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 "qapi/qobject-input-visitor.h" 42 #include "qapi/qapi-visit-block-core.h" 43 #include "crypto.h" 44 #include "block/aio_task.h" 45 46 /* 47 Differences with QCOW: 48 49 - Support for multiple incremental snapshots. 50 - Memory management by reference counts. 51 - Clusters which have a reference count of one have the bit 52 QCOW_OFLAG_COPIED to optimize write performance. 53 - Size of compressed clusters is stored in sectors to reduce bit usage 54 in the cluster offsets. 55 - Support for storing additional data (such as the VM state) in the 56 snapshots. 57 - If a backing store is used, the cluster size is not constrained 58 (could be backported to QCOW). 59 - L2 tables have always a size of one cluster. 60 */ 61 62 63 typedef struct { 64 uint32_t magic; 65 uint32_t len; 66 } QEMU_PACKED QCowExtension; 67 68 #define QCOW2_EXT_MAGIC_END 0 69 #define QCOW2_EXT_MAGIC_BACKING_FORMAT 0xe2792aca 70 #define QCOW2_EXT_MAGIC_FEATURE_TABLE 0x6803f857 71 #define QCOW2_EXT_MAGIC_CRYPTO_HEADER 0x0537be77 72 #define QCOW2_EXT_MAGIC_BITMAPS 0x23852875 73 #define QCOW2_EXT_MAGIC_DATA_FILE 0x44415441 74 75 static int coroutine_fn 76 qcow2_co_preadv_compressed(BlockDriverState *bs, 77 uint64_t cluster_descriptor, 78 uint64_t offset, 79 uint64_t bytes, 80 QEMUIOVector *qiov, 81 size_t qiov_offset); 82 83 static int qcow2_probe(const uint8_t *buf, int buf_size, const char *filename) 84 { 85 const QCowHeader *cow_header = (const void *)buf; 86 87 if (buf_size >= sizeof(QCowHeader) && 88 be32_to_cpu(cow_header->magic) == QCOW_MAGIC && 89 be32_to_cpu(cow_header->version) >= 2) 90 return 100; 91 else 92 return 0; 93 } 94 95 96 static ssize_t qcow2_crypto_hdr_read_func(QCryptoBlock *block, size_t offset, 97 uint8_t *buf, size_t buflen, 98 void *opaque, Error **errp) 99 { 100 BlockDriverState *bs = opaque; 101 BDRVQcow2State *s = bs->opaque; 102 ssize_t ret; 103 104 if ((offset + buflen) > s->crypto_header.length) { 105 error_setg(errp, "Request for data outside of extension header"); 106 return -1; 107 } 108 109 ret = bdrv_pread(bs->file, 110 s->crypto_header.offset + offset, buf, buflen); 111 if (ret < 0) { 112 error_setg_errno(errp, -ret, "Could not read encryption header"); 113 return -1; 114 } 115 return ret; 116 } 117 118 119 static ssize_t qcow2_crypto_hdr_init_func(QCryptoBlock *block, size_t headerlen, 120 void *opaque, Error **errp) 121 { 122 BlockDriverState *bs = opaque; 123 BDRVQcow2State *s = bs->opaque; 124 int64_t ret; 125 int64_t clusterlen; 126 127 ret = qcow2_alloc_clusters(bs, headerlen); 128 if (ret < 0) { 129 error_setg_errno(errp, -ret, 130 "Cannot allocate cluster for LUKS header size %zu", 131 headerlen); 132 return -1; 133 } 134 135 s->crypto_header.length = headerlen; 136 s->crypto_header.offset = ret; 137 138 /* 139 * Zero fill all space in cluster so it has predictable 140 * content, as we may not initialize some regions of the 141 * header (eg only 1 out of 8 key slots will be initialized) 142 */ 143 clusterlen = size_to_clusters(s, headerlen) * s->cluster_size; 144 assert(qcow2_pre_write_overlap_check(bs, 0, ret, clusterlen, false) == 0); 145 ret = bdrv_pwrite_zeroes(bs->file, 146 ret, 147 clusterlen, 0); 148 if (ret < 0) { 149 error_setg_errno(errp, -ret, "Could not zero fill encryption header"); 150 return -1; 151 } 152 153 return ret; 154 } 155 156 157 static ssize_t qcow2_crypto_hdr_write_func(QCryptoBlock *block, size_t offset, 158 const uint8_t *buf, size_t buflen, 159 void *opaque, Error **errp) 160 { 161 BlockDriverState *bs = opaque; 162 BDRVQcow2State *s = bs->opaque; 163 ssize_t ret; 164 165 if ((offset + buflen) > s->crypto_header.length) { 166 error_setg(errp, "Request for data outside of extension header"); 167 return -1; 168 } 169 170 ret = bdrv_pwrite(bs->file, 171 s->crypto_header.offset + offset, buf, buflen); 172 if (ret < 0) { 173 error_setg_errno(errp, -ret, "Could not read encryption header"); 174 return -1; 175 } 176 return ret; 177 } 178 179 static QDict* 180 qcow2_extract_crypto_opts(QemuOpts *opts, const char *fmt, Error **errp) 181 { 182 QDict *cryptoopts_qdict; 183 QDict *opts_qdict; 184 185 /* Extract "encrypt." options into a qdict */ 186 opts_qdict = qemu_opts_to_qdict(opts, NULL); 187 qdict_extract_subqdict(opts_qdict, &cryptoopts_qdict, "encrypt."); 188 qobject_unref(opts_qdict); 189 qdict_put_str(cryptoopts_qdict, "format", fmt); 190 return cryptoopts_qdict; 191 } 192 193 /* 194 * read qcow2 extension and fill bs 195 * start reading from start_offset 196 * finish reading upon magic of value 0 or when end_offset reached 197 * unknown magic is skipped (future extension this version knows nothing about) 198 * return 0 upon success, non-0 otherwise 199 */ 200 static int qcow2_read_extensions(BlockDriverState *bs, uint64_t start_offset, 201 uint64_t end_offset, void **p_feature_table, 202 int flags, bool *need_update_header, 203 Error **errp) 204 { 205 BDRVQcow2State *s = bs->opaque; 206 QCowExtension ext; 207 uint64_t offset; 208 int ret; 209 Qcow2BitmapHeaderExt bitmaps_ext; 210 211 if (need_update_header != NULL) { 212 *need_update_header = false; 213 } 214 215 #ifdef DEBUG_EXT 216 printf("qcow2_read_extensions: start=%ld end=%ld\n", start_offset, end_offset); 217 #endif 218 offset = start_offset; 219 while (offset < end_offset) { 220 221 #ifdef DEBUG_EXT 222 /* Sanity check */ 223 if (offset > s->cluster_size) 224 printf("qcow2_read_extension: suspicious offset %lu\n", offset); 225 226 printf("attempting to read extended header in offset %lu\n", offset); 227 #endif 228 229 ret = bdrv_pread(bs->file, offset, &ext, sizeof(ext)); 230 if (ret < 0) { 231 error_setg_errno(errp, -ret, "qcow2_read_extension: ERROR: " 232 "pread fail from offset %" PRIu64, offset); 233 return 1; 234 } 235 ext.magic = be32_to_cpu(ext.magic); 236 ext.len = be32_to_cpu(ext.len); 237 offset += sizeof(ext); 238 #ifdef DEBUG_EXT 239 printf("ext.magic = 0x%x\n", ext.magic); 240 #endif 241 if (offset > end_offset || ext.len > end_offset - offset) { 242 error_setg(errp, "Header extension too large"); 243 return -EINVAL; 244 } 245 246 switch (ext.magic) { 247 case QCOW2_EXT_MAGIC_END: 248 return 0; 249 250 case QCOW2_EXT_MAGIC_BACKING_FORMAT: 251 if (ext.len >= sizeof(bs->backing_format)) { 252 error_setg(errp, "ERROR: ext_backing_format: len=%" PRIu32 253 " too large (>=%zu)", ext.len, 254 sizeof(bs->backing_format)); 255 return 2; 256 } 257 ret = bdrv_pread(bs->file, offset, bs->backing_format, ext.len); 258 if (ret < 0) { 259 error_setg_errno(errp, -ret, "ERROR: ext_backing_format: " 260 "Could not read format name"); 261 return 3; 262 } 263 bs->backing_format[ext.len] = '\0'; 264 s->image_backing_format = g_strdup(bs->backing_format); 265 #ifdef DEBUG_EXT 266 printf("Qcow2: Got format extension %s\n", bs->backing_format); 267 #endif 268 break; 269 270 case QCOW2_EXT_MAGIC_FEATURE_TABLE: 271 if (p_feature_table != NULL) { 272 void *feature_table = g_malloc0(ext.len + 2 * sizeof(Qcow2Feature)); 273 ret = bdrv_pread(bs->file, offset , feature_table, ext.len); 274 if (ret < 0) { 275 error_setg_errno(errp, -ret, "ERROR: ext_feature_table: " 276 "Could not read table"); 277 return ret; 278 } 279 280 *p_feature_table = feature_table; 281 } 282 break; 283 284 case QCOW2_EXT_MAGIC_CRYPTO_HEADER: { 285 unsigned int cflags = 0; 286 if (s->crypt_method_header != QCOW_CRYPT_LUKS) { 287 error_setg(errp, "CRYPTO header extension only " 288 "expected with LUKS encryption method"); 289 return -EINVAL; 290 } 291 if (ext.len != sizeof(Qcow2CryptoHeaderExtension)) { 292 error_setg(errp, "CRYPTO header extension size %u, " 293 "but expected size %zu", ext.len, 294 sizeof(Qcow2CryptoHeaderExtension)); 295 return -EINVAL; 296 } 297 298 ret = bdrv_pread(bs->file, offset, &s->crypto_header, ext.len); 299 if (ret < 0) { 300 error_setg_errno(errp, -ret, 301 "Unable to read CRYPTO header extension"); 302 return ret; 303 } 304 s->crypto_header.offset = be64_to_cpu(s->crypto_header.offset); 305 s->crypto_header.length = be64_to_cpu(s->crypto_header.length); 306 307 if ((s->crypto_header.offset % s->cluster_size) != 0) { 308 error_setg(errp, "Encryption header offset '%" PRIu64 "' is " 309 "not a multiple of cluster size '%u'", 310 s->crypto_header.offset, s->cluster_size); 311 return -EINVAL; 312 } 313 314 if (flags & BDRV_O_NO_IO) { 315 cflags |= QCRYPTO_BLOCK_OPEN_NO_IO; 316 } 317 s->crypto = qcrypto_block_open(s->crypto_opts, "encrypt.", 318 qcow2_crypto_hdr_read_func, 319 bs, cflags, QCOW2_MAX_THREADS, errp); 320 if (!s->crypto) { 321 return -EINVAL; 322 } 323 } break; 324 325 case QCOW2_EXT_MAGIC_BITMAPS: 326 if (ext.len != sizeof(bitmaps_ext)) { 327 error_setg_errno(errp, -ret, "bitmaps_ext: " 328 "Invalid extension length"); 329 return -EINVAL; 330 } 331 332 if (!(s->autoclear_features & QCOW2_AUTOCLEAR_BITMAPS)) { 333 if (s->qcow_version < 3) { 334 /* Let's be a bit more specific */ 335 warn_report("This qcow2 v2 image contains bitmaps, but " 336 "they may have been modified by a program " 337 "without persistent bitmap support; so now " 338 "they must all be considered inconsistent"); 339 } else { 340 warn_report("a program lacking bitmap support " 341 "modified this file, so all bitmaps are now " 342 "considered inconsistent"); 343 } 344 error_printf("Some clusters may be leaked, " 345 "run 'qemu-img check -r' on the image " 346 "file to fix."); 347 if (need_update_header != NULL) { 348 /* Updating is needed to drop invalid bitmap extension. */ 349 *need_update_header = true; 350 } 351 break; 352 } 353 354 ret = bdrv_pread(bs->file, offset, &bitmaps_ext, ext.len); 355 if (ret < 0) { 356 error_setg_errno(errp, -ret, "bitmaps_ext: " 357 "Could not read ext header"); 358 return ret; 359 } 360 361 if (bitmaps_ext.reserved32 != 0) { 362 error_setg_errno(errp, -ret, "bitmaps_ext: " 363 "Reserved field is not zero"); 364 return -EINVAL; 365 } 366 367 bitmaps_ext.nb_bitmaps = be32_to_cpu(bitmaps_ext.nb_bitmaps); 368 bitmaps_ext.bitmap_directory_size = 369 be64_to_cpu(bitmaps_ext.bitmap_directory_size); 370 bitmaps_ext.bitmap_directory_offset = 371 be64_to_cpu(bitmaps_ext.bitmap_directory_offset); 372 373 if (bitmaps_ext.nb_bitmaps > QCOW2_MAX_BITMAPS) { 374 error_setg(errp, 375 "bitmaps_ext: Image has %" PRIu32 " bitmaps, " 376 "exceeding the QEMU supported maximum of %d", 377 bitmaps_ext.nb_bitmaps, QCOW2_MAX_BITMAPS); 378 return -EINVAL; 379 } 380 381 if (bitmaps_ext.nb_bitmaps == 0) { 382 error_setg(errp, "found bitmaps extension with zero bitmaps"); 383 return -EINVAL; 384 } 385 386 if (offset_into_cluster(s, bitmaps_ext.bitmap_directory_offset)) { 387 error_setg(errp, "bitmaps_ext: " 388 "invalid bitmap directory offset"); 389 return -EINVAL; 390 } 391 392 if (bitmaps_ext.bitmap_directory_size > 393 QCOW2_MAX_BITMAP_DIRECTORY_SIZE) { 394 error_setg(errp, "bitmaps_ext: " 395 "bitmap directory size (%" PRIu64 ") exceeds " 396 "the maximum supported size (%d)", 397 bitmaps_ext.bitmap_directory_size, 398 QCOW2_MAX_BITMAP_DIRECTORY_SIZE); 399 return -EINVAL; 400 } 401 402 s->nb_bitmaps = bitmaps_ext.nb_bitmaps; 403 s->bitmap_directory_offset = 404 bitmaps_ext.bitmap_directory_offset; 405 s->bitmap_directory_size = 406 bitmaps_ext.bitmap_directory_size; 407 408 #ifdef DEBUG_EXT 409 printf("Qcow2: Got bitmaps extension: " 410 "offset=%" PRIu64 " nb_bitmaps=%" PRIu32 "\n", 411 s->bitmap_directory_offset, s->nb_bitmaps); 412 #endif 413 break; 414 415 case QCOW2_EXT_MAGIC_DATA_FILE: 416 { 417 s->image_data_file = g_malloc0(ext.len + 1); 418 ret = bdrv_pread(bs->file, offset, s->image_data_file, ext.len); 419 if (ret < 0) { 420 error_setg_errno(errp, -ret, 421 "ERROR: Could not read data file name"); 422 return ret; 423 } 424 #ifdef DEBUG_EXT 425 printf("Qcow2: Got external data file %s\n", s->image_data_file); 426 #endif 427 break; 428 } 429 430 default: 431 /* unknown magic - save it in case we need to rewrite the header */ 432 /* If you add a new feature, make sure to also update the fast 433 * path of qcow2_make_empty() to deal with it. */ 434 { 435 Qcow2UnknownHeaderExtension *uext; 436 437 uext = g_malloc0(sizeof(*uext) + ext.len); 438 uext->magic = ext.magic; 439 uext->len = ext.len; 440 QLIST_INSERT_HEAD(&s->unknown_header_ext, uext, next); 441 442 ret = bdrv_pread(bs->file, offset , uext->data, uext->len); 443 if (ret < 0) { 444 error_setg_errno(errp, -ret, "ERROR: unknown extension: " 445 "Could not read data"); 446 return ret; 447 } 448 } 449 break; 450 } 451 452 offset += ((ext.len + 7) & ~7); 453 } 454 455 return 0; 456 } 457 458 static void cleanup_unknown_header_ext(BlockDriverState *bs) 459 { 460 BDRVQcow2State *s = bs->opaque; 461 Qcow2UnknownHeaderExtension *uext, *next; 462 463 QLIST_FOREACH_SAFE(uext, &s->unknown_header_ext, next, next) { 464 QLIST_REMOVE(uext, next); 465 g_free(uext); 466 } 467 } 468 469 static void report_unsupported_feature(Error **errp, Qcow2Feature *table, 470 uint64_t mask) 471 { 472 g_autoptr(GString) features = g_string_sized_new(60); 473 474 while (table && table->name[0] != '\0') { 475 if (table->type == QCOW2_FEAT_TYPE_INCOMPATIBLE) { 476 if (mask & (1ULL << table->bit)) { 477 if (features->len > 0) { 478 g_string_append(features, ", "); 479 } 480 g_string_append_printf(features, "%.46s", table->name); 481 mask &= ~(1ULL << table->bit); 482 } 483 } 484 table++; 485 } 486 487 if (mask) { 488 if (features->len > 0) { 489 g_string_append(features, ", "); 490 } 491 g_string_append_printf(features, 492 "Unknown incompatible feature: %" PRIx64, mask); 493 } 494 495 error_setg(errp, "Unsupported qcow2 feature(s): %s", features->str); 496 } 497 498 /* 499 * Sets the dirty bit and flushes afterwards if necessary. 500 * 501 * The incompatible_features bit is only set if the image file header was 502 * updated successfully. Therefore it is not required to check the return 503 * value of this function. 504 */ 505 int qcow2_mark_dirty(BlockDriverState *bs) 506 { 507 BDRVQcow2State *s = bs->opaque; 508 uint64_t val; 509 int ret; 510 511 assert(s->qcow_version >= 3); 512 513 if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) { 514 return 0; /* already dirty */ 515 } 516 517 val = cpu_to_be64(s->incompatible_features | QCOW2_INCOMPAT_DIRTY); 518 ret = bdrv_pwrite(bs->file, offsetof(QCowHeader, incompatible_features), 519 &val, sizeof(val)); 520 if (ret < 0) { 521 return ret; 522 } 523 ret = bdrv_flush(bs->file->bs); 524 if (ret < 0) { 525 return ret; 526 } 527 528 /* Only treat image as dirty if the header was updated successfully */ 529 s->incompatible_features |= QCOW2_INCOMPAT_DIRTY; 530 return 0; 531 } 532 533 /* 534 * Clears the dirty bit and flushes before if necessary. Only call this 535 * function when there are no pending requests, it does not guard against 536 * concurrent requests dirtying the image. 537 */ 538 static int qcow2_mark_clean(BlockDriverState *bs) 539 { 540 BDRVQcow2State *s = bs->opaque; 541 542 if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) { 543 int ret; 544 545 s->incompatible_features &= ~QCOW2_INCOMPAT_DIRTY; 546 547 ret = qcow2_flush_caches(bs); 548 if (ret < 0) { 549 return ret; 550 } 551 552 return qcow2_update_header(bs); 553 } 554 return 0; 555 } 556 557 /* 558 * Marks the image as corrupt. 559 */ 560 int qcow2_mark_corrupt(BlockDriverState *bs) 561 { 562 BDRVQcow2State *s = bs->opaque; 563 564 s->incompatible_features |= QCOW2_INCOMPAT_CORRUPT; 565 return qcow2_update_header(bs); 566 } 567 568 /* 569 * Marks the image as consistent, i.e., unsets the corrupt bit, and flushes 570 * before if necessary. 571 */ 572 int qcow2_mark_consistent(BlockDriverState *bs) 573 { 574 BDRVQcow2State *s = bs->opaque; 575 576 if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) { 577 int ret = qcow2_flush_caches(bs); 578 if (ret < 0) { 579 return ret; 580 } 581 582 s->incompatible_features &= ~QCOW2_INCOMPAT_CORRUPT; 583 return qcow2_update_header(bs); 584 } 585 return 0; 586 } 587 588 static void qcow2_add_check_result(BdrvCheckResult *out, 589 const BdrvCheckResult *src, 590 bool set_allocation_info) 591 { 592 out->corruptions += src->corruptions; 593 out->leaks += src->leaks; 594 out->check_errors += src->check_errors; 595 out->corruptions_fixed += src->corruptions_fixed; 596 out->leaks_fixed += src->leaks_fixed; 597 598 if (set_allocation_info) { 599 out->image_end_offset = src->image_end_offset; 600 out->bfi = src->bfi; 601 } 602 } 603 604 static int coroutine_fn qcow2_co_check_locked(BlockDriverState *bs, 605 BdrvCheckResult *result, 606 BdrvCheckMode fix) 607 { 608 BdrvCheckResult snapshot_res = {}; 609 BdrvCheckResult refcount_res = {}; 610 int ret; 611 612 memset(result, 0, sizeof(*result)); 613 614 ret = qcow2_check_read_snapshot_table(bs, &snapshot_res, fix); 615 if (ret < 0) { 616 qcow2_add_check_result(result, &snapshot_res, false); 617 return ret; 618 } 619 620 ret = qcow2_check_refcounts(bs, &refcount_res, fix); 621 qcow2_add_check_result(result, &refcount_res, true); 622 if (ret < 0) { 623 qcow2_add_check_result(result, &snapshot_res, false); 624 return ret; 625 } 626 627 ret = qcow2_check_fix_snapshot_table(bs, &snapshot_res, fix); 628 qcow2_add_check_result(result, &snapshot_res, false); 629 if (ret < 0) { 630 return ret; 631 } 632 633 if (fix && result->check_errors == 0 && result->corruptions == 0) { 634 ret = qcow2_mark_clean(bs); 635 if (ret < 0) { 636 return ret; 637 } 638 return qcow2_mark_consistent(bs); 639 } 640 return ret; 641 } 642 643 static int coroutine_fn qcow2_co_check(BlockDriverState *bs, 644 BdrvCheckResult *result, 645 BdrvCheckMode fix) 646 { 647 BDRVQcow2State *s = bs->opaque; 648 int ret; 649 650 qemu_co_mutex_lock(&s->lock); 651 ret = qcow2_co_check_locked(bs, result, fix); 652 qemu_co_mutex_unlock(&s->lock); 653 return ret; 654 } 655 656 int qcow2_validate_table(BlockDriverState *bs, uint64_t offset, 657 uint64_t entries, size_t entry_len, 658 int64_t max_size_bytes, const char *table_name, 659 Error **errp) 660 { 661 BDRVQcow2State *s = bs->opaque; 662 663 if (entries > max_size_bytes / entry_len) { 664 error_setg(errp, "%s too large", table_name); 665 return -EFBIG; 666 } 667 668 /* Use signed INT64_MAX as the maximum even for uint64_t header fields, 669 * because values will be passed to qemu functions taking int64_t. */ 670 if ((INT64_MAX - entries * entry_len < offset) || 671 (offset_into_cluster(s, offset) != 0)) { 672 error_setg(errp, "%s offset invalid", table_name); 673 return -EINVAL; 674 } 675 676 return 0; 677 } 678 679 static const char *const mutable_opts[] = { 680 QCOW2_OPT_LAZY_REFCOUNTS, 681 QCOW2_OPT_DISCARD_REQUEST, 682 QCOW2_OPT_DISCARD_SNAPSHOT, 683 QCOW2_OPT_DISCARD_OTHER, 684 QCOW2_OPT_OVERLAP, 685 QCOW2_OPT_OVERLAP_TEMPLATE, 686 QCOW2_OPT_OVERLAP_MAIN_HEADER, 687 QCOW2_OPT_OVERLAP_ACTIVE_L1, 688 QCOW2_OPT_OVERLAP_ACTIVE_L2, 689 QCOW2_OPT_OVERLAP_REFCOUNT_TABLE, 690 QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK, 691 QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE, 692 QCOW2_OPT_OVERLAP_INACTIVE_L1, 693 QCOW2_OPT_OVERLAP_INACTIVE_L2, 694 QCOW2_OPT_OVERLAP_BITMAP_DIRECTORY, 695 QCOW2_OPT_CACHE_SIZE, 696 QCOW2_OPT_L2_CACHE_SIZE, 697 QCOW2_OPT_L2_CACHE_ENTRY_SIZE, 698 QCOW2_OPT_REFCOUNT_CACHE_SIZE, 699 QCOW2_OPT_CACHE_CLEAN_INTERVAL, 700 NULL 701 }; 702 703 static QemuOptsList qcow2_runtime_opts = { 704 .name = "qcow2", 705 .head = QTAILQ_HEAD_INITIALIZER(qcow2_runtime_opts.head), 706 .desc = { 707 { 708 .name = QCOW2_OPT_LAZY_REFCOUNTS, 709 .type = QEMU_OPT_BOOL, 710 .help = "Postpone refcount updates", 711 }, 712 { 713 .name = QCOW2_OPT_DISCARD_REQUEST, 714 .type = QEMU_OPT_BOOL, 715 .help = "Pass guest discard requests to the layer below", 716 }, 717 { 718 .name = QCOW2_OPT_DISCARD_SNAPSHOT, 719 .type = QEMU_OPT_BOOL, 720 .help = "Generate discard requests when snapshot related space " 721 "is freed", 722 }, 723 { 724 .name = QCOW2_OPT_DISCARD_OTHER, 725 .type = QEMU_OPT_BOOL, 726 .help = "Generate discard requests when other clusters are freed", 727 }, 728 { 729 .name = QCOW2_OPT_OVERLAP, 730 .type = QEMU_OPT_STRING, 731 .help = "Selects which overlap checks to perform from a range of " 732 "templates (none, constant, cached, all)", 733 }, 734 { 735 .name = QCOW2_OPT_OVERLAP_TEMPLATE, 736 .type = QEMU_OPT_STRING, 737 .help = "Selects which overlap checks to perform from a range of " 738 "templates (none, constant, cached, all)", 739 }, 740 { 741 .name = QCOW2_OPT_OVERLAP_MAIN_HEADER, 742 .type = QEMU_OPT_BOOL, 743 .help = "Check for unintended writes into the main qcow2 header", 744 }, 745 { 746 .name = QCOW2_OPT_OVERLAP_ACTIVE_L1, 747 .type = QEMU_OPT_BOOL, 748 .help = "Check for unintended writes into the active L1 table", 749 }, 750 { 751 .name = QCOW2_OPT_OVERLAP_ACTIVE_L2, 752 .type = QEMU_OPT_BOOL, 753 .help = "Check for unintended writes into an active L2 table", 754 }, 755 { 756 .name = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE, 757 .type = QEMU_OPT_BOOL, 758 .help = "Check for unintended writes into the refcount table", 759 }, 760 { 761 .name = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK, 762 .type = QEMU_OPT_BOOL, 763 .help = "Check for unintended writes into a refcount block", 764 }, 765 { 766 .name = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE, 767 .type = QEMU_OPT_BOOL, 768 .help = "Check for unintended writes into the snapshot table", 769 }, 770 { 771 .name = QCOW2_OPT_OVERLAP_INACTIVE_L1, 772 .type = QEMU_OPT_BOOL, 773 .help = "Check for unintended writes into an inactive L1 table", 774 }, 775 { 776 .name = QCOW2_OPT_OVERLAP_INACTIVE_L2, 777 .type = QEMU_OPT_BOOL, 778 .help = "Check for unintended writes into an inactive L2 table", 779 }, 780 { 781 .name = QCOW2_OPT_OVERLAP_BITMAP_DIRECTORY, 782 .type = QEMU_OPT_BOOL, 783 .help = "Check for unintended writes into the bitmap directory", 784 }, 785 { 786 .name = QCOW2_OPT_CACHE_SIZE, 787 .type = QEMU_OPT_SIZE, 788 .help = "Maximum combined metadata (L2 tables and refcount blocks) " 789 "cache size", 790 }, 791 { 792 .name = QCOW2_OPT_L2_CACHE_SIZE, 793 .type = QEMU_OPT_SIZE, 794 .help = "Maximum L2 table cache size", 795 }, 796 { 797 .name = QCOW2_OPT_L2_CACHE_ENTRY_SIZE, 798 .type = QEMU_OPT_SIZE, 799 .help = "Size of each entry in the L2 cache", 800 }, 801 { 802 .name = QCOW2_OPT_REFCOUNT_CACHE_SIZE, 803 .type = QEMU_OPT_SIZE, 804 .help = "Maximum refcount block cache size", 805 }, 806 { 807 .name = QCOW2_OPT_CACHE_CLEAN_INTERVAL, 808 .type = QEMU_OPT_NUMBER, 809 .help = "Clean unused cache entries after this time (in seconds)", 810 }, 811 BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.", 812 "ID of secret providing qcow2 AES key or LUKS passphrase"), 813 { /* end of list */ } 814 }, 815 }; 816 817 static const char *overlap_bool_option_names[QCOW2_OL_MAX_BITNR] = { 818 [QCOW2_OL_MAIN_HEADER_BITNR] = QCOW2_OPT_OVERLAP_MAIN_HEADER, 819 [QCOW2_OL_ACTIVE_L1_BITNR] = QCOW2_OPT_OVERLAP_ACTIVE_L1, 820 [QCOW2_OL_ACTIVE_L2_BITNR] = QCOW2_OPT_OVERLAP_ACTIVE_L2, 821 [QCOW2_OL_REFCOUNT_TABLE_BITNR] = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE, 822 [QCOW2_OL_REFCOUNT_BLOCK_BITNR] = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK, 823 [QCOW2_OL_SNAPSHOT_TABLE_BITNR] = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE, 824 [QCOW2_OL_INACTIVE_L1_BITNR] = QCOW2_OPT_OVERLAP_INACTIVE_L1, 825 [QCOW2_OL_INACTIVE_L2_BITNR] = QCOW2_OPT_OVERLAP_INACTIVE_L2, 826 [QCOW2_OL_BITMAP_DIRECTORY_BITNR] = QCOW2_OPT_OVERLAP_BITMAP_DIRECTORY, 827 }; 828 829 static void cache_clean_timer_cb(void *opaque) 830 { 831 BlockDriverState *bs = opaque; 832 BDRVQcow2State *s = bs->opaque; 833 qcow2_cache_clean_unused(s->l2_table_cache); 834 qcow2_cache_clean_unused(s->refcount_block_cache); 835 timer_mod(s->cache_clean_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) + 836 (int64_t) s->cache_clean_interval * 1000); 837 } 838 839 static void cache_clean_timer_init(BlockDriverState *bs, AioContext *context) 840 { 841 BDRVQcow2State *s = bs->opaque; 842 if (s->cache_clean_interval > 0) { 843 s->cache_clean_timer = 844 aio_timer_new_with_attrs(context, QEMU_CLOCK_VIRTUAL, 845 SCALE_MS, QEMU_TIMER_ATTR_EXTERNAL, 846 cache_clean_timer_cb, bs); 847 timer_mod(s->cache_clean_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) + 848 (int64_t) s->cache_clean_interval * 1000); 849 } 850 } 851 852 static void cache_clean_timer_del(BlockDriverState *bs) 853 { 854 BDRVQcow2State *s = bs->opaque; 855 if (s->cache_clean_timer) { 856 timer_free(s->cache_clean_timer); 857 s->cache_clean_timer = NULL; 858 } 859 } 860 861 static void qcow2_detach_aio_context(BlockDriverState *bs) 862 { 863 cache_clean_timer_del(bs); 864 } 865 866 static void qcow2_attach_aio_context(BlockDriverState *bs, 867 AioContext *new_context) 868 { 869 cache_clean_timer_init(bs, new_context); 870 } 871 872 static bool read_cache_sizes(BlockDriverState *bs, QemuOpts *opts, 873 uint64_t *l2_cache_size, 874 uint64_t *l2_cache_entry_size, 875 uint64_t *refcount_cache_size, Error **errp) 876 { 877 BDRVQcow2State *s = bs->opaque; 878 uint64_t combined_cache_size, l2_cache_max_setting; 879 bool l2_cache_size_set, refcount_cache_size_set, combined_cache_size_set; 880 bool l2_cache_entry_size_set; 881 int min_refcount_cache = MIN_REFCOUNT_CACHE_SIZE * s->cluster_size; 882 uint64_t virtual_disk_size = bs->total_sectors * BDRV_SECTOR_SIZE; 883 uint64_t max_l2_entries = DIV_ROUND_UP(virtual_disk_size, s->cluster_size); 884 /* An L2 table is always one cluster in size so the max cache size 885 * should be a multiple of the cluster size. */ 886 uint64_t max_l2_cache = ROUND_UP(max_l2_entries * l2_entry_size(s), 887 s->cluster_size); 888 889 combined_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_CACHE_SIZE); 890 l2_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_L2_CACHE_SIZE); 891 refcount_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_REFCOUNT_CACHE_SIZE); 892 l2_cache_entry_size_set = qemu_opt_get(opts, QCOW2_OPT_L2_CACHE_ENTRY_SIZE); 893 894 combined_cache_size = qemu_opt_get_size(opts, QCOW2_OPT_CACHE_SIZE, 0); 895 l2_cache_max_setting = qemu_opt_get_size(opts, QCOW2_OPT_L2_CACHE_SIZE, 896 DEFAULT_L2_CACHE_MAX_SIZE); 897 *refcount_cache_size = qemu_opt_get_size(opts, 898 QCOW2_OPT_REFCOUNT_CACHE_SIZE, 0); 899 900 *l2_cache_entry_size = qemu_opt_get_size( 901 opts, QCOW2_OPT_L2_CACHE_ENTRY_SIZE, s->cluster_size); 902 903 *l2_cache_size = MIN(max_l2_cache, l2_cache_max_setting); 904 905 if (combined_cache_size_set) { 906 if (l2_cache_size_set && refcount_cache_size_set) { 907 error_setg(errp, QCOW2_OPT_CACHE_SIZE ", " QCOW2_OPT_L2_CACHE_SIZE 908 " and " QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not be set " 909 "at the same time"); 910 return false; 911 } else if (l2_cache_size_set && 912 (l2_cache_max_setting > combined_cache_size)) { 913 error_setg(errp, QCOW2_OPT_L2_CACHE_SIZE " may not exceed " 914 QCOW2_OPT_CACHE_SIZE); 915 return false; 916 } else if (*refcount_cache_size > combined_cache_size) { 917 error_setg(errp, QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not exceed " 918 QCOW2_OPT_CACHE_SIZE); 919 return false; 920 } 921 922 if (l2_cache_size_set) { 923 *refcount_cache_size = combined_cache_size - *l2_cache_size; 924 } else if (refcount_cache_size_set) { 925 *l2_cache_size = combined_cache_size - *refcount_cache_size; 926 } else { 927 /* Assign as much memory as possible to the L2 cache, and 928 * use the remainder for the refcount cache */ 929 if (combined_cache_size >= max_l2_cache + min_refcount_cache) { 930 *l2_cache_size = max_l2_cache; 931 *refcount_cache_size = combined_cache_size - *l2_cache_size; 932 } else { 933 *refcount_cache_size = 934 MIN(combined_cache_size, min_refcount_cache); 935 *l2_cache_size = combined_cache_size - *refcount_cache_size; 936 } 937 } 938 } 939 940 /* 941 * If the L2 cache is not enough to cover the whole disk then 942 * default to 4KB entries. Smaller entries reduce the cost of 943 * loads and evictions and increase I/O performance. 944 */ 945 if (*l2_cache_size < max_l2_cache && !l2_cache_entry_size_set) { 946 *l2_cache_entry_size = MIN(s->cluster_size, 4096); 947 } 948 949 /* l2_cache_size and refcount_cache_size are ensured to have at least 950 * their minimum values in qcow2_update_options_prepare() */ 951 952 if (*l2_cache_entry_size < (1 << MIN_CLUSTER_BITS) || 953 *l2_cache_entry_size > s->cluster_size || 954 !is_power_of_2(*l2_cache_entry_size)) { 955 error_setg(errp, "L2 cache entry size must be a power of two " 956 "between %d and the cluster size (%d)", 957 1 << MIN_CLUSTER_BITS, s->cluster_size); 958 return false; 959 } 960 961 return true; 962 } 963 964 typedef struct Qcow2ReopenState { 965 Qcow2Cache *l2_table_cache; 966 Qcow2Cache *refcount_block_cache; 967 int l2_slice_size; /* Number of entries in a slice of the L2 table */ 968 bool use_lazy_refcounts; 969 int overlap_check; 970 bool discard_passthrough[QCOW2_DISCARD_MAX]; 971 uint64_t cache_clean_interval; 972 QCryptoBlockOpenOptions *crypto_opts; /* Disk encryption runtime options */ 973 } Qcow2ReopenState; 974 975 static int qcow2_update_options_prepare(BlockDriverState *bs, 976 Qcow2ReopenState *r, 977 QDict *options, int flags, 978 Error **errp) 979 { 980 BDRVQcow2State *s = bs->opaque; 981 QemuOpts *opts = NULL; 982 const char *opt_overlap_check, *opt_overlap_check_template; 983 int overlap_check_template = 0; 984 uint64_t l2_cache_size, l2_cache_entry_size, refcount_cache_size; 985 int i; 986 const char *encryptfmt; 987 QDict *encryptopts = NULL; 988 int ret; 989 990 qdict_extract_subqdict(options, &encryptopts, "encrypt."); 991 encryptfmt = qdict_get_try_str(encryptopts, "format"); 992 993 opts = qemu_opts_create(&qcow2_runtime_opts, NULL, 0, &error_abort); 994 if (!qemu_opts_absorb_qdict(opts, options, errp)) { 995 ret = -EINVAL; 996 goto fail; 997 } 998 999 /* get L2 table/refcount block cache size from command line options */ 1000 if (!read_cache_sizes(bs, opts, &l2_cache_size, &l2_cache_entry_size, 1001 &refcount_cache_size, errp)) { 1002 ret = -EINVAL; 1003 goto fail; 1004 } 1005 1006 l2_cache_size /= l2_cache_entry_size; 1007 if (l2_cache_size < MIN_L2_CACHE_SIZE) { 1008 l2_cache_size = MIN_L2_CACHE_SIZE; 1009 } 1010 if (l2_cache_size > INT_MAX) { 1011 error_setg(errp, "L2 cache size too big"); 1012 ret = -EINVAL; 1013 goto fail; 1014 } 1015 1016 refcount_cache_size /= s->cluster_size; 1017 if (refcount_cache_size < MIN_REFCOUNT_CACHE_SIZE) { 1018 refcount_cache_size = MIN_REFCOUNT_CACHE_SIZE; 1019 } 1020 if (refcount_cache_size > INT_MAX) { 1021 error_setg(errp, "Refcount cache size too big"); 1022 ret = -EINVAL; 1023 goto fail; 1024 } 1025 1026 /* alloc new L2 table/refcount block cache, flush old one */ 1027 if (s->l2_table_cache) { 1028 ret = qcow2_cache_flush(bs, s->l2_table_cache); 1029 if (ret) { 1030 error_setg_errno(errp, -ret, "Failed to flush the L2 table cache"); 1031 goto fail; 1032 } 1033 } 1034 1035 if (s->refcount_block_cache) { 1036 ret = qcow2_cache_flush(bs, s->refcount_block_cache); 1037 if (ret) { 1038 error_setg_errno(errp, -ret, 1039 "Failed to flush the refcount block cache"); 1040 goto fail; 1041 } 1042 } 1043 1044 r->l2_slice_size = l2_cache_entry_size / l2_entry_size(s); 1045 r->l2_table_cache = qcow2_cache_create(bs, l2_cache_size, 1046 l2_cache_entry_size); 1047 r->refcount_block_cache = qcow2_cache_create(bs, refcount_cache_size, 1048 s->cluster_size); 1049 if (r->l2_table_cache == NULL || r->refcount_block_cache == NULL) { 1050 error_setg(errp, "Could not allocate metadata caches"); 1051 ret = -ENOMEM; 1052 goto fail; 1053 } 1054 1055 /* New interval for cache cleanup timer */ 1056 r->cache_clean_interval = 1057 qemu_opt_get_number(opts, QCOW2_OPT_CACHE_CLEAN_INTERVAL, 1058 DEFAULT_CACHE_CLEAN_INTERVAL); 1059 #ifndef CONFIG_LINUX 1060 if (r->cache_clean_interval != 0) { 1061 error_setg(errp, QCOW2_OPT_CACHE_CLEAN_INTERVAL 1062 " not supported on this host"); 1063 ret = -EINVAL; 1064 goto fail; 1065 } 1066 #endif 1067 if (r->cache_clean_interval > UINT_MAX) { 1068 error_setg(errp, "Cache clean interval too big"); 1069 ret = -EINVAL; 1070 goto fail; 1071 } 1072 1073 /* lazy-refcounts; flush if going from enabled to disabled */ 1074 r->use_lazy_refcounts = qemu_opt_get_bool(opts, QCOW2_OPT_LAZY_REFCOUNTS, 1075 (s->compatible_features & QCOW2_COMPAT_LAZY_REFCOUNTS)); 1076 if (r->use_lazy_refcounts && s->qcow_version < 3) { 1077 error_setg(errp, "Lazy refcounts require a qcow2 image with at least " 1078 "qemu 1.1 compatibility level"); 1079 ret = -EINVAL; 1080 goto fail; 1081 } 1082 1083 if (s->use_lazy_refcounts && !r->use_lazy_refcounts) { 1084 ret = qcow2_mark_clean(bs); 1085 if (ret < 0) { 1086 error_setg_errno(errp, -ret, "Failed to disable lazy refcounts"); 1087 goto fail; 1088 } 1089 } 1090 1091 /* Overlap check options */ 1092 opt_overlap_check = qemu_opt_get(opts, QCOW2_OPT_OVERLAP); 1093 opt_overlap_check_template = qemu_opt_get(opts, QCOW2_OPT_OVERLAP_TEMPLATE); 1094 if (opt_overlap_check_template && opt_overlap_check && 1095 strcmp(opt_overlap_check_template, opt_overlap_check)) 1096 { 1097 error_setg(errp, "Conflicting values for qcow2 options '" 1098 QCOW2_OPT_OVERLAP "' ('%s') and '" QCOW2_OPT_OVERLAP_TEMPLATE 1099 "' ('%s')", opt_overlap_check, opt_overlap_check_template); 1100 ret = -EINVAL; 1101 goto fail; 1102 } 1103 if (!opt_overlap_check) { 1104 opt_overlap_check = opt_overlap_check_template ?: "cached"; 1105 } 1106 1107 if (!strcmp(opt_overlap_check, "none")) { 1108 overlap_check_template = 0; 1109 } else if (!strcmp(opt_overlap_check, "constant")) { 1110 overlap_check_template = QCOW2_OL_CONSTANT; 1111 } else if (!strcmp(opt_overlap_check, "cached")) { 1112 overlap_check_template = QCOW2_OL_CACHED; 1113 } else if (!strcmp(opt_overlap_check, "all")) { 1114 overlap_check_template = QCOW2_OL_ALL; 1115 } else { 1116 error_setg(errp, "Unsupported value '%s' for qcow2 option " 1117 "'overlap-check'. Allowed are any of the following: " 1118 "none, constant, cached, all", opt_overlap_check); 1119 ret = -EINVAL; 1120 goto fail; 1121 } 1122 1123 r->overlap_check = 0; 1124 for (i = 0; i < QCOW2_OL_MAX_BITNR; i++) { 1125 /* overlap-check defines a template bitmask, but every flag may be 1126 * overwritten through the associated boolean option */ 1127 r->overlap_check |= 1128 qemu_opt_get_bool(opts, overlap_bool_option_names[i], 1129 overlap_check_template & (1 << i)) << i; 1130 } 1131 1132 r->discard_passthrough[QCOW2_DISCARD_NEVER] = false; 1133 r->discard_passthrough[QCOW2_DISCARD_ALWAYS] = true; 1134 r->discard_passthrough[QCOW2_DISCARD_REQUEST] = 1135 qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_REQUEST, 1136 flags & BDRV_O_UNMAP); 1137 r->discard_passthrough[QCOW2_DISCARD_SNAPSHOT] = 1138 qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_SNAPSHOT, true); 1139 r->discard_passthrough[QCOW2_DISCARD_OTHER] = 1140 qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_OTHER, false); 1141 1142 switch (s->crypt_method_header) { 1143 case QCOW_CRYPT_NONE: 1144 if (encryptfmt) { 1145 error_setg(errp, "No encryption in image header, but options " 1146 "specified format '%s'", encryptfmt); 1147 ret = -EINVAL; 1148 goto fail; 1149 } 1150 break; 1151 1152 case QCOW_CRYPT_AES: 1153 if (encryptfmt && !g_str_equal(encryptfmt, "aes")) { 1154 error_setg(errp, 1155 "Header reported 'aes' encryption format but " 1156 "options specify '%s'", encryptfmt); 1157 ret = -EINVAL; 1158 goto fail; 1159 } 1160 qdict_put_str(encryptopts, "format", "qcow"); 1161 r->crypto_opts = block_crypto_open_opts_init(encryptopts, errp); 1162 if (!r->crypto_opts) { 1163 ret = -EINVAL; 1164 goto fail; 1165 } 1166 break; 1167 1168 case QCOW_CRYPT_LUKS: 1169 if (encryptfmt && !g_str_equal(encryptfmt, "luks")) { 1170 error_setg(errp, 1171 "Header reported 'luks' encryption format but " 1172 "options specify '%s'", encryptfmt); 1173 ret = -EINVAL; 1174 goto fail; 1175 } 1176 qdict_put_str(encryptopts, "format", "luks"); 1177 r->crypto_opts = block_crypto_open_opts_init(encryptopts, errp); 1178 if (!r->crypto_opts) { 1179 ret = -EINVAL; 1180 goto fail; 1181 } 1182 break; 1183 1184 default: 1185 error_setg(errp, "Unsupported encryption method %d", 1186 s->crypt_method_header); 1187 ret = -EINVAL; 1188 goto fail; 1189 } 1190 1191 ret = 0; 1192 fail: 1193 qobject_unref(encryptopts); 1194 qemu_opts_del(opts); 1195 opts = NULL; 1196 return ret; 1197 } 1198 1199 static void qcow2_update_options_commit(BlockDriverState *bs, 1200 Qcow2ReopenState *r) 1201 { 1202 BDRVQcow2State *s = bs->opaque; 1203 int i; 1204 1205 if (s->l2_table_cache) { 1206 qcow2_cache_destroy(s->l2_table_cache); 1207 } 1208 if (s->refcount_block_cache) { 1209 qcow2_cache_destroy(s->refcount_block_cache); 1210 } 1211 s->l2_table_cache = r->l2_table_cache; 1212 s->refcount_block_cache = r->refcount_block_cache; 1213 s->l2_slice_size = r->l2_slice_size; 1214 1215 s->overlap_check = r->overlap_check; 1216 s->use_lazy_refcounts = r->use_lazy_refcounts; 1217 1218 for (i = 0; i < QCOW2_DISCARD_MAX; i++) { 1219 s->discard_passthrough[i] = r->discard_passthrough[i]; 1220 } 1221 1222 if (s->cache_clean_interval != r->cache_clean_interval) { 1223 cache_clean_timer_del(bs); 1224 s->cache_clean_interval = r->cache_clean_interval; 1225 cache_clean_timer_init(bs, bdrv_get_aio_context(bs)); 1226 } 1227 1228 qapi_free_QCryptoBlockOpenOptions(s->crypto_opts); 1229 s->crypto_opts = r->crypto_opts; 1230 } 1231 1232 static void qcow2_update_options_abort(BlockDriverState *bs, 1233 Qcow2ReopenState *r) 1234 { 1235 if (r->l2_table_cache) { 1236 qcow2_cache_destroy(r->l2_table_cache); 1237 } 1238 if (r->refcount_block_cache) { 1239 qcow2_cache_destroy(r->refcount_block_cache); 1240 } 1241 qapi_free_QCryptoBlockOpenOptions(r->crypto_opts); 1242 } 1243 1244 static int qcow2_update_options(BlockDriverState *bs, QDict *options, 1245 int flags, Error **errp) 1246 { 1247 Qcow2ReopenState r = {}; 1248 int ret; 1249 1250 ret = qcow2_update_options_prepare(bs, &r, options, flags, errp); 1251 if (ret >= 0) { 1252 qcow2_update_options_commit(bs, &r); 1253 } else { 1254 qcow2_update_options_abort(bs, &r); 1255 } 1256 1257 return ret; 1258 } 1259 1260 static int validate_compression_type(BDRVQcow2State *s, Error **errp) 1261 { 1262 switch (s->compression_type) { 1263 case QCOW2_COMPRESSION_TYPE_ZLIB: 1264 #ifdef CONFIG_ZSTD 1265 case QCOW2_COMPRESSION_TYPE_ZSTD: 1266 #endif 1267 break; 1268 1269 default: 1270 error_setg(errp, "qcow2: unknown compression type: %u", 1271 s->compression_type); 1272 return -ENOTSUP; 1273 } 1274 1275 /* 1276 * if the compression type differs from QCOW2_COMPRESSION_TYPE_ZLIB 1277 * the incompatible feature flag must be set 1278 */ 1279 if (s->compression_type == QCOW2_COMPRESSION_TYPE_ZLIB) { 1280 if (s->incompatible_features & QCOW2_INCOMPAT_COMPRESSION) { 1281 error_setg(errp, "qcow2: Compression type incompatible feature " 1282 "bit must not be set"); 1283 return -EINVAL; 1284 } 1285 } else { 1286 if (!(s->incompatible_features & QCOW2_INCOMPAT_COMPRESSION)) { 1287 error_setg(errp, "qcow2: Compression type incompatible feature " 1288 "bit must be set"); 1289 return -EINVAL; 1290 } 1291 } 1292 1293 return 0; 1294 } 1295 1296 /* Called with s->lock held. */ 1297 static int coroutine_fn qcow2_do_open(BlockDriverState *bs, QDict *options, 1298 int flags, Error **errp) 1299 { 1300 ERRP_GUARD(); 1301 BDRVQcow2State *s = bs->opaque; 1302 unsigned int len, i; 1303 int ret = 0; 1304 QCowHeader header; 1305 uint64_t ext_end; 1306 uint64_t l1_vm_state_index; 1307 bool update_header = false; 1308 1309 ret = bdrv_pread(bs->file, 0, &header, sizeof(header)); 1310 if (ret < 0) { 1311 error_setg_errno(errp, -ret, "Could not read qcow2 header"); 1312 goto fail; 1313 } 1314 header.magic = be32_to_cpu(header.magic); 1315 header.version = be32_to_cpu(header.version); 1316 header.backing_file_offset = be64_to_cpu(header.backing_file_offset); 1317 header.backing_file_size = be32_to_cpu(header.backing_file_size); 1318 header.size = be64_to_cpu(header.size); 1319 header.cluster_bits = be32_to_cpu(header.cluster_bits); 1320 header.crypt_method = be32_to_cpu(header.crypt_method); 1321 header.l1_table_offset = be64_to_cpu(header.l1_table_offset); 1322 header.l1_size = be32_to_cpu(header.l1_size); 1323 header.refcount_table_offset = be64_to_cpu(header.refcount_table_offset); 1324 header.refcount_table_clusters = 1325 be32_to_cpu(header.refcount_table_clusters); 1326 header.snapshots_offset = be64_to_cpu(header.snapshots_offset); 1327 header.nb_snapshots = be32_to_cpu(header.nb_snapshots); 1328 1329 if (header.magic != QCOW_MAGIC) { 1330 error_setg(errp, "Image is not in qcow2 format"); 1331 ret = -EINVAL; 1332 goto fail; 1333 } 1334 if (header.version < 2 || header.version > 3) { 1335 error_setg(errp, "Unsupported qcow2 version %" PRIu32, header.version); 1336 ret = -ENOTSUP; 1337 goto fail; 1338 } 1339 1340 s->qcow_version = header.version; 1341 1342 /* Initialise cluster size */ 1343 if (header.cluster_bits < MIN_CLUSTER_BITS || 1344 header.cluster_bits > MAX_CLUSTER_BITS) { 1345 error_setg(errp, "Unsupported cluster size: 2^%" PRIu32, 1346 header.cluster_bits); 1347 ret = -EINVAL; 1348 goto fail; 1349 } 1350 1351 s->cluster_bits = header.cluster_bits; 1352 s->cluster_size = 1 << s->cluster_bits; 1353 1354 /* Initialise version 3 header fields */ 1355 if (header.version == 2) { 1356 header.incompatible_features = 0; 1357 header.compatible_features = 0; 1358 header.autoclear_features = 0; 1359 header.refcount_order = 4; 1360 header.header_length = 72; 1361 } else { 1362 header.incompatible_features = 1363 be64_to_cpu(header.incompatible_features); 1364 header.compatible_features = be64_to_cpu(header.compatible_features); 1365 header.autoclear_features = be64_to_cpu(header.autoclear_features); 1366 header.refcount_order = be32_to_cpu(header.refcount_order); 1367 header.header_length = be32_to_cpu(header.header_length); 1368 1369 if (header.header_length < 104) { 1370 error_setg(errp, "qcow2 header too short"); 1371 ret = -EINVAL; 1372 goto fail; 1373 } 1374 } 1375 1376 if (header.header_length > s->cluster_size) { 1377 error_setg(errp, "qcow2 header exceeds cluster size"); 1378 ret = -EINVAL; 1379 goto fail; 1380 } 1381 1382 if (header.header_length > sizeof(header)) { 1383 s->unknown_header_fields_size = header.header_length - sizeof(header); 1384 s->unknown_header_fields = g_malloc(s->unknown_header_fields_size); 1385 ret = bdrv_pread(bs->file, sizeof(header), s->unknown_header_fields, 1386 s->unknown_header_fields_size); 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_table, 1582 s->l1_size * L1E_SIZE); 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 /* Open external data file */ 1617 s->data_file = bdrv_open_child(NULL, options, "data-file", bs, 1618 &child_of_bds, BDRV_CHILD_DATA, 1619 true, errp); 1620 if (*errp) { 1621 ret = -EINVAL; 1622 goto fail; 1623 } 1624 1625 if (s->incompatible_features & QCOW2_INCOMPAT_DATA_FILE) { 1626 if (!s->data_file && s->image_data_file) { 1627 s->data_file = bdrv_open_child(s->image_data_file, options, 1628 "data-file", bs, &child_of_bds, 1629 BDRV_CHILD_DATA, false, errp); 1630 if (!s->data_file) { 1631 ret = -EINVAL; 1632 goto fail; 1633 } 1634 } 1635 if (!s->data_file) { 1636 error_setg(errp, "'data-file' is required for this image"); 1637 ret = -EINVAL; 1638 goto fail; 1639 } 1640 1641 /* No data here */ 1642 bs->file->role &= ~BDRV_CHILD_DATA; 1643 1644 /* Must succeed because we have given up permissions if anything */ 1645 bdrv_child_refresh_perms(bs, bs->file, &error_abort); 1646 } else { 1647 if (s->data_file) { 1648 error_setg(errp, "'data-file' can only be set for images with an " 1649 "external data file"); 1650 ret = -EINVAL; 1651 goto fail; 1652 } 1653 1654 s->data_file = bs->file; 1655 1656 if (data_file_is_raw(bs)) { 1657 error_setg(errp, "data-file-raw requires a data file"); 1658 ret = -EINVAL; 1659 goto fail; 1660 } 1661 } 1662 1663 /* qcow2_read_extension may have set up the crypto context 1664 * if the crypt method needs a header region, some methods 1665 * don't need header extensions, so must check here 1666 */ 1667 if (s->crypt_method_header && !s->crypto) { 1668 if (s->crypt_method_header == QCOW_CRYPT_AES) { 1669 unsigned int cflags = 0; 1670 if (flags & BDRV_O_NO_IO) { 1671 cflags |= QCRYPTO_BLOCK_OPEN_NO_IO; 1672 } 1673 s->crypto = qcrypto_block_open(s->crypto_opts, "encrypt.", 1674 NULL, NULL, cflags, 1675 QCOW2_MAX_THREADS, errp); 1676 if (!s->crypto) { 1677 ret = -EINVAL; 1678 goto fail; 1679 } 1680 } else if (!(flags & BDRV_O_NO_IO)) { 1681 error_setg(errp, "Missing CRYPTO header for crypt method %d", 1682 s->crypt_method_header); 1683 ret = -EINVAL; 1684 goto fail; 1685 } 1686 } 1687 1688 /* read the backing file name */ 1689 if (header.backing_file_offset != 0) { 1690 len = header.backing_file_size; 1691 if (len > MIN(1023, s->cluster_size - header.backing_file_offset) || 1692 len >= sizeof(bs->backing_file)) { 1693 error_setg(errp, "Backing file name too long"); 1694 ret = -EINVAL; 1695 goto fail; 1696 } 1697 ret = bdrv_pread(bs->file, header.backing_file_offset, 1698 bs->auto_backing_file, len); 1699 if (ret < 0) { 1700 error_setg_errno(errp, -ret, "Could not read backing file name"); 1701 goto fail; 1702 } 1703 bs->auto_backing_file[len] = '\0'; 1704 pstrcpy(bs->backing_file, sizeof(bs->backing_file), 1705 bs->auto_backing_file); 1706 s->image_backing_file = g_strdup(bs->auto_backing_file); 1707 } 1708 1709 /* 1710 * Internal snapshots; skip reading them in check mode, because 1711 * we do not need them then, and we do not want to abort because 1712 * of a broken table. 1713 */ 1714 if (!(flags & BDRV_O_CHECK)) { 1715 s->snapshots_offset = header.snapshots_offset; 1716 s->nb_snapshots = header.nb_snapshots; 1717 1718 ret = qcow2_read_snapshots(bs, errp); 1719 if (ret < 0) { 1720 goto fail; 1721 } 1722 } 1723 1724 /* Clear unknown autoclear feature bits */ 1725 update_header |= s->autoclear_features & ~QCOW2_AUTOCLEAR_MASK; 1726 update_header = update_header && bdrv_is_writable(bs); 1727 if (update_header) { 1728 s->autoclear_features &= QCOW2_AUTOCLEAR_MASK; 1729 } 1730 1731 /* == Handle persistent dirty bitmaps == 1732 * 1733 * We want load dirty bitmaps in three cases: 1734 * 1735 * 1. Normal open of the disk in active mode, not related to invalidation 1736 * after migration. 1737 * 1738 * 2. Invalidation of the target vm after pre-copy phase of migration, if 1739 * bitmaps are _not_ migrating through migration channel, i.e. 1740 * 'dirty-bitmaps' capability is disabled. 1741 * 1742 * 3. Invalidation of source vm after failed or canceled migration. 1743 * This is a very interesting case. There are two possible types of 1744 * bitmaps: 1745 * 1746 * A. Stored on inactivation and removed. They should be loaded from the 1747 * image. 1748 * 1749 * B. Not stored: not-persistent bitmaps and bitmaps, migrated through 1750 * the migration channel (with dirty-bitmaps capability). 1751 * 1752 * On the other hand, there are two possible sub-cases: 1753 * 1754 * 3.1 disk was changed by somebody else while were inactive. In this 1755 * case all in-RAM dirty bitmaps (both persistent and not) are 1756 * definitely invalid. And we don't have any method to determine 1757 * this. 1758 * 1759 * Simple and safe thing is to just drop all the bitmaps of type B on 1760 * inactivation. But in this case we lose bitmaps in valid 4.2 case. 1761 * 1762 * On the other hand, resuming source vm, if disk was already changed 1763 * is a bad thing anyway: not only bitmaps, the whole vm state is 1764 * out of sync with disk. 1765 * 1766 * This means, that user or management tool, who for some reason 1767 * decided to resume source vm, after disk was already changed by 1768 * target vm, should at least drop all dirty bitmaps by hand. 1769 * 1770 * So, we can ignore this case for now, but TODO: "generation" 1771 * extension for qcow2, to determine, that image was changed after 1772 * last inactivation. And if it is changed, we will drop (or at least 1773 * mark as 'invalid' all the bitmaps of type B, both persistent 1774 * and not). 1775 * 1776 * 3.2 disk was _not_ changed while were inactive. Bitmaps may be saved 1777 * to disk ('dirty-bitmaps' capability disabled), or not saved 1778 * ('dirty-bitmaps' capability enabled), but we don't need to care 1779 * of: let's load bitmaps as always: stored bitmaps will be loaded, 1780 * and not stored has flag IN_USE=1 in the image and will be skipped 1781 * on loading. 1782 * 1783 * One remaining possible case when we don't want load bitmaps: 1784 * 1785 * 4. Open disk in inactive mode in target vm (bitmaps are migrating or 1786 * will be loaded on invalidation, no needs try loading them before) 1787 */ 1788 1789 if (!(bdrv_get_flags(bs) & BDRV_O_INACTIVE)) { 1790 /* It's case 1, 2 or 3.2. Or 3.1 which is BUG in management layer. */ 1791 bool header_updated; 1792 if (!qcow2_load_dirty_bitmaps(bs, &header_updated, errp)) { 1793 ret = -EINVAL; 1794 goto fail; 1795 } 1796 1797 update_header = update_header && !header_updated; 1798 } 1799 1800 if (update_header) { 1801 ret = qcow2_update_header(bs); 1802 if (ret < 0) { 1803 error_setg_errno(errp, -ret, "Could not update qcow2 header"); 1804 goto fail; 1805 } 1806 } 1807 1808 bs->supported_zero_flags = header.version >= 3 ? 1809 BDRV_REQ_MAY_UNMAP | BDRV_REQ_NO_FALLBACK : 0; 1810 bs->supported_truncate_flags = BDRV_REQ_ZERO_WRITE; 1811 1812 /* Repair image if dirty */ 1813 if (!(flags & BDRV_O_CHECK) && bdrv_is_writable(bs) && 1814 (s->incompatible_features & QCOW2_INCOMPAT_DIRTY)) { 1815 BdrvCheckResult result = {0}; 1816 1817 ret = qcow2_co_check_locked(bs, &result, 1818 BDRV_FIX_ERRORS | BDRV_FIX_LEAKS); 1819 if (ret < 0 || result.check_errors) { 1820 if (ret >= 0) { 1821 ret = -EIO; 1822 } 1823 error_setg_errno(errp, -ret, "Could not repair dirty image"); 1824 goto fail; 1825 } 1826 } 1827 1828 #ifdef DEBUG_ALLOC 1829 { 1830 BdrvCheckResult result = {0}; 1831 qcow2_check_refcounts(bs, &result, 0); 1832 } 1833 #endif 1834 1835 qemu_co_queue_init(&s->thread_task_queue); 1836 1837 return ret; 1838 1839 fail: 1840 g_free(s->image_data_file); 1841 if (has_data_file(bs)) { 1842 bdrv_unref_child(bs, s->data_file); 1843 s->data_file = NULL; 1844 } 1845 g_free(s->unknown_header_fields); 1846 cleanup_unknown_header_ext(bs); 1847 qcow2_free_snapshots(bs); 1848 qcow2_refcount_close(bs); 1849 qemu_vfree(s->l1_table); 1850 /* else pre-write overlap checks in cache_destroy may crash */ 1851 s->l1_table = NULL; 1852 cache_clean_timer_del(bs); 1853 if (s->l2_table_cache) { 1854 qcow2_cache_destroy(s->l2_table_cache); 1855 } 1856 if (s->refcount_block_cache) { 1857 qcow2_cache_destroy(s->refcount_block_cache); 1858 } 1859 qcrypto_block_free(s->crypto); 1860 qapi_free_QCryptoBlockOpenOptions(s->crypto_opts); 1861 return ret; 1862 } 1863 1864 typedef struct QCow2OpenCo { 1865 BlockDriverState *bs; 1866 QDict *options; 1867 int flags; 1868 Error **errp; 1869 int ret; 1870 } QCow2OpenCo; 1871 1872 static void coroutine_fn qcow2_open_entry(void *opaque) 1873 { 1874 QCow2OpenCo *qoc = opaque; 1875 BDRVQcow2State *s = qoc->bs->opaque; 1876 1877 qemu_co_mutex_lock(&s->lock); 1878 qoc->ret = qcow2_do_open(qoc->bs, qoc->options, qoc->flags, qoc->errp); 1879 qemu_co_mutex_unlock(&s->lock); 1880 } 1881 1882 static int qcow2_open(BlockDriverState *bs, QDict *options, int flags, 1883 Error **errp) 1884 { 1885 BDRVQcow2State *s = bs->opaque; 1886 QCow2OpenCo qoc = { 1887 .bs = bs, 1888 .options = options, 1889 .flags = flags, 1890 .errp = errp, 1891 .ret = -EINPROGRESS 1892 }; 1893 1894 bs->file = bdrv_open_child(NULL, options, "file", bs, &child_of_bds, 1895 BDRV_CHILD_IMAGE, false, errp); 1896 if (!bs->file) { 1897 return -EINVAL; 1898 } 1899 1900 /* Initialise locks */ 1901 qemu_co_mutex_init(&s->lock); 1902 1903 if (qemu_in_coroutine()) { 1904 /* From bdrv_co_create. */ 1905 qcow2_open_entry(&qoc); 1906 } else { 1907 assert(qemu_get_current_aio_context() == qemu_get_aio_context()); 1908 qemu_coroutine_enter(qemu_coroutine_create(qcow2_open_entry, &qoc)); 1909 BDRV_POLL_WHILE(bs, qoc.ret == -EINPROGRESS); 1910 } 1911 return qoc.ret; 1912 } 1913 1914 static void qcow2_refresh_limits(BlockDriverState *bs, Error **errp) 1915 { 1916 BDRVQcow2State *s = bs->opaque; 1917 1918 if (bs->encrypted) { 1919 /* Encryption works on a sector granularity */ 1920 bs->bl.request_alignment = qcrypto_block_get_sector_size(s->crypto); 1921 } 1922 bs->bl.pwrite_zeroes_alignment = s->subcluster_size; 1923 bs->bl.pdiscard_alignment = s->cluster_size; 1924 } 1925 1926 static int qcow2_reopen_prepare(BDRVReopenState *state, 1927 BlockReopenQueue *queue, Error **errp) 1928 { 1929 BDRVQcow2State *s = state->bs->opaque; 1930 Qcow2ReopenState *r; 1931 int ret; 1932 1933 r = g_new0(Qcow2ReopenState, 1); 1934 state->opaque = r; 1935 1936 ret = qcow2_update_options_prepare(state->bs, r, state->options, 1937 state->flags, errp); 1938 if (ret < 0) { 1939 goto fail; 1940 } 1941 1942 /* We need to write out any unwritten data if we reopen read-only. */ 1943 if ((state->flags & BDRV_O_RDWR) == 0) { 1944 ret = qcow2_reopen_bitmaps_ro(state->bs, errp); 1945 if (ret < 0) { 1946 goto fail; 1947 } 1948 1949 ret = bdrv_flush(state->bs); 1950 if (ret < 0) { 1951 goto fail; 1952 } 1953 1954 ret = qcow2_mark_clean(state->bs); 1955 if (ret < 0) { 1956 goto fail; 1957 } 1958 } 1959 1960 /* 1961 * Without an external data file, s->data_file points to the same BdrvChild 1962 * as bs->file. It needs to be resynced after reopen because bs->file may 1963 * be changed. We can't use it in the meantime. 1964 */ 1965 if (!has_data_file(state->bs)) { 1966 assert(s->data_file == state->bs->file); 1967 s->data_file = NULL; 1968 } 1969 1970 return 0; 1971 1972 fail: 1973 qcow2_update_options_abort(state->bs, r); 1974 g_free(r); 1975 return ret; 1976 } 1977 1978 static void qcow2_reopen_commit(BDRVReopenState *state) 1979 { 1980 BDRVQcow2State *s = state->bs->opaque; 1981 1982 qcow2_update_options_commit(state->bs, state->opaque); 1983 if (!s->data_file) { 1984 /* 1985 * If we don't have an external data file, s->data_file was cleared by 1986 * qcow2_reopen_prepare() and needs to be updated. 1987 */ 1988 s->data_file = state->bs->file; 1989 } 1990 g_free(state->opaque); 1991 } 1992 1993 static void qcow2_reopen_commit_post(BDRVReopenState *state) 1994 { 1995 if (state->flags & BDRV_O_RDWR) { 1996 Error *local_err = NULL; 1997 1998 if (qcow2_reopen_bitmaps_rw(state->bs, &local_err) < 0) { 1999 /* 2000 * This is not fatal, bitmaps just left read-only, so all following 2001 * writes will fail. User can remove read-only bitmaps to unblock 2002 * writes or retry reopen. 2003 */ 2004 error_reportf_err(local_err, 2005 "%s: Failed to make dirty bitmaps writable: ", 2006 bdrv_get_node_name(state->bs)); 2007 } 2008 } 2009 } 2010 2011 static void qcow2_reopen_abort(BDRVReopenState *state) 2012 { 2013 BDRVQcow2State *s = state->bs->opaque; 2014 2015 if (!s->data_file) { 2016 /* 2017 * If we don't have an external data file, s->data_file was cleared by 2018 * qcow2_reopen_prepare() and needs to be restored. 2019 */ 2020 s->data_file = state->bs->file; 2021 } 2022 qcow2_update_options_abort(state->bs, state->opaque); 2023 g_free(state->opaque); 2024 } 2025 2026 static void qcow2_join_options(QDict *options, QDict *old_options) 2027 { 2028 bool has_new_overlap_template = 2029 qdict_haskey(options, QCOW2_OPT_OVERLAP) || 2030 qdict_haskey(options, QCOW2_OPT_OVERLAP_TEMPLATE); 2031 bool has_new_total_cache_size = 2032 qdict_haskey(options, QCOW2_OPT_CACHE_SIZE); 2033 bool has_all_cache_options; 2034 2035 /* New overlap template overrides all old overlap options */ 2036 if (has_new_overlap_template) { 2037 qdict_del(old_options, QCOW2_OPT_OVERLAP); 2038 qdict_del(old_options, QCOW2_OPT_OVERLAP_TEMPLATE); 2039 qdict_del(old_options, QCOW2_OPT_OVERLAP_MAIN_HEADER); 2040 qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L1); 2041 qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L2); 2042 qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_TABLE); 2043 qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK); 2044 qdict_del(old_options, QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE); 2045 qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L1); 2046 qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L2); 2047 } 2048 2049 /* New total cache size overrides all old options */ 2050 if (qdict_haskey(options, QCOW2_OPT_CACHE_SIZE)) { 2051 qdict_del(old_options, QCOW2_OPT_L2_CACHE_SIZE); 2052 qdict_del(old_options, QCOW2_OPT_REFCOUNT_CACHE_SIZE); 2053 } 2054 2055 qdict_join(options, old_options, false); 2056 2057 /* 2058 * If after merging all cache size options are set, an old total size is 2059 * overwritten. Do keep all options, however, if all three are new. The 2060 * resulting error message is what we want to happen. 2061 */ 2062 has_all_cache_options = 2063 qdict_haskey(options, QCOW2_OPT_CACHE_SIZE) || 2064 qdict_haskey(options, QCOW2_OPT_L2_CACHE_SIZE) || 2065 qdict_haskey(options, QCOW2_OPT_REFCOUNT_CACHE_SIZE); 2066 2067 if (has_all_cache_options && !has_new_total_cache_size) { 2068 qdict_del(options, QCOW2_OPT_CACHE_SIZE); 2069 } 2070 } 2071 2072 static int coroutine_fn qcow2_co_block_status(BlockDriverState *bs, 2073 bool want_zero, 2074 int64_t offset, int64_t count, 2075 int64_t *pnum, int64_t *map, 2076 BlockDriverState **file) 2077 { 2078 BDRVQcow2State *s = bs->opaque; 2079 uint64_t host_offset; 2080 unsigned int bytes; 2081 QCow2SubclusterType type; 2082 int ret, status = 0; 2083 2084 qemu_co_mutex_lock(&s->lock); 2085 2086 if (!s->metadata_preallocation_checked) { 2087 ret = qcow2_detect_metadata_preallocation(bs); 2088 s->metadata_preallocation = (ret == 1); 2089 s->metadata_preallocation_checked = true; 2090 } 2091 2092 bytes = MIN(INT_MAX, count); 2093 ret = qcow2_get_host_offset(bs, offset, &bytes, &host_offset, &type); 2094 qemu_co_mutex_unlock(&s->lock); 2095 if (ret < 0) { 2096 return ret; 2097 } 2098 2099 *pnum = bytes; 2100 2101 if ((type == QCOW2_SUBCLUSTER_NORMAL || 2102 type == QCOW2_SUBCLUSTER_ZERO_ALLOC || 2103 type == QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC) && !s->crypto) { 2104 *map = host_offset; 2105 *file = s->data_file->bs; 2106 status |= BDRV_BLOCK_OFFSET_VALID; 2107 } 2108 if (type == QCOW2_SUBCLUSTER_ZERO_PLAIN || 2109 type == QCOW2_SUBCLUSTER_ZERO_ALLOC) { 2110 status |= BDRV_BLOCK_ZERO; 2111 } else if (type != QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN && 2112 type != QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC) { 2113 status |= BDRV_BLOCK_DATA; 2114 } 2115 if (s->metadata_preallocation && (status & BDRV_BLOCK_DATA) && 2116 (status & BDRV_BLOCK_OFFSET_VALID)) 2117 { 2118 status |= BDRV_BLOCK_RECURSE; 2119 } 2120 return status; 2121 } 2122 2123 static coroutine_fn int qcow2_handle_l2meta(BlockDriverState *bs, 2124 QCowL2Meta **pl2meta, 2125 bool link_l2) 2126 { 2127 int ret = 0; 2128 QCowL2Meta *l2meta = *pl2meta; 2129 2130 while (l2meta != NULL) { 2131 QCowL2Meta *next; 2132 2133 if (link_l2) { 2134 ret = qcow2_alloc_cluster_link_l2(bs, l2meta); 2135 if (ret) { 2136 goto out; 2137 } 2138 } else { 2139 qcow2_alloc_cluster_abort(bs, l2meta); 2140 } 2141 2142 /* Take the request off the list of running requests */ 2143 QLIST_REMOVE(l2meta, next_in_flight); 2144 2145 qemu_co_queue_restart_all(&l2meta->dependent_requests); 2146 2147 next = l2meta->next; 2148 g_free(l2meta); 2149 l2meta = next; 2150 } 2151 out: 2152 *pl2meta = l2meta; 2153 return ret; 2154 } 2155 2156 static coroutine_fn int 2157 qcow2_co_preadv_encrypted(BlockDriverState *bs, 2158 uint64_t host_offset, 2159 uint64_t offset, 2160 uint64_t bytes, 2161 QEMUIOVector *qiov, 2162 uint64_t qiov_offset) 2163 { 2164 int ret; 2165 BDRVQcow2State *s = bs->opaque; 2166 uint8_t *buf; 2167 2168 assert(bs->encrypted && s->crypto); 2169 assert(bytes <= QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size); 2170 2171 /* 2172 * For encrypted images, read everything into a temporary 2173 * contiguous buffer on which the AES functions can work. 2174 * Also, decryption in a separate buffer is better as it 2175 * prevents the guest from learning information about the 2176 * encrypted nature of the virtual disk. 2177 */ 2178 2179 buf = qemu_try_blockalign(s->data_file->bs, bytes); 2180 if (buf == NULL) { 2181 return -ENOMEM; 2182 } 2183 2184 BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO); 2185 ret = bdrv_co_pread(s->data_file, host_offset, bytes, buf, 0); 2186 if (ret < 0) { 2187 goto fail; 2188 } 2189 2190 if (qcow2_co_decrypt(bs, host_offset, offset, buf, bytes) < 0) 2191 { 2192 ret = -EIO; 2193 goto fail; 2194 } 2195 qemu_iovec_from_buf(qiov, qiov_offset, buf, bytes); 2196 2197 fail: 2198 qemu_vfree(buf); 2199 2200 return ret; 2201 } 2202 2203 typedef struct Qcow2AioTask { 2204 AioTask task; 2205 2206 BlockDriverState *bs; 2207 QCow2SubclusterType subcluster_type; /* only for read */ 2208 uint64_t host_offset; /* or full descriptor in compressed clusters */ 2209 uint64_t offset; 2210 uint64_t bytes; 2211 QEMUIOVector *qiov; 2212 uint64_t qiov_offset; 2213 QCowL2Meta *l2meta; /* only for write */ 2214 } Qcow2AioTask; 2215 2216 static coroutine_fn int qcow2_co_preadv_task_entry(AioTask *task); 2217 static coroutine_fn int qcow2_add_task(BlockDriverState *bs, 2218 AioTaskPool *pool, 2219 AioTaskFunc func, 2220 QCow2SubclusterType subcluster_type, 2221 uint64_t host_offset, 2222 uint64_t offset, 2223 uint64_t bytes, 2224 QEMUIOVector *qiov, 2225 size_t qiov_offset, 2226 QCowL2Meta *l2meta) 2227 { 2228 Qcow2AioTask local_task; 2229 Qcow2AioTask *task = pool ? g_new(Qcow2AioTask, 1) : &local_task; 2230 2231 *task = (Qcow2AioTask) { 2232 .task.func = func, 2233 .bs = bs, 2234 .subcluster_type = subcluster_type, 2235 .qiov = qiov, 2236 .host_offset = host_offset, 2237 .offset = offset, 2238 .bytes = bytes, 2239 .qiov_offset = qiov_offset, 2240 .l2meta = l2meta, 2241 }; 2242 2243 trace_qcow2_add_task(qemu_coroutine_self(), bs, pool, 2244 func == qcow2_co_preadv_task_entry ? "read" : "write", 2245 subcluster_type, host_offset, offset, bytes, 2246 qiov, qiov_offset); 2247 2248 if (!pool) { 2249 return func(&task->task); 2250 } 2251 2252 aio_task_pool_start_task(pool, &task->task); 2253 2254 return 0; 2255 } 2256 2257 static coroutine_fn int qcow2_co_preadv_task(BlockDriverState *bs, 2258 QCow2SubclusterType subc_type, 2259 uint64_t host_offset, 2260 uint64_t offset, uint64_t bytes, 2261 QEMUIOVector *qiov, 2262 size_t qiov_offset) 2263 { 2264 BDRVQcow2State *s = bs->opaque; 2265 2266 switch (subc_type) { 2267 case QCOW2_SUBCLUSTER_ZERO_PLAIN: 2268 case QCOW2_SUBCLUSTER_ZERO_ALLOC: 2269 /* Both zero types are handled in qcow2_co_preadv_part */ 2270 g_assert_not_reached(); 2271 2272 case QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN: 2273 case QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC: 2274 assert(bs->backing); /* otherwise handled in qcow2_co_preadv_part */ 2275 2276 BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO); 2277 return bdrv_co_preadv_part(bs->backing, offset, bytes, 2278 qiov, qiov_offset, 0); 2279 2280 case QCOW2_SUBCLUSTER_COMPRESSED: 2281 return qcow2_co_preadv_compressed(bs, host_offset, 2282 offset, bytes, qiov, qiov_offset); 2283 2284 case QCOW2_SUBCLUSTER_NORMAL: 2285 if (bs->encrypted) { 2286 return qcow2_co_preadv_encrypted(bs, host_offset, 2287 offset, bytes, qiov, qiov_offset); 2288 } 2289 2290 BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO); 2291 return bdrv_co_preadv_part(s->data_file, host_offset, 2292 bytes, qiov, qiov_offset, 0); 2293 2294 default: 2295 g_assert_not_reached(); 2296 } 2297 2298 g_assert_not_reached(); 2299 } 2300 2301 static coroutine_fn int qcow2_co_preadv_task_entry(AioTask *task) 2302 { 2303 Qcow2AioTask *t = container_of(task, Qcow2AioTask, task); 2304 2305 assert(!t->l2meta); 2306 2307 return qcow2_co_preadv_task(t->bs, t->subcluster_type, 2308 t->host_offset, t->offset, t->bytes, 2309 t->qiov, t->qiov_offset); 2310 } 2311 2312 static coroutine_fn int qcow2_co_preadv_part(BlockDriverState *bs, 2313 uint64_t offset, uint64_t bytes, 2314 QEMUIOVector *qiov, 2315 size_t qiov_offset, int flags) 2316 { 2317 BDRVQcow2State *s = bs->opaque; 2318 int ret = 0; 2319 unsigned int cur_bytes; /* number of bytes in current iteration */ 2320 uint64_t host_offset = 0; 2321 QCow2SubclusterType type; 2322 AioTaskPool *aio = NULL; 2323 2324 while (bytes != 0 && aio_task_pool_status(aio) == 0) { 2325 /* prepare next request */ 2326 cur_bytes = MIN(bytes, INT_MAX); 2327 if (s->crypto) { 2328 cur_bytes = MIN(cur_bytes, 2329 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size); 2330 } 2331 2332 qemu_co_mutex_lock(&s->lock); 2333 ret = qcow2_get_host_offset(bs, offset, &cur_bytes, 2334 &host_offset, &type); 2335 qemu_co_mutex_unlock(&s->lock); 2336 if (ret < 0) { 2337 goto out; 2338 } 2339 2340 if (type == QCOW2_SUBCLUSTER_ZERO_PLAIN || 2341 type == QCOW2_SUBCLUSTER_ZERO_ALLOC || 2342 (type == QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN && !bs->backing) || 2343 (type == QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC && !bs->backing)) 2344 { 2345 qemu_iovec_memset(qiov, qiov_offset, 0, cur_bytes); 2346 } else { 2347 if (!aio && cur_bytes != bytes) { 2348 aio = aio_task_pool_new(QCOW2_MAX_WORKERS); 2349 } 2350 ret = qcow2_add_task(bs, aio, qcow2_co_preadv_task_entry, type, 2351 host_offset, offset, cur_bytes, 2352 qiov, qiov_offset, NULL); 2353 if (ret < 0) { 2354 goto out; 2355 } 2356 } 2357 2358 bytes -= cur_bytes; 2359 offset += cur_bytes; 2360 qiov_offset += cur_bytes; 2361 } 2362 2363 out: 2364 if (aio) { 2365 aio_task_pool_wait_all(aio); 2366 if (ret == 0) { 2367 ret = aio_task_pool_status(aio); 2368 } 2369 g_free(aio); 2370 } 2371 2372 return ret; 2373 } 2374 2375 /* Check if it's possible to merge a write request with the writing of 2376 * the data from the COW regions */ 2377 static bool merge_cow(uint64_t offset, unsigned bytes, 2378 QEMUIOVector *qiov, size_t qiov_offset, 2379 QCowL2Meta *l2meta) 2380 { 2381 QCowL2Meta *m; 2382 2383 for (m = l2meta; m != NULL; m = m->next) { 2384 /* If both COW regions are empty then there's nothing to merge */ 2385 if (m->cow_start.nb_bytes == 0 && m->cow_end.nb_bytes == 0) { 2386 continue; 2387 } 2388 2389 /* If COW regions are handled already, skip this too */ 2390 if (m->skip_cow) { 2391 continue; 2392 } 2393 2394 /* 2395 * The write request should start immediately after the first 2396 * COW region. This does not always happen because the area 2397 * touched by the request can be larger than the one defined 2398 * by @m (a single request can span an area consisting of a 2399 * mix of previously unallocated and allocated clusters, that 2400 * is why @l2meta is a list). 2401 */ 2402 if (l2meta_cow_start(m) + m->cow_start.nb_bytes != offset) { 2403 /* In this case the request starts before this region */ 2404 assert(offset < l2meta_cow_start(m)); 2405 assert(m->cow_start.nb_bytes == 0); 2406 continue; 2407 } 2408 2409 /* The write request should end immediately before the second 2410 * COW region (see above for why it does not always happen) */ 2411 if (m->offset + m->cow_end.offset != offset + bytes) { 2412 assert(offset + bytes > m->offset + m->cow_end.offset); 2413 assert(m->cow_end.nb_bytes == 0); 2414 continue; 2415 } 2416 2417 /* Make sure that adding both COW regions to the QEMUIOVector 2418 * does not exceed IOV_MAX */ 2419 if (qemu_iovec_subvec_niov(qiov, qiov_offset, bytes) > IOV_MAX - 2) { 2420 continue; 2421 } 2422 2423 m->data_qiov = qiov; 2424 m->data_qiov_offset = qiov_offset; 2425 return true; 2426 } 2427 2428 return false; 2429 } 2430 2431 /* 2432 * Return 1 if the COW regions read as zeroes, 0 if not, < 0 on error. 2433 * Note that returning 0 does not guarantee non-zero data. 2434 */ 2435 static int is_zero_cow(BlockDriverState *bs, QCowL2Meta *m) 2436 { 2437 /* 2438 * This check is designed for optimization shortcut so it must be 2439 * efficient. 2440 * Instead of is_zero(), use bdrv_co_is_zero_fast() as it is 2441 * faster (but not as accurate and can result in false negatives). 2442 */ 2443 int ret = bdrv_co_is_zero_fast(bs, m->offset + m->cow_start.offset, 2444 m->cow_start.nb_bytes); 2445 if (ret <= 0) { 2446 return ret; 2447 } 2448 2449 return bdrv_co_is_zero_fast(bs, m->offset + m->cow_end.offset, 2450 m->cow_end.nb_bytes); 2451 } 2452 2453 static int handle_alloc_space(BlockDriverState *bs, QCowL2Meta *l2meta) 2454 { 2455 BDRVQcow2State *s = bs->opaque; 2456 QCowL2Meta *m; 2457 2458 if (!(s->data_file->bs->supported_zero_flags & BDRV_REQ_NO_FALLBACK)) { 2459 return 0; 2460 } 2461 2462 if (bs->encrypted) { 2463 return 0; 2464 } 2465 2466 for (m = l2meta; m != NULL; m = m->next) { 2467 int ret; 2468 uint64_t start_offset = m->alloc_offset + m->cow_start.offset; 2469 unsigned nb_bytes = m->cow_end.offset + m->cow_end.nb_bytes - 2470 m->cow_start.offset; 2471 2472 if (!m->cow_start.nb_bytes && !m->cow_end.nb_bytes) { 2473 continue; 2474 } 2475 2476 ret = is_zero_cow(bs, m); 2477 if (ret < 0) { 2478 return ret; 2479 } else if (ret == 0) { 2480 continue; 2481 } 2482 2483 /* 2484 * instead of writing zero COW buffers, 2485 * efficiently zero out the whole clusters 2486 */ 2487 2488 ret = qcow2_pre_write_overlap_check(bs, 0, start_offset, nb_bytes, 2489 true); 2490 if (ret < 0) { 2491 return ret; 2492 } 2493 2494 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC_SPACE); 2495 ret = bdrv_co_pwrite_zeroes(s->data_file, start_offset, nb_bytes, 2496 BDRV_REQ_NO_FALLBACK); 2497 if (ret < 0) { 2498 if (ret != -ENOTSUP && ret != -EAGAIN) { 2499 return ret; 2500 } 2501 continue; 2502 } 2503 2504 trace_qcow2_skip_cow(qemu_coroutine_self(), m->offset, m->nb_clusters); 2505 m->skip_cow = true; 2506 } 2507 return 0; 2508 } 2509 2510 /* 2511 * qcow2_co_pwritev_task 2512 * Called with s->lock unlocked 2513 * l2meta - if not NULL, qcow2_co_pwritev_task() will consume it. Caller must 2514 * not use it somehow after qcow2_co_pwritev_task() call 2515 */ 2516 static coroutine_fn int qcow2_co_pwritev_task(BlockDriverState *bs, 2517 uint64_t host_offset, 2518 uint64_t offset, uint64_t bytes, 2519 QEMUIOVector *qiov, 2520 uint64_t qiov_offset, 2521 QCowL2Meta *l2meta) 2522 { 2523 int ret; 2524 BDRVQcow2State *s = bs->opaque; 2525 void *crypt_buf = NULL; 2526 QEMUIOVector encrypted_qiov; 2527 2528 if (bs->encrypted) { 2529 assert(s->crypto); 2530 assert(bytes <= QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size); 2531 crypt_buf = qemu_try_blockalign(bs->file->bs, bytes); 2532 if (crypt_buf == NULL) { 2533 ret = -ENOMEM; 2534 goto out_unlocked; 2535 } 2536 qemu_iovec_to_buf(qiov, qiov_offset, crypt_buf, bytes); 2537 2538 if (qcow2_co_encrypt(bs, host_offset, offset, crypt_buf, bytes) < 0) { 2539 ret = -EIO; 2540 goto out_unlocked; 2541 } 2542 2543 qemu_iovec_init_buf(&encrypted_qiov, crypt_buf, bytes); 2544 qiov = &encrypted_qiov; 2545 qiov_offset = 0; 2546 } 2547 2548 /* Try to efficiently initialize the physical space with zeroes */ 2549 ret = handle_alloc_space(bs, l2meta); 2550 if (ret < 0) { 2551 goto out_unlocked; 2552 } 2553 2554 /* 2555 * If we need to do COW, check if it's possible to merge the 2556 * writing of the guest data together with that of the COW regions. 2557 * If it's not possible (or not necessary) then write the 2558 * guest data now. 2559 */ 2560 if (!merge_cow(offset, bytes, qiov, qiov_offset, l2meta)) { 2561 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO); 2562 trace_qcow2_writev_data(qemu_coroutine_self(), host_offset); 2563 ret = bdrv_co_pwritev_part(s->data_file, host_offset, 2564 bytes, qiov, qiov_offset, 0); 2565 if (ret < 0) { 2566 goto out_unlocked; 2567 } 2568 } 2569 2570 qemu_co_mutex_lock(&s->lock); 2571 2572 ret = qcow2_handle_l2meta(bs, &l2meta, true); 2573 goto out_locked; 2574 2575 out_unlocked: 2576 qemu_co_mutex_lock(&s->lock); 2577 2578 out_locked: 2579 qcow2_handle_l2meta(bs, &l2meta, false); 2580 qemu_co_mutex_unlock(&s->lock); 2581 2582 qemu_vfree(crypt_buf); 2583 2584 return ret; 2585 } 2586 2587 static coroutine_fn int qcow2_co_pwritev_task_entry(AioTask *task) 2588 { 2589 Qcow2AioTask *t = container_of(task, Qcow2AioTask, task); 2590 2591 assert(!t->subcluster_type); 2592 2593 return qcow2_co_pwritev_task(t->bs, t->host_offset, 2594 t->offset, t->bytes, t->qiov, t->qiov_offset, 2595 t->l2meta); 2596 } 2597 2598 static coroutine_fn int qcow2_co_pwritev_part( 2599 BlockDriverState *bs, uint64_t offset, uint64_t bytes, 2600 QEMUIOVector *qiov, size_t qiov_offset, int flags) 2601 { 2602 BDRVQcow2State *s = bs->opaque; 2603 int offset_in_cluster; 2604 int ret; 2605 unsigned int cur_bytes; /* number of sectors in current iteration */ 2606 uint64_t host_offset; 2607 QCowL2Meta *l2meta = NULL; 2608 AioTaskPool *aio = NULL; 2609 2610 trace_qcow2_writev_start_req(qemu_coroutine_self(), offset, bytes); 2611 2612 while (bytes != 0 && aio_task_pool_status(aio) == 0) { 2613 2614 l2meta = NULL; 2615 2616 trace_qcow2_writev_start_part(qemu_coroutine_self()); 2617 offset_in_cluster = offset_into_cluster(s, offset); 2618 cur_bytes = MIN(bytes, INT_MAX); 2619 if (bs->encrypted) { 2620 cur_bytes = MIN(cur_bytes, 2621 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size 2622 - offset_in_cluster); 2623 } 2624 2625 qemu_co_mutex_lock(&s->lock); 2626 2627 ret = qcow2_alloc_host_offset(bs, offset, &cur_bytes, 2628 &host_offset, &l2meta); 2629 if (ret < 0) { 2630 goto out_locked; 2631 } 2632 2633 ret = qcow2_pre_write_overlap_check(bs, 0, host_offset, 2634 cur_bytes, true); 2635 if (ret < 0) { 2636 goto out_locked; 2637 } 2638 2639 qemu_co_mutex_unlock(&s->lock); 2640 2641 if (!aio && cur_bytes != bytes) { 2642 aio = aio_task_pool_new(QCOW2_MAX_WORKERS); 2643 } 2644 ret = qcow2_add_task(bs, aio, qcow2_co_pwritev_task_entry, 0, 2645 host_offset, offset, 2646 cur_bytes, qiov, qiov_offset, l2meta); 2647 l2meta = NULL; /* l2meta is consumed by qcow2_co_pwritev_task() */ 2648 if (ret < 0) { 2649 goto fail_nometa; 2650 } 2651 2652 bytes -= cur_bytes; 2653 offset += cur_bytes; 2654 qiov_offset += cur_bytes; 2655 trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_bytes); 2656 } 2657 ret = 0; 2658 2659 qemu_co_mutex_lock(&s->lock); 2660 2661 out_locked: 2662 qcow2_handle_l2meta(bs, &l2meta, false); 2663 2664 qemu_co_mutex_unlock(&s->lock); 2665 2666 fail_nometa: 2667 if (aio) { 2668 aio_task_pool_wait_all(aio); 2669 if (ret == 0) { 2670 ret = aio_task_pool_status(aio); 2671 } 2672 g_free(aio); 2673 } 2674 2675 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret); 2676 2677 return ret; 2678 } 2679 2680 static int qcow2_inactivate(BlockDriverState *bs) 2681 { 2682 BDRVQcow2State *s = bs->opaque; 2683 int ret, result = 0; 2684 Error *local_err = NULL; 2685 2686 qcow2_store_persistent_dirty_bitmaps(bs, true, &local_err); 2687 if (local_err != NULL) { 2688 result = -EINVAL; 2689 error_reportf_err(local_err, "Lost persistent bitmaps during " 2690 "inactivation of node '%s': ", 2691 bdrv_get_device_or_node_name(bs)); 2692 } 2693 2694 ret = qcow2_cache_flush(bs, s->l2_table_cache); 2695 if (ret) { 2696 result = ret; 2697 error_report("Failed to flush the L2 table cache: %s", 2698 strerror(-ret)); 2699 } 2700 2701 ret = qcow2_cache_flush(bs, s->refcount_block_cache); 2702 if (ret) { 2703 result = ret; 2704 error_report("Failed to flush the refcount block cache: %s", 2705 strerror(-ret)); 2706 } 2707 2708 if (result == 0) { 2709 qcow2_mark_clean(bs); 2710 } 2711 2712 return result; 2713 } 2714 2715 static void qcow2_close(BlockDriverState *bs) 2716 { 2717 BDRVQcow2State *s = bs->opaque; 2718 qemu_vfree(s->l1_table); 2719 /* else pre-write overlap checks in cache_destroy may crash */ 2720 s->l1_table = NULL; 2721 2722 if (!(s->flags & BDRV_O_INACTIVE)) { 2723 qcow2_inactivate(bs); 2724 } 2725 2726 cache_clean_timer_del(bs); 2727 qcow2_cache_destroy(s->l2_table_cache); 2728 qcow2_cache_destroy(s->refcount_block_cache); 2729 2730 qcrypto_block_free(s->crypto); 2731 s->crypto = NULL; 2732 qapi_free_QCryptoBlockOpenOptions(s->crypto_opts); 2733 2734 g_free(s->unknown_header_fields); 2735 cleanup_unknown_header_ext(bs); 2736 2737 g_free(s->image_data_file); 2738 g_free(s->image_backing_file); 2739 g_free(s->image_backing_format); 2740 2741 if (has_data_file(bs)) { 2742 bdrv_unref_child(bs, s->data_file); 2743 s->data_file = NULL; 2744 } 2745 2746 qcow2_refcount_close(bs); 2747 qcow2_free_snapshots(bs); 2748 } 2749 2750 static void coroutine_fn qcow2_co_invalidate_cache(BlockDriverState *bs, 2751 Error **errp) 2752 { 2753 ERRP_GUARD(); 2754 BDRVQcow2State *s = bs->opaque; 2755 int flags = s->flags; 2756 QCryptoBlock *crypto = NULL; 2757 QDict *options; 2758 int ret; 2759 2760 /* 2761 * Backing files are read-only which makes all of their metadata immutable, 2762 * that means we don't have to worry about reopening them here. 2763 */ 2764 2765 crypto = s->crypto; 2766 s->crypto = NULL; 2767 2768 qcow2_close(bs); 2769 2770 memset(s, 0, sizeof(BDRVQcow2State)); 2771 options = qdict_clone_shallow(bs->options); 2772 2773 flags &= ~BDRV_O_INACTIVE; 2774 qemu_co_mutex_lock(&s->lock); 2775 ret = qcow2_do_open(bs, options, flags, errp); 2776 qemu_co_mutex_unlock(&s->lock); 2777 qobject_unref(options); 2778 if (ret < 0) { 2779 error_prepend(errp, "Could not reopen qcow2 layer: "); 2780 bs->drv = NULL; 2781 return; 2782 } 2783 2784 s->crypto = crypto; 2785 } 2786 2787 static size_t header_ext_add(char *buf, uint32_t magic, const void *s, 2788 size_t len, size_t buflen) 2789 { 2790 QCowExtension *ext_backing_fmt = (QCowExtension*) buf; 2791 size_t ext_len = sizeof(QCowExtension) + ((len + 7) & ~7); 2792 2793 if (buflen < ext_len) { 2794 return -ENOSPC; 2795 } 2796 2797 *ext_backing_fmt = (QCowExtension) { 2798 .magic = cpu_to_be32(magic), 2799 .len = cpu_to_be32(len), 2800 }; 2801 2802 if (len) { 2803 memcpy(buf + sizeof(QCowExtension), s, len); 2804 } 2805 2806 return ext_len; 2807 } 2808 2809 /* 2810 * Updates the qcow2 header, including the variable length parts of it, i.e. 2811 * the backing file name and all extensions. qcow2 was not designed to allow 2812 * such changes, so if we run out of space (we can only use the first cluster) 2813 * this function may fail. 2814 * 2815 * Returns 0 on success, -errno in error cases. 2816 */ 2817 int qcow2_update_header(BlockDriverState *bs) 2818 { 2819 BDRVQcow2State *s = bs->opaque; 2820 QCowHeader *header; 2821 char *buf; 2822 size_t buflen = s->cluster_size; 2823 int ret; 2824 uint64_t total_size; 2825 uint32_t refcount_table_clusters; 2826 size_t header_length; 2827 Qcow2UnknownHeaderExtension *uext; 2828 2829 buf = qemu_blockalign(bs, buflen); 2830 2831 /* Header structure */ 2832 header = (QCowHeader*) buf; 2833 2834 if (buflen < sizeof(*header)) { 2835 ret = -ENOSPC; 2836 goto fail; 2837 } 2838 2839 header_length = sizeof(*header) + s->unknown_header_fields_size; 2840 total_size = bs->total_sectors * BDRV_SECTOR_SIZE; 2841 refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3); 2842 2843 ret = validate_compression_type(s, NULL); 2844 if (ret) { 2845 goto fail; 2846 } 2847 2848 *header = (QCowHeader) { 2849 /* Version 2 fields */ 2850 .magic = cpu_to_be32(QCOW_MAGIC), 2851 .version = cpu_to_be32(s->qcow_version), 2852 .backing_file_offset = 0, 2853 .backing_file_size = 0, 2854 .cluster_bits = cpu_to_be32(s->cluster_bits), 2855 .size = cpu_to_be64(total_size), 2856 .crypt_method = cpu_to_be32(s->crypt_method_header), 2857 .l1_size = cpu_to_be32(s->l1_size), 2858 .l1_table_offset = cpu_to_be64(s->l1_table_offset), 2859 .refcount_table_offset = cpu_to_be64(s->refcount_table_offset), 2860 .refcount_table_clusters = cpu_to_be32(refcount_table_clusters), 2861 .nb_snapshots = cpu_to_be32(s->nb_snapshots), 2862 .snapshots_offset = cpu_to_be64(s->snapshots_offset), 2863 2864 /* Version 3 fields */ 2865 .incompatible_features = cpu_to_be64(s->incompatible_features), 2866 .compatible_features = cpu_to_be64(s->compatible_features), 2867 .autoclear_features = cpu_to_be64(s->autoclear_features), 2868 .refcount_order = cpu_to_be32(s->refcount_order), 2869 .header_length = cpu_to_be32(header_length), 2870 .compression_type = s->compression_type, 2871 }; 2872 2873 /* For older versions, write a shorter header */ 2874 switch (s->qcow_version) { 2875 case 2: 2876 ret = offsetof(QCowHeader, incompatible_features); 2877 break; 2878 case 3: 2879 ret = sizeof(*header); 2880 break; 2881 default: 2882 ret = -EINVAL; 2883 goto fail; 2884 } 2885 2886 buf += ret; 2887 buflen -= ret; 2888 memset(buf, 0, buflen); 2889 2890 /* Preserve any unknown field in the header */ 2891 if (s->unknown_header_fields_size) { 2892 if (buflen < s->unknown_header_fields_size) { 2893 ret = -ENOSPC; 2894 goto fail; 2895 } 2896 2897 memcpy(buf, s->unknown_header_fields, s->unknown_header_fields_size); 2898 buf += s->unknown_header_fields_size; 2899 buflen -= s->unknown_header_fields_size; 2900 } 2901 2902 /* Backing file format header extension */ 2903 if (s->image_backing_format) { 2904 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BACKING_FORMAT, 2905 s->image_backing_format, 2906 strlen(s->image_backing_format), 2907 buflen); 2908 if (ret < 0) { 2909 goto fail; 2910 } 2911 2912 buf += ret; 2913 buflen -= ret; 2914 } 2915 2916 /* External data file header extension */ 2917 if (has_data_file(bs) && s->image_data_file) { 2918 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_DATA_FILE, 2919 s->image_data_file, strlen(s->image_data_file), 2920 buflen); 2921 if (ret < 0) { 2922 goto fail; 2923 } 2924 2925 buf += ret; 2926 buflen -= ret; 2927 } 2928 2929 /* Full disk encryption header pointer extension */ 2930 if (s->crypto_header.offset != 0) { 2931 s->crypto_header.offset = cpu_to_be64(s->crypto_header.offset); 2932 s->crypto_header.length = cpu_to_be64(s->crypto_header.length); 2933 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_CRYPTO_HEADER, 2934 &s->crypto_header, sizeof(s->crypto_header), 2935 buflen); 2936 s->crypto_header.offset = be64_to_cpu(s->crypto_header.offset); 2937 s->crypto_header.length = be64_to_cpu(s->crypto_header.length); 2938 if (ret < 0) { 2939 goto fail; 2940 } 2941 buf += ret; 2942 buflen -= ret; 2943 } 2944 2945 /* 2946 * Feature table. A mere 8 feature names occupies 392 bytes, and 2947 * when coupled with the v3 minimum header of 104 bytes plus the 2948 * 8-byte end-of-extension marker, that would leave only 8 bytes 2949 * for a backing file name in an image with 512-byte clusters. 2950 * Thus, we choose to omit this header for cluster sizes 4k and 2951 * smaller. 2952 */ 2953 if (s->qcow_version >= 3 && s->cluster_size > 4096) { 2954 static const Qcow2Feature features[] = { 2955 { 2956 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE, 2957 .bit = QCOW2_INCOMPAT_DIRTY_BITNR, 2958 .name = "dirty bit", 2959 }, 2960 { 2961 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE, 2962 .bit = QCOW2_INCOMPAT_CORRUPT_BITNR, 2963 .name = "corrupt bit", 2964 }, 2965 { 2966 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE, 2967 .bit = QCOW2_INCOMPAT_DATA_FILE_BITNR, 2968 .name = "external data file", 2969 }, 2970 { 2971 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE, 2972 .bit = QCOW2_INCOMPAT_COMPRESSION_BITNR, 2973 .name = "compression type", 2974 }, 2975 { 2976 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE, 2977 .bit = QCOW2_INCOMPAT_EXTL2_BITNR, 2978 .name = "extended L2 entries", 2979 }, 2980 { 2981 .type = QCOW2_FEAT_TYPE_COMPATIBLE, 2982 .bit = QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR, 2983 .name = "lazy refcounts", 2984 }, 2985 { 2986 .type = QCOW2_FEAT_TYPE_AUTOCLEAR, 2987 .bit = QCOW2_AUTOCLEAR_BITMAPS_BITNR, 2988 .name = "bitmaps", 2989 }, 2990 { 2991 .type = QCOW2_FEAT_TYPE_AUTOCLEAR, 2992 .bit = QCOW2_AUTOCLEAR_DATA_FILE_RAW_BITNR, 2993 .name = "raw external data", 2994 }, 2995 }; 2996 2997 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_FEATURE_TABLE, 2998 features, sizeof(features), buflen); 2999 if (ret < 0) { 3000 goto fail; 3001 } 3002 buf += ret; 3003 buflen -= ret; 3004 } 3005 3006 /* Bitmap extension */ 3007 if (s->nb_bitmaps > 0) { 3008 Qcow2BitmapHeaderExt bitmaps_header = { 3009 .nb_bitmaps = cpu_to_be32(s->nb_bitmaps), 3010 .bitmap_directory_size = 3011 cpu_to_be64(s->bitmap_directory_size), 3012 .bitmap_directory_offset = 3013 cpu_to_be64(s->bitmap_directory_offset) 3014 }; 3015 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BITMAPS, 3016 &bitmaps_header, sizeof(bitmaps_header), 3017 buflen); 3018 if (ret < 0) { 3019 goto fail; 3020 } 3021 buf += ret; 3022 buflen -= ret; 3023 } 3024 3025 /* Keep unknown header extensions */ 3026 QLIST_FOREACH(uext, &s->unknown_header_ext, next) { 3027 ret = header_ext_add(buf, uext->magic, uext->data, uext->len, buflen); 3028 if (ret < 0) { 3029 goto fail; 3030 } 3031 3032 buf += ret; 3033 buflen -= ret; 3034 } 3035 3036 /* End of header extensions */ 3037 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_END, NULL, 0, buflen); 3038 if (ret < 0) { 3039 goto fail; 3040 } 3041 3042 buf += ret; 3043 buflen -= ret; 3044 3045 /* Backing file name */ 3046 if (s->image_backing_file) { 3047 size_t backing_file_len = strlen(s->image_backing_file); 3048 3049 if (buflen < backing_file_len) { 3050 ret = -ENOSPC; 3051 goto fail; 3052 } 3053 3054 /* Using strncpy is ok here, since buf is not NUL-terminated. */ 3055 strncpy(buf, s->image_backing_file, buflen); 3056 3057 header->backing_file_offset = cpu_to_be64(buf - ((char*) header)); 3058 header->backing_file_size = cpu_to_be32(backing_file_len); 3059 } 3060 3061 /* Write the new header */ 3062 ret = bdrv_pwrite(bs->file, 0, header, s->cluster_size); 3063 if (ret < 0) { 3064 goto fail; 3065 } 3066 3067 ret = 0; 3068 fail: 3069 qemu_vfree(header); 3070 return ret; 3071 } 3072 3073 static int qcow2_change_backing_file(BlockDriverState *bs, 3074 const char *backing_file, const char *backing_fmt) 3075 { 3076 BDRVQcow2State *s = bs->opaque; 3077 3078 /* Adding a backing file means that the external data file alone won't be 3079 * enough to make sense of the content */ 3080 if (backing_file && data_file_is_raw(bs)) { 3081 return -EINVAL; 3082 } 3083 3084 if (backing_file && strlen(backing_file) > 1023) { 3085 return -EINVAL; 3086 } 3087 3088 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file), 3089 backing_file ?: ""); 3090 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: ""); 3091 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: ""); 3092 3093 g_free(s->image_backing_file); 3094 g_free(s->image_backing_format); 3095 3096 s->image_backing_file = backing_file ? g_strdup(bs->backing_file) : NULL; 3097 s->image_backing_format = backing_fmt ? g_strdup(bs->backing_format) : NULL; 3098 3099 return qcow2_update_header(bs); 3100 } 3101 3102 static int qcow2_set_up_encryption(BlockDriverState *bs, 3103 QCryptoBlockCreateOptions *cryptoopts, 3104 Error **errp) 3105 { 3106 BDRVQcow2State *s = bs->opaque; 3107 QCryptoBlock *crypto = NULL; 3108 int fmt, ret; 3109 3110 switch (cryptoopts->format) { 3111 case Q_CRYPTO_BLOCK_FORMAT_LUKS: 3112 fmt = QCOW_CRYPT_LUKS; 3113 break; 3114 case Q_CRYPTO_BLOCK_FORMAT_QCOW: 3115 fmt = QCOW_CRYPT_AES; 3116 break; 3117 default: 3118 error_setg(errp, "Crypto format not supported in qcow2"); 3119 return -EINVAL; 3120 } 3121 3122 s->crypt_method_header = fmt; 3123 3124 crypto = qcrypto_block_create(cryptoopts, "encrypt.", 3125 qcow2_crypto_hdr_init_func, 3126 qcow2_crypto_hdr_write_func, 3127 bs, errp); 3128 if (!crypto) { 3129 return -EINVAL; 3130 } 3131 3132 ret = qcow2_update_header(bs); 3133 if (ret < 0) { 3134 error_setg_errno(errp, -ret, "Could not write encryption header"); 3135 goto out; 3136 } 3137 3138 ret = 0; 3139 out: 3140 qcrypto_block_free(crypto); 3141 return ret; 3142 } 3143 3144 /** 3145 * Preallocates metadata structures for data clusters between @offset (in the 3146 * guest disk) and @new_length (which is thus generally the new guest disk 3147 * size). 3148 * 3149 * Returns: 0 on success, -errno on failure. 3150 */ 3151 static int coroutine_fn preallocate_co(BlockDriverState *bs, uint64_t offset, 3152 uint64_t new_length, PreallocMode mode, 3153 Error **errp) 3154 { 3155 BDRVQcow2State *s = bs->opaque; 3156 uint64_t bytes; 3157 uint64_t host_offset = 0; 3158 int64_t file_length; 3159 unsigned int cur_bytes; 3160 int ret; 3161 QCowL2Meta *meta = NULL, *m; 3162 3163 assert(offset <= new_length); 3164 bytes = new_length - offset; 3165 3166 while (bytes) { 3167 cur_bytes = MIN(bytes, QEMU_ALIGN_DOWN(INT_MAX, s->cluster_size)); 3168 ret = qcow2_alloc_host_offset(bs, offset, &cur_bytes, 3169 &host_offset, &meta); 3170 if (ret < 0) { 3171 error_setg_errno(errp, -ret, "Allocating clusters failed"); 3172 goto out; 3173 } 3174 3175 for (m = meta; m != NULL; m = m->next) { 3176 m->prealloc = true; 3177 } 3178 3179 ret = qcow2_handle_l2meta(bs, &meta, true); 3180 if (ret < 0) { 3181 error_setg_errno(errp, -ret, "Mapping clusters failed"); 3182 goto out; 3183 } 3184 3185 /* TODO Preallocate data if requested */ 3186 3187 bytes -= cur_bytes; 3188 offset += cur_bytes; 3189 } 3190 3191 /* 3192 * It is expected that the image file is large enough to actually contain 3193 * all of the allocated clusters (otherwise we get failing reads after 3194 * EOF). Extend the image to the last allocated sector. 3195 */ 3196 file_length = bdrv_getlength(s->data_file->bs); 3197 if (file_length < 0) { 3198 error_setg_errno(errp, -file_length, "Could not get file size"); 3199 ret = file_length; 3200 goto out; 3201 } 3202 3203 if (host_offset + cur_bytes > file_length) { 3204 if (mode == PREALLOC_MODE_METADATA) { 3205 mode = PREALLOC_MODE_OFF; 3206 } 3207 ret = bdrv_co_truncate(s->data_file, host_offset + cur_bytes, false, 3208 mode, 0, errp); 3209 if (ret < 0) { 3210 goto out; 3211 } 3212 } 3213 3214 ret = 0; 3215 3216 out: 3217 qcow2_handle_l2meta(bs, &meta, false); 3218 return ret; 3219 } 3220 3221 /* qcow2_refcount_metadata_size: 3222 * @clusters: number of clusters to refcount (including data and L1/L2 tables) 3223 * @cluster_size: size of a cluster, in bytes 3224 * @refcount_order: refcount bits power-of-2 exponent 3225 * @generous_increase: allow for the refcount table to be 1.5x as large as it 3226 * needs to be 3227 * 3228 * Returns: Number of bytes required for refcount blocks and table metadata. 3229 */ 3230 int64_t qcow2_refcount_metadata_size(int64_t clusters, size_t cluster_size, 3231 int refcount_order, bool generous_increase, 3232 uint64_t *refblock_count) 3233 { 3234 /* 3235 * Every host cluster is reference-counted, including metadata (even 3236 * refcount metadata is recursively included). 3237 * 3238 * An accurate formula for the size of refcount metadata size is difficult 3239 * to derive. An easier method of calculation is finding the fixed point 3240 * where no further refcount blocks or table clusters are required to 3241 * reference count every cluster. 3242 */ 3243 int64_t blocks_per_table_cluster = cluster_size / REFTABLE_ENTRY_SIZE; 3244 int64_t refcounts_per_block = cluster_size * 8 / (1 << refcount_order); 3245 int64_t table = 0; /* number of refcount table clusters */ 3246 int64_t blocks = 0; /* number of refcount block clusters */ 3247 int64_t last; 3248 int64_t n = 0; 3249 3250 do { 3251 last = n; 3252 blocks = DIV_ROUND_UP(clusters + table + blocks, refcounts_per_block); 3253 table = DIV_ROUND_UP(blocks, blocks_per_table_cluster); 3254 n = clusters + blocks + table; 3255 3256 if (n == last && generous_increase) { 3257 clusters += DIV_ROUND_UP(table, 2); 3258 n = 0; /* force another loop */ 3259 generous_increase = false; 3260 } 3261 } while (n != last); 3262 3263 if (refblock_count) { 3264 *refblock_count = blocks; 3265 } 3266 3267 return (blocks + table) * cluster_size; 3268 } 3269 3270 /** 3271 * qcow2_calc_prealloc_size: 3272 * @total_size: virtual disk size in bytes 3273 * @cluster_size: cluster size in bytes 3274 * @refcount_order: refcount bits power-of-2 exponent 3275 * @extended_l2: true if the image has extended L2 entries 3276 * 3277 * Returns: Total number of bytes required for the fully allocated image 3278 * (including metadata). 3279 */ 3280 static int64_t qcow2_calc_prealloc_size(int64_t total_size, 3281 size_t cluster_size, 3282 int refcount_order, 3283 bool extended_l2) 3284 { 3285 int64_t meta_size = 0; 3286 uint64_t nl1e, nl2e; 3287 int64_t aligned_total_size = ROUND_UP(total_size, cluster_size); 3288 size_t l2e_size = extended_l2 ? L2E_SIZE_EXTENDED : L2E_SIZE_NORMAL; 3289 3290 /* header: 1 cluster */ 3291 meta_size += cluster_size; 3292 3293 /* total size of L2 tables */ 3294 nl2e = aligned_total_size / cluster_size; 3295 nl2e = ROUND_UP(nl2e, cluster_size / l2e_size); 3296 meta_size += nl2e * l2e_size; 3297 3298 /* total size of L1 tables */ 3299 nl1e = nl2e * l2e_size / cluster_size; 3300 nl1e = ROUND_UP(nl1e, cluster_size / L1E_SIZE); 3301 meta_size += nl1e * L1E_SIZE; 3302 3303 /* total size of refcount table and blocks */ 3304 meta_size += qcow2_refcount_metadata_size( 3305 (meta_size + aligned_total_size) / cluster_size, 3306 cluster_size, refcount_order, false, NULL); 3307 3308 return meta_size + aligned_total_size; 3309 } 3310 3311 static bool validate_cluster_size(size_t cluster_size, bool extended_l2, 3312 Error **errp) 3313 { 3314 int cluster_bits = ctz32(cluster_size); 3315 if (cluster_bits < MIN_CLUSTER_BITS || cluster_bits > MAX_CLUSTER_BITS || 3316 (1 << cluster_bits) != cluster_size) 3317 { 3318 error_setg(errp, "Cluster size must be a power of two between %d and " 3319 "%dk", 1 << MIN_CLUSTER_BITS, 1 << (MAX_CLUSTER_BITS - 10)); 3320 return false; 3321 } 3322 3323 if (extended_l2) { 3324 unsigned min_cluster_size = 3325 (1 << MIN_CLUSTER_BITS) * QCOW_EXTL2_SUBCLUSTERS_PER_CLUSTER; 3326 if (cluster_size < min_cluster_size) { 3327 error_setg(errp, "Extended L2 entries are only supported with " 3328 "cluster sizes of at least %u bytes", min_cluster_size); 3329 return false; 3330 } 3331 } 3332 3333 return true; 3334 } 3335 3336 static size_t qcow2_opt_get_cluster_size_del(QemuOpts *opts, bool extended_l2, 3337 Error **errp) 3338 { 3339 size_t cluster_size; 3340 3341 cluster_size = qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE, 3342 DEFAULT_CLUSTER_SIZE); 3343 if (!validate_cluster_size(cluster_size, extended_l2, errp)) { 3344 return 0; 3345 } 3346 return cluster_size; 3347 } 3348 3349 static int qcow2_opt_get_version_del(QemuOpts *opts, Error **errp) 3350 { 3351 char *buf; 3352 int ret; 3353 3354 buf = qemu_opt_get_del(opts, BLOCK_OPT_COMPAT_LEVEL); 3355 if (!buf) { 3356 ret = 3; /* default */ 3357 } else if (!strcmp(buf, "0.10")) { 3358 ret = 2; 3359 } else if (!strcmp(buf, "1.1")) { 3360 ret = 3; 3361 } else { 3362 error_setg(errp, "Invalid compatibility level: '%s'", buf); 3363 ret = -EINVAL; 3364 } 3365 g_free(buf); 3366 return ret; 3367 } 3368 3369 static uint64_t qcow2_opt_get_refcount_bits_del(QemuOpts *opts, int version, 3370 Error **errp) 3371 { 3372 uint64_t refcount_bits; 3373 3374 refcount_bits = qemu_opt_get_number_del(opts, BLOCK_OPT_REFCOUNT_BITS, 16); 3375 if (refcount_bits > 64 || !is_power_of_2(refcount_bits)) { 3376 error_setg(errp, "Refcount width must be a power of two and may not " 3377 "exceed 64 bits"); 3378 return 0; 3379 } 3380 3381 if (version < 3 && refcount_bits != 16) { 3382 error_setg(errp, "Different refcount widths than 16 bits require " 3383 "compatibility level 1.1 or above (use compat=1.1 or " 3384 "greater)"); 3385 return 0; 3386 } 3387 3388 return refcount_bits; 3389 } 3390 3391 static int coroutine_fn 3392 qcow2_co_create(BlockdevCreateOptions *create_options, Error **errp) 3393 { 3394 BlockdevCreateOptionsQcow2 *qcow2_opts; 3395 QDict *options; 3396 3397 /* 3398 * Open the image file and write a minimal qcow2 header. 3399 * 3400 * We keep things simple and start with a zero-sized image. We also 3401 * do without refcount blocks or a L1 table for now. We'll fix the 3402 * inconsistency later. 3403 * 3404 * We do need a refcount table because growing the refcount table means 3405 * allocating two new refcount blocks - the second of which would be at 3406 * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file 3407 * size for any qcow2 image. 3408 */ 3409 BlockBackend *blk = NULL; 3410 BlockDriverState *bs = NULL; 3411 BlockDriverState *data_bs = NULL; 3412 QCowHeader *header; 3413 size_t cluster_size; 3414 int version; 3415 int refcount_order; 3416 uint64_t *refcount_table; 3417 int ret; 3418 uint8_t compression_type = QCOW2_COMPRESSION_TYPE_ZLIB; 3419 3420 assert(create_options->driver == BLOCKDEV_DRIVER_QCOW2); 3421 qcow2_opts = &create_options->u.qcow2; 3422 3423 bs = bdrv_open_blockdev_ref(qcow2_opts->file, errp); 3424 if (bs == NULL) { 3425 return -EIO; 3426 } 3427 3428 /* Validate options and set default values */ 3429 if (!QEMU_IS_ALIGNED(qcow2_opts->size, BDRV_SECTOR_SIZE)) { 3430 error_setg(errp, "Image size must be a multiple of %u bytes", 3431 (unsigned) BDRV_SECTOR_SIZE); 3432 ret = -EINVAL; 3433 goto out; 3434 } 3435 3436 if (qcow2_opts->has_version) { 3437 switch (qcow2_opts->version) { 3438 case BLOCKDEV_QCOW2_VERSION_V2: 3439 version = 2; 3440 break; 3441 case BLOCKDEV_QCOW2_VERSION_V3: 3442 version = 3; 3443 break; 3444 default: 3445 g_assert_not_reached(); 3446 } 3447 } else { 3448 version = 3; 3449 } 3450 3451 if (qcow2_opts->has_cluster_size) { 3452 cluster_size = qcow2_opts->cluster_size; 3453 } else { 3454 cluster_size = DEFAULT_CLUSTER_SIZE; 3455 } 3456 3457 if (!qcow2_opts->has_extended_l2) { 3458 qcow2_opts->extended_l2 = false; 3459 } 3460 if (qcow2_opts->extended_l2) { 3461 if (version < 3) { 3462 error_setg(errp, "Extended L2 entries are only supported with " 3463 "compatibility level 1.1 and above (use version=v3 or " 3464 "greater)"); 3465 ret = -EINVAL; 3466 goto out; 3467 } 3468 } 3469 3470 if (!validate_cluster_size(cluster_size, qcow2_opts->extended_l2, errp)) { 3471 ret = -EINVAL; 3472 goto out; 3473 } 3474 3475 if (!qcow2_opts->has_preallocation) { 3476 qcow2_opts->preallocation = PREALLOC_MODE_OFF; 3477 } 3478 if (qcow2_opts->has_backing_file && 3479 qcow2_opts->preallocation != PREALLOC_MODE_OFF && 3480 !qcow2_opts->extended_l2) 3481 { 3482 error_setg(errp, "Backing file and preallocation can only be used at " 3483 "the same time if extended_l2 is on"); 3484 ret = -EINVAL; 3485 goto out; 3486 } 3487 if (qcow2_opts->has_backing_fmt && !qcow2_opts->has_backing_file) { 3488 error_setg(errp, "Backing format cannot be used without backing file"); 3489 ret = -EINVAL; 3490 goto out; 3491 } 3492 3493 if (!qcow2_opts->has_lazy_refcounts) { 3494 qcow2_opts->lazy_refcounts = false; 3495 } 3496 if (version < 3 && qcow2_opts->lazy_refcounts) { 3497 error_setg(errp, "Lazy refcounts only supported with compatibility " 3498 "level 1.1 and above (use version=v3 or greater)"); 3499 ret = -EINVAL; 3500 goto out; 3501 } 3502 3503 if (!qcow2_opts->has_refcount_bits) { 3504 qcow2_opts->refcount_bits = 16; 3505 } 3506 if (qcow2_opts->refcount_bits > 64 || 3507 !is_power_of_2(qcow2_opts->refcount_bits)) 3508 { 3509 error_setg(errp, "Refcount width must be a power of two and may not " 3510 "exceed 64 bits"); 3511 ret = -EINVAL; 3512 goto out; 3513 } 3514 if (version < 3 && qcow2_opts->refcount_bits != 16) { 3515 error_setg(errp, "Different refcount widths than 16 bits require " 3516 "compatibility level 1.1 or above (use version=v3 or " 3517 "greater)"); 3518 ret = -EINVAL; 3519 goto out; 3520 } 3521 refcount_order = ctz32(qcow2_opts->refcount_bits); 3522 3523 if (qcow2_opts->data_file_raw && !qcow2_opts->data_file) { 3524 error_setg(errp, "data-file-raw requires data-file"); 3525 ret = -EINVAL; 3526 goto out; 3527 } 3528 if (qcow2_opts->data_file_raw && qcow2_opts->has_backing_file) { 3529 error_setg(errp, "Backing file and data-file-raw cannot be used at " 3530 "the same time"); 3531 ret = -EINVAL; 3532 goto out; 3533 } 3534 if (qcow2_opts->data_file_raw && 3535 qcow2_opts->preallocation == PREALLOC_MODE_OFF) 3536 { 3537 /* 3538 * data-file-raw means that "the external data file can be 3539 * read as a consistent standalone raw image without looking 3540 * at the qcow2 metadata." It does not say that the metadata 3541 * must be ignored, though (and the qcow2 driver in fact does 3542 * not ignore it), so the L1/L2 tables must be present and 3543 * give a 1:1 mapping, so you get the same result regardless 3544 * of whether you look at the metadata or whether you ignore 3545 * it. 3546 */ 3547 qcow2_opts->preallocation = PREALLOC_MODE_METADATA; 3548 3549 /* 3550 * Cannot use preallocation with backing files, but giving a 3551 * backing file when specifying data_file_raw is an error 3552 * anyway. 3553 */ 3554 assert(!qcow2_opts->has_backing_file); 3555 } 3556 3557 if (qcow2_opts->data_file) { 3558 if (version < 3) { 3559 error_setg(errp, "External data files are only supported with " 3560 "compatibility level 1.1 and above (use version=v3 or " 3561 "greater)"); 3562 ret = -EINVAL; 3563 goto out; 3564 } 3565 data_bs = bdrv_open_blockdev_ref(qcow2_opts->data_file, errp); 3566 if (data_bs == NULL) { 3567 ret = -EIO; 3568 goto out; 3569 } 3570 } 3571 3572 if (qcow2_opts->has_compression_type && 3573 qcow2_opts->compression_type != QCOW2_COMPRESSION_TYPE_ZLIB) { 3574 3575 ret = -EINVAL; 3576 3577 if (version < 3) { 3578 error_setg(errp, "Non-zlib compression type is only supported with " 3579 "compatibility level 1.1 and above (use version=v3 or " 3580 "greater)"); 3581 goto out; 3582 } 3583 3584 switch (qcow2_opts->compression_type) { 3585 #ifdef CONFIG_ZSTD 3586 case QCOW2_COMPRESSION_TYPE_ZSTD: 3587 break; 3588 #endif 3589 default: 3590 error_setg(errp, "Unknown compression type"); 3591 goto out; 3592 } 3593 3594 compression_type = qcow2_opts->compression_type; 3595 } 3596 3597 /* Create BlockBackend to write to the image */ 3598 blk = blk_new_with_bs(bs, BLK_PERM_WRITE | BLK_PERM_RESIZE, BLK_PERM_ALL, 3599 errp); 3600 if (!blk) { 3601 ret = -EPERM; 3602 goto out; 3603 } 3604 blk_set_allow_write_beyond_eof(blk, true); 3605 3606 /* Write the header */ 3607 QEMU_BUILD_BUG_ON((1 << MIN_CLUSTER_BITS) < sizeof(*header)); 3608 header = g_malloc0(cluster_size); 3609 *header = (QCowHeader) { 3610 .magic = cpu_to_be32(QCOW_MAGIC), 3611 .version = cpu_to_be32(version), 3612 .cluster_bits = cpu_to_be32(ctz32(cluster_size)), 3613 .size = cpu_to_be64(0), 3614 .l1_table_offset = cpu_to_be64(0), 3615 .l1_size = cpu_to_be32(0), 3616 .refcount_table_offset = cpu_to_be64(cluster_size), 3617 .refcount_table_clusters = cpu_to_be32(1), 3618 .refcount_order = cpu_to_be32(refcount_order), 3619 /* don't deal with endianness since compression_type is 1 byte long */ 3620 .compression_type = compression_type, 3621 .header_length = cpu_to_be32(sizeof(*header)), 3622 }; 3623 3624 /* We'll update this to correct value later */ 3625 header->crypt_method = cpu_to_be32(QCOW_CRYPT_NONE); 3626 3627 if (qcow2_opts->lazy_refcounts) { 3628 header->compatible_features |= 3629 cpu_to_be64(QCOW2_COMPAT_LAZY_REFCOUNTS); 3630 } 3631 if (data_bs) { 3632 header->incompatible_features |= 3633 cpu_to_be64(QCOW2_INCOMPAT_DATA_FILE); 3634 } 3635 if (qcow2_opts->data_file_raw) { 3636 header->autoclear_features |= 3637 cpu_to_be64(QCOW2_AUTOCLEAR_DATA_FILE_RAW); 3638 } 3639 if (compression_type != QCOW2_COMPRESSION_TYPE_ZLIB) { 3640 header->incompatible_features |= 3641 cpu_to_be64(QCOW2_INCOMPAT_COMPRESSION); 3642 } 3643 3644 if (qcow2_opts->extended_l2) { 3645 header->incompatible_features |= 3646 cpu_to_be64(QCOW2_INCOMPAT_EXTL2); 3647 } 3648 3649 ret = blk_pwrite(blk, 0, header, cluster_size, 0); 3650 g_free(header); 3651 if (ret < 0) { 3652 error_setg_errno(errp, -ret, "Could not write qcow2 header"); 3653 goto out; 3654 } 3655 3656 /* Write a refcount table with one refcount block */ 3657 refcount_table = g_malloc0(2 * cluster_size); 3658 refcount_table[0] = cpu_to_be64(2 * cluster_size); 3659 ret = blk_pwrite(blk, cluster_size, refcount_table, 2 * cluster_size, 0); 3660 g_free(refcount_table); 3661 3662 if (ret < 0) { 3663 error_setg_errno(errp, -ret, "Could not write refcount table"); 3664 goto out; 3665 } 3666 3667 blk_unref(blk); 3668 blk = NULL; 3669 3670 /* 3671 * And now open the image and make it consistent first (i.e. increase the 3672 * refcount of the cluster that is occupied by the header and the refcount 3673 * table) 3674 */ 3675 options = qdict_new(); 3676 qdict_put_str(options, "driver", "qcow2"); 3677 qdict_put_str(options, "file", bs->node_name); 3678 if (data_bs) { 3679 qdict_put_str(options, "data-file", data_bs->node_name); 3680 } 3681 blk = blk_new_open(NULL, NULL, options, 3682 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_NO_FLUSH, 3683 errp); 3684 if (blk == NULL) { 3685 ret = -EIO; 3686 goto out; 3687 } 3688 3689 ret = qcow2_alloc_clusters(blk_bs(blk), 3 * cluster_size); 3690 if (ret < 0) { 3691 error_setg_errno(errp, -ret, "Could not allocate clusters for qcow2 " 3692 "header and refcount table"); 3693 goto out; 3694 3695 } else if (ret != 0) { 3696 error_report("Huh, first cluster in empty image is already in use?"); 3697 abort(); 3698 } 3699 3700 /* Set the external data file if necessary */ 3701 if (data_bs) { 3702 BDRVQcow2State *s = blk_bs(blk)->opaque; 3703 s->image_data_file = g_strdup(data_bs->filename); 3704 } 3705 3706 /* Create a full header (including things like feature table) */ 3707 ret = qcow2_update_header(blk_bs(blk)); 3708 if (ret < 0) { 3709 error_setg_errno(errp, -ret, "Could not update qcow2 header"); 3710 goto out; 3711 } 3712 3713 /* Okay, now that we have a valid image, let's give it the right size */ 3714 ret = blk_truncate(blk, qcow2_opts->size, false, qcow2_opts->preallocation, 3715 0, errp); 3716 if (ret < 0) { 3717 error_prepend(errp, "Could not resize image: "); 3718 goto out; 3719 } 3720 3721 /* Want a backing file? There you go. */ 3722 if (qcow2_opts->has_backing_file) { 3723 const char *backing_format = NULL; 3724 3725 if (qcow2_opts->has_backing_fmt) { 3726 backing_format = BlockdevDriver_str(qcow2_opts->backing_fmt); 3727 } 3728 3729 ret = bdrv_change_backing_file(blk_bs(blk), qcow2_opts->backing_file, 3730 backing_format, false); 3731 if (ret < 0) { 3732 error_setg_errno(errp, -ret, "Could not assign backing file '%s' " 3733 "with format '%s'", qcow2_opts->backing_file, 3734 backing_format); 3735 goto out; 3736 } 3737 } 3738 3739 /* Want encryption? There you go. */ 3740 if (qcow2_opts->has_encrypt) { 3741 ret = qcow2_set_up_encryption(blk_bs(blk), qcow2_opts->encrypt, errp); 3742 if (ret < 0) { 3743 goto out; 3744 } 3745 } 3746 3747 blk_unref(blk); 3748 blk = NULL; 3749 3750 /* Reopen the image without BDRV_O_NO_FLUSH to flush it before returning. 3751 * Using BDRV_O_NO_IO, since encryption is now setup we don't want to 3752 * have to setup decryption context. We're not doing any I/O on the top 3753 * level BlockDriverState, only lower layers, where BDRV_O_NO_IO does 3754 * not have effect. 3755 */ 3756 options = qdict_new(); 3757 qdict_put_str(options, "driver", "qcow2"); 3758 qdict_put_str(options, "file", bs->node_name); 3759 if (data_bs) { 3760 qdict_put_str(options, "data-file", data_bs->node_name); 3761 } 3762 blk = blk_new_open(NULL, NULL, options, 3763 BDRV_O_RDWR | BDRV_O_NO_BACKING | BDRV_O_NO_IO, 3764 errp); 3765 if (blk == NULL) { 3766 ret = -EIO; 3767 goto out; 3768 } 3769 3770 ret = 0; 3771 out: 3772 blk_unref(blk); 3773 bdrv_unref(bs); 3774 bdrv_unref(data_bs); 3775 return ret; 3776 } 3777 3778 static int coroutine_fn qcow2_co_create_opts(BlockDriver *drv, 3779 const char *filename, 3780 QemuOpts *opts, 3781 Error **errp) 3782 { 3783 BlockdevCreateOptions *create_options = NULL; 3784 QDict *qdict; 3785 Visitor *v; 3786 BlockDriverState *bs = NULL; 3787 BlockDriverState *data_bs = NULL; 3788 const char *val; 3789 int ret; 3790 3791 /* Only the keyval visitor supports the dotted syntax needed for 3792 * encryption, so go through a QDict before getting a QAPI type. Ignore 3793 * options meant for the protocol layer so that the visitor doesn't 3794 * complain. */ 3795 qdict = qemu_opts_to_qdict_filtered(opts, NULL, bdrv_qcow2.create_opts, 3796 true); 3797 3798 /* Handle encryption options */ 3799 val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT); 3800 if (val && !strcmp(val, "on")) { 3801 qdict_put_str(qdict, BLOCK_OPT_ENCRYPT, "qcow"); 3802 } else if (val && !strcmp(val, "off")) { 3803 qdict_del(qdict, BLOCK_OPT_ENCRYPT); 3804 } 3805 3806 val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT); 3807 if (val && !strcmp(val, "aes")) { 3808 qdict_put_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT, "qcow"); 3809 } 3810 3811 /* Convert compat=0.10/1.1 into compat=v2/v3, to be renamed into 3812 * version=v2/v3 below. */ 3813 val = qdict_get_try_str(qdict, BLOCK_OPT_COMPAT_LEVEL); 3814 if (val && !strcmp(val, "0.10")) { 3815 qdict_put_str(qdict, BLOCK_OPT_COMPAT_LEVEL, "v2"); 3816 } else if (val && !strcmp(val, "1.1")) { 3817 qdict_put_str(qdict, BLOCK_OPT_COMPAT_LEVEL, "v3"); 3818 } 3819 3820 /* Change legacy command line options into QMP ones */ 3821 static const QDictRenames opt_renames[] = { 3822 { BLOCK_OPT_BACKING_FILE, "backing-file" }, 3823 { BLOCK_OPT_BACKING_FMT, "backing-fmt" }, 3824 { BLOCK_OPT_CLUSTER_SIZE, "cluster-size" }, 3825 { BLOCK_OPT_LAZY_REFCOUNTS, "lazy-refcounts" }, 3826 { BLOCK_OPT_EXTL2, "extended-l2" }, 3827 { BLOCK_OPT_REFCOUNT_BITS, "refcount-bits" }, 3828 { BLOCK_OPT_ENCRYPT, BLOCK_OPT_ENCRYPT_FORMAT }, 3829 { BLOCK_OPT_COMPAT_LEVEL, "version" }, 3830 { BLOCK_OPT_DATA_FILE_RAW, "data-file-raw" }, 3831 { BLOCK_OPT_COMPRESSION_TYPE, "compression-type" }, 3832 { NULL, NULL }, 3833 }; 3834 3835 if (!qdict_rename_keys(qdict, opt_renames, errp)) { 3836 ret = -EINVAL; 3837 goto finish; 3838 } 3839 3840 /* Create and open the file (protocol layer) */ 3841 ret = bdrv_create_file(filename, opts, errp); 3842 if (ret < 0) { 3843 goto finish; 3844 } 3845 3846 bs = bdrv_open(filename, NULL, NULL, 3847 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp); 3848 if (bs == NULL) { 3849 ret = -EIO; 3850 goto finish; 3851 } 3852 3853 /* Create and open an external data file (protocol layer) */ 3854 val = qdict_get_try_str(qdict, BLOCK_OPT_DATA_FILE); 3855 if (val) { 3856 ret = bdrv_create_file(val, opts, errp); 3857 if (ret < 0) { 3858 goto finish; 3859 } 3860 3861 data_bs = bdrv_open(val, NULL, NULL, 3862 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, 3863 errp); 3864 if (data_bs == NULL) { 3865 ret = -EIO; 3866 goto finish; 3867 } 3868 3869 qdict_del(qdict, BLOCK_OPT_DATA_FILE); 3870 qdict_put_str(qdict, "data-file", data_bs->node_name); 3871 } 3872 3873 /* Set 'driver' and 'node' options */ 3874 qdict_put_str(qdict, "driver", "qcow2"); 3875 qdict_put_str(qdict, "file", bs->node_name); 3876 3877 /* Now get the QAPI type BlockdevCreateOptions */ 3878 v = qobject_input_visitor_new_flat_confused(qdict, errp); 3879 if (!v) { 3880 ret = -EINVAL; 3881 goto finish; 3882 } 3883 3884 visit_type_BlockdevCreateOptions(v, NULL, &create_options, errp); 3885 visit_free(v); 3886 if (!create_options) { 3887 ret = -EINVAL; 3888 goto finish; 3889 } 3890 3891 /* Silently round up size */ 3892 create_options->u.qcow2.size = ROUND_UP(create_options->u.qcow2.size, 3893 BDRV_SECTOR_SIZE); 3894 3895 /* Create the qcow2 image (format layer) */ 3896 ret = qcow2_co_create(create_options, errp); 3897 finish: 3898 if (ret < 0) { 3899 bdrv_co_delete_file_noerr(bs); 3900 bdrv_co_delete_file_noerr(data_bs); 3901 } else { 3902 ret = 0; 3903 } 3904 3905 qobject_unref(qdict); 3906 bdrv_unref(bs); 3907 bdrv_unref(data_bs); 3908 qapi_free_BlockdevCreateOptions(create_options); 3909 return ret; 3910 } 3911 3912 3913 static bool is_zero(BlockDriverState *bs, int64_t offset, int64_t bytes) 3914 { 3915 int64_t nr; 3916 int res; 3917 3918 /* Clamp to image length, before checking status of underlying sectors */ 3919 if (offset + bytes > bs->total_sectors * BDRV_SECTOR_SIZE) { 3920 bytes = bs->total_sectors * BDRV_SECTOR_SIZE - offset; 3921 } 3922 3923 if (!bytes) { 3924 return true; 3925 } 3926 3927 /* 3928 * bdrv_block_status_above doesn't merge different types of zeros, for 3929 * example, zeros which come from the region which is unallocated in 3930 * the whole backing chain, and zeros which come because of a short 3931 * backing file. So, we need a loop. 3932 */ 3933 do { 3934 res = bdrv_block_status_above(bs, NULL, offset, bytes, &nr, NULL, NULL); 3935 offset += nr; 3936 bytes -= nr; 3937 } while (res >= 0 && (res & BDRV_BLOCK_ZERO) && nr && bytes); 3938 3939 return res >= 0 && (res & BDRV_BLOCK_ZERO) && bytes == 0; 3940 } 3941 3942 static coroutine_fn int qcow2_co_pwrite_zeroes(BlockDriverState *bs, 3943 int64_t offset, int bytes, BdrvRequestFlags flags) 3944 { 3945 int ret; 3946 BDRVQcow2State *s = bs->opaque; 3947 3948 uint32_t head = offset_into_subcluster(s, offset); 3949 uint32_t tail = ROUND_UP(offset + bytes, s->subcluster_size) - 3950 (offset + bytes); 3951 3952 trace_qcow2_pwrite_zeroes_start_req(qemu_coroutine_self(), offset, bytes); 3953 if (offset + bytes == bs->total_sectors * BDRV_SECTOR_SIZE) { 3954 tail = 0; 3955 } 3956 3957 if (head || tail) { 3958 uint64_t off; 3959 unsigned int nr; 3960 QCow2SubclusterType type; 3961 3962 assert(head + bytes + tail <= s->subcluster_size); 3963 3964 /* check whether remainder of cluster already reads as zero */ 3965 if (!(is_zero(bs, offset - head, head) && 3966 is_zero(bs, offset + bytes, tail))) { 3967 return -ENOTSUP; 3968 } 3969 3970 qemu_co_mutex_lock(&s->lock); 3971 /* We can have new write after previous check */ 3972 offset -= head; 3973 bytes = s->subcluster_size; 3974 nr = s->subcluster_size; 3975 ret = qcow2_get_host_offset(bs, offset, &nr, &off, &type); 3976 if (ret < 0 || 3977 (type != QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN && 3978 type != QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC && 3979 type != QCOW2_SUBCLUSTER_ZERO_PLAIN && 3980 type != QCOW2_SUBCLUSTER_ZERO_ALLOC)) { 3981 qemu_co_mutex_unlock(&s->lock); 3982 return ret < 0 ? ret : -ENOTSUP; 3983 } 3984 } else { 3985 qemu_co_mutex_lock(&s->lock); 3986 } 3987 3988 trace_qcow2_pwrite_zeroes(qemu_coroutine_self(), offset, bytes); 3989 3990 /* Whatever is left can use real zero subclusters */ 3991 ret = qcow2_subcluster_zeroize(bs, offset, bytes, flags); 3992 qemu_co_mutex_unlock(&s->lock); 3993 3994 return ret; 3995 } 3996 3997 static coroutine_fn int qcow2_co_pdiscard(BlockDriverState *bs, 3998 int64_t offset, int bytes) 3999 { 4000 int ret; 4001 BDRVQcow2State *s = bs->opaque; 4002 4003 /* If the image does not support QCOW_OFLAG_ZERO then discarding 4004 * clusters could expose stale data from the backing file. */ 4005 if (s->qcow_version < 3 && bs->backing) { 4006 return -ENOTSUP; 4007 } 4008 4009 if (!QEMU_IS_ALIGNED(offset | bytes, s->cluster_size)) { 4010 assert(bytes < s->cluster_size); 4011 /* Ignore partial clusters, except for the special case of the 4012 * complete partial cluster at the end of an unaligned file */ 4013 if (!QEMU_IS_ALIGNED(offset, s->cluster_size) || 4014 offset + bytes != bs->total_sectors * BDRV_SECTOR_SIZE) { 4015 return -ENOTSUP; 4016 } 4017 } 4018 4019 qemu_co_mutex_lock(&s->lock); 4020 ret = qcow2_cluster_discard(bs, offset, bytes, QCOW2_DISCARD_REQUEST, 4021 false); 4022 qemu_co_mutex_unlock(&s->lock); 4023 return ret; 4024 } 4025 4026 static int coroutine_fn 4027 qcow2_co_copy_range_from(BlockDriverState *bs, 4028 BdrvChild *src, uint64_t src_offset, 4029 BdrvChild *dst, uint64_t dst_offset, 4030 uint64_t bytes, BdrvRequestFlags read_flags, 4031 BdrvRequestFlags write_flags) 4032 { 4033 BDRVQcow2State *s = bs->opaque; 4034 int ret; 4035 unsigned int cur_bytes; /* number of bytes in current iteration */ 4036 BdrvChild *child = NULL; 4037 BdrvRequestFlags cur_write_flags; 4038 4039 assert(!bs->encrypted); 4040 qemu_co_mutex_lock(&s->lock); 4041 4042 while (bytes != 0) { 4043 uint64_t copy_offset = 0; 4044 QCow2SubclusterType type; 4045 /* prepare next request */ 4046 cur_bytes = MIN(bytes, INT_MAX); 4047 cur_write_flags = write_flags; 4048 4049 ret = qcow2_get_host_offset(bs, src_offset, &cur_bytes, 4050 ©_offset, &type); 4051 if (ret < 0) { 4052 goto out; 4053 } 4054 4055 switch (type) { 4056 case QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN: 4057 case QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC: 4058 if (bs->backing && bs->backing->bs) { 4059 int64_t backing_length = bdrv_getlength(bs->backing->bs); 4060 if (src_offset >= backing_length) { 4061 cur_write_flags |= BDRV_REQ_ZERO_WRITE; 4062 } else { 4063 child = bs->backing; 4064 cur_bytes = MIN(cur_bytes, backing_length - src_offset); 4065 copy_offset = src_offset; 4066 } 4067 } else { 4068 cur_write_flags |= BDRV_REQ_ZERO_WRITE; 4069 } 4070 break; 4071 4072 case QCOW2_SUBCLUSTER_ZERO_PLAIN: 4073 case QCOW2_SUBCLUSTER_ZERO_ALLOC: 4074 cur_write_flags |= BDRV_REQ_ZERO_WRITE; 4075 break; 4076 4077 case QCOW2_SUBCLUSTER_COMPRESSED: 4078 ret = -ENOTSUP; 4079 goto out; 4080 4081 case QCOW2_SUBCLUSTER_NORMAL: 4082 child = s->data_file; 4083 break; 4084 4085 default: 4086 abort(); 4087 } 4088 qemu_co_mutex_unlock(&s->lock); 4089 ret = bdrv_co_copy_range_from(child, 4090 copy_offset, 4091 dst, dst_offset, 4092 cur_bytes, read_flags, cur_write_flags); 4093 qemu_co_mutex_lock(&s->lock); 4094 if (ret < 0) { 4095 goto out; 4096 } 4097 4098 bytes -= cur_bytes; 4099 src_offset += cur_bytes; 4100 dst_offset += cur_bytes; 4101 } 4102 ret = 0; 4103 4104 out: 4105 qemu_co_mutex_unlock(&s->lock); 4106 return ret; 4107 } 4108 4109 static int coroutine_fn 4110 qcow2_co_copy_range_to(BlockDriverState *bs, 4111 BdrvChild *src, uint64_t src_offset, 4112 BdrvChild *dst, uint64_t dst_offset, 4113 uint64_t bytes, BdrvRequestFlags read_flags, 4114 BdrvRequestFlags write_flags) 4115 { 4116 BDRVQcow2State *s = bs->opaque; 4117 int ret; 4118 unsigned int cur_bytes; /* number of sectors in current iteration */ 4119 uint64_t host_offset; 4120 QCowL2Meta *l2meta = NULL; 4121 4122 assert(!bs->encrypted); 4123 4124 qemu_co_mutex_lock(&s->lock); 4125 4126 while (bytes != 0) { 4127 4128 l2meta = NULL; 4129 4130 cur_bytes = MIN(bytes, INT_MAX); 4131 4132 /* TODO: 4133 * If src->bs == dst->bs, we could simply copy by incrementing 4134 * the refcnt, without copying user data. 4135 * Or if src->bs == dst->bs->backing->bs, we could copy by discarding. */ 4136 ret = qcow2_alloc_host_offset(bs, dst_offset, &cur_bytes, 4137 &host_offset, &l2meta); 4138 if (ret < 0) { 4139 goto fail; 4140 } 4141 4142 ret = qcow2_pre_write_overlap_check(bs, 0, host_offset, cur_bytes, 4143 true); 4144 if (ret < 0) { 4145 goto fail; 4146 } 4147 4148 qemu_co_mutex_unlock(&s->lock); 4149 ret = bdrv_co_copy_range_to(src, src_offset, s->data_file, host_offset, 4150 cur_bytes, read_flags, write_flags); 4151 qemu_co_mutex_lock(&s->lock); 4152 if (ret < 0) { 4153 goto fail; 4154 } 4155 4156 ret = qcow2_handle_l2meta(bs, &l2meta, true); 4157 if (ret) { 4158 goto fail; 4159 } 4160 4161 bytes -= cur_bytes; 4162 src_offset += cur_bytes; 4163 dst_offset += cur_bytes; 4164 } 4165 ret = 0; 4166 4167 fail: 4168 qcow2_handle_l2meta(bs, &l2meta, false); 4169 4170 qemu_co_mutex_unlock(&s->lock); 4171 4172 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret); 4173 4174 return ret; 4175 } 4176 4177 static int coroutine_fn qcow2_co_truncate(BlockDriverState *bs, int64_t offset, 4178 bool exact, PreallocMode prealloc, 4179 BdrvRequestFlags flags, Error **errp) 4180 { 4181 BDRVQcow2State *s = bs->opaque; 4182 uint64_t old_length; 4183 int64_t new_l1_size; 4184 int ret; 4185 QDict *options; 4186 4187 if (prealloc != PREALLOC_MODE_OFF && prealloc != PREALLOC_MODE_METADATA && 4188 prealloc != PREALLOC_MODE_FALLOC && prealloc != PREALLOC_MODE_FULL) 4189 { 4190 error_setg(errp, "Unsupported preallocation mode '%s'", 4191 PreallocMode_str(prealloc)); 4192 return -ENOTSUP; 4193 } 4194 4195 if (!QEMU_IS_ALIGNED(offset, BDRV_SECTOR_SIZE)) { 4196 error_setg(errp, "The new size must be a multiple of %u", 4197 (unsigned) BDRV_SECTOR_SIZE); 4198 return -EINVAL; 4199 } 4200 4201 qemu_co_mutex_lock(&s->lock); 4202 4203 /* 4204 * Even though we store snapshot size for all images, it was not 4205 * required until v3, so it is not safe to proceed for v2. 4206 */ 4207 if (s->nb_snapshots && s->qcow_version < 3) { 4208 error_setg(errp, "Can't resize a v2 image which has snapshots"); 4209 ret = -ENOTSUP; 4210 goto fail; 4211 } 4212 4213 /* See qcow2-bitmap.c for which bitmap scenarios prevent a resize. */ 4214 if (qcow2_truncate_bitmaps_check(bs, errp)) { 4215 ret = -ENOTSUP; 4216 goto fail; 4217 } 4218 4219 old_length = bs->total_sectors * BDRV_SECTOR_SIZE; 4220 new_l1_size = size_to_l1(s, offset); 4221 4222 if (offset < old_length) { 4223 int64_t last_cluster, old_file_size; 4224 if (prealloc != PREALLOC_MODE_OFF) { 4225 error_setg(errp, 4226 "Preallocation can't be used for shrinking an image"); 4227 ret = -EINVAL; 4228 goto fail; 4229 } 4230 4231 ret = qcow2_cluster_discard(bs, ROUND_UP(offset, s->cluster_size), 4232 old_length - ROUND_UP(offset, 4233 s->cluster_size), 4234 QCOW2_DISCARD_ALWAYS, true); 4235 if (ret < 0) { 4236 error_setg_errno(errp, -ret, "Failed to discard cropped clusters"); 4237 goto fail; 4238 } 4239 4240 ret = qcow2_shrink_l1_table(bs, new_l1_size); 4241 if (ret < 0) { 4242 error_setg_errno(errp, -ret, 4243 "Failed to reduce the number of L2 tables"); 4244 goto fail; 4245 } 4246 4247 ret = qcow2_shrink_reftable(bs); 4248 if (ret < 0) { 4249 error_setg_errno(errp, -ret, 4250 "Failed to discard unused refblocks"); 4251 goto fail; 4252 } 4253 4254 old_file_size = bdrv_getlength(bs->file->bs); 4255 if (old_file_size < 0) { 4256 error_setg_errno(errp, -old_file_size, 4257 "Failed to inquire current file length"); 4258 ret = old_file_size; 4259 goto fail; 4260 } 4261 last_cluster = qcow2_get_last_cluster(bs, old_file_size); 4262 if (last_cluster < 0) { 4263 error_setg_errno(errp, -last_cluster, 4264 "Failed to find the last cluster"); 4265 ret = last_cluster; 4266 goto fail; 4267 } 4268 if ((last_cluster + 1) * s->cluster_size < old_file_size) { 4269 Error *local_err = NULL; 4270 4271 /* 4272 * Do not pass @exact here: It will not help the user if 4273 * we get an error here just because they wanted to shrink 4274 * their qcow2 image (on a block device) with qemu-img. 4275 * (And on the qcow2 layer, the @exact requirement is 4276 * always fulfilled, so there is no need to pass it on.) 4277 */ 4278 bdrv_co_truncate(bs->file, (last_cluster + 1) * s->cluster_size, 4279 false, PREALLOC_MODE_OFF, 0, &local_err); 4280 if (local_err) { 4281 warn_reportf_err(local_err, 4282 "Failed to truncate the tail of the image: "); 4283 } 4284 } 4285 } else { 4286 ret = qcow2_grow_l1_table(bs, new_l1_size, true); 4287 if (ret < 0) { 4288 error_setg_errno(errp, -ret, "Failed to grow the L1 table"); 4289 goto fail; 4290 } 4291 4292 if (data_file_is_raw(bs) && prealloc == PREALLOC_MODE_OFF) { 4293 /* 4294 * When creating a qcow2 image with data-file-raw, we enforce 4295 * at least prealloc=metadata, so that the L1/L2 tables are 4296 * fully allocated and reading from the data file will return 4297 * the same data as reading from the qcow2 image. When the 4298 * image is grown, we must consequently preallocate the 4299 * metadata structures to cover the added area. 4300 */ 4301 prealloc = PREALLOC_MODE_METADATA; 4302 } 4303 } 4304 4305 switch (prealloc) { 4306 case PREALLOC_MODE_OFF: 4307 if (has_data_file(bs)) { 4308 /* 4309 * If the caller wants an exact resize, the external data 4310 * file should be resized to the exact target size, too, 4311 * so we pass @exact here. 4312 */ 4313 ret = bdrv_co_truncate(s->data_file, offset, exact, prealloc, 0, 4314 errp); 4315 if (ret < 0) { 4316 goto fail; 4317 } 4318 } 4319 break; 4320 4321 case PREALLOC_MODE_METADATA: 4322 ret = preallocate_co(bs, old_length, offset, prealloc, errp); 4323 if (ret < 0) { 4324 goto fail; 4325 } 4326 break; 4327 4328 case PREALLOC_MODE_FALLOC: 4329 case PREALLOC_MODE_FULL: 4330 { 4331 int64_t allocation_start, host_offset, guest_offset; 4332 int64_t clusters_allocated; 4333 int64_t old_file_size, last_cluster, new_file_size; 4334 uint64_t nb_new_data_clusters, nb_new_l2_tables; 4335 bool subclusters_need_allocation = false; 4336 4337 /* With a data file, preallocation means just allocating the metadata 4338 * and forwarding the truncate request to the data file */ 4339 if (has_data_file(bs)) { 4340 ret = preallocate_co(bs, old_length, offset, prealloc, errp); 4341 if (ret < 0) { 4342 goto fail; 4343 } 4344 break; 4345 } 4346 4347 old_file_size = bdrv_getlength(bs->file->bs); 4348 if (old_file_size < 0) { 4349 error_setg_errno(errp, -old_file_size, 4350 "Failed to inquire current file length"); 4351 ret = old_file_size; 4352 goto fail; 4353 } 4354 4355 last_cluster = qcow2_get_last_cluster(bs, old_file_size); 4356 if (last_cluster >= 0) { 4357 old_file_size = (last_cluster + 1) * s->cluster_size; 4358 } else { 4359 old_file_size = ROUND_UP(old_file_size, s->cluster_size); 4360 } 4361 4362 nb_new_data_clusters = (ROUND_UP(offset, s->cluster_size) - 4363 start_of_cluster(s, old_length)) >> s->cluster_bits; 4364 4365 /* This is an overestimation; we will not actually allocate space for 4366 * these in the file but just make sure the new refcount structures are 4367 * able to cover them so we will not have to allocate new refblocks 4368 * while entering the data blocks in the potentially new L2 tables. 4369 * (We do not actually care where the L2 tables are placed. Maybe they 4370 * are already allocated or they can be placed somewhere before 4371 * @old_file_size. It does not matter because they will be fully 4372 * allocated automatically, so they do not need to be covered by the 4373 * preallocation. All that matters is that we will not have to allocate 4374 * new refcount structures for them.) */ 4375 nb_new_l2_tables = DIV_ROUND_UP(nb_new_data_clusters, 4376 s->cluster_size / l2_entry_size(s)); 4377 /* The cluster range may not be aligned to L2 boundaries, so add one L2 4378 * table for a potential head/tail */ 4379 nb_new_l2_tables++; 4380 4381 allocation_start = qcow2_refcount_area(bs, old_file_size, 4382 nb_new_data_clusters + 4383 nb_new_l2_tables, 4384 true, 0, 0); 4385 if (allocation_start < 0) { 4386 error_setg_errno(errp, -allocation_start, 4387 "Failed to resize refcount structures"); 4388 ret = allocation_start; 4389 goto fail; 4390 } 4391 4392 clusters_allocated = qcow2_alloc_clusters_at(bs, allocation_start, 4393 nb_new_data_clusters); 4394 if (clusters_allocated < 0) { 4395 error_setg_errno(errp, -clusters_allocated, 4396 "Failed to allocate data clusters"); 4397 ret = clusters_allocated; 4398 goto fail; 4399 } 4400 4401 assert(clusters_allocated == nb_new_data_clusters); 4402 4403 /* Allocate the data area */ 4404 new_file_size = allocation_start + 4405 nb_new_data_clusters * s->cluster_size; 4406 /* 4407 * Image file grows, so @exact does not matter. 4408 * 4409 * If we need to zero out the new area, try first whether the protocol 4410 * driver can already take care of this. 4411 */ 4412 if (flags & BDRV_REQ_ZERO_WRITE) { 4413 ret = bdrv_co_truncate(bs->file, new_file_size, false, prealloc, 4414 BDRV_REQ_ZERO_WRITE, NULL); 4415 if (ret >= 0) { 4416 flags &= ~BDRV_REQ_ZERO_WRITE; 4417 /* Ensure that we read zeroes and not backing file data */ 4418 subclusters_need_allocation = true; 4419 } 4420 } else { 4421 ret = -1; 4422 } 4423 if (ret < 0) { 4424 ret = bdrv_co_truncate(bs->file, new_file_size, false, prealloc, 0, 4425 errp); 4426 } 4427 if (ret < 0) { 4428 error_prepend(errp, "Failed to resize underlying file: "); 4429 qcow2_free_clusters(bs, allocation_start, 4430 nb_new_data_clusters * s->cluster_size, 4431 QCOW2_DISCARD_OTHER); 4432 goto fail; 4433 } 4434 4435 /* Create the necessary L2 entries */ 4436 host_offset = allocation_start; 4437 guest_offset = old_length; 4438 while (nb_new_data_clusters) { 4439 int64_t nb_clusters = MIN( 4440 nb_new_data_clusters, 4441 s->l2_slice_size - offset_to_l2_slice_index(s, guest_offset)); 4442 unsigned cow_start_length = offset_into_cluster(s, guest_offset); 4443 QCowL2Meta allocation; 4444 guest_offset = start_of_cluster(s, guest_offset); 4445 allocation = (QCowL2Meta) { 4446 .offset = guest_offset, 4447 .alloc_offset = host_offset, 4448 .nb_clusters = nb_clusters, 4449 .cow_start = { 4450 .offset = 0, 4451 .nb_bytes = cow_start_length, 4452 }, 4453 .cow_end = { 4454 .offset = nb_clusters << s->cluster_bits, 4455 .nb_bytes = 0, 4456 }, 4457 .prealloc = !subclusters_need_allocation, 4458 }; 4459 qemu_co_queue_init(&allocation.dependent_requests); 4460 4461 ret = qcow2_alloc_cluster_link_l2(bs, &allocation); 4462 if (ret < 0) { 4463 error_setg_errno(errp, -ret, "Failed to update L2 tables"); 4464 qcow2_free_clusters(bs, host_offset, 4465 nb_new_data_clusters * s->cluster_size, 4466 QCOW2_DISCARD_OTHER); 4467 goto fail; 4468 } 4469 4470 guest_offset += nb_clusters * s->cluster_size; 4471 host_offset += nb_clusters * s->cluster_size; 4472 nb_new_data_clusters -= nb_clusters; 4473 } 4474 break; 4475 } 4476 4477 default: 4478 g_assert_not_reached(); 4479 } 4480 4481 if ((flags & BDRV_REQ_ZERO_WRITE) && offset > old_length) { 4482 uint64_t zero_start = QEMU_ALIGN_UP(old_length, s->subcluster_size); 4483 4484 /* 4485 * Use zero clusters as much as we can. qcow2_subcluster_zeroize() 4486 * requires a subcluster-aligned start. The end may be unaligned if 4487 * it is at the end of the image (which it is here). 4488 */ 4489 if (offset > zero_start) { 4490 ret = qcow2_subcluster_zeroize(bs, zero_start, offset - zero_start, 4491 0); 4492 if (ret < 0) { 4493 error_setg_errno(errp, -ret, "Failed to zero out new clusters"); 4494 goto fail; 4495 } 4496 } 4497 4498 /* Write explicit zeros for the unaligned head */ 4499 if (zero_start > old_length) { 4500 uint64_t len = MIN(zero_start, offset) - old_length; 4501 uint8_t *buf = qemu_blockalign0(bs, len); 4502 QEMUIOVector qiov; 4503 qemu_iovec_init_buf(&qiov, buf, len); 4504 4505 qemu_co_mutex_unlock(&s->lock); 4506 ret = qcow2_co_pwritev_part(bs, old_length, len, &qiov, 0, 0); 4507 qemu_co_mutex_lock(&s->lock); 4508 4509 qemu_vfree(buf); 4510 if (ret < 0) { 4511 error_setg_errno(errp, -ret, "Failed to zero out the new area"); 4512 goto fail; 4513 } 4514 } 4515 } 4516 4517 if (prealloc != PREALLOC_MODE_OFF) { 4518 /* Flush metadata before actually changing the image size */ 4519 ret = qcow2_write_caches(bs); 4520 if (ret < 0) { 4521 error_setg_errno(errp, -ret, 4522 "Failed to flush the preallocated area to disk"); 4523 goto fail; 4524 } 4525 } 4526 4527 bs->total_sectors = offset / BDRV_SECTOR_SIZE; 4528 4529 /* write updated header.size */ 4530 offset = cpu_to_be64(offset); 4531 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, size), 4532 &offset, sizeof(offset)); 4533 if (ret < 0) { 4534 error_setg_errno(errp, -ret, "Failed to update the image size"); 4535 goto fail; 4536 } 4537 4538 s->l1_vm_state_index = new_l1_size; 4539 4540 /* Update cache sizes */ 4541 options = qdict_clone_shallow(bs->options); 4542 ret = qcow2_update_options(bs, options, s->flags, errp); 4543 qobject_unref(options); 4544 if (ret < 0) { 4545 goto fail; 4546 } 4547 ret = 0; 4548 fail: 4549 qemu_co_mutex_unlock(&s->lock); 4550 return ret; 4551 } 4552 4553 static coroutine_fn int 4554 qcow2_co_pwritev_compressed_task(BlockDriverState *bs, 4555 uint64_t offset, uint64_t bytes, 4556 QEMUIOVector *qiov, size_t qiov_offset) 4557 { 4558 BDRVQcow2State *s = bs->opaque; 4559 int ret; 4560 ssize_t out_len; 4561 uint8_t *buf, *out_buf; 4562 uint64_t cluster_offset; 4563 4564 assert(bytes == s->cluster_size || (bytes < s->cluster_size && 4565 (offset + bytes == bs->total_sectors << BDRV_SECTOR_BITS))); 4566 4567 buf = qemu_blockalign(bs, s->cluster_size); 4568 if (bytes < s->cluster_size) { 4569 /* Zero-pad last write if image size is not cluster aligned */ 4570 memset(buf + bytes, 0, s->cluster_size - bytes); 4571 } 4572 qemu_iovec_to_buf(qiov, qiov_offset, buf, bytes); 4573 4574 out_buf = g_malloc(s->cluster_size); 4575 4576 out_len = qcow2_co_compress(bs, out_buf, s->cluster_size - 1, 4577 buf, s->cluster_size); 4578 if (out_len == -ENOMEM) { 4579 /* could not compress: write normal cluster */ 4580 ret = qcow2_co_pwritev_part(bs, offset, bytes, qiov, qiov_offset, 0); 4581 if (ret < 0) { 4582 goto fail; 4583 } 4584 goto success; 4585 } else if (out_len < 0) { 4586 ret = -EINVAL; 4587 goto fail; 4588 } 4589 4590 qemu_co_mutex_lock(&s->lock); 4591 ret = qcow2_alloc_compressed_cluster_offset(bs, offset, out_len, 4592 &cluster_offset); 4593 if (ret < 0) { 4594 qemu_co_mutex_unlock(&s->lock); 4595 goto fail; 4596 } 4597 4598 ret = qcow2_pre_write_overlap_check(bs, 0, cluster_offset, out_len, true); 4599 qemu_co_mutex_unlock(&s->lock); 4600 if (ret < 0) { 4601 goto fail; 4602 } 4603 4604 BLKDBG_EVENT(s->data_file, BLKDBG_WRITE_COMPRESSED); 4605 ret = bdrv_co_pwrite(s->data_file, cluster_offset, out_len, out_buf, 0); 4606 if (ret < 0) { 4607 goto fail; 4608 } 4609 success: 4610 ret = 0; 4611 fail: 4612 qemu_vfree(buf); 4613 g_free(out_buf); 4614 return ret; 4615 } 4616 4617 static coroutine_fn int qcow2_co_pwritev_compressed_task_entry(AioTask *task) 4618 { 4619 Qcow2AioTask *t = container_of(task, Qcow2AioTask, task); 4620 4621 assert(!t->subcluster_type && !t->l2meta); 4622 4623 return qcow2_co_pwritev_compressed_task(t->bs, t->offset, t->bytes, t->qiov, 4624 t->qiov_offset); 4625 } 4626 4627 /* 4628 * XXX: put compressed sectors first, then all the cluster aligned 4629 * tables to avoid losing bytes in alignment 4630 */ 4631 static coroutine_fn int 4632 qcow2_co_pwritev_compressed_part(BlockDriverState *bs, 4633 uint64_t offset, uint64_t bytes, 4634 QEMUIOVector *qiov, size_t qiov_offset) 4635 { 4636 BDRVQcow2State *s = bs->opaque; 4637 AioTaskPool *aio = NULL; 4638 int ret = 0; 4639 4640 if (has_data_file(bs)) { 4641 return -ENOTSUP; 4642 } 4643 4644 if (bytes == 0) { 4645 /* 4646 * align end of file to a sector boundary to ease reading with 4647 * sector based I/Os 4648 */ 4649 int64_t len = bdrv_getlength(bs->file->bs); 4650 if (len < 0) { 4651 return len; 4652 } 4653 return bdrv_co_truncate(bs->file, len, false, PREALLOC_MODE_OFF, 0, 4654 NULL); 4655 } 4656 4657 if (offset_into_cluster(s, offset)) { 4658 return -EINVAL; 4659 } 4660 4661 if (offset_into_cluster(s, bytes) && 4662 (offset + bytes) != (bs->total_sectors << BDRV_SECTOR_BITS)) { 4663 return -EINVAL; 4664 } 4665 4666 while (bytes && aio_task_pool_status(aio) == 0) { 4667 uint64_t chunk_size = MIN(bytes, s->cluster_size); 4668 4669 if (!aio && chunk_size != bytes) { 4670 aio = aio_task_pool_new(QCOW2_MAX_WORKERS); 4671 } 4672 4673 ret = qcow2_add_task(bs, aio, qcow2_co_pwritev_compressed_task_entry, 4674 0, 0, offset, chunk_size, qiov, qiov_offset, NULL); 4675 if (ret < 0) { 4676 break; 4677 } 4678 qiov_offset += chunk_size; 4679 offset += chunk_size; 4680 bytes -= chunk_size; 4681 } 4682 4683 if (aio) { 4684 aio_task_pool_wait_all(aio); 4685 if (ret == 0) { 4686 ret = aio_task_pool_status(aio); 4687 } 4688 g_free(aio); 4689 } 4690 4691 return ret; 4692 } 4693 4694 static int coroutine_fn 4695 qcow2_co_preadv_compressed(BlockDriverState *bs, 4696 uint64_t cluster_descriptor, 4697 uint64_t offset, 4698 uint64_t bytes, 4699 QEMUIOVector *qiov, 4700 size_t qiov_offset) 4701 { 4702 BDRVQcow2State *s = bs->opaque; 4703 int ret = 0, csize, nb_csectors; 4704 uint64_t coffset; 4705 uint8_t *buf, *out_buf; 4706 int offset_in_cluster = offset_into_cluster(s, offset); 4707 4708 coffset = cluster_descriptor & s->cluster_offset_mask; 4709 nb_csectors = ((cluster_descriptor >> s->csize_shift) & s->csize_mask) + 1; 4710 csize = nb_csectors * QCOW2_COMPRESSED_SECTOR_SIZE - 4711 (coffset & ~QCOW2_COMPRESSED_SECTOR_MASK); 4712 4713 buf = g_try_malloc(csize); 4714 if (!buf) { 4715 return -ENOMEM; 4716 } 4717 4718 out_buf = qemu_blockalign(bs, s->cluster_size); 4719 4720 BLKDBG_EVENT(bs->file, BLKDBG_READ_COMPRESSED); 4721 ret = bdrv_co_pread(bs->file, coffset, csize, buf, 0); 4722 if (ret < 0) { 4723 goto fail; 4724 } 4725 4726 if (qcow2_co_decompress(bs, out_buf, s->cluster_size, buf, csize) < 0) { 4727 ret = -EIO; 4728 goto fail; 4729 } 4730 4731 qemu_iovec_from_buf(qiov, qiov_offset, out_buf + offset_in_cluster, bytes); 4732 4733 fail: 4734 qemu_vfree(out_buf); 4735 g_free(buf); 4736 4737 return ret; 4738 } 4739 4740 static int make_completely_empty(BlockDriverState *bs) 4741 { 4742 BDRVQcow2State *s = bs->opaque; 4743 Error *local_err = NULL; 4744 int ret, l1_clusters; 4745 int64_t offset; 4746 uint64_t *new_reftable = NULL; 4747 uint64_t rt_entry, l1_size2; 4748 struct { 4749 uint64_t l1_offset; 4750 uint64_t reftable_offset; 4751 uint32_t reftable_clusters; 4752 } QEMU_PACKED l1_ofs_rt_ofs_cls; 4753 4754 ret = qcow2_cache_empty(bs, s->l2_table_cache); 4755 if (ret < 0) { 4756 goto fail; 4757 } 4758 4759 ret = qcow2_cache_empty(bs, s->refcount_block_cache); 4760 if (ret < 0) { 4761 goto fail; 4762 } 4763 4764 /* Refcounts will be broken utterly */ 4765 ret = qcow2_mark_dirty(bs); 4766 if (ret < 0) { 4767 goto fail; 4768 } 4769 4770 BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE); 4771 4772 l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / L1E_SIZE); 4773 l1_size2 = (uint64_t)s->l1_size * L1E_SIZE; 4774 4775 /* After this call, neither the in-memory nor the on-disk refcount 4776 * information accurately describe the actual references */ 4777 4778 ret = bdrv_pwrite_zeroes(bs->file, s->l1_table_offset, 4779 l1_clusters * s->cluster_size, 0); 4780 if (ret < 0) { 4781 goto fail_broken_refcounts; 4782 } 4783 memset(s->l1_table, 0, l1_size2); 4784 4785 BLKDBG_EVENT(bs->file, BLKDBG_EMPTY_IMAGE_PREPARE); 4786 4787 /* Overwrite enough clusters at the beginning of the sectors to place 4788 * the refcount table, a refcount block and the L1 table in; this may 4789 * overwrite parts of the existing refcount and L1 table, which is not 4790 * an issue because the dirty flag is set, complete data loss is in fact 4791 * desired and partial data loss is consequently fine as well */ 4792 ret = bdrv_pwrite_zeroes(bs->file, s->cluster_size, 4793 (2 + l1_clusters) * s->cluster_size, 0); 4794 /* This call (even if it failed overall) may have overwritten on-disk 4795 * refcount structures; in that case, the in-memory refcount information 4796 * will probably differ from the on-disk information which makes the BDS 4797 * unusable */ 4798 if (ret < 0) { 4799 goto fail_broken_refcounts; 4800 } 4801 4802 BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE); 4803 BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_UPDATE); 4804 4805 /* "Create" an empty reftable (one cluster) directly after the image 4806 * header and an empty L1 table three clusters after the image header; 4807 * the cluster between those two will be used as the first refblock */ 4808 l1_ofs_rt_ofs_cls.l1_offset = cpu_to_be64(3 * s->cluster_size); 4809 l1_ofs_rt_ofs_cls.reftable_offset = cpu_to_be64(s->cluster_size); 4810 l1_ofs_rt_ofs_cls.reftable_clusters = cpu_to_be32(1); 4811 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, l1_table_offset), 4812 &l1_ofs_rt_ofs_cls, sizeof(l1_ofs_rt_ofs_cls)); 4813 if (ret < 0) { 4814 goto fail_broken_refcounts; 4815 } 4816 4817 s->l1_table_offset = 3 * s->cluster_size; 4818 4819 new_reftable = g_try_new0(uint64_t, s->cluster_size / REFTABLE_ENTRY_SIZE); 4820 if (!new_reftable) { 4821 ret = -ENOMEM; 4822 goto fail_broken_refcounts; 4823 } 4824 4825 s->refcount_table_offset = s->cluster_size; 4826 s->refcount_table_size = s->cluster_size / REFTABLE_ENTRY_SIZE; 4827 s->max_refcount_table_index = 0; 4828 4829 g_free(s->refcount_table); 4830 s->refcount_table = new_reftable; 4831 new_reftable = NULL; 4832 4833 /* Now the in-memory refcount information again corresponds to the on-disk 4834 * information (reftable is empty and no refblocks (the refblock cache is 4835 * empty)); however, this means some clusters (e.g. the image header) are 4836 * referenced, but not refcounted, but the normal qcow2 code assumes that 4837 * the in-memory information is always correct */ 4838 4839 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC); 4840 4841 /* Enter the first refblock into the reftable */ 4842 rt_entry = cpu_to_be64(2 * s->cluster_size); 4843 ret = bdrv_pwrite_sync(bs->file, s->cluster_size, 4844 &rt_entry, sizeof(rt_entry)); 4845 if (ret < 0) { 4846 goto fail_broken_refcounts; 4847 } 4848 s->refcount_table[0] = 2 * s->cluster_size; 4849 4850 s->free_cluster_index = 0; 4851 assert(3 + l1_clusters <= s->refcount_block_size); 4852 offset = qcow2_alloc_clusters(bs, 3 * s->cluster_size + l1_size2); 4853 if (offset < 0) { 4854 ret = offset; 4855 goto fail_broken_refcounts; 4856 } else if (offset > 0) { 4857 error_report("First cluster in emptied image is in use"); 4858 abort(); 4859 } 4860 4861 /* Now finally the in-memory information corresponds to the on-disk 4862 * structures and is correct */ 4863 ret = qcow2_mark_clean(bs); 4864 if (ret < 0) { 4865 goto fail; 4866 } 4867 4868 ret = bdrv_truncate(bs->file, (3 + l1_clusters) * s->cluster_size, false, 4869 PREALLOC_MODE_OFF, 0, &local_err); 4870 if (ret < 0) { 4871 error_report_err(local_err); 4872 goto fail; 4873 } 4874 4875 return 0; 4876 4877 fail_broken_refcounts: 4878 /* The BDS is unusable at this point. If we wanted to make it usable, we 4879 * would have to call qcow2_refcount_close(), qcow2_refcount_init(), 4880 * qcow2_check_refcounts(), qcow2_refcount_close() and qcow2_refcount_init() 4881 * again. However, because the functions which could have caused this error 4882 * path to be taken are used by those functions as well, it's very likely 4883 * that that sequence will fail as well. Therefore, just eject the BDS. */ 4884 bs->drv = NULL; 4885 4886 fail: 4887 g_free(new_reftable); 4888 return ret; 4889 } 4890 4891 static int qcow2_make_empty(BlockDriverState *bs) 4892 { 4893 BDRVQcow2State *s = bs->opaque; 4894 uint64_t offset, end_offset; 4895 int step = QEMU_ALIGN_DOWN(INT_MAX, s->cluster_size); 4896 int l1_clusters, ret = 0; 4897 4898 l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / L1E_SIZE); 4899 4900 if (s->qcow_version >= 3 && !s->snapshots && !s->nb_bitmaps && 4901 3 + l1_clusters <= s->refcount_block_size && 4902 s->crypt_method_header != QCOW_CRYPT_LUKS && 4903 !has_data_file(bs)) { 4904 /* The following function only works for qcow2 v3 images (it 4905 * requires the dirty flag) and only as long as there are no 4906 * features that reserve extra clusters (such as snapshots, 4907 * LUKS header, or persistent bitmaps), because it completely 4908 * empties the image. Furthermore, the L1 table and three 4909 * additional clusters (image header, refcount table, one 4910 * refcount block) have to fit inside one refcount block. It 4911 * only resets the image file, i.e. does not work with an 4912 * external data file. */ 4913 return make_completely_empty(bs); 4914 } 4915 4916 /* This fallback code simply discards every active cluster; this is slow, 4917 * but works in all cases */ 4918 end_offset = bs->total_sectors * BDRV_SECTOR_SIZE; 4919 for (offset = 0; offset < end_offset; offset += step) { 4920 /* As this function is generally used after committing an external 4921 * snapshot, QCOW2_DISCARD_SNAPSHOT seems appropriate. Also, the 4922 * default action for this kind of discard is to pass the discard, 4923 * which will ideally result in an actually smaller image file, as 4924 * is probably desired. */ 4925 ret = qcow2_cluster_discard(bs, offset, MIN(step, end_offset - offset), 4926 QCOW2_DISCARD_SNAPSHOT, true); 4927 if (ret < 0) { 4928 break; 4929 } 4930 } 4931 4932 return ret; 4933 } 4934 4935 static coroutine_fn int qcow2_co_flush_to_os(BlockDriverState *bs) 4936 { 4937 BDRVQcow2State *s = bs->opaque; 4938 int ret; 4939 4940 qemu_co_mutex_lock(&s->lock); 4941 ret = qcow2_write_caches(bs); 4942 qemu_co_mutex_unlock(&s->lock); 4943 4944 return ret; 4945 } 4946 4947 static BlockMeasureInfo *qcow2_measure(QemuOpts *opts, BlockDriverState *in_bs, 4948 Error **errp) 4949 { 4950 Error *local_err = NULL; 4951 BlockMeasureInfo *info; 4952 uint64_t required = 0; /* bytes that contribute to required size */ 4953 uint64_t virtual_size; /* disk size as seen by guest */ 4954 uint64_t refcount_bits; 4955 uint64_t l2_tables; 4956 uint64_t luks_payload_size = 0; 4957 size_t cluster_size; 4958 int version; 4959 char *optstr; 4960 PreallocMode prealloc; 4961 bool has_backing_file; 4962 bool has_luks; 4963 bool extended_l2; 4964 size_t l2e_size; 4965 4966 /* Parse image creation options */ 4967 extended_l2 = qemu_opt_get_bool_del(opts, BLOCK_OPT_EXTL2, false); 4968 4969 cluster_size = qcow2_opt_get_cluster_size_del(opts, extended_l2, 4970 &local_err); 4971 if (local_err) { 4972 goto err; 4973 } 4974 4975 version = qcow2_opt_get_version_del(opts, &local_err); 4976 if (local_err) { 4977 goto err; 4978 } 4979 4980 refcount_bits = qcow2_opt_get_refcount_bits_del(opts, version, &local_err); 4981 if (local_err) { 4982 goto err; 4983 } 4984 4985 optstr = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC); 4986 prealloc = qapi_enum_parse(&PreallocMode_lookup, optstr, 4987 PREALLOC_MODE_OFF, &local_err); 4988 g_free(optstr); 4989 if (local_err) { 4990 goto err; 4991 } 4992 4993 optstr = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE); 4994 has_backing_file = !!optstr; 4995 g_free(optstr); 4996 4997 optstr = qemu_opt_get_del(opts, BLOCK_OPT_ENCRYPT_FORMAT); 4998 has_luks = optstr && strcmp(optstr, "luks") == 0; 4999 g_free(optstr); 5000 5001 if (has_luks) { 5002 g_autoptr(QCryptoBlockCreateOptions) create_opts = NULL; 5003 QDict *cryptoopts = qcow2_extract_crypto_opts(opts, "luks", errp); 5004 size_t headerlen; 5005 5006 create_opts = block_crypto_create_opts_init(cryptoopts, errp); 5007 qobject_unref(cryptoopts); 5008 if (!create_opts) { 5009 goto err; 5010 } 5011 5012 if (!qcrypto_block_calculate_payload_offset(create_opts, 5013 "encrypt.", 5014 &headerlen, 5015 &local_err)) { 5016 goto err; 5017 } 5018 5019 luks_payload_size = ROUND_UP(headerlen, cluster_size); 5020 } 5021 5022 virtual_size = qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0); 5023 virtual_size = ROUND_UP(virtual_size, cluster_size); 5024 5025 /* Check that virtual disk size is valid */ 5026 l2e_size = extended_l2 ? L2E_SIZE_EXTENDED : L2E_SIZE_NORMAL; 5027 l2_tables = DIV_ROUND_UP(virtual_size / cluster_size, 5028 cluster_size / l2e_size); 5029 if (l2_tables * L1E_SIZE > QCOW_MAX_L1_SIZE) { 5030 error_setg(&local_err, "The image size is too large " 5031 "(try using a larger cluster size)"); 5032 goto err; 5033 } 5034 5035 /* Account for input image */ 5036 if (in_bs) { 5037 int64_t ssize = bdrv_getlength(in_bs); 5038 if (ssize < 0) { 5039 error_setg_errno(&local_err, -ssize, 5040 "Unable to get image virtual_size"); 5041 goto err; 5042 } 5043 5044 virtual_size = ROUND_UP(ssize, cluster_size); 5045 5046 if (has_backing_file) { 5047 /* We don't how much of the backing chain is shared by the input 5048 * image and the new image file. In the worst case the new image's 5049 * backing file has nothing in common with the input image. Be 5050 * conservative and assume all clusters need to be written. 5051 */ 5052 required = virtual_size; 5053 } else { 5054 int64_t offset; 5055 int64_t pnum = 0; 5056 5057 for (offset = 0; offset < ssize; offset += pnum) { 5058 int ret; 5059 5060 ret = bdrv_block_status_above(in_bs, NULL, offset, 5061 ssize - offset, &pnum, NULL, 5062 NULL); 5063 if (ret < 0) { 5064 error_setg_errno(&local_err, -ret, 5065 "Unable to get block status"); 5066 goto err; 5067 } 5068 5069 if (ret & BDRV_BLOCK_ZERO) { 5070 /* Skip zero regions (safe with no backing file) */ 5071 } else if ((ret & (BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED)) == 5072 (BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED)) { 5073 /* Extend pnum to end of cluster for next iteration */ 5074 pnum = ROUND_UP(offset + pnum, cluster_size) - offset; 5075 5076 /* Count clusters we've seen */ 5077 required += offset % cluster_size + pnum; 5078 } 5079 } 5080 } 5081 } 5082 5083 /* Take into account preallocation. Nothing special is needed for 5084 * PREALLOC_MODE_METADATA since metadata is always counted. 5085 */ 5086 if (prealloc == PREALLOC_MODE_FULL || prealloc == PREALLOC_MODE_FALLOC) { 5087 required = virtual_size; 5088 } 5089 5090 info = g_new0(BlockMeasureInfo, 1); 5091 info->fully_allocated = luks_payload_size + 5092 qcow2_calc_prealloc_size(virtual_size, cluster_size, 5093 ctz32(refcount_bits), extended_l2); 5094 5095 /* 5096 * Remove data clusters that are not required. This overestimates the 5097 * required size because metadata needed for the fully allocated file is 5098 * still counted. Show bitmaps only if both source and destination 5099 * would support them. 5100 */ 5101 info->required = info->fully_allocated - virtual_size + required; 5102 info->has_bitmaps = version >= 3 && in_bs && 5103 bdrv_supports_persistent_dirty_bitmap(in_bs); 5104 if (info->has_bitmaps) { 5105 info->bitmaps = qcow2_get_persistent_dirty_bitmap_size(in_bs, 5106 cluster_size); 5107 } 5108 return info; 5109 5110 err: 5111 error_propagate(errp, local_err); 5112 return NULL; 5113 } 5114 5115 static int qcow2_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) 5116 { 5117 BDRVQcow2State *s = bs->opaque; 5118 bdi->cluster_size = s->cluster_size; 5119 bdi->vm_state_offset = qcow2_vm_state_offset(s); 5120 bdi->is_dirty = s->incompatible_features & QCOW2_INCOMPAT_DIRTY; 5121 return 0; 5122 } 5123 5124 static ImageInfoSpecific *qcow2_get_specific_info(BlockDriverState *bs, 5125 Error **errp) 5126 { 5127 BDRVQcow2State *s = bs->opaque; 5128 ImageInfoSpecific *spec_info; 5129 QCryptoBlockInfo *encrypt_info = NULL; 5130 5131 if (s->crypto != NULL) { 5132 encrypt_info = qcrypto_block_get_info(s->crypto, errp); 5133 if (!encrypt_info) { 5134 return NULL; 5135 } 5136 } 5137 5138 spec_info = g_new(ImageInfoSpecific, 1); 5139 *spec_info = (ImageInfoSpecific){ 5140 .type = IMAGE_INFO_SPECIFIC_KIND_QCOW2, 5141 .u.qcow2.data = g_new0(ImageInfoSpecificQCow2, 1), 5142 }; 5143 if (s->qcow_version == 2) { 5144 *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){ 5145 .compat = g_strdup("0.10"), 5146 .refcount_bits = s->refcount_bits, 5147 }; 5148 } else if (s->qcow_version == 3) { 5149 Qcow2BitmapInfoList *bitmaps; 5150 if (!qcow2_get_bitmap_info_list(bs, &bitmaps, errp)) { 5151 qapi_free_ImageInfoSpecific(spec_info); 5152 qapi_free_QCryptoBlockInfo(encrypt_info); 5153 return NULL; 5154 } 5155 *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){ 5156 .compat = g_strdup("1.1"), 5157 .lazy_refcounts = s->compatible_features & 5158 QCOW2_COMPAT_LAZY_REFCOUNTS, 5159 .has_lazy_refcounts = true, 5160 .corrupt = s->incompatible_features & 5161 QCOW2_INCOMPAT_CORRUPT, 5162 .has_corrupt = true, 5163 .has_extended_l2 = true, 5164 .extended_l2 = has_subclusters(s), 5165 .refcount_bits = s->refcount_bits, 5166 .has_bitmaps = !!bitmaps, 5167 .bitmaps = bitmaps, 5168 .has_data_file = !!s->image_data_file, 5169 .data_file = g_strdup(s->image_data_file), 5170 .has_data_file_raw = has_data_file(bs), 5171 .data_file_raw = data_file_is_raw(bs), 5172 .compression_type = s->compression_type, 5173 }; 5174 } else { 5175 /* if this assertion fails, this probably means a new version was 5176 * added without having it covered here */ 5177 assert(false); 5178 } 5179 5180 if (encrypt_info) { 5181 ImageInfoSpecificQCow2Encryption *qencrypt = 5182 g_new(ImageInfoSpecificQCow2Encryption, 1); 5183 switch (encrypt_info->format) { 5184 case Q_CRYPTO_BLOCK_FORMAT_QCOW: 5185 qencrypt->format = BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_AES; 5186 break; 5187 case Q_CRYPTO_BLOCK_FORMAT_LUKS: 5188 qencrypt->format = BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_LUKS; 5189 qencrypt->u.luks = encrypt_info->u.luks; 5190 break; 5191 default: 5192 abort(); 5193 } 5194 /* Since we did shallow copy above, erase any pointers 5195 * in the original info */ 5196 memset(&encrypt_info->u, 0, sizeof(encrypt_info->u)); 5197 qapi_free_QCryptoBlockInfo(encrypt_info); 5198 5199 spec_info->u.qcow2.data->has_encrypt = true; 5200 spec_info->u.qcow2.data->encrypt = qencrypt; 5201 } 5202 5203 return spec_info; 5204 } 5205 5206 static int qcow2_has_zero_init(BlockDriverState *bs) 5207 { 5208 BDRVQcow2State *s = bs->opaque; 5209 bool preallocated; 5210 5211 if (qemu_in_coroutine()) { 5212 qemu_co_mutex_lock(&s->lock); 5213 } 5214 /* 5215 * Check preallocation status: Preallocated images have all L2 5216 * tables allocated, nonpreallocated images have none. It is 5217 * therefore enough to check the first one. 5218 */ 5219 preallocated = s->l1_size > 0 && s->l1_table[0] != 0; 5220 if (qemu_in_coroutine()) { 5221 qemu_co_mutex_unlock(&s->lock); 5222 } 5223 5224 if (!preallocated) { 5225 return 1; 5226 } else if (bs->encrypted) { 5227 return 0; 5228 } else { 5229 return bdrv_has_zero_init(s->data_file->bs); 5230 } 5231 } 5232 5233 static int qcow2_save_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, 5234 int64_t pos) 5235 { 5236 BDRVQcow2State *s = bs->opaque; 5237 5238 BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_SAVE); 5239 return bs->drv->bdrv_co_pwritev_part(bs, qcow2_vm_state_offset(s) + pos, 5240 qiov->size, qiov, 0, 0); 5241 } 5242 5243 static int qcow2_load_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, 5244 int64_t pos) 5245 { 5246 BDRVQcow2State *s = bs->opaque; 5247 5248 BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_LOAD); 5249 return bs->drv->bdrv_co_preadv_part(bs, qcow2_vm_state_offset(s) + pos, 5250 qiov->size, qiov, 0, 0); 5251 } 5252 5253 /* 5254 * Downgrades an image's version. To achieve this, any incompatible features 5255 * have to be removed. 5256 */ 5257 static int qcow2_downgrade(BlockDriverState *bs, int target_version, 5258 BlockDriverAmendStatusCB *status_cb, void *cb_opaque, 5259 Error **errp) 5260 { 5261 BDRVQcow2State *s = bs->opaque; 5262 int current_version = s->qcow_version; 5263 int ret; 5264 int i; 5265 5266 /* This is qcow2_downgrade(), not qcow2_upgrade() */ 5267 assert(target_version < current_version); 5268 5269 /* There are no other versions (now) that you can downgrade to */ 5270 assert(target_version == 2); 5271 5272 if (s->refcount_order != 4) { 5273 error_setg(errp, "compat=0.10 requires refcount_bits=16"); 5274 return -ENOTSUP; 5275 } 5276 5277 if (has_data_file(bs)) { 5278 error_setg(errp, "Cannot downgrade an image with a data file"); 5279 return -ENOTSUP; 5280 } 5281 5282 /* 5283 * If any internal snapshot has a different size than the current 5284 * image size, or VM state size that exceeds 32 bits, downgrading 5285 * is unsafe. Even though we would still use v3-compliant output 5286 * to preserve that data, other v2 programs might not realize 5287 * those optional fields are important. 5288 */ 5289 for (i = 0; i < s->nb_snapshots; i++) { 5290 if (s->snapshots[i].vm_state_size > UINT32_MAX || 5291 s->snapshots[i].disk_size != bs->total_sectors * BDRV_SECTOR_SIZE) { 5292 error_setg(errp, "Internal snapshots prevent downgrade of image"); 5293 return -ENOTSUP; 5294 } 5295 } 5296 5297 /* clear incompatible features */ 5298 if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) { 5299 ret = qcow2_mark_clean(bs); 5300 if (ret < 0) { 5301 error_setg_errno(errp, -ret, "Failed to make the image clean"); 5302 return ret; 5303 } 5304 } 5305 5306 /* with QCOW2_INCOMPAT_CORRUPT, it is pretty much impossible to get here in 5307 * the first place; if that happens nonetheless, returning -ENOTSUP is the 5308 * best thing to do anyway */ 5309 5310 if (s->incompatible_features) { 5311 error_setg(errp, "Cannot downgrade an image with incompatible features " 5312 "%#" PRIx64 " set", s->incompatible_features); 5313 return -ENOTSUP; 5314 } 5315 5316 /* since we can ignore compatible features, we can set them to 0 as well */ 5317 s->compatible_features = 0; 5318 /* if lazy refcounts have been used, they have already been fixed through 5319 * clearing the dirty flag */ 5320 5321 /* clearing autoclear features is trivial */ 5322 s->autoclear_features = 0; 5323 5324 ret = qcow2_expand_zero_clusters(bs, status_cb, cb_opaque); 5325 if (ret < 0) { 5326 error_setg_errno(errp, -ret, "Failed to turn zero into data clusters"); 5327 return ret; 5328 } 5329 5330 s->qcow_version = target_version; 5331 ret = qcow2_update_header(bs); 5332 if (ret < 0) { 5333 s->qcow_version = current_version; 5334 error_setg_errno(errp, -ret, "Failed to update the image header"); 5335 return ret; 5336 } 5337 return 0; 5338 } 5339 5340 /* 5341 * Upgrades an image's version. While newer versions encompass all 5342 * features of older versions, some things may have to be presented 5343 * differently. 5344 */ 5345 static int qcow2_upgrade(BlockDriverState *bs, int target_version, 5346 BlockDriverAmendStatusCB *status_cb, void *cb_opaque, 5347 Error **errp) 5348 { 5349 BDRVQcow2State *s = bs->opaque; 5350 bool need_snapshot_update; 5351 int current_version = s->qcow_version; 5352 int i; 5353 int ret; 5354 5355 /* This is qcow2_upgrade(), not qcow2_downgrade() */ 5356 assert(target_version > current_version); 5357 5358 /* There are no other versions (yet) that you can upgrade to */ 5359 assert(target_version == 3); 5360 5361 status_cb(bs, 0, 2, cb_opaque); 5362 5363 /* 5364 * In v2, snapshots do not need to have extra data. v3 requires 5365 * the 64-bit VM state size and the virtual disk size to be 5366 * present. 5367 * qcow2_write_snapshots() will always write the list in the 5368 * v3-compliant format. 5369 */ 5370 need_snapshot_update = false; 5371 for (i = 0; i < s->nb_snapshots; i++) { 5372 if (s->snapshots[i].extra_data_size < 5373 sizeof_field(QCowSnapshotExtraData, vm_state_size_large) + 5374 sizeof_field(QCowSnapshotExtraData, disk_size)) 5375 { 5376 need_snapshot_update = true; 5377 break; 5378 } 5379 } 5380 if (need_snapshot_update) { 5381 ret = qcow2_write_snapshots(bs); 5382 if (ret < 0) { 5383 error_setg_errno(errp, -ret, "Failed to update the snapshot table"); 5384 return ret; 5385 } 5386 } 5387 status_cb(bs, 1, 2, cb_opaque); 5388 5389 s->qcow_version = target_version; 5390 ret = qcow2_update_header(bs); 5391 if (ret < 0) { 5392 s->qcow_version = current_version; 5393 error_setg_errno(errp, -ret, "Failed to update the image header"); 5394 return ret; 5395 } 5396 status_cb(bs, 2, 2, cb_opaque); 5397 5398 return 0; 5399 } 5400 5401 typedef enum Qcow2AmendOperation { 5402 /* This is the value Qcow2AmendHelperCBInfo::last_operation will be 5403 * statically initialized to so that the helper CB can discern the first 5404 * invocation from an operation change */ 5405 QCOW2_NO_OPERATION = 0, 5406 5407 QCOW2_UPGRADING, 5408 QCOW2_UPDATING_ENCRYPTION, 5409 QCOW2_CHANGING_REFCOUNT_ORDER, 5410 QCOW2_DOWNGRADING, 5411 } Qcow2AmendOperation; 5412 5413 typedef struct Qcow2AmendHelperCBInfo { 5414 /* The code coordinating the amend operations should only modify 5415 * these four fields; the rest will be managed by the CB */ 5416 BlockDriverAmendStatusCB *original_status_cb; 5417 void *original_cb_opaque; 5418 5419 Qcow2AmendOperation current_operation; 5420 5421 /* Total number of operations to perform (only set once) */ 5422 int total_operations; 5423 5424 /* The following fields are managed by the CB */ 5425 5426 /* Number of operations completed */ 5427 int operations_completed; 5428 5429 /* Cumulative offset of all completed operations */ 5430 int64_t offset_completed; 5431 5432 Qcow2AmendOperation last_operation; 5433 int64_t last_work_size; 5434 } Qcow2AmendHelperCBInfo; 5435 5436 static void qcow2_amend_helper_cb(BlockDriverState *bs, 5437 int64_t operation_offset, 5438 int64_t operation_work_size, void *opaque) 5439 { 5440 Qcow2AmendHelperCBInfo *info = opaque; 5441 int64_t current_work_size; 5442 int64_t projected_work_size; 5443 5444 if (info->current_operation != info->last_operation) { 5445 if (info->last_operation != QCOW2_NO_OPERATION) { 5446 info->offset_completed += info->last_work_size; 5447 info->operations_completed++; 5448 } 5449 5450 info->last_operation = info->current_operation; 5451 } 5452 5453 assert(info->total_operations > 0); 5454 assert(info->operations_completed < info->total_operations); 5455 5456 info->last_work_size = operation_work_size; 5457 5458 current_work_size = info->offset_completed + operation_work_size; 5459 5460 /* current_work_size is the total work size for (operations_completed + 1) 5461 * operations (which includes this one), so multiply it by the number of 5462 * operations not covered and divide it by the number of operations 5463 * covered to get a projection for the operations not covered */ 5464 projected_work_size = current_work_size * (info->total_operations - 5465 info->operations_completed - 1) 5466 / (info->operations_completed + 1); 5467 5468 info->original_status_cb(bs, info->offset_completed + operation_offset, 5469 current_work_size + projected_work_size, 5470 info->original_cb_opaque); 5471 } 5472 5473 static int qcow2_amend_options(BlockDriverState *bs, QemuOpts *opts, 5474 BlockDriverAmendStatusCB *status_cb, 5475 void *cb_opaque, 5476 bool force, 5477 Error **errp) 5478 { 5479 BDRVQcow2State *s = bs->opaque; 5480 int old_version = s->qcow_version, new_version = old_version; 5481 uint64_t new_size = 0; 5482 const char *backing_file = NULL, *backing_format = NULL, *data_file = NULL; 5483 bool lazy_refcounts = s->use_lazy_refcounts; 5484 bool data_file_raw = data_file_is_raw(bs); 5485 const char *compat = NULL; 5486 int refcount_bits = s->refcount_bits; 5487 int ret; 5488 QemuOptDesc *desc = opts->list->desc; 5489 Qcow2AmendHelperCBInfo helper_cb_info; 5490 bool encryption_update = false; 5491 5492 while (desc && desc->name) { 5493 if (!qemu_opt_find(opts, desc->name)) { 5494 /* only change explicitly defined options */ 5495 desc++; 5496 continue; 5497 } 5498 5499 if (!strcmp(desc->name, BLOCK_OPT_COMPAT_LEVEL)) { 5500 compat = qemu_opt_get(opts, BLOCK_OPT_COMPAT_LEVEL); 5501 if (!compat) { 5502 /* preserve default */ 5503 } else if (!strcmp(compat, "0.10") || !strcmp(compat, "v2")) { 5504 new_version = 2; 5505 } else if (!strcmp(compat, "1.1") || !strcmp(compat, "v3")) { 5506 new_version = 3; 5507 } else { 5508 error_setg(errp, "Unknown compatibility level %s", compat); 5509 return -EINVAL; 5510 } 5511 } else if (!strcmp(desc->name, BLOCK_OPT_SIZE)) { 5512 new_size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, 0); 5513 } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FILE)) { 5514 backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE); 5515 } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FMT)) { 5516 backing_format = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT); 5517 } else if (g_str_has_prefix(desc->name, "encrypt.")) { 5518 if (!s->crypto) { 5519 error_setg(errp, 5520 "Can't amend encryption options - encryption not present"); 5521 return -EINVAL; 5522 } 5523 if (s->crypt_method_header != QCOW_CRYPT_LUKS) { 5524 error_setg(errp, 5525 "Only LUKS encryption options can be amended"); 5526 return -ENOTSUP; 5527 } 5528 encryption_update = true; 5529 } else if (!strcmp(desc->name, BLOCK_OPT_LAZY_REFCOUNTS)) { 5530 lazy_refcounts = qemu_opt_get_bool(opts, BLOCK_OPT_LAZY_REFCOUNTS, 5531 lazy_refcounts); 5532 } else if (!strcmp(desc->name, BLOCK_OPT_REFCOUNT_BITS)) { 5533 refcount_bits = qemu_opt_get_number(opts, BLOCK_OPT_REFCOUNT_BITS, 5534 refcount_bits); 5535 5536 if (refcount_bits <= 0 || refcount_bits > 64 || 5537 !is_power_of_2(refcount_bits)) 5538 { 5539 error_setg(errp, "Refcount width must be a power of two and " 5540 "may not exceed 64 bits"); 5541 return -EINVAL; 5542 } 5543 } else if (!strcmp(desc->name, BLOCK_OPT_DATA_FILE)) { 5544 data_file = qemu_opt_get(opts, BLOCK_OPT_DATA_FILE); 5545 if (data_file && !has_data_file(bs)) { 5546 error_setg(errp, "data-file can only be set for images that " 5547 "use an external data file"); 5548 return -EINVAL; 5549 } 5550 } else if (!strcmp(desc->name, BLOCK_OPT_DATA_FILE_RAW)) { 5551 data_file_raw = qemu_opt_get_bool(opts, BLOCK_OPT_DATA_FILE_RAW, 5552 data_file_raw); 5553 if (data_file_raw && !data_file_is_raw(bs)) { 5554 error_setg(errp, "data-file-raw cannot be set on existing " 5555 "images"); 5556 return -EINVAL; 5557 } 5558 } else { 5559 /* if this point is reached, this probably means a new option was 5560 * added without having it covered here */ 5561 abort(); 5562 } 5563 5564 desc++; 5565 } 5566 5567 helper_cb_info = (Qcow2AmendHelperCBInfo){ 5568 .original_status_cb = status_cb, 5569 .original_cb_opaque = cb_opaque, 5570 .total_operations = (new_version != old_version) 5571 + (s->refcount_bits != refcount_bits) + 5572 (encryption_update == true) 5573 }; 5574 5575 /* Upgrade first (some features may require compat=1.1) */ 5576 if (new_version > old_version) { 5577 helper_cb_info.current_operation = QCOW2_UPGRADING; 5578 ret = qcow2_upgrade(bs, new_version, &qcow2_amend_helper_cb, 5579 &helper_cb_info, errp); 5580 if (ret < 0) { 5581 return ret; 5582 } 5583 } 5584 5585 if (encryption_update) { 5586 QDict *amend_opts_dict; 5587 QCryptoBlockAmendOptions *amend_opts; 5588 5589 helper_cb_info.current_operation = QCOW2_UPDATING_ENCRYPTION; 5590 amend_opts_dict = qcow2_extract_crypto_opts(opts, "luks", errp); 5591 if (!amend_opts_dict) { 5592 return -EINVAL; 5593 } 5594 amend_opts = block_crypto_amend_opts_init(amend_opts_dict, errp); 5595 qobject_unref(amend_opts_dict); 5596 if (!amend_opts) { 5597 return -EINVAL; 5598 } 5599 ret = qcrypto_block_amend_options(s->crypto, 5600 qcow2_crypto_hdr_read_func, 5601 qcow2_crypto_hdr_write_func, 5602 bs, 5603 amend_opts, 5604 force, 5605 errp); 5606 qapi_free_QCryptoBlockAmendOptions(amend_opts); 5607 if (ret < 0) { 5608 return ret; 5609 } 5610 } 5611 5612 if (s->refcount_bits != refcount_bits) { 5613 int refcount_order = ctz32(refcount_bits); 5614 5615 if (new_version < 3 && refcount_bits != 16) { 5616 error_setg(errp, "Refcount widths other than 16 bits require " 5617 "compatibility level 1.1 or above (use compat=1.1 or " 5618 "greater)"); 5619 return -EINVAL; 5620 } 5621 5622 helper_cb_info.current_operation = QCOW2_CHANGING_REFCOUNT_ORDER; 5623 ret = qcow2_change_refcount_order(bs, refcount_order, 5624 &qcow2_amend_helper_cb, 5625 &helper_cb_info, errp); 5626 if (ret < 0) { 5627 return ret; 5628 } 5629 } 5630 5631 /* data-file-raw blocks backing files, so clear it first if requested */ 5632 if (data_file_raw) { 5633 s->autoclear_features |= QCOW2_AUTOCLEAR_DATA_FILE_RAW; 5634 } else { 5635 s->autoclear_features &= ~QCOW2_AUTOCLEAR_DATA_FILE_RAW; 5636 } 5637 5638 if (data_file) { 5639 g_free(s->image_data_file); 5640 s->image_data_file = *data_file ? g_strdup(data_file) : NULL; 5641 } 5642 5643 ret = qcow2_update_header(bs); 5644 if (ret < 0) { 5645 error_setg_errno(errp, -ret, "Failed to update the image header"); 5646 return ret; 5647 } 5648 5649 if (backing_file || backing_format) { 5650 if (g_strcmp0(backing_file, s->image_backing_file) || 5651 g_strcmp0(backing_format, s->image_backing_format)) { 5652 error_setg(errp, "Cannot amend the backing file"); 5653 error_append_hint(errp, 5654 "You can use 'qemu-img rebase' instead.\n"); 5655 return -EINVAL; 5656 } 5657 } 5658 5659 if (s->use_lazy_refcounts != lazy_refcounts) { 5660 if (lazy_refcounts) { 5661 if (new_version < 3) { 5662 error_setg(errp, "Lazy refcounts only supported with " 5663 "compatibility level 1.1 and above (use compat=1.1 " 5664 "or greater)"); 5665 return -EINVAL; 5666 } 5667 s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS; 5668 ret = qcow2_update_header(bs); 5669 if (ret < 0) { 5670 s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS; 5671 error_setg_errno(errp, -ret, "Failed to update the image header"); 5672 return ret; 5673 } 5674 s->use_lazy_refcounts = true; 5675 } else { 5676 /* make image clean first */ 5677 ret = qcow2_mark_clean(bs); 5678 if (ret < 0) { 5679 error_setg_errno(errp, -ret, "Failed to make the image clean"); 5680 return ret; 5681 } 5682 /* now disallow lazy refcounts */ 5683 s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS; 5684 ret = qcow2_update_header(bs); 5685 if (ret < 0) { 5686 s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS; 5687 error_setg_errno(errp, -ret, "Failed to update the image header"); 5688 return ret; 5689 } 5690 s->use_lazy_refcounts = false; 5691 } 5692 } 5693 5694 if (new_size) { 5695 BlockBackend *blk = blk_new_with_bs(bs, BLK_PERM_RESIZE, BLK_PERM_ALL, 5696 errp); 5697 if (!blk) { 5698 return -EPERM; 5699 } 5700 5701 /* 5702 * Amending image options should ensure that the image has 5703 * exactly the given new values, so pass exact=true here. 5704 */ 5705 ret = blk_truncate(blk, new_size, true, PREALLOC_MODE_OFF, 0, errp); 5706 blk_unref(blk); 5707 if (ret < 0) { 5708 return ret; 5709 } 5710 } 5711 5712 /* Downgrade last (so unsupported features can be removed before) */ 5713 if (new_version < old_version) { 5714 helper_cb_info.current_operation = QCOW2_DOWNGRADING; 5715 ret = qcow2_downgrade(bs, new_version, &qcow2_amend_helper_cb, 5716 &helper_cb_info, errp); 5717 if (ret < 0) { 5718 return ret; 5719 } 5720 } 5721 5722 return 0; 5723 } 5724 5725 static int coroutine_fn qcow2_co_amend(BlockDriverState *bs, 5726 BlockdevAmendOptions *opts, 5727 bool force, 5728 Error **errp) 5729 { 5730 BlockdevAmendOptionsQcow2 *qopts = &opts->u.qcow2; 5731 BDRVQcow2State *s = bs->opaque; 5732 int ret = 0; 5733 5734 if (qopts->has_encrypt) { 5735 if (!s->crypto) { 5736 error_setg(errp, "image is not encrypted, can't amend"); 5737 return -EOPNOTSUPP; 5738 } 5739 5740 if (qopts->encrypt->format != Q_CRYPTO_BLOCK_FORMAT_LUKS) { 5741 error_setg(errp, 5742 "Amend can't be used to change the qcow2 encryption format"); 5743 return -EOPNOTSUPP; 5744 } 5745 5746 if (s->crypt_method_header != QCOW_CRYPT_LUKS) { 5747 error_setg(errp, 5748 "Only LUKS encryption options can be amended for qcow2 with blockdev-amend"); 5749 return -EOPNOTSUPP; 5750 } 5751 5752 ret = qcrypto_block_amend_options(s->crypto, 5753 qcow2_crypto_hdr_read_func, 5754 qcow2_crypto_hdr_write_func, 5755 bs, 5756 qopts->encrypt, 5757 force, 5758 errp); 5759 } 5760 return ret; 5761 } 5762 5763 /* 5764 * If offset or size are negative, respectively, they will not be included in 5765 * the BLOCK_IMAGE_CORRUPTED event emitted. 5766 * fatal will be ignored for read-only BDS; corruptions found there will always 5767 * be considered non-fatal. 5768 */ 5769 void qcow2_signal_corruption(BlockDriverState *bs, bool fatal, int64_t offset, 5770 int64_t size, const char *message_format, ...) 5771 { 5772 BDRVQcow2State *s = bs->opaque; 5773 const char *node_name; 5774 char *message; 5775 va_list ap; 5776 5777 fatal = fatal && bdrv_is_writable(bs); 5778 5779 if (s->signaled_corruption && 5780 (!fatal || (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT))) 5781 { 5782 return; 5783 } 5784 5785 va_start(ap, message_format); 5786 message = g_strdup_vprintf(message_format, ap); 5787 va_end(ap); 5788 5789 if (fatal) { 5790 fprintf(stderr, "qcow2: Marking image as corrupt: %s; further " 5791 "corruption events will be suppressed\n", message); 5792 } else { 5793 fprintf(stderr, "qcow2: Image is corrupt: %s; further non-fatal " 5794 "corruption events will be suppressed\n", message); 5795 } 5796 5797 node_name = bdrv_get_node_name(bs); 5798 qapi_event_send_block_image_corrupted(bdrv_get_device_name(bs), 5799 *node_name != '\0', node_name, 5800 message, offset >= 0, offset, 5801 size >= 0, size, 5802 fatal); 5803 g_free(message); 5804 5805 if (fatal) { 5806 qcow2_mark_corrupt(bs); 5807 bs->drv = NULL; /* make BDS unusable */ 5808 } 5809 5810 s->signaled_corruption = true; 5811 } 5812 5813 #define QCOW_COMMON_OPTIONS \ 5814 { \ 5815 .name = BLOCK_OPT_SIZE, \ 5816 .type = QEMU_OPT_SIZE, \ 5817 .help = "Virtual disk size" \ 5818 }, \ 5819 { \ 5820 .name = BLOCK_OPT_COMPAT_LEVEL, \ 5821 .type = QEMU_OPT_STRING, \ 5822 .help = "Compatibility level (v2 [0.10] or v3 [1.1])" \ 5823 }, \ 5824 { \ 5825 .name = BLOCK_OPT_BACKING_FILE, \ 5826 .type = QEMU_OPT_STRING, \ 5827 .help = "File name of a base image" \ 5828 }, \ 5829 { \ 5830 .name = BLOCK_OPT_BACKING_FMT, \ 5831 .type = QEMU_OPT_STRING, \ 5832 .help = "Image format of the base image" \ 5833 }, \ 5834 { \ 5835 .name = BLOCK_OPT_DATA_FILE, \ 5836 .type = QEMU_OPT_STRING, \ 5837 .help = "File name of an external data file" \ 5838 }, \ 5839 { \ 5840 .name = BLOCK_OPT_DATA_FILE_RAW, \ 5841 .type = QEMU_OPT_BOOL, \ 5842 .help = "The external data file must stay valid " \ 5843 "as a raw image" \ 5844 }, \ 5845 { \ 5846 .name = BLOCK_OPT_LAZY_REFCOUNTS, \ 5847 .type = QEMU_OPT_BOOL, \ 5848 .help = "Postpone refcount updates", \ 5849 .def_value_str = "off" \ 5850 }, \ 5851 { \ 5852 .name = BLOCK_OPT_REFCOUNT_BITS, \ 5853 .type = QEMU_OPT_NUMBER, \ 5854 .help = "Width of a reference count entry in bits", \ 5855 .def_value_str = "16" \ 5856 } 5857 5858 static QemuOptsList qcow2_create_opts = { 5859 .name = "qcow2-create-opts", 5860 .head = QTAILQ_HEAD_INITIALIZER(qcow2_create_opts.head), 5861 .desc = { 5862 { \ 5863 .name = BLOCK_OPT_ENCRYPT, \ 5864 .type = QEMU_OPT_BOOL, \ 5865 .help = "Encrypt the image with format 'aes'. (Deprecated " \ 5866 "in favor of " BLOCK_OPT_ENCRYPT_FORMAT "=aes)", \ 5867 }, \ 5868 { \ 5869 .name = BLOCK_OPT_ENCRYPT_FORMAT, \ 5870 .type = QEMU_OPT_STRING, \ 5871 .help = "Encrypt the image, format choices: 'aes', 'luks'", \ 5872 }, \ 5873 BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.", \ 5874 "ID of secret providing qcow AES key or LUKS passphrase"), \ 5875 BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_ALG("encrypt."), \ 5876 BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_MODE("encrypt."), \ 5877 BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_ALG("encrypt."), \ 5878 BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_HASH_ALG("encrypt."), \ 5879 BLOCK_CRYPTO_OPT_DEF_LUKS_HASH_ALG("encrypt."), \ 5880 BLOCK_CRYPTO_OPT_DEF_LUKS_ITER_TIME("encrypt."), \ 5881 { \ 5882 .name = BLOCK_OPT_CLUSTER_SIZE, \ 5883 .type = QEMU_OPT_SIZE, \ 5884 .help = "qcow2 cluster size", \ 5885 .def_value_str = stringify(DEFAULT_CLUSTER_SIZE) \ 5886 }, \ 5887 { \ 5888 .name = BLOCK_OPT_EXTL2, \ 5889 .type = QEMU_OPT_BOOL, \ 5890 .help = "Extended L2 tables", \ 5891 .def_value_str = "off" \ 5892 }, \ 5893 { \ 5894 .name = BLOCK_OPT_PREALLOC, \ 5895 .type = QEMU_OPT_STRING, \ 5896 .help = "Preallocation mode (allowed values: off, " \ 5897 "metadata, falloc, full)" \ 5898 }, \ 5899 { \ 5900 .name = BLOCK_OPT_COMPRESSION_TYPE, \ 5901 .type = QEMU_OPT_STRING, \ 5902 .help = "Compression method used for image cluster " \ 5903 "compression", \ 5904 .def_value_str = "zlib" \ 5905 }, 5906 QCOW_COMMON_OPTIONS, 5907 { /* end of list */ } 5908 } 5909 }; 5910 5911 static QemuOptsList qcow2_amend_opts = { 5912 .name = "qcow2-amend-opts", 5913 .head = QTAILQ_HEAD_INITIALIZER(qcow2_amend_opts.head), 5914 .desc = { 5915 BLOCK_CRYPTO_OPT_DEF_LUKS_STATE("encrypt."), 5916 BLOCK_CRYPTO_OPT_DEF_LUKS_KEYSLOT("encrypt."), 5917 BLOCK_CRYPTO_OPT_DEF_LUKS_OLD_SECRET("encrypt."), 5918 BLOCK_CRYPTO_OPT_DEF_LUKS_NEW_SECRET("encrypt."), 5919 BLOCK_CRYPTO_OPT_DEF_LUKS_ITER_TIME("encrypt."), 5920 QCOW_COMMON_OPTIONS, 5921 { /* end of list */ } 5922 } 5923 }; 5924 5925 static const char *const qcow2_strong_runtime_opts[] = { 5926 "encrypt." BLOCK_CRYPTO_OPT_QCOW_KEY_SECRET, 5927 5928 NULL 5929 }; 5930 5931 BlockDriver bdrv_qcow2 = { 5932 .format_name = "qcow2", 5933 .instance_size = sizeof(BDRVQcow2State), 5934 .bdrv_probe = qcow2_probe, 5935 .bdrv_open = qcow2_open, 5936 .bdrv_close = qcow2_close, 5937 .bdrv_reopen_prepare = qcow2_reopen_prepare, 5938 .bdrv_reopen_commit = qcow2_reopen_commit, 5939 .bdrv_reopen_commit_post = qcow2_reopen_commit_post, 5940 .bdrv_reopen_abort = qcow2_reopen_abort, 5941 .bdrv_join_options = qcow2_join_options, 5942 .bdrv_child_perm = bdrv_default_perms, 5943 .bdrv_co_create_opts = qcow2_co_create_opts, 5944 .bdrv_co_create = qcow2_co_create, 5945 .bdrv_has_zero_init = qcow2_has_zero_init, 5946 .bdrv_co_block_status = qcow2_co_block_status, 5947 5948 .bdrv_co_preadv_part = qcow2_co_preadv_part, 5949 .bdrv_co_pwritev_part = qcow2_co_pwritev_part, 5950 .bdrv_co_flush_to_os = qcow2_co_flush_to_os, 5951 5952 .bdrv_co_pwrite_zeroes = qcow2_co_pwrite_zeroes, 5953 .bdrv_co_pdiscard = qcow2_co_pdiscard, 5954 .bdrv_co_copy_range_from = qcow2_co_copy_range_from, 5955 .bdrv_co_copy_range_to = qcow2_co_copy_range_to, 5956 .bdrv_co_truncate = qcow2_co_truncate, 5957 .bdrv_co_pwritev_compressed_part = qcow2_co_pwritev_compressed_part, 5958 .bdrv_make_empty = qcow2_make_empty, 5959 5960 .bdrv_snapshot_create = qcow2_snapshot_create, 5961 .bdrv_snapshot_goto = qcow2_snapshot_goto, 5962 .bdrv_snapshot_delete = qcow2_snapshot_delete, 5963 .bdrv_snapshot_list = qcow2_snapshot_list, 5964 .bdrv_snapshot_load_tmp = qcow2_snapshot_load_tmp, 5965 .bdrv_measure = qcow2_measure, 5966 .bdrv_get_info = qcow2_get_info, 5967 .bdrv_get_specific_info = qcow2_get_specific_info, 5968 5969 .bdrv_save_vmstate = qcow2_save_vmstate, 5970 .bdrv_load_vmstate = qcow2_load_vmstate, 5971 5972 .is_format = true, 5973 .supports_backing = true, 5974 .bdrv_change_backing_file = qcow2_change_backing_file, 5975 5976 .bdrv_refresh_limits = qcow2_refresh_limits, 5977 .bdrv_co_invalidate_cache = qcow2_co_invalidate_cache, 5978 .bdrv_inactivate = qcow2_inactivate, 5979 5980 .create_opts = &qcow2_create_opts, 5981 .amend_opts = &qcow2_amend_opts, 5982 .strong_runtime_opts = qcow2_strong_runtime_opts, 5983 .mutable_opts = mutable_opts, 5984 .bdrv_co_check = qcow2_co_check, 5985 .bdrv_amend_options = qcow2_amend_options, 5986 .bdrv_co_amend = qcow2_co_amend, 5987 5988 .bdrv_detach_aio_context = qcow2_detach_aio_context, 5989 .bdrv_attach_aio_context = qcow2_attach_aio_context, 5990 5991 .bdrv_supports_persistent_dirty_bitmap = 5992 qcow2_supports_persistent_dirty_bitmap, 5993 .bdrv_co_can_store_new_dirty_bitmap = qcow2_co_can_store_new_dirty_bitmap, 5994 .bdrv_co_remove_persistent_dirty_bitmap = 5995 qcow2_co_remove_persistent_dirty_bitmap, 5996 }; 5997 5998 static void bdrv_qcow2_init(void) 5999 { 6000 bdrv_register(&bdrv_qcow2); 6001 } 6002 6003 block_init(bdrv_qcow2_init); 6004