1 /* 2 * Block driver for the VMDK format 3 * 4 * Copyright (c) 2004 Fabrice Bellard 5 * Copyright (c) 2005 Filip Navara 6 * 7 * Permission is hereby granted, free of charge, to any person obtaining a copy 8 * of this software and associated documentation files (the "Software"), to deal 9 * in the Software without restriction, including without limitation the rights 10 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 11 * copies of the Software, and to permit persons to whom the Software is 12 * furnished to do so, subject to the following conditions: 13 * 14 * The above copyright notice and this permission notice shall be included in 15 * all copies or substantial portions of the Software. 16 * 17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 18 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 20 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 21 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 22 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN 23 * THE SOFTWARE. 24 */ 25 26 #include "qemu-common.h" 27 #include "block/block_int.h" 28 #include "qemu/module.h" 29 #include "migration/migration.h" 30 #include <zlib.h> 31 32 #define VMDK3_MAGIC (('C' << 24) | ('O' << 16) | ('W' << 8) | 'D') 33 #define VMDK4_MAGIC (('K' << 24) | ('D' << 16) | ('M' << 8) | 'V') 34 #define VMDK4_COMPRESSION_DEFLATE 1 35 #define VMDK4_FLAG_NL_DETECT (1 << 0) 36 #define VMDK4_FLAG_RGD (1 << 1) 37 /* Zeroed-grain enable bit */ 38 #define VMDK4_FLAG_ZERO_GRAIN (1 << 2) 39 #define VMDK4_FLAG_COMPRESS (1 << 16) 40 #define VMDK4_FLAG_MARKER (1 << 17) 41 #define VMDK4_GD_AT_END 0xffffffffffffffffULL 42 43 #define VMDK_GTE_ZEROED 0x1 44 45 /* VMDK internal error codes */ 46 #define VMDK_OK 0 47 #define VMDK_ERROR (-1) 48 /* Cluster not allocated */ 49 #define VMDK_UNALLOC (-2) 50 #define VMDK_ZEROED (-3) 51 52 #define BLOCK_OPT_ZEROED_GRAIN "zeroed_grain" 53 54 typedef struct { 55 uint32_t version; 56 uint32_t flags; 57 uint32_t disk_sectors; 58 uint32_t granularity; 59 uint32_t l1dir_offset; 60 uint32_t l1dir_size; 61 uint32_t file_sectors; 62 uint32_t cylinders; 63 uint32_t heads; 64 uint32_t sectors_per_track; 65 } QEMU_PACKED VMDK3Header; 66 67 typedef struct { 68 uint32_t version; 69 uint32_t flags; 70 uint64_t capacity; 71 uint64_t granularity; 72 uint64_t desc_offset; 73 uint64_t desc_size; 74 /* Number of GrainTableEntries per GrainTable */ 75 uint32_t num_gtes_per_gt; 76 uint64_t rgd_offset; 77 uint64_t gd_offset; 78 uint64_t grain_offset; 79 char filler[1]; 80 char check_bytes[4]; 81 uint16_t compressAlgorithm; 82 } QEMU_PACKED VMDK4Header; 83 84 #define L2_CACHE_SIZE 16 85 86 typedef struct VmdkExtent { 87 BlockDriverState *file; 88 bool flat; 89 bool compressed; 90 bool has_marker; 91 bool has_zero_grain; 92 int version; 93 int64_t sectors; 94 int64_t end_sector; 95 int64_t flat_start_offset; 96 int64_t l1_table_offset; 97 int64_t l1_backup_table_offset; 98 uint32_t *l1_table; 99 uint32_t *l1_backup_table; 100 unsigned int l1_size; 101 uint32_t l1_entry_sectors; 102 103 unsigned int l2_size; 104 uint32_t *l2_cache; 105 uint32_t l2_cache_offsets[L2_CACHE_SIZE]; 106 uint32_t l2_cache_counts[L2_CACHE_SIZE]; 107 108 unsigned int cluster_sectors; 109 } VmdkExtent; 110 111 typedef struct BDRVVmdkState { 112 CoMutex lock; 113 uint64_t desc_offset; 114 bool cid_updated; 115 uint32_t parent_cid; 116 int num_extents; 117 /* Extent array with num_extents entries, ascend ordered by address */ 118 VmdkExtent *extents; 119 Error *migration_blocker; 120 } BDRVVmdkState; 121 122 typedef struct VmdkMetaData { 123 uint32_t offset; 124 unsigned int l1_index; 125 unsigned int l2_index; 126 unsigned int l2_offset; 127 int valid; 128 uint32_t *l2_cache_entry; 129 } VmdkMetaData; 130 131 typedef struct VmdkGrainMarker { 132 uint64_t lba; 133 uint32_t size; 134 uint8_t data[0]; 135 } QEMU_PACKED VmdkGrainMarker; 136 137 enum { 138 MARKER_END_OF_STREAM = 0, 139 MARKER_GRAIN_TABLE = 1, 140 MARKER_GRAIN_DIRECTORY = 2, 141 MARKER_FOOTER = 3, 142 }; 143 144 static int vmdk_probe(const uint8_t *buf, int buf_size, const char *filename) 145 { 146 uint32_t magic; 147 148 if (buf_size < 4) { 149 return 0; 150 } 151 magic = be32_to_cpu(*(uint32_t *)buf); 152 if (magic == VMDK3_MAGIC || 153 magic == VMDK4_MAGIC) { 154 return 100; 155 } else { 156 const char *p = (const char *)buf; 157 const char *end = p + buf_size; 158 while (p < end) { 159 if (*p == '#') { 160 /* skip comment line */ 161 while (p < end && *p != '\n') { 162 p++; 163 } 164 p++; 165 continue; 166 } 167 if (*p == ' ') { 168 while (p < end && *p == ' ') { 169 p++; 170 } 171 /* skip '\r' if windows line endings used. */ 172 if (p < end && *p == '\r') { 173 p++; 174 } 175 /* only accept blank lines before 'version=' line */ 176 if (p == end || *p != '\n') { 177 return 0; 178 } 179 p++; 180 continue; 181 } 182 if (end - p >= strlen("version=X\n")) { 183 if (strncmp("version=1\n", p, strlen("version=1\n")) == 0 || 184 strncmp("version=2\n", p, strlen("version=2\n")) == 0) { 185 return 100; 186 } 187 } 188 if (end - p >= strlen("version=X\r\n")) { 189 if (strncmp("version=1\r\n", p, strlen("version=1\r\n")) == 0 || 190 strncmp("version=2\r\n", p, strlen("version=2\r\n")) == 0) { 191 return 100; 192 } 193 } 194 return 0; 195 } 196 return 0; 197 } 198 } 199 200 #define CHECK_CID 1 201 202 #define SECTOR_SIZE 512 203 #define DESC_SIZE (20 * SECTOR_SIZE) /* 20 sectors of 512 bytes each */ 204 #define BUF_SIZE 4096 205 #define HEADER_SIZE 512 /* first sector of 512 bytes */ 206 207 static void vmdk_free_extents(BlockDriverState *bs) 208 { 209 int i; 210 BDRVVmdkState *s = bs->opaque; 211 VmdkExtent *e; 212 213 for (i = 0; i < s->num_extents; i++) { 214 e = &s->extents[i]; 215 g_free(e->l1_table); 216 g_free(e->l2_cache); 217 g_free(e->l1_backup_table); 218 if (e->file != bs->file) { 219 bdrv_unref(e->file); 220 } 221 } 222 g_free(s->extents); 223 } 224 225 static void vmdk_free_last_extent(BlockDriverState *bs) 226 { 227 BDRVVmdkState *s = bs->opaque; 228 229 if (s->num_extents == 0) { 230 return; 231 } 232 s->num_extents--; 233 s->extents = g_realloc(s->extents, s->num_extents * sizeof(VmdkExtent)); 234 } 235 236 static uint32_t vmdk_read_cid(BlockDriverState *bs, int parent) 237 { 238 char desc[DESC_SIZE]; 239 uint32_t cid = 0xffffffff; 240 const char *p_name, *cid_str; 241 size_t cid_str_size; 242 BDRVVmdkState *s = bs->opaque; 243 int ret; 244 245 ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE); 246 if (ret < 0) { 247 return 0; 248 } 249 250 if (parent) { 251 cid_str = "parentCID"; 252 cid_str_size = sizeof("parentCID"); 253 } else { 254 cid_str = "CID"; 255 cid_str_size = sizeof("CID"); 256 } 257 258 desc[DESC_SIZE - 1] = '\0'; 259 p_name = strstr(desc, cid_str); 260 if (p_name != NULL) { 261 p_name += cid_str_size; 262 sscanf(p_name, "%x", &cid); 263 } 264 265 return cid; 266 } 267 268 static int vmdk_write_cid(BlockDriverState *bs, uint32_t cid) 269 { 270 char desc[DESC_SIZE], tmp_desc[DESC_SIZE]; 271 char *p_name, *tmp_str; 272 BDRVVmdkState *s = bs->opaque; 273 int ret; 274 275 ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE); 276 if (ret < 0) { 277 return ret; 278 } 279 280 desc[DESC_SIZE - 1] = '\0'; 281 tmp_str = strstr(desc, "parentCID"); 282 if (tmp_str == NULL) { 283 return -EINVAL; 284 } 285 286 pstrcpy(tmp_desc, sizeof(tmp_desc), tmp_str); 287 p_name = strstr(desc, "CID"); 288 if (p_name != NULL) { 289 p_name += sizeof("CID"); 290 snprintf(p_name, sizeof(desc) - (p_name - desc), "%x\n", cid); 291 pstrcat(desc, sizeof(desc), tmp_desc); 292 } 293 294 ret = bdrv_pwrite_sync(bs->file, s->desc_offset, desc, DESC_SIZE); 295 if (ret < 0) { 296 return ret; 297 } 298 299 return 0; 300 } 301 302 static int vmdk_is_cid_valid(BlockDriverState *bs) 303 { 304 #ifdef CHECK_CID 305 BDRVVmdkState *s = bs->opaque; 306 BlockDriverState *p_bs = bs->backing_hd; 307 uint32_t cur_pcid; 308 309 if (p_bs) { 310 cur_pcid = vmdk_read_cid(p_bs, 0); 311 if (s->parent_cid != cur_pcid) { 312 /* CID not valid */ 313 return 0; 314 } 315 } 316 #endif 317 /* CID valid */ 318 return 1; 319 } 320 321 /* Queue extents, if any, for reopen() */ 322 static int vmdk_reopen_prepare(BDRVReopenState *state, 323 BlockReopenQueue *queue, Error **errp) 324 { 325 BDRVVmdkState *s; 326 int ret = -1; 327 int i; 328 VmdkExtent *e; 329 330 assert(state != NULL); 331 assert(state->bs != NULL); 332 333 if (queue == NULL) { 334 error_set(errp, ERROR_CLASS_GENERIC_ERROR, 335 "No reopen queue for VMDK extents"); 336 goto exit; 337 } 338 339 s = state->bs->opaque; 340 341 assert(s != NULL); 342 343 for (i = 0; i < s->num_extents; i++) { 344 e = &s->extents[i]; 345 if (e->file != state->bs->file) { 346 bdrv_reopen_queue(queue, e->file, state->flags); 347 } 348 } 349 ret = 0; 350 351 exit: 352 return ret; 353 } 354 355 static int vmdk_parent_open(BlockDriverState *bs) 356 { 357 char *p_name; 358 char desc[DESC_SIZE + 1]; 359 BDRVVmdkState *s = bs->opaque; 360 int ret; 361 362 desc[DESC_SIZE] = '\0'; 363 ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE); 364 if (ret < 0) { 365 return ret; 366 } 367 368 p_name = strstr(desc, "parentFileNameHint"); 369 if (p_name != NULL) { 370 char *end_name; 371 372 p_name += sizeof("parentFileNameHint") + 1; 373 end_name = strchr(p_name, '\"'); 374 if (end_name == NULL) { 375 return -EINVAL; 376 } 377 if ((end_name - p_name) > sizeof(bs->backing_file) - 1) { 378 return -EINVAL; 379 } 380 381 pstrcpy(bs->backing_file, end_name - p_name + 1, p_name); 382 } 383 384 return 0; 385 } 386 387 /* Create and append extent to the extent array. Return the added VmdkExtent 388 * address. return NULL if allocation failed. */ 389 static int vmdk_add_extent(BlockDriverState *bs, 390 BlockDriverState *file, bool flat, int64_t sectors, 391 int64_t l1_offset, int64_t l1_backup_offset, 392 uint32_t l1_size, 393 int l2_size, uint64_t cluster_sectors, 394 VmdkExtent **new_extent) 395 { 396 VmdkExtent *extent; 397 BDRVVmdkState *s = bs->opaque; 398 399 if (cluster_sectors > 0x200000) { 400 /* 0x200000 * 512Bytes = 1GB for one cluster is unrealistic */ 401 error_report("invalid granularity, image may be corrupt"); 402 return -EINVAL; 403 } 404 if (l1_size > 512 * 1024 * 1024) { 405 /* Although with big capacity and small l1_entry_sectors, we can get a 406 * big l1_size, we don't want unbounded value to allocate the table. 407 * Limit it to 512M, which is 16PB for default cluster and L2 table 408 * size */ 409 error_report("L1 size too big"); 410 return -EFBIG; 411 } 412 413 s->extents = g_realloc(s->extents, 414 (s->num_extents + 1) * sizeof(VmdkExtent)); 415 extent = &s->extents[s->num_extents]; 416 s->num_extents++; 417 418 memset(extent, 0, sizeof(VmdkExtent)); 419 extent->file = file; 420 extent->flat = flat; 421 extent->sectors = sectors; 422 extent->l1_table_offset = l1_offset; 423 extent->l1_backup_table_offset = l1_backup_offset; 424 extent->l1_size = l1_size; 425 extent->l1_entry_sectors = l2_size * cluster_sectors; 426 extent->l2_size = l2_size; 427 extent->cluster_sectors = cluster_sectors; 428 429 if (s->num_extents > 1) { 430 extent->end_sector = (*(extent - 1)).end_sector + extent->sectors; 431 } else { 432 extent->end_sector = extent->sectors; 433 } 434 bs->total_sectors = extent->end_sector; 435 if (new_extent) { 436 *new_extent = extent; 437 } 438 return 0; 439 } 440 441 static int vmdk_init_tables(BlockDriverState *bs, VmdkExtent *extent) 442 { 443 int ret; 444 int l1_size, i; 445 446 /* read the L1 table */ 447 l1_size = extent->l1_size * sizeof(uint32_t); 448 extent->l1_table = g_malloc(l1_size); 449 ret = bdrv_pread(extent->file, 450 extent->l1_table_offset, 451 extent->l1_table, 452 l1_size); 453 if (ret < 0) { 454 goto fail_l1; 455 } 456 for (i = 0; i < extent->l1_size; i++) { 457 le32_to_cpus(&extent->l1_table[i]); 458 } 459 460 if (extent->l1_backup_table_offset) { 461 extent->l1_backup_table = g_malloc(l1_size); 462 ret = bdrv_pread(extent->file, 463 extent->l1_backup_table_offset, 464 extent->l1_backup_table, 465 l1_size); 466 if (ret < 0) { 467 goto fail_l1b; 468 } 469 for (i = 0; i < extent->l1_size; i++) { 470 le32_to_cpus(&extent->l1_backup_table[i]); 471 } 472 } 473 474 extent->l2_cache = 475 g_malloc(extent->l2_size * L2_CACHE_SIZE * sizeof(uint32_t)); 476 return 0; 477 fail_l1b: 478 g_free(extent->l1_backup_table); 479 fail_l1: 480 g_free(extent->l1_table); 481 return ret; 482 } 483 484 static int vmdk_open_vmfs_sparse(BlockDriverState *bs, 485 BlockDriverState *file, 486 int flags) 487 { 488 int ret; 489 uint32_t magic; 490 VMDK3Header header; 491 VmdkExtent *extent; 492 493 ret = bdrv_pread(file, sizeof(magic), &header, sizeof(header)); 494 if (ret < 0) { 495 return ret; 496 } 497 ret = vmdk_add_extent(bs, file, false, 498 le32_to_cpu(header.disk_sectors), 499 le32_to_cpu(header.l1dir_offset) << 9, 500 0, 501 le32_to_cpu(header.l1dir_size), 502 4096, 503 le32_to_cpu(header.granularity), 504 &extent); 505 if (ret < 0) { 506 return ret; 507 } 508 ret = vmdk_init_tables(bs, extent); 509 if (ret) { 510 /* free extent allocated by vmdk_add_extent */ 511 vmdk_free_last_extent(bs); 512 } 513 return ret; 514 } 515 516 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, 517 uint64_t desc_offset); 518 519 static int vmdk_open_vmdk4(BlockDriverState *bs, 520 BlockDriverState *file, 521 int flags) 522 { 523 int ret; 524 uint32_t magic; 525 uint32_t l1_size, l1_entry_sectors; 526 VMDK4Header header; 527 VmdkExtent *extent; 528 int64_t l1_backup_offset = 0; 529 530 ret = bdrv_pread(file, sizeof(magic), &header, sizeof(header)); 531 if (ret < 0) { 532 return ret; 533 } 534 if (header.capacity == 0) { 535 uint64_t desc_offset = le64_to_cpu(header.desc_offset); 536 if (desc_offset) { 537 return vmdk_open_desc_file(bs, flags, desc_offset << 9); 538 } 539 } 540 541 if (le64_to_cpu(header.gd_offset) == VMDK4_GD_AT_END) { 542 /* 543 * The footer takes precedence over the header, so read it in. The 544 * footer starts at offset -1024 from the end: One sector for the 545 * footer, and another one for the end-of-stream marker. 546 */ 547 struct { 548 struct { 549 uint64_t val; 550 uint32_t size; 551 uint32_t type; 552 uint8_t pad[512 - 16]; 553 } QEMU_PACKED footer_marker; 554 555 uint32_t magic; 556 VMDK4Header header; 557 uint8_t pad[512 - 4 - sizeof(VMDK4Header)]; 558 559 struct { 560 uint64_t val; 561 uint32_t size; 562 uint32_t type; 563 uint8_t pad[512 - 16]; 564 } QEMU_PACKED eos_marker; 565 } QEMU_PACKED footer; 566 567 ret = bdrv_pread(file, 568 bs->file->total_sectors * 512 - 1536, 569 &footer, sizeof(footer)); 570 if (ret < 0) { 571 return ret; 572 } 573 574 /* Some sanity checks for the footer */ 575 if (be32_to_cpu(footer.magic) != VMDK4_MAGIC || 576 le32_to_cpu(footer.footer_marker.size) != 0 || 577 le32_to_cpu(footer.footer_marker.type) != MARKER_FOOTER || 578 le64_to_cpu(footer.eos_marker.val) != 0 || 579 le32_to_cpu(footer.eos_marker.size) != 0 || 580 le32_to_cpu(footer.eos_marker.type) != MARKER_END_OF_STREAM) 581 { 582 return -EINVAL; 583 } 584 585 header = footer.header; 586 } 587 588 if (le32_to_cpu(header.version) >= 3) { 589 char buf[64]; 590 snprintf(buf, sizeof(buf), "VMDK version %d", 591 le32_to_cpu(header.version)); 592 qerror_report(QERR_UNKNOWN_BLOCK_FORMAT_FEATURE, 593 bs->device_name, "vmdk", buf); 594 return -ENOTSUP; 595 } 596 597 if (le32_to_cpu(header.num_gtes_per_gt) > 512) { 598 error_report("L2 table size too big"); 599 return -EINVAL; 600 } 601 602 l1_entry_sectors = le32_to_cpu(header.num_gtes_per_gt) 603 * le64_to_cpu(header.granularity); 604 if (l1_entry_sectors == 0) { 605 return -EINVAL; 606 } 607 l1_size = (le64_to_cpu(header.capacity) + l1_entry_sectors - 1) 608 / l1_entry_sectors; 609 if (le32_to_cpu(header.flags) & VMDK4_FLAG_RGD) { 610 l1_backup_offset = le64_to_cpu(header.rgd_offset) << 9; 611 } 612 ret = vmdk_add_extent(bs, file, false, 613 le64_to_cpu(header.capacity), 614 le64_to_cpu(header.gd_offset) << 9, 615 l1_backup_offset, 616 l1_size, 617 le32_to_cpu(header.num_gtes_per_gt), 618 le64_to_cpu(header.granularity), 619 &extent); 620 if (ret < 0) { 621 return ret; 622 } 623 extent->compressed = 624 le16_to_cpu(header.compressAlgorithm) == VMDK4_COMPRESSION_DEFLATE; 625 extent->has_marker = le32_to_cpu(header.flags) & VMDK4_FLAG_MARKER; 626 extent->version = le32_to_cpu(header.version); 627 extent->has_zero_grain = le32_to_cpu(header.flags) & VMDK4_FLAG_ZERO_GRAIN; 628 ret = vmdk_init_tables(bs, extent); 629 if (ret) { 630 /* free extent allocated by vmdk_add_extent */ 631 vmdk_free_last_extent(bs); 632 } 633 return ret; 634 } 635 636 /* find an option value out of descriptor file */ 637 static int vmdk_parse_description(const char *desc, const char *opt_name, 638 char *buf, int buf_size) 639 { 640 char *opt_pos, *opt_end; 641 const char *end = desc + strlen(desc); 642 643 opt_pos = strstr(desc, opt_name); 644 if (!opt_pos) { 645 return VMDK_ERROR; 646 } 647 /* Skip "=\"" following opt_name */ 648 opt_pos += strlen(opt_name) + 2; 649 if (opt_pos >= end) { 650 return VMDK_ERROR; 651 } 652 opt_end = opt_pos; 653 while (opt_end < end && *opt_end != '"') { 654 opt_end++; 655 } 656 if (opt_end == end || buf_size < opt_end - opt_pos + 1) { 657 return VMDK_ERROR; 658 } 659 pstrcpy(buf, opt_end - opt_pos + 1, opt_pos); 660 return VMDK_OK; 661 } 662 663 /* Open an extent file and append to bs array */ 664 static int vmdk_open_sparse(BlockDriverState *bs, 665 BlockDriverState *file, 666 int flags) 667 { 668 uint32_t magic; 669 670 if (bdrv_pread(file, 0, &magic, sizeof(magic)) != sizeof(magic)) { 671 return -EIO; 672 } 673 674 magic = be32_to_cpu(magic); 675 switch (magic) { 676 case VMDK3_MAGIC: 677 return vmdk_open_vmfs_sparse(bs, file, flags); 678 break; 679 case VMDK4_MAGIC: 680 return vmdk_open_vmdk4(bs, file, flags); 681 break; 682 default: 683 return -EMEDIUMTYPE; 684 break; 685 } 686 } 687 688 static int vmdk_parse_extents(const char *desc, BlockDriverState *bs, 689 const char *desc_file_path) 690 { 691 int ret; 692 char access[11]; 693 char type[11]; 694 char fname[512]; 695 const char *p = desc; 696 int64_t sectors = 0; 697 int64_t flat_offset; 698 char extent_path[PATH_MAX]; 699 BlockDriverState *extent_file; 700 701 while (*p) { 702 /* parse extent line: 703 * RW [size in sectors] FLAT "file-name.vmdk" OFFSET 704 * or 705 * RW [size in sectors] SPARSE "file-name.vmdk" 706 */ 707 flat_offset = -1; 708 ret = sscanf(p, "%10s %" SCNd64 " %10s \"%511[^\n\r\"]\" %" SCNd64, 709 access, §ors, type, fname, &flat_offset); 710 if (ret < 4 || strcmp(access, "RW")) { 711 goto next_line; 712 } else if (!strcmp(type, "FLAT")) { 713 if (ret != 5 || flat_offset < 0) { 714 return -EINVAL; 715 } 716 } else if (ret != 4) { 717 return -EINVAL; 718 } 719 720 if (sectors <= 0 || 721 (strcmp(type, "FLAT") && strcmp(type, "SPARSE") && 722 strcmp(type, "VMFS") && strcmp(type, "VMFSSPARSE")) || 723 (strcmp(access, "RW"))) { 724 goto next_line; 725 } 726 727 path_combine(extent_path, sizeof(extent_path), 728 desc_file_path, fname); 729 ret = bdrv_file_open(&extent_file, extent_path, NULL, bs->open_flags); 730 if (ret) { 731 return ret; 732 } 733 734 /* save to extents array */ 735 if (!strcmp(type, "FLAT") || !strcmp(type, "VMFS")) { 736 /* FLAT extent */ 737 VmdkExtent *extent; 738 739 ret = vmdk_add_extent(bs, extent_file, true, sectors, 740 0, 0, 0, 0, sectors, &extent); 741 if (ret < 0) { 742 return ret; 743 } 744 extent->flat_start_offset = flat_offset << 9; 745 } else if (!strcmp(type, "SPARSE") || !strcmp(type, "VMFSSPARSE")) { 746 /* SPARSE extent and VMFSSPARSE extent are both "COWD" sparse file*/ 747 ret = vmdk_open_sparse(bs, extent_file, bs->open_flags); 748 if (ret) { 749 bdrv_unref(extent_file); 750 return ret; 751 } 752 } else { 753 fprintf(stderr, 754 "VMDK: Not supported extent type \"%s\""".\n", type); 755 return -ENOTSUP; 756 } 757 next_line: 758 /* move to next line */ 759 while (*p && *p != '\n') { 760 p++; 761 } 762 p++; 763 } 764 return 0; 765 } 766 767 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, 768 uint64_t desc_offset) 769 { 770 int ret; 771 char *buf = NULL; 772 char ct[128]; 773 BDRVVmdkState *s = bs->opaque; 774 int64_t size; 775 776 size = bdrv_getlength(bs->file); 777 if (size < 0) { 778 return -EINVAL; 779 } 780 781 size = MIN(size, 1 << 20); /* avoid unbounded allocation */ 782 buf = g_malloc0(size + 1); 783 784 ret = bdrv_pread(bs->file, desc_offset, buf, size); 785 if (ret < 0) { 786 goto exit; 787 } 788 if (vmdk_parse_description(buf, "createType", ct, sizeof(ct))) { 789 ret = -EMEDIUMTYPE; 790 goto exit; 791 } 792 if (strcmp(ct, "monolithicFlat") && 793 strcmp(ct, "vmfs") && 794 strcmp(ct, "vmfsSparse") && 795 strcmp(ct, "twoGbMaxExtentSparse") && 796 strcmp(ct, "twoGbMaxExtentFlat")) { 797 fprintf(stderr, 798 "VMDK: Not supported image type \"%s\""".\n", ct); 799 ret = -ENOTSUP; 800 goto exit; 801 } 802 s->desc_offset = 0; 803 ret = vmdk_parse_extents(buf, bs, bs->file->filename); 804 exit: 805 g_free(buf); 806 return ret; 807 } 808 809 static int vmdk_open(BlockDriverState *bs, QDict *options, int flags) 810 { 811 int ret; 812 BDRVVmdkState *s = bs->opaque; 813 814 if (vmdk_open_sparse(bs, bs->file, flags) == 0) { 815 s->desc_offset = 0x200; 816 } else { 817 ret = vmdk_open_desc_file(bs, flags, 0); 818 if (ret) { 819 goto fail; 820 } 821 } 822 /* try to open parent images, if exist */ 823 ret = vmdk_parent_open(bs); 824 if (ret) { 825 goto fail; 826 } 827 s->parent_cid = vmdk_read_cid(bs, 1); 828 qemu_co_mutex_init(&s->lock); 829 830 /* Disable migration when VMDK images are used */ 831 error_set(&s->migration_blocker, 832 QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED, 833 "vmdk", bs->device_name, "live migration"); 834 migrate_add_blocker(s->migration_blocker); 835 836 return 0; 837 838 fail: 839 vmdk_free_extents(bs); 840 return ret; 841 } 842 843 static int get_whole_cluster(BlockDriverState *bs, 844 VmdkExtent *extent, 845 uint64_t cluster_offset, 846 uint64_t offset, 847 bool allocate) 848 { 849 int ret = VMDK_OK; 850 uint8_t *whole_grain = NULL; 851 852 /* we will be here if it's first write on non-exist grain(cluster). 853 * try to read from parent image, if exist */ 854 if (bs->backing_hd) { 855 whole_grain = 856 qemu_blockalign(bs, extent->cluster_sectors << BDRV_SECTOR_BITS); 857 if (!vmdk_is_cid_valid(bs)) { 858 ret = VMDK_ERROR; 859 goto exit; 860 } 861 862 /* floor offset to cluster */ 863 offset -= offset % (extent->cluster_sectors * 512); 864 ret = bdrv_read(bs->backing_hd, offset >> 9, whole_grain, 865 extent->cluster_sectors); 866 if (ret < 0) { 867 ret = VMDK_ERROR; 868 goto exit; 869 } 870 871 /* Write grain only into the active image */ 872 ret = bdrv_write(extent->file, cluster_offset, whole_grain, 873 extent->cluster_sectors); 874 if (ret < 0) { 875 ret = VMDK_ERROR; 876 goto exit; 877 } 878 } 879 exit: 880 qemu_vfree(whole_grain); 881 return ret; 882 } 883 884 static int vmdk_L2update(VmdkExtent *extent, VmdkMetaData *m_data) 885 { 886 uint32_t offset; 887 QEMU_BUILD_BUG_ON(sizeof(offset) != sizeof(m_data->offset)); 888 offset = cpu_to_le32(m_data->offset); 889 /* update L2 table */ 890 if (bdrv_pwrite_sync( 891 extent->file, 892 ((int64_t)m_data->l2_offset * 512) 893 + (m_data->l2_index * sizeof(m_data->offset)), 894 &offset, sizeof(offset)) < 0) { 895 return VMDK_ERROR; 896 } 897 /* update backup L2 table */ 898 if (extent->l1_backup_table_offset != 0) { 899 m_data->l2_offset = extent->l1_backup_table[m_data->l1_index]; 900 if (bdrv_pwrite_sync( 901 extent->file, 902 ((int64_t)m_data->l2_offset * 512) 903 + (m_data->l2_index * sizeof(m_data->offset)), 904 &offset, sizeof(offset)) < 0) { 905 return VMDK_ERROR; 906 } 907 } 908 if (m_data->l2_cache_entry) { 909 *m_data->l2_cache_entry = offset; 910 } 911 912 return VMDK_OK; 913 } 914 915 static int get_cluster_offset(BlockDriverState *bs, 916 VmdkExtent *extent, 917 VmdkMetaData *m_data, 918 uint64_t offset, 919 int allocate, 920 uint64_t *cluster_offset) 921 { 922 unsigned int l1_index, l2_offset, l2_index; 923 int min_index, i, j; 924 uint32_t min_count, *l2_table; 925 bool zeroed = false; 926 927 if (m_data) { 928 m_data->valid = 0; 929 } 930 if (extent->flat) { 931 *cluster_offset = extent->flat_start_offset; 932 return VMDK_OK; 933 } 934 935 offset -= (extent->end_sector - extent->sectors) * SECTOR_SIZE; 936 l1_index = (offset >> 9) / extent->l1_entry_sectors; 937 if (l1_index >= extent->l1_size) { 938 return VMDK_ERROR; 939 } 940 l2_offset = extent->l1_table[l1_index]; 941 if (!l2_offset) { 942 return VMDK_UNALLOC; 943 } 944 for (i = 0; i < L2_CACHE_SIZE; i++) { 945 if (l2_offset == extent->l2_cache_offsets[i]) { 946 /* increment the hit count */ 947 if (++extent->l2_cache_counts[i] == 0xffffffff) { 948 for (j = 0; j < L2_CACHE_SIZE; j++) { 949 extent->l2_cache_counts[j] >>= 1; 950 } 951 } 952 l2_table = extent->l2_cache + (i * extent->l2_size); 953 goto found; 954 } 955 } 956 /* not found: load a new entry in the least used one */ 957 min_index = 0; 958 min_count = 0xffffffff; 959 for (i = 0; i < L2_CACHE_SIZE; i++) { 960 if (extent->l2_cache_counts[i] < min_count) { 961 min_count = extent->l2_cache_counts[i]; 962 min_index = i; 963 } 964 } 965 l2_table = extent->l2_cache + (min_index * extent->l2_size); 966 if (bdrv_pread( 967 extent->file, 968 (int64_t)l2_offset * 512, 969 l2_table, 970 extent->l2_size * sizeof(uint32_t) 971 ) != extent->l2_size * sizeof(uint32_t)) { 972 return VMDK_ERROR; 973 } 974 975 extent->l2_cache_offsets[min_index] = l2_offset; 976 extent->l2_cache_counts[min_index] = 1; 977 found: 978 l2_index = ((offset >> 9) / extent->cluster_sectors) % extent->l2_size; 979 *cluster_offset = le32_to_cpu(l2_table[l2_index]); 980 981 if (m_data) { 982 m_data->valid = 1; 983 m_data->l1_index = l1_index; 984 m_data->l2_index = l2_index; 985 m_data->offset = *cluster_offset; 986 m_data->l2_offset = l2_offset; 987 m_data->l2_cache_entry = &l2_table[l2_index]; 988 } 989 if (extent->has_zero_grain && *cluster_offset == VMDK_GTE_ZEROED) { 990 zeroed = true; 991 } 992 993 if (!*cluster_offset || zeroed) { 994 if (!allocate) { 995 return zeroed ? VMDK_ZEROED : VMDK_UNALLOC; 996 } 997 998 /* Avoid the L2 tables update for the images that have snapshots. */ 999 *cluster_offset = bdrv_getlength(extent->file); 1000 if (!extent->compressed) { 1001 bdrv_truncate( 1002 extent->file, 1003 *cluster_offset + (extent->cluster_sectors << 9) 1004 ); 1005 } 1006 1007 *cluster_offset >>= 9; 1008 l2_table[l2_index] = cpu_to_le32(*cluster_offset); 1009 1010 /* First of all we write grain itself, to avoid race condition 1011 * that may to corrupt the image. 1012 * This problem may occur because of insufficient space on host disk 1013 * or inappropriate VM shutdown. 1014 */ 1015 if (get_whole_cluster( 1016 bs, extent, *cluster_offset, offset, allocate) == -1) { 1017 return VMDK_ERROR; 1018 } 1019 1020 if (m_data) { 1021 m_data->offset = *cluster_offset; 1022 } 1023 } 1024 *cluster_offset <<= 9; 1025 return VMDK_OK; 1026 } 1027 1028 static VmdkExtent *find_extent(BDRVVmdkState *s, 1029 int64_t sector_num, VmdkExtent *start_hint) 1030 { 1031 VmdkExtent *extent = start_hint; 1032 1033 if (!extent) { 1034 extent = &s->extents[0]; 1035 } 1036 while (extent < &s->extents[s->num_extents]) { 1037 if (sector_num < extent->end_sector) { 1038 return extent; 1039 } 1040 extent++; 1041 } 1042 return NULL; 1043 } 1044 1045 static int64_t coroutine_fn vmdk_co_get_block_status(BlockDriverState *bs, 1046 int64_t sector_num, int nb_sectors, int *pnum) 1047 { 1048 BDRVVmdkState *s = bs->opaque; 1049 int64_t index_in_cluster, n, ret; 1050 uint64_t offset; 1051 VmdkExtent *extent; 1052 1053 extent = find_extent(s, sector_num, NULL); 1054 if (!extent) { 1055 return 0; 1056 } 1057 qemu_co_mutex_lock(&s->lock); 1058 ret = get_cluster_offset(bs, extent, NULL, 1059 sector_num * 512, 0, &offset); 1060 qemu_co_mutex_unlock(&s->lock); 1061 1062 switch (ret) { 1063 case VMDK_ERROR: 1064 ret = -EIO; 1065 break; 1066 case VMDK_UNALLOC: 1067 ret = 0; 1068 break; 1069 case VMDK_ZEROED: 1070 ret = BDRV_BLOCK_ZERO; 1071 break; 1072 case VMDK_OK: 1073 ret = BDRV_BLOCK_DATA; 1074 if (extent->file == bs->file) { 1075 ret |= BDRV_BLOCK_OFFSET_VALID | offset; 1076 } 1077 1078 break; 1079 } 1080 1081 index_in_cluster = sector_num % extent->cluster_sectors; 1082 n = extent->cluster_sectors - index_in_cluster; 1083 if (n > nb_sectors) { 1084 n = nb_sectors; 1085 } 1086 *pnum = n; 1087 return ret; 1088 } 1089 1090 static int vmdk_write_extent(VmdkExtent *extent, int64_t cluster_offset, 1091 int64_t offset_in_cluster, const uint8_t *buf, 1092 int nb_sectors, int64_t sector_num) 1093 { 1094 int ret; 1095 VmdkGrainMarker *data = NULL; 1096 uLongf buf_len; 1097 const uint8_t *write_buf = buf; 1098 int write_len = nb_sectors * 512; 1099 1100 if (extent->compressed) { 1101 if (!extent->has_marker) { 1102 ret = -EINVAL; 1103 goto out; 1104 } 1105 buf_len = (extent->cluster_sectors << 9) * 2; 1106 data = g_malloc(buf_len + sizeof(VmdkGrainMarker)); 1107 if (compress(data->data, &buf_len, buf, nb_sectors << 9) != Z_OK || 1108 buf_len == 0) { 1109 ret = -EINVAL; 1110 goto out; 1111 } 1112 data->lba = sector_num; 1113 data->size = buf_len; 1114 write_buf = (uint8_t *)data; 1115 write_len = buf_len + sizeof(VmdkGrainMarker); 1116 } 1117 ret = bdrv_pwrite(extent->file, 1118 cluster_offset + offset_in_cluster, 1119 write_buf, 1120 write_len); 1121 if (ret != write_len) { 1122 ret = ret < 0 ? ret : -EIO; 1123 goto out; 1124 } 1125 ret = 0; 1126 out: 1127 g_free(data); 1128 return ret; 1129 } 1130 1131 static int vmdk_read_extent(VmdkExtent *extent, int64_t cluster_offset, 1132 int64_t offset_in_cluster, uint8_t *buf, 1133 int nb_sectors) 1134 { 1135 int ret; 1136 int cluster_bytes, buf_bytes; 1137 uint8_t *cluster_buf, *compressed_data; 1138 uint8_t *uncomp_buf; 1139 uint32_t data_len; 1140 VmdkGrainMarker *marker; 1141 uLongf buf_len; 1142 1143 1144 if (!extent->compressed) { 1145 ret = bdrv_pread(extent->file, 1146 cluster_offset + offset_in_cluster, 1147 buf, nb_sectors * 512); 1148 if (ret == nb_sectors * 512) { 1149 return 0; 1150 } else { 1151 return -EIO; 1152 } 1153 } 1154 cluster_bytes = extent->cluster_sectors * 512; 1155 /* Read two clusters in case GrainMarker + compressed data > one cluster */ 1156 buf_bytes = cluster_bytes * 2; 1157 cluster_buf = g_malloc(buf_bytes); 1158 uncomp_buf = g_malloc(cluster_bytes); 1159 ret = bdrv_pread(extent->file, 1160 cluster_offset, 1161 cluster_buf, buf_bytes); 1162 if (ret < 0) { 1163 goto out; 1164 } 1165 compressed_data = cluster_buf; 1166 buf_len = cluster_bytes; 1167 data_len = cluster_bytes; 1168 if (extent->has_marker) { 1169 marker = (VmdkGrainMarker *)cluster_buf; 1170 compressed_data = marker->data; 1171 data_len = le32_to_cpu(marker->size); 1172 } 1173 if (!data_len || data_len > buf_bytes) { 1174 ret = -EINVAL; 1175 goto out; 1176 } 1177 ret = uncompress(uncomp_buf, &buf_len, compressed_data, data_len); 1178 if (ret != Z_OK) { 1179 ret = -EINVAL; 1180 goto out; 1181 1182 } 1183 if (offset_in_cluster < 0 || 1184 offset_in_cluster + nb_sectors * 512 > buf_len) { 1185 ret = -EINVAL; 1186 goto out; 1187 } 1188 memcpy(buf, uncomp_buf + offset_in_cluster, nb_sectors * 512); 1189 ret = 0; 1190 1191 out: 1192 g_free(uncomp_buf); 1193 g_free(cluster_buf); 1194 return ret; 1195 } 1196 1197 static int vmdk_read(BlockDriverState *bs, int64_t sector_num, 1198 uint8_t *buf, int nb_sectors) 1199 { 1200 BDRVVmdkState *s = bs->opaque; 1201 int ret; 1202 uint64_t n, index_in_cluster; 1203 uint64_t extent_begin_sector, extent_relative_sector_num; 1204 VmdkExtent *extent = NULL; 1205 uint64_t cluster_offset; 1206 1207 while (nb_sectors > 0) { 1208 extent = find_extent(s, sector_num, extent); 1209 if (!extent) { 1210 return -EIO; 1211 } 1212 ret = get_cluster_offset( 1213 bs, extent, NULL, 1214 sector_num << 9, 0, &cluster_offset); 1215 extent_begin_sector = extent->end_sector - extent->sectors; 1216 extent_relative_sector_num = sector_num - extent_begin_sector; 1217 index_in_cluster = extent_relative_sector_num % extent->cluster_sectors; 1218 n = extent->cluster_sectors - index_in_cluster; 1219 if (n > nb_sectors) { 1220 n = nb_sectors; 1221 } 1222 if (ret != VMDK_OK) { 1223 /* if not allocated, try to read from parent image, if exist */ 1224 if (bs->backing_hd && ret != VMDK_ZEROED) { 1225 if (!vmdk_is_cid_valid(bs)) { 1226 return -EINVAL; 1227 } 1228 ret = bdrv_read(bs->backing_hd, sector_num, buf, n); 1229 if (ret < 0) { 1230 return ret; 1231 } 1232 } else { 1233 memset(buf, 0, 512 * n); 1234 } 1235 } else { 1236 ret = vmdk_read_extent(extent, 1237 cluster_offset, index_in_cluster * 512, 1238 buf, n); 1239 if (ret) { 1240 return ret; 1241 } 1242 } 1243 nb_sectors -= n; 1244 sector_num += n; 1245 buf += n * 512; 1246 } 1247 return 0; 1248 } 1249 1250 static coroutine_fn int vmdk_co_read(BlockDriverState *bs, int64_t sector_num, 1251 uint8_t *buf, int nb_sectors) 1252 { 1253 int ret; 1254 BDRVVmdkState *s = bs->opaque; 1255 qemu_co_mutex_lock(&s->lock); 1256 ret = vmdk_read(bs, sector_num, buf, nb_sectors); 1257 qemu_co_mutex_unlock(&s->lock); 1258 return ret; 1259 } 1260 1261 /** 1262 * vmdk_write: 1263 * @zeroed: buf is ignored (data is zero), use zeroed_grain GTE feature 1264 * if possible, otherwise return -ENOTSUP. 1265 * @zero_dry_run: used for zeroed == true only, don't update L2 table, just try 1266 * with each cluster. By dry run we can find if the zero write 1267 * is possible without modifying image data. 1268 * 1269 * Returns: error code with 0 for success. 1270 */ 1271 static int vmdk_write(BlockDriverState *bs, int64_t sector_num, 1272 const uint8_t *buf, int nb_sectors, 1273 bool zeroed, bool zero_dry_run) 1274 { 1275 BDRVVmdkState *s = bs->opaque; 1276 VmdkExtent *extent = NULL; 1277 int n, ret; 1278 int64_t index_in_cluster; 1279 uint64_t extent_begin_sector, extent_relative_sector_num; 1280 uint64_t cluster_offset; 1281 VmdkMetaData m_data; 1282 1283 if (sector_num > bs->total_sectors) { 1284 fprintf(stderr, 1285 "(VMDK) Wrong offset: sector_num=0x%" PRIx64 1286 " total_sectors=0x%" PRIx64 "\n", 1287 sector_num, bs->total_sectors); 1288 return -EIO; 1289 } 1290 1291 while (nb_sectors > 0) { 1292 extent = find_extent(s, sector_num, extent); 1293 if (!extent) { 1294 return -EIO; 1295 } 1296 ret = get_cluster_offset( 1297 bs, 1298 extent, 1299 &m_data, 1300 sector_num << 9, !extent->compressed, 1301 &cluster_offset); 1302 if (extent->compressed) { 1303 if (ret == VMDK_OK) { 1304 /* Refuse write to allocated cluster for streamOptimized */ 1305 fprintf(stderr, 1306 "VMDK: can't write to allocated cluster" 1307 " for streamOptimized\n"); 1308 return -EIO; 1309 } else { 1310 /* allocate */ 1311 ret = get_cluster_offset( 1312 bs, 1313 extent, 1314 &m_data, 1315 sector_num << 9, 1, 1316 &cluster_offset); 1317 } 1318 } 1319 if (ret == VMDK_ERROR) { 1320 return -EINVAL; 1321 } 1322 extent_begin_sector = extent->end_sector - extent->sectors; 1323 extent_relative_sector_num = sector_num - extent_begin_sector; 1324 index_in_cluster = extent_relative_sector_num % extent->cluster_sectors; 1325 n = extent->cluster_sectors - index_in_cluster; 1326 if (n > nb_sectors) { 1327 n = nb_sectors; 1328 } 1329 if (zeroed) { 1330 /* Do zeroed write, buf is ignored */ 1331 if (extent->has_zero_grain && 1332 index_in_cluster == 0 && 1333 n >= extent->cluster_sectors) { 1334 n = extent->cluster_sectors; 1335 if (!zero_dry_run) { 1336 m_data.offset = VMDK_GTE_ZEROED; 1337 /* update L2 tables */ 1338 if (vmdk_L2update(extent, &m_data) != VMDK_OK) { 1339 return -EIO; 1340 } 1341 } 1342 } else { 1343 return -ENOTSUP; 1344 } 1345 } else { 1346 ret = vmdk_write_extent(extent, 1347 cluster_offset, index_in_cluster * 512, 1348 buf, n, sector_num); 1349 if (ret) { 1350 return ret; 1351 } 1352 if (m_data.valid) { 1353 /* update L2 tables */ 1354 if (vmdk_L2update(extent, &m_data) != VMDK_OK) { 1355 return -EIO; 1356 } 1357 } 1358 } 1359 nb_sectors -= n; 1360 sector_num += n; 1361 buf += n * 512; 1362 1363 /* update CID on the first write every time the virtual disk is 1364 * opened */ 1365 if (!s->cid_updated) { 1366 ret = vmdk_write_cid(bs, time(NULL)); 1367 if (ret < 0) { 1368 return ret; 1369 } 1370 s->cid_updated = true; 1371 } 1372 } 1373 return 0; 1374 } 1375 1376 static coroutine_fn int vmdk_co_write(BlockDriverState *bs, int64_t sector_num, 1377 const uint8_t *buf, int nb_sectors) 1378 { 1379 int ret; 1380 BDRVVmdkState *s = bs->opaque; 1381 qemu_co_mutex_lock(&s->lock); 1382 ret = vmdk_write(bs, sector_num, buf, nb_sectors, false, false); 1383 qemu_co_mutex_unlock(&s->lock); 1384 return ret; 1385 } 1386 1387 static int coroutine_fn vmdk_co_write_zeroes(BlockDriverState *bs, 1388 int64_t sector_num, 1389 int nb_sectors) 1390 { 1391 int ret; 1392 BDRVVmdkState *s = bs->opaque; 1393 qemu_co_mutex_lock(&s->lock); 1394 /* write zeroes could fail if sectors not aligned to cluster, test it with 1395 * dry_run == true before really updating image */ 1396 ret = vmdk_write(bs, sector_num, NULL, nb_sectors, true, true); 1397 if (!ret) { 1398 ret = vmdk_write(bs, sector_num, NULL, nb_sectors, true, false); 1399 } 1400 qemu_co_mutex_unlock(&s->lock); 1401 return ret; 1402 } 1403 1404 static int vmdk_create_extent(const char *filename, int64_t filesize, 1405 bool flat, bool compress, bool zeroed_grain) 1406 { 1407 int ret, i; 1408 int fd = 0; 1409 VMDK4Header header; 1410 uint32_t tmp, magic, grains, gd_size, gt_size, gt_count; 1411 1412 fd = qemu_open(filename, 1413 O_WRONLY | O_CREAT | O_TRUNC | O_BINARY | O_LARGEFILE, 1414 0644); 1415 if (fd < 0) { 1416 return -errno; 1417 } 1418 if (flat) { 1419 ret = ftruncate(fd, filesize); 1420 if (ret < 0) { 1421 ret = -errno; 1422 } 1423 goto exit; 1424 } 1425 magic = cpu_to_be32(VMDK4_MAGIC); 1426 memset(&header, 0, sizeof(header)); 1427 header.version = zeroed_grain ? 2 : 1; 1428 header.flags = VMDK4_FLAG_RGD | VMDK4_FLAG_NL_DETECT 1429 | (compress ? VMDK4_FLAG_COMPRESS | VMDK4_FLAG_MARKER : 0) 1430 | (zeroed_grain ? VMDK4_FLAG_ZERO_GRAIN : 0); 1431 header.compressAlgorithm = compress ? VMDK4_COMPRESSION_DEFLATE : 0; 1432 header.capacity = filesize / 512; 1433 header.granularity = 128; 1434 header.num_gtes_per_gt = 512; 1435 1436 grains = (filesize / 512 + header.granularity - 1) / header.granularity; 1437 gt_size = ((header.num_gtes_per_gt * sizeof(uint32_t)) + 511) >> 9; 1438 gt_count = 1439 (grains + header.num_gtes_per_gt - 1) / header.num_gtes_per_gt; 1440 gd_size = (gt_count * sizeof(uint32_t) + 511) >> 9; 1441 1442 header.desc_offset = 1; 1443 header.desc_size = 20; 1444 header.rgd_offset = header.desc_offset + header.desc_size; 1445 header.gd_offset = header.rgd_offset + gd_size + (gt_size * gt_count); 1446 header.grain_offset = 1447 ((header.gd_offset + gd_size + (gt_size * gt_count) + 1448 header.granularity - 1) / header.granularity) * 1449 header.granularity; 1450 /* swap endianness for all header fields */ 1451 header.version = cpu_to_le32(header.version); 1452 header.flags = cpu_to_le32(header.flags); 1453 header.capacity = cpu_to_le64(header.capacity); 1454 header.granularity = cpu_to_le64(header.granularity); 1455 header.num_gtes_per_gt = cpu_to_le32(header.num_gtes_per_gt); 1456 header.desc_offset = cpu_to_le64(header.desc_offset); 1457 header.desc_size = cpu_to_le64(header.desc_size); 1458 header.rgd_offset = cpu_to_le64(header.rgd_offset); 1459 header.gd_offset = cpu_to_le64(header.gd_offset); 1460 header.grain_offset = cpu_to_le64(header.grain_offset); 1461 header.compressAlgorithm = cpu_to_le16(header.compressAlgorithm); 1462 1463 header.check_bytes[0] = 0xa; 1464 header.check_bytes[1] = 0x20; 1465 header.check_bytes[2] = 0xd; 1466 header.check_bytes[3] = 0xa; 1467 1468 /* write all the data */ 1469 ret = qemu_write_full(fd, &magic, sizeof(magic)); 1470 if (ret != sizeof(magic)) { 1471 ret = -errno; 1472 goto exit; 1473 } 1474 ret = qemu_write_full(fd, &header, sizeof(header)); 1475 if (ret != sizeof(header)) { 1476 ret = -errno; 1477 goto exit; 1478 } 1479 1480 ret = ftruncate(fd, le64_to_cpu(header.grain_offset) << 9); 1481 if (ret < 0) { 1482 ret = -errno; 1483 goto exit; 1484 } 1485 1486 /* write grain directory */ 1487 lseek(fd, le64_to_cpu(header.rgd_offset) << 9, SEEK_SET); 1488 for (i = 0, tmp = le64_to_cpu(header.rgd_offset) + gd_size; 1489 i < gt_count; i++, tmp += gt_size) { 1490 ret = qemu_write_full(fd, &tmp, sizeof(tmp)); 1491 if (ret != sizeof(tmp)) { 1492 ret = -errno; 1493 goto exit; 1494 } 1495 } 1496 1497 /* write backup grain directory */ 1498 lseek(fd, le64_to_cpu(header.gd_offset) << 9, SEEK_SET); 1499 for (i = 0, tmp = le64_to_cpu(header.gd_offset) + gd_size; 1500 i < gt_count; i++, tmp += gt_size) { 1501 ret = qemu_write_full(fd, &tmp, sizeof(tmp)); 1502 if (ret != sizeof(tmp)) { 1503 ret = -errno; 1504 goto exit; 1505 } 1506 } 1507 1508 ret = 0; 1509 exit: 1510 qemu_close(fd); 1511 return ret; 1512 } 1513 1514 static int filename_decompose(const char *filename, char *path, char *prefix, 1515 char *postfix, size_t buf_len) 1516 { 1517 const char *p, *q; 1518 1519 if (filename == NULL || !strlen(filename)) { 1520 fprintf(stderr, "Vmdk: no filename provided.\n"); 1521 return VMDK_ERROR; 1522 } 1523 p = strrchr(filename, '/'); 1524 if (p == NULL) { 1525 p = strrchr(filename, '\\'); 1526 } 1527 if (p == NULL) { 1528 p = strrchr(filename, ':'); 1529 } 1530 if (p != NULL) { 1531 p++; 1532 if (p - filename >= buf_len) { 1533 return VMDK_ERROR; 1534 } 1535 pstrcpy(path, p - filename + 1, filename); 1536 } else { 1537 p = filename; 1538 path[0] = '\0'; 1539 } 1540 q = strrchr(p, '.'); 1541 if (q == NULL) { 1542 pstrcpy(prefix, buf_len, p); 1543 postfix[0] = '\0'; 1544 } else { 1545 if (q - p >= buf_len) { 1546 return VMDK_ERROR; 1547 } 1548 pstrcpy(prefix, q - p + 1, p); 1549 pstrcpy(postfix, buf_len, q); 1550 } 1551 return VMDK_OK; 1552 } 1553 1554 static int vmdk_create(const char *filename, QEMUOptionParameter *options) 1555 { 1556 int fd, idx = 0; 1557 char desc[BUF_SIZE]; 1558 int64_t total_size = 0, filesize; 1559 const char *adapter_type = NULL; 1560 const char *backing_file = NULL; 1561 const char *fmt = NULL; 1562 int flags = 0; 1563 int ret = 0; 1564 bool flat, split, compress; 1565 char ext_desc_lines[BUF_SIZE] = ""; 1566 char path[PATH_MAX], prefix[PATH_MAX], postfix[PATH_MAX]; 1567 const int64_t split_size = 0x80000000; /* VMDK has constant split size */ 1568 const char *desc_extent_line; 1569 char parent_desc_line[BUF_SIZE] = ""; 1570 uint32_t parent_cid = 0xffffffff; 1571 uint32_t number_heads = 16; 1572 bool zeroed_grain = false; 1573 const char desc_template[] = 1574 "# Disk DescriptorFile\n" 1575 "version=1\n" 1576 "CID=%x\n" 1577 "parentCID=%x\n" 1578 "createType=\"%s\"\n" 1579 "%s" 1580 "\n" 1581 "# Extent description\n" 1582 "%s" 1583 "\n" 1584 "# The Disk Data Base\n" 1585 "#DDB\n" 1586 "\n" 1587 "ddb.virtualHWVersion = \"%d\"\n" 1588 "ddb.geometry.cylinders = \"%" PRId64 "\"\n" 1589 "ddb.geometry.heads = \"%d\"\n" 1590 "ddb.geometry.sectors = \"63\"\n" 1591 "ddb.adapterType = \"%s\"\n"; 1592 1593 if (filename_decompose(filename, path, prefix, postfix, PATH_MAX)) { 1594 return -EINVAL; 1595 } 1596 /* Read out options */ 1597 while (options && options->name) { 1598 if (!strcmp(options->name, BLOCK_OPT_SIZE)) { 1599 total_size = options->value.n; 1600 } else if (!strcmp(options->name, BLOCK_OPT_ADAPTER_TYPE)) { 1601 adapter_type = options->value.s; 1602 } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) { 1603 backing_file = options->value.s; 1604 } else if (!strcmp(options->name, BLOCK_OPT_COMPAT6)) { 1605 flags |= options->value.n ? BLOCK_FLAG_COMPAT6 : 0; 1606 } else if (!strcmp(options->name, BLOCK_OPT_SUBFMT)) { 1607 fmt = options->value.s; 1608 } else if (!strcmp(options->name, BLOCK_OPT_ZEROED_GRAIN)) { 1609 zeroed_grain |= options->value.n; 1610 } 1611 options++; 1612 } 1613 if (!adapter_type) { 1614 adapter_type = "ide"; 1615 } else if (strcmp(adapter_type, "ide") && 1616 strcmp(adapter_type, "buslogic") && 1617 strcmp(adapter_type, "lsilogic") && 1618 strcmp(adapter_type, "legacyESX")) { 1619 fprintf(stderr, "VMDK: Unknown adapter type: '%s'.\n", adapter_type); 1620 return -EINVAL; 1621 } 1622 if (strcmp(adapter_type, "ide") != 0) { 1623 /* that's the number of heads with which vmware operates when 1624 creating, exporting, etc. vmdk files with a non-ide adapter type */ 1625 number_heads = 255; 1626 } 1627 if (!fmt) { 1628 /* Default format to monolithicSparse */ 1629 fmt = "monolithicSparse"; 1630 } else if (strcmp(fmt, "monolithicFlat") && 1631 strcmp(fmt, "monolithicSparse") && 1632 strcmp(fmt, "twoGbMaxExtentSparse") && 1633 strcmp(fmt, "twoGbMaxExtentFlat") && 1634 strcmp(fmt, "streamOptimized")) { 1635 fprintf(stderr, "VMDK: Unknown subformat: %s\n", fmt); 1636 return -EINVAL; 1637 } 1638 split = !(strcmp(fmt, "twoGbMaxExtentFlat") && 1639 strcmp(fmt, "twoGbMaxExtentSparse")); 1640 flat = !(strcmp(fmt, "monolithicFlat") && 1641 strcmp(fmt, "twoGbMaxExtentFlat")); 1642 compress = !strcmp(fmt, "streamOptimized"); 1643 if (flat) { 1644 desc_extent_line = "RW %lld FLAT \"%s\" 0\n"; 1645 } else { 1646 desc_extent_line = "RW %lld SPARSE \"%s\"\n"; 1647 } 1648 if (flat && backing_file) { 1649 /* not supporting backing file for flat image */ 1650 return -ENOTSUP; 1651 } 1652 if (backing_file) { 1653 BlockDriverState *bs = bdrv_new(""); 1654 ret = bdrv_open(bs, backing_file, NULL, 0, NULL); 1655 if (ret != 0) { 1656 bdrv_unref(bs); 1657 return ret; 1658 } 1659 if (strcmp(bs->drv->format_name, "vmdk")) { 1660 bdrv_unref(bs); 1661 return -EINVAL; 1662 } 1663 parent_cid = vmdk_read_cid(bs, 0); 1664 bdrv_unref(bs); 1665 snprintf(parent_desc_line, sizeof(parent_desc_line), 1666 "parentFileNameHint=\"%s\"", backing_file); 1667 } 1668 1669 /* Create extents */ 1670 filesize = total_size; 1671 while (filesize > 0) { 1672 char desc_line[BUF_SIZE]; 1673 char ext_filename[PATH_MAX]; 1674 char desc_filename[PATH_MAX]; 1675 int64_t size = filesize; 1676 1677 if (split && size > split_size) { 1678 size = split_size; 1679 } 1680 if (split) { 1681 snprintf(desc_filename, sizeof(desc_filename), "%s-%c%03d%s", 1682 prefix, flat ? 'f' : 's', ++idx, postfix); 1683 } else if (flat) { 1684 snprintf(desc_filename, sizeof(desc_filename), "%s-flat%s", 1685 prefix, postfix); 1686 } else { 1687 snprintf(desc_filename, sizeof(desc_filename), "%s%s", 1688 prefix, postfix); 1689 } 1690 snprintf(ext_filename, sizeof(ext_filename), "%s%s", 1691 path, desc_filename); 1692 1693 if (vmdk_create_extent(ext_filename, size, 1694 flat, compress, zeroed_grain)) { 1695 return -EINVAL; 1696 } 1697 filesize -= size; 1698 1699 /* Format description line */ 1700 snprintf(desc_line, sizeof(desc_line), 1701 desc_extent_line, size / 512, desc_filename); 1702 pstrcat(ext_desc_lines, sizeof(ext_desc_lines), desc_line); 1703 } 1704 /* generate descriptor file */ 1705 snprintf(desc, sizeof(desc), desc_template, 1706 (unsigned int)time(NULL), 1707 parent_cid, 1708 fmt, 1709 parent_desc_line, 1710 ext_desc_lines, 1711 (flags & BLOCK_FLAG_COMPAT6 ? 6 : 4), 1712 total_size / (int64_t)(63 * number_heads * 512), number_heads, 1713 adapter_type); 1714 if (split || flat) { 1715 fd = qemu_open(filename, 1716 O_WRONLY | O_CREAT | O_TRUNC | O_BINARY | O_LARGEFILE, 1717 0644); 1718 } else { 1719 fd = qemu_open(filename, 1720 O_WRONLY | O_BINARY | O_LARGEFILE, 1721 0644); 1722 } 1723 if (fd < 0) { 1724 return -errno; 1725 } 1726 /* the descriptor offset = 0x200 */ 1727 if (!split && !flat && 0x200 != lseek(fd, 0x200, SEEK_SET)) { 1728 ret = -errno; 1729 goto exit; 1730 } 1731 ret = qemu_write_full(fd, desc, strlen(desc)); 1732 if (ret != strlen(desc)) { 1733 ret = -errno; 1734 goto exit; 1735 } 1736 ret = 0; 1737 exit: 1738 qemu_close(fd); 1739 return ret; 1740 } 1741 1742 static void vmdk_close(BlockDriverState *bs) 1743 { 1744 BDRVVmdkState *s = bs->opaque; 1745 1746 vmdk_free_extents(bs); 1747 1748 migrate_del_blocker(s->migration_blocker); 1749 error_free(s->migration_blocker); 1750 } 1751 1752 static coroutine_fn int vmdk_co_flush(BlockDriverState *bs) 1753 { 1754 BDRVVmdkState *s = bs->opaque; 1755 int i, err; 1756 int ret = 0; 1757 1758 for (i = 0; i < s->num_extents; i++) { 1759 err = bdrv_co_flush(s->extents[i].file); 1760 if (err < 0) { 1761 ret = err; 1762 } 1763 } 1764 return ret; 1765 } 1766 1767 static int64_t vmdk_get_allocated_file_size(BlockDriverState *bs) 1768 { 1769 int i; 1770 int64_t ret = 0; 1771 int64_t r; 1772 BDRVVmdkState *s = bs->opaque; 1773 1774 ret = bdrv_get_allocated_file_size(bs->file); 1775 if (ret < 0) { 1776 return ret; 1777 } 1778 for (i = 0; i < s->num_extents; i++) { 1779 if (s->extents[i].file == bs->file) { 1780 continue; 1781 } 1782 r = bdrv_get_allocated_file_size(s->extents[i].file); 1783 if (r < 0) { 1784 return r; 1785 } 1786 ret += r; 1787 } 1788 return ret; 1789 } 1790 1791 static int vmdk_has_zero_init(BlockDriverState *bs) 1792 { 1793 int i; 1794 BDRVVmdkState *s = bs->opaque; 1795 1796 /* If has a flat extent and its underlying storage doesn't have zero init, 1797 * return 0. */ 1798 for (i = 0; i < s->num_extents; i++) { 1799 if (s->extents[i].flat) { 1800 if (!bdrv_has_zero_init(s->extents[i].file)) { 1801 return 0; 1802 } 1803 } 1804 } 1805 return 1; 1806 } 1807 1808 static QEMUOptionParameter vmdk_create_options[] = { 1809 { 1810 .name = BLOCK_OPT_SIZE, 1811 .type = OPT_SIZE, 1812 .help = "Virtual disk size" 1813 }, 1814 { 1815 .name = BLOCK_OPT_ADAPTER_TYPE, 1816 .type = OPT_STRING, 1817 .help = "Virtual adapter type, can be one of " 1818 "ide (default), lsilogic, buslogic or legacyESX" 1819 }, 1820 { 1821 .name = BLOCK_OPT_BACKING_FILE, 1822 .type = OPT_STRING, 1823 .help = "File name of a base image" 1824 }, 1825 { 1826 .name = BLOCK_OPT_COMPAT6, 1827 .type = OPT_FLAG, 1828 .help = "VMDK version 6 image" 1829 }, 1830 { 1831 .name = BLOCK_OPT_SUBFMT, 1832 .type = OPT_STRING, 1833 .help = 1834 "VMDK flat extent format, can be one of " 1835 "{monolithicSparse (default) | monolithicFlat | twoGbMaxExtentSparse | twoGbMaxExtentFlat | streamOptimized} " 1836 }, 1837 { 1838 .name = BLOCK_OPT_ZEROED_GRAIN, 1839 .type = OPT_FLAG, 1840 .help = "Enable efficient zero writes using the zeroed-grain GTE feature" 1841 }, 1842 { NULL } 1843 }; 1844 1845 static BlockDriver bdrv_vmdk = { 1846 .format_name = "vmdk", 1847 .instance_size = sizeof(BDRVVmdkState), 1848 .bdrv_probe = vmdk_probe, 1849 .bdrv_open = vmdk_open, 1850 .bdrv_reopen_prepare = vmdk_reopen_prepare, 1851 .bdrv_read = vmdk_co_read, 1852 .bdrv_write = vmdk_co_write, 1853 .bdrv_co_write_zeroes = vmdk_co_write_zeroes, 1854 .bdrv_close = vmdk_close, 1855 .bdrv_create = vmdk_create, 1856 .bdrv_co_flush_to_disk = vmdk_co_flush, 1857 .bdrv_co_get_block_status = vmdk_co_get_block_status, 1858 .bdrv_get_allocated_file_size = vmdk_get_allocated_file_size, 1859 .bdrv_has_zero_init = vmdk_has_zero_init, 1860 1861 .create_options = vmdk_create_options, 1862 }; 1863 1864 static void bdrv_vmdk_init(void) 1865 { 1866 bdrv_register(&bdrv_vmdk); 1867 } 1868 1869 block_init(bdrv_vmdk_init); 1870