1 /* 2 * Block driver for Hyper-V VHDX Images 3 * 4 * Copyright (c) 2013 Red Hat, Inc., 5 * 6 * Authors: 7 * Jeff Cody <jcody@redhat.com> 8 * 9 * This is based on the "VHDX Format Specification v1.00", published 8/25/2012 10 * by Microsoft: 11 * https://www.microsoft.com/en-us/download/details.aspx?id=34750 12 * 13 * This work is licensed under the terms of the GNU LGPL, version 2 or later. 14 * See the COPYING.LIB file in the top-level directory. 15 * 16 */ 17 18 #include "qemu/osdep.h" 19 #include "qapi/error.h" 20 #include "block/block_int.h" 21 #include "block/qdict.h" 22 #include "sysemu/block-backend.h" 23 #include "qemu/module.h" 24 #include "qemu/option.h" 25 #include "qemu/crc32c.h" 26 #include "qemu/bswap.h" 27 #include "qemu/error-report.h" 28 #include "vhdx.h" 29 #include "migration/blocker.h" 30 #include "qemu/uuid.h" 31 #include "qapi/qmp/qdict.h" 32 #include "qapi/qobject-input-visitor.h" 33 #include "qapi/qapi-visit-block-core.h" 34 35 /* Options for VHDX creation */ 36 37 #define VHDX_BLOCK_OPT_LOG_SIZE "log_size" 38 #define VHDX_BLOCK_OPT_BLOCK_SIZE "block_size" 39 #define VHDX_BLOCK_OPT_ZERO "block_state_zero" 40 41 typedef enum VHDXImageType { 42 VHDX_TYPE_DYNAMIC = 0, 43 VHDX_TYPE_FIXED, 44 VHDX_TYPE_DIFFERENCING, /* Currently unsupported */ 45 } VHDXImageType; 46 47 static QemuOptsList vhdx_create_opts; 48 49 /* Several metadata and region table data entries are identified by 50 * guids in a MS-specific GUID format. */ 51 52 53 /* ------- Known Region Table GUIDs ---------------------- */ 54 static const MSGUID bat_guid = { .data1 = 0x2dc27766, 55 .data2 = 0xf623, 56 .data3 = 0x4200, 57 .data4 = { 0x9d, 0x64, 0x11, 0x5e, 58 0x9b, 0xfd, 0x4a, 0x08} }; 59 60 static const MSGUID metadata_guid = { .data1 = 0x8b7ca206, 61 .data2 = 0x4790, 62 .data3 = 0x4b9a, 63 .data4 = { 0xb8, 0xfe, 0x57, 0x5f, 64 0x05, 0x0f, 0x88, 0x6e} }; 65 66 67 68 /* ------- Known Metadata Entry GUIDs ---------------------- */ 69 static const MSGUID file_param_guid = { .data1 = 0xcaa16737, 70 .data2 = 0xfa36, 71 .data3 = 0x4d43, 72 .data4 = { 0xb3, 0xb6, 0x33, 0xf0, 73 0xaa, 0x44, 0xe7, 0x6b} }; 74 75 static const MSGUID virtual_size_guid = { .data1 = 0x2FA54224, 76 .data2 = 0xcd1b, 77 .data3 = 0x4876, 78 .data4 = { 0xb2, 0x11, 0x5d, 0xbe, 79 0xd8, 0x3b, 0xf4, 0xb8} }; 80 81 static const MSGUID page83_guid = { .data1 = 0xbeca12ab, 82 .data2 = 0xb2e6, 83 .data3 = 0x4523, 84 .data4 = { 0x93, 0xef, 0xc3, 0x09, 85 0xe0, 0x00, 0xc7, 0x46} }; 86 87 88 static const MSGUID phys_sector_guid = { .data1 = 0xcda348c7, 89 .data2 = 0x445d, 90 .data3 = 0x4471, 91 .data4 = { 0x9c, 0xc9, 0xe9, 0x88, 92 0x52, 0x51, 0xc5, 0x56} }; 93 94 static const MSGUID parent_locator_guid = { .data1 = 0xa8d35f2d, 95 .data2 = 0xb30b, 96 .data3 = 0x454d, 97 .data4 = { 0xab, 0xf7, 0xd3, 98 0xd8, 0x48, 0x34, 99 0xab, 0x0c} }; 100 101 static const MSGUID logical_sector_guid = { .data1 = 0x8141bf1d, 102 .data2 = 0xa96f, 103 .data3 = 0x4709, 104 .data4 = { 0xba, 0x47, 0xf2, 105 0x33, 0xa8, 0xfa, 106 0xab, 0x5f} }; 107 108 /* Each parent type must have a valid GUID; this is for parent images 109 * of type 'VHDX'. If we were to allow e.g. a QCOW2 parent, we would 110 * need to make up our own QCOW2 GUID type */ 111 static const MSGUID parent_vhdx_guid __attribute__((unused)) 112 = { .data1 = 0xb04aefb7, 113 .data2 = 0xd19e, 114 .data3 = 0x4a81, 115 .data4 = { 0xb7, 0x89, 0x25, 0xb8, 116 0xe9, 0x44, 0x59, 0x13} }; 117 118 119 #define META_FILE_PARAMETER_PRESENT 0x01 120 #define META_VIRTUAL_DISK_SIZE_PRESENT 0x02 121 #define META_PAGE_83_PRESENT 0x04 122 #define META_LOGICAL_SECTOR_SIZE_PRESENT 0x08 123 #define META_PHYS_SECTOR_SIZE_PRESENT 0x10 124 #define META_PARENT_LOCATOR_PRESENT 0x20 125 126 #define META_ALL_PRESENT \ 127 (META_FILE_PARAMETER_PRESENT | META_VIRTUAL_DISK_SIZE_PRESENT | \ 128 META_PAGE_83_PRESENT | META_LOGICAL_SECTOR_SIZE_PRESENT | \ 129 META_PHYS_SECTOR_SIZE_PRESENT) 130 131 132 typedef struct VHDXSectorInfo { 133 uint32_t bat_idx; /* BAT entry index */ 134 uint32_t sectors_avail; /* sectors available in payload block */ 135 uint32_t bytes_left; /* bytes left in the block after data to r/w */ 136 uint32_t bytes_avail; /* bytes available in payload block */ 137 uint64_t file_offset; /* absolute offset in bytes, in file */ 138 uint64_t block_offset; /* block offset, in bytes */ 139 } VHDXSectorInfo; 140 141 /* Calculates new checksum. 142 * 143 * Zero is substituted during crc calculation for the original crc field 144 * crc_offset: byte offset in buf of the buffer crc 145 * buf: buffer pointer 146 * size: size of buffer (must be > crc_offset+4) 147 * 148 * Note: The buffer should have all multi-byte data in little-endian format, 149 * and the resulting checksum is in little endian format. 150 */ 151 uint32_t vhdx_update_checksum(uint8_t *buf, size_t size, int crc_offset) 152 { 153 uint32_t crc; 154 155 assert(buf != NULL); 156 assert(size > (crc_offset + sizeof(crc))); 157 158 memset(buf + crc_offset, 0, sizeof(crc)); 159 crc = crc32c(0xffffffff, buf, size); 160 crc = cpu_to_le32(crc); 161 memcpy(buf + crc_offset, &crc, sizeof(crc)); 162 163 return crc; 164 } 165 166 uint32_t vhdx_checksum_calc(uint32_t crc, uint8_t *buf, size_t size, 167 int crc_offset) 168 { 169 uint32_t crc_new; 170 uint32_t crc_orig; 171 assert(buf != NULL); 172 173 if (crc_offset > 0) { 174 memcpy(&crc_orig, buf + crc_offset, sizeof(crc_orig)); 175 memset(buf + crc_offset, 0, sizeof(crc_orig)); 176 } 177 178 crc_new = crc32c(crc, buf, size); 179 if (crc_offset > 0) { 180 memcpy(buf + crc_offset, &crc_orig, sizeof(crc_orig)); 181 } 182 183 return crc_new; 184 } 185 186 /* Validates the checksum of the buffer, with an in-place CRC. 187 * 188 * Zero is substituted during crc calculation for the original crc field, 189 * and the crc field is restored afterwards. But the buffer will be modified 190 * during the calculation, so this may not be not suitable for multi-threaded 191 * use. 192 * 193 * crc_offset: byte offset in buf of the buffer crc 194 * buf: buffer pointer 195 * size: size of buffer (must be > crc_offset+4) 196 * 197 * returns true if checksum is valid, false otherwise 198 */ 199 bool vhdx_checksum_is_valid(uint8_t *buf, size_t size, int crc_offset) 200 { 201 uint32_t crc_orig; 202 uint32_t crc; 203 204 assert(buf != NULL); 205 assert(size > (crc_offset + 4)); 206 207 memcpy(&crc_orig, buf + crc_offset, sizeof(crc_orig)); 208 crc_orig = le32_to_cpu(crc_orig); 209 210 crc = vhdx_checksum_calc(0xffffffff, buf, size, crc_offset); 211 212 return crc == crc_orig; 213 } 214 215 216 /* 217 * This generates a UUID that is compliant with the MS GUIDs used 218 * in the VHDX spec (and elsewhere). 219 */ 220 void vhdx_guid_generate(MSGUID *guid) 221 { 222 QemuUUID uuid; 223 assert(guid != NULL); 224 225 qemu_uuid_generate(&uuid); 226 memcpy(guid, &uuid, sizeof(MSGUID)); 227 } 228 229 /* Check for region overlaps inside the VHDX image */ 230 static int vhdx_region_check(BDRVVHDXState *s, uint64_t start, uint64_t length) 231 { 232 int ret = 0; 233 uint64_t end; 234 VHDXRegionEntry *r; 235 236 end = start + length; 237 QLIST_FOREACH(r, &s->regions, entries) { 238 if (!((start >= r->end) || (end <= r->start))) { 239 error_report("VHDX region %" PRIu64 "-%" PRIu64 " overlaps with " 240 "region %" PRIu64 "-%." PRIu64, start, end, r->start, 241 r->end); 242 ret = -EINVAL; 243 goto exit; 244 } 245 } 246 247 exit: 248 return ret; 249 } 250 251 /* Register a region for future checks */ 252 static void vhdx_region_register(BDRVVHDXState *s, 253 uint64_t start, uint64_t length) 254 { 255 VHDXRegionEntry *r; 256 257 r = g_malloc0(sizeof(*r)); 258 259 r->start = start; 260 r->end = start + length; 261 262 QLIST_INSERT_HEAD(&s->regions, r, entries); 263 } 264 265 /* Free all registered regions */ 266 static void vhdx_region_unregister_all(BDRVVHDXState *s) 267 { 268 VHDXRegionEntry *r, *r_next; 269 270 QLIST_FOREACH_SAFE(r, &s->regions, entries, r_next) { 271 QLIST_REMOVE(r, entries); 272 g_free(r); 273 } 274 } 275 276 static void vhdx_set_shift_bits(BDRVVHDXState *s) 277 { 278 s->logical_sector_size_bits = ctz32(s->logical_sector_size); 279 s->sectors_per_block_bits = ctz32(s->sectors_per_block); 280 s->chunk_ratio_bits = ctz64(s->chunk_ratio); 281 s->block_size_bits = ctz32(s->block_size); 282 } 283 284 /* 285 * Per the MS VHDX Specification, for every VHDX file: 286 * - The header section is fixed size - 1 MB 287 * - The header section is always the first "object" 288 * - The first 64KB of the header is the File Identifier 289 * - The first uint64 (8 bytes) is the VHDX Signature ("vhdxfile") 290 * - The following 512 bytes constitute a UTF-16 string identifiying the 291 * software that created the file, and is optional and diagnostic only. 292 * 293 * Therefore, we probe by looking for the vhdxfile signature "vhdxfile" 294 */ 295 static int vhdx_probe(const uint8_t *buf, int buf_size, const char *filename) 296 { 297 if (buf_size >= 8 && !memcmp(buf, "vhdxfile", 8)) { 298 return 100; 299 } 300 return 0; 301 } 302 303 /* 304 * Writes the header to the specified offset. 305 * 306 * This will optionally read in buffer data from disk (otherwise zero-fill), 307 * and then update the header checksum. Header is converted to proper 308 * endianness before being written to the specified file offset 309 */ 310 static int vhdx_write_header(BdrvChild *file, VHDXHeader *hdr, 311 uint64_t offset, bool read) 312 { 313 BlockDriverState *bs_file = file->bs; 314 uint8_t *buffer = NULL; 315 int ret; 316 VHDXHeader *header_le; 317 318 assert(bs_file != NULL); 319 assert(hdr != NULL); 320 321 /* the header checksum is not over just the packed size of VHDXHeader, 322 * but rather over the entire 'reserved' range for the header, which is 323 * 4KB (VHDX_HEADER_SIZE). */ 324 325 buffer = qemu_blockalign(bs_file, VHDX_HEADER_SIZE); 326 if (read) { 327 /* if true, we can't assume the extra reserved bytes are 0 */ 328 ret = bdrv_pread(file, offset, buffer, VHDX_HEADER_SIZE); 329 if (ret < 0) { 330 goto exit; 331 } 332 } else { 333 memset(buffer, 0, VHDX_HEADER_SIZE); 334 } 335 336 /* overwrite the actual VHDXHeader portion */ 337 header_le = (VHDXHeader *)buffer; 338 memcpy(header_le, hdr, sizeof(VHDXHeader)); 339 vhdx_header_le_export(hdr, header_le); 340 vhdx_update_checksum(buffer, VHDX_HEADER_SIZE, 341 offsetof(VHDXHeader, checksum)); 342 ret = bdrv_pwrite_sync(file, offset, header_le, sizeof(VHDXHeader)); 343 344 exit: 345 qemu_vfree(buffer); 346 return ret; 347 } 348 349 /* Update the VHDX headers 350 * 351 * This follows the VHDX spec procedures for header updates. 352 * 353 * - non-current header is updated with largest sequence number 354 */ 355 static int vhdx_update_header(BlockDriverState *bs, BDRVVHDXState *s, 356 bool generate_data_write_guid, MSGUID *log_guid) 357 { 358 int ret = 0; 359 int hdr_idx = 0; 360 uint64_t header_offset = VHDX_HEADER1_OFFSET; 361 362 VHDXHeader *active_header; 363 VHDXHeader *inactive_header; 364 365 /* operate on the non-current header */ 366 if (s->curr_header == 0) { 367 hdr_idx = 1; 368 header_offset = VHDX_HEADER2_OFFSET; 369 } 370 371 active_header = s->headers[s->curr_header]; 372 inactive_header = s->headers[hdr_idx]; 373 374 inactive_header->sequence_number = active_header->sequence_number + 1; 375 376 /* a new file guid must be generated before any file write, including 377 * headers */ 378 inactive_header->file_write_guid = s->session_guid; 379 380 /* a new data guid only needs to be generated before any guest-visible 381 * writes (i.e. something observable via virtual disk read) */ 382 if (generate_data_write_guid) { 383 vhdx_guid_generate(&inactive_header->data_write_guid); 384 } 385 386 /* update the log guid if present */ 387 if (log_guid) { 388 inactive_header->log_guid = *log_guid; 389 } 390 391 ret = vhdx_write_header(bs->file, inactive_header, header_offset, true); 392 if (ret < 0) { 393 goto exit; 394 } 395 s->curr_header = hdr_idx; 396 397 exit: 398 return ret; 399 } 400 401 /* 402 * The VHDX spec calls for header updates to be performed twice, so that both 403 * the current and non-current header have valid info 404 */ 405 int vhdx_update_headers(BlockDriverState *bs, BDRVVHDXState *s, 406 bool generate_data_write_guid, MSGUID *log_guid) 407 { 408 int ret; 409 410 ret = vhdx_update_header(bs, s, generate_data_write_guid, log_guid); 411 if (ret < 0) { 412 return ret; 413 } 414 return vhdx_update_header(bs, s, generate_data_write_guid, log_guid); 415 } 416 417 /* opens the specified header block from the VHDX file header section */ 418 static void vhdx_parse_header(BlockDriverState *bs, BDRVVHDXState *s, 419 Error **errp) 420 { 421 int ret; 422 VHDXHeader *header1; 423 VHDXHeader *header2; 424 bool h1_valid = false; 425 bool h2_valid = false; 426 uint64_t h1_seq = 0; 427 uint64_t h2_seq = 0; 428 uint8_t *buffer; 429 430 /* header1 & header2 are freed in vhdx_close() */ 431 header1 = qemu_blockalign(bs, sizeof(VHDXHeader)); 432 header2 = qemu_blockalign(bs, sizeof(VHDXHeader)); 433 434 buffer = qemu_blockalign(bs, VHDX_HEADER_SIZE); 435 436 s->headers[0] = header1; 437 s->headers[1] = header2; 438 439 /* We have to read the whole VHDX_HEADER_SIZE instead of 440 * sizeof(VHDXHeader), because the checksum is over the whole 441 * region */ 442 ret = bdrv_pread(bs->file, VHDX_HEADER1_OFFSET, buffer, 443 VHDX_HEADER_SIZE); 444 if (ret < 0) { 445 goto fail; 446 } 447 /* copy over just the relevant portion that we need */ 448 memcpy(header1, buffer, sizeof(VHDXHeader)); 449 450 if (vhdx_checksum_is_valid(buffer, VHDX_HEADER_SIZE, 4)) { 451 vhdx_header_le_import(header1); 452 if (header1->signature == VHDX_HEADER_SIGNATURE && 453 header1->version == 1) { 454 h1_seq = header1->sequence_number; 455 h1_valid = true; 456 } 457 } 458 459 ret = bdrv_pread(bs->file, VHDX_HEADER2_OFFSET, buffer, 460 VHDX_HEADER_SIZE); 461 if (ret < 0) { 462 goto fail; 463 } 464 /* copy over just the relevant portion that we need */ 465 memcpy(header2, buffer, sizeof(VHDXHeader)); 466 467 if (vhdx_checksum_is_valid(buffer, VHDX_HEADER_SIZE, 4)) { 468 vhdx_header_le_import(header2); 469 if (header2->signature == VHDX_HEADER_SIGNATURE && 470 header2->version == 1) { 471 h2_seq = header2->sequence_number; 472 h2_valid = true; 473 } 474 } 475 476 /* If there is only 1 valid header (or no valid headers), we 477 * don't care what the sequence numbers are */ 478 if (h1_valid && !h2_valid) { 479 s->curr_header = 0; 480 } else if (!h1_valid && h2_valid) { 481 s->curr_header = 1; 482 } else if (!h1_valid && !h2_valid) { 483 goto fail; 484 } else { 485 /* If both headers are valid, then we choose the active one by the 486 * highest sequence number. If the sequence numbers are equal, that is 487 * invalid */ 488 if (h1_seq > h2_seq) { 489 s->curr_header = 0; 490 } else if (h2_seq > h1_seq) { 491 s->curr_header = 1; 492 } else { 493 /* The Microsoft Disk2VHD tool will create 2 identical 494 * headers, with identical sequence numbers. If the headers are 495 * identical, don't consider the file corrupt */ 496 if (!memcmp(header1, header2, sizeof(VHDXHeader))) { 497 s->curr_header = 0; 498 } else { 499 goto fail; 500 } 501 } 502 } 503 504 vhdx_region_register(s, s->headers[s->curr_header]->log_offset, 505 s->headers[s->curr_header]->log_length); 506 goto exit; 507 508 fail: 509 error_setg_errno(errp, -ret, "No valid VHDX header found"); 510 qemu_vfree(header1); 511 qemu_vfree(header2); 512 s->headers[0] = NULL; 513 s->headers[1] = NULL; 514 exit: 515 qemu_vfree(buffer); 516 } 517 518 519 static int vhdx_open_region_tables(BlockDriverState *bs, BDRVVHDXState *s) 520 { 521 int ret = 0; 522 uint8_t *buffer; 523 int offset = 0; 524 VHDXRegionTableEntry rt_entry; 525 uint32_t i; 526 bool bat_rt_found = false; 527 bool metadata_rt_found = false; 528 529 /* We have to read the whole 64KB block, because the crc32 is over the 530 * whole block */ 531 buffer = qemu_blockalign(bs, VHDX_HEADER_BLOCK_SIZE); 532 533 ret = bdrv_pread(bs->file, VHDX_REGION_TABLE_OFFSET, buffer, 534 VHDX_HEADER_BLOCK_SIZE); 535 if (ret < 0) { 536 goto fail; 537 } 538 memcpy(&s->rt, buffer, sizeof(s->rt)); 539 offset += sizeof(s->rt); 540 541 if (!vhdx_checksum_is_valid(buffer, VHDX_HEADER_BLOCK_SIZE, 4)) { 542 ret = -EINVAL; 543 goto fail; 544 } 545 546 vhdx_region_header_le_import(&s->rt); 547 548 if (s->rt.signature != VHDX_REGION_SIGNATURE) { 549 ret = -EINVAL; 550 goto fail; 551 } 552 553 554 /* Per spec, maximum region table entry count is 2047 */ 555 if (s->rt.entry_count > 2047) { 556 ret = -EINVAL; 557 goto fail; 558 } 559 560 for (i = 0; i < s->rt.entry_count; i++) { 561 memcpy(&rt_entry, buffer + offset, sizeof(rt_entry)); 562 offset += sizeof(rt_entry); 563 564 vhdx_region_entry_le_import(&rt_entry); 565 566 /* check for region overlap between these entries, and any 567 * other memory regions in the file */ 568 ret = vhdx_region_check(s, rt_entry.file_offset, rt_entry.length); 569 if (ret < 0) { 570 goto fail; 571 } 572 573 vhdx_region_register(s, rt_entry.file_offset, rt_entry.length); 574 575 /* see if we recognize the entry */ 576 if (guid_eq(rt_entry.guid, bat_guid)) { 577 /* must be unique; if we have already found it this is invalid */ 578 if (bat_rt_found) { 579 ret = -EINVAL; 580 goto fail; 581 } 582 bat_rt_found = true; 583 s->bat_rt = rt_entry; 584 continue; 585 } 586 587 if (guid_eq(rt_entry.guid, metadata_guid)) { 588 /* must be unique; if we have already found it this is invalid */ 589 if (metadata_rt_found) { 590 ret = -EINVAL; 591 goto fail; 592 } 593 metadata_rt_found = true; 594 s->metadata_rt = rt_entry; 595 continue; 596 } 597 598 if (rt_entry.data_bits & VHDX_REGION_ENTRY_REQUIRED) { 599 /* cannot read vhdx file - required region table entry that 600 * we do not understand. per spec, we must fail to open */ 601 ret = -ENOTSUP; 602 goto fail; 603 } 604 } 605 606 if (!bat_rt_found || !metadata_rt_found) { 607 ret = -EINVAL; 608 goto fail; 609 } 610 611 ret = 0; 612 613 fail: 614 qemu_vfree(buffer); 615 return ret; 616 } 617 618 619 620 /* Metadata initial parser 621 * 622 * This loads all the metadata entry fields. This may cause additional 623 * fields to be processed (e.g. parent locator, etc..). 624 * 625 * There are 5 Metadata items that are always required: 626 * - File Parameters (block size, has a parent) 627 * - Virtual Disk Size (size, in bytes, of the virtual drive) 628 * - Page 83 Data (scsi page 83 guid) 629 * - Logical Sector Size (logical sector size in bytes, either 512 or 630 * 4096. We only support 512 currently) 631 * - Physical Sector Size (512 or 4096) 632 * 633 * Also, if the File Parameters indicate this is a differencing file, 634 * we must also look for the Parent Locator metadata item. 635 */ 636 static int vhdx_parse_metadata(BlockDriverState *bs, BDRVVHDXState *s) 637 { 638 int ret = 0; 639 uint8_t *buffer; 640 int offset = 0; 641 uint32_t i = 0; 642 VHDXMetadataTableEntry md_entry; 643 644 buffer = qemu_blockalign(bs, VHDX_METADATA_TABLE_MAX_SIZE); 645 646 ret = bdrv_pread(bs->file, s->metadata_rt.file_offset, buffer, 647 VHDX_METADATA_TABLE_MAX_SIZE); 648 if (ret < 0) { 649 goto exit; 650 } 651 memcpy(&s->metadata_hdr, buffer, sizeof(s->metadata_hdr)); 652 offset += sizeof(s->metadata_hdr); 653 654 vhdx_metadata_header_le_import(&s->metadata_hdr); 655 656 if (s->metadata_hdr.signature != VHDX_METADATA_SIGNATURE) { 657 ret = -EINVAL; 658 goto exit; 659 } 660 661 s->metadata_entries.present = 0; 662 663 if ((s->metadata_hdr.entry_count * sizeof(md_entry)) > 664 (VHDX_METADATA_TABLE_MAX_SIZE - offset)) { 665 ret = -EINVAL; 666 goto exit; 667 } 668 669 for (i = 0; i < s->metadata_hdr.entry_count; i++) { 670 memcpy(&md_entry, buffer + offset, sizeof(md_entry)); 671 offset += sizeof(md_entry); 672 673 vhdx_metadata_entry_le_import(&md_entry); 674 675 if (guid_eq(md_entry.item_id, file_param_guid)) { 676 if (s->metadata_entries.present & META_FILE_PARAMETER_PRESENT) { 677 ret = -EINVAL; 678 goto exit; 679 } 680 s->metadata_entries.file_parameters_entry = md_entry; 681 s->metadata_entries.present |= META_FILE_PARAMETER_PRESENT; 682 continue; 683 } 684 685 if (guid_eq(md_entry.item_id, virtual_size_guid)) { 686 if (s->metadata_entries.present & META_VIRTUAL_DISK_SIZE_PRESENT) { 687 ret = -EINVAL; 688 goto exit; 689 } 690 s->metadata_entries.virtual_disk_size_entry = md_entry; 691 s->metadata_entries.present |= META_VIRTUAL_DISK_SIZE_PRESENT; 692 continue; 693 } 694 695 if (guid_eq(md_entry.item_id, page83_guid)) { 696 if (s->metadata_entries.present & META_PAGE_83_PRESENT) { 697 ret = -EINVAL; 698 goto exit; 699 } 700 s->metadata_entries.page83_data_entry = md_entry; 701 s->metadata_entries.present |= META_PAGE_83_PRESENT; 702 continue; 703 } 704 705 if (guid_eq(md_entry.item_id, logical_sector_guid)) { 706 if (s->metadata_entries.present & 707 META_LOGICAL_SECTOR_SIZE_PRESENT) { 708 ret = -EINVAL; 709 goto exit; 710 } 711 s->metadata_entries.logical_sector_size_entry = md_entry; 712 s->metadata_entries.present |= META_LOGICAL_SECTOR_SIZE_PRESENT; 713 continue; 714 } 715 716 if (guid_eq(md_entry.item_id, phys_sector_guid)) { 717 if (s->metadata_entries.present & META_PHYS_SECTOR_SIZE_PRESENT) { 718 ret = -EINVAL; 719 goto exit; 720 } 721 s->metadata_entries.phys_sector_size_entry = md_entry; 722 s->metadata_entries.present |= META_PHYS_SECTOR_SIZE_PRESENT; 723 continue; 724 } 725 726 if (guid_eq(md_entry.item_id, parent_locator_guid)) { 727 if (s->metadata_entries.present & META_PARENT_LOCATOR_PRESENT) { 728 ret = -EINVAL; 729 goto exit; 730 } 731 s->metadata_entries.parent_locator_entry = md_entry; 732 s->metadata_entries.present |= META_PARENT_LOCATOR_PRESENT; 733 continue; 734 } 735 736 if (md_entry.data_bits & VHDX_META_FLAGS_IS_REQUIRED) { 737 /* cannot read vhdx file - required region table entry that 738 * we do not understand. per spec, we must fail to open */ 739 ret = -ENOTSUP; 740 goto exit; 741 } 742 } 743 744 if (s->metadata_entries.present != META_ALL_PRESENT) { 745 ret = -ENOTSUP; 746 goto exit; 747 } 748 749 ret = bdrv_pread(bs->file, 750 s->metadata_entries.file_parameters_entry.offset 751 + s->metadata_rt.file_offset, 752 &s->params, 753 sizeof(s->params)); 754 755 if (ret < 0) { 756 goto exit; 757 } 758 759 s->params.block_size = le32_to_cpu(s->params.block_size); 760 s->params.data_bits = le32_to_cpu(s->params.data_bits); 761 762 763 /* We now have the file parameters, so we can tell if this is a 764 * differencing file (i.e.. has_parent), is dynamic or fixed 765 * sized (leave_blocks_allocated), and the block size */ 766 767 /* The parent locator required iff the file parameters has_parent set */ 768 if (s->params.data_bits & VHDX_PARAMS_HAS_PARENT) { 769 if (s->metadata_entries.present & META_PARENT_LOCATOR_PRESENT) { 770 /* TODO: parse parent locator fields */ 771 ret = -ENOTSUP; /* temp, until differencing files are supported */ 772 goto exit; 773 } else { 774 /* if has_parent is set, but there is not parent locator present, 775 * then that is an invalid combination */ 776 ret = -EINVAL; 777 goto exit; 778 } 779 } 780 781 /* determine virtual disk size, logical sector size, 782 * and phys sector size */ 783 784 ret = bdrv_pread(bs->file, 785 s->metadata_entries.virtual_disk_size_entry.offset 786 + s->metadata_rt.file_offset, 787 &s->virtual_disk_size, 788 sizeof(uint64_t)); 789 if (ret < 0) { 790 goto exit; 791 } 792 ret = bdrv_pread(bs->file, 793 s->metadata_entries.logical_sector_size_entry.offset 794 + s->metadata_rt.file_offset, 795 &s->logical_sector_size, 796 sizeof(uint32_t)); 797 if (ret < 0) { 798 goto exit; 799 } 800 ret = bdrv_pread(bs->file, 801 s->metadata_entries.phys_sector_size_entry.offset 802 + s->metadata_rt.file_offset, 803 &s->physical_sector_size, 804 sizeof(uint32_t)); 805 if (ret < 0) { 806 goto exit; 807 } 808 809 s->virtual_disk_size = le64_to_cpu(s->virtual_disk_size); 810 s->logical_sector_size = le32_to_cpu(s->logical_sector_size); 811 s->physical_sector_size = le32_to_cpu(s->physical_sector_size); 812 813 if (s->params.block_size < VHDX_BLOCK_SIZE_MIN || 814 s->params.block_size > VHDX_BLOCK_SIZE_MAX) { 815 ret = -EINVAL; 816 goto exit; 817 } 818 819 /* only 2 supported sector sizes */ 820 if (s->logical_sector_size != 512 && s->logical_sector_size != 4096) { 821 ret = -EINVAL; 822 goto exit; 823 } 824 825 /* Both block_size and sector_size are guaranteed powers of 2, below. 826 Due to range checks above, s->sectors_per_block can never be < 256 */ 827 s->sectors_per_block = s->params.block_size / s->logical_sector_size; 828 s->chunk_ratio = (VHDX_MAX_SECTORS_PER_BLOCK) * 829 (uint64_t)s->logical_sector_size / 830 (uint64_t)s->params.block_size; 831 832 /* These values are ones we will want to use for division / multiplication 833 * later on, and they are all guaranteed (per the spec) to be powers of 2, 834 * so we can take advantage of that for shift operations during 835 * reads/writes */ 836 if (s->logical_sector_size & (s->logical_sector_size - 1)) { 837 ret = -EINVAL; 838 goto exit; 839 } 840 if (s->sectors_per_block & (s->sectors_per_block - 1)) { 841 ret = -EINVAL; 842 goto exit; 843 } 844 if (s->chunk_ratio & (s->chunk_ratio - 1)) { 845 ret = -EINVAL; 846 goto exit; 847 } 848 s->block_size = s->params.block_size; 849 if (s->block_size & (s->block_size - 1)) { 850 ret = -EINVAL; 851 goto exit; 852 } 853 854 vhdx_set_shift_bits(s); 855 856 ret = 0; 857 858 exit: 859 qemu_vfree(buffer); 860 return ret; 861 } 862 863 /* 864 * Calculate the number of BAT entries, including sector 865 * bitmap entries. 866 */ 867 static void vhdx_calc_bat_entries(BDRVVHDXState *s) 868 { 869 uint32_t data_blocks_cnt, bitmap_blocks_cnt; 870 871 data_blocks_cnt = DIV_ROUND_UP(s->virtual_disk_size, s->block_size); 872 bitmap_blocks_cnt = DIV_ROUND_UP(data_blocks_cnt, s->chunk_ratio); 873 874 if (s->parent_entries) { 875 s->bat_entries = bitmap_blocks_cnt * (s->chunk_ratio + 1); 876 } else { 877 s->bat_entries = data_blocks_cnt + 878 ((data_blocks_cnt - 1) >> s->chunk_ratio_bits); 879 } 880 881 } 882 883 static int vhdx_check_bat_entries(BlockDriverState *bs, int *errcnt) 884 { 885 BDRVVHDXState *s = bs->opaque; 886 int64_t image_file_size = bdrv_getlength(bs->file->bs); 887 uint64_t payblocks = s->chunk_ratio; 888 uint64_t i; 889 int ret = 0; 890 891 if (image_file_size < 0) { 892 error_report("Could not determinate VHDX image file size."); 893 return image_file_size; 894 } 895 896 for (i = 0; i < s->bat_entries; i++) { 897 if ((s->bat[i] & VHDX_BAT_STATE_BIT_MASK) == 898 PAYLOAD_BLOCK_FULLY_PRESENT) { 899 uint64_t offset = s->bat[i] & VHDX_BAT_FILE_OFF_MASK; 900 /* 901 * Allow that the last block exists only partially. The VHDX spec 902 * states that the image file can only grow in blocksize increments, 903 * but QEMU created images with partial last blocks in the past. 904 */ 905 uint32_t block_length = MIN(s->block_size, 906 bs->total_sectors * BDRV_SECTOR_SIZE - i * s->block_size); 907 /* 908 * Check for BAT entry overflow. 909 */ 910 if (offset > INT64_MAX - s->block_size) { 911 error_report("VHDX BAT entry %" PRIu64 " offset overflow.", i); 912 ret = -EINVAL; 913 if (!errcnt) { 914 break; 915 } 916 (*errcnt)++; 917 } 918 /* 919 * Check if fully allocated BAT entries do not reside after 920 * end of the image file. 921 */ 922 if (offset >= image_file_size) { 923 error_report("VHDX BAT entry %" PRIu64 " start offset %" PRIu64 924 " points after end of file (%" PRIi64 "). Image" 925 " has probably been truncated.", 926 i, offset, image_file_size); 927 ret = -EINVAL; 928 if (!errcnt) { 929 break; 930 } 931 (*errcnt)++; 932 } else if (offset + block_length > image_file_size) { 933 error_report("VHDX BAT entry %" PRIu64 " end offset %" PRIu64 934 " points after end of file (%" PRIi64 "). Image" 935 " has probably been truncated.", 936 i, offset + block_length - 1, image_file_size); 937 ret = -EINVAL; 938 if (!errcnt) { 939 break; 940 } 941 (*errcnt)++; 942 } 943 944 /* 945 * verify populated BAT field file offsets against 946 * region table and log entries 947 */ 948 if (payblocks--) { 949 /* payload bat entries */ 950 int ret2; 951 ret2 = vhdx_region_check(s, offset, s->block_size); 952 if (ret2 < 0) { 953 ret = -EINVAL; 954 if (!errcnt) { 955 break; 956 } 957 (*errcnt)++; 958 } 959 } else { 960 payblocks = s->chunk_ratio; 961 /* 962 * Once differencing files are supported, verify sector bitmap 963 * blocks here 964 */ 965 } 966 } 967 } 968 969 return ret; 970 } 971 972 static void vhdx_close(BlockDriverState *bs) 973 { 974 BDRVVHDXState *s = bs->opaque; 975 qemu_vfree(s->headers[0]); 976 s->headers[0] = NULL; 977 qemu_vfree(s->headers[1]); 978 s->headers[1] = NULL; 979 qemu_vfree(s->bat); 980 s->bat = NULL; 981 qemu_vfree(s->parent_entries); 982 s->parent_entries = NULL; 983 migrate_del_blocker(s->migration_blocker); 984 error_free(s->migration_blocker); 985 qemu_vfree(s->log.hdr); 986 s->log.hdr = NULL; 987 vhdx_region_unregister_all(s); 988 } 989 990 static int vhdx_open(BlockDriverState *bs, QDict *options, int flags, 991 Error **errp) 992 { 993 BDRVVHDXState *s = bs->opaque; 994 int ret = 0; 995 uint32_t i; 996 uint64_t signature; 997 Error *local_err = NULL; 998 999 bs->file = bdrv_open_child(NULL, options, "file", bs, &child_file, 1000 false, errp); 1001 if (!bs->file) { 1002 return -EINVAL; 1003 } 1004 1005 s->bat = NULL; 1006 s->first_visible_write = true; 1007 1008 qemu_co_mutex_init(&s->lock); 1009 QLIST_INIT(&s->regions); 1010 1011 /* validate the file signature */ 1012 ret = bdrv_pread(bs->file, 0, &signature, sizeof(uint64_t)); 1013 if (ret < 0) { 1014 goto fail; 1015 } 1016 if (memcmp(&signature, "vhdxfile", 8)) { 1017 ret = -EINVAL; 1018 goto fail; 1019 } 1020 1021 /* This is used for any header updates, for the file_write_guid. 1022 * The spec dictates that a new value should be used for the first 1023 * header update */ 1024 vhdx_guid_generate(&s->session_guid); 1025 1026 vhdx_parse_header(bs, s, &local_err); 1027 if (local_err != NULL) { 1028 error_propagate(errp, local_err); 1029 ret = -EINVAL; 1030 goto fail; 1031 } 1032 1033 ret = vhdx_parse_log(bs, s, &s->log_replayed_on_open, errp); 1034 if (ret < 0) { 1035 goto fail; 1036 } 1037 1038 ret = vhdx_open_region_tables(bs, s); 1039 if (ret < 0) { 1040 goto fail; 1041 } 1042 1043 ret = vhdx_parse_metadata(bs, s); 1044 if (ret < 0) { 1045 goto fail; 1046 } 1047 1048 s->block_size = s->params.block_size; 1049 1050 /* the VHDX spec dictates that virtual_disk_size is always a multiple of 1051 * logical_sector_size */ 1052 bs->total_sectors = s->virtual_disk_size >> s->logical_sector_size_bits; 1053 1054 vhdx_calc_bat_entries(s); 1055 1056 s->bat_offset = s->bat_rt.file_offset; 1057 1058 if (s->bat_entries > s->bat_rt.length / sizeof(VHDXBatEntry)) { 1059 /* BAT allocation is not large enough for all entries */ 1060 ret = -EINVAL; 1061 goto fail; 1062 } 1063 1064 /* s->bat is freed in vhdx_close() */ 1065 s->bat = qemu_try_blockalign(bs->file->bs, s->bat_rt.length); 1066 if (s->bat == NULL) { 1067 ret = -ENOMEM; 1068 goto fail; 1069 } 1070 1071 ret = bdrv_pread(bs->file, s->bat_offset, s->bat, s->bat_rt.length); 1072 if (ret < 0) { 1073 goto fail; 1074 } 1075 1076 /* endian convert populated BAT field entires */ 1077 for (i = 0; i < s->bat_entries; i++) { 1078 s->bat[i] = le64_to_cpu(s->bat[i]); 1079 } 1080 1081 if (!(flags & BDRV_O_CHECK)) { 1082 ret = vhdx_check_bat_entries(bs, NULL); 1083 if (ret < 0) { 1084 goto fail; 1085 } 1086 } 1087 1088 /* Disable migration when VHDX images are used */ 1089 error_setg(&s->migration_blocker, "The vhdx format used by node '%s' " 1090 "does not support live migration", 1091 bdrv_get_device_or_node_name(bs)); 1092 ret = migrate_add_blocker(s->migration_blocker, &local_err); 1093 if (local_err) { 1094 error_propagate(errp, local_err); 1095 error_free(s->migration_blocker); 1096 goto fail; 1097 } 1098 1099 /* TODO: differencing files */ 1100 1101 return 0; 1102 fail: 1103 vhdx_close(bs); 1104 return ret; 1105 } 1106 1107 static int vhdx_reopen_prepare(BDRVReopenState *state, 1108 BlockReopenQueue *queue, Error **errp) 1109 { 1110 return 0; 1111 } 1112 1113 1114 /* 1115 * Perform sector to block offset translations, to get various 1116 * sector and file offsets into the image. See VHDXSectorInfo 1117 */ 1118 static void vhdx_block_translate(BDRVVHDXState *s, int64_t sector_num, 1119 int nb_sectors, VHDXSectorInfo *sinfo) 1120 { 1121 uint32_t block_offset; 1122 1123 sinfo->bat_idx = sector_num >> s->sectors_per_block_bits; 1124 /* effectively a modulo - this gives us the offset into the block 1125 * (in sector sizes) for our sector number */ 1126 block_offset = sector_num - (sinfo->bat_idx << s->sectors_per_block_bits); 1127 /* the chunk ratio gives us the interleaving of the sector 1128 * bitmaps, so we need to advance our page block index by the 1129 * sector bitmaps entry number */ 1130 sinfo->bat_idx += sinfo->bat_idx >> s->chunk_ratio_bits; 1131 1132 /* the number of sectors we can read/write in this cycle */ 1133 sinfo->sectors_avail = s->sectors_per_block - block_offset; 1134 1135 sinfo->bytes_left = sinfo->sectors_avail << s->logical_sector_size_bits; 1136 1137 if (sinfo->sectors_avail > nb_sectors) { 1138 sinfo->sectors_avail = nb_sectors; 1139 } 1140 1141 sinfo->bytes_avail = sinfo->sectors_avail << s->logical_sector_size_bits; 1142 1143 sinfo->file_offset = s->bat[sinfo->bat_idx] & VHDX_BAT_FILE_OFF_MASK; 1144 1145 sinfo->block_offset = block_offset << s->logical_sector_size_bits; 1146 1147 /* The file offset must be past the header section, so must be > 0 */ 1148 if (sinfo->file_offset == 0) { 1149 return; 1150 } 1151 1152 /* block offset is the offset in vhdx logical sectors, in 1153 * the payload data block. Convert that to a byte offset 1154 * in the block, and add in the payload data block offset 1155 * in the file, in bytes, to get the final read address */ 1156 1157 sinfo->file_offset += sinfo->block_offset; 1158 } 1159 1160 1161 static int vhdx_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) 1162 { 1163 BDRVVHDXState *s = bs->opaque; 1164 1165 bdi->cluster_size = s->block_size; 1166 1167 bdi->unallocated_blocks_are_zero = 1168 (s->params.data_bits & VHDX_PARAMS_HAS_PARENT) == 0; 1169 1170 return 0; 1171 } 1172 1173 1174 static coroutine_fn int vhdx_co_readv(BlockDriverState *bs, int64_t sector_num, 1175 int nb_sectors, QEMUIOVector *qiov) 1176 { 1177 BDRVVHDXState *s = bs->opaque; 1178 int ret = 0; 1179 VHDXSectorInfo sinfo; 1180 uint64_t bytes_done = 0; 1181 QEMUIOVector hd_qiov; 1182 1183 qemu_iovec_init(&hd_qiov, qiov->niov); 1184 1185 qemu_co_mutex_lock(&s->lock); 1186 1187 while (nb_sectors > 0) { 1188 /* We are a differencing file, so we need to inspect the sector bitmap 1189 * to see if we have the data or not */ 1190 if (s->params.data_bits & VHDX_PARAMS_HAS_PARENT) { 1191 /* not supported yet */ 1192 ret = -ENOTSUP; 1193 goto exit; 1194 } else { 1195 vhdx_block_translate(s, sector_num, nb_sectors, &sinfo); 1196 1197 qemu_iovec_reset(&hd_qiov); 1198 qemu_iovec_concat(&hd_qiov, qiov, bytes_done, sinfo.bytes_avail); 1199 1200 /* check the payload block state */ 1201 switch (s->bat[sinfo.bat_idx] & VHDX_BAT_STATE_BIT_MASK) { 1202 case PAYLOAD_BLOCK_NOT_PRESENT: /* fall through */ 1203 case PAYLOAD_BLOCK_UNDEFINED: 1204 case PAYLOAD_BLOCK_UNMAPPED: 1205 case PAYLOAD_BLOCK_UNMAPPED_v095: 1206 case PAYLOAD_BLOCK_ZERO: 1207 /* return zero */ 1208 qemu_iovec_memset(&hd_qiov, 0, 0, sinfo.bytes_avail); 1209 break; 1210 case PAYLOAD_BLOCK_FULLY_PRESENT: 1211 qemu_co_mutex_unlock(&s->lock); 1212 ret = bdrv_co_preadv(bs->file, sinfo.file_offset, 1213 sinfo.sectors_avail * BDRV_SECTOR_SIZE, 1214 &hd_qiov, 0); 1215 qemu_co_mutex_lock(&s->lock); 1216 if (ret < 0) { 1217 goto exit; 1218 } 1219 break; 1220 case PAYLOAD_BLOCK_PARTIALLY_PRESENT: 1221 /* we don't yet support difference files, fall through 1222 * to error */ 1223 default: 1224 ret = -EIO; 1225 goto exit; 1226 break; 1227 } 1228 nb_sectors -= sinfo.sectors_avail; 1229 sector_num += sinfo.sectors_avail; 1230 bytes_done += sinfo.bytes_avail; 1231 } 1232 } 1233 ret = 0; 1234 exit: 1235 qemu_co_mutex_unlock(&s->lock); 1236 qemu_iovec_destroy(&hd_qiov); 1237 return ret; 1238 } 1239 1240 /* 1241 * Allocate a new payload block at the end of the file. 1242 * 1243 * Allocation will happen at 1MB alignment inside the file 1244 * 1245 * Returns the file offset start of the new payload block 1246 */ 1247 static int vhdx_allocate_block(BlockDriverState *bs, BDRVVHDXState *s, 1248 uint64_t *new_offset) 1249 { 1250 int64_t current_len; 1251 1252 current_len = bdrv_getlength(bs->file->bs); 1253 if (current_len < 0) { 1254 return current_len; 1255 } 1256 1257 *new_offset = current_len; 1258 1259 /* per the spec, the address for a block is in units of 1MB */ 1260 *new_offset = ROUND_UP(*new_offset, 1 * MiB); 1261 if (*new_offset > INT64_MAX) { 1262 return -EINVAL; 1263 } 1264 1265 return bdrv_truncate(bs->file, *new_offset + s->block_size, false, 1266 PREALLOC_MODE_OFF, 0, NULL); 1267 } 1268 1269 /* 1270 * Update the BAT table entry with the new file offset, and the new entry 1271 * state */ 1272 static void vhdx_update_bat_table_entry(BlockDriverState *bs, BDRVVHDXState *s, 1273 VHDXSectorInfo *sinfo, 1274 uint64_t *bat_entry_le, 1275 uint64_t *bat_offset, int state) 1276 { 1277 /* The BAT entry is a uint64, with 44 bits for the file offset in units of 1278 * 1MB, and 3 bits for the block state. */ 1279 if ((state == PAYLOAD_BLOCK_ZERO) || 1280 (state == PAYLOAD_BLOCK_UNDEFINED) || 1281 (state == PAYLOAD_BLOCK_NOT_PRESENT) || 1282 (state == PAYLOAD_BLOCK_UNMAPPED)) { 1283 s->bat[sinfo->bat_idx] = 0; /* For PAYLOAD_BLOCK_ZERO, the 1284 FileOffsetMB field is denoted as 1285 'reserved' in the v1.0 spec. If it is 1286 non-zero, MS Hyper-V will fail to read 1287 the disk image */ 1288 } else { 1289 s->bat[sinfo->bat_idx] = sinfo->file_offset; 1290 } 1291 1292 s->bat[sinfo->bat_idx] |= state & VHDX_BAT_STATE_BIT_MASK; 1293 1294 *bat_entry_le = cpu_to_le64(s->bat[sinfo->bat_idx]); 1295 *bat_offset = s->bat_offset + sinfo->bat_idx * sizeof(VHDXBatEntry); 1296 1297 } 1298 1299 /* Per the spec, on the first write of guest-visible data to the file the 1300 * data write guid must be updated in the header */ 1301 int vhdx_user_visible_write(BlockDriverState *bs, BDRVVHDXState *s) 1302 { 1303 int ret = 0; 1304 if (s->first_visible_write) { 1305 s->first_visible_write = false; 1306 ret = vhdx_update_headers(bs, s, true, NULL); 1307 } 1308 return ret; 1309 } 1310 1311 static coroutine_fn int vhdx_co_writev(BlockDriverState *bs, int64_t sector_num, 1312 int nb_sectors, QEMUIOVector *qiov, 1313 int flags) 1314 { 1315 int ret = -ENOTSUP; 1316 BDRVVHDXState *s = bs->opaque; 1317 VHDXSectorInfo sinfo; 1318 uint64_t bytes_done = 0; 1319 uint64_t bat_entry = 0; 1320 uint64_t bat_entry_offset = 0; 1321 QEMUIOVector hd_qiov; 1322 struct iovec iov1 = { 0 }; 1323 struct iovec iov2 = { 0 }; 1324 int sectors_to_write; 1325 int bat_state; 1326 uint64_t bat_prior_offset = 0; 1327 bool bat_update = false; 1328 1329 assert(!flags); 1330 qemu_iovec_init(&hd_qiov, qiov->niov); 1331 1332 qemu_co_mutex_lock(&s->lock); 1333 1334 ret = vhdx_user_visible_write(bs, s); 1335 if (ret < 0) { 1336 goto exit; 1337 } 1338 1339 while (nb_sectors > 0) { 1340 bool use_zero_buffers = false; 1341 bat_update = false; 1342 if (s->params.data_bits & VHDX_PARAMS_HAS_PARENT) { 1343 /* not supported yet */ 1344 ret = -ENOTSUP; 1345 goto exit; 1346 } else { 1347 vhdx_block_translate(s, sector_num, nb_sectors, &sinfo); 1348 sectors_to_write = sinfo.sectors_avail; 1349 1350 qemu_iovec_reset(&hd_qiov); 1351 /* check the payload block state */ 1352 bat_state = s->bat[sinfo.bat_idx] & VHDX_BAT_STATE_BIT_MASK; 1353 switch (bat_state) { 1354 case PAYLOAD_BLOCK_ZERO: 1355 /* in this case, we need to preserve zero writes for 1356 * data that is not part of this write, so we must pad 1357 * the rest of the buffer to zeroes */ 1358 1359 /* if we are on a posix system with ftruncate() that extends 1360 * a file, then it is zero-filled for us. On Win32, the raw 1361 * layer uses SetFilePointer and SetFileEnd, which does not 1362 * zero fill AFAIK */ 1363 1364 /* Queue another write of zero buffers if the underlying file 1365 * does not zero-fill on file extension */ 1366 1367 if (bdrv_has_zero_init_truncate(bs->file->bs) == 0) { 1368 use_zero_buffers = true; 1369 1370 /* zero fill the front, if any */ 1371 if (sinfo.block_offset) { 1372 iov1.iov_len = sinfo.block_offset; 1373 iov1.iov_base = qemu_blockalign(bs, iov1.iov_len); 1374 memset(iov1.iov_base, 0, iov1.iov_len); 1375 qemu_iovec_concat_iov(&hd_qiov, &iov1, 1, 0, 1376 iov1.iov_len); 1377 sectors_to_write += iov1.iov_len >> BDRV_SECTOR_BITS; 1378 } 1379 1380 /* our actual data */ 1381 qemu_iovec_concat(&hd_qiov, qiov, bytes_done, 1382 sinfo.bytes_avail); 1383 1384 /* zero fill the back, if any */ 1385 if ((sinfo.bytes_avail - sinfo.block_offset) < 1386 s->block_size) { 1387 iov2.iov_len = s->block_size - 1388 (sinfo.bytes_avail + sinfo.block_offset); 1389 iov2.iov_base = qemu_blockalign(bs, iov2.iov_len); 1390 memset(iov2.iov_base, 0, iov2.iov_len); 1391 qemu_iovec_concat_iov(&hd_qiov, &iov2, 1, 0, 1392 iov2.iov_len); 1393 sectors_to_write += iov2.iov_len >> BDRV_SECTOR_BITS; 1394 } 1395 } 1396 /* fall through */ 1397 case PAYLOAD_BLOCK_NOT_PRESENT: /* fall through */ 1398 case PAYLOAD_BLOCK_UNMAPPED: 1399 case PAYLOAD_BLOCK_UNMAPPED_v095: 1400 case PAYLOAD_BLOCK_UNDEFINED: 1401 bat_prior_offset = sinfo.file_offset; 1402 ret = vhdx_allocate_block(bs, s, &sinfo.file_offset); 1403 if (ret < 0) { 1404 goto exit; 1405 } 1406 /* once we support differencing files, this may also be 1407 * partially present */ 1408 /* update block state to the newly specified state */ 1409 vhdx_update_bat_table_entry(bs, s, &sinfo, &bat_entry, 1410 &bat_entry_offset, 1411 PAYLOAD_BLOCK_FULLY_PRESENT); 1412 bat_update = true; 1413 /* since we just allocated a block, file_offset is the 1414 * beginning of the payload block. It needs to be the 1415 * write address, which includes the offset into the block */ 1416 if (!use_zero_buffers) { 1417 sinfo.file_offset += sinfo.block_offset; 1418 } 1419 /* fall through */ 1420 case PAYLOAD_BLOCK_FULLY_PRESENT: 1421 /* if the file offset address is in the header zone, 1422 * there is a problem */ 1423 if (sinfo.file_offset < (1 * MiB)) { 1424 ret = -EFAULT; 1425 goto error_bat_restore; 1426 } 1427 1428 if (!use_zero_buffers) { 1429 qemu_iovec_concat(&hd_qiov, qiov, bytes_done, 1430 sinfo.bytes_avail); 1431 } 1432 /* block exists, so we can just overwrite it */ 1433 qemu_co_mutex_unlock(&s->lock); 1434 ret = bdrv_co_pwritev(bs->file, sinfo.file_offset, 1435 sectors_to_write * BDRV_SECTOR_SIZE, 1436 &hd_qiov, 0); 1437 qemu_co_mutex_lock(&s->lock); 1438 if (ret < 0) { 1439 goto error_bat_restore; 1440 } 1441 break; 1442 case PAYLOAD_BLOCK_PARTIALLY_PRESENT: 1443 /* we don't yet support difference files, fall through 1444 * to error */ 1445 default: 1446 ret = -EIO; 1447 goto exit; 1448 break; 1449 } 1450 1451 if (bat_update) { 1452 /* this will update the BAT entry into the log journal, and 1453 * then flush the log journal out to disk */ 1454 ret = vhdx_log_write_and_flush(bs, s, &bat_entry, 1455 sizeof(VHDXBatEntry), 1456 bat_entry_offset); 1457 if (ret < 0) { 1458 goto exit; 1459 } 1460 } 1461 1462 nb_sectors -= sinfo.sectors_avail; 1463 sector_num += sinfo.sectors_avail; 1464 bytes_done += sinfo.bytes_avail; 1465 1466 } 1467 } 1468 1469 goto exit; 1470 1471 error_bat_restore: 1472 if (bat_update) { 1473 /* keep metadata in sync, and restore the bat entry state 1474 * if error. */ 1475 sinfo.file_offset = bat_prior_offset; 1476 vhdx_update_bat_table_entry(bs, s, &sinfo, &bat_entry, 1477 &bat_entry_offset, bat_state); 1478 } 1479 exit: 1480 qemu_vfree(iov1.iov_base); 1481 qemu_vfree(iov2.iov_base); 1482 qemu_co_mutex_unlock(&s->lock); 1483 qemu_iovec_destroy(&hd_qiov); 1484 return ret; 1485 } 1486 1487 1488 1489 /* 1490 * Create VHDX Headers 1491 * 1492 * There are 2 headers, and the highest sequence number will represent 1493 * the active header 1494 */ 1495 static int vhdx_create_new_headers(BlockBackend *blk, uint64_t image_size, 1496 uint32_t log_size) 1497 { 1498 BlockDriverState *bs = blk_bs(blk); 1499 BdrvChild *child; 1500 int ret = 0; 1501 VHDXHeader *hdr = NULL; 1502 1503 hdr = g_new0(VHDXHeader, 1); 1504 1505 hdr->signature = VHDX_HEADER_SIGNATURE; 1506 hdr->sequence_number = g_random_int(); 1507 hdr->log_version = 0; 1508 hdr->version = 1; 1509 hdr->log_length = log_size; 1510 hdr->log_offset = VHDX_HEADER_SECTION_END; 1511 vhdx_guid_generate(&hdr->file_write_guid); 1512 vhdx_guid_generate(&hdr->data_write_guid); 1513 1514 /* XXX Ugly way to get blk->root, but that's a feature, not a bug. This 1515 * hack makes it obvious that vhdx_write_header() bypasses the BlockBackend 1516 * here, which it really shouldn't be doing. */ 1517 child = QLIST_FIRST(&bs->parents); 1518 assert(!QLIST_NEXT(child, next_parent)); 1519 1520 ret = vhdx_write_header(child, hdr, VHDX_HEADER1_OFFSET, false); 1521 if (ret < 0) { 1522 goto exit; 1523 } 1524 hdr->sequence_number++; 1525 ret = vhdx_write_header(child, hdr, VHDX_HEADER2_OFFSET, false); 1526 if (ret < 0) { 1527 goto exit; 1528 } 1529 1530 exit: 1531 g_free(hdr); 1532 return ret; 1533 } 1534 1535 #define VHDX_METADATA_ENTRY_BUFFER_SIZE \ 1536 (sizeof(VHDXFileParameters) +\ 1537 sizeof(VHDXVirtualDiskSize) +\ 1538 sizeof(VHDXPage83Data) +\ 1539 sizeof(VHDXVirtualDiskLogicalSectorSize) +\ 1540 sizeof(VHDXVirtualDiskPhysicalSectorSize)) 1541 1542 /* 1543 * Create the Metadata entries. 1544 * 1545 * For more details on the entries, see section 3.5 (pg 29) in the 1546 * VHDX 1.00 specification. 1547 * 1548 * We support 5 metadata entries (all required by spec): 1549 * File Parameters, 1550 * Virtual Disk Size, 1551 * Page 83 Data, 1552 * Logical Sector Size, 1553 * Physical Sector Size 1554 * 1555 * The first 64KB of the Metadata section is reserved for the metadata 1556 * header and entries; beyond that, the metadata items themselves reside. 1557 */ 1558 static int vhdx_create_new_metadata(BlockBackend *blk, 1559 uint64_t image_size, 1560 uint32_t block_size, 1561 uint32_t sector_size, 1562 uint64_t metadata_offset, 1563 VHDXImageType type) 1564 { 1565 int ret = 0; 1566 uint32_t offset = 0; 1567 void *buffer = NULL; 1568 void *entry_buffer; 1569 VHDXMetadataTableHeader *md_table; 1570 VHDXMetadataTableEntry *md_table_entry; 1571 1572 /* Metadata entries */ 1573 VHDXFileParameters *mt_file_params; 1574 VHDXVirtualDiskSize *mt_virtual_size; 1575 VHDXPage83Data *mt_page83; 1576 VHDXVirtualDiskLogicalSectorSize *mt_log_sector_size; 1577 VHDXVirtualDiskPhysicalSectorSize *mt_phys_sector_size; 1578 1579 entry_buffer = g_malloc0(VHDX_METADATA_ENTRY_BUFFER_SIZE); 1580 1581 mt_file_params = entry_buffer; 1582 offset += sizeof(VHDXFileParameters); 1583 mt_virtual_size = entry_buffer + offset; 1584 offset += sizeof(VHDXVirtualDiskSize); 1585 mt_page83 = entry_buffer + offset; 1586 offset += sizeof(VHDXPage83Data); 1587 mt_log_sector_size = entry_buffer + offset; 1588 offset += sizeof(VHDXVirtualDiskLogicalSectorSize); 1589 mt_phys_sector_size = entry_buffer + offset; 1590 1591 mt_file_params->block_size = cpu_to_le32(block_size); 1592 if (type == VHDX_TYPE_FIXED) { 1593 mt_file_params->data_bits |= VHDX_PARAMS_LEAVE_BLOCKS_ALLOCED; 1594 mt_file_params->data_bits = cpu_to_le32(mt_file_params->data_bits); 1595 } 1596 1597 vhdx_guid_generate(&mt_page83->page_83_data); 1598 cpu_to_leguids(&mt_page83->page_83_data); 1599 mt_virtual_size->virtual_disk_size = cpu_to_le64(image_size); 1600 mt_log_sector_size->logical_sector_size = cpu_to_le32(sector_size); 1601 mt_phys_sector_size->physical_sector_size = cpu_to_le32(sector_size); 1602 1603 buffer = g_malloc0(VHDX_HEADER_BLOCK_SIZE); 1604 md_table = buffer; 1605 1606 md_table->signature = VHDX_METADATA_SIGNATURE; 1607 md_table->entry_count = 5; 1608 vhdx_metadata_header_le_export(md_table); 1609 1610 1611 /* This will reference beyond the reserved table portion */ 1612 offset = 64 * KiB; 1613 1614 md_table_entry = buffer + sizeof(VHDXMetadataTableHeader); 1615 1616 md_table_entry[0].item_id = file_param_guid; 1617 md_table_entry[0].offset = offset; 1618 md_table_entry[0].length = sizeof(VHDXFileParameters); 1619 md_table_entry[0].data_bits |= VHDX_META_FLAGS_IS_REQUIRED; 1620 offset += md_table_entry[0].length; 1621 vhdx_metadata_entry_le_export(&md_table_entry[0]); 1622 1623 md_table_entry[1].item_id = virtual_size_guid; 1624 md_table_entry[1].offset = offset; 1625 md_table_entry[1].length = sizeof(VHDXVirtualDiskSize); 1626 md_table_entry[1].data_bits |= VHDX_META_FLAGS_IS_REQUIRED | 1627 VHDX_META_FLAGS_IS_VIRTUAL_DISK; 1628 offset += md_table_entry[1].length; 1629 vhdx_metadata_entry_le_export(&md_table_entry[1]); 1630 1631 md_table_entry[2].item_id = page83_guid; 1632 md_table_entry[2].offset = offset; 1633 md_table_entry[2].length = sizeof(VHDXPage83Data); 1634 md_table_entry[2].data_bits |= VHDX_META_FLAGS_IS_REQUIRED | 1635 VHDX_META_FLAGS_IS_VIRTUAL_DISK; 1636 offset += md_table_entry[2].length; 1637 vhdx_metadata_entry_le_export(&md_table_entry[2]); 1638 1639 md_table_entry[3].item_id = logical_sector_guid; 1640 md_table_entry[3].offset = offset; 1641 md_table_entry[3].length = sizeof(VHDXVirtualDiskLogicalSectorSize); 1642 md_table_entry[3].data_bits |= VHDX_META_FLAGS_IS_REQUIRED | 1643 VHDX_META_FLAGS_IS_VIRTUAL_DISK; 1644 offset += md_table_entry[3].length; 1645 vhdx_metadata_entry_le_export(&md_table_entry[3]); 1646 1647 md_table_entry[4].item_id = phys_sector_guid; 1648 md_table_entry[4].offset = offset; 1649 md_table_entry[4].length = sizeof(VHDXVirtualDiskPhysicalSectorSize); 1650 md_table_entry[4].data_bits |= VHDX_META_FLAGS_IS_REQUIRED | 1651 VHDX_META_FLAGS_IS_VIRTUAL_DISK; 1652 vhdx_metadata_entry_le_export(&md_table_entry[4]); 1653 1654 ret = blk_pwrite(blk, metadata_offset, buffer, VHDX_HEADER_BLOCK_SIZE, 0); 1655 if (ret < 0) { 1656 goto exit; 1657 } 1658 1659 ret = blk_pwrite(blk, metadata_offset + (64 * KiB), entry_buffer, 1660 VHDX_METADATA_ENTRY_BUFFER_SIZE, 0); 1661 if (ret < 0) { 1662 goto exit; 1663 } 1664 1665 1666 exit: 1667 g_free(buffer); 1668 g_free(entry_buffer); 1669 return ret; 1670 } 1671 1672 /* This create the actual BAT itself. We currently only support 1673 * 'Dynamic' and 'Fixed' image types. 1674 * 1675 * Dynamic images: default state of the BAT is all zeroes. 1676 * 1677 * Fixed images: default state of the BAT is fully populated, with 1678 * file offsets and state PAYLOAD_BLOCK_FULLY_PRESENT. 1679 */ 1680 static int vhdx_create_bat(BlockBackend *blk, BDRVVHDXState *s, 1681 uint64_t image_size, VHDXImageType type, 1682 bool use_zero_blocks, uint64_t file_offset, 1683 uint32_t length, Error **errp) 1684 { 1685 int ret = 0; 1686 uint64_t data_file_offset; 1687 uint64_t total_sectors = 0; 1688 uint64_t sector_num = 0; 1689 uint64_t unused; 1690 int block_state; 1691 VHDXSectorInfo sinfo; 1692 1693 assert(s->bat == NULL); 1694 1695 /* this gives a data start after BAT/bitmap entries, and well 1696 * past any metadata entries (with a 4 MB buffer for future 1697 * expansion */ 1698 data_file_offset = file_offset + length + 5 * MiB; 1699 total_sectors = image_size >> s->logical_sector_size_bits; 1700 1701 if (type == VHDX_TYPE_DYNAMIC) { 1702 /* All zeroes, so we can just extend the file - the end of the BAT 1703 * is the furthest thing we have written yet */ 1704 ret = blk_truncate(blk, data_file_offset, false, PREALLOC_MODE_OFF, 1705 0, errp); 1706 if (ret < 0) { 1707 goto exit; 1708 } 1709 } else if (type == VHDX_TYPE_FIXED) { 1710 ret = blk_truncate(blk, data_file_offset + image_size, false, 1711 PREALLOC_MODE_OFF, 0, errp); 1712 if (ret < 0) { 1713 goto exit; 1714 } 1715 } else { 1716 error_setg(errp, "Unsupported image type"); 1717 ret = -ENOTSUP; 1718 goto exit; 1719 } 1720 1721 if (type == VHDX_TYPE_FIXED || 1722 use_zero_blocks || 1723 bdrv_has_zero_init(blk_bs(blk)) == 0) { 1724 /* for a fixed file, the default BAT entry is not zero */ 1725 s->bat = g_try_malloc0(length); 1726 if (length && s->bat == NULL) { 1727 error_setg(errp, "Failed to allocate memory for the BAT"); 1728 ret = -ENOMEM; 1729 goto exit; 1730 } 1731 block_state = type == VHDX_TYPE_FIXED ? PAYLOAD_BLOCK_FULLY_PRESENT : 1732 PAYLOAD_BLOCK_NOT_PRESENT; 1733 block_state = use_zero_blocks ? PAYLOAD_BLOCK_ZERO : block_state; 1734 /* fill the BAT by emulating sector writes of sectors_per_block size */ 1735 while (sector_num < total_sectors) { 1736 vhdx_block_translate(s, sector_num, s->sectors_per_block, &sinfo); 1737 sinfo.file_offset = data_file_offset + 1738 (sector_num << s->logical_sector_size_bits); 1739 sinfo.file_offset = ROUND_UP(sinfo.file_offset, MiB); 1740 vhdx_update_bat_table_entry(blk_bs(blk), s, &sinfo, &unused, &unused, 1741 block_state); 1742 s->bat[sinfo.bat_idx] = cpu_to_le64(s->bat[sinfo.bat_idx]); 1743 sector_num += s->sectors_per_block; 1744 } 1745 ret = blk_pwrite(blk, file_offset, s->bat, length, 0); 1746 if (ret < 0) { 1747 error_setg_errno(errp, -ret, "Failed to write the BAT"); 1748 goto exit; 1749 } 1750 } 1751 1752 1753 1754 exit: 1755 g_free(s->bat); 1756 return ret; 1757 } 1758 1759 /* Creates the region table header, and region table entries. 1760 * There are 2 supported region table entries: BAT, and Metadata/ 1761 * 1762 * As the calculations for the BAT region table are also needed 1763 * to create the BAT itself, we will also cause the BAT to be 1764 * created. 1765 */ 1766 static int vhdx_create_new_region_table(BlockBackend *blk, 1767 uint64_t image_size, 1768 uint32_t block_size, 1769 uint32_t sector_size, 1770 uint32_t log_size, 1771 bool use_zero_blocks, 1772 VHDXImageType type, 1773 uint64_t *metadata_offset, 1774 Error **errp) 1775 { 1776 int ret = 0; 1777 uint32_t offset = 0; 1778 void *buffer = NULL; 1779 uint64_t bat_file_offset; 1780 uint32_t bat_length; 1781 BDRVVHDXState *s = NULL; 1782 VHDXRegionTableHeader *region_table; 1783 VHDXRegionTableEntry *rt_bat; 1784 VHDXRegionTableEntry *rt_metadata; 1785 1786 assert(metadata_offset != NULL); 1787 1788 /* Populate enough of the BDRVVHDXState to be able to use the 1789 * pre-existing BAT calculation, translation, and update functions */ 1790 s = g_new0(BDRVVHDXState, 1); 1791 1792 s->chunk_ratio = (VHDX_MAX_SECTORS_PER_BLOCK) * 1793 (uint64_t) sector_size / (uint64_t) block_size; 1794 1795 s->sectors_per_block = block_size / sector_size; 1796 s->virtual_disk_size = image_size; 1797 s->block_size = block_size; 1798 s->logical_sector_size = sector_size; 1799 1800 vhdx_set_shift_bits(s); 1801 1802 vhdx_calc_bat_entries(s); 1803 1804 /* At this point the VHDX state is populated enough for creation */ 1805 1806 /* a single buffer is used so we can calculate the checksum over the 1807 * entire 64KB block */ 1808 buffer = g_malloc0(VHDX_HEADER_BLOCK_SIZE); 1809 region_table = buffer; 1810 offset += sizeof(VHDXRegionTableHeader); 1811 rt_bat = buffer + offset; 1812 offset += sizeof(VHDXRegionTableEntry); 1813 rt_metadata = buffer + offset; 1814 1815 region_table->signature = VHDX_REGION_SIGNATURE; 1816 region_table->entry_count = 2; /* BAT and Metadata */ 1817 1818 rt_bat->guid = bat_guid; 1819 rt_bat->length = ROUND_UP(s->bat_entries * sizeof(VHDXBatEntry), MiB); 1820 rt_bat->file_offset = ROUND_UP(VHDX_HEADER_SECTION_END + log_size, MiB); 1821 s->bat_offset = rt_bat->file_offset; 1822 1823 rt_metadata->guid = metadata_guid; 1824 rt_metadata->file_offset = ROUND_UP(rt_bat->file_offset + rt_bat->length, 1825 MiB); 1826 rt_metadata->length = 1 * MiB; /* min size, and more than enough */ 1827 *metadata_offset = rt_metadata->file_offset; 1828 1829 bat_file_offset = rt_bat->file_offset; 1830 bat_length = rt_bat->length; 1831 1832 vhdx_region_header_le_export(region_table); 1833 vhdx_region_entry_le_export(rt_bat); 1834 vhdx_region_entry_le_export(rt_metadata); 1835 1836 vhdx_update_checksum(buffer, VHDX_HEADER_BLOCK_SIZE, 1837 offsetof(VHDXRegionTableHeader, checksum)); 1838 1839 1840 /* The region table gives us the data we need to create the BAT, 1841 * so do that now */ 1842 ret = vhdx_create_bat(blk, s, image_size, type, use_zero_blocks, 1843 bat_file_offset, bat_length, errp); 1844 if (ret < 0) { 1845 goto exit; 1846 } 1847 1848 /* Now write out the region headers to disk */ 1849 ret = blk_pwrite(blk, VHDX_REGION_TABLE_OFFSET, buffer, 1850 VHDX_HEADER_BLOCK_SIZE, 0); 1851 if (ret < 0) { 1852 error_setg_errno(errp, -ret, "Failed to write first region table"); 1853 goto exit; 1854 } 1855 1856 ret = blk_pwrite(blk, VHDX_REGION_TABLE2_OFFSET, buffer, 1857 VHDX_HEADER_BLOCK_SIZE, 0); 1858 if (ret < 0) { 1859 error_setg_errno(errp, -ret, "Failed to write second region table"); 1860 goto exit; 1861 } 1862 1863 exit: 1864 g_free(s); 1865 g_free(buffer); 1866 return ret; 1867 } 1868 1869 /* We need to create the following elements: 1870 * 1871 * .-----------------------------------------------------------------. 1872 * | (A) | (B) | (C) | (D) | (E) | 1873 * | File ID | Header1 | Header 2 | Region Tbl 1 | Region Tbl 2 | 1874 * | | | | | | 1875 * .-----------------------------------------------------------------. 1876 * 0 64KB 128KB 192KB 256KB 320KB 1877 * 1878 * 1879 * .---- ~ ----------- ~ ------------ ~ ---------------- ~ -----------. 1880 * | (F) | (G) | (H) | | 1881 * | Journal Log | BAT / Bitmap | Metadata | .... data ...... | 1882 * | | | | | 1883 * .---- ~ ----------- ~ ------------ ~ ---------------- ~ -----------. 1884 * 1MB 1885 */ 1886 static int coroutine_fn vhdx_co_create(BlockdevCreateOptions *opts, 1887 Error **errp) 1888 { 1889 BlockdevCreateOptionsVhdx *vhdx_opts; 1890 BlockBackend *blk = NULL; 1891 BlockDriverState *bs = NULL; 1892 1893 int ret = 0; 1894 uint64_t image_size; 1895 uint32_t log_size; 1896 uint32_t block_size; 1897 uint64_t signature; 1898 uint64_t metadata_offset; 1899 bool use_zero_blocks = false; 1900 1901 gunichar2 *creator = NULL; 1902 glong creator_items; 1903 VHDXImageType image_type; 1904 1905 assert(opts->driver == BLOCKDEV_DRIVER_VHDX); 1906 vhdx_opts = &opts->u.vhdx; 1907 1908 /* Validate options and set default values */ 1909 image_size = vhdx_opts->size; 1910 if (image_size > VHDX_MAX_IMAGE_SIZE) { 1911 error_setg(errp, "Image size too large; max of 64TB"); 1912 return -EINVAL; 1913 } 1914 1915 if (!vhdx_opts->has_log_size) { 1916 log_size = DEFAULT_LOG_SIZE; 1917 } else { 1918 if (vhdx_opts->log_size > UINT32_MAX) { 1919 error_setg(errp, "Log size must be smaller than 4 GB"); 1920 return -EINVAL; 1921 } 1922 log_size = vhdx_opts->log_size; 1923 } 1924 if (log_size < MiB || (log_size % MiB) != 0) { 1925 error_setg(errp, "Log size must be a multiple of 1 MB"); 1926 return -EINVAL; 1927 } 1928 1929 if (!vhdx_opts->has_block_state_zero) { 1930 use_zero_blocks = true; 1931 } else { 1932 use_zero_blocks = vhdx_opts->block_state_zero; 1933 } 1934 1935 if (!vhdx_opts->has_subformat) { 1936 vhdx_opts->subformat = BLOCKDEV_VHDX_SUBFORMAT_DYNAMIC; 1937 } 1938 1939 switch (vhdx_opts->subformat) { 1940 case BLOCKDEV_VHDX_SUBFORMAT_DYNAMIC: 1941 image_type = VHDX_TYPE_DYNAMIC; 1942 break; 1943 case BLOCKDEV_VHDX_SUBFORMAT_FIXED: 1944 image_type = VHDX_TYPE_FIXED; 1945 break; 1946 default: 1947 g_assert_not_reached(); 1948 } 1949 1950 /* These are pretty arbitrary, and mainly designed to keep the BAT 1951 * size reasonable to load into RAM */ 1952 if (vhdx_opts->has_block_size) { 1953 block_size = vhdx_opts->block_size; 1954 } else { 1955 if (image_size > 32 * TiB) { 1956 block_size = 64 * MiB; 1957 } else if (image_size > (uint64_t) 100 * GiB) { 1958 block_size = 32 * MiB; 1959 } else if (image_size > 1 * GiB) { 1960 block_size = 16 * MiB; 1961 } else { 1962 block_size = 8 * MiB; 1963 } 1964 } 1965 1966 if (block_size < MiB || (block_size % MiB) != 0) { 1967 error_setg(errp, "Block size must be a multiple of 1 MB"); 1968 return -EINVAL; 1969 } 1970 if (!is_power_of_2(block_size)) { 1971 error_setg(errp, "Block size must be a power of two"); 1972 return -EINVAL; 1973 } 1974 if (block_size > VHDX_BLOCK_SIZE_MAX) { 1975 error_setg(errp, "Block size must not exceed %" PRId64, 1976 VHDX_BLOCK_SIZE_MAX); 1977 return -EINVAL; 1978 } 1979 1980 /* Create BlockBackend to write to the image */ 1981 bs = bdrv_open_blockdev_ref(vhdx_opts->file, errp); 1982 if (bs == NULL) { 1983 return -EIO; 1984 } 1985 1986 blk = blk_new(bdrv_get_aio_context(bs), 1987 BLK_PERM_WRITE | BLK_PERM_RESIZE, BLK_PERM_ALL); 1988 ret = blk_insert_bs(blk, bs, errp); 1989 if (ret < 0) { 1990 goto delete_and_exit; 1991 } 1992 blk_set_allow_write_beyond_eof(blk, true); 1993 1994 /* Create (A) */ 1995 1996 /* The creator field is optional, but may be useful for 1997 * debugging / diagnostics */ 1998 creator = g_utf8_to_utf16("QEMU v" QEMU_VERSION, -1, NULL, 1999 &creator_items, NULL); 2000 signature = cpu_to_le64(VHDX_FILE_SIGNATURE); 2001 ret = blk_pwrite(blk, VHDX_FILE_ID_OFFSET, &signature, sizeof(signature), 2002 0); 2003 if (ret < 0) { 2004 error_setg_errno(errp, -ret, "Failed to write file signature"); 2005 goto delete_and_exit; 2006 } 2007 if (creator) { 2008 ret = blk_pwrite(blk, VHDX_FILE_ID_OFFSET + sizeof(signature), 2009 creator, creator_items * sizeof(gunichar2), 0); 2010 if (ret < 0) { 2011 error_setg_errno(errp, -ret, "Failed to write creator field"); 2012 goto delete_and_exit; 2013 } 2014 } 2015 2016 2017 /* Creates (B),(C) */ 2018 ret = vhdx_create_new_headers(blk, image_size, log_size); 2019 if (ret < 0) { 2020 error_setg_errno(errp, -ret, "Failed to write image headers"); 2021 goto delete_and_exit; 2022 } 2023 2024 /* Creates (D),(E),(G) explicitly. (F) created as by-product */ 2025 ret = vhdx_create_new_region_table(blk, image_size, block_size, 512, 2026 log_size, use_zero_blocks, image_type, 2027 &metadata_offset, errp); 2028 if (ret < 0) { 2029 goto delete_and_exit; 2030 } 2031 2032 /* Creates (H) */ 2033 ret = vhdx_create_new_metadata(blk, image_size, block_size, 512, 2034 metadata_offset, image_type); 2035 if (ret < 0) { 2036 error_setg_errno(errp, -ret, "Failed to initialize metadata"); 2037 goto delete_and_exit; 2038 } 2039 2040 ret = 0; 2041 delete_and_exit: 2042 blk_unref(blk); 2043 bdrv_unref(bs); 2044 g_free(creator); 2045 return ret; 2046 } 2047 2048 static int coroutine_fn vhdx_co_create_opts(BlockDriver *drv, 2049 const char *filename, 2050 QemuOpts *opts, 2051 Error **errp) 2052 { 2053 BlockdevCreateOptions *create_options = NULL; 2054 QDict *qdict; 2055 Visitor *v; 2056 BlockDriverState *bs = NULL; 2057 Error *local_err = NULL; 2058 int ret; 2059 2060 static const QDictRenames opt_renames[] = { 2061 { VHDX_BLOCK_OPT_LOG_SIZE, "log-size" }, 2062 { VHDX_BLOCK_OPT_BLOCK_SIZE, "block-size" }, 2063 { VHDX_BLOCK_OPT_ZERO, "block-state-zero" }, 2064 { NULL, NULL }, 2065 }; 2066 2067 /* Parse options and convert legacy syntax */ 2068 qdict = qemu_opts_to_qdict_filtered(opts, NULL, &vhdx_create_opts, true); 2069 2070 if (!qdict_rename_keys(qdict, opt_renames, errp)) { 2071 ret = -EINVAL; 2072 goto fail; 2073 } 2074 2075 /* Create and open the file (protocol layer) */ 2076 ret = bdrv_create_file(filename, opts, &local_err); 2077 if (ret < 0) { 2078 error_propagate(errp, local_err); 2079 goto fail; 2080 } 2081 2082 bs = bdrv_open(filename, NULL, NULL, 2083 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp); 2084 if (bs == NULL) { 2085 ret = -EIO; 2086 goto fail; 2087 } 2088 2089 /* Now get the QAPI type BlockdevCreateOptions */ 2090 qdict_put_str(qdict, "driver", "vhdx"); 2091 qdict_put_str(qdict, "file", bs->node_name); 2092 2093 v = qobject_input_visitor_new_flat_confused(qdict, errp); 2094 if (!v) { 2095 ret = -EINVAL; 2096 goto fail; 2097 } 2098 2099 visit_type_BlockdevCreateOptions(v, NULL, &create_options, &local_err); 2100 visit_free(v); 2101 2102 if (local_err) { 2103 error_propagate(errp, local_err); 2104 ret = -EINVAL; 2105 goto fail; 2106 } 2107 2108 /* Silently round up sizes: 2109 * The image size is rounded to 512 bytes. Make the block and log size 2110 * close to what was specified, but must be at least 1MB, and a multiple of 2111 * 1 MB. Also respect VHDX_BLOCK_SIZE_MAX for block sizes. block_size = 0 2112 * means auto, which is represented by a missing key in QAPI. */ 2113 assert(create_options->driver == BLOCKDEV_DRIVER_VHDX); 2114 create_options->u.vhdx.size = 2115 ROUND_UP(create_options->u.vhdx.size, BDRV_SECTOR_SIZE); 2116 2117 if (create_options->u.vhdx.has_log_size) { 2118 create_options->u.vhdx.log_size = 2119 ROUND_UP(create_options->u.vhdx.log_size, MiB); 2120 } 2121 if (create_options->u.vhdx.has_block_size) { 2122 create_options->u.vhdx.block_size = 2123 ROUND_UP(create_options->u.vhdx.block_size, MiB); 2124 2125 if (create_options->u.vhdx.block_size == 0) { 2126 create_options->u.vhdx.has_block_size = false; 2127 } 2128 if (create_options->u.vhdx.block_size > VHDX_BLOCK_SIZE_MAX) { 2129 create_options->u.vhdx.block_size = VHDX_BLOCK_SIZE_MAX; 2130 } 2131 } 2132 2133 /* Create the vhdx image (format layer) */ 2134 ret = vhdx_co_create(create_options, errp); 2135 2136 fail: 2137 qobject_unref(qdict); 2138 bdrv_unref(bs); 2139 qapi_free_BlockdevCreateOptions(create_options); 2140 return ret; 2141 } 2142 2143 /* If opened r/w, the VHDX driver will automatically replay the log, 2144 * if one is present, inside the vhdx_open() call. 2145 * 2146 * If qemu-img check -r all is called, the image is automatically opened 2147 * r/w and any log has already been replayed, so there is nothing (currently) 2148 * for us to do here 2149 */ 2150 static int coroutine_fn vhdx_co_check(BlockDriverState *bs, 2151 BdrvCheckResult *result, 2152 BdrvCheckMode fix) 2153 { 2154 BDRVVHDXState *s = bs->opaque; 2155 2156 if (s->log_replayed_on_open) { 2157 result->corruptions_fixed++; 2158 } 2159 2160 vhdx_check_bat_entries(bs, &result->corruptions); 2161 2162 return 0; 2163 } 2164 2165 static int vhdx_has_zero_init(BlockDriverState *bs) 2166 { 2167 BDRVVHDXState *s = bs->opaque; 2168 int state; 2169 2170 /* 2171 * Check the subformat: Fixed images have all BAT entries present, 2172 * dynamic images have none (right after creation). It is 2173 * therefore enough to check the first BAT entry. 2174 */ 2175 if (!s->bat_entries) { 2176 return 1; 2177 } 2178 2179 state = s->bat[0] & VHDX_BAT_STATE_BIT_MASK; 2180 if (state == PAYLOAD_BLOCK_FULLY_PRESENT) { 2181 /* Fixed subformat */ 2182 return bdrv_has_zero_init(bs->file->bs); 2183 } 2184 2185 /* Dynamic subformat */ 2186 return 1; 2187 } 2188 2189 static QemuOptsList vhdx_create_opts = { 2190 .name = "vhdx-create-opts", 2191 .head = QTAILQ_HEAD_INITIALIZER(vhdx_create_opts.head), 2192 .desc = { 2193 { 2194 .name = BLOCK_OPT_SIZE, 2195 .type = QEMU_OPT_SIZE, 2196 .help = "Virtual disk size; max of 64TB." 2197 }, 2198 { 2199 .name = VHDX_BLOCK_OPT_LOG_SIZE, 2200 .type = QEMU_OPT_SIZE, 2201 .def_value_str = stringify(DEFAULT_LOG_SIZE), 2202 .help = "Log size; min 1MB." 2203 }, 2204 { 2205 .name = VHDX_BLOCK_OPT_BLOCK_SIZE, 2206 .type = QEMU_OPT_SIZE, 2207 .def_value_str = stringify(0), 2208 .help = "Block Size; min 1MB, max 256MB. " 2209 "0 means auto-calculate based on image size." 2210 }, 2211 { 2212 .name = BLOCK_OPT_SUBFMT, 2213 .type = QEMU_OPT_STRING, 2214 .help = "VHDX format type, can be either 'dynamic' or 'fixed'. " 2215 "Default is 'dynamic'." 2216 }, 2217 { 2218 .name = VHDX_BLOCK_OPT_ZERO, 2219 .type = QEMU_OPT_BOOL, 2220 .help = "Force use of payload blocks of type 'ZERO'. " 2221 "Non-standard, but default. Do not set to 'off' when " 2222 "using 'qemu-img convert' with subformat=dynamic." 2223 }, 2224 { NULL } 2225 } 2226 }; 2227 2228 static BlockDriver bdrv_vhdx = { 2229 .format_name = "vhdx", 2230 .instance_size = sizeof(BDRVVHDXState), 2231 .bdrv_probe = vhdx_probe, 2232 .bdrv_open = vhdx_open, 2233 .bdrv_close = vhdx_close, 2234 .bdrv_reopen_prepare = vhdx_reopen_prepare, 2235 .bdrv_child_perm = bdrv_format_default_perms, 2236 .bdrv_co_readv = vhdx_co_readv, 2237 .bdrv_co_writev = vhdx_co_writev, 2238 .bdrv_co_create = vhdx_co_create, 2239 .bdrv_co_create_opts = vhdx_co_create_opts, 2240 .bdrv_get_info = vhdx_get_info, 2241 .bdrv_co_check = vhdx_co_check, 2242 .bdrv_has_zero_init = vhdx_has_zero_init, 2243 2244 .create_opts = &vhdx_create_opts, 2245 }; 2246 2247 static void bdrv_vhdx_init(void) 2248 { 2249 bdrv_register(&bdrv_vhdx); 2250 } 2251 2252 block_init(bdrv_vhdx_init); 2253