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 "qcow2.h" 29 #include "qemu/range.h" 30 #include "qemu/bswap.h" 31 #include "qemu/cutils.h" 32 33 static int64_t alloc_clusters_noref(BlockDriverState *bs, uint64_t size, 34 uint64_t max); 35 static int QEMU_WARN_UNUSED_RESULT update_refcount(BlockDriverState *bs, 36 int64_t offset, int64_t length, uint64_t addend, 37 bool decrease, enum qcow2_discard_type type); 38 39 static uint64_t get_refcount_ro0(const void *refcount_array, uint64_t index); 40 static uint64_t get_refcount_ro1(const void *refcount_array, uint64_t index); 41 static uint64_t get_refcount_ro2(const void *refcount_array, uint64_t index); 42 static uint64_t get_refcount_ro3(const void *refcount_array, uint64_t index); 43 static uint64_t get_refcount_ro4(const void *refcount_array, uint64_t index); 44 static uint64_t get_refcount_ro5(const void *refcount_array, uint64_t index); 45 static uint64_t get_refcount_ro6(const void *refcount_array, uint64_t index); 46 47 static void set_refcount_ro0(void *refcount_array, uint64_t index, 48 uint64_t value); 49 static void set_refcount_ro1(void *refcount_array, uint64_t index, 50 uint64_t value); 51 static void set_refcount_ro2(void *refcount_array, uint64_t index, 52 uint64_t value); 53 static void set_refcount_ro3(void *refcount_array, uint64_t index, 54 uint64_t value); 55 static void set_refcount_ro4(void *refcount_array, uint64_t index, 56 uint64_t value); 57 static void set_refcount_ro5(void *refcount_array, uint64_t index, 58 uint64_t value); 59 static void set_refcount_ro6(void *refcount_array, uint64_t index, 60 uint64_t value); 61 62 63 static Qcow2GetRefcountFunc *const get_refcount_funcs[] = { 64 &get_refcount_ro0, 65 &get_refcount_ro1, 66 &get_refcount_ro2, 67 &get_refcount_ro3, 68 &get_refcount_ro4, 69 &get_refcount_ro5, 70 &get_refcount_ro6 71 }; 72 73 static Qcow2SetRefcountFunc *const set_refcount_funcs[] = { 74 &set_refcount_ro0, 75 &set_refcount_ro1, 76 &set_refcount_ro2, 77 &set_refcount_ro3, 78 &set_refcount_ro4, 79 &set_refcount_ro5, 80 &set_refcount_ro6 81 }; 82 83 84 /*********************************************************/ 85 /* refcount handling */ 86 87 static void update_max_refcount_table_index(BDRVQcow2State *s) 88 { 89 unsigned i = s->refcount_table_size - 1; 90 while (i > 0 && (s->refcount_table[i] & REFT_OFFSET_MASK) == 0) { 91 i--; 92 } 93 /* Set s->max_refcount_table_index to the index of the last used entry */ 94 s->max_refcount_table_index = i; 95 } 96 97 int qcow2_refcount_init(BlockDriverState *bs) 98 { 99 BDRVQcow2State *s = bs->opaque; 100 unsigned int refcount_table_size2, i; 101 int ret; 102 103 assert(s->refcount_order >= 0 && s->refcount_order <= 6); 104 105 s->get_refcount = get_refcount_funcs[s->refcount_order]; 106 s->set_refcount = set_refcount_funcs[s->refcount_order]; 107 108 assert(s->refcount_table_size <= INT_MAX / sizeof(uint64_t)); 109 refcount_table_size2 = s->refcount_table_size * sizeof(uint64_t); 110 s->refcount_table = g_try_malloc(refcount_table_size2); 111 112 if (s->refcount_table_size > 0) { 113 if (s->refcount_table == NULL) { 114 ret = -ENOMEM; 115 goto fail; 116 } 117 BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_LOAD); 118 ret = bdrv_pread(bs->file, s->refcount_table_offset, 119 s->refcount_table, refcount_table_size2); 120 if (ret < 0) { 121 goto fail; 122 } 123 for(i = 0; i < s->refcount_table_size; i++) 124 be64_to_cpus(&s->refcount_table[i]); 125 update_max_refcount_table_index(s); 126 } 127 return 0; 128 fail: 129 return ret; 130 } 131 132 void qcow2_refcount_close(BlockDriverState *bs) 133 { 134 BDRVQcow2State *s = bs->opaque; 135 g_free(s->refcount_table); 136 } 137 138 139 static uint64_t get_refcount_ro0(const void *refcount_array, uint64_t index) 140 { 141 return (((const uint8_t *)refcount_array)[index / 8] >> (index % 8)) & 0x1; 142 } 143 144 static void set_refcount_ro0(void *refcount_array, uint64_t index, 145 uint64_t value) 146 { 147 assert(!(value >> 1)); 148 ((uint8_t *)refcount_array)[index / 8] &= ~(0x1 << (index % 8)); 149 ((uint8_t *)refcount_array)[index / 8] |= value << (index % 8); 150 } 151 152 static uint64_t get_refcount_ro1(const void *refcount_array, uint64_t index) 153 { 154 return (((const uint8_t *)refcount_array)[index / 4] >> (2 * (index % 4))) 155 & 0x3; 156 } 157 158 static void set_refcount_ro1(void *refcount_array, uint64_t index, 159 uint64_t value) 160 { 161 assert(!(value >> 2)); 162 ((uint8_t *)refcount_array)[index / 4] &= ~(0x3 << (2 * (index % 4))); 163 ((uint8_t *)refcount_array)[index / 4] |= value << (2 * (index % 4)); 164 } 165 166 static uint64_t get_refcount_ro2(const void *refcount_array, uint64_t index) 167 { 168 return (((const uint8_t *)refcount_array)[index / 2] >> (4 * (index % 2))) 169 & 0xf; 170 } 171 172 static void set_refcount_ro2(void *refcount_array, uint64_t index, 173 uint64_t value) 174 { 175 assert(!(value >> 4)); 176 ((uint8_t *)refcount_array)[index / 2] &= ~(0xf << (4 * (index % 2))); 177 ((uint8_t *)refcount_array)[index / 2] |= value << (4 * (index % 2)); 178 } 179 180 static uint64_t get_refcount_ro3(const void *refcount_array, uint64_t index) 181 { 182 return ((const uint8_t *)refcount_array)[index]; 183 } 184 185 static void set_refcount_ro3(void *refcount_array, uint64_t index, 186 uint64_t value) 187 { 188 assert(!(value >> 8)); 189 ((uint8_t *)refcount_array)[index] = value; 190 } 191 192 static uint64_t get_refcount_ro4(const void *refcount_array, uint64_t index) 193 { 194 return be16_to_cpu(((const uint16_t *)refcount_array)[index]); 195 } 196 197 static void set_refcount_ro4(void *refcount_array, uint64_t index, 198 uint64_t value) 199 { 200 assert(!(value >> 16)); 201 ((uint16_t *)refcount_array)[index] = cpu_to_be16(value); 202 } 203 204 static uint64_t get_refcount_ro5(const void *refcount_array, uint64_t index) 205 { 206 return be32_to_cpu(((const uint32_t *)refcount_array)[index]); 207 } 208 209 static void set_refcount_ro5(void *refcount_array, uint64_t index, 210 uint64_t value) 211 { 212 assert(!(value >> 32)); 213 ((uint32_t *)refcount_array)[index] = cpu_to_be32(value); 214 } 215 216 static uint64_t get_refcount_ro6(const void *refcount_array, uint64_t index) 217 { 218 return be64_to_cpu(((const uint64_t *)refcount_array)[index]); 219 } 220 221 static void set_refcount_ro6(void *refcount_array, uint64_t index, 222 uint64_t value) 223 { 224 ((uint64_t *)refcount_array)[index] = cpu_to_be64(value); 225 } 226 227 228 static int load_refcount_block(BlockDriverState *bs, 229 int64_t refcount_block_offset, 230 void **refcount_block) 231 { 232 BDRVQcow2State *s = bs->opaque; 233 234 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_LOAD); 235 return qcow2_cache_get(bs, s->refcount_block_cache, refcount_block_offset, 236 refcount_block); 237 } 238 239 /* 240 * Retrieves the refcount of the cluster given by its index and stores it in 241 * *refcount. Returns 0 on success and -errno on failure. 242 */ 243 int qcow2_get_refcount(BlockDriverState *bs, int64_t cluster_index, 244 uint64_t *refcount) 245 { 246 BDRVQcow2State *s = bs->opaque; 247 uint64_t refcount_table_index, block_index; 248 int64_t refcount_block_offset; 249 int ret; 250 void *refcount_block; 251 252 refcount_table_index = cluster_index >> s->refcount_block_bits; 253 if (refcount_table_index >= s->refcount_table_size) { 254 *refcount = 0; 255 return 0; 256 } 257 refcount_block_offset = 258 s->refcount_table[refcount_table_index] & REFT_OFFSET_MASK; 259 if (!refcount_block_offset) { 260 *refcount = 0; 261 return 0; 262 } 263 264 if (offset_into_cluster(s, refcount_block_offset)) { 265 qcow2_signal_corruption(bs, true, -1, -1, "Refblock offset %#" PRIx64 266 " unaligned (reftable index: %#" PRIx64 ")", 267 refcount_block_offset, refcount_table_index); 268 return -EIO; 269 } 270 271 ret = qcow2_cache_get(bs, s->refcount_block_cache, refcount_block_offset, 272 &refcount_block); 273 if (ret < 0) { 274 return ret; 275 } 276 277 block_index = cluster_index & (s->refcount_block_size - 1); 278 *refcount = s->get_refcount(refcount_block, block_index); 279 280 qcow2_cache_put(s->refcount_block_cache, &refcount_block); 281 282 return 0; 283 } 284 285 /* Checks if two offsets are described by the same refcount block */ 286 static int in_same_refcount_block(BDRVQcow2State *s, uint64_t offset_a, 287 uint64_t offset_b) 288 { 289 uint64_t block_a = offset_a >> (s->cluster_bits + s->refcount_block_bits); 290 uint64_t block_b = offset_b >> (s->cluster_bits + s->refcount_block_bits); 291 292 return (block_a == block_b); 293 } 294 295 /* 296 * Loads a refcount block. If it doesn't exist yet, it is allocated first 297 * (including growing the refcount table if needed). 298 * 299 * Returns 0 on success or -errno in error case 300 */ 301 static int alloc_refcount_block(BlockDriverState *bs, 302 int64_t cluster_index, void **refcount_block) 303 { 304 BDRVQcow2State *s = bs->opaque; 305 unsigned int refcount_table_index; 306 int64_t ret; 307 308 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC); 309 310 /* Find the refcount block for the given cluster */ 311 refcount_table_index = cluster_index >> s->refcount_block_bits; 312 313 if (refcount_table_index < s->refcount_table_size) { 314 315 uint64_t refcount_block_offset = 316 s->refcount_table[refcount_table_index] & REFT_OFFSET_MASK; 317 318 /* If it's already there, we're done */ 319 if (refcount_block_offset) { 320 if (offset_into_cluster(s, refcount_block_offset)) { 321 qcow2_signal_corruption(bs, true, -1, -1, "Refblock offset %#" 322 PRIx64 " unaligned (reftable index: " 323 "%#x)", refcount_block_offset, 324 refcount_table_index); 325 return -EIO; 326 } 327 328 return load_refcount_block(bs, refcount_block_offset, 329 refcount_block); 330 } 331 } 332 333 /* 334 * If we came here, we need to allocate something. Something is at least 335 * a cluster for the new refcount block. It may also include a new refcount 336 * table if the old refcount table is too small. 337 * 338 * Note that allocating clusters here needs some special care: 339 * 340 * - We can't use the normal qcow2_alloc_clusters(), it would try to 341 * increase the refcount and very likely we would end up with an endless 342 * recursion. Instead we must place the refcount blocks in a way that 343 * they can describe them themselves. 344 * 345 * - We need to consider that at this point we are inside update_refcounts 346 * and potentially doing an initial refcount increase. This means that 347 * some clusters have already been allocated by the caller, but their 348 * refcount isn't accurate yet. If we allocate clusters for metadata, we 349 * need to return -EAGAIN to signal the caller that it needs to restart 350 * the search for free clusters. 351 * 352 * - alloc_clusters_noref and qcow2_free_clusters may load a different 353 * refcount block into the cache 354 */ 355 356 *refcount_block = NULL; 357 358 /* We write to the refcount table, so we might depend on L2 tables */ 359 ret = qcow2_cache_flush(bs, s->l2_table_cache); 360 if (ret < 0) { 361 return ret; 362 } 363 364 /* Allocate the refcount block itself and mark it as used */ 365 int64_t new_block = alloc_clusters_noref(bs, s->cluster_size, INT64_MAX); 366 if (new_block < 0) { 367 return new_block; 368 } 369 370 /* The offset must fit in the offset field of the refcount table entry */ 371 assert((new_block & REFT_OFFSET_MASK) == new_block); 372 373 /* If we're allocating the block at offset 0 then something is wrong */ 374 if (new_block == 0) { 375 qcow2_signal_corruption(bs, true, -1, -1, "Preventing invalid " 376 "allocation of refcount block at offset 0"); 377 return -EIO; 378 } 379 380 #ifdef DEBUG_ALLOC2 381 fprintf(stderr, "qcow2: Allocate refcount block %d for %" PRIx64 382 " at %" PRIx64 "\n", 383 refcount_table_index, cluster_index << s->cluster_bits, new_block); 384 #endif 385 386 if (in_same_refcount_block(s, new_block, cluster_index << s->cluster_bits)) { 387 /* Zero the new refcount block before updating it */ 388 ret = qcow2_cache_get_empty(bs, s->refcount_block_cache, new_block, 389 refcount_block); 390 if (ret < 0) { 391 goto fail; 392 } 393 394 memset(*refcount_block, 0, s->cluster_size); 395 396 /* The block describes itself, need to update the cache */ 397 int block_index = (new_block >> s->cluster_bits) & 398 (s->refcount_block_size - 1); 399 s->set_refcount(*refcount_block, block_index, 1); 400 } else { 401 /* Described somewhere else. This can recurse at most twice before we 402 * arrive at a block that describes itself. */ 403 ret = update_refcount(bs, new_block, s->cluster_size, 1, false, 404 QCOW2_DISCARD_NEVER); 405 if (ret < 0) { 406 goto fail; 407 } 408 409 ret = qcow2_cache_flush(bs, s->refcount_block_cache); 410 if (ret < 0) { 411 goto fail; 412 } 413 414 /* Initialize the new refcount block only after updating its refcount, 415 * update_refcount uses the refcount cache itself */ 416 ret = qcow2_cache_get_empty(bs, s->refcount_block_cache, new_block, 417 refcount_block); 418 if (ret < 0) { 419 goto fail; 420 } 421 422 memset(*refcount_block, 0, s->cluster_size); 423 } 424 425 /* Now the new refcount block needs to be written to disk */ 426 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE); 427 qcow2_cache_entry_mark_dirty(s->refcount_block_cache, *refcount_block); 428 ret = qcow2_cache_flush(bs, s->refcount_block_cache); 429 if (ret < 0) { 430 goto fail; 431 } 432 433 /* If the refcount table is big enough, just hook the block up there */ 434 if (refcount_table_index < s->refcount_table_size) { 435 uint64_t data64 = cpu_to_be64(new_block); 436 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_HOOKUP); 437 ret = bdrv_pwrite_sync(bs->file, 438 s->refcount_table_offset + refcount_table_index * sizeof(uint64_t), 439 &data64, sizeof(data64)); 440 if (ret < 0) { 441 goto fail; 442 } 443 444 s->refcount_table[refcount_table_index] = new_block; 445 /* If there's a hole in s->refcount_table then it can happen 446 * that refcount_table_index < s->max_refcount_table_index */ 447 s->max_refcount_table_index = 448 MAX(s->max_refcount_table_index, refcount_table_index); 449 450 /* The new refcount block may be where the caller intended to put its 451 * data, so let it restart the search. */ 452 return -EAGAIN; 453 } 454 455 qcow2_cache_put(s->refcount_block_cache, refcount_block); 456 457 /* 458 * If we come here, we need to grow the refcount table. Again, a new 459 * refcount table needs some space and we can't simply allocate to avoid 460 * endless recursion. 461 * 462 * Therefore let's grab new refcount blocks at the end of the image, which 463 * will describe themselves and the new refcount table. This way we can 464 * reference them only in the new table and do the switch to the new 465 * refcount table at once without producing an inconsistent state in 466 * between. 467 */ 468 BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_GROW); 469 470 /* Calculate the number of refcount blocks needed so far; this will be the 471 * basis for calculating the index of the first cluster used for the 472 * self-describing refcount structures which we are about to create. 473 * 474 * Because we reached this point, there cannot be any refcount entries for 475 * cluster_index or higher indices yet. However, because new_block has been 476 * allocated to describe that cluster (and it will assume this role later 477 * on), we cannot use that index; also, new_block may actually have a higher 478 * cluster index than cluster_index, so it needs to be taken into account 479 * here (and 1 needs to be added to its value because that cluster is used). 480 */ 481 uint64_t blocks_used = DIV_ROUND_UP(MAX(cluster_index + 1, 482 (new_block >> s->cluster_bits) + 1), 483 s->refcount_block_size); 484 485 /* Create the new refcount table and blocks */ 486 uint64_t meta_offset = (blocks_used * s->refcount_block_size) * 487 s->cluster_size; 488 489 ret = qcow2_refcount_area(bs, meta_offset, 0, false, 490 refcount_table_index, new_block); 491 if (ret < 0) { 492 return ret; 493 } 494 495 ret = load_refcount_block(bs, new_block, refcount_block); 496 if (ret < 0) { 497 return ret; 498 } 499 500 /* If we were trying to do the initial refcount update for some cluster 501 * allocation, we might have used the same clusters to store newly 502 * allocated metadata. Make the caller search some new space. */ 503 return -EAGAIN; 504 505 fail: 506 if (*refcount_block != NULL) { 507 qcow2_cache_put(s->refcount_block_cache, refcount_block); 508 } 509 return ret; 510 } 511 512 /* 513 * Starting at @start_offset, this function creates new self-covering refcount 514 * structures: A new refcount table and refcount blocks which cover all of 515 * themselves, and a number of @additional_clusters beyond their end. 516 * @start_offset must be at the end of the image file, that is, there must be 517 * only empty space beyond it. 518 * If @exact_size is false, the refcount table will have 50 % more entries than 519 * necessary so it will not need to grow again soon. 520 * If @new_refblock_offset is not zero, it contains the offset of a refcount 521 * block that should be entered into the new refcount table at index 522 * @new_refblock_index. 523 * 524 * Returns: The offset after the new refcount structures (i.e. where the 525 * @additional_clusters may be placed) on success, -errno on error. 526 */ 527 int64_t qcow2_refcount_area(BlockDriverState *bs, uint64_t start_offset, 528 uint64_t additional_clusters, bool exact_size, 529 int new_refblock_index, 530 uint64_t new_refblock_offset) 531 { 532 BDRVQcow2State *s = bs->opaque; 533 uint64_t total_refblock_count_u64, additional_refblock_count; 534 int total_refblock_count, table_size, area_reftable_index, table_clusters; 535 int i; 536 uint64_t table_offset, block_offset, end_offset; 537 int ret; 538 uint64_t *new_table; 539 540 assert(!(start_offset % s->cluster_size)); 541 542 qcow2_refcount_metadata_size(start_offset / s->cluster_size + 543 additional_clusters, 544 s->cluster_size, s->refcount_order, 545 !exact_size, &total_refblock_count_u64); 546 if (total_refblock_count_u64 > QCOW_MAX_REFTABLE_SIZE) { 547 return -EFBIG; 548 } 549 total_refblock_count = total_refblock_count_u64; 550 551 /* Index in the refcount table of the first refcount block to cover the area 552 * of refcount structures we are about to create; we know that 553 * @total_refblock_count can cover @start_offset, so this will definitely 554 * fit into an int. */ 555 area_reftable_index = (start_offset / s->cluster_size) / 556 s->refcount_block_size; 557 558 if (exact_size) { 559 table_size = total_refblock_count; 560 } else { 561 table_size = total_refblock_count + 562 DIV_ROUND_UP(total_refblock_count, 2); 563 } 564 /* The qcow2 file can only store the reftable size in number of clusters */ 565 table_size = ROUND_UP(table_size, s->cluster_size / sizeof(uint64_t)); 566 table_clusters = (table_size * sizeof(uint64_t)) / s->cluster_size; 567 568 if (table_size > QCOW_MAX_REFTABLE_SIZE) { 569 return -EFBIG; 570 } 571 572 new_table = g_try_new0(uint64_t, table_size); 573 574 assert(table_size > 0); 575 if (new_table == NULL) { 576 ret = -ENOMEM; 577 goto fail; 578 } 579 580 /* Fill the new refcount table */ 581 if (table_size > s->max_refcount_table_index) { 582 /* We're actually growing the reftable */ 583 memcpy(new_table, s->refcount_table, 584 (s->max_refcount_table_index + 1) * sizeof(uint64_t)); 585 } else { 586 /* Improbable case: We're shrinking the reftable. However, the caller 587 * has assured us that there is only empty space beyond @start_offset, 588 * so we can simply drop all of the refblocks that won't fit into the 589 * new reftable. */ 590 memcpy(new_table, s->refcount_table, table_size * sizeof(uint64_t)); 591 } 592 593 if (new_refblock_offset) { 594 assert(new_refblock_index < total_refblock_count); 595 new_table[new_refblock_index] = new_refblock_offset; 596 } 597 598 /* Count how many new refblocks we have to create */ 599 additional_refblock_count = 0; 600 for (i = area_reftable_index; i < total_refblock_count; i++) { 601 if (!new_table[i]) { 602 additional_refblock_count++; 603 } 604 } 605 606 table_offset = start_offset + additional_refblock_count * s->cluster_size; 607 end_offset = table_offset + table_clusters * s->cluster_size; 608 609 /* Fill the refcount blocks, and create new ones, if necessary */ 610 block_offset = start_offset; 611 for (i = area_reftable_index; i < total_refblock_count; i++) { 612 void *refblock_data; 613 uint64_t first_offset_covered; 614 615 /* Reuse an existing refblock if possible, create a new one otherwise */ 616 if (new_table[i]) { 617 ret = qcow2_cache_get(bs, s->refcount_block_cache, new_table[i], 618 &refblock_data); 619 if (ret < 0) { 620 goto fail; 621 } 622 } else { 623 ret = qcow2_cache_get_empty(bs, s->refcount_block_cache, 624 block_offset, &refblock_data); 625 if (ret < 0) { 626 goto fail; 627 } 628 memset(refblock_data, 0, s->cluster_size); 629 qcow2_cache_entry_mark_dirty(s->refcount_block_cache, 630 refblock_data); 631 632 new_table[i] = block_offset; 633 block_offset += s->cluster_size; 634 } 635 636 /* First host offset covered by this refblock */ 637 first_offset_covered = (uint64_t)i * s->refcount_block_size * 638 s->cluster_size; 639 if (first_offset_covered < end_offset) { 640 int j, end_index; 641 642 /* Set the refcount of all of the new refcount structures to 1 */ 643 644 if (first_offset_covered < start_offset) { 645 assert(i == area_reftable_index); 646 j = (start_offset - first_offset_covered) / s->cluster_size; 647 assert(j < s->refcount_block_size); 648 } else { 649 j = 0; 650 } 651 652 end_index = MIN((end_offset - first_offset_covered) / 653 s->cluster_size, 654 s->refcount_block_size); 655 656 for (; j < end_index; j++) { 657 /* The caller guaranteed us this space would be empty */ 658 assert(s->get_refcount(refblock_data, j) == 0); 659 s->set_refcount(refblock_data, j, 1); 660 } 661 662 qcow2_cache_entry_mark_dirty(s->refcount_block_cache, 663 refblock_data); 664 } 665 666 qcow2_cache_put(s->refcount_block_cache, &refblock_data); 667 } 668 669 assert(block_offset == table_offset); 670 671 /* Write refcount blocks to disk */ 672 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE_BLOCKS); 673 ret = qcow2_cache_flush(bs, s->refcount_block_cache); 674 if (ret < 0) { 675 goto fail; 676 } 677 678 /* Write refcount table to disk */ 679 for (i = 0; i < total_refblock_count; i++) { 680 cpu_to_be64s(&new_table[i]); 681 } 682 683 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE_TABLE); 684 ret = bdrv_pwrite_sync(bs->file, table_offset, new_table, 685 table_size * sizeof(uint64_t)); 686 if (ret < 0) { 687 goto fail; 688 } 689 690 for (i = 0; i < total_refblock_count; i++) { 691 be64_to_cpus(&new_table[i]); 692 } 693 694 /* Hook up the new refcount table in the qcow2 header */ 695 struct QEMU_PACKED { 696 uint64_t d64; 697 uint32_t d32; 698 } data; 699 data.d64 = cpu_to_be64(table_offset); 700 data.d32 = cpu_to_be32(table_clusters); 701 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_SWITCH_TABLE); 702 ret = bdrv_pwrite_sync(bs->file, 703 offsetof(QCowHeader, refcount_table_offset), 704 &data, sizeof(data)); 705 if (ret < 0) { 706 goto fail; 707 } 708 709 /* And switch it in memory */ 710 uint64_t old_table_offset = s->refcount_table_offset; 711 uint64_t old_table_size = s->refcount_table_size; 712 713 g_free(s->refcount_table); 714 s->refcount_table = new_table; 715 s->refcount_table_size = table_size; 716 s->refcount_table_offset = table_offset; 717 update_max_refcount_table_index(s); 718 719 /* Free old table. */ 720 qcow2_free_clusters(bs, old_table_offset, old_table_size * sizeof(uint64_t), 721 QCOW2_DISCARD_OTHER); 722 723 return end_offset; 724 725 fail: 726 g_free(new_table); 727 return ret; 728 } 729 730 void qcow2_process_discards(BlockDriverState *bs, int ret) 731 { 732 BDRVQcow2State *s = bs->opaque; 733 Qcow2DiscardRegion *d, *next; 734 735 QTAILQ_FOREACH_SAFE(d, &s->discards, next, next) { 736 QTAILQ_REMOVE(&s->discards, d, next); 737 738 /* Discard is optional, ignore the return value */ 739 if (ret >= 0) { 740 bdrv_pdiscard(bs->file, d->offset, d->bytes); 741 } 742 743 g_free(d); 744 } 745 } 746 747 static void update_refcount_discard(BlockDriverState *bs, 748 uint64_t offset, uint64_t length) 749 { 750 BDRVQcow2State *s = bs->opaque; 751 Qcow2DiscardRegion *d, *p, *next; 752 753 QTAILQ_FOREACH(d, &s->discards, next) { 754 uint64_t new_start = MIN(offset, d->offset); 755 uint64_t new_end = MAX(offset + length, d->offset + d->bytes); 756 757 if (new_end - new_start <= length + d->bytes) { 758 /* There can't be any overlap, areas ending up here have no 759 * references any more and therefore shouldn't get freed another 760 * time. */ 761 assert(d->bytes + length == new_end - new_start); 762 d->offset = new_start; 763 d->bytes = new_end - new_start; 764 goto found; 765 } 766 } 767 768 d = g_malloc(sizeof(*d)); 769 *d = (Qcow2DiscardRegion) { 770 .bs = bs, 771 .offset = offset, 772 .bytes = length, 773 }; 774 QTAILQ_INSERT_TAIL(&s->discards, d, next); 775 776 found: 777 /* Merge discard requests if they are adjacent now */ 778 QTAILQ_FOREACH_SAFE(p, &s->discards, next, next) { 779 if (p == d 780 || p->offset > d->offset + d->bytes 781 || d->offset > p->offset + p->bytes) 782 { 783 continue; 784 } 785 786 /* Still no overlap possible */ 787 assert(p->offset == d->offset + d->bytes 788 || d->offset == p->offset + p->bytes); 789 790 QTAILQ_REMOVE(&s->discards, p, next); 791 d->offset = MIN(d->offset, p->offset); 792 d->bytes += p->bytes; 793 g_free(p); 794 } 795 } 796 797 /* XXX: cache several refcount block clusters ? */ 798 /* @addend is the absolute value of the addend; if @decrease is set, @addend 799 * will be subtracted from the current refcount, otherwise it will be added */ 800 static int QEMU_WARN_UNUSED_RESULT update_refcount(BlockDriverState *bs, 801 int64_t offset, 802 int64_t length, 803 uint64_t addend, 804 bool decrease, 805 enum qcow2_discard_type type) 806 { 807 BDRVQcow2State *s = bs->opaque; 808 int64_t start, last, cluster_offset; 809 void *refcount_block = NULL; 810 int64_t old_table_index = -1; 811 int ret; 812 813 #ifdef DEBUG_ALLOC2 814 fprintf(stderr, "update_refcount: offset=%" PRId64 " size=%" PRId64 815 " addend=%s%" PRIu64 "\n", offset, length, decrease ? "-" : "", 816 addend); 817 #endif 818 if (length < 0) { 819 return -EINVAL; 820 } else if (length == 0) { 821 return 0; 822 } 823 824 if (decrease) { 825 qcow2_cache_set_dependency(bs, s->refcount_block_cache, 826 s->l2_table_cache); 827 } 828 829 start = start_of_cluster(s, offset); 830 last = start_of_cluster(s, offset + length - 1); 831 for(cluster_offset = start; cluster_offset <= last; 832 cluster_offset += s->cluster_size) 833 { 834 int block_index; 835 uint64_t refcount; 836 int64_t cluster_index = cluster_offset >> s->cluster_bits; 837 int64_t table_index = cluster_index >> s->refcount_block_bits; 838 839 /* Load the refcount block and allocate it if needed */ 840 if (table_index != old_table_index) { 841 if (refcount_block) { 842 qcow2_cache_put(s->refcount_block_cache, &refcount_block); 843 } 844 ret = alloc_refcount_block(bs, cluster_index, &refcount_block); 845 /* If the caller needs to restart the search for free clusters, 846 * try the same ones first to see if they're still free. */ 847 if (ret == -EAGAIN) { 848 if (s->free_cluster_index > (start >> s->cluster_bits)) { 849 s->free_cluster_index = (start >> s->cluster_bits); 850 } 851 } 852 if (ret < 0) { 853 goto fail; 854 } 855 } 856 old_table_index = table_index; 857 858 qcow2_cache_entry_mark_dirty(s->refcount_block_cache, refcount_block); 859 860 /* we can update the count and save it */ 861 block_index = cluster_index & (s->refcount_block_size - 1); 862 863 refcount = s->get_refcount(refcount_block, block_index); 864 if (decrease ? (refcount - addend > refcount) 865 : (refcount + addend < refcount || 866 refcount + addend > s->refcount_max)) 867 { 868 ret = -EINVAL; 869 goto fail; 870 } 871 if (decrease) { 872 refcount -= addend; 873 } else { 874 refcount += addend; 875 } 876 if (refcount == 0 && cluster_index < s->free_cluster_index) { 877 s->free_cluster_index = cluster_index; 878 } 879 s->set_refcount(refcount_block, block_index, refcount); 880 881 if (refcount == 0) { 882 void *table; 883 884 table = qcow2_cache_is_table_offset(s->refcount_block_cache, 885 offset); 886 if (table != NULL) { 887 qcow2_cache_put(s->refcount_block_cache, &refcount_block); 888 qcow2_cache_discard(s->refcount_block_cache, table); 889 } 890 891 table = qcow2_cache_is_table_offset(s->l2_table_cache, offset); 892 if (table != NULL) { 893 qcow2_cache_discard(s->l2_table_cache, table); 894 } 895 896 if (s->discard_passthrough[type]) { 897 update_refcount_discard(bs, cluster_offset, s->cluster_size); 898 } 899 } 900 } 901 902 ret = 0; 903 fail: 904 if (!s->cache_discards) { 905 qcow2_process_discards(bs, ret); 906 } 907 908 /* Write last changed block to disk */ 909 if (refcount_block) { 910 qcow2_cache_put(s->refcount_block_cache, &refcount_block); 911 } 912 913 /* 914 * Try do undo any updates if an error is returned (This may succeed in 915 * some cases like ENOSPC for allocating a new refcount block) 916 */ 917 if (ret < 0) { 918 int dummy; 919 dummy = update_refcount(bs, offset, cluster_offset - offset, addend, 920 !decrease, QCOW2_DISCARD_NEVER); 921 (void)dummy; 922 } 923 924 return ret; 925 } 926 927 /* 928 * Increases or decreases the refcount of a given cluster. 929 * 930 * @addend is the absolute value of the addend; if @decrease is set, @addend 931 * will be subtracted from the current refcount, otherwise it will be added. 932 * 933 * On success 0 is returned; on failure -errno is returned. 934 */ 935 int qcow2_update_cluster_refcount(BlockDriverState *bs, 936 int64_t cluster_index, 937 uint64_t addend, bool decrease, 938 enum qcow2_discard_type type) 939 { 940 BDRVQcow2State *s = bs->opaque; 941 int ret; 942 943 ret = update_refcount(bs, cluster_index << s->cluster_bits, 1, addend, 944 decrease, type); 945 if (ret < 0) { 946 return ret; 947 } 948 949 return 0; 950 } 951 952 953 954 /*********************************************************/ 955 /* cluster allocation functions */ 956 957 958 959 /* return < 0 if error */ 960 static int64_t alloc_clusters_noref(BlockDriverState *bs, uint64_t size, 961 uint64_t max) 962 { 963 BDRVQcow2State *s = bs->opaque; 964 uint64_t i, nb_clusters, refcount; 965 int ret; 966 967 /* We can't allocate clusters if they may still be queued for discard. */ 968 if (s->cache_discards) { 969 qcow2_process_discards(bs, 0); 970 } 971 972 nb_clusters = size_to_clusters(s, size); 973 retry: 974 for(i = 0; i < nb_clusters; i++) { 975 uint64_t next_cluster_index = s->free_cluster_index++; 976 ret = qcow2_get_refcount(bs, next_cluster_index, &refcount); 977 978 if (ret < 0) { 979 return ret; 980 } else if (refcount != 0) { 981 goto retry; 982 } 983 } 984 985 /* Make sure that all offsets in the "allocated" range are representable 986 * in the requested max */ 987 if (s->free_cluster_index > 0 && 988 s->free_cluster_index - 1 > (max >> s->cluster_bits)) 989 { 990 return -EFBIG; 991 } 992 993 #ifdef DEBUG_ALLOC2 994 fprintf(stderr, "alloc_clusters: size=%" PRId64 " -> %" PRId64 "\n", 995 size, 996 (s->free_cluster_index - nb_clusters) << s->cluster_bits); 997 #endif 998 return (s->free_cluster_index - nb_clusters) << s->cluster_bits; 999 } 1000 1001 int64_t qcow2_alloc_clusters(BlockDriverState *bs, uint64_t size) 1002 { 1003 int64_t offset; 1004 int ret; 1005 1006 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC); 1007 do { 1008 offset = alloc_clusters_noref(bs, size, QCOW_MAX_CLUSTER_OFFSET); 1009 if (offset < 0) { 1010 return offset; 1011 } 1012 1013 ret = update_refcount(bs, offset, size, 1, false, QCOW2_DISCARD_NEVER); 1014 } while (ret == -EAGAIN); 1015 1016 if (ret < 0) { 1017 return ret; 1018 } 1019 1020 return offset; 1021 } 1022 1023 int64_t qcow2_alloc_clusters_at(BlockDriverState *bs, uint64_t offset, 1024 int64_t nb_clusters) 1025 { 1026 BDRVQcow2State *s = bs->opaque; 1027 uint64_t cluster_index, refcount; 1028 uint64_t i; 1029 int ret; 1030 1031 assert(nb_clusters >= 0); 1032 if (nb_clusters == 0) { 1033 return 0; 1034 } 1035 1036 do { 1037 /* Check how many clusters there are free */ 1038 cluster_index = offset >> s->cluster_bits; 1039 for(i = 0; i < nb_clusters; i++) { 1040 ret = qcow2_get_refcount(bs, cluster_index++, &refcount); 1041 if (ret < 0) { 1042 return ret; 1043 } else if (refcount != 0) { 1044 break; 1045 } 1046 } 1047 1048 /* And then allocate them */ 1049 ret = update_refcount(bs, offset, i << s->cluster_bits, 1, false, 1050 QCOW2_DISCARD_NEVER); 1051 } while (ret == -EAGAIN); 1052 1053 if (ret < 0) { 1054 return ret; 1055 } 1056 1057 return i; 1058 } 1059 1060 /* only used to allocate compressed sectors. We try to allocate 1061 contiguous sectors. size must be <= cluster_size */ 1062 int64_t qcow2_alloc_bytes(BlockDriverState *bs, int size) 1063 { 1064 BDRVQcow2State *s = bs->opaque; 1065 int64_t offset; 1066 size_t free_in_cluster; 1067 int ret; 1068 1069 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC_BYTES); 1070 assert(size > 0 && size <= s->cluster_size); 1071 assert(!s->free_byte_offset || offset_into_cluster(s, s->free_byte_offset)); 1072 1073 offset = s->free_byte_offset; 1074 1075 if (offset) { 1076 uint64_t refcount; 1077 ret = qcow2_get_refcount(bs, offset >> s->cluster_bits, &refcount); 1078 if (ret < 0) { 1079 return ret; 1080 } 1081 1082 if (refcount == s->refcount_max) { 1083 offset = 0; 1084 } 1085 } 1086 1087 free_in_cluster = s->cluster_size - offset_into_cluster(s, offset); 1088 do { 1089 if (!offset || free_in_cluster < size) { 1090 int64_t new_cluster; 1091 1092 new_cluster = alloc_clusters_noref(bs, s->cluster_size, 1093 MIN(s->cluster_offset_mask, 1094 QCOW_MAX_CLUSTER_OFFSET)); 1095 if (new_cluster < 0) { 1096 return new_cluster; 1097 } 1098 1099 if (new_cluster == 0) { 1100 qcow2_signal_corruption(bs, true, -1, -1, "Preventing invalid " 1101 "allocation of compressed cluster " 1102 "at offset 0"); 1103 return -EIO; 1104 } 1105 1106 if (!offset || ROUND_UP(offset, s->cluster_size) != new_cluster) { 1107 offset = new_cluster; 1108 free_in_cluster = s->cluster_size; 1109 } else { 1110 free_in_cluster += s->cluster_size; 1111 } 1112 } 1113 1114 assert(offset); 1115 ret = update_refcount(bs, offset, size, 1, false, QCOW2_DISCARD_NEVER); 1116 if (ret < 0) { 1117 offset = 0; 1118 } 1119 } while (ret == -EAGAIN); 1120 if (ret < 0) { 1121 return ret; 1122 } 1123 1124 /* The cluster refcount was incremented; refcount blocks must be flushed 1125 * before the caller's L2 table updates. */ 1126 qcow2_cache_set_dependency(bs, s->l2_table_cache, s->refcount_block_cache); 1127 1128 s->free_byte_offset = offset + size; 1129 if (!offset_into_cluster(s, s->free_byte_offset)) { 1130 s->free_byte_offset = 0; 1131 } 1132 1133 return offset; 1134 } 1135 1136 void qcow2_free_clusters(BlockDriverState *bs, 1137 int64_t offset, int64_t size, 1138 enum qcow2_discard_type type) 1139 { 1140 int ret; 1141 1142 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_FREE); 1143 ret = update_refcount(bs, offset, size, 1, true, type); 1144 if (ret < 0) { 1145 fprintf(stderr, "qcow2_free_clusters failed: %s\n", strerror(-ret)); 1146 /* TODO Remember the clusters to free them later and avoid leaking */ 1147 } 1148 } 1149 1150 /* 1151 * Free a cluster using its L2 entry (handles clusters of all types, e.g. 1152 * normal cluster, compressed cluster, etc.) 1153 */ 1154 void qcow2_free_any_clusters(BlockDriverState *bs, uint64_t l2_entry, 1155 int nb_clusters, enum qcow2_discard_type type) 1156 { 1157 BDRVQcow2State *s = bs->opaque; 1158 QCow2ClusterType ctype = qcow2_get_cluster_type(bs, l2_entry); 1159 1160 if (has_data_file(bs)) { 1161 if (s->discard_passthrough[type] && 1162 (ctype == QCOW2_CLUSTER_NORMAL || 1163 ctype == QCOW2_CLUSTER_ZERO_ALLOC)) 1164 { 1165 bdrv_pdiscard(s->data_file, l2_entry & L2E_OFFSET_MASK, 1166 nb_clusters << s->cluster_bits); 1167 } 1168 return; 1169 } 1170 1171 switch (ctype) { 1172 case QCOW2_CLUSTER_COMPRESSED: 1173 { 1174 int64_t offset = (l2_entry & s->cluster_offset_mask) 1175 & QCOW2_COMPRESSED_SECTOR_MASK; 1176 int size = QCOW2_COMPRESSED_SECTOR_SIZE * 1177 (((l2_entry >> s->csize_shift) & s->csize_mask) + 1); 1178 qcow2_free_clusters(bs, offset, size, type); 1179 } 1180 break; 1181 case QCOW2_CLUSTER_NORMAL: 1182 case QCOW2_CLUSTER_ZERO_ALLOC: 1183 if (offset_into_cluster(s, l2_entry & L2E_OFFSET_MASK)) { 1184 qcow2_signal_corruption(bs, false, -1, -1, 1185 "Cannot free unaligned cluster %#llx", 1186 l2_entry & L2E_OFFSET_MASK); 1187 } else { 1188 qcow2_free_clusters(bs, l2_entry & L2E_OFFSET_MASK, 1189 nb_clusters << s->cluster_bits, type); 1190 } 1191 break; 1192 case QCOW2_CLUSTER_ZERO_PLAIN: 1193 case QCOW2_CLUSTER_UNALLOCATED: 1194 break; 1195 default: 1196 abort(); 1197 } 1198 } 1199 1200 int coroutine_fn qcow2_write_caches(BlockDriverState *bs) 1201 { 1202 BDRVQcow2State *s = bs->opaque; 1203 int ret; 1204 1205 ret = qcow2_cache_write(bs, s->l2_table_cache); 1206 if (ret < 0) { 1207 return ret; 1208 } 1209 1210 if (qcow2_need_accurate_refcounts(s)) { 1211 ret = qcow2_cache_write(bs, s->refcount_block_cache); 1212 if (ret < 0) { 1213 return ret; 1214 } 1215 } 1216 1217 return 0; 1218 } 1219 1220 int coroutine_fn qcow2_flush_caches(BlockDriverState *bs) 1221 { 1222 int ret = qcow2_write_caches(bs); 1223 if (ret < 0) { 1224 return ret; 1225 } 1226 1227 return bdrv_flush(bs->file->bs); 1228 } 1229 1230 /*********************************************************/ 1231 /* snapshots and image creation */ 1232 1233 1234 1235 /* update the refcounts of snapshots and the copied flag */ 1236 int qcow2_update_snapshot_refcount(BlockDriverState *bs, 1237 int64_t l1_table_offset, int l1_size, int addend) 1238 { 1239 BDRVQcow2State *s = bs->opaque; 1240 uint64_t *l1_table, *l2_slice, l2_offset, entry, l1_size2, refcount; 1241 bool l1_allocated = false; 1242 int64_t old_entry, old_l2_offset; 1243 unsigned slice, slice_size2, n_slices; 1244 int i, j, l1_modified = 0, nb_csectors; 1245 int ret; 1246 1247 assert(addend >= -1 && addend <= 1); 1248 1249 l2_slice = NULL; 1250 l1_table = NULL; 1251 l1_size2 = l1_size * sizeof(uint64_t); 1252 slice_size2 = s->l2_slice_size * sizeof(uint64_t); 1253 n_slices = s->cluster_size / slice_size2; 1254 1255 s->cache_discards = true; 1256 1257 /* WARNING: qcow2_snapshot_goto relies on this function not using the 1258 * l1_table_offset when it is the current s->l1_table_offset! Be careful 1259 * when changing this! */ 1260 if (l1_table_offset != s->l1_table_offset) { 1261 l1_table = g_try_malloc0(ROUND_UP(l1_size2, 512)); 1262 if (l1_size2 && l1_table == NULL) { 1263 ret = -ENOMEM; 1264 goto fail; 1265 } 1266 l1_allocated = true; 1267 1268 ret = bdrv_pread(bs->file, l1_table_offset, l1_table, l1_size2); 1269 if (ret < 0) { 1270 goto fail; 1271 } 1272 1273 for (i = 0; i < l1_size; i++) { 1274 be64_to_cpus(&l1_table[i]); 1275 } 1276 } else { 1277 assert(l1_size == s->l1_size); 1278 l1_table = s->l1_table; 1279 l1_allocated = false; 1280 } 1281 1282 for (i = 0; i < l1_size; i++) { 1283 l2_offset = l1_table[i]; 1284 if (l2_offset) { 1285 old_l2_offset = l2_offset; 1286 l2_offset &= L1E_OFFSET_MASK; 1287 1288 if (offset_into_cluster(s, l2_offset)) { 1289 qcow2_signal_corruption(bs, true, -1, -1, "L2 table offset %#" 1290 PRIx64 " unaligned (L1 index: %#x)", 1291 l2_offset, i); 1292 ret = -EIO; 1293 goto fail; 1294 } 1295 1296 for (slice = 0; slice < n_slices; slice++) { 1297 ret = qcow2_cache_get(bs, s->l2_table_cache, 1298 l2_offset + slice * slice_size2, 1299 (void **) &l2_slice); 1300 if (ret < 0) { 1301 goto fail; 1302 } 1303 1304 for (j = 0; j < s->l2_slice_size; j++) { 1305 uint64_t cluster_index; 1306 uint64_t offset; 1307 1308 entry = be64_to_cpu(l2_slice[j]); 1309 old_entry = entry; 1310 entry &= ~QCOW_OFLAG_COPIED; 1311 offset = entry & L2E_OFFSET_MASK; 1312 1313 switch (qcow2_get_cluster_type(bs, entry)) { 1314 case QCOW2_CLUSTER_COMPRESSED: 1315 nb_csectors = ((entry >> s->csize_shift) & 1316 s->csize_mask) + 1; 1317 if (addend != 0) { 1318 uint64_t coffset = (entry & s->cluster_offset_mask) 1319 & QCOW2_COMPRESSED_SECTOR_MASK; 1320 ret = update_refcount( 1321 bs, coffset, 1322 nb_csectors * QCOW2_COMPRESSED_SECTOR_SIZE, 1323 abs(addend), addend < 0, 1324 QCOW2_DISCARD_SNAPSHOT); 1325 if (ret < 0) { 1326 goto fail; 1327 } 1328 } 1329 /* compressed clusters are never modified */ 1330 refcount = 2; 1331 break; 1332 1333 case QCOW2_CLUSTER_NORMAL: 1334 case QCOW2_CLUSTER_ZERO_ALLOC: 1335 if (offset_into_cluster(s, offset)) { 1336 /* Here l2_index means table (not slice) index */ 1337 int l2_index = slice * s->l2_slice_size + j; 1338 qcow2_signal_corruption( 1339 bs, true, -1, -1, "Cluster " 1340 "allocation offset %#" PRIx64 1341 " unaligned (L2 offset: %#" 1342 PRIx64 ", L2 index: %#x)", 1343 offset, l2_offset, l2_index); 1344 ret = -EIO; 1345 goto fail; 1346 } 1347 1348 cluster_index = offset >> s->cluster_bits; 1349 assert(cluster_index); 1350 if (addend != 0) { 1351 ret = qcow2_update_cluster_refcount( 1352 bs, cluster_index, abs(addend), addend < 0, 1353 QCOW2_DISCARD_SNAPSHOT); 1354 if (ret < 0) { 1355 goto fail; 1356 } 1357 } 1358 1359 ret = qcow2_get_refcount(bs, cluster_index, &refcount); 1360 if (ret < 0) { 1361 goto fail; 1362 } 1363 break; 1364 1365 case QCOW2_CLUSTER_ZERO_PLAIN: 1366 case QCOW2_CLUSTER_UNALLOCATED: 1367 refcount = 0; 1368 break; 1369 1370 default: 1371 abort(); 1372 } 1373 1374 if (refcount == 1) { 1375 entry |= QCOW_OFLAG_COPIED; 1376 } 1377 if (entry != old_entry) { 1378 if (addend > 0) { 1379 qcow2_cache_set_dependency(bs, s->l2_table_cache, 1380 s->refcount_block_cache); 1381 } 1382 l2_slice[j] = cpu_to_be64(entry); 1383 qcow2_cache_entry_mark_dirty(s->l2_table_cache, 1384 l2_slice); 1385 } 1386 } 1387 1388 qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice); 1389 } 1390 1391 if (addend != 0) { 1392 ret = qcow2_update_cluster_refcount(bs, l2_offset >> 1393 s->cluster_bits, 1394 abs(addend), addend < 0, 1395 QCOW2_DISCARD_SNAPSHOT); 1396 if (ret < 0) { 1397 goto fail; 1398 } 1399 } 1400 ret = qcow2_get_refcount(bs, l2_offset >> s->cluster_bits, 1401 &refcount); 1402 if (ret < 0) { 1403 goto fail; 1404 } else if (refcount == 1) { 1405 l2_offset |= QCOW_OFLAG_COPIED; 1406 } 1407 if (l2_offset != old_l2_offset) { 1408 l1_table[i] = l2_offset; 1409 l1_modified = 1; 1410 } 1411 } 1412 } 1413 1414 ret = bdrv_flush(bs); 1415 fail: 1416 if (l2_slice) { 1417 qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice); 1418 } 1419 1420 s->cache_discards = false; 1421 qcow2_process_discards(bs, ret); 1422 1423 /* Update L1 only if it isn't deleted anyway (addend = -1) */ 1424 if (ret == 0 && addend >= 0 && l1_modified) { 1425 for (i = 0; i < l1_size; i++) { 1426 cpu_to_be64s(&l1_table[i]); 1427 } 1428 1429 ret = bdrv_pwrite_sync(bs->file, l1_table_offset, 1430 l1_table, l1_size2); 1431 1432 for (i = 0; i < l1_size; i++) { 1433 be64_to_cpus(&l1_table[i]); 1434 } 1435 } 1436 if (l1_allocated) 1437 g_free(l1_table); 1438 return ret; 1439 } 1440 1441 1442 1443 1444 /*********************************************************/ 1445 /* refcount checking functions */ 1446 1447 1448 static uint64_t refcount_array_byte_size(BDRVQcow2State *s, uint64_t entries) 1449 { 1450 /* This assertion holds because there is no way we can address more than 1451 * 2^(64 - 9) clusters at once (with cluster size 512 = 2^9, and because 1452 * offsets have to be representable in bytes); due to every cluster 1453 * corresponding to one refcount entry, we are well below that limit */ 1454 assert(entries < (UINT64_C(1) << (64 - 9))); 1455 1456 /* Thanks to the assertion this will not overflow, because 1457 * s->refcount_order < 7. 1458 * (note: x << s->refcount_order == x * s->refcount_bits) */ 1459 return DIV_ROUND_UP(entries << s->refcount_order, 8); 1460 } 1461 1462 /** 1463 * Reallocates *array so that it can hold new_size entries. *size must contain 1464 * the current number of entries in *array. If the reallocation fails, *array 1465 * and *size will not be modified and -errno will be returned. If the 1466 * reallocation is successful, *array will be set to the new buffer, *size 1467 * will be set to new_size and 0 will be returned. The size of the reallocated 1468 * refcount array buffer will be aligned to a cluster boundary, and the newly 1469 * allocated area will be zeroed. 1470 */ 1471 static int realloc_refcount_array(BDRVQcow2State *s, void **array, 1472 int64_t *size, int64_t new_size) 1473 { 1474 int64_t old_byte_size, new_byte_size; 1475 void *new_ptr; 1476 1477 /* Round to clusters so the array can be directly written to disk */ 1478 old_byte_size = size_to_clusters(s, refcount_array_byte_size(s, *size)) 1479 * s->cluster_size; 1480 new_byte_size = size_to_clusters(s, refcount_array_byte_size(s, new_size)) 1481 * s->cluster_size; 1482 1483 if (new_byte_size == old_byte_size) { 1484 *size = new_size; 1485 return 0; 1486 } 1487 1488 assert(new_byte_size > 0); 1489 1490 if (new_byte_size > SIZE_MAX) { 1491 return -ENOMEM; 1492 } 1493 1494 new_ptr = g_try_realloc(*array, new_byte_size); 1495 if (!new_ptr) { 1496 return -ENOMEM; 1497 } 1498 1499 if (new_byte_size > old_byte_size) { 1500 memset((char *)new_ptr + old_byte_size, 0, 1501 new_byte_size - old_byte_size); 1502 } 1503 1504 *array = new_ptr; 1505 *size = new_size; 1506 1507 return 0; 1508 } 1509 1510 /* 1511 * Increases the refcount for a range of clusters in a given refcount table. 1512 * This is used to construct a temporary refcount table out of L1 and L2 tables 1513 * which can be compared to the refcount table saved in the image. 1514 * 1515 * Modifies the number of errors in res. 1516 */ 1517 int qcow2_inc_refcounts_imrt(BlockDriverState *bs, BdrvCheckResult *res, 1518 void **refcount_table, 1519 int64_t *refcount_table_size, 1520 int64_t offset, int64_t size) 1521 { 1522 BDRVQcow2State *s = bs->opaque; 1523 uint64_t start, last, cluster_offset, k, refcount; 1524 int64_t file_len; 1525 int ret; 1526 1527 if (size <= 0) { 1528 return 0; 1529 } 1530 1531 file_len = bdrv_getlength(bs->file->bs); 1532 if (file_len < 0) { 1533 return file_len; 1534 } 1535 1536 /* 1537 * Last cluster of qcow2 image may be semi-allocated, so it may be OK to 1538 * reference some space after file end but it should be less than one 1539 * cluster. 1540 */ 1541 if (offset + size - file_len >= s->cluster_size) { 1542 fprintf(stderr, "ERROR: counting reference for region exceeding the " 1543 "end of the file by one cluster or more: offset 0x%" PRIx64 1544 " size 0x%" PRIx64 "\n", offset, size); 1545 res->corruptions++; 1546 return 0; 1547 } 1548 1549 start = start_of_cluster(s, offset); 1550 last = start_of_cluster(s, offset + size - 1); 1551 for(cluster_offset = start; cluster_offset <= last; 1552 cluster_offset += s->cluster_size) { 1553 k = cluster_offset >> s->cluster_bits; 1554 if (k >= *refcount_table_size) { 1555 ret = realloc_refcount_array(s, refcount_table, 1556 refcount_table_size, k + 1); 1557 if (ret < 0) { 1558 res->check_errors++; 1559 return ret; 1560 } 1561 } 1562 1563 refcount = s->get_refcount(*refcount_table, k); 1564 if (refcount == s->refcount_max) { 1565 fprintf(stderr, "ERROR: overflow cluster offset=0x%" PRIx64 1566 "\n", cluster_offset); 1567 fprintf(stderr, "Use qemu-img amend to increase the refcount entry " 1568 "width or qemu-img convert to create a clean copy if the " 1569 "image cannot be opened for writing\n"); 1570 res->corruptions++; 1571 continue; 1572 } 1573 s->set_refcount(*refcount_table, k, refcount + 1); 1574 } 1575 1576 return 0; 1577 } 1578 1579 /* Flags for check_refcounts_l1() and check_refcounts_l2() */ 1580 enum { 1581 CHECK_FRAG_INFO = 0x2, /* update BlockFragInfo counters */ 1582 }; 1583 1584 /* 1585 * Increases the refcount in the given refcount table for the all clusters 1586 * referenced in the L2 table. While doing so, performs some checks on L2 1587 * entries. 1588 * 1589 * Returns the number of errors found by the checks or -errno if an internal 1590 * error occurred. 1591 */ 1592 static int check_refcounts_l2(BlockDriverState *bs, BdrvCheckResult *res, 1593 void **refcount_table, 1594 int64_t *refcount_table_size, int64_t l2_offset, 1595 int flags, BdrvCheckMode fix, bool active) 1596 { 1597 BDRVQcow2State *s = bs->opaque; 1598 uint64_t *l2_table, l2_entry; 1599 uint64_t next_contiguous_offset = 0; 1600 int i, l2_size, nb_csectors, ret; 1601 1602 /* Read L2 table from disk */ 1603 l2_size = s->l2_size * sizeof(uint64_t); 1604 l2_table = g_malloc(l2_size); 1605 1606 ret = bdrv_pread(bs->file, l2_offset, l2_table, l2_size); 1607 if (ret < 0) { 1608 fprintf(stderr, "ERROR: I/O error in check_refcounts_l2\n"); 1609 res->check_errors++; 1610 goto fail; 1611 } 1612 1613 /* Do the actual checks */ 1614 for(i = 0; i < s->l2_size; i++) { 1615 l2_entry = be64_to_cpu(l2_table[i]); 1616 1617 switch (qcow2_get_cluster_type(bs, l2_entry)) { 1618 case QCOW2_CLUSTER_COMPRESSED: 1619 /* Compressed clusters don't have QCOW_OFLAG_COPIED */ 1620 if (l2_entry & QCOW_OFLAG_COPIED) { 1621 fprintf(stderr, "ERROR: coffset=0x%" PRIx64 ": " 1622 "copied flag must never be set for compressed " 1623 "clusters\n", l2_entry & s->cluster_offset_mask); 1624 l2_entry &= ~QCOW_OFLAG_COPIED; 1625 res->corruptions++; 1626 } 1627 1628 if (has_data_file(bs)) { 1629 fprintf(stderr, "ERROR compressed cluster %d with data file, " 1630 "entry=0x%" PRIx64 "\n", i, l2_entry); 1631 res->corruptions++; 1632 break; 1633 } 1634 1635 /* Mark cluster as used */ 1636 nb_csectors = ((l2_entry >> s->csize_shift) & 1637 s->csize_mask) + 1; 1638 l2_entry &= s->cluster_offset_mask; 1639 ret = qcow2_inc_refcounts_imrt( 1640 bs, res, refcount_table, refcount_table_size, 1641 l2_entry & QCOW2_COMPRESSED_SECTOR_MASK, 1642 nb_csectors * QCOW2_COMPRESSED_SECTOR_SIZE); 1643 if (ret < 0) { 1644 goto fail; 1645 } 1646 1647 if (flags & CHECK_FRAG_INFO) { 1648 res->bfi.allocated_clusters++; 1649 res->bfi.compressed_clusters++; 1650 1651 /* Compressed clusters are fragmented by nature. Since they 1652 * take up sub-sector space but we only have sector granularity 1653 * I/O we need to re-read the same sectors even for adjacent 1654 * compressed clusters. 1655 */ 1656 res->bfi.fragmented_clusters++; 1657 } 1658 break; 1659 1660 case QCOW2_CLUSTER_ZERO_ALLOC: 1661 case QCOW2_CLUSTER_NORMAL: 1662 { 1663 uint64_t offset = l2_entry & L2E_OFFSET_MASK; 1664 1665 /* Correct offsets are cluster aligned */ 1666 if (offset_into_cluster(s, offset)) { 1667 res->corruptions++; 1668 1669 if (qcow2_get_cluster_type(bs, l2_entry) == 1670 QCOW2_CLUSTER_ZERO_ALLOC) 1671 { 1672 fprintf(stderr, "%s offset=%" PRIx64 ": Preallocated zero " 1673 "cluster is not properly aligned; L2 entry " 1674 "corrupted.\n", 1675 fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR", 1676 offset); 1677 if (fix & BDRV_FIX_ERRORS) { 1678 uint64_t l2e_offset = 1679 l2_offset + (uint64_t)i * sizeof(uint64_t); 1680 int ign = active ? QCOW2_OL_ACTIVE_L2 : 1681 QCOW2_OL_INACTIVE_L2; 1682 1683 l2_entry = QCOW_OFLAG_ZERO; 1684 l2_table[i] = cpu_to_be64(l2_entry); 1685 ret = qcow2_pre_write_overlap_check(bs, ign, 1686 l2e_offset, sizeof(uint64_t), false); 1687 if (ret < 0) { 1688 fprintf(stderr, "ERROR: Overlap check failed\n"); 1689 res->check_errors++; 1690 /* Something is seriously wrong, so abort checking 1691 * this L2 table */ 1692 goto fail; 1693 } 1694 1695 ret = bdrv_pwrite_sync(bs->file, l2e_offset, 1696 &l2_table[i], sizeof(uint64_t)); 1697 if (ret < 0) { 1698 fprintf(stderr, "ERROR: Failed to overwrite L2 " 1699 "table entry: %s\n", strerror(-ret)); 1700 res->check_errors++; 1701 /* Do not abort, continue checking the rest of this 1702 * L2 table's entries */ 1703 } else { 1704 res->corruptions--; 1705 res->corruptions_fixed++; 1706 /* Skip marking the cluster as used 1707 * (it is unused now) */ 1708 continue; 1709 } 1710 } 1711 } else { 1712 fprintf(stderr, "ERROR offset=%" PRIx64 ": Data cluster is " 1713 "not properly aligned; L2 entry corrupted.\n", offset); 1714 } 1715 } 1716 1717 if (flags & CHECK_FRAG_INFO) { 1718 res->bfi.allocated_clusters++; 1719 if (next_contiguous_offset && 1720 offset != next_contiguous_offset) { 1721 res->bfi.fragmented_clusters++; 1722 } 1723 next_contiguous_offset = offset + s->cluster_size; 1724 } 1725 1726 /* Mark cluster as used */ 1727 if (!has_data_file(bs)) { 1728 ret = qcow2_inc_refcounts_imrt(bs, res, refcount_table, 1729 refcount_table_size, 1730 offset, s->cluster_size); 1731 if (ret < 0) { 1732 goto fail; 1733 } 1734 } 1735 break; 1736 } 1737 1738 case QCOW2_CLUSTER_ZERO_PLAIN: 1739 case QCOW2_CLUSTER_UNALLOCATED: 1740 break; 1741 1742 default: 1743 abort(); 1744 } 1745 } 1746 1747 g_free(l2_table); 1748 return 0; 1749 1750 fail: 1751 g_free(l2_table); 1752 return ret; 1753 } 1754 1755 /* 1756 * Increases the refcount for the L1 table, its L2 tables and all referenced 1757 * clusters in the given refcount table. While doing so, performs some checks 1758 * on L1 and L2 entries. 1759 * 1760 * Returns the number of errors found by the checks or -errno if an internal 1761 * error occurred. 1762 */ 1763 static int check_refcounts_l1(BlockDriverState *bs, 1764 BdrvCheckResult *res, 1765 void **refcount_table, 1766 int64_t *refcount_table_size, 1767 int64_t l1_table_offset, int l1_size, 1768 int flags, BdrvCheckMode fix, bool active) 1769 { 1770 BDRVQcow2State *s = bs->opaque; 1771 uint64_t *l1_table = NULL, l2_offset, l1_size2; 1772 int i, ret; 1773 1774 l1_size2 = l1_size * sizeof(uint64_t); 1775 1776 /* Mark L1 table as used */ 1777 ret = qcow2_inc_refcounts_imrt(bs, res, refcount_table, refcount_table_size, 1778 l1_table_offset, l1_size2); 1779 if (ret < 0) { 1780 goto fail; 1781 } 1782 1783 /* Read L1 table entries from disk */ 1784 if (l1_size2 > 0) { 1785 l1_table = g_try_malloc(l1_size2); 1786 if (l1_table == NULL) { 1787 ret = -ENOMEM; 1788 res->check_errors++; 1789 goto fail; 1790 } 1791 ret = bdrv_pread(bs->file, l1_table_offset, l1_table, l1_size2); 1792 if (ret < 0) { 1793 fprintf(stderr, "ERROR: I/O error in check_refcounts_l1\n"); 1794 res->check_errors++; 1795 goto fail; 1796 } 1797 for(i = 0;i < l1_size; i++) 1798 be64_to_cpus(&l1_table[i]); 1799 } 1800 1801 /* Do the actual checks */ 1802 for(i = 0; i < l1_size; i++) { 1803 l2_offset = l1_table[i]; 1804 if (l2_offset) { 1805 /* Mark L2 table as used */ 1806 l2_offset &= L1E_OFFSET_MASK; 1807 ret = qcow2_inc_refcounts_imrt(bs, res, 1808 refcount_table, refcount_table_size, 1809 l2_offset, s->cluster_size); 1810 if (ret < 0) { 1811 goto fail; 1812 } 1813 1814 /* L2 tables are cluster aligned */ 1815 if (offset_into_cluster(s, l2_offset)) { 1816 fprintf(stderr, "ERROR l2_offset=%" PRIx64 ": Table is not " 1817 "cluster aligned; L1 entry corrupted\n", l2_offset); 1818 res->corruptions++; 1819 } 1820 1821 /* Process and check L2 entries */ 1822 ret = check_refcounts_l2(bs, res, refcount_table, 1823 refcount_table_size, l2_offset, flags, 1824 fix, active); 1825 if (ret < 0) { 1826 goto fail; 1827 } 1828 } 1829 } 1830 g_free(l1_table); 1831 return 0; 1832 1833 fail: 1834 g_free(l1_table); 1835 return ret; 1836 } 1837 1838 /* 1839 * Checks the OFLAG_COPIED flag for all L1 and L2 entries. 1840 * 1841 * This function does not print an error message nor does it increment 1842 * check_errors if qcow2_get_refcount fails (this is because such an error will 1843 * have been already detected and sufficiently signaled by the calling function 1844 * (qcow2_check_refcounts) by the time this function is called). 1845 */ 1846 static int check_oflag_copied(BlockDriverState *bs, BdrvCheckResult *res, 1847 BdrvCheckMode fix) 1848 { 1849 BDRVQcow2State *s = bs->opaque; 1850 uint64_t *l2_table = qemu_blockalign(bs, s->cluster_size); 1851 int ret; 1852 uint64_t refcount; 1853 int i, j; 1854 bool repair; 1855 1856 if (fix & BDRV_FIX_ERRORS) { 1857 /* Always repair */ 1858 repair = true; 1859 } else if (fix & BDRV_FIX_LEAKS) { 1860 /* Repair only if that seems safe: This function is always 1861 * called after the refcounts have been fixed, so the refcount 1862 * is accurate if that repair was successful */ 1863 repair = !res->check_errors && !res->corruptions && !res->leaks; 1864 } else { 1865 repair = false; 1866 } 1867 1868 for (i = 0; i < s->l1_size; i++) { 1869 uint64_t l1_entry = s->l1_table[i]; 1870 uint64_t l2_offset = l1_entry & L1E_OFFSET_MASK; 1871 int l2_dirty = 0; 1872 1873 if (!l2_offset) { 1874 continue; 1875 } 1876 1877 ret = qcow2_get_refcount(bs, l2_offset >> s->cluster_bits, 1878 &refcount); 1879 if (ret < 0) { 1880 /* don't print message nor increment check_errors */ 1881 continue; 1882 } 1883 if ((refcount == 1) != ((l1_entry & QCOW_OFLAG_COPIED) != 0)) { 1884 res->corruptions++; 1885 fprintf(stderr, "%s OFLAG_COPIED L2 cluster: l1_index=%d " 1886 "l1_entry=%" PRIx64 " refcount=%" PRIu64 "\n", 1887 repair ? "Repairing" : "ERROR", i, l1_entry, refcount); 1888 if (repair) { 1889 s->l1_table[i] = refcount == 1 1890 ? l1_entry | QCOW_OFLAG_COPIED 1891 : l1_entry & ~QCOW_OFLAG_COPIED; 1892 ret = qcow2_write_l1_entry(bs, i); 1893 if (ret < 0) { 1894 res->check_errors++; 1895 goto fail; 1896 } 1897 res->corruptions--; 1898 res->corruptions_fixed++; 1899 } 1900 } 1901 1902 ret = bdrv_pread(bs->file, l2_offset, l2_table, 1903 s->l2_size * sizeof(uint64_t)); 1904 if (ret < 0) { 1905 fprintf(stderr, "ERROR: Could not read L2 table: %s\n", 1906 strerror(-ret)); 1907 res->check_errors++; 1908 goto fail; 1909 } 1910 1911 for (j = 0; j < s->l2_size; j++) { 1912 uint64_t l2_entry = be64_to_cpu(l2_table[j]); 1913 uint64_t data_offset = l2_entry & L2E_OFFSET_MASK; 1914 QCow2ClusterType cluster_type = qcow2_get_cluster_type(bs, l2_entry); 1915 1916 if (cluster_type == QCOW2_CLUSTER_NORMAL || 1917 cluster_type == QCOW2_CLUSTER_ZERO_ALLOC) { 1918 if (has_data_file(bs)) { 1919 refcount = 1; 1920 } else { 1921 ret = qcow2_get_refcount(bs, 1922 data_offset >> s->cluster_bits, 1923 &refcount); 1924 if (ret < 0) { 1925 /* don't print message nor increment check_errors */ 1926 continue; 1927 } 1928 } 1929 if ((refcount == 1) != ((l2_entry & QCOW_OFLAG_COPIED) != 0)) { 1930 res->corruptions++; 1931 fprintf(stderr, "%s OFLAG_COPIED data cluster: " 1932 "l2_entry=%" PRIx64 " refcount=%" PRIu64 "\n", 1933 repair ? "Repairing" : "ERROR", l2_entry, refcount); 1934 if (repair) { 1935 l2_table[j] = cpu_to_be64(refcount == 1 1936 ? l2_entry | QCOW_OFLAG_COPIED 1937 : l2_entry & ~QCOW_OFLAG_COPIED); 1938 l2_dirty++; 1939 } 1940 } 1941 } 1942 } 1943 1944 if (l2_dirty > 0) { 1945 ret = qcow2_pre_write_overlap_check(bs, QCOW2_OL_ACTIVE_L2, 1946 l2_offset, s->cluster_size, 1947 false); 1948 if (ret < 0) { 1949 fprintf(stderr, "ERROR: Could not write L2 table; metadata " 1950 "overlap check failed: %s\n", strerror(-ret)); 1951 res->check_errors++; 1952 goto fail; 1953 } 1954 1955 ret = bdrv_pwrite(bs->file, l2_offset, l2_table, 1956 s->cluster_size); 1957 if (ret < 0) { 1958 fprintf(stderr, "ERROR: Could not write L2 table: %s\n", 1959 strerror(-ret)); 1960 res->check_errors++; 1961 goto fail; 1962 } 1963 res->corruptions -= l2_dirty; 1964 res->corruptions_fixed += l2_dirty; 1965 } 1966 } 1967 1968 ret = 0; 1969 1970 fail: 1971 qemu_vfree(l2_table); 1972 return ret; 1973 } 1974 1975 /* 1976 * Checks consistency of refblocks and accounts for each refblock in 1977 * *refcount_table. 1978 */ 1979 static int check_refblocks(BlockDriverState *bs, BdrvCheckResult *res, 1980 BdrvCheckMode fix, bool *rebuild, 1981 void **refcount_table, int64_t *nb_clusters) 1982 { 1983 BDRVQcow2State *s = bs->opaque; 1984 int64_t i, size; 1985 int ret; 1986 1987 for(i = 0; i < s->refcount_table_size; i++) { 1988 uint64_t offset, cluster; 1989 offset = s->refcount_table[i]; 1990 cluster = offset >> s->cluster_bits; 1991 1992 /* Refcount blocks are cluster aligned */ 1993 if (offset_into_cluster(s, offset)) { 1994 fprintf(stderr, "ERROR refcount block %" PRId64 " is not " 1995 "cluster aligned; refcount table entry corrupted\n", i); 1996 res->corruptions++; 1997 *rebuild = true; 1998 continue; 1999 } 2000 2001 if (cluster >= *nb_clusters) { 2002 res->corruptions++; 2003 fprintf(stderr, "%s refcount block %" PRId64 " is outside image\n", 2004 fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR", i); 2005 2006 if (fix & BDRV_FIX_ERRORS) { 2007 int64_t new_nb_clusters; 2008 Error *local_err = NULL; 2009 2010 if (offset > INT64_MAX - s->cluster_size) { 2011 ret = -EINVAL; 2012 goto resize_fail; 2013 } 2014 2015 ret = bdrv_truncate(bs->file, offset + s->cluster_size, 2016 PREALLOC_MODE_OFF, &local_err); 2017 if (ret < 0) { 2018 error_report_err(local_err); 2019 goto resize_fail; 2020 } 2021 size = bdrv_getlength(bs->file->bs); 2022 if (size < 0) { 2023 ret = size; 2024 goto resize_fail; 2025 } 2026 2027 new_nb_clusters = size_to_clusters(s, size); 2028 assert(new_nb_clusters >= *nb_clusters); 2029 2030 ret = realloc_refcount_array(s, refcount_table, 2031 nb_clusters, new_nb_clusters); 2032 if (ret < 0) { 2033 res->check_errors++; 2034 return ret; 2035 } 2036 2037 if (cluster >= *nb_clusters) { 2038 ret = -EINVAL; 2039 goto resize_fail; 2040 } 2041 2042 res->corruptions--; 2043 res->corruptions_fixed++; 2044 ret = qcow2_inc_refcounts_imrt(bs, res, 2045 refcount_table, nb_clusters, 2046 offset, s->cluster_size); 2047 if (ret < 0) { 2048 return ret; 2049 } 2050 /* No need to check whether the refcount is now greater than 1: 2051 * This area was just allocated and zeroed, so it can only be 2052 * exactly 1 after qcow2_inc_refcounts_imrt() */ 2053 continue; 2054 2055 resize_fail: 2056 *rebuild = true; 2057 fprintf(stderr, "ERROR could not resize image: %s\n", 2058 strerror(-ret)); 2059 } 2060 continue; 2061 } 2062 2063 if (offset != 0) { 2064 ret = qcow2_inc_refcounts_imrt(bs, res, refcount_table, nb_clusters, 2065 offset, s->cluster_size); 2066 if (ret < 0) { 2067 return ret; 2068 } 2069 if (s->get_refcount(*refcount_table, cluster) != 1) { 2070 fprintf(stderr, "ERROR refcount block %" PRId64 2071 " refcount=%" PRIu64 "\n", i, 2072 s->get_refcount(*refcount_table, cluster)); 2073 res->corruptions++; 2074 *rebuild = true; 2075 } 2076 } 2077 } 2078 2079 return 0; 2080 } 2081 2082 /* 2083 * Calculates an in-memory refcount table. 2084 */ 2085 static int calculate_refcounts(BlockDriverState *bs, BdrvCheckResult *res, 2086 BdrvCheckMode fix, bool *rebuild, 2087 void **refcount_table, int64_t *nb_clusters) 2088 { 2089 BDRVQcow2State *s = bs->opaque; 2090 int64_t i; 2091 QCowSnapshot *sn; 2092 int ret; 2093 2094 if (!*refcount_table) { 2095 int64_t old_size = 0; 2096 ret = realloc_refcount_array(s, refcount_table, 2097 &old_size, *nb_clusters); 2098 if (ret < 0) { 2099 res->check_errors++; 2100 return ret; 2101 } 2102 } 2103 2104 /* header */ 2105 ret = qcow2_inc_refcounts_imrt(bs, res, refcount_table, nb_clusters, 2106 0, s->cluster_size); 2107 if (ret < 0) { 2108 return ret; 2109 } 2110 2111 /* current L1 table */ 2112 ret = check_refcounts_l1(bs, res, refcount_table, nb_clusters, 2113 s->l1_table_offset, s->l1_size, CHECK_FRAG_INFO, 2114 fix, true); 2115 if (ret < 0) { 2116 return ret; 2117 } 2118 2119 /* snapshots */ 2120 if (has_data_file(bs) && s->nb_snapshots) { 2121 fprintf(stderr, "ERROR %d snapshots in image with data file\n", 2122 s->nb_snapshots); 2123 res->corruptions++; 2124 } 2125 2126 for (i = 0; i < s->nb_snapshots; i++) { 2127 sn = s->snapshots + i; 2128 if (offset_into_cluster(s, sn->l1_table_offset)) { 2129 fprintf(stderr, "ERROR snapshot %s (%s) l1_offset=%#" PRIx64 ": " 2130 "L1 table is not cluster aligned; snapshot table entry " 2131 "corrupted\n", sn->id_str, sn->name, sn->l1_table_offset); 2132 res->corruptions++; 2133 continue; 2134 } 2135 if (sn->l1_size > QCOW_MAX_L1_SIZE / sizeof(uint64_t)) { 2136 fprintf(stderr, "ERROR snapshot %s (%s) l1_size=%#" PRIx32 ": " 2137 "L1 table is too large; snapshot table entry corrupted\n", 2138 sn->id_str, sn->name, sn->l1_size); 2139 res->corruptions++; 2140 continue; 2141 } 2142 ret = check_refcounts_l1(bs, res, refcount_table, nb_clusters, 2143 sn->l1_table_offset, sn->l1_size, 0, fix, 2144 false); 2145 if (ret < 0) { 2146 return ret; 2147 } 2148 } 2149 ret = qcow2_inc_refcounts_imrt(bs, res, refcount_table, nb_clusters, 2150 s->snapshots_offset, s->snapshots_size); 2151 if (ret < 0) { 2152 return ret; 2153 } 2154 2155 /* refcount data */ 2156 ret = qcow2_inc_refcounts_imrt(bs, res, refcount_table, nb_clusters, 2157 s->refcount_table_offset, 2158 s->refcount_table_size * sizeof(uint64_t)); 2159 if (ret < 0) { 2160 return ret; 2161 } 2162 2163 /* encryption */ 2164 if (s->crypto_header.length) { 2165 ret = qcow2_inc_refcounts_imrt(bs, res, refcount_table, nb_clusters, 2166 s->crypto_header.offset, 2167 s->crypto_header.length); 2168 if (ret < 0) { 2169 return ret; 2170 } 2171 } 2172 2173 /* bitmaps */ 2174 ret = qcow2_check_bitmaps_refcounts(bs, res, refcount_table, nb_clusters); 2175 if (ret < 0) { 2176 return ret; 2177 } 2178 2179 return check_refblocks(bs, res, fix, rebuild, refcount_table, nb_clusters); 2180 } 2181 2182 /* 2183 * Compares the actual reference count for each cluster in the image against the 2184 * refcount as reported by the refcount structures on-disk. 2185 */ 2186 static void compare_refcounts(BlockDriverState *bs, BdrvCheckResult *res, 2187 BdrvCheckMode fix, bool *rebuild, 2188 int64_t *highest_cluster, 2189 void *refcount_table, int64_t nb_clusters) 2190 { 2191 BDRVQcow2State *s = bs->opaque; 2192 int64_t i; 2193 uint64_t refcount1, refcount2; 2194 int ret; 2195 2196 for (i = 0, *highest_cluster = 0; i < nb_clusters; i++) { 2197 ret = qcow2_get_refcount(bs, i, &refcount1); 2198 if (ret < 0) { 2199 fprintf(stderr, "Can't get refcount for cluster %" PRId64 ": %s\n", 2200 i, strerror(-ret)); 2201 res->check_errors++; 2202 continue; 2203 } 2204 2205 refcount2 = s->get_refcount(refcount_table, i); 2206 2207 if (refcount1 > 0 || refcount2 > 0) { 2208 *highest_cluster = i; 2209 } 2210 2211 if (refcount1 != refcount2) { 2212 /* Check if we're allowed to fix the mismatch */ 2213 int *num_fixed = NULL; 2214 if (refcount1 == 0) { 2215 *rebuild = true; 2216 } else if (refcount1 > refcount2 && (fix & BDRV_FIX_LEAKS)) { 2217 num_fixed = &res->leaks_fixed; 2218 } else if (refcount1 < refcount2 && (fix & BDRV_FIX_ERRORS)) { 2219 num_fixed = &res->corruptions_fixed; 2220 } 2221 2222 fprintf(stderr, "%s cluster %" PRId64 " refcount=%" PRIu64 2223 " reference=%" PRIu64 "\n", 2224 num_fixed != NULL ? "Repairing" : 2225 refcount1 < refcount2 ? "ERROR" : 2226 "Leaked", 2227 i, refcount1, refcount2); 2228 2229 if (num_fixed) { 2230 ret = update_refcount(bs, i << s->cluster_bits, 1, 2231 refcount_diff(refcount1, refcount2), 2232 refcount1 > refcount2, 2233 QCOW2_DISCARD_ALWAYS); 2234 if (ret >= 0) { 2235 (*num_fixed)++; 2236 continue; 2237 } 2238 } 2239 2240 /* And if we couldn't, print an error */ 2241 if (refcount1 < refcount2) { 2242 res->corruptions++; 2243 } else { 2244 res->leaks++; 2245 } 2246 } 2247 } 2248 } 2249 2250 /* 2251 * Allocates clusters using an in-memory refcount table (IMRT) in contrast to 2252 * the on-disk refcount structures. 2253 * 2254 * On input, *first_free_cluster tells where to start looking, and need not 2255 * actually be a free cluster; the returned offset will not be before that 2256 * cluster. On output, *first_free_cluster points to the first gap found, even 2257 * if that gap was too small to be used as the returned offset. 2258 * 2259 * Note that *first_free_cluster is a cluster index whereas the return value is 2260 * an offset. 2261 */ 2262 static int64_t alloc_clusters_imrt(BlockDriverState *bs, 2263 int cluster_count, 2264 void **refcount_table, 2265 int64_t *imrt_nb_clusters, 2266 int64_t *first_free_cluster) 2267 { 2268 BDRVQcow2State *s = bs->opaque; 2269 int64_t cluster = *first_free_cluster, i; 2270 bool first_gap = true; 2271 int contiguous_free_clusters; 2272 int ret; 2273 2274 /* Starting at *first_free_cluster, find a range of at least cluster_count 2275 * continuously free clusters */ 2276 for (contiguous_free_clusters = 0; 2277 cluster < *imrt_nb_clusters && 2278 contiguous_free_clusters < cluster_count; 2279 cluster++) 2280 { 2281 if (!s->get_refcount(*refcount_table, cluster)) { 2282 contiguous_free_clusters++; 2283 if (first_gap) { 2284 /* If this is the first free cluster found, update 2285 * *first_free_cluster accordingly */ 2286 *first_free_cluster = cluster; 2287 first_gap = false; 2288 } 2289 } else if (contiguous_free_clusters) { 2290 contiguous_free_clusters = 0; 2291 } 2292 } 2293 2294 /* If contiguous_free_clusters is greater than zero, it contains the number 2295 * of continuously free clusters until the current cluster; the first free 2296 * cluster in the current "gap" is therefore 2297 * cluster - contiguous_free_clusters */ 2298 2299 /* If no such range could be found, grow the in-memory refcount table 2300 * accordingly to append free clusters at the end of the image */ 2301 if (contiguous_free_clusters < cluster_count) { 2302 /* contiguous_free_clusters clusters are already empty at the image end; 2303 * we need cluster_count clusters; therefore, we have to allocate 2304 * cluster_count - contiguous_free_clusters new clusters at the end of 2305 * the image (which is the current value of cluster; note that cluster 2306 * may exceed old_imrt_nb_clusters if *first_free_cluster pointed beyond 2307 * the image end) */ 2308 ret = realloc_refcount_array(s, refcount_table, imrt_nb_clusters, 2309 cluster + cluster_count 2310 - contiguous_free_clusters); 2311 if (ret < 0) { 2312 return ret; 2313 } 2314 } 2315 2316 /* Go back to the first free cluster */ 2317 cluster -= contiguous_free_clusters; 2318 for (i = 0; i < cluster_count; i++) { 2319 s->set_refcount(*refcount_table, cluster + i, 1); 2320 } 2321 2322 return cluster << s->cluster_bits; 2323 } 2324 2325 /* 2326 * Creates a new refcount structure based solely on the in-memory information 2327 * given through *refcount_table. All necessary allocations will be reflected 2328 * in that array. 2329 * 2330 * On success, the old refcount structure is leaked (it will be covered by the 2331 * new refcount structure). 2332 */ 2333 static int rebuild_refcount_structure(BlockDriverState *bs, 2334 BdrvCheckResult *res, 2335 void **refcount_table, 2336 int64_t *nb_clusters) 2337 { 2338 BDRVQcow2State *s = bs->opaque; 2339 int64_t first_free_cluster = 0, reftable_offset = -1, cluster = 0; 2340 int64_t refblock_offset, refblock_start, refblock_index; 2341 uint32_t reftable_size = 0; 2342 uint64_t *on_disk_reftable = NULL; 2343 void *on_disk_refblock; 2344 int ret = 0; 2345 struct { 2346 uint64_t reftable_offset; 2347 uint32_t reftable_clusters; 2348 } QEMU_PACKED reftable_offset_and_clusters; 2349 2350 qcow2_cache_empty(bs, s->refcount_block_cache); 2351 2352 write_refblocks: 2353 for (; cluster < *nb_clusters; cluster++) { 2354 if (!s->get_refcount(*refcount_table, cluster)) { 2355 continue; 2356 } 2357 2358 refblock_index = cluster >> s->refcount_block_bits; 2359 refblock_start = refblock_index << s->refcount_block_bits; 2360 2361 /* Don't allocate a cluster in a refblock already written to disk */ 2362 if (first_free_cluster < refblock_start) { 2363 first_free_cluster = refblock_start; 2364 } 2365 refblock_offset = alloc_clusters_imrt(bs, 1, refcount_table, 2366 nb_clusters, &first_free_cluster); 2367 if (refblock_offset < 0) { 2368 fprintf(stderr, "ERROR allocating refblock: %s\n", 2369 strerror(-refblock_offset)); 2370 res->check_errors++; 2371 ret = refblock_offset; 2372 goto fail; 2373 } 2374 2375 if (reftable_size <= refblock_index) { 2376 uint32_t old_reftable_size = reftable_size; 2377 uint64_t *new_on_disk_reftable; 2378 2379 reftable_size = ROUND_UP((refblock_index + 1) * sizeof(uint64_t), 2380 s->cluster_size) / sizeof(uint64_t); 2381 new_on_disk_reftable = g_try_realloc(on_disk_reftable, 2382 reftable_size * 2383 sizeof(uint64_t)); 2384 if (!new_on_disk_reftable) { 2385 res->check_errors++; 2386 ret = -ENOMEM; 2387 goto fail; 2388 } 2389 on_disk_reftable = new_on_disk_reftable; 2390 2391 memset(on_disk_reftable + old_reftable_size, 0, 2392 (reftable_size - old_reftable_size) * sizeof(uint64_t)); 2393 2394 /* The offset we have for the reftable is now no longer valid; 2395 * this will leak that range, but we can easily fix that by running 2396 * a leak-fixing check after this rebuild operation */ 2397 reftable_offset = -1; 2398 } else { 2399 assert(on_disk_reftable); 2400 } 2401 on_disk_reftable[refblock_index] = refblock_offset; 2402 2403 /* If this is apparently the last refblock (for now), try to squeeze the 2404 * reftable in */ 2405 if (refblock_index == (*nb_clusters - 1) >> s->refcount_block_bits && 2406 reftable_offset < 0) 2407 { 2408 uint64_t reftable_clusters = size_to_clusters(s, reftable_size * 2409 sizeof(uint64_t)); 2410 reftable_offset = alloc_clusters_imrt(bs, reftable_clusters, 2411 refcount_table, nb_clusters, 2412 &first_free_cluster); 2413 if (reftable_offset < 0) { 2414 fprintf(stderr, "ERROR allocating reftable: %s\n", 2415 strerror(-reftable_offset)); 2416 res->check_errors++; 2417 ret = reftable_offset; 2418 goto fail; 2419 } 2420 } 2421 2422 ret = qcow2_pre_write_overlap_check(bs, 0, refblock_offset, 2423 s->cluster_size, false); 2424 if (ret < 0) { 2425 fprintf(stderr, "ERROR writing refblock: %s\n", strerror(-ret)); 2426 goto fail; 2427 } 2428 2429 /* The size of *refcount_table is always cluster-aligned, therefore the 2430 * write operation will not overflow */ 2431 on_disk_refblock = (void *)((char *) *refcount_table + 2432 refblock_index * s->cluster_size); 2433 2434 ret = bdrv_pwrite(bs->file, refblock_offset, on_disk_refblock, 2435 s->cluster_size); 2436 if (ret < 0) { 2437 fprintf(stderr, "ERROR writing refblock: %s\n", strerror(-ret)); 2438 goto fail; 2439 } 2440 2441 /* Go to the end of this refblock */ 2442 cluster = refblock_start + s->refcount_block_size - 1; 2443 } 2444 2445 if (reftable_offset < 0) { 2446 uint64_t post_refblock_start, reftable_clusters; 2447 2448 post_refblock_start = ROUND_UP(*nb_clusters, s->refcount_block_size); 2449 reftable_clusters = size_to_clusters(s, 2450 reftable_size * sizeof(uint64_t)); 2451 /* Not pretty but simple */ 2452 if (first_free_cluster < post_refblock_start) { 2453 first_free_cluster = post_refblock_start; 2454 } 2455 reftable_offset = alloc_clusters_imrt(bs, reftable_clusters, 2456 refcount_table, nb_clusters, 2457 &first_free_cluster); 2458 if (reftable_offset < 0) { 2459 fprintf(stderr, "ERROR allocating reftable: %s\n", 2460 strerror(-reftable_offset)); 2461 res->check_errors++; 2462 ret = reftable_offset; 2463 goto fail; 2464 } 2465 2466 goto write_refblocks; 2467 } 2468 2469 for (refblock_index = 0; refblock_index < reftable_size; refblock_index++) { 2470 cpu_to_be64s(&on_disk_reftable[refblock_index]); 2471 } 2472 2473 ret = qcow2_pre_write_overlap_check(bs, 0, reftable_offset, 2474 reftable_size * sizeof(uint64_t), 2475 false); 2476 if (ret < 0) { 2477 fprintf(stderr, "ERROR writing reftable: %s\n", strerror(-ret)); 2478 goto fail; 2479 } 2480 2481 assert(reftable_size < INT_MAX / sizeof(uint64_t)); 2482 ret = bdrv_pwrite(bs->file, reftable_offset, on_disk_reftable, 2483 reftable_size * sizeof(uint64_t)); 2484 if (ret < 0) { 2485 fprintf(stderr, "ERROR writing reftable: %s\n", strerror(-ret)); 2486 goto fail; 2487 } 2488 2489 /* Enter new reftable into the image header */ 2490 reftable_offset_and_clusters.reftable_offset = cpu_to_be64(reftable_offset); 2491 reftable_offset_and_clusters.reftable_clusters = 2492 cpu_to_be32(size_to_clusters(s, reftable_size * sizeof(uint64_t))); 2493 ret = bdrv_pwrite_sync(bs->file, 2494 offsetof(QCowHeader, refcount_table_offset), 2495 &reftable_offset_and_clusters, 2496 sizeof(reftable_offset_and_clusters)); 2497 if (ret < 0) { 2498 fprintf(stderr, "ERROR setting reftable: %s\n", strerror(-ret)); 2499 goto fail; 2500 } 2501 2502 for (refblock_index = 0; refblock_index < reftable_size; refblock_index++) { 2503 be64_to_cpus(&on_disk_reftable[refblock_index]); 2504 } 2505 s->refcount_table = on_disk_reftable; 2506 s->refcount_table_offset = reftable_offset; 2507 s->refcount_table_size = reftable_size; 2508 update_max_refcount_table_index(s); 2509 2510 return 0; 2511 2512 fail: 2513 g_free(on_disk_reftable); 2514 return ret; 2515 } 2516 2517 /* 2518 * Checks an image for refcount consistency. 2519 * 2520 * Returns 0 if no errors are found, the number of errors in case the image is 2521 * detected as corrupted, and -errno when an internal error occurred. 2522 */ 2523 int qcow2_check_refcounts(BlockDriverState *bs, BdrvCheckResult *res, 2524 BdrvCheckMode fix) 2525 { 2526 BDRVQcow2State *s = bs->opaque; 2527 BdrvCheckResult pre_compare_res; 2528 int64_t size, highest_cluster, nb_clusters; 2529 void *refcount_table = NULL; 2530 bool rebuild = false; 2531 int ret; 2532 2533 size = bdrv_getlength(bs->file->bs); 2534 if (size < 0) { 2535 res->check_errors++; 2536 return size; 2537 } 2538 2539 nb_clusters = size_to_clusters(s, size); 2540 if (nb_clusters > INT_MAX) { 2541 res->check_errors++; 2542 return -EFBIG; 2543 } 2544 2545 res->bfi.total_clusters = 2546 size_to_clusters(s, bs->total_sectors * BDRV_SECTOR_SIZE); 2547 2548 ret = calculate_refcounts(bs, res, fix, &rebuild, &refcount_table, 2549 &nb_clusters); 2550 if (ret < 0) { 2551 goto fail; 2552 } 2553 2554 /* In case we don't need to rebuild the refcount structure (but want to fix 2555 * something), this function is immediately called again, in which case the 2556 * result should be ignored */ 2557 pre_compare_res = *res; 2558 compare_refcounts(bs, res, 0, &rebuild, &highest_cluster, refcount_table, 2559 nb_clusters); 2560 2561 if (rebuild && (fix & BDRV_FIX_ERRORS)) { 2562 BdrvCheckResult old_res = *res; 2563 int fresh_leaks = 0; 2564 2565 fprintf(stderr, "Rebuilding refcount structure\n"); 2566 ret = rebuild_refcount_structure(bs, res, &refcount_table, 2567 &nb_clusters); 2568 if (ret < 0) { 2569 goto fail; 2570 } 2571 2572 res->corruptions = 0; 2573 res->leaks = 0; 2574 2575 /* Because the old reftable has been exchanged for a new one the 2576 * references have to be recalculated */ 2577 rebuild = false; 2578 memset(refcount_table, 0, refcount_array_byte_size(s, nb_clusters)); 2579 ret = calculate_refcounts(bs, res, 0, &rebuild, &refcount_table, 2580 &nb_clusters); 2581 if (ret < 0) { 2582 goto fail; 2583 } 2584 2585 if (fix & BDRV_FIX_LEAKS) { 2586 /* The old refcount structures are now leaked, fix it; the result 2587 * can be ignored, aside from leaks which were introduced by 2588 * rebuild_refcount_structure() that could not be fixed */ 2589 BdrvCheckResult saved_res = *res; 2590 *res = (BdrvCheckResult){ 0 }; 2591 2592 compare_refcounts(bs, res, BDRV_FIX_LEAKS, &rebuild, 2593 &highest_cluster, refcount_table, nb_clusters); 2594 if (rebuild) { 2595 fprintf(stderr, "ERROR rebuilt refcount structure is still " 2596 "broken\n"); 2597 } 2598 2599 /* Any leaks accounted for here were introduced by 2600 * rebuild_refcount_structure() because that function has created a 2601 * new refcount structure from scratch */ 2602 fresh_leaks = res->leaks; 2603 *res = saved_res; 2604 } 2605 2606 if (res->corruptions < old_res.corruptions) { 2607 res->corruptions_fixed += old_res.corruptions - res->corruptions; 2608 } 2609 if (res->leaks < old_res.leaks) { 2610 res->leaks_fixed += old_res.leaks - res->leaks; 2611 } 2612 res->leaks += fresh_leaks; 2613 } else if (fix) { 2614 if (rebuild) { 2615 fprintf(stderr, "ERROR need to rebuild refcount structures\n"); 2616 res->check_errors++; 2617 ret = -EIO; 2618 goto fail; 2619 } 2620 2621 if (res->leaks || res->corruptions) { 2622 *res = pre_compare_res; 2623 compare_refcounts(bs, res, fix, &rebuild, &highest_cluster, 2624 refcount_table, nb_clusters); 2625 } 2626 } 2627 2628 /* check OFLAG_COPIED */ 2629 ret = check_oflag_copied(bs, res, fix); 2630 if (ret < 0) { 2631 goto fail; 2632 } 2633 2634 res->image_end_offset = (highest_cluster + 1) * s->cluster_size; 2635 ret = 0; 2636 2637 fail: 2638 g_free(refcount_table); 2639 2640 return ret; 2641 } 2642 2643 #define overlaps_with(ofs, sz) \ 2644 ranges_overlap(offset, size, ofs, sz) 2645 2646 /* 2647 * Checks if the given offset into the image file is actually free to use by 2648 * looking for overlaps with important metadata sections (L1/L2 tables etc.), 2649 * i.e. a sanity check without relying on the refcount tables. 2650 * 2651 * The ign parameter specifies what checks not to perform (being a bitmask of 2652 * QCow2MetadataOverlap values), i.e., what sections to ignore. 2653 * 2654 * Returns: 2655 * - 0 if writing to this offset will not affect the mentioned metadata 2656 * - a positive QCow2MetadataOverlap value indicating one overlapping section 2657 * - a negative value (-errno) indicating an error while performing a check, 2658 * e.g. when bdrv_read failed on QCOW2_OL_INACTIVE_L2 2659 */ 2660 int qcow2_check_metadata_overlap(BlockDriverState *bs, int ign, int64_t offset, 2661 int64_t size) 2662 { 2663 BDRVQcow2State *s = bs->opaque; 2664 int chk = s->overlap_check & ~ign; 2665 int i, j; 2666 2667 if (!size) { 2668 return 0; 2669 } 2670 2671 if (chk & QCOW2_OL_MAIN_HEADER) { 2672 if (offset < s->cluster_size) { 2673 return QCOW2_OL_MAIN_HEADER; 2674 } 2675 } 2676 2677 /* align range to test to cluster boundaries */ 2678 size = ROUND_UP(offset_into_cluster(s, offset) + size, s->cluster_size); 2679 offset = start_of_cluster(s, offset); 2680 2681 if ((chk & QCOW2_OL_ACTIVE_L1) && s->l1_size) { 2682 if (overlaps_with(s->l1_table_offset, s->l1_size * sizeof(uint64_t))) { 2683 return QCOW2_OL_ACTIVE_L1; 2684 } 2685 } 2686 2687 if ((chk & QCOW2_OL_REFCOUNT_TABLE) && s->refcount_table_size) { 2688 if (overlaps_with(s->refcount_table_offset, 2689 s->refcount_table_size * sizeof(uint64_t))) { 2690 return QCOW2_OL_REFCOUNT_TABLE; 2691 } 2692 } 2693 2694 if ((chk & QCOW2_OL_SNAPSHOT_TABLE) && s->snapshots_size) { 2695 if (overlaps_with(s->snapshots_offset, s->snapshots_size)) { 2696 return QCOW2_OL_SNAPSHOT_TABLE; 2697 } 2698 } 2699 2700 if ((chk & QCOW2_OL_INACTIVE_L1) && s->snapshots) { 2701 for (i = 0; i < s->nb_snapshots; i++) { 2702 if (s->snapshots[i].l1_size && 2703 overlaps_with(s->snapshots[i].l1_table_offset, 2704 s->snapshots[i].l1_size * sizeof(uint64_t))) { 2705 return QCOW2_OL_INACTIVE_L1; 2706 } 2707 } 2708 } 2709 2710 if ((chk & QCOW2_OL_ACTIVE_L2) && s->l1_table) { 2711 for (i = 0; i < s->l1_size; i++) { 2712 if ((s->l1_table[i] & L1E_OFFSET_MASK) && 2713 overlaps_with(s->l1_table[i] & L1E_OFFSET_MASK, 2714 s->cluster_size)) { 2715 return QCOW2_OL_ACTIVE_L2; 2716 } 2717 } 2718 } 2719 2720 if ((chk & QCOW2_OL_REFCOUNT_BLOCK) && s->refcount_table) { 2721 unsigned last_entry = s->max_refcount_table_index; 2722 assert(last_entry < s->refcount_table_size); 2723 assert(last_entry + 1 == s->refcount_table_size || 2724 (s->refcount_table[last_entry + 1] & REFT_OFFSET_MASK) == 0); 2725 for (i = 0; i <= last_entry; i++) { 2726 if ((s->refcount_table[i] & REFT_OFFSET_MASK) && 2727 overlaps_with(s->refcount_table[i] & REFT_OFFSET_MASK, 2728 s->cluster_size)) { 2729 return QCOW2_OL_REFCOUNT_BLOCK; 2730 } 2731 } 2732 } 2733 2734 if ((chk & QCOW2_OL_INACTIVE_L2) && s->snapshots) { 2735 for (i = 0; i < s->nb_snapshots; i++) { 2736 uint64_t l1_ofs = s->snapshots[i].l1_table_offset; 2737 uint32_t l1_sz = s->snapshots[i].l1_size; 2738 uint64_t l1_sz2 = l1_sz * sizeof(uint64_t); 2739 uint64_t *l1; 2740 int ret; 2741 2742 ret = qcow2_validate_table(bs, l1_ofs, l1_sz, sizeof(uint64_t), 2743 QCOW_MAX_L1_SIZE, "", NULL); 2744 if (ret < 0) { 2745 return ret; 2746 } 2747 2748 l1 = g_try_malloc(l1_sz2); 2749 2750 if (l1_sz2 && l1 == NULL) { 2751 return -ENOMEM; 2752 } 2753 2754 ret = bdrv_pread(bs->file, l1_ofs, l1, l1_sz2); 2755 if (ret < 0) { 2756 g_free(l1); 2757 return ret; 2758 } 2759 2760 for (j = 0; j < l1_sz; j++) { 2761 uint64_t l2_ofs = be64_to_cpu(l1[j]) & L1E_OFFSET_MASK; 2762 if (l2_ofs && overlaps_with(l2_ofs, s->cluster_size)) { 2763 g_free(l1); 2764 return QCOW2_OL_INACTIVE_L2; 2765 } 2766 } 2767 2768 g_free(l1); 2769 } 2770 } 2771 2772 if ((chk & QCOW2_OL_BITMAP_DIRECTORY) && 2773 (s->autoclear_features & QCOW2_AUTOCLEAR_BITMAPS)) 2774 { 2775 if (overlaps_with(s->bitmap_directory_offset, 2776 s->bitmap_directory_size)) 2777 { 2778 return QCOW2_OL_BITMAP_DIRECTORY; 2779 } 2780 } 2781 2782 return 0; 2783 } 2784 2785 static const char *metadata_ol_names[] = { 2786 [QCOW2_OL_MAIN_HEADER_BITNR] = "qcow2_header", 2787 [QCOW2_OL_ACTIVE_L1_BITNR] = "active L1 table", 2788 [QCOW2_OL_ACTIVE_L2_BITNR] = "active L2 table", 2789 [QCOW2_OL_REFCOUNT_TABLE_BITNR] = "refcount table", 2790 [QCOW2_OL_REFCOUNT_BLOCK_BITNR] = "refcount block", 2791 [QCOW2_OL_SNAPSHOT_TABLE_BITNR] = "snapshot table", 2792 [QCOW2_OL_INACTIVE_L1_BITNR] = "inactive L1 table", 2793 [QCOW2_OL_INACTIVE_L2_BITNR] = "inactive L2 table", 2794 [QCOW2_OL_BITMAP_DIRECTORY_BITNR] = "bitmap directory", 2795 }; 2796 QEMU_BUILD_BUG_ON(QCOW2_OL_MAX_BITNR != ARRAY_SIZE(metadata_ol_names)); 2797 2798 /* 2799 * First performs a check for metadata overlaps (through 2800 * qcow2_check_metadata_overlap); if that fails with a negative value (error 2801 * while performing a check), that value is returned. If an impending overlap 2802 * is detected, the BDS will be made unusable, the qcow2 file marked corrupt 2803 * and -EIO returned. 2804 * 2805 * Returns 0 if there were neither overlaps nor errors while checking for 2806 * overlaps; or a negative value (-errno) on error. 2807 */ 2808 int qcow2_pre_write_overlap_check(BlockDriverState *bs, int ign, int64_t offset, 2809 int64_t size, bool data_file) 2810 { 2811 int ret; 2812 2813 if (data_file && has_data_file(bs)) { 2814 return 0; 2815 } 2816 2817 ret = qcow2_check_metadata_overlap(bs, ign, offset, size); 2818 if (ret < 0) { 2819 return ret; 2820 } else if (ret > 0) { 2821 int metadata_ol_bitnr = ctz32(ret); 2822 assert(metadata_ol_bitnr < QCOW2_OL_MAX_BITNR); 2823 2824 qcow2_signal_corruption(bs, true, offset, size, "Preventing invalid " 2825 "write on metadata (overlaps with %s)", 2826 metadata_ol_names[metadata_ol_bitnr]); 2827 return -EIO; 2828 } 2829 2830 return 0; 2831 } 2832 2833 /* A pointer to a function of this type is given to walk_over_reftable(). That 2834 * function will create refblocks and pass them to a RefblockFinishOp once they 2835 * are completed (@refblock). @refblock_empty is set if the refblock is 2836 * completely empty. 2837 * 2838 * Along with the refblock, a corresponding reftable entry is passed, in the 2839 * reftable @reftable (which may be reallocated) at @reftable_index. 2840 * 2841 * @allocated should be set to true if a new cluster has been allocated. 2842 */ 2843 typedef int (RefblockFinishOp)(BlockDriverState *bs, uint64_t **reftable, 2844 uint64_t reftable_index, uint64_t *reftable_size, 2845 void *refblock, bool refblock_empty, 2846 bool *allocated, Error **errp); 2847 2848 /** 2849 * This "operation" for walk_over_reftable() allocates the refblock on disk (if 2850 * it is not empty) and inserts its offset into the new reftable. The size of 2851 * this new reftable is increased as required. 2852 */ 2853 static int alloc_refblock(BlockDriverState *bs, uint64_t **reftable, 2854 uint64_t reftable_index, uint64_t *reftable_size, 2855 void *refblock, bool refblock_empty, bool *allocated, 2856 Error **errp) 2857 { 2858 BDRVQcow2State *s = bs->opaque; 2859 int64_t offset; 2860 2861 if (!refblock_empty && reftable_index >= *reftable_size) { 2862 uint64_t *new_reftable; 2863 uint64_t new_reftable_size; 2864 2865 new_reftable_size = ROUND_UP(reftable_index + 1, 2866 s->cluster_size / sizeof(uint64_t)); 2867 if (new_reftable_size > QCOW_MAX_REFTABLE_SIZE / sizeof(uint64_t)) { 2868 error_setg(errp, 2869 "This operation would make the refcount table grow " 2870 "beyond the maximum size supported by QEMU, aborting"); 2871 return -ENOTSUP; 2872 } 2873 2874 new_reftable = g_try_realloc(*reftable, new_reftable_size * 2875 sizeof(uint64_t)); 2876 if (!new_reftable) { 2877 error_setg(errp, "Failed to increase reftable buffer size"); 2878 return -ENOMEM; 2879 } 2880 2881 memset(new_reftable + *reftable_size, 0, 2882 (new_reftable_size - *reftable_size) * sizeof(uint64_t)); 2883 2884 *reftable = new_reftable; 2885 *reftable_size = new_reftable_size; 2886 } 2887 2888 if (!refblock_empty && !(*reftable)[reftable_index]) { 2889 offset = qcow2_alloc_clusters(bs, s->cluster_size); 2890 if (offset < 0) { 2891 error_setg_errno(errp, -offset, "Failed to allocate refblock"); 2892 return offset; 2893 } 2894 (*reftable)[reftable_index] = offset; 2895 *allocated = true; 2896 } 2897 2898 return 0; 2899 } 2900 2901 /** 2902 * This "operation" for walk_over_reftable() writes the refblock to disk at the 2903 * offset specified by the new reftable's entry. It does not modify the new 2904 * reftable or change any refcounts. 2905 */ 2906 static int flush_refblock(BlockDriverState *bs, uint64_t **reftable, 2907 uint64_t reftable_index, uint64_t *reftable_size, 2908 void *refblock, bool refblock_empty, bool *allocated, 2909 Error **errp) 2910 { 2911 BDRVQcow2State *s = bs->opaque; 2912 int64_t offset; 2913 int ret; 2914 2915 if (reftable_index < *reftable_size && (*reftable)[reftable_index]) { 2916 offset = (*reftable)[reftable_index]; 2917 2918 ret = qcow2_pre_write_overlap_check(bs, 0, offset, s->cluster_size, 2919 false); 2920 if (ret < 0) { 2921 error_setg_errno(errp, -ret, "Overlap check failed"); 2922 return ret; 2923 } 2924 2925 ret = bdrv_pwrite(bs->file, offset, refblock, s->cluster_size); 2926 if (ret < 0) { 2927 error_setg_errno(errp, -ret, "Failed to write refblock"); 2928 return ret; 2929 } 2930 } else { 2931 assert(refblock_empty); 2932 } 2933 2934 return 0; 2935 } 2936 2937 /** 2938 * This function walks over the existing reftable and every referenced refblock; 2939 * if @new_set_refcount is non-NULL, it is called for every refcount entry to 2940 * create an equal new entry in the passed @new_refblock. Once that 2941 * @new_refblock is completely filled, @operation will be called. 2942 * 2943 * @status_cb and @cb_opaque are used for the amend operation's status callback. 2944 * @index is the index of the walk_over_reftable() calls and @total is the total 2945 * number of walk_over_reftable() calls per amend operation. Both are used for 2946 * calculating the parameters for the status callback. 2947 * 2948 * @allocated is set to true if a new cluster has been allocated. 2949 */ 2950 static int walk_over_reftable(BlockDriverState *bs, uint64_t **new_reftable, 2951 uint64_t *new_reftable_index, 2952 uint64_t *new_reftable_size, 2953 void *new_refblock, int new_refblock_size, 2954 int new_refcount_bits, 2955 RefblockFinishOp *operation, bool *allocated, 2956 Qcow2SetRefcountFunc *new_set_refcount, 2957 BlockDriverAmendStatusCB *status_cb, 2958 void *cb_opaque, int index, int total, 2959 Error **errp) 2960 { 2961 BDRVQcow2State *s = bs->opaque; 2962 uint64_t reftable_index; 2963 bool new_refblock_empty = true; 2964 int refblock_index; 2965 int new_refblock_index = 0; 2966 int ret; 2967 2968 for (reftable_index = 0; reftable_index < s->refcount_table_size; 2969 reftable_index++) 2970 { 2971 uint64_t refblock_offset = s->refcount_table[reftable_index] 2972 & REFT_OFFSET_MASK; 2973 2974 status_cb(bs, (uint64_t)index * s->refcount_table_size + reftable_index, 2975 (uint64_t)total * s->refcount_table_size, cb_opaque); 2976 2977 if (refblock_offset) { 2978 void *refblock; 2979 2980 if (offset_into_cluster(s, refblock_offset)) { 2981 qcow2_signal_corruption(bs, true, -1, -1, "Refblock offset %#" 2982 PRIx64 " unaligned (reftable index: %#" 2983 PRIx64 ")", refblock_offset, 2984 reftable_index); 2985 error_setg(errp, 2986 "Image is corrupt (unaligned refblock offset)"); 2987 return -EIO; 2988 } 2989 2990 ret = qcow2_cache_get(bs, s->refcount_block_cache, refblock_offset, 2991 &refblock); 2992 if (ret < 0) { 2993 error_setg_errno(errp, -ret, "Failed to retrieve refblock"); 2994 return ret; 2995 } 2996 2997 for (refblock_index = 0; refblock_index < s->refcount_block_size; 2998 refblock_index++) 2999 { 3000 uint64_t refcount; 3001 3002 if (new_refblock_index >= new_refblock_size) { 3003 /* new_refblock is now complete */ 3004 ret = operation(bs, new_reftable, *new_reftable_index, 3005 new_reftable_size, new_refblock, 3006 new_refblock_empty, allocated, errp); 3007 if (ret < 0) { 3008 qcow2_cache_put(s->refcount_block_cache, &refblock); 3009 return ret; 3010 } 3011 3012 (*new_reftable_index)++; 3013 new_refblock_index = 0; 3014 new_refblock_empty = true; 3015 } 3016 3017 refcount = s->get_refcount(refblock, refblock_index); 3018 if (new_refcount_bits < 64 && refcount >> new_refcount_bits) { 3019 uint64_t offset; 3020 3021 qcow2_cache_put(s->refcount_block_cache, &refblock); 3022 3023 offset = ((reftable_index << s->refcount_block_bits) 3024 + refblock_index) << s->cluster_bits; 3025 3026 error_setg(errp, "Cannot decrease refcount entry width to " 3027 "%i bits: Cluster at offset %#" PRIx64 " has a " 3028 "refcount of %" PRIu64, new_refcount_bits, 3029 offset, refcount); 3030 return -EINVAL; 3031 } 3032 3033 if (new_set_refcount) { 3034 new_set_refcount(new_refblock, new_refblock_index++, 3035 refcount); 3036 } else { 3037 new_refblock_index++; 3038 } 3039 new_refblock_empty = new_refblock_empty && refcount == 0; 3040 } 3041 3042 qcow2_cache_put(s->refcount_block_cache, &refblock); 3043 } else { 3044 /* No refblock means every refcount is 0 */ 3045 for (refblock_index = 0; refblock_index < s->refcount_block_size; 3046 refblock_index++) 3047 { 3048 if (new_refblock_index >= new_refblock_size) { 3049 /* new_refblock is now complete */ 3050 ret = operation(bs, new_reftable, *new_reftable_index, 3051 new_reftable_size, new_refblock, 3052 new_refblock_empty, allocated, errp); 3053 if (ret < 0) { 3054 return ret; 3055 } 3056 3057 (*new_reftable_index)++; 3058 new_refblock_index = 0; 3059 new_refblock_empty = true; 3060 } 3061 3062 if (new_set_refcount) { 3063 new_set_refcount(new_refblock, new_refblock_index++, 0); 3064 } else { 3065 new_refblock_index++; 3066 } 3067 } 3068 } 3069 } 3070 3071 if (new_refblock_index > 0) { 3072 /* Complete the potentially existing partially filled final refblock */ 3073 if (new_set_refcount) { 3074 for (; new_refblock_index < new_refblock_size; 3075 new_refblock_index++) 3076 { 3077 new_set_refcount(new_refblock, new_refblock_index, 0); 3078 } 3079 } 3080 3081 ret = operation(bs, new_reftable, *new_reftable_index, 3082 new_reftable_size, new_refblock, new_refblock_empty, 3083 allocated, errp); 3084 if (ret < 0) { 3085 return ret; 3086 } 3087 3088 (*new_reftable_index)++; 3089 } 3090 3091 status_cb(bs, (uint64_t)(index + 1) * s->refcount_table_size, 3092 (uint64_t)total * s->refcount_table_size, cb_opaque); 3093 3094 return 0; 3095 } 3096 3097 int qcow2_change_refcount_order(BlockDriverState *bs, int refcount_order, 3098 BlockDriverAmendStatusCB *status_cb, 3099 void *cb_opaque, Error **errp) 3100 { 3101 BDRVQcow2State *s = bs->opaque; 3102 Qcow2GetRefcountFunc *new_get_refcount; 3103 Qcow2SetRefcountFunc *new_set_refcount; 3104 void *new_refblock = qemu_blockalign(bs->file->bs, s->cluster_size); 3105 uint64_t *new_reftable = NULL, new_reftable_size = 0; 3106 uint64_t *old_reftable, old_reftable_size, old_reftable_offset; 3107 uint64_t new_reftable_index = 0; 3108 uint64_t i; 3109 int64_t new_reftable_offset = 0, allocated_reftable_size = 0; 3110 int new_refblock_size, new_refcount_bits = 1 << refcount_order; 3111 int old_refcount_order; 3112 int walk_index = 0; 3113 int ret; 3114 bool new_allocation; 3115 3116 assert(s->qcow_version >= 3); 3117 assert(refcount_order >= 0 && refcount_order <= 6); 3118 3119 /* see qcow2_open() */ 3120 new_refblock_size = 1 << (s->cluster_bits - (refcount_order - 3)); 3121 3122 new_get_refcount = get_refcount_funcs[refcount_order]; 3123 new_set_refcount = set_refcount_funcs[refcount_order]; 3124 3125 3126 do { 3127 int total_walks; 3128 3129 new_allocation = false; 3130 3131 /* At least we have to do this walk and the one which writes the 3132 * refblocks; also, at least we have to do this loop here at least 3133 * twice (normally), first to do the allocations, and second to 3134 * determine that everything is correctly allocated, this then makes 3135 * three walks in total */ 3136 total_walks = MAX(walk_index + 2, 3); 3137 3138 /* First, allocate the structures so they are present in the refcount 3139 * structures */ 3140 ret = walk_over_reftable(bs, &new_reftable, &new_reftable_index, 3141 &new_reftable_size, NULL, new_refblock_size, 3142 new_refcount_bits, &alloc_refblock, 3143 &new_allocation, NULL, status_cb, cb_opaque, 3144 walk_index++, total_walks, errp); 3145 if (ret < 0) { 3146 goto done; 3147 } 3148 3149 new_reftable_index = 0; 3150 3151 if (new_allocation) { 3152 if (new_reftable_offset) { 3153 qcow2_free_clusters(bs, new_reftable_offset, 3154 allocated_reftable_size * sizeof(uint64_t), 3155 QCOW2_DISCARD_NEVER); 3156 } 3157 3158 new_reftable_offset = qcow2_alloc_clusters(bs, new_reftable_size * 3159 sizeof(uint64_t)); 3160 if (new_reftable_offset < 0) { 3161 error_setg_errno(errp, -new_reftable_offset, 3162 "Failed to allocate the new reftable"); 3163 ret = new_reftable_offset; 3164 goto done; 3165 } 3166 allocated_reftable_size = new_reftable_size; 3167 } 3168 } while (new_allocation); 3169 3170 /* Second, write the new refblocks */ 3171 ret = walk_over_reftable(bs, &new_reftable, &new_reftable_index, 3172 &new_reftable_size, new_refblock, 3173 new_refblock_size, new_refcount_bits, 3174 &flush_refblock, &new_allocation, new_set_refcount, 3175 status_cb, cb_opaque, walk_index, walk_index + 1, 3176 errp); 3177 if (ret < 0) { 3178 goto done; 3179 } 3180 assert(!new_allocation); 3181 3182 3183 /* Write the new reftable */ 3184 ret = qcow2_pre_write_overlap_check(bs, 0, new_reftable_offset, 3185 new_reftable_size * sizeof(uint64_t), 3186 false); 3187 if (ret < 0) { 3188 error_setg_errno(errp, -ret, "Overlap check failed"); 3189 goto done; 3190 } 3191 3192 for (i = 0; i < new_reftable_size; i++) { 3193 cpu_to_be64s(&new_reftable[i]); 3194 } 3195 3196 ret = bdrv_pwrite(bs->file, new_reftable_offset, new_reftable, 3197 new_reftable_size * sizeof(uint64_t)); 3198 3199 for (i = 0; i < new_reftable_size; i++) { 3200 be64_to_cpus(&new_reftable[i]); 3201 } 3202 3203 if (ret < 0) { 3204 error_setg_errno(errp, -ret, "Failed to write the new reftable"); 3205 goto done; 3206 } 3207 3208 3209 /* Empty the refcount cache */ 3210 ret = qcow2_cache_flush(bs, s->refcount_block_cache); 3211 if (ret < 0) { 3212 error_setg_errno(errp, -ret, "Failed to flush the refblock cache"); 3213 goto done; 3214 } 3215 3216 /* Update the image header to point to the new reftable; this only updates 3217 * the fields which are relevant to qcow2_update_header(); other fields 3218 * such as s->refcount_table or s->refcount_bits stay stale for now 3219 * (because we have to restore everything if qcow2_update_header() fails) */ 3220 old_refcount_order = s->refcount_order; 3221 old_reftable_size = s->refcount_table_size; 3222 old_reftable_offset = s->refcount_table_offset; 3223 3224 s->refcount_order = refcount_order; 3225 s->refcount_table_size = new_reftable_size; 3226 s->refcount_table_offset = new_reftable_offset; 3227 3228 ret = qcow2_update_header(bs); 3229 if (ret < 0) { 3230 s->refcount_order = old_refcount_order; 3231 s->refcount_table_size = old_reftable_size; 3232 s->refcount_table_offset = old_reftable_offset; 3233 error_setg_errno(errp, -ret, "Failed to update the qcow2 header"); 3234 goto done; 3235 } 3236 3237 /* Now update the rest of the in-memory information */ 3238 old_reftable = s->refcount_table; 3239 s->refcount_table = new_reftable; 3240 update_max_refcount_table_index(s); 3241 3242 s->refcount_bits = 1 << refcount_order; 3243 s->refcount_max = UINT64_C(1) << (s->refcount_bits - 1); 3244 s->refcount_max += s->refcount_max - 1; 3245 3246 s->refcount_block_bits = s->cluster_bits - (refcount_order - 3); 3247 s->refcount_block_size = 1 << s->refcount_block_bits; 3248 3249 s->get_refcount = new_get_refcount; 3250 s->set_refcount = new_set_refcount; 3251 3252 /* For cleaning up all old refblocks and the old reftable below the "done" 3253 * label */ 3254 new_reftable = old_reftable; 3255 new_reftable_size = old_reftable_size; 3256 new_reftable_offset = old_reftable_offset; 3257 3258 done: 3259 if (new_reftable) { 3260 /* On success, new_reftable actually points to the old reftable (and 3261 * new_reftable_size is the old reftable's size); but that is just 3262 * fine */ 3263 for (i = 0; i < new_reftable_size; i++) { 3264 uint64_t offset = new_reftable[i] & REFT_OFFSET_MASK; 3265 if (offset) { 3266 qcow2_free_clusters(bs, offset, s->cluster_size, 3267 QCOW2_DISCARD_OTHER); 3268 } 3269 } 3270 g_free(new_reftable); 3271 3272 if (new_reftable_offset > 0) { 3273 qcow2_free_clusters(bs, new_reftable_offset, 3274 new_reftable_size * sizeof(uint64_t), 3275 QCOW2_DISCARD_OTHER); 3276 } 3277 } 3278 3279 qemu_vfree(new_refblock); 3280 return ret; 3281 } 3282 3283 static int64_t get_refblock_offset(BlockDriverState *bs, uint64_t offset) 3284 { 3285 BDRVQcow2State *s = bs->opaque; 3286 uint32_t index = offset_to_reftable_index(s, offset); 3287 int64_t covering_refblock_offset = 0; 3288 3289 if (index < s->refcount_table_size) { 3290 covering_refblock_offset = s->refcount_table[index] & REFT_OFFSET_MASK; 3291 } 3292 if (!covering_refblock_offset) { 3293 qcow2_signal_corruption(bs, true, -1, -1, "Refblock at %#" PRIx64 " is " 3294 "not covered by the refcount structures", 3295 offset); 3296 return -EIO; 3297 } 3298 3299 return covering_refblock_offset; 3300 } 3301 3302 static int qcow2_discard_refcount_block(BlockDriverState *bs, 3303 uint64_t discard_block_offs) 3304 { 3305 BDRVQcow2State *s = bs->opaque; 3306 int64_t refblock_offs; 3307 uint64_t cluster_index = discard_block_offs >> s->cluster_bits; 3308 uint32_t block_index = cluster_index & (s->refcount_block_size - 1); 3309 void *refblock; 3310 int ret; 3311 3312 refblock_offs = get_refblock_offset(bs, discard_block_offs); 3313 if (refblock_offs < 0) { 3314 return refblock_offs; 3315 } 3316 3317 assert(discard_block_offs != 0); 3318 3319 ret = qcow2_cache_get(bs, s->refcount_block_cache, refblock_offs, 3320 &refblock); 3321 if (ret < 0) { 3322 return ret; 3323 } 3324 3325 if (s->get_refcount(refblock, block_index) != 1) { 3326 qcow2_signal_corruption(bs, true, -1, -1, "Invalid refcount:" 3327 " refblock offset %#" PRIx64 3328 ", reftable index %u" 3329 ", block offset %#" PRIx64 3330 ", refcount %#" PRIx64, 3331 refblock_offs, 3332 offset_to_reftable_index(s, discard_block_offs), 3333 discard_block_offs, 3334 s->get_refcount(refblock, block_index)); 3335 qcow2_cache_put(s->refcount_block_cache, &refblock); 3336 return -EINVAL; 3337 } 3338 s->set_refcount(refblock, block_index, 0); 3339 3340 qcow2_cache_entry_mark_dirty(s->refcount_block_cache, refblock); 3341 3342 qcow2_cache_put(s->refcount_block_cache, &refblock); 3343 3344 if (cluster_index < s->free_cluster_index) { 3345 s->free_cluster_index = cluster_index; 3346 } 3347 3348 refblock = qcow2_cache_is_table_offset(s->refcount_block_cache, 3349 discard_block_offs); 3350 if (refblock) { 3351 /* discard refblock from the cache if refblock is cached */ 3352 qcow2_cache_discard(s->refcount_block_cache, refblock); 3353 } 3354 update_refcount_discard(bs, discard_block_offs, s->cluster_size); 3355 3356 return 0; 3357 } 3358 3359 int qcow2_shrink_reftable(BlockDriverState *bs) 3360 { 3361 BDRVQcow2State *s = bs->opaque; 3362 uint64_t *reftable_tmp = 3363 g_malloc(s->refcount_table_size * sizeof(uint64_t)); 3364 int i, ret; 3365 3366 for (i = 0; i < s->refcount_table_size; i++) { 3367 int64_t refblock_offs = s->refcount_table[i] & REFT_OFFSET_MASK; 3368 void *refblock; 3369 bool unused_block; 3370 3371 if (refblock_offs == 0) { 3372 reftable_tmp[i] = 0; 3373 continue; 3374 } 3375 ret = qcow2_cache_get(bs, s->refcount_block_cache, refblock_offs, 3376 &refblock); 3377 if (ret < 0) { 3378 goto out; 3379 } 3380 3381 /* the refblock has own reference */ 3382 if (i == offset_to_reftable_index(s, refblock_offs)) { 3383 uint64_t block_index = (refblock_offs >> s->cluster_bits) & 3384 (s->refcount_block_size - 1); 3385 uint64_t refcount = s->get_refcount(refblock, block_index); 3386 3387 s->set_refcount(refblock, block_index, 0); 3388 3389 unused_block = buffer_is_zero(refblock, s->cluster_size); 3390 3391 s->set_refcount(refblock, block_index, refcount); 3392 } else { 3393 unused_block = buffer_is_zero(refblock, s->cluster_size); 3394 } 3395 qcow2_cache_put(s->refcount_block_cache, &refblock); 3396 3397 reftable_tmp[i] = unused_block ? 0 : cpu_to_be64(s->refcount_table[i]); 3398 } 3399 3400 ret = bdrv_pwrite_sync(bs->file, s->refcount_table_offset, reftable_tmp, 3401 s->refcount_table_size * sizeof(uint64_t)); 3402 /* 3403 * If the write in the reftable failed the image may contain a partially 3404 * overwritten reftable. In this case it would be better to clear the 3405 * reftable in memory to avoid possible image corruption. 3406 */ 3407 for (i = 0; i < s->refcount_table_size; i++) { 3408 if (s->refcount_table[i] && !reftable_tmp[i]) { 3409 if (ret == 0) { 3410 ret = qcow2_discard_refcount_block(bs, s->refcount_table[i] & 3411 REFT_OFFSET_MASK); 3412 } 3413 s->refcount_table[i] = 0; 3414 } 3415 } 3416 3417 if (!s->cache_discards) { 3418 qcow2_process_discards(bs, ret); 3419 } 3420 3421 out: 3422 g_free(reftable_tmp); 3423 return ret; 3424 } 3425 3426 int64_t qcow2_get_last_cluster(BlockDriverState *bs, int64_t size) 3427 { 3428 BDRVQcow2State *s = bs->opaque; 3429 int64_t i; 3430 3431 for (i = size_to_clusters(s, size) - 1; i >= 0; i--) { 3432 uint64_t refcount; 3433 int ret = qcow2_get_refcount(bs, i, &refcount); 3434 if (ret < 0) { 3435 fprintf(stderr, "Can't get refcount for cluster %" PRId64 ": %s\n", 3436 i, strerror(-ret)); 3437 return ret; 3438 } 3439 if (refcount > 0) { 3440 return i; 3441 } 3442 } 3443 qcow2_signal_corruption(bs, true, -1, -1, 3444 "There are no references in the refcount table."); 3445 return -EIO; 3446 } 3447