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 #define ZLIB_CONST 28 #include <zlib.h> 29 30 #include "block/block_int.h" 31 #include "block/qdict.h" 32 #include "sysemu/block-backend.h" 33 #include "qemu/module.h" 34 #include "qcow2.h" 35 #include "qemu/error-report.h" 36 #include "qapi/error.h" 37 #include "qapi/qapi-events-block-core.h" 38 #include "qapi/qmp/qdict.h" 39 #include "qapi/qmp/qstring.h" 40 #include "trace.h" 41 #include "qemu/option_int.h" 42 #include "qemu/cutils.h" 43 #include "qemu/bswap.h" 44 #include "qapi/qobject-input-visitor.h" 45 #include "qapi/qapi-visit-block-core.h" 46 #include "crypto.h" 47 #include "block/thread-pool.h" 48 49 /* 50 Differences with QCOW: 51 52 - Support for multiple incremental snapshots. 53 - Memory management by reference counts. 54 - Clusters which have a reference count of one have the bit 55 QCOW_OFLAG_COPIED to optimize write performance. 56 - Size of compressed clusters is stored in sectors to reduce bit usage 57 in the cluster offsets. 58 - Support for storing additional data (such as the VM state) in the 59 snapshots. 60 - If a backing store is used, the cluster size is not constrained 61 (could be backported to QCOW). 62 - L2 tables have always a size of one cluster. 63 */ 64 65 66 typedef struct { 67 uint32_t magic; 68 uint32_t len; 69 } QEMU_PACKED QCowExtension; 70 71 #define QCOW2_EXT_MAGIC_END 0 72 #define QCOW2_EXT_MAGIC_BACKING_FORMAT 0xE2792ACA 73 #define QCOW2_EXT_MAGIC_FEATURE_TABLE 0x6803f857 74 #define QCOW2_EXT_MAGIC_CRYPTO_HEADER 0x0537be77 75 #define QCOW2_EXT_MAGIC_BITMAPS 0x23852875 76 77 static int coroutine_fn 78 qcow2_co_preadv_compressed(BlockDriverState *bs, 79 uint64_t file_cluster_offset, 80 uint64_t offset, 81 uint64_t bytes, 82 QEMUIOVector *qiov); 83 84 static int qcow2_probe(const uint8_t *buf, int buf_size, const char *filename) 85 { 86 const QCowHeader *cow_header = (const void *)buf; 87 88 if (buf_size >= sizeof(QCowHeader) && 89 be32_to_cpu(cow_header->magic) == QCOW_MAGIC && 90 be32_to_cpu(cow_header->version) >= 2) 91 return 100; 92 else 93 return 0; 94 } 95 96 97 static ssize_t qcow2_crypto_hdr_read_func(QCryptoBlock *block, size_t offset, 98 uint8_t *buf, size_t buflen, 99 void *opaque, Error **errp) 100 { 101 BlockDriverState *bs = opaque; 102 BDRVQcow2State *s = bs->opaque; 103 ssize_t ret; 104 105 if ((offset + buflen) > s->crypto_header.length) { 106 error_setg(errp, "Request for data outside of extension header"); 107 return -1; 108 } 109 110 ret = bdrv_pread(bs->file, 111 s->crypto_header.offset + offset, buf, buflen); 112 if (ret < 0) { 113 error_setg_errno(errp, -ret, "Could not read encryption header"); 114 return -1; 115 } 116 return ret; 117 } 118 119 120 static ssize_t qcow2_crypto_hdr_init_func(QCryptoBlock *block, size_t headerlen, 121 void *opaque, Error **errp) 122 { 123 BlockDriverState *bs = opaque; 124 BDRVQcow2State *s = bs->opaque; 125 int64_t ret; 126 int64_t clusterlen; 127 128 ret = qcow2_alloc_clusters(bs, headerlen); 129 if (ret < 0) { 130 error_setg_errno(errp, -ret, 131 "Cannot allocate cluster for LUKS header size %zu", 132 headerlen); 133 return -1; 134 } 135 136 s->crypto_header.length = headerlen; 137 s->crypto_header.offset = ret; 138 139 /* Zero fill remaining space in cluster so it has predictable 140 * content in case of future spec changes */ 141 clusterlen = size_to_clusters(s, headerlen) * s->cluster_size; 142 assert(qcow2_pre_write_overlap_check(bs, 0, ret, clusterlen) == 0); 143 ret = bdrv_pwrite_zeroes(bs->file, 144 ret + headerlen, 145 clusterlen - headerlen, 0); 146 if (ret < 0) { 147 error_setg_errno(errp, -ret, "Could not zero fill encryption header"); 148 return -1; 149 } 150 151 return ret; 152 } 153 154 155 static ssize_t qcow2_crypto_hdr_write_func(QCryptoBlock *block, size_t offset, 156 const uint8_t *buf, size_t buflen, 157 void *opaque, Error **errp) 158 { 159 BlockDriverState *bs = opaque; 160 BDRVQcow2State *s = bs->opaque; 161 ssize_t ret; 162 163 if ((offset + buflen) > s->crypto_header.length) { 164 error_setg(errp, "Request for data outside of extension header"); 165 return -1; 166 } 167 168 ret = bdrv_pwrite(bs->file, 169 s->crypto_header.offset + offset, buf, buflen); 170 if (ret < 0) { 171 error_setg_errno(errp, -ret, "Could not read encryption header"); 172 return -1; 173 } 174 return ret; 175 } 176 177 178 /* 179 * read qcow2 extension and fill bs 180 * start reading from start_offset 181 * finish reading upon magic of value 0 or when end_offset reached 182 * unknown magic is skipped (future extension this version knows nothing about) 183 * return 0 upon success, non-0 otherwise 184 */ 185 static int qcow2_read_extensions(BlockDriverState *bs, uint64_t start_offset, 186 uint64_t end_offset, void **p_feature_table, 187 int flags, bool *need_update_header, 188 Error **errp) 189 { 190 BDRVQcow2State *s = bs->opaque; 191 QCowExtension ext; 192 uint64_t offset; 193 int ret; 194 Qcow2BitmapHeaderExt bitmaps_ext; 195 196 if (need_update_header != NULL) { 197 *need_update_header = false; 198 } 199 200 #ifdef DEBUG_EXT 201 printf("qcow2_read_extensions: start=%ld end=%ld\n", start_offset, end_offset); 202 #endif 203 offset = start_offset; 204 while (offset < end_offset) { 205 206 #ifdef DEBUG_EXT 207 /* Sanity check */ 208 if (offset > s->cluster_size) 209 printf("qcow2_read_extension: suspicious offset %lu\n", offset); 210 211 printf("attempting to read extended header in offset %lu\n", offset); 212 #endif 213 214 ret = bdrv_pread(bs->file, offset, &ext, sizeof(ext)); 215 if (ret < 0) { 216 error_setg_errno(errp, -ret, "qcow2_read_extension: ERROR: " 217 "pread fail from offset %" PRIu64, offset); 218 return 1; 219 } 220 ext.magic = be32_to_cpu(ext.magic); 221 ext.len = be32_to_cpu(ext.len); 222 offset += sizeof(ext); 223 #ifdef DEBUG_EXT 224 printf("ext.magic = 0x%x\n", ext.magic); 225 #endif 226 if (offset > end_offset || ext.len > end_offset - offset) { 227 error_setg(errp, "Header extension too large"); 228 return -EINVAL; 229 } 230 231 switch (ext.magic) { 232 case QCOW2_EXT_MAGIC_END: 233 return 0; 234 235 case QCOW2_EXT_MAGIC_BACKING_FORMAT: 236 if (ext.len >= sizeof(bs->backing_format)) { 237 error_setg(errp, "ERROR: ext_backing_format: len=%" PRIu32 238 " too large (>=%zu)", ext.len, 239 sizeof(bs->backing_format)); 240 return 2; 241 } 242 ret = bdrv_pread(bs->file, offset, bs->backing_format, ext.len); 243 if (ret < 0) { 244 error_setg_errno(errp, -ret, "ERROR: ext_backing_format: " 245 "Could not read format name"); 246 return 3; 247 } 248 bs->backing_format[ext.len] = '\0'; 249 s->image_backing_format = g_strdup(bs->backing_format); 250 #ifdef DEBUG_EXT 251 printf("Qcow2: Got format extension %s\n", bs->backing_format); 252 #endif 253 break; 254 255 case QCOW2_EXT_MAGIC_FEATURE_TABLE: 256 if (p_feature_table != NULL) { 257 void* feature_table = g_malloc0(ext.len + 2 * sizeof(Qcow2Feature)); 258 ret = bdrv_pread(bs->file, offset , feature_table, ext.len); 259 if (ret < 0) { 260 error_setg_errno(errp, -ret, "ERROR: ext_feature_table: " 261 "Could not read table"); 262 return ret; 263 } 264 265 *p_feature_table = feature_table; 266 } 267 break; 268 269 case QCOW2_EXT_MAGIC_CRYPTO_HEADER: { 270 unsigned int cflags = 0; 271 if (s->crypt_method_header != QCOW_CRYPT_LUKS) { 272 error_setg(errp, "CRYPTO header extension only " 273 "expected with LUKS encryption method"); 274 return -EINVAL; 275 } 276 if (ext.len != sizeof(Qcow2CryptoHeaderExtension)) { 277 error_setg(errp, "CRYPTO header extension size %u, " 278 "but expected size %zu", ext.len, 279 sizeof(Qcow2CryptoHeaderExtension)); 280 return -EINVAL; 281 } 282 283 ret = bdrv_pread(bs->file, offset, &s->crypto_header, ext.len); 284 if (ret < 0) { 285 error_setg_errno(errp, -ret, 286 "Unable to read CRYPTO header extension"); 287 return ret; 288 } 289 s->crypto_header.offset = be64_to_cpu(s->crypto_header.offset); 290 s->crypto_header.length = be64_to_cpu(s->crypto_header.length); 291 292 if ((s->crypto_header.offset % s->cluster_size) != 0) { 293 error_setg(errp, "Encryption header offset '%" PRIu64 "' is " 294 "not a multiple of cluster size '%u'", 295 s->crypto_header.offset, s->cluster_size); 296 return -EINVAL; 297 } 298 299 if (flags & BDRV_O_NO_IO) { 300 cflags |= QCRYPTO_BLOCK_OPEN_NO_IO; 301 } 302 s->crypto = qcrypto_block_open(s->crypto_opts, "encrypt.", 303 qcow2_crypto_hdr_read_func, 304 bs, cflags, 1, errp); 305 if (!s->crypto) { 306 return -EINVAL; 307 } 308 } break; 309 310 case QCOW2_EXT_MAGIC_BITMAPS: 311 if (ext.len != sizeof(bitmaps_ext)) { 312 error_setg_errno(errp, -ret, "bitmaps_ext: " 313 "Invalid extension length"); 314 return -EINVAL; 315 } 316 317 if (!(s->autoclear_features & QCOW2_AUTOCLEAR_BITMAPS)) { 318 if (s->qcow_version < 3) { 319 /* Let's be a bit more specific */ 320 warn_report("This qcow2 v2 image contains bitmaps, but " 321 "they may have been modified by a program " 322 "without persistent bitmap support; so now " 323 "they must all be considered inconsistent"); 324 } else { 325 warn_report("a program lacking bitmap support " 326 "modified this file, so all bitmaps are now " 327 "considered inconsistent"); 328 } 329 error_printf("Some clusters may be leaked, " 330 "run 'qemu-img check -r' on the image " 331 "file to fix."); 332 if (need_update_header != NULL) { 333 /* Updating is needed to drop invalid bitmap extension. */ 334 *need_update_header = true; 335 } 336 break; 337 } 338 339 ret = bdrv_pread(bs->file, offset, &bitmaps_ext, ext.len); 340 if (ret < 0) { 341 error_setg_errno(errp, -ret, "bitmaps_ext: " 342 "Could not read ext header"); 343 return ret; 344 } 345 346 if (bitmaps_ext.reserved32 != 0) { 347 error_setg_errno(errp, -ret, "bitmaps_ext: " 348 "Reserved field is not zero"); 349 return -EINVAL; 350 } 351 352 bitmaps_ext.nb_bitmaps = be32_to_cpu(bitmaps_ext.nb_bitmaps); 353 bitmaps_ext.bitmap_directory_size = 354 be64_to_cpu(bitmaps_ext.bitmap_directory_size); 355 bitmaps_ext.bitmap_directory_offset = 356 be64_to_cpu(bitmaps_ext.bitmap_directory_offset); 357 358 if (bitmaps_ext.nb_bitmaps > QCOW2_MAX_BITMAPS) { 359 error_setg(errp, 360 "bitmaps_ext: Image has %" PRIu32 " bitmaps, " 361 "exceeding the QEMU supported maximum of %d", 362 bitmaps_ext.nb_bitmaps, QCOW2_MAX_BITMAPS); 363 return -EINVAL; 364 } 365 366 if (bitmaps_ext.nb_bitmaps == 0) { 367 error_setg(errp, "found bitmaps extension with zero bitmaps"); 368 return -EINVAL; 369 } 370 371 if (bitmaps_ext.bitmap_directory_offset & (s->cluster_size - 1)) { 372 error_setg(errp, "bitmaps_ext: " 373 "invalid bitmap directory offset"); 374 return -EINVAL; 375 } 376 377 if (bitmaps_ext.bitmap_directory_size > 378 QCOW2_MAX_BITMAP_DIRECTORY_SIZE) { 379 error_setg(errp, "bitmaps_ext: " 380 "bitmap directory size (%" PRIu64 ") exceeds " 381 "the maximum supported size (%d)", 382 bitmaps_ext.bitmap_directory_size, 383 QCOW2_MAX_BITMAP_DIRECTORY_SIZE); 384 return -EINVAL; 385 } 386 387 s->nb_bitmaps = bitmaps_ext.nb_bitmaps; 388 s->bitmap_directory_offset = 389 bitmaps_ext.bitmap_directory_offset; 390 s->bitmap_directory_size = 391 bitmaps_ext.bitmap_directory_size; 392 393 #ifdef DEBUG_EXT 394 printf("Qcow2: Got bitmaps extension: " 395 "offset=%" PRIu64 " nb_bitmaps=%" PRIu32 "\n", 396 s->bitmap_directory_offset, s->nb_bitmaps); 397 #endif 398 break; 399 400 default: 401 /* unknown magic - save it in case we need to rewrite the header */ 402 /* If you add a new feature, make sure to also update the fast 403 * path of qcow2_make_empty() to deal with it. */ 404 { 405 Qcow2UnknownHeaderExtension *uext; 406 407 uext = g_malloc0(sizeof(*uext) + ext.len); 408 uext->magic = ext.magic; 409 uext->len = ext.len; 410 QLIST_INSERT_HEAD(&s->unknown_header_ext, uext, next); 411 412 ret = bdrv_pread(bs->file, offset , uext->data, uext->len); 413 if (ret < 0) { 414 error_setg_errno(errp, -ret, "ERROR: unknown extension: " 415 "Could not read data"); 416 return ret; 417 } 418 } 419 break; 420 } 421 422 offset += ((ext.len + 7) & ~7); 423 } 424 425 return 0; 426 } 427 428 static void cleanup_unknown_header_ext(BlockDriverState *bs) 429 { 430 BDRVQcow2State *s = bs->opaque; 431 Qcow2UnknownHeaderExtension *uext, *next; 432 433 QLIST_FOREACH_SAFE(uext, &s->unknown_header_ext, next, next) { 434 QLIST_REMOVE(uext, next); 435 g_free(uext); 436 } 437 } 438 439 static void report_unsupported_feature(Error **errp, Qcow2Feature *table, 440 uint64_t mask) 441 { 442 char *features = g_strdup(""); 443 char *old; 444 445 while (table && table->name[0] != '\0') { 446 if (table->type == QCOW2_FEAT_TYPE_INCOMPATIBLE) { 447 if (mask & (1ULL << table->bit)) { 448 old = features; 449 features = g_strdup_printf("%s%s%.46s", old, *old ? ", " : "", 450 table->name); 451 g_free(old); 452 mask &= ~(1ULL << table->bit); 453 } 454 } 455 table++; 456 } 457 458 if (mask) { 459 old = features; 460 features = g_strdup_printf("%s%sUnknown incompatible feature: %" PRIx64, 461 old, *old ? ", " : "", mask); 462 g_free(old); 463 } 464 465 error_setg(errp, "Unsupported qcow2 feature(s): %s", features); 466 g_free(features); 467 } 468 469 /* 470 * Sets the dirty bit and flushes afterwards if necessary. 471 * 472 * The incompatible_features bit is only set if the image file header was 473 * updated successfully. Therefore it is not required to check the return 474 * value of this function. 475 */ 476 int qcow2_mark_dirty(BlockDriverState *bs) 477 { 478 BDRVQcow2State *s = bs->opaque; 479 uint64_t val; 480 int ret; 481 482 assert(s->qcow_version >= 3); 483 484 if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) { 485 return 0; /* already dirty */ 486 } 487 488 val = cpu_to_be64(s->incompatible_features | QCOW2_INCOMPAT_DIRTY); 489 ret = bdrv_pwrite(bs->file, offsetof(QCowHeader, incompatible_features), 490 &val, sizeof(val)); 491 if (ret < 0) { 492 return ret; 493 } 494 ret = bdrv_flush(bs->file->bs); 495 if (ret < 0) { 496 return ret; 497 } 498 499 /* Only treat image as dirty if the header was updated successfully */ 500 s->incompatible_features |= QCOW2_INCOMPAT_DIRTY; 501 return 0; 502 } 503 504 /* 505 * Clears the dirty bit and flushes before if necessary. Only call this 506 * function when there are no pending requests, it does not guard against 507 * concurrent requests dirtying the image. 508 */ 509 static int qcow2_mark_clean(BlockDriverState *bs) 510 { 511 BDRVQcow2State *s = bs->opaque; 512 513 if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) { 514 int ret; 515 516 s->incompatible_features &= ~QCOW2_INCOMPAT_DIRTY; 517 518 ret = qcow2_flush_caches(bs); 519 if (ret < 0) { 520 return ret; 521 } 522 523 return qcow2_update_header(bs); 524 } 525 return 0; 526 } 527 528 /* 529 * Marks the image as corrupt. 530 */ 531 int qcow2_mark_corrupt(BlockDriverState *bs) 532 { 533 BDRVQcow2State *s = bs->opaque; 534 535 s->incompatible_features |= QCOW2_INCOMPAT_CORRUPT; 536 return qcow2_update_header(bs); 537 } 538 539 /* 540 * Marks the image as consistent, i.e., unsets the corrupt bit, and flushes 541 * before if necessary. 542 */ 543 int qcow2_mark_consistent(BlockDriverState *bs) 544 { 545 BDRVQcow2State *s = bs->opaque; 546 547 if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) { 548 int ret = qcow2_flush_caches(bs); 549 if (ret < 0) { 550 return ret; 551 } 552 553 s->incompatible_features &= ~QCOW2_INCOMPAT_CORRUPT; 554 return qcow2_update_header(bs); 555 } 556 return 0; 557 } 558 559 static int coroutine_fn qcow2_co_check_locked(BlockDriverState *bs, 560 BdrvCheckResult *result, 561 BdrvCheckMode fix) 562 { 563 int ret = qcow2_check_refcounts(bs, result, fix); 564 if (ret < 0) { 565 return ret; 566 } 567 568 if (fix && result->check_errors == 0 && result->corruptions == 0) { 569 ret = qcow2_mark_clean(bs); 570 if (ret < 0) { 571 return ret; 572 } 573 return qcow2_mark_consistent(bs); 574 } 575 return ret; 576 } 577 578 static int coroutine_fn qcow2_co_check(BlockDriverState *bs, 579 BdrvCheckResult *result, 580 BdrvCheckMode fix) 581 { 582 BDRVQcow2State *s = bs->opaque; 583 int ret; 584 585 qemu_co_mutex_lock(&s->lock); 586 ret = qcow2_co_check_locked(bs, result, fix); 587 qemu_co_mutex_unlock(&s->lock); 588 return ret; 589 } 590 591 int qcow2_validate_table(BlockDriverState *bs, uint64_t offset, 592 uint64_t entries, size_t entry_len, 593 int64_t max_size_bytes, const char *table_name, 594 Error **errp) 595 { 596 BDRVQcow2State *s = bs->opaque; 597 598 if (entries > max_size_bytes / entry_len) { 599 error_setg(errp, "%s too large", table_name); 600 return -EFBIG; 601 } 602 603 /* Use signed INT64_MAX as the maximum even for uint64_t header fields, 604 * because values will be passed to qemu functions taking int64_t. */ 605 if ((INT64_MAX - entries * entry_len < offset) || 606 (offset_into_cluster(s, offset) != 0)) { 607 error_setg(errp, "%s offset invalid", table_name); 608 return -EINVAL; 609 } 610 611 return 0; 612 } 613 614 static QemuOptsList qcow2_runtime_opts = { 615 .name = "qcow2", 616 .head = QTAILQ_HEAD_INITIALIZER(qcow2_runtime_opts.head), 617 .desc = { 618 { 619 .name = QCOW2_OPT_LAZY_REFCOUNTS, 620 .type = QEMU_OPT_BOOL, 621 .help = "Postpone refcount updates", 622 }, 623 { 624 .name = QCOW2_OPT_DISCARD_REQUEST, 625 .type = QEMU_OPT_BOOL, 626 .help = "Pass guest discard requests to the layer below", 627 }, 628 { 629 .name = QCOW2_OPT_DISCARD_SNAPSHOT, 630 .type = QEMU_OPT_BOOL, 631 .help = "Generate discard requests when snapshot related space " 632 "is freed", 633 }, 634 { 635 .name = QCOW2_OPT_DISCARD_OTHER, 636 .type = QEMU_OPT_BOOL, 637 .help = "Generate discard requests when other clusters are freed", 638 }, 639 { 640 .name = QCOW2_OPT_OVERLAP, 641 .type = QEMU_OPT_STRING, 642 .help = "Selects which overlap checks to perform from a range of " 643 "templates (none, constant, cached, all)", 644 }, 645 { 646 .name = QCOW2_OPT_OVERLAP_TEMPLATE, 647 .type = QEMU_OPT_STRING, 648 .help = "Selects which overlap checks to perform from a range of " 649 "templates (none, constant, cached, all)", 650 }, 651 { 652 .name = QCOW2_OPT_OVERLAP_MAIN_HEADER, 653 .type = QEMU_OPT_BOOL, 654 .help = "Check for unintended writes into the main qcow2 header", 655 }, 656 { 657 .name = QCOW2_OPT_OVERLAP_ACTIVE_L1, 658 .type = QEMU_OPT_BOOL, 659 .help = "Check for unintended writes into the active L1 table", 660 }, 661 { 662 .name = QCOW2_OPT_OVERLAP_ACTIVE_L2, 663 .type = QEMU_OPT_BOOL, 664 .help = "Check for unintended writes into an active L2 table", 665 }, 666 { 667 .name = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE, 668 .type = QEMU_OPT_BOOL, 669 .help = "Check for unintended writes into the refcount table", 670 }, 671 { 672 .name = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK, 673 .type = QEMU_OPT_BOOL, 674 .help = "Check for unintended writes into a refcount block", 675 }, 676 { 677 .name = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE, 678 .type = QEMU_OPT_BOOL, 679 .help = "Check for unintended writes into the snapshot table", 680 }, 681 { 682 .name = QCOW2_OPT_OVERLAP_INACTIVE_L1, 683 .type = QEMU_OPT_BOOL, 684 .help = "Check for unintended writes into an inactive L1 table", 685 }, 686 { 687 .name = QCOW2_OPT_OVERLAP_INACTIVE_L2, 688 .type = QEMU_OPT_BOOL, 689 .help = "Check for unintended writes into an inactive L2 table", 690 }, 691 { 692 .name = QCOW2_OPT_OVERLAP_BITMAP_DIRECTORY, 693 .type = QEMU_OPT_BOOL, 694 .help = "Check for unintended writes into the bitmap directory", 695 }, 696 { 697 .name = QCOW2_OPT_CACHE_SIZE, 698 .type = QEMU_OPT_SIZE, 699 .help = "Maximum combined metadata (L2 tables and refcount blocks) " 700 "cache size", 701 }, 702 { 703 .name = QCOW2_OPT_L2_CACHE_SIZE, 704 .type = QEMU_OPT_SIZE, 705 .help = "Maximum L2 table cache size", 706 }, 707 { 708 .name = QCOW2_OPT_L2_CACHE_ENTRY_SIZE, 709 .type = QEMU_OPT_SIZE, 710 .help = "Size of each entry in the L2 cache", 711 }, 712 { 713 .name = QCOW2_OPT_REFCOUNT_CACHE_SIZE, 714 .type = QEMU_OPT_SIZE, 715 .help = "Maximum refcount block cache size", 716 }, 717 { 718 .name = QCOW2_OPT_CACHE_CLEAN_INTERVAL, 719 .type = QEMU_OPT_NUMBER, 720 .help = "Clean unused cache entries after this time (in seconds)", 721 }, 722 BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.", 723 "ID of secret providing qcow2 AES key or LUKS passphrase"), 724 { /* end of list */ } 725 }, 726 }; 727 728 static const char *overlap_bool_option_names[QCOW2_OL_MAX_BITNR] = { 729 [QCOW2_OL_MAIN_HEADER_BITNR] = QCOW2_OPT_OVERLAP_MAIN_HEADER, 730 [QCOW2_OL_ACTIVE_L1_BITNR] = QCOW2_OPT_OVERLAP_ACTIVE_L1, 731 [QCOW2_OL_ACTIVE_L2_BITNR] = QCOW2_OPT_OVERLAP_ACTIVE_L2, 732 [QCOW2_OL_REFCOUNT_TABLE_BITNR] = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE, 733 [QCOW2_OL_REFCOUNT_BLOCK_BITNR] = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK, 734 [QCOW2_OL_SNAPSHOT_TABLE_BITNR] = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE, 735 [QCOW2_OL_INACTIVE_L1_BITNR] = QCOW2_OPT_OVERLAP_INACTIVE_L1, 736 [QCOW2_OL_INACTIVE_L2_BITNR] = QCOW2_OPT_OVERLAP_INACTIVE_L2, 737 [QCOW2_OL_BITMAP_DIRECTORY_BITNR] = QCOW2_OPT_OVERLAP_BITMAP_DIRECTORY, 738 }; 739 740 static void cache_clean_timer_cb(void *opaque) 741 { 742 BlockDriverState *bs = opaque; 743 BDRVQcow2State *s = bs->opaque; 744 qcow2_cache_clean_unused(s->l2_table_cache); 745 qcow2_cache_clean_unused(s->refcount_block_cache); 746 timer_mod(s->cache_clean_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) + 747 (int64_t) s->cache_clean_interval * 1000); 748 } 749 750 static void cache_clean_timer_init(BlockDriverState *bs, AioContext *context) 751 { 752 BDRVQcow2State *s = bs->opaque; 753 if (s->cache_clean_interval > 0) { 754 s->cache_clean_timer = aio_timer_new(context, QEMU_CLOCK_VIRTUAL, 755 SCALE_MS, cache_clean_timer_cb, 756 bs); 757 timer_mod(s->cache_clean_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) + 758 (int64_t) s->cache_clean_interval * 1000); 759 } 760 } 761 762 static void cache_clean_timer_del(BlockDriverState *bs) 763 { 764 BDRVQcow2State *s = bs->opaque; 765 if (s->cache_clean_timer) { 766 timer_del(s->cache_clean_timer); 767 timer_free(s->cache_clean_timer); 768 s->cache_clean_timer = NULL; 769 } 770 } 771 772 static void qcow2_detach_aio_context(BlockDriverState *bs) 773 { 774 cache_clean_timer_del(bs); 775 } 776 777 static void qcow2_attach_aio_context(BlockDriverState *bs, 778 AioContext *new_context) 779 { 780 cache_clean_timer_init(bs, new_context); 781 } 782 783 static void read_cache_sizes(BlockDriverState *bs, QemuOpts *opts, 784 uint64_t *l2_cache_size, 785 uint64_t *l2_cache_entry_size, 786 uint64_t *refcount_cache_size, Error **errp) 787 { 788 BDRVQcow2State *s = bs->opaque; 789 uint64_t combined_cache_size, l2_cache_max_setting; 790 bool l2_cache_size_set, refcount_cache_size_set, combined_cache_size_set; 791 int min_refcount_cache = MIN_REFCOUNT_CACHE_SIZE * s->cluster_size; 792 uint64_t virtual_disk_size = bs->total_sectors * BDRV_SECTOR_SIZE; 793 uint64_t max_l2_cache = virtual_disk_size / (s->cluster_size / 8); 794 795 combined_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_CACHE_SIZE); 796 l2_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_L2_CACHE_SIZE); 797 refcount_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_REFCOUNT_CACHE_SIZE); 798 799 combined_cache_size = qemu_opt_get_size(opts, QCOW2_OPT_CACHE_SIZE, 0); 800 l2_cache_max_setting = qemu_opt_get_size(opts, QCOW2_OPT_L2_CACHE_SIZE, 801 DEFAULT_L2_CACHE_MAX_SIZE); 802 *refcount_cache_size = qemu_opt_get_size(opts, 803 QCOW2_OPT_REFCOUNT_CACHE_SIZE, 0); 804 805 *l2_cache_entry_size = qemu_opt_get_size( 806 opts, QCOW2_OPT_L2_CACHE_ENTRY_SIZE, s->cluster_size); 807 808 *l2_cache_size = MIN(max_l2_cache, l2_cache_max_setting); 809 810 if (combined_cache_size_set) { 811 if (l2_cache_size_set && refcount_cache_size_set) { 812 error_setg(errp, QCOW2_OPT_CACHE_SIZE ", " QCOW2_OPT_L2_CACHE_SIZE 813 " and " QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not be set " 814 "at the same time"); 815 return; 816 } else if (l2_cache_size_set && 817 (l2_cache_max_setting > combined_cache_size)) { 818 error_setg(errp, QCOW2_OPT_L2_CACHE_SIZE " may not exceed " 819 QCOW2_OPT_CACHE_SIZE); 820 return; 821 } else if (*refcount_cache_size > combined_cache_size) { 822 error_setg(errp, QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not exceed " 823 QCOW2_OPT_CACHE_SIZE); 824 return; 825 } 826 827 if (l2_cache_size_set) { 828 *refcount_cache_size = combined_cache_size - *l2_cache_size; 829 } else if (refcount_cache_size_set) { 830 *l2_cache_size = combined_cache_size - *refcount_cache_size; 831 } else { 832 /* Assign as much memory as possible to the L2 cache, and 833 * use the remainder for the refcount cache */ 834 if (combined_cache_size >= max_l2_cache + min_refcount_cache) { 835 *l2_cache_size = max_l2_cache; 836 *refcount_cache_size = combined_cache_size - *l2_cache_size; 837 } else { 838 *refcount_cache_size = 839 MIN(combined_cache_size, min_refcount_cache); 840 *l2_cache_size = combined_cache_size - *refcount_cache_size; 841 } 842 } 843 } 844 /* l2_cache_size and refcount_cache_size are ensured to have at least 845 * their minimum values in qcow2_update_options_prepare() */ 846 847 if (*l2_cache_entry_size < (1 << MIN_CLUSTER_BITS) || 848 *l2_cache_entry_size > s->cluster_size || 849 !is_power_of_2(*l2_cache_entry_size)) { 850 error_setg(errp, "L2 cache entry size must be a power of two " 851 "between %d and the cluster size (%d)", 852 1 << MIN_CLUSTER_BITS, s->cluster_size); 853 return; 854 } 855 } 856 857 typedef struct Qcow2ReopenState { 858 Qcow2Cache *l2_table_cache; 859 Qcow2Cache *refcount_block_cache; 860 int l2_slice_size; /* Number of entries in a slice of the L2 table */ 861 bool use_lazy_refcounts; 862 int overlap_check; 863 bool discard_passthrough[QCOW2_DISCARD_MAX]; 864 uint64_t cache_clean_interval; 865 QCryptoBlockOpenOptions *crypto_opts; /* Disk encryption runtime options */ 866 } Qcow2ReopenState; 867 868 static int qcow2_update_options_prepare(BlockDriverState *bs, 869 Qcow2ReopenState *r, 870 QDict *options, int flags, 871 Error **errp) 872 { 873 BDRVQcow2State *s = bs->opaque; 874 QemuOpts *opts = NULL; 875 const char *opt_overlap_check, *opt_overlap_check_template; 876 int overlap_check_template = 0; 877 uint64_t l2_cache_size, l2_cache_entry_size, refcount_cache_size; 878 int i; 879 const char *encryptfmt; 880 QDict *encryptopts = NULL; 881 Error *local_err = NULL; 882 int ret; 883 884 qdict_extract_subqdict(options, &encryptopts, "encrypt."); 885 encryptfmt = qdict_get_try_str(encryptopts, "format"); 886 887 opts = qemu_opts_create(&qcow2_runtime_opts, NULL, 0, &error_abort); 888 qemu_opts_absorb_qdict(opts, options, &local_err); 889 if (local_err) { 890 error_propagate(errp, local_err); 891 ret = -EINVAL; 892 goto fail; 893 } 894 895 /* get L2 table/refcount block cache size from command line options */ 896 read_cache_sizes(bs, opts, &l2_cache_size, &l2_cache_entry_size, 897 &refcount_cache_size, &local_err); 898 if (local_err) { 899 error_propagate(errp, local_err); 900 ret = -EINVAL; 901 goto fail; 902 } 903 904 l2_cache_size /= l2_cache_entry_size; 905 if (l2_cache_size < MIN_L2_CACHE_SIZE) { 906 l2_cache_size = MIN_L2_CACHE_SIZE; 907 } 908 if (l2_cache_size > INT_MAX) { 909 error_setg(errp, "L2 cache size too big"); 910 ret = -EINVAL; 911 goto fail; 912 } 913 914 refcount_cache_size /= s->cluster_size; 915 if (refcount_cache_size < MIN_REFCOUNT_CACHE_SIZE) { 916 refcount_cache_size = MIN_REFCOUNT_CACHE_SIZE; 917 } 918 if (refcount_cache_size > INT_MAX) { 919 error_setg(errp, "Refcount cache size too big"); 920 ret = -EINVAL; 921 goto fail; 922 } 923 924 /* alloc new L2 table/refcount block cache, flush old one */ 925 if (s->l2_table_cache) { 926 ret = qcow2_cache_flush(bs, s->l2_table_cache); 927 if (ret) { 928 error_setg_errno(errp, -ret, "Failed to flush the L2 table cache"); 929 goto fail; 930 } 931 } 932 933 if (s->refcount_block_cache) { 934 ret = qcow2_cache_flush(bs, s->refcount_block_cache); 935 if (ret) { 936 error_setg_errno(errp, -ret, 937 "Failed to flush the refcount block cache"); 938 goto fail; 939 } 940 } 941 942 r->l2_slice_size = l2_cache_entry_size / sizeof(uint64_t); 943 r->l2_table_cache = qcow2_cache_create(bs, l2_cache_size, 944 l2_cache_entry_size); 945 r->refcount_block_cache = qcow2_cache_create(bs, refcount_cache_size, 946 s->cluster_size); 947 if (r->l2_table_cache == NULL || r->refcount_block_cache == NULL) { 948 error_setg(errp, "Could not allocate metadata caches"); 949 ret = -ENOMEM; 950 goto fail; 951 } 952 953 /* New interval for cache cleanup timer */ 954 r->cache_clean_interval = 955 qemu_opt_get_number(opts, QCOW2_OPT_CACHE_CLEAN_INTERVAL, 956 DEFAULT_CACHE_CLEAN_INTERVAL); 957 #ifndef CONFIG_LINUX 958 if (r->cache_clean_interval != 0) { 959 error_setg(errp, QCOW2_OPT_CACHE_CLEAN_INTERVAL 960 " not supported on this host"); 961 ret = -EINVAL; 962 goto fail; 963 } 964 #endif 965 if (r->cache_clean_interval > UINT_MAX) { 966 error_setg(errp, "Cache clean interval too big"); 967 ret = -EINVAL; 968 goto fail; 969 } 970 971 /* lazy-refcounts; flush if going from enabled to disabled */ 972 r->use_lazy_refcounts = qemu_opt_get_bool(opts, QCOW2_OPT_LAZY_REFCOUNTS, 973 (s->compatible_features & QCOW2_COMPAT_LAZY_REFCOUNTS)); 974 if (r->use_lazy_refcounts && s->qcow_version < 3) { 975 error_setg(errp, "Lazy refcounts require a qcow2 image with at least " 976 "qemu 1.1 compatibility level"); 977 ret = -EINVAL; 978 goto fail; 979 } 980 981 if (s->use_lazy_refcounts && !r->use_lazy_refcounts) { 982 ret = qcow2_mark_clean(bs); 983 if (ret < 0) { 984 error_setg_errno(errp, -ret, "Failed to disable lazy refcounts"); 985 goto fail; 986 } 987 } 988 989 /* Overlap check options */ 990 opt_overlap_check = qemu_opt_get(opts, QCOW2_OPT_OVERLAP); 991 opt_overlap_check_template = qemu_opt_get(opts, QCOW2_OPT_OVERLAP_TEMPLATE); 992 if (opt_overlap_check_template && opt_overlap_check && 993 strcmp(opt_overlap_check_template, opt_overlap_check)) 994 { 995 error_setg(errp, "Conflicting values for qcow2 options '" 996 QCOW2_OPT_OVERLAP "' ('%s') and '" QCOW2_OPT_OVERLAP_TEMPLATE 997 "' ('%s')", opt_overlap_check, opt_overlap_check_template); 998 ret = -EINVAL; 999 goto fail; 1000 } 1001 if (!opt_overlap_check) { 1002 opt_overlap_check = opt_overlap_check_template ?: "cached"; 1003 } 1004 1005 if (!strcmp(opt_overlap_check, "none")) { 1006 overlap_check_template = 0; 1007 } else if (!strcmp(opt_overlap_check, "constant")) { 1008 overlap_check_template = QCOW2_OL_CONSTANT; 1009 } else if (!strcmp(opt_overlap_check, "cached")) { 1010 overlap_check_template = QCOW2_OL_CACHED; 1011 } else if (!strcmp(opt_overlap_check, "all")) { 1012 overlap_check_template = QCOW2_OL_ALL; 1013 } else { 1014 error_setg(errp, "Unsupported value '%s' for qcow2 option " 1015 "'overlap-check'. Allowed are any of the following: " 1016 "none, constant, cached, all", opt_overlap_check); 1017 ret = -EINVAL; 1018 goto fail; 1019 } 1020 1021 r->overlap_check = 0; 1022 for (i = 0; i < QCOW2_OL_MAX_BITNR; i++) { 1023 /* overlap-check defines a template bitmask, but every flag may be 1024 * overwritten through the associated boolean option */ 1025 r->overlap_check |= 1026 qemu_opt_get_bool(opts, overlap_bool_option_names[i], 1027 overlap_check_template & (1 << i)) << i; 1028 } 1029 1030 r->discard_passthrough[QCOW2_DISCARD_NEVER] = false; 1031 r->discard_passthrough[QCOW2_DISCARD_ALWAYS] = true; 1032 r->discard_passthrough[QCOW2_DISCARD_REQUEST] = 1033 qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_REQUEST, 1034 flags & BDRV_O_UNMAP); 1035 r->discard_passthrough[QCOW2_DISCARD_SNAPSHOT] = 1036 qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_SNAPSHOT, true); 1037 r->discard_passthrough[QCOW2_DISCARD_OTHER] = 1038 qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_OTHER, false); 1039 1040 switch (s->crypt_method_header) { 1041 case QCOW_CRYPT_NONE: 1042 if (encryptfmt) { 1043 error_setg(errp, "No encryption in image header, but options " 1044 "specified format '%s'", encryptfmt); 1045 ret = -EINVAL; 1046 goto fail; 1047 } 1048 break; 1049 1050 case QCOW_CRYPT_AES: 1051 if (encryptfmt && !g_str_equal(encryptfmt, "aes")) { 1052 error_setg(errp, 1053 "Header reported 'aes' encryption format but " 1054 "options specify '%s'", encryptfmt); 1055 ret = -EINVAL; 1056 goto fail; 1057 } 1058 qdict_put_str(encryptopts, "format", "qcow"); 1059 r->crypto_opts = block_crypto_open_opts_init(encryptopts, errp); 1060 break; 1061 1062 case QCOW_CRYPT_LUKS: 1063 if (encryptfmt && !g_str_equal(encryptfmt, "luks")) { 1064 error_setg(errp, 1065 "Header reported 'luks' encryption format but " 1066 "options specify '%s'", encryptfmt); 1067 ret = -EINVAL; 1068 goto fail; 1069 } 1070 qdict_put_str(encryptopts, "format", "luks"); 1071 r->crypto_opts = block_crypto_open_opts_init(encryptopts, errp); 1072 break; 1073 1074 default: 1075 error_setg(errp, "Unsupported encryption method %d", 1076 s->crypt_method_header); 1077 break; 1078 } 1079 if (s->crypt_method_header != QCOW_CRYPT_NONE && !r->crypto_opts) { 1080 ret = -EINVAL; 1081 goto fail; 1082 } 1083 1084 ret = 0; 1085 fail: 1086 qobject_unref(encryptopts); 1087 qemu_opts_del(opts); 1088 opts = NULL; 1089 return ret; 1090 } 1091 1092 static void qcow2_update_options_commit(BlockDriverState *bs, 1093 Qcow2ReopenState *r) 1094 { 1095 BDRVQcow2State *s = bs->opaque; 1096 int i; 1097 1098 if (s->l2_table_cache) { 1099 qcow2_cache_destroy(s->l2_table_cache); 1100 } 1101 if (s->refcount_block_cache) { 1102 qcow2_cache_destroy(s->refcount_block_cache); 1103 } 1104 s->l2_table_cache = r->l2_table_cache; 1105 s->refcount_block_cache = r->refcount_block_cache; 1106 s->l2_slice_size = r->l2_slice_size; 1107 1108 s->overlap_check = r->overlap_check; 1109 s->use_lazy_refcounts = r->use_lazy_refcounts; 1110 1111 for (i = 0; i < QCOW2_DISCARD_MAX; i++) { 1112 s->discard_passthrough[i] = r->discard_passthrough[i]; 1113 } 1114 1115 if (s->cache_clean_interval != r->cache_clean_interval) { 1116 cache_clean_timer_del(bs); 1117 s->cache_clean_interval = r->cache_clean_interval; 1118 cache_clean_timer_init(bs, bdrv_get_aio_context(bs)); 1119 } 1120 1121 qapi_free_QCryptoBlockOpenOptions(s->crypto_opts); 1122 s->crypto_opts = r->crypto_opts; 1123 } 1124 1125 static void qcow2_update_options_abort(BlockDriverState *bs, 1126 Qcow2ReopenState *r) 1127 { 1128 if (r->l2_table_cache) { 1129 qcow2_cache_destroy(r->l2_table_cache); 1130 } 1131 if (r->refcount_block_cache) { 1132 qcow2_cache_destroy(r->refcount_block_cache); 1133 } 1134 qapi_free_QCryptoBlockOpenOptions(r->crypto_opts); 1135 } 1136 1137 static int qcow2_update_options(BlockDriverState *bs, QDict *options, 1138 int flags, Error **errp) 1139 { 1140 Qcow2ReopenState r = {}; 1141 int ret; 1142 1143 ret = qcow2_update_options_prepare(bs, &r, options, flags, errp); 1144 if (ret >= 0) { 1145 qcow2_update_options_commit(bs, &r); 1146 } else { 1147 qcow2_update_options_abort(bs, &r); 1148 } 1149 1150 return ret; 1151 } 1152 1153 /* Called with s->lock held. */ 1154 static int coroutine_fn qcow2_do_open(BlockDriverState *bs, QDict *options, 1155 int flags, Error **errp) 1156 { 1157 BDRVQcow2State *s = bs->opaque; 1158 unsigned int len, i; 1159 int ret = 0; 1160 QCowHeader header; 1161 Error *local_err = NULL; 1162 uint64_t ext_end; 1163 uint64_t l1_vm_state_index; 1164 bool update_header = false; 1165 1166 ret = bdrv_pread(bs->file, 0, &header, sizeof(header)); 1167 if (ret < 0) { 1168 error_setg_errno(errp, -ret, "Could not read qcow2 header"); 1169 goto fail; 1170 } 1171 header.magic = be32_to_cpu(header.magic); 1172 header.version = be32_to_cpu(header.version); 1173 header.backing_file_offset = be64_to_cpu(header.backing_file_offset); 1174 header.backing_file_size = be32_to_cpu(header.backing_file_size); 1175 header.size = be64_to_cpu(header.size); 1176 header.cluster_bits = be32_to_cpu(header.cluster_bits); 1177 header.crypt_method = be32_to_cpu(header.crypt_method); 1178 header.l1_table_offset = be64_to_cpu(header.l1_table_offset); 1179 header.l1_size = be32_to_cpu(header.l1_size); 1180 header.refcount_table_offset = be64_to_cpu(header.refcount_table_offset); 1181 header.refcount_table_clusters = 1182 be32_to_cpu(header.refcount_table_clusters); 1183 header.snapshots_offset = be64_to_cpu(header.snapshots_offset); 1184 header.nb_snapshots = be32_to_cpu(header.nb_snapshots); 1185 1186 if (header.magic != QCOW_MAGIC) { 1187 error_setg(errp, "Image is not in qcow2 format"); 1188 ret = -EINVAL; 1189 goto fail; 1190 } 1191 if (header.version < 2 || header.version > 3) { 1192 error_setg(errp, "Unsupported qcow2 version %" PRIu32, header.version); 1193 ret = -ENOTSUP; 1194 goto fail; 1195 } 1196 1197 s->qcow_version = header.version; 1198 1199 /* Initialise cluster size */ 1200 if (header.cluster_bits < MIN_CLUSTER_BITS || 1201 header.cluster_bits > MAX_CLUSTER_BITS) { 1202 error_setg(errp, "Unsupported cluster size: 2^%" PRIu32, 1203 header.cluster_bits); 1204 ret = -EINVAL; 1205 goto fail; 1206 } 1207 1208 s->cluster_bits = header.cluster_bits; 1209 s->cluster_size = 1 << s->cluster_bits; 1210 s->cluster_sectors = 1 << (s->cluster_bits - BDRV_SECTOR_BITS); 1211 1212 /* Initialise version 3 header fields */ 1213 if (header.version == 2) { 1214 header.incompatible_features = 0; 1215 header.compatible_features = 0; 1216 header.autoclear_features = 0; 1217 header.refcount_order = 4; 1218 header.header_length = 72; 1219 } else { 1220 header.incompatible_features = 1221 be64_to_cpu(header.incompatible_features); 1222 header.compatible_features = be64_to_cpu(header.compatible_features); 1223 header.autoclear_features = be64_to_cpu(header.autoclear_features); 1224 header.refcount_order = be32_to_cpu(header.refcount_order); 1225 header.header_length = be32_to_cpu(header.header_length); 1226 1227 if (header.header_length < 104) { 1228 error_setg(errp, "qcow2 header too short"); 1229 ret = -EINVAL; 1230 goto fail; 1231 } 1232 } 1233 1234 if (header.header_length > s->cluster_size) { 1235 error_setg(errp, "qcow2 header exceeds cluster size"); 1236 ret = -EINVAL; 1237 goto fail; 1238 } 1239 1240 if (header.header_length > sizeof(header)) { 1241 s->unknown_header_fields_size = header.header_length - sizeof(header); 1242 s->unknown_header_fields = g_malloc(s->unknown_header_fields_size); 1243 ret = bdrv_pread(bs->file, sizeof(header), s->unknown_header_fields, 1244 s->unknown_header_fields_size); 1245 if (ret < 0) { 1246 error_setg_errno(errp, -ret, "Could not read unknown qcow2 header " 1247 "fields"); 1248 goto fail; 1249 } 1250 } 1251 1252 if (header.backing_file_offset > s->cluster_size) { 1253 error_setg(errp, "Invalid backing file offset"); 1254 ret = -EINVAL; 1255 goto fail; 1256 } 1257 1258 if (header.backing_file_offset) { 1259 ext_end = header.backing_file_offset; 1260 } else { 1261 ext_end = 1 << header.cluster_bits; 1262 } 1263 1264 /* Handle feature bits */ 1265 s->incompatible_features = header.incompatible_features; 1266 s->compatible_features = header.compatible_features; 1267 s->autoclear_features = header.autoclear_features; 1268 1269 if (s->incompatible_features & ~QCOW2_INCOMPAT_MASK) { 1270 void *feature_table = NULL; 1271 qcow2_read_extensions(bs, header.header_length, ext_end, 1272 &feature_table, flags, NULL, NULL); 1273 report_unsupported_feature(errp, feature_table, 1274 s->incompatible_features & 1275 ~QCOW2_INCOMPAT_MASK); 1276 ret = -ENOTSUP; 1277 g_free(feature_table); 1278 goto fail; 1279 } 1280 1281 if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) { 1282 /* Corrupt images may not be written to unless they are being repaired 1283 */ 1284 if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_CHECK)) { 1285 error_setg(errp, "qcow2: Image is corrupt; cannot be opened " 1286 "read/write"); 1287 ret = -EACCES; 1288 goto fail; 1289 } 1290 } 1291 1292 /* Check support for various header values */ 1293 if (header.refcount_order > 6) { 1294 error_setg(errp, "Reference count entry width too large; may not " 1295 "exceed 64 bits"); 1296 ret = -EINVAL; 1297 goto fail; 1298 } 1299 s->refcount_order = header.refcount_order; 1300 s->refcount_bits = 1 << s->refcount_order; 1301 s->refcount_max = UINT64_C(1) << (s->refcount_bits - 1); 1302 s->refcount_max += s->refcount_max - 1; 1303 1304 s->crypt_method_header = header.crypt_method; 1305 if (s->crypt_method_header) { 1306 if (bdrv_uses_whitelist() && 1307 s->crypt_method_header == QCOW_CRYPT_AES) { 1308 error_setg(errp, 1309 "Use of AES-CBC encrypted qcow2 images is no longer " 1310 "supported in system emulators"); 1311 error_append_hint(errp, 1312 "You can use 'qemu-img convert' to convert your " 1313 "image to an alternative supported format, such " 1314 "as unencrypted qcow2, or raw with the LUKS " 1315 "format instead.\n"); 1316 ret = -ENOSYS; 1317 goto fail; 1318 } 1319 1320 if (s->crypt_method_header == QCOW_CRYPT_AES) { 1321 s->crypt_physical_offset = false; 1322 } else { 1323 /* Assuming LUKS and any future crypt methods we 1324 * add will all use physical offsets, due to the 1325 * fact that the alternative is insecure... */ 1326 s->crypt_physical_offset = true; 1327 } 1328 1329 bs->encrypted = true; 1330 } 1331 1332 s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */ 1333 s->l2_size = 1 << s->l2_bits; 1334 /* 2^(s->refcount_order - 3) is the refcount width in bytes */ 1335 s->refcount_block_bits = s->cluster_bits - (s->refcount_order - 3); 1336 s->refcount_block_size = 1 << s->refcount_block_bits; 1337 bs->total_sectors = header.size / BDRV_SECTOR_SIZE; 1338 s->csize_shift = (62 - (s->cluster_bits - 8)); 1339 s->csize_mask = (1 << (s->cluster_bits - 8)) - 1; 1340 s->cluster_offset_mask = (1LL << s->csize_shift) - 1; 1341 1342 s->refcount_table_offset = header.refcount_table_offset; 1343 s->refcount_table_size = 1344 header.refcount_table_clusters << (s->cluster_bits - 3); 1345 1346 if (header.refcount_table_clusters == 0 && !(flags & BDRV_O_CHECK)) { 1347 error_setg(errp, "Image does not contain a reference count table"); 1348 ret = -EINVAL; 1349 goto fail; 1350 } 1351 1352 ret = qcow2_validate_table(bs, s->refcount_table_offset, 1353 header.refcount_table_clusters, 1354 s->cluster_size, QCOW_MAX_REFTABLE_SIZE, 1355 "Reference count table", errp); 1356 if (ret < 0) { 1357 goto fail; 1358 } 1359 1360 /* The total size in bytes of the snapshot table is checked in 1361 * qcow2_read_snapshots() because the size of each snapshot is 1362 * variable and we don't know it yet. 1363 * Here we only check the offset and number of snapshots. */ 1364 ret = qcow2_validate_table(bs, header.snapshots_offset, 1365 header.nb_snapshots, 1366 sizeof(QCowSnapshotHeader), 1367 sizeof(QCowSnapshotHeader) * QCOW_MAX_SNAPSHOTS, 1368 "Snapshot table", errp); 1369 if (ret < 0) { 1370 goto fail; 1371 } 1372 1373 /* read the level 1 table */ 1374 ret = qcow2_validate_table(bs, header.l1_table_offset, 1375 header.l1_size, sizeof(uint64_t), 1376 QCOW_MAX_L1_SIZE, "Active L1 table", errp); 1377 if (ret < 0) { 1378 goto fail; 1379 } 1380 s->l1_size = header.l1_size; 1381 s->l1_table_offset = header.l1_table_offset; 1382 1383 l1_vm_state_index = size_to_l1(s, header.size); 1384 if (l1_vm_state_index > INT_MAX) { 1385 error_setg(errp, "Image is too big"); 1386 ret = -EFBIG; 1387 goto fail; 1388 } 1389 s->l1_vm_state_index = l1_vm_state_index; 1390 1391 /* the L1 table must contain at least enough entries to put 1392 header.size bytes */ 1393 if (s->l1_size < s->l1_vm_state_index) { 1394 error_setg(errp, "L1 table is too small"); 1395 ret = -EINVAL; 1396 goto fail; 1397 } 1398 1399 if (s->l1_size > 0) { 1400 s->l1_table = qemu_try_blockalign(bs->file->bs, 1401 ROUND_UP(s->l1_size * sizeof(uint64_t), 512)); 1402 if (s->l1_table == NULL) { 1403 error_setg(errp, "Could not allocate L1 table"); 1404 ret = -ENOMEM; 1405 goto fail; 1406 } 1407 ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table, 1408 s->l1_size * sizeof(uint64_t)); 1409 if (ret < 0) { 1410 error_setg_errno(errp, -ret, "Could not read L1 table"); 1411 goto fail; 1412 } 1413 for(i = 0;i < s->l1_size; i++) { 1414 s->l1_table[i] = be64_to_cpu(s->l1_table[i]); 1415 } 1416 } 1417 1418 /* Parse driver-specific options */ 1419 ret = qcow2_update_options(bs, options, flags, errp); 1420 if (ret < 0) { 1421 goto fail; 1422 } 1423 1424 s->flags = flags; 1425 1426 ret = qcow2_refcount_init(bs); 1427 if (ret != 0) { 1428 error_setg_errno(errp, -ret, "Could not initialize refcount handling"); 1429 goto fail; 1430 } 1431 1432 QLIST_INIT(&s->cluster_allocs); 1433 QTAILQ_INIT(&s->discards); 1434 1435 /* read qcow2 extensions */ 1436 if (qcow2_read_extensions(bs, header.header_length, ext_end, NULL, 1437 flags, &update_header, &local_err)) { 1438 error_propagate(errp, local_err); 1439 ret = -EINVAL; 1440 goto fail; 1441 } 1442 1443 /* qcow2_read_extension may have set up the crypto context 1444 * if the crypt method needs a header region, some methods 1445 * don't need header extensions, so must check here 1446 */ 1447 if (s->crypt_method_header && !s->crypto) { 1448 if (s->crypt_method_header == QCOW_CRYPT_AES) { 1449 unsigned int cflags = 0; 1450 if (flags & BDRV_O_NO_IO) { 1451 cflags |= QCRYPTO_BLOCK_OPEN_NO_IO; 1452 } 1453 s->crypto = qcrypto_block_open(s->crypto_opts, "encrypt.", 1454 NULL, NULL, cflags, 1, errp); 1455 if (!s->crypto) { 1456 ret = -EINVAL; 1457 goto fail; 1458 } 1459 } else if (!(flags & BDRV_O_NO_IO)) { 1460 error_setg(errp, "Missing CRYPTO header for crypt method %d", 1461 s->crypt_method_header); 1462 ret = -EINVAL; 1463 goto fail; 1464 } 1465 } 1466 1467 /* read the backing file name */ 1468 if (header.backing_file_offset != 0) { 1469 len = header.backing_file_size; 1470 if (len > MIN(1023, s->cluster_size - header.backing_file_offset) || 1471 len >= sizeof(bs->backing_file)) { 1472 error_setg(errp, "Backing file name too long"); 1473 ret = -EINVAL; 1474 goto fail; 1475 } 1476 ret = bdrv_pread(bs->file, header.backing_file_offset, 1477 bs->backing_file, len); 1478 if (ret < 0) { 1479 error_setg_errno(errp, -ret, "Could not read backing file name"); 1480 goto fail; 1481 } 1482 bs->backing_file[len] = '\0'; 1483 s->image_backing_file = g_strdup(bs->backing_file); 1484 } 1485 1486 /* Internal snapshots */ 1487 s->snapshots_offset = header.snapshots_offset; 1488 s->nb_snapshots = header.nb_snapshots; 1489 1490 ret = qcow2_read_snapshots(bs); 1491 if (ret < 0) { 1492 error_setg_errno(errp, -ret, "Could not read snapshots"); 1493 goto fail; 1494 } 1495 1496 /* Clear unknown autoclear feature bits */ 1497 update_header |= s->autoclear_features & ~QCOW2_AUTOCLEAR_MASK; 1498 update_header = 1499 update_header && !bs->read_only && !(flags & BDRV_O_INACTIVE); 1500 if (update_header) { 1501 s->autoclear_features &= QCOW2_AUTOCLEAR_MASK; 1502 } 1503 1504 /* == Handle persistent dirty bitmaps == 1505 * 1506 * We want load dirty bitmaps in three cases: 1507 * 1508 * 1. Normal open of the disk in active mode, not related to invalidation 1509 * after migration. 1510 * 1511 * 2. Invalidation of the target vm after pre-copy phase of migration, if 1512 * bitmaps are _not_ migrating through migration channel, i.e. 1513 * 'dirty-bitmaps' capability is disabled. 1514 * 1515 * 3. Invalidation of source vm after failed or canceled migration. 1516 * This is a very interesting case. There are two possible types of 1517 * bitmaps: 1518 * 1519 * A. Stored on inactivation and removed. They should be loaded from the 1520 * image. 1521 * 1522 * B. Not stored: not-persistent bitmaps and bitmaps, migrated through 1523 * the migration channel (with dirty-bitmaps capability). 1524 * 1525 * On the other hand, there are two possible sub-cases: 1526 * 1527 * 3.1 disk was changed by somebody else while were inactive. In this 1528 * case all in-RAM dirty bitmaps (both persistent and not) are 1529 * definitely invalid. And we don't have any method to determine 1530 * this. 1531 * 1532 * Simple and safe thing is to just drop all the bitmaps of type B on 1533 * inactivation. But in this case we lose bitmaps in valid 4.2 case. 1534 * 1535 * On the other hand, resuming source vm, if disk was already changed 1536 * is a bad thing anyway: not only bitmaps, the whole vm state is 1537 * out of sync with disk. 1538 * 1539 * This means, that user or management tool, who for some reason 1540 * decided to resume source vm, after disk was already changed by 1541 * target vm, should at least drop all dirty bitmaps by hand. 1542 * 1543 * So, we can ignore this case for now, but TODO: "generation" 1544 * extension for qcow2, to determine, that image was changed after 1545 * last inactivation. And if it is changed, we will drop (or at least 1546 * mark as 'invalid' all the bitmaps of type B, both persistent 1547 * and not). 1548 * 1549 * 3.2 disk was _not_ changed while were inactive. Bitmaps may be saved 1550 * to disk ('dirty-bitmaps' capability disabled), or not saved 1551 * ('dirty-bitmaps' capability enabled), but we don't need to care 1552 * of: let's load bitmaps as always: stored bitmaps will be loaded, 1553 * and not stored has flag IN_USE=1 in the image and will be skipped 1554 * on loading. 1555 * 1556 * One remaining possible case when we don't want load bitmaps: 1557 * 1558 * 4. Open disk in inactive mode in target vm (bitmaps are migrating or 1559 * will be loaded on invalidation, no needs try loading them before) 1560 */ 1561 1562 if (!(bdrv_get_flags(bs) & BDRV_O_INACTIVE)) { 1563 /* It's case 1, 2 or 3.2. Or 3.1 which is BUG in management layer. */ 1564 bool header_updated = qcow2_load_dirty_bitmaps(bs, &local_err); 1565 1566 update_header = update_header && !header_updated; 1567 } 1568 if (local_err != NULL) { 1569 error_propagate(errp, local_err); 1570 ret = -EINVAL; 1571 goto fail; 1572 } 1573 1574 if (update_header) { 1575 ret = qcow2_update_header(bs); 1576 if (ret < 0) { 1577 error_setg_errno(errp, -ret, "Could not update qcow2 header"); 1578 goto fail; 1579 } 1580 } 1581 1582 bs->supported_zero_flags = header.version >= 3 ? BDRV_REQ_MAY_UNMAP : 0; 1583 1584 /* Repair image if dirty */ 1585 if (!(flags & (BDRV_O_CHECK | BDRV_O_INACTIVE)) && !bs->read_only && 1586 (s->incompatible_features & QCOW2_INCOMPAT_DIRTY)) { 1587 BdrvCheckResult result = {0}; 1588 1589 ret = qcow2_co_check_locked(bs, &result, 1590 BDRV_FIX_ERRORS | BDRV_FIX_LEAKS); 1591 if (ret < 0 || result.check_errors) { 1592 if (ret >= 0) { 1593 ret = -EIO; 1594 } 1595 error_setg_errno(errp, -ret, "Could not repair dirty image"); 1596 goto fail; 1597 } 1598 } 1599 1600 #ifdef DEBUG_ALLOC 1601 { 1602 BdrvCheckResult result = {0}; 1603 qcow2_check_refcounts(bs, &result, 0); 1604 } 1605 #endif 1606 1607 qemu_co_queue_init(&s->compress_wait_queue); 1608 1609 return ret; 1610 1611 fail: 1612 g_free(s->unknown_header_fields); 1613 cleanup_unknown_header_ext(bs); 1614 qcow2_free_snapshots(bs); 1615 qcow2_refcount_close(bs); 1616 qemu_vfree(s->l1_table); 1617 /* else pre-write overlap checks in cache_destroy may crash */ 1618 s->l1_table = NULL; 1619 cache_clean_timer_del(bs); 1620 if (s->l2_table_cache) { 1621 qcow2_cache_destroy(s->l2_table_cache); 1622 } 1623 if (s->refcount_block_cache) { 1624 qcow2_cache_destroy(s->refcount_block_cache); 1625 } 1626 qcrypto_block_free(s->crypto); 1627 qapi_free_QCryptoBlockOpenOptions(s->crypto_opts); 1628 return ret; 1629 } 1630 1631 typedef struct QCow2OpenCo { 1632 BlockDriverState *bs; 1633 QDict *options; 1634 int flags; 1635 Error **errp; 1636 int ret; 1637 } QCow2OpenCo; 1638 1639 static void coroutine_fn qcow2_open_entry(void *opaque) 1640 { 1641 QCow2OpenCo *qoc = opaque; 1642 BDRVQcow2State *s = qoc->bs->opaque; 1643 1644 qemu_co_mutex_lock(&s->lock); 1645 qoc->ret = qcow2_do_open(qoc->bs, qoc->options, qoc->flags, qoc->errp); 1646 qemu_co_mutex_unlock(&s->lock); 1647 } 1648 1649 static int qcow2_open(BlockDriverState *bs, QDict *options, int flags, 1650 Error **errp) 1651 { 1652 BDRVQcow2State *s = bs->opaque; 1653 QCow2OpenCo qoc = { 1654 .bs = bs, 1655 .options = options, 1656 .flags = flags, 1657 .errp = errp, 1658 .ret = -EINPROGRESS 1659 }; 1660 1661 bs->file = bdrv_open_child(NULL, options, "file", bs, &child_file, 1662 false, errp); 1663 if (!bs->file) { 1664 return -EINVAL; 1665 } 1666 1667 /* Initialise locks */ 1668 qemu_co_mutex_init(&s->lock); 1669 1670 if (qemu_in_coroutine()) { 1671 /* From bdrv_co_create. */ 1672 qcow2_open_entry(&qoc); 1673 } else { 1674 assert(qemu_get_current_aio_context() == qemu_get_aio_context()); 1675 qemu_coroutine_enter(qemu_coroutine_create(qcow2_open_entry, &qoc)); 1676 BDRV_POLL_WHILE(bs, qoc.ret == -EINPROGRESS); 1677 } 1678 return qoc.ret; 1679 } 1680 1681 static void qcow2_refresh_limits(BlockDriverState *bs, Error **errp) 1682 { 1683 BDRVQcow2State *s = bs->opaque; 1684 1685 if (bs->encrypted) { 1686 /* Encryption works on a sector granularity */ 1687 bs->bl.request_alignment = qcrypto_block_get_sector_size(s->crypto); 1688 } 1689 bs->bl.pwrite_zeroes_alignment = s->cluster_size; 1690 bs->bl.pdiscard_alignment = s->cluster_size; 1691 } 1692 1693 static int qcow2_reopen_prepare(BDRVReopenState *state, 1694 BlockReopenQueue *queue, Error **errp) 1695 { 1696 Qcow2ReopenState *r; 1697 int ret; 1698 1699 r = g_new0(Qcow2ReopenState, 1); 1700 state->opaque = r; 1701 1702 ret = qcow2_update_options_prepare(state->bs, r, state->options, 1703 state->flags, errp); 1704 if (ret < 0) { 1705 goto fail; 1706 } 1707 1708 /* We need to write out any unwritten data if we reopen read-only. */ 1709 if ((state->flags & BDRV_O_RDWR) == 0) { 1710 ret = qcow2_reopen_bitmaps_ro(state->bs, errp); 1711 if (ret < 0) { 1712 goto fail; 1713 } 1714 1715 ret = bdrv_flush(state->bs); 1716 if (ret < 0) { 1717 goto fail; 1718 } 1719 1720 ret = qcow2_mark_clean(state->bs); 1721 if (ret < 0) { 1722 goto fail; 1723 } 1724 } 1725 1726 return 0; 1727 1728 fail: 1729 qcow2_update_options_abort(state->bs, r); 1730 g_free(r); 1731 return ret; 1732 } 1733 1734 static void qcow2_reopen_commit(BDRVReopenState *state) 1735 { 1736 qcow2_update_options_commit(state->bs, state->opaque); 1737 g_free(state->opaque); 1738 } 1739 1740 static void qcow2_reopen_abort(BDRVReopenState *state) 1741 { 1742 qcow2_update_options_abort(state->bs, state->opaque); 1743 g_free(state->opaque); 1744 } 1745 1746 static void qcow2_join_options(QDict *options, QDict *old_options) 1747 { 1748 bool has_new_overlap_template = 1749 qdict_haskey(options, QCOW2_OPT_OVERLAP) || 1750 qdict_haskey(options, QCOW2_OPT_OVERLAP_TEMPLATE); 1751 bool has_new_total_cache_size = 1752 qdict_haskey(options, QCOW2_OPT_CACHE_SIZE); 1753 bool has_all_cache_options; 1754 1755 /* New overlap template overrides all old overlap options */ 1756 if (has_new_overlap_template) { 1757 qdict_del(old_options, QCOW2_OPT_OVERLAP); 1758 qdict_del(old_options, QCOW2_OPT_OVERLAP_TEMPLATE); 1759 qdict_del(old_options, QCOW2_OPT_OVERLAP_MAIN_HEADER); 1760 qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L1); 1761 qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L2); 1762 qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_TABLE); 1763 qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK); 1764 qdict_del(old_options, QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE); 1765 qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L1); 1766 qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L2); 1767 } 1768 1769 /* New total cache size overrides all old options */ 1770 if (qdict_haskey(options, QCOW2_OPT_CACHE_SIZE)) { 1771 qdict_del(old_options, QCOW2_OPT_L2_CACHE_SIZE); 1772 qdict_del(old_options, QCOW2_OPT_REFCOUNT_CACHE_SIZE); 1773 } 1774 1775 qdict_join(options, old_options, false); 1776 1777 /* 1778 * If after merging all cache size options are set, an old total size is 1779 * overwritten. Do keep all options, however, if all three are new. The 1780 * resulting error message is what we want to happen. 1781 */ 1782 has_all_cache_options = 1783 qdict_haskey(options, QCOW2_OPT_CACHE_SIZE) || 1784 qdict_haskey(options, QCOW2_OPT_L2_CACHE_SIZE) || 1785 qdict_haskey(options, QCOW2_OPT_REFCOUNT_CACHE_SIZE); 1786 1787 if (has_all_cache_options && !has_new_total_cache_size) { 1788 qdict_del(options, QCOW2_OPT_CACHE_SIZE); 1789 } 1790 } 1791 1792 static int coroutine_fn qcow2_co_block_status(BlockDriverState *bs, 1793 bool want_zero, 1794 int64_t offset, int64_t count, 1795 int64_t *pnum, int64_t *map, 1796 BlockDriverState **file) 1797 { 1798 BDRVQcow2State *s = bs->opaque; 1799 uint64_t cluster_offset; 1800 int index_in_cluster, ret; 1801 unsigned int bytes; 1802 int status = 0; 1803 1804 bytes = MIN(INT_MAX, count); 1805 qemu_co_mutex_lock(&s->lock); 1806 ret = qcow2_get_cluster_offset(bs, offset, &bytes, &cluster_offset); 1807 qemu_co_mutex_unlock(&s->lock); 1808 if (ret < 0) { 1809 return ret; 1810 } 1811 1812 *pnum = bytes; 1813 1814 if (cluster_offset != 0 && ret != QCOW2_CLUSTER_COMPRESSED && 1815 !s->crypto) { 1816 index_in_cluster = offset & (s->cluster_size - 1); 1817 *map = cluster_offset | index_in_cluster; 1818 *file = bs->file->bs; 1819 status |= BDRV_BLOCK_OFFSET_VALID; 1820 } 1821 if (ret == QCOW2_CLUSTER_ZERO_PLAIN || ret == QCOW2_CLUSTER_ZERO_ALLOC) { 1822 status |= BDRV_BLOCK_ZERO; 1823 } else if (ret != QCOW2_CLUSTER_UNALLOCATED) { 1824 status |= BDRV_BLOCK_DATA; 1825 } 1826 return status; 1827 } 1828 1829 static coroutine_fn int qcow2_handle_l2meta(BlockDriverState *bs, 1830 QCowL2Meta **pl2meta, 1831 bool link_l2) 1832 { 1833 int ret = 0; 1834 QCowL2Meta *l2meta = *pl2meta; 1835 1836 while (l2meta != NULL) { 1837 QCowL2Meta *next; 1838 1839 if (link_l2) { 1840 ret = qcow2_alloc_cluster_link_l2(bs, l2meta); 1841 if (ret) { 1842 goto out; 1843 } 1844 } else { 1845 qcow2_alloc_cluster_abort(bs, l2meta); 1846 } 1847 1848 /* Take the request off the list of running requests */ 1849 if (l2meta->nb_clusters != 0) { 1850 QLIST_REMOVE(l2meta, next_in_flight); 1851 } 1852 1853 qemu_co_queue_restart_all(&l2meta->dependent_requests); 1854 1855 next = l2meta->next; 1856 g_free(l2meta); 1857 l2meta = next; 1858 } 1859 out: 1860 *pl2meta = l2meta; 1861 return ret; 1862 } 1863 1864 static coroutine_fn int qcow2_co_preadv(BlockDriverState *bs, uint64_t offset, 1865 uint64_t bytes, QEMUIOVector *qiov, 1866 int flags) 1867 { 1868 BDRVQcow2State *s = bs->opaque; 1869 int offset_in_cluster; 1870 int ret; 1871 unsigned int cur_bytes; /* number of bytes in current iteration */ 1872 uint64_t cluster_offset = 0; 1873 uint64_t bytes_done = 0; 1874 QEMUIOVector hd_qiov; 1875 uint8_t *cluster_data = NULL; 1876 1877 qemu_iovec_init(&hd_qiov, qiov->niov); 1878 1879 qemu_co_mutex_lock(&s->lock); 1880 1881 while (bytes != 0) { 1882 1883 /* prepare next request */ 1884 cur_bytes = MIN(bytes, INT_MAX); 1885 if (s->crypto) { 1886 cur_bytes = MIN(cur_bytes, 1887 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size); 1888 } 1889 1890 ret = qcow2_get_cluster_offset(bs, offset, &cur_bytes, &cluster_offset); 1891 if (ret < 0) { 1892 goto fail; 1893 } 1894 1895 offset_in_cluster = offset_into_cluster(s, offset); 1896 1897 qemu_iovec_reset(&hd_qiov); 1898 qemu_iovec_concat(&hd_qiov, qiov, bytes_done, cur_bytes); 1899 1900 switch (ret) { 1901 case QCOW2_CLUSTER_UNALLOCATED: 1902 1903 if (bs->backing) { 1904 BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO); 1905 qemu_co_mutex_unlock(&s->lock); 1906 ret = bdrv_co_preadv(bs->backing, offset, cur_bytes, 1907 &hd_qiov, 0); 1908 qemu_co_mutex_lock(&s->lock); 1909 if (ret < 0) { 1910 goto fail; 1911 } 1912 } else { 1913 /* Note: in this case, no need to wait */ 1914 qemu_iovec_memset(&hd_qiov, 0, 0, cur_bytes); 1915 } 1916 break; 1917 1918 case QCOW2_CLUSTER_ZERO_PLAIN: 1919 case QCOW2_CLUSTER_ZERO_ALLOC: 1920 qemu_iovec_memset(&hd_qiov, 0, 0, cur_bytes); 1921 break; 1922 1923 case QCOW2_CLUSTER_COMPRESSED: 1924 qemu_co_mutex_unlock(&s->lock); 1925 ret = qcow2_co_preadv_compressed(bs, cluster_offset, 1926 offset, cur_bytes, 1927 &hd_qiov); 1928 qemu_co_mutex_lock(&s->lock); 1929 if (ret < 0) { 1930 goto fail; 1931 } 1932 1933 break; 1934 1935 case QCOW2_CLUSTER_NORMAL: 1936 if ((cluster_offset & 511) != 0) { 1937 ret = -EIO; 1938 goto fail; 1939 } 1940 1941 if (bs->encrypted) { 1942 assert(s->crypto); 1943 1944 /* 1945 * For encrypted images, read everything into a temporary 1946 * contiguous buffer on which the AES functions can work. 1947 */ 1948 if (!cluster_data) { 1949 cluster_data = 1950 qemu_try_blockalign(bs->file->bs, 1951 QCOW_MAX_CRYPT_CLUSTERS 1952 * s->cluster_size); 1953 if (cluster_data == NULL) { 1954 ret = -ENOMEM; 1955 goto fail; 1956 } 1957 } 1958 1959 assert(cur_bytes <= QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size); 1960 qemu_iovec_reset(&hd_qiov); 1961 qemu_iovec_add(&hd_qiov, cluster_data, cur_bytes); 1962 } 1963 1964 BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO); 1965 qemu_co_mutex_unlock(&s->lock); 1966 ret = bdrv_co_preadv(bs->file, 1967 cluster_offset + offset_in_cluster, 1968 cur_bytes, &hd_qiov, 0); 1969 qemu_co_mutex_lock(&s->lock); 1970 if (ret < 0) { 1971 goto fail; 1972 } 1973 if (bs->encrypted) { 1974 assert(s->crypto); 1975 assert((offset & (BDRV_SECTOR_SIZE - 1)) == 0); 1976 assert((cur_bytes & (BDRV_SECTOR_SIZE - 1)) == 0); 1977 if (qcrypto_block_decrypt(s->crypto, 1978 (s->crypt_physical_offset ? 1979 cluster_offset + offset_in_cluster : 1980 offset), 1981 cluster_data, 1982 cur_bytes, 1983 NULL) < 0) { 1984 ret = -EIO; 1985 goto fail; 1986 } 1987 qemu_iovec_from_buf(qiov, bytes_done, cluster_data, cur_bytes); 1988 } 1989 break; 1990 1991 default: 1992 g_assert_not_reached(); 1993 ret = -EIO; 1994 goto fail; 1995 } 1996 1997 bytes -= cur_bytes; 1998 offset += cur_bytes; 1999 bytes_done += cur_bytes; 2000 } 2001 ret = 0; 2002 2003 fail: 2004 qemu_co_mutex_unlock(&s->lock); 2005 2006 qemu_iovec_destroy(&hd_qiov); 2007 qemu_vfree(cluster_data); 2008 2009 return ret; 2010 } 2011 2012 /* Check if it's possible to merge a write request with the writing of 2013 * the data from the COW regions */ 2014 static bool merge_cow(uint64_t offset, unsigned bytes, 2015 QEMUIOVector *hd_qiov, QCowL2Meta *l2meta) 2016 { 2017 QCowL2Meta *m; 2018 2019 for (m = l2meta; m != NULL; m = m->next) { 2020 /* If both COW regions are empty then there's nothing to merge */ 2021 if (m->cow_start.nb_bytes == 0 && m->cow_end.nb_bytes == 0) { 2022 continue; 2023 } 2024 2025 /* The data (middle) region must be immediately after the 2026 * start region */ 2027 if (l2meta_cow_start(m) + m->cow_start.nb_bytes != offset) { 2028 continue; 2029 } 2030 2031 /* The end region must be immediately after the data (middle) 2032 * region */ 2033 if (m->offset + m->cow_end.offset != offset + bytes) { 2034 continue; 2035 } 2036 2037 /* Make sure that adding both COW regions to the QEMUIOVector 2038 * does not exceed IOV_MAX */ 2039 if (hd_qiov->niov > IOV_MAX - 2) { 2040 continue; 2041 } 2042 2043 m->data_qiov = hd_qiov; 2044 return true; 2045 } 2046 2047 return false; 2048 } 2049 2050 static coroutine_fn int qcow2_co_pwritev(BlockDriverState *bs, uint64_t offset, 2051 uint64_t bytes, QEMUIOVector *qiov, 2052 int flags) 2053 { 2054 BDRVQcow2State *s = bs->opaque; 2055 int offset_in_cluster; 2056 int ret; 2057 unsigned int cur_bytes; /* number of sectors in current iteration */ 2058 uint64_t cluster_offset; 2059 QEMUIOVector hd_qiov; 2060 uint64_t bytes_done = 0; 2061 uint8_t *cluster_data = NULL; 2062 QCowL2Meta *l2meta = NULL; 2063 2064 trace_qcow2_writev_start_req(qemu_coroutine_self(), offset, bytes); 2065 2066 qemu_iovec_init(&hd_qiov, qiov->niov); 2067 2068 qemu_co_mutex_lock(&s->lock); 2069 2070 while (bytes != 0) { 2071 2072 l2meta = NULL; 2073 2074 trace_qcow2_writev_start_part(qemu_coroutine_self()); 2075 offset_in_cluster = offset_into_cluster(s, offset); 2076 cur_bytes = MIN(bytes, INT_MAX); 2077 if (bs->encrypted) { 2078 cur_bytes = MIN(cur_bytes, 2079 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size 2080 - offset_in_cluster); 2081 } 2082 2083 ret = qcow2_alloc_cluster_offset(bs, offset, &cur_bytes, 2084 &cluster_offset, &l2meta); 2085 if (ret < 0) { 2086 goto fail; 2087 } 2088 2089 assert((cluster_offset & 511) == 0); 2090 2091 qemu_iovec_reset(&hd_qiov); 2092 qemu_iovec_concat(&hd_qiov, qiov, bytes_done, cur_bytes); 2093 2094 if (bs->encrypted) { 2095 assert(s->crypto); 2096 if (!cluster_data) { 2097 cluster_data = qemu_try_blockalign(bs->file->bs, 2098 QCOW_MAX_CRYPT_CLUSTERS 2099 * s->cluster_size); 2100 if (cluster_data == NULL) { 2101 ret = -ENOMEM; 2102 goto fail; 2103 } 2104 } 2105 2106 assert(hd_qiov.size <= 2107 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size); 2108 qemu_iovec_to_buf(&hd_qiov, 0, cluster_data, hd_qiov.size); 2109 2110 if (qcrypto_block_encrypt(s->crypto, 2111 (s->crypt_physical_offset ? 2112 cluster_offset + offset_in_cluster : 2113 offset), 2114 cluster_data, 2115 cur_bytes, NULL) < 0) { 2116 ret = -EIO; 2117 goto fail; 2118 } 2119 2120 qemu_iovec_reset(&hd_qiov); 2121 qemu_iovec_add(&hd_qiov, cluster_data, cur_bytes); 2122 } 2123 2124 ret = qcow2_pre_write_overlap_check(bs, 0, 2125 cluster_offset + offset_in_cluster, cur_bytes); 2126 if (ret < 0) { 2127 goto fail; 2128 } 2129 2130 /* If we need to do COW, check if it's possible to merge the 2131 * writing of the guest data together with that of the COW regions. 2132 * If it's not possible (or not necessary) then write the 2133 * guest data now. */ 2134 if (!merge_cow(offset, cur_bytes, &hd_qiov, l2meta)) { 2135 qemu_co_mutex_unlock(&s->lock); 2136 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO); 2137 trace_qcow2_writev_data(qemu_coroutine_self(), 2138 cluster_offset + offset_in_cluster); 2139 ret = bdrv_co_pwritev(bs->file, 2140 cluster_offset + offset_in_cluster, 2141 cur_bytes, &hd_qiov, 0); 2142 qemu_co_mutex_lock(&s->lock); 2143 if (ret < 0) { 2144 goto fail; 2145 } 2146 } 2147 2148 ret = qcow2_handle_l2meta(bs, &l2meta, true); 2149 if (ret) { 2150 goto fail; 2151 } 2152 2153 bytes -= cur_bytes; 2154 offset += cur_bytes; 2155 bytes_done += cur_bytes; 2156 trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_bytes); 2157 } 2158 ret = 0; 2159 2160 fail: 2161 qcow2_handle_l2meta(bs, &l2meta, false); 2162 2163 qemu_co_mutex_unlock(&s->lock); 2164 2165 qemu_iovec_destroy(&hd_qiov); 2166 qemu_vfree(cluster_data); 2167 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret); 2168 2169 return ret; 2170 } 2171 2172 static int qcow2_inactivate(BlockDriverState *bs) 2173 { 2174 BDRVQcow2State *s = bs->opaque; 2175 int ret, result = 0; 2176 Error *local_err = NULL; 2177 2178 qcow2_store_persistent_dirty_bitmaps(bs, &local_err); 2179 if (local_err != NULL) { 2180 result = -EINVAL; 2181 error_reportf_err(local_err, "Lost persistent bitmaps during " 2182 "inactivation of node '%s': ", 2183 bdrv_get_device_or_node_name(bs)); 2184 } 2185 2186 ret = qcow2_cache_flush(bs, s->l2_table_cache); 2187 if (ret) { 2188 result = ret; 2189 error_report("Failed to flush the L2 table cache: %s", 2190 strerror(-ret)); 2191 } 2192 2193 ret = qcow2_cache_flush(bs, s->refcount_block_cache); 2194 if (ret) { 2195 result = ret; 2196 error_report("Failed to flush the refcount block cache: %s", 2197 strerror(-ret)); 2198 } 2199 2200 if (result == 0) { 2201 qcow2_mark_clean(bs); 2202 } 2203 2204 return result; 2205 } 2206 2207 static void qcow2_close(BlockDriverState *bs) 2208 { 2209 BDRVQcow2State *s = bs->opaque; 2210 qemu_vfree(s->l1_table); 2211 /* else pre-write overlap checks in cache_destroy may crash */ 2212 s->l1_table = NULL; 2213 2214 if (!(s->flags & BDRV_O_INACTIVE)) { 2215 qcow2_inactivate(bs); 2216 } 2217 2218 cache_clean_timer_del(bs); 2219 qcow2_cache_destroy(s->l2_table_cache); 2220 qcow2_cache_destroy(s->refcount_block_cache); 2221 2222 qcrypto_block_free(s->crypto); 2223 s->crypto = NULL; 2224 2225 g_free(s->unknown_header_fields); 2226 cleanup_unknown_header_ext(bs); 2227 2228 g_free(s->image_backing_file); 2229 g_free(s->image_backing_format); 2230 2231 qcow2_refcount_close(bs); 2232 qcow2_free_snapshots(bs); 2233 } 2234 2235 static void coroutine_fn qcow2_co_invalidate_cache(BlockDriverState *bs, 2236 Error **errp) 2237 { 2238 BDRVQcow2State *s = bs->opaque; 2239 int flags = s->flags; 2240 QCryptoBlock *crypto = NULL; 2241 QDict *options; 2242 Error *local_err = NULL; 2243 int ret; 2244 2245 /* 2246 * Backing files are read-only which makes all of their metadata immutable, 2247 * that means we don't have to worry about reopening them here. 2248 */ 2249 2250 crypto = s->crypto; 2251 s->crypto = NULL; 2252 2253 qcow2_close(bs); 2254 2255 memset(s, 0, sizeof(BDRVQcow2State)); 2256 options = qdict_clone_shallow(bs->options); 2257 2258 flags &= ~BDRV_O_INACTIVE; 2259 qemu_co_mutex_lock(&s->lock); 2260 ret = qcow2_do_open(bs, options, flags, &local_err); 2261 qemu_co_mutex_unlock(&s->lock); 2262 qobject_unref(options); 2263 if (local_err) { 2264 error_propagate_prepend(errp, local_err, 2265 "Could not reopen qcow2 layer: "); 2266 bs->drv = NULL; 2267 return; 2268 } else if (ret < 0) { 2269 error_setg_errno(errp, -ret, "Could not reopen qcow2 layer"); 2270 bs->drv = NULL; 2271 return; 2272 } 2273 2274 s->crypto = crypto; 2275 } 2276 2277 static size_t header_ext_add(char *buf, uint32_t magic, const void *s, 2278 size_t len, size_t buflen) 2279 { 2280 QCowExtension *ext_backing_fmt = (QCowExtension*) buf; 2281 size_t ext_len = sizeof(QCowExtension) + ((len + 7) & ~7); 2282 2283 if (buflen < ext_len) { 2284 return -ENOSPC; 2285 } 2286 2287 *ext_backing_fmt = (QCowExtension) { 2288 .magic = cpu_to_be32(magic), 2289 .len = cpu_to_be32(len), 2290 }; 2291 2292 if (len) { 2293 memcpy(buf + sizeof(QCowExtension), s, len); 2294 } 2295 2296 return ext_len; 2297 } 2298 2299 /* 2300 * Updates the qcow2 header, including the variable length parts of it, i.e. 2301 * the backing file name and all extensions. qcow2 was not designed to allow 2302 * such changes, so if we run out of space (we can only use the first cluster) 2303 * this function may fail. 2304 * 2305 * Returns 0 on success, -errno in error cases. 2306 */ 2307 int qcow2_update_header(BlockDriverState *bs) 2308 { 2309 BDRVQcow2State *s = bs->opaque; 2310 QCowHeader *header; 2311 char *buf; 2312 size_t buflen = s->cluster_size; 2313 int ret; 2314 uint64_t total_size; 2315 uint32_t refcount_table_clusters; 2316 size_t header_length; 2317 Qcow2UnknownHeaderExtension *uext; 2318 2319 buf = qemu_blockalign(bs, buflen); 2320 2321 /* Header structure */ 2322 header = (QCowHeader*) buf; 2323 2324 if (buflen < sizeof(*header)) { 2325 ret = -ENOSPC; 2326 goto fail; 2327 } 2328 2329 header_length = sizeof(*header) + s->unknown_header_fields_size; 2330 total_size = bs->total_sectors * BDRV_SECTOR_SIZE; 2331 refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3); 2332 2333 *header = (QCowHeader) { 2334 /* Version 2 fields */ 2335 .magic = cpu_to_be32(QCOW_MAGIC), 2336 .version = cpu_to_be32(s->qcow_version), 2337 .backing_file_offset = 0, 2338 .backing_file_size = 0, 2339 .cluster_bits = cpu_to_be32(s->cluster_bits), 2340 .size = cpu_to_be64(total_size), 2341 .crypt_method = cpu_to_be32(s->crypt_method_header), 2342 .l1_size = cpu_to_be32(s->l1_size), 2343 .l1_table_offset = cpu_to_be64(s->l1_table_offset), 2344 .refcount_table_offset = cpu_to_be64(s->refcount_table_offset), 2345 .refcount_table_clusters = cpu_to_be32(refcount_table_clusters), 2346 .nb_snapshots = cpu_to_be32(s->nb_snapshots), 2347 .snapshots_offset = cpu_to_be64(s->snapshots_offset), 2348 2349 /* Version 3 fields */ 2350 .incompatible_features = cpu_to_be64(s->incompatible_features), 2351 .compatible_features = cpu_to_be64(s->compatible_features), 2352 .autoclear_features = cpu_to_be64(s->autoclear_features), 2353 .refcount_order = cpu_to_be32(s->refcount_order), 2354 .header_length = cpu_to_be32(header_length), 2355 }; 2356 2357 /* For older versions, write a shorter header */ 2358 switch (s->qcow_version) { 2359 case 2: 2360 ret = offsetof(QCowHeader, incompatible_features); 2361 break; 2362 case 3: 2363 ret = sizeof(*header); 2364 break; 2365 default: 2366 ret = -EINVAL; 2367 goto fail; 2368 } 2369 2370 buf += ret; 2371 buflen -= ret; 2372 memset(buf, 0, buflen); 2373 2374 /* Preserve any unknown field in the header */ 2375 if (s->unknown_header_fields_size) { 2376 if (buflen < s->unknown_header_fields_size) { 2377 ret = -ENOSPC; 2378 goto fail; 2379 } 2380 2381 memcpy(buf, s->unknown_header_fields, s->unknown_header_fields_size); 2382 buf += s->unknown_header_fields_size; 2383 buflen -= s->unknown_header_fields_size; 2384 } 2385 2386 /* Backing file format header extension */ 2387 if (s->image_backing_format) { 2388 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BACKING_FORMAT, 2389 s->image_backing_format, 2390 strlen(s->image_backing_format), 2391 buflen); 2392 if (ret < 0) { 2393 goto fail; 2394 } 2395 2396 buf += ret; 2397 buflen -= ret; 2398 } 2399 2400 /* Full disk encryption header pointer extension */ 2401 if (s->crypto_header.offset != 0) { 2402 s->crypto_header.offset = cpu_to_be64(s->crypto_header.offset); 2403 s->crypto_header.length = cpu_to_be64(s->crypto_header.length); 2404 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_CRYPTO_HEADER, 2405 &s->crypto_header, sizeof(s->crypto_header), 2406 buflen); 2407 s->crypto_header.offset = be64_to_cpu(s->crypto_header.offset); 2408 s->crypto_header.length = be64_to_cpu(s->crypto_header.length); 2409 if (ret < 0) { 2410 goto fail; 2411 } 2412 buf += ret; 2413 buflen -= ret; 2414 } 2415 2416 /* Feature table */ 2417 if (s->qcow_version >= 3) { 2418 Qcow2Feature features[] = { 2419 { 2420 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE, 2421 .bit = QCOW2_INCOMPAT_DIRTY_BITNR, 2422 .name = "dirty bit", 2423 }, 2424 { 2425 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE, 2426 .bit = QCOW2_INCOMPAT_CORRUPT_BITNR, 2427 .name = "corrupt bit", 2428 }, 2429 { 2430 .type = QCOW2_FEAT_TYPE_COMPATIBLE, 2431 .bit = QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR, 2432 .name = "lazy refcounts", 2433 }, 2434 }; 2435 2436 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_FEATURE_TABLE, 2437 features, sizeof(features), buflen); 2438 if (ret < 0) { 2439 goto fail; 2440 } 2441 buf += ret; 2442 buflen -= ret; 2443 } 2444 2445 /* Bitmap extension */ 2446 if (s->nb_bitmaps > 0) { 2447 Qcow2BitmapHeaderExt bitmaps_header = { 2448 .nb_bitmaps = cpu_to_be32(s->nb_bitmaps), 2449 .bitmap_directory_size = 2450 cpu_to_be64(s->bitmap_directory_size), 2451 .bitmap_directory_offset = 2452 cpu_to_be64(s->bitmap_directory_offset) 2453 }; 2454 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BITMAPS, 2455 &bitmaps_header, sizeof(bitmaps_header), 2456 buflen); 2457 if (ret < 0) { 2458 goto fail; 2459 } 2460 buf += ret; 2461 buflen -= ret; 2462 } 2463 2464 /* Keep unknown header extensions */ 2465 QLIST_FOREACH(uext, &s->unknown_header_ext, next) { 2466 ret = header_ext_add(buf, uext->magic, uext->data, uext->len, buflen); 2467 if (ret < 0) { 2468 goto fail; 2469 } 2470 2471 buf += ret; 2472 buflen -= ret; 2473 } 2474 2475 /* End of header extensions */ 2476 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_END, NULL, 0, buflen); 2477 if (ret < 0) { 2478 goto fail; 2479 } 2480 2481 buf += ret; 2482 buflen -= ret; 2483 2484 /* Backing file name */ 2485 if (s->image_backing_file) { 2486 size_t backing_file_len = strlen(s->image_backing_file); 2487 2488 if (buflen < backing_file_len) { 2489 ret = -ENOSPC; 2490 goto fail; 2491 } 2492 2493 /* Using strncpy is ok here, since buf is not NUL-terminated. */ 2494 strncpy(buf, s->image_backing_file, buflen); 2495 2496 header->backing_file_offset = cpu_to_be64(buf - ((char*) header)); 2497 header->backing_file_size = cpu_to_be32(backing_file_len); 2498 } 2499 2500 /* Write the new header */ 2501 ret = bdrv_pwrite(bs->file, 0, header, s->cluster_size); 2502 if (ret < 0) { 2503 goto fail; 2504 } 2505 2506 ret = 0; 2507 fail: 2508 qemu_vfree(header); 2509 return ret; 2510 } 2511 2512 static int qcow2_change_backing_file(BlockDriverState *bs, 2513 const char *backing_file, const char *backing_fmt) 2514 { 2515 BDRVQcow2State *s = bs->opaque; 2516 2517 if (backing_file && strlen(backing_file) > 1023) { 2518 return -EINVAL; 2519 } 2520 2521 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: ""); 2522 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: ""); 2523 2524 g_free(s->image_backing_file); 2525 g_free(s->image_backing_format); 2526 2527 s->image_backing_file = backing_file ? g_strdup(bs->backing_file) : NULL; 2528 s->image_backing_format = backing_fmt ? g_strdup(bs->backing_format) : NULL; 2529 2530 return qcow2_update_header(bs); 2531 } 2532 2533 static int qcow2_crypt_method_from_format(const char *encryptfmt) 2534 { 2535 if (g_str_equal(encryptfmt, "luks")) { 2536 return QCOW_CRYPT_LUKS; 2537 } else if (g_str_equal(encryptfmt, "aes")) { 2538 return QCOW_CRYPT_AES; 2539 } else { 2540 return -EINVAL; 2541 } 2542 } 2543 2544 static int qcow2_set_up_encryption(BlockDriverState *bs, 2545 QCryptoBlockCreateOptions *cryptoopts, 2546 Error **errp) 2547 { 2548 BDRVQcow2State *s = bs->opaque; 2549 QCryptoBlock *crypto = NULL; 2550 int fmt, ret; 2551 2552 switch (cryptoopts->format) { 2553 case Q_CRYPTO_BLOCK_FORMAT_LUKS: 2554 fmt = QCOW_CRYPT_LUKS; 2555 break; 2556 case Q_CRYPTO_BLOCK_FORMAT_QCOW: 2557 fmt = QCOW_CRYPT_AES; 2558 break; 2559 default: 2560 error_setg(errp, "Crypto format not supported in qcow2"); 2561 return -EINVAL; 2562 } 2563 2564 s->crypt_method_header = fmt; 2565 2566 crypto = qcrypto_block_create(cryptoopts, "encrypt.", 2567 qcow2_crypto_hdr_init_func, 2568 qcow2_crypto_hdr_write_func, 2569 bs, errp); 2570 if (!crypto) { 2571 return -EINVAL; 2572 } 2573 2574 ret = qcow2_update_header(bs); 2575 if (ret < 0) { 2576 error_setg_errno(errp, -ret, "Could not write encryption header"); 2577 goto out; 2578 } 2579 2580 ret = 0; 2581 out: 2582 qcrypto_block_free(crypto); 2583 return ret; 2584 } 2585 2586 /** 2587 * Preallocates metadata structures for data clusters between @offset (in the 2588 * guest disk) and @new_length (which is thus generally the new guest disk 2589 * size). 2590 * 2591 * Returns: 0 on success, -errno on failure. 2592 */ 2593 static int coroutine_fn preallocate_co(BlockDriverState *bs, uint64_t offset, 2594 uint64_t new_length) 2595 { 2596 uint64_t bytes; 2597 uint64_t host_offset = 0; 2598 unsigned int cur_bytes; 2599 int ret; 2600 QCowL2Meta *meta; 2601 2602 assert(offset <= new_length); 2603 bytes = new_length - offset; 2604 2605 while (bytes) { 2606 cur_bytes = MIN(bytes, INT_MAX); 2607 ret = qcow2_alloc_cluster_offset(bs, offset, &cur_bytes, 2608 &host_offset, &meta); 2609 if (ret < 0) { 2610 return ret; 2611 } 2612 2613 while (meta) { 2614 QCowL2Meta *next = meta->next; 2615 2616 ret = qcow2_alloc_cluster_link_l2(bs, meta); 2617 if (ret < 0) { 2618 qcow2_free_any_clusters(bs, meta->alloc_offset, 2619 meta->nb_clusters, QCOW2_DISCARD_NEVER); 2620 return ret; 2621 } 2622 2623 /* There are no dependent requests, but we need to remove our 2624 * request from the list of in-flight requests */ 2625 QLIST_REMOVE(meta, next_in_flight); 2626 2627 g_free(meta); 2628 meta = next; 2629 } 2630 2631 /* TODO Preallocate data if requested */ 2632 2633 bytes -= cur_bytes; 2634 offset += cur_bytes; 2635 } 2636 2637 /* 2638 * It is expected that the image file is large enough to actually contain 2639 * all of the allocated clusters (otherwise we get failing reads after 2640 * EOF). Extend the image to the last allocated sector. 2641 */ 2642 if (host_offset != 0) { 2643 uint8_t data = 0; 2644 ret = bdrv_pwrite(bs->file, (host_offset + cur_bytes) - 1, 2645 &data, 1); 2646 if (ret < 0) { 2647 return ret; 2648 } 2649 } 2650 2651 return 0; 2652 } 2653 2654 /* qcow2_refcount_metadata_size: 2655 * @clusters: number of clusters to refcount (including data and L1/L2 tables) 2656 * @cluster_size: size of a cluster, in bytes 2657 * @refcount_order: refcount bits power-of-2 exponent 2658 * @generous_increase: allow for the refcount table to be 1.5x as large as it 2659 * needs to be 2660 * 2661 * Returns: Number of bytes required for refcount blocks and table metadata. 2662 */ 2663 int64_t qcow2_refcount_metadata_size(int64_t clusters, size_t cluster_size, 2664 int refcount_order, bool generous_increase, 2665 uint64_t *refblock_count) 2666 { 2667 /* 2668 * Every host cluster is reference-counted, including metadata (even 2669 * refcount metadata is recursively included). 2670 * 2671 * An accurate formula for the size of refcount metadata size is difficult 2672 * to derive. An easier method of calculation is finding the fixed point 2673 * where no further refcount blocks or table clusters are required to 2674 * reference count every cluster. 2675 */ 2676 int64_t blocks_per_table_cluster = cluster_size / sizeof(uint64_t); 2677 int64_t refcounts_per_block = cluster_size * 8 / (1 << refcount_order); 2678 int64_t table = 0; /* number of refcount table clusters */ 2679 int64_t blocks = 0; /* number of refcount block clusters */ 2680 int64_t last; 2681 int64_t n = 0; 2682 2683 do { 2684 last = n; 2685 blocks = DIV_ROUND_UP(clusters + table + blocks, refcounts_per_block); 2686 table = DIV_ROUND_UP(blocks, blocks_per_table_cluster); 2687 n = clusters + blocks + table; 2688 2689 if (n == last && generous_increase) { 2690 clusters += DIV_ROUND_UP(table, 2); 2691 n = 0; /* force another loop */ 2692 generous_increase = false; 2693 } 2694 } while (n != last); 2695 2696 if (refblock_count) { 2697 *refblock_count = blocks; 2698 } 2699 2700 return (blocks + table) * cluster_size; 2701 } 2702 2703 /** 2704 * qcow2_calc_prealloc_size: 2705 * @total_size: virtual disk size in bytes 2706 * @cluster_size: cluster size in bytes 2707 * @refcount_order: refcount bits power-of-2 exponent 2708 * 2709 * Returns: Total number of bytes required for the fully allocated image 2710 * (including metadata). 2711 */ 2712 static int64_t qcow2_calc_prealloc_size(int64_t total_size, 2713 size_t cluster_size, 2714 int refcount_order) 2715 { 2716 int64_t meta_size = 0; 2717 uint64_t nl1e, nl2e; 2718 int64_t aligned_total_size = ROUND_UP(total_size, cluster_size); 2719 2720 /* header: 1 cluster */ 2721 meta_size += cluster_size; 2722 2723 /* total size of L2 tables */ 2724 nl2e = aligned_total_size / cluster_size; 2725 nl2e = ROUND_UP(nl2e, cluster_size / sizeof(uint64_t)); 2726 meta_size += nl2e * sizeof(uint64_t); 2727 2728 /* total size of L1 tables */ 2729 nl1e = nl2e * sizeof(uint64_t) / cluster_size; 2730 nl1e = ROUND_UP(nl1e, cluster_size / sizeof(uint64_t)); 2731 meta_size += nl1e * sizeof(uint64_t); 2732 2733 /* total size of refcount table and blocks */ 2734 meta_size += qcow2_refcount_metadata_size( 2735 (meta_size + aligned_total_size) / cluster_size, 2736 cluster_size, refcount_order, false, NULL); 2737 2738 return meta_size + aligned_total_size; 2739 } 2740 2741 static bool validate_cluster_size(size_t cluster_size, Error **errp) 2742 { 2743 int cluster_bits = ctz32(cluster_size); 2744 if (cluster_bits < MIN_CLUSTER_BITS || cluster_bits > MAX_CLUSTER_BITS || 2745 (1 << cluster_bits) != cluster_size) 2746 { 2747 error_setg(errp, "Cluster size must be a power of two between %d and " 2748 "%dk", 1 << MIN_CLUSTER_BITS, 1 << (MAX_CLUSTER_BITS - 10)); 2749 return false; 2750 } 2751 return true; 2752 } 2753 2754 static size_t qcow2_opt_get_cluster_size_del(QemuOpts *opts, Error **errp) 2755 { 2756 size_t cluster_size; 2757 2758 cluster_size = qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE, 2759 DEFAULT_CLUSTER_SIZE); 2760 if (!validate_cluster_size(cluster_size, errp)) { 2761 return 0; 2762 } 2763 return cluster_size; 2764 } 2765 2766 static int qcow2_opt_get_version_del(QemuOpts *opts, Error **errp) 2767 { 2768 char *buf; 2769 int ret; 2770 2771 buf = qemu_opt_get_del(opts, BLOCK_OPT_COMPAT_LEVEL); 2772 if (!buf) { 2773 ret = 3; /* default */ 2774 } else if (!strcmp(buf, "0.10")) { 2775 ret = 2; 2776 } else if (!strcmp(buf, "1.1")) { 2777 ret = 3; 2778 } else { 2779 error_setg(errp, "Invalid compatibility level: '%s'", buf); 2780 ret = -EINVAL; 2781 } 2782 g_free(buf); 2783 return ret; 2784 } 2785 2786 static uint64_t qcow2_opt_get_refcount_bits_del(QemuOpts *opts, int version, 2787 Error **errp) 2788 { 2789 uint64_t refcount_bits; 2790 2791 refcount_bits = qemu_opt_get_number_del(opts, BLOCK_OPT_REFCOUNT_BITS, 16); 2792 if (refcount_bits > 64 || !is_power_of_2(refcount_bits)) { 2793 error_setg(errp, "Refcount width must be a power of two and may not " 2794 "exceed 64 bits"); 2795 return 0; 2796 } 2797 2798 if (version < 3 && refcount_bits != 16) { 2799 error_setg(errp, "Different refcount widths than 16 bits require " 2800 "compatibility level 1.1 or above (use compat=1.1 or " 2801 "greater)"); 2802 return 0; 2803 } 2804 2805 return refcount_bits; 2806 } 2807 2808 static int coroutine_fn 2809 qcow2_co_create(BlockdevCreateOptions *create_options, Error **errp) 2810 { 2811 BlockdevCreateOptionsQcow2 *qcow2_opts; 2812 QDict *options; 2813 2814 /* 2815 * Open the image file and write a minimal qcow2 header. 2816 * 2817 * We keep things simple and start with a zero-sized image. We also 2818 * do without refcount blocks or a L1 table for now. We'll fix the 2819 * inconsistency later. 2820 * 2821 * We do need a refcount table because growing the refcount table means 2822 * allocating two new refcount blocks - the seconds of which would be at 2823 * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file 2824 * size for any qcow2 image. 2825 */ 2826 BlockBackend *blk = NULL; 2827 BlockDriverState *bs = NULL; 2828 QCowHeader *header; 2829 size_t cluster_size; 2830 int version; 2831 int refcount_order; 2832 uint64_t* refcount_table; 2833 Error *local_err = NULL; 2834 int ret; 2835 2836 assert(create_options->driver == BLOCKDEV_DRIVER_QCOW2); 2837 qcow2_opts = &create_options->u.qcow2; 2838 2839 bs = bdrv_open_blockdev_ref(qcow2_opts->file, errp); 2840 if (bs == NULL) { 2841 return -EIO; 2842 } 2843 2844 /* Validate options and set default values */ 2845 if (!QEMU_IS_ALIGNED(qcow2_opts->size, BDRV_SECTOR_SIZE)) { 2846 error_setg(errp, "Image size must be a multiple of 512 bytes"); 2847 ret = -EINVAL; 2848 goto out; 2849 } 2850 2851 if (qcow2_opts->has_version) { 2852 switch (qcow2_opts->version) { 2853 case BLOCKDEV_QCOW2_VERSION_V2: 2854 version = 2; 2855 break; 2856 case BLOCKDEV_QCOW2_VERSION_V3: 2857 version = 3; 2858 break; 2859 default: 2860 g_assert_not_reached(); 2861 } 2862 } else { 2863 version = 3; 2864 } 2865 2866 if (qcow2_opts->has_cluster_size) { 2867 cluster_size = qcow2_opts->cluster_size; 2868 } else { 2869 cluster_size = DEFAULT_CLUSTER_SIZE; 2870 } 2871 2872 if (!validate_cluster_size(cluster_size, errp)) { 2873 ret = -EINVAL; 2874 goto out; 2875 } 2876 2877 if (!qcow2_opts->has_preallocation) { 2878 qcow2_opts->preallocation = PREALLOC_MODE_OFF; 2879 } 2880 if (qcow2_opts->has_backing_file && 2881 qcow2_opts->preallocation != PREALLOC_MODE_OFF) 2882 { 2883 error_setg(errp, "Backing file and preallocation cannot be used at " 2884 "the same time"); 2885 ret = -EINVAL; 2886 goto out; 2887 } 2888 if (qcow2_opts->has_backing_fmt && !qcow2_opts->has_backing_file) { 2889 error_setg(errp, "Backing format cannot be used without backing file"); 2890 ret = -EINVAL; 2891 goto out; 2892 } 2893 2894 if (!qcow2_opts->has_lazy_refcounts) { 2895 qcow2_opts->lazy_refcounts = false; 2896 } 2897 if (version < 3 && qcow2_opts->lazy_refcounts) { 2898 error_setg(errp, "Lazy refcounts only supported with compatibility " 2899 "level 1.1 and above (use version=v3 or greater)"); 2900 ret = -EINVAL; 2901 goto out; 2902 } 2903 2904 if (!qcow2_opts->has_refcount_bits) { 2905 qcow2_opts->refcount_bits = 16; 2906 } 2907 if (qcow2_opts->refcount_bits > 64 || 2908 !is_power_of_2(qcow2_opts->refcount_bits)) 2909 { 2910 error_setg(errp, "Refcount width must be a power of two and may not " 2911 "exceed 64 bits"); 2912 ret = -EINVAL; 2913 goto out; 2914 } 2915 if (version < 3 && qcow2_opts->refcount_bits != 16) { 2916 error_setg(errp, "Different refcount widths than 16 bits require " 2917 "compatibility level 1.1 or above (use version=v3 or " 2918 "greater)"); 2919 ret = -EINVAL; 2920 goto out; 2921 } 2922 refcount_order = ctz32(qcow2_opts->refcount_bits); 2923 2924 2925 /* Create BlockBackend to write to the image */ 2926 blk = blk_new(BLK_PERM_WRITE | BLK_PERM_RESIZE, BLK_PERM_ALL); 2927 ret = blk_insert_bs(blk, bs, errp); 2928 if (ret < 0) { 2929 goto out; 2930 } 2931 blk_set_allow_write_beyond_eof(blk, true); 2932 2933 /* Clear the protocol layer and preallocate it if necessary */ 2934 ret = blk_truncate(blk, 0, PREALLOC_MODE_OFF, errp); 2935 if (ret < 0) { 2936 goto out; 2937 } 2938 2939 if (qcow2_opts->preallocation == PREALLOC_MODE_FULL || 2940 qcow2_opts->preallocation == PREALLOC_MODE_FALLOC) 2941 { 2942 int64_t prealloc_size = 2943 qcow2_calc_prealloc_size(qcow2_opts->size, cluster_size, 2944 refcount_order); 2945 2946 ret = blk_truncate(blk, prealloc_size, qcow2_opts->preallocation, errp); 2947 if (ret < 0) { 2948 goto out; 2949 } 2950 } 2951 2952 /* Write the header */ 2953 QEMU_BUILD_BUG_ON((1 << MIN_CLUSTER_BITS) < sizeof(*header)); 2954 header = g_malloc0(cluster_size); 2955 *header = (QCowHeader) { 2956 .magic = cpu_to_be32(QCOW_MAGIC), 2957 .version = cpu_to_be32(version), 2958 .cluster_bits = cpu_to_be32(ctz32(cluster_size)), 2959 .size = cpu_to_be64(0), 2960 .l1_table_offset = cpu_to_be64(0), 2961 .l1_size = cpu_to_be32(0), 2962 .refcount_table_offset = cpu_to_be64(cluster_size), 2963 .refcount_table_clusters = cpu_to_be32(1), 2964 .refcount_order = cpu_to_be32(refcount_order), 2965 .header_length = cpu_to_be32(sizeof(*header)), 2966 }; 2967 2968 /* We'll update this to correct value later */ 2969 header->crypt_method = cpu_to_be32(QCOW_CRYPT_NONE); 2970 2971 if (qcow2_opts->lazy_refcounts) { 2972 header->compatible_features |= 2973 cpu_to_be64(QCOW2_COMPAT_LAZY_REFCOUNTS); 2974 } 2975 2976 ret = blk_pwrite(blk, 0, header, cluster_size, 0); 2977 g_free(header); 2978 if (ret < 0) { 2979 error_setg_errno(errp, -ret, "Could not write qcow2 header"); 2980 goto out; 2981 } 2982 2983 /* Write a refcount table with one refcount block */ 2984 refcount_table = g_malloc0(2 * cluster_size); 2985 refcount_table[0] = cpu_to_be64(2 * cluster_size); 2986 ret = blk_pwrite(blk, cluster_size, refcount_table, 2 * cluster_size, 0); 2987 g_free(refcount_table); 2988 2989 if (ret < 0) { 2990 error_setg_errno(errp, -ret, "Could not write refcount table"); 2991 goto out; 2992 } 2993 2994 blk_unref(blk); 2995 blk = NULL; 2996 2997 /* 2998 * And now open the image and make it consistent first (i.e. increase the 2999 * refcount of the cluster that is occupied by the header and the refcount 3000 * table) 3001 */ 3002 options = qdict_new(); 3003 qdict_put_str(options, "driver", "qcow2"); 3004 qdict_put_str(options, "file", bs->node_name); 3005 blk = blk_new_open(NULL, NULL, options, 3006 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_NO_FLUSH, 3007 &local_err); 3008 if (blk == NULL) { 3009 error_propagate(errp, local_err); 3010 ret = -EIO; 3011 goto out; 3012 } 3013 3014 ret = qcow2_alloc_clusters(blk_bs(blk), 3 * cluster_size); 3015 if (ret < 0) { 3016 error_setg_errno(errp, -ret, "Could not allocate clusters for qcow2 " 3017 "header and refcount table"); 3018 goto out; 3019 3020 } else if (ret != 0) { 3021 error_report("Huh, first cluster in empty image is already in use?"); 3022 abort(); 3023 } 3024 3025 /* Create a full header (including things like feature table) */ 3026 ret = qcow2_update_header(blk_bs(blk)); 3027 if (ret < 0) { 3028 error_setg_errno(errp, -ret, "Could not update qcow2 header"); 3029 goto out; 3030 } 3031 3032 /* Okay, now that we have a valid image, let's give it the right size */ 3033 ret = blk_truncate(blk, qcow2_opts->size, PREALLOC_MODE_OFF, errp); 3034 if (ret < 0) { 3035 error_prepend(errp, "Could not resize image: "); 3036 goto out; 3037 } 3038 3039 /* Want a backing file? There you go.*/ 3040 if (qcow2_opts->has_backing_file) { 3041 const char *backing_format = NULL; 3042 3043 if (qcow2_opts->has_backing_fmt) { 3044 backing_format = BlockdevDriver_str(qcow2_opts->backing_fmt); 3045 } 3046 3047 ret = bdrv_change_backing_file(blk_bs(blk), qcow2_opts->backing_file, 3048 backing_format); 3049 if (ret < 0) { 3050 error_setg_errno(errp, -ret, "Could not assign backing file '%s' " 3051 "with format '%s'", qcow2_opts->backing_file, 3052 backing_format); 3053 goto out; 3054 } 3055 } 3056 3057 /* Want encryption? There you go. */ 3058 if (qcow2_opts->has_encrypt) { 3059 ret = qcow2_set_up_encryption(blk_bs(blk), qcow2_opts->encrypt, errp); 3060 if (ret < 0) { 3061 goto out; 3062 } 3063 } 3064 3065 /* And if we're supposed to preallocate metadata, do that now */ 3066 if (qcow2_opts->preallocation != PREALLOC_MODE_OFF) { 3067 BDRVQcow2State *s = blk_bs(blk)->opaque; 3068 qemu_co_mutex_lock(&s->lock); 3069 ret = preallocate_co(blk_bs(blk), 0, qcow2_opts->size); 3070 qemu_co_mutex_unlock(&s->lock); 3071 3072 if (ret < 0) { 3073 error_setg_errno(errp, -ret, "Could not preallocate metadata"); 3074 goto out; 3075 } 3076 } 3077 3078 blk_unref(blk); 3079 blk = NULL; 3080 3081 /* Reopen the image without BDRV_O_NO_FLUSH to flush it before returning. 3082 * Using BDRV_O_NO_IO, since encryption is now setup we don't want to 3083 * have to setup decryption context. We're not doing any I/O on the top 3084 * level BlockDriverState, only lower layers, where BDRV_O_NO_IO does 3085 * not have effect. 3086 */ 3087 options = qdict_new(); 3088 qdict_put_str(options, "driver", "qcow2"); 3089 qdict_put_str(options, "file", bs->node_name); 3090 blk = blk_new_open(NULL, NULL, options, 3091 BDRV_O_RDWR | BDRV_O_NO_BACKING | BDRV_O_NO_IO, 3092 &local_err); 3093 if (blk == NULL) { 3094 error_propagate(errp, local_err); 3095 ret = -EIO; 3096 goto out; 3097 } 3098 3099 ret = 0; 3100 out: 3101 blk_unref(blk); 3102 bdrv_unref(bs); 3103 return ret; 3104 } 3105 3106 static int coroutine_fn qcow2_co_create_opts(const char *filename, QemuOpts *opts, 3107 Error **errp) 3108 { 3109 BlockdevCreateOptions *create_options = NULL; 3110 QDict *qdict; 3111 Visitor *v; 3112 BlockDriverState *bs = NULL; 3113 Error *local_err = NULL; 3114 const char *val; 3115 int ret; 3116 3117 /* Only the keyval visitor supports the dotted syntax needed for 3118 * encryption, so go through a QDict before getting a QAPI type. Ignore 3119 * options meant for the protocol layer so that the visitor doesn't 3120 * complain. */ 3121 qdict = qemu_opts_to_qdict_filtered(opts, NULL, bdrv_qcow2.create_opts, 3122 true); 3123 3124 /* Handle encryption options */ 3125 val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT); 3126 if (val && !strcmp(val, "on")) { 3127 qdict_put_str(qdict, BLOCK_OPT_ENCRYPT, "qcow"); 3128 } else if (val && !strcmp(val, "off")) { 3129 qdict_del(qdict, BLOCK_OPT_ENCRYPT); 3130 } 3131 3132 val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT); 3133 if (val && !strcmp(val, "aes")) { 3134 qdict_put_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT, "qcow"); 3135 } 3136 3137 /* Convert compat=0.10/1.1 into compat=v2/v3, to be renamed into 3138 * version=v2/v3 below. */ 3139 val = qdict_get_try_str(qdict, BLOCK_OPT_COMPAT_LEVEL); 3140 if (val && !strcmp(val, "0.10")) { 3141 qdict_put_str(qdict, BLOCK_OPT_COMPAT_LEVEL, "v2"); 3142 } else if (val && !strcmp(val, "1.1")) { 3143 qdict_put_str(qdict, BLOCK_OPT_COMPAT_LEVEL, "v3"); 3144 } 3145 3146 /* Change legacy command line options into QMP ones */ 3147 static const QDictRenames opt_renames[] = { 3148 { BLOCK_OPT_BACKING_FILE, "backing-file" }, 3149 { BLOCK_OPT_BACKING_FMT, "backing-fmt" }, 3150 { BLOCK_OPT_CLUSTER_SIZE, "cluster-size" }, 3151 { BLOCK_OPT_LAZY_REFCOUNTS, "lazy-refcounts" }, 3152 { BLOCK_OPT_REFCOUNT_BITS, "refcount-bits" }, 3153 { BLOCK_OPT_ENCRYPT, BLOCK_OPT_ENCRYPT_FORMAT }, 3154 { BLOCK_OPT_COMPAT_LEVEL, "version" }, 3155 { NULL, NULL }, 3156 }; 3157 3158 if (!qdict_rename_keys(qdict, opt_renames, errp)) { 3159 ret = -EINVAL; 3160 goto finish; 3161 } 3162 3163 /* Create and open the file (protocol layer) */ 3164 ret = bdrv_create_file(filename, opts, errp); 3165 if (ret < 0) { 3166 goto finish; 3167 } 3168 3169 bs = bdrv_open(filename, NULL, NULL, 3170 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp); 3171 if (bs == NULL) { 3172 ret = -EIO; 3173 goto finish; 3174 } 3175 3176 /* Set 'driver' and 'node' options */ 3177 qdict_put_str(qdict, "driver", "qcow2"); 3178 qdict_put_str(qdict, "file", bs->node_name); 3179 3180 /* Now get the QAPI type BlockdevCreateOptions */ 3181 v = qobject_input_visitor_new_flat_confused(qdict, errp); 3182 if (!v) { 3183 ret = -EINVAL; 3184 goto finish; 3185 } 3186 3187 visit_type_BlockdevCreateOptions(v, NULL, &create_options, &local_err); 3188 visit_free(v); 3189 3190 if (local_err) { 3191 error_propagate(errp, local_err); 3192 ret = -EINVAL; 3193 goto finish; 3194 } 3195 3196 /* Silently round up size */ 3197 create_options->u.qcow2.size = ROUND_UP(create_options->u.qcow2.size, 3198 BDRV_SECTOR_SIZE); 3199 3200 /* Create the qcow2 image (format layer) */ 3201 ret = qcow2_co_create(create_options, errp); 3202 if (ret < 0) { 3203 goto finish; 3204 } 3205 3206 ret = 0; 3207 finish: 3208 qobject_unref(qdict); 3209 bdrv_unref(bs); 3210 qapi_free_BlockdevCreateOptions(create_options); 3211 return ret; 3212 } 3213 3214 3215 static bool is_zero(BlockDriverState *bs, int64_t offset, int64_t bytes) 3216 { 3217 int64_t nr; 3218 int res; 3219 3220 /* Clamp to image length, before checking status of underlying sectors */ 3221 if (offset + bytes > bs->total_sectors * BDRV_SECTOR_SIZE) { 3222 bytes = bs->total_sectors * BDRV_SECTOR_SIZE - offset; 3223 } 3224 3225 if (!bytes) { 3226 return true; 3227 } 3228 res = bdrv_block_status_above(bs, NULL, offset, bytes, &nr, NULL, NULL); 3229 return res >= 0 && (res & BDRV_BLOCK_ZERO) && nr == bytes; 3230 } 3231 3232 static coroutine_fn int qcow2_co_pwrite_zeroes(BlockDriverState *bs, 3233 int64_t offset, int bytes, BdrvRequestFlags flags) 3234 { 3235 int ret; 3236 BDRVQcow2State *s = bs->opaque; 3237 3238 uint32_t head = offset % s->cluster_size; 3239 uint32_t tail = (offset + bytes) % s->cluster_size; 3240 3241 trace_qcow2_pwrite_zeroes_start_req(qemu_coroutine_self(), offset, bytes); 3242 if (offset + bytes == bs->total_sectors * BDRV_SECTOR_SIZE) { 3243 tail = 0; 3244 } 3245 3246 if (head || tail) { 3247 uint64_t off; 3248 unsigned int nr; 3249 3250 assert(head + bytes <= s->cluster_size); 3251 3252 /* check whether remainder of cluster already reads as zero */ 3253 if (!(is_zero(bs, offset - head, head) && 3254 is_zero(bs, offset + bytes, 3255 tail ? s->cluster_size - tail : 0))) { 3256 return -ENOTSUP; 3257 } 3258 3259 qemu_co_mutex_lock(&s->lock); 3260 /* We can have new write after previous check */ 3261 offset = QEMU_ALIGN_DOWN(offset, s->cluster_size); 3262 bytes = s->cluster_size; 3263 nr = s->cluster_size; 3264 ret = qcow2_get_cluster_offset(bs, offset, &nr, &off); 3265 if (ret != QCOW2_CLUSTER_UNALLOCATED && 3266 ret != QCOW2_CLUSTER_ZERO_PLAIN && 3267 ret != QCOW2_CLUSTER_ZERO_ALLOC) { 3268 qemu_co_mutex_unlock(&s->lock); 3269 return -ENOTSUP; 3270 } 3271 } else { 3272 qemu_co_mutex_lock(&s->lock); 3273 } 3274 3275 trace_qcow2_pwrite_zeroes(qemu_coroutine_self(), offset, bytes); 3276 3277 /* Whatever is left can use real zero clusters */ 3278 ret = qcow2_cluster_zeroize(bs, offset, bytes, flags); 3279 qemu_co_mutex_unlock(&s->lock); 3280 3281 return ret; 3282 } 3283 3284 static coroutine_fn int qcow2_co_pdiscard(BlockDriverState *bs, 3285 int64_t offset, int bytes) 3286 { 3287 int ret; 3288 BDRVQcow2State *s = bs->opaque; 3289 3290 if (!QEMU_IS_ALIGNED(offset | bytes, s->cluster_size)) { 3291 assert(bytes < s->cluster_size); 3292 /* Ignore partial clusters, except for the special case of the 3293 * complete partial cluster at the end of an unaligned file */ 3294 if (!QEMU_IS_ALIGNED(offset, s->cluster_size) || 3295 offset + bytes != bs->total_sectors * BDRV_SECTOR_SIZE) { 3296 return -ENOTSUP; 3297 } 3298 } 3299 3300 qemu_co_mutex_lock(&s->lock); 3301 ret = qcow2_cluster_discard(bs, offset, bytes, QCOW2_DISCARD_REQUEST, 3302 false); 3303 qemu_co_mutex_unlock(&s->lock); 3304 return ret; 3305 } 3306 3307 static int coroutine_fn 3308 qcow2_co_copy_range_from(BlockDriverState *bs, 3309 BdrvChild *src, uint64_t src_offset, 3310 BdrvChild *dst, uint64_t dst_offset, 3311 uint64_t bytes, BdrvRequestFlags read_flags, 3312 BdrvRequestFlags write_flags) 3313 { 3314 BDRVQcow2State *s = bs->opaque; 3315 int ret; 3316 unsigned int cur_bytes; /* number of bytes in current iteration */ 3317 BdrvChild *child = NULL; 3318 BdrvRequestFlags cur_write_flags; 3319 3320 assert(!bs->encrypted); 3321 qemu_co_mutex_lock(&s->lock); 3322 3323 while (bytes != 0) { 3324 uint64_t copy_offset = 0; 3325 /* prepare next request */ 3326 cur_bytes = MIN(bytes, INT_MAX); 3327 cur_write_flags = write_flags; 3328 3329 ret = qcow2_get_cluster_offset(bs, src_offset, &cur_bytes, ©_offset); 3330 if (ret < 0) { 3331 goto out; 3332 } 3333 3334 switch (ret) { 3335 case QCOW2_CLUSTER_UNALLOCATED: 3336 if (bs->backing && bs->backing->bs) { 3337 int64_t backing_length = bdrv_getlength(bs->backing->bs); 3338 if (src_offset >= backing_length) { 3339 cur_write_flags |= BDRV_REQ_ZERO_WRITE; 3340 } else { 3341 child = bs->backing; 3342 cur_bytes = MIN(cur_bytes, backing_length - src_offset); 3343 copy_offset = src_offset; 3344 } 3345 } else { 3346 cur_write_flags |= BDRV_REQ_ZERO_WRITE; 3347 } 3348 break; 3349 3350 case QCOW2_CLUSTER_ZERO_PLAIN: 3351 case QCOW2_CLUSTER_ZERO_ALLOC: 3352 cur_write_flags |= BDRV_REQ_ZERO_WRITE; 3353 break; 3354 3355 case QCOW2_CLUSTER_COMPRESSED: 3356 ret = -ENOTSUP; 3357 goto out; 3358 3359 case QCOW2_CLUSTER_NORMAL: 3360 child = bs->file; 3361 copy_offset += offset_into_cluster(s, src_offset); 3362 if ((copy_offset & 511) != 0) { 3363 ret = -EIO; 3364 goto out; 3365 } 3366 break; 3367 3368 default: 3369 abort(); 3370 } 3371 qemu_co_mutex_unlock(&s->lock); 3372 ret = bdrv_co_copy_range_from(child, 3373 copy_offset, 3374 dst, dst_offset, 3375 cur_bytes, read_flags, cur_write_flags); 3376 qemu_co_mutex_lock(&s->lock); 3377 if (ret < 0) { 3378 goto out; 3379 } 3380 3381 bytes -= cur_bytes; 3382 src_offset += cur_bytes; 3383 dst_offset += cur_bytes; 3384 } 3385 ret = 0; 3386 3387 out: 3388 qemu_co_mutex_unlock(&s->lock); 3389 return ret; 3390 } 3391 3392 static int coroutine_fn 3393 qcow2_co_copy_range_to(BlockDriverState *bs, 3394 BdrvChild *src, uint64_t src_offset, 3395 BdrvChild *dst, uint64_t dst_offset, 3396 uint64_t bytes, BdrvRequestFlags read_flags, 3397 BdrvRequestFlags write_flags) 3398 { 3399 BDRVQcow2State *s = bs->opaque; 3400 int offset_in_cluster; 3401 int ret; 3402 unsigned int cur_bytes; /* number of sectors in current iteration */ 3403 uint64_t cluster_offset; 3404 QCowL2Meta *l2meta = NULL; 3405 3406 assert(!bs->encrypted); 3407 3408 qemu_co_mutex_lock(&s->lock); 3409 3410 while (bytes != 0) { 3411 3412 l2meta = NULL; 3413 3414 offset_in_cluster = offset_into_cluster(s, dst_offset); 3415 cur_bytes = MIN(bytes, INT_MAX); 3416 3417 /* TODO: 3418 * If src->bs == dst->bs, we could simply copy by incrementing 3419 * the refcnt, without copying user data. 3420 * Or if src->bs == dst->bs->backing->bs, we could copy by discarding. */ 3421 ret = qcow2_alloc_cluster_offset(bs, dst_offset, &cur_bytes, 3422 &cluster_offset, &l2meta); 3423 if (ret < 0) { 3424 goto fail; 3425 } 3426 3427 assert((cluster_offset & 511) == 0); 3428 3429 ret = qcow2_pre_write_overlap_check(bs, 0, 3430 cluster_offset + offset_in_cluster, cur_bytes); 3431 if (ret < 0) { 3432 goto fail; 3433 } 3434 3435 qemu_co_mutex_unlock(&s->lock); 3436 ret = bdrv_co_copy_range_to(src, src_offset, 3437 bs->file, 3438 cluster_offset + offset_in_cluster, 3439 cur_bytes, read_flags, write_flags); 3440 qemu_co_mutex_lock(&s->lock); 3441 if (ret < 0) { 3442 goto fail; 3443 } 3444 3445 ret = qcow2_handle_l2meta(bs, &l2meta, true); 3446 if (ret) { 3447 goto fail; 3448 } 3449 3450 bytes -= cur_bytes; 3451 src_offset += cur_bytes; 3452 dst_offset += cur_bytes; 3453 } 3454 ret = 0; 3455 3456 fail: 3457 qcow2_handle_l2meta(bs, &l2meta, false); 3458 3459 qemu_co_mutex_unlock(&s->lock); 3460 3461 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret); 3462 3463 return ret; 3464 } 3465 3466 static int coroutine_fn qcow2_co_truncate(BlockDriverState *bs, int64_t offset, 3467 PreallocMode prealloc, Error **errp) 3468 { 3469 BDRVQcow2State *s = bs->opaque; 3470 uint64_t old_length; 3471 int64_t new_l1_size; 3472 int ret; 3473 QDict *options; 3474 3475 if (prealloc != PREALLOC_MODE_OFF && prealloc != PREALLOC_MODE_METADATA && 3476 prealloc != PREALLOC_MODE_FALLOC && prealloc != PREALLOC_MODE_FULL) 3477 { 3478 error_setg(errp, "Unsupported preallocation mode '%s'", 3479 PreallocMode_str(prealloc)); 3480 return -ENOTSUP; 3481 } 3482 3483 if (offset & 511) { 3484 error_setg(errp, "The new size must be a multiple of 512"); 3485 return -EINVAL; 3486 } 3487 3488 qemu_co_mutex_lock(&s->lock); 3489 3490 /* cannot proceed if image has snapshots */ 3491 if (s->nb_snapshots) { 3492 error_setg(errp, "Can't resize an image which has snapshots"); 3493 ret = -ENOTSUP; 3494 goto fail; 3495 } 3496 3497 /* cannot proceed if image has bitmaps */ 3498 if (s->nb_bitmaps) { 3499 /* TODO: resize bitmaps in the image */ 3500 error_setg(errp, "Can't resize an image which has bitmaps"); 3501 ret = -ENOTSUP; 3502 goto fail; 3503 } 3504 3505 old_length = bs->total_sectors * BDRV_SECTOR_SIZE; 3506 new_l1_size = size_to_l1(s, offset); 3507 3508 if (offset < old_length) { 3509 int64_t last_cluster, old_file_size; 3510 if (prealloc != PREALLOC_MODE_OFF) { 3511 error_setg(errp, 3512 "Preallocation can't be used for shrinking an image"); 3513 ret = -EINVAL; 3514 goto fail; 3515 } 3516 3517 ret = qcow2_cluster_discard(bs, ROUND_UP(offset, s->cluster_size), 3518 old_length - ROUND_UP(offset, 3519 s->cluster_size), 3520 QCOW2_DISCARD_ALWAYS, true); 3521 if (ret < 0) { 3522 error_setg_errno(errp, -ret, "Failed to discard cropped clusters"); 3523 goto fail; 3524 } 3525 3526 ret = qcow2_shrink_l1_table(bs, new_l1_size); 3527 if (ret < 0) { 3528 error_setg_errno(errp, -ret, 3529 "Failed to reduce the number of L2 tables"); 3530 goto fail; 3531 } 3532 3533 ret = qcow2_shrink_reftable(bs); 3534 if (ret < 0) { 3535 error_setg_errno(errp, -ret, 3536 "Failed to discard unused refblocks"); 3537 goto fail; 3538 } 3539 3540 old_file_size = bdrv_getlength(bs->file->bs); 3541 if (old_file_size < 0) { 3542 error_setg_errno(errp, -old_file_size, 3543 "Failed to inquire current file length"); 3544 ret = old_file_size; 3545 goto fail; 3546 } 3547 last_cluster = qcow2_get_last_cluster(bs, old_file_size); 3548 if (last_cluster < 0) { 3549 error_setg_errno(errp, -last_cluster, 3550 "Failed to find the last cluster"); 3551 ret = last_cluster; 3552 goto fail; 3553 } 3554 if ((last_cluster + 1) * s->cluster_size < old_file_size) { 3555 Error *local_err = NULL; 3556 3557 bdrv_co_truncate(bs->file, (last_cluster + 1) * s->cluster_size, 3558 PREALLOC_MODE_OFF, &local_err); 3559 if (local_err) { 3560 warn_reportf_err(local_err, 3561 "Failed to truncate the tail of the image: "); 3562 } 3563 } 3564 } else { 3565 ret = qcow2_grow_l1_table(bs, new_l1_size, true); 3566 if (ret < 0) { 3567 error_setg_errno(errp, -ret, "Failed to grow the L1 table"); 3568 goto fail; 3569 } 3570 } 3571 3572 switch (prealloc) { 3573 case PREALLOC_MODE_OFF: 3574 break; 3575 3576 case PREALLOC_MODE_METADATA: 3577 ret = preallocate_co(bs, old_length, offset); 3578 if (ret < 0) { 3579 error_setg_errno(errp, -ret, "Preallocation failed"); 3580 goto fail; 3581 } 3582 break; 3583 3584 case PREALLOC_MODE_FALLOC: 3585 case PREALLOC_MODE_FULL: 3586 { 3587 int64_t allocation_start, host_offset, guest_offset; 3588 int64_t clusters_allocated; 3589 int64_t old_file_size, new_file_size; 3590 uint64_t nb_new_data_clusters, nb_new_l2_tables; 3591 3592 old_file_size = bdrv_getlength(bs->file->bs); 3593 if (old_file_size < 0) { 3594 error_setg_errno(errp, -old_file_size, 3595 "Failed to inquire current file length"); 3596 ret = old_file_size; 3597 goto fail; 3598 } 3599 old_file_size = ROUND_UP(old_file_size, s->cluster_size); 3600 3601 nb_new_data_clusters = DIV_ROUND_UP(offset - old_length, 3602 s->cluster_size); 3603 3604 /* This is an overestimation; we will not actually allocate space for 3605 * these in the file but just make sure the new refcount structures are 3606 * able to cover them so we will not have to allocate new refblocks 3607 * while entering the data blocks in the potentially new L2 tables. 3608 * (We do not actually care where the L2 tables are placed. Maybe they 3609 * are already allocated or they can be placed somewhere before 3610 * @old_file_size. It does not matter because they will be fully 3611 * allocated automatically, so they do not need to be covered by the 3612 * preallocation. All that matters is that we will not have to allocate 3613 * new refcount structures for them.) */ 3614 nb_new_l2_tables = DIV_ROUND_UP(nb_new_data_clusters, 3615 s->cluster_size / sizeof(uint64_t)); 3616 /* The cluster range may not be aligned to L2 boundaries, so add one L2 3617 * table for a potential head/tail */ 3618 nb_new_l2_tables++; 3619 3620 allocation_start = qcow2_refcount_area(bs, old_file_size, 3621 nb_new_data_clusters + 3622 nb_new_l2_tables, 3623 true, 0, 0); 3624 if (allocation_start < 0) { 3625 error_setg_errno(errp, -allocation_start, 3626 "Failed to resize refcount structures"); 3627 ret = allocation_start; 3628 goto fail; 3629 } 3630 3631 clusters_allocated = qcow2_alloc_clusters_at(bs, allocation_start, 3632 nb_new_data_clusters); 3633 if (clusters_allocated < 0) { 3634 error_setg_errno(errp, -clusters_allocated, 3635 "Failed to allocate data clusters"); 3636 ret = clusters_allocated; 3637 goto fail; 3638 } 3639 3640 assert(clusters_allocated == nb_new_data_clusters); 3641 3642 /* Allocate the data area */ 3643 new_file_size = allocation_start + 3644 nb_new_data_clusters * s->cluster_size; 3645 ret = bdrv_co_truncate(bs->file, new_file_size, prealloc, errp); 3646 if (ret < 0) { 3647 error_prepend(errp, "Failed to resize underlying file: "); 3648 qcow2_free_clusters(bs, allocation_start, 3649 nb_new_data_clusters * s->cluster_size, 3650 QCOW2_DISCARD_OTHER); 3651 goto fail; 3652 } 3653 3654 /* Create the necessary L2 entries */ 3655 host_offset = allocation_start; 3656 guest_offset = old_length; 3657 while (nb_new_data_clusters) { 3658 int64_t nb_clusters = MIN( 3659 nb_new_data_clusters, 3660 s->l2_slice_size - offset_to_l2_slice_index(s, guest_offset)); 3661 QCowL2Meta allocation = { 3662 .offset = guest_offset, 3663 .alloc_offset = host_offset, 3664 .nb_clusters = nb_clusters, 3665 }; 3666 qemu_co_queue_init(&allocation.dependent_requests); 3667 3668 ret = qcow2_alloc_cluster_link_l2(bs, &allocation); 3669 if (ret < 0) { 3670 error_setg_errno(errp, -ret, "Failed to update L2 tables"); 3671 qcow2_free_clusters(bs, host_offset, 3672 nb_new_data_clusters * s->cluster_size, 3673 QCOW2_DISCARD_OTHER); 3674 goto fail; 3675 } 3676 3677 guest_offset += nb_clusters * s->cluster_size; 3678 host_offset += nb_clusters * s->cluster_size; 3679 nb_new_data_clusters -= nb_clusters; 3680 } 3681 break; 3682 } 3683 3684 default: 3685 g_assert_not_reached(); 3686 } 3687 3688 if (prealloc != PREALLOC_MODE_OFF) { 3689 /* Flush metadata before actually changing the image size */ 3690 ret = qcow2_write_caches(bs); 3691 if (ret < 0) { 3692 error_setg_errno(errp, -ret, 3693 "Failed to flush the preallocated area to disk"); 3694 goto fail; 3695 } 3696 } 3697 3698 bs->total_sectors = offset / BDRV_SECTOR_SIZE; 3699 3700 /* write updated header.size */ 3701 offset = cpu_to_be64(offset); 3702 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, size), 3703 &offset, sizeof(uint64_t)); 3704 if (ret < 0) { 3705 error_setg_errno(errp, -ret, "Failed to update the image size"); 3706 goto fail; 3707 } 3708 3709 s->l1_vm_state_index = new_l1_size; 3710 3711 /* Update cache sizes */ 3712 options = qdict_clone_shallow(bs->options); 3713 ret = qcow2_update_options(bs, options, s->flags, errp); 3714 qobject_unref(options); 3715 if (ret < 0) { 3716 goto fail; 3717 } 3718 ret = 0; 3719 fail: 3720 qemu_co_mutex_unlock(&s->lock); 3721 return ret; 3722 } 3723 3724 /* 3725 * qcow2_compress() 3726 * 3727 * @dest - destination buffer, @dest_size bytes 3728 * @src - source buffer, @src_size bytes 3729 * 3730 * Returns: compressed size on success 3731 * -1 destination buffer is not enough to store compressed data 3732 * -2 on any other error 3733 */ 3734 static ssize_t qcow2_compress(void *dest, size_t dest_size, 3735 const void *src, size_t src_size) 3736 { 3737 ssize_t ret; 3738 z_stream strm; 3739 3740 /* best compression, small window, no zlib header */ 3741 memset(&strm, 0, sizeof(strm)); 3742 ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION, Z_DEFLATED, 3743 -12, 9, Z_DEFAULT_STRATEGY); 3744 if (ret != Z_OK) { 3745 return -2; 3746 } 3747 3748 /* strm.next_in is not const in old zlib versions, such as those used on 3749 * OpenBSD/NetBSD, so cast the const away */ 3750 strm.avail_in = src_size; 3751 strm.next_in = (void *) src; 3752 strm.avail_out = dest_size; 3753 strm.next_out = dest; 3754 3755 ret = deflate(&strm, Z_FINISH); 3756 if (ret == Z_STREAM_END) { 3757 ret = dest_size - strm.avail_out; 3758 } else { 3759 ret = (ret == Z_OK ? -1 : -2); 3760 } 3761 3762 deflateEnd(&strm); 3763 3764 return ret; 3765 } 3766 3767 /* 3768 * qcow2_decompress() 3769 * 3770 * Decompress some data (not more than @src_size bytes) to produce exactly 3771 * @dest_size bytes. 3772 * 3773 * @dest - destination buffer, @dest_size bytes 3774 * @src - source buffer, @src_size bytes 3775 * 3776 * Returns: 0 on success 3777 * -1 on fail 3778 */ 3779 static ssize_t qcow2_decompress(void *dest, size_t dest_size, 3780 const void *src, size_t src_size) 3781 { 3782 int ret = 0; 3783 z_stream strm; 3784 3785 memset(&strm, 0, sizeof(strm)); 3786 strm.avail_in = src_size; 3787 strm.next_in = (void *) src; 3788 strm.avail_out = dest_size; 3789 strm.next_out = dest; 3790 3791 ret = inflateInit2(&strm, -12); 3792 if (ret != Z_OK) { 3793 return -1; 3794 } 3795 3796 ret = inflate(&strm, Z_FINISH); 3797 if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) || strm.avail_out != 0) { 3798 /* We approve Z_BUF_ERROR because we need @dest buffer to be filled, but 3799 * @src buffer may be processed partly (because in qcow2 we know size of 3800 * compressed data with precision of one sector) */ 3801 ret = -1; 3802 } 3803 3804 inflateEnd(&strm); 3805 3806 return ret; 3807 } 3808 3809 #define MAX_COMPRESS_THREADS 4 3810 3811 typedef ssize_t (*Qcow2CompressFunc)(void *dest, size_t dest_size, 3812 const void *src, size_t src_size); 3813 typedef struct Qcow2CompressData { 3814 void *dest; 3815 size_t dest_size; 3816 const void *src; 3817 size_t src_size; 3818 ssize_t ret; 3819 3820 Qcow2CompressFunc func; 3821 } Qcow2CompressData; 3822 3823 static int qcow2_compress_pool_func(void *opaque) 3824 { 3825 Qcow2CompressData *data = opaque; 3826 3827 data->ret = data->func(data->dest, data->dest_size, 3828 data->src, data->src_size); 3829 3830 return 0; 3831 } 3832 3833 static void qcow2_compress_complete(void *opaque, int ret) 3834 { 3835 qemu_coroutine_enter(opaque); 3836 } 3837 3838 static ssize_t coroutine_fn 3839 qcow2_co_do_compress(BlockDriverState *bs, void *dest, size_t dest_size, 3840 const void *src, size_t src_size, Qcow2CompressFunc func) 3841 { 3842 BDRVQcow2State *s = bs->opaque; 3843 BlockAIOCB *acb; 3844 ThreadPool *pool = aio_get_thread_pool(bdrv_get_aio_context(bs)); 3845 Qcow2CompressData arg = { 3846 .dest = dest, 3847 .dest_size = dest_size, 3848 .src = src, 3849 .src_size = src_size, 3850 .func = func, 3851 }; 3852 3853 while (s->nb_compress_threads >= MAX_COMPRESS_THREADS) { 3854 qemu_co_queue_wait(&s->compress_wait_queue, NULL); 3855 } 3856 3857 s->nb_compress_threads++; 3858 acb = thread_pool_submit_aio(pool, qcow2_compress_pool_func, &arg, 3859 qcow2_compress_complete, 3860 qemu_coroutine_self()); 3861 3862 if (!acb) { 3863 s->nb_compress_threads--; 3864 return -EINVAL; 3865 } 3866 qemu_coroutine_yield(); 3867 s->nb_compress_threads--; 3868 qemu_co_queue_next(&s->compress_wait_queue); 3869 3870 return arg.ret; 3871 } 3872 3873 static ssize_t coroutine_fn 3874 qcow2_co_compress(BlockDriverState *bs, void *dest, size_t dest_size, 3875 const void *src, size_t src_size) 3876 { 3877 return qcow2_co_do_compress(bs, dest, dest_size, src, src_size, 3878 qcow2_compress); 3879 } 3880 3881 static ssize_t coroutine_fn 3882 qcow2_co_decompress(BlockDriverState *bs, void *dest, size_t dest_size, 3883 const void *src, size_t src_size) 3884 { 3885 return qcow2_co_do_compress(bs, dest, dest_size, src, src_size, 3886 qcow2_decompress); 3887 } 3888 3889 /* XXX: put compressed sectors first, then all the cluster aligned 3890 tables to avoid losing bytes in alignment */ 3891 static coroutine_fn int 3892 qcow2_co_pwritev_compressed(BlockDriverState *bs, uint64_t offset, 3893 uint64_t bytes, QEMUIOVector *qiov) 3894 { 3895 BDRVQcow2State *s = bs->opaque; 3896 QEMUIOVector hd_qiov; 3897 struct iovec iov; 3898 int ret; 3899 size_t out_len; 3900 uint8_t *buf, *out_buf; 3901 int64_t cluster_offset; 3902 3903 if (bytes == 0) { 3904 /* align end of file to a sector boundary to ease reading with 3905 sector based I/Os */ 3906 cluster_offset = bdrv_getlength(bs->file->bs); 3907 if (cluster_offset < 0) { 3908 return cluster_offset; 3909 } 3910 return bdrv_co_truncate(bs->file, cluster_offset, PREALLOC_MODE_OFF, 3911 NULL); 3912 } 3913 3914 if (offset_into_cluster(s, offset)) { 3915 return -EINVAL; 3916 } 3917 3918 buf = qemu_blockalign(bs, s->cluster_size); 3919 if (bytes != s->cluster_size) { 3920 if (bytes > s->cluster_size || 3921 offset + bytes != bs->total_sectors << BDRV_SECTOR_BITS) 3922 { 3923 qemu_vfree(buf); 3924 return -EINVAL; 3925 } 3926 /* Zero-pad last write if image size is not cluster aligned */ 3927 memset(buf + bytes, 0, s->cluster_size - bytes); 3928 } 3929 qemu_iovec_to_buf(qiov, 0, buf, bytes); 3930 3931 out_buf = g_malloc(s->cluster_size); 3932 3933 out_len = qcow2_co_compress(bs, out_buf, s->cluster_size - 1, 3934 buf, s->cluster_size); 3935 if (out_len == -2) { 3936 ret = -EINVAL; 3937 goto fail; 3938 } else if (out_len == -1) { 3939 /* could not compress: write normal cluster */ 3940 ret = qcow2_co_pwritev(bs, offset, bytes, qiov, 0); 3941 if (ret < 0) { 3942 goto fail; 3943 } 3944 goto success; 3945 } 3946 3947 qemu_co_mutex_lock(&s->lock); 3948 cluster_offset = 3949 qcow2_alloc_compressed_cluster_offset(bs, offset, out_len); 3950 if (!cluster_offset) { 3951 qemu_co_mutex_unlock(&s->lock); 3952 ret = -EIO; 3953 goto fail; 3954 } 3955 cluster_offset &= s->cluster_offset_mask; 3956 3957 ret = qcow2_pre_write_overlap_check(bs, 0, cluster_offset, out_len); 3958 qemu_co_mutex_unlock(&s->lock); 3959 if (ret < 0) { 3960 goto fail; 3961 } 3962 3963 iov = (struct iovec) { 3964 .iov_base = out_buf, 3965 .iov_len = out_len, 3966 }; 3967 qemu_iovec_init_external(&hd_qiov, &iov, 1); 3968 3969 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_COMPRESSED); 3970 ret = bdrv_co_pwritev(bs->file, cluster_offset, out_len, &hd_qiov, 0); 3971 if (ret < 0) { 3972 goto fail; 3973 } 3974 success: 3975 ret = 0; 3976 fail: 3977 qemu_vfree(buf); 3978 g_free(out_buf); 3979 return ret; 3980 } 3981 3982 static int coroutine_fn 3983 qcow2_co_preadv_compressed(BlockDriverState *bs, 3984 uint64_t file_cluster_offset, 3985 uint64_t offset, 3986 uint64_t bytes, 3987 QEMUIOVector *qiov) 3988 { 3989 BDRVQcow2State *s = bs->opaque; 3990 int ret = 0, csize, nb_csectors; 3991 uint64_t coffset; 3992 uint8_t *buf, *out_buf; 3993 struct iovec iov; 3994 QEMUIOVector local_qiov; 3995 int offset_in_cluster = offset_into_cluster(s, offset); 3996 3997 coffset = file_cluster_offset & s->cluster_offset_mask; 3998 nb_csectors = ((file_cluster_offset >> s->csize_shift) & s->csize_mask) + 1; 3999 csize = nb_csectors * 512 - (coffset & 511); 4000 4001 buf = g_try_malloc(csize); 4002 if (!buf) { 4003 return -ENOMEM; 4004 } 4005 iov.iov_base = buf; 4006 iov.iov_len = csize; 4007 qemu_iovec_init_external(&local_qiov, &iov, 1); 4008 4009 out_buf = qemu_blockalign(bs, s->cluster_size); 4010 4011 BLKDBG_EVENT(bs->file, BLKDBG_READ_COMPRESSED); 4012 ret = bdrv_co_preadv(bs->file, coffset, csize, &local_qiov, 0); 4013 if (ret < 0) { 4014 goto fail; 4015 } 4016 4017 if (qcow2_co_decompress(bs, out_buf, s->cluster_size, buf, csize) < 0) { 4018 ret = -EIO; 4019 goto fail; 4020 } 4021 4022 qemu_iovec_from_buf(qiov, 0, out_buf + offset_in_cluster, bytes); 4023 4024 fail: 4025 qemu_vfree(out_buf); 4026 g_free(buf); 4027 4028 return ret; 4029 } 4030 4031 static int make_completely_empty(BlockDriverState *bs) 4032 { 4033 BDRVQcow2State *s = bs->opaque; 4034 Error *local_err = NULL; 4035 int ret, l1_clusters; 4036 int64_t offset; 4037 uint64_t *new_reftable = NULL; 4038 uint64_t rt_entry, l1_size2; 4039 struct { 4040 uint64_t l1_offset; 4041 uint64_t reftable_offset; 4042 uint32_t reftable_clusters; 4043 } QEMU_PACKED l1_ofs_rt_ofs_cls; 4044 4045 ret = qcow2_cache_empty(bs, s->l2_table_cache); 4046 if (ret < 0) { 4047 goto fail; 4048 } 4049 4050 ret = qcow2_cache_empty(bs, s->refcount_block_cache); 4051 if (ret < 0) { 4052 goto fail; 4053 } 4054 4055 /* Refcounts will be broken utterly */ 4056 ret = qcow2_mark_dirty(bs); 4057 if (ret < 0) { 4058 goto fail; 4059 } 4060 4061 BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE); 4062 4063 l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / sizeof(uint64_t)); 4064 l1_size2 = (uint64_t)s->l1_size * sizeof(uint64_t); 4065 4066 /* After this call, neither the in-memory nor the on-disk refcount 4067 * information accurately describe the actual references */ 4068 4069 ret = bdrv_pwrite_zeroes(bs->file, s->l1_table_offset, 4070 l1_clusters * s->cluster_size, 0); 4071 if (ret < 0) { 4072 goto fail_broken_refcounts; 4073 } 4074 memset(s->l1_table, 0, l1_size2); 4075 4076 BLKDBG_EVENT(bs->file, BLKDBG_EMPTY_IMAGE_PREPARE); 4077 4078 /* Overwrite enough clusters at the beginning of the sectors to place 4079 * the refcount table, a refcount block and the L1 table in; this may 4080 * overwrite parts of the existing refcount and L1 table, which is not 4081 * an issue because the dirty flag is set, complete data loss is in fact 4082 * desired and partial data loss is consequently fine as well */ 4083 ret = bdrv_pwrite_zeroes(bs->file, s->cluster_size, 4084 (2 + l1_clusters) * s->cluster_size, 0); 4085 /* This call (even if it failed overall) may have overwritten on-disk 4086 * refcount structures; in that case, the in-memory refcount information 4087 * will probably differ from the on-disk information which makes the BDS 4088 * unusable */ 4089 if (ret < 0) { 4090 goto fail_broken_refcounts; 4091 } 4092 4093 BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE); 4094 BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_UPDATE); 4095 4096 /* "Create" an empty reftable (one cluster) directly after the image 4097 * header and an empty L1 table three clusters after the image header; 4098 * the cluster between those two will be used as the first refblock */ 4099 l1_ofs_rt_ofs_cls.l1_offset = cpu_to_be64(3 * s->cluster_size); 4100 l1_ofs_rt_ofs_cls.reftable_offset = cpu_to_be64(s->cluster_size); 4101 l1_ofs_rt_ofs_cls.reftable_clusters = cpu_to_be32(1); 4102 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, l1_table_offset), 4103 &l1_ofs_rt_ofs_cls, sizeof(l1_ofs_rt_ofs_cls)); 4104 if (ret < 0) { 4105 goto fail_broken_refcounts; 4106 } 4107 4108 s->l1_table_offset = 3 * s->cluster_size; 4109 4110 new_reftable = g_try_new0(uint64_t, s->cluster_size / sizeof(uint64_t)); 4111 if (!new_reftable) { 4112 ret = -ENOMEM; 4113 goto fail_broken_refcounts; 4114 } 4115 4116 s->refcount_table_offset = s->cluster_size; 4117 s->refcount_table_size = s->cluster_size / sizeof(uint64_t); 4118 s->max_refcount_table_index = 0; 4119 4120 g_free(s->refcount_table); 4121 s->refcount_table = new_reftable; 4122 new_reftable = NULL; 4123 4124 /* Now the in-memory refcount information again corresponds to the on-disk 4125 * information (reftable is empty and no refblocks (the refblock cache is 4126 * empty)); however, this means some clusters (e.g. the image header) are 4127 * referenced, but not refcounted, but the normal qcow2 code assumes that 4128 * the in-memory information is always correct */ 4129 4130 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC); 4131 4132 /* Enter the first refblock into the reftable */ 4133 rt_entry = cpu_to_be64(2 * s->cluster_size); 4134 ret = bdrv_pwrite_sync(bs->file, s->cluster_size, 4135 &rt_entry, sizeof(rt_entry)); 4136 if (ret < 0) { 4137 goto fail_broken_refcounts; 4138 } 4139 s->refcount_table[0] = 2 * s->cluster_size; 4140 4141 s->free_cluster_index = 0; 4142 assert(3 + l1_clusters <= s->refcount_block_size); 4143 offset = qcow2_alloc_clusters(bs, 3 * s->cluster_size + l1_size2); 4144 if (offset < 0) { 4145 ret = offset; 4146 goto fail_broken_refcounts; 4147 } else if (offset > 0) { 4148 error_report("First cluster in emptied image is in use"); 4149 abort(); 4150 } 4151 4152 /* Now finally the in-memory information corresponds to the on-disk 4153 * structures and is correct */ 4154 ret = qcow2_mark_clean(bs); 4155 if (ret < 0) { 4156 goto fail; 4157 } 4158 4159 ret = bdrv_truncate(bs->file, (3 + l1_clusters) * s->cluster_size, 4160 PREALLOC_MODE_OFF, &local_err); 4161 if (ret < 0) { 4162 error_report_err(local_err); 4163 goto fail; 4164 } 4165 4166 return 0; 4167 4168 fail_broken_refcounts: 4169 /* The BDS is unusable at this point. If we wanted to make it usable, we 4170 * would have to call qcow2_refcount_close(), qcow2_refcount_init(), 4171 * qcow2_check_refcounts(), qcow2_refcount_close() and qcow2_refcount_init() 4172 * again. However, because the functions which could have caused this error 4173 * path to be taken are used by those functions as well, it's very likely 4174 * that that sequence will fail as well. Therefore, just eject the BDS. */ 4175 bs->drv = NULL; 4176 4177 fail: 4178 g_free(new_reftable); 4179 return ret; 4180 } 4181 4182 static int qcow2_make_empty(BlockDriverState *bs) 4183 { 4184 BDRVQcow2State *s = bs->opaque; 4185 uint64_t offset, end_offset; 4186 int step = QEMU_ALIGN_DOWN(INT_MAX, s->cluster_size); 4187 int l1_clusters, ret = 0; 4188 4189 l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / sizeof(uint64_t)); 4190 4191 if (s->qcow_version >= 3 && !s->snapshots && !s->nb_bitmaps && 4192 3 + l1_clusters <= s->refcount_block_size && 4193 s->crypt_method_header != QCOW_CRYPT_LUKS) { 4194 /* The following function only works for qcow2 v3 images (it 4195 * requires the dirty flag) and only as long as there are no 4196 * features that reserve extra clusters (such as snapshots, 4197 * LUKS header, or persistent bitmaps), because it completely 4198 * empties the image. Furthermore, the L1 table and three 4199 * additional clusters (image header, refcount table, one 4200 * refcount block) have to fit inside one refcount block. */ 4201 return make_completely_empty(bs); 4202 } 4203 4204 /* This fallback code simply discards every active cluster; this is slow, 4205 * but works in all cases */ 4206 end_offset = bs->total_sectors * BDRV_SECTOR_SIZE; 4207 for (offset = 0; offset < end_offset; offset += step) { 4208 /* As this function is generally used after committing an external 4209 * snapshot, QCOW2_DISCARD_SNAPSHOT seems appropriate. Also, the 4210 * default action for this kind of discard is to pass the discard, 4211 * which will ideally result in an actually smaller image file, as 4212 * is probably desired. */ 4213 ret = qcow2_cluster_discard(bs, offset, MIN(step, end_offset - offset), 4214 QCOW2_DISCARD_SNAPSHOT, true); 4215 if (ret < 0) { 4216 break; 4217 } 4218 } 4219 4220 return ret; 4221 } 4222 4223 static coroutine_fn int qcow2_co_flush_to_os(BlockDriverState *bs) 4224 { 4225 BDRVQcow2State *s = bs->opaque; 4226 int ret; 4227 4228 qemu_co_mutex_lock(&s->lock); 4229 ret = qcow2_write_caches(bs); 4230 qemu_co_mutex_unlock(&s->lock); 4231 4232 return ret; 4233 } 4234 4235 static BlockMeasureInfo *qcow2_measure(QemuOpts *opts, BlockDriverState *in_bs, 4236 Error **errp) 4237 { 4238 Error *local_err = NULL; 4239 BlockMeasureInfo *info; 4240 uint64_t required = 0; /* bytes that contribute to required size */ 4241 uint64_t virtual_size; /* disk size as seen by guest */ 4242 uint64_t refcount_bits; 4243 uint64_t l2_tables; 4244 size_t cluster_size; 4245 int version; 4246 char *optstr; 4247 PreallocMode prealloc; 4248 bool has_backing_file; 4249 4250 /* Parse image creation options */ 4251 cluster_size = qcow2_opt_get_cluster_size_del(opts, &local_err); 4252 if (local_err) { 4253 goto err; 4254 } 4255 4256 version = qcow2_opt_get_version_del(opts, &local_err); 4257 if (local_err) { 4258 goto err; 4259 } 4260 4261 refcount_bits = qcow2_opt_get_refcount_bits_del(opts, version, &local_err); 4262 if (local_err) { 4263 goto err; 4264 } 4265 4266 optstr = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC); 4267 prealloc = qapi_enum_parse(&PreallocMode_lookup, optstr, 4268 PREALLOC_MODE_OFF, &local_err); 4269 g_free(optstr); 4270 if (local_err) { 4271 goto err; 4272 } 4273 4274 optstr = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE); 4275 has_backing_file = !!optstr; 4276 g_free(optstr); 4277 4278 virtual_size = qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0); 4279 virtual_size = ROUND_UP(virtual_size, cluster_size); 4280 4281 /* Check that virtual disk size is valid */ 4282 l2_tables = DIV_ROUND_UP(virtual_size / cluster_size, 4283 cluster_size / sizeof(uint64_t)); 4284 if (l2_tables * sizeof(uint64_t) > QCOW_MAX_L1_SIZE) { 4285 error_setg(&local_err, "The image size is too large " 4286 "(try using a larger cluster size)"); 4287 goto err; 4288 } 4289 4290 /* Account for input image */ 4291 if (in_bs) { 4292 int64_t ssize = bdrv_getlength(in_bs); 4293 if (ssize < 0) { 4294 error_setg_errno(&local_err, -ssize, 4295 "Unable to get image virtual_size"); 4296 goto err; 4297 } 4298 4299 virtual_size = ROUND_UP(ssize, cluster_size); 4300 4301 if (has_backing_file) { 4302 /* We don't how much of the backing chain is shared by the input 4303 * image and the new image file. In the worst case the new image's 4304 * backing file has nothing in common with the input image. Be 4305 * conservative and assume all clusters need to be written. 4306 */ 4307 required = virtual_size; 4308 } else { 4309 int64_t offset; 4310 int64_t pnum = 0; 4311 4312 for (offset = 0; offset < ssize; offset += pnum) { 4313 int ret; 4314 4315 ret = bdrv_block_status_above(in_bs, NULL, offset, 4316 ssize - offset, &pnum, NULL, 4317 NULL); 4318 if (ret < 0) { 4319 error_setg_errno(&local_err, -ret, 4320 "Unable to get block status"); 4321 goto err; 4322 } 4323 4324 if (ret & BDRV_BLOCK_ZERO) { 4325 /* Skip zero regions (safe with no backing file) */ 4326 } else if ((ret & (BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED)) == 4327 (BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED)) { 4328 /* Extend pnum to end of cluster for next iteration */ 4329 pnum = ROUND_UP(offset + pnum, cluster_size) - offset; 4330 4331 /* Count clusters we've seen */ 4332 required += offset % cluster_size + pnum; 4333 } 4334 } 4335 } 4336 } 4337 4338 /* Take into account preallocation. Nothing special is needed for 4339 * PREALLOC_MODE_METADATA since metadata is always counted. 4340 */ 4341 if (prealloc == PREALLOC_MODE_FULL || prealloc == PREALLOC_MODE_FALLOC) { 4342 required = virtual_size; 4343 } 4344 4345 info = g_new(BlockMeasureInfo, 1); 4346 info->fully_allocated = 4347 qcow2_calc_prealloc_size(virtual_size, cluster_size, 4348 ctz32(refcount_bits)); 4349 4350 /* Remove data clusters that are not required. This overestimates the 4351 * required size because metadata needed for the fully allocated file is 4352 * still counted. 4353 */ 4354 info->required = info->fully_allocated - virtual_size + required; 4355 return info; 4356 4357 err: 4358 error_propagate(errp, local_err); 4359 return NULL; 4360 } 4361 4362 static int qcow2_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) 4363 { 4364 BDRVQcow2State *s = bs->opaque; 4365 bdi->unallocated_blocks_are_zero = true; 4366 bdi->cluster_size = s->cluster_size; 4367 bdi->vm_state_offset = qcow2_vm_state_offset(s); 4368 return 0; 4369 } 4370 4371 static ImageInfoSpecific *qcow2_get_specific_info(BlockDriverState *bs) 4372 { 4373 BDRVQcow2State *s = bs->opaque; 4374 ImageInfoSpecific *spec_info; 4375 QCryptoBlockInfo *encrypt_info = NULL; 4376 4377 if (s->crypto != NULL) { 4378 encrypt_info = qcrypto_block_get_info(s->crypto, &error_abort); 4379 } 4380 4381 spec_info = g_new(ImageInfoSpecific, 1); 4382 *spec_info = (ImageInfoSpecific){ 4383 .type = IMAGE_INFO_SPECIFIC_KIND_QCOW2, 4384 .u.qcow2.data = g_new(ImageInfoSpecificQCow2, 1), 4385 }; 4386 if (s->qcow_version == 2) { 4387 *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){ 4388 .compat = g_strdup("0.10"), 4389 .refcount_bits = s->refcount_bits, 4390 }; 4391 } else if (s->qcow_version == 3) { 4392 *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){ 4393 .compat = g_strdup("1.1"), 4394 .lazy_refcounts = s->compatible_features & 4395 QCOW2_COMPAT_LAZY_REFCOUNTS, 4396 .has_lazy_refcounts = true, 4397 .corrupt = s->incompatible_features & 4398 QCOW2_INCOMPAT_CORRUPT, 4399 .has_corrupt = true, 4400 .refcount_bits = s->refcount_bits, 4401 }; 4402 } else { 4403 /* if this assertion fails, this probably means a new version was 4404 * added without having it covered here */ 4405 assert(false); 4406 } 4407 4408 if (encrypt_info) { 4409 ImageInfoSpecificQCow2Encryption *qencrypt = 4410 g_new(ImageInfoSpecificQCow2Encryption, 1); 4411 switch (encrypt_info->format) { 4412 case Q_CRYPTO_BLOCK_FORMAT_QCOW: 4413 qencrypt->format = BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_AES; 4414 break; 4415 case Q_CRYPTO_BLOCK_FORMAT_LUKS: 4416 qencrypt->format = BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_LUKS; 4417 qencrypt->u.luks = encrypt_info->u.luks; 4418 break; 4419 default: 4420 abort(); 4421 } 4422 /* Since we did shallow copy above, erase any pointers 4423 * in the original info */ 4424 memset(&encrypt_info->u, 0, sizeof(encrypt_info->u)); 4425 qapi_free_QCryptoBlockInfo(encrypt_info); 4426 4427 spec_info->u.qcow2.data->has_encrypt = true; 4428 spec_info->u.qcow2.data->encrypt = qencrypt; 4429 } 4430 4431 return spec_info; 4432 } 4433 4434 static int qcow2_save_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, 4435 int64_t pos) 4436 { 4437 BDRVQcow2State *s = bs->opaque; 4438 4439 BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_SAVE); 4440 return bs->drv->bdrv_co_pwritev(bs, qcow2_vm_state_offset(s) + pos, 4441 qiov->size, qiov, 0); 4442 } 4443 4444 static int qcow2_load_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, 4445 int64_t pos) 4446 { 4447 BDRVQcow2State *s = bs->opaque; 4448 4449 BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_LOAD); 4450 return bs->drv->bdrv_co_preadv(bs, qcow2_vm_state_offset(s) + pos, 4451 qiov->size, qiov, 0); 4452 } 4453 4454 /* 4455 * Downgrades an image's version. To achieve this, any incompatible features 4456 * have to be removed. 4457 */ 4458 static int qcow2_downgrade(BlockDriverState *bs, int target_version, 4459 BlockDriverAmendStatusCB *status_cb, void *cb_opaque, 4460 Error **errp) 4461 { 4462 BDRVQcow2State *s = bs->opaque; 4463 int current_version = s->qcow_version; 4464 int ret; 4465 4466 /* This is qcow2_downgrade(), not qcow2_upgrade() */ 4467 assert(target_version < current_version); 4468 4469 /* There are no other versions (now) that you can downgrade to */ 4470 assert(target_version == 2); 4471 4472 if (s->refcount_order != 4) { 4473 error_setg(errp, "compat=0.10 requires refcount_bits=16"); 4474 return -ENOTSUP; 4475 } 4476 4477 /* clear incompatible features */ 4478 if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) { 4479 ret = qcow2_mark_clean(bs); 4480 if (ret < 0) { 4481 error_setg_errno(errp, -ret, "Failed to make the image clean"); 4482 return ret; 4483 } 4484 } 4485 4486 /* with QCOW2_INCOMPAT_CORRUPT, it is pretty much impossible to get here in 4487 * the first place; if that happens nonetheless, returning -ENOTSUP is the 4488 * best thing to do anyway */ 4489 4490 if (s->incompatible_features) { 4491 error_setg(errp, "Cannot downgrade an image with incompatible features " 4492 "%#" PRIx64 " set", s->incompatible_features); 4493 return -ENOTSUP; 4494 } 4495 4496 /* since we can ignore compatible features, we can set them to 0 as well */ 4497 s->compatible_features = 0; 4498 /* if lazy refcounts have been used, they have already been fixed through 4499 * clearing the dirty flag */ 4500 4501 /* clearing autoclear features is trivial */ 4502 s->autoclear_features = 0; 4503 4504 ret = qcow2_expand_zero_clusters(bs, status_cb, cb_opaque); 4505 if (ret < 0) { 4506 error_setg_errno(errp, -ret, "Failed to turn zero into data clusters"); 4507 return ret; 4508 } 4509 4510 s->qcow_version = target_version; 4511 ret = qcow2_update_header(bs); 4512 if (ret < 0) { 4513 s->qcow_version = current_version; 4514 error_setg_errno(errp, -ret, "Failed to update the image header"); 4515 return ret; 4516 } 4517 return 0; 4518 } 4519 4520 typedef enum Qcow2AmendOperation { 4521 /* This is the value Qcow2AmendHelperCBInfo::last_operation will be 4522 * statically initialized to so that the helper CB can discern the first 4523 * invocation from an operation change */ 4524 QCOW2_NO_OPERATION = 0, 4525 4526 QCOW2_CHANGING_REFCOUNT_ORDER, 4527 QCOW2_DOWNGRADING, 4528 } Qcow2AmendOperation; 4529 4530 typedef struct Qcow2AmendHelperCBInfo { 4531 /* The code coordinating the amend operations should only modify 4532 * these four fields; the rest will be managed by the CB */ 4533 BlockDriverAmendStatusCB *original_status_cb; 4534 void *original_cb_opaque; 4535 4536 Qcow2AmendOperation current_operation; 4537 4538 /* Total number of operations to perform (only set once) */ 4539 int total_operations; 4540 4541 /* The following fields are managed by the CB */ 4542 4543 /* Number of operations completed */ 4544 int operations_completed; 4545 4546 /* Cumulative offset of all completed operations */ 4547 int64_t offset_completed; 4548 4549 Qcow2AmendOperation last_operation; 4550 int64_t last_work_size; 4551 } Qcow2AmendHelperCBInfo; 4552 4553 static void qcow2_amend_helper_cb(BlockDriverState *bs, 4554 int64_t operation_offset, 4555 int64_t operation_work_size, void *opaque) 4556 { 4557 Qcow2AmendHelperCBInfo *info = opaque; 4558 int64_t current_work_size; 4559 int64_t projected_work_size; 4560 4561 if (info->current_operation != info->last_operation) { 4562 if (info->last_operation != QCOW2_NO_OPERATION) { 4563 info->offset_completed += info->last_work_size; 4564 info->operations_completed++; 4565 } 4566 4567 info->last_operation = info->current_operation; 4568 } 4569 4570 assert(info->total_operations > 0); 4571 assert(info->operations_completed < info->total_operations); 4572 4573 info->last_work_size = operation_work_size; 4574 4575 current_work_size = info->offset_completed + operation_work_size; 4576 4577 /* current_work_size is the total work size for (operations_completed + 1) 4578 * operations (which includes this one), so multiply it by the number of 4579 * operations not covered and divide it by the number of operations 4580 * covered to get a projection for the operations not covered */ 4581 projected_work_size = current_work_size * (info->total_operations - 4582 info->operations_completed - 1) 4583 / (info->operations_completed + 1); 4584 4585 info->original_status_cb(bs, info->offset_completed + operation_offset, 4586 current_work_size + projected_work_size, 4587 info->original_cb_opaque); 4588 } 4589 4590 static int qcow2_amend_options(BlockDriverState *bs, QemuOpts *opts, 4591 BlockDriverAmendStatusCB *status_cb, 4592 void *cb_opaque, 4593 Error **errp) 4594 { 4595 BDRVQcow2State *s = bs->opaque; 4596 int old_version = s->qcow_version, new_version = old_version; 4597 uint64_t new_size = 0; 4598 const char *backing_file = NULL, *backing_format = NULL; 4599 bool lazy_refcounts = s->use_lazy_refcounts; 4600 const char *compat = NULL; 4601 uint64_t cluster_size = s->cluster_size; 4602 bool encrypt; 4603 int encformat; 4604 int refcount_bits = s->refcount_bits; 4605 int ret; 4606 QemuOptDesc *desc = opts->list->desc; 4607 Qcow2AmendHelperCBInfo helper_cb_info; 4608 4609 while (desc && desc->name) { 4610 if (!qemu_opt_find(opts, desc->name)) { 4611 /* only change explicitly defined options */ 4612 desc++; 4613 continue; 4614 } 4615 4616 if (!strcmp(desc->name, BLOCK_OPT_COMPAT_LEVEL)) { 4617 compat = qemu_opt_get(opts, BLOCK_OPT_COMPAT_LEVEL); 4618 if (!compat) { 4619 /* preserve default */ 4620 } else if (!strcmp(compat, "0.10")) { 4621 new_version = 2; 4622 } else if (!strcmp(compat, "1.1")) { 4623 new_version = 3; 4624 } else { 4625 error_setg(errp, "Unknown compatibility level %s", compat); 4626 return -EINVAL; 4627 } 4628 } else if (!strcmp(desc->name, BLOCK_OPT_PREALLOC)) { 4629 error_setg(errp, "Cannot change preallocation mode"); 4630 return -ENOTSUP; 4631 } else if (!strcmp(desc->name, BLOCK_OPT_SIZE)) { 4632 new_size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, 0); 4633 } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FILE)) { 4634 backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE); 4635 } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FMT)) { 4636 backing_format = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT); 4637 } else if (!strcmp(desc->name, BLOCK_OPT_ENCRYPT)) { 4638 encrypt = qemu_opt_get_bool(opts, BLOCK_OPT_ENCRYPT, 4639 !!s->crypto); 4640 4641 if (encrypt != !!s->crypto) { 4642 error_setg(errp, 4643 "Changing the encryption flag is not supported"); 4644 return -ENOTSUP; 4645 } 4646 } else if (!strcmp(desc->name, BLOCK_OPT_ENCRYPT_FORMAT)) { 4647 encformat = qcow2_crypt_method_from_format( 4648 qemu_opt_get(opts, BLOCK_OPT_ENCRYPT_FORMAT)); 4649 4650 if (encformat != s->crypt_method_header) { 4651 error_setg(errp, 4652 "Changing the encryption format is not supported"); 4653 return -ENOTSUP; 4654 } 4655 } else if (g_str_has_prefix(desc->name, "encrypt.")) { 4656 error_setg(errp, 4657 "Changing the encryption parameters is not supported"); 4658 return -ENOTSUP; 4659 } else if (!strcmp(desc->name, BLOCK_OPT_CLUSTER_SIZE)) { 4660 cluster_size = qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE, 4661 cluster_size); 4662 if (cluster_size != s->cluster_size) { 4663 error_setg(errp, "Changing the cluster size is not supported"); 4664 return -ENOTSUP; 4665 } 4666 } else if (!strcmp(desc->name, BLOCK_OPT_LAZY_REFCOUNTS)) { 4667 lazy_refcounts = qemu_opt_get_bool(opts, BLOCK_OPT_LAZY_REFCOUNTS, 4668 lazy_refcounts); 4669 } else if (!strcmp(desc->name, BLOCK_OPT_REFCOUNT_BITS)) { 4670 refcount_bits = qemu_opt_get_number(opts, BLOCK_OPT_REFCOUNT_BITS, 4671 refcount_bits); 4672 4673 if (refcount_bits <= 0 || refcount_bits > 64 || 4674 !is_power_of_2(refcount_bits)) 4675 { 4676 error_setg(errp, "Refcount width must be a power of two and " 4677 "may not exceed 64 bits"); 4678 return -EINVAL; 4679 } 4680 } else { 4681 /* if this point is reached, this probably means a new option was 4682 * added without having it covered here */ 4683 abort(); 4684 } 4685 4686 desc++; 4687 } 4688 4689 helper_cb_info = (Qcow2AmendHelperCBInfo){ 4690 .original_status_cb = status_cb, 4691 .original_cb_opaque = cb_opaque, 4692 .total_operations = (new_version < old_version) 4693 + (s->refcount_bits != refcount_bits) 4694 }; 4695 4696 /* Upgrade first (some features may require compat=1.1) */ 4697 if (new_version > old_version) { 4698 s->qcow_version = new_version; 4699 ret = qcow2_update_header(bs); 4700 if (ret < 0) { 4701 s->qcow_version = old_version; 4702 error_setg_errno(errp, -ret, "Failed to update the image header"); 4703 return ret; 4704 } 4705 } 4706 4707 if (s->refcount_bits != refcount_bits) { 4708 int refcount_order = ctz32(refcount_bits); 4709 4710 if (new_version < 3 && refcount_bits != 16) { 4711 error_setg(errp, "Refcount widths other than 16 bits require " 4712 "compatibility level 1.1 or above (use compat=1.1 or " 4713 "greater)"); 4714 return -EINVAL; 4715 } 4716 4717 helper_cb_info.current_operation = QCOW2_CHANGING_REFCOUNT_ORDER; 4718 ret = qcow2_change_refcount_order(bs, refcount_order, 4719 &qcow2_amend_helper_cb, 4720 &helper_cb_info, errp); 4721 if (ret < 0) { 4722 return ret; 4723 } 4724 } 4725 4726 if (backing_file || backing_format) { 4727 ret = qcow2_change_backing_file(bs, 4728 backing_file ?: s->image_backing_file, 4729 backing_format ?: s->image_backing_format); 4730 if (ret < 0) { 4731 error_setg_errno(errp, -ret, "Failed to change the backing file"); 4732 return ret; 4733 } 4734 } 4735 4736 if (s->use_lazy_refcounts != lazy_refcounts) { 4737 if (lazy_refcounts) { 4738 if (new_version < 3) { 4739 error_setg(errp, "Lazy refcounts only supported with " 4740 "compatibility level 1.1 and above (use compat=1.1 " 4741 "or greater)"); 4742 return -EINVAL; 4743 } 4744 s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS; 4745 ret = qcow2_update_header(bs); 4746 if (ret < 0) { 4747 s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS; 4748 error_setg_errno(errp, -ret, "Failed to update the image header"); 4749 return ret; 4750 } 4751 s->use_lazy_refcounts = true; 4752 } else { 4753 /* make image clean first */ 4754 ret = qcow2_mark_clean(bs); 4755 if (ret < 0) { 4756 error_setg_errno(errp, -ret, "Failed to make the image clean"); 4757 return ret; 4758 } 4759 /* now disallow lazy refcounts */ 4760 s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS; 4761 ret = qcow2_update_header(bs); 4762 if (ret < 0) { 4763 s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS; 4764 error_setg_errno(errp, -ret, "Failed to update the image header"); 4765 return ret; 4766 } 4767 s->use_lazy_refcounts = false; 4768 } 4769 } 4770 4771 if (new_size) { 4772 BlockBackend *blk = blk_new(BLK_PERM_RESIZE, BLK_PERM_ALL); 4773 ret = blk_insert_bs(blk, bs, errp); 4774 if (ret < 0) { 4775 blk_unref(blk); 4776 return ret; 4777 } 4778 4779 ret = blk_truncate(blk, new_size, PREALLOC_MODE_OFF, errp); 4780 blk_unref(blk); 4781 if (ret < 0) { 4782 return ret; 4783 } 4784 } 4785 4786 /* Downgrade last (so unsupported features can be removed before) */ 4787 if (new_version < old_version) { 4788 helper_cb_info.current_operation = QCOW2_DOWNGRADING; 4789 ret = qcow2_downgrade(bs, new_version, &qcow2_amend_helper_cb, 4790 &helper_cb_info, errp); 4791 if (ret < 0) { 4792 return ret; 4793 } 4794 } 4795 4796 return 0; 4797 } 4798 4799 /* 4800 * If offset or size are negative, respectively, they will not be included in 4801 * the BLOCK_IMAGE_CORRUPTED event emitted. 4802 * fatal will be ignored for read-only BDS; corruptions found there will always 4803 * be considered non-fatal. 4804 */ 4805 void qcow2_signal_corruption(BlockDriverState *bs, bool fatal, int64_t offset, 4806 int64_t size, const char *message_format, ...) 4807 { 4808 BDRVQcow2State *s = bs->opaque; 4809 const char *node_name; 4810 char *message; 4811 va_list ap; 4812 4813 fatal = fatal && bdrv_is_writable(bs); 4814 4815 if (s->signaled_corruption && 4816 (!fatal || (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT))) 4817 { 4818 return; 4819 } 4820 4821 va_start(ap, message_format); 4822 message = g_strdup_vprintf(message_format, ap); 4823 va_end(ap); 4824 4825 if (fatal) { 4826 fprintf(stderr, "qcow2: Marking image as corrupt: %s; further " 4827 "corruption events will be suppressed\n", message); 4828 } else { 4829 fprintf(stderr, "qcow2: Image is corrupt: %s; further non-fatal " 4830 "corruption events will be suppressed\n", message); 4831 } 4832 4833 node_name = bdrv_get_node_name(bs); 4834 qapi_event_send_block_image_corrupted(bdrv_get_device_name(bs), 4835 *node_name != '\0', node_name, 4836 message, offset >= 0, offset, 4837 size >= 0, size, 4838 fatal); 4839 g_free(message); 4840 4841 if (fatal) { 4842 qcow2_mark_corrupt(bs); 4843 bs->drv = NULL; /* make BDS unusable */ 4844 } 4845 4846 s->signaled_corruption = true; 4847 } 4848 4849 static QemuOptsList qcow2_create_opts = { 4850 .name = "qcow2-create-opts", 4851 .head = QTAILQ_HEAD_INITIALIZER(qcow2_create_opts.head), 4852 .desc = { 4853 { 4854 .name = BLOCK_OPT_SIZE, 4855 .type = QEMU_OPT_SIZE, 4856 .help = "Virtual disk size" 4857 }, 4858 { 4859 .name = BLOCK_OPT_COMPAT_LEVEL, 4860 .type = QEMU_OPT_STRING, 4861 .help = "Compatibility level (0.10 or 1.1)" 4862 }, 4863 { 4864 .name = BLOCK_OPT_BACKING_FILE, 4865 .type = QEMU_OPT_STRING, 4866 .help = "File name of a base image" 4867 }, 4868 { 4869 .name = BLOCK_OPT_BACKING_FMT, 4870 .type = QEMU_OPT_STRING, 4871 .help = "Image format of the base image" 4872 }, 4873 { 4874 .name = BLOCK_OPT_ENCRYPT, 4875 .type = QEMU_OPT_BOOL, 4876 .help = "Encrypt the image with format 'aes'. (Deprecated " 4877 "in favor of " BLOCK_OPT_ENCRYPT_FORMAT "=aes)", 4878 }, 4879 { 4880 .name = BLOCK_OPT_ENCRYPT_FORMAT, 4881 .type = QEMU_OPT_STRING, 4882 .help = "Encrypt the image, format choices: 'aes', 'luks'", 4883 }, 4884 BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.", 4885 "ID of secret providing qcow AES key or LUKS passphrase"), 4886 BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_ALG("encrypt."), 4887 BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_MODE("encrypt."), 4888 BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_ALG("encrypt."), 4889 BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_HASH_ALG("encrypt."), 4890 BLOCK_CRYPTO_OPT_DEF_LUKS_HASH_ALG("encrypt."), 4891 BLOCK_CRYPTO_OPT_DEF_LUKS_ITER_TIME("encrypt."), 4892 { 4893 .name = BLOCK_OPT_CLUSTER_SIZE, 4894 .type = QEMU_OPT_SIZE, 4895 .help = "qcow2 cluster size", 4896 .def_value_str = stringify(DEFAULT_CLUSTER_SIZE) 4897 }, 4898 { 4899 .name = BLOCK_OPT_PREALLOC, 4900 .type = QEMU_OPT_STRING, 4901 .help = "Preallocation mode (allowed values: off, metadata, " 4902 "falloc, full)" 4903 }, 4904 { 4905 .name = BLOCK_OPT_LAZY_REFCOUNTS, 4906 .type = QEMU_OPT_BOOL, 4907 .help = "Postpone refcount updates", 4908 .def_value_str = "off" 4909 }, 4910 { 4911 .name = BLOCK_OPT_REFCOUNT_BITS, 4912 .type = QEMU_OPT_NUMBER, 4913 .help = "Width of a reference count entry in bits", 4914 .def_value_str = "16" 4915 }, 4916 { /* end of list */ } 4917 } 4918 }; 4919 4920 BlockDriver bdrv_qcow2 = { 4921 .format_name = "qcow2", 4922 .instance_size = sizeof(BDRVQcow2State), 4923 .bdrv_probe = qcow2_probe, 4924 .bdrv_open = qcow2_open, 4925 .bdrv_close = qcow2_close, 4926 .bdrv_reopen_prepare = qcow2_reopen_prepare, 4927 .bdrv_reopen_commit = qcow2_reopen_commit, 4928 .bdrv_reopen_abort = qcow2_reopen_abort, 4929 .bdrv_join_options = qcow2_join_options, 4930 .bdrv_child_perm = bdrv_format_default_perms, 4931 .bdrv_co_create_opts = qcow2_co_create_opts, 4932 .bdrv_co_create = qcow2_co_create, 4933 .bdrv_has_zero_init = bdrv_has_zero_init_1, 4934 .bdrv_co_block_status = qcow2_co_block_status, 4935 4936 .bdrv_co_preadv = qcow2_co_preadv, 4937 .bdrv_co_pwritev = qcow2_co_pwritev, 4938 .bdrv_co_flush_to_os = qcow2_co_flush_to_os, 4939 4940 .bdrv_co_pwrite_zeroes = qcow2_co_pwrite_zeroes, 4941 .bdrv_co_pdiscard = qcow2_co_pdiscard, 4942 .bdrv_co_copy_range_from = qcow2_co_copy_range_from, 4943 .bdrv_co_copy_range_to = qcow2_co_copy_range_to, 4944 .bdrv_co_truncate = qcow2_co_truncate, 4945 .bdrv_co_pwritev_compressed = qcow2_co_pwritev_compressed, 4946 .bdrv_make_empty = qcow2_make_empty, 4947 4948 .bdrv_snapshot_create = qcow2_snapshot_create, 4949 .bdrv_snapshot_goto = qcow2_snapshot_goto, 4950 .bdrv_snapshot_delete = qcow2_snapshot_delete, 4951 .bdrv_snapshot_list = qcow2_snapshot_list, 4952 .bdrv_snapshot_load_tmp = qcow2_snapshot_load_tmp, 4953 .bdrv_measure = qcow2_measure, 4954 .bdrv_get_info = qcow2_get_info, 4955 .bdrv_get_specific_info = qcow2_get_specific_info, 4956 4957 .bdrv_save_vmstate = qcow2_save_vmstate, 4958 .bdrv_load_vmstate = qcow2_load_vmstate, 4959 4960 .supports_backing = true, 4961 .bdrv_change_backing_file = qcow2_change_backing_file, 4962 4963 .bdrv_refresh_limits = qcow2_refresh_limits, 4964 .bdrv_co_invalidate_cache = qcow2_co_invalidate_cache, 4965 .bdrv_inactivate = qcow2_inactivate, 4966 4967 .create_opts = &qcow2_create_opts, 4968 .bdrv_co_check = qcow2_co_check, 4969 .bdrv_amend_options = qcow2_amend_options, 4970 4971 .bdrv_detach_aio_context = qcow2_detach_aio_context, 4972 .bdrv_attach_aio_context = qcow2_attach_aio_context, 4973 4974 .bdrv_reopen_bitmaps_rw = qcow2_reopen_bitmaps_rw, 4975 .bdrv_can_store_new_dirty_bitmap = qcow2_can_store_new_dirty_bitmap, 4976 .bdrv_remove_persistent_dirty_bitmap = qcow2_remove_persistent_dirty_bitmap, 4977 }; 4978 4979 static void bdrv_qcow2_init(void) 4980 { 4981 bdrv_register(&bdrv_qcow2); 4982 } 4983 4984 block_init(bdrv_qcow2_init); 4985