block/vdi.c

   1 /*
   2  * Block driver for the Virtual Disk Image (VDI) format
   3  *
   4  * Copyright (c) 2009, 2012 Stefan Weil
   5  *
   6  * This program is free software: you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License as published by
   8  * the Free Software Foundation, either version 2 of the License, or
   9  * (at your option) version 3 or any later version.
  10  *
  11  * This program is distributed in the hope that it will be useful,
  12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  14  * GNU General Public License for more details.
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * along with this program.  If not, see <http://www.gnu.org/licenses/>.
  18  *
  19  * Reference:
  20  * http://forums.virtualbox.org/viewtopic.php?t=8046
  21  *
  22  * This driver supports create / read / write operations on VDI images.
  23  *
  24  * Todo (see also TODO in code):
  25  *
  26  * Some features like snapshots are still missing.
  27  *
  28  * Deallocation of zero-filled blocks and shrinking images are missing, too
  29  * (might be added to common block layer).
  30  *
  31  * Allocation of blocks could be optimized (less writes to block map and
  32  * header).
  33  *
  34  * Read and write of adjacents blocks could be done in one operation
  35  * (current code uses one operation per block (1 MiB).
  36  *
  37  * The code is not thread safe (missing locks for changes in header and
  38  * block table, no problem with current QEMU).
  39  *
  40  * Hints:
  41  *
  42  * Blocks (VDI documentation) correspond to clusters (QEMU).
  43  * QEMU's backing files could be implemented using VDI snapshot files (TODO).
  44  * VDI snapshot files may also contain the complete machine state.
  45  * Maybe this machine state can be converted to QEMU PC machine snapshot data.
  46  *
  47  * The driver keeps a block cache (little endian entries) in memory.
  48  * For the standard block size (1 MiB), a 1 TiB disk will use 4 MiB RAM,
  49  * so this seems to be reasonable.
  50  */
  51
  52 #include "qemu-common.h"
  53 #include "block/block_int.h"
  54 #include "qemu/module.h"
  55 #include "migration/migration.h"
  56
  57 #if defined(CONFIG_UUID)
  58 #include <uuid/uuid.h>
  59 #else
  60 /* TODO: move uuid emulation to some central place in QEMU. */
  61 #include "sysemu/sysemu.h"     /* UUID_FMT */
  62 typedef unsigned char uuid_t[16];
  63 #endif
  64
  65 /* Code configuration options. */
  66
  67 /* Enable debug messages. */
  68 //~ #define CONFIG_VDI_DEBUG
  69
  70 /* Support write operations on VDI images. */
  71 #define CONFIG_VDI_WRITE
  72
  73 /* Support non-standard block (cluster) size. This is untested.
  74  * Maybe it will be needed for very large images.
  75  */
  76 //~ #define CONFIG_VDI_BLOCK_SIZE
  77
  78 /* Support static (fixed, pre-allocated) images. */
  79 #define CONFIG_VDI_STATIC_IMAGE
  80
  81 /* Command line option for static images. */
  82 #define BLOCK_OPT_STATIC "static"
  83
  84 #define KiB     1024
  85 #define MiB     (KiB * KiB)
  86
  87 #define SECTOR_SIZE 512
  88 #define DEFAULT_CLUSTER_SIZE (1 * MiB)
  89
  90 #if defined(CONFIG_VDI_DEBUG)
  91 #define logout(fmt, ...) \
  92                 fprintf(stderr, "vdi\t%-24s" fmt, __func__, ##__VA_ARGS__)
  93 #else
  94 #define logout(fmt, ...) ((void)0)
  95 #endif
  96
  97 /* Image signature. */
  98 #define VDI_SIGNATURE 0xbeda107f
  99
 100 /* Image version. */
 101 #define VDI_VERSION_1_1 0x00010001
 102
 103 /* Image type. */
 104 #define VDI_TYPE_DYNAMIC 1
 105 #define VDI_TYPE_STATIC  2
 106
 107 /* Innotek / SUN images use these strings in header.text:
 108  * "<<< innotek VirtualBox Disk Image >>>\n"
 109  * "<<< Sun xVM VirtualBox Disk Image >>>\n"
 110  * "<<< Sun VirtualBox Disk Image >>>\n"
 111  * The value does not matter, so QEMU created images use a different text.
 112  */
 113 #define VDI_TEXT "<<< QEMU VM Virtual Disk Image >>>\n"
 114
 115 /* A never-allocated block; semantically arbitrary content. */
 116 #define VDI_UNALLOCATED 0xffffffffU
 117
 118 /* A discarded (no longer allocated) block; semantically zero-filled. */
 119 #define VDI_DISCARDED   0xfffffffeU
 120
 121 #define VDI_IS_ALLOCATED(X) ((X) < VDI_DISCARDED)
 122
 123 #if !defined(CONFIG_UUID)
 124 static inline void uuid_generate(uuid_t out)
 125 {
 126     memset(out, 0, sizeof(uuid_t));
 127 }
 128
 129 static inline int uuid_is_null(const uuid_t uu)
 130 {
 131     uuid_t null_uuid = { 0 };
 132     return memcmp(uu, null_uuid, sizeof(uuid_t)) == 0;
 133 }
 134
 135 static inline void uuid_unparse(const uuid_t uu, char *out)
 136 {
 137     snprintf(out, 37, UUID_FMT,
 138             uu[0], uu[1], uu[2], uu[3], uu[4], uu[5], uu[6], uu[7],
 139             uu[8], uu[9], uu[10], uu[11], uu[12], uu[13], uu[14], uu[15]);
 140 }
 141 #endif
 142
 143 typedef struct {
 144     char text[0x40];
 145     uint32_t signature;
 146     uint32_t version;
 147     uint32_t header_size;
 148     uint32_t image_type;
 149     uint32_t image_flags;
 150     char description[256];
 151     uint32_t offset_bmap;
 152     uint32_t offset_data;
 153     uint32_t cylinders;         /* disk geometry, unused here */
 154     uint32_t heads;             /* disk geometry, unused here */
 155     uint32_t sectors;           /* disk geometry, unused here */
 156     uint32_t sector_size;
 157     uint32_t unused1;
 158     uint64_t disk_size;
 159     uint32_t block_size;
 160     uint32_t block_extra;       /* unused here */
 161     uint32_t blocks_in_image;
 162     uint32_t blocks_allocated;
 163     uuid_t uuid_image;
 164     uuid_t uuid_last_snap;
 165     uuid_t uuid_link;
 166     uuid_t uuid_parent;
 167     uint64_t unused2[7];
 168 } VdiHeader;
 169
 170 typedef struct {
 171     /* The block map entries are little endian (even in memory). */
 172     uint32_t *bmap;
 173     /* Size of block (bytes). */
 174     uint32_t block_size;
 175     /* Size of block (sectors). */
 176     uint32_t block_sectors;
 177     /* First sector of block map. */
 178     uint32_t bmap_sector;
 179     /* VDI header (converted to host endianness). */
 180     VdiHeader header;
 181
 182     Error *migration_blocker;
 183 } BDRVVdiState;
 184
 185 /* Change UUID from little endian (IPRT = VirtualBox format) to big endian
 186  * format (network byte order, standard, see RFC 4122) and vice versa.
 187  */
 188 static void uuid_convert(uuid_t uuid)
 189 {
 190     bswap32s((uint32_t *)&uuid[0]);
 191     bswap16s((uint16_t *)&uuid[4]);
 192     bswap16s((uint16_t *)&uuid[6]);
 193 }
 194
 195 static void vdi_header_to_cpu(VdiHeader *header)
 196 {
 197     le32_to_cpus(&header->signature);
 198     le32_to_cpus(&header->version);
 199     le32_to_cpus(&header->header_size);
 200     le32_to_cpus(&header->image_type);
 201     le32_to_cpus(&header->image_flags);
 202     le32_to_cpus(&header->offset_bmap);
 203     le32_to_cpus(&header->offset_data);
 204     le32_to_cpus(&header->cylinders);
 205     le32_to_cpus(&header->heads);
 206     le32_to_cpus(&header->sectors);
 207     le32_to_cpus(&header->sector_size);
 208     le64_to_cpus(&header->disk_size);
 209     le32_to_cpus(&header->block_size);
 210     le32_to_cpus(&header->block_extra);
 211     le32_to_cpus(&header->blocks_in_image);
 212     le32_to_cpus(&header->blocks_allocated);
 213     uuid_convert(header->uuid_image);
 214     uuid_convert(header->uuid_last_snap);
 215     uuid_convert(header->uuid_link);
 216     uuid_convert(header->uuid_parent);
 217 }
 218
 219 static void vdi_header_to_le(VdiHeader *header)
 220 {
 221     cpu_to_le32s(&header->signature);
 222     cpu_to_le32s(&header->version);
 223     cpu_to_le32s(&header->header_size);
 224     cpu_to_le32s(&header->image_type);
 225     cpu_to_le32s(&header->image_flags);
 226     cpu_to_le32s(&header->offset_bmap);
 227     cpu_to_le32s(&header->offset_data);
 228     cpu_to_le32s(&header->cylinders);
 229     cpu_to_le32s(&header->heads);
 230     cpu_to_le32s(&header->sectors);
 231     cpu_to_le32s(&header->sector_size);
 232     cpu_to_le64s(&header->disk_size);
 233     cpu_to_le32s(&header->block_size);
 234     cpu_to_le32s(&header->block_extra);
 235     cpu_to_le32s(&header->blocks_in_image);
 236     cpu_to_le32s(&header->blocks_allocated);
 237     cpu_to_le32s(&header->blocks_allocated);
 238     uuid_convert(header->uuid_image);
 239     uuid_convert(header->uuid_last_snap);
 240     uuid_convert(header->uuid_link);
 241     uuid_convert(header->uuid_parent);
 242 }
 243
 244 #if defined(CONFIG_VDI_DEBUG)
 245 static void vdi_header_print(VdiHeader *header)
 246 {
 247     char uuid[37];
 248     logout("text        %s", header->text);
 249     logout("signature   0x%04x\n", header->signature);
 250     logout("header size 0x%04x\n", header->header_size);
 251     logout("image type  0x%04x\n", header->image_type);
 252     logout("image flags 0x%04x\n", header->image_flags);
 253     logout("description %s\n", header->description);
 254     logout("offset bmap 0x%04x\n", header->offset_bmap);
 255     logout("offset data 0x%04x\n", header->offset_data);
 256     logout("cylinders   0x%04x\n", header->cylinders);
 257     logout("heads       0x%04x\n", header->heads);
 258     logout("sectors     0x%04x\n", header->sectors);
 259     logout("sector size 0x%04x\n", header->sector_size);
 260     logout("image size  0x%" PRIx64 " B (%" PRIu64 " MiB)\n",
 261            header->disk_size, header->disk_size / MiB);
 262     logout("block size  0x%04x\n", header->block_size);
 263     logout("block extra 0x%04x\n", header->block_extra);
 264     logout("blocks tot. 0x%04x\n", header->blocks_in_image);
 265     logout("blocks all. 0x%04x\n", header->blocks_allocated);
 266     uuid_unparse(header->uuid_image, uuid);
 267     logout("uuid image  %s\n", uuid);
 268     uuid_unparse(header->uuid_last_snap, uuid);
 269     logout("uuid snap   %s\n", uuid);
 270     uuid_unparse(header->uuid_link, uuid);
 271     logout("uuid link   %s\n", uuid);
 272     uuid_unparse(header->uuid_parent, uuid);
 273     logout("uuid parent %s\n", uuid);
 274 }
 275 #endif
 276
 277 static int vdi_check(BlockDriverState *bs, BdrvCheckResult *res,
 278                      BdrvCheckMode fix)
 279 {
 280     /* TODO: additional checks possible. */
 281     BDRVVdiState *s = (BDRVVdiState *)bs->opaque;
 282     uint32_t blocks_allocated = 0;
 283     uint32_t block;
 284     uint32_t *bmap;
 285     logout("\n");
 286
 287     if (fix) {
 288         return -ENOTSUP;
 289     }
 290
 291     bmap = g_malloc(s->header.blocks_in_image * sizeof(uint32_t));
 292     memset(bmap, 0xff, s->header.blocks_in_image * sizeof(uint32_t));
 293
 294     /* Check block map and value of blocks_allocated. */
 295     for (block = 0; block < s->header.blocks_in_image; block++) {
 296         uint32_t bmap_entry = le32_to_cpu(s->bmap[block]);
 297         if (VDI_IS_ALLOCATED(bmap_entry)) {
 298             if (bmap_entry < s->header.blocks_in_image) {
 299                 blocks_allocated++;
 300                 if (!VDI_IS_ALLOCATED(bmap[bmap_entry])) {
 301                     bmap[bmap_entry] = bmap_entry;
 302                 } else {
 303                     fprintf(stderr, "ERROR: block index %" PRIu32
 304                             " also used by %" PRIu32 "\n", bmap[bmap_entry], bmap_entry);
 305                     res->corruptions++;
 306                 }
 307             } else {
 308                 fprintf(stderr, "ERROR: block index %" PRIu32
 309                         " too large, is %" PRIu32 "\n", block, bmap_entry);
 310                 res->corruptions++;
 311             }
 312         }
 313     }
 314     if (blocks_allocated != s->header.blocks_allocated) {
 315         fprintf(stderr, "ERROR: allocated blocks mismatch, is %" PRIu32
 316                ", should be %" PRIu32 "\n",
 317                blocks_allocated, s->header.blocks_allocated);
 318         res->corruptions++;
 319     }
 320
 321     g_free(bmap);
 322
 323     return 0;
 324 }
 325
 326 static int vdi_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
 327 {
 328     /* TODO: vdi_get_info would be needed for machine snapshots.
 329        vm_state_offset is still missing. */
 330     BDRVVdiState *s = (BDRVVdiState *)bs->opaque;
 331     logout("\n");
 332     bdi->cluster_size = s->block_size;
 333     bdi->vm_state_offset = 0;
 334     return 0;
 335 }
 336
 337 static int vdi_make_empty(BlockDriverState *bs)
 338 {
 339     /* TODO: missing code. */
 340     logout("\n");
 341     /* The return value for missing code must be 0, see block.c. */
 342     return 0;
 343 }
 344
 345 static int vdi_probe(const uint8_t *buf, int buf_size, const char *filename)
 346 {
 347     const VdiHeader *header = (const VdiHeader *)buf;
 348     int result = 0;
 349
 350     logout("\n");
 351
 352     if (buf_size < sizeof(*header)) {
 353         /* Header too small, no VDI. */
 354     } else if (le32_to_cpu(header->signature) == VDI_SIGNATURE) {
 355         result = 100;
 356     }
 357
 358     if (result == 0) {
 359         logout("no vdi image\n");
 360     } else {
 361         logout("%s", header->text);
 362     }
 363
 364     return result;
 365 }
 366
 367 static int vdi_open(BlockDriverState *bs, int flags)
 368 {
 369     BDRVVdiState *s = bs->opaque;
 370     VdiHeader header;
 371     size_t bmap_size;
 372
 373     logout("\n");
 374
 375     if (bdrv_read(bs->file, 0, (uint8_t *)&header, 1) < 0) {
 376         goto fail;
 377     }
 378
 379     vdi_header_to_cpu(&header);
 380 #if defined(CONFIG_VDI_DEBUG)
 381     vdi_header_print(&header);
 382 #endif
 383
 384     if (header.disk_size % SECTOR_SIZE != 0) {
 385         /* 'VBoxManage convertfromraw' can create images with odd disk sizes.
 386            We accept them but round the disk size to the next multiple of
 387            SECTOR_SIZE. */
 388         logout("odd disk size %" PRIu64 " B, round up\n", header.disk_size);
 389         header.disk_size += SECTOR_SIZE - 1;
 390         header.disk_size &= ~(SECTOR_SIZE - 1);
 391     }
 392
 393     if (header.version != VDI_VERSION_1_1) {
 394         logout("unsupported version %u.%u\n",
 395                header.version >> 16, header.version & 0xffff);
 396         goto fail;
 397     } else if (header.offset_bmap % SECTOR_SIZE != 0) {
 398         /* We only support block maps which start on a sector boundary. */
 399         logout("unsupported block map offset 0x%x B\n", header.offset_bmap);
 400         goto fail;
 401     } else if (header.offset_data % SECTOR_SIZE != 0) {
 402         /* We only support data blocks which start on a sector boundary. */
 403         logout("unsupported data offset 0x%x B\n", header.offset_data);
 404         goto fail;
 405     } else if (header.sector_size != SECTOR_SIZE) {
 406         logout("unsupported sector size %u B\n", header.sector_size);
 407         goto fail;
 408     } else if (header.block_size != 1 * MiB) {
 409         logout("unsupported block size %u B\n", header.block_size);
 410         goto fail;
 411     } else if (header.disk_size >
 412                (uint64_t)header.blocks_in_image * header.block_size) {
 413         logout("unsupported disk size %" PRIu64 " B\n", header.disk_size);
 414         goto fail;
 415     } else if (!uuid_is_null(header.uuid_link)) {
 416         logout("link uuid != 0, unsupported\n");
 417         goto fail;
 418     } else if (!uuid_is_null(header.uuid_parent)) {
 419         logout("parent uuid != 0, unsupported\n");
 420         goto fail;
 421     }
 422
 423     bs->total_sectors = header.disk_size / SECTOR_SIZE;
 424
 425     s->block_size = header.block_size;
 426     s->block_sectors = header.block_size / SECTOR_SIZE;
 427     s->bmap_sector = header.offset_bmap / SECTOR_SIZE;
 428     s->header = header;
 429
 430     bmap_size = header.blocks_in_image * sizeof(uint32_t);
 431     bmap_size = (bmap_size + SECTOR_SIZE - 1) / SECTOR_SIZE;
 432     if (bmap_size > 0) {
 433         s->bmap = g_malloc(bmap_size * SECTOR_SIZE);
 434     }
 435     if (bdrv_read(bs->file, s->bmap_sector, (uint8_t *)s->bmap, bmap_size) < 0) {
 436         goto fail_free_bmap;
 437     }
 438
 439     /* Disable migration when vdi images are used */
 440     error_set(&s->migration_blocker,
 441               QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED,
 442               "vdi", bs->device_name, "live migration");
 443     migrate_add_blocker(s->migration_blocker);
 444
 445     return 0;
 446
 447  fail_free_bmap:
 448     g_free(s->bmap);
 449
 450  fail:
 451     return -1;
 452 }
 453
 454 static int vdi_reopen_prepare(BDRVReopenState *state,
 455                               BlockReopenQueue *queue, Error **errp)
 456 {
 457     return 0;
 458 }
 459
 460 static int coroutine_fn vdi_co_is_allocated(BlockDriverState *bs,
 461         int64_t sector_num, int nb_sectors, int *pnum)
 462 {
 463     /* TODO: Check for too large sector_num (in bdrv_is_allocated or here). */
 464     BDRVVdiState *s = (BDRVVdiState *)bs->opaque;
 465     size_t bmap_index = sector_num / s->block_sectors;
 466     size_t sector_in_block = sector_num % s->block_sectors;
 467     int n_sectors = s->block_sectors - sector_in_block;
 468     uint32_t bmap_entry = le32_to_cpu(s->bmap[bmap_index]);
 469     logout("%p, %" PRId64 ", %d, %p\n", bs, sector_num, nb_sectors, pnum);
 470     if (n_sectors > nb_sectors) {
 471         n_sectors = nb_sectors;
 472     }
 473     *pnum = n_sectors;
 474     return VDI_IS_ALLOCATED(bmap_entry);
 475 }
 476
 477 static int vdi_co_read(BlockDriverState *bs,
 478         int64_t sector_num, uint8_t *buf, int nb_sectors)
 479 {
 480     BDRVVdiState *s = bs->opaque;
 481     uint32_t bmap_entry;
 482     uint32_t block_index;
 483     uint32_t sector_in_block;
 484     uint32_t n_sectors;
 485     int ret = 0;
 486
 487     logout("\n");
 488
 489     while (ret >= 0 && nb_sectors > 0) {
 490         block_index = sector_num / s->block_sectors;
 491         sector_in_block = sector_num % s->block_sectors;
 492         n_sectors = s->block_sectors - sector_in_block;
 493         if (n_sectors > nb_sectors) {
 494             n_sectors = nb_sectors;
 495         }
 496
 497         logout("will read %u sectors starting at sector %" PRIu64 "\n",
 498                n_sectors, sector_num);
 499
 500         /* prepare next AIO request */
 501         bmap_entry = le32_to_cpu(s->bmap[block_index]);
 502         if (!VDI_IS_ALLOCATED(bmap_entry)) {
 503             /* Block not allocated, return zeros, no need to wait. */
 504             memset(buf, 0, n_sectors * SECTOR_SIZE);
 505             ret = 0;
 506         } else {
 507             uint64_t offset = s->header.offset_data / SECTOR_SIZE +
 508                               (uint64_t)bmap_entry * s->block_sectors +
 509                               sector_in_block;
 510             ret = bdrv_read(bs->file, offset, buf, n_sectors);
 511         }
 512         logout("%u sectors read\n", n_sectors);
 513
 514         nb_sectors -= n_sectors;
 515         sector_num += n_sectors;
 516         buf += n_sectors * SECTOR_SIZE;
 517     }
 518
 519     return ret;
 520 }
 521
 522 static int vdi_co_write(BlockDriverState *bs,
 523         int64_t sector_num, const uint8_t *buf, int nb_sectors)
 524 {
 525     BDRVVdiState *s = bs->opaque;
 526     uint32_t bmap_entry;
 527     uint32_t block_index;
 528     uint32_t sector_in_block;
 529     uint32_t n_sectors;
 530     uint32_t bmap_first = VDI_UNALLOCATED;
 531     uint32_t bmap_last = VDI_UNALLOCATED;
 532     uint8_t *block = NULL;
 533     int ret = 0;
 534
 535     logout("\n");
 536
 537     while (ret >= 0 && nb_sectors > 0) {
 538         block_index = sector_num / s->block_sectors;
 539         sector_in_block = sector_num % s->block_sectors;
 540         n_sectors = s->block_sectors - sector_in_block;
 541         if (n_sectors > nb_sectors) {
 542             n_sectors = nb_sectors;
 543         }
 544
 545         logout("will write %u sectors starting at sector %" PRIu64 "\n",
 546                n_sectors, sector_num);
 547
 548         /* prepare next AIO request */
 549         bmap_entry = le32_to_cpu(s->bmap[block_index]);
 550         if (!VDI_IS_ALLOCATED(bmap_entry)) {
 551             /* Allocate new block and write to it. */
 552             uint64_t offset;
 553             bmap_entry = s->header.blocks_allocated;
 554             s->bmap[block_index] = cpu_to_le32(bmap_entry);
 555             s->header.blocks_allocated++;
 556             offset = s->header.offset_data / SECTOR_SIZE +
 557                      (uint64_t)bmap_entry * s->block_sectors;
 558             if (block == NULL) {
 559                 block = g_malloc(s->block_size);
 560                 bmap_first = block_index;
 561             }
 562             bmap_last = block_index;
 563             /* Copy data to be written to new block and zero unused parts. */
 564             memset(block, 0, sector_in_block * SECTOR_SIZE);
 565             memcpy(block + sector_in_block * SECTOR_SIZE,
 566                    buf, n_sectors * SECTOR_SIZE);
 567             memset(block + (sector_in_block + n_sectors) * SECTOR_SIZE, 0,
 568                    (s->block_sectors - n_sectors - sector_in_block) * SECTOR_SIZE);
 569             ret = bdrv_write(bs->file, offset, block, s->block_sectors);
 570         } else {
 571             uint64_t offset = s->header.offset_data / SECTOR_SIZE +
 572                               (uint64_t)bmap_entry * s->block_sectors +
 573                               sector_in_block;
 574             ret = bdrv_write(bs->file, offset, buf, n_sectors);
 575         }
 576
 577         nb_sectors -= n_sectors;
 578         sector_num += n_sectors;
 579         buf += n_sectors * SECTOR_SIZE;
 580
 581         logout("%u sectors written\n", n_sectors);
 582     }
 583
 584     logout("finished data write\n");
 585     if (ret < 0) {
 586         return ret;
 587     }
 588
 589     if (block) {
 590         /* One or more new blocks were allocated. */
 591         VdiHeader *header = (VdiHeader *) block;
 592         uint8_t *base;
 593         uint64_t offset;
 594
 595         logout("now writing modified header\n");
 596         assert(VDI_IS_ALLOCATED(bmap_first));
 597         *header = s->header;
 598         vdi_header_to_le(header);
 599         ret = bdrv_write(bs->file, 0, block, 1);
 600         g_free(block);
 601         block = NULL;
 602
 603         if (ret < 0) {
 604             return ret;
 605         }
 606
 607         logout("now writing modified block map entry %u...%u\n",
 608                bmap_first, bmap_last);
 609         /* Write modified sectors from block map. */
 610         bmap_first /= (SECTOR_SIZE / sizeof(uint32_t));
 611         bmap_last /= (SECTOR_SIZE / sizeof(uint32_t));
 612         n_sectors = bmap_last - bmap_first + 1;
 613         offset = s->bmap_sector + bmap_first;
 614         base = ((uint8_t *)&s->bmap[0]) + bmap_first * SECTOR_SIZE;
 615         logout("will write %u block map sectors starting from entry %u\n",
 616                n_sectors, bmap_first);
 617         ret = bdrv_write(bs->file, offset, base, n_sectors);
 618     }
 619
 620     return ret;
 621 }
 622
 623 static int vdi_create(const char *filename, QEMUOptionParameter *options)
 624 {
 625     int fd;
 626     int result = 0;
 627     uint64_t bytes = 0;
 628     uint32_t blocks;
 629     size_t block_size = DEFAULT_CLUSTER_SIZE;
 630     uint32_t image_type = VDI_TYPE_DYNAMIC;
 631     VdiHeader header;
 632     size_t i;
 633     size_t bmap_size;
 634
 635     logout("\n");
 636
 637     /* Read out options. */
 638     while (options && options->name) {
 639         if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
 640             bytes = options->value.n;
 641 #if defined(CONFIG_VDI_BLOCK_SIZE)
 642         } else if (!strcmp(options->name, BLOCK_OPT_CLUSTER_SIZE)) {
 643             if (options->value.n) {
 644                 /* TODO: Additional checks (SECTOR_SIZE * 2^n, ...). */
 645                 block_size = options->value.n;
 646             }
 647 #endif
 648 #if defined(CONFIG_VDI_STATIC_IMAGE)
 649         } else if (!strcmp(options->name, BLOCK_OPT_STATIC)) {
 650             if (options->value.n) {
 651                 image_type = VDI_TYPE_STATIC;
 652             }
 653 #endif
 654         }
 655         options++;
 656     }
 657
 658     fd = qemu_open(filename,
 659                    O_WRONLY | O_CREAT | O_TRUNC | O_BINARY | O_LARGEFILE,
 660                    0644);
 661     if (fd < 0) {
 662         return -errno;
 663     }
 664
 665     /* We need enough blocks to store the given disk size,
 666        so always round up. */
 667     blocks = (bytes + block_size - 1) / block_size;
 668
 669     bmap_size = blocks * sizeof(uint32_t);
 670     bmap_size = ((bmap_size + SECTOR_SIZE - 1) & ~(SECTOR_SIZE -1));
 671
 672     memset(&header, 0, sizeof(header));
 673     pstrcpy(header.text, sizeof(header.text), VDI_TEXT);
 674     header.signature = VDI_SIGNATURE;
 675     header.version = VDI_VERSION_1_1;
 676     header.header_size = 0x180;
 677     header.image_type = image_type;
 678     header.offset_bmap = 0x200;
 679     header.offset_data = 0x200 + bmap_size;
 680     header.sector_size = SECTOR_SIZE;
 681     header.disk_size = bytes;
 682     header.block_size = block_size;
 683     header.blocks_in_image = blocks;
 684     if (image_type == VDI_TYPE_STATIC) {
 685         header.blocks_allocated = blocks;
 686     }
 687     uuid_generate(header.uuid_image);
 688     uuid_generate(header.uuid_last_snap);
 689     /* There is no need to set header.uuid_link or header.uuid_parent here. */
 690 #if defined(CONFIG_VDI_DEBUG)
 691     vdi_header_print(&header);
 692 #endif
 693     vdi_header_to_le(&header);
 694     if (write(fd, &header, sizeof(header)) < 0) {
 695         result = -errno;
 696     }
 697
 698     if (bmap_size > 0) {
 699         uint32_t *bmap = g_malloc0(bmap_size);
 700         for (i = 0; i < blocks; i++) {
 701             if (image_type == VDI_TYPE_STATIC) {
 702                 bmap[i] = i;
 703             } else {
 704                 bmap[i] = VDI_UNALLOCATED;
 705             }
 706         }
 707         if (write(fd, bmap, bmap_size) < 0) {
 708             result = -errno;
 709         }
 710         g_free(bmap);
 711     }
 712
 713     if (image_type == VDI_TYPE_STATIC) {
 714         if (ftruncate(fd, sizeof(header) + bmap_size + blocks * block_size)) {
 715             result = -errno;
 716         }
 717     }
 718
 719     if (close(fd) < 0) {
 720         result = -errno;
 721     }
 722
 723     return result;
 724 }
 725
 726 static void vdi_close(BlockDriverState *bs)
 727 {
 728     BDRVVdiState *s = bs->opaque;
 729
 730     g_free(s->bmap);
 731
 732     migrate_del_blocker(s->migration_blocker);
 733     error_free(s->migration_blocker);
 734 }
 735
 736 static QEMUOptionParameter vdi_create_options[] = {
 737     {
 738         .name = BLOCK_OPT_SIZE,
 739         .type = OPT_SIZE,
 740         .help = "Virtual disk size"
 741     },
 742 #if defined(CONFIG_VDI_BLOCK_SIZE)
 743     {
 744         .name = BLOCK_OPT_CLUSTER_SIZE,
 745         .type = OPT_SIZE,
 746         .help = "VDI cluster (block) size",
 747         .value = { .n = DEFAULT_CLUSTER_SIZE },
 748     },
 749 #endif
 750 #if defined(CONFIG_VDI_STATIC_IMAGE)
 751     {
 752         .name = BLOCK_OPT_STATIC,
 753         .type = OPT_FLAG,
 754         .help = "VDI static (pre-allocated) image"
 755     },
 756 #endif
 757     /* TODO: An additional option to set UUID values might be useful. */
 758     { NULL }
 759 };
 760
 761 static BlockDriver bdrv_vdi = {
 762     .format_name = "vdi",
 763     .instance_size = sizeof(BDRVVdiState),
 764     .bdrv_probe = vdi_probe,
 765     .bdrv_open = vdi_open,
 766     .bdrv_close = vdi_close,
 767     .bdrv_reopen_prepare = vdi_reopen_prepare,
 768     .bdrv_create = vdi_create,
 769     .bdrv_co_is_allocated = vdi_co_is_allocated,
 770     .bdrv_make_empty = vdi_make_empty,
 771
 772     .bdrv_read = vdi_co_read,
 773 #if defined(CONFIG_VDI_WRITE)
 774     .bdrv_write = vdi_co_write,
 775 #endif
 776
 777     .bdrv_get_info = vdi_get_info,
 778
 779     .create_options = vdi_create_options,
 780     .bdrv_check = vdi_check,
 781 };
 782
 783 static void bdrv_vdi_init(void)
 784 {
 785     logout("\n");
 786     bdrv_register(&bdrv_vdi);
 787 }
 788
 789 block_init(bdrv_vdi_init);