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 "sysemu/block-backend.h" 30 #include "qapi/qmp/qerror.h" 31 #include "qemu/error-report.h" 32 #include "qemu/module.h" 33 #include "migration/migration.h" 34 #include <zlib.h> 35 #include <glib.h> 36 37 #define VMDK3_MAGIC (('C' << 24) | ('O' << 16) | ('W' << 8) | 'D') 38 #define VMDK4_MAGIC (('K' << 24) | ('D' << 16) | ('M' << 8) | 'V') 39 #define VMDK4_COMPRESSION_DEFLATE 1 40 #define VMDK4_FLAG_NL_DETECT (1 << 0) 41 #define VMDK4_FLAG_RGD (1 << 1) 42 /* Zeroed-grain enable bit */ 43 #define VMDK4_FLAG_ZERO_GRAIN (1 << 2) 44 #define VMDK4_FLAG_COMPRESS (1 << 16) 45 #define VMDK4_FLAG_MARKER (1 << 17) 46 #define VMDK4_GD_AT_END 0xffffffffffffffffULL 47 48 #define VMDK_GTE_ZEROED 0x1 49 50 /* VMDK internal error codes */ 51 #define VMDK_OK 0 52 #define VMDK_ERROR (-1) 53 /* Cluster not allocated */ 54 #define VMDK_UNALLOC (-2) 55 #define VMDK_ZEROED (-3) 56 57 #define BLOCK_OPT_ZEROED_GRAIN "zeroed_grain" 58 59 typedef struct { 60 uint32_t version; 61 uint32_t flags; 62 uint32_t disk_sectors; 63 uint32_t granularity; 64 uint32_t l1dir_offset; 65 uint32_t l1dir_size; 66 uint32_t file_sectors; 67 uint32_t cylinders; 68 uint32_t heads; 69 uint32_t sectors_per_track; 70 } QEMU_PACKED VMDK3Header; 71 72 typedef struct { 73 uint32_t version; 74 uint32_t flags; 75 uint64_t capacity; 76 uint64_t granularity; 77 uint64_t desc_offset; 78 uint64_t desc_size; 79 /* Number of GrainTableEntries per GrainTable */ 80 uint32_t num_gtes_per_gt; 81 uint64_t rgd_offset; 82 uint64_t gd_offset; 83 uint64_t grain_offset; 84 char filler[1]; 85 char check_bytes[4]; 86 uint16_t compressAlgorithm; 87 } QEMU_PACKED VMDK4Header; 88 89 #define L2_CACHE_SIZE 16 90 91 typedef struct VmdkExtent { 92 BdrvChild *file; 93 bool flat; 94 bool compressed; 95 bool has_marker; 96 bool has_zero_grain; 97 int version; 98 int64_t sectors; 99 int64_t end_sector; 100 int64_t flat_start_offset; 101 int64_t l1_table_offset; 102 int64_t l1_backup_table_offset; 103 uint32_t *l1_table; 104 uint32_t *l1_backup_table; 105 unsigned int l1_size; 106 uint32_t l1_entry_sectors; 107 108 unsigned int l2_size; 109 uint32_t *l2_cache; 110 uint32_t l2_cache_offsets[L2_CACHE_SIZE]; 111 uint32_t l2_cache_counts[L2_CACHE_SIZE]; 112 113 int64_t cluster_sectors; 114 int64_t next_cluster_sector; 115 char *type; 116 } VmdkExtent; 117 118 typedef struct BDRVVmdkState { 119 CoMutex lock; 120 uint64_t desc_offset; 121 bool cid_updated; 122 bool cid_checked; 123 uint32_t cid; 124 uint32_t parent_cid; 125 int num_extents; 126 /* Extent array with num_extents entries, ascend ordered by address */ 127 VmdkExtent *extents; 128 Error *migration_blocker; 129 char *create_type; 130 } BDRVVmdkState; 131 132 typedef struct VmdkMetaData { 133 unsigned int l1_index; 134 unsigned int l2_index; 135 unsigned int l2_offset; 136 int valid; 137 uint32_t *l2_cache_entry; 138 } VmdkMetaData; 139 140 typedef struct VmdkGrainMarker { 141 uint64_t lba; 142 uint32_t size; 143 uint8_t data[0]; 144 } QEMU_PACKED VmdkGrainMarker; 145 146 enum { 147 MARKER_END_OF_STREAM = 0, 148 MARKER_GRAIN_TABLE = 1, 149 MARKER_GRAIN_DIRECTORY = 2, 150 MARKER_FOOTER = 3, 151 }; 152 153 static int vmdk_probe(const uint8_t *buf, int buf_size, const char *filename) 154 { 155 uint32_t magic; 156 157 if (buf_size < 4) { 158 return 0; 159 } 160 magic = be32_to_cpu(*(uint32_t *)buf); 161 if (magic == VMDK3_MAGIC || 162 magic == VMDK4_MAGIC) { 163 return 100; 164 } else { 165 const char *p = (const char *)buf; 166 const char *end = p + buf_size; 167 while (p < end) { 168 if (*p == '#') { 169 /* skip comment line */ 170 while (p < end && *p != '\n') { 171 p++; 172 } 173 p++; 174 continue; 175 } 176 if (*p == ' ') { 177 while (p < end && *p == ' ') { 178 p++; 179 } 180 /* skip '\r' if windows line endings used. */ 181 if (p < end && *p == '\r') { 182 p++; 183 } 184 /* only accept blank lines before 'version=' line */ 185 if (p == end || *p != '\n') { 186 return 0; 187 } 188 p++; 189 continue; 190 } 191 if (end - p >= strlen("version=X\n")) { 192 if (strncmp("version=1\n", p, strlen("version=1\n")) == 0 || 193 strncmp("version=2\n", p, strlen("version=2\n")) == 0) { 194 return 100; 195 } 196 } 197 if (end - p >= strlen("version=X\r\n")) { 198 if (strncmp("version=1\r\n", p, strlen("version=1\r\n")) == 0 || 199 strncmp("version=2\r\n", p, strlen("version=2\r\n")) == 0) { 200 return 100; 201 } 202 } 203 return 0; 204 } 205 return 0; 206 } 207 } 208 209 #define SECTOR_SIZE 512 210 #define DESC_SIZE (20 * SECTOR_SIZE) /* 20 sectors of 512 bytes each */ 211 #define BUF_SIZE 4096 212 #define HEADER_SIZE 512 /* first sector of 512 bytes */ 213 214 static void vmdk_free_extents(BlockDriverState *bs) 215 { 216 int i; 217 BDRVVmdkState *s = bs->opaque; 218 VmdkExtent *e; 219 220 for (i = 0; i < s->num_extents; i++) { 221 e = &s->extents[i]; 222 g_free(e->l1_table); 223 g_free(e->l2_cache); 224 g_free(e->l1_backup_table); 225 g_free(e->type); 226 if (e->file != bs->file) { 227 bdrv_unref_child(bs, e->file); 228 } 229 } 230 g_free(s->extents); 231 } 232 233 static void vmdk_free_last_extent(BlockDriverState *bs) 234 { 235 BDRVVmdkState *s = bs->opaque; 236 237 if (s->num_extents == 0) { 238 return; 239 } 240 s->num_extents--; 241 s->extents = g_renew(VmdkExtent, s->extents, s->num_extents); 242 } 243 244 static uint32_t vmdk_read_cid(BlockDriverState *bs, int parent) 245 { 246 char *desc; 247 uint32_t cid = 0xffffffff; 248 const char *p_name, *cid_str; 249 size_t cid_str_size; 250 BDRVVmdkState *s = bs->opaque; 251 int ret; 252 253 desc = g_malloc0(DESC_SIZE); 254 ret = bdrv_pread(bs->file->bs, s->desc_offset, desc, DESC_SIZE); 255 if (ret < 0) { 256 g_free(desc); 257 return 0; 258 } 259 260 if (parent) { 261 cid_str = "parentCID"; 262 cid_str_size = sizeof("parentCID"); 263 } else { 264 cid_str = "CID"; 265 cid_str_size = sizeof("CID"); 266 } 267 268 desc[DESC_SIZE - 1] = '\0'; 269 p_name = strstr(desc, cid_str); 270 if (p_name != NULL) { 271 p_name += cid_str_size; 272 sscanf(p_name, "%" SCNx32, &cid); 273 } 274 275 g_free(desc); 276 return cid; 277 } 278 279 static int vmdk_write_cid(BlockDriverState *bs, uint32_t cid) 280 { 281 char *desc, *tmp_desc; 282 char *p_name, *tmp_str; 283 BDRVVmdkState *s = bs->opaque; 284 int ret = 0; 285 286 desc = g_malloc0(DESC_SIZE); 287 tmp_desc = g_malloc0(DESC_SIZE); 288 ret = bdrv_pread(bs->file->bs, s->desc_offset, desc, DESC_SIZE); 289 if (ret < 0) { 290 goto out; 291 } 292 293 desc[DESC_SIZE - 1] = '\0'; 294 tmp_str = strstr(desc, "parentCID"); 295 if (tmp_str == NULL) { 296 ret = -EINVAL; 297 goto out; 298 } 299 300 pstrcpy(tmp_desc, DESC_SIZE, tmp_str); 301 p_name = strstr(desc, "CID"); 302 if (p_name != NULL) { 303 p_name += sizeof("CID"); 304 snprintf(p_name, DESC_SIZE - (p_name - desc), "%" PRIx32 "\n", cid); 305 pstrcat(desc, DESC_SIZE, tmp_desc); 306 } 307 308 ret = bdrv_pwrite_sync(bs->file->bs, s->desc_offset, desc, DESC_SIZE); 309 310 out: 311 g_free(desc); 312 g_free(tmp_desc); 313 return ret; 314 } 315 316 static int vmdk_is_cid_valid(BlockDriverState *bs) 317 { 318 BDRVVmdkState *s = bs->opaque; 319 uint32_t cur_pcid; 320 321 if (!s->cid_checked && bs->backing) { 322 BlockDriverState *p_bs = bs->backing->bs; 323 324 cur_pcid = vmdk_read_cid(p_bs, 0); 325 if (s->parent_cid != cur_pcid) { 326 /* CID not valid */ 327 return 0; 328 } 329 } 330 s->cid_checked = true; 331 /* CID valid */ 332 return 1; 333 } 334 335 /* We have nothing to do for VMDK reopen, stubs just return success */ 336 static int vmdk_reopen_prepare(BDRVReopenState *state, 337 BlockReopenQueue *queue, Error **errp) 338 { 339 assert(state != NULL); 340 assert(state->bs != NULL); 341 return 0; 342 } 343 344 static int vmdk_parent_open(BlockDriverState *bs) 345 { 346 char *p_name; 347 char *desc; 348 BDRVVmdkState *s = bs->opaque; 349 int ret; 350 351 desc = g_malloc0(DESC_SIZE + 1); 352 ret = bdrv_pread(bs->file->bs, s->desc_offset, desc, DESC_SIZE); 353 if (ret < 0) { 354 goto out; 355 } 356 ret = 0; 357 358 p_name = strstr(desc, "parentFileNameHint"); 359 if (p_name != NULL) { 360 char *end_name; 361 362 p_name += sizeof("parentFileNameHint") + 1; 363 end_name = strchr(p_name, '\"'); 364 if (end_name == NULL) { 365 ret = -EINVAL; 366 goto out; 367 } 368 if ((end_name - p_name) > sizeof(bs->backing_file) - 1) { 369 ret = -EINVAL; 370 goto out; 371 } 372 373 pstrcpy(bs->backing_file, end_name - p_name + 1, p_name); 374 } 375 376 out: 377 g_free(desc); 378 return ret; 379 } 380 381 /* Create and append extent to the extent array. Return the added VmdkExtent 382 * address. return NULL if allocation failed. */ 383 static int vmdk_add_extent(BlockDriverState *bs, 384 BdrvChild *file, bool flat, int64_t sectors, 385 int64_t l1_offset, int64_t l1_backup_offset, 386 uint32_t l1_size, 387 int l2_size, uint64_t cluster_sectors, 388 VmdkExtent **new_extent, 389 Error **errp) 390 { 391 VmdkExtent *extent; 392 BDRVVmdkState *s = bs->opaque; 393 int64_t nb_sectors; 394 395 if (cluster_sectors > 0x200000) { 396 /* 0x200000 * 512Bytes = 1GB for one cluster is unrealistic */ 397 error_setg(errp, "Invalid granularity, image may be corrupt"); 398 return -EFBIG; 399 } 400 if (l1_size > 512 * 1024 * 1024) { 401 /* Although with big capacity and small l1_entry_sectors, we can get a 402 * big l1_size, we don't want unbounded value to allocate the table. 403 * Limit it to 512M, which is 16PB for default cluster and L2 table 404 * size */ 405 error_setg(errp, "L1 size too big"); 406 return -EFBIG; 407 } 408 409 nb_sectors = bdrv_nb_sectors(file->bs); 410 if (nb_sectors < 0) { 411 return nb_sectors; 412 } 413 414 s->extents = g_renew(VmdkExtent, s->extents, s->num_extents + 1); 415 extent = &s->extents[s->num_extents]; 416 s->num_extents++; 417 418 memset(extent, 0, sizeof(VmdkExtent)); 419 extent->file = file; 420 extent->flat = flat; 421 extent->sectors = sectors; 422 extent->l1_table_offset = l1_offset; 423 extent->l1_backup_table_offset = l1_backup_offset; 424 extent->l1_size = l1_size; 425 extent->l1_entry_sectors = l2_size * cluster_sectors; 426 extent->l2_size = l2_size; 427 extent->cluster_sectors = flat ? sectors : cluster_sectors; 428 extent->next_cluster_sector = ROUND_UP(nb_sectors, cluster_sectors); 429 430 if (s->num_extents > 1) { 431 extent->end_sector = (*(extent - 1)).end_sector + extent->sectors; 432 } else { 433 extent->end_sector = extent->sectors; 434 } 435 bs->total_sectors = extent->end_sector; 436 if (new_extent) { 437 *new_extent = extent; 438 } 439 return 0; 440 } 441 442 static int vmdk_init_tables(BlockDriverState *bs, VmdkExtent *extent, 443 Error **errp) 444 { 445 int ret; 446 size_t l1_size; 447 int i; 448 449 /* read the L1 table */ 450 l1_size = extent->l1_size * sizeof(uint32_t); 451 extent->l1_table = g_try_malloc(l1_size); 452 if (l1_size && extent->l1_table == NULL) { 453 return -ENOMEM; 454 } 455 456 ret = bdrv_pread(extent->file->bs, 457 extent->l1_table_offset, 458 extent->l1_table, 459 l1_size); 460 if (ret < 0) { 461 error_setg_errno(errp, -ret, 462 "Could not read l1 table from extent '%s'", 463 extent->file->bs->filename); 464 goto fail_l1; 465 } 466 for (i = 0; i < extent->l1_size; i++) { 467 le32_to_cpus(&extent->l1_table[i]); 468 } 469 470 if (extent->l1_backup_table_offset) { 471 extent->l1_backup_table = g_try_malloc(l1_size); 472 if (l1_size && extent->l1_backup_table == NULL) { 473 ret = -ENOMEM; 474 goto fail_l1; 475 } 476 ret = bdrv_pread(extent->file->bs, 477 extent->l1_backup_table_offset, 478 extent->l1_backup_table, 479 l1_size); 480 if (ret < 0) { 481 error_setg_errno(errp, -ret, 482 "Could not read l1 backup table from extent '%s'", 483 extent->file->bs->filename); 484 goto fail_l1b; 485 } 486 for (i = 0; i < extent->l1_size; i++) { 487 le32_to_cpus(&extent->l1_backup_table[i]); 488 } 489 } 490 491 extent->l2_cache = 492 g_new(uint32_t, extent->l2_size * L2_CACHE_SIZE); 493 return 0; 494 fail_l1b: 495 g_free(extent->l1_backup_table); 496 fail_l1: 497 g_free(extent->l1_table); 498 return ret; 499 } 500 501 static int vmdk_open_vmfs_sparse(BlockDriverState *bs, 502 BdrvChild *file, 503 int flags, Error **errp) 504 { 505 int ret; 506 uint32_t magic; 507 VMDK3Header header; 508 VmdkExtent *extent; 509 510 ret = bdrv_pread(file->bs, sizeof(magic), &header, sizeof(header)); 511 if (ret < 0) { 512 error_setg_errno(errp, -ret, 513 "Could not read header from file '%s'", 514 file->bs->filename); 515 return ret; 516 } 517 ret = vmdk_add_extent(bs, file, false, 518 le32_to_cpu(header.disk_sectors), 519 (int64_t)le32_to_cpu(header.l1dir_offset) << 9, 520 0, 521 le32_to_cpu(header.l1dir_size), 522 4096, 523 le32_to_cpu(header.granularity), 524 &extent, 525 errp); 526 if (ret < 0) { 527 return ret; 528 } 529 ret = vmdk_init_tables(bs, extent, errp); 530 if (ret) { 531 /* free extent allocated by vmdk_add_extent */ 532 vmdk_free_last_extent(bs); 533 } 534 return ret; 535 } 536 537 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf, 538 QDict *options, Error **errp); 539 540 static char *vmdk_read_desc(BlockDriverState *file, uint64_t desc_offset, 541 Error **errp) 542 { 543 int64_t size; 544 char *buf; 545 int ret; 546 547 size = bdrv_getlength(file); 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->bs, 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->bs, 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->bs, 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 char buf[64]; 665 snprintf(buf, sizeof(buf), "VMDK version %" PRId32, 666 le32_to_cpu(header.version)); 667 error_setg(errp, QERR_UNKNOWN_BLOCK_FORMAT_FEATURE, 668 bdrv_get_device_or_node_name(bs), "vmdk", buf); 669 return -ENOTSUP; 670 } else if (le32_to_cpu(header.version) == 3 && (flags & BDRV_O_RDWR) && 671 !compressed) { 672 /* VMware KB 2064959 explains that version 3 added support for 673 * persistent changed block tracking (CBT), and backup software can 674 * read it as version=1 if it doesn't care about the changed area 675 * information. So we are safe to enable read only. */ 676 error_setg(errp, "VMDK version 3 must be read only"); 677 return -EINVAL; 678 } 679 680 if (le32_to_cpu(header.num_gtes_per_gt) > 512) { 681 error_setg(errp, "L2 table size too big"); 682 return -EINVAL; 683 } 684 685 l1_entry_sectors = le32_to_cpu(header.num_gtes_per_gt) 686 * le64_to_cpu(header.granularity); 687 if (l1_entry_sectors == 0) { 688 error_setg(errp, "L1 entry size is invalid"); 689 return -EINVAL; 690 } 691 l1_size = (le64_to_cpu(header.capacity) + l1_entry_sectors - 1) 692 / l1_entry_sectors; 693 if (le32_to_cpu(header.flags) & VMDK4_FLAG_RGD) { 694 l1_backup_offset = le64_to_cpu(header.rgd_offset) << 9; 695 } 696 if (bdrv_nb_sectors(file->bs) < le64_to_cpu(header.grain_offset)) { 697 error_setg(errp, "File truncated, expecting at least %" PRId64 " bytes", 698 (int64_t)(le64_to_cpu(header.grain_offset) 699 * BDRV_SECTOR_SIZE)); 700 return -EINVAL; 701 } 702 703 ret = vmdk_add_extent(bs, file, false, 704 le64_to_cpu(header.capacity), 705 le64_to_cpu(header.gd_offset) << 9, 706 l1_backup_offset, 707 l1_size, 708 le32_to_cpu(header.num_gtes_per_gt), 709 le64_to_cpu(header.granularity), 710 &extent, 711 errp); 712 if (ret < 0) { 713 return ret; 714 } 715 extent->compressed = 716 le16_to_cpu(header.compressAlgorithm) == VMDK4_COMPRESSION_DEFLATE; 717 if (extent->compressed) { 718 g_free(s->create_type); 719 s->create_type = g_strdup("streamOptimized"); 720 } 721 extent->has_marker = le32_to_cpu(header.flags) & VMDK4_FLAG_MARKER; 722 extent->version = le32_to_cpu(header.version); 723 extent->has_zero_grain = le32_to_cpu(header.flags) & VMDK4_FLAG_ZERO_GRAIN; 724 ret = vmdk_init_tables(bs, extent, errp); 725 if (ret) { 726 /* free extent allocated by vmdk_add_extent */ 727 vmdk_free_last_extent(bs); 728 } 729 return ret; 730 } 731 732 /* find an option value out of descriptor file */ 733 static int vmdk_parse_description(const char *desc, const char *opt_name, 734 char *buf, int buf_size) 735 { 736 char *opt_pos, *opt_end; 737 const char *end = desc + strlen(desc); 738 739 opt_pos = strstr(desc, opt_name); 740 if (!opt_pos) { 741 return VMDK_ERROR; 742 } 743 /* Skip "=\"" following opt_name */ 744 opt_pos += strlen(opt_name) + 2; 745 if (opt_pos >= end) { 746 return VMDK_ERROR; 747 } 748 opt_end = opt_pos; 749 while (opt_end < end && *opt_end != '"') { 750 opt_end++; 751 } 752 if (opt_end == end || buf_size < opt_end - opt_pos + 1) { 753 return VMDK_ERROR; 754 } 755 pstrcpy(buf, opt_end - opt_pos + 1, opt_pos); 756 return VMDK_OK; 757 } 758 759 /* Open an extent file and append to bs array */ 760 static int vmdk_open_sparse(BlockDriverState *bs, BdrvChild *file, int flags, 761 char *buf, QDict *options, Error **errp) 762 { 763 uint32_t magic; 764 765 magic = ldl_be_p(buf); 766 switch (magic) { 767 case VMDK3_MAGIC: 768 return vmdk_open_vmfs_sparse(bs, file, flags, errp); 769 break; 770 case VMDK4_MAGIC: 771 return vmdk_open_vmdk4(bs, file, flags, options, errp); 772 break; 773 default: 774 error_setg(errp, "Image not in VMDK format"); 775 return -EINVAL; 776 break; 777 } 778 } 779 780 static const char *next_line(const char *s) 781 { 782 while (*s) { 783 if (*s == '\n') { 784 return s + 1; 785 } 786 s++; 787 } 788 return s; 789 } 790 791 static int vmdk_parse_extents(const char *desc, BlockDriverState *bs, 792 const char *desc_file_path, QDict *options, 793 Error **errp) 794 { 795 int ret; 796 int matches; 797 char access[11]; 798 char type[11]; 799 char fname[512]; 800 const char *p, *np; 801 int64_t sectors = 0; 802 int64_t flat_offset; 803 char *extent_path; 804 BdrvChild *extent_file; 805 BDRVVmdkState *s = bs->opaque; 806 VmdkExtent *extent; 807 char extent_opt_prefix[32]; 808 Error *local_err = NULL; 809 810 for (p = desc; *p; p = next_line(p)) { 811 /* parse extent line in one of below formats: 812 * 813 * RW [size in sectors] FLAT "file-name.vmdk" OFFSET 814 * RW [size in sectors] SPARSE "file-name.vmdk" 815 * RW [size in sectors] VMFS "file-name.vmdk" 816 * RW [size in sectors] VMFSSPARSE "file-name.vmdk" 817 */ 818 flat_offset = -1; 819 matches = sscanf(p, "%10s %" SCNd64 " %10s \"%511[^\n\r\"]\" %" SCNd64, 820 access, §ors, type, fname, &flat_offset); 821 if (matches < 4 || strcmp(access, "RW")) { 822 continue; 823 } else if (!strcmp(type, "FLAT")) { 824 if (matches != 5 || flat_offset < 0) { 825 goto invalid; 826 } 827 } else if (!strcmp(type, "VMFS")) { 828 if (matches == 4) { 829 flat_offset = 0; 830 } else { 831 goto invalid; 832 } 833 } else if (matches != 4) { 834 goto invalid; 835 } 836 837 if (sectors <= 0 || 838 (strcmp(type, "FLAT") && strcmp(type, "SPARSE") && 839 strcmp(type, "VMFS") && strcmp(type, "VMFSSPARSE")) || 840 (strcmp(access, "RW"))) { 841 continue; 842 } 843 844 if (!path_is_absolute(fname) && !path_has_protocol(fname) && 845 !desc_file_path[0]) 846 { 847 error_setg(errp, "Cannot use relative extent paths with VMDK " 848 "descriptor file '%s'", bs->file->bs->filename); 849 return -EINVAL; 850 } 851 852 extent_path = g_malloc0(PATH_MAX); 853 path_combine(extent_path, PATH_MAX, desc_file_path, fname); 854 855 ret = snprintf(extent_opt_prefix, 32, "extents.%d", s->num_extents); 856 assert(ret < 32); 857 858 extent_file = bdrv_open_child(extent_path, options, extent_opt_prefix, 859 bs, &child_file, false, &local_err); 860 g_free(extent_path); 861 if (local_err) { 862 error_propagate(errp, local_err); 863 return -EINVAL; 864 } 865 866 /* save to extents array */ 867 if (!strcmp(type, "FLAT") || !strcmp(type, "VMFS")) { 868 /* FLAT extent */ 869 870 ret = vmdk_add_extent(bs, extent_file, true, sectors, 871 0, 0, 0, 0, 0, &extent, errp); 872 if (ret < 0) { 873 bdrv_unref_child(bs, extent_file); 874 return ret; 875 } 876 extent->flat_start_offset = flat_offset << 9; 877 } else if (!strcmp(type, "SPARSE") || !strcmp(type, "VMFSSPARSE")) { 878 /* SPARSE extent and VMFSSPARSE extent are both "COWD" sparse file*/ 879 char *buf = vmdk_read_desc(extent_file->bs, 0, errp); 880 if (!buf) { 881 ret = -EINVAL; 882 } else { 883 ret = vmdk_open_sparse(bs, extent_file, bs->open_flags, buf, 884 options, errp); 885 } 886 g_free(buf); 887 if (ret) { 888 bdrv_unref_child(bs, extent_file); 889 return ret; 890 } 891 extent = &s->extents[s->num_extents - 1]; 892 } else { 893 error_setg(errp, "Unsupported extent type '%s'", type); 894 bdrv_unref_child(bs, extent_file); 895 return -ENOTSUP; 896 } 897 extent->type = g_strdup(type); 898 } 899 return 0; 900 901 invalid: 902 np = next_line(p); 903 assert(np != p); 904 if (np[-1] == '\n') { 905 np--; 906 } 907 error_setg(errp, "Invalid extent line: %.*s", (int)(np - p), p); 908 return -EINVAL; 909 } 910 911 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf, 912 QDict *options, Error **errp) 913 { 914 int ret; 915 char ct[128]; 916 BDRVVmdkState *s = bs->opaque; 917 918 if (vmdk_parse_description(buf, "createType", ct, sizeof(ct))) { 919 error_setg(errp, "invalid VMDK image descriptor"); 920 ret = -EINVAL; 921 goto exit; 922 } 923 if (strcmp(ct, "monolithicFlat") && 924 strcmp(ct, "vmfs") && 925 strcmp(ct, "vmfsSparse") && 926 strcmp(ct, "twoGbMaxExtentSparse") && 927 strcmp(ct, "twoGbMaxExtentFlat")) { 928 error_setg(errp, "Unsupported image type '%s'", ct); 929 ret = -ENOTSUP; 930 goto exit; 931 } 932 s->create_type = g_strdup(ct); 933 s->desc_offset = 0; 934 ret = vmdk_parse_extents(buf, bs, bs->file->bs->exact_filename, options, 935 errp); 936 exit: 937 return ret; 938 } 939 940 static int vmdk_open(BlockDriverState *bs, QDict *options, int flags, 941 Error **errp) 942 { 943 char *buf; 944 int ret; 945 BDRVVmdkState *s = bs->opaque; 946 uint32_t magic; 947 948 buf = vmdk_read_desc(bs->file->bs, 0, errp); 949 if (!buf) { 950 return -EINVAL; 951 } 952 953 magic = ldl_be_p(buf); 954 switch (magic) { 955 case VMDK3_MAGIC: 956 case VMDK4_MAGIC: 957 ret = vmdk_open_sparse(bs, bs->file, flags, buf, options, 958 errp); 959 s->desc_offset = 0x200; 960 break; 961 default: 962 ret = vmdk_open_desc_file(bs, flags, buf, options, errp); 963 break; 964 } 965 if (ret) { 966 goto fail; 967 } 968 969 /* try to open parent images, if exist */ 970 ret = vmdk_parent_open(bs); 971 if (ret) { 972 goto fail; 973 } 974 s->cid = vmdk_read_cid(bs, 0); 975 s->parent_cid = vmdk_read_cid(bs, 1); 976 qemu_co_mutex_init(&s->lock); 977 978 /* Disable migration when VMDK images are used */ 979 error_setg(&s->migration_blocker, "The vmdk format used by node '%s' " 980 "does not support live migration", 981 bdrv_get_device_or_node_name(bs)); 982 migrate_add_blocker(s->migration_blocker); 983 g_free(buf); 984 return 0; 985 986 fail: 987 g_free(buf); 988 g_free(s->create_type); 989 s->create_type = NULL; 990 vmdk_free_extents(bs); 991 return ret; 992 } 993 994 995 static void vmdk_refresh_limits(BlockDriverState *bs, Error **errp) 996 { 997 BDRVVmdkState *s = bs->opaque; 998 int i; 999 1000 for (i = 0; i < s->num_extents; i++) { 1001 if (!s->extents[i].flat) { 1002 bs->bl.write_zeroes_alignment = 1003 MAX(bs->bl.write_zeroes_alignment, 1004 s->extents[i].cluster_sectors); 1005 } 1006 } 1007 } 1008 1009 /** 1010 * get_whole_cluster 1011 * 1012 * Copy backing file's cluster that covers @sector_num, otherwise write zero, 1013 * to the cluster at @cluster_sector_num. 1014 * 1015 * If @skip_start_sector < @skip_end_sector, the relative range 1016 * [@skip_start_sector, @skip_end_sector) is not copied or written, and leave 1017 * it for call to write user data in the request. 1018 */ 1019 static int get_whole_cluster(BlockDriverState *bs, 1020 VmdkExtent *extent, 1021 uint64_t cluster_sector_num, 1022 uint64_t sector_num, 1023 uint64_t skip_start_sector, 1024 uint64_t skip_end_sector) 1025 { 1026 int ret = VMDK_OK; 1027 int64_t cluster_bytes; 1028 uint8_t *whole_grain; 1029 1030 /* For COW, align request sector_num to cluster start */ 1031 sector_num = QEMU_ALIGN_DOWN(sector_num, extent->cluster_sectors); 1032 cluster_bytes = extent->cluster_sectors << BDRV_SECTOR_BITS; 1033 whole_grain = qemu_blockalign(bs, cluster_bytes); 1034 1035 if (!bs->backing) { 1036 memset(whole_grain, 0, skip_start_sector << BDRV_SECTOR_BITS); 1037 memset(whole_grain + (skip_end_sector << BDRV_SECTOR_BITS), 0, 1038 cluster_bytes - (skip_end_sector << BDRV_SECTOR_BITS)); 1039 } 1040 1041 assert(skip_end_sector <= extent->cluster_sectors); 1042 /* we will be here if it's first write on non-exist grain(cluster). 1043 * try to read from parent image, if exist */ 1044 if (bs->backing && !vmdk_is_cid_valid(bs)) { 1045 ret = VMDK_ERROR; 1046 goto exit; 1047 } 1048 1049 /* Read backing data before skip range */ 1050 if (skip_start_sector > 0) { 1051 if (bs->backing) { 1052 ret = bdrv_read(bs->backing->bs, sector_num, 1053 whole_grain, skip_start_sector); 1054 if (ret < 0) { 1055 ret = VMDK_ERROR; 1056 goto exit; 1057 } 1058 } 1059 ret = bdrv_write(extent->file->bs, cluster_sector_num, whole_grain, 1060 skip_start_sector); 1061 if (ret < 0) { 1062 ret = VMDK_ERROR; 1063 goto exit; 1064 } 1065 } 1066 /* Read backing data after skip range */ 1067 if (skip_end_sector < extent->cluster_sectors) { 1068 if (bs->backing) { 1069 ret = bdrv_read(bs->backing->bs, sector_num + skip_end_sector, 1070 whole_grain + (skip_end_sector << BDRV_SECTOR_BITS), 1071 extent->cluster_sectors - skip_end_sector); 1072 if (ret < 0) { 1073 ret = VMDK_ERROR; 1074 goto exit; 1075 } 1076 } 1077 ret = bdrv_write(extent->file->bs, cluster_sector_num + skip_end_sector, 1078 whole_grain + (skip_end_sector << BDRV_SECTOR_BITS), 1079 extent->cluster_sectors - skip_end_sector); 1080 if (ret < 0) { 1081 ret = VMDK_ERROR; 1082 goto exit; 1083 } 1084 } 1085 1086 exit: 1087 qemu_vfree(whole_grain); 1088 return ret; 1089 } 1090 1091 static int vmdk_L2update(VmdkExtent *extent, VmdkMetaData *m_data, 1092 uint32_t offset) 1093 { 1094 offset = cpu_to_le32(offset); 1095 /* update L2 table */ 1096 if (bdrv_pwrite_sync( 1097 extent->file->bs, 1098 ((int64_t)m_data->l2_offset * 512) 1099 + (m_data->l2_index * sizeof(offset)), 1100 &offset, sizeof(offset)) < 0) { 1101 return VMDK_ERROR; 1102 } 1103 /* update backup L2 table */ 1104 if (extent->l1_backup_table_offset != 0) { 1105 m_data->l2_offset = extent->l1_backup_table[m_data->l1_index]; 1106 if (bdrv_pwrite_sync( 1107 extent->file->bs, 1108 ((int64_t)m_data->l2_offset * 512) 1109 + (m_data->l2_index * sizeof(offset)), 1110 &offset, sizeof(offset)) < 0) { 1111 return VMDK_ERROR; 1112 } 1113 } 1114 if (m_data->l2_cache_entry) { 1115 *m_data->l2_cache_entry = offset; 1116 } 1117 1118 return VMDK_OK; 1119 } 1120 1121 /** 1122 * get_cluster_offset 1123 * 1124 * Look up cluster offset in extent file by sector number, and store in 1125 * @cluster_offset. 1126 * 1127 * For flat extents, the start offset as parsed from the description file is 1128 * returned. 1129 * 1130 * For sparse extents, look up in L1, L2 table. If allocate is true, return an 1131 * offset for a new cluster and update L2 cache. If there is a backing file, 1132 * COW is done before returning; otherwise, zeroes are written to the allocated 1133 * cluster. Both COW and zero writing skips the sector range 1134 * [@skip_start_sector, @skip_end_sector) passed in by caller, because caller 1135 * has new data to write there. 1136 * 1137 * Returns: VMDK_OK if cluster exists and mapped in the image. 1138 * VMDK_UNALLOC if cluster is not mapped and @allocate is false. 1139 * VMDK_ERROR if failed. 1140 */ 1141 static int get_cluster_offset(BlockDriverState *bs, 1142 VmdkExtent *extent, 1143 VmdkMetaData *m_data, 1144 uint64_t offset, 1145 bool allocate, 1146 uint64_t *cluster_offset, 1147 uint64_t skip_start_sector, 1148 uint64_t skip_end_sector) 1149 { 1150 unsigned int l1_index, l2_offset, l2_index; 1151 int min_index, i, j; 1152 uint32_t min_count, *l2_table; 1153 bool zeroed = false; 1154 int64_t ret; 1155 int64_t cluster_sector; 1156 1157 if (m_data) { 1158 m_data->valid = 0; 1159 } 1160 if (extent->flat) { 1161 *cluster_offset = extent->flat_start_offset; 1162 return VMDK_OK; 1163 } 1164 1165 offset -= (extent->end_sector - extent->sectors) * SECTOR_SIZE; 1166 l1_index = (offset >> 9) / extent->l1_entry_sectors; 1167 if (l1_index >= extent->l1_size) { 1168 return VMDK_ERROR; 1169 } 1170 l2_offset = extent->l1_table[l1_index]; 1171 if (!l2_offset) { 1172 return VMDK_UNALLOC; 1173 } 1174 for (i = 0; i < L2_CACHE_SIZE; i++) { 1175 if (l2_offset == extent->l2_cache_offsets[i]) { 1176 /* increment the hit count */ 1177 if (++extent->l2_cache_counts[i] == 0xffffffff) { 1178 for (j = 0; j < L2_CACHE_SIZE; j++) { 1179 extent->l2_cache_counts[j] >>= 1; 1180 } 1181 } 1182 l2_table = extent->l2_cache + (i * extent->l2_size); 1183 goto found; 1184 } 1185 } 1186 /* not found: load a new entry in the least used one */ 1187 min_index = 0; 1188 min_count = 0xffffffff; 1189 for (i = 0; i < L2_CACHE_SIZE; i++) { 1190 if (extent->l2_cache_counts[i] < min_count) { 1191 min_count = extent->l2_cache_counts[i]; 1192 min_index = i; 1193 } 1194 } 1195 l2_table = extent->l2_cache + (min_index * extent->l2_size); 1196 if (bdrv_pread( 1197 extent->file->bs, 1198 (int64_t)l2_offset * 512, 1199 l2_table, 1200 extent->l2_size * sizeof(uint32_t) 1201 ) != extent->l2_size * sizeof(uint32_t)) { 1202 return VMDK_ERROR; 1203 } 1204 1205 extent->l2_cache_offsets[min_index] = l2_offset; 1206 extent->l2_cache_counts[min_index] = 1; 1207 found: 1208 l2_index = ((offset >> 9) / extent->cluster_sectors) % extent->l2_size; 1209 cluster_sector = le32_to_cpu(l2_table[l2_index]); 1210 1211 if (m_data) { 1212 m_data->valid = 1; 1213 m_data->l1_index = l1_index; 1214 m_data->l2_index = l2_index; 1215 m_data->l2_offset = l2_offset; 1216 m_data->l2_cache_entry = &l2_table[l2_index]; 1217 } 1218 if (extent->has_zero_grain && cluster_sector == VMDK_GTE_ZEROED) { 1219 zeroed = true; 1220 } 1221 1222 if (!cluster_sector || zeroed) { 1223 if (!allocate) { 1224 return zeroed ? VMDK_ZEROED : VMDK_UNALLOC; 1225 } 1226 1227 cluster_sector = extent->next_cluster_sector; 1228 extent->next_cluster_sector += extent->cluster_sectors; 1229 1230 /* First of all we write grain itself, to avoid race condition 1231 * that may to corrupt the image. 1232 * This problem may occur because of insufficient space on host disk 1233 * or inappropriate VM shutdown. 1234 */ 1235 ret = get_whole_cluster(bs, extent, 1236 cluster_sector, 1237 offset >> BDRV_SECTOR_BITS, 1238 skip_start_sector, skip_end_sector); 1239 if (ret) { 1240 return ret; 1241 } 1242 } 1243 *cluster_offset = cluster_sector << BDRV_SECTOR_BITS; 1244 return VMDK_OK; 1245 } 1246 1247 static VmdkExtent *find_extent(BDRVVmdkState *s, 1248 int64_t sector_num, VmdkExtent *start_hint) 1249 { 1250 VmdkExtent *extent = start_hint; 1251 1252 if (!extent) { 1253 extent = &s->extents[0]; 1254 } 1255 while (extent < &s->extents[s->num_extents]) { 1256 if (sector_num < extent->end_sector) { 1257 return extent; 1258 } 1259 extent++; 1260 } 1261 return NULL; 1262 } 1263 1264 static inline uint64_t vmdk_find_index_in_cluster(VmdkExtent *extent, 1265 int64_t sector_num) 1266 { 1267 uint64_t index_in_cluster, extent_begin_sector, extent_relative_sector_num; 1268 1269 extent_begin_sector = extent->end_sector - extent->sectors; 1270 extent_relative_sector_num = sector_num - extent_begin_sector; 1271 index_in_cluster = extent_relative_sector_num % extent->cluster_sectors; 1272 return index_in_cluster; 1273 } 1274 1275 static int64_t coroutine_fn vmdk_co_get_block_status(BlockDriverState *bs, 1276 int64_t sector_num, int nb_sectors, int *pnum, BlockDriverState **file) 1277 { 1278 BDRVVmdkState *s = bs->opaque; 1279 int64_t index_in_cluster, n, ret; 1280 uint64_t offset; 1281 VmdkExtent *extent; 1282 1283 extent = find_extent(s, sector_num, NULL); 1284 if (!extent) { 1285 return 0; 1286 } 1287 qemu_co_mutex_lock(&s->lock); 1288 ret = get_cluster_offset(bs, extent, NULL, 1289 sector_num * 512, false, &offset, 1290 0, 0); 1291 qemu_co_mutex_unlock(&s->lock); 1292 1293 index_in_cluster = vmdk_find_index_in_cluster(extent, sector_num); 1294 switch (ret) { 1295 case VMDK_ERROR: 1296 ret = -EIO; 1297 break; 1298 case VMDK_UNALLOC: 1299 ret = 0; 1300 break; 1301 case VMDK_ZEROED: 1302 ret = BDRV_BLOCK_ZERO; 1303 break; 1304 case VMDK_OK: 1305 ret = BDRV_BLOCK_DATA; 1306 if (!extent->compressed) { 1307 ret |= BDRV_BLOCK_OFFSET_VALID; 1308 ret |= (offset + (index_in_cluster << BDRV_SECTOR_BITS)) 1309 & BDRV_BLOCK_OFFSET_MASK; 1310 } 1311 *file = extent->file->bs; 1312 break; 1313 } 1314 1315 n = extent->cluster_sectors - index_in_cluster; 1316 if (n > nb_sectors) { 1317 n = nb_sectors; 1318 } 1319 *pnum = n; 1320 return ret; 1321 } 1322 1323 static int vmdk_write_extent(VmdkExtent *extent, int64_t cluster_offset, 1324 int64_t offset_in_cluster, const uint8_t *buf, 1325 int nb_sectors, int64_t sector_num) 1326 { 1327 int ret; 1328 VmdkGrainMarker *data = NULL; 1329 uLongf buf_len; 1330 const uint8_t *write_buf = buf; 1331 int write_len = nb_sectors * 512; 1332 int64_t write_offset; 1333 int64_t write_end_sector; 1334 1335 if (extent->compressed) { 1336 if (!extent->has_marker) { 1337 ret = -EINVAL; 1338 goto out; 1339 } 1340 buf_len = (extent->cluster_sectors << 9) * 2; 1341 data = g_malloc(buf_len + sizeof(VmdkGrainMarker)); 1342 if (compress(data->data, &buf_len, buf, nb_sectors << 9) != Z_OK || 1343 buf_len == 0) { 1344 ret = -EINVAL; 1345 goto out; 1346 } 1347 data->lba = sector_num; 1348 data->size = buf_len; 1349 write_buf = (uint8_t *)data; 1350 write_len = buf_len + sizeof(VmdkGrainMarker); 1351 } 1352 write_offset = cluster_offset + offset_in_cluster, 1353 ret = bdrv_pwrite(extent->file->bs, write_offset, write_buf, write_len); 1354 1355 write_end_sector = DIV_ROUND_UP(write_offset + write_len, BDRV_SECTOR_SIZE); 1356 1357 if (extent->compressed) { 1358 extent->next_cluster_sector = write_end_sector; 1359 } else { 1360 extent->next_cluster_sector = MAX(extent->next_cluster_sector, 1361 write_end_sector); 1362 } 1363 1364 if (ret != write_len) { 1365 ret = ret < 0 ? ret : -EIO; 1366 goto out; 1367 } 1368 ret = 0; 1369 out: 1370 g_free(data); 1371 return ret; 1372 } 1373 1374 static int vmdk_read_extent(VmdkExtent *extent, int64_t cluster_offset, 1375 int64_t offset_in_cluster, uint8_t *buf, 1376 int nb_sectors) 1377 { 1378 int ret; 1379 int cluster_bytes, buf_bytes; 1380 uint8_t *cluster_buf, *compressed_data; 1381 uint8_t *uncomp_buf; 1382 uint32_t data_len; 1383 VmdkGrainMarker *marker; 1384 uLongf buf_len; 1385 1386 1387 if (!extent->compressed) { 1388 ret = bdrv_pread(extent->file->bs, 1389 cluster_offset + offset_in_cluster, 1390 buf, nb_sectors * 512); 1391 if (ret == nb_sectors * 512) { 1392 return 0; 1393 } else { 1394 return -EIO; 1395 } 1396 } 1397 cluster_bytes = extent->cluster_sectors * 512; 1398 /* Read two clusters in case GrainMarker + compressed data > one cluster */ 1399 buf_bytes = cluster_bytes * 2; 1400 cluster_buf = g_malloc(buf_bytes); 1401 uncomp_buf = g_malloc(cluster_bytes); 1402 ret = bdrv_pread(extent->file->bs, 1403 cluster_offset, 1404 cluster_buf, buf_bytes); 1405 if (ret < 0) { 1406 goto out; 1407 } 1408 compressed_data = cluster_buf; 1409 buf_len = cluster_bytes; 1410 data_len = cluster_bytes; 1411 if (extent->has_marker) { 1412 marker = (VmdkGrainMarker *)cluster_buf; 1413 compressed_data = marker->data; 1414 data_len = le32_to_cpu(marker->size); 1415 } 1416 if (!data_len || data_len > buf_bytes) { 1417 ret = -EINVAL; 1418 goto out; 1419 } 1420 ret = uncompress(uncomp_buf, &buf_len, compressed_data, data_len); 1421 if (ret != Z_OK) { 1422 ret = -EINVAL; 1423 goto out; 1424 1425 } 1426 if (offset_in_cluster < 0 || 1427 offset_in_cluster + nb_sectors * 512 > buf_len) { 1428 ret = -EINVAL; 1429 goto out; 1430 } 1431 memcpy(buf, uncomp_buf + offset_in_cluster, nb_sectors * 512); 1432 ret = 0; 1433 1434 out: 1435 g_free(uncomp_buf); 1436 g_free(cluster_buf); 1437 return ret; 1438 } 1439 1440 static int vmdk_read(BlockDriverState *bs, int64_t sector_num, 1441 uint8_t *buf, int nb_sectors) 1442 { 1443 BDRVVmdkState *s = bs->opaque; 1444 int ret; 1445 uint64_t n, index_in_cluster; 1446 VmdkExtent *extent = NULL; 1447 uint64_t cluster_offset; 1448 1449 while (nb_sectors > 0) { 1450 extent = find_extent(s, sector_num, extent); 1451 if (!extent) { 1452 return -EIO; 1453 } 1454 ret = get_cluster_offset(bs, extent, NULL, 1455 sector_num << 9, false, &cluster_offset, 1456 0, 0); 1457 index_in_cluster = vmdk_find_index_in_cluster(extent, sector_num); 1458 n = extent->cluster_sectors - index_in_cluster; 1459 if (n > nb_sectors) { 1460 n = nb_sectors; 1461 } 1462 if (ret != VMDK_OK) { 1463 /* if not allocated, try to read from parent image, if exist */ 1464 if (bs->backing && ret != VMDK_ZEROED) { 1465 if (!vmdk_is_cid_valid(bs)) { 1466 return -EINVAL; 1467 } 1468 ret = bdrv_read(bs->backing->bs, sector_num, buf, n); 1469 if (ret < 0) { 1470 return ret; 1471 } 1472 } else { 1473 memset(buf, 0, 512 * n); 1474 } 1475 } else { 1476 ret = vmdk_read_extent(extent, 1477 cluster_offset, index_in_cluster * 512, 1478 buf, n); 1479 if (ret) { 1480 return ret; 1481 } 1482 } 1483 nb_sectors -= n; 1484 sector_num += n; 1485 buf += n * 512; 1486 } 1487 return 0; 1488 } 1489 1490 static coroutine_fn int vmdk_co_read(BlockDriverState *bs, int64_t sector_num, 1491 uint8_t *buf, int nb_sectors) 1492 { 1493 int ret; 1494 BDRVVmdkState *s = bs->opaque; 1495 qemu_co_mutex_lock(&s->lock); 1496 ret = vmdk_read(bs, sector_num, buf, nb_sectors); 1497 qemu_co_mutex_unlock(&s->lock); 1498 return ret; 1499 } 1500 1501 /** 1502 * vmdk_write: 1503 * @zeroed: buf is ignored (data is zero), use zeroed_grain GTE feature 1504 * if possible, otherwise return -ENOTSUP. 1505 * @zero_dry_run: used for zeroed == true only, don't update L2 table, just try 1506 * with each cluster. By dry run we can find if the zero write 1507 * is possible without modifying image data. 1508 * 1509 * Returns: error code with 0 for success. 1510 */ 1511 static int vmdk_write(BlockDriverState *bs, int64_t sector_num, 1512 const uint8_t *buf, int nb_sectors, 1513 bool zeroed, bool zero_dry_run) 1514 { 1515 BDRVVmdkState *s = bs->opaque; 1516 VmdkExtent *extent = NULL; 1517 int ret; 1518 int64_t index_in_cluster, n; 1519 uint64_t cluster_offset; 1520 VmdkMetaData m_data; 1521 1522 if (sector_num > bs->total_sectors) { 1523 error_report("Wrong offset: sector_num=0x%" PRIx64 1524 " total_sectors=0x%" PRIx64, 1525 sector_num, bs->total_sectors); 1526 return -EIO; 1527 } 1528 1529 while (nb_sectors > 0) { 1530 extent = find_extent(s, sector_num, extent); 1531 if (!extent) { 1532 return -EIO; 1533 } 1534 index_in_cluster = vmdk_find_index_in_cluster(extent, sector_num); 1535 n = extent->cluster_sectors - index_in_cluster; 1536 if (n > nb_sectors) { 1537 n = nb_sectors; 1538 } 1539 ret = get_cluster_offset(bs, extent, &m_data, sector_num << 9, 1540 !(extent->compressed || zeroed), 1541 &cluster_offset, 1542 index_in_cluster, index_in_cluster + n); 1543 if (extent->compressed) { 1544 if (ret == VMDK_OK) { 1545 /* Refuse write to allocated cluster for streamOptimized */ 1546 error_report("Could not write to allocated cluster" 1547 " for streamOptimized"); 1548 return -EIO; 1549 } else { 1550 /* allocate */ 1551 ret = get_cluster_offset(bs, extent, &m_data, sector_num << 9, 1552 true, &cluster_offset, 0, 0); 1553 } 1554 } 1555 if (ret == VMDK_ERROR) { 1556 return -EINVAL; 1557 } 1558 if (zeroed) { 1559 /* Do zeroed write, buf is ignored */ 1560 if (extent->has_zero_grain && 1561 index_in_cluster == 0 && 1562 n >= extent->cluster_sectors) { 1563 n = extent->cluster_sectors; 1564 if (!zero_dry_run) { 1565 /* update L2 tables */ 1566 if (vmdk_L2update(extent, &m_data, VMDK_GTE_ZEROED) 1567 != VMDK_OK) { 1568 return -EIO; 1569 } 1570 } 1571 } else { 1572 return -ENOTSUP; 1573 } 1574 } else { 1575 ret = vmdk_write_extent(extent, 1576 cluster_offset, index_in_cluster * 512, 1577 buf, n, sector_num); 1578 if (ret) { 1579 return ret; 1580 } 1581 if (m_data.valid) { 1582 /* update L2 tables */ 1583 if (vmdk_L2update(extent, &m_data, 1584 cluster_offset >> BDRV_SECTOR_BITS) 1585 != VMDK_OK) { 1586 return -EIO; 1587 } 1588 } 1589 } 1590 nb_sectors -= n; 1591 sector_num += n; 1592 buf += n * 512; 1593 1594 /* update CID on the first write every time the virtual disk is 1595 * opened */ 1596 if (!s->cid_updated) { 1597 ret = vmdk_write_cid(bs, g_random_int()); 1598 if (ret < 0) { 1599 return ret; 1600 } 1601 s->cid_updated = true; 1602 } 1603 } 1604 return 0; 1605 } 1606 1607 static coroutine_fn int vmdk_co_write(BlockDriverState *bs, int64_t sector_num, 1608 const uint8_t *buf, int nb_sectors) 1609 { 1610 int ret; 1611 BDRVVmdkState *s = bs->opaque; 1612 qemu_co_mutex_lock(&s->lock); 1613 ret = vmdk_write(bs, sector_num, buf, nb_sectors, false, false); 1614 qemu_co_mutex_unlock(&s->lock); 1615 return ret; 1616 } 1617 1618 static int vmdk_write_compressed(BlockDriverState *bs, 1619 int64_t sector_num, 1620 const uint8_t *buf, 1621 int nb_sectors) 1622 { 1623 BDRVVmdkState *s = bs->opaque; 1624 if (s->num_extents == 1 && s->extents[0].compressed) { 1625 return vmdk_write(bs, sector_num, buf, nb_sectors, false, false); 1626 } else { 1627 return -ENOTSUP; 1628 } 1629 } 1630 1631 static int coroutine_fn vmdk_co_write_zeroes(BlockDriverState *bs, 1632 int64_t sector_num, 1633 int nb_sectors, 1634 BdrvRequestFlags flags) 1635 { 1636 int ret; 1637 BDRVVmdkState *s = bs->opaque; 1638 qemu_co_mutex_lock(&s->lock); 1639 /* write zeroes could fail if sectors not aligned to cluster, test it with 1640 * dry_run == true before really updating image */ 1641 ret = vmdk_write(bs, sector_num, NULL, nb_sectors, true, true); 1642 if (!ret) { 1643 ret = vmdk_write(bs, sector_num, NULL, nb_sectors, true, false); 1644 } 1645 qemu_co_mutex_unlock(&s->lock); 1646 return ret; 1647 } 1648 1649 static int vmdk_create_extent(const char *filename, int64_t filesize, 1650 bool flat, bool compress, bool zeroed_grain, 1651 QemuOpts *opts, Error **errp) 1652 { 1653 int ret, i; 1654 BlockBackend *blk = NULL; 1655 VMDK4Header header; 1656 Error *local_err = NULL; 1657 uint32_t tmp, magic, grains, gd_sectors, gt_size, gt_count; 1658 uint32_t *gd_buf = NULL; 1659 int gd_buf_size; 1660 1661 ret = bdrv_create_file(filename, opts, &local_err); 1662 if (ret < 0) { 1663 error_propagate(errp, local_err); 1664 goto exit; 1665 } 1666 1667 blk = blk_new_open("extent", filename, NULL, NULL, 1668 BDRV_O_RDWR | BDRV_O_CACHE_WB | BDRV_O_PROTOCOL, 1669 &local_err); 1670 if (blk == NULL) { 1671 error_propagate(errp, local_err); 1672 ret = -EIO; 1673 goto exit; 1674 } 1675 1676 blk_set_allow_write_beyond_eof(blk, true); 1677 1678 if (flat) { 1679 ret = blk_truncate(blk, filesize); 1680 if (ret < 0) { 1681 error_setg_errno(errp, -ret, "Could not truncate file"); 1682 } 1683 goto exit; 1684 } 1685 magic = cpu_to_be32(VMDK4_MAGIC); 1686 memset(&header, 0, sizeof(header)); 1687 if (compress) { 1688 header.version = 3; 1689 } else if (zeroed_grain) { 1690 header.version = 2; 1691 } else { 1692 header.version = 1; 1693 } 1694 header.flags = VMDK4_FLAG_RGD | VMDK4_FLAG_NL_DETECT 1695 | (compress ? VMDK4_FLAG_COMPRESS | VMDK4_FLAG_MARKER : 0) 1696 | (zeroed_grain ? VMDK4_FLAG_ZERO_GRAIN : 0); 1697 header.compressAlgorithm = compress ? VMDK4_COMPRESSION_DEFLATE : 0; 1698 header.capacity = filesize / BDRV_SECTOR_SIZE; 1699 header.granularity = 128; 1700 header.num_gtes_per_gt = BDRV_SECTOR_SIZE; 1701 1702 grains = DIV_ROUND_UP(filesize / BDRV_SECTOR_SIZE, header.granularity); 1703 gt_size = DIV_ROUND_UP(header.num_gtes_per_gt * sizeof(uint32_t), 1704 BDRV_SECTOR_SIZE); 1705 gt_count = DIV_ROUND_UP(grains, header.num_gtes_per_gt); 1706 gd_sectors = DIV_ROUND_UP(gt_count * sizeof(uint32_t), BDRV_SECTOR_SIZE); 1707 1708 header.desc_offset = 1; 1709 header.desc_size = 20; 1710 header.rgd_offset = header.desc_offset + header.desc_size; 1711 header.gd_offset = header.rgd_offset + gd_sectors + (gt_size * gt_count); 1712 header.grain_offset = 1713 ROUND_UP(header.gd_offset + gd_sectors + (gt_size * gt_count), 1714 header.granularity); 1715 /* swap endianness for all header fields */ 1716 header.version = cpu_to_le32(header.version); 1717 header.flags = cpu_to_le32(header.flags); 1718 header.capacity = cpu_to_le64(header.capacity); 1719 header.granularity = cpu_to_le64(header.granularity); 1720 header.num_gtes_per_gt = cpu_to_le32(header.num_gtes_per_gt); 1721 header.desc_offset = cpu_to_le64(header.desc_offset); 1722 header.desc_size = cpu_to_le64(header.desc_size); 1723 header.rgd_offset = cpu_to_le64(header.rgd_offset); 1724 header.gd_offset = cpu_to_le64(header.gd_offset); 1725 header.grain_offset = cpu_to_le64(header.grain_offset); 1726 header.compressAlgorithm = cpu_to_le16(header.compressAlgorithm); 1727 1728 header.check_bytes[0] = 0xa; 1729 header.check_bytes[1] = 0x20; 1730 header.check_bytes[2] = 0xd; 1731 header.check_bytes[3] = 0xa; 1732 1733 /* write all the data */ 1734 ret = blk_pwrite(blk, 0, &magic, sizeof(magic)); 1735 if (ret < 0) { 1736 error_setg(errp, QERR_IO_ERROR); 1737 goto exit; 1738 } 1739 ret = blk_pwrite(blk, sizeof(magic), &header, sizeof(header)); 1740 if (ret < 0) { 1741 error_setg(errp, QERR_IO_ERROR); 1742 goto exit; 1743 } 1744 1745 ret = blk_truncate(blk, le64_to_cpu(header.grain_offset) << 9); 1746 if (ret < 0) { 1747 error_setg_errno(errp, -ret, "Could not truncate file"); 1748 goto exit; 1749 } 1750 1751 /* write grain directory */ 1752 gd_buf_size = gd_sectors * BDRV_SECTOR_SIZE; 1753 gd_buf = g_malloc0(gd_buf_size); 1754 for (i = 0, tmp = le64_to_cpu(header.rgd_offset) + gd_sectors; 1755 i < gt_count; i++, tmp += gt_size) { 1756 gd_buf[i] = cpu_to_le32(tmp); 1757 } 1758 ret = blk_pwrite(blk, le64_to_cpu(header.rgd_offset) * BDRV_SECTOR_SIZE, 1759 gd_buf, gd_buf_size); 1760 if (ret < 0) { 1761 error_setg(errp, QERR_IO_ERROR); 1762 goto exit; 1763 } 1764 1765 /* write backup grain directory */ 1766 for (i = 0, tmp = le64_to_cpu(header.gd_offset) + gd_sectors; 1767 i < gt_count; i++, tmp += gt_size) { 1768 gd_buf[i] = cpu_to_le32(tmp); 1769 } 1770 ret = blk_pwrite(blk, le64_to_cpu(header.gd_offset) * BDRV_SECTOR_SIZE, 1771 gd_buf, gd_buf_size); 1772 if (ret < 0) { 1773 error_setg(errp, QERR_IO_ERROR); 1774 goto exit; 1775 } 1776 1777 ret = 0; 1778 exit: 1779 if (blk) { 1780 blk_unref(blk); 1781 } 1782 g_free(gd_buf); 1783 return ret; 1784 } 1785 1786 static int filename_decompose(const char *filename, char *path, char *prefix, 1787 char *postfix, size_t buf_len, Error **errp) 1788 { 1789 const char *p, *q; 1790 1791 if (filename == NULL || !strlen(filename)) { 1792 error_setg(errp, "No filename provided"); 1793 return VMDK_ERROR; 1794 } 1795 p = strrchr(filename, '/'); 1796 if (p == NULL) { 1797 p = strrchr(filename, '\\'); 1798 } 1799 if (p == NULL) { 1800 p = strrchr(filename, ':'); 1801 } 1802 if (p != NULL) { 1803 p++; 1804 if (p - filename >= buf_len) { 1805 return VMDK_ERROR; 1806 } 1807 pstrcpy(path, p - filename + 1, filename); 1808 } else { 1809 p = filename; 1810 path[0] = '\0'; 1811 } 1812 q = strrchr(p, '.'); 1813 if (q == NULL) { 1814 pstrcpy(prefix, buf_len, p); 1815 postfix[0] = '\0'; 1816 } else { 1817 if (q - p >= buf_len) { 1818 return VMDK_ERROR; 1819 } 1820 pstrcpy(prefix, q - p + 1, p); 1821 pstrcpy(postfix, buf_len, q); 1822 } 1823 return VMDK_OK; 1824 } 1825 1826 static int vmdk_create(const char *filename, QemuOpts *opts, Error **errp) 1827 { 1828 int idx = 0; 1829 BlockBackend *new_blk = NULL; 1830 Error *local_err = NULL; 1831 char *desc = NULL; 1832 int64_t total_size = 0, filesize; 1833 char *adapter_type = NULL; 1834 char *backing_file = NULL; 1835 char *fmt = NULL; 1836 int flags = 0; 1837 int ret = 0; 1838 bool flat, split, compress; 1839 GString *ext_desc_lines; 1840 char *path = g_malloc0(PATH_MAX); 1841 char *prefix = g_malloc0(PATH_MAX); 1842 char *postfix = g_malloc0(PATH_MAX); 1843 char *desc_line = g_malloc0(BUF_SIZE); 1844 char *ext_filename = g_malloc0(PATH_MAX); 1845 char *desc_filename = g_malloc0(PATH_MAX); 1846 const int64_t split_size = 0x80000000; /* VMDK has constant split size */ 1847 const char *desc_extent_line; 1848 char *parent_desc_line = g_malloc0(BUF_SIZE); 1849 uint32_t parent_cid = 0xffffffff; 1850 uint32_t number_heads = 16; 1851 bool zeroed_grain = false; 1852 uint32_t desc_offset = 0, desc_len; 1853 const char desc_template[] = 1854 "# Disk DescriptorFile\n" 1855 "version=1\n" 1856 "CID=%" PRIx32 "\n" 1857 "parentCID=%" PRIx32 "\n" 1858 "createType=\"%s\"\n" 1859 "%s" 1860 "\n" 1861 "# Extent description\n" 1862 "%s" 1863 "\n" 1864 "# The Disk Data Base\n" 1865 "#DDB\n" 1866 "\n" 1867 "ddb.virtualHWVersion = \"%d\"\n" 1868 "ddb.geometry.cylinders = \"%" PRId64 "\"\n" 1869 "ddb.geometry.heads = \"%" PRIu32 "\"\n" 1870 "ddb.geometry.sectors = \"63\"\n" 1871 "ddb.adapterType = \"%s\"\n"; 1872 1873 ext_desc_lines = g_string_new(NULL); 1874 1875 if (filename_decompose(filename, path, prefix, postfix, PATH_MAX, errp)) { 1876 ret = -EINVAL; 1877 goto exit; 1878 } 1879 /* Read out options */ 1880 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0), 1881 BDRV_SECTOR_SIZE); 1882 adapter_type = qemu_opt_get_del(opts, BLOCK_OPT_ADAPTER_TYPE); 1883 backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE); 1884 if (qemu_opt_get_bool_del(opts, BLOCK_OPT_COMPAT6, false)) { 1885 flags |= BLOCK_FLAG_COMPAT6; 1886 } 1887 fmt = qemu_opt_get_del(opts, BLOCK_OPT_SUBFMT); 1888 if (qemu_opt_get_bool_del(opts, BLOCK_OPT_ZEROED_GRAIN, false)) { 1889 zeroed_grain = true; 1890 } 1891 1892 if (!adapter_type) { 1893 adapter_type = g_strdup("ide"); 1894 } else if (strcmp(adapter_type, "ide") && 1895 strcmp(adapter_type, "buslogic") && 1896 strcmp(adapter_type, "lsilogic") && 1897 strcmp(adapter_type, "legacyESX")) { 1898 error_setg(errp, "Unknown adapter type: '%s'", adapter_type); 1899 ret = -EINVAL; 1900 goto exit; 1901 } 1902 if (strcmp(adapter_type, "ide") != 0) { 1903 /* that's the number of heads with which vmware operates when 1904 creating, exporting, etc. vmdk files with a non-ide adapter type */ 1905 number_heads = 255; 1906 } 1907 if (!fmt) { 1908 /* Default format to monolithicSparse */ 1909 fmt = g_strdup("monolithicSparse"); 1910 } else if (strcmp(fmt, "monolithicFlat") && 1911 strcmp(fmt, "monolithicSparse") && 1912 strcmp(fmt, "twoGbMaxExtentSparse") && 1913 strcmp(fmt, "twoGbMaxExtentFlat") && 1914 strcmp(fmt, "streamOptimized")) { 1915 error_setg(errp, "Unknown subformat: '%s'", fmt); 1916 ret = -EINVAL; 1917 goto exit; 1918 } 1919 split = !(strcmp(fmt, "twoGbMaxExtentFlat") && 1920 strcmp(fmt, "twoGbMaxExtentSparse")); 1921 flat = !(strcmp(fmt, "monolithicFlat") && 1922 strcmp(fmt, "twoGbMaxExtentFlat")); 1923 compress = !strcmp(fmt, "streamOptimized"); 1924 if (flat) { 1925 desc_extent_line = "RW %" PRId64 " FLAT \"%s\" 0\n"; 1926 } else { 1927 desc_extent_line = "RW %" PRId64 " SPARSE \"%s\"\n"; 1928 } 1929 if (flat && backing_file) { 1930 error_setg(errp, "Flat image can't have backing file"); 1931 ret = -ENOTSUP; 1932 goto exit; 1933 } 1934 if (flat && zeroed_grain) { 1935 error_setg(errp, "Flat image can't enable zeroed grain"); 1936 ret = -ENOTSUP; 1937 goto exit; 1938 } 1939 if (backing_file) { 1940 BlockBackend *blk; 1941 char *full_backing = g_new0(char, PATH_MAX); 1942 bdrv_get_full_backing_filename_from_filename(filename, backing_file, 1943 full_backing, PATH_MAX, 1944 &local_err); 1945 if (local_err) { 1946 g_free(full_backing); 1947 error_propagate(errp, local_err); 1948 ret = -ENOENT; 1949 goto exit; 1950 } 1951 1952 blk = blk_new_open("backing", full_backing, NULL, NULL, 1953 BDRV_O_NO_BACKING | BDRV_O_CACHE_WB, errp); 1954 g_free(full_backing); 1955 if (blk == NULL) { 1956 ret = -EIO; 1957 goto exit; 1958 } 1959 if (strcmp(blk_bs(blk)->drv->format_name, "vmdk")) { 1960 blk_unref(blk); 1961 ret = -EINVAL; 1962 goto exit; 1963 } 1964 parent_cid = vmdk_read_cid(blk_bs(blk), 0); 1965 blk_unref(blk); 1966 snprintf(parent_desc_line, BUF_SIZE, 1967 "parentFileNameHint=\"%s\"", backing_file); 1968 } 1969 1970 /* Create extents */ 1971 filesize = total_size; 1972 while (filesize > 0) { 1973 int64_t size = filesize; 1974 1975 if (split && size > split_size) { 1976 size = split_size; 1977 } 1978 if (split) { 1979 snprintf(desc_filename, PATH_MAX, "%s-%c%03d%s", 1980 prefix, flat ? 'f' : 's', ++idx, postfix); 1981 } else if (flat) { 1982 snprintf(desc_filename, PATH_MAX, "%s-flat%s", prefix, postfix); 1983 } else { 1984 snprintf(desc_filename, PATH_MAX, "%s%s", prefix, postfix); 1985 } 1986 snprintf(ext_filename, PATH_MAX, "%s%s", path, desc_filename); 1987 1988 if (vmdk_create_extent(ext_filename, size, 1989 flat, compress, zeroed_grain, opts, errp)) { 1990 ret = -EINVAL; 1991 goto exit; 1992 } 1993 filesize -= size; 1994 1995 /* Format description line */ 1996 snprintf(desc_line, BUF_SIZE, 1997 desc_extent_line, size / BDRV_SECTOR_SIZE, desc_filename); 1998 g_string_append(ext_desc_lines, desc_line); 1999 } 2000 /* generate descriptor file */ 2001 desc = g_strdup_printf(desc_template, 2002 g_random_int(), 2003 parent_cid, 2004 fmt, 2005 parent_desc_line, 2006 ext_desc_lines->str, 2007 (flags & BLOCK_FLAG_COMPAT6 ? 6 : 4), 2008 total_size / 2009 (int64_t)(63 * number_heads * BDRV_SECTOR_SIZE), 2010 number_heads, 2011 adapter_type); 2012 desc_len = strlen(desc); 2013 /* the descriptor offset = 0x200 */ 2014 if (!split && !flat) { 2015 desc_offset = 0x200; 2016 } else { 2017 ret = bdrv_create_file(filename, opts, &local_err); 2018 if (ret < 0) { 2019 error_propagate(errp, local_err); 2020 goto exit; 2021 } 2022 } 2023 2024 new_blk = blk_new_open("descriptor", filename, NULL, NULL, 2025 BDRV_O_RDWR | BDRV_O_CACHE_WB | BDRV_O_PROTOCOL, 2026 &local_err); 2027 if (new_blk == NULL) { 2028 error_propagate(errp, local_err); 2029 ret = -EIO; 2030 goto exit; 2031 } 2032 2033 blk_set_allow_write_beyond_eof(new_blk, true); 2034 2035 ret = blk_pwrite(new_blk, desc_offset, desc, desc_len); 2036 if (ret < 0) { 2037 error_setg_errno(errp, -ret, "Could not write description"); 2038 goto exit; 2039 } 2040 /* bdrv_pwrite write padding zeros to align to sector, we don't need that 2041 * for description file */ 2042 if (desc_offset == 0) { 2043 ret = blk_truncate(new_blk, desc_len); 2044 if (ret < 0) { 2045 error_setg_errno(errp, -ret, "Could not truncate file"); 2046 } 2047 } 2048 exit: 2049 if (new_blk) { 2050 blk_unref(new_blk); 2051 } 2052 g_free(adapter_type); 2053 g_free(backing_file); 2054 g_free(fmt); 2055 g_free(desc); 2056 g_free(path); 2057 g_free(prefix); 2058 g_free(postfix); 2059 g_free(desc_line); 2060 g_free(ext_filename); 2061 g_free(desc_filename); 2062 g_free(parent_desc_line); 2063 g_string_free(ext_desc_lines, true); 2064 return ret; 2065 } 2066 2067 static void vmdk_close(BlockDriverState *bs) 2068 { 2069 BDRVVmdkState *s = bs->opaque; 2070 2071 vmdk_free_extents(bs); 2072 g_free(s->create_type); 2073 2074 migrate_del_blocker(s->migration_blocker); 2075 error_free(s->migration_blocker); 2076 } 2077 2078 static coroutine_fn int vmdk_co_flush(BlockDriverState *bs) 2079 { 2080 BDRVVmdkState *s = bs->opaque; 2081 int i, err; 2082 int ret = 0; 2083 2084 for (i = 0; i < s->num_extents; i++) { 2085 err = bdrv_co_flush(s->extents[i].file->bs); 2086 if (err < 0) { 2087 ret = err; 2088 } 2089 } 2090 return ret; 2091 } 2092 2093 static int64_t vmdk_get_allocated_file_size(BlockDriverState *bs) 2094 { 2095 int i; 2096 int64_t ret = 0; 2097 int64_t r; 2098 BDRVVmdkState *s = bs->opaque; 2099 2100 ret = bdrv_get_allocated_file_size(bs->file->bs); 2101 if (ret < 0) { 2102 return ret; 2103 } 2104 for (i = 0; i < s->num_extents; i++) { 2105 if (s->extents[i].file == bs->file) { 2106 continue; 2107 } 2108 r = bdrv_get_allocated_file_size(s->extents[i].file->bs); 2109 if (r < 0) { 2110 return r; 2111 } 2112 ret += r; 2113 } 2114 return ret; 2115 } 2116 2117 static int vmdk_has_zero_init(BlockDriverState *bs) 2118 { 2119 int i; 2120 BDRVVmdkState *s = bs->opaque; 2121 2122 /* If has a flat extent and its underlying storage doesn't have zero init, 2123 * return 0. */ 2124 for (i = 0; i < s->num_extents; i++) { 2125 if (s->extents[i].flat) { 2126 if (!bdrv_has_zero_init(s->extents[i].file->bs)) { 2127 return 0; 2128 } 2129 } 2130 } 2131 return 1; 2132 } 2133 2134 static ImageInfo *vmdk_get_extent_info(VmdkExtent *extent) 2135 { 2136 ImageInfo *info = g_new0(ImageInfo, 1); 2137 2138 *info = (ImageInfo){ 2139 .filename = g_strdup(extent->file->bs->filename), 2140 .format = g_strdup(extent->type), 2141 .virtual_size = extent->sectors * BDRV_SECTOR_SIZE, 2142 .compressed = extent->compressed, 2143 .has_compressed = extent->compressed, 2144 .cluster_size = extent->cluster_sectors * BDRV_SECTOR_SIZE, 2145 .has_cluster_size = !extent->flat, 2146 }; 2147 2148 return info; 2149 } 2150 2151 static int vmdk_check(BlockDriverState *bs, BdrvCheckResult *result, 2152 BdrvCheckMode fix) 2153 { 2154 BDRVVmdkState *s = bs->opaque; 2155 VmdkExtent *extent = NULL; 2156 int64_t sector_num = 0; 2157 int64_t total_sectors = bdrv_nb_sectors(bs); 2158 int ret; 2159 uint64_t cluster_offset; 2160 2161 if (fix) { 2162 return -ENOTSUP; 2163 } 2164 2165 for (;;) { 2166 if (sector_num >= total_sectors) { 2167 return 0; 2168 } 2169 extent = find_extent(s, sector_num, extent); 2170 if (!extent) { 2171 fprintf(stderr, 2172 "ERROR: could not find extent for sector %" PRId64 "\n", 2173 sector_num); 2174 break; 2175 } 2176 ret = get_cluster_offset(bs, extent, NULL, 2177 sector_num << BDRV_SECTOR_BITS, 2178 false, &cluster_offset, 0, 0); 2179 if (ret == VMDK_ERROR) { 2180 fprintf(stderr, 2181 "ERROR: could not get cluster_offset for sector %" 2182 PRId64 "\n", sector_num); 2183 break; 2184 } 2185 if (ret == VMDK_OK && 2186 cluster_offset >= bdrv_getlength(extent->file->bs)) 2187 { 2188 fprintf(stderr, 2189 "ERROR: cluster offset for sector %" 2190 PRId64 " points after EOF\n", sector_num); 2191 break; 2192 } 2193 sector_num += extent->cluster_sectors; 2194 } 2195 2196 result->corruptions++; 2197 return 0; 2198 } 2199 2200 static ImageInfoSpecific *vmdk_get_specific_info(BlockDriverState *bs) 2201 { 2202 int i; 2203 BDRVVmdkState *s = bs->opaque; 2204 ImageInfoSpecific *spec_info = g_new0(ImageInfoSpecific, 1); 2205 ImageInfoList **next; 2206 2207 *spec_info = (ImageInfoSpecific){ 2208 .type = IMAGE_INFO_SPECIFIC_KIND_VMDK, 2209 { 2210 .vmdk = g_new0(ImageInfoSpecificVmdk, 1), 2211 }, 2212 }; 2213 2214 *spec_info->u.vmdk = (ImageInfoSpecificVmdk) { 2215 .create_type = g_strdup(s->create_type), 2216 .cid = s->cid, 2217 .parent_cid = s->parent_cid, 2218 }; 2219 2220 next = &spec_info->u.vmdk->extents; 2221 for (i = 0; i < s->num_extents; i++) { 2222 *next = g_new0(ImageInfoList, 1); 2223 (*next)->value = vmdk_get_extent_info(&s->extents[i]); 2224 (*next)->next = NULL; 2225 next = &(*next)->next; 2226 } 2227 2228 return spec_info; 2229 } 2230 2231 static bool vmdk_extents_type_eq(const VmdkExtent *a, const VmdkExtent *b) 2232 { 2233 return a->flat == b->flat && 2234 a->compressed == b->compressed && 2235 (a->flat || a->cluster_sectors == b->cluster_sectors); 2236 } 2237 2238 static int vmdk_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) 2239 { 2240 int i; 2241 BDRVVmdkState *s = bs->opaque; 2242 assert(s->num_extents); 2243 2244 /* See if we have multiple extents but they have different cases */ 2245 for (i = 1; i < s->num_extents; i++) { 2246 if (!vmdk_extents_type_eq(&s->extents[0], &s->extents[i])) { 2247 return -ENOTSUP; 2248 } 2249 } 2250 bdi->needs_compressed_writes = s->extents[0].compressed; 2251 if (!s->extents[0].flat) { 2252 bdi->cluster_size = s->extents[0].cluster_sectors << BDRV_SECTOR_BITS; 2253 } 2254 return 0; 2255 } 2256 2257 static void vmdk_detach_aio_context(BlockDriverState *bs) 2258 { 2259 BDRVVmdkState *s = bs->opaque; 2260 int i; 2261 2262 for (i = 0; i < s->num_extents; i++) { 2263 bdrv_detach_aio_context(s->extents[i].file->bs); 2264 } 2265 } 2266 2267 static void vmdk_attach_aio_context(BlockDriverState *bs, 2268 AioContext *new_context) 2269 { 2270 BDRVVmdkState *s = bs->opaque; 2271 int i; 2272 2273 for (i = 0; i < s->num_extents; i++) { 2274 bdrv_attach_aio_context(s->extents[i].file->bs, new_context); 2275 } 2276 } 2277 2278 static QemuOptsList vmdk_create_opts = { 2279 .name = "vmdk-create-opts", 2280 .head = QTAILQ_HEAD_INITIALIZER(vmdk_create_opts.head), 2281 .desc = { 2282 { 2283 .name = BLOCK_OPT_SIZE, 2284 .type = QEMU_OPT_SIZE, 2285 .help = "Virtual disk size" 2286 }, 2287 { 2288 .name = BLOCK_OPT_ADAPTER_TYPE, 2289 .type = QEMU_OPT_STRING, 2290 .help = "Virtual adapter type, can be one of " 2291 "ide (default), lsilogic, buslogic or legacyESX" 2292 }, 2293 { 2294 .name = BLOCK_OPT_BACKING_FILE, 2295 .type = QEMU_OPT_STRING, 2296 .help = "File name of a base image" 2297 }, 2298 { 2299 .name = BLOCK_OPT_COMPAT6, 2300 .type = QEMU_OPT_BOOL, 2301 .help = "VMDK version 6 image", 2302 .def_value_str = "off" 2303 }, 2304 { 2305 .name = BLOCK_OPT_SUBFMT, 2306 .type = QEMU_OPT_STRING, 2307 .help = 2308 "VMDK flat extent format, can be one of " 2309 "{monolithicSparse (default) | monolithicFlat | twoGbMaxExtentSparse | twoGbMaxExtentFlat | streamOptimized} " 2310 }, 2311 { 2312 .name = BLOCK_OPT_ZEROED_GRAIN, 2313 .type = QEMU_OPT_BOOL, 2314 .help = "Enable efficient zero writes " 2315 "using the zeroed-grain GTE feature" 2316 }, 2317 { /* end of list */ } 2318 } 2319 }; 2320 2321 static BlockDriver bdrv_vmdk = { 2322 .format_name = "vmdk", 2323 .instance_size = sizeof(BDRVVmdkState), 2324 .bdrv_probe = vmdk_probe, 2325 .bdrv_open = vmdk_open, 2326 .bdrv_check = vmdk_check, 2327 .bdrv_reopen_prepare = vmdk_reopen_prepare, 2328 .bdrv_read = vmdk_co_read, 2329 .bdrv_write = vmdk_co_write, 2330 .bdrv_write_compressed = vmdk_write_compressed, 2331 .bdrv_co_write_zeroes = vmdk_co_write_zeroes, 2332 .bdrv_close = vmdk_close, 2333 .bdrv_create = vmdk_create, 2334 .bdrv_co_flush_to_disk = vmdk_co_flush, 2335 .bdrv_co_get_block_status = vmdk_co_get_block_status, 2336 .bdrv_get_allocated_file_size = vmdk_get_allocated_file_size, 2337 .bdrv_has_zero_init = vmdk_has_zero_init, 2338 .bdrv_get_specific_info = vmdk_get_specific_info, 2339 .bdrv_refresh_limits = vmdk_refresh_limits, 2340 .bdrv_get_info = vmdk_get_info, 2341 .bdrv_detach_aio_context = vmdk_detach_aio_context, 2342 .bdrv_attach_aio_context = vmdk_attach_aio_context, 2343 2344 .supports_backing = true, 2345 .create_opts = &vmdk_create_opts, 2346 }; 2347 2348 static void bdrv_vmdk_init(void) 2349 { 2350 bdrv_register(&bdrv_vmdk); 2351 } 2352 2353 block_init(bdrv_vmdk_init); 2354