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 #include "qemu-common.h" 25 #include "block/block_int.h" 26 #include "qemu/module.h" 27 #include <zlib.h> 28 #include "qemu/aes.h" 29 #include "block/qcow2.h" 30 #include "qemu/error-report.h" 31 #include "qapi/qmp/qerror.h" 32 #include "qapi/qmp/qbool.h" 33 #include "qapi/util.h" 34 #include "qapi/qmp/types.h" 35 #include "qapi-event.h" 36 #include "trace.h" 37 #include "qemu/option_int.h" 38 39 /* 40 Differences with QCOW: 41 42 - Support for multiple incremental snapshots. 43 - Memory management by reference counts. 44 - Clusters which have a reference count of one have the bit 45 QCOW_OFLAG_COPIED to optimize write performance. 46 - Size of compressed clusters is stored in sectors to reduce bit usage 47 in the cluster offsets. 48 - Support for storing additional data (such as the VM state) in the 49 snapshots. 50 - If a backing store is used, the cluster size is not constrained 51 (could be backported to QCOW). 52 - L2 tables have always a size of one cluster. 53 */ 54 55 56 typedef struct { 57 uint32_t magic; 58 uint32_t len; 59 } QEMU_PACKED QCowExtension; 60 61 #define QCOW2_EXT_MAGIC_END 0 62 #define QCOW2_EXT_MAGIC_BACKING_FORMAT 0xE2792ACA 63 #define QCOW2_EXT_MAGIC_FEATURE_TABLE 0x6803f857 64 65 static int qcow2_probe(const uint8_t *buf, int buf_size, const char *filename) 66 { 67 const QCowHeader *cow_header = (const void *)buf; 68 69 if (buf_size >= sizeof(QCowHeader) && 70 be32_to_cpu(cow_header->magic) == QCOW_MAGIC && 71 be32_to_cpu(cow_header->version) >= 2) 72 return 100; 73 else 74 return 0; 75 } 76 77 78 /* 79 * read qcow2 extension and fill bs 80 * start reading from start_offset 81 * finish reading upon magic of value 0 or when end_offset reached 82 * unknown magic is skipped (future extension this version knows nothing about) 83 * return 0 upon success, non-0 otherwise 84 */ 85 static int qcow2_read_extensions(BlockDriverState *bs, uint64_t start_offset, 86 uint64_t end_offset, void **p_feature_table, 87 Error **errp) 88 { 89 BDRVQcowState *s = bs->opaque; 90 QCowExtension ext; 91 uint64_t offset; 92 int ret; 93 94 #ifdef DEBUG_EXT 95 printf("qcow2_read_extensions: start=%ld end=%ld\n", start_offset, end_offset); 96 #endif 97 offset = start_offset; 98 while (offset < end_offset) { 99 100 #ifdef DEBUG_EXT 101 /* Sanity check */ 102 if (offset > s->cluster_size) 103 printf("qcow2_read_extension: suspicious offset %lu\n", offset); 104 105 printf("attempting to read extended header in offset %lu\n", offset); 106 #endif 107 108 ret = bdrv_pread(bs->file, offset, &ext, sizeof(ext)); 109 if (ret < 0) { 110 error_setg_errno(errp, -ret, "qcow2_read_extension: ERROR: " 111 "pread fail from offset %" PRIu64, offset); 112 return 1; 113 } 114 be32_to_cpus(&ext.magic); 115 be32_to_cpus(&ext.len); 116 offset += sizeof(ext); 117 #ifdef DEBUG_EXT 118 printf("ext.magic = 0x%x\n", ext.magic); 119 #endif 120 if (offset > end_offset || ext.len > end_offset - offset) { 121 error_setg(errp, "Header extension too large"); 122 return -EINVAL; 123 } 124 125 switch (ext.magic) { 126 case QCOW2_EXT_MAGIC_END: 127 return 0; 128 129 case QCOW2_EXT_MAGIC_BACKING_FORMAT: 130 if (ext.len >= sizeof(bs->backing_format)) { 131 error_setg(errp, "ERROR: ext_backing_format: len=%" PRIu32 132 " too large (>=%zu)", ext.len, 133 sizeof(bs->backing_format)); 134 return 2; 135 } 136 ret = bdrv_pread(bs->file, offset, bs->backing_format, ext.len); 137 if (ret < 0) { 138 error_setg_errno(errp, -ret, "ERROR: ext_backing_format: " 139 "Could not read format name"); 140 return 3; 141 } 142 bs->backing_format[ext.len] = '\0'; 143 #ifdef DEBUG_EXT 144 printf("Qcow2: Got format extension %s\n", bs->backing_format); 145 #endif 146 break; 147 148 case QCOW2_EXT_MAGIC_FEATURE_TABLE: 149 if (p_feature_table != NULL) { 150 void* feature_table = g_malloc0(ext.len + 2 * sizeof(Qcow2Feature)); 151 ret = bdrv_pread(bs->file, offset , feature_table, ext.len); 152 if (ret < 0) { 153 error_setg_errno(errp, -ret, "ERROR: ext_feature_table: " 154 "Could not read table"); 155 return ret; 156 } 157 158 *p_feature_table = feature_table; 159 } 160 break; 161 162 default: 163 /* unknown magic - save it in case we need to rewrite the header */ 164 { 165 Qcow2UnknownHeaderExtension *uext; 166 167 uext = g_malloc0(sizeof(*uext) + ext.len); 168 uext->magic = ext.magic; 169 uext->len = ext.len; 170 QLIST_INSERT_HEAD(&s->unknown_header_ext, uext, next); 171 172 ret = bdrv_pread(bs->file, offset , uext->data, uext->len); 173 if (ret < 0) { 174 error_setg_errno(errp, -ret, "ERROR: unknown extension: " 175 "Could not read data"); 176 return ret; 177 } 178 } 179 break; 180 } 181 182 offset += ((ext.len + 7) & ~7); 183 } 184 185 return 0; 186 } 187 188 static void cleanup_unknown_header_ext(BlockDriverState *bs) 189 { 190 BDRVQcowState *s = bs->opaque; 191 Qcow2UnknownHeaderExtension *uext, *next; 192 193 QLIST_FOREACH_SAFE(uext, &s->unknown_header_ext, next, next) { 194 QLIST_REMOVE(uext, next); 195 g_free(uext); 196 } 197 } 198 199 static void GCC_FMT_ATTR(3, 4) report_unsupported(BlockDriverState *bs, 200 Error **errp, const char *fmt, ...) 201 { 202 char msg[64]; 203 va_list ap; 204 205 va_start(ap, fmt); 206 vsnprintf(msg, sizeof(msg), fmt, ap); 207 va_end(ap); 208 209 error_set(errp, QERR_UNKNOWN_BLOCK_FORMAT_FEATURE, 210 bdrv_get_device_name(bs), "qcow2", msg); 211 } 212 213 static void report_unsupported_feature(BlockDriverState *bs, 214 Error **errp, Qcow2Feature *table, uint64_t mask) 215 { 216 char *features = g_strdup(""); 217 char *old; 218 219 while (table && table->name[0] != '\0') { 220 if (table->type == QCOW2_FEAT_TYPE_INCOMPATIBLE) { 221 if (mask & (1ULL << table->bit)) { 222 old = features; 223 features = g_strdup_printf("%s%s%.46s", old, *old ? ", " : "", 224 table->name); 225 g_free(old); 226 mask &= ~(1ULL << table->bit); 227 } 228 } 229 table++; 230 } 231 232 if (mask) { 233 old = features; 234 features = g_strdup_printf("%s%sUnknown incompatible feature: %" PRIx64, 235 old, *old ? ", " : "", mask); 236 g_free(old); 237 } 238 239 report_unsupported(bs, errp, "%s", features); 240 g_free(features); 241 } 242 243 /* 244 * Sets the dirty bit and flushes afterwards if necessary. 245 * 246 * The incompatible_features bit is only set if the image file header was 247 * updated successfully. Therefore it is not required to check the return 248 * value of this function. 249 */ 250 int qcow2_mark_dirty(BlockDriverState *bs) 251 { 252 BDRVQcowState *s = bs->opaque; 253 uint64_t val; 254 int ret; 255 256 assert(s->qcow_version >= 3); 257 258 if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) { 259 return 0; /* already dirty */ 260 } 261 262 val = cpu_to_be64(s->incompatible_features | QCOW2_INCOMPAT_DIRTY); 263 ret = bdrv_pwrite(bs->file, offsetof(QCowHeader, incompatible_features), 264 &val, sizeof(val)); 265 if (ret < 0) { 266 return ret; 267 } 268 ret = bdrv_flush(bs->file); 269 if (ret < 0) { 270 return ret; 271 } 272 273 /* Only treat image as dirty if the header was updated successfully */ 274 s->incompatible_features |= QCOW2_INCOMPAT_DIRTY; 275 return 0; 276 } 277 278 /* 279 * Clears the dirty bit and flushes before if necessary. Only call this 280 * function when there are no pending requests, it does not guard against 281 * concurrent requests dirtying the image. 282 */ 283 static int qcow2_mark_clean(BlockDriverState *bs) 284 { 285 BDRVQcowState *s = bs->opaque; 286 287 if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) { 288 int ret; 289 290 s->incompatible_features &= ~QCOW2_INCOMPAT_DIRTY; 291 292 ret = bdrv_flush(bs); 293 if (ret < 0) { 294 return ret; 295 } 296 297 return qcow2_update_header(bs); 298 } 299 return 0; 300 } 301 302 /* 303 * Marks the image as corrupt. 304 */ 305 int qcow2_mark_corrupt(BlockDriverState *bs) 306 { 307 BDRVQcowState *s = bs->opaque; 308 309 s->incompatible_features |= QCOW2_INCOMPAT_CORRUPT; 310 return qcow2_update_header(bs); 311 } 312 313 /* 314 * Marks the image as consistent, i.e., unsets the corrupt bit, and flushes 315 * before if necessary. 316 */ 317 int qcow2_mark_consistent(BlockDriverState *bs) 318 { 319 BDRVQcowState *s = bs->opaque; 320 321 if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) { 322 int ret = bdrv_flush(bs); 323 if (ret < 0) { 324 return ret; 325 } 326 327 s->incompatible_features &= ~QCOW2_INCOMPAT_CORRUPT; 328 return qcow2_update_header(bs); 329 } 330 return 0; 331 } 332 333 static int qcow2_check(BlockDriverState *bs, BdrvCheckResult *result, 334 BdrvCheckMode fix) 335 { 336 int ret = qcow2_check_refcounts(bs, result, fix); 337 if (ret < 0) { 338 return ret; 339 } 340 341 if (fix && result->check_errors == 0 && result->corruptions == 0) { 342 ret = qcow2_mark_clean(bs); 343 if (ret < 0) { 344 return ret; 345 } 346 return qcow2_mark_consistent(bs); 347 } 348 return ret; 349 } 350 351 static int validate_table_offset(BlockDriverState *bs, uint64_t offset, 352 uint64_t entries, size_t entry_len) 353 { 354 BDRVQcowState *s = bs->opaque; 355 uint64_t size; 356 357 /* Use signed INT64_MAX as the maximum even for uint64_t header fields, 358 * because values will be passed to qemu functions taking int64_t. */ 359 if (entries > INT64_MAX / entry_len) { 360 return -EINVAL; 361 } 362 363 size = entries * entry_len; 364 365 if (INT64_MAX - size < offset) { 366 return -EINVAL; 367 } 368 369 /* Tables must be cluster aligned */ 370 if (offset & (s->cluster_size - 1)) { 371 return -EINVAL; 372 } 373 374 return 0; 375 } 376 377 static QemuOptsList qcow2_runtime_opts = { 378 .name = "qcow2", 379 .head = QTAILQ_HEAD_INITIALIZER(qcow2_runtime_opts.head), 380 .desc = { 381 { 382 .name = QCOW2_OPT_LAZY_REFCOUNTS, 383 .type = QEMU_OPT_BOOL, 384 .help = "Postpone refcount updates", 385 }, 386 { 387 .name = QCOW2_OPT_DISCARD_REQUEST, 388 .type = QEMU_OPT_BOOL, 389 .help = "Pass guest discard requests to the layer below", 390 }, 391 { 392 .name = QCOW2_OPT_DISCARD_SNAPSHOT, 393 .type = QEMU_OPT_BOOL, 394 .help = "Generate discard requests when snapshot related space " 395 "is freed", 396 }, 397 { 398 .name = QCOW2_OPT_DISCARD_OTHER, 399 .type = QEMU_OPT_BOOL, 400 .help = "Generate discard requests when other clusters are freed", 401 }, 402 { 403 .name = QCOW2_OPT_OVERLAP, 404 .type = QEMU_OPT_STRING, 405 .help = "Selects which overlap checks to perform from a range of " 406 "templates (none, constant, cached, all)", 407 }, 408 { 409 .name = QCOW2_OPT_OVERLAP_TEMPLATE, 410 .type = QEMU_OPT_STRING, 411 .help = "Selects which overlap checks to perform from a range of " 412 "templates (none, constant, cached, all)", 413 }, 414 { 415 .name = QCOW2_OPT_OVERLAP_MAIN_HEADER, 416 .type = QEMU_OPT_BOOL, 417 .help = "Check for unintended writes into the main qcow2 header", 418 }, 419 { 420 .name = QCOW2_OPT_OVERLAP_ACTIVE_L1, 421 .type = QEMU_OPT_BOOL, 422 .help = "Check for unintended writes into the active L1 table", 423 }, 424 { 425 .name = QCOW2_OPT_OVERLAP_ACTIVE_L2, 426 .type = QEMU_OPT_BOOL, 427 .help = "Check for unintended writes into an active L2 table", 428 }, 429 { 430 .name = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE, 431 .type = QEMU_OPT_BOOL, 432 .help = "Check for unintended writes into the refcount table", 433 }, 434 { 435 .name = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK, 436 .type = QEMU_OPT_BOOL, 437 .help = "Check for unintended writes into a refcount block", 438 }, 439 { 440 .name = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE, 441 .type = QEMU_OPT_BOOL, 442 .help = "Check for unintended writes into the snapshot table", 443 }, 444 { 445 .name = QCOW2_OPT_OVERLAP_INACTIVE_L1, 446 .type = QEMU_OPT_BOOL, 447 .help = "Check for unintended writes into an inactive L1 table", 448 }, 449 { 450 .name = QCOW2_OPT_OVERLAP_INACTIVE_L2, 451 .type = QEMU_OPT_BOOL, 452 .help = "Check for unintended writes into an inactive L2 table", 453 }, 454 { 455 .name = QCOW2_OPT_CACHE_SIZE, 456 .type = QEMU_OPT_SIZE, 457 .help = "Maximum combined metadata (L2 tables and refcount blocks) " 458 "cache size", 459 }, 460 { 461 .name = QCOW2_OPT_L2_CACHE_SIZE, 462 .type = QEMU_OPT_SIZE, 463 .help = "Maximum L2 table cache size", 464 }, 465 { 466 .name = QCOW2_OPT_REFCOUNT_CACHE_SIZE, 467 .type = QEMU_OPT_SIZE, 468 .help = "Maximum refcount block cache size", 469 }, 470 { /* end of list */ } 471 }, 472 }; 473 474 static const char *overlap_bool_option_names[QCOW2_OL_MAX_BITNR] = { 475 [QCOW2_OL_MAIN_HEADER_BITNR] = QCOW2_OPT_OVERLAP_MAIN_HEADER, 476 [QCOW2_OL_ACTIVE_L1_BITNR] = QCOW2_OPT_OVERLAP_ACTIVE_L1, 477 [QCOW2_OL_ACTIVE_L2_BITNR] = QCOW2_OPT_OVERLAP_ACTIVE_L2, 478 [QCOW2_OL_REFCOUNT_TABLE_BITNR] = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE, 479 [QCOW2_OL_REFCOUNT_BLOCK_BITNR] = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK, 480 [QCOW2_OL_SNAPSHOT_TABLE_BITNR] = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE, 481 [QCOW2_OL_INACTIVE_L1_BITNR] = QCOW2_OPT_OVERLAP_INACTIVE_L1, 482 [QCOW2_OL_INACTIVE_L2_BITNR] = QCOW2_OPT_OVERLAP_INACTIVE_L2, 483 }; 484 485 static void read_cache_sizes(QemuOpts *opts, uint64_t *l2_cache_size, 486 uint64_t *refcount_cache_size, Error **errp) 487 { 488 uint64_t combined_cache_size; 489 bool l2_cache_size_set, refcount_cache_size_set, combined_cache_size_set; 490 491 combined_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_CACHE_SIZE); 492 l2_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_L2_CACHE_SIZE); 493 refcount_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_REFCOUNT_CACHE_SIZE); 494 495 combined_cache_size = qemu_opt_get_size(opts, QCOW2_OPT_CACHE_SIZE, 0); 496 *l2_cache_size = qemu_opt_get_size(opts, QCOW2_OPT_L2_CACHE_SIZE, 0); 497 *refcount_cache_size = qemu_opt_get_size(opts, 498 QCOW2_OPT_REFCOUNT_CACHE_SIZE, 0); 499 500 if (combined_cache_size_set) { 501 if (l2_cache_size_set && refcount_cache_size_set) { 502 error_setg(errp, QCOW2_OPT_CACHE_SIZE ", " QCOW2_OPT_L2_CACHE_SIZE 503 " and " QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not be set " 504 "the same time"); 505 return; 506 } else if (*l2_cache_size > combined_cache_size) { 507 error_setg(errp, QCOW2_OPT_L2_CACHE_SIZE " may not exceed " 508 QCOW2_OPT_CACHE_SIZE); 509 return; 510 } else if (*refcount_cache_size > combined_cache_size) { 511 error_setg(errp, QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not exceed " 512 QCOW2_OPT_CACHE_SIZE); 513 return; 514 } 515 516 if (l2_cache_size_set) { 517 *refcount_cache_size = combined_cache_size - *l2_cache_size; 518 } else if (refcount_cache_size_set) { 519 *l2_cache_size = combined_cache_size - *refcount_cache_size; 520 } else { 521 *refcount_cache_size = combined_cache_size 522 / (DEFAULT_L2_REFCOUNT_SIZE_RATIO + 1); 523 *l2_cache_size = combined_cache_size - *refcount_cache_size; 524 } 525 } else { 526 if (!l2_cache_size_set && !refcount_cache_size_set) { 527 *l2_cache_size = DEFAULT_L2_CACHE_BYTE_SIZE; 528 *refcount_cache_size = *l2_cache_size 529 / DEFAULT_L2_REFCOUNT_SIZE_RATIO; 530 } else if (!l2_cache_size_set) { 531 *l2_cache_size = *refcount_cache_size 532 * DEFAULT_L2_REFCOUNT_SIZE_RATIO; 533 } else if (!refcount_cache_size_set) { 534 *refcount_cache_size = *l2_cache_size 535 / DEFAULT_L2_REFCOUNT_SIZE_RATIO; 536 } 537 } 538 } 539 540 static int qcow2_open(BlockDriverState *bs, QDict *options, int flags, 541 Error **errp) 542 { 543 BDRVQcowState *s = bs->opaque; 544 unsigned int len, i; 545 int ret = 0; 546 QCowHeader header; 547 QemuOpts *opts = NULL; 548 Error *local_err = NULL; 549 uint64_t ext_end; 550 uint64_t l1_vm_state_index; 551 const char *opt_overlap_check, *opt_overlap_check_template; 552 int overlap_check_template = 0; 553 uint64_t l2_cache_size, refcount_cache_size; 554 555 ret = bdrv_pread(bs->file, 0, &header, sizeof(header)); 556 if (ret < 0) { 557 error_setg_errno(errp, -ret, "Could not read qcow2 header"); 558 goto fail; 559 } 560 be32_to_cpus(&header.magic); 561 be32_to_cpus(&header.version); 562 be64_to_cpus(&header.backing_file_offset); 563 be32_to_cpus(&header.backing_file_size); 564 be64_to_cpus(&header.size); 565 be32_to_cpus(&header.cluster_bits); 566 be32_to_cpus(&header.crypt_method); 567 be64_to_cpus(&header.l1_table_offset); 568 be32_to_cpus(&header.l1_size); 569 be64_to_cpus(&header.refcount_table_offset); 570 be32_to_cpus(&header.refcount_table_clusters); 571 be64_to_cpus(&header.snapshots_offset); 572 be32_to_cpus(&header.nb_snapshots); 573 574 if (header.magic != QCOW_MAGIC) { 575 error_setg(errp, "Image is not in qcow2 format"); 576 ret = -EINVAL; 577 goto fail; 578 } 579 if (header.version < 2 || header.version > 3) { 580 report_unsupported(bs, errp, "QCOW version %" PRIu32, header.version); 581 ret = -ENOTSUP; 582 goto fail; 583 } 584 585 s->qcow_version = header.version; 586 587 /* Initialise cluster size */ 588 if (header.cluster_bits < MIN_CLUSTER_BITS || 589 header.cluster_bits > MAX_CLUSTER_BITS) { 590 error_setg(errp, "Unsupported cluster size: 2^%" PRIu32, 591 header.cluster_bits); 592 ret = -EINVAL; 593 goto fail; 594 } 595 596 s->cluster_bits = header.cluster_bits; 597 s->cluster_size = 1 << s->cluster_bits; 598 s->cluster_sectors = 1 << (s->cluster_bits - 9); 599 600 /* Initialise version 3 header fields */ 601 if (header.version == 2) { 602 header.incompatible_features = 0; 603 header.compatible_features = 0; 604 header.autoclear_features = 0; 605 header.refcount_order = 4; 606 header.header_length = 72; 607 } else { 608 be64_to_cpus(&header.incompatible_features); 609 be64_to_cpus(&header.compatible_features); 610 be64_to_cpus(&header.autoclear_features); 611 be32_to_cpus(&header.refcount_order); 612 be32_to_cpus(&header.header_length); 613 614 if (header.header_length < 104) { 615 error_setg(errp, "qcow2 header too short"); 616 ret = -EINVAL; 617 goto fail; 618 } 619 } 620 621 if (header.header_length > s->cluster_size) { 622 error_setg(errp, "qcow2 header exceeds cluster size"); 623 ret = -EINVAL; 624 goto fail; 625 } 626 627 if (header.header_length > sizeof(header)) { 628 s->unknown_header_fields_size = header.header_length - sizeof(header); 629 s->unknown_header_fields = g_malloc(s->unknown_header_fields_size); 630 ret = bdrv_pread(bs->file, sizeof(header), s->unknown_header_fields, 631 s->unknown_header_fields_size); 632 if (ret < 0) { 633 error_setg_errno(errp, -ret, "Could not read unknown qcow2 header " 634 "fields"); 635 goto fail; 636 } 637 } 638 639 if (header.backing_file_offset > s->cluster_size) { 640 error_setg(errp, "Invalid backing file offset"); 641 ret = -EINVAL; 642 goto fail; 643 } 644 645 if (header.backing_file_offset) { 646 ext_end = header.backing_file_offset; 647 } else { 648 ext_end = 1 << header.cluster_bits; 649 } 650 651 /* Handle feature bits */ 652 s->incompatible_features = header.incompatible_features; 653 s->compatible_features = header.compatible_features; 654 s->autoclear_features = header.autoclear_features; 655 656 if (s->incompatible_features & ~QCOW2_INCOMPAT_MASK) { 657 void *feature_table = NULL; 658 qcow2_read_extensions(bs, header.header_length, ext_end, 659 &feature_table, NULL); 660 report_unsupported_feature(bs, errp, feature_table, 661 s->incompatible_features & 662 ~QCOW2_INCOMPAT_MASK); 663 ret = -ENOTSUP; 664 g_free(feature_table); 665 goto fail; 666 } 667 668 if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) { 669 /* Corrupt images may not be written to unless they are being repaired 670 */ 671 if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_CHECK)) { 672 error_setg(errp, "qcow2: Image is corrupt; cannot be opened " 673 "read/write"); 674 ret = -EACCES; 675 goto fail; 676 } 677 } 678 679 /* Check support for various header values */ 680 if (header.refcount_order != 4) { 681 report_unsupported(bs, errp, "%d bit reference counts", 682 1 << header.refcount_order); 683 ret = -ENOTSUP; 684 goto fail; 685 } 686 s->refcount_order = header.refcount_order; 687 s->refcount_bits = 1 << s->refcount_order; 688 s->refcount_max = UINT64_C(1) << (s->refcount_bits - 1); 689 s->refcount_max += s->refcount_max - 1; 690 691 if (header.crypt_method > QCOW_CRYPT_AES) { 692 error_setg(errp, "Unsupported encryption method: %" PRIu32, 693 header.crypt_method); 694 ret = -EINVAL; 695 goto fail; 696 } 697 s->crypt_method_header = header.crypt_method; 698 if (s->crypt_method_header) { 699 bs->encrypted = 1; 700 } 701 702 s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */ 703 s->l2_size = 1 << s->l2_bits; 704 /* 2^(s->refcount_order - 3) is the refcount width in bytes */ 705 s->refcount_block_bits = s->cluster_bits - (s->refcount_order - 3); 706 s->refcount_block_size = 1 << s->refcount_block_bits; 707 bs->total_sectors = header.size / 512; 708 s->csize_shift = (62 - (s->cluster_bits - 8)); 709 s->csize_mask = (1 << (s->cluster_bits - 8)) - 1; 710 s->cluster_offset_mask = (1LL << s->csize_shift) - 1; 711 712 s->refcount_table_offset = header.refcount_table_offset; 713 s->refcount_table_size = 714 header.refcount_table_clusters << (s->cluster_bits - 3); 715 716 if (header.refcount_table_clusters > qcow2_max_refcount_clusters(s)) { 717 error_setg(errp, "Reference count table too large"); 718 ret = -EINVAL; 719 goto fail; 720 } 721 722 ret = validate_table_offset(bs, s->refcount_table_offset, 723 s->refcount_table_size, sizeof(uint64_t)); 724 if (ret < 0) { 725 error_setg(errp, "Invalid reference count table offset"); 726 goto fail; 727 } 728 729 /* Snapshot table offset/length */ 730 if (header.nb_snapshots > QCOW_MAX_SNAPSHOTS) { 731 error_setg(errp, "Too many snapshots"); 732 ret = -EINVAL; 733 goto fail; 734 } 735 736 ret = validate_table_offset(bs, header.snapshots_offset, 737 header.nb_snapshots, 738 sizeof(QCowSnapshotHeader)); 739 if (ret < 0) { 740 error_setg(errp, "Invalid snapshot table offset"); 741 goto fail; 742 } 743 744 /* read the level 1 table */ 745 if (header.l1_size > QCOW_MAX_L1_SIZE) { 746 error_setg(errp, "Active L1 table too large"); 747 ret = -EFBIG; 748 goto fail; 749 } 750 s->l1_size = header.l1_size; 751 752 l1_vm_state_index = size_to_l1(s, header.size); 753 if (l1_vm_state_index > INT_MAX) { 754 error_setg(errp, "Image is too big"); 755 ret = -EFBIG; 756 goto fail; 757 } 758 s->l1_vm_state_index = l1_vm_state_index; 759 760 /* the L1 table must contain at least enough entries to put 761 header.size bytes */ 762 if (s->l1_size < s->l1_vm_state_index) { 763 error_setg(errp, "L1 table is too small"); 764 ret = -EINVAL; 765 goto fail; 766 } 767 768 ret = validate_table_offset(bs, header.l1_table_offset, 769 header.l1_size, sizeof(uint64_t)); 770 if (ret < 0) { 771 error_setg(errp, "Invalid L1 table offset"); 772 goto fail; 773 } 774 s->l1_table_offset = header.l1_table_offset; 775 776 777 if (s->l1_size > 0) { 778 s->l1_table = qemu_try_blockalign(bs->file, 779 align_offset(s->l1_size * sizeof(uint64_t), 512)); 780 if (s->l1_table == NULL) { 781 error_setg(errp, "Could not allocate L1 table"); 782 ret = -ENOMEM; 783 goto fail; 784 } 785 ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table, 786 s->l1_size * sizeof(uint64_t)); 787 if (ret < 0) { 788 error_setg_errno(errp, -ret, "Could not read L1 table"); 789 goto fail; 790 } 791 for(i = 0;i < s->l1_size; i++) { 792 be64_to_cpus(&s->l1_table[i]); 793 } 794 } 795 796 /* get L2 table/refcount block cache size from command line options */ 797 opts = qemu_opts_create(&qcow2_runtime_opts, NULL, 0, &error_abort); 798 qemu_opts_absorb_qdict(opts, options, &local_err); 799 if (local_err) { 800 error_propagate(errp, local_err); 801 ret = -EINVAL; 802 goto fail; 803 } 804 805 read_cache_sizes(opts, &l2_cache_size, &refcount_cache_size, &local_err); 806 if (local_err) { 807 error_propagate(errp, local_err); 808 ret = -EINVAL; 809 goto fail; 810 } 811 812 l2_cache_size /= s->cluster_size; 813 if (l2_cache_size < MIN_L2_CACHE_SIZE) { 814 l2_cache_size = MIN_L2_CACHE_SIZE; 815 } 816 if (l2_cache_size > INT_MAX) { 817 error_setg(errp, "L2 cache size too big"); 818 ret = -EINVAL; 819 goto fail; 820 } 821 822 refcount_cache_size /= s->cluster_size; 823 if (refcount_cache_size < MIN_REFCOUNT_CACHE_SIZE) { 824 refcount_cache_size = MIN_REFCOUNT_CACHE_SIZE; 825 } 826 if (refcount_cache_size > INT_MAX) { 827 error_setg(errp, "Refcount cache size too big"); 828 ret = -EINVAL; 829 goto fail; 830 } 831 832 /* alloc L2 table/refcount block cache */ 833 s->l2_table_cache = qcow2_cache_create(bs, l2_cache_size); 834 s->refcount_block_cache = qcow2_cache_create(bs, refcount_cache_size); 835 if (s->l2_table_cache == NULL || s->refcount_block_cache == NULL) { 836 error_setg(errp, "Could not allocate metadata caches"); 837 ret = -ENOMEM; 838 goto fail; 839 } 840 841 s->cluster_cache = g_malloc(s->cluster_size); 842 /* one more sector for decompressed data alignment */ 843 s->cluster_data = qemu_try_blockalign(bs->file, QCOW_MAX_CRYPT_CLUSTERS 844 * s->cluster_size + 512); 845 if (s->cluster_data == NULL) { 846 error_setg(errp, "Could not allocate temporary cluster buffer"); 847 ret = -ENOMEM; 848 goto fail; 849 } 850 851 s->cluster_cache_offset = -1; 852 s->flags = flags; 853 854 ret = qcow2_refcount_init(bs); 855 if (ret != 0) { 856 error_setg_errno(errp, -ret, "Could not initialize refcount handling"); 857 goto fail; 858 } 859 860 QLIST_INIT(&s->cluster_allocs); 861 QTAILQ_INIT(&s->discards); 862 863 /* read qcow2 extensions */ 864 if (qcow2_read_extensions(bs, header.header_length, ext_end, NULL, 865 &local_err)) { 866 error_propagate(errp, local_err); 867 ret = -EINVAL; 868 goto fail; 869 } 870 871 /* read the backing file name */ 872 if (header.backing_file_offset != 0) { 873 len = header.backing_file_size; 874 if (len > MIN(1023, s->cluster_size - header.backing_file_offset) || 875 len >= sizeof(bs->backing_file)) { 876 error_setg(errp, "Backing file name too long"); 877 ret = -EINVAL; 878 goto fail; 879 } 880 ret = bdrv_pread(bs->file, header.backing_file_offset, 881 bs->backing_file, len); 882 if (ret < 0) { 883 error_setg_errno(errp, -ret, "Could not read backing file name"); 884 goto fail; 885 } 886 bs->backing_file[len] = '\0'; 887 } 888 889 /* Internal snapshots */ 890 s->snapshots_offset = header.snapshots_offset; 891 s->nb_snapshots = header.nb_snapshots; 892 893 ret = qcow2_read_snapshots(bs); 894 if (ret < 0) { 895 error_setg_errno(errp, -ret, "Could not read snapshots"); 896 goto fail; 897 } 898 899 /* Clear unknown autoclear feature bits */ 900 if (!bs->read_only && !(flags & BDRV_O_INCOMING) && s->autoclear_features) { 901 s->autoclear_features = 0; 902 ret = qcow2_update_header(bs); 903 if (ret < 0) { 904 error_setg_errno(errp, -ret, "Could not update qcow2 header"); 905 goto fail; 906 } 907 } 908 909 /* Initialise locks */ 910 qemu_co_mutex_init(&s->lock); 911 912 /* Repair image if dirty */ 913 if (!(flags & (BDRV_O_CHECK | BDRV_O_INCOMING)) && !bs->read_only && 914 (s->incompatible_features & QCOW2_INCOMPAT_DIRTY)) { 915 BdrvCheckResult result = {0}; 916 917 ret = qcow2_check(bs, &result, BDRV_FIX_ERRORS | BDRV_FIX_LEAKS); 918 if (ret < 0) { 919 error_setg_errno(errp, -ret, "Could not repair dirty image"); 920 goto fail; 921 } 922 } 923 924 /* Enable lazy_refcounts according to image and command line options */ 925 s->use_lazy_refcounts = qemu_opt_get_bool(opts, QCOW2_OPT_LAZY_REFCOUNTS, 926 (s->compatible_features & QCOW2_COMPAT_LAZY_REFCOUNTS)); 927 928 s->discard_passthrough[QCOW2_DISCARD_NEVER] = false; 929 s->discard_passthrough[QCOW2_DISCARD_ALWAYS] = true; 930 s->discard_passthrough[QCOW2_DISCARD_REQUEST] = 931 qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_REQUEST, 932 flags & BDRV_O_UNMAP); 933 s->discard_passthrough[QCOW2_DISCARD_SNAPSHOT] = 934 qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_SNAPSHOT, true); 935 s->discard_passthrough[QCOW2_DISCARD_OTHER] = 936 qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_OTHER, false); 937 938 opt_overlap_check = qemu_opt_get(opts, QCOW2_OPT_OVERLAP); 939 opt_overlap_check_template = qemu_opt_get(opts, QCOW2_OPT_OVERLAP_TEMPLATE); 940 if (opt_overlap_check_template && opt_overlap_check && 941 strcmp(opt_overlap_check_template, opt_overlap_check)) 942 { 943 error_setg(errp, "Conflicting values for qcow2 options '" 944 QCOW2_OPT_OVERLAP "' ('%s') and '" QCOW2_OPT_OVERLAP_TEMPLATE 945 "' ('%s')", opt_overlap_check, opt_overlap_check_template); 946 ret = -EINVAL; 947 goto fail; 948 } 949 if (!opt_overlap_check) { 950 opt_overlap_check = opt_overlap_check_template ?: "cached"; 951 } 952 953 if (!strcmp(opt_overlap_check, "none")) { 954 overlap_check_template = 0; 955 } else if (!strcmp(opt_overlap_check, "constant")) { 956 overlap_check_template = QCOW2_OL_CONSTANT; 957 } else if (!strcmp(opt_overlap_check, "cached")) { 958 overlap_check_template = QCOW2_OL_CACHED; 959 } else if (!strcmp(opt_overlap_check, "all")) { 960 overlap_check_template = QCOW2_OL_ALL; 961 } else { 962 error_setg(errp, "Unsupported value '%s' for qcow2 option " 963 "'overlap-check'. Allowed are either of the following: " 964 "none, constant, cached, all", opt_overlap_check); 965 ret = -EINVAL; 966 goto fail; 967 } 968 969 s->overlap_check = 0; 970 for (i = 0; i < QCOW2_OL_MAX_BITNR; i++) { 971 /* overlap-check defines a template bitmask, but every flag may be 972 * overwritten through the associated boolean option */ 973 s->overlap_check |= 974 qemu_opt_get_bool(opts, overlap_bool_option_names[i], 975 overlap_check_template & (1 << i)) << i; 976 } 977 978 qemu_opts_del(opts); 979 opts = NULL; 980 981 if (s->use_lazy_refcounts && s->qcow_version < 3) { 982 error_setg(errp, "Lazy refcounts require a qcow2 image with at least " 983 "qemu 1.1 compatibility level"); 984 ret = -EINVAL; 985 goto fail; 986 } 987 988 #ifdef DEBUG_ALLOC 989 { 990 BdrvCheckResult result = {0}; 991 qcow2_check_refcounts(bs, &result, 0); 992 } 993 #endif 994 return ret; 995 996 fail: 997 qemu_opts_del(opts); 998 g_free(s->unknown_header_fields); 999 cleanup_unknown_header_ext(bs); 1000 qcow2_free_snapshots(bs); 1001 qcow2_refcount_close(bs); 1002 qemu_vfree(s->l1_table); 1003 /* else pre-write overlap checks in cache_destroy may crash */ 1004 s->l1_table = NULL; 1005 if (s->l2_table_cache) { 1006 qcow2_cache_destroy(bs, s->l2_table_cache); 1007 } 1008 if (s->refcount_block_cache) { 1009 qcow2_cache_destroy(bs, s->refcount_block_cache); 1010 } 1011 g_free(s->cluster_cache); 1012 qemu_vfree(s->cluster_data); 1013 return ret; 1014 } 1015 1016 static void qcow2_refresh_limits(BlockDriverState *bs, Error **errp) 1017 { 1018 BDRVQcowState *s = bs->opaque; 1019 1020 bs->bl.write_zeroes_alignment = s->cluster_sectors; 1021 } 1022 1023 static int qcow2_set_key(BlockDriverState *bs, const char *key) 1024 { 1025 BDRVQcowState *s = bs->opaque; 1026 uint8_t keybuf[16]; 1027 int len, i; 1028 1029 memset(keybuf, 0, 16); 1030 len = strlen(key); 1031 if (len > 16) 1032 len = 16; 1033 /* XXX: we could compress the chars to 7 bits to increase 1034 entropy */ 1035 for(i = 0;i < len;i++) { 1036 keybuf[i] = key[i]; 1037 } 1038 s->crypt_method = s->crypt_method_header; 1039 1040 if (AES_set_encrypt_key(keybuf, 128, &s->aes_encrypt_key) != 0) 1041 return -1; 1042 if (AES_set_decrypt_key(keybuf, 128, &s->aes_decrypt_key) != 0) 1043 return -1; 1044 #if 0 1045 /* test */ 1046 { 1047 uint8_t in[16]; 1048 uint8_t out[16]; 1049 uint8_t tmp[16]; 1050 for(i=0;i<16;i++) 1051 in[i] = i; 1052 AES_encrypt(in, tmp, &s->aes_encrypt_key); 1053 AES_decrypt(tmp, out, &s->aes_decrypt_key); 1054 for(i = 0; i < 16; i++) 1055 printf(" %02x", tmp[i]); 1056 printf("\n"); 1057 for(i = 0; i < 16; i++) 1058 printf(" %02x", out[i]); 1059 printf("\n"); 1060 } 1061 #endif 1062 return 0; 1063 } 1064 1065 /* We have no actual commit/abort logic for qcow2, but we need to write out any 1066 * unwritten data if we reopen read-only. */ 1067 static int qcow2_reopen_prepare(BDRVReopenState *state, 1068 BlockReopenQueue *queue, Error **errp) 1069 { 1070 int ret; 1071 1072 if ((state->flags & BDRV_O_RDWR) == 0) { 1073 ret = bdrv_flush(state->bs); 1074 if (ret < 0) { 1075 return ret; 1076 } 1077 1078 ret = qcow2_mark_clean(state->bs); 1079 if (ret < 0) { 1080 return ret; 1081 } 1082 } 1083 1084 return 0; 1085 } 1086 1087 static int64_t coroutine_fn qcow2_co_get_block_status(BlockDriverState *bs, 1088 int64_t sector_num, int nb_sectors, int *pnum) 1089 { 1090 BDRVQcowState *s = bs->opaque; 1091 uint64_t cluster_offset; 1092 int index_in_cluster, ret; 1093 int64_t status = 0; 1094 1095 *pnum = nb_sectors; 1096 qemu_co_mutex_lock(&s->lock); 1097 ret = qcow2_get_cluster_offset(bs, sector_num << 9, pnum, &cluster_offset); 1098 qemu_co_mutex_unlock(&s->lock); 1099 if (ret < 0) { 1100 return ret; 1101 } 1102 1103 if (cluster_offset != 0 && ret != QCOW2_CLUSTER_COMPRESSED && 1104 !s->crypt_method) { 1105 index_in_cluster = sector_num & (s->cluster_sectors - 1); 1106 cluster_offset |= (index_in_cluster << BDRV_SECTOR_BITS); 1107 status |= BDRV_BLOCK_OFFSET_VALID | cluster_offset; 1108 } 1109 if (ret == QCOW2_CLUSTER_ZERO) { 1110 status |= BDRV_BLOCK_ZERO; 1111 } else if (ret != QCOW2_CLUSTER_UNALLOCATED) { 1112 status |= BDRV_BLOCK_DATA; 1113 } 1114 return status; 1115 } 1116 1117 /* handle reading after the end of the backing file */ 1118 int qcow2_backing_read1(BlockDriverState *bs, QEMUIOVector *qiov, 1119 int64_t sector_num, int nb_sectors) 1120 { 1121 int n1; 1122 if ((sector_num + nb_sectors) <= bs->total_sectors) 1123 return nb_sectors; 1124 if (sector_num >= bs->total_sectors) 1125 n1 = 0; 1126 else 1127 n1 = bs->total_sectors - sector_num; 1128 1129 qemu_iovec_memset(qiov, 512 * n1, 0, 512 * (nb_sectors - n1)); 1130 1131 return n1; 1132 } 1133 1134 static coroutine_fn int qcow2_co_readv(BlockDriverState *bs, int64_t sector_num, 1135 int remaining_sectors, QEMUIOVector *qiov) 1136 { 1137 BDRVQcowState *s = bs->opaque; 1138 int index_in_cluster, n1; 1139 int ret; 1140 int cur_nr_sectors; /* number of sectors in current iteration */ 1141 uint64_t cluster_offset = 0; 1142 uint64_t bytes_done = 0; 1143 QEMUIOVector hd_qiov; 1144 uint8_t *cluster_data = NULL; 1145 1146 qemu_iovec_init(&hd_qiov, qiov->niov); 1147 1148 qemu_co_mutex_lock(&s->lock); 1149 1150 while (remaining_sectors != 0) { 1151 1152 /* prepare next request */ 1153 cur_nr_sectors = remaining_sectors; 1154 if (s->crypt_method) { 1155 cur_nr_sectors = MIN(cur_nr_sectors, 1156 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors); 1157 } 1158 1159 ret = qcow2_get_cluster_offset(bs, sector_num << 9, 1160 &cur_nr_sectors, &cluster_offset); 1161 if (ret < 0) { 1162 goto fail; 1163 } 1164 1165 index_in_cluster = sector_num & (s->cluster_sectors - 1); 1166 1167 qemu_iovec_reset(&hd_qiov); 1168 qemu_iovec_concat(&hd_qiov, qiov, bytes_done, 1169 cur_nr_sectors * 512); 1170 1171 switch (ret) { 1172 case QCOW2_CLUSTER_UNALLOCATED: 1173 1174 if (bs->backing_hd) { 1175 /* read from the base image */ 1176 n1 = qcow2_backing_read1(bs->backing_hd, &hd_qiov, 1177 sector_num, cur_nr_sectors); 1178 if (n1 > 0) { 1179 QEMUIOVector local_qiov; 1180 1181 qemu_iovec_init(&local_qiov, hd_qiov.niov); 1182 qemu_iovec_concat(&local_qiov, &hd_qiov, 0, 1183 n1 * BDRV_SECTOR_SIZE); 1184 1185 BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO); 1186 qemu_co_mutex_unlock(&s->lock); 1187 ret = bdrv_co_readv(bs->backing_hd, sector_num, 1188 n1, &local_qiov); 1189 qemu_co_mutex_lock(&s->lock); 1190 1191 qemu_iovec_destroy(&local_qiov); 1192 1193 if (ret < 0) { 1194 goto fail; 1195 } 1196 } 1197 } else { 1198 /* Note: in this case, no need to wait */ 1199 qemu_iovec_memset(&hd_qiov, 0, 0, 512 * cur_nr_sectors); 1200 } 1201 break; 1202 1203 case QCOW2_CLUSTER_ZERO: 1204 qemu_iovec_memset(&hd_qiov, 0, 0, 512 * cur_nr_sectors); 1205 break; 1206 1207 case QCOW2_CLUSTER_COMPRESSED: 1208 /* add AIO support for compressed blocks ? */ 1209 ret = qcow2_decompress_cluster(bs, cluster_offset); 1210 if (ret < 0) { 1211 goto fail; 1212 } 1213 1214 qemu_iovec_from_buf(&hd_qiov, 0, 1215 s->cluster_cache + index_in_cluster * 512, 1216 512 * cur_nr_sectors); 1217 break; 1218 1219 case QCOW2_CLUSTER_NORMAL: 1220 if ((cluster_offset & 511) != 0) { 1221 ret = -EIO; 1222 goto fail; 1223 } 1224 1225 if (s->crypt_method) { 1226 /* 1227 * For encrypted images, read everything into a temporary 1228 * contiguous buffer on which the AES functions can work. 1229 */ 1230 if (!cluster_data) { 1231 cluster_data = 1232 qemu_try_blockalign(bs->file, QCOW_MAX_CRYPT_CLUSTERS 1233 * s->cluster_size); 1234 if (cluster_data == NULL) { 1235 ret = -ENOMEM; 1236 goto fail; 1237 } 1238 } 1239 1240 assert(cur_nr_sectors <= 1241 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors); 1242 qemu_iovec_reset(&hd_qiov); 1243 qemu_iovec_add(&hd_qiov, cluster_data, 1244 512 * cur_nr_sectors); 1245 } 1246 1247 BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO); 1248 qemu_co_mutex_unlock(&s->lock); 1249 ret = bdrv_co_readv(bs->file, 1250 (cluster_offset >> 9) + index_in_cluster, 1251 cur_nr_sectors, &hd_qiov); 1252 qemu_co_mutex_lock(&s->lock); 1253 if (ret < 0) { 1254 goto fail; 1255 } 1256 if (s->crypt_method) { 1257 qcow2_encrypt_sectors(s, sector_num, cluster_data, 1258 cluster_data, cur_nr_sectors, 0, &s->aes_decrypt_key); 1259 qemu_iovec_from_buf(qiov, bytes_done, 1260 cluster_data, 512 * cur_nr_sectors); 1261 } 1262 break; 1263 1264 default: 1265 g_assert_not_reached(); 1266 ret = -EIO; 1267 goto fail; 1268 } 1269 1270 remaining_sectors -= cur_nr_sectors; 1271 sector_num += cur_nr_sectors; 1272 bytes_done += cur_nr_sectors * 512; 1273 } 1274 ret = 0; 1275 1276 fail: 1277 qemu_co_mutex_unlock(&s->lock); 1278 1279 qemu_iovec_destroy(&hd_qiov); 1280 qemu_vfree(cluster_data); 1281 1282 return ret; 1283 } 1284 1285 static coroutine_fn int qcow2_co_writev(BlockDriverState *bs, 1286 int64_t sector_num, 1287 int remaining_sectors, 1288 QEMUIOVector *qiov) 1289 { 1290 BDRVQcowState *s = bs->opaque; 1291 int index_in_cluster; 1292 int ret; 1293 int cur_nr_sectors; /* number of sectors in current iteration */ 1294 uint64_t cluster_offset; 1295 QEMUIOVector hd_qiov; 1296 uint64_t bytes_done = 0; 1297 uint8_t *cluster_data = NULL; 1298 QCowL2Meta *l2meta = NULL; 1299 1300 trace_qcow2_writev_start_req(qemu_coroutine_self(), sector_num, 1301 remaining_sectors); 1302 1303 qemu_iovec_init(&hd_qiov, qiov->niov); 1304 1305 s->cluster_cache_offset = -1; /* disable compressed cache */ 1306 1307 qemu_co_mutex_lock(&s->lock); 1308 1309 while (remaining_sectors != 0) { 1310 1311 l2meta = NULL; 1312 1313 trace_qcow2_writev_start_part(qemu_coroutine_self()); 1314 index_in_cluster = sector_num & (s->cluster_sectors - 1); 1315 cur_nr_sectors = remaining_sectors; 1316 if (s->crypt_method && 1317 cur_nr_sectors > 1318 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors - index_in_cluster) { 1319 cur_nr_sectors = 1320 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors - index_in_cluster; 1321 } 1322 1323 ret = qcow2_alloc_cluster_offset(bs, sector_num << 9, 1324 &cur_nr_sectors, &cluster_offset, &l2meta); 1325 if (ret < 0) { 1326 goto fail; 1327 } 1328 1329 assert((cluster_offset & 511) == 0); 1330 1331 qemu_iovec_reset(&hd_qiov); 1332 qemu_iovec_concat(&hd_qiov, qiov, bytes_done, 1333 cur_nr_sectors * 512); 1334 1335 if (s->crypt_method) { 1336 if (!cluster_data) { 1337 cluster_data = qemu_try_blockalign(bs->file, 1338 QCOW_MAX_CRYPT_CLUSTERS 1339 * s->cluster_size); 1340 if (cluster_data == NULL) { 1341 ret = -ENOMEM; 1342 goto fail; 1343 } 1344 } 1345 1346 assert(hd_qiov.size <= 1347 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size); 1348 qemu_iovec_to_buf(&hd_qiov, 0, cluster_data, hd_qiov.size); 1349 1350 qcow2_encrypt_sectors(s, sector_num, cluster_data, 1351 cluster_data, cur_nr_sectors, 1, &s->aes_encrypt_key); 1352 1353 qemu_iovec_reset(&hd_qiov); 1354 qemu_iovec_add(&hd_qiov, cluster_data, 1355 cur_nr_sectors * 512); 1356 } 1357 1358 ret = qcow2_pre_write_overlap_check(bs, 0, 1359 cluster_offset + index_in_cluster * BDRV_SECTOR_SIZE, 1360 cur_nr_sectors * BDRV_SECTOR_SIZE); 1361 if (ret < 0) { 1362 goto fail; 1363 } 1364 1365 qemu_co_mutex_unlock(&s->lock); 1366 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO); 1367 trace_qcow2_writev_data(qemu_coroutine_self(), 1368 (cluster_offset >> 9) + index_in_cluster); 1369 ret = bdrv_co_writev(bs->file, 1370 (cluster_offset >> 9) + index_in_cluster, 1371 cur_nr_sectors, &hd_qiov); 1372 qemu_co_mutex_lock(&s->lock); 1373 if (ret < 0) { 1374 goto fail; 1375 } 1376 1377 while (l2meta != NULL) { 1378 QCowL2Meta *next; 1379 1380 ret = qcow2_alloc_cluster_link_l2(bs, l2meta); 1381 if (ret < 0) { 1382 goto fail; 1383 } 1384 1385 /* Take the request off the list of running requests */ 1386 if (l2meta->nb_clusters != 0) { 1387 QLIST_REMOVE(l2meta, next_in_flight); 1388 } 1389 1390 qemu_co_queue_restart_all(&l2meta->dependent_requests); 1391 1392 next = l2meta->next; 1393 g_free(l2meta); 1394 l2meta = next; 1395 } 1396 1397 remaining_sectors -= cur_nr_sectors; 1398 sector_num += cur_nr_sectors; 1399 bytes_done += cur_nr_sectors * 512; 1400 trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_nr_sectors); 1401 } 1402 ret = 0; 1403 1404 fail: 1405 qemu_co_mutex_unlock(&s->lock); 1406 1407 while (l2meta != NULL) { 1408 QCowL2Meta *next; 1409 1410 if (l2meta->nb_clusters != 0) { 1411 QLIST_REMOVE(l2meta, next_in_flight); 1412 } 1413 qemu_co_queue_restart_all(&l2meta->dependent_requests); 1414 1415 next = l2meta->next; 1416 g_free(l2meta); 1417 l2meta = next; 1418 } 1419 1420 qemu_iovec_destroy(&hd_qiov); 1421 qemu_vfree(cluster_data); 1422 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret); 1423 1424 return ret; 1425 } 1426 1427 static void qcow2_close(BlockDriverState *bs) 1428 { 1429 BDRVQcowState *s = bs->opaque; 1430 qemu_vfree(s->l1_table); 1431 /* else pre-write overlap checks in cache_destroy may crash */ 1432 s->l1_table = NULL; 1433 1434 if (!(bs->open_flags & BDRV_O_INCOMING)) { 1435 int ret1, ret2; 1436 1437 ret1 = qcow2_cache_flush(bs, s->l2_table_cache); 1438 ret2 = qcow2_cache_flush(bs, s->refcount_block_cache); 1439 1440 if (ret1) { 1441 error_report("Failed to flush the L2 table cache: %s", 1442 strerror(-ret1)); 1443 } 1444 if (ret2) { 1445 error_report("Failed to flush the refcount block cache: %s", 1446 strerror(-ret2)); 1447 } 1448 1449 if (!ret1 && !ret2) { 1450 qcow2_mark_clean(bs); 1451 } 1452 } 1453 1454 qcow2_cache_destroy(bs, s->l2_table_cache); 1455 qcow2_cache_destroy(bs, s->refcount_block_cache); 1456 1457 g_free(s->unknown_header_fields); 1458 cleanup_unknown_header_ext(bs); 1459 1460 g_free(s->cluster_cache); 1461 qemu_vfree(s->cluster_data); 1462 qcow2_refcount_close(bs); 1463 qcow2_free_snapshots(bs); 1464 } 1465 1466 static void qcow2_invalidate_cache(BlockDriverState *bs, Error **errp) 1467 { 1468 BDRVQcowState *s = bs->opaque; 1469 int flags = s->flags; 1470 AES_KEY aes_encrypt_key; 1471 AES_KEY aes_decrypt_key; 1472 uint32_t crypt_method = 0; 1473 QDict *options; 1474 Error *local_err = NULL; 1475 int ret; 1476 1477 /* 1478 * Backing files are read-only which makes all of their metadata immutable, 1479 * that means we don't have to worry about reopening them here. 1480 */ 1481 1482 if (s->crypt_method) { 1483 crypt_method = s->crypt_method; 1484 memcpy(&aes_encrypt_key, &s->aes_encrypt_key, sizeof(aes_encrypt_key)); 1485 memcpy(&aes_decrypt_key, &s->aes_decrypt_key, sizeof(aes_decrypt_key)); 1486 } 1487 1488 qcow2_close(bs); 1489 1490 bdrv_invalidate_cache(bs->file, &local_err); 1491 if (local_err) { 1492 error_propagate(errp, local_err); 1493 return; 1494 } 1495 1496 memset(s, 0, sizeof(BDRVQcowState)); 1497 options = qdict_clone_shallow(bs->options); 1498 1499 ret = qcow2_open(bs, options, flags, &local_err); 1500 QDECREF(options); 1501 if (local_err) { 1502 error_setg(errp, "Could not reopen qcow2 layer: %s", 1503 error_get_pretty(local_err)); 1504 error_free(local_err); 1505 return; 1506 } else if (ret < 0) { 1507 error_setg_errno(errp, -ret, "Could not reopen qcow2 layer"); 1508 return; 1509 } 1510 1511 if (crypt_method) { 1512 s->crypt_method = crypt_method; 1513 memcpy(&s->aes_encrypt_key, &aes_encrypt_key, sizeof(aes_encrypt_key)); 1514 memcpy(&s->aes_decrypt_key, &aes_decrypt_key, sizeof(aes_decrypt_key)); 1515 } 1516 } 1517 1518 static size_t header_ext_add(char *buf, uint32_t magic, const void *s, 1519 size_t len, size_t buflen) 1520 { 1521 QCowExtension *ext_backing_fmt = (QCowExtension*) buf; 1522 size_t ext_len = sizeof(QCowExtension) + ((len + 7) & ~7); 1523 1524 if (buflen < ext_len) { 1525 return -ENOSPC; 1526 } 1527 1528 *ext_backing_fmt = (QCowExtension) { 1529 .magic = cpu_to_be32(magic), 1530 .len = cpu_to_be32(len), 1531 }; 1532 memcpy(buf + sizeof(QCowExtension), s, len); 1533 1534 return ext_len; 1535 } 1536 1537 /* 1538 * Updates the qcow2 header, including the variable length parts of it, i.e. 1539 * the backing file name and all extensions. qcow2 was not designed to allow 1540 * such changes, so if we run out of space (we can only use the first cluster) 1541 * this function may fail. 1542 * 1543 * Returns 0 on success, -errno in error cases. 1544 */ 1545 int qcow2_update_header(BlockDriverState *bs) 1546 { 1547 BDRVQcowState *s = bs->opaque; 1548 QCowHeader *header; 1549 char *buf; 1550 size_t buflen = s->cluster_size; 1551 int ret; 1552 uint64_t total_size; 1553 uint32_t refcount_table_clusters; 1554 size_t header_length; 1555 Qcow2UnknownHeaderExtension *uext; 1556 1557 buf = qemu_blockalign(bs, buflen); 1558 1559 /* Header structure */ 1560 header = (QCowHeader*) buf; 1561 1562 if (buflen < sizeof(*header)) { 1563 ret = -ENOSPC; 1564 goto fail; 1565 } 1566 1567 header_length = sizeof(*header) + s->unknown_header_fields_size; 1568 total_size = bs->total_sectors * BDRV_SECTOR_SIZE; 1569 refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3); 1570 1571 *header = (QCowHeader) { 1572 /* Version 2 fields */ 1573 .magic = cpu_to_be32(QCOW_MAGIC), 1574 .version = cpu_to_be32(s->qcow_version), 1575 .backing_file_offset = 0, 1576 .backing_file_size = 0, 1577 .cluster_bits = cpu_to_be32(s->cluster_bits), 1578 .size = cpu_to_be64(total_size), 1579 .crypt_method = cpu_to_be32(s->crypt_method_header), 1580 .l1_size = cpu_to_be32(s->l1_size), 1581 .l1_table_offset = cpu_to_be64(s->l1_table_offset), 1582 .refcount_table_offset = cpu_to_be64(s->refcount_table_offset), 1583 .refcount_table_clusters = cpu_to_be32(refcount_table_clusters), 1584 .nb_snapshots = cpu_to_be32(s->nb_snapshots), 1585 .snapshots_offset = cpu_to_be64(s->snapshots_offset), 1586 1587 /* Version 3 fields */ 1588 .incompatible_features = cpu_to_be64(s->incompatible_features), 1589 .compatible_features = cpu_to_be64(s->compatible_features), 1590 .autoclear_features = cpu_to_be64(s->autoclear_features), 1591 .refcount_order = cpu_to_be32(s->refcount_order), 1592 .header_length = cpu_to_be32(header_length), 1593 }; 1594 1595 /* For older versions, write a shorter header */ 1596 switch (s->qcow_version) { 1597 case 2: 1598 ret = offsetof(QCowHeader, incompatible_features); 1599 break; 1600 case 3: 1601 ret = sizeof(*header); 1602 break; 1603 default: 1604 ret = -EINVAL; 1605 goto fail; 1606 } 1607 1608 buf += ret; 1609 buflen -= ret; 1610 memset(buf, 0, buflen); 1611 1612 /* Preserve any unknown field in the header */ 1613 if (s->unknown_header_fields_size) { 1614 if (buflen < s->unknown_header_fields_size) { 1615 ret = -ENOSPC; 1616 goto fail; 1617 } 1618 1619 memcpy(buf, s->unknown_header_fields, s->unknown_header_fields_size); 1620 buf += s->unknown_header_fields_size; 1621 buflen -= s->unknown_header_fields_size; 1622 } 1623 1624 /* Backing file format header extension */ 1625 if (*bs->backing_format) { 1626 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BACKING_FORMAT, 1627 bs->backing_format, strlen(bs->backing_format), 1628 buflen); 1629 if (ret < 0) { 1630 goto fail; 1631 } 1632 1633 buf += ret; 1634 buflen -= ret; 1635 } 1636 1637 /* Feature table */ 1638 Qcow2Feature features[] = { 1639 { 1640 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE, 1641 .bit = QCOW2_INCOMPAT_DIRTY_BITNR, 1642 .name = "dirty bit", 1643 }, 1644 { 1645 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE, 1646 .bit = QCOW2_INCOMPAT_CORRUPT_BITNR, 1647 .name = "corrupt bit", 1648 }, 1649 { 1650 .type = QCOW2_FEAT_TYPE_COMPATIBLE, 1651 .bit = QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR, 1652 .name = "lazy refcounts", 1653 }, 1654 }; 1655 1656 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_FEATURE_TABLE, 1657 features, sizeof(features), buflen); 1658 if (ret < 0) { 1659 goto fail; 1660 } 1661 buf += ret; 1662 buflen -= ret; 1663 1664 /* Keep unknown header extensions */ 1665 QLIST_FOREACH(uext, &s->unknown_header_ext, next) { 1666 ret = header_ext_add(buf, uext->magic, uext->data, uext->len, buflen); 1667 if (ret < 0) { 1668 goto fail; 1669 } 1670 1671 buf += ret; 1672 buflen -= ret; 1673 } 1674 1675 /* End of header extensions */ 1676 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_END, NULL, 0, buflen); 1677 if (ret < 0) { 1678 goto fail; 1679 } 1680 1681 buf += ret; 1682 buflen -= ret; 1683 1684 /* Backing file name */ 1685 if (*bs->backing_file) { 1686 size_t backing_file_len = strlen(bs->backing_file); 1687 1688 if (buflen < backing_file_len) { 1689 ret = -ENOSPC; 1690 goto fail; 1691 } 1692 1693 /* Using strncpy is ok here, since buf is not NUL-terminated. */ 1694 strncpy(buf, bs->backing_file, buflen); 1695 1696 header->backing_file_offset = cpu_to_be64(buf - ((char*) header)); 1697 header->backing_file_size = cpu_to_be32(backing_file_len); 1698 } 1699 1700 /* Write the new header */ 1701 ret = bdrv_pwrite(bs->file, 0, header, s->cluster_size); 1702 if (ret < 0) { 1703 goto fail; 1704 } 1705 1706 ret = 0; 1707 fail: 1708 qemu_vfree(header); 1709 return ret; 1710 } 1711 1712 static int qcow2_change_backing_file(BlockDriverState *bs, 1713 const char *backing_file, const char *backing_fmt) 1714 { 1715 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: ""); 1716 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: ""); 1717 1718 return qcow2_update_header(bs); 1719 } 1720 1721 static int preallocate(BlockDriverState *bs) 1722 { 1723 uint64_t nb_sectors; 1724 uint64_t offset; 1725 uint64_t host_offset = 0; 1726 int num; 1727 int ret; 1728 QCowL2Meta *meta; 1729 1730 nb_sectors = bdrv_nb_sectors(bs); 1731 offset = 0; 1732 1733 while (nb_sectors) { 1734 num = MIN(nb_sectors, INT_MAX >> BDRV_SECTOR_BITS); 1735 ret = qcow2_alloc_cluster_offset(bs, offset, &num, 1736 &host_offset, &meta); 1737 if (ret < 0) { 1738 return ret; 1739 } 1740 1741 while (meta) { 1742 QCowL2Meta *next = meta->next; 1743 1744 ret = qcow2_alloc_cluster_link_l2(bs, meta); 1745 if (ret < 0) { 1746 qcow2_free_any_clusters(bs, meta->alloc_offset, 1747 meta->nb_clusters, QCOW2_DISCARD_NEVER); 1748 return ret; 1749 } 1750 1751 /* There are no dependent requests, but we need to remove our 1752 * request from the list of in-flight requests */ 1753 QLIST_REMOVE(meta, next_in_flight); 1754 1755 g_free(meta); 1756 meta = next; 1757 } 1758 1759 /* TODO Preallocate data if requested */ 1760 1761 nb_sectors -= num; 1762 offset += num << BDRV_SECTOR_BITS; 1763 } 1764 1765 /* 1766 * It is expected that the image file is large enough to actually contain 1767 * all of the allocated clusters (otherwise we get failing reads after 1768 * EOF). Extend the image to the last allocated sector. 1769 */ 1770 if (host_offset != 0) { 1771 uint8_t buf[BDRV_SECTOR_SIZE]; 1772 memset(buf, 0, BDRV_SECTOR_SIZE); 1773 ret = bdrv_write(bs->file, (host_offset >> BDRV_SECTOR_BITS) + num - 1, 1774 buf, 1); 1775 if (ret < 0) { 1776 return ret; 1777 } 1778 } 1779 1780 return 0; 1781 } 1782 1783 static int qcow2_create2(const char *filename, int64_t total_size, 1784 const char *backing_file, const char *backing_format, 1785 int flags, size_t cluster_size, PreallocMode prealloc, 1786 QemuOpts *opts, int version, 1787 Error **errp) 1788 { 1789 /* Calculate cluster_bits */ 1790 int cluster_bits; 1791 cluster_bits = ffs(cluster_size) - 1; 1792 if (cluster_bits < MIN_CLUSTER_BITS || cluster_bits > MAX_CLUSTER_BITS || 1793 (1 << cluster_bits) != cluster_size) 1794 { 1795 error_setg(errp, "Cluster size must be a power of two between %d and " 1796 "%dk", 1 << MIN_CLUSTER_BITS, 1 << (MAX_CLUSTER_BITS - 10)); 1797 return -EINVAL; 1798 } 1799 1800 /* 1801 * Open the image file and write a minimal qcow2 header. 1802 * 1803 * We keep things simple and start with a zero-sized image. We also 1804 * do without refcount blocks or a L1 table for now. We'll fix the 1805 * inconsistency later. 1806 * 1807 * We do need a refcount table because growing the refcount table means 1808 * allocating two new refcount blocks - the seconds of which would be at 1809 * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file 1810 * size for any qcow2 image. 1811 */ 1812 BlockDriverState* bs; 1813 QCowHeader *header; 1814 uint64_t* refcount_table; 1815 Error *local_err = NULL; 1816 int ret; 1817 1818 if (prealloc == PREALLOC_MODE_FULL || prealloc == PREALLOC_MODE_FALLOC) { 1819 int64_t meta_size = 0; 1820 uint64_t nreftablee, nrefblocke, nl1e, nl2e; 1821 int64_t aligned_total_size = align_offset(total_size, cluster_size); 1822 1823 /* header: 1 cluster */ 1824 meta_size += cluster_size; 1825 1826 /* total size of L2 tables */ 1827 nl2e = aligned_total_size / cluster_size; 1828 nl2e = align_offset(nl2e, cluster_size / sizeof(uint64_t)); 1829 meta_size += nl2e * sizeof(uint64_t); 1830 1831 /* total size of L1 tables */ 1832 nl1e = nl2e * sizeof(uint64_t) / cluster_size; 1833 nl1e = align_offset(nl1e, cluster_size / sizeof(uint64_t)); 1834 meta_size += nl1e * sizeof(uint64_t); 1835 1836 /* total size of refcount blocks 1837 * 1838 * note: every host cluster is reference-counted, including metadata 1839 * (even refcount blocks are recursively included). 1840 * Let: 1841 * a = total_size (this is the guest disk size) 1842 * m = meta size not including refcount blocks and refcount tables 1843 * c = cluster size 1844 * y1 = number of refcount blocks entries 1845 * y2 = meta size including everything 1846 * then, 1847 * y1 = (y2 + a)/c 1848 * y2 = y1 * sizeof(u16) + y1 * sizeof(u16) * sizeof(u64) / c + m 1849 * we can get y1: 1850 * y1 = (a + m) / (c - sizeof(u16) - sizeof(u16) * sizeof(u64) / c) 1851 */ 1852 nrefblocke = (aligned_total_size + meta_size + cluster_size) / 1853 (cluster_size - sizeof(uint16_t) - 1854 1.0 * sizeof(uint16_t) * sizeof(uint64_t) / cluster_size); 1855 nrefblocke = align_offset(nrefblocke, cluster_size / sizeof(uint16_t)); 1856 meta_size += nrefblocke * sizeof(uint16_t); 1857 1858 /* total size of refcount tables */ 1859 nreftablee = nrefblocke * sizeof(uint16_t) / cluster_size; 1860 nreftablee = align_offset(nreftablee, cluster_size / sizeof(uint64_t)); 1861 meta_size += nreftablee * sizeof(uint64_t); 1862 1863 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, 1864 aligned_total_size + meta_size, &error_abort); 1865 qemu_opt_set(opts, BLOCK_OPT_PREALLOC, PreallocMode_lookup[prealloc], 1866 &error_abort); 1867 } 1868 1869 ret = bdrv_create_file(filename, opts, &local_err); 1870 if (ret < 0) { 1871 error_propagate(errp, local_err); 1872 return ret; 1873 } 1874 1875 bs = NULL; 1876 ret = bdrv_open(&bs, filename, NULL, NULL, BDRV_O_RDWR | BDRV_O_PROTOCOL, 1877 NULL, &local_err); 1878 if (ret < 0) { 1879 error_propagate(errp, local_err); 1880 return ret; 1881 } 1882 1883 /* Write the header */ 1884 QEMU_BUILD_BUG_ON((1 << MIN_CLUSTER_BITS) < sizeof(*header)); 1885 header = g_malloc0(cluster_size); 1886 *header = (QCowHeader) { 1887 .magic = cpu_to_be32(QCOW_MAGIC), 1888 .version = cpu_to_be32(version), 1889 .cluster_bits = cpu_to_be32(cluster_bits), 1890 .size = cpu_to_be64(0), 1891 .l1_table_offset = cpu_to_be64(0), 1892 .l1_size = cpu_to_be32(0), 1893 .refcount_table_offset = cpu_to_be64(cluster_size), 1894 .refcount_table_clusters = cpu_to_be32(1), 1895 .refcount_order = cpu_to_be32(4), 1896 .header_length = cpu_to_be32(sizeof(*header)), 1897 }; 1898 1899 if (flags & BLOCK_FLAG_ENCRYPT) { 1900 header->crypt_method = cpu_to_be32(QCOW_CRYPT_AES); 1901 } else { 1902 header->crypt_method = cpu_to_be32(QCOW_CRYPT_NONE); 1903 } 1904 1905 if (flags & BLOCK_FLAG_LAZY_REFCOUNTS) { 1906 header->compatible_features |= 1907 cpu_to_be64(QCOW2_COMPAT_LAZY_REFCOUNTS); 1908 } 1909 1910 ret = bdrv_pwrite(bs, 0, header, cluster_size); 1911 g_free(header); 1912 if (ret < 0) { 1913 error_setg_errno(errp, -ret, "Could not write qcow2 header"); 1914 goto out; 1915 } 1916 1917 /* Write a refcount table with one refcount block */ 1918 refcount_table = g_malloc0(2 * cluster_size); 1919 refcount_table[0] = cpu_to_be64(2 * cluster_size); 1920 ret = bdrv_pwrite(bs, cluster_size, refcount_table, 2 * cluster_size); 1921 g_free(refcount_table); 1922 1923 if (ret < 0) { 1924 error_setg_errno(errp, -ret, "Could not write refcount table"); 1925 goto out; 1926 } 1927 1928 bdrv_unref(bs); 1929 bs = NULL; 1930 1931 /* 1932 * And now open the image and make it consistent first (i.e. increase the 1933 * refcount of the cluster that is occupied by the header and the refcount 1934 * table) 1935 */ 1936 ret = bdrv_open(&bs, filename, NULL, NULL, 1937 BDRV_O_RDWR | BDRV_O_CACHE_WB | BDRV_O_NO_FLUSH, 1938 &bdrv_qcow2, &local_err); 1939 if (ret < 0) { 1940 error_propagate(errp, local_err); 1941 goto out; 1942 } 1943 1944 ret = qcow2_alloc_clusters(bs, 3 * cluster_size); 1945 if (ret < 0) { 1946 error_setg_errno(errp, -ret, "Could not allocate clusters for qcow2 " 1947 "header and refcount table"); 1948 goto out; 1949 1950 } else if (ret != 0) { 1951 error_report("Huh, first cluster in empty image is already in use?"); 1952 abort(); 1953 } 1954 1955 /* Okay, now that we have a valid image, let's give it the right size */ 1956 ret = bdrv_truncate(bs, total_size); 1957 if (ret < 0) { 1958 error_setg_errno(errp, -ret, "Could not resize image"); 1959 goto out; 1960 } 1961 1962 /* Want a backing file? There you go.*/ 1963 if (backing_file) { 1964 ret = bdrv_change_backing_file(bs, backing_file, backing_format); 1965 if (ret < 0) { 1966 error_setg_errno(errp, -ret, "Could not assign backing file '%s' " 1967 "with format '%s'", backing_file, backing_format); 1968 goto out; 1969 } 1970 } 1971 1972 /* And if we're supposed to preallocate metadata, do that now */ 1973 if (prealloc != PREALLOC_MODE_OFF) { 1974 BDRVQcowState *s = bs->opaque; 1975 qemu_co_mutex_lock(&s->lock); 1976 ret = preallocate(bs); 1977 qemu_co_mutex_unlock(&s->lock); 1978 if (ret < 0) { 1979 error_setg_errno(errp, -ret, "Could not preallocate metadata"); 1980 goto out; 1981 } 1982 } 1983 1984 bdrv_unref(bs); 1985 bs = NULL; 1986 1987 /* Reopen the image without BDRV_O_NO_FLUSH to flush it before returning */ 1988 ret = bdrv_open(&bs, filename, NULL, NULL, 1989 BDRV_O_RDWR | BDRV_O_CACHE_WB | BDRV_O_NO_BACKING, 1990 &bdrv_qcow2, &local_err); 1991 if (local_err) { 1992 error_propagate(errp, local_err); 1993 goto out; 1994 } 1995 1996 ret = 0; 1997 out: 1998 if (bs) { 1999 bdrv_unref(bs); 2000 } 2001 return ret; 2002 } 2003 2004 static int qcow2_create(const char *filename, QemuOpts *opts, Error **errp) 2005 { 2006 char *backing_file = NULL; 2007 char *backing_fmt = NULL; 2008 char *buf = NULL; 2009 uint64_t size = 0; 2010 int flags = 0; 2011 size_t cluster_size = DEFAULT_CLUSTER_SIZE; 2012 PreallocMode prealloc; 2013 int version = 3; 2014 Error *local_err = NULL; 2015 int ret; 2016 2017 /* Read out options */ 2018 size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0), 2019 BDRV_SECTOR_SIZE); 2020 backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE); 2021 backing_fmt = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FMT); 2022 if (qemu_opt_get_bool_del(opts, BLOCK_OPT_ENCRYPT, false)) { 2023 flags |= BLOCK_FLAG_ENCRYPT; 2024 } 2025 cluster_size = qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE, 2026 DEFAULT_CLUSTER_SIZE); 2027 buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC); 2028 prealloc = qapi_enum_parse(PreallocMode_lookup, buf, 2029 PREALLOC_MODE_MAX, PREALLOC_MODE_OFF, 2030 &local_err); 2031 if (local_err) { 2032 error_propagate(errp, local_err); 2033 ret = -EINVAL; 2034 goto finish; 2035 } 2036 g_free(buf); 2037 buf = qemu_opt_get_del(opts, BLOCK_OPT_COMPAT_LEVEL); 2038 if (!buf) { 2039 /* keep the default */ 2040 } else if (!strcmp(buf, "0.10")) { 2041 version = 2; 2042 } else if (!strcmp(buf, "1.1")) { 2043 version = 3; 2044 } else { 2045 error_setg(errp, "Invalid compatibility level: '%s'", buf); 2046 ret = -EINVAL; 2047 goto finish; 2048 } 2049 2050 if (qemu_opt_get_bool_del(opts, BLOCK_OPT_LAZY_REFCOUNTS, false)) { 2051 flags |= BLOCK_FLAG_LAZY_REFCOUNTS; 2052 } 2053 2054 if (backing_file && prealloc != PREALLOC_MODE_OFF) { 2055 error_setg(errp, "Backing file and preallocation cannot be used at " 2056 "the same time"); 2057 ret = -EINVAL; 2058 goto finish; 2059 } 2060 2061 if (version < 3 && (flags & BLOCK_FLAG_LAZY_REFCOUNTS)) { 2062 error_setg(errp, "Lazy refcounts only supported with compatibility " 2063 "level 1.1 and above (use compat=1.1 or greater)"); 2064 ret = -EINVAL; 2065 goto finish; 2066 } 2067 2068 ret = qcow2_create2(filename, size, backing_file, backing_fmt, flags, 2069 cluster_size, prealloc, opts, version, &local_err); 2070 if (local_err) { 2071 error_propagate(errp, local_err); 2072 } 2073 2074 finish: 2075 g_free(backing_file); 2076 g_free(backing_fmt); 2077 g_free(buf); 2078 return ret; 2079 } 2080 2081 static coroutine_fn int qcow2_co_write_zeroes(BlockDriverState *bs, 2082 int64_t sector_num, int nb_sectors, BdrvRequestFlags flags) 2083 { 2084 int ret; 2085 BDRVQcowState *s = bs->opaque; 2086 2087 /* Emulate misaligned zero writes */ 2088 if (sector_num % s->cluster_sectors || nb_sectors % s->cluster_sectors) { 2089 return -ENOTSUP; 2090 } 2091 2092 /* Whatever is left can use real zero clusters */ 2093 qemu_co_mutex_lock(&s->lock); 2094 ret = qcow2_zero_clusters(bs, sector_num << BDRV_SECTOR_BITS, 2095 nb_sectors); 2096 qemu_co_mutex_unlock(&s->lock); 2097 2098 return ret; 2099 } 2100 2101 static coroutine_fn int qcow2_co_discard(BlockDriverState *bs, 2102 int64_t sector_num, int nb_sectors) 2103 { 2104 int ret; 2105 BDRVQcowState *s = bs->opaque; 2106 2107 qemu_co_mutex_lock(&s->lock); 2108 ret = qcow2_discard_clusters(bs, sector_num << BDRV_SECTOR_BITS, 2109 nb_sectors, QCOW2_DISCARD_REQUEST, false); 2110 qemu_co_mutex_unlock(&s->lock); 2111 return ret; 2112 } 2113 2114 static int qcow2_truncate(BlockDriverState *bs, int64_t offset) 2115 { 2116 BDRVQcowState *s = bs->opaque; 2117 int64_t new_l1_size; 2118 int ret; 2119 2120 if (offset & 511) { 2121 error_report("The new size must be a multiple of 512"); 2122 return -EINVAL; 2123 } 2124 2125 /* cannot proceed if image has snapshots */ 2126 if (s->nb_snapshots) { 2127 error_report("Can't resize an image which has snapshots"); 2128 return -ENOTSUP; 2129 } 2130 2131 /* shrinking is currently not supported */ 2132 if (offset < bs->total_sectors * 512) { 2133 error_report("qcow2 doesn't support shrinking images yet"); 2134 return -ENOTSUP; 2135 } 2136 2137 new_l1_size = size_to_l1(s, offset); 2138 ret = qcow2_grow_l1_table(bs, new_l1_size, true); 2139 if (ret < 0) { 2140 return ret; 2141 } 2142 2143 /* write updated header.size */ 2144 offset = cpu_to_be64(offset); 2145 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, size), 2146 &offset, sizeof(uint64_t)); 2147 if (ret < 0) { 2148 return ret; 2149 } 2150 2151 s->l1_vm_state_index = new_l1_size; 2152 return 0; 2153 } 2154 2155 /* XXX: put compressed sectors first, then all the cluster aligned 2156 tables to avoid losing bytes in alignment */ 2157 static int qcow2_write_compressed(BlockDriverState *bs, int64_t sector_num, 2158 const uint8_t *buf, int nb_sectors) 2159 { 2160 BDRVQcowState *s = bs->opaque; 2161 z_stream strm; 2162 int ret, out_len; 2163 uint8_t *out_buf; 2164 uint64_t cluster_offset; 2165 2166 if (nb_sectors == 0) { 2167 /* align end of file to a sector boundary to ease reading with 2168 sector based I/Os */ 2169 cluster_offset = bdrv_getlength(bs->file); 2170 return bdrv_truncate(bs->file, cluster_offset); 2171 } 2172 2173 if (nb_sectors != s->cluster_sectors) { 2174 ret = -EINVAL; 2175 2176 /* Zero-pad last write if image size is not cluster aligned */ 2177 if (sector_num + nb_sectors == bs->total_sectors && 2178 nb_sectors < s->cluster_sectors) { 2179 uint8_t *pad_buf = qemu_blockalign(bs, s->cluster_size); 2180 memset(pad_buf, 0, s->cluster_size); 2181 memcpy(pad_buf, buf, nb_sectors * BDRV_SECTOR_SIZE); 2182 ret = qcow2_write_compressed(bs, sector_num, 2183 pad_buf, s->cluster_sectors); 2184 qemu_vfree(pad_buf); 2185 } 2186 return ret; 2187 } 2188 2189 out_buf = g_malloc(s->cluster_size + (s->cluster_size / 1000) + 128); 2190 2191 /* best compression, small window, no zlib header */ 2192 memset(&strm, 0, sizeof(strm)); 2193 ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION, 2194 Z_DEFLATED, -12, 2195 9, Z_DEFAULT_STRATEGY); 2196 if (ret != 0) { 2197 ret = -EINVAL; 2198 goto fail; 2199 } 2200 2201 strm.avail_in = s->cluster_size; 2202 strm.next_in = (uint8_t *)buf; 2203 strm.avail_out = s->cluster_size; 2204 strm.next_out = out_buf; 2205 2206 ret = deflate(&strm, Z_FINISH); 2207 if (ret != Z_STREAM_END && ret != Z_OK) { 2208 deflateEnd(&strm); 2209 ret = -EINVAL; 2210 goto fail; 2211 } 2212 out_len = strm.next_out - out_buf; 2213 2214 deflateEnd(&strm); 2215 2216 if (ret != Z_STREAM_END || out_len >= s->cluster_size) { 2217 /* could not compress: write normal cluster */ 2218 ret = bdrv_write(bs, sector_num, buf, s->cluster_sectors); 2219 if (ret < 0) { 2220 goto fail; 2221 } 2222 } else { 2223 cluster_offset = qcow2_alloc_compressed_cluster_offset(bs, 2224 sector_num << 9, out_len); 2225 if (!cluster_offset) { 2226 ret = -EIO; 2227 goto fail; 2228 } 2229 cluster_offset &= s->cluster_offset_mask; 2230 2231 ret = qcow2_pre_write_overlap_check(bs, 0, cluster_offset, out_len); 2232 if (ret < 0) { 2233 goto fail; 2234 } 2235 2236 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_COMPRESSED); 2237 ret = bdrv_pwrite(bs->file, cluster_offset, out_buf, out_len); 2238 if (ret < 0) { 2239 goto fail; 2240 } 2241 } 2242 2243 ret = 0; 2244 fail: 2245 g_free(out_buf); 2246 return ret; 2247 } 2248 2249 static int make_completely_empty(BlockDriverState *bs) 2250 { 2251 BDRVQcowState *s = bs->opaque; 2252 int ret, l1_clusters; 2253 int64_t offset; 2254 uint64_t *new_reftable = NULL; 2255 uint64_t rt_entry, l1_size2; 2256 struct { 2257 uint64_t l1_offset; 2258 uint64_t reftable_offset; 2259 uint32_t reftable_clusters; 2260 } QEMU_PACKED l1_ofs_rt_ofs_cls; 2261 2262 ret = qcow2_cache_empty(bs, s->l2_table_cache); 2263 if (ret < 0) { 2264 goto fail; 2265 } 2266 2267 ret = qcow2_cache_empty(bs, s->refcount_block_cache); 2268 if (ret < 0) { 2269 goto fail; 2270 } 2271 2272 /* Refcounts will be broken utterly */ 2273 ret = qcow2_mark_dirty(bs); 2274 if (ret < 0) { 2275 goto fail; 2276 } 2277 2278 BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE); 2279 2280 l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / sizeof(uint64_t)); 2281 l1_size2 = (uint64_t)s->l1_size * sizeof(uint64_t); 2282 2283 /* After this call, neither the in-memory nor the on-disk refcount 2284 * information accurately describe the actual references */ 2285 2286 ret = bdrv_write_zeroes(bs->file, s->l1_table_offset / BDRV_SECTOR_SIZE, 2287 l1_clusters * s->cluster_sectors, 0); 2288 if (ret < 0) { 2289 goto fail_broken_refcounts; 2290 } 2291 memset(s->l1_table, 0, l1_size2); 2292 2293 BLKDBG_EVENT(bs->file, BLKDBG_EMPTY_IMAGE_PREPARE); 2294 2295 /* Overwrite enough clusters at the beginning of the sectors to place 2296 * the refcount table, a refcount block and the L1 table in; this may 2297 * overwrite parts of the existing refcount and L1 table, which is not 2298 * an issue because the dirty flag is set, complete data loss is in fact 2299 * desired and partial data loss is consequently fine as well */ 2300 ret = bdrv_write_zeroes(bs->file, s->cluster_size / BDRV_SECTOR_SIZE, 2301 (2 + l1_clusters) * s->cluster_size / 2302 BDRV_SECTOR_SIZE, 0); 2303 /* This call (even if it failed overall) may have overwritten on-disk 2304 * refcount structures; in that case, the in-memory refcount information 2305 * will probably differ from the on-disk information which makes the BDS 2306 * unusable */ 2307 if (ret < 0) { 2308 goto fail_broken_refcounts; 2309 } 2310 2311 BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE); 2312 BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_UPDATE); 2313 2314 /* "Create" an empty reftable (one cluster) directly after the image 2315 * header and an empty L1 table three clusters after the image header; 2316 * the cluster between those two will be used as the first refblock */ 2317 cpu_to_be64w(&l1_ofs_rt_ofs_cls.l1_offset, 3 * s->cluster_size); 2318 cpu_to_be64w(&l1_ofs_rt_ofs_cls.reftable_offset, s->cluster_size); 2319 cpu_to_be32w(&l1_ofs_rt_ofs_cls.reftable_clusters, 1); 2320 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, l1_table_offset), 2321 &l1_ofs_rt_ofs_cls, sizeof(l1_ofs_rt_ofs_cls)); 2322 if (ret < 0) { 2323 goto fail_broken_refcounts; 2324 } 2325 2326 s->l1_table_offset = 3 * s->cluster_size; 2327 2328 new_reftable = g_try_new0(uint64_t, s->cluster_size / sizeof(uint64_t)); 2329 if (!new_reftable) { 2330 ret = -ENOMEM; 2331 goto fail_broken_refcounts; 2332 } 2333 2334 s->refcount_table_offset = s->cluster_size; 2335 s->refcount_table_size = s->cluster_size / sizeof(uint64_t); 2336 2337 g_free(s->refcount_table); 2338 s->refcount_table = new_reftable; 2339 new_reftable = NULL; 2340 2341 /* Now the in-memory refcount information again corresponds to the on-disk 2342 * information (reftable is empty and no refblocks (the refblock cache is 2343 * empty)); however, this means some clusters (e.g. the image header) are 2344 * referenced, but not refcounted, but the normal qcow2 code assumes that 2345 * the in-memory information is always correct */ 2346 2347 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC); 2348 2349 /* Enter the first refblock into the reftable */ 2350 rt_entry = cpu_to_be64(2 * s->cluster_size); 2351 ret = bdrv_pwrite_sync(bs->file, s->cluster_size, 2352 &rt_entry, sizeof(rt_entry)); 2353 if (ret < 0) { 2354 goto fail_broken_refcounts; 2355 } 2356 s->refcount_table[0] = 2 * s->cluster_size; 2357 2358 s->free_cluster_index = 0; 2359 assert(3 + l1_clusters <= s->refcount_block_size); 2360 offset = qcow2_alloc_clusters(bs, 3 * s->cluster_size + l1_size2); 2361 if (offset < 0) { 2362 ret = offset; 2363 goto fail_broken_refcounts; 2364 } else if (offset > 0) { 2365 error_report("First cluster in emptied image is in use"); 2366 abort(); 2367 } 2368 2369 /* Now finally the in-memory information corresponds to the on-disk 2370 * structures and is correct */ 2371 ret = qcow2_mark_clean(bs); 2372 if (ret < 0) { 2373 goto fail; 2374 } 2375 2376 ret = bdrv_truncate(bs->file, (3 + l1_clusters) * s->cluster_size); 2377 if (ret < 0) { 2378 goto fail; 2379 } 2380 2381 return 0; 2382 2383 fail_broken_refcounts: 2384 /* The BDS is unusable at this point. If we wanted to make it usable, we 2385 * would have to call qcow2_refcount_close(), qcow2_refcount_init(), 2386 * qcow2_check_refcounts(), qcow2_refcount_close() and qcow2_refcount_init() 2387 * again. However, because the functions which could have caused this error 2388 * path to be taken are used by those functions as well, it's very likely 2389 * that that sequence will fail as well. Therefore, just eject the BDS. */ 2390 bs->drv = NULL; 2391 2392 fail: 2393 g_free(new_reftable); 2394 return ret; 2395 } 2396 2397 static int qcow2_make_empty(BlockDriverState *bs) 2398 { 2399 BDRVQcowState *s = bs->opaque; 2400 uint64_t start_sector; 2401 int sector_step = INT_MAX / BDRV_SECTOR_SIZE; 2402 int l1_clusters, ret = 0; 2403 2404 l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / sizeof(uint64_t)); 2405 2406 if (s->qcow_version >= 3 && !s->snapshots && 2407 3 + l1_clusters <= s->refcount_block_size) { 2408 /* The following function only works for qcow2 v3 images (it requires 2409 * the dirty flag) and only as long as there are no snapshots (because 2410 * it completely empties the image). Furthermore, the L1 table and three 2411 * additional clusters (image header, refcount table, one refcount 2412 * block) have to fit inside one refcount block. */ 2413 return make_completely_empty(bs); 2414 } 2415 2416 /* This fallback code simply discards every active cluster; this is slow, 2417 * but works in all cases */ 2418 for (start_sector = 0; start_sector < bs->total_sectors; 2419 start_sector += sector_step) 2420 { 2421 /* As this function is generally used after committing an external 2422 * snapshot, QCOW2_DISCARD_SNAPSHOT seems appropriate. Also, the 2423 * default action for this kind of discard is to pass the discard, 2424 * which will ideally result in an actually smaller image file, as 2425 * is probably desired. */ 2426 ret = qcow2_discard_clusters(bs, start_sector * BDRV_SECTOR_SIZE, 2427 MIN(sector_step, 2428 bs->total_sectors - start_sector), 2429 QCOW2_DISCARD_SNAPSHOT, true); 2430 if (ret < 0) { 2431 break; 2432 } 2433 } 2434 2435 return ret; 2436 } 2437 2438 static coroutine_fn int qcow2_co_flush_to_os(BlockDriverState *bs) 2439 { 2440 BDRVQcowState *s = bs->opaque; 2441 int ret; 2442 2443 qemu_co_mutex_lock(&s->lock); 2444 ret = qcow2_cache_flush(bs, s->l2_table_cache); 2445 if (ret < 0) { 2446 qemu_co_mutex_unlock(&s->lock); 2447 return ret; 2448 } 2449 2450 if (qcow2_need_accurate_refcounts(s)) { 2451 ret = qcow2_cache_flush(bs, s->refcount_block_cache); 2452 if (ret < 0) { 2453 qemu_co_mutex_unlock(&s->lock); 2454 return ret; 2455 } 2456 } 2457 qemu_co_mutex_unlock(&s->lock); 2458 2459 return 0; 2460 } 2461 2462 static int qcow2_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) 2463 { 2464 BDRVQcowState *s = bs->opaque; 2465 bdi->unallocated_blocks_are_zero = true; 2466 bdi->can_write_zeroes_with_unmap = (s->qcow_version >= 3); 2467 bdi->cluster_size = s->cluster_size; 2468 bdi->vm_state_offset = qcow2_vm_state_offset(s); 2469 return 0; 2470 } 2471 2472 static ImageInfoSpecific *qcow2_get_specific_info(BlockDriverState *bs) 2473 { 2474 BDRVQcowState *s = bs->opaque; 2475 ImageInfoSpecific *spec_info = g_new(ImageInfoSpecific, 1); 2476 2477 *spec_info = (ImageInfoSpecific){ 2478 .kind = IMAGE_INFO_SPECIFIC_KIND_QCOW2, 2479 { 2480 .qcow2 = g_new(ImageInfoSpecificQCow2, 1), 2481 }, 2482 }; 2483 if (s->qcow_version == 2) { 2484 *spec_info->qcow2 = (ImageInfoSpecificQCow2){ 2485 .compat = g_strdup("0.10"), 2486 .refcount_bits = s->refcount_bits, 2487 }; 2488 } else if (s->qcow_version == 3) { 2489 *spec_info->qcow2 = (ImageInfoSpecificQCow2){ 2490 .compat = g_strdup("1.1"), 2491 .lazy_refcounts = s->compatible_features & 2492 QCOW2_COMPAT_LAZY_REFCOUNTS, 2493 .has_lazy_refcounts = true, 2494 .corrupt = s->incompatible_features & 2495 QCOW2_INCOMPAT_CORRUPT, 2496 .has_corrupt = true, 2497 .refcount_bits = s->refcount_bits, 2498 }; 2499 } 2500 2501 return spec_info; 2502 } 2503 2504 #if 0 2505 static void dump_refcounts(BlockDriverState *bs) 2506 { 2507 BDRVQcowState *s = bs->opaque; 2508 int64_t nb_clusters, k, k1, size; 2509 int refcount; 2510 2511 size = bdrv_getlength(bs->file); 2512 nb_clusters = size_to_clusters(s, size); 2513 for(k = 0; k < nb_clusters;) { 2514 k1 = k; 2515 refcount = get_refcount(bs, k); 2516 k++; 2517 while (k < nb_clusters && get_refcount(bs, k) == refcount) 2518 k++; 2519 printf("%" PRId64 ": refcount=%d nb=%" PRId64 "\n", k, refcount, 2520 k - k1); 2521 } 2522 } 2523 #endif 2524 2525 static int qcow2_save_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, 2526 int64_t pos) 2527 { 2528 BDRVQcowState *s = bs->opaque; 2529 int64_t total_sectors = bs->total_sectors; 2530 bool zero_beyond_eof = bs->zero_beyond_eof; 2531 int ret; 2532 2533 BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_SAVE); 2534 bs->zero_beyond_eof = false; 2535 ret = bdrv_pwritev(bs, qcow2_vm_state_offset(s) + pos, qiov); 2536 bs->zero_beyond_eof = zero_beyond_eof; 2537 2538 /* bdrv_co_do_writev will have increased the total_sectors value to include 2539 * the VM state - the VM state is however not an actual part of the block 2540 * device, therefore, we need to restore the old value. */ 2541 bs->total_sectors = total_sectors; 2542 2543 return ret; 2544 } 2545 2546 static int qcow2_load_vmstate(BlockDriverState *bs, uint8_t *buf, 2547 int64_t pos, int size) 2548 { 2549 BDRVQcowState *s = bs->opaque; 2550 bool zero_beyond_eof = bs->zero_beyond_eof; 2551 int ret; 2552 2553 BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_LOAD); 2554 bs->zero_beyond_eof = false; 2555 ret = bdrv_pread(bs, qcow2_vm_state_offset(s) + pos, buf, size); 2556 bs->zero_beyond_eof = zero_beyond_eof; 2557 2558 return ret; 2559 } 2560 2561 /* 2562 * Downgrades an image's version. To achieve this, any incompatible features 2563 * have to be removed. 2564 */ 2565 static int qcow2_downgrade(BlockDriverState *bs, int target_version, 2566 BlockDriverAmendStatusCB *status_cb) 2567 { 2568 BDRVQcowState *s = bs->opaque; 2569 int current_version = s->qcow_version; 2570 int ret; 2571 2572 if (target_version == current_version) { 2573 return 0; 2574 } else if (target_version > current_version) { 2575 return -EINVAL; 2576 } else if (target_version != 2) { 2577 return -EINVAL; 2578 } 2579 2580 if (s->refcount_order != 4) { 2581 /* we would have to convert the image to a refcount_order == 4 image 2582 * here; however, since qemu (at the time of writing this) does not 2583 * support anything different than 4 anyway, there is no point in doing 2584 * so right now; however, we should error out (if qemu supports this in 2585 * the future and this code has not been adapted) */ 2586 error_report("qcow2_downgrade: Image refcount orders other than 4 are " 2587 "currently not supported."); 2588 return -ENOTSUP; 2589 } 2590 2591 /* clear incompatible features */ 2592 if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) { 2593 ret = qcow2_mark_clean(bs); 2594 if (ret < 0) { 2595 return ret; 2596 } 2597 } 2598 2599 /* with QCOW2_INCOMPAT_CORRUPT, it is pretty much impossible to get here in 2600 * the first place; if that happens nonetheless, returning -ENOTSUP is the 2601 * best thing to do anyway */ 2602 2603 if (s->incompatible_features) { 2604 return -ENOTSUP; 2605 } 2606 2607 /* since we can ignore compatible features, we can set them to 0 as well */ 2608 s->compatible_features = 0; 2609 /* if lazy refcounts have been used, they have already been fixed through 2610 * clearing the dirty flag */ 2611 2612 /* clearing autoclear features is trivial */ 2613 s->autoclear_features = 0; 2614 2615 ret = qcow2_expand_zero_clusters(bs, status_cb); 2616 if (ret < 0) { 2617 return ret; 2618 } 2619 2620 s->qcow_version = target_version; 2621 ret = qcow2_update_header(bs); 2622 if (ret < 0) { 2623 s->qcow_version = current_version; 2624 return ret; 2625 } 2626 return 0; 2627 } 2628 2629 static int qcow2_amend_options(BlockDriverState *bs, QemuOpts *opts, 2630 BlockDriverAmendStatusCB *status_cb) 2631 { 2632 BDRVQcowState *s = bs->opaque; 2633 int old_version = s->qcow_version, new_version = old_version; 2634 uint64_t new_size = 0; 2635 const char *backing_file = NULL, *backing_format = NULL; 2636 bool lazy_refcounts = s->use_lazy_refcounts; 2637 const char *compat = NULL; 2638 uint64_t cluster_size = s->cluster_size; 2639 bool encrypt; 2640 int ret; 2641 QemuOptDesc *desc = opts->list->desc; 2642 2643 while (desc && desc->name) { 2644 if (!qemu_opt_find(opts, desc->name)) { 2645 /* only change explicitly defined options */ 2646 desc++; 2647 continue; 2648 } 2649 2650 if (!strcmp(desc->name, "compat")) { 2651 compat = qemu_opt_get(opts, "compat"); 2652 if (!compat) { 2653 /* preserve default */ 2654 } else if (!strcmp(compat, "0.10")) { 2655 new_version = 2; 2656 } else if (!strcmp(compat, "1.1")) { 2657 new_version = 3; 2658 } else { 2659 fprintf(stderr, "Unknown compatibility level %s.\n", compat); 2660 return -EINVAL; 2661 } 2662 } else if (!strcmp(desc->name, "preallocation")) { 2663 fprintf(stderr, "Cannot change preallocation mode.\n"); 2664 return -ENOTSUP; 2665 } else if (!strcmp(desc->name, "size")) { 2666 new_size = qemu_opt_get_size(opts, "size", 0); 2667 } else if (!strcmp(desc->name, "backing_file")) { 2668 backing_file = qemu_opt_get(opts, "backing_file"); 2669 } else if (!strcmp(desc->name, "backing_fmt")) { 2670 backing_format = qemu_opt_get(opts, "backing_fmt"); 2671 } else if (!strcmp(desc->name, "encryption")) { 2672 encrypt = qemu_opt_get_bool(opts, "encryption", s->crypt_method); 2673 if (encrypt != !!s->crypt_method) { 2674 fprintf(stderr, "Changing the encryption flag is not " 2675 "supported.\n"); 2676 return -ENOTSUP; 2677 } 2678 } else if (!strcmp(desc->name, "cluster_size")) { 2679 cluster_size = qemu_opt_get_size(opts, "cluster_size", 2680 cluster_size); 2681 if (cluster_size != s->cluster_size) { 2682 fprintf(stderr, "Changing the cluster size is not " 2683 "supported.\n"); 2684 return -ENOTSUP; 2685 } 2686 } else if (!strcmp(desc->name, "lazy_refcounts")) { 2687 lazy_refcounts = qemu_opt_get_bool(opts, "lazy_refcounts", 2688 lazy_refcounts); 2689 } else { 2690 /* if this assertion fails, this probably means a new option was 2691 * added without having it covered here */ 2692 assert(false); 2693 } 2694 2695 desc++; 2696 } 2697 2698 if (new_version != old_version) { 2699 if (new_version > old_version) { 2700 /* Upgrade */ 2701 s->qcow_version = new_version; 2702 ret = qcow2_update_header(bs); 2703 if (ret < 0) { 2704 s->qcow_version = old_version; 2705 return ret; 2706 } 2707 } else { 2708 ret = qcow2_downgrade(bs, new_version, status_cb); 2709 if (ret < 0) { 2710 return ret; 2711 } 2712 } 2713 } 2714 2715 if (backing_file || backing_format) { 2716 ret = qcow2_change_backing_file(bs, backing_file ?: bs->backing_file, 2717 backing_format ?: bs->backing_format); 2718 if (ret < 0) { 2719 return ret; 2720 } 2721 } 2722 2723 if (s->use_lazy_refcounts != lazy_refcounts) { 2724 if (lazy_refcounts) { 2725 if (s->qcow_version < 3) { 2726 fprintf(stderr, "Lazy refcounts only supported with compatibility " 2727 "level 1.1 and above (use compat=1.1 or greater)\n"); 2728 return -EINVAL; 2729 } 2730 s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS; 2731 ret = qcow2_update_header(bs); 2732 if (ret < 0) { 2733 s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS; 2734 return ret; 2735 } 2736 s->use_lazy_refcounts = true; 2737 } else { 2738 /* make image clean first */ 2739 ret = qcow2_mark_clean(bs); 2740 if (ret < 0) { 2741 return ret; 2742 } 2743 /* now disallow lazy refcounts */ 2744 s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS; 2745 ret = qcow2_update_header(bs); 2746 if (ret < 0) { 2747 s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS; 2748 return ret; 2749 } 2750 s->use_lazy_refcounts = false; 2751 } 2752 } 2753 2754 if (new_size) { 2755 ret = bdrv_truncate(bs, new_size); 2756 if (ret < 0) { 2757 return ret; 2758 } 2759 } 2760 2761 return 0; 2762 } 2763 2764 /* 2765 * If offset or size are negative, respectively, they will not be included in 2766 * the BLOCK_IMAGE_CORRUPTED event emitted. 2767 * fatal will be ignored for read-only BDS; corruptions found there will always 2768 * be considered non-fatal. 2769 */ 2770 void qcow2_signal_corruption(BlockDriverState *bs, bool fatal, int64_t offset, 2771 int64_t size, const char *message_format, ...) 2772 { 2773 BDRVQcowState *s = bs->opaque; 2774 char *message; 2775 va_list ap; 2776 2777 fatal = fatal && !bs->read_only; 2778 2779 if (s->signaled_corruption && 2780 (!fatal || (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT))) 2781 { 2782 return; 2783 } 2784 2785 va_start(ap, message_format); 2786 message = g_strdup_vprintf(message_format, ap); 2787 va_end(ap); 2788 2789 if (fatal) { 2790 fprintf(stderr, "qcow2: Marking image as corrupt: %s; further " 2791 "corruption events will be suppressed\n", message); 2792 } else { 2793 fprintf(stderr, "qcow2: Image is corrupt: %s; further non-fatal " 2794 "corruption events will be suppressed\n", message); 2795 } 2796 2797 qapi_event_send_block_image_corrupted(bdrv_get_device_name(bs), message, 2798 offset >= 0, offset, size >= 0, size, 2799 fatal, &error_abort); 2800 g_free(message); 2801 2802 if (fatal) { 2803 qcow2_mark_corrupt(bs); 2804 bs->drv = NULL; /* make BDS unusable */ 2805 } 2806 2807 s->signaled_corruption = true; 2808 } 2809 2810 static QemuOptsList qcow2_create_opts = { 2811 .name = "qcow2-create-opts", 2812 .head = QTAILQ_HEAD_INITIALIZER(qcow2_create_opts.head), 2813 .desc = { 2814 { 2815 .name = BLOCK_OPT_SIZE, 2816 .type = QEMU_OPT_SIZE, 2817 .help = "Virtual disk size" 2818 }, 2819 { 2820 .name = BLOCK_OPT_COMPAT_LEVEL, 2821 .type = QEMU_OPT_STRING, 2822 .help = "Compatibility level (0.10 or 1.1)" 2823 }, 2824 { 2825 .name = BLOCK_OPT_BACKING_FILE, 2826 .type = QEMU_OPT_STRING, 2827 .help = "File name of a base image" 2828 }, 2829 { 2830 .name = BLOCK_OPT_BACKING_FMT, 2831 .type = QEMU_OPT_STRING, 2832 .help = "Image format of the base image" 2833 }, 2834 { 2835 .name = BLOCK_OPT_ENCRYPT, 2836 .type = QEMU_OPT_BOOL, 2837 .help = "Encrypt the image", 2838 .def_value_str = "off" 2839 }, 2840 { 2841 .name = BLOCK_OPT_CLUSTER_SIZE, 2842 .type = QEMU_OPT_SIZE, 2843 .help = "qcow2 cluster size", 2844 .def_value_str = stringify(DEFAULT_CLUSTER_SIZE) 2845 }, 2846 { 2847 .name = BLOCK_OPT_PREALLOC, 2848 .type = QEMU_OPT_STRING, 2849 .help = "Preallocation mode (allowed values: off, metadata, " 2850 "falloc, full)" 2851 }, 2852 { 2853 .name = BLOCK_OPT_LAZY_REFCOUNTS, 2854 .type = QEMU_OPT_BOOL, 2855 .help = "Postpone refcount updates", 2856 .def_value_str = "off" 2857 }, 2858 { /* end of list */ } 2859 } 2860 }; 2861 2862 BlockDriver bdrv_qcow2 = { 2863 .format_name = "qcow2", 2864 .instance_size = sizeof(BDRVQcowState), 2865 .bdrv_probe = qcow2_probe, 2866 .bdrv_open = qcow2_open, 2867 .bdrv_close = qcow2_close, 2868 .bdrv_reopen_prepare = qcow2_reopen_prepare, 2869 .bdrv_create = qcow2_create, 2870 .bdrv_has_zero_init = bdrv_has_zero_init_1, 2871 .bdrv_co_get_block_status = qcow2_co_get_block_status, 2872 .bdrv_set_key = qcow2_set_key, 2873 2874 .bdrv_co_readv = qcow2_co_readv, 2875 .bdrv_co_writev = qcow2_co_writev, 2876 .bdrv_co_flush_to_os = qcow2_co_flush_to_os, 2877 2878 .bdrv_co_write_zeroes = qcow2_co_write_zeroes, 2879 .bdrv_co_discard = qcow2_co_discard, 2880 .bdrv_truncate = qcow2_truncate, 2881 .bdrv_write_compressed = qcow2_write_compressed, 2882 .bdrv_make_empty = qcow2_make_empty, 2883 2884 .bdrv_snapshot_create = qcow2_snapshot_create, 2885 .bdrv_snapshot_goto = qcow2_snapshot_goto, 2886 .bdrv_snapshot_delete = qcow2_snapshot_delete, 2887 .bdrv_snapshot_list = qcow2_snapshot_list, 2888 .bdrv_snapshot_load_tmp = qcow2_snapshot_load_tmp, 2889 .bdrv_get_info = qcow2_get_info, 2890 .bdrv_get_specific_info = qcow2_get_specific_info, 2891 2892 .bdrv_save_vmstate = qcow2_save_vmstate, 2893 .bdrv_load_vmstate = qcow2_load_vmstate, 2894 2895 .supports_backing = true, 2896 .bdrv_change_backing_file = qcow2_change_backing_file, 2897 2898 .bdrv_refresh_limits = qcow2_refresh_limits, 2899 .bdrv_invalidate_cache = qcow2_invalidate_cache, 2900 2901 .create_opts = &qcow2_create_opts, 2902 .bdrv_check = qcow2_check, 2903 .bdrv_amend_options = qcow2_amend_options, 2904 }; 2905 2906 static void bdrv_qcow2_init(void) 2907 { 2908 bdrv_register(&bdrv_qcow2); 2909 } 2910 2911 block_init(bdrv_qcow2_init); 2912