1/* 2 * Block driver for Connectix / Microsoft Virtual PC images 3 * 4 * Copyright (c) 2005 Alex Beregszaszi 5 * Copyright (c) 2009 Kevin Wolf <kwolf@suse.de> 6 * 7 * Permission is hereby granted, free of charge, to any person obtaining a copy 8 * of this software and associated documentation files (the "Software"), to deal 9 * in the Software without restriction, including without limitation the rights 10 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 11 * copies of the Software, and to permit persons to whom the Software is 12 * furnished to do so, subject to the following conditions: 13 * 14 * The above copyright notice and this permission notice shall be included in 15 * all copies or substantial portions of the Software. 16 * 17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 18 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 20 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 21 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 22 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN 23 * THE SOFTWARE. 24 */ 25#include "qemu-common.h" 26#include "block_int.h" 27#include "module.h" 28 29/**************************************************************/ 30 31#define HEADER_SIZE 512 32 33//#define CACHE 34 35enum vhd_type { 36 VHD_FIXED = 2, 37 VHD_DYNAMIC = 3, 38 VHD_DIFFERENCING = 4, 39}; 40 41// Seconds since Jan 1, 2000 0:00:00 (UTC) 42#define VHD_TIMESTAMP_BASE 946684800 43 44// always big-endian 45struct vhd_footer { 46 char creator[8]; // "conectix" 47 uint32_t features; 48 uint32_t version; 49 50 // Offset of next header structure, 0xFFFFFFFF if none 51 uint64_t data_offset; 52 53 // Seconds since Jan 1, 2000 0:00:00 (UTC) 54 uint32_t timestamp; 55 56 char creator_app[4]; // "vpc " 57 uint16_t major; 58 uint16_t minor; 59 char creator_os[4]; // "Wi2k" 60 61 uint64_t orig_size; 62 uint64_t size; 63 64 uint16_t cyls; 65 uint8_t heads; 66 uint8_t secs_per_cyl; 67 68 uint32_t type; 69 70 // Checksum of the Hard Disk Footer ("one's complement of the sum of all 71 // the bytes in the footer without the checksum field") 72 uint32_t checksum; 73 74 // UUID used to identify a parent hard disk (backing file) 75 uint8_t uuid[16]; 76 77 uint8_t in_saved_state; 78}; 79 80struct vhd_dyndisk_header { 81 char magic[8]; // "cxsparse" 82 83 // Offset of next header structure, 0xFFFFFFFF if none 84 uint64_t data_offset; 85 86 // Offset of the Block Allocation Table (BAT) 87 uint64_t table_offset; 88 89 uint32_t version; 90 uint32_t max_table_entries; // 32bit/entry 91 92 // 2 MB by default, must be a power of two 93 uint32_t block_size; 94 95 uint32_t checksum; 96 uint8_t parent_uuid[16]; 97 uint32_t parent_timestamp; 98 uint32_t reserved; 99 100 // Backing file name (in UTF-16) 101 uint8_t parent_name[512]; 102 103 struct { 104 uint32_t platform; 105 uint32_t data_space; 106 uint32_t data_length; 107 uint32_t reserved; 108 uint64_t data_offset; 109 } parent_locator[8]; 110}; 111 112typedef struct BDRVVPCState { 113 BlockDriverState *hd; 114 115 uint8_t footer_buf[HEADER_SIZE]; 116 uint64_t free_data_block_offset; 117 int max_table_entries; 118 uint32_t *pagetable; 119 uint64_t bat_offset; 120 uint64_t last_bitmap_offset; 121 122 uint32_t block_size; 123 uint32_t bitmap_size; 124 125#ifdef CACHE 126 uint8_t *pageentry_u8; 127 uint32_t *pageentry_u32; 128 uint16_t *pageentry_u16; 129 130 uint64_t last_bitmap; 131#endif 132} BDRVVPCState; 133 134static uint32_t vpc_checksum(uint8_t* buf, size_t size) 135{ 136 uint32_t res = 0; 137 int i; 138 139 for (i = 0; i < size; i++) 140 res += buf[i]; 141 142 return ~res; 143} 144 145 146static int vpc_probe(const uint8_t *buf, int buf_size, const char *filename) 147{ 148 if (buf_size >= 8 && !strncmp((char *)buf, "conectix", 8)) 149 return 100; 150 return 0; 151} 152 153static int vpc_open(BlockDriverState *bs, int flags) 154{ 155 BDRVVPCState *s = bs->opaque; 156 int i; 157 struct vhd_footer* footer; 158 struct vhd_dyndisk_header* dyndisk_header; 159 uint8_t buf[HEADER_SIZE]; 160 uint32_t checksum; 161 162 if (bdrv_pread(bs->file, 0, s->footer_buf, HEADER_SIZE) != HEADER_SIZE) 163 goto fail; 164 165 footer = (struct vhd_footer*) s->footer_buf; 166 if (strncmp(footer->creator, "conectix", 8)) 167 goto fail; 168 169 checksum = be32_to_cpu(footer->checksum); 170 footer->checksum = 0; 171 if (vpc_checksum(s->footer_buf, HEADER_SIZE) != checksum) 172 fprintf(stderr, "block-vpc: The header checksum of '%s' is " 173 "incorrect.\n", bs->filename); 174 175 // The visible size of a image in Virtual PC depends on the geometry 176 // rather than on the size stored in the footer (the size in the footer 177 // is too large usually) 178 bs->total_sectors = (int64_t) 179 be16_to_cpu(footer->cyls) * footer->heads * footer->secs_per_cyl; 180 181 if (bdrv_pread(bs->file, be64_to_cpu(footer->data_offset), buf, HEADER_SIZE) 182 != HEADER_SIZE) 183 goto fail; 184 185 dyndisk_header = (struct vhd_dyndisk_header*) buf; 186 187 if (strncmp(dyndisk_header->magic, "cxsparse", 8)) 188 goto fail; 189 190 191 s->block_size = be32_to_cpu(dyndisk_header->block_size); 192 s->bitmap_size = ((s->block_size / (8 * 512)) + 511) & ~511; 193 194 s->max_table_entries = be32_to_cpu(dyndisk_header->max_table_entries); 195 s->pagetable = qemu_malloc(s->max_table_entries * 4); 196 197 s->bat_offset = be64_to_cpu(dyndisk_header->table_offset); 198 if (bdrv_pread(bs->file, s->bat_offset, s->pagetable, 199 s->max_table_entries * 4) != s->max_table_entries * 4) 200 goto fail; 201 202 s->free_data_block_offset = 203 (s->bat_offset + (s->max_table_entries * 4) + 511) & ~511; 204 205 for (i = 0; i < s->max_table_entries; i++) { 206 be32_to_cpus(&s->pagetable[i]); 207 if (s->pagetable[i] != 0xFFFFFFFF) { 208 int64_t next = (512 * (int64_t) s->pagetable[i]) + 209 s->bitmap_size + s->block_size; 210 211 if (next> s->free_data_block_offset) 212 s->free_data_block_offset = next; 213 } 214 } 215 216 s->last_bitmap_offset = (int64_t) -1; 217 218#ifdef CACHE 219 s->pageentry_u8 = qemu_malloc(512); 220 s->pageentry_u32 = s->pageentry_u8; 221 s->pageentry_u16 = s->pageentry_u8; 222 s->last_pagetable = -1; 223#endif 224 225 return 0; 226 fail: 227 return -1; 228} 229 230/* 231 * Returns the absolute byte offset of the given sector in the image file. 232 * If the sector is not allocated, -1 is returned instead. 233 * 234 * The parameter write must be 1 if the offset will be used for a write 235 * operation (the block bitmaps is updated then), 0 otherwise. 236 */ 237static inline int64_t get_sector_offset(BlockDriverState *bs, 238 int64_t sector_num, int write) 239{ 240 BDRVVPCState *s = bs->opaque; 241 uint64_t offset = sector_num * 512; 242 uint64_t bitmap_offset, block_offset; 243 uint32_t pagetable_index, pageentry_index; 244 245 pagetable_index = offset / s->block_size; 246 pageentry_index = (offset % s->block_size) / 512; 247 248 if (pagetable_index >= s->max_table_entries || s->pagetable[pagetable_index] == 0xffffffff) 249 return -1; // not allocated 250 251 bitmap_offset = 512 * (uint64_t) s->pagetable[pagetable_index]; 252 block_offset = bitmap_offset + s->bitmap_size + (512 * pageentry_index); 253 254 // We must ensure that we don't write to any sectors which are marked as 255 // unused in the bitmap. We get away with setting all bits in the block 256 // bitmap each time we write to a new block. This might cause Virtual PC to 257 // miss sparse read optimization, but it's not a problem in terms of 258 // correctness. 259 if (write && (s->last_bitmap_offset != bitmap_offset)) { 260 uint8_t bitmap[s->bitmap_size]; 261 262 s->last_bitmap_offset = bitmap_offset; 263 memset(bitmap, 0xff, s->bitmap_size); 264 bdrv_pwrite_sync(bs->file, bitmap_offset, bitmap, s->bitmap_size); 265 } 266 267// printf("sector: %" PRIx64 ", index: %x, offset: %x, bioff: %" PRIx64 ", bloff: %" PRIx64 "\n", 268// sector_num, pagetable_index, pageentry_index, 269// bitmap_offset, block_offset); 270 271// disabled by reason 272#if 0 273#ifdef CACHE 274 if (bitmap_offset != s->last_bitmap) 275 { 276 lseek(s->fd, bitmap_offset, SEEK_SET); 277 278 s->last_bitmap = bitmap_offset; 279 280 // Scary! Bitmap is stored as big endian 32bit entries, 281 // while we used to look it up byte by byte 282 read(s->fd, s->pageentry_u8, 512); 283 for (i = 0; i < 128; i++) 284 be32_to_cpus(&s->pageentry_u32[i]); 285 } 286 287 if ((s->pageentry_u8[pageentry_index / 8] >> (pageentry_index % 8)) & 1) 288 return -1; 289#else 290 lseek(s->fd, bitmap_offset + (pageentry_index / 8), SEEK_SET); 291 292 read(s->fd, &bitmap_entry, 1); 293 294 if ((bitmap_entry >> (pageentry_index % 8)) & 1) 295 return -1; // not allocated 296#endif 297#endif 298 299 return block_offset; 300} 301 302/* 303 * Writes the footer to the end of the image file. This is needed when the 304 * file grows as it overwrites the old footer 305 * 306 * Returns 0 on success and < 0 on error 307 */ 308static int rewrite_footer(BlockDriverState* bs) 309{ 310 int ret; 311 BDRVVPCState *s = bs->opaque; 312 int64_t offset = s->free_data_block_offset; 313 314 ret = bdrv_pwrite_sync(bs->file, offset, s->footer_buf, HEADER_SIZE); 315 if (ret < 0) 316 return ret; 317 318 return 0; 319} 320 321/* 322 * Allocates a new block. This involves writing a new footer and updating 323 * the Block Allocation Table to use the space at the old end of the image 324 * file (overwriting the old footer) 325 * 326 * Returns the sectors' offset in the image file on success and < 0 on error 327 */ 328static int64_t alloc_block(BlockDriverState* bs, int64_t sector_num) 329{ 330 BDRVVPCState *s = bs->opaque; 331 int64_t bat_offset; 332 uint32_t index, bat_value; 333 int ret; 334 uint8_t bitmap[s->bitmap_size]; 335 336 // Check if sector_num is valid 337 if ((sector_num < 0) || (sector_num > bs->total_sectors)) 338 return -1; 339 340 // Write entry into in-memory BAT 341 index = (sector_num * 512) / s->block_size; 342 if (s->pagetable[index] != 0xFFFFFFFF) 343 return -1; 344 345 s->pagetable[index] = s->free_data_block_offset / 512; 346 347 // Initialize the block's bitmap 348 memset(bitmap, 0xff, s->bitmap_size); 349 bdrv_pwrite_sync(bs->file, s->free_data_block_offset, bitmap, 350 s->bitmap_size); 351 352 // Write new footer (the old one will be overwritten) 353 s->free_data_block_offset += s->block_size + s->bitmap_size; 354 ret = rewrite_footer(bs); 355 if (ret < 0) 356 goto fail; 357 358 // Write BAT entry to disk 359 bat_offset = s->bat_offset + (4 * index); 360 bat_value = be32_to_cpu(s->pagetable[index]); 361 ret = bdrv_pwrite_sync(bs->file, bat_offset, &bat_value, 4); 362 if (ret < 0) 363 goto fail; 364 365 return get_sector_offset(bs, sector_num, 0); 366 367fail: 368 s->free_data_block_offset -= (s->block_size + s->bitmap_size); 369 return -1; 370} 371 372static int vpc_read(BlockDriverState *bs, int64_t sector_num, 373 uint8_t *buf, int nb_sectors) 374{ 375 BDRVVPCState *s = bs->opaque; 376 int ret; 377 int64_t offset; 378 int64_t sectors, sectors_per_block; 379 380 while (nb_sectors > 0) { 381 offset = get_sector_offset(bs, sector_num, 0); 382 383 sectors_per_block = s->block_size >> BDRV_SECTOR_BITS; 384 sectors = sectors_per_block - (sector_num % sectors_per_block); 385 if (sectors > nb_sectors) { 386 sectors = nb_sectors; 387 } 388 389 if (offset == -1) { 390 memset(buf, 0, sectors * BDRV_SECTOR_SIZE); 391 } else { 392 ret = bdrv_pread(bs->file, offset, buf, 393 sectors * BDRV_SECTOR_SIZE); 394 if (ret != sectors * BDRV_SECTOR_SIZE) { 395 return -1; 396 } 397 } 398 399 nb_sectors -= sectors; 400 sector_num += sectors; 401 buf += sectors * BDRV_SECTOR_SIZE; 402 } 403 return 0; 404} 405 406static int vpc_write(BlockDriverState *bs, int64_t sector_num, 407 const uint8_t *buf, int nb_sectors) 408{ 409 BDRVVPCState *s = bs->opaque; 410 int64_t offset; 411 int64_t sectors, sectors_per_block; 412 int ret; 413 414 while (nb_sectors > 0) { 415 offset = get_sector_offset(bs, sector_num, 1); 416 417 sectors_per_block = s->block_size >> BDRV_SECTOR_BITS; 418 sectors = sectors_per_block - (sector_num % sectors_per_block); 419 if (sectors > nb_sectors) { 420 sectors = nb_sectors; 421 } 422 423 if (offset == -1) { 424 offset = alloc_block(bs, sector_num); 425 if (offset < 0) 426 return -1; 427 } 428 429 ret = bdrv_pwrite(bs->file, offset, buf, sectors * BDRV_SECTOR_SIZE); 430 if (ret != sectors * BDRV_SECTOR_SIZE) { 431 return -1; 432 } 433 434 nb_sectors -= sectors; 435 sector_num += sectors; 436 buf += sectors * BDRV_SECTOR_SIZE; 437 } 438 439 return 0; 440} 441 442 443/* 444 * Calculates the number of cylinders, heads and sectors per cylinder 445 * based on a given number of sectors. This is the algorithm described 446 * in the VHD specification. 447 * 448 * Note that the geometry doesn't always exactly match total_sectors but 449 * may round it down. 450 * 451 * Returns 0 on success, -EFBIG if the size is larger than 127 GB 452 */ 453static int calculate_geometry(int64_t total_sectors, uint16_t* cyls, 454 uint8_t* heads, uint8_t* secs_per_cyl) 455{ 456 uint32_t cyls_times_heads; 457 458 if (total_sectors > 65535 * 16 * 255) 459 return -EFBIG; 460 461 if (total_sectors > 65535 * 16 * 63) { 462 *secs_per_cyl = 255; 463 *heads = 16; 464 cyls_times_heads = total_sectors / *secs_per_cyl; 465 } else { 466 *secs_per_cyl = 17; 467 cyls_times_heads = total_sectors / *secs_per_cyl; 468 *heads = (cyls_times_heads + 1023) / 1024; 469 470 if (*heads < 4) 471 *heads = 4; 472 473 if (cyls_times_heads >= (*heads * 1024) || *heads > 16) { 474 *secs_per_cyl = 31; 475 *heads = 16; 476 cyls_times_heads = total_sectors / *secs_per_cyl; 477 } 478 479 if (cyls_times_heads >= (*heads * 1024)) { 480 *secs_per_cyl = 63; 481 *heads = 16; 482 cyls_times_heads = total_sectors / *secs_per_cyl; 483 } 484 } 485 486 *cyls = cyls_times_heads / *heads; 487 488 return 0; 489} 490 491static int vpc_create(const char *filename, QEMUOptionParameter *options) 492{ 493 uint8_t buf[1024]; 494 struct vhd_footer* footer = (struct vhd_footer*) buf; 495 struct vhd_dyndisk_header* dyndisk_header = 496 (struct vhd_dyndisk_header*) buf; 497 int fd, i; 498 uint16_t cyls = 0; 499 uint8_t heads = 0; 500 uint8_t secs_per_cyl = 0; 501 size_t block_size, num_bat_entries; 502 int64_t total_sectors = 0; 503 504 // Read out options 505 while (options && options->name) { 506 if (!strcmp(options->name, "size")) { 507 total_sectors = options->value.n / 512; 508 } 509 options++; 510 } 511 512 // Create the file 513 fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, 0644); 514 if (fd < 0) 515 return -EIO; 516 517 /* Calculate matching total_size and geometry. Increase the number of 518 sectors requested until we get enough (or fail). */ 519 for (i = 0; total_sectors > (int64_t)cyls * heads * secs_per_cyl; i++) { 520 if (calculate_geometry(total_sectors + i, 521 &cyls, &heads, &secs_per_cyl)) { 522 return -EFBIG; 523 } 524 } 525 total_sectors = (int64_t) cyls * heads * secs_per_cyl; 526 527 // Prepare the Hard Disk Footer 528 memset(buf, 0, 1024); 529 530 memcpy(footer->creator, "conectix", 8); 531 // TODO Check if "qemu" creator_app is ok for VPC 532 memcpy(footer->creator_app, "qemu", 4); 533 memcpy(footer->creator_os, "Wi2k", 4); 534 535 footer->features = be32_to_cpu(0x02); 536 footer->version = be32_to_cpu(0x00010000); 537 footer->data_offset = be64_to_cpu(HEADER_SIZE); 538 footer->timestamp = be32_to_cpu(time(NULL) - VHD_TIMESTAMP_BASE); 539 540 // Version of Virtual PC 2007 541 footer->major = be16_to_cpu(0x0005); 542 footer->minor =be16_to_cpu(0x0003); 543 544 footer->orig_size = be64_to_cpu(total_sectors * 512); 545 footer->size = be64_to_cpu(total_sectors * 512); 546 547 footer->cyls = be16_to_cpu(cyls); 548 footer->heads = heads; 549 footer->secs_per_cyl = secs_per_cyl; 550 551 footer->type = be32_to_cpu(VHD_DYNAMIC); 552 553 // TODO uuid is missing 554 555 footer->checksum = be32_to_cpu(vpc_checksum(buf, HEADER_SIZE)); 556 557 // Write the footer (twice: at the beginning and at the end) 558 block_size = 0x200000; 559 num_bat_entries = (total_sectors + block_size / 512) / (block_size / 512); 560 561 if (write(fd, buf, HEADER_SIZE) != HEADER_SIZE) 562 return -EIO; 563 564 if (lseek(fd, 1536 + ((num_bat_entries * 4 + 511) & ~511), SEEK_SET) < 0) 565 return -EIO; 566 if (write(fd, buf, HEADER_SIZE) != HEADER_SIZE) 567 return -EIO; 568 569 // Write the initial BAT 570 if (lseek(fd, 3 * 512, SEEK_SET) < 0) 571 return -EIO; 572 573 memset(buf, 0xFF, 512); 574 for (i = 0; i < (num_bat_entries * 4 + 511) / 512; i++) 575 if (write(fd, buf, 512) != 512) 576 return -EIO; 577 578 579 // Prepare the Dynamic Disk Header 580 memset(buf, 0, 1024); 581 582 memcpy(dyndisk_header->magic, "cxsparse", 8); 583 584 dyndisk_header->data_offset = be64_to_cpu(0xFFFFFFFF); 585 dyndisk_header->table_offset = be64_to_cpu(3 * 512); 586 dyndisk_header->version = be32_to_cpu(0x00010000); 587 dyndisk_header->block_size = be32_to_cpu(block_size); 588 dyndisk_header->max_table_entries = be32_to_cpu(num_bat_entries); 589 590 dyndisk_header->checksum = be32_to_cpu(vpc_checksum(buf, 1024)); 591 592 // Write the header 593 if (lseek(fd, 512, SEEK_SET) < 0) 594 return -EIO; 595 if (write(fd, buf, 1024) != 1024) 596 return -EIO; 597 598 close(fd); 599 return 0; 600} 601 602static void vpc_close(BlockDriverState *bs) 603{ 604 BDRVVPCState *s = bs->opaque; 605 qemu_free(s->pagetable); 606#ifdef CACHE 607 qemu_free(s->pageentry_u8); 608#endif 609} 610 611static QEMUOptionParameter vpc_create_options[] = { 612 { 613 .name = BLOCK_OPT_SIZE, 614 .type = OPT_SIZE, 615 .help = "Virtual disk size" 616 }, 617 { NULL } 618}; 619 620static BlockDriver bdrv_vpc = { 621 .format_name = "vpc", 622 .instance_size = sizeof(BDRVVPCState), 623 .bdrv_probe = vpc_probe, 624 .bdrv_open = vpc_open, 625 .bdrv_read = vpc_read, 626 .bdrv_write = vpc_write, 627 .bdrv_close = vpc_close, 628 .bdrv_create = vpc_create, 629 630 .create_options = vpc_create_options, 631}; 632 633static void bdrv_vpc_init(void) 634{ 635 bdrv_register(&bdrv_vpc); 636} 637 638block_init(bdrv_vpc_init); 639