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 #include <glib.h> 32 33 #define VMDK3_MAGIC (('C' << 24) | ('O' << 16) | ('W' << 8) | 'D') 34 #define VMDK4_MAGIC (('K' << 24) | ('D' << 16) | ('M' << 8) | 'V') 35 #define VMDK4_COMPRESSION_DEFLATE 1 36 #define VMDK4_FLAG_NL_DETECT (1 << 0) 37 #define VMDK4_FLAG_RGD (1 << 1) 38 /* Zeroed-grain enable bit */ 39 #define VMDK4_FLAG_ZERO_GRAIN (1 << 2) 40 #define VMDK4_FLAG_COMPRESS (1 << 16) 41 #define VMDK4_FLAG_MARKER (1 << 17) 42 #define VMDK4_GD_AT_END 0xffffffffffffffffULL 43 44 #define VMDK_GTE_ZEROED 0x1 45 46 /* VMDK internal error codes */ 47 #define VMDK_OK 0 48 #define VMDK_ERROR (-1) 49 /* Cluster not allocated */ 50 #define VMDK_UNALLOC (-2) 51 #define VMDK_ZEROED (-3) 52 53 #define BLOCK_OPT_ZEROED_GRAIN "zeroed_grain" 54 55 typedef struct { 56 uint32_t version; 57 uint32_t flags; 58 uint32_t disk_sectors; 59 uint32_t granularity; 60 uint32_t l1dir_offset; 61 uint32_t l1dir_size; 62 uint32_t file_sectors; 63 uint32_t cylinders; 64 uint32_t heads; 65 uint32_t sectors_per_track; 66 } QEMU_PACKED VMDK3Header; 67 68 typedef struct { 69 uint32_t version; 70 uint32_t flags; 71 uint64_t capacity; 72 uint64_t granularity; 73 uint64_t desc_offset; 74 uint64_t desc_size; 75 /* Number of GrainTableEntries per GrainTable */ 76 uint32_t num_gtes_per_gt; 77 uint64_t rgd_offset; 78 uint64_t gd_offset; 79 uint64_t grain_offset; 80 char filler[1]; 81 char check_bytes[4]; 82 uint16_t compressAlgorithm; 83 } QEMU_PACKED VMDK4Header; 84 85 #define L2_CACHE_SIZE 16 86 87 typedef struct VmdkExtent { 88 BlockDriverState *file; 89 bool flat; 90 bool compressed; 91 bool has_marker; 92 bool has_zero_grain; 93 int version; 94 int64_t sectors; 95 int64_t end_sector; 96 int64_t flat_start_offset; 97 int64_t l1_table_offset; 98 int64_t l1_backup_table_offset; 99 uint32_t *l1_table; 100 uint32_t *l1_backup_table; 101 unsigned int l1_size; 102 uint32_t l1_entry_sectors; 103 104 unsigned int l2_size; 105 uint32_t *l2_cache; 106 uint32_t l2_cache_offsets[L2_CACHE_SIZE]; 107 uint32_t l2_cache_counts[L2_CACHE_SIZE]; 108 109 int64_t cluster_sectors; 110 int64_t next_cluster_sector; 111 char *type; 112 } VmdkExtent; 113 114 typedef struct BDRVVmdkState { 115 CoMutex lock; 116 uint64_t desc_offset; 117 bool cid_updated; 118 bool cid_checked; 119 uint32_t cid; 120 uint32_t parent_cid; 121 int num_extents; 122 /* Extent array with num_extents entries, ascend ordered by address */ 123 VmdkExtent *extents; 124 Error *migration_blocker; 125 char *create_type; 126 } BDRVVmdkState; 127 128 typedef struct VmdkMetaData { 129 unsigned int l1_index; 130 unsigned int l2_index; 131 unsigned int l2_offset; 132 int valid; 133 uint32_t *l2_cache_entry; 134 } VmdkMetaData; 135 136 typedef struct VmdkGrainMarker { 137 uint64_t lba; 138 uint32_t size; 139 uint8_t data[0]; 140 } QEMU_PACKED VmdkGrainMarker; 141 142 enum { 143 MARKER_END_OF_STREAM = 0, 144 MARKER_GRAIN_TABLE = 1, 145 MARKER_GRAIN_DIRECTORY = 2, 146 MARKER_FOOTER = 3, 147 }; 148 149 static int vmdk_probe(const uint8_t *buf, int buf_size, const char *filename) 150 { 151 uint32_t magic; 152 153 if (buf_size < 4) { 154 return 0; 155 } 156 magic = be32_to_cpu(*(uint32_t *)buf); 157 if (magic == VMDK3_MAGIC || 158 magic == VMDK4_MAGIC) { 159 return 100; 160 } else { 161 const char *p = (const char *)buf; 162 const char *end = p + buf_size; 163 while (p < end) { 164 if (*p == '#') { 165 /* skip comment line */ 166 while (p < end && *p != '\n') { 167 p++; 168 } 169 p++; 170 continue; 171 } 172 if (*p == ' ') { 173 while (p < end && *p == ' ') { 174 p++; 175 } 176 /* skip '\r' if windows line endings used. */ 177 if (p < end && *p == '\r') { 178 p++; 179 } 180 /* only accept blank lines before 'version=' line */ 181 if (p == end || *p != '\n') { 182 return 0; 183 } 184 p++; 185 continue; 186 } 187 if (end - p >= strlen("version=X\n")) { 188 if (strncmp("version=1\n", p, strlen("version=1\n")) == 0 || 189 strncmp("version=2\n", p, strlen("version=2\n")) == 0) { 190 return 100; 191 } 192 } 193 if (end - p >= strlen("version=X\r\n")) { 194 if (strncmp("version=1\r\n", p, strlen("version=1\r\n")) == 0 || 195 strncmp("version=2\r\n", p, strlen("version=2\r\n")) == 0) { 196 return 100; 197 } 198 } 199 return 0; 200 } 201 return 0; 202 } 203 } 204 205 #define SECTOR_SIZE 512 206 #define DESC_SIZE (20 * SECTOR_SIZE) /* 20 sectors of 512 bytes each */ 207 #define BUF_SIZE 4096 208 #define HEADER_SIZE 512 /* first sector of 512 bytes */ 209 210 static void vmdk_free_extents(BlockDriverState *bs) 211 { 212 int i; 213 BDRVVmdkState *s = bs->opaque; 214 VmdkExtent *e; 215 216 for (i = 0; i < s->num_extents; i++) { 217 e = &s->extents[i]; 218 g_free(e->l1_table); 219 g_free(e->l2_cache); 220 g_free(e->l1_backup_table); 221 g_free(e->type); 222 if (e->file != bs->file) { 223 bdrv_unref(e->file); 224 } 225 } 226 g_free(s->extents); 227 } 228 229 static void vmdk_free_last_extent(BlockDriverState *bs) 230 { 231 BDRVVmdkState *s = bs->opaque; 232 233 if (s->num_extents == 0) { 234 return; 235 } 236 s->num_extents--; 237 s->extents = g_renew(VmdkExtent, s->extents, s->num_extents); 238 } 239 240 static uint32_t vmdk_read_cid(BlockDriverState *bs, int parent) 241 { 242 char desc[DESC_SIZE]; 243 uint32_t cid = 0xffffffff; 244 const char *p_name, *cid_str; 245 size_t cid_str_size; 246 BDRVVmdkState *s = bs->opaque; 247 int ret; 248 249 ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE); 250 if (ret < 0) { 251 return 0; 252 } 253 254 if (parent) { 255 cid_str = "parentCID"; 256 cid_str_size = sizeof("parentCID"); 257 } else { 258 cid_str = "CID"; 259 cid_str_size = sizeof("CID"); 260 } 261 262 desc[DESC_SIZE - 1] = '\0'; 263 p_name = strstr(desc, cid_str); 264 if (p_name != NULL) { 265 p_name += cid_str_size; 266 sscanf(p_name, "%" SCNx32, &cid); 267 } 268 269 return cid; 270 } 271 272 static int vmdk_write_cid(BlockDriverState *bs, uint32_t cid) 273 { 274 char desc[DESC_SIZE], tmp_desc[DESC_SIZE]; 275 char *p_name, *tmp_str; 276 BDRVVmdkState *s = bs->opaque; 277 int ret; 278 279 ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE); 280 if (ret < 0) { 281 return ret; 282 } 283 284 desc[DESC_SIZE - 1] = '\0'; 285 tmp_str = strstr(desc, "parentCID"); 286 if (tmp_str == NULL) { 287 return -EINVAL; 288 } 289 290 pstrcpy(tmp_desc, sizeof(tmp_desc), tmp_str); 291 p_name = strstr(desc, "CID"); 292 if (p_name != NULL) { 293 p_name += sizeof("CID"); 294 snprintf(p_name, sizeof(desc) - (p_name - desc), "%" PRIx32 "\n", cid); 295 pstrcat(desc, sizeof(desc), tmp_desc); 296 } 297 298 ret = bdrv_pwrite_sync(bs->file, s->desc_offset, desc, DESC_SIZE); 299 if (ret < 0) { 300 return ret; 301 } 302 303 return 0; 304 } 305 306 static int vmdk_is_cid_valid(BlockDriverState *bs) 307 { 308 BDRVVmdkState *s = bs->opaque; 309 BlockDriverState *p_bs = bs->backing_hd; 310 uint32_t cur_pcid; 311 312 if (!s->cid_checked && p_bs) { 313 cur_pcid = vmdk_read_cid(p_bs, 0); 314 if (s->parent_cid != cur_pcid) { 315 /* CID not valid */ 316 return 0; 317 } 318 } 319 s->cid_checked = true; 320 /* CID valid */ 321 return 1; 322 } 323 324 /* Queue extents, if any, for reopen() */ 325 static int vmdk_reopen_prepare(BDRVReopenState *state, 326 BlockReopenQueue *queue, Error **errp) 327 { 328 BDRVVmdkState *s; 329 int ret = -1; 330 int i; 331 VmdkExtent *e; 332 333 assert(state != NULL); 334 assert(state->bs != NULL); 335 336 if (queue == NULL) { 337 error_setg(errp, "No reopen queue for VMDK extents"); 338 goto exit; 339 } 340 341 s = state->bs->opaque; 342 343 assert(s != NULL); 344 345 for (i = 0; i < s->num_extents; i++) { 346 e = &s->extents[i]; 347 if (e->file != state->bs->file) { 348 bdrv_reopen_queue(queue, e->file, state->flags); 349 } 350 } 351 ret = 0; 352 353 exit: 354 return ret; 355 } 356 357 static int vmdk_parent_open(BlockDriverState *bs) 358 { 359 char *p_name; 360 char desc[DESC_SIZE + 1]; 361 BDRVVmdkState *s = bs->opaque; 362 int ret; 363 364 desc[DESC_SIZE] = '\0'; 365 ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE); 366 if (ret < 0) { 367 return ret; 368 } 369 370 p_name = strstr(desc, "parentFileNameHint"); 371 if (p_name != NULL) { 372 char *end_name; 373 374 p_name += sizeof("parentFileNameHint") + 1; 375 end_name = strchr(p_name, '\"'); 376 if (end_name == NULL) { 377 return -EINVAL; 378 } 379 if ((end_name - p_name) > sizeof(bs->backing_file) - 1) { 380 return -EINVAL; 381 } 382 383 pstrcpy(bs->backing_file, end_name - p_name + 1, p_name); 384 } 385 386 return 0; 387 } 388 389 /* Create and append extent to the extent array. Return the added VmdkExtent 390 * address. return NULL if allocation failed. */ 391 static int vmdk_add_extent(BlockDriverState *bs, 392 BlockDriverState *file, bool flat, int64_t sectors, 393 int64_t l1_offset, int64_t l1_backup_offset, 394 uint32_t l1_size, 395 int l2_size, uint64_t cluster_sectors, 396 VmdkExtent **new_extent, 397 Error **errp) 398 { 399 VmdkExtent *extent; 400 BDRVVmdkState *s = bs->opaque; 401 int64_t nb_sectors; 402 403 if (cluster_sectors > 0x200000) { 404 /* 0x200000 * 512Bytes = 1GB for one cluster is unrealistic */ 405 error_setg(errp, "Invalid granularity, image may be corrupt"); 406 return -EFBIG; 407 } 408 if (l1_size > 512 * 1024 * 1024) { 409 /* Although with big capacity and small l1_entry_sectors, we can get a 410 * big l1_size, we don't want unbounded value to allocate the table. 411 * Limit it to 512M, which is 16PB for default cluster and L2 table 412 * size */ 413 error_setg(errp, "L1 size too big"); 414 return -EFBIG; 415 } 416 417 nb_sectors = bdrv_nb_sectors(file); 418 if (nb_sectors < 0) { 419 return nb_sectors; 420 } 421 422 s->extents = g_renew(VmdkExtent, s->extents, s->num_extents + 1); 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 = ROUND_UP(nb_sectors, 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 size_t l1_size; 455 int 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_new(uint32_t, extent->l2_size * L2_CACHE_SIZE); 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 (int64_t)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 if (size < 4) { 562 /* Both descriptor file and sparse image must be much larger than 4 563 * bytes, also callers of vmdk_read_desc want to compare the first 4 564 * bytes with VMDK4_MAGIC, let's error out if less is read. */ 565 error_setg(errp, "File is too small, not a valid image"); 566 return NULL; 567 } 568 569 size = MIN(size, (1 << 20) - 1); /* avoid unbounded allocation */ 570 buf = g_malloc(size + 1); 571 572 ret = bdrv_pread(file, desc_offset, buf, size); 573 if (ret < 0) { 574 error_setg_errno(errp, -ret, "Could not read from file"); 575 g_free(buf); 576 return NULL; 577 } 578 buf[ret] = 0; 579 580 return buf; 581 } 582 583 static int vmdk_open_vmdk4(BlockDriverState *bs, 584 BlockDriverState *file, 585 int flags, Error **errp) 586 { 587 int ret; 588 uint32_t magic; 589 uint32_t l1_size, l1_entry_sectors; 590 VMDK4Header header; 591 VmdkExtent *extent; 592 BDRVVmdkState *s = bs->opaque; 593 int64_t l1_backup_offset = 0; 594 595 ret = bdrv_pread(file, sizeof(magic), &header, sizeof(header)); 596 if (ret < 0) { 597 error_setg_errno(errp, -ret, 598 "Could not read header from file '%s'", 599 file->filename); 600 return -EINVAL; 601 } 602 if (header.capacity == 0) { 603 uint64_t desc_offset = le64_to_cpu(header.desc_offset); 604 if (desc_offset) { 605 char *buf = vmdk_read_desc(file, desc_offset << 9, errp); 606 if (!buf) { 607 return -EINVAL; 608 } 609 ret = vmdk_open_desc_file(bs, flags, buf, errp); 610 g_free(buf); 611 return ret; 612 } 613 } 614 615 if (!s->create_type) { 616 s->create_type = g_strdup("monolithicSparse"); 617 } 618 619 if (le64_to_cpu(header.gd_offset) == VMDK4_GD_AT_END) { 620 /* 621 * The footer takes precedence over the header, so read it in. The 622 * footer starts at offset -1024 from the end: One sector for the 623 * footer, and another one for the end-of-stream marker. 624 */ 625 struct { 626 struct { 627 uint64_t val; 628 uint32_t size; 629 uint32_t type; 630 uint8_t pad[512 - 16]; 631 } QEMU_PACKED footer_marker; 632 633 uint32_t magic; 634 VMDK4Header header; 635 uint8_t pad[512 - 4 - sizeof(VMDK4Header)]; 636 637 struct { 638 uint64_t val; 639 uint32_t size; 640 uint32_t type; 641 uint8_t pad[512 - 16]; 642 } QEMU_PACKED eos_marker; 643 } QEMU_PACKED footer; 644 645 ret = bdrv_pread(file, 646 bs->file->total_sectors * 512 - 1536, 647 &footer, sizeof(footer)); 648 if (ret < 0) { 649 error_setg_errno(errp, -ret, "Failed to read footer"); 650 return ret; 651 } 652 653 /* Some sanity checks for the footer */ 654 if (be32_to_cpu(footer.magic) != VMDK4_MAGIC || 655 le32_to_cpu(footer.footer_marker.size) != 0 || 656 le32_to_cpu(footer.footer_marker.type) != MARKER_FOOTER || 657 le64_to_cpu(footer.eos_marker.val) != 0 || 658 le32_to_cpu(footer.eos_marker.size) != 0 || 659 le32_to_cpu(footer.eos_marker.type) != MARKER_END_OF_STREAM) 660 { 661 error_setg(errp, "Invalid footer"); 662 return -EINVAL; 663 } 664 665 header = footer.header; 666 } 667 668 if (le32_to_cpu(header.version) > 3) { 669 char buf[64]; 670 snprintf(buf, sizeof(buf), "VMDK version %" PRId32, 671 le32_to_cpu(header.version)); 672 error_set(errp, QERR_UNKNOWN_BLOCK_FORMAT_FEATURE, 673 bdrv_get_device_or_node_name(bs), "vmdk", buf); 674 return -ENOTSUP; 675 } else if (le32_to_cpu(header.version) == 3 && (flags & BDRV_O_RDWR)) { 676 /* VMware KB 2064959 explains that version 3 added support for 677 * persistent changed block tracking (CBT), and backup software can 678 * read it as version=1 if it doesn't care about the changed area 679 * information. So we are safe to enable read only. */ 680 error_setg(errp, "VMDK version 3 must be read only"); 681 return -EINVAL; 682 } 683 684 if (le32_to_cpu(header.num_gtes_per_gt) > 512) { 685 error_setg(errp, "L2 table size too big"); 686 return -EINVAL; 687 } 688 689 l1_entry_sectors = le32_to_cpu(header.num_gtes_per_gt) 690 * le64_to_cpu(header.granularity); 691 if (l1_entry_sectors == 0) { 692 error_setg(errp, "L1 entry size is invalid"); 693 return -EINVAL; 694 } 695 l1_size = (le64_to_cpu(header.capacity) + l1_entry_sectors - 1) 696 / l1_entry_sectors; 697 if (le32_to_cpu(header.flags) & VMDK4_FLAG_RGD) { 698 l1_backup_offset = le64_to_cpu(header.rgd_offset) << 9; 699 } 700 if (bdrv_nb_sectors(file) < le64_to_cpu(header.grain_offset)) { 701 error_setg(errp, "File truncated, expecting at least %" PRId64 " bytes", 702 (int64_t)(le64_to_cpu(header.grain_offset) 703 * BDRV_SECTOR_SIZE)); 704 return -EINVAL; 705 } 706 707 ret = vmdk_add_extent(bs, file, false, 708 le64_to_cpu(header.capacity), 709 le64_to_cpu(header.gd_offset) << 9, 710 l1_backup_offset, 711 l1_size, 712 le32_to_cpu(header.num_gtes_per_gt), 713 le64_to_cpu(header.granularity), 714 &extent, 715 errp); 716 if (ret < 0) { 717 return ret; 718 } 719 extent->compressed = 720 le16_to_cpu(header.compressAlgorithm) == VMDK4_COMPRESSION_DEFLATE; 721 if (extent->compressed) { 722 g_free(s->create_type); 723 s->create_type = g_strdup("streamOptimized"); 724 } 725 extent->has_marker = le32_to_cpu(header.flags) & VMDK4_FLAG_MARKER; 726 extent->version = le32_to_cpu(header.version); 727 extent->has_zero_grain = le32_to_cpu(header.flags) & VMDK4_FLAG_ZERO_GRAIN; 728 ret = vmdk_init_tables(bs, extent, errp); 729 if (ret) { 730 /* free extent allocated by vmdk_add_extent */ 731 vmdk_free_last_extent(bs); 732 } 733 return ret; 734 } 735 736 /* find an option value out of descriptor file */ 737 static int vmdk_parse_description(const char *desc, const char *opt_name, 738 char *buf, int buf_size) 739 { 740 char *opt_pos, *opt_end; 741 const char *end = desc + strlen(desc); 742 743 opt_pos = strstr(desc, opt_name); 744 if (!opt_pos) { 745 return VMDK_ERROR; 746 } 747 /* Skip "=\"" following opt_name */ 748 opt_pos += strlen(opt_name) + 2; 749 if (opt_pos >= end) { 750 return VMDK_ERROR; 751 } 752 opt_end = opt_pos; 753 while (opt_end < end && *opt_end != '"') { 754 opt_end++; 755 } 756 if (opt_end == end || buf_size < opt_end - opt_pos + 1) { 757 return VMDK_ERROR; 758 } 759 pstrcpy(buf, opt_end - opt_pos + 1, opt_pos); 760 return VMDK_OK; 761 } 762 763 /* Open an extent file and append to bs array */ 764 static int vmdk_open_sparse(BlockDriverState *bs, 765 BlockDriverState *file, int flags, 766 char *buf, Error **errp) 767 { 768 uint32_t magic; 769 770 magic = ldl_be_p(buf); 771 switch (magic) { 772 case VMDK3_MAGIC: 773 return vmdk_open_vmfs_sparse(bs, file, flags, errp); 774 break; 775 case VMDK4_MAGIC: 776 return vmdk_open_vmdk4(bs, file, flags, errp); 777 break; 778 default: 779 error_setg(errp, "Image not in VMDK format"); 780 return -EINVAL; 781 break; 782 } 783 } 784 785 static int vmdk_parse_extents(const char *desc, BlockDriverState *bs, 786 const char *desc_file_path, Error **errp) 787 { 788 int ret; 789 int matches; 790 char access[11]; 791 char type[11]; 792 char fname[512]; 793 const char *p = desc; 794 int64_t sectors = 0; 795 int64_t flat_offset; 796 char *extent_path; 797 BlockDriverState *extent_file; 798 BDRVVmdkState *s = bs->opaque; 799 VmdkExtent *extent; 800 801 while (*p) { 802 /* parse extent line in one of below formats: 803 * 804 * RW [size in sectors] FLAT "file-name.vmdk" OFFSET 805 * RW [size in sectors] SPARSE "file-name.vmdk" 806 * RW [size in sectors] VMFS "file-name.vmdk" 807 * RW [size in sectors] VMFSSPARSE "file-name.vmdk" 808 */ 809 flat_offset = -1; 810 matches = sscanf(p, "%10s %" SCNd64 " %10s \"%511[^\n\r\"]\" %" SCNd64, 811 access, §ors, type, fname, &flat_offset); 812 if (matches < 4 || strcmp(access, "RW")) { 813 goto next_line; 814 } else if (!strcmp(type, "FLAT")) { 815 if (matches != 5 || flat_offset < 0) { 816 error_setg(errp, "Invalid extent lines: \n%s", p); 817 return -EINVAL; 818 } 819 } else if (!strcmp(type, "VMFS")) { 820 if (matches == 4) { 821 flat_offset = 0; 822 } else { 823 error_setg(errp, "Invalid extent lines:\n%s", p); 824 return -EINVAL; 825 } 826 } else if (matches != 4) { 827 error_setg(errp, "Invalid extent lines:\n%s", p); 828 return -EINVAL; 829 } 830 831 if (sectors <= 0 || 832 (strcmp(type, "FLAT") && strcmp(type, "SPARSE") && 833 strcmp(type, "VMFS") && strcmp(type, "VMFSSPARSE")) || 834 (strcmp(access, "RW"))) { 835 goto next_line; 836 } 837 838 if (!path_is_absolute(fname) && !path_has_protocol(fname) && 839 !desc_file_path[0]) 840 { 841 error_setg(errp, "Cannot use relative extent paths with VMDK " 842 "descriptor file '%s'", bs->file->filename); 843 return -EINVAL; 844 } 845 846 extent_path = g_malloc0(PATH_MAX); 847 path_combine(extent_path, PATH_MAX, desc_file_path, fname); 848 extent_file = NULL; 849 ret = bdrv_open(&extent_file, extent_path, NULL, NULL, 850 bs->open_flags | BDRV_O_PROTOCOL, NULL, errp); 851 g_free(extent_path); 852 if (ret) { 853 return ret; 854 } 855 856 /* save to extents array */ 857 if (!strcmp(type, "FLAT") || !strcmp(type, "VMFS")) { 858 /* FLAT extent */ 859 860 ret = vmdk_add_extent(bs, extent_file, true, sectors, 861 0, 0, 0, 0, 0, &extent, errp); 862 if (ret < 0) { 863 bdrv_unref(extent_file); 864 return ret; 865 } 866 extent->flat_start_offset = flat_offset << 9; 867 } else if (!strcmp(type, "SPARSE") || !strcmp(type, "VMFSSPARSE")) { 868 /* SPARSE extent and VMFSSPARSE extent are both "COWD" sparse file*/ 869 char *buf = vmdk_read_desc(extent_file, 0, errp); 870 if (!buf) { 871 ret = -EINVAL; 872 } else { 873 ret = vmdk_open_sparse(bs, extent_file, bs->open_flags, buf, errp); 874 } 875 g_free(buf); 876 if (ret) { 877 bdrv_unref(extent_file); 878 return ret; 879 } 880 extent = &s->extents[s->num_extents - 1]; 881 } else { 882 error_setg(errp, "Unsupported extent type '%s'", type); 883 bdrv_unref(extent_file); 884 return -ENOTSUP; 885 } 886 extent->type = g_strdup(type); 887 next_line: 888 /* move to next line */ 889 while (*p) { 890 if (*p == '\n') { 891 p++; 892 break; 893 } 894 p++; 895 } 896 } 897 return 0; 898 } 899 900 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf, 901 Error **errp) 902 { 903 int ret; 904 char ct[128]; 905 BDRVVmdkState *s = bs->opaque; 906 907 if (vmdk_parse_description(buf, "createType", ct, sizeof(ct))) { 908 error_setg(errp, "invalid VMDK image descriptor"); 909 ret = -EINVAL; 910 goto exit; 911 } 912 if (strcmp(ct, "monolithicFlat") && 913 strcmp(ct, "vmfs") && 914 strcmp(ct, "vmfsSparse") && 915 strcmp(ct, "twoGbMaxExtentSparse") && 916 strcmp(ct, "twoGbMaxExtentFlat")) { 917 error_setg(errp, "Unsupported image type '%s'", ct); 918 ret = -ENOTSUP; 919 goto exit; 920 } 921 s->create_type = g_strdup(ct); 922 s->desc_offset = 0; 923 ret = vmdk_parse_extents(buf, bs, bs->file->exact_filename, errp); 924 exit: 925 return ret; 926 } 927 928 static int vmdk_open(BlockDriverState *bs, QDict *options, int flags, 929 Error **errp) 930 { 931 char *buf; 932 int ret; 933 BDRVVmdkState *s = bs->opaque; 934 uint32_t magic; 935 936 buf = vmdk_read_desc(bs->file, 0, errp); 937 if (!buf) { 938 return -EINVAL; 939 } 940 941 magic = ldl_be_p(buf); 942 switch (magic) { 943 case VMDK3_MAGIC: 944 case VMDK4_MAGIC: 945 ret = vmdk_open_sparse(bs, bs->file, flags, buf, errp); 946 s->desc_offset = 0x200; 947 break; 948 default: 949 ret = vmdk_open_desc_file(bs, flags, buf, errp); 950 break; 951 } 952 if (ret) { 953 goto fail; 954 } 955 956 /* try to open parent images, if exist */ 957 ret = vmdk_parent_open(bs); 958 if (ret) { 959 goto fail; 960 } 961 s->cid = vmdk_read_cid(bs, 0); 962 s->parent_cid = vmdk_read_cid(bs, 1); 963 qemu_co_mutex_init(&s->lock); 964 965 /* Disable migration when VMDK images are used */ 966 error_setg(&s->migration_blocker, "The vmdk format used by node '%s' " 967 "does not support live migration", 968 bdrv_get_device_or_node_name(bs)); 969 migrate_add_blocker(s->migration_blocker); 970 g_free(buf); 971 return 0; 972 973 fail: 974 g_free(buf); 975 g_free(s->create_type); 976 s->create_type = NULL; 977 vmdk_free_extents(bs); 978 return ret; 979 } 980 981 982 static void vmdk_refresh_limits(BlockDriverState *bs, Error **errp) 983 { 984 BDRVVmdkState *s = bs->opaque; 985 int i; 986 987 for (i = 0; i < s->num_extents; i++) { 988 if (!s->extents[i].flat) { 989 bs->bl.write_zeroes_alignment = 990 MAX(bs->bl.write_zeroes_alignment, 991 s->extents[i].cluster_sectors); 992 } 993 } 994 } 995 996 /** 997 * get_whole_cluster 998 * 999 * Copy backing file's cluster that covers @sector_num, otherwise write zero, 1000 * to the cluster at @cluster_sector_num. 1001 * 1002 * If @skip_start_sector < @skip_end_sector, the relative range 1003 * [@skip_start_sector, @skip_end_sector) is not copied or written, and leave 1004 * it for call to write user data in the request. 1005 */ 1006 static int get_whole_cluster(BlockDriverState *bs, 1007 VmdkExtent *extent, 1008 uint64_t cluster_sector_num, 1009 uint64_t sector_num, 1010 uint64_t skip_start_sector, 1011 uint64_t skip_end_sector) 1012 { 1013 int ret = VMDK_OK; 1014 int64_t cluster_bytes; 1015 uint8_t *whole_grain; 1016 1017 /* For COW, align request sector_num to cluster start */ 1018 sector_num = QEMU_ALIGN_DOWN(sector_num, extent->cluster_sectors); 1019 cluster_bytes = extent->cluster_sectors << BDRV_SECTOR_BITS; 1020 whole_grain = qemu_blockalign(bs, cluster_bytes); 1021 1022 if (!bs->backing_hd) { 1023 memset(whole_grain, 0, skip_start_sector << BDRV_SECTOR_BITS); 1024 memset(whole_grain + (skip_end_sector << BDRV_SECTOR_BITS), 0, 1025 cluster_bytes - (skip_end_sector << BDRV_SECTOR_BITS)); 1026 } 1027 1028 assert(skip_end_sector <= extent->cluster_sectors); 1029 /* we will be here if it's first write on non-exist grain(cluster). 1030 * try to read from parent image, if exist */ 1031 if (bs->backing_hd && !vmdk_is_cid_valid(bs)) { 1032 ret = VMDK_ERROR; 1033 goto exit; 1034 } 1035 1036 /* Read backing data before skip range */ 1037 if (skip_start_sector > 0) { 1038 if (bs->backing_hd) { 1039 ret = bdrv_read(bs->backing_hd, sector_num, 1040 whole_grain, skip_start_sector); 1041 if (ret < 0) { 1042 ret = VMDK_ERROR; 1043 goto exit; 1044 } 1045 } 1046 ret = bdrv_write(extent->file, cluster_sector_num, whole_grain, 1047 skip_start_sector); 1048 if (ret < 0) { 1049 ret = VMDK_ERROR; 1050 goto exit; 1051 } 1052 } 1053 /* Read backing data after skip range */ 1054 if (skip_end_sector < extent->cluster_sectors) { 1055 if (bs->backing_hd) { 1056 ret = bdrv_read(bs->backing_hd, sector_num + skip_end_sector, 1057 whole_grain + (skip_end_sector << BDRV_SECTOR_BITS), 1058 extent->cluster_sectors - skip_end_sector); 1059 if (ret < 0) { 1060 ret = VMDK_ERROR; 1061 goto exit; 1062 } 1063 } 1064 ret = bdrv_write(extent->file, cluster_sector_num + skip_end_sector, 1065 whole_grain + (skip_end_sector << BDRV_SECTOR_BITS), 1066 extent->cluster_sectors - skip_end_sector); 1067 if (ret < 0) { 1068 ret = VMDK_ERROR; 1069 goto exit; 1070 } 1071 } 1072 1073 exit: 1074 qemu_vfree(whole_grain); 1075 return ret; 1076 } 1077 1078 static int vmdk_L2update(VmdkExtent *extent, VmdkMetaData *m_data, 1079 uint32_t offset) 1080 { 1081 offset = cpu_to_le32(offset); 1082 /* update L2 table */ 1083 if (bdrv_pwrite_sync( 1084 extent->file, 1085 ((int64_t)m_data->l2_offset * 512) 1086 + (m_data->l2_index * sizeof(offset)), 1087 &offset, sizeof(offset)) < 0) { 1088 return VMDK_ERROR; 1089 } 1090 /* update backup L2 table */ 1091 if (extent->l1_backup_table_offset != 0) { 1092 m_data->l2_offset = extent->l1_backup_table[m_data->l1_index]; 1093 if (bdrv_pwrite_sync( 1094 extent->file, 1095 ((int64_t)m_data->l2_offset * 512) 1096 + (m_data->l2_index * sizeof(offset)), 1097 &offset, sizeof(offset)) < 0) { 1098 return VMDK_ERROR; 1099 } 1100 } 1101 if (m_data->l2_cache_entry) { 1102 *m_data->l2_cache_entry = offset; 1103 } 1104 1105 return VMDK_OK; 1106 } 1107 1108 /** 1109 * get_cluster_offset 1110 * 1111 * Look up cluster offset in extent file by sector number, and store in 1112 * @cluster_offset. 1113 * 1114 * For flat extents, the start offset as parsed from the description file is 1115 * returned. 1116 * 1117 * For sparse extents, look up in L1, L2 table. If allocate is true, return an 1118 * offset for a new cluster and update L2 cache. If there is a backing file, 1119 * COW is done before returning; otherwise, zeroes are written to the allocated 1120 * cluster. Both COW and zero writing skips the sector range 1121 * [@skip_start_sector, @skip_end_sector) passed in by caller, because caller 1122 * has new data to write there. 1123 * 1124 * Returns: VMDK_OK if cluster exists and mapped in the image. 1125 * VMDK_UNALLOC if cluster is not mapped and @allocate is false. 1126 * VMDK_ERROR if failed. 1127 */ 1128 static int get_cluster_offset(BlockDriverState *bs, 1129 VmdkExtent *extent, 1130 VmdkMetaData *m_data, 1131 uint64_t offset, 1132 bool allocate, 1133 uint64_t *cluster_offset, 1134 uint64_t skip_start_sector, 1135 uint64_t skip_end_sector) 1136 { 1137 unsigned int l1_index, l2_offset, l2_index; 1138 int min_index, i, j; 1139 uint32_t min_count, *l2_table; 1140 bool zeroed = false; 1141 int64_t ret; 1142 int64_t cluster_sector; 1143 1144 if (m_data) { 1145 m_data->valid = 0; 1146 } 1147 if (extent->flat) { 1148 *cluster_offset = extent->flat_start_offset; 1149 return VMDK_OK; 1150 } 1151 1152 offset -= (extent->end_sector - extent->sectors) * SECTOR_SIZE; 1153 l1_index = (offset >> 9) / extent->l1_entry_sectors; 1154 if (l1_index >= extent->l1_size) { 1155 return VMDK_ERROR; 1156 } 1157 l2_offset = extent->l1_table[l1_index]; 1158 if (!l2_offset) { 1159 return VMDK_UNALLOC; 1160 } 1161 for (i = 0; i < L2_CACHE_SIZE; i++) { 1162 if (l2_offset == extent->l2_cache_offsets[i]) { 1163 /* increment the hit count */ 1164 if (++extent->l2_cache_counts[i] == 0xffffffff) { 1165 for (j = 0; j < L2_CACHE_SIZE; j++) { 1166 extent->l2_cache_counts[j] >>= 1; 1167 } 1168 } 1169 l2_table = extent->l2_cache + (i * extent->l2_size); 1170 goto found; 1171 } 1172 } 1173 /* not found: load a new entry in the least used one */ 1174 min_index = 0; 1175 min_count = 0xffffffff; 1176 for (i = 0; i < L2_CACHE_SIZE; i++) { 1177 if (extent->l2_cache_counts[i] < min_count) { 1178 min_count = extent->l2_cache_counts[i]; 1179 min_index = i; 1180 } 1181 } 1182 l2_table = extent->l2_cache + (min_index * extent->l2_size); 1183 if (bdrv_pread( 1184 extent->file, 1185 (int64_t)l2_offset * 512, 1186 l2_table, 1187 extent->l2_size * sizeof(uint32_t) 1188 ) != extent->l2_size * sizeof(uint32_t)) { 1189 return VMDK_ERROR; 1190 } 1191 1192 extent->l2_cache_offsets[min_index] = l2_offset; 1193 extent->l2_cache_counts[min_index] = 1; 1194 found: 1195 l2_index = ((offset >> 9) / extent->cluster_sectors) % extent->l2_size; 1196 cluster_sector = le32_to_cpu(l2_table[l2_index]); 1197 1198 if (m_data) { 1199 m_data->valid = 1; 1200 m_data->l1_index = l1_index; 1201 m_data->l2_index = l2_index; 1202 m_data->l2_offset = l2_offset; 1203 m_data->l2_cache_entry = &l2_table[l2_index]; 1204 } 1205 if (extent->has_zero_grain && cluster_sector == VMDK_GTE_ZEROED) { 1206 zeroed = true; 1207 } 1208 1209 if (!cluster_sector || zeroed) { 1210 if (!allocate) { 1211 return zeroed ? VMDK_ZEROED : VMDK_UNALLOC; 1212 } 1213 1214 cluster_sector = extent->next_cluster_sector; 1215 extent->next_cluster_sector += extent->cluster_sectors; 1216 1217 /* First of all we write grain itself, to avoid race condition 1218 * that may to corrupt the image. 1219 * This problem may occur because of insufficient space on host disk 1220 * or inappropriate VM shutdown. 1221 */ 1222 ret = get_whole_cluster(bs, extent, 1223 cluster_sector, 1224 offset >> BDRV_SECTOR_BITS, 1225 skip_start_sector, skip_end_sector); 1226 if (ret) { 1227 return ret; 1228 } 1229 } 1230 *cluster_offset = cluster_sector << BDRV_SECTOR_BITS; 1231 return VMDK_OK; 1232 } 1233 1234 static VmdkExtent *find_extent(BDRVVmdkState *s, 1235 int64_t sector_num, VmdkExtent *start_hint) 1236 { 1237 VmdkExtent *extent = start_hint; 1238 1239 if (!extent) { 1240 extent = &s->extents[0]; 1241 } 1242 while (extent < &s->extents[s->num_extents]) { 1243 if (sector_num < extent->end_sector) { 1244 return extent; 1245 } 1246 extent++; 1247 } 1248 return NULL; 1249 } 1250 1251 static inline uint64_t vmdk_find_index_in_cluster(VmdkExtent *extent, 1252 int64_t sector_num) 1253 { 1254 uint64_t index_in_cluster, extent_begin_sector, extent_relative_sector_num; 1255 1256 extent_begin_sector = extent->end_sector - extent->sectors; 1257 extent_relative_sector_num = sector_num - extent_begin_sector; 1258 index_in_cluster = extent_relative_sector_num % extent->cluster_sectors; 1259 return index_in_cluster; 1260 } 1261 1262 static int64_t coroutine_fn vmdk_co_get_block_status(BlockDriverState *bs, 1263 int64_t sector_num, int nb_sectors, int *pnum) 1264 { 1265 BDRVVmdkState *s = bs->opaque; 1266 int64_t index_in_cluster, n, ret; 1267 uint64_t offset; 1268 VmdkExtent *extent; 1269 1270 extent = find_extent(s, sector_num, NULL); 1271 if (!extent) { 1272 return 0; 1273 } 1274 qemu_co_mutex_lock(&s->lock); 1275 ret = get_cluster_offset(bs, extent, NULL, 1276 sector_num * 512, false, &offset, 1277 0, 0); 1278 qemu_co_mutex_unlock(&s->lock); 1279 1280 switch (ret) { 1281 case VMDK_ERROR: 1282 ret = -EIO; 1283 break; 1284 case VMDK_UNALLOC: 1285 ret = 0; 1286 break; 1287 case VMDK_ZEROED: 1288 ret = BDRV_BLOCK_ZERO; 1289 break; 1290 case VMDK_OK: 1291 ret = BDRV_BLOCK_DATA; 1292 if (extent->file == bs->file && !extent->compressed) { 1293 ret |= BDRV_BLOCK_OFFSET_VALID | offset; 1294 } 1295 1296 break; 1297 } 1298 1299 index_in_cluster = vmdk_find_index_in_cluster(extent, sector_num); 1300 n = extent->cluster_sectors - index_in_cluster; 1301 if (n > nb_sectors) { 1302 n = nb_sectors; 1303 } 1304 *pnum = n; 1305 return ret; 1306 } 1307 1308 static int vmdk_write_extent(VmdkExtent *extent, int64_t cluster_offset, 1309 int64_t offset_in_cluster, const uint8_t *buf, 1310 int nb_sectors, int64_t sector_num) 1311 { 1312 int ret; 1313 VmdkGrainMarker *data = NULL; 1314 uLongf buf_len; 1315 const uint8_t *write_buf = buf; 1316 int write_len = nb_sectors * 512; 1317 int64_t write_offset; 1318 int64_t write_end_sector; 1319 1320 if (extent->compressed) { 1321 if (!extent->has_marker) { 1322 ret = -EINVAL; 1323 goto out; 1324 } 1325 buf_len = (extent->cluster_sectors << 9) * 2; 1326 data = g_malloc(buf_len + sizeof(VmdkGrainMarker)); 1327 if (compress(data->data, &buf_len, buf, nb_sectors << 9) != Z_OK || 1328 buf_len == 0) { 1329 ret = -EINVAL; 1330 goto out; 1331 } 1332 data->lba = sector_num; 1333 data->size = buf_len; 1334 write_buf = (uint8_t *)data; 1335 write_len = buf_len + sizeof(VmdkGrainMarker); 1336 } 1337 write_offset = cluster_offset + offset_in_cluster, 1338 ret = bdrv_pwrite(extent->file, write_offset, write_buf, write_len); 1339 1340 write_end_sector = DIV_ROUND_UP(write_offset + write_len, BDRV_SECTOR_SIZE); 1341 1342 extent->next_cluster_sector = MAX(extent->next_cluster_sector, 1343 write_end_sector); 1344 1345 if (ret != write_len) { 1346 ret = ret < 0 ? ret : -EIO; 1347 goto out; 1348 } 1349 ret = 0; 1350 out: 1351 g_free(data); 1352 return ret; 1353 } 1354 1355 static int vmdk_read_extent(VmdkExtent *extent, int64_t cluster_offset, 1356 int64_t offset_in_cluster, uint8_t *buf, 1357 int nb_sectors) 1358 { 1359 int ret; 1360 int cluster_bytes, buf_bytes; 1361 uint8_t *cluster_buf, *compressed_data; 1362 uint8_t *uncomp_buf; 1363 uint32_t data_len; 1364 VmdkGrainMarker *marker; 1365 uLongf buf_len; 1366 1367 1368 if (!extent->compressed) { 1369 ret = bdrv_pread(extent->file, 1370 cluster_offset + offset_in_cluster, 1371 buf, nb_sectors * 512); 1372 if (ret == nb_sectors * 512) { 1373 return 0; 1374 } else { 1375 return -EIO; 1376 } 1377 } 1378 cluster_bytes = extent->cluster_sectors * 512; 1379 /* Read two clusters in case GrainMarker + compressed data > one cluster */ 1380 buf_bytes = cluster_bytes * 2; 1381 cluster_buf = g_malloc(buf_bytes); 1382 uncomp_buf = g_malloc(cluster_bytes); 1383 ret = bdrv_pread(extent->file, 1384 cluster_offset, 1385 cluster_buf, buf_bytes); 1386 if (ret < 0) { 1387 goto out; 1388 } 1389 compressed_data = cluster_buf; 1390 buf_len = cluster_bytes; 1391 data_len = cluster_bytes; 1392 if (extent->has_marker) { 1393 marker = (VmdkGrainMarker *)cluster_buf; 1394 compressed_data = marker->data; 1395 data_len = le32_to_cpu(marker->size); 1396 } 1397 if (!data_len || data_len > buf_bytes) { 1398 ret = -EINVAL; 1399 goto out; 1400 } 1401 ret = uncompress(uncomp_buf, &buf_len, compressed_data, data_len); 1402 if (ret != Z_OK) { 1403 ret = -EINVAL; 1404 goto out; 1405 1406 } 1407 if (offset_in_cluster < 0 || 1408 offset_in_cluster + nb_sectors * 512 > buf_len) { 1409 ret = -EINVAL; 1410 goto out; 1411 } 1412 memcpy(buf, uncomp_buf + offset_in_cluster, nb_sectors * 512); 1413 ret = 0; 1414 1415 out: 1416 g_free(uncomp_buf); 1417 g_free(cluster_buf); 1418 return ret; 1419 } 1420 1421 static int vmdk_read(BlockDriverState *bs, int64_t sector_num, 1422 uint8_t *buf, int nb_sectors) 1423 { 1424 BDRVVmdkState *s = bs->opaque; 1425 int ret; 1426 uint64_t n, index_in_cluster; 1427 VmdkExtent *extent = NULL; 1428 uint64_t cluster_offset; 1429 1430 while (nb_sectors > 0) { 1431 extent = find_extent(s, sector_num, extent); 1432 if (!extent) { 1433 return -EIO; 1434 } 1435 ret = get_cluster_offset(bs, extent, NULL, 1436 sector_num << 9, false, &cluster_offset, 1437 0, 0); 1438 index_in_cluster = vmdk_find_index_in_cluster(extent, sector_num); 1439 n = extent->cluster_sectors - index_in_cluster; 1440 if (n > nb_sectors) { 1441 n = nb_sectors; 1442 } 1443 if (ret != VMDK_OK) { 1444 /* if not allocated, try to read from parent image, if exist */ 1445 if (bs->backing_hd && ret != VMDK_ZEROED) { 1446 if (!vmdk_is_cid_valid(bs)) { 1447 return -EINVAL; 1448 } 1449 ret = bdrv_read(bs->backing_hd, sector_num, buf, n); 1450 if (ret < 0) { 1451 return ret; 1452 } 1453 } else { 1454 memset(buf, 0, 512 * n); 1455 } 1456 } else { 1457 ret = vmdk_read_extent(extent, 1458 cluster_offset, index_in_cluster * 512, 1459 buf, n); 1460 if (ret) { 1461 return ret; 1462 } 1463 } 1464 nb_sectors -= n; 1465 sector_num += n; 1466 buf += n * 512; 1467 } 1468 return 0; 1469 } 1470 1471 static coroutine_fn int vmdk_co_read(BlockDriverState *bs, int64_t sector_num, 1472 uint8_t *buf, int nb_sectors) 1473 { 1474 int ret; 1475 BDRVVmdkState *s = bs->opaque; 1476 qemu_co_mutex_lock(&s->lock); 1477 ret = vmdk_read(bs, sector_num, buf, nb_sectors); 1478 qemu_co_mutex_unlock(&s->lock); 1479 return ret; 1480 } 1481 1482 /** 1483 * vmdk_write: 1484 * @zeroed: buf is ignored (data is zero), use zeroed_grain GTE feature 1485 * if possible, otherwise return -ENOTSUP. 1486 * @zero_dry_run: used for zeroed == true only, don't update L2 table, just try 1487 * with each cluster. By dry run we can find if the zero write 1488 * is possible without modifying image data. 1489 * 1490 * Returns: error code with 0 for success. 1491 */ 1492 static int vmdk_write(BlockDriverState *bs, int64_t sector_num, 1493 const uint8_t *buf, int nb_sectors, 1494 bool zeroed, bool zero_dry_run) 1495 { 1496 BDRVVmdkState *s = bs->opaque; 1497 VmdkExtent *extent = NULL; 1498 int ret; 1499 int64_t index_in_cluster, n; 1500 uint64_t cluster_offset; 1501 VmdkMetaData m_data; 1502 1503 if (sector_num > bs->total_sectors) { 1504 error_report("Wrong offset: sector_num=0x%" PRIx64 1505 " total_sectors=0x%" PRIx64 "\n", 1506 sector_num, bs->total_sectors); 1507 return -EIO; 1508 } 1509 1510 while (nb_sectors > 0) { 1511 extent = find_extent(s, sector_num, extent); 1512 if (!extent) { 1513 return -EIO; 1514 } 1515 index_in_cluster = vmdk_find_index_in_cluster(extent, sector_num); 1516 n = extent->cluster_sectors - index_in_cluster; 1517 if (n > nb_sectors) { 1518 n = nb_sectors; 1519 } 1520 ret = get_cluster_offset(bs, extent, &m_data, sector_num << 9, 1521 !(extent->compressed || zeroed), 1522 &cluster_offset, 1523 index_in_cluster, index_in_cluster + n); 1524 if (extent->compressed) { 1525 if (ret == VMDK_OK) { 1526 /* Refuse write to allocated cluster for streamOptimized */ 1527 error_report("Could not write to allocated cluster" 1528 " for streamOptimized"); 1529 return -EIO; 1530 } else { 1531 /* allocate */ 1532 ret = get_cluster_offset(bs, extent, &m_data, sector_num << 9, 1533 true, &cluster_offset, 0, 0); 1534 } 1535 } 1536 if (ret == VMDK_ERROR) { 1537 return -EINVAL; 1538 } 1539 if (zeroed) { 1540 /* Do zeroed write, buf is ignored */ 1541 if (extent->has_zero_grain && 1542 index_in_cluster == 0 && 1543 n >= extent->cluster_sectors) { 1544 n = extent->cluster_sectors; 1545 if (!zero_dry_run) { 1546 /* update L2 tables */ 1547 if (vmdk_L2update(extent, &m_data, VMDK_GTE_ZEROED) 1548 != VMDK_OK) { 1549 return -EIO; 1550 } 1551 } 1552 } else { 1553 return -ENOTSUP; 1554 } 1555 } else { 1556 ret = vmdk_write_extent(extent, 1557 cluster_offset, index_in_cluster * 512, 1558 buf, n, sector_num); 1559 if (ret) { 1560 return ret; 1561 } 1562 if (m_data.valid) { 1563 /* update L2 tables */ 1564 if (vmdk_L2update(extent, &m_data, 1565 cluster_offset >> BDRV_SECTOR_BITS) 1566 != VMDK_OK) { 1567 return -EIO; 1568 } 1569 } 1570 } 1571 nb_sectors -= n; 1572 sector_num += n; 1573 buf += n * 512; 1574 1575 /* update CID on the first write every time the virtual disk is 1576 * opened */ 1577 if (!s->cid_updated) { 1578 ret = vmdk_write_cid(bs, g_random_int()); 1579 if (ret < 0) { 1580 return ret; 1581 } 1582 s->cid_updated = true; 1583 } 1584 } 1585 return 0; 1586 } 1587 1588 static coroutine_fn int vmdk_co_write(BlockDriverState *bs, int64_t sector_num, 1589 const uint8_t *buf, int nb_sectors) 1590 { 1591 int ret; 1592 BDRVVmdkState *s = bs->opaque; 1593 qemu_co_mutex_lock(&s->lock); 1594 ret = vmdk_write(bs, sector_num, buf, nb_sectors, false, false); 1595 qemu_co_mutex_unlock(&s->lock); 1596 return ret; 1597 } 1598 1599 static int vmdk_write_compressed(BlockDriverState *bs, 1600 int64_t sector_num, 1601 const uint8_t *buf, 1602 int nb_sectors) 1603 { 1604 BDRVVmdkState *s = bs->opaque; 1605 if (s->num_extents == 1 && s->extents[0].compressed) { 1606 return vmdk_write(bs, sector_num, buf, nb_sectors, false, false); 1607 } else { 1608 return -ENOTSUP; 1609 } 1610 } 1611 1612 static int coroutine_fn vmdk_co_write_zeroes(BlockDriverState *bs, 1613 int64_t sector_num, 1614 int nb_sectors, 1615 BdrvRequestFlags flags) 1616 { 1617 int ret; 1618 BDRVVmdkState *s = bs->opaque; 1619 qemu_co_mutex_lock(&s->lock); 1620 /* write zeroes could fail if sectors not aligned to cluster, test it with 1621 * dry_run == true before really updating image */ 1622 ret = vmdk_write(bs, sector_num, NULL, nb_sectors, true, true); 1623 if (!ret) { 1624 ret = vmdk_write(bs, sector_num, NULL, nb_sectors, true, false); 1625 } 1626 qemu_co_mutex_unlock(&s->lock); 1627 return ret; 1628 } 1629 1630 static int vmdk_create_extent(const char *filename, int64_t filesize, 1631 bool flat, bool compress, bool zeroed_grain, 1632 QemuOpts *opts, Error **errp) 1633 { 1634 int ret, i; 1635 BlockDriverState *bs = NULL; 1636 VMDK4Header header; 1637 Error *local_err = NULL; 1638 uint32_t tmp, magic, grains, gd_sectors, gt_size, gt_count; 1639 uint32_t *gd_buf = NULL; 1640 int gd_buf_size; 1641 1642 ret = bdrv_create_file(filename, opts, &local_err); 1643 if (ret < 0) { 1644 error_propagate(errp, local_err); 1645 goto exit; 1646 } 1647 1648 assert(bs == NULL); 1649 ret = bdrv_open(&bs, filename, NULL, NULL, BDRV_O_RDWR | BDRV_O_PROTOCOL, 1650 NULL, &local_err); 1651 if (ret < 0) { 1652 error_propagate(errp, local_err); 1653 goto exit; 1654 } 1655 1656 if (flat) { 1657 ret = bdrv_truncate(bs, filesize); 1658 if (ret < 0) { 1659 error_setg_errno(errp, -ret, "Could not truncate file"); 1660 } 1661 goto exit; 1662 } 1663 magic = cpu_to_be32(VMDK4_MAGIC); 1664 memset(&header, 0, sizeof(header)); 1665 header.version = zeroed_grain ? 2 : 1; 1666 header.flags = VMDK4_FLAG_RGD | VMDK4_FLAG_NL_DETECT 1667 | (compress ? VMDK4_FLAG_COMPRESS | VMDK4_FLAG_MARKER : 0) 1668 | (zeroed_grain ? VMDK4_FLAG_ZERO_GRAIN : 0); 1669 header.compressAlgorithm = compress ? VMDK4_COMPRESSION_DEFLATE : 0; 1670 header.capacity = filesize / BDRV_SECTOR_SIZE; 1671 header.granularity = 128; 1672 header.num_gtes_per_gt = BDRV_SECTOR_SIZE; 1673 1674 grains = DIV_ROUND_UP(filesize / BDRV_SECTOR_SIZE, header.granularity); 1675 gt_size = DIV_ROUND_UP(header.num_gtes_per_gt * sizeof(uint32_t), 1676 BDRV_SECTOR_SIZE); 1677 gt_count = DIV_ROUND_UP(grains, header.num_gtes_per_gt); 1678 gd_sectors = DIV_ROUND_UP(gt_count * sizeof(uint32_t), BDRV_SECTOR_SIZE); 1679 1680 header.desc_offset = 1; 1681 header.desc_size = 20; 1682 header.rgd_offset = header.desc_offset + header.desc_size; 1683 header.gd_offset = header.rgd_offset + gd_sectors + (gt_size * gt_count); 1684 header.grain_offset = 1685 ROUND_UP(header.gd_offset + gd_sectors + (gt_size * gt_count), 1686 header.granularity); 1687 /* swap endianness for all header fields */ 1688 header.version = cpu_to_le32(header.version); 1689 header.flags = cpu_to_le32(header.flags); 1690 header.capacity = cpu_to_le64(header.capacity); 1691 header.granularity = cpu_to_le64(header.granularity); 1692 header.num_gtes_per_gt = cpu_to_le32(header.num_gtes_per_gt); 1693 header.desc_offset = cpu_to_le64(header.desc_offset); 1694 header.desc_size = cpu_to_le64(header.desc_size); 1695 header.rgd_offset = cpu_to_le64(header.rgd_offset); 1696 header.gd_offset = cpu_to_le64(header.gd_offset); 1697 header.grain_offset = cpu_to_le64(header.grain_offset); 1698 header.compressAlgorithm = cpu_to_le16(header.compressAlgorithm); 1699 1700 header.check_bytes[0] = 0xa; 1701 header.check_bytes[1] = 0x20; 1702 header.check_bytes[2] = 0xd; 1703 header.check_bytes[3] = 0xa; 1704 1705 /* write all the data */ 1706 ret = bdrv_pwrite(bs, 0, &magic, sizeof(magic)); 1707 if (ret < 0) { 1708 error_set(errp, QERR_IO_ERROR); 1709 goto exit; 1710 } 1711 ret = bdrv_pwrite(bs, sizeof(magic), &header, sizeof(header)); 1712 if (ret < 0) { 1713 error_set(errp, QERR_IO_ERROR); 1714 goto exit; 1715 } 1716 1717 ret = bdrv_truncate(bs, le64_to_cpu(header.grain_offset) << 9); 1718 if (ret < 0) { 1719 error_setg_errno(errp, -ret, "Could not truncate file"); 1720 goto exit; 1721 } 1722 1723 /* write grain directory */ 1724 gd_buf_size = gd_sectors * BDRV_SECTOR_SIZE; 1725 gd_buf = g_malloc0(gd_buf_size); 1726 for (i = 0, tmp = le64_to_cpu(header.rgd_offset) + gd_sectors; 1727 i < gt_count; i++, tmp += gt_size) { 1728 gd_buf[i] = cpu_to_le32(tmp); 1729 } 1730 ret = bdrv_pwrite(bs, le64_to_cpu(header.rgd_offset) * BDRV_SECTOR_SIZE, 1731 gd_buf, gd_buf_size); 1732 if (ret < 0) { 1733 error_set(errp, QERR_IO_ERROR); 1734 goto exit; 1735 } 1736 1737 /* write backup grain directory */ 1738 for (i = 0, tmp = le64_to_cpu(header.gd_offset) + gd_sectors; 1739 i < gt_count; i++, tmp += gt_size) { 1740 gd_buf[i] = cpu_to_le32(tmp); 1741 } 1742 ret = bdrv_pwrite(bs, le64_to_cpu(header.gd_offset) * BDRV_SECTOR_SIZE, 1743 gd_buf, gd_buf_size); 1744 if (ret < 0) { 1745 error_set(errp, QERR_IO_ERROR); 1746 goto exit; 1747 } 1748 1749 ret = 0; 1750 exit: 1751 if (bs) { 1752 bdrv_unref(bs); 1753 } 1754 g_free(gd_buf); 1755 return ret; 1756 } 1757 1758 static int filename_decompose(const char *filename, char *path, char *prefix, 1759 char *postfix, size_t buf_len, Error **errp) 1760 { 1761 const char *p, *q; 1762 1763 if (filename == NULL || !strlen(filename)) { 1764 error_setg(errp, "No filename provided"); 1765 return VMDK_ERROR; 1766 } 1767 p = strrchr(filename, '/'); 1768 if (p == NULL) { 1769 p = strrchr(filename, '\\'); 1770 } 1771 if (p == NULL) { 1772 p = strrchr(filename, ':'); 1773 } 1774 if (p != NULL) { 1775 p++; 1776 if (p - filename >= buf_len) { 1777 return VMDK_ERROR; 1778 } 1779 pstrcpy(path, p - filename + 1, filename); 1780 } else { 1781 p = filename; 1782 path[0] = '\0'; 1783 } 1784 q = strrchr(p, '.'); 1785 if (q == NULL) { 1786 pstrcpy(prefix, buf_len, p); 1787 postfix[0] = '\0'; 1788 } else { 1789 if (q - p >= buf_len) { 1790 return VMDK_ERROR; 1791 } 1792 pstrcpy(prefix, q - p + 1, p); 1793 pstrcpy(postfix, buf_len, q); 1794 } 1795 return VMDK_OK; 1796 } 1797 1798 static int vmdk_create(const char *filename, QemuOpts *opts, Error **errp) 1799 { 1800 int idx = 0; 1801 BlockDriverState *new_bs = NULL; 1802 Error *local_err = NULL; 1803 char *desc = NULL; 1804 int64_t total_size = 0, filesize; 1805 char *adapter_type = NULL; 1806 char *backing_file = NULL; 1807 char *fmt = NULL; 1808 int flags = 0; 1809 int ret = 0; 1810 bool flat, split, compress; 1811 GString *ext_desc_lines; 1812 char *path = g_malloc0(PATH_MAX); 1813 char *prefix = g_malloc0(PATH_MAX); 1814 char *postfix = g_malloc0(PATH_MAX); 1815 char *desc_line = g_malloc0(BUF_SIZE); 1816 char *ext_filename = g_malloc0(PATH_MAX); 1817 char *desc_filename = g_malloc0(PATH_MAX); 1818 const int64_t split_size = 0x80000000; /* VMDK has constant split size */ 1819 const char *desc_extent_line; 1820 char *parent_desc_line = g_malloc0(BUF_SIZE); 1821 uint32_t parent_cid = 0xffffffff; 1822 uint32_t number_heads = 16; 1823 bool zeroed_grain = false; 1824 uint32_t desc_offset = 0, desc_len; 1825 const char desc_template[] = 1826 "# Disk DescriptorFile\n" 1827 "version=1\n" 1828 "CID=%" PRIx32 "\n" 1829 "parentCID=%" PRIx32 "\n" 1830 "createType=\"%s\"\n" 1831 "%s" 1832 "\n" 1833 "# Extent description\n" 1834 "%s" 1835 "\n" 1836 "# The Disk Data Base\n" 1837 "#DDB\n" 1838 "\n" 1839 "ddb.virtualHWVersion = \"%d\"\n" 1840 "ddb.geometry.cylinders = \"%" PRId64 "\"\n" 1841 "ddb.geometry.heads = \"%" PRIu32 "\"\n" 1842 "ddb.geometry.sectors = \"63\"\n" 1843 "ddb.adapterType = \"%s\"\n"; 1844 1845 ext_desc_lines = g_string_new(NULL); 1846 1847 if (filename_decompose(filename, path, prefix, postfix, PATH_MAX, errp)) { 1848 ret = -EINVAL; 1849 goto exit; 1850 } 1851 /* Read out options */ 1852 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0), 1853 BDRV_SECTOR_SIZE); 1854 adapter_type = qemu_opt_get_del(opts, BLOCK_OPT_ADAPTER_TYPE); 1855 backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE); 1856 if (qemu_opt_get_bool_del(opts, BLOCK_OPT_COMPAT6, false)) { 1857 flags |= BLOCK_FLAG_COMPAT6; 1858 } 1859 fmt = qemu_opt_get_del(opts, BLOCK_OPT_SUBFMT); 1860 if (qemu_opt_get_bool_del(opts, BLOCK_OPT_ZEROED_GRAIN, false)) { 1861 zeroed_grain = true; 1862 } 1863 1864 if (!adapter_type) { 1865 adapter_type = g_strdup("ide"); 1866 } else if (strcmp(adapter_type, "ide") && 1867 strcmp(adapter_type, "buslogic") && 1868 strcmp(adapter_type, "lsilogic") && 1869 strcmp(adapter_type, "legacyESX")) { 1870 error_setg(errp, "Unknown adapter type: '%s'", adapter_type); 1871 ret = -EINVAL; 1872 goto exit; 1873 } 1874 if (strcmp(adapter_type, "ide") != 0) { 1875 /* that's the number of heads with which vmware operates when 1876 creating, exporting, etc. vmdk files with a non-ide adapter type */ 1877 number_heads = 255; 1878 } 1879 if (!fmt) { 1880 /* Default format to monolithicSparse */ 1881 fmt = g_strdup("monolithicSparse"); 1882 } else if (strcmp(fmt, "monolithicFlat") && 1883 strcmp(fmt, "monolithicSparse") && 1884 strcmp(fmt, "twoGbMaxExtentSparse") && 1885 strcmp(fmt, "twoGbMaxExtentFlat") && 1886 strcmp(fmt, "streamOptimized")) { 1887 error_setg(errp, "Unknown subformat: '%s'", fmt); 1888 ret = -EINVAL; 1889 goto exit; 1890 } 1891 split = !(strcmp(fmt, "twoGbMaxExtentFlat") && 1892 strcmp(fmt, "twoGbMaxExtentSparse")); 1893 flat = !(strcmp(fmt, "monolithicFlat") && 1894 strcmp(fmt, "twoGbMaxExtentFlat")); 1895 compress = !strcmp(fmt, "streamOptimized"); 1896 if (flat) { 1897 desc_extent_line = "RW %" PRId64 " FLAT \"%s\" 0\n"; 1898 } else { 1899 desc_extent_line = "RW %" PRId64 " SPARSE \"%s\"\n"; 1900 } 1901 if (flat && backing_file) { 1902 error_setg(errp, "Flat image can't have backing file"); 1903 ret = -ENOTSUP; 1904 goto exit; 1905 } 1906 if (flat && zeroed_grain) { 1907 error_setg(errp, "Flat image can't enable zeroed grain"); 1908 ret = -ENOTSUP; 1909 goto exit; 1910 } 1911 if (backing_file) { 1912 BlockDriverState *bs = NULL; 1913 char *full_backing = g_new0(char, PATH_MAX); 1914 bdrv_get_full_backing_filename_from_filename(filename, backing_file, 1915 full_backing, PATH_MAX, 1916 &local_err); 1917 if (local_err) { 1918 g_free(full_backing); 1919 error_propagate(errp, local_err); 1920 ret = -ENOENT; 1921 goto exit; 1922 } 1923 ret = bdrv_open(&bs, full_backing, NULL, NULL, BDRV_O_NO_BACKING, NULL, 1924 errp); 1925 g_free(full_backing); 1926 if (ret != 0) { 1927 goto exit; 1928 } 1929 if (strcmp(bs->drv->format_name, "vmdk")) { 1930 bdrv_unref(bs); 1931 ret = -EINVAL; 1932 goto exit; 1933 } 1934 parent_cid = vmdk_read_cid(bs, 0); 1935 bdrv_unref(bs); 1936 snprintf(parent_desc_line, BUF_SIZE, 1937 "parentFileNameHint=\"%s\"", backing_file); 1938 } 1939 1940 /* Create extents */ 1941 filesize = total_size; 1942 while (filesize > 0) { 1943 int64_t size = filesize; 1944 1945 if (split && size > split_size) { 1946 size = split_size; 1947 } 1948 if (split) { 1949 snprintf(desc_filename, PATH_MAX, "%s-%c%03d%s", 1950 prefix, flat ? 'f' : 's', ++idx, postfix); 1951 } else if (flat) { 1952 snprintf(desc_filename, PATH_MAX, "%s-flat%s", prefix, postfix); 1953 } else { 1954 snprintf(desc_filename, PATH_MAX, "%s%s", prefix, postfix); 1955 } 1956 snprintf(ext_filename, PATH_MAX, "%s%s", path, desc_filename); 1957 1958 if (vmdk_create_extent(ext_filename, size, 1959 flat, compress, zeroed_grain, opts, errp)) { 1960 ret = -EINVAL; 1961 goto exit; 1962 } 1963 filesize -= size; 1964 1965 /* Format description line */ 1966 snprintf(desc_line, BUF_SIZE, 1967 desc_extent_line, size / BDRV_SECTOR_SIZE, desc_filename); 1968 g_string_append(ext_desc_lines, desc_line); 1969 } 1970 /* generate descriptor file */ 1971 desc = g_strdup_printf(desc_template, 1972 g_random_int(), 1973 parent_cid, 1974 fmt, 1975 parent_desc_line, 1976 ext_desc_lines->str, 1977 (flags & BLOCK_FLAG_COMPAT6 ? 6 : 4), 1978 total_size / 1979 (int64_t)(63 * number_heads * BDRV_SECTOR_SIZE), 1980 number_heads, 1981 adapter_type); 1982 desc_len = strlen(desc); 1983 /* the descriptor offset = 0x200 */ 1984 if (!split && !flat) { 1985 desc_offset = 0x200; 1986 } else { 1987 ret = bdrv_create_file(filename, opts, &local_err); 1988 if (ret < 0) { 1989 error_propagate(errp, local_err); 1990 goto exit; 1991 } 1992 } 1993 assert(new_bs == NULL); 1994 ret = bdrv_open(&new_bs, filename, NULL, NULL, 1995 BDRV_O_RDWR | BDRV_O_PROTOCOL, NULL, &local_err); 1996 if (ret < 0) { 1997 error_propagate(errp, local_err); 1998 goto exit; 1999 } 2000 ret = bdrv_pwrite(new_bs, desc_offset, desc, desc_len); 2001 if (ret < 0) { 2002 error_setg_errno(errp, -ret, "Could not write description"); 2003 goto exit; 2004 } 2005 /* bdrv_pwrite write padding zeros to align to sector, we don't need that 2006 * for description file */ 2007 if (desc_offset == 0) { 2008 ret = bdrv_truncate(new_bs, desc_len); 2009 if (ret < 0) { 2010 error_setg_errno(errp, -ret, "Could not truncate file"); 2011 } 2012 } 2013 exit: 2014 if (new_bs) { 2015 bdrv_unref(new_bs); 2016 } 2017 g_free(adapter_type); 2018 g_free(backing_file); 2019 g_free(fmt); 2020 g_free(desc); 2021 g_free(path); 2022 g_free(prefix); 2023 g_free(postfix); 2024 g_free(desc_line); 2025 g_free(ext_filename); 2026 g_free(desc_filename); 2027 g_free(parent_desc_line); 2028 g_string_free(ext_desc_lines, true); 2029 return ret; 2030 } 2031 2032 static void vmdk_close(BlockDriverState *bs) 2033 { 2034 BDRVVmdkState *s = bs->opaque; 2035 2036 vmdk_free_extents(bs); 2037 g_free(s->create_type); 2038 2039 migrate_del_blocker(s->migration_blocker); 2040 error_free(s->migration_blocker); 2041 } 2042 2043 static coroutine_fn int vmdk_co_flush(BlockDriverState *bs) 2044 { 2045 BDRVVmdkState *s = bs->opaque; 2046 int i, err; 2047 int ret = 0; 2048 2049 for (i = 0; i < s->num_extents; i++) { 2050 err = bdrv_co_flush(s->extents[i].file); 2051 if (err < 0) { 2052 ret = err; 2053 } 2054 } 2055 return ret; 2056 } 2057 2058 static int64_t vmdk_get_allocated_file_size(BlockDriverState *bs) 2059 { 2060 int i; 2061 int64_t ret = 0; 2062 int64_t r; 2063 BDRVVmdkState *s = bs->opaque; 2064 2065 ret = bdrv_get_allocated_file_size(bs->file); 2066 if (ret < 0) { 2067 return ret; 2068 } 2069 for (i = 0; i < s->num_extents; i++) { 2070 if (s->extents[i].file == bs->file) { 2071 continue; 2072 } 2073 r = bdrv_get_allocated_file_size(s->extents[i].file); 2074 if (r < 0) { 2075 return r; 2076 } 2077 ret += r; 2078 } 2079 return ret; 2080 } 2081 2082 static int vmdk_has_zero_init(BlockDriverState *bs) 2083 { 2084 int i; 2085 BDRVVmdkState *s = bs->opaque; 2086 2087 /* If has a flat extent and its underlying storage doesn't have zero init, 2088 * return 0. */ 2089 for (i = 0; i < s->num_extents; i++) { 2090 if (s->extents[i].flat) { 2091 if (!bdrv_has_zero_init(s->extents[i].file)) { 2092 return 0; 2093 } 2094 } 2095 } 2096 return 1; 2097 } 2098 2099 static ImageInfo *vmdk_get_extent_info(VmdkExtent *extent) 2100 { 2101 ImageInfo *info = g_new0(ImageInfo, 1); 2102 2103 *info = (ImageInfo){ 2104 .filename = g_strdup(extent->file->filename), 2105 .format = g_strdup(extent->type), 2106 .virtual_size = extent->sectors * BDRV_SECTOR_SIZE, 2107 .compressed = extent->compressed, 2108 .has_compressed = extent->compressed, 2109 .cluster_size = extent->cluster_sectors * BDRV_SECTOR_SIZE, 2110 .has_cluster_size = !extent->flat, 2111 }; 2112 2113 return info; 2114 } 2115 2116 static int vmdk_check(BlockDriverState *bs, BdrvCheckResult *result, 2117 BdrvCheckMode fix) 2118 { 2119 BDRVVmdkState *s = bs->opaque; 2120 VmdkExtent *extent = NULL; 2121 int64_t sector_num = 0; 2122 int64_t total_sectors = bdrv_nb_sectors(bs); 2123 int ret; 2124 uint64_t cluster_offset; 2125 2126 if (fix) { 2127 return -ENOTSUP; 2128 } 2129 2130 for (;;) { 2131 if (sector_num >= total_sectors) { 2132 return 0; 2133 } 2134 extent = find_extent(s, sector_num, extent); 2135 if (!extent) { 2136 fprintf(stderr, 2137 "ERROR: could not find extent for sector %" PRId64 "\n", 2138 sector_num); 2139 break; 2140 } 2141 ret = get_cluster_offset(bs, extent, NULL, 2142 sector_num << BDRV_SECTOR_BITS, 2143 false, &cluster_offset, 0, 0); 2144 if (ret == VMDK_ERROR) { 2145 fprintf(stderr, 2146 "ERROR: could not get cluster_offset for sector %" 2147 PRId64 "\n", sector_num); 2148 break; 2149 } 2150 if (ret == VMDK_OK && cluster_offset >= bdrv_getlength(extent->file)) { 2151 fprintf(stderr, 2152 "ERROR: cluster offset for sector %" 2153 PRId64 " points after EOF\n", sector_num); 2154 break; 2155 } 2156 sector_num += extent->cluster_sectors; 2157 } 2158 2159 result->corruptions++; 2160 return 0; 2161 } 2162 2163 static ImageInfoSpecific *vmdk_get_specific_info(BlockDriverState *bs) 2164 { 2165 int i; 2166 BDRVVmdkState *s = bs->opaque; 2167 ImageInfoSpecific *spec_info = g_new0(ImageInfoSpecific, 1); 2168 ImageInfoList **next; 2169 2170 *spec_info = (ImageInfoSpecific){ 2171 .kind = IMAGE_INFO_SPECIFIC_KIND_VMDK, 2172 { 2173 .vmdk = g_new0(ImageInfoSpecificVmdk, 1), 2174 }, 2175 }; 2176 2177 *spec_info->vmdk = (ImageInfoSpecificVmdk) { 2178 .create_type = g_strdup(s->create_type), 2179 .cid = s->cid, 2180 .parent_cid = s->parent_cid, 2181 }; 2182 2183 next = &spec_info->vmdk->extents; 2184 for (i = 0; i < s->num_extents; i++) { 2185 *next = g_new0(ImageInfoList, 1); 2186 (*next)->value = vmdk_get_extent_info(&s->extents[i]); 2187 (*next)->next = NULL; 2188 next = &(*next)->next; 2189 } 2190 2191 return spec_info; 2192 } 2193 2194 static bool vmdk_extents_type_eq(const VmdkExtent *a, const VmdkExtent *b) 2195 { 2196 return a->flat == b->flat && 2197 a->compressed == b->compressed && 2198 (a->flat || a->cluster_sectors == b->cluster_sectors); 2199 } 2200 2201 static int vmdk_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) 2202 { 2203 int i; 2204 BDRVVmdkState *s = bs->opaque; 2205 assert(s->num_extents); 2206 2207 /* See if we have multiple extents but they have different cases */ 2208 for (i = 1; i < s->num_extents; i++) { 2209 if (!vmdk_extents_type_eq(&s->extents[0], &s->extents[i])) { 2210 return -ENOTSUP; 2211 } 2212 } 2213 bdi->needs_compressed_writes = s->extents[0].compressed; 2214 if (!s->extents[0].flat) { 2215 bdi->cluster_size = s->extents[0].cluster_sectors << BDRV_SECTOR_BITS; 2216 } 2217 return 0; 2218 } 2219 2220 static void vmdk_detach_aio_context(BlockDriverState *bs) 2221 { 2222 BDRVVmdkState *s = bs->opaque; 2223 int i; 2224 2225 for (i = 0; i < s->num_extents; i++) { 2226 bdrv_detach_aio_context(s->extents[i].file); 2227 } 2228 } 2229 2230 static void vmdk_attach_aio_context(BlockDriverState *bs, 2231 AioContext *new_context) 2232 { 2233 BDRVVmdkState *s = bs->opaque; 2234 int i; 2235 2236 for (i = 0; i < s->num_extents; i++) { 2237 bdrv_attach_aio_context(s->extents[i].file, new_context); 2238 } 2239 } 2240 2241 static QemuOptsList vmdk_create_opts = { 2242 .name = "vmdk-create-opts", 2243 .head = QTAILQ_HEAD_INITIALIZER(vmdk_create_opts.head), 2244 .desc = { 2245 { 2246 .name = BLOCK_OPT_SIZE, 2247 .type = QEMU_OPT_SIZE, 2248 .help = "Virtual disk size" 2249 }, 2250 { 2251 .name = BLOCK_OPT_ADAPTER_TYPE, 2252 .type = QEMU_OPT_STRING, 2253 .help = "Virtual adapter type, can be one of " 2254 "ide (default), lsilogic, buslogic or legacyESX" 2255 }, 2256 { 2257 .name = BLOCK_OPT_BACKING_FILE, 2258 .type = QEMU_OPT_STRING, 2259 .help = "File name of a base image" 2260 }, 2261 { 2262 .name = BLOCK_OPT_COMPAT6, 2263 .type = QEMU_OPT_BOOL, 2264 .help = "VMDK version 6 image", 2265 .def_value_str = "off" 2266 }, 2267 { 2268 .name = BLOCK_OPT_SUBFMT, 2269 .type = QEMU_OPT_STRING, 2270 .help = 2271 "VMDK flat extent format, can be one of " 2272 "{monolithicSparse (default) | monolithicFlat | twoGbMaxExtentSparse | twoGbMaxExtentFlat | streamOptimized} " 2273 }, 2274 { 2275 .name = BLOCK_OPT_ZEROED_GRAIN, 2276 .type = QEMU_OPT_BOOL, 2277 .help = "Enable efficient zero writes " 2278 "using the zeroed-grain GTE feature" 2279 }, 2280 { /* end of list */ } 2281 } 2282 }; 2283 2284 static BlockDriver bdrv_vmdk = { 2285 .format_name = "vmdk", 2286 .instance_size = sizeof(BDRVVmdkState), 2287 .bdrv_probe = vmdk_probe, 2288 .bdrv_open = vmdk_open, 2289 .bdrv_check = vmdk_check, 2290 .bdrv_reopen_prepare = vmdk_reopen_prepare, 2291 .bdrv_read = vmdk_co_read, 2292 .bdrv_write = vmdk_co_write, 2293 .bdrv_write_compressed = vmdk_write_compressed, 2294 .bdrv_co_write_zeroes = vmdk_co_write_zeroes, 2295 .bdrv_close = vmdk_close, 2296 .bdrv_create = vmdk_create, 2297 .bdrv_co_flush_to_disk = vmdk_co_flush, 2298 .bdrv_co_get_block_status = vmdk_co_get_block_status, 2299 .bdrv_get_allocated_file_size = vmdk_get_allocated_file_size, 2300 .bdrv_has_zero_init = vmdk_has_zero_init, 2301 .bdrv_get_specific_info = vmdk_get_specific_info, 2302 .bdrv_refresh_limits = vmdk_refresh_limits, 2303 .bdrv_get_info = vmdk_get_info, 2304 .bdrv_detach_aio_context = vmdk_detach_aio_context, 2305 .bdrv_attach_aio_context = vmdk_attach_aio_context, 2306 2307 .supports_backing = true, 2308 .create_opts = &vmdk_create_opts, 2309 }; 2310 2311 static void bdrv_vmdk_init(void) 2312 { 2313 bdrv_register(&bdrv_vmdk); 2314 } 2315 2316 block_init(bdrv_vmdk_init); 2317