1 /* 2 * Block driver for the QCOW version 2 format 3 * 4 * Copyright (c) 2004-2006 Fabrice Bellard 5 * 6 * Permission is hereby granted, free of charge, to any person obtaining a copy 7 * of this software and associated documentation files (the "Software"), to deal 8 * in the Software without restriction, including without limitation the rights 9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 10 * copies of the Software, and to permit persons to whom the Software is 11 * furnished to do so, subject to the following conditions: 12 * 13 * The above copyright notice and this permission notice shall be included in 14 * all copies or substantial portions of the Software. 15 * 16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN 22 * THE SOFTWARE. 23 */ 24 25 #include "qemu-common.h" 26 #include "block_int.h" 27 #include "block/qcow2.h" 28 29 static int64_t alloc_clusters_noref(BlockDriverState *bs, int64_t size); 30 static int QEMU_WARN_UNUSED_RESULT update_refcount(BlockDriverState *bs, 31 int64_t offset, int64_t length, 32 int addend); 33 34 35 /*********************************************************/ 36 /* refcount handling */ 37 38 int qcow2_refcount_init(BlockDriverState *bs) 39 { 40 BDRVQcowState *s = bs->opaque; 41 int ret, refcount_table_size2, i; 42 43 refcount_table_size2 = s->refcount_table_size * sizeof(uint64_t); 44 s->refcount_table = g_malloc(refcount_table_size2); 45 if (s->refcount_table_size > 0) { 46 BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_LOAD); 47 ret = bdrv_pread(bs->file, s->refcount_table_offset, 48 s->refcount_table, refcount_table_size2); 49 if (ret != refcount_table_size2) 50 goto fail; 51 for(i = 0; i < s->refcount_table_size; i++) 52 be64_to_cpus(&s->refcount_table[i]); 53 } 54 return 0; 55 fail: 56 return -ENOMEM; 57 } 58 59 void qcow2_refcount_close(BlockDriverState *bs) 60 { 61 BDRVQcowState *s = bs->opaque; 62 g_free(s->refcount_table); 63 } 64 65 66 static int load_refcount_block(BlockDriverState *bs, 67 int64_t refcount_block_offset, 68 void **refcount_block) 69 { 70 BDRVQcowState *s = bs->opaque; 71 int ret; 72 73 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_LOAD); 74 ret = qcow2_cache_get(bs, s->refcount_block_cache, refcount_block_offset, 75 refcount_block); 76 77 return ret; 78 } 79 80 /* 81 * Returns the refcount of the cluster given by its index. Any non-negative 82 * return value is the refcount of the cluster, negative values are -errno 83 * and indicate an error. 84 */ 85 static int get_refcount(BlockDriverState *bs, int64_t cluster_index) 86 { 87 BDRVQcowState *s = bs->opaque; 88 int refcount_table_index, block_index; 89 int64_t refcount_block_offset; 90 int ret; 91 uint16_t *refcount_block; 92 uint16_t refcount; 93 94 refcount_table_index = cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT); 95 if (refcount_table_index >= s->refcount_table_size) 96 return 0; 97 refcount_block_offset = s->refcount_table[refcount_table_index]; 98 if (!refcount_block_offset) 99 return 0; 100 101 ret = qcow2_cache_get(bs, s->refcount_block_cache, refcount_block_offset, 102 (void**) &refcount_block); 103 if (ret < 0) { 104 return ret; 105 } 106 107 block_index = cluster_index & 108 ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1); 109 refcount = be16_to_cpu(refcount_block[block_index]); 110 111 ret = qcow2_cache_put(bs, s->refcount_block_cache, 112 (void**) &refcount_block); 113 if (ret < 0) { 114 return ret; 115 } 116 117 return refcount; 118 } 119 120 /* 121 * Rounds the refcount table size up to avoid growing the table for each single 122 * refcount block that is allocated. 123 */ 124 static unsigned int next_refcount_table_size(BDRVQcowState *s, 125 unsigned int min_size) 126 { 127 unsigned int min_clusters = (min_size >> (s->cluster_bits - 3)) + 1; 128 unsigned int refcount_table_clusters = 129 MAX(1, s->refcount_table_size >> (s->cluster_bits - 3)); 130 131 while (min_clusters > refcount_table_clusters) { 132 refcount_table_clusters = (refcount_table_clusters * 3 + 1) / 2; 133 } 134 135 return refcount_table_clusters << (s->cluster_bits - 3); 136 } 137 138 139 /* Checks if two offsets are described by the same refcount block */ 140 static int in_same_refcount_block(BDRVQcowState *s, uint64_t offset_a, 141 uint64_t offset_b) 142 { 143 uint64_t block_a = offset_a >> (2 * s->cluster_bits - REFCOUNT_SHIFT); 144 uint64_t block_b = offset_b >> (2 * s->cluster_bits - REFCOUNT_SHIFT); 145 146 return (block_a == block_b); 147 } 148 149 /* 150 * Loads a refcount block. If it doesn't exist yet, it is allocated first 151 * (including growing the refcount table if needed). 152 * 153 * Returns 0 on success or -errno in error case 154 */ 155 static int alloc_refcount_block(BlockDriverState *bs, 156 int64_t cluster_index, uint16_t **refcount_block) 157 { 158 BDRVQcowState *s = bs->opaque; 159 unsigned int refcount_table_index; 160 int ret; 161 162 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC); 163 164 /* Find the refcount block for the given cluster */ 165 refcount_table_index = cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT); 166 167 if (refcount_table_index < s->refcount_table_size) { 168 169 uint64_t refcount_block_offset = 170 s->refcount_table[refcount_table_index] & REFT_OFFSET_MASK; 171 172 /* If it's already there, we're done */ 173 if (refcount_block_offset) { 174 return load_refcount_block(bs, refcount_block_offset, 175 (void**) refcount_block); 176 } 177 } 178 179 /* 180 * If we came here, we need to allocate something. Something is at least 181 * a cluster for the new refcount block. It may also include a new refcount 182 * table if the old refcount table is too small. 183 * 184 * Note that allocating clusters here needs some special care: 185 * 186 * - We can't use the normal qcow2_alloc_clusters(), it would try to 187 * increase the refcount and very likely we would end up with an endless 188 * recursion. Instead we must place the refcount blocks in a way that 189 * they can describe them themselves. 190 * 191 * - We need to consider that at this point we are inside update_refcounts 192 * and doing the initial refcount increase. This means that some clusters 193 * have already been allocated by the caller, but their refcount isn't 194 * accurate yet. free_cluster_index tells us where this allocation ends 195 * as long as we don't overwrite it by freeing clusters. 196 * 197 * - alloc_clusters_noref and qcow2_free_clusters may load a different 198 * refcount block into the cache 199 */ 200 201 *refcount_block = NULL; 202 203 /* We write to the refcount table, so we might depend on L2 tables */ 204 qcow2_cache_flush(bs, s->l2_table_cache); 205 206 /* Allocate the refcount block itself and mark it as used */ 207 int64_t new_block = alloc_clusters_noref(bs, s->cluster_size); 208 if (new_block < 0) { 209 return new_block; 210 } 211 212 #ifdef DEBUG_ALLOC2 213 fprintf(stderr, "qcow2: Allocate refcount block %d for %" PRIx64 214 " at %" PRIx64 "\n", 215 refcount_table_index, cluster_index << s->cluster_bits, new_block); 216 #endif 217 218 if (in_same_refcount_block(s, new_block, cluster_index << s->cluster_bits)) { 219 /* Zero the new refcount block before updating it */ 220 ret = qcow2_cache_get_empty(bs, s->refcount_block_cache, new_block, 221 (void**) refcount_block); 222 if (ret < 0) { 223 goto fail_block; 224 } 225 226 memset(*refcount_block, 0, s->cluster_size); 227 228 /* The block describes itself, need to update the cache */ 229 int block_index = (new_block >> s->cluster_bits) & 230 ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1); 231 (*refcount_block)[block_index] = cpu_to_be16(1); 232 } else { 233 /* Described somewhere else. This can recurse at most twice before we 234 * arrive at a block that describes itself. */ 235 ret = update_refcount(bs, new_block, s->cluster_size, 1); 236 if (ret < 0) { 237 goto fail_block; 238 } 239 240 bdrv_flush(bs->file); 241 242 /* Initialize the new refcount block only after updating its refcount, 243 * update_refcount uses the refcount cache itself */ 244 ret = qcow2_cache_get_empty(bs, s->refcount_block_cache, new_block, 245 (void**) refcount_block); 246 if (ret < 0) { 247 goto fail_block; 248 } 249 250 memset(*refcount_block, 0, s->cluster_size); 251 } 252 253 /* Now the new refcount block needs to be written to disk */ 254 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE); 255 qcow2_cache_entry_mark_dirty(s->refcount_block_cache, *refcount_block); 256 ret = qcow2_cache_flush(bs, s->refcount_block_cache); 257 if (ret < 0) { 258 goto fail_block; 259 } 260 261 /* If the refcount table is big enough, just hook the block up there */ 262 if (refcount_table_index < s->refcount_table_size) { 263 uint64_t data64 = cpu_to_be64(new_block); 264 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_HOOKUP); 265 ret = bdrv_pwrite_sync(bs->file, 266 s->refcount_table_offset + refcount_table_index * sizeof(uint64_t), 267 &data64, sizeof(data64)); 268 if (ret < 0) { 269 goto fail_block; 270 } 271 272 s->refcount_table[refcount_table_index] = new_block; 273 return 0; 274 } 275 276 ret = qcow2_cache_put(bs, s->refcount_block_cache, (void**) refcount_block); 277 if (ret < 0) { 278 goto fail_block; 279 } 280 281 /* 282 * If we come here, we need to grow the refcount table. Again, a new 283 * refcount table needs some space and we can't simply allocate to avoid 284 * endless recursion. 285 * 286 * Therefore let's grab new refcount blocks at the end of the image, which 287 * will describe themselves and the new refcount table. This way we can 288 * reference them only in the new table and do the switch to the new 289 * refcount table at once without producing an inconsistent state in 290 * between. 291 */ 292 BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_GROW); 293 294 /* Calculate the number of refcount blocks needed so far */ 295 uint64_t refcount_block_clusters = 1 << (s->cluster_bits - REFCOUNT_SHIFT); 296 uint64_t blocks_used = (s->free_cluster_index + 297 refcount_block_clusters - 1) / refcount_block_clusters; 298 299 /* And now we need at least one block more for the new metadata */ 300 uint64_t table_size = next_refcount_table_size(s, blocks_used + 1); 301 uint64_t last_table_size; 302 uint64_t blocks_clusters; 303 do { 304 uint64_t table_clusters = size_to_clusters(s, table_size); 305 blocks_clusters = 1 + 306 ((table_clusters + refcount_block_clusters - 1) 307 / refcount_block_clusters); 308 uint64_t meta_clusters = table_clusters + blocks_clusters; 309 310 last_table_size = table_size; 311 table_size = next_refcount_table_size(s, blocks_used + 312 ((meta_clusters + refcount_block_clusters - 1) 313 / refcount_block_clusters)); 314 315 } while (last_table_size != table_size); 316 317 #ifdef DEBUG_ALLOC2 318 fprintf(stderr, "qcow2: Grow refcount table %" PRId32 " => %" PRId64 "\n", 319 s->refcount_table_size, table_size); 320 #endif 321 322 /* Create the new refcount table and blocks */ 323 uint64_t meta_offset = (blocks_used * refcount_block_clusters) * 324 s->cluster_size; 325 uint64_t table_offset = meta_offset + blocks_clusters * s->cluster_size; 326 uint16_t *new_blocks = g_malloc0(blocks_clusters * s->cluster_size); 327 uint64_t *new_table = g_malloc0(table_size * sizeof(uint64_t)); 328 329 assert(meta_offset >= (s->free_cluster_index * s->cluster_size)); 330 331 /* Fill the new refcount table */ 332 memcpy(new_table, s->refcount_table, 333 s->refcount_table_size * sizeof(uint64_t)); 334 new_table[refcount_table_index] = new_block; 335 336 int i; 337 for (i = 0; i < blocks_clusters; i++) { 338 new_table[blocks_used + i] = meta_offset + (i * s->cluster_size); 339 } 340 341 /* Fill the refcount blocks */ 342 uint64_t table_clusters = size_to_clusters(s, table_size * sizeof(uint64_t)); 343 int block = 0; 344 for (i = 0; i < table_clusters + blocks_clusters; i++) { 345 new_blocks[block++] = cpu_to_be16(1); 346 } 347 348 /* Write refcount blocks to disk */ 349 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE_BLOCKS); 350 ret = bdrv_pwrite_sync(bs->file, meta_offset, new_blocks, 351 blocks_clusters * s->cluster_size); 352 g_free(new_blocks); 353 if (ret < 0) { 354 goto fail_table; 355 } 356 357 /* Write refcount table to disk */ 358 for(i = 0; i < table_size; i++) { 359 cpu_to_be64s(&new_table[i]); 360 } 361 362 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE_TABLE); 363 ret = bdrv_pwrite_sync(bs->file, table_offset, new_table, 364 table_size * sizeof(uint64_t)); 365 if (ret < 0) { 366 goto fail_table; 367 } 368 369 for(i = 0; i < table_size; i++) { 370 be64_to_cpus(&new_table[i]); 371 } 372 373 /* Hook up the new refcount table in the qcow2 header */ 374 uint8_t data[12]; 375 cpu_to_be64w((uint64_t*)data, table_offset); 376 cpu_to_be32w((uint32_t*)(data + 8), table_clusters); 377 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_SWITCH_TABLE); 378 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, refcount_table_offset), 379 data, sizeof(data)); 380 if (ret < 0) { 381 goto fail_table; 382 } 383 384 /* And switch it in memory */ 385 uint64_t old_table_offset = s->refcount_table_offset; 386 uint64_t old_table_size = s->refcount_table_size; 387 388 g_free(s->refcount_table); 389 s->refcount_table = new_table; 390 s->refcount_table_size = table_size; 391 s->refcount_table_offset = table_offset; 392 393 /* Free old table. Remember, we must not change free_cluster_index */ 394 uint64_t old_free_cluster_index = s->free_cluster_index; 395 qcow2_free_clusters(bs, old_table_offset, old_table_size * sizeof(uint64_t)); 396 s->free_cluster_index = old_free_cluster_index; 397 398 ret = load_refcount_block(bs, new_block, (void**) refcount_block); 399 if (ret < 0) { 400 return ret; 401 } 402 403 return 0; 404 405 fail_table: 406 g_free(new_table); 407 fail_block: 408 if (*refcount_block != NULL) { 409 qcow2_cache_put(bs, s->refcount_block_cache, (void**) refcount_block); 410 } 411 return ret; 412 } 413 414 /* XXX: cache several refcount block clusters ? */ 415 static int QEMU_WARN_UNUSED_RESULT update_refcount(BlockDriverState *bs, 416 int64_t offset, int64_t length, int addend) 417 { 418 BDRVQcowState *s = bs->opaque; 419 int64_t start, last, cluster_offset; 420 uint16_t *refcount_block = NULL; 421 int64_t old_table_index = -1; 422 int ret; 423 424 #ifdef DEBUG_ALLOC2 425 fprintf(stderr, "update_refcount: offset=%" PRId64 " size=%" PRId64 " addend=%d\n", 426 offset, length, addend); 427 #endif 428 if (length < 0) { 429 return -EINVAL; 430 } else if (length == 0) { 431 return 0; 432 } 433 434 if (addend < 0) { 435 qcow2_cache_set_dependency(bs, s->refcount_block_cache, 436 s->l2_table_cache); 437 } 438 439 start = offset & ~(s->cluster_size - 1); 440 last = (offset + length - 1) & ~(s->cluster_size - 1); 441 for(cluster_offset = start; cluster_offset <= last; 442 cluster_offset += s->cluster_size) 443 { 444 int block_index, refcount; 445 int64_t cluster_index = cluster_offset >> s->cluster_bits; 446 int64_t table_index = 447 cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT); 448 449 /* Load the refcount block and allocate it if needed */ 450 if (table_index != old_table_index) { 451 if (refcount_block) { 452 ret = qcow2_cache_put(bs, s->refcount_block_cache, 453 (void**) &refcount_block); 454 if (ret < 0) { 455 goto fail; 456 } 457 } 458 459 ret = alloc_refcount_block(bs, cluster_index, &refcount_block); 460 if (ret < 0) { 461 goto fail; 462 } 463 } 464 old_table_index = table_index; 465 466 qcow2_cache_entry_mark_dirty(s->refcount_block_cache, refcount_block); 467 468 /* we can update the count and save it */ 469 block_index = cluster_index & 470 ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1); 471 472 refcount = be16_to_cpu(refcount_block[block_index]); 473 refcount += addend; 474 if (refcount < 0 || refcount > 0xffff) { 475 ret = -EINVAL; 476 goto fail; 477 } 478 if (refcount == 0 && cluster_index < s->free_cluster_index) { 479 s->free_cluster_index = cluster_index; 480 } 481 refcount_block[block_index] = cpu_to_be16(refcount); 482 } 483 484 ret = 0; 485 fail: 486 /* Write last changed block to disk */ 487 if (refcount_block) { 488 int wret; 489 wret = qcow2_cache_put(bs, s->refcount_block_cache, 490 (void**) &refcount_block); 491 if (wret < 0) { 492 return ret < 0 ? ret : wret; 493 } 494 } 495 496 /* 497 * Try do undo any updates if an error is returned (This may succeed in 498 * some cases like ENOSPC for allocating a new refcount block) 499 */ 500 if (ret < 0) { 501 int dummy; 502 dummy = update_refcount(bs, offset, cluster_offset - offset, -addend); 503 (void)dummy; 504 } 505 506 return ret; 507 } 508 509 /* 510 * Increases or decreases the refcount of a given cluster by one. 511 * addend must be 1 or -1. 512 * 513 * If the return value is non-negative, it is the new refcount of the cluster. 514 * If it is negative, it is -errno and indicates an error. 515 */ 516 static int update_cluster_refcount(BlockDriverState *bs, 517 int64_t cluster_index, 518 int addend) 519 { 520 BDRVQcowState *s = bs->opaque; 521 int ret; 522 523 ret = update_refcount(bs, cluster_index << s->cluster_bits, 1, addend); 524 if (ret < 0) { 525 return ret; 526 } 527 528 bdrv_flush(bs->file); 529 530 return get_refcount(bs, cluster_index); 531 } 532 533 534 535 /*********************************************************/ 536 /* cluster allocation functions */ 537 538 539 540 /* return < 0 if error */ 541 static int64_t alloc_clusters_noref(BlockDriverState *bs, int64_t size) 542 { 543 BDRVQcowState *s = bs->opaque; 544 int i, nb_clusters, refcount; 545 546 nb_clusters = size_to_clusters(s, size); 547 retry: 548 for(i = 0; i < nb_clusters; i++) { 549 int64_t next_cluster_index = s->free_cluster_index++; 550 refcount = get_refcount(bs, next_cluster_index); 551 552 if (refcount < 0) { 553 return refcount; 554 } else if (refcount != 0) { 555 goto retry; 556 } 557 } 558 #ifdef DEBUG_ALLOC2 559 fprintf(stderr, "alloc_clusters: size=%" PRId64 " -> %" PRId64 "\n", 560 size, 561 (s->free_cluster_index - nb_clusters) << s->cluster_bits); 562 #endif 563 return (s->free_cluster_index - nb_clusters) << s->cluster_bits; 564 } 565 566 int64_t qcow2_alloc_clusters(BlockDriverState *bs, int64_t size) 567 { 568 int64_t offset; 569 int ret; 570 571 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC); 572 offset = alloc_clusters_noref(bs, size); 573 if (offset < 0) { 574 return offset; 575 } 576 577 ret = update_refcount(bs, offset, size, 1); 578 if (ret < 0) { 579 return ret; 580 } 581 582 return offset; 583 } 584 585 int qcow2_alloc_clusters_at(BlockDriverState *bs, uint64_t offset, 586 int nb_clusters) 587 { 588 BDRVQcowState *s = bs->opaque; 589 uint64_t cluster_index; 590 uint64_t old_free_cluster_index; 591 int i, refcount, ret; 592 593 /* Check how many clusters there are free */ 594 cluster_index = offset >> s->cluster_bits; 595 for(i = 0; i < nb_clusters; i++) { 596 refcount = get_refcount(bs, cluster_index++); 597 598 if (refcount < 0) { 599 return refcount; 600 } else if (refcount != 0) { 601 break; 602 } 603 } 604 605 /* And then allocate them */ 606 old_free_cluster_index = s->free_cluster_index; 607 s->free_cluster_index = cluster_index + i; 608 609 ret = update_refcount(bs, offset, i << s->cluster_bits, 1); 610 if (ret < 0) { 611 return ret; 612 } 613 614 s->free_cluster_index = old_free_cluster_index; 615 616 return i; 617 } 618 619 /* only used to allocate compressed sectors. We try to allocate 620 contiguous sectors. size must be <= cluster_size */ 621 int64_t qcow2_alloc_bytes(BlockDriverState *bs, int size) 622 { 623 BDRVQcowState *s = bs->opaque; 624 int64_t offset, cluster_offset; 625 int free_in_cluster; 626 627 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC_BYTES); 628 assert(size > 0 && size <= s->cluster_size); 629 if (s->free_byte_offset == 0) { 630 s->free_byte_offset = qcow2_alloc_clusters(bs, s->cluster_size); 631 if (s->free_byte_offset < 0) { 632 return s->free_byte_offset; 633 } 634 } 635 redo: 636 free_in_cluster = s->cluster_size - 637 (s->free_byte_offset & (s->cluster_size - 1)); 638 if (size <= free_in_cluster) { 639 /* enough space in current cluster */ 640 offset = s->free_byte_offset; 641 s->free_byte_offset += size; 642 free_in_cluster -= size; 643 if (free_in_cluster == 0) 644 s->free_byte_offset = 0; 645 if ((offset & (s->cluster_size - 1)) != 0) 646 update_cluster_refcount(bs, offset >> s->cluster_bits, 1); 647 } else { 648 offset = qcow2_alloc_clusters(bs, s->cluster_size); 649 if (offset < 0) { 650 return offset; 651 } 652 cluster_offset = s->free_byte_offset & ~(s->cluster_size - 1); 653 if ((cluster_offset + s->cluster_size) == offset) { 654 /* we are lucky: contiguous data */ 655 offset = s->free_byte_offset; 656 update_cluster_refcount(bs, offset >> s->cluster_bits, 1); 657 s->free_byte_offset += size; 658 } else { 659 s->free_byte_offset = offset; 660 goto redo; 661 } 662 } 663 664 bdrv_flush(bs->file); 665 return offset; 666 } 667 668 void qcow2_free_clusters(BlockDriverState *bs, 669 int64_t offset, int64_t size) 670 { 671 int ret; 672 673 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_FREE); 674 ret = update_refcount(bs, offset, size, -1); 675 if (ret < 0) { 676 fprintf(stderr, "qcow2_free_clusters failed: %s\n", strerror(-ret)); 677 /* TODO Remember the clusters to free them later and avoid leaking */ 678 } 679 } 680 681 /* 682 * Free a cluster using its L2 entry (handles clusters of all types, e.g. 683 * normal cluster, compressed cluster, etc.) 684 */ 685 void qcow2_free_any_clusters(BlockDriverState *bs, 686 uint64_t l2_entry, int nb_clusters) 687 { 688 BDRVQcowState *s = bs->opaque; 689 690 switch (qcow2_get_cluster_type(l2_entry)) { 691 case QCOW2_CLUSTER_COMPRESSED: 692 { 693 int nb_csectors; 694 nb_csectors = ((l2_entry >> s->csize_shift) & 695 s->csize_mask) + 1; 696 qcow2_free_clusters(bs, 697 (l2_entry & s->cluster_offset_mask) & ~511, 698 nb_csectors * 512); 699 } 700 break; 701 case QCOW2_CLUSTER_NORMAL: 702 qcow2_free_clusters(bs, l2_entry & L2E_OFFSET_MASK, 703 nb_clusters << s->cluster_bits); 704 break; 705 case QCOW2_CLUSTER_UNALLOCATED: 706 case QCOW2_CLUSTER_ZERO: 707 break; 708 default: 709 abort(); 710 } 711 } 712 713 714 715 /*********************************************************/ 716 /* snapshots and image creation */ 717 718 719 720 /* update the refcounts of snapshots and the copied flag */ 721 int qcow2_update_snapshot_refcount(BlockDriverState *bs, 722 int64_t l1_table_offset, int l1_size, int addend) 723 { 724 BDRVQcowState *s = bs->opaque; 725 uint64_t *l1_table, *l2_table, l2_offset, offset, l1_size2, l1_allocated; 726 int64_t old_offset, old_l2_offset; 727 int i, j, l1_modified = 0, nb_csectors, refcount; 728 int ret; 729 730 l2_table = NULL; 731 l1_table = NULL; 732 l1_size2 = l1_size * sizeof(uint64_t); 733 734 /* WARNING: qcow2_snapshot_goto relies on this function not using the 735 * l1_table_offset when it is the current s->l1_table_offset! Be careful 736 * when changing this! */ 737 if (l1_table_offset != s->l1_table_offset) { 738 if (l1_size2 != 0) { 739 l1_table = g_malloc0(align_offset(l1_size2, 512)); 740 } else { 741 l1_table = NULL; 742 } 743 l1_allocated = 1; 744 if (bdrv_pread(bs->file, l1_table_offset, 745 l1_table, l1_size2) != l1_size2) 746 { 747 ret = -EIO; 748 goto fail; 749 } 750 751 for(i = 0;i < l1_size; i++) 752 be64_to_cpus(&l1_table[i]); 753 } else { 754 assert(l1_size == s->l1_size); 755 l1_table = s->l1_table; 756 l1_allocated = 0; 757 } 758 759 for(i = 0; i < l1_size; i++) { 760 l2_offset = l1_table[i]; 761 if (l2_offset) { 762 old_l2_offset = l2_offset; 763 l2_offset &= L1E_OFFSET_MASK; 764 765 ret = qcow2_cache_get(bs, s->l2_table_cache, l2_offset, 766 (void**) &l2_table); 767 if (ret < 0) { 768 goto fail; 769 } 770 771 for(j = 0; j < s->l2_size; j++) { 772 offset = be64_to_cpu(l2_table[j]); 773 if (offset != 0) { 774 old_offset = offset; 775 offset &= ~QCOW_OFLAG_COPIED; 776 if (offset & QCOW_OFLAG_COMPRESSED) { 777 nb_csectors = ((offset >> s->csize_shift) & 778 s->csize_mask) + 1; 779 if (addend != 0) { 780 int ret; 781 ret = update_refcount(bs, 782 (offset & s->cluster_offset_mask) & ~511, 783 nb_csectors * 512, addend); 784 if (ret < 0) { 785 goto fail; 786 } 787 788 /* TODO Flushing once for the whole function should 789 * be enough */ 790 bdrv_flush(bs->file); 791 } 792 /* compressed clusters are never modified */ 793 refcount = 2; 794 } else { 795 uint64_t cluster_index = (offset & L2E_OFFSET_MASK) >> s->cluster_bits; 796 if (addend != 0) { 797 refcount = update_cluster_refcount(bs, cluster_index, addend); 798 } else { 799 refcount = get_refcount(bs, cluster_index); 800 } 801 802 if (refcount < 0) { 803 ret = -EIO; 804 goto fail; 805 } 806 } 807 808 if (refcount == 1) { 809 offset |= QCOW_OFLAG_COPIED; 810 } 811 if (offset != old_offset) { 812 if (addend > 0) { 813 qcow2_cache_set_dependency(bs, s->l2_table_cache, 814 s->refcount_block_cache); 815 } 816 l2_table[j] = cpu_to_be64(offset); 817 qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table); 818 } 819 } 820 } 821 822 ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table); 823 if (ret < 0) { 824 goto fail; 825 } 826 827 828 if (addend != 0) { 829 refcount = update_cluster_refcount(bs, l2_offset >> s->cluster_bits, addend); 830 } else { 831 refcount = get_refcount(bs, l2_offset >> s->cluster_bits); 832 } 833 if (refcount < 0) { 834 ret = -EIO; 835 goto fail; 836 } else if (refcount == 1) { 837 l2_offset |= QCOW_OFLAG_COPIED; 838 } 839 if (l2_offset != old_l2_offset) { 840 l1_table[i] = l2_offset; 841 l1_modified = 1; 842 } 843 } 844 } 845 846 ret = 0; 847 fail: 848 if (l2_table) { 849 qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table); 850 } 851 852 /* Update L1 only if it isn't deleted anyway (addend = -1) */ 853 if (addend >= 0 && l1_modified) { 854 for(i = 0; i < l1_size; i++) 855 cpu_to_be64s(&l1_table[i]); 856 if (bdrv_pwrite_sync(bs->file, l1_table_offset, l1_table, 857 l1_size2) < 0) 858 goto fail; 859 for(i = 0; i < l1_size; i++) 860 be64_to_cpus(&l1_table[i]); 861 } 862 if (l1_allocated) 863 g_free(l1_table); 864 return ret; 865 } 866 867 868 869 870 /*********************************************************/ 871 /* refcount checking functions */ 872 873 874 875 /* 876 * Increases the refcount for a range of clusters in a given refcount table. 877 * This is used to construct a temporary refcount table out of L1 and L2 tables 878 * which can be compared the the refcount table saved in the image. 879 * 880 * Modifies the number of errors in res. 881 */ 882 static void inc_refcounts(BlockDriverState *bs, 883 BdrvCheckResult *res, 884 uint16_t *refcount_table, 885 int refcount_table_size, 886 int64_t offset, int64_t size) 887 { 888 BDRVQcowState *s = bs->opaque; 889 int64_t start, last, cluster_offset; 890 int k; 891 892 if (size <= 0) 893 return; 894 895 start = offset & ~(s->cluster_size - 1); 896 last = (offset + size - 1) & ~(s->cluster_size - 1); 897 for(cluster_offset = start; cluster_offset <= last; 898 cluster_offset += s->cluster_size) { 899 k = cluster_offset >> s->cluster_bits; 900 if (k < 0) { 901 fprintf(stderr, "ERROR: invalid cluster offset=0x%" PRIx64 "\n", 902 cluster_offset); 903 res->corruptions++; 904 } else if (k >= refcount_table_size) { 905 fprintf(stderr, "Warning: cluster offset=0x%" PRIx64 " is after " 906 "the end of the image file, can't properly check refcounts.\n", 907 cluster_offset); 908 res->check_errors++; 909 } else { 910 if (++refcount_table[k] == 0) { 911 fprintf(stderr, "ERROR: overflow cluster offset=0x%" PRIx64 912 "\n", cluster_offset); 913 res->corruptions++; 914 } 915 } 916 } 917 } 918 919 /* 920 * Increases the refcount in the given refcount table for the all clusters 921 * referenced in the L2 table. While doing so, performs some checks on L2 922 * entries. 923 * 924 * Returns the number of errors found by the checks or -errno if an internal 925 * error occurred. 926 */ 927 static int check_refcounts_l2(BlockDriverState *bs, BdrvCheckResult *res, 928 uint16_t *refcount_table, int refcount_table_size, int64_t l2_offset, 929 int check_copied) 930 { 931 BDRVQcowState *s = bs->opaque; 932 uint64_t *l2_table, l2_entry; 933 int i, l2_size, nb_csectors, refcount; 934 935 /* Read L2 table from disk */ 936 l2_size = s->l2_size * sizeof(uint64_t); 937 l2_table = g_malloc(l2_size); 938 939 if (bdrv_pread(bs->file, l2_offset, l2_table, l2_size) != l2_size) 940 goto fail; 941 942 /* Do the actual checks */ 943 for(i = 0; i < s->l2_size; i++) { 944 l2_entry = be64_to_cpu(l2_table[i]); 945 946 switch (qcow2_get_cluster_type(l2_entry)) { 947 case QCOW2_CLUSTER_COMPRESSED: 948 /* Compressed clusters don't have QCOW_OFLAG_COPIED */ 949 if (l2_entry & QCOW_OFLAG_COPIED) { 950 fprintf(stderr, "ERROR: cluster %" PRId64 ": " 951 "copied flag must never be set for compressed " 952 "clusters\n", l2_entry >> s->cluster_bits); 953 l2_entry &= ~QCOW_OFLAG_COPIED; 954 res->corruptions++; 955 } 956 957 /* Mark cluster as used */ 958 nb_csectors = ((l2_entry >> s->csize_shift) & 959 s->csize_mask) + 1; 960 l2_entry &= s->cluster_offset_mask; 961 inc_refcounts(bs, res, refcount_table, refcount_table_size, 962 l2_entry & ~511, nb_csectors * 512); 963 break; 964 965 case QCOW2_CLUSTER_ZERO: 966 if ((l2_entry & L2E_OFFSET_MASK) == 0) { 967 break; 968 } 969 /* fall through */ 970 971 case QCOW2_CLUSTER_NORMAL: 972 { 973 /* QCOW_OFLAG_COPIED must be set iff refcount == 1 */ 974 uint64_t offset = l2_entry & L2E_OFFSET_MASK; 975 976 if (check_copied) { 977 refcount = get_refcount(bs, offset >> s->cluster_bits); 978 if (refcount < 0) { 979 fprintf(stderr, "Can't get refcount for offset %" 980 PRIx64 ": %s\n", l2_entry, strerror(-refcount)); 981 goto fail; 982 } 983 if ((refcount == 1) != ((l2_entry & QCOW_OFLAG_COPIED) != 0)) { 984 fprintf(stderr, "ERROR OFLAG_COPIED: offset=%" 985 PRIx64 " refcount=%d\n", l2_entry, refcount); 986 res->corruptions++; 987 } 988 } 989 990 /* Mark cluster as used */ 991 inc_refcounts(bs, res, refcount_table,refcount_table_size, 992 offset, s->cluster_size); 993 994 /* Correct offsets are cluster aligned */ 995 if (offset & (s->cluster_size - 1)) { 996 fprintf(stderr, "ERROR offset=%" PRIx64 ": Cluster is not " 997 "properly aligned; L2 entry corrupted.\n", offset); 998 res->corruptions++; 999 } 1000 break; 1001 } 1002 1003 case QCOW2_CLUSTER_UNALLOCATED: 1004 break; 1005 1006 default: 1007 abort(); 1008 } 1009 } 1010 1011 g_free(l2_table); 1012 return 0; 1013 1014 fail: 1015 fprintf(stderr, "ERROR: I/O error in check_refcounts_l2\n"); 1016 g_free(l2_table); 1017 return -EIO; 1018 } 1019 1020 /* 1021 * Increases the refcount for the L1 table, its L2 tables and all referenced 1022 * clusters in the given refcount table. While doing so, performs some checks 1023 * on L1 and L2 entries. 1024 * 1025 * Returns the number of errors found by the checks or -errno if an internal 1026 * error occurred. 1027 */ 1028 static int check_refcounts_l1(BlockDriverState *bs, 1029 BdrvCheckResult *res, 1030 uint16_t *refcount_table, 1031 int refcount_table_size, 1032 int64_t l1_table_offset, int l1_size, 1033 int check_copied) 1034 { 1035 BDRVQcowState *s = bs->opaque; 1036 uint64_t *l1_table, l2_offset, l1_size2; 1037 int i, refcount, ret; 1038 1039 l1_size2 = l1_size * sizeof(uint64_t); 1040 1041 /* Mark L1 table as used */ 1042 inc_refcounts(bs, res, refcount_table, refcount_table_size, 1043 l1_table_offset, l1_size2); 1044 1045 /* Read L1 table entries from disk */ 1046 if (l1_size2 == 0) { 1047 l1_table = NULL; 1048 } else { 1049 l1_table = g_malloc(l1_size2); 1050 if (bdrv_pread(bs->file, l1_table_offset, 1051 l1_table, l1_size2) != l1_size2) 1052 goto fail; 1053 for(i = 0;i < l1_size; i++) 1054 be64_to_cpus(&l1_table[i]); 1055 } 1056 1057 /* Do the actual checks */ 1058 for(i = 0; i < l1_size; i++) { 1059 l2_offset = l1_table[i]; 1060 if (l2_offset) { 1061 /* QCOW_OFLAG_COPIED must be set iff refcount == 1 */ 1062 if (check_copied) { 1063 refcount = get_refcount(bs, (l2_offset & ~QCOW_OFLAG_COPIED) 1064 >> s->cluster_bits); 1065 if (refcount < 0) { 1066 fprintf(stderr, "Can't get refcount for l2_offset %" 1067 PRIx64 ": %s\n", l2_offset, strerror(-refcount)); 1068 goto fail; 1069 } 1070 if ((refcount == 1) != ((l2_offset & QCOW_OFLAG_COPIED) != 0)) { 1071 fprintf(stderr, "ERROR OFLAG_COPIED: l2_offset=%" PRIx64 1072 " refcount=%d\n", l2_offset, refcount); 1073 res->corruptions++; 1074 } 1075 } 1076 1077 /* Mark L2 table as used */ 1078 l2_offset &= L1E_OFFSET_MASK; 1079 inc_refcounts(bs, res, refcount_table, refcount_table_size, 1080 l2_offset, s->cluster_size); 1081 1082 /* L2 tables are cluster aligned */ 1083 if (l2_offset & (s->cluster_size - 1)) { 1084 fprintf(stderr, "ERROR l2_offset=%" PRIx64 ": Table is not " 1085 "cluster aligned; L1 entry corrupted\n", l2_offset); 1086 res->corruptions++; 1087 } 1088 1089 /* Process and check L2 entries */ 1090 ret = check_refcounts_l2(bs, res, refcount_table, 1091 refcount_table_size, l2_offset, check_copied); 1092 if (ret < 0) { 1093 goto fail; 1094 } 1095 } 1096 } 1097 g_free(l1_table); 1098 return 0; 1099 1100 fail: 1101 fprintf(stderr, "ERROR: I/O error in check_refcounts_l1\n"); 1102 res->check_errors++; 1103 g_free(l1_table); 1104 return -EIO; 1105 } 1106 1107 /* 1108 * Checks an image for refcount consistency. 1109 * 1110 * Returns 0 if no errors are found, the number of errors in case the image is 1111 * detected as corrupted, and -errno when an internal error occurred. 1112 */ 1113 int qcow2_check_refcounts(BlockDriverState *bs, BdrvCheckResult *res, 1114 BdrvCheckMode fix) 1115 { 1116 BDRVQcowState *s = bs->opaque; 1117 int64_t size, i; 1118 int nb_clusters, refcount1, refcount2; 1119 QCowSnapshot *sn; 1120 uint16_t *refcount_table; 1121 int ret; 1122 1123 size = bdrv_getlength(bs->file); 1124 nb_clusters = size_to_clusters(s, size); 1125 refcount_table = g_malloc0(nb_clusters * sizeof(uint16_t)); 1126 1127 /* header */ 1128 inc_refcounts(bs, res, refcount_table, nb_clusters, 1129 0, s->cluster_size); 1130 1131 /* current L1 table */ 1132 ret = check_refcounts_l1(bs, res, refcount_table, nb_clusters, 1133 s->l1_table_offset, s->l1_size, 1); 1134 if (ret < 0) { 1135 goto fail; 1136 } 1137 1138 /* snapshots */ 1139 for(i = 0; i < s->nb_snapshots; i++) { 1140 sn = s->snapshots + i; 1141 ret = check_refcounts_l1(bs, res, refcount_table, nb_clusters, 1142 sn->l1_table_offset, sn->l1_size, 0); 1143 if (ret < 0) { 1144 goto fail; 1145 } 1146 } 1147 inc_refcounts(bs, res, refcount_table, nb_clusters, 1148 s->snapshots_offset, s->snapshots_size); 1149 1150 /* refcount data */ 1151 inc_refcounts(bs, res, refcount_table, nb_clusters, 1152 s->refcount_table_offset, 1153 s->refcount_table_size * sizeof(uint64_t)); 1154 1155 for(i = 0; i < s->refcount_table_size; i++) { 1156 uint64_t offset, cluster; 1157 offset = s->refcount_table[i]; 1158 cluster = offset >> s->cluster_bits; 1159 1160 /* Refcount blocks are cluster aligned */ 1161 if (offset & (s->cluster_size - 1)) { 1162 fprintf(stderr, "ERROR refcount block %" PRId64 " is not " 1163 "cluster aligned; refcount table entry corrupted\n", i); 1164 res->corruptions++; 1165 continue; 1166 } 1167 1168 if (cluster >= nb_clusters) { 1169 fprintf(stderr, "ERROR refcount block %" PRId64 1170 " is outside image\n", i); 1171 res->corruptions++; 1172 continue; 1173 } 1174 1175 if (offset != 0) { 1176 inc_refcounts(bs, res, refcount_table, nb_clusters, 1177 offset, s->cluster_size); 1178 if (refcount_table[cluster] != 1) { 1179 fprintf(stderr, "ERROR refcount block %" PRId64 1180 " refcount=%d\n", 1181 i, refcount_table[cluster]); 1182 res->corruptions++; 1183 } 1184 } 1185 } 1186 1187 /* compare ref counts */ 1188 for(i = 0; i < nb_clusters; i++) { 1189 refcount1 = get_refcount(bs, i); 1190 if (refcount1 < 0) { 1191 fprintf(stderr, "Can't get refcount for cluster %" PRId64 ": %s\n", 1192 i, strerror(-refcount1)); 1193 res->check_errors++; 1194 continue; 1195 } 1196 1197 refcount2 = refcount_table[i]; 1198 if (refcount1 != refcount2) { 1199 1200 /* Check if we're allowed to fix the mismatch */ 1201 int *num_fixed = NULL; 1202 if (refcount1 > refcount2 && (fix & BDRV_FIX_LEAKS)) { 1203 num_fixed = &res->leaks_fixed; 1204 } else if (refcount1 < refcount2 && (fix & BDRV_FIX_ERRORS)) { 1205 num_fixed = &res->corruptions_fixed; 1206 } 1207 1208 fprintf(stderr, "%s cluster %" PRId64 " refcount=%d reference=%d\n", 1209 num_fixed != NULL ? "Repairing" : 1210 refcount1 < refcount2 ? "ERROR" : 1211 "Leaked", 1212 i, refcount1, refcount2); 1213 1214 if (num_fixed) { 1215 ret = update_refcount(bs, i << s->cluster_bits, 1, 1216 refcount2 - refcount1); 1217 if (ret >= 0) { 1218 (*num_fixed)++; 1219 continue; 1220 } 1221 } 1222 1223 /* And if we couldn't, print an error */ 1224 if (refcount1 < refcount2) { 1225 res->corruptions++; 1226 } else { 1227 res->leaks++; 1228 } 1229 } 1230 } 1231 1232 ret = 0; 1233 1234 fail: 1235 g_free(refcount_table); 1236 1237 return ret; 1238 } 1239 1240