1 /* 2 * Block driver for the QCOW version 2 format 3 * 4 * Copyright (c) 2004-2006 Fabrice Bellard 5 * 6 * Permission is hereby granted, free of charge, to any person obtaining a copy 7 * of this software and associated documentation files (the "Software"), to deal 8 * in the Software without restriction, including without limitation the rights 9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 10 * copies of the Software, and to permit persons to whom the Software is 11 * furnished to do so, subject to the following conditions: 12 * 13 * The above copyright notice and this permission notice shall be included in 14 * all copies or substantial portions of the Software. 15 * 16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN 22 * THE SOFTWARE. 23 */ 24 25 #include "qemu/osdep.h" 26 27 #include "block/qdict.h" 28 #include "sysemu/block-backend.h" 29 #include "qemu/main-loop.h" 30 #include "qemu/module.h" 31 #include "qcow2.h" 32 #include "qemu/error-report.h" 33 #include "qapi/error.h" 34 #include "qapi/qapi-events-block-core.h" 35 #include "qapi/qmp/qdict.h" 36 #include "qapi/qmp/qstring.h" 37 #include "trace.h" 38 #include "qemu/option_int.h" 39 #include "qemu/cutils.h" 40 #include "qemu/bswap.h" 41 #include "qapi/qobject-input-visitor.h" 42 #include "qapi/qapi-visit-block-core.h" 43 #include "crypto.h" 44 #include "block/aio_task.h" 45 46 /* 47 Differences with QCOW: 48 49 - Support for multiple incremental snapshots. 50 - Memory management by reference counts. 51 - Clusters which have a reference count of one have the bit 52 QCOW_OFLAG_COPIED to optimize write performance. 53 - Size of compressed clusters is stored in sectors to reduce bit usage 54 in the cluster offsets. 55 - Support for storing additional data (such as the VM state) in the 56 snapshots. 57 - If a backing store is used, the cluster size is not constrained 58 (could be backported to QCOW). 59 - L2 tables have always a size of one cluster. 60 */ 61 62 63 typedef struct { 64 uint32_t magic; 65 uint32_t len; 66 } QEMU_PACKED QCowExtension; 67 68 #define QCOW2_EXT_MAGIC_END 0 69 #define QCOW2_EXT_MAGIC_BACKING_FORMAT 0xe2792aca 70 #define QCOW2_EXT_MAGIC_FEATURE_TABLE 0x6803f857 71 #define QCOW2_EXT_MAGIC_CRYPTO_HEADER 0x0537be77 72 #define QCOW2_EXT_MAGIC_BITMAPS 0x23852875 73 #define QCOW2_EXT_MAGIC_DATA_FILE 0x44415441 74 75 static int coroutine_fn 76 qcow2_co_preadv_compressed(BlockDriverState *bs, 77 uint64_t cluster_descriptor, 78 uint64_t offset, 79 uint64_t bytes, 80 QEMUIOVector *qiov, 81 size_t qiov_offset); 82 83 static int qcow2_probe(const uint8_t *buf, int buf_size, const char *filename) 84 { 85 const QCowHeader *cow_header = (const void *)buf; 86 87 if (buf_size >= sizeof(QCowHeader) && 88 be32_to_cpu(cow_header->magic) == QCOW_MAGIC && 89 be32_to_cpu(cow_header->version) >= 2) 90 return 100; 91 else 92 return 0; 93 } 94 95 96 static ssize_t qcow2_crypto_hdr_read_func(QCryptoBlock *block, size_t offset, 97 uint8_t *buf, size_t buflen, 98 void *opaque, Error **errp) 99 { 100 BlockDriverState *bs = opaque; 101 BDRVQcow2State *s = bs->opaque; 102 ssize_t ret; 103 104 if ((offset + buflen) > s->crypto_header.length) { 105 error_setg(errp, "Request for data outside of extension header"); 106 return -1; 107 } 108 109 ret = bdrv_pread(bs->file, 110 s->crypto_header.offset + offset, buf, buflen); 111 if (ret < 0) { 112 error_setg_errno(errp, -ret, "Could not read encryption header"); 113 return -1; 114 } 115 return ret; 116 } 117 118 119 static ssize_t qcow2_crypto_hdr_init_func(QCryptoBlock *block, size_t headerlen, 120 void *opaque, Error **errp) 121 { 122 BlockDriverState *bs = opaque; 123 BDRVQcow2State *s = bs->opaque; 124 int64_t ret; 125 int64_t clusterlen; 126 127 ret = qcow2_alloc_clusters(bs, headerlen); 128 if (ret < 0) { 129 error_setg_errno(errp, -ret, 130 "Cannot allocate cluster for LUKS header size %zu", 131 headerlen); 132 return -1; 133 } 134 135 s->crypto_header.length = headerlen; 136 s->crypto_header.offset = ret; 137 138 /* 139 * Zero fill all space in cluster so it has predictable 140 * content, as we may not initialize some regions of the 141 * header (eg only 1 out of 8 key slots will be initialized) 142 */ 143 clusterlen = size_to_clusters(s, headerlen) * s->cluster_size; 144 assert(qcow2_pre_write_overlap_check(bs, 0, ret, clusterlen, false) == 0); 145 ret = bdrv_pwrite_zeroes(bs->file, 146 ret, 147 clusterlen, 0); 148 if (ret < 0) { 149 error_setg_errno(errp, -ret, "Could not zero fill encryption header"); 150 return -1; 151 } 152 153 return ret; 154 } 155 156 157 static ssize_t qcow2_crypto_hdr_write_func(QCryptoBlock *block, size_t offset, 158 const uint8_t *buf, size_t buflen, 159 void *opaque, Error **errp) 160 { 161 BlockDriverState *bs = opaque; 162 BDRVQcow2State *s = bs->opaque; 163 ssize_t ret; 164 165 if ((offset + buflen) > s->crypto_header.length) { 166 error_setg(errp, "Request for data outside of extension header"); 167 return -1; 168 } 169 170 ret = bdrv_pwrite(bs->file, 171 s->crypto_header.offset + offset, buf, buflen); 172 if (ret < 0) { 173 error_setg_errno(errp, -ret, "Could not read encryption header"); 174 return -1; 175 } 176 return ret; 177 } 178 179 static QDict* 180 qcow2_extract_crypto_opts(QemuOpts *opts, const char *fmt, Error **errp) 181 { 182 QDict *cryptoopts_qdict; 183 QDict *opts_qdict; 184 185 /* Extract "encrypt." options into a qdict */ 186 opts_qdict = qemu_opts_to_qdict(opts, NULL); 187 qdict_extract_subqdict(opts_qdict, &cryptoopts_qdict, "encrypt."); 188 qobject_unref(opts_qdict); 189 qdict_put_str(cryptoopts_qdict, "format", fmt); 190 return cryptoopts_qdict; 191 } 192 193 /* 194 * read qcow2 extension and fill bs 195 * start reading from start_offset 196 * finish reading upon magic of value 0 or when end_offset reached 197 * unknown magic is skipped (future extension this version knows nothing about) 198 * return 0 upon success, non-0 otherwise 199 */ 200 static int qcow2_read_extensions(BlockDriverState *bs, uint64_t start_offset, 201 uint64_t end_offset, void **p_feature_table, 202 int flags, bool *need_update_header, 203 Error **errp) 204 { 205 BDRVQcow2State *s = bs->opaque; 206 QCowExtension ext; 207 uint64_t offset; 208 int ret; 209 Qcow2BitmapHeaderExt bitmaps_ext; 210 211 if (need_update_header != NULL) { 212 *need_update_header = false; 213 } 214 215 #ifdef DEBUG_EXT 216 printf("qcow2_read_extensions: start=%ld end=%ld\n", start_offset, end_offset); 217 #endif 218 offset = start_offset; 219 while (offset < end_offset) { 220 221 #ifdef DEBUG_EXT 222 /* Sanity check */ 223 if (offset > s->cluster_size) 224 printf("qcow2_read_extension: suspicious offset %lu\n", offset); 225 226 printf("attempting to read extended header in offset %lu\n", offset); 227 #endif 228 229 ret = bdrv_pread(bs->file, offset, &ext, sizeof(ext)); 230 if (ret < 0) { 231 error_setg_errno(errp, -ret, "qcow2_read_extension: ERROR: " 232 "pread fail from offset %" PRIu64, offset); 233 return 1; 234 } 235 ext.magic = be32_to_cpu(ext.magic); 236 ext.len = be32_to_cpu(ext.len); 237 offset += sizeof(ext); 238 #ifdef DEBUG_EXT 239 printf("ext.magic = 0x%x\n", ext.magic); 240 #endif 241 if (offset > end_offset || ext.len > end_offset - offset) { 242 error_setg(errp, "Header extension too large"); 243 return -EINVAL; 244 } 245 246 switch (ext.magic) { 247 case QCOW2_EXT_MAGIC_END: 248 return 0; 249 250 case QCOW2_EXT_MAGIC_BACKING_FORMAT: 251 if (ext.len >= sizeof(bs->backing_format)) { 252 error_setg(errp, "ERROR: ext_backing_format: len=%" PRIu32 253 " too large (>=%zu)", ext.len, 254 sizeof(bs->backing_format)); 255 return 2; 256 } 257 ret = bdrv_pread(bs->file, offset, bs->backing_format, ext.len); 258 if (ret < 0) { 259 error_setg_errno(errp, -ret, "ERROR: ext_backing_format: " 260 "Could not read format name"); 261 return 3; 262 } 263 bs->backing_format[ext.len] = '\0'; 264 s->image_backing_format = g_strdup(bs->backing_format); 265 #ifdef DEBUG_EXT 266 printf("Qcow2: Got format extension %s\n", bs->backing_format); 267 #endif 268 break; 269 270 case QCOW2_EXT_MAGIC_FEATURE_TABLE: 271 if (p_feature_table != NULL) { 272 void* feature_table = g_malloc0(ext.len + 2 * sizeof(Qcow2Feature)); 273 ret = bdrv_pread(bs->file, offset , feature_table, ext.len); 274 if (ret < 0) { 275 error_setg_errno(errp, -ret, "ERROR: ext_feature_table: " 276 "Could not read table"); 277 return ret; 278 } 279 280 *p_feature_table = feature_table; 281 } 282 break; 283 284 case QCOW2_EXT_MAGIC_CRYPTO_HEADER: { 285 unsigned int cflags = 0; 286 if (s->crypt_method_header != QCOW_CRYPT_LUKS) { 287 error_setg(errp, "CRYPTO header extension only " 288 "expected with LUKS encryption method"); 289 return -EINVAL; 290 } 291 if (ext.len != sizeof(Qcow2CryptoHeaderExtension)) { 292 error_setg(errp, "CRYPTO header extension size %u, " 293 "but expected size %zu", ext.len, 294 sizeof(Qcow2CryptoHeaderExtension)); 295 return -EINVAL; 296 } 297 298 ret = bdrv_pread(bs->file, offset, &s->crypto_header, ext.len); 299 if (ret < 0) { 300 error_setg_errno(errp, -ret, 301 "Unable to read CRYPTO header extension"); 302 return ret; 303 } 304 s->crypto_header.offset = be64_to_cpu(s->crypto_header.offset); 305 s->crypto_header.length = be64_to_cpu(s->crypto_header.length); 306 307 if ((s->crypto_header.offset % s->cluster_size) != 0) { 308 error_setg(errp, "Encryption header offset '%" PRIu64 "' is " 309 "not a multiple of cluster size '%u'", 310 s->crypto_header.offset, s->cluster_size); 311 return -EINVAL; 312 } 313 314 if (flags & BDRV_O_NO_IO) { 315 cflags |= QCRYPTO_BLOCK_OPEN_NO_IO; 316 } 317 s->crypto = qcrypto_block_open(s->crypto_opts, "encrypt.", 318 qcow2_crypto_hdr_read_func, 319 bs, cflags, QCOW2_MAX_THREADS, errp); 320 if (!s->crypto) { 321 return -EINVAL; 322 } 323 } break; 324 325 case QCOW2_EXT_MAGIC_BITMAPS: 326 if (ext.len != sizeof(bitmaps_ext)) { 327 error_setg_errno(errp, -ret, "bitmaps_ext: " 328 "Invalid extension length"); 329 return -EINVAL; 330 } 331 332 if (!(s->autoclear_features & QCOW2_AUTOCLEAR_BITMAPS)) { 333 if (s->qcow_version < 3) { 334 /* Let's be a bit more specific */ 335 warn_report("This qcow2 v2 image contains bitmaps, but " 336 "they may have been modified by a program " 337 "without persistent bitmap support; so now " 338 "they must all be considered inconsistent"); 339 } else { 340 warn_report("a program lacking bitmap support " 341 "modified this file, so all bitmaps are now " 342 "considered inconsistent"); 343 } 344 error_printf("Some clusters may be leaked, " 345 "run 'qemu-img check -r' on the image " 346 "file to fix."); 347 if (need_update_header != NULL) { 348 /* Updating is needed to drop invalid bitmap extension. */ 349 *need_update_header = true; 350 } 351 break; 352 } 353 354 ret = bdrv_pread(bs->file, offset, &bitmaps_ext, ext.len); 355 if (ret < 0) { 356 error_setg_errno(errp, -ret, "bitmaps_ext: " 357 "Could not read ext header"); 358 return ret; 359 } 360 361 if (bitmaps_ext.reserved32 != 0) { 362 error_setg_errno(errp, -ret, "bitmaps_ext: " 363 "Reserved field is not zero"); 364 return -EINVAL; 365 } 366 367 bitmaps_ext.nb_bitmaps = be32_to_cpu(bitmaps_ext.nb_bitmaps); 368 bitmaps_ext.bitmap_directory_size = 369 be64_to_cpu(bitmaps_ext.bitmap_directory_size); 370 bitmaps_ext.bitmap_directory_offset = 371 be64_to_cpu(bitmaps_ext.bitmap_directory_offset); 372 373 if (bitmaps_ext.nb_bitmaps > QCOW2_MAX_BITMAPS) { 374 error_setg(errp, 375 "bitmaps_ext: Image has %" PRIu32 " bitmaps, " 376 "exceeding the QEMU supported maximum of %d", 377 bitmaps_ext.nb_bitmaps, QCOW2_MAX_BITMAPS); 378 return -EINVAL; 379 } 380 381 if (bitmaps_ext.nb_bitmaps == 0) { 382 error_setg(errp, "found bitmaps extension with zero bitmaps"); 383 return -EINVAL; 384 } 385 386 if (offset_into_cluster(s, bitmaps_ext.bitmap_directory_offset)) { 387 error_setg(errp, "bitmaps_ext: " 388 "invalid bitmap directory offset"); 389 return -EINVAL; 390 } 391 392 if (bitmaps_ext.bitmap_directory_size > 393 QCOW2_MAX_BITMAP_DIRECTORY_SIZE) { 394 error_setg(errp, "bitmaps_ext: " 395 "bitmap directory size (%" PRIu64 ") exceeds " 396 "the maximum supported size (%d)", 397 bitmaps_ext.bitmap_directory_size, 398 QCOW2_MAX_BITMAP_DIRECTORY_SIZE); 399 return -EINVAL; 400 } 401 402 s->nb_bitmaps = bitmaps_ext.nb_bitmaps; 403 s->bitmap_directory_offset = 404 bitmaps_ext.bitmap_directory_offset; 405 s->bitmap_directory_size = 406 bitmaps_ext.bitmap_directory_size; 407 408 #ifdef DEBUG_EXT 409 printf("Qcow2: Got bitmaps extension: " 410 "offset=%" PRIu64 " nb_bitmaps=%" PRIu32 "\n", 411 s->bitmap_directory_offset, s->nb_bitmaps); 412 #endif 413 break; 414 415 case QCOW2_EXT_MAGIC_DATA_FILE: 416 { 417 s->image_data_file = g_malloc0(ext.len + 1); 418 ret = bdrv_pread(bs->file, offset, s->image_data_file, ext.len); 419 if (ret < 0) { 420 error_setg_errno(errp, -ret, 421 "ERROR: Could not read data file name"); 422 return ret; 423 } 424 #ifdef DEBUG_EXT 425 printf("Qcow2: Got external data file %s\n", s->image_data_file); 426 #endif 427 break; 428 } 429 430 default: 431 /* unknown magic - save it in case we need to rewrite the header */ 432 /* If you add a new feature, make sure to also update the fast 433 * path of qcow2_make_empty() to deal with it. */ 434 { 435 Qcow2UnknownHeaderExtension *uext; 436 437 uext = g_malloc0(sizeof(*uext) + ext.len); 438 uext->magic = ext.magic; 439 uext->len = ext.len; 440 QLIST_INSERT_HEAD(&s->unknown_header_ext, uext, next); 441 442 ret = bdrv_pread(bs->file, offset , uext->data, uext->len); 443 if (ret < 0) { 444 error_setg_errno(errp, -ret, "ERROR: unknown extension: " 445 "Could not read data"); 446 return ret; 447 } 448 } 449 break; 450 } 451 452 offset += ((ext.len + 7) & ~7); 453 } 454 455 return 0; 456 } 457 458 static void cleanup_unknown_header_ext(BlockDriverState *bs) 459 { 460 BDRVQcow2State *s = bs->opaque; 461 Qcow2UnknownHeaderExtension *uext, *next; 462 463 QLIST_FOREACH_SAFE(uext, &s->unknown_header_ext, next, next) { 464 QLIST_REMOVE(uext, next); 465 g_free(uext); 466 } 467 } 468 469 static void report_unsupported_feature(Error **errp, Qcow2Feature *table, 470 uint64_t mask) 471 { 472 g_autoptr(GString) features = g_string_sized_new(60); 473 474 while (table && table->name[0] != '\0') { 475 if (table->type == QCOW2_FEAT_TYPE_INCOMPATIBLE) { 476 if (mask & (1ULL << table->bit)) { 477 if (features->len > 0) { 478 g_string_append(features, ", "); 479 } 480 g_string_append_printf(features, "%.46s", table->name); 481 mask &= ~(1ULL << table->bit); 482 } 483 } 484 table++; 485 } 486 487 if (mask) { 488 if (features->len > 0) { 489 g_string_append(features, ", "); 490 } 491 g_string_append_printf(features, 492 "Unknown incompatible feature: %" PRIx64, mask); 493 } 494 495 error_setg(errp, "Unsupported qcow2 feature(s): %s", features->str); 496 } 497 498 /* 499 * Sets the dirty bit and flushes afterwards if necessary. 500 * 501 * The incompatible_features bit is only set if the image file header was 502 * updated successfully. Therefore it is not required to check the return 503 * value of this function. 504 */ 505 int qcow2_mark_dirty(BlockDriverState *bs) 506 { 507 BDRVQcow2State *s = bs->opaque; 508 uint64_t val; 509 int ret; 510 511 assert(s->qcow_version >= 3); 512 513 if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) { 514 return 0; /* already dirty */ 515 } 516 517 val = cpu_to_be64(s->incompatible_features | QCOW2_INCOMPAT_DIRTY); 518 ret = bdrv_pwrite(bs->file, offsetof(QCowHeader, incompatible_features), 519 &val, sizeof(val)); 520 if (ret < 0) { 521 return ret; 522 } 523 ret = bdrv_flush(bs->file->bs); 524 if (ret < 0) { 525 return ret; 526 } 527 528 /* Only treat image as dirty if the header was updated successfully */ 529 s->incompatible_features |= QCOW2_INCOMPAT_DIRTY; 530 return 0; 531 } 532 533 /* 534 * Clears the dirty bit and flushes before if necessary. Only call this 535 * function when there are no pending requests, it does not guard against 536 * concurrent requests dirtying the image. 537 */ 538 static int qcow2_mark_clean(BlockDriverState *bs) 539 { 540 BDRVQcow2State *s = bs->opaque; 541 542 if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) { 543 int ret; 544 545 s->incompatible_features &= ~QCOW2_INCOMPAT_DIRTY; 546 547 ret = qcow2_flush_caches(bs); 548 if (ret < 0) { 549 return ret; 550 } 551 552 return qcow2_update_header(bs); 553 } 554 return 0; 555 } 556 557 /* 558 * Marks the image as corrupt. 559 */ 560 int qcow2_mark_corrupt(BlockDriverState *bs) 561 { 562 BDRVQcow2State *s = bs->opaque; 563 564 s->incompatible_features |= QCOW2_INCOMPAT_CORRUPT; 565 return qcow2_update_header(bs); 566 } 567 568 /* 569 * Marks the image as consistent, i.e., unsets the corrupt bit, and flushes 570 * before if necessary. 571 */ 572 int qcow2_mark_consistent(BlockDriverState *bs) 573 { 574 BDRVQcow2State *s = bs->opaque; 575 576 if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) { 577 int ret = qcow2_flush_caches(bs); 578 if (ret < 0) { 579 return ret; 580 } 581 582 s->incompatible_features &= ~QCOW2_INCOMPAT_CORRUPT; 583 return qcow2_update_header(bs); 584 } 585 return 0; 586 } 587 588 static void qcow2_add_check_result(BdrvCheckResult *out, 589 const BdrvCheckResult *src, 590 bool set_allocation_info) 591 { 592 out->corruptions += src->corruptions; 593 out->leaks += src->leaks; 594 out->check_errors += src->check_errors; 595 out->corruptions_fixed += src->corruptions_fixed; 596 out->leaks_fixed += src->leaks_fixed; 597 598 if (set_allocation_info) { 599 out->image_end_offset = src->image_end_offset; 600 out->bfi = src->bfi; 601 } 602 } 603 604 static int coroutine_fn qcow2_co_check_locked(BlockDriverState *bs, 605 BdrvCheckResult *result, 606 BdrvCheckMode fix) 607 { 608 BdrvCheckResult snapshot_res = {}; 609 BdrvCheckResult refcount_res = {}; 610 int ret; 611 612 memset(result, 0, sizeof(*result)); 613 614 ret = qcow2_check_read_snapshot_table(bs, &snapshot_res, fix); 615 if (ret < 0) { 616 qcow2_add_check_result(result, &snapshot_res, false); 617 return ret; 618 } 619 620 ret = qcow2_check_refcounts(bs, &refcount_res, fix); 621 qcow2_add_check_result(result, &refcount_res, true); 622 if (ret < 0) { 623 qcow2_add_check_result(result, &snapshot_res, false); 624 return ret; 625 } 626 627 ret = qcow2_check_fix_snapshot_table(bs, &snapshot_res, fix); 628 qcow2_add_check_result(result, &snapshot_res, false); 629 if (ret < 0) { 630 return ret; 631 } 632 633 if (fix && result->check_errors == 0 && result->corruptions == 0) { 634 ret = qcow2_mark_clean(bs); 635 if (ret < 0) { 636 return ret; 637 } 638 return qcow2_mark_consistent(bs); 639 } 640 return ret; 641 } 642 643 static int coroutine_fn qcow2_co_check(BlockDriverState *bs, 644 BdrvCheckResult *result, 645 BdrvCheckMode fix) 646 { 647 BDRVQcow2State *s = bs->opaque; 648 int ret; 649 650 qemu_co_mutex_lock(&s->lock); 651 ret = qcow2_co_check_locked(bs, result, fix); 652 qemu_co_mutex_unlock(&s->lock); 653 return ret; 654 } 655 656 int qcow2_validate_table(BlockDriverState *bs, uint64_t offset, 657 uint64_t entries, size_t entry_len, 658 int64_t max_size_bytes, const char *table_name, 659 Error **errp) 660 { 661 BDRVQcow2State *s = bs->opaque; 662 663 if (entries > max_size_bytes / entry_len) { 664 error_setg(errp, "%s too large", table_name); 665 return -EFBIG; 666 } 667 668 /* Use signed INT64_MAX as the maximum even for uint64_t header fields, 669 * because values will be passed to qemu functions taking int64_t. */ 670 if ((INT64_MAX - entries * entry_len < offset) || 671 (offset_into_cluster(s, offset) != 0)) { 672 error_setg(errp, "%s offset invalid", table_name); 673 return -EINVAL; 674 } 675 676 return 0; 677 } 678 679 static const char *const mutable_opts[] = { 680 QCOW2_OPT_LAZY_REFCOUNTS, 681 QCOW2_OPT_DISCARD_REQUEST, 682 QCOW2_OPT_DISCARD_SNAPSHOT, 683 QCOW2_OPT_DISCARD_OTHER, 684 QCOW2_OPT_OVERLAP, 685 QCOW2_OPT_OVERLAP_TEMPLATE, 686 QCOW2_OPT_OVERLAP_MAIN_HEADER, 687 QCOW2_OPT_OVERLAP_ACTIVE_L1, 688 QCOW2_OPT_OVERLAP_ACTIVE_L2, 689 QCOW2_OPT_OVERLAP_REFCOUNT_TABLE, 690 QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK, 691 QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE, 692 QCOW2_OPT_OVERLAP_INACTIVE_L1, 693 QCOW2_OPT_OVERLAP_INACTIVE_L2, 694 QCOW2_OPT_OVERLAP_BITMAP_DIRECTORY, 695 QCOW2_OPT_CACHE_SIZE, 696 QCOW2_OPT_L2_CACHE_SIZE, 697 QCOW2_OPT_L2_CACHE_ENTRY_SIZE, 698 QCOW2_OPT_REFCOUNT_CACHE_SIZE, 699 QCOW2_OPT_CACHE_CLEAN_INTERVAL, 700 NULL 701 }; 702 703 static QemuOptsList qcow2_runtime_opts = { 704 .name = "qcow2", 705 .head = QTAILQ_HEAD_INITIALIZER(qcow2_runtime_opts.head), 706 .desc = { 707 { 708 .name = QCOW2_OPT_LAZY_REFCOUNTS, 709 .type = QEMU_OPT_BOOL, 710 .help = "Postpone refcount updates", 711 }, 712 { 713 .name = QCOW2_OPT_DISCARD_REQUEST, 714 .type = QEMU_OPT_BOOL, 715 .help = "Pass guest discard requests to the layer below", 716 }, 717 { 718 .name = QCOW2_OPT_DISCARD_SNAPSHOT, 719 .type = QEMU_OPT_BOOL, 720 .help = "Generate discard requests when snapshot related space " 721 "is freed", 722 }, 723 { 724 .name = QCOW2_OPT_DISCARD_OTHER, 725 .type = QEMU_OPT_BOOL, 726 .help = "Generate discard requests when other clusters are freed", 727 }, 728 { 729 .name = QCOW2_OPT_OVERLAP, 730 .type = QEMU_OPT_STRING, 731 .help = "Selects which overlap checks to perform from a range of " 732 "templates (none, constant, cached, all)", 733 }, 734 { 735 .name = QCOW2_OPT_OVERLAP_TEMPLATE, 736 .type = QEMU_OPT_STRING, 737 .help = "Selects which overlap checks to perform from a range of " 738 "templates (none, constant, cached, all)", 739 }, 740 { 741 .name = QCOW2_OPT_OVERLAP_MAIN_HEADER, 742 .type = QEMU_OPT_BOOL, 743 .help = "Check for unintended writes into the main qcow2 header", 744 }, 745 { 746 .name = QCOW2_OPT_OVERLAP_ACTIVE_L1, 747 .type = QEMU_OPT_BOOL, 748 .help = "Check for unintended writes into the active L1 table", 749 }, 750 { 751 .name = QCOW2_OPT_OVERLAP_ACTIVE_L2, 752 .type = QEMU_OPT_BOOL, 753 .help = "Check for unintended writes into an active L2 table", 754 }, 755 { 756 .name = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE, 757 .type = QEMU_OPT_BOOL, 758 .help = "Check for unintended writes into the refcount table", 759 }, 760 { 761 .name = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK, 762 .type = QEMU_OPT_BOOL, 763 .help = "Check for unintended writes into a refcount block", 764 }, 765 { 766 .name = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE, 767 .type = QEMU_OPT_BOOL, 768 .help = "Check for unintended writes into the snapshot table", 769 }, 770 { 771 .name = QCOW2_OPT_OVERLAP_INACTIVE_L1, 772 .type = QEMU_OPT_BOOL, 773 .help = "Check for unintended writes into an inactive L1 table", 774 }, 775 { 776 .name = QCOW2_OPT_OVERLAP_INACTIVE_L2, 777 .type = QEMU_OPT_BOOL, 778 .help = "Check for unintended writes into an inactive L2 table", 779 }, 780 { 781 .name = QCOW2_OPT_OVERLAP_BITMAP_DIRECTORY, 782 .type = QEMU_OPT_BOOL, 783 .help = "Check for unintended writes into the bitmap directory", 784 }, 785 { 786 .name = QCOW2_OPT_CACHE_SIZE, 787 .type = QEMU_OPT_SIZE, 788 .help = "Maximum combined metadata (L2 tables and refcount blocks) " 789 "cache size", 790 }, 791 { 792 .name = QCOW2_OPT_L2_CACHE_SIZE, 793 .type = QEMU_OPT_SIZE, 794 .help = "Maximum L2 table cache size", 795 }, 796 { 797 .name = QCOW2_OPT_L2_CACHE_ENTRY_SIZE, 798 .type = QEMU_OPT_SIZE, 799 .help = "Size of each entry in the L2 cache", 800 }, 801 { 802 .name = QCOW2_OPT_REFCOUNT_CACHE_SIZE, 803 .type = QEMU_OPT_SIZE, 804 .help = "Maximum refcount block cache size", 805 }, 806 { 807 .name = QCOW2_OPT_CACHE_CLEAN_INTERVAL, 808 .type = QEMU_OPT_NUMBER, 809 .help = "Clean unused cache entries after this time (in seconds)", 810 }, 811 BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.", 812 "ID of secret providing qcow2 AES key or LUKS passphrase"), 813 { /* end of list */ } 814 }, 815 }; 816 817 static const char *overlap_bool_option_names[QCOW2_OL_MAX_BITNR] = { 818 [QCOW2_OL_MAIN_HEADER_BITNR] = QCOW2_OPT_OVERLAP_MAIN_HEADER, 819 [QCOW2_OL_ACTIVE_L1_BITNR] = QCOW2_OPT_OVERLAP_ACTIVE_L1, 820 [QCOW2_OL_ACTIVE_L2_BITNR] = QCOW2_OPT_OVERLAP_ACTIVE_L2, 821 [QCOW2_OL_REFCOUNT_TABLE_BITNR] = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE, 822 [QCOW2_OL_REFCOUNT_BLOCK_BITNR] = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK, 823 [QCOW2_OL_SNAPSHOT_TABLE_BITNR] = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE, 824 [QCOW2_OL_INACTIVE_L1_BITNR] = QCOW2_OPT_OVERLAP_INACTIVE_L1, 825 [QCOW2_OL_INACTIVE_L2_BITNR] = QCOW2_OPT_OVERLAP_INACTIVE_L2, 826 [QCOW2_OL_BITMAP_DIRECTORY_BITNR] = QCOW2_OPT_OVERLAP_BITMAP_DIRECTORY, 827 }; 828 829 static void cache_clean_timer_cb(void *opaque) 830 { 831 BlockDriverState *bs = opaque; 832 BDRVQcow2State *s = bs->opaque; 833 qcow2_cache_clean_unused(s->l2_table_cache); 834 qcow2_cache_clean_unused(s->refcount_block_cache); 835 timer_mod(s->cache_clean_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) + 836 (int64_t) s->cache_clean_interval * 1000); 837 } 838 839 static void cache_clean_timer_init(BlockDriverState *bs, AioContext *context) 840 { 841 BDRVQcow2State *s = bs->opaque; 842 if (s->cache_clean_interval > 0) { 843 s->cache_clean_timer = aio_timer_new(context, QEMU_CLOCK_VIRTUAL, 844 SCALE_MS, cache_clean_timer_cb, 845 bs); 846 timer_mod(s->cache_clean_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) + 847 (int64_t) s->cache_clean_interval * 1000); 848 } 849 } 850 851 static void cache_clean_timer_del(BlockDriverState *bs) 852 { 853 BDRVQcow2State *s = bs->opaque; 854 if (s->cache_clean_timer) { 855 timer_del(s->cache_clean_timer); 856 timer_free(s->cache_clean_timer); 857 s->cache_clean_timer = NULL; 858 } 859 } 860 861 static void qcow2_detach_aio_context(BlockDriverState *bs) 862 { 863 cache_clean_timer_del(bs); 864 } 865 866 static void qcow2_attach_aio_context(BlockDriverState *bs, 867 AioContext *new_context) 868 { 869 cache_clean_timer_init(bs, new_context); 870 } 871 872 static void read_cache_sizes(BlockDriverState *bs, QemuOpts *opts, 873 uint64_t *l2_cache_size, 874 uint64_t *l2_cache_entry_size, 875 uint64_t *refcount_cache_size, Error **errp) 876 { 877 BDRVQcow2State *s = bs->opaque; 878 uint64_t combined_cache_size, l2_cache_max_setting; 879 bool l2_cache_size_set, refcount_cache_size_set, combined_cache_size_set; 880 bool l2_cache_entry_size_set; 881 int min_refcount_cache = MIN_REFCOUNT_CACHE_SIZE * s->cluster_size; 882 uint64_t virtual_disk_size = bs->total_sectors * BDRV_SECTOR_SIZE; 883 uint64_t max_l2_entries = DIV_ROUND_UP(virtual_disk_size, s->cluster_size); 884 /* An L2 table is always one cluster in size so the max cache size 885 * should be a multiple of the cluster size. */ 886 uint64_t max_l2_cache = ROUND_UP(max_l2_entries * l2_entry_size(s), 887 s->cluster_size); 888 889 combined_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_CACHE_SIZE); 890 l2_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_L2_CACHE_SIZE); 891 refcount_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_REFCOUNT_CACHE_SIZE); 892 l2_cache_entry_size_set = qemu_opt_get(opts, QCOW2_OPT_L2_CACHE_ENTRY_SIZE); 893 894 combined_cache_size = qemu_opt_get_size(opts, QCOW2_OPT_CACHE_SIZE, 0); 895 l2_cache_max_setting = qemu_opt_get_size(opts, QCOW2_OPT_L2_CACHE_SIZE, 896 DEFAULT_L2_CACHE_MAX_SIZE); 897 *refcount_cache_size = qemu_opt_get_size(opts, 898 QCOW2_OPT_REFCOUNT_CACHE_SIZE, 0); 899 900 *l2_cache_entry_size = qemu_opt_get_size( 901 opts, QCOW2_OPT_L2_CACHE_ENTRY_SIZE, s->cluster_size); 902 903 *l2_cache_size = MIN(max_l2_cache, l2_cache_max_setting); 904 905 if (combined_cache_size_set) { 906 if (l2_cache_size_set && refcount_cache_size_set) { 907 error_setg(errp, QCOW2_OPT_CACHE_SIZE ", " QCOW2_OPT_L2_CACHE_SIZE 908 " and " QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not be set " 909 "at the same time"); 910 return; 911 } else if (l2_cache_size_set && 912 (l2_cache_max_setting > combined_cache_size)) { 913 error_setg(errp, QCOW2_OPT_L2_CACHE_SIZE " may not exceed " 914 QCOW2_OPT_CACHE_SIZE); 915 return; 916 } else if (*refcount_cache_size > combined_cache_size) { 917 error_setg(errp, QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not exceed " 918 QCOW2_OPT_CACHE_SIZE); 919 return; 920 } 921 922 if (l2_cache_size_set) { 923 *refcount_cache_size = combined_cache_size - *l2_cache_size; 924 } else if (refcount_cache_size_set) { 925 *l2_cache_size = combined_cache_size - *refcount_cache_size; 926 } else { 927 /* Assign as much memory as possible to the L2 cache, and 928 * use the remainder for the refcount cache */ 929 if (combined_cache_size >= max_l2_cache + min_refcount_cache) { 930 *l2_cache_size = max_l2_cache; 931 *refcount_cache_size = combined_cache_size - *l2_cache_size; 932 } else { 933 *refcount_cache_size = 934 MIN(combined_cache_size, min_refcount_cache); 935 *l2_cache_size = combined_cache_size - *refcount_cache_size; 936 } 937 } 938 } 939 940 /* 941 * If the L2 cache is not enough to cover the whole disk then 942 * default to 4KB entries. Smaller entries reduce the cost of 943 * loads and evictions and increase I/O performance. 944 */ 945 if (*l2_cache_size < max_l2_cache && !l2_cache_entry_size_set) { 946 *l2_cache_entry_size = MIN(s->cluster_size, 4096); 947 } 948 949 /* l2_cache_size and refcount_cache_size are ensured to have at least 950 * their minimum values in qcow2_update_options_prepare() */ 951 952 if (*l2_cache_entry_size < (1 << MIN_CLUSTER_BITS) || 953 *l2_cache_entry_size > s->cluster_size || 954 !is_power_of_2(*l2_cache_entry_size)) { 955 error_setg(errp, "L2 cache entry size must be a power of two " 956 "between %d and the cluster size (%d)", 957 1 << MIN_CLUSTER_BITS, s->cluster_size); 958 return; 959 } 960 } 961 962 typedef struct Qcow2ReopenState { 963 Qcow2Cache *l2_table_cache; 964 Qcow2Cache *refcount_block_cache; 965 int l2_slice_size; /* Number of entries in a slice of the L2 table */ 966 bool use_lazy_refcounts; 967 int overlap_check; 968 bool discard_passthrough[QCOW2_DISCARD_MAX]; 969 uint64_t cache_clean_interval; 970 QCryptoBlockOpenOptions *crypto_opts; /* Disk encryption runtime options */ 971 } Qcow2ReopenState; 972 973 static int qcow2_update_options_prepare(BlockDriverState *bs, 974 Qcow2ReopenState *r, 975 QDict *options, int flags, 976 Error **errp) 977 { 978 BDRVQcow2State *s = bs->opaque; 979 QemuOpts *opts = NULL; 980 const char *opt_overlap_check, *opt_overlap_check_template; 981 int overlap_check_template = 0; 982 uint64_t l2_cache_size, l2_cache_entry_size, refcount_cache_size; 983 int i; 984 const char *encryptfmt; 985 QDict *encryptopts = NULL; 986 Error *local_err = NULL; 987 int ret; 988 989 qdict_extract_subqdict(options, &encryptopts, "encrypt."); 990 encryptfmt = qdict_get_try_str(encryptopts, "format"); 991 992 opts = qemu_opts_create(&qcow2_runtime_opts, NULL, 0, &error_abort); 993 if (!qemu_opts_absorb_qdict(opts, options, errp)) { 994 ret = -EINVAL; 995 goto fail; 996 } 997 998 /* get L2 table/refcount block cache size from command line options */ 999 read_cache_sizes(bs, opts, &l2_cache_size, &l2_cache_entry_size, 1000 &refcount_cache_size, &local_err); 1001 if (local_err) { 1002 error_propagate(errp, local_err); 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 break; 1164 1165 case QCOW_CRYPT_LUKS: 1166 if (encryptfmt && !g_str_equal(encryptfmt, "luks")) { 1167 error_setg(errp, 1168 "Header reported 'luks' encryption format but " 1169 "options specify '%s'", encryptfmt); 1170 ret = -EINVAL; 1171 goto fail; 1172 } 1173 qdict_put_str(encryptopts, "format", "luks"); 1174 r->crypto_opts = block_crypto_open_opts_init(encryptopts, errp); 1175 break; 1176 1177 default: 1178 error_setg(errp, "Unsupported encryption method %d", 1179 s->crypt_method_header); 1180 break; 1181 } 1182 if (s->crypt_method_header != QCOW_CRYPT_NONE && !r->crypto_opts) { 1183 ret = -EINVAL; 1184 goto fail; 1185 } 1186 1187 ret = 0; 1188 fail: 1189 qobject_unref(encryptopts); 1190 qemu_opts_del(opts); 1191 opts = NULL; 1192 return ret; 1193 } 1194 1195 static void qcow2_update_options_commit(BlockDriverState *bs, 1196 Qcow2ReopenState *r) 1197 { 1198 BDRVQcow2State *s = bs->opaque; 1199 int i; 1200 1201 if (s->l2_table_cache) { 1202 qcow2_cache_destroy(s->l2_table_cache); 1203 } 1204 if (s->refcount_block_cache) { 1205 qcow2_cache_destroy(s->refcount_block_cache); 1206 } 1207 s->l2_table_cache = r->l2_table_cache; 1208 s->refcount_block_cache = r->refcount_block_cache; 1209 s->l2_slice_size = r->l2_slice_size; 1210 1211 s->overlap_check = r->overlap_check; 1212 s->use_lazy_refcounts = r->use_lazy_refcounts; 1213 1214 for (i = 0; i < QCOW2_DISCARD_MAX; i++) { 1215 s->discard_passthrough[i] = r->discard_passthrough[i]; 1216 } 1217 1218 if (s->cache_clean_interval != r->cache_clean_interval) { 1219 cache_clean_timer_del(bs); 1220 s->cache_clean_interval = r->cache_clean_interval; 1221 cache_clean_timer_init(bs, bdrv_get_aio_context(bs)); 1222 } 1223 1224 qapi_free_QCryptoBlockOpenOptions(s->crypto_opts); 1225 s->crypto_opts = r->crypto_opts; 1226 } 1227 1228 static void qcow2_update_options_abort(BlockDriverState *bs, 1229 Qcow2ReopenState *r) 1230 { 1231 if (r->l2_table_cache) { 1232 qcow2_cache_destroy(r->l2_table_cache); 1233 } 1234 if (r->refcount_block_cache) { 1235 qcow2_cache_destroy(r->refcount_block_cache); 1236 } 1237 qapi_free_QCryptoBlockOpenOptions(r->crypto_opts); 1238 } 1239 1240 static int qcow2_update_options(BlockDriverState *bs, QDict *options, 1241 int flags, Error **errp) 1242 { 1243 Qcow2ReopenState r = {}; 1244 int ret; 1245 1246 ret = qcow2_update_options_prepare(bs, &r, options, flags, errp); 1247 if (ret >= 0) { 1248 qcow2_update_options_commit(bs, &r); 1249 } else { 1250 qcow2_update_options_abort(bs, &r); 1251 } 1252 1253 return ret; 1254 } 1255 1256 static int validate_compression_type(BDRVQcow2State *s, Error **errp) 1257 { 1258 switch (s->compression_type) { 1259 case QCOW2_COMPRESSION_TYPE_ZLIB: 1260 #ifdef CONFIG_ZSTD 1261 case QCOW2_COMPRESSION_TYPE_ZSTD: 1262 #endif 1263 break; 1264 1265 default: 1266 error_setg(errp, "qcow2: unknown compression type: %u", 1267 s->compression_type); 1268 return -ENOTSUP; 1269 } 1270 1271 /* 1272 * if the compression type differs from QCOW2_COMPRESSION_TYPE_ZLIB 1273 * the incompatible feature flag must be set 1274 */ 1275 if (s->compression_type == QCOW2_COMPRESSION_TYPE_ZLIB) { 1276 if (s->incompatible_features & QCOW2_INCOMPAT_COMPRESSION) { 1277 error_setg(errp, "qcow2: Compression type incompatible feature " 1278 "bit must not be set"); 1279 return -EINVAL; 1280 } 1281 } else { 1282 if (!(s->incompatible_features & QCOW2_INCOMPAT_COMPRESSION)) { 1283 error_setg(errp, "qcow2: Compression type incompatible feature " 1284 "bit must be set"); 1285 return -EINVAL; 1286 } 1287 } 1288 1289 return 0; 1290 } 1291 1292 /* Called with s->lock held. */ 1293 static int coroutine_fn qcow2_do_open(BlockDriverState *bs, QDict *options, 1294 int flags, Error **errp) 1295 { 1296 BDRVQcow2State *s = bs->opaque; 1297 unsigned int len, i; 1298 int ret = 0; 1299 QCowHeader header; 1300 Error *local_err = NULL; 1301 uint64_t ext_end; 1302 uint64_t l1_vm_state_index; 1303 bool update_header = false; 1304 1305 ret = bdrv_pread(bs->file, 0, &header, sizeof(header)); 1306 if (ret < 0) { 1307 error_setg_errno(errp, -ret, "Could not read qcow2 header"); 1308 goto fail; 1309 } 1310 header.magic = be32_to_cpu(header.magic); 1311 header.version = be32_to_cpu(header.version); 1312 header.backing_file_offset = be64_to_cpu(header.backing_file_offset); 1313 header.backing_file_size = be32_to_cpu(header.backing_file_size); 1314 header.size = be64_to_cpu(header.size); 1315 header.cluster_bits = be32_to_cpu(header.cluster_bits); 1316 header.crypt_method = be32_to_cpu(header.crypt_method); 1317 header.l1_table_offset = be64_to_cpu(header.l1_table_offset); 1318 header.l1_size = be32_to_cpu(header.l1_size); 1319 header.refcount_table_offset = be64_to_cpu(header.refcount_table_offset); 1320 header.refcount_table_clusters = 1321 be32_to_cpu(header.refcount_table_clusters); 1322 header.snapshots_offset = be64_to_cpu(header.snapshots_offset); 1323 header.nb_snapshots = be32_to_cpu(header.nb_snapshots); 1324 1325 if (header.magic != QCOW_MAGIC) { 1326 error_setg(errp, "Image is not in qcow2 format"); 1327 ret = -EINVAL; 1328 goto fail; 1329 } 1330 if (header.version < 2 || header.version > 3) { 1331 error_setg(errp, "Unsupported qcow2 version %" PRIu32, header.version); 1332 ret = -ENOTSUP; 1333 goto fail; 1334 } 1335 1336 s->qcow_version = header.version; 1337 1338 /* Initialise cluster size */ 1339 if (header.cluster_bits < MIN_CLUSTER_BITS || 1340 header.cluster_bits > MAX_CLUSTER_BITS) { 1341 error_setg(errp, "Unsupported cluster size: 2^%" PRIu32, 1342 header.cluster_bits); 1343 ret = -EINVAL; 1344 goto fail; 1345 } 1346 1347 s->cluster_bits = header.cluster_bits; 1348 s->cluster_size = 1 << s->cluster_bits; 1349 1350 /* Initialise version 3 header fields */ 1351 if (header.version == 2) { 1352 header.incompatible_features = 0; 1353 header.compatible_features = 0; 1354 header.autoclear_features = 0; 1355 header.refcount_order = 4; 1356 header.header_length = 72; 1357 } else { 1358 header.incompatible_features = 1359 be64_to_cpu(header.incompatible_features); 1360 header.compatible_features = be64_to_cpu(header.compatible_features); 1361 header.autoclear_features = be64_to_cpu(header.autoclear_features); 1362 header.refcount_order = be32_to_cpu(header.refcount_order); 1363 header.header_length = be32_to_cpu(header.header_length); 1364 1365 if (header.header_length < 104) { 1366 error_setg(errp, "qcow2 header too short"); 1367 ret = -EINVAL; 1368 goto fail; 1369 } 1370 } 1371 1372 if (header.header_length > s->cluster_size) { 1373 error_setg(errp, "qcow2 header exceeds cluster size"); 1374 ret = -EINVAL; 1375 goto fail; 1376 } 1377 1378 if (header.header_length > sizeof(header)) { 1379 s->unknown_header_fields_size = header.header_length - sizeof(header); 1380 s->unknown_header_fields = g_malloc(s->unknown_header_fields_size); 1381 ret = bdrv_pread(bs->file, sizeof(header), s->unknown_header_fields, 1382 s->unknown_header_fields_size); 1383 if (ret < 0) { 1384 error_setg_errno(errp, -ret, "Could not read unknown qcow2 header " 1385 "fields"); 1386 goto fail; 1387 } 1388 } 1389 1390 if (header.backing_file_offset > s->cluster_size) { 1391 error_setg(errp, "Invalid backing file offset"); 1392 ret = -EINVAL; 1393 goto fail; 1394 } 1395 1396 if (header.backing_file_offset) { 1397 ext_end = header.backing_file_offset; 1398 } else { 1399 ext_end = 1 << header.cluster_bits; 1400 } 1401 1402 /* Handle feature bits */ 1403 s->incompatible_features = header.incompatible_features; 1404 s->compatible_features = header.compatible_features; 1405 s->autoclear_features = header.autoclear_features; 1406 1407 /* 1408 * Handle compression type 1409 * Older qcow2 images don't contain the compression type header. 1410 * Distinguish them by the header length and use 1411 * the only valid (default) compression type in that case 1412 */ 1413 if (header.header_length > offsetof(QCowHeader, compression_type)) { 1414 s->compression_type = header.compression_type; 1415 } else { 1416 s->compression_type = QCOW2_COMPRESSION_TYPE_ZLIB; 1417 } 1418 1419 ret = validate_compression_type(s, errp); 1420 if (ret) { 1421 goto fail; 1422 } 1423 1424 if (s->incompatible_features & ~QCOW2_INCOMPAT_MASK) { 1425 void *feature_table = NULL; 1426 qcow2_read_extensions(bs, header.header_length, ext_end, 1427 &feature_table, flags, NULL, NULL); 1428 report_unsupported_feature(errp, feature_table, 1429 s->incompatible_features & 1430 ~QCOW2_INCOMPAT_MASK); 1431 ret = -ENOTSUP; 1432 g_free(feature_table); 1433 goto fail; 1434 } 1435 1436 if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) { 1437 /* Corrupt images may not be written to unless they are being repaired 1438 */ 1439 if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_CHECK)) { 1440 error_setg(errp, "qcow2: Image is corrupt; cannot be opened " 1441 "read/write"); 1442 ret = -EACCES; 1443 goto fail; 1444 } 1445 } 1446 1447 s->subclusters_per_cluster = 1448 has_subclusters(s) ? QCOW_EXTL2_SUBCLUSTERS_PER_CLUSTER : 1; 1449 s->subcluster_size = s->cluster_size / s->subclusters_per_cluster; 1450 s->subcluster_bits = ctz32(s->subcluster_size); 1451 1452 if (s->subcluster_size < (1 << MIN_CLUSTER_BITS)) { 1453 error_setg(errp, "Unsupported subcluster size: %d", s->subcluster_size); 1454 ret = -EINVAL; 1455 goto fail; 1456 } 1457 1458 /* Check support for various header values */ 1459 if (header.refcount_order > 6) { 1460 error_setg(errp, "Reference count entry width too large; may not " 1461 "exceed 64 bits"); 1462 ret = -EINVAL; 1463 goto fail; 1464 } 1465 s->refcount_order = header.refcount_order; 1466 s->refcount_bits = 1 << s->refcount_order; 1467 s->refcount_max = UINT64_C(1) << (s->refcount_bits - 1); 1468 s->refcount_max += s->refcount_max - 1; 1469 1470 s->crypt_method_header = header.crypt_method; 1471 if (s->crypt_method_header) { 1472 if (bdrv_uses_whitelist() && 1473 s->crypt_method_header == QCOW_CRYPT_AES) { 1474 error_setg(errp, 1475 "Use of AES-CBC encrypted qcow2 images is no longer " 1476 "supported in system emulators"); 1477 error_append_hint(errp, 1478 "You can use 'qemu-img convert' to convert your " 1479 "image to an alternative supported format, such " 1480 "as unencrypted qcow2, or raw with the LUKS " 1481 "format instead.\n"); 1482 ret = -ENOSYS; 1483 goto fail; 1484 } 1485 1486 if (s->crypt_method_header == QCOW_CRYPT_AES) { 1487 s->crypt_physical_offset = false; 1488 } else { 1489 /* Assuming LUKS and any future crypt methods we 1490 * add will all use physical offsets, due to the 1491 * fact that the alternative is insecure... */ 1492 s->crypt_physical_offset = true; 1493 } 1494 1495 bs->encrypted = true; 1496 } 1497 1498 s->l2_bits = s->cluster_bits - ctz32(l2_entry_size(s)); 1499 s->l2_size = 1 << s->l2_bits; 1500 /* 2^(s->refcount_order - 3) is the refcount width in bytes */ 1501 s->refcount_block_bits = s->cluster_bits - (s->refcount_order - 3); 1502 s->refcount_block_size = 1 << s->refcount_block_bits; 1503 bs->total_sectors = header.size / BDRV_SECTOR_SIZE; 1504 s->csize_shift = (62 - (s->cluster_bits - 8)); 1505 s->csize_mask = (1 << (s->cluster_bits - 8)) - 1; 1506 s->cluster_offset_mask = (1LL << s->csize_shift) - 1; 1507 1508 s->refcount_table_offset = header.refcount_table_offset; 1509 s->refcount_table_size = 1510 header.refcount_table_clusters << (s->cluster_bits - 3); 1511 1512 if (header.refcount_table_clusters == 0 && !(flags & BDRV_O_CHECK)) { 1513 error_setg(errp, "Image does not contain a reference count table"); 1514 ret = -EINVAL; 1515 goto fail; 1516 } 1517 1518 ret = qcow2_validate_table(bs, s->refcount_table_offset, 1519 header.refcount_table_clusters, 1520 s->cluster_size, QCOW_MAX_REFTABLE_SIZE, 1521 "Reference count table", errp); 1522 if (ret < 0) { 1523 goto fail; 1524 } 1525 1526 if (!(flags & BDRV_O_CHECK)) { 1527 /* 1528 * The total size in bytes of the snapshot table is checked in 1529 * qcow2_read_snapshots() because the size of each snapshot is 1530 * variable and we don't know it yet. 1531 * Here we only check the offset and number of snapshots. 1532 */ 1533 ret = qcow2_validate_table(bs, header.snapshots_offset, 1534 header.nb_snapshots, 1535 sizeof(QCowSnapshotHeader), 1536 sizeof(QCowSnapshotHeader) * 1537 QCOW_MAX_SNAPSHOTS, 1538 "Snapshot table", errp); 1539 if (ret < 0) { 1540 goto fail; 1541 } 1542 } 1543 1544 /* read the level 1 table */ 1545 ret = qcow2_validate_table(bs, header.l1_table_offset, 1546 header.l1_size, sizeof(uint64_t), 1547 QCOW_MAX_L1_SIZE, "Active L1 table", errp); 1548 if (ret < 0) { 1549 goto fail; 1550 } 1551 s->l1_size = header.l1_size; 1552 s->l1_table_offset = header.l1_table_offset; 1553 1554 l1_vm_state_index = size_to_l1(s, header.size); 1555 if (l1_vm_state_index > INT_MAX) { 1556 error_setg(errp, "Image is too big"); 1557 ret = -EFBIG; 1558 goto fail; 1559 } 1560 s->l1_vm_state_index = l1_vm_state_index; 1561 1562 /* the L1 table must contain at least enough entries to put 1563 header.size bytes */ 1564 if (s->l1_size < s->l1_vm_state_index) { 1565 error_setg(errp, "L1 table is too small"); 1566 ret = -EINVAL; 1567 goto fail; 1568 } 1569 1570 if (s->l1_size > 0) { 1571 s->l1_table = qemu_try_blockalign(bs->file->bs, 1572 s->l1_size * sizeof(uint64_t)); 1573 if (s->l1_table == NULL) { 1574 error_setg(errp, "Could not allocate L1 table"); 1575 ret = -ENOMEM; 1576 goto fail; 1577 } 1578 ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table, 1579 s->l1_size * sizeof(uint64_t)); 1580 if (ret < 0) { 1581 error_setg_errno(errp, -ret, "Could not read L1 table"); 1582 goto fail; 1583 } 1584 for(i = 0;i < s->l1_size; i++) { 1585 s->l1_table[i] = be64_to_cpu(s->l1_table[i]); 1586 } 1587 } 1588 1589 /* Parse driver-specific options */ 1590 ret = qcow2_update_options(bs, options, flags, errp); 1591 if (ret < 0) { 1592 goto fail; 1593 } 1594 1595 s->flags = flags; 1596 1597 ret = qcow2_refcount_init(bs); 1598 if (ret != 0) { 1599 error_setg_errno(errp, -ret, "Could not initialize refcount handling"); 1600 goto fail; 1601 } 1602 1603 QLIST_INIT(&s->cluster_allocs); 1604 QTAILQ_INIT(&s->discards); 1605 1606 /* read qcow2 extensions */ 1607 if (qcow2_read_extensions(bs, header.header_length, ext_end, NULL, 1608 flags, &update_header, errp)) { 1609 ret = -EINVAL; 1610 goto fail; 1611 } 1612 1613 /* Open external data file */ 1614 s->data_file = bdrv_open_child(NULL, options, "data-file", bs, 1615 &child_of_bds, BDRV_CHILD_DATA, 1616 true, &local_err); 1617 if (local_err) { 1618 error_propagate(errp, local_err); 1619 ret = -EINVAL; 1620 goto fail; 1621 } 1622 1623 if (s->incompatible_features & QCOW2_INCOMPAT_DATA_FILE) { 1624 if (!s->data_file && s->image_data_file) { 1625 s->data_file = bdrv_open_child(s->image_data_file, options, 1626 "data-file", bs, &child_of_bds, 1627 BDRV_CHILD_DATA, false, errp); 1628 if (!s->data_file) { 1629 ret = -EINVAL; 1630 goto fail; 1631 } 1632 } 1633 if (!s->data_file) { 1634 error_setg(errp, "'data-file' is required for this image"); 1635 ret = -EINVAL; 1636 goto fail; 1637 } 1638 1639 /* No data here */ 1640 bs->file->role &= ~BDRV_CHILD_DATA; 1641 1642 /* Must succeed because we have given up permissions if anything */ 1643 bdrv_child_refresh_perms(bs, bs->file, &error_abort); 1644 } else { 1645 if (s->data_file) { 1646 error_setg(errp, "'data-file' can only be set for images with an " 1647 "external data file"); 1648 ret = -EINVAL; 1649 goto fail; 1650 } 1651 1652 s->data_file = bs->file; 1653 1654 if (data_file_is_raw(bs)) { 1655 error_setg(errp, "data-file-raw requires a data file"); 1656 ret = -EINVAL; 1657 goto fail; 1658 } 1659 } 1660 1661 /* qcow2_read_extension may have set up the crypto context 1662 * if the crypt method needs a header region, some methods 1663 * don't need header extensions, so must check here 1664 */ 1665 if (s->crypt_method_header && !s->crypto) { 1666 if (s->crypt_method_header == QCOW_CRYPT_AES) { 1667 unsigned int cflags = 0; 1668 if (flags & BDRV_O_NO_IO) { 1669 cflags |= QCRYPTO_BLOCK_OPEN_NO_IO; 1670 } 1671 s->crypto = qcrypto_block_open(s->crypto_opts, "encrypt.", 1672 NULL, NULL, cflags, 1673 QCOW2_MAX_THREADS, errp); 1674 if (!s->crypto) { 1675 ret = -EINVAL; 1676 goto fail; 1677 } 1678 } else if (!(flags & BDRV_O_NO_IO)) { 1679 error_setg(errp, "Missing CRYPTO header for crypt method %d", 1680 s->crypt_method_header); 1681 ret = -EINVAL; 1682 goto fail; 1683 } 1684 } 1685 1686 /* read the backing file name */ 1687 if (header.backing_file_offset != 0) { 1688 len = header.backing_file_size; 1689 if (len > MIN(1023, s->cluster_size - header.backing_file_offset) || 1690 len >= sizeof(bs->backing_file)) { 1691 error_setg(errp, "Backing file name too long"); 1692 ret = -EINVAL; 1693 goto fail; 1694 } 1695 ret = bdrv_pread(bs->file, header.backing_file_offset, 1696 bs->auto_backing_file, len); 1697 if (ret < 0) { 1698 error_setg_errno(errp, -ret, "Could not read backing file name"); 1699 goto fail; 1700 } 1701 bs->auto_backing_file[len] = '\0'; 1702 pstrcpy(bs->backing_file, sizeof(bs->backing_file), 1703 bs->auto_backing_file); 1704 s->image_backing_file = g_strdup(bs->auto_backing_file); 1705 } 1706 1707 /* 1708 * Internal snapshots; skip reading them in check mode, because 1709 * we do not need them then, and we do not want to abort because 1710 * of a broken table. 1711 */ 1712 if (!(flags & BDRV_O_CHECK)) { 1713 s->snapshots_offset = header.snapshots_offset; 1714 s->nb_snapshots = header.nb_snapshots; 1715 1716 ret = qcow2_read_snapshots(bs, errp); 1717 if (ret < 0) { 1718 goto fail; 1719 } 1720 } 1721 1722 /* Clear unknown autoclear feature bits */ 1723 update_header |= s->autoclear_features & ~QCOW2_AUTOCLEAR_MASK; 1724 update_header = 1725 update_header && !bs->read_only && !(flags & BDRV_O_INACTIVE); 1726 if (update_header) { 1727 s->autoclear_features &= QCOW2_AUTOCLEAR_MASK; 1728 } 1729 1730 /* == Handle persistent dirty bitmaps == 1731 * 1732 * We want load dirty bitmaps in three cases: 1733 * 1734 * 1. Normal open of the disk in active mode, not related to invalidation 1735 * after migration. 1736 * 1737 * 2. Invalidation of the target vm after pre-copy phase of migration, if 1738 * bitmaps are _not_ migrating through migration channel, i.e. 1739 * 'dirty-bitmaps' capability is disabled. 1740 * 1741 * 3. Invalidation of source vm after failed or canceled migration. 1742 * This is a very interesting case. There are two possible types of 1743 * bitmaps: 1744 * 1745 * A. Stored on inactivation and removed. They should be loaded from the 1746 * image. 1747 * 1748 * B. Not stored: not-persistent bitmaps and bitmaps, migrated through 1749 * the migration channel (with dirty-bitmaps capability). 1750 * 1751 * On the other hand, there are two possible sub-cases: 1752 * 1753 * 3.1 disk was changed by somebody else while were inactive. In this 1754 * case all in-RAM dirty bitmaps (both persistent and not) are 1755 * definitely invalid. And we don't have any method to determine 1756 * this. 1757 * 1758 * Simple and safe thing is to just drop all the bitmaps of type B on 1759 * inactivation. But in this case we lose bitmaps in valid 4.2 case. 1760 * 1761 * On the other hand, resuming source vm, if disk was already changed 1762 * is a bad thing anyway: not only bitmaps, the whole vm state is 1763 * out of sync with disk. 1764 * 1765 * This means, that user or management tool, who for some reason 1766 * decided to resume source vm, after disk was already changed by 1767 * target vm, should at least drop all dirty bitmaps by hand. 1768 * 1769 * So, we can ignore this case for now, but TODO: "generation" 1770 * extension for qcow2, to determine, that image was changed after 1771 * last inactivation. And if it is changed, we will drop (or at least 1772 * mark as 'invalid' all the bitmaps of type B, both persistent 1773 * and not). 1774 * 1775 * 3.2 disk was _not_ changed while were inactive. Bitmaps may be saved 1776 * to disk ('dirty-bitmaps' capability disabled), or not saved 1777 * ('dirty-bitmaps' capability enabled), but we don't need to care 1778 * of: let's load bitmaps as always: stored bitmaps will be loaded, 1779 * and not stored has flag IN_USE=1 in the image and will be skipped 1780 * on loading. 1781 * 1782 * One remaining possible case when we don't want load bitmaps: 1783 * 1784 * 4. Open disk in inactive mode in target vm (bitmaps are migrating or 1785 * will be loaded on invalidation, no needs try loading them before) 1786 */ 1787 1788 if (!(bdrv_get_flags(bs) & BDRV_O_INACTIVE)) { 1789 /* It's case 1, 2 or 3.2. Or 3.1 which is BUG in management layer. */ 1790 bool header_updated = qcow2_load_dirty_bitmaps(bs, &local_err); 1791 if (local_err != NULL) { 1792 error_propagate(errp, local_err); 1793 ret = -EINVAL; 1794 goto fail; 1795 } 1796 1797 update_header = update_header && !header_updated; 1798 } 1799 1800 if (update_header) { 1801 ret = qcow2_update_header(bs); 1802 if (ret < 0) { 1803 error_setg_errno(errp, -ret, "Could not update qcow2 header"); 1804 goto fail; 1805 } 1806 } 1807 1808 bs->supported_zero_flags = header.version >= 3 ? 1809 BDRV_REQ_MAY_UNMAP | BDRV_REQ_NO_FALLBACK : 0; 1810 bs->supported_truncate_flags = BDRV_REQ_ZERO_WRITE; 1811 1812 /* Repair image if dirty */ 1813 if (!(flags & (BDRV_O_CHECK | BDRV_O_INACTIVE)) && !bs->read_only && 1814 (s->incompatible_features & QCOW2_INCOMPAT_DIRTY)) { 1815 BdrvCheckResult result = {0}; 1816 1817 ret = qcow2_co_check_locked(bs, &result, 1818 BDRV_FIX_ERRORS | BDRV_FIX_LEAKS); 1819 if (ret < 0 || result.check_errors) { 1820 if (ret >= 0) { 1821 ret = -EIO; 1822 } 1823 error_setg_errno(errp, -ret, "Could not repair dirty image"); 1824 goto fail; 1825 } 1826 } 1827 1828 #ifdef DEBUG_ALLOC 1829 { 1830 BdrvCheckResult result = {0}; 1831 qcow2_check_refcounts(bs, &result, 0); 1832 } 1833 #endif 1834 1835 qemu_co_queue_init(&s->thread_task_queue); 1836 1837 return ret; 1838 1839 fail: 1840 g_free(s->image_data_file); 1841 if (has_data_file(bs)) { 1842 bdrv_unref_child(bs, s->data_file); 1843 s->data_file = NULL; 1844 } 1845 g_free(s->unknown_header_fields); 1846 cleanup_unknown_header_ext(bs); 1847 qcow2_free_snapshots(bs); 1848 qcow2_refcount_close(bs); 1849 qemu_vfree(s->l1_table); 1850 /* else pre-write overlap checks in cache_destroy may crash */ 1851 s->l1_table = NULL; 1852 cache_clean_timer_del(bs); 1853 if (s->l2_table_cache) { 1854 qcow2_cache_destroy(s->l2_table_cache); 1855 } 1856 if (s->refcount_block_cache) { 1857 qcow2_cache_destroy(s->refcount_block_cache); 1858 } 1859 qcrypto_block_free(s->crypto); 1860 qapi_free_QCryptoBlockOpenOptions(s->crypto_opts); 1861 return ret; 1862 } 1863 1864 typedef struct QCow2OpenCo { 1865 BlockDriverState *bs; 1866 QDict *options; 1867 int flags; 1868 Error **errp; 1869 int ret; 1870 } QCow2OpenCo; 1871 1872 static void coroutine_fn qcow2_open_entry(void *opaque) 1873 { 1874 QCow2OpenCo *qoc = opaque; 1875 BDRVQcow2State *s = qoc->bs->opaque; 1876 1877 qemu_co_mutex_lock(&s->lock); 1878 qoc->ret = qcow2_do_open(qoc->bs, qoc->options, qoc->flags, qoc->errp); 1879 qemu_co_mutex_unlock(&s->lock); 1880 } 1881 1882 static int qcow2_open(BlockDriverState *bs, QDict *options, int flags, 1883 Error **errp) 1884 { 1885 BDRVQcow2State *s = bs->opaque; 1886 QCow2OpenCo qoc = { 1887 .bs = bs, 1888 .options = options, 1889 .flags = flags, 1890 .errp = errp, 1891 .ret = -EINPROGRESS 1892 }; 1893 1894 bs->file = bdrv_open_child(NULL, options, "file", bs, &child_of_bds, 1895 BDRV_CHILD_IMAGE, false, errp); 1896 if (!bs->file) { 1897 return -EINVAL; 1898 } 1899 1900 /* Initialise locks */ 1901 qemu_co_mutex_init(&s->lock); 1902 1903 if (qemu_in_coroutine()) { 1904 /* From bdrv_co_create. */ 1905 qcow2_open_entry(&qoc); 1906 } else { 1907 assert(qemu_get_current_aio_context() == qemu_get_aio_context()); 1908 qemu_coroutine_enter(qemu_coroutine_create(qcow2_open_entry, &qoc)); 1909 BDRV_POLL_WHILE(bs, qoc.ret == -EINPROGRESS); 1910 } 1911 return qoc.ret; 1912 } 1913 1914 static void qcow2_refresh_limits(BlockDriverState *bs, Error **errp) 1915 { 1916 BDRVQcow2State *s = bs->opaque; 1917 1918 if (bs->encrypted) { 1919 /* Encryption works on a sector granularity */ 1920 bs->bl.request_alignment = qcrypto_block_get_sector_size(s->crypto); 1921 } 1922 bs->bl.pwrite_zeroes_alignment = s->subcluster_size; 1923 bs->bl.pdiscard_alignment = s->cluster_size; 1924 } 1925 1926 static int qcow2_reopen_prepare(BDRVReopenState *state, 1927 BlockReopenQueue *queue, Error **errp) 1928 { 1929 Qcow2ReopenState *r; 1930 int ret; 1931 1932 r = g_new0(Qcow2ReopenState, 1); 1933 state->opaque = r; 1934 1935 ret = qcow2_update_options_prepare(state->bs, r, state->options, 1936 state->flags, errp); 1937 if (ret < 0) { 1938 goto fail; 1939 } 1940 1941 /* We need to write out any unwritten data if we reopen read-only. */ 1942 if ((state->flags & BDRV_O_RDWR) == 0) { 1943 ret = qcow2_reopen_bitmaps_ro(state->bs, errp); 1944 if (ret < 0) { 1945 goto fail; 1946 } 1947 1948 ret = bdrv_flush(state->bs); 1949 if (ret < 0) { 1950 goto fail; 1951 } 1952 1953 ret = qcow2_mark_clean(state->bs); 1954 if (ret < 0) { 1955 goto fail; 1956 } 1957 } 1958 1959 return 0; 1960 1961 fail: 1962 qcow2_update_options_abort(state->bs, r); 1963 g_free(r); 1964 return ret; 1965 } 1966 1967 static void qcow2_reopen_commit(BDRVReopenState *state) 1968 { 1969 qcow2_update_options_commit(state->bs, state->opaque); 1970 g_free(state->opaque); 1971 } 1972 1973 static void qcow2_reopen_commit_post(BDRVReopenState *state) 1974 { 1975 if (state->flags & BDRV_O_RDWR) { 1976 Error *local_err = NULL; 1977 1978 if (qcow2_reopen_bitmaps_rw(state->bs, &local_err) < 0) { 1979 /* 1980 * This is not fatal, bitmaps just left read-only, so all following 1981 * writes will fail. User can remove read-only bitmaps to unblock 1982 * writes or retry reopen. 1983 */ 1984 error_reportf_err(local_err, 1985 "%s: Failed to make dirty bitmaps writable: ", 1986 bdrv_get_node_name(state->bs)); 1987 } 1988 } 1989 } 1990 1991 static void qcow2_reopen_abort(BDRVReopenState *state) 1992 { 1993 qcow2_update_options_abort(state->bs, state->opaque); 1994 g_free(state->opaque); 1995 } 1996 1997 static void qcow2_join_options(QDict *options, QDict *old_options) 1998 { 1999 bool has_new_overlap_template = 2000 qdict_haskey(options, QCOW2_OPT_OVERLAP) || 2001 qdict_haskey(options, QCOW2_OPT_OVERLAP_TEMPLATE); 2002 bool has_new_total_cache_size = 2003 qdict_haskey(options, QCOW2_OPT_CACHE_SIZE); 2004 bool has_all_cache_options; 2005 2006 /* New overlap template overrides all old overlap options */ 2007 if (has_new_overlap_template) { 2008 qdict_del(old_options, QCOW2_OPT_OVERLAP); 2009 qdict_del(old_options, QCOW2_OPT_OVERLAP_TEMPLATE); 2010 qdict_del(old_options, QCOW2_OPT_OVERLAP_MAIN_HEADER); 2011 qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L1); 2012 qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L2); 2013 qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_TABLE); 2014 qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK); 2015 qdict_del(old_options, QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE); 2016 qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L1); 2017 qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L2); 2018 } 2019 2020 /* New total cache size overrides all old options */ 2021 if (qdict_haskey(options, QCOW2_OPT_CACHE_SIZE)) { 2022 qdict_del(old_options, QCOW2_OPT_L2_CACHE_SIZE); 2023 qdict_del(old_options, QCOW2_OPT_REFCOUNT_CACHE_SIZE); 2024 } 2025 2026 qdict_join(options, old_options, false); 2027 2028 /* 2029 * If after merging all cache size options are set, an old total size is 2030 * overwritten. Do keep all options, however, if all three are new. The 2031 * resulting error message is what we want to happen. 2032 */ 2033 has_all_cache_options = 2034 qdict_haskey(options, QCOW2_OPT_CACHE_SIZE) || 2035 qdict_haskey(options, QCOW2_OPT_L2_CACHE_SIZE) || 2036 qdict_haskey(options, QCOW2_OPT_REFCOUNT_CACHE_SIZE); 2037 2038 if (has_all_cache_options && !has_new_total_cache_size) { 2039 qdict_del(options, QCOW2_OPT_CACHE_SIZE); 2040 } 2041 } 2042 2043 static int coroutine_fn qcow2_co_block_status(BlockDriverState *bs, 2044 bool want_zero, 2045 int64_t offset, int64_t count, 2046 int64_t *pnum, int64_t *map, 2047 BlockDriverState **file) 2048 { 2049 BDRVQcow2State *s = bs->opaque; 2050 uint64_t host_offset; 2051 unsigned int bytes; 2052 QCow2SubclusterType type; 2053 int ret, status = 0; 2054 2055 qemu_co_mutex_lock(&s->lock); 2056 2057 if (!s->metadata_preallocation_checked) { 2058 ret = qcow2_detect_metadata_preallocation(bs); 2059 s->metadata_preallocation = (ret == 1); 2060 s->metadata_preallocation_checked = true; 2061 } 2062 2063 bytes = MIN(INT_MAX, count); 2064 ret = qcow2_get_host_offset(bs, offset, &bytes, &host_offset, &type); 2065 qemu_co_mutex_unlock(&s->lock); 2066 if (ret < 0) { 2067 return ret; 2068 } 2069 2070 *pnum = bytes; 2071 2072 if ((type == QCOW2_SUBCLUSTER_NORMAL || 2073 type == QCOW2_SUBCLUSTER_ZERO_ALLOC || 2074 type == QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC) && !s->crypto) { 2075 *map = host_offset; 2076 *file = s->data_file->bs; 2077 status |= BDRV_BLOCK_OFFSET_VALID; 2078 } 2079 if (type == QCOW2_SUBCLUSTER_ZERO_PLAIN || 2080 type == QCOW2_SUBCLUSTER_ZERO_ALLOC) { 2081 status |= BDRV_BLOCK_ZERO; 2082 } else if (type != QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN && 2083 type != QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC) { 2084 status |= BDRV_BLOCK_DATA; 2085 } 2086 if (s->metadata_preallocation && (status & BDRV_BLOCK_DATA) && 2087 (status & BDRV_BLOCK_OFFSET_VALID)) 2088 { 2089 status |= BDRV_BLOCK_RECURSE; 2090 } 2091 return status; 2092 } 2093 2094 static coroutine_fn int qcow2_handle_l2meta(BlockDriverState *bs, 2095 QCowL2Meta **pl2meta, 2096 bool link_l2) 2097 { 2098 int ret = 0; 2099 QCowL2Meta *l2meta = *pl2meta; 2100 2101 while (l2meta != NULL) { 2102 QCowL2Meta *next; 2103 2104 if (link_l2) { 2105 assert(!l2meta->prealloc); 2106 ret = qcow2_alloc_cluster_link_l2(bs, l2meta); 2107 if (ret) { 2108 goto out; 2109 } 2110 } else { 2111 qcow2_alloc_cluster_abort(bs, l2meta); 2112 } 2113 2114 /* Take the request off the list of running requests */ 2115 if (l2meta->nb_clusters != 0) { 2116 QLIST_REMOVE(l2meta, next_in_flight); 2117 } 2118 2119 qemu_co_queue_restart_all(&l2meta->dependent_requests); 2120 2121 next = l2meta->next; 2122 g_free(l2meta); 2123 l2meta = next; 2124 } 2125 out: 2126 *pl2meta = l2meta; 2127 return ret; 2128 } 2129 2130 static coroutine_fn int 2131 qcow2_co_preadv_encrypted(BlockDriverState *bs, 2132 uint64_t host_offset, 2133 uint64_t offset, 2134 uint64_t bytes, 2135 QEMUIOVector *qiov, 2136 uint64_t qiov_offset) 2137 { 2138 int ret; 2139 BDRVQcow2State *s = bs->opaque; 2140 uint8_t *buf; 2141 2142 assert(bs->encrypted && s->crypto); 2143 assert(bytes <= QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size); 2144 2145 /* 2146 * For encrypted images, read everything into a temporary 2147 * contiguous buffer on which the AES functions can work. 2148 * Also, decryption in a separate buffer is better as it 2149 * prevents the guest from learning information about the 2150 * encrypted nature of the virtual disk. 2151 */ 2152 2153 buf = qemu_try_blockalign(s->data_file->bs, bytes); 2154 if (buf == NULL) { 2155 return -ENOMEM; 2156 } 2157 2158 BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO); 2159 ret = bdrv_co_pread(s->data_file, host_offset, bytes, buf, 0); 2160 if (ret < 0) { 2161 goto fail; 2162 } 2163 2164 if (qcow2_co_decrypt(bs, host_offset, offset, buf, bytes) < 0) 2165 { 2166 ret = -EIO; 2167 goto fail; 2168 } 2169 qemu_iovec_from_buf(qiov, qiov_offset, buf, bytes); 2170 2171 fail: 2172 qemu_vfree(buf); 2173 2174 return ret; 2175 } 2176 2177 typedef struct Qcow2AioTask { 2178 AioTask task; 2179 2180 BlockDriverState *bs; 2181 QCow2SubclusterType subcluster_type; /* only for read */ 2182 uint64_t host_offset; /* or full descriptor in compressed clusters */ 2183 uint64_t offset; 2184 uint64_t bytes; 2185 QEMUIOVector *qiov; 2186 uint64_t qiov_offset; 2187 QCowL2Meta *l2meta; /* only for write */ 2188 } Qcow2AioTask; 2189 2190 static coroutine_fn int qcow2_co_preadv_task_entry(AioTask *task); 2191 static coroutine_fn int qcow2_add_task(BlockDriverState *bs, 2192 AioTaskPool *pool, 2193 AioTaskFunc func, 2194 QCow2SubclusterType subcluster_type, 2195 uint64_t host_offset, 2196 uint64_t offset, 2197 uint64_t bytes, 2198 QEMUIOVector *qiov, 2199 size_t qiov_offset, 2200 QCowL2Meta *l2meta) 2201 { 2202 Qcow2AioTask local_task; 2203 Qcow2AioTask *task = pool ? g_new(Qcow2AioTask, 1) : &local_task; 2204 2205 *task = (Qcow2AioTask) { 2206 .task.func = func, 2207 .bs = bs, 2208 .subcluster_type = subcluster_type, 2209 .qiov = qiov, 2210 .host_offset = host_offset, 2211 .offset = offset, 2212 .bytes = bytes, 2213 .qiov_offset = qiov_offset, 2214 .l2meta = l2meta, 2215 }; 2216 2217 trace_qcow2_add_task(qemu_coroutine_self(), bs, pool, 2218 func == qcow2_co_preadv_task_entry ? "read" : "write", 2219 subcluster_type, host_offset, offset, bytes, 2220 qiov, qiov_offset); 2221 2222 if (!pool) { 2223 return func(&task->task); 2224 } 2225 2226 aio_task_pool_start_task(pool, &task->task); 2227 2228 return 0; 2229 } 2230 2231 static coroutine_fn int qcow2_co_preadv_task(BlockDriverState *bs, 2232 QCow2SubclusterType subc_type, 2233 uint64_t host_offset, 2234 uint64_t offset, uint64_t bytes, 2235 QEMUIOVector *qiov, 2236 size_t qiov_offset) 2237 { 2238 BDRVQcow2State *s = bs->opaque; 2239 2240 switch (subc_type) { 2241 case QCOW2_SUBCLUSTER_ZERO_PLAIN: 2242 case QCOW2_SUBCLUSTER_ZERO_ALLOC: 2243 /* Both zero types are handled in qcow2_co_preadv_part */ 2244 g_assert_not_reached(); 2245 2246 case QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN: 2247 case QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC: 2248 assert(bs->backing); /* otherwise handled in qcow2_co_preadv_part */ 2249 2250 BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO); 2251 return bdrv_co_preadv_part(bs->backing, offset, bytes, 2252 qiov, qiov_offset, 0); 2253 2254 case QCOW2_SUBCLUSTER_COMPRESSED: 2255 return qcow2_co_preadv_compressed(bs, host_offset, 2256 offset, bytes, qiov, qiov_offset); 2257 2258 case QCOW2_SUBCLUSTER_NORMAL: 2259 if (bs->encrypted) { 2260 return qcow2_co_preadv_encrypted(bs, host_offset, 2261 offset, bytes, qiov, qiov_offset); 2262 } 2263 2264 BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO); 2265 return bdrv_co_preadv_part(s->data_file, host_offset, 2266 bytes, qiov, qiov_offset, 0); 2267 2268 default: 2269 g_assert_not_reached(); 2270 } 2271 2272 g_assert_not_reached(); 2273 } 2274 2275 static coroutine_fn int qcow2_co_preadv_task_entry(AioTask *task) 2276 { 2277 Qcow2AioTask *t = container_of(task, Qcow2AioTask, task); 2278 2279 assert(!t->l2meta); 2280 2281 return qcow2_co_preadv_task(t->bs, t->subcluster_type, 2282 t->host_offset, t->offset, t->bytes, 2283 t->qiov, t->qiov_offset); 2284 } 2285 2286 static coroutine_fn int qcow2_co_preadv_part(BlockDriverState *bs, 2287 uint64_t offset, uint64_t bytes, 2288 QEMUIOVector *qiov, 2289 size_t qiov_offset, int flags) 2290 { 2291 BDRVQcow2State *s = bs->opaque; 2292 int ret = 0; 2293 unsigned int cur_bytes; /* number of bytes in current iteration */ 2294 uint64_t host_offset = 0; 2295 QCow2SubclusterType type; 2296 AioTaskPool *aio = NULL; 2297 2298 while (bytes != 0 && aio_task_pool_status(aio) == 0) { 2299 /* prepare next request */ 2300 cur_bytes = MIN(bytes, INT_MAX); 2301 if (s->crypto) { 2302 cur_bytes = MIN(cur_bytes, 2303 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size); 2304 } 2305 2306 qemu_co_mutex_lock(&s->lock); 2307 ret = qcow2_get_host_offset(bs, offset, &cur_bytes, 2308 &host_offset, &type); 2309 qemu_co_mutex_unlock(&s->lock); 2310 if (ret < 0) { 2311 goto out; 2312 } 2313 2314 if (type == QCOW2_SUBCLUSTER_ZERO_PLAIN || 2315 type == QCOW2_SUBCLUSTER_ZERO_ALLOC || 2316 (type == QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN && !bs->backing) || 2317 (type == QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC && !bs->backing)) 2318 { 2319 qemu_iovec_memset(qiov, qiov_offset, 0, cur_bytes); 2320 } else { 2321 if (!aio && cur_bytes != bytes) { 2322 aio = aio_task_pool_new(QCOW2_MAX_WORKERS); 2323 } 2324 ret = qcow2_add_task(bs, aio, qcow2_co_preadv_task_entry, type, 2325 host_offset, offset, cur_bytes, 2326 qiov, qiov_offset, NULL); 2327 if (ret < 0) { 2328 goto out; 2329 } 2330 } 2331 2332 bytes -= cur_bytes; 2333 offset += cur_bytes; 2334 qiov_offset += cur_bytes; 2335 } 2336 2337 out: 2338 if (aio) { 2339 aio_task_pool_wait_all(aio); 2340 if (ret == 0) { 2341 ret = aio_task_pool_status(aio); 2342 } 2343 g_free(aio); 2344 } 2345 2346 return ret; 2347 } 2348 2349 /* Check if it's possible to merge a write request with the writing of 2350 * the data from the COW regions */ 2351 static bool merge_cow(uint64_t offset, unsigned bytes, 2352 QEMUIOVector *qiov, size_t qiov_offset, 2353 QCowL2Meta *l2meta) 2354 { 2355 QCowL2Meta *m; 2356 2357 for (m = l2meta; m != NULL; m = m->next) { 2358 /* If both COW regions are empty then there's nothing to merge */ 2359 if (m->cow_start.nb_bytes == 0 && m->cow_end.nb_bytes == 0) { 2360 continue; 2361 } 2362 2363 /* If COW regions are handled already, skip this too */ 2364 if (m->skip_cow) { 2365 continue; 2366 } 2367 2368 /* The data (middle) region must be immediately after the 2369 * start region */ 2370 if (l2meta_cow_start(m) + m->cow_start.nb_bytes != offset) { 2371 continue; 2372 } 2373 2374 /* The end region must be immediately after the data (middle) 2375 * region */ 2376 if (m->offset + m->cow_end.offset != offset + bytes) { 2377 continue; 2378 } 2379 2380 /* Make sure that adding both COW regions to the QEMUIOVector 2381 * does not exceed IOV_MAX */ 2382 if (qemu_iovec_subvec_niov(qiov, qiov_offset, bytes) > IOV_MAX - 2) { 2383 continue; 2384 } 2385 2386 m->data_qiov = qiov; 2387 m->data_qiov_offset = qiov_offset; 2388 return true; 2389 } 2390 2391 return false; 2392 } 2393 2394 static bool is_unallocated(BlockDriverState *bs, int64_t offset, int64_t bytes) 2395 { 2396 int64_t nr; 2397 return !bytes || 2398 (!bdrv_is_allocated_above(bs, NULL, false, offset, bytes, &nr) && 2399 nr == bytes); 2400 } 2401 2402 static bool is_zero_cow(BlockDriverState *bs, QCowL2Meta *m) 2403 { 2404 /* 2405 * This check is designed for optimization shortcut so it must be 2406 * efficient. 2407 * Instead of is_zero(), use is_unallocated() as it is faster (but not 2408 * as accurate and can result in false negatives). 2409 */ 2410 return is_unallocated(bs, m->offset + m->cow_start.offset, 2411 m->cow_start.nb_bytes) && 2412 is_unallocated(bs, m->offset + m->cow_end.offset, 2413 m->cow_end.nb_bytes); 2414 } 2415 2416 static int handle_alloc_space(BlockDriverState *bs, QCowL2Meta *l2meta) 2417 { 2418 BDRVQcow2State *s = bs->opaque; 2419 QCowL2Meta *m; 2420 2421 if (!(s->data_file->bs->supported_zero_flags & BDRV_REQ_NO_FALLBACK)) { 2422 return 0; 2423 } 2424 2425 if (bs->encrypted) { 2426 return 0; 2427 } 2428 2429 for (m = l2meta; m != NULL; m = m->next) { 2430 int ret; 2431 uint64_t start_offset = m->alloc_offset + m->cow_start.offset; 2432 unsigned nb_bytes = m->cow_end.offset + m->cow_end.nb_bytes - 2433 m->cow_start.offset; 2434 2435 if (!m->cow_start.nb_bytes && !m->cow_end.nb_bytes) { 2436 continue; 2437 } 2438 2439 if (!is_zero_cow(bs, m)) { 2440 continue; 2441 } 2442 2443 /* 2444 * instead of writing zero COW buffers, 2445 * efficiently zero out the whole clusters 2446 */ 2447 2448 ret = qcow2_pre_write_overlap_check(bs, 0, start_offset, nb_bytes, 2449 true); 2450 if (ret < 0) { 2451 return ret; 2452 } 2453 2454 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC_SPACE); 2455 ret = bdrv_co_pwrite_zeroes(s->data_file, start_offset, nb_bytes, 2456 BDRV_REQ_NO_FALLBACK); 2457 if (ret < 0) { 2458 if (ret != -ENOTSUP && ret != -EAGAIN) { 2459 return ret; 2460 } 2461 continue; 2462 } 2463 2464 trace_qcow2_skip_cow(qemu_coroutine_self(), m->offset, m->nb_clusters); 2465 m->skip_cow = true; 2466 } 2467 return 0; 2468 } 2469 2470 /* 2471 * qcow2_co_pwritev_task 2472 * Called with s->lock unlocked 2473 * l2meta - if not NULL, qcow2_co_pwritev_task() will consume it. Caller must 2474 * not use it somehow after qcow2_co_pwritev_task() call 2475 */ 2476 static coroutine_fn int qcow2_co_pwritev_task(BlockDriverState *bs, 2477 uint64_t host_offset, 2478 uint64_t offset, uint64_t bytes, 2479 QEMUIOVector *qiov, 2480 uint64_t qiov_offset, 2481 QCowL2Meta *l2meta) 2482 { 2483 int ret; 2484 BDRVQcow2State *s = bs->opaque; 2485 void *crypt_buf = NULL; 2486 QEMUIOVector encrypted_qiov; 2487 2488 if (bs->encrypted) { 2489 assert(s->crypto); 2490 assert(bytes <= QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size); 2491 crypt_buf = qemu_try_blockalign(bs->file->bs, bytes); 2492 if (crypt_buf == NULL) { 2493 ret = -ENOMEM; 2494 goto out_unlocked; 2495 } 2496 qemu_iovec_to_buf(qiov, qiov_offset, crypt_buf, bytes); 2497 2498 if (qcow2_co_encrypt(bs, host_offset, offset, crypt_buf, bytes) < 0) { 2499 ret = -EIO; 2500 goto out_unlocked; 2501 } 2502 2503 qemu_iovec_init_buf(&encrypted_qiov, crypt_buf, bytes); 2504 qiov = &encrypted_qiov; 2505 qiov_offset = 0; 2506 } 2507 2508 /* Try to efficiently initialize the physical space with zeroes */ 2509 ret = handle_alloc_space(bs, l2meta); 2510 if (ret < 0) { 2511 goto out_unlocked; 2512 } 2513 2514 /* 2515 * If we need to do COW, check if it's possible to merge the 2516 * writing of the guest data together with that of the COW regions. 2517 * If it's not possible (or not necessary) then write the 2518 * guest data now. 2519 */ 2520 if (!merge_cow(offset, bytes, qiov, qiov_offset, l2meta)) { 2521 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO); 2522 trace_qcow2_writev_data(qemu_coroutine_self(), host_offset); 2523 ret = bdrv_co_pwritev_part(s->data_file, host_offset, 2524 bytes, qiov, qiov_offset, 0); 2525 if (ret < 0) { 2526 goto out_unlocked; 2527 } 2528 } 2529 2530 qemu_co_mutex_lock(&s->lock); 2531 2532 ret = qcow2_handle_l2meta(bs, &l2meta, true); 2533 goto out_locked; 2534 2535 out_unlocked: 2536 qemu_co_mutex_lock(&s->lock); 2537 2538 out_locked: 2539 qcow2_handle_l2meta(bs, &l2meta, false); 2540 qemu_co_mutex_unlock(&s->lock); 2541 2542 qemu_vfree(crypt_buf); 2543 2544 return ret; 2545 } 2546 2547 static coroutine_fn int qcow2_co_pwritev_task_entry(AioTask *task) 2548 { 2549 Qcow2AioTask *t = container_of(task, Qcow2AioTask, task); 2550 2551 assert(!t->subcluster_type); 2552 2553 return qcow2_co_pwritev_task(t->bs, t->host_offset, 2554 t->offset, t->bytes, t->qiov, t->qiov_offset, 2555 t->l2meta); 2556 } 2557 2558 static coroutine_fn int qcow2_co_pwritev_part( 2559 BlockDriverState *bs, uint64_t offset, uint64_t bytes, 2560 QEMUIOVector *qiov, size_t qiov_offset, int flags) 2561 { 2562 BDRVQcow2State *s = bs->opaque; 2563 int offset_in_cluster; 2564 int ret; 2565 unsigned int cur_bytes; /* number of sectors in current iteration */ 2566 uint64_t cluster_offset; 2567 QCowL2Meta *l2meta = NULL; 2568 AioTaskPool *aio = NULL; 2569 2570 trace_qcow2_writev_start_req(qemu_coroutine_self(), offset, bytes); 2571 2572 while (bytes != 0 && aio_task_pool_status(aio) == 0) { 2573 2574 l2meta = NULL; 2575 2576 trace_qcow2_writev_start_part(qemu_coroutine_self()); 2577 offset_in_cluster = offset_into_cluster(s, offset); 2578 cur_bytes = MIN(bytes, INT_MAX); 2579 if (bs->encrypted) { 2580 cur_bytes = MIN(cur_bytes, 2581 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size 2582 - offset_in_cluster); 2583 } 2584 2585 qemu_co_mutex_lock(&s->lock); 2586 2587 ret = qcow2_alloc_cluster_offset(bs, offset, &cur_bytes, 2588 &cluster_offset, &l2meta); 2589 if (ret < 0) { 2590 goto out_locked; 2591 } 2592 2593 assert(offset_into_cluster(s, cluster_offset) == 0); 2594 2595 ret = qcow2_pre_write_overlap_check(bs, 0, 2596 cluster_offset + offset_in_cluster, 2597 cur_bytes, true); 2598 if (ret < 0) { 2599 goto out_locked; 2600 } 2601 2602 qemu_co_mutex_unlock(&s->lock); 2603 2604 if (!aio && cur_bytes != bytes) { 2605 aio = aio_task_pool_new(QCOW2_MAX_WORKERS); 2606 } 2607 ret = qcow2_add_task(bs, aio, qcow2_co_pwritev_task_entry, 0, 2608 cluster_offset + offset_in_cluster, offset, 2609 cur_bytes, qiov, qiov_offset, l2meta); 2610 l2meta = NULL; /* l2meta is consumed by qcow2_co_pwritev_task() */ 2611 if (ret < 0) { 2612 goto fail_nometa; 2613 } 2614 2615 bytes -= cur_bytes; 2616 offset += cur_bytes; 2617 qiov_offset += cur_bytes; 2618 trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_bytes); 2619 } 2620 ret = 0; 2621 2622 qemu_co_mutex_lock(&s->lock); 2623 2624 out_locked: 2625 qcow2_handle_l2meta(bs, &l2meta, false); 2626 2627 qemu_co_mutex_unlock(&s->lock); 2628 2629 fail_nometa: 2630 if (aio) { 2631 aio_task_pool_wait_all(aio); 2632 if (ret == 0) { 2633 ret = aio_task_pool_status(aio); 2634 } 2635 g_free(aio); 2636 } 2637 2638 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret); 2639 2640 return ret; 2641 } 2642 2643 static int qcow2_inactivate(BlockDriverState *bs) 2644 { 2645 BDRVQcow2State *s = bs->opaque; 2646 int ret, result = 0; 2647 Error *local_err = NULL; 2648 2649 qcow2_store_persistent_dirty_bitmaps(bs, true, &local_err); 2650 if (local_err != NULL) { 2651 result = -EINVAL; 2652 error_reportf_err(local_err, "Lost persistent bitmaps during " 2653 "inactivation of node '%s': ", 2654 bdrv_get_device_or_node_name(bs)); 2655 } 2656 2657 ret = qcow2_cache_flush(bs, s->l2_table_cache); 2658 if (ret) { 2659 result = ret; 2660 error_report("Failed to flush the L2 table cache: %s", 2661 strerror(-ret)); 2662 } 2663 2664 ret = qcow2_cache_flush(bs, s->refcount_block_cache); 2665 if (ret) { 2666 result = ret; 2667 error_report("Failed to flush the refcount block cache: %s", 2668 strerror(-ret)); 2669 } 2670 2671 if (result == 0) { 2672 qcow2_mark_clean(bs); 2673 } 2674 2675 return result; 2676 } 2677 2678 static void qcow2_close(BlockDriverState *bs) 2679 { 2680 BDRVQcow2State *s = bs->opaque; 2681 qemu_vfree(s->l1_table); 2682 /* else pre-write overlap checks in cache_destroy may crash */ 2683 s->l1_table = NULL; 2684 2685 if (!(s->flags & BDRV_O_INACTIVE)) { 2686 qcow2_inactivate(bs); 2687 } 2688 2689 cache_clean_timer_del(bs); 2690 qcow2_cache_destroy(s->l2_table_cache); 2691 qcow2_cache_destroy(s->refcount_block_cache); 2692 2693 qcrypto_block_free(s->crypto); 2694 s->crypto = NULL; 2695 qapi_free_QCryptoBlockOpenOptions(s->crypto_opts); 2696 2697 g_free(s->unknown_header_fields); 2698 cleanup_unknown_header_ext(bs); 2699 2700 g_free(s->image_data_file); 2701 g_free(s->image_backing_file); 2702 g_free(s->image_backing_format); 2703 2704 if (has_data_file(bs)) { 2705 bdrv_unref_child(bs, s->data_file); 2706 s->data_file = NULL; 2707 } 2708 2709 qcow2_refcount_close(bs); 2710 qcow2_free_snapshots(bs); 2711 } 2712 2713 static void coroutine_fn qcow2_co_invalidate_cache(BlockDriverState *bs, 2714 Error **errp) 2715 { 2716 BDRVQcow2State *s = bs->opaque; 2717 int flags = s->flags; 2718 QCryptoBlock *crypto = NULL; 2719 QDict *options; 2720 Error *local_err = NULL; 2721 int ret; 2722 2723 /* 2724 * Backing files are read-only which makes all of their metadata immutable, 2725 * that means we don't have to worry about reopening them here. 2726 */ 2727 2728 crypto = s->crypto; 2729 s->crypto = NULL; 2730 2731 qcow2_close(bs); 2732 2733 memset(s, 0, sizeof(BDRVQcow2State)); 2734 options = qdict_clone_shallow(bs->options); 2735 2736 flags &= ~BDRV_O_INACTIVE; 2737 qemu_co_mutex_lock(&s->lock); 2738 ret = qcow2_do_open(bs, options, flags, &local_err); 2739 qemu_co_mutex_unlock(&s->lock); 2740 qobject_unref(options); 2741 if (local_err) { 2742 error_propagate_prepend(errp, local_err, 2743 "Could not reopen qcow2 layer: "); 2744 bs->drv = NULL; 2745 return; 2746 } else if (ret < 0) { 2747 error_setg_errno(errp, -ret, "Could not reopen qcow2 layer"); 2748 bs->drv = NULL; 2749 return; 2750 } 2751 2752 s->crypto = crypto; 2753 } 2754 2755 static size_t header_ext_add(char *buf, uint32_t magic, const void *s, 2756 size_t len, size_t buflen) 2757 { 2758 QCowExtension *ext_backing_fmt = (QCowExtension*) buf; 2759 size_t ext_len = sizeof(QCowExtension) + ((len + 7) & ~7); 2760 2761 if (buflen < ext_len) { 2762 return -ENOSPC; 2763 } 2764 2765 *ext_backing_fmt = (QCowExtension) { 2766 .magic = cpu_to_be32(magic), 2767 .len = cpu_to_be32(len), 2768 }; 2769 2770 if (len) { 2771 memcpy(buf + sizeof(QCowExtension), s, len); 2772 } 2773 2774 return ext_len; 2775 } 2776 2777 /* 2778 * Updates the qcow2 header, including the variable length parts of it, i.e. 2779 * the backing file name and all extensions. qcow2 was not designed to allow 2780 * such changes, so if we run out of space (we can only use the first cluster) 2781 * this function may fail. 2782 * 2783 * Returns 0 on success, -errno in error cases. 2784 */ 2785 int qcow2_update_header(BlockDriverState *bs) 2786 { 2787 BDRVQcow2State *s = bs->opaque; 2788 QCowHeader *header; 2789 char *buf; 2790 size_t buflen = s->cluster_size; 2791 int ret; 2792 uint64_t total_size; 2793 uint32_t refcount_table_clusters; 2794 size_t header_length; 2795 Qcow2UnknownHeaderExtension *uext; 2796 2797 buf = qemu_blockalign(bs, buflen); 2798 2799 /* Header structure */ 2800 header = (QCowHeader*) buf; 2801 2802 if (buflen < sizeof(*header)) { 2803 ret = -ENOSPC; 2804 goto fail; 2805 } 2806 2807 header_length = sizeof(*header) + s->unknown_header_fields_size; 2808 total_size = bs->total_sectors * BDRV_SECTOR_SIZE; 2809 refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3); 2810 2811 ret = validate_compression_type(s, NULL); 2812 if (ret) { 2813 goto fail; 2814 } 2815 2816 *header = (QCowHeader) { 2817 /* Version 2 fields */ 2818 .magic = cpu_to_be32(QCOW_MAGIC), 2819 .version = cpu_to_be32(s->qcow_version), 2820 .backing_file_offset = 0, 2821 .backing_file_size = 0, 2822 .cluster_bits = cpu_to_be32(s->cluster_bits), 2823 .size = cpu_to_be64(total_size), 2824 .crypt_method = cpu_to_be32(s->crypt_method_header), 2825 .l1_size = cpu_to_be32(s->l1_size), 2826 .l1_table_offset = cpu_to_be64(s->l1_table_offset), 2827 .refcount_table_offset = cpu_to_be64(s->refcount_table_offset), 2828 .refcount_table_clusters = cpu_to_be32(refcount_table_clusters), 2829 .nb_snapshots = cpu_to_be32(s->nb_snapshots), 2830 .snapshots_offset = cpu_to_be64(s->snapshots_offset), 2831 2832 /* Version 3 fields */ 2833 .incompatible_features = cpu_to_be64(s->incompatible_features), 2834 .compatible_features = cpu_to_be64(s->compatible_features), 2835 .autoclear_features = cpu_to_be64(s->autoclear_features), 2836 .refcount_order = cpu_to_be32(s->refcount_order), 2837 .header_length = cpu_to_be32(header_length), 2838 .compression_type = s->compression_type, 2839 }; 2840 2841 /* For older versions, write a shorter header */ 2842 switch (s->qcow_version) { 2843 case 2: 2844 ret = offsetof(QCowHeader, incompatible_features); 2845 break; 2846 case 3: 2847 ret = sizeof(*header); 2848 break; 2849 default: 2850 ret = -EINVAL; 2851 goto fail; 2852 } 2853 2854 buf += ret; 2855 buflen -= ret; 2856 memset(buf, 0, buflen); 2857 2858 /* Preserve any unknown field in the header */ 2859 if (s->unknown_header_fields_size) { 2860 if (buflen < s->unknown_header_fields_size) { 2861 ret = -ENOSPC; 2862 goto fail; 2863 } 2864 2865 memcpy(buf, s->unknown_header_fields, s->unknown_header_fields_size); 2866 buf += s->unknown_header_fields_size; 2867 buflen -= s->unknown_header_fields_size; 2868 } 2869 2870 /* Backing file format header extension */ 2871 if (s->image_backing_format) { 2872 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BACKING_FORMAT, 2873 s->image_backing_format, 2874 strlen(s->image_backing_format), 2875 buflen); 2876 if (ret < 0) { 2877 goto fail; 2878 } 2879 2880 buf += ret; 2881 buflen -= ret; 2882 } 2883 2884 /* External data file header extension */ 2885 if (has_data_file(bs) && s->image_data_file) { 2886 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_DATA_FILE, 2887 s->image_data_file, strlen(s->image_data_file), 2888 buflen); 2889 if (ret < 0) { 2890 goto fail; 2891 } 2892 2893 buf += ret; 2894 buflen -= ret; 2895 } 2896 2897 /* Full disk encryption header pointer extension */ 2898 if (s->crypto_header.offset != 0) { 2899 s->crypto_header.offset = cpu_to_be64(s->crypto_header.offset); 2900 s->crypto_header.length = cpu_to_be64(s->crypto_header.length); 2901 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_CRYPTO_HEADER, 2902 &s->crypto_header, sizeof(s->crypto_header), 2903 buflen); 2904 s->crypto_header.offset = be64_to_cpu(s->crypto_header.offset); 2905 s->crypto_header.length = be64_to_cpu(s->crypto_header.length); 2906 if (ret < 0) { 2907 goto fail; 2908 } 2909 buf += ret; 2910 buflen -= ret; 2911 } 2912 2913 /* 2914 * Feature table. A mere 8 feature names occupies 392 bytes, and 2915 * when coupled with the v3 minimum header of 104 bytes plus the 2916 * 8-byte end-of-extension marker, that would leave only 8 bytes 2917 * for a backing file name in an image with 512-byte clusters. 2918 * Thus, we choose to omit this header for cluster sizes 4k and 2919 * smaller. 2920 */ 2921 if (s->qcow_version >= 3 && s->cluster_size > 4096) { 2922 static const Qcow2Feature features[] = { 2923 { 2924 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE, 2925 .bit = QCOW2_INCOMPAT_DIRTY_BITNR, 2926 .name = "dirty bit", 2927 }, 2928 { 2929 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE, 2930 .bit = QCOW2_INCOMPAT_CORRUPT_BITNR, 2931 .name = "corrupt bit", 2932 }, 2933 { 2934 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE, 2935 .bit = QCOW2_INCOMPAT_DATA_FILE_BITNR, 2936 .name = "external data file", 2937 }, 2938 { 2939 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE, 2940 .bit = QCOW2_INCOMPAT_COMPRESSION_BITNR, 2941 .name = "compression type", 2942 }, 2943 { 2944 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE, 2945 .bit = QCOW2_INCOMPAT_EXTL2_BITNR, 2946 .name = "extended L2 entries", 2947 }, 2948 { 2949 .type = QCOW2_FEAT_TYPE_COMPATIBLE, 2950 .bit = QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR, 2951 .name = "lazy refcounts", 2952 }, 2953 { 2954 .type = QCOW2_FEAT_TYPE_AUTOCLEAR, 2955 .bit = QCOW2_AUTOCLEAR_BITMAPS_BITNR, 2956 .name = "bitmaps", 2957 }, 2958 { 2959 .type = QCOW2_FEAT_TYPE_AUTOCLEAR, 2960 .bit = QCOW2_AUTOCLEAR_DATA_FILE_RAW_BITNR, 2961 .name = "raw external data", 2962 }, 2963 }; 2964 2965 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_FEATURE_TABLE, 2966 features, sizeof(features), buflen); 2967 if (ret < 0) { 2968 goto fail; 2969 } 2970 buf += ret; 2971 buflen -= ret; 2972 } 2973 2974 /* Bitmap extension */ 2975 if (s->nb_bitmaps > 0) { 2976 Qcow2BitmapHeaderExt bitmaps_header = { 2977 .nb_bitmaps = cpu_to_be32(s->nb_bitmaps), 2978 .bitmap_directory_size = 2979 cpu_to_be64(s->bitmap_directory_size), 2980 .bitmap_directory_offset = 2981 cpu_to_be64(s->bitmap_directory_offset) 2982 }; 2983 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BITMAPS, 2984 &bitmaps_header, sizeof(bitmaps_header), 2985 buflen); 2986 if (ret < 0) { 2987 goto fail; 2988 } 2989 buf += ret; 2990 buflen -= ret; 2991 } 2992 2993 /* Keep unknown header extensions */ 2994 QLIST_FOREACH(uext, &s->unknown_header_ext, next) { 2995 ret = header_ext_add(buf, uext->magic, uext->data, uext->len, buflen); 2996 if (ret < 0) { 2997 goto fail; 2998 } 2999 3000 buf += ret; 3001 buflen -= ret; 3002 } 3003 3004 /* End of header extensions */ 3005 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_END, NULL, 0, buflen); 3006 if (ret < 0) { 3007 goto fail; 3008 } 3009 3010 buf += ret; 3011 buflen -= ret; 3012 3013 /* Backing file name */ 3014 if (s->image_backing_file) { 3015 size_t backing_file_len = strlen(s->image_backing_file); 3016 3017 if (buflen < backing_file_len) { 3018 ret = -ENOSPC; 3019 goto fail; 3020 } 3021 3022 /* Using strncpy is ok here, since buf is not NUL-terminated. */ 3023 strncpy(buf, s->image_backing_file, buflen); 3024 3025 header->backing_file_offset = cpu_to_be64(buf - ((char*) header)); 3026 header->backing_file_size = cpu_to_be32(backing_file_len); 3027 } 3028 3029 /* Write the new header */ 3030 ret = bdrv_pwrite(bs->file, 0, header, s->cluster_size); 3031 if (ret < 0) { 3032 goto fail; 3033 } 3034 3035 ret = 0; 3036 fail: 3037 qemu_vfree(header); 3038 return ret; 3039 } 3040 3041 static int qcow2_change_backing_file(BlockDriverState *bs, 3042 const char *backing_file, const char *backing_fmt) 3043 { 3044 BDRVQcow2State *s = bs->opaque; 3045 3046 /* Adding a backing file means that the external data file alone won't be 3047 * enough to make sense of the content */ 3048 if (backing_file && data_file_is_raw(bs)) { 3049 return -EINVAL; 3050 } 3051 3052 if (backing_file && strlen(backing_file) > 1023) { 3053 return -EINVAL; 3054 } 3055 3056 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file), 3057 backing_file ?: ""); 3058 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: ""); 3059 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: ""); 3060 3061 g_free(s->image_backing_file); 3062 g_free(s->image_backing_format); 3063 3064 s->image_backing_file = backing_file ? g_strdup(bs->backing_file) : NULL; 3065 s->image_backing_format = backing_fmt ? g_strdup(bs->backing_format) : NULL; 3066 3067 return qcow2_update_header(bs); 3068 } 3069 3070 static int qcow2_set_up_encryption(BlockDriverState *bs, 3071 QCryptoBlockCreateOptions *cryptoopts, 3072 Error **errp) 3073 { 3074 BDRVQcow2State *s = bs->opaque; 3075 QCryptoBlock *crypto = NULL; 3076 int fmt, ret; 3077 3078 switch (cryptoopts->format) { 3079 case Q_CRYPTO_BLOCK_FORMAT_LUKS: 3080 fmt = QCOW_CRYPT_LUKS; 3081 break; 3082 case Q_CRYPTO_BLOCK_FORMAT_QCOW: 3083 fmt = QCOW_CRYPT_AES; 3084 break; 3085 default: 3086 error_setg(errp, "Crypto format not supported in qcow2"); 3087 return -EINVAL; 3088 } 3089 3090 s->crypt_method_header = fmt; 3091 3092 crypto = qcrypto_block_create(cryptoopts, "encrypt.", 3093 qcow2_crypto_hdr_init_func, 3094 qcow2_crypto_hdr_write_func, 3095 bs, errp); 3096 if (!crypto) { 3097 return -EINVAL; 3098 } 3099 3100 ret = qcow2_update_header(bs); 3101 if (ret < 0) { 3102 error_setg_errno(errp, -ret, "Could not write encryption header"); 3103 goto out; 3104 } 3105 3106 ret = 0; 3107 out: 3108 qcrypto_block_free(crypto); 3109 return ret; 3110 } 3111 3112 /** 3113 * Preallocates metadata structures for data clusters between @offset (in the 3114 * guest disk) and @new_length (which is thus generally the new guest disk 3115 * size). 3116 * 3117 * Returns: 0 on success, -errno on failure. 3118 */ 3119 static int coroutine_fn preallocate_co(BlockDriverState *bs, uint64_t offset, 3120 uint64_t new_length, PreallocMode mode, 3121 Error **errp) 3122 { 3123 BDRVQcow2State *s = bs->opaque; 3124 uint64_t bytes; 3125 uint64_t host_offset = 0; 3126 int64_t file_length; 3127 unsigned int cur_bytes; 3128 int ret; 3129 QCowL2Meta *meta; 3130 3131 assert(offset <= new_length); 3132 bytes = new_length - offset; 3133 3134 while (bytes) { 3135 cur_bytes = MIN(bytes, QEMU_ALIGN_DOWN(INT_MAX, s->cluster_size)); 3136 ret = qcow2_alloc_cluster_offset(bs, offset, &cur_bytes, 3137 &host_offset, &meta); 3138 if (ret < 0) { 3139 error_setg_errno(errp, -ret, "Allocating clusters failed"); 3140 return ret; 3141 } 3142 3143 while (meta) { 3144 QCowL2Meta *next = meta->next; 3145 meta->prealloc = true; 3146 3147 ret = qcow2_alloc_cluster_link_l2(bs, meta); 3148 if (ret < 0) { 3149 error_setg_errno(errp, -ret, "Mapping clusters failed"); 3150 qcow2_free_any_clusters(bs, meta->alloc_offset, 3151 meta->nb_clusters, QCOW2_DISCARD_NEVER); 3152 return ret; 3153 } 3154 3155 /* There are no dependent requests, but we need to remove our 3156 * request from the list of in-flight requests */ 3157 QLIST_REMOVE(meta, next_in_flight); 3158 3159 g_free(meta); 3160 meta = next; 3161 } 3162 3163 /* TODO Preallocate data if requested */ 3164 3165 bytes -= cur_bytes; 3166 offset += cur_bytes; 3167 } 3168 3169 /* 3170 * It is expected that the image file is large enough to actually contain 3171 * all of the allocated clusters (otherwise we get failing reads after 3172 * EOF). Extend the image to the last allocated sector. 3173 */ 3174 file_length = bdrv_getlength(s->data_file->bs); 3175 if (file_length < 0) { 3176 error_setg_errno(errp, -file_length, "Could not get file size"); 3177 return file_length; 3178 } 3179 3180 if (host_offset + cur_bytes > file_length) { 3181 if (mode == PREALLOC_MODE_METADATA) { 3182 mode = PREALLOC_MODE_OFF; 3183 } 3184 ret = bdrv_co_truncate(s->data_file, host_offset + cur_bytes, false, 3185 mode, 0, errp); 3186 if (ret < 0) { 3187 return ret; 3188 } 3189 } 3190 3191 return 0; 3192 } 3193 3194 /* qcow2_refcount_metadata_size: 3195 * @clusters: number of clusters to refcount (including data and L1/L2 tables) 3196 * @cluster_size: size of a cluster, in bytes 3197 * @refcount_order: refcount bits power-of-2 exponent 3198 * @generous_increase: allow for the refcount table to be 1.5x as large as it 3199 * needs to be 3200 * 3201 * Returns: Number of bytes required for refcount blocks and table metadata. 3202 */ 3203 int64_t qcow2_refcount_metadata_size(int64_t clusters, size_t cluster_size, 3204 int refcount_order, bool generous_increase, 3205 uint64_t *refblock_count) 3206 { 3207 /* 3208 * Every host cluster is reference-counted, including metadata (even 3209 * refcount metadata is recursively included). 3210 * 3211 * An accurate formula for the size of refcount metadata size is difficult 3212 * to derive. An easier method of calculation is finding the fixed point 3213 * where no further refcount blocks or table clusters are required to 3214 * reference count every cluster. 3215 */ 3216 int64_t blocks_per_table_cluster = cluster_size / sizeof(uint64_t); 3217 int64_t refcounts_per_block = cluster_size * 8 / (1 << refcount_order); 3218 int64_t table = 0; /* number of refcount table clusters */ 3219 int64_t blocks = 0; /* number of refcount block clusters */ 3220 int64_t last; 3221 int64_t n = 0; 3222 3223 do { 3224 last = n; 3225 blocks = DIV_ROUND_UP(clusters + table + blocks, refcounts_per_block); 3226 table = DIV_ROUND_UP(blocks, blocks_per_table_cluster); 3227 n = clusters + blocks + table; 3228 3229 if (n == last && generous_increase) { 3230 clusters += DIV_ROUND_UP(table, 2); 3231 n = 0; /* force another loop */ 3232 generous_increase = false; 3233 } 3234 } while (n != last); 3235 3236 if (refblock_count) { 3237 *refblock_count = blocks; 3238 } 3239 3240 return (blocks + table) * cluster_size; 3241 } 3242 3243 /** 3244 * qcow2_calc_prealloc_size: 3245 * @total_size: virtual disk size in bytes 3246 * @cluster_size: cluster size in bytes 3247 * @refcount_order: refcount bits power-of-2 exponent 3248 * @extended_l2: true if the image has extended L2 entries 3249 * 3250 * Returns: Total number of bytes required for the fully allocated image 3251 * (including metadata). 3252 */ 3253 static int64_t qcow2_calc_prealloc_size(int64_t total_size, 3254 size_t cluster_size, 3255 int refcount_order, 3256 bool extended_l2) 3257 { 3258 int64_t meta_size = 0; 3259 uint64_t nl1e, nl2e; 3260 int64_t aligned_total_size = ROUND_UP(total_size, cluster_size); 3261 size_t l2e_size = extended_l2 ? L2E_SIZE_EXTENDED : L2E_SIZE_NORMAL; 3262 3263 /* header: 1 cluster */ 3264 meta_size += cluster_size; 3265 3266 /* total size of L2 tables */ 3267 nl2e = aligned_total_size / cluster_size; 3268 nl2e = ROUND_UP(nl2e, cluster_size / l2e_size); 3269 meta_size += nl2e * l2e_size; 3270 3271 /* total size of L1 tables */ 3272 nl1e = nl2e * l2e_size / cluster_size; 3273 nl1e = ROUND_UP(nl1e, cluster_size / sizeof(uint64_t)); 3274 meta_size += nl1e * sizeof(uint64_t); 3275 3276 /* total size of refcount table and blocks */ 3277 meta_size += qcow2_refcount_metadata_size( 3278 (meta_size + aligned_total_size) / cluster_size, 3279 cluster_size, refcount_order, false, NULL); 3280 3281 return meta_size + aligned_total_size; 3282 } 3283 3284 static bool validate_cluster_size(size_t cluster_size, bool extended_l2, 3285 Error **errp) 3286 { 3287 int cluster_bits = ctz32(cluster_size); 3288 if (cluster_bits < MIN_CLUSTER_BITS || cluster_bits > MAX_CLUSTER_BITS || 3289 (1 << cluster_bits) != cluster_size) 3290 { 3291 error_setg(errp, "Cluster size must be a power of two between %d and " 3292 "%dk", 1 << MIN_CLUSTER_BITS, 1 << (MAX_CLUSTER_BITS - 10)); 3293 return false; 3294 } 3295 3296 if (extended_l2) { 3297 unsigned min_cluster_size = 3298 (1 << MIN_CLUSTER_BITS) * QCOW_EXTL2_SUBCLUSTERS_PER_CLUSTER; 3299 if (cluster_size < min_cluster_size) { 3300 error_setg(errp, "Extended L2 entries are only supported with " 3301 "cluster sizes of at least %u bytes", min_cluster_size); 3302 return false; 3303 } 3304 } 3305 3306 return true; 3307 } 3308 3309 static size_t qcow2_opt_get_cluster_size_del(QemuOpts *opts, bool extended_l2, 3310 Error **errp) 3311 { 3312 size_t cluster_size; 3313 3314 cluster_size = qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE, 3315 DEFAULT_CLUSTER_SIZE); 3316 if (!validate_cluster_size(cluster_size, extended_l2, errp)) { 3317 return 0; 3318 } 3319 return cluster_size; 3320 } 3321 3322 static int qcow2_opt_get_version_del(QemuOpts *opts, Error **errp) 3323 { 3324 char *buf; 3325 int ret; 3326 3327 buf = qemu_opt_get_del(opts, BLOCK_OPT_COMPAT_LEVEL); 3328 if (!buf) { 3329 ret = 3; /* default */ 3330 } else if (!strcmp(buf, "0.10")) { 3331 ret = 2; 3332 } else if (!strcmp(buf, "1.1")) { 3333 ret = 3; 3334 } else { 3335 error_setg(errp, "Invalid compatibility level: '%s'", buf); 3336 ret = -EINVAL; 3337 } 3338 g_free(buf); 3339 return ret; 3340 } 3341 3342 static uint64_t qcow2_opt_get_refcount_bits_del(QemuOpts *opts, int version, 3343 Error **errp) 3344 { 3345 uint64_t refcount_bits; 3346 3347 refcount_bits = qemu_opt_get_number_del(opts, BLOCK_OPT_REFCOUNT_BITS, 16); 3348 if (refcount_bits > 64 || !is_power_of_2(refcount_bits)) { 3349 error_setg(errp, "Refcount width must be a power of two and may not " 3350 "exceed 64 bits"); 3351 return 0; 3352 } 3353 3354 if (version < 3 && refcount_bits != 16) { 3355 error_setg(errp, "Different refcount widths than 16 bits require " 3356 "compatibility level 1.1 or above (use compat=1.1 or " 3357 "greater)"); 3358 return 0; 3359 } 3360 3361 return refcount_bits; 3362 } 3363 3364 static int coroutine_fn 3365 qcow2_co_create(BlockdevCreateOptions *create_options, Error **errp) 3366 { 3367 BlockdevCreateOptionsQcow2 *qcow2_opts; 3368 QDict *options; 3369 3370 /* 3371 * Open the image file and write a minimal qcow2 header. 3372 * 3373 * We keep things simple and start with a zero-sized image. We also 3374 * do without refcount blocks or a L1 table for now. We'll fix the 3375 * inconsistency later. 3376 * 3377 * We do need a refcount table because growing the refcount table means 3378 * allocating two new refcount blocks - the second of which would be at 3379 * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file 3380 * size for any qcow2 image. 3381 */ 3382 BlockBackend *blk = NULL; 3383 BlockDriverState *bs = NULL; 3384 BlockDriverState *data_bs = NULL; 3385 QCowHeader *header; 3386 size_t cluster_size; 3387 int version; 3388 int refcount_order; 3389 uint64_t* refcount_table; 3390 int ret; 3391 uint8_t compression_type = QCOW2_COMPRESSION_TYPE_ZLIB; 3392 3393 assert(create_options->driver == BLOCKDEV_DRIVER_QCOW2); 3394 qcow2_opts = &create_options->u.qcow2; 3395 3396 bs = bdrv_open_blockdev_ref(qcow2_opts->file, errp); 3397 if (bs == NULL) { 3398 return -EIO; 3399 } 3400 3401 /* Validate options and set default values */ 3402 if (!QEMU_IS_ALIGNED(qcow2_opts->size, BDRV_SECTOR_SIZE)) { 3403 error_setg(errp, "Image size must be a multiple of %u bytes", 3404 (unsigned) BDRV_SECTOR_SIZE); 3405 ret = -EINVAL; 3406 goto out; 3407 } 3408 3409 if (qcow2_opts->has_version) { 3410 switch (qcow2_opts->version) { 3411 case BLOCKDEV_QCOW2_VERSION_V2: 3412 version = 2; 3413 break; 3414 case BLOCKDEV_QCOW2_VERSION_V3: 3415 version = 3; 3416 break; 3417 default: 3418 g_assert_not_reached(); 3419 } 3420 } else { 3421 version = 3; 3422 } 3423 3424 if (qcow2_opts->has_cluster_size) { 3425 cluster_size = qcow2_opts->cluster_size; 3426 } else { 3427 cluster_size = DEFAULT_CLUSTER_SIZE; 3428 } 3429 3430 if (!qcow2_opts->has_extended_l2) { 3431 qcow2_opts->extended_l2 = false; 3432 } 3433 if (qcow2_opts->extended_l2) { 3434 if (version < 3) { 3435 error_setg(errp, "Extended L2 entries are only supported with " 3436 "compatibility level 1.1 and above (use version=v3 or " 3437 "greater)"); 3438 ret = -EINVAL; 3439 goto out; 3440 } 3441 } 3442 3443 if (!validate_cluster_size(cluster_size, qcow2_opts->extended_l2, errp)) { 3444 ret = -EINVAL; 3445 goto out; 3446 } 3447 3448 if (!qcow2_opts->has_preallocation) { 3449 qcow2_opts->preallocation = PREALLOC_MODE_OFF; 3450 } 3451 if (qcow2_opts->has_backing_file && 3452 qcow2_opts->preallocation != PREALLOC_MODE_OFF && 3453 !qcow2_opts->extended_l2) 3454 { 3455 error_setg(errp, "Backing file and preallocation can only be used at " 3456 "the same time if extended_l2 is on"); 3457 ret = -EINVAL; 3458 goto out; 3459 } 3460 if (qcow2_opts->has_backing_fmt && !qcow2_opts->has_backing_file) { 3461 error_setg(errp, "Backing format cannot be used without backing file"); 3462 ret = -EINVAL; 3463 goto out; 3464 } 3465 3466 if (!qcow2_opts->has_lazy_refcounts) { 3467 qcow2_opts->lazy_refcounts = false; 3468 } 3469 if (version < 3 && qcow2_opts->lazy_refcounts) { 3470 error_setg(errp, "Lazy refcounts only supported with compatibility " 3471 "level 1.1 and above (use version=v3 or greater)"); 3472 ret = -EINVAL; 3473 goto out; 3474 } 3475 3476 if (!qcow2_opts->has_refcount_bits) { 3477 qcow2_opts->refcount_bits = 16; 3478 } 3479 if (qcow2_opts->refcount_bits > 64 || 3480 !is_power_of_2(qcow2_opts->refcount_bits)) 3481 { 3482 error_setg(errp, "Refcount width must be a power of two and may not " 3483 "exceed 64 bits"); 3484 ret = -EINVAL; 3485 goto out; 3486 } 3487 if (version < 3 && qcow2_opts->refcount_bits != 16) { 3488 error_setg(errp, "Different refcount widths than 16 bits require " 3489 "compatibility level 1.1 or above (use version=v3 or " 3490 "greater)"); 3491 ret = -EINVAL; 3492 goto out; 3493 } 3494 refcount_order = ctz32(qcow2_opts->refcount_bits); 3495 3496 if (qcow2_opts->data_file_raw && !qcow2_opts->data_file) { 3497 error_setg(errp, "data-file-raw requires data-file"); 3498 ret = -EINVAL; 3499 goto out; 3500 } 3501 if (qcow2_opts->data_file_raw && qcow2_opts->has_backing_file) { 3502 error_setg(errp, "Backing file and data-file-raw cannot be used at " 3503 "the same time"); 3504 ret = -EINVAL; 3505 goto out; 3506 } 3507 3508 if (qcow2_opts->data_file) { 3509 if (version < 3) { 3510 error_setg(errp, "External data files are only supported with " 3511 "compatibility level 1.1 and above (use version=v3 or " 3512 "greater)"); 3513 ret = -EINVAL; 3514 goto out; 3515 } 3516 data_bs = bdrv_open_blockdev_ref(qcow2_opts->data_file, errp); 3517 if (data_bs == NULL) { 3518 ret = -EIO; 3519 goto out; 3520 } 3521 } 3522 3523 if (qcow2_opts->has_compression_type && 3524 qcow2_opts->compression_type != QCOW2_COMPRESSION_TYPE_ZLIB) { 3525 3526 ret = -EINVAL; 3527 3528 if (version < 3) { 3529 error_setg(errp, "Non-zlib compression type is only supported with " 3530 "compatibility level 1.1 and above (use version=v3 or " 3531 "greater)"); 3532 goto out; 3533 } 3534 3535 switch (qcow2_opts->compression_type) { 3536 #ifdef CONFIG_ZSTD 3537 case QCOW2_COMPRESSION_TYPE_ZSTD: 3538 break; 3539 #endif 3540 default: 3541 error_setg(errp, "Unknown compression type"); 3542 goto out; 3543 } 3544 3545 compression_type = qcow2_opts->compression_type; 3546 } 3547 3548 /* Create BlockBackend to write to the image */ 3549 blk = blk_new_with_bs(bs, BLK_PERM_WRITE | BLK_PERM_RESIZE, BLK_PERM_ALL, 3550 errp); 3551 if (!blk) { 3552 ret = -EPERM; 3553 goto out; 3554 } 3555 blk_set_allow_write_beyond_eof(blk, true); 3556 3557 /* Write the header */ 3558 QEMU_BUILD_BUG_ON((1 << MIN_CLUSTER_BITS) < sizeof(*header)); 3559 header = g_malloc0(cluster_size); 3560 *header = (QCowHeader) { 3561 .magic = cpu_to_be32(QCOW_MAGIC), 3562 .version = cpu_to_be32(version), 3563 .cluster_bits = cpu_to_be32(ctz32(cluster_size)), 3564 .size = cpu_to_be64(0), 3565 .l1_table_offset = cpu_to_be64(0), 3566 .l1_size = cpu_to_be32(0), 3567 .refcount_table_offset = cpu_to_be64(cluster_size), 3568 .refcount_table_clusters = cpu_to_be32(1), 3569 .refcount_order = cpu_to_be32(refcount_order), 3570 /* don't deal with endianness since compression_type is 1 byte long */ 3571 .compression_type = compression_type, 3572 .header_length = cpu_to_be32(sizeof(*header)), 3573 }; 3574 3575 /* We'll update this to correct value later */ 3576 header->crypt_method = cpu_to_be32(QCOW_CRYPT_NONE); 3577 3578 if (qcow2_opts->lazy_refcounts) { 3579 header->compatible_features |= 3580 cpu_to_be64(QCOW2_COMPAT_LAZY_REFCOUNTS); 3581 } 3582 if (data_bs) { 3583 header->incompatible_features |= 3584 cpu_to_be64(QCOW2_INCOMPAT_DATA_FILE); 3585 } 3586 if (qcow2_opts->data_file_raw) { 3587 header->autoclear_features |= 3588 cpu_to_be64(QCOW2_AUTOCLEAR_DATA_FILE_RAW); 3589 } 3590 if (compression_type != QCOW2_COMPRESSION_TYPE_ZLIB) { 3591 header->incompatible_features |= 3592 cpu_to_be64(QCOW2_INCOMPAT_COMPRESSION); 3593 } 3594 3595 if (qcow2_opts->extended_l2) { 3596 header->incompatible_features |= 3597 cpu_to_be64(QCOW2_INCOMPAT_EXTL2); 3598 } 3599 3600 ret = blk_pwrite(blk, 0, header, cluster_size, 0); 3601 g_free(header); 3602 if (ret < 0) { 3603 error_setg_errno(errp, -ret, "Could not write qcow2 header"); 3604 goto out; 3605 } 3606 3607 /* Write a refcount table with one refcount block */ 3608 refcount_table = g_malloc0(2 * cluster_size); 3609 refcount_table[0] = cpu_to_be64(2 * cluster_size); 3610 ret = blk_pwrite(blk, cluster_size, refcount_table, 2 * cluster_size, 0); 3611 g_free(refcount_table); 3612 3613 if (ret < 0) { 3614 error_setg_errno(errp, -ret, "Could not write refcount table"); 3615 goto out; 3616 } 3617 3618 blk_unref(blk); 3619 blk = NULL; 3620 3621 /* 3622 * And now open the image and make it consistent first (i.e. increase the 3623 * refcount of the cluster that is occupied by the header and the refcount 3624 * table) 3625 */ 3626 options = qdict_new(); 3627 qdict_put_str(options, "driver", "qcow2"); 3628 qdict_put_str(options, "file", bs->node_name); 3629 if (data_bs) { 3630 qdict_put_str(options, "data-file", data_bs->node_name); 3631 } 3632 blk = blk_new_open(NULL, NULL, options, 3633 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_NO_FLUSH, 3634 errp); 3635 if (blk == NULL) { 3636 ret = -EIO; 3637 goto out; 3638 } 3639 3640 ret = qcow2_alloc_clusters(blk_bs(blk), 3 * cluster_size); 3641 if (ret < 0) { 3642 error_setg_errno(errp, -ret, "Could not allocate clusters for qcow2 " 3643 "header and refcount table"); 3644 goto out; 3645 3646 } else if (ret != 0) { 3647 error_report("Huh, first cluster in empty image is already in use?"); 3648 abort(); 3649 } 3650 3651 /* Set the external data file if necessary */ 3652 if (data_bs) { 3653 BDRVQcow2State *s = blk_bs(blk)->opaque; 3654 s->image_data_file = g_strdup(data_bs->filename); 3655 } 3656 3657 /* Create a full header (including things like feature table) */ 3658 ret = qcow2_update_header(blk_bs(blk)); 3659 if (ret < 0) { 3660 error_setg_errno(errp, -ret, "Could not update qcow2 header"); 3661 goto out; 3662 } 3663 3664 /* Okay, now that we have a valid image, let's give it the right size */ 3665 ret = blk_truncate(blk, qcow2_opts->size, false, qcow2_opts->preallocation, 3666 0, errp); 3667 if (ret < 0) { 3668 error_prepend(errp, "Could not resize image: "); 3669 goto out; 3670 } 3671 3672 /* Want a backing file? There you go. */ 3673 if (qcow2_opts->has_backing_file) { 3674 const char *backing_format = NULL; 3675 3676 if (qcow2_opts->has_backing_fmt) { 3677 backing_format = BlockdevDriver_str(qcow2_opts->backing_fmt); 3678 } 3679 3680 ret = bdrv_change_backing_file(blk_bs(blk), qcow2_opts->backing_file, 3681 backing_format, false); 3682 if (ret < 0) { 3683 error_setg_errno(errp, -ret, "Could not assign backing file '%s' " 3684 "with format '%s'", qcow2_opts->backing_file, 3685 backing_format); 3686 goto out; 3687 } 3688 } 3689 3690 /* Want encryption? There you go. */ 3691 if (qcow2_opts->has_encrypt) { 3692 ret = qcow2_set_up_encryption(blk_bs(blk), qcow2_opts->encrypt, errp); 3693 if (ret < 0) { 3694 goto out; 3695 } 3696 } 3697 3698 blk_unref(blk); 3699 blk = NULL; 3700 3701 /* Reopen the image without BDRV_O_NO_FLUSH to flush it before returning. 3702 * Using BDRV_O_NO_IO, since encryption is now setup we don't want to 3703 * have to setup decryption context. We're not doing any I/O on the top 3704 * level BlockDriverState, only lower layers, where BDRV_O_NO_IO does 3705 * not have effect. 3706 */ 3707 options = qdict_new(); 3708 qdict_put_str(options, "driver", "qcow2"); 3709 qdict_put_str(options, "file", bs->node_name); 3710 if (data_bs) { 3711 qdict_put_str(options, "data-file", data_bs->node_name); 3712 } 3713 blk = blk_new_open(NULL, NULL, options, 3714 BDRV_O_RDWR | BDRV_O_NO_BACKING | BDRV_O_NO_IO, 3715 errp); 3716 if (blk == NULL) { 3717 ret = -EIO; 3718 goto out; 3719 } 3720 3721 ret = 0; 3722 out: 3723 blk_unref(blk); 3724 bdrv_unref(bs); 3725 bdrv_unref(data_bs); 3726 return ret; 3727 } 3728 3729 static int coroutine_fn qcow2_co_create_opts(BlockDriver *drv, 3730 const char *filename, 3731 QemuOpts *opts, 3732 Error **errp) 3733 { 3734 BlockdevCreateOptions *create_options = NULL; 3735 QDict *qdict; 3736 Visitor *v; 3737 BlockDriverState *bs = NULL; 3738 BlockDriverState *data_bs = NULL; 3739 const char *val; 3740 int ret; 3741 3742 /* Only the keyval visitor supports the dotted syntax needed for 3743 * encryption, so go through a QDict before getting a QAPI type. Ignore 3744 * options meant for the protocol layer so that the visitor doesn't 3745 * complain. */ 3746 qdict = qemu_opts_to_qdict_filtered(opts, NULL, bdrv_qcow2.create_opts, 3747 true); 3748 3749 /* Handle encryption options */ 3750 val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT); 3751 if (val && !strcmp(val, "on")) { 3752 qdict_put_str(qdict, BLOCK_OPT_ENCRYPT, "qcow"); 3753 } else if (val && !strcmp(val, "off")) { 3754 qdict_del(qdict, BLOCK_OPT_ENCRYPT); 3755 } 3756 3757 val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT); 3758 if (val && !strcmp(val, "aes")) { 3759 qdict_put_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT, "qcow"); 3760 } 3761 3762 /* Convert compat=0.10/1.1 into compat=v2/v3, to be renamed into 3763 * version=v2/v3 below. */ 3764 val = qdict_get_try_str(qdict, BLOCK_OPT_COMPAT_LEVEL); 3765 if (val && !strcmp(val, "0.10")) { 3766 qdict_put_str(qdict, BLOCK_OPT_COMPAT_LEVEL, "v2"); 3767 } else if (val && !strcmp(val, "1.1")) { 3768 qdict_put_str(qdict, BLOCK_OPT_COMPAT_LEVEL, "v3"); 3769 } 3770 3771 /* Change legacy command line options into QMP ones */ 3772 static const QDictRenames opt_renames[] = { 3773 { BLOCK_OPT_BACKING_FILE, "backing-file" }, 3774 { BLOCK_OPT_BACKING_FMT, "backing-fmt" }, 3775 { BLOCK_OPT_CLUSTER_SIZE, "cluster-size" }, 3776 { BLOCK_OPT_LAZY_REFCOUNTS, "lazy-refcounts" }, 3777 { BLOCK_OPT_EXTL2, "extended-l2" }, 3778 { BLOCK_OPT_REFCOUNT_BITS, "refcount-bits" }, 3779 { BLOCK_OPT_ENCRYPT, BLOCK_OPT_ENCRYPT_FORMAT }, 3780 { BLOCK_OPT_COMPAT_LEVEL, "version" }, 3781 { BLOCK_OPT_DATA_FILE_RAW, "data-file-raw" }, 3782 { BLOCK_OPT_COMPRESSION_TYPE, "compression-type" }, 3783 { NULL, NULL }, 3784 }; 3785 3786 if (!qdict_rename_keys(qdict, opt_renames, errp)) { 3787 ret = -EINVAL; 3788 goto finish; 3789 } 3790 3791 /* Create and open the file (protocol layer) */ 3792 ret = bdrv_create_file(filename, opts, errp); 3793 if (ret < 0) { 3794 goto finish; 3795 } 3796 3797 bs = bdrv_open(filename, NULL, NULL, 3798 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp); 3799 if (bs == NULL) { 3800 ret = -EIO; 3801 goto finish; 3802 } 3803 3804 /* Create and open an external data file (protocol layer) */ 3805 val = qdict_get_try_str(qdict, BLOCK_OPT_DATA_FILE); 3806 if (val) { 3807 ret = bdrv_create_file(val, opts, errp); 3808 if (ret < 0) { 3809 goto finish; 3810 } 3811 3812 data_bs = bdrv_open(val, NULL, NULL, 3813 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, 3814 errp); 3815 if (data_bs == NULL) { 3816 ret = -EIO; 3817 goto finish; 3818 } 3819 3820 qdict_del(qdict, BLOCK_OPT_DATA_FILE); 3821 qdict_put_str(qdict, "data-file", data_bs->node_name); 3822 } 3823 3824 /* Set 'driver' and 'node' options */ 3825 qdict_put_str(qdict, "driver", "qcow2"); 3826 qdict_put_str(qdict, "file", bs->node_name); 3827 3828 /* Now get the QAPI type BlockdevCreateOptions */ 3829 v = qobject_input_visitor_new_flat_confused(qdict, errp); 3830 if (!v) { 3831 ret = -EINVAL; 3832 goto finish; 3833 } 3834 3835 visit_type_BlockdevCreateOptions(v, NULL, &create_options, errp); 3836 visit_free(v); 3837 if (!create_options) { 3838 ret = -EINVAL; 3839 goto finish; 3840 } 3841 3842 /* Silently round up size */ 3843 create_options->u.qcow2.size = ROUND_UP(create_options->u.qcow2.size, 3844 BDRV_SECTOR_SIZE); 3845 3846 /* Create the qcow2 image (format layer) */ 3847 ret = qcow2_co_create(create_options, errp); 3848 if (ret < 0) { 3849 goto finish; 3850 } 3851 3852 ret = 0; 3853 finish: 3854 qobject_unref(qdict); 3855 bdrv_unref(bs); 3856 bdrv_unref(data_bs); 3857 qapi_free_BlockdevCreateOptions(create_options); 3858 return ret; 3859 } 3860 3861 3862 static bool is_zero(BlockDriverState *bs, int64_t offset, int64_t bytes) 3863 { 3864 int64_t nr; 3865 int res; 3866 3867 /* Clamp to image length, before checking status of underlying sectors */ 3868 if (offset + bytes > bs->total_sectors * BDRV_SECTOR_SIZE) { 3869 bytes = bs->total_sectors * BDRV_SECTOR_SIZE - offset; 3870 } 3871 3872 if (!bytes) { 3873 return true; 3874 } 3875 res = bdrv_block_status_above(bs, NULL, offset, bytes, &nr, NULL, NULL); 3876 return res >= 0 && (res & BDRV_BLOCK_ZERO) && nr == bytes; 3877 } 3878 3879 static coroutine_fn int qcow2_co_pwrite_zeroes(BlockDriverState *bs, 3880 int64_t offset, int bytes, BdrvRequestFlags flags) 3881 { 3882 int ret; 3883 BDRVQcow2State *s = bs->opaque; 3884 3885 uint32_t head = offset_into_subcluster(s, offset); 3886 uint32_t tail = ROUND_UP(offset + bytes, s->subcluster_size) - 3887 (offset + bytes); 3888 3889 trace_qcow2_pwrite_zeroes_start_req(qemu_coroutine_self(), offset, bytes); 3890 if (offset + bytes == bs->total_sectors * BDRV_SECTOR_SIZE) { 3891 tail = 0; 3892 } 3893 3894 if (head || tail) { 3895 uint64_t off; 3896 unsigned int nr; 3897 QCow2SubclusterType type; 3898 3899 assert(head + bytes + tail <= s->subcluster_size); 3900 3901 /* check whether remainder of cluster already reads as zero */ 3902 if (!(is_zero(bs, offset - head, head) && 3903 is_zero(bs, offset + bytes, tail))) { 3904 return -ENOTSUP; 3905 } 3906 3907 qemu_co_mutex_lock(&s->lock); 3908 /* We can have new write after previous check */ 3909 offset -= head; 3910 bytes = s->subcluster_size; 3911 nr = s->subcluster_size; 3912 ret = qcow2_get_host_offset(bs, offset, &nr, &off, &type); 3913 if (ret < 0 || 3914 (type != QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN && 3915 type != QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC && 3916 type != QCOW2_SUBCLUSTER_ZERO_PLAIN && 3917 type != QCOW2_SUBCLUSTER_ZERO_ALLOC)) { 3918 qemu_co_mutex_unlock(&s->lock); 3919 return -ENOTSUP; 3920 } 3921 } else { 3922 qemu_co_mutex_lock(&s->lock); 3923 } 3924 3925 trace_qcow2_pwrite_zeroes(qemu_coroutine_self(), offset, bytes); 3926 3927 /* Whatever is left can use real zero subclusters */ 3928 ret = qcow2_subcluster_zeroize(bs, offset, bytes, flags); 3929 qemu_co_mutex_unlock(&s->lock); 3930 3931 return ret; 3932 } 3933 3934 static coroutine_fn int qcow2_co_pdiscard(BlockDriverState *bs, 3935 int64_t offset, int bytes) 3936 { 3937 int ret; 3938 BDRVQcow2State *s = bs->opaque; 3939 3940 /* If the image does not support QCOW_OFLAG_ZERO then discarding 3941 * clusters could expose stale data from the backing file. */ 3942 if (s->qcow_version < 3 && bs->backing) { 3943 return -ENOTSUP; 3944 } 3945 3946 if (!QEMU_IS_ALIGNED(offset | bytes, s->cluster_size)) { 3947 assert(bytes < s->cluster_size); 3948 /* Ignore partial clusters, except for the special case of the 3949 * complete partial cluster at the end of an unaligned file */ 3950 if (!QEMU_IS_ALIGNED(offset, s->cluster_size) || 3951 offset + bytes != bs->total_sectors * BDRV_SECTOR_SIZE) { 3952 return -ENOTSUP; 3953 } 3954 } 3955 3956 qemu_co_mutex_lock(&s->lock); 3957 ret = qcow2_cluster_discard(bs, offset, bytes, QCOW2_DISCARD_REQUEST, 3958 false); 3959 qemu_co_mutex_unlock(&s->lock); 3960 return ret; 3961 } 3962 3963 static int coroutine_fn 3964 qcow2_co_copy_range_from(BlockDriverState *bs, 3965 BdrvChild *src, uint64_t src_offset, 3966 BdrvChild *dst, uint64_t dst_offset, 3967 uint64_t bytes, BdrvRequestFlags read_flags, 3968 BdrvRequestFlags write_flags) 3969 { 3970 BDRVQcow2State *s = bs->opaque; 3971 int ret; 3972 unsigned int cur_bytes; /* number of bytes in current iteration */ 3973 BdrvChild *child = NULL; 3974 BdrvRequestFlags cur_write_flags; 3975 3976 assert(!bs->encrypted); 3977 qemu_co_mutex_lock(&s->lock); 3978 3979 while (bytes != 0) { 3980 uint64_t copy_offset = 0; 3981 QCow2SubclusterType type; 3982 /* prepare next request */ 3983 cur_bytes = MIN(bytes, INT_MAX); 3984 cur_write_flags = write_flags; 3985 3986 ret = qcow2_get_host_offset(bs, src_offset, &cur_bytes, 3987 ©_offset, &type); 3988 if (ret < 0) { 3989 goto out; 3990 } 3991 3992 switch (type) { 3993 case QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN: 3994 case QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC: 3995 if (bs->backing && bs->backing->bs) { 3996 int64_t backing_length = bdrv_getlength(bs->backing->bs); 3997 if (src_offset >= backing_length) { 3998 cur_write_flags |= BDRV_REQ_ZERO_WRITE; 3999 } else { 4000 child = bs->backing; 4001 cur_bytes = MIN(cur_bytes, backing_length - src_offset); 4002 copy_offset = src_offset; 4003 } 4004 } else { 4005 cur_write_flags |= BDRV_REQ_ZERO_WRITE; 4006 } 4007 break; 4008 4009 case QCOW2_SUBCLUSTER_ZERO_PLAIN: 4010 case QCOW2_SUBCLUSTER_ZERO_ALLOC: 4011 cur_write_flags |= BDRV_REQ_ZERO_WRITE; 4012 break; 4013 4014 case QCOW2_SUBCLUSTER_COMPRESSED: 4015 ret = -ENOTSUP; 4016 goto out; 4017 4018 case QCOW2_SUBCLUSTER_NORMAL: 4019 child = s->data_file; 4020 break; 4021 4022 default: 4023 abort(); 4024 } 4025 qemu_co_mutex_unlock(&s->lock); 4026 ret = bdrv_co_copy_range_from(child, 4027 copy_offset, 4028 dst, dst_offset, 4029 cur_bytes, read_flags, cur_write_flags); 4030 qemu_co_mutex_lock(&s->lock); 4031 if (ret < 0) { 4032 goto out; 4033 } 4034 4035 bytes -= cur_bytes; 4036 src_offset += cur_bytes; 4037 dst_offset += cur_bytes; 4038 } 4039 ret = 0; 4040 4041 out: 4042 qemu_co_mutex_unlock(&s->lock); 4043 return ret; 4044 } 4045 4046 static int coroutine_fn 4047 qcow2_co_copy_range_to(BlockDriverState *bs, 4048 BdrvChild *src, uint64_t src_offset, 4049 BdrvChild *dst, uint64_t dst_offset, 4050 uint64_t bytes, BdrvRequestFlags read_flags, 4051 BdrvRequestFlags write_flags) 4052 { 4053 BDRVQcow2State *s = bs->opaque; 4054 int offset_in_cluster; 4055 int ret; 4056 unsigned int cur_bytes; /* number of sectors in current iteration */ 4057 uint64_t cluster_offset; 4058 QCowL2Meta *l2meta = NULL; 4059 4060 assert(!bs->encrypted); 4061 4062 qemu_co_mutex_lock(&s->lock); 4063 4064 while (bytes != 0) { 4065 4066 l2meta = NULL; 4067 4068 offset_in_cluster = offset_into_cluster(s, dst_offset); 4069 cur_bytes = MIN(bytes, INT_MAX); 4070 4071 /* TODO: 4072 * If src->bs == dst->bs, we could simply copy by incrementing 4073 * the refcnt, without copying user data. 4074 * Or if src->bs == dst->bs->backing->bs, we could copy by discarding. */ 4075 ret = qcow2_alloc_cluster_offset(bs, dst_offset, &cur_bytes, 4076 &cluster_offset, &l2meta); 4077 if (ret < 0) { 4078 goto fail; 4079 } 4080 4081 assert(offset_into_cluster(s, cluster_offset) == 0); 4082 4083 ret = qcow2_pre_write_overlap_check(bs, 0, 4084 cluster_offset + offset_in_cluster, cur_bytes, true); 4085 if (ret < 0) { 4086 goto fail; 4087 } 4088 4089 qemu_co_mutex_unlock(&s->lock); 4090 ret = bdrv_co_copy_range_to(src, src_offset, 4091 s->data_file, 4092 cluster_offset + offset_in_cluster, 4093 cur_bytes, read_flags, write_flags); 4094 qemu_co_mutex_lock(&s->lock); 4095 if (ret < 0) { 4096 goto fail; 4097 } 4098 4099 ret = qcow2_handle_l2meta(bs, &l2meta, true); 4100 if (ret) { 4101 goto fail; 4102 } 4103 4104 bytes -= cur_bytes; 4105 src_offset += cur_bytes; 4106 dst_offset += cur_bytes; 4107 } 4108 ret = 0; 4109 4110 fail: 4111 qcow2_handle_l2meta(bs, &l2meta, false); 4112 4113 qemu_co_mutex_unlock(&s->lock); 4114 4115 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret); 4116 4117 return ret; 4118 } 4119 4120 static int coroutine_fn qcow2_co_truncate(BlockDriverState *bs, int64_t offset, 4121 bool exact, PreallocMode prealloc, 4122 BdrvRequestFlags flags, Error **errp) 4123 { 4124 BDRVQcow2State *s = bs->opaque; 4125 uint64_t old_length; 4126 int64_t new_l1_size; 4127 int ret; 4128 QDict *options; 4129 4130 if (prealloc != PREALLOC_MODE_OFF && prealloc != PREALLOC_MODE_METADATA && 4131 prealloc != PREALLOC_MODE_FALLOC && prealloc != PREALLOC_MODE_FULL) 4132 { 4133 error_setg(errp, "Unsupported preallocation mode '%s'", 4134 PreallocMode_str(prealloc)); 4135 return -ENOTSUP; 4136 } 4137 4138 if (!QEMU_IS_ALIGNED(offset, BDRV_SECTOR_SIZE)) { 4139 error_setg(errp, "The new size must be a multiple of %u", 4140 (unsigned) BDRV_SECTOR_SIZE); 4141 return -EINVAL; 4142 } 4143 4144 qemu_co_mutex_lock(&s->lock); 4145 4146 /* 4147 * Even though we store snapshot size for all images, it was not 4148 * required until v3, so it is not safe to proceed for v2. 4149 */ 4150 if (s->nb_snapshots && s->qcow_version < 3) { 4151 error_setg(errp, "Can't resize a v2 image which has snapshots"); 4152 ret = -ENOTSUP; 4153 goto fail; 4154 } 4155 4156 /* See qcow2-bitmap.c for which bitmap scenarios prevent a resize. */ 4157 if (qcow2_truncate_bitmaps_check(bs, errp)) { 4158 ret = -ENOTSUP; 4159 goto fail; 4160 } 4161 4162 old_length = bs->total_sectors * BDRV_SECTOR_SIZE; 4163 new_l1_size = size_to_l1(s, offset); 4164 4165 if (offset < old_length) { 4166 int64_t last_cluster, old_file_size; 4167 if (prealloc != PREALLOC_MODE_OFF) { 4168 error_setg(errp, 4169 "Preallocation can't be used for shrinking an image"); 4170 ret = -EINVAL; 4171 goto fail; 4172 } 4173 4174 ret = qcow2_cluster_discard(bs, ROUND_UP(offset, s->cluster_size), 4175 old_length - ROUND_UP(offset, 4176 s->cluster_size), 4177 QCOW2_DISCARD_ALWAYS, true); 4178 if (ret < 0) { 4179 error_setg_errno(errp, -ret, "Failed to discard cropped clusters"); 4180 goto fail; 4181 } 4182 4183 ret = qcow2_shrink_l1_table(bs, new_l1_size); 4184 if (ret < 0) { 4185 error_setg_errno(errp, -ret, 4186 "Failed to reduce the number of L2 tables"); 4187 goto fail; 4188 } 4189 4190 ret = qcow2_shrink_reftable(bs); 4191 if (ret < 0) { 4192 error_setg_errno(errp, -ret, 4193 "Failed to discard unused refblocks"); 4194 goto fail; 4195 } 4196 4197 old_file_size = bdrv_getlength(bs->file->bs); 4198 if (old_file_size < 0) { 4199 error_setg_errno(errp, -old_file_size, 4200 "Failed to inquire current file length"); 4201 ret = old_file_size; 4202 goto fail; 4203 } 4204 last_cluster = qcow2_get_last_cluster(bs, old_file_size); 4205 if (last_cluster < 0) { 4206 error_setg_errno(errp, -last_cluster, 4207 "Failed to find the last cluster"); 4208 ret = last_cluster; 4209 goto fail; 4210 } 4211 if ((last_cluster + 1) * s->cluster_size < old_file_size) { 4212 Error *local_err = NULL; 4213 4214 /* 4215 * Do not pass @exact here: It will not help the user if 4216 * we get an error here just because they wanted to shrink 4217 * their qcow2 image (on a block device) with qemu-img. 4218 * (And on the qcow2 layer, the @exact requirement is 4219 * always fulfilled, so there is no need to pass it on.) 4220 */ 4221 bdrv_co_truncate(bs->file, (last_cluster + 1) * s->cluster_size, 4222 false, PREALLOC_MODE_OFF, 0, &local_err); 4223 if (local_err) { 4224 warn_reportf_err(local_err, 4225 "Failed to truncate the tail of the image: "); 4226 } 4227 } 4228 } else { 4229 ret = qcow2_grow_l1_table(bs, new_l1_size, true); 4230 if (ret < 0) { 4231 error_setg_errno(errp, -ret, "Failed to grow the L1 table"); 4232 goto fail; 4233 } 4234 } 4235 4236 switch (prealloc) { 4237 case PREALLOC_MODE_OFF: 4238 if (has_data_file(bs)) { 4239 /* 4240 * If the caller wants an exact resize, the external data 4241 * file should be resized to the exact target size, too, 4242 * so we pass @exact here. 4243 */ 4244 ret = bdrv_co_truncate(s->data_file, offset, exact, prealloc, 0, 4245 errp); 4246 if (ret < 0) { 4247 goto fail; 4248 } 4249 } 4250 break; 4251 4252 case PREALLOC_MODE_METADATA: 4253 ret = preallocate_co(bs, old_length, offset, prealloc, errp); 4254 if (ret < 0) { 4255 goto fail; 4256 } 4257 break; 4258 4259 case PREALLOC_MODE_FALLOC: 4260 case PREALLOC_MODE_FULL: 4261 { 4262 int64_t allocation_start, host_offset, guest_offset; 4263 int64_t clusters_allocated; 4264 int64_t old_file_size, last_cluster, new_file_size; 4265 uint64_t nb_new_data_clusters, nb_new_l2_tables; 4266 bool subclusters_need_allocation = false; 4267 4268 /* With a data file, preallocation means just allocating the metadata 4269 * and forwarding the truncate request to the data file */ 4270 if (has_data_file(bs)) { 4271 ret = preallocate_co(bs, old_length, offset, prealloc, errp); 4272 if (ret < 0) { 4273 goto fail; 4274 } 4275 break; 4276 } 4277 4278 old_file_size = bdrv_getlength(bs->file->bs); 4279 if (old_file_size < 0) { 4280 error_setg_errno(errp, -old_file_size, 4281 "Failed to inquire current file length"); 4282 ret = old_file_size; 4283 goto fail; 4284 } 4285 4286 last_cluster = qcow2_get_last_cluster(bs, old_file_size); 4287 if (last_cluster >= 0) { 4288 old_file_size = (last_cluster + 1) * s->cluster_size; 4289 } else { 4290 old_file_size = ROUND_UP(old_file_size, s->cluster_size); 4291 } 4292 4293 nb_new_data_clusters = (ROUND_UP(offset, s->cluster_size) - 4294 start_of_cluster(s, old_length)) >> s->cluster_bits; 4295 4296 /* This is an overestimation; we will not actually allocate space for 4297 * these in the file but just make sure the new refcount structures are 4298 * able to cover them so we will not have to allocate new refblocks 4299 * while entering the data blocks in the potentially new L2 tables. 4300 * (We do not actually care where the L2 tables are placed. Maybe they 4301 * are already allocated or they can be placed somewhere before 4302 * @old_file_size. It does not matter because they will be fully 4303 * allocated automatically, so they do not need to be covered by the 4304 * preallocation. All that matters is that we will not have to allocate 4305 * new refcount structures for them.) */ 4306 nb_new_l2_tables = DIV_ROUND_UP(nb_new_data_clusters, 4307 s->cluster_size / l2_entry_size(s)); 4308 /* The cluster range may not be aligned to L2 boundaries, so add one L2 4309 * table for a potential head/tail */ 4310 nb_new_l2_tables++; 4311 4312 allocation_start = qcow2_refcount_area(bs, old_file_size, 4313 nb_new_data_clusters + 4314 nb_new_l2_tables, 4315 true, 0, 0); 4316 if (allocation_start < 0) { 4317 error_setg_errno(errp, -allocation_start, 4318 "Failed to resize refcount structures"); 4319 ret = allocation_start; 4320 goto fail; 4321 } 4322 4323 clusters_allocated = qcow2_alloc_clusters_at(bs, allocation_start, 4324 nb_new_data_clusters); 4325 if (clusters_allocated < 0) { 4326 error_setg_errno(errp, -clusters_allocated, 4327 "Failed to allocate data clusters"); 4328 ret = clusters_allocated; 4329 goto fail; 4330 } 4331 4332 assert(clusters_allocated == nb_new_data_clusters); 4333 4334 /* Allocate the data area */ 4335 new_file_size = allocation_start + 4336 nb_new_data_clusters * s->cluster_size; 4337 /* 4338 * Image file grows, so @exact does not matter. 4339 * 4340 * If we need to zero out the new area, try first whether the protocol 4341 * driver can already take care of this. 4342 */ 4343 if (flags & BDRV_REQ_ZERO_WRITE) { 4344 ret = bdrv_co_truncate(bs->file, new_file_size, false, prealloc, 4345 BDRV_REQ_ZERO_WRITE, NULL); 4346 if (ret >= 0) { 4347 flags &= ~BDRV_REQ_ZERO_WRITE; 4348 /* Ensure that we read zeroes and not backing file data */ 4349 subclusters_need_allocation = true; 4350 } 4351 } else { 4352 ret = -1; 4353 } 4354 if (ret < 0) { 4355 ret = bdrv_co_truncate(bs->file, new_file_size, false, prealloc, 0, 4356 errp); 4357 } 4358 if (ret < 0) { 4359 error_prepend(errp, "Failed to resize underlying file: "); 4360 qcow2_free_clusters(bs, allocation_start, 4361 nb_new_data_clusters * s->cluster_size, 4362 QCOW2_DISCARD_OTHER); 4363 goto fail; 4364 } 4365 4366 /* Create the necessary L2 entries */ 4367 host_offset = allocation_start; 4368 guest_offset = old_length; 4369 while (nb_new_data_clusters) { 4370 int64_t nb_clusters = MIN( 4371 nb_new_data_clusters, 4372 s->l2_slice_size - offset_to_l2_slice_index(s, guest_offset)); 4373 unsigned cow_start_length = offset_into_cluster(s, guest_offset); 4374 QCowL2Meta allocation; 4375 guest_offset = start_of_cluster(s, guest_offset); 4376 allocation = (QCowL2Meta) { 4377 .offset = guest_offset, 4378 .alloc_offset = host_offset, 4379 .nb_clusters = nb_clusters, 4380 .cow_start = { 4381 .offset = 0, 4382 .nb_bytes = cow_start_length, 4383 }, 4384 .cow_end = { 4385 .offset = nb_clusters << s->cluster_bits, 4386 .nb_bytes = 0, 4387 }, 4388 .prealloc = !subclusters_need_allocation, 4389 }; 4390 qemu_co_queue_init(&allocation.dependent_requests); 4391 4392 ret = qcow2_alloc_cluster_link_l2(bs, &allocation); 4393 if (ret < 0) { 4394 error_setg_errno(errp, -ret, "Failed to update L2 tables"); 4395 qcow2_free_clusters(bs, host_offset, 4396 nb_new_data_clusters * s->cluster_size, 4397 QCOW2_DISCARD_OTHER); 4398 goto fail; 4399 } 4400 4401 guest_offset += nb_clusters * s->cluster_size; 4402 host_offset += nb_clusters * s->cluster_size; 4403 nb_new_data_clusters -= nb_clusters; 4404 } 4405 break; 4406 } 4407 4408 default: 4409 g_assert_not_reached(); 4410 } 4411 4412 if ((flags & BDRV_REQ_ZERO_WRITE) && offset > old_length) { 4413 uint64_t zero_start = QEMU_ALIGN_UP(old_length, s->subcluster_size); 4414 4415 /* 4416 * Use zero clusters as much as we can. qcow2_subcluster_zeroize() 4417 * requires a subcluster-aligned start. The end may be unaligned if 4418 * it is at the end of the image (which it is here). 4419 */ 4420 if (offset > zero_start) { 4421 ret = qcow2_subcluster_zeroize(bs, zero_start, offset - zero_start, 4422 0); 4423 if (ret < 0) { 4424 error_setg_errno(errp, -ret, "Failed to zero out new clusters"); 4425 goto fail; 4426 } 4427 } 4428 4429 /* Write explicit zeros for the unaligned head */ 4430 if (zero_start > old_length) { 4431 uint64_t len = MIN(zero_start, offset) - old_length; 4432 uint8_t *buf = qemu_blockalign0(bs, len); 4433 QEMUIOVector qiov; 4434 qemu_iovec_init_buf(&qiov, buf, len); 4435 4436 qemu_co_mutex_unlock(&s->lock); 4437 ret = qcow2_co_pwritev_part(bs, old_length, len, &qiov, 0, 0); 4438 qemu_co_mutex_lock(&s->lock); 4439 4440 qemu_vfree(buf); 4441 if (ret < 0) { 4442 error_setg_errno(errp, -ret, "Failed to zero out the new area"); 4443 goto fail; 4444 } 4445 } 4446 } 4447 4448 if (prealloc != PREALLOC_MODE_OFF) { 4449 /* Flush metadata before actually changing the image size */ 4450 ret = qcow2_write_caches(bs); 4451 if (ret < 0) { 4452 error_setg_errno(errp, -ret, 4453 "Failed to flush the preallocated area to disk"); 4454 goto fail; 4455 } 4456 } 4457 4458 bs->total_sectors = offset / BDRV_SECTOR_SIZE; 4459 4460 /* write updated header.size */ 4461 offset = cpu_to_be64(offset); 4462 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, size), 4463 &offset, sizeof(uint64_t)); 4464 if (ret < 0) { 4465 error_setg_errno(errp, -ret, "Failed to update the image size"); 4466 goto fail; 4467 } 4468 4469 s->l1_vm_state_index = new_l1_size; 4470 4471 /* Update cache sizes */ 4472 options = qdict_clone_shallow(bs->options); 4473 ret = qcow2_update_options(bs, options, s->flags, errp); 4474 qobject_unref(options); 4475 if (ret < 0) { 4476 goto fail; 4477 } 4478 ret = 0; 4479 fail: 4480 qemu_co_mutex_unlock(&s->lock); 4481 return ret; 4482 } 4483 4484 static coroutine_fn int 4485 qcow2_co_pwritev_compressed_task(BlockDriverState *bs, 4486 uint64_t offset, uint64_t bytes, 4487 QEMUIOVector *qiov, size_t qiov_offset) 4488 { 4489 BDRVQcow2State *s = bs->opaque; 4490 int ret; 4491 ssize_t out_len; 4492 uint8_t *buf, *out_buf; 4493 uint64_t cluster_offset; 4494 4495 assert(bytes == s->cluster_size || (bytes < s->cluster_size && 4496 (offset + bytes == bs->total_sectors << BDRV_SECTOR_BITS))); 4497 4498 buf = qemu_blockalign(bs, s->cluster_size); 4499 if (bytes < s->cluster_size) { 4500 /* Zero-pad last write if image size is not cluster aligned */ 4501 memset(buf + bytes, 0, s->cluster_size - bytes); 4502 } 4503 qemu_iovec_to_buf(qiov, qiov_offset, buf, bytes); 4504 4505 out_buf = g_malloc(s->cluster_size); 4506 4507 out_len = qcow2_co_compress(bs, out_buf, s->cluster_size - 1, 4508 buf, s->cluster_size); 4509 if (out_len == -ENOMEM) { 4510 /* could not compress: write normal cluster */ 4511 ret = qcow2_co_pwritev_part(bs, offset, bytes, qiov, qiov_offset, 0); 4512 if (ret < 0) { 4513 goto fail; 4514 } 4515 goto success; 4516 } else if (out_len < 0) { 4517 ret = -EINVAL; 4518 goto fail; 4519 } 4520 4521 qemu_co_mutex_lock(&s->lock); 4522 ret = qcow2_alloc_compressed_cluster_offset(bs, offset, out_len, 4523 &cluster_offset); 4524 if (ret < 0) { 4525 qemu_co_mutex_unlock(&s->lock); 4526 goto fail; 4527 } 4528 4529 ret = qcow2_pre_write_overlap_check(bs, 0, cluster_offset, out_len, true); 4530 qemu_co_mutex_unlock(&s->lock); 4531 if (ret < 0) { 4532 goto fail; 4533 } 4534 4535 BLKDBG_EVENT(s->data_file, BLKDBG_WRITE_COMPRESSED); 4536 ret = bdrv_co_pwrite(s->data_file, cluster_offset, out_len, out_buf, 0); 4537 if (ret < 0) { 4538 goto fail; 4539 } 4540 success: 4541 ret = 0; 4542 fail: 4543 qemu_vfree(buf); 4544 g_free(out_buf); 4545 return ret; 4546 } 4547 4548 static coroutine_fn int qcow2_co_pwritev_compressed_task_entry(AioTask *task) 4549 { 4550 Qcow2AioTask *t = container_of(task, Qcow2AioTask, task); 4551 4552 assert(!t->subcluster_type && !t->l2meta); 4553 4554 return qcow2_co_pwritev_compressed_task(t->bs, t->offset, t->bytes, t->qiov, 4555 t->qiov_offset); 4556 } 4557 4558 /* 4559 * XXX: put compressed sectors first, then all the cluster aligned 4560 * tables to avoid losing bytes in alignment 4561 */ 4562 static coroutine_fn int 4563 qcow2_co_pwritev_compressed_part(BlockDriverState *bs, 4564 uint64_t offset, uint64_t bytes, 4565 QEMUIOVector *qiov, size_t qiov_offset) 4566 { 4567 BDRVQcow2State *s = bs->opaque; 4568 AioTaskPool *aio = NULL; 4569 int ret = 0; 4570 4571 if (has_data_file(bs)) { 4572 return -ENOTSUP; 4573 } 4574 4575 if (bytes == 0) { 4576 /* 4577 * align end of file to a sector boundary to ease reading with 4578 * sector based I/Os 4579 */ 4580 int64_t len = bdrv_getlength(bs->file->bs); 4581 if (len < 0) { 4582 return len; 4583 } 4584 return bdrv_co_truncate(bs->file, len, false, PREALLOC_MODE_OFF, 0, 4585 NULL); 4586 } 4587 4588 if (offset_into_cluster(s, offset)) { 4589 return -EINVAL; 4590 } 4591 4592 if (offset_into_cluster(s, bytes) && 4593 (offset + bytes) != (bs->total_sectors << BDRV_SECTOR_BITS)) { 4594 return -EINVAL; 4595 } 4596 4597 while (bytes && aio_task_pool_status(aio) == 0) { 4598 uint64_t chunk_size = MIN(bytes, s->cluster_size); 4599 4600 if (!aio && chunk_size != bytes) { 4601 aio = aio_task_pool_new(QCOW2_MAX_WORKERS); 4602 } 4603 4604 ret = qcow2_add_task(bs, aio, qcow2_co_pwritev_compressed_task_entry, 4605 0, 0, offset, chunk_size, qiov, qiov_offset, NULL); 4606 if (ret < 0) { 4607 break; 4608 } 4609 qiov_offset += chunk_size; 4610 offset += chunk_size; 4611 bytes -= chunk_size; 4612 } 4613 4614 if (aio) { 4615 aio_task_pool_wait_all(aio); 4616 if (ret == 0) { 4617 ret = aio_task_pool_status(aio); 4618 } 4619 g_free(aio); 4620 } 4621 4622 return ret; 4623 } 4624 4625 static int coroutine_fn 4626 qcow2_co_preadv_compressed(BlockDriverState *bs, 4627 uint64_t cluster_descriptor, 4628 uint64_t offset, 4629 uint64_t bytes, 4630 QEMUIOVector *qiov, 4631 size_t qiov_offset) 4632 { 4633 BDRVQcow2State *s = bs->opaque; 4634 int ret = 0, csize, nb_csectors; 4635 uint64_t coffset; 4636 uint8_t *buf, *out_buf; 4637 int offset_in_cluster = offset_into_cluster(s, offset); 4638 4639 coffset = cluster_descriptor & s->cluster_offset_mask; 4640 nb_csectors = ((cluster_descriptor >> s->csize_shift) & s->csize_mask) + 1; 4641 csize = nb_csectors * QCOW2_COMPRESSED_SECTOR_SIZE - 4642 (coffset & ~QCOW2_COMPRESSED_SECTOR_MASK); 4643 4644 buf = g_try_malloc(csize); 4645 if (!buf) { 4646 return -ENOMEM; 4647 } 4648 4649 out_buf = qemu_blockalign(bs, s->cluster_size); 4650 4651 BLKDBG_EVENT(bs->file, BLKDBG_READ_COMPRESSED); 4652 ret = bdrv_co_pread(bs->file, coffset, csize, buf, 0); 4653 if (ret < 0) { 4654 goto fail; 4655 } 4656 4657 if (qcow2_co_decompress(bs, out_buf, s->cluster_size, buf, csize) < 0) { 4658 ret = -EIO; 4659 goto fail; 4660 } 4661 4662 qemu_iovec_from_buf(qiov, qiov_offset, out_buf + offset_in_cluster, bytes); 4663 4664 fail: 4665 qemu_vfree(out_buf); 4666 g_free(buf); 4667 4668 return ret; 4669 } 4670 4671 static int make_completely_empty(BlockDriverState *bs) 4672 { 4673 BDRVQcow2State *s = bs->opaque; 4674 Error *local_err = NULL; 4675 int ret, l1_clusters; 4676 int64_t offset; 4677 uint64_t *new_reftable = NULL; 4678 uint64_t rt_entry, l1_size2; 4679 struct { 4680 uint64_t l1_offset; 4681 uint64_t reftable_offset; 4682 uint32_t reftable_clusters; 4683 } QEMU_PACKED l1_ofs_rt_ofs_cls; 4684 4685 ret = qcow2_cache_empty(bs, s->l2_table_cache); 4686 if (ret < 0) { 4687 goto fail; 4688 } 4689 4690 ret = qcow2_cache_empty(bs, s->refcount_block_cache); 4691 if (ret < 0) { 4692 goto fail; 4693 } 4694 4695 /* Refcounts will be broken utterly */ 4696 ret = qcow2_mark_dirty(bs); 4697 if (ret < 0) { 4698 goto fail; 4699 } 4700 4701 BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE); 4702 4703 l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / sizeof(uint64_t)); 4704 l1_size2 = (uint64_t)s->l1_size * sizeof(uint64_t); 4705 4706 /* After this call, neither the in-memory nor the on-disk refcount 4707 * information accurately describe the actual references */ 4708 4709 ret = bdrv_pwrite_zeroes(bs->file, s->l1_table_offset, 4710 l1_clusters * s->cluster_size, 0); 4711 if (ret < 0) { 4712 goto fail_broken_refcounts; 4713 } 4714 memset(s->l1_table, 0, l1_size2); 4715 4716 BLKDBG_EVENT(bs->file, BLKDBG_EMPTY_IMAGE_PREPARE); 4717 4718 /* Overwrite enough clusters at the beginning of the sectors to place 4719 * the refcount table, a refcount block and the L1 table in; this may 4720 * overwrite parts of the existing refcount and L1 table, which is not 4721 * an issue because the dirty flag is set, complete data loss is in fact 4722 * desired and partial data loss is consequently fine as well */ 4723 ret = bdrv_pwrite_zeroes(bs->file, s->cluster_size, 4724 (2 + l1_clusters) * s->cluster_size, 0); 4725 /* This call (even if it failed overall) may have overwritten on-disk 4726 * refcount structures; in that case, the in-memory refcount information 4727 * will probably differ from the on-disk information which makes the BDS 4728 * unusable */ 4729 if (ret < 0) { 4730 goto fail_broken_refcounts; 4731 } 4732 4733 BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE); 4734 BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_UPDATE); 4735 4736 /* "Create" an empty reftable (one cluster) directly after the image 4737 * header and an empty L1 table three clusters after the image header; 4738 * the cluster between those two will be used as the first refblock */ 4739 l1_ofs_rt_ofs_cls.l1_offset = cpu_to_be64(3 * s->cluster_size); 4740 l1_ofs_rt_ofs_cls.reftable_offset = cpu_to_be64(s->cluster_size); 4741 l1_ofs_rt_ofs_cls.reftable_clusters = cpu_to_be32(1); 4742 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, l1_table_offset), 4743 &l1_ofs_rt_ofs_cls, sizeof(l1_ofs_rt_ofs_cls)); 4744 if (ret < 0) { 4745 goto fail_broken_refcounts; 4746 } 4747 4748 s->l1_table_offset = 3 * s->cluster_size; 4749 4750 new_reftable = g_try_new0(uint64_t, s->cluster_size / sizeof(uint64_t)); 4751 if (!new_reftable) { 4752 ret = -ENOMEM; 4753 goto fail_broken_refcounts; 4754 } 4755 4756 s->refcount_table_offset = s->cluster_size; 4757 s->refcount_table_size = s->cluster_size / sizeof(uint64_t); 4758 s->max_refcount_table_index = 0; 4759 4760 g_free(s->refcount_table); 4761 s->refcount_table = new_reftable; 4762 new_reftable = NULL; 4763 4764 /* Now the in-memory refcount information again corresponds to the on-disk 4765 * information (reftable is empty and no refblocks (the refblock cache is 4766 * empty)); however, this means some clusters (e.g. the image header) are 4767 * referenced, but not refcounted, but the normal qcow2 code assumes that 4768 * the in-memory information is always correct */ 4769 4770 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC); 4771 4772 /* Enter the first refblock into the reftable */ 4773 rt_entry = cpu_to_be64(2 * s->cluster_size); 4774 ret = bdrv_pwrite_sync(bs->file, s->cluster_size, 4775 &rt_entry, sizeof(rt_entry)); 4776 if (ret < 0) { 4777 goto fail_broken_refcounts; 4778 } 4779 s->refcount_table[0] = 2 * s->cluster_size; 4780 4781 s->free_cluster_index = 0; 4782 assert(3 + l1_clusters <= s->refcount_block_size); 4783 offset = qcow2_alloc_clusters(bs, 3 * s->cluster_size + l1_size2); 4784 if (offset < 0) { 4785 ret = offset; 4786 goto fail_broken_refcounts; 4787 } else if (offset > 0) { 4788 error_report("First cluster in emptied image is in use"); 4789 abort(); 4790 } 4791 4792 /* Now finally the in-memory information corresponds to the on-disk 4793 * structures and is correct */ 4794 ret = qcow2_mark_clean(bs); 4795 if (ret < 0) { 4796 goto fail; 4797 } 4798 4799 ret = bdrv_truncate(bs->file, (3 + l1_clusters) * s->cluster_size, false, 4800 PREALLOC_MODE_OFF, 0, &local_err); 4801 if (ret < 0) { 4802 error_report_err(local_err); 4803 goto fail; 4804 } 4805 4806 return 0; 4807 4808 fail_broken_refcounts: 4809 /* The BDS is unusable at this point. If we wanted to make it usable, we 4810 * would have to call qcow2_refcount_close(), qcow2_refcount_init(), 4811 * qcow2_check_refcounts(), qcow2_refcount_close() and qcow2_refcount_init() 4812 * again. However, because the functions which could have caused this error 4813 * path to be taken are used by those functions as well, it's very likely 4814 * that that sequence will fail as well. Therefore, just eject the BDS. */ 4815 bs->drv = NULL; 4816 4817 fail: 4818 g_free(new_reftable); 4819 return ret; 4820 } 4821 4822 static int qcow2_make_empty(BlockDriverState *bs) 4823 { 4824 BDRVQcow2State *s = bs->opaque; 4825 uint64_t offset, end_offset; 4826 int step = QEMU_ALIGN_DOWN(INT_MAX, s->cluster_size); 4827 int l1_clusters, ret = 0; 4828 4829 l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / sizeof(uint64_t)); 4830 4831 if (s->qcow_version >= 3 && !s->snapshots && !s->nb_bitmaps && 4832 3 + l1_clusters <= s->refcount_block_size && 4833 s->crypt_method_header != QCOW_CRYPT_LUKS && 4834 !has_data_file(bs)) { 4835 /* The following function only works for qcow2 v3 images (it 4836 * requires the dirty flag) and only as long as there are no 4837 * features that reserve extra clusters (such as snapshots, 4838 * LUKS header, or persistent bitmaps), because it completely 4839 * empties the image. Furthermore, the L1 table and three 4840 * additional clusters (image header, refcount table, one 4841 * refcount block) have to fit inside one refcount block. It 4842 * only resets the image file, i.e. does not work with an 4843 * external data file. */ 4844 return make_completely_empty(bs); 4845 } 4846 4847 /* This fallback code simply discards every active cluster; this is slow, 4848 * but works in all cases */ 4849 end_offset = bs->total_sectors * BDRV_SECTOR_SIZE; 4850 for (offset = 0; offset < end_offset; offset += step) { 4851 /* As this function is generally used after committing an external 4852 * snapshot, QCOW2_DISCARD_SNAPSHOT seems appropriate. Also, the 4853 * default action for this kind of discard is to pass the discard, 4854 * which will ideally result in an actually smaller image file, as 4855 * is probably desired. */ 4856 ret = qcow2_cluster_discard(bs, offset, MIN(step, end_offset - offset), 4857 QCOW2_DISCARD_SNAPSHOT, true); 4858 if (ret < 0) { 4859 break; 4860 } 4861 } 4862 4863 return ret; 4864 } 4865 4866 static coroutine_fn int qcow2_co_flush_to_os(BlockDriverState *bs) 4867 { 4868 BDRVQcow2State *s = bs->opaque; 4869 int ret; 4870 4871 qemu_co_mutex_lock(&s->lock); 4872 ret = qcow2_write_caches(bs); 4873 qemu_co_mutex_unlock(&s->lock); 4874 4875 return ret; 4876 } 4877 4878 static BlockMeasureInfo *qcow2_measure(QemuOpts *opts, BlockDriverState *in_bs, 4879 Error **errp) 4880 { 4881 Error *local_err = NULL; 4882 BlockMeasureInfo *info; 4883 uint64_t required = 0; /* bytes that contribute to required size */ 4884 uint64_t virtual_size; /* disk size as seen by guest */ 4885 uint64_t refcount_bits; 4886 uint64_t l2_tables; 4887 uint64_t luks_payload_size = 0; 4888 size_t cluster_size; 4889 int version; 4890 char *optstr; 4891 PreallocMode prealloc; 4892 bool has_backing_file; 4893 bool has_luks; 4894 bool extended_l2; 4895 size_t l2e_size; 4896 4897 /* Parse image creation options */ 4898 extended_l2 = qemu_opt_get_bool_del(opts, BLOCK_OPT_EXTL2, false); 4899 4900 cluster_size = qcow2_opt_get_cluster_size_del(opts, extended_l2, 4901 &local_err); 4902 if (local_err) { 4903 goto err; 4904 } 4905 4906 version = qcow2_opt_get_version_del(opts, &local_err); 4907 if (local_err) { 4908 goto err; 4909 } 4910 4911 refcount_bits = qcow2_opt_get_refcount_bits_del(opts, version, &local_err); 4912 if (local_err) { 4913 goto err; 4914 } 4915 4916 optstr = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC); 4917 prealloc = qapi_enum_parse(&PreallocMode_lookup, optstr, 4918 PREALLOC_MODE_OFF, &local_err); 4919 g_free(optstr); 4920 if (local_err) { 4921 goto err; 4922 } 4923 4924 optstr = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE); 4925 has_backing_file = !!optstr; 4926 g_free(optstr); 4927 4928 optstr = qemu_opt_get_del(opts, BLOCK_OPT_ENCRYPT_FORMAT); 4929 has_luks = optstr && strcmp(optstr, "luks") == 0; 4930 g_free(optstr); 4931 4932 if (has_luks) { 4933 g_autoptr(QCryptoBlockCreateOptions) create_opts = NULL; 4934 QDict *cryptoopts = qcow2_extract_crypto_opts(opts, "luks", errp); 4935 size_t headerlen; 4936 4937 create_opts = block_crypto_create_opts_init(cryptoopts, errp); 4938 qobject_unref(cryptoopts); 4939 if (!create_opts) { 4940 goto err; 4941 } 4942 4943 if (!qcrypto_block_calculate_payload_offset(create_opts, 4944 "encrypt.", 4945 &headerlen, 4946 &local_err)) { 4947 goto err; 4948 } 4949 4950 luks_payload_size = ROUND_UP(headerlen, cluster_size); 4951 } 4952 4953 virtual_size = qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0); 4954 virtual_size = ROUND_UP(virtual_size, cluster_size); 4955 4956 /* Check that virtual disk size is valid */ 4957 l2e_size = extended_l2 ? L2E_SIZE_EXTENDED : L2E_SIZE_NORMAL; 4958 l2_tables = DIV_ROUND_UP(virtual_size / cluster_size, 4959 cluster_size / l2e_size); 4960 if (l2_tables * sizeof(uint64_t) > QCOW_MAX_L1_SIZE) { 4961 error_setg(&local_err, "The image size is too large " 4962 "(try using a larger cluster size)"); 4963 goto err; 4964 } 4965 4966 /* Account for input image */ 4967 if (in_bs) { 4968 int64_t ssize = bdrv_getlength(in_bs); 4969 if (ssize < 0) { 4970 error_setg_errno(&local_err, -ssize, 4971 "Unable to get image virtual_size"); 4972 goto err; 4973 } 4974 4975 virtual_size = ROUND_UP(ssize, cluster_size); 4976 4977 if (has_backing_file) { 4978 /* We don't how much of the backing chain is shared by the input 4979 * image and the new image file. In the worst case the new image's 4980 * backing file has nothing in common with the input image. Be 4981 * conservative and assume all clusters need to be written. 4982 */ 4983 required = virtual_size; 4984 } else { 4985 int64_t offset; 4986 int64_t pnum = 0; 4987 4988 for (offset = 0; offset < ssize; offset += pnum) { 4989 int ret; 4990 4991 ret = bdrv_block_status_above(in_bs, NULL, offset, 4992 ssize - offset, &pnum, NULL, 4993 NULL); 4994 if (ret < 0) { 4995 error_setg_errno(&local_err, -ret, 4996 "Unable to get block status"); 4997 goto err; 4998 } 4999 5000 if (ret & BDRV_BLOCK_ZERO) { 5001 /* Skip zero regions (safe with no backing file) */ 5002 } else if ((ret & (BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED)) == 5003 (BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED)) { 5004 /* Extend pnum to end of cluster for next iteration */ 5005 pnum = ROUND_UP(offset + pnum, cluster_size) - offset; 5006 5007 /* Count clusters we've seen */ 5008 required += offset % cluster_size + pnum; 5009 } 5010 } 5011 } 5012 } 5013 5014 /* Take into account preallocation. Nothing special is needed for 5015 * PREALLOC_MODE_METADATA since metadata is always counted. 5016 */ 5017 if (prealloc == PREALLOC_MODE_FULL || prealloc == PREALLOC_MODE_FALLOC) { 5018 required = virtual_size; 5019 } 5020 5021 info = g_new0(BlockMeasureInfo, 1); 5022 info->fully_allocated = luks_payload_size + 5023 qcow2_calc_prealloc_size(virtual_size, cluster_size, 5024 ctz32(refcount_bits), extended_l2); 5025 5026 /* 5027 * Remove data clusters that are not required. This overestimates the 5028 * required size because metadata needed for the fully allocated file is 5029 * still counted. Show bitmaps only if both source and destination 5030 * would support them. 5031 */ 5032 info->required = info->fully_allocated - virtual_size + required; 5033 info->has_bitmaps = version >= 3 && in_bs && 5034 bdrv_supports_persistent_dirty_bitmap(in_bs); 5035 if (info->has_bitmaps) { 5036 info->bitmaps = qcow2_get_persistent_dirty_bitmap_size(in_bs, 5037 cluster_size); 5038 } 5039 return info; 5040 5041 err: 5042 error_propagate(errp, local_err); 5043 return NULL; 5044 } 5045 5046 static int qcow2_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) 5047 { 5048 BDRVQcow2State *s = bs->opaque; 5049 bdi->cluster_size = s->cluster_size; 5050 bdi->vm_state_offset = qcow2_vm_state_offset(s); 5051 return 0; 5052 } 5053 5054 static ImageInfoSpecific *qcow2_get_specific_info(BlockDriverState *bs, 5055 Error **errp) 5056 { 5057 BDRVQcow2State *s = bs->opaque; 5058 ImageInfoSpecific *spec_info; 5059 QCryptoBlockInfo *encrypt_info = NULL; 5060 Error *local_err = NULL; 5061 5062 if (s->crypto != NULL) { 5063 encrypt_info = qcrypto_block_get_info(s->crypto, &local_err); 5064 if (local_err) { 5065 error_propagate(errp, local_err); 5066 return NULL; 5067 } 5068 } 5069 5070 spec_info = g_new(ImageInfoSpecific, 1); 5071 *spec_info = (ImageInfoSpecific){ 5072 .type = IMAGE_INFO_SPECIFIC_KIND_QCOW2, 5073 .u.qcow2.data = g_new0(ImageInfoSpecificQCow2, 1), 5074 }; 5075 if (s->qcow_version == 2) { 5076 *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){ 5077 .compat = g_strdup("0.10"), 5078 .refcount_bits = s->refcount_bits, 5079 }; 5080 } else if (s->qcow_version == 3) { 5081 Qcow2BitmapInfoList *bitmaps; 5082 bitmaps = qcow2_get_bitmap_info_list(bs, &local_err); 5083 if (local_err) { 5084 error_propagate(errp, local_err); 5085 qapi_free_ImageInfoSpecific(spec_info); 5086 qapi_free_QCryptoBlockInfo(encrypt_info); 5087 return NULL; 5088 } 5089 *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){ 5090 .compat = g_strdup("1.1"), 5091 .lazy_refcounts = s->compatible_features & 5092 QCOW2_COMPAT_LAZY_REFCOUNTS, 5093 .has_lazy_refcounts = true, 5094 .corrupt = s->incompatible_features & 5095 QCOW2_INCOMPAT_CORRUPT, 5096 .has_corrupt = true, 5097 .has_extended_l2 = true, 5098 .extended_l2 = has_subclusters(s), 5099 .refcount_bits = s->refcount_bits, 5100 .has_bitmaps = !!bitmaps, 5101 .bitmaps = bitmaps, 5102 .has_data_file = !!s->image_data_file, 5103 .data_file = g_strdup(s->image_data_file), 5104 .has_data_file_raw = has_data_file(bs), 5105 .data_file_raw = data_file_is_raw(bs), 5106 .compression_type = s->compression_type, 5107 }; 5108 } else { 5109 /* if this assertion fails, this probably means a new version was 5110 * added without having it covered here */ 5111 assert(false); 5112 } 5113 5114 if (encrypt_info) { 5115 ImageInfoSpecificQCow2Encryption *qencrypt = 5116 g_new(ImageInfoSpecificQCow2Encryption, 1); 5117 switch (encrypt_info->format) { 5118 case Q_CRYPTO_BLOCK_FORMAT_QCOW: 5119 qencrypt->format = BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_AES; 5120 break; 5121 case Q_CRYPTO_BLOCK_FORMAT_LUKS: 5122 qencrypt->format = BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_LUKS; 5123 qencrypt->u.luks = encrypt_info->u.luks; 5124 break; 5125 default: 5126 abort(); 5127 } 5128 /* Since we did shallow copy above, erase any pointers 5129 * in the original info */ 5130 memset(&encrypt_info->u, 0, sizeof(encrypt_info->u)); 5131 qapi_free_QCryptoBlockInfo(encrypt_info); 5132 5133 spec_info->u.qcow2.data->has_encrypt = true; 5134 spec_info->u.qcow2.data->encrypt = qencrypt; 5135 } 5136 5137 return spec_info; 5138 } 5139 5140 static int qcow2_has_zero_init(BlockDriverState *bs) 5141 { 5142 BDRVQcow2State *s = bs->opaque; 5143 bool preallocated; 5144 5145 if (qemu_in_coroutine()) { 5146 qemu_co_mutex_lock(&s->lock); 5147 } 5148 /* 5149 * Check preallocation status: Preallocated images have all L2 5150 * tables allocated, nonpreallocated images have none. It is 5151 * therefore enough to check the first one. 5152 */ 5153 preallocated = s->l1_size > 0 && s->l1_table[0] != 0; 5154 if (qemu_in_coroutine()) { 5155 qemu_co_mutex_unlock(&s->lock); 5156 } 5157 5158 if (!preallocated) { 5159 return 1; 5160 } else if (bs->encrypted) { 5161 return 0; 5162 } else { 5163 return bdrv_has_zero_init(s->data_file->bs); 5164 } 5165 } 5166 5167 static int qcow2_save_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, 5168 int64_t pos) 5169 { 5170 BDRVQcow2State *s = bs->opaque; 5171 5172 BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_SAVE); 5173 return bs->drv->bdrv_co_pwritev_part(bs, qcow2_vm_state_offset(s) + pos, 5174 qiov->size, qiov, 0, 0); 5175 } 5176 5177 static int qcow2_load_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, 5178 int64_t pos) 5179 { 5180 BDRVQcow2State *s = bs->opaque; 5181 5182 BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_LOAD); 5183 return bs->drv->bdrv_co_preadv_part(bs, qcow2_vm_state_offset(s) + pos, 5184 qiov->size, qiov, 0, 0); 5185 } 5186 5187 /* 5188 * Downgrades an image's version. To achieve this, any incompatible features 5189 * have to be removed. 5190 */ 5191 static int qcow2_downgrade(BlockDriverState *bs, int target_version, 5192 BlockDriverAmendStatusCB *status_cb, void *cb_opaque, 5193 Error **errp) 5194 { 5195 BDRVQcow2State *s = bs->opaque; 5196 int current_version = s->qcow_version; 5197 int ret; 5198 int i; 5199 5200 /* This is qcow2_downgrade(), not qcow2_upgrade() */ 5201 assert(target_version < current_version); 5202 5203 /* There are no other versions (now) that you can downgrade to */ 5204 assert(target_version == 2); 5205 5206 if (s->refcount_order != 4) { 5207 error_setg(errp, "compat=0.10 requires refcount_bits=16"); 5208 return -ENOTSUP; 5209 } 5210 5211 if (has_data_file(bs)) { 5212 error_setg(errp, "Cannot downgrade an image with a data file"); 5213 return -ENOTSUP; 5214 } 5215 5216 /* 5217 * If any internal snapshot has a different size than the current 5218 * image size, or VM state size that exceeds 32 bits, downgrading 5219 * is unsafe. Even though we would still use v3-compliant output 5220 * to preserve that data, other v2 programs might not realize 5221 * those optional fields are important. 5222 */ 5223 for (i = 0; i < s->nb_snapshots; i++) { 5224 if (s->snapshots[i].vm_state_size > UINT32_MAX || 5225 s->snapshots[i].disk_size != bs->total_sectors * BDRV_SECTOR_SIZE) { 5226 error_setg(errp, "Internal snapshots prevent downgrade of image"); 5227 return -ENOTSUP; 5228 } 5229 } 5230 5231 /* clear incompatible features */ 5232 if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) { 5233 ret = qcow2_mark_clean(bs); 5234 if (ret < 0) { 5235 error_setg_errno(errp, -ret, "Failed to make the image clean"); 5236 return ret; 5237 } 5238 } 5239 5240 /* with QCOW2_INCOMPAT_CORRUPT, it is pretty much impossible to get here in 5241 * the first place; if that happens nonetheless, returning -ENOTSUP is the 5242 * best thing to do anyway */ 5243 5244 if (s->incompatible_features) { 5245 error_setg(errp, "Cannot downgrade an image with incompatible features " 5246 "%#" PRIx64 " set", s->incompatible_features); 5247 return -ENOTSUP; 5248 } 5249 5250 /* since we can ignore compatible features, we can set them to 0 as well */ 5251 s->compatible_features = 0; 5252 /* if lazy refcounts have been used, they have already been fixed through 5253 * clearing the dirty flag */ 5254 5255 /* clearing autoclear features is trivial */ 5256 s->autoclear_features = 0; 5257 5258 ret = qcow2_expand_zero_clusters(bs, status_cb, cb_opaque); 5259 if (ret < 0) { 5260 error_setg_errno(errp, -ret, "Failed to turn zero into data clusters"); 5261 return ret; 5262 } 5263 5264 s->qcow_version = target_version; 5265 ret = qcow2_update_header(bs); 5266 if (ret < 0) { 5267 s->qcow_version = current_version; 5268 error_setg_errno(errp, -ret, "Failed to update the image header"); 5269 return ret; 5270 } 5271 return 0; 5272 } 5273 5274 /* 5275 * Upgrades an image's version. While newer versions encompass all 5276 * features of older versions, some things may have to be presented 5277 * differently. 5278 */ 5279 static int qcow2_upgrade(BlockDriverState *bs, int target_version, 5280 BlockDriverAmendStatusCB *status_cb, void *cb_opaque, 5281 Error **errp) 5282 { 5283 BDRVQcow2State *s = bs->opaque; 5284 bool need_snapshot_update; 5285 int current_version = s->qcow_version; 5286 int i; 5287 int ret; 5288 5289 /* This is qcow2_upgrade(), not qcow2_downgrade() */ 5290 assert(target_version > current_version); 5291 5292 /* There are no other versions (yet) that you can upgrade to */ 5293 assert(target_version == 3); 5294 5295 status_cb(bs, 0, 2, cb_opaque); 5296 5297 /* 5298 * In v2, snapshots do not need to have extra data. v3 requires 5299 * the 64-bit VM state size and the virtual disk size to be 5300 * present. 5301 * qcow2_write_snapshots() will always write the list in the 5302 * v3-compliant format. 5303 */ 5304 need_snapshot_update = false; 5305 for (i = 0; i < s->nb_snapshots; i++) { 5306 if (s->snapshots[i].extra_data_size < 5307 sizeof_field(QCowSnapshotExtraData, vm_state_size_large) + 5308 sizeof_field(QCowSnapshotExtraData, disk_size)) 5309 { 5310 need_snapshot_update = true; 5311 break; 5312 } 5313 } 5314 if (need_snapshot_update) { 5315 ret = qcow2_write_snapshots(bs); 5316 if (ret < 0) { 5317 error_setg_errno(errp, -ret, "Failed to update the snapshot table"); 5318 return ret; 5319 } 5320 } 5321 status_cb(bs, 1, 2, cb_opaque); 5322 5323 s->qcow_version = target_version; 5324 ret = qcow2_update_header(bs); 5325 if (ret < 0) { 5326 s->qcow_version = current_version; 5327 error_setg_errno(errp, -ret, "Failed to update the image header"); 5328 return ret; 5329 } 5330 status_cb(bs, 2, 2, cb_opaque); 5331 5332 return 0; 5333 } 5334 5335 typedef enum Qcow2AmendOperation { 5336 /* This is the value Qcow2AmendHelperCBInfo::last_operation will be 5337 * statically initialized to so that the helper CB can discern the first 5338 * invocation from an operation change */ 5339 QCOW2_NO_OPERATION = 0, 5340 5341 QCOW2_UPGRADING, 5342 QCOW2_UPDATING_ENCRYPTION, 5343 QCOW2_CHANGING_REFCOUNT_ORDER, 5344 QCOW2_DOWNGRADING, 5345 } Qcow2AmendOperation; 5346 5347 typedef struct Qcow2AmendHelperCBInfo { 5348 /* The code coordinating the amend operations should only modify 5349 * these four fields; the rest will be managed by the CB */ 5350 BlockDriverAmendStatusCB *original_status_cb; 5351 void *original_cb_opaque; 5352 5353 Qcow2AmendOperation current_operation; 5354 5355 /* Total number of operations to perform (only set once) */ 5356 int total_operations; 5357 5358 /* The following fields are managed by the CB */ 5359 5360 /* Number of operations completed */ 5361 int operations_completed; 5362 5363 /* Cumulative offset of all completed operations */ 5364 int64_t offset_completed; 5365 5366 Qcow2AmendOperation last_operation; 5367 int64_t last_work_size; 5368 } Qcow2AmendHelperCBInfo; 5369 5370 static void qcow2_amend_helper_cb(BlockDriverState *bs, 5371 int64_t operation_offset, 5372 int64_t operation_work_size, void *opaque) 5373 { 5374 Qcow2AmendHelperCBInfo *info = opaque; 5375 int64_t current_work_size; 5376 int64_t projected_work_size; 5377 5378 if (info->current_operation != info->last_operation) { 5379 if (info->last_operation != QCOW2_NO_OPERATION) { 5380 info->offset_completed += info->last_work_size; 5381 info->operations_completed++; 5382 } 5383 5384 info->last_operation = info->current_operation; 5385 } 5386 5387 assert(info->total_operations > 0); 5388 assert(info->operations_completed < info->total_operations); 5389 5390 info->last_work_size = operation_work_size; 5391 5392 current_work_size = info->offset_completed + operation_work_size; 5393 5394 /* current_work_size is the total work size for (operations_completed + 1) 5395 * operations (which includes this one), so multiply it by the number of 5396 * operations not covered and divide it by the number of operations 5397 * covered to get a projection for the operations not covered */ 5398 projected_work_size = current_work_size * (info->total_operations - 5399 info->operations_completed - 1) 5400 / (info->operations_completed + 1); 5401 5402 info->original_status_cb(bs, info->offset_completed + operation_offset, 5403 current_work_size + projected_work_size, 5404 info->original_cb_opaque); 5405 } 5406 5407 static int qcow2_amend_options(BlockDriverState *bs, QemuOpts *opts, 5408 BlockDriverAmendStatusCB *status_cb, 5409 void *cb_opaque, 5410 bool force, 5411 Error **errp) 5412 { 5413 BDRVQcow2State *s = bs->opaque; 5414 int old_version = s->qcow_version, new_version = old_version; 5415 uint64_t new_size = 0; 5416 const char *backing_file = NULL, *backing_format = NULL, *data_file = NULL; 5417 bool lazy_refcounts = s->use_lazy_refcounts; 5418 bool data_file_raw = data_file_is_raw(bs); 5419 const char *compat = NULL; 5420 int refcount_bits = s->refcount_bits; 5421 int ret; 5422 QemuOptDesc *desc = opts->list->desc; 5423 Qcow2AmendHelperCBInfo helper_cb_info; 5424 bool encryption_update = false; 5425 5426 while (desc && desc->name) { 5427 if (!qemu_opt_find(opts, desc->name)) { 5428 /* only change explicitly defined options */ 5429 desc++; 5430 continue; 5431 } 5432 5433 if (!strcmp(desc->name, BLOCK_OPT_COMPAT_LEVEL)) { 5434 compat = qemu_opt_get(opts, BLOCK_OPT_COMPAT_LEVEL); 5435 if (!compat) { 5436 /* preserve default */ 5437 } else if (!strcmp(compat, "0.10") || !strcmp(compat, "v2")) { 5438 new_version = 2; 5439 } else if (!strcmp(compat, "1.1") || !strcmp(compat, "v3")) { 5440 new_version = 3; 5441 } else { 5442 error_setg(errp, "Unknown compatibility level %s", compat); 5443 return -EINVAL; 5444 } 5445 } else if (!strcmp(desc->name, BLOCK_OPT_SIZE)) { 5446 new_size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, 0); 5447 } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FILE)) { 5448 backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE); 5449 } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FMT)) { 5450 backing_format = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT); 5451 } else if (g_str_has_prefix(desc->name, "encrypt.")) { 5452 if (!s->crypto) { 5453 error_setg(errp, 5454 "Can't amend encryption options - encryption not present"); 5455 return -EINVAL; 5456 } 5457 if (s->crypt_method_header != QCOW_CRYPT_LUKS) { 5458 error_setg(errp, 5459 "Only LUKS encryption options can be amended"); 5460 return -ENOTSUP; 5461 } 5462 encryption_update = true; 5463 } else if (!strcmp(desc->name, BLOCK_OPT_LAZY_REFCOUNTS)) { 5464 lazy_refcounts = qemu_opt_get_bool(opts, BLOCK_OPT_LAZY_REFCOUNTS, 5465 lazy_refcounts); 5466 } else if (!strcmp(desc->name, BLOCK_OPT_REFCOUNT_BITS)) { 5467 refcount_bits = qemu_opt_get_number(opts, BLOCK_OPT_REFCOUNT_BITS, 5468 refcount_bits); 5469 5470 if (refcount_bits <= 0 || refcount_bits > 64 || 5471 !is_power_of_2(refcount_bits)) 5472 { 5473 error_setg(errp, "Refcount width must be a power of two and " 5474 "may not exceed 64 bits"); 5475 return -EINVAL; 5476 } 5477 } else if (!strcmp(desc->name, BLOCK_OPT_DATA_FILE)) { 5478 data_file = qemu_opt_get(opts, BLOCK_OPT_DATA_FILE); 5479 if (data_file && !has_data_file(bs)) { 5480 error_setg(errp, "data-file can only be set for images that " 5481 "use an external data file"); 5482 return -EINVAL; 5483 } 5484 } else if (!strcmp(desc->name, BLOCK_OPT_DATA_FILE_RAW)) { 5485 data_file_raw = qemu_opt_get_bool(opts, BLOCK_OPT_DATA_FILE_RAW, 5486 data_file_raw); 5487 if (data_file_raw && !data_file_is_raw(bs)) { 5488 error_setg(errp, "data-file-raw cannot be set on existing " 5489 "images"); 5490 return -EINVAL; 5491 } 5492 } else { 5493 /* if this point is reached, this probably means a new option was 5494 * added without having it covered here */ 5495 abort(); 5496 } 5497 5498 desc++; 5499 } 5500 5501 helper_cb_info = (Qcow2AmendHelperCBInfo){ 5502 .original_status_cb = status_cb, 5503 .original_cb_opaque = cb_opaque, 5504 .total_operations = (new_version != old_version) 5505 + (s->refcount_bits != refcount_bits) + 5506 (encryption_update == true) 5507 }; 5508 5509 /* Upgrade first (some features may require compat=1.1) */ 5510 if (new_version > old_version) { 5511 helper_cb_info.current_operation = QCOW2_UPGRADING; 5512 ret = qcow2_upgrade(bs, new_version, &qcow2_amend_helper_cb, 5513 &helper_cb_info, errp); 5514 if (ret < 0) { 5515 return ret; 5516 } 5517 } 5518 5519 if (encryption_update) { 5520 QDict *amend_opts_dict; 5521 QCryptoBlockAmendOptions *amend_opts; 5522 5523 helper_cb_info.current_operation = QCOW2_UPDATING_ENCRYPTION; 5524 amend_opts_dict = qcow2_extract_crypto_opts(opts, "luks", errp); 5525 if (!amend_opts_dict) { 5526 return -EINVAL; 5527 } 5528 amend_opts = block_crypto_amend_opts_init(amend_opts_dict, errp); 5529 qobject_unref(amend_opts_dict); 5530 if (!amend_opts) { 5531 return -EINVAL; 5532 } 5533 ret = qcrypto_block_amend_options(s->crypto, 5534 qcow2_crypto_hdr_read_func, 5535 qcow2_crypto_hdr_write_func, 5536 bs, 5537 amend_opts, 5538 force, 5539 errp); 5540 qapi_free_QCryptoBlockAmendOptions(amend_opts); 5541 if (ret < 0) { 5542 return ret; 5543 } 5544 } 5545 5546 if (s->refcount_bits != refcount_bits) { 5547 int refcount_order = ctz32(refcount_bits); 5548 5549 if (new_version < 3 && refcount_bits != 16) { 5550 error_setg(errp, "Refcount widths other than 16 bits require " 5551 "compatibility level 1.1 or above (use compat=1.1 or " 5552 "greater)"); 5553 return -EINVAL; 5554 } 5555 5556 helper_cb_info.current_operation = QCOW2_CHANGING_REFCOUNT_ORDER; 5557 ret = qcow2_change_refcount_order(bs, refcount_order, 5558 &qcow2_amend_helper_cb, 5559 &helper_cb_info, errp); 5560 if (ret < 0) { 5561 return ret; 5562 } 5563 } 5564 5565 /* data-file-raw blocks backing files, so clear it first if requested */ 5566 if (data_file_raw) { 5567 s->autoclear_features |= QCOW2_AUTOCLEAR_DATA_FILE_RAW; 5568 } else { 5569 s->autoclear_features &= ~QCOW2_AUTOCLEAR_DATA_FILE_RAW; 5570 } 5571 5572 if (data_file) { 5573 g_free(s->image_data_file); 5574 s->image_data_file = *data_file ? g_strdup(data_file) : NULL; 5575 } 5576 5577 ret = qcow2_update_header(bs); 5578 if (ret < 0) { 5579 error_setg_errno(errp, -ret, "Failed to update the image header"); 5580 return ret; 5581 } 5582 5583 if (backing_file || backing_format) { 5584 if (g_strcmp0(backing_file, s->image_backing_file) || 5585 g_strcmp0(backing_format, s->image_backing_format)) { 5586 warn_report("Deprecated use of amend to alter the backing file; " 5587 "use qemu-img rebase instead"); 5588 } 5589 ret = qcow2_change_backing_file(bs, 5590 backing_file ?: s->image_backing_file, 5591 backing_format ?: s->image_backing_format); 5592 if (ret < 0) { 5593 error_setg_errno(errp, -ret, "Failed to change the backing file"); 5594 return ret; 5595 } 5596 } 5597 5598 if (s->use_lazy_refcounts != lazy_refcounts) { 5599 if (lazy_refcounts) { 5600 if (new_version < 3) { 5601 error_setg(errp, "Lazy refcounts only supported with " 5602 "compatibility level 1.1 and above (use compat=1.1 " 5603 "or greater)"); 5604 return -EINVAL; 5605 } 5606 s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS; 5607 ret = qcow2_update_header(bs); 5608 if (ret < 0) { 5609 s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS; 5610 error_setg_errno(errp, -ret, "Failed to update the image header"); 5611 return ret; 5612 } 5613 s->use_lazy_refcounts = true; 5614 } else { 5615 /* make image clean first */ 5616 ret = qcow2_mark_clean(bs); 5617 if (ret < 0) { 5618 error_setg_errno(errp, -ret, "Failed to make the image clean"); 5619 return ret; 5620 } 5621 /* now disallow lazy refcounts */ 5622 s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS; 5623 ret = qcow2_update_header(bs); 5624 if (ret < 0) { 5625 s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS; 5626 error_setg_errno(errp, -ret, "Failed to update the image header"); 5627 return ret; 5628 } 5629 s->use_lazy_refcounts = false; 5630 } 5631 } 5632 5633 if (new_size) { 5634 BlockBackend *blk = blk_new_with_bs(bs, BLK_PERM_RESIZE, BLK_PERM_ALL, 5635 errp); 5636 if (!blk) { 5637 return -EPERM; 5638 } 5639 5640 /* 5641 * Amending image options should ensure that the image has 5642 * exactly the given new values, so pass exact=true here. 5643 */ 5644 ret = blk_truncate(blk, new_size, true, PREALLOC_MODE_OFF, 0, errp); 5645 blk_unref(blk); 5646 if (ret < 0) { 5647 return ret; 5648 } 5649 } 5650 5651 /* Downgrade last (so unsupported features can be removed before) */ 5652 if (new_version < old_version) { 5653 helper_cb_info.current_operation = QCOW2_DOWNGRADING; 5654 ret = qcow2_downgrade(bs, new_version, &qcow2_amend_helper_cb, 5655 &helper_cb_info, errp); 5656 if (ret < 0) { 5657 return ret; 5658 } 5659 } 5660 5661 return 0; 5662 } 5663 5664 static int coroutine_fn qcow2_co_amend(BlockDriverState *bs, 5665 BlockdevAmendOptions *opts, 5666 bool force, 5667 Error **errp) 5668 { 5669 BlockdevAmendOptionsQcow2 *qopts = &opts->u.qcow2; 5670 BDRVQcow2State *s = bs->opaque; 5671 int ret = 0; 5672 5673 if (qopts->has_encrypt) { 5674 if (!s->crypto) { 5675 error_setg(errp, "image is not encrypted, can't amend"); 5676 return -EOPNOTSUPP; 5677 } 5678 5679 if (qopts->encrypt->format != Q_CRYPTO_BLOCK_FORMAT_LUKS) { 5680 error_setg(errp, 5681 "Amend can't be used to change the qcow2 encryption format"); 5682 return -EOPNOTSUPP; 5683 } 5684 5685 if (s->crypt_method_header != QCOW_CRYPT_LUKS) { 5686 error_setg(errp, 5687 "Only LUKS encryption options can be amended for qcow2 with blockdev-amend"); 5688 return -EOPNOTSUPP; 5689 } 5690 5691 ret = qcrypto_block_amend_options(s->crypto, 5692 qcow2_crypto_hdr_read_func, 5693 qcow2_crypto_hdr_write_func, 5694 bs, 5695 qopts->encrypt, 5696 force, 5697 errp); 5698 } 5699 return ret; 5700 } 5701 5702 /* 5703 * If offset or size are negative, respectively, they will not be included in 5704 * the BLOCK_IMAGE_CORRUPTED event emitted. 5705 * fatal will be ignored for read-only BDS; corruptions found there will always 5706 * be considered non-fatal. 5707 */ 5708 void qcow2_signal_corruption(BlockDriverState *bs, bool fatal, int64_t offset, 5709 int64_t size, const char *message_format, ...) 5710 { 5711 BDRVQcow2State *s = bs->opaque; 5712 const char *node_name; 5713 char *message; 5714 va_list ap; 5715 5716 fatal = fatal && bdrv_is_writable(bs); 5717 5718 if (s->signaled_corruption && 5719 (!fatal || (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT))) 5720 { 5721 return; 5722 } 5723 5724 va_start(ap, message_format); 5725 message = g_strdup_vprintf(message_format, ap); 5726 va_end(ap); 5727 5728 if (fatal) { 5729 fprintf(stderr, "qcow2: Marking image as corrupt: %s; further " 5730 "corruption events will be suppressed\n", message); 5731 } else { 5732 fprintf(stderr, "qcow2: Image is corrupt: %s; further non-fatal " 5733 "corruption events will be suppressed\n", message); 5734 } 5735 5736 node_name = bdrv_get_node_name(bs); 5737 qapi_event_send_block_image_corrupted(bdrv_get_device_name(bs), 5738 *node_name != '\0', node_name, 5739 message, offset >= 0, offset, 5740 size >= 0, size, 5741 fatal); 5742 g_free(message); 5743 5744 if (fatal) { 5745 qcow2_mark_corrupt(bs); 5746 bs->drv = NULL; /* make BDS unusable */ 5747 } 5748 5749 s->signaled_corruption = true; 5750 } 5751 5752 #define QCOW_COMMON_OPTIONS \ 5753 { \ 5754 .name = BLOCK_OPT_SIZE, \ 5755 .type = QEMU_OPT_SIZE, \ 5756 .help = "Virtual disk size" \ 5757 }, \ 5758 { \ 5759 .name = BLOCK_OPT_COMPAT_LEVEL, \ 5760 .type = QEMU_OPT_STRING, \ 5761 .help = "Compatibility level (v2 [0.10] or v3 [1.1])" \ 5762 }, \ 5763 { \ 5764 .name = BLOCK_OPT_BACKING_FILE, \ 5765 .type = QEMU_OPT_STRING, \ 5766 .help = "File name of a base image" \ 5767 }, \ 5768 { \ 5769 .name = BLOCK_OPT_BACKING_FMT, \ 5770 .type = QEMU_OPT_STRING, \ 5771 .help = "Image format of the base image" \ 5772 }, \ 5773 { \ 5774 .name = BLOCK_OPT_DATA_FILE, \ 5775 .type = QEMU_OPT_STRING, \ 5776 .help = "File name of an external data file" \ 5777 }, \ 5778 { \ 5779 .name = BLOCK_OPT_DATA_FILE_RAW, \ 5780 .type = QEMU_OPT_BOOL, \ 5781 .help = "The external data file must stay valid " \ 5782 "as a raw image" \ 5783 }, \ 5784 { \ 5785 .name = BLOCK_OPT_LAZY_REFCOUNTS, \ 5786 .type = QEMU_OPT_BOOL, \ 5787 .help = "Postpone refcount updates", \ 5788 .def_value_str = "off" \ 5789 }, \ 5790 { \ 5791 .name = BLOCK_OPT_REFCOUNT_BITS, \ 5792 .type = QEMU_OPT_NUMBER, \ 5793 .help = "Width of a reference count entry in bits", \ 5794 .def_value_str = "16" \ 5795 } 5796 5797 static QemuOptsList qcow2_create_opts = { 5798 .name = "qcow2-create-opts", 5799 .head = QTAILQ_HEAD_INITIALIZER(qcow2_create_opts.head), 5800 .desc = { 5801 { \ 5802 .name = BLOCK_OPT_ENCRYPT, \ 5803 .type = QEMU_OPT_BOOL, \ 5804 .help = "Encrypt the image with format 'aes'. (Deprecated " \ 5805 "in favor of " BLOCK_OPT_ENCRYPT_FORMAT "=aes)", \ 5806 }, \ 5807 { \ 5808 .name = BLOCK_OPT_ENCRYPT_FORMAT, \ 5809 .type = QEMU_OPT_STRING, \ 5810 .help = "Encrypt the image, format choices: 'aes', 'luks'", \ 5811 }, \ 5812 BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.", \ 5813 "ID of secret providing qcow AES key or LUKS passphrase"), \ 5814 BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_ALG("encrypt."), \ 5815 BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_MODE("encrypt."), \ 5816 BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_ALG("encrypt."), \ 5817 BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_HASH_ALG("encrypt."), \ 5818 BLOCK_CRYPTO_OPT_DEF_LUKS_HASH_ALG("encrypt."), \ 5819 BLOCK_CRYPTO_OPT_DEF_LUKS_ITER_TIME("encrypt."), \ 5820 { \ 5821 .name = BLOCK_OPT_CLUSTER_SIZE, \ 5822 .type = QEMU_OPT_SIZE, \ 5823 .help = "qcow2 cluster size", \ 5824 .def_value_str = stringify(DEFAULT_CLUSTER_SIZE) \ 5825 }, \ 5826 { \ 5827 .name = BLOCK_OPT_EXTL2, \ 5828 .type = QEMU_OPT_BOOL, \ 5829 .help = "Extended L2 tables", \ 5830 .def_value_str = "off" \ 5831 }, \ 5832 { \ 5833 .name = BLOCK_OPT_PREALLOC, \ 5834 .type = QEMU_OPT_STRING, \ 5835 .help = "Preallocation mode (allowed values: off, " \ 5836 "metadata, falloc, full)" \ 5837 }, \ 5838 { \ 5839 .name = BLOCK_OPT_COMPRESSION_TYPE, \ 5840 .type = QEMU_OPT_STRING, \ 5841 .help = "Compression method used for image cluster " \ 5842 "compression", \ 5843 .def_value_str = "zlib" \ 5844 }, 5845 QCOW_COMMON_OPTIONS, 5846 { /* end of list */ } 5847 } 5848 }; 5849 5850 static QemuOptsList qcow2_amend_opts = { 5851 .name = "qcow2-amend-opts", 5852 .head = QTAILQ_HEAD_INITIALIZER(qcow2_amend_opts.head), 5853 .desc = { 5854 BLOCK_CRYPTO_OPT_DEF_LUKS_STATE("encrypt."), 5855 BLOCK_CRYPTO_OPT_DEF_LUKS_KEYSLOT("encrypt."), 5856 BLOCK_CRYPTO_OPT_DEF_LUKS_OLD_SECRET("encrypt."), 5857 BLOCK_CRYPTO_OPT_DEF_LUKS_NEW_SECRET("encrypt."), 5858 BLOCK_CRYPTO_OPT_DEF_LUKS_ITER_TIME("encrypt."), 5859 QCOW_COMMON_OPTIONS, 5860 { /* end of list */ } 5861 } 5862 }; 5863 5864 static const char *const qcow2_strong_runtime_opts[] = { 5865 "encrypt." BLOCK_CRYPTO_OPT_QCOW_KEY_SECRET, 5866 5867 NULL 5868 }; 5869 5870 BlockDriver bdrv_qcow2 = { 5871 .format_name = "qcow2", 5872 .instance_size = sizeof(BDRVQcow2State), 5873 .bdrv_probe = qcow2_probe, 5874 .bdrv_open = qcow2_open, 5875 .bdrv_close = qcow2_close, 5876 .bdrv_reopen_prepare = qcow2_reopen_prepare, 5877 .bdrv_reopen_commit = qcow2_reopen_commit, 5878 .bdrv_reopen_commit_post = qcow2_reopen_commit_post, 5879 .bdrv_reopen_abort = qcow2_reopen_abort, 5880 .bdrv_join_options = qcow2_join_options, 5881 .bdrv_child_perm = bdrv_default_perms, 5882 .bdrv_co_create_opts = qcow2_co_create_opts, 5883 .bdrv_co_create = qcow2_co_create, 5884 .bdrv_has_zero_init = qcow2_has_zero_init, 5885 .bdrv_co_block_status = qcow2_co_block_status, 5886 5887 .bdrv_co_preadv_part = qcow2_co_preadv_part, 5888 .bdrv_co_pwritev_part = qcow2_co_pwritev_part, 5889 .bdrv_co_flush_to_os = qcow2_co_flush_to_os, 5890 5891 .bdrv_co_pwrite_zeroes = qcow2_co_pwrite_zeroes, 5892 .bdrv_co_pdiscard = qcow2_co_pdiscard, 5893 .bdrv_co_copy_range_from = qcow2_co_copy_range_from, 5894 .bdrv_co_copy_range_to = qcow2_co_copy_range_to, 5895 .bdrv_co_truncate = qcow2_co_truncate, 5896 .bdrv_co_pwritev_compressed_part = qcow2_co_pwritev_compressed_part, 5897 .bdrv_make_empty = qcow2_make_empty, 5898 5899 .bdrv_snapshot_create = qcow2_snapshot_create, 5900 .bdrv_snapshot_goto = qcow2_snapshot_goto, 5901 .bdrv_snapshot_delete = qcow2_snapshot_delete, 5902 .bdrv_snapshot_list = qcow2_snapshot_list, 5903 .bdrv_snapshot_load_tmp = qcow2_snapshot_load_tmp, 5904 .bdrv_measure = qcow2_measure, 5905 .bdrv_get_info = qcow2_get_info, 5906 .bdrv_get_specific_info = qcow2_get_specific_info, 5907 5908 .bdrv_save_vmstate = qcow2_save_vmstate, 5909 .bdrv_load_vmstate = qcow2_load_vmstate, 5910 5911 .is_format = true, 5912 .supports_backing = true, 5913 .bdrv_change_backing_file = qcow2_change_backing_file, 5914 5915 .bdrv_refresh_limits = qcow2_refresh_limits, 5916 .bdrv_co_invalidate_cache = qcow2_co_invalidate_cache, 5917 .bdrv_inactivate = qcow2_inactivate, 5918 5919 .create_opts = &qcow2_create_opts, 5920 .amend_opts = &qcow2_amend_opts, 5921 .strong_runtime_opts = qcow2_strong_runtime_opts, 5922 .mutable_opts = mutable_opts, 5923 .bdrv_co_check = qcow2_co_check, 5924 .bdrv_amend_options = qcow2_amend_options, 5925 .bdrv_co_amend = qcow2_co_amend, 5926 5927 .bdrv_detach_aio_context = qcow2_detach_aio_context, 5928 .bdrv_attach_aio_context = qcow2_attach_aio_context, 5929 5930 .bdrv_supports_persistent_dirty_bitmap = 5931 qcow2_supports_persistent_dirty_bitmap, 5932 .bdrv_co_can_store_new_dirty_bitmap = qcow2_co_can_store_new_dirty_bitmap, 5933 .bdrv_co_remove_persistent_dirty_bitmap = 5934 qcow2_co_remove_persistent_dirty_bitmap, 5935 }; 5936 5937 static void bdrv_qcow2_init(void) 5938 { 5939 bdrv_register(&bdrv_qcow2); 5940 } 5941 5942 block_init(bdrv_qcow2_init); 5943