ |
| 252 | static inline int log_test_bit(uint32_t *bs, unsigned bit) |
| 253 | { |
| 254 | return ext2_test_bit(bit, (unsigned long *) bs) ? 1 : 0; |
| 255 | } |
| 256 | |
| 257 | static inline void log_set_bit(struct log_c *l, |
| 258 | uint32_t *bs, unsigned bit) |
| 259 | { |
| 260 | ext2_set_bit(bit, (unsigned long *) bs); |
| 261 | l->touched_cleaned = 1; |
| 262 | } |
| 263 | |
| 264 | static inline void log_clear_bit(struct log_c *l, |
| 265 | uint32_t *bs, unsigned bit) |
| 266 | { |
| 267 | ext2_clear_bit(bit, (unsigned long *) bs); |
| 268 | l->touched_dirtied = 1; |
| 269 | } |
| 270 | |
| 271 | /*---------------------------------------------------------------- |
| 272 | * Header IO |
| 273 | *--------------------------------------------------------------*/ |
| 274 | static void header_to_disk(struct log_header *core, struct log_header *disk) |
| 275 | { |
| 276 | disk->magic = cpu_to_le32(core->magic); |
| 277 | disk->version = cpu_to_le32(core->version); |
| 278 | disk->nr_regions = cpu_to_le64(core->nr_regions); |
| 279 | } |
| 280 | |
| 281 | static void header_from_disk(struct log_header *core, struct log_header *disk) |
| 282 | { |
| 283 | core->magic = le32_to_cpu(disk->magic); |
| 284 | core->version = le32_to_cpu(disk->version); |
| 285 | core->nr_regions = le64_to_cpu(disk->nr_regions); |
| 286 | } |
| 287 | |
| 288 | static int rw_header(struct log_c *lc, int rw) |
| 289 | { |
| 290 | lc->io_req.bi_rw = rw; |
| 291 | |
| 292 | return dm_io(&lc->io_req, 1, &lc->header_location, NULL); |
| 293 | } |
| 294 | |
| 295 | static int flush_header(struct log_c *lc) |
| 296 | { |
| 297 | struct dm_io_region null_location = { |
| 298 | .bdev = lc->header_location.bdev, |
| 299 | .sector = 0, |
| 300 | .count = 0, |
| 301 | }; |
| 302 | |
| 303 | lc->io_req.bi_rw = WRITE_FLUSH; |
| 304 | |
| 305 | return dm_io(&lc->io_req, 1, &null_location, NULL); |
| 306 | } |
| 307 | |
| 308 | static int read_header(struct log_c *log) |
| 309 | { |
| 310 | int r; |
| 311 | |
| 312 | r = rw_header(log, READ); |
| 313 | if (r) |
| 314 | return r; |
| 315 | |
| 316 | header_from_disk(&log->header, log->disk_header); |
| 317 | |
| 318 | /* New log required? */ |
| 319 | if (log->sync != DEFAULTSYNC || log->header.magic != MIRROR_MAGIC) { |
| 320 | log->header.magic = MIRROR_MAGIC; |
| 321 | log->header.version = MIRROR_DISK_VERSION; |
| 322 | log->header.nr_regions = 0; |
| 323 | } |
| 324 | |
| 325 | #ifdef __LITTLE_ENDIAN |
| 326 | if (log->header.version == 1) |
| 327 | log->header.version = 2; |
| 328 | #endif |
| 329 | |
| 330 | if (log->header.version != MIRROR_DISK_VERSION) { |
| 331 | DMWARN("incompatible disk log version"); |
| 332 | return -EINVAL; |
| 333 | } |
| 334 | |
| 335 | return 0; |
| 336 | } |
| 337 | |
| 338 | static int _check_region_size(struct dm_target *ti, uint32_t region_size) |
| 339 | { |
| 340 | if (region_size < 2 || region_size > ti->len) |
| 341 | return 0; |
| 342 | |
| 343 | if (!is_power_of_2(region_size)) |
| 344 | return 0; |
| 345 | |
| 346 | return 1; |
| 347 | } |
| 348 | |
| 349 | /*---------------------------------------------------------------- |
| 350 | * core log constructor/destructor |
| 351 | * |
| 352 | * argv contains region_size followed optionally by [no]sync |
| 353 | *--------------------------------------------------------------*/ |
| 354 | #define BYTE_SHIFT 3 |
| 355 | static int create_log_context(struct dm_dirty_log *log, struct dm_target *ti, |
| 356 | unsigned int argc, char **argv, |
| 357 | struct dm_dev *dev) |
| 358 | { |
| 359 | enum sync sync = DEFAULTSYNC; |
| 360 | |
| 361 | struct log_c *lc; |
| 362 | uint32_t region_size; |
| 363 | unsigned int region_count; |
| 364 | size_t bitset_size, buf_size; |
| 365 | int r; |
| 366 | |
| 367 | if (argc < 1 || argc > 2) { |
| 368 | DMWARN("wrong number of arguments to dirty region log"); |
| 369 | return -EINVAL; |
| 370 | } |
| 371 | |
| 372 | if (argc > 1) { |
| 373 | if (!strcmp(argv[1], "sync")) |
| 374 | sync = FORCESYNC; |
| 375 | else if (!strcmp(argv[1], "nosync")) |
| 376 | sync = NOSYNC; |
| 377 | else { |
| 378 | DMWARN("unrecognised sync argument to " |
| 379 | "dirty region log: %s", argv[1]); |
| 380 | return -EINVAL; |
| 381 | } |
| 382 | } |
| 383 | |
| 384 | if (sscanf(argv[0], "%u", ®ion_size) != 1 || |
| 385 | !_check_region_size(ti, region_size)) { |
| 386 | DMWARN("invalid region size %s", argv[0]); |
| 387 | return -EINVAL; |
| 388 | } |
| 389 | |
| 390 | region_count = dm_sector_div_up(ti->len, region_size); |
| 391 | |
| 392 | lc = kmalloc(sizeof(*lc), GFP_KERNEL); |
| 393 | if (!lc) { |
| 394 | DMWARN("couldn't allocate core log"); |
| 395 | return -ENOMEM; |
| 396 | } |
| 397 | |
| 398 | lc->ti = ti; |
| 399 | lc->touched_dirtied = 0; |
| 400 | lc->touched_cleaned = 0; |
| 401 | lc->flush_failed = 0; |
| 402 | lc->region_size = region_size; |
| 403 | lc->region_count = region_count; |
| 404 | lc->sync = sync; |
| 405 | |
| 406 | /* |
| 407 | * Work out how many "unsigned long"s we need to hold the bitset. |
| 408 | */ |
| 409 | bitset_size = dm_round_up(region_count, |
| 410 | sizeof(*lc->clean_bits) << BYTE_SHIFT); |
| 411 | bitset_size >>= BYTE_SHIFT; |
| 412 | |
| 413 | lc->bitset_uint32_count = bitset_size / sizeof(*lc->clean_bits); |
| 414 | |
| 415 | /* |
| 416 | * Disk log? |
| 417 | */ |
| 418 | if (!dev) { |
| 419 | lc->clean_bits = vmalloc(bitset_size); |
| 420 | if (!lc->clean_bits) { |
| 421 | DMWARN("couldn't allocate clean bitset"); |
| 422 | kfree(lc); |
| 423 | return -ENOMEM; |
| 424 | } |
| 425 | lc->disk_header = NULL; |
| 426 | } else { |
| 427 | lc->log_dev = dev; |
| 428 | lc->log_dev_failed = 0; |
| 429 | lc->log_dev_flush_failed = 0; |
| 430 | lc->header_location.bdev = lc->log_dev->bdev; |
| 431 | lc->header_location.sector = 0; |
| 432 | |
| 433 | /* |
| 434 | * Buffer holds both header and bitset. |
| 435 | */ |
| 436 | buf_size = |
| 437 | dm_round_up((LOG_OFFSET << SECTOR_SHIFT) + bitset_size, |
| 438 | bdev_logical_block_size(lc->header_location. |
| 439 | bdev)); |
| 440 | |
| 441 | if (buf_size > i_size_read(dev->bdev->bd_inode)) { |
| 442 | DMWARN("log device %s too small: need %llu bytes", |
| 443 | dev->name, (unsigned long long)buf_size); |
| 444 | kfree(lc); |
| 445 | return -EINVAL; |
| 446 | } |
| 447 | |
| 448 | lc->header_location.count = buf_size >> SECTOR_SHIFT; |
| 449 | |
| 450 | lc->io_req.mem.type = DM_IO_VMA; |
| 451 | lc->io_req.notify.fn = NULL; |
| 452 | lc->io_req.client = dm_io_client_create(dm_div_up(buf_size, |
| 453 | PAGE_SIZE)); |
| 454 | if (IS_ERR(lc->io_req.client)) { |
| 455 | r = PTR_ERR(lc->io_req.client); |
| 456 | DMWARN("couldn't allocate disk io client"); |
| 457 | kfree(lc); |
| 458 | return -ENOMEM; |
| 459 | } |
| 460 | |
| 461 | lc->disk_header = vmalloc(buf_size); |
| 462 | if (!lc->disk_header) { |
| 463 | DMWARN("couldn't allocate disk log buffer"); |
| 464 | dm_io_client_destroy(lc->io_req.client); |
| 465 | kfree(lc); |
| 466 | return -ENOMEM; |
| 467 | } |
| 468 | |
| 469 | lc->io_req.mem.ptr.vma = lc->disk_header; |
| 470 | lc->clean_bits = (void *)lc->disk_header + |
| 471 | (LOG_OFFSET << SECTOR_SHIFT); |
| 472 | } |
| 473 | |
| 474 | memset(lc->clean_bits, -1, bitset_size); |
| 475 | |
| 476 | lc->sync_bits = vmalloc(bitset_size); |
| 477 | if (!lc->sync_bits) { |
| 478 | DMWARN("couldn't allocate sync bitset"); |
| 479 | if (!dev) |
| 480 | vfree(lc->clean_bits); |
| 481 | else |
| 482 | dm_io_client_destroy(lc->io_req.client); |
| 483 | vfree(lc->disk_header); |
| 484 | kfree(lc); |
| 485 | return -ENOMEM; |
| 486 | } |
| 487 | memset(lc->sync_bits, (sync == NOSYNC) ? -1 : 0, bitset_size); |
| 488 | lc->sync_count = (sync == NOSYNC) ? region_count : 0; |
| 489 | |
| 490 | lc->recovering_bits = vmalloc(bitset_size); |
| 491 | if (!lc->recovering_bits) { |
| 492 | DMWARN("couldn't allocate sync bitset"); |
| 493 | vfree(lc->sync_bits); |
| 494 | if (!dev) |
| 495 | vfree(lc->clean_bits); |
| 496 | else |
| 497 | dm_io_client_destroy(lc->io_req.client); |
| 498 | vfree(lc->disk_header); |
| 499 | kfree(lc); |
| 500 | return -ENOMEM; |
| 501 | } |
| 502 | memset(lc->recovering_bits, 0, bitset_size); |
| 503 | lc->sync_search = 0; |
| 504 | log->context = lc; |
| 505 | |
| 506 | return 0; |
| 507 | } |
| 508 | |
| 509 | static int core_ctr(struct dm_dirty_log *log, struct dm_target *ti, |
| 510 | unsigned int argc, char **argv) |
| 511 | { |
| 512 | return create_log_context(log, ti, argc, argv, NULL); |
| 513 | } |
| 514 | |
| 515 | static void destroy_log_context(struct log_c *lc) |
| 516 | { |
| 517 | vfree(lc->sync_bits); |
| 518 | vfree(lc->recovering_bits); |
| 519 | kfree(lc); |
| 520 | } |
| 521 | |
| 522 | static void core_dtr(struct dm_dirty_log *log) |
| 523 | { |
| 524 | struct log_c *lc = (struct log_c *) log->context; |
| 525 | |
| 526 | vfree(lc->clean_bits); |
| 527 | destroy_log_context(lc); |
| 528 | } |
| 529 | |
| 530 | /*---------------------------------------------------------------- |
| 531 | * disk log constructor/destructor |
| 532 | * |
| 533 | * argv contains log_device region_size followed optionally by [no]sync |
| 534 | *--------------------------------------------------------------*/ |
| 535 | static int disk_ctr(struct dm_dirty_log *log, struct dm_target *ti, |
| 536 | unsigned int argc, char **argv) |
| 537 | { |
| 538 | int r; |
| 539 | struct dm_dev *dev; |
| 540 | |
| 541 | if (argc < 2 || argc > 3) { |
| 542 | DMWARN("wrong number of arguments to disk dirty region log"); |
| 543 | return -EINVAL; |
| 544 | } |
| 545 | |
| 546 | r = dm_get_device(ti, argv[0], FMODE_READ | FMODE_WRITE, &dev); |
| 547 | if (r) |
| 548 | return r; |
| 549 | |
| 550 | r = create_log_context(log, ti, argc - 1, argv + 1, dev); |
| 551 | if (r) { |
| 552 | dm_put_device(ti, dev); |
| 553 | return r; |
| 554 | } |
| 555 | |
| 556 | return 0; |
| 557 | } |
| 558 | |
| 559 | static void disk_dtr(struct dm_dirty_log *log) |
| 560 | { |
| 561 | struct log_c *lc = (struct log_c *) log->context; |
| 562 | |
| 563 | dm_put_device(lc->ti, lc->log_dev); |
| 564 | vfree(lc->disk_header); |
| 565 | dm_io_client_destroy(lc->io_req.client); |
| 566 | destroy_log_context(lc); |
| 567 | } |
| 568 | |
| 569 | static int count_bits32(uint32_t *addr, unsigned size) |
| 570 | { |
| 571 | int count = 0, i; |
| 572 | |
| 573 | for (i = 0; i < size; i++) { |
| 574 | count += hweight32(*(addr+i)); |
| 575 | } |
| 576 | return count; |
| 577 | } |
| 578 | |
| 579 | static void fail_log_device(struct log_c *lc) |
| 580 | { |
| 581 | if (lc->log_dev_failed) |
| 582 | return; |
| 583 | |
| 584 | lc->log_dev_failed = 1; |
| 585 | dm_table_event(lc->ti->table); |
| 586 | } |
| 587 | |
| 588 | static int disk_resume(struct dm_dirty_log *log) |
| 589 | { |
| 590 | int r; |
| 591 | unsigned i; |
| 592 | struct log_c *lc = (struct log_c *) log->context; |
| 593 | size_t size = lc->bitset_uint32_count * sizeof(uint32_t); |
| 594 | |
| 595 | /* read the disk header */ |
| 596 | r = read_header(lc); |
| 597 | if (r) { |
| 598 | DMWARN("%s: Failed to read header on dirty region log device", |
| 599 | lc->log_dev->name); |
| 600 | fail_log_device(lc); |
| 601 | /* |
| 602 | * If the log device cannot be read, we must assume |
| 603 | * all regions are out-of-sync. If we simply return |
| 604 | * here, the state will be uninitialized and could |
| 605 | * lead us to return 'in-sync' status for regions |
| 606 | * that are actually 'out-of-sync'. |
| 607 | */ |
| 608 | lc->header.nr_regions = 0; |
| 609 | } |
| 610 | |
| 611 | /* set or clear any new bits -- device has grown */ |
| 612 | if (lc->sync == NOSYNC) |
| 613 | for (i = lc->header.nr_regions; i < lc->region_count; i++) |
| 614 | /* FIXME: amazingly inefficient */ |
| 615 | log_set_bit(lc, lc->clean_bits, i); |
| 616 | else |
| 617 | for (i = lc->header.nr_regions; i < lc->region_count; i++) |
| 618 | /* FIXME: amazingly inefficient */ |
| 619 | log_clear_bit(lc, lc->clean_bits, i); |
| 620 | |
| 621 | /* clear any old bits -- device has shrunk */ |
| 622 | for (i = lc->region_count; i % (sizeof(*lc->clean_bits) << BYTE_SHIFT); i++) |
| 623 | log_clear_bit(lc, lc->clean_bits, i); |
| 624 | |
| 625 | /* copy clean across to sync */ |
| 626 | memcpy(lc->sync_bits, lc->clean_bits, size); |
| 627 | lc->sync_count = count_bits32(lc->clean_bits, lc->bitset_uint32_count); |
| 628 | lc->sync_search = 0; |
| 629 | |
| 630 | /* set the correct number of regions in the header */ |
| 631 | lc->header.nr_regions = lc->region_count; |
| 632 | |
| 633 | header_to_disk(&lc->header, lc->disk_header); |
| 634 | |
| 635 | /* write the new header */ |
| 636 | r = rw_header(lc, WRITE); |
| 637 | if (!r) { |
| 638 | r = flush_header(lc); |
| 639 | if (r) |
| 640 | lc->log_dev_flush_failed = 1; |
| 641 | } |
| 642 | if (r) { |
| 643 | DMWARN("%s: Failed to write header on dirty region log device", |
| 644 | lc->log_dev->name); |
| 645 | fail_log_device(lc); |
| 646 | } |
| 647 | |
| 648 | return r; |
| 649 | } |
| 650 | |
| 651 | static uint32_t core_get_region_size(struct dm_dirty_log *log) |
| 652 | { |
| 653 | struct log_c *lc = (struct log_c *) log->context; |
| 654 | return lc->region_size; |
| 655 | } |
| 656 | |
| 657 | static int core_resume(struct dm_dirty_log *log) |
| 658 | { |
| 659 | struct log_c *lc = (struct log_c *) log->context; |
| 660 | lc->sync_search = 0; |
| 661 | return 0; |
| 662 | } |
| 663 | |
| 664 | static int core_is_clean(struct dm_dirty_log *log, region_t region) |
| 665 | { |
| 666 | struct log_c *lc = (struct log_c *) log->context; |
| 667 | return log_test_bit(lc->clean_bits, region); |
| 668 | } |
| 669 | |
| 670 | static int core_in_sync(struct dm_dirty_log *log, region_t region, int block) |
| 671 | { |
| 672 | struct log_c *lc = (struct log_c *) log->context; |
| 673 | return log_test_bit(lc->sync_bits, region); |
| 674 | } |
| 675 | |
| 676 | static int core_flush(struct dm_dirty_log *log) |
| 677 | { |
| 678 | /* no op */ |
| 679 | return 0; |
| 680 | } |
| 681 | |
| 682 | static int disk_flush(struct dm_dirty_log *log) |
| 683 | { |
| 684 | int r, i; |
| 685 | struct log_c *lc = log->context; |
| 686 | |
| 687 | /* only write if the log has changed */ |
| 688 | if (!lc->touched_cleaned && !lc->touched_dirtied) |
| 689 | return 0; |
| 690 | |
| 691 | if (lc->touched_cleaned && log->flush_callback_fn && |
| 692 | log->flush_callback_fn(lc->ti)) { |
| 693 | /* |
| 694 | * At this point it is impossible to determine which |
| 695 | * regions are clean and which are dirty (without |
| 696 | * re-reading the log off disk). So mark all of them |
| 697 | * dirty. |
| 698 | */ |
| 699 | lc->flush_failed = 1; |
| 700 | for (i = 0; i < lc->region_count; i++) |
| 701 | log_clear_bit(lc, lc->clean_bits, i); |
| 702 | } |
| 703 | |
| 704 | r = rw_header(lc, WRITE); |
| 705 | if (r) |
| 706 | fail_log_device(lc); |
| 707 | else { |
| 708 | if (lc->touched_dirtied) { |
| 709 | r = flush_header(lc); |
| 710 | if (r) { |
| 711 | lc->log_dev_flush_failed = 1; |
| 712 | fail_log_device(lc); |
| 713 | } else |
| 714 | lc->touched_dirtied = 0; |
| 715 | } |
| 716 | lc->touched_cleaned = 0; |
| 717 | } |
| 718 | |
| 719 | return r; |
| 720 | } |
| 721 | |
| 722 | static void core_mark_region(struct dm_dirty_log *log, region_t region) |
| 723 | { |
| 724 | struct log_c *lc = (struct log_c *) log->context; |
| 725 | log_clear_bit(lc, lc->clean_bits, region); |
| 726 | } |
| 727 | |
| 728 | static void core_clear_region(struct dm_dirty_log *log, region_t region) |
| 729 | { |
| 730 | struct log_c *lc = (struct log_c *) log->context; |
| 731 | if (likely(!lc->flush_failed)) |
| 732 | log_set_bit(lc, lc->clean_bits, region); |
| 733 | } |
| 734 | |
| 735 | static int core_get_resync_work(struct dm_dirty_log *log, region_t *region) |
| 736 | { |
| 737 | struct log_c *lc = (struct log_c *) log->context; |
| 738 | |
| 739 | if (lc->sync_search >= lc->region_count) |
| 740 | return 0; |
| 741 | |
| 742 | do { |
| 743 | *region = ext2_find_next_zero_bit( |
| 744 | (unsigned long *) lc->sync_bits, |
| 745 | lc->region_count, |
| 746 | lc->sync_search); |
| 747 | lc->sync_search = *region + 1; |
| 748 | |
| 749 | if (*region >= lc->region_count) |
| 750 | return 0; |
| 751 | |
| 752 | } while (log_test_bit(lc->recovering_bits, *region)); |
| 753 | |
| 754 | log_set_bit(lc, lc->recovering_bits, *region); |
| 755 | return 1; |
| 756 | } |
| 757 | |
| 758 | static void core_set_region_sync(struct dm_dirty_log *log, region_t region, |
| 759 | int in_sync) |
| 760 | { |
| 761 | struct log_c *lc = (struct log_c *) log->context; |
| 762 | |
| 763 | log_clear_bit(lc, lc->recovering_bits, region); |
| 764 | if (in_sync) { |
| 765 | log_set_bit(lc, lc->sync_bits, region); |
| 766 | lc->sync_count++; |
| 767 | } else if (log_test_bit(lc->sync_bits, region)) { |
| 768 | lc->sync_count--; |
| 769 | log_clear_bit(lc, lc->sync_bits, region); |
| 770 | } |
| 771 | } |
| 772 | |
| 773 | static region_t core_get_sync_count(struct dm_dirty_log *log) |
| 774 | { |
| 775 | struct log_c *lc = (struct log_c *) log->context; |
| 776 | |
| 777 | return lc->sync_count; |
| 778 | } |
| 779 | |
| 780 | #define DMEMIT_SYNC \ |
| 781 | if (lc->sync != DEFAULTSYNC) \ |
| 782 | DMEMIT("%ssync ", lc->sync == NOSYNC ? "no" : "") |
| 783 | |
| 784 | static int core_status(struct dm_dirty_log *log, status_type_t status, |
| 785 | char *result, unsigned int maxlen) |
| 786 | { |
| 787 | int sz = 0; |
| 788 | struct log_c *lc = log->context; |
| 789 | |
| 790 | switch(status) { |
| 791 | case STATUSTYPE_INFO: |
| 792 | DMEMIT("1 %s", log->type->name); |
| 793 | break; |
| 794 | |
| 795 | case STATUSTYPE_TABLE: |
| 796 | DMEMIT("%s %u %u ", log->type->name, |
| 797 | lc->sync == DEFAULTSYNC ? 1 : 2, lc->region_size); |
| 798 | DMEMIT_SYNC; |
| 799 | } |
| 800 | |
| 801 | return sz; |
| 802 | } |
| 803 | |
| 804 | static int disk_status(struct dm_dirty_log *log, status_type_t status, |
| 805 | char *result, unsigned int maxlen) |
| 806 | { |
| 807 | int sz = 0; |
| 808 | struct log_c *lc = log->context; |
| 809 | |
| 810 | switch(status) { |
| 811 | case STATUSTYPE_INFO: |
| 812 | DMEMIT("3 %s %s %c", log->type->name, lc->log_dev->name, |
| 813 | lc->log_dev_flush_failed ? 'F' : |
| 814 | lc->log_dev_failed ? 'D' : |
| 815 | 'A'); |
| 816 | break; |
| 817 | |
| 818 | case STATUSTYPE_TABLE: |
| 819 | DMEMIT("%s %u %s %u ", log->type->name, |
| 820 | lc->sync == DEFAULTSYNC ? 2 : 3, lc->log_dev->name, |
| 821 | lc->region_size); |
| 822 | DMEMIT_SYNC; |
| 823 | } |
| 824 | |
| 825 | return sz; |
| 826 | } |
| 827 | |
| 828 | static struct dm_dirty_log_type _core_type = { |
| 829 | .name = "core", |
| 830 | .module = THIS_MODULE, |
| 831 | .ctr = core_ctr, |
| 832 | .dtr = core_dtr, |
| 833 | .resume = core_resume, |
| 834 | .get_region_size = core_get_region_size, |
| 835 | .is_clean = core_is_clean, |
| 836 | .in_sync = core_in_sync, |
| 837 | .flush = core_flush, |
| 838 | .mark_region = core_mark_region, |
| 839 | .clear_region = core_clear_region, |
| 840 | .get_resync_work = core_get_resync_work, |
| 841 | .set_region_sync = core_set_region_sync, |
| 842 | .get_sync_count = core_get_sync_count, |
| 843 | .status = core_status, |
| 844 | }; |
| 845 | |
| 846 | static struct dm_dirty_log_type _disk_type = { |
| 847 | .name = "disk", |
| 848 | .module = THIS_MODULE, |
| 849 | .ctr = disk_ctr, |
| 850 | .dtr = disk_dtr, |
| 851 | .postsuspend = disk_flush, |
| 852 | .resume = disk_resume, |
| 853 | .get_region_size = core_get_region_size, |
| 854 | .is_clean = core_is_clean, |
| 855 | .in_sync = core_in_sync, |
| 856 | .flush = disk_flush, |
| 857 | .mark_region = core_mark_region, |
| 858 | .clear_region = core_clear_region, |
| 859 | .get_resync_work = core_get_resync_work, |
| 860 | .set_region_sync = core_set_region_sync, |
| 861 | .get_sync_count = core_get_sync_count, |
| 862 | .status = disk_status, |
| 863 | }; |
| 864 | |
| 865 | static int __init dm_dirty_log_init(void) |
| 866 | { |
| 867 | int r; |
| 868 | |
| 869 | r = dm_dirty_log_type_register(&_core_type); |
| 870 | if (r) |
| 871 | DMWARN("couldn't register core log"); |
| 872 | |
| 873 | r = dm_dirty_log_type_register(&_disk_type); |
| 874 | if (r) { |
| 875 | DMWARN("couldn't register disk type"); |
| 876 | dm_dirty_log_type_unregister(&_core_type); |
| 877 | } |
| 878 | |
| 879 | return r; |
| 880 | } |
| 881 | |
| 882 | static void __exit dm_dirty_log_exit(void) |
| 883 | { |
| 884 | dm_dirty_log_type_unregister(&_disk_type); |
| 885 | dm_dirty_log_type_unregister(&_core_type); |
| 886 | } |
| 887 | |
| 888 | module_init(dm_dirty_log_init); |
| 889 | module_exit(dm_dirty_log_exit); |
| 890 | |
| 891 | MODULE_DESCRIPTION(DM_NAME " dirty region log"); |
| 892 | MODULE_AUTHOR("Joe Thornber, Heinz Mauelshagen <dm-devel@redhat.com>"); |
| 893 | MODULE_LICENSE("GPL"); |