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 int64_t next_cluster_sector; 110 char *type; 111 } VmdkExtent; 112 113 typedef struct BDRVVmdkState { 114 CoMutex lock; 115 uint64_t desc_offset; 116 bool cid_updated; 117 bool cid_checked; 118 uint32_t cid; 119 uint32_t parent_cid; 120 int num_extents; 121 /* Extent array with num_extents entries, ascend ordered by address */ 122 VmdkExtent *extents; 123 Error *migration_blocker; 124 char *create_type; 125 } BDRVVmdkState; 126 127 typedef struct VmdkMetaData { 128 unsigned int l1_index; 129 unsigned int l2_index; 130 unsigned int l2_offset; 131 int valid; 132 uint32_t *l2_cache_entry; 133 } VmdkMetaData; 134 135 typedef struct VmdkGrainMarker { 136 uint64_t lba; 137 uint32_t size; 138 uint8_t data[0]; 139 } QEMU_PACKED VmdkGrainMarker; 140 141 enum { 142 MARKER_END_OF_STREAM = 0, 143 MARKER_GRAIN_TABLE = 1, 144 MARKER_GRAIN_DIRECTORY = 2, 145 MARKER_FOOTER = 3, 146 }; 147 148 static int vmdk_probe(const uint8_t *buf, int buf_size, const char *filename) 149 { 150 uint32_t magic; 151 152 if (buf_size < 4) { 153 return 0; 154 } 155 magic = be32_to_cpu(*(uint32_t *)buf); 156 if (magic == VMDK3_MAGIC || 157 magic == VMDK4_MAGIC) { 158 return 100; 159 } else { 160 const char *p = (const char *)buf; 161 const char *end = p + buf_size; 162 while (p < end) { 163 if (*p == '#') { 164 /* skip comment line */ 165 while (p < end && *p != '\n') { 166 p++; 167 } 168 p++; 169 continue; 170 } 171 if (*p == ' ') { 172 while (p < end && *p == ' ') { 173 p++; 174 } 175 /* skip '\r' if windows line endings used. */ 176 if (p < end && *p == '\r') { 177 p++; 178 } 179 /* only accept blank lines before 'version=' line */ 180 if (p == end || *p != '\n') { 181 return 0; 182 } 183 p++; 184 continue; 185 } 186 if (end - p >= strlen("version=X\n")) { 187 if (strncmp("version=1\n", p, strlen("version=1\n")) == 0 || 188 strncmp("version=2\n", p, strlen("version=2\n")) == 0) { 189 return 100; 190 } 191 } 192 if (end - p >= strlen("version=X\r\n")) { 193 if (strncmp("version=1\r\n", p, strlen("version=1\r\n")) == 0 || 194 strncmp("version=2\r\n", p, strlen("version=2\r\n")) == 0) { 195 return 100; 196 } 197 } 198 return 0; 199 } 200 return 0; 201 } 202 } 203 204 #define SECTOR_SIZE 512 205 #define DESC_SIZE (20 * SECTOR_SIZE) /* 20 sectors of 512 bytes each */ 206 #define BUF_SIZE 4096 207 #define HEADER_SIZE 512 /* first sector of 512 bytes */ 208 209 static void vmdk_free_extents(BlockDriverState *bs) 210 { 211 int i; 212 BDRVVmdkState *s = bs->opaque; 213 VmdkExtent *e; 214 215 for (i = 0; i < s->num_extents; i++) { 216 e = &s->extents[i]; 217 g_free(e->l1_table); 218 g_free(e->l2_cache); 219 g_free(e->l1_backup_table); 220 g_free(e->type); 221 if (e->file != bs->file) { 222 bdrv_unref(e->file); 223 } 224 } 225 g_free(s->extents); 226 } 227 228 static void vmdk_free_last_extent(BlockDriverState *bs) 229 { 230 BDRVVmdkState *s = bs->opaque; 231 232 if (s->num_extents == 0) { 233 return; 234 } 235 s->num_extents--; 236 s->extents = g_realloc(s->extents, s->num_extents * sizeof(VmdkExtent)); 237 } 238 239 static uint32_t vmdk_read_cid(BlockDriverState *bs, int parent) 240 { 241 char desc[DESC_SIZE]; 242 uint32_t cid = 0xffffffff; 243 const char *p_name, *cid_str; 244 size_t cid_str_size; 245 BDRVVmdkState *s = bs->opaque; 246 int ret; 247 248 ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE); 249 if (ret < 0) { 250 return 0; 251 } 252 253 if (parent) { 254 cid_str = "parentCID"; 255 cid_str_size = sizeof("parentCID"); 256 } else { 257 cid_str = "CID"; 258 cid_str_size = sizeof("CID"); 259 } 260 261 desc[DESC_SIZE - 1] = '\0'; 262 p_name = strstr(desc, cid_str); 263 if (p_name != NULL) { 264 p_name += cid_str_size; 265 sscanf(p_name, "%" SCNx32, &cid); 266 } 267 268 return cid; 269 } 270 271 static int vmdk_write_cid(BlockDriverState *bs, uint32_t cid) 272 { 273 char desc[DESC_SIZE], tmp_desc[DESC_SIZE]; 274 char *p_name, *tmp_str; 275 BDRVVmdkState *s = bs->opaque; 276 int ret; 277 278 ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE); 279 if (ret < 0) { 280 return ret; 281 } 282 283 desc[DESC_SIZE - 1] = '\0'; 284 tmp_str = strstr(desc, "parentCID"); 285 if (tmp_str == NULL) { 286 return -EINVAL; 287 } 288 289 pstrcpy(tmp_desc, sizeof(tmp_desc), tmp_str); 290 p_name = strstr(desc, "CID"); 291 if (p_name != NULL) { 292 p_name += sizeof("CID"); 293 snprintf(p_name, sizeof(desc) - (p_name - desc), "%" PRIx32 "\n", cid); 294 pstrcat(desc, sizeof(desc), tmp_desc); 295 } 296 297 ret = bdrv_pwrite_sync(bs->file, s->desc_offset, desc, DESC_SIZE); 298 if (ret < 0) { 299 return ret; 300 } 301 302 return 0; 303 } 304 305 static int vmdk_is_cid_valid(BlockDriverState *bs) 306 { 307 BDRVVmdkState *s = bs->opaque; 308 BlockDriverState *p_bs = bs->backing_hd; 309 uint32_t cur_pcid; 310 311 if (!s->cid_checked && p_bs) { 312 cur_pcid = vmdk_read_cid(p_bs, 0); 313 if (s->parent_cid != cur_pcid) { 314 /* CID not valid */ 315 return 0; 316 } 317 } 318 s->cid_checked = true; 319 /* CID valid */ 320 return 1; 321 } 322 323 /* Queue extents, if any, for reopen() */ 324 static int vmdk_reopen_prepare(BDRVReopenState *state, 325 BlockReopenQueue *queue, Error **errp) 326 { 327 BDRVVmdkState *s; 328 int ret = -1; 329 int i; 330 VmdkExtent *e; 331 332 assert(state != NULL); 333 assert(state->bs != NULL); 334 335 if (queue == NULL) { 336 error_setg(errp, "No reopen queue for VMDK extents"); 337 goto exit; 338 } 339 340 s = state->bs->opaque; 341 342 assert(s != NULL); 343 344 for (i = 0; i < s->num_extents; i++) { 345 e = &s->extents[i]; 346 if (e->file != state->bs->file) { 347 bdrv_reopen_queue(queue, e->file, state->flags); 348 } 349 } 350 ret = 0; 351 352 exit: 353 return ret; 354 } 355 356 static int vmdk_parent_open(BlockDriverState *bs) 357 { 358 char *p_name; 359 char desc[DESC_SIZE + 1]; 360 BDRVVmdkState *s = bs->opaque; 361 int ret; 362 363 desc[DESC_SIZE] = '\0'; 364 ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE); 365 if (ret < 0) { 366 return ret; 367 } 368 369 p_name = strstr(desc, "parentFileNameHint"); 370 if (p_name != NULL) { 371 char *end_name; 372 373 p_name += sizeof("parentFileNameHint") + 1; 374 end_name = strchr(p_name, '\"'); 375 if (end_name == NULL) { 376 return -EINVAL; 377 } 378 if ((end_name - p_name) > sizeof(bs->backing_file) - 1) { 379 return -EINVAL; 380 } 381 382 pstrcpy(bs->backing_file, end_name - p_name + 1, p_name); 383 } 384 385 return 0; 386 } 387 388 /* Create and append extent to the extent array. Return the added VmdkExtent 389 * address. return NULL if allocation failed. */ 390 static int vmdk_add_extent(BlockDriverState *bs, 391 BlockDriverState *file, bool flat, int64_t sectors, 392 int64_t l1_offset, int64_t l1_backup_offset, 393 uint32_t l1_size, 394 int l2_size, uint64_t cluster_sectors, 395 VmdkExtent **new_extent, 396 Error **errp) 397 { 398 VmdkExtent *extent; 399 BDRVVmdkState *s = bs->opaque; 400 int64_t length; 401 402 if (cluster_sectors > 0x200000) { 403 /* 0x200000 * 512Bytes = 1GB for one cluster is unrealistic */ 404 error_setg(errp, "Invalid granularity, image may be corrupt"); 405 return -EFBIG; 406 } 407 if (l1_size > 512 * 1024 * 1024) { 408 /* Although with big capacity and small l1_entry_sectors, we can get a 409 * big l1_size, we don't want unbounded value to allocate the table. 410 * Limit it to 512M, which is 16PB for default cluster and L2 table 411 * size */ 412 error_setg(errp, "L1 size too big"); 413 return -EFBIG; 414 } 415 416 length = bdrv_getlength(file); 417 if (length < 0) { 418 return length; 419 } 420 421 s->extents = g_realloc(s->extents, 422 (s->num_extents + 1) * sizeof(VmdkExtent)); 423 extent = &s->extents[s->num_extents]; 424 s->num_extents++; 425 426 memset(extent, 0, sizeof(VmdkExtent)); 427 extent->file = file; 428 extent->flat = flat; 429 extent->sectors = sectors; 430 extent->l1_table_offset = l1_offset; 431 extent->l1_backup_table_offset = l1_backup_offset; 432 extent->l1_size = l1_size; 433 extent->l1_entry_sectors = l2_size * cluster_sectors; 434 extent->l2_size = l2_size; 435 extent->cluster_sectors = flat ? sectors : cluster_sectors; 436 extent->next_cluster_sector = 437 ROUND_UP(DIV_ROUND_UP(length, BDRV_SECTOR_SIZE), cluster_sectors); 438 439 if (s->num_extents > 1) { 440 extent->end_sector = (*(extent - 1)).end_sector + extent->sectors; 441 } else { 442 extent->end_sector = extent->sectors; 443 } 444 bs->total_sectors = extent->end_sector; 445 if (new_extent) { 446 *new_extent = extent; 447 } 448 return 0; 449 } 450 451 static int vmdk_init_tables(BlockDriverState *bs, VmdkExtent *extent, 452 Error **errp) 453 { 454 int ret; 455 int l1_size, i; 456 457 /* read the L1 table */ 458 l1_size = extent->l1_size * sizeof(uint32_t); 459 extent->l1_table = g_try_malloc(l1_size); 460 if (l1_size && extent->l1_table == NULL) { 461 return -ENOMEM; 462 } 463 464 ret = bdrv_pread(extent->file, 465 extent->l1_table_offset, 466 extent->l1_table, 467 l1_size); 468 if (ret < 0) { 469 error_setg_errno(errp, -ret, 470 "Could not read l1 table from extent '%s'", 471 extent->file->filename); 472 goto fail_l1; 473 } 474 for (i = 0; i < extent->l1_size; i++) { 475 le32_to_cpus(&extent->l1_table[i]); 476 } 477 478 if (extent->l1_backup_table_offset) { 479 extent->l1_backup_table = g_try_malloc(l1_size); 480 if (l1_size && extent->l1_backup_table == NULL) { 481 ret = -ENOMEM; 482 goto fail_l1; 483 } 484 ret = bdrv_pread(extent->file, 485 extent->l1_backup_table_offset, 486 extent->l1_backup_table, 487 l1_size); 488 if (ret < 0) { 489 error_setg_errno(errp, -ret, 490 "Could not read l1 backup table from extent '%s'", 491 extent->file->filename); 492 goto fail_l1b; 493 } 494 for (i = 0; i < extent->l1_size; i++) { 495 le32_to_cpus(&extent->l1_backup_table[i]); 496 } 497 } 498 499 extent->l2_cache = 500 g_malloc(extent->l2_size * L2_CACHE_SIZE * sizeof(uint32_t)); 501 return 0; 502 fail_l1b: 503 g_free(extent->l1_backup_table); 504 fail_l1: 505 g_free(extent->l1_table); 506 return ret; 507 } 508 509 static int vmdk_open_vmfs_sparse(BlockDriverState *bs, 510 BlockDriverState *file, 511 int flags, Error **errp) 512 { 513 int ret; 514 uint32_t magic; 515 VMDK3Header header; 516 VmdkExtent *extent; 517 518 ret = bdrv_pread(file, sizeof(magic), &header, sizeof(header)); 519 if (ret < 0) { 520 error_setg_errno(errp, -ret, 521 "Could not read header from file '%s'", 522 file->filename); 523 return ret; 524 } 525 ret = vmdk_add_extent(bs, file, false, 526 le32_to_cpu(header.disk_sectors), 527 le32_to_cpu(header.l1dir_offset) << 9, 528 0, 529 le32_to_cpu(header.l1dir_size), 530 4096, 531 le32_to_cpu(header.granularity), 532 &extent, 533 errp); 534 if (ret < 0) { 535 return ret; 536 } 537 ret = vmdk_init_tables(bs, extent, errp); 538 if (ret) { 539 /* free extent allocated by vmdk_add_extent */ 540 vmdk_free_last_extent(bs); 541 } 542 return ret; 543 } 544 545 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf, 546 Error **errp); 547 548 static char *vmdk_read_desc(BlockDriverState *file, uint64_t desc_offset, 549 Error **errp) 550 { 551 int64_t size; 552 char *buf; 553 int ret; 554 555 size = bdrv_getlength(file); 556 if (size < 0) { 557 error_setg_errno(errp, -size, "Could not access file"); 558 return NULL; 559 } 560 561 size = MIN(size, 1 << 20); /* avoid unbounded allocation */ 562 buf = g_malloc0(size + 1); 563 564 ret = bdrv_pread(file, desc_offset, buf, size); 565 if (ret < 0) { 566 error_setg_errno(errp, -ret, "Could not read from file"); 567 g_free(buf); 568 return NULL; 569 } 570 571 return buf; 572 } 573 574 static int vmdk_open_vmdk4(BlockDriverState *bs, 575 BlockDriverState *file, 576 int flags, Error **errp) 577 { 578 int ret; 579 uint32_t magic; 580 uint32_t l1_size, l1_entry_sectors; 581 VMDK4Header header; 582 VmdkExtent *extent; 583 BDRVVmdkState *s = bs->opaque; 584 int64_t l1_backup_offset = 0; 585 586 ret = bdrv_pread(file, sizeof(magic), &header, sizeof(header)); 587 if (ret < 0) { 588 error_setg_errno(errp, -ret, 589 "Could not read header from file '%s'", 590 file->filename); 591 return -EINVAL; 592 } 593 if (header.capacity == 0) { 594 uint64_t desc_offset = le64_to_cpu(header.desc_offset); 595 if (desc_offset) { 596 char *buf = vmdk_read_desc(file, desc_offset << 9, errp); 597 if (!buf) { 598 return -EINVAL; 599 } 600 ret = vmdk_open_desc_file(bs, flags, buf, errp); 601 g_free(buf); 602 return ret; 603 } 604 } 605 606 if (!s->create_type) { 607 s->create_type = g_strdup("monolithicSparse"); 608 } 609 610 if (le64_to_cpu(header.gd_offset) == VMDK4_GD_AT_END) { 611 /* 612 * The footer takes precedence over the header, so read it in. The 613 * footer starts at offset -1024 from the end: One sector for the 614 * footer, and another one for the end-of-stream marker. 615 */ 616 struct { 617 struct { 618 uint64_t val; 619 uint32_t size; 620 uint32_t type; 621 uint8_t pad[512 - 16]; 622 } QEMU_PACKED footer_marker; 623 624 uint32_t magic; 625 VMDK4Header header; 626 uint8_t pad[512 - 4 - sizeof(VMDK4Header)]; 627 628 struct { 629 uint64_t val; 630 uint32_t size; 631 uint32_t type; 632 uint8_t pad[512 - 16]; 633 } QEMU_PACKED eos_marker; 634 } QEMU_PACKED footer; 635 636 ret = bdrv_pread(file, 637 bs->file->total_sectors * 512 - 1536, 638 &footer, sizeof(footer)); 639 if (ret < 0) { 640 return ret; 641 } 642 643 /* Some sanity checks for the footer */ 644 if (be32_to_cpu(footer.magic) != VMDK4_MAGIC || 645 le32_to_cpu(footer.footer_marker.size) != 0 || 646 le32_to_cpu(footer.footer_marker.type) != MARKER_FOOTER || 647 le64_to_cpu(footer.eos_marker.val) != 0 || 648 le32_to_cpu(footer.eos_marker.size) != 0 || 649 le32_to_cpu(footer.eos_marker.type) != MARKER_END_OF_STREAM) 650 { 651 return -EINVAL; 652 } 653 654 header = footer.header; 655 } 656 657 if (le32_to_cpu(header.version) > 3) { 658 char buf[64]; 659 snprintf(buf, sizeof(buf), "VMDK version %" PRId32, 660 le32_to_cpu(header.version)); 661 error_set(errp, QERR_UNKNOWN_BLOCK_FORMAT_FEATURE, 662 bs->device_name, "vmdk", buf); 663 return -ENOTSUP; 664 } else if (le32_to_cpu(header.version) == 3 && (flags & BDRV_O_RDWR)) { 665 /* VMware KB 2064959 explains that version 3 added support for 666 * persistent changed block tracking (CBT), and backup software can 667 * read it as version=1 if it doesn't care about the changed area 668 * information. So we are safe to enable read only. */ 669 error_setg(errp, "VMDK version 3 must be read only"); 670 return -EINVAL; 671 } 672 673 if (le32_to_cpu(header.num_gtes_per_gt) > 512) { 674 error_setg(errp, "L2 table size too big"); 675 return -EINVAL; 676 } 677 678 l1_entry_sectors = le32_to_cpu(header.num_gtes_per_gt) 679 * le64_to_cpu(header.granularity); 680 if (l1_entry_sectors == 0) { 681 return -EINVAL; 682 } 683 l1_size = (le64_to_cpu(header.capacity) + l1_entry_sectors - 1) 684 / l1_entry_sectors; 685 if (le32_to_cpu(header.flags) & VMDK4_FLAG_RGD) { 686 l1_backup_offset = le64_to_cpu(header.rgd_offset) << 9; 687 } 688 if (bdrv_nb_sectors(file) < le64_to_cpu(header.grain_offset)) { 689 error_setg(errp, "File truncated, expecting at least %" PRId64 " bytes", 690 (int64_t)(le64_to_cpu(header.grain_offset) 691 * BDRV_SECTOR_SIZE)); 692 return -EINVAL; 693 } 694 695 ret = vmdk_add_extent(bs, file, false, 696 le64_to_cpu(header.capacity), 697 le64_to_cpu(header.gd_offset) << 9, 698 l1_backup_offset, 699 l1_size, 700 le32_to_cpu(header.num_gtes_per_gt), 701 le64_to_cpu(header.granularity), 702 &extent, 703 errp); 704 if (ret < 0) { 705 return ret; 706 } 707 extent->compressed = 708 le16_to_cpu(header.compressAlgorithm) == VMDK4_COMPRESSION_DEFLATE; 709 if (extent->compressed) { 710 g_free(s->create_type); 711 s->create_type = g_strdup("streamOptimized"); 712 } 713 extent->has_marker = le32_to_cpu(header.flags) & VMDK4_FLAG_MARKER; 714 extent->version = le32_to_cpu(header.version); 715 extent->has_zero_grain = le32_to_cpu(header.flags) & VMDK4_FLAG_ZERO_GRAIN; 716 ret = vmdk_init_tables(bs, extent, errp); 717 if (ret) { 718 /* free extent allocated by vmdk_add_extent */ 719 vmdk_free_last_extent(bs); 720 } 721 return ret; 722 } 723 724 /* find an option value out of descriptor file */ 725 static int vmdk_parse_description(const char *desc, const char *opt_name, 726 char *buf, int buf_size) 727 { 728 char *opt_pos, *opt_end; 729 const char *end = desc + strlen(desc); 730 731 opt_pos = strstr(desc, opt_name); 732 if (!opt_pos) { 733 return VMDK_ERROR; 734 } 735 /* Skip "=\"" following opt_name */ 736 opt_pos += strlen(opt_name) + 2; 737 if (opt_pos >= end) { 738 return VMDK_ERROR; 739 } 740 opt_end = opt_pos; 741 while (opt_end < end && *opt_end != '"') { 742 opt_end++; 743 } 744 if (opt_end == end || buf_size < opt_end - opt_pos + 1) { 745 return VMDK_ERROR; 746 } 747 pstrcpy(buf, opt_end - opt_pos + 1, opt_pos); 748 return VMDK_OK; 749 } 750 751 /* Open an extent file and append to bs array */ 752 static int vmdk_open_sparse(BlockDriverState *bs, 753 BlockDriverState *file, int flags, 754 char *buf, Error **errp) 755 { 756 uint32_t magic; 757 758 magic = ldl_be_p(buf); 759 switch (magic) { 760 case VMDK3_MAGIC: 761 return vmdk_open_vmfs_sparse(bs, file, flags, errp); 762 break; 763 case VMDK4_MAGIC: 764 return vmdk_open_vmdk4(bs, file, flags, errp); 765 break; 766 default: 767 error_setg(errp, "Image not in VMDK format"); 768 return -EINVAL; 769 break; 770 } 771 } 772 773 static int vmdk_parse_extents(const char *desc, BlockDriverState *bs, 774 const char *desc_file_path, Error **errp) 775 { 776 int ret; 777 char access[11]; 778 char type[11]; 779 char fname[512]; 780 const char *p = desc; 781 int64_t sectors = 0; 782 int64_t flat_offset; 783 char extent_path[PATH_MAX]; 784 BlockDriverState *extent_file; 785 BDRVVmdkState *s = bs->opaque; 786 VmdkExtent *extent; 787 788 while (*p) { 789 /* parse extent line: 790 * RW [size in sectors] FLAT "file-name.vmdk" OFFSET 791 * or 792 * RW [size in sectors] SPARSE "file-name.vmdk" 793 */ 794 flat_offset = -1; 795 ret = sscanf(p, "%10s %" SCNd64 " %10s \"%511[^\n\r\"]\" %" SCNd64, 796 access, §ors, type, fname, &flat_offset); 797 if (ret < 4 || strcmp(access, "RW")) { 798 goto next_line; 799 } else if (!strcmp(type, "FLAT")) { 800 if (ret != 5 || flat_offset < 0) { 801 error_setg(errp, "Invalid extent lines: \n%s", p); 802 return -EINVAL; 803 } 804 } else if (!strcmp(type, "VMFS")) { 805 if (ret == 4) { 806 flat_offset = 0; 807 } else { 808 error_setg(errp, "Invalid extent lines:\n%s", p); 809 return -EINVAL; 810 } 811 } else if (ret != 4) { 812 error_setg(errp, "Invalid extent lines:\n%s", p); 813 return -EINVAL; 814 } 815 816 if (sectors <= 0 || 817 (strcmp(type, "FLAT") && strcmp(type, "SPARSE") && 818 strcmp(type, "VMFS") && strcmp(type, "VMFSSPARSE")) || 819 (strcmp(access, "RW"))) { 820 goto next_line; 821 } 822 823 path_combine(extent_path, sizeof(extent_path), 824 desc_file_path, fname); 825 extent_file = NULL; 826 ret = bdrv_open(&extent_file, extent_path, NULL, NULL, 827 bs->open_flags | BDRV_O_PROTOCOL, NULL, errp); 828 if (ret) { 829 return ret; 830 } 831 832 /* save to extents array */ 833 if (!strcmp(type, "FLAT") || !strcmp(type, "VMFS")) { 834 /* FLAT extent */ 835 836 ret = vmdk_add_extent(bs, extent_file, true, sectors, 837 0, 0, 0, 0, 0, &extent, errp); 838 if (ret < 0) { 839 return ret; 840 } 841 extent->flat_start_offset = flat_offset << 9; 842 } else if (!strcmp(type, "SPARSE") || !strcmp(type, "VMFSSPARSE")) { 843 /* SPARSE extent and VMFSSPARSE extent are both "COWD" sparse file*/ 844 char *buf = vmdk_read_desc(extent_file, 0, errp); 845 if (!buf) { 846 ret = -EINVAL; 847 } else { 848 ret = vmdk_open_sparse(bs, extent_file, bs->open_flags, buf, errp); 849 } 850 if (ret) { 851 g_free(buf); 852 bdrv_unref(extent_file); 853 return ret; 854 } 855 extent = &s->extents[s->num_extents - 1]; 856 } else { 857 error_setg(errp, "Unsupported extent type '%s'", type); 858 return -ENOTSUP; 859 } 860 extent->type = g_strdup(type); 861 next_line: 862 /* move to next line */ 863 while (*p) { 864 if (*p == '\n') { 865 p++; 866 break; 867 } 868 p++; 869 } 870 } 871 return 0; 872 } 873 874 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf, 875 Error **errp) 876 { 877 int ret; 878 char ct[128]; 879 BDRVVmdkState *s = bs->opaque; 880 881 if (vmdk_parse_description(buf, "createType", ct, sizeof(ct))) { 882 error_setg(errp, "invalid VMDK image descriptor"); 883 ret = -EINVAL; 884 goto exit; 885 } 886 if (strcmp(ct, "monolithicFlat") && 887 strcmp(ct, "vmfs") && 888 strcmp(ct, "vmfsSparse") && 889 strcmp(ct, "twoGbMaxExtentSparse") && 890 strcmp(ct, "twoGbMaxExtentFlat")) { 891 error_setg(errp, "Unsupported image type '%s'", ct); 892 ret = -ENOTSUP; 893 goto exit; 894 } 895 s->create_type = g_strdup(ct); 896 s->desc_offset = 0; 897 ret = vmdk_parse_extents(buf, bs, bs->file->filename, errp); 898 exit: 899 return ret; 900 } 901 902 static int vmdk_open(BlockDriverState *bs, QDict *options, int flags, 903 Error **errp) 904 { 905 char *buf = NULL; 906 int ret; 907 BDRVVmdkState *s = bs->opaque; 908 uint32_t magic; 909 910 buf = vmdk_read_desc(bs->file, 0, errp); 911 if (!buf) { 912 return -EINVAL; 913 } 914 915 magic = ldl_be_p(buf); 916 switch (magic) { 917 case VMDK3_MAGIC: 918 case VMDK4_MAGIC: 919 ret = vmdk_open_sparse(bs, bs->file, flags, buf, errp); 920 s->desc_offset = 0x200; 921 break; 922 default: 923 ret = vmdk_open_desc_file(bs, flags, buf, errp); 924 break; 925 } 926 if (ret) { 927 goto fail; 928 } 929 930 /* try to open parent images, if exist */ 931 ret = vmdk_parent_open(bs); 932 if (ret) { 933 goto fail; 934 } 935 s->cid = vmdk_read_cid(bs, 0); 936 s->parent_cid = vmdk_read_cid(bs, 1); 937 qemu_co_mutex_init(&s->lock); 938 939 /* Disable migration when VMDK images are used */ 940 error_set(&s->migration_blocker, 941 QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED, 942 "vmdk", bs->device_name, "live migration"); 943 migrate_add_blocker(s->migration_blocker); 944 g_free(buf); 945 return 0; 946 947 fail: 948 g_free(buf); 949 g_free(s->create_type); 950 s->create_type = NULL; 951 vmdk_free_extents(bs); 952 return ret; 953 } 954 955 956 static void vmdk_refresh_limits(BlockDriverState *bs, Error **errp) 957 { 958 BDRVVmdkState *s = bs->opaque; 959 int i; 960 961 for (i = 0; i < s->num_extents; i++) { 962 if (!s->extents[i].flat) { 963 bs->bl.write_zeroes_alignment = 964 MAX(bs->bl.write_zeroes_alignment, 965 s->extents[i].cluster_sectors); 966 } 967 } 968 } 969 970 /** 971 * get_whole_cluster 972 * 973 * Copy backing file's cluster that covers @sector_num, otherwise write zero, 974 * to the cluster at @cluster_sector_num. 975 * 976 * If @skip_start_sector < @skip_end_sector, the relative range 977 * [@skip_start_sector, @skip_end_sector) is not copied or written, and leave 978 * it for call to write user data in the request. 979 */ 980 static int get_whole_cluster(BlockDriverState *bs, 981 VmdkExtent *extent, 982 uint64_t cluster_sector_num, 983 uint64_t sector_num, 984 uint64_t skip_start_sector, 985 uint64_t skip_end_sector) 986 { 987 int ret = VMDK_OK; 988 int64_t cluster_bytes; 989 uint8_t *whole_grain; 990 991 /* For COW, align request sector_num to cluster start */ 992 sector_num = QEMU_ALIGN_DOWN(sector_num, extent->cluster_sectors); 993 cluster_bytes = extent->cluster_sectors << BDRV_SECTOR_BITS; 994 whole_grain = qemu_blockalign(bs, cluster_bytes); 995 996 if (!bs->backing_hd) { 997 memset(whole_grain, 0, skip_start_sector << BDRV_SECTOR_BITS); 998 memset(whole_grain + (skip_end_sector << BDRV_SECTOR_BITS), 0, 999 cluster_bytes - (skip_end_sector << BDRV_SECTOR_BITS)); 1000 } 1001 1002 assert(skip_end_sector <= extent->cluster_sectors); 1003 /* we will be here if it's first write on non-exist grain(cluster). 1004 * try to read from parent image, if exist */ 1005 if (bs->backing_hd && !vmdk_is_cid_valid(bs)) { 1006 ret = VMDK_ERROR; 1007 goto exit; 1008 } 1009 1010 /* Read backing data before skip range */ 1011 if (skip_start_sector > 0) { 1012 if (bs->backing_hd) { 1013 ret = bdrv_read(bs->backing_hd, sector_num, 1014 whole_grain, skip_start_sector); 1015 if (ret < 0) { 1016 ret = VMDK_ERROR; 1017 goto exit; 1018 } 1019 } 1020 ret = bdrv_write(extent->file, cluster_sector_num, whole_grain, 1021 skip_start_sector); 1022 if (ret < 0) { 1023 ret = VMDK_ERROR; 1024 goto exit; 1025 } 1026 } 1027 /* Read backing data after skip range */ 1028 if (skip_end_sector < extent->cluster_sectors) { 1029 if (bs->backing_hd) { 1030 ret = bdrv_read(bs->backing_hd, sector_num + skip_end_sector, 1031 whole_grain + (skip_end_sector << BDRV_SECTOR_BITS), 1032 extent->cluster_sectors - skip_end_sector); 1033 if (ret < 0) { 1034 ret = VMDK_ERROR; 1035 goto exit; 1036 } 1037 } 1038 ret = bdrv_write(extent->file, cluster_sector_num + skip_end_sector, 1039 whole_grain + (skip_end_sector << BDRV_SECTOR_BITS), 1040 extent->cluster_sectors - skip_end_sector); 1041 if (ret < 0) { 1042 ret = VMDK_ERROR; 1043 goto exit; 1044 } 1045 } 1046 1047 exit: 1048 qemu_vfree(whole_grain); 1049 return ret; 1050 } 1051 1052 static int vmdk_L2update(VmdkExtent *extent, VmdkMetaData *m_data, 1053 uint32_t offset) 1054 { 1055 offset = cpu_to_le32(offset); 1056 /* update L2 table */ 1057 if (bdrv_pwrite_sync( 1058 extent->file, 1059 ((int64_t)m_data->l2_offset * 512) 1060 + (m_data->l2_index * sizeof(offset)), 1061 &offset, sizeof(offset)) < 0) { 1062 return VMDK_ERROR; 1063 } 1064 /* update backup L2 table */ 1065 if (extent->l1_backup_table_offset != 0) { 1066 m_data->l2_offset = extent->l1_backup_table[m_data->l1_index]; 1067 if (bdrv_pwrite_sync( 1068 extent->file, 1069 ((int64_t)m_data->l2_offset * 512) 1070 + (m_data->l2_index * sizeof(offset)), 1071 &offset, sizeof(offset)) < 0) { 1072 return VMDK_ERROR; 1073 } 1074 } 1075 if (m_data->l2_cache_entry) { 1076 *m_data->l2_cache_entry = offset; 1077 } 1078 1079 return VMDK_OK; 1080 } 1081 1082 /** 1083 * get_cluster_offset 1084 * 1085 * Look up cluster offset in extent file by sector number, and store in 1086 * @cluster_offset. 1087 * 1088 * For flat extents, the start offset as parsed from the description file is 1089 * returned. 1090 * 1091 * For sparse extents, look up in L1, L2 table. If allocate is true, return an 1092 * offset for a new cluster and update L2 cache. If there is a backing file, 1093 * COW is done before returning; otherwise, zeroes are written to the allocated 1094 * cluster. Both COW and zero writing skips the sector range 1095 * [@skip_start_sector, @skip_end_sector) passed in by caller, because caller 1096 * has new data to write there. 1097 * 1098 * Returns: VMDK_OK if cluster exists and mapped in the image. 1099 * VMDK_UNALLOC if cluster is not mapped and @allocate is false. 1100 * VMDK_ERROR if failed. 1101 */ 1102 static int get_cluster_offset(BlockDriverState *bs, 1103 VmdkExtent *extent, 1104 VmdkMetaData *m_data, 1105 uint64_t offset, 1106 bool allocate, 1107 uint64_t *cluster_offset, 1108 uint64_t skip_start_sector, 1109 uint64_t skip_end_sector) 1110 { 1111 unsigned int l1_index, l2_offset, l2_index; 1112 int min_index, i, j; 1113 uint32_t min_count, *l2_table; 1114 bool zeroed = false; 1115 int64_t ret; 1116 int32_t cluster_sector; 1117 1118 if (m_data) { 1119 m_data->valid = 0; 1120 } 1121 if (extent->flat) { 1122 *cluster_offset = extent->flat_start_offset; 1123 return VMDK_OK; 1124 } 1125 1126 offset -= (extent->end_sector - extent->sectors) * SECTOR_SIZE; 1127 l1_index = (offset >> 9) / extent->l1_entry_sectors; 1128 if (l1_index >= extent->l1_size) { 1129 return VMDK_ERROR; 1130 } 1131 l2_offset = extent->l1_table[l1_index]; 1132 if (!l2_offset) { 1133 return VMDK_UNALLOC; 1134 } 1135 for (i = 0; i < L2_CACHE_SIZE; i++) { 1136 if (l2_offset == extent->l2_cache_offsets[i]) { 1137 /* increment the hit count */ 1138 if (++extent->l2_cache_counts[i] == 0xffffffff) { 1139 for (j = 0; j < L2_CACHE_SIZE; j++) { 1140 extent->l2_cache_counts[j] >>= 1; 1141 } 1142 } 1143 l2_table = extent->l2_cache + (i * extent->l2_size); 1144 goto found; 1145 } 1146 } 1147 /* not found: load a new entry in the least used one */ 1148 min_index = 0; 1149 min_count = 0xffffffff; 1150 for (i = 0; i < L2_CACHE_SIZE; i++) { 1151 if (extent->l2_cache_counts[i] < min_count) { 1152 min_count = extent->l2_cache_counts[i]; 1153 min_index = i; 1154 } 1155 } 1156 l2_table = extent->l2_cache + (min_index * extent->l2_size); 1157 if (bdrv_pread( 1158 extent->file, 1159 (int64_t)l2_offset * 512, 1160 l2_table, 1161 extent->l2_size * sizeof(uint32_t) 1162 ) != extent->l2_size * sizeof(uint32_t)) { 1163 return VMDK_ERROR; 1164 } 1165 1166 extent->l2_cache_offsets[min_index] = l2_offset; 1167 extent->l2_cache_counts[min_index] = 1; 1168 found: 1169 l2_index = ((offset >> 9) / extent->cluster_sectors) % extent->l2_size; 1170 cluster_sector = le32_to_cpu(l2_table[l2_index]); 1171 1172 if (m_data) { 1173 m_data->valid = 1; 1174 m_data->l1_index = l1_index; 1175 m_data->l2_index = l2_index; 1176 m_data->l2_offset = l2_offset; 1177 m_data->l2_cache_entry = &l2_table[l2_index]; 1178 } 1179 if (extent->has_zero_grain && cluster_sector == VMDK_GTE_ZEROED) { 1180 zeroed = true; 1181 } 1182 1183 if (!cluster_sector || zeroed) { 1184 if (!allocate) { 1185 return zeroed ? VMDK_ZEROED : VMDK_UNALLOC; 1186 } 1187 1188 cluster_sector = extent->next_cluster_sector; 1189 extent->next_cluster_sector += extent->cluster_sectors; 1190 1191 /* First of all we write grain itself, to avoid race condition 1192 * that may to corrupt the image. 1193 * This problem may occur because of insufficient space on host disk 1194 * or inappropriate VM shutdown. 1195 */ 1196 ret = get_whole_cluster(bs, extent, 1197 cluster_sector, 1198 offset >> BDRV_SECTOR_BITS, 1199 skip_start_sector, skip_end_sector); 1200 if (ret) { 1201 return ret; 1202 } 1203 } 1204 *cluster_offset = cluster_sector << BDRV_SECTOR_BITS; 1205 return VMDK_OK; 1206 } 1207 1208 static VmdkExtent *find_extent(BDRVVmdkState *s, 1209 int64_t sector_num, VmdkExtent *start_hint) 1210 { 1211 VmdkExtent *extent = start_hint; 1212 1213 if (!extent) { 1214 extent = &s->extents[0]; 1215 } 1216 while (extent < &s->extents[s->num_extents]) { 1217 if (sector_num < extent->end_sector) { 1218 return extent; 1219 } 1220 extent++; 1221 } 1222 return NULL; 1223 } 1224 1225 static int64_t coroutine_fn vmdk_co_get_block_status(BlockDriverState *bs, 1226 int64_t sector_num, int nb_sectors, int *pnum) 1227 { 1228 BDRVVmdkState *s = bs->opaque; 1229 int64_t index_in_cluster, n, ret; 1230 uint64_t offset; 1231 VmdkExtent *extent; 1232 1233 extent = find_extent(s, sector_num, NULL); 1234 if (!extent) { 1235 return 0; 1236 } 1237 qemu_co_mutex_lock(&s->lock); 1238 ret = get_cluster_offset(bs, extent, NULL, 1239 sector_num * 512, false, &offset, 1240 0, 0); 1241 qemu_co_mutex_unlock(&s->lock); 1242 1243 switch (ret) { 1244 case VMDK_ERROR: 1245 ret = -EIO; 1246 break; 1247 case VMDK_UNALLOC: 1248 ret = 0; 1249 break; 1250 case VMDK_ZEROED: 1251 ret = BDRV_BLOCK_ZERO; 1252 break; 1253 case VMDK_OK: 1254 ret = BDRV_BLOCK_DATA; 1255 if (extent->file == bs->file && !extent->compressed) { 1256 ret |= BDRV_BLOCK_OFFSET_VALID | offset; 1257 } 1258 1259 break; 1260 } 1261 1262 index_in_cluster = sector_num % extent->cluster_sectors; 1263 n = extent->cluster_sectors - index_in_cluster; 1264 if (n > nb_sectors) { 1265 n = nb_sectors; 1266 } 1267 *pnum = n; 1268 return ret; 1269 } 1270 1271 static int vmdk_write_extent(VmdkExtent *extent, int64_t cluster_offset, 1272 int64_t offset_in_cluster, const uint8_t *buf, 1273 int nb_sectors, int64_t sector_num) 1274 { 1275 int ret; 1276 VmdkGrainMarker *data = NULL; 1277 uLongf buf_len; 1278 const uint8_t *write_buf = buf; 1279 int write_len = nb_sectors * 512; 1280 1281 if (extent->compressed) { 1282 if (!extent->has_marker) { 1283 ret = -EINVAL; 1284 goto out; 1285 } 1286 buf_len = (extent->cluster_sectors << 9) * 2; 1287 data = g_malloc(buf_len + sizeof(VmdkGrainMarker)); 1288 if (compress(data->data, &buf_len, buf, nb_sectors << 9) != Z_OK || 1289 buf_len == 0) { 1290 ret = -EINVAL; 1291 goto out; 1292 } 1293 data->lba = sector_num; 1294 data->size = buf_len; 1295 write_buf = (uint8_t *)data; 1296 write_len = buf_len + sizeof(VmdkGrainMarker); 1297 } 1298 ret = bdrv_pwrite(extent->file, 1299 cluster_offset + offset_in_cluster, 1300 write_buf, 1301 write_len); 1302 if (ret != write_len) { 1303 ret = ret < 0 ? ret : -EIO; 1304 goto out; 1305 } 1306 ret = 0; 1307 out: 1308 g_free(data); 1309 return ret; 1310 } 1311 1312 static int vmdk_read_extent(VmdkExtent *extent, int64_t cluster_offset, 1313 int64_t offset_in_cluster, uint8_t *buf, 1314 int nb_sectors) 1315 { 1316 int ret; 1317 int cluster_bytes, buf_bytes; 1318 uint8_t *cluster_buf, *compressed_data; 1319 uint8_t *uncomp_buf; 1320 uint32_t data_len; 1321 VmdkGrainMarker *marker; 1322 uLongf buf_len; 1323 1324 1325 if (!extent->compressed) { 1326 ret = bdrv_pread(extent->file, 1327 cluster_offset + offset_in_cluster, 1328 buf, nb_sectors * 512); 1329 if (ret == nb_sectors * 512) { 1330 return 0; 1331 } else { 1332 return -EIO; 1333 } 1334 } 1335 cluster_bytes = extent->cluster_sectors * 512; 1336 /* Read two clusters in case GrainMarker + compressed data > one cluster */ 1337 buf_bytes = cluster_bytes * 2; 1338 cluster_buf = g_malloc(buf_bytes); 1339 uncomp_buf = g_malloc(cluster_bytes); 1340 ret = bdrv_pread(extent->file, 1341 cluster_offset, 1342 cluster_buf, buf_bytes); 1343 if (ret < 0) { 1344 goto out; 1345 } 1346 compressed_data = cluster_buf; 1347 buf_len = cluster_bytes; 1348 data_len = cluster_bytes; 1349 if (extent->has_marker) { 1350 marker = (VmdkGrainMarker *)cluster_buf; 1351 compressed_data = marker->data; 1352 data_len = le32_to_cpu(marker->size); 1353 } 1354 if (!data_len || data_len > buf_bytes) { 1355 ret = -EINVAL; 1356 goto out; 1357 } 1358 ret = uncompress(uncomp_buf, &buf_len, compressed_data, data_len); 1359 if (ret != Z_OK) { 1360 ret = -EINVAL; 1361 goto out; 1362 1363 } 1364 if (offset_in_cluster < 0 || 1365 offset_in_cluster + nb_sectors * 512 > buf_len) { 1366 ret = -EINVAL; 1367 goto out; 1368 } 1369 memcpy(buf, uncomp_buf + offset_in_cluster, nb_sectors * 512); 1370 ret = 0; 1371 1372 out: 1373 g_free(uncomp_buf); 1374 g_free(cluster_buf); 1375 return ret; 1376 } 1377 1378 static int vmdk_read(BlockDriverState *bs, int64_t sector_num, 1379 uint8_t *buf, int nb_sectors) 1380 { 1381 BDRVVmdkState *s = bs->opaque; 1382 int ret; 1383 uint64_t n, index_in_cluster; 1384 uint64_t extent_begin_sector, extent_relative_sector_num; 1385 VmdkExtent *extent = NULL; 1386 uint64_t cluster_offset; 1387 1388 while (nb_sectors > 0) { 1389 extent = find_extent(s, sector_num, extent); 1390 if (!extent) { 1391 return -EIO; 1392 } 1393 ret = get_cluster_offset(bs, extent, NULL, 1394 sector_num << 9, false, &cluster_offset, 1395 0, 0); 1396 extent_begin_sector = extent->end_sector - extent->sectors; 1397 extent_relative_sector_num = sector_num - extent_begin_sector; 1398 index_in_cluster = extent_relative_sector_num % extent->cluster_sectors; 1399 n = extent->cluster_sectors - index_in_cluster; 1400 if (n > nb_sectors) { 1401 n = nb_sectors; 1402 } 1403 if (ret != VMDK_OK) { 1404 /* if not allocated, try to read from parent image, if exist */ 1405 if (bs->backing_hd && ret != VMDK_ZEROED) { 1406 if (!vmdk_is_cid_valid(bs)) { 1407 return -EINVAL; 1408 } 1409 ret = bdrv_read(bs->backing_hd, sector_num, buf, n); 1410 if (ret < 0) { 1411 return ret; 1412 } 1413 } else { 1414 memset(buf, 0, 512 * n); 1415 } 1416 } else { 1417 ret = vmdk_read_extent(extent, 1418 cluster_offset, index_in_cluster * 512, 1419 buf, n); 1420 if (ret) { 1421 return ret; 1422 } 1423 } 1424 nb_sectors -= n; 1425 sector_num += n; 1426 buf += n * 512; 1427 } 1428 return 0; 1429 } 1430 1431 static coroutine_fn int vmdk_co_read(BlockDriverState *bs, int64_t sector_num, 1432 uint8_t *buf, int nb_sectors) 1433 { 1434 int ret; 1435 BDRVVmdkState *s = bs->opaque; 1436 qemu_co_mutex_lock(&s->lock); 1437 ret = vmdk_read(bs, sector_num, buf, nb_sectors); 1438 qemu_co_mutex_unlock(&s->lock); 1439 return ret; 1440 } 1441 1442 /** 1443 * vmdk_write: 1444 * @zeroed: buf is ignored (data is zero), use zeroed_grain GTE feature 1445 * if possible, otherwise return -ENOTSUP. 1446 * @zero_dry_run: used for zeroed == true only, don't update L2 table, just try 1447 * with each cluster. By dry run we can find if the zero write 1448 * is possible without modifying image data. 1449 * 1450 * Returns: error code with 0 for success. 1451 */ 1452 static int vmdk_write(BlockDriverState *bs, int64_t sector_num, 1453 const uint8_t *buf, int nb_sectors, 1454 bool zeroed, bool zero_dry_run) 1455 { 1456 BDRVVmdkState *s = bs->opaque; 1457 VmdkExtent *extent = NULL; 1458 int ret; 1459 int64_t index_in_cluster, n; 1460 uint64_t extent_begin_sector, extent_relative_sector_num; 1461 uint64_t cluster_offset; 1462 VmdkMetaData m_data; 1463 1464 if (sector_num > bs->total_sectors) { 1465 error_report("Wrong offset: sector_num=0x%" PRIx64 1466 " total_sectors=0x%" PRIx64 "\n", 1467 sector_num, bs->total_sectors); 1468 return -EIO; 1469 } 1470 1471 while (nb_sectors > 0) { 1472 extent = find_extent(s, sector_num, extent); 1473 if (!extent) { 1474 return -EIO; 1475 } 1476 extent_begin_sector = extent->end_sector - extent->sectors; 1477 extent_relative_sector_num = sector_num - extent_begin_sector; 1478 index_in_cluster = extent_relative_sector_num % extent->cluster_sectors; 1479 n = extent->cluster_sectors - index_in_cluster; 1480 if (n > nb_sectors) { 1481 n = nb_sectors; 1482 } 1483 ret = get_cluster_offset(bs, extent, &m_data, sector_num << 9, 1484 !(extent->compressed || zeroed), 1485 &cluster_offset, 1486 index_in_cluster, index_in_cluster + n); 1487 if (extent->compressed) { 1488 if (ret == VMDK_OK) { 1489 /* Refuse write to allocated cluster for streamOptimized */ 1490 error_report("Could not write to allocated cluster" 1491 " for streamOptimized"); 1492 return -EIO; 1493 } else { 1494 /* allocate */ 1495 ret = get_cluster_offset(bs, extent, &m_data, sector_num << 9, 1496 true, &cluster_offset, 0, 0); 1497 } 1498 } 1499 if (ret == VMDK_ERROR) { 1500 return -EINVAL; 1501 } 1502 if (zeroed) { 1503 /* Do zeroed write, buf is ignored */ 1504 if (extent->has_zero_grain && 1505 index_in_cluster == 0 && 1506 n >= extent->cluster_sectors) { 1507 n = extent->cluster_sectors; 1508 if (!zero_dry_run) { 1509 /* update L2 tables */ 1510 if (vmdk_L2update(extent, &m_data, VMDK_GTE_ZEROED) 1511 != VMDK_OK) { 1512 return -EIO; 1513 } 1514 } 1515 } else { 1516 return -ENOTSUP; 1517 } 1518 } else { 1519 ret = vmdk_write_extent(extent, 1520 cluster_offset, index_in_cluster * 512, 1521 buf, n, sector_num); 1522 if (ret) { 1523 return ret; 1524 } 1525 if (m_data.valid) { 1526 /* update L2 tables */ 1527 if (vmdk_L2update(extent, &m_data, 1528 cluster_offset >> BDRV_SECTOR_BITS) 1529 != VMDK_OK) { 1530 return -EIO; 1531 } 1532 } 1533 } 1534 nb_sectors -= n; 1535 sector_num += n; 1536 buf += n * 512; 1537 1538 /* update CID on the first write every time the virtual disk is 1539 * opened */ 1540 if (!s->cid_updated) { 1541 ret = vmdk_write_cid(bs, time(NULL)); 1542 if (ret < 0) { 1543 return ret; 1544 } 1545 s->cid_updated = true; 1546 } 1547 } 1548 return 0; 1549 } 1550 1551 static coroutine_fn int vmdk_co_write(BlockDriverState *bs, int64_t sector_num, 1552 const uint8_t *buf, int nb_sectors) 1553 { 1554 int ret; 1555 BDRVVmdkState *s = bs->opaque; 1556 qemu_co_mutex_lock(&s->lock); 1557 ret = vmdk_write(bs, sector_num, buf, nb_sectors, false, false); 1558 qemu_co_mutex_unlock(&s->lock); 1559 return ret; 1560 } 1561 1562 static int vmdk_write_compressed(BlockDriverState *bs, 1563 int64_t sector_num, 1564 const uint8_t *buf, 1565 int nb_sectors) 1566 { 1567 BDRVVmdkState *s = bs->opaque; 1568 if (s->num_extents == 1 && s->extents[0].compressed) { 1569 return vmdk_write(bs, sector_num, buf, nb_sectors, false, false); 1570 } else { 1571 return -ENOTSUP; 1572 } 1573 } 1574 1575 static int coroutine_fn vmdk_co_write_zeroes(BlockDriverState *bs, 1576 int64_t sector_num, 1577 int nb_sectors, 1578 BdrvRequestFlags flags) 1579 { 1580 int ret; 1581 BDRVVmdkState *s = bs->opaque; 1582 qemu_co_mutex_lock(&s->lock); 1583 /* write zeroes could fail if sectors not aligned to cluster, test it with 1584 * dry_run == true before really updating image */ 1585 ret = vmdk_write(bs, sector_num, NULL, nb_sectors, true, true); 1586 if (!ret) { 1587 ret = vmdk_write(bs, sector_num, NULL, nb_sectors, true, false); 1588 } 1589 qemu_co_mutex_unlock(&s->lock); 1590 return ret; 1591 } 1592 1593 static int vmdk_create_extent(const char *filename, int64_t filesize, 1594 bool flat, bool compress, bool zeroed_grain, 1595 QemuOpts *opts, Error **errp) 1596 { 1597 int ret, i; 1598 BlockDriverState *bs = NULL; 1599 VMDK4Header header; 1600 Error *local_err = NULL; 1601 uint32_t tmp, magic, grains, gd_sectors, gt_size, gt_count; 1602 uint32_t *gd_buf = NULL; 1603 int gd_buf_size; 1604 1605 ret = bdrv_create_file(filename, opts, &local_err); 1606 if (ret < 0) { 1607 error_propagate(errp, local_err); 1608 goto exit; 1609 } 1610 1611 assert(bs == NULL); 1612 ret = bdrv_open(&bs, filename, NULL, NULL, BDRV_O_RDWR | BDRV_O_PROTOCOL, 1613 NULL, &local_err); 1614 if (ret < 0) { 1615 error_propagate(errp, local_err); 1616 goto exit; 1617 } 1618 1619 if (flat) { 1620 ret = bdrv_truncate(bs, filesize); 1621 if (ret < 0) { 1622 error_setg_errno(errp, -ret, "Could not truncate file"); 1623 } 1624 goto exit; 1625 } 1626 magic = cpu_to_be32(VMDK4_MAGIC); 1627 memset(&header, 0, sizeof(header)); 1628 header.version = zeroed_grain ? 2 : 1; 1629 header.flags = VMDK4_FLAG_RGD | VMDK4_FLAG_NL_DETECT 1630 | (compress ? VMDK4_FLAG_COMPRESS | VMDK4_FLAG_MARKER : 0) 1631 | (zeroed_grain ? VMDK4_FLAG_ZERO_GRAIN : 0); 1632 header.compressAlgorithm = compress ? VMDK4_COMPRESSION_DEFLATE : 0; 1633 header.capacity = filesize / BDRV_SECTOR_SIZE; 1634 header.granularity = 128; 1635 header.num_gtes_per_gt = BDRV_SECTOR_SIZE; 1636 1637 grains = DIV_ROUND_UP(filesize / BDRV_SECTOR_SIZE, header.granularity); 1638 gt_size = DIV_ROUND_UP(header.num_gtes_per_gt * sizeof(uint32_t), 1639 BDRV_SECTOR_SIZE); 1640 gt_count = DIV_ROUND_UP(grains, header.num_gtes_per_gt); 1641 gd_sectors = DIV_ROUND_UP(gt_count * sizeof(uint32_t), BDRV_SECTOR_SIZE); 1642 1643 header.desc_offset = 1; 1644 header.desc_size = 20; 1645 header.rgd_offset = header.desc_offset + header.desc_size; 1646 header.gd_offset = header.rgd_offset + gd_sectors + (gt_size * gt_count); 1647 header.grain_offset = 1648 ROUND_UP(header.gd_offset + gd_sectors + (gt_size * gt_count), 1649 header.granularity); 1650 /* swap endianness for all header fields */ 1651 header.version = cpu_to_le32(header.version); 1652 header.flags = cpu_to_le32(header.flags); 1653 header.capacity = cpu_to_le64(header.capacity); 1654 header.granularity = cpu_to_le64(header.granularity); 1655 header.num_gtes_per_gt = cpu_to_le32(header.num_gtes_per_gt); 1656 header.desc_offset = cpu_to_le64(header.desc_offset); 1657 header.desc_size = cpu_to_le64(header.desc_size); 1658 header.rgd_offset = cpu_to_le64(header.rgd_offset); 1659 header.gd_offset = cpu_to_le64(header.gd_offset); 1660 header.grain_offset = cpu_to_le64(header.grain_offset); 1661 header.compressAlgorithm = cpu_to_le16(header.compressAlgorithm); 1662 1663 header.check_bytes[0] = 0xa; 1664 header.check_bytes[1] = 0x20; 1665 header.check_bytes[2] = 0xd; 1666 header.check_bytes[3] = 0xa; 1667 1668 /* write all the data */ 1669 ret = bdrv_pwrite(bs, 0, &magic, sizeof(magic)); 1670 if (ret < 0) { 1671 error_set(errp, QERR_IO_ERROR); 1672 goto exit; 1673 } 1674 ret = bdrv_pwrite(bs, sizeof(magic), &header, sizeof(header)); 1675 if (ret < 0) { 1676 error_set(errp, QERR_IO_ERROR); 1677 goto exit; 1678 } 1679 1680 ret = bdrv_truncate(bs, le64_to_cpu(header.grain_offset) << 9); 1681 if (ret < 0) { 1682 error_setg_errno(errp, -ret, "Could not truncate file"); 1683 goto exit; 1684 } 1685 1686 /* write grain directory */ 1687 gd_buf_size = gd_sectors * BDRV_SECTOR_SIZE; 1688 gd_buf = g_malloc0(gd_buf_size); 1689 for (i = 0, tmp = le64_to_cpu(header.rgd_offset) + gd_sectors; 1690 i < gt_count; i++, tmp += gt_size) { 1691 gd_buf[i] = cpu_to_le32(tmp); 1692 } 1693 ret = bdrv_pwrite(bs, le64_to_cpu(header.rgd_offset) * BDRV_SECTOR_SIZE, 1694 gd_buf, gd_buf_size); 1695 if (ret < 0) { 1696 error_set(errp, QERR_IO_ERROR); 1697 goto exit; 1698 } 1699 1700 /* write backup grain directory */ 1701 for (i = 0, tmp = le64_to_cpu(header.gd_offset) + gd_sectors; 1702 i < gt_count; i++, tmp += gt_size) { 1703 gd_buf[i] = cpu_to_le32(tmp); 1704 } 1705 ret = bdrv_pwrite(bs, le64_to_cpu(header.gd_offset) * BDRV_SECTOR_SIZE, 1706 gd_buf, gd_buf_size); 1707 if (ret < 0) { 1708 error_set(errp, QERR_IO_ERROR); 1709 goto exit; 1710 } 1711 1712 ret = 0; 1713 exit: 1714 if (bs) { 1715 bdrv_unref(bs); 1716 } 1717 g_free(gd_buf); 1718 return ret; 1719 } 1720 1721 static int filename_decompose(const char *filename, char *path, char *prefix, 1722 char *postfix, size_t buf_len, Error **errp) 1723 { 1724 const char *p, *q; 1725 1726 if (filename == NULL || !strlen(filename)) { 1727 error_setg(errp, "No filename provided"); 1728 return VMDK_ERROR; 1729 } 1730 p = strrchr(filename, '/'); 1731 if (p == NULL) { 1732 p = strrchr(filename, '\\'); 1733 } 1734 if (p == NULL) { 1735 p = strrchr(filename, ':'); 1736 } 1737 if (p != NULL) { 1738 p++; 1739 if (p - filename >= buf_len) { 1740 return VMDK_ERROR; 1741 } 1742 pstrcpy(path, p - filename + 1, filename); 1743 } else { 1744 p = filename; 1745 path[0] = '\0'; 1746 } 1747 q = strrchr(p, '.'); 1748 if (q == NULL) { 1749 pstrcpy(prefix, buf_len, p); 1750 postfix[0] = '\0'; 1751 } else { 1752 if (q - p >= buf_len) { 1753 return VMDK_ERROR; 1754 } 1755 pstrcpy(prefix, q - p + 1, p); 1756 pstrcpy(postfix, buf_len, q); 1757 } 1758 return VMDK_OK; 1759 } 1760 1761 static int vmdk_create(const char *filename, QemuOpts *opts, Error **errp) 1762 { 1763 int idx = 0; 1764 BlockDriverState *new_bs = NULL; 1765 Error *local_err = NULL; 1766 char *desc = NULL; 1767 int64_t total_size = 0, filesize; 1768 char *adapter_type = NULL; 1769 char *backing_file = NULL; 1770 char *fmt = NULL; 1771 int flags = 0; 1772 int ret = 0; 1773 bool flat, split, compress; 1774 GString *ext_desc_lines; 1775 char path[PATH_MAX], prefix[PATH_MAX], postfix[PATH_MAX]; 1776 const int64_t split_size = 0x80000000; /* VMDK has constant split size */ 1777 const char *desc_extent_line; 1778 char parent_desc_line[BUF_SIZE] = ""; 1779 uint32_t parent_cid = 0xffffffff; 1780 uint32_t number_heads = 16; 1781 bool zeroed_grain = false; 1782 uint32_t desc_offset = 0, desc_len; 1783 const char desc_template[] = 1784 "# Disk DescriptorFile\n" 1785 "version=1\n" 1786 "CID=%" PRIx32 "\n" 1787 "parentCID=%" PRIx32 "\n" 1788 "createType=\"%s\"\n" 1789 "%s" 1790 "\n" 1791 "# Extent description\n" 1792 "%s" 1793 "\n" 1794 "# The Disk Data Base\n" 1795 "#DDB\n" 1796 "\n" 1797 "ddb.virtualHWVersion = \"%d\"\n" 1798 "ddb.geometry.cylinders = \"%" PRId64 "\"\n" 1799 "ddb.geometry.heads = \"%" PRIu32 "\"\n" 1800 "ddb.geometry.sectors = \"63\"\n" 1801 "ddb.adapterType = \"%s\"\n"; 1802 1803 ext_desc_lines = g_string_new(NULL); 1804 1805 if (filename_decompose(filename, path, prefix, postfix, PATH_MAX, errp)) { 1806 ret = -EINVAL; 1807 goto exit; 1808 } 1809 /* Read out options */ 1810 total_size = qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0); 1811 adapter_type = qemu_opt_get_del(opts, BLOCK_OPT_ADAPTER_TYPE); 1812 backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE); 1813 if (qemu_opt_get_bool_del(opts, BLOCK_OPT_COMPAT6, false)) { 1814 flags |= BLOCK_FLAG_COMPAT6; 1815 } 1816 fmt = qemu_opt_get_del(opts, BLOCK_OPT_SUBFMT); 1817 if (qemu_opt_get_bool_del(opts, BLOCK_OPT_ZEROED_GRAIN, false)) { 1818 zeroed_grain = true; 1819 } 1820 1821 if (!adapter_type) { 1822 adapter_type = g_strdup("ide"); 1823 } else if (strcmp(adapter_type, "ide") && 1824 strcmp(adapter_type, "buslogic") && 1825 strcmp(adapter_type, "lsilogic") && 1826 strcmp(adapter_type, "legacyESX")) { 1827 error_setg(errp, "Unknown adapter type: '%s'", adapter_type); 1828 ret = -EINVAL; 1829 goto exit; 1830 } 1831 if (strcmp(adapter_type, "ide") != 0) { 1832 /* that's the number of heads with which vmware operates when 1833 creating, exporting, etc. vmdk files with a non-ide adapter type */ 1834 number_heads = 255; 1835 } 1836 if (!fmt) { 1837 /* Default format to monolithicSparse */ 1838 fmt = g_strdup("monolithicSparse"); 1839 } else if (strcmp(fmt, "monolithicFlat") && 1840 strcmp(fmt, "monolithicSparse") && 1841 strcmp(fmt, "twoGbMaxExtentSparse") && 1842 strcmp(fmt, "twoGbMaxExtentFlat") && 1843 strcmp(fmt, "streamOptimized")) { 1844 error_setg(errp, "Unknown subformat: '%s'", fmt); 1845 ret = -EINVAL; 1846 goto exit; 1847 } 1848 split = !(strcmp(fmt, "twoGbMaxExtentFlat") && 1849 strcmp(fmt, "twoGbMaxExtentSparse")); 1850 flat = !(strcmp(fmt, "monolithicFlat") && 1851 strcmp(fmt, "twoGbMaxExtentFlat")); 1852 compress = !strcmp(fmt, "streamOptimized"); 1853 if (flat) { 1854 desc_extent_line = "RW %" PRId64 " FLAT \"%s\" 0\n"; 1855 } else { 1856 desc_extent_line = "RW %" PRId64 " SPARSE \"%s\"\n"; 1857 } 1858 if (flat && backing_file) { 1859 error_setg(errp, "Flat image can't have backing file"); 1860 ret = -ENOTSUP; 1861 goto exit; 1862 } 1863 if (flat && zeroed_grain) { 1864 error_setg(errp, "Flat image can't enable zeroed grain"); 1865 ret = -ENOTSUP; 1866 goto exit; 1867 } 1868 if (backing_file) { 1869 BlockDriverState *bs = NULL; 1870 ret = bdrv_open(&bs, backing_file, NULL, NULL, BDRV_O_NO_BACKING, NULL, 1871 errp); 1872 if (ret != 0) { 1873 goto exit; 1874 } 1875 if (strcmp(bs->drv->format_name, "vmdk")) { 1876 bdrv_unref(bs); 1877 ret = -EINVAL; 1878 goto exit; 1879 } 1880 parent_cid = vmdk_read_cid(bs, 0); 1881 bdrv_unref(bs); 1882 snprintf(parent_desc_line, sizeof(parent_desc_line), 1883 "parentFileNameHint=\"%s\"", backing_file); 1884 } 1885 1886 /* Create extents */ 1887 filesize = total_size; 1888 while (filesize > 0) { 1889 char desc_line[BUF_SIZE]; 1890 char ext_filename[PATH_MAX]; 1891 char desc_filename[PATH_MAX]; 1892 int64_t size = filesize; 1893 1894 if (split && size > split_size) { 1895 size = split_size; 1896 } 1897 if (split) { 1898 snprintf(desc_filename, sizeof(desc_filename), "%s-%c%03d%s", 1899 prefix, flat ? 'f' : 's', ++idx, postfix); 1900 } else if (flat) { 1901 snprintf(desc_filename, sizeof(desc_filename), "%s-flat%s", 1902 prefix, postfix); 1903 } else { 1904 snprintf(desc_filename, sizeof(desc_filename), "%s%s", 1905 prefix, postfix); 1906 } 1907 snprintf(ext_filename, sizeof(ext_filename), "%s%s", 1908 path, desc_filename); 1909 1910 if (vmdk_create_extent(ext_filename, size, 1911 flat, compress, zeroed_grain, opts, errp)) { 1912 ret = -EINVAL; 1913 goto exit; 1914 } 1915 filesize -= size; 1916 1917 /* Format description line */ 1918 snprintf(desc_line, sizeof(desc_line), 1919 desc_extent_line, size / BDRV_SECTOR_SIZE, desc_filename); 1920 g_string_append(ext_desc_lines, desc_line); 1921 } 1922 /* generate descriptor file */ 1923 desc = g_strdup_printf(desc_template, 1924 (uint32_t)time(NULL), 1925 parent_cid, 1926 fmt, 1927 parent_desc_line, 1928 ext_desc_lines->str, 1929 (flags & BLOCK_FLAG_COMPAT6 ? 6 : 4), 1930 total_size / 1931 (int64_t)(63 * number_heads * BDRV_SECTOR_SIZE), 1932 number_heads, 1933 adapter_type); 1934 desc_len = strlen(desc); 1935 /* the descriptor offset = 0x200 */ 1936 if (!split && !flat) { 1937 desc_offset = 0x200; 1938 } else { 1939 ret = bdrv_create_file(filename, opts, &local_err); 1940 if (ret < 0) { 1941 error_propagate(errp, local_err); 1942 goto exit; 1943 } 1944 } 1945 assert(new_bs == NULL); 1946 ret = bdrv_open(&new_bs, filename, NULL, NULL, 1947 BDRV_O_RDWR | BDRV_O_PROTOCOL, NULL, &local_err); 1948 if (ret < 0) { 1949 error_propagate(errp, local_err); 1950 goto exit; 1951 } 1952 ret = bdrv_pwrite(new_bs, desc_offset, desc, desc_len); 1953 if (ret < 0) { 1954 error_setg_errno(errp, -ret, "Could not write description"); 1955 goto exit; 1956 } 1957 /* bdrv_pwrite write padding zeros to align to sector, we don't need that 1958 * for description file */ 1959 if (desc_offset == 0) { 1960 ret = bdrv_truncate(new_bs, desc_len); 1961 if (ret < 0) { 1962 error_setg_errno(errp, -ret, "Could not truncate file"); 1963 } 1964 } 1965 exit: 1966 if (new_bs) { 1967 bdrv_unref(new_bs); 1968 } 1969 g_free(adapter_type); 1970 g_free(backing_file); 1971 g_free(fmt); 1972 g_free(desc); 1973 g_string_free(ext_desc_lines, true); 1974 return ret; 1975 } 1976 1977 static void vmdk_close(BlockDriverState *bs) 1978 { 1979 BDRVVmdkState *s = bs->opaque; 1980 1981 vmdk_free_extents(bs); 1982 g_free(s->create_type); 1983 1984 migrate_del_blocker(s->migration_blocker); 1985 error_free(s->migration_blocker); 1986 } 1987 1988 static coroutine_fn int vmdk_co_flush(BlockDriverState *bs) 1989 { 1990 BDRVVmdkState *s = bs->opaque; 1991 int i, err; 1992 int ret = 0; 1993 1994 for (i = 0; i < s->num_extents; i++) { 1995 err = bdrv_co_flush(s->extents[i].file); 1996 if (err < 0) { 1997 ret = err; 1998 } 1999 } 2000 return ret; 2001 } 2002 2003 static int64_t vmdk_get_allocated_file_size(BlockDriverState *bs) 2004 { 2005 int i; 2006 int64_t ret = 0; 2007 int64_t r; 2008 BDRVVmdkState *s = bs->opaque; 2009 2010 ret = bdrv_get_allocated_file_size(bs->file); 2011 if (ret < 0) { 2012 return ret; 2013 } 2014 for (i = 0; i < s->num_extents; i++) { 2015 if (s->extents[i].file == bs->file) { 2016 continue; 2017 } 2018 r = bdrv_get_allocated_file_size(s->extents[i].file); 2019 if (r < 0) { 2020 return r; 2021 } 2022 ret += r; 2023 } 2024 return ret; 2025 } 2026 2027 static int vmdk_has_zero_init(BlockDriverState *bs) 2028 { 2029 int i; 2030 BDRVVmdkState *s = bs->opaque; 2031 2032 /* If has a flat extent and its underlying storage doesn't have zero init, 2033 * return 0. */ 2034 for (i = 0; i < s->num_extents; i++) { 2035 if (s->extents[i].flat) { 2036 if (!bdrv_has_zero_init(s->extents[i].file)) { 2037 return 0; 2038 } 2039 } 2040 } 2041 return 1; 2042 } 2043 2044 static ImageInfo *vmdk_get_extent_info(VmdkExtent *extent) 2045 { 2046 ImageInfo *info = g_new0(ImageInfo, 1); 2047 2048 *info = (ImageInfo){ 2049 .filename = g_strdup(extent->file->filename), 2050 .format = g_strdup(extent->type), 2051 .virtual_size = extent->sectors * BDRV_SECTOR_SIZE, 2052 .compressed = extent->compressed, 2053 .has_compressed = extent->compressed, 2054 .cluster_size = extent->cluster_sectors * BDRV_SECTOR_SIZE, 2055 .has_cluster_size = !extent->flat, 2056 }; 2057 2058 return info; 2059 } 2060 2061 static int vmdk_check(BlockDriverState *bs, BdrvCheckResult *result, 2062 BdrvCheckMode fix) 2063 { 2064 BDRVVmdkState *s = bs->opaque; 2065 VmdkExtent *extent = NULL; 2066 int64_t sector_num = 0; 2067 int64_t total_sectors = bdrv_nb_sectors(bs); 2068 int ret; 2069 uint64_t cluster_offset; 2070 2071 if (fix) { 2072 return -ENOTSUP; 2073 } 2074 2075 for (;;) { 2076 if (sector_num >= total_sectors) { 2077 return 0; 2078 } 2079 extent = find_extent(s, sector_num, extent); 2080 if (!extent) { 2081 fprintf(stderr, 2082 "ERROR: could not find extent for sector %" PRId64 "\n", 2083 sector_num); 2084 break; 2085 } 2086 ret = get_cluster_offset(bs, extent, NULL, 2087 sector_num << BDRV_SECTOR_BITS, 2088 false, &cluster_offset, 0, 0); 2089 if (ret == VMDK_ERROR) { 2090 fprintf(stderr, 2091 "ERROR: could not get cluster_offset for sector %" 2092 PRId64 "\n", sector_num); 2093 break; 2094 } 2095 if (ret == VMDK_OK && cluster_offset >= bdrv_getlength(extent->file)) { 2096 fprintf(stderr, 2097 "ERROR: cluster offset for sector %" 2098 PRId64 " points after EOF\n", sector_num); 2099 break; 2100 } 2101 sector_num += extent->cluster_sectors; 2102 } 2103 2104 result->corruptions++; 2105 return 0; 2106 } 2107 2108 static ImageInfoSpecific *vmdk_get_specific_info(BlockDriverState *bs) 2109 { 2110 int i; 2111 BDRVVmdkState *s = bs->opaque; 2112 ImageInfoSpecific *spec_info = g_new0(ImageInfoSpecific, 1); 2113 ImageInfoList **next; 2114 2115 *spec_info = (ImageInfoSpecific){ 2116 .kind = IMAGE_INFO_SPECIFIC_KIND_VMDK, 2117 { 2118 .vmdk = g_new0(ImageInfoSpecificVmdk, 1), 2119 }, 2120 }; 2121 2122 *spec_info->vmdk = (ImageInfoSpecificVmdk) { 2123 .create_type = g_strdup(s->create_type), 2124 .cid = s->cid, 2125 .parent_cid = s->parent_cid, 2126 }; 2127 2128 next = &spec_info->vmdk->extents; 2129 for (i = 0; i < s->num_extents; i++) { 2130 *next = g_new0(ImageInfoList, 1); 2131 (*next)->value = vmdk_get_extent_info(&s->extents[i]); 2132 (*next)->next = NULL; 2133 next = &(*next)->next; 2134 } 2135 2136 return spec_info; 2137 } 2138 2139 static int vmdk_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) 2140 { 2141 int i; 2142 BDRVVmdkState *s = bs->opaque; 2143 assert(s->num_extents); 2144 bdi->needs_compressed_writes = s->extents[0].compressed; 2145 if (!s->extents[0].flat) { 2146 bdi->cluster_size = s->extents[0].cluster_sectors << BDRV_SECTOR_BITS; 2147 } 2148 /* See if we have multiple extents but they have different cases */ 2149 for (i = 1; i < s->num_extents; i++) { 2150 if (bdi->needs_compressed_writes != s->extents[i].compressed || 2151 (bdi->cluster_size && bdi->cluster_size != 2152 s->extents[i].cluster_sectors << BDRV_SECTOR_BITS)) { 2153 return -ENOTSUP; 2154 } 2155 } 2156 return 0; 2157 } 2158 2159 static void vmdk_detach_aio_context(BlockDriverState *bs) 2160 { 2161 BDRVVmdkState *s = bs->opaque; 2162 int i; 2163 2164 for (i = 0; i < s->num_extents; i++) { 2165 bdrv_detach_aio_context(s->extents[i].file); 2166 } 2167 } 2168 2169 static void vmdk_attach_aio_context(BlockDriverState *bs, 2170 AioContext *new_context) 2171 { 2172 BDRVVmdkState *s = bs->opaque; 2173 int i; 2174 2175 for (i = 0; i < s->num_extents; i++) { 2176 bdrv_attach_aio_context(s->extents[i].file, new_context); 2177 } 2178 } 2179 2180 static QemuOptsList vmdk_create_opts = { 2181 .name = "vmdk-create-opts", 2182 .head = QTAILQ_HEAD_INITIALIZER(vmdk_create_opts.head), 2183 .desc = { 2184 { 2185 .name = BLOCK_OPT_SIZE, 2186 .type = QEMU_OPT_SIZE, 2187 .help = "Virtual disk size" 2188 }, 2189 { 2190 .name = BLOCK_OPT_ADAPTER_TYPE, 2191 .type = QEMU_OPT_STRING, 2192 .help = "Virtual adapter type, can be one of " 2193 "ide (default), lsilogic, buslogic or legacyESX" 2194 }, 2195 { 2196 .name = BLOCK_OPT_BACKING_FILE, 2197 .type = QEMU_OPT_STRING, 2198 .help = "File name of a base image" 2199 }, 2200 { 2201 .name = BLOCK_OPT_COMPAT6, 2202 .type = QEMU_OPT_BOOL, 2203 .help = "VMDK version 6 image", 2204 .def_value_str = "off" 2205 }, 2206 { 2207 .name = BLOCK_OPT_SUBFMT, 2208 .type = QEMU_OPT_STRING, 2209 .help = 2210 "VMDK flat extent format, can be one of " 2211 "{monolithicSparse (default) | monolithicFlat | twoGbMaxExtentSparse | twoGbMaxExtentFlat | streamOptimized} " 2212 }, 2213 { 2214 .name = BLOCK_OPT_ZEROED_GRAIN, 2215 .type = QEMU_OPT_BOOL, 2216 .help = "Enable efficient zero writes " 2217 "using the zeroed-grain GTE feature" 2218 }, 2219 { /* end of list */ } 2220 } 2221 }; 2222 2223 static BlockDriver bdrv_vmdk = { 2224 .format_name = "vmdk", 2225 .instance_size = sizeof(BDRVVmdkState), 2226 .bdrv_probe = vmdk_probe, 2227 .bdrv_open = vmdk_open, 2228 .bdrv_check = vmdk_check, 2229 .bdrv_reopen_prepare = vmdk_reopen_prepare, 2230 .bdrv_read = vmdk_co_read, 2231 .bdrv_write = vmdk_co_write, 2232 .bdrv_write_compressed = vmdk_write_compressed, 2233 .bdrv_co_write_zeroes = vmdk_co_write_zeroes, 2234 .bdrv_close = vmdk_close, 2235 .bdrv_create = vmdk_create, 2236 .bdrv_co_flush_to_disk = vmdk_co_flush, 2237 .bdrv_co_get_block_status = vmdk_co_get_block_status, 2238 .bdrv_get_allocated_file_size = vmdk_get_allocated_file_size, 2239 .bdrv_has_zero_init = vmdk_has_zero_init, 2240 .bdrv_get_specific_info = vmdk_get_specific_info, 2241 .bdrv_refresh_limits = vmdk_refresh_limits, 2242 .bdrv_get_info = vmdk_get_info, 2243 .bdrv_detach_aio_context = vmdk_detach_aio_context, 2244 .bdrv_attach_aio_context = vmdk_attach_aio_context, 2245 2246 .supports_backing = true, 2247 .create_opts = &vmdk_create_opts, 2248 }; 2249 2250 static void bdrv_vmdk_init(void) 2251 { 2252 bdrv_register(&bdrv_vmdk); 2253 } 2254 2255 block_init(bdrv_vmdk_init); 2256