1 /* 2 * Block driver for the QCOW version 2 format 3 * 4 * Copyright (c) 2004-2006 Fabrice Bellard 5 * 6 * Permission is hereby granted, free of charge, to any person obtaining a copy 7 * of this software and associated documentation files (the "Software"), to deal 8 * in the Software without restriction, including without limitation the rights 9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 10 * copies of the Software, and to permit persons to whom the Software is 11 * furnished to do so, subject to the following conditions: 12 * 13 * The above copyright notice and this permission notice shall be included in 14 * all copies or substantial portions of the Software. 15 * 16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN 22 * THE SOFTWARE. 23 */ 24 #include "qemu/osdep.h" 25 #include "block/block_int.h" 26 #include "sysemu/block-backend.h" 27 #include "qemu/module.h" 28 #include <zlib.h> 29 #include "block/qcow2.h" 30 #include "qemu/error-report.h" 31 #include "qapi/qmp/qerror.h" 32 #include "qapi/qmp/qbool.h" 33 #include "qapi/qmp/types.h" 34 #include "qapi-event.h" 35 #include "trace.h" 36 #include "qemu/option_int.h" 37 #include "qemu/cutils.h" 38 #include "qemu/bswap.h" 39 #include "qapi/opts-visitor.h" 40 #include "qapi-visit.h" 41 #include "block/crypto.h" 42 43 /* 44 Differences with QCOW: 45 46 - Support for multiple incremental snapshots. 47 - Memory management by reference counts. 48 - Clusters which have a reference count of one have the bit 49 QCOW_OFLAG_COPIED to optimize write performance. 50 - Size of compressed clusters is stored in sectors to reduce bit usage 51 in the cluster offsets. 52 - Support for storing additional data (such as the VM state) in the 53 snapshots. 54 - If a backing store is used, the cluster size is not constrained 55 (could be backported to QCOW). 56 - L2 tables have always a size of one cluster. 57 */ 58 59 60 typedef struct { 61 uint32_t magic; 62 uint32_t len; 63 } QEMU_PACKED QCowExtension; 64 65 #define QCOW2_EXT_MAGIC_END 0 66 #define QCOW2_EXT_MAGIC_BACKING_FORMAT 0xE2792ACA 67 #define QCOW2_EXT_MAGIC_FEATURE_TABLE 0x6803f857 68 #define QCOW2_EXT_MAGIC_CRYPTO_HEADER 0x0537be77 69 #define QCOW2_EXT_MAGIC_BITMAPS 0x23852875 70 71 static int qcow2_probe(const uint8_t *buf, int buf_size, const char *filename) 72 { 73 const QCowHeader *cow_header = (const void *)buf; 74 75 if (buf_size >= sizeof(QCowHeader) && 76 be32_to_cpu(cow_header->magic) == QCOW_MAGIC && 77 be32_to_cpu(cow_header->version) >= 2) 78 return 100; 79 else 80 return 0; 81 } 82 83 84 static ssize_t qcow2_crypto_hdr_read_func(QCryptoBlock *block, size_t offset, 85 uint8_t *buf, size_t buflen, 86 void *opaque, Error **errp) 87 { 88 BlockDriverState *bs = opaque; 89 BDRVQcow2State *s = bs->opaque; 90 ssize_t ret; 91 92 if ((offset + buflen) > s->crypto_header.length) { 93 error_setg(errp, "Request for data outside of extension header"); 94 return -1; 95 } 96 97 ret = bdrv_pread(bs->file, 98 s->crypto_header.offset + offset, buf, buflen); 99 if (ret < 0) { 100 error_setg_errno(errp, -ret, "Could not read encryption header"); 101 return -1; 102 } 103 return ret; 104 } 105 106 107 static ssize_t qcow2_crypto_hdr_init_func(QCryptoBlock *block, size_t headerlen, 108 void *opaque, Error **errp) 109 { 110 BlockDriverState *bs = opaque; 111 BDRVQcow2State *s = bs->opaque; 112 int64_t ret; 113 int64_t clusterlen; 114 115 ret = qcow2_alloc_clusters(bs, headerlen); 116 if (ret < 0) { 117 error_setg_errno(errp, -ret, 118 "Cannot allocate cluster for LUKS header size %zu", 119 headerlen); 120 return -1; 121 } 122 123 s->crypto_header.length = headerlen; 124 s->crypto_header.offset = ret; 125 126 /* Zero fill remaining space in cluster so it has predictable 127 * content in case of future spec changes */ 128 clusterlen = size_to_clusters(s, headerlen) * s->cluster_size; 129 assert(qcow2_pre_write_overlap_check(bs, 0, ret, clusterlen) == 0); 130 ret = bdrv_pwrite_zeroes(bs->file, 131 ret + headerlen, 132 clusterlen - headerlen, 0); 133 if (ret < 0) { 134 error_setg_errno(errp, -ret, "Could not zero fill encryption header"); 135 return -1; 136 } 137 138 return ret; 139 } 140 141 142 static ssize_t qcow2_crypto_hdr_write_func(QCryptoBlock *block, size_t offset, 143 const uint8_t *buf, size_t buflen, 144 void *opaque, Error **errp) 145 { 146 BlockDriverState *bs = opaque; 147 BDRVQcow2State *s = bs->opaque; 148 ssize_t ret; 149 150 if ((offset + buflen) > s->crypto_header.length) { 151 error_setg(errp, "Request for data outside of extension header"); 152 return -1; 153 } 154 155 ret = bdrv_pwrite(bs->file, 156 s->crypto_header.offset + offset, buf, buflen); 157 if (ret < 0) { 158 error_setg_errno(errp, -ret, "Could not read encryption header"); 159 return -1; 160 } 161 return ret; 162 } 163 164 165 /* 166 * read qcow2 extension and fill bs 167 * start reading from start_offset 168 * finish reading upon magic of value 0 or when end_offset reached 169 * unknown magic is skipped (future extension this version knows nothing about) 170 * return 0 upon success, non-0 otherwise 171 */ 172 static int qcow2_read_extensions(BlockDriverState *bs, uint64_t start_offset, 173 uint64_t end_offset, void **p_feature_table, 174 int flags, bool *need_update_header, 175 Error **errp) 176 { 177 BDRVQcow2State *s = bs->opaque; 178 QCowExtension ext; 179 uint64_t offset; 180 int ret; 181 Qcow2BitmapHeaderExt bitmaps_ext; 182 183 if (need_update_header != NULL) { 184 *need_update_header = false; 185 } 186 187 #ifdef DEBUG_EXT 188 printf("qcow2_read_extensions: start=%ld end=%ld\n", start_offset, end_offset); 189 #endif 190 offset = start_offset; 191 while (offset < end_offset) { 192 193 #ifdef DEBUG_EXT 194 /* Sanity check */ 195 if (offset > s->cluster_size) 196 printf("qcow2_read_extension: suspicious offset %lu\n", offset); 197 198 printf("attempting to read extended header in offset %lu\n", offset); 199 #endif 200 201 ret = bdrv_pread(bs->file, offset, &ext, sizeof(ext)); 202 if (ret < 0) { 203 error_setg_errno(errp, -ret, "qcow2_read_extension: ERROR: " 204 "pread fail from offset %" PRIu64, offset); 205 return 1; 206 } 207 be32_to_cpus(&ext.magic); 208 be32_to_cpus(&ext.len); 209 offset += sizeof(ext); 210 #ifdef DEBUG_EXT 211 printf("ext.magic = 0x%x\n", ext.magic); 212 #endif 213 if (offset > end_offset || ext.len > end_offset - offset) { 214 error_setg(errp, "Header extension too large"); 215 return -EINVAL; 216 } 217 218 switch (ext.magic) { 219 case QCOW2_EXT_MAGIC_END: 220 return 0; 221 222 case QCOW2_EXT_MAGIC_BACKING_FORMAT: 223 if (ext.len >= sizeof(bs->backing_format)) { 224 error_setg(errp, "ERROR: ext_backing_format: len=%" PRIu32 225 " too large (>=%zu)", ext.len, 226 sizeof(bs->backing_format)); 227 return 2; 228 } 229 ret = bdrv_pread(bs->file, offset, bs->backing_format, ext.len); 230 if (ret < 0) { 231 error_setg_errno(errp, -ret, "ERROR: ext_backing_format: " 232 "Could not read format name"); 233 return 3; 234 } 235 bs->backing_format[ext.len] = '\0'; 236 s->image_backing_format = g_strdup(bs->backing_format); 237 #ifdef DEBUG_EXT 238 printf("Qcow2: Got format extension %s\n", bs->backing_format); 239 #endif 240 break; 241 242 case QCOW2_EXT_MAGIC_FEATURE_TABLE: 243 if (p_feature_table != NULL) { 244 void* feature_table = g_malloc0(ext.len + 2 * sizeof(Qcow2Feature)); 245 ret = bdrv_pread(bs->file, offset , feature_table, ext.len); 246 if (ret < 0) { 247 error_setg_errno(errp, -ret, "ERROR: ext_feature_table: " 248 "Could not read table"); 249 return ret; 250 } 251 252 *p_feature_table = feature_table; 253 } 254 break; 255 256 case QCOW2_EXT_MAGIC_CRYPTO_HEADER: { 257 unsigned int cflags = 0; 258 if (s->crypt_method_header != QCOW_CRYPT_LUKS) { 259 error_setg(errp, "CRYPTO header extension only " 260 "expected with LUKS encryption method"); 261 return -EINVAL; 262 } 263 if (ext.len != sizeof(Qcow2CryptoHeaderExtension)) { 264 error_setg(errp, "CRYPTO header extension size %u, " 265 "but expected size %zu", ext.len, 266 sizeof(Qcow2CryptoHeaderExtension)); 267 return -EINVAL; 268 } 269 270 ret = bdrv_pread(bs->file, offset, &s->crypto_header, ext.len); 271 if (ret < 0) { 272 error_setg_errno(errp, -ret, 273 "Unable to read CRYPTO header extension"); 274 return ret; 275 } 276 be64_to_cpus(&s->crypto_header.offset); 277 be64_to_cpus(&s->crypto_header.length); 278 279 if ((s->crypto_header.offset % s->cluster_size) != 0) { 280 error_setg(errp, "Encryption header offset '%" PRIu64 "' is " 281 "not a multiple of cluster size '%u'", 282 s->crypto_header.offset, s->cluster_size); 283 return -EINVAL; 284 } 285 286 if (flags & BDRV_O_NO_IO) { 287 cflags |= QCRYPTO_BLOCK_OPEN_NO_IO; 288 } 289 s->crypto = qcrypto_block_open(s->crypto_opts, "encrypt.", 290 qcow2_crypto_hdr_read_func, 291 bs, cflags, errp); 292 if (!s->crypto) { 293 return -EINVAL; 294 } 295 } break; 296 297 case QCOW2_EXT_MAGIC_BITMAPS: 298 if (ext.len != sizeof(bitmaps_ext)) { 299 error_setg_errno(errp, -ret, "bitmaps_ext: " 300 "Invalid extension length"); 301 return -EINVAL; 302 } 303 304 if (!(s->autoclear_features & QCOW2_AUTOCLEAR_BITMAPS)) { 305 warn_report("a program lacking bitmap support " 306 "modified this file, so all bitmaps are now " 307 "considered inconsistent"); 308 error_printf("Some clusters may be leaked, " 309 "run 'qemu-img check -r' on the image " 310 "file to fix."); 311 if (need_update_header != NULL) { 312 /* Updating is needed to drop invalid bitmap extension. */ 313 *need_update_header = true; 314 } 315 break; 316 } 317 318 ret = bdrv_pread(bs->file, offset, &bitmaps_ext, ext.len); 319 if (ret < 0) { 320 error_setg_errno(errp, -ret, "bitmaps_ext: " 321 "Could not read ext header"); 322 return ret; 323 } 324 325 if (bitmaps_ext.reserved32 != 0) { 326 error_setg_errno(errp, -ret, "bitmaps_ext: " 327 "Reserved field is not zero"); 328 return -EINVAL; 329 } 330 331 be32_to_cpus(&bitmaps_ext.nb_bitmaps); 332 be64_to_cpus(&bitmaps_ext.bitmap_directory_size); 333 be64_to_cpus(&bitmaps_ext.bitmap_directory_offset); 334 335 if (bitmaps_ext.nb_bitmaps > QCOW2_MAX_BITMAPS) { 336 error_setg(errp, 337 "bitmaps_ext: Image has %" PRIu32 " bitmaps, " 338 "exceeding the QEMU supported maximum of %d", 339 bitmaps_ext.nb_bitmaps, QCOW2_MAX_BITMAPS); 340 return -EINVAL; 341 } 342 343 if (bitmaps_ext.nb_bitmaps == 0) { 344 error_setg(errp, "found bitmaps extension with zero bitmaps"); 345 return -EINVAL; 346 } 347 348 if (bitmaps_ext.bitmap_directory_offset & (s->cluster_size - 1)) { 349 error_setg(errp, "bitmaps_ext: " 350 "invalid bitmap directory offset"); 351 return -EINVAL; 352 } 353 354 if (bitmaps_ext.bitmap_directory_size > 355 QCOW2_MAX_BITMAP_DIRECTORY_SIZE) { 356 error_setg(errp, "bitmaps_ext: " 357 "bitmap directory size (%" PRIu64 ") exceeds " 358 "the maximum supported size (%d)", 359 bitmaps_ext.bitmap_directory_size, 360 QCOW2_MAX_BITMAP_DIRECTORY_SIZE); 361 return -EINVAL; 362 } 363 364 s->nb_bitmaps = bitmaps_ext.nb_bitmaps; 365 s->bitmap_directory_offset = 366 bitmaps_ext.bitmap_directory_offset; 367 s->bitmap_directory_size = 368 bitmaps_ext.bitmap_directory_size; 369 370 #ifdef DEBUG_EXT 371 printf("Qcow2: Got bitmaps extension: " 372 "offset=%" PRIu64 " nb_bitmaps=%" PRIu32 "\n", 373 s->bitmap_directory_offset, s->nb_bitmaps); 374 #endif 375 break; 376 377 default: 378 /* unknown magic - save it in case we need to rewrite the header */ 379 /* If you add a new feature, make sure to also update the fast 380 * path of qcow2_make_empty() to deal with it. */ 381 { 382 Qcow2UnknownHeaderExtension *uext; 383 384 uext = g_malloc0(sizeof(*uext) + ext.len); 385 uext->magic = ext.magic; 386 uext->len = ext.len; 387 QLIST_INSERT_HEAD(&s->unknown_header_ext, uext, next); 388 389 ret = bdrv_pread(bs->file, offset , uext->data, uext->len); 390 if (ret < 0) { 391 error_setg_errno(errp, -ret, "ERROR: unknown extension: " 392 "Could not read data"); 393 return ret; 394 } 395 } 396 break; 397 } 398 399 offset += ((ext.len + 7) & ~7); 400 } 401 402 return 0; 403 } 404 405 static void cleanup_unknown_header_ext(BlockDriverState *bs) 406 { 407 BDRVQcow2State *s = bs->opaque; 408 Qcow2UnknownHeaderExtension *uext, *next; 409 410 QLIST_FOREACH_SAFE(uext, &s->unknown_header_ext, next, next) { 411 QLIST_REMOVE(uext, next); 412 g_free(uext); 413 } 414 } 415 416 static void report_unsupported_feature(Error **errp, Qcow2Feature *table, 417 uint64_t mask) 418 { 419 char *features = g_strdup(""); 420 char *old; 421 422 while (table && table->name[0] != '\0') { 423 if (table->type == QCOW2_FEAT_TYPE_INCOMPATIBLE) { 424 if (mask & (1ULL << table->bit)) { 425 old = features; 426 features = g_strdup_printf("%s%s%.46s", old, *old ? ", " : "", 427 table->name); 428 g_free(old); 429 mask &= ~(1ULL << table->bit); 430 } 431 } 432 table++; 433 } 434 435 if (mask) { 436 old = features; 437 features = g_strdup_printf("%s%sUnknown incompatible feature: %" PRIx64, 438 old, *old ? ", " : "", mask); 439 g_free(old); 440 } 441 442 error_setg(errp, "Unsupported qcow2 feature(s): %s", features); 443 g_free(features); 444 } 445 446 /* 447 * Sets the dirty bit and flushes afterwards if necessary. 448 * 449 * The incompatible_features bit is only set if the image file header was 450 * updated successfully. Therefore it is not required to check the return 451 * value of this function. 452 */ 453 int qcow2_mark_dirty(BlockDriverState *bs) 454 { 455 BDRVQcow2State *s = bs->opaque; 456 uint64_t val; 457 int ret; 458 459 assert(s->qcow_version >= 3); 460 461 if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) { 462 return 0; /* already dirty */ 463 } 464 465 val = cpu_to_be64(s->incompatible_features | QCOW2_INCOMPAT_DIRTY); 466 ret = bdrv_pwrite(bs->file, offsetof(QCowHeader, incompatible_features), 467 &val, sizeof(val)); 468 if (ret < 0) { 469 return ret; 470 } 471 ret = bdrv_flush(bs->file->bs); 472 if (ret < 0) { 473 return ret; 474 } 475 476 /* Only treat image as dirty if the header was updated successfully */ 477 s->incompatible_features |= QCOW2_INCOMPAT_DIRTY; 478 return 0; 479 } 480 481 /* 482 * Clears the dirty bit and flushes before if necessary. Only call this 483 * function when there are no pending requests, it does not guard against 484 * concurrent requests dirtying the image. 485 */ 486 static int qcow2_mark_clean(BlockDriverState *bs) 487 { 488 BDRVQcow2State *s = bs->opaque; 489 490 if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) { 491 int ret; 492 493 s->incompatible_features &= ~QCOW2_INCOMPAT_DIRTY; 494 495 ret = bdrv_flush(bs); 496 if (ret < 0) { 497 return ret; 498 } 499 500 return qcow2_update_header(bs); 501 } 502 return 0; 503 } 504 505 /* 506 * Marks the image as corrupt. 507 */ 508 int qcow2_mark_corrupt(BlockDriverState *bs) 509 { 510 BDRVQcow2State *s = bs->opaque; 511 512 s->incompatible_features |= QCOW2_INCOMPAT_CORRUPT; 513 return qcow2_update_header(bs); 514 } 515 516 /* 517 * Marks the image as consistent, i.e., unsets the corrupt bit, and flushes 518 * before if necessary. 519 */ 520 int qcow2_mark_consistent(BlockDriverState *bs) 521 { 522 BDRVQcow2State *s = bs->opaque; 523 524 if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) { 525 int ret = bdrv_flush(bs); 526 if (ret < 0) { 527 return ret; 528 } 529 530 s->incompatible_features &= ~QCOW2_INCOMPAT_CORRUPT; 531 return qcow2_update_header(bs); 532 } 533 return 0; 534 } 535 536 static int qcow2_check(BlockDriverState *bs, BdrvCheckResult *result, 537 BdrvCheckMode fix) 538 { 539 int ret = qcow2_check_refcounts(bs, result, fix); 540 if (ret < 0) { 541 return ret; 542 } 543 544 if (fix && result->check_errors == 0 && result->corruptions == 0) { 545 ret = qcow2_mark_clean(bs); 546 if (ret < 0) { 547 return ret; 548 } 549 return qcow2_mark_consistent(bs); 550 } 551 return ret; 552 } 553 554 static int validate_table_offset(BlockDriverState *bs, uint64_t offset, 555 uint64_t entries, size_t entry_len) 556 { 557 BDRVQcow2State *s = bs->opaque; 558 uint64_t size; 559 560 /* Use signed INT64_MAX as the maximum even for uint64_t header fields, 561 * because values will be passed to qemu functions taking int64_t. */ 562 if (entries > INT64_MAX / entry_len) { 563 return -EINVAL; 564 } 565 566 size = entries * entry_len; 567 568 if (INT64_MAX - size < offset) { 569 return -EINVAL; 570 } 571 572 /* Tables must be cluster aligned */ 573 if (offset_into_cluster(s, offset) != 0) { 574 return -EINVAL; 575 } 576 577 return 0; 578 } 579 580 static QemuOptsList qcow2_runtime_opts = { 581 .name = "qcow2", 582 .head = QTAILQ_HEAD_INITIALIZER(qcow2_runtime_opts.head), 583 .desc = { 584 { 585 .name = QCOW2_OPT_LAZY_REFCOUNTS, 586 .type = QEMU_OPT_BOOL, 587 .help = "Postpone refcount updates", 588 }, 589 { 590 .name = QCOW2_OPT_DISCARD_REQUEST, 591 .type = QEMU_OPT_BOOL, 592 .help = "Pass guest discard requests to the layer below", 593 }, 594 { 595 .name = QCOW2_OPT_DISCARD_SNAPSHOT, 596 .type = QEMU_OPT_BOOL, 597 .help = "Generate discard requests when snapshot related space " 598 "is freed", 599 }, 600 { 601 .name = QCOW2_OPT_DISCARD_OTHER, 602 .type = QEMU_OPT_BOOL, 603 .help = "Generate discard requests when other clusters are freed", 604 }, 605 { 606 .name = QCOW2_OPT_OVERLAP, 607 .type = QEMU_OPT_STRING, 608 .help = "Selects which overlap checks to perform from a range of " 609 "templates (none, constant, cached, all)", 610 }, 611 { 612 .name = QCOW2_OPT_OVERLAP_TEMPLATE, 613 .type = QEMU_OPT_STRING, 614 .help = "Selects which overlap checks to perform from a range of " 615 "templates (none, constant, cached, all)", 616 }, 617 { 618 .name = QCOW2_OPT_OVERLAP_MAIN_HEADER, 619 .type = QEMU_OPT_BOOL, 620 .help = "Check for unintended writes into the main qcow2 header", 621 }, 622 { 623 .name = QCOW2_OPT_OVERLAP_ACTIVE_L1, 624 .type = QEMU_OPT_BOOL, 625 .help = "Check for unintended writes into the active L1 table", 626 }, 627 { 628 .name = QCOW2_OPT_OVERLAP_ACTIVE_L2, 629 .type = QEMU_OPT_BOOL, 630 .help = "Check for unintended writes into an active L2 table", 631 }, 632 { 633 .name = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE, 634 .type = QEMU_OPT_BOOL, 635 .help = "Check for unintended writes into the refcount table", 636 }, 637 { 638 .name = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK, 639 .type = QEMU_OPT_BOOL, 640 .help = "Check for unintended writes into a refcount block", 641 }, 642 { 643 .name = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE, 644 .type = QEMU_OPT_BOOL, 645 .help = "Check for unintended writes into the snapshot table", 646 }, 647 { 648 .name = QCOW2_OPT_OVERLAP_INACTIVE_L1, 649 .type = QEMU_OPT_BOOL, 650 .help = "Check for unintended writes into an inactive L1 table", 651 }, 652 { 653 .name = QCOW2_OPT_OVERLAP_INACTIVE_L2, 654 .type = QEMU_OPT_BOOL, 655 .help = "Check for unintended writes into an inactive L2 table", 656 }, 657 { 658 .name = QCOW2_OPT_CACHE_SIZE, 659 .type = QEMU_OPT_SIZE, 660 .help = "Maximum combined metadata (L2 tables and refcount blocks) " 661 "cache size", 662 }, 663 { 664 .name = QCOW2_OPT_L2_CACHE_SIZE, 665 .type = QEMU_OPT_SIZE, 666 .help = "Maximum L2 table cache size", 667 }, 668 { 669 .name = QCOW2_OPT_REFCOUNT_CACHE_SIZE, 670 .type = QEMU_OPT_SIZE, 671 .help = "Maximum refcount block cache size", 672 }, 673 { 674 .name = QCOW2_OPT_CACHE_CLEAN_INTERVAL, 675 .type = QEMU_OPT_NUMBER, 676 .help = "Clean unused cache entries after this time (in seconds)", 677 }, 678 BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.", 679 "ID of secret providing qcow2 AES key or LUKS passphrase"), 680 { /* end of list */ } 681 }, 682 }; 683 684 static const char *overlap_bool_option_names[QCOW2_OL_MAX_BITNR] = { 685 [QCOW2_OL_MAIN_HEADER_BITNR] = QCOW2_OPT_OVERLAP_MAIN_HEADER, 686 [QCOW2_OL_ACTIVE_L1_BITNR] = QCOW2_OPT_OVERLAP_ACTIVE_L1, 687 [QCOW2_OL_ACTIVE_L2_BITNR] = QCOW2_OPT_OVERLAP_ACTIVE_L2, 688 [QCOW2_OL_REFCOUNT_TABLE_BITNR] = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE, 689 [QCOW2_OL_REFCOUNT_BLOCK_BITNR] = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK, 690 [QCOW2_OL_SNAPSHOT_TABLE_BITNR] = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE, 691 [QCOW2_OL_INACTIVE_L1_BITNR] = QCOW2_OPT_OVERLAP_INACTIVE_L1, 692 [QCOW2_OL_INACTIVE_L2_BITNR] = QCOW2_OPT_OVERLAP_INACTIVE_L2, 693 }; 694 695 static void cache_clean_timer_cb(void *opaque) 696 { 697 BlockDriverState *bs = opaque; 698 BDRVQcow2State *s = bs->opaque; 699 qcow2_cache_clean_unused(bs, s->l2_table_cache); 700 qcow2_cache_clean_unused(bs, s->refcount_block_cache); 701 timer_mod(s->cache_clean_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) + 702 (int64_t) s->cache_clean_interval * 1000); 703 } 704 705 static void cache_clean_timer_init(BlockDriverState *bs, AioContext *context) 706 { 707 BDRVQcow2State *s = bs->opaque; 708 if (s->cache_clean_interval > 0) { 709 s->cache_clean_timer = aio_timer_new(context, QEMU_CLOCK_VIRTUAL, 710 SCALE_MS, cache_clean_timer_cb, 711 bs); 712 timer_mod(s->cache_clean_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) + 713 (int64_t) s->cache_clean_interval * 1000); 714 } 715 } 716 717 static void cache_clean_timer_del(BlockDriverState *bs) 718 { 719 BDRVQcow2State *s = bs->opaque; 720 if (s->cache_clean_timer) { 721 timer_del(s->cache_clean_timer); 722 timer_free(s->cache_clean_timer); 723 s->cache_clean_timer = NULL; 724 } 725 } 726 727 static void qcow2_detach_aio_context(BlockDriverState *bs) 728 { 729 cache_clean_timer_del(bs); 730 } 731 732 static void qcow2_attach_aio_context(BlockDriverState *bs, 733 AioContext *new_context) 734 { 735 cache_clean_timer_init(bs, new_context); 736 } 737 738 static void read_cache_sizes(BlockDriverState *bs, QemuOpts *opts, 739 uint64_t *l2_cache_size, 740 uint64_t *refcount_cache_size, Error **errp) 741 { 742 BDRVQcow2State *s = bs->opaque; 743 uint64_t combined_cache_size; 744 bool l2_cache_size_set, refcount_cache_size_set, combined_cache_size_set; 745 746 combined_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_CACHE_SIZE); 747 l2_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_L2_CACHE_SIZE); 748 refcount_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_REFCOUNT_CACHE_SIZE); 749 750 combined_cache_size = qemu_opt_get_size(opts, QCOW2_OPT_CACHE_SIZE, 0); 751 *l2_cache_size = qemu_opt_get_size(opts, QCOW2_OPT_L2_CACHE_SIZE, 0); 752 *refcount_cache_size = qemu_opt_get_size(opts, 753 QCOW2_OPT_REFCOUNT_CACHE_SIZE, 0); 754 755 if (combined_cache_size_set) { 756 if (l2_cache_size_set && refcount_cache_size_set) { 757 error_setg(errp, QCOW2_OPT_CACHE_SIZE ", " QCOW2_OPT_L2_CACHE_SIZE 758 " and " QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not be set " 759 "the same time"); 760 return; 761 } else if (*l2_cache_size > combined_cache_size) { 762 error_setg(errp, QCOW2_OPT_L2_CACHE_SIZE " may not exceed " 763 QCOW2_OPT_CACHE_SIZE); 764 return; 765 } else if (*refcount_cache_size > combined_cache_size) { 766 error_setg(errp, QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not exceed " 767 QCOW2_OPT_CACHE_SIZE); 768 return; 769 } 770 771 if (l2_cache_size_set) { 772 *refcount_cache_size = combined_cache_size - *l2_cache_size; 773 } else if (refcount_cache_size_set) { 774 *l2_cache_size = combined_cache_size - *refcount_cache_size; 775 } else { 776 *refcount_cache_size = combined_cache_size 777 / (DEFAULT_L2_REFCOUNT_SIZE_RATIO + 1); 778 *l2_cache_size = combined_cache_size - *refcount_cache_size; 779 } 780 } else { 781 if (!l2_cache_size_set && !refcount_cache_size_set) { 782 *l2_cache_size = MAX(DEFAULT_L2_CACHE_BYTE_SIZE, 783 (uint64_t)DEFAULT_L2_CACHE_CLUSTERS 784 * s->cluster_size); 785 *refcount_cache_size = *l2_cache_size 786 / DEFAULT_L2_REFCOUNT_SIZE_RATIO; 787 } else if (!l2_cache_size_set) { 788 *l2_cache_size = *refcount_cache_size 789 * DEFAULT_L2_REFCOUNT_SIZE_RATIO; 790 } else if (!refcount_cache_size_set) { 791 *refcount_cache_size = *l2_cache_size 792 / DEFAULT_L2_REFCOUNT_SIZE_RATIO; 793 } 794 } 795 } 796 797 typedef struct Qcow2ReopenState { 798 Qcow2Cache *l2_table_cache; 799 Qcow2Cache *refcount_block_cache; 800 bool use_lazy_refcounts; 801 int overlap_check; 802 bool discard_passthrough[QCOW2_DISCARD_MAX]; 803 uint64_t cache_clean_interval; 804 QCryptoBlockOpenOptions *crypto_opts; /* Disk encryption runtime options */ 805 } Qcow2ReopenState; 806 807 static int qcow2_update_options_prepare(BlockDriverState *bs, 808 Qcow2ReopenState *r, 809 QDict *options, int flags, 810 Error **errp) 811 { 812 BDRVQcow2State *s = bs->opaque; 813 QemuOpts *opts = NULL; 814 const char *opt_overlap_check, *opt_overlap_check_template; 815 int overlap_check_template = 0; 816 uint64_t l2_cache_size, refcount_cache_size; 817 int i; 818 const char *encryptfmt; 819 QDict *encryptopts = NULL; 820 Error *local_err = NULL; 821 int ret; 822 823 qdict_extract_subqdict(options, &encryptopts, "encrypt."); 824 encryptfmt = qdict_get_try_str(encryptopts, "format"); 825 826 opts = qemu_opts_create(&qcow2_runtime_opts, NULL, 0, &error_abort); 827 qemu_opts_absorb_qdict(opts, options, &local_err); 828 if (local_err) { 829 error_propagate(errp, local_err); 830 ret = -EINVAL; 831 goto fail; 832 } 833 834 /* get L2 table/refcount block cache size from command line options */ 835 read_cache_sizes(bs, opts, &l2_cache_size, &refcount_cache_size, 836 &local_err); 837 if (local_err) { 838 error_propagate(errp, local_err); 839 ret = -EINVAL; 840 goto fail; 841 } 842 843 l2_cache_size /= s->cluster_size; 844 if (l2_cache_size < MIN_L2_CACHE_SIZE) { 845 l2_cache_size = MIN_L2_CACHE_SIZE; 846 } 847 if (l2_cache_size > INT_MAX) { 848 error_setg(errp, "L2 cache size too big"); 849 ret = -EINVAL; 850 goto fail; 851 } 852 853 refcount_cache_size /= s->cluster_size; 854 if (refcount_cache_size < MIN_REFCOUNT_CACHE_SIZE) { 855 refcount_cache_size = MIN_REFCOUNT_CACHE_SIZE; 856 } 857 if (refcount_cache_size > INT_MAX) { 858 error_setg(errp, "Refcount cache size too big"); 859 ret = -EINVAL; 860 goto fail; 861 } 862 863 /* alloc new L2 table/refcount block cache, flush old one */ 864 if (s->l2_table_cache) { 865 ret = qcow2_cache_flush(bs, s->l2_table_cache); 866 if (ret) { 867 error_setg_errno(errp, -ret, "Failed to flush the L2 table cache"); 868 goto fail; 869 } 870 } 871 872 if (s->refcount_block_cache) { 873 ret = qcow2_cache_flush(bs, s->refcount_block_cache); 874 if (ret) { 875 error_setg_errno(errp, -ret, 876 "Failed to flush the refcount block cache"); 877 goto fail; 878 } 879 } 880 881 r->l2_table_cache = qcow2_cache_create(bs, l2_cache_size); 882 r->refcount_block_cache = qcow2_cache_create(bs, refcount_cache_size); 883 if (r->l2_table_cache == NULL || r->refcount_block_cache == NULL) { 884 error_setg(errp, "Could not allocate metadata caches"); 885 ret = -ENOMEM; 886 goto fail; 887 } 888 889 /* New interval for cache cleanup timer */ 890 r->cache_clean_interval = 891 qemu_opt_get_number(opts, QCOW2_OPT_CACHE_CLEAN_INTERVAL, 892 s->cache_clean_interval); 893 #ifndef CONFIG_LINUX 894 if (r->cache_clean_interval != 0) { 895 error_setg(errp, QCOW2_OPT_CACHE_CLEAN_INTERVAL 896 " not supported on this host"); 897 ret = -EINVAL; 898 goto fail; 899 } 900 #endif 901 if (r->cache_clean_interval > UINT_MAX) { 902 error_setg(errp, "Cache clean interval too big"); 903 ret = -EINVAL; 904 goto fail; 905 } 906 907 /* lazy-refcounts; flush if going from enabled to disabled */ 908 r->use_lazy_refcounts = qemu_opt_get_bool(opts, QCOW2_OPT_LAZY_REFCOUNTS, 909 (s->compatible_features & QCOW2_COMPAT_LAZY_REFCOUNTS)); 910 if (r->use_lazy_refcounts && s->qcow_version < 3) { 911 error_setg(errp, "Lazy refcounts require a qcow2 image with at least " 912 "qemu 1.1 compatibility level"); 913 ret = -EINVAL; 914 goto fail; 915 } 916 917 if (s->use_lazy_refcounts && !r->use_lazy_refcounts) { 918 ret = qcow2_mark_clean(bs); 919 if (ret < 0) { 920 error_setg_errno(errp, -ret, "Failed to disable lazy refcounts"); 921 goto fail; 922 } 923 } 924 925 /* Overlap check options */ 926 opt_overlap_check = qemu_opt_get(opts, QCOW2_OPT_OVERLAP); 927 opt_overlap_check_template = qemu_opt_get(opts, QCOW2_OPT_OVERLAP_TEMPLATE); 928 if (opt_overlap_check_template && opt_overlap_check && 929 strcmp(opt_overlap_check_template, opt_overlap_check)) 930 { 931 error_setg(errp, "Conflicting values for qcow2 options '" 932 QCOW2_OPT_OVERLAP "' ('%s') and '" QCOW2_OPT_OVERLAP_TEMPLATE 933 "' ('%s')", opt_overlap_check, opt_overlap_check_template); 934 ret = -EINVAL; 935 goto fail; 936 } 937 if (!opt_overlap_check) { 938 opt_overlap_check = opt_overlap_check_template ?: "cached"; 939 } 940 941 if (!strcmp(opt_overlap_check, "none")) { 942 overlap_check_template = 0; 943 } else if (!strcmp(opt_overlap_check, "constant")) { 944 overlap_check_template = QCOW2_OL_CONSTANT; 945 } else if (!strcmp(opt_overlap_check, "cached")) { 946 overlap_check_template = QCOW2_OL_CACHED; 947 } else if (!strcmp(opt_overlap_check, "all")) { 948 overlap_check_template = QCOW2_OL_ALL; 949 } else { 950 error_setg(errp, "Unsupported value '%s' for qcow2 option " 951 "'overlap-check'. Allowed are any of the following: " 952 "none, constant, cached, all", opt_overlap_check); 953 ret = -EINVAL; 954 goto fail; 955 } 956 957 r->overlap_check = 0; 958 for (i = 0; i < QCOW2_OL_MAX_BITNR; i++) { 959 /* overlap-check defines a template bitmask, but every flag may be 960 * overwritten through the associated boolean option */ 961 r->overlap_check |= 962 qemu_opt_get_bool(opts, overlap_bool_option_names[i], 963 overlap_check_template & (1 << i)) << i; 964 } 965 966 r->discard_passthrough[QCOW2_DISCARD_NEVER] = false; 967 r->discard_passthrough[QCOW2_DISCARD_ALWAYS] = true; 968 r->discard_passthrough[QCOW2_DISCARD_REQUEST] = 969 qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_REQUEST, 970 flags & BDRV_O_UNMAP); 971 r->discard_passthrough[QCOW2_DISCARD_SNAPSHOT] = 972 qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_SNAPSHOT, true); 973 r->discard_passthrough[QCOW2_DISCARD_OTHER] = 974 qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_OTHER, false); 975 976 switch (s->crypt_method_header) { 977 case QCOW_CRYPT_NONE: 978 if (encryptfmt) { 979 error_setg(errp, "No encryption in image header, but options " 980 "specified format '%s'", encryptfmt); 981 ret = -EINVAL; 982 goto fail; 983 } 984 break; 985 986 case QCOW_CRYPT_AES: 987 if (encryptfmt && !g_str_equal(encryptfmt, "aes")) { 988 error_setg(errp, 989 "Header reported 'aes' encryption format but " 990 "options specify '%s'", encryptfmt); 991 ret = -EINVAL; 992 goto fail; 993 } 994 qdict_del(encryptopts, "format"); 995 r->crypto_opts = block_crypto_open_opts_init( 996 Q_CRYPTO_BLOCK_FORMAT_QCOW, encryptopts, errp); 997 break; 998 999 case QCOW_CRYPT_LUKS: 1000 if (encryptfmt && !g_str_equal(encryptfmt, "luks")) { 1001 error_setg(errp, 1002 "Header reported 'luks' encryption format but " 1003 "options specify '%s'", encryptfmt); 1004 ret = -EINVAL; 1005 goto fail; 1006 } 1007 qdict_del(encryptopts, "format"); 1008 r->crypto_opts = block_crypto_open_opts_init( 1009 Q_CRYPTO_BLOCK_FORMAT_LUKS, encryptopts, errp); 1010 break; 1011 1012 default: 1013 error_setg(errp, "Unsupported encryption method %d", 1014 s->crypt_method_header); 1015 break; 1016 } 1017 if (s->crypt_method_header != QCOW_CRYPT_NONE && !r->crypto_opts) { 1018 ret = -EINVAL; 1019 goto fail; 1020 } 1021 1022 ret = 0; 1023 fail: 1024 QDECREF(encryptopts); 1025 qemu_opts_del(opts); 1026 opts = NULL; 1027 return ret; 1028 } 1029 1030 static void qcow2_update_options_commit(BlockDriverState *bs, 1031 Qcow2ReopenState *r) 1032 { 1033 BDRVQcow2State *s = bs->opaque; 1034 int i; 1035 1036 if (s->l2_table_cache) { 1037 qcow2_cache_destroy(bs, s->l2_table_cache); 1038 } 1039 if (s->refcount_block_cache) { 1040 qcow2_cache_destroy(bs, s->refcount_block_cache); 1041 } 1042 s->l2_table_cache = r->l2_table_cache; 1043 s->refcount_block_cache = r->refcount_block_cache; 1044 1045 s->overlap_check = r->overlap_check; 1046 s->use_lazy_refcounts = r->use_lazy_refcounts; 1047 1048 for (i = 0; i < QCOW2_DISCARD_MAX; i++) { 1049 s->discard_passthrough[i] = r->discard_passthrough[i]; 1050 } 1051 1052 if (s->cache_clean_interval != r->cache_clean_interval) { 1053 cache_clean_timer_del(bs); 1054 s->cache_clean_interval = r->cache_clean_interval; 1055 cache_clean_timer_init(bs, bdrv_get_aio_context(bs)); 1056 } 1057 1058 qapi_free_QCryptoBlockOpenOptions(s->crypto_opts); 1059 s->crypto_opts = r->crypto_opts; 1060 } 1061 1062 static void qcow2_update_options_abort(BlockDriverState *bs, 1063 Qcow2ReopenState *r) 1064 { 1065 if (r->l2_table_cache) { 1066 qcow2_cache_destroy(bs, r->l2_table_cache); 1067 } 1068 if (r->refcount_block_cache) { 1069 qcow2_cache_destroy(bs, r->refcount_block_cache); 1070 } 1071 qapi_free_QCryptoBlockOpenOptions(r->crypto_opts); 1072 } 1073 1074 static int qcow2_update_options(BlockDriverState *bs, QDict *options, 1075 int flags, Error **errp) 1076 { 1077 Qcow2ReopenState r = {}; 1078 int ret; 1079 1080 ret = qcow2_update_options_prepare(bs, &r, options, flags, errp); 1081 if (ret >= 0) { 1082 qcow2_update_options_commit(bs, &r); 1083 } else { 1084 qcow2_update_options_abort(bs, &r); 1085 } 1086 1087 return ret; 1088 } 1089 1090 static int qcow2_do_open(BlockDriverState *bs, QDict *options, int flags, 1091 Error **errp) 1092 { 1093 BDRVQcow2State *s = bs->opaque; 1094 unsigned int len, i; 1095 int ret = 0; 1096 QCowHeader header; 1097 Error *local_err = NULL; 1098 uint64_t ext_end; 1099 uint64_t l1_vm_state_index; 1100 bool update_header = false; 1101 1102 ret = bdrv_pread(bs->file, 0, &header, sizeof(header)); 1103 if (ret < 0) { 1104 error_setg_errno(errp, -ret, "Could not read qcow2 header"); 1105 goto fail; 1106 } 1107 be32_to_cpus(&header.magic); 1108 be32_to_cpus(&header.version); 1109 be64_to_cpus(&header.backing_file_offset); 1110 be32_to_cpus(&header.backing_file_size); 1111 be64_to_cpus(&header.size); 1112 be32_to_cpus(&header.cluster_bits); 1113 be32_to_cpus(&header.crypt_method); 1114 be64_to_cpus(&header.l1_table_offset); 1115 be32_to_cpus(&header.l1_size); 1116 be64_to_cpus(&header.refcount_table_offset); 1117 be32_to_cpus(&header.refcount_table_clusters); 1118 be64_to_cpus(&header.snapshots_offset); 1119 be32_to_cpus(&header.nb_snapshots); 1120 1121 if (header.magic != QCOW_MAGIC) { 1122 error_setg(errp, "Image is not in qcow2 format"); 1123 ret = -EINVAL; 1124 goto fail; 1125 } 1126 if (header.version < 2 || header.version > 3) { 1127 error_setg(errp, "Unsupported qcow2 version %" PRIu32, header.version); 1128 ret = -ENOTSUP; 1129 goto fail; 1130 } 1131 1132 s->qcow_version = header.version; 1133 1134 /* Initialise cluster size */ 1135 if (header.cluster_bits < MIN_CLUSTER_BITS || 1136 header.cluster_bits > MAX_CLUSTER_BITS) { 1137 error_setg(errp, "Unsupported cluster size: 2^%" PRIu32, 1138 header.cluster_bits); 1139 ret = -EINVAL; 1140 goto fail; 1141 } 1142 1143 s->cluster_bits = header.cluster_bits; 1144 s->cluster_size = 1 << s->cluster_bits; 1145 s->cluster_sectors = 1 << (s->cluster_bits - BDRV_SECTOR_BITS); 1146 1147 /* Initialise version 3 header fields */ 1148 if (header.version == 2) { 1149 header.incompatible_features = 0; 1150 header.compatible_features = 0; 1151 header.autoclear_features = 0; 1152 header.refcount_order = 4; 1153 header.header_length = 72; 1154 } else { 1155 be64_to_cpus(&header.incompatible_features); 1156 be64_to_cpus(&header.compatible_features); 1157 be64_to_cpus(&header.autoclear_features); 1158 be32_to_cpus(&header.refcount_order); 1159 be32_to_cpus(&header.header_length); 1160 1161 if (header.header_length < 104) { 1162 error_setg(errp, "qcow2 header too short"); 1163 ret = -EINVAL; 1164 goto fail; 1165 } 1166 } 1167 1168 if (header.header_length > s->cluster_size) { 1169 error_setg(errp, "qcow2 header exceeds cluster size"); 1170 ret = -EINVAL; 1171 goto fail; 1172 } 1173 1174 if (header.header_length > sizeof(header)) { 1175 s->unknown_header_fields_size = header.header_length - sizeof(header); 1176 s->unknown_header_fields = g_malloc(s->unknown_header_fields_size); 1177 ret = bdrv_pread(bs->file, sizeof(header), s->unknown_header_fields, 1178 s->unknown_header_fields_size); 1179 if (ret < 0) { 1180 error_setg_errno(errp, -ret, "Could not read unknown qcow2 header " 1181 "fields"); 1182 goto fail; 1183 } 1184 } 1185 1186 if (header.backing_file_offset > s->cluster_size) { 1187 error_setg(errp, "Invalid backing file offset"); 1188 ret = -EINVAL; 1189 goto fail; 1190 } 1191 1192 if (header.backing_file_offset) { 1193 ext_end = header.backing_file_offset; 1194 } else { 1195 ext_end = 1 << header.cluster_bits; 1196 } 1197 1198 /* Handle feature bits */ 1199 s->incompatible_features = header.incompatible_features; 1200 s->compatible_features = header.compatible_features; 1201 s->autoclear_features = header.autoclear_features; 1202 1203 if (s->incompatible_features & ~QCOW2_INCOMPAT_MASK) { 1204 void *feature_table = NULL; 1205 qcow2_read_extensions(bs, header.header_length, ext_end, 1206 &feature_table, flags, NULL, NULL); 1207 report_unsupported_feature(errp, feature_table, 1208 s->incompatible_features & 1209 ~QCOW2_INCOMPAT_MASK); 1210 ret = -ENOTSUP; 1211 g_free(feature_table); 1212 goto fail; 1213 } 1214 1215 if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) { 1216 /* Corrupt images may not be written to unless they are being repaired 1217 */ 1218 if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_CHECK)) { 1219 error_setg(errp, "qcow2: Image is corrupt; cannot be opened " 1220 "read/write"); 1221 ret = -EACCES; 1222 goto fail; 1223 } 1224 } 1225 1226 /* Check support for various header values */ 1227 if (header.refcount_order > 6) { 1228 error_setg(errp, "Reference count entry width too large; may not " 1229 "exceed 64 bits"); 1230 ret = -EINVAL; 1231 goto fail; 1232 } 1233 s->refcount_order = header.refcount_order; 1234 s->refcount_bits = 1 << s->refcount_order; 1235 s->refcount_max = UINT64_C(1) << (s->refcount_bits - 1); 1236 s->refcount_max += s->refcount_max - 1; 1237 1238 s->crypt_method_header = header.crypt_method; 1239 if (s->crypt_method_header) { 1240 if (bdrv_uses_whitelist() && 1241 s->crypt_method_header == QCOW_CRYPT_AES) { 1242 error_setg(errp, 1243 "Use of AES-CBC encrypted qcow2 images is no longer " 1244 "supported in system emulators"); 1245 error_append_hint(errp, 1246 "You can use 'qemu-img convert' to convert your " 1247 "image to an alternative supported format, such " 1248 "as unencrypted qcow2, or raw with the LUKS " 1249 "format instead.\n"); 1250 ret = -ENOSYS; 1251 goto fail; 1252 } 1253 1254 if (s->crypt_method_header == QCOW_CRYPT_AES) { 1255 s->crypt_physical_offset = false; 1256 } else { 1257 /* Assuming LUKS and any future crypt methods we 1258 * add will all use physical offsets, due to the 1259 * fact that the alternative is insecure... */ 1260 s->crypt_physical_offset = true; 1261 } 1262 1263 bs->encrypted = true; 1264 } 1265 1266 s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */ 1267 s->l2_size = 1 << s->l2_bits; 1268 /* 2^(s->refcount_order - 3) is the refcount width in bytes */ 1269 s->refcount_block_bits = s->cluster_bits - (s->refcount_order - 3); 1270 s->refcount_block_size = 1 << s->refcount_block_bits; 1271 bs->total_sectors = header.size / 512; 1272 s->csize_shift = (62 - (s->cluster_bits - 8)); 1273 s->csize_mask = (1 << (s->cluster_bits - 8)) - 1; 1274 s->cluster_offset_mask = (1LL << s->csize_shift) - 1; 1275 1276 s->refcount_table_offset = header.refcount_table_offset; 1277 s->refcount_table_size = 1278 header.refcount_table_clusters << (s->cluster_bits - 3); 1279 1280 if (header.refcount_table_clusters > qcow2_max_refcount_clusters(s)) { 1281 error_setg(errp, "Reference count table too large"); 1282 ret = -EINVAL; 1283 goto fail; 1284 } 1285 1286 if (header.refcount_table_clusters == 0 && !(flags & BDRV_O_CHECK)) { 1287 error_setg(errp, "Image does not contain a reference count table"); 1288 ret = -EINVAL; 1289 goto fail; 1290 } 1291 1292 ret = validate_table_offset(bs, s->refcount_table_offset, 1293 s->refcount_table_size, sizeof(uint64_t)); 1294 if (ret < 0) { 1295 error_setg(errp, "Invalid reference count table offset"); 1296 goto fail; 1297 } 1298 1299 /* Snapshot table offset/length */ 1300 if (header.nb_snapshots > QCOW_MAX_SNAPSHOTS) { 1301 error_setg(errp, "Too many snapshots"); 1302 ret = -EINVAL; 1303 goto fail; 1304 } 1305 1306 ret = validate_table_offset(bs, header.snapshots_offset, 1307 header.nb_snapshots, 1308 sizeof(QCowSnapshotHeader)); 1309 if (ret < 0) { 1310 error_setg(errp, "Invalid snapshot table offset"); 1311 goto fail; 1312 } 1313 1314 /* read the level 1 table */ 1315 if (header.l1_size > QCOW_MAX_L1_SIZE / sizeof(uint64_t)) { 1316 error_setg(errp, "Active L1 table too large"); 1317 ret = -EFBIG; 1318 goto fail; 1319 } 1320 s->l1_size = header.l1_size; 1321 1322 l1_vm_state_index = size_to_l1(s, header.size); 1323 if (l1_vm_state_index > INT_MAX) { 1324 error_setg(errp, "Image is too big"); 1325 ret = -EFBIG; 1326 goto fail; 1327 } 1328 s->l1_vm_state_index = l1_vm_state_index; 1329 1330 /* the L1 table must contain at least enough entries to put 1331 header.size bytes */ 1332 if (s->l1_size < s->l1_vm_state_index) { 1333 error_setg(errp, "L1 table is too small"); 1334 ret = -EINVAL; 1335 goto fail; 1336 } 1337 1338 ret = validate_table_offset(bs, header.l1_table_offset, 1339 header.l1_size, sizeof(uint64_t)); 1340 if (ret < 0) { 1341 error_setg(errp, "Invalid L1 table offset"); 1342 goto fail; 1343 } 1344 s->l1_table_offset = header.l1_table_offset; 1345 1346 1347 if (s->l1_size > 0) { 1348 s->l1_table = qemu_try_blockalign(bs->file->bs, 1349 align_offset(s->l1_size * sizeof(uint64_t), 512)); 1350 if (s->l1_table == NULL) { 1351 error_setg(errp, "Could not allocate L1 table"); 1352 ret = -ENOMEM; 1353 goto fail; 1354 } 1355 ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table, 1356 s->l1_size * sizeof(uint64_t)); 1357 if (ret < 0) { 1358 error_setg_errno(errp, -ret, "Could not read L1 table"); 1359 goto fail; 1360 } 1361 for(i = 0;i < s->l1_size; i++) { 1362 be64_to_cpus(&s->l1_table[i]); 1363 } 1364 } 1365 1366 /* Parse driver-specific options */ 1367 ret = qcow2_update_options(bs, options, flags, errp); 1368 if (ret < 0) { 1369 goto fail; 1370 } 1371 1372 s->cluster_cache_offset = -1; 1373 s->flags = flags; 1374 1375 ret = qcow2_refcount_init(bs); 1376 if (ret != 0) { 1377 error_setg_errno(errp, -ret, "Could not initialize refcount handling"); 1378 goto fail; 1379 } 1380 1381 QLIST_INIT(&s->cluster_allocs); 1382 QTAILQ_INIT(&s->discards); 1383 1384 /* read qcow2 extensions */ 1385 if (qcow2_read_extensions(bs, header.header_length, ext_end, NULL, 1386 flags, &update_header, &local_err)) { 1387 error_propagate(errp, local_err); 1388 ret = -EINVAL; 1389 goto fail; 1390 } 1391 1392 /* qcow2_read_extension may have set up the crypto context 1393 * if the crypt method needs a header region, some methods 1394 * don't need header extensions, so must check here 1395 */ 1396 if (s->crypt_method_header && !s->crypto) { 1397 if (s->crypt_method_header == QCOW_CRYPT_AES) { 1398 unsigned int cflags = 0; 1399 if (flags & BDRV_O_NO_IO) { 1400 cflags |= QCRYPTO_BLOCK_OPEN_NO_IO; 1401 } 1402 s->crypto = qcrypto_block_open(s->crypto_opts, "encrypt.", 1403 NULL, NULL, cflags, errp); 1404 if (!s->crypto) { 1405 ret = -EINVAL; 1406 goto fail; 1407 } 1408 } else if (!(flags & BDRV_O_NO_IO)) { 1409 error_setg(errp, "Missing CRYPTO header for crypt method %d", 1410 s->crypt_method_header); 1411 ret = -EINVAL; 1412 goto fail; 1413 } 1414 } 1415 1416 /* read the backing file name */ 1417 if (header.backing_file_offset != 0) { 1418 len = header.backing_file_size; 1419 if (len > MIN(1023, s->cluster_size - header.backing_file_offset) || 1420 len >= sizeof(bs->backing_file)) { 1421 error_setg(errp, "Backing file name too long"); 1422 ret = -EINVAL; 1423 goto fail; 1424 } 1425 ret = bdrv_pread(bs->file, header.backing_file_offset, 1426 bs->backing_file, len); 1427 if (ret < 0) { 1428 error_setg_errno(errp, -ret, "Could not read backing file name"); 1429 goto fail; 1430 } 1431 bs->backing_file[len] = '\0'; 1432 s->image_backing_file = g_strdup(bs->backing_file); 1433 } 1434 1435 /* Internal snapshots */ 1436 s->snapshots_offset = header.snapshots_offset; 1437 s->nb_snapshots = header.nb_snapshots; 1438 1439 ret = qcow2_read_snapshots(bs); 1440 if (ret < 0) { 1441 error_setg_errno(errp, -ret, "Could not read snapshots"); 1442 goto fail; 1443 } 1444 1445 /* Clear unknown autoclear feature bits */ 1446 update_header |= s->autoclear_features & ~QCOW2_AUTOCLEAR_MASK; 1447 update_header = 1448 update_header && !bs->read_only && !(flags & BDRV_O_INACTIVE); 1449 if (update_header) { 1450 s->autoclear_features &= QCOW2_AUTOCLEAR_MASK; 1451 } 1452 1453 if (qcow2_load_autoloading_dirty_bitmaps(bs, &local_err)) { 1454 update_header = false; 1455 } 1456 if (local_err != NULL) { 1457 error_propagate(errp, local_err); 1458 ret = -EINVAL; 1459 goto fail; 1460 } 1461 1462 if (update_header) { 1463 ret = qcow2_update_header(bs); 1464 if (ret < 0) { 1465 error_setg_errno(errp, -ret, "Could not update qcow2 header"); 1466 goto fail; 1467 } 1468 } 1469 1470 /* Initialise locks */ 1471 qemu_co_mutex_init(&s->lock); 1472 bs->supported_zero_flags = BDRV_REQ_MAY_UNMAP; 1473 1474 /* Repair image if dirty */ 1475 if (!(flags & (BDRV_O_CHECK | BDRV_O_INACTIVE)) && !bs->read_only && 1476 (s->incompatible_features & QCOW2_INCOMPAT_DIRTY)) { 1477 BdrvCheckResult result = {0}; 1478 1479 ret = qcow2_check(bs, &result, BDRV_FIX_ERRORS | BDRV_FIX_LEAKS); 1480 if (ret < 0 || result.check_errors) { 1481 if (ret >= 0) { 1482 ret = -EIO; 1483 } 1484 error_setg_errno(errp, -ret, "Could not repair dirty image"); 1485 goto fail; 1486 } 1487 } 1488 1489 #ifdef DEBUG_ALLOC 1490 { 1491 BdrvCheckResult result = {0}; 1492 qcow2_check_refcounts(bs, &result, 0); 1493 } 1494 #endif 1495 return ret; 1496 1497 fail: 1498 g_free(s->unknown_header_fields); 1499 cleanup_unknown_header_ext(bs); 1500 qcow2_free_snapshots(bs); 1501 qcow2_refcount_close(bs); 1502 qemu_vfree(s->l1_table); 1503 /* else pre-write overlap checks in cache_destroy may crash */ 1504 s->l1_table = NULL; 1505 cache_clean_timer_del(bs); 1506 if (s->l2_table_cache) { 1507 qcow2_cache_destroy(bs, s->l2_table_cache); 1508 } 1509 if (s->refcount_block_cache) { 1510 qcow2_cache_destroy(bs, s->refcount_block_cache); 1511 } 1512 qcrypto_block_free(s->crypto); 1513 qapi_free_QCryptoBlockOpenOptions(s->crypto_opts); 1514 return ret; 1515 } 1516 1517 static int qcow2_open(BlockDriverState *bs, QDict *options, int flags, 1518 Error **errp) 1519 { 1520 bs->file = bdrv_open_child(NULL, options, "file", bs, &child_file, 1521 false, errp); 1522 if (!bs->file) { 1523 return -EINVAL; 1524 } 1525 1526 return qcow2_do_open(bs, options, flags, errp); 1527 } 1528 1529 static void qcow2_refresh_limits(BlockDriverState *bs, Error **errp) 1530 { 1531 BDRVQcow2State *s = bs->opaque; 1532 1533 if (bs->encrypted) { 1534 /* Encryption works on a sector granularity */ 1535 bs->bl.request_alignment = BDRV_SECTOR_SIZE; 1536 } 1537 bs->bl.pwrite_zeroes_alignment = s->cluster_size; 1538 bs->bl.pdiscard_alignment = s->cluster_size; 1539 } 1540 1541 static int qcow2_reopen_prepare(BDRVReopenState *state, 1542 BlockReopenQueue *queue, Error **errp) 1543 { 1544 Qcow2ReopenState *r; 1545 int ret; 1546 1547 r = g_new0(Qcow2ReopenState, 1); 1548 state->opaque = r; 1549 1550 ret = qcow2_update_options_prepare(state->bs, r, state->options, 1551 state->flags, errp); 1552 if (ret < 0) { 1553 goto fail; 1554 } 1555 1556 /* We need to write out any unwritten data if we reopen read-only. */ 1557 if ((state->flags & BDRV_O_RDWR) == 0) { 1558 ret = qcow2_reopen_bitmaps_ro(state->bs, errp); 1559 if (ret < 0) { 1560 goto fail; 1561 } 1562 1563 ret = bdrv_flush(state->bs); 1564 if (ret < 0) { 1565 goto fail; 1566 } 1567 1568 ret = qcow2_mark_clean(state->bs); 1569 if (ret < 0) { 1570 goto fail; 1571 } 1572 } 1573 1574 return 0; 1575 1576 fail: 1577 qcow2_update_options_abort(state->bs, r); 1578 g_free(r); 1579 return ret; 1580 } 1581 1582 static void qcow2_reopen_commit(BDRVReopenState *state) 1583 { 1584 qcow2_update_options_commit(state->bs, state->opaque); 1585 g_free(state->opaque); 1586 } 1587 1588 static void qcow2_reopen_abort(BDRVReopenState *state) 1589 { 1590 qcow2_update_options_abort(state->bs, state->opaque); 1591 g_free(state->opaque); 1592 } 1593 1594 static void qcow2_join_options(QDict *options, QDict *old_options) 1595 { 1596 bool has_new_overlap_template = 1597 qdict_haskey(options, QCOW2_OPT_OVERLAP) || 1598 qdict_haskey(options, QCOW2_OPT_OVERLAP_TEMPLATE); 1599 bool has_new_total_cache_size = 1600 qdict_haskey(options, QCOW2_OPT_CACHE_SIZE); 1601 bool has_all_cache_options; 1602 1603 /* New overlap template overrides all old overlap options */ 1604 if (has_new_overlap_template) { 1605 qdict_del(old_options, QCOW2_OPT_OVERLAP); 1606 qdict_del(old_options, QCOW2_OPT_OVERLAP_TEMPLATE); 1607 qdict_del(old_options, QCOW2_OPT_OVERLAP_MAIN_HEADER); 1608 qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L1); 1609 qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L2); 1610 qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_TABLE); 1611 qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK); 1612 qdict_del(old_options, QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE); 1613 qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L1); 1614 qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L2); 1615 } 1616 1617 /* New total cache size overrides all old options */ 1618 if (qdict_haskey(options, QCOW2_OPT_CACHE_SIZE)) { 1619 qdict_del(old_options, QCOW2_OPT_L2_CACHE_SIZE); 1620 qdict_del(old_options, QCOW2_OPT_REFCOUNT_CACHE_SIZE); 1621 } 1622 1623 qdict_join(options, old_options, false); 1624 1625 /* 1626 * If after merging all cache size options are set, an old total size is 1627 * overwritten. Do keep all options, however, if all three are new. The 1628 * resulting error message is what we want to happen. 1629 */ 1630 has_all_cache_options = 1631 qdict_haskey(options, QCOW2_OPT_CACHE_SIZE) || 1632 qdict_haskey(options, QCOW2_OPT_L2_CACHE_SIZE) || 1633 qdict_haskey(options, QCOW2_OPT_REFCOUNT_CACHE_SIZE); 1634 1635 if (has_all_cache_options && !has_new_total_cache_size) { 1636 qdict_del(options, QCOW2_OPT_CACHE_SIZE); 1637 } 1638 } 1639 1640 static int64_t coroutine_fn qcow2_co_get_block_status(BlockDriverState *bs, 1641 int64_t sector_num, int nb_sectors, int *pnum, BlockDriverState **file) 1642 { 1643 BDRVQcow2State *s = bs->opaque; 1644 uint64_t cluster_offset; 1645 int index_in_cluster, ret; 1646 unsigned int bytes; 1647 int64_t status = 0; 1648 1649 bytes = MIN(INT_MAX, nb_sectors * BDRV_SECTOR_SIZE); 1650 qemu_co_mutex_lock(&s->lock); 1651 ret = qcow2_get_cluster_offset(bs, sector_num << BDRV_SECTOR_BITS, &bytes, 1652 &cluster_offset); 1653 qemu_co_mutex_unlock(&s->lock); 1654 if (ret < 0) { 1655 return ret; 1656 } 1657 1658 *pnum = bytes >> BDRV_SECTOR_BITS; 1659 1660 if (cluster_offset != 0 && ret != QCOW2_CLUSTER_COMPRESSED && 1661 !s->crypto) { 1662 index_in_cluster = sector_num & (s->cluster_sectors - 1); 1663 cluster_offset |= (index_in_cluster << BDRV_SECTOR_BITS); 1664 *file = bs->file->bs; 1665 status |= BDRV_BLOCK_OFFSET_VALID | cluster_offset; 1666 } 1667 if (ret == QCOW2_CLUSTER_ZERO_PLAIN || ret == QCOW2_CLUSTER_ZERO_ALLOC) { 1668 status |= BDRV_BLOCK_ZERO; 1669 } else if (ret != QCOW2_CLUSTER_UNALLOCATED) { 1670 status |= BDRV_BLOCK_DATA; 1671 } 1672 return status; 1673 } 1674 1675 static coroutine_fn int qcow2_co_preadv(BlockDriverState *bs, uint64_t offset, 1676 uint64_t bytes, QEMUIOVector *qiov, 1677 int flags) 1678 { 1679 BDRVQcow2State *s = bs->opaque; 1680 int offset_in_cluster; 1681 int ret; 1682 unsigned int cur_bytes; /* number of bytes in current iteration */ 1683 uint64_t cluster_offset = 0; 1684 uint64_t bytes_done = 0; 1685 QEMUIOVector hd_qiov; 1686 uint8_t *cluster_data = NULL; 1687 1688 qemu_iovec_init(&hd_qiov, qiov->niov); 1689 1690 qemu_co_mutex_lock(&s->lock); 1691 1692 while (bytes != 0) { 1693 1694 /* prepare next request */ 1695 cur_bytes = MIN(bytes, INT_MAX); 1696 if (s->crypto) { 1697 cur_bytes = MIN(cur_bytes, 1698 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size); 1699 } 1700 1701 ret = qcow2_get_cluster_offset(bs, offset, &cur_bytes, &cluster_offset); 1702 if (ret < 0) { 1703 goto fail; 1704 } 1705 1706 offset_in_cluster = offset_into_cluster(s, offset); 1707 1708 qemu_iovec_reset(&hd_qiov); 1709 qemu_iovec_concat(&hd_qiov, qiov, bytes_done, cur_bytes); 1710 1711 switch (ret) { 1712 case QCOW2_CLUSTER_UNALLOCATED: 1713 1714 if (bs->backing) { 1715 BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO); 1716 qemu_co_mutex_unlock(&s->lock); 1717 ret = bdrv_co_preadv(bs->backing, offset, cur_bytes, 1718 &hd_qiov, 0); 1719 qemu_co_mutex_lock(&s->lock); 1720 if (ret < 0) { 1721 goto fail; 1722 } 1723 } else { 1724 /* Note: in this case, no need to wait */ 1725 qemu_iovec_memset(&hd_qiov, 0, 0, cur_bytes); 1726 } 1727 break; 1728 1729 case QCOW2_CLUSTER_ZERO_PLAIN: 1730 case QCOW2_CLUSTER_ZERO_ALLOC: 1731 qemu_iovec_memset(&hd_qiov, 0, 0, cur_bytes); 1732 break; 1733 1734 case QCOW2_CLUSTER_COMPRESSED: 1735 /* add AIO support for compressed blocks ? */ 1736 ret = qcow2_decompress_cluster(bs, cluster_offset); 1737 if (ret < 0) { 1738 goto fail; 1739 } 1740 1741 qemu_iovec_from_buf(&hd_qiov, 0, 1742 s->cluster_cache + offset_in_cluster, 1743 cur_bytes); 1744 break; 1745 1746 case QCOW2_CLUSTER_NORMAL: 1747 if ((cluster_offset & 511) != 0) { 1748 ret = -EIO; 1749 goto fail; 1750 } 1751 1752 if (bs->encrypted) { 1753 assert(s->crypto); 1754 1755 /* 1756 * For encrypted images, read everything into a temporary 1757 * contiguous buffer on which the AES functions can work. 1758 */ 1759 if (!cluster_data) { 1760 cluster_data = 1761 qemu_try_blockalign(bs->file->bs, 1762 QCOW_MAX_CRYPT_CLUSTERS 1763 * s->cluster_size); 1764 if (cluster_data == NULL) { 1765 ret = -ENOMEM; 1766 goto fail; 1767 } 1768 } 1769 1770 assert(cur_bytes <= QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size); 1771 qemu_iovec_reset(&hd_qiov); 1772 qemu_iovec_add(&hd_qiov, cluster_data, cur_bytes); 1773 } 1774 1775 BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO); 1776 qemu_co_mutex_unlock(&s->lock); 1777 ret = bdrv_co_preadv(bs->file, 1778 cluster_offset + offset_in_cluster, 1779 cur_bytes, &hd_qiov, 0); 1780 qemu_co_mutex_lock(&s->lock); 1781 if (ret < 0) { 1782 goto fail; 1783 } 1784 if (bs->encrypted) { 1785 assert(s->crypto); 1786 assert((offset & (BDRV_SECTOR_SIZE - 1)) == 0); 1787 assert((cur_bytes & (BDRV_SECTOR_SIZE - 1)) == 0); 1788 if (qcrypto_block_decrypt(s->crypto, 1789 (s->crypt_physical_offset ? 1790 cluster_offset + offset_in_cluster : 1791 offset), 1792 cluster_data, 1793 cur_bytes, 1794 NULL) < 0) { 1795 ret = -EIO; 1796 goto fail; 1797 } 1798 qemu_iovec_from_buf(qiov, bytes_done, cluster_data, cur_bytes); 1799 } 1800 break; 1801 1802 default: 1803 g_assert_not_reached(); 1804 ret = -EIO; 1805 goto fail; 1806 } 1807 1808 bytes -= cur_bytes; 1809 offset += cur_bytes; 1810 bytes_done += cur_bytes; 1811 } 1812 ret = 0; 1813 1814 fail: 1815 qemu_co_mutex_unlock(&s->lock); 1816 1817 qemu_iovec_destroy(&hd_qiov); 1818 qemu_vfree(cluster_data); 1819 1820 return ret; 1821 } 1822 1823 /* Check if it's possible to merge a write request with the writing of 1824 * the data from the COW regions */ 1825 static bool merge_cow(uint64_t offset, unsigned bytes, 1826 QEMUIOVector *hd_qiov, QCowL2Meta *l2meta) 1827 { 1828 QCowL2Meta *m; 1829 1830 for (m = l2meta; m != NULL; m = m->next) { 1831 /* If both COW regions are empty then there's nothing to merge */ 1832 if (m->cow_start.nb_bytes == 0 && m->cow_end.nb_bytes == 0) { 1833 continue; 1834 } 1835 1836 /* The data (middle) region must be immediately after the 1837 * start region */ 1838 if (l2meta_cow_start(m) + m->cow_start.nb_bytes != offset) { 1839 continue; 1840 } 1841 1842 /* The end region must be immediately after the data (middle) 1843 * region */ 1844 if (m->offset + m->cow_end.offset != offset + bytes) { 1845 continue; 1846 } 1847 1848 /* Make sure that adding both COW regions to the QEMUIOVector 1849 * does not exceed IOV_MAX */ 1850 if (hd_qiov->niov > IOV_MAX - 2) { 1851 continue; 1852 } 1853 1854 m->data_qiov = hd_qiov; 1855 return true; 1856 } 1857 1858 return false; 1859 } 1860 1861 static coroutine_fn int qcow2_co_pwritev(BlockDriverState *bs, uint64_t offset, 1862 uint64_t bytes, QEMUIOVector *qiov, 1863 int flags) 1864 { 1865 BDRVQcow2State *s = bs->opaque; 1866 int offset_in_cluster; 1867 int ret; 1868 unsigned int cur_bytes; /* number of sectors in current iteration */ 1869 uint64_t cluster_offset; 1870 QEMUIOVector hd_qiov; 1871 uint64_t bytes_done = 0; 1872 uint8_t *cluster_data = NULL; 1873 QCowL2Meta *l2meta = NULL; 1874 1875 trace_qcow2_writev_start_req(qemu_coroutine_self(), offset, bytes); 1876 1877 qemu_iovec_init(&hd_qiov, qiov->niov); 1878 1879 s->cluster_cache_offset = -1; /* disable compressed cache */ 1880 1881 qemu_co_mutex_lock(&s->lock); 1882 1883 while (bytes != 0) { 1884 1885 l2meta = NULL; 1886 1887 trace_qcow2_writev_start_part(qemu_coroutine_self()); 1888 offset_in_cluster = offset_into_cluster(s, offset); 1889 cur_bytes = MIN(bytes, INT_MAX); 1890 if (bs->encrypted) { 1891 cur_bytes = MIN(cur_bytes, 1892 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size 1893 - offset_in_cluster); 1894 } 1895 1896 ret = qcow2_alloc_cluster_offset(bs, offset, &cur_bytes, 1897 &cluster_offset, &l2meta); 1898 if (ret < 0) { 1899 goto fail; 1900 } 1901 1902 assert((cluster_offset & 511) == 0); 1903 1904 qemu_iovec_reset(&hd_qiov); 1905 qemu_iovec_concat(&hd_qiov, qiov, bytes_done, cur_bytes); 1906 1907 if (bs->encrypted) { 1908 assert(s->crypto); 1909 if (!cluster_data) { 1910 cluster_data = qemu_try_blockalign(bs->file->bs, 1911 QCOW_MAX_CRYPT_CLUSTERS 1912 * s->cluster_size); 1913 if (cluster_data == NULL) { 1914 ret = -ENOMEM; 1915 goto fail; 1916 } 1917 } 1918 1919 assert(hd_qiov.size <= 1920 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size); 1921 qemu_iovec_to_buf(&hd_qiov, 0, cluster_data, hd_qiov.size); 1922 1923 if (qcrypto_block_encrypt(s->crypto, 1924 (s->crypt_physical_offset ? 1925 cluster_offset + offset_in_cluster : 1926 offset), 1927 cluster_data, 1928 cur_bytes, NULL) < 0) { 1929 ret = -EIO; 1930 goto fail; 1931 } 1932 1933 qemu_iovec_reset(&hd_qiov); 1934 qemu_iovec_add(&hd_qiov, cluster_data, cur_bytes); 1935 } 1936 1937 ret = qcow2_pre_write_overlap_check(bs, 0, 1938 cluster_offset + offset_in_cluster, cur_bytes); 1939 if (ret < 0) { 1940 goto fail; 1941 } 1942 1943 /* If we need to do COW, check if it's possible to merge the 1944 * writing of the guest data together with that of the COW regions. 1945 * If it's not possible (or not necessary) then write the 1946 * guest data now. */ 1947 if (!merge_cow(offset, cur_bytes, &hd_qiov, l2meta)) { 1948 qemu_co_mutex_unlock(&s->lock); 1949 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO); 1950 trace_qcow2_writev_data(qemu_coroutine_self(), 1951 cluster_offset + offset_in_cluster); 1952 ret = bdrv_co_pwritev(bs->file, 1953 cluster_offset + offset_in_cluster, 1954 cur_bytes, &hd_qiov, 0); 1955 qemu_co_mutex_lock(&s->lock); 1956 if (ret < 0) { 1957 goto fail; 1958 } 1959 } 1960 1961 while (l2meta != NULL) { 1962 QCowL2Meta *next; 1963 1964 ret = qcow2_alloc_cluster_link_l2(bs, l2meta); 1965 if (ret < 0) { 1966 goto fail; 1967 } 1968 1969 /* Take the request off the list of running requests */ 1970 if (l2meta->nb_clusters != 0) { 1971 QLIST_REMOVE(l2meta, next_in_flight); 1972 } 1973 1974 qemu_co_queue_restart_all(&l2meta->dependent_requests); 1975 1976 next = l2meta->next; 1977 g_free(l2meta); 1978 l2meta = next; 1979 } 1980 1981 bytes -= cur_bytes; 1982 offset += cur_bytes; 1983 bytes_done += cur_bytes; 1984 trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_bytes); 1985 } 1986 ret = 0; 1987 1988 fail: 1989 while (l2meta != NULL) { 1990 QCowL2Meta *next; 1991 1992 if (l2meta->nb_clusters != 0) { 1993 QLIST_REMOVE(l2meta, next_in_flight); 1994 } 1995 qemu_co_queue_restart_all(&l2meta->dependent_requests); 1996 1997 next = l2meta->next; 1998 g_free(l2meta); 1999 l2meta = next; 2000 } 2001 2002 qemu_co_mutex_unlock(&s->lock); 2003 2004 qemu_iovec_destroy(&hd_qiov); 2005 qemu_vfree(cluster_data); 2006 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret); 2007 2008 return ret; 2009 } 2010 2011 static int qcow2_inactivate(BlockDriverState *bs) 2012 { 2013 BDRVQcow2State *s = bs->opaque; 2014 int ret, result = 0; 2015 Error *local_err = NULL; 2016 2017 qcow2_store_persistent_dirty_bitmaps(bs, &local_err); 2018 if (local_err != NULL) { 2019 result = -EINVAL; 2020 error_report_err(local_err); 2021 error_report("Persistent bitmaps are lost for node '%s'", 2022 bdrv_get_device_or_node_name(bs)); 2023 } 2024 2025 ret = qcow2_cache_flush(bs, s->l2_table_cache); 2026 if (ret) { 2027 result = ret; 2028 error_report("Failed to flush the L2 table cache: %s", 2029 strerror(-ret)); 2030 } 2031 2032 ret = qcow2_cache_flush(bs, s->refcount_block_cache); 2033 if (ret) { 2034 result = ret; 2035 error_report("Failed to flush the refcount block cache: %s", 2036 strerror(-ret)); 2037 } 2038 2039 if (result == 0) { 2040 qcow2_mark_clean(bs); 2041 } 2042 2043 return result; 2044 } 2045 2046 static void qcow2_close(BlockDriverState *bs) 2047 { 2048 BDRVQcow2State *s = bs->opaque; 2049 qemu_vfree(s->l1_table); 2050 /* else pre-write overlap checks in cache_destroy may crash */ 2051 s->l1_table = NULL; 2052 2053 if (!(s->flags & BDRV_O_INACTIVE)) { 2054 qcow2_inactivate(bs); 2055 } 2056 2057 cache_clean_timer_del(bs); 2058 qcow2_cache_destroy(bs, s->l2_table_cache); 2059 qcow2_cache_destroy(bs, s->refcount_block_cache); 2060 2061 qcrypto_block_free(s->crypto); 2062 s->crypto = NULL; 2063 2064 g_free(s->unknown_header_fields); 2065 cleanup_unknown_header_ext(bs); 2066 2067 g_free(s->image_backing_file); 2068 g_free(s->image_backing_format); 2069 2070 g_free(s->cluster_cache); 2071 qemu_vfree(s->cluster_data); 2072 qcow2_refcount_close(bs); 2073 qcow2_free_snapshots(bs); 2074 } 2075 2076 static void qcow2_invalidate_cache(BlockDriverState *bs, Error **errp) 2077 { 2078 BDRVQcow2State *s = bs->opaque; 2079 int flags = s->flags; 2080 QCryptoBlock *crypto = NULL; 2081 QDict *options; 2082 Error *local_err = NULL; 2083 int ret; 2084 2085 /* 2086 * Backing files are read-only which makes all of their metadata immutable, 2087 * that means we don't have to worry about reopening them here. 2088 */ 2089 2090 crypto = s->crypto; 2091 s->crypto = NULL; 2092 2093 qcow2_close(bs); 2094 2095 memset(s, 0, sizeof(BDRVQcow2State)); 2096 options = qdict_clone_shallow(bs->options); 2097 2098 flags &= ~BDRV_O_INACTIVE; 2099 ret = qcow2_do_open(bs, options, flags, &local_err); 2100 QDECREF(options); 2101 if (local_err) { 2102 error_propagate(errp, local_err); 2103 error_prepend(errp, "Could not reopen qcow2 layer: "); 2104 bs->drv = NULL; 2105 return; 2106 } else if (ret < 0) { 2107 error_setg_errno(errp, -ret, "Could not reopen qcow2 layer"); 2108 bs->drv = NULL; 2109 return; 2110 } 2111 2112 s->crypto = crypto; 2113 } 2114 2115 static size_t header_ext_add(char *buf, uint32_t magic, const void *s, 2116 size_t len, size_t buflen) 2117 { 2118 QCowExtension *ext_backing_fmt = (QCowExtension*) buf; 2119 size_t ext_len = sizeof(QCowExtension) + ((len + 7) & ~7); 2120 2121 if (buflen < ext_len) { 2122 return -ENOSPC; 2123 } 2124 2125 *ext_backing_fmt = (QCowExtension) { 2126 .magic = cpu_to_be32(magic), 2127 .len = cpu_to_be32(len), 2128 }; 2129 2130 if (len) { 2131 memcpy(buf + sizeof(QCowExtension), s, len); 2132 } 2133 2134 return ext_len; 2135 } 2136 2137 /* 2138 * Updates the qcow2 header, including the variable length parts of it, i.e. 2139 * the backing file name and all extensions. qcow2 was not designed to allow 2140 * such changes, so if we run out of space (we can only use the first cluster) 2141 * this function may fail. 2142 * 2143 * Returns 0 on success, -errno in error cases. 2144 */ 2145 int qcow2_update_header(BlockDriverState *bs) 2146 { 2147 BDRVQcow2State *s = bs->opaque; 2148 QCowHeader *header; 2149 char *buf; 2150 size_t buflen = s->cluster_size; 2151 int ret; 2152 uint64_t total_size; 2153 uint32_t refcount_table_clusters; 2154 size_t header_length; 2155 Qcow2UnknownHeaderExtension *uext; 2156 2157 buf = qemu_blockalign(bs, buflen); 2158 2159 /* Header structure */ 2160 header = (QCowHeader*) buf; 2161 2162 if (buflen < sizeof(*header)) { 2163 ret = -ENOSPC; 2164 goto fail; 2165 } 2166 2167 header_length = sizeof(*header) + s->unknown_header_fields_size; 2168 total_size = bs->total_sectors * BDRV_SECTOR_SIZE; 2169 refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3); 2170 2171 *header = (QCowHeader) { 2172 /* Version 2 fields */ 2173 .magic = cpu_to_be32(QCOW_MAGIC), 2174 .version = cpu_to_be32(s->qcow_version), 2175 .backing_file_offset = 0, 2176 .backing_file_size = 0, 2177 .cluster_bits = cpu_to_be32(s->cluster_bits), 2178 .size = cpu_to_be64(total_size), 2179 .crypt_method = cpu_to_be32(s->crypt_method_header), 2180 .l1_size = cpu_to_be32(s->l1_size), 2181 .l1_table_offset = cpu_to_be64(s->l1_table_offset), 2182 .refcount_table_offset = cpu_to_be64(s->refcount_table_offset), 2183 .refcount_table_clusters = cpu_to_be32(refcount_table_clusters), 2184 .nb_snapshots = cpu_to_be32(s->nb_snapshots), 2185 .snapshots_offset = cpu_to_be64(s->snapshots_offset), 2186 2187 /* Version 3 fields */ 2188 .incompatible_features = cpu_to_be64(s->incompatible_features), 2189 .compatible_features = cpu_to_be64(s->compatible_features), 2190 .autoclear_features = cpu_to_be64(s->autoclear_features), 2191 .refcount_order = cpu_to_be32(s->refcount_order), 2192 .header_length = cpu_to_be32(header_length), 2193 }; 2194 2195 /* For older versions, write a shorter header */ 2196 switch (s->qcow_version) { 2197 case 2: 2198 ret = offsetof(QCowHeader, incompatible_features); 2199 break; 2200 case 3: 2201 ret = sizeof(*header); 2202 break; 2203 default: 2204 ret = -EINVAL; 2205 goto fail; 2206 } 2207 2208 buf += ret; 2209 buflen -= ret; 2210 memset(buf, 0, buflen); 2211 2212 /* Preserve any unknown field in the header */ 2213 if (s->unknown_header_fields_size) { 2214 if (buflen < s->unknown_header_fields_size) { 2215 ret = -ENOSPC; 2216 goto fail; 2217 } 2218 2219 memcpy(buf, s->unknown_header_fields, s->unknown_header_fields_size); 2220 buf += s->unknown_header_fields_size; 2221 buflen -= s->unknown_header_fields_size; 2222 } 2223 2224 /* Backing file format header extension */ 2225 if (s->image_backing_format) { 2226 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BACKING_FORMAT, 2227 s->image_backing_format, 2228 strlen(s->image_backing_format), 2229 buflen); 2230 if (ret < 0) { 2231 goto fail; 2232 } 2233 2234 buf += ret; 2235 buflen -= ret; 2236 } 2237 2238 /* Full disk encryption header pointer extension */ 2239 if (s->crypto_header.offset != 0) { 2240 cpu_to_be64s(&s->crypto_header.offset); 2241 cpu_to_be64s(&s->crypto_header.length); 2242 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_CRYPTO_HEADER, 2243 &s->crypto_header, sizeof(s->crypto_header), 2244 buflen); 2245 be64_to_cpus(&s->crypto_header.offset); 2246 be64_to_cpus(&s->crypto_header.length); 2247 if (ret < 0) { 2248 goto fail; 2249 } 2250 buf += ret; 2251 buflen -= ret; 2252 } 2253 2254 /* Feature table */ 2255 if (s->qcow_version >= 3) { 2256 Qcow2Feature features[] = { 2257 { 2258 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE, 2259 .bit = QCOW2_INCOMPAT_DIRTY_BITNR, 2260 .name = "dirty bit", 2261 }, 2262 { 2263 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE, 2264 .bit = QCOW2_INCOMPAT_CORRUPT_BITNR, 2265 .name = "corrupt bit", 2266 }, 2267 { 2268 .type = QCOW2_FEAT_TYPE_COMPATIBLE, 2269 .bit = QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR, 2270 .name = "lazy refcounts", 2271 }, 2272 }; 2273 2274 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_FEATURE_TABLE, 2275 features, sizeof(features), buflen); 2276 if (ret < 0) { 2277 goto fail; 2278 } 2279 buf += ret; 2280 buflen -= ret; 2281 } 2282 2283 /* Bitmap extension */ 2284 if (s->nb_bitmaps > 0) { 2285 Qcow2BitmapHeaderExt bitmaps_header = { 2286 .nb_bitmaps = cpu_to_be32(s->nb_bitmaps), 2287 .bitmap_directory_size = 2288 cpu_to_be64(s->bitmap_directory_size), 2289 .bitmap_directory_offset = 2290 cpu_to_be64(s->bitmap_directory_offset) 2291 }; 2292 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BITMAPS, 2293 &bitmaps_header, sizeof(bitmaps_header), 2294 buflen); 2295 if (ret < 0) { 2296 goto fail; 2297 } 2298 buf += ret; 2299 buflen -= ret; 2300 } 2301 2302 /* Keep unknown header extensions */ 2303 QLIST_FOREACH(uext, &s->unknown_header_ext, next) { 2304 ret = header_ext_add(buf, uext->magic, uext->data, uext->len, buflen); 2305 if (ret < 0) { 2306 goto fail; 2307 } 2308 2309 buf += ret; 2310 buflen -= ret; 2311 } 2312 2313 /* End of header extensions */ 2314 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_END, NULL, 0, buflen); 2315 if (ret < 0) { 2316 goto fail; 2317 } 2318 2319 buf += ret; 2320 buflen -= ret; 2321 2322 /* Backing file name */ 2323 if (s->image_backing_file) { 2324 size_t backing_file_len = strlen(s->image_backing_file); 2325 2326 if (buflen < backing_file_len) { 2327 ret = -ENOSPC; 2328 goto fail; 2329 } 2330 2331 /* Using strncpy is ok here, since buf is not NUL-terminated. */ 2332 strncpy(buf, s->image_backing_file, buflen); 2333 2334 header->backing_file_offset = cpu_to_be64(buf - ((char*) header)); 2335 header->backing_file_size = cpu_to_be32(backing_file_len); 2336 } 2337 2338 /* Write the new header */ 2339 ret = bdrv_pwrite(bs->file, 0, header, s->cluster_size); 2340 if (ret < 0) { 2341 goto fail; 2342 } 2343 2344 ret = 0; 2345 fail: 2346 qemu_vfree(header); 2347 return ret; 2348 } 2349 2350 static int qcow2_change_backing_file(BlockDriverState *bs, 2351 const char *backing_file, const char *backing_fmt) 2352 { 2353 BDRVQcow2State *s = bs->opaque; 2354 2355 if (backing_file && strlen(backing_file) > 1023) { 2356 return -EINVAL; 2357 } 2358 2359 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: ""); 2360 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: ""); 2361 2362 g_free(s->image_backing_file); 2363 g_free(s->image_backing_format); 2364 2365 s->image_backing_file = backing_file ? g_strdup(bs->backing_file) : NULL; 2366 s->image_backing_format = backing_fmt ? g_strdup(bs->backing_format) : NULL; 2367 2368 return qcow2_update_header(bs); 2369 } 2370 2371 static int qcow2_crypt_method_from_format(const char *encryptfmt) 2372 { 2373 if (g_str_equal(encryptfmt, "luks")) { 2374 return QCOW_CRYPT_LUKS; 2375 } else if (g_str_equal(encryptfmt, "aes")) { 2376 return QCOW_CRYPT_AES; 2377 } else { 2378 return -EINVAL; 2379 } 2380 } 2381 2382 static int qcow2_set_up_encryption(BlockDriverState *bs, const char *encryptfmt, 2383 QemuOpts *opts, Error **errp) 2384 { 2385 BDRVQcow2State *s = bs->opaque; 2386 QCryptoBlockCreateOptions *cryptoopts = NULL; 2387 QCryptoBlock *crypto = NULL; 2388 int ret = -EINVAL; 2389 QDict *options, *encryptopts; 2390 int fmt; 2391 2392 options = qemu_opts_to_qdict(opts, NULL); 2393 qdict_extract_subqdict(options, &encryptopts, "encrypt."); 2394 QDECREF(options); 2395 2396 fmt = qcow2_crypt_method_from_format(encryptfmt); 2397 2398 switch (fmt) { 2399 case QCOW_CRYPT_LUKS: 2400 cryptoopts = block_crypto_create_opts_init( 2401 Q_CRYPTO_BLOCK_FORMAT_LUKS, encryptopts, errp); 2402 break; 2403 case QCOW_CRYPT_AES: 2404 cryptoopts = block_crypto_create_opts_init( 2405 Q_CRYPTO_BLOCK_FORMAT_QCOW, encryptopts, errp); 2406 break; 2407 default: 2408 error_setg(errp, "Unknown encryption format '%s'", encryptfmt); 2409 break; 2410 } 2411 if (!cryptoopts) { 2412 ret = -EINVAL; 2413 goto out; 2414 } 2415 s->crypt_method_header = fmt; 2416 2417 crypto = qcrypto_block_create(cryptoopts, "encrypt.", 2418 qcow2_crypto_hdr_init_func, 2419 qcow2_crypto_hdr_write_func, 2420 bs, errp); 2421 if (!crypto) { 2422 ret = -EINVAL; 2423 goto out; 2424 } 2425 2426 ret = qcow2_update_header(bs); 2427 if (ret < 0) { 2428 error_setg_errno(errp, -ret, "Could not write encryption header"); 2429 goto out; 2430 } 2431 2432 out: 2433 QDECREF(encryptopts); 2434 qcrypto_block_free(crypto); 2435 qapi_free_QCryptoBlockCreateOptions(cryptoopts); 2436 return ret; 2437 } 2438 2439 2440 typedef struct PreallocCo { 2441 BlockDriverState *bs; 2442 uint64_t offset; 2443 uint64_t new_length; 2444 2445 int ret; 2446 } PreallocCo; 2447 2448 /** 2449 * Preallocates metadata structures for data clusters between @offset (in the 2450 * guest disk) and @new_length (which is thus generally the new guest disk 2451 * size). 2452 * 2453 * Returns: 0 on success, -errno on failure. 2454 */ 2455 static void coroutine_fn preallocate_co(void *opaque) 2456 { 2457 PreallocCo *params = opaque; 2458 BlockDriverState *bs = params->bs; 2459 uint64_t offset = params->offset; 2460 uint64_t new_length = params->new_length; 2461 BDRVQcow2State *s = bs->opaque; 2462 uint64_t bytes; 2463 uint64_t host_offset = 0; 2464 unsigned int cur_bytes; 2465 int ret; 2466 QCowL2Meta *meta; 2467 2468 qemu_co_mutex_lock(&s->lock); 2469 2470 assert(offset <= new_length); 2471 bytes = new_length - offset; 2472 2473 while (bytes) { 2474 cur_bytes = MIN(bytes, INT_MAX); 2475 ret = qcow2_alloc_cluster_offset(bs, offset, &cur_bytes, 2476 &host_offset, &meta); 2477 if (ret < 0) { 2478 goto done; 2479 } 2480 2481 while (meta) { 2482 QCowL2Meta *next = meta->next; 2483 2484 ret = qcow2_alloc_cluster_link_l2(bs, meta); 2485 if (ret < 0) { 2486 qcow2_free_any_clusters(bs, meta->alloc_offset, 2487 meta->nb_clusters, QCOW2_DISCARD_NEVER); 2488 goto done; 2489 } 2490 2491 /* There are no dependent requests, but we need to remove our 2492 * request from the list of in-flight requests */ 2493 QLIST_REMOVE(meta, next_in_flight); 2494 2495 g_free(meta); 2496 meta = next; 2497 } 2498 2499 /* TODO Preallocate data if requested */ 2500 2501 bytes -= cur_bytes; 2502 offset += cur_bytes; 2503 } 2504 2505 /* 2506 * It is expected that the image file is large enough to actually contain 2507 * all of the allocated clusters (otherwise we get failing reads after 2508 * EOF). Extend the image to the last allocated sector. 2509 */ 2510 if (host_offset != 0) { 2511 uint8_t data = 0; 2512 ret = bdrv_pwrite(bs->file, (host_offset + cur_bytes) - 1, 2513 &data, 1); 2514 if (ret < 0) { 2515 goto done; 2516 } 2517 } 2518 2519 ret = 0; 2520 2521 done: 2522 qemu_co_mutex_unlock(&s->lock); 2523 params->ret = ret; 2524 } 2525 2526 static int preallocate(BlockDriverState *bs, 2527 uint64_t offset, uint64_t new_length) 2528 { 2529 PreallocCo params = { 2530 .bs = bs, 2531 .offset = offset, 2532 .new_length = new_length, 2533 .ret = -EINPROGRESS, 2534 }; 2535 2536 if (qemu_in_coroutine()) { 2537 preallocate_co(¶ms); 2538 } else { 2539 Coroutine *co = qemu_coroutine_create(preallocate_co, ¶ms); 2540 bdrv_coroutine_enter(bs, co); 2541 BDRV_POLL_WHILE(bs, params.ret == -EINPROGRESS); 2542 } 2543 return params.ret; 2544 } 2545 2546 /* qcow2_refcount_metadata_size: 2547 * @clusters: number of clusters to refcount (including data and L1/L2 tables) 2548 * @cluster_size: size of a cluster, in bytes 2549 * @refcount_order: refcount bits power-of-2 exponent 2550 * @generous_increase: allow for the refcount table to be 1.5x as large as it 2551 * needs to be 2552 * 2553 * Returns: Number of bytes required for refcount blocks and table metadata. 2554 */ 2555 int64_t qcow2_refcount_metadata_size(int64_t clusters, size_t cluster_size, 2556 int refcount_order, bool generous_increase, 2557 uint64_t *refblock_count) 2558 { 2559 /* 2560 * Every host cluster is reference-counted, including metadata (even 2561 * refcount metadata is recursively included). 2562 * 2563 * An accurate formula for the size of refcount metadata size is difficult 2564 * to derive. An easier method of calculation is finding the fixed point 2565 * where no further refcount blocks or table clusters are required to 2566 * reference count every cluster. 2567 */ 2568 int64_t blocks_per_table_cluster = cluster_size / sizeof(uint64_t); 2569 int64_t refcounts_per_block = cluster_size * 8 / (1 << refcount_order); 2570 int64_t table = 0; /* number of refcount table clusters */ 2571 int64_t blocks = 0; /* number of refcount block clusters */ 2572 int64_t last; 2573 int64_t n = 0; 2574 2575 do { 2576 last = n; 2577 blocks = DIV_ROUND_UP(clusters + table + blocks, refcounts_per_block); 2578 table = DIV_ROUND_UP(blocks, blocks_per_table_cluster); 2579 n = clusters + blocks + table; 2580 2581 if (n == last && generous_increase) { 2582 clusters += DIV_ROUND_UP(table, 2); 2583 n = 0; /* force another loop */ 2584 generous_increase = false; 2585 } 2586 } while (n != last); 2587 2588 if (refblock_count) { 2589 *refblock_count = blocks; 2590 } 2591 2592 return (blocks + table) * cluster_size; 2593 } 2594 2595 /** 2596 * qcow2_calc_prealloc_size: 2597 * @total_size: virtual disk size in bytes 2598 * @cluster_size: cluster size in bytes 2599 * @refcount_order: refcount bits power-of-2 exponent 2600 * 2601 * Returns: Total number of bytes required for the fully allocated image 2602 * (including metadata). 2603 */ 2604 static int64_t qcow2_calc_prealloc_size(int64_t total_size, 2605 size_t cluster_size, 2606 int refcount_order) 2607 { 2608 int64_t meta_size = 0; 2609 uint64_t nl1e, nl2e; 2610 int64_t aligned_total_size = align_offset(total_size, cluster_size); 2611 2612 /* header: 1 cluster */ 2613 meta_size += cluster_size; 2614 2615 /* total size of L2 tables */ 2616 nl2e = aligned_total_size / cluster_size; 2617 nl2e = align_offset(nl2e, cluster_size / sizeof(uint64_t)); 2618 meta_size += nl2e * sizeof(uint64_t); 2619 2620 /* total size of L1 tables */ 2621 nl1e = nl2e * sizeof(uint64_t) / cluster_size; 2622 nl1e = align_offset(nl1e, cluster_size / sizeof(uint64_t)); 2623 meta_size += nl1e * sizeof(uint64_t); 2624 2625 /* total size of refcount table and blocks */ 2626 meta_size += qcow2_refcount_metadata_size( 2627 (meta_size + aligned_total_size) / cluster_size, 2628 cluster_size, refcount_order, false, NULL); 2629 2630 return meta_size + aligned_total_size; 2631 } 2632 2633 static size_t qcow2_opt_get_cluster_size_del(QemuOpts *opts, Error **errp) 2634 { 2635 size_t cluster_size; 2636 int cluster_bits; 2637 2638 cluster_size = qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE, 2639 DEFAULT_CLUSTER_SIZE); 2640 cluster_bits = ctz32(cluster_size); 2641 if (cluster_bits < MIN_CLUSTER_BITS || cluster_bits > MAX_CLUSTER_BITS || 2642 (1 << cluster_bits) != cluster_size) 2643 { 2644 error_setg(errp, "Cluster size must be a power of two between %d and " 2645 "%dk", 1 << MIN_CLUSTER_BITS, 1 << (MAX_CLUSTER_BITS - 10)); 2646 return 0; 2647 } 2648 return cluster_size; 2649 } 2650 2651 static int qcow2_opt_get_version_del(QemuOpts *opts, Error **errp) 2652 { 2653 char *buf; 2654 int ret; 2655 2656 buf = qemu_opt_get_del(opts, BLOCK_OPT_COMPAT_LEVEL); 2657 if (!buf) { 2658 ret = 3; /* default */ 2659 } else if (!strcmp(buf, "0.10")) { 2660 ret = 2; 2661 } else if (!strcmp(buf, "1.1")) { 2662 ret = 3; 2663 } else { 2664 error_setg(errp, "Invalid compatibility level: '%s'", buf); 2665 ret = -EINVAL; 2666 } 2667 g_free(buf); 2668 return ret; 2669 } 2670 2671 static uint64_t qcow2_opt_get_refcount_bits_del(QemuOpts *opts, int version, 2672 Error **errp) 2673 { 2674 uint64_t refcount_bits; 2675 2676 refcount_bits = qemu_opt_get_number_del(opts, BLOCK_OPT_REFCOUNT_BITS, 16); 2677 if (refcount_bits > 64 || !is_power_of_2(refcount_bits)) { 2678 error_setg(errp, "Refcount width must be a power of two and may not " 2679 "exceed 64 bits"); 2680 return 0; 2681 } 2682 2683 if (version < 3 && refcount_bits != 16) { 2684 error_setg(errp, "Different refcount widths than 16 bits require " 2685 "compatibility level 1.1 or above (use compat=1.1 or " 2686 "greater)"); 2687 return 0; 2688 } 2689 2690 return refcount_bits; 2691 } 2692 2693 static int qcow2_create2(const char *filename, int64_t total_size, 2694 const char *backing_file, const char *backing_format, 2695 int flags, size_t cluster_size, PreallocMode prealloc, 2696 QemuOpts *opts, int version, int refcount_order, 2697 const char *encryptfmt, Error **errp) 2698 { 2699 QDict *options; 2700 2701 /* 2702 * Open the image file and write a minimal qcow2 header. 2703 * 2704 * We keep things simple and start with a zero-sized image. We also 2705 * do without refcount blocks or a L1 table for now. We'll fix the 2706 * inconsistency later. 2707 * 2708 * We do need a refcount table because growing the refcount table means 2709 * allocating two new refcount blocks - the seconds of which would be at 2710 * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file 2711 * size for any qcow2 image. 2712 */ 2713 BlockBackend *blk; 2714 QCowHeader *header; 2715 uint64_t* refcount_table; 2716 Error *local_err = NULL; 2717 int ret; 2718 2719 if (prealloc == PREALLOC_MODE_FULL || prealloc == PREALLOC_MODE_FALLOC) { 2720 int64_t prealloc_size = 2721 qcow2_calc_prealloc_size(total_size, cluster_size, refcount_order); 2722 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, prealloc_size, &error_abort); 2723 qemu_opt_set(opts, BLOCK_OPT_PREALLOC, PreallocMode_str(prealloc), 2724 &error_abort); 2725 } 2726 2727 ret = bdrv_create_file(filename, opts, &local_err); 2728 if (ret < 0) { 2729 error_propagate(errp, local_err); 2730 return ret; 2731 } 2732 2733 blk = blk_new_open(filename, NULL, NULL, 2734 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, 2735 &local_err); 2736 if (blk == NULL) { 2737 error_propagate(errp, local_err); 2738 return -EIO; 2739 } 2740 2741 blk_set_allow_write_beyond_eof(blk, true); 2742 2743 /* Write the header */ 2744 QEMU_BUILD_BUG_ON((1 << MIN_CLUSTER_BITS) < sizeof(*header)); 2745 header = g_malloc0(cluster_size); 2746 *header = (QCowHeader) { 2747 .magic = cpu_to_be32(QCOW_MAGIC), 2748 .version = cpu_to_be32(version), 2749 .cluster_bits = cpu_to_be32(ctz32(cluster_size)), 2750 .size = cpu_to_be64(0), 2751 .l1_table_offset = cpu_to_be64(0), 2752 .l1_size = cpu_to_be32(0), 2753 .refcount_table_offset = cpu_to_be64(cluster_size), 2754 .refcount_table_clusters = cpu_to_be32(1), 2755 .refcount_order = cpu_to_be32(refcount_order), 2756 .header_length = cpu_to_be32(sizeof(*header)), 2757 }; 2758 2759 /* We'll update this to correct value later */ 2760 header->crypt_method = cpu_to_be32(QCOW_CRYPT_NONE); 2761 2762 if (flags & BLOCK_FLAG_LAZY_REFCOUNTS) { 2763 header->compatible_features |= 2764 cpu_to_be64(QCOW2_COMPAT_LAZY_REFCOUNTS); 2765 } 2766 2767 ret = blk_pwrite(blk, 0, header, cluster_size, 0); 2768 g_free(header); 2769 if (ret < 0) { 2770 error_setg_errno(errp, -ret, "Could not write qcow2 header"); 2771 goto out; 2772 } 2773 2774 /* Write a refcount table with one refcount block */ 2775 refcount_table = g_malloc0(2 * cluster_size); 2776 refcount_table[0] = cpu_to_be64(2 * cluster_size); 2777 ret = blk_pwrite(blk, cluster_size, refcount_table, 2 * cluster_size, 0); 2778 g_free(refcount_table); 2779 2780 if (ret < 0) { 2781 error_setg_errno(errp, -ret, "Could not write refcount table"); 2782 goto out; 2783 } 2784 2785 blk_unref(blk); 2786 blk = NULL; 2787 2788 /* 2789 * And now open the image and make it consistent first (i.e. increase the 2790 * refcount of the cluster that is occupied by the header and the refcount 2791 * table) 2792 */ 2793 options = qdict_new(); 2794 qdict_put_str(options, "driver", "qcow2"); 2795 blk = blk_new_open(filename, NULL, options, 2796 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_NO_FLUSH, 2797 &local_err); 2798 if (blk == NULL) { 2799 error_propagate(errp, local_err); 2800 ret = -EIO; 2801 goto out; 2802 } 2803 2804 ret = qcow2_alloc_clusters(blk_bs(blk), 3 * cluster_size); 2805 if (ret < 0) { 2806 error_setg_errno(errp, -ret, "Could not allocate clusters for qcow2 " 2807 "header and refcount table"); 2808 goto out; 2809 2810 } else if (ret != 0) { 2811 error_report("Huh, first cluster in empty image is already in use?"); 2812 abort(); 2813 } 2814 2815 /* Create a full header (including things like feature table) */ 2816 ret = qcow2_update_header(blk_bs(blk)); 2817 if (ret < 0) { 2818 error_setg_errno(errp, -ret, "Could not update qcow2 header"); 2819 goto out; 2820 } 2821 2822 /* Okay, now that we have a valid image, let's give it the right size */ 2823 ret = blk_truncate(blk, total_size, PREALLOC_MODE_OFF, errp); 2824 if (ret < 0) { 2825 error_prepend(errp, "Could not resize image: "); 2826 goto out; 2827 } 2828 2829 /* Want a backing file? There you go.*/ 2830 if (backing_file) { 2831 ret = bdrv_change_backing_file(blk_bs(blk), backing_file, backing_format); 2832 if (ret < 0) { 2833 error_setg_errno(errp, -ret, "Could not assign backing file '%s' " 2834 "with format '%s'", backing_file, backing_format); 2835 goto out; 2836 } 2837 } 2838 2839 /* Want encryption? There you go. */ 2840 if (encryptfmt) { 2841 ret = qcow2_set_up_encryption(blk_bs(blk), encryptfmt, opts, errp); 2842 if (ret < 0) { 2843 goto out; 2844 } 2845 } 2846 2847 /* And if we're supposed to preallocate metadata, do that now */ 2848 if (prealloc != PREALLOC_MODE_OFF) { 2849 ret = preallocate(blk_bs(blk), 0, total_size); 2850 if (ret < 0) { 2851 error_setg_errno(errp, -ret, "Could not preallocate metadata"); 2852 goto out; 2853 } 2854 } 2855 2856 blk_unref(blk); 2857 blk = NULL; 2858 2859 /* Reopen the image without BDRV_O_NO_FLUSH to flush it before returning. 2860 * Using BDRV_O_NO_IO, since encryption is now setup we don't want to 2861 * have to setup decryption context. We're not doing any I/O on the top 2862 * level BlockDriverState, only lower layers, where BDRV_O_NO_IO does 2863 * not have effect. 2864 */ 2865 options = qdict_new(); 2866 qdict_put_str(options, "driver", "qcow2"); 2867 blk = blk_new_open(filename, NULL, options, 2868 BDRV_O_RDWR | BDRV_O_NO_BACKING | BDRV_O_NO_IO, 2869 &local_err); 2870 if (blk == NULL) { 2871 error_propagate(errp, local_err); 2872 ret = -EIO; 2873 goto out; 2874 } 2875 2876 ret = 0; 2877 out: 2878 if (blk) { 2879 blk_unref(blk); 2880 } 2881 return ret; 2882 } 2883 2884 static int qcow2_create(const char *filename, QemuOpts *opts, Error **errp) 2885 { 2886 char *backing_file = NULL; 2887 char *backing_fmt = NULL; 2888 char *buf = NULL; 2889 uint64_t size = 0; 2890 int flags = 0; 2891 size_t cluster_size = DEFAULT_CLUSTER_SIZE; 2892 PreallocMode prealloc; 2893 int version; 2894 uint64_t refcount_bits; 2895 int refcount_order; 2896 char *encryptfmt = NULL; 2897 Error *local_err = NULL; 2898 int ret; 2899 2900 /* Read out options */ 2901 size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0), 2902 BDRV_SECTOR_SIZE); 2903 backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE); 2904 backing_fmt = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FMT); 2905 encryptfmt = qemu_opt_get_del(opts, BLOCK_OPT_ENCRYPT_FORMAT); 2906 if (encryptfmt) { 2907 if (qemu_opt_get(opts, BLOCK_OPT_ENCRYPT)) { 2908 error_setg(errp, "Options " BLOCK_OPT_ENCRYPT " and " 2909 BLOCK_OPT_ENCRYPT_FORMAT " are mutually exclusive"); 2910 ret = -EINVAL; 2911 goto finish; 2912 } 2913 } else if (qemu_opt_get_bool_del(opts, BLOCK_OPT_ENCRYPT, false)) { 2914 encryptfmt = g_strdup("aes"); 2915 } 2916 cluster_size = qcow2_opt_get_cluster_size_del(opts, &local_err); 2917 if (local_err) { 2918 error_propagate(errp, local_err); 2919 ret = -EINVAL; 2920 goto finish; 2921 } 2922 buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC); 2923 prealloc = qapi_enum_parse(&PreallocMode_lookup, buf, 2924 PREALLOC_MODE_OFF, &local_err); 2925 if (local_err) { 2926 error_propagate(errp, local_err); 2927 ret = -EINVAL; 2928 goto finish; 2929 } 2930 2931 version = qcow2_opt_get_version_del(opts, &local_err); 2932 if (local_err) { 2933 error_propagate(errp, local_err); 2934 ret = -EINVAL; 2935 goto finish; 2936 } 2937 2938 if (qemu_opt_get_bool_del(opts, BLOCK_OPT_LAZY_REFCOUNTS, false)) { 2939 flags |= BLOCK_FLAG_LAZY_REFCOUNTS; 2940 } 2941 2942 if (backing_file && prealloc != PREALLOC_MODE_OFF) { 2943 error_setg(errp, "Backing file and preallocation cannot be used at " 2944 "the same time"); 2945 ret = -EINVAL; 2946 goto finish; 2947 } 2948 2949 if (version < 3 && (flags & BLOCK_FLAG_LAZY_REFCOUNTS)) { 2950 error_setg(errp, "Lazy refcounts only supported with compatibility " 2951 "level 1.1 and above (use compat=1.1 or greater)"); 2952 ret = -EINVAL; 2953 goto finish; 2954 } 2955 2956 refcount_bits = qcow2_opt_get_refcount_bits_del(opts, version, &local_err); 2957 if (local_err) { 2958 error_propagate(errp, local_err); 2959 ret = -EINVAL; 2960 goto finish; 2961 } 2962 2963 refcount_order = ctz32(refcount_bits); 2964 2965 ret = qcow2_create2(filename, size, backing_file, backing_fmt, flags, 2966 cluster_size, prealloc, opts, version, refcount_order, 2967 encryptfmt, &local_err); 2968 error_propagate(errp, local_err); 2969 2970 finish: 2971 g_free(backing_file); 2972 g_free(backing_fmt); 2973 g_free(encryptfmt); 2974 g_free(buf); 2975 return ret; 2976 } 2977 2978 2979 static bool is_zero(BlockDriverState *bs, int64_t offset, int64_t bytes) 2980 { 2981 int64_t nr; 2982 int res; 2983 2984 /* Clamp to image length, before checking status of underlying sectors */ 2985 if (offset + bytes > bs->total_sectors * BDRV_SECTOR_SIZE) { 2986 bytes = bs->total_sectors * BDRV_SECTOR_SIZE - offset; 2987 } 2988 2989 if (!bytes) { 2990 return true; 2991 } 2992 res = bdrv_block_status_above(bs, NULL, offset, bytes, &nr, NULL, NULL); 2993 return res >= 0 && (res & BDRV_BLOCK_ZERO) && nr == bytes; 2994 } 2995 2996 static coroutine_fn int qcow2_co_pwrite_zeroes(BlockDriverState *bs, 2997 int64_t offset, int bytes, BdrvRequestFlags flags) 2998 { 2999 int ret; 3000 BDRVQcow2State *s = bs->opaque; 3001 3002 uint32_t head = offset % s->cluster_size; 3003 uint32_t tail = (offset + bytes) % s->cluster_size; 3004 3005 trace_qcow2_pwrite_zeroes_start_req(qemu_coroutine_self(), offset, bytes); 3006 if (offset + bytes == bs->total_sectors * BDRV_SECTOR_SIZE) { 3007 tail = 0; 3008 } 3009 3010 if (head || tail) { 3011 uint64_t off; 3012 unsigned int nr; 3013 3014 assert(head + bytes <= s->cluster_size); 3015 3016 /* check whether remainder of cluster already reads as zero */ 3017 if (!(is_zero(bs, offset - head, head) && 3018 is_zero(bs, offset + bytes, 3019 tail ? s->cluster_size - tail : 0))) { 3020 return -ENOTSUP; 3021 } 3022 3023 qemu_co_mutex_lock(&s->lock); 3024 /* We can have new write after previous check */ 3025 offset = QEMU_ALIGN_DOWN(offset, s->cluster_size); 3026 bytes = s->cluster_size; 3027 nr = s->cluster_size; 3028 ret = qcow2_get_cluster_offset(bs, offset, &nr, &off); 3029 if (ret != QCOW2_CLUSTER_UNALLOCATED && 3030 ret != QCOW2_CLUSTER_ZERO_PLAIN && 3031 ret != QCOW2_CLUSTER_ZERO_ALLOC) { 3032 qemu_co_mutex_unlock(&s->lock); 3033 return -ENOTSUP; 3034 } 3035 } else { 3036 qemu_co_mutex_lock(&s->lock); 3037 } 3038 3039 trace_qcow2_pwrite_zeroes(qemu_coroutine_self(), offset, bytes); 3040 3041 /* Whatever is left can use real zero clusters */ 3042 ret = qcow2_cluster_zeroize(bs, offset, bytes, flags); 3043 qemu_co_mutex_unlock(&s->lock); 3044 3045 return ret; 3046 } 3047 3048 static coroutine_fn int qcow2_co_pdiscard(BlockDriverState *bs, 3049 int64_t offset, int bytes) 3050 { 3051 int ret; 3052 BDRVQcow2State *s = bs->opaque; 3053 3054 if (!QEMU_IS_ALIGNED(offset | bytes, s->cluster_size)) { 3055 assert(bytes < s->cluster_size); 3056 /* Ignore partial clusters, except for the special case of the 3057 * complete partial cluster at the end of an unaligned file */ 3058 if (!QEMU_IS_ALIGNED(offset, s->cluster_size) || 3059 offset + bytes != bs->total_sectors * BDRV_SECTOR_SIZE) { 3060 return -ENOTSUP; 3061 } 3062 } 3063 3064 qemu_co_mutex_lock(&s->lock); 3065 ret = qcow2_cluster_discard(bs, offset, bytes, QCOW2_DISCARD_REQUEST, 3066 false); 3067 qemu_co_mutex_unlock(&s->lock); 3068 return ret; 3069 } 3070 3071 static int qcow2_truncate(BlockDriverState *bs, int64_t offset, 3072 PreallocMode prealloc, Error **errp) 3073 { 3074 BDRVQcow2State *s = bs->opaque; 3075 uint64_t old_length; 3076 int64_t new_l1_size; 3077 int ret; 3078 3079 if (prealloc != PREALLOC_MODE_OFF && prealloc != PREALLOC_MODE_METADATA && 3080 prealloc != PREALLOC_MODE_FALLOC && prealloc != PREALLOC_MODE_FULL) 3081 { 3082 error_setg(errp, "Unsupported preallocation mode '%s'", 3083 PreallocMode_str(prealloc)); 3084 return -ENOTSUP; 3085 } 3086 3087 if (offset & 511) { 3088 error_setg(errp, "The new size must be a multiple of 512"); 3089 return -EINVAL; 3090 } 3091 3092 /* cannot proceed if image has snapshots */ 3093 if (s->nb_snapshots) { 3094 error_setg(errp, "Can't resize an image which has snapshots"); 3095 return -ENOTSUP; 3096 } 3097 3098 /* cannot proceed if image has bitmaps */ 3099 if (s->nb_bitmaps) { 3100 /* TODO: resize bitmaps in the image */ 3101 error_setg(errp, "Can't resize an image which has bitmaps"); 3102 return -ENOTSUP; 3103 } 3104 3105 old_length = bs->total_sectors * 512; 3106 new_l1_size = size_to_l1(s, offset); 3107 3108 if (offset < old_length) { 3109 int64_t last_cluster, old_file_size; 3110 if (prealloc != PREALLOC_MODE_OFF) { 3111 error_setg(errp, 3112 "Preallocation can't be used for shrinking an image"); 3113 return -EINVAL; 3114 } 3115 3116 ret = qcow2_cluster_discard(bs, ROUND_UP(offset, s->cluster_size), 3117 old_length - ROUND_UP(offset, 3118 s->cluster_size), 3119 QCOW2_DISCARD_ALWAYS, true); 3120 if (ret < 0) { 3121 error_setg_errno(errp, -ret, "Failed to discard cropped clusters"); 3122 return ret; 3123 } 3124 3125 ret = qcow2_shrink_l1_table(bs, new_l1_size); 3126 if (ret < 0) { 3127 error_setg_errno(errp, -ret, 3128 "Failed to reduce the number of L2 tables"); 3129 return ret; 3130 } 3131 3132 ret = qcow2_shrink_reftable(bs); 3133 if (ret < 0) { 3134 error_setg_errno(errp, -ret, 3135 "Failed to discard unused refblocks"); 3136 return ret; 3137 } 3138 3139 old_file_size = bdrv_getlength(bs->file->bs); 3140 if (old_file_size < 0) { 3141 error_setg_errno(errp, -old_file_size, 3142 "Failed to inquire current file length"); 3143 return old_file_size; 3144 } 3145 last_cluster = qcow2_get_last_cluster(bs, old_file_size); 3146 if (last_cluster < 0) { 3147 error_setg_errno(errp, -last_cluster, 3148 "Failed to find the last cluster"); 3149 return last_cluster; 3150 } 3151 if ((last_cluster + 1) * s->cluster_size < old_file_size) { 3152 Error *local_err = NULL; 3153 3154 bdrv_truncate(bs->file, (last_cluster + 1) * s->cluster_size, 3155 PREALLOC_MODE_OFF, &local_err); 3156 if (local_err) { 3157 warn_reportf_err(local_err, 3158 "Failed to truncate the tail of the image: "); 3159 } 3160 } 3161 } else { 3162 ret = qcow2_grow_l1_table(bs, new_l1_size, true); 3163 if (ret < 0) { 3164 error_setg_errno(errp, -ret, "Failed to grow the L1 table"); 3165 return ret; 3166 } 3167 } 3168 3169 switch (prealloc) { 3170 case PREALLOC_MODE_OFF: 3171 break; 3172 3173 case PREALLOC_MODE_METADATA: 3174 ret = preallocate(bs, old_length, offset); 3175 if (ret < 0) { 3176 error_setg_errno(errp, -ret, "Preallocation failed"); 3177 return ret; 3178 } 3179 break; 3180 3181 case PREALLOC_MODE_FALLOC: 3182 case PREALLOC_MODE_FULL: 3183 { 3184 int64_t allocation_start, host_offset, guest_offset; 3185 int64_t clusters_allocated; 3186 int64_t old_file_size, new_file_size; 3187 uint64_t nb_new_data_clusters, nb_new_l2_tables; 3188 3189 old_file_size = bdrv_getlength(bs->file->bs); 3190 if (old_file_size < 0) { 3191 error_setg_errno(errp, -old_file_size, 3192 "Failed to inquire current file length"); 3193 return old_file_size; 3194 } 3195 old_file_size = ROUND_UP(old_file_size, s->cluster_size); 3196 3197 nb_new_data_clusters = DIV_ROUND_UP(offset - old_length, 3198 s->cluster_size); 3199 3200 /* This is an overestimation; we will not actually allocate space for 3201 * these in the file but just make sure the new refcount structures are 3202 * able to cover them so we will not have to allocate new refblocks 3203 * while entering the data blocks in the potentially new L2 tables. 3204 * (We do not actually care where the L2 tables are placed. Maybe they 3205 * are already allocated or they can be placed somewhere before 3206 * @old_file_size. It does not matter because they will be fully 3207 * allocated automatically, so they do not need to be covered by the 3208 * preallocation. All that matters is that we will not have to allocate 3209 * new refcount structures for them.) */ 3210 nb_new_l2_tables = DIV_ROUND_UP(nb_new_data_clusters, 3211 s->cluster_size / sizeof(uint64_t)); 3212 /* The cluster range may not be aligned to L2 boundaries, so add one L2 3213 * table for a potential head/tail */ 3214 nb_new_l2_tables++; 3215 3216 allocation_start = qcow2_refcount_area(bs, old_file_size, 3217 nb_new_data_clusters + 3218 nb_new_l2_tables, 3219 true, 0, 0); 3220 if (allocation_start < 0) { 3221 error_setg_errno(errp, -allocation_start, 3222 "Failed to resize refcount structures"); 3223 return allocation_start; 3224 } 3225 3226 clusters_allocated = qcow2_alloc_clusters_at(bs, allocation_start, 3227 nb_new_data_clusters); 3228 if (clusters_allocated < 0) { 3229 error_setg_errno(errp, -clusters_allocated, 3230 "Failed to allocate data clusters"); 3231 return -clusters_allocated; 3232 } 3233 3234 assert(clusters_allocated == nb_new_data_clusters); 3235 3236 /* Allocate the data area */ 3237 new_file_size = allocation_start + 3238 nb_new_data_clusters * s->cluster_size; 3239 ret = bdrv_truncate(bs->file, new_file_size, prealloc, errp); 3240 if (ret < 0) { 3241 error_prepend(errp, "Failed to resize underlying file: "); 3242 qcow2_free_clusters(bs, allocation_start, 3243 nb_new_data_clusters * s->cluster_size, 3244 QCOW2_DISCARD_OTHER); 3245 return ret; 3246 } 3247 3248 /* Create the necessary L2 entries */ 3249 host_offset = allocation_start; 3250 guest_offset = old_length; 3251 while (nb_new_data_clusters) { 3252 int64_t guest_cluster = guest_offset >> s->cluster_bits; 3253 int64_t nb_clusters = MIN(nb_new_data_clusters, 3254 s->l2_size - guest_cluster % s->l2_size); 3255 QCowL2Meta allocation = { 3256 .offset = guest_offset, 3257 .alloc_offset = host_offset, 3258 .nb_clusters = nb_clusters, 3259 }; 3260 qemu_co_queue_init(&allocation.dependent_requests); 3261 3262 ret = qcow2_alloc_cluster_link_l2(bs, &allocation); 3263 if (ret < 0) { 3264 error_setg_errno(errp, -ret, "Failed to update L2 tables"); 3265 qcow2_free_clusters(bs, host_offset, 3266 nb_new_data_clusters * s->cluster_size, 3267 QCOW2_DISCARD_OTHER); 3268 return ret; 3269 } 3270 3271 guest_offset += nb_clusters * s->cluster_size; 3272 host_offset += nb_clusters * s->cluster_size; 3273 nb_new_data_clusters -= nb_clusters; 3274 } 3275 break; 3276 } 3277 3278 default: 3279 g_assert_not_reached(); 3280 } 3281 3282 if (prealloc != PREALLOC_MODE_OFF) { 3283 /* Flush metadata before actually changing the image size */ 3284 ret = bdrv_flush(bs); 3285 if (ret < 0) { 3286 error_setg_errno(errp, -ret, 3287 "Failed to flush the preallocated area to disk"); 3288 return ret; 3289 } 3290 } 3291 3292 /* write updated header.size */ 3293 offset = cpu_to_be64(offset); 3294 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, size), 3295 &offset, sizeof(uint64_t)); 3296 if (ret < 0) { 3297 error_setg_errno(errp, -ret, "Failed to update the image size"); 3298 return ret; 3299 } 3300 3301 s->l1_vm_state_index = new_l1_size; 3302 return 0; 3303 } 3304 3305 /* XXX: put compressed sectors first, then all the cluster aligned 3306 tables to avoid losing bytes in alignment */ 3307 static coroutine_fn int 3308 qcow2_co_pwritev_compressed(BlockDriverState *bs, uint64_t offset, 3309 uint64_t bytes, QEMUIOVector *qiov) 3310 { 3311 BDRVQcow2State *s = bs->opaque; 3312 QEMUIOVector hd_qiov; 3313 struct iovec iov; 3314 z_stream strm; 3315 int ret, out_len; 3316 uint8_t *buf, *out_buf; 3317 int64_t cluster_offset; 3318 3319 if (bytes == 0) { 3320 /* align end of file to a sector boundary to ease reading with 3321 sector based I/Os */ 3322 cluster_offset = bdrv_getlength(bs->file->bs); 3323 if (cluster_offset < 0) { 3324 return cluster_offset; 3325 } 3326 return bdrv_truncate(bs->file, cluster_offset, PREALLOC_MODE_OFF, NULL); 3327 } 3328 3329 if (offset_into_cluster(s, offset)) { 3330 return -EINVAL; 3331 } 3332 3333 buf = qemu_blockalign(bs, s->cluster_size); 3334 if (bytes != s->cluster_size) { 3335 if (bytes > s->cluster_size || 3336 offset + bytes != bs->total_sectors << BDRV_SECTOR_BITS) 3337 { 3338 qemu_vfree(buf); 3339 return -EINVAL; 3340 } 3341 /* Zero-pad last write if image size is not cluster aligned */ 3342 memset(buf + bytes, 0, s->cluster_size - bytes); 3343 } 3344 qemu_iovec_to_buf(qiov, 0, buf, bytes); 3345 3346 out_buf = g_malloc(s->cluster_size); 3347 3348 /* best compression, small window, no zlib header */ 3349 memset(&strm, 0, sizeof(strm)); 3350 ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION, 3351 Z_DEFLATED, -12, 3352 9, Z_DEFAULT_STRATEGY); 3353 if (ret != 0) { 3354 ret = -EINVAL; 3355 goto fail; 3356 } 3357 3358 strm.avail_in = s->cluster_size; 3359 strm.next_in = (uint8_t *)buf; 3360 strm.avail_out = s->cluster_size; 3361 strm.next_out = out_buf; 3362 3363 ret = deflate(&strm, Z_FINISH); 3364 if (ret != Z_STREAM_END && ret != Z_OK) { 3365 deflateEnd(&strm); 3366 ret = -EINVAL; 3367 goto fail; 3368 } 3369 out_len = strm.next_out - out_buf; 3370 3371 deflateEnd(&strm); 3372 3373 if (ret != Z_STREAM_END || out_len >= s->cluster_size) { 3374 /* could not compress: write normal cluster */ 3375 ret = qcow2_co_pwritev(bs, offset, bytes, qiov, 0); 3376 if (ret < 0) { 3377 goto fail; 3378 } 3379 goto success; 3380 } 3381 3382 qemu_co_mutex_lock(&s->lock); 3383 cluster_offset = 3384 qcow2_alloc_compressed_cluster_offset(bs, offset, out_len); 3385 if (!cluster_offset) { 3386 qemu_co_mutex_unlock(&s->lock); 3387 ret = -EIO; 3388 goto fail; 3389 } 3390 cluster_offset &= s->cluster_offset_mask; 3391 3392 ret = qcow2_pre_write_overlap_check(bs, 0, cluster_offset, out_len); 3393 qemu_co_mutex_unlock(&s->lock); 3394 if (ret < 0) { 3395 goto fail; 3396 } 3397 3398 iov = (struct iovec) { 3399 .iov_base = out_buf, 3400 .iov_len = out_len, 3401 }; 3402 qemu_iovec_init_external(&hd_qiov, &iov, 1); 3403 3404 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_COMPRESSED); 3405 ret = bdrv_co_pwritev(bs->file, cluster_offset, out_len, &hd_qiov, 0); 3406 if (ret < 0) { 3407 goto fail; 3408 } 3409 success: 3410 ret = 0; 3411 fail: 3412 qemu_vfree(buf); 3413 g_free(out_buf); 3414 return ret; 3415 } 3416 3417 static int make_completely_empty(BlockDriverState *bs) 3418 { 3419 BDRVQcow2State *s = bs->opaque; 3420 Error *local_err = NULL; 3421 int ret, l1_clusters; 3422 int64_t offset; 3423 uint64_t *new_reftable = NULL; 3424 uint64_t rt_entry, l1_size2; 3425 struct { 3426 uint64_t l1_offset; 3427 uint64_t reftable_offset; 3428 uint32_t reftable_clusters; 3429 } QEMU_PACKED l1_ofs_rt_ofs_cls; 3430 3431 ret = qcow2_cache_empty(bs, s->l2_table_cache); 3432 if (ret < 0) { 3433 goto fail; 3434 } 3435 3436 ret = qcow2_cache_empty(bs, s->refcount_block_cache); 3437 if (ret < 0) { 3438 goto fail; 3439 } 3440 3441 /* Refcounts will be broken utterly */ 3442 ret = qcow2_mark_dirty(bs); 3443 if (ret < 0) { 3444 goto fail; 3445 } 3446 3447 BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE); 3448 3449 l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / sizeof(uint64_t)); 3450 l1_size2 = (uint64_t)s->l1_size * sizeof(uint64_t); 3451 3452 /* After this call, neither the in-memory nor the on-disk refcount 3453 * information accurately describe the actual references */ 3454 3455 ret = bdrv_pwrite_zeroes(bs->file, s->l1_table_offset, 3456 l1_clusters * s->cluster_size, 0); 3457 if (ret < 0) { 3458 goto fail_broken_refcounts; 3459 } 3460 memset(s->l1_table, 0, l1_size2); 3461 3462 BLKDBG_EVENT(bs->file, BLKDBG_EMPTY_IMAGE_PREPARE); 3463 3464 /* Overwrite enough clusters at the beginning of the sectors to place 3465 * the refcount table, a refcount block and the L1 table in; this may 3466 * overwrite parts of the existing refcount and L1 table, which is not 3467 * an issue because the dirty flag is set, complete data loss is in fact 3468 * desired and partial data loss is consequently fine as well */ 3469 ret = bdrv_pwrite_zeroes(bs->file, s->cluster_size, 3470 (2 + l1_clusters) * s->cluster_size, 0); 3471 /* This call (even if it failed overall) may have overwritten on-disk 3472 * refcount structures; in that case, the in-memory refcount information 3473 * will probably differ from the on-disk information which makes the BDS 3474 * unusable */ 3475 if (ret < 0) { 3476 goto fail_broken_refcounts; 3477 } 3478 3479 BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE); 3480 BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_UPDATE); 3481 3482 /* "Create" an empty reftable (one cluster) directly after the image 3483 * header and an empty L1 table three clusters after the image header; 3484 * the cluster between those two will be used as the first refblock */ 3485 l1_ofs_rt_ofs_cls.l1_offset = cpu_to_be64(3 * s->cluster_size); 3486 l1_ofs_rt_ofs_cls.reftable_offset = cpu_to_be64(s->cluster_size); 3487 l1_ofs_rt_ofs_cls.reftable_clusters = cpu_to_be32(1); 3488 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, l1_table_offset), 3489 &l1_ofs_rt_ofs_cls, sizeof(l1_ofs_rt_ofs_cls)); 3490 if (ret < 0) { 3491 goto fail_broken_refcounts; 3492 } 3493 3494 s->l1_table_offset = 3 * s->cluster_size; 3495 3496 new_reftable = g_try_new0(uint64_t, s->cluster_size / sizeof(uint64_t)); 3497 if (!new_reftable) { 3498 ret = -ENOMEM; 3499 goto fail_broken_refcounts; 3500 } 3501 3502 s->refcount_table_offset = s->cluster_size; 3503 s->refcount_table_size = s->cluster_size / sizeof(uint64_t); 3504 s->max_refcount_table_index = 0; 3505 3506 g_free(s->refcount_table); 3507 s->refcount_table = new_reftable; 3508 new_reftable = NULL; 3509 3510 /* Now the in-memory refcount information again corresponds to the on-disk 3511 * information (reftable is empty and no refblocks (the refblock cache is 3512 * empty)); however, this means some clusters (e.g. the image header) are 3513 * referenced, but not refcounted, but the normal qcow2 code assumes that 3514 * the in-memory information is always correct */ 3515 3516 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC); 3517 3518 /* Enter the first refblock into the reftable */ 3519 rt_entry = cpu_to_be64(2 * s->cluster_size); 3520 ret = bdrv_pwrite_sync(bs->file, s->cluster_size, 3521 &rt_entry, sizeof(rt_entry)); 3522 if (ret < 0) { 3523 goto fail_broken_refcounts; 3524 } 3525 s->refcount_table[0] = 2 * s->cluster_size; 3526 3527 s->free_cluster_index = 0; 3528 assert(3 + l1_clusters <= s->refcount_block_size); 3529 offset = qcow2_alloc_clusters(bs, 3 * s->cluster_size + l1_size2); 3530 if (offset < 0) { 3531 ret = offset; 3532 goto fail_broken_refcounts; 3533 } else if (offset > 0) { 3534 error_report("First cluster in emptied image is in use"); 3535 abort(); 3536 } 3537 3538 /* Now finally the in-memory information corresponds to the on-disk 3539 * structures and is correct */ 3540 ret = qcow2_mark_clean(bs); 3541 if (ret < 0) { 3542 goto fail; 3543 } 3544 3545 ret = bdrv_truncate(bs->file, (3 + l1_clusters) * s->cluster_size, 3546 PREALLOC_MODE_OFF, &local_err); 3547 if (ret < 0) { 3548 error_report_err(local_err); 3549 goto fail; 3550 } 3551 3552 return 0; 3553 3554 fail_broken_refcounts: 3555 /* The BDS is unusable at this point. If we wanted to make it usable, we 3556 * would have to call qcow2_refcount_close(), qcow2_refcount_init(), 3557 * qcow2_check_refcounts(), qcow2_refcount_close() and qcow2_refcount_init() 3558 * again. However, because the functions which could have caused this error 3559 * path to be taken are used by those functions as well, it's very likely 3560 * that that sequence will fail as well. Therefore, just eject the BDS. */ 3561 bs->drv = NULL; 3562 3563 fail: 3564 g_free(new_reftable); 3565 return ret; 3566 } 3567 3568 static int qcow2_make_empty(BlockDriverState *bs) 3569 { 3570 BDRVQcow2State *s = bs->opaque; 3571 uint64_t offset, end_offset; 3572 int step = QEMU_ALIGN_DOWN(INT_MAX, s->cluster_size); 3573 int l1_clusters, ret = 0; 3574 3575 l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / sizeof(uint64_t)); 3576 3577 if (s->qcow_version >= 3 && !s->snapshots && !s->nb_bitmaps && 3578 3 + l1_clusters <= s->refcount_block_size && 3579 s->crypt_method_header != QCOW_CRYPT_LUKS) { 3580 /* The following function only works for qcow2 v3 images (it 3581 * requires the dirty flag) and only as long as there are no 3582 * features that reserve extra clusters (such as snapshots, 3583 * LUKS header, or persistent bitmaps), because it completely 3584 * empties the image. Furthermore, the L1 table and three 3585 * additional clusters (image header, refcount table, one 3586 * refcount block) have to fit inside one refcount block. */ 3587 return make_completely_empty(bs); 3588 } 3589 3590 /* This fallback code simply discards every active cluster; this is slow, 3591 * but works in all cases */ 3592 end_offset = bs->total_sectors * BDRV_SECTOR_SIZE; 3593 for (offset = 0; offset < end_offset; offset += step) { 3594 /* As this function is generally used after committing an external 3595 * snapshot, QCOW2_DISCARD_SNAPSHOT seems appropriate. Also, the 3596 * default action for this kind of discard is to pass the discard, 3597 * which will ideally result in an actually smaller image file, as 3598 * is probably desired. */ 3599 ret = qcow2_cluster_discard(bs, offset, MIN(step, end_offset - offset), 3600 QCOW2_DISCARD_SNAPSHOT, true); 3601 if (ret < 0) { 3602 break; 3603 } 3604 } 3605 3606 return ret; 3607 } 3608 3609 static coroutine_fn int qcow2_co_flush_to_os(BlockDriverState *bs) 3610 { 3611 BDRVQcow2State *s = bs->opaque; 3612 int ret; 3613 3614 qemu_co_mutex_lock(&s->lock); 3615 ret = qcow2_cache_write(bs, s->l2_table_cache); 3616 if (ret < 0) { 3617 qemu_co_mutex_unlock(&s->lock); 3618 return ret; 3619 } 3620 3621 if (qcow2_need_accurate_refcounts(s)) { 3622 ret = qcow2_cache_write(bs, s->refcount_block_cache); 3623 if (ret < 0) { 3624 qemu_co_mutex_unlock(&s->lock); 3625 return ret; 3626 } 3627 } 3628 qemu_co_mutex_unlock(&s->lock); 3629 3630 return 0; 3631 } 3632 3633 static BlockMeasureInfo *qcow2_measure(QemuOpts *opts, BlockDriverState *in_bs, 3634 Error **errp) 3635 { 3636 Error *local_err = NULL; 3637 BlockMeasureInfo *info; 3638 uint64_t required = 0; /* bytes that contribute to required size */ 3639 uint64_t virtual_size; /* disk size as seen by guest */ 3640 uint64_t refcount_bits; 3641 uint64_t l2_tables; 3642 size_t cluster_size; 3643 int version; 3644 char *optstr; 3645 PreallocMode prealloc; 3646 bool has_backing_file; 3647 3648 /* Parse image creation options */ 3649 cluster_size = qcow2_opt_get_cluster_size_del(opts, &local_err); 3650 if (local_err) { 3651 goto err; 3652 } 3653 3654 version = qcow2_opt_get_version_del(opts, &local_err); 3655 if (local_err) { 3656 goto err; 3657 } 3658 3659 refcount_bits = qcow2_opt_get_refcount_bits_del(opts, version, &local_err); 3660 if (local_err) { 3661 goto err; 3662 } 3663 3664 optstr = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC); 3665 prealloc = qapi_enum_parse(&PreallocMode_lookup, optstr, 3666 PREALLOC_MODE_OFF, &local_err); 3667 g_free(optstr); 3668 if (local_err) { 3669 goto err; 3670 } 3671 3672 optstr = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE); 3673 has_backing_file = !!optstr; 3674 g_free(optstr); 3675 3676 virtual_size = align_offset(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0), 3677 cluster_size); 3678 3679 /* Check that virtual disk size is valid */ 3680 l2_tables = DIV_ROUND_UP(virtual_size / cluster_size, 3681 cluster_size / sizeof(uint64_t)); 3682 if (l2_tables * sizeof(uint64_t) > QCOW_MAX_L1_SIZE) { 3683 error_setg(&local_err, "The image size is too large " 3684 "(try using a larger cluster size)"); 3685 goto err; 3686 } 3687 3688 /* Account for input image */ 3689 if (in_bs) { 3690 int64_t ssize = bdrv_getlength(in_bs); 3691 if (ssize < 0) { 3692 error_setg_errno(&local_err, -ssize, 3693 "Unable to get image virtual_size"); 3694 goto err; 3695 } 3696 3697 virtual_size = align_offset(ssize, cluster_size); 3698 3699 if (has_backing_file) { 3700 /* We don't how much of the backing chain is shared by the input 3701 * image and the new image file. In the worst case the new image's 3702 * backing file has nothing in common with the input image. Be 3703 * conservative and assume all clusters need to be written. 3704 */ 3705 required = virtual_size; 3706 } else { 3707 int64_t offset; 3708 int64_t pnum = 0; 3709 3710 for (offset = 0; offset < ssize; offset += pnum) { 3711 int ret; 3712 3713 ret = bdrv_block_status_above(in_bs, NULL, offset, 3714 ssize - offset, &pnum, NULL, 3715 NULL); 3716 if (ret < 0) { 3717 error_setg_errno(&local_err, -ret, 3718 "Unable to get block status"); 3719 goto err; 3720 } 3721 3722 if (ret & BDRV_BLOCK_ZERO) { 3723 /* Skip zero regions (safe with no backing file) */ 3724 } else if ((ret & (BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED)) == 3725 (BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED)) { 3726 /* Extend pnum to end of cluster for next iteration */ 3727 pnum = ROUND_UP(offset + pnum, cluster_size) - offset; 3728 3729 /* Count clusters we've seen */ 3730 required += offset % cluster_size + pnum; 3731 } 3732 } 3733 } 3734 } 3735 3736 /* Take into account preallocation. Nothing special is needed for 3737 * PREALLOC_MODE_METADATA since metadata is always counted. 3738 */ 3739 if (prealloc == PREALLOC_MODE_FULL || prealloc == PREALLOC_MODE_FALLOC) { 3740 required = virtual_size; 3741 } 3742 3743 info = g_new(BlockMeasureInfo, 1); 3744 info->fully_allocated = 3745 qcow2_calc_prealloc_size(virtual_size, cluster_size, 3746 ctz32(refcount_bits)); 3747 3748 /* Remove data clusters that are not required. This overestimates the 3749 * required size because metadata needed for the fully allocated file is 3750 * still counted. 3751 */ 3752 info->required = info->fully_allocated - virtual_size + required; 3753 return info; 3754 3755 err: 3756 error_propagate(errp, local_err); 3757 return NULL; 3758 } 3759 3760 static int qcow2_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) 3761 { 3762 BDRVQcow2State *s = bs->opaque; 3763 bdi->unallocated_blocks_are_zero = true; 3764 bdi->can_write_zeroes_with_unmap = (s->qcow_version >= 3); 3765 bdi->cluster_size = s->cluster_size; 3766 bdi->vm_state_offset = qcow2_vm_state_offset(s); 3767 return 0; 3768 } 3769 3770 static ImageInfoSpecific *qcow2_get_specific_info(BlockDriverState *bs) 3771 { 3772 BDRVQcow2State *s = bs->opaque; 3773 ImageInfoSpecific *spec_info; 3774 QCryptoBlockInfo *encrypt_info = NULL; 3775 3776 if (s->crypto != NULL) { 3777 encrypt_info = qcrypto_block_get_info(s->crypto, &error_abort); 3778 } 3779 3780 spec_info = g_new(ImageInfoSpecific, 1); 3781 *spec_info = (ImageInfoSpecific){ 3782 .type = IMAGE_INFO_SPECIFIC_KIND_QCOW2, 3783 .u.qcow2.data = g_new(ImageInfoSpecificQCow2, 1), 3784 }; 3785 if (s->qcow_version == 2) { 3786 *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){ 3787 .compat = g_strdup("0.10"), 3788 .refcount_bits = s->refcount_bits, 3789 }; 3790 } else if (s->qcow_version == 3) { 3791 *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){ 3792 .compat = g_strdup("1.1"), 3793 .lazy_refcounts = s->compatible_features & 3794 QCOW2_COMPAT_LAZY_REFCOUNTS, 3795 .has_lazy_refcounts = true, 3796 .corrupt = s->incompatible_features & 3797 QCOW2_INCOMPAT_CORRUPT, 3798 .has_corrupt = true, 3799 .refcount_bits = s->refcount_bits, 3800 }; 3801 } else { 3802 /* if this assertion fails, this probably means a new version was 3803 * added without having it covered here */ 3804 assert(false); 3805 } 3806 3807 if (encrypt_info) { 3808 ImageInfoSpecificQCow2Encryption *qencrypt = 3809 g_new(ImageInfoSpecificQCow2Encryption, 1); 3810 switch (encrypt_info->format) { 3811 case Q_CRYPTO_BLOCK_FORMAT_QCOW: 3812 qencrypt->format = BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_AES; 3813 qencrypt->u.aes = encrypt_info->u.qcow; 3814 break; 3815 case Q_CRYPTO_BLOCK_FORMAT_LUKS: 3816 qencrypt->format = BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_LUKS; 3817 qencrypt->u.luks = encrypt_info->u.luks; 3818 break; 3819 default: 3820 abort(); 3821 } 3822 /* Since we did shallow copy above, erase any pointers 3823 * in the original info */ 3824 memset(&encrypt_info->u, 0, sizeof(encrypt_info->u)); 3825 qapi_free_QCryptoBlockInfo(encrypt_info); 3826 3827 spec_info->u.qcow2.data->has_encrypt = true; 3828 spec_info->u.qcow2.data->encrypt = qencrypt; 3829 } 3830 3831 return spec_info; 3832 } 3833 3834 static int qcow2_save_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, 3835 int64_t pos) 3836 { 3837 BDRVQcow2State *s = bs->opaque; 3838 3839 BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_SAVE); 3840 return bs->drv->bdrv_co_pwritev(bs, qcow2_vm_state_offset(s) + pos, 3841 qiov->size, qiov, 0); 3842 } 3843 3844 static int qcow2_load_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, 3845 int64_t pos) 3846 { 3847 BDRVQcow2State *s = bs->opaque; 3848 3849 BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_LOAD); 3850 return bs->drv->bdrv_co_preadv(bs, qcow2_vm_state_offset(s) + pos, 3851 qiov->size, qiov, 0); 3852 } 3853 3854 /* 3855 * Downgrades an image's version. To achieve this, any incompatible features 3856 * have to be removed. 3857 */ 3858 static int qcow2_downgrade(BlockDriverState *bs, int target_version, 3859 BlockDriverAmendStatusCB *status_cb, void *cb_opaque) 3860 { 3861 BDRVQcow2State *s = bs->opaque; 3862 int current_version = s->qcow_version; 3863 int ret; 3864 3865 if (target_version == current_version) { 3866 return 0; 3867 } else if (target_version > current_version) { 3868 return -EINVAL; 3869 } else if (target_version != 2) { 3870 return -EINVAL; 3871 } 3872 3873 if (s->refcount_order != 4) { 3874 error_report("compat=0.10 requires refcount_bits=16"); 3875 return -ENOTSUP; 3876 } 3877 3878 /* clear incompatible features */ 3879 if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) { 3880 ret = qcow2_mark_clean(bs); 3881 if (ret < 0) { 3882 return ret; 3883 } 3884 } 3885 3886 /* with QCOW2_INCOMPAT_CORRUPT, it is pretty much impossible to get here in 3887 * the first place; if that happens nonetheless, returning -ENOTSUP is the 3888 * best thing to do anyway */ 3889 3890 if (s->incompatible_features) { 3891 return -ENOTSUP; 3892 } 3893 3894 /* since we can ignore compatible features, we can set them to 0 as well */ 3895 s->compatible_features = 0; 3896 /* if lazy refcounts have been used, they have already been fixed through 3897 * clearing the dirty flag */ 3898 3899 /* clearing autoclear features is trivial */ 3900 s->autoclear_features = 0; 3901 3902 ret = qcow2_expand_zero_clusters(bs, status_cb, cb_opaque); 3903 if (ret < 0) { 3904 return ret; 3905 } 3906 3907 s->qcow_version = target_version; 3908 ret = qcow2_update_header(bs); 3909 if (ret < 0) { 3910 s->qcow_version = current_version; 3911 return ret; 3912 } 3913 return 0; 3914 } 3915 3916 typedef enum Qcow2AmendOperation { 3917 /* This is the value Qcow2AmendHelperCBInfo::last_operation will be 3918 * statically initialized to so that the helper CB can discern the first 3919 * invocation from an operation change */ 3920 QCOW2_NO_OPERATION = 0, 3921 3922 QCOW2_CHANGING_REFCOUNT_ORDER, 3923 QCOW2_DOWNGRADING, 3924 } Qcow2AmendOperation; 3925 3926 typedef struct Qcow2AmendHelperCBInfo { 3927 /* The code coordinating the amend operations should only modify 3928 * these four fields; the rest will be managed by the CB */ 3929 BlockDriverAmendStatusCB *original_status_cb; 3930 void *original_cb_opaque; 3931 3932 Qcow2AmendOperation current_operation; 3933 3934 /* Total number of operations to perform (only set once) */ 3935 int total_operations; 3936 3937 /* The following fields are managed by the CB */ 3938 3939 /* Number of operations completed */ 3940 int operations_completed; 3941 3942 /* Cumulative offset of all completed operations */ 3943 int64_t offset_completed; 3944 3945 Qcow2AmendOperation last_operation; 3946 int64_t last_work_size; 3947 } Qcow2AmendHelperCBInfo; 3948 3949 static void qcow2_amend_helper_cb(BlockDriverState *bs, 3950 int64_t operation_offset, 3951 int64_t operation_work_size, void *opaque) 3952 { 3953 Qcow2AmendHelperCBInfo *info = opaque; 3954 int64_t current_work_size; 3955 int64_t projected_work_size; 3956 3957 if (info->current_operation != info->last_operation) { 3958 if (info->last_operation != QCOW2_NO_OPERATION) { 3959 info->offset_completed += info->last_work_size; 3960 info->operations_completed++; 3961 } 3962 3963 info->last_operation = info->current_operation; 3964 } 3965 3966 assert(info->total_operations > 0); 3967 assert(info->operations_completed < info->total_operations); 3968 3969 info->last_work_size = operation_work_size; 3970 3971 current_work_size = info->offset_completed + operation_work_size; 3972 3973 /* current_work_size is the total work size for (operations_completed + 1) 3974 * operations (which includes this one), so multiply it by the number of 3975 * operations not covered and divide it by the number of operations 3976 * covered to get a projection for the operations not covered */ 3977 projected_work_size = current_work_size * (info->total_operations - 3978 info->operations_completed - 1) 3979 / (info->operations_completed + 1); 3980 3981 info->original_status_cb(bs, info->offset_completed + operation_offset, 3982 current_work_size + projected_work_size, 3983 info->original_cb_opaque); 3984 } 3985 3986 static int qcow2_amend_options(BlockDriverState *bs, QemuOpts *opts, 3987 BlockDriverAmendStatusCB *status_cb, 3988 void *cb_opaque) 3989 { 3990 BDRVQcow2State *s = bs->opaque; 3991 int old_version = s->qcow_version, new_version = old_version; 3992 uint64_t new_size = 0; 3993 const char *backing_file = NULL, *backing_format = NULL; 3994 bool lazy_refcounts = s->use_lazy_refcounts; 3995 const char *compat = NULL; 3996 uint64_t cluster_size = s->cluster_size; 3997 bool encrypt; 3998 int encformat; 3999 int refcount_bits = s->refcount_bits; 4000 Error *local_err = NULL; 4001 int ret; 4002 QemuOptDesc *desc = opts->list->desc; 4003 Qcow2AmendHelperCBInfo helper_cb_info; 4004 4005 while (desc && desc->name) { 4006 if (!qemu_opt_find(opts, desc->name)) { 4007 /* only change explicitly defined options */ 4008 desc++; 4009 continue; 4010 } 4011 4012 if (!strcmp(desc->name, BLOCK_OPT_COMPAT_LEVEL)) { 4013 compat = qemu_opt_get(opts, BLOCK_OPT_COMPAT_LEVEL); 4014 if (!compat) { 4015 /* preserve default */ 4016 } else if (!strcmp(compat, "0.10")) { 4017 new_version = 2; 4018 } else if (!strcmp(compat, "1.1")) { 4019 new_version = 3; 4020 } else { 4021 error_report("Unknown compatibility level %s", compat); 4022 return -EINVAL; 4023 } 4024 } else if (!strcmp(desc->name, BLOCK_OPT_PREALLOC)) { 4025 error_report("Cannot change preallocation mode"); 4026 return -ENOTSUP; 4027 } else if (!strcmp(desc->name, BLOCK_OPT_SIZE)) { 4028 new_size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, 0); 4029 } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FILE)) { 4030 backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE); 4031 } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FMT)) { 4032 backing_format = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT); 4033 } else if (!strcmp(desc->name, BLOCK_OPT_ENCRYPT)) { 4034 encrypt = qemu_opt_get_bool(opts, BLOCK_OPT_ENCRYPT, 4035 !!s->crypto); 4036 4037 if (encrypt != !!s->crypto) { 4038 error_report("Changing the encryption flag is not supported"); 4039 return -ENOTSUP; 4040 } 4041 } else if (!strcmp(desc->name, BLOCK_OPT_ENCRYPT_FORMAT)) { 4042 encformat = qcow2_crypt_method_from_format( 4043 qemu_opt_get(opts, BLOCK_OPT_ENCRYPT_FORMAT)); 4044 4045 if (encformat != s->crypt_method_header) { 4046 error_report("Changing the encryption format is not supported"); 4047 return -ENOTSUP; 4048 } 4049 } else if (g_str_has_prefix(desc->name, "encrypt.")) { 4050 error_report("Changing the encryption parameters is not supported"); 4051 return -ENOTSUP; 4052 } else if (!strcmp(desc->name, BLOCK_OPT_CLUSTER_SIZE)) { 4053 cluster_size = qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE, 4054 cluster_size); 4055 if (cluster_size != s->cluster_size) { 4056 error_report("Changing the cluster size is not supported"); 4057 return -ENOTSUP; 4058 } 4059 } else if (!strcmp(desc->name, BLOCK_OPT_LAZY_REFCOUNTS)) { 4060 lazy_refcounts = qemu_opt_get_bool(opts, BLOCK_OPT_LAZY_REFCOUNTS, 4061 lazy_refcounts); 4062 } else if (!strcmp(desc->name, BLOCK_OPT_REFCOUNT_BITS)) { 4063 refcount_bits = qemu_opt_get_number(opts, BLOCK_OPT_REFCOUNT_BITS, 4064 refcount_bits); 4065 4066 if (refcount_bits <= 0 || refcount_bits > 64 || 4067 !is_power_of_2(refcount_bits)) 4068 { 4069 error_report("Refcount width must be a power of two and may " 4070 "not exceed 64 bits"); 4071 return -EINVAL; 4072 } 4073 } else { 4074 /* if this point is reached, this probably means a new option was 4075 * added without having it covered here */ 4076 abort(); 4077 } 4078 4079 desc++; 4080 } 4081 4082 helper_cb_info = (Qcow2AmendHelperCBInfo){ 4083 .original_status_cb = status_cb, 4084 .original_cb_opaque = cb_opaque, 4085 .total_operations = (new_version < old_version) 4086 + (s->refcount_bits != refcount_bits) 4087 }; 4088 4089 /* Upgrade first (some features may require compat=1.1) */ 4090 if (new_version > old_version) { 4091 s->qcow_version = new_version; 4092 ret = qcow2_update_header(bs); 4093 if (ret < 0) { 4094 s->qcow_version = old_version; 4095 return ret; 4096 } 4097 } 4098 4099 if (s->refcount_bits != refcount_bits) { 4100 int refcount_order = ctz32(refcount_bits); 4101 4102 if (new_version < 3 && refcount_bits != 16) { 4103 error_report("Different refcount widths than 16 bits require " 4104 "compatibility level 1.1 or above (use compat=1.1 or " 4105 "greater)"); 4106 return -EINVAL; 4107 } 4108 4109 helper_cb_info.current_operation = QCOW2_CHANGING_REFCOUNT_ORDER; 4110 ret = qcow2_change_refcount_order(bs, refcount_order, 4111 &qcow2_amend_helper_cb, 4112 &helper_cb_info, &local_err); 4113 if (ret < 0) { 4114 error_report_err(local_err); 4115 return ret; 4116 } 4117 } 4118 4119 if (backing_file || backing_format) { 4120 ret = qcow2_change_backing_file(bs, 4121 backing_file ?: s->image_backing_file, 4122 backing_format ?: s->image_backing_format); 4123 if (ret < 0) { 4124 return ret; 4125 } 4126 } 4127 4128 if (s->use_lazy_refcounts != lazy_refcounts) { 4129 if (lazy_refcounts) { 4130 if (new_version < 3) { 4131 error_report("Lazy refcounts only supported with compatibility " 4132 "level 1.1 and above (use compat=1.1 or greater)"); 4133 return -EINVAL; 4134 } 4135 s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS; 4136 ret = qcow2_update_header(bs); 4137 if (ret < 0) { 4138 s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS; 4139 return ret; 4140 } 4141 s->use_lazy_refcounts = true; 4142 } else { 4143 /* make image clean first */ 4144 ret = qcow2_mark_clean(bs); 4145 if (ret < 0) { 4146 return ret; 4147 } 4148 /* now disallow lazy refcounts */ 4149 s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS; 4150 ret = qcow2_update_header(bs); 4151 if (ret < 0) { 4152 s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS; 4153 return ret; 4154 } 4155 s->use_lazy_refcounts = false; 4156 } 4157 } 4158 4159 if (new_size) { 4160 BlockBackend *blk = blk_new(BLK_PERM_RESIZE, BLK_PERM_ALL); 4161 ret = blk_insert_bs(blk, bs, &local_err); 4162 if (ret < 0) { 4163 error_report_err(local_err); 4164 blk_unref(blk); 4165 return ret; 4166 } 4167 4168 ret = blk_truncate(blk, new_size, PREALLOC_MODE_OFF, &local_err); 4169 blk_unref(blk); 4170 if (ret < 0) { 4171 error_report_err(local_err); 4172 return ret; 4173 } 4174 } 4175 4176 /* Downgrade last (so unsupported features can be removed before) */ 4177 if (new_version < old_version) { 4178 helper_cb_info.current_operation = QCOW2_DOWNGRADING; 4179 ret = qcow2_downgrade(bs, new_version, &qcow2_amend_helper_cb, 4180 &helper_cb_info); 4181 if (ret < 0) { 4182 return ret; 4183 } 4184 } 4185 4186 return 0; 4187 } 4188 4189 /* 4190 * If offset or size are negative, respectively, they will not be included in 4191 * the BLOCK_IMAGE_CORRUPTED event emitted. 4192 * fatal will be ignored for read-only BDS; corruptions found there will always 4193 * be considered non-fatal. 4194 */ 4195 void qcow2_signal_corruption(BlockDriverState *bs, bool fatal, int64_t offset, 4196 int64_t size, const char *message_format, ...) 4197 { 4198 BDRVQcow2State *s = bs->opaque; 4199 const char *node_name; 4200 char *message; 4201 va_list ap; 4202 4203 fatal = fatal && !bs->read_only; 4204 4205 if (s->signaled_corruption && 4206 (!fatal || (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT))) 4207 { 4208 return; 4209 } 4210 4211 va_start(ap, message_format); 4212 message = g_strdup_vprintf(message_format, ap); 4213 va_end(ap); 4214 4215 if (fatal) { 4216 fprintf(stderr, "qcow2: Marking image as corrupt: %s; further " 4217 "corruption events will be suppressed\n", message); 4218 } else { 4219 fprintf(stderr, "qcow2: Image is corrupt: %s; further non-fatal " 4220 "corruption events will be suppressed\n", message); 4221 } 4222 4223 node_name = bdrv_get_node_name(bs); 4224 qapi_event_send_block_image_corrupted(bdrv_get_device_name(bs), 4225 *node_name != '\0', node_name, 4226 message, offset >= 0, offset, 4227 size >= 0, size, 4228 fatal, &error_abort); 4229 g_free(message); 4230 4231 if (fatal) { 4232 qcow2_mark_corrupt(bs); 4233 bs->drv = NULL; /* make BDS unusable */ 4234 } 4235 4236 s->signaled_corruption = true; 4237 } 4238 4239 static QemuOptsList qcow2_create_opts = { 4240 .name = "qcow2-create-opts", 4241 .head = QTAILQ_HEAD_INITIALIZER(qcow2_create_opts.head), 4242 .desc = { 4243 { 4244 .name = BLOCK_OPT_SIZE, 4245 .type = QEMU_OPT_SIZE, 4246 .help = "Virtual disk size" 4247 }, 4248 { 4249 .name = BLOCK_OPT_COMPAT_LEVEL, 4250 .type = QEMU_OPT_STRING, 4251 .help = "Compatibility level (0.10 or 1.1)" 4252 }, 4253 { 4254 .name = BLOCK_OPT_BACKING_FILE, 4255 .type = QEMU_OPT_STRING, 4256 .help = "File name of a base image" 4257 }, 4258 { 4259 .name = BLOCK_OPT_BACKING_FMT, 4260 .type = QEMU_OPT_STRING, 4261 .help = "Image format of the base image" 4262 }, 4263 { 4264 .name = BLOCK_OPT_ENCRYPT, 4265 .type = QEMU_OPT_BOOL, 4266 .help = "Encrypt the image with format 'aes'. (Deprecated " 4267 "in favor of " BLOCK_OPT_ENCRYPT_FORMAT "=aes)", 4268 }, 4269 { 4270 .name = BLOCK_OPT_ENCRYPT_FORMAT, 4271 .type = QEMU_OPT_STRING, 4272 .help = "Encrypt the image, format choices: 'aes', 'luks'", 4273 }, 4274 BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.", 4275 "ID of secret providing qcow AES key or LUKS passphrase"), 4276 BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_ALG("encrypt."), 4277 BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_MODE("encrypt."), 4278 BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_ALG("encrypt."), 4279 BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_HASH_ALG("encrypt."), 4280 BLOCK_CRYPTO_OPT_DEF_LUKS_HASH_ALG("encrypt."), 4281 BLOCK_CRYPTO_OPT_DEF_LUKS_ITER_TIME("encrypt."), 4282 { 4283 .name = BLOCK_OPT_CLUSTER_SIZE, 4284 .type = QEMU_OPT_SIZE, 4285 .help = "qcow2 cluster size", 4286 .def_value_str = stringify(DEFAULT_CLUSTER_SIZE) 4287 }, 4288 { 4289 .name = BLOCK_OPT_PREALLOC, 4290 .type = QEMU_OPT_STRING, 4291 .help = "Preallocation mode (allowed values: off, metadata, " 4292 "falloc, full)" 4293 }, 4294 { 4295 .name = BLOCK_OPT_LAZY_REFCOUNTS, 4296 .type = QEMU_OPT_BOOL, 4297 .help = "Postpone refcount updates", 4298 .def_value_str = "off" 4299 }, 4300 { 4301 .name = BLOCK_OPT_REFCOUNT_BITS, 4302 .type = QEMU_OPT_NUMBER, 4303 .help = "Width of a reference count entry in bits", 4304 .def_value_str = "16" 4305 }, 4306 { /* end of list */ } 4307 } 4308 }; 4309 4310 BlockDriver bdrv_qcow2 = { 4311 .format_name = "qcow2", 4312 .instance_size = sizeof(BDRVQcow2State), 4313 .bdrv_probe = qcow2_probe, 4314 .bdrv_open = qcow2_open, 4315 .bdrv_close = qcow2_close, 4316 .bdrv_reopen_prepare = qcow2_reopen_prepare, 4317 .bdrv_reopen_commit = qcow2_reopen_commit, 4318 .bdrv_reopen_abort = qcow2_reopen_abort, 4319 .bdrv_join_options = qcow2_join_options, 4320 .bdrv_child_perm = bdrv_format_default_perms, 4321 .bdrv_create = qcow2_create, 4322 .bdrv_has_zero_init = bdrv_has_zero_init_1, 4323 .bdrv_co_get_block_status = qcow2_co_get_block_status, 4324 4325 .bdrv_co_preadv = qcow2_co_preadv, 4326 .bdrv_co_pwritev = qcow2_co_pwritev, 4327 .bdrv_co_flush_to_os = qcow2_co_flush_to_os, 4328 4329 .bdrv_co_pwrite_zeroes = qcow2_co_pwrite_zeroes, 4330 .bdrv_co_pdiscard = qcow2_co_pdiscard, 4331 .bdrv_truncate = qcow2_truncate, 4332 .bdrv_co_pwritev_compressed = qcow2_co_pwritev_compressed, 4333 .bdrv_make_empty = qcow2_make_empty, 4334 4335 .bdrv_snapshot_create = qcow2_snapshot_create, 4336 .bdrv_snapshot_goto = qcow2_snapshot_goto, 4337 .bdrv_snapshot_delete = qcow2_snapshot_delete, 4338 .bdrv_snapshot_list = qcow2_snapshot_list, 4339 .bdrv_snapshot_load_tmp = qcow2_snapshot_load_tmp, 4340 .bdrv_measure = qcow2_measure, 4341 .bdrv_get_info = qcow2_get_info, 4342 .bdrv_get_specific_info = qcow2_get_specific_info, 4343 4344 .bdrv_save_vmstate = qcow2_save_vmstate, 4345 .bdrv_load_vmstate = qcow2_load_vmstate, 4346 4347 .supports_backing = true, 4348 .bdrv_change_backing_file = qcow2_change_backing_file, 4349 4350 .bdrv_refresh_limits = qcow2_refresh_limits, 4351 .bdrv_invalidate_cache = qcow2_invalidate_cache, 4352 .bdrv_inactivate = qcow2_inactivate, 4353 4354 .create_opts = &qcow2_create_opts, 4355 .bdrv_check = qcow2_check, 4356 .bdrv_amend_options = qcow2_amend_options, 4357 4358 .bdrv_detach_aio_context = qcow2_detach_aio_context, 4359 .bdrv_attach_aio_context = qcow2_attach_aio_context, 4360 4361 .bdrv_reopen_bitmaps_rw = qcow2_reopen_bitmaps_rw, 4362 .bdrv_can_store_new_dirty_bitmap = qcow2_can_store_new_dirty_bitmap, 4363 .bdrv_remove_persistent_dirty_bitmap = qcow2_remove_persistent_dirty_bitmap, 4364 }; 4365 4366 static void bdrv_qcow2_init(void) 4367 { 4368 bdrv_register(&bdrv_qcow2); 4369 } 4370 4371 block_init(bdrv_qcow2_init); 4372