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 } VMDK3Header; 66 67 typedef struct { 68 uint32_t version; 69 uint32_t flags; 70 int64_t capacity; 71 int64_t granularity; 72 int64_t desc_offset; 73 int64_t desc_size; 74 int32_t num_gtes_per_gte; 75 int64_t rgd_offset; 76 int64_t gd_offset; 77 int64_t grain_offset; 78 char filler[1]; 79 char check_bytes[4]; 80 uint16_t compressAlgorithm; 81 } QEMU_PACKED VMDK4Header; 82 83 #define L2_CACHE_SIZE 16 84 85 typedef struct VmdkExtent { 86 BlockDriverState *file; 87 bool flat; 88 bool compressed; 89 bool has_marker; 90 bool has_zero_grain; 91 int version; 92 int64_t sectors; 93 int64_t end_sector; 94 int64_t flat_start_offset; 95 int64_t l1_table_offset; 96 int64_t l1_backup_table_offset; 97 uint32_t *l1_table; 98 uint32_t *l1_backup_table; 99 unsigned int l1_size; 100 uint32_t l1_entry_sectors; 101 102 unsigned int l2_size; 103 uint32_t *l2_cache; 104 uint32_t l2_cache_offsets[L2_CACHE_SIZE]; 105 uint32_t l2_cache_counts[L2_CACHE_SIZE]; 106 107 unsigned int cluster_sectors; 108 } VmdkExtent; 109 110 typedef struct BDRVVmdkState { 111 CoMutex lock; 112 int desc_offset; 113 bool cid_updated; 114 uint32_t parent_cid; 115 int num_extents; 116 /* Extent array with num_extents entries, ascend ordered by address */ 117 VmdkExtent *extents; 118 Error *migration_blocker; 119 } BDRVVmdkState; 120 121 typedef struct VmdkMetaData { 122 uint32_t offset; 123 unsigned int l1_index; 124 unsigned int l2_index; 125 unsigned int l2_offset; 126 int valid; 127 uint32_t *l2_cache_entry; 128 } VmdkMetaData; 129 130 typedef struct VmdkGrainMarker { 131 uint64_t lba; 132 uint32_t size; 133 uint8_t data[0]; 134 } VmdkGrainMarker; 135 136 enum { 137 MARKER_END_OF_STREAM = 0, 138 MARKER_GRAIN_TABLE = 1, 139 MARKER_GRAIN_DIRECTORY = 2, 140 MARKER_FOOTER = 3, 141 }; 142 143 static int vmdk_probe(const uint8_t *buf, int buf_size, const char *filename) 144 { 145 uint32_t magic; 146 147 if (buf_size < 4) { 148 return 0; 149 } 150 magic = be32_to_cpu(*(uint32_t *)buf); 151 if (magic == VMDK3_MAGIC || 152 magic == VMDK4_MAGIC) { 153 return 100; 154 } else { 155 const char *p = (const char *)buf; 156 const char *end = p + buf_size; 157 while (p < end) { 158 if (*p == '#') { 159 /* skip comment line */ 160 while (p < end && *p != '\n') { 161 p++; 162 } 163 p++; 164 continue; 165 } 166 if (*p == ' ') { 167 while (p < end && *p == ' ') { 168 p++; 169 } 170 /* skip '\r' if windows line endings used. */ 171 if (p < end && *p == '\r') { 172 p++; 173 } 174 /* only accept blank lines before 'version=' line */ 175 if (p == end || *p != '\n') { 176 return 0; 177 } 178 p++; 179 continue; 180 } 181 if (end - p >= strlen("version=X\n")) { 182 if (strncmp("version=1\n", p, strlen("version=1\n")) == 0 || 183 strncmp("version=2\n", p, strlen("version=2\n")) == 0) { 184 return 100; 185 } 186 } 187 if (end - p >= strlen("version=X\r\n")) { 188 if (strncmp("version=1\r\n", p, strlen("version=1\r\n")) == 0 || 189 strncmp("version=2\r\n", p, strlen("version=2\r\n")) == 0) { 190 return 100; 191 } 192 } 193 return 0; 194 } 195 return 0; 196 } 197 } 198 199 #define CHECK_CID 1 200 201 #define SECTOR_SIZE 512 202 #define DESC_SIZE (20 * SECTOR_SIZE) /* 20 sectors of 512 bytes each */ 203 #define BUF_SIZE 4096 204 #define HEADER_SIZE 512 /* first sector of 512 bytes */ 205 206 static void vmdk_free_extents(BlockDriverState *bs) 207 { 208 int i; 209 BDRVVmdkState *s = bs->opaque; 210 VmdkExtent *e; 211 212 for (i = 0; i < s->num_extents; i++) { 213 e = &s->extents[i]; 214 g_free(e->l1_table); 215 g_free(e->l2_cache); 216 g_free(e->l1_backup_table); 217 if (e->file != bs->file) { 218 bdrv_delete(e->file); 219 } 220 } 221 g_free(s->extents); 222 } 223 224 static void vmdk_free_last_extent(BlockDriverState *bs) 225 { 226 BDRVVmdkState *s = bs->opaque; 227 228 if (s->num_extents == 0) { 229 return; 230 } 231 s->num_extents--; 232 s->extents = g_realloc(s->extents, s->num_extents * sizeof(VmdkExtent)); 233 } 234 235 static uint32_t vmdk_read_cid(BlockDriverState *bs, int parent) 236 { 237 char desc[DESC_SIZE]; 238 uint32_t cid = 0xffffffff; 239 const char *p_name, *cid_str; 240 size_t cid_str_size; 241 BDRVVmdkState *s = bs->opaque; 242 int ret; 243 244 ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE); 245 if (ret < 0) { 246 return 0; 247 } 248 249 if (parent) { 250 cid_str = "parentCID"; 251 cid_str_size = sizeof("parentCID"); 252 } else { 253 cid_str = "CID"; 254 cid_str_size = sizeof("CID"); 255 } 256 257 desc[DESC_SIZE - 1] = '\0'; 258 p_name = strstr(desc, cid_str); 259 if (p_name != NULL) { 260 p_name += cid_str_size; 261 sscanf(p_name, "%x", &cid); 262 } 263 264 return cid; 265 } 266 267 static int vmdk_write_cid(BlockDriverState *bs, uint32_t cid) 268 { 269 char desc[DESC_SIZE], tmp_desc[DESC_SIZE]; 270 char *p_name, *tmp_str; 271 BDRVVmdkState *s = bs->opaque; 272 int ret; 273 274 ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE); 275 if (ret < 0) { 276 return ret; 277 } 278 279 desc[DESC_SIZE - 1] = '\0'; 280 tmp_str = strstr(desc, "parentCID"); 281 if (tmp_str == NULL) { 282 return -EINVAL; 283 } 284 285 pstrcpy(tmp_desc, sizeof(tmp_desc), tmp_str); 286 p_name = strstr(desc, "CID"); 287 if (p_name != NULL) { 288 p_name += sizeof("CID"); 289 snprintf(p_name, sizeof(desc) - (p_name - desc), "%x\n", cid); 290 pstrcat(desc, sizeof(desc), tmp_desc); 291 } 292 293 ret = bdrv_pwrite_sync(bs->file, s->desc_offset, desc, DESC_SIZE); 294 if (ret < 0) { 295 return ret; 296 } 297 298 return 0; 299 } 300 301 static int vmdk_is_cid_valid(BlockDriverState *bs) 302 { 303 #ifdef CHECK_CID 304 BDRVVmdkState *s = bs->opaque; 305 BlockDriverState *p_bs = bs->backing_hd; 306 uint32_t cur_pcid; 307 308 if (p_bs) { 309 cur_pcid = vmdk_read_cid(p_bs, 0); 310 if (s->parent_cid != cur_pcid) { 311 /* CID not valid */ 312 return 0; 313 } 314 } 315 #endif 316 /* CID valid */ 317 return 1; 318 } 319 320 /* Queue extents, if any, for reopen() */ 321 static int vmdk_reopen_prepare(BDRVReopenState *state, 322 BlockReopenQueue *queue, Error **errp) 323 { 324 BDRVVmdkState *s; 325 int ret = -1; 326 int i; 327 VmdkExtent *e; 328 329 assert(state != NULL); 330 assert(state->bs != NULL); 331 332 if (queue == NULL) { 333 error_set(errp, ERROR_CLASS_GENERIC_ERROR, 334 "No reopen queue for VMDK extents"); 335 goto exit; 336 } 337 338 s = state->bs->opaque; 339 340 assert(s != NULL); 341 342 for (i = 0; i < s->num_extents; i++) { 343 e = &s->extents[i]; 344 if (e->file != state->bs->file) { 345 bdrv_reopen_queue(queue, e->file, state->flags); 346 } 347 } 348 ret = 0; 349 350 exit: 351 return ret; 352 } 353 354 static int vmdk_parent_open(BlockDriverState *bs) 355 { 356 char *p_name; 357 char desc[DESC_SIZE + 1]; 358 BDRVVmdkState *s = bs->opaque; 359 int ret; 360 361 desc[DESC_SIZE] = '\0'; 362 ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE); 363 if (ret < 0) { 364 return ret; 365 } 366 367 p_name = strstr(desc, "parentFileNameHint"); 368 if (p_name != NULL) { 369 char *end_name; 370 371 p_name += sizeof("parentFileNameHint") + 1; 372 end_name = strchr(p_name, '\"'); 373 if (end_name == NULL) { 374 return -EINVAL; 375 } 376 if ((end_name - p_name) > sizeof(bs->backing_file) - 1) { 377 return -EINVAL; 378 } 379 380 pstrcpy(bs->backing_file, end_name - p_name + 1, p_name); 381 } 382 383 return 0; 384 } 385 386 /* Create and append extent to the extent array. Return the added VmdkExtent 387 * address. return NULL if allocation failed. */ 388 static VmdkExtent *vmdk_add_extent(BlockDriverState *bs, 389 BlockDriverState *file, bool flat, int64_t sectors, 390 int64_t l1_offset, int64_t l1_backup_offset, 391 uint32_t l1_size, 392 int l2_size, unsigned int cluster_sectors) 393 { 394 VmdkExtent *extent; 395 BDRVVmdkState *s = bs->opaque; 396 397 s->extents = g_realloc(s->extents, 398 (s->num_extents + 1) * sizeof(VmdkExtent)); 399 extent = &s->extents[s->num_extents]; 400 s->num_extents++; 401 402 memset(extent, 0, sizeof(VmdkExtent)); 403 extent->file = file; 404 extent->flat = flat; 405 extent->sectors = sectors; 406 extent->l1_table_offset = l1_offset; 407 extent->l1_backup_table_offset = l1_backup_offset; 408 extent->l1_size = l1_size; 409 extent->l1_entry_sectors = l2_size * cluster_sectors; 410 extent->l2_size = l2_size; 411 extent->cluster_sectors = cluster_sectors; 412 413 if (s->num_extents > 1) { 414 extent->end_sector = (*(extent - 1)).end_sector + extent->sectors; 415 } else { 416 extent->end_sector = extent->sectors; 417 } 418 bs->total_sectors = extent->end_sector; 419 return extent; 420 } 421 422 static int vmdk_init_tables(BlockDriverState *bs, VmdkExtent *extent) 423 { 424 int ret; 425 int l1_size, i; 426 427 /* read the L1 table */ 428 l1_size = extent->l1_size * sizeof(uint32_t); 429 extent->l1_table = g_malloc(l1_size); 430 ret = bdrv_pread(extent->file, 431 extent->l1_table_offset, 432 extent->l1_table, 433 l1_size); 434 if (ret < 0) { 435 goto fail_l1; 436 } 437 for (i = 0; i < extent->l1_size; i++) { 438 le32_to_cpus(&extent->l1_table[i]); 439 } 440 441 if (extent->l1_backup_table_offset) { 442 extent->l1_backup_table = g_malloc(l1_size); 443 ret = bdrv_pread(extent->file, 444 extent->l1_backup_table_offset, 445 extent->l1_backup_table, 446 l1_size); 447 if (ret < 0) { 448 goto fail_l1b; 449 } 450 for (i = 0; i < extent->l1_size; i++) { 451 le32_to_cpus(&extent->l1_backup_table[i]); 452 } 453 } 454 455 extent->l2_cache = 456 g_malloc(extent->l2_size * L2_CACHE_SIZE * sizeof(uint32_t)); 457 return 0; 458 fail_l1b: 459 g_free(extent->l1_backup_table); 460 fail_l1: 461 g_free(extent->l1_table); 462 return ret; 463 } 464 465 static int vmdk_open_vmdk3(BlockDriverState *bs, 466 BlockDriverState *file, 467 int flags) 468 { 469 int ret; 470 uint32_t magic; 471 VMDK3Header header; 472 VmdkExtent *extent; 473 474 ret = bdrv_pread(file, sizeof(magic), &header, sizeof(header)); 475 if (ret < 0) { 476 return ret; 477 } 478 extent = vmdk_add_extent(bs, 479 bs->file, false, 480 le32_to_cpu(header.disk_sectors), 481 le32_to_cpu(header.l1dir_offset) << 9, 482 0, 1 << 6, 1 << 9, 483 le32_to_cpu(header.granularity)); 484 ret = vmdk_init_tables(bs, extent); 485 if (ret) { 486 /* free extent allocated by vmdk_add_extent */ 487 vmdk_free_last_extent(bs); 488 } 489 return ret; 490 } 491 492 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, 493 int64_t desc_offset); 494 495 static int vmdk_open_vmdk4(BlockDriverState *bs, 496 BlockDriverState *file, 497 int flags) 498 { 499 int ret; 500 uint32_t magic; 501 uint32_t l1_size, l1_entry_sectors; 502 VMDK4Header header; 503 VmdkExtent *extent; 504 int64_t l1_backup_offset = 0; 505 506 ret = bdrv_pread(file, sizeof(magic), &header, sizeof(header)); 507 if (ret < 0) { 508 return ret; 509 } 510 if (header.capacity == 0) { 511 int64_t desc_offset = le64_to_cpu(header.desc_offset); 512 if (desc_offset) { 513 return vmdk_open_desc_file(bs, flags, desc_offset << 9); 514 } 515 } 516 517 if (le64_to_cpu(header.gd_offset) == VMDK4_GD_AT_END) { 518 /* 519 * The footer takes precedence over the header, so read it in. The 520 * footer starts at offset -1024 from the end: One sector for the 521 * footer, and another one for the end-of-stream marker. 522 */ 523 struct { 524 struct { 525 uint64_t val; 526 uint32_t size; 527 uint32_t type; 528 uint8_t pad[512 - 16]; 529 } QEMU_PACKED footer_marker; 530 531 uint32_t magic; 532 VMDK4Header header; 533 uint8_t pad[512 - 4 - sizeof(VMDK4Header)]; 534 535 struct { 536 uint64_t val; 537 uint32_t size; 538 uint32_t type; 539 uint8_t pad[512 - 16]; 540 } QEMU_PACKED eos_marker; 541 } QEMU_PACKED footer; 542 543 ret = bdrv_pread(file, 544 bs->file->total_sectors * 512 - 1536, 545 &footer, sizeof(footer)); 546 if (ret < 0) { 547 return ret; 548 } 549 550 /* Some sanity checks for the footer */ 551 if (be32_to_cpu(footer.magic) != VMDK4_MAGIC || 552 le32_to_cpu(footer.footer_marker.size) != 0 || 553 le32_to_cpu(footer.footer_marker.type) != MARKER_FOOTER || 554 le64_to_cpu(footer.eos_marker.val) != 0 || 555 le32_to_cpu(footer.eos_marker.size) != 0 || 556 le32_to_cpu(footer.eos_marker.type) != MARKER_END_OF_STREAM) 557 { 558 return -EINVAL; 559 } 560 561 header = footer.header; 562 } 563 564 l1_entry_sectors = le32_to_cpu(header.num_gtes_per_gte) 565 * le64_to_cpu(header.granularity); 566 if (l1_entry_sectors == 0) { 567 return -EINVAL; 568 } 569 l1_size = (le64_to_cpu(header.capacity) + l1_entry_sectors - 1) 570 / l1_entry_sectors; 571 if (le32_to_cpu(header.flags) & VMDK4_FLAG_RGD) { 572 l1_backup_offset = le64_to_cpu(header.rgd_offset) << 9; 573 } 574 extent = vmdk_add_extent(bs, file, false, 575 le64_to_cpu(header.capacity), 576 le64_to_cpu(header.gd_offset) << 9, 577 l1_backup_offset, 578 l1_size, 579 le32_to_cpu(header.num_gtes_per_gte), 580 le64_to_cpu(header.granularity)); 581 extent->compressed = 582 le16_to_cpu(header.compressAlgorithm) == VMDK4_COMPRESSION_DEFLATE; 583 extent->has_marker = le32_to_cpu(header.flags) & VMDK4_FLAG_MARKER; 584 extent->version = le32_to_cpu(header.version); 585 extent->has_zero_grain = le32_to_cpu(header.flags) & VMDK4_FLAG_ZERO_GRAIN; 586 ret = vmdk_init_tables(bs, extent); 587 if (ret) { 588 /* free extent allocated by vmdk_add_extent */ 589 vmdk_free_last_extent(bs); 590 } 591 return ret; 592 } 593 594 /* find an option value out of descriptor file */ 595 static int vmdk_parse_description(const char *desc, const char *opt_name, 596 char *buf, int buf_size) 597 { 598 char *opt_pos, *opt_end; 599 const char *end = desc + strlen(desc); 600 601 opt_pos = strstr(desc, opt_name); 602 if (!opt_pos) { 603 return VMDK_ERROR; 604 } 605 /* Skip "=\"" following opt_name */ 606 opt_pos += strlen(opt_name) + 2; 607 if (opt_pos >= end) { 608 return VMDK_ERROR; 609 } 610 opt_end = opt_pos; 611 while (opt_end < end && *opt_end != '"') { 612 opt_end++; 613 } 614 if (opt_end == end || buf_size < opt_end - opt_pos + 1) { 615 return VMDK_ERROR; 616 } 617 pstrcpy(buf, opt_end - opt_pos + 1, opt_pos); 618 return VMDK_OK; 619 } 620 621 /* Open an extent file and append to bs array */ 622 static int vmdk_open_sparse(BlockDriverState *bs, 623 BlockDriverState *file, 624 int flags) 625 { 626 uint32_t magic; 627 628 if (bdrv_pread(file, 0, &magic, sizeof(magic)) != sizeof(magic)) { 629 return -EIO; 630 } 631 632 magic = be32_to_cpu(magic); 633 switch (magic) { 634 case VMDK3_MAGIC: 635 return vmdk_open_vmdk3(bs, file, flags); 636 break; 637 case VMDK4_MAGIC: 638 return vmdk_open_vmdk4(bs, file, flags); 639 break; 640 default: 641 return -EMEDIUMTYPE; 642 break; 643 } 644 } 645 646 static int vmdk_parse_extents(const char *desc, BlockDriverState *bs, 647 const char *desc_file_path) 648 { 649 int ret; 650 char access[11]; 651 char type[11]; 652 char fname[512]; 653 const char *p = desc; 654 int64_t sectors = 0; 655 int64_t flat_offset; 656 char extent_path[PATH_MAX]; 657 BlockDriverState *extent_file; 658 659 while (*p) { 660 /* parse extent line: 661 * RW [size in sectors] FLAT "file-name.vmdk" OFFSET 662 * or 663 * RW [size in sectors] SPARSE "file-name.vmdk" 664 */ 665 flat_offset = -1; 666 ret = sscanf(p, "%10s %" SCNd64 " %10s \"%511[^\n\r\"]\" %" SCNd64, 667 access, §ors, type, fname, &flat_offset); 668 if (ret < 4 || strcmp(access, "RW")) { 669 goto next_line; 670 } else if (!strcmp(type, "FLAT")) { 671 if (ret != 5 || flat_offset < 0) { 672 return -EINVAL; 673 } 674 } else if (ret != 4) { 675 return -EINVAL; 676 } 677 678 if (sectors <= 0 || 679 (strcmp(type, "FLAT") && strcmp(type, "SPARSE")) || 680 (strcmp(access, "RW"))) { 681 goto next_line; 682 } 683 684 path_combine(extent_path, sizeof(extent_path), 685 desc_file_path, fname); 686 ret = bdrv_file_open(&extent_file, extent_path, NULL, bs->open_flags); 687 if (ret) { 688 return ret; 689 } 690 691 /* save to extents array */ 692 if (!strcmp(type, "FLAT")) { 693 /* FLAT extent */ 694 VmdkExtent *extent; 695 696 extent = vmdk_add_extent(bs, extent_file, true, sectors, 697 0, 0, 0, 0, sectors); 698 extent->flat_start_offset = flat_offset << 9; 699 } else if (!strcmp(type, "SPARSE")) { 700 /* SPARSE extent */ 701 ret = vmdk_open_sparse(bs, extent_file, bs->open_flags); 702 if (ret) { 703 bdrv_delete(extent_file); 704 return ret; 705 } 706 } else { 707 fprintf(stderr, 708 "VMDK: Not supported extent type \"%s\""".\n", type); 709 return -ENOTSUP; 710 } 711 next_line: 712 /* move to next line */ 713 while (*p && *p != '\n') { 714 p++; 715 } 716 p++; 717 } 718 return 0; 719 } 720 721 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, 722 int64_t desc_offset) 723 { 724 int ret; 725 char *buf = NULL; 726 char ct[128]; 727 BDRVVmdkState *s = bs->opaque; 728 int64_t size; 729 730 size = bdrv_getlength(bs->file); 731 if (size < 0) { 732 return -EINVAL; 733 } 734 735 size = MIN(size, 1 << 20); /* avoid unbounded allocation */ 736 buf = g_malloc0(size + 1); 737 738 ret = bdrv_pread(bs->file, desc_offset, buf, size); 739 if (ret < 0) { 740 goto exit; 741 } 742 if (vmdk_parse_description(buf, "createType", ct, sizeof(ct))) { 743 ret = -EMEDIUMTYPE; 744 goto exit; 745 } 746 if (strcmp(ct, "monolithicFlat") && 747 strcmp(ct, "twoGbMaxExtentSparse") && 748 strcmp(ct, "twoGbMaxExtentFlat")) { 749 fprintf(stderr, 750 "VMDK: Not supported image type \"%s\""".\n", ct); 751 ret = -ENOTSUP; 752 goto exit; 753 } 754 s->desc_offset = 0; 755 ret = vmdk_parse_extents(buf, bs, bs->file->filename); 756 exit: 757 g_free(buf); 758 return ret; 759 } 760 761 static int vmdk_open(BlockDriverState *bs, QDict *options, int flags) 762 { 763 int ret; 764 BDRVVmdkState *s = bs->opaque; 765 766 if (vmdk_open_sparse(bs, bs->file, flags) == 0) { 767 s->desc_offset = 0x200; 768 } else { 769 ret = vmdk_open_desc_file(bs, flags, 0); 770 if (ret) { 771 goto fail; 772 } 773 } 774 /* try to open parent images, if exist */ 775 ret = vmdk_parent_open(bs); 776 if (ret) { 777 goto fail; 778 } 779 s->parent_cid = vmdk_read_cid(bs, 1); 780 qemu_co_mutex_init(&s->lock); 781 782 /* Disable migration when VMDK images are used */ 783 error_set(&s->migration_blocker, 784 QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED, 785 "vmdk", bs->device_name, "live migration"); 786 migrate_add_blocker(s->migration_blocker); 787 788 return 0; 789 790 fail: 791 vmdk_free_extents(bs); 792 return ret; 793 } 794 795 static int get_whole_cluster(BlockDriverState *bs, 796 VmdkExtent *extent, 797 uint64_t cluster_offset, 798 uint64_t offset, 799 bool allocate) 800 { 801 /* 128 sectors * 512 bytes each = grain size 64KB */ 802 uint8_t whole_grain[extent->cluster_sectors * 512]; 803 804 /* we will be here if it's first write on non-exist grain(cluster). 805 * try to read from parent image, if exist */ 806 if (bs->backing_hd) { 807 int ret; 808 809 if (!vmdk_is_cid_valid(bs)) { 810 return VMDK_ERROR; 811 } 812 813 /* floor offset to cluster */ 814 offset -= offset % (extent->cluster_sectors * 512); 815 ret = bdrv_read(bs->backing_hd, offset >> 9, whole_grain, 816 extent->cluster_sectors); 817 if (ret < 0) { 818 return VMDK_ERROR; 819 } 820 821 /* Write grain only into the active image */ 822 ret = bdrv_write(extent->file, cluster_offset, whole_grain, 823 extent->cluster_sectors); 824 if (ret < 0) { 825 return VMDK_ERROR; 826 } 827 } 828 return VMDK_OK; 829 } 830 831 static int vmdk_L2update(VmdkExtent *extent, VmdkMetaData *m_data) 832 { 833 uint32_t offset; 834 QEMU_BUILD_BUG_ON(sizeof(offset) != sizeof(m_data->offset)); 835 offset = cpu_to_le32(m_data->offset); 836 /* update L2 table */ 837 if (bdrv_pwrite_sync( 838 extent->file, 839 ((int64_t)m_data->l2_offset * 512) 840 + (m_data->l2_index * sizeof(m_data->offset)), 841 &offset, sizeof(offset)) < 0) { 842 return VMDK_ERROR; 843 } 844 /* update backup L2 table */ 845 if (extent->l1_backup_table_offset != 0) { 846 m_data->l2_offset = extent->l1_backup_table[m_data->l1_index]; 847 if (bdrv_pwrite_sync( 848 extent->file, 849 ((int64_t)m_data->l2_offset * 512) 850 + (m_data->l2_index * sizeof(m_data->offset)), 851 &offset, sizeof(offset)) < 0) { 852 return VMDK_ERROR; 853 } 854 } 855 if (m_data->l2_cache_entry) { 856 *m_data->l2_cache_entry = offset; 857 } 858 859 return VMDK_OK; 860 } 861 862 static int get_cluster_offset(BlockDriverState *bs, 863 VmdkExtent *extent, 864 VmdkMetaData *m_data, 865 uint64_t offset, 866 int allocate, 867 uint64_t *cluster_offset) 868 { 869 unsigned int l1_index, l2_offset, l2_index; 870 int min_index, i, j; 871 uint32_t min_count, *l2_table; 872 bool zeroed = false; 873 874 if (m_data) { 875 m_data->valid = 0; 876 } 877 if (extent->flat) { 878 *cluster_offset = extent->flat_start_offset; 879 return VMDK_OK; 880 } 881 882 offset -= (extent->end_sector - extent->sectors) * SECTOR_SIZE; 883 l1_index = (offset >> 9) / extent->l1_entry_sectors; 884 if (l1_index >= extent->l1_size) { 885 return VMDK_ERROR; 886 } 887 l2_offset = extent->l1_table[l1_index]; 888 if (!l2_offset) { 889 return VMDK_UNALLOC; 890 } 891 for (i = 0; i < L2_CACHE_SIZE; i++) { 892 if (l2_offset == extent->l2_cache_offsets[i]) { 893 /* increment the hit count */ 894 if (++extent->l2_cache_counts[i] == 0xffffffff) { 895 for (j = 0; j < L2_CACHE_SIZE; j++) { 896 extent->l2_cache_counts[j] >>= 1; 897 } 898 } 899 l2_table = extent->l2_cache + (i * extent->l2_size); 900 goto found; 901 } 902 } 903 /* not found: load a new entry in the least used one */ 904 min_index = 0; 905 min_count = 0xffffffff; 906 for (i = 0; i < L2_CACHE_SIZE; i++) { 907 if (extent->l2_cache_counts[i] < min_count) { 908 min_count = extent->l2_cache_counts[i]; 909 min_index = i; 910 } 911 } 912 l2_table = extent->l2_cache + (min_index * extent->l2_size); 913 if (bdrv_pread( 914 extent->file, 915 (int64_t)l2_offset * 512, 916 l2_table, 917 extent->l2_size * sizeof(uint32_t) 918 ) != extent->l2_size * sizeof(uint32_t)) { 919 return VMDK_ERROR; 920 } 921 922 extent->l2_cache_offsets[min_index] = l2_offset; 923 extent->l2_cache_counts[min_index] = 1; 924 found: 925 l2_index = ((offset >> 9) / extent->cluster_sectors) % extent->l2_size; 926 *cluster_offset = le32_to_cpu(l2_table[l2_index]); 927 928 if (m_data) { 929 m_data->valid = 1; 930 m_data->l1_index = l1_index; 931 m_data->l2_index = l2_index; 932 m_data->offset = *cluster_offset; 933 m_data->l2_offset = l2_offset; 934 m_data->l2_cache_entry = &l2_table[l2_index]; 935 } 936 if (extent->has_zero_grain && *cluster_offset == VMDK_GTE_ZEROED) { 937 zeroed = true; 938 } 939 940 if (!*cluster_offset || zeroed) { 941 if (!allocate) { 942 return zeroed ? VMDK_ZEROED : VMDK_UNALLOC; 943 } 944 945 /* Avoid the L2 tables update for the images that have snapshots. */ 946 *cluster_offset = bdrv_getlength(extent->file); 947 if (!extent->compressed) { 948 bdrv_truncate( 949 extent->file, 950 *cluster_offset + (extent->cluster_sectors << 9) 951 ); 952 } 953 954 *cluster_offset >>= 9; 955 l2_table[l2_index] = cpu_to_le32(*cluster_offset); 956 957 /* First of all we write grain itself, to avoid race condition 958 * that may to corrupt the image. 959 * This problem may occur because of insufficient space on host disk 960 * or inappropriate VM shutdown. 961 */ 962 if (get_whole_cluster( 963 bs, extent, *cluster_offset, offset, allocate) == -1) { 964 return VMDK_ERROR; 965 } 966 967 if (m_data) { 968 m_data->offset = *cluster_offset; 969 } 970 } 971 *cluster_offset <<= 9; 972 return VMDK_OK; 973 } 974 975 static VmdkExtent *find_extent(BDRVVmdkState *s, 976 int64_t sector_num, VmdkExtent *start_hint) 977 { 978 VmdkExtent *extent = start_hint; 979 980 if (!extent) { 981 extent = &s->extents[0]; 982 } 983 while (extent < &s->extents[s->num_extents]) { 984 if (sector_num < extent->end_sector) { 985 return extent; 986 } 987 extent++; 988 } 989 return NULL; 990 } 991 992 static int coroutine_fn vmdk_co_is_allocated(BlockDriverState *bs, 993 int64_t sector_num, int nb_sectors, int *pnum) 994 { 995 BDRVVmdkState *s = bs->opaque; 996 int64_t index_in_cluster, n, ret; 997 uint64_t offset; 998 VmdkExtent *extent; 999 1000 extent = find_extent(s, sector_num, NULL); 1001 if (!extent) { 1002 return 0; 1003 } 1004 qemu_co_mutex_lock(&s->lock); 1005 ret = get_cluster_offset(bs, extent, NULL, 1006 sector_num * 512, 0, &offset); 1007 qemu_co_mutex_unlock(&s->lock); 1008 1009 ret = (ret == VMDK_OK || ret == VMDK_ZEROED); 1010 1011 index_in_cluster = sector_num % extent->cluster_sectors; 1012 n = extent->cluster_sectors - index_in_cluster; 1013 if (n > nb_sectors) { 1014 n = nb_sectors; 1015 } 1016 *pnum = n; 1017 return ret; 1018 } 1019 1020 static int vmdk_write_extent(VmdkExtent *extent, int64_t cluster_offset, 1021 int64_t offset_in_cluster, const uint8_t *buf, 1022 int nb_sectors, int64_t sector_num) 1023 { 1024 int ret; 1025 VmdkGrainMarker *data = NULL; 1026 uLongf buf_len; 1027 const uint8_t *write_buf = buf; 1028 int write_len = nb_sectors * 512; 1029 1030 if (extent->compressed) { 1031 if (!extent->has_marker) { 1032 ret = -EINVAL; 1033 goto out; 1034 } 1035 buf_len = (extent->cluster_sectors << 9) * 2; 1036 data = g_malloc(buf_len + sizeof(VmdkGrainMarker)); 1037 if (compress(data->data, &buf_len, buf, nb_sectors << 9) != Z_OK || 1038 buf_len == 0) { 1039 ret = -EINVAL; 1040 goto out; 1041 } 1042 data->lba = sector_num; 1043 data->size = buf_len; 1044 write_buf = (uint8_t *)data; 1045 write_len = buf_len + sizeof(VmdkGrainMarker); 1046 } 1047 ret = bdrv_pwrite(extent->file, 1048 cluster_offset + offset_in_cluster, 1049 write_buf, 1050 write_len); 1051 if (ret != write_len) { 1052 ret = ret < 0 ? ret : -EIO; 1053 goto out; 1054 } 1055 ret = 0; 1056 out: 1057 g_free(data); 1058 return ret; 1059 } 1060 1061 static int vmdk_read_extent(VmdkExtent *extent, int64_t cluster_offset, 1062 int64_t offset_in_cluster, uint8_t *buf, 1063 int nb_sectors) 1064 { 1065 int ret; 1066 int cluster_bytes, buf_bytes; 1067 uint8_t *cluster_buf, *compressed_data; 1068 uint8_t *uncomp_buf; 1069 uint32_t data_len; 1070 VmdkGrainMarker *marker; 1071 uLongf buf_len; 1072 1073 1074 if (!extent->compressed) { 1075 ret = bdrv_pread(extent->file, 1076 cluster_offset + offset_in_cluster, 1077 buf, nb_sectors * 512); 1078 if (ret == nb_sectors * 512) { 1079 return 0; 1080 } else { 1081 return -EIO; 1082 } 1083 } 1084 cluster_bytes = extent->cluster_sectors * 512; 1085 /* Read two clusters in case GrainMarker + compressed data > one cluster */ 1086 buf_bytes = cluster_bytes * 2; 1087 cluster_buf = g_malloc(buf_bytes); 1088 uncomp_buf = g_malloc(cluster_bytes); 1089 ret = bdrv_pread(extent->file, 1090 cluster_offset, 1091 cluster_buf, buf_bytes); 1092 if (ret < 0) { 1093 goto out; 1094 } 1095 compressed_data = cluster_buf; 1096 buf_len = cluster_bytes; 1097 data_len = cluster_bytes; 1098 if (extent->has_marker) { 1099 marker = (VmdkGrainMarker *)cluster_buf; 1100 compressed_data = marker->data; 1101 data_len = le32_to_cpu(marker->size); 1102 } 1103 if (!data_len || data_len > buf_bytes) { 1104 ret = -EINVAL; 1105 goto out; 1106 } 1107 ret = uncompress(uncomp_buf, &buf_len, compressed_data, data_len); 1108 if (ret != Z_OK) { 1109 ret = -EINVAL; 1110 goto out; 1111 1112 } 1113 if (offset_in_cluster < 0 || 1114 offset_in_cluster + nb_sectors * 512 > buf_len) { 1115 ret = -EINVAL; 1116 goto out; 1117 } 1118 memcpy(buf, uncomp_buf + offset_in_cluster, nb_sectors * 512); 1119 ret = 0; 1120 1121 out: 1122 g_free(uncomp_buf); 1123 g_free(cluster_buf); 1124 return ret; 1125 } 1126 1127 static int vmdk_read(BlockDriverState *bs, int64_t sector_num, 1128 uint8_t *buf, int nb_sectors) 1129 { 1130 BDRVVmdkState *s = bs->opaque; 1131 int ret; 1132 uint64_t n, index_in_cluster; 1133 uint64_t extent_begin_sector, extent_relative_sector_num; 1134 VmdkExtent *extent = NULL; 1135 uint64_t cluster_offset; 1136 1137 while (nb_sectors > 0) { 1138 extent = find_extent(s, sector_num, extent); 1139 if (!extent) { 1140 return -EIO; 1141 } 1142 ret = get_cluster_offset( 1143 bs, extent, NULL, 1144 sector_num << 9, 0, &cluster_offset); 1145 extent_begin_sector = extent->end_sector - extent->sectors; 1146 extent_relative_sector_num = sector_num - extent_begin_sector; 1147 index_in_cluster = extent_relative_sector_num % extent->cluster_sectors; 1148 n = extent->cluster_sectors - index_in_cluster; 1149 if (n > nb_sectors) { 1150 n = nb_sectors; 1151 } 1152 if (ret != VMDK_OK) { 1153 /* if not allocated, try to read from parent image, if exist */ 1154 if (bs->backing_hd && ret != VMDK_ZEROED) { 1155 if (!vmdk_is_cid_valid(bs)) { 1156 return -EINVAL; 1157 } 1158 ret = bdrv_read(bs->backing_hd, sector_num, buf, n); 1159 if (ret < 0) { 1160 return ret; 1161 } 1162 } else { 1163 memset(buf, 0, 512 * n); 1164 } 1165 } else { 1166 ret = vmdk_read_extent(extent, 1167 cluster_offset, index_in_cluster * 512, 1168 buf, n); 1169 if (ret) { 1170 return ret; 1171 } 1172 } 1173 nb_sectors -= n; 1174 sector_num += n; 1175 buf += n * 512; 1176 } 1177 return 0; 1178 } 1179 1180 static coroutine_fn int vmdk_co_read(BlockDriverState *bs, int64_t sector_num, 1181 uint8_t *buf, int nb_sectors) 1182 { 1183 int ret; 1184 BDRVVmdkState *s = bs->opaque; 1185 qemu_co_mutex_lock(&s->lock); 1186 ret = vmdk_read(bs, sector_num, buf, nb_sectors); 1187 qemu_co_mutex_unlock(&s->lock); 1188 return ret; 1189 } 1190 1191 /** 1192 * vmdk_write: 1193 * @zeroed: buf is ignored (data is zero), use zeroed_grain GTE feature 1194 * if possible, otherwise return -ENOTSUP. 1195 * @zero_dry_run: used for zeroed == true only, don't update L2 table, just 1196 * 1197 * Returns: error code with 0 for success. 1198 */ 1199 static int vmdk_write(BlockDriverState *bs, int64_t sector_num, 1200 const uint8_t *buf, int nb_sectors, 1201 bool zeroed, bool zero_dry_run) 1202 { 1203 BDRVVmdkState *s = bs->opaque; 1204 VmdkExtent *extent = NULL; 1205 int n, ret; 1206 int64_t index_in_cluster; 1207 uint64_t extent_begin_sector, extent_relative_sector_num; 1208 uint64_t cluster_offset; 1209 VmdkMetaData m_data; 1210 1211 if (sector_num > bs->total_sectors) { 1212 fprintf(stderr, 1213 "(VMDK) Wrong offset: sector_num=0x%" PRIx64 1214 " total_sectors=0x%" PRIx64 "\n", 1215 sector_num, bs->total_sectors); 1216 return -EIO; 1217 } 1218 1219 while (nb_sectors > 0) { 1220 extent = find_extent(s, sector_num, extent); 1221 if (!extent) { 1222 return -EIO; 1223 } 1224 ret = get_cluster_offset( 1225 bs, 1226 extent, 1227 &m_data, 1228 sector_num << 9, !extent->compressed, 1229 &cluster_offset); 1230 if (extent->compressed) { 1231 if (ret == VMDK_OK) { 1232 /* Refuse write to allocated cluster for streamOptimized */ 1233 fprintf(stderr, 1234 "VMDK: can't write to allocated cluster" 1235 " for streamOptimized\n"); 1236 return -EIO; 1237 } else { 1238 /* allocate */ 1239 ret = get_cluster_offset( 1240 bs, 1241 extent, 1242 &m_data, 1243 sector_num << 9, 1, 1244 &cluster_offset); 1245 } 1246 } 1247 if (ret == VMDK_ERROR) { 1248 return -EINVAL; 1249 } 1250 extent_begin_sector = extent->end_sector - extent->sectors; 1251 extent_relative_sector_num = sector_num - extent_begin_sector; 1252 index_in_cluster = extent_relative_sector_num % extent->cluster_sectors; 1253 n = extent->cluster_sectors - index_in_cluster; 1254 if (n > nb_sectors) { 1255 n = nb_sectors; 1256 } 1257 if (zeroed) { 1258 /* Do zeroed write, buf is ignored */ 1259 if (extent->has_zero_grain && 1260 index_in_cluster == 0 && 1261 n >= extent->cluster_sectors) { 1262 n = extent->cluster_sectors; 1263 if (!zero_dry_run) { 1264 m_data.offset = VMDK_GTE_ZEROED; 1265 /* update L2 tables */ 1266 if (vmdk_L2update(extent, &m_data) != VMDK_OK) { 1267 return -EIO; 1268 } 1269 } 1270 } else { 1271 return -ENOTSUP; 1272 } 1273 } else { 1274 ret = vmdk_write_extent(extent, 1275 cluster_offset, index_in_cluster * 512, 1276 buf, n, sector_num); 1277 if (ret) { 1278 return ret; 1279 } 1280 if (m_data.valid) { 1281 /* update L2 tables */ 1282 if (vmdk_L2update(extent, &m_data) != VMDK_OK) { 1283 return -EIO; 1284 } 1285 } 1286 } 1287 nb_sectors -= n; 1288 sector_num += n; 1289 buf += n * 512; 1290 1291 /* update CID on the first write every time the virtual disk is 1292 * opened */ 1293 if (!s->cid_updated) { 1294 ret = vmdk_write_cid(bs, time(NULL)); 1295 if (ret < 0) { 1296 return ret; 1297 } 1298 s->cid_updated = true; 1299 } 1300 } 1301 return 0; 1302 } 1303 1304 static coroutine_fn int vmdk_co_write(BlockDriverState *bs, int64_t sector_num, 1305 const uint8_t *buf, int nb_sectors) 1306 { 1307 int ret; 1308 BDRVVmdkState *s = bs->opaque; 1309 qemu_co_mutex_lock(&s->lock); 1310 ret = vmdk_write(bs, sector_num, buf, nb_sectors, false, false); 1311 qemu_co_mutex_unlock(&s->lock); 1312 return ret; 1313 } 1314 1315 static int coroutine_fn vmdk_co_write_zeroes(BlockDriverState *bs, 1316 int64_t sector_num, 1317 int nb_sectors) 1318 { 1319 int ret; 1320 BDRVVmdkState *s = bs->opaque; 1321 qemu_co_mutex_lock(&s->lock); 1322 ret = vmdk_write(bs, sector_num, NULL, nb_sectors, true, true); 1323 if (!ret) { 1324 ret = vmdk_write(bs, sector_num, NULL, nb_sectors, true, false); 1325 } 1326 qemu_co_mutex_unlock(&s->lock); 1327 return ret; 1328 } 1329 1330 1331 static int vmdk_create_extent(const char *filename, int64_t filesize, 1332 bool flat, bool compress, bool zeroed_grain) 1333 { 1334 int ret, i; 1335 int fd = 0; 1336 VMDK4Header header; 1337 uint32_t tmp, magic, grains, gd_size, gt_size, gt_count; 1338 1339 fd = qemu_open(filename, 1340 O_WRONLY | O_CREAT | O_TRUNC | O_BINARY | O_LARGEFILE, 1341 0644); 1342 if (fd < 0) { 1343 return -errno; 1344 } 1345 if (flat) { 1346 ret = ftruncate(fd, filesize); 1347 if (ret < 0) { 1348 ret = -errno; 1349 } 1350 goto exit; 1351 } 1352 magic = cpu_to_be32(VMDK4_MAGIC); 1353 memset(&header, 0, sizeof(header)); 1354 header.version = zeroed_grain ? 2 : 1; 1355 header.flags = VMDK4_FLAG_RGD | VMDK4_FLAG_NL_DETECT 1356 | (compress ? VMDK4_FLAG_COMPRESS | VMDK4_FLAG_MARKER : 0) 1357 | (zeroed_grain ? VMDK4_FLAG_ZERO_GRAIN : 0); 1358 header.compressAlgorithm = compress ? VMDK4_COMPRESSION_DEFLATE : 0; 1359 header.capacity = filesize / 512; 1360 header.granularity = 128; 1361 header.num_gtes_per_gte = 512; 1362 1363 grains = (filesize / 512 + header.granularity - 1) / header.granularity; 1364 gt_size = ((header.num_gtes_per_gte * sizeof(uint32_t)) + 511) >> 9; 1365 gt_count = 1366 (grains + header.num_gtes_per_gte - 1) / header.num_gtes_per_gte; 1367 gd_size = (gt_count * sizeof(uint32_t) + 511) >> 9; 1368 1369 header.desc_offset = 1; 1370 header.desc_size = 20; 1371 header.rgd_offset = header.desc_offset + header.desc_size; 1372 header.gd_offset = header.rgd_offset + gd_size + (gt_size * gt_count); 1373 header.grain_offset = 1374 ((header.gd_offset + gd_size + (gt_size * gt_count) + 1375 header.granularity - 1) / header.granularity) * 1376 header.granularity; 1377 /* swap endianness for all header fields */ 1378 header.version = cpu_to_le32(header.version); 1379 header.flags = cpu_to_le32(header.flags); 1380 header.capacity = cpu_to_le64(header.capacity); 1381 header.granularity = cpu_to_le64(header.granularity); 1382 header.num_gtes_per_gte = cpu_to_le32(header.num_gtes_per_gte); 1383 header.desc_offset = cpu_to_le64(header.desc_offset); 1384 header.desc_size = cpu_to_le64(header.desc_size); 1385 header.rgd_offset = cpu_to_le64(header.rgd_offset); 1386 header.gd_offset = cpu_to_le64(header.gd_offset); 1387 header.grain_offset = cpu_to_le64(header.grain_offset); 1388 header.compressAlgorithm = cpu_to_le16(header.compressAlgorithm); 1389 1390 header.check_bytes[0] = 0xa; 1391 header.check_bytes[1] = 0x20; 1392 header.check_bytes[2] = 0xd; 1393 header.check_bytes[3] = 0xa; 1394 1395 /* write all the data */ 1396 ret = qemu_write_full(fd, &magic, sizeof(magic)); 1397 if (ret != sizeof(magic)) { 1398 ret = -errno; 1399 goto exit; 1400 } 1401 ret = qemu_write_full(fd, &header, sizeof(header)); 1402 if (ret != sizeof(header)) { 1403 ret = -errno; 1404 goto exit; 1405 } 1406 1407 ret = ftruncate(fd, le64_to_cpu(header.grain_offset) << 9); 1408 if (ret < 0) { 1409 ret = -errno; 1410 goto exit; 1411 } 1412 1413 /* write grain directory */ 1414 lseek(fd, le64_to_cpu(header.rgd_offset) << 9, SEEK_SET); 1415 for (i = 0, tmp = le64_to_cpu(header.rgd_offset) + gd_size; 1416 i < gt_count; i++, tmp += gt_size) { 1417 ret = qemu_write_full(fd, &tmp, sizeof(tmp)); 1418 if (ret != sizeof(tmp)) { 1419 ret = -errno; 1420 goto exit; 1421 } 1422 } 1423 1424 /* write backup grain directory */ 1425 lseek(fd, le64_to_cpu(header.gd_offset) << 9, SEEK_SET); 1426 for (i = 0, tmp = le64_to_cpu(header.gd_offset) + gd_size; 1427 i < gt_count; i++, tmp += gt_size) { 1428 ret = qemu_write_full(fd, &tmp, sizeof(tmp)); 1429 if (ret != sizeof(tmp)) { 1430 ret = -errno; 1431 goto exit; 1432 } 1433 } 1434 1435 ret = 0; 1436 exit: 1437 qemu_close(fd); 1438 return ret; 1439 } 1440 1441 static int filename_decompose(const char *filename, char *path, char *prefix, 1442 char *postfix, size_t buf_len) 1443 { 1444 const char *p, *q; 1445 1446 if (filename == NULL || !strlen(filename)) { 1447 fprintf(stderr, "Vmdk: no filename provided.\n"); 1448 return VMDK_ERROR; 1449 } 1450 p = strrchr(filename, '/'); 1451 if (p == NULL) { 1452 p = strrchr(filename, '\\'); 1453 } 1454 if (p == NULL) { 1455 p = strrchr(filename, ':'); 1456 } 1457 if (p != NULL) { 1458 p++; 1459 if (p - filename >= buf_len) { 1460 return VMDK_ERROR; 1461 } 1462 pstrcpy(path, p - filename + 1, filename); 1463 } else { 1464 p = filename; 1465 path[0] = '\0'; 1466 } 1467 q = strrchr(p, '.'); 1468 if (q == NULL) { 1469 pstrcpy(prefix, buf_len, p); 1470 postfix[0] = '\0'; 1471 } else { 1472 if (q - p >= buf_len) { 1473 return VMDK_ERROR; 1474 } 1475 pstrcpy(prefix, q - p + 1, p); 1476 pstrcpy(postfix, buf_len, q); 1477 } 1478 return VMDK_OK; 1479 } 1480 1481 static int relative_path(char *dest, int dest_size, 1482 const char *base, const char *target) 1483 { 1484 int i = 0; 1485 int n = 0; 1486 const char *p, *q; 1487 #ifdef _WIN32 1488 const char *sep = "\\"; 1489 #else 1490 const char *sep = "/"; 1491 #endif 1492 1493 if (!(dest && base && target)) { 1494 return VMDK_ERROR; 1495 } 1496 if (path_is_absolute(target)) { 1497 pstrcpy(dest, dest_size, target); 1498 return VMDK_OK; 1499 } 1500 while (base[i] == target[i]) { 1501 i++; 1502 } 1503 p = &base[i]; 1504 q = &target[i]; 1505 while (*p) { 1506 if (*p == *sep) { 1507 n++; 1508 } 1509 p++; 1510 } 1511 dest[0] = '\0'; 1512 for (; n; n--) { 1513 pstrcat(dest, dest_size, ".."); 1514 pstrcat(dest, dest_size, sep); 1515 } 1516 pstrcat(dest, dest_size, q); 1517 return VMDK_OK; 1518 } 1519 1520 static int vmdk_create(const char *filename, QEMUOptionParameter *options) 1521 { 1522 int fd, idx = 0; 1523 char desc[BUF_SIZE]; 1524 int64_t total_size = 0, filesize; 1525 const char *adapter_type = NULL; 1526 const char *backing_file = NULL; 1527 const char *fmt = NULL; 1528 int flags = 0; 1529 int ret = 0; 1530 bool flat, split, compress; 1531 char ext_desc_lines[BUF_SIZE] = ""; 1532 char path[PATH_MAX], prefix[PATH_MAX], postfix[PATH_MAX]; 1533 const int64_t split_size = 0x80000000; /* VMDK has constant split size */ 1534 const char *desc_extent_line; 1535 char parent_desc_line[BUF_SIZE] = ""; 1536 uint32_t parent_cid = 0xffffffff; 1537 uint32_t number_heads = 16; 1538 bool zeroed_grain = false; 1539 const char desc_template[] = 1540 "# Disk DescriptorFile\n" 1541 "version=1\n" 1542 "CID=%x\n" 1543 "parentCID=%x\n" 1544 "createType=\"%s\"\n" 1545 "%s" 1546 "\n" 1547 "# Extent description\n" 1548 "%s" 1549 "\n" 1550 "# The Disk Data Base\n" 1551 "#DDB\n" 1552 "\n" 1553 "ddb.virtualHWVersion = \"%d\"\n" 1554 "ddb.geometry.cylinders = \"%" PRId64 "\"\n" 1555 "ddb.geometry.heads = \"%d\"\n" 1556 "ddb.geometry.sectors = \"63\"\n" 1557 "ddb.adapterType = \"%s\"\n"; 1558 1559 if (filename_decompose(filename, path, prefix, postfix, PATH_MAX)) { 1560 return -EINVAL; 1561 } 1562 /* Read out options */ 1563 while (options && options->name) { 1564 if (!strcmp(options->name, BLOCK_OPT_SIZE)) { 1565 total_size = options->value.n; 1566 } else if (!strcmp(options->name, BLOCK_OPT_ADAPTER_TYPE)) { 1567 adapter_type = options->value.s; 1568 } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) { 1569 backing_file = options->value.s; 1570 } else if (!strcmp(options->name, BLOCK_OPT_COMPAT6)) { 1571 flags |= options->value.n ? BLOCK_FLAG_COMPAT6 : 0; 1572 } else if (!strcmp(options->name, BLOCK_OPT_SUBFMT)) { 1573 fmt = options->value.s; 1574 } else if (!strcmp(options->name, BLOCK_OPT_ZEROED_GRAIN)) { 1575 zeroed_grain |= options->value.n; 1576 } 1577 options++; 1578 } 1579 if (!adapter_type) { 1580 adapter_type = "ide"; 1581 } else if (strcmp(adapter_type, "ide") && 1582 strcmp(adapter_type, "buslogic") && 1583 strcmp(adapter_type, "lsilogic") && 1584 strcmp(adapter_type, "legacyESX")) { 1585 fprintf(stderr, "VMDK: Unknown adapter type: '%s'.\n", adapter_type); 1586 return -EINVAL; 1587 } 1588 if (strcmp(adapter_type, "ide") != 0) { 1589 /* that's the number of heads with which vmware operates when 1590 creating, exporting, etc. vmdk files with a non-ide adapter type */ 1591 number_heads = 255; 1592 } 1593 if (!fmt) { 1594 /* Default format to monolithicSparse */ 1595 fmt = "monolithicSparse"; 1596 } else if (strcmp(fmt, "monolithicFlat") && 1597 strcmp(fmt, "monolithicSparse") && 1598 strcmp(fmt, "twoGbMaxExtentSparse") && 1599 strcmp(fmt, "twoGbMaxExtentFlat") && 1600 strcmp(fmt, "streamOptimized")) { 1601 fprintf(stderr, "VMDK: Unknown subformat: %s\n", fmt); 1602 return -EINVAL; 1603 } 1604 split = !(strcmp(fmt, "twoGbMaxExtentFlat") && 1605 strcmp(fmt, "twoGbMaxExtentSparse")); 1606 flat = !(strcmp(fmt, "monolithicFlat") && 1607 strcmp(fmt, "twoGbMaxExtentFlat")); 1608 compress = !strcmp(fmt, "streamOptimized"); 1609 if (flat) { 1610 desc_extent_line = "RW %lld FLAT \"%s\" 0\n"; 1611 } else { 1612 desc_extent_line = "RW %lld SPARSE \"%s\"\n"; 1613 } 1614 if (flat && backing_file) { 1615 /* not supporting backing file for flat image */ 1616 return -ENOTSUP; 1617 } 1618 if (backing_file) { 1619 char parent_filename[PATH_MAX]; 1620 BlockDriverState *bs = bdrv_new(""); 1621 ret = bdrv_open(bs, backing_file, NULL, 0, NULL); 1622 if (ret != 0) { 1623 bdrv_delete(bs); 1624 return ret; 1625 } 1626 if (strcmp(bs->drv->format_name, "vmdk")) { 1627 bdrv_delete(bs); 1628 return -EINVAL; 1629 } 1630 parent_cid = vmdk_read_cid(bs, 0); 1631 bdrv_delete(bs); 1632 relative_path(parent_filename, sizeof(parent_filename), 1633 filename, backing_file); 1634 snprintf(parent_desc_line, sizeof(parent_desc_line), 1635 "parentFileNameHint=\"%s\"", parent_filename); 1636 } 1637 1638 /* Create extents */ 1639 filesize = total_size; 1640 while (filesize > 0) { 1641 char desc_line[BUF_SIZE]; 1642 char ext_filename[PATH_MAX]; 1643 char desc_filename[PATH_MAX]; 1644 int64_t size = filesize; 1645 1646 if (split && size > split_size) { 1647 size = split_size; 1648 } 1649 if (split) { 1650 snprintf(desc_filename, sizeof(desc_filename), "%s-%c%03d%s", 1651 prefix, flat ? 'f' : 's', ++idx, postfix); 1652 } else if (flat) { 1653 snprintf(desc_filename, sizeof(desc_filename), "%s-flat%s", 1654 prefix, postfix); 1655 } else { 1656 snprintf(desc_filename, sizeof(desc_filename), "%s%s", 1657 prefix, postfix); 1658 } 1659 snprintf(ext_filename, sizeof(ext_filename), "%s%s", 1660 path, desc_filename); 1661 1662 if (vmdk_create_extent(ext_filename, size, 1663 flat, compress, zeroed_grain)) { 1664 return -EINVAL; 1665 } 1666 filesize -= size; 1667 1668 /* Format description line */ 1669 snprintf(desc_line, sizeof(desc_line), 1670 desc_extent_line, size / 512, desc_filename); 1671 pstrcat(ext_desc_lines, sizeof(ext_desc_lines), desc_line); 1672 } 1673 /* generate descriptor file */ 1674 snprintf(desc, sizeof(desc), desc_template, 1675 (unsigned int)time(NULL), 1676 parent_cid, 1677 fmt, 1678 parent_desc_line, 1679 ext_desc_lines, 1680 (flags & BLOCK_FLAG_COMPAT6 ? 6 : 4), 1681 total_size / (int64_t)(63 * number_heads * 512), number_heads, 1682 adapter_type); 1683 if (split || flat) { 1684 fd = qemu_open(filename, 1685 O_WRONLY | O_CREAT | O_TRUNC | O_BINARY | O_LARGEFILE, 1686 0644); 1687 } else { 1688 fd = qemu_open(filename, 1689 O_WRONLY | O_BINARY | O_LARGEFILE, 1690 0644); 1691 } 1692 if (fd < 0) { 1693 return -errno; 1694 } 1695 /* the descriptor offset = 0x200 */ 1696 if (!split && !flat && 0x200 != lseek(fd, 0x200, SEEK_SET)) { 1697 ret = -errno; 1698 goto exit; 1699 } 1700 ret = qemu_write_full(fd, desc, strlen(desc)); 1701 if (ret != strlen(desc)) { 1702 ret = -errno; 1703 goto exit; 1704 } 1705 ret = 0; 1706 exit: 1707 qemu_close(fd); 1708 return ret; 1709 } 1710 1711 static void vmdk_close(BlockDriverState *bs) 1712 { 1713 BDRVVmdkState *s = bs->opaque; 1714 1715 vmdk_free_extents(bs); 1716 1717 migrate_del_blocker(s->migration_blocker); 1718 error_free(s->migration_blocker); 1719 } 1720 1721 static coroutine_fn int vmdk_co_flush(BlockDriverState *bs) 1722 { 1723 BDRVVmdkState *s = bs->opaque; 1724 int i, err; 1725 int ret = 0; 1726 1727 for (i = 0; i < s->num_extents; i++) { 1728 err = bdrv_co_flush(s->extents[i].file); 1729 if (err < 0) { 1730 ret = err; 1731 } 1732 } 1733 return ret; 1734 } 1735 1736 static int64_t vmdk_get_allocated_file_size(BlockDriverState *bs) 1737 { 1738 int i; 1739 int64_t ret = 0; 1740 int64_t r; 1741 BDRVVmdkState *s = bs->opaque; 1742 1743 ret = bdrv_get_allocated_file_size(bs->file); 1744 if (ret < 0) { 1745 return ret; 1746 } 1747 for (i = 0; i < s->num_extents; i++) { 1748 if (s->extents[i].file == bs->file) { 1749 continue; 1750 } 1751 r = bdrv_get_allocated_file_size(s->extents[i].file); 1752 if (r < 0) { 1753 return r; 1754 } 1755 ret += r; 1756 } 1757 return ret; 1758 } 1759 1760 static QEMUOptionParameter vmdk_create_options[] = { 1761 { 1762 .name = BLOCK_OPT_SIZE, 1763 .type = OPT_SIZE, 1764 .help = "Virtual disk size" 1765 }, 1766 { 1767 .name = BLOCK_OPT_ADAPTER_TYPE, 1768 .type = OPT_STRING, 1769 .help = "Virtual adapter type, can be one of " 1770 "ide (default), lsilogic, buslogic or legacyESX" 1771 }, 1772 { 1773 .name = BLOCK_OPT_BACKING_FILE, 1774 .type = OPT_STRING, 1775 .help = "File name of a base image" 1776 }, 1777 { 1778 .name = BLOCK_OPT_COMPAT6, 1779 .type = OPT_FLAG, 1780 .help = "VMDK version 6 image" 1781 }, 1782 { 1783 .name = BLOCK_OPT_SUBFMT, 1784 .type = OPT_STRING, 1785 .help = 1786 "VMDK flat extent format, can be one of " 1787 "{monolithicSparse (default) | monolithicFlat | twoGbMaxExtentSparse | twoGbMaxExtentFlat | streamOptimized} " 1788 }, 1789 { 1790 .name = BLOCK_OPT_ZEROED_GRAIN, 1791 .type = OPT_FLAG, 1792 .help = "Enable efficient zero writes using the zeroed-grain GTE feature" 1793 }, 1794 { NULL } 1795 }; 1796 1797 static BlockDriver bdrv_vmdk = { 1798 .format_name = "vmdk", 1799 .instance_size = sizeof(BDRVVmdkState), 1800 .bdrv_probe = vmdk_probe, 1801 .bdrv_open = vmdk_open, 1802 .bdrv_reopen_prepare = vmdk_reopen_prepare, 1803 .bdrv_read = vmdk_co_read, 1804 .bdrv_write = vmdk_co_write, 1805 .bdrv_co_write_zeroes = vmdk_co_write_zeroes, 1806 .bdrv_close = vmdk_close, 1807 .bdrv_create = vmdk_create, 1808 .bdrv_co_flush_to_disk = vmdk_co_flush, 1809 .bdrv_co_is_allocated = vmdk_co_is_allocated, 1810 .bdrv_get_allocated_file_size = vmdk_get_allocated_file_size, 1811 1812 .create_options = vmdk_create_options, 1813 }; 1814 1815 static void bdrv_vmdk_init(void) 1816 { 1817 bdrv_register(&bdrv_vmdk); 1818 } 1819 1820 block_init(bdrv_vmdk_init); 1821