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