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 int64_t 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 = flat ? 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 Error *local_err = NULL; 701 702 while (*p) { 703 /* parse extent line: 704 * RW [size in sectors] FLAT "file-name.vmdk" OFFSET 705 * or 706 * RW [size in sectors] SPARSE "file-name.vmdk" 707 */ 708 flat_offset = -1; 709 ret = sscanf(p, "%10s %" SCNd64 " %10s \"%511[^\n\r\"]\" %" SCNd64, 710 access, §ors, type, fname, &flat_offset); 711 if (ret < 4 || strcmp(access, "RW")) { 712 goto next_line; 713 } else if (!strcmp(type, "FLAT")) { 714 if (ret != 5 || flat_offset < 0) { 715 return -EINVAL; 716 } 717 } else if (ret != 4) { 718 return -EINVAL; 719 } 720 721 if (sectors <= 0 || 722 (strcmp(type, "FLAT") && strcmp(type, "SPARSE") && 723 strcmp(type, "VMFS") && strcmp(type, "VMFSSPARSE")) || 724 (strcmp(access, "RW"))) { 725 goto next_line; 726 } 727 728 path_combine(extent_path, sizeof(extent_path), 729 desc_file_path, fname); 730 ret = bdrv_file_open(&extent_file, extent_path, NULL, bs->open_flags, 731 &local_err); 732 if (ret) { 733 qerror_report_err(local_err); 734 error_free(local_err); 735 return ret; 736 } 737 738 /* save to extents array */ 739 if (!strcmp(type, "FLAT") || !strcmp(type, "VMFS")) { 740 /* FLAT extent */ 741 VmdkExtent *extent; 742 743 ret = vmdk_add_extent(bs, extent_file, true, sectors, 744 0, 0, 0, 0, 0, &extent); 745 if (ret < 0) { 746 return ret; 747 } 748 extent->flat_start_offset = flat_offset << 9; 749 } else if (!strcmp(type, "SPARSE") || !strcmp(type, "VMFSSPARSE")) { 750 /* SPARSE extent and VMFSSPARSE extent are both "COWD" sparse file*/ 751 ret = vmdk_open_sparse(bs, extent_file, bs->open_flags); 752 if (ret) { 753 bdrv_unref(extent_file); 754 return ret; 755 } 756 } else { 757 fprintf(stderr, 758 "VMDK: Not supported extent type \"%s\""".\n", type); 759 return -ENOTSUP; 760 } 761 next_line: 762 /* move to next line */ 763 while (*p && *p != '\n') { 764 p++; 765 } 766 p++; 767 } 768 return 0; 769 } 770 771 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, 772 uint64_t desc_offset) 773 { 774 int ret; 775 char *buf = NULL; 776 char ct[128]; 777 BDRVVmdkState *s = bs->opaque; 778 int64_t size; 779 780 size = bdrv_getlength(bs->file); 781 if (size < 0) { 782 return -EINVAL; 783 } 784 785 size = MIN(size, 1 << 20); /* avoid unbounded allocation */ 786 buf = g_malloc0(size + 1); 787 788 ret = bdrv_pread(bs->file, desc_offset, buf, size); 789 if (ret < 0) { 790 goto exit; 791 } 792 if (vmdk_parse_description(buf, "createType", ct, sizeof(ct))) { 793 ret = -EMEDIUMTYPE; 794 goto exit; 795 } 796 if (strcmp(ct, "monolithicFlat") && 797 strcmp(ct, "vmfs") && 798 strcmp(ct, "vmfsSparse") && 799 strcmp(ct, "twoGbMaxExtentSparse") && 800 strcmp(ct, "twoGbMaxExtentFlat")) { 801 fprintf(stderr, 802 "VMDK: Not supported image type \"%s\""".\n", ct); 803 ret = -ENOTSUP; 804 goto exit; 805 } 806 s->desc_offset = 0; 807 ret = vmdk_parse_extents(buf, bs, bs->file->filename); 808 exit: 809 g_free(buf); 810 return ret; 811 } 812 813 static int vmdk_open(BlockDriverState *bs, QDict *options, int flags, 814 Error **errp) 815 { 816 int ret; 817 BDRVVmdkState *s = bs->opaque; 818 819 if (vmdk_open_sparse(bs, bs->file, flags) == 0) { 820 s->desc_offset = 0x200; 821 } else { 822 ret = vmdk_open_desc_file(bs, flags, 0); 823 if (ret) { 824 goto fail; 825 } 826 } 827 /* try to open parent images, if exist */ 828 ret = vmdk_parent_open(bs); 829 if (ret) { 830 goto fail; 831 } 832 s->parent_cid = vmdk_read_cid(bs, 1); 833 qemu_co_mutex_init(&s->lock); 834 835 /* Disable migration when VMDK images are used */ 836 error_set(&s->migration_blocker, 837 QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED, 838 "vmdk", bs->device_name, "live migration"); 839 migrate_add_blocker(s->migration_blocker); 840 841 return 0; 842 843 fail: 844 vmdk_free_extents(bs); 845 return ret; 846 } 847 848 static int get_whole_cluster(BlockDriverState *bs, 849 VmdkExtent *extent, 850 uint64_t cluster_offset, 851 uint64_t offset, 852 bool allocate) 853 { 854 int ret = VMDK_OK; 855 uint8_t *whole_grain = NULL; 856 857 /* we will be here if it's first write on non-exist grain(cluster). 858 * try to read from parent image, if exist */ 859 if (bs->backing_hd) { 860 whole_grain = 861 qemu_blockalign(bs, extent->cluster_sectors << BDRV_SECTOR_BITS); 862 if (!vmdk_is_cid_valid(bs)) { 863 ret = VMDK_ERROR; 864 goto exit; 865 } 866 867 /* floor offset to cluster */ 868 offset -= offset % (extent->cluster_sectors * 512); 869 ret = bdrv_read(bs->backing_hd, offset >> 9, whole_grain, 870 extent->cluster_sectors); 871 if (ret < 0) { 872 ret = VMDK_ERROR; 873 goto exit; 874 } 875 876 /* Write grain only into the active image */ 877 ret = bdrv_write(extent->file, cluster_offset, whole_grain, 878 extent->cluster_sectors); 879 if (ret < 0) { 880 ret = VMDK_ERROR; 881 goto exit; 882 } 883 } 884 exit: 885 qemu_vfree(whole_grain); 886 return ret; 887 } 888 889 static int vmdk_L2update(VmdkExtent *extent, VmdkMetaData *m_data) 890 { 891 uint32_t offset; 892 QEMU_BUILD_BUG_ON(sizeof(offset) != sizeof(m_data->offset)); 893 offset = cpu_to_le32(m_data->offset); 894 /* update L2 table */ 895 if (bdrv_pwrite_sync( 896 extent->file, 897 ((int64_t)m_data->l2_offset * 512) 898 + (m_data->l2_index * sizeof(m_data->offset)), 899 &offset, sizeof(offset)) < 0) { 900 return VMDK_ERROR; 901 } 902 /* update backup L2 table */ 903 if (extent->l1_backup_table_offset != 0) { 904 m_data->l2_offset = extent->l1_backup_table[m_data->l1_index]; 905 if (bdrv_pwrite_sync( 906 extent->file, 907 ((int64_t)m_data->l2_offset * 512) 908 + (m_data->l2_index * sizeof(m_data->offset)), 909 &offset, sizeof(offset)) < 0) { 910 return VMDK_ERROR; 911 } 912 } 913 if (m_data->l2_cache_entry) { 914 *m_data->l2_cache_entry = offset; 915 } 916 917 return VMDK_OK; 918 } 919 920 static int get_cluster_offset(BlockDriverState *bs, 921 VmdkExtent *extent, 922 VmdkMetaData *m_data, 923 uint64_t offset, 924 int allocate, 925 uint64_t *cluster_offset) 926 { 927 unsigned int l1_index, l2_offset, l2_index; 928 int min_index, i, j; 929 uint32_t min_count, *l2_table; 930 bool zeroed = false; 931 932 if (m_data) { 933 m_data->valid = 0; 934 } 935 if (extent->flat) { 936 *cluster_offset = extent->flat_start_offset; 937 return VMDK_OK; 938 } 939 940 offset -= (extent->end_sector - extent->sectors) * SECTOR_SIZE; 941 l1_index = (offset >> 9) / extent->l1_entry_sectors; 942 if (l1_index >= extent->l1_size) { 943 return VMDK_ERROR; 944 } 945 l2_offset = extent->l1_table[l1_index]; 946 if (!l2_offset) { 947 return VMDK_UNALLOC; 948 } 949 for (i = 0; i < L2_CACHE_SIZE; i++) { 950 if (l2_offset == extent->l2_cache_offsets[i]) { 951 /* increment the hit count */ 952 if (++extent->l2_cache_counts[i] == 0xffffffff) { 953 for (j = 0; j < L2_CACHE_SIZE; j++) { 954 extent->l2_cache_counts[j] >>= 1; 955 } 956 } 957 l2_table = extent->l2_cache + (i * extent->l2_size); 958 goto found; 959 } 960 } 961 /* not found: load a new entry in the least used one */ 962 min_index = 0; 963 min_count = 0xffffffff; 964 for (i = 0; i < L2_CACHE_SIZE; i++) { 965 if (extent->l2_cache_counts[i] < min_count) { 966 min_count = extent->l2_cache_counts[i]; 967 min_index = i; 968 } 969 } 970 l2_table = extent->l2_cache + (min_index * extent->l2_size); 971 if (bdrv_pread( 972 extent->file, 973 (int64_t)l2_offset * 512, 974 l2_table, 975 extent->l2_size * sizeof(uint32_t) 976 ) != extent->l2_size * sizeof(uint32_t)) { 977 return VMDK_ERROR; 978 } 979 980 extent->l2_cache_offsets[min_index] = l2_offset; 981 extent->l2_cache_counts[min_index] = 1; 982 found: 983 l2_index = ((offset >> 9) / extent->cluster_sectors) % extent->l2_size; 984 *cluster_offset = le32_to_cpu(l2_table[l2_index]); 985 986 if (m_data) { 987 m_data->valid = 1; 988 m_data->l1_index = l1_index; 989 m_data->l2_index = l2_index; 990 m_data->offset = *cluster_offset; 991 m_data->l2_offset = l2_offset; 992 m_data->l2_cache_entry = &l2_table[l2_index]; 993 } 994 if (extent->has_zero_grain && *cluster_offset == VMDK_GTE_ZEROED) { 995 zeroed = true; 996 } 997 998 if (!*cluster_offset || zeroed) { 999 if (!allocate) { 1000 return zeroed ? VMDK_ZEROED : VMDK_UNALLOC; 1001 } 1002 1003 /* Avoid the L2 tables update for the images that have snapshots. */ 1004 *cluster_offset = bdrv_getlength(extent->file); 1005 if (!extent->compressed) { 1006 bdrv_truncate( 1007 extent->file, 1008 *cluster_offset + (extent->cluster_sectors << 9) 1009 ); 1010 } 1011 1012 *cluster_offset >>= 9; 1013 l2_table[l2_index] = cpu_to_le32(*cluster_offset); 1014 1015 /* First of all we write grain itself, to avoid race condition 1016 * that may to corrupt the image. 1017 * This problem may occur because of insufficient space on host disk 1018 * or inappropriate VM shutdown. 1019 */ 1020 if (get_whole_cluster( 1021 bs, extent, *cluster_offset, offset, allocate) == -1) { 1022 return VMDK_ERROR; 1023 } 1024 1025 if (m_data) { 1026 m_data->offset = *cluster_offset; 1027 } 1028 } 1029 *cluster_offset <<= 9; 1030 return VMDK_OK; 1031 } 1032 1033 static VmdkExtent *find_extent(BDRVVmdkState *s, 1034 int64_t sector_num, VmdkExtent *start_hint) 1035 { 1036 VmdkExtent *extent = start_hint; 1037 1038 if (!extent) { 1039 extent = &s->extents[0]; 1040 } 1041 while (extent < &s->extents[s->num_extents]) { 1042 if (sector_num < extent->end_sector) { 1043 return extent; 1044 } 1045 extent++; 1046 } 1047 return NULL; 1048 } 1049 1050 static int64_t coroutine_fn vmdk_co_get_block_status(BlockDriverState *bs, 1051 int64_t sector_num, int nb_sectors, int *pnum) 1052 { 1053 BDRVVmdkState *s = bs->opaque; 1054 int64_t index_in_cluster, n, ret; 1055 uint64_t offset; 1056 VmdkExtent *extent; 1057 1058 extent = find_extent(s, sector_num, NULL); 1059 if (!extent) { 1060 return 0; 1061 } 1062 qemu_co_mutex_lock(&s->lock); 1063 ret = get_cluster_offset(bs, extent, NULL, 1064 sector_num * 512, 0, &offset); 1065 qemu_co_mutex_unlock(&s->lock); 1066 1067 switch (ret) { 1068 case VMDK_ERROR: 1069 ret = -EIO; 1070 break; 1071 case VMDK_UNALLOC: 1072 ret = 0; 1073 break; 1074 case VMDK_ZEROED: 1075 ret = BDRV_BLOCK_ZERO; 1076 break; 1077 case VMDK_OK: 1078 ret = BDRV_BLOCK_DATA; 1079 if (extent->file == bs->file) { 1080 ret |= BDRV_BLOCK_OFFSET_VALID | offset; 1081 } 1082 1083 break; 1084 } 1085 1086 index_in_cluster = sector_num % extent->cluster_sectors; 1087 n = extent->cluster_sectors - index_in_cluster; 1088 if (n > nb_sectors) { 1089 n = nb_sectors; 1090 } 1091 *pnum = n; 1092 return ret; 1093 } 1094 1095 static int vmdk_write_extent(VmdkExtent *extent, int64_t cluster_offset, 1096 int64_t offset_in_cluster, const uint8_t *buf, 1097 int nb_sectors, int64_t sector_num) 1098 { 1099 int ret; 1100 VmdkGrainMarker *data = NULL; 1101 uLongf buf_len; 1102 const uint8_t *write_buf = buf; 1103 int write_len = nb_sectors * 512; 1104 1105 if (extent->compressed) { 1106 if (!extent->has_marker) { 1107 ret = -EINVAL; 1108 goto out; 1109 } 1110 buf_len = (extent->cluster_sectors << 9) * 2; 1111 data = g_malloc(buf_len + sizeof(VmdkGrainMarker)); 1112 if (compress(data->data, &buf_len, buf, nb_sectors << 9) != Z_OK || 1113 buf_len == 0) { 1114 ret = -EINVAL; 1115 goto out; 1116 } 1117 data->lba = sector_num; 1118 data->size = buf_len; 1119 write_buf = (uint8_t *)data; 1120 write_len = buf_len + sizeof(VmdkGrainMarker); 1121 } 1122 ret = bdrv_pwrite(extent->file, 1123 cluster_offset + offset_in_cluster, 1124 write_buf, 1125 write_len); 1126 if (ret != write_len) { 1127 ret = ret < 0 ? ret : -EIO; 1128 goto out; 1129 } 1130 ret = 0; 1131 out: 1132 g_free(data); 1133 return ret; 1134 } 1135 1136 static int vmdk_read_extent(VmdkExtent *extent, int64_t cluster_offset, 1137 int64_t offset_in_cluster, uint8_t *buf, 1138 int nb_sectors) 1139 { 1140 int ret; 1141 int cluster_bytes, buf_bytes; 1142 uint8_t *cluster_buf, *compressed_data; 1143 uint8_t *uncomp_buf; 1144 uint32_t data_len; 1145 VmdkGrainMarker *marker; 1146 uLongf buf_len; 1147 1148 1149 if (!extent->compressed) { 1150 ret = bdrv_pread(extent->file, 1151 cluster_offset + offset_in_cluster, 1152 buf, nb_sectors * 512); 1153 if (ret == nb_sectors * 512) { 1154 return 0; 1155 } else { 1156 return -EIO; 1157 } 1158 } 1159 cluster_bytes = extent->cluster_sectors * 512; 1160 /* Read two clusters in case GrainMarker + compressed data > one cluster */ 1161 buf_bytes = cluster_bytes * 2; 1162 cluster_buf = g_malloc(buf_bytes); 1163 uncomp_buf = g_malloc(cluster_bytes); 1164 ret = bdrv_pread(extent->file, 1165 cluster_offset, 1166 cluster_buf, buf_bytes); 1167 if (ret < 0) { 1168 goto out; 1169 } 1170 compressed_data = cluster_buf; 1171 buf_len = cluster_bytes; 1172 data_len = cluster_bytes; 1173 if (extent->has_marker) { 1174 marker = (VmdkGrainMarker *)cluster_buf; 1175 compressed_data = marker->data; 1176 data_len = le32_to_cpu(marker->size); 1177 } 1178 if (!data_len || data_len > buf_bytes) { 1179 ret = -EINVAL; 1180 goto out; 1181 } 1182 ret = uncompress(uncomp_buf, &buf_len, compressed_data, data_len); 1183 if (ret != Z_OK) { 1184 ret = -EINVAL; 1185 goto out; 1186 1187 } 1188 if (offset_in_cluster < 0 || 1189 offset_in_cluster + nb_sectors * 512 > buf_len) { 1190 ret = -EINVAL; 1191 goto out; 1192 } 1193 memcpy(buf, uncomp_buf + offset_in_cluster, nb_sectors * 512); 1194 ret = 0; 1195 1196 out: 1197 g_free(uncomp_buf); 1198 g_free(cluster_buf); 1199 return ret; 1200 } 1201 1202 static int vmdk_read(BlockDriverState *bs, int64_t sector_num, 1203 uint8_t *buf, int nb_sectors) 1204 { 1205 BDRVVmdkState *s = bs->opaque; 1206 int ret; 1207 uint64_t n, index_in_cluster; 1208 uint64_t extent_begin_sector, extent_relative_sector_num; 1209 VmdkExtent *extent = NULL; 1210 uint64_t cluster_offset; 1211 1212 while (nb_sectors > 0) { 1213 extent = find_extent(s, sector_num, extent); 1214 if (!extent) { 1215 return -EIO; 1216 } 1217 ret = get_cluster_offset( 1218 bs, extent, NULL, 1219 sector_num << 9, 0, &cluster_offset); 1220 extent_begin_sector = extent->end_sector - extent->sectors; 1221 extent_relative_sector_num = sector_num - extent_begin_sector; 1222 index_in_cluster = extent_relative_sector_num % extent->cluster_sectors; 1223 n = extent->cluster_sectors - index_in_cluster; 1224 if (n > nb_sectors) { 1225 n = nb_sectors; 1226 } 1227 if (ret != VMDK_OK) { 1228 /* if not allocated, try to read from parent image, if exist */ 1229 if (bs->backing_hd && ret != VMDK_ZEROED) { 1230 if (!vmdk_is_cid_valid(bs)) { 1231 return -EINVAL; 1232 } 1233 ret = bdrv_read(bs->backing_hd, sector_num, buf, n); 1234 if (ret < 0) { 1235 return ret; 1236 } 1237 } else { 1238 memset(buf, 0, 512 * n); 1239 } 1240 } else { 1241 ret = vmdk_read_extent(extent, 1242 cluster_offset, index_in_cluster * 512, 1243 buf, n); 1244 if (ret) { 1245 return ret; 1246 } 1247 } 1248 nb_sectors -= n; 1249 sector_num += n; 1250 buf += n * 512; 1251 } 1252 return 0; 1253 } 1254 1255 static coroutine_fn int vmdk_co_read(BlockDriverState *bs, int64_t sector_num, 1256 uint8_t *buf, int nb_sectors) 1257 { 1258 int ret; 1259 BDRVVmdkState *s = bs->opaque; 1260 qemu_co_mutex_lock(&s->lock); 1261 ret = vmdk_read(bs, sector_num, buf, nb_sectors); 1262 qemu_co_mutex_unlock(&s->lock); 1263 return ret; 1264 } 1265 1266 /** 1267 * vmdk_write: 1268 * @zeroed: buf is ignored (data is zero), use zeroed_grain GTE feature 1269 * if possible, otherwise return -ENOTSUP. 1270 * @zero_dry_run: used for zeroed == true only, don't update L2 table, just try 1271 * with each cluster. By dry run we can find if the zero write 1272 * is possible without modifying image data. 1273 * 1274 * Returns: error code with 0 for success. 1275 */ 1276 static int vmdk_write(BlockDriverState *bs, int64_t sector_num, 1277 const uint8_t *buf, int nb_sectors, 1278 bool zeroed, bool zero_dry_run) 1279 { 1280 BDRVVmdkState *s = bs->opaque; 1281 VmdkExtent *extent = NULL; 1282 int n, ret; 1283 int64_t index_in_cluster; 1284 uint64_t extent_begin_sector, extent_relative_sector_num; 1285 uint64_t cluster_offset; 1286 VmdkMetaData m_data; 1287 1288 if (sector_num > bs->total_sectors) { 1289 fprintf(stderr, 1290 "(VMDK) Wrong offset: sector_num=0x%" PRIx64 1291 " total_sectors=0x%" PRIx64 "\n", 1292 sector_num, bs->total_sectors); 1293 return -EIO; 1294 } 1295 1296 while (nb_sectors > 0) { 1297 extent = find_extent(s, sector_num, extent); 1298 if (!extent) { 1299 return -EIO; 1300 } 1301 ret = get_cluster_offset( 1302 bs, 1303 extent, 1304 &m_data, 1305 sector_num << 9, !extent->compressed, 1306 &cluster_offset); 1307 if (extent->compressed) { 1308 if (ret == VMDK_OK) { 1309 /* Refuse write to allocated cluster for streamOptimized */ 1310 fprintf(stderr, 1311 "VMDK: can't write to allocated cluster" 1312 " for streamOptimized\n"); 1313 return -EIO; 1314 } else { 1315 /* allocate */ 1316 ret = get_cluster_offset( 1317 bs, 1318 extent, 1319 &m_data, 1320 sector_num << 9, 1, 1321 &cluster_offset); 1322 } 1323 } 1324 if (ret == VMDK_ERROR) { 1325 return -EINVAL; 1326 } 1327 extent_begin_sector = extent->end_sector - extent->sectors; 1328 extent_relative_sector_num = sector_num - extent_begin_sector; 1329 index_in_cluster = extent_relative_sector_num % extent->cluster_sectors; 1330 n = extent->cluster_sectors - index_in_cluster; 1331 if (n > nb_sectors) { 1332 n = nb_sectors; 1333 } 1334 if (zeroed) { 1335 /* Do zeroed write, buf is ignored */ 1336 if (extent->has_zero_grain && 1337 index_in_cluster == 0 && 1338 n >= extent->cluster_sectors) { 1339 n = extent->cluster_sectors; 1340 if (!zero_dry_run) { 1341 m_data.offset = VMDK_GTE_ZEROED; 1342 /* update L2 tables */ 1343 if (vmdk_L2update(extent, &m_data) != VMDK_OK) { 1344 return -EIO; 1345 } 1346 } 1347 } else { 1348 return -ENOTSUP; 1349 } 1350 } else { 1351 ret = vmdk_write_extent(extent, 1352 cluster_offset, index_in_cluster * 512, 1353 buf, n, sector_num); 1354 if (ret) { 1355 return ret; 1356 } 1357 if (m_data.valid) { 1358 /* update L2 tables */ 1359 if (vmdk_L2update(extent, &m_data) != VMDK_OK) { 1360 return -EIO; 1361 } 1362 } 1363 } 1364 nb_sectors -= n; 1365 sector_num += n; 1366 buf += n * 512; 1367 1368 /* update CID on the first write every time the virtual disk is 1369 * opened */ 1370 if (!s->cid_updated) { 1371 ret = vmdk_write_cid(bs, time(NULL)); 1372 if (ret < 0) { 1373 return ret; 1374 } 1375 s->cid_updated = true; 1376 } 1377 } 1378 return 0; 1379 } 1380 1381 static coroutine_fn int vmdk_co_write(BlockDriverState *bs, int64_t sector_num, 1382 const uint8_t *buf, int nb_sectors) 1383 { 1384 int ret; 1385 BDRVVmdkState *s = bs->opaque; 1386 qemu_co_mutex_lock(&s->lock); 1387 ret = vmdk_write(bs, sector_num, buf, nb_sectors, false, false); 1388 qemu_co_mutex_unlock(&s->lock); 1389 return ret; 1390 } 1391 1392 static int coroutine_fn vmdk_co_write_zeroes(BlockDriverState *bs, 1393 int64_t sector_num, 1394 int nb_sectors) 1395 { 1396 int ret; 1397 BDRVVmdkState *s = bs->opaque; 1398 qemu_co_mutex_lock(&s->lock); 1399 /* write zeroes could fail if sectors not aligned to cluster, test it with 1400 * dry_run == true before really updating image */ 1401 ret = vmdk_write(bs, sector_num, NULL, nb_sectors, true, true); 1402 if (!ret) { 1403 ret = vmdk_write(bs, sector_num, NULL, nb_sectors, true, false); 1404 } 1405 qemu_co_mutex_unlock(&s->lock); 1406 return ret; 1407 } 1408 1409 static int vmdk_create_extent(const char *filename, int64_t filesize, 1410 bool flat, bool compress, bool zeroed_grain) 1411 { 1412 int ret, i; 1413 int fd = 0; 1414 VMDK4Header header; 1415 uint32_t tmp, magic, grains, gd_size, gt_size, gt_count; 1416 1417 fd = qemu_open(filename, 1418 O_WRONLY | O_CREAT | O_TRUNC | O_BINARY | O_LARGEFILE, 1419 0644); 1420 if (fd < 0) { 1421 return -errno; 1422 } 1423 if (flat) { 1424 ret = ftruncate(fd, filesize); 1425 if (ret < 0) { 1426 ret = -errno; 1427 } 1428 goto exit; 1429 } 1430 magic = cpu_to_be32(VMDK4_MAGIC); 1431 memset(&header, 0, sizeof(header)); 1432 header.version = zeroed_grain ? 2 : 1; 1433 header.flags = VMDK4_FLAG_RGD | VMDK4_FLAG_NL_DETECT 1434 | (compress ? VMDK4_FLAG_COMPRESS | VMDK4_FLAG_MARKER : 0) 1435 | (zeroed_grain ? VMDK4_FLAG_ZERO_GRAIN : 0); 1436 header.compressAlgorithm = compress ? VMDK4_COMPRESSION_DEFLATE : 0; 1437 header.capacity = filesize / 512; 1438 header.granularity = 128; 1439 header.num_gtes_per_gt = 512; 1440 1441 grains = (filesize / 512 + header.granularity - 1) / header.granularity; 1442 gt_size = ((header.num_gtes_per_gt * sizeof(uint32_t)) + 511) >> 9; 1443 gt_count = 1444 (grains + header.num_gtes_per_gt - 1) / header.num_gtes_per_gt; 1445 gd_size = (gt_count * sizeof(uint32_t) + 511) >> 9; 1446 1447 header.desc_offset = 1; 1448 header.desc_size = 20; 1449 header.rgd_offset = header.desc_offset + header.desc_size; 1450 header.gd_offset = header.rgd_offset + gd_size + (gt_size * gt_count); 1451 header.grain_offset = 1452 ((header.gd_offset + gd_size + (gt_size * gt_count) + 1453 header.granularity - 1) / header.granularity) * 1454 header.granularity; 1455 /* swap endianness for all header fields */ 1456 header.version = cpu_to_le32(header.version); 1457 header.flags = cpu_to_le32(header.flags); 1458 header.capacity = cpu_to_le64(header.capacity); 1459 header.granularity = cpu_to_le64(header.granularity); 1460 header.num_gtes_per_gt = cpu_to_le32(header.num_gtes_per_gt); 1461 header.desc_offset = cpu_to_le64(header.desc_offset); 1462 header.desc_size = cpu_to_le64(header.desc_size); 1463 header.rgd_offset = cpu_to_le64(header.rgd_offset); 1464 header.gd_offset = cpu_to_le64(header.gd_offset); 1465 header.grain_offset = cpu_to_le64(header.grain_offset); 1466 header.compressAlgorithm = cpu_to_le16(header.compressAlgorithm); 1467 1468 header.check_bytes[0] = 0xa; 1469 header.check_bytes[1] = 0x20; 1470 header.check_bytes[2] = 0xd; 1471 header.check_bytes[3] = 0xa; 1472 1473 /* write all the data */ 1474 ret = qemu_write_full(fd, &magic, sizeof(magic)); 1475 if (ret != sizeof(magic)) { 1476 ret = -errno; 1477 goto exit; 1478 } 1479 ret = qemu_write_full(fd, &header, sizeof(header)); 1480 if (ret != sizeof(header)) { 1481 ret = -errno; 1482 goto exit; 1483 } 1484 1485 ret = ftruncate(fd, le64_to_cpu(header.grain_offset) << 9); 1486 if (ret < 0) { 1487 ret = -errno; 1488 goto exit; 1489 } 1490 1491 /* write grain directory */ 1492 lseek(fd, le64_to_cpu(header.rgd_offset) << 9, SEEK_SET); 1493 for (i = 0, tmp = le64_to_cpu(header.rgd_offset) + gd_size; 1494 i < gt_count; i++, tmp += gt_size) { 1495 ret = qemu_write_full(fd, &tmp, sizeof(tmp)); 1496 if (ret != sizeof(tmp)) { 1497 ret = -errno; 1498 goto exit; 1499 } 1500 } 1501 1502 /* write backup grain directory */ 1503 lseek(fd, le64_to_cpu(header.gd_offset) << 9, SEEK_SET); 1504 for (i = 0, tmp = le64_to_cpu(header.gd_offset) + gd_size; 1505 i < gt_count; i++, tmp += gt_size) { 1506 ret = qemu_write_full(fd, &tmp, sizeof(tmp)); 1507 if (ret != sizeof(tmp)) { 1508 ret = -errno; 1509 goto exit; 1510 } 1511 } 1512 1513 ret = 0; 1514 exit: 1515 qemu_close(fd); 1516 return ret; 1517 } 1518 1519 static int filename_decompose(const char *filename, char *path, char *prefix, 1520 char *postfix, size_t buf_len) 1521 { 1522 const char *p, *q; 1523 1524 if (filename == NULL || !strlen(filename)) { 1525 fprintf(stderr, "Vmdk: no filename provided.\n"); 1526 return VMDK_ERROR; 1527 } 1528 p = strrchr(filename, '/'); 1529 if (p == NULL) { 1530 p = strrchr(filename, '\\'); 1531 } 1532 if (p == NULL) { 1533 p = strrchr(filename, ':'); 1534 } 1535 if (p != NULL) { 1536 p++; 1537 if (p - filename >= buf_len) { 1538 return VMDK_ERROR; 1539 } 1540 pstrcpy(path, p - filename + 1, filename); 1541 } else { 1542 p = filename; 1543 path[0] = '\0'; 1544 } 1545 q = strrchr(p, '.'); 1546 if (q == NULL) { 1547 pstrcpy(prefix, buf_len, p); 1548 postfix[0] = '\0'; 1549 } else { 1550 if (q - p >= buf_len) { 1551 return VMDK_ERROR; 1552 } 1553 pstrcpy(prefix, q - p + 1, p); 1554 pstrcpy(postfix, buf_len, q); 1555 } 1556 return VMDK_OK; 1557 } 1558 1559 static int vmdk_create(const char *filename, QEMUOptionParameter *options, 1560 Error **errp) 1561 { 1562 int fd, idx = 0; 1563 char desc[BUF_SIZE]; 1564 int64_t total_size = 0, filesize; 1565 const char *adapter_type = NULL; 1566 const char *backing_file = NULL; 1567 const char *fmt = NULL; 1568 int flags = 0; 1569 int ret = 0; 1570 bool flat, split, compress; 1571 char ext_desc_lines[BUF_SIZE] = ""; 1572 char path[PATH_MAX], prefix[PATH_MAX], postfix[PATH_MAX]; 1573 const int64_t split_size = 0x80000000; /* VMDK has constant split size */ 1574 const char *desc_extent_line; 1575 char parent_desc_line[BUF_SIZE] = ""; 1576 uint32_t parent_cid = 0xffffffff; 1577 uint32_t number_heads = 16; 1578 bool zeroed_grain = false; 1579 const char desc_template[] = 1580 "# Disk DescriptorFile\n" 1581 "version=1\n" 1582 "CID=%x\n" 1583 "parentCID=%x\n" 1584 "createType=\"%s\"\n" 1585 "%s" 1586 "\n" 1587 "# Extent description\n" 1588 "%s" 1589 "\n" 1590 "# The Disk Data Base\n" 1591 "#DDB\n" 1592 "\n" 1593 "ddb.virtualHWVersion = \"%d\"\n" 1594 "ddb.geometry.cylinders = \"%" PRId64 "\"\n" 1595 "ddb.geometry.heads = \"%d\"\n" 1596 "ddb.geometry.sectors = \"63\"\n" 1597 "ddb.adapterType = \"%s\"\n"; 1598 Error *local_err = NULL; 1599 1600 if (filename_decompose(filename, path, prefix, postfix, PATH_MAX)) { 1601 return -EINVAL; 1602 } 1603 /* Read out options */ 1604 while (options && options->name) { 1605 if (!strcmp(options->name, BLOCK_OPT_SIZE)) { 1606 total_size = options->value.n; 1607 } else if (!strcmp(options->name, BLOCK_OPT_ADAPTER_TYPE)) { 1608 adapter_type = options->value.s; 1609 } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) { 1610 backing_file = options->value.s; 1611 } else if (!strcmp(options->name, BLOCK_OPT_COMPAT6)) { 1612 flags |= options->value.n ? BLOCK_FLAG_COMPAT6 : 0; 1613 } else if (!strcmp(options->name, BLOCK_OPT_SUBFMT)) { 1614 fmt = options->value.s; 1615 } else if (!strcmp(options->name, BLOCK_OPT_ZEROED_GRAIN)) { 1616 zeroed_grain |= options->value.n; 1617 } 1618 options++; 1619 } 1620 if (!adapter_type) { 1621 adapter_type = "ide"; 1622 } else if (strcmp(adapter_type, "ide") && 1623 strcmp(adapter_type, "buslogic") && 1624 strcmp(adapter_type, "lsilogic") && 1625 strcmp(adapter_type, "legacyESX")) { 1626 fprintf(stderr, "VMDK: Unknown adapter type: '%s'.\n", adapter_type); 1627 return -EINVAL; 1628 } 1629 if (strcmp(adapter_type, "ide") != 0) { 1630 /* that's the number of heads with which vmware operates when 1631 creating, exporting, etc. vmdk files with a non-ide adapter type */ 1632 number_heads = 255; 1633 } 1634 if (!fmt) { 1635 /* Default format to monolithicSparse */ 1636 fmt = "monolithicSparse"; 1637 } else if (strcmp(fmt, "monolithicFlat") && 1638 strcmp(fmt, "monolithicSparse") && 1639 strcmp(fmt, "twoGbMaxExtentSparse") && 1640 strcmp(fmt, "twoGbMaxExtentFlat") && 1641 strcmp(fmt, "streamOptimized")) { 1642 fprintf(stderr, "VMDK: Unknown subformat: %s\n", fmt); 1643 return -EINVAL; 1644 } 1645 split = !(strcmp(fmt, "twoGbMaxExtentFlat") && 1646 strcmp(fmt, "twoGbMaxExtentSparse")); 1647 flat = !(strcmp(fmt, "monolithicFlat") && 1648 strcmp(fmt, "twoGbMaxExtentFlat")); 1649 compress = !strcmp(fmt, "streamOptimized"); 1650 if (flat) { 1651 desc_extent_line = "RW %lld FLAT \"%s\" 0\n"; 1652 } else { 1653 desc_extent_line = "RW %lld SPARSE \"%s\"\n"; 1654 } 1655 if (flat && backing_file) { 1656 /* not supporting backing file for flat image */ 1657 return -ENOTSUP; 1658 } 1659 if (backing_file) { 1660 BlockDriverState *bs = bdrv_new(""); 1661 ret = bdrv_open(bs, backing_file, NULL, 0, NULL, &local_err); 1662 if (ret != 0) { 1663 qerror_report_err(local_err); 1664 error_free(local_err); 1665 bdrv_unref(bs); 1666 return ret; 1667 } 1668 if (strcmp(bs->drv->format_name, "vmdk")) { 1669 bdrv_unref(bs); 1670 return -EINVAL; 1671 } 1672 parent_cid = vmdk_read_cid(bs, 0); 1673 bdrv_unref(bs); 1674 snprintf(parent_desc_line, sizeof(parent_desc_line), 1675 "parentFileNameHint=\"%s\"", backing_file); 1676 } 1677 1678 /* Create extents */ 1679 filesize = total_size; 1680 while (filesize > 0) { 1681 char desc_line[BUF_SIZE]; 1682 char ext_filename[PATH_MAX]; 1683 char desc_filename[PATH_MAX]; 1684 int64_t size = filesize; 1685 1686 if (split && size > split_size) { 1687 size = split_size; 1688 } 1689 if (split) { 1690 snprintf(desc_filename, sizeof(desc_filename), "%s-%c%03d%s", 1691 prefix, flat ? 'f' : 's', ++idx, postfix); 1692 } else if (flat) { 1693 snprintf(desc_filename, sizeof(desc_filename), "%s-flat%s", 1694 prefix, postfix); 1695 } else { 1696 snprintf(desc_filename, sizeof(desc_filename), "%s%s", 1697 prefix, postfix); 1698 } 1699 snprintf(ext_filename, sizeof(ext_filename), "%s%s", 1700 path, desc_filename); 1701 1702 if (vmdk_create_extent(ext_filename, size, 1703 flat, compress, zeroed_grain)) { 1704 return -EINVAL; 1705 } 1706 filesize -= size; 1707 1708 /* Format description line */ 1709 snprintf(desc_line, sizeof(desc_line), 1710 desc_extent_line, size / 512, desc_filename); 1711 pstrcat(ext_desc_lines, sizeof(ext_desc_lines), desc_line); 1712 } 1713 /* generate descriptor file */ 1714 snprintf(desc, sizeof(desc), desc_template, 1715 (unsigned int)time(NULL), 1716 parent_cid, 1717 fmt, 1718 parent_desc_line, 1719 ext_desc_lines, 1720 (flags & BLOCK_FLAG_COMPAT6 ? 6 : 4), 1721 total_size / (int64_t)(63 * number_heads * 512), number_heads, 1722 adapter_type); 1723 if (split || flat) { 1724 fd = qemu_open(filename, 1725 O_WRONLY | O_CREAT | O_TRUNC | O_BINARY | O_LARGEFILE, 1726 0644); 1727 } else { 1728 fd = qemu_open(filename, 1729 O_WRONLY | O_BINARY | O_LARGEFILE, 1730 0644); 1731 } 1732 if (fd < 0) { 1733 return -errno; 1734 } 1735 /* the descriptor offset = 0x200 */ 1736 if (!split && !flat && 0x200 != lseek(fd, 0x200, SEEK_SET)) { 1737 ret = -errno; 1738 goto exit; 1739 } 1740 ret = qemu_write_full(fd, desc, strlen(desc)); 1741 if (ret != strlen(desc)) { 1742 ret = -errno; 1743 goto exit; 1744 } 1745 ret = 0; 1746 exit: 1747 qemu_close(fd); 1748 return ret; 1749 } 1750 1751 static void vmdk_close(BlockDriverState *bs) 1752 { 1753 BDRVVmdkState *s = bs->opaque; 1754 1755 vmdk_free_extents(bs); 1756 1757 migrate_del_blocker(s->migration_blocker); 1758 error_free(s->migration_blocker); 1759 } 1760 1761 static coroutine_fn int vmdk_co_flush(BlockDriverState *bs) 1762 { 1763 BDRVVmdkState *s = bs->opaque; 1764 int i, err; 1765 int ret = 0; 1766 1767 for (i = 0; i < s->num_extents; i++) { 1768 err = bdrv_co_flush(s->extents[i].file); 1769 if (err < 0) { 1770 ret = err; 1771 } 1772 } 1773 return ret; 1774 } 1775 1776 static int64_t vmdk_get_allocated_file_size(BlockDriverState *bs) 1777 { 1778 int i; 1779 int64_t ret = 0; 1780 int64_t r; 1781 BDRVVmdkState *s = bs->opaque; 1782 1783 ret = bdrv_get_allocated_file_size(bs->file); 1784 if (ret < 0) { 1785 return ret; 1786 } 1787 for (i = 0; i < s->num_extents; i++) { 1788 if (s->extents[i].file == bs->file) { 1789 continue; 1790 } 1791 r = bdrv_get_allocated_file_size(s->extents[i].file); 1792 if (r < 0) { 1793 return r; 1794 } 1795 ret += r; 1796 } 1797 return ret; 1798 } 1799 1800 static int vmdk_has_zero_init(BlockDriverState *bs) 1801 { 1802 int i; 1803 BDRVVmdkState *s = bs->opaque; 1804 1805 /* If has a flat extent and its underlying storage doesn't have zero init, 1806 * return 0. */ 1807 for (i = 0; i < s->num_extents; i++) { 1808 if (s->extents[i].flat) { 1809 if (!bdrv_has_zero_init(s->extents[i].file)) { 1810 return 0; 1811 } 1812 } 1813 } 1814 return 1; 1815 } 1816 1817 static QEMUOptionParameter vmdk_create_options[] = { 1818 { 1819 .name = BLOCK_OPT_SIZE, 1820 .type = OPT_SIZE, 1821 .help = "Virtual disk size" 1822 }, 1823 { 1824 .name = BLOCK_OPT_ADAPTER_TYPE, 1825 .type = OPT_STRING, 1826 .help = "Virtual adapter type, can be one of " 1827 "ide (default), lsilogic, buslogic or legacyESX" 1828 }, 1829 { 1830 .name = BLOCK_OPT_BACKING_FILE, 1831 .type = OPT_STRING, 1832 .help = "File name of a base image" 1833 }, 1834 { 1835 .name = BLOCK_OPT_COMPAT6, 1836 .type = OPT_FLAG, 1837 .help = "VMDK version 6 image" 1838 }, 1839 { 1840 .name = BLOCK_OPT_SUBFMT, 1841 .type = OPT_STRING, 1842 .help = 1843 "VMDK flat extent format, can be one of " 1844 "{monolithicSparse (default) | monolithicFlat | twoGbMaxExtentSparse | twoGbMaxExtentFlat | streamOptimized} " 1845 }, 1846 { 1847 .name = BLOCK_OPT_ZEROED_GRAIN, 1848 .type = OPT_FLAG, 1849 .help = "Enable efficient zero writes using the zeroed-grain GTE feature" 1850 }, 1851 { NULL } 1852 }; 1853 1854 static BlockDriver bdrv_vmdk = { 1855 .format_name = "vmdk", 1856 .instance_size = sizeof(BDRVVmdkState), 1857 .bdrv_probe = vmdk_probe, 1858 .bdrv_open = vmdk_open, 1859 .bdrv_reopen_prepare = vmdk_reopen_prepare, 1860 .bdrv_read = vmdk_co_read, 1861 .bdrv_write = vmdk_co_write, 1862 .bdrv_co_write_zeroes = vmdk_co_write_zeroes, 1863 .bdrv_close = vmdk_close, 1864 .bdrv_create = vmdk_create, 1865 .bdrv_co_flush_to_disk = vmdk_co_flush, 1866 .bdrv_co_get_block_status = vmdk_co_get_block_status, 1867 .bdrv_get_allocated_file_size = vmdk_get_allocated_file_size, 1868 .bdrv_has_zero_init = vmdk_has_zero_init, 1869 1870 .create_options = vmdk_create_options, 1871 }; 1872 1873 static void bdrv_vmdk_init(void) 1874 { 1875 bdrv_register(&bdrv_vmdk); 1876 } 1877 1878 block_init(bdrv_vmdk_init); 1879