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 GRAPH_RDLOCK 96 decompress_cluster(BlockDriverState *bs, 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 GRAPH_RDLOCK 354 get_cluster_offset(BlockDriverState *bs, uint64_t offset, int allocate, 355 int compressed_size, int n_start, int n_end, 356 uint64_t *result) 357 { 358 BDRVQcowState *s = bs->opaque; 359 int min_index, i, j, l1_index, l2_index, ret; 360 int64_t l2_offset; 361 uint64_t *l2_table, cluster_offset, tmp; 362 uint32_t min_count; 363 int new_l2_table; 364 365 *result = 0; 366 l1_index = offset >> (s->l2_bits + s->cluster_bits); 367 l2_offset = s->l1_table[l1_index]; 368 new_l2_table = 0; 369 if (!l2_offset) { 370 if (!allocate) 371 return 0; 372 /* allocate a new l2 entry */ 373 l2_offset = bdrv_co_getlength(bs->file->bs); 374 if (l2_offset < 0) { 375 return l2_offset; 376 } 377 /* round to cluster size */ 378 l2_offset = QEMU_ALIGN_UP(l2_offset, s->cluster_size); 379 /* update the L1 entry */ 380 s->l1_table[l1_index] = l2_offset; 381 tmp = cpu_to_be64(l2_offset); 382 BLKDBG_CO_EVENT(bs->file, BLKDBG_L1_UPDATE); 383 ret = bdrv_co_pwrite_sync(bs->file, 384 s->l1_table_offset + l1_index * sizeof(tmp), 385 sizeof(tmp), &tmp, 0); 386 if (ret < 0) { 387 return ret; 388 } 389 new_l2_table = 1; 390 } 391 for(i = 0; i < L2_CACHE_SIZE; i++) { 392 if (l2_offset == s->l2_cache_offsets[i]) { 393 /* increment the hit count */ 394 if (++s->l2_cache_counts[i] == 0xffffffff) { 395 for(j = 0; j < L2_CACHE_SIZE; j++) { 396 s->l2_cache_counts[j] >>= 1; 397 } 398 } 399 l2_table = s->l2_cache + (i << s->l2_bits); 400 goto found; 401 } 402 } 403 /* not found: load a new entry in the least used one */ 404 min_index = 0; 405 min_count = 0xffffffff; 406 for(i = 0; i < L2_CACHE_SIZE; i++) { 407 if (s->l2_cache_counts[i] < min_count) { 408 min_count = s->l2_cache_counts[i]; 409 min_index = i; 410 } 411 } 412 l2_table = s->l2_cache + (min_index << s->l2_bits); 413 BLKDBG_CO_EVENT(bs->file, BLKDBG_L2_LOAD); 414 if (new_l2_table) { 415 memset(l2_table, 0, s->l2_size * sizeof(uint64_t)); 416 ret = bdrv_co_pwrite_sync(bs->file, l2_offset, 417 s->l2_size * sizeof(uint64_t), l2_table, 0); 418 if (ret < 0) { 419 return ret; 420 } 421 } else { 422 ret = bdrv_co_pread(bs->file, l2_offset, 423 s->l2_size * sizeof(uint64_t), l2_table, 0); 424 if (ret < 0) { 425 return ret; 426 } 427 } 428 s->l2_cache_offsets[min_index] = l2_offset; 429 s->l2_cache_counts[min_index] = 1; 430 found: 431 l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1); 432 cluster_offset = be64_to_cpu(l2_table[l2_index]); 433 if (!cluster_offset || 434 ((cluster_offset & QCOW_OFLAG_COMPRESSED) && allocate == 1)) { 435 if (!allocate) 436 return 0; 437 BLKDBG_CO_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC); 438 assert(QEMU_IS_ALIGNED(n_start | n_end, BDRV_SECTOR_SIZE)); 439 /* allocate a new cluster */ 440 if ((cluster_offset & QCOW_OFLAG_COMPRESSED) && 441 (n_end - n_start) < s->cluster_size) { 442 /* if the cluster is already compressed, we must 443 decompress it in the case it is not completely 444 overwritten */ 445 if (decompress_cluster(bs, cluster_offset) < 0) { 446 return -EIO; 447 } 448 cluster_offset = bdrv_co_getlength(bs->file->bs); 449 if ((int64_t) cluster_offset < 0) { 450 return cluster_offset; 451 } 452 cluster_offset = QEMU_ALIGN_UP(cluster_offset, s->cluster_size); 453 /* write the cluster content */ 454 BLKDBG_CO_EVENT(bs->file, BLKDBG_WRITE_AIO); 455 ret = bdrv_co_pwrite(bs->file, cluster_offset, s->cluster_size, 456 s->cluster_cache, 0); 457 if (ret < 0) { 458 return ret; 459 } 460 } else { 461 cluster_offset = bdrv_co_getlength(bs->file->bs); 462 if ((int64_t) cluster_offset < 0) { 463 return cluster_offset; 464 } 465 if (allocate == 1) { 466 /* round to cluster size */ 467 cluster_offset = QEMU_ALIGN_UP(cluster_offset, s->cluster_size); 468 if (cluster_offset + s->cluster_size > INT64_MAX) { 469 return -E2BIG; 470 } 471 ret = bdrv_co_truncate(bs->file, 472 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_CO_EVENT(bs->file, BLKDBG_WRITE_AIO); 495 ret = bdrv_co_pwrite(bs->file, cluster_offset + i, 496 BDRV_SECTOR_SIZE, 497 s->cluster_data, 0); 498 if (ret < 0) { 499 return ret; 500 } 501 } 502 } 503 } 504 } else if (allocate == 2) { 505 cluster_offset |= QCOW_OFLAG_COMPRESSED | 506 (uint64_t)compressed_size << (63 - s->cluster_bits); 507 } 508 } 509 /* update L2 table */ 510 tmp = cpu_to_be64(cluster_offset); 511 l2_table[l2_index] = tmp; 512 if (allocate == 2) { 513 BLKDBG_CO_EVENT(bs->file, BLKDBG_L2_UPDATE_COMPRESSED); 514 } else { 515 BLKDBG_CO_EVENT(bs->file, BLKDBG_L2_UPDATE); 516 } 517 ret = bdrv_co_pwrite_sync(bs->file, l2_offset + l2_index * sizeof(tmp), 518 sizeof(tmp), &tmp, 0); 519 if (ret < 0) { 520 return ret; 521 } 522 } 523 *result = cluster_offset; 524 return 1; 525 } 526 527 static int coroutine_fn GRAPH_RDLOCK 528 qcow_co_block_status(BlockDriverState *bs, bool want_zero, 529 int64_t offset, int64_t bytes, int64_t *pnum, 530 int64_t *map, BlockDriverState **file) 531 { 532 BDRVQcowState *s = bs->opaque; 533 int index_in_cluster, ret; 534 int64_t n; 535 uint64_t cluster_offset; 536 537 qemu_co_mutex_lock(&s->lock); 538 ret = get_cluster_offset(bs, offset, 0, 0, 0, 0, &cluster_offset); 539 qemu_co_mutex_unlock(&s->lock); 540 if (ret < 0) { 541 return ret; 542 } 543 index_in_cluster = offset & (s->cluster_size - 1); 544 n = s->cluster_size - index_in_cluster; 545 if (n > bytes) { 546 n = bytes; 547 } 548 *pnum = n; 549 if (!cluster_offset) { 550 return 0; 551 } 552 if (cluster_offset & QCOW_OFLAG_COMPRESSED) { 553 return BDRV_BLOCK_DATA | BDRV_BLOCK_COMPRESSED; 554 } 555 if (s->crypto) { 556 return BDRV_BLOCK_DATA; 557 } 558 *map = cluster_offset | index_in_cluster; 559 *file = bs->file->bs; 560 return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID; 561 } 562 563 static int decompress_buffer(uint8_t *out_buf, int out_buf_size, 564 const uint8_t *buf, int buf_size) 565 { 566 z_stream strm1, *strm = &strm1; 567 int ret, out_len; 568 569 memset(strm, 0, sizeof(*strm)); 570 571 strm->next_in = (uint8_t *)buf; 572 strm->avail_in = buf_size; 573 strm->next_out = out_buf; 574 strm->avail_out = out_buf_size; 575 576 ret = inflateInit2(strm, -12); 577 if (ret != Z_OK) 578 return -1; 579 ret = inflate(strm, Z_FINISH); 580 out_len = strm->next_out - out_buf; 581 if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) || 582 out_len != out_buf_size) { 583 inflateEnd(strm); 584 return -1; 585 } 586 inflateEnd(strm); 587 return 0; 588 } 589 590 static int coroutine_fn GRAPH_RDLOCK 591 decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset) 592 { 593 BDRVQcowState *s = bs->opaque; 594 int ret, csize; 595 uint64_t coffset; 596 597 coffset = cluster_offset & s->cluster_offset_mask; 598 if (s->cluster_cache_offset != coffset) { 599 csize = cluster_offset >> (63 - s->cluster_bits); 600 csize &= (s->cluster_size - 1); 601 BLKDBG_CO_EVENT(bs->file, BLKDBG_READ_COMPRESSED); 602 ret = bdrv_co_pread(bs->file, coffset, csize, s->cluster_data, 0); 603 if (ret < 0) 604 return -1; 605 if (decompress_buffer(s->cluster_cache, s->cluster_size, 606 s->cluster_data, csize) < 0) { 607 return -1; 608 } 609 s->cluster_cache_offset = coffset; 610 } 611 return 0; 612 } 613 614 static void qcow_refresh_limits(BlockDriverState *bs, Error **errp) 615 { 616 /* At least encrypted images require 512-byte alignment. Apply the 617 * limit universally, rather than just on encrypted images, as 618 * it's easier to let the block layer handle rounding than to 619 * audit this code further. */ 620 bs->bl.request_alignment = BDRV_SECTOR_SIZE; 621 } 622 623 static int coroutine_fn GRAPH_RDLOCK 624 qcow_co_preadv(BlockDriverState *bs, int64_t offset, int64_t bytes, 625 QEMUIOVector *qiov, BdrvRequestFlags flags) 626 { 627 BDRVQcowState *s = bs->opaque; 628 int offset_in_cluster; 629 int ret = 0, n; 630 uint64_t cluster_offset; 631 uint8_t *buf; 632 void *orig_buf; 633 634 if (qiov->niov > 1) { 635 buf = orig_buf = qemu_try_blockalign(bs, qiov->size); 636 if (buf == NULL) { 637 return -ENOMEM; 638 } 639 } else { 640 orig_buf = NULL; 641 buf = (uint8_t *)qiov->iov->iov_base; 642 } 643 644 qemu_co_mutex_lock(&s->lock); 645 646 while (bytes != 0) { 647 /* prepare next request */ 648 ret = get_cluster_offset(bs, offset, 0, 0, 0, 0, &cluster_offset); 649 if (ret < 0) { 650 break; 651 } 652 offset_in_cluster = offset & (s->cluster_size - 1); 653 n = s->cluster_size - offset_in_cluster; 654 if (n > bytes) { 655 n = bytes; 656 } 657 658 if (!cluster_offset) { 659 if (bs->backing) { 660 /* read from the base image */ 661 qemu_co_mutex_unlock(&s->lock); 662 /* qcow2 emits this on bs->file instead of bs->backing */ 663 BLKDBG_CO_EVENT(bs->file, BLKDBG_READ_BACKING_AIO); 664 ret = bdrv_co_pread(bs->backing, offset, n, buf, 0); 665 qemu_co_mutex_lock(&s->lock); 666 if (ret < 0) { 667 break; 668 } 669 } else { 670 /* Note: in this case, no need to wait */ 671 memset(buf, 0, n); 672 } 673 } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) { 674 /* add AIO support for compressed blocks ? */ 675 if (decompress_cluster(bs, cluster_offset) < 0) { 676 ret = -EIO; 677 break; 678 } 679 memcpy(buf, s->cluster_cache + offset_in_cluster, n); 680 } else { 681 if ((cluster_offset & 511) != 0) { 682 ret = -EIO; 683 break; 684 } 685 qemu_co_mutex_unlock(&s->lock); 686 BLKDBG_CO_EVENT(bs->file, BLKDBG_READ_AIO); 687 ret = bdrv_co_pread(bs->file, cluster_offset + offset_in_cluster, 688 n, buf, 0); 689 qemu_co_mutex_lock(&s->lock); 690 if (ret < 0) { 691 break; 692 } 693 if (bs->encrypted) { 694 assert(s->crypto); 695 if (qcrypto_block_decrypt(s->crypto, 696 offset, buf, n, NULL) < 0) { 697 ret = -EIO; 698 break; 699 } 700 } 701 } 702 ret = 0; 703 704 bytes -= n; 705 offset += n; 706 buf += n; 707 } 708 709 qemu_co_mutex_unlock(&s->lock); 710 711 if (qiov->niov > 1) { 712 qemu_iovec_from_buf(qiov, 0, orig_buf, qiov->size); 713 qemu_vfree(orig_buf); 714 } 715 716 return ret; 717 } 718 719 static int coroutine_fn GRAPH_RDLOCK 720 qcow_co_pwritev(BlockDriverState *bs, int64_t offset, int64_t bytes, 721 QEMUIOVector *qiov, BdrvRequestFlags flags) 722 { 723 BDRVQcowState *s = bs->opaque; 724 int offset_in_cluster; 725 uint64_t cluster_offset; 726 int ret = 0, n; 727 uint8_t *buf; 728 void *orig_buf; 729 730 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_CO_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 GRAPH_UNLOCKED 807 qcow_co_create(BlockdevCreateOptions *opts, 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->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_co_open_blockdev_ref(qcow_opts->file, errp); 838 if (bs == NULL) { 839 return -EIO; 840 } 841 842 qcow_blk = blk_co_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->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->backing_file = NULL; 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->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_co_pwrite(qcow_blk, 0, sizeof(header), &header, 0); 895 if (ret < 0) { 896 goto exit; 897 } 898 899 if (qcow_opts->backing_file) { 900 ret = blk_co_pwrite(qcow_blk, sizeof(header), backing_filename_len, 901 qcow_opts->backing_file, 0); 902 if (ret < 0) { 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_co_pwrite(qcow_blk, header_size + BDRV_SECTOR_SIZE * i, 911 BDRV_SECTOR_SIZE, tmp, 0); 912 if (ret < 0) { 913 g_free(tmp); 914 goto exit; 915 } 916 } 917 918 g_free(tmp); 919 ret = 0; 920 exit: 921 blk_co_unref(qcow_blk); 922 bdrv_co_unref(bs); 923 qcrypto_block_free(crypto); 924 return ret; 925 } 926 927 static int coroutine_fn GRAPH_UNLOCKED 928 qcow_co_create_opts(BlockDriver *drv, 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_co_create_file(filename, opts, errp); 978 if (ret < 0) { 979 goto fail; 980 } 981 982 bs = bdrv_co_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_co_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, l1_length, s->l1_table, 1034 0) < 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 int coroutine_fn GRAPH_RDLOCK 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_CO_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 coroutine_fn 1134 qcow_co_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) 1135 { 1136 BDRVQcowState *s = bs->opaque; 1137 bdi->cluster_size = s->cluster_size; 1138 return 0; 1139 } 1140 1141 static QemuOptsList qcow_create_opts = { 1142 .name = "qcow-create-opts", 1143 .head = QTAILQ_HEAD_INITIALIZER(qcow_create_opts.head), 1144 .desc = { 1145 { 1146 .name = BLOCK_OPT_SIZE, 1147 .type = QEMU_OPT_SIZE, 1148 .help = "Virtual disk size" 1149 }, 1150 { 1151 .name = BLOCK_OPT_BACKING_FILE, 1152 .type = QEMU_OPT_STRING, 1153 .help = "File name of a base image" 1154 }, 1155 { 1156 .name = BLOCK_OPT_BACKING_FMT, 1157 .type = QEMU_OPT_STRING, 1158 .help = "Format of the backing image", 1159 }, 1160 { 1161 .name = BLOCK_OPT_ENCRYPT, 1162 .type = QEMU_OPT_BOOL, 1163 .help = "Encrypt the image with format 'aes'. (Deprecated " 1164 "in favor of " BLOCK_OPT_ENCRYPT_FORMAT "=aes)", 1165 }, 1166 { 1167 .name = BLOCK_OPT_ENCRYPT_FORMAT, 1168 .type = QEMU_OPT_STRING, 1169 .help = "Encrypt the image, format choices: 'aes'", 1170 }, 1171 BLOCK_CRYPTO_OPT_DEF_QCOW_KEY_SECRET("encrypt."), 1172 { /* end of list */ } 1173 } 1174 }; 1175 1176 static const char *const qcow_strong_runtime_opts[] = { 1177 "encrypt." BLOCK_CRYPTO_OPT_QCOW_KEY_SECRET, 1178 1179 NULL 1180 }; 1181 1182 static BlockDriver bdrv_qcow = { 1183 .format_name = "qcow", 1184 .instance_size = sizeof(BDRVQcowState), 1185 .bdrv_probe = qcow_probe, 1186 .bdrv_open = qcow_open, 1187 .bdrv_close = qcow_close, 1188 .bdrv_child_perm = bdrv_default_perms, 1189 .bdrv_reopen_prepare = qcow_reopen_prepare, 1190 .bdrv_co_create = qcow_co_create, 1191 .bdrv_co_create_opts = qcow_co_create_opts, 1192 .bdrv_has_zero_init = bdrv_has_zero_init_1, 1193 .is_format = true, 1194 .supports_backing = true, 1195 .bdrv_refresh_limits = qcow_refresh_limits, 1196 1197 .bdrv_co_preadv = qcow_co_preadv, 1198 .bdrv_co_pwritev = qcow_co_pwritev, 1199 .bdrv_co_block_status = qcow_co_block_status, 1200 1201 .bdrv_make_empty = qcow_make_empty, 1202 .bdrv_co_pwritev_compressed = qcow_co_pwritev_compressed, 1203 .bdrv_co_get_info = qcow_co_get_info, 1204 1205 .create_opts = &qcow_create_opts, 1206 .strong_runtime_opts = qcow_strong_runtime_opts, 1207 }; 1208 1209 static void bdrv_qcow_init(void) 1210 { 1211 bdrv_register(&bdrv_qcow); 1212 } 1213 1214 block_init(bdrv_qcow_init); 1215