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/osdep.h" 27 #include "qemu-common.h" 28 #include "block/block_int.h" 29 #include "qapi/qmp/qerror.h" 30 #include "qemu/error-report.h" 31 #include "qemu/module.h" 32 #include "migration/migration.h" 33 #include <zlib.h> 34 #include <glib.h> 35 36 #define VMDK3_MAGIC (('C' << 24) | ('O' << 16) | ('W' << 8) | 'D') 37 #define VMDK4_MAGIC (('K' << 24) | ('D' << 16) | ('M' << 8) | 'V') 38 #define VMDK4_COMPRESSION_DEFLATE 1 39 #define VMDK4_FLAG_NL_DETECT (1 << 0) 40 #define VMDK4_FLAG_RGD (1 << 1) 41 /* Zeroed-grain enable bit */ 42 #define VMDK4_FLAG_ZERO_GRAIN (1 << 2) 43 #define VMDK4_FLAG_COMPRESS (1 << 16) 44 #define VMDK4_FLAG_MARKER (1 << 17) 45 #define VMDK4_GD_AT_END 0xffffffffffffffffULL 46 47 #define VMDK_GTE_ZEROED 0x1 48 49 /* VMDK internal error codes */ 50 #define VMDK_OK 0 51 #define VMDK_ERROR (-1) 52 /* Cluster not allocated */ 53 #define VMDK_UNALLOC (-2) 54 #define VMDK_ZEROED (-3) 55 56 #define BLOCK_OPT_ZEROED_GRAIN "zeroed_grain" 57 58 typedef struct { 59 uint32_t version; 60 uint32_t flags; 61 uint32_t disk_sectors; 62 uint32_t granularity; 63 uint32_t l1dir_offset; 64 uint32_t l1dir_size; 65 uint32_t file_sectors; 66 uint32_t cylinders; 67 uint32_t heads; 68 uint32_t sectors_per_track; 69 } QEMU_PACKED VMDK3Header; 70 71 typedef struct { 72 uint32_t version; 73 uint32_t flags; 74 uint64_t capacity; 75 uint64_t granularity; 76 uint64_t desc_offset; 77 uint64_t desc_size; 78 /* Number of GrainTableEntries per GrainTable */ 79 uint32_t num_gtes_per_gt; 80 uint64_t rgd_offset; 81 uint64_t gd_offset; 82 uint64_t grain_offset; 83 char filler[1]; 84 char check_bytes[4]; 85 uint16_t compressAlgorithm; 86 } QEMU_PACKED VMDK4Header; 87 88 #define L2_CACHE_SIZE 16 89 90 typedef struct VmdkExtent { 91 BdrvChild *file; 92 bool flat; 93 bool compressed; 94 bool has_marker; 95 bool has_zero_grain; 96 int version; 97 int64_t sectors; 98 int64_t end_sector; 99 int64_t flat_start_offset; 100 int64_t l1_table_offset; 101 int64_t l1_backup_table_offset; 102 uint32_t *l1_table; 103 uint32_t *l1_backup_table; 104 unsigned int l1_size; 105 uint32_t l1_entry_sectors; 106 107 unsigned int l2_size; 108 uint32_t *l2_cache; 109 uint32_t l2_cache_offsets[L2_CACHE_SIZE]; 110 uint32_t l2_cache_counts[L2_CACHE_SIZE]; 111 112 int64_t cluster_sectors; 113 int64_t next_cluster_sector; 114 char *type; 115 } VmdkExtent; 116 117 typedef struct BDRVVmdkState { 118 CoMutex lock; 119 uint64_t desc_offset; 120 bool cid_updated; 121 bool cid_checked; 122 uint32_t cid; 123 uint32_t parent_cid; 124 int num_extents; 125 /* Extent array with num_extents entries, ascend ordered by address */ 126 VmdkExtent *extents; 127 Error *migration_blocker; 128 char *create_type; 129 } BDRVVmdkState; 130 131 typedef struct VmdkMetaData { 132 unsigned int l1_index; 133 unsigned int l2_index; 134 unsigned int l2_offset; 135 int valid; 136 uint32_t *l2_cache_entry; 137 } VmdkMetaData; 138 139 typedef struct VmdkGrainMarker { 140 uint64_t lba; 141 uint32_t size; 142 uint8_t data[0]; 143 } QEMU_PACKED VmdkGrainMarker; 144 145 enum { 146 MARKER_END_OF_STREAM = 0, 147 MARKER_GRAIN_TABLE = 1, 148 MARKER_GRAIN_DIRECTORY = 2, 149 MARKER_FOOTER = 3, 150 }; 151 152 static int vmdk_probe(const uint8_t *buf, int buf_size, const char *filename) 153 { 154 uint32_t magic; 155 156 if (buf_size < 4) { 157 return 0; 158 } 159 magic = be32_to_cpu(*(uint32_t *)buf); 160 if (magic == VMDK3_MAGIC || 161 magic == VMDK4_MAGIC) { 162 return 100; 163 } else { 164 const char *p = (const char *)buf; 165 const char *end = p + buf_size; 166 while (p < end) { 167 if (*p == '#') { 168 /* skip comment line */ 169 while (p < end && *p != '\n') { 170 p++; 171 } 172 p++; 173 continue; 174 } 175 if (*p == ' ') { 176 while (p < end && *p == ' ') { 177 p++; 178 } 179 /* skip '\r' if windows line endings used. */ 180 if (p < end && *p == '\r') { 181 p++; 182 } 183 /* only accept blank lines before 'version=' line */ 184 if (p == end || *p != '\n') { 185 return 0; 186 } 187 p++; 188 continue; 189 } 190 if (end - p >= strlen("version=X\n")) { 191 if (strncmp("version=1\n", p, strlen("version=1\n")) == 0 || 192 strncmp("version=2\n", p, strlen("version=2\n")) == 0) { 193 return 100; 194 } 195 } 196 if (end - p >= strlen("version=X\r\n")) { 197 if (strncmp("version=1\r\n", p, strlen("version=1\r\n")) == 0 || 198 strncmp("version=2\r\n", p, strlen("version=2\r\n")) == 0) { 199 return 100; 200 } 201 } 202 return 0; 203 } 204 return 0; 205 } 206 } 207 208 #define SECTOR_SIZE 512 209 #define DESC_SIZE (20 * SECTOR_SIZE) /* 20 sectors of 512 bytes each */ 210 #define BUF_SIZE 4096 211 #define HEADER_SIZE 512 /* first sector of 512 bytes */ 212 213 static void vmdk_free_extents(BlockDriverState *bs) 214 { 215 int i; 216 BDRVVmdkState *s = bs->opaque; 217 VmdkExtent *e; 218 219 for (i = 0; i < s->num_extents; i++) { 220 e = &s->extents[i]; 221 g_free(e->l1_table); 222 g_free(e->l2_cache); 223 g_free(e->l1_backup_table); 224 g_free(e->type); 225 if (e->file != bs->file) { 226 bdrv_unref_child(bs, e->file); 227 } 228 } 229 g_free(s->extents); 230 } 231 232 static void vmdk_free_last_extent(BlockDriverState *bs) 233 { 234 BDRVVmdkState *s = bs->opaque; 235 236 if (s->num_extents == 0) { 237 return; 238 } 239 s->num_extents--; 240 s->extents = g_renew(VmdkExtent, s->extents, s->num_extents); 241 } 242 243 static uint32_t vmdk_read_cid(BlockDriverState *bs, int parent) 244 { 245 char desc[DESC_SIZE]; 246 uint32_t cid = 0xffffffff; 247 const char *p_name, *cid_str; 248 size_t cid_str_size; 249 BDRVVmdkState *s = bs->opaque; 250 int ret; 251 252 ret = bdrv_pread(bs->file->bs, s->desc_offset, desc, DESC_SIZE); 253 if (ret < 0) { 254 return 0; 255 } 256 257 if (parent) { 258 cid_str = "parentCID"; 259 cid_str_size = sizeof("parentCID"); 260 } else { 261 cid_str = "CID"; 262 cid_str_size = sizeof("CID"); 263 } 264 265 desc[DESC_SIZE - 1] = '\0'; 266 p_name = strstr(desc, cid_str); 267 if (p_name != NULL) { 268 p_name += cid_str_size; 269 sscanf(p_name, "%" SCNx32, &cid); 270 } 271 272 return cid; 273 } 274 275 static int vmdk_write_cid(BlockDriverState *bs, uint32_t cid) 276 { 277 char desc[DESC_SIZE], tmp_desc[DESC_SIZE]; 278 char *p_name, *tmp_str; 279 BDRVVmdkState *s = bs->opaque; 280 int ret; 281 282 ret = bdrv_pread(bs->file->bs, s->desc_offset, desc, DESC_SIZE); 283 if (ret < 0) { 284 return ret; 285 } 286 287 desc[DESC_SIZE - 1] = '\0'; 288 tmp_str = strstr(desc, "parentCID"); 289 if (tmp_str == NULL) { 290 return -EINVAL; 291 } 292 293 pstrcpy(tmp_desc, sizeof(tmp_desc), tmp_str); 294 p_name = strstr(desc, "CID"); 295 if (p_name != NULL) { 296 p_name += sizeof("CID"); 297 snprintf(p_name, sizeof(desc) - (p_name - desc), "%" PRIx32 "\n", cid); 298 pstrcat(desc, sizeof(desc), tmp_desc); 299 } 300 301 ret = bdrv_pwrite_sync(bs->file->bs, s->desc_offset, desc, DESC_SIZE); 302 if (ret < 0) { 303 return ret; 304 } 305 306 return 0; 307 } 308 309 static int vmdk_is_cid_valid(BlockDriverState *bs) 310 { 311 BDRVVmdkState *s = bs->opaque; 312 uint32_t cur_pcid; 313 314 if (!s->cid_checked && bs->backing) { 315 BlockDriverState *p_bs = bs->backing->bs; 316 317 cur_pcid = vmdk_read_cid(p_bs, 0); 318 if (s->parent_cid != cur_pcid) { 319 /* CID not valid */ 320 return 0; 321 } 322 } 323 s->cid_checked = true; 324 /* CID valid */ 325 return 1; 326 } 327 328 /* We have nothing to do for VMDK reopen, stubs just return success */ 329 static int vmdk_reopen_prepare(BDRVReopenState *state, 330 BlockReopenQueue *queue, Error **errp) 331 { 332 assert(state != NULL); 333 assert(state->bs != NULL); 334 return 0; 335 } 336 337 static int vmdk_parent_open(BlockDriverState *bs) 338 { 339 char *p_name; 340 char desc[DESC_SIZE + 1]; 341 BDRVVmdkState *s = bs->opaque; 342 int ret; 343 344 desc[DESC_SIZE] = '\0'; 345 ret = bdrv_pread(bs->file->bs, s->desc_offset, desc, DESC_SIZE); 346 if (ret < 0) { 347 return ret; 348 } 349 350 p_name = strstr(desc, "parentFileNameHint"); 351 if (p_name != NULL) { 352 char *end_name; 353 354 p_name += sizeof("parentFileNameHint") + 1; 355 end_name = strchr(p_name, '\"'); 356 if (end_name == NULL) { 357 return -EINVAL; 358 } 359 if ((end_name - p_name) > sizeof(bs->backing_file) - 1) { 360 return -EINVAL; 361 } 362 363 pstrcpy(bs->backing_file, end_name - p_name + 1, p_name); 364 } 365 366 return 0; 367 } 368 369 /* Create and append extent to the extent array. Return the added VmdkExtent 370 * address. return NULL if allocation failed. */ 371 static int vmdk_add_extent(BlockDriverState *bs, 372 BdrvChild *file, bool flat, int64_t sectors, 373 int64_t l1_offset, int64_t l1_backup_offset, 374 uint32_t l1_size, 375 int l2_size, uint64_t cluster_sectors, 376 VmdkExtent **new_extent, 377 Error **errp) 378 { 379 VmdkExtent *extent; 380 BDRVVmdkState *s = bs->opaque; 381 int64_t nb_sectors; 382 383 if (cluster_sectors > 0x200000) { 384 /* 0x200000 * 512Bytes = 1GB for one cluster is unrealistic */ 385 error_setg(errp, "Invalid granularity, image may be corrupt"); 386 return -EFBIG; 387 } 388 if (l1_size > 512 * 1024 * 1024) { 389 /* Although with big capacity and small l1_entry_sectors, we can get a 390 * big l1_size, we don't want unbounded value to allocate the table. 391 * Limit it to 512M, which is 16PB for default cluster and L2 table 392 * size */ 393 error_setg(errp, "L1 size too big"); 394 return -EFBIG; 395 } 396 397 nb_sectors = bdrv_nb_sectors(file->bs); 398 if (nb_sectors < 0) { 399 return nb_sectors; 400 } 401 402 s->extents = g_renew(VmdkExtent, s->extents, s->num_extents + 1); 403 extent = &s->extents[s->num_extents]; 404 s->num_extents++; 405 406 memset(extent, 0, sizeof(VmdkExtent)); 407 extent->file = file; 408 extent->flat = flat; 409 extent->sectors = sectors; 410 extent->l1_table_offset = l1_offset; 411 extent->l1_backup_table_offset = l1_backup_offset; 412 extent->l1_size = l1_size; 413 extent->l1_entry_sectors = l2_size * cluster_sectors; 414 extent->l2_size = l2_size; 415 extent->cluster_sectors = flat ? sectors : cluster_sectors; 416 extent->next_cluster_sector = ROUND_UP(nb_sectors, cluster_sectors); 417 418 if (s->num_extents > 1) { 419 extent->end_sector = (*(extent - 1)).end_sector + extent->sectors; 420 } else { 421 extent->end_sector = extent->sectors; 422 } 423 bs->total_sectors = extent->end_sector; 424 if (new_extent) { 425 *new_extent = extent; 426 } 427 return 0; 428 } 429 430 static int vmdk_init_tables(BlockDriverState *bs, VmdkExtent *extent, 431 Error **errp) 432 { 433 int ret; 434 size_t l1_size; 435 int i; 436 437 /* read the L1 table */ 438 l1_size = extent->l1_size * sizeof(uint32_t); 439 extent->l1_table = g_try_malloc(l1_size); 440 if (l1_size && extent->l1_table == NULL) { 441 return -ENOMEM; 442 } 443 444 ret = bdrv_pread(extent->file->bs, 445 extent->l1_table_offset, 446 extent->l1_table, 447 l1_size); 448 if (ret < 0) { 449 error_setg_errno(errp, -ret, 450 "Could not read l1 table from extent '%s'", 451 extent->file->bs->filename); 452 goto fail_l1; 453 } 454 for (i = 0; i < extent->l1_size; i++) { 455 le32_to_cpus(&extent->l1_table[i]); 456 } 457 458 if (extent->l1_backup_table_offset) { 459 extent->l1_backup_table = g_try_malloc(l1_size); 460 if (l1_size && extent->l1_backup_table == NULL) { 461 ret = -ENOMEM; 462 goto fail_l1; 463 } 464 ret = bdrv_pread(extent->file->bs, 465 extent->l1_backup_table_offset, 466 extent->l1_backup_table, 467 l1_size); 468 if (ret < 0) { 469 error_setg_errno(errp, -ret, 470 "Could not read l1 backup table from extent '%s'", 471 extent->file->bs->filename); 472 goto fail_l1b; 473 } 474 for (i = 0; i < extent->l1_size; i++) { 475 le32_to_cpus(&extent->l1_backup_table[i]); 476 } 477 } 478 479 extent->l2_cache = 480 g_new(uint32_t, extent->l2_size * L2_CACHE_SIZE); 481 return 0; 482 fail_l1b: 483 g_free(extent->l1_backup_table); 484 fail_l1: 485 g_free(extent->l1_table); 486 return ret; 487 } 488 489 static int vmdk_open_vmfs_sparse(BlockDriverState *bs, 490 BdrvChild *file, 491 int flags, Error **errp) 492 { 493 int ret; 494 uint32_t magic; 495 VMDK3Header header; 496 VmdkExtent *extent; 497 498 ret = bdrv_pread(file->bs, sizeof(magic), &header, sizeof(header)); 499 if (ret < 0) { 500 error_setg_errno(errp, -ret, 501 "Could not read header from file '%s'", 502 file->bs->filename); 503 return ret; 504 } 505 ret = vmdk_add_extent(bs, file, false, 506 le32_to_cpu(header.disk_sectors), 507 (int64_t)le32_to_cpu(header.l1dir_offset) << 9, 508 0, 509 le32_to_cpu(header.l1dir_size), 510 4096, 511 le32_to_cpu(header.granularity), 512 &extent, 513 errp); 514 if (ret < 0) { 515 return ret; 516 } 517 ret = vmdk_init_tables(bs, extent, errp); 518 if (ret) { 519 /* free extent allocated by vmdk_add_extent */ 520 vmdk_free_last_extent(bs); 521 } 522 return ret; 523 } 524 525 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf, 526 QDict *options, Error **errp); 527 528 static char *vmdk_read_desc(BlockDriverState *file, uint64_t desc_offset, 529 Error **errp) 530 { 531 int64_t size; 532 char *buf; 533 int ret; 534 535 size = bdrv_getlength(file); 536 if (size < 0) { 537 error_setg_errno(errp, -size, "Could not access file"); 538 return NULL; 539 } 540 541 if (size < 4) { 542 /* Both descriptor file and sparse image must be much larger than 4 543 * bytes, also callers of vmdk_read_desc want to compare the first 4 544 * bytes with VMDK4_MAGIC, let's error out if less is read. */ 545 error_setg(errp, "File is too small, not a valid image"); 546 return NULL; 547 } 548 549 size = MIN(size, (1 << 20) - 1); /* avoid unbounded allocation */ 550 buf = g_malloc(size + 1); 551 552 ret = bdrv_pread(file, desc_offset, buf, size); 553 if (ret < 0) { 554 error_setg_errno(errp, -ret, "Could not read from file"); 555 g_free(buf); 556 return NULL; 557 } 558 buf[ret] = 0; 559 560 return buf; 561 } 562 563 static int vmdk_open_vmdk4(BlockDriverState *bs, 564 BdrvChild *file, 565 int flags, QDict *options, Error **errp) 566 { 567 int ret; 568 uint32_t magic; 569 uint32_t l1_size, l1_entry_sectors; 570 VMDK4Header header; 571 VmdkExtent *extent; 572 BDRVVmdkState *s = bs->opaque; 573 int64_t l1_backup_offset = 0; 574 bool compressed; 575 576 ret = bdrv_pread(file->bs, sizeof(magic), &header, sizeof(header)); 577 if (ret < 0) { 578 error_setg_errno(errp, -ret, 579 "Could not read header from file '%s'", 580 file->bs->filename); 581 return -EINVAL; 582 } 583 if (header.capacity == 0) { 584 uint64_t desc_offset = le64_to_cpu(header.desc_offset); 585 if (desc_offset) { 586 char *buf = vmdk_read_desc(file->bs, desc_offset << 9, errp); 587 if (!buf) { 588 return -EINVAL; 589 } 590 ret = vmdk_open_desc_file(bs, flags, buf, options, errp); 591 g_free(buf); 592 return ret; 593 } 594 } 595 596 if (!s->create_type) { 597 s->create_type = g_strdup("monolithicSparse"); 598 } 599 600 if (le64_to_cpu(header.gd_offset) == VMDK4_GD_AT_END) { 601 /* 602 * The footer takes precedence over the header, so read it in. The 603 * footer starts at offset -1024 from the end: One sector for the 604 * footer, and another one for the end-of-stream marker. 605 */ 606 struct { 607 struct { 608 uint64_t val; 609 uint32_t size; 610 uint32_t type; 611 uint8_t pad[512 - 16]; 612 } QEMU_PACKED footer_marker; 613 614 uint32_t magic; 615 VMDK4Header header; 616 uint8_t pad[512 - 4 - sizeof(VMDK4Header)]; 617 618 struct { 619 uint64_t val; 620 uint32_t size; 621 uint32_t type; 622 uint8_t pad[512 - 16]; 623 } QEMU_PACKED eos_marker; 624 } QEMU_PACKED footer; 625 626 ret = bdrv_pread(file->bs, 627 bs->file->bs->total_sectors * 512 - 1536, 628 &footer, sizeof(footer)); 629 if (ret < 0) { 630 error_setg_errno(errp, -ret, "Failed to read footer"); 631 return ret; 632 } 633 634 /* Some sanity checks for the footer */ 635 if (be32_to_cpu(footer.magic) != VMDK4_MAGIC || 636 le32_to_cpu(footer.footer_marker.size) != 0 || 637 le32_to_cpu(footer.footer_marker.type) != MARKER_FOOTER || 638 le64_to_cpu(footer.eos_marker.val) != 0 || 639 le32_to_cpu(footer.eos_marker.size) != 0 || 640 le32_to_cpu(footer.eos_marker.type) != MARKER_END_OF_STREAM) 641 { 642 error_setg(errp, "Invalid footer"); 643 return -EINVAL; 644 } 645 646 header = footer.header; 647 } 648 649 compressed = 650 le16_to_cpu(header.compressAlgorithm) == VMDK4_COMPRESSION_DEFLATE; 651 if (le32_to_cpu(header.version) > 3) { 652 char buf[64]; 653 snprintf(buf, sizeof(buf), "VMDK version %" PRId32, 654 le32_to_cpu(header.version)); 655 error_setg(errp, QERR_UNKNOWN_BLOCK_FORMAT_FEATURE, 656 bdrv_get_device_or_node_name(bs), "vmdk", buf); 657 return -ENOTSUP; 658 } else if (le32_to_cpu(header.version) == 3 && (flags & BDRV_O_RDWR) && 659 !compressed) { 660 /* VMware KB 2064959 explains that version 3 added support for 661 * persistent changed block tracking (CBT), and backup software can 662 * read it as version=1 if it doesn't care about the changed area 663 * information. So we are safe to enable read only. */ 664 error_setg(errp, "VMDK version 3 must be read only"); 665 return -EINVAL; 666 } 667 668 if (le32_to_cpu(header.num_gtes_per_gt) > 512) { 669 error_setg(errp, "L2 table size too big"); 670 return -EINVAL; 671 } 672 673 l1_entry_sectors = le32_to_cpu(header.num_gtes_per_gt) 674 * le64_to_cpu(header.granularity); 675 if (l1_entry_sectors == 0) { 676 error_setg(errp, "L1 entry size is invalid"); 677 return -EINVAL; 678 } 679 l1_size = (le64_to_cpu(header.capacity) + l1_entry_sectors - 1) 680 / l1_entry_sectors; 681 if (le32_to_cpu(header.flags) & VMDK4_FLAG_RGD) { 682 l1_backup_offset = le64_to_cpu(header.rgd_offset) << 9; 683 } 684 if (bdrv_nb_sectors(file->bs) < le64_to_cpu(header.grain_offset)) { 685 error_setg(errp, "File truncated, expecting at least %" PRId64 " bytes", 686 (int64_t)(le64_to_cpu(header.grain_offset) 687 * BDRV_SECTOR_SIZE)); 688 return -EINVAL; 689 } 690 691 ret = vmdk_add_extent(bs, file, false, 692 le64_to_cpu(header.capacity), 693 le64_to_cpu(header.gd_offset) << 9, 694 l1_backup_offset, 695 l1_size, 696 le32_to_cpu(header.num_gtes_per_gt), 697 le64_to_cpu(header.granularity), 698 &extent, 699 errp); 700 if (ret < 0) { 701 return ret; 702 } 703 extent->compressed = 704 le16_to_cpu(header.compressAlgorithm) == VMDK4_COMPRESSION_DEFLATE; 705 if (extent->compressed) { 706 g_free(s->create_type); 707 s->create_type = g_strdup("streamOptimized"); 708 } 709 extent->has_marker = le32_to_cpu(header.flags) & VMDK4_FLAG_MARKER; 710 extent->version = le32_to_cpu(header.version); 711 extent->has_zero_grain = le32_to_cpu(header.flags) & VMDK4_FLAG_ZERO_GRAIN; 712 ret = vmdk_init_tables(bs, extent, errp); 713 if (ret) { 714 /* free extent allocated by vmdk_add_extent */ 715 vmdk_free_last_extent(bs); 716 } 717 return ret; 718 } 719 720 /* find an option value out of descriptor file */ 721 static int vmdk_parse_description(const char *desc, const char *opt_name, 722 char *buf, int buf_size) 723 { 724 char *opt_pos, *opt_end; 725 const char *end = desc + strlen(desc); 726 727 opt_pos = strstr(desc, opt_name); 728 if (!opt_pos) { 729 return VMDK_ERROR; 730 } 731 /* Skip "=\"" following opt_name */ 732 opt_pos += strlen(opt_name) + 2; 733 if (opt_pos >= end) { 734 return VMDK_ERROR; 735 } 736 opt_end = opt_pos; 737 while (opt_end < end && *opt_end != '"') { 738 opt_end++; 739 } 740 if (opt_end == end || buf_size < opt_end - opt_pos + 1) { 741 return VMDK_ERROR; 742 } 743 pstrcpy(buf, opt_end - opt_pos + 1, opt_pos); 744 return VMDK_OK; 745 } 746 747 /* Open an extent file and append to bs array */ 748 static int vmdk_open_sparse(BlockDriverState *bs, BdrvChild *file, int flags, 749 char *buf, QDict *options, Error **errp) 750 { 751 uint32_t magic; 752 753 magic = ldl_be_p(buf); 754 switch (magic) { 755 case VMDK3_MAGIC: 756 return vmdk_open_vmfs_sparse(bs, file, flags, errp); 757 break; 758 case VMDK4_MAGIC: 759 return vmdk_open_vmdk4(bs, file, flags, options, errp); 760 break; 761 default: 762 error_setg(errp, "Image not in VMDK format"); 763 return -EINVAL; 764 break; 765 } 766 } 767 768 static const char *next_line(const char *s) 769 { 770 while (*s) { 771 if (*s == '\n') { 772 return s + 1; 773 } 774 s++; 775 } 776 return s; 777 } 778 779 static int vmdk_parse_extents(const char *desc, BlockDriverState *bs, 780 const char *desc_file_path, QDict *options, 781 Error **errp) 782 { 783 int ret; 784 int matches; 785 char access[11]; 786 char type[11]; 787 char fname[512]; 788 const char *p, *np; 789 int64_t sectors = 0; 790 int64_t flat_offset; 791 char *extent_path; 792 BdrvChild *extent_file; 793 BDRVVmdkState *s = bs->opaque; 794 VmdkExtent *extent; 795 char extent_opt_prefix[32]; 796 Error *local_err = NULL; 797 798 for (p = desc; *p; p = next_line(p)) { 799 /* parse extent line in one of below formats: 800 * 801 * RW [size in sectors] FLAT "file-name.vmdk" OFFSET 802 * RW [size in sectors] SPARSE "file-name.vmdk" 803 * RW [size in sectors] VMFS "file-name.vmdk" 804 * RW [size in sectors] VMFSSPARSE "file-name.vmdk" 805 */ 806 flat_offset = -1; 807 matches = sscanf(p, "%10s %" SCNd64 " %10s \"%511[^\n\r\"]\" %" SCNd64, 808 access, §ors, type, fname, &flat_offset); 809 if (matches < 4 || strcmp(access, "RW")) { 810 continue; 811 } else if (!strcmp(type, "FLAT")) { 812 if (matches != 5 || flat_offset < 0) { 813 goto invalid; 814 } 815 } else if (!strcmp(type, "VMFS")) { 816 if (matches == 4) { 817 flat_offset = 0; 818 } else { 819 goto invalid; 820 } 821 } else if (matches != 4) { 822 goto invalid; 823 } 824 825 if (sectors <= 0 || 826 (strcmp(type, "FLAT") && strcmp(type, "SPARSE") && 827 strcmp(type, "VMFS") && strcmp(type, "VMFSSPARSE")) || 828 (strcmp(access, "RW"))) { 829 continue; 830 } 831 832 if (!path_is_absolute(fname) && !path_has_protocol(fname) && 833 !desc_file_path[0]) 834 { 835 error_setg(errp, "Cannot use relative extent paths with VMDK " 836 "descriptor file '%s'", bs->file->bs->filename); 837 return -EINVAL; 838 } 839 840 extent_path = g_malloc0(PATH_MAX); 841 path_combine(extent_path, PATH_MAX, desc_file_path, fname); 842 843 ret = snprintf(extent_opt_prefix, 32, "extents.%d", s->num_extents); 844 assert(ret < 32); 845 846 extent_file = bdrv_open_child(extent_path, options, extent_opt_prefix, 847 bs, &child_file, false, &local_err); 848 g_free(extent_path); 849 if (local_err) { 850 error_propagate(errp, local_err); 851 return -EINVAL; 852 } 853 854 /* save to extents array */ 855 if (!strcmp(type, "FLAT") || !strcmp(type, "VMFS")) { 856 /* FLAT extent */ 857 858 ret = vmdk_add_extent(bs, extent_file, true, sectors, 859 0, 0, 0, 0, 0, &extent, errp); 860 if (ret < 0) { 861 bdrv_unref_child(bs, extent_file); 862 return ret; 863 } 864 extent->flat_start_offset = flat_offset << 9; 865 } else if (!strcmp(type, "SPARSE") || !strcmp(type, "VMFSSPARSE")) { 866 /* SPARSE extent and VMFSSPARSE extent are both "COWD" sparse file*/ 867 char *buf = vmdk_read_desc(extent_file->bs, 0, errp); 868 if (!buf) { 869 ret = -EINVAL; 870 } else { 871 ret = vmdk_open_sparse(bs, extent_file, bs->open_flags, buf, 872 options, errp); 873 } 874 g_free(buf); 875 if (ret) { 876 bdrv_unref_child(bs, extent_file); 877 return ret; 878 } 879 extent = &s->extents[s->num_extents - 1]; 880 } else { 881 error_setg(errp, "Unsupported extent type '%s'", type); 882 bdrv_unref_child(bs, extent_file); 883 return -ENOTSUP; 884 } 885 extent->type = g_strdup(type); 886 } 887 return 0; 888 889 invalid: 890 np = next_line(p); 891 assert(np != p); 892 if (np[-1] == '\n') { 893 np--; 894 } 895 error_setg(errp, "Invalid extent line: %.*s", (int)(np - p), p); 896 return -EINVAL; 897 } 898 899 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf, 900 QDict *options, Error **errp) 901 { 902 int ret; 903 char ct[128]; 904 BDRVVmdkState *s = bs->opaque; 905 906 if (vmdk_parse_description(buf, "createType", ct, sizeof(ct))) { 907 error_setg(errp, "invalid VMDK image descriptor"); 908 ret = -EINVAL; 909 goto exit; 910 } 911 if (strcmp(ct, "monolithicFlat") && 912 strcmp(ct, "vmfs") && 913 strcmp(ct, "vmfsSparse") && 914 strcmp(ct, "twoGbMaxExtentSparse") && 915 strcmp(ct, "twoGbMaxExtentFlat")) { 916 error_setg(errp, "Unsupported image type '%s'", ct); 917 ret = -ENOTSUP; 918 goto exit; 919 } 920 s->create_type = g_strdup(ct); 921 s->desc_offset = 0; 922 ret = vmdk_parse_extents(buf, bs, bs->file->bs->exact_filename, options, 923 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->bs, 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, options, 946 errp); 947 s->desc_offset = 0x200; 948 break; 949 default: 950 ret = vmdk_open_desc_file(bs, flags, buf, options, errp); 951 break; 952 } 953 if (ret) { 954 goto fail; 955 } 956 957 /* try to open parent images, if exist */ 958 ret = vmdk_parent_open(bs); 959 if (ret) { 960 goto fail; 961 } 962 s->cid = vmdk_read_cid(bs, 0); 963 s->parent_cid = vmdk_read_cid(bs, 1); 964 qemu_co_mutex_init(&s->lock); 965 966 /* Disable migration when VMDK images are used */ 967 error_setg(&s->migration_blocker, "The vmdk format used by node '%s' " 968 "does not support live migration", 969 bdrv_get_device_or_node_name(bs)); 970 migrate_add_blocker(s->migration_blocker); 971 g_free(buf); 972 return 0; 973 974 fail: 975 g_free(buf); 976 g_free(s->create_type); 977 s->create_type = NULL; 978 vmdk_free_extents(bs); 979 return ret; 980 } 981 982 983 static void vmdk_refresh_limits(BlockDriverState *bs, Error **errp) 984 { 985 BDRVVmdkState *s = bs->opaque; 986 int i; 987 988 for (i = 0; i < s->num_extents; i++) { 989 if (!s->extents[i].flat) { 990 bs->bl.write_zeroes_alignment = 991 MAX(bs->bl.write_zeroes_alignment, 992 s->extents[i].cluster_sectors); 993 } 994 } 995 } 996 997 /** 998 * get_whole_cluster 999 * 1000 * Copy backing file's cluster that covers @sector_num, otherwise write zero, 1001 * to the cluster at @cluster_sector_num. 1002 * 1003 * If @skip_start_sector < @skip_end_sector, the relative range 1004 * [@skip_start_sector, @skip_end_sector) is not copied or written, and leave 1005 * it for call to write user data in the request. 1006 */ 1007 static int get_whole_cluster(BlockDriverState *bs, 1008 VmdkExtent *extent, 1009 uint64_t cluster_sector_num, 1010 uint64_t sector_num, 1011 uint64_t skip_start_sector, 1012 uint64_t skip_end_sector) 1013 { 1014 int ret = VMDK_OK; 1015 int64_t cluster_bytes; 1016 uint8_t *whole_grain; 1017 1018 /* For COW, align request sector_num to cluster start */ 1019 sector_num = QEMU_ALIGN_DOWN(sector_num, extent->cluster_sectors); 1020 cluster_bytes = extent->cluster_sectors << BDRV_SECTOR_BITS; 1021 whole_grain = qemu_blockalign(bs, cluster_bytes); 1022 1023 if (!bs->backing) { 1024 memset(whole_grain, 0, skip_start_sector << BDRV_SECTOR_BITS); 1025 memset(whole_grain + (skip_end_sector << BDRV_SECTOR_BITS), 0, 1026 cluster_bytes - (skip_end_sector << BDRV_SECTOR_BITS)); 1027 } 1028 1029 assert(skip_end_sector <= extent->cluster_sectors); 1030 /* we will be here if it's first write on non-exist grain(cluster). 1031 * try to read from parent image, if exist */ 1032 if (bs->backing && !vmdk_is_cid_valid(bs)) { 1033 ret = VMDK_ERROR; 1034 goto exit; 1035 } 1036 1037 /* Read backing data before skip range */ 1038 if (skip_start_sector > 0) { 1039 if (bs->backing) { 1040 ret = bdrv_read(bs->backing->bs, sector_num, 1041 whole_grain, skip_start_sector); 1042 if (ret < 0) { 1043 ret = VMDK_ERROR; 1044 goto exit; 1045 } 1046 } 1047 ret = bdrv_write(extent->file->bs, cluster_sector_num, whole_grain, 1048 skip_start_sector); 1049 if (ret < 0) { 1050 ret = VMDK_ERROR; 1051 goto exit; 1052 } 1053 } 1054 /* Read backing data after skip range */ 1055 if (skip_end_sector < extent->cluster_sectors) { 1056 if (bs->backing) { 1057 ret = bdrv_read(bs->backing->bs, sector_num + skip_end_sector, 1058 whole_grain + (skip_end_sector << BDRV_SECTOR_BITS), 1059 extent->cluster_sectors - skip_end_sector); 1060 if (ret < 0) { 1061 ret = VMDK_ERROR; 1062 goto exit; 1063 } 1064 } 1065 ret = bdrv_write(extent->file->bs, cluster_sector_num + skip_end_sector, 1066 whole_grain + (skip_end_sector << BDRV_SECTOR_BITS), 1067 extent->cluster_sectors - skip_end_sector); 1068 if (ret < 0) { 1069 ret = VMDK_ERROR; 1070 goto exit; 1071 } 1072 } 1073 1074 exit: 1075 qemu_vfree(whole_grain); 1076 return ret; 1077 } 1078 1079 static int vmdk_L2update(VmdkExtent *extent, VmdkMetaData *m_data, 1080 uint32_t offset) 1081 { 1082 offset = cpu_to_le32(offset); 1083 /* update L2 table */ 1084 if (bdrv_pwrite_sync( 1085 extent->file->bs, 1086 ((int64_t)m_data->l2_offset * 512) 1087 + (m_data->l2_index * sizeof(offset)), 1088 &offset, sizeof(offset)) < 0) { 1089 return VMDK_ERROR; 1090 } 1091 /* update backup L2 table */ 1092 if (extent->l1_backup_table_offset != 0) { 1093 m_data->l2_offset = extent->l1_backup_table[m_data->l1_index]; 1094 if (bdrv_pwrite_sync( 1095 extent->file->bs, 1096 ((int64_t)m_data->l2_offset * 512) 1097 + (m_data->l2_index * sizeof(offset)), 1098 &offset, sizeof(offset)) < 0) { 1099 return VMDK_ERROR; 1100 } 1101 } 1102 if (m_data->l2_cache_entry) { 1103 *m_data->l2_cache_entry = offset; 1104 } 1105 1106 return VMDK_OK; 1107 } 1108 1109 /** 1110 * get_cluster_offset 1111 * 1112 * Look up cluster offset in extent file by sector number, and store in 1113 * @cluster_offset. 1114 * 1115 * For flat extents, the start offset as parsed from the description file is 1116 * returned. 1117 * 1118 * For sparse extents, look up in L1, L2 table. If allocate is true, return an 1119 * offset for a new cluster and update L2 cache. If there is a backing file, 1120 * COW is done before returning; otherwise, zeroes are written to the allocated 1121 * cluster. Both COW and zero writing skips the sector range 1122 * [@skip_start_sector, @skip_end_sector) passed in by caller, because caller 1123 * has new data to write there. 1124 * 1125 * Returns: VMDK_OK if cluster exists and mapped in the image. 1126 * VMDK_UNALLOC if cluster is not mapped and @allocate is false. 1127 * VMDK_ERROR if failed. 1128 */ 1129 static int get_cluster_offset(BlockDriverState *bs, 1130 VmdkExtent *extent, 1131 VmdkMetaData *m_data, 1132 uint64_t offset, 1133 bool allocate, 1134 uint64_t *cluster_offset, 1135 uint64_t skip_start_sector, 1136 uint64_t skip_end_sector) 1137 { 1138 unsigned int l1_index, l2_offset, l2_index; 1139 int min_index, i, j; 1140 uint32_t min_count, *l2_table; 1141 bool zeroed = false; 1142 int64_t ret; 1143 int64_t cluster_sector; 1144 1145 if (m_data) { 1146 m_data->valid = 0; 1147 } 1148 if (extent->flat) { 1149 *cluster_offset = extent->flat_start_offset; 1150 return VMDK_OK; 1151 } 1152 1153 offset -= (extent->end_sector - extent->sectors) * SECTOR_SIZE; 1154 l1_index = (offset >> 9) / extent->l1_entry_sectors; 1155 if (l1_index >= extent->l1_size) { 1156 return VMDK_ERROR; 1157 } 1158 l2_offset = extent->l1_table[l1_index]; 1159 if (!l2_offset) { 1160 return VMDK_UNALLOC; 1161 } 1162 for (i = 0; i < L2_CACHE_SIZE; i++) { 1163 if (l2_offset == extent->l2_cache_offsets[i]) { 1164 /* increment the hit count */ 1165 if (++extent->l2_cache_counts[i] == 0xffffffff) { 1166 for (j = 0; j < L2_CACHE_SIZE; j++) { 1167 extent->l2_cache_counts[j] >>= 1; 1168 } 1169 } 1170 l2_table = extent->l2_cache + (i * extent->l2_size); 1171 goto found; 1172 } 1173 } 1174 /* not found: load a new entry in the least used one */ 1175 min_index = 0; 1176 min_count = 0xffffffff; 1177 for (i = 0; i < L2_CACHE_SIZE; i++) { 1178 if (extent->l2_cache_counts[i] < min_count) { 1179 min_count = extent->l2_cache_counts[i]; 1180 min_index = i; 1181 } 1182 } 1183 l2_table = extent->l2_cache + (min_index * extent->l2_size); 1184 if (bdrv_pread( 1185 extent->file->bs, 1186 (int64_t)l2_offset * 512, 1187 l2_table, 1188 extent->l2_size * sizeof(uint32_t) 1189 ) != extent->l2_size * sizeof(uint32_t)) { 1190 return VMDK_ERROR; 1191 } 1192 1193 extent->l2_cache_offsets[min_index] = l2_offset; 1194 extent->l2_cache_counts[min_index] = 1; 1195 found: 1196 l2_index = ((offset >> 9) / extent->cluster_sectors) % extent->l2_size; 1197 cluster_sector = le32_to_cpu(l2_table[l2_index]); 1198 1199 if (m_data) { 1200 m_data->valid = 1; 1201 m_data->l1_index = l1_index; 1202 m_data->l2_index = l2_index; 1203 m_data->l2_offset = l2_offset; 1204 m_data->l2_cache_entry = &l2_table[l2_index]; 1205 } 1206 if (extent->has_zero_grain && cluster_sector == VMDK_GTE_ZEROED) { 1207 zeroed = true; 1208 } 1209 1210 if (!cluster_sector || zeroed) { 1211 if (!allocate) { 1212 return zeroed ? VMDK_ZEROED : VMDK_UNALLOC; 1213 } 1214 1215 cluster_sector = extent->next_cluster_sector; 1216 extent->next_cluster_sector += extent->cluster_sectors; 1217 1218 /* First of all we write grain itself, to avoid race condition 1219 * that may to corrupt the image. 1220 * This problem may occur because of insufficient space on host disk 1221 * or inappropriate VM shutdown. 1222 */ 1223 ret = get_whole_cluster(bs, extent, 1224 cluster_sector, 1225 offset >> BDRV_SECTOR_BITS, 1226 skip_start_sector, skip_end_sector); 1227 if (ret) { 1228 return ret; 1229 } 1230 } 1231 *cluster_offset = cluster_sector << BDRV_SECTOR_BITS; 1232 return VMDK_OK; 1233 } 1234 1235 static VmdkExtent *find_extent(BDRVVmdkState *s, 1236 int64_t sector_num, VmdkExtent *start_hint) 1237 { 1238 VmdkExtent *extent = start_hint; 1239 1240 if (!extent) { 1241 extent = &s->extents[0]; 1242 } 1243 while (extent < &s->extents[s->num_extents]) { 1244 if (sector_num < extent->end_sector) { 1245 return extent; 1246 } 1247 extent++; 1248 } 1249 return NULL; 1250 } 1251 1252 static inline uint64_t vmdk_find_index_in_cluster(VmdkExtent *extent, 1253 int64_t sector_num) 1254 { 1255 uint64_t index_in_cluster, extent_begin_sector, extent_relative_sector_num; 1256 1257 extent_begin_sector = extent->end_sector - extent->sectors; 1258 extent_relative_sector_num = sector_num - extent_begin_sector; 1259 index_in_cluster = extent_relative_sector_num % extent->cluster_sectors; 1260 return index_in_cluster; 1261 } 1262 1263 static int64_t coroutine_fn vmdk_co_get_block_status(BlockDriverState *bs, 1264 int64_t sector_num, int nb_sectors, int *pnum, BlockDriverState **file) 1265 { 1266 BDRVVmdkState *s = bs->opaque; 1267 int64_t index_in_cluster, n, ret; 1268 uint64_t offset; 1269 VmdkExtent *extent; 1270 1271 extent = find_extent(s, sector_num, NULL); 1272 if (!extent) { 1273 return 0; 1274 } 1275 qemu_co_mutex_lock(&s->lock); 1276 ret = get_cluster_offset(bs, extent, NULL, 1277 sector_num * 512, false, &offset, 1278 0, 0); 1279 qemu_co_mutex_unlock(&s->lock); 1280 1281 index_in_cluster = vmdk_find_index_in_cluster(extent, sector_num); 1282 switch (ret) { 1283 case VMDK_ERROR: 1284 ret = -EIO; 1285 break; 1286 case VMDK_UNALLOC: 1287 ret = 0; 1288 break; 1289 case VMDK_ZEROED: 1290 ret = BDRV_BLOCK_ZERO; 1291 break; 1292 case VMDK_OK: 1293 ret = BDRV_BLOCK_DATA; 1294 if (!extent->compressed) { 1295 ret |= BDRV_BLOCK_OFFSET_VALID; 1296 ret |= (offset + (index_in_cluster << BDRV_SECTOR_BITS)) 1297 & BDRV_BLOCK_OFFSET_MASK; 1298 } 1299 *file = extent->file->bs; 1300 break; 1301 } 1302 1303 n = extent->cluster_sectors - index_in_cluster; 1304 if (n > nb_sectors) { 1305 n = nb_sectors; 1306 } 1307 *pnum = n; 1308 return ret; 1309 } 1310 1311 static int vmdk_write_extent(VmdkExtent *extent, int64_t cluster_offset, 1312 int64_t offset_in_cluster, const uint8_t *buf, 1313 int nb_sectors, int64_t sector_num) 1314 { 1315 int ret; 1316 VmdkGrainMarker *data = NULL; 1317 uLongf buf_len; 1318 const uint8_t *write_buf = buf; 1319 int write_len = nb_sectors * 512; 1320 int64_t write_offset; 1321 int64_t write_end_sector; 1322 1323 if (extent->compressed) { 1324 if (!extent->has_marker) { 1325 ret = -EINVAL; 1326 goto out; 1327 } 1328 buf_len = (extent->cluster_sectors << 9) * 2; 1329 data = g_malloc(buf_len + sizeof(VmdkGrainMarker)); 1330 if (compress(data->data, &buf_len, buf, nb_sectors << 9) != Z_OK || 1331 buf_len == 0) { 1332 ret = -EINVAL; 1333 goto out; 1334 } 1335 data->lba = sector_num; 1336 data->size = buf_len; 1337 write_buf = (uint8_t *)data; 1338 write_len = buf_len + sizeof(VmdkGrainMarker); 1339 } 1340 write_offset = cluster_offset + offset_in_cluster, 1341 ret = bdrv_pwrite(extent->file->bs, write_offset, write_buf, write_len); 1342 1343 write_end_sector = DIV_ROUND_UP(write_offset + write_len, BDRV_SECTOR_SIZE); 1344 1345 if (extent->compressed) { 1346 extent->next_cluster_sector = write_end_sector; 1347 } else { 1348 extent->next_cluster_sector = MAX(extent->next_cluster_sector, 1349 write_end_sector); 1350 } 1351 1352 if (ret != write_len) { 1353 ret = ret < 0 ? ret : -EIO; 1354 goto out; 1355 } 1356 ret = 0; 1357 out: 1358 g_free(data); 1359 return ret; 1360 } 1361 1362 static int vmdk_read_extent(VmdkExtent *extent, int64_t cluster_offset, 1363 int64_t offset_in_cluster, uint8_t *buf, 1364 int nb_sectors) 1365 { 1366 int ret; 1367 int cluster_bytes, buf_bytes; 1368 uint8_t *cluster_buf, *compressed_data; 1369 uint8_t *uncomp_buf; 1370 uint32_t data_len; 1371 VmdkGrainMarker *marker; 1372 uLongf buf_len; 1373 1374 1375 if (!extent->compressed) { 1376 ret = bdrv_pread(extent->file->bs, 1377 cluster_offset + offset_in_cluster, 1378 buf, nb_sectors * 512); 1379 if (ret == nb_sectors * 512) { 1380 return 0; 1381 } else { 1382 return -EIO; 1383 } 1384 } 1385 cluster_bytes = extent->cluster_sectors * 512; 1386 /* Read two clusters in case GrainMarker + compressed data > one cluster */ 1387 buf_bytes = cluster_bytes * 2; 1388 cluster_buf = g_malloc(buf_bytes); 1389 uncomp_buf = g_malloc(cluster_bytes); 1390 ret = bdrv_pread(extent->file->bs, 1391 cluster_offset, 1392 cluster_buf, buf_bytes); 1393 if (ret < 0) { 1394 goto out; 1395 } 1396 compressed_data = cluster_buf; 1397 buf_len = cluster_bytes; 1398 data_len = cluster_bytes; 1399 if (extent->has_marker) { 1400 marker = (VmdkGrainMarker *)cluster_buf; 1401 compressed_data = marker->data; 1402 data_len = le32_to_cpu(marker->size); 1403 } 1404 if (!data_len || data_len > buf_bytes) { 1405 ret = -EINVAL; 1406 goto out; 1407 } 1408 ret = uncompress(uncomp_buf, &buf_len, compressed_data, data_len); 1409 if (ret != Z_OK) { 1410 ret = -EINVAL; 1411 goto out; 1412 1413 } 1414 if (offset_in_cluster < 0 || 1415 offset_in_cluster + nb_sectors * 512 > buf_len) { 1416 ret = -EINVAL; 1417 goto out; 1418 } 1419 memcpy(buf, uncomp_buf + offset_in_cluster, nb_sectors * 512); 1420 ret = 0; 1421 1422 out: 1423 g_free(uncomp_buf); 1424 g_free(cluster_buf); 1425 return ret; 1426 } 1427 1428 static int vmdk_read(BlockDriverState *bs, int64_t sector_num, 1429 uint8_t *buf, int nb_sectors) 1430 { 1431 BDRVVmdkState *s = bs->opaque; 1432 int ret; 1433 uint64_t n, index_in_cluster; 1434 VmdkExtent *extent = NULL; 1435 uint64_t cluster_offset; 1436 1437 while (nb_sectors > 0) { 1438 extent = find_extent(s, sector_num, extent); 1439 if (!extent) { 1440 return -EIO; 1441 } 1442 ret = get_cluster_offset(bs, extent, NULL, 1443 sector_num << 9, false, &cluster_offset, 1444 0, 0); 1445 index_in_cluster = vmdk_find_index_in_cluster(extent, sector_num); 1446 n = extent->cluster_sectors - index_in_cluster; 1447 if (n > nb_sectors) { 1448 n = nb_sectors; 1449 } 1450 if (ret != VMDK_OK) { 1451 /* if not allocated, try to read from parent image, if exist */ 1452 if (bs->backing && ret != VMDK_ZEROED) { 1453 if (!vmdk_is_cid_valid(bs)) { 1454 return -EINVAL; 1455 } 1456 ret = bdrv_read(bs->backing->bs, sector_num, buf, n); 1457 if (ret < 0) { 1458 return ret; 1459 } 1460 } else { 1461 memset(buf, 0, 512 * n); 1462 } 1463 } else { 1464 ret = vmdk_read_extent(extent, 1465 cluster_offset, index_in_cluster * 512, 1466 buf, n); 1467 if (ret) { 1468 return ret; 1469 } 1470 } 1471 nb_sectors -= n; 1472 sector_num += n; 1473 buf += n * 512; 1474 } 1475 return 0; 1476 } 1477 1478 static coroutine_fn int vmdk_co_read(BlockDriverState *bs, int64_t sector_num, 1479 uint8_t *buf, int nb_sectors) 1480 { 1481 int ret; 1482 BDRVVmdkState *s = bs->opaque; 1483 qemu_co_mutex_lock(&s->lock); 1484 ret = vmdk_read(bs, sector_num, buf, nb_sectors); 1485 qemu_co_mutex_unlock(&s->lock); 1486 return ret; 1487 } 1488 1489 /** 1490 * vmdk_write: 1491 * @zeroed: buf is ignored (data is zero), use zeroed_grain GTE feature 1492 * if possible, otherwise return -ENOTSUP. 1493 * @zero_dry_run: used for zeroed == true only, don't update L2 table, just try 1494 * with each cluster. By dry run we can find if the zero write 1495 * is possible without modifying image data. 1496 * 1497 * Returns: error code with 0 for success. 1498 */ 1499 static int vmdk_write(BlockDriverState *bs, int64_t sector_num, 1500 const uint8_t *buf, int nb_sectors, 1501 bool zeroed, bool zero_dry_run) 1502 { 1503 BDRVVmdkState *s = bs->opaque; 1504 VmdkExtent *extent = NULL; 1505 int ret; 1506 int64_t index_in_cluster, n; 1507 uint64_t cluster_offset; 1508 VmdkMetaData m_data; 1509 1510 if (sector_num > bs->total_sectors) { 1511 error_report("Wrong offset: sector_num=0x%" PRIx64 1512 " total_sectors=0x%" PRIx64, 1513 sector_num, bs->total_sectors); 1514 return -EIO; 1515 } 1516 1517 while (nb_sectors > 0) { 1518 extent = find_extent(s, sector_num, extent); 1519 if (!extent) { 1520 return -EIO; 1521 } 1522 index_in_cluster = vmdk_find_index_in_cluster(extent, sector_num); 1523 n = extent->cluster_sectors - index_in_cluster; 1524 if (n > nb_sectors) { 1525 n = nb_sectors; 1526 } 1527 ret = get_cluster_offset(bs, extent, &m_data, sector_num << 9, 1528 !(extent->compressed || zeroed), 1529 &cluster_offset, 1530 index_in_cluster, index_in_cluster + n); 1531 if (extent->compressed) { 1532 if (ret == VMDK_OK) { 1533 /* Refuse write to allocated cluster for streamOptimized */ 1534 error_report("Could not write to allocated cluster" 1535 " for streamOptimized"); 1536 return -EIO; 1537 } else { 1538 /* allocate */ 1539 ret = get_cluster_offset(bs, extent, &m_data, sector_num << 9, 1540 true, &cluster_offset, 0, 0); 1541 } 1542 } 1543 if (ret == VMDK_ERROR) { 1544 return -EINVAL; 1545 } 1546 if (zeroed) { 1547 /* Do zeroed write, buf is ignored */ 1548 if (extent->has_zero_grain && 1549 index_in_cluster == 0 && 1550 n >= extent->cluster_sectors) { 1551 n = extent->cluster_sectors; 1552 if (!zero_dry_run) { 1553 /* update L2 tables */ 1554 if (vmdk_L2update(extent, &m_data, VMDK_GTE_ZEROED) 1555 != VMDK_OK) { 1556 return -EIO; 1557 } 1558 } 1559 } else { 1560 return -ENOTSUP; 1561 } 1562 } else { 1563 ret = vmdk_write_extent(extent, 1564 cluster_offset, index_in_cluster * 512, 1565 buf, n, sector_num); 1566 if (ret) { 1567 return ret; 1568 } 1569 if (m_data.valid) { 1570 /* update L2 tables */ 1571 if (vmdk_L2update(extent, &m_data, 1572 cluster_offset >> BDRV_SECTOR_BITS) 1573 != VMDK_OK) { 1574 return -EIO; 1575 } 1576 } 1577 } 1578 nb_sectors -= n; 1579 sector_num += n; 1580 buf += n * 512; 1581 1582 /* update CID on the first write every time the virtual disk is 1583 * opened */ 1584 if (!s->cid_updated) { 1585 ret = vmdk_write_cid(bs, g_random_int()); 1586 if (ret < 0) { 1587 return ret; 1588 } 1589 s->cid_updated = true; 1590 } 1591 } 1592 return 0; 1593 } 1594 1595 static coroutine_fn int vmdk_co_write(BlockDriverState *bs, int64_t sector_num, 1596 const uint8_t *buf, int nb_sectors) 1597 { 1598 int ret; 1599 BDRVVmdkState *s = bs->opaque; 1600 qemu_co_mutex_lock(&s->lock); 1601 ret = vmdk_write(bs, sector_num, buf, nb_sectors, false, false); 1602 qemu_co_mutex_unlock(&s->lock); 1603 return ret; 1604 } 1605 1606 static int vmdk_write_compressed(BlockDriverState *bs, 1607 int64_t sector_num, 1608 const uint8_t *buf, 1609 int nb_sectors) 1610 { 1611 BDRVVmdkState *s = bs->opaque; 1612 if (s->num_extents == 1 && s->extents[0].compressed) { 1613 return vmdk_write(bs, sector_num, buf, nb_sectors, false, false); 1614 } else { 1615 return -ENOTSUP; 1616 } 1617 } 1618 1619 static int coroutine_fn vmdk_co_write_zeroes(BlockDriverState *bs, 1620 int64_t sector_num, 1621 int nb_sectors, 1622 BdrvRequestFlags flags) 1623 { 1624 int ret; 1625 BDRVVmdkState *s = bs->opaque; 1626 qemu_co_mutex_lock(&s->lock); 1627 /* write zeroes could fail if sectors not aligned to cluster, test it with 1628 * dry_run == true before really updating image */ 1629 ret = vmdk_write(bs, sector_num, NULL, nb_sectors, true, true); 1630 if (!ret) { 1631 ret = vmdk_write(bs, sector_num, NULL, nb_sectors, true, false); 1632 } 1633 qemu_co_mutex_unlock(&s->lock); 1634 return ret; 1635 } 1636 1637 static int vmdk_create_extent(const char *filename, int64_t filesize, 1638 bool flat, bool compress, bool zeroed_grain, 1639 QemuOpts *opts, Error **errp) 1640 { 1641 int ret, i; 1642 BlockDriverState *bs = NULL; 1643 VMDK4Header header; 1644 Error *local_err = NULL; 1645 uint32_t tmp, magic, grains, gd_sectors, gt_size, gt_count; 1646 uint32_t *gd_buf = NULL; 1647 int gd_buf_size; 1648 1649 ret = bdrv_create_file(filename, opts, &local_err); 1650 if (ret < 0) { 1651 error_propagate(errp, local_err); 1652 goto exit; 1653 } 1654 1655 assert(bs == NULL); 1656 ret = bdrv_open(&bs, filename, NULL, NULL, BDRV_O_RDWR | BDRV_O_PROTOCOL, 1657 &local_err); 1658 if (ret < 0) { 1659 error_propagate(errp, local_err); 1660 goto exit; 1661 } 1662 1663 if (flat) { 1664 ret = bdrv_truncate(bs, filesize); 1665 if (ret < 0) { 1666 error_setg_errno(errp, -ret, "Could not truncate file"); 1667 } 1668 goto exit; 1669 } 1670 magic = cpu_to_be32(VMDK4_MAGIC); 1671 memset(&header, 0, sizeof(header)); 1672 if (compress) { 1673 header.version = 3; 1674 } else if (zeroed_grain) { 1675 header.version = 2; 1676 } else { 1677 header.version = 1; 1678 } 1679 header.flags = VMDK4_FLAG_RGD | VMDK4_FLAG_NL_DETECT 1680 | (compress ? VMDK4_FLAG_COMPRESS | VMDK4_FLAG_MARKER : 0) 1681 | (zeroed_grain ? VMDK4_FLAG_ZERO_GRAIN : 0); 1682 header.compressAlgorithm = compress ? VMDK4_COMPRESSION_DEFLATE : 0; 1683 header.capacity = filesize / BDRV_SECTOR_SIZE; 1684 header.granularity = 128; 1685 header.num_gtes_per_gt = BDRV_SECTOR_SIZE; 1686 1687 grains = DIV_ROUND_UP(filesize / BDRV_SECTOR_SIZE, header.granularity); 1688 gt_size = DIV_ROUND_UP(header.num_gtes_per_gt * sizeof(uint32_t), 1689 BDRV_SECTOR_SIZE); 1690 gt_count = DIV_ROUND_UP(grains, header.num_gtes_per_gt); 1691 gd_sectors = DIV_ROUND_UP(gt_count * sizeof(uint32_t), BDRV_SECTOR_SIZE); 1692 1693 header.desc_offset = 1; 1694 header.desc_size = 20; 1695 header.rgd_offset = header.desc_offset + header.desc_size; 1696 header.gd_offset = header.rgd_offset + gd_sectors + (gt_size * gt_count); 1697 header.grain_offset = 1698 ROUND_UP(header.gd_offset + gd_sectors + (gt_size * gt_count), 1699 header.granularity); 1700 /* swap endianness for all header fields */ 1701 header.version = cpu_to_le32(header.version); 1702 header.flags = cpu_to_le32(header.flags); 1703 header.capacity = cpu_to_le64(header.capacity); 1704 header.granularity = cpu_to_le64(header.granularity); 1705 header.num_gtes_per_gt = cpu_to_le32(header.num_gtes_per_gt); 1706 header.desc_offset = cpu_to_le64(header.desc_offset); 1707 header.desc_size = cpu_to_le64(header.desc_size); 1708 header.rgd_offset = cpu_to_le64(header.rgd_offset); 1709 header.gd_offset = cpu_to_le64(header.gd_offset); 1710 header.grain_offset = cpu_to_le64(header.grain_offset); 1711 header.compressAlgorithm = cpu_to_le16(header.compressAlgorithm); 1712 1713 header.check_bytes[0] = 0xa; 1714 header.check_bytes[1] = 0x20; 1715 header.check_bytes[2] = 0xd; 1716 header.check_bytes[3] = 0xa; 1717 1718 /* write all the data */ 1719 ret = bdrv_pwrite(bs, 0, &magic, sizeof(magic)); 1720 if (ret < 0) { 1721 error_setg(errp, QERR_IO_ERROR); 1722 goto exit; 1723 } 1724 ret = bdrv_pwrite(bs, sizeof(magic), &header, sizeof(header)); 1725 if (ret < 0) { 1726 error_setg(errp, QERR_IO_ERROR); 1727 goto exit; 1728 } 1729 1730 ret = bdrv_truncate(bs, le64_to_cpu(header.grain_offset) << 9); 1731 if (ret < 0) { 1732 error_setg_errno(errp, -ret, "Could not truncate file"); 1733 goto exit; 1734 } 1735 1736 /* write grain directory */ 1737 gd_buf_size = gd_sectors * BDRV_SECTOR_SIZE; 1738 gd_buf = g_malloc0(gd_buf_size); 1739 for (i = 0, tmp = le64_to_cpu(header.rgd_offset) + gd_sectors; 1740 i < gt_count; i++, tmp += gt_size) { 1741 gd_buf[i] = cpu_to_le32(tmp); 1742 } 1743 ret = bdrv_pwrite(bs, le64_to_cpu(header.rgd_offset) * BDRV_SECTOR_SIZE, 1744 gd_buf, gd_buf_size); 1745 if (ret < 0) { 1746 error_setg(errp, QERR_IO_ERROR); 1747 goto exit; 1748 } 1749 1750 /* write backup grain directory */ 1751 for (i = 0, tmp = le64_to_cpu(header.gd_offset) + gd_sectors; 1752 i < gt_count; i++, tmp += gt_size) { 1753 gd_buf[i] = cpu_to_le32(tmp); 1754 } 1755 ret = bdrv_pwrite(bs, le64_to_cpu(header.gd_offset) * BDRV_SECTOR_SIZE, 1756 gd_buf, gd_buf_size); 1757 if (ret < 0) { 1758 error_setg(errp, QERR_IO_ERROR); 1759 goto exit; 1760 } 1761 1762 ret = 0; 1763 exit: 1764 if (bs) { 1765 bdrv_unref(bs); 1766 } 1767 g_free(gd_buf); 1768 return ret; 1769 } 1770 1771 static int filename_decompose(const char *filename, char *path, char *prefix, 1772 char *postfix, size_t buf_len, Error **errp) 1773 { 1774 const char *p, *q; 1775 1776 if (filename == NULL || !strlen(filename)) { 1777 error_setg(errp, "No filename provided"); 1778 return VMDK_ERROR; 1779 } 1780 p = strrchr(filename, '/'); 1781 if (p == NULL) { 1782 p = strrchr(filename, '\\'); 1783 } 1784 if (p == NULL) { 1785 p = strrchr(filename, ':'); 1786 } 1787 if (p != NULL) { 1788 p++; 1789 if (p - filename >= buf_len) { 1790 return VMDK_ERROR; 1791 } 1792 pstrcpy(path, p - filename + 1, filename); 1793 } else { 1794 p = filename; 1795 path[0] = '\0'; 1796 } 1797 q = strrchr(p, '.'); 1798 if (q == NULL) { 1799 pstrcpy(prefix, buf_len, p); 1800 postfix[0] = '\0'; 1801 } else { 1802 if (q - p >= buf_len) { 1803 return VMDK_ERROR; 1804 } 1805 pstrcpy(prefix, q - p + 1, p); 1806 pstrcpy(postfix, buf_len, q); 1807 } 1808 return VMDK_OK; 1809 } 1810 1811 static int vmdk_create(const char *filename, QemuOpts *opts, Error **errp) 1812 { 1813 int idx = 0; 1814 BlockDriverState *new_bs = NULL; 1815 Error *local_err = NULL; 1816 char *desc = NULL; 1817 int64_t total_size = 0, filesize; 1818 char *adapter_type = NULL; 1819 char *backing_file = NULL; 1820 char *fmt = NULL; 1821 int flags = 0; 1822 int ret = 0; 1823 bool flat, split, compress; 1824 GString *ext_desc_lines; 1825 char *path = g_malloc0(PATH_MAX); 1826 char *prefix = g_malloc0(PATH_MAX); 1827 char *postfix = g_malloc0(PATH_MAX); 1828 char *desc_line = g_malloc0(BUF_SIZE); 1829 char *ext_filename = g_malloc0(PATH_MAX); 1830 char *desc_filename = g_malloc0(PATH_MAX); 1831 const int64_t split_size = 0x80000000; /* VMDK has constant split size */ 1832 const char *desc_extent_line; 1833 char *parent_desc_line = g_malloc0(BUF_SIZE); 1834 uint32_t parent_cid = 0xffffffff; 1835 uint32_t number_heads = 16; 1836 bool zeroed_grain = false; 1837 uint32_t desc_offset = 0, desc_len; 1838 const char desc_template[] = 1839 "# Disk DescriptorFile\n" 1840 "version=1\n" 1841 "CID=%" PRIx32 "\n" 1842 "parentCID=%" PRIx32 "\n" 1843 "createType=\"%s\"\n" 1844 "%s" 1845 "\n" 1846 "# Extent description\n" 1847 "%s" 1848 "\n" 1849 "# The Disk Data Base\n" 1850 "#DDB\n" 1851 "\n" 1852 "ddb.virtualHWVersion = \"%d\"\n" 1853 "ddb.geometry.cylinders = \"%" PRId64 "\"\n" 1854 "ddb.geometry.heads = \"%" PRIu32 "\"\n" 1855 "ddb.geometry.sectors = \"63\"\n" 1856 "ddb.adapterType = \"%s\"\n"; 1857 1858 ext_desc_lines = g_string_new(NULL); 1859 1860 if (filename_decompose(filename, path, prefix, postfix, PATH_MAX, errp)) { 1861 ret = -EINVAL; 1862 goto exit; 1863 } 1864 /* Read out options */ 1865 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0), 1866 BDRV_SECTOR_SIZE); 1867 adapter_type = qemu_opt_get_del(opts, BLOCK_OPT_ADAPTER_TYPE); 1868 backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE); 1869 if (qemu_opt_get_bool_del(opts, BLOCK_OPT_COMPAT6, false)) { 1870 flags |= BLOCK_FLAG_COMPAT6; 1871 } 1872 fmt = qemu_opt_get_del(opts, BLOCK_OPT_SUBFMT); 1873 if (qemu_opt_get_bool_del(opts, BLOCK_OPT_ZEROED_GRAIN, false)) { 1874 zeroed_grain = true; 1875 } 1876 1877 if (!adapter_type) { 1878 adapter_type = g_strdup("ide"); 1879 } else if (strcmp(adapter_type, "ide") && 1880 strcmp(adapter_type, "buslogic") && 1881 strcmp(adapter_type, "lsilogic") && 1882 strcmp(adapter_type, "legacyESX")) { 1883 error_setg(errp, "Unknown adapter type: '%s'", adapter_type); 1884 ret = -EINVAL; 1885 goto exit; 1886 } 1887 if (strcmp(adapter_type, "ide") != 0) { 1888 /* that's the number of heads with which vmware operates when 1889 creating, exporting, etc. vmdk files with a non-ide adapter type */ 1890 number_heads = 255; 1891 } 1892 if (!fmt) { 1893 /* Default format to monolithicSparse */ 1894 fmt = g_strdup("monolithicSparse"); 1895 } else if (strcmp(fmt, "monolithicFlat") && 1896 strcmp(fmt, "monolithicSparse") && 1897 strcmp(fmt, "twoGbMaxExtentSparse") && 1898 strcmp(fmt, "twoGbMaxExtentFlat") && 1899 strcmp(fmt, "streamOptimized")) { 1900 error_setg(errp, "Unknown subformat: '%s'", fmt); 1901 ret = -EINVAL; 1902 goto exit; 1903 } 1904 split = !(strcmp(fmt, "twoGbMaxExtentFlat") && 1905 strcmp(fmt, "twoGbMaxExtentSparse")); 1906 flat = !(strcmp(fmt, "monolithicFlat") && 1907 strcmp(fmt, "twoGbMaxExtentFlat")); 1908 compress = !strcmp(fmt, "streamOptimized"); 1909 if (flat) { 1910 desc_extent_line = "RW %" PRId64 " FLAT \"%s\" 0\n"; 1911 } else { 1912 desc_extent_line = "RW %" PRId64 " SPARSE \"%s\"\n"; 1913 } 1914 if (flat && backing_file) { 1915 error_setg(errp, "Flat image can't have backing file"); 1916 ret = -ENOTSUP; 1917 goto exit; 1918 } 1919 if (flat && zeroed_grain) { 1920 error_setg(errp, "Flat image can't enable zeroed grain"); 1921 ret = -ENOTSUP; 1922 goto exit; 1923 } 1924 if (backing_file) { 1925 BlockDriverState *bs = NULL; 1926 char *full_backing = g_new0(char, PATH_MAX); 1927 bdrv_get_full_backing_filename_from_filename(filename, backing_file, 1928 full_backing, PATH_MAX, 1929 &local_err); 1930 if (local_err) { 1931 g_free(full_backing); 1932 error_propagate(errp, local_err); 1933 ret = -ENOENT; 1934 goto exit; 1935 } 1936 ret = bdrv_open(&bs, full_backing, NULL, NULL, BDRV_O_NO_BACKING, errp); 1937 g_free(full_backing); 1938 if (ret != 0) { 1939 goto exit; 1940 } 1941 if (strcmp(bs->drv->format_name, "vmdk")) { 1942 bdrv_unref(bs); 1943 ret = -EINVAL; 1944 goto exit; 1945 } 1946 parent_cid = vmdk_read_cid(bs, 0); 1947 bdrv_unref(bs); 1948 snprintf(parent_desc_line, BUF_SIZE, 1949 "parentFileNameHint=\"%s\"", backing_file); 1950 } 1951 1952 /* Create extents */ 1953 filesize = total_size; 1954 while (filesize > 0) { 1955 int64_t size = filesize; 1956 1957 if (split && size > split_size) { 1958 size = split_size; 1959 } 1960 if (split) { 1961 snprintf(desc_filename, PATH_MAX, "%s-%c%03d%s", 1962 prefix, flat ? 'f' : 's', ++idx, postfix); 1963 } else if (flat) { 1964 snprintf(desc_filename, PATH_MAX, "%s-flat%s", prefix, postfix); 1965 } else { 1966 snprintf(desc_filename, PATH_MAX, "%s%s", prefix, postfix); 1967 } 1968 snprintf(ext_filename, PATH_MAX, "%s%s", path, desc_filename); 1969 1970 if (vmdk_create_extent(ext_filename, size, 1971 flat, compress, zeroed_grain, opts, errp)) { 1972 ret = -EINVAL; 1973 goto exit; 1974 } 1975 filesize -= size; 1976 1977 /* Format description line */ 1978 snprintf(desc_line, BUF_SIZE, 1979 desc_extent_line, size / BDRV_SECTOR_SIZE, desc_filename); 1980 g_string_append(ext_desc_lines, desc_line); 1981 } 1982 /* generate descriptor file */ 1983 desc = g_strdup_printf(desc_template, 1984 g_random_int(), 1985 parent_cid, 1986 fmt, 1987 parent_desc_line, 1988 ext_desc_lines->str, 1989 (flags & BLOCK_FLAG_COMPAT6 ? 6 : 4), 1990 total_size / 1991 (int64_t)(63 * number_heads * BDRV_SECTOR_SIZE), 1992 number_heads, 1993 adapter_type); 1994 desc_len = strlen(desc); 1995 /* the descriptor offset = 0x200 */ 1996 if (!split && !flat) { 1997 desc_offset = 0x200; 1998 } else { 1999 ret = bdrv_create_file(filename, opts, &local_err); 2000 if (ret < 0) { 2001 error_propagate(errp, local_err); 2002 goto exit; 2003 } 2004 } 2005 assert(new_bs == NULL); 2006 ret = bdrv_open(&new_bs, filename, NULL, NULL, 2007 BDRV_O_RDWR | BDRV_O_PROTOCOL, &local_err); 2008 if (ret < 0) { 2009 error_propagate(errp, local_err); 2010 goto exit; 2011 } 2012 ret = bdrv_pwrite(new_bs, desc_offset, desc, desc_len); 2013 if (ret < 0) { 2014 error_setg_errno(errp, -ret, "Could not write description"); 2015 goto exit; 2016 } 2017 /* bdrv_pwrite write padding zeros to align to sector, we don't need that 2018 * for description file */ 2019 if (desc_offset == 0) { 2020 ret = bdrv_truncate(new_bs, desc_len); 2021 if (ret < 0) { 2022 error_setg_errno(errp, -ret, "Could not truncate file"); 2023 } 2024 } 2025 exit: 2026 if (new_bs) { 2027 bdrv_unref(new_bs); 2028 } 2029 g_free(adapter_type); 2030 g_free(backing_file); 2031 g_free(fmt); 2032 g_free(desc); 2033 g_free(path); 2034 g_free(prefix); 2035 g_free(postfix); 2036 g_free(desc_line); 2037 g_free(ext_filename); 2038 g_free(desc_filename); 2039 g_free(parent_desc_line); 2040 g_string_free(ext_desc_lines, true); 2041 return ret; 2042 } 2043 2044 static void vmdk_close(BlockDriverState *bs) 2045 { 2046 BDRVVmdkState *s = bs->opaque; 2047 2048 vmdk_free_extents(bs); 2049 g_free(s->create_type); 2050 2051 migrate_del_blocker(s->migration_blocker); 2052 error_free(s->migration_blocker); 2053 } 2054 2055 static coroutine_fn int vmdk_co_flush(BlockDriverState *bs) 2056 { 2057 BDRVVmdkState *s = bs->opaque; 2058 int i, err; 2059 int ret = 0; 2060 2061 for (i = 0; i < s->num_extents; i++) { 2062 err = bdrv_co_flush(s->extents[i].file->bs); 2063 if (err < 0) { 2064 ret = err; 2065 } 2066 } 2067 return ret; 2068 } 2069 2070 static int64_t vmdk_get_allocated_file_size(BlockDriverState *bs) 2071 { 2072 int i; 2073 int64_t ret = 0; 2074 int64_t r; 2075 BDRVVmdkState *s = bs->opaque; 2076 2077 ret = bdrv_get_allocated_file_size(bs->file->bs); 2078 if (ret < 0) { 2079 return ret; 2080 } 2081 for (i = 0; i < s->num_extents; i++) { 2082 if (s->extents[i].file == bs->file) { 2083 continue; 2084 } 2085 r = bdrv_get_allocated_file_size(s->extents[i].file->bs); 2086 if (r < 0) { 2087 return r; 2088 } 2089 ret += r; 2090 } 2091 return ret; 2092 } 2093 2094 static int vmdk_has_zero_init(BlockDriverState *bs) 2095 { 2096 int i; 2097 BDRVVmdkState *s = bs->opaque; 2098 2099 /* If has a flat extent and its underlying storage doesn't have zero init, 2100 * return 0. */ 2101 for (i = 0; i < s->num_extents; i++) { 2102 if (s->extents[i].flat) { 2103 if (!bdrv_has_zero_init(s->extents[i].file->bs)) { 2104 return 0; 2105 } 2106 } 2107 } 2108 return 1; 2109 } 2110 2111 static ImageInfo *vmdk_get_extent_info(VmdkExtent *extent) 2112 { 2113 ImageInfo *info = g_new0(ImageInfo, 1); 2114 2115 *info = (ImageInfo){ 2116 .filename = g_strdup(extent->file->bs->filename), 2117 .format = g_strdup(extent->type), 2118 .virtual_size = extent->sectors * BDRV_SECTOR_SIZE, 2119 .compressed = extent->compressed, 2120 .has_compressed = extent->compressed, 2121 .cluster_size = extent->cluster_sectors * BDRV_SECTOR_SIZE, 2122 .has_cluster_size = !extent->flat, 2123 }; 2124 2125 return info; 2126 } 2127 2128 static int vmdk_check(BlockDriverState *bs, BdrvCheckResult *result, 2129 BdrvCheckMode fix) 2130 { 2131 BDRVVmdkState *s = bs->opaque; 2132 VmdkExtent *extent = NULL; 2133 int64_t sector_num = 0; 2134 int64_t total_sectors = bdrv_nb_sectors(bs); 2135 int ret; 2136 uint64_t cluster_offset; 2137 2138 if (fix) { 2139 return -ENOTSUP; 2140 } 2141 2142 for (;;) { 2143 if (sector_num >= total_sectors) { 2144 return 0; 2145 } 2146 extent = find_extent(s, sector_num, extent); 2147 if (!extent) { 2148 fprintf(stderr, 2149 "ERROR: could not find extent for sector %" PRId64 "\n", 2150 sector_num); 2151 break; 2152 } 2153 ret = get_cluster_offset(bs, extent, NULL, 2154 sector_num << BDRV_SECTOR_BITS, 2155 false, &cluster_offset, 0, 0); 2156 if (ret == VMDK_ERROR) { 2157 fprintf(stderr, 2158 "ERROR: could not get cluster_offset for sector %" 2159 PRId64 "\n", sector_num); 2160 break; 2161 } 2162 if (ret == VMDK_OK && 2163 cluster_offset >= bdrv_getlength(extent->file->bs)) 2164 { 2165 fprintf(stderr, 2166 "ERROR: cluster offset for sector %" 2167 PRId64 " points after EOF\n", sector_num); 2168 break; 2169 } 2170 sector_num += extent->cluster_sectors; 2171 } 2172 2173 result->corruptions++; 2174 return 0; 2175 } 2176 2177 static ImageInfoSpecific *vmdk_get_specific_info(BlockDriverState *bs) 2178 { 2179 int i; 2180 BDRVVmdkState *s = bs->opaque; 2181 ImageInfoSpecific *spec_info = g_new0(ImageInfoSpecific, 1); 2182 ImageInfoList **next; 2183 2184 *spec_info = (ImageInfoSpecific){ 2185 .type = IMAGE_INFO_SPECIFIC_KIND_VMDK, 2186 { 2187 .vmdk = g_new0(ImageInfoSpecificVmdk, 1), 2188 }, 2189 }; 2190 2191 *spec_info->u.vmdk = (ImageInfoSpecificVmdk) { 2192 .create_type = g_strdup(s->create_type), 2193 .cid = s->cid, 2194 .parent_cid = s->parent_cid, 2195 }; 2196 2197 next = &spec_info->u.vmdk->extents; 2198 for (i = 0; i < s->num_extents; i++) { 2199 *next = g_new0(ImageInfoList, 1); 2200 (*next)->value = vmdk_get_extent_info(&s->extents[i]); 2201 (*next)->next = NULL; 2202 next = &(*next)->next; 2203 } 2204 2205 return spec_info; 2206 } 2207 2208 static bool vmdk_extents_type_eq(const VmdkExtent *a, const VmdkExtent *b) 2209 { 2210 return a->flat == b->flat && 2211 a->compressed == b->compressed && 2212 (a->flat || a->cluster_sectors == b->cluster_sectors); 2213 } 2214 2215 static int vmdk_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) 2216 { 2217 int i; 2218 BDRVVmdkState *s = bs->opaque; 2219 assert(s->num_extents); 2220 2221 /* See if we have multiple extents but they have different cases */ 2222 for (i = 1; i < s->num_extents; i++) { 2223 if (!vmdk_extents_type_eq(&s->extents[0], &s->extents[i])) { 2224 return -ENOTSUP; 2225 } 2226 } 2227 bdi->needs_compressed_writes = s->extents[0].compressed; 2228 if (!s->extents[0].flat) { 2229 bdi->cluster_size = s->extents[0].cluster_sectors << BDRV_SECTOR_BITS; 2230 } 2231 return 0; 2232 } 2233 2234 static void vmdk_detach_aio_context(BlockDriverState *bs) 2235 { 2236 BDRVVmdkState *s = bs->opaque; 2237 int i; 2238 2239 for (i = 0; i < s->num_extents; i++) { 2240 bdrv_detach_aio_context(s->extents[i].file->bs); 2241 } 2242 } 2243 2244 static void vmdk_attach_aio_context(BlockDriverState *bs, 2245 AioContext *new_context) 2246 { 2247 BDRVVmdkState *s = bs->opaque; 2248 int i; 2249 2250 for (i = 0; i < s->num_extents; i++) { 2251 bdrv_attach_aio_context(s->extents[i].file->bs, new_context); 2252 } 2253 } 2254 2255 static QemuOptsList vmdk_create_opts = { 2256 .name = "vmdk-create-opts", 2257 .head = QTAILQ_HEAD_INITIALIZER(vmdk_create_opts.head), 2258 .desc = { 2259 { 2260 .name = BLOCK_OPT_SIZE, 2261 .type = QEMU_OPT_SIZE, 2262 .help = "Virtual disk size" 2263 }, 2264 { 2265 .name = BLOCK_OPT_ADAPTER_TYPE, 2266 .type = QEMU_OPT_STRING, 2267 .help = "Virtual adapter type, can be one of " 2268 "ide (default), lsilogic, buslogic or legacyESX" 2269 }, 2270 { 2271 .name = BLOCK_OPT_BACKING_FILE, 2272 .type = QEMU_OPT_STRING, 2273 .help = "File name of a base image" 2274 }, 2275 { 2276 .name = BLOCK_OPT_COMPAT6, 2277 .type = QEMU_OPT_BOOL, 2278 .help = "VMDK version 6 image", 2279 .def_value_str = "off" 2280 }, 2281 { 2282 .name = BLOCK_OPT_SUBFMT, 2283 .type = QEMU_OPT_STRING, 2284 .help = 2285 "VMDK flat extent format, can be one of " 2286 "{monolithicSparse (default) | monolithicFlat | twoGbMaxExtentSparse | twoGbMaxExtentFlat | streamOptimized} " 2287 }, 2288 { 2289 .name = BLOCK_OPT_ZEROED_GRAIN, 2290 .type = QEMU_OPT_BOOL, 2291 .help = "Enable efficient zero writes " 2292 "using the zeroed-grain GTE feature" 2293 }, 2294 { /* end of list */ } 2295 } 2296 }; 2297 2298 static BlockDriver bdrv_vmdk = { 2299 .format_name = "vmdk", 2300 .instance_size = sizeof(BDRVVmdkState), 2301 .bdrv_probe = vmdk_probe, 2302 .bdrv_open = vmdk_open, 2303 .bdrv_check = vmdk_check, 2304 .bdrv_reopen_prepare = vmdk_reopen_prepare, 2305 .bdrv_read = vmdk_co_read, 2306 .bdrv_write = vmdk_co_write, 2307 .bdrv_write_compressed = vmdk_write_compressed, 2308 .bdrv_co_write_zeroes = vmdk_co_write_zeroes, 2309 .bdrv_close = vmdk_close, 2310 .bdrv_create = vmdk_create, 2311 .bdrv_co_flush_to_disk = vmdk_co_flush, 2312 .bdrv_co_get_block_status = vmdk_co_get_block_status, 2313 .bdrv_get_allocated_file_size = vmdk_get_allocated_file_size, 2314 .bdrv_has_zero_init = vmdk_has_zero_init, 2315 .bdrv_get_specific_info = vmdk_get_specific_info, 2316 .bdrv_refresh_limits = vmdk_refresh_limits, 2317 .bdrv_get_info = vmdk_get_info, 2318 .bdrv_detach_aio_context = vmdk_detach_aio_context, 2319 .bdrv_attach_aio_context = vmdk_attach_aio_context, 2320 2321 .supports_backing = true, 2322 .create_opts = &vmdk_create_opts, 2323 }; 2324 2325 static void bdrv_vmdk_init(void) 2326 { 2327 bdrv_register(&bdrv_vmdk); 2328 } 2329 2330 block_init(bdrv_vmdk_init); 2331