1 /* 2 * Block driver for the QCOW version 2 format 3 * 4 * Copyright (c) 2004-2006 Fabrice Bellard 5 * 6 * Permission is hereby granted, free of charge, to any person obtaining a copy 7 * of this software and associated documentation files (the "Software"), to deal 8 * in the Software without restriction, including without limitation the rights 9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 10 * copies of the Software, and to permit persons to whom the Software is 11 * furnished to do so, subject to the following conditions: 12 * 13 * The above copyright notice and this permission notice shall be included in 14 * all copies or substantial portions of the Software. 15 * 16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN 22 * THE SOFTWARE. 23 */ 24 25 #include "qemu/osdep.h" 26 27 #include "block/qdict.h" 28 #include "sysemu/block-backend.h" 29 #include "qemu/main-loop.h" 30 #include "qemu/module.h" 31 #include "qcow2.h" 32 #include "qemu/error-report.h" 33 #include "qapi/error.h" 34 #include "qapi/qapi-events-block-core.h" 35 #include "qapi/qmp/qdict.h" 36 #include "qapi/qmp/qstring.h" 37 #include "trace.h" 38 #include "qemu/option_int.h" 39 #include "qemu/cutils.h" 40 #include "qemu/bswap.h" 41 #include "qemu/memalign.h" 42 #include "qapi/qobject-input-visitor.h" 43 #include "qapi/qapi-visit-block-core.h" 44 #include "crypto.h" 45 #include "block/aio_task.h" 46 47 /* 48 Differences with QCOW: 49 50 - Support for multiple incremental snapshots. 51 - Memory management by reference counts. 52 - Clusters which have a reference count of one have the bit 53 QCOW_OFLAG_COPIED to optimize write performance. 54 - Size of compressed clusters is stored in sectors to reduce bit usage 55 in the cluster offsets. 56 - Support for storing additional data (such as the VM state) in the 57 snapshots. 58 - If a backing store is used, the cluster size is not constrained 59 (could be backported to QCOW). 60 - L2 tables have always a size of one cluster. 61 */ 62 63 64 typedef struct { 65 uint32_t magic; 66 uint32_t len; 67 } QEMU_PACKED QCowExtension; 68 69 #define QCOW2_EXT_MAGIC_END 0 70 #define QCOW2_EXT_MAGIC_BACKING_FORMAT 0xe2792aca 71 #define QCOW2_EXT_MAGIC_FEATURE_TABLE 0x6803f857 72 #define QCOW2_EXT_MAGIC_CRYPTO_HEADER 0x0537be77 73 #define QCOW2_EXT_MAGIC_BITMAPS 0x23852875 74 #define QCOW2_EXT_MAGIC_DATA_FILE 0x44415441 75 76 static int coroutine_fn 77 qcow2_co_preadv_compressed(BlockDriverState *bs, 78 uint64_t l2_entry, 79 uint64_t offset, 80 uint64_t bytes, 81 QEMUIOVector *qiov, 82 size_t qiov_offset); 83 84 static int qcow2_probe(const uint8_t *buf, int buf_size, const char *filename) 85 { 86 const QCowHeader *cow_header = (const void *)buf; 87 88 if (buf_size >= sizeof(QCowHeader) && 89 be32_to_cpu(cow_header->magic) == QCOW_MAGIC && 90 be32_to_cpu(cow_header->version) >= 2) 91 return 100; 92 else 93 return 0; 94 } 95 96 97 static ssize_t qcow2_crypto_hdr_read_func(QCryptoBlock *block, size_t offset, 98 uint8_t *buf, size_t buflen, 99 void *opaque, Error **errp) 100 { 101 BlockDriverState *bs = opaque; 102 BDRVQcow2State *s = bs->opaque; 103 ssize_t ret; 104 105 if ((offset + buflen) > s->crypto_header.length) { 106 error_setg(errp, "Request for data outside of extension header"); 107 return -1; 108 } 109 110 ret = bdrv_pread(bs->file, 111 s->crypto_header.offset + offset, buf, buflen); 112 if (ret < 0) { 113 error_setg_errno(errp, -ret, "Could not read encryption header"); 114 return -1; 115 } 116 return ret; 117 } 118 119 120 static ssize_t qcow2_crypto_hdr_init_func(QCryptoBlock *block, size_t headerlen, 121 void *opaque, Error **errp) 122 { 123 BlockDriverState *bs = opaque; 124 BDRVQcow2State *s = bs->opaque; 125 int64_t ret; 126 int64_t clusterlen; 127 128 ret = qcow2_alloc_clusters(bs, headerlen); 129 if (ret < 0) { 130 error_setg_errno(errp, -ret, 131 "Cannot allocate cluster for LUKS header size %zu", 132 headerlen); 133 return -1; 134 } 135 136 s->crypto_header.length = headerlen; 137 s->crypto_header.offset = ret; 138 139 /* 140 * Zero fill all space in cluster so it has predictable 141 * content, as we may not initialize some regions of the 142 * header (eg only 1 out of 8 key slots will be initialized) 143 */ 144 clusterlen = size_to_clusters(s, headerlen) * s->cluster_size; 145 assert(qcow2_pre_write_overlap_check(bs, 0, ret, clusterlen, false) == 0); 146 ret = bdrv_pwrite_zeroes(bs->file, 147 ret, 148 clusterlen, 0); 149 if (ret < 0) { 150 error_setg_errno(errp, -ret, "Could not zero fill encryption header"); 151 return -1; 152 } 153 154 return ret; 155 } 156 157 158 static ssize_t qcow2_crypto_hdr_write_func(QCryptoBlock *block, size_t offset, 159 const uint8_t *buf, size_t buflen, 160 void *opaque, Error **errp) 161 { 162 BlockDriverState *bs = opaque; 163 BDRVQcow2State *s = bs->opaque; 164 ssize_t ret; 165 166 if ((offset + buflen) > s->crypto_header.length) { 167 error_setg(errp, "Request for data outside of extension header"); 168 return -1; 169 } 170 171 ret = bdrv_pwrite(bs->file, 172 s->crypto_header.offset + offset, buf, buflen); 173 if (ret < 0) { 174 error_setg_errno(errp, -ret, "Could not read encryption header"); 175 return -1; 176 } 177 return ret; 178 } 179 180 static QDict* 181 qcow2_extract_crypto_opts(QemuOpts *opts, const char *fmt, Error **errp) 182 { 183 QDict *cryptoopts_qdict; 184 QDict *opts_qdict; 185 186 /* Extract "encrypt." options into a qdict */ 187 opts_qdict = qemu_opts_to_qdict(opts, NULL); 188 qdict_extract_subqdict(opts_qdict, &cryptoopts_qdict, "encrypt."); 189 qobject_unref(opts_qdict); 190 qdict_put_str(cryptoopts_qdict, "format", fmt); 191 return cryptoopts_qdict; 192 } 193 194 /* 195 * read qcow2 extension and fill bs 196 * start reading from start_offset 197 * finish reading upon magic of value 0 or when end_offset reached 198 * unknown magic is skipped (future extension this version knows nothing about) 199 * return 0 upon success, non-0 otherwise 200 */ 201 static int qcow2_read_extensions(BlockDriverState *bs, uint64_t start_offset, 202 uint64_t end_offset, void **p_feature_table, 203 int flags, bool *need_update_header, 204 Error **errp) 205 { 206 BDRVQcow2State *s = bs->opaque; 207 QCowExtension ext; 208 uint64_t offset; 209 int ret; 210 Qcow2BitmapHeaderExt bitmaps_ext; 211 212 if (need_update_header != NULL) { 213 *need_update_header = false; 214 } 215 216 #ifdef DEBUG_EXT 217 printf("qcow2_read_extensions: start=%ld end=%ld\n", start_offset, end_offset); 218 #endif 219 offset = start_offset; 220 while (offset < end_offset) { 221 222 #ifdef DEBUG_EXT 223 /* Sanity check */ 224 if (offset > s->cluster_size) 225 printf("qcow2_read_extension: suspicious offset %lu\n", offset); 226 227 printf("attempting to read extended header in offset %lu\n", offset); 228 #endif 229 230 ret = bdrv_pread(bs->file, offset, &ext, sizeof(ext)); 231 if (ret < 0) { 232 error_setg_errno(errp, -ret, "qcow2_read_extension: ERROR: " 233 "pread fail from offset %" PRIu64, offset); 234 return 1; 235 } 236 ext.magic = be32_to_cpu(ext.magic); 237 ext.len = be32_to_cpu(ext.len); 238 offset += sizeof(ext); 239 #ifdef DEBUG_EXT 240 printf("ext.magic = 0x%x\n", ext.magic); 241 #endif 242 if (offset > end_offset || ext.len > end_offset - offset) { 243 error_setg(errp, "Header extension too large"); 244 return -EINVAL; 245 } 246 247 switch (ext.magic) { 248 case QCOW2_EXT_MAGIC_END: 249 return 0; 250 251 case QCOW2_EXT_MAGIC_BACKING_FORMAT: 252 if (ext.len >= sizeof(bs->backing_format)) { 253 error_setg(errp, "ERROR: ext_backing_format: len=%" PRIu32 254 " too large (>=%zu)", ext.len, 255 sizeof(bs->backing_format)); 256 return 2; 257 } 258 ret = bdrv_pread(bs->file, offset, bs->backing_format, ext.len); 259 if (ret < 0) { 260 error_setg_errno(errp, -ret, "ERROR: ext_backing_format: " 261 "Could not read format name"); 262 return 3; 263 } 264 bs->backing_format[ext.len] = '\0'; 265 s->image_backing_format = g_strdup(bs->backing_format); 266 #ifdef DEBUG_EXT 267 printf("Qcow2: Got format extension %s\n", bs->backing_format); 268 #endif 269 break; 270 271 case QCOW2_EXT_MAGIC_FEATURE_TABLE: 272 if (p_feature_table != NULL) { 273 void *feature_table = g_malloc0(ext.len + 2 * sizeof(Qcow2Feature)); 274 ret = bdrv_pread(bs->file, offset , feature_table, ext.len); 275 if (ret < 0) { 276 error_setg_errno(errp, -ret, "ERROR: ext_feature_table: " 277 "Could not read table"); 278 return ret; 279 } 280 281 *p_feature_table = feature_table; 282 } 283 break; 284 285 case QCOW2_EXT_MAGIC_CRYPTO_HEADER: { 286 unsigned int cflags = 0; 287 if (s->crypt_method_header != QCOW_CRYPT_LUKS) { 288 error_setg(errp, "CRYPTO header extension only " 289 "expected with LUKS encryption method"); 290 return -EINVAL; 291 } 292 if (ext.len != sizeof(Qcow2CryptoHeaderExtension)) { 293 error_setg(errp, "CRYPTO header extension size %u, " 294 "but expected size %zu", ext.len, 295 sizeof(Qcow2CryptoHeaderExtension)); 296 return -EINVAL; 297 } 298 299 ret = bdrv_pread(bs->file, offset, &s->crypto_header, ext.len); 300 if (ret < 0) { 301 error_setg_errno(errp, -ret, 302 "Unable to read CRYPTO header extension"); 303 return ret; 304 } 305 s->crypto_header.offset = be64_to_cpu(s->crypto_header.offset); 306 s->crypto_header.length = be64_to_cpu(s->crypto_header.length); 307 308 if ((s->crypto_header.offset % s->cluster_size) != 0) { 309 error_setg(errp, "Encryption header offset '%" PRIu64 "' is " 310 "not a multiple of cluster size '%u'", 311 s->crypto_header.offset, s->cluster_size); 312 return -EINVAL; 313 } 314 315 if (flags & BDRV_O_NO_IO) { 316 cflags |= QCRYPTO_BLOCK_OPEN_NO_IO; 317 } 318 s->crypto = qcrypto_block_open(s->crypto_opts, "encrypt.", 319 qcow2_crypto_hdr_read_func, 320 bs, cflags, QCOW2_MAX_THREADS, errp); 321 if (!s->crypto) { 322 return -EINVAL; 323 } 324 } break; 325 326 case QCOW2_EXT_MAGIC_BITMAPS: 327 if (ext.len != sizeof(bitmaps_ext)) { 328 error_setg_errno(errp, -ret, "bitmaps_ext: " 329 "Invalid extension length"); 330 return -EINVAL; 331 } 332 333 if (!(s->autoclear_features & QCOW2_AUTOCLEAR_BITMAPS)) { 334 if (s->qcow_version < 3) { 335 /* Let's be a bit more specific */ 336 warn_report("This qcow2 v2 image contains bitmaps, but " 337 "they may have been modified by a program " 338 "without persistent bitmap support; so now " 339 "they must all be considered inconsistent"); 340 } else { 341 warn_report("a program lacking bitmap support " 342 "modified this file, so all bitmaps are now " 343 "considered inconsistent"); 344 } 345 error_printf("Some clusters may be leaked, " 346 "run 'qemu-img check -r' on the image " 347 "file to fix."); 348 if (need_update_header != NULL) { 349 /* Updating is needed to drop invalid bitmap extension. */ 350 *need_update_header = true; 351 } 352 break; 353 } 354 355 ret = bdrv_pread(bs->file, offset, &bitmaps_ext, ext.len); 356 if (ret < 0) { 357 error_setg_errno(errp, -ret, "bitmaps_ext: " 358 "Could not read ext header"); 359 return ret; 360 } 361 362 if (bitmaps_ext.reserved32 != 0) { 363 error_setg_errno(errp, -ret, "bitmaps_ext: " 364 "Reserved field is not zero"); 365 return -EINVAL; 366 } 367 368 bitmaps_ext.nb_bitmaps = be32_to_cpu(bitmaps_ext.nb_bitmaps); 369 bitmaps_ext.bitmap_directory_size = 370 be64_to_cpu(bitmaps_ext.bitmap_directory_size); 371 bitmaps_ext.bitmap_directory_offset = 372 be64_to_cpu(bitmaps_ext.bitmap_directory_offset); 373 374 if (bitmaps_ext.nb_bitmaps > QCOW2_MAX_BITMAPS) { 375 error_setg(errp, 376 "bitmaps_ext: Image has %" PRIu32 " bitmaps, " 377 "exceeding the QEMU supported maximum of %d", 378 bitmaps_ext.nb_bitmaps, QCOW2_MAX_BITMAPS); 379 return -EINVAL; 380 } 381 382 if (bitmaps_ext.nb_bitmaps == 0) { 383 error_setg(errp, "found bitmaps extension with zero bitmaps"); 384 return -EINVAL; 385 } 386 387 if (offset_into_cluster(s, bitmaps_ext.bitmap_directory_offset)) { 388 error_setg(errp, "bitmaps_ext: " 389 "invalid bitmap directory offset"); 390 return -EINVAL; 391 } 392 393 if (bitmaps_ext.bitmap_directory_size > 394 QCOW2_MAX_BITMAP_DIRECTORY_SIZE) { 395 error_setg(errp, "bitmaps_ext: " 396 "bitmap directory size (%" PRIu64 ") exceeds " 397 "the maximum supported size (%d)", 398 bitmaps_ext.bitmap_directory_size, 399 QCOW2_MAX_BITMAP_DIRECTORY_SIZE); 400 return -EINVAL; 401 } 402 403 s->nb_bitmaps = bitmaps_ext.nb_bitmaps; 404 s->bitmap_directory_offset = 405 bitmaps_ext.bitmap_directory_offset; 406 s->bitmap_directory_size = 407 bitmaps_ext.bitmap_directory_size; 408 409 #ifdef DEBUG_EXT 410 printf("Qcow2: Got bitmaps extension: " 411 "offset=%" PRIu64 " nb_bitmaps=%" PRIu32 "\n", 412 s->bitmap_directory_offset, s->nb_bitmaps); 413 #endif 414 break; 415 416 case QCOW2_EXT_MAGIC_DATA_FILE: 417 { 418 s->image_data_file = g_malloc0(ext.len + 1); 419 ret = bdrv_pread(bs->file, offset, s->image_data_file, ext.len); 420 if (ret < 0) { 421 error_setg_errno(errp, -ret, 422 "ERROR: Could not read data file name"); 423 return ret; 424 } 425 #ifdef DEBUG_EXT 426 printf("Qcow2: Got external data file %s\n", s->image_data_file); 427 #endif 428 break; 429 } 430 431 default: 432 /* unknown magic - save it in case we need to rewrite the header */ 433 /* If you add a new feature, make sure to also update the fast 434 * path of qcow2_make_empty() to deal with it. */ 435 { 436 Qcow2UnknownHeaderExtension *uext; 437 438 uext = g_malloc0(sizeof(*uext) + ext.len); 439 uext->magic = ext.magic; 440 uext->len = ext.len; 441 QLIST_INSERT_HEAD(&s->unknown_header_ext, uext, next); 442 443 ret = bdrv_pread(bs->file, offset , uext->data, uext->len); 444 if (ret < 0) { 445 error_setg_errno(errp, -ret, "ERROR: unknown extension: " 446 "Could not read data"); 447 return ret; 448 } 449 } 450 break; 451 } 452 453 offset += ((ext.len + 7) & ~7); 454 } 455 456 return 0; 457 } 458 459 static void cleanup_unknown_header_ext(BlockDriverState *bs) 460 { 461 BDRVQcow2State *s = bs->opaque; 462 Qcow2UnknownHeaderExtension *uext, *next; 463 464 QLIST_FOREACH_SAFE(uext, &s->unknown_header_ext, next, next) { 465 QLIST_REMOVE(uext, next); 466 g_free(uext); 467 } 468 } 469 470 static void report_unsupported_feature(Error **errp, Qcow2Feature *table, 471 uint64_t mask) 472 { 473 g_autoptr(GString) features = g_string_sized_new(60); 474 475 while (table && table->name[0] != '\0') { 476 if (table->type == QCOW2_FEAT_TYPE_INCOMPATIBLE) { 477 if (mask & (1ULL << table->bit)) { 478 if (features->len > 0) { 479 g_string_append(features, ", "); 480 } 481 g_string_append_printf(features, "%.46s", table->name); 482 mask &= ~(1ULL << table->bit); 483 } 484 } 485 table++; 486 } 487 488 if (mask) { 489 if (features->len > 0) { 490 g_string_append(features, ", "); 491 } 492 g_string_append_printf(features, 493 "Unknown incompatible feature: %" PRIx64, mask); 494 } 495 496 error_setg(errp, "Unsupported qcow2 feature(s): %s", features->str); 497 } 498 499 /* 500 * Sets the dirty bit and flushes afterwards if necessary. 501 * 502 * The incompatible_features bit is only set if the image file header was 503 * updated successfully. Therefore it is not required to check the return 504 * value of this function. 505 */ 506 int qcow2_mark_dirty(BlockDriverState *bs) 507 { 508 BDRVQcow2State *s = bs->opaque; 509 uint64_t val; 510 int ret; 511 512 assert(s->qcow_version >= 3); 513 514 if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) { 515 return 0; /* already dirty */ 516 } 517 518 val = cpu_to_be64(s->incompatible_features | QCOW2_INCOMPAT_DIRTY); 519 ret = bdrv_pwrite(bs->file, offsetof(QCowHeader, incompatible_features), 520 &val, sizeof(val)); 521 if (ret < 0) { 522 return ret; 523 } 524 ret = bdrv_flush(bs->file->bs); 525 if (ret < 0) { 526 return ret; 527 } 528 529 /* Only treat image as dirty if the header was updated successfully */ 530 s->incompatible_features |= QCOW2_INCOMPAT_DIRTY; 531 return 0; 532 } 533 534 /* 535 * Clears the dirty bit and flushes before if necessary. Only call this 536 * function when there are no pending requests, it does not guard against 537 * concurrent requests dirtying the image. 538 */ 539 static int qcow2_mark_clean(BlockDriverState *bs) 540 { 541 BDRVQcow2State *s = bs->opaque; 542 543 if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) { 544 int ret; 545 546 s->incompatible_features &= ~QCOW2_INCOMPAT_DIRTY; 547 548 ret = qcow2_flush_caches(bs); 549 if (ret < 0) { 550 return ret; 551 } 552 553 return qcow2_update_header(bs); 554 } 555 return 0; 556 } 557 558 /* 559 * Marks the image as corrupt. 560 */ 561 int qcow2_mark_corrupt(BlockDriverState *bs) 562 { 563 BDRVQcow2State *s = bs->opaque; 564 565 s->incompatible_features |= QCOW2_INCOMPAT_CORRUPT; 566 return qcow2_update_header(bs); 567 } 568 569 /* 570 * Marks the image as consistent, i.e., unsets the corrupt bit, and flushes 571 * before if necessary. 572 */ 573 int qcow2_mark_consistent(BlockDriverState *bs) 574 { 575 BDRVQcow2State *s = bs->opaque; 576 577 if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) { 578 int ret = qcow2_flush_caches(bs); 579 if (ret < 0) { 580 return ret; 581 } 582 583 s->incompatible_features &= ~QCOW2_INCOMPAT_CORRUPT; 584 return qcow2_update_header(bs); 585 } 586 return 0; 587 } 588 589 static void qcow2_add_check_result(BdrvCheckResult *out, 590 const BdrvCheckResult *src, 591 bool set_allocation_info) 592 { 593 out->corruptions += src->corruptions; 594 out->leaks += src->leaks; 595 out->check_errors += src->check_errors; 596 out->corruptions_fixed += src->corruptions_fixed; 597 out->leaks_fixed += src->leaks_fixed; 598 599 if (set_allocation_info) { 600 out->image_end_offset = src->image_end_offset; 601 out->bfi = src->bfi; 602 } 603 } 604 605 static int coroutine_fn qcow2_co_check_locked(BlockDriverState *bs, 606 BdrvCheckResult *result, 607 BdrvCheckMode fix) 608 { 609 BdrvCheckResult snapshot_res = {}; 610 BdrvCheckResult refcount_res = {}; 611 int ret; 612 613 memset(result, 0, sizeof(*result)); 614 615 ret = qcow2_check_read_snapshot_table(bs, &snapshot_res, fix); 616 if (ret < 0) { 617 qcow2_add_check_result(result, &snapshot_res, false); 618 return ret; 619 } 620 621 ret = qcow2_check_refcounts(bs, &refcount_res, fix); 622 qcow2_add_check_result(result, &refcount_res, true); 623 if (ret < 0) { 624 qcow2_add_check_result(result, &snapshot_res, false); 625 return ret; 626 } 627 628 ret = qcow2_check_fix_snapshot_table(bs, &snapshot_res, fix); 629 qcow2_add_check_result(result, &snapshot_res, false); 630 if (ret < 0) { 631 return ret; 632 } 633 634 if (fix && result->check_errors == 0 && result->corruptions == 0) { 635 ret = qcow2_mark_clean(bs); 636 if (ret < 0) { 637 return ret; 638 } 639 return qcow2_mark_consistent(bs); 640 } 641 return ret; 642 } 643 644 static int coroutine_fn qcow2_co_check(BlockDriverState *bs, 645 BdrvCheckResult *result, 646 BdrvCheckMode fix) 647 { 648 BDRVQcow2State *s = bs->opaque; 649 int ret; 650 651 qemu_co_mutex_lock(&s->lock); 652 ret = qcow2_co_check_locked(bs, result, fix); 653 qemu_co_mutex_unlock(&s->lock); 654 return ret; 655 } 656 657 int qcow2_validate_table(BlockDriverState *bs, uint64_t offset, 658 uint64_t entries, size_t entry_len, 659 int64_t max_size_bytes, const char *table_name, 660 Error **errp) 661 { 662 BDRVQcow2State *s = bs->opaque; 663 664 if (entries > max_size_bytes / entry_len) { 665 error_setg(errp, "%s too large", table_name); 666 return -EFBIG; 667 } 668 669 /* Use signed INT64_MAX as the maximum even for uint64_t header fields, 670 * because values will be passed to qemu functions taking int64_t. */ 671 if ((INT64_MAX - entries * entry_len < offset) || 672 (offset_into_cluster(s, offset) != 0)) { 673 error_setg(errp, "%s offset invalid", table_name); 674 return -EINVAL; 675 } 676 677 return 0; 678 } 679 680 static const char *const mutable_opts[] = { 681 QCOW2_OPT_LAZY_REFCOUNTS, 682 QCOW2_OPT_DISCARD_REQUEST, 683 QCOW2_OPT_DISCARD_SNAPSHOT, 684 QCOW2_OPT_DISCARD_OTHER, 685 QCOW2_OPT_OVERLAP, 686 QCOW2_OPT_OVERLAP_TEMPLATE, 687 QCOW2_OPT_OVERLAP_MAIN_HEADER, 688 QCOW2_OPT_OVERLAP_ACTIVE_L1, 689 QCOW2_OPT_OVERLAP_ACTIVE_L2, 690 QCOW2_OPT_OVERLAP_REFCOUNT_TABLE, 691 QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK, 692 QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE, 693 QCOW2_OPT_OVERLAP_INACTIVE_L1, 694 QCOW2_OPT_OVERLAP_INACTIVE_L2, 695 QCOW2_OPT_OVERLAP_BITMAP_DIRECTORY, 696 QCOW2_OPT_CACHE_SIZE, 697 QCOW2_OPT_L2_CACHE_SIZE, 698 QCOW2_OPT_L2_CACHE_ENTRY_SIZE, 699 QCOW2_OPT_REFCOUNT_CACHE_SIZE, 700 QCOW2_OPT_CACHE_CLEAN_INTERVAL, 701 NULL 702 }; 703 704 static QemuOptsList qcow2_runtime_opts = { 705 .name = "qcow2", 706 .head = QTAILQ_HEAD_INITIALIZER(qcow2_runtime_opts.head), 707 .desc = { 708 { 709 .name = QCOW2_OPT_LAZY_REFCOUNTS, 710 .type = QEMU_OPT_BOOL, 711 .help = "Postpone refcount updates", 712 }, 713 { 714 .name = QCOW2_OPT_DISCARD_REQUEST, 715 .type = QEMU_OPT_BOOL, 716 .help = "Pass guest discard requests to the layer below", 717 }, 718 { 719 .name = QCOW2_OPT_DISCARD_SNAPSHOT, 720 .type = QEMU_OPT_BOOL, 721 .help = "Generate discard requests when snapshot related space " 722 "is freed", 723 }, 724 { 725 .name = QCOW2_OPT_DISCARD_OTHER, 726 .type = QEMU_OPT_BOOL, 727 .help = "Generate discard requests when other clusters are freed", 728 }, 729 { 730 .name = QCOW2_OPT_OVERLAP, 731 .type = QEMU_OPT_STRING, 732 .help = "Selects which overlap checks to perform from a range of " 733 "templates (none, constant, cached, all)", 734 }, 735 { 736 .name = QCOW2_OPT_OVERLAP_TEMPLATE, 737 .type = QEMU_OPT_STRING, 738 .help = "Selects which overlap checks to perform from a range of " 739 "templates (none, constant, cached, all)", 740 }, 741 { 742 .name = QCOW2_OPT_OVERLAP_MAIN_HEADER, 743 .type = QEMU_OPT_BOOL, 744 .help = "Check for unintended writes into the main qcow2 header", 745 }, 746 { 747 .name = QCOW2_OPT_OVERLAP_ACTIVE_L1, 748 .type = QEMU_OPT_BOOL, 749 .help = "Check for unintended writes into the active L1 table", 750 }, 751 { 752 .name = QCOW2_OPT_OVERLAP_ACTIVE_L2, 753 .type = QEMU_OPT_BOOL, 754 .help = "Check for unintended writes into an active L2 table", 755 }, 756 { 757 .name = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE, 758 .type = QEMU_OPT_BOOL, 759 .help = "Check for unintended writes into the refcount table", 760 }, 761 { 762 .name = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK, 763 .type = QEMU_OPT_BOOL, 764 .help = "Check for unintended writes into a refcount block", 765 }, 766 { 767 .name = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE, 768 .type = QEMU_OPT_BOOL, 769 .help = "Check for unintended writes into the snapshot table", 770 }, 771 { 772 .name = QCOW2_OPT_OVERLAP_INACTIVE_L1, 773 .type = QEMU_OPT_BOOL, 774 .help = "Check for unintended writes into an inactive L1 table", 775 }, 776 { 777 .name = QCOW2_OPT_OVERLAP_INACTIVE_L2, 778 .type = QEMU_OPT_BOOL, 779 .help = "Check for unintended writes into an inactive L2 table", 780 }, 781 { 782 .name = QCOW2_OPT_OVERLAP_BITMAP_DIRECTORY, 783 .type = QEMU_OPT_BOOL, 784 .help = "Check for unintended writes into the bitmap directory", 785 }, 786 { 787 .name = QCOW2_OPT_CACHE_SIZE, 788 .type = QEMU_OPT_SIZE, 789 .help = "Maximum combined metadata (L2 tables and refcount blocks) " 790 "cache size", 791 }, 792 { 793 .name = QCOW2_OPT_L2_CACHE_SIZE, 794 .type = QEMU_OPT_SIZE, 795 .help = "Maximum L2 table cache size", 796 }, 797 { 798 .name = QCOW2_OPT_L2_CACHE_ENTRY_SIZE, 799 .type = QEMU_OPT_SIZE, 800 .help = "Size of each entry in the L2 cache", 801 }, 802 { 803 .name = QCOW2_OPT_REFCOUNT_CACHE_SIZE, 804 .type = QEMU_OPT_SIZE, 805 .help = "Maximum refcount block cache size", 806 }, 807 { 808 .name = QCOW2_OPT_CACHE_CLEAN_INTERVAL, 809 .type = QEMU_OPT_NUMBER, 810 .help = "Clean unused cache entries after this time (in seconds)", 811 }, 812 BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.", 813 "ID of secret providing qcow2 AES key or LUKS passphrase"), 814 { /* end of list */ } 815 }, 816 }; 817 818 static const char *overlap_bool_option_names[QCOW2_OL_MAX_BITNR] = { 819 [QCOW2_OL_MAIN_HEADER_BITNR] = QCOW2_OPT_OVERLAP_MAIN_HEADER, 820 [QCOW2_OL_ACTIVE_L1_BITNR] = QCOW2_OPT_OVERLAP_ACTIVE_L1, 821 [QCOW2_OL_ACTIVE_L2_BITNR] = QCOW2_OPT_OVERLAP_ACTIVE_L2, 822 [QCOW2_OL_REFCOUNT_TABLE_BITNR] = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE, 823 [QCOW2_OL_REFCOUNT_BLOCK_BITNR] = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK, 824 [QCOW2_OL_SNAPSHOT_TABLE_BITNR] = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE, 825 [QCOW2_OL_INACTIVE_L1_BITNR] = QCOW2_OPT_OVERLAP_INACTIVE_L1, 826 [QCOW2_OL_INACTIVE_L2_BITNR] = QCOW2_OPT_OVERLAP_INACTIVE_L2, 827 [QCOW2_OL_BITMAP_DIRECTORY_BITNR] = QCOW2_OPT_OVERLAP_BITMAP_DIRECTORY, 828 }; 829 830 static void cache_clean_timer_cb(void *opaque) 831 { 832 BlockDriverState *bs = opaque; 833 BDRVQcow2State *s = bs->opaque; 834 qcow2_cache_clean_unused(s->l2_table_cache); 835 qcow2_cache_clean_unused(s->refcount_block_cache); 836 timer_mod(s->cache_clean_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) + 837 (int64_t) s->cache_clean_interval * 1000); 838 } 839 840 static void cache_clean_timer_init(BlockDriverState *bs, AioContext *context) 841 { 842 BDRVQcow2State *s = bs->opaque; 843 if (s->cache_clean_interval > 0) { 844 s->cache_clean_timer = 845 aio_timer_new_with_attrs(context, QEMU_CLOCK_VIRTUAL, 846 SCALE_MS, QEMU_TIMER_ATTR_EXTERNAL, 847 cache_clean_timer_cb, bs); 848 timer_mod(s->cache_clean_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) + 849 (int64_t) s->cache_clean_interval * 1000); 850 } 851 } 852 853 static void cache_clean_timer_del(BlockDriverState *bs) 854 { 855 BDRVQcow2State *s = bs->opaque; 856 if (s->cache_clean_timer) { 857 timer_free(s->cache_clean_timer); 858 s->cache_clean_timer = NULL; 859 } 860 } 861 862 static void qcow2_detach_aio_context(BlockDriverState *bs) 863 { 864 cache_clean_timer_del(bs); 865 } 866 867 static void qcow2_attach_aio_context(BlockDriverState *bs, 868 AioContext *new_context) 869 { 870 cache_clean_timer_init(bs, new_context); 871 } 872 873 static bool read_cache_sizes(BlockDriverState *bs, QemuOpts *opts, 874 uint64_t *l2_cache_size, 875 uint64_t *l2_cache_entry_size, 876 uint64_t *refcount_cache_size, Error **errp) 877 { 878 BDRVQcow2State *s = bs->opaque; 879 uint64_t combined_cache_size, l2_cache_max_setting; 880 bool l2_cache_size_set, refcount_cache_size_set, combined_cache_size_set; 881 bool l2_cache_entry_size_set; 882 int min_refcount_cache = MIN_REFCOUNT_CACHE_SIZE * s->cluster_size; 883 uint64_t virtual_disk_size = bs->total_sectors * BDRV_SECTOR_SIZE; 884 uint64_t max_l2_entries = DIV_ROUND_UP(virtual_disk_size, s->cluster_size); 885 /* An L2 table is always one cluster in size so the max cache size 886 * should be a multiple of the cluster size. */ 887 uint64_t max_l2_cache = ROUND_UP(max_l2_entries * l2_entry_size(s), 888 s->cluster_size); 889 890 combined_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_CACHE_SIZE); 891 l2_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_L2_CACHE_SIZE); 892 refcount_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_REFCOUNT_CACHE_SIZE); 893 l2_cache_entry_size_set = qemu_opt_get(opts, QCOW2_OPT_L2_CACHE_ENTRY_SIZE); 894 895 combined_cache_size = qemu_opt_get_size(opts, QCOW2_OPT_CACHE_SIZE, 0); 896 l2_cache_max_setting = qemu_opt_get_size(opts, QCOW2_OPT_L2_CACHE_SIZE, 897 DEFAULT_L2_CACHE_MAX_SIZE); 898 *refcount_cache_size = qemu_opt_get_size(opts, 899 QCOW2_OPT_REFCOUNT_CACHE_SIZE, 0); 900 901 *l2_cache_entry_size = qemu_opt_get_size( 902 opts, QCOW2_OPT_L2_CACHE_ENTRY_SIZE, s->cluster_size); 903 904 *l2_cache_size = MIN(max_l2_cache, l2_cache_max_setting); 905 906 if (combined_cache_size_set) { 907 if (l2_cache_size_set && refcount_cache_size_set) { 908 error_setg(errp, QCOW2_OPT_CACHE_SIZE ", " QCOW2_OPT_L2_CACHE_SIZE 909 " and " QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not be set " 910 "at the same time"); 911 return false; 912 } else if (l2_cache_size_set && 913 (l2_cache_max_setting > combined_cache_size)) { 914 error_setg(errp, QCOW2_OPT_L2_CACHE_SIZE " may not exceed " 915 QCOW2_OPT_CACHE_SIZE); 916 return false; 917 } else if (*refcount_cache_size > combined_cache_size) { 918 error_setg(errp, QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not exceed " 919 QCOW2_OPT_CACHE_SIZE); 920 return false; 921 } 922 923 if (l2_cache_size_set) { 924 *refcount_cache_size = combined_cache_size - *l2_cache_size; 925 } else if (refcount_cache_size_set) { 926 *l2_cache_size = combined_cache_size - *refcount_cache_size; 927 } else { 928 /* Assign as much memory as possible to the L2 cache, and 929 * use the remainder for the refcount cache */ 930 if (combined_cache_size >= max_l2_cache + min_refcount_cache) { 931 *l2_cache_size = max_l2_cache; 932 *refcount_cache_size = combined_cache_size - *l2_cache_size; 933 } else { 934 *refcount_cache_size = 935 MIN(combined_cache_size, min_refcount_cache); 936 *l2_cache_size = combined_cache_size - *refcount_cache_size; 937 } 938 } 939 } 940 941 /* 942 * If the L2 cache is not enough to cover the whole disk then 943 * default to 4KB entries. Smaller entries reduce the cost of 944 * loads and evictions and increase I/O performance. 945 */ 946 if (*l2_cache_size < max_l2_cache && !l2_cache_entry_size_set) { 947 *l2_cache_entry_size = MIN(s->cluster_size, 4096); 948 } 949 950 /* l2_cache_size and refcount_cache_size are ensured to have at least 951 * their minimum values in qcow2_update_options_prepare() */ 952 953 if (*l2_cache_entry_size < (1 << MIN_CLUSTER_BITS) || 954 *l2_cache_entry_size > s->cluster_size || 955 !is_power_of_2(*l2_cache_entry_size)) { 956 error_setg(errp, "L2 cache entry size must be a power of two " 957 "between %d and the cluster size (%d)", 958 1 << MIN_CLUSTER_BITS, s->cluster_size); 959 return false; 960 } 961 962 return true; 963 } 964 965 typedef struct Qcow2ReopenState { 966 Qcow2Cache *l2_table_cache; 967 Qcow2Cache *refcount_block_cache; 968 int l2_slice_size; /* Number of entries in a slice of the L2 table */ 969 bool use_lazy_refcounts; 970 int overlap_check; 971 bool discard_passthrough[QCOW2_DISCARD_MAX]; 972 uint64_t cache_clean_interval; 973 QCryptoBlockOpenOptions *crypto_opts; /* Disk encryption runtime options */ 974 } Qcow2ReopenState; 975 976 static int qcow2_update_options_prepare(BlockDriverState *bs, 977 Qcow2ReopenState *r, 978 QDict *options, int flags, 979 Error **errp) 980 { 981 BDRVQcow2State *s = bs->opaque; 982 QemuOpts *opts = NULL; 983 const char *opt_overlap_check, *opt_overlap_check_template; 984 int overlap_check_template = 0; 985 uint64_t l2_cache_size, l2_cache_entry_size, refcount_cache_size; 986 int i; 987 const char *encryptfmt; 988 QDict *encryptopts = NULL; 989 int ret; 990 991 qdict_extract_subqdict(options, &encryptopts, "encrypt."); 992 encryptfmt = qdict_get_try_str(encryptopts, "format"); 993 994 opts = qemu_opts_create(&qcow2_runtime_opts, NULL, 0, &error_abort); 995 if (!qemu_opts_absorb_qdict(opts, options, errp)) { 996 ret = -EINVAL; 997 goto fail; 998 } 999 1000 /* get L2 table/refcount block cache size from command line options */ 1001 if (!read_cache_sizes(bs, opts, &l2_cache_size, &l2_cache_entry_size, 1002 &refcount_cache_size, errp)) { 1003 ret = -EINVAL; 1004 goto fail; 1005 } 1006 1007 l2_cache_size /= l2_cache_entry_size; 1008 if (l2_cache_size < MIN_L2_CACHE_SIZE) { 1009 l2_cache_size = MIN_L2_CACHE_SIZE; 1010 } 1011 if (l2_cache_size > INT_MAX) { 1012 error_setg(errp, "L2 cache size too big"); 1013 ret = -EINVAL; 1014 goto fail; 1015 } 1016 1017 refcount_cache_size /= s->cluster_size; 1018 if (refcount_cache_size < MIN_REFCOUNT_CACHE_SIZE) { 1019 refcount_cache_size = MIN_REFCOUNT_CACHE_SIZE; 1020 } 1021 if (refcount_cache_size > INT_MAX) { 1022 error_setg(errp, "Refcount cache size too big"); 1023 ret = -EINVAL; 1024 goto fail; 1025 } 1026 1027 /* alloc new L2 table/refcount block cache, flush old one */ 1028 if (s->l2_table_cache) { 1029 ret = qcow2_cache_flush(bs, s->l2_table_cache); 1030 if (ret) { 1031 error_setg_errno(errp, -ret, "Failed to flush the L2 table cache"); 1032 goto fail; 1033 } 1034 } 1035 1036 if (s->refcount_block_cache) { 1037 ret = qcow2_cache_flush(bs, s->refcount_block_cache); 1038 if (ret) { 1039 error_setg_errno(errp, -ret, 1040 "Failed to flush the refcount block cache"); 1041 goto fail; 1042 } 1043 } 1044 1045 r->l2_slice_size = l2_cache_entry_size / l2_entry_size(s); 1046 r->l2_table_cache = qcow2_cache_create(bs, l2_cache_size, 1047 l2_cache_entry_size); 1048 r->refcount_block_cache = qcow2_cache_create(bs, refcount_cache_size, 1049 s->cluster_size); 1050 if (r->l2_table_cache == NULL || r->refcount_block_cache == NULL) { 1051 error_setg(errp, "Could not allocate metadata caches"); 1052 ret = -ENOMEM; 1053 goto fail; 1054 } 1055 1056 /* New interval for cache cleanup timer */ 1057 r->cache_clean_interval = 1058 qemu_opt_get_number(opts, QCOW2_OPT_CACHE_CLEAN_INTERVAL, 1059 DEFAULT_CACHE_CLEAN_INTERVAL); 1060 #ifndef CONFIG_LINUX 1061 if (r->cache_clean_interval != 0) { 1062 error_setg(errp, QCOW2_OPT_CACHE_CLEAN_INTERVAL 1063 " not supported on this host"); 1064 ret = -EINVAL; 1065 goto fail; 1066 } 1067 #endif 1068 if (r->cache_clean_interval > UINT_MAX) { 1069 error_setg(errp, "Cache clean interval too big"); 1070 ret = -EINVAL; 1071 goto fail; 1072 } 1073 1074 /* lazy-refcounts; flush if going from enabled to disabled */ 1075 r->use_lazy_refcounts = qemu_opt_get_bool(opts, QCOW2_OPT_LAZY_REFCOUNTS, 1076 (s->compatible_features & QCOW2_COMPAT_LAZY_REFCOUNTS)); 1077 if (r->use_lazy_refcounts && s->qcow_version < 3) { 1078 error_setg(errp, "Lazy refcounts require a qcow2 image with at least " 1079 "qemu 1.1 compatibility level"); 1080 ret = -EINVAL; 1081 goto fail; 1082 } 1083 1084 if (s->use_lazy_refcounts && !r->use_lazy_refcounts) { 1085 ret = qcow2_mark_clean(bs); 1086 if (ret < 0) { 1087 error_setg_errno(errp, -ret, "Failed to disable lazy refcounts"); 1088 goto fail; 1089 } 1090 } 1091 1092 /* Overlap check options */ 1093 opt_overlap_check = qemu_opt_get(opts, QCOW2_OPT_OVERLAP); 1094 opt_overlap_check_template = qemu_opt_get(opts, QCOW2_OPT_OVERLAP_TEMPLATE); 1095 if (opt_overlap_check_template && opt_overlap_check && 1096 strcmp(opt_overlap_check_template, opt_overlap_check)) 1097 { 1098 error_setg(errp, "Conflicting values for qcow2 options '" 1099 QCOW2_OPT_OVERLAP "' ('%s') and '" QCOW2_OPT_OVERLAP_TEMPLATE 1100 "' ('%s')", opt_overlap_check, opt_overlap_check_template); 1101 ret = -EINVAL; 1102 goto fail; 1103 } 1104 if (!opt_overlap_check) { 1105 opt_overlap_check = opt_overlap_check_template ?: "cached"; 1106 } 1107 1108 if (!strcmp(opt_overlap_check, "none")) { 1109 overlap_check_template = 0; 1110 } else if (!strcmp(opt_overlap_check, "constant")) { 1111 overlap_check_template = QCOW2_OL_CONSTANT; 1112 } else if (!strcmp(opt_overlap_check, "cached")) { 1113 overlap_check_template = QCOW2_OL_CACHED; 1114 } else if (!strcmp(opt_overlap_check, "all")) { 1115 overlap_check_template = QCOW2_OL_ALL; 1116 } else { 1117 error_setg(errp, "Unsupported value '%s' for qcow2 option " 1118 "'overlap-check'. Allowed are any of the following: " 1119 "none, constant, cached, all", opt_overlap_check); 1120 ret = -EINVAL; 1121 goto fail; 1122 } 1123 1124 r->overlap_check = 0; 1125 for (i = 0; i < QCOW2_OL_MAX_BITNR; i++) { 1126 /* overlap-check defines a template bitmask, but every flag may be 1127 * overwritten through the associated boolean option */ 1128 r->overlap_check |= 1129 qemu_opt_get_bool(opts, overlap_bool_option_names[i], 1130 overlap_check_template & (1 << i)) << i; 1131 } 1132 1133 r->discard_passthrough[QCOW2_DISCARD_NEVER] = false; 1134 r->discard_passthrough[QCOW2_DISCARD_ALWAYS] = true; 1135 r->discard_passthrough[QCOW2_DISCARD_REQUEST] = 1136 qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_REQUEST, 1137 flags & BDRV_O_UNMAP); 1138 r->discard_passthrough[QCOW2_DISCARD_SNAPSHOT] = 1139 qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_SNAPSHOT, true); 1140 r->discard_passthrough[QCOW2_DISCARD_OTHER] = 1141 qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_OTHER, false); 1142 1143 switch (s->crypt_method_header) { 1144 case QCOW_CRYPT_NONE: 1145 if (encryptfmt) { 1146 error_setg(errp, "No encryption in image header, but options " 1147 "specified format '%s'", encryptfmt); 1148 ret = -EINVAL; 1149 goto fail; 1150 } 1151 break; 1152 1153 case QCOW_CRYPT_AES: 1154 if (encryptfmt && !g_str_equal(encryptfmt, "aes")) { 1155 error_setg(errp, 1156 "Header reported 'aes' encryption format but " 1157 "options specify '%s'", encryptfmt); 1158 ret = -EINVAL; 1159 goto fail; 1160 } 1161 qdict_put_str(encryptopts, "format", "qcow"); 1162 r->crypto_opts = block_crypto_open_opts_init(encryptopts, errp); 1163 if (!r->crypto_opts) { 1164 ret = -EINVAL; 1165 goto fail; 1166 } 1167 break; 1168 1169 case QCOW_CRYPT_LUKS: 1170 if (encryptfmt && !g_str_equal(encryptfmt, "luks")) { 1171 error_setg(errp, 1172 "Header reported 'luks' encryption format but " 1173 "options specify '%s'", encryptfmt); 1174 ret = -EINVAL; 1175 goto fail; 1176 } 1177 qdict_put_str(encryptopts, "format", "luks"); 1178 r->crypto_opts = block_crypto_open_opts_init(encryptopts, errp); 1179 if (!r->crypto_opts) { 1180 ret = -EINVAL; 1181 goto fail; 1182 } 1183 break; 1184 1185 default: 1186 error_setg(errp, "Unsupported encryption method %d", 1187 s->crypt_method_header); 1188 ret = -EINVAL; 1189 goto fail; 1190 } 1191 1192 ret = 0; 1193 fail: 1194 qobject_unref(encryptopts); 1195 qemu_opts_del(opts); 1196 opts = NULL; 1197 return ret; 1198 } 1199 1200 static void qcow2_update_options_commit(BlockDriverState *bs, 1201 Qcow2ReopenState *r) 1202 { 1203 BDRVQcow2State *s = bs->opaque; 1204 int i; 1205 1206 if (s->l2_table_cache) { 1207 qcow2_cache_destroy(s->l2_table_cache); 1208 } 1209 if (s->refcount_block_cache) { 1210 qcow2_cache_destroy(s->refcount_block_cache); 1211 } 1212 s->l2_table_cache = r->l2_table_cache; 1213 s->refcount_block_cache = r->refcount_block_cache; 1214 s->l2_slice_size = r->l2_slice_size; 1215 1216 s->overlap_check = r->overlap_check; 1217 s->use_lazy_refcounts = r->use_lazy_refcounts; 1218 1219 for (i = 0; i < QCOW2_DISCARD_MAX; i++) { 1220 s->discard_passthrough[i] = r->discard_passthrough[i]; 1221 } 1222 1223 if (s->cache_clean_interval != r->cache_clean_interval) { 1224 cache_clean_timer_del(bs); 1225 s->cache_clean_interval = r->cache_clean_interval; 1226 cache_clean_timer_init(bs, bdrv_get_aio_context(bs)); 1227 } 1228 1229 qapi_free_QCryptoBlockOpenOptions(s->crypto_opts); 1230 s->crypto_opts = r->crypto_opts; 1231 } 1232 1233 static void qcow2_update_options_abort(BlockDriverState *bs, 1234 Qcow2ReopenState *r) 1235 { 1236 if (r->l2_table_cache) { 1237 qcow2_cache_destroy(r->l2_table_cache); 1238 } 1239 if (r->refcount_block_cache) { 1240 qcow2_cache_destroy(r->refcount_block_cache); 1241 } 1242 qapi_free_QCryptoBlockOpenOptions(r->crypto_opts); 1243 } 1244 1245 static int qcow2_update_options(BlockDriverState *bs, QDict *options, 1246 int flags, Error **errp) 1247 { 1248 Qcow2ReopenState r = {}; 1249 int ret; 1250 1251 ret = qcow2_update_options_prepare(bs, &r, options, flags, errp); 1252 if (ret >= 0) { 1253 qcow2_update_options_commit(bs, &r); 1254 } else { 1255 qcow2_update_options_abort(bs, &r); 1256 } 1257 1258 return ret; 1259 } 1260 1261 static int validate_compression_type(BDRVQcow2State *s, Error **errp) 1262 { 1263 switch (s->compression_type) { 1264 case QCOW2_COMPRESSION_TYPE_ZLIB: 1265 #ifdef CONFIG_ZSTD 1266 case QCOW2_COMPRESSION_TYPE_ZSTD: 1267 #endif 1268 break; 1269 1270 default: 1271 error_setg(errp, "qcow2: unknown compression type: %u", 1272 s->compression_type); 1273 return -ENOTSUP; 1274 } 1275 1276 /* 1277 * if the compression type differs from QCOW2_COMPRESSION_TYPE_ZLIB 1278 * the incompatible feature flag must be set 1279 */ 1280 if (s->compression_type == QCOW2_COMPRESSION_TYPE_ZLIB) { 1281 if (s->incompatible_features & QCOW2_INCOMPAT_COMPRESSION) { 1282 error_setg(errp, "qcow2: Compression type incompatible feature " 1283 "bit must not be set"); 1284 return -EINVAL; 1285 } 1286 } else { 1287 if (!(s->incompatible_features & QCOW2_INCOMPAT_COMPRESSION)) { 1288 error_setg(errp, "qcow2: Compression type incompatible feature " 1289 "bit must be set"); 1290 return -EINVAL; 1291 } 1292 } 1293 1294 return 0; 1295 } 1296 1297 /* Called with s->lock held. */ 1298 static int coroutine_fn qcow2_do_open(BlockDriverState *bs, QDict *options, 1299 int flags, bool open_data_file, 1300 Error **errp) 1301 { 1302 ERRP_GUARD(); 1303 BDRVQcow2State *s = bs->opaque; 1304 unsigned int len, i; 1305 int ret = 0; 1306 QCowHeader header; 1307 uint64_t ext_end; 1308 uint64_t l1_vm_state_index; 1309 bool update_header = false; 1310 1311 ret = bdrv_pread(bs->file, 0, &header, sizeof(header)); 1312 if (ret < 0) { 1313 error_setg_errno(errp, -ret, "Could not read qcow2 header"); 1314 goto fail; 1315 } 1316 header.magic = be32_to_cpu(header.magic); 1317 header.version = be32_to_cpu(header.version); 1318 header.backing_file_offset = be64_to_cpu(header.backing_file_offset); 1319 header.backing_file_size = be32_to_cpu(header.backing_file_size); 1320 header.size = be64_to_cpu(header.size); 1321 header.cluster_bits = be32_to_cpu(header.cluster_bits); 1322 header.crypt_method = be32_to_cpu(header.crypt_method); 1323 header.l1_table_offset = be64_to_cpu(header.l1_table_offset); 1324 header.l1_size = be32_to_cpu(header.l1_size); 1325 header.refcount_table_offset = be64_to_cpu(header.refcount_table_offset); 1326 header.refcount_table_clusters = 1327 be32_to_cpu(header.refcount_table_clusters); 1328 header.snapshots_offset = be64_to_cpu(header.snapshots_offset); 1329 header.nb_snapshots = be32_to_cpu(header.nb_snapshots); 1330 1331 if (header.magic != QCOW_MAGIC) { 1332 error_setg(errp, "Image is not in qcow2 format"); 1333 ret = -EINVAL; 1334 goto fail; 1335 } 1336 if (header.version < 2 || header.version > 3) { 1337 error_setg(errp, "Unsupported qcow2 version %" PRIu32, header.version); 1338 ret = -ENOTSUP; 1339 goto fail; 1340 } 1341 1342 s->qcow_version = header.version; 1343 1344 /* Initialise cluster size */ 1345 if (header.cluster_bits < MIN_CLUSTER_BITS || 1346 header.cluster_bits > MAX_CLUSTER_BITS) { 1347 error_setg(errp, "Unsupported cluster size: 2^%" PRIu32, 1348 header.cluster_bits); 1349 ret = -EINVAL; 1350 goto fail; 1351 } 1352 1353 s->cluster_bits = header.cluster_bits; 1354 s->cluster_size = 1 << s->cluster_bits; 1355 1356 /* Initialise version 3 header fields */ 1357 if (header.version == 2) { 1358 header.incompatible_features = 0; 1359 header.compatible_features = 0; 1360 header.autoclear_features = 0; 1361 header.refcount_order = 4; 1362 header.header_length = 72; 1363 } else { 1364 header.incompatible_features = 1365 be64_to_cpu(header.incompatible_features); 1366 header.compatible_features = be64_to_cpu(header.compatible_features); 1367 header.autoclear_features = be64_to_cpu(header.autoclear_features); 1368 header.refcount_order = be32_to_cpu(header.refcount_order); 1369 header.header_length = be32_to_cpu(header.header_length); 1370 1371 if (header.header_length < 104) { 1372 error_setg(errp, "qcow2 header too short"); 1373 ret = -EINVAL; 1374 goto fail; 1375 } 1376 } 1377 1378 if (header.header_length > s->cluster_size) { 1379 error_setg(errp, "qcow2 header exceeds cluster size"); 1380 ret = -EINVAL; 1381 goto fail; 1382 } 1383 1384 if (header.header_length > sizeof(header)) { 1385 s->unknown_header_fields_size = header.header_length - sizeof(header); 1386 s->unknown_header_fields = g_malloc(s->unknown_header_fields_size); 1387 ret = bdrv_pread(bs->file, sizeof(header), s->unknown_header_fields, 1388 s->unknown_header_fields_size); 1389 if (ret < 0) { 1390 error_setg_errno(errp, -ret, "Could not read unknown qcow2 header " 1391 "fields"); 1392 goto fail; 1393 } 1394 } 1395 1396 if (header.backing_file_offset > s->cluster_size) { 1397 error_setg(errp, "Invalid backing file offset"); 1398 ret = -EINVAL; 1399 goto fail; 1400 } 1401 1402 if (header.backing_file_offset) { 1403 ext_end = header.backing_file_offset; 1404 } else { 1405 ext_end = 1 << header.cluster_bits; 1406 } 1407 1408 /* Handle feature bits */ 1409 s->incompatible_features = header.incompatible_features; 1410 s->compatible_features = header.compatible_features; 1411 s->autoclear_features = header.autoclear_features; 1412 1413 /* 1414 * Handle compression type 1415 * Older qcow2 images don't contain the compression type header. 1416 * Distinguish them by the header length and use 1417 * the only valid (default) compression type in that case 1418 */ 1419 if (header.header_length > offsetof(QCowHeader, compression_type)) { 1420 s->compression_type = header.compression_type; 1421 } else { 1422 s->compression_type = QCOW2_COMPRESSION_TYPE_ZLIB; 1423 } 1424 1425 ret = validate_compression_type(s, errp); 1426 if (ret) { 1427 goto fail; 1428 } 1429 1430 if (s->incompatible_features & ~QCOW2_INCOMPAT_MASK) { 1431 void *feature_table = NULL; 1432 qcow2_read_extensions(bs, header.header_length, ext_end, 1433 &feature_table, flags, NULL, NULL); 1434 report_unsupported_feature(errp, feature_table, 1435 s->incompatible_features & 1436 ~QCOW2_INCOMPAT_MASK); 1437 ret = -ENOTSUP; 1438 g_free(feature_table); 1439 goto fail; 1440 } 1441 1442 if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) { 1443 /* Corrupt images may not be written to unless they are being repaired 1444 */ 1445 if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_CHECK)) { 1446 error_setg(errp, "qcow2: Image is corrupt; cannot be opened " 1447 "read/write"); 1448 ret = -EACCES; 1449 goto fail; 1450 } 1451 } 1452 1453 s->subclusters_per_cluster = 1454 has_subclusters(s) ? QCOW_EXTL2_SUBCLUSTERS_PER_CLUSTER : 1; 1455 s->subcluster_size = s->cluster_size / s->subclusters_per_cluster; 1456 s->subcluster_bits = ctz32(s->subcluster_size); 1457 1458 if (s->subcluster_size < (1 << MIN_CLUSTER_BITS)) { 1459 error_setg(errp, "Unsupported subcluster size: %d", s->subcluster_size); 1460 ret = -EINVAL; 1461 goto fail; 1462 } 1463 1464 /* Check support for various header values */ 1465 if (header.refcount_order > 6) { 1466 error_setg(errp, "Reference count entry width too large; may not " 1467 "exceed 64 bits"); 1468 ret = -EINVAL; 1469 goto fail; 1470 } 1471 s->refcount_order = header.refcount_order; 1472 s->refcount_bits = 1 << s->refcount_order; 1473 s->refcount_max = UINT64_C(1) << (s->refcount_bits - 1); 1474 s->refcount_max += s->refcount_max - 1; 1475 1476 s->crypt_method_header = header.crypt_method; 1477 if (s->crypt_method_header) { 1478 if (bdrv_uses_whitelist() && 1479 s->crypt_method_header == QCOW_CRYPT_AES) { 1480 error_setg(errp, 1481 "Use of AES-CBC encrypted qcow2 images is no longer " 1482 "supported in system emulators"); 1483 error_append_hint(errp, 1484 "You can use 'qemu-img convert' to convert your " 1485 "image to an alternative supported format, such " 1486 "as unencrypted qcow2, or raw with the LUKS " 1487 "format instead.\n"); 1488 ret = -ENOSYS; 1489 goto fail; 1490 } 1491 1492 if (s->crypt_method_header == QCOW_CRYPT_AES) { 1493 s->crypt_physical_offset = false; 1494 } else { 1495 /* Assuming LUKS and any future crypt methods we 1496 * add will all use physical offsets, due to the 1497 * fact that the alternative is insecure... */ 1498 s->crypt_physical_offset = true; 1499 } 1500 1501 bs->encrypted = true; 1502 } 1503 1504 s->l2_bits = s->cluster_bits - ctz32(l2_entry_size(s)); 1505 s->l2_size = 1 << s->l2_bits; 1506 /* 2^(s->refcount_order - 3) is the refcount width in bytes */ 1507 s->refcount_block_bits = s->cluster_bits - (s->refcount_order - 3); 1508 s->refcount_block_size = 1 << s->refcount_block_bits; 1509 bs->total_sectors = header.size / BDRV_SECTOR_SIZE; 1510 s->csize_shift = (62 - (s->cluster_bits - 8)); 1511 s->csize_mask = (1 << (s->cluster_bits - 8)) - 1; 1512 s->cluster_offset_mask = (1LL << s->csize_shift) - 1; 1513 1514 s->refcount_table_offset = header.refcount_table_offset; 1515 s->refcount_table_size = 1516 header.refcount_table_clusters << (s->cluster_bits - 3); 1517 1518 if (header.refcount_table_clusters == 0 && !(flags & BDRV_O_CHECK)) { 1519 error_setg(errp, "Image does not contain a reference count table"); 1520 ret = -EINVAL; 1521 goto fail; 1522 } 1523 1524 ret = qcow2_validate_table(bs, s->refcount_table_offset, 1525 header.refcount_table_clusters, 1526 s->cluster_size, QCOW_MAX_REFTABLE_SIZE, 1527 "Reference count table", errp); 1528 if (ret < 0) { 1529 goto fail; 1530 } 1531 1532 if (!(flags & BDRV_O_CHECK)) { 1533 /* 1534 * The total size in bytes of the snapshot table is checked in 1535 * qcow2_read_snapshots() because the size of each snapshot is 1536 * variable and we don't know it yet. 1537 * Here we only check the offset and number of snapshots. 1538 */ 1539 ret = qcow2_validate_table(bs, header.snapshots_offset, 1540 header.nb_snapshots, 1541 sizeof(QCowSnapshotHeader), 1542 sizeof(QCowSnapshotHeader) * 1543 QCOW_MAX_SNAPSHOTS, 1544 "Snapshot table", errp); 1545 if (ret < 0) { 1546 goto fail; 1547 } 1548 } 1549 1550 /* read the level 1 table */ 1551 ret = qcow2_validate_table(bs, header.l1_table_offset, 1552 header.l1_size, L1E_SIZE, 1553 QCOW_MAX_L1_SIZE, "Active L1 table", errp); 1554 if (ret < 0) { 1555 goto fail; 1556 } 1557 s->l1_size = header.l1_size; 1558 s->l1_table_offset = header.l1_table_offset; 1559 1560 l1_vm_state_index = size_to_l1(s, header.size); 1561 if (l1_vm_state_index > INT_MAX) { 1562 error_setg(errp, "Image is too big"); 1563 ret = -EFBIG; 1564 goto fail; 1565 } 1566 s->l1_vm_state_index = l1_vm_state_index; 1567 1568 /* the L1 table must contain at least enough entries to put 1569 header.size bytes */ 1570 if (s->l1_size < s->l1_vm_state_index) { 1571 error_setg(errp, "L1 table is too small"); 1572 ret = -EINVAL; 1573 goto fail; 1574 } 1575 1576 if (s->l1_size > 0) { 1577 s->l1_table = qemu_try_blockalign(bs->file->bs, s->l1_size * L1E_SIZE); 1578 if (s->l1_table == NULL) { 1579 error_setg(errp, "Could not allocate L1 table"); 1580 ret = -ENOMEM; 1581 goto fail; 1582 } 1583 ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table, 1584 s->l1_size * L1E_SIZE); 1585 if (ret < 0) { 1586 error_setg_errno(errp, -ret, "Could not read L1 table"); 1587 goto fail; 1588 } 1589 for(i = 0;i < s->l1_size; i++) { 1590 s->l1_table[i] = be64_to_cpu(s->l1_table[i]); 1591 } 1592 } 1593 1594 /* Parse driver-specific options */ 1595 ret = qcow2_update_options(bs, options, flags, errp); 1596 if (ret < 0) { 1597 goto fail; 1598 } 1599 1600 s->flags = flags; 1601 1602 ret = qcow2_refcount_init(bs); 1603 if (ret != 0) { 1604 error_setg_errno(errp, -ret, "Could not initialize refcount handling"); 1605 goto fail; 1606 } 1607 1608 QLIST_INIT(&s->cluster_allocs); 1609 QTAILQ_INIT(&s->discards); 1610 1611 /* read qcow2 extensions */ 1612 if (qcow2_read_extensions(bs, header.header_length, ext_end, NULL, 1613 flags, &update_header, errp)) { 1614 ret = -EINVAL; 1615 goto fail; 1616 } 1617 1618 if (open_data_file) { 1619 /* Open external data file */ 1620 s->data_file = bdrv_open_child(NULL, options, "data-file", bs, 1621 &child_of_bds, BDRV_CHILD_DATA, 1622 true, errp); 1623 if (*errp) { 1624 ret = -EINVAL; 1625 goto fail; 1626 } 1627 1628 if (s->incompatible_features & QCOW2_INCOMPAT_DATA_FILE) { 1629 if (!s->data_file && s->image_data_file) { 1630 s->data_file = bdrv_open_child(s->image_data_file, options, 1631 "data-file", bs, &child_of_bds, 1632 BDRV_CHILD_DATA, false, errp); 1633 if (!s->data_file) { 1634 ret = -EINVAL; 1635 goto fail; 1636 } 1637 } 1638 if (!s->data_file) { 1639 error_setg(errp, "'data-file' is required for this image"); 1640 ret = -EINVAL; 1641 goto fail; 1642 } 1643 1644 /* No data here */ 1645 bs->file->role &= ~BDRV_CHILD_DATA; 1646 1647 /* Must succeed because we have given up permissions if anything */ 1648 bdrv_child_refresh_perms(bs, bs->file, &error_abort); 1649 } else { 1650 if (s->data_file) { 1651 error_setg(errp, "'data-file' can only be set for images with " 1652 "an external data file"); 1653 ret = -EINVAL; 1654 goto fail; 1655 } 1656 1657 s->data_file = bs->file; 1658 1659 if (data_file_is_raw(bs)) { 1660 error_setg(errp, "data-file-raw requires a data file"); 1661 ret = -EINVAL; 1662 goto fail; 1663 } 1664 } 1665 } 1666 1667 /* qcow2_read_extension may have set up the crypto context 1668 * if the crypt method needs a header region, some methods 1669 * don't need header extensions, so must check here 1670 */ 1671 if (s->crypt_method_header && !s->crypto) { 1672 if (s->crypt_method_header == QCOW_CRYPT_AES) { 1673 unsigned int cflags = 0; 1674 if (flags & BDRV_O_NO_IO) { 1675 cflags |= QCRYPTO_BLOCK_OPEN_NO_IO; 1676 } 1677 s->crypto = qcrypto_block_open(s->crypto_opts, "encrypt.", 1678 NULL, NULL, cflags, 1679 QCOW2_MAX_THREADS, errp); 1680 if (!s->crypto) { 1681 ret = -EINVAL; 1682 goto fail; 1683 } 1684 } else if (!(flags & BDRV_O_NO_IO)) { 1685 error_setg(errp, "Missing CRYPTO header for crypt method %d", 1686 s->crypt_method_header); 1687 ret = -EINVAL; 1688 goto fail; 1689 } 1690 } 1691 1692 /* read the backing file name */ 1693 if (header.backing_file_offset != 0) { 1694 len = header.backing_file_size; 1695 if (len > MIN(1023, s->cluster_size - header.backing_file_offset) || 1696 len >= sizeof(bs->backing_file)) { 1697 error_setg(errp, "Backing file name too long"); 1698 ret = -EINVAL; 1699 goto fail; 1700 } 1701 ret = bdrv_pread(bs->file, header.backing_file_offset, 1702 bs->auto_backing_file, len); 1703 if (ret < 0) { 1704 error_setg_errno(errp, -ret, "Could not read backing file name"); 1705 goto fail; 1706 } 1707 bs->auto_backing_file[len] = '\0'; 1708 pstrcpy(bs->backing_file, sizeof(bs->backing_file), 1709 bs->auto_backing_file); 1710 s->image_backing_file = g_strdup(bs->auto_backing_file); 1711 } 1712 1713 /* 1714 * Internal snapshots; skip reading them in check mode, because 1715 * we do not need them then, and we do not want to abort because 1716 * of a broken table. 1717 */ 1718 if (!(flags & BDRV_O_CHECK)) { 1719 s->snapshots_offset = header.snapshots_offset; 1720 s->nb_snapshots = header.nb_snapshots; 1721 1722 ret = qcow2_read_snapshots(bs, errp); 1723 if (ret < 0) { 1724 goto fail; 1725 } 1726 } 1727 1728 /* Clear unknown autoclear feature bits */ 1729 update_header |= s->autoclear_features & ~QCOW2_AUTOCLEAR_MASK; 1730 update_header = update_header && bdrv_is_writable(bs); 1731 if (update_header) { 1732 s->autoclear_features &= QCOW2_AUTOCLEAR_MASK; 1733 } 1734 1735 /* == Handle persistent dirty bitmaps == 1736 * 1737 * We want load dirty bitmaps in three cases: 1738 * 1739 * 1. Normal open of the disk in active mode, not related to invalidation 1740 * after migration. 1741 * 1742 * 2. Invalidation of the target vm after pre-copy phase of migration, if 1743 * bitmaps are _not_ migrating through migration channel, i.e. 1744 * 'dirty-bitmaps' capability is disabled. 1745 * 1746 * 3. Invalidation of source vm after failed or canceled migration. 1747 * This is a very interesting case. There are two possible types of 1748 * bitmaps: 1749 * 1750 * A. Stored on inactivation and removed. They should be loaded from the 1751 * image. 1752 * 1753 * B. Not stored: not-persistent bitmaps and bitmaps, migrated through 1754 * the migration channel (with dirty-bitmaps capability). 1755 * 1756 * On the other hand, there are two possible sub-cases: 1757 * 1758 * 3.1 disk was changed by somebody else while were inactive. In this 1759 * case all in-RAM dirty bitmaps (both persistent and not) are 1760 * definitely invalid. And we don't have any method to determine 1761 * this. 1762 * 1763 * Simple and safe thing is to just drop all the bitmaps of type B on 1764 * inactivation. But in this case we lose bitmaps in valid 4.2 case. 1765 * 1766 * On the other hand, resuming source vm, if disk was already changed 1767 * is a bad thing anyway: not only bitmaps, the whole vm state is 1768 * out of sync with disk. 1769 * 1770 * This means, that user or management tool, who for some reason 1771 * decided to resume source vm, after disk was already changed by 1772 * target vm, should at least drop all dirty bitmaps by hand. 1773 * 1774 * So, we can ignore this case for now, but TODO: "generation" 1775 * extension for qcow2, to determine, that image was changed after 1776 * last inactivation. And if it is changed, we will drop (or at least 1777 * mark as 'invalid' all the bitmaps of type B, both persistent 1778 * and not). 1779 * 1780 * 3.2 disk was _not_ changed while were inactive. Bitmaps may be saved 1781 * to disk ('dirty-bitmaps' capability disabled), or not saved 1782 * ('dirty-bitmaps' capability enabled), but we don't need to care 1783 * of: let's load bitmaps as always: stored bitmaps will be loaded, 1784 * and not stored has flag IN_USE=1 in the image and will be skipped 1785 * on loading. 1786 * 1787 * One remaining possible case when we don't want load bitmaps: 1788 * 1789 * 4. Open disk in inactive mode in target vm (bitmaps are migrating or 1790 * will be loaded on invalidation, no needs try loading them before) 1791 */ 1792 1793 if (!(bdrv_get_flags(bs) & BDRV_O_INACTIVE)) { 1794 /* It's case 1, 2 or 3.2. Or 3.1 which is BUG in management layer. */ 1795 bool header_updated; 1796 if (!qcow2_load_dirty_bitmaps(bs, &header_updated, errp)) { 1797 ret = -EINVAL; 1798 goto fail; 1799 } 1800 1801 update_header = update_header && !header_updated; 1802 } 1803 1804 if (update_header) { 1805 ret = qcow2_update_header(bs); 1806 if (ret < 0) { 1807 error_setg_errno(errp, -ret, "Could not update qcow2 header"); 1808 goto fail; 1809 } 1810 } 1811 1812 bs->supported_zero_flags = header.version >= 3 ? 1813 BDRV_REQ_MAY_UNMAP | BDRV_REQ_NO_FALLBACK : 0; 1814 bs->supported_truncate_flags = BDRV_REQ_ZERO_WRITE; 1815 1816 /* Repair image if dirty */ 1817 if (!(flags & BDRV_O_CHECK) && bdrv_is_writable(bs) && 1818 (s->incompatible_features & QCOW2_INCOMPAT_DIRTY)) { 1819 BdrvCheckResult result = {0}; 1820 1821 ret = qcow2_co_check_locked(bs, &result, 1822 BDRV_FIX_ERRORS | BDRV_FIX_LEAKS); 1823 if (ret < 0 || result.check_errors) { 1824 if (ret >= 0) { 1825 ret = -EIO; 1826 } 1827 error_setg_errno(errp, -ret, "Could not repair dirty image"); 1828 goto fail; 1829 } 1830 } 1831 1832 #ifdef DEBUG_ALLOC 1833 { 1834 BdrvCheckResult result = {0}; 1835 qcow2_check_refcounts(bs, &result, 0); 1836 } 1837 #endif 1838 1839 qemu_co_queue_init(&s->thread_task_queue); 1840 1841 return ret; 1842 1843 fail: 1844 g_free(s->image_data_file); 1845 if (open_data_file && has_data_file(bs)) { 1846 bdrv_unref_child(bs, s->data_file); 1847 s->data_file = NULL; 1848 } 1849 g_free(s->unknown_header_fields); 1850 cleanup_unknown_header_ext(bs); 1851 qcow2_free_snapshots(bs); 1852 qcow2_refcount_close(bs); 1853 qemu_vfree(s->l1_table); 1854 /* else pre-write overlap checks in cache_destroy may crash */ 1855 s->l1_table = NULL; 1856 cache_clean_timer_del(bs); 1857 if (s->l2_table_cache) { 1858 qcow2_cache_destroy(s->l2_table_cache); 1859 } 1860 if (s->refcount_block_cache) { 1861 qcow2_cache_destroy(s->refcount_block_cache); 1862 } 1863 qcrypto_block_free(s->crypto); 1864 qapi_free_QCryptoBlockOpenOptions(s->crypto_opts); 1865 return ret; 1866 } 1867 1868 typedef struct QCow2OpenCo { 1869 BlockDriverState *bs; 1870 QDict *options; 1871 int flags; 1872 Error **errp; 1873 int ret; 1874 } QCow2OpenCo; 1875 1876 static void coroutine_fn qcow2_open_entry(void *opaque) 1877 { 1878 QCow2OpenCo *qoc = opaque; 1879 BDRVQcow2State *s = qoc->bs->opaque; 1880 1881 qemu_co_mutex_lock(&s->lock); 1882 qoc->ret = qcow2_do_open(qoc->bs, qoc->options, qoc->flags, true, 1883 qoc->errp); 1884 qemu_co_mutex_unlock(&s->lock); 1885 } 1886 1887 static int qcow2_open(BlockDriverState *bs, QDict *options, int flags, 1888 Error **errp) 1889 { 1890 BDRVQcow2State *s = bs->opaque; 1891 QCow2OpenCo qoc = { 1892 .bs = bs, 1893 .options = options, 1894 .flags = flags, 1895 .errp = errp, 1896 .ret = -EINPROGRESS 1897 }; 1898 1899 bs->file = bdrv_open_child(NULL, options, "file", bs, &child_of_bds, 1900 BDRV_CHILD_IMAGE, false, errp); 1901 if (!bs->file) { 1902 return -EINVAL; 1903 } 1904 1905 /* Initialise locks */ 1906 qemu_co_mutex_init(&s->lock); 1907 1908 if (qemu_in_coroutine()) { 1909 /* From bdrv_co_create. */ 1910 qcow2_open_entry(&qoc); 1911 } else { 1912 assert(qemu_get_current_aio_context() == qemu_get_aio_context()); 1913 qemu_coroutine_enter(qemu_coroutine_create(qcow2_open_entry, &qoc)); 1914 BDRV_POLL_WHILE(bs, qoc.ret == -EINPROGRESS); 1915 } 1916 return qoc.ret; 1917 } 1918 1919 static void qcow2_refresh_limits(BlockDriverState *bs, Error **errp) 1920 { 1921 BDRVQcow2State *s = bs->opaque; 1922 1923 if (bs->encrypted) { 1924 /* Encryption works on a sector granularity */ 1925 bs->bl.request_alignment = qcrypto_block_get_sector_size(s->crypto); 1926 } 1927 bs->bl.pwrite_zeroes_alignment = s->subcluster_size; 1928 bs->bl.pdiscard_alignment = s->cluster_size; 1929 } 1930 1931 static int qcow2_reopen_prepare(BDRVReopenState *state, 1932 BlockReopenQueue *queue, Error **errp) 1933 { 1934 BDRVQcow2State *s = state->bs->opaque; 1935 Qcow2ReopenState *r; 1936 int ret; 1937 1938 r = g_new0(Qcow2ReopenState, 1); 1939 state->opaque = r; 1940 1941 ret = qcow2_update_options_prepare(state->bs, r, state->options, 1942 state->flags, errp); 1943 if (ret < 0) { 1944 goto fail; 1945 } 1946 1947 /* We need to write out any unwritten data if we reopen read-only. */ 1948 if ((state->flags & BDRV_O_RDWR) == 0) { 1949 ret = qcow2_reopen_bitmaps_ro(state->bs, errp); 1950 if (ret < 0) { 1951 goto fail; 1952 } 1953 1954 ret = bdrv_flush(state->bs); 1955 if (ret < 0) { 1956 goto fail; 1957 } 1958 1959 ret = qcow2_mark_clean(state->bs); 1960 if (ret < 0) { 1961 goto fail; 1962 } 1963 } 1964 1965 /* 1966 * Without an external data file, s->data_file points to the same BdrvChild 1967 * as bs->file. It needs to be resynced after reopen because bs->file may 1968 * be changed. We can't use it in the meantime. 1969 */ 1970 if (!has_data_file(state->bs)) { 1971 assert(s->data_file == state->bs->file); 1972 s->data_file = NULL; 1973 } 1974 1975 return 0; 1976 1977 fail: 1978 qcow2_update_options_abort(state->bs, r); 1979 g_free(r); 1980 return ret; 1981 } 1982 1983 static void qcow2_reopen_commit(BDRVReopenState *state) 1984 { 1985 BDRVQcow2State *s = state->bs->opaque; 1986 1987 qcow2_update_options_commit(state->bs, state->opaque); 1988 if (!s->data_file) { 1989 /* 1990 * If we don't have an external data file, s->data_file was cleared by 1991 * qcow2_reopen_prepare() and needs to be updated. 1992 */ 1993 s->data_file = state->bs->file; 1994 } 1995 g_free(state->opaque); 1996 } 1997 1998 static void qcow2_reopen_commit_post(BDRVReopenState *state) 1999 { 2000 if (state->flags & BDRV_O_RDWR) { 2001 Error *local_err = NULL; 2002 2003 if (qcow2_reopen_bitmaps_rw(state->bs, &local_err) < 0) { 2004 /* 2005 * This is not fatal, bitmaps just left read-only, so all following 2006 * writes will fail. User can remove read-only bitmaps to unblock 2007 * writes or retry reopen. 2008 */ 2009 error_reportf_err(local_err, 2010 "%s: Failed to make dirty bitmaps writable: ", 2011 bdrv_get_node_name(state->bs)); 2012 } 2013 } 2014 } 2015 2016 static void qcow2_reopen_abort(BDRVReopenState *state) 2017 { 2018 BDRVQcow2State *s = state->bs->opaque; 2019 2020 if (!s->data_file) { 2021 /* 2022 * If we don't have an external data file, s->data_file was cleared by 2023 * qcow2_reopen_prepare() and needs to be restored. 2024 */ 2025 s->data_file = state->bs->file; 2026 } 2027 qcow2_update_options_abort(state->bs, state->opaque); 2028 g_free(state->opaque); 2029 } 2030 2031 static void qcow2_join_options(QDict *options, QDict *old_options) 2032 { 2033 bool has_new_overlap_template = 2034 qdict_haskey(options, QCOW2_OPT_OVERLAP) || 2035 qdict_haskey(options, QCOW2_OPT_OVERLAP_TEMPLATE); 2036 bool has_new_total_cache_size = 2037 qdict_haskey(options, QCOW2_OPT_CACHE_SIZE); 2038 bool has_all_cache_options; 2039 2040 /* New overlap template overrides all old overlap options */ 2041 if (has_new_overlap_template) { 2042 qdict_del(old_options, QCOW2_OPT_OVERLAP); 2043 qdict_del(old_options, QCOW2_OPT_OVERLAP_TEMPLATE); 2044 qdict_del(old_options, QCOW2_OPT_OVERLAP_MAIN_HEADER); 2045 qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L1); 2046 qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L2); 2047 qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_TABLE); 2048 qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK); 2049 qdict_del(old_options, QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE); 2050 qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L1); 2051 qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L2); 2052 } 2053 2054 /* New total cache size overrides all old options */ 2055 if (qdict_haskey(options, QCOW2_OPT_CACHE_SIZE)) { 2056 qdict_del(old_options, QCOW2_OPT_L2_CACHE_SIZE); 2057 qdict_del(old_options, QCOW2_OPT_REFCOUNT_CACHE_SIZE); 2058 } 2059 2060 qdict_join(options, old_options, false); 2061 2062 /* 2063 * If after merging all cache size options are set, an old total size is 2064 * overwritten. Do keep all options, however, if all three are new. The 2065 * resulting error message is what we want to happen. 2066 */ 2067 has_all_cache_options = 2068 qdict_haskey(options, QCOW2_OPT_CACHE_SIZE) || 2069 qdict_haskey(options, QCOW2_OPT_L2_CACHE_SIZE) || 2070 qdict_haskey(options, QCOW2_OPT_REFCOUNT_CACHE_SIZE); 2071 2072 if (has_all_cache_options && !has_new_total_cache_size) { 2073 qdict_del(options, QCOW2_OPT_CACHE_SIZE); 2074 } 2075 } 2076 2077 static int coroutine_fn qcow2_co_block_status(BlockDriverState *bs, 2078 bool want_zero, 2079 int64_t offset, int64_t count, 2080 int64_t *pnum, int64_t *map, 2081 BlockDriverState **file) 2082 { 2083 BDRVQcow2State *s = bs->opaque; 2084 uint64_t host_offset; 2085 unsigned int bytes; 2086 QCow2SubclusterType type; 2087 int ret, status = 0; 2088 2089 qemu_co_mutex_lock(&s->lock); 2090 2091 if (!s->metadata_preallocation_checked) { 2092 ret = qcow2_detect_metadata_preallocation(bs); 2093 s->metadata_preallocation = (ret == 1); 2094 s->metadata_preallocation_checked = true; 2095 } 2096 2097 bytes = MIN(INT_MAX, count); 2098 ret = qcow2_get_host_offset(bs, offset, &bytes, &host_offset, &type); 2099 qemu_co_mutex_unlock(&s->lock); 2100 if (ret < 0) { 2101 return ret; 2102 } 2103 2104 *pnum = bytes; 2105 2106 if ((type == QCOW2_SUBCLUSTER_NORMAL || 2107 type == QCOW2_SUBCLUSTER_ZERO_ALLOC || 2108 type == QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC) && !s->crypto) { 2109 *map = host_offset; 2110 *file = s->data_file->bs; 2111 status |= BDRV_BLOCK_OFFSET_VALID; 2112 } 2113 if (type == QCOW2_SUBCLUSTER_ZERO_PLAIN || 2114 type == QCOW2_SUBCLUSTER_ZERO_ALLOC) { 2115 status |= BDRV_BLOCK_ZERO; 2116 } else if (type != QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN && 2117 type != QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC) { 2118 status |= BDRV_BLOCK_DATA; 2119 } 2120 if (s->metadata_preallocation && (status & BDRV_BLOCK_DATA) && 2121 (status & BDRV_BLOCK_OFFSET_VALID)) 2122 { 2123 status |= BDRV_BLOCK_RECURSE; 2124 } 2125 return status; 2126 } 2127 2128 static coroutine_fn int qcow2_handle_l2meta(BlockDriverState *bs, 2129 QCowL2Meta **pl2meta, 2130 bool link_l2) 2131 { 2132 int ret = 0; 2133 QCowL2Meta *l2meta = *pl2meta; 2134 2135 while (l2meta != NULL) { 2136 QCowL2Meta *next; 2137 2138 if (link_l2) { 2139 ret = qcow2_alloc_cluster_link_l2(bs, l2meta); 2140 if (ret) { 2141 goto out; 2142 } 2143 } else { 2144 qcow2_alloc_cluster_abort(bs, l2meta); 2145 } 2146 2147 /* Take the request off the list of running requests */ 2148 QLIST_REMOVE(l2meta, next_in_flight); 2149 2150 qemu_co_queue_restart_all(&l2meta->dependent_requests); 2151 2152 next = l2meta->next; 2153 g_free(l2meta); 2154 l2meta = next; 2155 } 2156 out: 2157 *pl2meta = l2meta; 2158 return ret; 2159 } 2160 2161 static coroutine_fn int 2162 qcow2_co_preadv_encrypted(BlockDriverState *bs, 2163 uint64_t host_offset, 2164 uint64_t offset, 2165 uint64_t bytes, 2166 QEMUIOVector *qiov, 2167 uint64_t qiov_offset) 2168 { 2169 int ret; 2170 BDRVQcow2State *s = bs->opaque; 2171 uint8_t *buf; 2172 2173 assert(bs->encrypted && s->crypto); 2174 assert(bytes <= QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size); 2175 2176 /* 2177 * For encrypted images, read everything into a temporary 2178 * contiguous buffer on which the AES functions can work. 2179 * Also, decryption in a separate buffer is better as it 2180 * prevents the guest from learning information about the 2181 * encrypted nature of the virtual disk. 2182 */ 2183 2184 buf = qemu_try_blockalign(s->data_file->bs, bytes); 2185 if (buf == NULL) { 2186 return -ENOMEM; 2187 } 2188 2189 BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO); 2190 ret = bdrv_co_pread(s->data_file, host_offset, bytes, buf, 0); 2191 if (ret < 0) { 2192 goto fail; 2193 } 2194 2195 if (qcow2_co_decrypt(bs, host_offset, offset, buf, bytes) < 0) 2196 { 2197 ret = -EIO; 2198 goto fail; 2199 } 2200 qemu_iovec_from_buf(qiov, qiov_offset, buf, bytes); 2201 2202 fail: 2203 qemu_vfree(buf); 2204 2205 return ret; 2206 } 2207 2208 typedef struct Qcow2AioTask { 2209 AioTask task; 2210 2211 BlockDriverState *bs; 2212 QCow2SubclusterType subcluster_type; /* only for read */ 2213 uint64_t host_offset; /* or l2_entry for compressed read */ 2214 uint64_t offset; 2215 uint64_t bytes; 2216 QEMUIOVector *qiov; 2217 uint64_t qiov_offset; 2218 QCowL2Meta *l2meta; /* only for write */ 2219 } Qcow2AioTask; 2220 2221 static coroutine_fn int qcow2_co_preadv_task_entry(AioTask *task); 2222 static coroutine_fn int qcow2_add_task(BlockDriverState *bs, 2223 AioTaskPool *pool, 2224 AioTaskFunc func, 2225 QCow2SubclusterType subcluster_type, 2226 uint64_t host_offset, 2227 uint64_t offset, 2228 uint64_t bytes, 2229 QEMUIOVector *qiov, 2230 size_t qiov_offset, 2231 QCowL2Meta *l2meta) 2232 { 2233 Qcow2AioTask local_task; 2234 Qcow2AioTask *task = pool ? g_new(Qcow2AioTask, 1) : &local_task; 2235 2236 *task = (Qcow2AioTask) { 2237 .task.func = func, 2238 .bs = bs, 2239 .subcluster_type = subcluster_type, 2240 .qiov = qiov, 2241 .host_offset = host_offset, 2242 .offset = offset, 2243 .bytes = bytes, 2244 .qiov_offset = qiov_offset, 2245 .l2meta = l2meta, 2246 }; 2247 2248 trace_qcow2_add_task(qemu_coroutine_self(), bs, pool, 2249 func == qcow2_co_preadv_task_entry ? "read" : "write", 2250 subcluster_type, host_offset, offset, bytes, 2251 qiov, qiov_offset); 2252 2253 if (!pool) { 2254 return func(&task->task); 2255 } 2256 2257 aio_task_pool_start_task(pool, &task->task); 2258 2259 return 0; 2260 } 2261 2262 static coroutine_fn int qcow2_co_preadv_task(BlockDriverState *bs, 2263 QCow2SubclusterType subc_type, 2264 uint64_t host_offset, 2265 uint64_t offset, uint64_t bytes, 2266 QEMUIOVector *qiov, 2267 size_t qiov_offset) 2268 { 2269 BDRVQcow2State *s = bs->opaque; 2270 2271 switch (subc_type) { 2272 case QCOW2_SUBCLUSTER_ZERO_PLAIN: 2273 case QCOW2_SUBCLUSTER_ZERO_ALLOC: 2274 /* Both zero types are handled in qcow2_co_preadv_part */ 2275 g_assert_not_reached(); 2276 2277 case QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN: 2278 case QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC: 2279 assert(bs->backing); /* otherwise handled in qcow2_co_preadv_part */ 2280 2281 BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO); 2282 return bdrv_co_preadv_part(bs->backing, offset, bytes, 2283 qiov, qiov_offset, 0); 2284 2285 case QCOW2_SUBCLUSTER_COMPRESSED: 2286 return qcow2_co_preadv_compressed(bs, host_offset, 2287 offset, bytes, qiov, qiov_offset); 2288 2289 case QCOW2_SUBCLUSTER_NORMAL: 2290 if (bs->encrypted) { 2291 return qcow2_co_preadv_encrypted(bs, host_offset, 2292 offset, bytes, qiov, qiov_offset); 2293 } 2294 2295 BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO); 2296 return bdrv_co_preadv_part(s->data_file, host_offset, 2297 bytes, qiov, qiov_offset, 0); 2298 2299 default: 2300 g_assert_not_reached(); 2301 } 2302 2303 g_assert_not_reached(); 2304 } 2305 2306 static coroutine_fn int qcow2_co_preadv_task_entry(AioTask *task) 2307 { 2308 Qcow2AioTask *t = container_of(task, Qcow2AioTask, task); 2309 2310 assert(!t->l2meta); 2311 2312 return qcow2_co_preadv_task(t->bs, t->subcluster_type, 2313 t->host_offset, t->offset, t->bytes, 2314 t->qiov, t->qiov_offset); 2315 } 2316 2317 static coroutine_fn int qcow2_co_preadv_part(BlockDriverState *bs, 2318 int64_t offset, int64_t bytes, 2319 QEMUIOVector *qiov, 2320 size_t qiov_offset, 2321 BdrvRequestFlags flags) 2322 { 2323 BDRVQcow2State *s = bs->opaque; 2324 int ret = 0; 2325 unsigned int cur_bytes; /* number of bytes in current iteration */ 2326 uint64_t host_offset = 0; 2327 QCow2SubclusterType type; 2328 AioTaskPool *aio = NULL; 2329 2330 while (bytes != 0 && aio_task_pool_status(aio) == 0) { 2331 /* prepare next request */ 2332 cur_bytes = MIN(bytes, INT_MAX); 2333 if (s->crypto) { 2334 cur_bytes = MIN(cur_bytes, 2335 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size); 2336 } 2337 2338 qemu_co_mutex_lock(&s->lock); 2339 ret = qcow2_get_host_offset(bs, offset, &cur_bytes, 2340 &host_offset, &type); 2341 qemu_co_mutex_unlock(&s->lock); 2342 if (ret < 0) { 2343 goto out; 2344 } 2345 2346 if (type == QCOW2_SUBCLUSTER_ZERO_PLAIN || 2347 type == QCOW2_SUBCLUSTER_ZERO_ALLOC || 2348 (type == QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN && !bs->backing) || 2349 (type == QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC && !bs->backing)) 2350 { 2351 qemu_iovec_memset(qiov, qiov_offset, 0, cur_bytes); 2352 } else { 2353 if (!aio && cur_bytes != bytes) { 2354 aio = aio_task_pool_new(QCOW2_MAX_WORKERS); 2355 } 2356 ret = qcow2_add_task(bs, aio, qcow2_co_preadv_task_entry, type, 2357 host_offset, offset, cur_bytes, 2358 qiov, qiov_offset, NULL); 2359 if (ret < 0) { 2360 goto out; 2361 } 2362 } 2363 2364 bytes -= cur_bytes; 2365 offset += cur_bytes; 2366 qiov_offset += cur_bytes; 2367 } 2368 2369 out: 2370 if (aio) { 2371 aio_task_pool_wait_all(aio); 2372 if (ret == 0) { 2373 ret = aio_task_pool_status(aio); 2374 } 2375 g_free(aio); 2376 } 2377 2378 return ret; 2379 } 2380 2381 /* Check if it's possible to merge a write request with the writing of 2382 * the data from the COW regions */ 2383 static bool merge_cow(uint64_t offset, unsigned bytes, 2384 QEMUIOVector *qiov, size_t qiov_offset, 2385 QCowL2Meta *l2meta) 2386 { 2387 QCowL2Meta *m; 2388 2389 for (m = l2meta; m != NULL; m = m->next) { 2390 /* If both COW regions are empty then there's nothing to merge */ 2391 if (m->cow_start.nb_bytes == 0 && m->cow_end.nb_bytes == 0) { 2392 continue; 2393 } 2394 2395 /* If COW regions are handled already, skip this too */ 2396 if (m->skip_cow) { 2397 continue; 2398 } 2399 2400 /* 2401 * The write request should start immediately after the first 2402 * COW region. This does not always happen because the area 2403 * touched by the request can be larger than the one defined 2404 * by @m (a single request can span an area consisting of a 2405 * mix of previously unallocated and allocated clusters, that 2406 * is why @l2meta is a list). 2407 */ 2408 if (l2meta_cow_start(m) + m->cow_start.nb_bytes != offset) { 2409 /* In this case the request starts before this region */ 2410 assert(offset < l2meta_cow_start(m)); 2411 assert(m->cow_start.nb_bytes == 0); 2412 continue; 2413 } 2414 2415 /* The write request should end immediately before the second 2416 * COW region (see above for why it does not always happen) */ 2417 if (m->offset + m->cow_end.offset != offset + bytes) { 2418 assert(offset + bytes > m->offset + m->cow_end.offset); 2419 assert(m->cow_end.nb_bytes == 0); 2420 continue; 2421 } 2422 2423 /* Make sure that adding both COW regions to the QEMUIOVector 2424 * does not exceed IOV_MAX */ 2425 if (qemu_iovec_subvec_niov(qiov, qiov_offset, bytes) > IOV_MAX - 2) { 2426 continue; 2427 } 2428 2429 m->data_qiov = qiov; 2430 m->data_qiov_offset = qiov_offset; 2431 return true; 2432 } 2433 2434 return false; 2435 } 2436 2437 /* 2438 * Return 1 if the COW regions read as zeroes, 0 if not, < 0 on error. 2439 * Note that returning 0 does not guarantee non-zero data. 2440 */ 2441 static int is_zero_cow(BlockDriverState *bs, QCowL2Meta *m) 2442 { 2443 /* 2444 * This check is designed for optimization shortcut so it must be 2445 * efficient. 2446 * Instead of is_zero(), use bdrv_co_is_zero_fast() as it is 2447 * faster (but not as accurate and can result in false negatives). 2448 */ 2449 int ret = bdrv_co_is_zero_fast(bs, m->offset + m->cow_start.offset, 2450 m->cow_start.nb_bytes); 2451 if (ret <= 0) { 2452 return ret; 2453 } 2454 2455 return bdrv_co_is_zero_fast(bs, m->offset + m->cow_end.offset, 2456 m->cow_end.nb_bytes); 2457 } 2458 2459 static int handle_alloc_space(BlockDriverState *bs, QCowL2Meta *l2meta) 2460 { 2461 BDRVQcow2State *s = bs->opaque; 2462 QCowL2Meta *m; 2463 2464 if (!(s->data_file->bs->supported_zero_flags & BDRV_REQ_NO_FALLBACK)) { 2465 return 0; 2466 } 2467 2468 if (bs->encrypted) { 2469 return 0; 2470 } 2471 2472 for (m = l2meta; m != NULL; m = m->next) { 2473 int ret; 2474 uint64_t start_offset = m->alloc_offset + m->cow_start.offset; 2475 unsigned nb_bytes = m->cow_end.offset + m->cow_end.nb_bytes - 2476 m->cow_start.offset; 2477 2478 if (!m->cow_start.nb_bytes && !m->cow_end.nb_bytes) { 2479 continue; 2480 } 2481 2482 ret = is_zero_cow(bs, m); 2483 if (ret < 0) { 2484 return ret; 2485 } else if (ret == 0) { 2486 continue; 2487 } 2488 2489 /* 2490 * instead of writing zero COW buffers, 2491 * efficiently zero out the whole clusters 2492 */ 2493 2494 ret = qcow2_pre_write_overlap_check(bs, 0, start_offset, nb_bytes, 2495 true); 2496 if (ret < 0) { 2497 return ret; 2498 } 2499 2500 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC_SPACE); 2501 ret = bdrv_co_pwrite_zeroes(s->data_file, start_offset, nb_bytes, 2502 BDRV_REQ_NO_FALLBACK); 2503 if (ret < 0) { 2504 if (ret != -ENOTSUP && ret != -EAGAIN) { 2505 return ret; 2506 } 2507 continue; 2508 } 2509 2510 trace_qcow2_skip_cow(qemu_coroutine_self(), m->offset, m->nb_clusters); 2511 m->skip_cow = true; 2512 } 2513 return 0; 2514 } 2515 2516 /* 2517 * qcow2_co_pwritev_task 2518 * Called with s->lock unlocked 2519 * l2meta - if not NULL, qcow2_co_pwritev_task() will consume it. Caller must 2520 * not use it somehow after qcow2_co_pwritev_task() call 2521 */ 2522 static coroutine_fn int qcow2_co_pwritev_task(BlockDriverState *bs, 2523 uint64_t host_offset, 2524 uint64_t offset, uint64_t bytes, 2525 QEMUIOVector *qiov, 2526 uint64_t qiov_offset, 2527 QCowL2Meta *l2meta) 2528 { 2529 int ret; 2530 BDRVQcow2State *s = bs->opaque; 2531 void *crypt_buf = NULL; 2532 QEMUIOVector encrypted_qiov; 2533 2534 if (bs->encrypted) { 2535 assert(s->crypto); 2536 assert(bytes <= QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size); 2537 crypt_buf = qemu_try_blockalign(bs->file->bs, bytes); 2538 if (crypt_buf == NULL) { 2539 ret = -ENOMEM; 2540 goto out_unlocked; 2541 } 2542 qemu_iovec_to_buf(qiov, qiov_offset, crypt_buf, bytes); 2543 2544 if (qcow2_co_encrypt(bs, host_offset, offset, crypt_buf, bytes) < 0) { 2545 ret = -EIO; 2546 goto out_unlocked; 2547 } 2548 2549 qemu_iovec_init_buf(&encrypted_qiov, crypt_buf, bytes); 2550 qiov = &encrypted_qiov; 2551 qiov_offset = 0; 2552 } 2553 2554 /* Try to efficiently initialize the physical space with zeroes */ 2555 ret = handle_alloc_space(bs, l2meta); 2556 if (ret < 0) { 2557 goto out_unlocked; 2558 } 2559 2560 /* 2561 * If we need to do COW, check if it's possible to merge the 2562 * writing of the guest data together with that of the COW regions. 2563 * If it's not possible (or not necessary) then write the 2564 * guest data now. 2565 */ 2566 if (!merge_cow(offset, bytes, qiov, qiov_offset, l2meta)) { 2567 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO); 2568 trace_qcow2_writev_data(qemu_coroutine_self(), host_offset); 2569 ret = bdrv_co_pwritev_part(s->data_file, host_offset, 2570 bytes, qiov, qiov_offset, 0); 2571 if (ret < 0) { 2572 goto out_unlocked; 2573 } 2574 } 2575 2576 qemu_co_mutex_lock(&s->lock); 2577 2578 ret = qcow2_handle_l2meta(bs, &l2meta, true); 2579 goto out_locked; 2580 2581 out_unlocked: 2582 qemu_co_mutex_lock(&s->lock); 2583 2584 out_locked: 2585 qcow2_handle_l2meta(bs, &l2meta, false); 2586 qemu_co_mutex_unlock(&s->lock); 2587 2588 qemu_vfree(crypt_buf); 2589 2590 return ret; 2591 } 2592 2593 static coroutine_fn int qcow2_co_pwritev_task_entry(AioTask *task) 2594 { 2595 Qcow2AioTask *t = container_of(task, Qcow2AioTask, task); 2596 2597 assert(!t->subcluster_type); 2598 2599 return qcow2_co_pwritev_task(t->bs, t->host_offset, 2600 t->offset, t->bytes, t->qiov, t->qiov_offset, 2601 t->l2meta); 2602 } 2603 2604 static coroutine_fn int qcow2_co_pwritev_part( 2605 BlockDriverState *bs, int64_t offset, int64_t bytes, 2606 QEMUIOVector *qiov, size_t qiov_offset, BdrvRequestFlags flags) 2607 { 2608 BDRVQcow2State *s = bs->opaque; 2609 int offset_in_cluster; 2610 int ret; 2611 unsigned int cur_bytes; /* number of sectors in current iteration */ 2612 uint64_t host_offset; 2613 QCowL2Meta *l2meta = NULL; 2614 AioTaskPool *aio = NULL; 2615 2616 trace_qcow2_writev_start_req(qemu_coroutine_self(), offset, bytes); 2617 2618 while (bytes != 0 && aio_task_pool_status(aio) == 0) { 2619 2620 l2meta = NULL; 2621 2622 trace_qcow2_writev_start_part(qemu_coroutine_self()); 2623 offset_in_cluster = offset_into_cluster(s, offset); 2624 cur_bytes = MIN(bytes, INT_MAX); 2625 if (bs->encrypted) { 2626 cur_bytes = MIN(cur_bytes, 2627 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size 2628 - offset_in_cluster); 2629 } 2630 2631 qemu_co_mutex_lock(&s->lock); 2632 2633 ret = qcow2_alloc_host_offset(bs, offset, &cur_bytes, 2634 &host_offset, &l2meta); 2635 if (ret < 0) { 2636 goto out_locked; 2637 } 2638 2639 ret = qcow2_pre_write_overlap_check(bs, 0, host_offset, 2640 cur_bytes, true); 2641 if (ret < 0) { 2642 goto out_locked; 2643 } 2644 2645 qemu_co_mutex_unlock(&s->lock); 2646 2647 if (!aio && cur_bytes != bytes) { 2648 aio = aio_task_pool_new(QCOW2_MAX_WORKERS); 2649 } 2650 ret = qcow2_add_task(bs, aio, qcow2_co_pwritev_task_entry, 0, 2651 host_offset, offset, 2652 cur_bytes, qiov, qiov_offset, l2meta); 2653 l2meta = NULL; /* l2meta is consumed by qcow2_co_pwritev_task() */ 2654 if (ret < 0) { 2655 goto fail_nometa; 2656 } 2657 2658 bytes -= cur_bytes; 2659 offset += cur_bytes; 2660 qiov_offset += cur_bytes; 2661 trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_bytes); 2662 } 2663 ret = 0; 2664 2665 qemu_co_mutex_lock(&s->lock); 2666 2667 out_locked: 2668 qcow2_handle_l2meta(bs, &l2meta, false); 2669 2670 qemu_co_mutex_unlock(&s->lock); 2671 2672 fail_nometa: 2673 if (aio) { 2674 aio_task_pool_wait_all(aio); 2675 if (ret == 0) { 2676 ret = aio_task_pool_status(aio); 2677 } 2678 g_free(aio); 2679 } 2680 2681 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret); 2682 2683 return ret; 2684 } 2685 2686 static int qcow2_inactivate(BlockDriverState *bs) 2687 { 2688 BDRVQcow2State *s = bs->opaque; 2689 int ret, result = 0; 2690 Error *local_err = NULL; 2691 2692 qcow2_store_persistent_dirty_bitmaps(bs, true, &local_err); 2693 if (local_err != NULL) { 2694 result = -EINVAL; 2695 error_reportf_err(local_err, "Lost persistent bitmaps during " 2696 "inactivation of node '%s': ", 2697 bdrv_get_device_or_node_name(bs)); 2698 } 2699 2700 ret = qcow2_cache_flush(bs, s->l2_table_cache); 2701 if (ret) { 2702 result = ret; 2703 error_report("Failed to flush the L2 table cache: %s", 2704 strerror(-ret)); 2705 } 2706 2707 ret = qcow2_cache_flush(bs, s->refcount_block_cache); 2708 if (ret) { 2709 result = ret; 2710 error_report("Failed to flush the refcount block cache: %s", 2711 strerror(-ret)); 2712 } 2713 2714 if (result == 0) { 2715 qcow2_mark_clean(bs); 2716 } 2717 2718 return result; 2719 } 2720 2721 static void qcow2_do_close(BlockDriverState *bs, bool close_data_file) 2722 { 2723 BDRVQcow2State *s = bs->opaque; 2724 qemu_vfree(s->l1_table); 2725 /* else pre-write overlap checks in cache_destroy may crash */ 2726 s->l1_table = NULL; 2727 2728 if (!(s->flags & BDRV_O_INACTIVE)) { 2729 qcow2_inactivate(bs); 2730 } 2731 2732 cache_clean_timer_del(bs); 2733 qcow2_cache_destroy(s->l2_table_cache); 2734 qcow2_cache_destroy(s->refcount_block_cache); 2735 2736 qcrypto_block_free(s->crypto); 2737 s->crypto = NULL; 2738 qapi_free_QCryptoBlockOpenOptions(s->crypto_opts); 2739 2740 g_free(s->unknown_header_fields); 2741 cleanup_unknown_header_ext(bs); 2742 2743 g_free(s->image_data_file); 2744 g_free(s->image_backing_file); 2745 g_free(s->image_backing_format); 2746 2747 if (close_data_file && has_data_file(bs)) { 2748 bdrv_unref_child(bs, s->data_file); 2749 s->data_file = NULL; 2750 } 2751 2752 qcow2_refcount_close(bs); 2753 qcow2_free_snapshots(bs); 2754 } 2755 2756 static void qcow2_close(BlockDriverState *bs) 2757 { 2758 qcow2_do_close(bs, true); 2759 } 2760 2761 static void coroutine_fn qcow2_co_invalidate_cache(BlockDriverState *bs, 2762 Error **errp) 2763 { 2764 ERRP_GUARD(); 2765 BDRVQcow2State *s = bs->opaque; 2766 BdrvChild *data_file; 2767 int flags = s->flags; 2768 QCryptoBlock *crypto = NULL; 2769 QDict *options; 2770 int ret; 2771 2772 /* 2773 * Backing files are read-only which makes all of their metadata immutable, 2774 * that means we don't have to worry about reopening them here. 2775 */ 2776 2777 crypto = s->crypto; 2778 s->crypto = NULL; 2779 2780 /* 2781 * Do not reopen s->data_file (i.e., have qcow2_do_close() not close it, 2782 * and then prevent qcow2_do_open() from opening it), because this function 2783 * runs in the I/O path and as such we must not invoke global-state 2784 * functions like bdrv_unref_child() and bdrv_open_child(). 2785 */ 2786 2787 qcow2_do_close(bs, false); 2788 2789 data_file = s->data_file; 2790 memset(s, 0, sizeof(BDRVQcow2State)); 2791 s->data_file = data_file; 2792 2793 options = qdict_clone_shallow(bs->options); 2794 2795 flags &= ~BDRV_O_INACTIVE; 2796 qemu_co_mutex_lock(&s->lock); 2797 ret = qcow2_do_open(bs, options, flags, false, errp); 2798 qemu_co_mutex_unlock(&s->lock); 2799 qobject_unref(options); 2800 if (ret < 0) { 2801 error_prepend(errp, "Could not reopen qcow2 layer: "); 2802 bs->drv = NULL; 2803 return; 2804 } 2805 2806 s->crypto = crypto; 2807 } 2808 2809 static size_t header_ext_add(char *buf, uint32_t magic, const void *s, 2810 size_t len, size_t buflen) 2811 { 2812 QCowExtension *ext_backing_fmt = (QCowExtension*) buf; 2813 size_t ext_len = sizeof(QCowExtension) + ((len + 7) & ~7); 2814 2815 if (buflen < ext_len) { 2816 return -ENOSPC; 2817 } 2818 2819 *ext_backing_fmt = (QCowExtension) { 2820 .magic = cpu_to_be32(magic), 2821 .len = cpu_to_be32(len), 2822 }; 2823 2824 if (len) { 2825 memcpy(buf + sizeof(QCowExtension), s, len); 2826 } 2827 2828 return ext_len; 2829 } 2830 2831 /* 2832 * Updates the qcow2 header, including the variable length parts of it, i.e. 2833 * the backing file name and all extensions. qcow2 was not designed to allow 2834 * such changes, so if we run out of space (we can only use the first cluster) 2835 * this function may fail. 2836 * 2837 * Returns 0 on success, -errno in error cases. 2838 */ 2839 int qcow2_update_header(BlockDriverState *bs) 2840 { 2841 BDRVQcow2State *s = bs->opaque; 2842 QCowHeader *header; 2843 char *buf; 2844 size_t buflen = s->cluster_size; 2845 int ret; 2846 uint64_t total_size; 2847 uint32_t refcount_table_clusters; 2848 size_t header_length; 2849 Qcow2UnknownHeaderExtension *uext; 2850 2851 buf = qemu_blockalign(bs, buflen); 2852 2853 /* Header structure */ 2854 header = (QCowHeader*) buf; 2855 2856 if (buflen < sizeof(*header)) { 2857 ret = -ENOSPC; 2858 goto fail; 2859 } 2860 2861 header_length = sizeof(*header) + s->unknown_header_fields_size; 2862 total_size = bs->total_sectors * BDRV_SECTOR_SIZE; 2863 refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3); 2864 2865 ret = validate_compression_type(s, NULL); 2866 if (ret) { 2867 goto fail; 2868 } 2869 2870 *header = (QCowHeader) { 2871 /* Version 2 fields */ 2872 .magic = cpu_to_be32(QCOW_MAGIC), 2873 .version = cpu_to_be32(s->qcow_version), 2874 .backing_file_offset = 0, 2875 .backing_file_size = 0, 2876 .cluster_bits = cpu_to_be32(s->cluster_bits), 2877 .size = cpu_to_be64(total_size), 2878 .crypt_method = cpu_to_be32(s->crypt_method_header), 2879 .l1_size = cpu_to_be32(s->l1_size), 2880 .l1_table_offset = cpu_to_be64(s->l1_table_offset), 2881 .refcount_table_offset = cpu_to_be64(s->refcount_table_offset), 2882 .refcount_table_clusters = cpu_to_be32(refcount_table_clusters), 2883 .nb_snapshots = cpu_to_be32(s->nb_snapshots), 2884 .snapshots_offset = cpu_to_be64(s->snapshots_offset), 2885 2886 /* Version 3 fields */ 2887 .incompatible_features = cpu_to_be64(s->incompatible_features), 2888 .compatible_features = cpu_to_be64(s->compatible_features), 2889 .autoclear_features = cpu_to_be64(s->autoclear_features), 2890 .refcount_order = cpu_to_be32(s->refcount_order), 2891 .header_length = cpu_to_be32(header_length), 2892 .compression_type = s->compression_type, 2893 }; 2894 2895 /* For older versions, write a shorter header */ 2896 switch (s->qcow_version) { 2897 case 2: 2898 ret = offsetof(QCowHeader, incompatible_features); 2899 break; 2900 case 3: 2901 ret = sizeof(*header); 2902 break; 2903 default: 2904 ret = -EINVAL; 2905 goto fail; 2906 } 2907 2908 buf += ret; 2909 buflen -= ret; 2910 memset(buf, 0, buflen); 2911 2912 /* Preserve any unknown field in the header */ 2913 if (s->unknown_header_fields_size) { 2914 if (buflen < s->unknown_header_fields_size) { 2915 ret = -ENOSPC; 2916 goto fail; 2917 } 2918 2919 memcpy(buf, s->unknown_header_fields, s->unknown_header_fields_size); 2920 buf += s->unknown_header_fields_size; 2921 buflen -= s->unknown_header_fields_size; 2922 } 2923 2924 /* Backing file format header extension */ 2925 if (s->image_backing_format) { 2926 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BACKING_FORMAT, 2927 s->image_backing_format, 2928 strlen(s->image_backing_format), 2929 buflen); 2930 if (ret < 0) { 2931 goto fail; 2932 } 2933 2934 buf += ret; 2935 buflen -= ret; 2936 } 2937 2938 /* External data file header extension */ 2939 if (has_data_file(bs) && s->image_data_file) { 2940 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_DATA_FILE, 2941 s->image_data_file, strlen(s->image_data_file), 2942 buflen); 2943 if (ret < 0) { 2944 goto fail; 2945 } 2946 2947 buf += ret; 2948 buflen -= ret; 2949 } 2950 2951 /* Full disk encryption header pointer extension */ 2952 if (s->crypto_header.offset != 0) { 2953 s->crypto_header.offset = cpu_to_be64(s->crypto_header.offset); 2954 s->crypto_header.length = cpu_to_be64(s->crypto_header.length); 2955 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_CRYPTO_HEADER, 2956 &s->crypto_header, sizeof(s->crypto_header), 2957 buflen); 2958 s->crypto_header.offset = be64_to_cpu(s->crypto_header.offset); 2959 s->crypto_header.length = be64_to_cpu(s->crypto_header.length); 2960 if (ret < 0) { 2961 goto fail; 2962 } 2963 buf += ret; 2964 buflen -= ret; 2965 } 2966 2967 /* 2968 * Feature table. A mere 8 feature names occupies 392 bytes, and 2969 * when coupled with the v3 minimum header of 104 bytes plus the 2970 * 8-byte end-of-extension marker, that would leave only 8 bytes 2971 * for a backing file name in an image with 512-byte clusters. 2972 * Thus, we choose to omit this header for cluster sizes 4k and 2973 * smaller. 2974 */ 2975 if (s->qcow_version >= 3 && s->cluster_size > 4096) { 2976 static const Qcow2Feature features[] = { 2977 { 2978 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE, 2979 .bit = QCOW2_INCOMPAT_DIRTY_BITNR, 2980 .name = "dirty bit", 2981 }, 2982 { 2983 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE, 2984 .bit = QCOW2_INCOMPAT_CORRUPT_BITNR, 2985 .name = "corrupt bit", 2986 }, 2987 { 2988 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE, 2989 .bit = QCOW2_INCOMPAT_DATA_FILE_BITNR, 2990 .name = "external data file", 2991 }, 2992 { 2993 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE, 2994 .bit = QCOW2_INCOMPAT_COMPRESSION_BITNR, 2995 .name = "compression type", 2996 }, 2997 { 2998 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE, 2999 .bit = QCOW2_INCOMPAT_EXTL2_BITNR, 3000 .name = "extended L2 entries", 3001 }, 3002 { 3003 .type = QCOW2_FEAT_TYPE_COMPATIBLE, 3004 .bit = QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR, 3005 .name = "lazy refcounts", 3006 }, 3007 { 3008 .type = QCOW2_FEAT_TYPE_AUTOCLEAR, 3009 .bit = QCOW2_AUTOCLEAR_BITMAPS_BITNR, 3010 .name = "bitmaps", 3011 }, 3012 { 3013 .type = QCOW2_FEAT_TYPE_AUTOCLEAR, 3014 .bit = QCOW2_AUTOCLEAR_DATA_FILE_RAW_BITNR, 3015 .name = "raw external data", 3016 }, 3017 }; 3018 3019 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_FEATURE_TABLE, 3020 features, sizeof(features), buflen); 3021 if (ret < 0) { 3022 goto fail; 3023 } 3024 buf += ret; 3025 buflen -= ret; 3026 } 3027 3028 /* Bitmap extension */ 3029 if (s->nb_bitmaps > 0) { 3030 Qcow2BitmapHeaderExt bitmaps_header = { 3031 .nb_bitmaps = cpu_to_be32(s->nb_bitmaps), 3032 .bitmap_directory_size = 3033 cpu_to_be64(s->bitmap_directory_size), 3034 .bitmap_directory_offset = 3035 cpu_to_be64(s->bitmap_directory_offset) 3036 }; 3037 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BITMAPS, 3038 &bitmaps_header, sizeof(bitmaps_header), 3039 buflen); 3040 if (ret < 0) { 3041 goto fail; 3042 } 3043 buf += ret; 3044 buflen -= ret; 3045 } 3046 3047 /* Keep unknown header extensions */ 3048 QLIST_FOREACH(uext, &s->unknown_header_ext, next) { 3049 ret = header_ext_add(buf, uext->magic, uext->data, uext->len, buflen); 3050 if (ret < 0) { 3051 goto fail; 3052 } 3053 3054 buf += ret; 3055 buflen -= ret; 3056 } 3057 3058 /* End of header extensions */ 3059 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_END, NULL, 0, buflen); 3060 if (ret < 0) { 3061 goto fail; 3062 } 3063 3064 buf += ret; 3065 buflen -= ret; 3066 3067 /* Backing file name */ 3068 if (s->image_backing_file) { 3069 size_t backing_file_len = strlen(s->image_backing_file); 3070 3071 if (buflen < backing_file_len) { 3072 ret = -ENOSPC; 3073 goto fail; 3074 } 3075 3076 /* Using strncpy is ok here, since buf is not NUL-terminated. */ 3077 strncpy(buf, s->image_backing_file, buflen); 3078 3079 header->backing_file_offset = cpu_to_be64(buf - ((char*) header)); 3080 header->backing_file_size = cpu_to_be32(backing_file_len); 3081 } 3082 3083 /* Write the new header */ 3084 ret = bdrv_pwrite(bs->file, 0, header, s->cluster_size); 3085 if (ret < 0) { 3086 goto fail; 3087 } 3088 3089 ret = 0; 3090 fail: 3091 qemu_vfree(header); 3092 return ret; 3093 } 3094 3095 static int qcow2_change_backing_file(BlockDriverState *bs, 3096 const char *backing_file, const char *backing_fmt) 3097 { 3098 BDRVQcow2State *s = bs->opaque; 3099 3100 /* Adding a backing file means that the external data file alone won't be 3101 * enough to make sense of the content */ 3102 if (backing_file && data_file_is_raw(bs)) { 3103 return -EINVAL; 3104 } 3105 3106 if (backing_file && strlen(backing_file) > 1023) { 3107 return -EINVAL; 3108 } 3109 3110 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file), 3111 backing_file ?: ""); 3112 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: ""); 3113 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: ""); 3114 3115 g_free(s->image_backing_file); 3116 g_free(s->image_backing_format); 3117 3118 s->image_backing_file = backing_file ? g_strdup(bs->backing_file) : NULL; 3119 s->image_backing_format = backing_fmt ? g_strdup(bs->backing_format) : NULL; 3120 3121 return qcow2_update_header(bs); 3122 } 3123 3124 static int qcow2_set_up_encryption(BlockDriverState *bs, 3125 QCryptoBlockCreateOptions *cryptoopts, 3126 Error **errp) 3127 { 3128 BDRVQcow2State *s = bs->opaque; 3129 QCryptoBlock *crypto = NULL; 3130 int fmt, ret; 3131 3132 switch (cryptoopts->format) { 3133 case Q_CRYPTO_BLOCK_FORMAT_LUKS: 3134 fmt = QCOW_CRYPT_LUKS; 3135 break; 3136 case Q_CRYPTO_BLOCK_FORMAT_QCOW: 3137 fmt = QCOW_CRYPT_AES; 3138 break; 3139 default: 3140 error_setg(errp, "Crypto format not supported in qcow2"); 3141 return -EINVAL; 3142 } 3143 3144 s->crypt_method_header = fmt; 3145 3146 crypto = qcrypto_block_create(cryptoopts, "encrypt.", 3147 qcow2_crypto_hdr_init_func, 3148 qcow2_crypto_hdr_write_func, 3149 bs, errp); 3150 if (!crypto) { 3151 return -EINVAL; 3152 } 3153 3154 ret = qcow2_update_header(bs); 3155 if (ret < 0) { 3156 error_setg_errno(errp, -ret, "Could not write encryption header"); 3157 goto out; 3158 } 3159 3160 ret = 0; 3161 out: 3162 qcrypto_block_free(crypto); 3163 return ret; 3164 } 3165 3166 /** 3167 * Preallocates metadata structures for data clusters between @offset (in the 3168 * guest disk) and @new_length (which is thus generally the new guest disk 3169 * size). 3170 * 3171 * Returns: 0 on success, -errno on failure. 3172 */ 3173 static int coroutine_fn preallocate_co(BlockDriverState *bs, uint64_t offset, 3174 uint64_t new_length, PreallocMode mode, 3175 Error **errp) 3176 { 3177 BDRVQcow2State *s = bs->opaque; 3178 uint64_t bytes; 3179 uint64_t host_offset = 0; 3180 int64_t file_length; 3181 unsigned int cur_bytes; 3182 int ret; 3183 QCowL2Meta *meta = NULL, *m; 3184 3185 assert(offset <= new_length); 3186 bytes = new_length - offset; 3187 3188 while (bytes) { 3189 cur_bytes = MIN(bytes, QEMU_ALIGN_DOWN(INT_MAX, s->cluster_size)); 3190 ret = qcow2_alloc_host_offset(bs, offset, &cur_bytes, 3191 &host_offset, &meta); 3192 if (ret < 0) { 3193 error_setg_errno(errp, -ret, "Allocating clusters failed"); 3194 goto out; 3195 } 3196 3197 for (m = meta; m != NULL; m = m->next) { 3198 m->prealloc = true; 3199 } 3200 3201 ret = qcow2_handle_l2meta(bs, &meta, true); 3202 if (ret < 0) { 3203 error_setg_errno(errp, -ret, "Mapping clusters failed"); 3204 goto out; 3205 } 3206 3207 /* TODO Preallocate data if requested */ 3208 3209 bytes -= cur_bytes; 3210 offset += cur_bytes; 3211 } 3212 3213 /* 3214 * It is expected that the image file is large enough to actually contain 3215 * all of the allocated clusters (otherwise we get failing reads after 3216 * EOF). Extend the image to the last allocated sector. 3217 */ 3218 file_length = bdrv_getlength(s->data_file->bs); 3219 if (file_length < 0) { 3220 error_setg_errno(errp, -file_length, "Could not get file size"); 3221 ret = file_length; 3222 goto out; 3223 } 3224 3225 if (host_offset + cur_bytes > file_length) { 3226 if (mode == PREALLOC_MODE_METADATA) { 3227 mode = PREALLOC_MODE_OFF; 3228 } 3229 ret = bdrv_co_truncate(s->data_file, host_offset + cur_bytes, false, 3230 mode, 0, errp); 3231 if (ret < 0) { 3232 goto out; 3233 } 3234 } 3235 3236 ret = 0; 3237 3238 out: 3239 qcow2_handle_l2meta(bs, &meta, false); 3240 return ret; 3241 } 3242 3243 /* qcow2_refcount_metadata_size: 3244 * @clusters: number of clusters to refcount (including data and L1/L2 tables) 3245 * @cluster_size: size of a cluster, in bytes 3246 * @refcount_order: refcount bits power-of-2 exponent 3247 * @generous_increase: allow for the refcount table to be 1.5x as large as it 3248 * needs to be 3249 * 3250 * Returns: Number of bytes required for refcount blocks and table metadata. 3251 */ 3252 int64_t qcow2_refcount_metadata_size(int64_t clusters, size_t cluster_size, 3253 int refcount_order, bool generous_increase, 3254 uint64_t *refblock_count) 3255 { 3256 /* 3257 * Every host cluster is reference-counted, including metadata (even 3258 * refcount metadata is recursively included). 3259 * 3260 * An accurate formula for the size of refcount metadata size is difficult 3261 * to derive. An easier method of calculation is finding the fixed point 3262 * where no further refcount blocks or table clusters are required to 3263 * reference count every cluster. 3264 */ 3265 int64_t blocks_per_table_cluster = cluster_size / REFTABLE_ENTRY_SIZE; 3266 int64_t refcounts_per_block = cluster_size * 8 / (1 << refcount_order); 3267 int64_t table = 0; /* number of refcount table clusters */ 3268 int64_t blocks = 0; /* number of refcount block clusters */ 3269 int64_t last; 3270 int64_t n = 0; 3271 3272 do { 3273 last = n; 3274 blocks = DIV_ROUND_UP(clusters + table + blocks, refcounts_per_block); 3275 table = DIV_ROUND_UP(blocks, blocks_per_table_cluster); 3276 n = clusters + blocks + table; 3277 3278 if (n == last && generous_increase) { 3279 clusters += DIV_ROUND_UP(table, 2); 3280 n = 0; /* force another loop */ 3281 generous_increase = false; 3282 } 3283 } while (n != last); 3284 3285 if (refblock_count) { 3286 *refblock_count = blocks; 3287 } 3288 3289 return (blocks + table) * cluster_size; 3290 } 3291 3292 /** 3293 * qcow2_calc_prealloc_size: 3294 * @total_size: virtual disk size in bytes 3295 * @cluster_size: cluster size in bytes 3296 * @refcount_order: refcount bits power-of-2 exponent 3297 * @extended_l2: true if the image has extended L2 entries 3298 * 3299 * Returns: Total number of bytes required for the fully allocated image 3300 * (including metadata). 3301 */ 3302 static int64_t qcow2_calc_prealloc_size(int64_t total_size, 3303 size_t cluster_size, 3304 int refcount_order, 3305 bool extended_l2) 3306 { 3307 int64_t meta_size = 0; 3308 uint64_t nl1e, nl2e; 3309 int64_t aligned_total_size = ROUND_UP(total_size, cluster_size); 3310 size_t l2e_size = extended_l2 ? L2E_SIZE_EXTENDED : L2E_SIZE_NORMAL; 3311 3312 /* header: 1 cluster */ 3313 meta_size += cluster_size; 3314 3315 /* total size of L2 tables */ 3316 nl2e = aligned_total_size / cluster_size; 3317 nl2e = ROUND_UP(nl2e, cluster_size / l2e_size); 3318 meta_size += nl2e * l2e_size; 3319 3320 /* total size of L1 tables */ 3321 nl1e = nl2e * l2e_size / cluster_size; 3322 nl1e = ROUND_UP(nl1e, cluster_size / L1E_SIZE); 3323 meta_size += nl1e * L1E_SIZE; 3324 3325 /* total size of refcount table and blocks */ 3326 meta_size += qcow2_refcount_metadata_size( 3327 (meta_size + aligned_total_size) / cluster_size, 3328 cluster_size, refcount_order, false, NULL); 3329 3330 return meta_size + aligned_total_size; 3331 } 3332 3333 static bool validate_cluster_size(size_t cluster_size, bool extended_l2, 3334 Error **errp) 3335 { 3336 int cluster_bits = ctz32(cluster_size); 3337 if (cluster_bits < MIN_CLUSTER_BITS || cluster_bits > MAX_CLUSTER_BITS || 3338 (1 << cluster_bits) != cluster_size) 3339 { 3340 error_setg(errp, "Cluster size must be a power of two between %d and " 3341 "%dk", 1 << MIN_CLUSTER_BITS, 1 << (MAX_CLUSTER_BITS - 10)); 3342 return false; 3343 } 3344 3345 if (extended_l2) { 3346 unsigned min_cluster_size = 3347 (1 << MIN_CLUSTER_BITS) * QCOW_EXTL2_SUBCLUSTERS_PER_CLUSTER; 3348 if (cluster_size < min_cluster_size) { 3349 error_setg(errp, "Extended L2 entries are only supported with " 3350 "cluster sizes of at least %u bytes", min_cluster_size); 3351 return false; 3352 } 3353 } 3354 3355 return true; 3356 } 3357 3358 static size_t qcow2_opt_get_cluster_size_del(QemuOpts *opts, bool extended_l2, 3359 Error **errp) 3360 { 3361 size_t cluster_size; 3362 3363 cluster_size = qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE, 3364 DEFAULT_CLUSTER_SIZE); 3365 if (!validate_cluster_size(cluster_size, extended_l2, errp)) { 3366 return 0; 3367 } 3368 return cluster_size; 3369 } 3370 3371 static int qcow2_opt_get_version_del(QemuOpts *opts, Error **errp) 3372 { 3373 char *buf; 3374 int ret; 3375 3376 buf = qemu_opt_get_del(opts, BLOCK_OPT_COMPAT_LEVEL); 3377 if (!buf) { 3378 ret = 3; /* default */ 3379 } else if (!strcmp(buf, "0.10")) { 3380 ret = 2; 3381 } else if (!strcmp(buf, "1.1")) { 3382 ret = 3; 3383 } else { 3384 error_setg(errp, "Invalid compatibility level: '%s'", buf); 3385 ret = -EINVAL; 3386 } 3387 g_free(buf); 3388 return ret; 3389 } 3390 3391 static uint64_t qcow2_opt_get_refcount_bits_del(QemuOpts *opts, int version, 3392 Error **errp) 3393 { 3394 uint64_t refcount_bits; 3395 3396 refcount_bits = qemu_opt_get_number_del(opts, BLOCK_OPT_REFCOUNT_BITS, 16); 3397 if (refcount_bits > 64 || !is_power_of_2(refcount_bits)) { 3398 error_setg(errp, "Refcount width must be a power of two and may not " 3399 "exceed 64 bits"); 3400 return 0; 3401 } 3402 3403 if (version < 3 && refcount_bits != 16) { 3404 error_setg(errp, "Different refcount widths than 16 bits require " 3405 "compatibility level 1.1 or above (use compat=1.1 or " 3406 "greater)"); 3407 return 0; 3408 } 3409 3410 return refcount_bits; 3411 } 3412 3413 static int coroutine_fn 3414 qcow2_co_create(BlockdevCreateOptions *create_options, Error **errp) 3415 { 3416 BlockdevCreateOptionsQcow2 *qcow2_opts; 3417 QDict *options; 3418 3419 /* 3420 * Open the image file and write a minimal qcow2 header. 3421 * 3422 * We keep things simple and start with a zero-sized image. We also 3423 * do without refcount blocks or a L1 table for now. We'll fix the 3424 * inconsistency later. 3425 * 3426 * We do need a refcount table because growing the refcount table means 3427 * allocating two new refcount blocks - the second of which would be at 3428 * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file 3429 * size for any qcow2 image. 3430 */ 3431 BlockBackend *blk = NULL; 3432 BlockDriverState *bs = NULL; 3433 BlockDriverState *data_bs = NULL; 3434 QCowHeader *header; 3435 size_t cluster_size; 3436 int version; 3437 int refcount_order; 3438 uint64_t *refcount_table; 3439 int ret; 3440 uint8_t compression_type = QCOW2_COMPRESSION_TYPE_ZLIB; 3441 3442 assert(create_options->driver == BLOCKDEV_DRIVER_QCOW2); 3443 qcow2_opts = &create_options->u.qcow2; 3444 3445 bs = bdrv_open_blockdev_ref(qcow2_opts->file, errp); 3446 if (bs == NULL) { 3447 return -EIO; 3448 } 3449 3450 /* Validate options and set default values */ 3451 if (!QEMU_IS_ALIGNED(qcow2_opts->size, BDRV_SECTOR_SIZE)) { 3452 error_setg(errp, "Image size must be a multiple of %u bytes", 3453 (unsigned) BDRV_SECTOR_SIZE); 3454 ret = -EINVAL; 3455 goto out; 3456 } 3457 3458 if (qcow2_opts->has_version) { 3459 switch (qcow2_opts->version) { 3460 case BLOCKDEV_QCOW2_VERSION_V2: 3461 version = 2; 3462 break; 3463 case BLOCKDEV_QCOW2_VERSION_V3: 3464 version = 3; 3465 break; 3466 default: 3467 g_assert_not_reached(); 3468 } 3469 } else { 3470 version = 3; 3471 } 3472 3473 if (qcow2_opts->has_cluster_size) { 3474 cluster_size = qcow2_opts->cluster_size; 3475 } else { 3476 cluster_size = DEFAULT_CLUSTER_SIZE; 3477 } 3478 3479 if (!qcow2_opts->has_extended_l2) { 3480 qcow2_opts->extended_l2 = false; 3481 } 3482 if (qcow2_opts->extended_l2) { 3483 if (version < 3) { 3484 error_setg(errp, "Extended L2 entries are only supported with " 3485 "compatibility level 1.1 and above (use version=v3 or " 3486 "greater)"); 3487 ret = -EINVAL; 3488 goto out; 3489 } 3490 } 3491 3492 if (!validate_cluster_size(cluster_size, qcow2_opts->extended_l2, errp)) { 3493 ret = -EINVAL; 3494 goto out; 3495 } 3496 3497 if (!qcow2_opts->has_preallocation) { 3498 qcow2_opts->preallocation = PREALLOC_MODE_OFF; 3499 } 3500 if (qcow2_opts->has_backing_file && 3501 qcow2_opts->preallocation != PREALLOC_MODE_OFF && 3502 !qcow2_opts->extended_l2) 3503 { 3504 error_setg(errp, "Backing file and preallocation can only be used at " 3505 "the same time if extended_l2 is on"); 3506 ret = -EINVAL; 3507 goto out; 3508 } 3509 if (qcow2_opts->has_backing_fmt && !qcow2_opts->has_backing_file) { 3510 error_setg(errp, "Backing format cannot be used without backing file"); 3511 ret = -EINVAL; 3512 goto out; 3513 } 3514 3515 if (!qcow2_opts->has_lazy_refcounts) { 3516 qcow2_opts->lazy_refcounts = false; 3517 } 3518 if (version < 3 && qcow2_opts->lazy_refcounts) { 3519 error_setg(errp, "Lazy refcounts only supported with compatibility " 3520 "level 1.1 and above (use version=v3 or greater)"); 3521 ret = -EINVAL; 3522 goto out; 3523 } 3524 3525 if (!qcow2_opts->has_refcount_bits) { 3526 qcow2_opts->refcount_bits = 16; 3527 } 3528 if (qcow2_opts->refcount_bits > 64 || 3529 !is_power_of_2(qcow2_opts->refcount_bits)) 3530 { 3531 error_setg(errp, "Refcount width must be a power of two and may not " 3532 "exceed 64 bits"); 3533 ret = -EINVAL; 3534 goto out; 3535 } 3536 if (version < 3 && qcow2_opts->refcount_bits != 16) { 3537 error_setg(errp, "Different refcount widths than 16 bits require " 3538 "compatibility level 1.1 or above (use version=v3 or " 3539 "greater)"); 3540 ret = -EINVAL; 3541 goto out; 3542 } 3543 refcount_order = ctz32(qcow2_opts->refcount_bits); 3544 3545 if (qcow2_opts->data_file_raw && !qcow2_opts->data_file) { 3546 error_setg(errp, "data-file-raw requires data-file"); 3547 ret = -EINVAL; 3548 goto out; 3549 } 3550 if (qcow2_opts->data_file_raw && qcow2_opts->has_backing_file) { 3551 error_setg(errp, "Backing file and data-file-raw cannot be used at " 3552 "the same time"); 3553 ret = -EINVAL; 3554 goto out; 3555 } 3556 if (qcow2_opts->data_file_raw && 3557 qcow2_opts->preallocation == PREALLOC_MODE_OFF) 3558 { 3559 /* 3560 * data-file-raw means that "the external data file can be 3561 * read as a consistent standalone raw image without looking 3562 * at the qcow2 metadata." It does not say that the metadata 3563 * must be ignored, though (and the qcow2 driver in fact does 3564 * not ignore it), so the L1/L2 tables must be present and 3565 * give a 1:1 mapping, so you get the same result regardless 3566 * of whether you look at the metadata or whether you ignore 3567 * it. 3568 */ 3569 qcow2_opts->preallocation = PREALLOC_MODE_METADATA; 3570 3571 /* 3572 * Cannot use preallocation with backing files, but giving a 3573 * backing file when specifying data_file_raw is an error 3574 * anyway. 3575 */ 3576 assert(!qcow2_opts->has_backing_file); 3577 } 3578 3579 if (qcow2_opts->data_file) { 3580 if (version < 3) { 3581 error_setg(errp, "External data files are only supported with " 3582 "compatibility level 1.1 and above (use version=v3 or " 3583 "greater)"); 3584 ret = -EINVAL; 3585 goto out; 3586 } 3587 data_bs = bdrv_open_blockdev_ref(qcow2_opts->data_file, errp); 3588 if (data_bs == NULL) { 3589 ret = -EIO; 3590 goto out; 3591 } 3592 } 3593 3594 if (qcow2_opts->has_compression_type && 3595 qcow2_opts->compression_type != QCOW2_COMPRESSION_TYPE_ZLIB) { 3596 3597 ret = -EINVAL; 3598 3599 if (version < 3) { 3600 error_setg(errp, "Non-zlib compression type is only supported with " 3601 "compatibility level 1.1 and above (use version=v3 or " 3602 "greater)"); 3603 goto out; 3604 } 3605 3606 switch (qcow2_opts->compression_type) { 3607 #ifdef CONFIG_ZSTD 3608 case QCOW2_COMPRESSION_TYPE_ZSTD: 3609 break; 3610 #endif 3611 default: 3612 error_setg(errp, "Unknown compression type"); 3613 goto out; 3614 } 3615 3616 compression_type = qcow2_opts->compression_type; 3617 } 3618 3619 /* Create BlockBackend to write to the image */ 3620 blk = blk_new_with_bs(bs, BLK_PERM_WRITE | BLK_PERM_RESIZE, BLK_PERM_ALL, 3621 errp); 3622 if (!blk) { 3623 ret = -EPERM; 3624 goto out; 3625 } 3626 blk_set_allow_write_beyond_eof(blk, true); 3627 3628 /* Write the header */ 3629 QEMU_BUILD_BUG_ON((1 << MIN_CLUSTER_BITS) < sizeof(*header)); 3630 header = g_malloc0(cluster_size); 3631 *header = (QCowHeader) { 3632 .magic = cpu_to_be32(QCOW_MAGIC), 3633 .version = cpu_to_be32(version), 3634 .cluster_bits = cpu_to_be32(ctz32(cluster_size)), 3635 .size = cpu_to_be64(0), 3636 .l1_table_offset = cpu_to_be64(0), 3637 .l1_size = cpu_to_be32(0), 3638 .refcount_table_offset = cpu_to_be64(cluster_size), 3639 .refcount_table_clusters = cpu_to_be32(1), 3640 .refcount_order = cpu_to_be32(refcount_order), 3641 /* don't deal with endianness since compression_type is 1 byte long */ 3642 .compression_type = compression_type, 3643 .header_length = cpu_to_be32(sizeof(*header)), 3644 }; 3645 3646 /* We'll update this to correct value later */ 3647 header->crypt_method = cpu_to_be32(QCOW_CRYPT_NONE); 3648 3649 if (qcow2_opts->lazy_refcounts) { 3650 header->compatible_features |= 3651 cpu_to_be64(QCOW2_COMPAT_LAZY_REFCOUNTS); 3652 } 3653 if (data_bs) { 3654 header->incompatible_features |= 3655 cpu_to_be64(QCOW2_INCOMPAT_DATA_FILE); 3656 } 3657 if (qcow2_opts->data_file_raw) { 3658 header->autoclear_features |= 3659 cpu_to_be64(QCOW2_AUTOCLEAR_DATA_FILE_RAW); 3660 } 3661 if (compression_type != QCOW2_COMPRESSION_TYPE_ZLIB) { 3662 header->incompatible_features |= 3663 cpu_to_be64(QCOW2_INCOMPAT_COMPRESSION); 3664 } 3665 3666 if (qcow2_opts->extended_l2) { 3667 header->incompatible_features |= 3668 cpu_to_be64(QCOW2_INCOMPAT_EXTL2); 3669 } 3670 3671 ret = blk_pwrite(blk, 0, header, cluster_size, 0); 3672 g_free(header); 3673 if (ret < 0) { 3674 error_setg_errno(errp, -ret, "Could not write qcow2 header"); 3675 goto out; 3676 } 3677 3678 /* Write a refcount table with one refcount block */ 3679 refcount_table = g_malloc0(2 * cluster_size); 3680 refcount_table[0] = cpu_to_be64(2 * cluster_size); 3681 ret = blk_pwrite(blk, cluster_size, refcount_table, 2 * cluster_size, 0); 3682 g_free(refcount_table); 3683 3684 if (ret < 0) { 3685 error_setg_errno(errp, -ret, "Could not write refcount table"); 3686 goto out; 3687 } 3688 3689 blk_unref(blk); 3690 blk = NULL; 3691 3692 /* 3693 * And now open the image and make it consistent first (i.e. increase the 3694 * refcount of the cluster that is occupied by the header and the refcount 3695 * table) 3696 */ 3697 options = qdict_new(); 3698 qdict_put_str(options, "driver", "qcow2"); 3699 qdict_put_str(options, "file", bs->node_name); 3700 if (data_bs) { 3701 qdict_put_str(options, "data-file", data_bs->node_name); 3702 } 3703 blk = blk_new_open(NULL, NULL, options, 3704 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_NO_FLUSH, 3705 errp); 3706 if (blk == NULL) { 3707 ret = -EIO; 3708 goto out; 3709 } 3710 3711 ret = qcow2_alloc_clusters(blk_bs(blk), 3 * cluster_size); 3712 if (ret < 0) { 3713 error_setg_errno(errp, -ret, "Could not allocate clusters for qcow2 " 3714 "header and refcount table"); 3715 goto out; 3716 3717 } else if (ret != 0) { 3718 error_report("Huh, first cluster in empty image is already in use?"); 3719 abort(); 3720 } 3721 3722 /* Set the external data file if necessary */ 3723 if (data_bs) { 3724 BDRVQcow2State *s = blk_bs(blk)->opaque; 3725 s->image_data_file = g_strdup(data_bs->filename); 3726 } 3727 3728 /* Create a full header (including things like feature table) */ 3729 ret = qcow2_update_header(blk_bs(blk)); 3730 if (ret < 0) { 3731 error_setg_errno(errp, -ret, "Could not update qcow2 header"); 3732 goto out; 3733 } 3734 3735 /* Okay, now that we have a valid image, let's give it the right size */ 3736 ret = blk_truncate(blk, qcow2_opts->size, false, qcow2_opts->preallocation, 3737 0, errp); 3738 if (ret < 0) { 3739 error_prepend(errp, "Could not resize image: "); 3740 goto out; 3741 } 3742 3743 /* Want a backing file? There you go. */ 3744 if (qcow2_opts->has_backing_file) { 3745 const char *backing_format = NULL; 3746 3747 if (qcow2_opts->has_backing_fmt) { 3748 backing_format = BlockdevDriver_str(qcow2_opts->backing_fmt); 3749 } 3750 3751 ret = bdrv_change_backing_file(blk_bs(blk), qcow2_opts->backing_file, 3752 backing_format, false); 3753 if (ret < 0) { 3754 error_setg_errno(errp, -ret, "Could not assign backing file '%s' " 3755 "with format '%s'", qcow2_opts->backing_file, 3756 backing_format); 3757 goto out; 3758 } 3759 } 3760 3761 /* Want encryption? There you go. */ 3762 if (qcow2_opts->has_encrypt) { 3763 ret = qcow2_set_up_encryption(blk_bs(blk), qcow2_opts->encrypt, errp); 3764 if (ret < 0) { 3765 goto out; 3766 } 3767 } 3768 3769 blk_unref(blk); 3770 blk = NULL; 3771 3772 /* Reopen the image without BDRV_O_NO_FLUSH to flush it before returning. 3773 * Using BDRV_O_NO_IO, since encryption is now setup we don't want to 3774 * have to setup decryption context. We're not doing any I/O on the top 3775 * level BlockDriverState, only lower layers, where BDRV_O_NO_IO does 3776 * not have effect. 3777 */ 3778 options = qdict_new(); 3779 qdict_put_str(options, "driver", "qcow2"); 3780 qdict_put_str(options, "file", bs->node_name); 3781 if (data_bs) { 3782 qdict_put_str(options, "data-file", data_bs->node_name); 3783 } 3784 blk = blk_new_open(NULL, NULL, options, 3785 BDRV_O_RDWR | BDRV_O_NO_BACKING | BDRV_O_NO_IO, 3786 errp); 3787 if (blk == NULL) { 3788 ret = -EIO; 3789 goto out; 3790 } 3791 3792 ret = 0; 3793 out: 3794 blk_unref(blk); 3795 bdrv_unref(bs); 3796 bdrv_unref(data_bs); 3797 return ret; 3798 } 3799 3800 static int coroutine_fn qcow2_co_create_opts(BlockDriver *drv, 3801 const char *filename, 3802 QemuOpts *opts, 3803 Error **errp) 3804 { 3805 BlockdevCreateOptions *create_options = NULL; 3806 QDict *qdict; 3807 Visitor *v; 3808 BlockDriverState *bs = NULL; 3809 BlockDriverState *data_bs = NULL; 3810 const char *val; 3811 int ret; 3812 3813 /* Only the keyval visitor supports the dotted syntax needed for 3814 * encryption, so go through a QDict before getting a QAPI type. Ignore 3815 * options meant for the protocol layer so that the visitor doesn't 3816 * complain. */ 3817 qdict = qemu_opts_to_qdict_filtered(opts, NULL, bdrv_qcow2.create_opts, 3818 true); 3819 3820 /* Handle encryption options */ 3821 val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT); 3822 if (val && !strcmp(val, "on")) { 3823 qdict_put_str(qdict, BLOCK_OPT_ENCRYPT, "qcow"); 3824 } else if (val && !strcmp(val, "off")) { 3825 qdict_del(qdict, BLOCK_OPT_ENCRYPT); 3826 } 3827 3828 val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT); 3829 if (val && !strcmp(val, "aes")) { 3830 qdict_put_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT, "qcow"); 3831 } 3832 3833 /* Convert compat=0.10/1.1 into compat=v2/v3, to be renamed into 3834 * version=v2/v3 below. */ 3835 val = qdict_get_try_str(qdict, BLOCK_OPT_COMPAT_LEVEL); 3836 if (val && !strcmp(val, "0.10")) { 3837 qdict_put_str(qdict, BLOCK_OPT_COMPAT_LEVEL, "v2"); 3838 } else if (val && !strcmp(val, "1.1")) { 3839 qdict_put_str(qdict, BLOCK_OPT_COMPAT_LEVEL, "v3"); 3840 } 3841 3842 /* Change legacy command line options into QMP ones */ 3843 static const QDictRenames opt_renames[] = { 3844 { BLOCK_OPT_BACKING_FILE, "backing-file" }, 3845 { BLOCK_OPT_BACKING_FMT, "backing-fmt" }, 3846 { BLOCK_OPT_CLUSTER_SIZE, "cluster-size" }, 3847 { BLOCK_OPT_LAZY_REFCOUNTS, "lazy-refcounts" }, 3848 { BLOCK_OPT_EXTL2, "extended-l2" }, 3849 { BLOCK_OPT_REFCOUNT_BITS, "refcount-bits" }, 3850 { BLOCK_OPT_ENCRYPT, BLOCK_OPT_ENCRYPT_FORMAT }, 3851 { BLOCK_OPT_COMPAT_LEVEL, "version" }, 3852 { BLOCK_OPT_DATA_FILE_RAW, "data-file-raw" }, 3853 { BLOCK_OPT_COMPRESSION_TYPE, "compression-type" }, 3854 { NULL, NULL }, 3855 }; 3856 3857 if (!qdict_rename_keys(qdict, opt_renames, errp)) { 3858 ret = -EINVAL; 3859 goto finish; 3860 } 3861 3862 /* Create and open the file (protocol layer) */ 3863 ret = bdrv_create_file(filename, opts, errp); 3864 if (ret < 0) { 3865 goto finish; 3866 } 3867 3868 bs = bdrv_open(filename, NULL, NULL, 3869 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp); 3870 if (bs == NULL) { 3871 ret = -EIO; 3872 goto finish; 3873 } 3874 3875 /* Create and open an external data file (protocol layer) */ 3876 val = qdict_get_try_str(qdict, BLOCK_OPT_DATA_FILE); 3877 if (val) { 3878 ret = bdrv_create_file(val, opts, errp); 3879 if (ret < 0) { 3880 goto finish; 3881 } 3882 3883 data_bs = bdrv_open(val, NULL, NULL, 3884 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, 3885 errp); 3886 if (data_bs == NULL) { 3887 ret = -EIO; 3888 goto finish; 3889 } 3890 3891 qdict_del(qdict, BLOCK_OPT_DATA_FILE); 3892 qdict_put_str(qdict, "data-file", data_bs->node_name); 3893 } 3894 3895 /* Set 'driver' and 'node' options */ 3896 qdict_put_str(qdict, "driver", "qcow2"); 3897 qdict_put_str(qdict, "file", bs->node_name); 3898 3899 /* Now get the QAPI type BlockdevCreateOptions */ 3900 v = qobject_input_visitor_new_flat_confused(qdict, errp); 3901 if (!v) { 3902 ret = -EINVAL; 3903 goto finish; 3904 } 3905 3906 visit_type_BlockdevCreateOptions(v, NULL, &create_options, errp); 3907 visit_free(v); 3908 if (!create_options) { 3909 ret = -EINVAL; 3910 goto finish; 3911 } 3912 3913 /* Silently round up size */ 3914 create_options->u.qcow2.size = ROUND_UP(create_options->u.qcow2.size, 3915 BDRV_SECTOR_SIZE); 3916 3917 /* Create the qcow2 image (format layer) */ 3918 ret = qcow2_co_create(create_options, errp); 3919 finish: 3920 if (ret < 0) { 3921 bdrv_co_delete_file_noerr(bs); 3922 bdrv_co_delete_file_noerr(data_bs); 3923 } else { 3924 ret = 0; 3925 } 3926 3927 qobject_unref(qdict); 3928 bdrv_unref(bs); 3929 bdrv_unref(data_bs); 3930 qapi_free_BlockdevCreateOptions(create_options); 3931 return ret; 3932 } 3933 3934 3935 static bool is_zero(BlockDriverState *bs, int64_t offset, int64_t bytes) 3936 { 3937 int64_t nr; 3938 int res; 3939 3940 /* Clamp to image length, before checking status of underlying sectors */ 3941 if (offset + bytes > bs->total_sectors * BDRV_SECTOR_SIZE) { 3942 bytes = bs->total_sectors * BDRV_SECTOR_SIZE - offset; 3943 } 3944 3945 if (!bytes) { 3946 return true; 3947 } 3948 3949 /* 3950 * bdrv_block_status_above doesn't merge different types of zeros, for 3951 * example, zeros which come from the region which is unallocated in 3952 * the whole backing chain, and zeros which come because of a short 3953 * backing file. So, we need a loop. 3954 */ 3955 do { 3956 res = bdrv_block_status_above(bs, NULL, offset, bytes, &nr, NULL, NULL); 3957 offset += nr; 3958 bytes -= nr; 3959 } while (res >= 0 && (res & BDRV_BLOCK_ZERO) && nr && bytes); 3960 3961 return res >= 0 && (res & BDRV_BLOCK_ZERO) && bytes == 0; 3962 } 3963 3964 static coroutine_fn int qcow2_co_pwrite_zeroes(BlockDriverState *bs, 3965 int64_t offset, int64_t bytes, BdrvRequestFlags flags) 3966 { 3967 int ret; 3968 BDRVQcow2State *s = bs->opaque; 3969 3970 uint32_t head = offset_into_subcluster(s, offset); 3971 uint32_t tail = ROUND_UP(offset + bytes, s->subcluster_size) - 3972 (offset + bytes); 3973 3974 trace_qcow2_pwrite_zeroes_start_req(qemu_coroutine_self(), offset, bytes); 3975 if (offset + bytes == bs->total_sectors * BDRV_SECTOR_SIZE) { 3976 tail = 0; 3977 } 3978 3979 if (head || tail) { 3980 uint64_t off; 3981 unsigned int nr; 3982 QCow2SubclusterType type; 3983 3984 assert(head + bytes + tail <= s->subcluster_size); 3985 3986 /* check whether remainder of cluster already reads as zero */ 3987 if (!(is_zero(bs, offset - head, head) && 3988 is_zero(bs, offset + bytes, tail))) { 3989 return -ENOTSUP; 3990 } 3991 3992 qemu_co_mutex_lock(&s->lock); 3993 /* We can have new write after previous check */ 3994 offset -= head; 3995 bytes = s->subcluster_size; 3996 nr = s->subcluster_size; 3997 ret = qcow2_get_host_offset(bs, offset, &nr, &off, &type); 3998 if (ret < 0 || 3999 (type != QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN && 4000 type != QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC && 4001 type != QCOW2_SUBCLUSTER_ZERO_PLAIN && 4002 type != QCOW2_SUBCLUSTER_ZERO_ALLOC)) { 4003 qemu_co_mutex_unlock(&s->lock); 4004 return ret < 0 ? ret : -ENOTSUP; 4005 } 4006 } else { 4007 qemu_co_mutex_lock(&s->lock); 4008 } 4009 4010 trace_qcow2_pwrite_zeroes(qemu_coroutine_self(), offset, bytes); 4011 4012 /* Whatever is left can use real zero subclusters */ 4013 ret = qcow2_subcluster_zeroize(bs, offset, bytes, flags); 4014 qemu_co_mutex_unlock(&s->lock); 4015 4016 return ret; 4017 } 4018 4019 static coroutine_fn int qcow2_co_pdiscard(BlockDriverState *bs, 4020 int64_t offset, int64_t bytes) 4021 { 4022 int ret; 4023 BDRVQcow2State *s = bs->opaque; 4024 4025 /* If the image does not support QCOW_OFLAG_ZERO then discarding 4026 * clusters could expose stale data from the backing file. */ 4027 if (s->qcow_version < 3 && bs->backing) { 4028 return -ENOTSUP; 4029 } 4030 4031 if (!QEMU_IS_ALIGNED(offset | bytes, s->cluster_size)) { 4032 assert(bytes < s->cluster_size); 4033 /* Ignore partial clusters, except for the special case of the 4034 * complete partial cluster at the end of an unaligned file */ 4035 if (!QEMU_IS_ALIGNED(offset, s->cluster_size) || 4036 offset + bytes != bs->total_sectors * BDRV_SECTOR_SIZE) { 4037 return -ENOTSUP; 4038 } 4039 } 4040 4041 qemu_co_mutex_lock(&s->lock); 4042 ret = qcow2_cluster_discard(bs, offset, bytes, QCOW2_DISCARD_REQUEST, 4043 false); 4044 qemu_co_mutex_unlock(&s->lock); 4045 return ret; 4046 } 4047 4048 static int coroutine_fn 4049 qcow2_co_copy_range_from(BlockDriverState *bs, 4050 BdrvChild *src, int64_t src_offset, 4051 BdrvChild *dst, int64_t dst_offset, 4052 int64_t bytes, BdrvRequestFlags read_flags, 4053 BdrvRequestFlags write_flags) 4054 { 4055 BDRVQcow2State *s = bs->opaque; 4056 int ret; 4057 unsigned int cur_bytes; /* number of bytes in current iteration */ 4058 BdrvChild *child = NULL; 4059 BdrvRequestFlags cur_write_flags; 4060 4061 assert(!bs->encrypted); 4062 qemu_co_mutex_lock(&s->lock); 4063 4064 while (bytes != 0) { 4065 uint64_t copy_offset = 0; 4066 QCow2SubclusterType type; 4067 /* prepare next request */ 4068 cur_bytes = MIN(bytes, INT_MAX); 4069 cur_write_flags = write_flags; 4070 4071 ret = qcow2_get_host_offset(bs, src_offset, &cur_bytes, 4072 ©_offset, &type); 4073 if (ret < 0) { 4074 goto out; 4075 } 4076 4077 switch (type) { 4078 case QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN: 4079 case QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC: 4080 if (bs->backing && bs->backing->bs) { 4081 int64_t backing_length = bdrv_getlength(bs->backing->bs); 4082 if (src_offset >= backing_length) { 4083 cur_write_flags |= BDRV_REQ_ZERO_WRITE; 4084 } else { 4085 child = bs->backing; 4086 cur_bytes = MIN(cur_bytes, backing_length - src_offset); 4087 copy_offset = src_offset; 4088 } 4089 } else { 4090 cur_write_flags |= BDRV_REQ_ZERO_WRITE; 4091 } 4092 break; 4093 4094 case QCOW2_SUBCLUSTER_ZERO_PLAIN: 4095 case QCOW2_SUBCLUSTER_ZERO_ALLOC: 4096 cur_write_flags |= BDRV_REQ_ZERO_WRITE; 4097 break; 4098 4099 case QCOW2_SUBCLUSTER_COMPRESSED: 4100 ret = -ENOTSUP; 4101 goto out; 4102 4103 case QCOW2_SUBCLUSTER_NORMAL: 4104 child = s->data_file; 4105 break; 4106 4107 default: 4108 abort(); 4109 } 4110 qemu_co_mutex_unlock(&s->lock); 4111 ret = bdrv_co_copy_range_from(child, 4112 copy_offset, 4113 dst, dst_offset, 4114 cur_bytes, read_flags, cur_write_flags); 4115 qemu_co_mutex_lock(&s->lock); 4116 if (ret < 0) { 4117 goto out; 4118 } 4119 4120 bytes -= cur_bytes; 4121 src_offset += cur_bytes; 4122 dst_offset += cur_bytes; 4123 } 4124 ret = 0; 4125 4126 out: 4127 qemu_co_mutex_unlock(&s->lock); 4128 return ret; 4129 } 4130 4131 static int coroutine_fn 4132 qcow2_co_copy_range_to(BlockDriverState *bs, 4133 BdrvChild *src, int64_t src_offset, 4134 BdrvChild *dst, int64_t dst_offset, 4135 int64_t bytes, BdrvRequestFlags read_flags, 4136 BdrvRequestFlags write_flags) 4137 { 4138 BDRVQcow2State *s = bs->opaque; 4139 int ret; 4140 unsigned int cur_bytes; /* number of sectors in current iteration */ 4141 uint64_t host_offset; 4142 QCowL2Meta *l2meta = NULL; 4143 4144 assert(!bs->encrypted); 4145 4146 qemu_co_mutex_lock(&s->lock); 4147 4148 while (bytes != 0) { 4149 4150 l2meta = NULL; 4151 4152 cur_bytes = MIN(bytes, INT_MAX); 4153 4154 /* TODO: 4155 * If src->bs == dst->bs, we could simply copy by incrementing 4156 * the refcnt, without copying user data. 4157 * Or if src->bs == dst->bs->backing->bs, we could copy by discarding. */ 4158 ret = qcow2_alloc_host_offset(bs, dst_offset, &cur_bytes, 4159 &host_offset, &l2meta); 4160 if (ret < 0) { 4161 goto fail; 4162 } 4163 4164 ret = qcow2_pre_write_overlap_check(bs, 0, host_offset, cur_bytes, 4165 true); 4166 if (ret < 0) { 4167 goto fail; 4168 } 4169 4170 qemu_co_mutex_unlock(&s->lock); 4171 ret = bdrv_co_copy_range_to(src, src_offset, s->data_file, host_offset, 4172 cur_bytes, read_flags, write_flags); 4173 qemu_co_mutex_lock(&s->lock); 4174 if (ret < 0) { 4175 goto fail; 4176 } 4177 4178 ret = qcow2_handle_l2meta(bs, &l2meta, true); 4179 if (ret) { 4180 goto fail; 4181 } 4182 4183 bytes -= cur_bytes; 4184 src_offset += cur_bytes; 4185 dst_offset += cur_bytes; 4186 } 4187 ret = 0; 4188 4189 fail: 4190 qcow2_handle_l2meta(bs, &l2meta, false); 4191 4192 qemu_co_mutex_unlock(&s->lock); 4193 4194 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret); 4195 4196 return ret; 4197 } 4198 4199 static int coroutine_fn qcow2_co_truncate(BlockDriverState *bs, int64_t offset, 4200 bool exact, PreallocMode prealloc, 4201 BdrvRequestFlags flags, Error **errp) 4202 { 4203 BDRVQcow2State *s = bs->opaque; 4204 uint64_t old_length; 4205 int64_t new_l1_size; 4206 int ret; 4207 QDict *options; 4208 4209 if (prealloc != PREALLOC_MODE_OFF && prealloc != PREALLOC_MODE_METADATA && 4210 prealloc != PREALLOC_MODE_FALLOC && prealloc != PREALLOC_MODE_FULL) 4211 { 4212 error_setg(errp, "Unsupported preallocation mode '%s'", 4213 PreallocMode_str(prealloc)); 4214 return -ENOTSUP; 4215 } 4216 4217 if (!QEMU_IS_ALIGNED(offset, BDRV_SECTOR_SIZE)) { 4218 error_setg(errp, "The new size must be a multiple of %u", 4219 (unsigned) BDRV_SECTOR_SIZE); 4220 return -EINVAL; 4221 } 4222 4223 qemu_co_mutex_lock(&s->lock); 4224 4225 /* 4226 * Even though we store snapshot size for all images, it was not 4227 * required until v3, so it is not safe to proceed for v2. 4228 */ 4229 if (s->nb_snapshots && s->qcow_version < 3) { 4230 error_setg(errp, "Can't resize a v2 image which has snapshots"); 4231 ret = -ENOTSUP; 4232 goto fail; 4233 } 4234 4235 /* See qcow2-bitmap.c for which bitmap scenarios prevent a resize. */ 4236 if (qcow2_truncate_bitmaps_check(bs, errp)) { 4237 ret = -ENOTSUP; 4238 goto fail; 4239 } 4240 4241 old_length = bs->total_sectors * BDRV_SECTOR_SIZE; 4242 new_l1_size = size_to_l1(s, offset); 4243 4244 if (offset < old_length) { 4245 int64_t last_cluster, old_file_size; 4246 if (prealloc != PREALLOC_MODE_OFF) { 4247 error_setg(errp, 4248 "Preallocation can't be used for shrinking an image"); 4249 ret = -EINVAL; 4250 goto fail; 4251 } 4252 4253 ret = qcow2_cluster_discard(bs, ROUND_UP(offset, s->cluster_size), 4254 old_length - ROUND_UP(offset, 4255 s->cluster_size), 4256 QCOW2_DISCARD_ALWAYS, true); 4257 if (ret < 0) { 4258 error_setg_errno(errp, -ret, "Failed to discard cropped clusters"); 4259 goto fail; 4260 } 4261 4262 ret = qcow2_shrink_l1_table(bs, new_l1_size); 4263 if (ret < 0) { 4264 error_setg_errno(errp, -ret, 4265 "Failed to reduce the number of L2 tables"); 4266 goto fail; 4267 } 4268 4269 ret = qcow2_shrink_reftable(bs); 4270 if (ret < 0) { 4271 error_setg_errno(errp, -ret, 4272 "Failed to discard unused refblocks"); 4273 goto fail; 4274 } 4275 4276 old_file_size = bdrv_getlength(bs->file->bs); 4277 if (old_file_size < 0) { 4278 error_setg_errno(errp, -old_file_size, 4279 "Failed to inquire current file length"); 4280 ret = old_file_size; 4281 goto fail; 4282 } 4283 last_cluster = qcow2_get_last_cluster(bs, old_file_size); 4284 if (last_cluster < 0) { 4285 error_setg_errno(errp, -last_cluster, 4286 "Failed to find the last cluster"); 4287 ret = last_cluster; 4288 goto fail; 4289 } 4290 if ((last_cluster + 1) * s->cluster_size < old_file_size) { 4291 Error *local_err = NULL; 4292 4293 /* 4294 * Do not pass @exact here: It will not help the user if 4295 * we get an error here just because they wanted to shrink 4296 * their qcow2 image (on a block device) with qemu-img. 4297 * (And on the qcow2 layer, the @exact requirement is 4298 * always fulfilled, so there is no need to pass it on.) 4299 */ 4300 bdrv_co_truncate(bs->file, (last_cluster + 1) * s->cluster_size, 4301 false, PREALLOC_MODE_OFF, 0, &local_err); 4302 if (local_err) { 4303 warn_reportf_err(local_err, 4304 "Failed to truncate the tail of the image: "); 4305 } 4306 } 4307 } else { 4308 ret = qcow2_grow_l1_table(bs, new_l1_size, true); 4309 if (ret < 0) { 4310 error_setg_errno(errp, -ret, "Failed to grow the L1 table"); 4311 goto fail; 4312 } 4313 4314 if (data_file_is_raw(bs) && prealloc == PREALLOC_MODE_OFF) { 4315 /* 4316 * When creating a qcow2 image with data-file-raw, we enforce 4317 * at least prealloc=metadata, so that the L1/L2 tables are 4318 * fully allocated and reading from the data file will return 4319 * the same data as reading from the qcow2 image. When the 4320 * image is grown, we must consequently preallocate the 4321 * metadata structures to cover the added area. 4322 */ 4323 prealloc = PREALLOC_MODE_METADATA; 4324 } 4325 } 4326 4327 switch (prealloc) { 4328 case PREALLOC_MODE_OFF: 4329 if (has_data_file(bs)) { 4330 /* 4331 * If the caller wants an exact resize, the external data 4332 * file should be resized to the exact target size, too, 4333 * so we pass @exact here. 4334 */ 4335 ret = bdrv_co_truncate(s->data_file, offset, exact, prealloc, 0, 4336 errp); 4337 if (ret < 0) { 4338 goto fail; 4339 } 4340 } 4341 break; 4342 4343 case PREALLOC_MODE_METADATA: 4344 ret = preallocate_co(bs, old_length, offset, prealloc, errp); 4345 if (ret < 0) { 4346 goto fail; 4347 } 4348 break; 4349 4350 case PREALLOC_MODE_FALLOC: 4351 case PREALLOC_MODE_FULL: 4352 { 4353 int64_t allocation_start, host_offset, guest_offset; 4354 int64_t clusters_allocated; 4355 int64_t old_file_size, last_cluster, new_file_size; 4356 uint64_t nb_new_data_clusters, nb_new_l2_tables; 4357 bool subclusters_need_allocation = false; 4358 4359 /* With a data file, preallocation means just allocating the metadata 4360 * and forwarding the truncate request to the data file */ 4361 if (has_data_file(bs)) { 4362 ret = preallocate_co(bs, old_length, offset, prealloc, errp); 4363 if (ret < 0) { 4364 goto fail; 4365 } 4366 break; 4367 } 4368 4369 old_file_size = bdrv_getlength(bs->file->bs); 4370 if (old_file_size < 0) { 4371 error_setg_errno(errp, -old_file_size, 4372 "Failed to inquire current file length"); 4373 ret = old_file_size; 4374 goto fail; 4375 } 4376 4377 last_cluster = qcow2_get_last_cluster(bs, old_file_size); 4378 if (last_cluster >= 0) { 4379 old_file_size = (last_cluster + 1) * s->cluster_size; 4380 } else { 4381 old_file_size = ROUND_UP(old_file_size, s->cluster_size); 4382 } 4383 4384 nb_new_data_clusters = (ROUND_UP(offset, s->cluster_size) - 4385 start_of_cluster(s, old_length)) >> s->cluster_bits; 4386 4387 /* This is an overestimation; we will not actually allocate space for 4388 * these in the file but just make sure the new refcount structures are 4389 * able to cover them so we will not have to allocate new refblocks 4390 * while entering the data blocks in the potentially new L2 tables. 4391 * (We do not actually care where the L2 tables are placed. Maybe they 4392 * are already allocated or they can be placed somewhere before 4393 * @old_file_size. It does not matter because they will be fully 4394 * allocated automatically, so they do not need to be covered by the 4395 * preallocation. All that matters is that we will not have to allocate 4396 * new refcount structures for them.) */ 4397 nb_new_l2_tables = DIV_ROUND_UP(nb_new_data_clusters, 4398 s->cluster_size / l2_entry_size(s)); 4399 /* The cluster range may not be aligned to L2 boundaries, so add one L2 4400 * table for a potential head/tail */ 4401 nb_new_l2_tables++; 4402 4403 allocation_start = qcow2_refcount_area(bs, old_file_size, 4404 nb_new_data_clusters + 4405 nb_new_l2_tables, 4406 true, 0, 0); 4407 if (allocation_start < 0) { 4408 error_setg_errno(errp, -allocation_start, 4409 "Failed to resize refcount structures"); 4410 ret = allocation_start; 4411 goto fail; 4412 } 4413 4414 clusters_allocated = qcow2_alloc_clusters_at(bs, allocation_start, 4415 nb_new_data_clusters); 4416 if (clusters_allocated < 0) { 4417 error_setg_errno(errp, -clusters_allocated, 4418 "Failed to allocate data clusters"); 4419 ret = clusters_allocated; 4420 goto fail; 4421 } 4422 4423 assert(clusters_allocated == nb_new_data_clusters); 4424 4425 /* Allocate the data area */ 4426 new_file_size = allocation_start + 4427 nb_new_data_clusters * s->cluster_size; 4428 /* 4429 * Image file grows, so @exact does not matter. 4430 * 4431 * If we need to zero out the new area, try first whether the protocol 4432 * driver can already take care of this. 4433 */ 4434 if (flags & BDRV_REQ_ZERO_WRITE) { 4435 ret = bdrv_co_truncate(bs->file, new_file_size, false, prealloc, 4436 BDRV_REQ_ZERO_WRITE, NULL); 4437 if (ret >= 0) { 4438 flags &= ~BDRV_REQ_ZERO_WRITE; 4439 /* Ensure that we read zeroes and not backing file data */ 4440 subclusters_need_allocation = true; 4441 } 4442 } else { 4443 ret = -1; 4444 } 4445 if (ret < 0) { 4446 ret = bdrv_co_truncate(bs->file, new_file_size, false, prealloc, 0, 4447 errp); 4448 } 4449 if (ret < 0) { 4450 error_prepend(errp, "Failed to resize underlying file: "); 4451 qcow2_free_clusters(bs, allocation_start, 4452 nb_new_data_clusters * s->cluster_size, 4453 QCOW2_DISCARD_OTHER); 4454 goto fail; 4455 } 4456 4457 /* Create the necessary L2 entries */ 4458 host_offset = allocation_start; 4459 guest_offset = old_length; 4460 while (nb_new_data_clusters) { 4461 int64_t nb_clusters = MIN( 4462 nb_new_data_clusters, 4463 s->l2_slice_size - offset_to_l2_slice_index(s, guest_offset)); 4464 unsigned cow_start_length = offset_into_cluster(s, guest_offset); 4465 QCowL2Meta allocation; 4466 guest_offset = start_of_cluster(s, guest_offset); 4467 allocation = (QCowL2Meta) { 4468 .offset = guest_offset, 4469 .alloc_offset = host_offset, 4470 .nb_clusters = nb_clusters, 4471 .cow_start = { 4472 .offset = 0, 4473 .nb_bytes = cow_start_length, 4474 }, 4475 .cow_end = { 4476 .offset = nb_clusters << s->cluster_bits, 4477 .nb_bytes = 0, 4478 }, 4479 .prealloc = !subclusters_need_allocation, 4480 }; 4481 qemu_co_queue_init(&allocation.dependent_requests); 4482 4483 ret = qcow2_alloc_cluster_link_l2(bs, &allocation); 4484 if (ret < 0) { 4485 error_setg_errno(errp, -ret, "Failed to update L2 tables"); 4486 qcow2_free_clusters(bs, host_offset, 4487 nb_new_data_clusters * s->cluster_size, 4488 QCOW2_DISCARD_OTHER); 4489 goto fail; 4490 } 4491 4492 guest_offset += nb_clusters * s->cluster_size; 4493 host_offset += nb_clusters * s->cluster_size; 4494 nb_new_data_clusters -= nb_clusters; 4495 } 4496 break; 4497 } 4498 4499 default: 4500 g_assert_not_reached(); 4501 } 4502 4503 if ((flags & BDRV_REQ_ZERO_WRITE) && offset > old_length) { 4504 uint64_t zero_start = QEMU_ALIGN_UP(old_length, s->subcluster_size); 4505 4506 /* 4507 * Use zero clusters as much as we can. qcow2_subcluster_zeroize() 4508 * requires a subcluster-aligned start. The end may be unaligned if 4509 * it is at the end of the image (which it is here). 4510 */ 4511 if (offset > zero_start) { 4512 ret = qcow2_subcluster_zeroize(bs, zero_start, offset - zero_start, 4513 0); 4514 if (ret < 0) { 4515 error_setg_errno(errp, -ret, "Failed to zero out new clusters"); 4516 goto fail; 4517 } 4518 } 4519 4520 /* Write explicit zeros for the unaligned head */ 4521 if (zero_start > old_length) { 4522 uint64_t len = MIN(zero_start, offset) - old_length; 4523 uint8_t *buf = qemu_blockalign0(bs, len); 4524 QEMUIOVector qiov; 4525 qemu_iovec_init_buf(&qiov, buf, len); 4526 4527 qemu_co_mutex_unlock(&s->lock); 4528 ret = qcow2_co_pwritev_part(bs, old_length, len, &qiov, 0, 0); 4529 qemu_co_mutex_lock(&s->lock); 4530 4531 qemu_vfree(buf); 4532 if (ret < 0) { 4533 error_setg_errno(errp, -ret, "Failed to zero out the new area"); 4534 goto fail; 4535 } 4536 } 4537 } 4538 4539 if (prealloc != PREALLOC_MODE_OFF) { 4540 /* Flush metadata before actually changing the image size */ 4541 ret = qcow2_write_caches(bs); 4542 if (ret < 0) { 4543 error_setg_errno(errp, -ret, 4544 "Failed to flush the preallocated area to disk"); 4545 goto fail; 4546 } 4547 } 4548 4549 bs->total_sectors = offset / BDRV_SECTOR_SIZE; 4550 4551 /* write updated header.size */ 4552 offset = cpu_to_be64(offset); 4553 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, size), 4554 &offset, sizeof(offset)); 4555 if (ret < 0) { 4556 error_setg_errno(errp, -ret, "Failed to update the image size"); 4557 goto fail; 4558 } 4559 4560 s->l1_vm_state_index = new_l1_size; 4561 4562 /* Update cache sizes */ 4563 options = qdict_clone_shallow(bs->options); 4564 ret = qcow2_update_options(bs, options, s->flags, errp); 4565 qobject_unref(options); 4566 if (ret < 0) { 4567 goto fail; 4568 } 4569 ret = 0; 4570 fail: 4571 qemu_co_mutex_unlock(&s->lock); 4572 return ret; 4573 } 4574 4575 static coroutine_fn int 4576 qcow2_co_pwritev_compressed_task(BlockDriverState *bs, 4577 uint64_t offset, uint64_t bytes, 4578 QEMUIOVector *qiov, size_t qiov_offset) 4579 { 4580 BDRVQcow2State *s = bs->opaque; 4581 int ret; 4582 ssize_t out_len; 4583 uint8_t *buf, *out_buf; 4584 uint64_t cluster_offset; 4585 4586 assert(bytes == s->cluster_size || (bytes < s->cluster_size && 4587 (offset + bytes == bs->total_sectors << BDRV_SECTOR_BITS))); 4588 4589 buf = qemu_blockalign(bs, s->cluster_size); 4590 if (bytes < s->cluster_size) { 4591 /* Zero-pad last write if image size is not cluster aligned */ 4592 memset(buf + bytes, 0, s->cluster_size - bytes); 4593 } 4594 qemu_iovec_to_buf(qiov, qiov_offset, buf, bytes); 4595 4596 out_buf = g_malloc(s->cluster_size); 4597 4598 out_len = qcow2_co_compress(bs, out_buf, s->cluster_size - 1, 4599 buf, s->cluster_size); 4600 if (out_len == -ENOMEM) { 4601 /* could not compress: write normal cluster */ 4602 ret = qcow2_co_pwritev_part(bs, offset, bytes, qiov, qiov_offset, 0); 4603 if (ret < 0) { 4604 goto fail; 4605 } 4606 goto success; 4607 } else if (out_len < 0) { 4608 ret = -EINVAL; 4609 goto fail; 4610 } 4611 4612 qemu_co_mutex_lock(&s->lock); 4613 ret = qcow2_alloc_compressed_cluster_offset(bs, offset, out_len, 4614 &cluster_offset); 4615 if (ret < 0) { 4616 qemu_co_mutex_unlock(&s->lock); 4617 goto fail; 4618 } 4619 4620 ret = qcow2_pre_write_overlap_check(bs, 0, cluster_offset, out_len, true); 4621 qemu_co_mutex_unlock(&s->lock); 4622 if (ret < 0) { 4623 goto fail; 4624 } 4625 4626 BLKDBG_EVENT(s->data_file, BLKDBG_WRITE_COMPRESSED); 4627 ret = bdrv_co_pwrite(s->data_file, cluster_offset, out_len, out_buf, 0); 4628 if (ret < 0) { 4629 goto fail; 4630 } 4631 success: 4632 ret = 0; 4633 fail: 4634 qemu_vfree(buf); 4635 g_free(out_buf); 4636 return ret; 4637 } 4638 4639 static coroutine_fn int qcow2_co_pwritev_compressed_task_entry(AioTask *task) 4640 { 4641 Qcow2AioTask *t = container_of(task, Qcow2AioTask, task); 4642 4643 assert(!t->subcluster_type && !t->l2meta); 4644 4645 return qcow2_co_pwritev_compressed_task(t->bs, t->offset, t->bytes, t->qiov, 4646 t->qiov_offset); 4647 } 4648 4649 /* 4650 * XXX: put compressed sectors first, then all the cluster aligned 4651 * tables to avoid losing bytes in alignment 4652 */ 4653 static coroutine_fn int 4654 qcow2_co_pwritev_compressed_part(BlockDriverState *bs, 4655 int64_t offset, int64_t bytes, 4656 QEMUIOVector *qiov, size_t qiov_offset) 4657 { 4658 BDRVQcow2State *s = bs->opaque; 4659 AioTaskPool *aio = NULL; 4660 int ret = 0; 4661 4662 if (has_data_file(bs)) { 4663 return -ENOTSUP; 4664 } 4665 4666 if (bytes == 0) { 4667 /* 4668 * align end of file to a sector boundary to ease reading with 4669 * sector based I/Os 4670 */ 4671 int64_t len = bdrv_getlength(bs->file->bs); 4672 if (len < 0) { 4673 return len; 4674 } 4675 return bdrv_co_truncate(bs->file, len, false, PREALLOC_MODE_OFF, 0, 4676 NULL); 4677 } 4678 4679 if (offset_into_cluster(s, offset)) { 4680 return -EINVAL; 4681 } 4682 4683 if (offset_into_cluster(s, bytes) && 4684 (offset + bytes) != (bs->total_sectors << BDRV_SECTOR_BITS)) { 4685 return -EINVAL; 4686 } 4687 4688 while (bytes && aio_task_pool_status(aio) == 0) { 4689 uint64_t chunk_size = MIN(bytes, s->cluster_size); 4690 4691 if (!aio && chunk_size != bytes) { 4692 aio = aio_task_pool_new(QCOW2_MAX_WORKERS); 4693 } 4694 4695 ret = qcow2_add_task(bs, aio, qcow2_co_pwritev_compressed_task_entry, 4696 0, 0, offset, chunk_size, qiov, qiov_offset, NULL); 4697 if (ret < 0) { 4698 break; 4699 } 4700 qiov_offset += chunk_size; 4701 offset += chunk_size; 4702 bytes -= chunk_size; 4703 } 4704 4705 if (aio) { 4706 aio_task_pool_wait_all(aio); 4707 if (ret == 0) { 4708 ret = aio_task_pool_status(aio); 4709 } 4710 g_free(aio); 4711 } 4712 4713 return ret; 4714 } 4715 4716 static int coroutine_fn 4717 qcow2_co_preadv_compressed(BlockDriverState *bs, 4718 uint64_t l2_entry, 4719 uint64_t offset, 4720 uint64_t bytes, 4721 QEMUIOVector *qiov, 4722 size_t qiov_offset) 4723 { 4724 BDRVQcow2State *s = bs->opaque; 4725 int ret = 0, csize; 4726 uint64_t coffset; 4727 uint8_t *buf, *out_buf; 4728 int offset_in_cluster = offset_into_cluster(s, offset); 4729 4730 qcow2_parse_compressed_l2_entry(bs, l2_entry, &coffset, &csize); 4731 4732 buf = g_try_malloc(csize); 4733 if (!buf) { 4734 return -ENOMEM; 4735 } 4736 4737 out_buf = qemu_blockalign(bs, s->cluster_size); 4738 4739 BLKDBG_EVENT(bs->file, BLKDBG_READ_COMPRESSED); 4740 ret = bdrv_co_pread(bs->file, coffset, csize, buf, 0); 4741 if (ret < 0) { 4742 goto fail; 4743 } 4744 4745 if (qcow2_co_decompress(bs, out_buf, s->cluster_size, buf, csize) < 0) { 4746 ret = -EIO; 4747 goto fail; 4748 } 4749 4750 qemu_iovec_from_buf(qiov, qiov_offset, out_buf + offset_in_cluster, bytes); 4751 4752 fail: 4753 qemu_vfree(out_buf); 4754 g_free(buf); 4755 4756 return ret; 4757 } 4758 4759 static int make_completely_empty(BlockDriverState *bs) 4760 { 4761 BDRVQcow2State *s = bs->opaque; 4762 Error *local_err = NULL; 4763 int ret, l1_clusters; 4764 int64_t offset; 4765 uint64_t *new_reftable = NULL; 4766 uint64_t rt_entry, l1_size2; 4767 struct { 4768 uint64_t l1_offset; 4769 uint64_t reftable_offset; 4770 uint32_t reftable_clusters; 4771 } QEMU_PACKED l1_ofs_rt_ofs_cls; 4772 4773 ret = qcow2_cache_empty(bs, s->l2_table_cache); 4774 if (ret < 0) { 4775 goto fail; 4776 } 4777 4778 ret = qcow2_cache_empty(bs, s->refcount_block_cache); 4779 if (ret < 0) { 4780 goto fail; 4781 } 4782 4783 /* Refcounts will be broken utterly */ 4784 ret = qcow2_mark_dirty(bs); 4785 if (ret < 0) { 4786 goto fail; 4787 } 4788 4789 BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE); 4790 4791 l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / L1E_SIZE); 4792 l1_size2 = (uint64_t)s->l1_size * L1E_SIZE; 4793 4794 /* After this call, neither the in-memory nor the on-disk refcount 4795 * information accurately describe the actual references */ 4796 4797 ret = bdrv_pwrite_zeroes(bs->file, s->l1_table_offset, 4798 l1_clusters * s->cluster_size, 0); 4799 if (ret < 0) { 4800 goto fail_broken_refcounts; 4801 } 4802 memset(s->l1_table, 0, l1_size2); 4803 4804 BLKDBG_EVENT(bs->file, BLKDBG_EMPTY_IMAGE_PREPARE); 4805 4806 /* Overwrite enough clusters at the beginning of the sectors to place 4807 * the refcount table, a refcount block and the L1 table in; this may 4808 * overwrite parts of the existing refcount and L1 table, which is not 4809 * an issue because the dirty flag is set, complete data loss is in fact 4810 * desired and partial data loss is consequently fine as well */ 4811 ret = bdrv_pwrite_zeroes(bs->file, s->cluster_size, 4812 (2 + l1_clusters) * s->cluster_size, 0); 4813 /* This call (even if it failed overall) may have overwritten on-disk 4814 * refcount structures; in that case, the in-memory refcount information 4815 * will probably differ from the on-disk information which makes the BDS 4816 * unusable */ 4817 if (ret < 0) { 4818 goto fail_broken_refcounts; 4819 } 4820 4821 BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE); 4822 BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_UPDATE); 4823 4824 /* "Create" an empty reftable (one cluster) directly after the image 4825 * header and an empty L1 table three clusters after the image header; 4826 * the cluster between those two will be used as the first refblock */ 4827 l1_ofs_rt_ofs_cls.l1_offset = cpu_to_be64(3 * s->cluster_size); 4828 l1_ofs_rt_ofs_cls.reftable_offset = cpu_to_be64(s->cluster_size); 4829 l1_ofs_rt_ofs_cls.reftable_clusters = cpu_to_be32(1); 4830 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, l1_table_offset), 4831 &l1_ofs_rt_ofs_cls, sizeof(l1_ofs_rt_ofs_cls)); 4832 if (ret < 0) { 4833 goto fail_broken_refcounts; 4834 } 4835 4836 s->l1_table_offset = 3 * s->cluster_size; 4837 4838 new_reftable = g_try_new0(uint64_t, s->cluster_size / REFTABLE_ENTRY_SIZE); 4839 if (!new_reftable) { 4840 ret = -ENOMEM; 4841 goto fail_broken_refcounts; 4842 } 4843 4844 s->refcount_table_offset = s->cluster_size; 4845 s->refcount_table_size = s->cluster_size / REFTABLE_ENTRY_SIZE; 4846 s->max_refcount_table_index = 0; 4847 4848 g_free(s->refcount_table); 4849 s->refcount_table = new_reftable; 4850 new_reftable = NULL; 4851 4852 /* Now the in-memory refcount information again corresponds to the on-disk 4853 * information (reftable is empty and no refblocks (the refblock cache is 4854 * empty)); however, this means some clusters (e.g. the image header) are 4855 * referenced, but not refcounted, but the normal qcow2 code assumes that 4856 * the in-memory information is always correct */ 4857 4858 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC); 4859 4860 /* Enter the first refblock into the reftable */ 4861 rt_entry = cpu_to_be64(2 * s->cluster_size); 4862 ret = bdrv_pwrite_sync(bs->file, s->cluster_size, 4863 &rt_entry, sizeof(rt_entry)); 4864 if (ret < 0) { 4865 goto fail_broken_refcounts; 4866 } 4867 s->refcount_table[0] = 2 * s->cluster_size; 4868 4869 s->free_cluster_index = 0; 4870 assert(3 + l1_clusters <= s->refcount_block_size); 4871 offset = qcow2_alloc_clusters(bs, 3 * s->cluster_size + l1_size2); 4872 if (offset < 0) { 4873 ret = offset; 4874 goto fail_broken_refcounts; 4875 } else if (offset > 0) { 4876 error_report("First cluster in emptied image is in use"); 4877 abort(); 4878 } 4879 4880 /* Now finally the in-memory information corresponds to the on-disk 4881 * structures and is correct */ 4882 ret = qcow2_mark_clean(bs); 4883 if (ret < 0) { 4884 goto fail; 4885 } 4886 4887 ret = bdrv_truncate(bs->file, (3 + l1_clusters) * s->cluster_size, false, 4888 PREALLOC_MODE_OFF, 0, &local_err); 4889 if (ret < 0) { 4890 error_report_err(local_err); 4891 goto fail; 4892 } 4893 4894 return 0; 4895 4896 fail_broken_refcounts: 4897 /* The BDS is unusable at this point. If we wanted to make it usable, we 4898 * would have to call qcow2_refcount_close(), qcow2_refcount_init(), 4899 * qcow2_check_refcounts(), qcow2_refcount_close() and qcow2_refcount_init() 4900 * again. However, because the functions which could have caused this error 4901 * path to be taken are used by those functions as well, it's very likely 4902 * that that sequence will fail as well. Therefore, just eject the BDS. */ 4903 bs->drv = NULL; 4904 4905 fail: 4906 g_free(new_reftable); 4907 return ret; 4908 } 4909 4910 static int qcow2_make_empty(BlockDriverState *bs) 4911 { 4912 BDRVQcow2State *s = bs->opaque; 4913 uint64_t offset, end_offset; 4914 int step = QEMU_ALIGN_DOWN(INT_MAX, s->cluster_size); 4915 int l1_clusters, ret = 0; 4916 4917 l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / L1E_SIZE); 4918 4919 if (s->qcow_version >= 3 && !s->snapshots && !s->nb_bitmaps && 4920 3 + l1_clusters <= s->refcount_block_size && 4921 s->crypt_method_header != QCOW_CRYPT_LUKS && 4922 !has_data_file(bs)) { 4923 /* The following function only works for qcow2 v3 images (it 4924 * requires the dirty flag) and only as long as there are no 4925 * features that reserve extra clusters (such as snapshots, 4926 * LUKS header, or persistent bitmaps), because it completely 4927 * empties the image. Furthermore, the L1 table and three 4928 * additional clusters (image header, refcount table, one 4929 * refcount block) have to fit inside one refcount block. It 4930 * only resets the image file, i.e. does not work with an 4931 * external data file. */ 4932 return make_completely_empty(bs); 4933 } 4934 4935 /* This fallback code simply discards every active cluster; this is slow, 4936 * but works in all cases */ 4937 end_offset = bs->total_sectors * BDRV_SECTOR_SIZE; 4938 for (offset = 0; offset < end_offset; offset += step) { 4939 /* As this function is generally used after committing an external 4940 * snapshot, QCOW2_DISCARD_SNAPSHOT seems appropriate. Also, the 4941 * default action for this kind of discard is to pass the discard, 4942 * which will ideally result in an actually smaller image file, as 4943 * is probably desired. */ 4944 ret = qcow2_cluster_discard(bs, offset, MIN(step, end_offset - offset), 4945 QCOW2_DISCARD_SNAPSHOT, true); 4946 if (ret < 0) { 4947 break; 4948 } 4949 } 4950 4951 return ret; 4952 } 4953 4954 static coroutine_fn int qcow2_co_flush_to_os(BlockDriverState *bs) 4955 { 4956 BDRVQcow2State *s = bs->opaque; 4957 int ret; 4958 4959 qemu_co_mutex_lock(&s->lock); 4960 ret = qcow2_write_caches(bs); 4961 qemu_co_mutex_unlock(&s->lock); 4962 4963 return ret; 4964 } 4965 4966 static BlockMeasureInfo *qcow2_measure(QemuOpts *opts, BlockDriverState *in_bs, 4967 Error **errp) 4968 { 4969 Error *local_err = NULL; 4970 BlockMeasureInfo *info; 4971 uint64_t required = 0; /* bytes that contribute to required size */ 4972 uint64_t virtual_size; /* disk size as seen by guest */ 4973 uint64_t refcount_bits; 4974 uint64_t l2_tables; 4975 uint64_t luks_payload_size = 0; 4976 size_t cluster_size; 4977 int version; 4978 char *optstr; 4979 PreallocMode prealloc; 4980 bool has_backing_file; 4981 bool has_luks; 4982 bool extended_l2; 4983 size_t l2e_size; 4984 4985 /* Parse image creation options */ 4986 extended_l2 = qemu_opt_get_bool_del(opts, BLOCK_OPT_EXTL2, false); 4987 4988 cluster_size = qcow2_opt_get_cluster_size_del(opts, extended_l2, 4989 &local_err); 4990 if (local_err) { 4991 goto err; 4992 } 4993 4994 version = qcow2_opt_get_version_del(opts, &local_err); 4995 if (local_err) { 4996 goto err; 4997 } 4998 4999 refcount_bits = qcow2_opt_get_refcount_bits_del(opts, version, &local_err); 5000 if (local_err) { 5001 goto err; 5002 } 5003 5004 optstr = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC); 5005 prealloc = qapi_enum_parse(&PreallocMode_lookup, optstr, 5006 PREALLOC_MODE_OFF, &local_err); 5007 g_free(optstr); 5008 if (local_err) { 5009 goto err; 5010 } 5011 5012 optstr = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE); 5013 has_backing_file = !!optstr; 5014 g_free(optstr); 5015 5016 optstr = qemu_opt_get_del(opts, BLOCK_OPT_ENCRYPT_FORMAT); 5017 has_luks = optstr && strcmp(optstr, "luks") == 0; 5018 g_free(optstr); 5019 5020 if (has_luks) { 5021 g_autoptr(QCryptoBlockCreateOptions) create_opts = NULL; 5022 QDict *cryptoopts = qcow2_extract_crypto_opts(opts, "luks", errp); 5023 size_t headerlen; 5024 5025 create_opts = block_crypto_create_opts_init(cryptoopts, errp); 5026 qobject_unref(cryptoopts); 5027 if (!create_opts) { 5028 goto err; 5029 } 5030 5031 if (!qcrypto_block_calculate_payload_offset(create_opts, 5032 "encrypt.", 5033 &headerlen, 5034 &local_err)) { 5035 goto err; 5036 } 5037 5038 luks_payload_size = ROUND_UP(headerlen, cluster_size); 5039 } 5040 5041 virtual_size = qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0); 5042 virtual_size = ROUND_UP(virtual_size, cluster_size); 5043 5044 /* Check that virtual disk size is valid */ 5045 l2e_size = extended_l2 ? L2E_SIZE_EXTENDED : L2E_SIZE_NORMAL; 5046 l2_tables = DIV_ROUND_UP(virtual_size / cluster_size, 5047 cluster_size / l2e_size); 5048 if (l2_tables * L1E_SIZE > QCOW_MAX_L1_SIZE) { 5049 error_setg(&local_err, "The image size is too large " 5050 "(try using a larger cluster size)"); 5051 goto err; 5052 } 5053 5054 /* Account for input image */ 5055 if (in_bs) { 5056 int64_t ssize = bdrv_getlength(in_bs); 5057 if (ssize < 0) { 5058 error_setg_errno(&local_err, -ssize, 5059 "Unable to get image virtual_size"); 5060 goto err; 5061 } 5062 5063 virtual_size = ROUND_UP(ssize, cluster_size); 5064 5065 if (has_backing_file) { 5066 /* We don't how much of the backing chain is shared by the input 5067 * image and the new image file. In the worst case the new image's 5068 * backing file has nothing in common with the input image. Be 5069 * conservative and assume all clusters need to be written. 5070 */ 5071 required = virtual_size; 5072 } else { 5073 int64_t offset; 5074 int64_t pnum = 0; 5075 5076 for (offset = 0; offset < ssize; offset += pnum) { 5077 int ret; 5078 5079 ret = bdrv_block_status_above(in_bs, NULL, offset, 5080 ssize - offset, &pnum, NULL, 5081 NULL); 5082 if (ret < 0) { 5083 error_setg_errno(&local_err, -ret, 5084 "Unable to get block status"); 5085 goto err; 5086 } 5087 5088 if (ret & BDRV_BLOCK_ZERO) { 5089 /* Skip zero regions (safe with no backing file) */ 5090 } else if ((ret & (BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED)) == 5091 (BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED)) { 5092 /* Extend pnum to end of cluster for next iteration */ 5093 pnum = ROUND_UP(offset + pnum, cluster_size) - offset; 5094 5095 /* Count clusters we've seen */ 5096 required += offset % cluster_size + pnum; 5097 } 5098 } 5099 } 5100 } 5101 5102 /* Take into account preallocation. Nothing special is needed for 5103 * PREALLOC_MODE_METADATA since metadata is always counted. 5104 */ 5105 if (prealloc == PREALLOC_MODE_FULL || prealloc == PREALLOC_MODE_FALLOC) { 5106 required = virtual_size; 5107 } 5108 5109 info = g_new0(BlockMeasureInfo, 1); 5110 info->fully_allocated = luks_payload_size + 5111 qcow2_calc_prealloc_size(virtual_size, cluster_size, 5112 ctz32(refcount_bits), extended_l2); 5113 5114 /* 5115 * Remove data clusters that are not required. This overestimates the 5116 * required size because metadata needed for the fully allocated file is 5117 * still counted. Show bitmaps only if both source and destination 5118 * would support them. 5119 */ 5120 info->required = info->fully_allocated - virtual_size + required; 5121 info->has_bitmaps = version >= 3 && in_bs && 5122 bdrv_supports_persistent_dirty_bitmap(in_bs); 5123 if (info->has_bitmaps) { 5124 info->bitmaps = qcow2_get_persistent_dirty_bitmap_size(in_bs, 5125 cluster_size); 5126 } 5127 return info; 5128 5129 err: 5130 error_propagate(errp, local_err); 5131 return NULL; 5132 } 5133 5134 static int qcow2_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) 5135 { 5136 BDRVQcow2State *s = bs->opaque; 5137 bdi->cluster_size = s->cluster_size; 5138 bdi->vm_state_offset = qcow2_vm_state_offset(s); 5139 bdi->is_dirty = s->incompatible_features & QCOW2_INCOMPAT_DIRTY; 5140 return 0; 5141 } 5142 5143 static ImageInfoSpecific *qcow2_get_specific_info(BlockDriverState *bs, 5144 Error **errp) 5145 { 5146 BDRVQcow2State *s = bs->opaque; 5147 ImageInfoSpecific *spec_info; 5148 QCryptoBlockInfo *encrypt_info = NULL; 5149 5150 if (s->crypto != NULL) { 5151 encrypt_info = qcrypto_block_get_info(s->crypto, errp); 5152 if (!encrypt_info) { 5153 return NULL; 5154 } 5155 } 5156 5157 spec_info = g_new(ImageInfoSpecific, 1); 5158 *spec_info = (ImageInfoSpecific){ 5159 .type = IMAGE_INFO_SPECIFIC_KIND_QCOW2, 5160 .u.qcow2.data = g_new0(ImageInfoSpecificQCow2, 1), 5161 }; 5162 if (s->qcow_version == 2) { 5163 *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){ 5164 .compat = g_strdup("0.10"), 5165 .refcount_bits = s->refcount_bits, 5166 }; 5167 } else if (s->qcow_version == 3) { 5168 Qcow2BitmapInfoList *bitmaps; 5169 if (!qcow2_get_bitmap_info_list(bs, &bitmaps, errp)) { 5170 qapi_free_ImageInfoSpecific(spec_info); 5171 qapi_free_QCryptoBlockInfo(encrypt_info); 5172 return NULL; 5173 } 5174 *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){ 5175 .compat = g_strdup("1.1"), 5176 .lazy_refcounts = s->compatible_features & 5177 QCOW2_COMPAT_LAZY_REFCOUNTS, 5178 .has_lazy_refcounts = true, 5179 .corrupt = s->incompatible_features & 5180 QCOW2_INCOMPAT_CORRUPT, 5181 .has_corrupt = true, 5182 .has_extended_l2 = true, 5183 .extended_l2 = has_subclusters(s), 5184 .refcount_bits = s->refcount_bits, 5185 .has_bitmaps = !!bitmaps, 5186 .bitmaps = bitmaps, 5187 .has_data_file = !!s->image_data_file, 5188 .data_file = g_strdup(s->image_data_file), 5189 .has_data_file_raw = has_data_file(bs), 5190 .data_file_raw = data_file_is_raw(bs), 5191 .compression_type = s->compression_type, 5192 }; 5193 } else { 5194 /* if this assertion fails, this probably means a new version was 5195 * added without having it covered here */ 5196 assert(false); 5197 } 5198 5199 if (encrypt_info) { 5200 ImageInfoSpecificQCow2Encryption *qencrypt = 5201 g_new(ImageInfoSpecificQCow2Encryption, 1); 5202 switch (encrypt_info->format) { 5203 case Q_CRYPTO_BLOCK_FORMAT_QCOW: 5204 qencrypt->format = BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_AES; 5205 break; 5206 case Q_CRYPTO_BLOCK_FORMAT_LUKS: 5207 qencrypt->format = BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_LUKS; 5208 qencrypt->u.luks = encrypt_info->u.luks; 5209 break; 5210 default: 5211 abort(); 5212 } 5213 /* Since we did shallow copy above, erase any pointers 5214 * in the original info */ 5215 memset(&encrypt_info->u, 0, sizeof(encrypt_info->u)); 5216 qapi_free_QCryptoBlockInfo(encrypt_info); 5217 5218 spec_info->u.qcow2.data->has_encrypt = true; 5219 spec_info->u.qcow2.data->encrypt = qencrypt; 5220 } 5221 5222 return spec_info; 5223 } 5224 5225 static int qcow2_has_zero_init(BlockDriverState *bs) 5226 { 5227 BDRVQcow2State *s = bs->opaque; 5228 bool preallocated; 5229 5230 if (qemu_in_coroutine()) { 5231 qemu_co_mutex_lock(&s->lock); 5232 } 5233 /* 5234 * Check preallocation status: Preallocated images have all L2 5235 * tables allocated, nonpreallocated images have none. It is 5236 * therefore enough to check the first one. 5237 */ 5238 preallocated = s->l1_size > 0 && s->l1_table[0] != 0; 5239 if (qemu_in_coroutine()) { 5240 qemu_co_mutex_unlock(&s->lock); 5241 } 5242 5243 if (!preallocated) { 5244 return 1; 5245 } else if (bs->encrypted) { 5246 return 0; 5247 } else { 5248 return bdrv_has_zero_init(s->data_file->bs); 5249 } 5250 } 5251 5252 /* 5253 * Check the request to vmstate. On success return 5254 * qcow2_vm_state_offset(bs) + @pos 5255 */ 5256 static int64_t qcow2_check_vmstate_request(BlockDriverState *bs, 5257 QEMUIOVector *qiov, int64_t pos) 5258 { 5259 BDRVQcow2State *s = bs->opaque; 5260 int64_t vmstate_offset = qcow2_vm_state_offset(s); 5261 int ret; 5262 5263 /* Incoming requests must be OK */ 5264 bdrv_check_qiov_request(pos, qiov->size, qiov, 0, &error_abort); 5265 5266 if (INT64_MAX - pos < vmstate_offset) { 5267 return -EIO; 5268 } 5269 5270 pos += vmstate_offset; 5271 ret = bdrv_check_qiov_request(pos, qiov->size, qiov, 0, NULL); 5272 if (ret < 0) { 5273 return ret; 5274 } 5275 5276 return pos; 5277 } 5278 5279 static int qcow2_save_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, 5280 int64_t pos) 5281 { 5282 int64_t offset = qcow2_check_vmstate_request(bs, qiov, pos); 5283 if (offset < 0) { 5284 return offset; 5285 } 5286 5287 BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_SAVE); 5288 return bs->drv->bdrv_co_pwritev_part(bs, offset, qiov->size, qiov, 0, 0); 5289 } 5290 5291 static int qcow2_load_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, 5292 int64_t pos) 5293 { 5294 int64_t offset = qcow2_check_vmstate_request(bs, qiov, pos); 5295 if (offset < 0) { 5296 return offset; 5297 } 5298 5299 BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_LOAD); 5300 return bs->drv->bdrv_co_preadv_part(bs, offset, qiov->size, qiov, 0, 0); 5301 } 5302 5303 static int qcow2_has_compressed_clusters(BlockDriverState *bs) 5304 { 5305 int64_t offset = 0; 5306 int64_t bytes = bdrv_getlength(bs); 5307 5308 if (bytes < 0) { 5309 return bytes; 5310 } 5311 5312 while (bytes != 0) { 5313 int ret; 5314 QCow2SubclusterType type; 5315 unsigned int cur_bytes = MIN(INT_MAX, bytes); 5316 uint64_t host_offset; 5317 5318 ret = qcow2_get_host_offset(bs, offset, &cur_bytes, &host_offset, 5319 &type); 5320 if (ret < 0) { 5321 return ret; 5322 } 5323 5324 if (type == QCOW2_SUBCLUSTER_COMPRESSED) { 5325 return 1; 5326 } 5327 5328 offset += cur_bytes; 5329 bytes -= cur_bytes; 5330 } 5331 5332 return 0; 5333 } 5334 5335 /* 5336 * Downgrades an image's version. To achieve this, any incompatible features 5337 * have to be removed. 5338 */ 5339 static int qcow2_downgrade(BlockDriverState *bs, int target_version, 5340 BlockDriverAmendStatusCB *status_cb, void *cb_opaque, 5341 Error **errp) 5342 { 5343 BDRVQcow2State *s = bs->opaque; 5344 int current_version = s->qcow_version; 5345 int ret; 5346 int i; 5347 5348 /* This is qcow2_downgrade(), not qcow2_upgrade() */ 5349 assert(target_version < current_version); 5350 5351 /* There are no other versions (now) that you can downgrade to */ 5352 assert(target_version == 2); 5353 5354 if (s->refcount_order != 4) { 5355 error_setg(errp, "compat=0.10 requires refcount_bits=16"); 5356 return -ENOTSUP; 5357 } 5358 5359 if (has_data_file(bs)) { 5360 error_setg(errp, "Cannot downgrade an image with a data file"); 5361 return -ENOTSUP; 5362 } 5363 5364 /* 5365 * If any internal snapshot has a different size than the current 5366 * image size, or VM state size that exceeds 32 bits, downgrading 5367 * is unsafe. Even though we would still use v3-compliant output 5368 * to preserve that data, other v2 programs might not realize 5369 * those optional fields are important. 5370 */ 5371 for (i = 0; i < s->nb_snapshots; i++) { 5372 if (s->snapshots[i].vm_state_size > UINT32_MAX || 5373 s->snapshots[i].disk_size != bs->total_sectors * BDRV_SECTOR_SIZE) { 5374 error_setg(errp, "Internal snapshots prevent downgrade of image"); 5375 return -ENOTSUP; 5376 } 5377 } 5378 5379 /* clear incompatible features */ 5380 if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) { 5381 ret = qcow2_mark_clean(bs); 5382 if (ret < 0) { 5383 error_setg_errno(errp, -ret, "Failed to make the image clean"); 5384 return ret; 5385 } 5386 } 5387 5388 /* with QCOW2_INCOMPAT_CORRUPT, it is pretty much impossible to get here in 5389 * the first place; if that happens nonetheless, returning -ENOTSUP is the 5390 * best thing to do anyway */ 5391 5392 if (s->incompatible_features & ~QCOW2_INCOMPAT_COMPRESSION) { 5393 error_setg(errp, "Cannot downgrade an image with incompatible features " 5394 "0x%" PRIx64 " set", 5395 s->incompatible_features & ~QCOW2_INCOMPAT_COMPRESSION); 5396 return -ENOTSUP; 5397 } 5398 5399 /* since we can ignore compatible features, we can set them to 0 as well */ 5400 s->compatible_features = 0; 5401 /* if lazy refcounts have been used, they have already been fixed through 5402 * clearing the dirty flag */ 5403 5404 /* clearing autoclear features is trivial */ 5405 s->autoclear_features = 0; 5406 5407 ret = qcow2_expand_zero_clusters(bs, status_cb, cb_opaque); 5408 if (ret < 0) { 5409 error_setg_errno(errp, -ret, "Failed to turn zero into data clusters"); 5410 return ret; 5411 } 5412 5413 if (s->incompatible_features & QCOW2_INCOMPAT_COMPRESSION) { 5414 ret = qcow2_has_compressed_clusters(bs); 5415 if (ret < 0) { 5416 error_setg(errp, "Failed to check block status"); 5417 return -EINVAL; 5418 } 5419 if (ret) { 5420 error_setg(errp, "Cannot downgrade an image with zstd compression " 5421 "type and existing compressed clusters"); 5422 return -ENOTSUP; 5423 } 5424 /* 5425 * No compressed clusters for now, so just chose default zlib 5426 * compression. 5427 */ 5428 s->incompatible_features &= ~QCOW2_INCOMPAT_COMPRESSION; 5429 s->compression_type = QCOW2_COMPRESSION_TYPE_ZLIB; 5430 } 5431 5432 assert(s->incompatible_features == 0); 5433 5434 s->qcow_version = target_version; 5435 ret = qcow2_update_header(bs); 5436 if (ret < 0) { 5437 s->qcow_version = current_version; 5438 error_setg_errno(errp, -ret, "Failed to update the image header"); 5439 return ret; 5440 } 5441 return 0; 5442 } 5443 5444 /* 5445 * Upgrades an image's version. While newer versions encompass all 5446 * features of older versions, some things may have to be presented 5447 * differently. 5448 */ 5449 static int qcow2_upgrade(BlockDriverState *bs, int target_version, 5450 BlockDriverAmendStatusCB *status_cb, void *cb_opaque, 5451 Error **errp) 5452 { 5453 BDRVQcow2State *s = bs->opaque; 5454 bool need_snapshot_update; 5455 int current_version = s->qcow_version; 5456 int i; 5457 int ret; 5458 5459 /* This is qcow2_upgrade(), not qcow2_downgrade() */ 5460 assert(target_version > current_version); 5461 5462 /* There are no other versions (yet) that you can upgrade to */ 5463 assert(target_version == 3); 5464 5465 status_cb(bs, 0, 2, cb_opaque); 5466 5467 /* 5468 * In v2, snapshots do not need to have extra data. v3 requires 5469 * the 64-bit VM state size and the virtual disk size to be 5470 * present. 5471 * qcow2_write_snapshots() will always write the list in the 5472 * v3-compliant format. 5473 */ 5474 need_snapshot_update = false; 5475 for (i = 0; i < s->nb_snapshots; i++) { 5476 if (s->snapshots[i].extra_data_size < 5477 sizeof_field(QCowSnapshotExtraData, vm_state_size_large) + 5478 sizeof_field(QCowSnapshotExtraData, disk_size)) 5479 { 5480 need_snapshot_update = true; 5481 break; 5482 } 5483 } 5484 if (need_snapshot_update) { 5485 ret = qcow2_write_snapshots(bs); 5486 if (ret < 0) { 5487 error_setg_errno(errp, -ret, "Failed to update the snapshot table"); 5488 return ret; 5489 } 5490 } 5491 status_cb(bs, 1, 2, cb_opaque); 5492 5493 s->qcow_version = target_version; 5494 ret = qcow2_update_header(bs); 5495 if (ret < 0) { 5496 s->qcow_version = current_version; 5497 error_setg_errno(errp, -ret, "Failed to update the image header"); 5498 return ret; 5499 } 5500 status_cb(bs, 2, 2, cb_opaque); 5501 5502 return 0; 5503 } 5504 5505 typedef enum Qcow2AmendOperation { 5506 /* This is the value Qcow2AmendHelperCBInfo::last_operation will be 5507 * statically initialized to so that the helper CB can discern the first 5508 * invocation from an operation change */ 5509 QCOW2_NO_OPERATION = 0, 5510 5511 QCOW2_UPGRADING, 5512 QCOW2_UPDATING_ENCRYPTION, 5513 QCOW2_CHANGING_REFCOUNT_ORDER, 5514 QCOW2_DOWNGRADING, 5515 } Qcow2AmendOperation; 5516 5517 typedef struct Qcow2AmendHelperCBInfo { 5518 /* The code coordinating the amend operations should only modify 5519 * these four fields; the rest will be managed by the CB */ 5520 BlockDriverAmendStatusCB *original_status_cb; 5521 void *original_cb_opaque; 5522 5523 Qcow2AmendOperation current_operation; 5524 5525 /* Total number of operations to perform (only set once) */ 5526 int total_operations; 5527 5528 /* The following fields are managed by the CB */ 5529 5530 /* Number of operations completed */ 5531 int operations_completed; 5532 5533 /* Cumulative offset of all completed operations */ 5534 int64_t offset_completed; 5535 5536 Qcow2AmendOperation last_operation; 5537 int64_t last_work_size; 5538 } Qcow2AmendHelperCBInfo; 5539 5540 static void qcow2_amend_helper_cb(BlockDriverState *bs, 5541 int64_t operation_offset, 5542 int64_t operation_work_size, void *opaque) 5543 { 5544 Qcow2AmendHelperCBInfo *info = opaque; 5545 int64_t current_work_size; 5546 int64_t projected_work_size; 5547 5548 if (info->current_operation != info->last_operation) { 5549 if (info->last_operation != QCOW2_NO_OPERATION) { 5550 info->offset_completed += info->last_work_size; 5551 info->operations_completed++; 5552 } 5553 5554 info->last_operation = info->current_operation; 5555 } 5556 5557 assert(info->total_operations > 0); 5558 assert(info->operations_completed < info->total_operations); 5559 5560 info->last_work_size = operation_work_size; 5561 5562 current_work_size = info->offset_completed + operation_work_size; 5563 5564 /* current_work_size is the total work size for (operations_completed + 1) 5565 * operations (which includes this one), so multiply it by the number of 5566 * operations not covered and divide it by the number of operations 5567 * covered to get a projection for the operations not covered */ 5568 projected_work_size = current_work_size * (info->total_operations - 5569 info->operations_completed - 1) 5570 / (info->operations_completed + 1); 5571 5572 info->original_status_cb(bs, info->offset_completed + operation_offset, 5573 current_work_size + projected_work_size, 5574 info->original_cb_opaque); 5575 } 5576 5577 static int qcow2_amend_options(BlockDriverState *bs, QemuOpts *opts, 5578 BlockDriverAmendStatusCB *status_cb, 5579 void *cb_opaque, 5580 bool force, 5581 Error **errp) 5582 { 5583 BDRVQcow2State *s = bs->opaque; 5584 int old_version = s->qcow_version, new_version = old_version; 5585 uint64_t new_size = 0; 5586 const char *backing_file = NULL, *backing_format = NULL, *data_file = NULL; 5587 bool lazy_refcounts = s->use_lazy_refcounts; 5588 bool data_file_raw = data_file_is_raw(bs); 5589 const char *compat = NULL; 5590 int refcount_bits = s->refcount_bits; 5591 int ret; 5592 QemuOptDesc *desc = opts->list->desc; 5593 Qcow2AmendHelperCBInfo helper_cb_info; 5594 bool encryption_update = false; 5595 5596 while (desc && desc->name) { 5597 if (!qemu_opt_find(opts, desc->name)) { 5598 /* only change explicitly defined options */ 5599 desc++; 5600 continue; 5601 } 5602 5603 if (!strcmp(desc->name, BLOCK_OPT_COMPAT_LEVEL)) { 5604 compat = qemu_opt_get(opts, BLOCK_OPT_COMPAT_LEVEL); 5605 if (!compat) { 5606 /* preserve default */ 5607 } else if (!strcmp(compat, "0.10") || !strcmp(compat, "v2")) { 5608 new_version = 2; 5609 } else if (!strcmp(compat, "1.1") || !strcmp(compat, "v3")) { 5610 new_version = 3; 5611 } else { 5612 error_setg(errp, "Unknown compatibility level %s", compat); 5613 return -EINVAL; 5614 } 5615 } else if (!strcmp(desc->name, BLOCK_OPT_SIZE)) { 5616 new_size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, 0); 5617 } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FILE)) { 5618 backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE); 5619 } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FMT)) { 5620 backing_format = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT); 5621 } else if (g_str_has_prefix(desc->name, "encrypt.")) { 5622 if (!s->crypto) { 5623 error_setg(errp, 5624 "Can't amend encryption options - encryption not present"); 5625 return -EINVAL; 5626 } 5627 if (s->crypt_method_header != QCOW_CRYPT_LUKS) { 5628 error_setg(errp, 5629 "Only LUKS encryption options can be amended"); 5630 return -ENOTSUP; 5631 } 5632 encryption_update = true; 5633 } else if (!strcmp(desc->name, BLOCK_OPT_LAZY_REFCOUNTS)) { 5634 lazy_refcounts = qemu_opt_get_bool(opts, BLOCK_OPT_LAZY_REFCOUNTS, 5635 lazy_refcounts); 5636 } else if (!strcmp(desc->name, BLOCK_OPT_REFCOUNT_BITS)) { 5637 refcount_bits = qemu_opt_get_number(opts, BLOCK_OPT_REFCOUNT_BITS, 5638 refcount_bits); 5639 5640 if (refcount_bits <= 0 || refcount_bits > 64 || 5641 !is_power_of_2(refcount_bits)) 5642 { 5643 error_setg(errp, "Refcount width must be a power of two and " 5644 "may not exceed 64 bits"); 5645 return -EINVAL; 5646 } 5647 } else if (!strcmp(desc->name, BLOCK_OPT_DATA_FILE)) { 5648 data_file = qemu_opt_get(opts, BLOCK_OPT_DATA_FILE); 5649 if (data_file && !has_data_file(bs)) { 5650 error_setg(errp, "data-file can only be set for images that " 5651 "use an external data file"); 5652 return -EINVAL; 5653 } 5654 } else if (!strcmp(desc->name, BLOCK_OPT_DATA_FILE_RAW)) { 5655 data_file_raw = qemu_opt_get_bool(opts, BLOCK_OPT_DATA_FILE_RAW, 5656 data_file_raw); 5657 if (data_file_raw && !data_file_is_raw(bs)) { 5658 error_setg(errp, "data-file-raw cannot be set on existing " 5659 "images"); 5660 return -EINVAL; 5661 } 5662 } else { 5663 /* if this point is reached, this probably means a new option was 5664 * added without having it covered here */ 5665 abort(); 5666 } 5667 5668 desc++; 5669 } 5670 5671 helper_cb_info = (Qcow2AmendHelperCBInfo){ 5672 .original_status_cb = status_cb, 5673 .original_cb_opaque = cb_opaque, 5674 .total_operations = (new_version != old_version) 5675 + (s->refcount_bits != refcount_bits) + 5676 (encryption_update == true) 5677 }; 5678 5679 /* Upgrade first (some features may require compat=1.1) */ 5680 if (new_version > old_version) { 5681 helper_cb_info.current_operation = QCOW2_UPGRADING; 5682 ret = qcow2_upgrade(bs, new_version, &qcow2_amend_helper_cb, 5683 &helper_cb_info, errp); 5684 if (ret < 0) { 5685 return ret; 5686 } 5687 } 5688 5689 if (encryption_update) { 5690 QDict *amend_opts_dict; 5691 QCryptoBlockAmendOptions *amend_opts; 5692 5693 helper_cb_info.current_operation = QCOW2_UPDATING_ENCRYPTION; 5694 amend_opts_dict = qcow2_extract_crypto_opts(opts, "luks", errp); 5695 if (!amend_opts_dict) { 5696 return -EINVAL; 5697 } 5698 amend_opts = block_crypto_amend_opts_init(amend_opts_dict, errp); 5699 qobject_unref(amend_opts_dict); 5700 if (!amend_opts) { 5701 return -EINVAL; 5702 } 5703 ret = qcrypto_block_amend_options(s->crypto, 5704 qcow2_crypto_hdr_read_func, 5705 qcow2_crypto_hdr_write_func, 5706 bs, 5707 amend_opts, 5708 force, 5709 errp); 5710 qapi_free_QCryptoBlockAmendOptions(amend_opts); 5711 if (ret < 0) { 5712 return ret; 5713 } 5714 } 5715 5716 if (s->refcount_bits != refcount_bits) { 5717 int refcount_order = ctz32(refcount_bits); 5718 5719 if (new_version < 3 && refcount_bits != 16) { 5720 error_setg(errp, "Refcount widths other than 16 bits require " 5721 "compatibility level 1.1 or above (use compat=1.1 or " 5722 "greater)"); 5723 return -EINVAL; 5724 } 5725 5726 helper_cb_info.current_operation = QCOW2_CHANGING_REFCOUNT_ORDER; 5727 ret = qcow2_change_refcount_order(bs, refcount_order, 5728 &qcow2_amend_helper_cb, 5729 &helper_cb_info, errp); 5730 if (ret < 0) { 5731 return ret; 5732 } 5733 } 5734 5735 /* data-file-raw blocks backing files, so clear it first if requested */ 5736 if (data_file_raw) { 5737 s->autoclear_features |= QCOW2_AUTOCLEAR_DATA_FILE_RAW; 5738 } else { 5739 s->autoclear_features &= ~QCOW2_AUTOCLEAR_DATA_FILE_RAW; 5740 } 5741 5742 if (data_file) { 5743 g_free(s->image_data_file); 5744 s->image_data_file = *data_file ? g_strdup(data_file) : NULL; 5745 } 5746 5747 ret = qcow2_update_header(bs); 5748 if (ret < 0) { 5749 error_setg_errno(errp, -ret, "Failed to update the image header"); 5750 return ret; 5751 } 5752 5753 if (backing_file || backing_format) { 5754 if (g_strcmp0(backing_file, s->image_backing_file) || 5755 g_strcmp0(backing_format, s->image_backing_format)) { 5756 error_setg(errp, "Cannot amend the backing file"); 5757 error_append_hint(errp, 5758 "You can use 'qemu-img rebase' instead.\n"); 5759 return -EINVAL; 5760 } 5761 } 5762 5763 if (s->use_lazy_refcounts != lazy_refcounts) { 5764 if (lazy_refcounts) { 5765 if (new_version < 3) { 5766 error_setg(errp, "Lazy refcounts only supported with " 5767 "compatibility level 1.1 and above (use compat=1.1 " 5768 "or greater)"); 5769 return -EINVAL; 5770 } 5771 s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS; 5772 ret = qcow2_update_header(bs); 5773 if (ret < 0) { 5774 s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS; 5775 error_setg_errno(errp, -ret, "Failed to update the image header"); 5776 return ret; 5777 } 5778 s->use_lazy_refcounts = true; 5779 } else { 5780 /* make image clean first */ 5781 ret = qcow2_mark_clean(bs); 5782 if (ret < 0) { 5783 error_setg_errno(errp, -ret, "Failed to make the image clean"); 5784 return ret; 5785 } 5786 /* now disallow lazy refcounts */ 5787 s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS; 5788 ret = qcow2_update_header(bs); 5789 if (ret < 0) { 5790 s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS; 5791 error_setg_errno(errp, -ret, "Failed to update the image header"); 5792 return ret; 5793 } 5794 s->use_lazy_refcounts = false; 5795 } 5796 } 5797 5798 if (new_size) { 5799 BlockBackend *blk = blk_new_with_bs(bs, BLK_PERM_RESIZE, BLK_PERM_ALL, 5800 errp); 5801 if (!blk) { 5802 return -EPERM; 5803 } 5804 5805 /* 5806 * Amending image options should ensure that the image has 5807 * exactly the given new values, so pass exact=true here. 5808 */ 5809 ret = blk_truncate(blk, new_size, true, PREALLOC_MODE_OFF, 0, errp); 5810 blk_unref(blk); 5811 if (ret < 0) { 5812 return ret; 5813 } 5814 } 5815 5816 /* Downgrade last (so unsupported features can be removed before) */ 5817 if (new_version < old_version) { 5818 helper_cb_info.current_operation = QCOW2_DOWNGRADING; 5819 ret = qcow2_downgrade(bs, new_version, &qcow2_amend_helper_cb, 5820 &helper_cb_info, errp); 5821 if (ret < 0) { 5822 return ret; 5823 } 5824 } 5825 5826 return 0; 5827 } 5828 5829 static int coroutine_fn qcow2_co_amend(BlockDriverState *bs, 5830 BlockdevAmendOptions *opts, 5831 bool force, 5832 Error **errp) 5833 { 5834 BlockdevAmendOptionsQcow2 *qopts = &opts->u.qcow2; 5835 BDRVQcow2State *s = bs->opaque; 5836 int ret = 0; 5837 5838 if (qopts->has_encrypt) { 5839 if (!s->crypto) { 5840 error_setg(errp, "image is not encrypted, can't amend"); 5841 return -EOPNOTSUPP; 5842 } 5843 5844 if (qopts->encrypt->format != Q_CRYPTO_BLOCK_FORMAT_LUKS) { 5845 error_setg(errp, 5846 "Amend can't be used to change the qcow2 encryption format"); 5847 return -EOPNOTSUPP; 5848 } 5849 5850 if (s->crypt_method_header != QCOW_CRYPT_LUKS) { 5851 error_setg(errp, 5852 "Only LUKS encryption options can be amended for qcow2 with blockdev-amend"); 5853 return -EOPNOTSUPP; 5854 } 5855 5856 ret = qcrypto_block_amend_options(s->crypto, 5857 qcow2_crypto_hdr_read_func, 5858 qcow2_crypto_hdr_write_func, 5859 bs, 5860 qopts->encrypt, 5861 force, 5862 errp); 5863 } 5864 return ret; 5865 } 5866 5867 /* 5868 * If offset or size are negative, respectively, they will not be included in 5869 * the BLOCK_IMAGE_CORRUPTED event emitted. 5870 * fatal will be ignored for read-only BDS; corruptions found there will always 5871 * be considered non-fatal. 5872 */ 5873 void qcow2_signal_corruption(BlockDriverState *bs, bool fatal, int64_t offset, 5874 int64_t size, const char *message_format, ...) 5875 { 5876 BDRVQcow2State *s = bs->opaque; 5877 const char *node_name; 5878 char *message; 5879 va_list ap; 5880 5881 fatal = fatal && bdrv_is_writable(bs); 5882 5883 if (s->signaled_corruption && 5884 (!fatal || (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT))) 5885 { 5886 return; 5887 } 5888 5889 va_start(ap, message_format); 5890 message = g_strdup_vprintf(message_format, ap); 5891 va_end(ap); 5892 5893 if (fatal) { 5894 fprintf(stderr, "qcow2: Marking image as corrupt: %s; further " 5895 "corruption events will be suppressed\n", message); 5896 } else { 5897 fprintf(stderr, "qcow2: Image is corrupt: %s; further non-fatal " 5898 "corruption events will be suppressed\n", message); 5899 } 5900 5901 node_name = bdrv_get_node_name(bs); 5902 qapi_event_send_block_image_corrupted(bdrv_get_device_name(bs), 5903 *node_name != '\0', node_name, 5904 message, offset >= 0, offset, 5905 size >= 0, size, 5906 fatal); 5907 g_free(message); 5908 5909 if (fatal) { 5910 qcow2_mark_corrupt(bs); 5911 bs->drv = NULL; /* make BDS unusable */ 5912 } 5913 5914 s->signaled_corruption = true; 5915 } 5916 5917 #define QCOW_COMMON_OPTIONS \ 5918 { \ 5919 .name = BLOCK_OPT_SIZE, \ 5920 .type = QEMU_OPT_SIZE, \ 5921 .help = "Virtual disk size" \ 5922 }, \ 5923 { \ 5924 .name = BLOCK_OPT_COMPAT_LEVEL, \ 5925 .type = QEMU_OPT_STRING, \ 5926 .help = "Compatibility level (v2 [0.10] or v3 [1.1])" \ 5927 }, \ 5928 { \ 5929 .name = BLOCK_OPT_BACKING_FILE, \ 5930 .type = QEMU_OPT_STRING, \ 5931 .help = "File name of a base image" \ 5932 }, \ 5933 { \ 5934 .name = BLOCK_OPT_BACKING_FMT, \ 5935 .type = QEMU_OPT_STRING, \ 5936 .help = "Image format of the base image" \ 5937 }, \ 5938 { \ 5939 .name = BLOCK_OPT_DATA_FILE, \ 5940 .type = QEMU_OPT_STRING, \ 5941 .help = "File name of an external data file" \ 5942 }, \ 5943 { \ 5944 .name = BLOCK_OPT_DATA_FILE_RAW, \ 5945 .type = QEMU_OPT_BOOL, \ 5946 .help = "The external data file must stay valid " \ 5947 "as a raw image" \ 5948 }, \ 5949 { \ 5950 .name = BLOCK_OPT_LAZY_REFCOUNTS, \ 5951 .type = QEMU_OPT_BOOL, \ 5952 .help = "Postpone refcount updates", \ 5953 .def_value_str = "off" \ 5954 }, \ 5955 { \ 5956 .name = BLOCK_OPT_REFCOUNT_BITS, \ 5957 .type = QEMU_OPT_NUMBER, \ 5958 .help = "Width of a reference count entry in bits", \ 5959 .def_value_str = "16" \ 5960 } 5961 5962 static QemuOptsList qcow2_create_opts = { 5963 .name = "qcow2-create-opts", 5964 .head = QTAILQ_HEAD_INITIALIZER(qcow2_create_opts.head), 5965 .desc = { 5966 { \ 5967 .name = BLOCK_OPT_ENCRYPT, \ 5968 .type = QEMU_OPT_BOOL, \ 5969 .help = "Encrypt the image with format 'aes'. (Deprecated " \ 5970 "in favor of " BLOCK_OPT_ENCRYPT_FORMAT "=aes)", \ 5971 }, \ 5972 { \ 5973 .name = BLOCK_OPT_ENCRYPT_FORMAT, \ 5974 .type = QEMU_OPT_STRING, \ 5975 .help = "Encrypt the image, format choices: 'aes', 'luks'", \ 5976 }, \ 5977 BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.", \ 5978 "ID of secret providing qcow AES key or LUKS passphrase"), \ 5979 BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_ALG("encrypt."), \ 5980 BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_MODE("encrypt."), \ 5981 BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_ALG("encrypt."), \ 5982 BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_HASH_ALG("encrypt."), \ 5983 BLOCK_CRYPTO_OPT_DEF_LUKS_HASH_ALG("encrypt."), \ 5984 BLOCK_CRYPTO_OPT_DEF_LUKS_ITER_TIME("encrypt."), \ 5985 { \ 5986 .name = BLOCK_OPT_CLUSTER_SIZE, \ 5987 .type = QEMU_OPT_SIZE, \ 5988 .help = "qcow2 cluster size", \ 5989 .def_value_str = stringify(DEFAULT_CLUSTER_SIZE) \ 5990 }, \ 5991 { \ 5992 .name = BLOCK_OPT_EXTL2, \ 5993 .type = QEMU_OPT_BOOL, \ 5994 .help = "Extended L2 tables", \ 5995 .def_value_str = "off" \ 5996 }, \ 5997 { \ 5998 .name = BLOCK_OPT_PREALLOC, \ 5999 .type = QEMU_OPT_STRING, \ 6000 .help = "Preallocation mode (allowed values: off, " \ 6001 "metadata, falloc, full)" \ 6002 }, \ 6003 { \ 6004 .name = BLOCK_OPT_COMPRESSION_TYPE, \ 6005 .type = QEMU_OPT_STRING, \ 6006 .help = "Compression method used for image cluster " \ 6007 "compression", \ 6008 .def_value_str = "zlib" \ 6009 }, 6010 QCOW_COMMON_OPTIONS, 6011 { /* end of list */ } 6012 } 6013 }; 6014 6015 static QemuOptsList qcow2_amend_opts = { 6016 .name = "qcow2-amend-opts", 6017 .head = QTAILQ_HEAD_INITIALIZER(qcow2_amend_opts.head), 6018 .desc = { 6019 BLOCK_CRYPTO_OPT_DEF_LUKS_STATE("encrypt."), 6020 BLOCK_CRYPTO_OPT_DEF_LUKS_KEYSLOT("encrypt."), 6021 BLOCK_CRYPTO_OPT_DEF_LUKS_OLD_SECRET("encrypt."), 6022 BLOCK_CRYPTO_OPT_DEF_LUKS_NEW_SECRET("encrypt."), 6023 BLOCK_CRYPTO_OPT_DEF_LUKS_ITER_TIME("encrypt."), 6024 QCOW_COMMON_OPTIONS, 6025 { /* end of list */ } 6026 } 6027 }; 6028 6029 static const char *const qcow2_strong_runtime_opts[] = { 6030 "encrypt." BLOCK_CRYPTO_OPT_QCOW_KEY_SECRET, 6031 6032 NULL 6033 }; 6034 6035 BlockDriver bdrv_qcow2 = { 6036 .format_name = "qcow2", 6037 .instance_size = sizeof(BDRVQcow2State), 6038 .bdrv_probe = qcow2_probe, 6039 .bdrv_open = qcow2_open, 6040 .bdrv_close = qcow2_close, 6041 .bdrv_reopen_prepare = qcow2_reopen_prepare, 6042 .bdrv_reopen_commit = qcow2_reopen_commit, 6043 .bdrv_reopen_commit_post = qcow2_reopen_commit_post, 6044 .bdrv_reopen_abort = qcow2_reopen_abort, 6045 .bdrv_join_options = qcow2_join_options, 6046 .bdrv_child_perm = bdrv_default_perms, 6047 .bdrv_co_create_opts = qcow2_co_create_opts, 6048 .bdrv_co_create = qcow2_co_create, 6049 .bdrv_has_zero_init = qcow2_has_zero_init, 6050 .bdrv_co_block_status = qcow2_co_block_status, 6051 6052 .bdrv_co_preadv_part = qcow2_co_preadv_part, 6053 .bdrv_co_pwritev_part = qcow2_co_pwritev_part, 6054 .bdrv_co_flush_to_os = qcow2_co_flush_to_os, 6055 6056 .bdrv_co_pwrite_zeroes = qcow2_co_pwrite_zeroes, 6057 .bdrv_co_pdiscard = qcow2_co_pdiscard, 6058 .bdrv_co_copy_range_from = qcow2_co_copy_range_from, 6059 .bdrv_co_copy_range_to = qcow2_co_copy_range_to, 6060 .bdrv_co_truncate = qcow2_co_truncate, 6061 .bdrv_co_pwritev_compressed_part = qcow2_co_pwritev_compressed_part, 6062 .bdrv_make_empty = qcow2_make_empty, 6063 6064 .bdrv_snapshot_create = qcow2_snapshot_create, 6065 .bdrv_snapshot_goto = qcow2_snapshot_goto, 6066 .bdrv_snapshot_delete = qcow2_snapshot_delete, 6067 .bdrv_snapshot_list = qcow2_snapshot_list, 6068 .bdrv_snapshot_load_tmp = qcow2_snapshot_load_tmp, 6069 .bdrv_measure = qcow2_measure, 6070 .bdrv_get_info = qcow2_get_info, 6071 .bdrv_get_specific_info = qcow2_get_specific_info, 6072 6073 .bdrv_save_vmstate = qcow2_save_vmstate, 6074 .bdrv_load_vmstate = qcow2_load_vmstate, 6075 6076 .is_format = true, 6077 .supports_backing = true, 6078 .bdrv_change_backing_file = qcow2_change_backing_file, 6079 6080 .bdrv_refresh_limits = qcow2_refresh_limits, 6081 .bdrv_co_invalidate_cache = qcow2_co_invalidate_cache, 6082 .bdrv_inactivate = qcow2_inactivate, 6083 6084 .create_opts = &qcow2_create_opts, 6085 .amend_opts = &qcow2_amend_opts, 6086 .strong_runtime_opts = qcow2_strong_runtime_opts, 6087 .mutable_opts = mutable_opts, 6088 .bdrv_co_check = qcow2_co_check, 6089 .bdrv_amend_options = qcow2_amend_options, 6090 .bdrv_co_amend = qcow2_co_amend, 6091 6092 .bdrv_detach_aio_context = qcow2_detach_aio_context, 6093 .bdrv_attach_aio_context = qcow2_attach_aio_context, 6094 6095 .bdrv_supports_persistent_dirty_bitmap = 6096 qcow2_supports_persistent_dirty_bitmap, 6097 .bdrv_co_can_store_new_dirty_bitmap = qcow2_co_can_store_new_dirty_bitmap, 6098 .bdrv_co_remove_persistent_dirty_bitmap = 6099 qcow2_co_remove_persistent_dirty_bitmap, 6100 }; 6101 6102 static void bdrv_qcow2_init(void) 6103 { 6104 bdrv_register(&bdrv_qcow2); 6105 } 6106 6107 block_init(bdrv_qcow2_init); 6108