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/osdep.h" 26 #include "qapi/error.h" 27 #include "qemu-common.h" 28 #include "block/block_int.h" 29 #include "qcow2.h" 30 #include "qemu/range.h" 31 #include "qemu/bswap.h" 32 #include "qemu/cutils.h" 33 34 static int64_t alloc_clusters_noref(BlockDriverState *bs, uint64_t size, 35 uint64_t max); 36 static int QEMU_WARN_UNUSED_RESULT update_refcount(BlockDriverState *bs, 37 int64_t offset, int64_t length, uint64_t addend, 38 bool decrease, enum qcow2_discard_type type); 39 40 static uint64_t get_refcount_ro0(const void *refcount_array, uint64_t index); 41 static uint64_t get_refcount_ro1(const void *refcount_array, uint64_t index); 42 static uint64_t get_refcount_ro2(const void *refcount_array, uint64_t index); 43 static uint64_t get_refcount_ro3(const void *refcount_array, uint64_t index); 44 static uint64_t get_refcount_ro4(const void *refcount_array, uint64_t index); 45 static uint64_t get_refcount_ro5(const void *refcount_array, uint64_t index); 46 static uint64_t get_refcount_ro6(const void *refcount_array, uint64_t index); 47 48 static void set_refcount_ro0(void *refcount_array, uint64_t index, 49 uint64_t value); 50 static void set_refcount_ro1(void *refcount_array, uint64_t index, 51 uint64_t value); 52 static void set_refcount_ro2(void *refcount_array, uint64_t index, 53 uint64_t value); 54 static void set_refcount_ro3(void *refcount_array, uint64_t index, 55 uint64_t value); 56 static void set_refcount_ro4(void *refcount_array, uint64_t index, 57 uint64_t value); 58 static void set_refcount_ro5(void *refcount_array, uint64_t index, 59 uint64_t value); 60 static void set_refcount_ro6(void *refcount_array, uint64_t index, 61 uint64_t value); 62 63 64 static Qcow2GetRefcountFunc *const get_refcount_funcs[] = { 65 &get_refcount_ro0, 66 &get_refcount_ro1, 67 &get_refcount_ro2, 68 &get_refcount_ro3, 69 &get_refcount_ro4, 70 &get_refcount_ro5, 71 &get_refcount_ro6 72 }; 73 74 static Qcow2SetRefcountFunc *const set_refcount_funcs[] = { 75 &set_refcount_ro0, 76 &set_refcount_ro1, 77 &set_refcount_ro2, 78 &set_refcount_ro3, 79 &set_refcount_ro4, 80 &set_refcount_ro5, 81 &set_refcount_ro6 82 }; 83 84 85 /*********************************************************/ 86 /* refcount handling */ 87 88 static void update_max_refcount_table_index(BDRVQcow2State *s) 89 { 90 unsigned i = s->refcount_table_size - 1; 91 while (i > 0 && (s->refcount_table[i] & REFT_OFFSET_MASK) == 0) { 92 i--; 93 } 94 /* Set s->max_refcount_table_index to the index of the last used entry */ 95 s->max_refcount_table_index = i; 96 } 97 98 int qcow2_refcount_init(BlockDriverState *bs) 99 { 100 BDRVQcow2State *s = bs->opaque; 101 unsigned int refcount_table_size2, i; 102 int ret; 103 104 assert(s->refcount_order >= 0 && s->refcount_order <= 6); 105 106 s->get_refcount = get_refcount_funcs[s->refcount_order]; 107 s->set_refcount = set_refcount_funcs[s->refcount_order]; 108 109 assert(s->refcount_table_size <= INT_MAX / sizeof(uint64_t)); 110 refcount_table_size2 = s->refcount_table_size * sizeof(uint64_t); 111 s->refcount_table = g_try_malloc(refcount_table_size2); 112 113 if (s->refcount_table_size > 0) { 114 if (s->refcount_table == NULL) { 115 ret = -ENOMEM; 116 goto fail; 117 } 118 BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_LOAD); 119 ret = bdrv_pread(bs->file, s->refcount_table_offset, 120 s->refcount_table, refcount_table_size2); 121 if (ret < 0) { 122 goto fail; 123 } 124 for(i = 0; i < s->refcount_table_size; i++) 125 be64_to_cpus(&s->refcount_table[i]); 126 update_max_refcount_table_index(s); 127 } 128 return 0; 129 fail: 130 return ret; 131 } 132 133 void qcow2_refcount_close(BlockDriverState *bs) 134 { 135 BDRVQcow2State *s = bs->opaque; 136 g_free(s->refcount_table); 137 } 138 139 140 static uint64_t get_refcount_ro0(const void *refcount_array, uint64_t index) 141 { 142 return (((const uint8_t *)refcount_array)[index / 8] >> (index % 8)) & 0x1; 143 } 144 145 static void set_refcount_ro0(void *refcount_array, uint64_t index, 146 uint64_t value) 147 { 148 assert(!(value >> 1)); 149 ((uint8_t *)refcount_array)[index / 8] &= ~(0x1 << (index % 8)); 150 ((uint8_t *)refcount_array)[index / 8] |= value << (index % 8); 151 } 152 153 static uint64_t get_refcount_ro1(const void *refcount_array, uint64_t index) 154 { 155 return (((const uint8_t *)refcount_array)[index / 4] >> (2 * (index % 4))) 156 & 0x3; 157 } 158 159 static void set_refcount_ro1(void *refcount_array, uint64_t index, 160 uint64_t value) 161 { 162 assert(!(value >> 2)); 163 ((uint8_t *)refcount_array)[index / 4] &= ~(0x3 << (2 * (index % 4))); 164 ((uint8_t *)refcount_array)[index / 4] |= value << (2 * (index % 4)); 165 } 166 167 static uint64_t get_refcount_ro2(const void *refcount_array, uint64_t index) 168 { 169 return (((const uint8_t *)refcount_array)[index / 2] >> (4 * (index % 2))) 170 & 0xf; 171 } 172 173 static void set_refcount_ro2(void *refcount_array, uint64_t index, 174 uint64_t value) 175 { 176 assert(!(value >> 4)); 177 ((uint8_t *)refcount_array)[index / 2] &= ~(0xf << (4 * (index % 2))); 178 ((uint8_t *)refcount_array)[index / 2] |= value << (4 * (index % 2)); 179 } 180 181 static uint64_t get_refcount_ro3(const void *refcount_array, uint64_t index) 182 { 183 return ((const uint8_t *)refcount_array)[index]; 184 } 185 186 static void set_refcount_ro3(void *refcount_array, uint64_t index, 187 uint64_t value) 188 { 189 assert(!(value >> 8)); 190 ((uint8_t *)refcount_array)[index] = value; 191 } 192 193 static uint64_t get_refcount_ro4(const void *refcount_array, uint64_t index) 194 { 195 return be16_to_cpu(((const uint16_t *)refcount_array)[index]); 196 } 197 198 static void set_refcount_ro4(void *refcount_array, uint64_t index, 199 uint64_t value) 200 { 201 assert(!(value >> 16)); 202 ((uint16_t *)refcount_array)[index] = cpu_to_be16(value); 203 } 204 205 static uint64_t get_refcount_ro5(const void *refcount_array, uint64_t index) 206 { 207 return be32_to_cpu(((const uint32_t *)refcount_array)[index]); 208 } 209 210 static void set_refcount_ro5(void *refcount_array, uint64_t index, 211 uint64_t value) 212 { 213 assert(!(value >> 32)); 214 ((uint32_t *)refcount_array)[index] = cpu_to_be32(value); 215 } 216 217 static uint64_t get_refcount_ro6(const void *refcount_array, uint64_t index) 218 { 219 return be64_to_cpu(((const uint64_t *)refcount_array)[index]); 220 } 221 222 static void set_refcount_ro6(void *refcount_array, uint64_t index, 223 uint64_t value) 224 { 225 ((uint64_t *)refcount_array)[index] = cpu_to_be64(value); 226 } 227 228 229 static int load_refcount_block(BlockDriverState *bs, 230 int64_t refcount_block_offset, 231 void **refcount_block) 232 { 233 BDRVQcow2State *s = bs->opaque; 234 235 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_LOAD); 236 return qcow2_cache_get(bs, s->refcount_block_cache, refcount_block_offset, 237 refcount_block); 238 } 239 240 /* 241 * Retrieves the refcount of the cluster given by its index and stores it in 242 * *refcount. Returns 0 on success and -errno on failure. 243 */ 244 int qcow2_get_refcount(BlockDriverState *bs, int64_t cluster_index, 245 uint64_t *refcount) 246 { 247 BDRVQcow2State *s = bs->opaque; 248 uint64_t refcount_table_index, block_index; 249 int64_t refcount_block_offset; 250 int ret; 251 void *refcount_block; 252 253 refcount_table_index = cluster_index >> s->refcount_block_bits; 254 if (refcount_table_index >= s->refcount_table_size) { 255 *refcount = 0; 256 return 0; 257 } 258 refcount_block_offset = 259 s->refcount_table[refcount_table_index] & REFT_OFFSET_MASK; 260 if (!refcount_block_offset) { 261 *refcount = 0; 262 return 0; 263 } 264 265 if (offset_into_cluster(s, refcount_block_offset)) { 266 qcow2_signal_corruption(bs, true, -1, -1, "Refblock offset %#" PRIx64 267 " unaligned (reftable index: %#" PRIx64 ")", 268 refcount_block_offset, refcount_table_index); 269 return -EIO; 270 } 271 272 ret = qcow2_cache_get(bs, s->refcount_block_cache, refcount_block_offset, 273 &refcount_block); 274 if (ret < 0) { 275 return ret; 276 } 277 278 block_index = cluster_index & (s->refcount_block_size - 1); 279 *refcount = s->get_refcount(refcount_block, block_index); 280 281 qcow2_cache_put(s->refcount_block_cache, &refcount_block); 282 283 return 0; 284 } 285 286 /* Checks if two offsets are described by the same refcount block */ 287 static int in_same_refcount_block(BDRVQcow2State *s, uint64_t offset_a, 288 uint64_t offset_b) 289 { 290 uint64_t block_a = offset_a >> (s->cluster_bits + s->refcount_block_bits); 291 uint64_t block_b = offset_b >> (s->cluster_bits + s->refcount_block_bits); 292 293 return (block_a == block_b); 294 } 295 296 /* 297 * Loads a refcount block. If it doesn't exist yet, it is allocated first 298 * (including growing the refcount table if needed). 299 * 300 * Returns 0 on success or -errno in error case 301 */ 302 static int alloc_refcount_block(BlockDriverState *bs, 303 int64_t cluster_index, void **refcount_block) 304 { 305 BDRVQcow2State *s = bs->opaque; 306 unsigned int refcount_table_index; 307 int64_t ret; 308 309 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC); 310 311 /* Find the refcount block for the given cluster */ 312 refcount_table_index = cluster_index >> s->refcount_block_bits; 313 314 if (refcount_table_index < s->refcount_table_size) { 315 316 uint64_t refcount_block_offset = 317 s->refcount_table[refcount_table_index] & REFT_OFFSET_MASK; 318 319 /* If it's already there, we're done */ 320 if (refcount_block_offset) { 321 if (offset_into_cluster(s, refcount_block_offset)) { 322 qcow2_signal_corruption(bs, true, -1, -1, "Refblock offset %#" 323 PRIx64 " unaligned (reftable index: " 324 "%#x)", refcount_block_offset, 325 refcount_table_index); 326 return -EIO; 327 } 328 329 return load_refcount_block(bs, refcount_block_offset, 330 refcount_block); 331 } 332 } 333 334 /* 335 * If we came here, we need to allocate something. Something is at least 336 * a cluster for the new refcount block. It may also include a new refcount 337 * table if the old refcount table is too small. 338 * 339 * Note that allocating clusters here needs some special care: 340 * 341 * - We can't use the normal qcow2_alloc_clusters(), it would try to 342 * increase the refcount and very likely we would end up with an endless 343 * recursion. Instead we must place the refcount blocks in a way that 344 * they can describe them themselves. 345 * 346 * - We need to consider that at this point we are inside update_refcounts 347 * and potentially doing an initial refcount increase. This means that 348 * some clusters have already been allocated by the caller, but their 349 * refcount isn't accurate yet. If we allocate clusters for metadata, we 350 * need to return -EAGAIN to signal the caller that it needs to restart 351 * the search for free clusters. 352 * 353 * - alloc_clusters_noref and qcow2_free_clusters may load a different 354 * refcount block into the cache 355 */ 356 357 *refcount_block = NULL; 358 359 /* We write to the refcount table, so we might depend on L2 tables */ 360 ret = qcow2_cache_flush(bs, s->l2_table_cache); 361 if (ret < 0) { 362 return ret; 363 } 364 365 /* Allocate the refcount block itself and mark it as used */ 366 int64_t new_block = alloc_clusters_noref(bs, s->cluster_size, INT64_MAX); 367 if (new_block < 0) { 368 return new_block; 369 } 370 371 /* The offset must fit in the offset field of the refcount table entry */ 372 assert((new_block & REFT_OFFSET_MASK) == new_block); 373 374 /* If we're allocating the block at offset 0 then something is wrong */ 375 if (new_block == 0) { 376 qcow2_signal_corruption(bs, true, -1, -1, "Preventing invalid " 377 "allocation of refcount block at offset 0"); 378 return -EIO; 379 } 380 381 #ifdef DEBUG_ALLOC2 382 fprintf(stderr, "qcow2: Allocate refcount block %d for %" PRIx64 383 " at %" PRIx64 "\n", 384 refcount_table_index, cluster_index << s->cluster_bits, new_block); 385 #endif 386 387 if (in_same_refcount_block(s, new_block, cluster_index << s->cluster_bits)) { 388 /* Zero the new refcount block before updating it */ 389 ret = qcow2_cache_get_empty(bs, s->refcount_block_cache, new_block, 390 refcount_block); 391 if (ret < 0) { 392 goto fail; 393 } 394 395 memset(*refcount_block, 0, s->cluster_size); 396 397 /* The block describes itself, need to update the cache */ 398 int block_index = (new_block >> s->cluster_bits) & 399 (s->refcount_block_size - 1); 400 s->set_refcount(*refcount_block, block_index, 1); 401 } else { 402 /* Described somewhere else. This can recurse at most twice before we 403 * arrive at a block that describes itself. */ 404 ret = update_refcount(bs, new_block, s->cluster_size, 1, false, 405 QCOW2_DISCARD_NEVER); 406 if (ret < 0) { 407 goto fail; 408 } 409 410 ret = qcow2_cache_flush(bs, s->refcount_block_cache); 411 if (ret < 0) { 412 goto fail; 413 } 414 415 /* Initialize the new refcount block only after updating its refcount, 416 * update_refcount uses the refcount cache itself */ 417 ret = qcow2_cache_get_empty(bs, s->refcount_block_cache, new_block, 418 refcount_block); 419 if (ret < 0) { 420 goto fail; 421 } 422 423 memset(*refcount_block, 0, s->cluster_size); 424 } 425 426 /* Now the new refcount block needs to be written to disk */ 427 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE); 428 qcow2_cache_entry_mark_dirty(s->refcount_block_cache, *refcount_block); 429 ret = qcow2_cache_flush(bs, s->refcount_block_cache); 430 if (ret < 0) { 431 goto fail; 432 } 433 434 /* If the refcount table is big enough, just hook the block up there */ 435 if (refcount_table_index < s->refcount_table_size) { 436 uint64_t data64 = cpu_to_be64(new_block); 437 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_HOOKUP); 438 ret = bdrv_pwrite_sync(bs->file, 439 s->refcount_table_offset + refcount_table_index * sizeof(uint64_t), 440 &data64, sizeof(data64)); 441 if (ret < 0) { 442 goto fail; 443 } 444 445 s->refcount_table[refcount_table_index] = new_block; 446 /* If there's a hole in s->refcount_table then it can happen 447 * that refcount_table_index < s->max_refcount_table_index */ 448 s->max_refcount_table_index = 449 MAX(s->max_refcount_table_index, refcount_table_index); 450 451 /* The new refcount block may be where the caller intended to put its 452 * data, so let it restart the search. */ 453 return -EAGAIN; 454 } 455 456 qcow2_cache_put(s->refcount_block_cache, refcount_block); 457 458 /* 459 * If we come here, we need to grow the refcount table. Again, a new 460 * refcount table needs some space and we can't simply allocate to avoid 461 * endless recursion. 462 * 463 * Therefore let's grab new refcount blocks at the end of the image, which 464 * will describe themselves and the new refcount table. This way we can 465 * reference them only in the new table and do the switch to the new 466 * refcount table at once without producing an inconsistent state in 467 * between. 468 */ 469 BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_GROW); 470 471 /* Calculate the number of refcount blocks needed so far; this will be the 472 * basis for calculating the index of the first cluster used for the 473 * self-describing refcount structures which we are about to create. 474 * 475 * Because we reached this point, there cannot be any refcount entries for 476 * cluster_index or higher indices yet. However, because new_block has been 477 * allocated to describe that cluster (and it will assume this role later 478 * on), we cannot use that index; also, new_block may actually have a higher 479 * cluster index than cluster_index, so it needs to be taken into account 480 * here (and 1 needs to be added to its value because that cluster is used). 481 */ 482 uint64_t blocks_used = DIV_ROUND_UP(MAX(cluster_index + 1, 483 (new_block >> s->cluster_bits) + 1), 484 s->refcount_block_size); 485 486 /* Create the new refcount table and blocks */ 487 uint64_t meta_offset = (blocks_used * s->refcount_block_size) * 488 s->cluster_size; 489 490 ret = qcow2_refcount_area(bs, meta_offset, 0, false, 491 refcount_table_index, new_block); 492 if (ret < 0) { 493 return ret; 494 } 495 496 ret = load_refcount_block(bs, new_block, refcount_block); 497 if (ret < 0) { 498 return ret; 499 } 500 501 /* If we were trying to do the initial refcount update for some cluster 502 * allocation, we might have used the same clusters to store newly 503 * allocated metadata. Make the caller search some new space. */ 504 return -EAGAIN; 505 506 fail: 507 if (*refcount_block != NULL) { 508 qcow2_cache_put(s->refcount_block_cache, refcount_block); 509 } 510 return ret; 511 } 512 513 /* 514 * Starting at @start_offset, this function creates new self-covering refcount 515 * structures: A new refcount table and refcount blocks which cover all of 516 * themselves, and a number of @additional_clusters beyond their end. 517 * @start_offset must be at the end of the image file, that is, there must be 518 * only empty space beyond it. 519 * If @exact_size is false, the refcount table will have 50 % more entries than 520 * necessary so it will not need to grow again soon. 521 * If @new_refblock_offset is not zero, it contains the offset of a refcount 522 * block that should be entered into the new refcount table at index 523 * @new_refblock_index. 524 * 525 * Returns: The offset after the new refcount structures (i.e. where the 526 * @additional_clusters may be placed) on success, -errno on error. 527 */ 528 int64_t qcow2_refcount_area(BlockDriverState *bs, uint64_t start_offset, 529 uint64_t additional_clusters, bool exact_size, 530 int new_refblock_index, 531 uint64_t new_refblock_offset) 532 { 533 BDRVQcow2State *s = bs->opaque; 534 uint64_t total_refblock_count_u64, additional_refblock_count; 535 int total_refblock_count, table_size, area_reftable_index, table_clusters; 536 int i; 537 uint64_t table_offset, block_offset, end_offset; 538 int ret; 539 uint64_t *new_table; 540 541 assert(!(start_offset % s->cluster_size)); 542 543 qcow2_refcount_metadata_size(start_offset / s->cluster_size + 544 additional_clusters, 545 s->cluster_size, s->refcount_order, 546 !exact_size, &total_refblock_count_u64); 547 if (total_refblock_count_u64 > QCOW_MAX_REFTABLE_SIZE) { 548 return -EFBIG; 549 } 550 total_refblock_count = total_refblock_count_u64; 551 552 /* Index in the refcount table of the first refcount block to cover the area 553 * of refcount structures we are about to create; we know that 554 * @total_refblock_count can cover @start_offset, so this will definitely 555 * fit into an int. */ 556 area_reftable_index = (start_offset / s->cluster_size) / 557 s->refcount_block_size; 558 559 if (exact_size) { 560 table_size = total_refblock_count; 561 } else { 562 table_size = total_refblock_count + 563 DIV_ROUND_UP(total_refblock_count, 2); 564 } 565 /* The qcow2 file can only store the reftable size in number of clusters */ 566 table_size = ROUND_UP(table_size, s->cluster_size / sizeof(uint64_t)); 567 table_clusters = (table_size * sizeof(uint64_t)) / s->cluster_size; 568 569 if (table_size > QCOW_MAX_REFTABLE_SIZE) { 570 return -EFBIG; 571 } 572 573 new_table = g_try_new0(uint64_t, table_size); 574 575 assert(table_size > 0); 576 if (new_table == NULL) { 577 ret = -ENOMEM; 578 goto fail; 579 } 580 581 /* Fill the new refcount table */ 582 if (table_size > s->max_refcount_table_index) { 583 /* We're actually growing the reftable */ 584 memcpy(new_table, s->refcount_table, 585 (s->max_refcount_table_index + 1) * sizeof(uint64_t)); 586 } else { 587 /* Improbable case: We're shrinking the reftable. However, the caller 588 * has assured us that there is only empty space beyond @start_offset, 589 * so we can simply drop all of the refblocks that won't fit into the 590 * new reftable. */ 591 memcpy(new_table, s->refcount_table, table_size * sizeof(uint64_t)); 592 } 593 594 if (new_refblock_offset) { 595 assert(new_refblock_index < total_refblock_count); 596 new_table[new_refblock_index] = new_refblock_offset; 597 } 598 599 /* Count how many new refblocks we have to create */ 600 additional_refblock_count = 0; 601 for (i = area_reftable_index; i < total_refblock_count; i++) { 602 if (!new_table[i]) { 603 additional_refblock_count++; 604 } 605 } 606 607 table_offset = start_offset + additional_refblock_count * s->cluster_size; 608 end_offset = table_offset + table_clusters * s->cluster_size; 609 610 /* Fill the refcount blocks, and create new ones, if necessary */ 611 block_offset = start_offset; 612 for (i = area_reftable_index; i < total_refblock_count; i++) { 613 void *refblock_data; 614 uint64_t first_offset_covered; 615 616 /* Reuse an existing refblock if possible, create a new one otherwise */ 617 if (new_table[i]) { 618 ret = qcow2_cache_get(bs, s->refcount_block_cache, new_table[i], 619 &refblock_data); 620 if (ret < 0) { 621 goto fail; 622 } 623 } else { 624 ret = qcow2_cache_get_empty(bs, s->refcount_block_cache, 625 block_offset, &refblock_data); 626 if (ret < 0) { 627 goto fail; 628 } 629 memset(refblock_data, 0, s->cluster_size); 630 qcow2_cache_entry_mark_dirty(s->refcount_block_cache, 631 refblock_data); 632 633 new_table[i] = block_offset; 634 block_offset += s->cluster_size; 635 } 636 637 /* First host offset covered by this refblock */ 638 first_offset_covered = (uint64_t)i * s->refcount_block_size * 639 s->cluster_size; 640 if (first_offset_covered < end_offset) { 641 int j, end_index; 642 643 /* Set the refcount of all of the new refcount structures to 1 */ 644 645 if (first_offset_covered < start_offset) { 646 assert(i == area_reftable_index); 647 j = (start_offset - first_offset_covered) / s->cluster_size; 648 assert(j < s->refcount_block_size); 649 } else { 650 j = 0; 651 } 652 653 end_index = MIN((end_offset - first_offset_covered) / 654 s->cluster_size, 655 s->refcount_block_size); 656 657 for (; j < end_index; j++) { 658 /* The caller guaranteed us this space would be empty */ 659 assert(s->get_refcount(refblock_data, j) == 0); 660 s->set_refcount(refblock_data, j, 1); 661 } 662 663 qcow2_cache_entry_mark_dirty(s->refcount_block_cache, 664 refblock_data); 665 } 666 667 qcow2_cache_put(s->refcount_block_cache, &refblock_data); 668 } 669 670 assert(block_offset == table_offset); 671 672 /* Write refcount blocks to disk */ 673 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE_BLOCKS); 674 ret = qcow2_cache_flush(bs, s->refcount_block_cache); 675 if (ret < 0) { 676 goto fail; 677 } 678 679 /* Write refcount table to disk */ 680 for (i = 0; i < total_refblock_count; i++) { 681 cpu_to_be64s(&new_table[i]); 682 } 683 684 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE_TABLE); 685 ret = bdrv_pwrite_sync(bs->file, table_offset, new_table, 686 table_size * sizeof(uint64_t)); 687 if (ret < 0) { 688 goto fail; 689 } 690 691 for (i = 0; i < total_refblock_count; i++) { 692 be64_to_cpus(&new_table[i]); 693 } 694 695 /* Hook up the new refcount table in the qcow2 header */ 696 struct QEMU_PACKED { 697 uint64_t d64; 698 uint32_t d32; 699 } data; 700 data.d64 = cpu_to_be64(table_offset); 701 data.d32 = cpu_to_be32(table_clusters); 702 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_SWITCH_TABLE); 703 ret = bdrv_pwrite_sync(bs->file, 704 offsetof(QCowHeader, refcount_table_offset), 705 &data, sizeof(data)); 706 if (ret < 0) { 707 goto fail; 708 } 709 710 /* And switch it in memory */ 711 uint64_t old_table_offset = s->refcount_table_offset; 712 uint64_t old_table_size = s->refcount_table_size; 713 714 g_free(s->refcount_table); 715 s->refcount_table = new_table; 716 s->refcount_table_size = table_size; 717 s->refcount_table_offset = table_offset; 718 update_max_refcount_table_index(s); 719 720 /* Free old table. */ 721 qcow2_free_clusters(bs, old_table_offset, old_table_size * sizeof(uint64_t), 722 QCOW2_DISCARD_OTHER); 723 724 return end_offset; 725 726 fail: 727 g_free(new_table); 728 return ret; 729 } 730 731 void qcow2_process_discards(BlockDriverState *bs, int ret) 732 { 733 BDRVQcow2State *s = bs->opaque; 734 Qcow2DiscardRegion *d, *next; 735 736 QTAILQ_FOREACH_SAFE(d, &s->discards, next, next) { 737 QTAILQ_REMOVE(&s->discards, d, next); 738 739 /* Discard is optional, ignore the return value */ 740 if (ret >= 0) { 741 bdrv_pdiscard(bs->file, d->offset, d->bytes); 742 } 743 744 g_free(d); 745 } 746 } 747 748 static void update_refcount_discard(BlockDriverState *bs, 749 uint64_t offset, uint64_t length) 750 { 751 BDRVQcow2State *s = bs->opaque; 752 Qcow2DiscardRegion *d, *p, *next; 753 754 QTAILQ_FOREACH(d, &s->discards, next) { 755 uint64_t new_start = MIN(offset, d->offset); 756 uint64_t new_end = MAX(offset + length, d->offset + d->bytes); 757 758 if (new_end - new_start <= length + d->bytes) { 759 /* There can't be any overlap, areas ending up here have no 760 * references any more and therefore shouldn't get freed another 761 * time. */ 762 assert(d->bytes + length == new_end - new_start); 763 d->offset = new_start; 764 d->bytes = new_end - new_start; 765 goto found; 766 } 767 } 768 769 d = g_malloc(sizeof(*d)); 770 *d = (Qcow2DiscardRegion) { 771 .bs = bs, 772 .offset = offset, 773 .bytes = length, 774 }; 775 QTAILQ_INSERT_TAIL(&s->discards, d, next); 776 777 found: 778 /* Merge discard requests if they are adjacent now */ 779 QTAILQ_FOREACH_SAFE(p, &s->discards, next, next) { 780 if (p == d 781 || p->offset > d->offset + d->bytes 782 || d->offset > p->offset + p->bytes) 783 { 784 continue; 785 } 786 787 /* Still no overlap possible */ 788 assert(p->offset == d->offset + d->bytes 789 || d->offset == p->offset + p->bytes); 790 791 QTAILQ_REMOVE(&s->discards, p, next); 792 d->offset = MIN(d->offset, p->offset); 793 d->bytes += p->bytes; 794 g_free(p); 795 } 796 } 797 798 /* XXX: cache several refcount block clusters ? */ 799 /* @addend is the absolute value of the addend; if @decrease is set, @addend 800 * will be subtracted from the current refcount, otherwise it will be added */ 801 static int QEMU_WARN_UNUSED_RESULT update_refcount(BlockDriverState *bs, 802 int64_t offset, 803 int64_t length, 804 uint64_t addend, 805 bool decrease, 806 enum qcow2_discard_type type) 807 { 808 BDRVQcow2State *s = bs->opaque; 809 int64_t start, last, cluster_offset; 810 void *refcount_block = NULL; 811 int64_t old_table_index = -1; 812 int ret; 813 814 #ifdef DEBUG_ALLOC2 815 fprintf(stderr, "update_refcount: offset=%" PRId64 " size=%" PRId64 816 " addend=%s%" PRIu64 "\n", offset, length, decrease ? "-" : "", 817 addend); 818 #endif 819 if (length < 0) { 820 return -EINVAL; 821 } else if (length == 0) { 822 return 0; 823 } 824 825 if (decrease) { 826 qcow2_cache_set_dependency(bs, s->refcount_block_cache, 827 s->l2_table_cache); 828 } 829 830 start = start_of_cluster(s, offset); 831 last = start_of_cluster(s, offset + length - 1); 832 for(cluster_offset = start; cluster_offset <= last; 833 cluster_offset += s->cluster_size) 834 { 835 int block_index; 836 uint64_t refcount; 837 int64_t cluster_index = cluster_offset >> s->cluster_bits; 838 int64_t table_index = cluster_index >> s->refcount_block_bits; 839 840 /* Load the refcount block and allocate it if needed */ 841 if (table_index != old_table_index) { 842 if (refcount_block) { 843 qcow2_cache_put(s->refcount_block_cache, &refcount_block); 844 } 845 ret = alloc_refcount_block(bs, cluster_index, &refcount_block); 846 /* If the caller needs to restart the search for free clusters, 847 * try the same ones first to see if they're still free. */ 848 if (ret == -EAGAIN) { 849 if (s->free_cluster_index > (start >> s->cluster_bits)) { 850 s->free_cluster_index = (start >> s->cluster_bits); 851 } 852 } 853 if (ret < 0) { 854 goto fail; 855 } 856 } 857 old_table_index = table_index; 858 859 qcow2_cache_entry_mark_dirty(s->refcount_block_cache, refcount_block); 860 861 /* we can update the count and save it */ 862 block_index = cluster_index & (s->refcount_block_size - 1); 863 864 refcount = s->get_refcount(refcount_block, block_index); 865 if (decrease ? (refcount - addend > refcount) 866 : (refcount + addend < refcount || 867 refcount + addend > s->refcount_max)) 868 { 869 ret = -EINVAL; 870 goto fail; 871 } 872 if (decrease) { 873 refcount -= addend; 874 } else { 875 refcount += addend; 876 } 877 if (refcount == 0 && cluster_index < s->free_cluster_index) { 878 s->free_cluster_index = cluster_index; 879 } 880 s->set_refcount(refcount_block, block_index, refcount); 881 882 if (refcount == 0) { 883 void *table; 884 885 table = qcow2_cache_is_table_offset(s->refcount_block_cache, 886 offset); 887 if (table != NULL) { 888 qcow2_cache_put(s->refcount_block_cache, &refcount_block); 889 qcow2_cache_discard(s->refcount_block_cache, table); 890 } 891 892 table = qcow2_cache_is_table_offset(s->l2_table_cache, offset); 893 if (table != NULL) { 894 qcow2_cache_discard(s->l2_table_cache, table); 895 } 896 897 if (s->discard_passthrough[type]) { 898 update_refcount_discard(bs, cluster_offset, s->cluster_size); 899 } 900 } 901 } 902 903 ret = 0; 904 fail: 905 if (!s->cache_discards) { 906 qcow2_process_discards(bs, ret); 907 } 908 909 /* Write last changed block to disk */ 910 if (refcount_block) { 911 qcow2_cache_put(s->refcount_block_cache, &refcount_block); 912 } 913 914 /* 915 * Try do undo any updates if an error is returned (This may succeed in 916 * some cases like ENOSPC for allocating a new refcount block) 917 */ 918 if (ret < 0) { 919 int dummy; 920 dummy = update_refcount(bs, offset, cluster_offset - offset, addend, 921 !decrease, QCOW2_DISCARD_NEVER); 922 (void)dummy; 923 } 924 925 return ret; 926 } 927 928 /* 929 * Increases or decreases the refcount of a given cluster. 930 * 931 * @addend is the absolute value of the addend; if @decrease is set, @addend 932 * will be subtracted from the current refcount, otherwise it will be added. 933 * 934 * On success 0 is returned; on failure -errno is returned. 935 */ 936 int qcow2_update_cluster_refcount(BlockDriverState *bs, 937 int64_t cluster_index, 938 uint64_t addend, bool decrease, 939 enum qcow2_discard_type type) 940 { 941 BDRVQcow2State *s = bs->opaque; 942 int ret; 943 944 ret = update_refcount(bs, cluster_index << s->cluster_bits, 1, addend, 945 decrease, type); 946 if (ret < 0) { 947 return ret; 948 } 949 950 return 0; 951 } 952 953 954 955 /*********************************************************/ 956 /* cluster allocation functions */ 957 958 959 960 /* return < 0 if error */ 961 static int64_t alloc_clusters_noref(BlockDriverState *bs, uint64_t size, 962 uint64_t max) 963 { 964 BDRVQcow2State *s = bs->opaque; 965 uint64_t i, nb_clusters, refcount; 966 int ret; 967 968 /* We can't allocate clusters if they may still be queued for discard. */ 969 if (s->cache_discards) { 970 qcow2_process_discards(bs, 0); 971 } 972 973 nb_clusters = size_to_clusters(s, size); 974 retry: 975 for(i = 0; i < nb_clusters; i++) { 976 uint64_t next_cluster_index = s->free_cluster_index++; 977 ret = qcow2_get_refcount(bs, next_cluster_index, &refcount); 978 979 if (ret < 0) { 980 return ret; 981 } else if (refcount != 0) { 982 goto retry; 983 } 984 } 985 986 /* Make sure that all offsets in the "allocated" range are representable 987 * in the requested max */ 988 if (s->free_cluster_index > 0 && 989 s->free_cluster_index - 1 > (max >> s->cluster_bits)) 990 { 991 return -EFBIG; 992 } 993 994 #ifdef DEBUG_ALLOC2 995 fprintf(stderr, "alloc_clusters: size=%" PRId64 " -> %" PRId64 "\n", 996 size, 997 (s->free_cluster_index - nb_clusters) << s->cluster_bits); 998 #endif 999 return (s->free_cluster_index - nb_clusters) << s->cluster_bits; 1000 } 1001 1002 int64_t qcow2_alloc_clusters(BlockDriverState *bs, uint64_t size) 1003 { 1004 int64_t offset; 1005 int ret; 1006 1007 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC); 1008 do { 1009 offset = alloc_clusters_noref(bs, size, QCOW_MAX_CLUSTER_OFFSET); 1010 if (offset < 0) { 1011 return offset; 1012 } 1013 1014 ret = update_refcount(bs, offset, size, 1, false, QCOW2_DISCARD_NEVER); 1015 } while (ret == -EAGAIN); 1016 1017 if (ret < 0) { 1018 return ret; 1019 } 1020 1021 return offset; 1022 } 1023 1024 int64_t qcow2_alloc_clusters_at(BlockDriverState *bs, uint64_t offset, 1025 int64_t nb_clusters) 1026 { 1027 BDRVQcow2State *s = bs->opaque; 1028 uint64_t cluster_index, refcount; 1029 uint64_t i; 1030 int ret; 1031 1032 assert(nb_clusters >= 0); 1033 if (nb_clusters == 0) { 1034 return 0; 1035 } 1036 1037 do { 1038 /* Check how many clusters there are free */ 1039 cluster_index = offset >> s->cluster_bits; 1040 for(i = 0; i < nb_clusters; i++) { 1041 ret = qcow2_get_refcount(bs, cluster_index++, &refcount); 1042 if (ret < 0) { 1043 return ret; 1044 } else if (refcount != 0) { 1045 break; 1046 } 1047 } 1048 1049 /* And then allocate them */ 1050 ret = update_refcount(bs, offset, i << s->cluster_bits, 1, false, 1051 QCOW2_DISCARD_NEVER); 1052 } while (ret == -EAGAIN); 1053 1054 if (ret < 0) { 1055 return ret; 1056 } 1057 1058 return i; 1059 } 1060 1061 /* only used to allocate compressed sectors. We try to allocate 1062 contiguous sectors. size must be <= cluster_size */ 1063 int64_t qcow2_alloc_bytes(BlockDriverState *bs, int size) 1064 { 1065 BDRVQcow2State *s = bs->opaque; 1066 int64_t offset; 1067 size_t free_in_cluster; 1068 int ret; 1069 1070 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC_BYTES); 1071 assert(size > 0 && size <= s->cluster_size); 1072 assert(!s->free_byte_offset || offset_into_cluster(s, s->free_byte_offset)); 1073 1074 offset = s->free_byte_offset; 1075 1076 if (offset) { 1077 uint64_t refcount; 1078 ret = qcow2_get_refcount(bs, offset >> s->cluster_bits, &refcount); 1079 if (ret < 0) { 1080 return ret; 1081 } 1082 1083 if (refcount == s->refcount_max) { 1084 offset = 0; 1085 } 1086 } 1087 1088 free_in_cluster = s->cluster_size - offset_into_cluster(s, offset); 1089 do { 1090 if (!offset || free_in_cluster < size) { 1091 int64_t new_cluster; 1092 1093 new_cluster = alloc_clusters_noref(bs, s->cluster_size, 1094 MIN(s->cluster_offset_mask, 1095 QCOW_MAX_CLUSTER_OFFSET)); 1096 if (new_cluster < 0) { 1097 return new_cluster; 1098 } 1099 1100 if (new_cluster == 0) { 1101 qcow2_signal_corruption(bs, true, -1, -1, "Preventing invalid " 1102 "allocation of compressed cluster " 1103 "at offset 0"); 1104 return -EIO; 1105 } 1106 1107 if (!offset || ROUND_UP(offset, s->cluster_size) != new_cluster) { 1108 offset = new_cluster; 1109 free_in_cluster = s->cluster_size; 1110 } else { 1111 free_in_cluster += s->cluster_size; 1112 } 1113 } 1114 1115 assert(offset); 1116 ret = update_refcount(bs, offset, size, 1, false, QCOW2_DISCARD_NEVER); 1117 if (ret < 0) { 1118 offset = 0; 1119 } 1120 } while (ret == -EAGAIN); 1121 if (ret < 0) { 1122 return ret; 1123 } 1124 1125 /* The cluster refcount was incremented; refcount blocks must be flushed 1126 * before the caller's L2 table updates. */ 1127 qcow2_cache_set_dependency(bs, s->l2_table_cache, s->refcount_block_cache); 1128 1129 s->free_byte_offset = offset + size; 1130 if (!offset_into_cluster(s, s->free_byte_offset)) { 1131 s->free_byte_offset = 0; 1132 } 1133 1134 return offset; 1135 } 1136 1137 void qcow2_free_clusters(BlockDriverState *bs, 1138 int64_t offset, int64_t size, 1139 enum qcow2_discard_type type) 1140 { 1141 int ret; 1142 1143 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_FREE); 1144 ret = update_refcount(bs, offset, size, 1, true, type); 1145 if (ret < 0) { 1146 fprintf(stderr, "qcow2_free_clusters failed: %s\n", strerror(-ret)); 1147 /* TODO Remember the clusters to free them later and avoid leaking */ 1148 } 1149 } 1150 1151 /* 1152 * Free a cluster using its L2 entry (handles clusters of all types, e.g. 1153 * normal cluster, compressed cluster, etc.) 1154 */ 1155 void qcow2_free_any_clusters(BlockDriverState *bs, uint64_t l2_entry, 1156 int nb_clusters, enum qcow2_discard_type type) 1157 { 1158 BDRVQcow2State *s = bs->opaque; 1159 QCow2ClusterType ctype = qcow2_get_cluster_type(bs, l2_entry); 1160 1161 if (has_data_file(bs)) { 1162 if (s->discard_passthrough[type] && 1163 (ctype == QCOW2_CLUSTER_NORMAL || 1164 ctype == QCOW2_CLUSTER_ZERO_ALLOC)) 1165 { 1166 bdrv_pdiscard(s->data_file, l2_entry & L2E_OFFSET_MASK, 1167 nb_clusters << s->cluster_bits); 1168 } 1169 return; 1170 } 1171 1172 switch (ctype) { 1173 case QCOW2_CLUSTER_COMPRESSED: 1174 { 1175 int nb_csectors; 1176 nb_csectors = ((l2_entry >> s->csize_shift) & 1177 s->csize_mask) + 1; 1178 qcow2_free_clusters(bs, 1179 (l2_entry & s->cluster_offset_mask) & ~511, 1180 nb_csectors * 512, type); 1181 } 1182 break; 1183 case QCOW2_CLUSTER_NORMAL: 1184 case QCOW2_CLUSTER_ZERO_ALLOC: 1185 if (offset_into_cluster(s, l2_entry & L2E_OFFSET_MASK)) { 1186 qcow2_signal_corruption(bs, false, -1, -1, 1187 "Cannot free unaligned cluster %#llx", 1188 l2_entry & L2E_OFFSET_MASK); 1189 } else { 1190 qcow2_free_clusters(bs, l2_entry & L2E_OFFSET_MASK, 1191 nb_clusters << s->cluster_bits, type); 1192 } 1193 break; 1194 case QCOW2_CLUSTER_ZERO_PLAIN: 1195 case QCOW2_CLUSTER_UNALLOCATED: 1196 break; 1197 default: 1198 abort(); 1199 } 1200 } 1201 1202 int coroutine_fn qcow2_write_caches(BlockDriverState *bs) 1203 { 1204 BDRVQcow2State *s = bs->opaque; 1205 int ret; 1206 1207 ret = qcow2_cache_write(bs, s->l2_table_cache); 1208 if (ret < 0) { 1209 return ret; 1210 } 1211 1212 if (qcow2_need_accurate_refcounts(s)) { 1213 ret = qcow2_cache_write(bs, s->refcount_block_cache); 1214 if (ret < 0) { 1215 return ret; 1216 } 1217 } 1218 1219 return 0; 1220 } 1221 1222 int coroutine_fn qcow2_flush_caches(BlockDriverState *bs) 1223 { 1224 int ret = qcow2_write_caches(bs); 1225 if (ret < 0) { 1226 return ret; 1227 } 1228 1229 return bdrv_flush(bs->file->bs); 1230 } 1231 1232 /*********************************************************/ 1233 /* snapshots and image creation */ 1234 1235 1236 1237 /* update the refcounts of snapshots and the copied flag */ 1238 int qcow2_update_snapshot_refcount(BlockDriverState *bs, 1239 int64_t l1_table_offset, int l1_size, int addend) 1240 { 1241 BDRVQcow2State *s = bs->opaque; 1242 uint64_t *l1_table, *l2_slice, l2_offset, entry, l1_size2, refcount; 1243 bool l1_allocated = false; 1244 int64_t old_entry, old_l2_offset; 1245 unsigned slice, slice_size2, n_slices; 1246 int i, j, l1_modified = 0, nb_csectors; 1247 int ret; 1248 1249 assert(addend >= -1 && addend <= 1); 1250 1251 l2_slice = NULL; 1252 l1_table = NULL; 1253 l1_size2 = l1_size * sizeof(uint64_t); 1254 slice_size2 = s->l2_slice_size * sizeof(uint64_t); 1255 n_slices = s->cluster_size / slice_size2; 1256 1257 s->cache_discards = true; 1258 1259 /* WARNING: qcow2_snapshot_goto relies on this function not using the 1260 * l1_table_offset when it is the current s->l1_table_offset! Be careful 1261 * when changing this! */ 1262 if (l1_table_offset != s->l1_table_offset) { 1263 l1_table = g_try_malloc0(ROUND_UP(l1_size2, 512)); 1264 if (l1_size2 && l1_table == NULL) { 1265 ret = -ENOMEM; 1266 goto fail; 1267 } 1268 l1_allocated = true; 1269 1270 ret = bdrv_pread(bs->file, l1_table_offset, l1_table, l1_size2); 1271 if (ret < 0) { 1272 goto fail; 1273 } 1274 1275 for (i = 0; i < l1_size; i++) { 1276 be64_to_cpus(&l1_table[i]); 1277 } 1278 } else { 1279 assert(l1_size == s->l1_size); 1280 l1_table = s->l1_table; 1281 l1_allocated = false; 1282 } 1283 1284 for (i = 0; i < l1_size; i++) { 1285 l2_offset = l1_table[i]; 1286 if (l2_offset) { 1287 old_l2_offset = l2_offset; 1288 l2_offset &= L1E_OFFSET_MASK; 1289 1290 if (offset_into_cluster(s, l2_offset)) { 1291 qcow2_signal_corruption(bs, true, -1, -1, "L2 table offset %#" 1292 PRIx64 " unaligned (L1 index: %#x)", 1293 l2_offset, i); 1294 ret = -EIO; 1295 goto fail; 1296 } 1297 1298 for (slice = 0; slice < n_slices; slice++) { 1299 ret = qcow2_cache_get(bs, s->l2_table_cache, 1300 l2_offset + slice * slice_size2, 1301 (void **) &l2_slice); 1302 if (ret < 0) { 1303 goto fail; 1304 } 1305 1306 for (j = 0; j < s->l2_slice_size; j++) { 1307 uint64_t cluster_index; 1308 uint64_t offset; 1309 1310 entry = be64_to_cpu(l2_slice[j]); 1311 old_entry = entry; 1312 entry &= ~QCOW_OFLAG_COPIED; 1313 offset = entry & L2E_OFFSET_MASK; 1314 1315 switch (qcow2_get_cluster_type(bs, entry)) { 1316 case QCOW2_CLUSTER_COMPRESSED: 1317 nb_csectors = ((entry >> s->csize_shift) & 1318 s->csize_mask) + 1; 1319 if (addend != 0) { 1320 ret = update_refcount( 1321 bs, (entry & s->cluster_offset_mask) & ~511, 1322 nb_csectors * 512, abs(addend), addend < 0, 1323 QCOW2_DISCARD_SNAPSHOT); 1324 if (ret < 0) { 1325 goto fail; 1326 } 1327 } 1328 /* compressed clusters are never modified */ 1329 refcount = 2; 1330 break; 1331 1332 case QCOW2_CLUSTER_NORMAL: 1333 case QCOW2_CLUSTER_ZERO_ALLOC: 1334 if (offset_into_cluster(s, offset)) { 1335 /* Here l2_index means table (not slice) index */ 1336 int l2_index = slice * s->l2_slice_size + j; 1337 qcow2_signal_corruption( 1338 bs, true, -1, -1, "Cluster " 1339 "allocation offset %#" PRIx64 1340 " unaligned (L2 offset: %#" 1341 PRIx64 ", L2 index: %#x)", 1342 offset, l2_offset, l2_index); 1343 ret = -EIO; 1344 goto fail; 1345 } 1346 1347 cluster_index = offset >> s->cluster_bits; 1348 assert(cluster_index); 1349 if (addend != 0) { 1350 ret = qcow2_update_cluster_refcount( 1351 bs, cluster_index, abs(addend), addend < 0, 1352 QCOW2_DISCARD_SNAPSHOT); 1353 if (ret < 0) { 1354 goto fail; 1355 } 1356 } 1357 1358 ret = qcow2_get_refcount(bs, cluster_index, &refcount); 1359 if (ret < 0) { 1360 goto fail; 1361 } 1362 break; 1363 1364 case QCOW2_CLUSTER_ZERO_PLAIN: 1365 case QCOW2_CLUSTER_UNALLOCATED: 1366 refcount = 0; 1367 break; 1368 1369 default: 1370 abort(); 1371 } 1372 1373 if (refcount == 1) { 1374 entry |= QCOW_OFLAG_COPIED; 1375 } 1376 if (entry != old_entry) { 1377 if (addend > 0) { 1378 qcow2_cache_set_dependency(bs, s->l2_table_cache, 1379 s->refcount_block_cache); 1380 } 1381 l2_slice[j] = cpu_to_be64(entry); 1382 qcow2_cache_entry_mark_dirty(s->l2_table_cache, 1383 l2_slice); 1384 } 1385 } 1386 1387 qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice); 1388 } 1389 1390 if (addend != 0) { 1391 ret = qcow2_update_cluster_refcount(bs, l2_offset >> 1392 s->cluster_bits, 1393 abs(addend), addend < 0, 1394 QCOW2_DISCARD_SNAPSHOT); 1395 if (ret < 0) { 1396 goto fail; 1397 } 1398 } 1399 ret = qcow2_get_refcount(bs, l2_offset >> s->cluster_bits, 1400 &refcount); 1401 if (ret < 0) { 1402 goto fail; 1403 } else if (refcount == 1) { 1404 l2_offset |= QCOW_OFLAG_COPIED; 1405 } 1406 if (l2_offset != old_l2_offset) { 1407 l1_table[i] = l2_offset; 1408 l1_modified = 1; 1409 } 1410 } 1411 } 1412 1413 ret = bdrv_flush(bs); 1414 fail: 1415 if (l2_slice) { 1416 qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice); 1417 } 1418 1419 s->cache_discards = false; 1420 qcow2_process_discards(bs, ret); 1421 1422 /* Update L1 only if it isn't deleted anyway (addend = -1) */ 1423 if (ret == 0 && addend >= 0 && l1_modified) { 1424 for (i = 0; i < l1_size; i++) { 1425 cpu_to_be64s(&l1_table[i]); 1426 } 1427 1428 ret = bdrv_pwrite_sync(bs->file, l1_table_offset, 1429 l1_table, l1_size2); 1430 1431 for (i = 0; i < l1_size; i++) { 1432 be64_to_cpus(&l1_table[i]); 1433 } 1434 } 1435 if (l1_allocated) 1436 g_free(l1_table); 1437 return ret; 1438 } 1439 1440 1441 1442 1443 /*********************************************************/ 1444 /* refcount checking functions */ 1445 1446 1447 static uint64_t refcount_array_byte_size(BDRVQcow2State *s, uint64_t entries) 1448 { 1449 /* This assertion holds because there is no way we can address more than 1450 * 2^(64 - 9) clusters at once (with cluster size 512 = 2^9, and because 1451 * offsets have to be representable in bytes); due to every cluster 1452 * corresponding to one refcount entry, we are well below that limit */ 1453 assert(entries < (UINT64_C(1) << (64 - 9))); 1454 1455 /* Thanks to the assertion this will not overflow, because 1456 * s->refcount_order < 7. 1457 * (note: x << s->refcount_order == x * s->refcount_bits) */ 1458 return DIV_ROUND_UP(entries << s->refcount_order, 8); 1459 } 1460 1461 /** 1462 * Reallocates *array so that it can hold new_size entries. *size must contain 1463 * the current number of entries in *array. If the reallocation fails, *array 1464 * and *size will not be modified and -errno will be returned. If the 1465 * reallocation is successful, *array will be set to the new buffer, *size 1466 * will be set to new_size and 0 will be returned. The size of the reallocated 1467 * refcount array buffer will be aligned to a cluster boundary, and the newly 1468 * allocated area will be zeroed. 1469 */ 1470 static int realloc_refcount_array(BDRVQcow2State *s, void **array, 1471 int64_t *size, int64_t new_size) 1472 { 1473 int64_t old_byte_size, new_byte_size; 1474 void *new_ptr; 1475 1476 /* Round to clusters so the array can be directly written to disk */ 1477 old_byte_size = size_to_clusters(s, refcount_array_byte_size(s, *size)) 1478 * s->cluster_size; 1479 new_byte_size = size_to_clusters(s, refcount_array_byte_size(s, new_size)) 1480 * s->cluster_size; 1481 1482 if (new_byte_size == old_byte_size) { 1483 *size = new_size; 1484 return 0; 1485 } 1486 1487 assert(new_byte_size > 0); 1488 1489 if (new_byte_size > SIZE_MAX) { 1490 return -ENOMEM; 1491 } 1492 1493 new_ptr = g_try_realloc(*array, new_byte_size); 1494 if (!new_ptr) { 1495 return -ENOMEM; 1496 } 1497 1498 if (new_byte_size > old_byte_size) { 1499 memset((char *)new_ptr + old_byte_size, 0, 1500 new_byte_size - old_byte_size); 1501 } 1502 1503 *array = new_ptr; 1504 *size = new_size; 1505 1506 return 0; 1507 } 1508 1509 /* 1510 * Increases the refcount for a range of clusters in a given refcount table. 1511 * This is used to construct a temporary refcount table out of L1 and L2 tables 1512 * which can be compared to the refcount table saved in the image. 1513 * 1514 * Modifies the number of errors in res. 1515 */ 1516 int qcow2_inc_refcounts_imrt(BlockDriverState *bs, BdrvCheckResult *res, 1517 void **refcount_table, 1518 int64_t *refcount_table_size, 1519 int64_t offset, int64_t size) 1520 { 1521 BDRVQcow2State *s = bs->opaque; 1522 uint64_t start, last, cluster_offset, k, refcount; 1523 int64_t file_len; 1524 int ret; 1525 1526 if (size <= 0) { 1527 return 0; 1528 } 1529 1530 file_len = bdrv_getlength(bs->file->bs); 1531 if (file_len < 0) { 1532 return file_len; 1533 } 1534 1535 /* 1536 * Last cluster of qcow2 image may be semi-allocated, so it may be OK to 1537 * reference some space after file end but it should be less than one 1538 * cluster. 1539 */ 1540 if (offset + size - file_len >= s->cluster_size) { 1541 fprintf(stderr, "ERROR: counting reference for region exceeding the " 1542 "end of the file by one cluster or more: offset 0x%" PRIx64 1543 " size 0x%" PRIx64 "\n", offset, size); 1544 res->corruptions++; 1545 return 0; 1546 } 1547 1548 start = start_of_cluster(s, offset); 1549 last = start_of_cluster(s, offset + size - 1); 1550 for(cluster_offset = start; cluster_offset <= last; 1551 cluster_offset += s->cluster_size) { 1552 k = cluster_offset >> s->cluster_bits; 1553 if (k >= *refcount_table_size) { 1554 ret = realloc_refcount_array(s, refcount_table, 1555 refcount_table_size, k + 1); 1556 if (ret < 0) { 1557 res->check_errors++; 1558 return ret; 1559 } 1560 } 1561 1562 refcount = s->get_refcount(*refcount_table, k); 1563 if (refcount == s->refcount_max) { 1564 fprintf(stderr, "ERROR: overflow cluster offset=0x%" PRIx64 1565 "\n", cluster_offset); 1566 fprintf(stderr, "Use qemu-img amend to increase the refcount entry " 1567 "width or qemu-img convert to create a clean copy if the " 1568 "image cannot be opened for writing\n"); 1569 res->corruptions++; 1570 continue; 1571 } 1572 s->set_refcount(*refcount_table, k, refcount + 1); 1573 } 1574 1575 return 0; 1576 } 1577 1578 /* Flags for check_refcounts_l1() and check_refcounts_l2() */ 1579 enum { 1580 CHECK_FRAG_INFO = 0x2, /* update BlockFragInfo counters */ 1581 }; 1582 1583 /* 1584 * Increases the refcount in the given refcount table for the all clusters 1585 * referenced in the L2 table. While doing so, performs some checks on L2 1586 * entries. 1587 * 1588 * Returns the number of errors found by the checks or -errno if an internal 1589 * error occurred. 1590 */ 1591 static int check_refcounts_l2(BlockDriverState *bs, BdrvCheckResult *res, 1592 void **refcount_table, 1593 int64_t *refcount_table_size, int64_t l2_offset, 1594 int flags, BdrvCheckMode fix, bool active) 1595 { 1596 BDRVQcow2State *s = bs->opaque; 1597 uint64_t *l2_table, l2_entry; 1598 uint64_t next_contiguous_offset = 0; 1599 int i, l2_size, nb_csectors, ret; 1600 1601 /* Read L2 table from disk */ 1602 l2_size = s->l2_size * sizeof(uint64_t); 1603 l2_table = g_malloc(l2_size); 1604 1605 ret = bdrv_pread(bs->file, l2_offset, l2_table, l2_size); 1606 if (ret < 0) { 1607 fprintf(stderr, "ERROR: I/O error in check_refcounts_l2\n"); 1608 res->check_errors++; 1609 goto fail; 1610 } 1611 1612 /* Do the actual checks */ 1613 for(i = 0; i < s->l2_size; i++) { 1614 l2_entry = be64_to_cpu(l2_table[i]); 1615 1616 switch (qcow2_get_cluster_type(bs, l2_entry)) { 1617 case QCOW2_CLUSTER_COMPRESSED: 1618 /* Compressed clusters don't have QCOW_OFLAG_COPIED */ 1619 if (l2_entry & QCOW_OFLAG_COPIED) { 1620 fprintf(stderr, "ERROR: coffset=0x%" PRIx64 ": " 1621 "copied flag must never be set for compressed " 1622 "clusters\n", l2_entry & s->cluster_offset_mask); 1623 l2_entry &= ~QCOW_OFLAG_COPIED; 1624 res->corruptions++; 1625 } 1626 1627 if (has_data_file(bs)) { 1628 fprintf(stderr, "ERROR compressed cluster %d with data file, " 1629 "entry=0x%" PRIx64 "\n", i, l2_entry); 1630 res->corruptions++; 1631 break; 1632 } 1633 1634 /* Mark cluster as used */ 1635 nb_csectors = ((l2_entry >> s->csize_shift) & 1636 s->csize_mask) + 1; 1637 l2_entry &= s->cluster_offset_mask; 1638 ret = qcow2_inc_refcounts_imrt(bs, res, 1639 refcount_table, refcount_table_size, 1640 l2_entry & ~511, nb_csectors * 512); 1641 if (ret < 0) { 1642 goto fail; 1643 } 1644 1645 if (flags & CHECK_FRAG_INFO) { 1646 res->bfi.allocated_clusters++; 1647 res->bfi.compressed_clusters++; 1648 1649 /* Compressed clusters are fragmented by nature. Since they 1650 * take up sub-sector space but we only have sector granularity 1651 * I/O we need to re-read the same sectors even for adjacent 1652 * compressed clusters. 1653 */ 1654 res->bfi.fragmented_clusters++; 1655 } 1656 break; 1657 1658 case QCOW2_CLUSTER_ZERO_ALLOC: 1659 case QCOW2_CLUSTER_NORMAL: 1660 { 1661 uint64_t offset = l2_entry & L2E_OFFSET_MASK; 1662 1663 /* Correct offsets are cluster aligned */ 1664 if (offset_into_cluster(s, offset)) { 1665 res->corruptions++; 1666 1667 if (qcow2_get_cluster_type(bs, l2_entry) == 1668 QCOW2_CLUSTER_ZERO_ALLOC) 1669 { 1670 fprintf(stderr, "%s offset=%" PRIx64 ": Preallocated zero " 1671 "cluster is not properly aligned; L2 entry " 1672 "corrupted.\n", 1673 fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR", 1674 offset); 1675 if (fix & BDRV_FIX_ERRORS) { 1676 uint64_t l2e_offset = 1677 l2_offset + (uint64_t)i * sizeof(uint64_t); 1678 int ign = active ? QCOW2_OL_ACTIVE_L2 : 1679 QCOW2_OL_INACTIVE_L2; 1680 1681 l2_entry = QCOW_OFLAG_ZERO; 1682 l2_table[i] = cpu_to_be64(l2_entry); 1683 ret = qcow2_pre_write_overlap_check(bs, ign, 1684 l2e_offset, sizeof(uint64_t), false); 1685 if (ret < 0) { 1686 fprintf(stderr, "ERROR: Overlap check failed\n"); 1687 res->check_errors++; 1688 /* Something is seriously wrong, so abort checking 1689 * this L2 table */ 1690 goto fail; 1691 } 1692 1693 ret = bdrv_pwrite_sync(bs->file, l2e_offset, 1694 &l2_table[i], sizeof(uint64_t)); 1695 if (ret < 0) { 1696 fprintf(stderr, "ERROR: Failed to overwrite L2 " 1697 "table entry: %s\n", strerror(-ret)); 1698 res->check_errors++; 1699 /* Do not abort, continue checking the rest of this 1700 * L2 table's entries */ 1701 } else { 1702 res->corruptions--; 1703 res->corruptions_fixed++; 1704 /* Skip marking the cluster as used 1705 * (it is unused now) */ 1706 continue; 1707 } 1708 } 1709 } else { 1710 fprintf(stderr, "ERROR offset=%" PRIx64 ": Data cluster is " 1711 "not properly aligned; L2 entry corrupted.\n", offset); 1712 } 1713 } 1714 1715 if (flags & CHECK_FRAG_INFO) { 1716 res->bfi.allocated_clusters++; 1717 if (next_contiguous_offset && 1718 offset != next_contiguous_offset) { 1719 res->bfi.fragmented_clusters++; 1720 } 1721 next_contiguous_offset = offset + s->cluster_size; 1722 } 1723 1724 /* Mark cluster as used */ 1725 if (!has_data_file(bs)) { 1726 ret = qcow2_inc_refcounts_imrt(bs, res, refcount_table, 1727 refcount_table_size, 1728 offset, s->cluster_size); 1729 if (ret < 0) { 1730 goto fail; 1731 } 1732 } 1733 break; 1734 } 1735 1736 case QCOW2_CLUSTER_ZERO_PLAIN: 1737 case QCOW2_CLUSTER_UNALLOCATED: 1738 break; 1739 1740 default: 1741 abort(); 1742 } 1743 } 1744 1745 g_free(l2_table); 1746 return 0; 1747 1748 fail: 1749 g_free(l2_table); 1750 return ret; 1751 } 1752 1753 /* 1754 * Increases the refcount for the L1 table, its L2 tables and all referenced 1755 * clusters in the given refcount table. While doing so, performs some checks 1756 * on L1 and L2 entries. 1757 * 1758 * Returns the number of errors found by the checks or -errno if an internal 1759 * error occurred. 1760 */ 1761 static int check_refcounts_l1(BlockDriverState *bs, 1762 BdrvCheckResult *res, 1763 void **refcount_table, 1764 int64_t *refcount_table_size, 1765 int64_t l1_table_offset, int l1_size, 1766 int flags, BdrvCheckMode fix, bool active) 1767 { 1768 BDRVQcow2State *s = bs->opaque; 1769 uint64_t *l1_table = NULL, l2_offset, l1_size2; 1770 int i, ret; 1771 1772 l1_size2 = l1_size * sizeof(uint64_t); 1773 1774 /* Mark L1 table as used */ 1775 ret = qcow2_inc_refcounts_imrt(bs, res, refcount_table, refcount_table_size, 1776 l1_table_offset, l1_size2); 1777 if (ret < 0) { 1778 goto fail; 1779 } 1780 1781 /* Read L1 table entries from disk */ 1782 if (l1_size2 > 0) { 1783 l1_table = g_try_malloc(l1_size2); 1784 if (l1_table == NULL) { 1785 ret = -ENOMEM; 1786 res->check_errors++; 1787 goto fail; 1788 } 1789 ret = bdrv_pread(bs->file, l1_table_offset, l1_table, l1_size2); 1790 if (ret < 0) { 1791 fprintf(stderr, "ERROR: I/O error in check_refcounts_l1\n"); 1792 res->check_errors++; 1793 goto fail; 1794 } 1795 for(i = 0;i < l1_size; i++) 1796 be64_to_cpus(&l1_table[i]); 1797 } 1798 1799 /* Do the actual checks */ 1800 for(i = 0; i < l1_size; i++) { 1801 l2_offset = l1_table[i]; 1802 if (l2_offset) { 1803 /* Mark L2 table as used */ 1804 l2_offset &= L1E_OFFSET_MASK; 1805 ret = qcow2_inc_refcounts_imrt(bs, res, 1806 refcount_table, refcount_table_size, 1807 l2_offset, s->cluster_size); 1808 if (ret < 0) { 1809 goto fail; 1810 } 1811 1812 /* L2 tables are cluster aligned */ 1813 if (offset_into_cluster(s, l2_offset)) { 1814 fprintf(stderr, "ERROR l2_offset=%" PRIx64 ": Table is not " 1815 "cluster aligned; L1 entry corrupted\n", l2_offset); 1816 res->corruptions++; 1817 } 1818 1819 /* Process and check L2 entries */ 1820 ret = check_refcounts_l2(bs, res, refcount_table, 1821 refcount_table_size, l2_offset, flags, 1822 fix, active); 1823 if (ret < 0) { 1824 goto fail; 1825 } 1826 } 1827 } 1828 g_free(l1_table); 1829 return 0; 1830 1831 fail: 1832 g_free(l1_table); 1833 return ret; 1834 } 1835 1836 /* 1837 * Checks the OFLAG_COPIED flag for all L1 and L2 entries. 1838 * 1839 * This function does not print an error message nor does it increment 1840 * check_errors if qcow2_get_refcount fails (this is because such an error will 1841 * have been already detected and sufficiently signaled by the calling function 1842 * (qcow2_check_refcounts) by the time this function is called). 1843 */ 1844 static int check_oflag_copied(BlockDriverState *bs, BdrvCheckResult *res, 1845 BdrvCheckMode fix) 1846 { 1847 BDRVQcow2State *s = bs->opaque; 1848 uint64_t *l2_table = qemu_blockalign(bs, s->cluster_size); 1849 int ret; 1850 uint64_t refcount; 1851 int i, j; 1852 bool repair; 1853 1854 if (fix & BDRV_FIX_ERRORS) { 1855 /* Always repair */ 1856 repair = true; 1857 } else if (fix & BDRV_FIX_LEAKS) { 1858 /* Repair only if that seems safe: This function is always 1859 * called after the refcounts have been fixed, so the refcount 1860 * is accurate if that repair was successful */ 1861 repair = !res->check_errors && !res->corruptions && !res->leaks; 1862 } else { 1863 repair = false; 1864 } 1865 1866 for (i = 0; i < s->l1_size; i++) { 1867 uint64_t l1_entry = s->l1_table[i]; 1868 uint64_t l2_offset = l1_entry & L1E_OFFSET_MASK; 1869 int l2_dirty = 0; 1870 1871 if (!l2_offset) { 1872 continue; 1873 } 1874 1875 ret = qcow2_get_refcount(bs, l2_offset >> s->cluster_bits, 1876 &refcount); 1877 if (ret < 0) { 1878 /* don't print message nor increment check_errors */ 1879 continue; 1880 } 1881 if ((refcount == 1) != ((l1_entry & QCOW_OFLAG_COPIED) != 0)) { 1882 res->corruptions++; 1883 fprintf(stderr, "%s OFLAG_COPIED L2 cluster: l1_index=%d " 1884 "l1_entry=%" PRIx64 " refcount=%" PRIu64 "\n", 1885 repair ? "Repairing" : "ERROR", i, l1_entry, refcount); 1886 if (repair) { 1887 s->l1_table[i] = refcount == 1 1888 ? l1_entry | QCOW_OFLAG_COPIED 1889 : l1_entry & ~QCOW_OFLAG_COPIED; 1890 ret = qcow2_write_l1_entry(bs, i); 1891 if (ret < 0) { 1892 res->check_errors++; 1893 goto fail; 1894 } 1895 res->corruptions--; 1896 res->corruptions_fixed++; 1897 } 1898 } 1899 1900 ret = bdrv_pread(bs->file, l2_offset, l2_table, 1901 s->l2_size * sizeof(uint64_t)); 1902 if (ret < 0) { 1903 fprintf(stderr, "ERROR: Could not read L2 table: %s\n", 1904 strerror(-ret)); 1905 res->check_errors++; 1906 goto fail; 1907 } 1908 1909 for (j = 0; j < s->l2_size; j++) { 1910 uint64_t l2_entry = be64_to_cpu(l2_table[j]); 1911 uint64_t data_offset = l2_entry & L2E_OFFSET_MASK; 1912 QCow2ClusterType cluster_type = qcow2_get_cluster_type(bs, l2_entry); 1913 1914 if (cluster_type == QCOW2_CLUSTER_NORMAL || 1915 cluster_type == QCOW2_CLUSTER_ZERO_ALLOC) { 1916 if (has_data_file(bs)) { 1917 refcount = 1; 1918 } else { 1919 ret = qcow2_get_refcount(bs, 1920 data_offset >> s->cluster_bits, 1921 &refcount); 1922 if (ret < 0) { 1923 /* don't print message nor increment check_errors */ 1924 continue; 1925 } 1926 } 1927 if ((refcount == 1) != ((l2_entry & QCOW_OFLAG_COPIED) != 0)) { 1928 res->corruptions++; 1929 fprintf(stderr, "%s OFLAG_COPIED data cluster: " 1930 "l2_entry=%" PRIx64 " refcount=%" PRIu64 "\n", 1931 repair ? "Repairing" : "ERROR", l2_entry, refcount); 1932 if (repair) { 1933 l2_table[j] = cpu_to_be64(refcount == 1 1934 ? l2_entry | QCOW_OFLAG_COPIED 1935 : l2_entry & ~QCOW_OFLAG_COPIED); 1936 l2_dirty++; 1937 } 1938 } 1939 } 1940 } 1941 1942 if (l2_dirty > 0) { 1943 ret = qcow2_pre_write_overlap_check(bs, QCOW2_OL_ACTIVE_L2, 1944 l2_offset, s->cluster_size, 1945 false); 1946 if (ret < 0) { 1947 fprintf(stderr, "ERROR: Could not write L2 table; metadata " 1948 "overlap check failed: %s\n", strerror(-ret)); 1949 res->check_errors++; 1950 goto fail; 1951 } 1952 1953 ret = bdrv_pwrite(bs->file, l2_offset, l2_table, 1954 s->cluster_size); 1955 if (ret < 0) { 1956 fprintf(stderr, "ERROR: Could not write L2 table: %s\n", 1957 strerror(-ret)); 1958 res->check_errors++; 1959 goto fail; 1960 } 1961 res->corruptions -= l2_dirty; 1962 res->corruptions_fixed += l2_dirty; 1963 } 1964 } 1965 1966 ret = 0; 1967 1968 fail: 1969 qemu_vfree(l2_table); 1970 return ret; 1971 } 1972 1973 /* 1974 * Checks consistency of refblocks and accounts for each refblock in 1975 * *refcount_table. 1976 */ 1977 static int check_refblocks(BlockDriverState *bs, BdrvCheckResult *res, 1978 BdrvCheckMode fix, bool *rebuild, 1979 void **refcount_table, int64_t *nb_clusters) 1980 { 1981 BDRVQcow2State *s = bs->opaque; 1982 int64_t i, size; 1983 int ret; 1984 1985 for(i = 0; i < s->refcount_table_size; i++) { 1986 uint64_t offset, cluster; 1987 offset = s->refcount_table[i]; 1988 cluster = offset >> s->cluster_bits; 1989 1990 /* Refcount blocks are cluster aligned */ 1991 if (offset_into_cluster(s, offset)) { 1992 fprintf(stderr, "ERROR refcount block %" PRId64 " is not " 1993 "cluster aligned; refcount table entry corrupted\n", i); 1994 res->corruptions++; 1995 *rebuild = true; 1996 continue; 1997 } 1998 1999 if (cluster >= *nb_clusters) { 2000 res->corruptions++; 2001 fprintf(stderr, "%s refcount block %" PRId64 " is outside image\n", 2002 fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR", i); 2003 2004 if (fix & BDRV_FIX_ERRORS) { 2005 int64_t new_nb_clusters; 2006 Error *local_err = NULL; 2007 2008 if (offset > INT64_MAX - s->cluster_size) { 2009 ret = -EINVAL; 2010 goto resize_fail; 2011 } 2012 2013 ret = bdrv_truncate(bs->file, offset + s->cluster_size, 2014 PREALLOC_MODE_OFF, &local_err); 2015 if (ret < 0) { 2016 error_report_err(local_err); 2017 goto resize_fail; 2018 } 2019 size = bdrv_getlength(bs->file->bs); 2020 if (size < 0) { 2021 ret = size; 2022 goto resize_fail; 2023 } 2024 2025 new_nb_clusters = size_to_clusters(s, size); 2026 assert(new_nb_clusters >= *nb_clusters); 2027 2028 ret = realloc_refcount_array(s, refcount_table, 2029 nb_clusters, new_nb_clusters); 2030 if (ret < 0) { 2031 res->check_errors++; 2032 return ret; 2033 } 2034 2035 if (cluster >= *nb_clusters) { 2036 ret = -EINVAL; 2037 goto resize_fail; 2038 } 2039 2040 res->corruptions--; 2041 res->corruptions_fixed++; 2042 ret = qcow2_inc_refcounts_imrt(bs, res, 2043 refcount_table, nb_clusters, 2044 offset, s->cluster_size); 2045 if (ret < 0) { 2046 return ret; 2047 } 2048 /* No need to check whether the refcount is now greater than 1: 2049 * This area was just allocated and zeroed, so it can only be 2050 * exactly 1 after qcow2_inc_refcounts_imrt() */ 2051 continue; 2052 2053 resize_fail: 2054 *rebuild = true; 2055 fprintf(stderr, "ERROR could not resize image: %s\n", 2056 strerror(-ret)); 2057 } 2058 continue; 2059 } 2060 2061 if (offset != 0) { 2062 ret = qcow2_inc_refcounts_imrt(bs, res, refcount_table, nb_clusters, 2063 offset, s->cluster_size); 2064 if (ret < 0) { 2065 return ret; 2066 } 2067 if (s->get_refcount(*refcount_table, cluster) != 1) { 2068 fprintf(stderr, "ERROR refcount block %" PRId64 2069 " refcount=%" PRIu64 "\n", i, 2070 s->get_refcount(*refcount_table, cluster)); 2071 res->corruptions++; 2072 *rebuild = true; 2073 } 2074 } 2075 } 2076 2077 return 0; 2078 } 2079 2080 /* 2081 * Calculates an in-memory refcount table. 2082 */ 2083 static int calculate_refcounts(BlockDriverState *bs, BdrvCheckResult *res, 2084 BdrvCheckMode fix, bool *rebuild, 2085 void **refcount_table, int64_t *nb_clusters) 2086 { 2087 BDRVQcow2State *s = bs->opaque; 2088 int64_t i; 2089 QCowSnapshot *sn; 2090 int ret; 2091 2092 if (!*refcount_table) { 2093 int64_t old_size = 0; 2094 ret = realloc_refcount_array(s, refcount_table, 2095 &old_size, *nb_clusters); 2096 if (ret < 0) { 2097 res->check_errors++; 2098 return ret; 2099 } 2100 } 2101 2102 /* header */ 2103 ret = qcow2_inc_refcounts_imrt(bs, res, refcount_table, nb_clusters, 2104 0, s->cluster_size); 2105 if (ret < 0) { 2106 return ret; 2107 } 2108 2109 /* current L1 table */ 2110 ret = check_refcounts_l1(bs, res, refcount_table, nb_clusters, 2111 s->l1_table_offset, s->l1_size, CHECK_FRAG_INFO, 2112 fix, true); 2113 if (ret < 0) { 2114 return ret; 2115 } 2116 2117 /* snapshots */ 2118 if (has_data_file(bs) && s->nb_snapshots) { 2119 fprintf(stderr, "ERROR %d snapshots in image with data file\n", 2120 s->nb_snapshots); 2121 res->corruptions++; 2122 } 2123 2124 for (i = 0; i < s->nb_snapshots; i++) { 2125 sn = s->snapshots + i; 2126 if (offset_into_cluster(s, sn->l1_table_offset)) { 2127 fprintf(stderr, "ERROR snapshot %s (%s) l1_offset=%#" PRIx64 ": " 2128 "L1 table is not cluster aligned; snapshot table entry " 2129 "corrupted\n", sn->id_str, sn->name, sn->l1_table_offset); 2130 res->corruptions++; 2131 continue; 2132 } 2133 if (sn->l1_size > QCOW_MAX_L1_SIZE / sizeof(uint64_t)) { 2134 fprintf(stderr, "ERROR snapshot %s (%s) l1_size=%#" PRIx32 ": " 2135 "L1 table is too large; snapshot table entry corrupted\n", 2136 sn->id_str, sn->name, sn->l1_size); 2137 res->corruptions++; 2138 continue; 2139 } 2140 ret = check_refcounts_l1(bs, res, refcount_table, nb_clusters, 2141 sn->l1_table_offset, sn->l1_size, 0, fix, 2142 false); 2143 if (ret < 0) { 2144 return ret; 2145 } 2146 } 2147 ret = qcow2_inc_refcounts_imrt(bs, res, refcount_table, nb_clusters, 2148 s->snapshots_offset, s->snapshots_size); 2149 if (ret < 0) { 2150 return ret; 2151 } 2152 2153 /* refcount data */ 2154 ret = qcow2_inc_refcounts_imrt(bs, res, refcount_table, nb_clusters, 2155 s->refcount_table_offset, 2156 s->refcount_table_size * sizeof(uint64_t)); 2157 if (ret < 0) { 2158 return ret; 2159 } 2160 2161 /* encryption */ 2162 if (s->crypto_header.length) { 2163 ret = qcow2_inc_refcounts_imrt(bs, res, refcount_table, nb_clusters, 2164 s->crypto_header.offset, 2165 s->crypto_header.length); 2166 if (ret < 0) { 2167 return ret; 2168 } 2169 } 2170 2171 /* bitmaps */ 2172 ret = qcow2_check_bitmaps_refcounts(bs, res, refcount_table, nb_clusters); 2173 if (ret < 0) { 2174 return ret; 2175 } 2176 2177 return check_refblocks(bs, res, fix, rebuild, refcount_table, nb_clusters); 2178 } 2179 2180 /* 2181 * Compares the actual reference count for each cluster in the image against the 2182 * refcount as reported by the refcount structures on-disk. 2183 */ 2184 static void compare_refcounts(BlockDriverState *bs, BdrvCheckResult *res, 2185 BdrvCheckMode fix, bool *rebuild, 2186 int64_t *highest_cluster, 2187 void *refcount_table, int64_t nb_clusters) 2188 { 2189 BDRVQcow2State *s = bs->opaque; 2190 int64_t i; 2191 uint64_t refcount1, refcount2; 2192 int ret; 2193 2194 for (i = 0, *highest_cluster = 0; i < nb_clusters; i++) { 2195 ret = qcow2_get_refcount(bs, i, &refcount1); 2196 if (ret < 0) { 2197 fprintf(stderr, "Can't get refcount for cluster %" PRId64 ": %s\n", 2198 i, strerror(-ret)); 2199 res->check_errors++; 2200 continue; 2201 } 2202 2203 refcount2 = s->get_refcount(refcount_table, i); 2204 2205 if (refcount1 > 0 || refcount2 > 0) { 2206 *highest_cluster = i; 2207 } 2208 2209 if (refcount1 != refcount2) { 2210 /* Check if we're allowed to fix the mismatch */ 2211 int *num_fixed = NULL; 2212 if (refcount1 == 0) { 2213 *rebuild = true; 2214 } else if (refcount1 > refcount2 && (fix & BDRV_FIX_LEAKS)) { 2215 num_fixed = &res->leaks_fixed; 2216 } else if (refcount1 < refcount2 && (fix & BDRV_FIX_ERRORS)) { 2217 num_fixed = &res->corruptions_fixed; 2218 } 2219 2220 fprintf(stderr, "%s cluster %" PRId64 " refcount=%" PRIu64 2221 " reference=%" PRIu64 "\n", 2222 num_fixed != NULL ? "Repairing" : 2223 refcount1 < refcount2 ? "ERROR" : 2224 "Leaked", 2225 i, refcount1, refcount2); 2226 2227 if (num_fixed) { 2228 ret = update_refcount(bs, i << s->cluster_bits, 1, 2229 refcount_diff(refcount1, refcount2), 2230 refcount1 > refcount2, 2231 QCOW2_DISCARD_ALWAYS); 2232 if (ret >= 0) { 2233 (*num_fixed)++; 2234 continue; 2235 } 2236 } 2237 2238 /* And if we couldn't, print an error */ 2239 if (refcount1 < refcount2) { 2240 res->corruptions++; 2241 } else { 2242 res->leaks++; 2243 } 2244 } 2245 } 2246 } 2247 2248 /* 2249 * Allocates clusters using an in-memory refcount table (IMRT) in contrast to 2250 * the on-disk refcount structures. 2251 * 2252 * On input, *first_free_cluster tells where to start looking, and need not 2253 * actually be a free cluster; the returned offset will not be before that 2254 * cluster. On output, *first_free_cluster points to the first gap found, even 2255 * if that gap was too small to be used as the returned offset. 2256 * 2257 * Note that *first_free_cluster is a cluster index whereas the return value is 2258 * an offset. 2259 */ 2260 static int64_t alloc_clusters_imrt(BlockDriverState *bs, 2261 int cluster_count, 2262 void **refcount_table, 2263 int64_t *imrt_nb_clusters, 2264 int64_t *first_free_cluster) 2265 { 2266 BDRVQcow2State *s = bs->opaque; 2267 int64_t cluster = *first_free_cluster, i; 2268 bool first_gap = true; 2269 int contiguous_free_clusters; 2270 int ret; 2271 2272 /* Starting at *first_free_cluster, find a range of at least cluster_count 2273 * continuously free clusters */ 2274 for (contiguous_free_clusters = 0; 2275 cluster < *imrt_nb_clusters && 2276 contiguous_free_clusters < cluster_count; 2277 cluster++) 2278 { 2279 if (!s->get_refcount(*refcount_table, cluster)) { 2280 contiguous_free_clusters++; 2281 if (first_gap) { 2282 /* If this is the first free cluster found, update 2283 * *first_free_cluster accordingly */ 2284 *first_free_cluster = cluster; 2285 first_gap = false; 2286 } 2287 } else if (contiguous_free_clusters) { 2288 contiguous_free_clusters = 0; 2289 } 2290 } 2291 2292 /* If contiguous_free_clusters is greater than zero, it contains the number 2293 * of continuously free clusters until the current cluster; the first free 2294 * cluster in the current "gap" is therefore 2295 * cluster - contiguous_free_clusters */ 2296 2297 /* If no such range could be found, grow the in-memory refcount table 2298 * accordingly to append free clusters at the end of the image */ 2299 if (contiguous_free_clusters < cluster_count) { 2300 /* contiguous_free_clusters clusters are already empty at the image end; 2301 * we need cluster_count clusters; therefore, we have to allocate 2302 * cluster_count - contiguous_free_clusters new clusters at the end of 2303 * the image (which is the current value of cluster; note that cluster 2304 * may exceed old_imrt_nb_clusters if *first_free_cluster pointed beyond 2305 * the image end) */ 2306 ret = realloc_refcount_array(s, refcount_table, imrt_nb_clusters, 2307 cluster + cluster_count 2308 - contiguous_free_clusters); 2309 if (ret < 0) { 2310 return ret; 2311 } 2312 } 2313 2314 /* Go back to the first free cluster */ 2315 cluster -= contiguous_free_clusters; 2316 for (i = 0; i < cluster_count; i++) { 2317 s->set_refcount(*refcount_table, cluster + i, 1); 2318 } 2319 2320 return cluster << s->cluster_bits; 2321 } 2322 2323 /* 2324 * Creates a new refcount structure based solely on the in-memory information 2325 * given through *refcount_table. All necessary allocations will be reflected 2326 * in that array. 2327 * 2328 * On success, the old refcount structure is leaked (it will be covered by the 2329 * new refcount structure). 2330 */ 2331 static int rebuild_refcount_structure(BlockDriverState *bs, 2332 BdrvCheckResult *res, 2333 void **refcount_table, 2334 int64_t *nb_clusters) 2335 { 2336 BDRVQcow2State *s = bs->opaque; 2337 int64_t first_free_cluster = 0, reftable_offset = -1, cluster = 0; 2338 int64_t refblock_offset, refblock_start, refblock_index; 2339 uint32_t reftable_size = 0; 2340 uint64_t *on_disk_reftable = NULL; 2341 void *on_disk_refblock; 2342 int ret = 0; 2343 struct { 2344 uint64_t reftable_offset; 2345 uint32_t reftable_clusters; 2346 } QEMU_PACKED reftable_offset_and_clusters; 2347 2348 qcow2_cache_empty(bs, s->refcount_block_cache); 2349 2350 write_refblocks: 2351 for (; cluster < *nb_clusters; cluster++) { 2352 if (!s->get_refcount(*refcount_table, cluster)) { 2353 continue; 2354 } 2355 2356 refblock_index = cluster >> s->refcount_block_bits; 2357 refblock_start = refblock_index << s->refcount_block_bits; 2358 2359 /* Don't allocate a cluster in a refblock already written to disk */ 2360 if (first_free_cluster < refblock_start) { 2361 first_free_cluster = refblock_start; 2362 } 2363 refblock_offset = alloc_clusters_imrt(bs, 1, refcount_table, 2364 nb_clusters, &first_free_cluster); 2365 if (refblock_offset < 0) { 2366 fprintf(stderr, "ERROR allocating refblock: %s\n", 2367 strerror(-refblock_offset)); 2368 res->check_errors++; 2369 ret = refblock_offset; 2370 goto fail; 2371 } 2372 2373 if (reftable_size <= refblock_index) { 2374 uint32_t old_reftable_size = reftable_size; 2375 uint64_t *new_on_disk_reftable; 2376 2377 reftable_size = ROUND_UP((refblock_index + 1) * sizeof(uint64_t), 2378 s->cluster_size) / sizeof(uint64_t); 2379 new_on_disk_reftable = g_try_realloc(on_disk_reftable, 2380 reftable_size * 2381 sizeof(uint64_t)); 2382 if (!new_on_disk_reftable) { 2383 res->check_errors++; 2384 ret = -ENOMEM; 2385 goto fail; 2386 } 2387 on_disk_reftable = new_on_disk_reftable; 2388 2389 memset(on_disk_reftable + old_reftable_size, 0, 2390 (reftable_size - old_reftable_size) * sizeof(uint64_t)); 2391 2392 /* The offset we have for the reftable is now no longer valid; 2393 * this will leak that range, but we can easily fix that by running 2394 * a leak-fixing check after this rebuild operation */ 2395 reftable_offset = -1; 2396 } else { 2397 assert(on_disk_reftable); 2398 } 2399 on_disk_reftable[refblock_index] = refblock_offset; 2400 2401 /* If this is apparently the last refblock (for now), try to squeeze the 2402 * reftable in */ 2403 if (refblock_index == (*nb_clusters - 1) >> s->refcount_block_bits && 2404 reftable_offset < 0) 2405 { 2406 uint64_t reftable_clusters = size_to_clusters(s, reftable_size * 2407 sizeof(uint64_t)); 2408 reftable_offset = alloc_clusters_imrt(bs, reftable_clusters, 2409 refcount_table, nb_clusters, 2410 &first_free_cluster); 2411 if (reftable_offset < 0) { 2412 fprintf(stderr, "ERROR allocating reftable: %s\n", 2413 strerror(-reftable_offset)); 2414 res->check_errors++; 2415 ret = reftable_offset; 2416 goto fail; 2417 } 2418 } 2419 2420 ret = qcow2_pre_write_overlap_check(bs, 0, refblock_offset, 2421 s->cluster_size, false); 2422 if (ret < 0) { 2423 fprintf(stderr, "ERROR writing refblock: %s\n", strerror(-ret)); 2424 goto fail; 2425 } 2426 2427 /* The size of *refcount_table is always cluster-aligned, therefore the 2428 * write operation will not overflow */ 2429 on_disk_refblock = (void *)((char *) *refcount_table + 2430 refblock_index * s->cluster_size); 2431 2432 ret = bdrv_pwrite(bs->file, refblock_offset, on_disk_refblock, 2433 s->cluster_size); 2434 if (ret < 0) { 2435 fprintf(stderr, "ERROR writing refblock: %s\n", strerror(-ret)); 2436 goto fail; 2437 } 2438 2439 /* Go to the end of this refblock */ 2440 cluster = refblock_start + s->refcount_block_size - 1; 2441 } 2442 2443 if (reftable_offset < 0) { 2444 uint64_t post_refblock_start, reftable_clusters; 2445 2446 post_refblock_start = ROUND_UP(*nb_clusters, s->refcount_block_size); 2447 reftable_clusters = size_to_clusters(s, 2448 reftable_size * sizeof(uint64_t)); 2449 /* Not pretty but simple */ 2450 if (first_free_cluster < post_refblock_start) { 2451 first_free_cluster = post_refblock_start; 2452 } 2453 reftable_offset = alloc_clusters_imrt(bs, reftable_clusters, 2454 refcount_table, nb_clusters, 2455 &first_free_cluster); 2456 if (reftable_offset < 0) { 2457 fprintf(stderr, "ERROR allocating reftable: %s\n", 2458 strerror(-reftable_offset)); 2459 res->check_errors++; 2460 ret = reftable_offset; 2461 goto fail; 2462 } 2463 2464 goto write_refblocks; 2465 } 2466 2467 for (refblock_index = 0; refblock_index < reftable_size; refblock_index++) { 2468 cpu_to_be64s(&on_disk_reftable[refblock_index]); 2469 } 2470 2471 ret = qcow2_pre_write_overlap_check(bs, 0, reftable_offset, 2472 reftable_size * sizeof(uint64_t), 2473 false); 2474 if (ret < 0) { 2475 fprintf(stderr, "ERROR writing reftable: %s\n", strerror(-ret)); 2476 goto fail; 2477 } 2478 2479 assert(reftable_size < INT_MAX / sizeof(uint64_t)); 2480 ret = bdrv_pwrite(bs->file, reftable_offset, on_disk_reftable, 2481 reftable_size * sizeof(uint64_t)); 2482 if (ret < 0) { 2483 fprintf(stderr, "ERROR writing reftable: %s\n", strerror(-ret)); 2484 goto fail; 2485 } 2486 2487 /* Enter new reftable into the image header */ 2488 reftable_offset_and_clusters.reftable_offset = cpu_to_be64(reftable_offset); 2489 reftable_offset_and_clusters.reftable_clusters = 2490 cpu_to_be32(size_to_clusters(s, reftable_size * sizeof(uint64_t))); 2491 ret = bdrv_pwrite_sync(bs->file, 2492 offsetof(QCowHeader, refcount_table_offset), 2493 &reftable_offset_and_clusters, 2494 sizeof(reftable_offset_and_clusters)); 2495 if (ret < 0) { 2496 fprintf(stderr, "ERROR setting reftable: %s\n", strerror(-ret)); 2497 goto fail; 2498 } 2499 2500 for (refblock_index = 0; refblock_index < reftable_size; refblock_index++) { 2501 be64_to_cpus(&on_disk_reftable[refblock_index]); 2502 } 2503 s->refcount_table = on_disk_reftable; 2504 s->refcount_table_offset = reftable_offset; 2505 s->refcount_table_size = reftable_size; 2506 update_max_refcount_table_index(s); 2507 2508 return 0; 2509 2510 fail: 2511 g_free(on_disk_reftable); 2512 return ret; 2513 } 2514 2515 /* 2516 * Checks an image for refcount consistency. 2517 * 2518 * Returns 0 if no errors are found, the number of errors in case the image is 2519 * detected as corrupted, and -errno when an internal error occurred. 2520 */ 2521 int qcow2_check_refcounts(BlockDriverState *bs, BdrvCheckResult *res, 2522 BdrvCheckMode fix) 2523 { 2524 BDRVQcow2State *s = bs->opaque; 2525 BdrvCheckResult pre_compare_res; 2526 int64_t size, highest_cluster, nb_clusters; 2527 void *refcount_table = NULL; 2528 bool rebuild = false; 2529 int ret; 2530 2531 size = bdrv_getlength(bs->file->bs); 2532 if (size < 0) { 2533 res->check_errors++; 2534 return size; 2535 } 2536 2537 nb_clusters = size_to_clusters(s, size); 2538 if (nb_clusters > INT_MAX) { 2539 res->check_errors++; 2540 return -EFBIG; 2541 } 2542 2543 res->bfi.total_clusters = 2544 size_to_clusters(s, bs->total_sectors * BDRV_SECTOR_SIZE); 2545 2546 ret = calculate_refcounts(bs, res, fix, &rebuild, &refcount_table, 2547 &nb_clusters); 2548 if (ret < 0) { 2549 goto fail; 2550 } 2551 2552 /* In case we don't need to rebuild the refcount structure (but want to fix 2553 * something), this function is immediately called again, in which case the 2554 * result should be ignored */ 2555 pre_compare_res = *res; 2556 compare_refcounts(bs, res, 0, &rebuild, &highest_cluster, refcount_table, 2557 nb_clusters); 2558 2559 if (rebuild && (fix & BDRV_FIX_ERRORS)) { 2560 BdrvCheckResult old_res = *res; 2561 int fresh_leaks = 0; 2562 2563 fprintf(stderr, "Rebuilding refcount structure\n"); 2564 ret = rebuild_refcount_structure(bs, res, &refcount_table, 2565 &nb_clusters); 2566 if (ret < 0) { 2567 goto fail; 2568 } 2569 2570 res->corruptions = 0; 2571 res->leaks = 0; 2572 2573 /* Because the old reftable has been exchanged for a new one the 2574 * references have to be recalculated */ 2575 rebuild = false; 2576 memset(refcount_table, 0, refcount_array_byte_size(s, nb_clusters)); 2577 ret = calculate_refcounts(bs, res, 0, &rebuild, &refcount_table, 2578 &nb_clusters); 2579 if (ret < 0) { 2580 goto fail; 2581 } 2582 2583 if (fix & BDRV_FIX_LEAKS) { 2584 /* The old refcount structures are now leaked, fix it; the result 2585 * can be ignored, aside from leaks which were introduced by 2586 * rebuild_refcount_structure() that could not be fixed */ 2587 BdrvCheckResult saved_res = *res; 2588 *res = (BdrvCheckResult){ 0 }; 2589 2590 compare_refcounts(bs, res, BDRV_FIX_LEAKS, &rebuild, 2591 &highest_cluster, refcount_table, nb_clusters); 2592 if (rebuild) { 2593 fprintf(stderr, "ERROR rebuilt refcount structure is still " 2594 "broken\n"); 2595 } 2596 2597 /* Any leaks accounted for here were introduced by 2598 * rebuild_refcount_structure() because that function has created a 2599 * new refcount structure from scratch */ 2600 fresh_leaks = res->leaks; 2601 *res = saved_res; 2602 } 2603 2604 if (res->corruptions < old_res.corruptions) { 2605 res->corruptions_fixed += old_res.corruptions - res->corruptions; 2606 } 2607 if (res->leaks < old_res.leaks) { 2608 res->leaks_fixed += old_res.leaks - res->leaks; 2609 } 2610 res->leaks += fresh_leaks; 2611 } else if (fix) { 2612 if (rebuild) { 2613 fprintf(stderr, "ERROR need to rebuild refcount structures\n"); 2614 res->check_errors++; 2615 ret = -EIO; 2616 goto fail; 2617 } 2618 2619 if (res->leaks || res->corruptions) { 2620 *res = pre_compare_res; 2621 compare_refcounts(bs, res, fix, &rebuild, &highest_cluster, 2622 refcount_table, nb_clusters); 2623 } 2624 } 2625 2626 /* check OFLAG_COPIED */ 2627 ret = check_oflag_copied(bs, res, fix); 2628 if (ret < 0) { 2629 goto fail; 2630 } 2631 2632 res->image_end_offset = (highest_cluster + 1) * s->cluster_size; 2633 ret = 0; 2634 2635 fail: 2636 g_free(refcount_table); 2637 2638 return ret; 2639 } 2640 2641 #define overlaps_with(ofs, sz) \ 2642 ranges_overlap(offset, size, ofs, sz) 2643 2644 /* 2645 * Checks if the given offset into the image file is actually free to use by 2646 * looking for overlaps with important metadata sections (L1/L2 tables etc.), 2647 * i.e. a sanity check without relying on the refcount tables. 2648 * 2649 * The ign parameter specifies what checks not to perform (being a bitmask of 2650 * QCow2MetadataOverlap values), i.e., what sections to ignore. 2651 * 2652 * Returns: 2653 * - 0 if writing to this offset will not affect the mentioned metadata 2654 * - a positive QCow2MetadataOverlap value indicating one overlapping section 2655 * - a negative value (-errno) indicating an error while performing a check, 2656 * e.g. when bdrv_read failed on QCOW2_OL_INACTIVE_L2 2657 */ 2658 int qcow2_check_metadata_overlap(BlockDriverState *bs, int ign, int64_t offset, 2659 int64_t size) 2660 { 2661 BDRVQcow2State *s = bs->opaque; 2662 int chk = s->overlap_check & ~ign; 2663 int i, j; 2664 2665 if (!size) { 2666 return 0; 2667 } 2668 2669 if (chk & QCOW2_OL_MAIN_HEADER) { 2670 if (offset < s->cluster_size) { 2671 return QCOW2_OL_MAIN_HEADER; 2672 } 2673 } 2674 2675 /* align range to test to cluster boundaries */ 2676 size = ROUND_UP(offset_into_cluster(s, offset) + size, s->cluster_size); 2677 offset = start_of_cluster(s, offset); 2678 2679 if ((chk & QCOW2_OL_ACTIVE_L1) && s->l1_size) { 2680 if (overlaps_with(s->l1_table_offset, s->l1_size * sizeof(uint64_t))) { 2681 return QCOW2_OL_ACTIVE_L1; 2682 } 2683 } 2684 2685 if ((chk & QCOW2_OL_REFCOUNT_TABLE) && s->refcount_table_size) { 2686 if (overlaps_with(s->refcount_table_offset, 2687 s->refcount_table_size * sizeof(uint64_t))) { 2688 return QCOW2_OL_REFCOUNT_TABLE; 2689 } 2690 } 2691 2692 if ((chk & QCOW2_OL_SNAPSHOT_TABLE) && s->snapshots_size) { 2693 if (overlaps_with(s->snapshots_offset, s->snapshots_size)) { 2694 return QCOW2_OL_SNAPSHOT_TABLE; 2695 } 2696 } 2697 2698 if ((chk & QCOW2_OL_INACTIVE_L1) && s->snapshots) { 2699 for (i = 0; i < s->nb_snapshots; i++) { 2700 if (s->snapshots[i].l1_size && 2701 overlaps_with(s->snapshots[i].l1_table_offset, 2702 s->snapshots[i].l1_size * sizeof(uint64_t))) { 2703 return QCOW2_OL_INACTIVE_L1; 2704 } 2705 } 2706 } 2707 2708 if ((chk & QCOW2_OL_ACTIVE_L2) && s->l1_table) { 2709 for (i = 0; i < s->l1_size; i++) { 2710 if ((s->l1_table[i] & L1E_OFFSET_MASK) && 2711 overlaps_with(s->l1_table[i] & L1E_OFFSET_MASK, 2712 s->cluster_size)) { 2713 return QCOW2_OL_ACTIVE_L2; 2714 } 2715 } 2716 } 2717 2718 if ((chk & QCOW2_OL_REFCOUNT_BLOCK) && s->refcount_table) { 2719 unsigned last_entry = s->max_refcount_table_index; 2720 assert(last_entry < s->refcount_table_size); 2721 assert(last_entry + 1 == s->refcount_table_size || 2722 (s->refcount_table[last_entry + 1] & REFT_OFFSET_MASK) == 0); 2723 for (i = 0; i <= last_entry; i++) { 2724 if ((s->refcount_table[i] & REFT_OFFSET_MASK) && 2725 overlaps_with(s->refcount_table[i] & REFT_OFFSET_MASK, 2726 s->cluster_size)) { 2727 return QCOW2_OL_REFCOUNT_BLOCK; 2728 } 2729 } 2730 } 2731 2732 if ((chk & QCOW2_OL_INACTIVE_L2) && s->snapshots) { 2733 for (i = 0; i < s->nb_snapshots; i++) { 2734 uint64_t l1_ofs = s->snapshots[i].l1_table_offset; 2735 uint32_t l1_sz = s->snapshots[i].l1_size; 2736 uint64_t l1_sz2 = l1_sz * sizeof(uint64_t); 2737 uint64_t *l1; 2738 int ret; 2739 2740 ret = qcow2_validate_table(bs, l1_ofs, l1_sz, sizeof(uint64_t), 2741 QCOW_MAX_L1_SIZE, "", NULL); 2742 if (ret < 0) { 2743 return ret; 2744 } 2745 2746 l1 = g_try_malloc(l1_sz2); 2747 2748 if (l1_sz2 && l1 == NULL) { 2749 return -ENOMEM; 2750 } 2751 2752 ret = bdrv_pread(bs->file, l1_ofs, l1, l1_sz2); 2753 if (ret < 0) { 2754 g_free(l1); 2755 return ret; 2756 } 2757 2758 for (j = 0; j < l1_sz; j++) { 2759 uint64_t l2_ofs = be64_to_cpu(l1[j]) & L1E_OFFSET_MASK; 2760 if (l2_ofs && overlaps_with(l2_ofs, s->cluster_size)) { 2761 g_free(l1); 2762 return QCOW2_OL_INACTIVE_L2; 2763 } 2764 } 2765 2766 g_free(l1); 2767 } 2768 } 2769 2770 if ((chk & QCOW2_OL_BITMAP_DIRECTORY) && 2771 (s->autoclear_features & QCOW2_AUTOCLEAR_BITMAPS)) 2772 { 2773 if (overlaps_with(s->bitmap_directory_offset, 2774 s->bitmap_directory_size)) 2775 { 2776 return QCOW2_OL_BITMAP_DIRECTORY; 2777 } 2778 } 2779 2780 return 0; 2781 } 2782 2783 static const char *metadata_ol_names[] = { 2784 [QCOW2_OL_MAIN_HEADER_BITNR] = "qcow2_header", 2785 [QCOW2_OL_ACTIVE_L1_BITNR] = "active L1 table", 2786 [QCOW2_OL_ACTIVE_L2_BITNR] = "active L2 table", 2787 [QCOW2_OL_REFCOUNT_TABLE_BITNR] = "refcount table", 2788 [QCOW2_OL_REFCOUNT_BLOCK_BITNR] = "refcount block", 2789 [QCOW2_OL_SNAPSHOT_TABLE_BITNR] = "snapshot table", 2790 [QCOW2_OL_INACTIVE_L1_BITNR] = "inactive L1 table", 2791 [QCOW2_OL_INACTIVE_L2_BITNR] = "inactive L2 table", 2792 [QCOW2_OL_BITMAP_DIRECTORY_BITNR] = "bitmap directory", 2793 }; 2794 QEMU_BUILD_BUG_ON(QCOW2_OL_MAX_BITNR != ARRAY_SIZE(metadata_ol_names)); 2795 2796 /* 2797 * First performs a check for metadata overlaps (through 2798 * qcow2_check_metadata_overlap); if that fails with a negative value (error 2799 * while performing a check), that value is returned. If an impending overlap 2800 * is detected, the BDS will be made unusable, the qcow2 file marked corrupt 2801 * and -EIO returned. 2802 * 2803 * Returns 0 if there were neither overlaps nor errors while checking for 2804 * overlaps; or a negative value (-errno) on error. 2805 */ 2806 int qcow2_pre_write_overlap_check(BlockDriverState *bs, int ign, int64_t offset, 2807 int64_t size, bool data_file) 2808 { 2809 int ret; 2810 2811 if (data_file && has_data_file(bs)) { 2812 return 0; 2813 } 2814 2815 ret = qcow2_check_metadata_overlap(bs, ign, offset, size); 2816 if (ret < 0) { 2817 return ret; 2818 } else if (ret > 0) { 2819 int metadata_ol_bitnr = ctz32(ret); 2820 assert(metadata_ol_bitnr < QCOW2_OL_MAX_BITNR); 2821 2822 qcow2_signal_corruption(bs, true, offset, size, "Preventing invalid " 2823 "write on metadata (overlaps with %s)", 2824 metadata_ol_names[metadata_ol_bitnr]); 2825 return -EIO; 2826 } 2827 2828 return 0; 2829 } 2830 2831 /* A pointer to a function of this type is given to walk_over_reftable(). That 2832 * function will create refblocks and pass them to a RefblockFinishOp once they 2833 * are completed (@refblock). @refblock_empty is set if the refblock is 2834 * completely empty. 2835 * 2836 * Along with the refblock, a corresponding reftable entry is passed, in the 2837 * reftable @reftable (which may be reallocated) at @reftable_index. 2838 * 2839 * @allocated should be set to true if a new cluster has been allocated. 2840 */ 2841 typedef int (RefblockFinishOp)(BlockDriverState *bs, uint64_t **reftable, 2842 uint64_t reftable_index, uint64_t *reftable_size, 2843 void *refblock, bool refblock_empty, 2844 bool *allocated, Error **errp); 2845 2846 /** 2847 * This "operation" for walk_over_reftable() allocates the refblock on disk (if 2848 * it is not empty) and inserts its offset into the new reftable. The size of 2849 * this new reftable is increased as required. 2850 */ 2851 static int alloc_refblock(BlockDriverState *bs, uint64_t **reftable, 2852 uint64_t reftable_index, uint64_t *reftable_size, 2853 void *refblock, bool refblock_empty, bool *allocated, 2854 Error **errp) 2855 { 2856 BDRVQcow2State *s = bs->opaque; 2857 int64_t offset; 2858 2859 if (!refblock_empty && reftable_index >= *reftable_size) { 2860 uint64_t *new_reftable; 2861 uint64_t new_reftable_size; 2862 2863 new_reftable_size = ROUND_UP(reftable_index + 1, 2864 s->cluster_size / sizeof(uint64_t)); 2865 if (new_reftable_size > QCOW_MAX_REFTABLE_SIZE / sizeof(uint64_t)) { 2866 error_setg(errp, 2867 "This operation would make the refcount table grow " 2868 "beyond the maximum size supported by QEMU, aborting"); 2869 return -ENOTSUP; 2870 } 2871 2872 new_reftable = g_try_realloc(*reftable, new_reftable_size * 2873 sizeof(uint64_t)); 2874 if (!new_reftable) { 2875 error_setg(errp, "Failed to increase reftable buffer size"); 2876 return -ENOMEM; 2877 } 2878 2879 memset(new_reftable + *reftable_size, 0, 2880 (new_reftable_size - *reftable_size) * sizeof(uint64_t)); 2881 2882 *reftable = new_reftable; 2883 *reftable_size = new_reftable_size; 2884 } 2885 2886 if (!refblock_empty && !(*reftable)[reftable_index]) { 2887 offset = qcow2_alloc_clusters(bs, s->cluster_size); 2888 if (offset < 0) { 2889 error_setg_errno(errp, -offset, "Failed to allocate refblock"); 2890 return offset; 2891 } 2892 (*reftable)[reftable_index] = offset; 2893 *allocated = true; 2894 } 2895 2896 return 0; 2897 } 2898 2899 /** 2900 * This "operation" for walk_over_reftable() writes the refblock to disk at the 2901 * offset specified by the new reftable's entry. It does not modify the new 2902 * reftable or change any refcounts. 2903 */ 2904 static int flush_refblock(BlockDriverState *bs, uint64_t **reftable, 2905 uint64_t reftable_index, uint64_t *reftable_size, 2906 void *refblock, bool refblock_empty, bool *allocated, 2907 Error **errp) 2908 { 2909 BDRVQcow2State *s = bs->opaque; 2910 int64_t offset; 2911 int ret; 2912 2913 if (reftable_index < *reftable_size && (*reftable)[reftable_index]) { 2914 offset = (*reftable)[reftable_index]; 2915 2916 ret = qcow2_pre_write_overlap_check(bs, 0, offset, s->cluster_size, 2917 false); 2918 if (ret < 0) { 2919 error_setg_errno(errp, -ret, "Overlap check failed"); 2920 return ret; 2921 } 2922 2923 ret = bdrv_pwrite(bs->file, offset, refblock, s->cluster_size); 2924 if (ret < 0) { 2925 error_setg_errno(errp, -ret, "Failed to write refblock"); 2926 return ret; 2927 } 2928 } else { 2929 assert(refblock_empty); 2930 } 2931 2932 return 0; 2933 } 2934 2935 /** 2936 * This function walks over the existing reftable and every referenced refblock; 2937 * if @new_set_refcount is non-NULL, it is called for every refcount entry to 2938 * create an equal new entry in the passed @new_refblock. Once that 2939 * @new_refblock is completely filled, @operation will be called. 2940 * 2941 * @status_cb and @cb_opaque are used for the amend operation's status callback. 2942 * @index is the index of the walk_over_reftable() calls and @total is the total 2943 * number of walk_over_reftable() calls per amend operation. Both are used for 2944 * calculating the parameters for the status callback. 2945 * 2946 * @allocated is set to true if a new cluster has been allocated. 2947 */ 2948 static int walk_over_reftable(BlockDriverState *bs, uint64_t **new_reftable, 2949 uint64_t *new_reftable_index, 2950 uint64_t *new_reftable_size, 2951 void *new_refblock, int new_refblock_size, 2952 int new_refcount_bits, 2953 RefblockFinishOp *operation, bool *allocated, 2954 Qcow2SetRefcountFunc *new_set_refcount, 2955 BlockDriverAmendStatusCB *status_cb, 2956 void *cb_opaque, int index, int total, 2957 Error **errp) 2958 { 2959 BDRVQcow2State *s = bs->opaque; 2960 uint64_t reftable_index; 2961 bool new_refblock_empty = true; 2962 int refblock_index; 2963 int new_refblock_index = 0; 2964 int ret; 2965 2966 for (reftable_index = 0; reftable_index < s->refcount_table_size; 2967 reftable_index++) 2968 { 2969 uint64_t refblock_offset = s->refcount_table[reftable_index] 2970 & REFT_OFFSET_MASK; 2971 2972 status_cb(bs, (uint64_t)index * s->refcount_table_size + reftable_index, 2973 (uint64_t)total * s->refcount_table_size, cb_opaque); 2974 2975 if (refblock_offset) { 2976 void *refblock; 2977 2978 if (offset_into_cluster(s, refblock_offset)) { 2979 qcow2_signal_corruption(bs, true, -1, -1, "Refblock offset %#" 2980 PRIx64 " unaligned (reftable index: %#" 2981 PRIx64 ")", refblock_offset, 2982 reftable_index); 2983 error_setg(errp, 2984 "Image is corrupt (unaligned refblock offset)"); 2985 return -EIO; 2986 } 2987 2988 ret = qcow2_cache_get(bs, s->refcount_block_cache, refblock_offset, 2989 &refblock); 2990 if (ret < 0) { 2991 error_setg_errno(errp, -ret, "Failed to retrieve refblock"); 2992 return ret; 2993 } 2994 2995 for (refblock_index = 0; refblock_index < s->refcount_block_size; 2996 refblock_index++) 2997 { 2998 uint64_t refcount; 2999 3000 if (new_refblock_index >= new_refblock_size) { 3001 /* new_refblock is now complete */ 3002 ret = operation(bs, new_reftable, *new_reftable_index, 3003 new_reftable_size, new_refblock, 3004 new_refblock_empty, allocated, errp); 3005 if (ret < 0) { 3006 qcow2_cache_put(s->refcount_block_cache, &refblock); 3007 return ret; 3008 } 3009 3010 (*new_reftable_index)++; 3011 new_refblock_index = 0; 3012 new_refblock_empty = true; 3013 } 3014 3015 refcount = s->get_refcount(refblock, refblock_index); 3016 if (new_refcount_bits < 64 && refcount >> new_refcount_bits) { 3017 uint64_t offset; 3018 3019 qcow2_cache_put(s->refcount_block_cache, &refblock); 3020 3021 offset = ((reftable_index << s->refcount_block_bits) 3022 + refblock_index) << s->cluster_bits; 3023 3024 error_setg(errp, "Cannot decrease refcount entry width to " 3025 "%i bits: Cluster at offset %#" PRIx64 " has a " 3026 "refcount of %" PRIu64, new_refcount_bits, 3027 offset, refcount); 3028 return -EINVAL; 3029 } 3030 3031 if (new_set_refcount) { 3032 new_set_refcount(new_refblock, new_refblock_index++, 3033 refcount); 3034 } else { 3035 new_refblock_index++; 3036 } 3037 new_refblock_empty = new_refblock_empty && refcount == 0; 3038 } 3039 3040 qcow2_cache_put(s->refcount_block_cache, &refblock); 3041 } else { 3042 /* No refblock means every refcount is 0 */ 3043 for (refblock_index = 0; refblock_index < s->refcount_block_size; 3044 refblock_index++) 3045 { 3046 if (new_refblock_index >= new_refblock_size) { 3047 /* new_refblock is now complete */ 3048 ret = operation(bs, new_reftable, *new_reftable_index, 3049 new_reftable_size, new_refblock, 3050 new_refblock_empty, allocated, errp); 3051 if (ret < 0) { 3052 return ret; 3053 } 3054 3055 (*new_reftable_index)++; 3056 new_refblock_index = 0; 3057 new_refblock_empty = true; 3058 } 3059 3060 if (new_set_refcount) { 3061 new_set_refcount(new_refblock, new_refblock_index++, 0); 3062 } else { 3063 new_refblock_index++; 3064 } 3065 } 3066 } 3067 } 3068 3069 if (new_refblock_index > 0) { 3070 /* Complete the potentially existing partially filled final refblock */ 3071 if (new_set_refcount) { 3072 for (; new_refblock_index < new_refblock_size; 3073 new_refblock_index++) 3074 { 3075 new_set_refcount(new_refblock, new_refblock_index, 0); 3076 } 3077 } 3078 3079 ret = operation(bs, new_reftable, *new_reftable_index, 3080 new_reftable_size, new_refblock, new_refblock_empty, 3081 allocated, errp); 3082 if (ret < 0) { 3083 return ret; 3084 } 3085 3086 (*new_reftable_index)++; 3087 } 3088 3089 status_cb(bs, (uint64_t)(index + 1) * s->refcount_table_size, 3090 (uint64_t)total * s->refcount_table_size, cb_opaque); 3091 3092 return 0; 3093 } 3094 3095 int qcow2_change_refcount_order(BlockDriverState *bs, int refcount_order, 3096 BlockDriverAmendStatusCB *status_cb, 3097 void *cb_opaque, Error **errp) 3098 { 3099 BDRVQcow2State *s = bs->opaque; 3100 Qcow2GetRefcountFunc *new_get_refcount; 3101 Qcow2SetRefcountFunc *new_set_refcount; 3102 void *new_refblock = qemu_blockalign(bs->file->bs, s->cluster_size); 3103 uint64_t *new_reftable = NULL, new_reftable_size = 0; 3104 uint64_t *old_reftable, old_reftable_size, old_reftable_offset; 3105 uint64_t new_reftable_index = 0; 3106 uint64_t i; 3107 int64_t new_reftable_offset = 0, allocated_reftable_size = 0; 3108 int new_refblock_size, new_refcount_bits = 1 << refcount_order; 3109 int old_refcount_order; 3110 int walk_index = 0; 3111 int ret; 3112 bool new_allocation; 3113 3114 assert(s->qcow_version >= 3); 3115 assert(refcount_order >= 0 && refcount_order <= 6); 3116 3117 /* see qcow2_open() */ 3118 new_refblock_size = 1 << (s->cluster_bits - (refcount_order - 3)); 3119 3120 new_get_refcount = get_refcount_funcs[refcount_order]; 3121 new_set_refcount = set_refcount_funcs[refcount_order]; 3122 3123 3124 do { 3125 int total_walks; 3126 3127 new_allocation = false; 3128 3129 /* At least we have to do this walk and the one which writes the 3130 * refblocks; also, at least we have to do this loop here at least 3131 * twice (normally), first to do the allocations, and second to 3132 * determine that everything is correctly allocated, this then makes 3133 * three walks in total */ 3134 total_walks = MAX(walk_index + 2, 3); 3135 3136 /* First, allocate the structures so they are present in the refcount 3137 * structures */ 3138 ret = walk_over_reftable(bs, &new_reftable, &new_reftable_index, 3139 &new_reftable_size, NULL, new_refblock_size, 3140 new_refcount_bits, &alloc_refblock, 3141 &new_allocation, NULL, status_cb, cb_opaque, 3142 walk_index++, total_walks, errp); 3143 if (ret < 0) { 3144 goto done; 3145 } 3146 3147 new_reftable_index = 0; 3148 3149 if (new_allocation) { 3150 if (new_reftable_offset) { 3151 qcow2_free_clusters(bs, new_reftable_offset, 3152 allocated_reftable_size * sizeof(uint64_t), 3153 QCOW2_DISCARD_NEVER); 3154 } 3155 3156 new_reftable_offset = qcow2_alloc_clusters(bs, new_reftable_size * 3157 sizeof(uint64_t)); 3158 if (new_reftable_offset < 0) { 3159 error_setg_errno(errp, -new_reftable_offset, 3160 "Failed to allocate the new reftable"); 3161 ret = new_reftable_offset; 3162 goto done; 3163 } 3164 allocated_reftable_size = new_reftable_size; 3165 } 3166 } while (new_allocation); 3167 3168 /* Second, write the new refblocks */ 3169 ret = walk_over_reftable(bs, &new_reftable, &new_reftable_index, 3170 &new_reftable_size, new_refblock, 3171 new_refblock_size, new_refcount_bits, 3172 &flush_refblock, &new_allocation, new_set_refcount, 3173 status_cb, cb_opaque, walk_index, walk_index + 1, 3174 errp); 3175 if (ret < 0) { 3176 goto done; 3177 } 3178 assert(!new_allocation); 3179 3180 3181 /* Write the new reftable */ 3182 ret = qcow2_pre_write_overlap_check(bs, 0, new_reftable_offset, 3183 new_reftable_size * sizeof(uint64_t), 3184 false); 3185 if (ret < 0) { 3186 error_setg_errno(errp, -ret, "Overlap check failed"); 3187 goto done; 3188 } 3189 3190 for (i = 0; i < new_reftable_size; i++) { 3191 cpu_to_be64s(&new_reftable[i]); 3192 } 3193 3194 ret = bdrv_pwrite(bs->file, new_reftable_offset, new_reftable, 3195 new_reftable_size * sizeof(uint64_t)); 3196 3197 for (i = 0; i < new_reftable_size; i++) { 3198 be64_to_cpus(&new_reftable[i]); 3199 } 3200 3201 if (ret < 0) { 3202 error_setg_errno(errp, -ret, "Failed to write the new reftable"); 3203 goto done; 3204 } 3205 3206 3207 /* Empty the refcount cache */ 3208 ret = qcow2_cache_flush(bs, s->refcount_block_cache); 3209 if (ret < 0) { 3210 error_setg_errno(errp, -ret, "Failed to flush the refblock cache"); 3211 goto done; 3212 } 3213 3214 /* Update the image header to point to the new reftable; this only updates 3215 * the fields which are relevant to qcow2_update_header(); other fields 3216 * such as s->refcount_table or s->refcount_bits stay stale for now 3217 * (because we have to restore everything if qcow2_update_header() fails) */ 3218 old_refcount_order = s->refcount_order; 3219 old_reftable_size = s->refcount_table_size; 3220 old_reftable_offset = s->refcount_table_offset; 3221 3222 s->refcount_order = refcount_order; 3223 s->refcount_table_size = new_reftable_size; 3224 s->refcount_table_offset = new_reftable_offset; 3225 3226 ret = qcow2_update_header(bs); 3227 if (ret < 0) { 3228 s->refcount_order = old_refcount_order; 3229 s->refcount_table_size = old_reftable_size; 3230 s->refcount_table_offset = old_reftable_offset; 3231 error_setg_errno(errp, -ret, "Failed to update the qcow2 header"); 3232 goto done; 3233 } 3234 3235 /* Now update the rest of the in-memory information */ 3236 old_reftable = s->refcount_table; 3237 s->refcount_table = new_reftable; 3238 update_max_refcount_table_index(s); 3239 3240 s->refcount_bits = 1 << refcount_order; 3241 s->refcount_max = UINT64_C(1) << (s->refcount_bits - 1); 3242 s->refcount_max += s->refcount_max - 1; 3243 3244 s->refcount_block_bits = s->cluster_bits - (refcount_order - 3); 3245 s->refcount_block_size = 1 << s->refcount_block_bits; 3246 3247 s->get_refcount = new_get_refcount; 3248 s->set_refcount = new_set_refcount; 3249 3250 /* For cleaning up all old refblocks and the old reftable below the "done" 3251 * label */ 3252 new_reftable = old_reftable; 3253 new_reftable_size = old_reftable_size; 3254 new_reftable_offset = old_reftable_offset; 3255 3256 done: 3257 if (new_reftable) { 3258 /* On success, new_reftable actually points to the old reftable (and 3259 * new_reftable_size is the old reftable's size); but that is just 3260 * fine */ 3261 for (i = 0; i < new_reftable_size; i++) { 3262 uint64_t offset = new_reftable[i] & REFT_OFFSET_MASK; 3263 if (offset) { 3264 qcow2_free_clusters(bs, offset, s->cluster_size, 3265 QCOW2_DISCARD_OTHER); 3266 } 3267 } 3268 g_free(new_reftable); 3269 3270 if (new_reftable_offset > 0) { 3271 qcow2_free_clusters(bs, new_reftable_offset, 3272 new_reftable_size * sizeof(uint64_t), 3273 QCOW2_DISCARD_OTHER); 3274 } 3275 } 3276 3277 qemu_vfree(new_refblock); 3278 return ret; 3279 } 3280 3281 static int64_t get_refblock_offset(BlockDriverState *bs, uint64_t offset) 3282 { 3283 BDRVQcow2State *s = bs->opaque; 3284 uint32_t index = offset_to_reftable_index(s, offset); 3285 int64_t covering_refblock_offset = 0; 3286 3287 if (index < s->refcount_table_size) { 3288 covering_refblock_offset = s->refcount_table[index] & REFT_OFFSET_MASK; 3289 } 3290 if (!covering_refblock_offset) { 3291 qcow2_signal_corruption(bs, true, -1, -1, "Refblock at %#" PRIx64 " is " 3292 "not covered by the refcount structures", 3293 offset); 3294 return -EIO; 3295 } 3296 3297 return covering_refblock_offset; 3298 } 3299 3300 static int qcow2_discard_refcount_block(BlockDriverState *bs, 3301 uint64_t discard_block_offs) 3302 { 3303 BDRVQcow2State *s = bs->opaque; 3304 int64_t refblock_offs; 3305 uint64_t cluster_index = discard_block_offs >> s->cluster_bits; 3306 uint32_t block_index = cluster_index & (s->refcount_block_size - 1); 3307 void *refblock; 3308 int ret; 3309 3310 refblock_offs = get_refblock_offset(bs, discard_block_offs); 3311 if (refblock_offs < 0) { 3312 return refblock_offs; 3313 } 3314 3315 assert(discard_block_offs != 0); 3316 3317 ret = qcow2_cache_get(bs, s->refcount_block_cache, refblock_offs, 3318 &refblock); 3319 if (ret < 0) { 3320 return ret; 3321 } 3322 3323 if (s->get_refcount(refblock, block_index) != 1) { 3324 qcow2_signal_corruption(bs, true, -1, -1, "Invalid refcount:" 3325 " refblock offset %#" PRIx64 3326 ", reftable index %u" 3327 ", block offset %#" PRIx64 3328 ", refcount %#" PRIx64, 3329 refblock_offs, 3330 offset_to_reftable_index(s, discard_block_offs), 3331 discard_block_offs, 3332 s->get_refcount(refblock, block_index)); 3333 qcow2_cache_put(s->refcount_block_cache, &refblock); 3334 return -EINVAL; 3335 } 3336 s->set_refcount(refblock, block_index, 0); 3337 3338 qcow2_cache_entry_mark_dirty(s->refcount_block_cache, refblock); 3339 3340 qcow2_cache_put(s->refcount_block_cache, &refblock); 3341 3342 if (cluster_index < s->free_cluster_index) { 3343 s->free_cluster_index = cluster_index; 3344 } 3345 3346 refblock = qcow2_cache_is_table_offset(s->refcount_block_cache, 3347 discard_block_offs); 3348 if (refblock) { 3349 /* discard refblock from the cache if refblock is cached */ 3350 qcow2_cache_discard(s->refcount_block_cache, refblock); 3351 } 3352 update_refcount_discard(bs, discard_block_offs, s->cluster_size); 3353 3354 return 0; 3355 } 3356 3357 int qcow2_shrink_reftable(BlockDriverState *bs) 3358 { 3359 BDRVQcow2State *s = bs->opaque; 3360 uint64_t *reftable_tmp = 3361 g_malloc(s->refcount_table_size * sizeof(uint64_t)); 3362 int i, ret; 3363 3364 for (i = 0; i < s->refcount_table_size; i++) { 3365 int64_t refblock_offs = s->refcount_table[i] & REFT_OFFSET_MASK; 3366 void *refblock; 3367 bool unused_block; 3368 3369 if (refblock_offs == 0) { 3370 reftable_tmp[i] = 0; 3371 continue; 3372 } 3373 ret = qcow2_cache_get(bs, s->refcount_block_cache, refblock_offs, 3374 &refblock); 3375 if (ret < 0) { 3376 goto out; 3377 } 3378 3379 /* the refblock has own reference */ 3380 if (i == offset_to_reftable_index(s, refblock_offs)) { 3381 uint64_t block_index = (refblock_offs >> s->cluster_bits) & 3382 (s->refcount_block_size - 1); 3383 uint64_t refcount = s->get_refcount(refblock, block_index); 3384 3385 s->set_refcount(refblock, block_index, 0); 3386 3387 unused_block = buffer_is_zero(refblock, s->cluster_size); 3388 3389 s->set_refcount(refblock, block_index, refcount); 3390 } else { 3391 unused_block = buffer_is_zero(refblock, s->cluster_size); 3392 } 3393 qcow2_cache_put(s->refcount_block_cache, &refblock); 3394 3395 reftable_tmp[i] = unused_block ? 0 : cpu_to_be64(s->refcount_table[i]); 3396 } 3397 3398 ret = bdrv_pwrite_sync(bs->file, s->refcount_table_offset, reftable_tmp, 3399 s->refcount_table_size * sizeof(uint64_t)); 3400 /* 3401 * If the write in the reftable failed the image may contain a partially 3402 * overwritten reftable. In this case it would be better to clear the 3403 * reftable in memory to avoid possible image corruption. 3404 */ 3405 for (i = 0; i < s->refcount_table_size; i++) { 3406 if (s->refcount_table[i] && !reftable_tmp[i]) { 3407 if (ret == 0) { 3408 ret = qcow2_discard_refcount_block(bs, s->refcount_table[i] & 3409 REFT_OFFSET_MASK); 3410 } 3411 s->refcount_table[i] = 0; 3412 } 3413 } 3414 3415 if (!s->cache_discards) { 3416 qcow2_process_discards(bs, ret); 3417 } 3418 3419 out: 3420 g_free(reftable_tmp); 3421 return ret; 3422 } 3423 3424 int64_t qcow2_get_last_cluster(BlockDriverState *bs, int64_t size) 3425 { 3426 BDRVQcow2State *s = bs->opaque; 3427 int64_t i; 3428 3429 for (i = size_to_clusters(s, size) - 1; i >= 0; i--) { 3430 uint64_t refcount; 3431 int ret = qcow2_get_refcount(bs, i, &refcount); 3432 if (ret < 0) { 3433 fprintf(stderr, "Can't get refcount for cluster %" PRId64 ": %s\n", 3434 i, strerror(-ret)); 3435 return ret; 3436 } 3437 if (refcount > 0) { 3438 return i; 3439 } 3440 } 3441 qcow2_signal_corruption(bs, true, -1, -1, 3442 "There are no references in the refcount table."); 3443 return -EIO; 3444 } 3445