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