1 /* 2 * Block driver for the QCOW 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 #include "qemu/osdep.h" 25 #include "qapi/error.h" 26 #include "qemu-common.h" 27 #include "qemu/error-report.h" 28 #include "block/block_int.h" 29 #include "sysemu/block-backend.h" 30 #include "qemu/module.h" 31 #include "qemu/bswap.h" 32 #include <zlib.h> 33 #include "qapi/qmp/qerror.h" 34 #include "qapi/qmp/qstring.h" 35 #include "crypto/block.h" 36 #include "migration/blocker.h" 37 #include "block/crypto.h" 38 39 /**************************************************************/ 40 /* QEMU COW block driver with compression and encryption support */ 41 42 #define QCOW_MAGIC (('Q' << 24) | ('F' << 16) | ('I' << 8) | 0xfb) 43 #define QCOW_VERSION 1 44 45 #define QCOW_CRYPT_NONE 0 46 #define QCOW_CRYPT_AES 1 47 48 #define QCOW_OFLAG_COMPRESSED (1LL << 63) 49 50 typedef struct QCowHeader { 51 uint32_t magic; 52 uint32_t version; 53 uint64_t backing_file_offset; 54 uint32_t backing_file_size; 55 uint32_t mtime; 56 uint64_t size; /* in bytes */ 57 uint8_t cluster_bits; 58 uint8_t l2_bits; 59 uint16_t padding; 60 uint32_t crypt_method; 61 uint64_t l1_table_offset; 62 } QEMU_PACKED QCowHeader; 63 64 #define L2_CACHE_SIZE 16 65 66 typedef struct BDRVQcowState { 67 int cluster_bits; 68 int cluster_size; 69 int cluster_sectors; 70 int l2_bits; 71 int l2_size; 72 unsigned int l1_size; 73 uint64_t cluster_offset_mask; 74 uint64_t l1_table_offset; 75 uint64_t *l1_table; 76 uint64_t *l2_cache; 77 uint64_t l2_cache_offsets[L2_CACHE_SIZE]; 78 uint32_t l2_cache_counts[L2_CACHE_SIZE]; 79 uint8_t *cluster_cache; 80 uint8_t *cluster_data; 81 uint64_t cluster_cache_offset; 82 QCryptoBlock *crypto; /* Disk encryption format driver */ 83 uint32_t crypt_method_header; 84 CoMutex lock; 85 Error *migration_blocker; 86 } BDRVQcowState; 87 88 static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset); 89 90 static int qcow_probe(const uint8_t *buf, int buf_size, const char *filename) 91 { 92 const QCowHeader *cow_header = (const void *)buf; 93 94 if (buf_size >= sizeof(QCowHeader) && 95 be32_to_cpu(cow_header->magic) == QCOW_MAGIC && 96 be32_to_cpu(cow_header->version) == QCOW_VERSION) 97 return 100; 98 else 99 return 0; 100 } 101 102 static QemuOptsList qcow_runtime_opts = { 103 .name = "qcow", 104 .head = QTAILQ_HEAD_INITIALIZER(qcow_runtime_opts.head), 105 .desc = { 106 BLOCK_CRYPTO_OPT_DEF_QCOW_KEY_SECRET("encrypt."), 107 { /* end of list */ } 108 }, 109 }; 110 111 static int qcow_open(BlockDriverState *bs, QDict *options, int flags, 112 Error **errp) 113 { 114 BDRVQcowState *s = bs->opaque; 115 unsigned int len, i, shift; 116 int ret; 117 QCowHeader header; 118 Error *local_err = NULL; 119 QCryptoBlockOpenOptions *crypto_opts = NULL; 120 unsigned int cflags = 0; 121 QDict *encryptopts = NULL; 122 const char *encryptfmt; 123 124 qdict_extract_subqdict(options, &encryptopts, "encrypt."); 125 encryptfmt = qdict_get_try_str(encryptopts, "format"); 126 127 bs->file = bdrv_open_child(NULL, options, "file", bs, &child_file, 128 false, errp); 129 if (!bs->file) { 130 ret = -EINVAL; 131 goto fail; 132 } 133 134 ret = bdrv_pread(bs->file, 0, &header, sizeof(header)); 135 if (ret < 0) { 136 goto fail; 137 } 138 be32_to_cpus(&header.magic); 139 be32_to_cpus(&header.version); 140 be64_to_cpus(&header.backing_file_offset); 141 be32_to_cpus(&header.backing_file_size); 142 be32_to_cpus(&header.mtime); 143 be64_to_cpus(&header.size); 144 be32_to_cpus(&header.crypt_method); 145 be64_to_cpus(&header.l1_table_offset); 146 147 if (header.magic != QCOW_MAGIC) { 148 error_setg(errp, "Image not in qcow format"); 149 ret = -EINVAL; 150 goto fail; 151 } 152 if (header.version != QCOW_VERSION) { 153 error_setg(errp, "Unsupported qcow version %" PRIu32, header.version); 154 ret = -ENOTSUP; 155 goto fail; 156 } 157 158 if (header.size <= 1) { 159 error_setg(errp, "Image size is too small (must be at least 2 bytes)"); 160 ret = -EINVAL; 161 goto fail; 162 } 163 if (header.cluster_bits < 9 || header.cluster_bits > 16) { 164 error_setg(errp, "Cluster size must be between 512 and 64k"); 165 ret = -EINVAL; 166 goto fail; 167 } 168 169 /* l2_bits specifies number of entries; storing a uint64_t in each entry, 170 * so bytes = num_entries << 3. */ 171 if (header.l2_bits < 9 - 3 || header.l2_bits > 16 - 3) { 172 error_setg(errp, "L2 table size must be between 512 and 64k"); 173 ret = -EINVAL; 174 goto fail; 175 } 176 177 s->crypt_method_header = header.crypt_method; 178 if (s->crypt_method_header) { 179 if (bdrv_uses_whitelist() && 180 s->crypt_method_header == QCOW_CRYPT_AES) { 181 error_setg(errp, 182 "Use of AES-CBC encrypted qcow images is no longer " 183 "supported in system emulators"); 184 error_append_hint(errp, 185 "You can use 'qemu-img convert' to convert your " 186 "image to an alternative supported format, such " 187 "as unencrypted qcow, or raw with the LUKS " 188 "format instead.\n"); 189 ret = -ENOSYS; 190 goto fail; 191 } 192 if (s->crypt_method_header == QCOW_CRYPT_AES) { 193 if (encryptfmt && !g_str_equal(encryptfmt, "aes")) { 194 error_setg(errp, 195 "Header reported 'aes' encryption format but " 196 "options specify '%s'", encryptfmt); 197 ret = -EINVAL; 198 goto fail; 199 } 200 qdict_del(encryptopts, "format"); 201 crypto_opts = block_crypto_open_opts_init( 202 Q_CRYPTO_BLOCK_FORMAT_QCOW, encryptopts, errp); 203 if (!crypto_opts) { 204 ret = -EINVAL; 205 goto fail; 206 } 207 208 if (flags & BDRV_O_NO_IO) { 209 cflags |= QCRYPTO_BLOCK_OPEN_NO_IO; 210 } 211 s->crypto = qcrypto_block_open(crypto_opts, "encrypt.", 212 NULL, NULL, cflags, errp); 213 if (!s->crypto) { 214 ret = -EINVAL; 215 goto fail; 216 } 217 } else { 218 error_setg(errp, "invalid encryption method in qcow header"); 219 ret = -EINVAL; 220 goto fail; 221 } 222 bs->encrypted = true; 223 } else { 224 if (encryptfmt) { 225 error_setg(errp, "No encryption in image header, but options " 226 "specified format '%s'", encryptfmt); 227 ret = -EINVAL; 228 goto fail; 229 } 230 } 231 s->cluster_bits = header.cluster_bits; 232 s->cluster_size = 1 << s->cluster_bits; 233 s->cluster_sectors = 1 << (s->cluster_bits - 9); 234 s->l2_bits = header.l2_bits; 235 s->l2_size = 1 << s->l2_bits; 236 bs->total_sectors = header.size / 512; 237 s->cluster_offset_mask = (1LL << (63 - s->cluster_bits)) - 1; 238 239 /* read the level 1 table */ 240 shift = s->cluster_bits + s->l2_bits; 241 if (header.size > UINT64_MAX - (1LL << shift)) { 242 error_setg(errp, "Image too large"); 243 ret = -EINVAL; 244 goto fail; 245 } else { 246 uint64_t l1_size = (header.size + (1LL << shift) - 1) >> shift; 247 if (l1_size > INT_MAX / sizeof(uint64_t)) { 248 error_setg(errp, "Image too large"); 249 ret = -EINVAL; 250 goto fail; 251 } 252 s->l1_size = l1_size; 253 } 254 255 s->l1_table_offset = header.l1_table_offset; 256 s->l1_table = g_try_new(uint64_t, s->l1_size); 257 if (s->l1_table == NULL) { 258 error_setg(errp, "Could not allocate memory for L1 table"); 259 ret = -ENOMEM; 260 goto fail; 261 } 262 263 ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table, 264 s->l1_size * sizeof(uint64_t)); 265 if (ret < 0) { 266 goto fail; 267 } 268 269 for(i = 0;i < s->l1_size; i++) { 270 be64_to_cpus(&s->l1_table[i]); 271 } 272 273 /* alloc L2 cache (max. 64k * 16 * 8 = 8 MB) */ 274 s->l2_cache = 275 qemu_try_blockalign(bs->file->bs, 276 s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t)); 277 if (s->l2_cache == NULL) { 278 error_setg(errp, "Could not allocate L2 table cache"); 279 ret = -ENOMEM; 280 goto fail; 281 } 282 s->cluster_cache = g_malloc(s->cluster_size); 283 s->cluster_data = g_malloc(s->cluster_size); 284 s->cluster_cache_offset = -1; 285 286 /* read the backing file name */ 287 if (header.backing_file_offset != 0) { 288 len = header.backing_file_size; 289 if (len > 1023 || len >= sizeof(bs->backing_file)) { 290 error_setg(errp, "Backing file name too long"); 291 ret = -EINVAL; 292 goto fail; 293 } 294 ret = bdrv_pread(bs->file, header.backing_file_offset, 295 bs->backing_file, len); 296 if (ret < 0) { 297 goto fail; 298 } 299 bs->backing_file[len] = '\0'; 300 } 301 302 /* Disable migration when qcow images are used */ 303 error_setg(&s->migration_blocker, "The qcow format used by node '%s' " 304 "does not support live migration", 305 bdrv_get_device_or_node_name(bs)); 306 ret = migrate_add_blocker(s->migration_blocker, &local_err); 307 if (local_err) { 308 error_propagate(errp, local_err); 309 error_free(s->migration_blocker); 310 goto fail; 311 } 312 313 QDECREF(encryptopts); 314 qapi_free_QCryptoBlockOpenOptions(crypto_opts); 315 qemu_co_mutex_init(&s->lock); 316 return 0; 317 318 fail: 319 g_free(s->l1_table); 320 qemu_vfree(s->l2_cache); 321 g_free(s->cluster_cache); 322 g_free(s->cluster_data); 323 qcrypto_block_free(s->crypto); 324 QDECREF(encryptopts); 325 qapi_free_QCryptoBlockOpenOptions(crypto_opts); 326 return ret; 327 } 328 329 330 /* We have nothing to do for QCOW reopen, stubs just return 331 * success */ 332 static int qcow_reopen_prepare(BDRVReopenState *state, 333 BlockReopenQueue *queue, Error **errp) 334 { 335 return 0; 336 } 337 338 339 /* 'allocate' is: 340 * 341 * 0 to not allocate. 342 * 343 * 1 to allocate a normal cluster (for sector indexes 'n_start' to 344 * 'n_end') 345 * 346 * 2 to allocate a compressed cluster of size 347 * 'compressed_size'. 'compressed_size' must be > 0 and < 348 * cluster_size 349 * 350 * return 0 if not allocated, 1 if *result is assigned, and negative 351 * errno on failure. 352 */ 353 static int get_cluster_offset(BlockDriverState *bs, 354 uint64_t offset, int allocate, 355 int compressed_size, 356 int n_start, int n_end, uint64_t *result) 357 { 358 BDRVQcowState *s = bs->opaque; 359 int min_index, i, j, l1_index, l2_index, ret; 360 int64_t l2_offset; 361 uint64_t *l2_table, cluster_offset, tmp; 362 uint32_t min_count; 363 int new_l2_table; 364 365 *result = 0; 366 l1_index = offset >> (s->l2_bits + s->cluster_bits); 367 l2_offset = s->l1_table[l1_index]; 368 new_l2_table = 0; 369 if (!l2_offset) { 370 if (!allocate) 371 return 0; 372 /* allocate a new l2 entry */ 373 l2_offset = bdrv_getlength(bs->file->bs); 374 if (l2_offset < 0) { 375 return l2_offset; 376 } 377 /* round to cluster size */ 378 l2_offset = QEMU_ALIGN_UP(l2_offset, s->cluster_size); 379 /* update the L1 entry */ 380 s->l1_table[l1_index] = l2_offset; 381 tmp = cpu_to_be64(l2_offset); 382 ret = bdrv_pwrite_sync(bs->file, 383 s->l1_table_offset + l1_index * sizeof(tmp), 384 &tmp, sizeof(tmp)); 385 if (ret < 0) { 386 return ret; 387 } 388 new_l2_table = 1; 389 } 390 for(i = 0; i < L2_CACHE_SIZE; i++) { 391 if (l2_offset == s->l2_cache_offsets[i]) { 392 /* increment the hit count */ 393 if (++s->l2_cache_counts[i] == 0xffffffff) { 394 for(j = 0; j < L2_CACHE_SIZE; j++) { 395 s->l2_cache_counts[j] >>= 1; 396 } 397 } 398 l2_table = s->l2_cache + (i << s->l2_bits); 399 goto found; 400 } 401 } 402 /* not found: load a new entry in the least used one */ 403 min_index = 0; 404 min_count = 0xffffffff; 405 for(i = 0; i < L2_CACHE_SIZE; i++) { 406 if (s->l2_cache_counts[i] < min_count) { 407 min_count = s->l2_cache_counts[i]; 408 min_index = i; 409 } 410 } 411 l2_table = s->l2_cache + (min_index << s->l2_bits); 412 if (new_l2_table) { 413 memset(l2_table, 0, s->l2_size * sizeof(uint64_t)); 414 ret = bdrv_pwrite_sync(bs->file, l2_offset, l2_table, 415 s->l2_size * sizeof(uint64_t)); 416 if (ret < 0) { 417 return ret; 418 } 419 } else { 420 ret = bdrv_pread(bs->file, l2_offset, l2_table, 421 s->l2_size * sizeof(uint64_t)); 422 if (ret < 0) { 423 return ret; 424 } 425 } 426 s->l2_cache_offsets[min_index] = l2_offset; 427 s->l2_cache_counts[min_index] = 1; 428 found: 429 l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1); 430 cluster_offset = be64_to_cpu(l2_table[l2_index]); 431 if (!cluster_offset || 432 ((cluster_offset & QCOW_OFLAG_COMPRESSED) && allocate == 1)) { 433 if (!allocate) 434 return 0; 435 /* allocate a new cluster */ 436 if ((cluster_offset & QCOW_OFLAG_COMPRESSED) && 437 (n_end - n_start) < s->cluster_sectors) { 438 /* if the cluster is already compressed, we must 439 decompress it in the case it is not completely 440 overwritten */ 441 if (decompress_cluster(bs, cluster_offset) < 0) { 442 return -EIO; 443 } 444 cluster_offset = bdrv_getlength(bs->file->bs); 445 if ((int64_t) cluster_offset < 0) { 446 return cluster_offset; 447 } 448 cluster_offset = QEMU_ALIGN_UP(cluster_offset, s->cluster_size); 449 /* write the cluster content */ 450 ret = bdrv_pwrite(bs->file, cluster_offset, s->cluster_cache, 451 s->cluster_size); 452 if (ret < 0) { 453 return ret; 454 } 455 } else { 456 cluster_offset = bdrv_getlength(bs->file->bs); 457 if ((int64_t) cluster_offset < 0) { 458 return cluster_offset; 459 } 460 if (allocate == 1) { 461 /* round to cluster size */ 462 cluster_offset = QEMU_ALIGN_UP(cluster_offset, s->cluster_size); 463 if (cluster_offset + s->cluster_size > INT64_MAX) { 464 return -E2BIG; 465 } 466 ret = bdrv_truncate(bs->file, cluster_offset + s->cluster_size, 467 PREALLOC_MODE_OFF, NULL); 468 if (ret < 0) { 469 return ret; 470 } 471 /* if encrypted, we must initialize the cluster 472 content which won't be written */ 473 if (bs->encrypted && 474 (n_end - n_start) < s->cluster_sectors) { 475 uint64_t start_sect; 476 assert(s->crypto); 477 start_sect = (offset & ~(s->cluster_size - 1)) >> 9; 478 for(i = 0; i < s->cluster_sectors; i++) { 479 if (i < n_start || i >= n_end) { 480 memset(s->cluster_data, 0x00, 512); 481 if (qcrypto_block_encrypt(s->crypto, 482 (start_sect + i) * 483 BDRV_SECTOR_SIZE, 484 s->cluster_data, 485 BDRV_SECTOR_SIZE, 486 NULL) < 0) { 487 return -EIO; 488 } 489 ret = bdrv_pwrite(bs->file, 490 cluster_offset + i * 512, 491 s->cluster_data, 512); 492 if (ret < 0) { 493 return ret; 494 } 495 } 496 } 497 } 498 } else if (allocate == 2) { 499 cluster_offset |= QCOW_OFLAG_COMPRESSED | 500 (uint64_t)compressed_size << (63 - s->cluster_bits); 501 } 502 } 503 /* update L2 table */ 504 tmp = cpu_to_be64(cluster_offset); 505 l2_table[l2_index] = tmp; 506 ret = bdrv_pwrite_sync(bs->file, l2_offset + l2_index * sizeof(tmp), 507 &tmp, sizeof(tmp)); 508 if (ret < 0) { 509 return ret; 510 } 511 } 512 *result = cluster_offset; 513 return 1; 514 } 515 516 static int64_t coroutine_fn qcow_co_get_block_status(BlockDriverState *bs, 517 int64_t sector_num, int nb_sectors, int *pnum, BlockDriverState **file) 518 { 519 BDRVQcowState *s = bs->opaque; 520 int index_in_cluster, n, ret; 521 uint64_t cluster_offset; 522 523 qemu_co_mutex_lock(&s->lock); 524 ret = get_cluster_offset(bs, sector_num << 9, 0, 0, 0, 0, &cluster_offset); 525 qemu_co_mutex_unlock(&s->lock); 526 if (ret < 0) { 527 return ret; 528 } 529 index_in_cluster = sector_num & (s->cluster_sectors - 1); 530 n = s->cluster_sectors - index_in_cluster; 531 if (n > nb_sectors) 532 n = nb_sectors; 533 *pnum = n; 534 if (!cluster_offset) { 535 return 0; 536 } 537 if ((cluster_offset & QCOW_OFLAG_COMPRESSED) || s->crypto) { 538 return BDRV_BLOCK_DATA; 539 } 540 cluster_offset |= (index_in_cluster << BDRV_SECTOR_BITS); 541 *file = bs->file->bs; 542 return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID | cluster_offset; 543 } 544 545 static int decompress_buffer(uint8_t *out_buf, int out_buf_size, 546 const uint8_t *buf, int buf_size) 547 { 548 z_stream strm1, *strm = &strm1; 549 int ret, out_len; 550 551 memset(strm, 0, sizeof(*strm)); 552 553 strm->next_in = (uint8_t *)buf; 554 strm->avail_in = buf_size; 555 strm->next_out = out_buf; 556 strm->avail_out = out_buf_size; 557 558 ret = inflateInit2(strm, -12); 559 if (ret != Z_OK) 560 return -1; 561 ret = inflate(strm, Z_FINISH); 562 out_len = strm->next_out - out_buf; 563 if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) || 564 out_len != out_buf_size) { 565 inflateEnd(strm); 566 return -1; 567 } 568 inflateEnd(strm); 569 return 0; 570 } 571 572 static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset) 573 { 574 BDRVQcowState *s = bs->opaque; 575 int ret, csize; 576 uint64_t coffset; 577 578 coffset = cluster_offset & s->cluster_offset_mask; 579 if (s->cluster_cache_offset != coffset) { 580 csize = cluster_offset >> (63 - s->cluster_bits); 581 csize &= (s->cluster_size - 1); 582 ret = bdrv_pread(bs->file, coffset, s->cluster_data, csize); 583 if (ret != csize) 584 return -1; 585 if (decompress_buffer(s->cluster_cache, s->cluster_size, 586 s->cluster_data, csize) < 0) { 587 return -1; 588 } 589 s->cluster_cache_offset = coffset; 590 } 591 return 0; 592 } 593 594 static coroutine_fn int qcow_co_readv(BlockDriverState *bs, int64_t sector_num, 595 int nb_sectors, QEMUIOVector *qiov) 596 { 597 BDRVQcowState *s = bs->opaque; 598 int index_in_cluster; 599 int ret = 0, n; 600 uint64_t cluster_offset; 601 struct iovec hd_iov; 602 QEMUIOVector hd_qiov; 603 uint8_t *buf; 604 void *orig_buf; 605 606 if (qiov->niov > 1) { 607 buf = orig_buf = qemu_try_blockalign(bs, qiov->size); 608 if (buf == NULL) { 609 return -ENOMEM; 610 } 611 } else { 612 orig_buf = NULL; 613 buf = (uint8_t *)qiov->iov->iov_base; 614 } 615 616 qemu_co_mutex_lock(&s->lock); 617 618 while (nb_sectors != 0) { 619 /* prepare next request */ 620 ret = get_cluster_offset(bs, sector_num << 9, 621 0, 0, 0, 0, &cluster_offset); 622 if (ret < 0) { 623 break; 624 } 625 index_in_cluster = sector_num & (s->cluster_sectors - 1); 626 n = s->cluster_sectors - index_in_cluster; 627 if (n > nb_sectors) { 628 n = nb_sectors; 629 } 630 631 if (!cluster_offset) { 632 if (bs->backing) { 633 /* read from the base image */ 634 hd_iov.iov_base = (void *)buf; 635 hd_iov.iov_len = n * 512; 636 qemu_iovec_init_external(&hd_qiov, &hd_iov, 1); 637 qemu_co_mutex_unlock(&s->lock); 638 ret = bdrv_co_readv(bs->backing, sector_num, n, &hd_qiov); 639 qemu_co_mutex_lock(&s->lock); 640 if (ret < 0) { 641 break; 642 } 643 } else { 644 /* Note: in this case, no need to wait */ 645 memset(buf, 0, 512 * n); 646 } 647 } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) { 648 /* add AIO support for compressed blocks ? */ 649 if (decompress_cluster(bs, cluster_offset) < 0) { 650 ret = -EIO; 651 break; 652 } 653 memcpy(buf, 654 s->cluster_cache + index_in_cluster * 512, 512 * n); 655 } else { 656 if ((cluster_offset & 511) != 0) { 657 ret = -EIO; 658 break; 659 } 660 hd_iov.iov_base = (void *)buf; 661 hd_iov.iov_len = n * 512; 662 qemu_iovec_init_external(&hd_qiov, &hd_iov, 1); 663 qemu_co_mutex_unlock(&s->lock); 664 ret = bdrv_co_readv(bs->file, 665 (cluster_offset >> 9) + index_in_cluster, 666 n, &hd_qiov); 667 qemu_co_mutex_lock(&s->lock); 668 if (ret < 0) { 669 break; 670 } 671 if (bs->encrypted) { 672 assert(s->crypto); 673 if (qcrypto_block_decrypt(s->crypto, 674 sector_num * BDRV_SECTOR_SIZE, buf, 675 n * BDRV_SECTOR_SIZE, NULL) < 0) { 676 ret = -EIO; 677 break; 678 } 679 } 680 } 681 ret = 0; 682 683 nb_sectors -= n; 684 sector_num += n; 685 buf += n * 512; 686 } 687 688 qemu_co_mutex_unlock(&s->lock); 689 690 if (qiov->niov > 1) { 691 qemu_iovec_from_buf(qiov, 0, orig_buf, qiov->size); 692 qemu_vfree(orig_buf); 693 } 694 695 return ret; 696 } 697 698 static coroutine_fn int qcow_co_writev(BlockDriverState *bs, int64_t sector_num, 699 int nb_sectors, QEMUIOVector *qiov) 700 { 701 BDRVQcowState *s = bs->opaque; 702 int index_in_cluster; 703 uint64_t cluster_offset; 704 int ret = 0, n; 705 struct iovec hd_iov; 706 QEMUIOVector hd_qiov; 707 uint8_t *buf; 708 void *orig_buf; 709 710 s->cluster_cache_offset = -1; /* disable compressed cache */ 711 712 /* We must always copy the iov when encrypting, so we 713 * don't modify the original data buffer during encryption */ 714 if (bs->encrypted || qiov->niov > 1) { 715 buf = orig_buf = qemu_try_blockalign(bs, qiov->size); 716 if (buf == NULL) { 717 return -ENOMEM; 718 } 719 qemu_iovec_to_buf(qiov, 0, buf, qiov->size); 720 } else { 721 orig_buf = NULL; 722 buf = (uint8_t *)qiov->iov->iov_base; 723 } 724 725 qemu_co_mutex_lock(&s->lock); 726 727 while (nb_sectors != 0) { 728 729 index_in_cluster = sector_num & (s->cluster_sectors - 1); 730 n = s->cluster_sectors - index_in_cluster; 731 if (n > nb_sectors) { 732 n = nb_sectors; 733 } 734 ret = get_cluster_offset(bs, sector_num << 9, 1, 0, 735 index_in_cluster, 736 index_in_cluster + n, &cluster_offset); 737 if (ret < 0) { 738 break; 739 } 740 if (!cluster_offset || (cluster_offset & 511) != 0) { 741 ret = -EIO; 742 break; 743 } 744 if (bs->encrypted) { 745 assert(s->crypto); 746 if (qcrypto_block_encrypt(s->crypto, sector_num * BDRV_SECTOR_SIZE, 747 buf, n * BDRV_SECTOR_SIZE, NULL) < 0) { 748 ret = -EIO; 749 break; 750 } 751 } 752 753 hd_iov.iov_base = (void *)buf; 754 hd_iov.iov_len = n * 512; 755 qemu_iovec_init_external(&hd_qiov, &hd_iov, 1); 756 qemu_co_mutex_unlock(&s->lock); 757 ret = bdrv_co_writev(bs->file, 758 (cluster_offset >> 9) + index_in_cluster, 759 n, &hd_qiov); 760 qemu_co_mutex_lock(&s->lock); 761 if (ret < 0) { 762 break; 763 } 764 ret = 0; 765 766 nb_sectors -= n; 767 sector_num += n; 768 buf += n * 512; 769 } 770 qemu_co_mutex_unlock(&s->lock); 771 772 qemu_vfree(orig_buf); 773 774 return ret; 775 } 776 777 static void qcow_close(BlockDriverState *bs) 778 { 779 BDRVQcowState *s = bs->opaque; 780 781 qcrypto_block_free(s->crypto); 782 s->crypto = NULL; 783 g_free(s->l1_table); 784 qemu_vfree(s->l2_cache); 785 g_free(s->cluster_cache); 786 g_free(s->cluster_data); 787 788 migrate_del_blocker(s->migration_blocker); 789 error_free(s->migration_blocker); 790 } 791 792 static int qcow_create(const char *filename, QemuOpts *opts, Error **errp) 793 { 794 int header_size, backing_filename_len, l1_size, shift, i; 795 QCowHeader header; 796 uint8_t *tmp; 797 int64_t total_size = 0; 798 char *backing_file = NULL; 799 Error *local_err = NULL; 800 int ret; 801 BlockBackend *qcow_blk; 802 char *encryptfmt = NULL; 803 QDict *options; 804 QDict *encryptopts = NULL; 805 QCryptoBlockCreateOptions *crypto_opts = NULL; 806 QCryptoBlock *crypto = NULL; 807 808 /* Read out options */ 809 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0), 810 BDRV_SECTOR_SIZE); 811 if (total_size == 0) { 812 error_setg(errp, "Image size is too small, cannot be zero length"); 813 ret = -EINVAL; 814 goto cleanup; 815 } 816 817 backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE); 818 encryptfmt = qemu_opt_get_del(opts, BLOCK_OPT_ENCRYPT_FORMAT); 819 if (encryptfmt) { 820 if (qemu_opt_get(opts, BLOCK_OPT_ENCRYPT)) { 821 error_setg(errp, "Options " BLOCK_OPT_ENCRYPT " and " 822 BLOCK_OPT_ENCRYPT_FORMAT " are mutually exclusive"); 823 ret = -EINVAL; 824 goto cleanup; 825 } 826 } else if (qemu_opt_get_bool_del(opts, BLOCK_OPT_ENCRYPT, false)) { 827 encryptfmt = g_strdup("aes"); 828 } 829 830 ret = bdrv_create_file(filename, opts, &local_err); 831 if (ret < 0) { 832 error_propagate(errp, local_err); 833 goto cleanup; 834 } 835 836 qcow_blk = blk_new_open(filename, NULL, NULL, 837 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, 838 &local_err); 839 if (qcow_blk == NULL) { 840 error_propagate(errp, local_err); 841 ret = -EIO; 842 goto cleanup; 843 } 844 845 blk_set_allow_write_beyond_eof(qcow_blk, true); 846 847 ret = blk_truncate(qcow_blk, 0, PREALLOC_MODE_OFF, errp); 848 if (ret < 0) { 849 goto exit; 850 } 851 852 memset(&header, 0, sizeof(header)); 853 header.magic = cpu_to_be32(QCOW_MAGIC); 854 header.version = cpu_to_be32(QCOW_VERSION); 855 header.size = cpu_to_be64(total_size); 856 header_size = sizeof(header); 857 backing_filename_len = 0; 858 if (backing_file) { 859 if (strcmp(backing_file, "fat:")) { 860 header.backing_file_offset = cpu_to_be64(header_size); 861 backing_filename_len = strlen(backing_file); 862 header.backing_file_size = cpu_to_be32(backing_filename_len); 863 header_size += backing_filename_len; 864 } else { 865 /* special backing file for vvfat */ 866 g_free(backing_file); 867 backing_file = NULL; 868 } 869 header.cluster_bits = 9; /* 512 byte cluster to avoid copying 870 unmodified sectors */ 871 header.l2_bits = 12; /* 32 KB L2 tables */ 872 } else { 873 header.cluster_bits = 12; /* 4 KB clusters */ 874 header.l2_bits = 9; /* 4 KB L2 tables */ 875 } 876 header_size = (header_size + 7) & ~7; 877 shift = header.cluster_bits + header.l2_bits; 878 l1_size = (total_size + (1LL << shift) - 1) >> shift; 879 880 header.l1_table_offset = cpu_to_be64(header_size); 881 882 options = qemu_opts_to_qdict(opts, NULL); 883 qdict_extract_subqdict(options, &encryptopts, "encrypt."); 884 QDECREF(options); 885 if (encryptfmt) { 886 if (!g_str_equal(encryptfmt, "aes")) { 887 error_setg(errp, "Unknown encryption format '%s', expected 'aes'", 888 encryptfmt); 889 ret = -EINVAL; 890 goto exit; 891 } 892 header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES); 893 894 crypto_opts = block_crypto_create_opts_init( 895 Q_CRYPTO_BLOCK_FORMAT_QCOW, encryptopts, errp); 896 if (!crypto_opts) { 897 ret = -EINVAL; 898 goto exit; 899 } 900 901 crypto = qcrypto_block_create(crypto_opts, "encrypt.", 902 NULL, NULL, NULL, errp); 903 if (!crypto) { 904 ret = -EINVAL; 905 goto exit; 906 } 907 } else { 908 header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE); 909 } 910 911 /* write all the data */ 912 ret = blk_pwrite(qcow_blk, 0, &header, sizeof(header), 0); 913 if (ret != sizeof(header)) { 914 goto exit; 915 } 916 917 if (backing_file) { 918 ret = blk_pwrite(qcow_blk, sizeof(header), 919 backing_file, backing_filename_len, 0); 920 if (ret != backing_filename_len) { 921 goto exit; 922 } 923 } 924 925 tmp = g_malloc0(BDRV_SECTOR_SIZE); 926 for (i = 0; i < DIV_ROUND_UP(sizeof(uint64_t) * l1_size, BDRV_SECTOR_SIZE); 927 i++) { 928 ret = blk_pwrite(qcow_blk, header_size + BDRV_SECTOR_SIZE * i, 929 tmp, BDRV_SECTOR_SIZE, 0); 930 if (ret != BDRV_SECTOR_SIZE) { 931 g_free(tmp); 932 goto exit; 933 } 934 } 935 936 g_free(tmp); 937 ret = 0; 938 exit: 939 blk_unref(qcow_blk); 940 cleanup: 941 QDECREF(encryptopts); 942 g_free(encryptfmt); 943 qcrypto_block_free(crypto); 944 qapi_free_QCryptoBlockCreateOptions(crypto_opts); 945 g_free(backing_file); 946 return ret; 947 } 948 949 static int qcow_make_empty(BlockDriverState *bs) 950 { 951 BDRVQcowState *s = bs->opaque; 952 uint32_t l1_length = s->l1_size * sizeof(uint64_t); 953 int ret; 954 955 memset(s->l1_table, 0, l1_length); 956 if (bdrv_pwrite_sync(bs->file, s->l1_table_offset, s->l1_table, 957 l1_length) < 0) 958 return -1; 959 ret = bdrv_truncate(bs->file, s->l1_table_offset + l1_length, 960 PREALLOC_MODE_OFF, NULL); 961 if (ret < 0) 962 return ret; 963 964 memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t)); 965 memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t)); 966 memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t)); 967 968 return 0; 969 } 970 971 /* XXX: put compressed sectors first, then all the cluster aligned 972 tables to avoid losing bytes in alignment */ 973 static coroutine_fn int 974 qcow_co_pwritev_compressed(BlockDriverState *bs, uint64_t offset, 975 uint64_t bytes, QEMUIOVector *qiov) 976 { 977 BDRVQcowState *s = bs->opaque; 978 QEMUIOVector hd_qiov; 979 struct iovec iov; 980 z_stream strm; 981 int ret, out_len; 982 uint8_t *buf, *out_buf; 983 uint64_t cluster_offset; 984 985 buf = qemu_blockalign(bs, s->cluster_size); 986 if (bytes != s->cluster_size) { 987 if (bytes > s->cluster_size || 988 offset + bytes != bs->total_sectors << BDRV_SECTOR_BITS) 989 { 990 qemu_vfree(buf); 991 return -EINVAL; 992 } 993 /* Zero-pad last write if image size is not cluster aligned */ 994 memset(buf + bytes, 0, s->cluster_size - bytes); 995 } 996 qemu_iovec_to_buf(qiov, 0, buf, qiov->size); 997 998 out_buf = g_malloc(s->cluster_size); 999 1000 /* best compression, small window, no zlib header */ 1001 memset(&strm, 0, sizeof(strm)); 1002 ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION, 1003 Z_DEFLATED, -12, 1004 9, Z_DEFAULT_STRATEGY); 1005 if (ret != 0) { 1006 ret = -EINVAL; 1007 goto fail; 1008 } 1009 1010 strm.avail_in = s->cluster_size; 1011 strm.next_in = (uint8_t *)buf; 1012 strm.avail_out = s->cluster_size; 1013 strm.next_out = out_buf; 1014 1015 ret = deflate(&strm, Z_FINISH); 1016 if (ret != Z_STREAM_END && ret != Z_OK) { 1017 deflateEnd(&strm); 1018 ret = -EINVAL; 1019 goto fail; 1020 } 1021 out_len = strm.next_out - out_buf; 1022 1023 deflateEnd(&strm); 1024 1025 if (ret != Z_STREAM_END || out_len >= s->cluster_size) { 1026 /* could not compress: write normal cluster */ 1027 ret = qcow_co_writev(bs, offset >> BDRV_SECTOR_BITS, 1028 bytes >> BDRV_SECTOR_BITS, qiov); 1029 if (ret < 0) { 1030 goto fail; 1031 } 1032 goto success; 1033 } 1034 qemu_co_mutex_lock(&s->lock); 1035 ret = get_cluster_offset(bs, offset, 2, out_len, 0, 0, &cluster_offset); 1036 qemu_co_mutex_unlock(&s->lock); 1037 if (ret < 0) { 1038 goto fail; 1039 } 1040 if (cluster_offset == 0) { 1041 ret = -EIO; 1042 goto fail; 1043 } 1044 cluster_offset &= s->cluster_offset_mask; 1045 1046 iov = (struct iovec) { 1047 .iov_base = out_buf, 1048 .iov_len = out_len, 1049 }; 1050 qemu_iovec_init_external(&hd_qiov, &iov, 1); 1051 ret = bdrv_co_pwritev(bs->file, cluster_offset, out_len, &hd_qiov, 0); 1052 if (ret < 0) { 1053 goto fail; 1054 } 1055 success: 1056 ret = 0; 1057 fail: 1058 qemu_vfree(buf); 1059 g_free(out_buf); 1060 return ret; 1061 } 1062 1063 static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) 1064 { 1065 BDRVQcowState *s = bs->opaque; 1066 bdi->cluster_size = s->cluster_size; 1067 return 0; 1068 } 1069 1070 static QemuOptsList qcow_create_opts = { 1071 .name = "qcow-create-opts", 1072 .head = QTAILQ_HEAD_INITIALIZER(qcow_create_opts.head), 1073 .desc = { 1074 { 1075 .name = BLOCK_OPT_SIZE, 1076 .type = QEMU_OPT_SIZE, 1077 .help = "Virtual disk size" 1078 }, 1079 { 1080 .name = BLOCK_OPT_BACKING_FILE, 1081 .type = QEMU_OPT_STRING, 1082 .help = "File name of a base image" 1083 }, 1084 { 1085 .name = BLOCK_OPT_ENCRYPT, 1086 .type = QEMU_OPT_BOOL, 1087 .help = "Encrypt the image with format 'aes'. (Deprecated " 1088 "in favor of " BLOCK_OPT_ENCRYPT_FORMAT "=aes)", 1089 }, 1090 { 1091 .name = BLOCK_OPT_ENCRYPT_FORMAT, 1092 .type = QEMU_OPT_STRING, 1093 .help = "Encrypt the image, format choices: 'aes'", 1094 }, 1095 BLOCK_CRYPTO_OPT_DEF_QCOW_KEY_SECRET("encrypt."), 1096 { /* end of list */ } 1097 } 1098 }; 1099 1100 static BlockDriver bdrv_qcow = { 1101 .format_name = "qcow", 1102 .instance_size = sizeof(BDRVQcowState), 1103 .bdrv_probe = qcow_probe, 1104 .bdrv_open = qcow_open, 1105 .bdrv_close = qcow_close, 1106 .bdrv_child_perm = bdrv_format_default_perms, 1107 .bdrv_reopen_prepare = qcow_reopen_prepare, 1108 .bdrv_create = qcow_create, 1109 .bdrv_has_zero_init = bdrv_has_zero_init_1, 1110 .supports_backing = true, 1111 1112 .bdrv_co_readv = qcow_co_readv, 1113 .bdrv_co_writev = qcow_co_writev, 1114 .bdrv_co_get_block_status = qcow_co_get_block_status, 1115 1116 .bdrv_make_empty = qcow_make_empty, 1117 .bdrv_co_pwritev_compressed = qcow_co_pwritev_compressed, 1118 .bdrv_get_info = qcow_get_info, 1119 1120 .create_opts = &qcow_create_opts, 1121 }; 1122 1123 static void bdrv_qcow_init(void) 1124 { 1125 bdrv_register(&bdrv_qcow); 1126 } 1127 1128 block_init(bdrv_qcow_init); 1129