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                      BdrvCheckMode fix)
 282 {
 283     /* TODO: additional checks possible. */
 284     BDRVVdiState *s = (BDRVVdiState *)bs->opaque;
 285     uint32_t blocks_allocated = 0;
 286     uint32_t block;
 287     uint32_t *bmap;
 288     logout("\n");
 289
 290     if (fix) {
 291         return -ENOTSUP;
 292     }
 293
 294     bmap = g_malloc(s->header.blocks_in_image * sizeof(uint32_t));
 295     memset(bmap, 0xff, s->header.blocks_in_image * sizeof(uint32_t));
 296
 297     /* Check block map and value of blocks_allocated. */
 298     for (block = 0; block < s->header.blocks_in_image; block++) {
 299         uint32_t bmap_entry = le32_to_cpu(s->bmap[block]);
 300         if (VDI_IS_ALLOCATED(bmap_entry)) {
 301             if (bmap_entry < s->header.blocks_in_image) {
 302                 blocks_allocated++;
 303                 if (!VDI_IS_ALLOCATED(bmap[bmap_entry])) {
 304                     bmap[bmap_entry] = bmap_entry;
 305                 } else {
 306                     fprintf(stderr, "ERROR: block index %" PRIu32
 307                             " also used by %" PRIu32 "\n", bmap[bmap_entry], bmap_entry);
 308                     res->corruptions++;
 309                 }
 310             } else {
 311                 fprintf(stderr, "ERROR: block index %" PRIu32
 312                         " too large, is %" PRIu32 "\n", block, bmap_entry);
 313                 res->corruptions++;
 314             }
 315         }
 316     }
 317     if (blocks_allocated != s->header.blocks_allocated) {
 318         fprintf(stderr, "ERROR: allocated blocks mismatch, is %" PRIu32
 319                ", should be %" PRIu32 "\n",
 320                blocks_allocated, s->header.blocks_allocated);
 321         res->corruptions++;
 322     }
 323
 324     g_free(bmap);
 325
 326     return 0;
 327 }
 328
 329 static int vdi_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
 330 {
 331     /* TODO: vdi_get_info would be needed for machine snapshots.
 332        vm_state_offset is still missing. */
 333     BDRVVdiState *s = (BDRVVdiState *)bs->opaque;
 334     logout("\n");
 335     bdi->cluster_size = s->block_size;
 336     bdi->vm_state_offset = 0;
 337     return 0;
 338 }
 339
 340 static int vdi_make_empty(BlockDriverState *bs)
 341 {
 342     /* TODO: missing code. */
 343     logout("\n");
 344     /* The return value for missing code must be 0, see block.c. */
 345     return 0;
 346 }
 347
 348 static int vdi_probe(const uint8_t *buf, int buf_size, const char *filename)
 349 {
 350     const VdiHeader *header = (const VdiHeader *)buf;
 351     int result = 0;
 352
 353     logout("\n");
 354
 355     if (buf_size < sizeof(*header)) {
 356         /* Header too small, no VDI. */
 357     } else if (le32_to_cpu(header->signature) == VDI_SIGNATURE) {
 358         result = 100;
 359     }
 360
 361     if (result == 0) {
 362         logout("no vdi image\n");
 363     } else {
 364         logout("%s", header->text);
 365     }
 366
 367     return result;
 368 }
 369
 370 static int vdi_open(BlockDriverState *bs, int flags)
 371 {
 372     BDRVVdiState *s = bs->opaque;
 373     VdiHeader header;
 374     size_t bmap_size;
 375
 376     logout("\n");
 377
 378     if (bdrv_read(bs->file, 0, (uint8_t *)&header, 1) < 0) {
 379         goto fail;
 380     }
 381
 382     vdi_header_to_cpu(&header);
 383 #if defined(CONFIG_VDI_DEBUG)
 384     vdi_header_print(&header);
 385 #endif
 386
 387     if (header.disk_size % SECTOR_SIZE != 0) {
 388         /* 'VBoxManage convertfromraw' can create images with odd disk sizes.
 389            We accept them but round the disk size to the next multiple of
 390            SECTOR_SIZE. */
 391         logout("odd disk size %" PRIu64 " B, round up\n", header.disk_size);
 392         header.disk_size += SECTOR_SIZE - 1;
 393         header.disk_size &= ~(SECTOR_SIZE - 1);
 394     }
 395
 396     if (header.version != VDI_VERSION_1_1) {
 397         logout("unsupported version %u.%u\n",
 398                header.version >> 16, header.version & 0xffff);
 399         goto fail;
 400     } else if (header.offset_bmap % SECTOR_SIZE != 0) {
 401         /* We only support block maps which start on a sector boundary. */
 402         logout("unsupported block map offset 0x%x B\n", header.offset_bmap);
 403         goto fail;
 404     } else if (header.offset_data % SECTOR_SIZE != 0) {
 405         /* We only support data blocks which start on a sector boundary. */
 406         logout("unsupported data offset 0x%x B\n", header.offset_data);
 407         goto fail;
 408     } else if (header.sector_size != SECTOR_SIZE) {
 409         logout("unsupported sector size %u B\n", header.sector_size);
 410         goto fail;
 411     } else if (header.block_size != 1 * MiB) {
 412         logout("unsupported block size %u B\n", header.block_size);
 413         goto fail;
 414     } else if (header.disk_size >
 415                (uint64_t)header.blocks_in_image * header.block_size) {
 416         logout("unsupported disk size %" PRIu64 " B\n", header.disk_size);
 417         goto fail;
 418     } else if (!uuid_is_null(header.uuid_link)) {
 419         logout("link uuid != 0, unsupported\n");
 420         goto fail;
 421     } else if (!uuid_is_null(header.uuid_parent)) {
 422         logout("parent uuid != 0, unsupported\n");
 423         goto fail;
 424     }
 425
 426     bs->total_sectors = header.disk_size / SECTOR_SIZE;
 427
 428     s->block_size = header.block_size;
 429     s->block_sectors = header.block_size / SECTOR_SIZE;
 430     s->bmap_sector = header.offset_bmap / SECTOR_SIZE;
 431     s->header = header;
 432
 433     bmap_size = header.blocks_in_image * sizeof(uint32_t);
 434     bmap_size = (bmap_size + SECTOR_SIZE - 1) / SECTOR_SIZE;
 435     if (bmap_size > 0) {
 436         s->bmap = g_malloc(bmap_size * SECTOR_SIZE);
 437     }
 438     if (bdrv_read(bs->file, s->bmap_sector, (uint8_t *)s->bmap, bmap_size) < 0) {
 439         goto fail_free_bmap;
 440     }
 441
 442     /* Disable migration when vdi images are used */
 443     error_set(&s->migration_blocker,
 444               QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED,
 445               "vdi", bs->device_name, "live migration");
 446     migrate_add_blocker(s->migration_blocker);
 447
 448     return 0;
 449
 450  fail_free_bmap:
 451     g_free(s->bmap);
 452
 453  fail:
 454     return -1;
 455 }
 456
 457 static int vdi_reopen_prepare(BDRVReopenState *state,
 458                               BlockReopenQueue *queue, Error **errp)
 459 {
 460     return 0;
 461 }
 462
 463 static int coroutine_fn vdi_co_is_allocated(BlockDriverState *bs,
 464         int64_t sector_num, int nb_sectors, int *pnum)
 465 {
 466     /* TODO: Check for too large sector_num (in bdrv_is_allocated or here). */
 467     BDRVVdiState *s = (BDRVVdiState *)bs->opaque;
 468     size_t bmap_index = sector_num / s->block_sectors;
 469     size_t sector_in_block = sector_num % s->block_sectors;
 470     int n_sectors = s->block_sectors - sector_in_block;
 471     uint32_t bmap_entry = le32_to_cpu(s->bmap[bmap_index]);
 472     logout("%p, %" PRId64 ", %d, %p\n", bs, sector_num, nb_sectors, pnum);
 473     if (n_sectors > nb_sectors) {
 474         n_sectors = nb_sectors;
 475     }
 476     *pnum = n_sectors;
 477     return VDI_IS_ALLOCATED(bmap_entry);
 478 }
 479
 480 static int vdi_co_read(BlockDriverState *bs,
 481         int64_t sector_num, uint8_t *buf, int nb_sectors)
 482 {
 483     BDRVVdiState *s = bs->opaque;
 484     uint32_t bmap_entry;
 485     uint32_t block_index;
 486     uint32_t sector_in_block;
 487     uint32_t n_sectors;
 488     int ret = 0;
 489
 490     logout("\n");
 491
 492     while (ret >= 0 && nb_sectors > 0) {
 493         block_index = sector_num / s->block_sectors;
 494         sector_in_block = sector_num % s->block_sectors;
 495         n_sectors = s->block_sectors - sector_in_block;
 496         if (n_sectors > nb_sectors) {
 497             n_sectors = nb_sectors;
 498         }
 499
 500         logout("will read %u sectors starting at sector %" PRIu64 "\n",
 501                n_sectors, sector_num);
 502
 503         /* prepare next AIO request */
 504         bmap_entry = le32_to_cpu(s->bmap[block_index]);
 505         if (!VDI_IS_ALLOCATED(bmap_entry)) {
 506             /* Block not allocated, return zeros, no need to wait. */
 507             memset(buf, 0, n_sectors * SECTOR_SIZE);
 508             ret = 0;
 509         } else {
 510             uint64_t offset = s->header.offset_data / SECTOR_SIZE +
 511                               (uint64_t)bmap_entry * s->block_sectors +
 512                               sector_in_block;
 513             ret = bdrv_read(bs->file, offset, buf, n_sectors);
 514         }
 515         logout("%u sectors read\n", n_sectors);
 516
 517         nb_sectors -= n_sectors;
 518         sector_num += n_sectors;
 519         buf += n_sectors * SECTOR_SIZE;
 520     }
 521
 522     return ret;
 523 }
 524
 525 static int vdi_co_write(BlockDriverState *bs,
 526         int64_t sector_num, const uint8_t *buf, int nb_sectors)
 527 {
 528     BDRVVdiState *s = bs->opaque;
 529     uint32_t bmap_entry;
 530     uint32_t block_index;
 531     uint32_t sector_in_block;
 532     uint32_t n_sectors;
 533     uint32_t bmap_first = VDI_UNALLOCATED;
 534     uint32_t bmap_last = VDI_UNALLOCATED;
 535     uint8_t *block = NULL;
 536     int ret = 0;
 537
 538     logout("\n");
 539
 540     while (ret >= 0 && nb_sectors > 0) {
 541         block_index = sector_num / s->block_sectors;
 542         sector_in_block = sector_num % s->block_sectors;
 543         n_sectors = s->block_sectors - sector_in_block;
 544         if (n_sectors > nb_sectors) {
 545             n_sectors = nb_sectors;
 546         }
 547
 548         logout("will write %u sectors starting at sector %" PRIu64 "\n",
 549                n_sectors, sector_num);
 550
 551         /* prepare next AIO request */
 552         bmap_entry = le32_to_cpu(s->bmap[block_index]);
 553         if (!VDI_IS_ALLOCATED(bmap_entry)) {
 554             /* Allocate new block and write to it. */
 555             uint64_t offset;
 556             bmap_entry = s->header.blocks_allocated;
 557             s->bmap[block_index] = cpu_to_le32(bmap_entry);
 558             s->header.blocks_allocated++;
 559             offset = s->header.offset_data / SECTOR_SIZE +
 560                      (uint64_t)bmap_entry * s->block_sectors;
 561             if (block == NULL) {
 562                 block = g_malloc(s->block_size);
 563                 bmap_first = block_index;
 564             }
 565             bmap_last = block_index;
 566             /* Copy data to be written to new block and zero unused parts. */
 567             memset(block, 0, sector_in_block * SECTOR_SIZE);
 568             memcpy(block + sector_in_block * SECTOR_SIZE,
 569                    buf, n_sectors * SECTOR_SIZE);
 570             memset(block + (sector_in_block + n_sectors) * SECTOR_SIZE, 0,
 571                    (s->block_sectors - n_sectors - sector_in_block) * SECTOR_SIZE);
 572             ret = bdrv_write(bs->file, offset, block, s->block_sectors);
 573         } else {
 574             uint64_t offset = s->header.offset_data / SECTOR_SIZE +
 575                               (uint64_t)bmap_entry * s->block_sectors +
 576                               sector_in_block;
 577             ret = bdrv_write(bs->file, offset, buf, n_sectors);
 578         }
 579
 580         nb_sectors -= n_sectors;
 581         sector_num += n_sectors;
 582         buf += n_sectors * SECTOR_SIZE;
 583
 584         logout("%u sectors written\n", n_sectors);
 585     }
 586
 587     logout("finished data write\n");
 588     if (ret < 0) {
 589         return ret;
 590     }
 591
 592     if (block) {
 593         /* One or more new blocks were allocated. */
 594         VdiHeader *header = (VdiHeader *) block;
 595         uint8_t *base;
 596         uint64_t offset;
 597
 598         logout("now writing modified header\n");
 599         assert(VDI_IS_ALLOCATED(bmap_first));
 600         *header = s->header;
 601         vdi_header_to_le(header);
 602         ret = bdrv_write(bs->file, 0, block, 1);
 603         g_free(block);
 604         block = NULL;
 605
 606         if (ret < 0) {
 607             return ret;
 608         }
 609
 610         logout("now writing modified block map entry %u...%u\n",
 611                bmap_first, bmap_last);
 612         /* Write modified sectors from block map. */
 613         bmap_first /= (SECTOR_SIZE / sizeof(uint32_t));
 614         bmap_last /= (SECTOR_SIZE / sizeof(uint32_t));
 615         n_sectors = bmap_last - bmap_first + 1;
 616         offset = s->bmap_sector + bmap_first;
 617         base = ((uint8_t *)&s->bmap[0]) + bmap_first * SECTOR_SIZE;
 618         logout("will write %u block map sectors starting from entry %u\n",
 619                n_sectors, bmap_first);
 620         ret = bdrv_write(bs->file, offset, base, n_sectors);
 621     }
 622
 623     return ret;
 624 }
 625
 626 static int vdi_create(const char *filename, QEMUOptionParameter *options)
 627 {
 628     int fd;
 629     int result = 0;
 630     uint64_t bytes = 0;
 631     uint32_t blocks;
 632     size_t block_size = DEFAULT_CLUSTER_SIZE;
 633     uint32_t image_type = VDI_TYPE_DYNAMIC;
 634     VdiHeader header;
 635     size_t i;
 636     size_t bmap_size;
 637
 638     logout("\n");
 639
 640     /* Read out options. */
 641     while (options && options->name) {
 642         if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
 643             bytes = options->value.n;
 644 #if defined(CONFIG_VDI_BLOCK_SIZE)
 645         } else if (!strcmp(options->name, BLOCK_OPT_CLUSTER_SIZE)) {
 646             if (options->value.n) {
 647                 /* TODO: Additional checks (SECTOR_SIZE * 2^n, ...). */
 648                 block_size = options->value.n;
 649             }
 650 #endif
 651 #if defined(CONFIG_VDI_STATIC_IMAGE)
 652         } else if (!strcmp(options->name, BLOCK_OPT_STATIC)) {
 653             if (options->value.n) {
 654                 image_type = VDI_TYPE_STATIC;
 655             }
 656 #endif
 657         }
 658         options++;
 659     }
 660
 661     fd = qemu_open(filename,
 662                    O_WRONLY | O_CREAT | O_TRUNC | O_BINARY | O_LARGEFILE,
 663                    0644);
 664     if (fd < 0) {
 665         return -errno;
 666     }
 667
 668     /* We need enough blocks to store the given disk size,
 669        so always round up. */
 670     blocks = (bytes + block_size - 1) / block_size;
 671
 672     bmap_size = blocks * sizeof(uint32_t);
 673     bmap_size = ((bmap_size + SECTOR_SIZE - 1) & ~(SECTOR_SIZE -1));
 674
 675     memset(&header, 0, sizeof(header));
 676     pstrcpy(header.text, sizeof(header.text), VDI_TEXT);
 677     header.signature = VDI_SIGNATURE;
 678     header.version = VDI_VERSION_1_1;
 679     header.header_size = 0x180;
 680     header.image_type = image_type;
 681     header.offset_bmap = 0x200;
 682     header.offset_data = 0x200 + bmap_size;
 683     header.sector_size = SECTOR_SIZE;
 684     header.disk_size = bytes;
 685     header.block_size = block_size;
 686     header.blocks_in_image = blocks;
 687     if (image_type == VDI_TYPE_STATIC) {
 688         header.blocks_allocated = blocks;
 689     }
 690     uuid_generate(header.uuid_image);
 691     uuid_generate(header.uuid_last_snap);
 692     /* There is no need to set header.uuid_link or header.uuid_parent here. */
 693 #if defined(CONFIG_VDI_DEBUG)
 694     vdi_header_print(&header);
 695 #endif
 696     vdi_header_to_le(&header);
 697     if (write(fd, &header, sizeof(header)) < 0) {
 698         result = -errno;
 699     }
 700
 701     if (bmap_size > 0) {
 702         uint32_t *bmap = g_malloc0(bmap_size);
 703         for (i = 0; i < blocks; i++) {
 704             if (image_type == VDI_TYPE_STATIC) {
 705                 bmap[i] = i;
 706             } else {
 707                 bmap[i] = VDI_UNALLOCATED;
 708             }
 709         }
 710         if (write(fd, bmap, bmap_size) < 0) {
 711             result = -errno;
 712         }
 713         g_free(bmap);
 714     }
 715
 716     if (image_type == VDI_TYPE_STATIC) {
 717         if (ftruncate(fd, sizeof(header) + bmap_size + blocks * block_size)) {
 718             result = -errno;
 719         }
 720     }
 721
 722     if (close(fd) < 0) {
 723         result = -errno;
 724     }
 725
 726     return result;
 727 }
 728
 729 static void vdi_close(BlockDriverState *bs)
 730 {
 731     BDRVVdiState *s = bs->opaque;
 732
 733     g_free(s->bmap);
 734
 735     migrate_del_blocker(s->migration_blocker);
 736     error_free(s->migration_blocker);
 737 }
 738
 739 static QEMUOptionParameter vdi_create_options[] = {
 740     {
 741         .name = BLOCK_OPT_SIZE,
 742         .type = OPT_SIZE,
 743         .help = "Virtual disk size"
 744     },
 745 #if defined(CONFIG_VDI_BLOCK_SIZE)
 746     {
 747         .name = BLOCK_OPT_CLUSTER_SIZE,
 748         .type = OPT_SIZE,
 749         .help = "VDI cluster (block) size",
 750         .value = { .n = DEFAULT_CLUSTER_SIZE },
 751     },
 752 #endif
 753 #if defined(CONFIG_VDI_STATIC_IMAGE)
 754     {
 755         .name = BLOCK_OPT_STATIC,
 756         .type = OPT_FLAG,
 757         .help = "VDI static (pre-allocated) image"
 758     },
 759 #endif
 760     /* TODO: An additional option to set UUID values might be useful. */
 761     { NULL }
 762 };
 763
 764 static BlockDriver bdrv_vdi = {
 765     .format_name = "vdi",
 766     .instance_size = sizeof(BDRVVdiState),
 767     .bdrv_probe = vdi_probe,
 768     .bdrv_open = vdi_open,
 769     .bdrv_close = vdi_close,
 770     .bdrv_reopen_prepare = vdi_reopen_prepare,
 771     .bdrv_create = vdi_create,
 772     .bdrv_co_is_allocated = vdi_co_is_allocated,
 773     .bdrv_make_empty = vdi_make_empty,
 774
 775     .bdrv_read = vdi_co_read,
 776 #if defined(CONFIG_VDI_WRITE)
 777     .bdrv_write = vdi_co_write,
 778 #endif
 779
 780     .bdrv_get_info = vdi_get_info,
 781
 782     .create_options = vdi_create_options,
 783     .bdrv_check = vdi_check,
 784 };
 785
 786 static void bdrv_vdi_init(void)
 787 {
 788     logout("\n");
 789     bdrv_register(&bdrv_vdi);
 790 }
 791
 792 block_init(bdrv_vdi_init);