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-common.h" 19 #include "block/block_int.h" 20 #include "qemu/module.h" 21 #include "qemu/crc32c.h" 22 #include "block/vhdx.h" 23 #include "migration/migration.h" 24 25 #include <uuid/uuid.h> 26 #include <glib.h> 27 28 /* Options for VHDX creation */ 29 30 #define VHDX_BLOCK_OPT_LOG_SIZE "log_size" 31 #define VHDX_BLOCK_OPT_BLOCK_SIZE "block_size" 32 #define VHDX_BLOCK_OPT_ZERO "block_state_zero" 33 34 typedef enum VHDXImageType { 35 VHDX_TYPE_DYNAMIC = 0, 36 VHDX_TYPE_FIXED, 37 VHDX_TYPE_DIFFERENCING, /* Currently unsupported */ 38 } VHDXImageType; 39 40 /* Several metadata and region table data entries are identified by 41 * guids in a MS-specific GUID format. */ 42 43 44 /* ------- Known Region Table GUIDs ---------------------- */ 45 static const MSGUID bat_guid = { .data1 = 0x2dc27766, 46 .data2 = 0xf623, 47 .data3 = 0x4200, 48 .data4 = { 0x9d, 0x64, 0x11, 0x5e, 49 0x9b, 0xfd, 0x4a, 0x08} }; 50 51 static const MSGUID metadata_guid = { .data1 = 0x8b7ca206, 52 .data2 = 0x4790, 53 .data3 = 0x4b9a, 54 .data4 = { 0xb8, 0xfe, 0x57, 0x5f, 55 0x05, 0x0f, 0x88, 0x6e} }; 56 57 58 59 /* ------- Known Metadata Entry GUIDs ---------------------- */ 60 static const MSGUID file_param_guid = { .data1 = 0xcaa16737, 61 .data2 = 0xfa36, 62 .data3 = 0x4d43, 63 .data4 = { 0xb3, 0xb6, 0x33, 0xf0, 64 0xaa, 0x44, 0xe7, 0x6b} }; 65 66 static const MSGUID virtual_size_guid = { .data1 = 0x2FA54224, 67 .data2 = 0xcd1b, 68 .data3 = 0x4876, 69 .data4 = { 0xb2, 0x11, 0x5d, 0xbe, 70 0xd8, 0x3b, 0xf4, 0xb8} }; 71 72 static const MSGUID page83_guid = { .data1 = 0xbeca12ab, 73 .data2 = 0xb2e6, 74 .data3 = 0x4523, 75 .data4 = { 0x93, 0xef, 0xc3, 0x09, 76 0xe0, 0x00, 0xc7, 0x46} }; 77 78 79 static const MSGUID phys_sector_guid = { .data1 = 0xcda348c7, 80 .data2 = 0x445d, 81 .data3 = 0x4471, 82 .data4 = { 0x9c, 0xc9, 0xe9, 0x88, 83 0x52, 0x51, 0xc5, 0x56} }; 84 85 static const MSGUID parent_locator_guid = { .data1 = 0xa8d35f2d, 86 .data2 = 0xb30b, 87 .data3 = 0x454d, 88 .data4 = { 0xab, 0xf7, 0xd3, 89 0xd8, 0x48, 0x34, 90 0xab, 0x0c} }; 91 92 static const MSGUID logical_sector_guid = { .data1 = 0x8141bf1d, 93 .data2 = 0xa96f, 94 .data3 = 0x4709, 95 .data4 = { 0xba, 0x47, 0xf2, 96 0x33, 0xa8, 0xfa, 97 0xab, 0x5f} }; 98 99 /* Each parent type must have a valid GUID; this is for parent images 100 * of type 'VHDX'. If we were to allow e.g. a QCOW2 parent, we would 101 * need to make up our own QCOW2 GUID type */ 102 static const MSGUID parent_vhdx_guid = { .data1 = 0xb04aefb7, 103 .data2 = 0xd19e, 104 .data3 = 0x4a81, 105 .data4 = { 0xb7, 0x89, 0x25, 0xb8, 106 0xe9, 0x44, 0x59, 0x13} }; 107 108 109 #define META_FILE_PARAMETER_PRESENT 0x01 110 #define META_VIRTUAL_DISK_SIZE_PRESENT 0x02 111 #define META_PAGE_83_PRESENT 0x04 112 #define META_LOGICAL_SECTOR_SIZE_PRESENT 0x08 113 #define META_PHYS_SECTOR_SIZE_PRESENT 0x10 114 #define META_PARENT_LOCATOR_PRESENT 0x20 115 116 #define META_ALL_PRESENT \ 117 (META_FILE_PARAMETER_PRESENT | META_VIRTUAL_DISK_SIZE_PRESENT | \ 118 META_PAGE_83_PRESENT | META_LOGICAL_SECTOR_SIZE_PRESENT | \ 119 META_PHYS_SECTOR_SIZE_PRESENT) 120 121 122 typedef struct VHDXSectorInfo { 123 uint32_t bat_idx; /* BAT entry index */ 124 uint32_t sectors_avail; /* sectors available in payload block */ 125 uint32_t bytes_left; /* bytes left in the block after data to r/w */ 126 uint32_t bytes_avail; /* bytes available in payload block */ 127 uint64_t file_offset; /* absolute offset in bytes, in file */ 128 uint64_t block_offset; /* block offset, in bytes */ 129 } VHDXSectorInfo; 130 131 /* Calculates new checksum. 132 * 133 * Zero is substituted during crc calculation for the original crc field 134 * crc_offset: byte offset in buf of the buffer crc 135 * buf: buffer pointer 136 * size: size of buffer (must be > crc_offset+4) 137 * 138 * Note: The resulting checksum is in the CPU endianness, not necessarily 139 * in the file format endianness (LE). Any header export to disk should 140 * make sure that vhdx_header_le_export() is used to convert to the 141 * correct endianness 142 */ 143 uint32_t vhdx_update_checksum(uint8_t *buf, size_t size, int crc_offset) 144 { 145 uint32_t crc; 146 147 assert(buf != NULL); 148 assert(size > (crc_offset + sizeof(crc))); 149 150 memset(buf + crc_offset, 0, sizeof(crc)); 151 crc = crc32c(0xffffffff, buf, size); 152 memcpy(buf + crc_offset, &crc, sizeof(crc)); 153 154 return crc; 155 } 156 157 uint32_t vhdx_checksum_calc(uint32_t crc, uint8_t *buf, size_t size, 158 int crc_offset) 159 { 160 uint32_t crc_new; 161 uint32_t crc_orig; 162 assert(buf != NULL); 163 164 if (crc_offset > 0) { 165 memcpy(&crc_orig, buf + crc_offset, sizeof(crc_orig)); 166 memset(buf + crc_offset, 0, sizeof(crc_orig)); 167 } 168 169 crc_new = crc32c(crc, buf, size); 170 if (crc_offset > 0) { 171 memcpy(buf + crc_offset, &crc_orig, sizeof(crc_orig)); 172 } 173 174 return crc_new; 175 } 176 177 /* Validates the checksum of the buffer, with an in-place CRC. 178 * 179 * Zero is substituted during crc calculation for the original crc field, 180 * and the crc field is restored afterwards. But the buffer will be modifed 181 * during the calculation, so this may not be not suitable for multi-threaded 182 * use. 183 * 184 * crc_offset: byte offset in buf of the buffer crc 185 * buf: buffer pointer 186 * size: size of buffer (must be > crc_offset+4) 187 * 188 * returns true if checksum is valid, false otherwise 189 */ 190 bool vhdx_checksum_is_valid(uint8_t *buf, size_t size, int crc_offset) 191 { 192 uint32_t crc_orig; 193 uint32_t crc; 194 195 assert(buf != NULL); 196 assert(size > (crc_offset + 4)); 197 198 memcpy(&crc_orig, buf + crc_offset, sizeof(crc_orig)); 199 crc_orig = le32_to_cpu(crc_orig); 200 201 crc = vhdx_checksum_calc(0xffffffff, buf, size, crc_offset); 202 203 return crc == crc_orig; 204 } 205 206 207 /* 208 * This generates a UUID that is compliant with the MS GUIDs used 209 * in the VHDX spec (and elsewhere). 210 */ 211 void vhdx_guid_generate(MSGUID *guid) 212 { 213 uuid_t uuid; 214 assert(guid != NULL); 215 216 uuid_generate(uuid); 217 memcpy(guid, uuid, sizeof(MSGUID)); 218 } 219 220 /* Check for region overlaps inside the VHDX image */ 221 static int vhdx_region_check(BDRVVHDXState *s, uint64_t start, uint64_t length) 222 { 223 int ret = 0; 224 uint64_t end; 225 VHDXRegionEntry *r; 226 227 end = start + length; 228 QLIST_FOREACH(r, &s->regions, entries) { 229 if (!((start >= r->end) || (end <= r->start))) { 230 ret = -EINVAL; 231 goto exit; 232 } 233 } 234 235 exit: 236 return ret; 237 } 238 239 /* Register a region for future checks */ 240 static void vhdx_region_register(BDRVVHDXState *s, 241 uint64_t start, uint64_t length) 242 { 243 VHDXRegionEntry *r; 244 245 r = g_malloc0(sizeof(*r)); 246 247 r->start = start; 248 r->end = start + length; 249 250 QLIST_INSERT_HEAD(&s->regions, r, entries); 251 } 252 253 /* Free all registered regions */ 254 static void vhdx_region_unregister_all(BDRVVHDXState *s) 255 { 256 VHDXRegionEntry *r, *r_next; 257 258 QLIST_FOREACH_SAFE(r, &s->regions, entries, r_next) { 259 QLIST_REMOVE(r, entries); 260 g_free(r); 261 } 262 } 263 264 static void vhdx_set_shift_bits(BDRVVHDXState *s) 265 { 266 s->logical_sector_size_bits = 31 - clz32(s->logical_sector_size); 267 s->sectors_per_block_bits = 31 - clz32(s->sectors_per_block); 268 s->chunk_ratio_bits = 63 - clz64(s->chunk_ratio); 269 s->block_size_bits = 31 - clz32(s->block_size); 270 } 271 272 /* 273 * Per the MS VHDX Specification, for every VHDX file: 274 * - The header section is fixed size - 1 MB 275 * - The header section is always the first "object" 276 * - The first 64KB of the header is the File Identifier 277 * - The first uint64 (8 bytes) is the VHDX Signature ("vhdxfile") 278 * - The following 512 bytes constitute a UTF-16 string identifiying the 279 * software that created the file, and is optional and diagnostic only. 280 * 281 * Therefore, we probe by looking for the vhdxfile signature "vhdxfile" 282 */ 283 static int vhdx_probe(const uint8_t *buf, int buf_size, const char *filename) 284 { 285 if (buf_size >= 8 && !memcmp(buf, "vhdxfile", 8)) { 286 return 100; 287 } 288 return 0; 289 } 290 291 /* 292 * Writes the header to the specified offset. 293 * 294 * This will optionally read in buffer data from disk (otherwise zero-fill), 295 * and then update the header checksum. Header is converted to proper 296 * endianness before being written to the specified file offset 297 */ 298 static int vhdx_write_header(BlockDriverState *bs_file, VHDXHeader *hdr, 299 uint64_t offset, bool read) 300 { 301 uint8_t *buffer = NULL; 302 int ret; 303 VHDXHeader header_le; 304 305 assert(bs_file != NULL); 306 assert(hdr != NULL); 307 308 /* the header checksum is not over just the packed size of VHDXHeader, 309 * but rather over the entire 'reserved' range for the header, which is 310 * 4KB (VHDX_HEADER_SIZE). */ 311 312 buffer = qemu_blockalign(bs_file, VHDX_HEADER_SIZE); 313 if (read) { 314 /* if true, we can't assume the extra reserved bytes are 0 */ 315 ret = bdrv_pread(bs_file, offset, buffer, VHDX_HEADER_SIZE); 316 if (ret < 0) { 317 goto exit; 318 } 319 } else { 320 memset(buffer, 0, VHDX_HEADER_SIZE); 321 } 322 323 /* overwrite the actual VHDXHeader portion */ 324 memcpy(buffer, hdr, sizeof(VHDXHeader)); 325 hdr->checksum = vhdx_update_checksum(buffer, VHDX_HEADER_SIZE, 326 offsetof(VHDXHeader, checksum)); 327 vhdx_header_le_export(hdr, &header_le); 328 ret = bdrv_pwrite_sync(bs_file, offset, &header_le, sizeof(VHDXHeader)); 329 330 exit: 331 qemu_vfree(buffer); 332 return ret; 333 } 334 335 /* Update the VHDX headers 336 * 337 * This follows the VHDX spec procedures for header updates. 338 * 339 * - non-current header is updated with largest sequence number 340 */ 341 static int vhdx_update_header(BlockDriverState *bs, BDRVVHDXState *s, 342 bool generate_data_write_guid, MSGUID *log_guid) 343 { 344 int ret = 0; 345 int hdr_idx = 0; 346 uint64_t header_offset = VHDX_HEADER1_OFFSET; 347 348 VHDXHeader *active_header; 349 VHDXHeader *inactive_header; 350 351 /* operate on the non-current header */ 352 if (s->curr_header == 0) { 353 hdr_idx = 1; 354 header_offset = VHDX_HEADER2_OFFSET; 355 } 356 357 active_header = s->headers[s->curr_header]; 358 inactive_header = s->headers[hdr_idx]; 359 360 inactive_header->sequence_number = active_header->sequence_number + 1; 361 362 /* a new file guid must be generated before any file write, including 363 * headers */ 364 inactive_header->file_write_guid = s->session_guid; 365 366 /* a new data guid only needs to be generated before any guest-visible 367 * writes (i.e. something observable via virtual disk read) */ 368 if (generate_data_write_guid) { 369 vhdx_guid_generate(&inactive_header->data_write_guid); 370 } 371 372 /* update the log guid if present */ 373 if (log_guid) { 374 inactive_header->log_guid = *log_guid; 375 } 376 377 vhdx_write_header(bs->file, inactive_header, header_offset, true); 378 if (ret < 0) { 379 goto exit; 380 } 381 s->curr_header = hdr_idx; 382 383 exit: 384 return ret; 385 } 386 387 /* 388 * The VHDX spec calls for header updates to be performed twice, so that both 389 * the current and non-current header have valid info 390 */ 391 int vhdx_update_headers(BlockDriverState *bs, BDRVVHDXState *s, 392 bool generate_data_write_guid, MSGUID *log_guid) 393 { 394 int ret; 395 396 ret = vhdx_update_header(bs, s, generate_data_write_guid, log_guid); 397 if (ret < 0) { 398 return ret; 399 } 400 ret = vhdx_update_header(bs, s, generate_data_write_guid, log_guid); 401 return ret; 402 } 403 404 /* opens the specified header block from the VHDX file header section */ 405 static int vhdx_parse_header(BlockDriverState *bs, BDRVVHDXState *s) 406 { 407 int ret = 0; 408 VHDXHeader *header1; 409 VHDXHeader *header2; 410 bool h1_valid = false; 411 bool h2_valid = false; 412 uint64_t h1_seq = 0; 413 uint64_t h2_seq = 0; 414 uint8_t *buffer; 415 416 /* header1 & header2 are freed in vhdx_close() */ 417 header1 = qemu_blockalign(bs, sizeof(VHDXHeader)); 418 header2 = qemu_blockalign(bs, sizeof(VHDXHeader)); 419 420 buffer = qemu_blockalign(bs, VHDX_HEADER_SIZE); 421 422 s->headers[0] = header1; 423 s->headers[1] = header2; 424 425 /* We have to read the whole VHDX_HEADER_SIZE instead of 426 * sizeof(VHDXHeader), because the checksum is over the whole 427 * region */ 428 ret = bdrv_pread(bs->file, VHDX_HEADER1_OFFSET, buffer, VHDX_HEADER_SIZE); 429 if (ret < 0) { 430 goto fail; 431 } 432 /* copy over just the relevant portion that we need */ 433 memcpy(header1, buffer, sizeof(VHDXHeader)); 434 vhdx_header_le_import(header1); 435 436 if (vhdx_checksum_is_valid(buffer, VHDX_HEADER_SIZE, 4) && 437 !memcmp(&header1->signature, "head", 4) && 438 header1->version == 1) { 439 h1_seq = header1->sequence_number; 440 h1_valid = true; 441 } 442 443 ret = bdrv_pread(bs->file, VHDX_HEADER2_OFFSET, buffer, VHDX_HEADER_SIZE); 444 if (ret < 0) { 445 goto fail; 446 } 447 /* copy over just the relevant portion that we need */ 448 memcpy(header2, buffer, sizeof(VHDXHeader)); 449 vhdx_header_le_import(header2); 450 451 if (vhdx_checksum_is_valid(buffer, VHDX_HEADER_SIZE, 4) && 452 !memcmp(&header2->signature, "head", 4) && 453 header2->version == 1) { 454 h2_seq = header2->sequence_number; 455 h2_valid = true; 456 } 457 458 /* If there is only 1 valid header (or no valid headers), we 459 * don't care what the sequence numbers are */ 460 if (h1_valid && !h2_valid) { 461 s->curr_header = 0; 462 } else if (!h1_valid && h2_valid) { 463 s->curr_header = 1; 464 } else if (!h1_valid && !h2_valid) { 465 ret = -EINVAL; 466 goto fail; 467 } else { 468 /* If both headers are valid, then we choose the active one by the 469 * highest sequence number. If the sequence numbers are equal, that is 470 * invalid */ 471 if (h1_seq > h2_seq) { 472 s->curr_header = 0; 473 } else if (h2_seq > h1_seq) { 474 s->curr_header = 1; 475 } else { 476 ret = -EINVAL; 477 goto fail; 478 } 479 } 480 481 vhdx_region_register(s, s->headers[s->curr_header]->log_offset, 482 s->headers[s->curr_header]->log_length); 483 484 ret = 0; 485 486 goto exit; 487 488 fail: 489 qerror_report(ERROR_CLASS_GENERIC_ERROR, "No valid VHDX header found"); 490 qemu_vfree(header1); 491 qemu_vfree(header2); 492 s->headers[0] = NULL; 493 s->headers[1] = NULL; 494 exit: 495 qemu_vfree(buffer); 496 return ret; 497 } 498 499 500 static int vhdx_open_region_tables(BlockDriverState *bs, BDRVVHDXState *s) 501 { 502 int ret = 0; 503 uint8_t *buffer; 504 int offset = 0; 505 VHDXRegionTableEntry rt_entry; 506 uint32_t i; 507 bool bat_rt_found = false; 508 bool metadata_rt_found = false; 509 510 /* We have to read the whole 64KB block, because the crc32 is over the 511 * whole block */ 512 buffer = qemu_blockalign(bs, VHDX_HEADER_BLOCK_SIZE); 513 514 ret = bdrv_pread(bs->file, VHDX_REGION_TABLE_OFFSET, buffer, 515 VHDX_HEADER_BLOCK_SIZE); 516 if (ret < 0) { 517 goto fail; 518 } 519 memcpy(&s->rt, buffer, sizeof(s->rt)); 520 vhdx_region_header_le_import(&s->rt); 521 offset += sizeof(s->rt); 522 523 if (!vhdx_checksum_is_valid(buffer, VHDX_HEADER_BLOCK_SIZE, 4) || 524 memcmp(&s->rt.signature, "regi", 4)) { 525 ret = -EINVAL; 526 goto fail; 527 } 528 529 /* Per spec, maximum region table entry count is 2047 */ 530 if (s->rt.entry_count > 2047) { 531 ret = -EINVAL; 532 goto fail; 533 } 534 535 for (i = 0; i < s->rt.entry_count; i++) { 536 memcpy(&rt_entry, buffer + offset, sizeof(rt_entry)); 537 offset += sizeof(rt_entry); 538 539 vhdx_region_entry_le_import(&rt_entry); 540 541 /* check for region overlap between these entries, and any 542 * other memory regions in the file */ 543 ret = vhdx_region_check(s, rt_entry.file_offset, rt_entry.length); 544 if (ret < 0) { 545 goto fail; 546 } 547 548 vhdx_region_register(s, rt_entry.file_offset, rt_entry.length); 549 550 /* see if we recognize the entry */ 551 if (guid_eq(rt_entry.guid, bat_guid)) { 552 /* must be unique; if we have already found it this is invalid */ 553 if (bat_rt_found) { 554 ret = -EINVAL; 555 goto fail; 556 } 557 bat_rt_found = true; 558 s->bat_rt = rt_entry; 559 continue; 560 } 561 562 if (guid_eq(rt_entry.guid, metadata_guid)) { 563 /* must be unique; if we have already found it this is invalid */ 564 if (metadata_rt_found) { 565 ret = -EINVAL; 566 goto fail; 567 } 568 metadata_rt_found = true; 569 s->metadata_rt = rt_entry; 570 continue; 571 } 572 573 if (rt_entry.data_bits & VHDX_REGION_ENTRY_REQUIRED) { 574 /* cannot read vhdx file - required region table entry that 575 * we do not understand. per spec, we must fail to open */ 576 ret = -ENOTSUP; 577 goto fail; 578 } 579 } 580 581 if (!bat_rt_found || !metadata_rt_found) { 582 ret = -EINVAL; 583 goto fail; 584 } 585 586 ret = 0; 587 588 fail: 589 qemu_vfree(buffer); 590 return ret; 591 } 592 593 594 595 /* Metadata initial parser 596 * 597 * This loads all the metadata entry fields. This may cause additional 598 * fields to be processed (e.g. parent locator, etc..). 599 * 600 * There are 5 Metadata items that are always required: 601 * - File Parameters (block size, has a parent) 602 * - Virtual Disk Size (size, in bytes, of the virtual drive) 603 * - Page 83 Data (scsi page 83 guid) 604 * - Logical Sector Size (logical sector size in bytes, either 512 or 605 * 4096. We only support 512 currently) 606 * - Physical Sector Size (512 or 4096) 607 * 608 * Also, if the File Parameters indicate this is a differencing file, 609 * we must also look for the Parent Locator metadata item. 610 */ 611 static int vhdx_parse_metadata(BlockDriverState *bs, BDRVVHDXState *s) 612 { 613 int ret = 0; 614 uint8_t *buffer; 615 int offset = 0; 616 uint32_t i = 0; 617 VHDXMetadataTableEntry md_entry; 618 619 buffer = qemu_blockalign(bs, VHDX_METADATA_TABLE_MAX_SIZE); 620 621 ret = bdrv_pread(bs->file, s->metadata_rt.file_offset, buffer, 622 VHDX_METADATA_TABLE_MAX_SIZE); 623 if (ret < 0) { 624 goto exit; 625 } 626 memcpy(&s->metadata_hdr, buffer, sizeof(s->metadata_hdr)); 627 offset += sizeof(s->metadata_hdr); 628 629 vhdx_metadata_header_le_import(&s->metadata_hdr); 630 631 if (memcmp(&s->metadata_hdr.signature, "metadata", 8)) { 632 ret = -EINVAL; 633 goto exit; 634 } 635 636 s->metadata_entries.present = 0; 637 638 if ((s->metadata_hdr.entry_count * sizeof(md_entry)) > 639 (VHDX_METADATA_TABLE_MAX_SIZE - offset)) { 640 ret = -EINVAL; 641 goto exit; 642 } 643 644 for (i = 0; i < s->metadata_hdr.entry_count; i++) { 645 memcpy(&md_entry, buffer + offset, sizeof(md_entry)); 646 offset += sizeof(md_entry); 647 648 vhdx_metadata_entry_le_import(&md_entry); 649 650 if (guid_eq(md_entry.item_id, file_param_guid)) { 651 if (s->metadata_entries.present & META_FILE_PARAMETER_PRESENT) { 652 ret = -EINVAL; 653 goto exit; 654 } 655 s->metadata_entries.file_parameters_entry = md_entry; 656 s->metadata_entries.present |= META_FILE_PARAMETER_PRESENT; 657 continue; 658 } 659 660 if (guid_eq(md_entry.item_id, virtual_size_guid)) { 661 if (s->metadata_entries.present & META_VIRTUAL_DISK_SIZE_PRESENT) { 662 ret = -EINVAL; 663 goto exit; 664 } 665 s->metadata_entries.virtual_disk_size_entry = md_entry; 666 s->metadata_entries.present |= META_VIRTUAL_DISK_SIZE_PRESENT; 667 continue; 668 } 669 670 if (guid_eq(md_entry.item_id, page83_guid)) { 671 if (s->metadata_entries.present & META_PAGE_83_PRESENT) { 672 ret = -EINVAL; 673 goto exit; 674 } 675 s->metadata_entries.page83_data_entry = md_entry; 676 s->metadata_entries.present |= META_PAGE_83_PRESENT; 677 continue; 678 } 679 680 if (guid_eq(md_entry.item_id, logical_sector_guid)) { 681 if (s->metadata_entries.present & 682 META_LOGICAL_SECTOR_SIZE_PRESENT) { 683 ret = -EINVAL; 684 goto exit; 685 } 686 s->metadata_entries.logical_sector_size_entry = md_entry; 687 s->metadata_entries.present |= META_LOGICAL_SECTOR_SIZE_PRESENT; 688 continue; 689 } 690 691 if (guid_eq(md_entry.item_id, phys_sector_guid)) { 692 if (s->metadata_entries.present & META_PHYS_SECTOR_SIZE_PRESENT) { 693 ret = -EINVAL; 694 goto exit; 695 } 696 s->metadata_entries.phys_sector_size_entry = md_entry; 697 s->metadata_entries.present |= META_PHYS_SECTOR_SIZE_PRESENT; 698 continue; 699 } 700 701 if (guid_eq(md_entry.item_id, parent_locator_guid)) { 702 if (s->metadata_entries.present & META_PARENT_LOCATOR_PRESENT) { 703 ret = -EINVAL; 704 goto exit; 705 } 706 s->metadata_entries.parent_locator_entry = md_entry; 707 s->metadata_entries.present |= META_PARENT_LOCATOR_PRESENT; 708 continue; 709 } 710 711 if (md_entry.data_bits & VHDX_META_FLAGS_IS_REQUIRED) { 712 /* cannot read vhdx file - required region table entry that 713 * we do not understand. per spec, we must fail to open */ 714 ret = -ENOTSUP; 715 goto exit; 716 } 717 } 718 719 if (s->metadata_entries.present != META_ALL_PRESENT) { 720 ret = -ENOTSUP; 721 goto exit; 722 } 723 724 ret = bdrv_pread(bs->file, 725 s->metadata_entries.file_parameters_entry.offset 726 + s->metadata_rt.file_offset, 727 &s->params, 728 sizeof(s->params)); 729 730 if (ret < 0) { 731 goto exit; 732 } 733 734 le32_to_cpus(&s->params.block_size); 735 le32_to_cpus(&s->params.data_bits); 736 737 738 /* We now have the file parameters, so we can tell if this is a 739 * differencing file (i.e.. has_parent), is dynamic or fixed 740 * sized (leave_blocks_allocated), and the block size */ 741 742 /* The parent locator required iff the file parameters has_parent set */ 743 if (s->params.data_bits & VHDX_PARAMS_HAS_PARENT) { 744 if (s->metadata_entries.present & META_PARENT_LOCATOR_PRESENT) { 745 /* TODO: parse parent locator fields */ 746 ret = -ENOTSUP; /* temp, until differencing files are supported */ 747 goto exit; 748 } else { 749 /* if has_parent is set, but there is not parent locator present, 750 * then that is an invalid combination */ 751 ret = -EINVAL; 752 goto exit; 753 } 754 } 755 756 /* determine virtual disk size, logical sector size, 757 * and phys sector size */ 758 759 ret = bdrv_pread(bs->file, 760 s->metadata_entries.virtual_disk_size_entry.offset 761 + s->metadata_rt.file_offset, 762 &s->virtual_disk_size, 763 sizeof(uint64_t)); 764 if (ret < 0) { 765 goto exit; 766 } 767 ret = bdrv_pread(bs->file, 768 s->metadata_entries.logical_sector_size_entry.offset 769 + s->metadata_rt.file_offset, 770 &s->logical_sector_size, 771 sizeof(uint32_t)); 772 if (ret < 0) { 773 goto exit; 774 } 775 ret = bdrv_pread(bs->file, 776 s->metadata_entries.phys_sector_size_entry.offset 777 + s->metadata_rt.file_offset, 778 &s->physical_sector_size, 779 sizeof(uint32_t)); 780 if (ret < 0) { 781 goto exit; 782 } 783 784 le64_to_cpus(&s->virtual_disk_size); 785 le32_to_cpus(&s->logical_sector_size); 786 le32_to_cpus(&s->physical_sector_size); 787 788 if (s->logical_sector_size == 0 || s->params.block_size == 0) { 789 ret = -EINVAL; 790 goto exit; 791 } 792 793 /* both block_size and sector_size are guaranteed powers of 2 */ 794 s->sectors_per_block = s->params.block_size / s->logical_sector_size; 795 s->chunk_ratio = (VHDX_MAX_SECTORS_PER_BLOCK) * 796 (uint64_t)s->logical_sector_size / 797 (uint64_t)s->params.block_size; 798 799 /* These values are ones we will want to use for division / multiplication 800 * later on, and they are all guaranteed (per the spec) to be powers of 2, 801 * so we can take advantage of that for shift operations during 802 * reads/writes */ 803 if (s->logical_sector_size & (s->logical_sector_size - 1)) { 804 ret = -EINVAL; 805 goto exit; 806 } 807 if (s->sectors_per_block & (s->sectors_per_block - 1)) { 808 ret = -EINVAL; 809 goto exit; 810 } 811 if (s->chunk_ratio & (s->chunk_ratio - 1)) { 812 ret = -EINVAL; 813 goto exit; 814 } 815 s->block_size = s->params.block_size; 816 if (s->block_size & (s->block_size - 1)) { 817 ret = -EINVAL; 818 goto exit; 819 } 820 821 vhdx_set_shift_bits(s); 822 823 ret = 0; 824 825 exit: 826 qemu_vfree(buffer); 827 return ret; 828 } 829 830 /* 831 * Calculate the number of BAT entries, including sector 832 * bitmap entries. 833 */ 834 static void vhdx_calc_bat_entries(BDRVVHDXState *s) 835 { 836 uint32_t data_blocks_cnt, bitmap_blocks_cnt; 837 838 data_blocks_cnt = s->virtual_disk_size >> s->block_size_bits; 839 if (s->virtual_disk_size - (data_blocks_cnt << s->block_size_bits)) { 840 data_blocks_cnt++; 841 } 842 bitmap_blocks_cnt = data_blocks_cnt >> s->chunk_ratio_bits; 843 if (data_blocks_cnt - (bitmap_blocks_cnt << s->chunk_ratio_bits)) { 844 bitmap_blocks_cnt++; 845 } 846 847 if (s->parent_entries) { 848 s->bat_entries = bitmap_blocks_cnt * (s->chunk_ratio + 1); 849 } else { 850 s->bat_entries = data_blocks_cnt + 851 ((data_blocks_cnt - 1) >> s->chunk_ratio_bits); 852 } 853 854 } 855 856 static void vhdx_close(BlockDriverState *bs) 857 { 858 BDRVVHDXState *s = bs->opaque; 859 qemu_vfree(s->headers[0]); 860 s->headers[0] = NULL; 861 qemu_vfree(s->headers[1]); 862 s->headers[1] = NULL; 863 qemu_vfree(s->bat); 864 s->bat = NULL; 865 qemu_vfree(s->parent_entries); 866 s->parent_entries = NULL; 867 migrate_del_blocker(s->migration_blocker); 868 error_free(s->migration_blocker); 869 qemu_vfree(s->log.hdr); 870 s->log.hdr = NULL; 871 vhdx_region_unregister_all(s); 872 } 873 874 static int vhdx_open(BlockDriverState *bs, QDict *options, int flags, 875 Error **errp) 876 { 877 BDRVVHDXState *s = bs->opaque; 878 int ret = 0; 879 uint32_t i; 880 uint64_t signature; 881 bool log_flushed = false; 882 883 884 s->bat = NULL; 885 s->first_visible_write = true; 886 887 qemu_co_mutex_init(&s->lock); 888 QLIST_INIT(&s->regions); 889 890 /* validate the file signature */ 891 ret = bdrv_pread(bs->file, 0, &signature, sizeof(uint64_t)); 892 if (ret < 0) { 893 goto fail; 894 } 895 if (memcmp(&signature, "vhdxfile", 8)) { 896 ret = -EINVAL; 897 goto fail; 898 } 899 900 /* This is used for any header updates, for the file_write_guid. 901 * The spec dictates that a new value should be used for the first 902 * header update */ 903 vhdx_guid_generate(&s->session_guid); 904 905 ret = vhdx_parse_header(bs, s); 906 if (ret < 0) { 907 goto fail; 908 } 909 910 ret = vhdx_parse_log(bs, s, &log_flushed); 911 if (ret < 0) { 912 goto fail; 913 } 914 915 ret = vhdx_open_region_tables(bs, s); 916 if (ret < 0) { 917 goto fail; 918 } 919 920 ret = vhdx_parse_metadata(bs, s); 921 if (ret < 0) { 922 goto fail; 923 } 924 925 s->block_size = s->params.block_size; 926 927 /* the VHDX spec dictates that virtual_disk_size is always a multiple of 928 * logical_sector_size */ 929 bs->total_sectors = s->virtual_disk_size >> s->logical_sector_size_bits; 930 931 vhdx_calc_bat_entries(s); 932 933 s->bat_offset = s->bat_rt.file_offset; 934 935 if (s->bat_entries > s->bat_rt.length / sizeof(VHDXBatEntry)) { 936 /* BAT allocation is not large enough for all entries */ 937 ret = -EINVAL; 938 goto fail; 939 } 940 941 /* s->bat is freed in vhdx_close() */ 942 s->bat = qemu_blockalign(bs, s->bat_rt.length); 943 944 ret = bdrv_pread(bs->file, s->bat_offset, s->bat, s->bat_rt.length); 945 if (ret < 0) { 946 goto fail; 947 } 948 949 uint64_t payblocks = s->chunk_ratio; 950 /* endian convert, and verify populated BAT field file offsets against 951 * region table and log entries */ 952 for (i = 0; i < s->bat_entries; i++) { 953 le64_to_cpus(&s->bat[i]); 954 if (payblocks--) { 955 /* payload bat entries */ 956 if ((s->bat[i] & VHDX_BAT_STATE_BIT_MASK) == 957 PAYLOAD_BLOCK_FULLY_PRESENT) { 958 ret = vhdx_region_check(s, s->bat[i] & VHDX_BAT_FILE_OFF_MASK, 959 s->block_size); 960 if (ret < 0) { 961 goto fail; 962 } 963 } 964 } else { 965 payblocks = s->chunk_ratio; 966 /* Once differencing files are supported, verify sector bitmap 967 * blocks here */ 968 } 969 } 970 971 if (flags & BDRV_O_RDWR) { 972 ret = vhdx_update_headers(bs, s, false, NULL); 973 if (ret < 0) { 974 goto fail; 975 } 976 } 977 978 /* TODO: differencing files */ 979 980 /* Disable migration when VHDX images are used */ 981 error_set(&s->migration_blocker, 982 QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED, 983 "vhdx", bs->device_name, "live migration"); 984 migrate_add_blocker(s->migration_blocker); 985 986 return 0; 987 fail: 988 vhdx_close(bs); 989 return ret; 990 } 991 992 static int vhdx_reopen_prepare(BDRVReopenState *state, 993 BlockReopenQueue *queue, Error **errp) 994 { 995 return 0; 996 } 997 998 999 /* 1000 * Perform sector to block offset translations, to get various 1001 * sector and file offsets into the image. See VHDXSectorInfo 1002 */ 1003 static void vhdx_block_translate(BDRVVHDXState *s, int64_t sector_num, 1004 int nb_sectors, VHDXSectorInfo *sinfo) 1005 { 1006 uint32_t block_offset; 1007 1008 sinfo->bat_idx = sector_num >> s->sectors_per_block_bits; 1009 /* effectively a modulo - this gives us the offset into the block 1010 * (in sector sizes) for our sector number */ 1011 block_offset = sector_num - (sinfo->bat_idx << s->sectors_per_block_bits); 1012 /* the chunk ratio gives us the interleaving of the sector 1013 * bitmaps, so we need to advance our page block index by the 1014 * sector bitmaps entry number */ 1015 sinfo->bat_idx += sinfo->bat_idx >> s->chunk_ratio_bits; 1016 1017 /* the number of sectors we can read/write in this cycle */ 1018 sinfo->sectors_avail = s->sectors_per_block - block_offset; 1019 1020 sinfo->bytes_left = sinfo->sectors_avail << s->logical_sector_size_bits; 1021 1022 if (sinfo->sectors_avail > nb_sectors) { 1023 sinfo->sectors_avail = nb_sectors; 1024 } 1025 1026 sinfo->bytes_avail = sinfo->sectors_avail << s->logical_sector_size_bits; 1027 1028 sinfo->file_offset = s->bat[sinfo->bat_idx] & VHDX_BAT_FILE_OFF_MASK; 1029 1030 sinfo->block_offset = block_offset << s->logical_sector_size_bits; 1031 1032 /* The file offset must be past the header section, so must be > 0 */ 1033 if (sinfo->file_offset == 0) { 1034 return; 1035 } 1036 1037 /* block offset is the offset in vhdx logical sectors, in 1038 * the payload data block. Convert that to a byte offset 1039 * in the block, and add in the payload data block offset 1040 * in the file, in bytes, to get the final read address */ 1041 1042 sinfo->file_offset += sinfo->block_offset; 1043 } 1044 1045 1046 static int vhdx_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) 1047 { 1048 BDRVVHDXState *s = bs->opaque; 1049 1050 bdi->cluster_size = s->block_size; 1051 1052 bdi->unallocated_blocks_are_zero = 1053 (s->params.data_bits & VHDX_PARAMS_HAS_PARENT) == 0; 1054 1055 return 0; 1056 } 1057 1058 1059 static coroutine_fn int vhdx_co_readv(BlockDriverState *bs, int64_t sector_num, 1060 int nb_sectors, QEMUIOVector *qiov) 1061 { 1062 BDRVVHDXState *s = bs->opaque; 1063 int ret = 0; 1064 VHDXSectorInfo sinfo; 1065 uint64_t bytes_done = 0; 1066 QEMUIOVector hd_qiov; 1067 1068 qemu_iovec_init(&hd_qiov, qiov->niov); 1069 1070 qemu_co_mutex_lock(&s->lock); 1071 1072 while (nb_sectors > 0) { 1073 /* We are a differencing file, so we need to inspect the sector bitmap 1074 * to see if we have the data or not */ 1075 if (s->params.data_bits & VHDX_PARAMS_HAS_PARENT) { 1076 /* not supported yet */ 1077 ret = -ENOTSUP; 1078 goto exit; 1079 } else { 1080 vhdx_block_translate(s, sector_num, nb_sectors, &sinfo); 1081 1082 qemu_iovec_reset(&hd_qiov); 1083 qemu_iovec_concat(&hd_qiov, qiov, bytes_done, sinfo.bytes_avail); 1084 1085 /* check the payload block state */ 1086 switch (s->bat[sinfo.bat_idx] & VHDX_BAT_STATE_BIT_MASK) { 1087 case PAYLOAD_BLOCK_NOT_PRESENT: /* fall through */ 1088 case PAYLOAD_BLOCK_UNDEFINED: /* fall through */ 1089 case PAYLOAD_BLOCK_UNMAPPED: /* fall through */ 1090 case PAYLOAD_BLOCK_ZERO: 1091 /* return zero */ 1092 qemu_iovec_memset(&hd_qiov, 0, 0, sinfo.bytes_avail); 1093 break; 1094 case PAYLOAD_BLOCK_FULLY_PRESENT: 1095 qemu_co_mutex_unlock(&s->lock); 1096 ret = bdrv_co_readv(bs->file, 1097 sinfo.file_offset >> BDRV_SECTOR_BITS, 1098 sinfo.sectors_avail, &hd_qiov); 1099 qemu_co_mutex_lock(&s->lock); 1100 if (ret < 0) { 1101 goto exit; 1102 } 1103 break; 1104 case PAYLOAD_BLOCK_PARTIALLY_PRESENT: 1105 /* we don't yet support difference files, fall through 1106 * to error */ 1107 default: 1108 ret = -EIO; 1109 goto exit; 1110 break; 1111 } 1112 nb_sectors -= sinfo.sectors_avail; 1113 sector_num += sinfo.sectors_avail; 1114 bytes_done += sinfo.bytes_avail; 1115 } 1116 } 1117 ret = 0; 1118 exit: 1119 qemu_co_mutex_unlock(&s->lock); 1120 qemu_iovec_destroy(&hd_qiov); 1121 return ret; 1122 } 1123 1124 /* 1125 * Allocate a new payload block at the end of the file. 1126 * 1127 * Allocation will happen at 1MB alignment inside the file 1128 * 1129 * Returns the file offset start of the new payload block 1130 */ 1131 static int vhdx_allocate_block(BlockDriverState *bs, BDRVVHDXState *s, 1132 uint64_t *new_offset) 1133 { 1134 *new_offset = bdrv_getlength(bs->file); 1135 1136 /* per the spec, the address for a block is in units of 1MB */ 1137 *new_offset = ROUND_UP(*new_offset, 1024 * 1024); 1138 1139 return bdrv_truncate(bs->file, *new_offset + s->block_size); 1140 } 1141 1142 /* 1143 * Update the BAT table entry with the new file offset, and the new entry 1144 * state */ 1145 static void vhdx_update_bat_table_entry(BlockDriverState *bs, BDRVVHDXState *s, 1146 VHDXSectorInfo *sinfo, 1147 uint64_t *bat_entry_le, 1148 uint64_t *bat_offset, int state) 1149 { 1150 /* The BAT entry is a uint64, with 44 bits for the file offset in units of 1151 * 1MB, and 3 bits for the block state. */ 1152 s->bat[sinfo->bat_idx] = sinfo->file_offset; 1153 1154 s->bat[sinfo->bat_idx] |= state & VHDX_BAT_STATE_BIT_MASK; 1155 1156 *bat_entry_le = cpu_to_le64(s->bat[sinfo->bat_idx]); 1157 *bat_offset = s->bat_offset + sinfo->bat_idx * sizeof(VHDXBatEntry); 1158 1159 } 1160 1161 /* Per the spec, on the first write of guest-visible data to the file the 1162 * data write guid must be updated in the header */ 1163 int vhdx_user_visible_write(BlockDriverState *bs, BDRVVHDXState *s) 1164 { 1165 int ret = 0; 1166 if (s->first_visible_write) { 1167 s->first_visible_write = false; 1168 ret = vhdx_update_headers(bs, s, true, NULL); 1169 } 1170 return ret; 1171 } 1172 1173 static coroutine_fn int vhdx_co_writev(BlockDriverState *bs, int64_t sector_num, 1174 int nb_sectors, QEMUIOVector *qiov) 1175 { 1176 int ret = -ENOTSUP; 1177 BDRVVHDXState *s = bs->opaque; 1178 VHDXSectorInfo sinfo; 1179 uint64_t bytes_done = 0; 1180 uint64_t bat_entry = 0; 1181 uint64_t bat_entry_offset = 0; 1182 QEMUIOVector hd_qiov; 1183 struct iovec iov1 = { 0 }; 1184 struct iovec iov2 = { 0 }; 1185 int sectors_to_write; 1186 int bat_state; 1187 uint64_t bat_prior_offset = 0; 1188 bool bat_update = false; 1189 1190 qemu_iovec_init(&hd_qiov, qiov->niov); 1191 1192 qemu_co_mutex_lock(&s->lock); 1193 1194 ret = vhdx_user_visible_write(bs, s); 1195 if (ret < 0) { 1196 goto exit; 1197 } 1198 1199 while (nb_sectors > 0) { 1200 bool use_zero_buffers = false; 1201 bat_update = false; 1202 if (s->params.data_bits & VHDX_PARAMS_HAS_PARENT) { 1203 /* not supported yet */ 1204 ret = -ENOTSUP; 1205 goto exit; 1206 } else { 1207 vhdx_block_translate(s, sector_num, nb_sectors, &sinfo); 1208 sectors_to_write = sinfo.sectors_avail; 1209 1210 qemu_iovec_reset(&hd_qiov); 1211 /* check the payload block state */ 1212 bat_state = s->bat[sinfo.bat_idx] & VHDX_BAT_STATE_BIT_MASK; 1213 switch (bat_state) { 1214 case PAYLOAD_BLOCK_ZERO: 1215 /* in this case, we need to preserve zero writes for 1216 * data that is not part of this write, so we must pad 1217 * the rest of the buffer to zeroes */ 1218 1219 /* if we are on a posix system with ftruncate() that extends 1220 * a file, then it is zero-filled for us. On Win32, the raw 1221 * layer uses SetFilePointer and SetFileEnd, which does not 1222 * zero fill AFAIK */ 1223 1224 /* Queue another write of zero buffers if the underlying file 1225 * does not zero-fill on file extension */ 1226 1227 if (bdrv_has_zero_init(bs->file) == 0) { 1228 use_zero_buffers = true; 1229 1230 /* zero fill the front, if any */ 1231 if (sinfo.block_offset) { 1232 iov1.iov_len = sinfo.block_offset; 1233 iov1.iov_base = qemu_blockalign(bs, iov1.iov_len); 1234 memset(iov1.iov_base, 0, iov1.iov_len); 1235 qemu_iovec_concat_iov(&hd_qiov, &iov1, 1, 0, 1236 sinfo.block_offset); 1237 sectors_to_write += iov1.iov_len >> BDRV_SECTOR_BITS; 1238 } 1239 1240 /* our actual data */ 1241 qemu_iovec_concat(&hd_qiov, qiov, bytes_done, 1242 sinfo.bytes_avail); 1243 1244 /* zero fill the back, if any */ 1245 if ((sinfo.bytes_avail - sinfo.block_offset) < 1246 s->block_size) { 1247 iov2.iov_len = s->block_size - 1248 (sinfo.bytes_avail + sinfo.block_offset); 1249 iov2.iov_base = qemu_blockalign(bs, iov2.iov_len); 1250 memset(iov2.iov_base, 0, iov2.iov_len); 1251 qemu_iovec_concat_iov(&hd_qiov, &iov2, 1, 0, 1252 sinfo.block_offset); 1253 sectors_to_write += iov2.iov_len >> BDRV_SECTOR_BITS; 1254 } 1255 } 1256 1257 /* fall through */ 1258 case PAYLOAD_BLOCK_NOT_PRESENT: /* fall through */ 1259 case PAYLOAD_BLOCK_UNMAPPED: /* fall through */ 1260 case PAYLOAD_BLOCK_UNDEFINED: /* fall through */ 1261 bat_prior_offset = sinfo.file_offset; 1262 ret = vhdx_allocate_block(bs, s, &sinfo.file_offset); 1263 if (ret < 0) { 1264 goto exit; 1265 } 1266 /* once we support differencing files, this may also be 1267 * partially present */ 1268 /* update block state to the newly specified state */ 1269 vhdx_update_bat_table_entry(bs, s, &sinfo, &bat_entry, 1270 &bat_entry_offset, 1271 PAYLOAD_BLOCK_FULLY_PRESENT); 1272 bat_update = true; 1273 /* since we just allocated a block, file_offset is the 1274 * beginning of the payload block. It needs to be the 1275 * write address, which includes the offset into the block */ 1276 if (!use_zero_buffers) { 1277 sinfo.file_offset += sinfo.block_offset; 1278 } 1279 /* fall through */ 1280 case PAYLOAD_BLOCK_FULLY_PRESENT: 1281 /* if the file offset address is in the header zone, 1282 * there is a problem */ 1283 if (sinfo.file_offset < (1024 * 1024)) { 1284 ret = -EFAULT; 1285 goto error_bat_restore; 1286 } 1287 1288 if (!use_zero_buffers) { 1289 qemu_iovec_concat(&hd_qiov, qiov, bytes_done, 1290 sinfo.bytes_avail); 1291 } 1292 /* block exists, so we can just overwrite it */ 1293 qemu_co_mutex_unlock(&s->lock); 1294 ret = bdrv_co_writev(bs->file, 1295 sinfo.file_offset >> BDRV_SECTOR_BITS, 1296 sectors_to_write, &hd_qiov); 1297 qemu_co_mutex_lock(&s->lock); 1298 if (ret < 0) { 1299 goto error_bat_restore; 1300 } 1301 break; 1302 case PAYLOAD_BLOCK_PARTIALLY_PRESENT: 1303 /* we don't yet support difference files, fall through 1304 * to error */ 1305 default: 1306 ret = -EIO; 1307 goto exit; 1308 break; 1309 } 1310 1311 if (bat_update) { 1312 /* this will update the BAT entry into the log journal, and 1313 * then flush the log journal out to disk */ 1314 ret = vhdx_log_write_and_flush(bs, s, &bat_entry, 1315 sizeof(VHDXBatEntry), 1316 bat_entry_offset); 1317 if (ret < 0) { 1318 goto exit; 1319 } 1320 } 1321 1322 nb_sectors -= sinfo.sectors_avail; 1323 sector_num += sinfo.sectors_avail; 1324 bytes_done += sinfo.bytes_avail; 1325 1326 } 1327 } 1328 1329 goto exit; 1330 1331 error_bat_restore: 1332 if (bat_update) { 1333 /* keep metadata in sync, and restore the bat entry state 1334 * if error. */ 1335 sinfo.file_offset = bat_prior_offset; 1336 vhdx_update_bat_table_entry(bs, s, &sinfo, &bat_entry, 1337 &bat_entry_offset, bat_state); 1338 } 1339 exit: 1340 qemu_vfree(iov1.iov_base); 1341 qemu_vfree(iov2.iov_base); 1342 qemu_co_mutex_unlock(&s->lock); 1343 qemu_iovec_destroy(&hd_qiov); 1344 return ret; 1345 } 1346 1347 1348 1349 /* 1350 * Create VHDX Headers 1351 * 1352 * There are 2 headers, and the highest sequence number will represent 1353 * the active header 1354 */ 1355 static int vhdx_create_new_headers(BlockDriverState *bs, uint64_t image_size, 1356 uint32_t log_size) 1357 { 1358 int ret = 0; 1359 VHDXHeader *hdr = NULL; 1360 1361 hdr = g_malloc0(sizeof(VHDXHeader)); 1362 1363 hdr->signature = VHDX_HEADER_SIGNATURE; 1364 hdr->sequence_number = g_random_int(); 1365 hdr->log_version = 0; 1366 hdr->version = 1; 1367 hdr->log_length = log_size; 1368 hdr->log_offset = VHDX_HEADER_SECTION_END; 1369 vhdx_guid_generate(&hdr->file_write_guid); 1370 vhdx_guid_generate(&hdr->data_write_guid); 1371 1372 ret = vhdx_write_header(bs, hdr, VHDX_HEADER1_OFFSET, false); 1373 if (ret < 0) { 1374 goto exit; 1375 } 1376 hdr->sequence_number++; 1377 ret = vhdx_write_header(bs, hdr, VHDX_HEADER2_OFFSET, false); 1378 if (ret < 0) { 1379 goto exit; 1380 } 1381 1382 exit: 1383 g_free(hdr); 1384 return ret; 1385 } 1386 1387 1388 /* 1389 * Create the Metadata entries. 1390 * 1391 * For more details on the entries, see section 3.5 (pg 29) in the 1392 * VHDX 1.00 specification. 1393 * 1394 * We support 5 metadata entries (all required by spec): 1395 * File Parameters, 1396 * Virtual Disk Size, 1397 * Page 83 Data, 1398 * Logical Sector Size, 1399 * Physical Sector Size 1400 * 1401 * The first 64KB of the Metadata section is reserved for the metadata 1402 * header and entries; beyond that, the metadata items themselves reside. 1403 */ 1404 static int vhdx_create_new_metadata(BlockDriverState *bs, 1405 uint64_t image_size, 1406 uint32_t block_size, 1407 uint32_t sector_size, 1408 uint64_t metadata_offset, 1409 VHDXImageType type) 1410 { 1411 int ret = 0; 1412 uint32_t offset = 0; 1413 void *buffer = NULL; 1414 void *entry_buffer; 1415 VHDXMetadataTableHeader *md_table;; 1416 VHDXMetadataTableEntry *md_table_entry; 1417 1418 /* Metadata entries */ 1419 VHDXFileParameters *mt_file_params; 1420 VHDXVirtualDiskSize *mt_virtual_size; 1421 VHDXPage83Data *mt_page83; 1422 VHDXVirtualDiskLogicalSectorSize *mt_log_sector_size; 1423 VHDXVirtualDiskPhysicalSectorSize *mt_phys_sector_size; 1424 1425 entry_buffer = g_malloc0(sizeof(VHDXFileParameters) + 1426 sizeof(VHDXVirtualDiskSize) + 1427 sizeof(VHDXPage83Data) + 1428 sizeof(VHDXVirtualDiskLogicalSectorSize) + 1429 sizeof(VHDXVirtualDiskPhysicalSectorSize)); 1430 1431 mt_file_params = entry_buffer; 1432 offset += sizeof(VHDXFileParameters); 1433 mt_virtual_size = entry_buffer + offset; 1434 offset += sizeof(VHDXVirtualDiskSize); 1435 mt_page83 = entry_buffer + offset; 1436 offset += sizeof(VHDXPage83Data); 1437 mt_log_sector_size = entry_buffer + offset; 1438 offset += sizeof(VHDXVirtualDiskLogicalSectorSize); 1439 mt_phys_sector_size = entry_buffer + offset; 1440 1441 mt_file_params->block_size = cpu_to_le32(block_size); 1442 if (type == VHDX_TYPE_FIXED) { 1443 mt_file_params->data_bits |= VHDX_PARAMS_LEAVE_BLOCKS_ALLOCED; 1444 cpu_to_le32s(&mt_file_params->data_bits); 1445 } 1446 1447 vhdx_guid_generate(&mt_page83->page_83_data); 1448 cpu_to_leguids(&mt_page83->page_83_data); 1449 mt_virtual_size->virtual_disk_size = cpu_to_le64(image_size); 1450 mt_log_sector_size->logical_sector_size = cpu_to_le32(sector_size); 1451 mt_phys_sector_size->physical_sector_size = cpu_to_le32(sector_size); 1452 1453 buffer = g_malloc0(VHDX_HEADER_BLOCK_SIZE); 1454 md_table = buffer; 1455 1456 md_table->signature = VHDX_METADATA_SIGNATURE; 1457 md_table->entry_count = 5; 1458 vhdx_metadata_header_le_export(md_table); 1459 1460 1461 /* This will reference beyond the reserved table portion */ 1462 offset = 64 * KiB; 1463 1464 md_table_entry = buffer + sizeof(VHDXMetadataTableHeader); 1465 1466 md_table_entry[0].item_id = file_param_guid; 1467 md_table_entry[0].offset = offset; 1468 md_table_entry[0].length = sizeof(VHDXFileParameters); 1469 md_table_entry[0].data_bits |= VHDX_META_FLAGS_IS_REQUIRED; 1470 offset += md_table_entry[0].length; 1471 vhdx_metadata_entry_le_export(&md_table_entry[0]); 1472 1473 md_table_entry[1].item_id = virtual_size_guid; 1474 md_table_entry[1].offset = offset; 1475 md_table_entry[1].length = sizeof(VHDXVirtualDiskSize); 1476 md_table_entry[1].data_bits |= VHDX_META_FLAGS_IS_REQUIRED | 1477 VHDX_META_FLAGS_IS_VIRTUAL_DISK; 1478 offset += md_table_entry[1].length; 1479 vhdx_metadata_entry_le_export(&md_table_entry[1]); 1480 1481 md_table_entry[2].item_id = page83_guid; 1482 md_table_entry[2].offset = offset; 1483 md_table_entry[2].length = sizeof(VHDXPage83Data); 1484 md_table_entry[2].data_bits |= VHDX_META_FLAGS_IS_REQUIRED | 1485 VHDX_META_FLAGS_IS_VIRTUAL_DISK; 1486 offset += md_table_entry[2].length; 1487 vhdx_metadata_entry_le_export(&md_table_entry[2]); 1488 1489 md_table_entry[3].item_id = logical_sector_guid; 1490 md_table_entry[3].offset = offset; 1491 md_table_entry[3].length = sizeof(VHDXVirtualDiskLogicalSectorSize); 1492 md_table_entry[3].data_bits |= VHDX_META_FLAGS_IS_REQUIRED | 1493 VHDX_META_FLAGS_IS_VIRTUAL_DISK; 1494 offset += md_table_entry[3].length; 1495 vhdx_metadata_entry_le_export(&md_table_entry[3]); 1496 1497 md_table_entry[4].item_id = phys_sector_guid; 1498 md_table_entry[4].offset = offset; 1499 md_table_entry[4].length = sizeof(VHDXVirtualDiskPhysicalSectorSize); 1500 md_table_entry[4].data_bits |= VHDX_META_FLAGS_IS_REQUIRED | 1501 VHDX_META_FLAGS_IS_VIRTUAL_DISK; 1502 vhdx_metadata_entry_le_export(&md_table_entry[4]); 1503 1504 ret = bdrv_pwrite(bs, metadata_offset, buffer, VHDX_HEADER_BLOCK_SIZE); 1505 if (ret < 0) { 1506 goto exit; 1507 } 1508 1509 ret = bdrv_pwrite(bs, metadata_offset + (64 * KiB), entry_buffer, 1510 VHDX_HEADER_BLOCK_SIZE); 1511 if (ret < 0) { 1512 goto exit; 1513 } 1514 1515 1516 exit: 1517 g_free(buffer); 1518 g_free(entry_buffer); 1519 return ret; 1520 } 1521 1522 /* This create the actual BAT itself. We currently only support 1523 * 'Dynamic' and 'Fixed' image types. 1524 * 1525 * Dynamic images: default state of the BAT is all zeroes. 1526 * 1527 * Fixed images: default state of the BAT is fully populated, with 1528 * file offsets and state PAYLOAD_BLOCK_FULLY_PRESENT. 1529 */ 1530 static int vhdx_create_bat(BlockDriverState *bs, BDRVVHDXState *s, 1531 uint64_t image_size, VHDXImageType type, 1532 bool use_zero_blocks, VHDXRegionTableEntry *rt_bat) 1533 { 1534 int ret = 0; 1535 uint64_t data_file_offset; 1536 uint64_t total_sectors = 0; 1537 uint64_t sector_num = 0; 1538 uint64_t unused; 1539 int block_state; 1540 VHDXSectorInfo sinfo; 1541 1542 assert(s->bat == NULL); 1543 1544 /* this gives a data start after BAT/bitmap entries, and well 1545 * past any metadata entries (with a 4 MB buffer for future 1546 * expansion */ 1547 data_file_offset = rt_bat->file_offset + rt_bat->length + 5 * MiB; 1548 total_sectors = image_size >> s->logical_sector_size_bits; 1549 1550 if (type == VHDX_TYPE_DYNAMIC) { 1551 /* All zeroes, so we can just extend the file - the end of the BAT 1552 * is the furthest thing we have written yet */ 1553 ret = bdrv_truncate(bs, data_file_offset); 1554 if (ret < 0) { 1555 goto exit; 1556 } 1557 } else if (type == VHDX_TYPE_FIXED) { 1558 ret = bdrv_truncate(bs, data_file_offset + image_size); 1559 if (ret < 0) { 1560 goto exit; 1561 } 1562 } else { 1563 ret = -ENOTSUP; 1564 goto exit; 1565 } 1566 1567 if (type == VHDX_TYPE_FIXED || 1568 use_zero_blocks || 1569 bdrv_has_zero_init(bs) == 0) { 1570 /* for a fixed file, the default BAT entry is not zero */ 1571 s->bat = g_malloc0(rt_bat->length); 1572 block_state = type == VHDX_TYPE_FIXED ? PAYLOAD_BLOCK_FULLY_PRESENT : 1573 PAYLOAD_BLOCK_NOT_PRESENT; 1574 block_state = use_zero_blocks ? PAYLOAD_BLOCK_ZERO : block_state; 1575 /* fill the BAT by emulating sector writes of sectors_per_block size */ 1576 while (sector_num < total_sectors) { 1577 vhdx_block_translate(s, sector_num, s->sectors_per_block, &sinfo); 1578 sinfo.file_offset = data_file_offset + 1579 (sector_num << s->logical_sector_size_bits); 1580 sinfo.file_offset = ROUND_UP(sinfo.file_offset, MiB); 1581 vhdx_update_bat_table_entry(bs, s, &sinfo, &unused, &unused, 1582 block_state); 1583 cpu_to_le64s(&s->bat[sinfo.bat_idx]); 1584 sector_num += s->sectors_per_block; 1585 } 1586 ret = bdrv_pwrite(bs, rt_bat->file_offset, s->bat, rt_bat->length); 1587 if (ret < 0) { 1588 goto exit; 1589 } 1590 } 1591 1592 1593 1594 exit: 1595 g_free(s->bat); 1596 return ret; 1597 } 1598 1599 /* Creates the region table header, and region table entries. 1600 * There are 2 supported region table entries: BAT, and Metadata/ 1601 * 1602 * As the calculations for the BAT region table are also needed 1603 * to create the BAT itself, we will also cause the BAT to be 1604 * created. 1605 */ 1606 static int vhdx_create_new_region_table(BlockDriverState *bs, 1607 uint64_t image_size, 1608 uint32_t block_size, 1609 uint32_t sector_size, 1610 uint32_t log_size, 1611 bool use_zero_blocks, 1612 VHDXImageType type, 1613 uint64_t *metadata_offset) 1614 { 1615 int ret = 0; 1616 uint32_t offset = 0; 1617 void *buffer = NULL; 1618 BDRVVHDXState *s = NULL; 1619 VHDXRegionTableHeader *region_table; 1620 VHDXRegionTableEntry *rt_bat; 1621 VHDXRegionTableEntry *rt_metadata; 1622 1623 assert(metadata_offset != NULL); 1624 1625 /* Populate enough of the BDRVVHDXState to be able to use the 1626 * pre-existing BAT calculation, translation, and update functions */ 1627 s = g_malloc0(sizeof(BDRVVHDXState)); 1628 1629 s->chunk_ratio = (VHDX_MAX_SECTORS_PER_BLOCK) * 1630 (uint64_t) sector_size / (uint64_t) block_size; 1631 1632 s->sectors_per_block = block_size / sector_size; 1633 s->virtual_disk_size = image_size; 1634 s->block_size = block_size; 1635 s->logical_sector_size = sector_size; 1636 1637 vhdx_set_shift_bits(s); 1638 1639 vhdx_calc_bat_entries(s); 1640 1641 /* At this point the VHDX state is populated enough for creation */ 1642 1643 /* a single buffer is used so we can calculate the checksum over the 1644 * entire 64KB block */ 1645 buffer = g_malloc0(VHDX_HEADER_BLOCK_SIZE); 1646 region_table = buffer; 1647 offset += sizeof(VHDXRegionTableHeader); 1648 rt_bat = buffer + offset; 1649 offset += sizeof(VHDXRegionTableEntry); 1650 rt_metadata = buffer + offset; 1651 1652 region_table->signature = VHDX_REGION_SIGNATURE; 1653 region_table->entry_count = 2; /* BAT and Metadata */ 1654 1655 rt_bat->guid = bat_guid; 1656 rt_bat->length = ROUND_UP(s->bat_entries * sizeof(VHDXBatEntry), MiB); 1657 rt_bat->file_offset = ROUND_UP(VHDX_HEADER_SECTION_END + log_size, MiB); 1658 s->bat_offset = rt_bat->file_offset; 1659 1660 rt_metadata->guid = metadata_guid; 1661 rt_metadata->file_offset = ROUND_UP(rt_bat->file_offset + rt_bat->length, 1662 MiB); 1663 rt_metadata->length = 1 * MiB; /* min size, and more than enough */ 1664 *metadata_offset = rt_metadata->file_offset; 1665 1666 vhdx_update_checksum(buffer, VHDX_HEADER_BLOCK_SIZE, 1667 offsetof(VHDXRegionTableHeader, checksum)); 1668 1669 1670 /* The region table gives us the data we need to create the BAT, 1671 * so do that now */ 1672 ret = vhdx_create_bat(bs, s, image_size, type, use_zero_blocks, rt_bat); 1673 1674 /* Now write out the region headers to disk */ 1675 vhdx_region_header_le_export(region_table); 1676 vhdx_region_entry_le_export(rt_bat); 1677 vhdx_region_entry_le_export(rt_metadata); 1678 1679 ret = bdrv_pwrite(bs, VHDX_REGION_TABLE_OFFSET, buffer, 1680 VHDX_HEADER_BLOCK_SIZE); 1681 if (ret < 0) { 1682 goto exit; 1683 } 1684 1685 ret = bdrv_pwrite(bs, VHDX_REGION_TABLE2_OFFSET, buffer, 1686 VHDX_HEADER_BLOCK_SIZE); 1687 if (ret < 0) { 1688 goto exit; 1689 } 1690 1691 1692 exit: 1693 g_free(s); 1694 g_free(buffer); 1695 return ret; 1696 } 1697 1698 /* We need to create the following elements: 1699 * 1700 * .-----------------------------------------------------------------. 1701 * | (A) | (B) | (C) | (D) | (E) | 1702 * | File ID | Header1 | Header 2 | Region Tbl 1 | Region Tbl 2 | 1703 * | | | | | | 1704 * .-----------------------------------------------------------------. 1705 * 0 64KB 128KB 192KB 256KB 320KB 1706 * 1707 * 1708 * .---- ~ ----------- ~ ------------ ~ ---------------- ~ -----------. 1709 * | (F) | (G) | (H) | | 1710 * | Journal Log | BAT / Bitmap | Metadata | .... data ...... | 1711 * | | | | | 1712 * .---- ~ ----------- ~ ------------ ~ ---------------- ~ -----------. 1713 * 1MB 1714 */ 1715 static int vhdx_create(const char *filename, QEMUOptionParameter *options, 1716 Error **errp) 1717 { 1718 int ret = 0; 1719 uint64_t image_size = (uint64_t) 2 * GiB; 1720 uint32_t log_size = 1 * MiB; 1721 uint32_t block_size = 0; 1722 uint64_t signature; 1723 uint64_t metadata_offset; 1724 bool use_zero_blocks = false; 1725 1726 gunichar2 *creator = NULL; 1727 glong creator_items; 1728 BlockDriverState *bs; 1729 const char *type = NULL; 1730 VHDXImageType image_type; 1731 Error *local_err = NULL; 1732 1733 while (options && options->name) { 1734 if (!strcmp(options->name, BLOCK_OPT_SIZE)) { 1735 image_size = options->value.n; 1736 } else if (!strcmp(options->name, VHDX_BLOCK_OPT_LOG_SIZE)) { 1737 log_size = options->value.n; 1738 } else if (!strcmp(options->name, VHDX_BLOCK_OPT_BLOCK_SIZE)) { 1739 block_size = options->value.n; 1740 } else if (!strcmp(options->name, BLOCK_OPT_SUBFMT)) { 1741 type = options->value.s; 1742 } else if (!strcmp(options->name, VHDX_BLOCK_OPT_ZERO)) { 1743 use_zero_blocks = options->value.n != 0; 1744 } 1745 options++; 1746 } 1747 1748 if (image_size > VHDX_MAX_IMAGE_SIZE) { 1749 error_setg_errno(errp, EINVAL, "Image size too large; max of 64TB"); 1750 ret = -EINVAL; 1751 goto exit; 1752 } 1753 1754 if (type == NULL) { 1755 type = "dynamic"; 1756 } 1757 1758 if (!strcmp(type, "dynamic")) { 1759 image_type = VHDX_TYPE_DYNAMIC; 1760 } else if (!strcmp(type, "fixed")) { 1761 image_type = VHDX_TYPE_FIXED; 1762 } else if (!strcmp(type, "differencing")) { 1763 error_setg_errno(errp, ENOTSUP, 1764 "Differencing files not yet supported"); 1765 ret = -ENOTSUP; 1766 goto exit; 1767 } else { 1768 ret = -EINVAL; 1769 goto exit; 1770 } 1771 1772 /* These are pretty arbitrary, and mainly designed to keep the BAT 1773 * size reasonable to load into RAM */ 1774 if (block_size == 0) { 1775 if (image_size > 32 * TiB) { 1776 block_size = 64 * MiB; 1777 } else if (image_size > (uint64_t) 100 * GiB) { 1778 block_size = 32 * MiB; 1779 } else if (image_size > 1 * GiB) { 1780 block_size = 16 * MiB; 1781 } else { 1782 block_size = 8 * MiB; 1783 } 1784 } 1785 1786 1787 /* make the log size close to what was specified, but must be 1788 * min 1MB, and multiple of 1MB */ 1789 log_size = ROUND_UP(log_size, MiB); 1790 1791 block_size = ROUND_UP(block_size, MiB); 1792 block_size = block_size > VHDX_BLOCK_SIZE_MAX ? VHDX_BLOCK_SIZE_MAX : 1793 block_size; 1794 1795 ret = bdrv_create_file(filename, options, &local_err); 1796 if (ret < 0) { 1797 error_propagate(errp, local_err); 1798 goto exit; 1799 } 1800 1801 ret = bdrv_file_open(&bs, filename, NULL, BDRV_O_RDWR, &local_err); 1802 if (ret < 0) { 1803 error_propagate(errp, local_err); 1804 goto exit; 1805 } 1806 1807 /* Create (A) */ 1808 1809 /* The creator field is optional, but may be useful for 1810 * debugging / diagnostics */ 1811 creator = g_utf8_to_utf16("QEMU v" QEMU_VERSION, -1, NULL, 1812 &creator_items, NULL); 1813 signature = cpu_to_le64(VHDX_FILE_SIGNATURE); 1814 bdrv_pwrite(bs, VHDX_FILE_ID_OFFSET, &signature, sizeof(signature)); 1815 if (ret < 0) { 1816 goto delete_and_exit; 1817 } 1818 if (creator) { 1819 bdrv_pwrite(bs, VHDX_FILE_ID_OFFSET + sizeof(signature), creator, 1820 creator_items * sizeof(gunichar2)); 1821 if (ret < 0) { 1822 goto delete_and_exit; 1823 } 1824 } 1825 1826 1827 /* Creates (B),(C) */ 1828 ret = vhdx_create_new_headers(bs, image_size, log_size); 1829 if (ret < 0) { 1830 goto delete_and_exit; 1831 } 1832 1833 /* Creates (D),(E),(G) explicitly. (F) created as by-product */ 1834 ret = vhdx_create_new_region_table(bs, image_size, block_size, 512, 1835 log_size, use_zero_blocks, image_type, 1836 &metadata_offset); 1837 if (ret < 0) { 1838 goto delete_and_exit; 1839 } 1840 1841 /* Creates (H) */ 1842 ret = vhdx_create_new_metadata(bs, image_size, block_size, 512, 1843 metadata_offset, image_type); 1844 if (ret < 0) { 1845 goto delete_and_exit; 1846 } 1847 1848 1849 1850 delete_and_exit: 1851 bdrv_unref(bs); 1852 exit: 1853 g_free(creator); 1854 return ret; 1855 } 1856 1857 static QEMUOptionParameter vhdx_create_options[] = { 1858 { 1859 .name = BLOCK_OPT_SIZE, 1860 .type = OPT_SIZE, 1861 .help = "Virtual disk size; max of 64TB." 1862 }, 1863 { 1864 .name = VHDX_BLOCK_OPT_LOG_SIZE, 1865 .type = OPT_SIZE, 1866 .value.n = 1 * MiB, 1867 .help = "Log size; min 1MB." 1868 }, 1869 { 1870 .name = VHDX_BLOCK_OPT_BLOCK_SIZE, 1871 .type = OPT_SIZE, 1872 .value.n = 0, 1873 .help = "Block Size; min 1MB, max 256MB. " \ 1874 "0 means auto-calculate based on image size." 1875 }, 1876 { 1877 .name = BLOCK_OPT_SUBFMT, 1878 .type = OPT_STRING, 1879 .help = "VHDX format type, can be either 'dynamic' or 'fixed'. "\ 1880 "Default is 'dynamic'." 1881 }, 1882 { 1883 .name = VHDX_BLOCK_OPT_ZERO, 1884 .type = OPT_FLAG, 1885 .help = "Force use of payload blocks of type 'ZERO'. Non-standard." 1886 }, 1887 { NULL } 1888 }; 1889 1890 static BlockDriver bdrv_vhdx = { 1891 .format_name = "vhdx", 1892 .instance_size = sizeof(BDRVVHDXState), 1893 .bdrv_probe = vhdx_probe, 1894 .bdrv_open = vhdx_open, 1895 .bdrv_close = vhdx_close, 1896 .bdrv_reopen_prepare = vhdx_reopen_prepare, 1897 .bdrv_co_readv = vhdx_co_readv, 1898 .bdrv_co_writev = vhdx_co_writev, 1899 .bdrv_create = vhdx_create, 1900 .bdrv_get_info = vhdx_get_info, 1901 1902 .create_options = vhdx_create_options, 1903 }; 1904 1905 static void bdrv_vhdx_init(void) 1906 { 1907 bdrv_register(&bdrv_vhdx); 1908 } 1909 1910 block_init(bdrv_vhdx_init); 1911