1 /* 2 * Block driver for the QCOW version 2 format 3 * 4 * Copyright (c) 2004-2006 Fabrice Bellard 5 * 6 * Permission is hereby granted, free of charge, to any person obtaining a copy 7 * of this software and associated documentation files (the "Software"), to deal 8 * in the Software without restriction, including without limitation the rights 9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 10 * copies of the Software, and to permit persons to whom the Software is 11 * furnished to do so, subject to the following conditions: 12 * 13 * The above copyright notice and this permission notice shall be included in 14 * all copies or substantial portions of the Software. 15 * 16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN 22 * THE SOFTWARE. 23 */ 24 25 #include "qemu/osdep.h" 26 27 #include "block/qdict.h" 28 #include "sysemu/block-backend.h" 29 #include "qemu/main-loop.h" 30 #include "qemu/module.h" 31 #include "qcow2.h" 32 #include "qemu/error-report.h" 33 #include "qapi/error.h" 34 #include "qapi/qapi-events-block-core.h" 35 #include "qapi/qmp/qdict.h" 36 #include "qapi/qmp/qstring.h" 37 #include "trace.h" 38 #include "qemu/option_int.h" 39 #include "qemu/cutils.h" 40 #include "qemu/bswap.h" 41 #include "qemu/memalign.h" 42 #include "qapi/qobject-input-visitor.h" 43 #include "qapi/qapi-visit-block-core.h" 44 #include "crypto.h" 45 #include "block/aio_task.h" 46 47 /* 48 Differences with QCOW: 49 50 - Support for multiple incremental snapshots. 51 - Memory management by reference counts. 52 - Clusters which have a reference count of one have the bit 53 QCOW_OFLAG_COPIED to optimize write performance. 54 - Size of compressed clusters is stored in sectors to reduce bit usage 55 in the cluster offsets. 56 - Support for storing additional data (such as the VM state) in the 57 snapshots. 58 - If a backing store is used, the cluster size is not constrained 59 (could be backported to QCOW). 60 - L2 tables have always a size of one cluster. 61 */ 62 63 64 typedef struct { 65 uint32_t magic; 66 uint32_t len; 67 } QEMU_PACKED QCowExtension; 68 69 #define QCOW2_EXT_MAGIC_END 0 70 #define QCOW2_EXT_MAGIC_BACKING_FORMAT 0xe2792aca 71 #define QCOW2_EXT_MAGIC_FEATURE_TABLE 0x6803f857 72 #define QCOW2_EXT_MAGIC_CRYPTO_HEADER 0x0537be77 73 #define QCOW2_EXT_MAGIC_BITMAPS 0x23852875 74 #define QCOW2_EXT_MAGIC_DATA_FILE 0x44415441 75 76 static int coroutine_fn 77 qcow2_co_preadv_compressed(BlockDriverState *bs, 78 uint64_t l2_entry, 79 uint64_t offset, 80 uint64_t bytes, 81 QEMUIOVector *qiov, 82 size_t qiov_offset); 83 84 static int qcow2_probe(const uint8_t *buf, int buf_size, const char *filename) 85 { 86 const QCowHeader *cow_header = (const void *)buf; 87 88 if (buf_size >= sizeof(QCowHeader) && 89 be32_to_cpu(cow_header->magic) == QCOW_MAGIC && 90 be32_to_cpu(cow_header->version) >= 2) 91 return 100; 92 else 93 return 0; 94 } 95 96 97 static int qcow2_crypto_hdr_read_func(QCryptoBlock *block, size_t offset, 98 uint8_t *buf, size_t buflen, 99 void *opaque, Error **errp) 100 { 101 BlockDriverState *bs = opaque; 102 BDRVQcow2State *s = bs->opaque; 103 ssize_t ret; 104 105 if ((offset + buflen) > s->crypto_header.length) { 106 error_setg(errp, "Request for data outside of extension header"); 107 return -1; 108 } 109 110 ret = bdrv_pread(bs->file, s->crypto_header.offset + offset, buflen, buf, 111 0); 112 if (ret < 0) { 113 error_setg_errno(errp, -ret, "Could not read encryption header"); 114 return -1; 115 } 116 return 0; 117 } 118 119 120 static int qcow2_crypto_hdr_init_func(QCryptoBlock *block, size_t headerlen, 121 void *opaque, Error **errp) 122 { 123 BlockDriverState *bs = opaque; 124 BDRVQcow2State *s = bs->opaque; 125 int64_t ret; 126 int64_t clusterlen; 127 128 ret = qcow2_alloc_clusters(bs, headerlen); 129 if (ret < 0) { 130 error_setg_errno(errp, -ret, 131 "Cannot allocate cluster for LUKS header size %zu", 132 headerlen); 133 return -1; 134 } 135 136 s->crypto_header.length = headerlen; 137 s->crypto_header.offset = ret; 138 139 /* 140 * Zero fill all space in cluster so it has predictable 141 * content, as we may not initialize some regions of the 142 * header (eg only 1 out of 8 key slots will be initialized) 143 */ 144 clusterlen = size_to_clusters(s, headerlen) * s->cluster_size; 145 assert(qcow2_pre_write_overlap_check(bs, 0, ret, clusterlen, false) == 0); 146 ret = bdrv_pwrite_zeroes(bs->file, 147 ret, 148 clusterlen, 0); 149 if (ret < 0) { 150 error_setg_errno(errp, -ret, "Could not zero fill encryption header"); 151 return -1; 152 } 153 154 return 0; 155 } 156 157 158 static int qcow2_crypto_hdr_write_func(QCryptoBlock *block, size_t offset, 159 const uint8_t *buf, size_t buflen, 160 void *opaque, Error **errp) 161 { 162 BlockDriverState *bs = opaque; 163 BDRVQcow2State *s = bs->opaque; 164 ssize_t ret; 165 166 if ((offset + buflen) > s->crypto_header.length) { 167 error_setg(errp, "Request for data outside of extension header"); 168 return -1; 169 } 170 171 ret = bdrv_pwrite(bs->file, s->crypto_header.offset + offset, buflen, buf, 172 0); 173 if (ret < 0) { 174 error_setg_errno(errp, -ret, "Could not read encryption header"); 175 return -1; 176 } 177 return 0; 178 } 179 180 static QDict* 181 qcow2_extract_crypto_opts(QemuOpts *opts, const char *fmt, Error **errp) 182 { 183 QDict *cryptoopts_qdict; 184 QDict *opts_qdict; 185 186 /* Extract "encrypt." options into a qdict */ 187 opts_qdict = qemu_opts_to_qdict(opts, NULL); 188 qdict_extract_subqdict(opts_qdict, &cryptoopts_qdict, "encrypt."); 189 qobject_unref(opts_qdict); 190 qdict_put_str(cryptoopts_qdict, "format", fmt); 191 return cryptoopts_qdict; 192 } 193 194 /* 195 * read qcow2 extension and fill bs 196 * start reading from start_offset 197 * finish reading upon magic of value 0 or when end_offset reached 198 * unknown magic is skipped (future extension this version knows nothing about) 199 * return 0 upon success, non-0 otherwise 200 */ 201 static int qcow2_read_extensions(BlockDriverState *bs, uint64_t start_offset, 202 uint64_t end_offset, void **p_feature_table, 203 int flags, bool *need_update_header, 204 Error **errp) 205 { 206 BDRVQcow2State *s = bs->opaque; 207 QCowExtension ext; 208 uint64_t offset; 209 int ret; 210 Qcow2BitmapHeaderExt bitmaps_ext; 211 212 if (need_update_header != NULL) { 213 *need_update_header = false; 214 } 215 216 #ifdef DEBUG_EXT 217 printf("qcow2_read_extensions: start=%ld end=%ld\n", start_offset, end_offset); 218 #endif 219 offset = start_offset; 220 while (offset < end_offset) { 221 222 #ifdef DEBUG_EXT 223 /* Sanity check */ 224 if (offset > s->cluster_size) 225 printf("qcow2_read_extension: suspicious offset %lu\n", offset); 226 227 printf("attempting to read extended header in offset %lu\n", offset); 228 #endif 229 230 ret = bdrv_pread(bs->file, offset, sizeof(ext), &ext, 0); 231 if (ret < 0) { 232 error_setg_errno(errp, -ret, "qcow2_read_extension: ERROR: " 233 "pread fail from offset %" PRIu64, offset); 234 return 1; 235 } 236 ext.magic = be32_to_cpu(ext.magic); 237 ext.len = be32_to_cpu(ext.len); 238 offset += sizeof(ext); 239 #ifdef DEBUG_EXT 240 printf("ext.magic = 0x%x\n", ext.magic); 241 #endif 242 if (offset > end_offset || ext.len > end_offset - offset) { 243 error_setg(errp, "Header extension too large"); 244 return -EINVAL; 245 } 246 247 switch (ext.magic) { 248 case QCOW2_EXT_MAGIC_END: 249 return 0; 250 251 case QCOW2_EXT_MAGIC_BACKING_FORMAT: 252 if (ext.len >= sizeof(bs->backing_format)) { 253 error_setg(errp, "ERROR: ext_backing_format: len=%" PRIu32 254 " too large (>=%zu)", ext.len, 255 sizeof(bs->backing_format)); 256 return 2; 257 } 258 ret = bdrv_pread(bs->file, offset, ext.len, bs->backing_format, 0); 259 if (ret < 0) { 260 error_setg_errno(errp, -ret, "ERROR: ext_backing_format: " 261 "Could not read format name"); 262 return 3; 263 } 264 bs->backing_format[ext.len] = '\0'; 265 s->image_backing_format = g_strdup(bs->backing_format); 266 #ifdef DEBUG_EXT 267 printf("Qcow2: Got format extension %s\n", bs->backing_format); 268 #endif 269 break; 270 271 case QCOW2_EXT_MAGIC_FEATURE_TABLE: 272 if (p_feature_table != NULL) { 273 void *feature_table = g_malloc0(ext.len + 2 * sizeof(Qcow2Feature)); 274 ret = bdrv_pread(bs->file, offset, ext.len, feature_table, 0); 275 if (ret < 0) { 276 error_setg_errno(errp, -ret, "ERROR: ext_feature_table: " 277 "Could not read table"); 278 return ret; 279 } 280 281 *p_feature_table = feature_table; 282 } 283 break; 284 285 case QCOW2_EXT_MAGIC_CRYPTO_HEADER: { 286 unsigned int cflags = 0; 287 if (s->crypt_method_header != QCOW_CRYPT_LUKS) { 288 error_setg(errp, "CRYPTO header extension only " 289 "expected with LUKS encryption method"); 290 return -EINVAL; 291 } 292 if (ext.len != sizeof(Qcow2CryptoHeaderExtension)) { 293 error_setg(errp, "CRYPTO header extension size %u, " 294 "but expected size %zu", ext.len, 295 sizeof(Qcow2CryptoHeaderExtension)); 296 return -EINVAL; 297 } 298 299 ret = bdrv_pread(bs->file, offset, ext.len, &s->crypto_header, 0); 300 if (ret < 0) { 301 error_setg_errno(errp, -ret, 302 "Unable to read CRYPTO header extension"); 303 return ret; 304 } 305 s->crypto_header.offset = be64_to_cpu(s->crypto_header.offset); 306 s->crypto_header.length = be64_to_cpu(s->crypto_header.length); 307 308 if ((s->crypto_header.offset % s->cluster_size) != 0) { 309 error_setg(errp, "Encryption header offset '%" PRIu64 "' is " 310 "not a multiple of cluster size '%u'", 311 s->crypto_header.offset, s->cluster_size); 312 return -EINVAL; 313 } 314 315 if (flags & BDRV_O_NO_IO) { 316 cflags |= QCRYPTO_BLOCK_OPEN_NO_IO; 317 } 318 s->crypto = qcrypto_block_open(s->crypto_opts, "encrypt.", 319 qcow2_crypto_hdr_read_func, 320 bs, cflags, QCOW2_MAX_THREADS, errp); 321 if (!s->crypto) { 322 return -EINVAL; 323 } 324 } break; 325 326 case QCOW2_EXT_MAGIC_BITMAPS: 327 if (ext.len != sizeof(bitmaps_ext)) { 328 error_setg_errno(errp, -ret, "bitmaps_ext: " 329 "Invalid extension length"); 330 return -EINVAL; 331 } 332 333 if (!(s->autoclear_features & QCOW2_AUTOCLEAR_BITMAPS)) { 334 if (s->qcow_version < 3) { 335 /* Let's be a bit more specific */ 336 warn_report("This qcow2 v2 image contains bitmaps, but " 337 "they may have been modified by a program " 338 "without persistent bitmap support; so now " 339 "they must all be considered inconsistent"); 340 } else { 341 warn_report("a program lacking bitmap support " 342 "modified this file, so all bitmaps are now " 343 "considered inconsistent"); 344 } 345 error_printf("Some clusters may be leaked, " 346 "run 'qemu-img check -r' on the image " 347 "file to fix."); 348 if (need_update_header != NULL) { 349 /* Updating is needed to drop invalid bitmap extension. */ 350 *need_update_header = true; 351 } 352 break; 353 } 354 355 ret = bdrv_pread(bs->file, offset, ext.len, &bitmaps_ext, 0); 356 if (ret < 0) { 357 error_setg_errno(errp, -ret, "bitmaps_ext: " 358 "Could not read ext header"); 359 return ret; 360 } 361 362 if (bitmaps_ext.reserved32 != 0) { 363 error_setg_errno(errp, -ret, "bitmaps_ext: " 364 "Reserved field is not zero"); 365 return -EINVAL; 366 } 367 368 bitmaps_ext.nb_bitmaps = be32_to_cpu(bitmaps_ext.nb_bitmaps); 369 bitmaps_ext.bitmap_directory_size = 370 be64_to_cpu(bitmaps_ext.bitmap_directory_size); 371 bitmaps_ext.bitmap_directory_offset = 372 be64_to_cpu(bitmaps_ext.bitmap_directory_offset); 373 374 if (bitmaps_ext.nb_bitmaps > QCOW2_MAX_BITMAPS) { 375 error_setg(errp, 376 "bitmaps_ext: Image has %" PRIu32 " bitmaps, " 377 "exceeding the QEMU supported maximum of %d", 378 bitmaps_ext.nb_bitmaps, QCOW2_MAX_BITMAPS); 379 return -EINVAL; 380 } 381 382 if (bitmaps_ext.nb_bitmaps == 0) { 383 error_setg(errp, "found bitmaps extension with zero bitmaps"); 384 return -EINVAL; 385 } 386 387 if (offset_into_cluster(s, bitmaps_ext.bitmap_directory_offset)) { 388 error_setg(errp, "bitmaps_ext: " 389 "invalid bitmap directory offset"); 390 return -EINVAL; 391 } 392 393 if (bitmaps_ext.bitmap_directory_size > 394 QCOW2_MAX_BITMAP_DIRECTORY_SIZE) { 395 error_setg(errp, "bitmaps_ext: " 396 "bitmap directory size (%" PRIu64 ") exceeds " 397 "the maximum supported size (%d)", 398 bitmaps_ext.bitmap_directory_size, 399 QCOW2_MAX_BITMAP_DIRECTORY_SIZE); 400 return -EINVAL; 401 } 402 403 s->nb_bitmaps = bitmaps_ext.nb_bitmaps; 404 s->bitmap_directory_offset = 405 bitmaps_ext.bitmap_directory_offset; 406 s->bitmap_directory_size = 407 bitmaps_ext.bitmap_directory_size; 408 409 #ifdef DEBUG_EXT 410 printf("Qcow2: Got bitmaps extension: " 411 "offset=%" PRIu64 " nb_bitmaps=%" PRIu32 "\n", 412 s->bitmap_directory_offset, s->nb_bitmaps); 413 #endif 414 break; 415 416 case QCOW2_EXT_MAGIC_DATA_FILE: 417 { 418 s->image_data_file = g_malloc0(ext.len + 1); 419 ret = bdrv_pread(bs->file, offset, ext.len, s->image_data_file, 0); 420 if (ret < 0) { 421 error_setg_errno(errp, -ret, 422 "ERROR: Could not read data file name"); 423 return ret; 424 } 425 #ifdef DEBUG_EXT 426 printf("Qcow2: Got external data file %s\n", s->image_data_file); 427 #endif 428 break; 429 } 430 431 default: 432 /* unknown magic - save it in case we need to rewrite the header */ 433 /* If you add a new feature, make sure to also update the fast 434 * path of qcow2_make_empty() to deal with it. */ 435 { 436 Qcow2UnknownHeaderExtension *uext; 437 438 uext = g_malloc0(sizeof(*uext) + ext.len); 439 uext->magic = ext.magic; 440 uext->len = ext.len; 441 QLIST_INSERT_HEAD(&s->unknown_header_ext, uext, next); 442 443 ret = bdrv_pread(bs->file, offset, uext->len, uext->data, 0); 444 if (ret < 0) { 445 error_setg_errno(errp, -ret, "ERROR: unknown extension: " 446 "Could not read data"); 447 return ret; 448 } 449 } 450 break; 451 } 452 453 offset += ((ext.len + 7) & ~7); 454 } 455 456 return 0; 457 } 458 459 static void cleanup_unknown_header_ext(BlockDriverState *bs) 460 { 461 BDRVQcow2State *s = bs->opaque; 462 Qcow2UnknownHeaderExtension *uext, *next; 463 464 QLIST_FOREACH_SAFE(uext, &s->unknown_header_ext, next, next) { 465 QLIST_REMOVE(uext, next); 466 g_free(uext); 467 } 468 } 469 470 static void report_unsupported_feature(Error **errp, Qcow2Feature *table, 471 uint64_t mask) 472 { 473 g_autoptr(GString) features = g_string_sized_new(60); 474 475 while (table && table->name[0] != '\0') { 476 if (table->type == QCOW2_FEAT_TYPE_INCOMPATIBLE) { 477 if (mask & (1ULL << table->bit)) { 478 if (features->len > 0) { 479 g_string_append(features, ", "); 480 } 481 g_string_append_printf(features, "%.46s", table->name); 482 mask &= ~(1ULL << table->bit); 483 } 484 } 485 table++; 486 } 487 488 if (mask) { 489 if (features->len > 0) { 490 g_string_append(features, ", "); 491 } 492 g_string_append_printf(features, 493 "Unknown incompatible feature: %" PRIx64, mask); 494 } 495 496 error_setg(errp, "Unsupported qcow2 feature(s): %s", features->str); 497 } 498 499 /* 500 * Sets the dirty bit and flushes afterwards if necessary. 501 * 502 * The incompatible_features bit is only set if the image file header was 503 * updated successfully. Therefore it is not required to check the return 504 * value of this function. 505 */ 506 int qcow2_mark_dirty(BlockDriverState *bs) 507 { 508 BDRVQcow2State *s = bs->opaque; 509 uint64_t val; 510 int ret; 511 512 assert(s->qcow_version >= 3); 513 514 if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) { 515 return 0; /* already dirty */ 516 } 517 518 val = cpu_to_be64(s->incompatible_features | QCOW2_INCOMPAT_DIRTY); 519 ret = bdrv_pwrite_sync(bs->file, 520 offsetof(QCowHeader, incompatible_features), 521 sizeof(val), &val, 0); 522 if (ret < 0) { 523 return ret; 524 } 525 526 /* Only treat image as dirty if the header was updated successfully */ 527 s->incompatible_features |= QCOW2_INCOMPAT_DIRTY; 528 return 0; 529 } 530 531 /* 532 * Clears the dirty bit and flushes before if necessary. Only call this 533 * function when there are no pending requests, it does not guard against 534 * concurrent requests dirtying the image. 535 */ 536 static int qcow2_mark_clean(BlockDriverState *bs) 537 { 538 BDRVQcow2State *s = bs->opaque; 539 540 if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) { 541 int ret; 542 543 s->incompatible_features &= ~QCOW2_INCOMPAT_DIRTY; 544 545 ret = qcow2_flush_caches(bs); 546 if (ret < 0) { 547 return ret; 548 } 549 550 return qcow2_update_header(bs); 551 } 552 return 0; 553 } 554 555 /* 556 * Marks the image as corrupt. 557 */ 558 int qcow2_mark_corrupt(BlockDriverState *bs) 559 { 560 BDRVQcow2State *s = bs->opaque; 561 562 s->incompatible_features |= QCOW2_INCOMPAT_CORRUPT; 563 return qcow2_update_header(bs); 564 } 565 566 /* 567 * Marks the image as consistent, i.e., unsets the corrupt bit, and flushes 568 * before if necessary. 569 */ 570 int qcow2_mark_consistent(BlockDriverState *bs) 571 { 572 BDRVQcow2State *s = bs->opaque; 573 574 if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) { 575 int ret = qcow2_flush_caches(bs); 576 if (ret < 0) { 577 return ret; 578 } 579 580 s->incompatible_features &= ~QCOW2_INCOMPAT_CORRUPT; 581 return qcow2_update_header(bs); 582 } 583 return 0; 584 } 585 586 static void qcow2_add_check_result(BdrvCheckResult *out, 587 const BdrvCheckResult *src, 588 bool set_allocation_info) 589 { 590 out->corruptions += src->corruptions; 591 out->leaks += src->leaks; 592 out->check_errors += src->check_errors; 593 out->corruptions_fixed += src->corruptions_fixed; 594 out->leaks_fixed += src->leaks_fixed; 595 596 if (set_allocation_info) { 597 out->image_end_offset = src->image_end_offset; 598 out->bfi = src->bfi; 599 } 600 } 601 602 static int coroutine_fn qcow2_co_check_locked(BlockDriverState *bs, 603 BdrvCheckResult *result, 604 BdrvCheckMode fix) 605 { 606 BdrvCheckResult snapshot_res = {}; 607 BdrvCheckResult refcount_res = {}; 608 int ret; 609 610 memset(result, 0, sizeof(*result)); 611 612 ret = qcow2_check_read_snapshot_table(bs, &snapshot_res, fix); 613 if (ret < 0) { 614 qcow2_add_check_result(result, &snapshot_res, false); 615 return ret; 616 } 617 618 ret = qcow2_check_refcounts(bs, &refcount_res, fix); 619 qcow2_add_check_result(result, &refcount_res, true); 620 if (ret < 0) { 621 qcow2_add_check_result(result, &snapshot_res, false); 622 return ret; 623 } 624 625 ret = qcow2_check_fix_snapshot_table(bs, &snapshot_res, fix); 626 qcow2_add_check_result(result, &snapshot_res, false); 627 if (ret < 0) { 628 return ret; 629 } 630 631 if (fix && result->check_errors == 0 && result->corruptions == 0) { 632 ret = qcow2_mark_clean(bs); 633 if (ret < 0) { 634 return ret; 635 } 636 return qcow2_mark_consistent(bs); 637 } 638 return ret; 639 } 640 641 static int coroutine_fn qcow2_co_check(BlockDriverState *bs, 642 BdrvCheckResult *result, 643 BdrvCheckMode fix) 644 { 645 BDRVQcow2State *s = bs->opaque; 646 int ret; 647 648 qemu_co_mutex_lock(&s->lock); 649 ret = qcow2_co_check_locked(bs, result, fix); 650 qemu_co_mutex_unlock(&s->lock); 651 return ret; 652 } 653 654 int qcow2_validate_table(BlockDriverState *bs, uint64_t offset, 655 uint64_t entries, size_t entry_len, 656 int64_t max_size_bytes, const char *table_name, 657 Error **errp) 658 { 659 BDRVQcow2State *s = bs->opaque; 660 661 if (entries > max_size_bytes / entry_len) { 662 error_setg(errp, "%s too large", table_name); 663 return -EFBIG; 664 } 665 666 /* Use signed INT64_MAX as the maximum even for uint64_t header fields, 667 * because values will be passed to qemu functions taking int64_t. */ 668 if ((INT64_MAX - entries * entry_len < offset) || 669 (offset_into_cluster(s, offset) != 0)) { 670 error_setg(errp, "%s offset invalid", table_name); 671 return -EINVAL; 672 } 673 674 return 0; 675 } 676 677 static const char *const mutable_opts[] = { 678 QCOW2_OPT_LAZY_REFCOUNTS, 679 QCOW2_OPT_DISCARD_REQUEST, 680 QCOW2_OPT_DISCARD_SNAPSHOT, 681 QCOW2_OPT_DISCARD_OTHER, 682 QCOW2_OPT_OVERLAP, 683 QCOW2_OPT_OVERLAP_TEMPLATE, 684 QCOW2_OPT_OVERLAP_MAIN_HEADER, 685 QCOW2_OPT_OVERLAP_ACTIVE_L1, 686 QCOW2_OPT_OVERLAP_ACTIVE_L2, 687 QCOW2_OPT_OVERLAP_REFCOUNT_TABLE, 688 QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK, 689 QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE, 690 QCOW2_OPT_OVERLAP_INACTIVE_L1, 691 QCOW2_OPT_OVERLAP_INACTIVE_L2, 692 QCOW2_OPT_OVERLAP_BITMAP_DIRECTORY, 693 QCOW2_OPT_CACHE_SIZE, 694 QCOW2_OPT_L2_CACHE_SIZE, 695 QCOW2_OPT_L2_CACHE_ENTRY_SIZE, 696 QCOW2_OPT_REFCOUNT_CACHE_SIZE, 697 QCOW2_OPT_CACHE_CLEAN_INTERVAL, 698 NULL 699 }; 700 701 static QemuOptsList qcow2_runtime_opts = { 702 .name = "qcow2", 703 .head = QTAILQ_HEAD_INITIALIZER(qcow2_runtime_opts.head), 704 .desc = { 705 { 706 .name = QCOW2_OPT_LAZY_REFCOUNTS, 707 .type = QEMU_OPT_BOOL, 708 .help = "Postpone refcount updates", 709 }, 710 { 711 .name = QCOW2_OPT_DISCARD_REQUEST, 712 .type = QEMU_OPT_BOOL, 713 .help = "Pass guest discard requests to the layer below", 714 }, 715 { 716 .name = QCOW2_OPT_DISCARD_SNAPSHOT, 717 .type = QEMU_OPT_BOOL, 718 .help = "Generate discard requests when snapshot related space " 719 "is freed", 720 }, 721 { 722 .name = QCOW2_OPT_DISCARD_OTHER, 723 .type = QEMU_OPT_BOOL, 724 .help = "Generate discard requests when other clusters are freed", 725 }, 726 { 727 .name = QCOW2_OPT_OVERLAP, 728 .type = QEMU_OPT_STRING, 729 .help = "Selects which overlap checks to perform from a range of " 730 "templates (none, constant, cached, all)", 731 }, 732 { 733 .name = QCOW2_OPT_OVERLAP_TEMPLATE, 734 .type = QEMU_OPT_STRING, 735 .help = "Selects which overlap checks to perform from a range of " 736 "templates (none, constant, cached, all)", 737 }, 738 { 739 .name = QCOW2_OPT_OVERLAP_MAIN_HEADER, 740 .type = QEMU_OPT_BOOL, 741 .help = "Check for unintended writes into the main qcow2 header", 742 }, 743 { 744 .name = QCOW2_OPT_OVERLAP_ACTIVE_L1, 745 .type = QEMU_OPT_BOOL, 746 .help = "Check for unintended writes into the active L1 table", 747 }, 748 { 749 .name = QCOW2_OPT_OVERLAP_ACTIVE_L2, 750 .type = QEMU_OPT_BOOL, 751 .help = "Check for unintended writes into an active L2 table", 752 }, 753 { 754 .name = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE, 755 .type = QEMU_OPT_BOOL, 756 .help = "Check for unintended writes into the refcount table", 757 }, 758 { 759 .name = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK, 760 .type = QEMU_OPT_BOOL, 761 .help = "Check for unintended writes into a refcount block", 762 }, 763 { 764 .name = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE, 765 .type = QEMU_OPT_BOOL, 766 .help = "Check for unintended writes into the snapshot table", 767 }, 768 { 769 .name = QCOW2_OPT_OVERLAP_INACTIVE_L1, 770 .type = QEMU_OPT_BOOL, 771 .help = "Check for unintended writes into an inactive L1 table", 772 }, 773 { 774 .name = QCOW2_OPT_OVERLAP_INACTIVE_L2, 775 .type = QEMU_OPT_BOOL, 776 .help = "Check for unintended writes into an inactive L2 table", 777 }, 778 { 779 .name = QCOW2_OPT_OVERLAP_BITMAP_DIRECTORY, 780 .type = QEMU_OPT_BOOL, 781 .help = "Check for unintended writes into the bitmap directory", 782 }, 783 { 784 .name = QCOW2_OPT_CACHE_SIZE, 785 .type = QEMU_OPT_SIZE, 786 .help = "Maximum combined metadata (L2 tables and refcount blocks) " 787 "cache size", 788 }, 789 { 790 .name = QCOW2_OPT_L2_CACHE_SIZE, 791 .type = QEMU_OPT_SIZE, 792 .help = "Maximum L2 table cache size", 793 }, 794 { 795 .name = QCOW2_OPT_L2_CACHE_ENTRY_SIZE, 796 .type = QEMU_OPT_SIZE, 797 .help = "Size of each entry in the L2 cache", 798 }, 799 { 800 .name = QCOW2_OPT_REFCOUNT_CACHE_SIZE, 801 .type = QEMU_OPT_SIZE, 802 .help = "Maximum refcount block cache size", 803 }, 804 { 805 .name = QCOW2_OPT_CACHE_CLEAN_INTERVAL, 806 .type = QEMU_OPT_NUMBER, 807 .help = "Clean unused cache entries after this time (in seconds)", 808 }, 809 BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.", 810 "ID of secret providing qcow2 AES key or LUKS passphrase"), 811 { /* end of list */ } 812 }, 813 }; 814 815 static const char *overlap_bool_option_names[QCOW2_OL_MAX_BITNR] = { 816 [QCOW2_OL_MAIN_HEADER_BITNR] = QCOW2_OPT_OVERLAP_MAIN_HEADER, 817 [QCOW2_OL_ACTIVE_L1_BITNR] = QCOW2_OPT_OVERLAP_ACTIVE_L1, 818 [QCOW2_OL_ACTIVE_L2_BITNR] = QCOW2_OPT_OVERLAP_ACTIVE_L2, 819 [QCOW2_OL_REFCOUNT_TABLE_BITNR] = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE, 820 [QCOW2_OL_REFCOUNT_BLOCK_BITNR] = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK, 821 [QCOW2_OL_SNAPSHOT_TABLE_BITNR] = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE, 822 [QCOW2_OL_INACTIVE_L1_BITNR] = QCOW2_OPT_OVERLAP_INACTIVE_L1, 823 [QCOW2_OL_INACTIVE_L2_BITNR] = QCOW2_OPT_OVERLAP_INACTIVE_L2, 824 [QCOW2_OL_BITMAP_DIRECTORY_BITNR] = QCOW2_OPT_OVERLAP_BITMAP_DIRECTORY, 825 }; 826 827 static void cache_clean_timer_cb(void *opaque) 828 { 829 BlockDriverState *bs = opaque; 830 BDRVQcow2State *s = bs->opaque; 831 qcow2_cache_clean_unused(s->l2_table_cache); 832 qcow2_cache_clean_unused(s->refcount_block_cache); 833 timer_mod(s->cache_clean_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) + 834 (int64_t) s->cache_clean_interval * 1000); 835 } 836 837 static void cache_clean_timer_init(BlockDriverState *bs, AioContext *context) 838 { 839 BDRVQcow2State *s = bs->opaque; 840 if (s->cache_clean_interval > 0) { 841 s->cache_clean_timer = 842 aio_timer_new_with_attrs(context, QEMU_CLOCK_VIRTUAL, 843 SCALE_MS, QEMU_TIMER_ATTR_EXTERNAL, 844 cache_clean_timer_cb, bs); 845 timer_mod(s->cache_clean_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) + 846 (int64_t) s->cache_clean_interval * 1000); 847 } 848 } 849 850 static void cache_clean_timer_del(BlockDriverState *bs) 851 { 852 BDRVQcow2State *s = bs->opaque; 853 if (s->cache_clean_timer) { 854 timer_free(s->cache_clean_timer); 855 s->cache_clean_timer = NULL; 856 } 857 } 858 859 static void qcow2_detach_aio_context(BlockDriverState *bs) 860 { 861 cache_clean_timer_del(bs); 862 } 863 864 static void qcow2_attach_aio_context(BlockDriverState *bs, 865 AioContext *new_context) 866 { 867 cache_clean_timer_init(bs, new_context); 868 } 869 870 static bool read_cache_sizes(BlockDriverState *bs, QemuOpts *opts, 871 uint64_t *l2_cache_size, 872 uint64_t *l2_cache_entry_size, 873 uint64_t *refcount_cache_size, Error **errp) 874 { 875 BDRVQcow2State *s = bs->opaque; 876 uint64_t combined_cache_size, l2_cache_max_setting; 877 bool l2_cache_size_set, refcount_cache_size_set, combined_cache_size_set; 878 bool l2_cache_entry_size_set; 879 int min_refcount_cache = MIN_REFCOUNT_CACHE_SIZE * s->cluster_size; 880 uint64_t virtual_disk_size = bs->total_sectors * BDRV_SECTOR_SIZE; 881 uint64_t max_l2_entries = DIV_ROUND_UP(virtual_disk_size, s->cluster_size); 882 /* An L2 table is always one cluster in size so the max cache size 883 * should be a multiple of the cluster size. */ 884 uint64_t max_l2_cache = ROUND_UP(max_l2_entries * l2_entry_size(s), 885 s->cluster_size); 886 887 combined_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_CACHE_SIZE); 888 l2_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_L2_CACHE_SIZE); 889 refcount_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_REFCOUNT_CACHE_SIZE); 890 l2_cache_entry_size_set = qemu_opt_get(opts, QCOW2_OPT_L2_CACHE_ENTRY_SIZE); 891 892 combined_cache_size = qemu_opt_get_size(opts, QCOW2_OPT_CACHE_SIZE, 0); 893 l2_cache_max_setting = qemu_opt_get_size(opts, QCOW2_OPT_L2_CACHE_SIZE, 894 DEFAULT_L2_CACHE_MAX_SIZE); 895 *refcount_cache_size = qemu_opt_get_size(opts, 896 QCOW2_OPT_REFCOUNT_CACHE_SIZE, 0); 897 898 *l2_cache_entry_size = qemu_opt_get_size( 899 opts, QCOW2_OPT_L2_CACHE_ENTRY_SIZE, s->cluster_size); 900 901 *l2_cache_size = MIN(max_l2_cache, l2_cache_max_setting); 902 903 if (combined_cache_size_set) { 904 if (l2_cache_size_set && refcount_cache_size_set) { 905 error_setg(errp, QCOW2_OPT_CACHE_SIZE ", " QCOW2_OPT_L2_CACHE_SIZE 906 " and " QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not be set " 907 "at the same time"); 908 return false; 909 } else if (l2_cache_size_set && 910 (l2_cache_max_setting > combined_cache_size)) { 911 error_setg(errp, QCOW2_OPT_L2_CACHE_SIZE " may not exceed " 912 QCOW2_OPT_CACHE_SIZE); 913 return false; 914 } else if (*refcount_cache_size > combined_cache_size) { 915 error_setg(errp, QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not exceed " 916 QCOW2_OPT_CACHE_SIZE); 917 return false; 918 } 919 920 if (l2_cache_size_set) { 921 *refcount_cache_size = combined_cache_size - *l2_cache_size; 922 } else if (refcount_cache_size_set) { 923 *l2_cache_size = combined_cache_size - *refcount_cache_size; 924 } else { 925 /* Assign as much memory as possible to the L2 cache, and 926 * use the remainder for the refcount cache */ 927 if (combined_cache_size >= max_l2_cache + min_refcount_cache) { 928 *l2_cache_size = max_l2_cache; 929 *refcount_cache_size = combined_cache_size - *l2_cache_size; 930 } else { 931 *refcount_cache_size = 932 MIN(combined_cache_size, min_refcount_cache); 933 *l2_cache_size = combined_cache_size - *refcount_cache_size; 934 } 935 } 936 } 937 938 /* 939 * If the L2 cache is not enough to cover the whole disk then 940 * default to 4KB entries. Smaller entries reduce the cost of 941 * loads and evictions and increase I/O performance. 942 */ 943 if (*l2_cache_size < max_l2_cache && !l2_cache_entry_size_set) { 944 *l2_cache_entry_size = MIN(s->cluster_size, 4096); 945 } 946 947 /* l2_cache_size and refcount_cache_size are ensured to have at least 948 * their minimum values in qcow2_update_options_prepare() */ 949 950 if (*l2_cache_entry_size < (1 << MIN_CLUSTER_BITS) || 951 *l2_cache_entry_size > s->cluster_size || 952 !is_power_of_2(*l2_cache_entry_size)) { 953 error_setg(errp, "L2 cache entry size must be a power of two " 954 "between %d and the cluster size (%d)", 955 1 << MIN_CLUSTER_BITS, s->cluster_size); 956 return false; 957 } 958 959 return true; 960 } 961 962 typedef struct Qcow2ReopenState { 963 Qcow2Cache *l2_table_cache; 964 Qcow2Cache *refcount_block_cache; 965 int l2_slice_size; /* Number of entries in a slice of the L2 table */ 966 bool use_lazy_refcounts; 967 int overlap_check; 968 bool discard_passthrough[QCOW2_DISCARD_MAX]; 969 uint64_t cache_clean_interval; 970 QCryptoBlockOpenOptions *crypto_opts; /* Disk encryption runtime options */ 971 } Qcow2ReopenState; 972 973 static int qcow2_update_options_prepare(BlockDriverState *bs, 974 Qcow2ReopenState *r, 975 QDict *options, int flags, 976 Error **errp) 977 { 978 BDRVQcow2State *s = bs->opaque; 979 QemuOpts *opts = NULL; 980 const char *opt_overlap_check, *opt_overlap_check_template; 981 int overlap_check_template = 0; 982 uint64_t l2_cache_size, l2_cache_entry_size, refcount_cache_size; 983 int i; 984 const char *encryptfmt; 985 QDict *encryptopts = NULL; 986 int ret; 987 988 qdict_extract_subqdict(options, &encryptopts, "encrypt."); 989 encryptfmt = qdict_get_try_str(encryptopts, "format"); 990 991 opts = qemu_opts_create(&qcow2_runtime_opts, NULL, 0, &error_abort); 992 if (!qemu_opts_absorb_qdict(opts, options, errp)) { 993 ret = -EINVAL; 994 goto fail; 995 } 996 997 /* get L2 table/refcount block cache size from command line options */ 998 if (!read_cache_sizes(bs, opts, &l2_cache_size, &l2_cache_entry_size, 999 &refcount_cache_size, errp)) { 1000 ret = -EINVAL; 1001 goto fail; 1002 } 1003 1004 l2_cache_size /= l2_cache_entry_size; 1005 if (l2_cache_size < MIN_L2_CACHE_SIZE) { 1006 l2_cache_size = MIN_L2_CACHE_SIZE; 1007 } 1008 if (l2_cache_size > INT_MAX) { 1009 error_setg(errp, "L2 cache size too big"); 1010 ret = -EINVAL; 1011 goto fail; 1012 } 1013 1014 refcount_cache_size /= s->cluster_size; 1015 if (refcount_cache_size < MIN_REFCOUNT_CACHE_SIZE) { 1016 refcount_cache_size = MIN_REFCOUNT_CACHE_SIZE; 1017 } 1018 if (refcount_cache_size > INT_MAX) { 1019 error_setg(errp, "Refcount cache size too big"); 1020 ret = -EINVAL; 1021 goto fail; 1022 } 1023 1024 /* alloc new L2 table/refcount block cache, flush old one */ 1025 if (s->l2_table_cache) { 1026 ret = qcow2_cache_flush(bs, s->l2_table_cache); 1027 if (ret) { 1028 error_setg_errno(errp, -ret, "Failed to flush the L2 table cache"); 1029 goto fail; 1030 } 1031 } 1032 1033 if (s->refcount_block_cache) { 1034 ret = qcow2_cache_flush(bs, s->refcount_block_cache); 1035 if (ret) { 1036 error_setg_errno(errp, -ret, 1037 "Failed to flush the refcount block cache"); 1038 goto fail; 1039 } 1040 } 1041 1042 r->l2_slice_size = l2_cache_entry_size / l2_entry_size(s); 1043 r->l2_table_cache = qcow2_cache_create(bs, l2_cache_size, 1044 l2_cache_entry_size); 1045 r->refcount_block_cache = qcow2_cache_create(bs, refcount_cache_size, 1046 s->cluster_size); 1047 if (r->l2_table_cache == NULL || r->refcount_block_cache == NULL) { 1048 error_setg(errp, "Could not allocate metadata caches"); 1049 ret = -ENOMEM; 1050 goto fail; 1051 } 1052 1053 /* New interval for cache cleanup timer */ 1054 r->cache_clean_interval = 1055 qemu_opt_get_number(opts, QCOW2_OPT_CACHE_CLEAN_INTERVAL, 1056 DEFAULT_CACHE_CLEAN_INTERVAL); 1057 #ifndef CONFIG_LINUX 1058 if (r->cache_clean_interval != 0) { 1059 error_setg(errp, QCOW2_OPT_CACHE_CLEAN_INTERVAL 1060 " not supported on this host"); 1061 ret = -EINVAL; 1062 goto fail; 1063 } 1064 #endif 1065 if (r->cache_clean_interval > UINT_MAX) { 1066 error_setg(errp, "Cache clean interval too big"); 1067 ret = -EINVAL; 1068 goto fail; 1069 } 1070 1071 /* lazy-refcounts; flush if going from enabled to disabled */ 1072 r->use_lazy_refcounts = qemu_opt_get_bool(opts, QCOW2_OPT_LAZY_REFCOUNTS, 1073 (s->compatible_features & QCOW2_COMPAT_LAZY_REFCOUNTS)); 1074 if (r->use_lazy_refcounts && s->qcow_version < 3) { 1075 error_setg(errp, "Lazy refcounts require a qcow2 image with at least " 1076 "qemu 1.1 compatibility level"); 1077 ret = -EINVAL; 1078 goto fail; 1079 } 1080 1081 if (s->use_lazy_refcounts && !r->use_lazy_refcounts) { 1082 ret = qcow2_mark_clean(bs); 1083 if (ret < 0) { 1084 error_setg_errno(errp, -ret, "Failed to disable lazy refcounts"); 1085 goto fail; 1086 } 1087 } 1088 1089 /* Overlap check options */ 1090 opt_overlap_check = qemu_opt_get(opts, QCOW2_OPT_OVERLAP); 1091 opt_overlap_check_template = qemu_opt_get(opts, QCOW2_OPT_OVERLAP_TEMPLATE); 1092 if (opt_overlap_check_template && opt_overlap_check && 1093 strcmp(opt_overlap_check_template, opt_overlap_check)) 1094 { 1095 error_setg(errp, "Conflicting values for qcow2 options '" 1096 QCOW2_OPT_OVERLAP "' ('%s') and '" QCOW2_OPT_OVERLAP_TEMPLATE 1097 "' ('%s')", opt_overlap_check, opt_overlap_check_template); 1098 ret = -EINVAL; 1099 goto fail; 1100 } 1101 if (!opt_overlap_check) { 1102 opt_overlap_check = opt_overlap_check_template ?: "cached"; 1103 } 1104 1105 if (!strcmp(opt_overlap_check, "none")) { 1106 overlap_check_template = 0; 1107 } else if (!strcmp(opt_overlap_check, "constant")) { 1108 overlap_check_template = QCOW2_OL_CONSTANT; 1109 } else if (!strcmp(opt_overlap_check, "cached")) { 1110 overlap_check_template = QCOW2_OL_CACHED; 1111 } else if (!strcmp(opt_overlap_check, "all")) { 1112 overlap_check_template = QCOW2_OL_ALL; 1113 } else { 1114 error_setg(errp, "Unsupported value '%s' for qcow2 option " 1115 "'overlap-check'. Allowed are any of the following: " 1116 "none, constant, cached, all", opt_overlap_check); 1117 ret = -EINVAL; 1118 goto fail; 1119 } 1120 1121 r->overlap_check = 0; 1122 for (i = 0; i < QCOW2_OL_MAX_BITNR; i++) { 1123 /* overlap-check defines a template bitmask, but every flag may be 1124 * overwritten through the associated boolean option */ 1125 r->overlap_check |= 1126 qemu_opt_get_bool(opts, overlap_bool_option_names[i], 1127 overlap_check_template & (1 << i)) << i; 1128 } 1129 1130 r->discard_passthrough[QCOW2_DISCARD_NEVER] = false; 1131 r->discard_passthrough[QCOW2_DISCARD_ALWAYS] = true; 1132 r->discard_passthrough[QCOW2_DISCARD_REQUEST] = 1133 qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_REQUEST, 1134 flags & BDRV_O_UNMAP); 1135 r->discard_passthrough[QCOW2_DISCARD_SNAPSHOT] = 1136 qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_SNAPSHOT, true); 1137 r->discard_passthrough[QCOW2_DISCARD_OTHER] = 1138 qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_OTHER, false); 1139 1140 switch (s->crypt_method_header) { 1141 case QCOW_CRYPT_NONE: 1142 if (encryptfmt) { 1143 error_setg(errp, "No encryption in image header, but options " 1144 "specified format '%s'", encryptfmt); 1145 ret = -EINVAL; 1146 goto fail; 1147 } 1148 break; 1149 1150 case QCOW_CRYPT_AES: 1151 if (encryptfmt && !g_str_equal(encryptfmt, "aes")) { 1152 error_setg(errp, 1153 "Header reported 'aes' encryption format but " 1154 "options specify '%s'", encryptfmt); 1155 ret = -EINVAL; 1156 goto fail; 1157 } 1158 qdict_put_str(encryptopts, "format", "qcow"); 1159 r->crypto_opts = block_crypto_open_opts_init(encryptopts, errp); 1160 if (!r->crypto_opts) { 1161 ret = -EINVAL; 1162 goto fail; 1163 } 1164 break; 1165 1166 case QCOW_CRYPT_LUKS: 1167 if (encryptfmt && !g_str_equal(encryptfmt, "luks")) { 1168 error_setg(errp, 1169 "Header reported 'luks' encryption format but " 1170 "options specify '%s'", encryptfmt); 1171 ret = -EINVAL; 1172 goto fail; 1173 } 1174 qdict_put_str(encryptopts, "format", "luks"); 1175 r->crypto_opts = block_crypto_open_opts_init(encryptopts, errp); 1176 if (!r->crypto_opts) { 1177 ret = -EINVAL; 1178 goto fail; 1179 } 1180 break; 1181 1182 default: 1183 error_setg(errp, "Unsupported encryption method %d", 1184 s->crypt_method_header); 1185 ret = -EINVAL; 1186 goto fail; 1187 } 1188 1189 ret = 0; 1190 fail: 1191 qobject_unref(encryptopts); 1192 qemu_opts_del(opts); 1193 opts = NULL; 1194 return ret; 1195 } 1196 1197 static void qcow2_update_options_commit(BlockDriverState *bs, 1198 Qcow2ReopenState *r) 1199 { 1200 BDRVQcow2State *s = bs->opaque; 1201 int i; 1202 1203 if (s->l2_table_cache) { 1204 qcow2_cache_destroy(s->l2_table_cache); 1205 } 1206 if (s->refcount_block_cache) { 1207 qcow2_cache_destroy(s->refcount_block_cache); 1208 } 1209 s->l2_table_cache = r->l2_table_cache; 1210 s->refcount_block_cache = r->refcount_block_cache; 1211 s->l2_slice_size = r->l2_slice_size; 1212 1213 s->overlap_check = r->overlap_check; 1214 s->use_lazy_refcounts = r->use_lazy_refcounts; 1215 1216 for (i = 0; i < QCOW2_DISCARD_MAX; i++) { 1217 s->discard_passthrough[i] = r->discard_passthrough[i]; 1218 } 1219 1220 if (s->cache_clean_interval != r->cache_clean_interval) { 1221 cache_clean_timer_del(bs); 1222 s->cache_clean_interval = r->cache_clean_interval; 1223 cache_clean_timer_init(bs, bdrv_get_aio_context(bs)); 1224 } 1225 1226 qapi_free_QCryptoBlockOpenOptions(s->crypto_opts); 1227 s->crypto_opts = r->crypto_opts; 1228 } 1229 1230 static void qcow2_update_options_abort(BlockDriverState *bs, 1231 Qcow2ReopenState *r) 1232 { 1233 if (r->l2_table_cache) { 1234 qcow2_cache_destroy(r->l2_table_cache); 1235 } 1236 if (r->refcount_block_cache) { 1237 qcow2_cache_destroy(r->refcount_block_cache); 1238 } 1239 qapi_free_QCryptoBlockOpenOptions(r->crypto_opts); 1240 } 1241 1242 static int qcow2_update_options(BlockDriverState *bs, QDict *options, 1243 int flags, Error **errp) 1244 { 1245 Qcow2ReopenState r = {}; 1246 int ret; 1247 1248 ret = qcow2_update_options_prepare(bs, &r, options, flags, errp); 1249 if (ret >= 0) { 1250 qcow2_update_options_commit(bs, &r); 1251 } else { 1252 qcow2_update_options_abort(bs, &r); 1253 } 1254 1255 return ret; 1256 } 1257 1258 static int validate_compression_type(BDRVQcow2State *s, Error **errp) 1259 { 1260 switch (s->compression_type) { 1261 case QCOW2_COMPRESSION_TYPE_ZLIB: 1262 #ifdef CONFIG_ZSTD 1263 case QCOW2_COMPRESSION_TYPE_ZSTD: 1264 #endif 1265 break; 1266 1267 default: 1268 error_setg(errp, "qcow2: unknown compression type: %u", 1269 s->compression_type); 1270 return -ENOTSUP; 1271 } 1272 1273 /* 1274 * if the compression type differs from QCOW2_COMPRESSION_TYPE_ZLIB 1275 * the incompatible feature flag must be set 1276 */ 1277 if (s->compression_type == QCOW2_COMPRESSION_TYPE_ZLIB) { 1278 if (s->incompatible_features & QCOW2_INCOMPAT_COMPRESSION) { 1279 error_setg(errp, "qcow2: Compression type incompatible feature " 1280 "bit must not be set"); 1281 return -EINVAL; 1282 } 1283 } else { 1284 if (!(s->incompatible_features & QCOW2_INCOMPAT_COMPRESSION)) { 1285 error_setg(errp, "qcow2: Compression type incompatible feature " 1286 "bit must be set"); 1287 return -EINVAL; 1288 } 1289 } 1290 1291 return 0; 1292 } 1293 1294 /* Called with s->lock held. */ 1295 static int coroutine_fn qcow2_do_open(BlockDriverState *bs, QDict *options, 1296 int flags, bool open_data_file, 1297 Error **errp) 1298 { 1299 ERRP_GUARD(); 1300 BDRVQcow2State *s = bs->opaque; 1301 unsigned int len, i; 1302 int ret = 0; 1303 QCowHeader header; 1304 uint64_t ext_end; 1305 uint64_t l1_vm_state_index; 1306 bool update_header = false; 1307 1308 ret = bdrv_pread(bs->file, 0, sizeof(header), &header, 0); 1309 if (ret < 0) { 1310 error_setg_errno(errp, -ret, "Could not read qcow2 header"); 1311 goto fail; 1312 } 1313 header.magic = be32_to_cpu(header.magic); 1314 header.version = be32_to_cpu(header.version); 1315 header.backing_file_offset = be64_to_cpu(header.backing_file_offset); 1316 header.backing_file_size = be32_to_cpu(header.backing_file_size); 1317 header.size = be64_to_cpu(header.size); 1318 header.cluster_bits = be32_to_cpu(header.cluster_bits); 1319 header.crypt_method = be32_to_cpu(header.crypt_method); 1320 header.l1_table_offset = be64_to_cpu(header.l1_table_offset); 1321 header.l1_size = be32_to_cpu(header.l1_size); 1322 header.refcount_table_offset = be64_to_cpu(header.refcount_table_offset); 1323 header.refcount_table_clusters = 1324 be32_to_cpu(header.refcount_table_clusters); 1325 header.snapshots_offset = be64_to_cpu(header.snapshots_offset); 1326 header.nb_snapshots = be32_to_cpu(header.nb_snapshots); 1327 1328 if (header.magic != QCOW_MAGIC) { 1329 error_setg(errp, "Image is not in qcow2 format"); 1330 ret = -EINVAL; 1331 goto fail; 1332 } 1333 if (header.version < 2 || header.version > 3) { 1334 error_setg(errp, "Unsupported qcow2 version %" PRIu32, header.version); 1335 ret = -ENOTSUP; 1336 goto fail; 1337 } 1338 1339 s->qcow_version = header.version; 1340 1341 /* Initialise cluster size */ 1342 if (header.cluster_bits < MIN_CLUSTER_BITS || 1343 header.cluster_bits > MAX_CLUSTER_BITS) { 1344 error_setg(errp, "Unsupported cluster size: 2^%" PRIu32, 1345 header.cluster_bits); 1346 ret = -EINVAL; 1347 goto fail; 1348 } 1349 1350 s->cluster_bits = header.cluster_bits; 1351 s->cluster_size = 1 << s->cluster_bits; 1352 1353 /* Initialise version 3 header fields */ 1354 if (header.version == 2) { 1355 header.incompatible_features = 0; 1356 header.compatible_features = 0; 1357 header.autoclear_features = 0; 1358 header.refcount_order = 4; 1359 header.header_length = 72; 1360 } else { 1361 header.incompatible_features = 1362 be64_to_cpu(header.incompatible_features); 1363 header.compatible_features = be64_to_cpu(header.compatible_features); 1364 header.autoclear_features = be64_to_cpu(header.autoclear_features); 1365 header.refcount_order = be32_to_cpu(header.refcount_order); 1366 header.header_length = be32_to_cpu(header.header_length); 1367 1368 if (header.header_length < 104) { 1369 error_setg(errp, "qcow2 header too short"); 1370 ret = -EINVAL; 1371 goto fail; 1372 } 1373 } 1374 1375 if (header.header_length > s->cluster_size) { 1376 error_setg(errp, "qcow2 header exceeds cluster size"); 1377 ret = -EINVAL; 1378 goto fail; 1379 } 1380 1381 if (header.header_length > sizeof(header)) { 1382 s->unknown_header_fields_size = header.header_length - sizeof(header); 1383 s->unknown_header_fields = g_malloc(s->unknown_header_fields_size); 1384 ret = bdrv_pread(bs->file, sizeof(header), 1385 s->unknown_header_fields_size, 1386 s->unknown_header_fields, 0); 1387 if (ret < 0) { 1388 error_setg_errno(errp, -ret, "Could not read unknown qcow2 header " 1389 "fields"); 1390 goto fail; 1391 } 1392 } 1393 1394 if (header.backing_file_offset > s->cluster_size) { 1395 error_setg(errp, "Invalid backing file offset"); 1396 ret = -EINVAL; 1397 goto fail; 1398 } 1399 1400 if (header.backing_file_offset) { 1401 ext_end = header.backing_file_offset; 1402 } else { 1403 ext_end = 1 << header.cluster_bits; 1404 } 1405 1406 /* Handle feature bits */ 1407 s->incompatible_features = header.incompatible_features; 1408 s->compatible_features = header.compatible_features; 1409 s->autoclear_features = header.autoclear_features; 1410 1411 /* 1412 * Handle compression type 1413 * Older qcow2 images don't contain the compression type header. 1414 * Distinguish them by the header length and use 1415 * the only valid (default) compression type in that case 1416 */ 1417 if (header.header_length > offsetof(QCowHeader, compression_type)) { 1418 s->compression_type = header.compression_type; 1419 } else { 1420 s->compression_type = QCOW2_COMPRESSION_TYPE_ZLIB; 1421 } 1422 1423 ret = validate_compression_type(s, errp); 1424 if (ret) { 1425 goto fail; 1426 } 1427 1428 if (s->incompatible_features & ~QCOW2_INCOMPAT_MASK) { 1429 void *feature_table = NULL; 1430 qcow2_read_extensions(bs, header.header_length, ext_end, 1431 &feature_table, flags, NULL, NULL); 1432 report_unsupported_feature(errp, feature_table, 1433 s->incompatible_features & 1434 ~QCOW2_INCOMPAT_MASK); 1435 ret = -ENOTSUP; 1436 g_free(feature_table); 1437 goto fail; 1438 } 1439 1440 if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) { 1441 /* Corrupt images may not be written to unless they are being repaired 1442 */ 1443 if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_CHECK)) { 1444 error_setg(errp, "qcow2: Image is corrupt; cannot be opened " 1445 "read/write"); 1446 ret = -EACCES; 1447 goto fail; 1448 } 1449 } 1450 1451 s->subclusters_per_cluster = 1452 has_subclusters(s) ? QCOW_EXTL2_SUBCLUSTERS_PER_CLUSTER : 1; 1453 s->subcluster_size = s->cluster_size / s->subclusters_per_cluster; 1454 s->subcluster_bits = ctz32(s->subcluster_size); 1455 1456 if (s->subcluster_size < (1 << MIN_CLUSTER_BITS)) { 1457 error_setg(errp, "Unsupported subcluster size: %d", s->subcluster_size); 1458 ret = -EINVAL; 1459 goto fail; 1460 } 1461 1462 /* Check support for various header values */ 1463 if (header.refcount_order > 6) { 1464 error_setg(errp, "Reference count entry width too large; may not " 1465 "exceed 64 bits"); 1466 ret = -EINVAL; 1467 goto fail; 1468 } 1469 s->refcount_order = header.refcount_order; 1470 s->refcount_bits = 1 << s->refcount_order; 1471 s->refcount_max = UINT64_C(1) << (s->refcount_bits - 1); 1472 s->refcount_max += s->refcount_max - 1; 1473 1474 s->crypt_method_header = header.crypt_method; 1475 if (s->crypt_method_header) { 1476 if (bdrv_uses_whitelist() && 1477 s->crypt_method_header == QCOW_CRYPT_AES) { 1478 error_setg(errp, 1479 "Use of AES-CBC encrypted qcow2 images is no longer " 1480 "supported in system emulators"); 1481 error_append_hint(errp, 1482 "You can use 'qemu-img convert' to convert your " 1483 "image to an alternative supported format, such " 1484 "as unencrypted qcow2, or raw with the LUKS " 1485 "format instead.\n"); 1486 ret = -ENOSYS; 1487 goto fail; 1488 } 1489 1490 if (s->crypt_method_header == QCOW_CRYPT_AES) { 1491 s->crypt_physical_offset = false; 1492 } else { 1493 /* Assuming LUKS and any future crypt methods we 1494 * add will all use physical offsets, due to the 1495 * fact that the alternative is insecure... */ 1496 s->crypt_physical_offset = true; 1497 } 1498 1499 bs->encrypted = true; 1500 } 1501 1502 s->l2_bits = s->cluster_bits - ctz32(l2_entry_size(s)); 1503 s->l2_size = 1 << s->l2_bits; 1504 /* 2^(s->refcount_order - 3) is the refcount width in bytes */ 1505 s->refcount_block_bits = s->cluster_bits - (s->refcount_order - 3); 1506 s->refcount_block_size = 1 << s->refcount_block_bits; 1507 bs->total_sectors = header.size / BDRV_SECTOR_SIZE; 1508 s->csize_shift = (62 - (s->cluster_bits - 8)); 1509 s->csize_mask = (1 << (s->cluster_bits - 8)) - 1; 1510 s->cluster_offset_mask = (1LL << s->csize_shift) - 1; 1511 1512 s->refcount_table_offset = header.refcount_table_offset; 1513 s->refcount_table_size = 1514 header.refcount_table_clusters << (s->cluster_bits - 3); 1515 1516 if (header.refcount_table_clusters == 0 && !(flags & BDRV_O_CHECK)) { 1517 error_setg(errp, "Image does not contain a reference count table"); 1518 ret = -EINVAL; 1519 goto fail; 1520 } 1521 1522 ret = qcow2_validate_table(bs, s->refcount_table_offset, 1523 header.refcount_table_clusters, 1524 s->cluster_size, QCOW_MAX_REFTABLE_SIZE, 1525 "Reference count table", errp); 1526 if (ret < 0) { 1527 goto fail; 1528 } 1529 1530 if (!(flags & BDRV_O_CHECK)) { 1531 /* 1532 * The total size in bytes of the snapshot table is checked in 1533 * qcow2_read_snapshots() because the size of each snapshot is 1534 * variable and we don't know it yet. 1535 * Here we only check the offset and number of snapshots. 1536 */ 1537 ret = qcow2_validate_table(bs, header.snapshots_offset, 1538 header.nb_snapshots, 1539 sizeof(QCowSnapshotHeader), 1540 sizeof(QCowSnapshotHeader) * 1541 QCOW_MAX_SNAPSHOTS, 1542 "Snapshot table", errp); 1543 if (ret < 0) { 1544 goto fail; 1545 } 1546 } 1547 1548 /* read the level 1 table */ 1549 ret = qcow2_validate_table(bs, header.l1_table_offset, 1550 header.l1_size, L1E_SIZE, 1551 QCOW_MAX_L1_SIZE, "Active L1 table", errp); 1552 if (ret < 0) { 1553 goto fail; 1554 } 1555 s->l1_size = header.l1_size; 1556 s->l1_table_offset = header.l1_table_offset; 1557 1558 l1_vm_state_index = size_to_l1(s, header.size); 1559 if (l1_vm_state_index > INT_MAX) { 1560 error_setg(errp, "Image is too big"); 1561 ret = -EFBIG; 1562 goto fail; 1563 } 1564 s->l1_vm_state_index = l1_vm_state_index; 1565 1566 /* the L1 table must contain at least enough entries to put 1567 header.size bytes */ 1568 if (s->l1_size < s->l1_vm_state_index) { 1569 error_setg(errp, "L1 table is too small"); 1570 ret = -EINVAL; 1571 goto fail; 1572 } 1573 1574 if (s->l1_size > 0) { 1575 s->l1_table = qemu_try_blockalign(bs->file->bs, s->l1_size * L1E_SIZE); 1576 if (s->l1_table == NULL) { 1577 error_setg(errp, "Could not allocate L1 table"); 1578 ret = -ENOMEM; 1579 goto fail; 1580 } 1581 ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_size * L1E_SIZE, 1582 s->l1_table, 0); 1583 if (ret < 0) { 1584 error_setg_errno(errp, -ret, "Could not read L1 table"); 1585 goto fail; 1586 } 1587 for(i = 0;i < s->l1_size; i++) { 1588 s->l1_table[i] = be64_to_cpu(s->l1_table[i]); 1589 } 1590 } 1591 1592 /* Parse driver-specific options */ 1593 ret = qcow2_update_options(bs, options, flags, errp); 1594 if (ret < 0) { 1595 goto fail; 1596 } 1597 1598 s->flags = flags; 1599 1600 ret = qcow2_refcount_init(bs); 1601 if (ret != 0) { 1602 error_setg_errno(errp, -ret, "Could not initialize refcount handling"); 1603 goto fail; 1604 } 1605 1606 QLIST_INIT(&s->cluster_allocs); 1607 QTAILQ_INIT(&s->discards); 1608 1609 /* read qcow2 extensions */ 1610 if (qcow2_read_extensions(bs, header.header_length, ext_end, NULL, 1611 flags, &update_header, errp)) { 1612 ret = -EINVAL; 1613 goto fail; 1614 } 1615 1616 if (open_data_file) { 1617 /* Open external data file */ 1618 s->data_file = bdrv_open_child(NULL, options, "data-file", bs, 1619 &child_of_bds, BDRV_CHILD_DATA, 1620 true, errp); 1621 if (*errp) { 1622 ret = -EINVAL; 1623 goto fail; 1624 } 1625 1626 if (s->incompatible_features & QCOW2_INCOMPAT_DATA_FILE) { 1627 if (!s->data_file && s->image_data_file) { 1628 s->data_file = bdrv_open_child(s->image_data_file, options, 1629 "data-file", bs, &child_of_bds, 1630 BDRV_CHILD_DATA, false, errp); 1631 if (!s->data_file) { 1632 ret = -EINVAL; 1633 goto fail; 1634 } 1635 } 1636 if (!s->data_file) { 1637 error_setg(errp, "'data-file' is required for this image"); 1638 ret = -EINVAL; 1639 goto fail; 1640 } 1641 1642 /* No data here */ 1643 bs->file->role &= ~BDRV_CHILD_DATA; 1644 1645 /* Must succeed because we have given up permissions if anything */ 1646 bdrv_child_refresh_perms(bs, bs->file, &error_abort); 1647 } else { 1648 if (s->data_file) { 1649 error_setg(errp, "'data-file' can only be set for images with " 1650 "an external data file"); 1651 ret = -EINVAL; 1652 goto fail; 1653 } 1654 1655 s->data_file = bs->file; 1656 1657 if (data_file_is_raw(bs)) { 1658 error_setg(errp, "data-file-raw requires a data file"); 1659 ret = -EINVAL; 1660 goto fail; 1661 } 1662 } 1663 } 1664 1665 /* qcow2_read_extension may have set up the crypto context 1666 * if the crypt method needs a header region, some methods 1667 * don't need header extensions, so must check here 1668 */ 1669 if (s->crypt_method_header && !s->crypto) { 1670 if (s->crypt_method_header == QCOW_CRYPT_AES) { 1671 unsigned int cflags = 0; 1672 if (flags & BDRV_O_NO_IO) { 1673 cflags |= QCRYPTO_BLOCK_OPEN_NO_IO; 1674 } 1675 s->crypto = qcrypto_block_open(s->crypto_opts, "encrypt.", 1676 NULL, NULL, cflags, 1677 QCOW2_MAX_THREADS, errp); 1678 if (!s->crypto) { 1679 ret = -EINVAL; 1680 goto fail; 1681 } 1682 } else if (!(flags & BDRV_O_NO_IO)) { 1683 error_setg(errp, "Missing CRYPTO header for crypt method %d", 1684 s->crypt_method_header); 1685 ret = -EINVAL; 1686 goto fail; 1687 } 1688 } 1689 1690 /* read the backing file name */ 1691 if (header.backing_file_offset != 0) { 1692 len = header.backing_file_size; 1693 if (len > MIN(1023, s->cluster_size - header.backing_file_offset) || 1694 len >= sizeof(bs->backing_file)) { 1695 error_setg(errp, "Backing file name too long"); 1696 ret = -EINVAL; 1697 goto fail; 1698 } 1699 ret = bdrv_pread(bs->file, header.backing_file_offset, len, 1700 bs->auto_backing_file, 0); 1701 if (ret < 0) { 1702 error_setg_errno(errp, -ret, "Could not read backing file name"); 1703 goto fail; 1704 } 1705 bs->auto_backing_file[len] = '\0'; 1706 pstrcpy(bs->backing_file, sizeof(bs->backing_file), 1707 bs->auto_backing_file); 1708 s->image_backing_file = g_strdup(bs->auto_backing_file); 1709 } 1710 1711 /* 1712 * Internal snapshots; skip reading them in check mode, because 1713 * we do not need them then, and we do not want to abort because 1714 * of a broken table. 1715 */ 1716 if (!(flags & BDRV_O_CHECK)) { 1717 s->snapshots_offset = header.snapshots_offset; 1718 s->nb_snapshots = header.nb_snapshots; 1719 1720 ret = qcow2_read_snapshots(bs, errp); 1721 if (ret < 0) { 1722 goto fail; 1723 } 1724 } 1725 1726 /* Clear unknown autoclear feature bits */ 1727 update_header |= s->autoclear_features & ~QCOW2_AUTOCLEAR_MASK; 1728 update_header = update_header && bdrv_is_writable(bs); 1729 if (update_header) { 1730 s->autoclear_features &= QCOW2_AUTOCLEAR_MASK; 1731 } 1732 1733 /* == Handle persistent dirty bitmaps == 1734 * 1735 * We want load dirty bitmaps in three cases: 1736 * 1737 * 1. Normal open of the disk in active mode, not related to invalidation 1738 * after migration. 1739 * 1740 * 2. Invalidation of the target vm after pre-copy phase of migration, if 1741 * bitmaps are _not_ migrating through migration channel, i.e. 1742 * 'dirty-bitmaps' capability is disabled. 1743 * 1744 * 3. Invalidation of source vm after failed or canceled migration. 1745 * This is a very interesting case. There are two possible types of 1746 * bitmaps: 1747 * 1748 * A. Stored on inactivation and removed. They should be loaded from the 1749 * image. 1750 * 1751 * B. Not stored: not-persistent bitmaps and bitmaps, migrated through 1752 * the migration channel (with dirty-bitmaps capability). 1753 * 1754 * On the other hand, there are two possible sub-cases: 1755 * 1756 * 3.1 disk was changed by somebody else while were inactive. In this 1757 * case all in-RAM dirty bitmaps (both persistent and not) are 1758 * definitely invalid. And we don't have any method to determine 1759 * this. 1760 * 1761 * Simple and safe thing is to just drop all the bitmaps of type B on 1762 * inactivation. But in this case we lose bitmaps in valid 4.2 case. 1763 * 1764 * On the other hand, resuming source vm, if disk was already changed 1765 * is a bad thing anyway: not only bitmaps, the whole vm state is 1766 * out of sync with disk. 1767 * 1768 * This means, that user or management tool, who for some reason 1769 * decided to resume source vm, after disk was already changed by 1770 * target vm, should at least drop all dirty bitmaps by hand. 1771 * 1772 * So, we can ignore this case for now, but TODO: "generation" 1773 * extension for qcow2, to determine, that image was changed after 1774 * last inactivation. And if it is changed, we will drop (or at least 1775 * mark as 'invalid' all the bitmaps of type B, both persistent 1776 * and not). 1777 * 1778 * 3.2 disk was _not_ changed while were inactive. Bitmaps may be saved 1779 * to disk ('dirty-bitmaps' capability disabled), or not saved 1780 * ('dirty-bitmaps' capability enabled), but we don't need to care 1781 * of: let's load bitmaps as always: stored bitmaps will be loaded, 1782 * and not stored has flag IN_USE=1 in the image and will be skipped 1783 * on loading. 1784 * 1785 * One remaining possible case when we don't want load bitmaps: 1786 * 1787 * 4. Open disk in inactive mode in target vm (bitmaps are migrating or 1788 * will be loaded on invalidation, no needs try loading them before) 1789 */ 1790 1791 if (!(bdrv_get_flags(bs) & BDRV_O_INACTIVE)) { 1792 /* It's case 1, 2 or 3.2. Or 3.1 which is BUG in management layer. */ 1793 bool header_updated; 1794 if (!qcow2_load_dirty_bitmaps(bs, &header_updated, errp)) { 1795 ret = -EINVAL; 1796 goto fail; 1797 } 1798 1799 update_header = update_header && !header_updated; 1800 } 1801 1802 if (update_header) { 1803 ret = qcow2_update_header(bs); 1804 if (ret < 0) { 1805 error_setg_errno(errp, -ret, "Could not update qcow2 header"); 1806 goto fail; 1807 } 1808 } 1809 1810 bs->supported_zero_flags = header.version >= 3 ? 1811 BDRV_REQ_MAY_UNMAP | BDRV_REQ_NO_FALLBACK : 0; 1812 bs->supported_truncate_flags = BDRV_REQ_ZERO_WRITE; 1813 1814 /* Repair image if dirty */ 1815 if (!(flags & BDRV_O_CHECK) && bdrv_is_writable(bs) && 1816 (s->incompatible_features & QCOW2_INCOMPAT_DIRTY)) { 1817 BdrvCheckResult result = {0}; 1818 1819 ret = qcow2_co_check_locked(bs, &result, 1820 BDRV_FIX_ERRORS | BDRV_FIX_LEAKS); 1821 if (ret < 0 || result.check_errors) { 1822 if (ret >= 0) { 1823 ret = -EIO; 1824 } 1825 error_setg_errno(errp, -ret, "Could not repair dirty image"); 1826 goto fail; 1827 } 1828 } 1829 1830 #ifdef DEBUG_ALLOC 1831 { 1832 BdrvCheckResult result = {0}; 1833 qcow2_check_refcounts(bs, &result, 0); 1834 } 1835 #endif 1836 1837 qemu_co_queue_init(&s->thread_task_queue); 1838 1839 return ret; 1840 1841 fail: 1842 g_free(s->image_data_file); 1843 if (open_data_file && has_data_file(bs)) { 1844 bdrv_unref_child(bs, s->data_file); 1845 s->data_file = NULL; 1846 } 1847 g_free(s->unknown_header_fields); 1848 cleanup_unknown_header_ext(bs); 1849 qcow2_free_snapshots(bs); 1850 qcow2_refcount_close(bs); 1851 qemu_vfree(s->l1_table); 1852 /* else pre-write overlap checks in cache_destroy may crash */ 1853 s->l1_table = NULL; 1854 cache_clean_timer_del(bs); 1855 if (s->l2_table_cache) { 1856 qcow2_cache_destroy(s->l2_table_cache); 1857 } 1858 if (s->refcount_block_cache) { 1859 qcow2_cache_destroy(s->refcount_block_cache); 1860 } 1861 qcrypto_block_free(s->crypto); 1862 qapi_free_QCryptoBlockOpenOptions(s->crypto_opts); 1863 return ret; 1864 } 1865 1866 typedef struct QCow2OpenCo { 1867 BlockDriverState *bs; 1868 QDict *options; 1869 int flags; 1870 Error **errp; 1871 int ret; 1872 } QCow2OpenCo; 1873 1874 static void coroutine_fn qcow2_open_entry(void *opaque) 1875 { 1876 QCow2OpenCo *qoc = opaque; 1877 BDRVQcow2State *s = qoc->bs->opaque; 1878 1879 qemu_co_mutex_lock(&s->lock); 1880 qoc->ret = qcow2_do_open(qoc->bs, qoc->options, qoc->flags, true, 1881 qoc->errp); 1882 qemu_co_mutex_unlock(&s->lock); 1883 } 1884 1885 static int qcow2_open(BlockDriverState *bs, QDict *options, int flags, 1886 Error **errp) 1887 { 1888 BDRVQcow2State *s = bs->opaque; 1889 QCow2OpenCo qoc = { 1890 .bs = bs, 1891 .options = options, 1892 .flags = flags, 1893 .errp = errp, 1894 .ret = -EINPROGRESS 1895 }; 1896 1897 bs->file = bdrv_open_child(NULL, options, "file", bs, &child_of_bds, 1898 BDRV_CHILD_IMAGE, false, errp); 1899 if (!bs->file) { 1900 return -EINVAL; 1901 } 1902 1903 /* Initialise locks */ 1904 qemu_co_mutex_init(&s->lock); 1905 1906 if (qemu_in_coroutine()) { 1907 /* From bdrv_co_create. */ 1908 qcow2_open_entry(&qoc); 1909 } else { 1910 assert(qemu_get_current_aio_context() == qemu_get_aio_context()); 1911 qemu_coroutine_enter(qemu_coroutine_create(qcow2_open_entry, &qoc)); 1912 BDRV_POLL_WHILE(bs, qoc.ret == -EINPROGRESS); 1913 } 1914 return qoc.ret; 1915 } 1916 1917 static void qcow2_refresh_limits(BlockDriverState *bs, Error **errp) 1918 { 1919 BDRVQcow2State *s = bs->opaque; 1920 1921 if (bs->encrypted) { 1922 /* Encryption works on a sector granularity */ 1923 bs->bl.request_alignment = qcrypto_block_get_sector_size(s->crypto); 1924 } 1925 bs->bl.pwrite_zeroes_alignment = s->subcluster_size; 1926 bs->bl.pdiscard_alignment = s->cluster_size; 1927 } 1928 1929 static int qcow2_reopen_prepare(BDRVReopenState *state, 1930 BlockReopenQueue *queue, Error **errp) 1931 { 1932 BDRVQcow2State *s = state->bs->opaque; 1933 Qcow2ReopenState *r; 1934 int ret; 1935 1936 r = g_new0(Qcow2ReopenState, 1); 1937 state->opaque = r; 1938 1939 ret = qcow2_update_options_prepare(state->bs, r, state->options, 1940 state->flags, errp); 1941 if (ret < 0) { 1942 goto fail; 1943 } 1944 1945 /* We need to write out any unwritten data if we reopen read-only. */ 1946 if ((state->flags & BDRV_O_RDWR) == 0) { 1947 ret = qcow2_reopen_bitmaps_ro(state->bs, errp); 1948 if (ret < 0) { 1949 goto fail; 1950 } 1951 1952 ret = bdrv_flush(state->bs); 1953 if (ret < 0) { 1954 goto fail; 1955 } 1956 1957 ret = qcow2_mark_clean(state->bs); 1958 if (ret < 0) { 1959 goto fail; 1960 } 1961 } 1962 1963 /* 1964 * Without an external data file, s->data_file points to the same BdrvChild 1965 * as bs->file. It needs to be resynced after reopen because bs->file may 1966 * be changed. We can't use it in the meantime. 1967 */ 1968 if (!has_data_file(state->bs)) { 1969 assert(s->data_file == state->bs->file); 1970 s->data_file = NULL; 1971 } 1972 1973 return 0; 1974 1975 fail: 1976 qcow2_update_options_abort(state->bs, r); 1977 g_free(r); 1978 return ret; 1979 } 1980 1981 static void qcow2_reopen_commit(BDRVReopenState *state) 1982 { 1983 BDRVQcow2State *s = state->bs->opaque; 1984 1985 qcow2_update_options_commit(state->bs, state->opaque); 1986 if (!s->data_file) { 1987 /* 1988 * If we don't have an external data file, s->data_file was cleared by 1989 * qcow2_reopen_prepare() and needs to be updated. 1990 */ 1991 s->data_file = state->bs->file; 1992 } 1993 g_free(state->opaque); 1994 } 1995 1996 static void qcow2_reopen_commit_post(BDRVReopenState *state) 1997 { 1998 if (state->flags & BDRV_O_RDWR) { 1999 Error *local_err = NULL; 2000 2001 if (qcow2_reopen_bitmaps_rw(state->bs, &local_err) < 0) { 2002 /* 2003 * This is not fatal, bitmaps just left read-only, so all following 2004 * writes will fail. User can remove read-only bitmaps to unblock 2005 * writes or retry reopen. 2006 */ 2007 error_reportf_err(local_err, 2008 "%s: Failed to make dirty bitmaps writable: ", 2009 bdrv_get_node_name(state->bs)); 2010 } 2011 } 2012 } 2013 2014 static void qcow2_reopen_abort(BDRVReopenState *state) 2015 { 2016 BDRVQcow2State *s = state->bs->opaque; 2017 2018 if (!s->data_file) { 2019 /* 2020 * If we don't have an external data file, s->data_file was cleared by 2021 * qcow2_reopen_prepare() and needs to be restored. 2022 */ 2023 s->data_file = state->bs->file; 2024 } 2025 qcow2_update_options_abort(state->bs, state->opaque); 2026 g_free(state->opaque); 2027 } 2028 2029 static void qcow2_join_options(QDict *options, QDict *old_options) 2030 { 2031 bool has_new_overlap_template = 2032 qdict_haskey(options, QCOW2_OPT_OVERLAP) || 2033 qdict_haskey(options, QCOW2_OPT_OVERLAP_TEMPLATE); 2034 bool has_new_total_cache_size = 2035 qdict_haskey(options, QCOW2_OPT_CACHE_SIZE); 2036 bool has_all_cache_options; 2037 2038 /* New overlap template overrides all old overlap options */ 2039 if (has_new_overlap_template) { 2040 qdict_del(old_options, QCOW2_OPT_OVERLAP); 2041 qdict_del(old_options, QCOW2_OPT_OVERLAP_TEMPLATE); 2042 qdict_del(old_options, QCOW2_OPT_OVERLAP_MAIN_HEADER); 2043 qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L1); 2044 qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L2); 2045 qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_TABLE); 2046 qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK); 2047 qdict_del(old_options, QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE); 2048 qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L1); 2049 qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L2); 2050 } 2051 2052 /* New total cache size overrides all old options */ 2053 if (qdict_haskey(options, QCOW2_OPT_CACHE_SIZE)) { 2054 qdict_del(old_options, QCOW2_OPT_L2_CACHE_SIZE); 2055 qdict_del(old_options, QCOW2_OPT_REFCOUNT_CACHE_SIZE); 2056 } 2057 2058 qdict_join(options, old_options, false); 2059 2060 /* 2061 * If after merging all cache size options are set, an old total size is 2062 * overwritten. Do keep all options, however, if all three are new. The 2063 * resulting error message is what we want to happen. 2064 */ 2065 has_all_cache_options = 2066 qdict_haskey(options, QCOW2_OPT_CACHE_SIZE) || 2067 qdict_haskey(options, QCOW2_OPT_L2_CACHE_SIZE) || 2068 qdict_haskey(options, QCOW2_OPT_REFCOUNT_CACHE_SIZE); 2069 2070 if (has_all_cache_options && !has_new_total_cache_size) { 2071 qdict_del(options, QCOW2_OPT_CACHE_SIZE); 2072 } 2073 } 2074 2075 static int coroutine_fn qcow2_co_block_status(BlockDriverState *bs, 2076 bool want_zero, 2077 int64_t offset, int64_t count, 2078 int64_t *pnum, int64_t *map, 2079 BlockDriverState **file) 2080 { 2081 BDRVQcow2State *s = bs->opaque; 2082 uint64_t host_offset; 2083 unsigned int bytes; 2084 QCow2SubclusterType type; 2085 int ret, status = 0; 2086 2087 qemu_co_mutex_lock(&s->lock); 2088 2089 if (!s->metadata_preallocation_checked) { 2090 ret = qcow2_detect_metadata_preallocation(bs); 2091 s->metadata_preallocation = (ret == 1); 2092 s->metadata_preallocation_checked = true; 2093 } 2094 2095 bytes = MIN(INT_MAX, count); 2096 ret = qcow2_get_host_offset(bs, offset, &bytes, &host_offset, &type); 2097 qemu_co_mutex_unlock(&s->lock); 2098 if (ret < 0) { 2099 return ret; 2100 } 2101 2102 *pnum = bytes; 2103 2104 if ((type == QCOW2_SUBCLUSTER_NORMAL || 2105 type == QCOW2_SUBCLUSTER_ZERO_ALLOC || 2106 type == QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC) && !s->crypto) { 2107 *map = host_offset; 2108 *file = s->data_file->bs; 2109 status |= BDRV_BLOCK_OFFSET_VALID; 2110 } 2111 if (type == QCOW2_SUBCLUSTER_ZERO_PLAIN || 2112 type == QCOW2_SUBCLUSTER_ZERO_ALLOC) { 2113 status |= BDRV_BLOCK_ZERO; 2114 } else if (type != QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN && 2115 type != QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC) { 2116 status |= BDRV_BLOCK_DATA; 2117 } 2118 if (s->metadata_preallocation && (status & BDRV_BLOCK_DATA) && 2119 (status & BDRV_BLOCK_OFFSET_VALID)) 2120 { 2121 status |= BDRV_BLOCK_RECURSE; 2122 } 2123 return status; 2124 } 2125 2126 static coroutine_fn int qcow2_handle_l2meta(BlockDriverState *bs, 2127 QCowL2Meta **pl2meta, 2128 bool link_l2) 2129 { 2130 int ret = 0; 2131 QCowL2Meta *l2meta = *pl2meta; 2132 2133 while (l2meta != NULL) { 2134 QCowL2Meta *next; 2135 2136 if (link_l2) { 2137 ret = qcow2_alloc_cluster_link_l2(bs, l2meta); 2138 if (ret) { 2139 goto out; 2140 } 2141 } else { 2142 qcow2_alloc_cluster_abort(bs, l2meta); 2143 } 2144 2145 /* Take the request off the list of running requests */ 2146 QLIST_REMOVE(l2meta, next_in_flight); 2147 2148 qemu_co_queue_restart_all(&l2meta->dependent_requests); 2149 2150 next = l2meta->next; 2151 g_free(l2meta); 2152 l2meta = next; 2153 } 2154 out: 2155 *pl2meta = l2meta; 2156 return ret; 2157 } 2158 2159 static coroutine_fn int 2160 qcow2_co_preadv_encrypted(BlockDriverState *bs, 2161 uint64_t host_offset, 2162 uint64_t offset, 2163 uint64_t bytes, 2164 QEMUIOVector *qiov, 2165 uint64_t qiov_offset) 2166 { 2167 int ret; 2168 BDRVQcow2State *s = bs->opaque; 2169 uint8_t *buf; 2170 2171 assert(bs->encrypted && s->crypto); 2172 assert(bytes <= QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size); 2173 2174 /* 2175 * For encrypted images, read everything into a temporary 2176 * contiguous buffer on which the AES functions can work. 2177 * Also, decryption in a separate buffer is better as it 2178 * prevents the guest from learning information about the 2179 * encrypted nature of the virtual disk. 2180 */ 2181 2182 buf = qemu_try_blockalign(s->data_file->bs, bytes); 2183 if (buf == NULL) { 2184 return -ENOMEM; 2185 } 2186 2187 BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO); 2188 ret = bdrv_co_pread(s->data_file, host_offset, bytes, buf, 0); 2189 if (ret < 0) { 2190 goto fail; 2191 } 2192 2193 if (qcow2_co_decrypt(bs, host_offset, offset, buf, bytes) < 0) 2194 { 2195 ret = -EIO; 2196 goto fail; 2197 } 2198 qemu_iovec_from_buf(qiov, qiov_offset, buf, bytes); 2199 2200 fail: 2201 qemu_vfree(buf); 2202 2203 return ret; 2204 } 2205 2206 typedef struct Qcow2AioTask { 2207 AioTask task; 2208 2209 BlockDriverState *bs; 2210 QCow2SubclusterType subcluster_type; /* only for read */ 2211 uint64_t host_offset; /* or l2_entry for compressed read */ 2212 uint64_t offset; 2213 uint64_t bytes; 2214 QEMUIOVector *qiov; 2215 uint64_t qiov_offset; 2216 QCowL2Meta *l2meta; /* only for write */ 2217 } Qcow2AioTask; 2218 2219 static coroutine_fn int qcow2_co_preadv_task_entry(AioTask *task); 2220 static coroutine_fn int qcow2_add_task(BlockDriverState *bs, 2221 AioTaskPool *pool, 2222 AioTaskFunc func, 2223 QCow2SubclusterType subcluster_type, 2224 uint64_t host_offset, 2225 uint64_t offset, 2226 uint64_t bytes, 2227 QEMUIOVector *qiov, 2228 size_t qiov_offset, 2229 QCowL2Meta *l2meta) 2230 { 2231 Qcow2AioTask local_task; 2232 Qcow2AioTask *task = pool ? g_new(Qcow2AioTask, 1) : &local_task; 2233 2234 *task = (Qcow2AioTask) { 2235 .task.func = func, 2236 .bs = bs, 2237 .subcluster_type = subcluster_type, 2238 .qiov = qiov, 2239 .host_offset = host_offset, 2240 .offset = offset, 2241 .bytes = bytes, 2242 .qiov_offset = qiov_offset, 2243 .l2meta = l2meta, 2244 }; 2245 2246 trace_qcow2_add_task(qemu_coroutine_self(), bs, pool, 2247 func == qcow2_co_preadv_task_entry ? "read" : "write", 2248 subcluster_type, host_offset, offset, bytes, 2249 qiov, qiov_offset); 2250 2251 if (!pool) { 2252 return func(&task->task); 2253 } 2254 2255 aio_task_pool_start_task(pool, &task->task); 2256 2257 return 0; 2258 } 2259 2260 static coroutine_fn int qcow2_co_preadv_task(BlockDriverState *bs, 2261 QCow2SubclusterType subc_type, 2262 uint64_t host_offset, 2263 uint64_t offset, uint64_t bytes, 2264 QEMUIOVector *qiov, 2265 size_t qiov_offset) 2266 { 2267 BDRVQcow2State *s = bs->opaque; 2268 2269 switch (subc_type) { 2270 case QCOW2_SUBCLUSTER_ZERO_PLAIN: 2271 case QCOW2_SUBCLUSTER_ZERO_ALLOC: 2272 /* Both zero types are handled in qcow2_co_preadv_part */ 2273 g_assert_not_reached(); 2274 2275 case QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN: 2276 case QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC: 2277 assert(bs->backing); /* otherwise handled in qcow2_co_preadv_part */ 2278 2279 BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO); 2280 return bdrv_co_preadv_part(bs->backing, offset, bytes, 2281 qiov, qiov_offset, 0); 2282 2283 case QCOW2_SUBCLUSTER_COMPRESSED: 2284 return qcow2_co_preadv_compressed(bs, host_offset, 2285 offset, bytes, qiov, qiov_offset); 2286 2287 case QCOW2_SUBCLUSTER_NORMAL: 2288 if (bs->encrypted) { 2289 return qcow2_co_preadv_encrypted(bs, host_offset, 2290 offset, bytes, qiov, qiov_offset); 2291 } 2292 2293 BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO); 2294 return bdrv_co_preadv_part(s->data_file, host_offset, 2295 bytes, qiov, qiov_offset, 0); 2296 2297 default: 2298 g_assert_not_reached(); 2299 } 2300 2301 g_assert_not_reached(); 2302 } 2303 2304 static coroutine_fn int qcow2_co_preadv_task_entry(AioTask *task) 2305 { 2306 Qcow2AioTask *t = container_of(task, Qcow2AioTask, task); 2307 2308 assert(!t->l2meta); 2309 2310 return qcow2_co_preadv_task(t->bs, t->subcluster_type, 2311 t->host_offset, t->offset, t->bytes, 2312 t->qiov, t->qiov_offset); 2313 } 2314 2315 static coroutine_fn int qcow2_co_preadv_part(BlockDriverState *bs, 2316 int64_t offset, int64_t bytes, 2317 QEMUIOVector *qiov, 2318 size_t qiov_offset, 2319 BdrvRequestFlags flags) 2320 { 2321 BDRVQcow2State *s = bs->opaque; 2322 int ret = 0; 2323 unsigned int cur_bytes; /* number of bytes in current iteration */ 2324 uint64_t host_offset = 0; 2325 QCow2SubclusterType type; 2326 AioTaskPool *aio = NULL; 2327 2328 while (bytes != 0 && aio_task_pool_status(aio) == 0) { 2329 /* prepare next request */ 2330 cur_bytes = MIN(bytes, INT_MAX); 2331 if (s->crypto) { 2332 cur_bytes = MIN(cur_bytes, 2333 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size); 2334 } 2335 2336 qemu_co_mutex_lock(&s->lock); 2337 ret = qcow2_get_host_offset(bs, offset, &cur_bytes, 2338 &host_offset, &type); 2339 qemu_co_mutex_unlock(&s->lock); 2340 if (ret < 0) { 2341 goto out; 2342 } 2343 2344 if (type == QCOW2_SUBCLUSTER_ZERO_PLAIN || 2345 type == QCOW2_SUBCLUSTER_ZERO_ALLOC || 2346 (type == QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN && !bs->backing) || 2347 (type == QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC && !bs->backing)) 2348 { 2349 qemu_iovec_memset(qiov, qiov_offset, 0, cur_bytes); 2350 } else { 2351 if (!aio && cur_bytes != bytes) { 2352 aio = aio_task_pool_new(QCOW2_MAX_WORKERS); 2353 } 2354 ret = qcow2_add_task(bs, aio, qcow2_co_preadv_task_entry, type, 2355 host_offset, offset, cur_bytes, 2356 qiov, qiov_offset, NULL); 2357 if (ret < 0) { 2358 goto out; 2359 } 2360 } 2361 2362 bytes -= cur_bytes; 2363 offset += cur_bytes; 2364 qiov_offset += cur_bytes; 2365 } 2366 2367 out: 2368 if (aio) { 2369 aio_task_pool_wait_all(aio); 2370 if (ret == 0) { 2371 ret = aio_task_pool_status(aio); 2372 } 2373 g_free(aio); 2374 } 2375 2376 return ret; 2377 } 2378 2379 /* Check if it's possible to merge a write request with the writing of 2380 * the data from the COW regions */ 2381 static bool merge_cow(uint64_t offset, unsigned bytes, 2382 QEMUIOVector *qiov, size_t qiov_offset, 2383 QCowL2Meta *l2meta) 2384 { 2385 QCowL2Meta *m; 2386 2387 for (m = l2meta; m != NULL; m = m->next) { 2388 /* If both COW regions are empty then there's nothing to merge */ 2389 if (m->cow_start.nb_bytes == 0 && m->cow_end.nb_bytes == 0) { 2390 continue; 2391 } 2392 2393 /* If COW regions are handled already, skip this too */ 2394 if (m->skip_cow) { 2395 continue; 2396 } 2397 2398 /* 2399 * The write request should start immediately after the first 2400 * COW region. This does not always happen because the area 2401 * touched by the request can be larger than the one defined 2402 * by @m (a single request can span an area consisting of a 2403 * mix of previously unallocated and allocated clusters, that 2404 * is why @l2meta is a list). 2405 */ 2406 if (l2meta_cow_start(m) + m->cow_start.nb_bytes != offset) { 2407 /* In this case the request starts before this region */ 2408 assert(offset < l2meta_cow_start(m)); 2409 assert(m->cow_start.nb_bytes == 0); 2410 continue; 2411 } 2412 2413 /* The write request should end immediately before the second 2414 * COW region (see above for why it does not always happen) */ 2415 if (m->offset + m->cow_end.offset != offset + bytes) { 2416 assert(offset + bytes > m->offset + m->cow_end.offset); 2417 assert(m->cow_end.nb_bytes == 0); 2418 continue; 2419 } 2420 2421 /* Make sure that adding both COW regions to the QEMUIOVector 2422 * does not exceed IOV_MAX */ 2423 if (qemu_iovec_subvec_niov(qiov, qiov_offset, bytes) > IOV_MAX - 2) { 2424 continue; 2425 } 2426 2427 m->data_qiov = qiov; 2428 m->data_qiov_offset = qiov_offset; 2429 return true; 2430 } 2431 2432 return false; 2433 } 2434 2435 /* 2436 * Return 1 if the COW regions read as zeroes, 0 if not, < 0 on error. 2437 * Note that returning 0 does not guarantee non-zero data. 2438 */ 2439 static int is_zero_cow(BlockDriverState *bs, QCowL2Meta *m) 2440 { 2441 /* 2442 * This check is designed for optimization shortcut so it must be 2443 * efficient. 2444 * Instead of is_zero(), use bdrv_co_is_zero_fast() as it is 2445 * faster (but not as accurate and can result in false negatives). 2446 */ 2447 int ret = bdrv_co_is_zero_fast(bs, m->offset + m->cow_start.offset, 2448 m->cow_start.nb_bytes); 2449 if (ret <= 0) { 2450 return ret; 2451 } 2452 2453 return bdrv_co_is_zero_fast(bs, m->offset + m->cow_end.offset, 2454 m->cow_end.nb_bytes); 2455 } 2456 2457 static int handle_alloc_space(BlockDriverState *bs, QCowL2Meta *l2meta) 2458 { 2459 BDRVQcow2State *s = bs->opaque; 2460 QCowL2Meta *m; 2461 2462 if (!(s->data_file->bs->supported_zero_flags & BDRV_REQ_NO_FALLBACK)) { 2463 return 0; 2464 } 2465 2466 if (bs->encrypted) { 2467 return 0; 2468 } 2469 2470 for (m = l2meta; m != NULL; m = m->next) { 2471 int ret; 2472 uint64_t start_offset = m->alloc_offset + m->cow_start.offset; 2473 unsigned nb_bytes = m->cow_end.offset + m->cow_end.nb_bytes - 2474 m->cow_start.offset; 2475 2476 if (!m->cow_start.nb_bytes && !m->cow_end.nb_bytes) { 2477 continue; 2478 } 2479 2480 ret = is_zero_cow(bs, m); 2481 if (ret < 0) { 2482 return ret; 2483 } else if (ret == 0) { 2484 continue; 2485 } 2486 2487 /* 2488 * instead of writing zero COW buffers, 2489 * efficiently zero out the whole clusters 2490 */ 2491 2492 ret = qcow2_pre_write_overlap_check(bs, 0, start_offset, nb_bytes, 2493 true); 2494 if (ret < 0) { 2495 return ret; 2496 } 2497 2498 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC_SPACE); 2499 ret = bdrv_co_pwrite_zeroes(s->data_file, start_offset, nb_bytes, 2500 BDRV_REQ_NO_FALLBACK); 2501 if (ret < 0) { 2502 if (ret != -ENOTSUP && ret != -EAGAIN) { 2503 return ret; 2504 } 2505 continue; 2506 } 2507 2508 trace_qcow2_skip_cow(qemu_coroutine_self(), m->offset, m->nb_clusters); 2509 m->skip_cow = true; 2510 } 2511 return 0; 2512 } 2513 2514 /* 2515 * qcow2_co_pwritev_task 2516 * Called with s->lock unlocked 2517 * l2meta - if not NULL, qcow2_co_pwritev_task() will consume it. Caller must 2518 * not use it somehow after qcow2_co_pwritev_task() call 2519 */ 2520 static coroutine_fn int qcow2_co_pwritev_task(BlockDriverState *bs, 2521 uint64_t host_offset, 2522 uint64_t offset, uint64_t bytes, 2523 QEMUIOVector *qiov, 2524 uint64_t qiov_offset, 2525 QCowL2Meta *l2meta) 2526 { 2527 int ret; 2528 BDRVQcow2State *s = bs->opaque; 2529 void *crypt_buf = NULL; 2530 QEMUIOVector encrypted_qiov; 2531 2532 if (bs->encrypted) { 2533 assert(s->crypto); 2534 assert(bytes <= QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size); 2535 crypt_buf = qemu_try_blockalign(bs->file->bs, bytes); 2536 if (crypt_buf == NULL) { 2537 ret = -ENOMEM; 2538 goto out_unlocked; 2539 } 2540 qemu_iovec_to_buf(qiov, qiov_offset, crypt_buf, bytes); 2541 2542 if (qcow2_co_encrypt(bs, host_offset, offset, crypt_buf, bytes) < 0) { 2543 ret = -EIO; 2544 goto out_unlocked; 2545 } 2546 2547 qemu_iovec_init_buf(&encrypted_qiov, crypt_buf, bytes); 2548 qiov = &encrypted_qiov; 2549 qiov_offset = 0; 2550 } 2551 2552 /* Try to efficiently initialize the physical space with zeroes */ 2553 ret = handle_alloc_space(bs, l2meta); 2554 if (ret < 0) { 2555 goto out_unlocked; 2556 } 2557 2558 /* 2559 * If we need to do COW, check if it's possible to merge the 2560 * writing of the guest data together with that of the COW regions. 2561 * If it's not possible (or not necessary) then write the 2562 * guest data now. 2563 */ 2564 if (!merge_cow(offset, bytes, qiov, qiov_offset, l2meta)) { 2565 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO); 2566 trace_qcow2_writev_data(qemu_coroutine_self(), host_offset); 2567 ret = bdrv_co_pwritev_part(s->data_file, host_offset, 2568 bytes, qiov, qiov_offset, 0); 2569 if (ret < 0) { 2570 goto out_unlocked; 2571 } 2572 } 2573 2574 qemu_co_mutex_lock(&s->lock); 2575 2576 ret = qcow2_handle_l2meta(bs, &l2meta, true); 2577 goto out_locked; 2578 2579 out_unlocked: 2580 qemu_co_mutex_lock(&s->lock); 2581 2582 out_locked: 2583 qcow2_handle_l2meta(bs, &l2meta, false); 2584 qemu_co_mutex_unlock(&s->lock); 2585 2586 qemu_vfree(crypt_buf); 2587 2588 return ret; 2589 } 2590 2591 static coroutine_fn int qcow2_co_pwritev_task_entry(AioTask *task) 2592 { 2593 Qcow2AioTask *t = container_of(task, Qcow2AioTask, task); 2594 2595 assert(!t->subcluster_type); 2596 2597 return qcow2_co_pwritev_task(t->bs, t->host_offset, 2598 t->offset, t->bytes, t->qiov, t->qiov_offset, 2599 t->l2meta); 2600 } 2601 2602 static coroutine_fn int qcow2_co_pwritev_part( 2603 BlockDriverState *bs, int64_t offset, int64_t bytes, 2604 QEMUIOVector *qiov, size_t qiov_offset, BdrvRequestFlags flags) 2605 { 2606 BDRVQcow2State *s = bs->opaque; 2607 int offset_in_cluster; 2608 int ret; 2609 unsigned int cur_bytes; /* number of sectors in current iteration */ 2610 uint64_t host_offset; 2611 QCowL2Meta *l2meta = NULL; 2612 AioTaskPool *aio = NULL; 2613 2614 trace_qcow2_writev_start_req(qemu_coroutine_self(), offset, bytes); 2615 2616 while (bytes != 0 && aio_task_pool_status(aio) == 0) { 2617 2618 l2meta = NULL; 2619 2620 trace_qcow2_writev_start_part(qemu_coroutine_self()); 2621 offset_in_cluster = offset_into_cluster(s, offset); 2622 cur_bytes = MIN(bytes, INT_MAX); 2623 if (bs->encrypted) { 2624 cur_bytes = MIN(cur_bytes, 2625 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size 2626 - offset_in_cluster); 2627 } 2628 2629 qemu_co_mutex_lock(&s->lock); 2630 2631 ret = qcow2_alloc_host_offset(bs, offset, &cur_bytes, 2632 &host_offset, &l2meta); 2633 if (ret < 0) { 2634 goto out_locked; 2635 } 2636 2637 ret = qcow2_pre_write_overlap_check(bs, 0, host_offset, 2638 cur_bytes, true); 2639 if (ret < 0) { 2640 goto out_locked; 2641 } 2642 2643 qemu_co_mutex_unlock(&s->lock); 2644 2645 if (!aio && cur_bytes != bytes) { 2646 aio = aio_task_pool_new(QCOW2_MAX_WORKERS); 2647 } 2648 ret = qcow2_add_task(bs, aio, qcow2_co_pwritev_task_entry, 0, 2649 host_offset, offset, 2650 cur_bytes, qiov, qiov_offset, l2meta); 2651 l2meta = NULL; /* l2meta is consumed by qcow2_co_pwritev_task() */ 2652 if (ret < 0) { 2653 goto fail_nometa; 2654 } 2655 2656 bytes -= cur_bytes; 2657 offset += cur_bytes; 2658 qiov_offset += cur_bytes; 2659 trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_bytes); 2660 } 2661 ret = 0; 2662 2663 qemu_co_mutex_lock(&s->lock); 2664 2665 out_locked: 2666 qcow2_handle_l2meta(bs, &l2meta, false); 2667 2668 qemu_co_mutex_unlock(&s->lock); 2669 2670 fail_nometa: 2671 if (aio) { 2672 aio_task_pool_wait_all(aio); 2673 if (ret == 0) { 2674 ret = aio_task_pool_status(aio); 2675 } 2676 g_free(aio); 2677 } 2678 2679 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret); 2680 2681 return ret; 2682 } 2683 2684 static int qcow2_inactivate(BlockDriverState *bs) 2685 { 2686 BDRVQcow2State *s = bs->opaque; 2687 int ret, result = 0; 2688 Error *local_err = NULL; 2689 2690 qcow2_store_persistent_dirty_bitmaps(bs, true, &local_err); 2691 if (local_err != NULL) { 2692 result = -EINVAL; 2693 error_reportf_err(local_err, "Lost persistent bitmaps during " 2694 "inactivation of node '%s': ", 2695 bdrv_get_device_or_node_name(bs)); 2696 } 2697 2698 ret = qcow2_cache_flush(bs, s->l2_table_cache); 2699 if (ret) { 2700 result = ret; 2701 error_report("Failed to flush the L2 table cache: %s", 2702 strerror(-ret)); 2703 } 2704 2705 ret = qcow2_cache_flush(bs, s->refcount_block_cache); 2706 if (ret) { 2707 result = ret; 2708 error_report("Failed to flush the refcount block cache: %s", 2709 strerror(-ret)); 2710 } 2711 2712 if (result == 0) { 2713 qcow2_mark_clean(bs); 2714 } 2715 2716 return result; 2717 } 2718 2719 static void qcow2_do_close(BlockDriverState *bs, bool close_data_file) 2720 { 2721 BDRVQcow2State *s = bs->opaque; 2722 qemu_vfree(s->l1_table); 2723 /* else pre-write overlap checks in cache_destroy may crash */ 2724 s->l1_table = NULL; 2725 2726 if (!(s->flags & BDRV_O_INACTIVE)) { 2727 qcow2_inactivate(bs); 2728 } 2729 2730 cache_clean_timer_del(bs); 2731 qcow2_cache_destroy(s->l2_table_cache); 2732 qcow2_cache_destroy(s->refcount_block_cache); 2733 2734 qcrypto_block_free(s->crypto); 2735 s->crypto = NULL; 2736 qapi_free_QCryptoBlockOpenOptions(s->crypto_opts); 2737 2738 g_free(s->unknown_header_fields); 2739 cleanup_unknown_header_ext(bs); 2740 2741 g_free(s->image_data_file); 2742 g_free(s->image_backing_file); 2743 g_free(s->image_backing_format); 2744 2745 if (close_data_file && has_data_file(bs)) { 2746 bdrv_unref_child(bs, s->data_file); 2747 s->data_file = NULL; 2748 } 2749 2750 qcow2_refcount_close(bs); 2751 qcow2_free_snapshots(bs); 2752 } 2753 2754 static void qcow2_close(BlockDriverState *bs) 2755 { 2756 qcow2_do_close(bs, true); 2757 } 2758 2759 static void coroutine_fn qcow2_co_invalidate_cache(BlockDriverState *bs, 2760 Error **errp) 2761 { 2762 ERRP_GUARD(); 2763 BDRVQcow2State *s = bs->opaque; 2764 BdrvChild *data_file; 2765 int flags = s->flags; 2766 QCryptoBlock *crypto = NULL; 2767 QDict *options; 2768 int ret; 2769 2770 /* 2771 * Backing files are read-only which makes all of their metadata immutable, 2772 * that means we don't have to worry about reopening them here. 2773 */ 2774 2775 crypto = s->crypto; 2776 s->crypto = NULL; 2777 2778 /* 2779 * Do not reopen s->data_file (i.e., have qcow2_do_close() not close it, 2780 * and then prevent qcow2_do_open() from opening it), because this function 2781 * runs in the I/O path and as such we must not invoke global-state 2782 * functions like bdrv_unref_child() and bdrv_open_child(). 2783 */ 2784 2785 qcow2_do_close(bs, false); 2786 2787 data_file = s->data_file; 2788 memset(s, 0, sizeof(BDRVQcow2State)); 2789 s->data_file = data_file; 2790 2791 options = qdict_clone_shallow(bs->options); 2792 2793 flags &= ~BDRV_O_INACTIVE; 2794 qemu_co_mutex_lock(&s->lock); 2795 ret = qcow2_do_open(bs, options, flags, false, errp); 2796 qemu_co_mutex_unlock(&s->lock); 2797 qobject_unref(options); 2798 if (ret < 0) { 2799 error_prepend(errp, "Could not reopen qcow2 layer: "); 2800 bs->drv = NULL; 2801 return; 2802 } 2803 2804 s->crypto = crypto; 2805 } 2806 2807 static size_t header_ext_add(char *buf, uint32_t magic, const void *s, 2808 size_t len, size_t buflen) 2809 { 2810 QCowExtension *ext_backing_fmt = (QCowExtension*) buf; 2811 size_t ext_len = sizeof(QCowExtension) + ((len + 7) & ~7); 2812 2813 if (buflen < ext_len) { 2814 return -ENOSPC; 2815 } 2816 2817 *ext_backing_fmt = (QCowExtension) { 2818 .magic = cpu_to_be32(magic), 2819 .len = cpu_to_be32(len), 2820 }; 2821 2822 if (len) { 2823 memcpy(buf + sizeof(QCowExtension), s, len); 2824 } 2825 2826 return ext_len; 2827 } 2828 2829 /* 2830 * Updates the qcow2 header, including the variable length parts of it, i.e. 2831 * the backing file name and all extensions. qcow2 was not designed to allow 2832 * such changes, so if we run out of space (we can only use the first cluster) 2833 * this function may fail. 2834 * 2835 * Returns 0 on success, -errno in error cases. 2836 */ 2837 int qcow2_update_header(BlockDriverState *bs) 2838 { 2839 BDRVQcow2State *s = bs->opaque; 2840 QCowHeader *header; 2841 char *buf; 2842 size_t buflen = s->cluster_size; 2843 int ret; 2844 uint64_t total_size; 2845 uint32_t refcount_table_clusters; 2846 size_t header_length; 2847 Qcow2UnknownHeaderExtension *uext; 2848 2849 buf = qemu_blockalign(bs, buflen); 2850 2851 /* Header structure */ 2852 header = (QCowHeader*) buf; 2853 2854 if (buflen < sizeof(*header)) { 2855 ret = -ENOSPC; 2856 goto fail; 2857 } 2858 2859 header_length = sizeof(*header) + s->unknown_header_fields_size; 2860 total_size = bs->total_sectors * BDRV_SECTOR_SIZE; 2861 refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3); 2862 2863 ret = validate_compression_type(s, NULL); 2864 if (ret) { 2865 goto fail; 2866 } 2867 2868 *header = (QCowHeader) { 2869 /* Version 2 fields */ 2870 .magic = cpu_to_be32(QCOW_MAGIC), 2871 .version = cpu_to_be32(s->qcow_version), 2872 .backing_file_offset = 0, 2873 .backing_file_size = 0, 2874 .cluster_bits = cpu_to_be32(s->cluster_bits), 2875 .size = cpu_to_be64(total_size), 2876 .crypt_method = cpu_to_be32(s->crypt_method_header), 2877 .l1_size = cpu_to_be32(s->l1_size), 2878 .l1_table_offset = cpu_to_be64(s->l1_table_offset), 2879 .refcount_table_offset = cpu_to_be64(s->refcount_table_offset), 2880 .refcount_table_clusters = cpu_to_be32(refcount_table_clusters), 2881 .nb_snapshots = cpu_to_be32(s->nb_snapshots), 2882 .snapshots_offset = cpu_to_be64(s->snapshots_offset), 2883 2884 /* Version 3 fields */ 2885 .incompatible_features = cpu_to_be64(s->incompatible_features), 2886 .compatible_features = cpu_to_be64(s->compatible_features), 2887 .autoclear_features = cpu_to_be64(s->autoclear_features), 2888 .refcount_order = cpu_to_be32(s->refcount_order), 2889 .header_length = cpu_to_be32(header_length), 2890 .compression_type = s->compression_type, 2891 }; 2892 2893 /* For older versions, write a shorter header */ 2894 switch (s->qcow_version) { 2895 case 2: 2896 ret = offsetof(QCowHeader, incompatible_features); 2897 break; 2898 case 3: 2899 ret = sizeof(*header); 2900 break; 2901 default: 2902 ret = -EINVAL; 2903 goto fail; 2904 } 2905 2906 buf += ret; 2907 buflen -= ret; 2908 memset(buf, 0, buflen); 2909 2910 /* Preserve any unknown field in the header */ 2911 if (s->unknown_header_fields_size) { 2912 if (buflen < s->unknown_header_fields_size) { 2913 ret = -ENOSPC; 2914 goto fail; 2915 } 2916 2917 memcpy(buf, s->unknown_header_fields, s->unknown_header_fields_size); 2918 buf += s->unknown_header_fields_size; 2919 buflen -= s->unknown_header_fields_size; 2920 } 2921 2922 /* Backing file format header extension */ 2923 if (s->image_backing_format) { 2924 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BACKING_FORMAT, 2925 s->image_backing_format, 2926 strlen(s->image_backing_format), 2927 buflen); 2928 if (ret < 0) { 2929 goto fail; 2930 } 2931 2932 buf += ret; 2933 buflen -= ret; 2934 } 2935 2936 /* External data file header extension */ 2937 if (has_data_file(bs) && s->image_data_file) { 2938 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_DATA_FILE, 2939 s->image_data_file, strlen(s->image_data_file), 2940 buflen); 2941 if (ret < 0) { 2942 goto fail; 2943 } 2944 2945 buf += ret; 2946 buflen -= ret; 2947 } 2948 2949 /* Full disk encryption header pointer extension */ 2950 if (s->crypto_header.offset != 0) { 2951 s->crypto_header.offset = cpu_to_be64(s->crypto_header.offset); 2952 s->crypto_header.length = cpu_to_be64(s->crypto_header.length); 2953 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_CRYPTO_HEADER, 2954 &s->crypto_header, sizeof(s->crypto_header), 2955 buflen); 2956 s->crypto_header.offset = be64_to_cpu(s->crypto_header.offset); 2957 s->crypto_header.length = be64_to_cpu(s->crypto_header.length); 2958 if (ret < 0) { 2959 goto fail; 2960 } 2961 buf += ret; 2962 buflen -= ret; 2963 } 2964 2965 /* 2966 * Feature table. A mere 8 feature names occupies 392 bytes, and 2967 * when coupled with the v3 minimum header of 104 bytes plus the 2968 * 8-byte end-of-extension marker, that would leave only 8 bytes 2969 * for a backing file name in an image with 512-byte clusters. 2970 * Thus, we choose to omit this header for cluster sizes 4k and 2971 * smaller. 2972 */ 2973 if (s->qcow_version >= 3 && s->cluster_size > 4096) { 2974 static const Qcow2Feature features[] = { 2975 { 2976 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE, 2977 .bit = QCOW2_INCOMPAT_DIRTY_BITNR, 2978 .name = "dirty bit", 2979 }, 2980 { 2981 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE, 2982 .bit = QCOW2_INCOMPAT_CORRUPT_BITNR, 2983 .name = "corrupt bit", 2984 }, 2985 { 2986 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE, 2987 .bit = QCOW2_INCOMPAT_DATA_FILE_BITNR, 2988 .name = "external data file", 2989 }, 2990 { 2991 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE, 2992 .bit = QCOW2_INCOMPAT_COMPRESSION_BITNR, 2993 .name = "compression type", 2994 }, 2995 { 2996 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE, 2997 .bit = QCOW2_INCOMPAT_EXTL2_BITNR, 2998 .name = "extended L2 entries", 2999 }, 3000 { 3001 .type = QCOW2_FEAT_TYPE_COMPATIBLE, 3002 .bit = QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR, 3003 .name = "lazy refcounts", 3004 }, 3005 { 3006 .type = QCOW2_FEAT_TYPE_AUTOCLEAR, 3007 .bit = QCOW2_AUTOCLEAR_BITMAPS_BITNR, 3008 .name = "bitmaps", 3009 }, 3010 { 3011 .type = QCOW2_FEAT_TYPE_AUTOCLEAR, 3012 .bit = QCOW2_AUTOCLEAR_DATA_FILE_RAW_BITNR, 3013 .name = "raw external data", 3014 }, 3015 }; 3016 3017 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_FEATURE_TABLE, 3018 features, sizeof(features), buflen); 3019 if (ret < 0) { 3020 goto fail; 3021 } 3022 buf += ret; 3023 buflen -= ret; 3024 } 3025 3026 /* Bitmap extension */ 3027 if (s->nb_bitmaps > 0) { 3028 Qcow2BitmapHeaderExt bitmaps_header = { 3029 .nb_bitmaps = cpu_to_be32(s->nb_bitmaps), 3030 .bitmap_directory_size = 3031 cpu_to_be64(s->bitmap_directory_size), 3032 .bitmap_directory_offset = 3033 cpu_to_be64(s->bitmap_directory_offset) 3034 }; 3035 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BITMAPS, 3036 &bitmaps_header, sizeof(bitmaps_header), 3037 buflen); 3038 if (ret < 0) { 3039 goto fail; 3040 } 3041 buf += ret; 3042 buflen -= ret; 3043 } 3044 3045 /* Keep unknown header extensions */ 3046 QLIST_FOREACH(uext, &s->unknown_header_ext, next) { 3047 ret = header_ext_add(buf, uext->magic, uext->data, uext->len, buflen); 3048 if (ret < 0) { 3049 goto fail; 3050 } 3051 3052 buf += ret; 3053 buflen -= ret; 3054 } 3055 3056 /* End of header extensions */ 3057 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_END, NULL, 0, buflen); 3058 if (ret < 0) { 3059 goto fail; 3060 } 3061 3062 buf += ret; 3063 buflen -= ret; 3064 3065 /* Backing file name */ 3066 if (s->image_backing_file) { 3067 size_t backing_file_len = strlen(s->image_backing_file); 3068 3069 if (buflen < backing_file_len) { 3070 ret = -ENOSPC; 3071 goto fail; 3072 } 3073 3074 /* Using strncpy is ok here, since buf is not NUL-terminated. */ 3075 strncpy(buf, s->image_backing_file, buflen); 3076 3077 header->backing_file_offset = cpu_to_be64(buf - ((char*) header)); 3078 header->backing_file_size = cpu_to_be32(backing_file_len); 3079 } 3080 3081 /* Write the new header */ 3082 ret = bdrv_pwrite(bs->file, 0, s->cluster_size, header, 0); 3083 if (ret < 0) { 3084 goto fail; 3085 } 3086 3087 ret = 0; 3088 fail: 3089 qemu_vfree(header); 3090 return ret; 3091 } 3092 3093 static int qcow2_change_backing_file(BlockDriverState *bs, 3094 const char *backing_file, const char *backing_fmt) 3095 { 3096 BDRVQcow2State *s = bs->opaque; 3097 3098 /* Adding a backing file means that the external data file alone won't be 3099 * enough to make sense of the content */ 3100 if (backing_file && data_file_is_raw(bs)) { 3101 return -EINVAL; 3102 } 3103 3104 if (backing_file && strlen(backing_file) > 1023) { 3105 return -EINVAL; 3106 } 3107 3108 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file), 3109 backing_file ?: ""); 3110 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: ""); 3111 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: ""); 3112 3113 g_free(s->image_backing_file); 3114 g_free(s->image_backing_format); 3115 3116 s->image_backing_file = backing_file ? g_strdup(bs->backing_file) : NULL; 3117 s->image_backing_format = backing_fmt ? g_strdup(bs->backing_format) : NULL; 3118 3119 return qcow2_update_header(bs); 3120 } 3121 3122 static int qcow2_set_up_encryption(BlockDriverState *bs, 3123 QCryptoBlockCreateOptions *cryptoopts, 3124 Error **errp) 3125 { 3126 BDRVQcow2State *s = bs->opaque; 3127 QCryptoBlock *crypto = NULL; 3128 int fmt, ret; 3129 3130 switch (cryptoopts->format) { 3131 case Q_CRYPTO_BLOCK_FORMAT_LUKS: 3132 fmt = QCOW_CRYPT_LUKS; 3133 break; 3134 case Q_CRYPTO_BLOCK_FORMAT_QCOW: 3135 fmt = QCOW_CRYPT_AES; 3136 break; 3137 default: 3138 error_setg(errp, "Crypto format not supported in qcow2"); 3139 return -EINVAL; 3140 } 3141 3142 s->crypt_method_header = fmt; 3143 3144 crypto = qcrypto_block_create(cryptoopts, "encrypt.", 3145 qcow2_crypto_hdr_init_func, 3146 qcow2_crypto_hdr_write_func, 3147 bs, errp); 3148 if (!crypto) { 3149 return -EINVAL; 3150 } 3151 3152 ret = qcow2_update_header(bs); 3153 if (ret < 0) { 3154 error_setg_errno(errp, -ret, "Could not write encryption header"); 3155 goto out; 3156 } 3157 3158 ret = 0; 3159 out: 3160 qcrypto_block_free(crypto); 3161 return ret; 3162 } 3163 3164 /** 3165 * Preallocates metadata structures for data clusters between @offset (in the 3166 * guest disk) and @new_length (which is thus generally the new guest disk 3167 * size). 3168 * 3169 * Returns: 0 on success, -errno on failure. 3170 */ 3171 static int coroutine_fn preallocate_co(BlockDriverState *bs, uint64_t offset, 3172 uint64_t new_length, PreallocMode mode, 3173 Error **errp) 3174 { 3175 BDRVQcow2State *s = bs->opaque; 3176 uint64_t bytes; 3177 uint64_t host_offset = 0; 3178 int64_t file_length; 3179 unsigned int cur_bytes; 3180 int ret; 3181 QCowL2Meta *meta = NULL, *m; 3182 3183 assert(offset <= new_length); 3184 bytes = new_length - offset; 3185 3186 while (bytes) { 3187 cur_bytes = MIN(bytes, QEMU_ALIGN_DOWN(INT_MAX, s->cluster_size)); 3188 ret = qcow2_alloc_host_offset(bs, offset, &cur_bytes, 3189 &host_offset, &meta); 3190 if (ret < 0) { 3191 error_setg_errno(errp, -ret, "Allocating clusters failed"); 3192 goto out; 3193 } 3194 3195 for (m = meta; m != NULL; m = m->next) { 3196 m->prealloc = true; 3197 } 3198 3199 ret = qcow2_handle_l2meta(bs, &meta, true); 3200 if (ret < 0) { 3201 error_setg_errno(errp, -ret, "Mapping clusters failed"); 3202 goto out; 3203 } 3204 3205 /* TODO Preallocate data if requested */ 3206 3207 bytes -= cur_bytes; 3208 offset += cur_bytes; 3209 } 3210 3211 /* 3212 * It is expected that the image file is large enough to actually contain 3213 * all of the allocated clusters (otherwise we get failing reads after 3214 * EOF). Extend the image to the last allocated sector. 3215 */ 3216 file_length = bdrv_getlength(s->data_file->bs); 3217 if (file_length < 0) { 3218 error_setg_errno(errp, -file_length, "Could not get file size"); 3219 ret = file_length; 3220 goto out; 3221 } 3222 3223 if (host_offset + cur_bytes > file_length) { 3224 if (mode == PREALLOC_MODE_METADATA) { 3225 mode = PREALLOC_MODE_OFF; 3226 } 3227 ret = bdrv_co_truncate(s->data_file, host_offset + cur_bytes, false, 3228 mode, 0, errp); 3229 if (ret < 0) { 3230 goto out; 3231 } 3232 } 3233 3234 ret = 0; 3235 3236 out: 3237 qcow2_handle_l2meta(bs, &meta, false); 3238 return ret; 3239 } 3240 3241 /* qcow2_refcount_metadata_size: 3242 * @clusters: number of clusters to refcount (including data and L1/L2 tables) 3243 * @cluster_size: size of a cluster, in bytes 3244 * @refcount_order: refcount bits power-of-2 exponent 3245 * @generous_increase: allow for the refcount table to be 1.5x as large as it 3246 * needs to be 3247 * 3248 * Returns: Number of bytes required for refcount blocks and table metadata. 3249 */ 3250 int64_t qcow2_refcount_metadata_size(int64_t clusters, size_t cluster_size, 3251 int refcount_order, bool generous_increase, 3252 uint64_t *refblock_count) 3253 { 3254 /* 3255 * Every host cluster is reference-counted, including metadata (even 3256 * refcount metadata is recursively included). 3257 * 3258 * An accurate formula for the size of refcount metadata size is difficult 3259 * to derive. An easier method of calculation is finding the fixed point 3260 * where no further refcount blocks or table clusters are required to 3261 * reference count every cluster. 3262 */ 3263 int64_t blocks_per_table_cluster = cluster_size / REFTABLE_ENTRY_SIZE; 3264 int64_t refcounts_per_block = cluster_size * 8 / (1 << refcount_order); 3265 int64_t table = 0; /* number of refcount table clusters */ 3266 int64_t blocks = 0; /* number of refcount block clusters */ 3267 int64_t last; 3268 int64_t n = 0; 3269 3270 do { 3271 last = n; 3272 blocks = DIV_ROUND_UP(clusters + table + blocks, refcounts_per_block); 3273 table = DIV_ROUND_UP(blocks, blocks_per_table_cluster); 3274 n = clusters + blocks + table; 3275 3276 if (n == last && generous_increase) { 3277 clusters += DIV_ROUND_UP(table, 2); 3278 n = 0; /* force another loop */ 3279 generous_increase = false; 3280 } 3281 } while (n != last); 3282 3283 if (refblock_count) { 3284 *refblock_count = blocks; 3285 } 3286 3287 return (blocks + table) * cluster_size; 3288 } 3289 3290 /** 3291 * qcow2_calc_prealloc_size: 3292 * @total_size: virtual disk size in bytes 3293 * @cluster_size: cluster size in bytes 3294 * @refcount_order: refcount bits power-of-2 exponent 3295 * @extended_l2: true if the image has extended L2 entries 3296 * 3297 * Returns: Total number of bytes required for the fully allocated image 3298 * (including metadata). 3299 */ 3300 static int64_t qcow2_calc_prealloc_size(int64_t total_size, 3301 size_t cluster_size, 3302 int refcount_order, 3303 bool extended_l2) 3304 { 3305 int64_t meta_size = 0; 3306 uint64_t nl1e, nl2e; 3307 int64_t aligned_total_size = ROUND_UP(total_size, cluster_size); 3308 size_t l2e_size = extended_l2 ? L2E_SIZE_EXTENDED : L2E_SIZE_NORMAL; 3309 3310 /* header: 1 cluster */ 3311 meta_size += cluster_size; 3312 3313 /* total size of L2 tables */ 3314 nl2e = aligned_total_size / cluster_size; 3315 nl2e = ROUND_UP(nl2e, cluster_size / l2e_size); 3316 meta_size += nl2e * l2e_size; 3317 3318 /* total size of L1 tables */ 3319 nl1e = nl2e * l2e_size / cluster_size; 3320 nl1e = ROUND_UP(nl1e, cluster_size / L1E_SIZE); 3321 meta_size += nl1e * L1E_SIZE; 3322 3323 /* total size of refcount table and blocks */ 3324 meta_size += qcow2_refcount_metadata_size( 3325 (meta_size + aligned_total_size) / cluster_size, 3326 cluster_size, refcount_order, false, NULL); 3327 3328 return meta_size + aligned_total_size; 3329 } 3330 3331 static bool validate_cluster_size(size_t cluster_size, bool extended_l2, 3332 Error **errp) 3333 { 3334 int cluster_bits = ctz32(cluster_size); 3335 if (cluster_bits < MIN_CLUSTER_BITS || cluster_bits > MAX_CLUSTER_BITS || 3336 (1 << cluster_bits) != cluster_size) 3337 { 3338 error_setg(errp, "Cluster size must be a power of two between %d and " 3339 "%dk", 1 << MIN_CLUSTER_BITS, 1 << (MAX_CLUSTER_BITS - 10)); 3340 return false; 3341 } 3342 3343 if (extended_l2) { 3344 unsigned min_cluster_size = 3345 (1 << MIN_CLUSTER_BITS) * QCOW_EXTL2_SUBCLUSTERS_PER_CLUSTER; 3346 if (cluster_size < min_cluster_size) { 3347 error_setg(errp, "Extended L2 entries are only supported with " 3348 "cluster sizes of at least %u bytes", min_cluster_size); 3349 return false; 3350 } 3351 } 3352 3353 return true; 3354 } 3355 3356 static size_t qcow2_opt_get_cluster_size_del(QemuOpts *opts, bool extended_l2, 3357 Error **errp) 3358 { 3359 size_t cluster_size; 3360 3361 cluster_size = qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE, 3362 DEFAULT_CLUSTER_SIZE); 3363 if (!validate_cluster_size(cluster_size, extended_l2, errp)) { 3364 return 0; 3365 } 3366 return cluster_size; 3367 } 3368 3369 static int qcow2_opt_get_version_del(QemuOpts *opts, Error **errp) 3370 { 3371 char *buf; 3372 int ret; 3373 3374 buf = qemu_opt_get_del(opts, BLOCK_OPT_COMPAT_LEVEL); 3375 if (!buf) { 3376 ret = 3; /* default */ 3377 } else if (!strcmp(buf, "0.10")) { 3378 ret = 2; 3379 } else if (!strcmp(buf, "1.1")) { 3380 ret = 3; 3381 } else { 3382 error_setg(errp, "Invalid compatibility level: '%s'", buf); 3383 ret = -EINVAL; 3384 } 3385 g_free(buf); 3386 return ret; 3387 } 3388 3389 static uint64_t qcow2_opt_get_refcount_bits_del(QemuOpts *opts, int version, 3390 Error **errp) 3391 { 3392 uint64_t refcount_bits; 3393 3394 refcount_bits = qemu_opt_get_number_del(opts, BLOCK_OPT_REFCOUNT_BITS, 16); 3395 if (refcount_bits > 64 || !is_power_of_2(refcount_bits)) { 3396 error_setg(errp, "Refcount width must be a power of two and may not " 3397 "exceed 64 bits"); 3398 return 0; 3399 } 3400 3401 if (version < 3 && refcount_bits != 16) { 3402 error_setg(errp, "Different refcount widths than 16 bits require " 3403 "compatibility level 1.1 or above (use compat=1.1 or " 3404 "greater)"); 3405 return 0; 3406 } 3407 3408 return refcount_bits; 3409 } 3410 3411 static int coroutine_fn 3412 qcow2_co_create(BlockdevCreateOptions *create_options, Error **errp) 3413 { 3414 BlockdevCreateOptionsQcow2 *qcow2_opts; 3415 QDict *options; 3416 3417 /* 3418 * Open the image file and write a minimal qcow2 header. 3419 * 3420 * We keep things simple and start with a zero-sized image. We also 3421 * do without refcount blocks or a L1 table for now. We'll fix the 3422 * inconsistency later. 3423 * 3424 * We do need a refcount table because growing the refcount table means 3425 * allocating two new refcount blocks - the second of which would be at 3426 * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file 3427 * size for any qcow2 image. 3428 */ 3429 BlockBackend *blk = NULL; 3430 BlockDriverState *bs = NULL; 3431 BlockDriverState *data_bs = NULL; 3432 QCowHeader *header; 3433 size_t cluster_size; 3434 int version; 3435 int refcount_order; 3436 uint64_t *refcount_table; 3437 int ret; 3438 uint8_t compression_type = QCOW2_COMPRESSION_TYPE_ZLIB; 3439 3440 assert(create_options->driver == BLOCKDEV_DRIVER_QCOW2); 3441 qcow2_opts = &create_options->u.qcow2; 3442 3443 bs = bdrv_open_blockdev_ref(qcow2_opts->file, errp); 3444 if (bs == NULL) { 3445 return -EIO; 3446 } 3447 3448 /* Validate options and set default values */ 3449 if (!QEMU_IS_ALIGNED(qcow2_opts->size, BDRV_SECTOR_SIZE)) { 3450 error_setg(errp, "Image size must be a multiple of %u bytes", 3451 (unsigned) BDRV_SECTOR_SIZE); 3452 ret = -EINVAL; 3453 goto out; 3454 } 3455 3456 if (qcow2_opts->has_version) { 3457 switch (qcow2_opts->version) { 3458 case BLOCKDEV_QCOW2_VERSION_V2: 3459 version = 2; 3460 break; 3461 case BLOCKDEV_QCOW2_VERSION_V3: 3462 version = 3; 3463 break; 3464 default: 3465 g_assert_not_reached(); 3466 } 3467 } else { 3468 version = 3; 3469 } 3470 3471 if (qcow2_opts->has_cluster_size) { 3472 cluster_size = qcow2_opts->cluster_size; 3473 } else { 3474 cluster_size = DEFAULT_CLUSTER_SIZE; 3475 } 3476 3477 if (!qcow2_opts->has_extended_l2) { 3478 qcow2_opts->extended_l2 = false; 3479 } 3480 if (qcow2_opts->extended_l2) { 3481 if (version < 3) { 3482 error_setg(errp, "Extended L2 entries are only supported with " 3483 "compatibility level 1.1 and above (use version=v3 or " 3484 "greater)"); 3485 ret = -EINVAL; 3486 goto out; 3487 } 3488 } 3489 3490 if (!validate_cluster_size(cluster_size, qcow2_opts->extended_l2, errp)) { 3491 ret = -EINVAL; 3492 goto out; 3493 } 3494 3495 if (!qcow2_opts->has_preallocation) { 3496 qcow2_opts->preallocation = PREALLOC_MODE_OFF; 3497 } 3498 if (qcow2_opts->has_backing_file && 3499 qcow2_opts->preallocation != PREALLOC_MODE_OFF && 3500 !qcow2_opts->extended_l2) 3501 { 3502 error_setg(errp, "Backing file and preallocation can only be used at " 3503 "the same time if extended_l2 is on"); 3504 ret = -EINVAL; 3505 goto out; 3506 } 3507 if (qcow2_opts->has_backing_fmt && !qcow2_opts->has_backing_file) { 3508 error_setg(errp, "Backing format cannot be used without backing file"); 3509 ret = -EINVAL; 3510 goto out; 3511 } 3512 3513 if (!qcow2_opts->has_lazy_refcounts) { 3514 qcow2_opts->lazy_refcounts = false; 3515 } 3516 if (version < 3 && qcow2_opts->lazy_refcounts) { 3517 error_setg(errp, "Lazy refcounts only supported with compatibility " 3518 "level 1.1 and above (use version=v3 or greater)"); 3519 ret = -EINVAL; 3520 goto out; 3521 } 3522 3523 if (!qcow2_opts->has_refcount_bits) { 3524 qcow2_opts->refcount_bits = 16; 3525 } 3526 if (qcow2_opts->refcount_bits > 64 || 3527 !is_power_of_2(qcow2_opts->refcount_bits)) 3528 { 3529 error_setg(errp, "Refcount width must be a power of two and may not " 3530 "exceed 64 bits"); 3531 ret = -EINVAL; 3532 goto out; 3533 } 3534 if (version < 3 && qcow2_opts->refcount_bits != 16) { 3535 error_setg(errp, "Different refcount widths than 16 bits require " 3536 "compatibility level 1.1 or above (use version=v3 or " 3537 "greater)"); 3538 ret = -EINVAL; 3539 goto out; 3540 } 3541 refcount_order = ctz32(qcow2_opts->refcount_bits); 3542 3543 if (qcow2_opts->data_file_raw && !qcow2_opts->data_file) { 3544 error_setg(errp, "data-file-raw requires data-file"); 3545 ret = -EINVAL; 3546 goto out; 3547 } 3548 if (qcow2_opts->data_file_raw && qcow2_opts->has_backing_file) { 3549 error_setg(errp, "Backing file and data-file-raw cannot be used at " 3550 "the same time"); 3551 ret = -EINVAL; 3552 goto out; 3553 } 3554 if (qcow2_opts->data_file_raw && 3555 qcow2_opts->preallocation == PREALLOC_MODE_OFF) 3556 { 3557 /* 3558 * data-file-raw means that "the external data file can be 3559 * read as a consistent standalone raw image without looking 3560 * at the qcow2 metadata." It does not say that the metadata 3561 * must be ignored, though (and the qcow2 driver in fact does 3562 * not ignore it), so the L1/L2 tables must be present and 3563 * give a 1:1 mapping, so you get the same result regardless 3564 * of whether you look at the metadata or whether you ignore 3565 * it. 3566 */ 3567 qcow2_opts->preallocation = PREALLOC_MODE_METADATA; 3568 3569 /* 3570 * Cannot use preallocation with backing files, but giving a 3571 * backing file when specifying data_file_raw is an error 3572 * anyway. 3573 */ 3574 assert(!qcow2_opts->has_backing_file); 3575 } 3576 3577 if (qcow2_opts->data_file) { 3578 if (version < 3) { 3579 error_setg(errp, "External data files are only supported with " 3580 "compatibility level 1.1 and above (use version=v3 or " 3581 "greater)"); 3582 ret = -EINVAL; 3583 goto out; 3584 } 3585 data_bs = bdrv_open_blockdev_ref(qcow2_opts->data_file, errp); 3586 if (data_bs == NULL) { 3587 ret = -EIO; 3588 goto out; 3589 } 3590 } 3591 3592 if (qcow2_opts->has_compression_type && 3593 qcow2_opts->compression_type != QCOW2_COMPRESSION_TYPE_ZLIB) { 3594 3595 ret = -EINVAL; 3596 3597 if (version < 3) { 3598 error_setg(errp, "Non-zlib compression type is only supported with " 3599 "compatibility level 1.1 and above (use version=v3 or " 3600 "greater)"); 3601 goto out; 3602 } 3603 3604 switch (qcow2_opts->compression_type) { 3605 #ifdef CONFIG_ZSTD 3606 case QCOW2_COMPRESSION_TYPE_ZSTD: 3607 break; 3608 #endif 3609 default: 3610 error_setg(errp, "Unknown compression type"); 3611 goto out; 3612 } 3613 3614 compression_type = qcow2_opts->compression_type; 3615 } 3616 3617 /* Create BlockBackend to write to the image */ 3618 blk = blk_new_with_bs(bs, BLK_PERM_WRITE | BLK_PERM_RESIZE, BLK_PERM_ALL, 3619 errp); 3620 if (!blk) { 3621 ret = -EPERM; 3622 goto out; 3623 } 3624 blk_set_allow_write_beyond_eof(blk, true); 3625 3626 /* Write the header */ 3627 QEMU_BUILD_BUG_ON((1 << MIN_CLUSTER_BITS) < sizeof(*header)); 3628 header = g_malloc0(cluster_size); 3629 *header = (QCowHeader) { 3630 .magic = cpu_to_be32(QCOW_MAGIC), 3631 .version = cpu_to_be32(version), 3632 .cluster_bits = cpu_to_be32(ctz32(cluster_size)), 3633 .size = cpu_to_be64(0), 3634 .l1_table_offset = cpu_to_be64(0), 3635 .l1_size = cpu_to_be32(0), 3636 .refcount_table_offset = cpu_to_be64(cluster_size), 3637 .refcount_table_clusters = cpu_to_be32(1), 3638 .refcount_order = cpu_to_be32(refcount_order), 3639 /* don't deal with endianness since compression_type is 1 byte long */ 3640 .compression_type = compression_type, 3641 .header_length = cpu_to_be32(sizeof(*header)), 3642 }; 3643 3644 /* We'll update this to correct value later */ 3645 header->crypt_method = cpu_to_be32(QCOW_CRYPT_NONE); 3646 3647 if (qcow2_opts->lazy_refcounts) { 3648 header->compatible_features |= 3649 cpu_to_be64(QCOW2_COMPAT_LAZY_REFCOUNTS); 3650 } 3651 if (data_bs) { 3652 header->incompatible_features |= 3653 cpu_to_be64(QCOW2_INCOMPAT_DATA_FILE); 3654 } 3655 if (qcow2_opts->data_file_raw) { 3656 header->autoclear_features |= 3657 cpu_to_be64(QCOW2_AUTOCLEAR_DATA_FILE_RAW); 3658 } 3659 if (compression_type != QCOW2_COMPRESSION_TYPE_ZLIB) { 3660 header->incompatible_features |= 3661 cpu_to_be64(QCOW2_INCOMPAT_COMPRESSION); 3662 } 3663 3664 if (qcow2_opts->extended_l2) { 3665 header->incompatible_features |= 3666 cpu_to_be64(QCOW2_INCOMPAT_EXTL2); 3667 } 3668 3669 ret = blk_pwrite(blk, 0, cluster_size, header, 0); 3670 g_free(header); 3671 if (ret < 0) { 3672 error_setg_errno(errp, -ret, "Could not write qcow2 header"); 3673 goto out; 3674 } 3675 3676 /* Write a refcount table with one refcount block */ 3677 refcount_table = g_malloc0(2 * cluster_size); 3678 refcount_table[0] = cpu_to_be64(2 * cluster_size); 3679 ret = blk_pwrite(blk, cluster_size, 2 * cluster_size, refcount_table, 0); 3680 g_free(refcount_table); 3681 3682 if (ret < 0) { 3683 error_setg_errno(errp, -ret, "Could not write refcount table"); 3684 goto out; 3685 } 3686 3687 blk_unref(blk); 3688 blk = NULL; 3689 3690 /* 3691 * And now open the image and make it consistent first (i.e. increase the 3692 * refcount of the cluster that is occupied by the header and the refcount 3693 * table) 3694 */ 3695 options = qdict_new(); 3696 qdict_put_str(options, "driver", "qcow2"); 3697 qdict_put_str(options, "file", bs->node_name); 3698 if (data_bs) { 3699 qdict_put_str(options, "data-file", data_bs->node_name); 3700 } 3701 blk = blk_new_open(NULL, NULL, options, 3702 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_NO_FLUSH, 3703 errp); 3704 if (blk == NULL) { 3705 ret = -EIO; 3706 goto out; 3707 } 3708 3709 ret = qcow2_alloc_clusters(blk_bs(blk), 3 * cluster_size); 3710 if (ret < 0) { 3711 error_setg_errno(errp, -ret, "Could not allocate clusters for qcow2 " 3712 "header and refcount table"); 3713 goto out; 3714 3715 } else if (ret != 0) { 3716 error_report("Huh, first cluster in empty image is already in use?"); 3717 abort(); 3718 } 3719 3720 /* Set the external data file if necessary */ 3721 if (data_bs) { 3722 BDRVQcow2State *s = blk_bs(blk)->opaque; 3723 s->image_data_file = g_strdup(data_bs->filename); 3724 } 3725 3726 /* Create a full header (including things like feature table) */ 3727 ret = qcow2_update_header(blk_bs(blk)); 3728 if (ret < 0) { 3729 error_setg_errno(errp, -ret, "Could not update qcow2 header"); 3730 goto out; 3731 } 3732 3733 /* Okay, now that we have a valid image, let's give it the right size */ 3734 ret = blk_truncate(blk, qcow2_opts->size, false, qcow2_opts->preallocation, 3735 0, errp); 3736 if (ret < 0) { 3737 error_prepend(errp, "Could not resize image: "); 3738 goto out; 3739 } 3740 3741 /* Want a backing file? There you go. */ 3742 if (qcow2_opts->has_backing_file) { 3743 const char *backing_format = NULL; 3744 3745 if (qcow2_opts->has_backing_fmt) { 3746 backing_format = BlockdevDriver_str(qcow2_opts->backing_fmt); 3747 } 3748 3749 ret = bdrv_change_backing_file(blk_bs(blk), qcow2_opts->backing_file, 3750 backing_format, false); 3751 if (ret < 0) { 3752 error_setg_errno(errp, -ret, "Could not assign backing file '%s' " 3753 "with format '%s'", qcow2_opts->backing_file, 3754 backing_format); 3755 goto out; 3756 } 3757 } 3758 3759 /* Want encryption? There you go. */ 3760 if (qcow2_opts->has_encrypt) { 3761 ret = qcow2_set_up_encryption(blk_bs(blk), qcow2_opts->encrypt, errp); 3762 if (ret < 0) { 3763 goto out; 3764 } 3765 } 3766 3767 blk_unref(blk); 3768 blk = NULL; 3769 3770 /* Reopen the image without BDRV_O_NO_FLUSH to flush it before returning. 3771 * Using BDRV_O_NO_IO, since encryption is now setup we don't want to 3772 * have to setup decryption context. We're not doing any I/O on the top 3773 * level BlockDriverState, only lower layers, where BDRV_O_NO_IO does 3774 * not have effect. 3775 */ 3776 options = qdict_new(); 3777 qdict_put_str(options, "driver", "qcow2"); 3778 qdict_put_str(options, "file", bs->node_name); 3779 if (data_bs) { 3780 qdict_put_str(options, "data-file", data_bs->node_name); 3781 } 3782 blk = blk_new_open(NULL, NULL, options, 3783 BDRV_O_RDWR | BDRV_O_NO_BACKING | BDRV_O_NO_IO, 3784 errp); 3785 if (blk == NULL) { 3786 ret = -EIO; 3787 goto out; 3788 } 3789 3790 ret = 0; 3791 out: 3792 blk_unref(blk); 3793 bdrv_unref(bs); 3794 bdrv_unref(data_bs); 3795 return ret; 3796 } 3797 3798 static int coroutine_fn qcow2_co_create_opts(BlockDriver *drv, 3799 const char *filename, 3800 QemuOpts *opts, 3801 Error **errp) 3802 { 3803 BlockdevCreateOptions *create_options = NULL; 3804 QDict *qdict; 3805 Visitor *v; 3806 BlockDriverState *bs = NULL; 3807 BlockDriverState *data_bs = NULL; 3808 const char *val; 3809 int ret; 3810 3811 /* Only the keyval visitor supports the dotted syntax needed for 3812 * encryption, so go through a QDict before getting a QAPI type. Ignore 3813 * options meant for the protocol layer so that the visitor doesn't 3814 * complain. */ 3815 qdict = qemu_opts_to_qdict_filtered(opts, NULL, bdrv_qcow2.create_opts, 3816 true); 3817 3818 /* Handle encryption options */ 3819 val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT); 3820 if (val && !strcmp(val, "on")) { 3821 qdict_put_str(qdict, BLOCK_OPT_ENCRYPT, "qcow"); 3822 } else if (val && !strcmp(val, "off")) { 3823 qdict_del(qdict, BLOCK_OPT_ENCRYPT); 3824 } 3825 3826 val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT); 3827 if (val && !strcmp(val, "aes")) { 3828 qdict_put_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT, "qcow"); 3829 } 3830 3831 /* Convert compat=0.10/1.1 into compat=v2/v3, to be renamed into 3832 * version=v2/v3 below. */ 3833 val = qdict_get_try_str(qdict, BLOCK_OPT_COMPAT_LEVEL); 3834 if (val && !strcmp(val, "0.10")) { 3835 qdict_put_str(qdict, BLOCK_OPT_COMPAT_LEVEL, "v2"); 3836 } else if (val && !strcmp(val, "1.1")) { 3837 qdict_put_str(qdict, BLOCK_OPT_COMPAT_LEVEL, "v3"); 3838 } 3839 3840 /* Change legacy command line options into QMP ones */ 3841 static const QDictRenames opt_renames[] = { 3842 { BLOCK_OPT_BACKING_FILE, "backing-file" }, 3843 { BLOCK_OPT_BACKING_FMT, "backing-fmt" }, 3844 { BLOCK_OPT_CLUSTER_SIZE, "cluster-size" }, 3845 { BLOCK_OPT_LAZY_REFCOUNTS, "lazy-refcounts" }, 3846 { BLOCK_OPT_EXTL2, "extended-l2" }, 3847 { BLOCK_OPT_REFCOUNT_BITS, "refcount-bits" }, 3848 { BLOCK_OPT_ENCRYPT, BLOCK_OPT_ENCRYPT_FORMAT }, 3849 { BLOCK_OPT_COMPAT_LEVEL, "version" }, 3850 { BLOCK_OPT_DATA_FILE_RAW, "data-file-raw" }, 3851 { BLOCK_OPT_COMPRESSION_TYPE, "compression-type" }, 3852 { NULL, NULL }, 3853 }; 3854 3855 if (!qdict_rename_keys(qdict, opt_renames, errp)) { 3856 ret = -EINVAL; 3857 goto finish; 3858 } 3859 3860 /* Create and open the file (protocol layer) */ 3861 ret = bdrv_create_file(filename, opts, errp); 3862 if (ret < 0) { 3863 goto finish; 3864 } 3865 3866 bs = bdrv_open(filename, NULL, NULL, 3867 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp); 3868 if (bs == NULL) { 3869 ret = -EIO; 3870 goto finish; 3871 } 3872 3873 /* Create and open an external data file (protocol layer) */ 3874 val = qdict_get_try_str(qdict, BLOCK_OPT_DATA_FILE); 3875 if (val) { 3876 ret = bdrv_create_file(val, opts, errp); 3877 if (ret < 0) { 3878 goto finish; 3879 } 3880 3881 data_bs = bdrv_open(val, NULL, NULL, 3882 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, 3883 errp); 3884 if (data_bs == NULL) { 3885 ret = -EIO; 3886 goto finish; 3887 } 3888 3889 qdict_del(qdict, BLOCK_OPT_DATA_FILE); 3890 qdict_put_str(qdict, "data-file", data_bs->node_name); 3891 } 3892 3893 /* Set 'driver' and 'node' options */ 3894 qdict_put_str(qdict, "driver", "qcow2"); 3895 qdict_put_str(qdict, "file", bs->node_name); 3896 3897 /* Now get the QAPI type BlockdevCreateOptions */ 3898 v = qobject_input_visitor_new_flat_confused(qdict, errp); 3899 if (!v) { 3900 ret = -EINVAL; 3901 goto finish; 3902 } 3903 3904 visit_type_BlockdevCreateOptions(v, NULL, &create_options, errp); 3905 visit_free(v); 3906 if (!create_options) { 3907 ret = -EINVAL; 3908 goto finish; 3909 } 3910 3911 /* Silently round up size */ 3912 create_options->u.qcow2.size = ROUND_UP(create_options->u.qcow2.size, 3913 BDRV_SECTOR_SIZE); 3914 3915 /* Create the qcow2 image (format layer) */ 3916 ret = qcow2_co_create(create_options, errp); 3917 finish: 3918 if (ret < 0) { 3919 bdrv_co_delete_file_noerr(bs); 3920 bdrv_co_delete_file_noerr(data_bs); 3921 } else { 3922 ret = 0; 3923 } 3924 3925 qobject_unref(qdict); 3926 bdrv_unref(bs); 3927 bdrv_unref(data_bs); 3928 qapi_free_BlockdevCreateOptions(create_options); 3929 return ret; 3930 } 3931 3932 3933 static bool is_zero(BlockDriverState *bs, int64_t offset, int64_t bytes) 3934 { 3935 int64_t nr; 3936 int res; 3937 3938 /* Clamp to image length, before checking status of underlying sectors */ 3939 if (offset + bytes > bs->total_sectors * BDRV_SECTOR_SIZE) { 3940 bytes = bs->total_sectors * BDRV_SECTOR_SIZE - offset; 3941 } 3942 3943 if (!bytes) { 3944 return true; 3945 } 3946 3947 /* 3948 * bdrv_block_status_above doesn't merge different types of zeros, for 3949 * example, zeros which come from the region which is unallocated in 3950 * the whole backing chain, and zeros which come because of a short 3951 * backing file. So, we need a loop. 3952 */ 3953 do { 3954 res = bdrv_block_status_above(bs, NULL, offset, bytes, &nr, NULL, NULL); 3955 offset += nr; 3956 bytes -= nr; 3957 } while (res >= 0 && (res & BDRV_BLOCK_ZERO) && nr && bytes); 3958 3959 return res >= 0 && (res & BDRV_BLOCK_ZERO) && bytes == 0; 3960 } 3961 3962 static coroutine_fn int qcow2_co_pwrite_zeroes(BlockDriverState *bs, 3963 int64_t offset, int64_t bytes, BdrvRequestFlags flags) 3964 { 3965 int ret; 3966 BDRVQcow2State *s = bs->opaque; 3967 3968 uint32_t head = offset_into_subcluster(s, offset); 3969 uint32_t tail = ROUND_UP(offset + bytes, s->subcluster_size) - 3970 (offset + bytes); 3971 3972 trace_qcow2_pwrite_zeroes_start_req(qemu_coroutine_self(), offset, bytes); 3973 if (offset + bytes == bs->total_sectors * BDRV_SECTOR_SIZE) { 3974 tail = 0; 3975 } 3976 3977 if (head || tail) { 3978 uint64_t off; 3979 unsigned int nr; 3980 QCow2SubclusterType type; 3981 3982 assert(head + bytes + tail <= s->subcluster_size); 3983 3984 /* check whether remainder of cluster already reads as zero */ 3985 if (!(is_zero(bs, offset - head, head) && 3986 is_zero(bs, offset + bytes, tail))) { 3987 return -ENOTSUP; 3988 } 3989 3990 qemu_co_mutex_lock(&s->lock); 3991 /* We can have new write after previous check */ 3992 offset -= head; 3993 bytes = s->subcluster_size; 3994 nr = s->subcluster_size; 3995 ret = qcow2_get_host_offset(bs, offset, &nr, &off, &type); 3996 if (ret < 0 || 3997 (type != QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN && 3998 type != QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC && 3999 type != QCOW2_SUBCLUSTER_ZERO_PLAIN && 4000 type != QCOW2_SUBCLUSTER_ZERO_ALLOC)) { 4001 qemu_co_mutex_unlock(&s->lock); 4002 return ret < 0 ? ret : -ENOTSUP; 4003 } 4004 } else { 4005 qemu_co_mutex_lock(&s->lock); 4006 } 4007 4008 trace_qcow2_pwrite_zeroes(qemu_coroutine_self(), offset, bytes); 4009 4010 /* Whatever is left can use real zero subclusters */ 4011 ret = qcow2_subcluster_zeroize(bs, offset, bytes, flags); 4012 qemu_co_mutex_unlock(&s->lock); 4013 4014 return ret; 4015 } 4016 4017 static coroutine_fn int qcow2_co_pdiscard(BlockDriverState *bs, 4018 int64_t offset, int64_t bytes) 4019 { 4020 int ret; 4021 BDRVQcow2State *s = bs->opaque; 4022 4023 /* If the image does not support QCOW_OFLAG_ZERO then discarding 4024 * clusters could expose stale data from the backing file. */ 4025 if (s->qcow_version < 3 && bs->backing) { 4026 return -ENOTSUP; 4027 } 4028 4029 if (!QEMU_IS_ALIGNED(offset | bytes, s->cluster_size)) { 4030 assert(bytes < s->cluster_size); 4031 /* Ignore partial clusters, except for the special case of the 4032 * complete partial cluster at the end of an unaligned file */ 4033 if (!QEMU_IS_ALIGNED(offset, s->cluster_size) || 4034 offset + bytes != bs->total_sectors * BDRV_SECTOR_SIZE) { 4035 return -ENOTSUP; 4036 } 4037 } 4038 4039 qemu_co_mutex_lock(&s->lock); 4040 ret = qcow2_cluster_discard(bs, offset, bytes, QCOW2_DISCARD_REQUEST, 4041 false); 4042 qemu_co_mutex_unlock(&s->lock); 4043 return ret; 4044 } 4045 4046 static int coroutine_fn 4047 qcow2_co_copy_range_from(BlockDriverState *bs, 4048 BdrvChild *src, int64_t src_offset, 4049 BdrvChild *dst, int64_t dst_offset, 4050 int64_t bytes, BdrvRequestFlags read_flags, 4051 BdrvRequestFlags write_flags) 4052 { 4053 BDRVQcow2State *s = bs->opaque; 4054 int ret; 4055 unsigned int cur_bytes; /* number of bytes in current iteration */ 4056 BdrvChild *child = NULL; 4057 BdrvRequestFlags cur_write_flags; 4058 4059 assert(!bs->encrypted); 4060 qemu_co_mutex_lock(&s->lock); 4061 4062 while (bytes != 0) { 4063 uint64_t copy_offset = 0; 4064 QCow2SubclusterType type; 4065 /* prepare next request */ 4066 cur_bytes = MIN(bytes, INT_MAX); 4067 cur_write_flags = write_flags; 4068 4069 ret = qcow2_get_host_offset(bs, src_offset, &cur_bytes, 4070 ©_offset, &type); 4071 if (ret < 0) { 4072 goto out; 4073 } 4074 4075 switch (type) { 4076 case QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN: 4077 case QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC: 4078 if (bs->backing && bs->backing->bs) { 4079 int64_t backing_length = bdrv_getlength(bs->backing->bs); 4080 if (src_offset >= backing_length) { 4081 cur_write_flags |= BDRV_REQ_ZERO_WRITE; 4082 } else { 4083 child = bs->backing; 4084 cur_bytes = MIN(cur_bytes, backing_length - src_offset); 4085 copy_offset = src_offset; 4086 } 4087 } else { 4088 cur_write_flags |= BDRV_REQ_ZERO_WRITE; 4089 } 4090 break; 4091 4092 case QCOW2_SUBCLUSTER_ZERO_PLAIN: 4093 case QCOW2_SUBCLUSTER_ZERO_ALLOC: 4094 cur_write_flags |= BDRV_REQ_ZERO_WRITE; 4095 break; 4096 4097 case QCOW2_SUBCLUSTER_COMPRESSED: 4098 ret = -ENOTSUP; 4099 goto out; 4100 4101 case QCOW2_SUBCLUSTER_NORMAL: 4102 child = s->data_file; 4103 break; 4104 4105 default: 4106 abort(); 4107 } 4108 qemu_co_mutex_unlock(&s->lock); 4109 ret = bdrv_co_copy_range_from(child, 4110 copy_offset, 4111 dst, dst_offset, 4112 cur_bytes, read_flags, cur_write_flags); 4113 qemu_co_mutex_lock(&s->lock); 4114 if (ret < 0) { 4115 goto out; 4116 } 4117 4118 bytes -= cur_bytes; 4119 src_offset += cur_bytes; 4120 dst_offset += cur_bytes; 4121 } 4122 ret = 0; 4123 4124 out: 4125 qemu_co_mutex_unlock(&s->lock); 4126 return ret; 4127 } 4128 4129 static int coroutine_fn 4130 qcow2_co_copy_range_to(BlockDriverState *bs, 4131 BdrvChild *src, int64_t src_offset, 4132 BdrvChild *dst, int64_t dst_offset, 4133 int64_t bytes, BdrvRequestFlags read_flags, 4134 BdrvRequestFlags write_flags) 4135 { 4136 BDRVQcow2State *s = bs->opaque; 4137 int ret; 4138 unsigned int cur_bytes; /* number of sectors in current iteration */ 4139 uint64_t host_offset; 4140 QCowL2Meta *l2meta = NULL; 4141 4142 assert(!bs->encrypted); 4143 4144 qemu_co_mutex_lock(&s->lock); 4145 4146 while (bytes != 0) { 4147 4148 l2meta = NULL; 4149 4150 cur_bytes = MIN(bytes, INT_MAX); 4151 4152 /* TODO: 4153 * If src->bs == dst->bs, we could simply copy by incrementing 4154 * the refcnt, without copying user data. 4155 * Or if src->bs == dst->bs->backing->bs, we could copy by discarding. */ 4156 ret = qcow2_alloc_host_offset(bs, dst_offset, &cur_bytes, 4157 &host_offset, &l2meta); 4158 if (ret < 0) { 4159 goto fail; 4160 } 4161 4162 ret = qcow2_pre_write_overlap_check(bs, 0, host_offset, cur_bytes, 4163 true); 4164 if (ret < 0) { 4165 goto fail; 4166 } 4167 4168 qemu_co_mutex_unlock(&s->lock); 4169 ret = bdrv_co_copy_range_to(src, src_offset, s->data_file, host_offset, 4170 cur_bytes, read_flags, write_flags); 4171 qemu_co_mutex_lock(&s->lock); 4172 if (ret < 0) { 4173 goto fail; 4174 } 4175 4176 ret = qcow2_handle_l2meta(bs, &l2meta, true); 4177 if (ret) { 4178 goto fail; 4179 } 4180 4181 bytes -= cur_bytes; 4182 src_offset += cur_bytes; 4183 dst_offset += cur_bytes; 4184 } 4185 ret = 0; 4186 4187 fail: 4188 qcow2_handle_l2meta(bs, &l2meta, false); 4189 4190 qemu_co_mutex_unlock(&s->lock); 4191 4192 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret); 4193 4194 return ret; 4195 } 4196 4197 static int coroutine_fn qcow2_co_truncate(BlockDriverState *bs, int64_t offset, 4198 bool exact, PreallocMode prealloc, 4199 BdrvRequestFlags flags, Error **errp) 4200 { 4201 BDRVQcow2State *s = bs->opaque; 4202 uint64_t old_length; 4203 int64_t new_l1_size; 4204 int ret; 4205 QDict *options; 4206 4207 if (prealloc != PREALLOC_MODE_OFF && prealloc != PREALLOC_MODE_METADATA && 4208 prealloc != PREALLOC_MODE_FALLOC && prealloc != PREALLOC_MODE_FULL) 4209 { 4210 error_setg(errp, "Unsupported preallocation mode '%s'", 4211 PreallocMode_str(prealloc)); 4212 return -ENOTSUP; 4213 } 4214 4215 if (!QEMU_IS_ALIGNED(offset, BDRV_SECTOR_SIZE)) { 4216 error_setg(errp, "The new size must be a multiple of %u", 4217 (unsigned) BDRV_SECTOR_SIZE); 4218 return -EINVAL; 4219 } 4220 4221 qemu_co_mutex_lock(&s->lock); 4222 4223 /* 4224 * Even though we store snapshot size for all images, it was not 4225 * required until v3, so it is not safe to proceed for v2. 4226 */ 4227 if (s->nb_snapshots && s->qcow_version < 3) { 4228 error_setg(errp, "Can't resize a v2 image which has snapshots"); 4229 ret = -ENOTSUP; 4230 goto fail; 4231 } 4232 4233 /* See qcow2-bitmap.c for which bitmap scenarios prevent a resize. */ 4234 if (qcow2_truncate_bitmaps_check(bs, errp)) { 4235 ret = -ENOTSUP; 4236 goto fail; 4237 } 4238 4239 old_length = bs->total_sectors * BDRV_SECTOR_SIZE; 4240 new_l1_size = size_to_l1(s, offset); 4241 4242 if (offset < old_length) { 4243 int64_t last_cluster, old_file_size; 4244 if (prealloc != PREALLOC_MODE_OFF) { 4245 error_setg(errp, 4246 "Preallocation can't be used for shrinking an image"); 4247 ret = -EINVAL; 4248 goto fail; 4249 } 4250 4251 ret = qcow2_cluster_discard(bs, ROUND_UP(offset, s->cluster_size), 4252 old_length - ROUND_UP(offset, 4253 s->cluster_size), 4254 QCOW2_DISCARD_ALWAYS, true); 4255 if (ret < 0) { 4256 error_setg_errno(errp, -ret, "Failed to discard cropped clusters"); 4257 goto fail; 4258 } 4259 4260 ret = qcow2_shrink_l1_table(bs, new_l1_size); 4261 if (ret < 0) { 4262 error_setg_errno(errp, -ret, 4263 "Failed to reduce the number of L2 tables"); 4264 goto fail; 4265 } 4266 4267 ret = qcow2_shrink_reftable(bs); 4268 if (ret < 0) { 4269 error_setg_errno(errp, -ret, 4270 "Failed to discard unused refblocks"); 4271 goto fail; 4272 } 4273 4274 old_file_size = bdrv_getlength(bs->file->bs); 4275 if (old_file_size < 0) { 4276 error_setg_errno(errp, -old_file_size, 4277 "Failed to inquire current file length"); 4278 ret = old_file_size; 4279 goto fail; 4280 } 4281 last_cluster = qcow2_get_last_cluster(bs, old_file_size); 4282 if (last_cluster < 0) { 4283 error_setg_errno(errp, -last_cluster, 4284 "Failed to find the last cluster"); 4285 ret = last_cluster; 4286 goto fail; 4287 } 4288 if ((last_cluster + 1) * s->cluster_size < old_file_size) { 4289 Error *local_err = NULL; 4290 4291 /* 4292 * Do not pass @exact here: It will not help the user if 4293 * we get an error here just because they wanted to shrink 4294 * their qcow2 image (on a block device) with qemu-img. 4295 * (And on the qcow2 layer, the @exact requirement is 4296 * always fulfilled, so there is no need to pass it on.) 4297 */ 4298 bdrv_co_truncate(bs->file, (last_cluster + 1) * s->cluster_size, 4299 false, PREALLOC_MODE_OFF, 0, &local_err); 4300 if (local_err) { 4301 warn_reportf_err(local_err, 4302 "Failed to truncate the tail of the image: "); 4303 } 4304 } 4305 } else { 4306 ret = qcow2_grow_l1_table(bs, new_l1_size, true); 4307 if (ret < 0) { 4308 error_setg_errno(errp, -ret, "Failed to grow the L1 table"); 4309 goto fail; 4310 } 4311 4312 if (data_file_is_raw(bs) && prealloc == PREALLOC_MODE_OFF) { 4313 /* 4314 * When creating a qcow2 image with data-file-raw, we enforce 4315 * at least prealloc=metadata, so that the L1/L2 tables are 4316 * fully allocated and reading from the data file will return 4317 * the same data as reading from the qcow2 image. When the 4318 * image is grown, we must consequently preallocate the 4319 * metadata structures to cover the added area. 4320 */ 4321 prealloc = PREALLOC_MODE_METADATA; 4322 } 4323 } 4324 4325 switch (prealloc) { 4326 case PREALLOC_MODE_OFF: 4327 if (has_data_file(bs)) { 4328 /* 4329 * If the caller wants an exact resize, the external data 4330 * file should be resized to the exact target size, too, 4331 * so we pass @exact here. 4332 */ 4333 ret = bdrv_co_truncate(s->data_file, offset, exact, prealloc, 0, 4334 errp); 4335 if (ret < 0) { 4336 goto fail; 4337 } 4338 } 4339 break; 4340 4341 case PREALLOC_MODE_METADATA: 4342 ret = preallocate_co(bs, old_length, offset, prealloc, errp); 4343 if (ret < 0) { 4344 goto fail; 4345 } 4346 break; 4347 4348 case PREALLOC_MODE_FALLOC: 4349 case PREALLOC_MODE_FULL: 4350 { 4351 int64_t allocation_start, host_offset, guest_offset; 4352 int64_t clusters_allocated; 4353 int64_t old_file_size, last_cluster, new_file_size; 4354 uint64_t nb_new_data_clusters, nb_new_l2_tables; 4355 bool subclusters_need_allocation = false; 4356 4357 /* With a data file, preallocation means just allocating the metadata 4358 * and forwarding the truncate request to the data file */ 4359 if (has_data_file(bs)) { 4360 ret = preallocate_co(bs, old_length, offset, prealloc, errp); 4361 if (ret < 0) { 4362 goto fail; 4363 } 4364 break; 4365 } 4366 4367 old_file_size = bdrv_getlength(bs->file->bs); 4368 if (old_file_size < 0) { 4369 error_setg_errno(errp, -old_file_size, 4370 "Failed to inquire current file length"); 4371 ret = old_file_size; 4372 goto fail; 4373 } 4374 4375 last_cluster = qcow2_get_last_cluster(bs, old_file_size); 4376 if (last_cluster >= 0) { 4377 old_file_size = (last_cluster + 1) * s->cluster_size; 4378 } else { 4379 old_file_size = ROUND_UP(old_file_size, s->cluster_size); 4380 } 4381 4382 nb_new_data_clusters = (ROUND_UP(offset, s->cluster_size) - 4383 start_of_cluster(s, old_length)) >> s->cluster_bits; 4384 4385 /* This is an overestimation; we will not actually allocate space for 4386 * these in the file but just make sure the new refcount structures are 4387 * able to cover them so we will not have to allocate new refblocks 4388 * while entering the data blocks in the potentially new L2 tables. 4389 * (We do not actually care where the L2 tables are placed. Maybe they 4390 * are already allocated or they can be placed somewhere before 4391 * @old_file_size. It does not matter because they will be fully 4392 * allocated automatically, so they do not need to be covered by the 4393 * preallocation. All that matters is that we will not have to allocate 4394 * new refcount structures for them.) */ 4395 nb_new_l2_tables = DIV_ROUND_UP(nb_new_data_clusters, 4396 s->cluster_size / l2_entry_size(s)); 4397 /* The cluster range may not be aligned to L2 boundaries, so add one L2 4398 * table for a potential head/tail */ 4399 nb_new_l2_tables++; 4400 4401 allocation_start = qcow2_refcount_area(bs, old_file_size, 4402 nb_new_data_clusters + 4403 nb_new_l2_tables, 4404 true, 0, 0); 4405 if (allocation_start < 0) { 4406 error_setg_errno(errp, -allocation_start, 4407 "Failed to resize refcount structures"); 4408 ret = allocation_start; 4409 goto fail; 4410 } 4411 4412 clusters_allocated = qcow2_alloc_clusters_at(bs, allocation_start, 4413 nb_new_data_clusters); 4414 if (clusters_allocated < 0) { 4415 error_setg_errno(errp, -clusters_allocated, 4416 "Failed to allocate data clusters"); 4417 ret = clusters_allocated; 4418 goto fail; 4419 } 4420 4421 assert(clusters_allocated == nb_new_data_clusters); 4422 4423 /* Allocate the data area */ 4424 new_file_size = allocation_start + 4425 nb_new_data_clusters * s->cluster_size; 4426 /* 4427 * Image file grows, so @exact does not matter. 4428 * 4429 * If we need to zero out the new area, try first whether the protocol 4430 * driver can already take care of this. 4431 */ 4432 if (flags & BDRV_REQ_ZERO_WRITE) { 4433 ret = bdrv_co_truncate(bs->file, new_file_size, false, prealloc, 4434 BDRV_REQ_ZERO_WRITE, NULL); 4435 if (ret >= 0) { 4436 flags &= ~BDRV_REQ_ZERO_WRITE; 4437 /* Ensure that we read zeroes and not backing file data */ 4438 subclusters_need_allocation = true; 4439 } 4440 } else { 4441 ret = -1; 4442 } 4443 if (ret < 0) { 4444 ret = bdrv_co_truncate(bs->file, new_file_size, false, prealloc, 0, 4445 errp); 4446 } 4447 if (ret < 0) { 4448 error_prepend(errp, "Failed to resize underlying file: "); 4449 qcow2_free_clusters(bs, allocation_start, 4450 nb_new_data_clusters * s->cluster_size, 4451 QCOW2_DISCARD_OTHER); 4452 goto fail; 4453 } 4454 4455 /* Create the necessary L2 entries */ 4456 host_offset = allocation_start; 4457 guest_offset = old_length; 4458 while (nb_new_data_clusters) { 4459 int64_t nb_clusters = MIN( 4460 nb_new_data_clusters, 4461 s->l2_slice_size - offset_to_l2_slice_index(s, guest_offset)); 4462 unsigned cow_start_length = offset_into_cluster(s, guest_offset); 4463 QCowL2Meta allocation; 4464 guest_offset = start_of_cluster(s, guest_offset); 4465 allocation = (QCowL2Meta) { 4466 .offset = guest_offset, 4467 .alloc_offset = host_offset, 4468 .nb_clusters = nb_clusters, 4469 .cow_start = { 4470 .offset = 0, 4471 .nb_bytes = cow_start_length, 4472 }, 4473 .cow_end = { 4474 .offset = nb_clusters << s->cluster_bits, 4475 .nb_bytes = 0, 4476 }, 4477 .prealloc = !subclusters_need_allocation, 4478 }; 4479 qemu_co_queue_init(&allocation.dependent_requests); 4480 4481 ret = qcow2_alloc_cluster_link_l2(bs, &allocation); 4482 if (ret < 0) { 4483 error_setg_errno(errp, -ret, "Failed to update L2 tables"); 4484 qcow2_free_clusters(bs, host_offset, 4485 nb_new_data_clusters * s->cluster_size, 4486 QCOW2_DISCARD_OTHER); 4487 goto fail; 4488 } 4489 4490 guest_offset += nb_clusters * s->cluster_size; 4491 host_offset += nb_clusters * s->cluster_size; 4492 nb_new_data_clusters -= nb_clusters; 4493 } 4494 break; 4495 } 4496 4497 default: 4498 g_assert_not_reached(); 4499 } 4500 4501 if ((flags & BDRV_REQ_ZERO_WRITE) && offset > old_length) { 4502 uint64_t zero_start = QEMU_ALIGN_UP(old_length, s->subcluster_size); 4503 4504 /* 4505 * Use zero clusters as much as we can. qcow2_subcluster_zeroize() 4506 * requires a subcluster-aligned start. The end may be unaligned if 4507 * it is at the end of the image (which it is here). 4508 */ 4509 if (offset > zero_start) { 4510 ret = qcow2_subcluster_zeroize(bs, zero_start, offset - zero_start, 4511 0); 4512 if (ret < 0) { 4513 error_setg_errno(errp, -ret, "Failed to zero out new clusters"); 4514 goto fail; 4515 } 4516 } 4517 4518 /* Write explicit zeros for the unaligned head */ 4519 if (zero_start > old_length) { 4520 uint64_t len = MIN(zero_start, offset) - old_length; 4521 uint8_t *buf = qemu_blockalign0(bs, len); 4522 QEMUIOVector qiov; 4523 qemu_iovec_init_buf(&qiov, buf, len); 4524 4525 qemu_co_mutex_unlock(&s->lock); 4526 ret = qcow2_co_pwritev_part(bs, old_length, len, &qiov, 0, 0); 4527 qemu_co_mutex_lock(&s->lock); 4528 4529 qemu_vfree(buf); 4530 if (ret < 0) { 4531 error_setg_errno(errp, -ret, "Failed to zero out the new area"); 4532 goto fail; 4533 } 4534 } 4535 } 4536 4537 if (prealloc != PREALLOC_MODE_OFF) { 4538 /* Flush metadata before actually changing the image size */ 4539 ret = qcow2_write_caches(bs); 4540 if (ret < 0) { 4541 error_setg_errno(errp, -ret, 4542 "Failed to flush the preallocated area to disk"); 4543 goto fail; 4544 } 4545 } 4546 4547 bs->total_sectors = offset / BDRV_SECTOR_SIZE; 4548 4549 /* write updated header.size */ 4550 offset = cpu_to_be64(offset); 4551 ret = bdrv_co_pwrite_sync(bs->file, offsetof(QCowHeader, size), 4552 sizeof(offset), &offset, 0); 4553 if (ret < 0) { 4554 error_setg_errno(errp, -ret, "Failed to update the image size"); 4555 goto fail; 4556 } 4557 4558 s->l1_vm_state_index = new_l1_size; 4559 4560 /* Update cache sizes */ 4561 options = qdict_clone_shallow(bs->options); 4562 ret = qcow2_update_options(bs, options, s->flags, errp); 4563 qobject_unref(options); 4564 if (ret < 0) { 4565 goto fail; 4566 } 4567 ret = 0; 4568 fail: 4569 qemu_co_mutex_unlock(&s->lock); 4570 return ret; 4571 } 4572 4573 static coroutine_fn int 4574 qcow2_co_pwritev_compressed_task(BlockDriverState *bs, 4575 uint64_t offset, uint64_t bytes, 4576 QEMUIOVector *qiov, size_t qiov_offset) 4577 { 4578 BDRVQcow2State *s = bs->opaque; 4579 int ret; 4580 ssize_t out_len; 4581 uint8_t *buf, *out_buf; 4582 uint64_t cluster_offset; 4583 4584 assert(bytes == s->cluster_size || (bytes < s->cluster_size && 4585 (offset + bytes == bs->total_sectors << BDRV_SECTOR_BITS))); 4586 4587 buf = qemu_blockalign(bs, s->cluster_size); 4588 if (bytes < s->cluster_size) { 4589 /* Zero-pad last write if image size is not cluster aligned */ 4590 memset(buf + bytes, 0, s->cluster_size - bytes); 4591 } 4592 qemu_iovec_to_buf(qiov, qiov_offset, buf, bytes); 4593 4594 out_buf = g_malloc(s->cluster_size); 4595 4596 out_len = qcow2_co_compress(bs, out_buf, s->cluster_size - 1, 4597 buf, s->cluster_size); 4598 if (out_len == -ENOMEM) { 4599 /* could not compress: write normal cluster */ 4600 ret = qcow2_co_pwritev_part(bs, offset, bytes, qiov, qiov_offset, 0); 4601 if (ret < 0) { 4602 goto fail; 4603 } 4604 goto success; 4605 } else if (out_len < 0) { 4606 ret = -EINVAL; 4607 goto fail; 4608 } 4609 4610 qemu_co_mutex_lock(&s->lock); 4611 ret = qcow2_alloc_compressed_cluster_offset(bs, offset, out_len, 4612 &cluster_offset); 4613 if (ret < 0) { 4614 qemu_co_mutex_unlock(&s->lock); 4615 goto fail; 4616 } 4617 4618 ret = qcow2_pre_write_overlap_check(bs, 0, cluster_offset, out_len, true); 4619 qemu_co_mutex_unlock(&s->lock); 4620 if (ret < 0) { 4621 goto fail; 4622 } 4623 4624 BLKDBG_EVENT(s->data_file, BLKDBG_WRITE_COMPRESSED); 4625 ret = bdrv_co_pwrite(s->data_file, cluster_offset, out_len, out_buf, 0); 4626 if (ret < 0) { 4627 goto fail; 4628 } 4629 success: 4630 ret = 0; 4631 fail: 4632 qemu_vfree(buf); 4633 g_free(out_buf); 4634 return ret; 4635 } 4636 4637 static coroutine_fn int qcow2_co_pwritev_compressed_task_entry(AioTask *task) 4638 { 4639 Qcow2AioTask *t = container_of(task, Qcow2AioTask, task); 4640 4641 assert(!t->subcluster_type && !t->l2meta); 4642 4643 return qcow2_co_pwritev_compressed_task(t->bs, t->offset, t->bytes, t->qiov, 4644 t->qiov_offset); 4645 } 4646 4647 /* 4648 * XXX: put compressed sectors first, then all the cluster aligned 4649 * tables to avoid losing bytes in alignment 4650 */ 4651 static coroutine_fn int 4652 qcow2_co_pwritev_compressed_part(BlockDriverState *bs, 4653 int64_t offset, int64_t bytes, 4654 QEMUIOVector *qiov, size_t qiov_offset) 4655 { 4656 BDRVQcow2State *s = bs->opaque; 4657 AioTaskPool *aio = NULL; 4658 int ret = 0; 4659 4660 if (has_data_file(bs)) { 4661 return -ENOTSUP; 4662 } 4663 4664 if (bytes == 0) { 4665 /* 4666 * align end of file to a sector boundary to ease reading with 4667 * sector based I/Os 4668 */ 4669 int64_t len = bdrv_getlength(bs->file->bs); 4670 if (len < 0) { 4671 return len; 4672 } 4673 return bdrv_co_truncate(bs->file, len, false, PREALLOC_MODE_OFF, 0, 4674 NULL); 4675 } 4676 4677 if (offset_into_cluster(s, offset)) { 4678 return -EINVAL; 4679 } 4680 4681 if (offset_into_cluster(s, bytes) && 4682 (offset + bytes) != (bs->total_sectors << BDRV_SECTOR_BITS)) { 4683 return -EINVAL; 4684 } 4685 4686 while (bytes && aio_task_pool_status(aio) == 0) { 4687 uint64_t chunk_size = MIN(bytes, s->cluster_size); 4688 4689 if (!aio && chunk_size != bytes) { 4690 aio = aio_task_pool_new(QCOW2_MAX_WORKERS); 4691 } 4692 4693 ret = qcow2_add_task(bs, aio, qcow2_co_pwritev_compressed_task_entry, 4694 0, 0, offset, chunk_size, qiov, qiov_offset, NULL); 4695 if (ret < 0) { 4696 break; 4697 } 4698 qiov_offset += chunk_size; 4699 offset += chunk_size; 4700 bytes -= chunk_size; 4701 } 4702 4703 if (aio) { 4704 aio_task_pool_wait_all(aio); 4705 if (ret == 0) { 4706 ret = aio_task_pool_status(aio); 4707 } 4708 g_free(aio); 4709 } 4710 4711 return ret; 4712 } 4713 4714 static int coroutine_fn 4715 qcow2_co_preadv_compressed(BlockDriverState *bs, 4716 uint64_t l2_entry, 4717 uint64_t offset, 4718 uint64_t bytes, 4719 QEMUIOVector *qiov, 4720 size_t qiov_offset) 4721 { 4722 BDRVQcow2State *s = bs->opaque; 4723 int ret = 0, csize; 4724 uint64_t coffset; 4725 uint8_t *buf, *out_buf; 4726 int offset_in_cluster = offset_into_cluster(s, offset); 4727 4728 qcow2_parse_compressed_l2_entry(bs, l2_entry, &coffset, &csize); 4729 4730 buf = g_try_malloc(csize); 4731 if (!buf) { 4732 return -ENOMEM; 4733 } 4734 4735 out_buf = qemu_blockalign(bs, s->cluster_size); 4736 4737 BLKDBG_EVENT(bs->file, BLKDBG_READ_COMPRESSED); 4738 ret = bdrv_co_pread(bs->file, coffset, csize, buf, 0); 4739 if (ret < 0) { 4740 goto fail; 4741 } 4742 4743 if (qcow2_co_decompress(bs, out_buf, s->cluster_size, buf, csize) < 0) { 4744 ret = -EIO; 4745 goto fail; 4746 } 4747 4748 qemu_iovec_from_buf(qiov, qiov_offset, out_buf + offset_in_cluster, bytes); 4749 4750 fail: 4751 qemu_vfree(out_buf); 4752 g_free(buf); 4753 4754 return ret; 4755 } 4756 4757 static int make_completely_empty(BlockDriverState *bs) 4758 { 4759 BDRVQcow2State *s = bs->opaque; 4760 Error *local_err = NULL; 4761 int ret, l1_clusters; 4762 int64_t offset; 4763 uint64_t *new_reftable = NULL; 4764 uint64_t rt_entry, l1_size2; 4765 struct { 4766 uint64_t l1_offset; 4767 uint64_t reftable_offset; 4768 uint32_t reftable_clusters; 4769 } QEMU_PACKED l1_ofs_rt_ofs_cls; 4770 4771 ret = qcow2_cache_empty(bs, s->l2_table_cache); 4772 if (ret < 0) { 4773 goto fail; 4774 } 4775 4776 ret = qcow2_cache_empty(bs, s->refcount_block_cache); 4777 if (ret < 0) { 4778 goto fail; 4779 } 4780 4781 /* Refcounts will be broken utterly */ 4782 ret = qcow2_mark_dirty(bs); 4783 if (ret < 0) { 4784 goto fail; 4785 } 4786 4787 BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE); 4788 4789 l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / L1E_SIZE); 4790 l1_size2 = (uint64_t)s->l1_size * L1E_SIZE; 4791 4792 /* After this call, neither the in-memory nor the on-disk refcount 4793 * information accurately describe the actual references */ 4794 4795 ret = bdrv_pwrite_zeroes(bs->file, s->l1_table_offset, 4796 l1_clusters * s->cluster_size, 0); 4797 if (ret < 0) { 4798 goto fail_broken_refcounts; 4799 } 4800 memset(s->l1_table, 0, l1_size2); 4801 4802 BLKDBG_EVENT(bs->file, BLKDBG_EMPTY_IMAGE_PREPARE); 4803 4804 /* Overwrite enough clusters at the beginning of the sectors to place 4805 * the refcount table, a refcount block and the L1 table in; this may 4806 * overwrite parts of the existing refcount and L1 table, which is not 4807 * an issue because the dirty flag is set, complete data loss is in fact 4808 * desired and partial data loss is consequently fine as well */ 4809 ret = bdrv_pwrite_zeroes(bs->file, s->cluster_size, 4810 (2 + l1_clusters) * s->cluster_size, 0); 4811 /* This call (even if it failed overall) may have overwritten on-disk 4812 * refcount structures; in that case, the in-memory refcount information 4813 * will probably differ from the on-disk information which makes the BDS 4814 * unusable */ 4815 if (ret < 0) { 4816 goto fail_broken_refcounts; 4817 } 4818 4819 BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE); 4820 BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_UPDATE); 4821 4822 /* "Create" an empty reftable (one cluster) directly after the image 4823 * header and an empty L1 table three clusters after the image header; 4824 * the cluster between those two will be used as the first refblock */ 4825 l1_ofs_rt_ofs_cls.l1_offset = cpu_to_be64(3 * s->cluster_size); 4826 l1_ofs_rt_ofs_cls.reftable_offset = cpu_to_be64(s->cluster_size); 4827 l1_ofs_rt_ofs_cls.reftable_clusters = cpu_to_be32(1); 4828 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, l1_table_offset), 4829 sizeof(l1_ofs_rt_ofs_cls), &l1_ofs_rt_ofs_cls, 0); 4830 if (ret < 0) { 4831 goto fail_broken_refcounts; 4832 } 4833 4834 s->l1_table_offset = 3 * s->cluster_size; 4835 4836 new_reftable = g_try_new0(uint64_t, s->cluster_size / REFTABLE_ENTRY_SIZE); 4837 if (!new_reftable) { 4838 ret = -ENOMEM; 4839 goto fail_broken_refcounts; 4840 } 4841 4842 s->refcount_table_offset = s->cluster_size; 4843 s->refcount_table_size = s->cluster_size / REFTABLE_ENTRY_SIZE; 4844 s->max_refcount_table_index = 0; 4845 4846 g_free(s->refcount_table); 4847 s->refcount_table = new_reftable; 4848 new_reftable = NULL; 4849 4850 /* Now the in-memory refcount information again corresponds to the on-disk 4851 * information (reftable is empty and no refblocks (the refblock cache is 4852 * empty)); however, this means some clusters (e.g. the image header) are 4853 * referenced, but not refcounted, but the normal qcow2 code assumes that 4854 * the in-memory information is always correct */ 4855 4856 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC); 4857 4858 /* Enter the first refblock into the reftable */ 4859 rt_entry = cpu_to_be64(2 * s->cluster_size); 4860 ret = bdrv_pwrite_sync(bs->file, s->cluster_size, sizeof(rt_entry), 4861 &rt_entry, 0); 4862 if (ret < 0) { 4863 goto fail_broken_refcounts; 4864 } 4865 s->refcount_table[0] = 2 * s->cluster_size; 4866 4867 s->free_cluster_index = 0; 4868 assert(3 + l1_clusters <= s->refcount_block_size); 4869 offset = qcow2_alloc_clusters(bs, 3 * s->cluster_size + l1_size2); 4870 if (offset < 0) { 4871 ret = offset; 4872 goto fail_broken_refcounts; 4873 } else if (offset > 0) { 4874 error_report("First cluster in emptied image is in use"); 4875 abort(); 4876 } 4877 4878 /* Now finally the in-memory information corresponds to the on-disk 4879 * structures and is correct */ 4880 ret = qcow2_mark_clean(bs); 4881 if (ret < 0) { 4882 goto fail; 4883 } 4884 4885 ret = bdrv_truncate(bs->file, (3 + l1_clusters) * s->cluster_size, false, 4886 PREALLOC_MODE_OFF, 0, &local_err); 4887 if (ret < 0) { 4888 error_report_err(local_err); 4889 goto fail; 4890 } 4891 4892 return 0; 4893 4894 fail_broken_refcounts: 4895 /* The BDS is unusable at this point. If we wanted to make it usable, we 4896 * would have to call qcow2_refcount_close(), qcow2_refcount_init(), 4897 * qcow2_check_refcounts(), qcow2_refcount_close() and qcow2_refcount_init() 4898 * again. However, because the functions which could have caused this error 4899 * path to be taken are used by those functions as well, it's very likely 4900 * that that sequence will fail as well. Therefore, just eject the BDS. */ 4901 bs->drv = NULL; 4902 4903 fail: 4904 g_free(new_reftable); 4905 return ret; 4906 } 4907 4908 static int qcow2_make_empty(BlockDriverState *bs) 4909 { 4910 BDRVQcow2State *s = bs->opaque; 4911 uint64_t offset, end_offset; 4912 int step = QEMU_ALIGN_DOWN(INT_MAX, s->cluster_size); 4913 int l1_clusters, ret = 0; 4914 4915 l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / L1E_SIZE); 4916 4917 if (s->qcow_version >= 3 && !s->snapshots && !s->nb_bitmaps && 4918 3 + l1_clusters <= s->refcount_block_size && 4919 s->crypt_method_header != QCOW_CRYPT_LUKS && 4920 !has_data_file(bs)) { 4921 /* The following function only works for qcow2 v3 images (it 4922 * requires the dirty flag) and only as long as there are no 4923 * features that reserve extra clusters (such as snapshots, 4924 * LUKS header, or persistent bitmaps), because it completely 4925 * empties the image. Furthermore, the L1 table and three 4926 * additional clusters (image header, refcount table, one 4927 * refcount block) have to fit inside one refcount block. It 4928 * only resets the image file, i.e. does not work with an 4929 * external data file. */ 4930 return make_completely_empty(bs); 4931 } 4932 4933 /* This fallback code simply discards every active cluster; this is slow, 4934 * but works in all cases */ 4935 end_offset = bs->total_sectors * BDRV_SECTOR_SIZE; 4936 for (offset = 0; offset < end_offset; offset += step) { 4937 /* As this function is generally used after committing an external 4938 * snapshot, QCOW2_DISCARD_SNAPSHOT seems appropriate. Also, the 4939 * default action for this kind of discard is to pass the discard, 4940 * which will ideally result in an actually smaller image file, as 4941 * is probably desired. */ 4942 ret = qcow2_cluster_discard(bs, offset, MIN(step, end_offset - offset), 4943 QCOW2_DISCARD_SNAPSHOT, true); 4944 if (ret < 0) { 4945 break; 4946 } 4947 } 4948 4949 return ret; 4950 } 4951 4952 static coroutine_fn int qcow2_co_flush_to_os(BlockDriverState *bs) 4953 { 4954 BDRVQcow2State *s = bs->opaque; 4955 int ret; 4956 4957 qemu_co_mutex_lock(&s->lock); 4958 ret = qcow2_write_caches(bs); 4959 qemu_co_mutex_unlock(&s->lock); 4960 4961 return ret; 4962 } 4963 4964 static BlockMeasureInfo *qcow2_measure(QemuOpts *opts, BlockDriverState *in_bs, 4965 Error **errp) 4966 { 4967 Error *local_err = NULL; 4968 BlockMeasureInfo *info; 4969 uint64_t required = 0; /* bytes that contribute to required size */ 4970 uint64_t virtual_size; /* disk size as seen by guest */ 4971 uint64_t refcount_bits; 4972 uint64_t l2_tables; 4973 uint64_t luks_payload_size = 0; 4974 size_t cluster_size; 4975 int version; 4976 char *optstr; 4977 PreallocMode prealloc; 4978 bool has_backing_file; 4979 bool has_luks; 4980 bool extended_l2; 4981 size_t l2e_size; 4982 4983 /* Parse image creation options */ 4984 extended_l2 = qemu_opt_get_bool_del(opts, BLOCK_OPT_EXTL2, false); 4985 4986 cluster_size = qcow2_opt_get_cluster_size_del(opts, extended_l2, 4987 &local_err); 4988 if (local_err) { 4989 goto err; 4990 } 4991 4992 version = qcow2_opt_get_version_del(opts, &local_err); 4993 if (local_err) { 4994 goto err; 4995 } 4996 4997 refcount_bits = qcow2_opt_get_refcount_bits_del(opts, version, &local_err); 4998 if (local_err) { 4999 goto err; 5000 } 5001 5002 optstr = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC); 5003 prealloc = qapi_enum_parse(&PreallocMode_lookup, optstr, 5004 PREALLOC_MODE_OFF, &local_err); 5005 g_free(optstr); 5006 if (local_err) { 5007 goto err; 5008 } 5009 5010 optstr = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE); 5011 has_backing_file = !!optstr; 5012 g_free(optstr); 5013 5014 optstr = qemu_opt_get_del(opts, BLOCK_OPT_ENCRYPT_FORMAT); 5015 has_luks = optstr && strcmp(optstr, "luks") == 0; 5016 g_free(optstr); 5017 5018 if (has_luks) { 5019 g_autoptr(QCryptoBlockCreateOptions) create_opts = NULL; 5020 QDict *cryptoopts = qcow2_extract_crypto_opts(opts, "luks", errp); 5021 size_t headerlen; 5022 5023 create_opts = block_crypto_create_opts_init(cryptoopts, errp); 5024 qobject_unref(cryptoopts); 5025 if (!create_opts) { 5026 goto err; 5027 } 5028 5029 if (!qcrypto_block_calculate_payload_offset(create_opts, 5030 "encrypt.", 5031 &headerlen, 5032 &local_err)) { 5033 goto err; 5034 } 5035 5036 luks_payload_size = ROUND_UP(headerlen, cluster_size); 5037 } 5038 5039 virtual_size = qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0); 5040 virtual_size = ROUND_UP(virtual_size, cluster_size); 5041 5042 /* Check that virtual disk size is valid */ 5043 l2e_size = extended_l2 ? L2E_SIZE_EXTENDED : L2E_SIZE_NORMAL; 5044 l2_tables = DIV_ROUND_UP(virtual_size / cluster_size, 5045 cluster_size / l2e_size); 5046 if (l2_tables * L1E_SIZE > QCOW_MAX_L1_SIZE) { 5047 error_setg(&local_err, "The image size is too large " 5048 "(try using a larger cluster size)"); 5049 goto err; 5050 } 5051 5052 /* Account for input image */ 5053 if (in_bs) { 5054 int64_t ssize = bdrv_getlength(in_bs); 5055 if (ssize < 0) { 5056 error_setg_errno(&local_err, -ssize, 5057 "Unable to get image virtual_size"); 5058 goto err; 5059 } 5060 5061 virtual_size = ROUND_UP(ssize, cluster_size); 5062 5063 if (has_backing_file) { 5064 /* We don't how much of the backing chain is shared by the input 5065 * image and the new image file. In the worst case the new image's 5066 * backing file has nothing in common with the input image. Be 5067 * conservative and assume all clusters need to be written. 5068 */ 5069 required = virtual_size; 5070 } else { 5071 int64_t offset; 5072 int64_t pnum = 0; 5073 5074 for (offset = 0; offset < ssize; offset += pnum) { 5075 int ret; 5076 5077 ret = bdrv_block_status_above(in_bs, NULL, offset, 5078 ssize - offset, &pnum, NULL, 5079 NULL); 5080 if (ret < 0) { 5081 error_setg_errno(&local_err, -ret, 5082 "Unable to get block status"); 5083 goto err; 5084 } 5085 5086 if (ret & BDRV_BLOCK_ZERO) { 5087 /* Skip zero regions (safe with no backing file) */ 5088 } else if ((ret & (BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED)) == 5089 (BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED)) { 5090 /* Extend pnum to end of cluster for next iteration */ 5091 pnum = ROUND_UP(offset + pnum, cluster_size) - offset; 5092 5093 /* Count clusters we've seen */ 5094 required += offset % cluster_size + pnum; 5095 } 5096 } 5097 } 5098 } 5099 5100 /* Take into account preallocation. Nothing special is needed for 5101 * PREALLOC_MODE_METADATA since metadata is always counted. 5102 */ 5103 if (prealloc == PREALLOC_MODE_FULL || prealloc == PREALLOC_MODE_FALLOC) { 5104 required = virtual_size; 5105 } 5106 5107 info = g_new0(BlockMeasureInfo, 1); 5108 info->fully_allocated = luks_payload_size + 5109 qcow2_calc_prealloc_size(virtual_size, cluster_size, 5110 ctz32(refcount_bits), extended_l2); 5111 5112 /* 5113 * Remove data clusters that are not required. This overestimates the 5114 * required size because metadata needed for the fully allocated file is 5115 * still counted. Show bitmaps only if both source and destination 5116 * would support them. 5117 */ 5118 info->required = info->fully_allocated - virtual_size + required; 5119 info->has_bitmaps = version >= 3 && in_bs && 5120 bdrv_supports_persistent_dirty_bitmap(in_bs); 5121 if (info->has_bitmaps) { 5122 info->bitmaps = qcow2_get_persistent_dirty_bitmap_size(in_bs, 5123 cluster_size); 5124 } 5125 return info; 5126 5127 err: 5128 error_propagate(errp, local_err); 5129 return NULL; 5130 } 5131 5132 static int qcow2_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) 5133 { 5134 BDRVQcow2State *s = bs->opaque; 5135 bdi->cluster_size = s->cluster_size; 5136 bdi->vm_state_offset = qcow2_vm_state_offset(s); 5137 bdi->is_dirty = s->incompatible_features & QCOW2_INCOMPAT_DIRTY; 5138 return 0; 5139 } 5140 5141 static ImageInfoSpecific *qcow2_get_specific_info(BlockDriverState *bs, 5142 Error **errp) 5143 { 5144 BDRVQcow2State *s = bs->opaque; 5145 ImageInfoSpecific *spec_info; 5146 QCryptoBlockInfo *encrypt_info = NULL; 5147 5148 if (s->crypto != NULL) { 5149 encrypt_info = qcrypto_block_get_info(s->crypto, errp); 5150 if (!encrypt_info) { 5151 return NULL; 5152 } 5153 } 5154 5155 spec_info = g_new(ImageInfoSpecific, 1); 5156 *spec_info = (ImageInfoSpecific){ 5157 .type = IMAGE_INFO_SPECIFIC_KIND_QCOW2, 5158 .u.qcow2.data = g_new0(ImageInfoSpecificQCow2, 1), 5159 }; 5160 if (s->qcow_version == 2) { 5161 *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){ 5162 .compat = g_strdup("0.10"), 5163 .refcount_bits = s->refcount_bits, 5164 }; 5165 } else if (s->qcow_version == 3) { 5166 Qcow2BitmapInfoList *bitmaps; 5167 if (!qcow2_get_bitmap_info_list(bs, &bitmaps, errp)) { 5168 qapi_free_ImageInfoSpecific(spec_info); 5169 qapi_free_QCryptoBlockInfo(encrypt_info); 5170 return NULL; 5171 } 5172 *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){ 5173 .compat = g_strdup("1.1"), 5174 .lazy_refcounts = s->compatible_features & 5175 QCOW2_COMPAT_LAZY_REFCOUNTS, 5176 .has_lazy_refcounts = true, 5177 .corrupt = s->incompatible_features & 5178 QCOW2_INCOMPAT_CORRUPT, 5179 .has_corrupt = true, 5180 .has_extended_l2 = true, 5181 .extended_l2 = has_subclusters(s), 5182 .refcount_bits = s->refcount_bits, 5183 .has_bitmaps = !!bitmaps, 5184 .bitmaps = bitmaps, 5185 .has_data_file = !!s->image_data_file, 5186 .data_file = g_strdup(s->image_data_file), 5187 .has_data_file_raw = has_data_file(bs), 5188 .data_file_raw = data_file_is_raw(bs), 5189 .compression_type = s->compression_type, 5190 }; 5191 } else { 5192 /* if this assertion fails, this probably means a new version was 5193 * added without having it covered here */ 5194 assert(false); 5195 } 5196 5197 if (encrypt_info) { 5198 ImageInfoSpecificQCow2Encryption *qencrypt = 5199 g_new(ImageInfoSpecificQCow2Encryption, 1); 5200 switch (encrypt_info->format) { 5201 case Q_CRYPTO_BLOCK_FORMAT_QCOW: 5202 qencrypt->format = BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_AES; 5203 break; 5204 case Q_CRYPTO_BLOCK_FORMAT_LUKS: 5205 qencrypt->format = BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_LUKS; 5206 qencrypt->u.luks = encrypt_info->u.luks; 5207 break; 5208 default: 5209 abort(); 5210 } 5211 /* Since we did shallow copy above, erase any pointers 5212 * in the original info */ 5213 memset(&encrypt_info->u, 0, sizeof(encrypt_info->u)); 5214 qapi_free_QCryptoBlockInfo(encrypt_info); 5215 5216 spec_info->u.qcow2.data->has_encrypt = true; 5217 spec_info->u.qcow2.data->encrypt = qencrypt; 5218 } 5219 5220 return spec_info; 5221 } 5222 5223 static int qcow2_has_zero_init(BlockDriverState *bs) 5224 { 5225 BDRVQcow2State *s = bs->opaque; 5226 bool preallocated; 5227 5228 if (qemu_in_coroutine()) { 5229 qemu_co_mutex_lock(&s->lock); 5230 } 5231 /* 5232 * Check preallocation status: Preallocated images have all L2 5233 * tables allocated, nonpreallocated images have none. It is 5234 * therefore enough to check the first one. 5235 */ 5236 preallocated = s->l1_size > 0 && s->l1_table[0] != 0; 5237 if (qemu_in_coroutine()) { 5238 qemu_co_mutex_unlock(&s->lock); 5239 } 5240 5241 if (!preallocated) { 5242 return 1; 5243 } else if (bs->encrypted) { 5244 return 0; 5245 } else { 5246 return bdrv_has_zero_init(s->data_file->bs); 5247 } 5248 } 5249 5250 /* 5251 * Check the request to vmstate. On success return 5252 * qcow2_vm_state_offset(bs) + @pos 5253 */ 5254 static int64_t qcow2_check_vmstate_request(BlockDriverState *bs, 5255 QEMUIOVector *qiov, int64_t pos) 5256 { 5257 BDRVQcow2State *s = bs->opaque; 5258 int64_t vmstate_offset = qcow2_vm_state_offset(s); 5259 int ret; 5260 5261 /* Incoming requests must be OK */ 5262 bdrv_check_qiov_request(pos, qiov->size, qiov, 0, &error_abort); 5263 5264 if (INT64_MAX - pos < vmstate_offset) { 5265 return -EIO; 5266 } 5267 5268 pos += vmstate_offset; 5269 ret = bdrv_check_qiov_request(pos, qiov->size, qiov, 0, NULL); 5270 if (ret < 0) { 5271 return ret; 5272 } 5273 5274 return pos; 5275 } 5276 5277 static int qcow2_save_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, 5278 int64_t pos) 5279 { 5280 int64_t offset = qcow2_check_vmstate_request(bs, qiov, pos); 5281 if (offset < 0) { 5282 return offset; 5283 } 5284 5285 BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_SAVE); 5286 return bs->drv->bdrv_co_pwritev_part(bs, offset, qiov->size, qiov, 0, 0); 5287 } 5288 5289 static int qcow2_load_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, 5290 int64_t pos) 5291 { 5292 int64_t offset = qcow2_check_vmstate_request(bs, qiov, pos); 5293 if (offset < 0) { 5294 return offset; 5295 } 5296 5297 BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_LOAD); 5298 return bs->drv->bdrv_co_preadv_part(bs, offset, qiov->size, qiov, 0, 0); 5299 } 5300 5301 static int qcow2_has_compressed_clusters(BlockDriverState *bs) 5302 { 5303 int64_t offset = 0; 5304 int64_t bytes = bdrv_getlength(bs); 5305 5306 if (bytes < 0) { 5307 return bytes; 5308 } 5309 5310 while (bytes != 0) { 5311 int ret; 5312 QCow2SubclusterType type; 5313 unsigned int cur_bytes = MIN(INT_MAX, bytes); 5314 uint64_t host_offset; 5315 5316 ret = qcow2_get_host_offset(bs, offset, &cur_bytes, &host_offset, 5317 &type); 5318 if (ret < 0) { 5319 return ret; 5320 } 5321 5322 if (type == QCOW2_SUBCLUSTER_COMPRESSED) { 5323 return 1; 5324 } 5325 5326 offset += cur_bytes; 5327 bytes -= cur_bytes; 5328 } 5329 5330 return 0; 5331 } 5332 5333 /* 5334 * Downgrades an image's version. To achieve this, any incompatible features 5335 * have to be removed. 5336 */ 5337 static int qcow2_downgrade(BlockDriverState *bs, int target_version, 5338 BlockDriverAmendStatusCB *status_cb, void *cb_opaque, 5339 Error **errp) 5340 { 5341 BDRVQcow2State *s = bs->opaque; 5342 int current_version = s->qcow_version; 5343 int ret; 5344 int i; 5345 5346 /* This is qcow2_downgrade(), not qcow2_upgrade() */ 5347 assert(target_version < current_version); 5348 5349 /* There are no other versions (now) that you can downgrade to */ 5350 assert(target_version == 2); 5351 5352 if (s->refcount_order != 4) { 5353 error_setg(errp, "compat=0.10 requires refcount_bits=16"); 5354 return -ENOTSUP; 5355 } 5356 5357 if (has_data_file(bs)) { 5358 error_setg(errp, "Cannot downgrade an image with a data file"); 5359 return -ENOTSUP; 5360 } 5361 5362 /* 5363 * If any internal snapshot has a different size than the current 5364 * image size, or VM state size that exceeds 32 bits, downgrading 5365 * is unsafe. Even though we would still use v3-compliant output 5366 * to preserve that data, other v2 programs might not realize 5367 * those optional fields are important. 5368 */ 5369 for (i = 0; i < s->nb_snapshots; i++) { 5370 if (s->snapshots[i].vm_state_size > UINT32_MAX || 5371 s->snapshots[i].disk_size != bs->total_sectors * BDRV_SECTOR_SIZE) { 5372 error_setg(errp, "Internal snapshots prevent downgrade of image"); 5373 return -ENOTSUP; 5374 } 5375 } 5376 5377 /* clear incompatible features */ 5378 if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) { 5379 ret = qcow2_mark_clean(bs); 5380 if (ret < 0) { 5381 error_setg_errno(errp, -ret, "Failed to make the image clean"); 5382 return ret; 5383 } 5384 } 5385 5386 /* with QCOW2_INCOMPAT_CORRUPT, it is pretty much impossible to get here in 5387 * the first place; if that happens nonetheless, returning -ENOTSUP is the 5388 * best thing to do anyway */ 5389 5390 if (s->incompatible_features & ~QCOW2_INCOMPAT_COMPRESSION) { 5391 error_setg(errp, "Cannot downgrade an image with incompatible features " 5392 "0x%" PRIx64 " set", 5393 s->incompatible_features & ~QCOW2_INCOMPAT_COMPRESSION); 5394 return -ENOTSUP; 5395 } 5396 5397 /* since we can ignore compatible features, we can set them to 0 as well */ 5398 s->compatible_features = 0; 5399 /* if lazy refcounts have been used, they have already been fixed through 5400 * clearing the dirty flag */ 5401 5402 /* clearing autoclear features is trivial */ 5403 s->autoclear_features = 0; 5404 5405 ret = qcow2_expand_zero_clusters(bs, status_cb, cb_opaque); 5406 if (ret < 0) { 5407 error_setg_errno(errp, -ret, "Failed to turn zero into data clusters"); 5408 return ret; 5409 } 5410 5411 if (s->incompatible_features & QCOW2_INCOMPAT_COMPRESSION) { 5412 ret = qcow2_has_compressed_clusters(bs); 5413 if (ret < 0) { 5414 error_setg(errp, "Failed to check block status"); 5415 return -EINVAL; 5416 } 5417 if (ret) { 5418 error_setg(errp, "Cannot downgrade an image with zstd compression " 5419 "type and existing compressed clusters"); 5420 return -ENOTSUP; 5421 } 5422 /* 5423 * No compressed clusters for now, so just chose default zlib 5424 * compression. 5425 */ 5426 s->incompatible_features &= ~QCOW2_INCOMPAT_COMPRESSION; 5427 s->compression_type = QCOW2_COMPRESSION_TYPE_ZLIB; 5428 } 5429 5430 assert(s->incompatible_features == 0); 5431 5432 s->qcow_version = target_version; 5433 ret = qcow2_update_header(bs); 5434 if (ret < 0) { 5435 s->qcow_version = current_version; 5436 error_setg_errno(errp, -ret, "Failed to update the image header"); 5437 return ret; 5438 } 5439 return 0; 5440 } 5441 5442 /* 5443 * Upgrades an image's version. While newer versions encompass all 5444 * features of older versions, some things may have to be presented 5445 * differently. 5446 */ 5447 static int qcow2_upgrade(BlockDriverState *bs, int target_version, 5448 BlockDriverAmendStatusCB *status_cb, void *cb_opaque, 5449 Error **errp) 5450 { 5451 BDRVQcow2State *s = bs->opaque; 5452 bool need_snapshot_update; 5453 int current_version = s->qcow_version; 5454 int i; 5455 int ret; 5456 5457 /* This is qcow2_upgrade(), not qcow2_downgrade() */ 5458 assert(target_version > current_version); 5459 5460 /* There are no other versions (yet) that you can upgrade to */ 5461 assert(target_version == 3); 5462 5463 status_cb(bs, 0, 2, cb_opaque); 5464 5465 /* 5466 * In v2, snapshots do not need to have extra data. v3 requires 5467 * the 64-bit VM state size and the virtual disk size to be 5468 * present. 5469 * qcow2_write_snapshots() will always write the list in the 5470 * v3-compliant format. 5471 */ 5472 need_snapshot_update = false; 5473 for (i = 0; i < s->nb_snapshots; i++) { 5474 if (s->snapshots[i].extra_data_size < 5475 sizeof_field(QCowSnapshotExtraData, vm_state_size_large) + 5476 sizeof_field(QCowSnapshotExtraData, disk_size)) 5477 { 5478 need_snapshot_update = true; 5479 break; 5480 } 5481 } 5482 if (need_snapshot_update) { 5483 ret = qcow2_write_snapshots(bs); 5484 if (ret < 0) { 5485 error_setg_errno(errp, -ret, "Failed to update the snapshot table"); 5486 return ret; 5487 } 5488 } 5489 status_cb(bs, 1, 2, cb_opaque); 5490 5491 s->qcow_version = target_version; 5492 ret = qcow2_update_header(bs); 5493 if (ret < 0) { 5494 s->qcow_version = current_version; 5495 error_setg_errno(errp, -ret, "Failed to update the image header"); 5496 return ret; 5497 } 5498 status_cb(bs, 2, 2, cb_opaque); 5499 5500 return 0; 5501 } 5502 5503 typedef enum Qcow2AmendOperation { 5504 /* This is the value Qcow2AmendHelperCBInfo::last_operation will be 5505 * statically initialized to so that the helper CB can discern the first 5506 * invocation from an operation change */ 5507 QCOW2_NO_OPERATION = 0, 5508 5509 QCOW2_UPGRADING, 5510 QCOW2_UPDATING_ENCRYPTION, 5511 QCOW2_CHANGING_REFCOUNT_ORDER, 5512 QCOW2_DOWNGRADING, 5513 } Qcow2AmendOperation; 5514 5515 typedef struct Qcow2AmendHelperCBInfo { 5516 /* The code coordinating the amend operations should only modify 5517 * these four fields; the rest will be managed by the CB */ 5518 BlockDriverAmendStatusCB *original_status_cb; 5519 void *original_cb_opaque; 5520 5521 Qcow2AmendOperation current_operation; 5522 5523 /* Total number of operations to perform (only set once) */ 5524 int total_operations; 5525 5526 /* The following fields are managed by the CB */ 5527 5528 /* Number of operations completed */ 5529 int operations_completed; 5530 5531 /* Cumulative offset of all completed operations */ 5532 int64_t offset_completed; 5533 5534 Qcow2AmendOperation last_operation; 5535 int64_t last_work_size; 5536 } Qcow2AmendHelperCBInfo; 5537 5538 static void qcow2_amend_helper_cb(BlockDriverState *bs, 5539 int64_t operation_offset, 5540 int64_t operation_work_size, void *opaque) 5541 { 5542 Qcow2AmendHelperCBInfo *info = opaque; 5543 int64_t current_work_size; 5544 int64_t projected_work_size; 5545 5546 if (info->current_operation != info->last_operation) { 5547 if (info->last_operation != QCOW2_NO_OPERATION) { 5548 info->offset_completed += info->last_work_size; 5549 info->operations_completed++; 5550 } 5551 5552 info->last_operation = info->current_operation; 5553 } 5554 5555 assert(info->total_operations > 0); 5556 assert(info->operations_completed < info->total_operations); 5557 5558 info->last_work_size = operation_work_size; 5559 5560 current_work_size = info->offset_completed + operation_work_size; 5561 5562 /* current_work_size is the total work size for (operations_completed + 1) 5563 * operations (which includes this one), so multiply it by the number of 5564 * operations not covered and divide it by the number of operations 5565 * covered to get a projection for the operations not covered */ 5566 projected_work_size = current_work_size * (info->total_operations - 5567 info->operations_completed - 1) 5568 / (info->operations_completed + 1); 5569 5570 info->original_status_cb(bs, info->offset_completed + operation_offset, 5571 current_work_size + projected_work_size, 5572 info->original_cb_opaque); 5573 } 5574 5575 static int qcow2_amend_options(BlockDriverState *bs, QemuOpts *opts, 5576 BlockDriverAmendStatusCB *status_cb, 5577 void *cb_opaque, 5578 bool force, 5579 Error **errp) 5580 { 5581 BDRVQcow2State *s = bs->opaque; 5582 int old_version = s->qcow_version, new_version = old_version; 5583 uint64_t new_size = 0; 5584 const char *backing_file = NULL, *backing_format = NULL, *data_file = NULL; 5585 bool lazy_refcounts = s->use_lazy_refcounts; 5586 bool data_file_raw = data_file_is_raw(bs); 5587 const char *compat = NULL; 5588 int refcount_bits = s->refcount_bits; 5589 int ret; 5590 QemuOptDesc *desc = opts->list->desc; 5591 Qcow2AmendHelperCBInfo helper_cb_info; 5592 bool encryption_update = false; 5593 5594 while (desc && desc->name) { 5595 if (!qemu_opt_find(opts, desc->name)) { 5596 /* only change explicitly defined options */ 5597 desc++; 5598 continue; 5599 } 5600 5601 if (!strcmp(desc->name, BLOCK_OPT_COMPAT_LEVEL)) { 5602 compat = qemu_opt_get(opts, BLOCK_OPT_COMPAT_LEVEL); 5603 if (!compat) { 5604 /* preserve default */ 5605 } else if (!strcmp(compat, "0.10") || !strcmp(compat, "v2")) { 5606 new_version = 2; 5607 } else if (!strcmp(compat, "1.1") || !strcmp(compat, "v3")) { 5608 new_version = 3; 5609 } else { 5610 error_setg(errp, "Unknown compatibility level %s", compat); 5611 return -EINVAL; 5612 } 5613 } else if (!strcmp(desc->name, BLOCK_OPT_SIZE)) { 5614 new_size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, 0); 5615 } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FILE)) { 5616 backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE); 5617 } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FMT)) { 5618 backing_format = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT); 5619 } else if (g_str_has_prefix(desc->name, "encrypt.")) { 5620 if (!s->crypto) { 5621 error_setg(errp, 5622 "Can't amend encryption options - encryption not present"); 5623 return -EINVAL; 5624 } 5625 if (s->crypt_method_header != QCOW_CRYPT_LUKS) { 5626 error_setg(errp, 5627 "Only LUKS encryption options can be amended"); 5628 return -ENOTSUP; 5629 } 5630 encryption_update = true; 5631 } else if (!strcmp(desc->name, BLOCK_OPT_LAZY_REFCOUNTS)) { 5632 lazy_refcounts = qemu_opt_get_bool(opts, BLOCK_OPT_LAZY_REFCOUNTS, 5633 lazy_refcounts); 5634 } else if (!strcmp(desc->name, BLOCK_OPT_REFCOUNT_BITS)) { 5635 refcount_bits = qemu_opt_get_number(opts, BLOCK_OPT_REFCOUNT_BITS, 5636 refcount_bits); 5637 5638 if (refcount_bits <= 0 || refcount_bits > 64 || 5639 !is_power_of_2(refcount_bits)) 5640 { 5641 error_setg(errp, "Refcount width must be a power of two and " 5642 "may not exceed 64 bits"); 5643 return -EINVAL; 5644 } 5645 } else if (!strcmp(desc->name, BLOCK_OPT_DATA_FILE)) { 5646 data_file = qemu_opt_get(opts, BLOCK_OPT_DATA_FILE); 5647 if (data_file && !has_data_file(bs)) { 5648 error_setg(errp, "data-file can only be set for images that " 5649 "use an external data file"); 5650 return -EINVAL; 5651 } 5652 } else if (!strcmp(desc->name, BLOCK_OPT_DATA_FILE_RAW)) { 5653 data_file_raw = qemu_opt_get_bool(opts, BLOCK_OPT_DATA_FILE_RAW, 5654 data_file_raw); 5655 if (data_file_raw && !data_file_is_raw(bs)) { 5656 error_setg(errp, "data-file-raw cannot be set on existing " 5657 "images"); 5658 return -EINVAL; 5659 } 5660 } else { 5661 /* if this point is reached, this probably means a new option was 5662 * added without having it covered here */ 5663 abort(); 5664 } 5665 5666 desc++; 5667 } 5668 5669 helper_cb_info = (Qcow2AmendHelperCBInfo){ 5670 .original_status_cb = status_cb, 5671 .original_cb_opaque = cb_opaque, 5672 .total_operations = (new_version != old_version) 5673 + (s->refcount_bits != refcount_bits) + 5674 (encryption_update == true) 5675 }; 5676 5677 /* Upgrade first (some features may require compat=1.1) */ 5678 if (new_version > old_version) { 5679 helper_cb_info.current_operation = QCOW2_UPGRADING; 5680 ret = qcow2_upgrade(bs, new_version, &qcow2_amend_helper_cb, 5681 &helper_cb_info, errp); 5682 if (ret < 0) { 5683 return ret; 5684 } 5685 } 5686 5687 if (encryption_update) { 5688 QDict *amend_opts_dict; 5689 QCryptoBlockAmendOptions *amend_opts; 5690 5691 helper_cb_info.current_operation = QCOW2_UPDATING_ENCRYPTION; 5692 amend_opts_dict = qcow2_extract_crypto_opts(opts, "luks", errp); 5693 if (!amend_opts_dict) { 5694 return -EINVAL; 5695 } 5696 amend_opts = block_crypto_amend_opts_init(amend_opts_dict, errp); 5697 qobject_unref(amend_opts_dict); 5698 if (!amend_opts) { 5699 return -EINVAL; 5700 } 5701 ret = qcrypto_block_amend_options(s->crypto, 5702 qcow2_crypto_hdr_read_func, 5703 qcow2_crypto_hdr_write_func, 5704 bs, 5705 amend_opts, 5706 force, 5707 errp); 5708 qapi_free_QCryptoBlockAmendOptions(amend_opts); 5709 if (ret < 0) { 5710 return ret; 5711 } 5712 } 5713 5714 if (s->refcount_bits != refcount_bits) { 5715 int refcount_order = ctz32(refcount_bits); 5716 5717 if (new_version < 3 && refcount_bits != 16) { 5718 error_setg(errp, "Refcount widths other than 16 bits require " 5719 "compatibility level 1.1 or above (use compat=1.1 or " 5720 "greater)"); 5721 return -EINVAL; 5722 } 5723 5724 helper_cb_info.current_operation = QCOW2_CHANGING_REFCOUNT_ORDER; 5725 ret = qcow2_change_refcount_order(bs, refcount_order, 5726 &qcow2_amend_helper_cb, 5727 &helper_cb_info, errp); 5728 if (ret < 0) { 5729 return ret; 5730 } 5731 } 5732 5733 /* data-file-raw blocks backing files, so clear it first if requested */ 5734 if (data_file_raw) { 5735 s->autoclear_features |= QCOW2_AUTOCLEAR_DATA_FILE_RAW; 5736 } else { 5737 s->autoclear_features &= ~QCOW2_AUTOCLEAR_DATA_FILE_RAW; 5738 } 5739 5740 if (data_file) { 5741 g_free(s->image_data_file); 5742 s->image_data_file = *data_file ? g_strdup(data_file) : NULL; 5743 } 5744 5745 ret = qcow2_update_header(bs); 5746 if (ret < 0) { 5747 error_setg_errno(errp, -ret, "Failed to update the image header"); 5748 return ret; 5749 } 5750 5751 if (backing_file || backing_format) { 5752 if (g_strcmp0(backing_file, s->image_backing_file) || 5753 g_strcmp0(backing_format, s->image_backing_format)) { 5754 error_setg(errp, "Cannot amend the backing file"); 5755 error_append_hint(errp, 5756 "You can use 'qemu-img rebase' instead.\n"); 5757 return -EINVAL; 5758 } 5759 } 5760 5761 if (s->use_lazy_refcounts != lazy_refcounts) { 5762 if (lazy_refcounts) { 5763 if (new_version < 3) { 5764 error_setg(errp, "Lazy refcounts only supported with " 5765 "compatibility level 1.1 and above (use compat=1.1 " 5766 "or greater)"); 5767 return -EINVAL; 5768 } 5769 s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS; 5770 ret = qcow2_update_header(bs); 5771 if (ret < 0) { 5772 s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS; 5773 error_setg_errno(errp, -ret, "Failed to update the image header"); 5774 return ret; 5775 } 5776 s->use_lazy_refcounts = true; 5777 } else { 5778 /* make image clean first */ 5779 ret = qcow2_mark_clean(bs); 5780 if (ret < 0) { 5781 error_setg_errno(errp, -ret, "Failed to make the image clean"); 5782 return ret; 5783 } 5784 /* now disallow lazy refcounts */ 5785 s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS; 5786 ret = qcow2_update_header(bs); 5787 if (ret < 0) { 5788 s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS; 5789 error_setg_errno(errp, -ret, "Failed to update the image header"); 5790 return ret; 5791 } 5792 s->use_lazy_refcounts = false; 5793 } 5794 } 5795 5796 if (new_size) { 5797 BlockBackend *blk = blk_new_with_bs(bs, BLK_PERM_RESIZE, BLK_PERM_ALL, 5798 errp); 5799 if (!blk) { 5800 return -EPERM; 5801 } 5802 5803 /* 5804 * Amending image options should ensure that the image has 5805 * exactly the given new values, so pass exact=true here. 5806 */ 5807 ret = blk_truncate(blk, new_size, true, PREALLOC_MODE_OFF, 0, errp); 5808 blk_unref(blk); 5809 if (ret < 0) { 5810 return ret; 5811 } 5812 } 5813 5814 /* Downgrade last (so unsupported features can be removed before) */ 5815 if (new_version < old_version) { 5816 helper_cb_info.current_operation = QCOW2_DOWNGRADING; 5817 ret = qcow2_downgrade(bs, new_version, &qcow2_amend_helper_cb, 5818 &helper_cb_info, errp); 5819 if (ret < 0) { 5820 return ret; 5821 } 5822 } 5823 5824 return 0; 5825 } 5826 5827 static int coroutine_fn qcow2_co_amend(BlockDriverState *bs, 5828 BlockdevAmendOptions *opts, 5829 bool force, 5830 Error **errp) 5831 { 5832 BlockdevAmendOptionsQcow2 *qopts = &opts->u.qcow2; 5833 BDRVQcow2State *s = bs->opaque; 5834 int ret = 0; 5835 5836 if (qopts->has_encrypt) { 5837 if (!s->crypto) { 5838 error_setg(errp, "image is not encrypted, can't amend"); 5839 return -EOPNOTSUPP; 5840 } 5841 5842 if (qopts->encrypt->format != Q_CRYPTO_BLOCK_FORMAT_LUKS) { 5843 error_setg(errp, 5844 "Amend can't be used to change the qcow2 encryption format"); 5845 return -EOPNOTSUPP; 5846 } 5847 5848 if (s->crypt_method_header != QCOW_CRYPT_LUKS) { 5849 error_setg(errp, 5850 "Only LUKS encryption options can be amended for qcow2 with blockdev-amend"); 5851 return -EOPNOTSUPP; 5852 } 5853 5854 ret = qcrypto_block_amend_options(s->crypto, 5855 qcow2_crypto_hdr_read_func, 5856 qcow2_crypto_hdr_write_func, 5857 bs, 5858 qopts->encrypt, 5859 force, 5860 errp); 5861 } 5862 return ret; 5863 } 5864 5865 /* 5866 * If offset or size are negative, respectively, they will not be included in 5867 * the BLOCK_IMAGE_CORRUPTED event emitted. 5868 * fatal will be ignored for read-only BDS; corruptions found there will always 5869 * be considered non-fatal. 5870 */ 5871 void qcow2_signal_corruption(BlockDriverState *bs, bool fatal, int64_t offset, 5872 int64_t size, const char *message_format, ...) 5873 { 5874 BDRVQcow2State *s = bs->opaque; 5875 const char *node_name; 5876 char *message; 5877 va_list ap; 5878 5879 fatal = fatal && bdrv_is_writable(bs); 5880 5881 if (s->signaled_corruption && 5882 (!fatal || (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT))) 5883 { 5884 return; 5885 } 5886 5887 va_start(ap, message_format); 5888 message = g_strdup_vprintf(message_format, ap); 5889 va_end(ap); 5890 5891 if (fatal) { 5892 fprintf(stderr, "qcow2: Marking image as corrupt: %s; further " 5893 "corruption events will be suppressed\n", message); 5894 } else { 5895 fprintf(stderr, "qcow2: Image is corrupt: %s; further non-fatal " 5896 "corruption events will be suppressed\n", message); 5897 } 5898 5899 node_name = bdrv_get_node_name(bs); 5900 qapi_event_send_block_image_corrupted(bdrv_get_device_name(bs), 5901 *node_name != '\0', node_name, 5902 message, offset >= 0, offset, 5903 size >= 0, size, 5904 fatal); 5905 g_free(message); 5906 5907 if (fatal) { 5908 qcow2_mark_corrupt(bs); 5909 bs->drv = NULL; /* make BDS unusable */ 5910 } 5911 5912 s->signaled_corruption = true; 5913 } 5914 5915 #define QCOW_COMMON_OPTIONS \ 5916 { \ 5917 .name = BLOCK_OPT_SIZE, \ 5918 .type = QEMU_OPT_SIZE, \ 5919 .help = "Virtual disk size" \ 5920 }, \ 5921 { \ 5922 .name = BLOCK_OPT_COMPAT_LEVEL, \ 5923 .type = QEMU_OPT_STRING, \ 5924 .help = "Compatibility level (v2 [0.10] or v3 [1.1])" \ 5925 }, \ 5926 { \ 5927 .name = BLOCK_OPT_BACKING_FILE, \ 5928 .type = QEMU_OPT_STRING, \ 5929 .help = "File name of a base image" \ 5930 }, \ 5931 { \ 5932 .name = BLOCK_OPT_BACKING_FMT, \ 5933 .type = QEMU_OPT_STRING, \ 5934 .help = "Image format of the base image" \ 5935 }, \ 5936 { \ 5937 .name = BLOCK_OPT_DATA_FILE, \ 5938 .type = QEMU_OPT_STRING, \ 5939 .help = "File name of an external data file" \ 5940 }, \ 5941 { \ 5942 .name = BLOCK_OPT_DATA_FILE_RAW, \ 5943 .type = QEMU_OPT_BOOL, \ 5944 .help = "The external data file must stay valid " \ 5945 "as a raw image" \ 5946 }, \ 5947 { \ 5948 .name = BLOCK_OPT_LAZY_REFCOUNTS, \ 5949 .type = QEMU_OPT_BOOL, \ 5950 .help = "Postpone refcount updates", \ 5951 .def_value_str = "off" \ 5952 }, \ 5953 { \ 5954 .name = BLOCK_OPT_REFCOUNT_BITS, \ 5955 .type = QEMU_OPT_NUMBER, \ 5956 .help = "Width of a reference count entry in bits", \ 5957 .def_value_str = "16" \ 5958 } 5959 5960 static QemuOptsList qcow2_create_opts = { 5961 .name = "qcow2-create-opts", 5962 .head = QTAILQ_HEAD_INITIALIZER(qcow2_create_opts.head), 5963 .desc = { 5964 { \ 5965 .name = BLOCK_OPT_ENCRYPT, \ 5966 .type = QEMU_OPT_BOOL, \ 5967 .help = "Encrypt the image with format 'aes'. (Deprecated " \ 5968 "in favor of " BLOCK_OPT_ENCRYPT_FORMAT "=aes)", \ 5969 }, \ 5970 { \ 5971 .name = BLOCK_OPT_ENCRYPT_FORMAT, \ 5972 .type = QEMU_OPT_STRING, \ 5973 .help = "Encrypt the image, format choices: 'aes', 'luks'", \ 5974 }, \ 5975 BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.", \ 5976 "ID of secret providing qcow AES key or LUKS passphrase"), \ 5977 BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_ALG("encrypt."), \ 5978 BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_MODE("encrypt."), \ 5979 BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_ALG("encrypt."), \ 5980 BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_HASH_ALG("encrypt."), \ 5981 BLOCK_CRYPTO_OPT_DEF_LUKS_HASH_ALG("encrypt."), \ 5982 BLOCK_CRYPTO_OPT_DEF_LUKS_ITER_TIME("encrypt."), \ 5983 { \ 5984 .name = BLOCK_OPT_CLUSTER_SIZE, \ 5985 .type = QEMU_OPT_SIZE, \ 5986 .help = "qcow2 cluster size", \ 5987 .def_value_str = stringify(DEFAULT_CLUSTER_SIZE) \ 5988 }, \ 5989 { \ 5990 .name = BLOCK_OPT_EXTL2, \ 5991 .type = QEMU_OPT_BOOL, \ 5992 .help = "Extended L2 tables", \ 5993 .def_value_str = "off" \ 5994 }, \ 5995 { \ 5996 .name = BLOCK_OPT_PREALLOC, \ 5997 .type = QEMU_OPT_STRING, \ 5998 .help = "Preallocation mode (allowed values: off, " \ 5999 "metadata, falloc, full)" \ 6000 }, \ 6001 { \ 6002 .name = BLOCK_OPT_COMPRESSION_TYPE, \ 6003 .type = QEMU_OPT_STRING, \ 6004 .help = "Compression method used for image cluster " \ 6005 "compression", \ 6006 .def_value_str = "zlib" \ 6007 }, 6008 QCOW_COMMON_OPTIONS, 6009 { /* end of list */ } 6010 } 6011 }; 6012 6013 static QemuOptsList qcow2_amend_opts = { 6014 .name = "qcow2-amend-opts", 6015 .head = QTAILQ_HEAD_INITIALIZER(qcow2_amend_opts.head), 6016 .desc = { 6017 BLOCK_CRYPTO_OPT_DEF_LUKS_STATE("encrypt."), 6018 BLOCK_CRYPTO_OPT_DEF_LUKS_KEYSLOT("encrypt."), 6019 BLOCK_CRYPTO_OPT_DEF_LUKS_OLD_SECRET("encrypt."), 6020 BLOCK_CRYPTO_OPT_DEF_LUKS_NEW_SECRET("encrypt."), 6021 BLOCK_CRYPTO_OPT_DEF_LUKS_ITER_TIME("encrypt."), 6022 QCOW_COMMON_OPTIONS, 6023 { /* end of list */ } 6024 } 6025 }; 6026 6027 static const char *const qcow2_strong_runtime_opts[] = { 6028 "encrypt." BLOCK_CRYPTO_OPT_QCOW_KEY_SECRET, 6029 6030 NULL 6031 }; 6032 6033 BlockDriver bdrv_qcow2 = { 6034 .format_name = "qcow2", 6035 .instance_size = sizeof(BDRVQcow2State), 6036 .bdrv_probe = qcow2_probe, 6037 .bdrv_open = qcow2_open, 6038 .bdrv_close = qcow2_close, 6039 .bdrv_reopen_prepare = qcow2_reopen_prepare, 6040 .bdrv_reopen_commit = qcow2_reopen_commit, 6041 .bdrv_reopen_commit_post = qcow2_reopen_commit_post, 6042 .bdrv_reopen_abort = qcow2_reopen_abort, 6043 .bdrv_join_options = qcow2_join_options, 6044 .bdrv_child_perm = bdrv_default_perms, 6045 .bdrv_co_create_opts = qcow2_co_create_opts, 6046 .bdrv_co_create = qcow2_co_create, 6047 .bdrv_has_zero_init = qcow2_has_zero_init, 6048 .bdrv_co_block_status = qcow2_co_block_status, 6049 6050 .bdrv_co_preadv_part = qcow2_co_preadv_part, 6051 .bdrv_co_pwritev_part = qcow2_co_pwritev_part, 6052 .bdrv_co_flush_to_os = qcow2_co_flush_to_os, 6053 6054 .bdrv_co_pwrite_zeroes = qcow2_co_pwrite_zeroes, 6055 .bdrv_co_pdiscard = qcow2_co_pdiscard, 6056 .bdrv_co_copy_range_from = qcow2_co_copy_range_from, 6057 .bdrv_co_copy_range_to = qcow2_co_copy_range_to, 6058 .bdrv_co_truncate = qcow2_co_truncate, 6059 .bdrv_co_pwritev_compressed_part = qcow2_co_pwritev_compressed_part, 6060 .bdrv_make_empty = qcow2_make_empty, 6061 6062 .bdrv_snapshot_create = qcow2_snapshot_create, 6063 .bdrv_snapshot_goto = qcow2_snapshot_goto, 6064 .bdrv_snapshot_delete = qcow2_snapshot_delete, 6065 .bdrv_snapshot_list = qcow2_snapshot_list, 6066 .bdrv_snapshot_load_tmp = qcow2_snapshot_load_tmp, 6067 .bdrv_measure = qcow2_measure, 6068 .bdrv_get_info = qcow2_get_info, 6069 .bdrv_get_specific_info = qcow2_get_specific_info, 6070 6071 .bdrv_save_vmstate = qcow2_save_vmstate, 6072 .bdrv_load_vmstate = qcow2_load_vmstate, 6073 6074 .is_format = true, 6075 .supports_backing = true, 6076 .bdrv_change_backing_file = qcow2_change_backing_file, 6077 6078 .bdrv_refresh_limits = qcow2_refresh_limits, 6079 .bdrv_co_invalidate_cache = qcow2_co_invalidate_cache, 6080 .bdrv_inactivate = qcow2_inactivate, 6081 6082 .create_opts = &qcow2_create_opts, 6083 .amend_opts = &qcow2_amend_opts, 6084 .strong_runtime_opts = qcow2_strong_runtime_opts, 6085 .mutable_opts = mutable_opts, 6086 .bdrv_co_check = qcow2_co_check, 6087 .bdrv_amend_options = qcow2_amend_options, 6088 .bdrv_co_amend = qcow2_co_amend, 6089 6090 .bdrv_detach_aio_context = qcow2_detach_aio_context, 6091 .bdrv_attach_aio_context = qcow2_attach_aio_context, 6092 6093 .bdrv_supports_persistent_dirty_bitmap = 6094 qcow2_supports_persistent_dirty_bitmap, 6095 .bdrv_co_can_store_new_dirty_bitmap = qcow2_co_can_store_new_dirty_bitmap, 6096 .bdrv_co_remove_persistent_dirty_bitmap = 6097 qcow2_co_remove_persistent_dirty_bitmap, 6098 }; 6099 6100 static void bdrv_qcow2_init(void) 6101 { 6102 bdrv_register(&bdrv_qcow2); 6103 } 6104 6105 block_init(bdrv_qcow2_init); 6106