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 "qapi/error.h" 28 #include "block/block_int.h" 29 #include "sysemu/block-backend.h" 30 #include "qapi/qmp/qerror.h" 31 #include "qemu/error-report.h" 32 #include "qemu/module.h" 33 #include "qemu/bswap.h" 34 #include "migration/migration.h" 35 #include "qemu/cutils.h" 36 #include <zlib.h> 37 38 #define VMDK3_MAGIC (('C' << 24) | ('O' << 16) | ('W' << 8) | 'D') 39 #define VMDK4_MAGIC (('K' << 24) | ('D' << 16) | ('M' << 8) | 'V') 40 #define VMDK4_COMPRESSION_DEFLATE 1 41 #define VMDK4_FLAG_NL_DETECT (1 << 0) 42 #define VMDK4_FLAG_RGD (1 << 1) 43 /* Zeroed-grain enable bit */ 44 #define VMDK4_FLAG_ZERO_GRAIN (1 << 2) 45 #define VMDK4_FLAG_COMPRESS (1 << 16) 46 #define VMDK4_FLAG_MARKER (1 << 17) 47 #define VMDK4_GD_AT_END 0xffffffffffffffffULL 48 49 #define VMDK_GTE_ZEROED 0x1 50 51 /* VMDK internal error codes */ 52 #define VMDK_OK 0 53 #define VMDK_ERROR (-1) 54 /* Cluster not allocated */ 55 #define VMDK_UNALLOC (-2) 56 #define VMDK_ZEROED (-3) 57 58 #define BLOCK_OPT_ZEROED_GRAIN "zeroed_grain" 59 60 typedef struct { 61 uint32_t version; 62 uint32_t flags; 63 uint32_t disk_sectors; 64 uint32_t granularity; 65 uint32_t l1dir_offset; 66 uint32_t l1dir_size; 67 uint32_t file_sectors; 68 uint32_t cylinders; 69 uint32_t heads; 70 uint32_t sectors_per_track; 71 } QEMU_PACKED VMDK3Header; 72 73 typedef struct { 74 uint32_t version; 75 uint32_t flags; 76 uint64_t capacity; 77 uint64_t granularity; 78 uint64_t desc_offset; 79 uint64_t desc_size; 80 /* Number of GrainTableEntries per GrainTable */ 81 uint32_t num_gtes_per_gt; 82 uint64_t rgd_offset; 83 uint64_t gd_offset; 84 uint64_t grain_offset; 85 char filler[1]; 86 char check_bytes[4]; 87 uint16_t compressAlgorithm; 88 } QEMU_PACKED VMDK4Header; 89 90 #define L2_CACHE_SIZE 16 91 92 typedef struct VmdkExtent { 93 BdrvChild *file; 94 bool flat; 95 bool compressed; 96 bool has_marker; 97 bool has_zero_grain; 98 int version; 99 int64_t sectors; 100 int64_t end_sector; 101 int64_t flat_start_offset; 102 int64_t l1_table_offset; 103 int64_t l1_backup_table_offset; 104 uint32_t *l1_table; 105 uint32_t *l1_backup_table; 106 unsigned int l1_size; 107 uint32_t l1_entry_sectors; 108 109 unsigned int l2_size; 110 uint32_t *l2_cache; 111 uint32_t l2_cache_offsets[L2_CACHE_SIZE]; 112 uint32_t l2_cache_counts[L2_CACHE_SIZE]; 113 114 int64_t cluster_sectors; 115 int64_t next_cluster_sector; 116 char *type; 117 } VmdkExtent; 118 119 typedef struct BDRVVmdkState { 120 CoMutex lock; 121 uint64_t desc_offset; 122 bool cid_updated; 123 bool cid_checked; 124 uint32_t cid; 125 uint32_t parent_cid; 126 int num_extents; 127 /* Extent array with num_extents entries, ascend ordered by address */ 128 VmdkExtent *extents; 129 Error *migration_blocker; 130 char *create_type; 131 } BDRVVmdkState; 132 133 typedef struct VmdkMetaData { 134 unsigned int l1_index; 135 unsigned int l2_index; 136 unsigned int l2_offset; 137 int valid; 138 uint32_t *l2_cache_entry; 139 } VmdkMetaData; 140 141 typedef struct VmdkGrainMarker { 142 uint64_t lba; 143 uint32_t size; 144 uint8_t data[0]; 145 } QEMU_PACKED VmdkGrainMarker; 146 147 enum { 148 MARKER_END_OF_STREAM = 0, 149 MARKER_GRAIN_TABLE = 1, 150 MARKER_GRAIN_DIRECTORY = 2, 151 MARKER_FOOTER = 3, 152 }; 153 154 static int vmdk_probe(const uint8_t *buf, int buf_size, const char *filename) 155 { 156 uint32_t magic; 157 158 if (buf_size < 4) { 159 return 0; 160 } 161 magic = be32_to_cpu(*(uint32_t *)buf); 162 if (magic == VMDK3_MAGIC || 163 magic == VMDK4_MAGIC) { 164 return 100; 165 } else { 166 const char *p = (const char *)buf; 167 const char *end = p + buf_size; 168 while (p < end) { 169 if (*p == '#') { 170 /* skip comment line */ 171 while (p < end && *p != '\n') { 172 p++; 173 } 174 p++; 175 continue; 176 } 177 if (*p == ' ') { 178 while (p < end && *p == ' ') { 179 p++; 180 } 181 /* skip '\r' if windows line endings used. */ 182 if (p < end && *p == '\r') { 183 p++; 184 } 185 /* only accept blank lines before 'version=' line */ 186 if (p == end || *p != '\n') { 187 return 0; 188 } 189 p++; 190 continue; 191 } 192 if (end - p >= strlen("version=X\n")) { 193 if (strncmp("version=1\n", p, strlen("version=1\n")) == 0 || 194 strncmp("version=2\n", p, strlen("version=2\n")) == 0) { 195 return 100; 196 } 197 } 198 if (end - p >= strlen("version=X\r\n")) { 199 if (strncmp("version=1\r\n", p, strlen("version=1\r\n")) == 0 || 200 strncmp("version=2\r\n", p, strlen("version=2\r\n")) == 0) { 201 return 100; 202 } 203 } 204 return 0; 205 } 206 return 0; 207 } 208 } 209 210 #define SECTOR_SIZE 512 211 #define DESC_SIZE (20 * SECTOR_SIZE) /* 20 sectors of 512 bytes each */ 212 #define BUF_SIZE 4096 213 #define HEADER_SIZE 512 /* first sector of 512 bytes */ 214 215 static void vmdk_free_extents(BlockDriverState *bs) 216 { 217 int i; 218 BDRVVmdkState *s = bs->opaque; 219 VmdkExtent *e; 220 221 for (i = 0; i < s->num_extents; i++) { 222 e = &s->extents[i]; 223 g_free(e->l1_table); 224 g_free(e->l2_cache); 225 g_free(e->l1_backup_table); 226 g_free(e->type); 227 if (e->file != bs->file) { 228 bdrv_unref_child(bs, e->file); 229 } 230 } 231 g_free(s->extents); 232 } 233 234 static void vmdk_free_last_extent(BlockDriverState *bs) 235 { 236 BDRVVmdkState *s = bs->opaque; 237 238 if (s->num_extents == 0) { 239 return; 240 } 241 s->num_extents--; 242 s->extents = g_renew(VmdkExtent, s->extents, s->num_extents); 243 } 244 245 static uint32_t vmdk_read_cid(BlockDriverState *bs, int parent) 246 { 247 char *desc; 248 uint32_t cid = 0xffffffff; 249 const char *p_name, *cid_str; 250 size_t cid_str_size; 251 BDRVVmdkState *s = bs->opaque; 252 int ret; 253 254 desc = g_malloc0(DESC_SIZE); 255 ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE); 256 if (ret < 0) { 257 g_free(desc); 258 return 0; 259 } 260 261 if (parent) { 262 cid_str = "parentCID"; 263 cid_str_size = sizeof("parentCID"); 264 } else { 265 cid_str = "CID"; 266 cid_str_size = sizeof("CID"); 267 } 268 269 desc[DESC_SIZE - 1] = '\0'; 270 p_name = strstr(desc, cid_str); 271 if (p_name != NULL) { 272 p_name += cid_str_size; 273 sscanf(p_name, "%" SCNx32, &cid); 274 } 275 276 g_free(desc); 277 return cid; 278 } 279 280 static int vmdk_write_cid(BlockDriverState *bs, uint32_t cid) 281 { 282 char *desc, *tmp_desc; 283 char *p_name, *tmp_str; 284 BDRVVmdkState *s = bs->opaque; 285 int ret = 0; 286 287 desc = g_malloc0(DESC_SIZE); 288 tmp_desc = g_malloc0(DESC_SIZE); 289 ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE); 290 if (ret < 0) { 291 goto out; 292 } 293 294 desc[DESC_SIZE - 1] = '\0'; 295 tmp_str = strstr(desc, "parentCID"); 296 if (tmp_str == NULL) { 297 ret = -EINVAL; 298 goto out; 299 } 300 301 pstrcpy(tmp_desc, DESC_SIZE, tmp_str); 302 p_name = strstr(desc, "CID"); 303 if (p_name != NULL) { 304 p_name += sizeof("CID"); 305 snprintf(p_name, DESC_SIZE - (p_name - desc), "%" PRIx32 "\n", cid); 306 pstrcat(desc, DESC_SIZE, tmp_desc); 307 } 308 309 ret = bdrv_pwrite_sync(bs->file, s->desc_offset, desc, DESC_SIZE); 310 311 out: 312 g_free(desc); 313 g_free(tmp_desc); 314 return ret; 315 } 316 317 static int vmdk_is_cid_valid(BlockDriverState *bs) 318 { 319 BDRVVmdkState *s = bs->opaque; 320 uint32_t cur_pcid; 321 322 if (!s->cid_checked && bs->backing) { 323 BlockDriverState *p_bs = bs->backing->bs; 324 325 cur_pcid = vmdk_read_cid(p_bs, 0); 326 if (s->parent_cid != cur_pcid) { 327 /* CID not valid */ 328 return 0; 329 } 330 } 331 s->cid_checked = true; 332 /* CID valid */ 333 return 1; 334 } 335 336 /* We have nothing to do for VMDK reopen, stubs just return success */ 337 static int vmdk_reopen_prepare(BDRVReopenState *state, 338 BlockReopenQueue *queue, Error **errp) 339 { 340 assert(state != NULL); 341 assert(state->bs != NULL); 342 return 0; 343 } 344 345 static int vmdk_parent_open(BlockDriverState *bs) 346 { 347 char *p_name; 348 char *desc; 349 BDRVVmdkState *s = bs->opaque; 350 int ret; 351 352 desc = g_malloc0(DESC_SIZE + 1); 353 ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE); 354 if (ret < 0) { 355 goto out; 356 } 357 ret = 0; 358 359 p_name = strstr(desc, "parentFileNameHint"); 360 if (p_name != NULL) { 361 char *end_name; 362 363 p_name += sizeof("parentFileNameHint") + 1; 364 end_name = strchr(p_name, '\"'); 365 if (end_name == NULL) { 366 ret = -EINVAL; 367 goto out; 368 } 369 if ((end_name - p_name) > sizeof(bs->backing_file) - 1) { 370 ret = -EINVAL; 371 goto out; 372 } 373 374 pstrcpy(bs->backing_file, end_name - p_name + 1, p_name); 375 } 376 377 out: 378 g_free(desc); 379 return ret; 380 } 381 382 /* Create and append extent to the extent array. Return the added VmdkExtent 383 * address. return NULL if allocation failed. */ 384 static int vmdk_add_extent(BlockDriverState *bs, 385 BdrvChild *file, bool flat, int64_t sectors, 386 int64_t l1_offset, int64_t l1_backup_offset, 387 uint32_t l1_size, 388 int l2_size, uint64_t cluster_sectors, 389 VmdkExtent **new_extent, 390 Error **errp) 391 { 392 VmdkExtent *extent; 393 BDRVVmdkState *s = bs->opaque; 394 int64_t nb_sectors; 395 396 if (cluster_sectors > 0x200000) { 397 /* 0x200000 * 512Bytes = 1GB for one cluster is unrealistic */ 398 error_setg(errp, "Invalid granularity, image may be corrupt"); 399 return -EFBIG; 400 } 401 if (l1_size > 512 * 1024 * 1024) { 402 /* Although with big capacity and small l1_entry_sectors, we can get a 403 * big l1_size, we don't want unbounded value to allocate the table. 404 * Limit it to 512M, which is 16PB for default cluster and L2 table 405 * size */ 406 error_setg(errp, "L1 size too big"); 407 return -EFBIG; 408 } 409 410 nb_sectors = bdrv_nb_sectors(file->bs); 411 if (nb_sectors < 0) { 412 return nb_sectors; 413 } 414 415 s->extents = g_renew(VmdkExtent, s->extents, s->num_extents + 1); 416 extent = &s->extents[s->num_extents]; 417 s->num_extents++; 418 419 memset(extent, 0, sizeof(VmdkExtent)); 420 extent->file = file; 421 extent->flat = flat; 422 extent->sectors = sectors; 423 extent->l1_table_offset = l1_offset; 424 extent->l1_backup_table_offset = l1_backup_offset; 425 extent->l1_size = l1_size; 426 extent->l1_entry_sectors = l2_size * cluster_sectors; 427 extent->l2_size = l2_size; 428 extent->cluster_sectors = flat ? sectors : cluster_sectors; 429 extent->next_cluster_sector = ROUND_UP(nb_sectors, cluster_sectors); 430 431 if (s->num_extents > 1) { 432 extent->end_sector = (*(extent - 1)).end_sector + extent->sectors; 433 } else { 434 extent->end_sector = extent->sectors; 435 } 436 bs->total_sectors = extent->end_sector; 437 if (new_extent) { 438 *new_extent = extent; 439 } 440 return 0; 441 } 442 443 static int vmdk_init_tables(BlockDriverState *bs, VmdkExtent *extent, 444 Error **errp) 445 { 446 int ret; 447 size_t l1_size; 448 int i; 449 450 /* read the L1 table */ 451 l1_size = extent->l1_size * sizeof(uint32_t); 452 extent->l1_table = g_try_malloc(l1_size); 453 if (l1_size && extent->l1_table == NULL) { 454 return -ENOMEM; 455 } 456 457 ret = bdrv_pread(extent->file, 458 extent->l1_table_offset, 459 extent->l1_table, 460 l1_size); 461 if (ret < 0) { 462 error_setg_errno(errp, -ret, 463 "Could not read l1 table from extent '%s'", 464 extent->file->bs->filename); 465 goto fail_l1; 466 } 467 for (i = 0; i < extent->l1_size; i++) { 468 le32_to_cpus(&extent->l1_table[i]); 469 } 470 471 if (extent->l1_backup_table_offset) { 472 extent->l1_backup_table = g_try_malloc(l1_size); 473 if (l1_size && extent->l1_backup_table == NULL) { 474 ret = -ENOMEM; 475 goto fail_l1; 476 } 477 ret = bdrv_pread(extent->file, 478 extent->l1_backup_table_offset, 479 extent->l1_backup_table, 480 l1_size); 481 if (ret < 0) { 482 error_setg_errno(errp, -ret, 483 "Could not read l1 backup table from extent '%s'", 484 extent->file->bs->filename); 485 goto fail_l1b; 486 } 487 for (i = 0; i < extent->l1_size; i++) { 488 le32_to_cpus(&extent->l1_backup_table[i]); 489 } 490 } 491 492 extent->l2_cache = 493 g_new(uint32_t, extent->l2_size * L2_CACHE_SIZE); 494 return 0; 495 fail_l1b: 496 g_free(extent->l1_backup_table); 497 fail_l1: 498 g_free(extent->l1_table); 499 return ret; 500 } 501 502 static int vmdk_open_vmfs_sparse(BlockDriverState *bs, 503 BdrvChild *file, 504 int flags, Error **errp) 505 { 506 int ret; 507 uint32_t magic; 508 VMDK3Header header; 509 VmdkExtent *extent; 510 511 ret = bdrv_pread(file, sizeof(magic), &header, sizeof(header)); 512 if (ret < 0) { 513 error_setg_errno(errp, -ret, 514 "Could not read header from file '%s'", 515 file->bs->filename); 516 return ret; 517 } 518 ret = vmdk_add_extent(bs, file, false, 519 le32_to_cpu(header.disk_sectors), 520 (int64_t)le32_to_cpu(header.l1dir_offset) << 9, 521 0, 522 le32_to_cpu(header.l1dir_size), 523 4096, 524 le32_to_cpu(header.granularity), 525 &extent, 526 errp); 527 if (ret < 0) { 528 return ret; 529 } 530 ret = vmdk_init_tables(bs, extent, errp); 531 if (ret) { 532 /* free extent allocated by vmdk_add_extent */ 533 vmdk_free_last_extent(bs); 534 } 535 return ret; 536 } 537 538 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf, 539 QDict *options, Error **errp); 540 541 static char *vmdk_read_desc(BdrvChild *file, uint64_t desc_offset, Error **errp) 542 { 543 int64_t size; 544 char *buf; 545 int ret; 546 547 size = bdrv_getlength(file->bs); 548 if (size < 0) { 549 error_setg_errno(errp, -size, "Could not access file"); 550 return NULL; 551 } 552 553 if (size < 4) { 554 /* Both descriptor file and sparse image must be much larger than 4 555 * bytes, also callers of vmdk_read_desc want to compare the first 4 556 * bytes with VMDK4_MAGIC, let's error out if less is read. */ 557 error_setg(errp, "File is too small, not a valid image"); 558 return NULL; 559 } 560 561 size = MIN(size, (1 << 20) - 1); /* avoid unbounded allocation */ 562 buf = g_malloc(size + 1); 563 564 ret = bdrv_pread(file, desc_offset, buf, size); 565 if (ret < 0) { 566 error_setg_errno(errp, -ret, "Could not read from file"); 567 g_free(buf); 568 return NULL; 569 } 570 buf[ret] = 0; 571 572 return buf; 573 } 574 575 static int vmdk_open_vmdk4(BlockDriverState *bs, 576 BdrvChild *file, 577 int flags, QDict *options, Error **errp) 578 { 579 int ret; 580 uint32_t magic; 581 uint32_t l1_size, l1_entry_sectors; 582 VMDK4Header header; 583 VmdkExtent *extent; 584 BDRVVmdkState *s = bs->opaque; 585 int64_t l1_backup_offset = 0; 586 bool compressed; 587 588 ret = bdrv_pread(file, sizeof(magic), &header, sizeof(header)); 589 if (ret < 0) { 590 error_setg_errno(errp, -ret, 591 "Could not read header from file '%s'", 592 file->bs->filename); 593 return -EINVAL; 594 } 595 if (header.capacity == 0) { 596 uint64_t desc_offset = le64_to_cpu(header.desc_offset); 597 if (desc_offset) { 598 char *buf = vmdk_read_desc(file, desc_offset << 9, errp); 599 if (!buf) { 600 return -EINVAL; 601 } 602 ret = vmdk_open_desc_file(bs, flags, buf, options, errp); 603 g_free(buf); 604 return ret; 605 } 606 } 607 608 if (!s->create_type) { 609 s->create_type = g_strdup("monolithicSparse"); 610 } 611 612 if (le64_to_cpu(header.gd_offset) == VMDK4_GD_AT_END) { 613 /* 614 * The footer takes precedence over the header, so read it in. The 615 * footer starts at offset -1024 from the end: One sector for the 616 * footer, and another one for the end-of-stream marker. 617 */ 618 struct { 619 struct { 620 uint64_t val; 621 uint32_t size; 622 uint32_t type; 623 uint8_t pad[512 - 16]; 624 } QEMU_PACKED footer_marker; 625 626 uint32_t magic; 627 VMDK4Header header; 628 uint8_t pad[512 - 4 - sizeof(VMDK4Header)]; 629 630 struct { 631 uint64_t val; 632 uint32_t size; 633 uint32_t type; 634 uint8_t pad[512 - 16]; 635 } QEMU_PACKED eos_marker; 636 } QEMU_PACKED footer; 637 638 ret = bdrv_pread(file, 639 bs->file->bs->total_sectors * 512 - 1536, 640 &footer, sizeof(footer)); 641 if (ret < 0) { 642 error_setg_errno(errp, -ret, "Failed to read footer"); 643 return ret; 644 } 645 646 /* Some sanity checks for the footer */ 647 if (be32_to_cpu(footer.magic) != VMDK4_MAGIC || 648 le32_to_cpu(footer.footer_marker.size) != 0 || 649 le32_to_cpu(footer.footer_marker.type) != MARKER_FOOTER || 650 le64_to_cpu(footer.eos_marker.val) != 0 || 651 le32_to_cpu(footer.eos_marker.size) != 0 || 652 le32_to_cpu(footer.eos_marker.type) != MARKER_END_OF_STREAM) 653 { 654 error_setg(errp, "Invalid footer"); 655 return -EINVAL; 656 } 657 658 header = footer.header; 659 } 660 661 compressed = 662 le16_to_cpu(header.compressAlgorithm) == VMDK4_COMPRESSION_DEFLATE; 663 if (le32_to_cpu(header.version) > 3) { 664 error_setg(errp, "Unsupported VMDK version %" PRIu32, 665 le32_to_cpu(header.version)); 666 return -ENOTSUP; 667 } else if (le32_to_cpu(header.version) == 3 && (flags & BDRV_O_RDWR) && 668 !compressed) { 669 /* VMware KB 2064959 explains that version 3 added support for 670 * persistent changed block tracking (CBT), and backup software can 671 * read it as version=1 if it doesn't care about the changed area 672 * information. So we are safe to enable read only. */ 673 error_setg(errp, "VMDK version 3 must be read only"); 674 return -EINVAL; 675 } 676 677 if (le32_to_cpu(header.num_gtes_per_gt) > 512) { 678 error_setg(errp, "L2 table size too big"); 679 return -EINVAL; 680 } 681 682 l1_entry_sectors = le32_to_cpu(header.num_gtes_per_gt) 683 * le64_to_cpu(header.granularity); 684 if (l1_entry_sectors == 0) { 685 error_setg(errp, "L1 entry size is invalid"); 686 return -EINVAL; 687 } 688 l1_size = (le64_to_cpu(header.capacity) + l1_entry_sectors - 1) 689 / l1_entry_sectors; 690 if (le32_to_cpu(header.flags) & VMDK4_FLAG_RGD) { 691 l1_backup_offset = le64_to_cpu(header.rgd_offset) << 9; 692 } 693 if (bdrv_nb_sectors(file->bs) < le64_to_cpu(header.grain_offset)) { 694 error_setg(errp, "File truncated, expecting at least %" PRId64 " bytes", 695 (int64_t)(le64_to_cpu(header.grain_offset) 696 * BDRV_SECTOR_SIZE)); 697 return -EINVAL; 698 } 699 700 ret = vmdk_add_extent(bs, file, false, 701 le64_to_cpu(header.capacity), 702 le64_to_cpu(header.gd_offset) << 9, 703 l1_backup_offset, 704 l1_size, 705 le32_to_cpu(header.num_gtes_per_gt), 706 le64_to_cpu(header.granularity), 707 &extent, 708 errp); 709 if (ret < 0) { 710 return ret; 711 } 712 extent->compressed = 713 le16_to_cpu(header.compressAlgorithm) == VMDK4_COMPRESSION_DEFLATE; 714 if (extent->compressed) { 715 g_free(s->create_type); 716 s->create_type = g_strdup("streamOptimized"); 717 } 718 extent->has_marker = le32_to_cpu(header.flags) & VMDK4_FLAG_MARKER; 719 extent->version = le32_to_cpu(header.version); 720 extent->has_zero_grain = le32_to_cpu(header.flags) & VMDK4_FLAG_ZERO_GRAIN; 721 ret = vmdk_init_tables(bs, extent, errp); 722 if (ret) { 723 /* free extent allocated by vmdk_add_extent */ 724 vmdk_free_last_extent(bs); 725 } 726 return ret; 727 } 728 729 /* find an option value out of descriptor file */ 730 static int vmdk_parse_description(const char *desc, const char *opt_name, 731 char *buf, int buf_size) 732 { 733 char *opt_pos, *opt_end; 734 const char *end = desc + strlen(desc); 735 736 opt_pos = strstr(desc, opt_name); 737 if (!opt_pos) { 738 return VMDK_ERROR; 739 } 740 /* Skip "=\"" following opt_name */ 741 opt_pos += strlen(opt_name) + 2; 742 if (opt_pos >= end) { 743 return VMDK_ERROR; 744 } 745 opt_end = opt_pos; 746 while (opt_end < end && *opt_end != '"') { 747 opt_end++; 748 } 749 if (opt_end == end || buf_size < opt_end - opt_pos + 1) { 750 return VMDK_ERROR; 751 } 752 pstrcpy(buf, opt_end - opt_pos + 1, opt_pos); 753 return VMDK_OK; 754 } 755 756 /* Open an extent file and append to bs array */ 757 static int vmdk_open_sparse(BlockDriverState *bs, BdrvChild *file, int flags, 758 char *buf, QDict *options, Error **errp) 759 { 760 uint32_t magic; 761 762 magic = ldl_be_p(buf); 763 switch (magic) { 764 case VMDK3_MAGIC: 765 return vmdk_open_vmfs_sparse(bs, file, flags, errp); 766 break; 767 case VMDK4_MAGIC: 768 return vmdk_open_vmdk4(bs, file, flags, options, errp); 769 break; 770 default: 771 error_setg(errp, "Image not in VMDK format"); 772 return -EINVAL; 773 break; 774 } 775 } 776 777 static const char *next_line(const char *s) 778 { 779 while (*s) { 780 if (*s == '\n') { 781 return s + 1; 782 } 783 s++; 784 } 785 return s; 786 } 787 788 static int vmdk_parse_extents(const char *desc, BlockDriverState *bs, 789 const char *desc_file_path, QDict *options, 790 Error **errp) 791 { 792 int ret; 793 int matches; 794 char access[11]; 795 char type[11]; 796 char fname[512]; 797 const char *p, *np; 798 int64_t sectors = 0; 799 int64_t flat_offset; 800 char *extent_path; 801 BdrvChild *extent_file; 802 BDRVVmdkState *s = bs->opaque; 803 VmdkExtent *extent; 804 char extent_opt_prefix[32]; 805 Error *local_err = NULL; 806 807 for (p = desc; *p; p = next_line(p)) { 808 /* parse extent line in one of below formats: 809 * 810 * RW [size in sectors] FLAT "file-name.vmdk" OFFSET 811 * RW [size in sectors] SPARSE "file-name.vmdk" 812 * RW [size in sectors] VMFS "file-name.vmdk" 813 * RW [size in sectors] VMFSSPARSE "file-name.vmdk" 814 */ 815 flat_offset = -1; 816 matches = sscanf(p, "%10s %" SCNd64 " %10s \"%511[^\n\r\"]\" %" SCNd64, 817 access, §ors, type, fname, &flat_offset); 818 if (matches < 4 || strcmp(access, "RW")) { 819 continue; 820 } else if (!strcmp(type, "FLAT")) { 821 if (matches != 5 || flat_offset < 0) { 822 goto invalid; 823 } 824 } else if (!strcmp(type, "VMFS")) { 825 if (matches == 4) { 826 flat_offset = 0; 827 } else { 828 goto invalid; 829 } 830 } else if (matches != 4) { 831 goto invalid; 832 } 833 834 if (sectors <= 0 || 835 (strcmp(type, "FLAT") && strcmp(type, "SPARSE") && 836 strcmp(type, "VMFS") && strcmp(type, "VMFSSPARSE")) || 837 (strcmp(access, "RW"))) { 838 continue; 839 } 840 841 if (!path_is_absolute(fname) && !path_has_protocol(fname) && 842 !desc_file_path[0]) 843 { 844 error_setg(errp, "Cannot use relative extent paths with VMDK " 845 "descriptor file '%s'", bs->file->bs->filename); 846 return -EINVAL; 847 } 848 849 extent_path = g_malloc0(PATH_MAX); 850 path_combine(extent_path, PATH_MAX, desc_file_path, fname); 851 852 ret = snprintf(extent_opt_prefix, 32, "extents.%d", s->num_extents); 853 assert(ret < 32); 854 855 extent_file = bdrv_open_child(extent_path, options, extent_opt_prefix, 856 bs, &child_file, false, &local_err); 857 g_free(extent_path); 858 if (local_err) { 859 error_propagate(errp, local_err); 860 return -EINVAL; 861 } 862 863 /* save to extents array */ 864 if (!strcmp(type, "FLAT") || !strcmp(type, "VMFS")) { 865 /* FLAT extent */ 866 867 ret = vmdk_add_extent(bs, extent_file, true, sectors, 868 0, 0, 0, 0, 0, &extent, errp); 869 if (ret < 0) { 870 bdrv_unref_child(bs, extent_file); 871 return ret; 872 } 873 extent->flat_start_offset = flat_offset << 9; 874 } else if (!strcmp(type, "SPARSE") || !strcmp(type, "VMFSSPARSE")) { 875 /* SPARSE extent and VMFSSPARSE extent are both "COWD" sparse file*/ 876 char *buf = vmdk_read_desc(extent_file, 0, errp); 877 if (!buf) { 878 ret = -EINVAL; 879 } else { 880 ret = vmdk_open_sparse(bs, extent_file, bs->open_flags, buf, 881 options, errp); 882 } 883 g_free(buf); 884 if (ret) { 885 bdrv_unref_child(bs, extent_file); 886 return ret; 887 } 888 extent = &s->extents[s->num_extents - 1]; 889 } else { 890 error_setg(errp, "Unsupported extent type '%s'", type); 891 bdrv_unref_child(bs, extent_file); 892 return -ENOTSUP; 893 } 894 extent->type = g_strdup(type); 895 } 896 return 0; 897 898 invalid: 899 np = next_line(p); 900 assert(np != p); 901 if (np[-1] == '\n') { 902 np--; 903 } 904 error_setg(errp, "Invalid extent line: %.*s", (int)(np - p), p); 905 return -EINVAL; 906 } 907 908 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf, 909 QDict *options, Error **errp) 910 { 911 int ret; 912 char ct[128]; 913 BDRVVmdkState *s = bs->opaque; 914 915 if (vmdk_parse_description(buf, "createType", ct, sizeof(ct))) { 916 error_setg(errp, "invalid VMDK image descriptor"); 917 ret = -EINVAL; 918 goto exit; 919 } 920 if (strcmp(ct, "monolithicFlat") && 921 strcmp(ct, "vmfs") && 922 strcmp(ct, "vmfsSparse") && 923 strcmp(ct, "twoGbMaxExtentSparse") && 924 strcmp(ct, "twoGbMaxExtentFlat")) { 925 error_setg(errp, "Unsupported image type '%s'", ct); 926 ret = -ENOTSUP; 927 goto exit; 928 } 929 s->create_type = g_strdup(ct); 930 s->desc_offset = 0; 931 ret = vmdk_parse_extents(buf, bs, bs->file->bs->exact_filename, options, 932 errp); 933 exit: 934 return ret; 935 } 936 937 static int vmdk_open(BlockDriverState *bs, QDict *options, int flags, 938 Error **errp) 939 { 940 char *buf; 941 int ret; 942 BDRVVmdkState *s = bs->opaque; 943 uint32_t magic; 944 Error *local_err = NULL; 945 946 buf = vmdk_read_desc(bs->file, 0, errp); 947 if (!buf) { 948 return -EINVAL; 949 } 950 951 magic = ldl_be_p(buf); 952 switch (magic) { 953 case VMDK3_MAGIC: 954 case VMDK4_MAGIC: 955 ret = vmdk_open_sparse(bs, bs->file, flags, buf, options, 956 errp); 957 s->desc_offset = 0x200; 958 break; 959 default: 960 ret = vmdk_open_desc_file(bs, flags, buf, options, errp); 961 break; 962 } 963 if (ret) { 964 goto fail; 965 } 966 967 /* try to open parent images, if exist */ 968 ret = vmdk_parent_open(bs); 969 if (ret) { 970 goto fail; 971 } 972 s->cid = vmdk_read_cid(bs, 0); 973 s->parent_cid = vmdk_read_cid(bs, 1); 974 qemu_co_mutex_init(&s->lock); 975 976 /* Disable migration when VMDK images are used */ 977 error_setg(&s->migration_blocker, "The vmdk format used by node '%s' " 978 "does not support live migration", 979 bdrv_get_device_or_node_name(bs)); 980 ret = migrate_add_blocker(s->migration_blocker, &local_err); 981 if (local_err) { 982 error_propagate(errp, local_err); 983 error_free(s->migration_blocker); 984 goto fail; 985 } 986 987 g_free(buf); 988 return 0; 989 990 fail: 991 g_free(buf); 992 g_free(s->create_type); 993 s->create_type = NULL; 994 vmdk_free_extents(bs); 995 return ret; 996 } 997 998 999 static void vmdk_refresh_limits(BlockDriverState *bs, Error **errp) 1000 { 1001 BDRVVmdkState *s = bs->opaque; 1002 int i; 1003 1004 for (i = 0; i < s->num_extents; i++) { 1005 if (!s->extents[i].flat) { 1006 bs->bl.pwrite_zeroes_alignment = 1007 MAX(bs->bl.pwrite_zeroes_alignment, 1008 s->extents[i].cluster_sectors << BDRV_SECTOR_BITS); 1009 } 1010 } 1011 } 1012 1013 /** 1014 * get_whole_cluster 1015 * 1016 * Copy backing file's cluster that covers @sector_num, otherwise write zero, 1017 * to the cluster at @cluster_sector_num. 1018 * 1019 * If @skip_start_sector < @skip_end_sector, the relative range 1020 * [@skip_start_sector, @skip_end_sector) is not copied or written, and leave 1021 * it for call to write user data in the request. 1022 */ 1023 static int get_whole_cluster(BlockDriverState *bs, 1024 VmdkExtent *extent, 1025 uint64_t cluster_offset, 1026 uint64_t offset, 1027 uint64_t skip_start_bytes, 1028 uint64_t skip_end_bytes) 1029 { 1030 int ret = VMDK_OK; 1031 int64_t cluster_bytes; 1032 uint8_t *whole_grain; 1033 1034 /* For COW, align request sector_num to cluster start */ 1035 cluster_bytes = extent->cluster_sectors << BDRV_SECTOR_BITS; 1036 offset = QEMU_ALIGN_DOWN(offset, cluster_bytes); 1037 whole_grain = qemu_blockalign(bs, cluster_bytes); 1038 1039 if (!bs->backing) { 1040 memset(whole_grain, 0, skip_start_bytes); 1041 memset(whole_grain + skip_end_bytes, 0, cluster_bytes - skip_end_bytes); 1042 } 1043 1044 assert(skip_end_bytes <= cluster_bytes); 1045 /* we will be here if it's first write on non-exist grain(cluster). 1046 * try to read from parent image, if exist */ 1047 if (bs->backing && !vmdk_is_cid_valid(bs)) { 1048 ret = VMDK_ERROR; 1049 goto exit; 1050 } 1051 1052 /* Read backing data before skip range */ 1053 if (skip_start_bytes > 0) { 1054 if (bs->backing) { 1055 ret = bdrv_pread(bs->backing, offset, whole_grain, 1056 skip_start_bytes); 1057 if (ret < 0) { 1058 ret = VMDK_ERROR; 1059 goto exit; 1060 } 1061 } 1062 ret = bdrv_pwrite(extent->file, cluster_offset, whole_grain, 1063 skip_start_bytes); 1064 if (ret < 0) { 1065 ret = VMDK_ERROR; 1066 goto exit; 1067 } 1068 } 1069 /* Read backing data after skip range */ 1070 if (skip_end_bytes < cluster_bytes) { 1071 if (bs->backing) { 1072 ret = bdrv_pread(bs->backing, offset + skip_end_bytes, 1073 whole_grain + skip_end_bytes, 1074 cluster_bytes - skip_end_bytes); 1075 if (ret < 0) { 1076 ret = VMDK_ERROR; 1077 goto exit; 1078 } 1079 } 1080 ret = bdrv_pwrite(extent->file, cluster_offset + skip_end_bytes, 1081 whole_grain + skip_end_bytes, 1082 cluster_bytes - skip_end_bytes); 1083 if (ret < 0) { 1084 ret = VMDK_ERROR; 1085 goto exit; 1086 } 1087 } 1088 1089 ret = VMDK_OK; 1090 exit: 1091 qemu_vfree(whole_grain); 1092 return ret; 1093 } 1094 1095 static int vmdk_L2update(VmdkExtent *extent, VmdkMetaData *m_data, 1096 uint32_t offset) 1097 { 1098 offset = cpu_to_le32(offset); 1099 /* update L2 table */ 1100 if (bdrv_pwrite_sync(extent->file, 1101 ((int64_t)m_data->l2_offset * 512) 1102 + (m_data->l2_index * sizeof(offset)), 1103 &offset, sizeof(offset)) < 0) { 1104 return VMDK_ERROR; 1105 } 1106 /* update backup L2 table */ 1107 if (extent->l1_backup_table_offset != 0) { 1108 m_data->l2_offset = extent->l1_backup_table[m_data->l1_index]; 1109 if (bdrv_pwrite_sync(extent->file, 1110 ((int64_t)m_data->l2_offset * 512) 1111 + (m_data->l2_index * sizeof(offset)), 1112 &offset, sizeof(offset)) < 0) { 1113 return VMDK_ERROR; 1114 } 1115 } 1116 if (m_data->l2_cache_entry) { 1117 *m_data->l2_cache_entry = offset; 1118 } 1119 1120 return VMDK_OK; 1121 } 1122 1123 /** 1124 * get_cluster_offset 1125 * 1126 * Look up cluster offset in extent file by sector number, and store in 1127 * @cluster_offset. 1128 * 1129 * For flat extents, the start offset as parsed from the description file is 1130 * returned. 1131 * 1132 * For sparse extents, look up in L1, L2 table. If allocate is true, return an 1133 * offset for a new cluster and update L2 cache. If there is a backing file, 1134 * COW is done before returning; otherwise, zeroes are written to the allocated 1135 * cluster. Both COW and zero writing skips the sector range 1136 * [@skip_start_sector, @skip_end_sector) passed in by caller, because caller 1137 * has new data to write there. 1138 * 1139 * Returns: VMDK_OK if cluster exists and mapped in the image. 1140 * VMDK_UNALLOC if cluster is not mapped and @allocate is false. 1141 * VMDK_ERROR if failed. 1142 */ 1143 static int get_cluster_offset(BlockDriverState *bs, 1144 VmdkExtent *extent, 1145 VmdkMetaData *m_data, 1146 uint64_t offset, 1147 bool allocate, 1148 uint64_t *cluster_offset, 1149 uint64_t skip_start_bytes, 1150 uint64_t skip_end_bytes) 1151 { 1152 unsigned int l1_index, l2_offset, l2_index; 1153 int min_index, i, j; 1154 uint32_t min_count, *l2_table; 1155 bool zeroed = false; 1156 int64_t ret; 1157 int64_t cluster_sector; 1158 1159 if (m_data) { 1160 m_data->valid = 0; 1161 } 1162 if (extent->flat) { 1163 *cluster_offset = extent->flat_start_offset; 1164 return VMDK_OK; 1165 } 1166 1167 offset -= (extent->end_sector - extent->sectors) * SECTOR_SIZE; 1168 l1_index = (offset >> 9) / extent->l1_entry_sectors; 1169 if (l1_index >= extent->l1_size) { 1170 return VMDK_ERROR; 1171 } 1172 l2_offset = extent->l1_table[l1_index]; 1173 if (!l2_offset) { 1174 return VMDK_UNALLOC; 1175 } 1176 for (i = 0; i < L2_CACHE_SIZE; i++) { 1177 if (l2_offset == extent->l2_cache_offsets[i]) { 1178 /* increment the hit count */ 1179 if (++extent->l2_cache_counts[i] == 0xffffffff) { 1180 for (j = 0; j < L2_CACHE_SIZE; j++) { 1181 extent->l2_cache_counts[j] >>= 1; 1182 } 1183 } 1184 l2_table = extent->l2_cache + (i * extent->l2_size); 1185 goto found; 1186 } 1187 } 1188 /* not found: load a new entry in the least used one */ 1189 min_index = 0; 1190 min_count = 0xffffffff; 1191 for (i = 0; i < L2_CACHE_SIZE; i++) { 1192 if (extent->l2_cache_counts[i] < min_count) { 1193 min_count = extent->l2_cache_counts[i]; 1194 min_index = i; 1195 } 1196 } 1197 l2_table = extent->l2_cache + (min_index * extent->l2_size); 1198 if (bdrv_pread(extent->file, 1199 (int64_t)l2_offset * 512, 1200 l2_table, 1201 extent->l2_size * sizeof(uint32_t) 1202 ) != extent->l2_size * sizeof(uint32_t)) { 1203 return VMDK_ERROR; 1204 } 1205 1206 extent->l2_cache_offsets[min_index] = l2_offset; 1207 extent->l2_cache_counts[min_index] = 1; 1208 found: 1209 l2_index = ((offset >> 9) / extent->cluster_sectors) % extent->l2_size; 1210 cluster_sector = le32_to_cpu(l2_table[l2_index]); 1211 1212 if (extent->has_zero_grain && cluster_sector == VMDK_GTE_ZEROED) { 1213 zeroed = true; 1214 } 1215 1216 if (!cluster_sector || zeroed) { 1217 if (!allocate) { 1218 return zeroed ? VMDK_ZEROED : VMDK_UNALLOC; 1219 } 1220 1221 cluster_sector = extent->next_cluster_sector; 1222 extent->next_cluster_sector += extent->cluster_sectors; 1223 1224 /* First of all we write grain itself, to avoid race condition 1225 * that may to corrupt the image. 1226 * This problem may occur because of insufficient space on host disk 1227 * or inappropriate VM shutdown. 1228 */ 1229 ret = get_whole_cluster(bs, extent, cluster_sector * BDRV_SECTOR_SIZE, 1230 offset, skip_start_bytes, skip_end_bytes); 1231 if (ret) { 1232 return ret; 1233 } 1234 if (m_data) { 1235 m_data->valid = 1; 1236 m_data->l1_index = l1_index; 1237 m_data->l2_index = l2_index; 1238 m_data->l2_offset = l2_offset; 1239 m_data->l2_cache_entry = &l2_table[l2_index]; 1240 } 1241 } 1242 *cluster_offset = cluster_sector << BDRV_SECTOR_BITS; 1243 return VMDK_OK; 1244 } 1245 1246 static VmdkExtent *find_extent(BDRVVmdkState *s, 1247 int64_t sector_num, VmdkExtent *start_hint) 1248 { 1249 VmdkExtent *extent = start_hint; 1250 1251 if (!extent) { 1252 extent = &s->extents[0]; 1253 } 1254 while (extent < &s->extents[s->num_extents]) { 1255 if (sector_num < extent->end_sector) { 1256 return extent; 1257 } 1258 extent++; 1259 } 1260 return NULL; 1261 } 1262 1263 static inline uint64_t vmdk_find_offset_in_cluster(VmdkExtent *extent, 1264 int64_t offset) 1265 { 1266 uint64_t extent_begin_offset, extent_relative_offset; 1267 uint64_t cluster_size = extent->cluster_sectors * BDRV_SECTOR_SIZE; 1268 1269 extent_begin_offset = 1270 (extent->end_sector - extent->sectors) * BDRV_SECTOR_SIZE; 1271 extent_relative_offset = offset - extent_begin_offset; 1272 return extent_relative_offset % cluster_size; 1273 } 1274 1275 static inline uint64_t vmdk_find_index_in_cluster(VmdkExtent *extent, 1276 int64_t sector_num) 1277 { 1278 uint64_t offset; 1279 offset = vmdk_find_offset_in_cluster(extent, sector_num * BDRV_SECTOR_SIZE); 1280 return offset / BDRV_SECTOR_SIZE; 1281 } 1282 1283 static int64_t coroutine_fn vmdk_co_get_block_status(BlockDriverState *bs, 1284 int64_t sector_num, int nb_sectors, int *pnum, BlockDriverState **file) 1285 { 1286 BDRVVmdkState *s = bs->opaque; 1287 int64_t index_in_cluster, n, ret; 1288 uint64_t offset; 1289 VmdkExtent *extent; 1290 1291 extent = find_extent(s, sector_num, NULL); 1292 if (!extent) { 1293 return 0; 1294 } 1295 qemu_co_mutex_lock(&s->lock); 1296 ret = get_cluster_offset(bs, extent, NULL, 1297 sector_num * 512, false, &offset, 1298 0, 0); 1299 qemu_co_mutex_unlock(&s->lock); 1300 1301 index_in_cluster = vmdk_find_index_in_cluster(extent, sector_num); 1302 switch (ret) { 1303 case VMDK_ERROR: 1304 ret = -EIO; 1305 break; 1306 case VMDK_UNALLOC: 1307 ret = 0; 1308 break; 1309 case VMDK_ZEROED: 1310 ret = BDRV_BLOCK_ZERO; 1311 break; 1312 case VMDK_OK: 1313 ret = BDRV_BLOCK_DATA; 1314 if (!extent->compressed) { 1315 ret |= BDRV_BLOCK_OFFSET_VALID; 1316 ret |= (offset + (index_in_cluster << BDRV_SECTOR_BITS)) 1317 & BDRV_BLOCK_OFFSET_MASK; 1318 } 1319 *file = extent->file->bs; 1320 break; 1321 } 1322 1323 n = extent->cluster_sectors - index_in_cluster; 1324 if (n > nb_sectors) { 1325 n = nb_sectors; 1326 } 1327 *pnum = n; 1328 return ret; 1329 } 1330 1331 static int vmdk_write_extent(VmdkExtent *extent, int64_t cluster_offset, 1332 int64_t offset_in_cluster, QEMUIOVector *qiov, 1333 uint64_t qiov_offset, uint64_t n_bytes, 1334 uint64_t offset) 1335 { 1336 int ret; 1337 VmdkGrainMarker *data = NULL; 1338 uLongf buf_len; 1339 QEMUIOVector local_qiov; 1340 struct iovec iov; 1341 int64_t write_offset; 1342 int64_t write_end_sector; 1343 1344 if (extent->compressed) { 1345 void *compressed_data; 1346 1347 if (!extent->has_marker) { 1348 ret = -EINVAL; 1349 goto out; 1350 } 1351 buf_len = (extent->cluster_sectors << 9) * 2; 1352 data = g_malloc(buf_len + sizeof(VmdkGrainMarker)); 1353 1354 compressed_data = g_malloc(n_bytes); 1355 qemu_iovec_to_buf(qiov, qiov_offset, compressed_data, n_bytes); 1356 ret = compress(data->data, &buf_len, compressed_data, n_bytes); 1357 g_free(compressed_data); 1358 1359 if (ret != Z_OK || buf_len == 0) { 1360 ret = -EINVAL; 1361 goto out; 1362 } 1363 1364 data->lba = offset >> BDRV_SECTOR_BITS; 1365 data->size = buf_len; 1366 1367 n_bytes = buf_len + sizeof(VmdkGrainMarker); 1368 iov = (struct iovec) { 1369 .iov_base = data, 1370 .iov_len = n_bytes, 1371 }; 1372 qemu_iovec_init_external(&local_qiov, &iov, 1); 1373 } else { 1374 qemu_iovec_init(&local_qiov, qiov->niov); 1375 qemu_iovec_concat(&local_qiov, qiov, qiov_offset, n_bytes); 1376 } 1377 1378 write_offset = cluster_offset + offset_in_cluster, 1379 ret = bdrv_co_pwritev(extent->file, write_offset, n_bytes, 1380 &local_qiov, 0); 1381 1382 write_end_sector = DIV_ROUND_UP(write_offset + n_bytes, BDRV_SECTOR_SIZE); 1383 1384 if (extent->compressed) { 1385 extent->next_cluster_sector = write_end_sector; 1386 } else { 1387 extent->next_cluster_sector = MAX(extent->next_cluster_sector, 1388 write_end_sector); 1389 } 1390 1391 if (ret < 0) { 1392 goto out; 1393 } 1394 ret = 0; 1395 out: 1396 g_free(data); 1397 if (!extent->compressed) { 1398 qemu_iovec_destroy(&local_qiov); 1399 } 1400 return ret; 1401 } 1402 1403 static int vmdk_read_extent(VmdkExtent *extent, int64_t cluster_offset, 1404 int64_t offset_in_cluster, QEMUIOVector *qiov, 1405 int bytes) 1406 { 1407 int ret; 1408 int cluster_bytes, buf_bytes; 1409 uint8_t *cluster_buf, *compressed_data; 1410 uint8_t *uncomp_buf; 1411 uint32_t data_len; 1412 VmdkGrainMarker *marker; 1413 uLongf buf_len; 1414 1415 1416 if (!extent->compressed) { 1417 ret = bdrv_co_preadv(extent->file, 1418 cluster_offset + offset_in_cluster, bytes, 1419 qiov, 0); 1420 if (ret < 0) { 1421 return ret; 1422 } 1423 return 0; 1424 } 1425 cluster_bytes = extent->cluster_sectors * 512; 1426 /* Read two clusters in case GrainMarker + compressed data > one cluster */ 1427 buf_bytes = cluster_bytes * 2; 1428 cluster_buf = g_malloc(buf_bytes); 1429 uncomp_buf = g_malloc(cluster_bytes); 1430 ret = bdrv_pread(extent->file, 1431 cluster_offset, 1432 cluster_buf, buf_bytes); 1433 if (ret < 0) { 1434 goto out; 1435 } 1436 compressed_data = cluster_buf; 1437 buf_len = cluster_bytes; 1438 data_len = cluster_bytes; 1439 if (extent->has_marker) { 1440 marker = (VmdkGrainMarker *)cluster_buf; 1441 compressed_data = marker->data; 1442 data_len = le32_to_cpu(marker->size); 1443 } 1444 if (!data_len || data_len > buf_bytes) { 1445 ret = -EINVAL; 1446 goto out; 1447 } 1448 ret = uncompress(uncomp_buf, &buf_len, compressed_data, data_len); 1449 if (ret != Z_OK) { 1450 ret = -EINVAL; 1451 goto out; 1452 1453 } 1454 if (offset_in_cluster < 0 || 1455 offset_in_cluster + bytes > buf_len) { 1456 ret = -EINVAL; 1457 goto out; 1458 } 1459 qemu_iovec_from_buf(qiov, 0, uncomp_buf + offset_in_cluster, bytes); 1460 ret = 0; 1461 1462 out: 1463 g_free(uncomp_buf); 1464 g_free(cluster_buf); 1465 return ret; 1466 } 1467 1468 static int coroutine_fn 1469 vmdk_co_preadv(BlockDriverState *bs, uint64_t offset, uint64_t bytes, 1470 QEMUIOVector *qiov, int flags) 1471 { 1472 BDRVVmdkState *s = bs->opaque; 1473 int ret; 1474 uint64_t n_bytes, offset_in_cluster; 1475 VmdkExtent *extent = NULL; 1476 QEMUIOVector local_qiov; 1477 uint64_t cluster_offset; 1478 uint64_t bytes_done = 0; 1479 1480 qemu_iovec_init(&local_qiov, qiov->niov); 1481 qemu_co_mutex_lock(&s->lock); 1482 1483 while (bytes > 0) { 1484 extent = find_extent(s, offset >> BDRV_SECTOR_BITS, extent); 1485 if (!extent) { 1486 ret = -EIO; 1487 goto fail; 1488 } 1489 ret = get_cluster_offset(bs, extent, NULL, 1490 offset, false, &cluster_offset, 0, 0); 1491 offset_in_cluster = vmdk_find_offset_in_cluster(extent, offset); 1492 1493 n_bytes = MIN(bytes, extent->cluster_sectors * BDRV_SECTOR_SIZE 1494 - offset_in_cluster); 1495 1496 if (ret != VMDK_OK) { 1497 /* if not allocated, try to read from parent image, if exist */ 1498 if (bs->backing && ret != VMDK_ZEROED) { 1499 if (!vmdk_is_cid_valid(bs)) { 1500 ret = -EINVAL; 1501 goto fail; 1502 } 1503 1504 qemu_iovec_reset(&local_qiov); 1505 qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes); 1506 1507 ret = bdrv_co_preadv(bs->backing, offset, n_bytes, 1508 &local_qiov, 0); 1509 if (ret < 0) { 1510 goto fail; 1511 } 1512 } else { 1513 qemu_iovec_memset(qiov, bytes_done, 0, n_bytes); 1514 } 1515 } else { 1516 qemu_iovec_reset(&local_qiov); 1517 qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes); 1518 1519 ret = vmdk_read_extent(extent, cluster_offset, offset_in_cluster, 1520 &local_qiov, n_bytes); 1521 if (ret) { 1522 goto fail; 1523 } 1524 } 1525 bytes -= n_bytes; 1526 offset += n_bytes; 1527 bytes_done += n_bytes; 1528 } 1529 1530 ret = 0; 1531 fail: 1532 qemu_co_mutex_unlock(&s->lock); 1533 qemu_iovec_destroy(&local_qiov); 1534 1535 return ret; 1536 } 1537 1538 /** 1539 * vmdk_write: 1540 * @zeroed: buf is ignored (data is zero), use zeroed_grain GTE feature 1541 * if possible, otherwise return -ENOTSUP. 1542 * @zero_dry_run: used for zeroed == true only, don't update L2 table, just try 1543 * with each cluster. By dry run we can find if the zero write 1544 * is possible without modifying image data. 1545 * 1546 * Returns: error code with 0 for success. 1547 */ 1548 static int vmdk_pwritev(BlockDriverState *bs, uint64_t offset, 1549 uint64_t bytes, QEMUIOVector *qiov, 1550 bool zeroed, bool zero_dry_run) 1551 { 1552 BDRVVmdkState *s = bs->opaque; 1553 VmdkExtent *extent = NULL; 1554 int ret; 1555 int64_t offset_in_cluster, n_bytes; 1556 uint64_t cluster_offset; 1557 uint64_t bytes_done = 0; 1558 VmdkMetaData m_data; 1559 1560 if (DIV_ROUND_UP(offset, BDRV_SECTOR_SIZE) > bs->total_sectors) { 1561 error_report("Wrong offset: offset=0x%" PRIx64 1562 " total_sectors=0x%" PRIx64, 1563 offset, bs->total_sectors); 1564 return -EIO; 1565 } 1566 1567 while (bytes > 0) { 1568 extent = find_extent(s, offset >> BDRV_SECTOR_BITS, extent); 1569 if (!extent) { 1570 return -EIO; 1571 } 1572 offset_in_cluster = vmdk_find_offset_in_cluster(extent, offset); 1573 n_bytes = MIN(bytes, extent->cluster_sectors * BDRV_SECTOR_SIZE 1574 - offset_in_cluster); 1575 1576 ret = get_cluster_offset(bs, extent, &m_data, offset, 1577 !(extent->compressed || zeroed), 1578 &cluster_offset, offset_in_cluster, 1579 offset_in_cluster + n_bytes); 1580 if (extent->compressed) { 1581 if (ret == VMDK_OK) { 1582 /* Refuse write to allocated cluster for streamOptimized */ 1583 error_report("Could not write to allocated cluster" 1584 " for streamOptimized"); 1585 return -EIO; 1586 } else { 1587 /* allocate */ 1588 ret = get_cluster_offset(bs, extent, &m_data, offset, 1589 true, &cluster_offset, 0, 0); 1590 } 1591 } 1592 if (ret == VMDK_ERROR) { 1593 return -EINVAL; 1594 } 1595 if (zeroed) { 1596 /* Do zeroed write, buf is ignored */ 1597 if (extent->has_zero_grain && 1598 offset_in_cluster == 0 && 1599 n_bytes >= extent->cluster_sectors * BDRV_SECTOR_SIZE) { 1600 n_bytes = extent->cluster_sectors * BDRV_SECTOR_SIZE; 1601 if (!zero_dry_run) { 1602 /* update L2 tables */ 1603 if (vmdk_L2update(extent, &m_data, VMDK_GTE_ZEROED) 1604 != VMDK_OK) { 1605 return -EIO; 1606 } 1607 } 1608 } else { 1609 return -ENOTSUP; 1610 } 1611 } else { 1612 ret = vmdk_write_extent(extent, cluster_offset, offset_in_cluster, 1613 qiov, bytes_done, n_bytes, offset); 1614 if (ret) { 1615 return ret; 1616 } 1617 if (m_data.valid) { 1618 /* update L2 tables */ 1619 if (vmdk_L2update(extent, &m_data, 1620 cluster_offset >> BDRV_SECTOR_BITS) 1621 != VMDK_OK) { 1622 return -EIO; 1623 } 1624 } 1625 } 1626 bytes -= n_bytes; 1627 offset += n_bytes; 1628 bytes_done += n_bytes; 1629 1630 /* update CID on the first write every time the virtual disk is 1631 * opened */ 1632 if (!s->cid_updated) { 1633 ret = vmdk_write_cid(bs, g_random_int()); 1634 if (ret < 0) { 1635 return ret; 1636 } 1637 s->cid_updated = true; 1638 } 1639 } 1640 return 0; 1641 } 1642 1643 static int coroutine_fn 1644 vmdk_co_pwritev(BlockDriverState *bs, uint64_t offset, uint64_t bytes, 1645 QEMUIOVector *qiov, int flags) 1646 { 1647 int ret; 1648 BDRVVmdkState *s = bs->opaque; 1649 qemu_co_mutex_lock(&s->lock); 1650 ret = vmdk_pwritev(bs, offset, bytes, qiov, false, false); 1651 qemu_co_mutex_unlock(&s->lock); 1652 return ret; 1653 } 1654 1655 static int coroutine_fn 1656 vmdk_co_pwritev_compressed(BlockDriverState *bs, uint64_t offset, 1657 uint64_t bytes, QEMUIOVector *qiov) 1658 { 1659 return vmdk_co_pwritev(bs, offset, bytes, qiov, 0); 1660 } 1661 1662 static int coroutine_fn vmdk_co_pwrite_zeroes(BlockDriverState *bs, 1663 int64_t offset, 1664 int bytes, 1665 BdrvRequestFlags flags) 1666 { 1667 int ret; 1668 BDRVVmdkState *s = bs->opaque; 1669 1670 qemu_co_mutex_lock(&s->lock); 1671 /* write zeroes could fail if sectors not aligned to cluster, test it with 1672 * dry_run == true before really updating image */ 1673 ret = vmdk_pwritev(bs, offset, bytes, NULL, true, true); 1674 if (!ret) { 1675 ret = vmdk_pwritev(bs, offset, bytes, NULL, true, false); 1676 } 1677 qemu_co_mutex_unlock(&s->lock); 1678 return ret; 1679 } 1680 1681 static int vmdk_create_extent(const char *filename, int64_t filesize, 1682 bool flat, bool compress, bool zeroed_grain, 1683 QemuOpts *opts, Error **errp) 1684 { 1685 int ret, i; 1686 BlockBackend *blk = NULL; 1687 VMDK4Header header; 1688 Error *local_err = NULL; 1689 uint32_t tmp, magic, grains, gd_sectors, gt_size, gt_count; 1690 uint32_t *gd_buf = NULL; 1691 int gd_buf_size; 1692 1693 ret = bdrv_create_file(filename, opts, &local_err); 1694 if (ret < 0) { 1695 error_propagate(errp, local_err); 1696 goto exit; 1697 } 1698 1699 blk = blk_new_open(filename, NULL, NULL, 1700 BDRV_O_RDWR | BDRV_O_PROTOCOL, &local_err); 1701 if (blk == NULL) { 1702 error_propagate(errp, local_err); 1703 ret = -EIO; 1704 goto exit; 1705 } 1706 1707 blk_set_allow_write_beyond_eof(blk, true); 1708 1709 if (flat) { 1710 ret = blk_truncate(blk, filesize); 1711 if (ret < 0) { 1712 error_setg_errno(errp, -ret, "Could not truncate file"); 1713 } 1714 goto exit; 1715 } 1716 magic = cpu_to_be32(VMDK4_MAGIC); 1717 memset(&header, 0, sizeof(header)); 1718 if (compress) { 1719 header.version = 3; 1720 } else if (zeroed_grain) { 1721 header.version = 2; 1722 } else { 1723 header.version = 1; 1724 } 1725 header.flags = VMDK4_FLAG_RGD | VMDK4_FLAG_NL_DETECT 1726 | (compress ? VMDK4_FLAG_COMPRESS | VMDK4_FLAG_MARKER : 0) 1727 | (zeroed_grain ? VMDK4_FLAG_ZERO_GRAIN : 0); 1728 header.compressAlgorithm = compress ? VMDK4_COMPRESSION_DEFLATE : 0; 1729 header.capacity = filesize / BDRV_SECTOR_SIZE; 1730 header.granularity = 128; 1731 header.num_gtes_per_gt = BDRV_SECTOR_SIZE; 1732 1733 grains = DIV_ROUND_UP(filesize / BDRV_SECTOR_SIZE, header.granularity); 1734 gt_size = DIV_ROUND_UP(header.num_gtes_per_gt * sizeof(uint32_t), 1735 BDRV_SECTOR_SIZE); 1736 gt_count = DIV_ROUND_UP(grains, header.num_gtes_per_gt); 1737 gd_sectors = DIV_ROUND_UP(gt_count * sizeof(uint32_t), BDRV_SECTOR_SIZE); 1738 1739 header.desc_offset = 1; 1740 header.desc_size = 20; 1741 header.rgd_offset = header.desc_offset + header.desc_size; 1742 header.gd_offset = header.rgd_offset + gd_sectors + (gt_size * gt_count); 1743 header.grain_offset = 1744 ROUND_UP(header.gd_offset + gd_sectors + (gt_size * gt_count), 1745 header.granularity); 1746 /* swap endianness for all header fields */ 1747 header.version = cpu_to_le32(header.version); 1748 header.flags = cpu_to_le32(header.flags); 1749 header.capacity = cpu_to_le64(header.capacity); 1750 header.granularity = cpu_to_le64(header.granularity); 1751 header.num_gtes_per_gt = cpu_to_le32(header.num_gtes_per_gt); 1752 header.desc_offset = cpu_to_le64(header.desc_offset); 1753 header.desc_size = cpu_to_le64(header.desc_size); 1754 header.rgd_offset = cpu_to_le64(header.rgd_offset); 1755 header.gd_offset = cpu_to_le64(header.gd_offset); 1756 header.grain_offset = cpu_to_le64(header.grain_offset); 1757 header.compressAlgorithm = cpu_to_le16(header.compressAlgorithm); 1758 1759 header.check_bytes[0] = 0xa; 1760 header.check_bytes[1] = 0x20; 1761 header.check_bytes[2] = 0xd; 1762 header.check_bytes[3] = 0xa; 1763 1764 /* write all the data */ 1765 ret = blk_pwrite(blk, 0, &magic, sizeof(magic), 0); 1766 if (ret < 0) { 1767 error_setg(errp, QERR_IO_ERROR); 1768 goto exit; 1769 } 1770 ret = blk_pwrite(blk, sizeof(magic), &header, sizeof(header), 0); 1771 if (ret < 0) { 1772 error_setg(errp, QERR_IO_ERROR); 1773 goto exit; 1774 } 1775 1776 ret = blk_truncate(blk, le64_to_cpu(header.grain_offset) << 9); 1777 if (ret < 0) { 1778 error_setg_errno(errp, -ret, "Could not truncate file"); 1779 goto exit; 1780 } 1781 1782 /* write grain directory */ 1783 gd_buf_size = gd_sectors * BDRV_SECTOR_SIZE; 1784 gd_buf = g_malloc0(gd_buf_size); 1785 for (i = 0, tmp = le64_to_cpu(header.rgd_offset) + gd_sectors; 1786 i < gt_count; i++, tmp += gt_size) { 1787 gd_buf[i] = cpu_to_le32(tmp); 1788 } 1789 ret = blk_pwrite(blk, le64_to_cpu(header.rgd_offset) * BDRV_SECTOR_SIZE, 1790 gd_buf, gd_buf_size, 0); 1791 if (ret < 0) { 1792 error_setg(errp, QERR_IO_ERROR); 1793 goto exit; 1794 } 1795 1796 /* write backup grain directory */ 1797 for (i = 0, tmp = le64_to_cpu(header.gd_offset) + gd_sectors; 1798 i < gt_count; i++, tmp += gt_size) { 1799 gd_buf[i] = cpu_to_le32(tmp); 1800 } 1801 ret = blk_pwrite(blk, le64_to_cpu(header.gd_offset) * BDRV_SECTOR_SIZE, 1802 gd_buf, gd_buf_size, 0); 1803 if (ret < 0) { 1804 error_setg(errp, QERR_IO_ERROR); 1805 goto exit; 1806 } 1807 1808 ret = 0; 1809 exit: 1810 if (blk) { 1811 blk_unref(blk); 1812 } 1813 g_free(gd_buf); 1814 return ret; 1815 } 1816 1817 static int filename_decompose(const char *filename, char *path, char *prefix, 1818 char *postfix, size_t buf_len, Error **errp) 1819 { 1820 const char *p, *q; 1821 1822 if (filename == NULL || !strlen(filename)) { 1823 error_setg(errp, "No filename provided"); 1824 return VMDK_ERROR; 1825 } 1826 p = strrchr(filename, '/'); 1827 if (p == NULL) { 1828 p = strrchr(filename, '\\'); 1829 } 1830 if (p == NULL) { 1831 p = strrchr(filename, ':'); 1832 } 1833 if (p != NULL) { 1834 p++; 1835 if (p - filename >= buf_len) { 1836 return VMDK_ERROR; 1837 } 1838 pstrcpy(path, p - filename + 1, filename); 1839 } else { 1840 p = filename; 1841 path[0] = '\0'; 1842 } 1843 q = strrchr(p, '.'); 1844 if (q == NULL) { 1845 pstrcpy(prefix, buf_len, p); 1846 postfix[0] = '\0'; 1847 } else { 1848 if (q - p >= buf_len) { 1849 return VMDK_ERROR; 1850 } 1851 pstrcpy(prefix, q - p + 1, p); 1852 pstrcpy(postfix, buf_len, q); 1853 } 1854 return VMDK_OK; 1855 } 1856 1857 static int vmdk_create(const char *filename, QemuOpts *opts, Error **errp) 1858 { 1859 int idx = 0; 1860 BlockBackend *new_blk = NULL; 1861 Error *local_err = NULL; 1862 char *desc = NULL; 1863 int64_t total_size = 0, filesize; 1864 char *adapter_type = NULL; 1865 char *backing_file = NULL; 1866 char *hw_version = NULL; 1867 char *fmt = NULL; 1868 int ret = 0; 1869 bool flat, split, compress; 1870 GString *ext_desc_lines; 1871 char *path = g_malloc0(PATH_MAX); 1872 char *prefix = g_malloc0(PATH_MAX); 1873 char *postfix = g_malloc0(PATH_MAX); 1874 char *desc_line = g_malloc0(BUF_SIZE); 1875 char *ext_filename = g_malloc0(PATH_MAX); 1876 char *desc_filename = g_malloc0(PATH_MAX); 1877 const int64_t split_size = 0x80000000; /* VMDK has constant split size */ 1878 const char *desc_extent_line; 1879 char *parent_desc_line = g_malloc0(BUF_SIZE); 1880 uint32_t parent_cid = 0xffffffff; 1881 uint32_t number_heads = 16; 1882 bool zeroed_grain = false; 1883 uint32_t desc_offset = 0, desc_len; 1884 const char desc_template[] = 1885 "# Disk DescriptorFile\n" 1886 "version=1\n" 1887 "CID=%" PRIx32 "\n" 1888 "parentCID=%" PRIx32 "\n" 1889 "createType=\"%s\"\n" 1890 "%s" 1891 "\n" 1892 "# Extent description\n" 1893 "%s" 1894 "\n" 1895 "# The Disk Data Base\n" 1896 "#DDB\n" 1897 "\n" 1898 "ddb.virtualHWVersion = \"%s\"\n" 1899 "ddb.geometry.cylinders = \"%" PRId64 "\"\n" 1900 "ddb.geometry.heads = \"%" PRIu32 "\"\n" 1901 "ddb.geometry.sectors = \"63\"\n" 1902 "ddb.adapterType = \"%s\"\n"; 1903 1904 ext_desc_lines = g_string_new(NULL); 1905 1906 if (filename_decompose(filename, path, prefix, postfix, PATH_MAX, errp)) { 1907 ret = -EINVAL; 1908 goto exit; 1909 } 1910 /* Read out options */ 1911 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0), 1912 BDRV_SECTOR_SIZE); 1913 adapter_type = qemu_opt_get_del(opts, BLOCK_OPT_ADAPTER_TYPE); 1914 backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE); 1915 hw_version = qemu_opt_get_del(opts, BLOCK_OPT_HWVERSION); 1916 if (qemu_opt_get_bool_del(opts, BLOCK_OPT_COMPAT6, false)) { 1917 if (strcmp(hw_version, "undefined")) { 1918 error_setg(errp, 1919 "compat6 cannot be enabled with hwversion set"); 1920 ret = -EINVAL; 1921 goto exit; 1922 } 1923 g_free(hw_version); 1924 hw_version = g_strdup("6"); 1925 } 1926 if (strcmp(hw_version, "undefined") == 0) { 1927 g_free(hw_version); 1928 hw_version = g_strdup("4"); 1929 } 1930 fmt = qemu_opt_get_del(opts, BLOCK_OPT_SUBFMT); 1931 if (qemu_opt_get_bool_del(opts, BLOCK_OPT_ZEROED_GRAIN, false)) { 1932 zeroed_grain = true; 1933 } 1934 1935 if (!adapter_type) { 1936 adapter_type = g_strdup("ide"); 1937 } else if (strcmp(adapter_type, "ide") && 1938 strcmp(adapter_type, "buslogic") && 1939 strcmp(adapter_type, "lsilogic") && 1940 strcmp(adapter_type, "legacyESX")) { 1941 error_setg(errp, "Unknown adapter type: '%s'", adapter_type); 1942 ret = -EINVAL; 1943 goto exit; 1944 } 1945 if (strcmp(adapter_type, "ide") != 0) { 1946 /* that's the number of heads with which vmware operates when 1947 creating, exporting, etc. vmdk files with a non-ide adapter type */ 1948 number_heads = 255; 1949 } 1950 if (!fmt) { 1951 /* Default format to monolithicSparse */ 1952 fmt = g_strdup("monolithicSparse"); 1953 } else if (strcmp(fmt, "monolithicFlat") && 1954 strcmp(fmt, "monolithicSparse") && 1955 strcmp(fmt, "twoGbMaxExtentSparse") && 1956 strcmp(fmt, "twoGbMaxExtentFlat") && 1957 strcmp(fmt, "streamOptimized")) { 1958 error_setg(errp, "Unknown subformat: '%s'", fmt); 1959 ret = -EINVAL; 1960 goto exit; 1961 } 1962 split = !(strcmp(fmt, "twoGbMaxExtentFlat") && 1963 strcmp(fmt, "twoGbMaxExtentSparse")); 1964 flat = !(strcmp(fmt, "monolithicFlat") && 1965 strcmp(fmt, "twoGbMaxExtentFlat")); 1966 compress = !strcmp(fmt, "streamOptimized"); 1967 if (flat) { 1968 desc_extent_line = "RW %" PRId64 " FLAT \"%s\" 0\n"; 1969 } else { 1970 desc_extent_line = "RW %" PRId64 " SPARSE \"%s\"\n"; 1971 } 1972 if (flat && backing_file) { 1973 error_setg(errp, "Flat image can't have backing file"); 1974 ret = -ENOTSUP; 1975 goto exit; 1976 } 1977 if (flat && zeroed_grain) { 1978 error_setg(errp, "Flat image can't enable zeroed grain"); 1979 ret = -ENOTSUP; 1980 goto exit; 1981 } 1982 if (backing_file) { 1983 BlockBackend *blk; 1984 char *full_backing = g_new0(char, PATH_MAX); 1985 bdrv_get_full_backing_filename_from_filename(filename, backing_file, 1986 full_backing, PATH_MAX, 1987 &local_err); 1988 if (local_err) { 1989 g_free(full_backing); 1990 error_propagate(errp, local_err); 1991 ret = -ENOENT; 1992 goto exit; 1993 } 1994 1995 blk = blk_new_open(full_backing, NULL, NULL, 1996 BDRV_O_NO_BACKING, errp); 1997 g_free(full_backing); 1998 if (blk == NULL) { 1999 ret = -EIO; 2000 goto exit; 2001 } 2002 if (strcmp(blk_bs(blk)->drv->format_name, "vmdk")) { 2003 blk_unref(blk); 2004 ret = -EINVAL; 2005 goto exit; 2006 } 2007 parent_cid = vmdk_read_cid(blk_bs(blk), 0); 2008 blk_unref(blk); 2009 snprintf(parent_desc_line, BUF_SIZE, 2010 "parentFileNameHint=\"%s\"", backing_file); 2011 } 2012 2013 /* Create extents */ 2014 filesize = total_size; 2015 while (filesize > 0) { 2016 int64_t size = filesize; 2017 2018 if (split && size > split_size) { 2019 size = split_size; 2020 } 2021 if (split) { 2022 snprintf(desc_filename, PATH_MAX, "%s-%c%03d%s", 2023 prefix, flat ? 'f' : 's', ++idx, postfix); 2024 } else if (flat) { 2025 snprintf(desc_filename, PATH_MAX, "%s-flat%s", prefix, postfix); 2026 } else { 2027 snprintf(desc_filename, PATH_MAX, "%s%s", prefix, postfix); 2028 } 2029 snprintf(ext_filename, PATH_MAX, "%s%s", path, desc_filename); 2030 2031 if (vmdk_create_extent(ext_filename, size, 2032 flat, compress, zeroed_grain, opts, errp)) { 2033 ret = -EINVAL; 2034 goto exit; 2035 } 2036 filesize -= size; 2037 2038 /* Format description line */ 2039 snprintf(desc_line, BUF_SIZE, 2040 desc_extent_line, size / BDRV_SECTOR_SIZE, desc_filename); 2041 g_string_append(ext_desc_lines, desc_line); 2042 } 2043 /* generate descriptor file */ 2044 desc = g_strdup_printf(desc_template, 2045 g_random_int(), 2046 parent_cid, 2047 fmt, 2048 parent_desc_line, 2049 ext_desc_lines->str, 2050 hw_version, 2051 total_size / 2052 (int64_t)(63 * number_heads * BDRV_SECTOR_SIZE), 2053 number_heads, 2054 adapter_type); 2055 desc_len = strlen(desc); 2056 /* the descriptor offset = 0x200 */ 2057 if (!split && !flat) { 2058 desc_offset = 0x200; 2059 } else { 2060 ret = bdrv_create_file(filename, opts, &local_err); 2061 if (ret < 0) { 2062 error_propagate(errp, local_err); 2063 goto exit; 2064 } 2065 } 2066 2067 new_blk = blk_new_open(filename, NULL, NULL, 2068 BDRV_O_RDWR | BDRV_O_PROTOCOL, &local_err); 2069 if (new_blk == NULL) { 2070 error_propagate(errp, local_err); 2071 ret = -EIO; 2072 goto exit; 2073 } 2074 2075 blk_set_allow_write_beyond_eof(new_blk, true); 2076 2077 ret = blk_pwrite(new_blk, desc_offset, desc, desc_len, 0); 2078 if (ret < 0) { 2079 error_setg_errno(errp, -ret, "Could not write description"); 2080 goto exit; 2081 } 2082 /* bdrv_pwrite write padding zeros to align to sector, we don't need that 2083 * for description file */ 2084 if (desc_offset == 0) { 2085 ret = blk_truncate(new_blk, desc_len); 2086 if (ret < 0) { 2087 error_setg_errno(errp, -ret, "Could not truncate file"); 2088 } 2089 } 2090 exit: 2091 if (new_blk) { 2092 blk_unref(new_blk); 2093 } 2094 g_free(adapter_type); 2095 g_free(backing_file); 2096 g_free(hw_version); 2097 g_free(fmt); 2098 g_free(desc); 2099 g_free(path); 2100 g_free(prefix); 2101 g_free(postfix); 2102 g_free(desc_line); 2103 g_free(ext_filename); 2104 g_free(desc_filename); 2105 g_free(parent_desc_line); 2106 g_string_free(ext_desc_lines, true); 2107 return ret; 2108 } 2109 2110 static void vmdk_close(BlockDriverState *bs) 2111 { 2112 BDRVVmdkState *s = bs->opaque; 2113 2114 vmdk_free_extents(bs); 2115 g_free(s->create_type); 2116 2117 migrate_del_blocker(s->migration_blocker); 2118 error_free(s->migration_blocker); 2119 } 2120 2121 static coroutine_fn int vmdk_co_flush(BlockDriverState *bs) 2122 { 2123 BDRVVmdkState *s = bs->opaque; 2124 int i, err; 2125 int ret = 0; 2126 2127 for (i = 0; i < s->num_extents; i++) { 2128 err = bdrv_co_flush(s->extents[i].file->bs); 2129 if (err < 0) { 2130 ret = err; 2131 } 2132 } 2133 return ret; 2134 } 2135 2136 static int64_t vmdk_get_allocated_file_size(BlockDriverState *bs) 2137 { 2138 int i; 2139 int64_t ret = 0; 2140 int64_t r; 2141 BDRVVmdkState *s = bs->opaque; 2142 2143 ret = bdrv_get_allocated_file_size(bs->file->bs); 2144 if (ret < 0) { 2145 return ret; 2146 } 2147 for (i = 0; i < s->num_extents; i++) { 2148 if (s->extents[i].file == bs->file) { 2149 continue; 2150 } 2151 r = bdrv_get_allocated_file_size(s->extents[i].file->bs); 2152 if (r < 0) { 2153 return r; 2154 } 2155 ret += r; 2156 } 2157 return ret; 2158 } 2159 2160 static int vmdk_has_zero_init(BlockDriverState *bs) 2161 { 2162 int i; 2163 BDRVVmdkState *s = bs->opaque; 2164 2165 /* If has a flat extent and its underlying storage doesn't have zero init, 2166 * return 0. */ 2167 for (i = 0; i < s->num_extents; i++) { 2168 if (s->extents[i].flat) { 2169 if (!bdrv_has_zero_init(s->extents[i].file->bs)) { 2170 return 0; 2171 } 2172 } 2173 } 2174 return 1; 2175 } 2176 2177 static ImageInfo *vmdk_get_extent_info(VmdkExtent *extent) 2178 { 2179 ImageInfo *info = g_new0(ImageInfo, 1); 2180 2181 *info = (ImageInfo){ 2182 .filename = g_strdup(extent->file->bs->filename), 2183 .format = g_strdup(extent->type), 2184 .virtual_size = extent->sectors * BDRV_SECTOR_SIZE, 2185 .compressed = extent->compressed, 2186 .has_compressed = extent->compressed, 2187 .cluster_size = extent->cluster_sectors * BDRV_SECTOR_SIZE, 2188 .has_cluster_size = !extent->flat, 2189 }; 2190 2191 return info; 2192 } 2193 2194 static int vmdk_check(BlockDriverState *bs, BdrvCheckResult *result, 2195 BdrvCheckMode fix) 2196 { 2197 BDRVVmdkState *s = bs->opaque; 2198 VmdkExtent *extent = NULL; 2199 int64_t sector_num = 0; 2200 int64_t total_sectors = bdrv_nb_sectors(bs); 2201 int ret; 2202 uint64_t cluster_offset; 2203 2204 if (fix) { 2205 return -ENOTSUP; 2206 } 2207 2208 for (;;) { 2209 if (sector_num >= total_sectors) { 2210 return 0; 2211 } 2212 extent = find_extent(s, sector_num, extent); 2213 if (!extent) { 2214 fprintf(stderr, 2215 "ERROR: could not find extent for sector %" PRId64 "\n", 2216 sector_num); 2217 break; 2218 } 2219 ret = get_cluster_offset(bs, extent, NULL, 2220 sector_num << BDRV_SECTOR_BITS, 2221 false, &cluster_offset, 0, 0); 2222 if (ret == VMDK_ERROR) { 2223 fprintf(stderr, 2224 "ERROR: could not get cluster_offset for sector %" 2225 PRId64 "\n", sector_num); 2226 break; 2227 } 2228 if (ret == VMDK_OK && 2229 cluster_offset >= bdrv_getlength(extent->file->bs)) 2230 { 2231 fprintf(stderr, 2232 "ERROR: cluster offset for sector %" 2233 PRId64 " points after EOF\n", sector_num); 2234 break; 2235 } 2236 sector_num += extent->cluster_sectors; 2237 } 2238 2239 result->corruptions++; 2240 return 0; 2241 } 2242 2243 static ImageInfoSpecific *vmdk_get_specific_info(BlockDriverState *bs) 2244 { 2245 int i; 2246 BDRVVmdkState *s = bs->opaque; 2247 ImageInfoSpecific *spec_info = g_new0(ImageInfoSpecific, 1); 2248 ImageInfoList **next; 2249 2250 *spec_info = (ImageInfoSpecific){ 2251 .type = IMAGE_INFO_SPECIFIC_KIND_VMDK, 2252 .u = { 2253 .vmdk.data = g_new0(ImageInfoSpecificVmdk, 1), 2254 }, 2255 }; 2256 2257 *spec_info->u.vmdk.data = (ImageInfoSpecificVmdk) { 2258 .create_type = g_strdup(s->create_type), 2259 .cid = s->cid, 2260 .parent_cid = s->parent_cid, 2261 }; 2262 2263 next = &spec_info->u.vmdk.data->extents; 2264 for (i = 0; i < s->num_extents; i++) { 2265 *next = g_new0(ImageInfoList, 1); 2266 (*next)->value = vmdk_get_extent_info(&s->extents[i]); 2267 (*next)->next = NULL; 2268 next = &(*next)->next; 2269 } 2270 2271 return spec_info; 2272 } 2273 2274 static bool vmdk_extents_type_eq(const VmdkExtent *a, const VmdkExtent *b) 2275 { 2276 return a->flat == b->flat && 2277 a->compressed == b->compressed && 2278 (a->flat || a->cluster_sectors == b->cluster_sectors); 2279 } 2280 2281 static int vmdk_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) 2282 { 2283 int i; 2284 BDRVVmdkState *s = bs->opaque; 2285 assert(s->num_extents); 2286 2287 /* See if we have multiple extents but they have different cases */ 2288 for (i = 1; i < s->num_extents; i++) { 2289 if (!vmdk_extents_type_eq(&s->extents[0], &s->extents[i])) { 2290 return -ENOTSUP; 2291 } 2292 } 2293 bdi->needs_compressed_writes = s->extents[0].compressed; 2294 if (!s->extents[0].flat) { 2295 bdi->cluster_size = s->extents[0].cluster_sectors << BDRV_SECTOR_BITS; 2296 } 2297 return 0; 2298 } 2299 2300 static QemuOptsList vmdk_create_opts = { 2301 .name = "vmdk-create-opts", 2302 .head = QTAILQ_HEAD_INITIALIZER(vmdk_create_opts.head), 2303 .desc = { 2304 { 2305 .name = BLOCK_OPT_SIZE, 2306 .type = QEMU_OPT_SIZE, 2307 .help = "Virtual disk size" 2308 }, 2309 { 2310 .name = BLOCK_OPT_ADAPTER_TYPE, 2311 .type = QEMU_OPT_STRING, 2312 .help = "Virtual adapter type, can be one of " 2313 "ide (default), lsilogic, buslogic or legacyESX" 2314 }, 2315 { 2316 .name = BLOCK_OPT_BACKING_FILE, 2317 .type = QEMU_OPT_STRING, 2318 .help = "File name of a base image" 2319 }, 2320 { 2321 .name = BLOCK_OPT_COMPAT6, 2322 .type = QEMU_OPT_BOOL, 2323 .help = "VMDK version 6 image", 2324 .def_value_str = "off" 2325 }, 2326 { 2327 .name = BLOCK_OPT_HWVERSION, 2328 .type = QEMU_OPT_STRING, 2329 .help = "VMDK hardware version", 2330 .def_value_str = "undefined" 2331 }, 2332 { 2333 .name = BLOCK_OPT_SUBFMT, 2334 .type = QEMU_OPT_STRING, 2335 .help = 2336 "VMDK flat extent format, can be one of " 2337 "{monolithicSparse (default) | monolithicFlat | twoGbMaxExtentSparse | twoGbMaxExtentFlat | streamOptimized} " 2338 }, 2339 { 2340 .name = BLOCK_OPT_ZEROED_GRAIN, 2341 .type = QEMU_OPT_BOOL, 2342 .help = "Enable efficient zero writes " 2343 "using the zeroed-grain GTE feature" 2344 }, 2345 { /* end of list */ } 2346 } 2347 }; 2348 2349 static BlockDriver bdrv_vmdk = { 2350 .format_name = "vmdk", 2351 .instance_size = sizeof(BDRVVmdkState), 2352 .bdrv_probe = vmdk_probe, 2353 .bdrv_open = vmdk_open, 2354 .bdrv_check = vmdk_check, 2355 .bdrv_reopen_prepare = vmdk_reopen_prepare, 2356 .bdrv_co_preadv = vmdk_co_preadv, 2357 .bdrv_co_pwritev = vmdk_co_pwritev, 2358 .bdrv_co_pwritev_compressed = vmdk_co_pwritev_compressed, 2359 .bdrv_co_pwrite_zeroes = vmdk_co_pwrite_zeroes, 2360 .bdrv_close = vmdk_close, 2361 .bdrv_create = vmdk_create, 2362 .bdrv_co_flush_to_disk = vmdk_co_flush, 2363 .bdrv_co_get_block_status = vmdk_co_get_block_status, 2364 .bdrv_get_allocated_file_size = vmdk_get_allocated_file_size, 2365 .bdrv_has_zero_init = vmdk_has_zero_init, 2366 .bdrv_get_specific_info = vmdk_get_specific_info, 2367 .bdrv_refresh_limits = vmdk_refresh_limits, 2368 .bdrv_get_info = vmdk_get_info, 2369 2370 .supports_backing = true, 2371 .create_opts = &vmdk_create_opts, 2372 }; 2373 2374 static void bdrv_vmdk_init(void) 2375 { 2376 bdrv_register(&bdrv_vmdk); 2377 } 2378 2379 block_init(bdrv_vmdk_init); 2380