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