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 "qemu/memalign.h" 29 #include "vhdx.h" 30 #include "migration/blocker.h" 31 #include "qemu/uuid.h" 32 #include "qapi/qmp/qdict.h" 33 #include "qapi/qobject-input-visitor.h" 34 #include "qapi/qapi-visit-block-core.h" 35 36 /* Options for VHDX creation */ 37 38 #define VHDX_BLOCK_OPT_LOG_SIZE "log_size" 39 #define VHDX_BLOCK_OPT_BLOCK_SIZE "block_size" 40 #define VHDX_BLOCK_OPT_ZERO "block_state_zero" 41 42 typedef enum VHDXImageType { 43 VHDX_TYPE_DYNAMIC = 0, 44 VHDX_TYPE_FIXED, 45 VHDX_TYPE_DIFFERENCING, /* Currently unsupported */ 46 } VHDXImageType; 47 48 static QemuOptsList vhdx_create_opts; 49 50 /* Several metadata and region table data entries are identified by 51 * guids in a MS-specific GUID format. */ 52 53 54 /* ------- Known Region Table GUIDs ---------------------- */ 55 static const MSGUID bat_guid = { .data1 = 0x2dc27766, 56 .data2 = 0xf623, 57 .data3 = 0x4200, 58 .data4 = { 0x9d, 0x64, 0x11, 0x5e, 59 0x9b, 0xfd, 0x4a, 0x08} }; 60 61 static const MSGUID metadata_guid = { .data1 = 0x8b7ca206, 62 .data2 = 0x4790, 63 .data3 = 0x4b9a, 64 .data4 = { 0xb8, 0xfe, 0x57, 0x5f, 65 0x05, 0x0f, 0x88, 0x6e} }; 66 67 68 69 /* ------- Known Metadata Entry GUIDs ---------------------- */ 70 static const MSGUID file_param_guid = { .data1 = 0xcaa16737, 71 .data2 = 0xfa36, 72 .data3 = 0x4d43, 73 .data4 = { 0xb3, 0xb6, 0x33, 0xf0, 74 0xaa, 0x44, 0xe7, 0x6b} }; 75 76 static const MSGUID virtual_size_guid = { .data1 = 0x2FA54224, 77 .data2 = 0xcd1b, 78 .data3 = 0x4876, 79 .data4 = { 0xb2, 0x11, 0x5d, 0xbe, 80 0xd8, 0x3b, 0xf4, 0xb8} }; 81 82 static const MSGUID page83_guid = { .data1 = 0xbeca12ab, 83 .data2 = 0xb2e6, 84 .data3 = 0x4523, 85 .data4 = { 0x93, 0xef, 0xc3, 0x09, 86 0xe0, 0x00, 0xc7, 0x46} }; 87 88 89 static const MSGUID phys_sector_guid = { .data1 = 0xcda348c7, 90 .data2 = 0x445d, 91 .data3 = 0x4471, 92 .data4 = { 0x9c, 0xc9, 0xe9, 0x88, 93 0x52, 0x51, 0xc5, 0x56} }; 94 95 static const MSGUID parent_locator_guid = { .data1 = 0xa8d35f2d, 96 .data2 = 0xb30b, 97 .data3 = 0x454d, 98 .data4 = { 0xab, 0xf7, 0xd3, 99 0xd8, 0x48, 0x34, 100 0xab, 0x0c} }; 101 102 static const MSGUID logical_sector_guid = { .data1 = 0x8141bf1d, 103 .data2 = 0xa96f, 104 .data3 = 0x4709, 105 .data4 = { 0xba, 0x47, 0xf2, 106 0x33, 0xa8, 0xfa, 107 0xab, 0x5f} }; 108 109 /* Each parent type must have a valid GUID; this is for parent images 110 * of type 'VHDX'. If we were to allow e.g. a QCOW2 parent, we would 111 * need to make up our own QCOW2 GUID type */ 112 static const MSGUID parent_vhdx_guid __attribute__((unused)) 113 = { .data1 = 0xb04aefb7, 114 .data2 = 0xd19e, 115 .data3 = 0x4a81, 116 .data4 = { 0xb7, 0x89, 0x25, 0xb8, 117 0xe9, 0x44, 0x59, 0x13} }; 118 119 120 #define META_FILE_PARAMETER_PRESENT 0x01 121 #define META_VIRTUAL_DISK_SIZE_PRESENT 0x02 122 #define META_PAGE_83_PRESENT 0x04 123 #define META_LOGICAL_SECTOR_SIZE_PRESENT 0x08 124 #define META_PHYS_SECTOR_SIZE_PRESENT 0x10 125 #define META_PARENT_LOCATOR_PRESENT 0x20 126 127 #define META_ALL_PRESENT \ 128 (META_FILE_PARAMETER_PRESENT | META_VIRTUAL_DISK_SIZE_PRESENT | \ 129 META_PAGE_83_PRESENT | META_LOGICAL_SECTOR_SIZE_PRESENT | \ 130 META_PHYS_SECTOR_SIZE_PRESENT) 131 132 133 typedef struct VHDXSectorInfo { 134 uint32_t bat_idx; /* BAT entry index */ 135 uint32_t sectors_avail; /* sectors available in payload block */ 136 uint32_t bytes_left; /* bytes left in the block after data to r/w */ 137 uint32_t bytes_avail; /* bytes available in payload block */ 138 uint64_t file_offset; /* absolute offset in bytes, in file */ 139 uint64_t block_offset; /* block offset, in bytes */ 140 } VHDXSectorInfo; 141 142 /* Calculates new checksum. 143 * 144 * Zero is substituted during crc calculation for the original crc field 145 * crc_offset: byte offset in buf of the buffer crc 146 * buf: buffer pointer 147 * size: size of buffer (must be > crc_offset+4) 148 * 149 * Note: The buffer should have all multi-byte data in little-endian format, 150 * and the resulting checksum is in little endian format. 151 */ 152 uint32_t vhdx_update_checksum(uint8_t *buf, size_t size, int crc_offset) 153 { 154 uint32_t crc; 155 156 assert(buf != NULL); 157 assert(size > (crc_offset + sizeof(crc))); 158 159 memset(buf + crc_offset, 0, sizeof(crc)); 160 crc = crc32c(0xffffffff, buf, size); 161 crc = cpu_to_le32(crc); 162 memcpy(buf + crc_offset, &crc, sizeof(crc)); 163 164 return crc; 165 } 166 167 uint32_t vhdx_checksum_calc(uint32_t crc, uint8_t *buf, size_t size, 168 int crc_offset) 169 { 170 uint32_t crc_new; 171 uint32_t crc_orig; 172 assert(buf != NULL); 173 174 if (crc_offset > 0) { 175 memcpy(&crc_orig, buf + crc_offset, sizeof(crc_orig)); 176 memset(buf + crc_offset, 0, sizeof(crc_orig)); 177 } 178 179 crc_new = crc32c(crc, buf, size); 180 if (crc_offset > 0) { 181 memcpy(buf + crc_offset, &crc_orig, sizeof(crc_orig)); 182 } 183 184 return crc_new; 185 } 186 187 /* Validates the checksum of the buffer, with an in-place CRC. 188 * 189 * Zero is substituted during crc calculation for the original crc field, 190 * and the crc field is restored afterwards. But the buffer will be modified 191 * during the calculation, so this may not be not suitable for multi-threaded 192 * use. 193 * 194 * crc_offset: byte offset in buf of the buffer crc 195 * buf: buffer pointer 196 * size: size of buffer (must be > crc_offset+4) 197 * 198 * returns true if checksum is valid, false otherwise 199 */ 200 bool vhdx_checksum_is_valid(uint8_t *buf, size_t size, int crc_offset) 201 { 202 uint32_t crc_orig; 203 uint32_t crc; 204 205 assert(buf != NULL); 206 assert(size > (crc_offset + 4)); 207 208 memcpy(&crc_orig, buf + crc_offset, sizeof(crc_orig)); 209 crc_orig = le32_to_cpu(crc_orig); 210 211 crc = vhdx_checksum_calc(0xffffffff, buf, size, crc_offset); 212 213 return crc == crc_orig; 214 } 215 216 217 /* 218 * This generates a UUID that is compliant with the MS GUIDs used 219 * in the VHDX spec (and elsewhere). 220 */ 221 void vhdx_guid_generate(MSGUID *guid) 222 { 223 QemuUUID uuid; 224 assert(guid != NULL); 225 226 qemu_uuid_generate(&uuid); 227 memcpy(guid, &uuid, sizeof(MSGUID)); 228 } 229 230 /* Check for region overlaps inside the VHDX image */ 231 static int vhdx_region_check(BDRVVHDXState *s, uint64_t start, uint64_t length) 232 { 233 int ret = 0; 234 uint64_t end; 235 VHDXRegionEntry *r; 236 237 end = start + length; 238 QLIST_FOREACH(r, &s->regions, entries) { 239 if (!((start >= r->end) || (end <= r->start))) { 240 error_report("VHDX region %" PRIu64 "-%" PRIu64 " overlaps with " 241 "region %" PRIu64 "-%." PRIu64, start, end, r->start, 242 r->end); 243 ret = -EINVAL; 244 goto exit; 245 } 246 } 247 248 exit: 249 return ret; 250 } 251 252 /* Register a region for future checks */ 253 static void vhdx_region_register(BDRVVHDXState *s, 254 uint64_t start, uint64_t length) 255 { 256 VHDXRegionEntry *r; 257 258 r = g_malloc0(sizeof(*r)); 259 260 r->start = start; 261 r->end = start + length; 262 263 QLIST_INSERT_HEAD(&s->regions, r, entries); 264 } 265 266 /* Free all registered regions */ 267 static void vhdx_region_unregister_all(BDRVVHDXState *s) 268 { 269 VHDXRegionEntry *r, *r_next; 270 271 QLIST_FOREACH_SAFE(r, &s->regions, entries, r_next) { 272 QLIST_REMOVE(r, entries); 273 g_free(r); 274 } 275 } 276 277 static void vhdx_set_shift_bits(BDRVVHDXState *s) 278 { 279 s->logical_sector_size_bits = ctz32(s->logical_sector_size); 280 s->sectors_per_block_bits = ctz32(s->sectors_per_block); 281 s->chunk_ratio_bits = ctz64(s->chunk_ratio); 282 s->block_size_bits = ctz32(s->block_size); 283 } 284 285 /* 286 * Per the MS VHDX Specification, for every VHDX file: 287 * - The header section is fixed size - 1 MB 288 * - The header section is always the first "object" 289 * - The first 64KB of the header is the File Identifier 290 * - The first uint64 (8 bytes) is the VHDX Signature ("vhdxfile") 291 * - The following 512 bytes constitute a UTF-16 string identifiying the 292 * software that created the file, and is optional and diagnostic only. 293 * 294 * Therefore, we probe by looking for the vhdxfile signature "vhdxfile" 295 */ 296 static int vhdx_probe(const uint8_t *buf, int buf_size, const char *filename) 297 { 298 if (buf_size >= 8 && !memcmp(buf, "vhdxfile", 8)) { 299 return 100; 300 } 301 return 0; 302 } 303 304 /* 305 * Writes the header to the specified offset. 306 * 307 * This will optionally read in buffer data from disk (otherwise zero-fill), 308 * and then update the header checksum. Header is converted to proper 309 * endianness before being written to the specified file offset 310 */ 311 static int vhdx_write_header(BdrvChild *file, VHDXHeader *hdr, 312 uint64_t offset, bool read) 313 { 314 BlockDriverState *bs_file = file->bs; 315 uint8_t *buffer = NULL; 316 int ret; 317 VHDXHeader *header_le; 318 319 assert(bs_file != NULL); 320 assert(hdr != NULL); 321 322 /* the header checksum is not over just the packed size of VHDXHeader, 323 * but rather over the entire 'reserved' range for the header, which is 324 * 4KB (VHDX_HEADER_SIZE). */ 325 326 buffer = qemu_blockalign(bs_file, VHDX_HEADER_SIZE); 327 if (read) { 328 /* if true, we can't assume the extra reserved bytes are 0 */ 329 ret = bdrv_pread(file, offset, VHDX_HEADER_SIZE, buffer, 0); 330 if (ret < 0) { 331 goto exit; 332 } 333 } else { 334 memset(buffer, 0, VHDX_HEADER_SIZE); 335 } 336 337 /* overwrite the actual VHDXHeader portion */ 338 header_le = (VHDXHeader *)buffer; 339 memcpy(header_le, hdr, sizeof(VHDXHeader)); 340 vhdx_header_le_export(hdr, header_le); 341 vhdx_update_checksum(buffer, VHDX_HEADER_SIZE, 342 offsetof(VHDXHeader, checksum)); 343 ret = bdrv_pwrite_sync(file, offset, sizeof(VHDXHeader), header_le, 0); 344 345 exit: 346 qemu_vfree(buffer); 347 return ret; 348 } 349 350 /* Update the VHDX headers 351 * 352 * This follows the VHDX spec procedures for header updates. 353 * 354 * - non-current header is updated with largest sequence number 355 */ 356 static int vhdx_update_header(BlockDriverState *bs, BDRVVHDXState *s, 357 bool generate_data_write_guid, MSGUID *log_guid) 358 { 359 int ret = 0; 360 int hdr_idx = 0; 361 uint64_t header_offset = VHDX_HEADER1_OFFSET; 362 363 VHDXHeader *active_header; 364 VHDXHeader *inactive_header; 365 366 /* operate on the non-current header */ 367 if (s->curr_header == 0) { 368 hdr_idx = 1; 369 header_offset = VHDX_HEADER2_OFFSET; 370 } 371 372 active_header = s->headers[s->curr_header]; 373 inactive_header = s->headers[hdr_idx]; 374 375 inactive_header->sequence_number = active_header->sequence_number + 1; 376 377 /* a new file guid must be generated before any file write, including 378 * headers */ 379 inactive_header->file_write_guid = s->session_guid; 380 381 /* a new data guid only needs to be generated before any guest-visible 382 * writes (i.e. something observable via virtual disk read) */ 383 if (generate_data_write_guid) { 384 vhdx_guid_generate(&inactive_header->data_write_guid); 385 } 386 387 /* update the log guid if present */ 388 if (log_guid) { 389 inactive_header->log_guid = *log_guid; 390 } 391 392 ret = vhdx_write_header(bs->file, inactive_header, header_offset, true); 393 if (ret < 0) { 394 goto exit; 395 } 396 s->curr_header = hdr_idx; 397 398 exit: 399 return ret; 400 } 401 402 /* 403 * The VHDX spec calls for header updates to be performed twice, so that both 404 * the current and non-current header have valid info 405 */ 406 int vhdx_update_headers(BlockDriverState *bs, BDRVVHDXState *s, 407 bool generate_data_write_guid, MSGUID *log_guid) 408 { 409 int ret; 410 411 ret = vhdx_update_header(bs, s, generate_data_write_guid, log_guid); 412 if (ret < 0) { 413 return ret; 414 } 415 return vhdx_update_header(bs, s, generate_data_write_guid, log_guid); 416 } 417 418 /* opens the specified header block from the VHDX file header section */ 419 static void vhdx_parse_header(BlockDriverState *bs, BDRVVHDXState *s, 420 Error **errp) 421 { 422 int ret; 423 VHDXHeader *header1; 424 VHDXHeader *header2; 425 bool h1_valid = false; 426 bool h2_valid = false; 427 uint64_t h1_seq = 0; 428 uint64_t h2_seq = 0; 429 uint8_t *buffer; 430 431 /* header1 & header2 are freed in vhdx_close() */ 432 header1 = qemu_blockalign(bs, sizeof(VHDXHeader)); 433 header2 = qemu_blockalign(bs, sizeof(VHDXHeader)); 434 435 buffer = qemu_blockalign(bs, VHDX_HEADER_SIZE); 436 437 s->headers[0] = header1; 438 s->headers[1] = header2; 439 440 /* We have to read the whole VHDX_HEADER_SIZE instead of 441 * sizeof(VHDXHeader), because the checksum is over the whole 442 * region */ 443 ret = bdrv_pread(bs->file, VHDX_HEADER1_OFFSET, VHDX_HEADER_SIZE, buffer, 444 0); 445 if (ret < 0) { 446 goto fail; 447 } 448 /* copy over just the relevant portion that we need */ 449 memcpy(header1, buffer, sizeof(VHDXHeader)); 450 451 if (vhdx_checksum_is_valid(buffer, VHDX_HEADER_SIZE, 4)) { 452 vhdx_header_le_import(header1); 453 if (header1->signature == VHDX_HEADER_SIGNATURE && 454 header1->version == 1) { 455 h1_seq = header1->sequence_number; 456 h1_valid = true; 457 } 458 } 459 460 ret = bdrv_pread(bs->file, VHDX_HEADER2_OFFSET, VHDX_HEADER_SIZE, buffer, 461 0); 462 if (ret < 0) { 463 goto fail; 464 } 465 /* copy over just the relevant portion that we need */ 466 memcpy(header2, buffer, sizeof(VHDXHeader)); 467 468 if (vhdx_checksum_is_valid(buffer, VHDX_HEADER_SIZE, 4)) { 469 vhdx_header_le_import(header2); 470 if (header2->signature == VHDX_HEADER_SIGNATURE && 471 header2->version == 1) { 472 h2_seq = header2->sequence_number; 473 h2_valid = true; 474 } 475 } 476 477 /* If there is only 1 valid header (or no valid headers), we 478 * don't care what the sequence numbers are */ 479 if (h1_valid && !h2_valid) { 480 s->curr_header = 0; 481 } else if (!h1_valid && h2_valid) { 482 s->curr_header = 1; 483 } else if (!h1_valid && !h2_valid) { 484 goto fail; 485 } else { 486 /* If both headers are valid, then we choose the active one by the 487 * highest sequence number. If the sequence numbers are equal, that is 488 * invalid */ 489 if (h1_seq > h2_seq) { 490 s->curr_header = 0; 491 } else if (h2_seq > h1_seq) { 492 s->curr_header = 1; 493 } else { 494 /* The Microsoft Disk2VHD tool will create 2 identical 495 * headers, with identical sequence numbers. If the headers are 496 * identical, don't consider the file corrupt */ 497 if (!memcmp(header1, header2, sizeof(VHDXHeader))) { 498 s->curr_header = 0; 499 } else { 500 goto fail; 501 } 502 } 503 } 504 505 vhdx_region_register(s, s->headers[s->curr_header]->log_offset, 506 s->headers[s->curr_header]->log_length); 507 goto exit; 508 509 fail: 510 error_setg_errno(errp, -ret, "No valid VHDX header found"); 511 qemu_vfree(header1); 512 qemu_vfree(header2); 513 s->headers[0] = NULL; 514 s->headers[1] = NULL; 515 exit: 516 qemu_vfree(buffer); 517 } 518 519 520 static int vhdx_open_region_tables(BlockDriverState *bs, BDRVVHDXState *s) 521 { 522 int ret = 0; 523 uint8_t *buffer; 524 int offset = 0; 525 VHDXRegionTableEntry rt_entry; 526 uint32_t i; 527 bool bat_rt_found = false; 528 bool metadata_rt_found = false; 529 530 /* We have to read the whole 64KB block, because the crc32 is over the 531 * whole block */ 532 buffer = qemu_blockalign(bs, VHDX_HEADER_BLOCK_SIZE); 533 534 ret = bdrv_pread(bs->file, VHDX_REGION_TABLE_OFFSET, 535 VHDX_HEADER_BLOCK_SIZE, buffer, 0); 536 if (ret < 0) { 537 goto fail; 538 } 539 memcpy(&s->rt, buffer, sizeof(s->rt)); 540 offset += sizeof(s->rt); 541 542 if (!vhdx_checksum_is_valid(buffer, VHDX_HEADER_BLOCK_SIZE, 4)) { 543 ret = -EINVAL; 544 goto fail; 545 } 546 547 vhdx_region_header_le_import(&s->rt); 548 549 if (s->rt.signature != VHDX_REGION_SIGNATURE) { 550 ret = -EINVAL; 551 goto fail; 552 } 553 554 555 /* Per spec, maximum region table entry count is 2047 */ 556 if (s->rt.entry_count > 2047) { 557 ret = -EINVAL; 558 goto fail; 559 } 560 561 for (i = 0; i < s->rt.entry_count; i++) { 562 memcpy(&rt_entry, buffer + offset, sizeof(rt_entry)); 563 offset += sizeof(rt_entry); 564 565 vhdx_region_entry_le_import(&rt_entry); 566 567 /* check for region overlap between these entries, and any 568 * other memory regions in the file */ 569 ret = vhdx_region_check(s, rt_entry.file_offset, rt_entry.length); 570 if (ret < 0) { 571 goto fail; 572 } 573 574 vhdx_region_register(s, rt_entry.file_offset, rt_entry.length); 575 576 /* see if we recognize the entry */ 577 if (guid_eq(rt_entry.guid, bat_guid)) { 578 /* must be unique; if we have already found it this is invalid */ 579 if (bat_rt_found) { 580 ret = -EINVAL; 581 goto fail; 582 } 583 bat_rt_found = true; 584 s->bat_rt = rt_entry; 585 continue; 586 } 587 588 if (guid_eq(rt_entry.guid, metadata_guid)) { 589 /* must be unique; if we have already found it this is invalid */ 590 if (metadata_rt_found) { 591 ret = -EINVAL; 592 goto fail; 593 } 594 metadata_rt_found = true; 595 s->metadata_rt = rt_entry; 596 continue; 597 } 598 599 if (rt_entry.data_bits & VHDX_REGION_ENTRY_REQUIRED) { 600 /* cannot read vhdx file - required region table entry that 601 * we do not understand. per spec, we must fail to open */ 602 ret = -ENOTSUP; 603 goto fail; 604 } 605 } 606 607 if (!bat_rt_found || !metadata_rt_found) { 608 ret = -EINVAL; 609 goto fail; 610 } 611 612 ret = 0; 613 614 fail: 615 qemu_vfree(buffer); 616 return ret; 617 } 618 619 620 621 /* Metadata initial parser 622 * 623 * This loads all the metadata entry fields. This may cause additional 624 * fields to be processed (e.g. parent locator, etc..). 625 * 626 * There are 5 Metadata items that are always required: 627 * - File Parameters (block size, has a parent) 628 * - Virtual Disk Size (size, in bytes, of the virtual drive) 629 * - Page 83 Data (scsi page 83 guid) 630 * - Logical Sector Size (logical sector size in bytes, either 512 or 631 * 4096. We only support 512 currently) 632 * - Physical Sector Size (512 or 4096) 633 * 634 * Also, if the File Parameters indicate this is a differencing file, 635 * we must also look for the Parent Locator metadata item. 636 */ 637 static int vhdx_parse_metadata(BlockDriverState *bs, BDRVVHDXState *s) 638 { 639 int ret = 0; 640 uint8_t *buffer; 641 int offset = 0; 642 uint32_t i = 0; 643 VHDXMetadataTableEntry md_entry; 644 645 buffer = qemu_blockalign(bs, VHDX_METADATA_TABLE_MAX_SIZE); 646 647 ret = bdrv_pread(bs->file, s->metadata_rt.file_offset, 648 VHDX_METADATA_TABLE_MAX_SIZE, buffer, 0); 649 if (ret < 0) { 650 goto exit; 651 } 652 memcpy(&s->metadata_hdr, buffer, sizeof(s->metadata_hdr)); 653 offset += sizeof(s->metadata_hdr); 654 655 vhdx_metadata_header_le_import(&s->metadata_hdr); 656 657 if (s->metadata_hdr.signature != VHDX_METADATA_SIGNATURE) { 658 ret = -EINVAL; 659 goto exit; 660 } 661 662 s->metadata_entries.present = 0; 663 664 if ((s->metadata_hdr.entry_count * sizeof(md_entry)) > 665 (VHDX_METADATA_TABLE_MAX_SIZE - offset)) { 666 ret = -EINVAL; 667 goto exit; 668 } 669 670 for (i = 0; i < s->metadata_hdr.entry_count; i++) { 671 memcpy(&md_entry, buffer + offset, sizeof(md_entry)); 672 offset += sizeof(md_entry); 673 674 vhdx_metadata_entry_le_import(&md_entry); 675 676 if (guid_eq(md_entry.item_id, file_param_guid)) { 677 if (s->metadata_entries.present & META_FILE_PARAMETER_PRESENT) { 678 ret = -EINVAL; 679 goto exit; 680 } 681 s->metadata_entries.file_parameters_entry = md_entry; 682 s->metadata_entries.present |= META_FILE_PARAMETER_PRESENT; 683 continue; 684 } 685 686 if (guid_eq(md_entry.item_id, virtual_size_guid)) { 687 if (s->metadata_entries.present & META_VIRTUAL_DISK_SIZE_PRESENT) { 688 ret = -EINVAL; 689 goto exit; 690 } 691 s->metadata_entries.virtual_disk_size_entry = md_entry; 692 s->metadata_entries.present |= META_VIRTUAL_DISK_SIZE_PRESENT; 693 continue; 694 } 695 696 if (guid_eq(md_entry.item_id, page83_guid)) { 697 if (s->metadata_entries.present & META_PAGE_83_PRESENT) { 698 ret = -EINVAL; 699 goto exit; 700 } 701 s->metadata_entries.page83_data_entry = md_entry; 702 s->metadata_entries.present |= META_PAGE_83_PRESENT; 703 continue; 704 } 705 706 if (guid_eq(md_entry.item_id, logical_sector_guid)) { 707 if (s->metadata_entries.present & 708 META_LOGICAL_SECTOR_SIZE_PRESENT) { 709 ret = -EINVAL; 710 goto exit; 711 } 712 s->metadata_entries.logical_sector_size_entry = md_entry; 713 s->metadata_entries.present |= META_LOGICAL_SECTOR_SIZE_PRESENT; 714 continue; 715 } 716 717 if (guid_eq(md_entry.item_id, phys_sector_guid)) { 718 if (s->metadata_entries.present & META_PHYS_SECTOR_SIZE_PRESENT) { 719 ret = -EINVAL; 720 goto exit; 721 } 722 s->metadata_entries.phys_sector_size_entry = md_entry; 723 s->metadata_entries.present |= META_PHYS_SECTOR_SIZE_PRESENT; 724 continue; 725 } 726 727 if (guid_eq(md_entry.item_id, parent_locator_guid)) { 728 if (s->metadata_entries.present & META_PARENT_LOCATOR_PRESENT) { 729 ret = -EINVAL; 730 goto exit; 731 } 732 s->metadata_entries.parent_locator_entry = md_entry; 733 s->metadata_entries.present |= META_PARENT_LOCATOR_PRESENT; 734 continue; 735 } 736 737 if (md_entry.data_bits & VHDX_META_FLAGS_IS_REQUIRED) { 738 /* cannot read vhdx file - required region table entry that 739 * we do not understand. per spec, we must fail to open */ 740 ret = -ENOTSUP; 741 goto exit; 742 } 743 } 744 745 if (s->metadata_entries.present != META_ALL_PRESENT) { 746 ret = -ENOTSUP; 747 goto exit; 748 } 749 750 ret = bdrv_pread(bs->file, 751 s->metadata_entries.file_parameters_entry.offset 752 + s->metadata_rt.file_offset, 753 sizeof(s->params), 754 &s->params, 755 0); 756 757 if (ret < 0) { 758 goto exit; 759 } 760 761 s->params.block_size = le32_to_cpu(s->params.block_size); 762 s->params.data_bits = le32_to_cpu(s->params.data_bits); 763 764 765 /* We now have the file parameters, so we can tell if this is a 766 * differencing file (i.e.. has_parent), is dynamic or fixed 767 * sized (leave_blocks_allocated), and the block size */ 768 769 /* The parent locator required iff the file parameters has_parent set */ 770 if (s->params.data_bits & VHDX_PARAMS_HAS_PARENT) { 771 if (s->metadata_entries.present & META_PARENT_LOCATOR_PRESENT) { 772 /* TODO: parse parent locator fields */ 773 ret = -ENOTSUP; /* temp, until differencing files are supported */ 774 goto exit; 775 } else { 776 /* if has_parent is set, but there is not parent locator present, 777 * then that is an invalid combination */ 778 ret = -EINVAL; 779 goto exit; 780 } 781 } 782 783 /* determine virtual disk size, logical sector size, 784 * and phys sector size */ 785 786 ret = bdrv_pread(bs->file, 787 s->metadata_entries.virtual_disk_size_entry.offset 788 + s->metadata_rt.file_offset, 789 sizeof(uint64_t), 790 &s->virtual_disk_size, 791 0); 792 if (ret < 0) { 793 goto exit; 794 } 795 ret = bdrv_pread(bs->file, 796 s->metadata_entries.logical_sector_size_entry.offset 797 + s->metadata_rt.file_offset, 798 sizeof(uint32_t), 799 &s->logical_sector_size, 800 0); 801 if (ret < 0) { 802 goto exit; 803 } 804 ret = bdrv_pread(bs->file, 805 s->metadata_entries.phys_sector_size_entry.offset 806 + s->metadata_rt.file_offset, 807 sizeof(uint32_t), 808 &s->physical_sector_size, 809 0); 810 if (ret < 0) { 811 goto exit; 812 } 813 814 s->virtual_disk_size = le64_to_cpu(s->virtual_disk_size); 815 s->logical_sector_size = le32_to_cpu(s->logical_sector_size); 816 s->physical_sector_size = le32_to_cpu(s->physical_sector_size); 817 818 if (s->params.block_size < VHDX_BLOCK_SIZE_MIN || 819 s->params.block_size > VHDX_BLOCK_SIZE_MAX) { 820 ret = -EINVAL; 821 goto exit; 822 } 823 824 /* Currently we only support 512 */ 825 if (s->logical_sector_size != 512) { 826 ret = -ENOTSUP; 827 goto exit; 828 } 829 830 /* Both block_size and sector_size are guaranteed powers of 2, below. 831 Due to range checks above, s->sectors_per_block can never be < 256 */ 832 s->sectors_per_block = s->params.block_size / s->logical_sector_size; 833 s->chunk_ratio = (VHDX_MAX_SECTORS_PER_BLOCK) * 834 (uint64_t)s->logical_sector_size / 835 (uint64_t)s->params.block_size; 836 837 /* These values are ones we will want to use for division / multiplication 838 * later on, and they are all guaranteed (per the spec) to be powers of 2, 839 * so we can take advantage of that for shift operations during 840 * reads/writes */ 841 if (s->logical_sector_size & (s->logical_sector_size - 1)) { 842 ret = -EINVAL; 843 goto exit; 844 } 845 if (s->sectors_per_block & (s->sectors_per_block - 1)) { 846 ret = -EINVAL; 847 goto exit; 848 } 849 if (s->chunk_ratio & (s->chunk_ratio - 1)) { 850 ret = -EINVAL; 851 goto exit; 852 } 853 s->block_size = s->params.block_size; 854 if (s->block_size & (s->block_size - 1)) { 855 ret = -EINVAL; 856 goto exit; 857 } 858 859 vhdx_set_shift_bits(s); 860 861 ret = 0; 862 863 exit: 864 qemu_vfree(buffer); 865 return ret; 866 } 867 868 /* 869 * Calculate the number of BAT entries, including sector 870 * bitmap entries. 871 */ 872 static void vhdx_calc_bat_entries(BDRVVHDXState *s) 873 { 874 uint32_t data_blocks_cnt, bitmap_blocks_cnt; 875 876 data_blocks_cnt = DIV_ROUND_UP(s->virtual_disk_size, s->block_size); 877 bitmap_blocks_cnt = DIV_ROUND_UP(data_blocks_cnt, s->chunk_ratio); 878 879 if (s->parent_entries) { 880 s->bat_entries = bitmap_blocks_cnt * (s->chunk_ratio + 1); 881 } else { 882 s->bat_entries = data_blocks_cnt + 883 ((data_blocks_cnt - 1) >> s->chunk_ratio_bits); 884 } 885 886 } 887 888 static int vhdx_check_bat_entries(BlockDriverState *bs, int *errcnt) 889 { 890 BDRVVHDXState *s = bs->opaque; 891 int64_t image_file_size = bdrv_getlength(bs->file->bs); 892 uint64_t payblocks = s->chunk_ratio; 893 uint64_t i; 894 int ret = 0; 895 896 if (image_file_size < 0) { 897 error_report("Could not determinate VHDX image file size."); 898 return image_file_size; 899 } 900 901 for (i = 0; i < s->bat_entries; i++) { 902 if ((s->bat[i] & VHDX_BAT_STATE_BIT_MASK) == 903 PAYLOAD_BLOCK_FULLY_PRESENT) { 904 uint64_t offset = s->bat[i] & VHDX_BAT_FILE_OFF_MASK; 905 /* 906 * Allow that the last block exists only partially. The VHDX spec 907 * states that the image file can only grow in blocksize increments, 908 * but QEMU created images with partial last blocks in the past. 909 */ 910 uint32_t block_length = MIN(s->block_size, 911 bs->total_sectors * BDRV_SECTOR_SIZE - i * s->block_size); 912 /* 913 * Check for BAT entry overflow. 914 */ 915 if (offset > INT64_MAX - s->block_size) { 916 error_report("VHDX BAT entry %" PRIu64 " offset overflow.", i); 917 ret = -EINVAL; 918 if (!errcnt) { 919 break; 920 } 921 (*errcnt)++; 922 } 923 /* 924 * Check if fully allocated BAT entries do not reside after 925 * end of the image file. 926 */ 927 if (offset >= image_file_size) { 928 error_report("VHDX BAT entry %" PRIu64 " start offset %" PRIu64 929 " points after end of file (%" PRIi64 "). Image" 930 " has probably been truncated.", 931 i, offset, image_file_size); 932 ret = -EINVAL; 933 if (!errcnt) { 934 break; 935 } 936 (*errcnt)++; 937 } else if (offset + block_length > image_file_size) { 938 error_report("VHDX BAT entry %" PRIu64 " end offset %" PRIu64 939 " points after end of file (%" PRIi64 "). Image" 940 " has probably been truncated.", 941 i, offset + block_length - 1, image_file_size); 942 ret = -EINVAL; 943 if (!errcnt) { 944 break; 945 } 946 (*errcnt)++; 947 } 948 949 /* 950 * verify populated BAT field file offsets against 951 * region table and log entries 952 */ 953 if (payblocks--) { 954 /* payload bat entries */ 955 int ret2; 956 ret2 = vhdx_region_check(s, offset, s->block_size); 957 if (ret2 < 0) { 958 ret = -EINVAL; 959 if (!errcnt) { 960 break; 961 } 962 (*errcnt)++; 963 } 964 } else { 965 payblocks = s->chunk_ratio; 966 /* 967 * Once differencing files are supported, verify sector bitmap 968 * blocks here 969 */ 970 } 971 } 972 } 973 974 return ret; 975 } 976 977 static void vhdx_close(BlockDriverState *bs) 978 { 979 BDRVVHDXState *s = bs->opaque; 980 qemu_vfree(s->headers[0]); 981 s->headers[0] = NULL; 982 qemu_vfree(s->headers[1]); 983 s->headers[1] = NULL; 984 qemu_vfree(s->bat); 985 s->bat = NULL; 986 qemu_vfree(s->parent_entries); 987 s->parent_entries = NULL; 988 migrate_del_blocker(s->migration_blocker); 989 error_free(s->migration_blocker); 990 qemu_vfree(s->log.hdr); 991 s->log.hdr = NULL; 992 vhdx_region_unregister_all(s); 993 } 994 995 static int vhdx_open(BlockDriverState *bs, QDict *options, int flags, 996 Error **errp) 997 { 998 BDRVVHDXState *s = bs->opaque; 999 int ret = 0; 1000 uint32_t i; 1001 uint64_t signature; 1002 Error *local_err = NULL; 1003 1004 ret = bdrv_open_file_child(NULL, options, "file", bs, errp); 1005 if (ret < 0) { 1006 return ret; 1007 } 1008 1009 s->bat = NULL; 1010 s->first_visible_write = true; 1011 1012 qemu_co_mutex_init(&s->lock); 1013 QLIST_INIT(&s->regions); 1014 1015 /* validate the file signature */ 1016 ret = bdrv_pread(bs->file, 0, sizeof(uint64_t), &signature, 0); 1017 if (ret < 0) { 1018 goto fail; 1019 } 1020 if (memcmp(&signature, "vhdxfile", 8)) { 1021 ret = -EINVAL; 1022 goto fail; 1023 } 1024 1025 /* This is used for any header updates, for the file_write_guid. 1026 * The spec dictates that a new value should be used for the first 1027 * header update */ 1028 vhdx_guid_generate(&s->session_guid); 1029 1030 vhdx_parse_header(bs, s, &local_err); 1031 if (local_err != NULL) { 1032 error_propagate(errp, local_err); 1033 ret = -EINVAL; 1034 goto fail; 1035 } 1036 1037 ret = vhdx_parse_log(bs, s, &s->log_replayed_on_open, errp); 1038 if (ret < 0) { 1039 goto fail; 1040 } 1041 1042 ret = vhdx_open_region_tables(bs, s); 1043 if (ret < 0) { 1044 goto fail; 1045 } 1046 1047 ret = vhdx_parse_metadata(bs, s); 1048 if (ret < 0) { 1049 goto fail; 1050 } 1051 1052 s->block_size = s->params.block_size; 1053 1054 /* the VHDX spec dictates that virtual_disk_size is always a multiple of 1055 * logical_sector_size */ 1056 bs->total_sectors = s->virtual_disk_size >> s->logical_sector_size_bits; 1057 1058 vhdx_calc_bat_entries(s); 1059 1060 s->bat_offset = s->bat_rt.file_offset; 1061 1062 if (s->bat_entries > s->bat_rt.length / sizeof(VHDXBatEntry)) { 1063 /* BAT allocation is not large enough for all entries */ 1064 ret = -EINVAL; 1065 goto fail; 1066 } 1067 1068 /* s->bat is freed in vhdx_close() */ 1069 s->bat = qemu_try_blockalign(bs->file->bs, s->bat_rt.length); 1070 if (s->bat == NULL) { 1071 ret = -ENOMEM; 1072 goto fail; 1073 } 1074 1075 ret = bdrv_pread(bs->file, s->bat_offset, s->bat_rt.length, s->bat, 0); 1076 if (ret < 0) { 1077 goto fail; 1078 } 1079 1080 /* endian convert populated BAT field entries */ 1081 for (i = 0; i < s->bat_entries; i++) { 1082 s->bat[i] = le64_to_cpu(s->bat[i]); 1083 } 1084 1085 if (!(flags & BDRV_O_CHECK)) { 1086 ret = vhdx_check_bat_entries(bs, NULL); 1087 if (ret < 0) { 1088 goto fail; 1089 } 1090 } 1091 1092 /* Disable migration when VHDX images are used */ 1093 error_setg(&s->migration_blocker, "The vhdx format used by node '%s' " 1094 "does not support live migration", 1095 bdrv_get_device_or_node_name(bs)); 1096 ret = migrate_add_blocker(s->migration_blocker, errp); 1097 if (ret < 0) { 1098 error_free(s->migration_blocker); 1099 goto fail; 1100 } 1101 1102 /* TODO: differencing files */ 1103 1104 return 0; 1105 fail: 1106 vhdx_close(bs); 1107 return ret; 1108 } 1109 1110 static int vhdx_reopen_prepare(BDRVReopenState *state, 1111 BlockReopenQueue *queue, Error **errp) 1112 { 1113 return 0; 1114 } 1115 1116 1117 /* 1118 * Perform sector to block offset translations, to get various 1119 * sector and file offsets into the image. See VHDXSectorInfo 1120 */ 1121 static void vhdx_block_translate(BDRVVHDXState *s, int64_t sector_num, 1122 int nb_sectors, VHDXSectorInfo *sinfo) 1123 { 1124 uint32_t block_offset; 1125 1126 sinfo->bat_idx = sector_num >> s->sectors_per_block_bits; 1127 /* effectively a modulo - this gives us the offset into the block 1128 * (in sector sizes) for our sector number */ 1129 block_offset = sector_num - (sinfo->bat_idx << s->sectors_per_block_bits); 1130 /* the chunk ratio gives us the interleaving of the sector 1131 * bitmaps, so we need to advance our page block index by the 1132 * sector bitmaps entry number */ 1133 sinfo->bat_idx += sinfo->bat_idx >> s->chunk_ratio_bits; 1134 1135 /* the number of sectors we can read/write in this cycle */ 1136 sinfo->sectors_avail = s->sectors_per_block - block_offset; 1137 1138 sinfo->bytes_left = sinfo->sectors_avail << s->logical_sector_size_bits; 1139 1140 if (sinfo->sectors_avail > nb_sectors) { 1141 sinfo->sectors_avail = nb_sectors; 1142 } 1143 1144 sinfo->bytes_avail = sinfo->sectors_avail << s->logical_sector_size_bits; 1145 1146 sinfo->file_offset = s->bat[sinfo->bat_idx] & VHDX_BAT_FILE_OFF_MASK; 1147 1148 sinfo->block_offset = block_offset << s->logical_sector_size_bits; 1149 1150 /* The file offset must be past the header section, so must be > 0 */ 1151 if (sinfo->file_offset == 0) { 1152 return; 1153 } 1154 1155 /* block offset is the offset in vhdx logical sectors, in 1156 * the payload data block. Convert that to a byte offset 1157 * in the block, and add in the payload data block offset 1158 * in the file, in bytes, to get the final read address */ 1159 1160 sinfo->file_offset += sinfo->block_offset; 1161 } 1162 1163 1164 static int coroutine_fn 1165 vhdx_co_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) 1166 { 1167 BDRVVHDXState *s = bs->opaque; 1168 1169 bdi->cluster_size = s->block_size; 1170 1171 return 0; 1172 } 1173 1174 1175 static int coroutine_fn GRAPH_RDLOCK 1176 vhdx_co_readv(BlockDriverState *bs, int64_t sector_num, int nb_sectors, 1177 QEMUIOVector *qiov) 1178 { 1179 BDRVVHDXState *s = bs->opaque; 1180 int ret = 0; 1181 VHDXSectorInfo sinfo; 1182 uint64_t bytes_done = 0; 1183 QEMUIOVector hd_qiov; 1184 1185 qemu_iovec_init(&hd_qiov, qiov->niov); 1186 1187 qemu_co_mutex_lock(&s->lock); 1188 1189 while (nb_sectors > 0) { 1190 /* We are a differencing file, so we need to inspect the sector bitmap 1191 * to see if we have the data or not */ 1192 if (s->params.data_bits & VHDX_PARAMS_HAS_PARENT) { 1193 /* not supported yet */ 1194 ret = -ENOTSUP; 1195 goto exit; 1196 } else { 1197 vhdx_block_translate(s, sector_num, nb_sectors, &sinfo); 1198 1199 qemu_iovec_reset(&hd_qiov); 1200 qemu_iovec_concat(&hd_qiov, qiov, bytes_done, sinfo.bytes_avail); 1201 1202 /* check the payload block state */ 1203 switch (s->bat[sinfo.bat_idx] & VHDX_BAT_STATE_BIT_MASK) { 1204 case PAYLOAD_BLOCK_NOT_PRESENT: /* fall through */ 1205 case PAYLOAD_BLOCK_UNDEFINED: 1206 case PAYLOAD_BLOCK_UNMAPPED: 1207 case PAYLOAD_BLOCK_UNMAPPED_v095: 1208 case PAYLOAD_BLOCK_ZERO: 1209 /* return zero */ 1210 qemu_iovec_memset(&hd_qiov, 0, 0, sinfo.bytes_avail); 1211 break; 1212 case PAYLOAD_BLOCK_FULLY_PRESENT: 1213 qemu_co_mutex_unlock(&s->lock); 1214 ret = bdrv_co_preadv(bs->file, sinfo.file_offset, 1215 sinfo.sectors_avail * BDRV_SECTOR_SIZE, 1216 &hd_qiov, 0); 1217 qemu_co_mutex_lock(&s->lock); 1218 if (ret < 0) { 1219 goto exit; 1220 } 1221 break; 1222 case PAYLOAD_BLOCK_PARTIALLY_PRESENT: 1223 /* we don't yet support difference files, fall through 1224 * to error */ 1225 default: 1226 ret = -EIO; 1227 goto exit; 1228 break; 1229 } 1230 nb_sectors -= sinfo.sectors_avail; 1231 sector_num += sinfo.sectors_avail; 1232 bytes_done += sinfo.bytes_avail; 1233 } 1234 } 1235 ret = 0; 1236 exit: 1237 qemu_co_mutex_unlock(&s->lock); 1238 qemu_iovec_destroy(&hd_qiov); 1239 return ret; 1240 } 1241 1242 /* 1243 * Allocate a new payload block at the end of the file. 1244 * 1245 * Allocation will happen at 1MB alignment inside the file. 1246 * 1247 * If @need_zero is set on entry but not cleared on return, then truncation 1248 * could not guarantee that the new portion reads as zero, and the caller 1249 * will take care of it instead. 1250 * 1251 * Returns the file offset start of the new payload block 1252 */ 1253 static int coroutine_fn GRAPH_RDLOCK 1254 vhdx_allocate_block(BlockDriverState *bs, BDRVVHDXState *s, 1255 uint64_t *new_offset, bool *need_zero) 1256 { 1257 int64_t current_len; 1258 1259 current_len = bdrv_co_getlength(bs->file->bs); 1260 if (current_len < 0) { 1261 return current_len; 1262 } 1263 1264 *new_offset = current_len; 1265 1266 /* per the spec, the address for a block is in units of 1MB */ 1267 *new_offset = ROUND_UP(*new_offset, 1 * MiB); 1268 if (*new_offset > INT64_MAX) { 1269 return -EINVAL; 1270 } 1271 1272 if (*need_zero) { 1273 int ret; 1274 1275 ret = bdrv_co_truncate(bs->file, *new_offset + s->block_size, false, 1276 PREALLOC_MODE_OFF, BDRV_REQ_ZERO_WRITE, NULL); 1277 if (ret != -ENOTSUP) { 1278 *need_zero = false; 1279 return ret; 1280 } 1281 } 1282 1283 return bdrv_co_truncate(bs->file, *new_offset + s->block_size, false, 1284 PREALLOC_MODE_OFF, 0, NULL); 1285 } 1286 1287 /* 1288 * Update the BAT table entry with the new file offset, and the new entry 1289 * state */ 1290 static void vhdx_update_bat_table_entry(BlockDriverState *bs, BDRVVHDXState *s, 1291 VHDXSectorInfo *sinfo, 1292 uint64_t *bat_entry_le, 1293 uint64_t *bat_offset, int state) 1294 { 1295 /* The BAT entry is a uint64, with 44 bits for the file offset in units of 1296 * 1MB, and 3 bits for the block state. */ 1297 if ((state == PAYLOAD_BLOCK_ZERO) || 1298 (state == PAYLOAD_BLOCK_UNDEFINED) || 1299 (state == PAYLOAD_BLOCK_NOT_PRESENT) || 1300 (state == PAYLOAD_BLOCK_UNMAPPED)) { 1301 s->bat[sinfo->bat_idx] = 0; /* For PAYLOAD_BLOCK_ZERO, the 1302 FileOffsetMB field is denoted as 1303 'reserved' in the v1.0 spec. If it is 1304 non-zero, MS Hyper-V will fail to read 1305 the disk image */ 1306 } else { 1307 s->bat[sinfo->bat_idx] = sinfo->file_offset; 1308 } 1309 1310 s->bat[sinfo->bat_idx] |= state & VHDX_BAT_STATE_BIT_MASK; 1311 1312 *bat_entry_le = cpu_to_le64(s->bat[sinfo->bat_idx]); 1313 *bat_offset = s->bat_offset + sinfo->bat_idx * sizeof(VHDXBatEntry); 1314 1315 } 1316 1317 /* Per the spec, on the first write of guest-visible data to the file the 1318 * data write guid must be updated in the header */ 1319 int vhdx_user_visible_write(BlockDriverState *bs, BDRVVHDXState *s) 1320 { 1321 int ret = 0; 1322 if (s->first_visible_write) { 1323 s->first_visible_write = false; 1324 ret = vhdx_update_headers(bs, s, true, NULL); 1325 } 1326 return ret; 1327 } 1328 1329 static int coroutine_fn GRAPH_RDLOCK 1330 vhdx_co_writev(BlockDriverState *bs, int64_t sector_num, int nb_sectors, 1331 QEMUIOVector *qiov, int flags) 1332 { 1333 int ret = -ENOTSUP; 1334 BDRVVHDXState *s = bs->opaque; 1335 VHDXSectorInfo sinfo; 1336 uint64_t bytes_done = 0; 1337 uint64_t bat_entry = 0; 1338 uint64_t bat_entry_offset = 0; 1339 QEMUIOVector hd_qiov; 1340 struct iovec iov1 = { 0 }; 1341 struct iovec iov2 = { 0 }; 1342 int sectors_to_write; 1343 int bat_state; 1344 uint64_t bat_prior_offset = 0; 1345 bool bat_update = false; 1346 1347 qemu_iovec_init(&hd_qiov, qiov->niov); 1348 1349 qemu_co_mutex_lock(&s->lock); 1350 1351 ret = vhdx_user_visible_write(bs, s); 1352 if (ret < 0) { 1353 goto exit; 1354 } 1355 1356 while (nb_sectors > 0) { 1357 bool use_zero_buffers = false; 1358 bat_update = false; 1359 if (s->params.data_bits & VHDX_PARAMS_HAS_PARENT) { 1360 /* not supported yet */ 1361 ret = -ENOTSUP; 1362 goto exit; 1363 } else { 1364 vhdx_block_translate(s, sector_num, nb_sectors, &sinfo); 1365 sectors_to_write = sinfo.sectors_avail; 1366 1367 qemu_iovec_reset(&hd_qiov); 1368 /* check the payload block state */ 1369 bat_state = s->bat[sinfo.bat_idx] & VHDX_BAT_STATE_BIT_MASK; 1370 switch (bat_state) { 1371 case PAYLOAD_BLOCK_ZERO: 1372 /* in this case, we need to preserve zero writes for 1373 * data that is not part of this write, so we must pad 1374 * the rest of the buffer to zeroes */ 1375 use_zero_buffers = true; 1376 /* fall through */ 1377 case PAYLOAD_BLOCK_NOT_PRESENT: /* fall through */ 1378 case PAYLOAD_BLOCK_UNMAPPED: 1379 case PAYLOAD_BLOCK_UNMAPPED_v095: 1380 case PAYLOAD_BLOCK_UNDEFINED: 1381 bat_prior_offset = sinfo.file_offset; 1382 ret = vhdx_allocate_block(bs, s, &sinfo.file_offset, 1383 &use_zero_buffers); 1384 if (ret < 0) { 1385 goto exit; 1386 } 1387 /* 1388 * once we support differencing files, this may also be 1389 * partially present 1390 */ 1391 /* update block state to the newly specified state */ 1392 vhdx_update_bat_table_entry(bs, s, &sinfo, &bat_entry, 1393 &bat_entry_offset, 1394 PAYLOAD_BLOCK_FULLY_PRESENT); 1395 bat_update = true; 1396 /* 1397 * Since we just allocated a block, file_offset is the 1398 * beginning of the payload block. It needs to be the 1399 * write address, which includes the offset into the 1400 * block, unless the entire block needs to read as 1401 * zeroes but truncation was not able to provide them, 1402 * in which case we need to fill in the rest. 1403 */ 1404 if (!use_zero_buffers) { 1405 sinfo.file_offset += sinfo.block_offset; 1406 } else { 1407 /* zero fill the front, if any */ 1408 if (sinfo.block_offset) { 1409 iov1.iov_len = sinfo.block_offset; 1410 iov1.iov_base = qemu_blockalign(bs, iov1.iov_len); 1411 memset(iov1.iov_base, 0, iov1.iov_len); 1412 qemu_iovec_concat_iov(&hd_qiov, &iov1, 1, 0, 1413 iov1.iov_len); 1414 sectors_to_write += iov1.iov_len >> BDRV_SECTOR_BITS; 1415 } 1416 1417 /* our actual data */ 1418 qemu_iovec_concat(&hd_qiov, qiov, bytes_done, 1419 sinfo.bytes_avail); 1420 1421 /* zero fill the back, if any */ 1422 if ((sinfo.bytes_avail - sinfo.block_offset) < 1423 s->block_size) { 1424 iov2.iov_len = s->block_size - 1425 (sinfo.bytes_avail + sinfo.block_offset); 1426 iov2.iov_base = qemu_blockalign(bs, iov2.iov_len); 1427 memset(iov2.iov_base, 0, iov2.iov_len); 1428 qemu_iovec_concat_iov(&hd_qiov, &iov2, 1, 0, 1429 iov2.iov_len); 1430 sectors_to_write += iov2.iov_len >> BDRV_SECTOR_BITS; 1431 } 1432 } 1433 1434 /* fall through */ 1435 case PAYLOAD_BLOCK_FULLY_PRESENT: 1436 /* if the file offset address is in the header zone, 1437 * there is a problem */ 1438 if (sinfo.file_offset < (1 * MiB)) { 1439 ret = -EFAULT; 1440 goto error_bat_restore; 1441 } 1442 1443 if (!use_zero_buffers) { 1444 qemu_iovec_concat(&hd_qiov, qiov, bytes_done, 1445 sinfo.bytes_avail); 1446 } 1447 /* block exists, so we can just overwrite it */ 1448 qemu_co_mutex_unlock(&s->lock); 1449 ret = bdrv_co_pwritev(bs->file, sinfo.file_offset, 1450 sectors_to_write * BDRV_SECTOR_SIZE, 1451 &hd_qiov, 0); 1452 qemu_co_mutex_lock(&s->lock); 1453 if (ret < 0) { 1454 goto error_bat_restore; 1455 } 1456 break; 1457 case PAYLOAD_BLOCK_PARTIALLY_PRESENT: 1458 /* we don't yet support difference files, fall through 1459 * to error */ 1460 default: 1461 ret = -EIO; 1462 goto exit; 1463 break; 1464 } 1465 1466 if (bat_update) { 1467 /* this will update the BAT entry into the log journal, and 1468 * then flush the log journal out to disk */ 1469 ret = vhdx_log_write_and_flush(bs, s, &bat_entry, 1470 sizeof(VHDXBatEntry), 1471 bat_entry_offset); 1472 if (ret < 0) { 1473 goto exit; 1474 } 1475 } 1476 1477 nb_sectors -= sinfo.sectors_avail; 1478 sector_num += sinfo.sectors_avail; 1479 bytes_done += sinfo.bytes_avail; 1480 1481 } 1482 } 1483 1484 goto exit; 1485 1486 error_bat_restore: 1487 if (bat_update) { 1488 /* keep metadata in sync, and restore the bat entry state 1489 * if error. */ 1490 sinfo.file_offset = bat_prior_offset; 1491 vhdx_update_bat_table_entry(bs, s, &sinfo, &bat_entry, 1492 &bat_entry_offset, bat_state); 1493 } 1494 exit: 1495 qemu_vfree(iov1.iov_base); 1496 qemu_vfree(iov2.iov_base); 1497 qemu_co_mutex_unlock(&s->lock); 1498 qemu_iovec_destroy(&hd_qiov); 1499 return ret; 1500 } 1501 1502 1503 1504 /* 1505 * Create VHDX Headers 1506 * 1507 * There are 2 headers, and the highest sequence number will represent 1508 * the active header 1509 */ 1510 static int coroutine_fn GRAPH_UNLOCKED 1511 vhdx_create_new_headers(BlockBackend *blk, uint64_t image_size, 1512 uint32_t log_size) 1513 { 1514 BlockDriverState *bs = blk_bs(blk); 1515 BdrvChild *child; 1516 int ret = 0; 1517 VHDXHeader *hdr = NULL; 1518 1519 GRAPH_RDLOCK_GUARD(); 1520 1521 hdr = g_new0(VHDXHeader, 1); 1522 1523 hdr->signature = VHDX_HEADER_SIGNATURE; 1524 hdr->sequence_number = g_random_int(); 1525 hdr->log_version = 0; 1526 hdr->version = 1; 1527 hdr->log_length = log_size; 1528 hdr->log_offset = VHDX_HEADER_SECTION_END; 1529 vhdx_guid_generate(&hdr->file_write_guid); 1530 vhdx_guid_generate(&hdr->data_write_guid); 1531 1532 /* XXX Ugly way to get blk->root, but that's a feature, not a bug. This 1533 * hack makes it obvious that vhdx_write_header() bypasses the BlockBackend 1534 * here, which it really shouldn't be doing. */ 1535 child = QLIST_FIRST(&bs->parents); 1536 assert(!QLIST_NEXT(child, next_parent)); 1537 1538 ret = vhdx_write_header(child, hdr, VHDX_HEADER1_OFFSET, false); 1539 if (ret < 0) { 1540 goto exit; 1541 } 1542 hdr->sequence_number++; 1543 ret = vhdx_write_header(child, hdr, VHDX_HEADER2_OFFSET, false); 1544 if (ret < 0) { 1545 goto exit; 1546 } 1547 1548 exit: 1549 g_free(hdr); 1550 return ret; 1551 } 1552 1553 #define VHDX_METADATA_ENTRY_BUFFER_SIZE \ 1554 (sizeof(VHDXFileParameters) +\ 1555 sizeof(VHDXVirtualDiskSize) +\ 1556 sizeof(VHDXPage83Data) +\ 1557 sizeof(VHDXVirtualDiskLogicalSectorSize) +\ 1558 sizeof(VHDXVirtualDiskPhysicalSectorSize)) 1559 1560 /* 1561 * Create the Metadata entries. 1562 * 1563 * For more details on the entries, see section 3.5 (pg 29) in the 1564 * VHDX 1.00 specification. 1565 * 1566 * We support 5 metadata entries (all required by spec): 1567 * File Parameters, 1568 * Virtual Disk Size, 1569 * Page 83 Data, 1570 * Logical Sector Size, 1571 * Physical Sector Size 1572 * 1573 * The first 64KB of the Metadata section is reserved for the metadata 1574 * header and entries; beyond that, the metadata items themselves reside. 1575 */ 1576 static int coroutine_fn 1577 vhdx_create_new_metadata(BlockBackend *blk, uint64_t image_size, 1578 uint32_t block_size, uint32_t sector_size, 1579 uint64_t metadata_offset, VHDXImageType type) 1580 { 1581 int ret = 0; 1582 uint32_t offset = 0; 1583 void *buffer = NULL; 1584 void *entry_buffer; 1585 VHDXMetadataTableHeader *md_table; 1586 VHDXMetadataTableEntry *md_table_entry; 1587 1588 /* Metadata entries */ 1589 VHDXFileParameters *mt_file_params; 1590 VHDXVirtualDiskSize *mt_virtual_size; 1591 VHDXPage83Data *mt_page83; 1592 VHDXVirtualDiskLogicalSectorSize *mt_log_sector_size; 1593 VHDXVirtualDiskPhysicalSectorSize *mt_phys_sector_size; 1594 1595 entry_buffer = g_malloc0(VHDX_METADATA_ENTRY_BUFFER_SIZE); 1596 1597 mt_file_params = entry_buffer; 1598 offset += sizeof(VHDXFileParameters); 1599 mt_virtual_size = entry_buffer + offset; 1600 offset += sizeof(VHDXVirtualDiskSize); 1601 mt_page83 = entry_buffer + offset; 1602 offset += sizeof(VHDXPage83Data); 1603 mt_log_sector_size = entry_buffer + offset; 1604 offset += sizeof(VHDXVirtualDiskLogicalSectorSize); 1605 mt_phys_sector_size = entry_buffer + offset; 1606 1607 mt_file_params->block_size = cpu_to_le32(block_size); 1608 if (type == VHDX_TYPE_FIXED) { 1609 mt_file_params->data_bits |= VHDX_PARAMS_LEAVE_BLOCKS_ALLOCED; 1610 mt_file_params->data_bits = cpu_to_le32(mt_file_params->data_bits); 1611 } 1612 1613 vhdx_guid_generate(&mt_page83->page_83_data); 1614 cpu_to_leguids(&mt_page83->page_83_data); 1615 mt_virtual_size->virtual_disk_size = cpu_to_le64(image_size); 1616 mt_log_sector_size->logical_sector_size = cpu_to_le32(sector_size); 1617 mt_phys_sector_size->physical_sector_size = cpu_to_le32(sector_size); 1618 1619 buffer = g_malloc0(VHDX_HEADER_BLOCK_SIZE); 1620 md_table = buffer; 1621 1622 md_table->signature = VHDX_METADATA_SIGNATURE; 1623 md_table->entry_count = 5; 1624 vhdx_metadata_header_le_export(md_table); 1625 1626 1627 /* This will reference beyond the reserved table portion */ 1628 offset = 64 * KiB; 1629 1630 md_table_entry = buffer + sizeof(VHDXMetadataTableHeader); 1631 1632 md_table_entry[0].item_id = file_param_guid; 1633 md_table_entry[0].offset = offset; 1634 md_table_entry[0].length = sizeof(VHDXFileParameters); 1635 md_table_entry[0].data_bits |= VHDX_META_FLAGS_IS_REQUIRED; 1636 offset += md_table_entry[0].length; 1637 vhdx_metadata_entry_le_export(&md_table_entry[0]); 1638 1639 md_table_entry[1].item_id = virtual_size_guid; 1640 md_table_entry[1].offset = offset; 1641 md_table_entry[1].length = sizeof(VHDXVirtualDiskSize); 1642 md_table_entry[1].data_bits |= VHDX_META_FLAGS_IS_REQUIRED | 1643 VHDX_META_FLAGS_IS_VIRTUAL_DISK; 1644 offset += md_table_entry[1].length; 1645 vhdx_metadata_entry_le_export(&md_table_entry[1]); 1646 1647 md_table_entry[2].item_id = page83_guid; 1648 md_table_entry[2].offset = offset; 1649 md_table_entry[2].length = sizeof(VHDXPage83Data); 1650 md_table_entry[2].data_bits |= VHDX_META_FLAGS_IS_REQUIRED | 1651 VHDX_META_FLAGS_IS_VIRTUAL_DISK; 1652 offset += md_table_entry[2].length; 1653 vhdx_metadata_entry_le_export(&md_table_entry[2]); 1654 1655 md_table_entry[3].item_id = logical_sector_guid; 1656 md_table_entry[3].offset = offset; 1657 md_table_entry[3].length = sizeof(VHDXVirtualDiskLogicalSectorSize); 1658 md_table_entry[3].data_bits |= VHDX_META_FLAGS_IS_REQUIRED | 1659 VHDX_META_FLAGS_IS_VIRTUAL_DISK; 1660 offset += md_table_entry[3].length; 1661 vhdx_metadata_entry_le_export(&md_table_entry[3]); 1662 1663 md_table_entry[4].item_id = phys_sector_guid; 1664 md_table_entry[4].offset = offset; 1665 md_table_entry[4].length = sizeof(VHDXVirtualDiskPhysicalSectorSize); 1666 md_table_entry[4].data_bits |= VHDX_META_FLAGS_IS_REQUIRED | 1667 VHDX_META_FLAGS_IS_VIRTUAL_DISK; 1668 vhdx_metadata_entry_le_export(&md_table_entry[4]); 1669 1670 ret = blk_co_pwrite(blk, metadata_offset, VHDX_HEADER_BLOCK_SIZE, buffer, 0); 1671 if (ret < 0) { 1672 goto exit; 1673 } 1674 1675 ret = blk_co_pwrite(blk, metadata_offset + (64 * KiB), 1676 VHDX_METADATA_ENTRY_BUFFER_SIZE, entry_buffer, 0); 1677 if (ret < 0) { 1678 goto exit; 1679 } 1680 1681 1682 exit: 1683 g_free(buffer); 1684 g_free(entry_buffer); 1685 return ret; 1686 } 1687 1688 /* This create the actual BAT itself. We currently only support 1689 * 'Dynamic' and 'Fixed' image types. 1690 * 1691 * Dynamic images: default state of the BAT is all zeroes. 1692 * 1693 * Fixed images: default state of the BAT is fully populated, with 1694 * file offsets and state PAYLOAD_BLOCK_FULLY_PRESENT. 1695 */ 1696 static int coroutine_fn 1697 vhdx_create_bat(BlockBackend *blk, BDRVVHDXState *s, 1698 uint64_t image_size, VHDXImageType type, 1699 bool use_zero_blocks, uint64_t file_offset, 1700 uint32_t length, Error **errp) 1701 { 1702 int ret = 0; 1703 uint64_t data_file_offset; 1704 uint64_t total_sectors = 0; 1705 uint64_t sector_num = 0; 1706 uint64_t unused; 1707 int block_state; 1708 VHDXSectorInfo sinfo; 1709 1710 assert(s->bat == NULL); 1711 1712 /* this gives a data start after BAT/bitmap entries, and well 1713 * past any metadata entries (with a 4 MB buffer for future 1714 * expansion */ 1715 data_file_offset = file_offset + length + 5 * MiB; 1716 total_sectors = image_size >> s->logical_sector_size_bits; 1717 1718 if (type == VHDX_TYPE_DYNAMIC) { 1719 /* All zeroes, so we can just extend the file - the end of the BAT 1720 * is the furthest thing we have written yet */ 1721 ret = blk_co_truncate(blk, data_file_offset, false, PREALLOC_MODE_OFF, 1722 0, errp); 1723 if (ret < 0) { 1724 goto exit; 1725 } 1726 } else if (type == VHDX_TYPE_FIXED) { 1727 ret = blk_co_truncate(blk, data_file_offset + image_size, false, 1728 PREALLOC_MODE_OFF, 0, errp); 1729 if (ret < 0) { 1730 goto exit; 1731 } 1732 } else { 1733 error_setg(errp, "Unsupported image type"); 1734 ret = -ENOTSUP; 1735 goto exit; 1736 } 1737 1738 if (type == VHDX_TYPE_FIXED || 1739 use_zero_blocks || 1740 bdrv_has_zero_init(blk_bs(blk)) == 0) { 1741 /* for a fixed file, the default BAT entry is not zero */ 1742 s->bat = g_try_malloc0(length); 1743 if (length && s->bat == NULL) { 1744 error_setg(errp, "Failed to allocate memory for the BAT"); 1745 ret = -ENOMEM; 1746 goto exit; 1747 } 1748 block_state = type == VHDX_TYPE_FIXED ? PAYLOAD_BLOCK_FULLY_PRESENT : 1749 PAYLOAD_BLOCK_NOT_PRESENT; 1750 block_state = use_zero_blocks ? PAYLOAD_BLOCK_ZERO : block_state; 1751 /* fill the BAT by emulating sector writes of sectors_per_block size */ 1752 while (sector_num < total_sectors) { 1753 vhdx_block_translate(s, sector_num, s->sectors_per_block, &sinfo); 1754 sinfo.file_offset = data_file_offset + 1755 (sector_num << s->logical_sector_size_bits); 1756 sinfo.file_offset = ROUND_UP(sinfo.file_offset, MiB); 1757 vhdx_update_bat_table_entry(blk_bs(blk), s, &sinfo, &unused, &unused, 1758 block_state); 1759 s->bat[sinfo.bat_idx] = cpu_to_le64(s->bat[sinfo.bat_idx]); 1760 sector_num += s->sectors_per_block; 1761 } 1762 ret = blk_co_pwrite(blk, file_offset, length, s->bat, 0); 1763 if (ret < 0) { 1764 error_setg_errno(errp, -ret, "Failed to write the BAT"); 1765 goto exit; 1766 } 1767 } 1768 1769 1770 1771 exit: 1772 g_free(s->bat); 1773 return ret; 1774 } 1775 1776 /* Creates the region table header, and region table entries. 1777 * There are 2 supported region table entries: BAT, and Metadata/ 1778 * 1779 * As the calculations for the BAT region table are also needed 1780 * to create the BAT itself, we will also cause the BAT to be 1781 * created. 1782 */ 1783 static int coroutine_fn 1784 vhdx_create_new_region_table(BlockBackend *blk, uint64_t image_size, 1785 uint32_t block_size, uint32_t sector_size, 1786 uint32_t log_size, bool use_zero_blocks, 1787 VHDXImageType type, uint64_t *metadata_offset, 1788 Error **errp) 1789 { 1790 int ret = 0; 1791 uint32_t offset = 0; 1792 void *buffer = NULL; 1793 uint64_t bat_file_offset; 1794 uint32_t bat_length; 1795 BDRVVHDXState *s = NULL; 1796 VHDXRegionTableHeader *region_table; 1797 VHDXRegionTableEntry *rt_bat; 1798 VHDXRegionTableEntry *rt_metadata; 1799 1800 assert(metadata_offset != NULL); 1801 1802 /* Populate enough of the BDRVVHDXState to be able to use the 1803 * pre-existing BAT calculation, translation, and update functions */ 1804 s = g_new0(BDRVVHDXState, 1); 1805 1806 s->chunk_ratio = (VHDX_MAX_SECTORS_PER_BLOCK) * 1807 (uint64_t) sector_size / (uint64_t) block_size; 1808 1809 s->sectors_per_block = block_size / sector_size; 1810 s->virtual_disk_size = image_size; 1811 s->block_size = block_size; 1812 s->logical_sector_size = sector_size; 1813 1814 vhdx_set_shift_bits(s); 1815 1816 vhdx_calc_bat_entries(s); 1817 1818 /* At this point the VHDX state is populated enough for creation */ 1819 1820 /* a single buffer is used so we can calculate the checksum over the 1821 * entire 64KB block */ 1822 buffer = g_malloc0(VHDX_HEADER_BLOCK_SIZE); 1823 region_table = buffer; 1824 offset += sizeof(VHDXRegionTableHeader); 1825 rt_bat = buffer + offset; 1826 offset += sizeof(VHDXRegionTableEntry); 1827 rt_metadata = buffer + offset; 1828 1829 region_table->signature = VHDX_REGION_SIGNATURE; 1830 region_table->entry_count = 2; /* BAT and Metadata */ 1831 1832 rt_bat->guid = bat_guid; 1833 rt_bat->length = ROUND_UP(s->bat_entries * sizeof(VHDXBatEntry), MiB); 1834 rt_bat->file_offset = ROUND_UP(VHDX_HEADER_SECTION_END + log_size, MiB); 1835 s->bat_offset = rt_bat->file_offset; 1836 1837 rt_metadata->guid = metadata_guid; 1838 rt_metadata->file_offset = ROUND_UP(rt_bat->file_offset + rt_bat->length, 1839 MiB); 1840 rt_metadata->length = 1 * MiB; /* min size, and more than enough */ 1841 *metadata_offset = rt_metadata->file_offset; 1842 1843 bat_file_offset = rt_bat->file_offset; 1844 bat_length = rt_bat->length; 1845 1846 vhdx_region_header_le_export(region_table); 1847 vhdx_region_entry_le_export(rt_bat); 1848 vhdx_region_entry_le_export(rt_metadata); 1849 1850 vhdx_update_checksum(buffer, VHDX_HEADER_BLOCK_SIZE, 1851 offsetof(VHDXRegionTableHeader, checksum)); 1852 1853 1854 /* The region table gives us the data we need to create the BAT, 1855 * so do that now */ 1856 ret = vhdx_create_bat(blk, s, image_size, type, use_zero_blocks, 1857 bat_file_offset, bat_length, errp); 1858 if (ret < 0) { 1859 goto exit; 1860 } 1861 1862 /* Now write out the region headers to disk */ 1863 ret = blk_co_pwrite(blk, VHDX_REGION_TABLE_OFFSET, VHDX_HEADER_BLOCK_SIZE, 1864 buffer, 0); 1865 if (ret < 0) { 1866 error_setg_errno(errp, -ret, "Failed to write first region table"); 1867 goto exit; 1868 } 1869 1870 ret = blk_co_pwrite(blk, VHDX_REGION_TABLE2_OFFSET, VHDX_HEADER_BLOCK_SIZE, 1871 buffer, 0); 1872 if (ret < 0) { 1873 error_setg_errno(errp, -ret, "Failed to write second region table"); 1874 goto exit; 1875 } 1876 1877 exit: 1878 g_free(s); 1879 g_free(buffer); 1880 return ret; 1881 } 1882 1883 /* We need to create the following elements: 1884 * 1885 * .-----------------------------------------------------------------. 1886 * | (A) | (B) | (C) | (D) | (E) | 1887 * | File ID | Header1 | Header 2 | Region Tbl 1 | Region Tbl 2 | 1888 * | | | | | | 1889 * .-----------------------------------------------------------------. 1890 * 0 64KB 128KB 192KB 256KB 320KB 1891 * 1892 * 1893 * .---- ~ ----------- ~ ------------ ~ ---------------- ~ -----------. 1894 * | (F) | (G) | (H) | | 1895 * | Journal Log | BAT / Bitmap | Metadata | .... data ...... | 1896 * | | | | | 1897 * .---- ~ ----------- ~ ------------ ~ ---------------- ~ -----------. 1898 * 1MB 1899 */ 1900 static int coroutine_fn GRAPH_UNLOCKED 1901 vhdx_co_create(BlockdevCreateOptions *opts, Error **errp) 1902 { 1903 BlockdevCreateOptionsVhdx *vhdx_opts; 1904 BlockBackend *blk = NULL; 1905 BlockDriverState *bs = NULL; 1906 1907 int ret = 0; 1908 uint64_t image_size; 1909 uint32_t log_size; 1910 uint32_t block_size; 1911 uint64_t signature; 1912 uint64_t metadata_offset; 1913 bool use_zero_blocks = false; 1914 1915 gunichar2 *creator = NULL; 1916 glong creator_items; 1917 VHDXImageType image_type; 1918 1919 assert(opts->driver == BLOCKDEV_DRIVER_VHDX); 1920 vhdx_opts = &opts->u.vhdx; 1921 1922 /* Validate options and set default values */ 1923 image_size = vhdx_opts->size; 1924 if (image_size > VHDX_MAX_IMAGE_SIZE) { 1925 error_setg(errp, "Image size too large; max of 64TB"); 1926 return -EINVAL; 1927 } 1928 1929 if (!vhdx_opts->has_log_size) { 1930 log_size = DEFAULT_LOG_SIZE; 1931 } else { 1932 if (vhdx_opts->log_size > UINT32_MAX) { 1933 error_setg(errp, "Log size must be smaller than 4 GB"); 1934 return -EINVAL; 1935 } 1936 log_size = vhdx_opts->log_size; 1937 } 1938 if (log_size < MiB || (log_size % MiB) != 0) { 1939 error_setg(errp, "Log size must be a multiple of 1 MB"); 1940 return -EINVAL; 1941 } 1942 1943 if (!vhdx_opts->has_block_state_zero) { 1944 use_zero_blocks = true; 1945 } else { 1946 use_zero_blocks = vhdx_opts->block_state_zero; 1947 } 1948 1949 if (!vhdx_opts->has_subformat) { 1950 vhdx_opts->subformat = BLOCKDEV_VHDX_SUBFORMAT_DYNAMIC; 1951 } 1952 1953 switch (vhdx_opts->subformat) { 1954 case BLOCKDEV_VHDX_SUBFORMAT_DYNAMIC: 1955 image_type = VHDX_TYPE_DYNAMIC; 1956 break; 1957 case BLOCKDEV_VHDX_SUBFORMAT_FIXED: 1958 image_type = VHDX_TYPE_FIXED; 1959 break; 1960 default: 1961 g_assert_not_reached(); 1962 } 1963 1964 /* These are pretty arbitrary, and mainly designed to keep the BAT 1965 * size reasonable to load into RAM */ 1966 if (vhdx_opts->has_block_size) { 1967 block_size = vhdx_opts->block_size; 1968 } else { 1969 if (image_size > 32 * TiB) { 1970 block_size = 64 * MiB; 1971 } else if (image_size > (uint64_t) 100 * GiB) { 1972 block_size = 32 * MiB; 1973 } else if (image_size > 1 * GiB) { 1974 block_size = 16 * MiB; 1975 } else { 1976 block_size = 8 * MiB; 1977 } 1978 } 1979 1980 if (block_size < MiB || (block_size % MiB) != 0) { 1981 error_setg(errp, "Block size must be a multiple of 1 MB"); 1982 return -EINVAL; 1983 } 1984 if (!is_power_of_2(block_size)) { 1985 error_setg(errp, "Block size must be a power of two"); 1986 return -EINVAL; 1987 } 1988 if (block_size > VHDX_BLOCK_SIZE_MAX) { 1989 error_setg(errp, "Block size must not exceed %" PRId64, 1990 VHDX_BLOCK_SIZE_MAX); 1991 return -EINVAL; 1992 } 1993 1994 /* Create BlockBackend to write to the image */ 1995 bs = bdrv_co_open_blockdev_ref(vhdx_opts->file, errp); 1996 if (bs == NULL) { 1997 return -EIO; 1998 } 1999 2000 blk = blk_co_new_with_bs(bs, BLK_PERM_WRITE | BLK_PERM_RESIZE, BLK_PERM_ALL, 2001 errp); 2002 if (!blk) { 2003 ret = -EPERM; 2004 goto delete_and_exit; 2005 } 2006 blk_set_allow_write_beyond_eof(blk, true); 2007 2008 /* Create (A) */ 2009 2010 /* The creator field is optional, but may be useful for 2011 * debugging / diagnostics */ 2012 creator = g_utf8_to_utf16("QEMU v" QEMU_VERSION, -1, NULL, 2013 &creator_items, NULL); 2014 signature = cpu_to_le64(VHDX_FILE_SIGNATURE); 2015 ret = blk_co_pwrite(blk, VHDX_FILE_ID_OFFSET, sizeof(signature), &signature, 2016 0); 2017 if (ret < 0) { 2018 error_setg_errno(errp, -ret, "Failed to write file signature"); 2019 goto delete_and_exit; 2020 } 2021 if (creator) { 2022 ret = blk_co_pwrite(blk, VHDX_FILE_ID_OFFSET + sizeof(signature), 2023 creator_items * sizeof(gunichar2), creator, 0); 2024 if (ret < 0) { 2025 error_setg_errno(errp, -ret, "Failed to write creator field"); 2026 goto delete_and_exit; 2027 } 2028 } 2029 2030 2031 /* Creates (B),(C) */ 2032 ret = vhdx_create_new_headers(blk, image_size, log_size); 2033 if (ret < 0) { 2034 error_setg_errno(errp, -ret, "Failed to write image headers"); 2035 goto delete_and_exit; 2036 } 2037 2038 /* Creates (D),(E),(G) explicitly. (F) created as by-product */ 2039 ret = vhdx_create_new_region_table(blk, image_size, block_size, 512, 2040 log_size, use_zero_blocks, image_type, 2041 &metadata_offset, errp); 2042 if (ret < 0) { 2043 goto delete_and_exit; 2044 } 2045 2046 /* Creates (H) */ 2047 ret = vhdx_create_new_metadata(blk, image_size, block_size, 512, 2048 metadata_offset, image_type); 2049 if (ret < 0) { 2050 error_setg_errno(errp, -ret, "Failed to initialize metadata"); 2051 goto delete_and_exit; 2052 } 2053 2054 ret = 0; 2055 delete_and_exit: 2056 blk_co_unref(blk); 2057 bdrv_co_unref(bs); 2058 g_free(creator); 2059 return ret; 2060 } 2061 2062 static int coroutine_fn GRAPH_UNLOCKED 2063 vhdx_co_create_opts(BlockDriver *drv, const char *filename, 2064 QemuOpts *opts, Error **errp) 2065 { 2066 BlockdevCreateOptions *create_options = NULL; 2067 QDict *qdict; 2068 Visitor *v; 2069 BlockDriverState *bs = NULL; 2070 int ret; 2071 2072 static const QDictRenames opt_renames[] = { 2073 { VHDX_BLOCK_OPT_LOG_SIZE, "log-size" }, 2074 { VHDX_BLOCK_OPT_BLOCK_SIZE, "block-size" }, 2075 { VHDX_BLOCK_OPT_ZERO, "block-state-zero" }, 2076 { NULL, NULL }, 2077 }; 2078 2079 /* Parse options and convert legacy syntax */ 2080 qdict = qemu_opts_to_qdict_filtered(opts, NULL, &vhdx_create_opts, true); 2081 2082 if (!qdict_rename_keys(qdict, opt_renames, errp)) { 2083 ret = -EINVAL; 2084 goto fail; 2085 } 2086 2087 /* Create and open the file (protocol layer) */ 2088 ret = bdrv_co_create_file(filename, opts, errp); 2089 if (ret < 0) { 2090 goto fail; 2091 } 2092 2093 bs = bdrv_co_open(filename, NULL, NULL, 2094 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp); 2095 if (bs == NULL) { 2096 ret = -EIO; 2097 goto fail; 2098 } 2099 2100 /* Now get the QAPI type BlockdevCreateOptions */ 2101 qdict_put_str(qdict, "driver", "vhdx"); 2102 qdict_put_str(qdict, "file", bs->node_name); 2103 2104 v = qobject_input_visitor_new_flat_confused(qdict, errp); 2105 if (!v) { 2106 ret = -EINVAL; 2107 goto fail; 2108 } 2109 2110 visit_type_BlockdevCreateOptions(v, NULL, &create_options, errp); 2111 visit_free(v); 2112 if (!create_options) { 2113 ret = -EINVAL; 2114 goto fail; 2115 } 2116 2117 /* Silently round up sizes: 2118 * The image size is rounded to 512 bytes. Make the block and log size 2119 * close to what was specified, but must be at least 1MB, and a multiple of 2120 * 1 MB. Also respect VHDX_BLOCK_SIZE_MAX for block sizes. block_size = 0 2121 * means auto, which is represented by a missing key in QAPI. */ 2122 assert(create_options->driver == BLOCKDEV_DRIVER_VHDX); 2123 create_options->u.vhdx.size = 2124 ROUND_UP(create_options->u.vhdx.size, BDRV_SECTOR_SIZE); 2125 2126 if (create_options->u.vhdx.has_log_size) { 2127 create_options->u.vhdx.log_size = 2128 ROUND_UP(create_options->u.vhdx.log_size, MiB); 2129 } 2130 if (create_options->u.vhdx.has_block_size) { 2131 create_options->u.vhdx.block_size = 2132 ROUND_UP(create_options->u.vhdx.block_size, MiB); 2133 2134 if (create_options->u.vhdx.block_size == 0) { 2135 create_options->u.vhdx.has_block_size = false; 2136 } 2137 if (create_options->u.vhdx.block_size > VHDX_BLOCK_SIZE_MAX) { 2138 create_options->u.vhdx.block_size = VHDX_BLOCK_SIZE_MAX; 2139 } 2140 } 2141 2142 /* Create the vhdx image (format layer) */ 2143 ret = vhdx_co_create(create_options, errp); 2144 2145 fail: 2146 qobject_unref(qdict); 2147 bdrv_co_unref(bs); 2148 qapi_free_BlockdevCreateOptions(create_options); 2149 return ret; 2150 } 2151 2152 /* If opened r/w, the VHDX driver will automatically replay the log, 2153 * if one is present, inside the vhdx_open() call. 2154 * 2155 * If qemu-img check -r all is called, the image is automatically opened 2156 * r/w and any log has already been replayed, so there is nothing (currently) 2157 * for us to do here 2158 */ 2159 static int coroutine_fn vhdx_co_check(BlockDriverState *bs, 2160 BdrvCheckResult *result, 2161 BdrvCheckMode fix) 2162 { 2163 BDRVVHDXState *s = bs->opaque; 2164 2165 if (s->log_replayed_on_open) { 2166 result->corruptions_fixed++; 2167 } 2168 2169 vhdx_check_bat_entries(bs, &result->corruptions); 2170 2171 return 0; 2172 } 2173 2174 static int vhdx_has_zero_init(BlockDriverState *bs) 2175 { 2176 BDRVVHDXState *s = bs->opaque; 2177 int state; 2178 2179 /* 2180 * Check the subformat: Fixed images have all BAT entries present, 2181 * dynamic images have none (right after creation). It is 2182 * therefore enough to check the first BAT entry. 2183 */ 2184 if (!s->bat_entries) { 2185 return 1; 2186 } 2187 2188 state = s->bat[0] & VHDX_BAT_STATE_BIT_MASK; 2189 if (state == PAYLOAD_BLOCK_FULLY_PRESENT) { 2190 /* Fixed subformat */ 2191 return bdrv_has_zero_init(bs->file->bs); 2192 } 2193 2194 /* Dynamic subformat */ 2195 return 1; 2196 } 2197 2198 static QemuOptsList vhdx_create_opts = { 2199 .name = "vhdx-create-opts", 2200 .head = QTAILQ_HEAD_INITIALIZER(vhdx_create_opts.head), 2201 .desc = { 2202 { 2203 .name = BLOCK_OPT_SIZE, 2204 .type = QEMU_OPT_SIZE, 2205 .help = "Virtual disk size; max of 64TB." 2206 }, 2207 { 2208 .name = VHDX_BLOCK_OPT_LOG_SIZE, 2209 .type = QEMU_OPT_SIZE, 2210 .def_value_str = stringify(DEFAULT_LOG_SIZE), 2211 .help = "Log size; min 1MB." 2212 }, 2213 { 2214 .name = VHDX_BLOCK_OPT_BLOCK_SIZE, 2215 .type = QEMU_OPT_SIZE, 2216 .def_value_str = stringify(0), 2217 .help = "Block Size; min 1MB, max 256MB. " 2218 "0 means auto-calculate based on image size." 2219 }, 2220 { 2221 .name = BLOCK_OPT_SUBFMT, 2222 .type = QEMU_OPT_STRING, 2223 .help = "VHDX format type, can be either 'dynamic' or 'fixed'. " 2224 "Default is 'dynamic'." 2225 }, 2226 { 2227 .name = VHDX_BLOCK_OPT_ZERO, 2228 .type = QEMU_OPT_BOOL, 2229 .help = "Force use of payload blocks of type 'ZERO'. " 2230 "Non-standard, but default. Do not set to 'off' when " 2231 "using 'qemu-img convert' with subformat=dynamic." 2232 }, 2233 { NULL } 2234 } 2235 }; 2236 2237 static BlockDriver bdrv_vhdx = { 2238 .format_name = "vhdx", 2239 .instance_size = sizeof(BDRVVHDXState), 2240 .bdrv_probe = vhdx_probe, 2241 .bdrv_open = vhdx_open, 2242 .bdrv_close = vhdx_close, 2243 .bdrv_reopen_prepare = vhdx_reopen_prepare, 2244 .bdrv_child_perm = bdrv_default_perms, 2245 .bdrv_co_readv = vhdx_co_readv, 2246 .bdrv_co_writev = vhdx_co_writev, 2247 .bdrv_co_create = vhdx_co_create, 2248 .bdrv_co_create_opts = vhdx_co_create_opts, 2249 .bdrv_co_get_info = vhdx_co_get_info, 2250 .bdrv_co_check = vhdx_co_check, 2251 .bdrv_has_zero_init = vhdx_has_zero_init, 2252 2253 .is_format = true, 2254 .create_opts = &vhdx_create_opts, 2255 }; 2256 2257 static void bdrv_vhdx_init(void) 2258 { 2259 bdrv_register(&bdrv_vhdx); 2260 } 2261 2262 block_init(bdrv_vhdx_init); 2263