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 "qemu/memalign.h" 36 #include <zlib.h> 37 #include "qapi/qmp/qdict.h" 38 #include "qapi/qmp/qstring.h" 39 #include "qapi/qobject-input-visitor.h" 40 #include "qapi/qapi-visit-block-core.h" 41 #include "crypto/block.h" 42 #include "migration/blocker.h" 43 #include "crypto.h" 44 45 /**************************************************************/ 46 /* QEMU COW block driver with compression and encryption support */ 47 48 #define QCOW_MAGIC (('Q' << 24) | ('F' << 16) | ('I' << 8) | 0xfb) 49 #define QCOW_VERSION 1 50 51 #define QCOW_CRYPT_NONE 0 52 #define QCOW_CRYPT_AES 1 53 54 #define QCOW_OFLAG_COMPRESSED (1LL << 63) 55 56 typedef struct QCowHeader { 57 uint32_t magic; 58 uint32_t version; 59 uint64_t backing_file_offset; 60 uint32_t backing_file_size; 61 uint32_t mtime; 62 uint64_t size; /* in bytes */ 63 uint8_t cluster_bits; 64 uint8_t l2_bits; 65 uint16_t padding; 66 uint32_t crypt_method; 67 uint64_t l1_table_offset; 68 } QEMU_PACKED QCowHeader; 69 70 #define L2_CACHE_SIZE 16 71 72 typedef struct BDRVQcowState { 73 int cluster_bits; 74 int cluster_size; 75 int l2_bits; 76 int l2_size; 77 unsigned int l1_size; 78 uint64_t cluster_offset_mask; 79 uint64_t l1_table_offset; 80 uint64_t *l1_table; 81 uint64_t *l2_cache; 82 uint64_t l2_cache_offsets[L2_CACHE_SIZE]; 83 uint32_t l2_cache_counts[L2_CACHE_SIZE]; 84 uint8_t *cluster_cache; 85 uint8_t *cluster_data; 86 uint64_t cluster_cache_offset; 87 QCryptoBlock *crypto; /* Disk encryption format driver */ 88 uint32_t crypt_method_header; 89 CoMutex lock; 90 Error *migration_blocker; 91 } BDRVQcowState; 92 93 static QemuOptsList qcow_create_opts; 94 95 static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset); 96 97 static int qcow_probe(const uint8_t *buf, int buf_size, const char *filename) 98 { 99 const QCowHeader *cow_header = (const void *)buf; 100 101 if (buf_size >= sizeof(QCowHeader) && 102 be32_to_cpu(cow_header->magic) == QCOW_MAGIC && 103 be32_to_cpu(cow_header->version) == QCOW_VERSION) 104 return 100; 105 else 106 return 0; 107 } 108 109 static int qcow_open(BlockDriverState *bs, QDict *options, int flags, 110 Error **errp) 111 { 112 BDRVQcowState *s = bs->opaque; 113 unsigned int len, i, shift; 114 int ret; 115 QCowHeader header; 116 QCryptoBlockOpenOptions *crypto_opts = NULL; 117 unsigned int cflags = 0; 118 QDict *encryptopts = NULL; 119 const char *encryptfmt; 120 121 qdict_extract_subqdict(options, &encryptopts, "encrypt."); 122 encryptfmt = qdict_get_try_str(encryptopts, "format"); 123 124 bs->file = bdrv_open_child(NULL, options, "file", bs, &child_of_bds, 125 BDRV_CHILD_IMAGE, false, errp); 126 if (!bs->file) { 127 ret = -EINVAL; 128 goto fail; 129 } 130 131 ret = bdrv_pread(bs->file, 0, &header, sizeof(header)); 132 if (ret < 0) { 133 goto fail; 134 } 135 header.magic = be32_to_cpu(header.magic); 136 header.version = be32_to_cpu(header.version); 137 header.backing_file_offset = be64_to_cpu(header.backing_file_offset); 138 header.backing_file_size = be32_to_cpu(header.backing_file_size); 139 header.mtime = be32_to_cpu(header.mtime); 140 header.size = be64_to_cpu(header.size); 141 header.crypt_method = be32_to_cpu(header.crypt_method); 142 header.l1_table_offset = be64_to_cpu(header.l1_table_offset); 143 144 if (header.magic != QCOW_MAGIC) { 145 error_setg(errp, "Image not in qcow format"); 146 ret = -EINVAL; 147 goto fail; 148 } 149 if (header.version != QCOW_VERSION) { 150 error_setg(errp, "qcow (v%d) does not support qcow version %" PRIu32, 151 QCOW_VERSION, header.version); 152 if (header.version == 2 || header.version == 3) { 153 error_append_hint(errp, "Try the 'qcow2' driver instead.\n"); 154 } 155 156 ret = -ENOTSUP; 157 goto fail; 158 } 159 160 if (header.size <= 1) { 161 error_setg(errp, "Image size is too small (must be at least 2 bytes)"); 162 ret = -EINVAL; 163 goto fail; 164 } 165 if (header.cluster_bits < 9 || header.cluster_bits > 16) { 166 error_setg(errp, "Cluster size must be between 512 and 64k"); 167 ret = -EINVAL; 168 goto fail; 169 } 170 171 /* l2_bits specifies number of entries; storing a uint64_t in each entry, 172 * so bytes = num_entries << 3. */ 173 if (header.l2_bits < 9 - 3 || header.l2_bits > 16 - 3) { 174 error_setg(errp, "L2 table size must be between 512 and 64k"); 175 ret = -EINVAL; 176 goto fail; 177 } 178 179 s->crypt_method_header = header.crypt_method; 180 if (s->crypt_method_header) { 181 if (bdrv_uses_whitelist() && 182 s->crypt_method_header == QCOW_CRYPT_AES) { 183 error_setg(errp, 184 "Use of AES-CBC encrypted qcow images is no longer " 185 "supported in system emulators"); 186 error_append_hint(errp, 187 "You can use 'qemu-img convert' to convert your " 188 "image to an alternative supported format, such " 189 "as unencrypted qcow, or raw with the LUKS " 190 "format instead.\n"); 191 ret = -ENOSYS; 192 goto fail; 193 } 194 if (s->crypt_method_header == QCOW_CRYPT_AES) { 195 if (encryptfmt && !g_str_equal(encryptfmt, "aes")) { 196 error_setg(errp, 197 "Header reported 'aes' encryption format but " 198 "options specify '%s'", encryptfmt); 199 ret = -EINVAL; 200 goto fail; 201 } 202 qdict_put_str(encryptopts, "format", "qcow"); 203 crypto_opts = block_crypto_open_opts_init(encryptopts, errp); 204 if (!crypto_opts) { 205 ret = -EINVAL; 206 goto fail; 207 } 208 209 if (flags & BDRV_O_NO_IO) { 210 cflags |= QCRYPTO_BLOCK_OPEN_NO_IO; 211 } 212 s->crypto = qcrypto_block_open(crypto_opts, "encrypt.", 213 NULL, NULL, cflags, 1, errp); 214 if (!s->crypto) { 215 ret = -EINVAL; 216 goto fail; 217 } 218 } else { 219 error_setg(errp, "invalid encryption method in qcow header"); 220 ret = -EINVAL; 221 goto fail; 222 } 223 bs->encrypted = true; 224 } else { 225 if (encryptfmt) { 226 error_setg(errp, "No encryption in image header, but options " 227 "specified format '%s'", encryptfmt); 228 ret = -EINVAL; 229 goto fail; 230 } 231 } 232 s->cluster_bits = header.cluster_bits; 233 s->cluster_size = 1 << s->cluster_bits; 234 s->l2_bits = header.l2_bits; 235 s->l2_size = 1 << s->l2_bits; 236 bs->total_sectors = header.size / 512; 237 s->cluster_offset_mask = (1LL << (63 - s->cluster_bits)) - 1; 238 239 /* read the level 1 table */ 240 shift = s->cluster_bits + s->l2_bits; 241 if (header.size > UINT64_MAX - (1LL << shift)) { 242 error_setg(errp, "Image too large"); 243 ret = -EINVAL; 244 goto fail; 245 } else { 246 uint64_t l1_size = (header.size + (1LL << shift) - 1) >> shift; 247 if (l1_size > INT_MAX / sizeof(uint64_t)) { 248 error_setg(errp, "Image too large"); 249 ret = -EINVAL; 250 goto fail; 251 } 252 s->l1_size = l1_size; 253 } 254 255 s->l1_table_offset = header.l1_table_offset; 256 s->l1_table = g_try_new(uint64_t, s->l1_size); 257 if (s->l1_table == NULL) { 258 error_setg(errp, "Could not allocate memory for L1 table"); 259 ret = -ENOMEM; 260 goto fail; 261 } 262 263 ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table, 264 s->l1_size * sizeof(uint64_t)); 265 if (ret < 0) { 266 goto fail; 267 } 268 269 for(i = 0;i < s->l1_size; i++) { 270 s->l1_table[i] = be64_to_cpu(s->l1_table[i]); 271 } 272 273 /* alloc L2 cache (max. 64k * 16 * 8 = 8 MB) */ 274 s->l2_cache = 275 qemu_try_blockalign(bs->file->bs, 276 s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t)); 277 if (s->l2_cache == NULL) { 278 error_setg(errp, "Could not allocate L2 table cache"); 279 ret = -ENOMEM; 280 goto fail; 281 } 282 s->cluster_cache = g_malloc(s->cluster_size); 283 s->cluster_data = g_malloc(s->cluster_size); 284 s->cluster_cache_offset = -1; 285 286 /* read the backing file name */ 287 if (header.backing_file_offset != 0) { 288 len = header.backing_file_size; 289 if (len > 1023 || len >= sizeof(bs->backing_file)) { 290 error_setg(errp, "Backing file name too long"); 291 ret = -EINVAL; 292 goto fail; 293 } 294 ret = bdrv_pread(bs->file, header.backing_file_offset, 295 bs->auto_backing_file, len); 296 if (ret < 0) { 297 goto fail; 298 } 299 bs->auto_backing_file[len] = '\0'; 300 pstrcpy(bs->backing_file, sizeof(bs->backing_file), 301 bs->auto_backing_file); 302 } 303 304 /* Disable migration when qcow images are used */ 305 error_setg(&s->migration_blocker, "The qcow format used by node '%s' " 306 "does not support live migration", 307 bdrv_get_device_or_node_name(bs)); 308 ret = migrate_add_blocker(s->migration_blocker, errp); 309 if (ret < 0) { 310 error_free(s->migration_blocker); 311 goto fail; 312 } 313 314 qobject_unref(encryptopts); 315 qapi_free_QCryptoBlockOpenOptions(crypto_opts); 316 qemu_co_mutex_init(&s->lock); 317 return 0; 318 319 fail: 320 g_free(s->l1_table); 321 qemu_vfree(s->l2_cache); 322 g_free(s->cluster_cache); 323 g_free(s->cluster_data); 324 qcrypto_block_free(s->crypto); 325 qobject_unref(encryptopts); 326 qapi_free_QCryptoBlockOpenOptions(crypto_opts); 327 return ret; 328 } 329 330 331 /* We have nothing to do for QCOW reopen, stubs just return 332 * success */ 333 static int qcow_reopen_prepare(BDRVReopenState *state, 334 BlockReopenQueue *queue, Error **errp) 335 { 336 return 0; 337 } 338 339 340 /* 'allocate' is: 341 * 342 * 0 to not allocate. 343 * 344 * 1 to allocate a normal cluster (for sector-aligned byte offsets 'n_start' 345 * to 'n_end' within the cluster) 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, 1 if *result is assigned, and negative 352 * errno on failure. 353 */ 354 static int get_cluster_offset(BlockDriverState *bs, 355 uint64_t offset, int allocate, 356 int compressed_size, 357 int n_start, int n_end, uint64_t *result) 358 { 359 BDRVQcowState *s = bs->opaque; 360 int min_index, i, j, l1_index, l2_index, ret; 361 int64_t l2_offset; 362 uint64_t *l2_table, cluster_offset, tmp; 363 uint32_t min_count; 364 int new_l2_table; 365 366 *result = 0; 367 l1_index = offset >> (s->l2_bits + s->cluster_bits); 368 l2_offset = s->l1_table[l1_index]; 369 new_l2_table = 0; 370 if (!l2_offset) { 371 if (!allocate) 372 return 0; 373 /* allocate a new l2 entry */ 374 l2_offset = bdrv_getlength(bs->file->bs); 375 if (l2_offset < 0) { 376 return l2_offset; 377 } 378 /* round to cluster size */ 379 l2_offset = QEMU_ALIGN_UP(l2_offset, s->cluster_size); 380 /* update the L1 entry */ 381 s->l1_table[l1_index] = l2_offset; 382 tmp = cpu_to_be64(l2_offset); 383 BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE); 384 ret = bdrv_pwrite_sync(bs->file, 385 s->l1_table_offset + l1_index * sizeof(tmp), 386 &tmp, sizeof(tmp)); 387 if (ret < 0) { 388 return ret; 389 } 390 new_l2_table = 1; 391 } 392 for(i = 0; i < L2_CACHE_SIZE; i++) { 393 if (l2_offset == s->l2_cache_offsets[i]) { 394 /* increment the hit count */ 395 if (++s->l2_cache_counts[i] == 0xffffffff) { 396 for(j = 0; j < L2_CACHE_SIZE; j++) { 397 s->l2_cache_counts[j] >>= 1; 398 } 399 } 400 l2_table = s->l2_cache + (i << s->l2_bits); 401 goto found; 402 } 403 } 404 /* not found: load a new entry in the least used one */ 405 min_index = 0; 406 min_count = 0xffffffff; 407 for(i = 0; i < L2_CACHE_SIZE; i++) { 408 if (s->l2_cache_counts[i] < min_count) { 409 min_count = s->l2_cache_counts[i]; 410 min_index = i; 411 } 412 } 413 l2_table = s->l2_cache + (min_index << s->l2_bits); 414 BLKDBG_EVENT(bs->file, BLKDBG_L2_LOAD); 415 if (new_l2_table) { 416 memset(l2_table, 0, s->l2_size * sizeof(uint64_t)); 417 ret = bdrv_pwrite_sync(bs->file, l2_offset, l2_table, 418 s->l2_size * sizeof(uint64_t)); 419 if (ret < 0) { 420 return ret; 421 } 422 } else { 423 ret = bdrv_pread(bs->file, l2_offset, l2_table, 424 s->l2_size * sizeof(uint64_t)); 425 if (ret < 0) { 426 return ret; 427 } 428 } 429 s->l2_cache_offsets[min_index] = l2_offset; 430 s->l2_cache_counts[min_index] = 1; 431 found: 432 l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1); 433 cluster_offset = be64_to_cpu(l2_table[l2_index]); 434 if (!cluster_offset || 435 ((cluster_offset & QCOW_OFLAG_COMPRESSED) && allocate == 1)) { 436 if (!allocate) 437 return 0; 438 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC); 439 assert(QEMU_IS_ALIGNED(n_start | n_end, BDRV_SECTOR_SIZE)); 440 /* allocate a new cluster */ 441 if ((cluster_offset & QCOW_OFLAG_COMPRESSED) && 442 (n_end - n_start) < s->cluster_size) { 443 /* if the cluster is already compressed, we must 444 decompress it in the case it is not completely 445 overwritten */ 446 if (decompress_cluster(bs, cluster_offset) < 0) { 447 return -EIO; 448 } 449 cluster_offset = bdrv_getlength(bs->file->bs); 450 if ((int64_t) cluster_offset < 0) { 451 return cluster_offset; 452 } 453 cluster_offset = QEMU_ALIGN_UP(cluster_offset, s->cluster_size); 454 /* write the cluster content */ 455 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO); 456 ret = bdrv_pwrite(bs->file, cluster_offset, s->cluster_cache, 457 s->cluster_size); 458 if (ret < 0) { 459 return ret; 460 } 461 } else { 462 cluster_offset = bdrv_getlength(bs->file->bs); 463 if ((int64_t) cluster_offset < 0) { 464 return cluster_offset; 465 } 466 if (allocate == 1) { 467 /* round to cluster size */ 468 cluster_offset = QEMU_ALIGN_UP(cluster_offset, s->cluster_size); 469 if (cluster_offset + s->cluster_size > INT64_MAX) { 470 return -E2BIG; 471 } 472 ret = bdrv_truncate(bs->file, cluster_offset + s->cluster_size, 473 false, PREALLOC_MODE_OFF, 0, NULL); 474 if (ret < 0) { 475 return ret; 476 } 477 /* if encrypted, we must initialize the cluster 478 content which won't be written */ 479 if (bs->encrypted && 480 (n_end - n_start) < s->cluster_size) { 481 uint64_t start_offset; 482 assert(s->crypto); 483 start_offset = offset & ~(s->cluster_size - 1); 484 for (i = 0; i < s->cluster_size; i += BDRV_SECTOR_SIZE) { 485 if (i < n_start || i >= n_end) { 486 memset(s->cluster_data, 0x00, BDRV_SECTOR_SIZE); 487 if (qcrypto_block_encrypt(s->crypto, 488 start_offset + i, 489 s->cluster_data, 490 BDRV_SECTOR_SIZE, 491 NULL) < 0) { 492 return -EIO; 493 } 494 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO); 495 ret = bdrv_pwrite(bs->file, 496 cluster_offset + i, 497 s->cluster_data, 498 BDRV_SECTOR_SIZE); 499 if (ret < 0) { 500 return ret; 501 } 502 } 503 } 504 } 505 } else if (allocate == 2) { 506 cluster_offset |= QCOW_OFLAG_COMPRESSED | 507 (uint64_t)compressed_size << (63 - s->cluster_bits); 508 } 509 } 510 /* update L2 table */ 511 tmp = cpu_to_be64(cluster_offset); 512 l2_table[l2_index] = tmp; 513 if (allocate == 2) { 514 BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE_COMPRESSED); 515 } else { 516 BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE); 517 } 518 ret = bdrv_pwrite_sync(bs->file, l2_offset + l2_index * sizeof(tmp), 519 &tmp, sizeof(tmp)); 520 if (ret < 0) { 521 return ret; 522 } 523 } 524 *result = cluster_offset; 525 return 1; 526 } 527 528 static int coroutine_fn qcow_co_block_status(BlockDriverState *bs, 529 bool want_zero, 530 int64_t offset, int64_t bytes, 531 int64_t *pnum, int64_t *map, 532 BlockDriverState **file) 533 { 534 BDRVQcowState *s = bs->opaque; 535 int index_in_cluster, ret; 536 int64_t n; 537 uint64_t cluster_offset; 538 539 qemu_co_mutex_lock(&s->lock); 540 ret = get_cluster_offset(bs, offset, 0, 0, 0, 0, &cluster_offset); 541 qemu_co_mutex_unlock(&s->lock); 542 if (ret < 0) { 543 return ret; 544 } 545 index_in_cluster = offset & (s->cluster_size - 1); 546 n = s->cluster_size - index_in_cluster; 547 if (n > bytes) { 548 n = bytes; 549 } 550 *pnum = n; 551 if (!cluster_offset) { 552 return 0; 553 } 554 if ((cluster_offset & QCOW_OFLAG_COMPRESSED) || s->crypto) { 555 return BDRV_BLOCK_DATA; 556 } 557 *map = cluster_offset | index_in_cluster; 558 *file = bs->file->bs; 559 return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID; 560 } 561 562 static int decompress_buffer(uint8_t *out_buf, int out_buf_size, 563 const uint8_t *buf, int buf_size) 564 { 565 z_stream strm1, *strm = &strm1; 566 int ret, out_len; 567 568 memset(strm, 0, sizeof(*strm)); 569 570 strm->next_in = (uint8_t *)buf; 571 strm->avail_in = buf_size; 572 strm->next_out = out_buf; 573 strm->avail_out = out_buf_size; 574 575 ret = inflateInit2(strm, -12); 576 if (ret != Z_OK) 577 return -1; 578 ret = inflate(strm, Z_FINISH); 579 out_len = strm->next_out - out_buf; 580 if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) || 581 out_len != out_buf_size) { 582 inflateEnd(strm); 583 return -1; 584 } 585 inflateEnd(strm); 586 return 0; 587 } 588 589 static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset) 590 { 591 BDRVQcowState *s = bs->opaque; 592 int ret, csize; 593 uint64_t coffset; 594 595 coffset = cluster_offset & s->cluster_offset_mask; 596 if (s->cluster_cache_offset != coffset) { 597 csize = cluster_offset >> (63 - s->cluster_bits); 598 csize &= (s->cluster_size - 1); 599 BLKDBG_EVENT(bs->file, BLKDBG_READ_COMPRESSED); 600 ret = bdrv_pread(bs->file, coffset, s->cluster_data, csize); 601 if (ret != csize) 602 return -1; 603 if (decompress_buffer(s->cluster_cache, s->cluster_size, 604 s->cluster_data, csize) < 0) { 605 return -1; 606 } 607 s->cluster_cache_offset = coffset; 608 } 609 return 0; 610 } 611 612 static void qcow_refresh_limits(BlockDriverState *bs, Error **errp) 613 { 614 /* At least encrypted images require 512-byte alignment. Apply the 615 * limit universally, rather than just on encrypted images, as 616 * it's easier to let the block layer handle rounding than to 617 * audit this code further. */ 618 bs->bl.request_alignment = BDRV_SECTOR_SIZE; 619 } 620 621 static coroutine_fn int qcow_co_preadv(BlockDriverState *bs, int64_t offset, 622 int64_t bytes, QEMUIOVector *qiov, 623 BdrvRequestFlags flags) 624 { 625 BDRVQcowState *s = bs->opaque; 626 int offset_in_cluster; 627 int ret = 0, n; 628 uint64_t cluster_offset; 629 uint8_t *buf; 630 void *orig_buf; 631 632 assert(!flags); 633 if (qiov->niov > 1) { 634 buf = orig_buf = qemu_try_blockalign(bs, qiov->size); 635 if (buf == NULL) { 636 return -ENOMEM; 637 } 638 } else { 639 orig_buf = NULL; 640 buf = (uint8_t *)qiov->iov->iov_base; 641 } 642 643 qemu_co_mutex_lock(&s->lock); 644 645 while (bytes != 0) { 646 /* prepare next request */ 647 ret = get_cluster_offset(bs, offset, 0, 0, 0, 0, &cluster_offset); 648 if (ret < 0) { 649 break; 650 } 651 offset_in_cluster = offset & (s->cluster_size - 1); 652 n = s->cluster_size - offset_in_cluster; 653 if (n > bytes) { 654 n = bytes; 655 } 656 657 if (!cluster_offset) { 658 if (bs->backing) { 659 /* read from the base image */ 660 qemu_co_mutex_unlock(&s->lock); 661 /* qcow2 emits this on bs->file instead of bs->backing */ 662 BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO); 663 ret = bdrv_co_pread(bs->backing, offset, n, buf, 0); 664 qemu_co_mutex_lock(&s->lock); 665 if (ret < 0) { 666 break; 667 } 668 } else { 669 /* Note: in this case, no need to wait */ 670 memset(buf, 0, n); 671 } 672 } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) { 673 /* add AIO support for compressed blocks ? */ 674 if (decompress_cluster(bs, cluster_offset) < 0) { 675 ret = -EIO; 676 break; 677 } 678 memcpy(buf, s->cluster_cache + offset_in_cluster, n); 679 } else { 680 if ((cluster_offset & 511) != 0) { 681 ret = -EIO; 682 break; 683 } 684 qemu_co_mutex_unlock(&s->lock); 685 BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO); 686 ret = bdrv_co_pread(bs->file, cluster_offset + offset_in_cluster, 687 n, buf, 0); 688 qemu_co_mutex_lock(&s->lock); 689 if (ret < 0) { 690 break; 691 } 692 if (bs->encrypted) { 693 assert(s->crypto); 694 if (qcrypto_block_decrypt(s->crypto, 695 offset, buf, n, NULL) < 0) { 696 ret = -EIO; 697 break; 698 } 699 } 700 } 701 ret = 0; 702 703 bytes -= n; 704 offset += n; 705 buf += n; 706 } 707 708 qemu_co_mutex_unlock(&s->lock); 709 710 if (qiov->niov > 1) { 711 qemu_iovec_from_buf(qiov, 0, orig_buf, qiov->size); 712 qemu_vfree(orig_buf); 713 } 714 715 return ret; 716 } 717 718 static coroutine_fn int qcow_co_pwritev(BlockDriverState *bs, int64_t offset, 719 int64_t bytes, QEMUIOVector *qiov, 720 BdrvRequestFlags flags) 721 { 722 BDRVQcowState *s = bs->opaque; 723 int offset_in_cluster; 724 uint64_t cluster_offset; 725 int ret = 0, n; 726 uint8_t *buf; 727 void *orig_buf; 728 729 assert(!flags); 730 s->cluster_cache_offset = -1; /* disable compressed cache */ 731 732 /* We must always copy the iov when encrypting, so we 733 * don't modify the original data buffer during encryption */ 734 if (bs->encrypted || qiov->niov > 1) { 735 buf = orig_buf = qemu_try_blockalign(bs, qiov->size); 736 if (buf == NULL) { 737 return -ENOMEM; 738 } 739 qemu_iovec_to_buf(qiov, 0, buf, qiov->size); 740 } else { 741 orig_buf = NULL; 742 buf = (uint8_t *)qiov->iov->iov_base; 743 } 744 745 qemu_co_mutex_lock(&s->lock); 746 747 while (bytes != 0) { 748 offset_in_cluster = offset & (s->cluster_size - 1); 749 n = s->cluster_size - offset_in_cluster; 750 if (n > bytes) { 751 n = bytes; 752 } 753 ret = get_cluster_offset(bs, offset, 1, 0, offset_in_cluster, 754 offset_in_cluster + n, &cluster_offset); 755 if (ret < 0) { 756 break; 757 } 758 if (!cluster_offset || (cluster_offset & 511) != 0) { 759 ret = -EIO; 760 break; 761 } 762 if (bs->encrypted) { 763 assert(s->crypto); 764 if (qcrypto_block_encrypt(s->crypto, offset, buf, n, NULL) < 0) { 765 ret = -EIO; 766 break; 767 } 768 } 769 770 qemu_co_mutex_unlock(&s->lock); 771 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO); 772 ret = bdrv_co_pwrite(bs->file, cluster_offset + offset_in_cluster, 773 n, buf, 0); 774 qemu_co_mutex_lock(&s->lock); 775 if (ret < 0) { 776 break; 777 } 778 ret = 0; 779 780 bytes -= n; 781 offset += n; 782 buf += n; 783 } 784 qemu_co_mutex_unlock(&s->lock); 785 786 qemu_vfree(orig_buf); 787 788 return ret; 789 } 790 791 static void qcow_close(BlockDriverState *bs) 792 { 793 BDRVQcowState *s = bs->opaque; 794 795 qcrypto_block_free(s->crypto); 796 s->crypto = NULL; 797 g_free(s->l1_table); 798 qemu_vfree(s->l2_cache); 799 g_free(s->cluster_cache); 800 g_free(s->cluster_data); 801 802 migrate_del_blocker(s->migration_blocker); 803 error_free(s->migration_blocker); 804 } 805 806 static int coroutine_fn qcow_co_create(BlockdevCreateOptions *opts, 807 Error **errp) 808 { 809 BlockdevCreateOptionsQcow *qcow_opts; 810 int header_size, backing_filename_len, l1_size, shift, i; 811 QCowHeader header; 812 uint8_t *tmp; 813 int64_t total_size = 0; 814 int ret; 815 BlockDriverState *bs; 816 BlockBackend *qcow_blk; 817 QCryptoBlock *crypto = NULL; 818 819 assert(opts->driver == BLOCKDEV_DRIVER_QCOW); 820 qcow_opts = &opts->u.qcow; 821 822 /* Sanity checks */ 823 total_size = qcow_opts->size; 824 if (total_size == 0) { 825 error_setg(errp, "Image size is too small, cannot be zero length"); 826 return -EINVAL; 827 } 828 829 if (qcow_opts->has_encrypt && 830 qcow_opts->encrypt->format != Q_CRYPTO_BLOCK_FORMAT_QCOW) 831 { 832 error_setg(errp, "Unsupported encryption format"); 833 return -EINVAL; 834 } 835 836 /* Create BlockBackend to write to the image */ 837 bs = bdrv_open_blockdev_ref(qcow_opts->file, errp); 838 if (bs == NULL) { 839 return -EIO; 840 } 841 842 qcow_blk = blk_new_with_bs(bs, BLK_PERM_WRITE | BLK_PERM_RESIZE, 843 BLK_PERM_ALL, errp); 844 if (!qcow_blk) { 845 ret = -EPERM; 846 goto exit; 847 } 848 blk_set_allow_write_beyond_eof(qcow_blk, true); 849 850 /* Create image format */ 851 memset(&header, 0, sizeof(header)); 852 header.magic = cpu_to_be32(QCOW_MAGIC); 853 header.version = cpu_to_be32(QCOW_VERSION); 854 header.size = cpu_to_be64(total_size); 855 header_size = sizeof(header); 856 backing_filename_len = 0; 857 if (qcow_opts->has_backing_file) { 858 if (strcmp(qcow_opts->backing_file, "fat:")) { 859 header.backing_file_offset = cpu_to_be64(header_size); 860 backing_filename_len = strlen(qcow_opts->backing_file); 861 header.backing_file_size = cpu_to_be32(backing_filename_len); 862 header_size += backing_filename_len; 863 } else { 864 /* special backing file for vvfat */ 865 qcow_opts->has_backing_file = false; 866 } 867 header.cluster_bits = 9; /* 512 byte cluster to avoid copying 868 unmodified sectors */ 869 header.l2_bits = 12; /* 32 KB L2 tables */ 870 } else { 871 header.cluster_bits = 12; /* 4 KB clusters */ 872 header.l2_bits = 9; /* 4 KB L2 tables */ 873 } 874 header_size = (header_size + 7) & ~7; 875 shift = header.cluster_bits + header.l2_bits; 876 l1_size = (total_size + (1LL << shift) - 1) >> shift; 877 878 header.l1_table_offset = cpu_to_be64(header_size); 879 880 if (qcow_opts->has_encrypt) { 881 header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES); 882 883 crypto = qcrypto_block_create(qcow_opts->encrypt, "encrypt.", 884 NULL, NULL, NULL, errp); 885 if (!crypto) { 886 ret = -EINVAL; 887 goto exit; 888 } 889 } else { 890 header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE); 891 } 892 893 /* write all the data */ 894 ret = blk_pwrite(qcow_blk, 0, &header, sizeof(header), 0); 895 if (ret != sizeof(header)) { 896 goto exit; 897 } 898 899 if (qcow_opts->has_backing_file) { 900 ret = blk_pwrite(qcow_blk, sizeof(header), 901 qcow_opts->backing_file, backing_filename_len, 0); 902 if (ret != backing_filename_len) { 903 goto exit; 904 } 905 } 906 907 tmp = g_malloc0(BDRV_SECTOR_SIZE); 908 for (i = 0; i < DIV_ROUND_UP(sizeof(uint64_t) * l1_size, BDRV_SECTOR_SIZE); 909 i++) { 910 ret = blk_pwrite(qcow_blk, header_size + BDRV_SECTOR_SIZE * i, 911 tmp, BDRV_SECTOR_SIZE, 0); 912 if (ret != BDRV_SECTOR_SIZE) { 913 g_free(tmp); 914 goto exit; 915 } 916 } 917 918 g_free(tmp); 919 ret = 0; 920 exit: 921 blk_unref(qcow_blk); 922 bdrv_unref(bs); 923 qcrypto_block_free(crypto); 924 return ret; 925 } 926 927 static int coroutine_fn qcow_co_create_opts(BlockDriver *drv, 928 const char *filename, 929 QemuOpts *opts, Error **errp) 930 { 931 BlockdevCreateOptions *create_options = NULL; 932 BlockDriverState *bs = NULL; 933 QDict *qdict = NULL; 934 Visitor *v; 935 const char *val; 936 int ret; 937 char *backing_fmt; 938 939 static const QDictRenames opt_renames[] = { 940 { BLOCK_OPT_BACKING_FILE, "backing-file" }, 941 { BLOCK_OPT_ENCRYPT, BLOCK_OPT_ENCRYPT_FORMAT }, 942 { NULL, NULL }, 943 }; 944 945 /* 946 * We can't actually store a backing format, but can check that 947 * the user's request made sense. 948 */ 949 backing_fmt = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FMT); 950 if (backing_fmt && !bdrv_find_format(backing_fmt)) { 951 error_setg(errp, "unrecognized backing format '%s'", backing_fmt); 952 ret = -EINVAL; 953 goto fail; 954 } 955 956 /* Parse options and convert legacy syntax */ 957 qdict = qemu_opts_to_qdict_filtered(opts, NULL, &qcow_create_opts, true); 958 959 val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT); 960 if (val && !strcmp(val, "on")) { 961 qdict_put_str(qdict, BLOCK_OPT_ENCRYPT, "qcow"); 962 } else if (val && !strcmp(val, "off")) { 963 qdict_del(qdict, BLOCK_OPT_ENCRYPT); 964 } 965 966 val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT); 967 if (val && !strcmp(val, "aes")) { 968 qdict_put_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT, "qcow"); 969 } 970 971 if (!qdict_rename_keys(qdict, opt_renames, errp)) { 972 ret = -EINVAL; 973 goto fail; 974 } 975 976 /* Create and open the file (protocol layer) */ 977 ret = bdrv_create_file(filename, opts, errp); 978 if (ret < 0) { 979 goto fail; 980 } 981 982 bs = bdrv_open(filename, NULL, NULL, 983 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp); 984 if (bs == NULL) { 985 ret = -EIO; 986 goto fail; 987 } 988 989 /* Now get the QAPI type BlockdevCreateOptions */ 990 qdict_put_str(qdict, "driver", "qcow"); 991 qdict_put_str(qdict, "file", bs->node_name); 992 993 v = qobject_input_visitor_new_flat_confused(qdict, errp); 994 if (!v) { 995 ret = -EINVAL; 996 goto fail; 997 } 998 999 visit_type_BlockdevCreateOptions(v, NULL, &create_options, errp); 1000 visit_free(v); 1001 if (!create_options) { 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 g_free(backing_fmt); 1020 qobject_unref(qdict); 1021 bdrv_unref(bs); 1022 qapi_free_BlockdevCreateOptions(create_options); 1023 return ret; 1024 } 1025 1026 static int qcow_make_empty(BlockDriverState *bs) 1027 { 1028 BDRVQcowState *s = bs->opaque; 1029 uint32_t l1_length = s->l1_size * sizeof(uint64_t); 1030 int ret; 1031 1032 memset(s->l1_table, 0, l1_length); 1033 if (bdrv_pwrite_sync(bs->file, s->l1_table_offset, s->l1_table, 1034 l1_length) < 0) 1035 return -1; 1036 ret = bdrv_truncate(bs->file, s->l1_table_offset + l1_length, false, 1037 PREALLOC_MODE_OFF, 0, NULL); 1038 if (ret < 0) 1039 return ret; 1040 1041 memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t)); 1042 memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t)); 1043 memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t)); 1044 1045 return 0; 1046 } 1047 1048 /* XXX: put compressed sectors first, then all the cluster aligned 1049 tables to avoid losing bytes in alignment */ 1050 static coroutine_fn int 1051 qcow_co_pwritev_compressed(BlockDriverState *bs, int64_t offset, int64_t bytes, 1052 QEMUIOVector *qiov) 1053 { 1054 BDRVQcowState *s = bs->opaque; 1055 z_stream strm; 1056 int ret, out_len; 1057 uint8_t *buf, *out_buf; 1058 uint64_t cluster_offset; 1059 1060 buf = qemu_blockalign(bs, s->cluster_size); 1061 if (bytes != s->cluster_size) { 1062 if (bytes > s->cluster_size || 1063 offset + bytes != bs->total_sectors << BDRV_SECTOR_BITS) 1064 { 1065 qemu_vfree(buf); 1066 return -EINVAL; 1067 } 1068 /* Zero-pad last write if image size is not cluster aligned */ 1069 memset(buf + bytes, 0, s->cluster_size - bytes); 1070 } 1071 qemu_iovec_to_buf(qiov, 0, buf, qiov->size); 1072 1073 out_buf = g_malloc(s->cluster_size); 1074 1075 /* best compression, small window, no zlib header */ 1076 memset(&strm, 0, sizeof(strm)); 1077 ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION, 1078 Z_DEFLATED, -12, 1079 9, Z_DEFAULT_STRATEGY); 1080 if (ret != 0) { 1081 ret = -EINVAL; 1082 goto fail; 1083 } 1084 1085 strm.avail_in = s->cluster_size; 1086 strm.next_in = (uint8_t *)buf; 1087 strm.avail_out = s->cluster_size; 1088 strm.next_out = out_buf; 1089 1090 ret = deflate(&strm, Z_FINISH); 1091 if (ret != Z_STREAM_END && ret != Z_OK) { 1092 deflateEnd(&strm); 1093 ret = -EINVAL; 1094 goto fail; 1095 } 1096 out_len = strm.next_out - out_buf; 1097 1098 deflateEnd(&strm); 1099 1100 if (ret != Z_STREAM_END || out_len >= s->cluster_size) { 1101 /* could not compress: write normal cluster */ 1102 ret = qcow_co_pwritev(bs, offset, bytes, qiov, 0); 1103 if (ret < 0) { 1104 goto fail; 1105 } 1106 goto success; 1107 } 1108 qemu_co_mutex_lock(&s->lock); 1109 ret = get_cluster_offset(bs, offset, 2, out_len, 0, 0, &cluster_offset); 1110 qemu_co_mutex_unlock(&s->lock); 1111 if (ret < 0) { 1112 goto fail; 1113 } 1114 if (cluster_offset == 0) { 1115 ret = -EIO; 1116 goto fail; 1117 } 1118 cluster_offset &= s->cluster_offset_mask; 1119 1120 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_COMPRESSED); 1121 ret = bdrv_co_pwrite(bs->file, cluster_offset, out_len, out_buf, 0); 1122 if (ret < 0) { 1123 goto fail; 1124 } 1125 success: 1126 ret = 0; 1127 fail: 1128 qemu_vfree(buf); 1129 g_free(out_buf); 1130 return ret; 1131 } 1132 1133 static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) 1134 { 1135 BDRVQcowState *s = bs->opaque; 1136 bdi->cluster_size = s->cluster_size; 1137 return 0; 1138 } 1139 1140 static QemuOptsList qcow_create_opts = { 1141 .name = "qcow-create-opts", 1142 .head = QTAILQ_HEAD_INITIALIZER(qcow_create_opts.head), 1143 .desc = { 1144 { 1145 .name = BLOCK_OPT_SIZE, 1146 .type = QEMU_OPT_SIZE, 1147 .help = "Virtual disk size" 1148 }, 1149 { 1150 .name = BLOCK_OPT_BACKING_FILE, 1151 .type = QEMU_OPT_STRING, 1152 .help = "File name of a base image" 1153 }, 1154 { 1155 .name = BLOCK_OPT_BACKING_FMT, 1156 .type = QEMU_OPT_STRING, 1157 .help = "Format of the backing image", 1158 }, 1159 { 1160 .name = BLOCK_OPT_ENCRYPT, 1161 .type = QEMU_OPT_BOOL, 1162 .help = "Encrypt the image with format 'aes'. (Deprecated " 1163 "in favor of " BLOCK_OPT_ENCRYPT_FORMAT "=aes)", 1164 }, 1165 { 1166 .name = BLOCK_OPT_ENCRYPT_FORMAT, 1167 .type = QEMU_OPT_STRING, 1168 .help = "Encrypt the image, format choices: 'aes'", 1169 }, 1170 BLOCK_CRYPTO_OPT_DEF_QCOW_KEY_SECRET("encrypt."), 1171 { /* end of list */ } 1172 } 1173 }; 1174 1175 static const char *const qcow_strong_runtime_opts[] = { 1176 "encrypt." BLOCK_CRYPTO_OPT_QCOW_KEY_SECRET, 1177 1178 NULL 1179 }; 1180 1181 static BlockDriver bdrv_qcow = { 1182 .format_name = "qcow", 1183 .instance_size = sizeof(BDRVQcowState), 1184 .bdrv_probe = qcow_probe, 1185 .bdrv_open = qcow_open, 1186 .bdrv_close = qcow_close, 1187 .bdrv_child_perm = bdrv_default_perms, 1188 .bdrv_reopen_prepare = qcow_reopen_prepare, 1189 .bdrv_co_create = qcow_co_create, 1190 .bdrv_co_create_opts = qcow_co_create_opts, 1191 .bdrv_has_zero_init = bdrv_has_zero_init_1, 1192 .is_format = true, 1193 .supports_backing = true, 1194 .bdrv_refresh_limits = qcow_refresh_limits, 1195 1196 .bdrv_co_preadv = qcow_co_preadv, 1197 .bdrv_co_pwritev = qcow_co_pwritev, 1198 .bdrv_co_block_status = qcow_co_block_status, 1199 1200 .bdrv_make_empty = qcow_make_empty, 1201 .bdrv_co_pwritev_compressed = qcow_co_pwritev_compressed, 1202 .bdrv_get_info = qcow_get_info, 1203 1204 .create_opts = &qcow_create_opts, 1205 .strong_runtime_opts = qcow_strong_runtime_opts, 1206 }; 1207 1208 static void bdrv_qcow_init(void) 1209 { 1210 bdrv_register(&bdrv_qcow); 1211 } 1212 1213 block_init(bdrv_qcow_init); 1214