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 25 #include "qemu/osdep.h" 26 #include "qapi/error.h" 27 #include "qemu/error-report.h" 28 #include "block/block_int.h" 29 #include "block/qdict.h" 30 #include "sysemu/block-backend.h" 31 #include "qemu/module.h" 32 #include "qemu/option.h" 33 #include "qemu/bswap.h" 34 #include "qemu/cutils.h" 35 #include <zlib.h> 36 #include "qapi/qmp/qdict.h" 37 #include "qapi/qmp/qstring.h" 38 #include "qapi/qobject-input-visitor.h" 39 #include "qapi/qapi-visit-block-core.h" 40 #include "crypto/block.h" 41 #include "migration/blocker.h" 42 #include "crypto.h" 43 44 /**************************************************************/ 45 /* QEMU COW block driver with compression and encryption support */ 46 47 #define QCOW_MAGIC (('Q' << 24) | ('F' << 16) | ('I' << 8) | 0xfb) 48 #define QCOW_VERSION 1 49 50 #define QCOW_CRYPT_NONE 0 51 #define QCOW_CRYPT_AES 1 52 53 #define QCOW_OFLAG_COMPRESSED (1LL << 63) 54 55 typedef struct QCowHeader { 56 uint32_t magic; 57 uint32_t version; 58 uint64_t backing_file_offset; 59 uint32_t backing_file_size; 60 uint32_t mtime; 61 uint64_t size; /* in bytes */ 62 uint8_t cluster_bits; 63 uint8_t l2_bits; 64 uint16_t padding; 65 uint32_t crypt_method; 66 uint64_t l1_table_offset; 67 } QEMU_PACKED QCowHeader; 68 69 #define L2_CACHE_SIZE 16 70 71 typedef struct BDRVQcowState { 72 int cluster_bits; 73 int cluster_size; 74 int l2_bits; 75 int l2_size; 76 unsigned int l1_size; 77 uint64_t cluster_offset_mask; 78 uint64_t l1_table_offset; 79 uint64_t *l1_table; 80 uint64_t *l2_cache; 81 uint64_t l2_cache_offsets[L2_CACHE_SIZE]; 82 uint32_t l2_cache_counts[L2_CACHE_SIZE]; 83 uint8_t *cluster_cache; 84 uint8_t *cluster_data; 85 uint64_t cluster_cache_offset; 86 QCryptoBlock *crypto; /* Disk encryption format driver */ 87 uint32_t crypt_method_header; 88 CoMutex lock; 89 Error *migration_blocker; 90 } BDRVQcowState; 91 92 static QemuOptsList qcow_create_opts; 93 94 static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset); 95 96 static int qcow_probe(const uint8_t *buf, int buf_size, const char *filename) 97 { 98 const QCowHeader *cow_header = (const void *)buf; 99 100 if (buf_size >= sizeof(QCowHeader) && 101 be32_to_cpu(cow_header->magic) == QCOW_MAGIC && 102 be32_to_cpu(cow_header->version) == QCOW_VERSION) 103 return 100; 104 else 105 return 0; 106 } 107 108 static QemuOptsList qcow_runtime_opts = { 109 .name = "qcow", 110 .head = QTAILQ_HEAD_INITIALIZER(qcow_runtime_opts.head), 111 .desc = { 112 BLOCK_CRYPTO_OPT_DEF_QCOW_KEY_SECRET("encrypt."), 113 { /* end of list */ } 114 }, 115 }; 116 117 static int qcow_open(BlockDriverState *bs, QDict *options, int flags, 118 Error **errp) 119 { 120 BDRVQcowState *s = bs->opaque; 121 unsigned int len, i, shift; 122 int ret; 123 QCowHeader header; 124 Error *local_err = NULL; 125 QCryptoBlockOpenOptions *crypto_opts = NULL; 126 unsigned int cflags = 0; 127 QDict *encryptopts = NULL; 128 const char *encryptfmt; 129 130 qdict_extract_subqdict(options, &encryptopts, "encrypt."); 131 encryptfmt = qdict_get_try_str(encryptopts, "format"); 132 133 bs->file = bdrv_open_child(NULL, options, "file", bs, &child_file, 134 false, errp); 135 if (!bs->file) { 136 ret = -EINVAL; 137 goto fail; 138 } 139 140 ret = bdrv_pread(bs->file, 0, &header, sizeof(header)); 141 if (ret < 0) { 142 goto fail; 143 } 144 header.magic = be32_to_cpu(header.magic); 145 header.version = be32_to_cpu(header.version); 146 header.backing_file_offset = be64_to_cpu(header.backing_file_offset); 147 header.backing_file_size = be32_to_cpu(header.backing_file_size); 148 header.mtime = be32_to_cpu(header.mtime); 149 header.size = be64_to_cpu(header.size); 150 header.crypt_method = be32_to_cpu(header.crypt_method); 151 header.l1_table_offset = be64_to_cpu(header.l1_table_offset); 152 153 if (header.magic != QCOW_MAGIC) { 154 error_setg(errp, "Image not in qcow format"); 155 ret = -EINVAL; 156 goto fail; 157 } 158 if (header.version != QCOW_VERSION) { 159 error_setg(errp, "Unsupported qcow version %" PRIu32, header.version); 160 ret = -ENOTSUP; 161 goto fail; 162 } 163 164 if (header.size <= 1) { 165 error_setg(errp, "Image size is too small (must be at least 2 bytes)"); 166 ret = -EINVAL; 167 goto fail; 168 } 169 if (header.cluster_bits < 9 || header.cluster_bits > 16) { 170 error_setg(errp, "Cluster size must be between 512 and 64k"); 171 ret = -EINVAL; 172 goto fail; 173 } 174 175 /* l2_bits specifies number of entries; storing a uint64_t in each entry, 176 * so bytes = num_entries << 3. */ 177 if (header.l2_bits < 9 - 3 || header.l2_bits > 16 - 3) { 178 error_setg(errp, "L2 table size must be between 512 and 64k"); 179 ret = -EINVAL; 180 goto fail; 181 } 182 183 s->crypt_method_header = header.crypt_method; 184 if (s->crypt_method_header) { 185 if (bdrv_uses_whitelist() && 186 s->crypt_method_header == QCOW_CRYPT_AES) { 187 error_setg(errp, 188 "Use of AES-CBC encrypted qcow images is no longer " 189 "supported in system emulators"); 190 error_append_hint(errp, 191 "You can use 'qemu-img convert' to convert your " 192 "image to an alternative supported format, such " 193 "as unencrypted qcow, or raw with the LUKS " 194 "format instead.\n"); 195 ret = -ENOSYS; 196 goto fail; 197 } 198 if (s->crypt_method_header == QCOW_CRYPT_AES) { 199 if (encryptfmt && !g_str_equal(encryptfmt, "aes")) { 200 error_setg(errp, 201 "Header reported 'aes' encryption format but " 202 "options specify '%s'", encryptfmt); 203 ret = -EINVAL; 204 goto fail; 205 } 206 qdict_put_str(encryptopts, "format", "qcow"); 207 crypto_opts = block_crypto_open_opts_init(encryptopts, errp); 208 if (!crypto_opts) { 209 ret = -EINVAL; 210 goto fail; 211 } 212 213 if (flags & BDRV_O_NO_IO) { 214 cflags |= QCRYPTO_BLOCK_OPEN_NO_IO; 215 } 216 s->crypto = qcrypto_block_open(crypto_opts, "encrypt.", 217 NULL, NULL, cflags, 1, errp); 218 if (!s->crypto) { 219 ret = -EINVAL; 220 goto fail; 221 } 222 } else { 223 error_setg(errp, "invalid encryption method in qcow header"); 224 ret = -EINVAL; 225 goto fail; 226 } 227 bs->encrypted = true; 228 } else { 229 if (encryptfmt) { 230 error_setg(errp, "No encryption in image header, but options " 231 "specified format '%s'", encryptfmt); 232 ret = -EINVAL; 233 goto fail; 234 } 235 } 236 s->cluster_bits = header.cluster_bits; 237 s->cluster_size = 1 << s->cluster_bits; 238 s->l2_bits = header.l2_bits; 239 s->l2_size = 1 << s->l2_bits; 240 bs->total_sectors = header.size / 512; 241 s->cluster_offset_mask = (1LL << (63 - s->cluster_bits)) - 1; 242 243 /* read the level 1 table */ 244 shift = s->cluster_bits + s->l2_bits; 245 if (header.size > UINT64_MAX - (1LL << shift)) { 246 error_setg(errp, "Image too large"); 247 ret = -EINVAL; 248 goto fail; 249 } else { 250 uint64_t l1_size = (header.size + (1LL << shift) - 1) >> shift; 251 if (l1_size > INT_MAX / sizeof(uint64_t)) { 252 error_setg(errp, "Image too large"); 253 ret = -EINVAL; 254 goto fail; 255 } 256 s->l1_size = l1_size; 257 } 258 259 s->l1_table_offset = header.l1_table_offset; 260 s->l1_table = g_try_new(uint64_t, s->l1_size); 261 if (s->l1_table == NULL) { 262 error_setg(errp, "Could not allocate memory for L1 table"); 263 ret = -ENOMEM; 264 goto fail; 265 } 266 267 ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table, 268 s->l1_size * sizeof(uint64_t)); 269 if (ret < 0) { 270 goto fail; 271 } 272 273 for(i = 0;i < s->l1_size; i++) { 274 s->l1_table[i] = be64_to_cpu(s->l1_table[i]); 275 } 276 277 /* alloc L2 cache (max. 64k * 16 * 8 = 8 MB) */ 278 s->l2_cache = 279 qemu_try_blockalign(bs->file->bs, 280 s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t)); 281 if (s->l2_cache == NULL) { 282 error_setg(errp, "Could not allocate L2 table cache"); 283 ret = -ENOMEM; 284 goto fail; 285 } 286 s->cluster_cache = g_malloc(s->cluster_size); 287 s->cluster_data = g_malloc(s->cluster_size); 288 s->cluster_cache_offset = -1; 289 290 /* read the backing file name */ 291 if (header.backing_file_offset != 0) { 292 len = header.backing_file_size; 293 if (len > 1023 || len >= sizeof(bs->backing_file)) { 294 error_setg(errp, "Backing file name too long"); 295 ret = -EINVAL; 296 goto fail; 297 } 298 ret = bdrv_pread(bs->file, header.backing_file_offset, 299 bs->auto_backing_file, len); 300 if (ret < 0) { 301 goto fail; 302 } 303 bs->auto_backing_file[len] = '\0'; 304 pstrcpy(bs->backing_file, sizeof(bs->backing_file), 305 bs->auto_backing_file); 306 } 307 308 /* Disable migration when qcow images are used */ 309 error_setg(&s->migration_blocker, "The qcow format used by node '%s' " 310 "does not support live migration", 311 bdrv_get_device_or_node_name(bs)); 312 ret = migrate_add_blocker(s->migration_blocker, &local_err); 313 if (local_err) { 314 error_propagate(errp, local_err); 315 error_free(s->migration_blocker); 316 goto fail; 317 } 318 319 qobject_unref(encryptopts); 320 qapi_free_QCryptoBlockOpenOptions(crypto_opts); 321 qemu_co_mutex_init(&s->lock); 322 return 0; 323 324 fail: 325 g_free(s->l1_table); 326 qemu_vfree(s->l2_cache); 327 g_free(s->cluster_cache); 328 g_free(s->cluster_data); 329 qcrypto_block_free(s->crypto); 330 qobject_unref(encryptopts); 331 qapi_free_QCryptoBlockOpenOptions(crypto_opts); 332 return ret; 333 } 334 335 336 /* We have nothing to do for QCOW reopen, stubs just return 337 * success */ 338 static int qcow_reopen_prepare(BDRVReopenState *state, 339 BlockReopenQueue *queue, Error **errp) 340 { 341 return 0; 342 } 343 344 345 /* 'allocate' is: 346 * 347 * 0 to not allocate. 348 * 349 * 1 to allocate a normal cluster (for sector-aligned byte offsets 'n_start' 350 * to 'n_end' within the cluster) 351 * 352 * 2 to allocate a compressed cluster of size 353 * 'compressed_size'. 'compressed_size' must be > 0 and < 354 * cluster_size 355 * 356 * return 0 if not allocated, 1 if *result is assigned, and negative 357 * errno on failure. 358 */ 359 static int get_cluster_offset(BlockDriverState *bs, 360 uint64_t offset, int allocate, 361 int compressed_size, 362 int n_start, int n_end, uint64_t *result) 363 { 364 BDRVQcowState *s = bs->opaque; 365 int min_index, i, j, l1_index, l2_index, ret; 366 int64_t l2_offset; 367 uint64_t *l2_table, cluster_offset, tmp; 368 uint32_t min_count; 369 int new_l2_table; 370 371 *result = 0; 372 l1_index = offset >> (s->l2_bits + s->cluster_bits); 373 l2_offset = s->l1_table[l1_index]; 374 new_l2_table = 0; 375 if (!l2_offset) { 376 if (!allocate) 377 return 0; 378 /* allocate a new l2 entry */ 379 l2_offset = bdrv_getlength(bs->file->bs); 380 if (l2_offset < 0) { 381 return l2_offset; 382 } 383 /* round to cluster size */ 384 l2_offset = QEMU_ALIGN_UP(l2_offset, s->cluster_size); 385 /* update the L1 entry */ 386 s->l1_table[l1_index] = l2_offset; 387 tmp = cpu_to_be64(l2_offset); 388 BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE); 389 ret = bdrv_pwrite_sync(bs->file, 390 s->l1_table_offset + l1_index * sizeof(tmp), 391 &tmp, sizeof(tmp)); 392 if (ret < 0) { 393 return ret; 394 } 395 new_l2_table = 1; 396 } 397 for(i = 0; i < L2_CACHE_SIZE; i++) { 398 if (l2_offset == s->l2_cache_offsets[i]) { 399 /* increment the hit count */ 400 if (++s->l2_cache_counts[i] == 0xffffffff) { 401 for(j = 0; j < L2_CACHE_SIZE; j++) { 402 s->l2_cache_counts[j] >>= 1; 403 } 404 } 405 l2_table = s->l2_cache + (i << s->l2_bits); 406 goto found; 407 } 408 } 409 /* not found: load a new entry in the least used one */ 410 min_index = 0; 411 min_count = 0xffffffff; 412 for(i = 0; i < L2_CACHE_SIZE; i++) { 413 if (s->l2_cache_counts[i] < min_count) { 414 min_count = s->l2_cache_counts[i]; 415 min_index = i; 416 } 417 } 418 l2_table = s->l2_cache + (min_index << s->l2_bits); 419 BLKDBG_EVENT(bs->file, BLKDBG_L2_LOAD); 420 if (new_l2_table) { 421 memset(l2_table, 0, s->l2_size * sizeof(uint64_t)); 422 ret = bdrv_pwrite_sync(bs->file, l2_offset, l2_table, 423 s->l2_size * sizeof(uint64_t)); 424 if (ret < 0) { 425 return ret; 426 } 427 } else { 428 ret = bdrv_pread(bs->file, l2_offset, l2_table, 429 s->l2_size * sizeof(uint64_t)); 430 if (ret < 0) { 431 return ret; 432 } 433 } 434 s->l2_cache_offsets[min_index] = l2_offset; 435 s->l2_cache_counts[min_index] = 1; 436 found: 437 l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1); 438 cluster_offset = be64_to_cpu(l2_table[l2_index]); 439 if (!cluster_offset || 440 ((cluster_offset & QCOW_OFLAG_COMPRESSED) && allocate == 1)) { 441 if (!allocate) 442 return 0; 443 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC); 444 assert(QEMU_IS_ALIGNED(n_start | n_end, BDRV_SECTOR_SIZE)); 445 /* allocate a new cluster */ 446 if ((cluster_offset & QCOW_OFLAG_COMPRESSED) && 447 (n_end - n_start) < s->cluster_size) { 448 /* if the cluster is already compressed, we must 449 decompress it in the case it is not completely 450 overwritten */ 451 if (decompress_cluster(bs, cluster_offset) < 0) { 452 return -EIO; 453 } 454 cluster_offset = bdrv_getlength(bs->file->bs); 455 if ((int64_t) cluster_offset < 0) { 456 return cluster_offset; 457 } 458 cluster_offset = QEMU_ALIGN_UP(cluster_offset, s->cluster_size); 459 /* write the cluster content */ 460 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO); 461 ret = bdrv_pwrite(bs->file, cluster_offset, s->cluster_cache, 462 s->cluster_size); 463 if (ret < 0) { 464 return ret; 465 } 466 } else { 467 cluster_offset = bdrv_getlength(bs->file->bs); 468 if ((int64_t) cluster_offset < 0) { 469 return cluster_offset; 470 } 471 if (allocate == 1) { 472 /* round to cluster size */ 473 cluster_offset = QEMU_ALIGN_UP(cluster_offset, s->cluster_size); 474 if (cluster_offset + s->cluster_size > INT64_MAX) { 475 return -E2BIG; 476 } 477 ret = bdrv_truncate(bs->file, cluster_offset + s->cluster_size, 478 PREALLOC_MODE_OFF, NULL); 479 if (ret < 0) { 480 return ret; 481 } 482 /* if encrypted, we must initialize the cluster 483 content which won't be written */ 484 if (bs->encrypted && 485 (n_end - n_start) < s->cluster_size) { 486 uint64_t start_offset; 487 assert(s->crypto); 488 start_offset = offset & ~(s->cluster_size - 1); 489 for (i = 0; i < s->cluster_size; i += BDRV_SECTOR_SIZE) { 490 if (i < n_start || i >= n_end) { 491 memset(s->cluster_data, 0x00, BDRV_SECTOR_SIZE); 492 if (qcrypto_block_encrypt(s->crypto, 493 start_offset + i, 494 s->cluster_data, 495 BDRV_SECTOR_SIZE, 496 NULL) < 0) { 497 return -EIO; 498 } 499 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO); 500 ret = bdrv_pwrite(bs->file, 501 cluster_offset + i, 502 s->cluster_data, 503 BDRV_SECTOR_SIZE); 504 if (ret < 0) { 505 return ret; 506 } 507 } 508 } 509 } 510 } else if (allocate == 2) { 511 cluster_offset |= QCOW_OFLAG_COMPRESSED | 512 (uint64_t)compressed_size << (63 - s->cluster_bits); 513 } 514 } 515 /* update L2 table */ 516 tmp = cpu_to_be64(cluster_offset); 517 l2_table[l2_index] = tmp; 518 if (allocate == 2) { 519 BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE_COMPRESSED); 520 } else { 521 BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE); 522 } 523 ret = bdrv_pwrite_sync(bs->file, l2_offset + l2_index * sizeof(tmp), 524 &tmp, sizeof(tmp)); 525 if (ret < 0) { 526 return ret; 527 } 528 } 529 *result = cluster_offset; 530 return 1; 531 } 532 533 static int coroutine_fn qcow_co_block_status(BlockDriverState *bs, 534 bool want_zero, 535 int64_t offset, int64_t bytes, 536 int64_t *pnum, int64_t *map, 537 BlockDriverState **file) 538 { 539 BDRVQcowState *s = bs->opaque; 540 int index_in_cluster, ret; 541 int64_t n; 542 uint64_t cluster_offset; 543 544 qemu_co_mutex_lock(&s->lock); 545 ret = get_cluster_offset(bs, offset, 0, 0, 0, 0, &cluster_offset); 546 qemu_co_mutex_unlock(&s->lock); 547 if (ret < 0) { 548 return ret; 549 } 550 index_in_cluster = offset & (s->cluster_size - 1); 551 n = s->cluster_size - index_in_cluster; 552 if (n > bytes) { 553 n = bytes; 554 } 555 *pnum = n; 556 if (!cluster_offset) { 557 return 0; 558 } 559 if ((cluster_offset & QCOW_OFLAG_COMPRESSED) || s->crypto) { 560 return BDRV_BLOCK_DATA; 561 } 562 *map = cluster_offset | index_in_cluster; 563 *file = bs->file->bs; 564 return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID; 565 } 566 567 static int decompress_buffer(uint8_t *out_buf, int out_buf_size, 568 const uint8_t *buf, int buf_size) 569 { 570 z_stream strm1, *strm = &strm1; 571 int ret, out_len; 572 573 memset(strm, 0, sizeof(*strm)); 574 575 strm->next_in = (uint8_t *)buf; 576 strm->avail_in = buf_size; 577 strm->next_out = out_buf; 578 strm->avail_out = out_buf_size; 579 580 ret = inflateInit2(strm, -12); 581 if (ret != Z_OK) 582 return -1; 583 ret = inflate(strm, Z_FINISH); 584 out_len = strm->next_out - out_buf; 585 if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) || 586 out_len != out_buf_size) { 587 inflateEnd(strm); 588 return -1; 589 } 590 inflateEnd(strm); 591 return 0; 592 } 593 594 static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset) 595 { 596 BDRVQcowState *s = bs->opaque; 597 int ret, csize; 598 uint64_t coffset; 599 600 coffset = cluster_offset & s->cluster_offset_mask; 601 if (s->cluster_cache_offset != coffset) { 602 csize = cluster_offset >> (63 - s->cluster_bits); 603 csize &= (s->cluster_size - 1); 604 BLKDBG_EVENT(bs->file, BLKDBG_READ_COMPRESSED); 605 ret = bdrv_pread(bs->file, coffset, s->cluster_data, csize); 606 if (ret != csize) 607 return -1; 608 if (decompress_buffer(s->cluster_cache, s->cluster_size, 609 s->cluster_data, csize) < 0) { 610 return -1; 611 } 612 s->cluster_cache_offset = coffset; 613 } 614 return 0; 615 } 616 617 static void qcow_refresh_limits(BlockDriverState *bs, Error **errp) 618 { 619 /* At least encrypted images require 512-byte alignment. Apply the 620 * limit universally, rather than just on encrypted images, as 621 * it's easier to let the block layer handle rounding than to 622 * audit this code further. */ 623 bs->bl.request_alignment = BDRV_SECTOR_SIZE; 624 } 625 626 static coroutine_fn int qcow_co_preadv(BlockDriverState *bs, uint64_t offset, 627 uint64_t bytes, QEMUIOVector *qiov, 628 int flags) 629 { 630 BDRVQcowState *s = bs->opaque; 631 int offset_in_cluster; 632 int ret = 0, n; 633 uint64_t cluster_offset; 634 uint8_t *buf; 635 void *orig_buf; 636 637 assert(!flags); 638 if (qiov->niov > 1) { 639 buf = orig_buf = qemu_try_blockalign(bs, qiov->size); 640 if (buf == NULL) { 641 return -ENOMEM; 642 } 643 } else { 644 orig_buf = NULL; 645 buf = (uint8_t *)qiov->iov->iov_base; 646 } 647 648 qemu_co_mutex_lock(&s->lock); 649 650 while (bytes != 0) { 651 /* prepare next request */ 652 ret = get_cluster_offset(bs, offset, 0, 0, 0, 0, &cluster_offset); 653 if (ret < 0) { 654 break; 655 } 656 offset_in_cluster = offset & (s->cluster_size - 1); 657 n = s->cluster_size - offset_in_cluster; 658 if (n > bytes) { 659 n = bytes; 660 } 661 662 if (!cluster_offset) { 663 if (bs->backing) { 664 /* read from the base image */ 665 qemu_co_mutex_unlock(&s->lock); 666 /* qcow2 emits this on bs->file instead of bs->backing */ 667 BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO); 668 ret = bdrv_co_pread(bs->backing, offset, n, buf, 0); 669 qemu_co_mutex_lock(&s->lock); 670 if (ret < 0) { 671 break; 672 } 673 } else { 674 /* Note: in this case, no need to wait */ 675 memset(buf, 0, n); 676 } 677 } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) { 678 /* add AIO support for compressed blocks ? */ 679 if (decompress_cluster(bs, cluster_offset) < 0) { 680 ret = -EIO; 681 break; 682 } 683 memcpy(buf, s->cluster_cache + offset_in_cluster, n); 684 } else { 685 if ((cluster_offset & 511) != 0) { 686 ret = -EIO; 687 break; 688 } 689 qemu_co_mutex_unlock(&s->lock); 690 BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO); 691 ret = bdrv_co_pread(bs->file, cluster_offset + offset_in_cluster, 692 n, buf, 0); 693 qemu_co_mutex_lock(&s->lock); 694 if (ret < 0) { 695 break; 696 } 697 if (bs->encrypted) { 698 assert(s->crypto); 699 if (qcrypto_block_decrypt(s->crypto, 700 offset, buf, n, NULL) < 0) { 701 ret = -EIO; 702 break; 703 } 704 } 705 } 706 ret = 0; 707 708 bytes -= n; 709 offset += n; 710 buf += n; 711 } 712 713 qemu_co_mutex_unlock(&s->lock); 714 715 if (qiov->niov > 1) { 716 qemu_iovec_from_buf(qiov, 0, orig_buf, qiov->size); 717 qemu_vfree(orig_buf); 718 } 719 720 return ret; 721 } 722 723 static coroutine_fn int qcow_co_pwritev(BlockDriverState *bs, uint64_t offset, 724 uint64_t bytes, QEMUIOVector *qiov, 725 int flags) 726 { 727 BDRVQcowState *s = bs->opaque; 728 int offset_in_cluster; 729 uint64_t cluster_offset; 730 int ret = 0, n; 731 uint8_t *buf; 732 void *orig_buf; 733 734 assert(!flags); 735 s->cluster_cache_offset = -1; /* disable compressed cache */ 736 737 /* We must always copy the iov when encrypting, so we 738 * don't modify the original data buffer during encryption */ 739 if (bs->encrypted || qiov->niov > 1) { 740 buf = orig_buf = qemu_try_blockalign(bs, qiov->size); 741 if (buf == NULL) { 742 return -ENOMEM; 743 } 744 qemu_iovec_to_buf(qiov, 0, buf, qiov->size); 745 } else { 746 orig_buf = NULL; 747 buf = (uint8_t *)qiov->iov->iov_base; 748 } 749 750 qemu_co_mutex_lock(&s->lock); 751 752 while (bytes != 0) { 753 offset_in_cluster = offset & (s->cluster_size - 1); 754 n = s->cluster_size - offset_in_cluster; 755 if (n > bytes) { 756 n = bytes; 757 } 758 ret = get_cluster_offset(bs, offset, 1, 0, offset_in_cluster, 759 offset_in_cluster + n, &cluster_offset); 760 if (ret < 0) { 761 break; 762 } 763 if (!cluster_offset || (cluster_offset & 511) != 0) { 764 ret = -EIO; 765 break; 766 } 767 if (bs->encrypted) { 768 assert(s->crypto); 769 if (qcrypto_block_encrypt(s->crypto, offset, buf, n, NULL) < 0) { 770 ret = -EIO; 771 break; 772 } 773 } 774 775 qemu_co_mutex_unlock(&s->lock); 776 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO); 777 ret = bdrv_co_pwrite(bs->file, cluster_offset + offset_in_cluster, 778 n, buf, 0); 779 qemu_co_mutex_lock(&s->lock); 780 if (ret < 0) { 781 break; 782 } 783 ret = 0; 784 785 bytes -= n; 786 offset += n; 787 buf += n; 788 } 789 qemu_co_mutex_unlock(&s->lock); 790 791 qemu_vfree(orig_buf); 792 793 return ret; 794 } 795 796 static void qcow_close(BlockDriverState *bs) 797 { 798 BDRVQcowState *s = bs->opaque; 799 800 qcrypto_block_free(s->crypto); 801 s->crypto = NULL; 802 g_free(s->l1_table); 803 qemu_vfree(s->l2_cache); 804 g_free(s->cluster_cache); 805 g_free(s->cluster_data); 806 807 migrate_del_blocker(s->migration_blocker); 808 error_free(s->migration_blocker); 809 } 810 811 static int coroutine_fn qcow_co_create(BlockdevCreateOptions *opts, 812 Error **errp) 813 { 814 BlockdevCreateOptionsQcow *qcow_opts; 815 int header_size, backing_filename_len, l1_size, shift, i; 816 QCowHeader header; 817 uint8_t *tmp; 818 int64_t total_size = 0; 819 int ret; 820 BlockDriverState *bs; 821 BlockBackend *qcow_blk; 822 QCryptoBlock *crypto = NULL; 823 824 assert(opts->driver == BLOCKDEV_DRIVER_QCOW); 825 qcow_opts = &opts->u.qcow; 826 827 /* Sanity checks */ 828 total_size = qcow_opts->size; 829 if (total_size == 0) { 830 error_setg(errp, "Image size is too small, cannot be zero length"); 831 return -EINVAL; 832 } 833 834 if (qcow_opts->has_encrypt && 835 qcow_opts->encrypt->format != Q_CRYPTO_BLOCK_FORMAT_QCOW) 836 { 837 error_setg(errp, "Unsupported encryption format"); 838 return -EINVAL; 839 } 840 841 /* Create BlockBackend to write to the image */ 842 bs = bdrv_open_blockdev_ref(qcow_opts->file, errp); 843 if (bs == NULL) { 844 return -EIO; 845 } 846 847 qcow_blk = blk_new(BLK_PERM_WRITE | BLK_PERM_RESIZE, BLK_PERM_ALL); 848 ret = blk_insert_bs(qcow_blk, bs, errp); 849 if (ret < 0) { 850 goto exit; 851 } 852 blk_set_allow_write_beyond_eof(qcow_blk, true); 853 854 /* Create image format */ 855 ret = blk_truncate(qcow_blk, 0, PREALLOC_MODE_OFF, errp); 856 if (ret < 0) { 857 goto exit; 858 } 859 860 memset(&header, 0, sizeof(header)); 861 header.magic = cpu_to_be32(QCOW_MAGIC); 862 header.version = cpu_to_be32(QCOW_VERSION); 863 header.size = cpu_to_be64(total_size); 864 header_size = sizeof(header); 865 backing_filename_len = 0; 866 if (qcow_opts->has_backing_file) { 867 if (strcmp(qcow_opts->backing_file, "fat:")) { 868 header.backing_file_offset = cpu_to_be64(header_size); 869 backing_filename_len = strlen(qcow_opts->backing_file); 870 header.backing_file_size = cpu_to_be32(backing_filename_len); 871 header_size += backing_filename_len; 872 } else { 873 /* special backing file for vvfat */ 874 qcow_opts->has_backing_file = false; 875 } 876 header.cluster_bits = 9; /* 512 byte cluster to avoid copying 877 unmodified sectors */ 878 header.l2_bits = 12; /* 32 KB L2 tables */ 879 } else { 880 header.cluster_bits = 12; /* 4 KB clusters */ 881 header.l2_bits = 9; /* 4 KB L2 tables */ 882 } 883 header_size = (header_size + 7) & ~7; 884 shift = header.cluster_bits + header.l2_bits; 885 l1_size = (total_size + (1LL << shift) - 1) >> shift; 886 887 header.l1_table_offset = cpu_to_be64(header_size); 888 889 if (qcow_opts->has_encrypt) { 890 header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES); 891 892 crypto = qcrypto_block_create(qcow_opts->encrypt, "encrypt.", 893 NULL, NULL, NULL, errp); 894 if (!crypto) { 895 ret = -EINVAL; 896 goto exit; 897 } 898 } else { 899 header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE); 900 } 901 902 /* write all the data */ 903 ret = blk_pwrite(qcow_blk, 0, &header, sizeof(header), 0); 904 if (ret != sizeof(header)) { 905 goto exit; 906 } 907 908 if (qcow_opts->has_backing_file) { 909 ret = blk_pwrite(qcow_blk, sizeof(header), 910 qcow_opts->backing_file, backing_filename_len, 0); 911 if (ret != backing_filename_len) { 912 goto exit; 913 } 914 } 915 916 tmp = g_malloc0(BDRV_SECTOR_SIZE); 917 for (i = 0; i < DIV_ROUND_UP(sizeof(uint64_t) * l1_size, BDRV_SECTOR_SIZE); 918 i++) { 919 ret = blk_pwrite(qcow_blk, header_size + BDRV_SECTOR_SIZE * i, 920 tmp, BDRV_SECTOR_SIZE, 0); 921 if (ret != BDRV_SECTOR_SIZE) { 922 g_free(tmp); 923 goto exit; 924 } 925 } 926 927 g_free(tmp); 928 ret = 0; 929 exit: 930 blk_unref(qcow_blk); 931 bdrv_unref(bs); 932 qcrypto_block_free(crypto); 933 return ret; 934 } 935 936 static int coroutine_fn qcow_co_create_opts(const char *filename, 937 QemuOpts *opts, Error **errp) 938 { 939 BlockdevCreateOptions *create_options = NULL; 940 BlockDriverState *bs = NULL; 941 QDict *qdict; 942 Visitor *v; 943 const char *val; 944 Error *local_err = NULL; 945 int ret; 946 947 static const QDictRenames opt_renames[] = { 948 { BLOCK_OPT_BACKING_FILE, "backing-file" }, 949 { BLOCK_OPT_ENCRYPT, BLOCK_OPT_ENCRYPT_FORMAT }, 950 { NULL, NULL }, 951 }; 952 953 /* Parse options and convert legacy syntax */ 954 qdict = qemu_opts_to_qdict_filtered(opts, NULL, &qcow_create_opts, true); 955 956 val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT); 957 if (val && !strcmp(val, "on")) { 958 qdict_put_str(qdict, BLOCK_OPT_ENCRYPT, "qcow"); 959 } else if (val && !strcmp(val, "off")) { 960 qdict_del(qdict, BLOCK_OPT_ENCRYPT); 961 } 962 963 val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT); 964 if (val && !strcmp(val, "aes")) { 965 qdict_put_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT, "qcow"); 966 } 967 968 if (!qdict_rename_keys(qdict, opt_renames, errp)) { 969 ret = -EINVAL; 970 goto fail; 971 } 972 973 /* Create and open the file (protocol layer) */ 974 ret = bdrv_create_file(filename, opts, &local_err); 975 if (ret < 0) { 976 error_propagate(errp, local_err); 977 goto fail; 978 } 979 980 bs = bdrv_open(filename, NULL, NULL, 981 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp); 982 if (bs == NULL) { 983 ret = -EIO; 984 goto fail; 985 } 986 987 /* Now get the QAPI type BlockdevCreateOptions */ 988 qdict_put_str(qdict, "driver", "qcow"); 989 qdict_put_str(qdict, "file", bs->node_name); 990 991 v = qobject_input_visitor_new_flat_confused(qdict, errp); 992 if (!v) { 993 ret = -EINVAL; 994 goto fail; 995 } 996 997 visit_type_BlockdevCreateOptions(v, NULL, &create_options, &local_err); 998 visit_free(v); 999 1000 if (local_err) { 1001 error_propagate(errp, local_err); 1002 ret = -EINVAL; 1003 goto fail; 1004 } 1005 1006 /* Silently round up size */ 1007 assert(create_options->driver == BLOCKDEV_DRIVER_QCOW); 1008 create_options->u.qcow.size = 1009 ROUND_UP(create_options->u.qcow.size, BDRV_SECTOR_SIZE); 1010 1011 /* Create the qcow image (format layer) */ 1012 ret = qcow_co_create(create_options, errp); 1013 if (ret < 0) { 1014 goto fail; 1015 } 1016 1017 ret = 0; 1018 fail: 1019 qobject_unref(qdict); 1020 bdrv_unref(bs); 1021 qapi_free_BlockdevCreateOptions(create_options); 1022 return ret; 1023 } 1024 1025 static int qcow_make_empty(BlockDriverState *bs) 1026 { 1027 BDRVQcowState *s = bs->opaque; 1028 uint32_t l1_length = s->l1_size * sizeof(uint64_t); 1029 int ret; 1030 1031 memset(s->l1_table, 0, l1_length); 1032 if (bdrv_pwrite_sync(bs->file, s->l1_table_offset, s->l1_table, 1033 l1_length) < 0) 1034 return -1; 1035 ret = bdrv_truncate(bs->file, s->l1_table_offset + l1_length, 1036 PREALLOC_MODE_OFF, NULL); 1037 if (ret < 0) 1038 return ret; 1039 1040 memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t)); 1041 memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t)); 1042 memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t)); 1043 1044 return 0; 1045 } 1046 1047 /* XXX: put compressed sectors first, then all the cluster aligned 1048 tables to avoid losing bytes in alignment */ 1049 static coroutine_fn int 1050 qcow_co_pwritev_compressed(BlockDriverState *bs, uint64_t offset, 1051 uint64_t bytes, QEMUIOVector *qiov) 1052 { 1053 BDRVQcowState *s = bs->opaque; 1054 z_stream strm; 1055 int ret, out_len; 1056 uint8_t *buf, *out_buf; 1057 uint64_t cluster_offset; 1058 1059 buf = qemu_blockalign(bs, s->cluster_size); 1060 if (bytes != s->cluster_size) { 1061 if (bytes > s->cluster_size || 1062 offset + bytes != bs->total_sectors << BDRV_SECTOR_BITS) 1063 { 1064 qemu_vfree(buf); 1065 return -EINVAL; 1066 } 1067 /* Zero-pad last write if image size is not cluster aligned */ 1068 memset(buf + bytes, 0, s->cluster_size - bytes); 1069 } 1070 qemu_iovec_to_buf(qiov, 0, buf, qiov->size); 1071 1072 out_buf = g_malloc(s->cluster_size); 1073 1074 /* best compression, small window, no zlib header */ 1075 memset(&strm, 0, sizeof(strm)); 1076 ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION, 1077 Z_DEFLATED, -12, 1078 9, Z_DEFAULT_STRATEGY); 1079 if (ret != 0) { 1080 ret = -EINVAL; 1081 goto fail; 1082 } 1083 1084 strm.avail_in = s->cluster_size; 1085 strm.next_in = (uint8_t *)buf; 1086 strm.avail_out = s->cluster_size; 1087 strm.next_out = out_buf; 1088 1089 ret = deflate(&strm, Z_FINISH); 1090 if (ret != Z_STREAM_END && ret != Z_OK) { 1091 deflateEnd(&strm); 1092 ret = -EINVAL; 1093 goto fail; 1094 } 1095 out_len = strm.next_out - out_buf; 1096 1097 deflateEnd(&strm); 1098 1099 if (ret != Z_STREAM_END || out_len >= s->cluster_size) { 1100 /* could not compress: write normal cluster */ 1101 ret = qcow_co_pwritev(bs, offset, bytes, qiov, 0); 1102 if (ret < 0) { 1103 goto fail; 1104 } 1105 goto success; 1106 } 1107 qemu_co_mutex_lock(&s->lock); 1108 ret = get_cluster_offset(bs, offset, 2, out_len, 0, 0, &cluster_offset); 1109 qemu_co_mutex_unlock(&s->lock); 1110 if (ret < 0) { 1111 goto fail; 1112 } 1113 if (cluster_offset == 0) { 1114 ret = -EIO; 1115 goto fail; 1116 } 1117 cluster_offset &= s->cluster_offset_mask; 1118 1119 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_COMPRESSED); 1120 ret = bdrv_co_pwrite(bs->file, cluster_offset, out_len, out_buf, 0); 1121 if (ret < 0) { 1122 goto fail; 1123 } 1124 success: 1125 ret = 0; 1126 fail: 1127 qemu_vfree(buf); 1128 g_free(out_buf); 1129 return ret; 1130 } 1131 1132 static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) 1133 { 1134 BDRVQcowState *s = bs->opaque; 1135 bdi->cluster_size = s->cluster_size; 1136 return 0; 1137 } 1138 1139 static QemuOptsList qcow_create_opts = { 1140 .name = "qcow-create-opts", 1141 .head = QTAILQ_HEAD_INITIALIZER(qcow_create_opts.head), 1142 .desc = { 1143 { 1144 .name = BLOCK_OPT_SIZE, 1145 .type = QEMU_OPT_SIZE, 1146 .help = "Virtual disk size" 1147 }, 1148 { 1149 .name = BLOCK_OPT_BACKING_FILE, 1150 .type = QEMU_OPT_STRING, 1151 .help = "File name of a base image" 1152 }, 1153 { 1154 .name = BLOCK_OPT_ENCRYPT, 1155 .type = QEMU_OPT_BOOL, 1156 .help = "Encrypt the image with format 'aes'. (Deprecated " 1157 "in favor of " BLOCK_OPT_ENCRYPT_FORMAT "=aes)", 1158 }, 1159 { 1160 .name = BLOCK_OPT_ENCRYPT_FORMAT, 1161 .type = QEMU_OPT_STRING, 1162 .help = "Encrypt the image, format choices: 'aes'", 1163 }, 1164 BLOCK_CRYPTO_OPT_DEF_QCOW_KEY_SECRET("encrypt."), 1165 { /* end of list */ } 1166 } 1167 }; 1168 1169 static const char *const qcow_strong_runtime_opts[] = { 1170 "encrypt." BLOCK_CRYPTO_OPT_QCOW_KEY_SECRET, 1171 1172 NULL 1173 }; 1174 1175 static BlockDriver bdrv_qcow = { 1176 .format_name = "qcow", 1177 .instance_size = sizeof(BDRVQcowState), 1178 .bdrv_probe = qcow_probe, 1179 .bdrv_open = qcow_open, 1180 .bdrv_close = qcow_close, 1181 .bdrv_child_perm = bdrv_format_default_perms, 1182 .bdrv_reopen_prepare = qcow_reopen_prepare, 1183 .bdrv_co_create = qcow_co_create, 1184 .bdrv_co_create_opts = qcow_co_create_opts, 1185 .bdrv_has_zero_init = bdrv_has_zero_init_1, 1186 .supports_backing = true, 1187 .bdrv_refresh_limits = qcow_refresh_limits, 1188 1189 .bdrv_co_preadv = qcow_co_preadv, 1190 .bdrv_co_pwritev = qcow_co_pwritev, 1191 .bdrv_co_block_status = qcow_co_block_status, 1192 1193 .bdrv_make_empty = qcow_make_empty, 1194 .bdrv_co_pwritev_compressed = qcow_co_pwritev_compressed, 1195 .bdrv_get_info = qcow_get_info, 1196 1197 .create_opts = &qcow_create_opts, 1198 .strong_runtime_opts = qcow_strong_runtime_opts, 1199 }; 1200 1201 static void bdrv_qcow_init(void) 1202 { 1203 bdrv_register(&bdrv_qcow); 1204 } 1205 1206 block_init(bdrv_qcow_init); 1207