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 coroutine_fn decompress_cluster(BlockDriverState *bs, 96 uint64_t cluster_offset); 97 98 static int qcow_probe(const uint8_t *buf, int buf_size, const char *filename) 99 { 100 const QCowHeader *cow_header = (const void *)buf; 101 102 if (buf_size >= sizeof(QCowHeader) && 103 be32_to_cpu(cow_header->magic) == QCOW_MAGIC && 104 be32_to_cpu(cow_header->version) == QCOW_VERSION) 105 return 100; 106 else 107 return 0; 108 } 109 110 static int qcow_open(BlockDriverState *bs, QDict *options, int flags, 111 Error **errp) 112 { 113 BDRVQcowState *s = bs->opaque; 114 unsigned int len, i, shift; 115 int ret; 116 QCowHeader header; 117 QCryptoBlockOpenOptions *crypto_opts = NULL; 118 unsigned int cflags = 0; 119 QDict *encryptopts = NULL; 120 const char *encryptfmt; 121 122 qdict_extract_subqdict(options, &encryptopts, "encrypt."); 123 encryptfmt = qdict_get_try_str(encryptopts, "format"); 124 125 ret = bdrv_open_file_child(NULL, options, "file", bs, errp); 126 if (ret < 0) { 127 goto fail; 128 } 129 130 ret = bdrv_pread(bs->file, 0, sizeof(header), &header, 0); 131 if (ret < 0) { 132 goto fail; 133 } 134 header.magic = be32_to_cpu(header.magic); 135 header.version = be32_to_cpu(header.version); 136 header.backing_file_offset = be64_to_cpu(header.backing_file_offset); 137 header.backing_file_size = be32_to_cpu(header.backing_file_size); 138 header.mtime = be32_to_cpu(header.mtime); 139 header.size = be64_to_cpu(header.size); 140 header.crypt_method = be32_to_cpu(header.crypt_method); 141 header.l1_table_offset = be64_to_cpu(header.l1_table_offset); 142 143 if (header.magic != QCOW_MAGIC) { 144 error_setg(errp, "Image not in qcow format"); 145 ret = -EINVAL; 146 goto fail; 147 } 148 if (header.version != QCOW_VERSION) { 149 error_setg(errp, "qcow (v%d) does not support qcow version %" PRIu32, 150 QCOW_VERSION, header.version); 151 if (header.version == 2 || header.version == 3) { 152 error_append_hint(errp, "Try the 'qcow2' driver instead.\n"); 153 } 154 155 ret = -ENOTSUP; 156 goto fail; 157 } 158 159 if (header.size <= 1) { 160 error_setg(errp, "Image size is too small (must be at least 2 bytes)"); 161 ret = -EINVAL; 162 goto fail; 163 } 164 if (header.cluster_bits < 9 || header.cluster_bits > 16) { 165 error_setg(errp, "Cluster size must be between 512 and 64k"); 166 ret = -EINVAL; 167 goto fail; 168 } 169 170 /* l2_bits specifies number of entries; storing a uint64_t in each entry, 171 * so bytes = num_entries << 3. */ 172 if (header.l2_bits < 9 - 3 || header.l2_bits > 16 - 3) { 173 error_setg(errp, "L2 table size must be between 512 and 64k"); 174 ret = -EINVAL; 175 goto fail; 176 } 177 178 s->crypt_method_header = header.crypt_method; 179 if (s->crypt_method_header) { 180 if (bdrv_uses_whitelist() && 181 s->crypt_method_header == QCOW_CRYPT_AES) { 182 error_setg(errp, 183 "Use of AES-CBC encrypted qcow images is no longer " 184 "supported in system emulators"); 185 error_append_hint(errp, 186 "You can use 'qemu-img convert' to convert your " 187 "image to an alternative supported format, such " 188 "as unencrypted qcow, or raw with the LUKS " 189 "format instead.\n"); 190 ret = -ENOSYS; 191 goto fail; 192 } 193 if (s->crypt_method_header == QCOW_CRYPT_AES) { 194 if (encryptfmt && !g_str_equal(encryptfmt, "aes")) { 195 error_setg(errp, 196 "Header reported 'aes' encryption format but " 197 "options specify '%s'", encryptfmt); 198 ret = -EINVAL; 199 goto fail; 200 } 201 qdict_put_str(encryptopts, "format", "qcow"); 202 crypto_opts = block_crypto_open_opts_init(encryptopts, errp); 203 if (!crypto_opts) { 204 ret = -EINVAL; 205 goto fail; 206 } 207 208 if (flags & BDRV_O_NO_IO) { 209 cflags |= QCRYPTO_BLOCK_OPEN_NO_IO; 210 } 211 s->crypto = qcrypto_block_open(crypto_opts, "encrypt.", 212 NULL, NULL, cflags, 1, errp); 213 if (!s->crypto) { 214 ret = -EINVAL; 215 goto fail; 216 } 217 } else { 218 error_setg(errp, "invalid encryption method in qcow header"); 219 ret = -EINVAL; 220 goto fail; 221 } 222 bs->encrypted = true; 223 } else { 224 if (encryptfmt) { 225 error_setg(errp, "No encryption in image header, but options " 226 "specified format '%s'", encryptfmt); 227 ret = -EINVAL; 228 goto fail; 229 } 230 } 231 s->cluster_bits = header.cluster_bits; 232 s->cluster_size = 1 << s->cluster_bits; 233 s->l2_bits = header.l2_bits; 234 s->l2_size = 1 << s->l2_bits; 235 bs->total_sectors = header.size / 512; 236 s->cluster_offset_mask = (1LL << (63 - s->cluster_bits)) - 1; 237 238 /* read the level 1 table */ 239 shift = s->cluster_bits + s->l2_bits; 240 if (header.size > UINT64_MAX - (1LL << shift)) { 241 error_setg(errp, "Image too large"); 242 ret = -EINVAL; 243 goto fail; 244 } else { 245 uint64_t l1_size = (header.size + (1LL << shift) - 1) >> shift; 246 if (l1_size > INT_MAX / sizeof(uint64_t)) { 247 error_setg(errp, "Image too large"); 248 ret = -EINVAL; 249 goto fail; 250 } 251 s->l1_size = l1_size; 252 } 253 254 s->l1_table_offset = header.l1_table_offset; 255 s->l1_table = g_try_new(uint64_t, s->l1_size); 256 if (s->l1_table == NULL) { 257 error_setg(errp, "Could not allocate memory for L1 table"); 258 ret = -ENOMEM; 259 goto fail; 260 } 261 262 ret = bdrv_pread(bs->file, s->l1_table_offset, 263 s->l1_size * sizeof(uint64_t), s->l1_table, 0); 264 if (ret < 0) { 265 goto fail; 266 } 267 268 for(i = 0;i < s->l1_size; i++) { 269 s->l1_table[i] = be64_to_cpu(s->l1_table[i]); 270 } 271 272 /* alloc L2 cache (max. 64k * 16 * 8 = 8 MB) */ 273 s->l2_cache = 274 qemu_try_blockalign(bs->file->bs, 275 s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t)); 276 if (s->l2_cache == NULL) { 277 error_setg(errp, "Could not allocate L2 table cache"); 278 ret = -ENOMEM; 279 goto fail; 280 } 281 s->cluster_cache = g_malloc(s->cluster_size); 282 s->cluster_data = g_malloc(s->cluster_size); 283 s->cluster_cache_offset = -1; 284 285 /* read the backing file name */ 286 if (header.backing_file_offset != 0) { 287 len = header.backing_file_size; 288 if (len > 1023 || len >= sizeof(bs->backing_file)) { 289 error_setg(errp, "Backing file name too long"); 290 ret = -EINVAL; 291 goto fail; 292 } 293 ret = bdrv_pread(bs->file, header.backing_file_offset, len, 294 bs->auto_backing_file, 0); 295 if (ret < 0) { 296 goto fail; 297 } 298 bs->auto_backing_file[len] = '\0'; 299 pstrcpy(bs->backing_file, sizeof(bs->backing_file), 300 bs->auto_backing_file); 301 } 302 303 /* Disable migration when qcow images are used */ 304 error_setg(&s->migration_blocker, "The qcow format used by node '%s' " 305 "does not support live migration", 306 bdrv_get_device_or_node_name(bs)); 307 ret = migrate_add_blocker(s->migration_blocker, errp); 308 if (ret < 0) { 309 error_free(s->migration_blocker); 310 goto fail; 311 } 312 313 qobject_unref(encryptopts); 314 qapi_free_QCryptoBlockOpenOptions(crypto_opts); 315 qemu_co_mutex_init(&s->lock); 316 return 0; 317 318 fail: 319 g_free(s->l1_table); 320 qemu_vfree(s->l2_cache); 321 g_free(s->cluster_cache); 322 g_free(s->cluster_data); 323 qcrypto_block_free(s->crypto); 324 qobject_unref(encryptopts); 325 qapi_free_QCryptoBlockOpenOptions(crypto_opts); 326 return ret; 327 } 328 329 330 /* We have nothing to do for QCOW reopen, stubs just return 331 * success */ 332 static int qcow_reopen_prepare(BDRVReopenState *state, 333 BlockReopenQueue *queue, Error **errp) 334 { 335 return 0; 336 } 337 338 339 /* 'allocate' is: 340 * 341 * 0 to not allocate. 342 * 343 * 1 to allocate a normal cluster (for sector-aligned byte offsets 'n_start' 344 * to 'n_end' within the cluster) 345 * 346 * 2 to allocate a compressed cluster of size 347 * 'compressed_size'. 'compressed_size' must be > 0 and < 348 * cluster_size 349 * 350 * return 0 if not allocated, 1 if *result is assigned, and negative 351 * errno on failure. 352 */ 353 static int coroutine_fn get_cluster_offset(BlockDriverState *bs, 354 uint64_t offset, int allocate, 355 int compressed_size, 356 int n_start, int n_end, 357 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_co_pwrite_sync(bs->file, 385 s->l1_table_offset + l1_index * sizeof(tmp), 386 sizeof(tmp), &tmp, 0); 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_co_pwrite_sync(bs->file, l2_offset, 418 s->l2_size * sizeof(uint64_t), l2_table, 0); 419 if (ret < 0) { 420 return ret; 421 } 422 } else { 423 ret = bdrv_co_pread(bs->file, l2_offset, 424 s->l2_size * sizeof(uint64_t), l2_table, 0); 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_co_pwrite(bs->file, cluster_offset, s->cluster_size, 457 s->cluster_cache, 0); 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_co_truncate(bs->file, 473 cluster_offset + s->cluster_size, 474 false, PREALLOC_MODE_OFF, 0, NULL); 475 if (ret < 0) { 476 return ret; 477 } 478 /* if encrypted, we must initialize the cluster 479 content which won't be written */ 480 if (bs->encrypted && 481 (n_end - n_start) < s->cluster_size) { 482 uint64_t start_offset; 483 assert(s->crypto); 484 start_offset = offset & ~(s->cluster_size - 1); 485 for (i = 0; i < s->cluster_size; i += BDRV_SECTOR_SIZE) { 486 if (i < n_start || i >= n_end) { 487 memset(s->cluster_data, 0x00, BDRV_SECTOR_SIZE); 488 if (qcrypto_block_encrypt(s->crypto, 489 start_offset + i, 490 s->cluster_data, 491 BDRV_SECTOR_SIZE, 492 NULL) < 0) { 493 return -EIO; 494 } 495 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO); 496 ret = bdrv_co_pwrite(bs->file, cluster_offset + i, 497 BDRV_SECTOR_SIZE, 498 s->cluster_data, 0); 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_co_pwrite_sync(bs->file, l2_offset + l2_index * sizeof(tmp), 519 sizeof(tmp), &tmp, 0); 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 coroutine_fn decompress_cluster(BlockDriverState *bs, 590 uint64_t cluster_offset) 591 { 592 BDRVQcowState *s = bs->opaque; 593 int ret, csize; 594 uint64_t coffset; 595 596 coffset = cluster_offset & s->cluster_offset_mask; 597 if (s->cluster_cache_offset != coffset) { 598 csize = cluster_offset >> (63 - s->cluster_bits); 599 csize &= (s->cluster_size - 1); 600 BLKDBG_EVENT(bs->file, BLKDBG_READ_COMPRESSED); 601 ret = bdrv_co_pread(bs->file, coffset, csize, s->cluster_data, 0); 602 if (ret < 0) 603 return -1; 604 if (decompress_buffer(s->cluster_cache, s->cluster_size, 605 s->cluster_data, csize) < 0) { 606 return -1; 607 } 608 s->cluster_cache_offset = coffset; 609 } 610 return 0; 611 } 612 613 static void qcow_refresh_limits(BlockDriverState *bs, Error **errp) 614 { 615 /* At least encrypted images require 512-byte alignment. Apply the 616 * limit universally, rather than just on encrypted images, as 617 * it's easier to let the block layer handle rounding than to 618 * audit this code further. */ 619 bs->bl.request_alignment = BDRV_SECTOR_SIZE; 620 } 621 622 static coroutine_fn int qcow_co_preadv(BlockDriverState *bs, int64_t offset, 623 int64_t bytes, QEMUIOVector *qiov, 624 BdrvRequestFlags flags) 625 { 626 BDRVQcowState *s = bs->opaque; 627 int offset_in_cluster; 628 int ret = 0, n; 629 uint64_t cluster_offset; 630 uint8_t *buf; 631 void *orig_buf; 632 633 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 s->cluster_cache_offset = -1; /* disable compressed cache */ 730 731 /* We must always copy the iov when encrypting, so we 732 * don't modify the original data buffer during encryption */ 733 if (bs->encrypted || qiov->niov > 1) { 734 buf = orig_buf = qemu_try_blockalign(bs, qiov->size); 735 if (buf == NULL) { 736 return -ENOMEM; 737 } 738 qemu_iovec_to_buf(qiov, 0, buf, qiov->size); 739 } else { 740 orig_buf = NULL; 741 buf = (uint8_t *)qiov->iov->iov_base; 742 } 743 744 qemu_co_mutex_lock(&s->lock); 745 746 while (bytes != 0) { 747 offset_in_cluster = offset & (s->cluster_size - 1); 748 n = s->cluster_size - offset_in_cluster; 749 if (n > bytes) { 750 n = bytes; 751 } 752 ret = get_cluster_offset(bs, offset, 1, 0, offset_in_cluster, 753 offset_in_cluster + n, &cluster_offset); 754 if (ret < 0) { 755 break; 756 } 757 if (!cluster_offset || (cluster_offset & 511) != 0) { 758 ret = -EIO; 759 break; 760 } 761 if (bs->encrypted) { 762 assert(s->crypto); 763 if (qcrypto_block_encrypt(s->crypto, offset, buf, n, NULL) < 0) { 764 ret = -EIO; 765 break; 766 } 767 } 768 769 qemu_co_mutex_unlock(&s->lock); 770 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO); 771 ret = bdrv_co_pwrite(bs->file, cluster_offset + offset_in_cluster, 772 n, buf, 0); 773 qemu_co_mutex_lock(&s->lock); 774 if (ret < 0) { 775 break; 776 } 777 ret = 0; 778 779 bytes -= n; 780 offset += n; 781 buf += n; 782 } 783 qemu_co_mutex_unlock(&s->lock); 784 785 qemu_vfree(orig_buf); 786 787 return ret; 788 } 789 790 static void qcow_close(BlockDriverState *bs) 791 { 792 BDRVQcowState *s = bs->opaque; 793 794 qcrypto_block_free(s->crypto); 795 s->crypto = NULL; 796 g_free(s->l1_table); 797 qemu_vfree(s->l2_cache); 798 g_free(s->cluster_cache); 799 g_free(s->cluster_data); 800 801 migrate_del_blocker(s->migration_blocker); 802 error_free(s->migration_blocker); 803 } 804 805 static int coroutine_fn qcow_co_create(BlockdevCreateOptions *opts, 806 Error **errp) 807 { 808 BlockdevCreateOptionsQcow *qcow_opts; 809 int header_size, backing_filename_len, l1_size, shift, i; 810 QCowHeader header; 811 uint8_t *tmp; 812 int64_t total_size = 0; 813 int ret; 814 BlockDriverState *bs; 815 BlockBackend *qcow_blk; 816 QCryptoBlock *crypto = NULL; 817 818 assert(opts->driver == BLOCKDEV_DRIVER_QCOW); 819 qcow_opts = &opts->u.qcow; 820 821 /* Sanity checks */ 822 total_size = qcow_opts->size; 823 if (total_size == 0) { 824 error_setg(errp, "Image size is too small, cannot be zero length"); 825 return -EINVAL; 826 } 827 828 if (qcow_opts->encrypt && 829 qcow_opts->encrypt->format != Q_CRYPTO_BLOCK_FORMAT_QCOW) 830 { 831 error_setg(errp, "Unsupported encryption format"); 832 return -EINVAL; 833 } 834 835 /* Create BlockBackend to write to the image */ 836 bs = bdrv_open_blockdev_ref(qcow_opts->file, errp); 837 if (bs == NULL) { 838 return -EIO; 839 } 840 841 qcow_blk = blk_new_with_bs(bs, BLK_PERM_WRITE | BLK_PERM_RESIZE, 842 BLK_PERM_ALL, errp); 843 if (!qcow_blk) { 844 ret = -EPERM; 845 goto exit; 846 } 847 blk_set_allow_write_beyond_eof(qcow_blk, true); 848 849 /* Create image format */ 850 memset(&header, 0, sizeof(header)); 851 header.magic = cpu_to_be32(QCOW_MAGIC); 852 header.version = cpu_to_be32(QCOW_VERSION); 853 header.size = cpu_to_be64(total_size); 854 header_size = sizeof(header); 855 backing_filename_len = 0; 856 if (qcow_opts->backing_file) { 857 if (strcmp(qcow_opts->backing_file, "fat:")) { 858 header.backing_file_offset = cpu_to_be64(header_size); 859 backing_filename_len = strlen(qcow_opts->backing_file); 860 header.backing_file_size = cpu_to_be32(backing_filename_len); 861 header_size += backing_filename_len; 862 } else { 863 /* special backing file for vvfat */ 864 qcow_opts->backing_file = NULL; 865 } 866 header.cluster_bits = 9; /* 512 byte cluster to avoid copying 867 unmodified sectors */ 868 header.l2_bits = 12; /* 32 KB L2 tables */ 869 } else { 870 header.cluster_bits = 12; /* 4 KB clusters */ 871 header.l2_bits = 9; /* 4 KB L2 tables */ 872 } 873 header_size = (header_size + 7) & ~7; 874 shift = header.cluster_bits + header.l2_bits; 875 l1_size = (total_size + (1LL << shift) - 1) >> shift; 876 877 header.l1_table_offset = cpu_to_be64(header_size); 878 879 if (qcow_opts->encrypt) { 880 header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES); 881 882 crypto = qcrypto_block_create(qcow_opts->encrypt, "encrypt.", 883 NULL, NULL, NULL, errp); 884 if (!crypto) { 885 ret = -EINVAL; 886 goto exit; 887 } 888 } else { 889 header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE); 890 } 891 892 /* write all the data */ 893 ret = blk_co_pwrite(qcow_blk, 0, sizeof(header), &header, 0); 894 if (ret < 0) { 895 goto exit; 896 } 897 898 if (qcow_opts->backing_file) { 899 ret = blk_co_pwrite(qcow_blk, sizeof(header), backing_filename_len, 900 qcow_opts->backing_file, 0); 901 if (ret < 0) { 902 goto exit; 903 } 904 } 905 906 tmp = g_malloc0(BDRV_SECTOR_SIZE); 907 for (i = 0; i < DIV_ROUND_UP(sizeof(uint64_t) * l1_size, BDRV_SECTOR_SIZE); 908 i++) { 909 ret = blk_co_pwrite(qcow_blk, header_size + BDRV_SECTOR_SIZE * i, 910 BDRV_SECTOR_SIZE, tmp, 0); 911 if (ret < 0) { 912 g_free(tmp); 913 goto exit; 914 } 915 } 916 917 g_free(tmp); 918 ret = 0; 919 exit: 920 blk_unref(qcow_blk); 921 bdrv_unref(bs); 922 qcrypto_block_free(crypto); 923 return ret; 924 } 925 926 static int coroutine_fn qcow_co_create_opts(BlockDriver *drv, 927 const char *filename, 928 QemuOpts *opts, Error **errp) 929 { 930 BlockdevCreateOptions *create_options = NULL; 931 BlockDriverState *bs = NULL; 932 QDict *qdict = NULL; 933 Visitor *v; 934 const char *val; 935 int ret; 936 char *backing_fmt; 937 938 static const QDictRenames opt_renames[] = { 939 { BLOCK_OPT_BACKING_FILE, "backing-file" }, 940 { BLOCK_OPT_ENCRYPT, BLOCK_OPT_ENCRYPT_FORMAT }, 941 { NULL, NULL }, 942 }; 943 944 /* 945 * We can't actually store a backing format, but can check that 946 * the user's request made sense. 947 */ 948 backing_fmt = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FMT); 949 if (backing_fmt && !bdrv_find_format(backing_fmt)) { 950 error_setg(errp, "unrecognized backing format '%s'", backing_fmt); 951 ret = -EINVAL; 952 goto fail; 953 } 954 955 /* Parse options and convert legacy syntax */ 956 qdict = qemu_opts_to_qdict_filtered(opts, NULL, &qcow_create_opts, true); 957 958 val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT); 959 if (val && !strcmp(val, "on")) { 960 qdict_put_str(qdict, BLOCK_OPT_ENCRYPT, "qcow"); 961 } else if (val && !strcmp(val, "off")) { 962 qdict_del(qdict, BLOCK_OPT_ENCRYPT); 963 } 964 965 val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT); 966 if (val && !strcmp(val, "aes")) { 967 qdict_put_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT, "qcow"); 968 } 969 970 if (!qdict_rename_keys(qdict, opt_renames, errp)) { 971 ret = -EINVAL; 972 goto fail; 973 } 974 975 /* Create and open the file (protocol layer) */ 976 ret = bdrv_create_file(filename, opts, errp); 977 if (ret < 0) { 978 goto fail; 979 } 980 981 bs = bdrv_open(filename, NULL, NULL, 982 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp); 983 if (bs == NULL) { 984 ret = -EIO; 985 goto fail; 986 } 987 988 /* Now get the QAPI type BlockdevCreateOptions */ 989 qdict_put_str(qdict, "driver", "qcow"); 990 qdict_put_str(qdict, "file", bs->node_name); 991 992 v = qobject_input_visitor_new_flat_confused(qdict, errp); 993 if (!v) { 994 ret = -EINVAL; 995 goto fail; 996 } 997 998 visit_type_BlockdevCreateOptions(v, NULL, &create_options, errp); 999 visit_free(v); 1000 if (!create_options) { 1001 ret = -EINVAL; 1002 goto fail; 1003 } 1004 1005 /* Silently round up size */ 1006 assert(create_options->driver == BLOCKDEV_DRIVER_QCOW); 1007 create_options->u.qcow.size = 1008 ROUND_UP(create_options->u.qcow.size, BDRV_SECTOR_SIZE); 1009 1010 /* Create the qcow image (format layer) */ 1011 ret = qcow_co_create(create_options, errp); 1012 if (ret < 0) { 1013 goto fail; 1014 } 1015 1016 ret = 0; 1017 fail: 1018 g_free(backing_fmt); 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, l1_length, s->l1_table, 1033 0) < 0) 1034 return -1; 1035 ret = bdrv_truncate(bs->file, s->l1_table_offset + l1_length, false, 1036 PREALLOC_MODE_OFF, 0, 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, int64_t offset, int64_t bytes, 1051 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_BACKING_FMT, 1155 .type = QEMU_OPT_STRING, 1156 .help = "Format of the backing image", 1157 }, 1158 { 1159 .name = BLOCK_OPT_ENCRYPT, 1160 .type = QEMU_OPT_BOOL, 1161 .help = "Encrypt the image with format 'aes'. (Deprecated " 1162 "in favor of " BLOCK_OPT_ENCRYPT_FORMAT "=aes)", 1163 }, 1164 { 1165 .name = BLOCK_OPT_ENCRYPT_FORMAT, 1166 .type = QEMU_OPT_STRING, 1167 .help = "Encrypt the image, format choices: 'aes'", 1168 }, 1169 BLOCK_CRYPTO_OPT_DEF_QCOW_KEY_SECRET("encrypt."), 1170 { /* end of list */ } 1171 } 1172 }; 1173 1174 static const char *const qcow_strong_runtime_opts[] = { 1175 "encrypt." BLOCK_CRYPTO_OPT_QCOW_KEY_SECRET, 1176 1177 NULL 1178 }; 1179 1180 static BlockDriver bdrv_qcow = { 1181 .format_name = "qcow", 1182 .instance_size = sizeof(BDRVQcowState), 1183 .bdrv_probe = qcow_probe, 1184 .bdrv_open = qcow_open, 1185 .bdrv_close = qcow_close, 1186 .bdrv_child_perm = bdrv_default_perms, 1187 .bdrv_reopen_prepare = qcow_reopen_prepare, 1188 .bdrv_co_create = qcow_co_create, 1189 .bdrv_co_create_opts = qcow_co_create_opts, 1190 .bdrv_has_zero_init = bdrv_has_zero_init_1, 1191 .is_format = true, 1192 .supports_backing = true, 1193 .bdrv_refresh_limits = qcow_refresh_limits, 1194 1195 .bdrv_co_preadv = qcow_co_preadv, 1196 .bdrv_co_pwritev = qcow_co_pwritev, 1197 .bdrv_co_block_status = qcow_co_block_status, 1198 1199 .bdrv_make_empty = qcow_make_empty, 1200 .bdrv_co_pwritev_compressed = qcow_co_pwritev_compressed, 1201 .bdrv_get_info = qcow_get_info, 1202 1203 .create_opts = &qcow_create_opts, 1204 .strong_runtime_opts = qcow_strong_runtime_opts, 1205 }; 1206 1207 static void bdrv_qcow_init(void) 1208 { 1209 bdrv_register(&bdrv_qcow); 1210 } 1211 1212 block_init(bdrv_qcow_init); 1213