| 1 | |
| 2 | |
| 3 | * Copyright (C) 2003 Sistina Software
|
| 4 | * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
|
| 5 | *
|
| 6 | * This file is released under the LGPL.
|
| 7 |
|
| 8 |
|
| 9 | #include <linux/init.h>
|
| 10 | #include <linux/slab.h>
|
| 11 | #include <linux/module.h>
|
| 12 | #include <linux/vmalloc.h>
|
| 13 | #include <linux/dm-io.h>
|
| 14 | #include <linux/dm-dirty-log.h>
|
| 15 |
|
| 16 | #include <linux/device-mapper.h>
|
| 17 |
|
| 18 | #define DM_MSG_PREFIX "dirty region log"
|
| 19 |
|
| 20 | static LIST_HEAD(_log_types);
|
| 21 | static DEFINE_SPINLOCK(_lock);
|
| 22 |
|
| 23 | static struct dm_dirty_log_type *__find_dirty_log_type(const char *name)
|
| 24 | {
|
| 25 | struct dm_dirty_log_type *log_type;
|
| 26 |
|
| 27 | list_for_each_entry(log_type, &_log_types, list)
|
| 28 | if (!strcmp(name, log_type->name))
|
| 29 | return log_type;
|
| 30 |
|
| 31 | return NULL;
|
| 32 | }
|
| 33 |
|
| 34 | static struct dm_dirty_log_type *_get_dirty_log_type(const char *name)
|
| 35 | {
|
| 36 | struct dm_dirty_log_type *log_type;
|
| 37 |
|
| 38 | spin_lock(&_lock);
|
| 39 |
|
| 40 | log_type = __find_dirty_log_type(name);
|
| 41 | if (log_type && !try_module_get(log_type->module))
|
| 42 | log_type = NULL;
|
| 43 |
|
| 44 | spin_unlock(&_lock);
|
| 45 |
|
| 46 | return log_type;
|
| 47 | }
|
| 48 |
|
| 49 | |
| 50 | * get_type
|
| 51 | * @type_name
|
| 52 | *
|
| 53 | * Attempt to retrieve the dm_dirty_log_type by name. If not already
|
| 54 | * available, attempt to load the appropriate module.
|
| 55 | *
|
| 56 | * Log modules are named "dm-log-" followed by the 'type_name'.
|
| 57 | * Modules may contain multiple types.
|
| 58 | * This function will first try the module "dm-log-<type_name>",
|
| 59 | * then truncate 'type_name' on the last '-' and try again.
|
| 60 | *
|
| 61 | * For example, if type_name was "clustered-disk", it would search
|
| 62 | * 'dm-log-clustered-disk' then 'dm-log-clustered'.
|
| 63 | *
|
| 64 | * Returns: dirty_log_type* on success, NULL on failure
|
| 65 |
|
| 66 | static struct dm_dirty_log_type *get_type(const char *type_name)
|
| 67 | {
|
| 68 | char *p, *type_name_dup;
|
| 69 | struct dm_dirty_log_type *log_type;
|
| 70 |
|
| 71 | if (!type_name)
|
| 72 | return NULL;
|
| 73 |
|
| 74 | log_type = _get_dirty_log_type(type_name);
|
| 75 | if (log_type)
|
| 76 | return log_type;
|
| 77 |
|
| 78 | type_name_dup = kstrdup(type_name, GFP_KERNEL);
|
| 79 | if (!type_name_dup) {
|
| 80 | DMWARN("No memory left to attempt log module load for \"%s\"",
|
| 81 | type_name);
|
| 82 | return NULL;
|
| 83 | }
|
| 84 |
|
| 85 | while (request_module("dm-log-%s", type_name_dup) ||
|
| 86 | !(log_type = _get_dirty_log_type(type_name))) {
|
| 87 | p = strrchr(type_name_dup, '-');
|
| 88 | if (!p)
|
| 89 | break;
|
| 90 | p[0] = '\0';
|
| 91 | }
|
| 92 |
|
| 93 | if (!log_type)
|
| 94 | DMWARN("Module for logging type \"%s\" not found.", type_name);
|
| 95 |
|
| 96 | kfree(type_name_dup);
|
| 97 |
|
| 98 | return log_type;
|
| 99 | }
|
| 100 |
|
| 101 | static void put_type(struct dm_dirty_log_type *type)
|
| 102 | {
|
| 103 | if (!type)
|
| 104 | return;
|
| 105 |
|
| 106 | spin_lock(&_lock);
|
| 107 | if (!__find_dirty_log_type(type->name))
|
| 108 | goto out;
|
| 109 |
|
| 110 | module_put(type->module);
|
| 111 |
|
| 112 | out:
|
| 113 | spin_unlock(&_lock);
|
| 114 | }
|
| 115 |
|
| 116 | int dm_dirty_log_type_register(struct dm_dirty_log_type *type)
|
| 117 | {
|
| 118 | int r = 0;
|
| 119 |
|
| 120 | spin_lock(&_lock);
|
| 121 | if (!__find_dirty_log_type(type->name))
|
| 122 | list_add(&type->list, &_log_types);
|
| 123 | else
|
| 124 | r = -EBUSY;
|
| 125 | spin_unlock(&_lock);
|
| 126 |
|
| 127 | return r;
|
| 128 | }
|
| 129 | EXPORT_SYMBOL(dm_dirty_log_type_register);
|
| 130 |
|
| 131 | int dm_dirty_log_type_unregister(struct dm_dirty_log_type *type)
|
| 132 | {
|
| 133 | spin_lock(&_lock);
|
| 134 |
|
| 135 | if (!__find_dirty_log_type(type->name)) {
|
| 136 | spin_unlock(&_lock);
|
| 137 | return -EINVAL;
|
| 138 | }
|
| 139 |
|
| 140 | list_del(&type->list);
|
| 141 |
|
| 142 | spin_unlock(&_lock);
|
| 143 |
|
| 144 | return 0;
|
| 145 | }
|
| 146 | EXPORT_SYMBOL(dm_dirty_log_type_unregister);
|
| 147 |
|
| 148 | struct dm_dirty_log *dm_dirty_log_create(const char *type_name,
|
| 149 | struct dm_target *ti,
|
| 150 | int (*flush_callback_fn)(struct dm_target *ti),
|
| 151 | unsigned int argc, char **argv)
|
| 152 | {
|
| 153 | struct dm_dirty_log_type *type;
|
| 154 | struct dm_dirty_log *log;
|
| 155 |
|
| 156 | log = kmalloc_obj(*log);
|
| 157 | if (!log)
|
| 158 | return NULL;
|
| 159 |
|
| 160 | type = get_type(type_name);
|
| 161 | if (!type) {
|
| 162 | kfree(log);
|
| 163 | return NULL;
|
| 164 | }
|
| 165 |
|
| 166 | log->flush_callback_fn = flush_callback_fn;
|
| 167 | log->type = type;
|
| 168 | if (type->ctr(log, ti, argc, argv)) {
|
| 169 | kfree(log);
|
| 170 | put_type(type);
|
| 171 | return NULL;
|
| 172 | }
|
| 173 |
|
| 174 | return log;
|
| 175 | }
|
| 176 | EXPORT_SYMBOL(dm_dirty_log_create);
|
| 177 |
|
| 178 | void dm_dirty_log_destroy(struct dm_dirty_log *log)
|
| 179 | {
|
| 180 | log->type->dtr(log);
|
| 181 | put_type(log->type);
|
| 182 | kfree(log);
|
| 183 | }
|
| 184 | EXPORT_SYMBOL(dm_dirty_log_destroy);
|
| 185 |
|
| 186 | |
| 187 | *---------------------------------------------------------------
|
| 188 | * Persistent and core logs share a lot of their implementation.
|
| 189 | * FIXME: need a reload method to be called from a resume
|
| 190 | *---------------------------------------------------------------
|
| 191 |
|
| 192 | |
| 193 | * Magic for persistent mirrors: "MiRr"
|
| 194 |
|
| 195 | #define MIRROR_MAGIC 0x4D695272
|
| 196 |
|
| 197 | |
| 198 | * The on-disk version of the metadata.
|
| 199 |
|
| 200 | #define MIRROR_DISK_VERSION 2
|
| 201 | #define LOG_OFFSET 2
|
| 202 |
|
| 203 | struct log_header_disk {
|
| 204 | __le32 magic;
|
| 205 |
|
| 206 | |
| 207 | * Simple, incrementing version. no backward
|
| 208 | * compatibility.
|
| 209 |
|
| 210 | __le32 version;
|
| 211 | __le64 nr_regions;
|
| 212 | } __packed;
|
| 213 |
|
| 214 | struct log_header_core {
|
| 215 | uint32_t magic;
|
| 216 | uint32_t version;
|
| 217 | uint64_t nr_regions;
|
| 218 | };
|
| 219 |
|
| 220 | struct log_c {
|
| 221 | struct dm_target *ti;
|
| 222 | int touched_dirtied;
|
| 223 | int touched_cleaned;
|
| 224 | int flush_failed;
|
| 225 | uint32_t region_size;
|
| 226 | unsigned int region_count;
|
| 227 | region_t sync_count;
|
| 228 |
|
| 229 | unsigned int bitset_uint32_count;
|
| 230 | uint32_t *clean_bits;
|
| 231 | uint32_t *sync_bits;
|
| 232 | uint32_t *recovering_bits;
|
| 233 |
|
| 234 | int sync_search;
|
| 235 |
|
| 236 |
|
| 237 | enum sync {
|
| 238 | DEFAULTSYNC,
|
| 239 | NOSYNC,
|
| 240 | FORCESYNC,
|
| 241 | } sync;
|
| 242 |
|
| 243 | struct dm_io_request io_req;
|
| 244 |
|
| 245 | |
| 246 | * Disk log fields
|
| 247 |
|
| 248 | int log_dev_failed;
|
| 249 | int log_dev_flush_failed;
|
| 250 | struct dm_dev *log_dev;
|
| 251 | struct log_header_core header;
|
| 252 |
|
| 253 | struct dm_io_region header_location;
|
| 254 | struct log_header_disk *disk_header;
|
| 255 | };
|
| 256 |
|
| 257 | |
| 258 | * The touched member needs to be updated every time we access
|
| 259 | * one of the bitsets.
|
| 260 |
|
| 261 | static inline int log_test_bit(uint32_t *bs, unsigned int bit)
|
| 262 | {
|
| 263 | return test_bit_le(bit, bs) ? 1 : 0;
|
| 264 | }
|
| 265 |
|
| 266 | static inline void log_set_bit(struct log_c *l,
|
| 267 | uint32_t *bs, unsigned int bit)
|
| 268 | {
|
| 269 | __set_bit_le(bit, bs);
|
| 270 | l->touched_cleaned = 1;
|
| 271 | }
|
| 272 |
|
| 273 | static inline void log_clear_bit(struct log_c *l,
|
| 274 | uint32_t *bs, unsigned int bit)
|
| 275 | {
|
| 276 | __clear_bit_le(bit, bs);
|
| 277 | l->touched_dirtied = 1;
|
| 278 | }
|
| 279 |
|
| 280 | |
| 281 | *---------------------------------------------------------------
|
| 282 | * Header IO
|
| 283 | *--------------------------------------------------------------
|
| 284 |
|
| 285 | static void header_to_disk(struct log_header_core *core, struct log_header_disk *disk)
|
| 286 | {
|
| 287 | disk->magic = cpu_to_le32(core->magic);
|
| 288 | disk->version = cpu_to_le32(core->version);
|
| 289 | disk->nr_regions = cpu_to_le64(core->nr_regions);
|
| 290 | }
|
| 291 |
|
| 292 | static void header_from_disk(struct log_header_core *core, struct log_header_disk *disk)
|
| 293 | {
|
| 294 | core->magic = le32_to_cpu(disk->magic);
|
| 295 | core->version = le32_to_cpu(disk->version);
|
| 296 | core->nr_regions = le64_to_cpu(disk->nr_regions);
|
| 297 | }
|
| 298 |
|
| 299 | static int rw_header(struct log_c *lc, enum req_op op)
|
| 300 | {
|
| 301 | lc->io_req.bi_opf = op;
|
| 302 |
|
| 303 | return dm_io(&lc->io_req, 1, &lc->header_location, NULL, NULL, IOPRIO_DEFAULT);
|
| 304 | }
|
| 305 |
|
| 306 | static int flush_header(struct log_c *lc)
|
| 307 | {
|
| 308 | struct dm_io_region null_location = {
|
| 309 | .bdev = lc->header_location.bdev,
|
| 310 | .sector = 0,
|
| 311 | .count = 0,
|
| 312 | };
|
| 313 |
|
| 314 | lc->io_req.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH;
|
| 315 |
|
| 316 | return dm_io(&lc->io_req, 1, &null_location, NULL, NULL, IOPRIO_DEFAULT);
|
| 317 | }
|
| 318 |
|
| 319 | static int read_header(struct log_c *log)
|
| 320 | {
|
| 321 | int r;
|
| 322 |
|
| 323 | r = rw_header(log, REQ_OP_READ);
|
| 324 | if (r)
|
| 325 | return r;
|
| 326 |
|
| 327 | header_from_disk(&log->header, log->disk_header);
|
| 328 |
|
| 329 |
|
| 330 | if (log->sync != DEFAULTSYNC || log->header.magic != MIRROR_MAGIC) {
|
| 331 | log->header.magic = MIRROR_MAGIC;
|
| 332 | log->header.version = MIRROR_DISK_VERSION;
|
| 333 | log->header.nr_regions = 0;
|
| 334 | }
|
| 335 |
|
| 336 | #ifdef __LITTLE_ENDIAN
|
| 337 | if (log->header.version == 1)
|
| 338 | log->header.version = 2;
|
| 339 | #endif
|
| 340 |
|
| 341 | if (log->header.version != MIRROR_DISK_VERSION) {
|
| 342 | DMWARN("incompatible disk log version");
|
| 343 | return -EINVAL;
|
| 344 | }
|
| 345 |
|
| 346 | return 0;
|
| 347 | }
|
| 348 |
|
| 349 | static int _check_region_size(struct dm_target *ti, uint32_t region_size)
|
| 350 | {
|
| 351 | if (region_size < 2 || region_size > ti->len)
|
| 352 | return 0;
|
| 353 |
|
| 354 | if (!is_power_of_2(region_size))
|
| 355 | return 0;
|
| 356 |
|
| 357 | return 1;
|
| 358 | }
|
| 359 |
|
| 360 | |
| 361 | *--------------------------------------------------------------
|
| 362 | * core log constructor/destructor
|
| 363 | *
|
| 364 | * argv contains region_size followed optionally by [no]sync
|
| 365 | *--------------------------------------------------------------
|
| 366 |
|
| 367 | #define BYTE_SHIFT 3
|
| 368 | static int create_log_context(struct dm_dirty_log *log, struct dm_target *ti,
|
| 369 | unsigned int argc, char **argv,
|
| 370 | struct dm_dev *dev)
|
| 371 | {
|
| 372 | enum sync sync = DEFAULTSYNC;
|
| 373 |
|
| 374 | struct log_c *lc;
|
| 375 | uint32_t region_size;
|
| 376 | sector_t region_count;
|
| 377 | size_t bitset_size, buf_size;
|
| 378 | int r;
|
| 379 | char dummy;
|
| 380 |
|
| 381 | if (argc < 1 || argc > 2) {
|
| 382 | DMWARN("wrong number of arguments to dirty region log");
|
| 383 | return -EINVAL;
|
| 384 | }
|
| 385 |
|
| 386 | if (argc > 1) {
|
| 387 | if (!strcmp(argv[1], "sync"))
|
| 388 | sync = FORCESYNC;
|
| 389 | else if (!strcmp(argv[1], "nosync"))
|
| 390 | sync = NOSYNC;
|
| 391 | else {
|
| 392 | DMWARN("unrecognised sync argument to dirty region log: %s", argv[1]);
|
| 393 | return -EINVAL;
|
| 394 | }
|
| 395 | }
|
| 396 |
|
| 397 | if (sscanf(argv[0], "%u%c", ®ion_size, &dummy) != 1 ||
|
| 398 | !_check_region_size(ti, region_size)) {
|
| 399 | DMWARN("invalid region size %s", argv[0]);
|
| 400 | return -EINVAL;
|
| 401 | }
|
| 402 |
|
| 403 | region_count = dm_sector_div_up(ti->len, region_size);
|
| 404 | if (region_count > UINT_MAX) {
|
| 405 | DMWARN("region count exceeds limit of %u", UINT_MAX);
|
| 406 | return -EINVAL;
|
| 407 | }
|
| 408 |
|
| 409 | lc = kmalloc_obj(*lc);
|
| 410 | if (!lc) {
|
| 411 | DMWARN("couldn't allocate core log");
|
| 412 | return -ENOMEM;
|
| 413 | }
|
| 414 |
|
| 415 | lc->ti = ti;
|
| 416 | lc->touched_dirtied = 0;
|
| 417 | lc->touched_cleaned = 0;
|
| 418 | lc->flush_failed = 0;
|
| 419 | lc->region_size = region_size;
|
| 420 | lc->region_count = region_count;
|
| 421 | lc->sync = sync;
|
| 422 |
|
| 423 | |
| 424 | * Work out how many "unsigned long"s we need to hold the bitset.
|
| 425 |
|
| 426 | bitset_size = dm_round_up(region_count, BITS_PER_LONG);
|
| 427 | bitset_size >>= BYTE_SHIFT;
|
| 428 |
|
| 429 | if (!bitset_size)
|
| 430 | bitset_size = 1UL << (BITS_PER_LONG - BYTE_SHIFT);
|
| 431 |
|
| 432 | lc->bitset_uint32_count = bitset_size / sizeof(*lc->clean_bits);
|
| 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;
|
| 447 | lc->log_dev_failed = 0;
|
| 448 | lc->log_dev_flush_failed = 0;
|
| 449 | lc->header_location.bdev = lc->log_dev->bdev;
|
| 450 | lc->header_location.sector = 0;
|
| 451 |
|
| 452 | |
| 453 | * Buffer holds both header and bitset.
|
| 454 |
|
| 455 | buf_size =
|
| 456 | dm_round_up((LOG_OFFSET << SECTOR_SHIFT) + bitset_size,
|
| 457 | bdev_logical_block_size(lc->header_location.bdev));
|
| 458 |
|
| 459 | if (buf_size > bdev_nr_bytes(dev->bdev)) {
|
| 460 | DMWARN("log device %s too small: need %llu bytes",
|
| 461 | dev->name, (unsigned long long)buf_size);
|
| 462 | kfree(lc);
|
| 463 | return -EINVAL;
|
| 464 | }
|
| 465 |
|
| 466 | lc->header_location.count = buf_size >> SECTOR_SHIFT;
|
| 467 |
|
| 468 | lc->io_req.mem.type = DM_IO_VMA;
|
| 469 | lc->io_req.notify.fn = NULL;
|
| 470 | lc->io_req.client = dm_io_client_create();
|
| 471 | if (IS_ERR(lc->io_req.client)) {
|
| 472 | r = PTR_ERR(lc->io_req.client);
|
| 473 | DMWARN("couldn't allocate disk io client");
|
| 474 | kfree(lc);
|
| 475 | return r;
|
| 476 | }
|
| 477 |
|
| 478 | lc->disk_header = vmalloc(buf_size);
|
| 479 | if (!lc->disk_header) {
|
| 480 | DMWARN("couldn't allocate disk log buffer");
|
| 481 | dm_io_client_destroy(lc->io_req.client);
|
| 482 | kfree(lc);
|
| 483 | return -ENOMEM;
|
| 484 | }
|
| 485 |
|
| 486 | lc->io_req.mem.ptr.vma = lc->disk_header;
|
| 487 | lc->clean_bits = (void *)lc->disk_header +
|
| 488 | (LOG_OFFSET << SECTOR_SHIFT);
|
| 489 | }
|
| 490 |
|
| 491 | memset(lc->clean_bits, -1, bitset_size);
|
| 492 |
|
| 493 | lc->sync_bits = vmalloc(bitset_size);
|
| 494 | if (!lc->sync_bits) {
|
| 495 | DMWARN("couldn't allocate sync bitset");
|
| 496 | if (!dev)
|
| 497 | vfree(lc->clean_bits);
|
| 498 | else
|
| 499 | dm_io_client_destroy(lc->io_req.client);
|
| 500 | vfree(lc->disk_header);
|
| 501 | kfree(lc);
|
| 502 | return -ENOMEM;
|
| 503 | }
|
| 504 | memset(lc->sync_bits, (sync == NOSYNC) ? -1 : 0, bitset_size);
|
| 505 | lc->sync_count = (sync == NOSYNC) ? region_count : 0;
|
| 506 |
|
| 507 | lc->recovering_bits = vzalloc(bitset_size);
|
| 508 | if (!lc->recovering_bits) {
|
| 509 | DMWARN("couldn't allocate sync bitset");
|
| 510 | vfree(lc->sync_bits);
|
| 511 | if (!dev)
|
| 512 | vfree(lc->clean_bits);
|
| 513 | else
|
| 514 | dm_io_client_destroy(lc->io_req.client);
|
| 515 | vfree(lc->disk_header);
|
| 516 | kfree(lc);
|
| 517 | return -ENOMEM;
|
| 518 | }
|
| 519 | lc->sync_search = 0;
|
| 520 | log->context = lc;
|
| 521 |
|
| 522 | return 0;
|
| 523 | }
|
| 524 |
|
| 525 | static int core_ctr(struct dm_dirty_log *log, struct dm_target *ti,
|
| 526 | unsigned int argc, char **argv)
|
| 527 | {
|
| 528 | return create_log_context(log, ti, argc, argv, NULL);
|
| 529 | }
|
| 530 |
|
| 531 | static void destroy_log_context(struct log_c *lc)
|
| 532 | {
|
| 533 | vfree(lc->sync_bits);
|
| 534 | vfree(lc->recovering_bits);
|
| 535 | kfree(lc);
|
| 536 | }
|
| 537 |
|
| 538 | static void core_dtr(struct dm_dirty_log *log)
|
| 539 | {
|
| 540 | struct log_c *lc = log->context;
|
| 541 |
|
| 542 | vfree(lc->clean_bits);
|
| 543 | destroy_log_context(lc);
|
| 544 | }
|
| 545 |
|
| 546 | |
| 547 | *---------------------------------------------------------------------
|
| 548 | * disk log constructor/destructor
|
| 549 | *
|
| 550 | * argv contains log_device region_size followed optionally by [no]sync
|
| 551 | *---------------------------------------------------------------------
|
| 552 |
|
| 553 | static int disk_ctr(struct dm_dirty_log *log, struct dm_target *ti,
|
| 554 | unsigned int argc, char **argv)
|
| 555 | {
|
| 556 | int r;
|
| 557 | struct dm_dev *dev;
|
| 558 |
|
| 559 | if (argc < 2 || argc > 3) {
|
| 560 | DMWARN("wrong number of arguments to disk dirty region log");
|
| 561 | return -EINVAL;
|
| 562 | }
|
| 563 |
|
| 564 | r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &dev);
|
| 565 | if (r)
|
| 566 | return r;
|
| 567 |
|
| 568 | r = create_log_context(log, ti, argc - 1, argv + 1, dev);
|
| 569 | if (r) {
|
| 570 | dm_put_device(ti, dev);
|
| 571 | return r;
|
| 572 | }
|
| 573 |
|
| 574 | return 0;
|
| 575 | }
|
| 576 |
|
| 577 | static void disk_dtr(struct dm_dirty_log *log)
|
| 578 | {
|
| 579 | struct log_c *lc = log->context;
|
| 580 |
|
| 581 | dm_put_device(lc->ti, lc->log_dev);
|
| 582 | vfree(lc->disk_header);
|
| 583 | dm_io_client_destroy(lc->io_req.client);
|
| 584 | destroy_log_context(lc);
|
| 585 | }
|
| 586 |
|
| 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 |
|
| 596 | static int disk_resume(struct dm_dirty_log *log)
|
| 597 | {
|
| 598 | int r;
|
| 599 | unsigned int i;
|
| 600 | struct log_c *lc = log->context;
|
| 601 | size_t size = lc->bitset_uint32_count * sizeof(uint32_t);
|
| 602 |
|
| 603 |
|
| 604 | r = read_header(lc);
|
| 605 | if (r) {
|
| 606 | DMWARN("%s: Failed to read header on dirty region log device",
|
| 607 | lc->log_dev->name);
|
| 608 | fail_log_device(lc);
|
| 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;
|
| 617 | }
|
| 618 |
|
| 619 |
|
| 620 | if (lc->sync == NOSYNC)
|
| 621 | for (i = lc->header.nr_regions; i < lc->region_count; i++)
|
| 622 |
|
| 623 | log_set_bit(lc, lc->clean_bits, i);
|
| 624 | else
|
| 625 | for (i = lc->header.nr_regions; i < lc->region_count; i++)
|
| 626 |
|
| 627 | log_clear_bit(lc, lc->clean_bits, i);
|
| 628 |
|
| 629 |
|
| 630 | for (i = lc->region_count; i % BITS_PER_LONG; i++)
|
| 631 | log_clear_bit(lc, lc->clean_bits, i);
|
| 632 |
|
| 633 |
|
| 634 | memcpy(lc->sync_bits, lc->clean_bits, size);
|
| 635 | lc->sync_count = memweight(lc->clean_bits,
|
| 636 | lc->bitset_uint32_count * sizeof(uint32_t));
|
| 637 | lc->sync_search = 0;
|
| 638 |
|
| 639 |
|
| 640 | lc->header.nr_regions = lc->region_count;
|
| 641 |
|
| 642 | header_to_disk(&lc->header, lc->disk_header);
|
| 643 |
|
| 644 |
|
| 645 | r = rw_header(lc, REQ_OP_WRITE);
|
| 646 | if (!r) {
|
| 647 | r = flush_header(lc);
|
| 648 | if (r)
|
| 649 | lc->log_dev_flush_failed = 1;
|
| 650 | }
|
| 651 | if (r) {
|
| 652 | DMWARN("%s: Failed to write header on dirty region log device",
|
| 653 | lc->log_dev->name);
|
| 654 | fail_log_device(lc);
|
| 655 | }
|
| 656 |
|
| 657 | return r;
|
| 658 | }
|
| 659 |
|
| 660 | static uint32_t core_get_region_size(struct dm_dirty_log *log)
|
| 661 | {
|
| 662 | struct log_c *lc = log->context;
|
| 663 |
|
| 664 | return lc->region_size;
|
| 665 | }
|
| 666 |
|
| 667 | static int core_resume(struct dm_dirty_log *log)
|
| 668 | {
|
| 669 | struct log_c *lc = log->context;
|
| 670 |
|
| 671 | lc->sync_search = 0;
|
| 672 | return 0;
|
| 673 | }
|
| 674 |
|
| 675 | static int core_is_clean(struct dm_dirty_log *log, region_t region)
|
| 676 | {
|
| 677 | struct log_c *lc = log->context;
|
| 678 |
|
| 679 | return log_test_bit(lc->clean_bits, region);
|
| 680 | }
|
| 681 |
|
| 682 | static int core_in_sync(struct dm_dirty_log *log, region_t region, int block)
|
| 683 | {
|
| 684 | struct log_c *lc = log->context;
|
| 685 |
|
| 686 | return log_test_bit(lc->sync_bits, region);
|
| 687 | }
|
| 688 |
|
| 689 | static int core_flush(struct dm_dirty_log *log)
|
| 690 | {
|
| 691 |
|
| 692 | return 0;
|
| 693 | }
|
| 694 |
|
| 695 | static int disk_flush(struct dm_dirty_log *log)
|
| 696 | {
|
| 697 | int r, i;
|
| 698 | struct log_c *lc = log->context;
|
| 699 |
|
| 700 |
|
| 701 | if (!lc->touched_cleaned && !lc->touched_dirtied)
|
| 702 | return 0;
|
| 703 |
|
| 704 | if (lc->touched_cleaned && log->flush_callback_fn &&
|
| 705 | log->flush_callback_fn(lc->ti)) {
|
| 706 | |
| 707 | * At this point it is impossible to determine which
|
| 708 | * regions are clean and which are dirty (without
|
| 709 | * re-reading the log off disk). So mark all of them
|
| 710 | * dirty.
|
| 711 |
|
| 712 | lc->flush_failed = 1;
|
| 713 | for (i = 0; i < lc->region_count; i++)
|
| 714 | log_clear_bit(lc, lc->clean_bits, i);
|
| 715 | }
|
| 716 |
|
| 717 | r = rw_header(lc, REQ_OP_WRITE);
|
| 718 | if (r)
|
| 719 | fail_log_device(lc);
|
| 720 | else {
|
| 721 | if (lc->touched_dirtied) {
|
| 722 | r = flush_header(lc);
|
| 723 | if (r) {
|
| 724 | lc->log_dev_flush_failed = 1;
|
| 725 | fail_log_device(lc);
|
| 726 | } else
|
| 727 | lc->touched_dirtied = 0;
|
| 728 | }
|
| 729 | lc->touched_cleaned = 0;
|
| 730 | }
|
| 731 |
|
| 732 | return r;
|
| 733 | }
|
| 734 |
|
| 735 | static void core_mark_region(struct dm_dirty_log *log, region_t region)
|
| 736 | {
|
| 737 | struct log_c *lc = log->context;
|
| 738 |
|
| 739 | log_clear_bit(lc, lc->clean_bits, region);
|
| 740 | }
|
| 741 |
|
| 742 | static void core_clear_region(struct dm_dirty_log *log, region_t region)
|
| 743 | {
|
| 744 | struct log_c *lc = log->context;
|
| 745 |
|
| 746 | if (likely(!lc->flush_failed))
|
| 747 | log_set_bit(lc, lc->clean_bits, region);
|
| 748 | }
|
| 749 |
|
| 750 | static int core_get_resync_work(struct dm_dirty_log *log, region_t *region)
|
| 751 | {
|
| 752 | struct log_c *lc = log->context;
|
| 753 |
|
| 754 | if (lc->sync_search >= lc->region_count)
|
| 755 | return 0;
|
| 756 |
|
| 757 | do {
|
| 758 | *region = find_next_zero_bit_le(lc->sync_bits,
|
| 759 | lc->region_count,
|
| 760 | lc->sync_search);
|
| 761 | lc->sync_search = *region + 1;
|
| 762 |
|
| 763 | if (*region >= lc->region_count)
|
| 764 | return 0;
|
| 765 |
|
| 766 | } while (log_test_bit(lc->recovering_bits, *region));
|
| 767 |
|
| 768 | log_set_bit(lc, lc->recovering_bits, *region);
|
| 769 | return 1;
|
| 770 | }
|
| 771 |
|
| 772 | static void core_set_region_sync(struct dm_dirty_log *log, region_t region,
|
| 773 | int in_sync)
|
| 774 | {
|
| 775 | struct log_c *lc = log->context;
|
| 776 |
|
| 777 | log_clear_bit(lc, lc->recovering_bits, region);
|
| 778 | if (in_sync) {
|
| 779 | log_set_bit(lc, lc->sync_bits, region);
|
| 780 | lc->sync_count++;
|
| 781 | } else if (log_test_bit(lc->sync_bits, region)) {
|
| 782 | lc->sync_count--;
|
| 783 | log_clear_bit(lc, lc->sync_bits, region);
|
| 784 | }
|
| 785 | }
|
| 786 |
|
| 787 | static region_t core_get_sync_count(struct dm_dirty_log *log)
|
| 788 | {
|
| 789 | struct log_c *lc = log->context;
|
| 790 |
|
| 791 | return lc->sync_count;
|
| 792 | }
|
| 793 |
|
| 794 | #define DMEMIT_SYNC \
|
| 795 | do { \
|
| 796 | if (lc->sync != DEFAULTSYNC) \
|
| 797 | DMEMIT("%ssync ", lc->sync == NOSYNC ? "no" : ""); \
|
| 798 | } while (0)
|
| 799 |
|
| 800 | static int core_status(struct dm_dirty_log *log, status_type_t status,
|
| 801 | char *result, unsigned int maxlen)
|
| 802 | {
|
| 803 | int sz = 0;
|
| 804 | struct log_c *lc = log->context;
|
| 805 |
|
| 806 | switch (status) {
|
| 807 | case STATUSTYPE_INFO:
|
| 808 | DMEMIT("1 %s", log->type->name);
|
| 809 | break;
|
| 810 |
|
| 811 | case STATUSTYPE_TABLE:
|
| 812 | DMEMIT("%s %u %u ", log->type->name,
|
| 813 | lc->sync == DEFAULTSYNC ? 1 : 2, lc->region_size);
|
| 814 | DMEMIT_SYNC;
|
| 815 | break;
|
| 816 |
|
| 817 | case STATUSTYPE_IMA:
|
| 818 | *result = '\0';
|
| 819 | break;
|
| 820 | }
|
| 821 |
|
| 822 | return sz;
|
| 823 | }
|
| 824 |
|
| 825 | static int disk_status(struct dm_dirty_log *log, status_type_t status,
|
| 826 | char *result, unsigned int maxlen)
|
| 827 | {
|
| 828 | int sz = 0;
|
| 829 | struct log_c *lc = log->context;
|
| 830 |
|
| 831 | switch (status) {
|
| 832 | case STATUSTYPE_INFO:
|
| 833 | DMEMIT("3 %s %s %c", log->type->name, lc->log_dev->name,
|
| 834 | lc->log_dev_flush_failed ? 'F' :
|
| 835 | lc->log_dev_failed ? 'D' :
|
| 836 | 'A');
|
| 837 | break;
|
| 838 |
|
| 839 | case STATUSTYPE_TABLE:
|
| 840 | DMEMIT("%s %u %s %u ", log->type->name,
|
| 841 | lc->sync == DEFAULTSYNC ? 2 : 3, lc->log_dev->name,
|
| 842 | lc->region_size);
|
| 843 | DMEMIT_SYNC;
|
| 844 | break;
|
| 845 |
|
| 846 | case STATUSTYPE_IMA:
|
| 847 | *result = '\0';
|
| 848 | break;
|
| 849 | }
|
| 850 |
|
| 851 | return sz;
|
| 852 | }
|
| 853 |
|
| 854 | static struct dm_dirty_log_type _core_type = {
|
| 855 | .name = "core",
|
| 856 | .module = THIS_MODULE,
|
| 857 | .ctr = core_ctr,
|
| 858 | .dtr = core_dtr,
|
| 859 | .resume = core_resume,
|
| 860 | .get_region_size = core_get_region_size,
|
| 861 | .is_clean = core_is_clean,
|
| 862 | .in_sync = core_in_sync,
|
| 863 | .flush = core_flush,
|
| 864 | .mark_region = core_mark_region,
|
| 865 | .clear_region = core_clear_region,
|
| 866 | .get_resync_work = core_get_resync_work,
|
| 867 | .set_region_sync = core_set_region_sync,
|
| 868 | .get_sync_count = core_get_sync_count,
|
| 869 | .status = core_status,
|
| 870 | };
|
| 871 |
|
| 872 | static struct dm_dirty_log_type _disk_type = {
|
| 873 | .name = "disk",
|
| 874 | .module = THIS_MODULE,
|
| 875 | .ctr = disk_ctr,
|
| 876 | .dtr = disk_dtr,
|
| 877 | .postsuspend = disk_flush,
|
| 878 | .resume = disk_resume,
|
| 879 | .get_region_size = core_get_region_size,
|
| 880 | .is_clean = core_is_clean,
|
| 881 | .in_sync = core_in_sync,
|
| 882 | .flush = disk_flush,
|
| 883 | .mark_region = core_mark_region,
|
| 884 | .clear_region = core_clear_region,
|
| 885 | .get_resync_work = core_get_resync_work,
|
| 886 | .set_region_sync = core_set_region_sync,
|
| 887 | .get_sync_count = core_get_sync_count,
|
| 888 | .status = disk_status,
|
| 889 | };
|
| 890 |
|
| 891 | static int __init dm_dirty_log_init(void)
|
| 892 | {
|
| 893 | int r;
|
| 894 |
|
| 895 | r = dm_dirty_log_type_register(&_core_type);
|
| 896 | if (r)
|
| 897 | DMWARN("couldn't register core log");
|
| 898 |
|
| 899 | r = dm_dirty_log_type_register(&_disk_type);
|
| 900 | if (r) {
|
| 901 | DMWARN("couldn't register disk type");
|
| 902 | dm_dirty_log_type_unregister(&_core_type);
|
| 903 | }
|
| 904 |
|
| 905 | return r;
|
| 906 | }
|
| 907 |
|
| 908 | static void __exit dm_dirty_log_exit(void)
|
| 909 | {
|
| 910 | dm_dirty_log_type_unregister(&_disk_type);
|
| 911 | dm_dirty_log_type_unregister(&_core_type);
|
| 912 | }
|
| 913 |
|
| 914 | module_init(dm_dirty_log_init);
|
| 915 | module_exit(dm_dirty_log_exit);
|
| 916 |
|
| 917 | MODULE_DESCRIPTION(DM_NAME " dirty region log");
|
| 918 | MODULE_AUTHOR("Joe Thornber, Heinz Mauelshagen <[email protected]>");
|
| 919 | MODULE_LICENSE("GPL");
|