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/qdict.h" 31 #include "qapi/qmp/qerror.h" 32 #include "qemu/error-report.h" 33 #include "qemu/module.h" 34 #include "qemu/option.h" 35 #include "qemu/bswap.h" 36 #include "qemu/memalign.h" 37 #include "migration/blocker.h" 38 #include "qemu/cutils.h" 39 #include <zlib.h> 40 41 #define VMDK3_MAGIC (('C' << 24) | ('O' << 16) | ('W' << 8) | 'D') 42 #define VMDK4_MAGIC (('K' << 24) | ('D' << 16) | ('M' << 8) | 'V') 43 #define VMDK4_COMPRESSION_DEFLATE 1 44 #define VMDK4_FLAG_NL_DETECT (1 << 0) 45 #define VMDK4_FLAG_RGD (1 << 1) 46 /* Zeroed-grain enable bit */ 47 #define VMDK4_FLAG_ZERO_GRAIN (1 << 2) 48 #define VMDK4_FLAG_COMPRESS (1 << 16) 49 #define VMDK4_FLAG_MARKER (1 << 17) 50 #define VMDK4_GD_AT_END 0xffffffffffffffffULL 51 52 #define VMDK_EXTENT_MAX_SECTORS (1ULL << 32) 53 54 #define VMDK_GTE_ZEROED 0x1 55 56 /* VMDK internal error codes */ 57 #define VMDK_OK 0 58 #define VMDK_ERROR (-1) 59 /* Cluster not allocated */ 60 #define VMDK_UNALLOC (-2) 61 #define VMDK_ZEROED (-3) 62 63 #define BLOCK_OPT_ZEROED_GRAIN "zeroed_grain" 64 #define BLOCK_OPT_TOOLSVERSION "toolsversion" 65 66 typedef struct { 67 uint32_t version; 68 uint32_t flags; 69 uint32_t disk_sectors; 70 uint32_t granularity; 71 uint32_t l1dir_offset; 72 uint32_t l1dir_size; 73 uint32_t file_sectors; 74 uint32_t cylinders; 75 uint32_t heads; 76 uint32_t sectors_per_track; 77 } QEMU_PACKED VMDK3Header; 78 79 typedef struct { 80 uint32_t version; 81 uint32_t flags; 82 uint64_t capacity; 83 uint64_t granularity; 84 uint64_t desc_offset; 85 uint64_t desc_size; 86 /* Number of GrainTableEntries per GrainTable */ 87 uint32_t num_gtes_per_gt; 88 uint64_t rgd_offset; 89 uint64_t gd_offset; 90 uint64_t grain_offset; 91 char filler[1]; 92 char check_bytes[4]; 93 uint16_t compressAlgorithm; 94 } QEMU_PACKED VMDK4Header; 95 96 typedef struct VMDKSESparseConstHeader { 97 uint64_t magic; 98 uint64_t version; 99 uint64_t capacity; 100 uint64_t grain_size; 101 uint64_t grain_table_size; 102 uint64_t flags; 103 uint64_t reserved1; 104 uint64_t reserved2; 105 uint64_t reserved3; 106 uint64_t reserved4; 107 uint64_t volatile_header_offset; 108 uint64_t volatile_header_size; 109 uint64_t journal_header_offset; 110 uint64_t journal_header_size; 111 uint64_t journal_offset; 112 uint64_t journal_size; 113 uint64_t grain_dir_offset; 114 uint64_t grain_dir_size; 115 uint64_t grain_tables_offset; 116 uint64_t grain_tables_size; 117 uint64_t free_bitmap_offset; 118 uint64_t free_bitmap_size; 119 uint64_t backmap_offset; 120 uint64_t backmap_size; 121 uint64_t grains_offset; 122 uint64_t grains_size; 123 uint8_t pad[304]; 124 } QEMU_PACKED VMDKSESparseConstHeader; 125 126 typedef struct VMDKSESparseVolatileHeader { 127 uint64_t magic; 128 uint64_t free_gt_number; 129 uint64_t next_txn_seq_number; 130 uint64_t replay_journal; 131 uint8_t pad[480]; 132 } QEMU_PACKED VMDKSESparseVolatileHeader; 133 134 #define L2_CACHE_SIZE 16 135 136 typedef struct VmdkExtent { 137 BdrvChild *file; 138 bool flat; 139 bool compressed; 140 bool has_marker; 141 bool has_zero_grain; 142 bool sesparse; 143 uint64_t sesparse_l2_tables_offset; 144 uint64_t sesparse_clusters_offset; 145 int32_t entry_size; 146 int version; 147 int64_t sectors; 148 int64_t end_sector; 149 int64_t flat_start_offset; 150 int64_t l1_table_offset; 151 int64_t l1_backup_table_offset; 152 void *l1_table; 153 uint32_t *l1_backup_table; 154 unsigned int l1_size; 155 uint32_t l1_entry_sectors; 156 157 unsigned int l2_size; 158 void *l2_cache; 159 uint32_t l2_cache_offsets[L2_CACHE_SIZE]; 160 uint32_t l2_cache_counts[L2_CACHE_SIZE]; 161 162 int64_t cluster_sectors; 163 int64_t next_cluster_sector; 164 char *type; 165 } VmdkExtent; 166 167 typedef struct BDRVVmdkState { 168 CoMutex lock; 169 uint64_t desc_offset; 170 bool cid_updated; 171 bool cid_checked; 172 uint32_t cid; 173 uint32_t parent_cid; 174 int num_extents; 175 /* Extent array with num_extents entries, ascend ordered by address */ 176 VmdkExtent *extents; 177 Error *migration_blocker; 178 char *create_type; 179 } BDRVVmdkState; 180 181 typedef struct BDRVVmdkReopenState { 182 bool *extents_using_bs_file; 183 } BDRVVmdkReopenState; 184 185 typedef struct VmdkMetaData { 186 unsigned int l1_index; 187 unsigned int l2_index; 188 unsigned int l2_offset; 189 bool new_allocation; 190 uint32_t *l2_cache_entry; 191 } VmdkMetaData; 192 193 typedef struct VmdkGrainMarker { 194 uint64_t lba; 195 uint32_t size; 196 uint8_t data[]; 197 } QEMU_PACKED VmdkGrainMarker; 198 199 enum { 200 MARKER_END_OF_STREAM = 0, 201 MARKER_GRAIN_TABLE = 1, 202 MARKER_GRAIN_DIRECTORY = 2, 203 MARKER_FOOTER = 3, 204 }; 205 206 static int vmdk_probe(const uint8_t *buf, int buf_size, const char *filename) 207 { 208 uint32_t magic; 209 210 if (buf_size < 4) { 211 return 0; 212 } 213 magic = be32_to_cpu(*(uint32_t *)buf); 214 if (magic == VMDK3_MAGIC || 215 magic == VMDK4_MAGIC) { 216 return 100; 217 } else { 218 const char *p = (const char *)buf; 219 const char *end = p + buf_size; 220 while (p < end) { 221 if (*p == '#') { 222 /* skip comment line */ 223 while (p < end && *p != '\n') { 224 p++; 225 } 226 p++; 227 continue; 228 } 229 if (*p == ' ') { 230 while (p < end && *p == ' ') { 231 p++; 232 } 233 /* skip '\r' if windows line endings used. */ 234 if (p < end && *p == '\r') { 235 p++; 236 } 237 /* only accept blank lines before 'version=' line */ 238 if (p == end || *p != '\n') { 239 return 0; 240 } 241 p++; 242 continue; 243 } 244 if (end - p >= strlen("version=X\n")) { 245 if (strncmp("version=1\n", p, strlen("version=1\n")) == 0 || 246 strncmp("version=2\n", p, strlen("version=2\n")) == 0 || 247 strncmp("version=3\n", p, strlen("version=3\n")) == 0) { 248 return 100; 249 } 250 } 251 if (end - p >= strlen("version=X\r\n")) { 252 if (strncmp("version=1\r\n", p, strlen("version=1\r\n")) == 0 || 253 strncmp("version=2\r\n", p, strlen("version=2\r\n")) == 0 || 254 strncmp("version=3\r\n", p, strlen("version=3\r\n")) == 0) { 255 return 100; 256 } 257 } 258 return 0; 259 } 260 return 0; 261 } 262 } 263 264 #define SECTOR_SIZE 512 265 #define DESC_SIZE (20 * SECTOR_SIZE) /* 20 sectors of 512 bytes each */ 266 #define BUF_SIZE 4096 267 #define HEADER_SIZE 512 /* first sector of 512 bytes */ 268 269 static void vmdk_free_extents(BlockDriverState *bs) 270 { 271 int i; 272 BDRVVmdkState *s = bs->opaque; 273 VmdkExtent *e; 274 275 bdrv_graph_wrlock(NULL); 276 for (i = 0; i < s->num_extents; i++) { 277 e = &s->extents[i]; 278 g_free(e->l1_table); 279 g_free(e->l2_cache); 280 g_free(e->l1_backup_table); 281 g_free(e->type); 282 if (e->file != bs->file) { 283 bdrv_unref_child(bs, e->file); 284 } 285 } 286 bdrv_graph_wrunlock(); 287 288 g_free(s->extents); 289 } 290 291 static void vmdk_free_last_extent(BlockDriverState *bs) 292 { 293 BDRVVmdkState *s = bs->opaque; 294 295 if (s->num_extents == 0) { 296 return; 297 } 298 s->num_extents--; 299 s->extents = g_renew(VmdkExtent, s->extents, s->num_extents); 300 } 301 302 /* Return -ve errno, or 0 on success and write CID into *pcid. */ 303 static int vmdk_read_cid(BlockDriverState *bs, int parent, uint32_t *pcid) 304 { 305 char *desc; 306 uint32_t cid; 307 const char *p_name, *cid_str; 308 size_t cid_str_size; 309 BDRVVmdkState *s = bs->opaque; 310 int ret; 311 312 desc = g_malloc0(DESC_SIZE); 313 ret = bdrv_pread(bs->file, s->desc_offset, DESC_SIZE, desc, 0); 314 if (ret < 0) { 315 goto out; 316 } 317 318 if (parent) { 319 cid_str = "parentCID"; 320 cid_str_size = sizeof("parentCID"); 321 } else { 322 cid_str = "CID"; 323 cid_str_size = sizeof("CID"); 324 } 325 326 desc[DESC_SIZE - 1] = '\0'; 327 p_name = strstr(desc, cid_str); 328 if (p_name == NULL) { 329 ret = -EINVAL; 330 goto out; 331 } 332 p_name += cid_str_size; 333 if (sscanf(p_name, "%" SCNx32, &cid) != 1) { 334 ret = -EINVAL; 335 goto out; 336 } 337 *pcid = cid; 338 ret = 0; 339 340 out: 341 g_free(desc); 342 return ret; 343 } 344 345 static int coroutine_fn GRAPH_RDLOCK 346 vmdk_write_cid(BlockDriverState *bs, uint32_t cid) 347 { 348 char *desc, *tmp_desc; 349 char *p_name, *tmp_str; 350 BDRVVmdkState *s = bs->opaque; 351 int ret = 0; 352 353 desc = g_malloc0(DESC_SIZE); 354 tmp_desc = g_malloc0(DESC_SIZE); 355 ret = bdrv_co_pread(bs->file, s->desc_offset, DESC_SIZE, desc, 0); 356 if (ret < 0) { 357 goto out; 358 } 359 360 desc[DESC_SIZE - 1] = '\0'; 361 tmp_str = strstr(desc, "parentCID"); 362 if (tmp_str == NULL) { 363 ret = -EINVAL; 364 goto out; 365 } 366 367 pstrcpy(tmp_desc, DESC_SIZE, tmp_str); 368 p_name = strstr(desc, "CID"); 369 if (p_name != NULL) { 370 p_name += sizeof("CID"); 371 snprintf(p_name, DESC_SIZE - (p_name - desc), "%" PRIx32 "\n", cid); 372 pstrcat(desc, DESC_SIZE, tmp_desc); 373 } 374 375 ret = bdrv_co_pwrite_sync(bs->file, s->desc_offset, DESC_SIZE, desc, 0); 376 377 out: 378 g_free(desc); 379 g_free(tmp_desc); 380 return ret; 381 } 382 383 static int coroutine_fn vmdk_is_cid_valid(BlockDriverState *bs) 384 { 385 BDRVVmdkState *s = bs->opaque; 386 uint32_t cur_pcid; 387 388 if (!s->cid_checked && bs->backing) { 389 BlockDriverState *p_bs = bs->backing->bs; 390 391 if (strcmp(p_bs->drv->format_name, "vmdk")) { 392 /* Backing file is not in vmdk format, so it does not have 393 * a CID, which makes the overlay's parent CID invalid */ 394 return 0; 395 } 396 397 if (vmdk_read_cid(p_bs, 0, &cur_pcid) != 0) { 398 /* read failure: report as not valid */ 399 return 0; 400 } 401 if (s->parent_cid != cur_pcid) { 402 /* CID not valid */ 403 return 0; 404 } 405 } 406 s->cid_checked = true; 407 /* CID valid */ 408 return 1; 409 } 410 411 static int vmdk_reopen_prepare(BDRVReopenState *state, 412 BlockReopenQueue *queue, Error **errp) 413 { 414 BDRVVmdkState *s; 415 BDRVVmdkReopenState *rs; 416 int i; 417 418 assert(state != NULL); 419 assert(state->bs != NULL); 420 assert(state->opaque == NULL); 421 422 s = state->bs->opaque; 423 424 rs = g_new0(BDRVVmdkReopenState, 1); 425 state->opaque = rs; 426 427 /* 428 * Check whether there are any extents stored in bs->file; if bs->file 429 * changes, we will need to update their .file pointers to follow suit 430 */ 431 rs->extents_using_bs_file = g_new(bool, s->num_extents); 432 for (i = 0; i < s->num_extents; i++) { 433 rs->extents_using_bs_file[i] = s->extents[i].file == state->bs->file; 434 } 435 436 return 0; 437 } 438 439 static void vmdk_reopen_clean(BDRVReopenState *state) 440 { 441 BDRVVmdkReopenState *rs = state->opaque; 442 443 g_free(rs->extents_using_bs_file); 444 g_free(rs); 445 state->opaque = NULL; 446 } 447 448 static void vmdk_reopen_commit(BDRVReopenState *state) 449 { 450 BDRVVmdkState *s = state->bs->opaque; 451 BDRVVmdkReopenState *rs = state->opaque; 452 int i; 453 454 for (i = 0; i < s->num_extents; i++) { 455 if (rs->extents_using_bs_file[i]) { 456 s->extents[i].file = state->bs->file; 457 } 458 } 459 460 vmdk_reopen_clean(state); 461 } 462 463 static void vmdk_reopen_abort(BDRVReopenState *state) 464 { 465 vmdk_reopen_clean(state); 466 } 467 468 static int vmdk_parent_open(BlockDriverState *bs) 469 { 470 char *p_name; 471 char *desc; 472 BDRVVmdkState *s = bs->opaque; 473 int ret; 474 475 desc = g_malloc0(DESC_SIZE + 1); 476 ret = bdrv_pread(bs->file, s->desc_offset, DESC_SIZE, desc, 0); 477 if (ret < 0) { 478 goto out; 479 } 480 481 p_name = strstr(desc, "parentFileNameHint"); 482 if (p_name != NULL) { 483 char *end_name; 484 485 p_name += sizeof("parentFileNameHint") + 1; 486 end_name = strchr(p_name, '\"'); 487 if (end_name == NULL) { 488 ret = -EINVAL; 489 goto out; 490 } 491 if ((end_name - p_name) > sizeof(bs->auto_backing_file) - 1) { 492 ret = -EINVAL; 493 goto out; 494 } 495 496 pstrcpy(bs->auto_backing_file, end_name - p_name + 1, p_name); 497 pstrcpy(bs->backing_file, sizeof(bs->backing_file), 498 bs->auto_backing_file); 499 pstrcpy(bs->backing_format, sizeof(bs->backing_format), 500 "vmdk"); 501 } 502 503 out: 504 g_free(desc); 505 return ret; 506 } 507 508 /* Create and append extent to the extent array. Return the added VmdkExtent 509 * address. return NULL if allocation failed. */ 510 static int vmdk_add_extent(BlockDriverState *bs, 511 BdrvChild *file, bool flat, int64_t sectors, 512 int64_t l1_offset, int64_t l1_backup_offset, 513 uint32_t l1_size, 514 int l2_size, uint64_t cluster_sectors, 515 VmdkExtent **new_extent, 516 Error **errp) 517 { 518 VmdkExtent *extent; 519 BDRVVmdkState *s = bs->opaque; 520 int64_t nb_sectors; 521 522 if (cluster_sectors > 0x200000) { 523 /* 0x200000 * 512Bytes = 1GB for one cluster is unrealistic */ 524 error_setg(errp, "Invalid granularity, image may be corrupt"); 525 return -EFBIG; 526 } 527 if (l1_size > 32 * 1024 * 1024) { 528 /* 529 * Although with big capacity and small l1_entry_sectors, we can get a 530 * big l1_size, we don't want unbounded value to allocate the table. 531 * Limit it to 32M, which is enough to store: 532 * 8TB - for both VMDK3 & VMDK4 with 533 * minimal cluster size: 512B 534 * minimal L2 table size: 512 entries 535 * 8 TB is still more than the maximal value supported for 536 * VMDK3 & VMDK4 which is 2TB. 537 * 64TB - for "ESXi seSparse Extent" 538 * minimal cluster size: 512B (default is 4KB) 539 * L2 table size: 4096 entries (const). 540 * 64TB is more than the maximal value supported for 541 * seSparse VMDKs (which is slightly less than 64TB) 542 */ 543 error_setg(errp, "L1 size too big"); 544 return -EFBIG; 545 } 546 547 nb_sectors = bdrv_nb_sectors(file->bs); 548 if (nb_sectors < 0) { 549 return nb_sectors; 550 } 551 552 s->extents = g_renew(VmdkExtent, s->extents, s->num_extents + 1); 553 extent = &s->extents[s->num_extents]; 554 s->num_extents++; 555 556 memset(extent, 0, sizeof(VmdkExtent)); 557 extent->file = file; 558 extent->flat = flat; 559 extent->sectors = sectors; 560 extent->l1_table_offset = l1_offset; 561 extent->l1_backup_table_offset = l1_backup_offset; 562 extent->l1_size = l1_size; 563 extent->l1_entry_sectors = l2_size * cluster_sectors; 564 extent->l2_size = l2_size; 565 extent->cluster_sectors = flat ? sectors : cluster_sectors; 566 extent->next_cluster_sector = ROUND_UP(nb_sectors, cluster_sectors); 567 extent->entry_size = sizeof(uint32_t); 568 569 if (s->num_extents > 1) { 570 extent->end_sector = (*(extent - 1)).end_sector + extent->sectors; 571 } else { 572 extent->end_sector = extent->sectors; 573 } 574 bs->total_sectors = extent->end_sector; 575 if (new_extent) { 576 *new_extent = extent; 577 } 578 return 0; 579 } 580 581 static int GRAPH_RDLOCK 582 vmdk_init_tables(BlockDriverState *bs, VmdkExtent *extent, Error **errp) 583 { 584 int ret; 585 size_t l1_size; 586 int i; 587 588 /* read the L1 table */ 589 l1_size = extent->l1_size * extent->entry_size; 590 extent->l1_table = g_try_malloc(l1_size); 591 if (l1_size && extent->l1_table == NULL) { 592 return -ENOMEM; 593 } 594 595 ret = bdrv_pread(extent->file, extent->l1_table_offset, l1_size, 596 extent->l1_table, 0); 597 if (ret < 0) { 598 bdrv_refresh_filename(extent->file->bs); 599 error_setg_errno(errp, -ret, 600 "Could not read l1 table from extent '%s'", 601 extent->file->bs->filename); 602 goto fail_l1; 603 } 604 for (i = 0; i < extent->l1_size; i++) { 605 if (extent->entry_size == sizeof(uint64_t)) { 606 le64_to_cpus((uint64_t *)extent->l1_table + i); 607 } else { 608 assert(extent->entry_size == sizeof(uint32_t)); 609 le32_to_cpus((uint32_t *)extent->l1_table + i); 610 } 611 } 612 613 if (extent->l1_backup_table_offset) { 614 assert(!extent->sesparse); 615 extent->l1_backup_table = g_try_malloc(l1_size); 616 if (l1_size && extent->l1_backup_table == NULL) { 617 ret = -ENOMEM; 618 goto fail_l1; 619 } 620 ret = bdrv_pread(extent->file, extent->l1_backup_table_offset, 621 l1_size, extent->l1_backup_table, 0); 622 if (ret < 0) { 623 bdrv_refresh_filename(extent->file->bs); 624 error_setg_errno(errp, -ret, 625 "Could not read l1 backup table from extent '%s'", 626 extent->file->bs->filename); 627 goto fail_l1b; 628 } 629 for (i = 0; i < extent->l1_size; i++) { 630 le32_to_cpus(&extent->l1_backup_table[i]); 631 } 632 } 633 634 extent->l2_cache = 635 g_malloc(extent->entry_size * extent->l2_size * L2_CACHE_SIZE); 636 return 0; 637 fail_l1b: 638 g_free(extent->l1_backup_table); 639 fail_l1: 640 g_free(extent->l1_table); 641 return ret; 642 } 643 644 static int GRAPH_RDLOCK 645 vmdk_open_vmfs_sparse(BlockDriverState *bs, BdrvChild *file, int flags, 646 Error **errp) 647 { 648 int ret; 649 uint32_t magic; 650 VMDK3Header header; 651 VmdkExtent *extent = NULL; 652 653 ret = bdrv_pread(file, sizeof(magic), sizeof(header), &header, 0); 654 if (ret < 0) { 655 bdrv_refresh_filename(file->bs); 656 error_setg_errno(errp, -ret, 657 "Could not read header from file '%s'", 658 file->bs->filename); 659 return ret; 660 } 661 ret = vmdk_add_extent(bs, file, false, 662 le32_to_cpu(header.disk_sectors), 663 (int64_t)le32_to_cpu(header.l1dir_offset) << 9, 664 0, 665 le32_to_cpu(header.l1dir_size), 666 4096, 667 le32_to_cpu(header.granularity), 668 &extent, 669 errp); 670 if (ret < 0) { 671 return ret; 672 } 673 ret = vmdk_init_tables(bs, extent, errp); 674 if (ret) { 675 /* free extent allocated by vmdk_add_extent */ 676 vmdk_free_last_extent(bs); 677 } 678 return ret; 679 } 680 681 #define SESPARSE_CONST_HEADER_MAGIC UINT64_C(0x00000000cafebabe) 682 #define SESPARSE_VOLATILE_HEADER_MAGIC UINT64_C(0x00000000cafecafe) 683 684 /* Strict checks - format not officially documented */ 685 static int check_se_sparse_const_header(VMDKSESparseConstHeader *header, 686 Error **errp) 687 { 688 header->magic = le64_to_cpu(header->magic); 689 header->version = le64_to_cpu(header->version); 690 header->grain_size = le64_to_cpu(header->grain_size); 691 header->grain_table_size = le64_to_cpu(header->grain_table_size); 692 header->flags = le64_to_cpu(header->flags); 693 header->reserved1 = le64_to_cpu(header->reserved1); 694 header->reserved2 = le64_to_cpu(header->reserved2); 695 header->reserved3 = le64_to_cpu(header->reserved3); 696 header->reserved4 = le64_to_cpu(header->reserved4); 697 698 header->volatile_header_offset = 699 le64_to_cpu(header->volatile_header_offset); 700 header->volatile_header_size = le64_to_cpu(header->volatile_header_size); 701 702 header->journal_header_offset = le64_to_cpu(header->journal_header_offset); 703 header->journal_header_size = le64_to_cpu(header->journal_header_size); 704 705 header->journal_offset = le64_to_cpu(header->journal_offset); 706 header->journal_size = le64_to_cpu(header->journal_size); 707 708 header->grain_dir_offset = le64_to_cpu(header->grain_dir_offset); 709 header->grain_dir_size = le64_to_cpu(header->grain_dir_size); 710 711 header->grain_tables_offset = le64_to_cpu(header->grain_tables_offset); 712 header->grain_tables_size = le64_to_cpu(header->grain_tables_size); 713 714 header->free_bitmap_offset = le64_to_cpu(header->free_bitmap_offset); 715 header->free_bitmap_size = le64_to_cpu(header->free_bitmap_size); 716 717 header->backmap_offset = le64_to_cpu(header->backmap_offset); 718 header->backmap_size = le64_to_cpu(header->backmap_size); 719 720 header->grains_offset = le64_to_cpu(header->grains_offset); 721 header->grains_size = le64_to_cpu(header->grains_size); 722 723 if (header->magic != SESPARSE_CONST_HEADER_MAGIC) { 724 error_setg(errp, "Bad const header magic: 0x%016" PRIx64, 725 header->magic); 726 return -EINVAL; 727 } 728 729 if (header->version != 0x0000000200000001) { 730 error_setg(errp, "Unsupported version: 0x%016" PRIx64, 731 header->version); 732 return -ENOTSUP; 733 } 734 735 if (header->grain_size != 8) { 736 error_setg(errp, "Unsupported grain size: %" PRIu64, 737 header->grain_size); 738 return -ENOTSUP; 739 } 740 741 if (header->grain_table_size != 64) { 742 error_setg(errp, "Unsupported grain table size: %" PRIu64, 743 header->grain_table_size); 744 return -ENOTSUP; 745 } 746 747 if (header->flags != 0) { 748 error_setg(errp, "Unsupported flags: 0x%016" PRIx64, 749 header->flags); 750 return -ENOTSUP; 751 } 752 753 if (header->reserved1 != 0 || header->reserved2 != 0 || 754 header->reserved3 != 0 || header->reserved4 != 0) { 755 error_setg(errp, "Unsupported reserved bits:" 756 " 0x%016" PRIx64 " 0x%016" PRIx64 757 " 0x%016" PRIx64 " 0x%016" PRIx64, 758 header->reserved1, header->reserved2, 759 header->reserved3, header->reserved4); 760 return -ENOTSUP; 761 } 762 763 /* check that padding is 0 */ 764 if (!buffer_is_zero(header->pad, sizeof(header->pad))) { 765 error_setg(errp, "Unsupported non-zero const header padding"); 766 return -ENOTSUP; 767 } 768 769 return 0; 770 } 771 772 static int check_se_sparse_volatile_header(VMDKSESparseVolatileHeader *header, 773 Error **errp) 774 { 775 header->magic = le64_to_cpu(header->magic); 776 header->free_gt_number = le64_to_cpu(header->free_gt_number); 777 header->next_txn_seq_number = le64_to_cpu(header->next_txn_seq_number); 778 header->replay_journal = le64_to_cpu(header->replay_journal); 779 780 if (header->magic != SESPARSE_VOLATILE_HEADER_MAGIC) { 781 error_setg(errp, "Bad volatile header magic: 0x%016" PRIx64, 782 header->magic); 783 return -EINVAL; 784 } 785 786 if (header->replay_journal) { 787 error_setg(errp, "Image is dirty, Replaying journal not supported"); 788 return -ENOTSUP; 789 } 790 791 /* check that padding is 0 */ 792 if (!buffer_is_zero(header->pad, sizeof(header->pad))) { 793 error_setg(errp, "Unsupported non-zero volatile header padding"); 794 return -ENOTSUP; 795 } 796 797 return 0; 798 } 799 800 static int GRAPH_RDLOCK 801 vmdk_open_se_sparse(BlockDriverState *bs, BdrvChild *file, int flags, 802 Error **errp) 803 { 804 int ret; 805 VMDKSESparseConstHeader const_header; 806 VMDKSESparseVolatileHeader volatile_header; 807 VmdkExtent *extent = NULL; 808 809 ret = bdrv_apply_auto_read_only(bs, 810 "No write support for seSparse images available", errp); 811 if (ret < 0) { 812 return ret; 813 } 814 815 assert(sizeof(const_header) == SECTOR_SIZE); 816 817 ret = bdrv_pread(file, 0, sizeof(const_header), &const_header, 0); 818 if (ret < 0) { 819 bdrv_refresh_filename(file->bs); 820 error_setg_errno(errp, -ret, 821 "Could not read const header from file '%s'", 822 file->bs->filename); 823 return ret; 824 } 825 826 /* check const header */ 827 ret = check_se_sparse_const_header(&const_header, errp); 828 if (ret < 0) { 829 return ret; 830 } 831 832 assert(sizeof(volatile_header) == SECTOR_SIZE); 833 834 ret = bdrv_pread(file, const_header.volatile_header_offset * SECTOR_SIZE, 835 sizeof(volatile_header), &volatile_header, 0); 836 if (ret < 0) { 837 bdrv_refresh_filename(file->bs); 838 error_setg_errno(errp, -ret, 839 "Could not read volatile header from file '%s'", 840 file->bs->filename); 841 return ret; 842 } 843 844 /* check volatile header */ 845 ret = check_se_sparse_volatile_header(&volatile_header, errp); 846 if (ret < 0) { 847 return ret; 848 } 849 850 ret = vmdk_add_extent(bs, file, false, 851 const_header.capacity, 852 const_header.grain_dir_offset * SECTOR_SIZE, 853 0, 854 const_header.grain_dir_size * 855 SECTOR_SIZE / sizeof(uint64_t), 856 const_header.grain_table_size * 857 SECTOR_SIZE / sizeof(uint64_t), 858 const_header.grain_size, 859 &extent, 860 errp); 861 if (ret < 0) { 862 return ret; 863 } 864 865 extent->sesparse = true; 866 extent->sesparse_l2_tables_offset = const_header.grain_tables_offset; 867 extent->sesparse_clusters_offset = const_header.grains_offset; 868 extent->entry_size = sizeof(uint64_t); 869 870 ret = vmdk_init_tables(bs, extent, errp); 871 if (ret) { 872 /* free extent allocated by vmdk_add_extent */ 873 vmdk_free_last_extent(bs); 874 } 875 876 return ret; 877 } 878 879 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf, 880 QDict *options, Error **errp); 881 882 static char *vmdk_read_desc(BdrvChild *file, uint64_t desc_offset, Error **errp) 883 { 884 int64_t size; 885 char *buf; 886 int ret; 887 888 size = bdrv_getlength(file->bs); 889 if (size < 0) { 890 error_setg_errno(errp, -size, "Could not access file"); 891 return NULL; 892 } 893 894 if (size < 4) { 895 /* Both descriptor file and sparse image must be much larger than 4 896 * bytes, also callers of vmdk_read_desc want to compare the first 4 897 * bytes with VMDK4_MAGIC, let's error out if less is read. */ 898 error_setg(errp, "File is too small, not a valid image"); 899 return NULL; 900 } 901 902 size = MIN(size, (1 << 20) - 1); /* avoid unbounded allocation */ 903 buf = g_malloc(size + 1); 904 905 ret = bdrv_pread(file, desc_offset, size, buf, 0); 906 if (ret < 0) { 907 error_setg_errno(errp, -ret, "Could not read from file"); 908 g_free(buf); 909 return NULL; 910 } 911 buf[size] = 0; 912 913 return buf; 914 } 915 916 static int GRAPH_RDLOCK 917 vmdk_open_vmdk4(BlockDriverState *bs, BdrvChild *file, int flags, 918 QDict *options, Error **errp) 919 { 920 int ret; 921 uint32_t magic; 922 uint32_t l1_size, l1_entry_sectors; 923 VMDK4Header header; 924 VmdkExtent *extent = NULL; 925 BDRVVmdkState *s = bs->opaque; 926 int64_t l1_backup_offset = 0; 927 bool compressed; 928 929 ret = bdrv_pread(file, sizeof(magic), sizeof(header), &header, 0); 930 if (ret < 0) { 931 bdrv_refresh_filename(file->bs); 932 error_setg_errno(errp, -ret, 933 "Could not read header from file '%s'", 934 file->bs->filename); 935 return -EINVAL; 936 } 937 if (header.capacity == 0) { 938 uint64_t desc_offset = le64_to_cpu(header.desc_offset); 939 if (desc_offset) { 940 char *buf = vmdk_read_desc(file, desc_offset << 9, errp); 941 if (!buf) { 942 return -EINVAL; 943 } 944 ret = vmdk_open_desc_file(bs, flags, buf, options, errp); 945 g_free(buf); 946 return ret; 947 } 948 } 949 950 if (!s->create_type) { 951 s->create_type = g_strdup("monolithicSparse"); 952 } 953 954 if (le64_to_cpu(header.gd_offset) == VMDK4_GD_AT_END) { 955 /* 956 * The footer takes precedence over the header, so read it in. The 957 * footer starts at offset -1024 from the end: One sector for the 958 * footer, and another one for the end-of-stream marker. 959 */ 960 struct { 961 struct { 962 uint64_t val; 963 uint32_t size; 964 uint32_t type; 965 uint8_t pad[512 - 16]; 966 } QEMU_PACKED footer_marker; 967 968 uint32_t magic; 969 VMDK4Header header; 970 uint8_t pad[512 - 4 - sizeof(VMDK4Header)]; 971 972 struct { 973 uint64_t val; 974 uint32_t size; 975 uint32_t type; 976 uint8_t pad[512 - 16]; 977 } QEMU_PACKED eos_marker; 978 } QEMU_PACKED footer; 979 980 ret = bdrv_pread(file, bs->file->bs->total_sectors * 512 - 1536, 981 sizeof(footer), &footer, 0); 982 if (ret < 0) { 983 error_setg_errno(errp, -ret, "Failed to read footer"); 984 return ret; 985 } 986 987 /* Some sanity checks for the footer */ 988 if (be32_to_cpu(footer.magic) != VMDK4_MAGIC || 989 le32_to_cpu(footer.footer_marker.size) != 0 || 990 le32_to_cpu(footer.footer_marker.type) != MARKER_FOOTER || 991 le64_to_cpu(footer.eos_marker.val) != 0 || 992 le32_to_cpu(footer.eos_marker.size) != 0 || 993 le32_to_cpu(footer.eos_marker.type) != MARKER_END_OF_STREAM) 994 { 995 error_setg(errp, "Invalid footer"); 996 return -EINVAL; 997 } 998 999 header = footer.header; 1000 } 1001 1002 compressed = 1003 le16_to_cpu(header.compressAlgorithm) == VMDK4_COMPRESSION_DEFLATE; 1004 if (le32_to_cpu(header.version) > 3) { 1005 error_setg(errp, "Unsupported VMDK version %" PRIu32, 1006 le32_to_cpu(header.version)); 1007 return -ENOTSUP; 1008 } else if (le32_to_cpu(header.version) == 3 && (flags & BDRV_O_RDWR) && 1009 !compressed) { 1010 /* VMware KB 2064959 explains that version 3 added support for 1011 * persistent changed block tracking (CBT), and backup software can 1012 * read it as version=1 if it doesn't care about the changed area 1013 * information. So we are safe to enable read only. */ 1014 error_setg(errp, "VMDK version 3 must be read only"); 1015 return -EINVAL; 1016 } 1017 1018 if (le32_to_cpu(header.num_gtes_per_gt) > 512) { 1019 error_setg(errp, "L2 table size too big"); 1020 return -EINVAL; 1021 } 1022 1023 l1_entry_sectors = le32_to_cpu(header.num_gtes_per_gt) 1024 * le64_to_cpu(header.granularity); 1025 if (l1_entry_sectors == 0) { 1026 error_setg(errp, "L1 entry size is invalid"); 1027 return -EINVAL; 1028 } 1029 l1_size = (le64_to_cpu(header.capacity) + l1_entry_sectors - 1) 1030 / l1_entry_sectors; 1031 if (le32_to_cpu(header.flags) & VMDK4_FLAG_RGD) { 1032 l1_backup_offset = le64_to_cpu(header.rgd_offset) << 9; 1033 } 1034 if (bdrv_nb_sectors(file->bs) < le64_to_cpu(header.grain_offset)) { 1035 error_setg(errp, "File truncated, expecting at least %" PRId64 " bytes", 1036 (int64_t)(le64_to_cpu(header.grain_offset) 1037 * BDRV_SECTOR_SIZE)); 1038 return -EINVAL; 1039 } 1040 1041 ret = vmdk_add_extent(bs, file, false, 1042 le64_to_cpu(header.capacity), 1043 le64_to_cpu(header.gd_offset) << 9, 1044 l1_backup_offset, 1045 l1_size, 1046 le32_to_cpu(header.num_gtes_per_gt), 1047 le64_to_cpu(header.granularity), 1048 &extent, 1049 errp); 1050 if (ret < 0) { 1051 return ret; 1052 } 1053 extent->compressed = 1054 le16_to_cpu(header.compressAlgorithm) == VMDK4_COMPRESSION_DEFLATE; 1055 if (extent->compressed) { 1056 g_free(s->create_type); 1057 s->create_type = g_strdup("streamOptimized"); 1058 } 1059 extent->has_marker = le32_to_cpu(header.flags) & VMDK4_FLAG_MARKER; 1060 extent->version = le32_to_cpu(header.version); 1061 extent->has_zero_grain = le32_to_cpu(header.flags) & VMDK4_FLAG_ZERO_GRAIN; 1062 ret = vmdk_init_tables(bs, extent, errp); 1063 if (ret) { 1064 /* free extent allocated by vmdk_add_extent */ 1065 vmdk_free_last_extent(bs); 1066 } 1067 return ret; 1068 } 1069 1070 /* find an option value out of descriptor file */ 1071 static int vmdk_parse_description(const char *desc, const char *opt_name, 1072 char *buf, int buf_size) 1073 { 1074 char *opt_pos, *opt_end; 1075 const char *end = desc + strlen(desc); 1076 1077 opt_pos = strstr(desc, opt_name); 1078 if (!opt_pos) { 1079 return VMDK_ERROR; 1080 } 1081 /* Skip "=\"" following opt_name */ 1082 opt_pos += strlen(opt_name) + 2; 1083 if (opt_pos >= end) { 1084 return VMDK_ERROR; 1085 } 1086 opt_end = opt_pos; 1087 while (opt_end < end && *opt_end != '"') { 1088 opt_end++; 1089 } 1090 if (opt_end == end || buf_size < opt_end - opt_pos + 1) { 1091 return VMDK_ERROR; 1092 } 1093 pstrcpy(buf, opt_end - opt_pos + 1, opt_pos); 1094 return VMDK_OK; 1095 } 1096 1097 /* Open an extent file and append to bs array */ 1098 static int GRAPH_RDLOCK 1099 vmdk_open_sparse(BlockDriverState *bs, BdrvChild *file, int flags, 1100 char *buf, QDict *options, Error **errp) 1101 { 1102 uint32_t magic; 1103 1104 magic = ldl_be_p(buf); 1105 switch (magic) { 1106 case VMDK3_MAGIC: 1107 return vmdk_open_vmfs_sparse(bs, file, flags, errp); 1108 case VMDK4_MAGIC: 1109 return vmdk_open_vmdk4(bs, file, flags, options, errp); 1110 default: 1111 error_setg(errp, "Image not in VMDK format"); 1112 return -EINVAL; 1113 } 1114 } 1115 1116 static const char *next_line(const char *s) 1117 { 1118 while (*s) { 1119 if (*s == '\n') { 1120 return s + 1; 1121 } 1122 s++; 1123 } 1124 return s; 1125 } 1126 1127 static int GRAPH_RDLOCK 1128 vmdk_parse_extents(const char *desc, BlockDriverState *bs, QDict *options, 1129 Error **errp) 1130 { 1131 int ret; 1132 int matches; 1133 char access[11]; 1134 char type[11]; 1135 char fname[512]; 1136 const char *p, *np; 1137 int64_t sectors = 0; 1138 int64_t flat_offset; 1139 char *desc_file_dir = NULL; 1140 char *extent_path; 1141 BdrvChild *extent_file; 1142 BdrvChildRole extent_role; 1143 BDRVVmdkState *s = bs->opaque; 1144 VmdkExtent *extent = NULL; 1145 char extent_opt_prefix[32]; 1146 Error *local_err = NULL; 1147 1148 GLOBAL_STATE_CODE(); 1149 1150 for (p = desc; *p; p = next_line(p)) { 1151 /* parse extent line in one of below formats: 1152 * 1153 * RW [size in sectors] FLAT "file-name.vmdk" OFFSET 1154 * RW [size in sectors] SPARSE "file-name.vmdk" 1155 * RW [size in sectors] VMFS "file-name.vmdk" 1156 * RW [size in sectors] VMFSSPARSE "file-name.vmdk" 1157 * RW [size in sectors] SESPARSE "file-name.vmdk" 1158 */ 1159 flat_offset = -1; 1160 matches = sscanf(p, "%10s %" SCNd64 " %10s \"%511[^\n\r\"]\" %" SCNd64, 1161 access, §ors, type, fname, &flat_offset); 1162 if (matches < 4 || strcmp(access, "RW")) { 1163 continue; 1164 } else if (!strcmp(type, "FLAT")) { 1165 if (matches != 5 || flat_offset < 0) { 1166 goto invalid; 1167 } 1168 } else if (!strcmp(type, "VMFS")) { 1169 if (matches == 4) { 1170 flat_offset = 0; 1171 } else { 1172 goto invalid; 1173 } 1174 } else if (matches != 4) { 1175 goto invalid; 1176 } 1177 1178 if (sectors <= 0 || 1179 (strcmp(type, "FLAT") && strcmp(type, "SPARSE") && 1180 strcmp(type, "VMFS") && strcmp(type, "VMFSSPARSE") && 1181 strcmp(type, "SESPARSE")) || 1182 (strcmp(access, "RW"))) { 1183 continue; 1184 } 1185 1186 if (path_is_absolute(fname)) { 1187 extent_path = g_strdup(fname); 1188 } else { 1189 if (!desc_file_dir) { 1190 desc_file_dir = bdrv_dirname(bs->file->bs, errp); 1191 if (!desc_file_dir) { 1192 bdrv_refresh_filename(bs->file->bs); 1193 error_prepend(errp, "Cannot use relative paths with VMDK " 1194 "descriptor file '%s': ", 1195 bs->file->bs->filename); 1196 ret = -EINVAL; 1197 goto out; 1198 } 1199 } 1200 1201 extent_path = g_strconcat(desc_file_dir, fname, NULL); 1202 } 1203 1204 ret = snprintf(extent_opt_prefix, 32, "extents.%d", s->num_extents); 1205 assert(ret < 32); 1206 1207 extent_role = BDRV_CHILD_DATA; 1208 if (strcmp(type, "FLAT") != 0 && strcmp(type, "VMFS") != 0) { 1209 /* non-flat extents have metadata */ 1210 extent_role |= BDRV_CHILD_METADATA; 1211 } 1212 1213 extent_file = bdrv_open_child(extent_path, options, extent_opt_prefix, 1214 bs, &child_of_bds, extent_role, false, 1215 &local_err); 1216 g_free(extent_path); 1217 if (!extent_file) { 1218 error_propagate(errp, local_err); 1219 ret = -EINVAL; 1220 goto out; 1221 } 1222 1223 /* save to extents array */ 1224 if (!strcmp(type, "FLAT") || !strcmp(type, "VMFS")) { 1225 /* FLAT extent */ 1226 1227 ret = vmdk_add_extent(bs, extent_file, true, sectors, 1228 0, 0, 0, 0, 0, &extent, errp); 1229 if (ret < 0) { 1230 bdrv_graph_rdunlock_main_loop(); 1231 bdrv_graph_wrlock(NULL); 1232 bdrv_unref_child(bs, extent_file); 1233 bdrv_graph_wrunlock(); 1234 bdrv_graph_rdlock_main_loop(); 1235 goto out; 1236 } 1237 extent->flat_start_offset = flat_offset << 9; 1238 } else if (!strcmp(type, "SPARSE") || !strcmp(type, "VMFSSPARSE")) { 1239 /* SPARSE extent and VMFSSPARSE extent are both "COWD" sparse file*/ 1240 char *buf = vmdk_read_desc(extent_file, 0, errp); 1241 if (!buf) { 1242 ret = -EINVAL; 1243 } else { 1244 ret = vmdk_open_sparse(bs, extent_file, bs->open_flags, buf, 1245 options, errp); 1246 } 1247 g_free(buf); 1248 if (ret) { 1249 bdrv_graph_rdunlock_main_loop(); 1250 bdrv_graph_wrlock(NULL); 1251 bdrv_unref_child(bs, extent_file); 1252 bdrv_graph_wrunlock(); 1253 bdrv_graph_rdlock_main_loop(); 1254 goto out; 1255 } 1256 extent = &s->extents[s->num_extents - 1]; 1257 } else if (!strcmp(type, "SESPARSE")) { 1258 ret = vmdk_open_se_sparse(bs, extent_file, bs->open_flags, errp); 1259 if (ret) { 1260 bdrv_graph_rdunlock_main_loop(); 1261 bdrv_graph_wrlock(NULL); 1262 bdrv_unref_child(bs, extent_file); 1263 bdrv_graph_wrunlock(); 1264 bdrv_graph_rdlock_main_loop(); 1265 goto out; 1266 } 1267 extent = &s->extents[s->num_extents - 1]; 1268 } else { 1269 error_setg(errp, "Unsupported extent type '%s'", type); 1270 bdrv_graph_rdunlock_main_loop(); 1271 bdrv_graph_wrlock(NULL); 1272 bdrv_unref_child(bs, extent_file); 1273 bdrv_graph_wrunlock(); 1274 bdrv_graph_rdlock_main_loop(); 1275 ret = -ENOTSUP; 1276 goto out; 1277 } 1278 extent->type = g_strdup(type); 1279 } 1280 1281 ret = 0; 1282 goto out; 1283 1284 invalid: 1285 np = next_line(p); 1286 assert(np != p); 1287 if (np[-1] == '\n') { 1288 np--; 1289 } 1290 error_setg(errp, "Invalid extent line: %.*s", (int)(np - p), p); 1291 ret = -EINVAL; 1292 1293 out: 1294 g_free(desc_file_dir); 1295 return ret; 1296 } 1297 1298 static int GRAPH_RDLOCK 1299 vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf, QDict *options, 1300 Error **errp) 1301 { 1302 int ret; 1303 char ct[128]; 1304 BDRVVmdkState *s = bs->opaque; 1305 1306 if (vmdk_parse_description(buf, "createType", ct, sizeof(ct))) { 1307 error_setg(errp, "invalid VMDK image descriptor"); 1308 ret = -EINVAL; 1309 goto exit; 1310 } 1311 if (strcmp(ct, "monolithicFlat") && 1312 strcmp(ct, "vmfs") && 1313 strcmp(ct, "vmfsSparse") && 1314 strcmp(ct, "seSparse") && 1315 strcmp(ct, "twoGbMaxExtentSparse") && 1316 strcmp(ct, "twoGbMaxExtentFlat")) { 1317 error_setg(errp, "Unsupported image type '%s'", ct); 1318 ret = -ENOTSUP; 1319 goto exit; 1320 } 1321 s->create_type = g_strdup(ct); 1322 s->desc_offset = 0; 1323 ret = vmdk_parse_extents(buf, bs, options, errp); 1324 exit: 1325 return ret; 1326 } 1327 1328 static int vmdk_open(BlockDriverState *bs, QDict *options, int flags, 1329 Error **errp) 1330 { 1331 char *buf; 1332 int ret; 1333 BDRVVmdkState *s = bs->opaque; 1334 uint32_t magic; 1335 1336 GRAPH_RDLOCK_GUARD_MAINLOOP(); 1337 1338 ret = bdrv_open_file_child(NULL, options, "file", bs, errp); 1339 if (ret < 0) { 1340 return ret; 1341 } 1342 1343 buf = vmdk_read_desc(bs->file, 0, errp); 1344 if (!buf) { 1345 return -EINVAL; 1346 } 1347 1348 magic = ldl_be_p(buf); 1349 switch (magic) { 1350 case VMDK3_MAGIC: 1351 case VMDK4_MAGIC: 1352 ret = vmdk_open_sparse(bs, bs->file, flags, buf, options, 1353 errp); 1354 s->desc_offset = 0x200; 1355 break; 1356 default: 1357 /* No data in the descriptor file */ 1358 bs->file->role &= ~BDRV_CHILD_DATA; 1359 1360 /* Must succeed because we have given up permissions if anything */ 1361 bdrv_child_refresh_perms(bs, bs->file, &error_abort); 1362 1363 ret = vmdk_open_desc_file(bs, flags, buf, options, errp); 1364 break; 1365 } 1366 if (ret) { 1367 goto fail; 1368 } 1369 1370 /* try to open parent images, if exist */ 1371 ret = vmdk_parent_open(bs); 1372 if (ret) { 1373 goto fail; 1374 } 1375 ret = vmdk_read_cid(bs, 0, &s->cid); 1376 if (ret) { 1377 goto fail; 1378 } 1379 ret = vmdk_read_cid(bs, 1, &s->parent_cid); 1380 if (ret) { 1381 goto fail; 1382 } 1383 qemu_co_mutex_init(&s->lock); 1384 1385 /* Disable migration when VMDK images are used */ 1386 error_setg(&s->migration_blocker, "The vmdk format used by node '%s' " 1387 "does not support live migration", 1388 bdrv_get_device_or_node_name(bs)); 1389 ret = migrate_add_blocker(s->migration_blocker, errp); 1390 if (ret < 0) { 1391 error_free(s->migration_blocker); 1392 goto fail; 1393 } 1394 1395 g_free(buf); 1396 return 0; 1397 1398 fail: 1399 g_free(buf); 1400 g_free(s->create_type); 1401 s->create_type = NULL; 1402 vmdk_free_extents(bs); 1403 return ret; 1404 } 1405 1406 1407 static void vmdk_refresh_limits(BlockDriverState *bs, Error **errp) 1408 { 1409 BDRVVmdkState *s = bs->opaque; 1410 int i; 1411 1412 for (i = 0; i < s->num_extents; i++) { 1413 if (!s->extents[i].flat) { 1414 bs->bl.pwrite_zeroes_alignment = 1415 MAX(bs->bl.pwrite_zeroes_alignment, 1416 s->extents[i].cluster_sectors << BDRV_SECTOR_BITS); 1417 } 1418 } 1419 } 1420 1421 /** 1422 * get_whole_cluster 1423 * 1424 * Copy backing file's cluster that covers @sector_num, otherwise write zero, 1425 * to the cluster at @cluster_sector_num. If @zeroed is true, we're overwriting 1426 * a zeroed cluster in the current layer and must not copy data from the 1427 * backing file. 1428 * 1429 * If @skip_start_sector < @skip_end_sector, the relative range 1430 * [@skip_start_sector, @skip_end_sector) is not copied or written, and leave 1431 * it for call to write user data in the request. 1432 */ 1433 static int coroutine_fn GRAPH_RDLOCK 1434 get_whole_cluster(BlockDriverState *bs, VmdkExtent *extent, 1435 uint64_t cluster_offset, uint64_t offset, 1436 uint64_t skip_start_bytes, uint64_t skip_end_bytes, 1437 bool zeroed) 1438 { 1439 int ret = VMDK_OK; 1440 int64_t cluster_bytes; 1441 uint8_t *whole_grain; 1442 bool copy_from_backing; 1443 1444 /* For COW, align request sector_num to cluster start */ 1445 cluster_bytes = extent->cluster_sectors << BDRV_SECTOR_BITS; 1446 offset = QEMU_ALIGN_DOWN(offset, cluster_bytes); 1447 whole_grain = qemu_blockalign(bs, cluster_bytes); 1448 copy_from_backing = bs->backing && !zeroed; 1449 1450 if (!copy_from_backing) { 1451 memset(whole_grain, 0, skip_start_bytes); 1452 memset(whole_grain + skip_end_bytes, 0, cluster_bytes - skip_end_bytes); 1453 } 1454 1455 assert(skip_end_bytes <= cluster_bytes); 1456 /* we will be here if it's first write on non-exist grain(cluster). 1457 * try to read from parent image, if exist */ 1458 if (bs->backing && !vmdk_is_cid_valid(bs)) { 1459 ret = VMDK_ERROR; 1460 goto exit; 1461 } 1462 1463 /* Read backing data before skip range */ 1464 if (skip_start_bytes > 0) { 1465 if (copy_from_backing) { 1466 /* qcow2 emits this on bs->file instead of bs->backing */ 1467 BLKDBG_CO_EVENT(extent->file, BLKDBG_COW_READ); 1468 ret = bdrv_co_pread(bs->backing, offset, skip_start_bytes, 1469 whole_grain, 0); 1470 if (ret < 0) { 1471 ret = VMDK_ERROR; 1472 goto exit; 1473 } 1474 } 1475 BLKDBG_CO_EVENT(extent->file, BLKDBG_COW_WRITE); 1476 ret = bdrv_co_pwrite(extent->file, cluster_offset, skip_start_bytes, 1477 whole_grain, 0); 1478 if (ret < 0) { 1479 ret = VMDK_ERROR; 1480 goto exit; 1481 } 1482 } 1483 /* Read backing data after skip range */ 1484 if (skip_end_bytes < cluster_bytes) { 1485 if (copy_from_backing) { 1486 /* qcow2 emits this on bs->file instead of bs->backing */ 1487 BLKDBG_CO_EVENT(extent->file, BLKDBG_COW_READ); 1488 ret = bdrv_co_pread(bs->backing, offset + skip_end_bytes, 1489 cluster_bytes - skip_end_bytes, 1490 whole_grain + skip_end_bytes, 0); 1491 if (ret < 0) { 1492 ret = VMDK_ERROR; 1493 goto exit; 1494 } 1495 } 1496 BLKDBG_CO_EVENT(extent->file, BLKDBG_COW_WRITE); 1497 ret = bdrv_co_pwrite(extent->file, cluster_offset + skip_end_bytes, 1498 cluster_bytes - skip_end_bytes, 1499 whole_grain + skip_end_bytes, 0); 1500 if (ret < 0) { 1501 ret = VMDK_ERROR; 1502 goto exit; 1503 } 1504 } 1505 1506 ret = VMDK_OK; 1507 exit: 1508 qemu_vfree(whole_grain); 1509 return ret; 1510 } 1511 1512 static int coroutine_fn GRAPH_RDLOCK 1513 vmdk_L2update(VmdkExtent *extent, VmdkMetaData *m_data, uint32_t offset) 1514 { 1515 offset = cpu_to_le32(offset); 1516 /* update L2 table */ 1517 BLKDBG_CO_EVENT(extent->file, BLKDBG_L2_UPDATE); 1518 if (bdrv_co_pwrite(extent->file, 1519 ((int64_t)m_data->l2_offset * 512) 1520 + (m_data->l2_index * sizeof(offset)), 1521 sizeof(offset), &offset, 0) < 0) { 1522 return VMDK_ERROR; 1523 } 1524 /* update backup L2 table */ 1525 if (extent->l1_backup_table_offset != 0) { 1526 m_data->l2_offset = extent->l1_backup_table[m_data->l1_index]; 1527 if (bdrv_co_pwrite(extent->file, 1528 ((int64_t)m_data->l2_offset * 512) 1529 + (m_data->l2_index * sizeof(offset)), 1530 sizeof(offset), &offset, 0) < 0) { 1531 return VMDK_ERROR; 1532 } 1533 } 1534 if (bdrv_co_flush(extent->file->bs) < 0) { 1535 return VMDK_ERROR; 1536 } 1537 if (m_data->l2_cache_entry) { 1538 *m_data->l2_cache_entry = offset; 1539 } 1540 1541 return VMDK_OK; 1542 } 1543 1544 /** 1545 * get_cluster_offset 1546 * 1547 * Look up cluster offset in extent file by sector number, and store in 1548 * @cluster_offset. 1549 * 1550 * For flat extents, the start offset as parsed from the description file is 1551 * returned. 1552 * 1553 * For sparse extents, look up in L1, L2 table. If allocate is true, return an 1554 * offset for a new cluster and update L2 cache. If there is a backing file, 1555 * COW is done before returning; otherwise, zeroes are written to the allocated 1556 * cluster. Both COW and zero writing skips the sector range 1557 * [@skip_start_sector, @skip_end_sector) passed in by caller, because caller 1558 * has new data to write there. 1559 * 1560 * Returns: VMDK_OK if cluster exists and mapped in the image. 1561 * VMDK_UNALLOC if cluster is not mapped and @allocate is false. 1562 * VMDK_ERROR if failed. 1563 */ 1564 static int coroutine_fn GRAPH_RDLOCK 1565 get_cluster_offset(BlockDriverState *bs, VmdkExtent *extent, 1566 VmdkMetaData *m_data, uint64_t offset, bool allocate, 1567 uint64_t *cluster_offset, uint64_t skip_start_bytes, 1568 uint64_t skip_end_bytes) 1569 { 1570 unsigned int l1_index, l2_offset, l2_index; 1571 int min_index, i, j; 1572 uint32_t min_count; 1573 void *l2_table; 1574 bool zeroed = false; 1575 int64_t ret; 1576 int64_t cluster_sector; 1577 unsigned int l2_size_bytes = extent->l2_size * extent->entry_size; 1578 1579 if (m_data) { 1580 m_data->new_allocation = false; 1581 } 1582 if (extent->flat) { 1583 *cluster_offset = extent->flat_start_offset; 1584 return VMDK_OK; 1585 } 1586 1587 offset -= (extent->end_sector - extent->sectors) * SECTOR_SIZE; 1588 l1_index = (offset >> 9) / extent->l1_entry_sectors; 1589 if (l1_index >= extent->l1_size) { 1590 return VMDK_ERROR; 1591 } 1592 if (extent->sesparse) { 1593 uint64_t l2_offset_u64; 1594 1595 assert(extent->entry_size == sizeof(uint64_t)); 1596 1597 l2_offset_u64 = ((uint64_t *)extent->l1_table)[l1_index]; 1598 if (l2_offset_u64 == 0) { 1599 l2_offset = 0; 1600 } else if ((l2_offset_u64 & 0xffffffff00000000) != 0x1000000000000000) { 1601 /* 1602 * Top most nibble is 0x1 if grain table is allocated. 1603 * strict check - top most 4 bytes must be 0x10000000 since max 1604 * supported size is 64TB for disk - so no more than 64TB / 16MB 1605 * grain directories which is smaller than uint32, 1606 * where 16MB is the only supported default grain table coverage. 1607 */ 1608 return VMDK_ERROR; 1609 } else { 1610 l2_offset_u64 = l2_offset_u64 & 0x00000000ffffffff; 1611 l2_offset_u64 = extent->sesparse_l2_tables_offset + 1612 l2_offset_u64 * l2_size_bytes / SECTOR_SIZE; 1613 if (l2_offset_u64 > 0x00000000ffffffff) { 1614 return VMDK_ERROR; 1615 } 1616 l2_offset = (unsigned int)(l2_offset_u64); 1617 } 1618 } else { 1619 assert(extent->entry_size == sizeof(uint32_t)); 1620 l2_offset = ((uint32_t *)extent->l1_table)[l1_index]; 1621 } 1622 if (!l2_offset) { 1623 return VMDK_UNALLOC; 1624 } 1625 for (i = 0; i < L2_CACHE_SIZE; i++) { 1626 if (l2_offset == extent->l2_cache_offsets[i]) { 1627 /* increment the hit count */ 1628 if (++extent->l2_cache_counts[i] == 0xffffffff) { 1629 for (j = 0; j < L2_CACHE_SIZE; j++) { 1630 extent->l2_cache_counts[j] >>= 1; 1631 } 1632 } 1633 l2_table = (char *)extent->l2_cache + (i * l2_size_bytes); 1634 goto found; 1635 } 1636 } 1637 /* not found: load a new entry in the least used one */ 1638 min_index = 0; 1639 min_count = 0xffffffff; 1640 for (i = 0; i < L2_CACHE_SIZE; i++) { 1641 if (extent->l2_cache_counts[i] < min_count) { 1642 min_count = extent->l2_cache_counts[i]; 1643 min_index = i; 1644 } 1645 } 1646 l2_table = (char *)extent->l2_cache + (min_index * l2_size_bytes); 1647 BLKDBG_CO_EVENT(extent->file, BLKDBG_L2_LOAD); 1648 if (bdrv_co_pread(extent->file, 1649 (int64_t)l2_offset * 512, 1650 l2_size_bytes, 1651 l2_table, 0 1652 ) < 0) { 1653 return VMDK_ERROR; 1654 } 1655 1656 extent->l2_cache_offsets[min_index] = l2_offset; 1657 extent->l2_cache_counts[min_index] = 1; 1658 found: 1659 l2_index = ((offset >> 9) / extent->cluster_sectors) % extent->l2_size; 1660 if (m_data) { 1661 m_data->l1_index = l1_index; 1662 m_data->l2_index = l2_index; 1663 m_data->l2_offset = l2_offset; 1664 m_data->l2_cache_entry = ((uint32_t *)l2_table) + l2_index; 1665 } 1666 1667 if (extent->sesparse) { 1668 cluster_sector = le64_to_cpu(((uint64_t *)l2_table)[l2_index]); 1669 switch (cluster_sector & 0xf000000000000000) { 1670 case 0x0000000000000000: 1671 /* unallocated grain */ 1672 if (cluster_sector != 0) { 1673 return VMDK_ERROR; 1674 } 1675 break; 1676 case 0x1000000000000000: 1677 /* scsi-unmapped grain - fallthrough */ 1678 case 0x2000000000000000: 1679 /* zero grain */ 1680 zeroed = true; 1681 break; 1682 case 0x3000000000000000: 1683 /* allocated grain */ 1684 cluster_sector = (((cluster_sector & 0x0fff000000000000) >> 48) | 1685 ((cluster_sector & 0x0000ffffffffffff) << 12)); 1686 cluster_sector = extent->sesparse_clusters_offset + 1687 cluster_sector * extent->cluster_sectors; 1688 break; 1689 default: 1690 return VMDK_ERROR; 1691 } 1692 } else { 1693 cluster_sector = le32_to_cpu(((uint32_t *)l2_table)[l2_index]); 1694 1695 if (extent->has_zero_grain && cluster_sector == VMDK_GTE_ZEROED) { 1696 zeroed = true; 1697 } 1698 } 1699 1700 if (!cluster_sector || zeroed) { 1701 if (!allocate) { 1702 return zeroed ? VMDK_ZEROED : VMDK_UNALLOC; 1703 } 1704 assert(!extent->sesparse); 1705 1706 if (extent->next_cluster_sector >= VMDK_EXTENT_MAX_SECTORS) { 1707 return VMDK_ERROR; 1708 } 1709 1710 cluster_sector = extent->next_cluster_sector; 1711 extent->next_cluster_sector += extent->cluster_sectors; 1712 1713 /* First of all we write grain itself, to avoid race condition 1714 * that may to corrupt the image. 1715 * This problem may occur because of insufficient space on host disk 1716 * or inappropriate VM shutdown. 1717 */ 1718 ret = get_whole_cluster(bs, extent, cluster_sector * BDRV_SECTOR_SIZE, 1719 offset, skip_start_bytes, skip_end_bytes, 1720 zeroed); 1721 if (ret) { 1722 return ret; 1723 } 1724 if (m_data) { 1725 m_data->new_allocation = true; 1726 } 1727 } 1728 *cluster_offset = cluster_sector << BDRV_SECTOR_BITS; 1729 return VMDK_OK; 1730 } 1731 1732 static VmdkExtent *find_extent(BDRVVmdkState *s, 1733 int64_t sector_num, VmdkExtent *start_hint) 1734 { 1735 VmdkExtent *extent = start_hint; 1736 1737 if (!extent) { 1738 extent = &s->extents[0]; 1739 } 1740 while (extent < &s->extents[s->num_extents]) { 1741 if (sector_num < extent->end_sector) { 1742 return extent; 1743 } 1744 extent++; 1745 } 1746 return NULL; 1747 } 1748 1749 static inline uint64_t vmdk_find_offset_in_cluster(VmdkExtent *extent, 1750 int64_t offset) 1751 { 1752 uint64_t extent_begin_offset, extent_relative_offset; 1753 uint64_t cluster_size = extent->cluster_sectors * BDRV_SECTOR_SIZE; 1754 1755 extent_begin_offset = 1756 (extent->end_sector - extent->sectors) * BDRV_SECTOR_SIZE; 1757 extent_relative_offset = offset - extent_begin_offset; 1758 return extent_relative_offset % cluster_size; 1759 } 1760 1761 static int coroutine_fn GRAPH_RDLOCK 1762 vmdk_co_block_status(BlockDriverState *bs, bool want_zero, 1763 int64_t offset, int64_t bytes, int64_t *pnum, 1764 int64_t *map, BlockDriverState **file) 1765 { 1766 BDRVVmdkState *s = bs->opaque; 1767 int64_t index_in_cluster, n, ret; 1768 uint64_t cluster_offset; 1769 VmdkExtent *extent; 1770 1771 extent = find_extent(s, offset >> BDRV_SECTOR_BITS, NULL); 1772 if (!extent) { 1773 return -EIO; 1774 } 1775 qemu_co_mutex_lock(&s->lock); 1776 ret = get_cluster_offset(bs, extent, NULL, offset, false, &cluster_offset, 1777 0, 0); 1778 qemu_co_mutex_unlock(&s->lock); 1779 1780 index_in_cluster = vmdk_find_offset_in_cluster(extent, offset); 1781 switch (ret) { 1782 case VMDK_ERROR: 1783 ret = -EIO; 1784 break; 1785 case VMDK_UNALLOC: 1786 ret = 0; 1787 break; 1788 case VMDK_ZEROED: 1789 ret = BDRV_BLOCK_ZERO; 1790 break; 1791 case VMDK_OK: 1792 ret = BDRV_BLOCK_DATA; 1793 if (!extent->compressed) { 1794 ret |= BDRV_BLOCK_OFFSET_VALID; 1795 *map = cluster_offset + index_in_cluster; 1796 if (extent->flat) { 1797 ret |= BDRV_BLOCK_RECURSE; 1798 } 1799 } else { 1800 ret |= BDRV_BLOCK_COMPRESSED; 1801 } 1802 *file = extent->file->bs; 1803 break; 1804 } 1805 1806 n = extent->cluster_sectors * BDRV_SECTOR_SIZE - index_in_cluster; 1807 *pnum = MIN(n, bytes); 1808 return ret; 1809 } 1810 1811 static int coroutine_fn GRAPH_RDLOCK 1812 vmdk_write_extent(VmdkExtent *extent, int64_t cluster_offset, 1813 int64_t offset_in_cluster, QEMUIOVector *qiov, 1814 uint64_t qiov_offset, uint64_t n_bytes, 1815 uint64_t offset) 1816 { 1817 int ret; 1818 VmdkGrainMarker *data = NULL; 1819 uLongf buf_len; 1820 QEMUIOVector local_qiov; 1821 int64_t write_offset; 1822 int64_t write_end_sector; 1823 1824 if (extent->compressed) { 1825 void *compressed_data; 1826 1827 /* Only whole clusters */ 1828 if (offset_in_cluster || 1829 n_bytes > (extent->cluster_sectors * SECTOR_SIZE) || 1830 (n_bytes < (extent->cluster_sectors * SECTOR_SIZE) && 1831 offset + n_bytes != extent->end_sector * SECTOR_SIZE)) 1832 { 1833 ret = -EINVAL; 1834 goto out; 1835 } 1836 1837 if (!extent->has_marker) { 1838 ret = -EINVAL; 1839 goto out; 1840 } 1841 buf_len = (extent->cluster_sectors << 9) * 2; 1842 data = g_malloc(buf_len + sizeof(VmdkGrainMarker)); 1843 1844 compressed_data = g_malloc(n_bytes); 1845 qemu_iovec_to_buf(qiov, qiov_offset, compressed_data, n_bytes); 1846 ret = compress(data->data, &buf_len, compressed_data, n_bytes); 1847 g_free(compressed_data); 1848 1849 if (ret != Z_OK || buf_len == 0) { 1850 ret = -EINVAL; 1851 goto out; 1852 } 1853 1854 data->lba = cpu_to_le64(offset >> BDRV_SECTOR_BITS); 1855 data->size = cpu_to_le32(buf_len); 1856 1857 n_bytes = buf_len + sizeof(VmdkGrainMarker); 1858 qemu_iovec_init_buf(&local_qiov, data, n_bytes); 1859 1860 BLKDBG_CO_EVENT(extent->file, BLKDBG_WRITE_COMPRESSED); 1861 } else { 1862 qemu_iovec_init(&local_qiov, qiov->niov); 1863 qemu_iovec_concat(&local_qiov, qiov, qiov_offset, n_bytes); 1864 1865 BLKDBG_CO_EVENT(extent->file, BLKDBG_WRITE_AIO); 1866 } 1867 1868 write_offset = cluster_offset + offset_in_cluster; 1869 ret = bdrv_co_pwritev(extent->file, write_offset, n_bytes, 1870 &local_qiov, 0); 1871 1872 write_end_sector = DIV_ROUND_UP(write_offset + n_bytes, BDRV_SECTOR_SIZE); 1873 1874 if (extent->compressed) { 1875 extent->next_cluster_sector = write_end_sector; 1876 } else { 1877 extent->next_cluster_sector = MAX(extent->next_cluster_sector, 1878 write_end_sector); 1879 } 1880 1881 if (ret < 0) { 1882 goto out; 1883 } 1884 ret = 0; 1885 out: 1886 g_free(data); 1887 if (!extent->compressed) { 1888 qemu_iovec_destroy(&local_qiov); 1889 } 1890 return ret; 1891 } 1892 1893 static int coroutine_fn GRAPH_RDLOCK 1894 vmdk_read_extent(VmdkExtent *extent, int64_t cluster_offset, 1895 int64_t offset_in_cluster, QEMUIOVector *qiov, int bytes) 1896 { 1897 int ret; 1898 int cluster_bytes, buf_bytes; 1899 uint8_t *cluster_buf, *compressed_data; 1900 uint8_t *uncomp_buf; 1901 uint32_t data_len; 1902 VmdkGrainMarker *marker; 1903 uLongf buf_len; 1904 1905 1906 if (!extent->compressed) { 1907 BLKDBG_CO_EVENT(extent->file, BLKDBG_READ_AIO); 1908 ret = bdrv_co_preadv(extent->file, 1909 cluster_offset + offset_in_cluster, bytes, 1910 qiov, 0); 1911 if (ret < 0) { 1912 return ret; 1913 } 1914 return 0; 1915 } 1916 cluster_bytes = extent->cluster_sectors * 512; 1917 /* Read two clusters in case GrainMarker + compressed data > one cluster */ 1918 buf_bytes = cluster_bytes * 2; 1919 cluster_buf = g_malloc(buf_bytes); 1920 uncomp_buf = g_malloc(cluster_bytes); 1921 BLKDBG_CO_EVENT(extent->file, BLKDBG_READ_COMPRESSED); 1922 ret = bdrv_co_pread(extent->file, cluster_offset, buf_bytes, cluster_buf, 1923 0); 1924 if (ret < 0) { 1925 goto out; 1926 } 1927 compressed_data = cluster_buf; 1928 buf_len = cluster_bytes; 1929 data_len = cluster_bytes; 1930 if (extent->has_marker) { 1931 marker = (VmdkGrainMarker *)cluster_buf; 1932 compressed_data = marker->data; 1933 data_len = le32_to_cpu(marker->size); 1934 } 1935 if (!data_len || data_len > buf_bytes) { 1936 ret = -EINVAL; 1937 goto out; 1938 } 1939 ret = uncompress(uncomp_buf, &buf_len, compressed_data, data_len); 1940 if (ret != Z_OK) { 1941 ret = -EINVAL; 1942 goto out; 1943 1944 } 1945 if (offset_in_cluster < 0 || 1946 offset_in_cluster + bytes > buf_len) { 1947 ret = -EINVAL; 1948 goto out; 1949 } 1950 qemu_iovec_from_buf(qiov, 0, uncomp_buf + offset_in_cluster, bytes); 1951 ret = 0; 1952 1953 out: 1954 g_free(uncomp_buf); 1955 g_free(cluster_buf); 1956 return ret; 1957 } 1958 1959 static int coroutine_fn GRAPH_RDLOCK 1960 vmdk_co_preadv(BlockDriverState *bs, int64_t offset, int64_t bytes, 1961 QEMUIOVector *qiov, BdrvRequestFlags flags) 1962 { 1963 BDRVVmdkState *s = bs->opaque; 1964 int ret; 1965 uint64_t n_bytes, offset_in_cluster; 1966 VmdkExtent *extent = NULL; 1967 QEMUIOVector local_qiov; 1968 uint64_t cluster_offset; 1969 uint64_t bytes_done = 0; 1970 1971 qemu_iovec_init(&local_qiov, qiov->niov); 1972 qemu_co_mutex_lock(&s->lock); 1973 1974 while (bytes > 0) { 1975 extent = find_extent(s, offset >> BDRV_SECTOR_BITS, extent); 1976 if (!extent) { 1977 ret = -EIO; 1978 goto fail; 1979 } 1980 ret = get_cluster_offset(bs, extent, NULL, 1981 offset, false, &cluster_offset, 0, 0); 1982 offset_in_cluster = vmdk_find_offset_in_cluster(extent, offset); 1983 1984 n_bytes = MIN(bytes, extent->cluster_sectors * BDRV_SECTOR_SIZE 1985 - offset_in_cluster); 1986 1987 if (ret != VMDK_OK) { 1988 /* if not allocated, try to read from parent image, if exist */ 1989 if (bs->backing && ret != VMDK_ZEROED) { 1990 if (!vmdk_is_cid_valid(bs)) { 1991 ret = -EINVAL; 1992 goto fail; 1993 } 1994 1995 qemu_iovec_reset(&local_qiov); 1996 qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes); 1997 1998 /* qcow2 emits this on bs->file instead of bs->backing */ 1999 BLKDBG_CO_EVENT(bs->file, BLKDBG_READ_BACKING_AIO); 2000 ret = bdrv_co_preadv(bs->backing, offset, n_bytes, 2001 &local_qiov, 0); 2002 if (ret < 0) { 2003 goto fail; 2004 } 2005 } else { 2006 qemu_iovec_memset(qiov, bytes_done, 0, n_bytes); 2007 } 2008 } else { 2009 qemu_iovec_reset(&local_qiov); 2010 qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes); 2011 2012 ret = vmdk_read_extent(extent, cluster_offset, offset_in_cluster, 2013 &local_qiov, n_bytes); 2014 if (ret) { 2015 goto fail; 2016 } 2017 } 2018 bytes -= n_bytes; 2019 offset += n_bytes; 2020 bytes_done += n_bytes; 2021 } 2022 2023 ret = 0; 2024 fail: 2025 qemu_co_mutex_unlock(&s->lock); 2026 qemu_iovec_destroy(&local_qiov); 2027 2028 return ret; 2029 } 2030 2031 /** 2032 * vmdk_write: 2033 * @zeroed: buf is ignored (data is zero), use zeroed_grain GTE feature 2034 * if possible, otherwise return -ENOTSUP. 2035 * @zero_dry_run: used for zeroed == true only, don't update L2 table, just try 2036 * with each cluster. By dry run we can find if the zero write 2037 * is possible without modifying image data. 2038 * 2039 * Returns: error code with 0 for success. 2040 */ 2041 static int coroutine_fn GRAPH_RDLOCK 2042 vmdk_pwritev(BlockDriverState *bs, uint64_t offset, uint64_t bytes, 2043 QEMUIOVector *qiov, bool zeroed, bool zero_dry_run) 2044 { 2045 BDRVVmdkState *s = bs->opaque; 2046 VmdkExtent *extent = NULL; 2047 int ret; 2048 int64_t offset_in_cluster, n_bytes; 2049 uint64_t cluster_offset; 2050 uint64_t bytes_done = 0; 2051 VmdkMetaData m_data; 2052 2053 if (DIV_ROUND_UP(offset, BDRV_SECTOR_SIZE) > bs->total_sectors) { 2054 error_report("Wrong offset: offset=0x%" PRIx64 2055 " total_sectors=0x%" PRIx64, 2056 offset, bs->total_sectors); 2057 return -EIO; 2058 } 2059 2060 while (bytes > 0) { 2061 extent = find_extent(s, offset >> BDRV_SECTOR_BITS, extent); 2062 if (!extent) { 2063 return -EIO; 2064 } 2065 if (extent->sesparse) { 2066 return -ENOTSUP; 2067 } 2068 offset_in_cluster = vmdk_find_offset_in_cluster(extent, offset); 2069 n_bytes = MIN(bytes, extent->cluster_sectors * BDRV_SECTOR_SIZE 2070 - offset_in_cluster); 2071 2072 ret = get_cluster_offset(bs, extent, &m_data, offset, 2073 !(extent->compressed || zeroed), 2074 &cluster_offset, offset_in_cluster, 2075 offset_in_cluster + n_bytes); 2076 if (extent->compressed) { 2077 if (ret == VMDK_OK) { 2078 /* Refuse write to allocated cluster for streamOptimized */ 2079 error_report("Could not write to allocated cluster" 2080 " for streamOptimized"); 2081 return -EIO; 2082 } else if (!zeroed) { 2083 /* allocate */ 2084 ret = get_cluster_offset(bs, extent, &m_data, offset, 2085 true, &cluster_offset, 0, 0); 2086 } 2087 } 2088 if (ret == VMDK_ERROR) { 2089 return -EINVAL; 2090 } 2091 if (zeroed) { 2092 /* Do zeroed write, buf is ignored */ 2093 if (extent->has_zero_grain && 2094 offset_in_cluster == 0 && 2095 n_bytes >= extent->cluster_sectors * BDRV_SECTOR_SIZE) { 2096 n_bytes = extent->cluster_sectors * BDRV_SECTOR_SIZE; 2097 if (!zero_dry_run && ret != VMDK_ZEROED) { 2098 /* update L2 tables */ 2099 if (vmdk_L2update(extent, &m_data, VMDK_GTE_ZEROED) 2100 != VMDK_OK) { 2101 return -EIO; 2102 } 2103 } 2104 } else { 2105 return -ENOTSUP; 2106 } 2107 } else { 2108 ret = vmdk_write_extent(extent, cluster_offset, offset_in_cluster, 2109 qiov, bytes_done, n_bytes, offset); 2110 if (ret) { 2111 return ret; 2112 } 2113 if (m_data.new_allocation) { 2114 /* update L2 tables */ 2115 if (vmdk_L2update(extent, &m_data, 2116 cluster_offset >> BDRV_SECTOR_BITS) 2117 != VMDK_OK) { 2118 return -EIO; 2119 } 2120 } 2121 } 2122 bytes -= n_bytes; 2123 offset += n_bytes; 2124 bytes_done += n_bytes; 2125 2126 /* update CID on the first write every time the virtual disk is 2127 * opened */ 2128 if (!s->cid_updated) { 2129 ret = vmdk_write_cid(bs, g_random_int()); 2130 if (ret < 0) { 2131 return ret; 2132 } 2133 s->cid_updated = true; 2134 } 2135 } 2136 return 0; 2137 } 2138 2139 static int coroutine_fn GRAPH_RDLOCK 2140 vmdk_co_pwritev(BlockDriverState *bs, int64_t offset, int64_t bytes, 2141 QEMUIOVector *qiov, BdrvRequestFlags flags) 2142 { 2143 int ret; 2144 BDRVVmdkState *s = bs->opaque; 2145 qemu_co_mutex_lock(&s->lock); 2146 ret = vmdk_pwritev(bs, offset, bytes, qiov, false, false); 2147 qemu_co_mutex_unlock(&s->lock); 2148 return ret; 2149 } 2150 2151 static int coroutine_fn GRAPH_RDLOCK 2152 vmdk_co_pwritev_compressed(BlockDriverState *bs, int64_t offset, int64_t bytes, 2153 QEMUIOVector *qiov) 2154 { 2155 if (bytes == 0) { 2156 /* The caller will write bytes 0 to signal EOF. 2157 * When receive it, we align EOF to a sector boundary. */ 2158 BDRVVmdkState *s = bs->opaque; 2159 int i, ret; 2160 int64_t length; 2161 2162 for (i = 0; i < s->num_extents; i++) { 2163 length = bdrv_co_getlength(s->extents[i].file->bs); 2164 if (length < 0) { 2165 return length; 2166 } 2167 length = QEMU_ALIGN_UP(length, BDRV_SECTOR_SIZE); 2168 ret = bdrv_co_truncate(s->extents[i].file, length, false, 2169 PREALLOC_MODE_OFF, 0, NULL); 2170 if (ret < 0) { 2171 return ret; 2172 } 2173 } 2174 return 0; 2175 } 2176 return vmdk_co_pwritev(bs, offset, bytes, qiov, 0); 2177 } 2178 2179 static int coroutine_fn GRAPH_RDLOCK 2180 vmdk_co_pwrite_zeroes(BlockDriverState *bs, int64_t offset, int64_t bytes, 2181 BdrvRequestFlags flags) 2182 { 2183 int ret; 2184 BDRVVmdkState *s = bs->opaque; 2185 2186 qemu_co_mutex_lock(&s->lock); 2187 /* write zeroes could fail if sectors not aligned to cluster, test it with 2188 * dry_run == true before really updating image */ 2189 ret = vmdk_pwritev(bs, offset, bytes, NULL, true, true); 2190 if (!ret) { 2191 ret = vmdk_pwritev(bs, offset, bytes, NULL, true, false); 2192 } 2193 qemu_co_mutex_unlock(&s->lock); 2194 return ret; 2195 } 2196 2197 static int coroutine_fn GRAPH_UNLOCKED 2198 vmdk_init_extent(BlockBackend *blk, int64_t filesize, bool flat, bool compress, 2199 bool zeroed_grain, Error **errp) 2200 { 2201 int ret, i; 2202 VMDK4Header header; 2203 uint32_t tmp, magic, grains, gd_sectors, gt_size, gt_count; 2204 uint32_t *gd_buf = NULL; 2205 int gd_buf_size; 2206 2207 if (flat) { 2208 ret = blk_co_truncate(blk, filesize, false, PREALLOC_MODE_OFF, 0, errp); 2209 goto exit; 2210 } 2211 magic = cpu_to_be32(VMDK4_MAGIC); 2212 memset(&header, 0, sizeof(header)); 2213 if (compress) { 2214 header.version = 3; 2215 } else if (zeroed_grain) { 2216 header.version = 2; 2217 } else { 2218 header.version = 1; 2219 } 2220 header.flags = VMDK4_FLAG_RGD | VMDK4_FLAG_NL_DETECT 2221 | (compress ? VMDK4_FLAG_COMPRESS | VMDK4_FLAG_MARKER : 0) 2222 | (zeroed_grain ? VMDK4_FLAG_ZERO_GRAIN : 0); 2223 header.compressAlgorithm = compress ? VMDK4_COMPRESSION_DEFLATE : 0; 2224 header.capacity = filesize / BDRV_SECTOR_SIZE; 2225 header.granularity = 128; 2226 header.num_gtes_per_gt = BDRV_SECTOR_SIZE; 2227 2228 grains = DIV_ROUND_UP(filesize / BDRV_SECTOR_SIZE, header.granularity); 2229 gt_size = DIV_ROUND_UP(header.num_gtes_per_gt * sizeof(uint32_t), 2230 BDRV_SECTOR_SIZE); 2231 gt_count = DIV_ROUND_UP(grains, header.num_gtes_per_gt); 2232 gd_sectors = DIV_ROUND_UP(gt_count * sizeof(uint32_t), BDRV_SECTOR_SIZE); 2233 2234 header.desc_offset = 1; 2235 header.desc_size = 20; 2236 header.rgd_offset = header.desc_offset + header.desc_size; 2237 header.gd_offset = header.rgd_offset + gd_sectors + (gt_size * gt_count); 2238 header.grain_offset = 2239 ROUND_UP(header.gd_offset + gd_sectors + (gt_size * gt_count), 2240 header.granularity); 2241 /* swap endianness for all header fields */ 2242 header.version = cpu_to_le32(header.version); 2243 header.flags = cpu_to_le32(header.flags); 2244 header.capacity = cpu_to_le64(header.capacity); 2245 header.granularity = cpu_to_le64(header.granularity); 2246 header.num_gtes_per_gt = cpu_to_le32(header.num_gtes_per_gt); 2247 header.desc_offset = cpu_to_le64(header.desc_offset); 2248 header.desc_size = cpu_to_le64(header.desc_size); 2249 header.rgd_offset = cpu_to_le64(header.rgd_offset); 2250 header.gd_offset = cpu_to_le64(header.gd_offset); 2251 header.grain_offset = cpu_to_le64(header.grain_offset); 2252 header.compressAlgorithm = cpu_to_le16(header.compressAlgorithm); 2253 2254 header.check_bytes[0] = 0xa; 2255 header.check_bytes[1] = 0x20; 2256 header.check_bytes[2] = 0xd; 2257 header.check_bytes[3] = 0xa; 2258 2259 /* write all the data */ 2260 ret = blk_co_pwrite(blk, 0, sizeof(magic), &magic, 0); 2261 if (ret < 0) { 2262 error_setg(errp, QERR_IO_ERROR); 2263 goto exit; 2264 } 2265 ret = blk_co_pwrite(blk, sizeof(magic), sizeof(header), &header, 0); 2266 if (ret < 0) { 2267 error_setg(errp, QERR_IO_ERROR); 2268 goto exit; 2269 } 2270 2271 ret = blk_co_truncate(blk, le64_to_cpu(header.grain_offset) << 9, false, 2272 PREALLOC_MODE_OFF, 0, errp); 2273 if (ret < 0) { 2274 goto exit; 2275 } 2276 2277 /* write grain directory */ 2278 gd_buf_size = gd_sectors * BDRV_SECTOR_SIZE; 2279 gd_buf = g_malloc0(gd_buf_size); 2280 for (i = 0, tmp = le64_to_cpu(header.rgd_offset) + gd_sectors; 2281 i < gt_count; i++, tmp += gt_size) { 2282 gd_buf[i] = cpu_to_le32(tmp); 2283 } 2284 ret = blk_co_pwrite(blk, le64_to_cpu(header.rgd_offset) * BDRV_SECTOR_SIZE, 2285 gd_buf_size, gd_buf, 0); 2286 if (ret < 0) { 2287 error_setg(errp, QERR_IO_ERROR); 2288 goto exit; 2289 } 2290 2291 /* write backup grain directory */ 2292 for (i = 0, tmp = le64_to_cpu(header.gd_offset) + gd_sectors; 2293 i < gt_count; i++, tmp += gt_size) { 2294 gd_buf[i] = cpu_to_le32(tmp); 2295 } 2296 ret = blk_co_pwrite(blk, le64_to_cpu(header.gd_offset) * BDRV_SECTOR_SIZE, 2297 gd_buf_size, gd_buf, 0); 2298 if (ret < 0) { 2299 error_setg(errp, QERR_IO_ERROR); 2300 } 2301 2302 ret = 0; 2303 exit: 2304 g_free(gd_buf); 2305 return ret; 2306 } 2307 2308 static int coroutine_fn GRAPH_UNLOCKED 2309 vmdk_create_extent(const char *filename, int64_t filesize, bool flat, 2310 bool compress, bool zeroed_grain, BlockBackend **pbb, 2311 QemuOpts *opts, Error **errp) 2312 { 2313 int ret; 2314 BlockBackend *blk = NULL; 2315 2316 ret = bdrv_co_create_file(filename, opts, errp); 2317 if (ret < 0) { 2318 goto exit; 2319 } 2320 2321 blk = blk_co_new_open(filename, NULL, NULL, 2322 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, 2323 errp); 2324 if (blk == NULL) { 2325 ret = -EIO; 2326 goto exit; 2327 } 2328 2329 blk_set_allow_write_beyond_eof(blk, true); 2330 2331 ret = vmdk_init_extent(blk, filesize, flat, compress, zeroed_grain, errp); 2332 exit: 2333 if (blk) { 2334 if (pbb) { 2335 *pbb = blk; 2336 } else { 2337 blk_co_unref(blk); 2338 blk = NULL; 2339 } 2340 } 2341 return ret; 2342 } 2343 2344 static int filename_decompose(const char *filename, char *path, char *prefix, 2345 char *postfix, size_t buf_len, Error **errp) 2346 { 2347 const char *p, *q; 2348 2349 if (filename == NULL || !strlen(filename)) { 2350 error_setg(errp, "No filename provided"); 2351 return VMDK_ERROR; 2352 } 2353 p = strrchr(filename, '/'); 2354 if (p == NULL) { 2355 p = strrchr(filename, '\\'); 2356 } 2357 if (p == NULL) { 2358 p = strrchr(filename, ':'); 2359 } 2360 if (p != NULL) { 2361 p++; 2362 if (p - filename >= buf_len) { 2363 return VMDK_ERROR; 2364 } 2365 pstrcpy(path, p - filename + 1, filename); 2366 } else { 2367 p = filename; 2368 path[0] = '\0'; 2369 } 2370 q = strrchr(p, '.'); 2371 if (q == NULL) { 2372 pstrcpy(prefix, buf_len, p); 2373 postfix[0] = '\0'; 2374 } else { 2375 if (q - p >= buf_len) { 2376 return VMDK_ERROR; 2377 } 2378 pstrcpy(prefix, q - p + 1, p); 2379 pstrcpy(postfix, buf_len, q); 2380 } 2381 return VMDK_OK; 2382 } 2383 2384 /* 2385 * idx == 0: get or create the descriptor file (also the image file if in a 2386 * non-split format. 2387 * idx >= 1: get the n-th extent if in a split subformat 2388 */ 2389 typedef BlockBackend * coroutine_fn GRAPH_UNLOCKED_PTR 2390 (*vmdk_create_extent_fn)(int64_t size, int idx, bool flat, bool split, 2391 bool compress, bool zeroed_grain, void *opaque, 2392 Error **errp); 2393 2394 static void vmdk_desc_add_extent(GString *desc, 2395 const char *extent_line_fmt, 2396 int64_t size, const char *filename) 2397 { 2398 char *basename = g_path_get_basename(filename); 2399 2400 g_string_append_printf(desc, extent_line_fmt, 2401 DIV_ROUND_UP(size, BDRV_SECTOR_SIZE), basename); 2402 g_free(basename); 2403 } 2404 2405 static int coroutine_fn GRAPH_UNLOCKED 2406 vmdk_co_do_create(int64_t size, 2407 BlockdevVmdkSubformat subformat, 2408 BlockdevVmdkAdapterType adapter_type, 2409 const char *backing_file, 2410 const char *hw_version, 2411 const char *toolsversion, 2412 bool compat6, 2413 bool zeroed_grain, 2414 vmdk_create_extent_fn extent_fn, 2415 void *opaque, 2416 Error **errp) 2417 { 2418 int extent_idx; 2419 BlockBackend *blk = NULL; 2420 BlockBackend *extent_blk; 2421 Error *local_err = NULL; 2422 char *desc = NULL; 2423 int ret = 0; 2424 bool flat, split, compress; 2425 GString *ext_desc_lines; 2426 const int64_t split_size = 0x80000000; /* VMDK has constant split size */ 2427 int64_t extent_size; 2428 int64_t created_size = 0; 2429 const char *extent_line_fmt; 2430 char *parent_desc_line = g_malloc0(BUF_SIZE); 2431 uint32_t parent_cid = 0xffffffff; 2432 uint32_t number_heads = 16; 2433 uint32_t desc_offset = 0, desc_len; 2434 const char desc_template[] = 2435 "# Disk DescriptorFile\n" 2436 "version=1\n" 2437 "CID=%" PRIx32 "\n" 2438 "parentCID=%" PRIx32 "\n" 2439 "createType=\"%s\"\n" 2440 "%s" 2441 "\n" 2442 "# Extent description\n" 2443 "%s" 2444 "\n" 2445 "# The Disk Data Base\n" 2446 "#DDB\n" 2447 "\n" 2448 "ddb.virtualHWVersion = \"%s\"\n" 2449 "ddb.geometry.cylinders = \"%" PRId64 "\"\n" 2450 "ddb.geometry.heads = \"%" PRIu32 "\"\n" 2451 "ddb.geometry.sectors = \"63\"\n" 2452 "ddb.adapterType = \"%s\"\n" 2453 "ddb.toolsVersion = \"%s\"\n"; 2454 2455 ext_desc_lines = g_string_new(NULL); 2456 2457 /* Read out options */ 2458 if (compat6) { 2459 if (hw_version) { 2460 error_setg(errp, 2461 "compat6 cannot be enabled with hwversion set"); 2462 ret = -EINVAL; 2463 goto exit; 2464 } 2465 hw_version = "6"; 2466 } 2467 if (!hw_version) { 2468 hw_version = "4"; 2469 } 2470 if (!toolsversion) { 2471 toolsversion = "2147483647"; 2472 } 2473 2474 if (adapter_type != BLOCKDEV_VMDK_ADAPTER_TYPE_IDE) { 2475 /* that's the number of heads with which vmware operates when 2476 creating, exporting, etc. vmdk files with a non-ide adapter type */ 2477 number_heads = 255; 2478 } 2479 split = (subformat == BLOCKDEV_VMDK_SUBFORMAT_TWOGBMAXEXTENTFLAT) || 2480 (subformat == BLOCKDEV_VMDK_SUBFORMAT_TWOGBMAXEXTENTSPARSE); 2481 flat = (subformat == BLOCKDEV_VMDK_SUBFORMAT_MONOLITHICFLAT) || 2482 (subformat == BLOCKDEV_VMDK_SUBFORMAT_TWOGBMAXEXTENTFLAT); 2483 compress = subformat == BLOCKDEV_VMDK_SUBFORMAT_STREAMOPTIMIZED; 2484 2485 if (flat) { 2486 extent_line_fmt = "RW %" PRId64 " FLAT \"%s\" 0\n"; 2487 } else { 2488 extent_line_fmt = "RW %" PRId64 " SPARSE \"%s\"\n"; 2489 } 2490 if (flat && backing_file) { 2491 error_setg(errp, "Flat image can't have backing file"); 2492 ret = -ENOTSUP; 2493 goto exit; 2494 } 2495 if (flat && zeroed_grain) { 2496 error_setg(errp, "Flat image can't enable zeroed grain"); 2497 ret = -ENOTSUP; 2498 goto exit; 2499 } 2500 2501 /* Create extents */ 2502 if (split) { 2503 extent_size = split_size; 2504 } else { 2505 extent_size = size; 2506 } 2507 if (!split && !flat) { 2508 created_size = extent_size; 2509 } else { 2510 created_size = 0; 2511 } 2512 /* Get the descriptor file BDS */ 2513 blk = extent_fn(created_size, 0, flat, split, compress, zeroed_grain, 2514 opaque, errp); 2515 if (!blk) { 2516 ret = -EIO; 2517 goto exit; 2518 } 2519 if (!split && !flat) { 2520 vmdk_desc_add_extent(ext_desc_lines, extent_line_fmt, created_size, 2521 blk_bs(blk)->filename); 2522 } 2523 2524 if (backing_file) { 2525 BlockBackend *backing; 2526 char *full_backing = 2527 bdrv_get_full_backing_filename_from_filename(blk_bs(blk)->filename, 2528 backing_file, 2529 &local_err); 2530 if (local_err) { 2531 error_propagate(errp, local_err); 2532 ret = -ENOENT; 2533 goto exit; 2534 } 2535 assert(full_backing); 2536 2537 backing = blk_co_new_open(full_backing, NULL, NULL, 2538 BDRV_O_NO_BACKING, errp); 2539 g_free(full_backing); 2540 if (backing == NULL) { 2541 ret = -EIO; 2542 goto exit; 2543 } 2544 if (strcmp(blk_bs(backing)->drv->format_name, "vmdk")) { 2545 error_setg(errp, "Invalid backing file format: %s. Must be vmdk", 2546 blk_bs(backing)->drv->format_name); 2547 blk_co_unref(backing); 2548 ret = -EINVAL; 2549 goto exit; 2550 } 2551 ret = vmdk_read_cid(blk_bs(backing), 0, &parent_cid); 2552 blk_co_unref(backing); 2553 if (ret) { 2554 error_setg(errp, "Failed to read parent CID"); 2555 goto exit; 2556 } 2557 snprintf(parent_desc_line, BUF_SIZE, 2558 "parentFileNameHint=\"%s\"", backing_file); 2559 } 2560 extent_idx = 1; 2561 while (created_size < size) { 2562 int64_t cur_size = MIN(size - created_size, extent_size); 2563 extent_blk = extent_fn(cur_size, extent_idx, flat, split, compress, 2564 zeroed_grain, opaque, errp); 2565 if (!extent_blk) { 2566 ret = -EINVAL; 2567 goto exit; 2568 } 2569 vmdk_desc_add_extent(ext_desc_lines, extent_line_fmt, cur_size, 2570 blk_bs(extent_blk)->filename); 2571 created_size += cur_size; 2572 extent_idx++; 2573 blk_co_unref(extent_blk); 2574 } 2575 2576 /* Check whether we got excess extents */ 2577 extent_blk = extent_fn(-1, extent_idx, flat, split, compress, zeroed_grain, 2578 opaque, NULL); 2579 if (extent_blk) { 2580 blk_co_unref(extent_blk); 2581 error_setg(errp, "List of extents contains unused extents"); 2582 ret = -EINVAL; 2583 goto exit; 2584 } 2585 2586 /* generate descriptor file */ 2587 desc = g_strdup_printf(desc_template, 2588 g_random_int(), 2589 parent_cid, 2590 BlockdevVmdkSubformat_str(subformat), 2591 parent_desc_line, 2592 ext_desc_lines->str, 2593 hw_version, 2594 size / 2595 (int64_t)(63 * number_heads * BDRV_SECTOR_SIZE), 2596 number_heads, 2597 BlockdevVmdkAdapterType_str(adapter_type), 2598 toolsversion); 2599 desc_len = strlen(desc); 2600 /* the descriptor offset = 0x200 */ 2601 if (!split && !flat) { 2602 desc_offset = 0x200; 2603 } 2604 2605 ret = blk_co_pwrite(blk, desc_offset, desc_len, desc, 0); 2606 if (ret < 0) { 2607 error_setg_errno(errp, -ret, "Could not write description"); 2608 goto exit; 2609 } 2610 /* bdrv_pwrite write padding zeros to align to sector, we don't need that 2611 * for description file */ 2612 if (desc_offset == 0) { 2613 ret = blk_co_truncate(blk, desc_len, false, PREALLOC_MODE_OFF, 0, errp); 2614 if (ret < 0) { 2615 goto exit; 2616 } 2617 } 2618 ret = 0; 2619 exit: 2620 if (blk) { 2621 blk_co_unref(blk); 2622 } 2623 g_free(desc); 2624 g_free(parent_desc_line); 2625 g_string_free(ext_desc_lines, true); 2626 return ret; 2627 } 2628 2629 typedef struct { 2630 char *path; 2631 char *prefix; 2632 char *postfix; 2633 QemuOpts *opts; 2634 } VMDKCreateOptsData; 2635 2636 static BlockBackend * coroutine_fn GRAPH_UNLOCKED 2637 vmdk_co_create_opts_cb(int64_t size, int idx, bool flat, bool split, 2638 bool compress, bool zeroed_grain, void *opaque, 2639 Error **errp) 2640 { 2641 BlockBackend *blk = NULL; 2642 BlockDriverState *bs = NULL; 2643 VMDKCreateOptsData *data = opaque; 2644 char *ext_filename = NULL; 2645 char *rel_filename = NULL; 2646 2647 /* We're done, don't create excess extents. */ 2648 if (size == -1) { 2649 assert(errp == NULL); 2650 return NULL; 2651 } 2652 2653 if (idx == 0) { 2654 rel_filename = g_strdup_printf("%s%s", data->prefix, data->postfix); 2655 } else if (split) { 2656 rel_filename = g_strdup_printf("%s-%c%03d%s", 2657 data->prefix, 2658 flat ? 'f' : 's', idx, data->postfix); 2659 } else { 2660 assert(idx == 1); 2661 rel_filename = g_strdup_printf("%s-flat%s", data->prefix, data->postfix); 2662 } 2663 2664 ext_filename = g_strdup_printf("%s%s", data->path, rel_filename); 2665 g_free(rel_filename); 2666 2667 if (vmdk_create_extent(ext_filename, size, 2668 flat, compress, zeroed_grain, &blk, data->opts, 2669 errp)) { 2670 goto exit; 2671 } 2672 bdrv_co_unref(bs); 2673 exit: 2674 g_free(ext_filename); 2675 return blk; 2676 } 2677 2678 static int coroutine_fn GRAPH_UNLOCKED 2679 vmdk_co_create_opts(BlockDriver *drv, const char *filename, 2680 QemuOpts *opts, Error **errp) 2681 { 2682 Error *local_err = NULL; 2683 char *desc = NULL; 2684 int64_t total_size = 0; 2685 char *adapter_type = NULL; 2686 BlockdevVmdkAdapterType adapter_type_enum; 2687 char *backing_file = NULL; 2688 char *hw_version = NULL; 2689 char *toolsversion = NULL; 2690 char *fmt = NULL; 2691 BlockdevVmdkSubformat subformat; 2692 int ret = 0; 2693 char *path = g_malloc0(PATH_MAX); 2694 char *prefix = g_malloc0(PATH_MAX); 2695 char *postfix = g_malloc0(PATH_MAX); 2696 char *desc_line = g_malloc0(BUF_SIZE); 2697 char *ext_filename = g_malloc0(PATH_MAX); 2698 char *desc_filename = g_malloc0(PATH_MAX); 2699 char *parent_desc_line = g_malloc0(BUF_SIZE); 2700 bool zeroed_grain; 2701 bool compat6; 2702 VMDKCreateOptsData data; 2703 char *backing_fmt = NULL; 2704 2705 backing_fmt = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FMT); 2706 if (backing_fmt && strcmp(backing_fmt, "vmdk") != 0) { 2707 error_setg(errp, "backing_file must be a vmdk image"); 2708 ret = -EINVAL; 2709 goto exit; 2710 } 2711 2712 if (filename_decompose(filename, path, prefix, postfix, PATH_MAX, errp)) { 2713 ret = -EINVAL; 2714 goto exit; 2715 } 2716 /* Read out options */ 2717 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0), 2718 BDRV_SECTOR_SIZE); 2719 adapter_type = qemu_opt_get_del(opts, BLOCK_OPT_ADAPTER_TYPE); 2720 backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE); 2721 hw_version = qemu_opt_get_del(opts, BLOCK_OPT_HWVERSION); 2722 toolsversion = qemu_opt_get_del(opts, BLOCK_OPT_TOOLSVERSION); 2723 compat6 = qemu_opt_get_bool_del(opts, BLOCK_OPT_COMPAT6, false); 2724 if (strcmp(hw_version, "undefined") == 0) { 2725 g_free(hw_version); 2726 hw_version = NULL; 2727 } 2728 fmt = qemu_opt_get_del(opts, BLOCK_OPT_SUBFMT); 2729 zeroed_grain = qemu_opt_get_bool_del(opts, BLOCK_OPT_ZEROED_GRAIN, false); 2730 2731 if (adapter_type) { 2732 adapter_type_enum = qapi_enum_parse(&BlockdevVmdkAdapterType_lookup, 2733 adapter_type, 2734 BLOCKDEV_VMDK_ADAPTER_TYPE_IDE, 2735 &local_err); 2736 if (local_err) { 2737 error_propagate(errp, local_err); 2738 ret = -EINVAL; 2739 goto exit; 2740 } 2741 } else { 2742 adapter_type_enum = BLOCKDEV_VMDK_ADAPTER_TYPE_IDE; 2743 } 2744 2745 if (!fmt) { 2746 /* Default format to monolithicSparse */ 2747 subformat = BLOCKDEV_VMDK_SUBFORMAT_MONOLITHICSPARSE; 2748 } else { 2749 subformat = qapi_enum_parse(&BlockdevVmdkSubformat_lookup, 2750 fmt, 2751 BLOCKDEV_VMDK_SUBFORMAT_MONOLITHICSPARSE, 2752 &local_err); 2753 if (local_err) { 2754 error_propagate(errp, local_err); 2755 ret = -EINVAL; 2756 goto exit; 2757 } 2758 } 2759 data = (VMDKCreateOptsData){ 2760 .prefix = prefix, 2761 .postfix = postfix, 2762 .path = path, 2763 .opts = opts, 2764 }; 2765 ret = vmdk_co_do_create(total_size, subformat, adapter_type_enum, 2766 backing_file, hw_version, toolsversion, compat6, 2767 zeroed_grain, vmdk_co_create_opts_cb, &data, errp); 2768 2769 exit: 2770 g_free(backing_fmt); 2771 g_free(adapter_type); 2772 g_free(backing_file); 2773 g_free(hw_version); 2774 g_free(toolsversion); 2775 g_free(fmt); 2776 g_free(desc); 2777 g_free(path); 2778 g_free(prefix); 2779 g_free(postfix); 2780 g_free(desc_line); 2781 g_free(ext_filename); 2782 g_free(desc_filename); 2783 g_free(parent_desc_line); 2784 return ret; 2785 } 2786 2787 static BlockBackend * coroutine_fn GRAPH_UNLOCKED 2788 vmdk_co_create_cb(int64_t size, int idx, bool flat, bool split, bool compress, 2789 bool zeroed_grain, void *opaque, Error **errp) 2790 { 2791 int ret; 2792 BlockDriverState *bs; 2793 BlockBackend *blk; 2794 BlockdevCreateOptionsVmdk *opts = opaque; 2795 2796 if (idx == 0) { 2797 bs = bdrv_co_open_blockdev_ref(opts->file, errp); 2798 } else { 2799 int i; 2800 BlockdevRefList *list = opts->extents; 2801 for (i = 1; i < idx; i++) { 2802 if (!list || !list->next) { 2803 error_setg(errp, "Extent [%d] not specified", i); 2804 return NULL; 2805 } 2806 list = list->next; 2807 } 2808 if (!list) { 2809 error_setg(errp, "Extent [%d] not specified", idx - 1); 2810 return NULL; 2811 } 2812 bs = bdrv_co_open_blockdev_ref(list->value, errp); 2813 } 2814 if (!bs) { 2815 return NULL; 2816 } 2817 blk = blk_co_new_with_bs(bs, 2818 BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE | 2819 BLK_PERM_RESIZE, 2820 BLK_PERM_ALL, 2821 errp); 2822 if (!blk) { 2823 return NULL; 2824 } 2825 blk_set_allow_write_beyond_eof(blk, true); 2826 bdrv_co_unref(bs); 2827 2828 if (size != -1) { 2829 ret = vmdk_init_extent(blk, size, flat, compress, zeroed_grain, errp); 2830 if (ret) { 2831 blk_co_unref(blk); 2832 blk = NULL; 2833 } 2834 } 2835 return blk; 2836 } 2837 2838 static int coroutine_fn GRAPH_UNLOCKED 2839 vmdk_co_create(BlockdevCreateOptions *create_options, Error **errp) 2840 { 2841 BlockdevCreateOptionsVmdk *opts; 2842 2843 opts = &create_options->u.vmdk; 2844 2845 /* Validate options */ 2846 if (!QEMU_IS_ALIGNED(opts->size, BDRV_SECTOR_SIZE)) { 2847 error_setg(errp, "Image size must be a multiple of 512 bytes"); 2848 return -EINVAL; 2849 } 2850 2851 return vmdk_co_do_create(opts->size, 2852 opts->subformat, 2853 opts->adapter_type, 2854 opts->backing_file, 2855 opts->hwversion, 2856 opts->toolsversion, 2857 false, 2858 opts->zeroed_grain, 2859 vmdk_co_create_cb, 2860 opts, errp); 2861 } 2862 2863 static void vmdk_close(BlockDriverState *bs) 2864 { 2865 BDRVVmdkState *s = bs->opaque; 2866 2867 vmdk_free_extents(bs); 2868 g_free(s->create_type); 2869 2870 migrate_del_blocker(s->migration_blocker); 2871 error_free(s->migration_blocker); 2872 } 2873 2874 static int64_t coroutine_fn GRAPH_RDLOCK 2875 vmdk_co_get_allocated_file_size(BlockDriverState *bs) 2876 { 2877 int i; 2878 int64_t ret = 0; 2879 int64_t r; 2880 BDRVVmdkState *s = bs->opaque; 2881 2882 ret = bdrv_co_get_allocated_file_size(bs->file->bs); 2883 if (ret < 0) { 2884 return ret; 2885 } 2886 for (i = 0; i < s->num_extents; i++) { 2887 if (s->extents[i].file == bs->file) { 2888 continue; 2889 } 2890 r = bdrv_co_get_allocated_file_size(s->extents[i].file->bs); 2891 if (r < 0) { 2892 return r; 2893 } 2894 ret += r; 2895 } 2896 return ret; 2897 } 2898 2899 static int vmdk_has_zero_init(BlockDriverState *bs) 2900 { 2901 int i; 2902 BDRVVmdkState *s = bs->opaque; 2903 2904 /* If has a flat extent and its underlying storage doesn't have zero init, 2905 * return 0. */ 2906 for (i = 0; i < s->num_extents; i++) { 2907 if (s->extents[i].flat) { 2908 if (!bdrv_has_zero_init(s->extents[i].file->bs)) { 2909 return 0; 2910 } 2911 } 2912 } 2913 return 1; 2914 } 2915 2916 static VmdkExtentInfo * GRAPH_RDLOCK vmdk_get_extent_info(VmdkExtent *extent) 2917 { 2918 VmdkExtentInfo *info = g_new0(VmdkExtentInfo, 1); 2919 2920 bdrv_refresh_filename(extent->file->bs); 2921 *info = (VmdkExtentInfo){ 2922 .filename = g_strdup(extent->file->bs->filename), 2923 .format = g_strdup(extent->type), 2924 .virtual_size = extent->sectors * BDRV_SECTOR_SIZE, 2925 .compressed = extent->compressed, 2926 .has_compressed = extent->compressed, 2927 .cluster_size = extent->cluster_sectors * BDRV_SECTOR_SIZE, 2928 .has_cluster_size = !extent->flat, 2929 }; 2930 2931 return info; 2932 } 2933 2934 static int coroutine_fn GRAPH_RDLOCK 2935 vmdk_co_check(BlockDriverState *bs, BdrvCheckResult *result, BdrvCheckMode fix) 2936 { 2937 BDRVVmdkState *s = bs->opaque; 2938 VmdkExtent *extent = NULL; 2939 int64_t sector_num = 0; 2940 int64_t total_sectors = bdrv_co_nb_sectors(bs); 2941 int ret; 2942 uint64_t cluster_offset; 2943 2944 if (fix) { 2945 return -ENOTSUP; 2946 } 2947 2948 for (;;) { 2949 if (sector_num >= total_sectors) { 2950 return 0; 2951 } 2952 extent = find_extent(s, sector_num, extent); 2953 if (!extent) { 2954 fprintf(stderr, 2955 "ERROR: could not find extent for sector %" PRId64 "\n", 2956 sector_num); 2957 ret = -EINVAL; 2958 break; 2959 } 2960 ret = get_cluster_offset(bs, extent, NULL, 2961 sector_num << BDRV_SECTOR_BITS, 2962 false, &cluster_offset, 0, 0); 2963 if (ret == VMDK_ERROR) { 2964 fprintf(stderr, 2965 "ERROR: could not get cluster_offset for sector %" 2966 PRId64 "\n", sector_num); 2967 break; 2968 } 2969 if (ret == VMDK_OK) { 2970 int64_t extent_len = bdrv_co_getlength(extent->file->bs); 2971 if (extent_len < 0) { 2972 fprintf(stderr, 2973 "ERROR: could not get extent file length for sector %" 2974 PRId64 "\n", sector_num); 2975 ret = extent_len; 2976 break; 2977 } 2978 if (cluster_offset >= extent_len) { 2979 fprintf(stderr, 2980 "ERROR: cluster offset for sector %" 2981 PRId64 " points after EOF\n", sector_num); 2982 ret = -EINVAL; 2983 break; 2984 } 2985 } 2986 sector_num += extent->cluster_sectors; 2987 } 2988 2989 result->corruptions++; 2990 return ret; 2991 } 2992 2993 static ImageInfoSpecific * GRAPH_RDLOCK 2994 vmdk_get_specific_info(BlockDriverState *bs, Error **errp) 2995 { 2996 int i; 2997 BDRVVmdkState *s = bs->opaque; 2998 ImageInfoSpecific *spec_info = g_new0(ImageInfoSpecific, 1); 2999 VmdkExtentInfoList **tail; 3000 3001 *spec_info = (ImageInfoSpecific){ 3002 .type = IMAGE_INFO_SPECIFIC_KIND_VMDK, 3003 .u = { 3004 .vmdk.data = g_new0(ImageInfoSpecificVmdk, 1), 3005 }, 3006 }; 3007 3008 *spec_info->u.vmdk.data = (ImageInfoSpecificVmdk) { 3009 .create_type = g_strdup(s->create_type), 3010 .cid = s->cid, 3011 .parent_cid = s->parent_cid, 3012 }; 3013 3014 tail = &spec_info->u.vmdk.data->extents; 3015 for (i = 0; i < s->num_extents; i++) { 3016 QAPI_LIST_APPEND(tail, vmdk_get_extent_info(&s->extents[i])); 3017 } 3018 3019 return spec_info; 3020 } 3021 3022 static bool vmdk_extents_type_eq(const VmdkExtent *a, const VmdkExtent *b) 3023 { 3024 return a->flat == b->flat && 3025 a->compressed == b->compressed && 3026 (a->flat || a->cluster_sectors == b->cluster_sectors); 3027 } 3028 3029 static int coroutine_fn 3030 vmdk_co_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) 3031 { 3032 int i; 3033 BDRVVmdkState *s = bs->opaque; 3034 assert(s->num_extents); 3035 3036 /* See if we have multiple extents but they have different cases */ 3037 for (i = 1; i < s->num_extents; i++) { 3038 if (!vmdk_extents_type_eq(&s->extents[0], &s->extents[i])) { 3039 return -ENOTSUP; 3040 } 3041 } 3042 bdi->needs_compressed_writes = s->extents[0].compressed; 3043 if (!s->extents[0].flat) { 3044 bdi->cluster_size = s->extents[0].cluster_sectors << BDRV_SECTOR_BITS; 3045 } 3046 return 0; 3047 } 3048 3049 static void vmdk_gather_child_options(BlockDriverState *bs, QDict *target, 3050 bool backing_overridden) 3051 { 3052 /* No children but file and backing can be explicitly specified (TODO) */ 3053 qdict_put(target, "file", 3054 qobject_ref(bs->file->bs->full_open_options)); 3055 3056 if (backing_overridden) { 3057 if (bs->backing) { 3058 qdict_put(target, "backing", 3059 qobject_ref(bs->backing->bs->full_open_options)); 3060 } else { 3061 qdict_put_null(target, "backing"); 3062 } 3063 } 3064 } 3065 3066 static QemuOptsList vmdk_create_opts = { 3067 .name = "vmdk-create-opts", 3068 .head = QTAILQ_HEAD_INITIALIZER(vmdk_create_opts.head), 3069 .desc = { 3070 { 3071 .name = BLOCK_OPT_SIZE, 3072 .type = QEMU_OPT_SIZE, 3073 .help = "Virtual disk size" 3074 }, 3075 { 3076 .name = BLOCK_OPT_ADAPTER_TYPE, 3077 .type = QEMU_OPT_STRING, 3078 .help = "Virtual adapter type, can be one of " 3079 "ide (default), lsilogic, buslogic or legacyESX" 3080 }, 3081 { 3082 .name = BLOCK_OPT_BACKING_FILE, 3083 .type = QEMU_OPT_STRING, 3084 .help = "File name of a base image" 3085 }, 3086 { 3087 .name = BLOCK_OPT_BACKING_FMT, 3088 .type = QEMU_OPT_STRING, 3089 .help = "Must be 'vmdk' if present", 3090 }, 3091 { 3092 .name = BLOCK_OPT_COMPAT6, 3093 .type = QEMU_OPT_BOOL, 3094 .help = "VMDK version 6 image", 3095 .def_value_str = "off" 3096 }, 3097 { 3098 .name = BLOCK_OPT_HWVERSION, 3099 .type = QEMU_OPT_STRING, 3100 .help = "VMDK hardware version", 3101 .def_value_str = "undefined" 3102 }, 3103 { 3104 .name = BLOCK_OPT_TOOLSVERSION, 3105 .type = QEMU_OPT_STRING, 3106 .help = "VMware guest tools version", 3107 }, 3108 { 3109 .name = BLOCK_OPT_SUBFMT, 3110 .type = QEMU_OPT_STRING, 3111 .help = 3112 "VMDK flat extent format, can be one of " 3113 "{monolithicSparse (default) | monolithicFlat | twoGbMaxExtentSparse | twoGbMaxExtentFlat | streamOptimized} " 3114 }, 3115 { 3116 .name = BLOCK_OPT_ZEROED_GRAIN, 3117 .type = QEMU_OPT_BOOL, 3118 .help = "Enable efficient zero writes " 3119 "using the zeroed-grain GTE feature" 3120 }, 3121 { /* end of list */ } 3122 } 3123 }; 3124 3125 static BlockDriver bdrv_vmdk = { 3126 .format_name = "vmdk", 3127 .instance_size = sizeof(BDRVVmdkState), 3128 .bdrv_probe = vmdk_probe, 3129 .bdrv_open = vmdk_open, 3130 .bdrv_co_check = vmdk_co_check, 3131 .bdrv_reopen_prepare = vmdk_reopen_prepare, 3132 .bdrv_reopen_commit = vmdk_reopen_commit, 3133 .bdrv_reopen_abort = vmdk_reopen_abort, 3134 .bdrv_child_perm = bdrv_default_perms, 3135 .bdrv_co_preadv = vmdk_co_preadv, 3136 .bdrv_co_pwritev = vmdk_co_pwritev, 3137 .bdrv_co_pwritev_compressed = vmdk_co_pwritev_compressed, 3138 .bdrv_co_pwrite_zeroes = vmdk_co_pwrite_zeroes, 3139 .bdrv_close = vmdk_close, 3140 .bdrv_co_create_opts = vmdk_co_create_opts, 3141 .bdrv_co_create = vmdk_co_create, 3142 .bdrv_co_block_status = vmdk_co_block_status, 3143 .bdrv_co_get_allocated_file_size = vmdk_co_get_allocated_file_size, 3144 .bdrv_has_zero_init = vmdk_has_zero_init, 3145 .bdrv_get_specific_info = vmdk_get_specific_info, 3146 .bdrv_refresh_limits = vmdk_refresh_limits, 3147 .bdrv_co_get_info = vmdk_co_get_info, 3148 .bdrv_gather_child_options = vmdk_gather_child_options, 3149 3150 .is_format = true, 3151 .supports_backing = true, 3152 .create_opts = &vmdk_create_opts, 3153 }; 3154 3155 static void bdrv_vmdk_init(void) 3156 { 3157 bdrv_register(&bdrv_vmdk); 3158 } 3159 3160 block_init(bdrv_vmdk_init); 3161