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 "block/block_int.h" 28 #include "sysemu/block-backend.h" 29 #include "qemu/module.h" 30 #include <zlib.h> 31 #include "qapi/qmp/qerror.h" 32 #include "crypto/cipher.h" 33 #include "migration/migration.h" 34 35 /**************************************************************/ 36 /* QEMU COW block driver with compression and encryption support */ 37 38 #define QCOW_MAGIC (('Q' << 24) | ('F' << 16) | ('I' << 8) | 0xfb) 39 #define QCOW_VERSION 1 40 41 #define QCOW_CRYPT_NONE 0 42 #define QCOW_CRYPT_AES 1 43 44 #define QCOW_OFLAG_COMPRESSED (1LL << 63) 45 46 typedef struct QCowHeader { 47 uint32_t magic; 48 uint32_t version; 49 uint64_t backing_file_offset; 50 uint32_t backing_file_size; 51 uint32_t mtime; 52 uint64_t size; /* in bytes */ 53 uint8_t cluster_bits; 54 uint8_t l2_bits; 55 uint16_t padding; 56 uint32_t crypt_method; 57 uint64_t l1_table_offset; 58 } QEMU_PACKED QCowHeader; 59 60 #define L2_CACHE_SIZE 16 61 62 typedef struct BDRVQcowState { 63 int cluster_bits; 64 int cluster_size; 65 int cluster_sectors; 66 int l2_bits; 67 int l2_size; 68 unsigned int l1_size; 69 uint64_t cluster_offset_mask; 70 uint64_t l1_table_offset; 71 uint64_t *l1_table; 72 uint64_t *l2_cache; 73 uint64_t l2_cache_offsets[L2_CACHE_SIZE]; 74 uint32_t l2_cache_counts[L2_CACHE_SIZE]; 75 uint8_t *cluster_cache; 76 uint8_t *cluster_data; 77 uint64_t cluster_cache_offset; 78 QCryptoCipher *cipher; /* NULL if no key yet */ 79 uint32_t crypt_method_header; 80 CoMutex lock; 81 Error *migration_blocker; 82 } BDRVQcowState; 83 84 static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset); 85 86 static int qcow_probe(const uint8_t *buf, int buf_size, const char *filename) 87 { 88 const QCowHeader *cow_header = (const void *)buf; 89 90 if (buf_size >= sizeof(QCowHeader) && 91 be32_to_cpu(cow_header->magic) == QCOW_MAGIC && 92 be32_to_cpu(cow_header->version) == QCOW_VERSION) 93 return 100; 94 else 95 return 0; 96 } 97 98 static int qcow_open(BlockDriverState *bs, QDict *options, int flags, 99 Error **errp) 100 { 101 BDRVQcowState *s = bs->opaque; 102 unsigned int len, i, shift; 103 int ret; 104 QCowHeader header; 105 106 ret = bdrv_pread(bs->file->bs, 0, &header, sizeof(header)); 107 if (ret < 0) { 108 goto fail; 109 } 110 be32_to_cpus(&header.magic); 111 be32_to_cpus(&header.version); 112 be64_to_cpus(&header.backing_file_offset); 113 be32_to_cpus(&header.backing_file_size); 114 be32_to_cpus(&header.mtime); 115 be64_to_cpus(&header.size); 116 be32_to_cpus(&header.crypt_method); 117 be64_to_cpus(&header.l1_table_offset); 118 119 if (header.magic != QCOW_MAGIC) { 120 error_setg(errp, "Image not in qcow format"); 121 ret = -EINVAL; 122 goto fail; 123 } 124 if (header.version != QCOW_VERSION) { 125 error_setg(errp, "Unsupported qcow version %" PRIu32, header.version); 126 ret = -ENOTSUP; 127 goto fail; 128 } 129 130 if (header.size <= 1) { 131 error_setg(errp, "Image size is too small (must be at least 2 bytes)"); 132 ret = -EINVAL; 133 goto fail; 134 } 135 if (header.cluster_bits < 9 || header.cluster_bits > 16) { 136 error_setg(errp, "Cluster size must be between 512 and 64k"); 137 ret = -EINVAL; 138 goto fail; 139 } 140 141 /* l2_bits specifies number of entries; storing a uint64_t in each entry, 142 * so bytes = num_entries << 3. */ 143 if (header.l2_bits < 9 - 3 || header.l2_bits > 16 - 3) { 144 error_setg(errp, "L2 table size must be between 512 and 64k"); 145 ret = -EINVAL; 146 goto fail; 147 } 148 149 if (header.crypt_method > QCOW_CRYPT_AES) { 150 error_setg(errp, "invalid encryption method in qcow header"); 151 ret = -EINVAL; 152 goto fail; 153 } 154 if (!qcrypto_cipher_supports(QCRYPTO_CIPHER_ALG_AES_128)) { 155 error_setg(errp, "AES cipher not available"); 156 ret = -EINVAL; 157 goto fail; 158 } 159 s->crypt_method_header = header.crypt_method; 160 if (s->crypt_method_header) { 161 bs->encrypted = 1; 162 } 163 s->cluster_bits = header.cluster_bits; 164 s->cluster_size = 1 << s->cluster_bits; 165 s->cluster_sectors = 1 << (s->cluster_bits - 9); 166 s->l2_bits = header.l2_bits; 167 s->l2_size = 1 << s->l2_bits; 168 bs->total_sectors = header.size / 512; 169 s->cluster_offset_mask = (1LL << (63 - s->cluster_bits)) - 1; 170 171 /* read the level 1 table */ 172 shift = s->cluster_bits + s->l2_bits; 173 if (header.size > UINT64_MAX - (1LL << shift)) { 174 error_setg(errp, "Image too large"); 175 ret = -EINVAL; 176 goto fail; 177 } else { 178 uint64_t l1_size = (header.size + (1LL << shift) - 1) >> shift; 179 if (l1_size > INT_MAX / sizeof(uint64_t)) { 180 error_setg(errp, "Image too large"); 181 ret = -EINVAL; 182 goto fail; 183 } 184 s->l1_size = l1_size; 185 } 186 187 s->l1_table_offset = header.l1_table_offset; 188 s->l1_table = g_try_new(uint64_t, s->l1_size); 189 if (s->l1_table == NULL) { 190 error_setg(errp, "Could not allocate memory for L1 table"); 191 ret = -ENOMEM; 192 goto fail; 193 } 194 195 ret = bdrv_pread(bs->file->bs, s->l1_table_offset, s->l1_table, 196 s->l1_size * sizeof(uint64_t)); 197 if (ret < 0) { 198 goto fail; 199 } 200 201 for(i = 0;i < s->l1_size; i++) { 202 be64_to_cpus(&s->l1_table[i]); 203 } 204 205 /* alloc L2 cache (max. 64k * 16 * 8 = 8 MB) */ 206 s->l2_cache = 207 qemu_try_blockalign(bs->file->bs, 208 s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t)); 209 if (s->l2_cache == NULL) { 210 error_setg(errp, "Could not allocate L2 table cache"); 211 ret = -ENOMEM; 212 goto fail; 213 } 214 s->cluster_cache = g_malloc(s->cluster_size); 215 s->cluster_data = g_malloc(s->cluster_size); 216 s->cluster_cache_offset = -1; 217 218 /* read the backing file name */ 219 if (header.backing_file_offset != 0) { 220 len = header.backing_file_size; 221 if (len > 1023 || len >= sizeof(bs->backing_file)) { 222 error_setg(errp, "Backing file name too long"); 223 ret = -EINVAL; 224 goto fail; 225 } 226 ret = bdrv_pread(bs->file->bs, header.backing_file_offset, 227 bs->backing_file, len); 228 if (ret < 0) { 229 goto fail; 230 } 231 bs->backing_file[len] = '\0'; 232 } 233 234 /* Disable migration when qcow images are used */ 235 error_setg(&s->migration_blocker, "The qcow format used by node '%s' " 236 "does not support live migration", 237 bdrv_get_device_or_node_name(bs)); 238 migrate_add_blocker(s->migration_blocker); 239 240 qemu_co_mutex_init(&s->lock); 241 return 0; 242 243 fail: 244 g_free(s->l1_table); 245 qemu_vfree(s->l2_cache); 246 g_free(s->cluster_cache); 247 g_free(s->cluster_data); 248 return ret; 249 } 250 251 252 /* We have nothing to do for QCOW reopen, stubs just return 253 * success */ 254 static int qcow_reopen_prepare(BDRVReopenState *state, 255 BlockReopenQueue *queue, Error **errp) 256 { 257 return 0; 258 } 259 260 static int qcow_set_key(BlockDriverState *bs, const char *key) 261 { 262 BDRVQcowState *s = bs->opaque; 263 uint8_t keybuf[16]; 264 int len, i; 265 Error *err; 266 267 memset(keybuf, 0, 16); 268 len = strlen(key); 269 if (len > 16) 270 len = 16; 271 /* XXX: we could compress the chars to 7 bits to increase 272 entropy */ 273 for(i = 0;i < len;i++) { 274 keybuf[i] = key[i]; 275 } 276 assert(bs->encrypted); 277 278 qcrypto_cipher_free(s->cipher); 279 s->cipher = qcrypto_cipher_new( 280 QCRYPTO_CIPHER_ALG_AES_128, 281 QCRYPTO_CIPHER_MODE_CBC, 282 keybuf, G_N_ELEMENTS(keybuf), 283 &err); 284 285 if (!s->cipher) { 286 /* XXX would be nice if errors in this method could 287 * be properly propagate to the caller. Would need 288 * the bdrv_set_key() API signature to be fixed. */ 289 error_free(err); 290 return -1; 291 } 292 return 0; 293 } 294 295 /* The crypt function is compatible with the linux cryptoloop 296 algorithm for < 4 GB images. NOTE: out_buf == in_buf is 297 supported */ 298 static int encrypt_sectors(BDRVQcowState *s, int64_t sector_num, 299 uint8_t *out_buf, const uint8_t *in_buf, 300 int nb_sectors, bool enc, Error **errp) 301 { 302 union { 303 uint64_t ll[2]; 304 uint8_t b[16]; 305 } ivec; 306 int i; 307 int ret; 308 309 for(i = 0; i < nb_sectors; i++) { 310 ivec.ll[0] = cpu_to_le64(sector_num); 311 ivec.ll[1] = 0; 312 if (qcrypto_cipher_setiv(s->cipher, 313 ivec.b, G_N_ELEMENTS(ivec.b), 314 errp) < 0) { 315 return -1; 316 } 317 if (enc) { 318 ret = qcrypto_cipher_encrypt(s->cipher, 319 in_buf, 320 out_buf, 321 512, 322 errp); 323 } else { 324 ret = qcrypto_cipher_decrypt(s->cipher, 325 in_buf, 326 out_buf, 327 512, 328 errp); 329 } 330 if (ret < 0) { 331 return -1; 332 } 333 sector_num++; 334 in_buf += 512; 335 out_buf += 512; 336 } 337 return 0; 338 } 339 340 /* 'allocate' is: 341 * 342 * 0 to not allocate. 343 * 344 * 1 to allocate a normal cluster (for sector indexes 'n_start' to 345 * 'n_end') 346 * 347 * 2 to allocate a compressed cluster of size 348 * 'compressed_size'. 'compressed_size' must be > 0 and < 349 * cluster_size 350 * 351 * return 0 if not allocated. 352 */ 353 static uint64_t get_cluster_offset(BlockDriverState *bs, 354 uint64_t offset, int allocate, 355 int compressed_size, 356 int n_start, int n_end) 357 { 358 BDRVQcowState *s = bs->opaque; 359 int min_index, i, j, l1_index, l2_index; 360 uint64_t l2_offset, *l2_table, cluster_offset, tmp; 361 uint32_t min_count; 362 int new_l2_table; 363 364 l1_index = offset >> (s->l2_bits + s->cluster_bits); 365 l2_offset = s->l1_table[l1_index]; 366 new_l2_table = 0; 367 if (!l2_offset) { 368 if (!allocate) 369 return 0; 370 /* allocate a new l2 entry */ 371 l2_offset = bdrv_getlength(bs->file->bs); 372 /* round to cluster size */ 373 l2_offset = (l2_offset + s->cluster_size - 1) & ~(s->cluster_size - 1); 374 /* update the L1 entry */ 375 s->l1_table[l1_index] = l2_offset; 376 tmp = cpu_to_be64(l2_offset); 377 if (bdrv_pwrite_sync(bs->file->bs, 378 s->l1_table_offset + l1_index * sizeof(tmp), 379 &tmp, sizeof(tmp)) < 0) 380 return 0; 381 new_l2_table = 1; 382 } 383 for(i = 0; i < L2_CACHE_SIZE; i++) { 384 if (l2_offset == s->l2_cache_offsets[i]) { 385 /* increment the hit count */ 386 if (++s->l2_cache_counts[i] == 0xffffffff) { 387 for(j = 0; j < L2_CACHE_SIZE; j++) { 388 s->l2_cache_counts[j] >>= 1; 389 } 390 } 391 l2_table = s->l2_cache + (i << s->l2_bits); 392 goto found; 393 } 394 } 395 /* not found: load a new entry in the least used one */ 396 min_index = 0; 397 min_count = 0xffffffff; 398 for(i = 0; i < L2_CACHE_SIZE; i++) { 399 if (s->l2_cache_counts[i] < min_count) { 400 min_count = s->l2_cache_counts[i]; 401 min_index = i; 402 } 403 } 404 l2_table = s->l2_cache + (min_index << s->l2_bits); 405 if (new_l2_table) { 406 memset(l2_table, 0, s->l2_size * sizeof(uint64_t)); 407 if (bdrv_pwrite_sync(bs->file->bs, l2_offset, l2_table, 408 s->l2_size * sizeof(uint64_t)) < 0) 409 return 0; 410 } else { 411 if (bdrv_pread(bs->file->bs, l2_offset, l2_table, 412 s->l2_size * sizeof(uint64_t)) != 413 s->l2_size * sizeof(uint64_t)) 414 return 0; 415 } 416 s->l2_cache_offsets[min_index] = l2_offset; 417 s->l2_cache_counts[min_index] = 1; 418 found: 419 l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1); 420 cluster_offset = be64_to_cpu(l2_table[l2_index]); 421 if (!cluster_offset || 422 ((cluster_offset & QCOW_OFLAG_COMPRESSED) && allocate == 1)) { 423 if (!allocate) 424 return 0; 425 /* allocate a new cluster */ 426 if ((cluster_offset & QCOW_OFLAG_COMPRESSED) && 427 (n_end - n_start) < s->cluster_sectors) { 428 /* if the cluster is already compressed, we must 429 decompress it in the case it is not completely 430 overwritten */ 431 if (decompress_cluster(bs, cluster_offset) < 0) 432 return 0; 433 cluster_offset = bdrv_getlength(bs->file->bs); 434 cluster_offset = (cluster_offset + s->cluster_size - 1) & 435 ~(s->cluster_size - 1); 436 /* write the cluster content */ 437 if (bdrv_pwrite(bs->file->bs, cluster_offset, s->cluster_cache, 438 s->cluster_size) != 439 s->cluster_size) 440 return -1; 441 } else { 442 cluster_offset = bdrv_getlength(bs->file->bs); 443 if (allocate == 1) { 444 /* round to cluster size */ 445 cluster_offset = (cluster_offset + s->cluster_size - 1) & 446 ~(s->cluster_size - 1); 447 bdrv_truncate(bs->file->bs, cluster_offset + s->cluster_size); 448 /* if encrypted, we must initialize the cluster 449 content which won't be written */ 450 if (bs->encrypted && 451 (n_end - n_start) < s->cluster_sectors) { 452 uint64_t start_sect; 453 assert(s->cipher); 454 start_sect = (offset & ~(s->cluster_size - 1)) >> 9; 455 memset(s->cluster_data + 512, 0x00, 512); 456 for(i = 0; i < s->cluster_sectors; i++) { 457 if (i < n_start || i >= n_end) { 458 Error *err = NULL; 459 if (encrypt_sectors(s, start_sect + i, 460 s->cluster_data, 461 s->cluster_data + 512, 1, 462 true, &err) < 0) { 463 error_free(err); 464 errno = EIO; 465 return -1; 466 } 467 if (bdrv_pwrite(bs->file->bs, 468 cluster_offset + i * 512, 469 s->cluster_data, 512) != 512) 470 return -1; 471 } 472 } 473 } 474 } else if (allocate == 2) { 475 cluster_offset |= QCOW_OFLAG_COMPRESSED | 476 (uint64_t)compressed_size << (63 - s->cluster_bits); 477 } 478 } 479 /* update L2 table */ 480 tmp = cpu_to_be64(cluster_offset); 481 l2_table[l2_index] = tmp; 482 if (bdrv_pwrite_sync(bs->file->bs, l2_offset + l2_index * sizeof(tmp), 483 &tmp, sizeof(tmp)) < 0) 484 return 0; 485 } 486 return cluster_offset; 487 } 488 489 static int64_t coroutine_fn qcow_co_get_block_status(BlockDriverState *bs, 490 int64_t sector_num, int nb_sectors, int *pnum, BlockDriverState **file) 491 { 492 BDRVQcowState *s = bs->opaque; 493 int index_in_cluster, n; 494 uint64_t cluster_offset; 495 496 qemu_co_mutex_lock(&s->lock); 497 cluster_offset = get_cluster_offset(bs, sector_num << 9, 0, 0, 0, 0); 498 qemu_co_mutex_unlock(&s->lock); 499 index_in_cluster = sector_num & (s->cluster_sectors - 1); 500 n = s->cluster_sectors - index_in_cluster; 501 if (n > nb_sectors) 502 n = nb_sectors; 503 *pnum = n; 504 if (!cluster_offset) { 505 return 0; 506 } 507 if ((cluster_offset & QCOW_OFLAG_COMPRESSED) || s->cipher) { 508 return BDRV_BLOCK_DATA; 509 } 510 cluster_offset |= (index_in_cluster << BDRV_SECTOR_BITS); 511 *file = bs->file->bs; 512 return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID | cluster_offset; 513 } 514 515 static int decompress_buffer(uint8_t *out_buf, int out_buf_size, 516 const uint8_t *buf, int buf_size) 517 { 518 z_stream strm1, *strm = &strm1; 519 int ret, out_len; 520 521 memset(strm, 0, sizeof(*strm)); 522 523 strm->next_in = (uint8_t *)buf; 524 strm->avail_in = buf_size; 525 strm->next_out = out_buf; 526 strm->avail_out = out_buf_size; 527 528 ret = inflateInit2(strm, -12); 529 if (ret != Z_OK) 530 return -1; 531 ret = inflate(strm, Z_FINISH); 532 out_len = strm->next_out - out_buf; 533 if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) || 534 out_len != out_buf_size) { 535 inflateEnd(strm); 536 return -1; 537 } 538 inflateEnd(strm); 539 return 0; 540 } 541 542 static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset) 543 { 544 BDRVQcowState *s = bs->opaque; 545 int ret, csize; 546 uint64_t coffset; 547 548 coffset = cluster_offset & s->cluster_offset_mask; 549 if (s->cluster_cache_offset != coffset) { 550 csize = cluster_offset >> (63 - s->cluster_bits); 551 csize &= (s->cluster_size - 1); 552 ret = bdrv_pread(bs->file->bs, coffset, s->cluster_data, csize); 553 if (ret != csize) 554 return -1; 555 if (decompress_buffer(s->cluster_cache, s->cluster_size, 556 s->cluster_data, csize) < 0) { 557 return -1; 558 } 559 s->cluster_cache_offset = coffset; 560 } 561 return 0; 562 } 563 564 static coroutine_fn int qcow_co_readv(BlockDriverState *bs, int64_t sector_num, 565 int nb_sectors, QEMUIOVector *qiov) 566 { 567 BDRVQcowState *s = bs->opaque; 568 int index_in_cluster; 569 int ret = 0, n; 570 uint64_t cluster_offset; 571 struct iovec hd_iov; 572 QEMUIOVector hd_qiov; 573 uint8_t *buf; 574 void *orig_buf; 575 Error *err = NULL; 576 577 if (qiov->niov > 1) { 578 buf = orig_buf = qemu_try_blockalign(bs, qiov->size); 579 if (buf == NULL) { 580 return -ENOMEM; 581 } 582 } else { 583 orig_buf = NULL; 584 buf = (uint8_t *)qiov->iov->iov_base; 585 } 586 587 qemu_co_mutex_lock(&s->lock); 588 589 while (nb_sectors != 0) { 590 /* prepare next request */ 591 cluster_offset = get_cluster_offset(bs, sector_num << 9, 592 0, 0, 0, 0); 593 index_in_cluster = sector_num & (s->cluster_sectors - 1); 594 n = s->cluster_sectors - index_in_cluster; 595 if (n > nb_sectors) { 596 n = nb_sectors; 597 } 598 599 if (!cluster_offset) { 600 if (bs->backing) { 601 /* read from the base image */ 602 hd_iov.iov_base = (void *)buf; 603 hd_iov.iov_len = n * 512; 604 qemu_iovec_init_external(&hd_qiov, &hd_iov, 1); 605 qemu_co_mutex_unlock(&s->lock); 606 ret = bdrv_co_readv(bs->backing->bs, sector_num, 607 n, &hd_qiov); 608 qemu_co_mutex_lock(&s->lock); 609 if (ret < 0) { 610 goto fail; 611 } 612 } else { 613 /* Note: in this case, no need to wait */ 614 memset(buf, 0, 512 * n); 615 } 616 } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) { 617 /* add AIO support for compressed blocks ? */ 618 if (decompress_cluster(bs, cluster_offset) < 0) { 619 goto fail; 620 } 621 memcpy(buf, 622 s->cluster_cache + index_in_cluster * 512, 512 * n); 623 } else { 624 if ((cluster_offset & 511) != 0) { 625 goto fail; 626 } 627 hd_iov.iov_base = (void *)buf; 628 hd_iov.iov_len = n * 512; 629 qemu_iovec_init_external(&hd_qiov, &hd_iov, 1); 630 qemu_co_mutex_unlock(&s->lock); 631 ret = bdrv_co_readv(bs->file->bs, 632 (cluster_offset >> 9) + index_in_cluster, 633 n, &hd_qiov); 634 qemu_co_mutex_lock(&s->lock); 635 if (ret < 0) { 636 break; 637 } 638 if (bs->encrypted) { 639 assert(s->cipher); 640 if (encrypt_sectors(s, sector_num, buf, buf, 641 n, false, &err) < 0) { 642 goto fail; 643 } 644 } 645 } 646 ret = 0; 647 648 nb_sectors -= n; 649 sector_num += n; 650 buf += n * 512; 651 } 652 653 done: 654 qemu_co_mutex_unlock(&s->lock); 655 656 if (qiov->niov > 1) { 657 qemu_iovec_from_buf(qiov, 0, orig_buf, qiov->size); 658 qemu_vfree(orig_buf); 659 } 660 661 return ret; 662 663 fail: 664 error_free(err); 665 ret = -EIO; 666 goto done; 667 } 668 669 static coroutine_fn int qcow_co_writev(BlockDriverState *bs, int64_t sector_num, 670 int nb_sectors, QEMUIOVector *qiov) 671 { 672 BDRVQcowState *s = bs->opaque; 673 int index_in_cluster; 674 uint64_t cluster_offset; 675 const uint8_t *src_buf; 676 int ret = 0, n; 677 uint8_t *cluster_data = NULL; 678 struct iovec hd_iov; 679 QEMUIOVector hd_qiov; 680 uint8_t *buf; 681 void *orig_buf; 682 683 s->cluster_cache_offset = -1; /* disable compressed cache */ 684 685 if (qiov->niov > 1) { 686 buf = orig_buf = qemu_try_blockalign(bs, qiov->size); 687 if (buf == NULL) { 688 return -ENOMEM; 689 } 690 qemu_iovec_to_buf(qiov, 0, buf, qiov->size); 691 } else { 692 orig_buf = NULL; 693 buf = (uint8_t *)qiov->iov->iov_base; 694 } 695 696 qemu_co_mutex_lock(&s->lock); 697 698 while (nb_sectors != 0) { 699 700 index_in_cluster = sector_num & (s->cluster_sectors - 1); 701 n = s->cluster_sectors - index_in_cluster; 702 if (n > nb_sectors) { 703 n = nb_sectors; 704 } 705 cluster_offset = get_cluster_offset(bs, sector_num << 9, 1, 0, 706 index_in_cluster, 707 index_in_cluster + n); 708 if (!cluster_offset || (cluster_offset & 511) != 0) { 709 ret = -EIO; 710 break; 711 } 712 if (bs->encrypted) { 713 Error *err = NULL; 714 assert(s->cipher); 715 if (!cluster_data) { 716 cluster_data = g_malloc0(s->cluster_size); 717 } 718 if (encrypt_sectors(s, sector_num, cluster_data, buf, 719 n, true, &err) < 0) { 720 error_free(err); 721 ret = -EIO; 722 break; 723 } 724 src_buf = cluster_data; 725 } else { 726 src_buf = buf; 727 } 728 729 hd_iov.iov_base = (void *)src_buf; 730 hd_iov.iov_len = n * 512; 731 qemu_iovec_init_external(&hd_qiov, &hd_iov, 1); 732 qemu_co_mutex_unlock(&s->lock); 733 ret = bdrv_co_writev(bs->file->bs, 734 (cluster_offset >> 9) + index_in_cluster, 735 n, &hd_qiov); 736 qemu_co_mutex_lock(&s->lock); 737 if (ret < 0) { 738 break; 739 } 740 ret = 0; 741 742 nb_sectors -= n; 743 sector_num += n; 744 buf += n * 512; 745 } 746 qemu_co_mutex_unlock(&s->lock); 747 748 if (qiov->niov > 1) { 749 qemu_vfree(orig_buf); 750 } 751 g_free(cluster_data); 752 753 return ret; 754 } 755 756 static void qcow_close(BlockDriverState *bs) 757 { 758 BDRVQcowState *s = bs->opaque; 759 760 qcrypto_cipher_free(s->cipher); 761 s->cipher = NULL; 762 g_free(s->l1_table); 763 qemu_vfree(s->l2_cache); 764 g_free(s->cluster_cache); 765 g_free(s->cluster_data); 766 767 migrate_del_blocker(s->migration_blocker); 768 error_free(s->migration_blocker); 769 } 770 771 static int qcow_create(const char *filename, QemuOpts *opts, Error **errp) 772 { 773 int header_size, backing_filename_len, l1_size, shift, i; 774 QCowHeader header; 775 uint8_t *tmp; 776 int64_t total_size = 0; 777 char *backing_file = NULL; 778 int flags = 0; 779 Error *local_err = NULL; 780 int ret; 781 BlockBackend *qcow_blk; 782 783 /* Read out options */ 784 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0), 785 BDRV_SECTOR_SIZE); 786 backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE); 787 if (qemu_opt_get_bool_del(opts, BLOCK_OPT_ENCRYPT, false)) { 788 flags |= BLOCK_FLAG_ENCRYPT; 789 } 790 791 ret = bdrv_create_file(filename, opts, &local_err); 792 if (ret < 0) { 793 error_propagate(errp, local_err); 794 goto cleanup; 795 } 796 797 qcow_blk = blk_new_open(filename, NULL, NULL, 798 BDRV_O_RDWR | BDRV_O_CACHE_WB | BDRV_O_PROTOCOL, 799 &local_err); 800 if (qcow_blk == NULL) { 801 error_propagate(errp, local_err); 802 ret = -EIO; 803 goto cleanup; 804 } 805 806 blk_set_allow_write_beyond_eof(qcow_blk, true); 807 808 ret = blk_truncate(qcow_blk, 0); 809 if (ret < 0) { 810 goto exit; 811 } 812 813 memset(&header, 0, sizeof(header)); 814 header.magic = cpu_to_be32(QCOW_MAGIC); 815 header.version = cpu_to_be32(QCOW_VERSION); 816 header.size = cpu_to_be64(total_size); 817 header_size = sizeof(header); 818 backing_filename_len = 0; 819 if (backing_file) { 820 if (strcmp(backing_file, "fat:")) { 821 header.backing_file_offset = cpu_to_be64(header_size); 822 backing_filename_len = strlen(backing_file); 823 header.backing_file_size = cpu_to_be32(backing_filename_len); 824 header_size += backing_filename_len; 825 } else { 826 /* special backing file for vvfat */ 827 backing_file = NULL; 828 } 829 header.cluster_bits = 9; /* 512 byte cluster to avoid copying 830 unmodified sectors */ 831 header.l2_bits = 12; /* 32 KB L2 tables */ 832 } else { 833 header.cluster_bits = 12; /* 4 KB clusters */ 834 header.l2_bits = 9; /* 4 KB L2 tables */ 835 } 836 header_size = (header_size + 7) & ~7; 837 shift = header.cluster_bits + header.l2_bits; 838 l1_size = (total_size + (1LL << shift) - 1) >> shift; 839 840 header.l1_table_offset = cpu_to_be64(header_size); 841 if (flags & BLOCK_FLAG_ENCRYPT) { 842 header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES); 843 } else { 844 header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE); 845 } 846 847 /* write all the data */ 848 ret = blk_pwrite(qcow_blk, 0, &header, sizeof(header)); 849 if (ret != sizeof(header)) { 850 goto exit; 851 } 852 853 if (backing_file) { 854 ret = blk_pwrite(qcow_blk, sizeof(header), 855 backing_file, backing_filename_len); 856 if (ret != backing_filename_len) { 857 goto exit; 858 } 859 } 860 861 tmp = g_malloc0(BDRV_SECTOR_SIZE); 862 for (i = 0; i < ((sizeof(uint64_t)*l1_size + BDRV_SECTOR_SIZE - 1)/ 863 BDRV_SECTOR_SIZE); i++) { 864 ret = blk_pwrite(qcow_blk, header_size + 865 BDRV_SECTOR_SIZE*i, tmp, BDRV_SECTOR_SIZE); 866 if (ret != BDRV_SECTOR_SIZE) { 867 g_free(tmp); 868 goto exit; 869 } 870 } 871 872 g_free(tmp); 873 ret = 0; 874 exit: 875 blk_unref(qcow_blk); 876 cleanup: 877 g_free(backing_file); 878 return ret; 879 } 880 881 static int qcow_make_empty(BlockDriverState *bs) 882 { 883 BDRVQcowState *s = bs->opaque; 884 uint32_t l1_length = s->l1_size * sizeof(uint64_t); 885 int ret; 886 887 memset(s->l1_table, 0, l1_length); 888 if (bdrv_pwrite_sync(bs->file->bs, s->l1_table_offset, s->l1_table, 889 l1_length) < 0) 890 return -1; 891 ret = bdrv_truncate(bs->file->bs, s->l1_table_offset + l1_length); 892 if (ret < 0) 893 return ret; 894 895 memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t)); 896 memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t)); 897 memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t)); 898 899 return 0; 900 } 901 902 /* XXX: put compressed sectors first, then all the cluster aligned 903 tables to avoid losing bytes in alignment */ 904 static int qcow_write_compressed(BlockDriverState *bs, int64_t sector_num, 905 const uint8_t *buf, int nb_sectors) 906 { 907 BDRVQcowState *s = bs->opaque; 908 z_stream strm; 909 int ret, out_len; 910 uint8_t *out_buf; 911 uint64_t cluster_offset; 912 913 if (nb_sectors != s->cluster_sectors) { 914 ret = -EINVAL; 915 916 /* Zero-pad last write if image size is not cluster aligned */ 917 if (sector_num + nb_sectors == bs->total_sectors && 918 nb_sectors < s->cluster_sectors) { 919 uint8_t *pad_buf = qemu_blockalign(bs, s->cluster_size); 920 memset(pad_buf, 0, s->cluster_size); 921 memcpy(pad_buf, buf, nb_sectors * BDRV_SECTOR_SIZE); 922 ret = qcow_write_compressed(bs, sector_num, 923 pad_buf, s->cluster_sectors); 924 qemu_vfree(pad_buf); 925 } 926 return ret; 927 } 928 929 out_buf = g_malloc(s->cluster_size + (s->cluster_size / 1000) + 128); 930 931 /* best compression, small window, no zlib header */ 932 memset(&strm, 0, sizeof(strm)); 933 ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION, 934 Z_DEFLATED, -12, 935 9, Z_DEFAULT_STRATEGY); 936 if (ret != 0) { 937 ret = -EINVAL; 938 goto fail; 939 } 940 941 strm.avail_in = s->cluster_size; 942 strm.next_in = (uint8_t *)buf; 943 strm.avail_out = s->cluster_size; 944 strm.next_out = out_buf; 945 946 ret = deflate(&strm, Z_FINISH); 947 if (ret != Z_STREAM_END && ret != Z_OK) { 948 deflateEnd(&strm); 949 ret = -EINVAL; 950 goto fail; 951 } 952 out_len = strm.next_out - out_buf; 953 954 deflateEnd(&strm); 955 956 if (ret != Z_STREAM_END || out_len >= s->cluster_size) { 957 /* could not compress: write normal cluster */ 958 ret = bdrv_write(bs, sector_num, buf, s->cluster_sectors); 959 if (ret < 0) { 960 goto fail; 961 } 962 } else { 963 cluster_offset = get_cluster_offset(bs, sector_num << 9, 2, 964 out_len, 0, 0); 965 if (cluster_offset == 0) { 966 ret = -EIO; 967 goto fail; 968 } 969 970 cluster_offset &= s->cluster_offset_mask; 971 ret = bdrv_pwrite(bs->file->bs, cluster_offset, out_buf, out_len); 972 if (ret < 0) { 973 goto fail; 974 } 975 } 976 977 ret = 0; 978 fail: 979 g_free(out_buf); 980 return ret; 981 } 982 983 static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) 984 { 985 BDRVQcowState *s = bs->opaque; 986 bdi->cluster_size = s->cluster_size; 987 return 0; 988 } 989 990 static QemuOptsList qcow_create_opts = { 991 .name = "qcow-create-opts", 992 .head = QTAILQ_HEAD_INITIALIZER(qcow_create_opts.head), 993 .desc = { 994 { 995 .name = BLOCK_OPT_SIZE, 996 .type = QEMU_OPT_SIZE, 997 .help = "Virtual disk size" 998 }, 999 { 1000 .name = BLOCK_OPT_BACKING_FILE, 1001 .type = QEMU_OPT_STRING, 1002 .help = "File name of a base image" 1003 }, 1004 { 1005 .name = BLOCK_OPT_ENCRYPT, 1006 .type = QEMU_OPT_BOOL, 1007 .help = "Encrypt the image", 1008 .def_value_str = "off" 1009 }, 1010 { /* end of list */ } 1011 } 1012 }; 1013 1014 static BlockDriver bdrv_qcow = { 1015 .format_name = "qcow", 1016 .instance_size = sizeof(BDRVQcowState), 1017 .bdrv_probe = qcow_probe, 1018 .bdrv_open = qcow_open, 1019 .bdrv_close = qcow_close, 1020 .bdrv_reopen_prepare = qcow_reopen_prepare, 1021 .bdrv_create = qcow_create, 1022 .bdrv_has_zero_init = bdrv_has_zero_init_1, 1023 .supports_backing = true, 1024 1025 .bdrv_co_readv = qcow_co_readv, 1026 .bdrv_co_writev = qcow_co_writev, 1027 .bdrv_co_get_block_status = qcow_co_get_block_status, 1028 1029 .bdrv_set_key = qcow_set_key, 1030 .bdrv_make_empty = qcow_make_empty, 1031 .bdrv_write_compressed = qcow_write_compressed, 1032 .bdrv_get_info = qcow_get_info, 1033 1034 .create_opts = &qcow_create_opts, 1035 }; 1036 1037 static void bdrv_qcow_init(void) 1038 { 1039 bdrv_register(&bdrv_qcow); 1040 } 1041 1042 block_init(bdrv_qcow_init); 1043