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 27 #include "block/qdict.h" 28 #include "sysemu/block-backend.h" 29 #include "qemu/module.h" 30 #include "qcow2.h" 31 #include "qemu/error-report.h" 32 #include "qapi/error.h" 33 #include "qapi/qapi-events-block-core.h" 34 #include "qapi/qmp/qdict.h" 35 #include "qapi/qmp/qstring.h" 36 #include "trace.h" 37 #include "qemu/option_int.h" 38 #include "qemu/cutils.h" 39 #include "qemu/bswap.h" 40 #include "qapi/qobject-input-visitor.h" 41 #include "qapi/qapi-visit-block-core.h" 42 #include "crypto.h" 43 44 /* 45 Differences with QCOW: 46 47 - Support for multiple incremental snapshots. 48 - Memory management by reference counts. 49 - Clusters which have a reference count of one have the bit 50 QCOW_OFLAG_COPIED to optimize write performance. 51 - Size of compressed clusters is stored in sectors to reduce bit usage 52 in the cluster offsets. 53 - Support for storing additional data (such as the VM state) in the 54 snapshots. 55 - If a backing store is used, the cluster size is not constrained 56 (could be backported to QCOW). 57 - L2 tables have always a size of one cluster. 58 */ 59 60 61 typedef struct { 62 uint32_t magic; 63 uint32_t len; 64 } QEMU_PACKED QCowExtension; 65 66 #define QCOW2_EXT_MAGIC_END 0 67 #define QCOW2_EXT_MAGIC_BACKING_FORMAT 0xE2792ACA 68 #define QCOW2_EXT_MAGIC_FEATURE_TABLE 0x6803f857 69 #define QCOW2_EXT_MAGIC_CRYPTO_HEADER 0x0537be77 70 #define QCOW2_EXT_MAGIC_BITMAPS 0x23852875 71 #define QCOW2_EXT_MAGIC_DATA_FILE 0x44415441 72 73 static int coroutine_fn 74 qcow2_co_preadv_compressed(BlockDriverState *bs, 75 uint64_t file_cluster_offset, 76 uint64_t offset, 77 uint64_t bytes, 78 QEMUIOVector *qiov); 79 80 static int qcow2_probe(const uint8_t *buf, int buf_size, const char *filename) 81 { 82 const QCowHeader *cow_header = (const void *)buf; 83 84 if (buf_size >= sizeof(QCowHeader) && 85 be32_to_cpu(cow_header->magic) == QCOW_MAGIC && 86 be32_to_cpu(cow_header->version) >= 2) 87 return 100; 88 else 89 return 0; 90 } 91 92 93 static ssize_t qcow2_crypto_hdr_read_func(QCryptoBlock *block, size_t offset, 94 uint8_t *buf, size_t buflen, 95 void *opaque, Error **errp) 96 { 97 BlockDriverState *bs = opaque; 98 BDRVQcow2State *s = bs->opaque; 99 ssize_t ret; 100 101 if ((offset + buflen) > s->crypto_header.length) { 102 error_setg(errp, "Request for data outside of extension header"); 103 return -1; 104 } 105 106 ret = bdrv_pread(bs->file, 107 s->crypto_header.offset + offset, buf, buflen); 108 if (ret < 0) { 109 error_setg_errno(errp, -ret, "Could not read encryption header"); 110 return -1; 111 } 112 return ret; 113 } 114 115 116 static ssize_t qcow2_crypto_hdr_init_func(QCryptoBlock *block, size_t headerlen, 117 void *opaque, Error **errp) 118 { 119 BlockDriverState *bs = opaque; 120 BDRVQcow2State *s = bs->opaque; 121 int64_t ret; 122 int64_t clusterlen; 123 124 ret = qcow2_alloc_clusters(bs, headerlen); 125 if (ret < 0) { 126 error_setg_errno(errp, -ret, 127 "Cannot allocate cluster for LUKS header size %zu", 128 headerlen); 129 return -1; 130 } 131 132 s->crypto_header.length = headerlen; 133 s->crypto_header.offset = ret; 134 135 /* Zero fill remaining space in cluster so it has predictable 136 * content in case of future spec changes */ 137 clusterlen = size_to_clusters(s, headerlen) * s->cluster_size; 138 assert(qcow2_pre_write_overlap_check(bs, 0, ret, clusterlen, false) == 0); 139 ret = bdrv_pwrite_zeroes(bs->file, 140 ret + headerlen, 141 clusterlen - headerlen, 0); 142 if (ret < 0) { 143 error_setg_errno(errp, -ret, "Could not zero fill encryption header"); 144 return -1; 145 } 146 147 return ret; 148 } 149 150 151 static ssize_t qcow2_crypto_hdr_write_func(QCryptoBlock *block, size_t offset, 152 const uint8_t *buf, size_t buflen, 153 void *opaque, Error **errp) 154 { 155 BlockDriverState *bs = opaque; 156 BDRVQcow2State *s = bs->opaque; 157 ssize_t ret; 158 159 if ((offset + buflen) > s->crypto_header.length) { 160 error_setg(errp, "Request for data outside of extension header"); 161 return -1; 162 } 163 164 ret = bdrv_pwrite(bs->file, 165 s->crypto_header.offset + offset, buf, buflen); 166 if (ret < 0) { 167 error_setg_errno(errp, -ret, "Could not read encryption header"); 168 return -1; 169 } 170 return ret; 171 } 172 173 174 /* 175 * read qcow2 extension and fill bs 176 * start reading from start_offset 177 * finish reading upon magic of value 0 or when end_offset reached 178 * unknown magic is skipped (future extension this version knows nothing about) 179 * return 0 upon success, non-0 otherwise 180 */ 181 static int qcow2_read_extensions(BlockDriverState *bs, uint64_t start_offset, 182 uint64_t end_offset, void **p_feature_table, 183 int flags, bool *need_update_header, 184 Error **errp) 185 { 186 BDRVQcow2State *s = bs->opaque; 187 QCowExtension ext; 188 uint64_t offset; 189 int ret; 190 Qcow2BitmapHeaderExt bitmaps_ext; 191 192 if (need_update_header != NULL) { 193 *need_update_header = false; 194 } 195 196 #ifdef DEBUG_EXT 197 printf("qcow2_read_extensions: start=%ld end=%ld\n", start_offset, end_offset); 198 #endif 199 offset = start_offset; 200 while (offset < end_offset) { 201 202 #ifdef DEBUG_EXT 203 /* Sanity check */ 204 if (offset > s->cluster_size) 205 printf("qcow2_read_extension: suspicious offset %lu\n", offset); 206 207 printf("attempting to read extended header in offset %lu\n", offset); 208 #endif 209 210 ret = bdrv_pread(bs->file, offset, &ext, sizeof(ext)); 211 if (ret < 0) { 212 error_setg_errno(errp, -ret, "qcow2_read_extension: ERROR: " 213 "pread fail from offset %" PRIu64, offset); 214 return 1; 215 } 216 ext.magic = be32_to_cpu(ext.magic); 217 ext.len = be32_to_cpu(ext.len); 218 offset += sizeof(ext); 219 #ifdef DEBUG_EXT 220 printf("ext.magic = 0x%x\n", ext.magic); 221 #endif 222 if (offset > end_offset || ext.len > end_offset - offset) { 223 error_setg(errp, "Header extension too large"); 224 return -EINVAL; 225 } 226 227 switch (ext.magic) { 228 case QCOW2_EXT_MAGIC_END: 229 return 0; 230 231 case QCOW2_EXT_MAGIC_BACKING_FORMAT: 232 if (ext.len >= sizeof(bs->backing_format)) { 233 error_setg(errp, "ERROR: ext_backing_format: len=%" PRIu32 234 " too large (>=%zu)", ext.len, 235 sizeof(bs->backing_format)); 236 return 2; 237 } 238 ret = bdrv_pread(bs->file, offset, bs->backing_format, ext.len); 239 if (ret < 0) { 240 error_setg_errno(errp, -ret, "ERROR: ext_backing_format: " 241 "Could not read format name"); 242 return 3; 243 } 244 bs->backing_format[ext.len] = '\0'; 245 s->image_backing_format = g_strdup(bs->backing_format); 246 #ifdef DEBUG_EXT 247 printf("Qcow2: Got format extension %s\n", bs->backing_format); 248 #endif 249 break; 250 251 case QCOW2_EXT_MAGIC_FEATURE_TABLE: 252 if (p_feature_table != NULL) { 253 void* feature_table = g_malloc0(ext.len + 2 * sizeof(Qcow2Feature)); 254 ret = bdrv_pread(bs->file, offset , feature_table, ext.len); 255 if (ret < 0) { 256 error_setg_errno(errp, -ret, "ERROR: ext_feature_table: " 257 "Could not read table"); 258 return ret; 259 } 260 261 *p_feature_table = feature_table; 262 } 263 break; 264 265 case QCOW2_EXT_MAGIC_CRYPTO_HEADER: { 266 unsigned int cflags = 0; 267 if (s->crypt_method_header != QCOW_CRYPT_LUKS) { 268 error_setg(errp, "CRYPTO header extension only " 269 "expected with LUKS encryption method"); 270 return -EINVAL; 271 } 272 if (ext.len != sizeof(Qcow2CryptoHeaderExtension)) { 273 error_setg(errp, "CRYPTO header extension size %u, " 274 "but expected size %zu", ext.len, 275 sizeof(Qcow2CryptoHeaderExtension)); 276 return -EINVAL; 277 } 278 279 ret = bdrv_pread(bs->file, offset, &s->crypto_header, ext.len); 280 if (ret < 0) { 281 error_setg_errno(errp, -ret, 282 "Unable to read CRYPTO header extension"); 283 return ret; 284 } 285 s->crypto_header.offset = be64_to_cpu(s->crypto_header.offset); 286 s->crypto_header.length = be64_to_cpu(s->crypto_header.length); 287 288 if ((s->crypto_header.offset % s->cluster_size) != 0) { 289 error_setg(errp, "Encryption header offset '%" PRIu64 "' is " 290 "not a multiple of cluster size '%u'", 291 s->crypto_header.offset, s->cluster_size); 292 return -EINVAL; 293 } 294 295 if (flags & BDRV_O_NO_IO) { 296 cflags |= QCRYPTO_BLOCK_OPEN_NO_IO; 297 } 298 s->crypto = qcrypto_block_open(s->crypto_opts, "encrypt.", 299 qcow2_crypto_hdr_read_func, 300 bs, cflags, QCOW2_MAX_THREADS, errp); 301 if (!s->crypto) { 302 return -EINVAL; 303 } 304 } break; 305 306 case QCOW2_EXT_MAGIC_BITMAPS: 307 if (ext.len != sizeof(bitmaps_ext)) { 308 error_setg_errno(errp, -ret, "bitmaps_ext: " 309 "Invalid extension length"); 310 return -EINVAL; 311 } 312 313 if (!(s->autoclear_features & QCOW2_AUTOCLEAR_BITMAPS)) { 314 if (s->qcow_version < 3) { 315 /* Let's be a bit more specific */ 316 warn_report("This qcow2 v2 image contains bitmaps, but " 317 "they may have been modified by a program " 318 "without persistent bitmap support; so now " 319 "they must all be considered inconsistent"); 320 } else { 321 warn_report("a program lacking bitmap support " 322 "modified this file, so all bitmaps are now " 323 "considered inconsistent"); 324 } 325 error_printf("Some clusters may be leaked, " 326 "run 'qemu-img check -r' on the image " 327 "file to fix."); 328 if (need_update_header != NULL) { 329 /* Updating is needed to drop invalid bitmap extension. */ 330 *need_update_header = true; 331 } 332 break; 333 } 334 335 ret = bdrv_pread(bs->file, offset, &bitmaps_ext, ext.len); 336 if (ret < 0) { 337 error_setg_errno(errp, -ret, "bitmaps_ext: " 338 "Could not read ext header"); 339 return ret; 340 } 341 342 if (bitmaps_ext.reserved32 != 0) { 343 error_setg_errno(errp, -ret, "bitmaps_ext: " 344 "Reserved field is not zero"); 345 return -EINVAL; 346 } 347 348 bitmaps_ext.nb_bitmaps = be32_to_cpu(bitmaps_ext.nb_bitmaps); 349 bitmaps_ext.bitmap_directory_size = 350 be64_to_cpu(bitmaps_ext.bitmap_directory_size); 351 bitmaps_ext.bitmap_directory_offset = 352 be64_to_cpu(bitmaps_ext.bitmap_directory_offset); 353 354 if (bitmaps_ext.nb_bitmaps > QCOW2_MAX_BITMAPS) { 355 error_setg(errp, 356 "bitmaps_ext: Image has %" PRIu32 " bitmaps, " 357 "exceeding the QEMU supported maximum of %d", 358 bitmaps_ext.nb_bitmaps, QCOW2_MAX_BITMAPS); 359 return -EINVAL; 360 } 361 362 if (bitmaps_ext.nb_bitmaps == 0) { 363 error_setg(errp, "found bitmaps extension with zero bitmaps"); 364 return -EINVAL; 365 } 366 367 if (bitmaps_ext.bitmap_directory_offset & (s->cluster_size - 1)) { 368 error_setg(errp, "bitmaps_ext: " 369 "invalid bitmap directory offset"); 370 return -EINVAL; 371 } 372 373 if (bitmaps_ext.bitmap_directory_size > 374 QCOW2_MAX_BITMAP_DIRECTORY_SIZE) { 375 error_setg(errp, "bitmaps_ext: " 376 "bitmap directory size (%" PRIu64 ") exceeds " 377 "the maximum supported size (%d)", 378 bitmaps_ext.bitmap_directory_size, 379 QCOW2_MAX_BITMAP_DIRECTORY_SIZE); 380 return -EINVAL; 381 } 382 383 s->nb_bitmaps = bitmaps_ext.nb_bitmaps; 384 s->bitmap_directory_offset = 385 bitmaps_ext.bitmap_directory_offset; 386 s->bitmap_directory_size = 387 bitmaps_ext.bitmap_directory_size; 388 389 #ifdef DEBUG_EXT 390 printf("Qcow2: Got bitmaps extension: " 391 "offset=%" PRIu64 " nb_bitmaps=%" PRIu32 "\n", 392 s->bitmap_directory_offset, s->nb_bitmaps); 393 #endif 394 break; 395 396 case QCOW2_EXT_MAGIC_DATA_FILE: 397 { 398 s->image_data_file = g_malloc0(ext.len + 1); 399 ret = bdrv_pread(bs->file, offset, s->image_data_file, ext.len); 400 if (ret < 0) { 401 error_setg_errno(errp, -ret, 402 "ERROR: Could not read data file name"); 403 return ret; 404 } 405 #ifdef DEBUG_EXT 406 printf("Qcow2: Got external data file %s\n", s->image_data_file); 407 #endif 408 break; 409 } 410 411 default: 412 /* unknown magic - save it in case we need to rewrite the header */ 413 /* If you add a new feature, make sure to also update the fast 414 * path of qcow2_make_empty() to deal with it. */ 415 { 416 Qcow2UnknownHeaderExtension *uext; 417 418 uext = g_malloc0(sizeof(*uext) + ext.len); 419 uext->magic = ext.magic; 420 uext->len = ext.len; 421 QLIST_INSERT_HEAD(&s->unknown_header_ext, uext, next); 422 423 ret = bdrv_pread(bs->file, offset , uext->data, uext->len); 424 if (ret < 0) { 425 error_setg_errno(errp, -ret, "ERROR: unknown extension: " 426 "Could not read data"); 427 return ret; 428 } 429 } 430 break; 431 } 432 433 offset += ((ext.len + 7) & ~7); 434 } 435 436 return 0; 437 } 438 439 static void cleanup_unknown_header_ext(BlockDriverState *bs) 440 { 441 BDRVQcow2State *s = bs->opaque; 442 Qcow2UnknownHeaderExtension *uext, *next; 443 444 QLIST_FOREACH_SAFE(uext, &s->unknown_header_ext, next, next) { 445 QLIST_REMOVE(uext, next); 446 g_free(uext); 447 } 448 } 449 450 static void report_unsupported_feature(Error **errp, Qcow2Feature *table, 451 uint64_t mask) 452 { 453 char *features = g_strdup(""); 454 char *old; 455 456 while (table && table->name[0] != '\0') { 457 if (table->type == QCOW2_FEAT_TYPE_INCOMPATIBLE) { 458 if (mask & (1ULL << table->bit)) { 459 old = features; 460 features = g_strdup_printf("%s%s%.46s", old, *old ? ", " : "", 461 table->name); 462 g_free(old); 463 mask &= ~(1ULL << table->bit); 464 } 465 } 466 table++; 467 } 468 469 if (mask) { 470 old = features; 471 features = g_strdup_printf("%s%sUnknown incompatible feature: %" PRIx64, 472 old, *old ? ", " : "", mask); 473 g_free(old); 474 } 475 476 error_setg(errp, "Unsupported qcow2 feature(s): %s", features); 477 g_free(features); 478 } 479 480 /* 481 * Sets the dirty bit and flushes afterwards if necessary. 482 * 483 * The incompatible_features bit is only set if the image file header was 484 * updated successfully. Therefore it is not required to check the return 485 * value of this function. 486 */ 487 int qcow2_mark_dirty(BlockDriverState *bs) 488 { 489 BDRVQcow2State *s = bs->opaque; 490 uint64_t val; 491 int ret; 492 493 assert(s->qcow_version >= 3); 494 495 if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) { 496 return 0; /* already dirty */ 497 } 498 499 val = cpu_to_be64(s->incompatible_features | QCOW2_INCOMPAT_DIRTY); 500 ret = bdrv_pwrite(bs->file, offsetof(QCowHeader, incompatible_features), 501 &val, sizeof(val)); 502 if (ret < 0) { 503 return ret; 504 } 505 ret = bdrv_flush(bs->file->bs); 506 if (ret < 0) { 507 return ret; 508 } 509 510 /* Only treat image as dirty if the header was updated successfully */ 511 s->incompatible_features |= QCOW2_INCOMPAT_DIRTY; 512 return 0; 513 } 514 515 /* 516 * Clears the dirty bit and flushes before if necessary. Only call this 517 * function when there are no pending requests, it does not guard against 518 * concurrent requests dirtying the image. 519 */ 520 static int qcow2_mark_clean(BlockDriverState *bs) 521 { 522 BDRVQcow2State *s = bs->opaque; 523 524 if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) { 525 int ret; 526 527 s->incompatible_features &= ~QCOW2_INCOMPAT_DIRTY; 528 529 ret = qcow2_flush_caches(bs); 530 if (ret < 0) { 531 return ret; 532 } 533 534 return qcow2_update_header(bs); 535 } 536 return 0; 537 } 538 539 /* 540 * Marks the image as corrupt. 541 */ 542 int qcow2_mark_corrupt(BlockDriverState *bs) 543 { 544 BDRVQcow2State *s = bs->opaque; 545 546 s->incompatible_features |= QCOW2_INCOMPAT_CORRUPT; 547 return qcow2_update_header(bs); 548 } 549 550 /* 551 * Marks the image as consistent, i.e., unsets the corrupt bit, and flushes 552 * before if necessary. 553 */ 554 int qcow2_mark_consistent(BlockDriverState *bs) 555 { 556 BDRVQcow2State *s = bs->opaque; 557 558 if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) { 559 int ret = qcow2_flush_caches(bs); 560 if (ret < 0) { 561 return ret; 562 } 563 564 s->incompatible_features &= ~QCOW2_INCOMPAT_CORRUPT; 565 return qcow2_update_header(bs); 566 } 567 return 0; 568 } 569 570 static int coroutine_fn qcow2_co_check_locked(BlockDriverState *bs, 571 BdrvCheckResult *result, 572 BdrvCheckMode fix) 573 { 574 int ret = qcow2_check_refcounts(bs, result, fix); 575 if (ret < 0) { 576 return ret; 577 } 578 579 if (fix && result->check_errors == 0 && result->corruptions == 0) { 580 ret = qcow2_mark_clean(bs); 581 if (ret < 0) { 582 return ret; 583 } 584 return qcow2_mark_consistent(bs); 585 } 586 return ret; 587 } 588 589 static int coroutine_fn qcow2_co_check(BlockDriverState *bs, 590 BdrvCheckResult *result, 591 BdrvCheckMode fix) 592 { 593 BDRVQcow2State *s = bs->opaque; 594 int ret; 595 596 qemu_co_mutex_lock(&s->lock); 597 ret = qcow2_co_check_locked(bs, result, fix); 598 qemu_co_mutex_unlock(&s->lock); 599 return ret; 600 } 601 602 int qcow2_validate_table(BlockDriverState *bs, uint64_t offset, 603 uint64_t entries, size_t entry_len, 604 int64_t max_size_bytes, const char *table_name, 605 Error **errp) 606 { 607 BDRVQcow2State *s = bs->opaque; 608 609 if (entries > max_size_bytes / entry_len) { 610 error_setg(errp, "%s too large", table_name); 611 return -EFBIG; 612 } 613 614 /* Use signed INT64_MAX as the maximum even for uint64_t header fields, 615 * because values will be passed to qemu functions taking int64_t. */ 616 if ((INT64_MAX - entries * entry_len < offset) || 617 (offset_into_cluster(s, offset) != 0)) { 618 error_setg(errp, "%s offset invalid", table_name); 619 return -EINVAL; 620 } 621 622 return 0; 623 } 624 625 static const char *const mutable_opts[] = { 626 QCOW2_OPT_LAZY_REFCOUNTS, 627 QCOW2_OPT_DISCARD_REQUEST, 628 QCOW2_OPT_DISCARD_SNAPSHOT, 629 QCOW2_OPT_DISCARD_OTHER, 630 QCOW2_OPT_OVERLAP, 631 QCOW2_OPT_OVERLAP_TEMPLATE, 632 QCOW2_OPT_OVERLAP_MAIN_HEADER, 633 QCOW2_OPT_OVERLAP_ACTIVE_L1, 634 QCOW2_OPT_OVERLAP_ACTIVE_L2, 635 QCOW2_OPT_OVERLAP_REFCOUNT_TABLE, 636 QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK, 637 QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE, 638 QCOW2_OPT_OVERLAP_INACTIVE_L1, 639 QCOW2_OPT_OVERLAP_INACTIVE_L2, 640 QCOW2_OPT_OVERLAP_BITMAP_DIRECTORY, 641 QCOW2_OPT_CACHE_SIZE, 642 QCOW2_OPT_L2_CACHE_SIZE, 643 QCOW2_OPT_L2_CACHE_ENTRY_SIZE, 644 QCOW2_OPT_REFCOUNT_CACHE_SIZE, 645 QCOW2_OPT_CACHE_CLEAN_INTERVAL, 646 NULL 647 }; 648 649 static QemuOptsList qcow2_runtime_opts = { 650 .name = "qcow2", 651 .head = QTAILQ_HEAD_INITIALIZER(qcow2_runtime_opts.head), 652 .desc = { 653 { 654 .name = QCOW2_OPT_LAZY_REFCOUNTS, 655 .type = QEMU_OPT_BOOL, 656 .help = "Postpone refcount updates", 657 }, 658 { 659 .name = QCOW2_OPT_DISCARD_REQUEST, 660 .type = QEMU_OPT_BOOL, 661 .help = "Pass guest discard requests to the layer below", 662 }, 663 { 664 .name = QCOW2_OPT_DISCARD_SNAPSHOT, 665 .type = QEMU_OPT_BOOL, 666 .help = "Generate discard requests when snapshot related space " 667 "is freed", 668 }, 669 { 670 .name = QCOW2_OPT_DISCARD_OTHER, 671 .type = QEMU_OPT_BOOL, 672 .help = "Generate discard requests when other clusters are freed", 673 }, 674 { 675 .name = QCOW2_OPT_OVERLAP, 676 .type = QEMU_OPT_STRING, 677 .help = "Selects which overlap checks to perform from a range of " 678 "templates (none, constant, cached, all)", 679 }, 680 { 681 .name = QCOW2_OPT_OVERLAP_TEMPLATE, 682 .type = QEMU_OPT_STRING, 683 .help = "Selects which overlap checks to perform from a range of " 684 "templates (none, constant, cached, all)", 685 }, 686 { 687 .name = QCOW2_OPT_OVERLAP_MAIN_HEADER, 688 .type = QEMU_OPT_BOOL, 689 .help = "Check for unintended writes into the main qcow2 header", 690 }, 691 { 692 .name = QCOW2_OPT_OVERLAP_ACTIVE_L1, 693 .type = QEMU_OPT_BOOL, 694 .help = "Check for unintended writes into the active L1 table", 695 }, 696 { 697 .name = QCOW2_OPT_OVERLAP_ACTIVE_L2, 698 .type = QEMU_OPT_BOOL, 699 .help = "Check for unintended writes into an active L2 table", 700 }, 701 { 702 .name = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE, 703 .type = QEMU_OPT_BOOL, 704 .help = "Check for unintended writes into the refcount table", 705 }, 706 { 707 .name = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK, 708 .type = QEMU_OPT_BOOL, 709 .help = "Check for unintended writes into a refcount block", 710 }, 711 { 712 .name = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE, 713 .type = QEMU_OPT_BOOL, 714 .help = "Check for unintended writes into the snapshot table", 715 }, 716 { 717 .name = QCOW2_OPT_OVERLAP_INACTIVE_L1, 718 .type = QEMU_OPT_BOOL, 719 .help = "Check for unintended writes into an inactive L1 table", 720 }, 721 { 722 .name = QCOW2_OPT_OVERLAP_INACTIVE_L2, 723 .type = QEMU_OPT_BOOL, 724 .help = "Check for unintended writes into an inactive L2 table", 725 }, 726 { 727 .name = QCOW2_OPT_OVERLAP_BITMAP_DIRECTORY, 728 .type = QEMU_OPT_BOOL, 729 .help = "Check for unintended writes into the bitmap directory", 730 }, 731 { 732 .name = QCOW2_OPT_CACHE_SIZE, 733 .type = QEMU_OPT_SIZE, 734 .help = "Maximum combined metadata (L2 tables and refcount blocks) " 735 "cache size", 736 }, 737 { 738 .name = QCOW2_OPT_L2_CACHE_SIZE, 739 .type = QEMU_OPT_SIZE, 740 .help = "Maximum L2 table cache size", 741 }, 742 { 743 .name = QCOW2_OPT_L2_CACHE_ENTRY_SIZE, 744 .type = QEMU_OPT_SIZE, 745 .help = "Size of each entry in the L2 cache", 746 }, 747 { 748 .name = QCOW2_OPT_REFCOUNT_CACHE_SIZE, 749 .type = QEMU_OPT_SIZE, 750 .help = "Maximum refcount block cache size", 751 }, 752 { 753 .name = QCOW2_OPT_CACHE_CLEAN_INTERVAL, 754 .type = QEMU_OPT_NUMBER, 755 .help = "Clean unused cache entries after this time (in seconds)", 756 }, 757 BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.", 758 "ID of secret providing qcow2 AES key or LUKS passphrase"), 759 { /* end of list */ } 760 }, 761 }; 762 763 static const char *overlap_bool_option_names[QCOW2_OL_MAX_BITNR] = { 764 [QCOW2_OL_MAIN_HEADER_BITNR] = QCOW2_OPT_OVERLAP_MAIN_HEADER, 765 [QCOW2_OL_ACTIVE_L1_BITNR] = QCOW2_OPT_OVERLAP_ACTIVE_L1, 766 [QCOW2_OL_ACTIVE_L2_BITNR] = QCOW2_OPT_OVERLAP_ACTIVE_L2, 767 [QCOW2_OL_REFCOUNT_TABLE_BITNR] = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE, 768 [QCOW2_OL_REFCOUNT_BLOCK_BITNR] = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK, 769 [QCOW2_OL_SNAPSHOT_TABLE_BITNR] = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE, 770 [QCOW2_OL_INACTIVE_L1_BITNR] = QCOW2_OPT_OVERLAP_INACTIVE_L1, 771 [QCOW2_OL_INACTIVE_L2_BITNR] = QCOW2_OPT_OVERLAP_INACTIVE_L2, 772 [QCOW2_OL_BITMAP_DIRECTORY_BITNR] = QCOW2_OPT_OVERLAP_BITMAP_DIRECTORY, 773 }; 774 775 static void cache_clean_timer_cb(void *opaque) 776 { 777 BlockDriverState *bs = opaque; 778 BDRVQcow2State *s = bs->opaque; 779 qcow2_cache_clean_unused(s->l2_table_cache); 780 qcow2_cache_clean_unused(s->refcount_block_cache); 781 timer_mod(s->cache_clean_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) + 782 (int64_t) s->cache_clean_interval * 1000); 783 } 784 785 static void cache_clean_timer_init(BlockDriverState *bs, AioContext *context) 786 { 787 BDRVQcow2State *s = bs->opaque; 788 if (s->cache_clean_interval > 0) { 789 s->cache_clean_timer = aio_timer_new(context, QEMU_CLOCK_VIRTUAL, 790 SCALE_MS, cache_clean_timer_cb, 791 bs); 792 timer_mod(s->cache_clean_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) + 793 (int64_t) s->cache_clean_interval * 1000); 794 } 795 } 796 797 static void cache_clean_timer_del(BlockDriverState *bs) 798 { 799 BDRVQcow2State *s = bs->opaque; 800 if (s->cache_clean_timer) { 801 timer_del(s->cache_clean_timer); 802 timer_free(s->cache_clean_timer); 803 s->cache_clean_timer = NULL; 804 } 805 } 806 807 static void qcow2_detach_aio_context(BlockDriverState *bs) 808 { 809 cache_clean_timer_del(bs); 810 } 811 812 static void qcow2_attach_aio_context(BlockDriverState *bs, 813 AioContext *new_context) 814 { 815 cache_clean_timer_init(bs, new_context); 816 } 817 818 static void read_cache_sizes(BlockDriverState *bs, QemuOpts *opts, 819 uint64_t *l2_cache_size, 820 uint64_t *l2_cache_entry_size, 821 uint64_t *refcount_cache_size, Error **errp) 822 { 823 BDRVQcow2State *s = bs->opaque; 824 uint64_t combined_cache_size, l2_cache_max_setting; 825 bool l2_cache_size_set, refcount_cache_size_set, combined_cache_size_set; 826 bool l2_cache_entry_size_set; 827 int min_refcount_cache = MIN_REFCOUNT_CACHE_SIZE * s->cluster_size; 828 uint64_t virtual_disk_size = bs->total_sectors * BDRV_SECTOR_SIZE; 829 uint64_t max_l2_cache = virtual_disk_size / (s->cluster_size / 8); 830 831 combined_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_CACHE_SIZE); 832 l2_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_L2_CACHE_SIZE); 833 refcount_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_REFCOUNT_CACHE_SIZE); 834 l2_cache_entry_size_set = qemu_opt_get(opts, QCOW2_OPT_L2_CACHE_ENTRY_SIZE); 835 836 combined_cache_size = qemu_opt_get_size(opts, QCOW2_OPT_CACHE_SIZE, 0); 837 l2_cache_max_setting = qemu_opt_get_size(opts, QCOW2_OPT_L2_CACHE_SIZE, 838 DEFAULT_L2_CACHE_MAX_SIZE); 839 *refcount_cache_size = qemu_opt_get_size(opts, 840 QCOW2_OPT_REFCOUNT_CACHE_SIZE, 0); 841 842 *l2_cache_entry_size = qemu_opt_get_size( 843 opts, QCOW2_OPT_L2_CACHE_ENTRY_SIZE, s->cluster_size); 844 845 *l2_cache_size = MIN(max_l2_cache, l2_cache_max_setting); 846 847 if (combined_cache_size_set) { 848 if (l2_cache_size_set && refcount_cache_size_set) { 849 error_setg(errp, QCOW2_OPT_CACHE_SIZE ", " QCOW2_OPT_L2_CACHE_SIZE 850 " and " QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not be set " 851 "at the same time"); 852 return; 853 } else if (l2_cache_size_set && 854 (l2_cache_max_setting > combined_cache_size)) { 855 error_setg(errp, QCOW2_OPT_L2_CACHE_SIZE " may not exceed " 856 QCOW2_OPT_CACHE_SIZE); 857 return; 858 } else if (*refcount_cache_size > combined_cache_size) { 859 error_setg(errp, QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not exceed " 860 QCOW2_OPT_CACHE_SIZE); 861 return; 862 } 863 864 if (l2_cache_size_set) { 865 *refcount_cache_size = combined_cache_size - *l2_cache_size; 866 } else if (refcount_cache_size_set) { 867 *l2_cache_size = combined_cache_size - *refcount_cache_size; 868 } else { 869 /* Assign as much memory as possible to the L2 cache, and 870 * use the remainder for the refcount cache */ 871 if (combined_cache_size >= max_l2_cache + min_refcount_cache) { 872 *l2_cache_size = max_l2_cache; 873 *refcount_cache_size = combined_cache_size - *l2_cache_size; 874 } else { 875 *refcount_cache_size = 876 MIN(combined_cache_size, min_refcount_cache); 877 *l2_cache_size = combined_cache_size - *refcount_cache_size; 878 } 879 } 880 } 881 882 /* 883 * If the L2 cache is not enough to cover the whole disk then 884 * default to 4KB entries. Smaller entries reduce the cost of 885 * loads and evictions and increase I/O performance. 886 */ 887 if (*l2_cache_size < max_l2_cache && !l2_cache_entry_size_set) { 888 *l2_cache_entry_size = MIN(s->cluster_size, 4096); 889 } 890 891 /* l2_cache_size and refcount_cache_size are ensured to have at least 892 * their minimum values in qcow2_update_options_prepare() */ 893 894 if (*l2_cache_entry_size < (1 << MIN_CLUSTER_BITS) || 895 *l2_cache_entry_size > s->cluster_size || 896 !is_power_of_2(*l2_cache_entry_size)) { 897 error_setg(errp, "L2 cache entry size must be a power of two " 898 "between %d and the cluster size (%d)", 899 1 << MIN_CLUSTER_BITS, s->cluster_size); 900 return; 901 } 902 } 903 904 typedef struct Qcow2ReopenState { 905 Qcow2Cache *l2_table_cache; 906 Qcow2Cache *refcount_block_cache; 907 int l2_slice_size; /* Number of entries in a slice of the L2 table */ 908 bool use_lazy_refcounts; 909 int overlap_check; 910 bool discard_passthrough[QCOW2_DISCARD_MAX]; 911 uint64_t cache_clean_interval; 912 QCryptoBlockOpenOptions *crypto_opts; /* Disk encryption runtime options */ 913 } Qcow2ReopenState; 914 915 static int qcow2_update_options_prepare(BlockDriverState *bs, 916 Qcow2ReopenState *r, 917 QDict *options, int flags, 918 Error **errp) 919 { 920 BDRVQcow2State *s = bs->opaque; 921 QemuOpts *opts = NULL; 922 const char *opt_overlap_check, *opt_overlap_check_template; 923 int overlap_check_template = 0; 924 uint64_t l2_cache_size, l2_cache_entry_size, refcount_cache_size; 925 int i; 926 const char *encryptfmt; 927 QDict *encryptopts = NULL; 928 Error *local_err = NULL; 929 int ret; 930 931 qdict_extract_subqdict(options, &encryptopts, "encrypt."); 932 encryptfmt = qdict_get_try_str(encryptopts, "format"); 933 934 opts = qemu_opts_create(&qcow2_runtime_opts, NULL, 0, &error_abort); 935 qemu_opts_absorb_qdict(opts, options, &local_err); 936 if (local_err) { 937 error_propagate(errp, local_err); 938 ret = -EINVAL; 939 goto fail; 940 } 941 942 /* get L2 table/refcount block cache size from command line options */ 943 read_cache_sizes(bs, opts, &l2_cache_size, &l2_cache_entry_size, 944 &refcount_cache_size, &local_err); 945 if (local_err) { 946 error_propagate(errp, local_err); 947 ret = -EINVAL; 948 goto fail; 949 } 950 951 l2_cache_size /= l2_cache_entry_size; 952 if (l2_cache_size < MIN_L2_CACHE_SIZE) { 953 l2_cache_size = MIN_L2_CACHE_SIZE; 954 } 955 if (l2_cache_size > INT_MAX) { 956 error_setg(errp, "L2 cache size too big"); 957 ret = -EINVAL; 958 goto fail; 959 } 960 961 refcount_cache_size /= s->cluster_size; 962 if (refcount_cache_size < MIN_REFCOUNT_CACHE_SIZE) { 963 refcount_cache_size = MIN_REFCOUNT_CACHE_SIZE; 964 } 965 if (refcount_cache_size > INT_MAX) { 966 error_setg(errp, "Refcount cache size too big"); 967 ret = -EINVAL; 968 goto fail; 969 } 970 971 /* alloc new L2 table/refcount block cache, flush old one */ 972 if (s->l2_table_cache) { 973 ret = qcow2_cache_flush(bs, s->l2_table_cache); 974 if (ret) { 975 error_setg_errno(errp, -ret, "Failed to flush the L2 table cache"); 976 goto fail; 977 } 978 } 979 980 if (s->refcount_block_cache) { 981 ret = qcow2_cache_flush(bs, s->refcount_block_cache); 982 if (ret) { 983 error_setg_errno(errp, -ret, 984 "Failed to flush the refcount block cache"); 985 goto fail; 986 } 987 } 988 989 r->l2_slice_size = l2_cache_entry_size / sizeof(uint64_t); 990 r->l2_table_cache = qcow2_cache_create(bs, l2_cache_size, 991 l2_cache_entry_size); 992 r->refcount_block_cache = qcow2_cache_create(bs, refcount_cache_size, 993 s->cluster_size); 994 if (r->l2_table_cache == NULL || r->refcount_block_cache == NULL) { 995 error_setg(errp, "Could not allocate metadata caches"); 996 ret = -ENOMEM; 997 goto fail; 998 } 999 1000 /* New interval for cache cleanup timer */ 1001 r->cache_clean_interval = 1002 qemu_opt_get_number(opts, QCOW2_OPT_CACHE_CLEAN_INTERVAL, 1003 DEFAULT_CACHE_CLEAN_INTERVAL); 1004 #ifndef CONFIG_LINUX 1005 if (r->cache_clean_interval != 0) { 1006 error_setg(errp, QCOW2_OPT_CACHE_CLEAN_INTERVAL 1007 " not supported on this host"); 1008 ret = -EINVAL; 1009 goto fail; 1010 } 1011 #endif 1012 if (r->cache_clean_interval > UINT_MAX) { 1013 error_setg(errp, "Cache clean interval too big"); 1014 ret = -EINVAL; 1015 goto fail; 1016 } 1017 1018 /* lazy-refcounts; flush if going from enabled to disabled */ 1019 r->use_lazy_refcounts = qemu_opt_get_bool(opts, QCOW2_OPT_LAZY_REFCOUNTS, 1020 (s->compatible_features & QCOW2_COMPAT_LAZY_REFCOUNTS)); 1021 if (r->use_lazy_refcounts && s->qcow_version < 3) { 1022 error_setg(errp, "Lazy refcounts require a qcow2 image with at least " 1023 "qemu 1.1 compatibility level"); 1024 ret = -EINVAL; 1025 goto fail; 1026 } 1027 1028 if (s->use_lazy_refcounts && !r->use_lazy_refcounts) { 1029 ret = qcow2_mark_clean(bs); 1030 if (ret < 0) { 1031 error_setg_errno(errp, -ret, "Failed to disable lazy refcounts"); 1032 goto fail; 1033 } 1034 } 1035 1036 /* Overlap check options */ 1037 opt_overlap_check = qemu_opt_get(opts, QCOW2_OPT_OVERLAP); 1038 opt_overlap_check_template = qemu_opt_get(opts, QCOW2_OPT_OVERLAP_TEMPLATE); 1039 if (opt_overlap_check_template && opt_overlap_check && 1040 strcmp(opt_overlap_check_template, opt_overlap_check)) 1041 { 1042 error_setg(errp, "Conflicting values for qcow2 options '" 1043 QCOW2_OPT_OVERLAP "' ('%s') and '" QCOW2_OPT_OVERLAP_TEMPLATE 1044 "' ('%s')", opt_overlap_check, opt_overlap_check_template); 1045 ret = -EINVAL; 1046 goto fail; 1047 } 1048 if (!opt_overlap_check) { 1049 opt_overlap_check = opt_overlap_check_template ?: "cached"; 1050 } 1051 1052 if (!strcmp(opt_overlap_check, "none")) { 1053 overlap_check_template = 0; 1054 } else if (!strcmp(opt_overlap_check, "constant")) { 1055 overlap_check_template = QCOW2_OL_CONSTANT; 1056 } else if (!strcmp(opt_overlap_check, "cached")) { 1057 overlap_check_template = QCOW2_OL_CACHED; 1058 } else if (!strcmp(opt_overlap_check, "all")) { 1059 overlap_check_template = QCOW2_OL_ALL; 1060 } else { 1061 error_setg(errp, "Unsupported value '%s' for qcow2 option " 1062 "'overlap-check'. Allowed are any of the following: " 1063 "none, constant, cached, all", opt_overlap_check); 1064 ret = -EINVAL; 1065 goto fail; 1066 } 1067 1068 r->overlap_check = 0; 1069 for (i = 0; i < QCOW2_OL_MAX_BITNR; i++) { 1070 /* overlap-check defines a template bitmask, but every flag may be 1071 * overwritten through the associated boolean option */ 1072 r->overlap_check |= 1073 qemu_opt_get_bool(opts, overlap_bool_option_names[i], 1074 overlap_check_template & (1 << i)) << i; 1075 } 1076 1077 r->discard_passthrough[QCOW2_DISCARD_NEVER] = false; 1078 r->discard_passthrough[QCOW2_DISCARD_ALWAYS] = true; 1079 r->discard_passthrough[QCOW2_DISCARD_REQUEST] = 1080 qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_REQUEST, 1081 flags & BDRV_O_UNMAP); 1082 r->discard_passthrough[QCOW2_DISCARD_SNAPSHOT] = 1083 qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_SNAPSHOT, true); 1084 r->discard_passthrough[QCOW2_DISCARD_OTHER] = 1085 qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_OTHER, false); 1086 1087 switch (s->crypt_method_header) { 1088 case QCOW_CRYPT_NONE: 1089 if (encryptfmt) { 1090 error_setg(errp, "No encryption in image header, but options " 1091 "specified format '%s'", encryptfmt); 1092 ret = -EINVAL; 1093 goto fail; 1094 } 1095 break; 1096 1097 case QCOW_CRYPT_AES: 1098 if (encryptfmt && !g_str_equal(encryptfmt, "aes")) { 1099 error_setg(errp, 1100 "Header reported 'aes' encryption format but " 1101 "options specify '%s'", encryptfmt); 1102 ret = -EINVAL; 1103 goto fail; 1104 } 1105 qdict_put_str(encryptopts, "format", "qcow"); 1106 r->crypto_opts = block_crypto_open_opts_init(encryptopts, errp); 1107 break; 1108 1109 case QCOW_CRYPT_LUKS: 1110 if (encryptfmt && !g_str_equal(encryptfmt, "luks")) { 1111 error_setg(errp, 1112 "Header reported 'luks' encryption format but " 1113 "options specify '%s'", encryptfmt); 1114 ret = -EINVAL; 1115 goto fail; 1116 } 1117 qdict_put_str(encryptopts, "format", "luks"); 1118 r->crypto_opts = block_crypto_open_opts_init(encryptopts, errp); 1119 break; 1120 1121 default: 1122 error_setg(errp, "Unsupported encryption method %d", 1123 s->crypt_method_header); 1124 break; 1125 } 1126 if (s->crypt_method_header != QCOW_CRYPT_NONE && !r->crypto_opts) { 1127 ret = -EINVAL; 1128 goto fail; 1129 } 1130 1131 ret = 0; 1132 fail: 1133 qobject_unref(encryptopts); 1134 qemu_opts_del(opts); 1135 opts = NULL; 1136 return ret; 1137 } 1138 1139 static void qcow2_update_options_commit(BlockDriverState *bs, 1140 Qcow2ReopenState *r) 1141 { 1142 BDRVQcow2State *s = bs->opaque; 1143 int i; 1144 1145 if (s->l2_table_cache) { 1146 qcow2_cache_destroy(s->l2_table_cache); 1147 } 1148 if (s->refcount_block_cache) { 1149 qcow2_cache_destroy(s->refcount_block_cache); 1150 } 1151 s->l2_table_cache = r->l2_table_cache; 1152 s->refcount_block_cache = r->refcount_block_cache; 1153 s->l2_slice_size = r->l2_slice_size; 1154 1155 s->overlap_check = r->overlap_check; 1156 s->use_lazy_refcounts = r->use_lazy_refcounts; 1157 1158 for (i = 0; i < QCOW2_DISCARD_MAX; i++) { 1159 s->discard_passthrough[i] = r->discard_passthrough[i]; 1160 } 1161 1162 if (s->cache_clean_interval != r->cache_clean_interval) { 1163 cache_clean_timer_del(bs); 1164 s->cache_clean_interval = r->cache_clean_interval; 1165 cache_clean_timer_init(bs, bdrv_get_aio_context(bs)); 1166 } 1167 1168 qapi_free_QCryptoBlockOpenOptions(s->crypto_opts); 1169 s->crypto_opts = r->crypto_opts; 1170 } 1171 1172 static void qcow2_update_options_abort(BlockDriverState *bs, 1173 Qcow2ReopenState *r) 1174 { 1175 if (r->l2_table_cache) { 1176 qcow2_cache_destroy(r->l2_table_cache); 1177 } 1178 if (r->refcount_block_cache) { 1179 qcow2_cache_destroy(r->refcount_block_cache); 1180 } 1181 qapi_free_QCryptoBlockOpenOptions(r->crypto_opts); 1182 } 1183 1184 static int qcow2_update_options(BlockDriverState *bs, QDict *options, 1185 int flags, Error **errp) 1186 { 1187 Qcow2ReopenState r = {}; 1188 int ret; 1189 1190 ret = qcow2_update_options_prepare(bs, &r, options, flags, errp); 1191 if (ret >= 0) { 1192 qcow2_update_options_commit(bs, &r); 1193 } else { 1194 qcow2_update_options_abort(bs, &r); 1195 } 1196 1197 return ret; 1198 } 1199 1200 /* Called with s->lock held. */ 1201 static int coroutine_fn qcow2_do_open(BlockDriverState *bs, QDict *options, 1202 int flags, Error **errp) 1203 { 1204 BDRVQcow2State *s = bs->opaque; 1205 unsigned int len, i; 1206 int ret = 0; 1207 QCowHeader header; 1208 Error *local_err = NULL; 1209 uint64_t ext_end; 1210 uint64_t l1_vm_state_index; 1211 bool update_header = false; 1212 1213 ret = bdrv_pread(bs->file, 0, &header, sizeof(header)); 1214 if (ret < 0) { 1215 error_setg_errno(errp, -ret, "Could not read qcow2 header"); 1216 goto fail; 1217 } 1218 header.magic = be32_to_cpu(header.magic); 1219 header.version = be32_to_cpu(header.version); 1220 header.backing_file_offset = be64_to_cpu(header.backing_file_offset); 1221 header.backing_file_size = be32_to_cpu(header.backing_file_size); 1222 header.size = be64_to_cpu(header.size); 1223 header.cluster_bits = be32_to_cpu(header.cluster_bits); 1224 header.crypt_method = be32_to_cpu(header.crypt_method); 1225 header.l1_table_offset = be64_to_cpu(header.l1_table_offset); 1226 header.l1_size = be32_to_cpu(header.l1_size); 1227 header.refcount_table_offset = be64_to_cpu(header.refcount_table_offset); 1228 header.refcount_table_clusters = 1229 be32_to_cpu(header.refcount_table_clusters); 1230 header.snapshots_offset = be64_to_cpu(header.snapshots_offset); 1231 header.nb_snapshots = be32_to_cpu(header.nb_snapshots); 1232 1233 if (header.magic != QCOW_MAGIC) { 1234 error_setg(errp, "Image is not in qcow2 format"); 1235 ret = -EINVAL; 1236 goto fail; 1237 } 1238 if (header.version < 2 || header.version > 3) { 1239 error_setg(errp, "Unsupported qcow2 version %" PRIu32, header.version); 1240 ret = -ENOTSUP; 1241 goto fail; 1242 } 1243 1244 s->qcow_version = header.version; 1245 1246 /* Initialise cluster size */ 1247 if (header.cluster_bits < MIN_CLUSTER_BITS || 1248 header.cluster_bits > MAX_CLUSTER_BITS) { 1249 error_setg(errp, "Unsupported cluster size: 2^%" PRIu32, 1250 header.cluster_bits); 1251 ret = -EINVAL; 1252 goto fail; 1253 } 1254 1255 s->cluster_bits = header.cluster_bits; 1256 s->cluster_size = 1 << s->cluster_bits; 1257 1258 /* Initialise version 3 header fields */ 1259 if (header.version == 2) { 1260 header.incompatible_features = 0; 1261 header.compatible_features = 0; 1262 header.autoclear_features = 0; 1263 header.refcount_order = 4; 1264 header.header_length = 72; 1265 } else { 1266 header.incompatible_features = 1267 be64_to_cpu(header.incompatible_features); 1268 header.compatible_features = be64_to_cpu(header.compatible_features); 1269 header.autoclear_features = be64_to_cpu(header.autoclear_features); 1270 header.refcount_order = be32_to_cpu(header.refcount_order); 1271 header.header_length = be32_to_cpu(header.header_length); 1272 1273 if (header.header_length < 104) { 1274 error_setg(errp, "qcow2 header too short"); 1275 ret = -EINVAL; 1276 goto fail; 1277 } 1278 } 1279 1280 if (header.header_length > s->cluster_size) { 1281 error_setg(errp, "qcow2 header exceeds cluster size"); 1282 ret = -EINVAL; 1283 goto fail; 1284 } 1285 1286 if (header.header_length > sizeof(header)) { 1287 s->unknown_header_fields_size = header.header_length - sizeof(header); 1288 s->unknown_header_fields = g_malloc(s->unknown_header_fields_size); 1289 ret = bdrv_pread(bs->file, sizeof(header), s->unknown_header_fields, 1290 s->unknown_header_fields_size); 1291 if (ret < 0) { 1292 error_setg_errno(errp, -ret, "Could not read unknown qcow2 header " 1293 "fields"); 1294 goto fail; 1295 } 1296 } 1297 1298 if (header.backing_file_offset > s->cluster_size) { 1299 error_setg(errp, "Invalid backing file offset"); 1300 ret = -EINVAL; 1301 goto fail; 1302 } 1303 1304 if (header.backing_file_offset) { 1305 ext_end = header.backing_file_offset; 1306 } else { 1307 ext_end = 1 << header.cluster_bits; 1308 } 1309 1310 /* Handle feature bits */ 1311 s->incompatible_features = header.incompatible_features; 1312 s->compatible_features = header.compatible_features; 1313 s->autoclear_features = header.autoclear_features; 1314 1315 if (s->incompatible_features & ~QCOW2_INCOMPAT_MASK) { 1316 void *feature_table = NULL; 1317 qcow2_read_extensions(bs, header.header_length, ext_end, 1318 &feature_table, flags, NULL, NULL); 1319 report_unsupported_feature(errp, feature_table, 1320 s->incompatible_features & 1321 ~QCOW2_INCOMPAT_MASK); 1322 ret = -ENOTSUP; 1323 g_free(feature_table); 1324 goto fail; 1325 } 1326 1327 if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) { 1328 /* Corrupt images may not be written to unless they are being repaired 1329 */ 1330 if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_CHECK)) { 1331 error_setg(errp, "qcow2: Image is corrupt; cannot be opened " 1332 "read/write"); 1333 ret = -EACCES; 1334 goto fail; 1335 } 1336 } 1337 1338 /* Check support for various header values */ 1339 if (header.refcount_order > 6) { 1340 error_setg(errp, "Reference count entry width too large; may not " 1341 "exceed 64 bits"); 1342 ret = -EINVAL; 1343 goto fail; 1344 } 1345 s->refcount_order = header.refcount_order; 1346 s->refcount_bits = 1 << s->refcount_order; 1347 s->refcount_max = UINT64_C(1) << (s->refcount_bits - 1); 1348 s->refcount_max += s->refcount_max - 1; 1349 1350 s->crypt_method_header = header.crypt_method; 1351 if (s->crypt_method_header) { 1352 if (bdrv_uses_whitelist() && 1353 s->crypt_method_header == QCOW_CRYPT_AES) { 1354 error_setg(errp, 1355 "Use of AES-CBC encrypted qcow2 images is no longer " 1356 "supported in system emulators"); 1357 error_append_hint(errp, 1358 "You can use 'qemu-img convert' to convert your " 1359 "image to an alternative supported format, such " 1360 "as unencrypted qcow2, or raw with the LUKS " 1361 "format instead.\n"); 1362 ret = -ENOSYS; 1363 goto fail; 1364 } 1365 1366 if (s->crypt_method_header == QCOW_CRYPT_AES) { 1367 s->crypt_physical_offset = false; 1368 } else { 1369 /* Assuming LUKS and any future crypt methods we 1370 * add will all use physical offsets, due to the 1371 * fact that the alternative is insecure... */ 1372 s->crypt_physical_offset = true; 1373 } 1374 1375 bs->encrypted = true; 1376 } 1377 1378 s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */ 1379 s->l2_size = 1 << s->l2_bits; 1380 /* 2^(s->refcount_order - 3) is the refcount width in bytes */ 1381 s->refcount_block_bits = s->cluster_bits - (s->refcount_order - 3); 1382 s->refcount_block_size = 1 << s->refcount_block_bits; 1383 bs->total_sectors = header.size / BDRV_SECTOR_SIZE; 1384 s->csize_shift = (62 - (s->cluster_bits - 8)); 1385 s->csize_mask = (1 << (s->cluster_bits - 8)) - 1; 1386 s->cluster_offset_mask = (1LL << s->csize_shift) - 1; 1387 1388 s->refcount_table_offset = header.refcount_table_offset; 1389 s->refcount_table_size = 1390 header.refcount_table_clusters << (s->cluster_bits - 3); 1391 1392 if (header.refcount_table_clusters == 0 && !(flags & BDRV_O_CHECK)) { 1393 error_setg(errp, "Image does not contain a reference count table"); 1394 ret = -EINVAL; 1395 goto fail; 1396 } 1397 1398 ret = qcow2_validate_table(bs, s->refcount_table_offset, 1399 header.refcount_table_clusters, 1400 s->cluster_size, QCOW_MAX_REFTABLE_SIZE, 1401 "Reference count table", errp); 1402 if (ret < 0) { 1403 goto fail; 1404 } 1405 1406 /* The total size in bytes of the snapshot table is checked in 1407 * qcow2_read_snapshots() because the size of each snapshot is 1408 * variable and we don't know it yet. 1409 * Here we only check the offset and number of snapshots. */ 1410 ret = qcow2_validate_table(bs, header.snapshots_offset, 1411 header.nb_snapshots, 1412 sizeof(QCowSnapshotHeader), 1413 sizeof(QCowSnapshotHeader) * QCOW_MAX_SNAPSHOTS, 1414 "Snapshot table", errp); 1415 if (ret < 0) { 1416 goto fail; 1417 } 1418 1419 /* read the level 1 table */ 1420 ret = qcow2_validate_table(bs, header.l1_table_offset, 1421 header.l1_size, sizeof(uint64_t), 1422 QCOW_MAX_L1_SIZE, "Active L1 table", errp); 1423 if (ret < 0) { 1424 goto fail; 1425 } 1426 s->l1_size = header.l1_size; 1427 s->l1_table_offset = header.l1_table_offset; 1428 1429 l1_vm_state_index = size_to_l1(s, header.size); 1430 if (l1_vm_state_index > INT_MAX) { 1431 error_setg(errp, "Image is too big"); 1432 ret = -EFBIG; 1433 goto fail; 1434 } 1435 s->l1_vm_state_index = l1_vm_state_index; 1436 1437 /* the L1 table must contain at least enough entries to put 1438 header.size bytes */ 1439 if (s->l1_size < s->l1_vm_state_index) { 1440 error_setg(errp, "L1 table is too small"); 1441 ret = -EINVAL; 1442 goto fail; 1443 } 1444 1445 if (s->l1_size > 0) { 1446 s->l1_table = qemu_try_blockalign(bs->file->bs, 1447 ROUND_UP(s->l1_size * sizeof(uint64_t), 512)); 1448 if (s->l1_table == NULL) { 1449 error_setg(errp, "Could not allocate L1 table"); 1450 ret = -ENOMEM; 1451 goto fail; 1452 } 1453 ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table, 1454 s->l1_size * sizeof(uint64_t)); 1455 if (ret < 0) { 1456 error_setg_errno(errp, -ret, "Could not read L1 table"); 1457 goto fail; 1458 } 1459 for(i = 0;i < s->l1_size; i++) { 1460 s->l1_table[i] = be64_to_cpu(s->l1_table[i]); 1461 } 1462 } 1463 1464 /* Parse driver-specific options */ 1465 ret = qcow2_update_options(bs, options, flags, errp); 1466 if (ret < 0) { 1467 goto fail; 1468 } 1469 1470 s->flags = flags; 1471 1472 ret = qcow2_refcount_init(bs); 1473 if (ret != 0) { 1474 error_setg_errno(errp, -ret, "Could not initialize refcount handling"); 1475 goto fail; 1476 } 1477 1478 QLIST_INIT(&s->cluster_allocs); 1479 QTAILQ_INIT(&s->discards); 1480 1481 /* read qcow2 extensions */ 1482 if (qcow2_read_extensions(bs, header.header_length, ext_end, NULL, 1483 flags, &update_header, &local_err)) { 1484 error_propagate(errp, local_err); 1485 ret = -EINVAL; 1486 goto fail; 1487 } 1488 1489 /* Open external data file */ 1490 s->data_file = bdrv_open_child(NULL, options, "data-file", bs, &child_file, 1491 true, &local_err); 1492 if (local_err) { 1493 error_propagate(errp, local_err); 1494 ret = -EINVAL; 1495 goto fail; 1496 } 1497 1498 if (s->incompatible_features & QCOW2_INCOMPAT_DATA_FILE) { 1499 if (!s->data_file && s->image_data_file) { 1500 s->data_file = bdrv_open_child(s->image_data_file, options, 1501 "data-file", bs, &child_file, 1502 false, errp); 1503 if (!s->data_file) { 1504 ret = -EINVAL; 1505 goto fail; 1506 } 1507 } 1508 if (!s->data_file) { 1509 error_setg(errp, "'data-file' is required for this image"); 1510 ret = -EINVAL; 1511 goto fail; 1512 } 1513 } else { 1514 if (s->data_file) { 1515 error_setg(errp, "'data-file' can only be set for images with an " 1516 "external data file"); 1517 ret = -EINVAL; 1518 goto fail; 1519 } 1520 1521 s->data_file = bs->file; 1522 1523 if (data_file_is_raw(bs)) { 1524 error_setg(errp, "data-file-raw requires a data file"); 1525 ret = -EINVAL; 1526 goto fail; 1527 } 1528 } 1529 1530 /* qcow2_read_extension may have set up the crypto context 1531 * if the crypt method needs a header region, some methods 1532 * don't need header extensions, so must check here 1533 */ 1534 if (s->crypt_method_header && !s->crypto) { 1535 if (s->crypt_method_header == QCOW_CRYPT_AES) { 1536 unsigned int cflags = 0; 1537 if (flags & BDRV_O_NO_IO) { 1538 cflags |= QCRYPTO_BLOCK_OPEN_NO_IO; 1539 } 1540 s->crypto = qcrypto_block_open(s->crypto_opts, "encrypt.", 1541 NULL, NULL, cflags, 1542 QCOW2_MAX_THREADS, errp); 1543 if (!s->crypto) { 1544 ret = -EINVAL; 1545 goto fail; 1546 } 1547 } else if (!(flags & BDRV_O_NO_IO)) { 1548 error_setg(errp, "Missing CRYPTO header for crypt method %d", 1549 s->crypt_method_header); 1550 ret = -EINVAL; 1551 goto fail; 1552 } 1553 } 1554 1555 /* read the backing file name */ 1556 if (header.backing_file_offset != 0) { 1557 len = header.backing_file_size; 1558 if (len > MIN(1023, s->cluster_size - header.backing_file_offset) || 1559 len >= sizeof(bs->backing_file)) { 1560 error_setg(errp, "Backing file name too long"); 1561 ret = -EINVAL; 1562 goto fail; 1563 } 1564 ret = bdrv_pread(bs->file, header.backing_file_offset, 1565 bs->auto_backing_file, len); 1566 if (ret < 0) { 1567 error_setg_errno(errp, -ret, "Could not read backing file name"); 1568 goto fail; 1569 } 1570 bs->auto_backing_file[len] = '\0'; 1571 pstrcpy(bs->backing_file, sizeof(bs->backing_file), 1572 bs->auto_backing_file); 1573 s->image_backing_file = g_strdup(bs->auto_backing_file); 1574 } 1575 1576 /* Internal snapshots */ 1577 s->snapshots_offset = header.snapshots_offset; 1578 s->nb_snapshots = header.nb_snapshots; 1579 1580 ret = qcow2_read_snapshots(bs); 1581 if (ret < 0) { 1582 error_setg_errno(errp, -ret, "Could not read snapshots"); 1583 goto fail; 1584 } 1585 1586 /* Clear unknown autoclear feature bits */ 1587 update_header |= s->autoclear_features & ~QCOW2_AUTOCLEAR_MASK; 1588 update_header = 1589 update_header && !bs->read_only && !(flags & BDRV_O_INACTIVE); 1590 if (update_header) { 1591 s->autoclear_features &= QCOW2_AUTOCLEAR_MASK; 1592 } 1593 1594 /* == Handle persistent dirty bitmaps == 1595 * 1596 * We want load dirty bitmaps in three cases: 1597 * 1598 * 1. Normal open of the disk in active mode, not related to invalidation 1599 * after migration. 1600 * 1601 * 2. Invalidation of the target vm after pre-copy phase of migration, if 1602 * bitmaps are _not_ migrating through migration channel, i.e. 1603 * 'dirty-bitmaps' capability is disabled. 1604 * 1605 * 3. Invalidation of source vm after failed or canceled migration. 1606 * This is a very interesting case. There are two possible types of 1607 * bitmaps: 1608 * 1609 * A. Stored on inactivation and removed. They should be loaded from the 1610 * image. 1611 * 1612 * B. Not stored: not-persistent bitmaps and bitmaps, migrated through 1613 * the migration channel (with dirty-bitmaps capability). 1614 * 1615 * On the other hand, there are two possible sub-cases: 1616 * 1617 * 3.1 disk was changed by somebody else while were inactive. In this 1618 * case all in-RAM dirty bitmaps (both persistent and not) are 1619 * definitely invalid. And we don't have any method to determine 1620 * this. 1621 * 1622 * Simple and safe thing is to just drop all the bitmaps of type B on 1623 * inactivation. But in this case we lose bitmaps in valid 4.2 case. 1624 * 1625 * On the other hand, resuming source vm, if disk was already changed 1626 * is a bad thing anyway: not only bitmaps, the whole vm state is 1627 * out of sync with disk. 1628 * 1629 * This means, that user or management tool, who for some reason 1630 * decided to resume source vm, after disk was already changed by 1631 * target vm, should at least drop all dirty bitmaps by hand. 1632 * 1633 * So, we can ignore this case for now, but TODO: "generation" 1634 * extension for qcow2, to determine, that image was changed after 1635 * last inactivation. And if it is changed, we will drop (or at least 1636 * mark as 'invalid' all the bitmaps of type B, both persistent 1637 * and not). 1638 * 1639 * 3.2 disk was _not_ changed while were inactive. Bitmaps may be saved 1640 * to disk ('dirty-bitmaps' capability disabled), or not saved 1641 * ('dirty-bitmaps' capability enabled), but we don't need to care 1642 * of: let's load bitmaps as always: stored bitmaps will be loaded, 1643 * and not stored has flag IN_USE=1 in the image and will be skipped 1644 * on loading. 1645 * 1646 * One remaining possible case when we don't want load bitmaps: 1647 * 1648 * 4. Open disk in inactive mode in target vm (bitmaps are migrating or 1649 * will be loaded on invalidation, no needs try loading them before) 1650 */ 1651 1652 if (!(bdrv_get_flags(bs) & BDRV_O_INACTIVE)) { 1653 /* It's case 1, 2 or 3.2. Or 3.1 which is BUG in management layer. */ 1654 bool header_updated = qcow2_load_dirty_bitmaps(bs, &local_err); 1655 1656 update_header = update_header && !header_updated; 1657 } 1658 if (local_err != NULL) { 1659 error_propagate(errp, local_err); 1660 ret = -EINVAL; 1661 goto fail; 1662 } 1663 1664 if (update_header) { 1665 ret = qcow2_update_header(bs); 1666 if (ret < 0) { 1667 error_setg_errno(errp, -ret, "Could not update qcow2 header"); 1668 goto fail; 1669 } 1670 } 1671 1672 bs->supported_zero_flags = header.version >= 3 ? BDRV_REQ_MAY_UNMAP : 0; 1673 1674 /* Repair image if dirty */ 1675 if (!(flags & (BDRV_O_CHECK | BDRV_O_INACTIVE)) && !bs->read_only && 1676 (s->incompatible_features & QCOW2_INCOMPAT_DIRTY)) { 1677 BdrvCheckResult result = {0}; 1678 1679 ret = qcow2_co_check_locked(bs, &result, 1680 BDRV_FIX_ERRORS | BDRV_FIX_LEAKS); 1681 if (ret < 0 || result.check_errors) { 1682 if (ret >= 0) { 1683 ret = -EIO; 1684 } 1685 error_setg_errno(errp, -ret, "Could not repair dirty image"); 1686 goto fail; 1687 } 1688 } 1689 1690 #ifdef DEBUG_ALLOC 1691 { 1692 BdrvCheckResult result = {0}; 1693 qcow2_check_refcounts(bs, &result, 0); 1694 } 1695 #endif 1696 1697 qemu_co_queue_init(&s->thread_task_queue); 1698 1699 return ret; 1700 1701 fail: 1702 g_free(s->image_data_file); 1703 if (has_data_file(bs)) { 1704 bdrv_unref_child(bs, s->data_file); 1705 } 1706 g_free(s->unknown_header_fields); 1707 cleanup_unknown_header_ext(bs); 1708 qcow2_free_snapshots(bs); 1709 qcow2_refcount_close(bs); 1710 qemu_vfree(s->l1_table); 1711 /* else pre-write overlap checks in cache_destroy may crash */ 1712 s->l1_table = NULL; 1713 cache_clean_timer_del(bs); 1714 if (s->l2_table_cache) { 1715 qcow2_cache_destroy(s->l2_table_cache); 1716 } 1717 if (s->refcount_block_cache) { 1718 qcow2_cache_destroy(s->refcount_block_cache); 1719 } 1720 qcrypto_block_free(s->crypto); 1721 qapi_free_QCryptoBlockOpenOptions(s->crypto_opts); 1722 return ret; 1723 } 1724 1725 typedef struct QCow2OpenCo { 1726 BlockDriverState *bs; 1727 QDict *options; 1728 int flags; 1729 Error **errp; 1730 int ret; 1731 } QCow2OpenCo; 1732 1733 static void coroutine_fn qcow2_open_entry(void *opaque) 1734 { 1735 QCow2OpenCo *qoc = opaque; 1736 BDRVQcow2State *s = qoc->bs->opaque; 1737 1738 qemu_co_mutex_lock(&s->lock); 1739 qoc->ret = qcow2_do_open(qoc->bs, qoc->options, qoc->flags, qoc->errp); 1740 qemu_co_mutex_unlock(&s->lock); 1741 } 1742 1743 static int qcow2_open(BlockDriverState *bs, QDict *options, int flags, 1744 Error **errp) 1745 { 1746 BDRVQcow2State *s = bs->opaque; 1747 QCow2OpenCo qoc = { 1748 .bs = bs, 1749 .options = options, 1750 .flags = flags, 1751 .errp = errp, 1752 .ret = -EINPROGRESS 1753 }; 1754 1755 bs->file = bdrv_open_child(NULL, options, "file", bs, &child_file, 1756 false, errp); 1757 if (!bs->file) { 1758 return -EINVAL; 1759 } 1760 1761 /* Initialise locks */ 1762 qemu_co_mutex_init(&s->lock); 1763 1764 if (qemu_in_coroutine()) { 1765 /* From bdrv_co_create. */ 1766 qcow2_open_entry(&qoc); 1767 } else { 1768 assert(qemu_get_current_aio_context() == qemu_get_aio_context()); 1769 qemu_coroutine_enter(qemu_coroutine_create(qcow2_open_entry, &qoc)); 1770 BDRV_POLL_WHILE(bs, qoc.ret == -EINPROGRESS); 1771 } 1772 return qoc.ret; 1773 } 1774 1775 static void qcow2_refresh_limits(BlockDriverState *bs, Error **errp) 1776 { 1777 BDRVQcow2State *s = bs->opaque; 1778 1779 if (bs->encrypted) { 1780 /* Encryption works on a sector granularity */ 1781 bs->bl.request_alignment = qcrypto_block_get_sector_size(s->crypto); 1782 } 1783 bs->bl.pwrite_zeroes_alignment = s->cluster_size; 1784 bs->bl.pdiscard_alignment = s->cluster_size; 1785 } 1786 1787 static int qcow2_reopen_prepare(BDRVReopenState *state, 1788 BlockReopenQueue *queue, Error **errp) 1789 { 1790 Qcow2ReopenState *r; 1791 int ret; 1792 1793 r = g_new0(Qcow2ReopenState, 1); 1794 state->opaque = r; 1795 1796 ret = qcow2_update_options_prepare(state->bs, r, state->options, 1797 state->flags, errp); 1798 if (ret < 0) { 1799 goto fail; 1800 } 1801 1802 /* We need to write out any unwritten data if we reopen read-only. */ 1803 if ((state->flags & BDRV_O_RDWR) == 0) { 1804 ret = qcow2_reopen_bitmaps_ro(state->bs, errp); 1805 if (ret < 0) { 1806 goto fail; 1807 } 1808 1809 ret = bdrv_flush(state->bs); 1810 if (ret < 0) { 1811 goto fail; 1812 } 1813 1814 ret = qcow2_mark_clean(state->bs); 1815 if (ret < 0) { 1816 goto fail; 1817 } 1818 } 1819 1820 return 0; 1821 1822 fail: 1823 qcow2_update_options_abort(state->bs, r); 1824 g_free(r); 1825 return ret; 1826 } 1827 1828 static void qcow2_reopen_commit(BDRVReopenState *state) 1829 { 1830 qcow2_update_options_commit(state->bs, state->opaque); 1831 g_free(state->opaque); 1832 } 1833 1834 static void qcow2_reopen_abort(BDRVReopenState *state) 1835 { 1836 qcow2_update_options_abort(state->bs, state->opaque); 1837 g_free(state->opaque); 1838 } 1839 1840 static void qcow2_join_options(QDict *options, QDict *old_options) 1841 { 1842 bool has_new_overlap_template = 1843 qdict_haskey(options, QCOW2_OPT_OVERLAP) || 1844 qdict_haskey(options, QCOW2_OPT_OVERLAP_TEMPLATE); 1845 bool has_new_total_cache_size = 1846 qdict_haskey(options, QCOW2_OPT_CACHE_SIZE); 1847 bool has_all_cache_options; 1848 1849 /* New overlap template overrides all old overlap options */ 1850 if (has_new_overlap_template) { 1851 qdict_del(old_options, QCOW2_OPT_OVERLAP); 1852 qdict_del(old_options, QCOW2_OPT_OVERLAP_TEMPLATE); 1853 qdict_del(old_options, QCOW2_OPT_OVERLAP_MAIN_HEADER); 1854 qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L1); 1855 qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L2); 1856 qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_TABLE); 1857 qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK); 1858 qdict_del(old_options, QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE); 1859 qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L1); 1860 qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L2); 1861 } 1862 1863 /* New total cache size overrides all old options */ 1864 if (qdict_haskey(options, QCOW2_OPT_CACHE_SIZE)) { 1865 qdict_del(old_options, QCOW2_OPT_L2_CACHE_SIZE); 1866 qdict_del(old_options, QCOW2_OPT_REFCOUNT_CACHE_SIZE); 1867 } 1868 1869 qdict_join(options, old_options, false); 1870 1871 /* 1872 * If after merging all cache size options are set, an old total size is 1873 * overwritten. Do keep all options, however, if all three are new. The 1874 * resulting error message is what we want to happen. 1875 */ 1876 has_all_cache_options = 1877 qdict_haskey(options, QCOW2_OPT_CACHE_SIZE) || 1878 qdict_haskey(options, QCOW2_OPT_L2_CACHE_SIZE) || 1879 qdict_haskey(options, QCOW2_OPT_REFCOUNT_CACHE_SIZE); 1880 1881 if (has_all_cache_options && !has_new_total_cache_size) { 1882 qdict_del(options, QCOW2_OPT_CACHE_SIZE); 1883 } 1884 } 1885 1886 static int coroutine_fn qcow2_co_block_status(BlockDriverState *bs, 1887 bool want_zero, 1888 int64_t offset, int64_t count, 1889 int64_t *pnum, int64_t *map, 1890 BlockDriverState **file) 1891 { 1892 BDRVQcow2State *s = bs->opaque; 1893 uint64_t cluster_offset; 1894 int index_in_cluster, ret; 1895 unsigned int bytes; 1896 int status = 0; 1897 1898 if (!s->metadata_preallocation_checked) { 1899 ret = qcow2_detect_metadata_preallocation(bs); 1900 s->metadata_preallocation = (ret == 1); 1901 s->metadata_preallocation_checked = true; 1902 } 1903 1904 bytes = MIN(INT_MAX, count); 1905 qemu_co_mutex_lock(&s->lock); 1906 ret = qcow2_get_cluster_offset(bs, offset, &bytes, &cluster_offset); 1907 qemu_co_mutex_unlock(&s->lock); 1908 if (ret < 0) { 1909 return ret; 1910 } 1911 1912 *pnum = bytes; 1913 1914 if ((ret == QCOW2_CLUSTER_NORMAL || ret == QCOW2_CLUSTER_ZERO_ALLOC) && 1915 !s->crypto) { 1916 index_in_cluster = offset & (s->cluster_size - 1); 1917 *map = cluster_offset | index_in_cluster; 1918 *file = s->data_file->bs; 1919 status |= BDRV_BLOCK_OFFSET_VALID; 1920 } 1921 if (ret == QCOW2_CLUSTER_ZERO_PLAIN || ret == QCOW2_CLUSTER_ZERO_ALLOC) { 1922 status |= BDRV_BLOCK_ZERO; 1923 } else if (ret != QCOW2_CLUSTER_UNALLOCATED) { 1924 status |= BDRV_BLOCK_DATA; 1925 } 1926 if (s->metadata_preallocation && (status & BDRV_BLOCK_DATA) && 1927 (status & BDRV_BLOCK_OFFSET_VALID)) 1928 { 1929 status |= BDRV_BLOCK_RECURSE; 1930 } 1931 return status; 1932 } 1933 1934 static coroutine_fn int qcow2_handle_l2meta(BlockDriverState *bs, 1935 QCowL2Meta **pl2meta, 1936 bool link_l2) 1937 { 1938 int ret = 0; 1939 QCowL2Meta *l2meta = *pl2meta; 1940 1941 while (l2meta != NULL) { 1942 QCowL2Meta *next; 1943 1944 if (link_l2) { 1945 ret = qcow2_alloc_cluster_link_l2(bs, l2meta); 1946 if (ret) { 1947 goto out; 1948 } 1949 } else { 1950 qcow2_alloc_cluster_abort(bs, l2meta); 1951 } 1952 1953 /* Take the request off the list of running requests */ 1954 if (l2meta->nb_clusters != 0) { 1955 QLIST_REMOVE(l2meta, next_in_flight); 1956 } 1957 1958 qemu_co_queue_restart_all(&l2meta->dependent_requests); 1959 1960 next = l2meta->next; 1961 g_free(l2meta); 1962 l2meta = next; 1963 } 1964 out: 1965 *pl2meta = l2meta; 1966 return ret; 1967 } 1968 1969 static coroutine_fn int qcow2_co_preadv(BlockDriverState *bs, uint64_t offset, 1970 uint64_t bytes, QEMUIOVector *qiov, 1971 int flags) 1972 { 1973 BDRVQcow2State *s = bs->opaque; 1974 int offset_in_cluster; 1975 int ret; 1976 unsigned int cur_bytes; /* number of bytes in current iteration */ 1977 uint64_t cluster_offset = 0; 1978 uint64_t bytes_done = 0; 1979 QEMUIOVector hd_qiov; 1980 uint8_t *cluster_data = NULL; 1981 1982 qemu_iovec_init(&hd_qiov, qiov->niov); 1983 1984 while (bytes != 0) { 1985 1986 /* prepare next request */ 1987 cur_bytes = MIN(bytes, INT_MAX); 1988 if (s->crypto) { 1989 cur_bytes = MIN(cur_bytes, 1990 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size); 1991 } 1992 1993 qemu_co_mutex_lock(&s->lock); 1994 ret = qcow2_get_cluster_offset(bs, offset, &cur_bytes, &cluster_offset); 1995 qemu_co_mutex_unlock(&s->lock); 1996 if (ret < 0) { 1997 goto fail; 1998 } 1999 2000 offset_in_cluster = offset_into_cluster(s, offset); 2001 2002 qemu_iovec_reset(&hd_qiov); 2003 qemu_iovec_concat(&hd_qiov, qiov, bytes_done, cur_bytes); 2004 2005 switch (ret) { 2006 case QCOW2_CLUSTER_UNALLOCATED: 2007 2008 if (bs->backing) { 2009 BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO); 2010 ret = bdrv_co_preadv(bs->backing, offset, cur_bytes, 2011 &hd_qiov, 0); 2012 if (ret < 0) { 2013 goto fail; 2014 } 2015 } else { 2016 /* Note: in this case, no need to wait */ 2017 qemu_iovec_memset(&hd_qiov, 0, 0, cur_bytes); 2018 } 2019 break; 2020 2021 case QCOW2_CLUSTER_ZERO_PLAIN: 2022 case QCOW2_CLUSTER_ZERO_ALLOC: 2023 qemu_iovec_memset(&hd_qiov, 0, 0, cur_bytes); 2024 break; 2025 2026 case QCOW2_CLUSTER_COMPRESSED: 2027 ret = qcow2_co_preadv_compressed(bs, cluster_offset, 2028 offset, cur_bytes, 2029 &hd_qiov); 2030 if (ret < 0) { 2031 goto fail; 2032 } 2033 2034 break; 2035 2036 case QCOW2_CLUSTER_NORMAL: 2037 if ((cluster_offset & 511) != 0) { 2038 ret = -EIO; 2039 goto fail; 2040 } 2041 2042 if (bs->encrypted) { 2043 assert(s->crypto); 2044 2045 /* 2046 * For encrypted images, read everything into a temporary 2047 * contiguous buffer on which the AES functions can work. 2048 */ 2049 if (!cluster_data) { 2050 cluster_data = 2051 qemu_try_blockalign(s->data_file->bs, 2052 QCOW_MAX_CRYPT_CLUSTERS 2053 * s->cluster_size); 2054 if (cluster_data == NULL) { 2055 ret = -ENOMEM; 2056 goto fail; 2057 } 2058 } 2059 2060 assert(cur_bytes <= QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size); 2061 qemu_iovec_reset(&hd_qiov); 2062 qemu_iovec_add(&hd_qiov, cluster_data, cur_bytes); 2063 } 2064 2065 BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO); 2066 ret = bdrv_co_preadv(s->data_file, 2067 cluster_offset + offset_in_cluster, 2068 cur_bytes, &hd_qiov, 0); 2069 if (ret < 0) { 2070 goto fail; 2071 } 2072 if (bs->encrypted) { 2073 assert(s->crypto); 2074 assert((offset & (BDRV_SECTOR_SIZE - 1)) == 0); 2075 assert((cur_bytes & (BDRV_SECTOR_SIZE - 1)) == 0); 2076 if (qcow2_co_decrypt(bs, cluster_offset, offset, 2077 cluster_data, cur_bytes) < 0) { 2078 ret = -EIO; 2079 goto fail; 2080 } 2081 qemu_iovec_from_buf(qiov, bytes_done, cluster_data, cur_bytes); 2082 } 2083 break; 2084 2085 default: 2086 g_assert_not_reached(); 2087 ret = -EIO; 2088 goto fail; 2089 } 2090 2091 bytes -= cur_bytes; 2092 offset += cur_bytes; 2093 bytes_done += cur_bytes; 2094 } 2095 ret = 0; 2096 2097 fail: 2098 qemu_iovec_destroy(&hd_qiov); 2099 qemu_vfree(cluster_data); 2100 2101 return ret; 2102 } 2103 2104 /* Check if it's possible to merge a write request with the writing of 2105 * the data from the COW regions */ 2106 static bool merge_cow(uint64_t offset, unsigned bytes, 2107 QEMUIOVector *hd_qiov, QCowL2Meta *l2meta) 2108 { 2109 QCowL2Meta *m; 2110 2111 for (m = l2meta; m != NULL; m = m->next) { 2112 /* If both COW regions are empty then there's nothing to merge */ 2113 if (m->cow_start.nb_bytes == 0 && m->cow_end.nb_bytes == 0) { 2114 continue; 2115 } 2116 2117 /* If COW regions are handled already, skip this too */ 2118 if (m->skip_cow) { 2119 continue; 2120 } 2121 2122 /* The data (middle) region must be immediately after the 2123 * start region */ 2124 if (l2meta_cow_start(m) + m->cow_start.nb_bytes != offset) { 2125 continue; 2126 } 2127 2128 /* The end region must be immediately after the data (middle) 2129 * region */ 2130 if (m->offset + m->cow_end.offset != offset + bytes) { 2131 continue; 2132 } 2133 2134 /* Make sure that adding both COW regions to the QEMUIOVector 2135 * does not exceed IOV_MAX */ 2136 if (hd_qiov->niov > IOV_MAX - 2) { 2137 continue; 2138 } 2139 2140 m->data_qiov = hd_qiov; 2141 return true; 2142 } 2143 2144 return false; 2145 } 2146 2147 static bool is_unallocated(BlockDriverState *bs, int64_t offset, int64_t bytes) 2148 { 2149 int64_t nr; 2150 return !bytes || 2151 (!bdrv_is_allocated_above(bs, NULL, offset, bytes, &nr) && nr == bytes); 2152 } 2153 2154 static bool is_zero_cow(BlockDriverState *bs, QCowL2Meta *m) 2155 { 2156 /* 2157 * This check is designed for optimization shortcut so it must be 2158 * efficient. 2159 * Instead of is_zero(), use is_unallocated() as it is faster (but not 2160 * as accurate and can result in false negatives). 2161 */ 2162 return is_unallocated(bs, m->offset + m->cow_start.offset, 2163 m->cow_start.nb_bytes) && 2164 is_unallocated(bs, m->offset + m->cow_end.offset, 2165 m->cow_end.nb_bytes); 2166 } 2167 2168 static int handle_alloc_space(BlockDriverState *bs, QCowL2Meta *l2meta) 2169 { 2170 BDRVQcow2State *s = bs->opaque; 2171 QCowL2Meta *m; 2172 2173 if (!(s->data_file->bs->supported_zero_flags & BDRV_REQ_NO_FALLBACK)) { 2174 return 0; 2175 } 2176 2177 if (bs->encrypted) { 2178 return 0; 2179 } 2180 2181 for (m = l2meta; m != NULL; m = m->next) { 2182 int ret; 2183 2184 if (!m->cow_start.nb_bytes && !m->cow_end.nb_bytes) { 2185 continue; 2186 } 2187 2188 if (!is_zero_cow(bs, m)) { 2189 continue; 2190 } 2191 2192 /* 2193 * instead of writing zero COW buffers, 2194 * efficiently zero out the whole clusters 2195 */ 2196 2197 ret = qcow2_pre_write_overlap_check(bs, 0, m->alloc_offset, 2198 m->nb_clusters * s->cluster_size, 2199 true); 2200 if (ret < 0) { 2201 return ret; 2202 } 2203 2204 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC_SPACE); 2205 ret = bdrv_co_pwrite_zeroes(s->data_file, m->alloc_offset, 2206 m->nb_clusters * s->cluster_size, 2207 BDRV_REQ_NO_FALLBACK); 2208 if (ret < 0) { 2209 if (ret != -ENOTSUP && ret != -EAGAIN) { 2210 return ret; 2211 } 2212 continue; 2213 } 2214 2215 trace_qcow2_skip_cow(qemu_coroutine_self(), m->offset, m->nb_clusters); 2216 m->skip_cow = true; 2217 } 2218 return 0; 2219 } 2220 2221 static coroutine_fn int qcow2_co_pwritev(BlockDriverState *bs, uint64_t offset, 2222 uint64_t bytes, QEMUIOVector *qiov, 2223 int flags) 2224 { 2225 BDRVQcow2State *s = bs->opaque; 2226 int offset_in_cluster; 2227 int ret; 2228 unsigned int cur_bytes; /* number of sectors in current iteration */ 2229 uint64_t cluster_offset; 2230 QEMUIOVector hd_qiov; 2231 uint64_t bytes_done = 0; 2232 uint8_t *cluster_data = NULL; 2233 QCowL2Meta *l2meta = NULL; 2234 2235 trace_qcow2_writev_start_req(qemu_coroutine_self(), offset, bytes); 2236 2237 qemu_iovec_init(&hd_qiov, qiov->niov); 2238 2239 qemu_co_mutex_lock(&s->lock); 2240 2241 while (bytes != 0) { 2242 2243 l2meta = NULL; 2244 2245 trace_qcow2_writev_start_part(qemu_coroutine_self()); 2246 offset_in_cluster = offset_into_cluster(s, offset); 2247 cur_bytes = MIN(bytes, INT_MAX); 2248 if (bs->encrypted) { 2249 cur_bytes = MIN(cur_bytes, 2250 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size 2251 - offset_in_cluster); 2252 } 2253 2254 ret = qcow2_alloc_cluster_offset(bs, offset, &cur_bytes, 2255 &cluster_offset, &l2meta); 2256 if (ret < 0) { 2257 goto out_locked; 2258 } 2259 2260 assert((cluster_offset & 511) == 0); 2261 2262 ret = qcow2_pre_write_overlap_check(bs, 0, 2263 cluster_offset + offset_in_cluster, 2264 cur_bytes, true); 2265 if (ret < 0) { 2266 goto out_locked; 2267 } 2268 2269 qemu_co_mutex_unlock(&s->lock); 2270 2271 qemu_iovec_reset(&hd_qiov); 2272 qemu_iovec_concat(&hd_qiov, qiov, bytes_done, cur_bytes); 2273 2274 if (bs->encrypted) { 2275 assert(s->crypto); 2276 if (!cluster_data) { 2277 cluster_data = qemu_try_blockalign(bs->file->bs, 2278 QCOW_MAX_CRYPT_CLUSTERS 2279 * s->cluster_size); 2280 if (cluster_data == NULL) { 2281 ret = -ENOMEM; 2282 goto out_unlocked; 2283 } 2284 } 2285 2286 assert(hd_qiov.size <= 2287 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size); 2288 qemu_iovec_to_buf(&hd_qiov, 0, cluster_data, hd_qiov.size); 2289 2290 if (qcow2_co_encrypt(bs, cluster_offset, offset, 2291 cluster_data, cur_bytes) < 0) { 2292 ret = -EIO; 2293 goto out_unlocked; 2294 } 2295 2296 qemu_iovec_reset(&hd_qiov); 2297 qemu_iovec_add(&hd_qiov, cluster_data, cur_bytes); 2298 } 2299 2300 /* Try to efficiently initialize the physical space with zeroes */ 2301 ret = handle_alloc_space(bs, l2meta); 2302 if (ret < 0) { 2303 goto out_unlocked; 2304 } 2305 2306 /* If we need to do COW, check if it's possible to merge the 2307 * writing of the guest data together with that of the COW regions. 2308 * If it's not possible (or not necessary) then write the 2309 * guest data now. */ 2310 if (!merge_cow(offset, cur_bytes, &hd_qiov, l2meta)) { 2311 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO); 2312 trace_qcow2_writev_data(qemu_coroutine_self(), 2313 cluster_offset + offset_in_cluster); 2314 ret = bdrv_co_pwritev(s->data_file, 2315 cluster_offset + offset_in_cluster, 2316 cur_bytes, &hd_qiov, 0); 2317 if (ret < 0) { 2318 goto out_unlocked; 2319 } 2320 } 2321 2322 qemu_co_mutex_lock(&s->lock); 2323 2324 ret = qcow2_handle_l2meta(bs, &l2meta, true); 2325 if (ret) { 2326 goto out_locked; 2327 } 2328 2329 bytes -= cur_bytes; 2330 offset += cur_bytes; 2331 bytes_done += cur_bytes; 2332 trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_bytes); 2333 } 2334 ret = 0; 2335 goto out_locked; 2336 2337 out_unlocked: 2338 qemu_co_mutex_lock(&s->lock); 2339 2340 out_locked: 2341 qcow2_handle_l2meta(bs, &l2meta, false); 2342 2343 qemu_co_mutex_unlock(&s->lock); 2344 2345 qemu_iovec_destroy(&hd_qiov); 2346 qemu_vfree(cluster_data); 2347 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret); 2348 2349 return ret; 2350 } 2351 2352 static int qcow2_inactivate(BlockDriverState *bs) 2353 { 2354 BDRVQcow2State *s = bs->opaque; 2355 int ret, result = 0; 2356 Error *local_err = NULL; 2357 2358 qcow2_store_persistent_dirty_bitmaps(bs, &local_err); 2359 if (local_err != NULL) { 2360 result = -EINVAL; 2361 error_reportf_err(local_err, "Lost persistent bitmaps during " 2362 "inactivation of node '%s': ", 2363 bdrv_get_device_or_node_name(bs)); 2364 } 2365 2366 ret = qcow2_cache_flush(bs, s->l2_table_cache); 2367 if (ret) { 2368 result = ret; 2369 error_report("Failed to flush the L2 table cache: %s", 2370 strerror(-ret)); 2371 } 2372 2373 ret = qcow2_cache_flush(bs, s->refcount_block_cache); 2374 if (ret) { 2375 result = ret; 2376 error_report("Failed to flush the refcount block cache: %s", 2377 strerror(-ret)); 2378 } 2379 2380 if (result == 0) { 2381 qcow2_mark_clean(bs); 2382 } 2383 2384 return result; 2385 } 2386 2387 static void qcow2_close(BlockDriverState *bs) 2388 { 2389 BDRVQcow2State *s = bs->opaque; 2390 qemu_vfree(s->l1_table); 2391 /* else pre-write overlap checks in cache_destroy may crash */ 2392 s->l1_table = NULL; 2393 2394 if (!(s->flags & BDRV_O_INACTIVE)) { 2395 qcow2_inactivate(bs); 2396 } 2397 2398 cache_clean_timer_del(bs); 2399 qcow2_cache_destroy(s->l2_table_cache); 2400 qcow2_cache_destroy(s->refcount_block_cache); 2401 2402 qcrypto_block_free(s->crypto); 2403 s->crypto = NULL; 2404 2405 g_free(s->unknown_header_fields); 2406 cleanup_unknown_header_ext(bs); 2407 2408 g_free(s->image_data_file); 2409 g_free(s->image_backing_file); 2410 g_free(s->image_backing_format); 2411 2412 if (has_data_file(bs)) { 2413 bdrv_unref_child(bs, s->data_file); 2414 } 2415 2416 qcow2_refcount_close(bs); 2417 qcow2_free_snapshots(bs); 2418 } 2419 2420 static void coroutine_fn qcow2_co_invalidate_cache(BlockDriverState *bs, 2421 Error **errp) 2422 { 2423 BDRVQcow2State *s = bs->opaque; 2424 int flags = s->flags; 2425 QCryptoBlock *crypto = NULL; 2426 QDict *options; 2427 Error *local_err = NULL; 2428 int ret; 2429 2430 /* 2431 * Backing files are read-only which makes all of their metadata immutable, 2432 * that means we don't have to worry about reopening them here. 2433 */ 2434 2435 crypto = s->crypto; 2436 s->crypto = NULL; 2437 2438 qcow2_close(bs); 2439 2440 memset(s, 0, sizeof(BDRVQcow2State)); 2441 options = qdict_clone_shallow(bs->options); 2442 2443 flags &= ~BDRV_O_INACTIVE; 2444 qemu_co_mutex_lock(&s->lock); 2445 ret = qcow2_do_open(bs, options, flags, &local_err); 2446 qemu_co_mutex_unlock(&s->lock); 2447 qobject_unref(options); 2448 if (local_err) { 2449 error_propagate_prepend(errp, local_err, 2450 "Could not reopen qcow2 layer: "); 2451 bs->drv = NULL; 2452 return; 2453 } else if (ret < 0) { 2454 error_setg_errno(errp, -ret, "Could not reopen qcow2 layer"); 2455 bs->drv = NULL; 2456 return; 2457 } 2458 2459 s->crypto = crypto; 2460 } 2461 2462 static size_t header_ext_add(char *buf, uint32_t magic, const void *s, 2463 size_t len, size_t buflen) 2464 { 2465 QCowExtension *ext_backing_fmt = (QCowExtension*) buf; 2466 size_t ext_len = sizeof(QCowExtension) + ((len + 7) & ~7); 2467 2468 if (buflen < ext_len) { 2469 return -ENOSPC; 2470 } 2471 2472 *ext_backing_fmt = (QCowExtension) { 2473 .magic = cpu_to_be32(magic), 2474 .len = cpu_to_be32(len), 2475 }; 2476 2477 if (len) { 2478 memcpy(buf + sizeof(QCowExtension), s, len); 2479 } 2480 2481 return ext_len; 2482 } 2483 2484 /* 2485 * Updates the qcow2 header, including the variable length parts of it, i.e. 2486 * the backing file name and all extensions. qcow2 was not designed to allow 2487 * such changes, so if we run out of space (we can only use the first cluster) 2488 * this function may fail. 2489 * 2490 * Returns 0 on success, -errno in error cases. 2491 */ 2492 int qcow2_update_header(BlockDriverState *bs) 2493 { 2494 BDRVQcow2State *s = bs->opaque; 2495 QCowHeader *header; 2496 char *buf; 2497 size_t buflen = s->cluster_size; 2498 int ret; 2499 uint64_t total_size; 2500 uint32_t refcount_table_clusters; 2501 size_t header_length; 2502 Qcow2UnknownHeaderExtension *uext; 2503 2504 buf = qemu_blockalign(bs, buflen); 2505 2506 /* Header structure */ 2507 header = (QCowHeader*) buf; 2508 2509 if (buflen < sizeof(*header)) { 2510 ret = -ENOSPC; 2511 goto fail; 2512 } 2513 2514 header_length = sizeof(*header) + s->unknown_header_fields_size; 2515 total_size = bs->total_sectors * BDRV_SECTOR_SIZE; 2516 refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3); 2517 2518 *header = (QCowHeader) { 2519 /* Version 2 fields */ 2520 .magic = cpu_to_be32(QCOW_MAGIC), 2521 .version = cpu_to_be32(s->qcow_version), 2522 .backing_file_offset = 0, 2523 .backing_file_size = 0, 2524 .cluster_bits = cpu_to_be32(s->cluster_bits), 2525 .size = cpu_to_be64(total_size), 2526 .crypt_method = cpu_to_be32(s->crypt_method_header), 2527 .l1_size = cpu_to_be32(s->l1_size), 2528 .l1_table_offset = cpu_to_be64(s->l1_table_offset), 2529 .refcount_table_offset = cpu_to_be64(s->refcount_table_offset), 2530 .refcount_table_clusters = cpu_to_be32(refcount_table_clusters), 2531 .nb_snapshots = cpu_to_be32(s->nb_snapshots), 2532 .snapshots_offset = cpu_to_be64(s->snapshots_offset), 2533 2534 /* Version 3 fields */ 2535 .incompatible_features = cpu_to_be64(s->incompatible_features), 2536 .compatible_features = cpu_to_be64(s->compatible_features), 2537 .autoclear_features = cpu_to_be64(s->autoclear_features), 2538 .refcount_order = cpu_to_be32(s->refcount_order), 2539 .header_length = cpu_to_be32(header_length), 2540 }; 2541 2542 /* For older versions, write a shorter header */ 2543 switch (s->qcow_version) { 2544 case 2: 2545 ret = offsetof(QCowHeader, incompatible_features); 2546 break; 2547 case 3: 2548 ret = sizeof(*header); 2549 break; 2550 default: 2551 ret = -EINVAL; 2552 goto fail; 2553 } 2554 2555 buf += ret; 2556 buflen -= ret; 2557 memset(buf, 0, buflen); 2558 2559 /* Preserve any unknown field in the header */ 2560 if (s->unknown_header_fields_size) { 2561 if (buflen < s->unknown_header_fields_size) { 2562 ret = -ENOSPC; 2563 goto fail; 2564 } 2565 2566 memcpy(buf, s->unknown_header_fields, s->unknown_header_fields_size); 2567 buf += s->unknown_header_fields_size; 2568 buflen -= s->unknown_header_fields_size; 2569 } 2570 2571 /* Backing file format header extension */ 2572 if (s->image_backing_format) { 2573 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BACKING_FORMAT, 2574 s->image_backing_format, 2575 strlen(s->image_backing_format), 2576 buflen); 2577 if (ret < 0) { 2578 goto fail; 2579 } 2580 2581 buf += ret; 2582 buflen -= ret; 2583 } 2584 2585 /* External data file header extension */ 2586 if (has_data_file(bs) && s->image_data_file) { 2587 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_DATA_FILE, 2588 s->image_data_file, strlen(s->image_data_file), 2589 buflen); 2590 if (ret < 0) { 2591 goto fail; 2592 } 2593 2594 buf += ret; 2595 buflen -= ret; 2596 } 2597 2598 /* Full disk encryption header pointer extension */ 2599 if (s->crypto_header.offset != 0) { 2600 s->crypto_header.offset = cpu_to_be64(s->crypto_header.offset); 2601 s->crypto_header.length = cpu_to_be64(s->crypto_header.length); 2602 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_CRYPTO_HEADER, 2603 &s->crypto_header, sizeof(s->crypto_header), 2604 buflen); 2605 s->crypto_header.offset = be64_to_cpu(s->crypto_header.offset); 2606 s->crypto_header.length = be64_to_cpu(s->crypto_header.length); 2607 if (ret < 0) { 2608 goto fail; 2609 } 2610 buf += ret; 2611 buflen -= ret; 2612 } 2613 2614 /* Feature table */ 2615 if (s->qcow_version >= 3) { 2616 Qcow2Feature features[] = { 2617 { 2618 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE, 2619 .bit = QCOW2_INCOMPAT_DIRTY_BITNR, 2620 .name = "dirty bit", 2621 }, 2622 { 2623 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE, 2624 .bit = QCOW2_INCOMPAT_CORRUPT_BITNR, 2625 .name = "corrupt bit", 2626 }, 2627 { 2628 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE, 2629 .bit = QCOW2_INCOMPAT_DATA_FILE_BITNR, 2630 .name = "external data file", 2631 }, 2632 { 2633 .type = QCOW2_FEAT_TYPE_COMPATIBLE, 2634 .bit = QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR, 2635 .name = "lazy refcounts", 2636 }, 2637 }; 2638 2639 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_FEATURE_TABLE, 2640 features, sizeof(features), buflen); 2641 if (ret < 0) { 2642 goto fail; 2643 } 2644 buf += ret; 2645 buflen -= ret; 2646 } 2647 2648 /* Bitmap extension */ 2649 if (s->nb_bitmaps > 0) { 2650 Qcow2BitmapHeaderExt bitmaps_header = { 2651 .nb_bitmaps = cpu_to_be32(s->nb_bitmaps), 2652 .bitmap_directory_size = 2653 cpu_to_be64(s->bitmap_directory_size), 2654 .bitmap_directory_offset = 2655 cpu_to_be64(s->bitmap_directory_offset) 2656 }; 2657 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BITMAPS, 2658 &bitmaps_header, sizeof(bitmaps_header), 2659 buflen); 2660 if (ret < 0) { 2661 goto fail; 2662 } 2663 buf += ret; 2664 buflen -= ret; 2665 } 2666 2667 /* Keep unknown header extensions */ 2668 QLIST_FOREACH(uext, &s->unknown_header_ext, next) { 2669 ret = header_ext_add(buf, uext->magic, uext->data, uext->len, buflen); 2670 if (ret < 0) { 2671 goto fail; 2672 } 2673 2674 buf += ret; 2675 buflen -= ret; 2676 } 2677 2678 /* End of header extensions */ 2679 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_END, NULL, 0, buflen); 2680 if (ret < 0) { 2681 goto fail; 2682 } 2683 2684 buf += ret; 2685 buflen -= ret; 2686 2687 /* Backing file name */ 2688 if (s->image_backing_file) { 2689 size_t backing_file_len = strlen(s->image_backing_file); 2690 2691 if (buflen < backing_file_len) { 2692 ret = -ENOSPC; 2693 goto fail; 2694 } 2695 2696 /* Using strncpy is ok here, since buf is not NUL-terminated. */ 2697 strncpy(buf, s->image_backing_file, buflen); 2698 2699 header->backing_file_offset = cpu_to_be64(buf - ((char*) header)); 2700 header->backing_file_size = cpu_to_be32(backing_file_len); 2701 } 2702 2703 /* Write the new header */ 2704 ret = bdrv_pwrite(bs->file, 0, header, s->cluster_size); 2705 if (ret < 0) { 2706 goto fail; 2707 } 2708 2709 ret = 0; 2710 fail: 2711 qemu_vfree(header); 2712 return ret; 2713 } 2714 2715 static int qcow2_change_backing_file(BlockDriverState *bs, 2716 const char *backing_file, const char *backing_fmt) 2717 { 2718 BDRVQcow2State *s = bs->opaque; 2719 2720 /* Adding a backing file means that the external data file alone won't be 2721 * enough to make sense of the content */ 2722 if (backing_file && data_file_is_raw(bs)) { 2723 return -EINVAL; 2724 } 2725 2726 if (backing_file && strlen(backing_file) > 1023) { 2727 return -EINVAL; 2728 } 2729 2730 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file), 2731 backing_file ?: ""); 2732 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: ""); 2733 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: ""); 2734 2735 g_free(s->image_backing_file); 2736 g_free(s->image_backing_format); 2737 2738 s->image_backing_file = backing_file ? g_strdup(bs->backing_file) : NULL; 2739 s->image_backing_format = backing_fmt ? g_strdup(bs->backing_format) : NULL; 2740 2741 return qcow2_update_header(bs); 2742 } 2743 2744 static int qcow2_crypt_method_from_format(const char *encryptfmt) 2745 { 2746 if (g_str_equal(encryptfmt, "luks")) { 2747 return QCOW_CRYPT_LUKS; 2748 } else if (g_str_equal(encryptfmt, "aes")) { 2749 return QCOW_CRYPT_AES; 2750 } else { 2751 return -EINVAL; 2752 } 2753 } 2754 2755 static int qcow2_set_up_encryption(BlockDriverState *bs, 2756 QCryptoBlockCreateOptions *cryptoopts, 2757 Error **errp) 2758 { 2759 BDRVQcow2State *s = bs->opaque; 2760 QCryptoBlock *crypto = NULL; 2761 int fmt, ret; 2762 2763 switch (cryptoopts->format) { 2764 case Q_CRYPTO_BLOCK_FORMAT_LUKS: 2765 fmt = QCOW_CRYPT_LUKS; 2766 break; 2767 case Q_CRYPTO_BLOCK_FORMAT_QCOW: 2768 fmt = QCOW_CRYPT_AES; 2769 break; 2770 default: 2771 error_setg(errp, "Crypto format not supported in qcow2"); 2772 return -EINVAL; 2773 } 2774 2775 s->crypt_method_header = fmt; 2776 2777 crypto = qcrypto_block_create(cryptoopts, "encrypt.", 2778 qcow2_crypto_hdr_init_func, 2779 qcow2_crypto_hdr_write_func, 2780 bs, errp); 2781 if (!crypto) { 2782 return -EINVAL; 2783 } 2784 2785 ret = qcow2_update_header(bs); 2786 if (ret < 0) { 2787 error_setg_errno(errp, -ret, "Could not write encryption header"); 2788 goto out; 2789 } 2790 2791 ret = 0; 2792 out: 2793 qcrypto_block_free(crypto); 2794 return ret; 2795 } 2796 2797 /** 2798 * Preallocates metadata structures for data clusters between @offset (in the 2799 * guest disk) and @new_length (which is thus generally the new guest disk 2800 * size). 2801 * 2802 * Returns: 0 on success, -errno on failure. 2803 */ 2804 static int coroutine_fn preallocate_co(BlockDriverState *bs, uint64_t offset, 2805 uint64_t new_length, PreallocMode mode, 2806 Error **errp) 2807 { 2808 BDRVQcow2State *s = bs->opaque; 2809 uint64_t bytes; 2810 uint64_t host_offset = 0; 2811 int64_t file_length; 2812 unsigned int cur_bytes; 2813 int ret; 2814 QCowL2Meta *meta; 2815 2816 assert(offset <= new_length); 2817 bytes = new_length - offset; 2818 2819 while (bytes) { 2820 cur_bytes = MIN(bytes, QEMU_ALIGN_DOWN(INT_MAX, s->cluster_size)); 2821 ret = qcow2_alloc_cluster_offset(bs, offset, &cur_bytes, 2822 &host_offset, &meta); 2823 if (ret < 0) { 2824 error_setg_errno(errp, -ret, "Allocating clusters failed"); 2825 return ret; 2826 } 2827 2828 while (meta) { 2829 QCowL2Meta *next = meta->next; 2830 2831 ret = qcow2_alloc_cluster_link_l2(bs, meta); 2832 if (ret < 0) { 2833 error_setg_errno(errp, -ret, "Mapping clusters failed"); 2834 qcow2_free_any_clusters(bs, meta->alloc_offset, 2835 meta->nb_clusters, QCOW2_DISCARD_NEVER); 2836 return ret; 2837 } 2838 2839 /* There are no dependent requests, but we need to remove our 2840 * request from the list of in-flight requests */ 2841 QLIST_REMOVE(meta, next_in_flight); 2842 2843 g_free(meta); 2844 meta = next; 2845 } 2846 2847 /* TODO Preallocate data if requested */ 2848 2849 bytes -= cur_bytes; 2850 offset += cur_bytes; 2851 } 2852 2853 /* 2854 * It is expected that the image file is large enough to actually contain 2855 * all of the allocated clusters (otherwise we get failing reads after 2856 * EOF). Extend the image to the last allocated sector. 2857 */ 2858 file_length = bdrv_getlength(s->data_file->bs); 2859 if (file_length < 0) { 2860 error_setg_errno(errp, -file_length, "Could not get file size"); 2861 return file_length; 2862 } 2863 2864 if (host_offset + cur_bytes > file_length) { 2865 if (mode == PREALLOC_MODE_METADATA) { 2866 mode = PREALLOC_MODE_OFF; 2867 } 2868 ret = bdrv_co_truncate(s->data_file, host_offset + cur_bytes, mode, 2869 errp); 2870 if (ret < 0) { 2871 return ret; 2872 } 2873 } 2874 2875 return 0; 2876 } 2877 2878 /* qcow2_refcount_metadata_size: 2879 * @clusters: number of clusters to refcount (including data and L1/L2 tables) 2880 * @cluster_size: size of a cluster, in bytes 2881 * @refcount_order: refcount bits power-of-2 exponent 2882 * @generous_increase: allow for the refcount table to be 1.5x as large as it 2883 * needs to be 2884 * 2885 * Returns: Number of bytes required for refcount blocks and table metadata. 2886 */ 2887 int64_t qcow2_refcount_metadata_size(int64_t clusters, size_t cluster_size, 2888 int refcount_order, bool generous_increase, 2889 uint64_t *refblock_count) 2890 { 2891 /* 2892 * Every host cluster is reference-counted, including metadata (even 2893 * refcount metadata is recursively included). 2894 * 2895 * An accurate formula for the size of refcount metadata size is difficult 2896 * to derive. An easier method of calculation is finding the fixed point 2897 * where no further refcount blocks or table clusters are required to 2898 * reference count every cluster. 2899 */ 2900 int64_t blocks_per_table_cluster = cluster_size / sizeof(uint64_t); 2901 int64_t refcounts_per_block = cluster_size * 8 / (1 << refcount_order); 2902 int64_t table = 0; /* number of refcount table clusters */ 2903 int64_t blocks = 0; /* number of refcount block clusters */ 2904 int64_t last; 2905 int64_t n = 0; 2906 2907 do { 2908 last = n; 2909 blocks = DIV_ROUND_UP(clusters + table + blocks, refcounts_per_block); 2910 table = DIV_ROUND_UP(blocks, blocks_per_table_cluster); 2911 n = clusters + blocks + table; 2912 2913 if (n == last && generous_increase) { 2914 clusters += DIV_ROUND_UP(table, 2); 2915 n = 0; /* force another loop */ 2916 generous_increase = false; 2917 } 2918 } while (n != last); 2919 2920 if (refblock_count) { 2921 *refblock_count = blocks; 2922 } 2923 2924 return (blocks + table) * cluster_size; 2925 } 2926 2927 /** 2928 * qcow2_calc_prealloc_size: 2929 * @total_size: virtual disk size in bytes 2930 * @cluster_size: cluster size in bytes 2931 * @refcount_order: refcount bits power-of-2 exponent 2932 * 2933 * Returns: Total number of bytes required for the fully allocated image 2934 * (including metadata). 2935 */ 2936 static int64_t qcow2_calc_prealloc_size(int64_t total_size, 2937 size_t cluster_size, 2938 int refcount_order) 2939 { 2940 int64_t meta_size = 0; 2941 uint64_t nl1e, nl2e; 2942 int64_t aligned_total_size = ROUND_UP(total_size, cluster_size); 2943 2944 /* header: 1 cluster */ 2945 meta_size += cluster_size; 2946 2947 /* total size of L2 tables */ 2948 nl2e = aligned_total_size / cluster_size; 2949 nl2e = ROUND_UP(nl2e, cluster_size / sizeof(uint64_t)); 2950 meta_size += nl2e * sizeof(uint64_t); 2951 2952 /* total size of L1 tables */ 2953 nl1e = nl2e * sizeof(uint64_t) / cluster_size; 2954 nl1e = ROUND_UP(nl1e, cluster_size / sizeof(uint64_t)); 2955 meta_size += nl1e * sizeof(uint64_t); 2956 2957 /* total size of refcount table and blocks */ 2958 meta_size += qcow2_refcount_metadata_size( 2959 (meta_size + aligned_total_size) / cluster_size, 2960 cluster_size, refcount_order, false, NULL); 2961 2962 return meta_size + aligned_total_size; 2963 } 2964 2965 static bool validate_cluster_size(size_t cluster_size, Error **errp) 2966 { 2967 int cluster_bits = ctz32(cluster_size); 2968 if (cluster_bits < MIN_CLUSTER_BITS || cluster_bits > MAX_CLUSTER_BITS || 2969 (1 << cluster_bits) != cluster_size) 2970 { 2971 error_setg(errp, "Cluster size must be a power of two between %d and " 2972 "%dk", 1 << MIN_CLUSTER_BITS, 1 << (MAX_CLUSTER_BITS - 10)); 2973 return false; 2974 } 2975 return true; 2976 } 2977 2978 static size_t qcow2_opt_get_cluster_size_del(QemuOpts *opts, Error **errp) 2979 { 2980 size_t cluster_size; 2981 2982 cluster_size = qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE, 2983 DEFAULT_CLUSTER_SIZE); 2984 if (!validate_cluster_size(cluster_size, errp)) { 2985 return 0; 2986 } 2987 return cluster_size; 2988 } 2989 2990 static int qcow2_opt_get_version_del(QemuOpts *opts, Error **errp) 2991 { 2992 char *buf; 2993 int ret; 2994 2995 buf = qemu_opt_get_del(opts, BLOCK_OPT_COMPAT_LEVEL); 2996 if (!buf) { 2997 ret = 3; /* default */ 2998 } else if (!strcmp(buf, "0.10")) { 2999 ret = 2; 3000 } else if (!strcmp(buf, "1.1")) { 3001 ret = 3; 3002 } else { 3003 error_setg(errp, "Invalid compatibility level: '%s'", buf); 3004 ret = -EINVAL; 3005 } 3006 g_free(buf); 3007 return ret; 3008 } 3009 3010 static uint64_t qcow2_opt_get_refcount_bits_del(QemuOpts *opts, int version, 3011 Error **errp) 3012 { 3013 uint64_t refcount_bits; 3014 3015 refcount_bits = qemu_opt_get_number_del(opts, BLOCK_OPT_REFCOUNT_BITS, 16); 3016 if (refcount_bits > 64 || !is_power_of_2(refcount_bits)) { 3017 error_setg(errp, "Refcount width must be a power of two and may not " 3018 "exceed 64 bits"); 3019 return 0; 3020 } 3021 3022 if (version < 3 && refcount_bits != 16) { 3023 error_setg(errp, "Different refcount widths than 16 bits require " 3024 "compatibility level 1.1 or above (use compat=1.1 or " 3025 "greater)"); 3026 return 0; 3027 } 3028 3029 return refcount_bits; 3030 } 3031 3032 static int coroutine_fn 3033 qcow2_co_create(BlockdevCreateOptions *create_options, Error **errp) 3034 { 3035 BlockdevCreateOptionsQcow2 *qcow2_opts; 3036 QDict *options; 3037 3038 /* 3039 * Open the image file and write a minimal qcow2 header. 3040 * 3041 * We keep things simple and start with a zero-sized image. We also 3042 * do without refcount blocks or a L1 table for now. We'll fix the 3043 * inconsistency later. 3044 * 3045 * We do need a refcount table because growing the refcount table means 3046 * allocating two new refcount blocks - the seconds of which would be at 3047 * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file 3048 * size for any qcow2 image. 3049 */ 3050 BlockBackend *blk = NULL; 3051 BlockDriverState *bs = NULL; 3052 BlockDriverState *data_bs = NULL; 3053 QCowHeader *header; 3054 size_t cluster_size; 3055 int version; 3056 int refcount_order; 3057 uint64_t* refcount_table; 3058 Error *local_err = NULL; 3059 int ret; 3060 3061 assert(create_options->driver == BLOCKDEV_DRIVER_QCOW2); 3062 qcow2_opts = &create_options->u.qcow2; 3063 3064 bs = bdrv_open_blockdev_ref(qcow2_opts->file, errp); 3065 if (bs == NULL) { 3066 return -EIO; 3067 } 3068 3069 /* Validate options and set default values */ 3070 if (!QEMU_IS_ALIGNED(qcow2_opts->size, BDRV_SECTOR_SIZE)) { 3071 error_setg(errp, "Image size must be a multiple of 512 bytes"); 3072 ret = -EINVAL; 3073 goto out; 3074 } 3075 3076 if (qcow2_opts->has_version) { 3077 switch (qcow2_opts->version) { 3078 case BLOCKDEV_QCOW2_VERSION_V2: 3079 version = 2; 3080 break; 3081 case BLOCKDEV_QCOW2_VERSION_V3: 3082 version = 3; 3083 break; 3084 default: 3085 g_assert_not_reached(); 3086 } 3087 } else { 3088 version = 3; 3089 } 3090 3091 if (qcow2_opts->has_cluster_size) { 3092 cluster_size = qcow2_opts->cluster_size; 3093 } else { 3094 cluster_size = DEFAULT_CLUSTER_SIZE; 3095 } 3096 3097 if (!validate_cluster_size(cluster_size, errp)) { 3098 ret = -EINVAL; 3099 goto out; 3100 } 3101 3102 if (!qcow2_opts->has_preallocation) { 3103 qcow2_opts->preallocation = PREALLOC_MODE_OFF; 3104 } 3105 if (qcow2_opts->has_backing_file && 3106 qcow2_opts->preallocation != PREALLOC_MODE_OFF) 3107 { 3108 error_setg(errp, "Backing file and preallocation cannot be used at " 3109 "the same time"); 3110 ret = -EINVAL; 3111 goto out; 3112 } 3113 if (qcow2_opts->has_backing_fmt && !qcow2_opts->has_backing_file) { 3114 error_setg(errp, "Backing format cannot be used without backing file"); 3115 ret = -EINVAL; 3116 goto out; 3117 } 3118 3119 if (!qcow2_opts->has_lazy_refcounts) { 3120 qcow2_opts->lazy_refcounts = false; 3121 } 3122 if (version < 3 && qcow2_opts->lazy_refcounts) { 3123 error_setg(errp, "Lazy refcounts only supported with compatibility " 3124 "level 1.1 and above (use version=v3 or greater)"); 3125 ret = -EINVAL; 3126 goto out; 3127 } 3128 3129 if (!qcow2_opts->has_refcount_bits) { 3130 qcow2_opts->refcount_bits = 16; 3131 } 3132 if (qcow2_opts->refcount_bits > 64 || 3133 !is_power_of_2(qcow2_opts->refcount_bits)) 3134 { 3135 error_setg(errp, "Refcount width must be a power of two and may not " 3136 "exceed 64 bits"); 3137 ret = -EINVAL; 3138 goto out; 3139 } 3140 if (version < 3 && qcow2_opts->refcount_bits != 16) { 3141 error_setg(errp, "Different refcount widths than 16 bits require " 3142 "compatibility level 1.1 or above (use version=v3 or " 3143 "greater)"); 3144 ret = -EINVAL; 3145 goto out; 3146 } 3147 refcount_order = ctz32(qcow2_opts->refcount_bits); 3148 3149 if (qcow2_opts->data_file_raw && !qcow2_opts->data_file) { 3150 error_setg(errp, "data-file-raw requires data-file"); 3151 ret = -EINVAL; 3152 goto out; 3153 } 3154 if (qcow2_opts->data_file_raw && qcow2_opts->has_backing_file) { 3155 error_setg(errp, "Backing file and data-file-raw cannot be used at " 3156 "the same time"); 3157 ret = -EINVAL; 3158 goto out; 3159 } 3160 3161 if (qcow2_opts->data_file) { 3162 if (version < 3) { 3163 error_setg(errp, "External data files are only supported with " 3164 "compatibility level 1.1 and above (use version=v3 or " 3165 "greater)"); 3166 ret = -EINVAL; 3167 goto out; 3168 } 3169 data_bs = bdrv_open_blockdev_ref(qcow2_opts->data_file, errp); 3170 if (data_bs == NULL) { 3171 ret = -EIO; 3172 goto out; 3173 } 3174 } 3175 3176 /* Create BlockBackend to write to the image */ 3177 blk = blk_new(bdrv_get_aio_context(bs), 3178 BLK_PERM_WRITE | BLK_PERM_RESIZE, BLK_PERM_ALL); 3179 ret = blk_insert_bs(blk, bs, errp); 3180 if (ret < 0) { 3181 goto out; 3182 } 3183 blk_set_allow_write_beyond_eof(blk, true); 3184 3185 /* Clear the protocol layer and preallocate it if necessary */ 3186 ret = blk_truncate(blk, 0, PREALLOC_MODE_OFF, errp); 3187 if (ret < 0) { 3188 goto out; 3189 } 3190 3191 /* Write the header */ 3192 QEMU_BUILD_BUG_ON((1 << MIN_CLUSTER_BITS) < sizeof(*header)); 3193 header = g_malloc0(cluster_size); 3194 *header = (QCowHeader) { 3195 .magic = cpu_to_be32(QCOW_MAGIC), 3196 .version = cpu_to_be32(version), 3197 .cluster_bits = cpu_to_be32(ctz32(cluster_size)), 3198 .size = cpu_to_be64(0), 3199 .l1_table_offset = cpu_to_be64(0), 3200 .l1_size = cpu_to_be32(0), 3201 .refcount_table_offset = cpu_to_be64(cluster_size), 3202 .refcount_table_clusters = cpu_to_be32(1), 3203 .refcount_order = cpu_to_be32(refcount_order), 3204 .header_length = cpu_to_be32(sizeof(*header)), 3205 }; 3206 3207 /* We'll update this to correct value later */ 3208 header->crypt_method = cpu_to_be32(QCOW_CRYPT_NONE); 3209 3210 if (qcow2_opts->lazy_refcounts) { 3211 header->compatible_features |= 3212 cpu_to_be64(QCOW2_COMPAT_LAZY_REFCOUNTS); 3213 } 3214 if (data_bs) { 3215 header->incompatible_features |= 3216 cpu_to_be64(QCOW2_INCOMPAT_DATA_FILE); 3217 } 3218 if (qcow2_opts->data_file_raw) { 3219 header->autoclear_features |= 3220 cpu_to_be64(QCOW2_AUTOCLEAR_DATA_FILE_RAW); 3221 } 3222 3223 ret = blk_pwrite(blk, 0, header, cluster_size, 0); 3224 g_free(header); 3225 if (ret < 0) { 3226 error_setg_errno(errp, -ret, "Could not write qcow2 header"); 3227 goto out; 3228 } 3229 3230 /* Write a refcount table with one refcount block */ 3231 refcount_table = g_malloc0(2 * cluster_size); 3232 refcount_table[0] = cpu_to_be64(2 * cluster_size); 3233 ret = blk_pwrite(blk, cluster_size, refcount_table, 2 * cluster_size, 0); 3234 g_free(refcount_table); 3235 3236 if (ret < 0) { 3237 error_setg_errno(errp, -ret, "Could not write refcount table"); 3238 goto out; 3239 } 3240 3241 blk_unref(blk); 3242 blk = NULL; 3243 3244 /* 3245 * And now open the image and make it consistent first (i.e. increase the 3246 * refcount of the cluster that is occupied by the header and the refcount 3247 * table) 3248 */ 3249 options = qdict_new(); 3250 qdict_put_str(options, "driver", "qcow2"); 3251 qdict_put_str(options, "file", bs->node_name); 3252 if (data_bs) { 3253 qdict_put_str(options, "data-file", data_bs->node_name); 3254 } 3255 blk = blk_new_open(NULL, NULL, options, 3256 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_NO_FLUSH, 3257 &local_err); 3258 if (blk == NULL) { 3259 error_propagate(errp, local_err); 3260 ret = -EIO; 3261 goto out; 3262 } 3263 3264 ret = qcow2_alloc_clusters(blk_bs(blk), 3 * cluster_size); 3265 if (ret < 0) { 3266 error_setg_errno(errp, -ret, "Could not allocate clusters for qcow2 " 3267 "header and refcount table"); 3268 goto out; 3269 3270 } else if (ret != 0) { 3271 error_report("Huh, first cluster in empty image is already in use?"); 3272 abort(); 3273 } 3274 3275 /* Set the external data file if necessary */ 3276 if (data_bs) { 3277 BDRVQcow2State *s = blk_bs(blk)->opaque; 3278 s->image_data_file = g_strdup(data_bs->filename); 3279 } 3280 3281 /* Create a full header (including things like feature table) */ 3282 ret = qcow2_update_header(blk_bs(blk)); 3283 if (ret < 0) { 3284 error_setg_errno(errp, -ret, "Could not update qcow2 header"); 3285 goto out; 3286 } 3287 3288 /* Okay, now that we have a valid image, let's give it the right size */ 3289 ret = blk_truncate(blk, qcow2_opts->size, qcow2_opts->preallocation, errp); 3290 if (ret < 0) { 3291 error_prepend(errp, "Could not resize image: "); 3292 goto out; 3293 } 3294 3295 /* Want a backing file? There you go.*/ 3296 if (qcow2_opts->has_backing_file) { 3297 const char *backing_format = NULL; 3298 3299 if (qcow2_opts->has_backing_fmt) { 3300 backing_format = BlockdevDriver_str(qcow2_opts->backing_fmt); 3301 } 3302 3303 ret = bdrv_change_backing_file(blk_bs(blk), qcow2_opts->backing_file, 3304 backing_format); 3305 if (ret < 0) { 3306 error_setg_errno(errp, -ret, "Could not assign backing file '%s' " 3307 "with format '%s'", qcow2_opts->backing_file, 3308 backing_format); 3309 goto out; 3310 } 3311 } 3312 3313 /* Want encryption? There you go. */ 3314 if (qcow2_opts->has_encrypt) { 3315 ret = qcow2_set_up_encryption(blk_bs(blk), qcow2_opts->encrypt, errp); 3316 if (ret < 0) { 3317 goto out; 3318 } 3319 } 3320 3321 blk_unref(blk); 3322 blk = NULL; 3323 3324 /* Reopen the image without BDRV_O_NO_FLUSH to flush it before returning. 3325 * Using BDRV_O_NO_IO, since encryption is now setup we don't want to 3326 * have to setup decryption context. We're not doing any I/O on the top 3327 * level BlockDriverState, only lower layers, where BDRV_O_NO_IO does 3328 * not have effect. 3329 */ 3330 options = qdict_new(); 3331 qdict_put_str(options, "driver", "qcow2"); 3332 qdict_put_str(options, "file", bs->node_name); 3333 if (data_bs) { 3334 qdict_put_str(options, "data-file", data_bs->node_name); 3335 } 3336 blk = blk_new_open(NULL, NULL, options, 3337 BDRV_O_RDWR | BDRV_O_NO_BACKING | BDRV_O_NO_IO, 3338 &local_err); 3339 if (blk == NULL) { 3340 error_propagate(errp, local_err); 3341 ret = -EIO; 3342 goto out; 3343 } 3344 3345 ret = 0; 3346 out: 3347 blk_unref(blk); 3348 bdrv_unref(bs); 3349 bdrv_unref(data_bs); 3350 return ret; 3351 } 3352 3353 static int coroutine_fn qcow2_co_create_opts(const char *filename, QemuOpts *opts, 3354 Error **errp) 3355 { 3356 BlockdevCreateOptions *create_options = NULL; 3357 QDict *qdict; 3358 Visitor *v; 3359 BlockDriverState *bs = NULL; 3360 BlockDriverState *data_bs = NULL; 3361 Error *local_err = NULL; 3362 const char *val; 3363 int ret; 3364 3365 /* Only the keyval visitor supports the dotted syntax needed for 3366 * encryption, so go through a QDict before getting a QAPI type. Ignore 3367 * options meant for the protocol layer so that the visitor doesn't 3368 * complain. */ 3369 qdict = qemu_opts_to_qdict_filtered(opts, NULL, bdrv_qcow2.create_opts, 3370 true); 3371 3372 /* Handle encryption options */ 3373 val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT); 3374 if (val && !strcmp(val, "on")) { 3375 qdict_put_str(qdict, BLOCK_OPT_ENCRYPT, "qcow"); 3376 } else if (val && !strcmp(val, "off")) { 3377 qdict_del(qdict, BLOCK_OPT_ENCRYPT); 3378 } 3379 3380 val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT); 3381 if (val && !strcmp(val, "aes")) { 3382 qdict_put_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT, "qcow"); 3383 } 3384 3385 /* Convert compat=0.10/1.1 into compat=v2/v3, to be renamed into 3386 * version=v2/v3 below. */ 3387 val = qdict_get_try_str(qdict, BLOCK_OPT_COMPAT_LEVEL); 3388 if (val && !strcmp(val, "0.10")) { 3389 qdict_put_str(qdict, BLOCK_OPT_COMPAT_LEVEL, "v2"); 3390 } else if (val && !strcmp(val, "1.1")) { 3391 qdict_put_str(qdict, BLOCK_OPT_COMPAT_LEVEL, "v3"); 3392 } 3393 3394 /* Change legacy command line options into QMP ones */ 3395 static const QDictRenames opt_renames[] = { 3396 { BLOCK_OPT_BACKING_FILE, "backing-file" }, 3397 { BLOCK_OPT_BACKING_FMT, "backing-fmt" }, 3398 { BLOCK_OPT_CLUSTER_SIZE, "cluster-size" }, 3399 { BLOCK_OPT_LAZY_REFCOUNTS, "lazy-refcounts" }, 3400 { BLOCK_OPT_REFCOUNT_BITS, "refcount-bits" }, 3401 { BLOCK_OPT_ENCRYPT, BLOCK_OPT_ENCRYPT_FORMAT }, 3402 { BLOCK_OPT_COMPAT_LEVEL, "version" }, 3403 { BLOCK_OPT_DATA_FILE_RAW, "data-file-raw" }, 3404 { NULL, NULL }, 3405 }; 3406 3407 if (!qdict_rename_keys(qdict, opt_renames, errp)) { 3408 ret = -EINVAL; 3409 goto finish; 3410 } 3411 3412 /* Create and open the file (protocol layer) */ 3413 ret = bdrv_create_file(filename, opts, errp); 3414 if (ret < 0) { 3415 goto finish; 3416 } 3417 3418 bs = bdrv_open(filename, NULL, NULL, 3419 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp); 3420 if (bs == NULL) { 3421 ret = -EIO; 3422 goto finish; 3423 } 3424 3425 /* Create and open an external data file (protocol layer) */ 3426 val = qdict_get_try_str(qdict, BLOCK_OPT_DATA_FILE); 3427 if (val) { 3428 ret = bdrv_create_file(val, opts, errp); 3429 if (ret < 0) { 3430 goto finish; 3431 } 3432 3433 data_bs = bdrv_open(val, NULL, NULL, 3434 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, 3435 errp); 3436 if (data_bs == NULL) { 3437 ret = -EIO; 3438 goto finish; 3439 } 3440 3441 qdict_del(qdict, BLOCK_OPT_DATA_FILE); 3442 qdict_put_str(qdict, "data-file", data_bs->node_name); 3443 } 3444 3445 /* Set 'driver' and 'node' options */ 3446 qdict_put_str(qdict, "driver", "qcow2"); 3447 qdict_put_str(qdict, "file", bs->node_name); 3448 3449 /* Now get the QAPI type BlockdevCreateOptions */ 3450 v = qobject_input_visitor_new_flat_confused(qdict, errp); 3451 if (!v) { 3452 ret = -EINVAL; 3453 goto finish; 3454 } 3455 3456 visit_type_BlockdevCreateOptions(v, NULL, &create_options, &local_err); 3457 visit_free(v); 3458 3459 if (local_err) { 3460 error_propagate(errp, local_err); 3461 ret = -EINVAL; 3462 goto finish; 3463 } 3464 3465 /* Silently round up size */ 3466 create_options->u.qcow2.size = ROUND_UP(create_options->u.qcow2.size, 3467 BDRV_SECTOR_SIZE); 3468 3469 /* Create the qcow2 image (format layer) */ 3470 ret = qcow2_co_create(create_options, errp); 3471 if (ret < 0) { 3472 goto finish; 3473 } 3474 3475 ret = 0; 3476 finish: 3477 qobject_unref(qdict); 3478 bdrv_unref(bs); 3479 bdrv_unref(data_bs); 3480 qapi_free_BlockdevCreateOptions(create_options); 3481 return ret; 3482 } 3483 3484 3485 static bool is_zero(BlockDriverState *bs, int64_t offset, int64_t bytes) 3486 { 3487 int64_t nr; 3488 int res; 3489 3490 /* Clamp to image length, before checking status of underlying sectors */ 3491 if (offset + bytes > bs->total_sectors * BDRV_SECTOR_SIZE) { 3492 bytes = bs->total_sectors * BDRV_SECTOR_SIZE - offset; 3493 } 3494 3495 if (!bytes) { 3496 return true; 3497 } 3498 res = bdrv_block_status_above(bs, NULL, offset, bytes, &nr, NULL, NULL); 3499 return res >= 0 && (res & BDRV_BLOCK_ZERO) && nr == bytes; 3500 } 3501 3502 static coroutine_fn int qcow2_co_pwrite_zeroes(BlockDriverState *bs, 3503 int64_t offset, int bytes, BdrvRequestFlags flags) 3504 { 3505 int ret; 3506 BDRVQcow2State *s = bs->opaque; 3507 3508 uint32_t head = offset % s->cluster_size; 3509 uint32_t tail = (offset + bytes) % s->cluster_size; 3510 3511 trace_qcow2_pwrite_zeroes_start_req(qemu_coroutine_self(), offset, bytes); 3512 if (offset + bytes == bs->total_sectors * BDRV_SECTOR_SIZE) { 3513 tail = 0; 3514 } 3515 3516 if (head || tail) { 3517 uint64_t off; 3518 unsigned int nr; 3519 3520 assert(head + bytes <= s->cluster_size); 3521 3522 /* check whether remainder of cluster already reads as zero */ 3523 if (!(is_zero(bs, offset - head, head) && 3524 is_zero(bs, offset + bytes, 3525 tail ? s->cluster_size - tail : 0))) { 3526 return -ENOTSUP; 3527 } 3528 3529 qemu_co_mutex_lock(&s->lock); 3530 /* We can have new write after previous check */ 3531 offset = QEMU_ALIGN_DOWN(offset, s->cluster_size); 3532 bytes = s->cluster_size; 3533 nr = s->cluster_size; 3534 ret = qcow2_get_cluster_offset(bs, offset, &nr, &off); 3535 if (ret != QCOW2_CLUSTER_UNALLOCATED && 3536 ret != QCOW2_CLUSTER_ZERO_PLAIN && 3537 ret != QCOW2_CLUSTER_ZERO_ALLOC) { 3538 qemu_co_mutex_unlock(&s->lock); 3539 return -ENOTSUP; 3540 } 3541 } else { 3542 qemu_co_mutex_lock(&s->lock); 3543 } 3544 3545 trace_qcow2_pwrite_zeroes(qemu_coroutine_self(), offset, bytes); 3546 3547 /* Whatever is left can use real zero clusters */ 3548 ret = qcow2_cluster_zeroize(bs, offset, bytes, flags); 3549 qemu_co_mutex_unlock(&s->lock); 3550 3551 return ret; 3552 } 3553 3554 static coroutine_fn int qcow2_co_pdiscard(BlockDriverState *bs, 3555 int64_t offset, int bytes) 3556 { 3557 int ret; 3558 BDRVQcow2State *s = bs->opaque; 3559 3560 if (!QEMU_IS_ALIGNED(offset | bytes, s->cluster_size)) { 3561 assert(bytes < s->cluster_size); 3562 /* Ignore partial clusters, except for the special case of the 3563 * complete partial cluster at the end of an unaligned file */ 3564 if (!QEMU_IS_ALIGNED(offset, s->cluster_size) || 3565 offset + bytes != bs->total_sectors * BDRV_SECTOR_SIZE) { 3566 return -ENOTSUP; 3567 } 3568 } 3569 3570 qemu_co_mutex_lock(&s->lock); 3571 ret = qcow2_cluster_discard(bs, offset, bytes, QCOW2_DISCARD_REQUEST, 3572 false); 3573 qemu_co_mutex_unlock(&s->lock); 3574 return ret; 3575 } 3576 3577 static int coroutine_fn 3578 qcow2_co_copy_range_from(BlockDriverState *bs, 3579 BdrvChild *src, uint64_t src_offset, 3580 BdrvChild *dst, uint64_t dst_offset, 3581 uint64_t bytes, BdrvRequestFlags read_flags, 3582 BdrvRequestFlags write_flags) 3583 { 3584 BDRVQcow2State *s = bs->opaque; 3585 int ret; 3586 unsigned int cur_bytes; /* number of bytes in current iteration */ 3587 BdrvChild *child = NULL; 3588 BdrvRequestFlags cur_write_flags; 3589 3590 assert(!bs->encrypted); 3591 qemu_co_mutex_lock(&s->lock); 3592 3593 while (bytes != 0) { 3594 uint64_t copy_offset = 0; 3595 /* prepare next request */ 3596 cur_bytes = MIN(bytes, INT_MAX); 3597 cur_write_flags = write_flags; 3598 3599 ret = qcow2_get_cluster_offset(bs, src_offset, &cur_bytes, ©_offset); 3600 if (ret < 0) { 3601 goto out; 3602 } 3603 3604 switch (ret) { 3605 case QCOW2_CLUSTER_UNALLOCATED: 3606 if (bs->backing && bs->backing->bs) { 3607 int64_t backing_length = bdrv_getlength(bs->backing->bs); 3608 if (src_offset >= backing_length) { 3609 cur_write_flags |= BDRV_REQ_ZERO_WRITE; 3610 } else { 3611 child = bs->backing; 3612 cur_bytes = MIN(cur_bytes, backing_length - src_offset); 3613 copy_offset = src_offset; 3614 } 3615 } else { 3616 cur_write_flags |= BDRV_REQ_ZERO_WRITE; 3617 } 3618 break; 3619 3620 case QCOW2_CLUSTER_ZERO_PLAIN: 3621 case QCOW2_CLUSTER_ZERO_ALLOC: 3622 cur_write_flags |= BDRV_REQ_ZERO_WRITE; 3623 break; 3624 3625 case QCOW2_CLUSTER_COMPRESSED: 3626 ret = -ENOTSUP; 3627 goto out; 3628 3629 case QCOW2_CLUSTER_NORMAL: 3630 child = s->data_file; 3631 copy_offset += offset_into_cluster(s, src_offset); 3632 if ((copy_offset & 511) != 0) { 3633 ret = -EIO; 3634 goto out; 3635 } 3636 break; 3637 3638 default: 3639 abort(); 3640 } 3641 qemu_co_mutex_unlock(&s->lock); 3642 ret = bdrv_co_copy_range_from(child, 3643 copy_offset, 3644 dst, dst_offset, 3645 cur_bytes, read_flags, cur_write_flags); 3646 qemu_co_mutex_lock(&s->lock); 3647 if (ret < 0) { 3648 goto out; 3649 } 3650 3651 bytes -= cur_bytes; 3652 src_offset += cur_bytes; 3653 dst_offset += cur_bytes; 3654 } 3655 ret = 0; 3656 3657 out: 3658 qemu_co_mutex_unlock(&s->lock); 3659 return ret; 3660 } 3661 3662 static int coroutine_fn 3663 qcow2_co_copy_range_to(BlockDriverState *bs, 3664 BdrvChild *src, uint64_t src_offset, 3665 BdrvChild *dst, uint64_t dst_offset, 3666 uint64_t bytes, BdrvRequestFlags read_flags, 3667 BdrvRequestFlags write_flags) 3668 { 3669 BDRVQcow2State *s = bs->opaque; 3670 int offset_in_cluster; 3671 int ret; 3672 unsigned int cur_bytes; /* number of sectors in current iteration */ 3673 uint64_t cluster_offset; 3674 QCowL2Meta *l2meta = NULL; 3675 3676 assert(!bs->encrypted); 3677 3678 qemu_co_mutex_lock(&s->lock); 3679 3680 while (bytes != 0) { 3681 3682 l2meta = NULL; 3683 3684 offset_in_cluster = offset_into_cluster(s, dst_offset); 3685 cur_bytes = MIN(bytes, INT_MAX); 3686 3687 /* TODO: 3688 * If src->bs == dst->bs, we could simply copy by incrementing 3689 * the refcnt, without copying user data. 3690 * Or if src->bs == dst->bs->backing->bs, we could copy by discarding. */ 3691 ret = qcow2_alloc_cluster_offset(bs, dst_offset, &cur_bytes, 3692 &cluster_offset, &l2meta); 3693 if (ret < 0) { 3694 goto fail; 3695 } 3696 3697 assert((cluster_offset & 511) == 0); 3698 3699 ret = qcow2_pre_write_overlap_check(bs, 0, 3700 cluster_offset + offset_in_cluster, cur_bytes, true); 3701 if (ret < 0) { 3702 goto fail; 3703 } 3704 3705 qemu_co_mutex_unlock(&s->lock); 3706 ret = bdrv_co_copy_range_to(src, src_offset, 3707 s->data_file, 3708 cluster_offset + offset_in_cluster, 3709 cur_bytes, read_flags, write_flags); 3710 qemu_co_mutex_lock(&s->lock); 3711 if (ret < 0) { 3712 goto fail; 3713 } 3714 3715 ret = qcow2_handle_l2meta(bs, &l2meta, true); 3716 if (ret) { 3717 goto fail; 3718 } 3719 3720 bytes -= cur_bytes; 3721 src_offset += cur_bytes; 3722 dst_offset += cur_bytes; 3723 } 3724 ret = 0; 3725 3726 fail: 3727 qcow2_handle_l2meta(bs, &l2meta, false); 3728 3729 qemu_co_mutex_unlock(&s->lock); 3730 3731 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret); 3732 3733 return ret; 3734 } 3735 3736 static int coroutine_fn qcow2_co_truncate(BlockDriverState *bs, int64_t offset, 3737 PreallocMode prealloc, Error **errp) 3738 { 3739 BDRVQcow2State *s = bs->opaque; 3740 uint64_t old_length; 3741 int64_t new_l1_size; 3742 int ret; 3743 QDict *options; 3744 3745 if (prealloc != PREALLOC_MODE_OFF && prealloc != PREALLOC_MODE_METADATA && 3746 prealloc != PREALLOC_MODE_FALLOC && prealloc != PREALLOC_MODE_FULL) 3747 { 3748 error_setg(errp, "Unsupported preallocation mode '%s'", 3749 PreallocMode_str(prealloc)); 3750 return -ENOTSUP; 3751 } 3752 3753 if (offset & 511) { 3754 error_setg(errp, "The new size must be a multiple of 512"); 3755 return -EINVAL; 3756 } 3757 3758 qemu_co_mutex_lock(&s->lock); 3759 3760 /* cannot proceed if image has snapshots */ 3761 if (s->nb_snapshots) { 3762 error_setg(errp, "Can't resize an image which has snapshots"); 3763 ret = -ENOTSUP; 3764 goto fail; 3765 } 3766 3767 /* cannot proceed if image has bitmaps */ 3768 if (qcow2_truncate_bitmaps_check(bs, errp)) { 3769 ret = -ENOTSUP; 3770 goto fail; 3771 } 3772 3773 old_length = bs->total_sectors * BDRV_SECTOR_SIZE; 3774 new_l1_size = size_to_l1(s, offset); 3775 3776 if (offset < old_length) { 3777 int64_t last_cluster, old_file_size; 3778 if (prealloc != PREALLOC_MODE_OFF) { 3779 error_setg(errp, 3780 "Preallocation can't be used for shrinking an image"); 3781 ret = -EINVAL; 3782 goto fail; 3783 } 3784 3785 ret = qcow2_cluster_discard(bs, ROUND_UP(offset, s->cluster_size), 3786 old_length - ROUND_UP(offset, 3787 s->cluster_size), 3788 QCOW2_DISCARD_ALWAYS, true); 3789 if (ret < 0) { 3790 error_setg_errno(errp, -ret, "Failed to discard cropped clusters"); 3791 goto fail; 3792 } 3793 3794 ret = qcow2_shrink_l1_table(bs, new_l1_size); 3795 if (ret < 0) { 3796 error_setg_errno(errp, -ret, 3797 "Failed to reduce the number of L2 tables"); 3798 goto fail; 3799 } 3800 3801 ret = qcow2_shrink_reftable(bs); 3802 if (ret < 0) { 3803 error_setg_errno(errp, -ret, 3804 "Failed to discard unused refblocks"); 3805 goto fail; 3806 } 3807 3808 old_file_size = bdrv_getlength(bs->file->bs); 3809 if (old_file_size < 0) { 3810 error_setg_errno(errp, -old_file_size, 3811 "Failed to inquire current file length"); 3812 ret = old_file_size; 3813 goto fail; 3814 } 3815 last_cluster = qcow2_get_last_cluster(bs, old_file_size); 3816 if (last_cluster < 0) { 3817 error_setg_errno(errp, -last_cluster, 3818 "Failed to find the last cluster"); 3819 ret = last_cluster; 3820 goto fail; 3821 } 3822 if ((last_cluster + 1) * s->cluster_size < old_file_size) { 3823 Error *local_err = NULL; 3824 3825 bdrv_co_truncate(bs->file, (last_cluster + 1) * s->cluster_size, 3826 PREALLOC_MODE_OFF, &local_err); 3827 if (local_err) { 3828 warn_reportf_err(local_err, 3829 "Failed to truncate the tail of the image: "); 3830 } 3831 } 3832 } else { 3833 ret = qcow2_grow_l1_table(bs, new_l1_size, true); 3834 if (ret < 0) { 3835 error_setg_errno(errp, -ret, "Failed to grow the L1 table"); 3836 goto fail; 3837 } 3838 } 3839 3840 switch (prealloc) { 3841 case PREALLOC_MODE_OFF: 3842 if (has_data_file(bs)) { 3843 ret = bdrv_co_truncate(s->data_file, offset, prealloc, errp); 3844 if (ret < 0) { 3845 goto fail; 3846 } 3847 } 3848 break; 3849 3850 case PREALLOC_MODE_METADATA: 3851 ret = preallocate_co(bs, old_length, offset, prealloc, errp); 3852 if (ret < 0) { 3853 goto fail; 3854 } 3855 break; 3856 3857 case PREALLOC_MODE_FALLOC: 3858 case PREALLOC_MODE_FULL: 3859 { 3860 int64_t allocation_start, host_offset, guest_offset; 3861 int64_t clusters_allocated; 3862 int64_t old_file_size, new_file_size; 3863 uint64_t nb_new_data_clusters, nb_new_l2_tables; 3864 3865 /* With a data file, preallocation means just allocating the metadata 3866 * and forwarding the truncate request to the data file */ 3867 if (has_data_file(bs)) { 3868 ret = preallocate_co(bs, old_length, offset, prealloc, errp); 3869 if (ret < 0) { 3870 goto fail; 3871 } 3872 break; 3873 } 3874 3875 old_file_size = bdrv_getlength(bs->file->bs); 3876 if (old_file_size < 0) { 3877 error_setg_errno(errp, -old_file_size, 3878 "Failed to inquire current file length"); 3879 ret = old_file_size; 3880 goto fail; 3881 } 3882 old_file_size = ROUND_UP(old_file_size, s->cluster_size); 3883 3884 nb_new_data_clusters = DIV_ROUND_UP(offset - old_length, 3885 s->cluster_size); 3886 3887 /* This is an overestimation; we will not actually allocate space for 3888 * these in the file but just make sure the new refcount structures are 3889 * able to cover them so we will not have to allocate new refblocks 3890 * while entering the data blocks in the potentially new L2 tables. 3891 * (We do not actually care where the L2 tables are placed. Maybe they 3892 * are already allocated or they can be placed somewhere before 3893 * @old_file_size. It does not matter because they will be fully 3894 * allocated automatically, so they do not need to be covered by the 3895 * preallocation. All that matters is that we will not have to allocate 3896 * new refcount structures for them.) */ 3897 nb_new_l2_tables = DIV_ROUND_UP(nb_new_data_clusters, 3898 s->cluster_size / sizeof(uint64_t)); 3899 /* The cluster range may not be aligned to L2 boundaries, so add one L2 3900 * table for a potential head/tail */ 3901 nb_new_l2_tables++; 3902 3903 allocation_start = qcow2_refcount_area(bs, old_file_size, 3904 nb_new_data_clusters + 3905 nb_new_l2_tables, 3906 true, 0, 0); 3907 if (allocation_start < 0) { 3908 error_setg_errno(errp, -allocation_start, 3909 "Failed to resize refcount structures"); 3910 ret = allocation_start; 3911 goto fail; 3912 } 3913 3914 clusters_allocated = qcow2_alloc_clusters_at(bs, allocation_start, 3915 nb_new_data_clusters); 3916 if (clusters_allocated < 0) { 3917 error_setg_errno(errp, -clusters_allocated, 3918 "Failed to allocate data clusters"); 3919 ret = clusters_allocated; 3920 goto fail; 3921 } 3922 3923 assert(clusters_allocated == nb_new_data_clusters); 3924 3925 /* Allocate the data area */ 3926 new_file_size = allocation_start + 3927 nb_new_data_clusters * s->cluster_size; 3928 ret = bdrv_co_truncate(bs->file, new_file_size, prealloc, errp); 3929 if (ret < 0) { 3930 error_prepend(errp, "Failed to resize underlying file: "); 3931 qcow2_free_clusters(bs, allocation_start, 3932 nb_new_data_clusters * s->cluster_size, 3933 QCOW2_DISCARD_OTHER); 3934 goto fail; 3935 } 3936 3937 /* Create the necessary L2 entries */ 3938 host_offset = allocation_start; 3939 guest_offset = old_length; 3940 while (nb_new_data_clusters) { 3941 int64_t nb_clusters = MIN( 3942 nb_new_data_clusters, 3943 s->l2_slice_size - offset_to_l2_slice_index(s, guest_offset)); 3944 QCowL2Meta allocation = { 3945 .offset = guest_offset, 3946 .alloc_offset = host_offset, 3947 .nb_clusters = nb_clusters, 3948 }; 3949 qemu_co_queue_init(&allocation.dependent_requests); 3950 3951 ret = qcow2_alloc_cluster_link_l2(bs, &allocation); 3952 if (ret < 0) { 3953 error_setg_errno(errp, -ret, "Failed to update L2 tables"); 3954 qcow2_free_clusters(bs, host_offset, 3955 nb_new_data_clusters * s->cluster_size, 3956 QCOW2_DISCARD_OTHER); 3957 goto fail; 3958 } 3959 3960 guest_offset += nb_clusters * s->cluster_size; 3961 host_offset += nb_clusters * s->cluster_size; 3962 nb_new_data_clusters -= nb_clusters; 3963 } 3964 break; 3965 } 3966 3967 default: 3968 g_assert_not_reached(); 3969 } 3970 3971 if (prealloc != PREALLOC_MODE_OFF) { 3972 /* Flush metadata before actually changing the image size */ 3973 ret = qcow2_write_caches(bs); 3974 if (ret < 0) { 3975 error_setg_errno(errp, -ret, 3976 "Failed to flush the preallocated area to disk"); 3977 goto fail; 3978 } 3979 } 3980 3981 bs->total_sectors = offset / BDRV_SECTOR_SIZE; 3982 3983 /* write updated header.size */ 3984 offset = cpu_to_be64(offset); 3985 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, size), 3986 &offset, sizeof(uint64_t)); 3987 if (ret < 0) { 3988 error_setg_errno(errp, -ret, "Failed to update the image size"); 3989 goto fail; 3990 } 3991 3992 s->l1_vm_state_index = new_l1_size; 3993 3994 /* Update cache sizes */ 3995 options = qdict_clone_shallow(bs->options); 3996 ret = qcow2_update_options(bs, options, s->flags, errp); 3997 qobject_unref(options); 3998 if (ret < 0) { 3999 goto fail; 4000 } 4001 ret = 0; 4002 fail: 4003 qemu_co_mutex_unlock(&s->lock); 4004 return ret; 4005 } 4006 4007 /* XXX: put compressed sectors first, then all the cluster aligned 4008 tables to avoid losing bytes in alignment */ 4009 static coroutine_fn int 4010 qcow2_co_pwritev_compressed(BlockDriverState *bs, uint64_t offset, 4011 uint64_t bytes, QEMUIOVector *qiov) 4012 { 4013 BDRVQcow2State *s = bs->opaque; 4014 int ret; 4015 ssize_t out_len; 4016 uint8_t *buf, *out_buf; 4017 uint64_t cluster_offset; 4018 4019 if (has_data_file(bs)) { 4020 return -ENOTSUP; 4021 } 4022 4023 if (bytes == 0) { 4024 /* align end of file to a sector boundary to ease reading with 4025 sector based I/Os */ 4026 int64_t len = bdrv_getlength(bs->file->bs); 4027 if (len < 0) { 4028 return len; 4029 } 4030 return bdrv_co_truncate(bs->file, len, PREALLOC_MODE_OFF, NULL); 4031 } 4032 4033 if (offset_into_cluster(s, offset)) { 4034 return -EINVAL; 4035 } 4036 4037 buf = qemu_blockalign(bs, s->cluster_size); 4038 if (bytes != s->cluster_size) { 4039 if (bytes > s->cluster_size || 4040 offset + bytes != bs->total_sectors << BDRV_SECTOR_BITS) 4041 { 4042 qemu_vfree(buf); 4043 return -EINVAL; 4044 } 4045 /* Zero-pad last write if image size is not cluster aligned */ 4046 memset(buf + bytes, 0, s->cluster_size - bytes); 4047 } 4048 qemu_iovec_to_buf(qiov, 0, buf, bytes); 4049 4050 out_buf = g_malloc(s->cluster_size); 4051 4052 out_len = qcow2_co_compress(bs, out_buf, s->cluster_size - 1, 4053 buf, s->cluster_size); 4054 if (out_len == -ENOMEM) { 4055 /* could not compress: write normal cluster */ 4056 ret = qcow2_co_pwritev(bs, offset, bytes, qiov, 0); 4057 if (ret < 0) { 4058 goto fail; 4059 } 4060 goto success; 4061 } else if (out_len < 0) { 4062 ret = -EINVAL; 4063 goto fail; 4064 } 4065 4066 qemu_co_mutex_lock(&s->lock); 4067 ret = qcow2_alloc_compressed_cluster_offset(bs, offset, out_len, 4068 &cluster_offset); 4069 if (ret < 0) { 4070 qemu_co_mutex_unlock(&s->lock); 4071 goto fail; 4072 } 4073 4074 ret = qcow2_pre_write_overlap_check(bs, 0, cluster_offset, out_len, true); 4075 qemu_co_mutex_unlock(&s->lock); 4076 if (ret < 0) { 4077 goto fail; 4078 } 4079 4080 BLKDBG_EVENT(s->data_file, BLKDBG_WRITE_COMPRESSED); 4081 ret = bdrv_co_pwrite(s->data_file, cluster_offset, out_len, out_buf, 0); 4082 if (ret < 0) { 4083 goto fail; 4084 } 4085 success: 4086 ret = 0; 4087 fail: 4088 qemu_vfree(buf); 4089 g_free(out_buf); 4090 return ret; 4091 } 4092 4093 static int coroutine_fn 4094 qcow2_co_preadv_compressed(BlockDriverState *bs, 4095 uint64_t file_cluster_offset, 4096 uint64_t offset, 4097 uint64_t bytes, 4098 QEMUIOVector *qiov) 4099 { 4100 BDRVQcow2State *s = bs->opaque; 4101 int ret = 0, csize, nb_csectors; 4102 uint64_t coffset; 4103 uint8_t *buf, *out_buf; 4104 int offset_in_cluster = offset_into_cluster(s, offset); 4105 4106 coffset = file_cluster_offset & s->cluster_offset_mask; 4107 nb_csectors = ((file_cluster_offset >> s->csize_shift) & s->csize_mask) + 1; 4108 csize = nb_csectors * QCOW2_COMPRESSED_SECTOR_SIZE - 4109 (coffset & ~QCOW2_COMPRESSED_SECTOR_MASK); 4110 4111 buf = g_try_malloc(csize); 4112 if (!buf) { 4113 return -ENOMEM; 4114 } 4115 4116 out_buf = qemu_blockalign(bs, s->cluster_size); 4117 4118 BLKDBG_EVENT(bs->file, BLKDBG_READ_COMPRESSED); 4119 ret = bdrv_co_pread(bs->file, coffset, csize, buf, 0); 4120 if (ret < 0) { 4121 goto fail; 4122 } 4123 4124 if (qcow2_co_decompress(bs, out_buf, s->cluster_size, buf, csize) < 0) { 4125 ret = -EIO; 4126 goto fail; 4127 } 4128 4129 qemu_iovec_from_buf(qiov, 0, out_buf + offset_in_cluster, bytes); 4130 4131 fail: 4132 qemu_vfree(out_buf); 4133 g_free(buf); 4134 4135 return ret; 4136 } 4137 4138 static int make_completely_empty(BlockDriverState *bs) 4139 { 4140 BDRVQcow2State *s = bs->opaque; 4141 Error *local_err = NULL; 4142 int ret, l1_clusters; 4143 int64_t offset; 4144 uint64_t *new_reftable = NULL; 4145 uint64_t rt_entry, l1_size2; 4146 struct { 4147 uint64_t l1_offset; 4148 uint64_t reftable_offset; 4149 uint32_t reftable_clusters; 4150 } QEMU_PACKED l1_ofs_rt_ofs_cls; 4151 4152 ret = qcow2_cache_empty(bs, s->l2_table_cache); 4153 if (ret < 0) { 4154 goto fail; 4155 } 4156 4157 ret = qcow2_cache_empty(bs, s->refcount_block_cache); 4158 if (ret < 0) { 4159 goto fail; 4160 } 4161 4162 /* Refcounts will be broken utterly */ 4163 ret = qcow2_mark_dirty(bs); 4164 if (ret < 0) { 4165 goto fail; 4166 } 4167 4168 BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE); 4169 4170 l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / sizeof(uint64_t)); 4171 l1_size2 = (uint64_t)s->l1_size * sizeof(uint64_t); 4172 4173 /* After this call, neither the in-memory nor the on-disk refcount 4174 * information accurately describe the actual references */ 4175 4176 ret = bdrv_pwrite_zeroes(bs->file, s->l1_table_offset, 4177 l1_clusters * s->cluster_size, 0); 4178 if (ret < 0) { 4179 goto fail_broken_refcounts; 4180 } 4181 memset(s->l1_table, 0, l1_size2); 4182 4183 BLKDBG_EVENT(bs->file, BLKDBG_EMPTY_IMAGE_PREPARE); 4184 4185 /* Overwrite enough clusters at the beginning of the sectors to place 4186 * the refcount table, a refcount block and the L1 table in; this may 4187 * overwrite parts of the existing refcount and L1 table, which is not 4188 * an issue because the dirty flag is set, complete data loss is in fact 4189 * desired and partial data loss is consequently fine as well */ 4190 ret = bdrv_pwrite_zeroes(bs->file, s->cluster_size, 4191 (2 + l1_clusters) * s->cluster_size, 0); 4192 /* This call (even if it failed overall) may have overwritten on-disk 4193 * refcount structures; in that case, the in-memory refcount information 4194 * will probably differ from the on-disk information which makes the BDS 4195 * unusable */ 4196 if (ret < 0) { 4197 goto fail_broken_refcounts; 4198 } 4199 4200 BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE); 4201 BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_UPDATE); 4202 4203 /* "Create" an empty reftable (one cluster) directly after the image 4204 * header and an empty L1 table three clusters after the image header; 4205 * the cluster between those two will be used as the first refblock */ 4206 l1_ofs_rt_ofs_cls.l1_offset = cpu_to_be64(3 * s->cluster_size); 4207 l1_ofs_rt_ofs_cls.reftable_offset = cpu_to_be64(s->cluster_size); 4208 l1_ofs_rt_ofs_cls.reftable_clusters = cpu_to_be32(1); 4209 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, l1_table_offset), 4210 &l1_ofs_rt_ofs_cls, sizeof(l1_ofs_rt_ofs_cls)); 4211 if (ret < 0) { 4212 goto fail_broken_refcounts; 4213 } 4214 4215 s->l1_table_offset = 3 * s->cluster_size; 4216 4217 new_reftable = g_try_new0(uint64_t, s->cluster_size / sizeof(uint64_t)); 4218 if (!new_reftable) { 4219 ret = -ENOMEM; 4220 goto fail_broken_refcounts; 4221 } 4222 4223 s->refcount_table_offset = s->cluster_size; 4224 s->refcount_table_size = s->cluster_size / sizeof(uint64_t); 4225 s->max_refcount_table_index = 0; 4226 4227 g_free(s->refcount_table); 4228 s->refcount_table = new_reftable; 4229 new_reftable = NULL; 4230 4231 /* Now the in-memory refcount information again corresponds to the on-disk 4232 * information (reftable is empty and no refblocks (the refblock cache is 4233 * empty)); however, this means some clusters (e.g. the image header) are 4234 * referenced, but not refcounted, but the normal qcow2 code assumes that 4235 * the in-memory information is always correct */ 4236 4237 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC); 4238 4239 /* Enter the first refblock into the reftable */ 4240 rt_entry = cpu_to_be64(2 * s->cluster_size); 4241 ret = bdrv_pwrite_sync(bs->file, s->cluster_size, 4242 &rt_entry, sizeof(rt_entry)); 4243 if (ret < 0) { 4244 goto fail_broken_refcounts; 4245 } 4246 s->refcount_table[0] = 2 * s->cluster_size; 4247 4248 s->free_cluster_index = 0; 4249 assert(3 + l1_clusters <= s->refcount_block_size); 4250 offset = qcow2_alloc_clusters(bs, 3 * s->cluster_size + l1_size2); 4251 if (offset < 0) { 4252 ret = offset; 4253 goto fail_broken_refcounts; 4254 } else if (offset > 0) { 4255 error_report("First cluster in emptied image is in use"); 4256 abort(); 4257 } 4258 4259 /* Now finally the in-memory information corresponds to the on-disk 4260 * structures and is correct */ 4261 ret = qcow2_mark_clean(bs); 4262 if (ret < 0) { 4263 goto fail; 4264 } 4265 4266 ret = bdrv_truncate(bs->file, (3 + l1_clusters) * s->cluster_size, 4267 PREALLOC_MODE_OFF, &local_err); 4268 if (ret < 0) { 4269 error_report_err(local_err); 4270 goto fail; 4271 } 4272 4273 return 0; 4274 4275 fail_broken_refcounts: 4276 /* The BDS is unusable at this point. If we wanted to make it usable, we 4277 * would have to call qcow2_refcount_close(), qcow2_refcount_init(), 4278 * qcow2_check_refcounts(), qcow2_refcount_close() and qcow2_refcount_init() 4279 * again. However, because the functions which could have caused this error 4280 * path to be taken are used by those functions as well, it's very likely 4281 * that that sequence will fail as well. Therefore, just eject the BDS. */ 4282 bs->drv = NULL; 4283 4284 fail: 4285 g_free(new_reftable); 4286 return ret; 4287 } 4288 4289 static int qcow2_make_empty(BlockDriverState *bs) 4290 { 4291 BDRVQcow2State *s = bs->opaque; 4292 uint64_t offset, end_offset; 4293 int step = QEMU_ALIGN_DOWN(INT_MAX, s->cluster_size); 4294 int l1_clusters, ret = 0; 4295 4296 l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / sizeof(uint64_t)); 4297 4298 if (s->qcow_version >= 3 && !s->snapshots && !s->nb_bitmaps && 4299 3 + l1_clusters <= s->refcount_block_size && 4300 s->crypt_method_header != QCOW_CRYPT_LUKS && 4301 !has_data_file(bs)) { 4302 /* The following function only works for qcow2 v3 images (it 4303 * requires the dirty flag) and only as long as there are no 4304 * features that reserve extra clusters (such as snapshots, 4305 * LUKS header, or persistent bitmaps), because it completely 4306 * empties the image. Furthermore, the L1 table and three 4307 * additional clusters (image header, refcount table, one 4308 * refcount block) have to fit inside one refcount block. It 4309 * only resets the image file, i.e. does not work with an 4310 * external data file. */ 4311 return make_completely_empty(bs); 4312 } 4313 4314 /* This fallback code simply discards every active cluster; this is slow, 4315 * but works in all cases */ 4316 end_offset = bs->total_sectors * BDRV_SECTOR_SIZE; 4317 for (offset = 0; offset < end_offset; offset += step) { 4318 /* As this function is generally used after committing an external 4319 * snapshot, QCOW2_DISCARD_SNAPSHOT seems appropriate. Also, the 4320 * default action for this kind of discard is to pass the discard, 4321 * which will ideally result in an actually smaller image file, as 4322 * is probably desired. */ 4323 ret = qcow2_cluster_discard(bs, offset, MIN(step, end_offset - offset), 4324 QCOW2_DISCARD_SNAPSHOT, true); 4325 if (ret < 0) { 4326 break; 4327 } 4328 } 4329 4330 return ret; 4331 } 4332 4333 static coroutine_fn int qcow2_co_flush_to_os(BlockDriverState *bs) 4334 { 4335 BDRVQcow2State *s = bs->opaque; 4336 int ret; 4337 4338 qemu_co_mutex_lock(&s->lock); 4339 ret = qcow2_write_caches(bs); 4340 qemu_co_mutex_unlock(&s->lock); 4341 4342 return ret; 4343 } 4344 4345 static ssize_t qcow2_measure_crypto_hdr_init_func(QCryptoBlock *block, 4346 size_t headerlen, void *opaque, Error **errp) 4347 { 4348 size_t *headerlenp = opaque; 4349 4350 /* Stash away the payload size */ 4351 *headerlenp = headerlen; 4352 return 0; 4353 } 4354 4355 static ssize_t qcow2_measure_crypto_hdr_write_func(QCryptoBlock *block, 4356 size_t offset, const uint8_t *buf, size_t buflen, 4357 void *opaque, Error **errp) 4358 { 4359 /* Discard the bytes, we're not actually writing to an image */ 4360 return buflen; 4361 } 4362 4363 /* Determine the number of bytes for the LUKS payload */ 4364 static bool qcow2_measure_luks_headerlen(QemuOpts *opts, size_t *len, 4365 Error **errp) 4366 { 4367 QDict *opts_qdict; 4368 QDict *cryptoopts_qdict; 4369 QCryptoBlockCreateOptions *cryptoopts; 4370 QCryptoBlock *crypto; 4371 4372 /* Extract "encrypt." options into a qdict */ 4373 opts_qdict = qemu_opts_to_qdict(opts, NULL); 4374 qdict_extract_subqdict(opts_qdict, &cryptoopts_qdict, "encrypt."); 4375 qobject_unref(opts_qdict); 4376 4377 /* Build QCryptoBlockCreateOptions object from qdict */ 4378 qdict_put_str(cryptoopts_qdict, "format", "luks"); 4379 cryptoopts = block_crypto_create_opts_init(cryptoopts_qdict, errp); 4380 qobject_unref(cryptoopts_qdict); 4381 if (!cryptoopts) { 4382 return false; 4383 } 4384 4385 /* Fake LUKS creation in order to determine the payload size */ 4386 crypto = qcrypto_block_create(cryptoopts, "encrypt.", 4387 qcow2_measure_crypto_hdr_init_func, 4388 qcow2_measure_crypto_hdr_write_func, 4389 len, errp); 4390 qapi_free_QCryptoBlockCreateOptions(cryptoopts); 4391 if (!crypto) { 4392 return false; 4393 } 4394 4395 qcrypto_block_free(crypto); 4396 return true; 4397 } 4398 4399 static BlockMeasureInfo *qcow2_measure(QemuOpts *opts, BlockDriverState *in_bs, 4400 Error **errp) 4401 { 4402 Error *local_err = NULL; 4403 BlockMeasureInfo *info; 4404 uint64_t required = 0; /* bytes that contribute to required size */ 4405 uint64_t virtual_size; /* disk size as seen by guest */ 4406 uint64_t refcount_bits; 4407 uint64_t l2_tables; 4408 uint64_t luks_payload_size = 0; 4409 size_t cluster_size; 4410 int version; 4411 char *optstr; 4412 PreallocMode prealloc; 4413 bool has_backing_file; 4414 bool has_luks; 4415 4416 /* Parse image creation options */ 4417 cluster_size = qcow2_opt_get_cluster_size_del(opts, &local_err); 4418 if (local_err) { 4419 goto err; 4420 } 4421 4422 version = qcow2_opt_get_version_del(opts, &local_err); 4423 if (local_err) { 4424 goto err; 4425 } 4426 4427 refcount_bits = qcow2_opt_get_refcount_bits_del(opts, version, &local_err); 4428 if (local_err) { 4429 goto err; 4430 } 4431 4432 optstr = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC); 4433 prealloc = qapi_enum_parse(&PreallocMode_lookup, optstr, 4434 PREALLOC_MODE_OFF, &local_err); 4435 g_free(optstr); 4436 if (local_err) { 4437 goto err; 4438 } 4439 4440 optstr = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE); 4441 has_backing_file = !!optstr; 4442 g_free(optstr); 4443 4444 optstr = qemu_opt_get_del(opts, BLOCK_OPT_ENCRYPT_FORMAT); 4445 has_luks = optstr && strcmp(optstr, "luks") == 0; 4446 g_free(optstr); 4447 4448 if (has_luks) { 4449 size_t headerlen; 4450 4451 if (!qcow2_measure_luks_headerlen(opts, &headerlen, &local_err)) { 4452 goto err; 4453 } 4454 4455 luks_payload_size = ROUND_UP(headerlen, cluster_size); 4456 } 4457 4458 virtual_size = qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0); 4459 virtual_size = ROUND_UP(virtual_size, cluster_size); 4460 4461 /* Check that virtual disk size is valid */ 4462 l2_tables = DIV_ROUND_UP(virtual_size / cluster_size, 4463 cluster_size / sizeof(uint64_t)); 4464 if (l2_tables * sizeof(uint64_t) > QCOW_MAX_L1_SIZE) { 4465 error_setg(&local_err, "The image size is too large " 4466 "(try using a larger cluster size)"); 4467 goto err; 4468 } 4469 4470 /* Account for input image */ 4471 if (in_bs) { 4472 int64_t ssize = bdrv_getlength(in_bs); 4473 if (ssize < 0) { 4474 error_setg_errno(&local_err, -ssize, 4475 "Unable to get image virtual_size"); 4476 goto err; 4477 } 4478 4479 virtual_size = ROUND_UP(ssize, cluster_size); 4480 4481 if (has_backing_file) { 4482 /* We don't how much of the backing chain is shared by the input 4483 * image and the new image file. In the worst case the new image's 4484 * backing file has nothing in common with the input image. Be 4485 * conservative and assume all clusters need to be written. 4486 */ 4487 required = virtual_size; 4488 } else { 4489 int64_t offset; 4490 int64_t pnum = 0; 4491 4492 for (offset = 0; offset < ssize; offset += pnum) { 4493 int ret; 4494 4495 ret = bdrv_block_status_above(in_bs, NULL, offset, 4496 ssize - offset, &pnum, NULL, 4497 NULL); 4498 if (ret < 0) { 4499 error_setg_errno(&local_err, -ret, 4500 "Unable to get block status"); 4501 goto err; 4502 } 4503 4504 if (ret & BDRV_BLOCK_ZERO) { 4505 /* Skip zero regions (safe with no backing file) */ 4506 } else if ((ret & (BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED)) == 4507 (BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED)) { 4508 /* Extend pnum to end of cluster for next iteration */ 4509 pnum = ROUND_UP(offset + pnum, cluster_size) - offset; 4510 4511 /* Count clusters we've seen */ 4512 required += offset % cluster_size + pnum; 4513 } 4514 } 4515 } 4516 } 4517 4518 /* Take into account preallocation. Nothing special is needed for 4519 * PREALLOC_MODE_METADATA since metadata is always counted. 4520 */ 4521 if (prealloc == PREALLOC_MODE_FULL || prealloc == PREALLOC_MODE_FALLOC) { 4522 required = virtual_size; 4523 } 4524 4525 info = g_new(BlockMeasureInfo, 1); 4526 info->fully_allocated = 4527 qcow2_calc_prealloc_size(virtual_size, cluster_size, 4528 ctz32(refcount_bits)) + luks_payload_size; 4529 4530 /* Remove data clusters that are not required. This overestimates the 4531 * required size because metadata needed for the fully allocated file is 4532 * still counted. 4533 */ 4534 info->required = info->fully_allocated - virtual_size + required; 4535 return info; 4536 4537 err: 4538 error_propagate(errp, local_err); 4539 return NULL; 4540 } 4541 4542 static int qcow2_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) 4543 { 4544 BDRVQcow2State *s = bs->opaque; 4545 bdi->unallocated_blocks_are_zero = true; 4546 bdi->cluster_size = s->cluster_size; 4547 bdi->vm_state_offset = qcow2_vm_state_offset(s); 4548 return 0; 4549 } 4550 4551 static ImageInfoSpecific *qcow2_get_specific_info(BlockDriverState *bs, 4552 Error **errp) 4553 { 4554 BDRVQcow2State *s = bs->opaque; 4555 ImageInfoSpecific *spec_info; 4556 QCryptoBlockInfo *encrypt_info = NULL; 4557 Error *local_err = NULL; 4558 4559 if (s->crypto != NULL) { 4560 encrypt_info = qcrypto_block_get_info(s->crypto, &local_err); 4561 if (local_err) { 4562 error_propagate(errp, local_err); 4563 return NULL; 4564 } 4565 } 4566 4567 spec_info = g_new(ImageInfoSpecific, 1); 4568 *spec_info = (ImageInfoSpecific){ 4569 .type = IMAGE_INFO_SPECIFIC_KIND_QCOW2, 4570 .u.qcow2.data = g_new0(ImageInfoSpecificQCow2, 1), 4571 }; 4572 if (s->qcow_version == 2) { 4573 *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){ 4574 .compat = g_strdup("0.10"), 4575 .refcount_bits = s->refcount_bits, 4576 }; 4577 } else if (s->qcow_version == 3) { 4578 Qcow2BitmapInfoList *bitmaps; 4579 bitmaps = qcow2_get_bitmap_info_list(bs, &local_err); 4580 if (local_err) { 4581 error_propagate(errp, local_err); 4582 qapi_free_ImageInfoSpecific(spec_info); 4583 return NULL; 4584 } 4585 *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){ 4586 .compat = g_strdup("1.1"), 4587 .lazy_refcounts = s->compatible_features & 4588 QCOW2_COMPAT_LAZY_REFCOUNTS, 4589 .has_lazy_refcounts = true, 4590 .corrupt = s->incompatible_features & 4591 QCOW2_INCOMPAT_CORRUPT, 4592 .has_corrupt = true, 4593 .refcount_bits = s->refcount_bits, 4594 .has_bitmaps = !!bitmaps, 4595 .bitmaps = bitmaps, 4596 .has_data_file = !!s->image_data_file, 4597 .data_file = g_strdup(s->image_data_file), 4598 .has_data_file_raw = has_data_file(bs), 4599 .data_file_raw = data_file_is_raw(bs), 4600 }; 4601 } else { 4602 /* if this assertion fails, this probably means a new version was 4603 * added without having it covered here */ 4604 assert(false); 4605 } 4606 4607 if (encrypt_info) { 4608 ImageInfoSpecificQCow2Encryption *qencrypt = 4609 g_new(ImageInfoSpecificQCow2Encryption, 1); 4610 switch (encrypt_info->format) { 4611 case Q_CRYPTO_BLOCK_FORMAT_QCOW: 4612 qencrypt->format = BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_AES; 4613 break; 4614 case Q_CRYPTO_BLOCK_FORMAT_LUKS: 4615 qencrypt->format = BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_LUKS; 4616 qencrypt->u.luks = encrypt_info->u.luks; 4617 break; 4618 default: 4619 abort(); 4620 } 4621 /* Since we did shallow copy above, erase any pointers 4622 * in the original info */ 4623 memset(&encrypt_info->u, 0, sizeof(encrypt_info->u)); 4624 qapi_free_QCryptoBlockInfo(encrypt_info); 4625 4626 spec_info->u.qcow2.data->has_encrypt = true; 4627 spec_info->u.qcow2.data->encrypt = qencrypt; 4628 } 4629 4630 return spec_info; 4631 } 4632 4633 static int qcow2_save_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, 4634 int64_t pos) 4635 { 4636 BDRVQcow2State *s = bs->opaque; 4637 4638 BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_SAVE); 4639 return bs->drv->bdrv_co_pwritev(bs, qcow2_vm_state_offset(s) + pos, 4640 qiov->size, qiov, 0); 4641 } 4642 4643 static int qcow2_load_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, 4644 int64_t pos) 4645 { 4646 BDRVQcow2State *s = bs->opaque; 4647 4648 BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_LOAD); 4649 return bs->drv->bdrv_co_preadv(bs, qcow2_vm_state_offset(s) + pos, 4650 qiov->size, qiov, 0); 4651 } 4652 4653 /* 4654 * Downgrades an image's version. To achieve this, any incompatible features 4655 * have to be removed. 4656 */ 4657 static int qcow2_downgrade(BlockDriverState *bs, int target_version, 4658 BlockDriverAmendStatusCB *status_cb, void *cb_opaque, 4659 Error **errp) 4660 { 4661 BDRVQcow2State *s = bs->opaque; 4662 int current_version = s->qcow_version; 4663 int ret; 4664 4665 /* This is qcow2_downgrade(), not qcow2_upgrade() */ 4666 assert(target_version < current_version); 4667 4668 /* There are no other versions (now) that you can downgrade to */ 4669 assert(target_version == 2); 4670 4671 if (s->refcount_order != 4) { 4672 error_setg(errp, "compat=0.10 requires refcount_bits=16"); 4673 return -ENOTSUP; 4674 } 4675 4676 if (has_data_file(bs)) { 4677 error_setg(errp, "Cannot downgrade an image with a data file"); 4678 return -ENOTSUP; 4679 } 4680 4681 /* clear incompatible features */ 4682 if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) { 4683 ret = qcow2_mark_clean(bs); 4684 if (ret < 0) { 4685 error_setg_errno(errp, -ret, "Failed to make the image clean"); 4686 return ret; 4687 } 4688 } 4689 4690 /* with QCOW2_INCOMPAT_CORRUPT, it is pretty much impossible to get here in 4691 * the first place; if that happens nonetheless, returning -ENOTSUP is the 4692 * best thing to do anyway */ 4693 4694 if (s->incompatible_features) { 4695 error_setg(errp, "Cannot downgrade an image with incompatible features " 4696 "%#" PRIx64 " set", s->incompatible_features); 4697 return -ENOTSUP; 4698 } 4699 4700 /* since we can ignore compatible features, we can set them to 0 as well */ 4701 s->compatible_features = 0; 4702 /* if lazy refcounts have been used, they have already been fixed through 4703 * clearing the dirty flag */ 4704 4705 /* clearing autoclear features is trivial */ 4706 s->autoclear_features = 0; 4707 4708 ret = qcow2_expand_zero_clusters(bs, status_cb, cb_opaque); 4709 if (ret < 0) { 4710 error_setg_errno(errp, -ret, "Failed to turn zero into data clusters"); 4711 return ret; 4712 } 4713 4714 s->qcow_version = target_version; 4715 ret = qcow2_update_header(bs); 4716 if (ret < 0) { 4717 s->qcow_version = current_version; 4718 error_setg_errno(errp, -ret, "Failed to update the image header"); 4719 return ret; 4720 } 4721 return 0; 4722 } 4723 4724 typedef enum Qcow2AmendOperation { 4725 /* This is the value Qcow2AmendHelperCBInfo::last_operation will be 4726 * statically initialized to so that the helper CB can discern the first 4727 * invocation from an operation change */ 4728 QCOW2_NO_OPERATION = 0, 4729 4730 QCOW2_CHANGING_REFCOUNT_ORDER, 4731 QCOW2_DOWNGRADING, 4732 } Qcow2AmendOperation; 4733 4734 typedef struct Qcow2AmendHelperCBInfo { 4735 /* The code coordinating the amend operations should only modify 4736 * these four fields; the rest will be managed by the CB */ 4737 BlockDriverAmendStatusCB *original_status_cb; 4738 void *original_cb_opaque; 4739 4740 Qcow2AmendOperation current_operation; 4741 4742 /* Total number of operations to perform (only set once) */ 4743 int total_operations; 4744 4745 /* The following fields are managed by the CB */ 4746 4747 /* Number of operations completed */ 4748 int operations_completed; 4749 4750 /* Cumulative offset of all completed operations */ 4751 int64_t offset_completed; 4752 4753 Qcow2AmendOperation last_operation; 4754 int64_t last_work_size; 4755 } Qcow2AmendHelperCBInfo; 4756 4757 static void qcow2_amend_helper_cb(BlockDriverState *bs, 4758 int64_t operation_offset, 4759 int64_t operation_work_size, void *opaque) 4760 { 4761 Qcow2AmendHelperCBInfo *info = opaque; 4762 int64_t current_work_size; 4763 int64_t projected_work_size; 4764 4765 if (info->current_operation != info->last_operation) { 4766 if (info->last_operation != QCOW2_NO_OPERATION) { 4767 info->offset_completed += info->last_work_size; 4768 info->operations_completed++; 4769 } 4770 4771 info->last_operation = info->current_operation; 4772 } 4773 4774 assert(info->total_operations > 0); 4775 assert(info->operations_completed < info->total_operations); 4776 4777 info->last_work_size = operation_work_size; 4778 4779 current_work_size = info->offset_completed + operation_work_size; 4780 4781 /* current_work_size is the total work size for (operations_completed + 1) 4782 * operations (which includes this one), so multiply it by the number of 4783 * operations not covered and divide it by the number of operations 4784 * covered to get a projection for the operations not covered */ 4785 projected_work_size = current_work_size * (info->total_operations - 4786 info->operations_completed - 1) 4787 / (info->operations_completed + 1); 4788 4789 info->original_status_cb(bs, info->offset_completed + operation_offset, 4790 current_work_size + projected_work_size, 4791 info->original_cb_opaque); 4792 } 4793 4794 static int qcow2_amend_options(BlockDriverState *bs, QemuOpts *opts, 4795 BlockDriverAmendStatusCB *status_cb, 4796 void *cb_opaque, 4797 Error **errp) 4798 { 4799 BDRVQcow2State *s = bs->opaque; 4800 int old_version = s->qcow_version, new_version = old_version; 4801 uint64_t new_size = 0; 4802 const char *backing_file = NULL, *backing_format = NULL, *data_file = NULL; 4803 bool lazy_refcounts = s->use_lazy_refcounts; 4804 bool data_file_raw = data_file_is_raw(bs); 4805 const char *compat = NULL; 4806 uint64_t cluster_size = s->cluster_size; 4807 bool encrypt; 4808 int encformat; 4809 int refcount_bits = s->refcount_bits; 4810 int ret; 4811 QemuOptDesc *desc = opts->list->desc; 4812 Qcow2AmendHelperCBInfo helper_cb_info; 4813 4814 while (desc && desc->name) { 4815 if (!qemu_opt_find(opts, desc->name)) { 4816 /* only change explicitly defined options */ 4817 desc++; 4818 continue; 4819 } 4820 4821 if (!strcmp(desc->name, BLOCK_OPT_COMPAT_LEVEL)) { 4822 compat = qemu_opt_get(opts, BLOCK_OPT_COMPAT_LEVEL); 4823 if (!compat) { 4824 /* preserve default */ 4825 } else if (!strcmp(compat, "0.10")) { 4826 new_version = 2; 4827 } else if (!strcmp(compat, "1.1")) { 4828 new_version = 3; 4829 } else { 4830 error_setg(errp, "Unknown compatibility level %s", compat); 4831 return -EINVAL; 4832 } 4833 } else if (!strcmp(desc->name, BLOCK_OPT_PREALLOC)) { 4834 error_setg(errp, "Cannot change preallocation mode"); 4835 return -ENOTSUP; 4836 } else if (!strcmp(desc->name, BLOCK_OPT_SIZE)) { 4837 new_size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, 0); 4838 } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FILE)) { 4839 backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE); 4840 } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FMT)) { 4841 backing_format = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT); 4842 } else if (!strcmp(desc->name, BLOCK_OPT_ENCRYPT)) { 4843 encrypt = qemu_opt_get_bool(opts, BLOCK_OPT_ENCRYPT, 4844 !!s->crypto); 4845 4846 if (encrypt != !!s->crypto) { 4847 error_setg(errp, 4848 "Changing the encryption flag is not supported"); 4849 return -ENOTSUP; 4850 } 4851 } else if (!strcmp(desc->name, BLOCK_OPT_ENCRYPT_FORMAT)) { 4852 encformat = qcow2_crypt_method_from_format( 4853 qemu_opt_get(opts, BLOCK_OPT_ENCRYPT_FORMAT)); 4854 4855 if (encformat != s->crypt_method_header) { 4856 error_setg(errp, 4857 "Changing the encryption format is not supported"); 4858 return -ENOTSUP; 4859 } 4860 } else if (g_str_has_prefix(desc->name, "encrypt.")) { 4861 error_setg(errp, 4862 "Changing the encryption parameters is not supported"); 4863 return -ENOTSUP; 4864 } else if (!strcmp(desc->name, BLOCK_OPT_CLUSTER_SIZE)) { 4865 cluster_size = qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE, 4866 cluster_size); 4867 if (cluster_size != s->cluster_size) { 4868 error_setg(errp, "Changing the cluster size is not supported"); 4869 return -ENOTSUP; 4870 } 4871 } else if (!strcmp(desc->name, BLOCK_OPT_LAZY_REFCOUNTS)) { 4872 lazy_refcounts = qemu_opt_get_bool(opts, BLOCK_OPT_LAZY_REFCOUNTS, 4873 lazy_refcounts); 4874 } else if (!strcmp(desc->name, BLOCK_OPT_REFCOUNT_BITS)) { 4875 refcount_bits = qemu_opt_get_number(opts, BLOCK_OPT_REFCOUNT_BITS, 4876 refcount_bits); 4877 4878 if (refcount_bits <= 0 || refcount_bits > 64 || 4879 !is_power_of_2(refcount_bits)) 4880 { 4881 error_setg(errp, "Refcount width must be a power of two and " 4882 "may not exceed 64 bits"); 4883 return -EINVAL; 4884 } 4885 } else if (!strcmp(desc->name, BLOCK_OPT_DATA_FILE)) { 4886 data_file = qemu_opt_get(opts, BLOCK_OPT_DATA_FILE); 4887 if (data_file && !has_data_file(bs)) { 4888 error_setg(errp, "data-file can only be set for images that " 4889 "use an external data file"); 4890 return -EINVAL; 4891 } 4892 } else if (!strcmp(desc->name, BLOCK_OPT_DATA_FILE_RAW)) { 4893 data_file_raw = qemu_opt_get_bool(opts, BLOCK_OPT_DATA_FILE_RAW, 4894 data_file_raw); 4895 if (data_file_raw && !data_file_is_raw(bs)) { 4896 error_setg(errp, "data-file-raw cannot be set on existing " 4897 "images"); 4898 return -EINVAL; 4899 } 4900 } else { 4901 /* if this point is reached, this probably means a new option was 4902 * added without having it covered here */ 4903 abort(); 4904 } 4905 4906 desc++; 4907 } 4908 4909 helper_cb_info = (Qcow2AmendHelperCBInfo){ 4910 .original_status_cb = status_cb, 4911 .original_cb_opaque = cb_opaque, 4912 .total_operations = (new_version < old_version) 4913 + (s->refcount_bits != refcount_bits) 4914 }; 4915 4916 /* Upgrade first (some features may require compat=1.1) */ 4917 if (new_version > old_version) { 4918 s->qcow_version = new_version; 4919 ret = qcow2_update_header(bs); 4920 if (ret < 0) { 4921 s->qcow_version = old_version; 4922 error_setg_errno(errp, -ret, "Failed to update the image header"); 4923 return ret; 4924 } 4925 } 4926 4927 if (s->refcount_bits != refcount_bits) { 4928 int refcount_order = ctz32(refcount_bits); 4929 4930 if (new_version < 3 && refcount_bits != 16) { 4931 error_setg(errp, "Refcount widths other than 16 bits require " 4932 "compatibility level 1.1 or above (use compat=1.1 or " 4933 "greater)"); 4934 return -EINVAL; 4935 } 4936 4937 helper_cb_info.current_operation = QCOW2_CHANGING_REFCOUNT_ORDER; 4938 ret = qcow2_change_refcount_order(bs, refcount_order, 4939 &qcow2_amend_helper_cb, 4940 &helper_cb_info, errp); 4941 if (ret < 0) { 4942 return ret; 4943 } 4944 } 4945 4946 /* data-file-raw blocks backing files, so clear it first if requested */ 4947 if (data_file_raw) { 4948 s->autoclear_features |= QCOW2_AUTOCLEAR_DATA_FILE_RAW; 4949 } else { 4950 s->autoclear_features &= ~QCOW2_AUTOCLEAR_DATA_FILE_RAW; 4951 } 4952 4953 if (data_file) { 4954 g_free(s->image_data_file); 4955 s->image_data_file = *data_file ? g_strdup(data_file) : NULL; 4956 } 4957 4958 ret = qcow2_update_header(bs); 4959 if (ret < 0) { 4960 error_setg_errno(errp, -ret, "Failed to update the image header"); 4961 return ret; 4962 } 4963 4964 if (backing_file || backing_format) { 4965 ret = qcow2_change_backing_file(bs, 4966 backing_file ?: s->image_backing_file, 4967 backing_format ?: s->image_backing_format); 4968 if (ret < 0) { 4969 error_setg_errno(errp, -ret, "Failed to change the backing file"); 4970 return ret; 4971 } 4972 } 4973 4974 if (s->use_lazy_refcounts != lazy_refcounts) { 4975 if (lazy_refcounts) { 4976 if (new_version < 3) { 4977 error_setg(errp, "Lazy refcounts only supported with " 4978 "compatibility level 1.1 and above (use compat=1.1 " 4979 "or greater)"); 4980 return -EINVAL; 4981 } 4982 s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS; 4983 ret = qcow2_update_header(bs); 4984 if (ret < 0) { 4985 s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS; 4986 error_setg_errno(errp, -ret, "Failed to update the image header"); 4987 return ret; 4988 } 4989 s->use_lazy_refcounts = true; 4990 } else { 4991 /* make image clean first */ 4992 ret = qcow2_mark_clean(bs); 4993 if (ret < 0) { 4994 error_setg_errno(errp, -ret, "Failed to make the image clean"); 4995 return ret; 4996 } 4997 /* now disallow lazy refcounts */ 4998 s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS; 4999 ret = qcow2_update_header(bs); 5000 if (ret < 0) { 5001 s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS; 5002 error_setg_errno(errp, -ret, "Failed to update the image header"); 5003 return ret; 5004 } 5005 s->use_lazy_refcounts = false; 5006 } 5007 } 5008 5009 if (new_size) { 5010 BlockBackend *blk = blk_new(bdrv_get_aio_context(bs), 5011 BLK_PERM_RESIZE, BLK_PERM_ALL); 5012 ret = blk_insert_bs(blk, bs, errp); 5013 if (ret < 0) { 5014 blk_unref(blk); 5015 return ret; 5016 } 5017 5018 ret = blk_truncate(blk, new_size, PREALLOC_MODE_OFF, errp); 5019 blk_unref(blk); 5020 if (ret < 0) { 5021 return ret; 5022 } 5023 } 5024 5025 /* Downgrade last (so unsupported features can be removed before) */ 5026 if (new_version < old_version) { 5027 helper_cb_info.current_operation = QCOW2_DOWNGRADING; 5028 ret = qcow2_downgrade(bs, new_version, &qcow2_amend_helper_cb, 5029 &helper_cb_info, errp); 5030 if (ret < 0) { 5031 return ret; 5032 } 5033 } 5034 5035 return 0; 5036 } 5037 5038 /* 5039 * If offset or size are negative, respectively, they will not be included in 5040 * the BLOCK_IMAGE_CORRUPTED event emitted. 5041 * fatal will be ignored for read-only BDS; corruptions found there will always 5042 * be considered non-fatal. 5043 */ 5044 void qcow2_signal_corruption(BlockDriverState *bs, bool fatal, int64_t offset, 5045 int64_t size, const char *message_format, ...) 5046 { 5047 BDRVQcow2State *s = bs->opaque; 5048 const char *node_name; 5049 char *message; 5050 va_list ap; 5051 5052 fatal = fatal && bdrv_is_writable(bs); 5053 5054 if (s->signaled_corruption && 5055 (!fatal || (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT))) 5056 { 5057 return; 5058 } 5059 5060 va_start(ap, message_format); 5061 message = g_strdup_vprintf(message_format, ap); 5062 va_end(ap); 5063 5064 if (fatal) { 5065 fprintf(stderr, "qcow2: Marking image as corrupt: %s; further " 5066 "corruption events will be suppressed\n", message); 5067 } else { 5068 fprintf(stderr, "qcow2: Image is corrupt: %s; further non-fatal " 5069 "corruption events will be suppressed\n", message); 5070 } 5071 5072 node_name = bdrv_get_node_name(bs); 5073 qapi_event_send_block_image_corrupted(bdrv_get_device_name(bs), 5074 *node_name != '\0', node_name, 5075 message, offset >= 0, offset, 5076 size >= 0, size, 5077 fatal); 5078 g_free(message); 5079 5080 if (fatal) { 5081 qcow2_mark_corrupt(bs); 5082 bs->drv = NULL; /* make BDS unusable */ 5083 } 5084 5085 s->signaled_corruption = true; 5086 } 5087 5088 static QemuOptsList qcow2_create_opts = { 5089 .name = "qcow2-create-opts", 5090 .head = QTAILQ_HEAD_INITIALIZER(qcow2_create_opts.head), 5091 .desc = { 5092 { 5093 .name = BLOCK_OPT_SIZE, 5094 .type = QEMU_OPT_SIZE, 5095 .help = "Virtual disk size" 5096 }, 5097 { 5098 .name = BLOCK_OPT_COMPAT_LEVEL, 5099 .type = QEMU_OPT_STRING, 5100 .help = "Compatibility level (0.10 or 1.1)" 5101 }, 5102 { 5103 .name = BLOCK_OPT_BACKING_FILE, 5104 .type = QEMU_OPT_STRING, 5105 .help = "File name of a base image" 5106 }, 5107 { 5108 .name = BLOCK_OPT_BACKING_FMT, 5109 .type = QEMU_OPT_STRING, 5110 .help = "Image format of the base image" 5111 }, 5112 { 5113 .name = BLOCK_OPT_DATA_FILE, 5114 .type = QEMU_OPT_STRING, 5115 .help = "File name of an external data file" 5116 }, 5117 { 5118 .name = BLOCK_OPT_DATA_FILE_RAW, 5119 .type = QEMU_OPT_BOOL, 5120 .help = "The external data file must stay valid as a raw image" 5121 }, 5122 { 5123 .name = BLOCK_OPT_ENCRYPT, 5124 .type = QEMU_OPT_BOOL, 5125 .help = "Encrypt the image with format 'aes'. (Deprecated " 5126 "in favor of " BLOCK_OPT_ENCRYPT_FORMAT "=aes)", 5127 }, 5128 { 5129 .name = BLOCK_OPT_ENCRYPT_FORMAT, 5130 .type = QEMU_OPT_STRING, 5131 .help = "Encrypt the image, format choices: 'aes', 'luks'", 5132 }, 5133 BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.", 5134 "ID of secret providing qcow AES key or LUKS passphrase"), 5135 BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_ALG("encrypt."), 5136 BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_MODE("encrypt."), 5137 BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_ALG("encrypt."), 5138 BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_HASH_ALG("encrypt."), 5139 BLOCK_CRYPTO_OPT_DEF_LUKS_HASH_ALG("encrypt."), 5140 BLOCK_CRYPTO_OPT_DEF_LUKS_ITER_TIME("encrypt."), 5141 { 5142 .name = BLOCK_OPT_CLUSTER_SIZE, 5143 .type = QEMU_OPT_SIZE, 5144 .help = "qcow2 cluster size", 5145 .def_value_str = stringify(DEFAULT_CLUSTER_SIZE) 5146 }, 5147 { 5148 .name = BLOCK_OPT_PREALLOC, 5149 .type = QEMU_OPT_STRING, 5150 .help = "Preallocation mode (allowed values: off, metadata, " 5151 "falloc, full)" 5152 }, 5153 { 5154 .name = BLOCK_OPT_LAZY_REFCOUNTS, 5155 .type = QEMU_OPT_BOOL, 5156 .help = "Postpone refcount updates", 5157 .def_value_str = "off" 5158 }, 5159 { 5160 .name = BLOCK_OPT_REFCOUNT_BITS, 5161 .type = QEMU_OPT_NUMBER, 5162 .help = "Width of a reference count entry in bits", 5163 .def_value_str = "16" 5164 }, 5165 { /* end of list */ } 5166 } 5167 }; 5168 5169 static const char *const qcow2_strong_runtime_opts[] = { 5170 "encrypt." BLOCK_CRYPTO_OPT_QCOW_KEY_SECRET, 5171 5172 NULL 5173 }; 5174 5175 BlockDriver bdrv_qcow2 = { 5176 .format_name = "qcow2", 5177 .instance_size = sizeof(BDRVQcow2State), 5178 .bdrv_probe = qcow2_probe, 5179 .bdrv_open = qcow2_open, 5180 .bdrv_close = qcow2_close, 5181 .bdrv_reopen_prepare = qcow2_reopen_prepare, 5182 .bdrv_reopen_commit = qcow2_reopen_commit, 5183 .bdrv_reopen_abort = qcow2_reopen_abort, 5184 .bdrv_join_options = qcow2_join_options, 5185 .bdrv_child_perm = bdrv_format_default_perms, 5186 .bdrv_co_create_opts = qcow2_co_create_opts, 5187 .bdrv_co_create = qcow2_co_create, 5188 .bdrv_has_zero_init = bdrv_has_zero_init_1, 5189 .bdrv_co_block_status = qcow2_co_block_status, 5190 5191 .bdrv_co_preadv = qcow2_co_preadv, 5192 .bdrv_co_pwritev = qcow2_co_pwritev, 5193 .bdrv_co_flush_to_os = qcow2_co_flush_to_os, 5194 5195 .bdrv_co_pwrite_zeroes = qcow2_co_pwrite_zeroes, 5196 .bdrv_co_pdiscard = qcow2_co_pdiscard, 5197 .bdrv_co_copy_range_from = qcow2_co_copy_range_from, 5198 .bdrv_co_copy_range_to = qcow2_co_copy_range_to, 5199 .bdrv_co_truncate = qcow2_co_truncate, 5200 .bdrv_co_pwritev_compressed = qcow2_co_pwritev_compressed, 5201 .bdrv_make_empty = qcow2_make_empty, 5202 5203 .bdrv_snapshot_create = qcow2_snapshot_create, 5204 .bdrv_snapshot_goto = qcow2_snapshot_goto, 5205 .bdrv_snapshot_delete = qcow2_snapshot_delete, 5206 .bdrv_snapshot_list = qcow2_snapshot_list, 5207 .bdrv_snapshot_load_tmp = qcow2_snapshot_load_tmp, 5208 .bdrv_measure = qcow2_measure, 5209 .bdrv_get_info = qcow2_get_info, 5210 .bdrv_get_specific_info = qcow2_get_specific_info, 5211 5212 .bdrv_save_vmstate = qcow2_save_vmstate, 5213 .bdrv_load_vmstate = qcow2_load_vmstate, 5214 5215 .supports_backing = true, 5216 .bdrv_change_backing_file = qcow2_change_backing_file, 5217 5218 .bdrv_refresh_limits = qcow2_refresh_limits, 5219 .bdrv_co_invalidate_cache = qcow2_co_invalidate_cache, 5220 .bdrv_inactivate = qcow2_inactivate, 5221 5222 .create_opts = &qcow2_create_opts, 5223 .strong_runtime_opts = qcow2_strong_runtime_opts, 5224 .mutable_opts = mutable_opts, 5225 .bdrv_co_check = qcow2_co_check, 5226 .bdrv_amend_options = qcow2_amend_options, 5227 5228 .bdrv_detach_aio_context = qcow2_detach_aio_context, 5229 .bdrv_attach_aio_context = qcow2_attach_aio_context, 5230 5231 .bdrv_reopen_bitmaps_rw = qcow2_reopen_bitmaps_rw, 5232 .bdrv_can_store_new_dirty_bitmap = qcow2_can_store_new_dirty_bitmap, 5233 .bdrv_remove_persistent_dirty_bitmap = qcow2_remove_persistent_dirty_bitmap, 5234 }; 5235 5236 static void bdrv_qcow2_init(void) 5237 { 5238 bdrv_register(&bdrv_qcow2); 5239 } 5240 5241 block_init(bdrv_qcow2_init); 5242