1 /* 2 * Copyright (C) 2011 Red Hat, Inc. 3 * 4 * This file is released under the GPL. 5 */ 6 7 #include "dm-space-map-common.h" 8 #include "dm-transaction-manager.h" 9 10 #include <linux/bitops.h> 11 #include <linux/device-mapper.h> 12 13 #define DM_MSG_PREFIX "space map common" 14 15 /*----------------------------------------------------------------*/ 16 17 /* 18 * Index validator. 19 */ 20 #define INDEX_CSUM_XOR 160478 21 22 static void index_prepare_for_write(struct dm_block_validator *v, 23 struct dm_block *b, 24 size_t block_size) 25 { 26 struct disk_metadata_index *mi_le = dm_block_data(b); 27 28 mi_le->blocknr = cpu_to_le64(dm_block_location(b)); 29 mi_le->csum = cpu_to_le32(dm_bm_checksum(&mi_le->padding, 30 block_size - sizeof(__le32), 31 INDEX_CSUM_XOR)); 32 } 33 34 static int index_check(struct dm_block_validator *v, 35 struct dm_block *b, 36 size_t block_size) 37 { 38 struct disk_metadata_index *mi_le = dm_block_data(b); 39 __le32 csum_disk; 40 41 if (dm_block_location(b) != le64_to_cpu(mi_le->blocknr)) { 42 DMERR("index_check failed blocknr %llu wanted %llu", 43 le64_to_cpu(mi_le->blocknr), dm_block_location(b)); 44 return -ENOTBLK; 45 } 46 47 csum_disk = cpu_to_le32(dm_bm_checksum(&mi_le->padding, 48 block_size - sizeof(__le32), 49 INDEX_CSUM_XOR)); 50 if (csum_disk != mi_le->csum) { 51 DMERR("index_check failed csum %u wanted %u", 52 le32_to_cpu(csum_disk), le32_to_cpu(mi_le->csum)); 53 return -EILSEQ; 54 } 55 56 return 0; 57 } 58 59 static struct dm_block_validator index_validator = { 60 .name = "index", 61 .prepare_for_write = index_prepare_for_write, 62 .check = index_check 63 }; 64 65 /*----------------------------------------------------------------*/ 66 67 /* 68 * Bitmap validator 69 */ 70 #define BITMAP_CSUM_XOR 240779 71 72 static void bitmap_prepare_for_write(struct dm_block_validator *v, 73 struct dm_block *b, 74 size_t block_size) 75 { 76 struct disk_bitmap_header *disk_header = dm_block_data(b); 77 78 disk_header->blocknr = cpu_to_le64(dm_block_location(b)); 79 disk_header->csum = cpu_to_le32(dm_bm_checksum(&disk_header->not_used, 80 block_size - sizeof(__le32), 81 BITMAP_CSUM_XOR)); 82 } 83 84 static int bitmap_check(struct dm_block_validator *v, 85 struct dm_block *b, 86 size_t block_size) 87 { 88 struct disk_bitmap_header *disk_header = dm_block_data(b); 89 __le32 csum_disk; 90 91 if (dm_block_location(b) != le64_to_cpu(disk_header->blocknr)) { 92 DMERR("bitmap check failed blocknr %llu wanted %llu", 93 le64_to_cpu(disk_header->blocknr), dm_block_location(b)); 94 return -ENOTBLK; 95 } 96 97 csum_disk = cpu_to_le32(dm_bm_checksum(&disk_header->not_used, 98 block_size - sizeof(__le32), 99 BITMAP_CSUM_XOR)); 100 if (csum_disk != disk_header->csum) { 101 DMERR("bitmap check failed csum %u wanted %u", 102 le32_to_cpu(csum_disk), le32_to_cpu(disk_header->csum)); 103 return -EILSEQ; 104 } 105 106 return 0; 107 } 108 109 static struct dm_block_validator dm_sm_bitmap_validator = { 110 .name = "sm_bitmap", 111 .prepare_for_write = bitmap_prepare_for_write, 112 .check = bitmap_check 113 }; 114 115 /*----------------------------------------------------------------*/ 116 117 #define ENTRIES_PER_WORD 32 118 #define ENTRIES_SHIFT 5 119 120 static void *dm_bitmap_data(struct dm_block *b) 121 { 122 return dm_block_data(b) + sizeof(struct disk_bitmap_header); 123 } 124 125 #define WORD_MASK_HIGH 0xAAAAAAAAAAAAAAAAULL 126 127 static unsigned bitmap_word_used(void *addr, unsigned b) 128 { 129 __le64 *words_le = addr; 130 __le64 *w_le = words_le + (b >> ENTRIES_SHIFT); 131 132 uint64_t bits = le64_to_cpu(*w_le); 133 uint64_t mask = (bits + WORD_MASK_HIGH + 1) & WORD_MASK_HIGH; 134 135 return !(~bits & mask); 136 } 137 138 static unsigned sm_lookup_bitmap(void *addr, unsigned b) 139 { 140 __le64 *words_le = addr; 141 __le64 *w_le = words_le + (b >> ENTRIES_SHIFT); 142 unsigned hi, lo; 143 144 b = (b & (ENTRIES_PER_WORD - 1)) << 1; 145 hi = !!test_bit_le(b, (void *) w_le); 146 lo = !!test_bit_le(b + 1, (void *) w_le); 147 return (hi << 1) | lo; 148 } 149 150 static void sm_set_bitmap(void *addr, unsigned b, unsigned val) 151 { 152 __le64 *words_le = addr; 153 __le64 *w_le = words_le + (b >> ENTRIES_SHIFT); 154 155 b = (b & (ENTRIES_PER_WORD - 1)) << 1; 156 157 if (val & 2) 158 __set_bit_le(b, (void *) w_le); 159 else 160 __clear_bit_le(b, (void *) w_le); 161 162 if (val & 1) 163 __set_bit_le(b + 1, (void *) w_le); 164 else 165 __clear_bit_le(b + 1, (void *) w_le); 166 } 167 168 static int sm_find_free(void *addr, unsigned begin, unsigned end, 169 unsigned *result) 170 { 171 while (begin < end) { 172 if (!(begin & (ENTRIES_PER_WORD - 1)) && 173 bitmap_word_used(addr, begin)) { 174 begin += ENTRIES_PER_WORD; 175 continue; 176 } 177 178 if (!sm_lookup_bitmap(addr, begin)) { 179 *result = begin; 180 return 0; 181 } 182 183 begin++; 184 } 185 186 return -ENOSPC; 187 } 188 189 /*----------------------------------------------------------------*/ 190 191 static int sm_ll_init(struct ll_disk *ll, struct dm_transaction_manager *tm) 192 { 193 ll->tm = tm; 194 195 ll->bitmap_info.tm = tm; 196 ll->bitmap_info.levels = 1; 197 198 /* 199 * Because the new bitmap blocks are created via a shadow 200 * operation, the old entry has already had its reference count 201 * decremented and we don't need the btree to do any bookkeeping. 202 */ 203 ll->bitmap_info.value_type.size = sizeof(struct disk_index_entry); 204 ll->bitmap_info.value_type.inc = NULL; 205 ll->bitmap_info.value_type.dec = NULL; 206 ll->bitmap_info.value_type.equal = NULL; 207 208 ll->ref_count_info.tm = tm; 209 ll->ref_count_info.levels = 1; 210 ll->ref_count_info.value_type.size = sizeof(uint32_t); 211 ll->ref_count_info.value_type.inc = NULL; 212 ll->ref_count_info.value_type.dec = NULL; 213 ll->ref_count_info.value_type.equal = NULL; 214 215 ll->block_size = dm_bm_block_size(dm_tm_get_bm(tm)); 216 217 if (ll->block_size > (1 << 30)) { 218 DMERR("block size too big to hold bitmaps"); 219 return -EINVAL; 220 } 221 222 ll->entries_per_block = (ll->block_size - sizeof(struct disk_bitmap_header)) * 223 ENTRIES_PER_BYTE; 224 ll->nr_blocks = 0; 225 ll->bitmap_root = 0; 226 ll->ref_count_root = 0; 227 ll->bitmap_index_changed = false; 228 229 return 0; 230 } 231 232 int sm_ll_extend(struct ll_disk *ll, dm_block_t extra_blocks) 233 { 234 int r; 235 dm_block_t i, nr_blocks, nr_indexes; 236 unsigned old_blocks, blocks; 237 238 nr_blocks = ll->nr_blocks + extra_blocks; 239 old_blocks = dm_sector_div_up(ll->nr_blocks, ll->entries_per_block); 240 blocks = dm_sector_div_up(nr_blocks, ll->entries_per_block); 241 242 nr_indexes = dm_sector_div_up(nr_blocks, ll->entries_per_block); 243 if (nr_indexes > ll->max_entries(ll)) { 244 DMERR("space map too large"); 245 return -EINVAL; 246 } 247 248 for (i = old_blocks; i < blocks; i++) { 249 struct dm_block *b; 250 struct disk_index_entry idx; 251 252 r = dm_tm_new_block(ll->tm, &dm_sm_bitmap_validator, &b); 253 if (r < 0) 254 return r; 255 idx.blocknr = cpu_to_le64(dm_block_location(b)); 256 257 r = dm_tm_unlock(ll->tm, b); 258 if (r < 0) 259 return r; 260 261 idx.nr_free = cpu_to_le32(ll->entries_per_block); 262 idx.none_free_before = 0; 263 264 r = ll->save_ie(ll, i, &idx); 265 if (r < 0) 266 return r; 267 } 268 269 ll->nr_blocks = nr_blocks; 270 return 0; 271 } 272 273 int sm_ll_lookup_bitmap(struct ll_disk *ll, dm_block_t b, uint32_t *result) 274 { 275 int r; 276 dm_block_t index = b; 277 struct disk_index_entry ie_disk; 278 struct dm_block *blk; 279 280 b = do_div(index, ll->entries_per_block); 281 r = ll->load_ie(ll, index, &ie_disk); 282 if (r < 0) 283 return r; 284 285 r = dm_tm_read_lock(ll->tm, le64_to_cpu(ie_disk.blocknr), 286 &dm_sm_bitmap_validator, &blk); 287 if (r < 0) 288 return r; 289 290 *result = sm_lookup_bitmap(dm_bitmap_data(blk), b); 291 292 return dm_tm_unlock(ll->tm, blk); 293 } 294 295 int sm_ll_lookup(struct ll_disk *ll, dm_block_t b, uint32_t *result) 296 { 297 __le32 le_rc; 298 int r = sm_ll_lookup_bitmap(ll, b, result); 299 300 if (r) 301 return r; 302 303 if (*result != 3) 304 return r; 305 306 r = dm_btree_lookup(&ll->ref_count_info, ll->ref_count_root, &b, &le_rc); 307 if (r < 0) 308 return r; 309 310 *result = le32_to_cpu(le_rc); 311 312 return r; 313 } 314 315 int sm_ll_find_free_block(struct ll_disk *ll, dm_block_t begin, 316 dm_block_t end, dm_block_t *result) 317 { 318 int r; 319 struct disk_index_entry ie_disk; 320 dm_block_t i, index_begin = begin; 321 dm_block_t index_end = dm_sector_div_up(end, ll->entries_per_block); 322 323 /* 324 * FIXME: Use shifts 325 */ 326 begin = do_div(index_begin, ll->entries_per_block); 327 end = do_div(end, ll->entries_per_block); 328 329 for (i = index_begin; i < index_end; i++, begin = 0) { 330 struct dm_block *blk; 331 unsigned position; 332 uint32_t bit_end; 333 334 r = ll->load_ie(ll, i, &ie_disk); 335 if (r < 0) 336 return r; 337 338 if (le32_to_cpu(ie_disk.nr_free) == 0) 339 continue; 340 341 r = dm_tm_read_lock(ll->tm, le64_to_cpu(ie_disk.blocknr), 342 &dm_sm_bitmap_validator, &blk); 343 if (r < 0) 344 return r; 345 346 bit_end = (i == index_end - 1) ? end : ll->entries_per_block; 347 348 r = sm_find_free(dm_bitmap_data(blk), 349 max_t(unsigned, begin, le32_to_cpu(ie_disk.none_free_before)), 350 bit_end, &position); 351 if (r == -ENOSPC) { 352 /* 353 * This might happen because we started searching 354 * part way through the bitmap. 355 */ 356 dm_tm_unlock(ll->tm, blk); 357 continue; 358 359 } else if (r < 0) { 360 dm_tm_unlock(ll->tm, blk); 361 return r; 362 } 363 364 r = dm_tm_unlock(ll->tm, blk); 365 if (r < 0) 366 return r; 367 368 *result = i * ll->entries_per_block + (dm_block_t) position; 369 return 0; 370 } 371 372 return -ENOSPC; 373 } 374 375 int sm_ll_insert(struct ll_disk *ll, dm_block_t b, 376 uint32_t ref_count, enum allocation_event *ev) 377 { 378 int r; 379 uint32_t bit, old; 380 struct dm_block *nb; 381 dm_block_t index = b; 382 struct disk_index_entry ie_disk; 383 void *bm_le; 384 int inc; 385 386 bit = do_div(index, ll->entries_per_block); 387 r = ll->load_ie(ll, index, &ie_disk); 388 if (r < 0) 389 return r; 390 391 r = dm_tm_shadow_block(ll->tm, le64_to_cpu(ie_disk.blocknr), 392 &dm_sm_bitmap_validator, &nb, &inc); 393 if (r < 0) { 394 DMERR("dm_tm_shadow_block() failed"); 395 return r; 396 } 397 ie_disk.blocknr = cpu_to_le64(dm_block_location(nb)); 398 399 bm_le = dm_bitmap_data(nb); 400 old = sm_lookup_bitmap(bm_le, bit); 401 402 if (ref_count <= 2) { 403 sm_set_bitmap(bm_le, bit, ref_count); 404 405 r = dm_tm_unlock(ll->tm, nb); 406 if (r < 0) 407 return r; 408 409 if (old > 2) { 410 r = dm_btree_remove(&ll->ref_count_info, 411 ll->ref_count_root, 412 &b, &ll->ref_count_root); 413 if (r) 414 return r; 415 } 416 417 } else { 418 __le32 le_rc = cpu_to_le32(ref_count); 419 420 sm_set_bitmap(bm_le, bit, 3); 421 r = dm_tm_unlock(ll->tm, nb); 422 if (r < 0) 423 return r; 424 425 __dm_bless_for_disk(&le_rc); 426 r = dm_btree_insert(&ll->ref_count_info, ll->ref_count_root, 427 &b, &le_rc, &ll->ref_count_root); 428 if (r < 0) { 429 DMERR("ref count insert failed"); 430 return r; 431 } 432 } 433 434 if (ref_count && !old) { 435 *ev = SM_ALLOC; 436 ll->nr_allocated++; 437 le32_add_cpu(&ie_disk.nr_free, -1); 438 if (le32_to_cpu(ie_disk.none_free_before) == bit) 439 ie_disk.none_free_before = cpu_to_le32(bit + 1); 440 441 } else if (old && !ref_count) { 442 *ev = SM_FREE; 443 ll->nr_allocated--; 444 le32_add_cpu(&ie_disk.nr_free, 1); 445 ie_disk.none_free_before = cpu_to_le32(min(le32_to_cpu(ie_disk.none_free_before), bit)); 446 } 447 448 return ll->save_ie(ll, index, &ie_disk); 449 } 450 451 int sm_ll_inc(struct ll_disk *ll, dm_block_t b, enum allocation_event *ev) 452 { 453 int r; 454 uint32_t rc; 455 456 r = sm_ll_lookup(ll, b, &rc); 457 if (r) 458 return r; 459 460 return sm_ll_insert(ll, b, rc + 1, ev); 461 } 462 463 int sm_ll_dec(struct ll_disk *ll, dm_block_t b, enum allocation_event *ev) 464 { 465 int r; 466 uint32_t rc; 467 468 r = sm_ll_lookup(ll, b, &rc); 469 if (r) 470 return r; 471 472 if (!rc) 473 return -EINVAL; 474 475 return sm_ll_insert(ll, b, rc - 1, ev); 476 } 477 478 int sm_ll_commit(struct ll_disk *ll) 479 { 480 int r = 0; 481 482 if (ll->bitmap_index_changed) { 483 r = ll->commit(ll); 484 if (!r) 485 ll->bitmap_index_changed = false; 486 } 487 488 return r; 489 } 490 491 /*----------------------------------------------------------------*/ 492 493 static int metadata_ll_load_ie(struct ll_disk *ll, dm_block_t index, 494 struct disk_index_entry *ie) 495 { 496 memcpy(ie, ll->mi_le.index + index, sizeof(*ie)); 497 return 0; 498 } 499 500 static int metadata_ll_save_ie(struct ll_disk *ll, dm_block_t index, 501 struct disk_index_entry *ie) 502 { 503 ll->bitmap_index_changed = true; 504 memcpy(ll->mi_le.index + index, ie, sizeof(*ie)); 505 return 0; 506 } 507 508 static int metadata_ll_init_index(struct ll_disk *ll) 509 { 510 int r; 511 struct dm_block *b; 512 513 r = dm_tm_new_block(ll->tm, &index_validator, &b); 514 if (r < 0) 515 return r; 516 517 memcpy(dm_block_data(b), &ll->mi_le, sizeof(ll->mi_le)); 518 ll->bitmap_root = dm_block_location(b); 519 520 return dm_tm_unlock(ll->tm, b); 521 } 522 523 static int metadata_ll_open(struct ll_disk *ll) 524 { 525 int r; 526 struct dm_block *block; 527 528 r = dm_tm_read_lock(ll->tm, ll->bitmap_root, 529 &index_validator, &block); 530 if (r) 531 return r; 532 533 memcpy(&ll->mi_le, dm_block_data(block), sizeof(ll->mi_le)); 534 return dm_tm_unlock(ll->tm, block); 535 } 536 537 static dm_block_t metadata_ll_max_entries(struct ll_disk *ll) 538 { 539 return MAX_METADATA_BITMAPS; 540 } 541 542 static int metadata_ll_commit(struct ll_disk *ll) 543 { 544 int r, inc; 545 struct dm_block *b; 546 547 r = dm_tm_shadow_block(ll->tm, ll->bitmap_root, &index_validator, &b, &inc); 548 if (r) 549 return r; 550 551 memcpy(dm_block_data(b), &ll->mi_le, sizeof(ll->mi_le)); 552 ll->bitmap_root = dm_block_location(b); 553 554 return dm_tm_unlock(ll->tm, b); 555 } 556 557 int sm_ll_new_metadata(struct ll_disk *ll, struct dm_transaction_manager *tm) 558 { 559 int r; 560 561 r = sm_ll_init(ll, tm); 562 if (r < 0) 563 return r; 564 565 ll->load_ie = metadata_ll_load_ie; 566 ll->save_ie = metadata_ll_save_ie; 567 ll->init_index = metadata_ll_init_index; 568 ll->open_index = metadata_ll_open; 569 ll->max_entries = metadata_ll_max_entries; 570 ll->commit = metadata_ll_commit; 571 572 ll->nr_blocks = 0; 573 ll->nr_allocated = 0; 574 575 r = ll->init_index(ll); 576 if (r < 0) 577 return r; 578 579 r = dm_btree_empty(&ll->ref_count_info, &ll->ref_count_root); 580 if (r < 0) 581 return r; 582 583 return 0; 584 } 585 586 int sm_ll_open_metadata(struct ll_disk *ll, struct dm_transaction_manager *tm, 587 void *root_le, size_t len) 588 { 589 int r; 590 struct disk_sm_root *smr = root_le; 591 592 if (len < sizeof(struct disk_sm_root)) { 593 DMERR("sm_metadata root too small"); 594 return -ENOMEM; 595 } 596 597 r = sm_ll_init(ll, tm); 598 if (r < 0) 599 return r; 600 601 ll->load_ie = metadata_ll_load_ie; 602 ll->save_ie = metadata_ll_save_ie; 603 ll->init_index = metadata_ll_init_index; 604 ll->open_index = metadata_ll_open; 605 ll->max_entries = metadata_ll_max_entries; 606 ll->commit = metadata_ll_commit; 607 608 ll->nr_blocks = le64_to_cpu(smr->nr_blocks); 609 ll->nr_allocated = le64_to_cpu(smr->nr_allocated); 610 ll->bitmap_root = le64_to_cpu(smr->bitmap_root); 611 ll->ref_count_root = le64_to_cpu(smr->ref_count_root); 612 613 return ll->open_index(ll); 614 } 615 616 /*----------------------------------------------------------------*/ 617 618 static int disk_ll_load_ie(struct ll_disk *ll, dm_block_t index, 619 struct disk_index_entry *ie) 620 { 621 return dm_btree_lookup(&ll->bitmap_info, ll->bitmap_root, &index, ie); 622 } 623 624 static int disk_ll_save_ie(struct ll_disk *ll, dm_block_t index, 625 struct disk_index_entry *ie) 626 { 627 __dm_bless_for_disk(ie); 628 return dm_btree_insert(&ll->bitmap_info, ll->bitmap_root, 629 &index, ie, &ll->bitmap_root); 630 } 631 632 static int disk_ll_init_index(struct ll_disk *ll) 633 { 634 return dm_btree_empty(&ll->bitmap_info, &ll->bitmap_root); 635 } 636 637 static int disk_ll_open(struct ll_disk *ll) 638 { 639 /* nothing to do */ 640 return 0; 641 } 642 643 static dm_block_t disk_ll_max_entries(struct ll_disk *ll) 644 { 645 return -1ULL; 646 } 647 648 static int disk_ll_commit(struct ll_disk *ll) 649 { 650 return 0; 651 } 652 653 int sm_ll_new_disk(struct ll_disk *ll, struct dm_transaction_manager *tm) 654 { 655 int r; 656 657 r = sm_ll_init(ll, tm); 658 if (r < 0) 659 return r; 660 661 ll->load_ie = disk_ll_load_ie; 662 ll->save_ie = disk_ll_save_ie; 663 ll->init_index = disk_ll_init_index; 664 ll->open_index = disk_ll_open; 665 ll->max_entries = disk_ll_max_entries; 666 ll->commit = disk_ll_commit; 667 668 ll->nr_blocks = 0; 669 ll->nr_allocated = 0; 670 671 r = ll->init_index(ll); 672 if (r < 0) 673 return r; 674 675 r = dm_btree_empty(&ll->ref_count_info, &ll->ref_count_root); 676 if (r < 0) 677 return r; 678 679 return 0; 680 } 681 682 int sm_ll_open_disk(struct ll_disk *ll, struct dm_transaction_manager *tm, 683 void *root_le, size_t len) 684 { 685 int r; 686 struct disk_sm_root *smr = root_le; 687 688 if (len < sizeof(struct disk_sm_root)) { 689 DMERR("sm_metadata root too small"); 690 return -ENOMEM; 691 } 692 693 r = sm_ll_init(ll, tm); 694 if (r < 0) 695 return r; 696 697 ll->load_ie = disk_ll_load_ie; 698 ll->save_ie = disk_ll_save_ie; 699 ll->init_index = disk_ll_init_index; 700 ll->open_index = disk_ll_open; 701 ll->max_entries = disk_ll_max_entries; 702 ll->commit = disk_ll_commit; 703 704 ll->nr_blocks = le64_to_cpu(smr->nr_blocks); 705 ll->nr_allocated = le64_to_cpu(smr->nr_allocated); 706 ll->bitmap_root = le64_to_cpu(smr->bitmap_root); 707 ll->ref_count_root = le64_to_cpu(smr->ref_count_root); 708 709 return ll->open_index(ll); 710 } 711 712 /*----------------------------------------------------------------*/ 713