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 #include "qemu-common.h" 25 #include "block_int.h" 26 #include "module.h" 27 #include <zlib.h> 28 #include "aes.h" 29 #include "block/qcow2.h" 30 #include "qemu-error.h" 31 #include "qerror.h" 32 33 /* 34 Differences with QCOW: 35 36 - Support for multiple incremental snapshots. 37 - Memory management by reference counts. 38 - Clusters which have a reference count of one have the bit 39 QCOW_OFLAG_COPIED to optimize write performance. 40 - Size of compressed clusters is stored in sectors to reduce bit usage 41 in the cluster offsets. 42 - Support for storing additional data (such as the VM state) in the 43 snapshots. 44 - If a backing store is used, the cluster size is not constrained 45 (could be backported to QCOW). 46 - L2 tables have always a size of one cluster. 47 */ 48 49 50 typedef struct { 51 uint32_t magic; 52 uint32_t len; 53 } QCowExtension; 54 #define QCOW2_EXT_MAGIC_END 0 55 #define QCOW2_EXT_MAGIC_BACKING_FORMAT 0xE2792ACA 56 57 static int qcow2_probe(const uint8_t *buf, int buf_size, const char *filename) 58 { 59 const QCowHeader *cow_header = (const void *)buf; 60 61 if (buf_size >= sizeof(QCowHeader) && 62 be32_to_cpu(cow_header->magic) == QCOW_MAGIC && 63 be32_to_cpu(cow_header->version) >= QCOW_VERSION) 64 return 100; 65 else 66 return 0; 67 } 68 69 70 /* 71 * read qcow2 extension and fill bs 72 * start reading from start_offset 73 * finish reading upon magic of value 0 or when end_offset reached 74 * unknown magic is skipped (future extension this version knows nothing about) 75 * return 0 upon success, non-0 otherwise 76 */ 77 static int qcow2_read_extensions(BlockDriverState *bs, uint64_t start_offset, 78 uint64_t end_offset) 79 { 80 QCowExtension ext; 81 uint64_t offset; 82 83 #ifdef DEBUG_EXT 84 printf("qcow2_read_extensions: start=%ld end=%ld\n", start_offset, end_offset); 85 #endif 86 offset = start_offset; 87 while (offset < end_offset) { 88 89 #ifdef DEBUG_EXT 90 BDRVQcowState *s = bs->opaque; 91 /* Sanity check */ 92 if (offset > s->cluster_size) 93 printf("qcow2_read_extension: suspicious offset %lu\n", offset); 94 95 printf("attemting to read extended header in offset %lu\n", offset); 96 #endif 97 98 if (bdrv_pread(bs->file, offset, &ext, sizeof(ext)) != sizeof(ext)) { 99 fprintf(stderr, "qcow2_read_extension: ERROR: " 100 "pread fail from offset %" PRIu64 "\n", 101 offset); 102 return 1; 103 } 104 be32_to_cpus(&ext.magic); 105 be32_to_cpus(&ext.len); 106 offset += sizeof(ext); 107 #ifdef DEBUG_EXT 108 printf("ext.magic = 0x%x\n", ext.magic); 109 #endif 110 switch (ext.magic) { 111 case QCOW2_EXT_MAGIC_END: 112 return 0; 113 114 case QCOW2_EXT_MAGIC_BACKING_FORMAT: 115 if (ext.len >= sizeof(bs->backing_format)) { 116 fprintf(stderr, "ERROR: ext_backing_format: len=%u too large" 117 " (>=%zu)\n", 118 ext.len, sizeof(bs->backing_format)); 119 return 2; 120 } 121 if (bdrv_pread(bs->file, offset , bs->backing_format, 122 ext.len) != ext.len) 123 return 3; 124 bs->backing_format[ext.len] = '\0'; 125 #ifdef DEBUG_EXT 126 printf("Qcow2: Got format extension %s\n", bs->backing_format); 127 #endif 128 offset = ((offset + ext.len + 7) & ~7); 129 break; 130 131 default: 132 /* unknown magic -- just skip it */ 133 offset = ((offset + ext.len + 7) & ~7); 134 break; 135 } 136 } 137 138 return 0; 139 } 140 141 142 static int qcow2_open(BlockDriverState *bs, int flags) 143 { 144 BDRVQcowState *s = bs->opaque; 145 int len, i, ret = 0; 146 QCowHeader header; 147 uint64_t ext_end; 148 bool writethrough; 149 150 ret = bdrv_pread(bs->file, 0, &header, sizeof(header)); 151 if (ret < 0) { 152 goto fail; 153 } 154 be32_to_cpus(&header.magic); 155 be32_to_cpus(&header.version); 156 be64_to_cpus(&header.backing_file_offset); 157 be32_to_cpus(&header.backing_file_size); 158 be64_to_cpus(&header.size); 159 be32_to_cpus(&header.cluster_bits); 160 be32_to_cpus(&header.crypt_method); 161 be64_to_cpus(&header.l1_table_offset); 162 be32_to_cpus(&header.l1_size); 163 be64_to_cpus(&header.refcount_table_offset); 164 be32_to_cpus(&header.refcount_table_clusters); 165 be64_to_cpus(&header.snapshots_offset); 166 be32_to_cpus(&header.nb_snapshots); 167 168 if (header.magic != QCOW_MAGIC) { 169 ret = -EINVAL; 170 goto fail; 171 } 172 if (header.version != QCOW_VERSION) { 173 char version[64]; 174 snprintf(version, sizeof(version), "QCOW version %d", header.version); 175 qerror_report(QERR_UNKNOWN_BLOCK_FORMAT_FEATURE, 176 bs->device_name, "qcow2", version); 177 ret = -ENOTSUP; 178 goto fail; 179 } 180 if (header.cluster_bits < MIN_CLUSTER_BITS || 181 header.cluster_bits > MAX_CLUSTER_BITS) { 182 ret = -EINVAL; 183 goto fail; 184 } 185 if (header.crypt_method > QCOW_CRYPT_AES) { 186 ret = -EINVAL; 187 goto fail; 188 } 189 s->crypt_method_header = header.crypt_method; 190 if (s->crypt_method_header) { 191 bs->encrypted = 1; 192 } 193 s->cluster_bits = header.cluster_bits; 194 s->cluster_size = 1 << s->cluster_bits; 195 s->cluster_sectors = 1 << (s->cluster_bits - 9); 196 s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */ 197 s->l2_size = 1 << s->l2_bits; 198 bs->total_sectors = header.size / 512; 199 s->csize_shift = (62 - (s->cluster_bits - 8)); 200 s->csize_mask = (1 << (s->cluster_bits - 8)) - 1; 201 s->cluster_offset_mask = (1LL << s->csize_shift) - 1; 202 s->refcount_table_offset = header.refcount_table_offset; 203 s->refcount_table_size = 204 header.refcount_table_clusters << (s->cluster_bits - 3); 205 206 s->snapshots_offset = header.snapshots_offset; 207 s->nb_snapshots = header.nb_snapshots; 208 209 /* read the level 1 table */ 210 s->l1_size = header.l1_size; 211 s->l1_vm_state_index = size_to_l1(s, header.size); 212 /* the L1 table must contain at least enough entries to put 213 header.size bytes */ 214 if (s->l1_size < s->l1_vm_state_index) { 215 ret = -EINVAL; 216 goto fail; 217 } 218 s->l1_table_offset = header.l1_table_offset; 219 if (s->l1_size > 0) { 220 s->l1_table = g_malloc0( 221 align_offset(s->l1_size * sizeof(uint64_t), 512)); 222 ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table, 223 s->l1_size * sizeof(uint64_t)); 224 if (ret < 0) { 225 goto fail; 226 } 227 for(i = 0;i < s->l1_size; i++) { 228 be64_to_cpus(&s->l1_table[i]); 229 } 230 } 231 232 /* alloc L2 table/refcount block cache */ 233 writethrough = ((flags & BDRV_O_CACHE_WB) == 0); 234 s->l2_table_cache = qcow2_cache_create(bs, L2_CACHE_SIZE, writethrough); 235 s->refcount_block_cache = qcow2_cache_create(bs, REFCOUNT_CACHE_SIZE, 236 writethrough); 237 238 s->cluster_cache = g_malloc(s->cluster_size); 239 /* one more sector for decompressed data alignment */ 240 s->cluster_data = qemu_blockalign(bs, QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size 241 + 512); 242 s->cluster_cache_offset = -1; 243 244 ret = qcow2_refcount_init(bs); 245 if (ret != 0) { 246 goto fail; 247 } 248 249 QLIST_INIT(&s->cluster_allocs); 250 251 /* read qcow2 extensions */ 252 if (header.backing_file_offset) { 253 ext_end = header.backing_file_offset; 254 } else { 255 ext_end = s->cluster_size; 256 } 257 if (qcow2_read_extensions(bs, sizeof(header), ext_end)) { 258 ret = -EINVAL; 259 goto fail; 260 } 261 262 /* read the backing file name */ 263 if (header.backing_file_offset != 0) { 264 len = header.backing_file_size; 265 if (len > 1023) { 266 len = 1023; 267 } 268 ret = bdrv_pread(bs->file, header.backing_file_offset, 269 bs->backing_file, len); 270 if (ret < 0) { 271 goto fail; 272 } 273 bs->backing_file[len] = '\0'; 274 } 275 if (qcow2_read_snapshots(bs) < 0) { 276 ret = -EINVAL; 277 goto fail; 278 } 279 280 /* Initialise locks */ 281 qemu_co_mutex_init(&s->lock); 282 283 #ifdef DEBUG_ALLOC 284 { 285 BdrvCheckResult result = {0}; 286 qcow2_check_refcounts(bs, &result); 287 } 288 #endif 289 return ret; 290 291 fail: 292 qcow2_free_snapshots(bs); 293 qcow2_refcount_close(bs); 294 g_free(s->l1_table); 295 if (s->l2_table_cache) { 296 qcow2_cache_destroy(bs, s->l2_table_cache); 297 } 298 g_free(s->cluster_cache); 299 qemu_vfree(s->cluster_data); 300 return ret; 301 } 302 303 static int qcow2_set_key(BlockDriverState *bs, const char *key) 304 { 305 BDRVQcowState *s = bs->opaque; 306 uint8_t keybuf[16]; 307 int len, i; 308 309 memset(keybuf, 0, 16); 310 len = strlen(key); 311 if (len > 16) 312 len = 16; 313 /* XXX: we could compress the chars to 7 bits to increase 314 entropy */ 315 for(i = 0;i < len;i++) { 316 keybuf[i] = key[i]; 317 } 318 s->crypt_method = s->crypt_method_header; 319 320 if (AES_set_encrypt_key(keybuf, 128, &s->aes_encrypt_key) != 0) 321 return -1; 322 if (AES_set_decrypt_key(keybuf, 128, &s->aes_decrypt_key) != 0) 323 return -1; 324 #if 0 325 /* test */ 326 { 327 uint8_t in[16]; 328 uint8_t out[16]; 329 uint8_t tmp[16]; 330 for(i=0;i<16;i++) 331 in[i] = i; 332 AES_encrypt(in, tmp, &s->aes_encrypt_key); 333 AES_decrypt(tmp, out, &s->aes_decrypt_key); 334 for(i = 0; i < 16; i++) 335 printf(" %02x", tmp[i]); 336 printf("\n"); 337 for(i = 0; i < 16; i++) 338 printf(" %02x", out[i]); 339 printf("\n"); 340 } 341 #endif 342 return 0; 343 } 344 345 static int qcow2_is_allocated(BlockDriverState *bs, int64_t sector_num, 346 int nb_sectors, int *pnum) 347 { 348 uint64_t cluster_offset; 349 int ret; 350 351 *pnum = nb_sectors; 352 /* FIXME We can get errors here, but the bdrv_is_allocated interface can't 353 * pass them on today */ 354 ret = qcow2_get_cluster_offset(bs, sector_num << 9, pnum, &cluster_offset); 355 if (ret < 0) { 356 *pnum = 0; 357 } 358 359 return (cluster_offset != 0); 360 } 361 362 /* handle reading after the end of the backing file */ 363 int qcow2_backing_read1(BlockDriverState *bs, QEMUIOVector *qiov, 364 int64_t sector_num, int nb_sectors) 365 { 366 int n1; 367 if ((sector_num + nb_sectors) <= bs->total_sectors) 368 return nb_sectors; 369 if (sector_num >= bs->total_sectors) 370 n1 = 0; 371 else 372 n1 = bs->total_sectors - sector_num; 373 374 qemu_iovec_memset_skip(qiov, 0, 512 * (nb_sectors - n1), 512 * n1); 375 376 return n1; 377 } 378 379 static int qcow2_co_readv(BlockDriverState *bs, int64_t sector_num, 380 int remaining_sectors, QEMUIOVector *qiov) 381 { 382 BDRVQcowState *s = bs->opaque; 383 int index_in_cluster, n1; 384 int ret; 385 int cur_nr_sectors; /* number of sectors in current iteration */ 386 uint64_t cluster_offset = 0; 387 uint64_t bytes_done = 0; 388 QEMUIOVector hd_qiov; 389 uint8_t *cluster_data = NULL; 390 391 qemu_iovec_init(&hd_qiov, qiov->niov); 392 393 qemu_co_mutex_lock(&s->lock); 394 395 while (remaining_sectors != 0) { 396 397 /* prepare next request */ 398 cur_nr_sectors = remaining_sectors; 399 if (s->crypt_method) { 400 cur_nr_sectors = MIN(cur_nr_sectors, 401 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors); 402 } 403 404 ret = qcow2_get_cluster_offset(bs, sector_num << 9, 405 &cur_nr_sectors, &cluster_offset); 406 if (ret < 0) { 407 goto fail; 408 } 409 410 index_in_cluster = sector_num & (s->cluster_sectors - 1); 411 412 qemu_iovec_reset(&hd_qiov); 413 qemu_iovec_copy(&hd_qiov, qiov, bytes_done, 414 cur_nr_sectors * 512); 415 416 if (!cluster_offset) { 417 418 if (bs->backing_hd) { 419 /* read from the base image */ 420 n1 = qcow2_backing_read1(bs->backing_hd, &hd_qiov, 421 sector_num, cur_nr_sectors); 422 if (n1 > 0) { 423 BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO); 424 qemu_co_mutex_unlock(&s->lock); 425 ret = bdrv_co_readv(bs->backing_hd, sector_num, 426 n1, &hd_qiov); 427 qemu_co_mutex_lock(&s->lock); 428 if (ret < 0) { 429 goto fail; 430 } 431 } 432 } else { 433 /* Note: in this case, no need to wait */ 434 qemu_iovec_memset(&hd_qiov, 0, 512 * cur_nr_sectors); 435 } 436 } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) { 437 /* add AIO support for compressed blocks ? */ 438 ret = qcow2_decompress_cluster(bs, cluster_offset); 439 if (ret < 0) { 440 goto fail; 441 } 442 443 qemu_iovec_from_buffer(&hd_qiov, 444 s->cluster_cache + index_in_cluster * 512, 445 512 * cur_nr_sectors); 446 } else { 447 if ((cluster_offset & 511) != 0) { 448 ret = -EIO; 449 goto fail; 450 } 451 452 if (s->crypt_method) { 453 /* 454 * For encrypted images, read everything into a temporary 455 * contiguous buffer on which the AES functions can work. 456 */ 457 if (!cluster_data) { 458 cluster_data = 459 qemu_blockalign(bs, QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size); 460 } 461 462 assert(cur_nr_sectors <= 463 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors); 464 qemu_iovec_reset(&hd_qiov); 465 qemu_iovec_add(&hd_qiov, cluster_data, 466 512 * cur_nr_sectors); 467 } 468 469 BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO); 470 qemu_co_mutex_unlock(&s->lock); 471 ret = bdrv_co_readv(bs->file, 472 (cluster_offset >> 9) + index_in_cluster, 473 cur_nr_sectors, &hd_qiov); 474 qemu_co_mutex_lock(&s->lock); 475 if (ret < 0) { 476 goto fail; 477 } 478 if (s->crypt_method) { 479 qcow2_encrypt_sectors(s, sector_num, cluster_data, 480 cluster_data, cur_nr_sectors, 0, &s->aes_decrypt_key); 481 qemu_iovec_reset(&hd_qiov); 482 qemu_iovec_copy(&hd_qiov, qiov, bytes_done, 483 cur_nr_sectors * 512); 484 qemu_iovec_from_buffer(&hd_qiov, cluster_data, 485 512 * cur_nr_sectors); 486 } 487 } 488 489 remaining_sectors -= cur_nr_sectors; 490 sector_num += cur_nr_sectors; 491 bytes_done += cur_nr_sectors * 512; 492 } 493 ret = 0; 494 495 fail: 496 qemu_co_mutex_unlock(&s->lock); 497 498 qemu_iovec_destroy(&hd_qiov); 499 qemu_vfree(cluster_data); 500 501 return ret; 502 } 503 504 static void run_dependent_requests(BDRVQcowState *s, QCowL2Meta *m) 505 { 506 /* Take the request off the list of running requests */ 507 if (m->nb_clusters != 0) { 508 QLIST_REMOVE(m, next_in_flight); 509 } 510 511 /* Restart all dependent requests */ 512 if (!qemu_co_queue_empty(&m->dependent_requests)) { 513 qemu_co_mutex_unlock(&s->lock); 514 while(qemu_co_queue_next(&m->dependent_requests)); 515 qemu_co_mutex_lock(&s->lock); 516 } 517 } 518 519 static int qcow2_co_writev(BlockDriverState *bs, 520 int64_t sector_num, 521 int remaining_sectors, 522 QEMUIOVector *qiov) 523 { 524 BDRVQcowState *s = bs->opaque; 525 int index_in_cluster; 526 int n_end; 527 int ret; 528 int cur_nr_sectors; /* number of sectors in current iteration */ 529 uint64_t cluster_offset; 530 QEMUIOVector hd_qiov; 531 uint64_t bytes_done = 0; 532 uint8_t *cluster_data = NULL; 533 QCowL2Meta l2meta = { 534 .nb_clusters = 0, 535 }; 536 537 qemu_co_queue_init(&l2meta.dependent_requests); 538 539 qemu_iovec_init(&hd_qiov, qiov->niov); 540 541 s->cluster_cache_offset = -1; /* disable compressed cache */ 542 543 qemu_co_mutex_lock(&s->lock); 544 545 while (remaining_sectors != 0) { 546 547 index_in_cluster = sector_num & (s->cluster_sectors - 1); 548 n_end = index_in_cluster + remaining_sectors; 549 if (s->crypt_method && 550 n_end > QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors) { 551 n_end = QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors; 552 } 553 554 ret = qcow2_alloc_cluster_offset(bs, sector_num << 9, 555 index_in_cluster, n_end, &cur_nr_sectors, &l2meta); 556 if (ret < 0) { 557 goto fail; 558 } 559 560 cluster_offset = l2meta.cluster_offset; 561 assert((cluster_offset & 511) == 0); 562 563 qemu_iovec_reset(&hd_qiov); 564 qemu_iovec_copy(&hd_qiov, qiov, bytes_done, 565 cur_nr_sectors * 512); 566 567 if (s->crypt_method) { 568 if (!cluster_data) { 569 cluster_data = qemu_blockalign(bs, QCOW_MAX_CRYPT_CLUSTERS * 570 s->cluster_size); 571 } 572 573 assert(hd_qiov.size <= 574 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size); 575 qemu_iovec_to_buffer(&hd_qiov, cluster_data); 576 577 qcow2_encrypt_sectors(s, sector_num, cluster_data, 578 cluster_data, cur_nr_sectors, 1, &s->aes_encrypt_key); 579 580 qemu_iovec_reset(&hd_qiov); 581 qemu_iovec_add(&hd_qiov, cluster_data, 582 cur_nr_sectors * 512); 583 } 584 585 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO); 586 qemu_co_mutex_unlock(&s->lock); 587 ret = bdrv_co_writev(bs->file, 588 (cluster_offset >> 9) + index_in_cluster, 589 cur_nr_sectors, &hd_qiov); 590 qemu_co_mutex_lock(&s->lock); 591 if (ret < 0) { 592 goto fail; 593 } 594 595 ret = qcow2_alloc_cluster_link_l2(bs, &l2meta); 596 if (ret < 0) { 597 goto fail; 598 } 599 600 run_dependent_requests(s, &l2meta); 601 602 remaining_sectors -= cur_nr_sectors; 603 sector_num += cur_nr_sectors; 604 bytes_done += cur_nr_sectors * 512; 605 } 606 ret = 0; 607 608 fail: 609 run_dependent_requests(s, &l2meta); 610 611 qemu_co_mutex_unlock(&s->lock); 612 613 qemu_iovec_destroy(&hd_qiov); 614 qemu_vfree(cluster_data); 615 616 return ret; 617 } 618 619 static void qcow2_close(BlockDriverState *bs) 620 { 621 BDRVQcowState *s = bs->opaque; 622 g_free(s->l1_table); 623 624 qcow2_cache_flush(bs, s->l2_table_cache); 625 qcow2_cache_flush(bs, s->refcount_block_cache); 626 627 qcow2_cache_destroy(bs, s->l2_table_cache); 628 qcow2_cache_destroy(bs, s->refcount_block_cache); 629 630 g_free(s->cluster_cache); 631 qemu_vfree(s->cluster_data); 632 qcow2_refcount_close(bs); 633 } 634 635 /* 636 * Updates the variable length parts of the qcow2 header, i.e. the backing file 637 * name and all extensions. qcow2 was not designed to allow such changes, so if 638 * we run out of space (we can only use the first cluster) this function may 639 * fail. 640 * 641 * Returns 0 on success, -errno in error cases. 642 */ 643 static int qcow2_update_ext_header(BlockDriverState *bs, 644 const char *backing_file, const char *backing_fmt) 645 { 646 size_t backing_file_len = 0; 647 size_t backing_fmt_len = 0; 648 BDRVQcowState *s = bs->opaque; 649 QCowExtension ext_backing_fmt = {0, 0}; 650 int ret; 651 652 /* Backing file format doesn't make sense without a backing file */ 653 if (backing_fmt && !backing_file) { 654 return -EINVAL; 655 } 656 657 /* Prepare the backing file format extension if needed */ 658 if (backing_fmt) { 659 ext_backing_fmt.len = cpu_to_be32(strlen(backing_fmt)); 660 ext_backing_fmt.magic = cpu_to_be32(QCOW2_EXT_MAGIC_BACKING_FORMAT); 661 backing_fmt_len = ((sizeof(ext_backing_fmt) 662 + strlen(backing_fmt) + 7) & ~7); 663 } 664 665 /* Check if we can fit the new header into the first cluster */ 666 if (backing_file) { 667 backing_file_len = strlen(backing_file); 668 } 669 670 size_t header_size = sizeof(QCowHeader) + backing_file_len 671 + backing_fmt_len; 672 673 if (header_size > s->cluster_size) { 674 return -ENOSPC; 675 } 676 677 /* Rewrite backing file name and qcow2 extensions */ 678 size_t ext_size = header_size - sizeof(QCowHeader); 679 uint8_t buf[ext_size]; 680 size_t offset = 0; 681 size_t backing_file_offset = 0; 682 683 if (backing_file) { 684 if (backing_fmt) { 685 int padding = backing_fmt_len - 686 (sizeof(ext_backing_fmt) + strlen(backing_fmt)); 687 688 memcpy(buf + offset, &ext_backing_fmt, sizeof(ext_backing_fmt)); 689 offset += sizeof(ext_backing_fmt); 690 691 memcpy(buf + offset, backing_fmt, strlen(backing_fmt)); 692 offset += strlen(backing_fmt); 693 694 memset(buf + offset, 0, padding); 695 offset += padding; 696 } 697 698 memcpy(buf + offset, backing_file, backing_file_len); 699 backing_file_offset = sizeof(QCowHeader) + offset; 700 } 701 702 ret = bdrv_pwrite_sync(bs->file, sizeof(QCowHeader), buf, ext_size); 703 if (ret < 0) { 704 goto fail; 705 } 706 707 /* Update header fields */ 708 uint64_t be_backing_file_offset = cpu_to_be64(backing_file_offset); 709 uint32_t be_backing_file_size = cpu_to_be32(backing_file_len); 710 711 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, backing_file_offset), 712 &be_backing_file_offset, sizeof(uint64_t)); 713 if (ret < 0) { 714 goto fail; 715 } 716 717 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, backing_file_size), 718 &be_backing_file_size, sizeof(uint32_t)); 719 if (ret < 0) { 720 goto fail; 721 } 722 723 ret = 0; 724 fail: 725 return ret; 726 } 727 728 static int qcow2_change_backing_file(BlockDriverState *bs, 729 const char *backing_file, const char *backing_fmt) 730 { 731 return qcow2_update_ext_header(bs, backing_file, backing_fmt); 732 } 733 734 static int preallocate(BlockDriverState *bs) 735 { 736 uint64_t nb_sectors; 737 uint64_t offset; 738 int num; 739 int ret; 740 QCowL2Meta meta; 741 742 nb_sectors = bdrv_getlength(bs) >> 9; 743 offset = 0; 744 qemu_co_queue_init(&meta.dependent_requests); 745 meta.cluster_offset = 0; 746 747 while (nb_sectors) { 748 num = MIN(nb_sectors, INT_MAX >> 9); 749 ret = qcow2_alloc_cluster_offset(bs, offset, 0, num, &num, &meta); 750 if (ret < 0) { 751 return ret; 752 } 753 754 ret = qcow2_alloc_cluster_link_l2(bs, &meta); 755 if (ret < 0) { 756 qcow2_free_any_clusters(bs, meta.cluster_offset, meta.nb_clusters); 757 return ret; 758 } 759 760 /* There are no dependent requests, but we need to remove our request 761 * from the list of in-flight requests */ 762 run_dependent_requests(bs->opaque, &meta); 763 764 /* TODO Preallocate data if requested */ 765 766 nb_sectors -= num; 767 offset += num << 9; 768 } 769 770 /* 771 * It is expected that the image file is large enough to actually contain 772 * all of the allocated clusters (otherwise we get failing reads after 773 * EOF). Extend the image to the last allocated sector. 774 */ 775 if (meta.cluster_offset != 0) { 776 uint8_t buf[512]; 777 memset(buf, 0, 512); 778 ret = bdrv_write(bs->file, (meta.cluster_offset >> 9) + num - 1, buf, 1); 779 if (ret < 0) { 780 return ret; 781 } 782 } 783 784 return 0; 785 } 786 787 static int qcow2_create2(const char *filename, int64_t total_size, 788 const char *backing_file, const char *backing_format, 789 int flags, size_t cluster_size, int prealloc, 790 QEMUOptionParameter *options) 791 { 792 /* Calulate cluster_bits */ 793 int cluster_bits; 794 cluster_bits = ffs(cluster_size) - 1; 795 if (cluster_bits < MIN_CLUSTER_BITS || cluster_bits > MAX_CLUSTER_BITS || 796 (1 << cluster_bits) != cluster_size) 797 { 798 error_report( 799 "Cluster size must be a power of two between %d and %dk", 800 1 << MIN_CLUSTER_BITS, 1 << (MAX_CLUSTER_BITS - 10)); 801 return -EINVAL; 802 } 803 804 /* 805 * Open the image file and write a minimal qcow2 header. 806 * 807 * We keep things simple and start with a zero-sized image. We also 808 * do without refcount blocks or a L1 table for now. We'll fix the 809 * inconsistency later. 810 * 811 * We do need a refcount table because growing the refcount table means 812 * allocating two new refcount blocks - the seconds of which would be at 813 * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file 814 * size for any qcow2 image. 815 */ 816 BlockDriverState* bs; 817 QCowHeader header; 818 uint8_t* refcount_table; 819 int ret; 820 821 ret = bdrv_create_file(filename, options); 822 if (ret < 0) { 823 return ret; 824 } 825 826 ret = bdrv_file_open(&bs, filename, BDRV_O_RDWR); 827 if (ret < 0) { 828 return ret; 829 } 830 831 /* Write the header */ 832 memset(&header, 0, sizeof(header)); 833 header.magic = cpu_to_be32(QCOW_MAGIC); 834 header.version = cpu_to_be32(QCOW_VERSION); 835 header.cluster_bits = cpu_to_be32(cluster_bits); 836 header.size = cpu_to_be64(0); 837 header.l1_table_offset = cpu_to_be64(0); 838 header.l1_size = cpu_to_be32(0); 839 header.refcount_table_offset = cpu_to_be64(cluster_size); 840 header.refcount_table_clusters = cpu_to_be32(1); 841 842 if (flags & BLOCK_FLAG_ENCRYPT) { 843 header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES); 844 } else { 845 header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE); 846 } 847 848 ret = bdrv_pwrite(bs, 0, &header, sizeof(header)); 849 if (ret < 0) { 850 goto out; 851 } 852 853 /* Write an empty refcount table */ 854 refcount_table = g_malloc0(cluster_size); 855 ret = bdrv_pwrite(bs, cluster_size, refcount_table, cluster_size); 856 g_free(refcount_table); 857 858 if (ret < 0) { 859 goto out; 860 } 861 862 bdrv_close(bs); 863 864 /* 865 * And now open the image and make it consistent first (i.e. increase the 866 * refcount of the cluster that is occupied by the header and the refcount 867 * table) 868 */ 869 BlockDriver* drv = bdrv_find_format("qcow2"); 870 assert(drv != NULL); 871 ret = bdrv_open(bs, filename, 872 BDRV_O_RDWR | BDRV_O_CACHE_WB | BDRV_O_NO_FLUSH, drv); 873 if (ret < 0) { 874 goto out; 875 } 876 877 ret = qcow2_alloc_clusters(bs, 2 * cluster_size); 878 if (ret < 0) { 879 goto out; 880 881 } else if (ret != 0) { 882 error_report("Huh, first cluster in empty image is already in use?"); 883 abort(); 884 } 885 886 /* Okay, now that we have a valid image, let's give it the right size */ 887 ret = bdrv_truncate(bs, total_size * BDRV_SECTOR_SIZE); 888 if (ret < 0) { 889 goto out; 890 } 891 892 /* Want a backing file? There you go.*/ 893 if (backing_file) { 894 ret = bdrv_change_backing_file(bs, backing_file, backing_format); 895 if (ret < 0) { 896 goto out; 897 } 898 } 899 900 /* And if we're supposed to preallocate metadata, do that now */ 901 if (prealloc) { 902 ret = preallocate(bs); 903 if (ret < 0) { 904 goto out; 905 } 906 } 907 908 ret = 0; 909 out: 910 bdrv_delete(bs); 911 return ret; 912 } 913 914 static int qcow2_create(const char *filename, QEMUOptionParameter *options) 915 { 916 const char *backing_file = NULL; 917 const char *backing_fmt = NULL; 918 uint64_t sectors = 0; 919 int flags = 0; 920 size_t cluster_size = DEFAULT_CLUSTER_SIZE; 921 int prealloc = 0; 922 923 /* Read out options */ 924 while (options && options->name) { 925 if (!strcmp(options->name, BLOCK_OPT_SIZE)) { 926 sectors = options->value.n / 512; 927 } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) { 928 backing_file = options->value.s; 929 } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FMT)) { 930 backing_fmt = options->value.s; 931 } else if (!strcmp(options->name, BLOCK_OPT_ENCRYPT)) { 932 flags |= options->value.n ? BLOCK_FLAG_ENCRYPT : 0; 933 } else if (!strcmp(options->name, BLOCK_OPT_CLUSTER_SIZE)) { 934 if (options->value.n) { 935 cluster_size = options->value.n; 936 } 937 } else if (!strcmp(options->name, BLOCK_OPT_PREALLOC)) { 938 if (!options->value.s || !strcmp(options->value.s, "off")) { 939 prealloc = 0; 940 } else if (!strcmp(options->value.s, "metadata")) { 941 prealloc = 1; 942 } else { 943 fprintf(stderr, "Invalid preallocation mode: '%s'\n", 944 options->value.s); 945 return -EINVAL; 946 } 947 } 948 options++; 949 } 950 951 if (backing_file && prealloc) { 952 fprintf(stderr, "Backing file and preallocation cannot be used at " 953 "the same time\n"); 954 return -EINVAL; 955 } 956 957 return qcow2_create2(filename, sectors, backing_file, backing_fmt, flags, 958 cluster_size, prealloc, options); 959 } 960 961 static int qcow2_make_empty(BlockDriverState *bs) 962 { 963 #if 0 964 /* XXX: not correct */ 965 BDRVQcowState *s = bs->opaque; 966 uint32_t l1_length = s->l1_size * sizeof(uint64_t); 967 int ret; 968 969 memset(s->l1_table, 0, l1_length); 970 if (bdrv_pwrite(bs->file, s->l1_table_offset, s->l1_table, l1_length) < 0) 971 return -1; 972 ret = bdrv_truncate(bs->file, s->l1_table_offset + l1_length); 973 if (ret < 0) 974 return ret; 975 976 l2_cache_reset(bs); 977 #endif 978 return 0; 979 } 980 981 static int qcow2_discard(BlockDriverState *bs, int64_t sector_num, 982 int nb_sectors) 983 { 984 return qcow2_discard_clusters(bs, sector_num << BDRV_SECTOR_BITS, 985 nb_sectors); 986 } 987 988 static int qcow2_truncate(BlockDriverState *bs, int64_t offset) 989 { 990 BDRVQcowState *s = bs->opaque; 991 int ret, new_l1_size; 992 993 if (offset & 511) { 994 return -EINVAL; 995 } 996 997 /* cannot proceed if image has snapshots */ 998 if (s->nb_snapshots) { 999 return -ENOTSUP; 1000 } 1001 1002 /* shrinking is currently not supported */ 1003 if (offset < bs->total_sectors * 512) { 1004 return -ENOTSUP; 1005 } 1006 1007 new_l1_size = size_to_l1(s, offset); 1008 ret = qcow2_grow_l1_table(bs, new_l1_size, true); 1009 if (ret < 0) { 1010 return ret; 1011 } 1012 1013 /* write updated header.size */ 1014 offset = cpu_to_be64(offset); 1015 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, size), 1016 &offset, sizeof(uint64_t)); 1017 if (ret < 0) { 1018 return ret; 1019 } 1020 1021 s->l1_vm_state_index = new_l1_size; 1022 return 0; 1023 } 1024 1025 /* XXX: put compressed sectors first, then all the cluster aligned 1026 tables to avoid losing bytes in alignment */ 1027 static int qcow2_write_compressed(BlockDriverState *bs, int64_t sector_num, 1028 const uint8_t *buf, int nb_sectors) 1029 { 1030 BDRVQcowState *s = bs->opaque; 1031 z_stream strm; 1032 int ret, out_len; 1033 uint8_t *out_buf; 1034 uint64_t cluster_offset; 1035 1036 if (nb_sectors == 0) { 1037 /* align end of file to a sector boundary to ease reading with 1038 sector based I/Os */ 1039 cluster_offset = bdrv_getlength(bs->file); 1040 cluster_offset = (cluster_offset + 511) & ~511; 1041 bdrv_truncate(bs->file, cluster_offset); 1042 return 0; 1043 } 1044 1045 if (nb_sectors != s->cluster_sectors) 1046 return -EINVAL; 1047 1048 out_buf = g_malloc(s->cluster_size + (s->cluster_size / 1000) + 128); 1049 1050 /* best compression, small window, no zlib header */ 1051 memset(&strm, 0, sizeof(strm)); 1052 ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION, 1053 Z_DEFLATED, -12, 1054 9, Z_DEFAULT_STRATEGY); 1055 if (ret != 0) { 1056 g_free(out_buf); 1057 return -1; 1058 } 1059 1060 strm.avail_in = s->cluster_size; 1061 strm.next_in = (uint8_t *)buf; 1062 strm.avail_out = s->cluster_size; 1063 strm.next_out = out_buf; 1064 1065 ret = deflate(&strm, Z_FINISH); 1066 if (ret != Z_STREAM_END && ret != Z_OK) { 1067 g_free(out_buf); 1068 deflateEnd(&strm); 1069 return -1; 1070 } 1071 out_len = strm.next_out - out_buf; 1072 1073 deflateEnd(&strm); 1074 1075 if (ret != Z_STREAM_END || out_len >= s->cluster_size) { 1076 /* could not compress: write normal cluster */ 1077 bdrv_write(bs, sector_num, buf, s->cluster_sectors); 1078 } else { 1079 cluster_offset = qcow2_alloc_compressed_cluster_offset(bs, 1080 sector_num << 9, out_len); 1081 if (!cluster_offset) 1082 return -1; 1083 cluster_offset &= s->cluster_offset_mask; 1084 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_COMPRESSED); 1085 if (bdrv_pwrite(bs->file, cluster_offset, out_buf, out_len) != out_len) { 1086 g_free(out_buf); 1087 return -1; 1088 } 1089 } 1090 1091 g_free(out_buf); 1092 return 0; 1093 } 1094 1095 static int qcow2_flush(BlockDriverState *bs) 1096 { 1097 BDRVQcowState *s = bs->opaque; 1098 int ret; 1099 1100 ret = qcow2_cache_flush(bs, s->l2_table_cache); 1101 if (ret < 0) { 1102 return ret; 1103 } 1104 1105 ret = qcow2_cache_flush(bs, s->refcount_block_cache); 1106 if (ret < 0) { 1107 return ret; 1108 } 1109 1110 return bdrv_flush(bs->file); 1111 } 1112 1113 static BlockDriverAIOCB *qcow2_aio_flush(BlockDriverState *bs, 1114 BlockDriverCompletionFunc *cb, 1115 void *opaque) 1116 { 1117 BDRVQcowState *s = bs->opaque; 1118 int ret; 1119 1120 ret = qcow2_cache_flush(bs, s->l2_table_cache); 1121 if (ret < 0) { 1122 return NULL; 1123 } 1124 1125 ret = qcow2_cache_flush(bs, s->refcount_block_cache); 1126 if (ret < 0) { 1127 return NULL; 1128 } 1129 1130 return bdrv_aio_flush(bs->file, cb, opaque); 1131 } 1132 1133 static int64_t qcow2_vm_state_offset(BDRVQcowState *s) 1134 { 1135 return (int64_t)s->l1_vm_state_index << (s->cluster_bits + s->l2_bits); 1136 } 1137 1138 static int qcow2_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) 1139 { 1140 BDRVQcowState *s = bs->opaque; 1141 bdi->cluster_size = s->cluster_size; 1142 bdi->vm_state_offset = qcow2_vm_state_offset(s); 1143 return 0; 1144 } 1145 1146 1147 static int qcow2_check(BlockDriverState *bs, BdrvCheckResult *result) 1148 { 1149 return qcow2_check_refcounts(bs, result); 1150 } 1151 1152 #if 0 1153 static void dump_refcounts(BlockDriverState *bs) 1154 { 1155 BDRVQcowState *s = bs->opaque; 1156 int64_t nb_clusters, k, k1, size; 1157 int refcount; 1158 1159 size = bdrv_getlength(bs->file); 1160 nb_clusters = size_to_clusters(s, size); 1161 for(k = 0; k < nb_clusters;) { 1162 k1 = k; 1163 refcount = get_refcount(bs, k); 1164 k++; 1165 while (k < nb_clusters && get_refcount(bs, k) == refcount) 1166 k++; 1167 printf("%" PRId64 ": refcount=%d nb=%" PRId64 "\n", k, refcount, 1168 k - k1); 1169 } 1170 } 1171 #endif 1172 1173 static int qcow2_save_vmstate(BlockDriverState *bs, const uint8_t *buf, 1174 int64_t pos, int size) 1175 { 1176 BDRVQcowState *s = bs->opaque; 1177 int growable = bs->growable; 1178 int ret; 1179 1180 BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_SAVE); 1181 bs->growable = 1; 1182 ret = bdrv_pwrite(bs, qcow2_vm_state_offset(s) + pos, buf, size); 1183 bs->growable = growable; 1184 1185 return ret; 1186 } 1187 1188 static int qcow2_load_vmstate(BlockDriverState *bs, uint8_t *buf, 1189 int64_t pos, int size) 1190 { 1191 BDRVQcowState *s = bs->opaque; 1192 int growable = bs->growable; 1193 int ret; 1194 1195 BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_LOAD); 1196 bs->growable = 1; 1197 ret = bdrv_pread(bs, qcow2_vm_state_offset(s) + pos, buf, size); 1198 bs->growable = growable; 1199 1200 return ret; 1201 } 1202 1203 static QEMUOptionParameter qcow2_create_options[] = { 1204 { 1205 .name = BLOCK_OPT_SIZE, 1206 .type = OPT_SIZE, 1207 .help = "Virtual disk size" 1208 }, 1209 { 1210 .name = BLOCK_OPT_BACKING_FILE, 1211 .type = OPT_STRING, 1212 .help = "File name of a base image" 1213 }, 1214 { 1215 .name = BLOCK_OPT_BACKING_FMT, 1216 .type = OPT_STRING, 1217 .help = "Image format of the base image" 1218 }, 1219 { 1220 .name = BLOCK_OPT_ENCRYPT, 1221 .type = OPT_FLAG, 1222 .help = "Encrypt the image" 1223 }, 1224 { 1225 .name = BLOCK_OPT_CLUSTER_SIZE, 1226 .type = OPT_SIZE, 1227 .help = "qcow2 cluster size", 1228 .value = { .n = DEFAULT_CLUSTER_SIZE }, 1229 }, 1230 { 1231 .name = BLOCK_OPT_PREALLOC, 1232 .type = OPT_STRING, 1233 .help = "Preallocation mode (allowed values: off, metadata)" 1234 }, 1235 { NULL } 1236 }; 1237 1238 static BlockDriver bdrv_qcow2 = { 1239 .format_name = "qcow2", 1240 .instance_size = sizeof(BDRVQcowState), 1241 .bdrv_probe = qcow2_probe, 1242 .bdrv_open = qcow2_open, 1243 .bdrv_close = qcow2_close, 1244 .bdrv_create = qcow2_create, 1245 .bdrv_flush = qcow2_flush, 1246 .bdrv_is_allocated = qcow2_is_allocated, 1247 .bdrv_set_key = qcow2_set_key, 1248 .bdrv_make_empty = qcow2_make_empty, 1249 1250 .bdrv_co_readv = qcow2_co_readv, 1251 .bdrv_co_writev = qcow2_co_writev, 1252 .bdrv_aio_flush = qcow2_aio_flush, 1253 1254 .bdrv_discard = qcow2_discard, 1255 .bdrv_truncate = qcow2_truncate, 1256 .bdrv_write_compressed = qcow2_write_compressed, 1257 1258 .bdrv_snapshot_create = qcow2_snapshot_create, 1259 .bdrv_snapshot_goto = qcow2_snapshot_goto, 1260 .bdrv_snapshot_delete = qcow2_snapshot_delete, 1261 .bdrv_snapshot_list = qcow2_snapshot_list, 1262 .bdrv_snapshot_load_tmp = qcow2_snapshot_load_tmp, 1263 .bdrv_get_info = qcow2_get_info, 1264 1265 .bdrv_save_vmstate = qcow2_save_vmstate, 1266 .bdrv_load_vmstate = qcow2_load_vmstate, 1267 1268 .bdrv_change_backing_file = qcow2_change_backing_file, 1269 1270 .create_options = qcow2_create_options, 1271 .bdrv_check = qcow2_check, 1272 }; 1273 1274 static void bdrv_qcow2_init(void) 1275 { 1276 bdrv_register(&bdrv_qcow2); 1277 } 1278 1279 block_init(bdrv_qcow2_init); 1280