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 adjacent 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 /* max blocks in image is (0xffffffff / 4) */
 124 #define VDI_BLOCKS_IN_IMAGE_MAX  0x3fffffff
 125 #define VDI_DISK_SIZE_MAX        ((uint64_t)VDI_BLOCKS_IN_IMAGE_MAX * \
 126                                   (uint64_t)DEFAULT_CLUSTER_SIZE)
 127
 128 #if !defined(CONFIG_UUID)
 129 static inline void uuid_generate(uuid_t out)
 130 {
 131     memset(out, 0, sizeof(uuid_t));
 132 }
 133
 134 static inline int uuid_is_null(const uuid_t uu)
 135 {
 136     uuid_t null_uuid = { 0 };
 137     return memcmp(uu, null_uuid, sizeof(uuid_t)) == 0;
 138 }
 139
 140 static inline void uuid_unparse(const uuid_t uu, char *out)
 141 {
 142     snprintf(out, 37, UUID_FMT,
 143             uu[0], uu[1], uu[2], uu[3], uu[4], uu[5], uu[6], uu[7],
 144             uu[8], uu[9], uu[10], uu[11], uu[12], uu[13], uu[14], uu[15]);
 145 }
 146 #endif
 147
 148 typedef struct {
 149     char text[0x40];
 150     uint32_t signature;
 151     uint32_t version;
 152     uint32_t header_size;
 153     uint32_t image_type;
 154     uint32_t image_flags;
 155     char description[256];
 156     uint32_t offset_bmap;
 157     uint32_t offset_data;
 158     uint32_t cylinders;         /* disk geometry, unused here */
 159     uint32_t heads;             /* disk geometry, unused here */
 160     uint32_t sectors;           /* disk geometry, unused here */
 161     uint32_t sector_size;
 162     uint32_t unused1;
 163     uint64_t disk_size;
 164     uint32_t block_size;
 165     uint32_t block_extra;       /* unused here */
 166     uint32_t blocks_in_image;
 167     uint32_t blocks_allocated;
 168     uuid_t uuid_image;
 169     uuid_t uuid_last_snap;
 170     uuid_t uuid_link;
 171     uuid_t uuid_parent;
 172     uint64_t unused2[7];
 173 } QEMU_PACKED VdiHeader;
 174
 175 typedef struct {
 176     /* The block map entries are little endian (even in memory). */
 177     uint32_t *bmap;
 178     /* Size of block (bytes). */
 179     uint32_t block_size;
 180     /* Size of block (sectors). */
 181     uint32_t block_sectors;
 182     /* First sector of block map. */
 183     uint32_t bmap_sector;
 184     /* VDI header (converted to host endianness). */
 185     VdiHeader header;
 186
 187     Error *migration_blocker;
 188 } BDRVVdiState;
 189
 190 /* Change UUID from little endian (IPRT = VirtualBox format) to big endian
 191  * format (network byte order, standard, see RFC 4122) and vice versa.
 192  */
 193 static void uuid_convert(uuid_t uuid)
 194 {
 195     bswap32s((uint32_t *)&uuid[0]);
 196     bswap16s((uint16_t *)&uuid[4]);
 197     bswap16s((uint16_t *)&uuid[6]);
 198 }
 199
 200 static void vdi_header_to_cpu(VdiHeader *header)
 201 {
 202     le32_to_cpus(&header->signature);
 203     le32_to_cpus(&header->version);
 204     le32_to_cpus(&header->header_size);
 205     le32_to_cpus(&header->image_type);
 206     le32_to_cpus(&header->image_flags);
 207     le32_to_cpus(&header->offset_bmap);
 208     le32_to_cpus(&header->offset_data);
 209     le32_to_cpus(&header->cylinders);
 210     le32_to_cpus(&header->heads);
 211     le32_to_cpus(&header->sectors);
 212     le32_to_cpus(&header->sector_size);
 213     le64_to_cpus(&header->disk_size);
 214     le32_to_cpus(&header->block_size);
 215     le32_to_cpus(&header->block_extra);
 216     le32_to_cpus(&header->blocks_in_image);
 217     le32_to_cpus(&header->blocks_allocated);
 218     uuid_convert(header->uuid_image);
 219     uuid_convert(header->uuid_last_snap);
 220     uuid_convert(header->uuid_link);
 221     uuid_convert(header->uuid_parent);
 222 }
 223
 224 static void vdi_header_to_le(VdiHeader *header)
 225 {
 226     cpu_to_le32s(&header->signature);
 227     cpu_to_le32s(&header->version);
 228     cpu_to_le32s(&header->header_size);
 229     cpu_to_le32s(&header->image_type);
 230     cpu_to_le32s(&header->image_flags);
 231     cpu_to_le32s(&header->offset_bmap);
 232     cpu_to_le32s(&header->offset_data);
 233     cpu_to_le32s(&header->cylinders);
 234     cpu_to_le32s(&header->heads);
 235     cpu_to_le32s(&header->sectors);
 236     cpu_to_le32s(&header->sector_size);
 237     cpu_to_le64s(&header->disk_size);
 238     cpu_to_le32s(&header->block_size);
 239     cpu_to_le32s(&header->block_extra);
 240     cpu_to_le32s(&header->blocks_in_image);
 241     cpu_to_le32s(&header->blocks_allocated);
 242     cpu_to_le32s(&header->blocks_allocated);
 243     uuid_convert(header->uuid_image);
 244     uuid_convert(header->uuid_last_snap);
 245     uuid_convert(header->uuid_link);
 246     uuid_convert(header->uuid_parent);
 247 }
 248
 249 #if defined(CONFIG_VDI_DEBUG)
 250 static void vdi_header_print(VdiHeader *header)
 251 {
 252     char uuid[37];
 253     logout("text        %s", header->text);
 254     logout("signature   0x%08x\n", header->signature);
 255     logout("header size 0x%04x\n", header->header_size);
 256     logout("image type  0x%04x\n", header->image_type);
 257     logout("image flags 0x%04x\n", header->image_flags);
 258     logout("description %s\n", header->description);
 259     logout("offset bmap 0x%04x\n", header->offset_bmap);
 260     logout("offset data 0x%04x\n", header->offset_data);
 261     logout("cylinders   0x%04x\n", header->cylinders);
 262     logout("heads       0x%04x\n", header->heads);
 263     logout("sectors     0x%04x\n", header->sectors);
 264     logout("sector size 0x%04x\n", header->sector_size);
 265     logout("image size  0x%" PRIx64 " B (%" PRIu64 " MiB)\n",
 266            header->disk_size, header->disk_size / MiB);
 267     logout("block size  0x%04x\n", header->block_size);
 268     logout("block extra 0x%04x\n", header->block_extra);
 269     logout("blocks tot. 0x%04x\n", header->blocks_in_image);
 270     logout("blocks all. 0x%04x\n", header->blocks_allocated);
 271     uuid_unparse(header->uuid_image, uuid);
 272     logout("uuid image  %s\n", uuid);
 273     uuid_unparse(header->uuid_last_snap, uuid);
 274     logout("uuid snap   %s\n", uuid);
 275     uuid_unparse(header->uuid_link, uuid);
 276     logout("uuid link   %s\n", uuid);
 277     uuid_unparse(header->uuid_parent, uuid);
 278     logout("uuid parent %s\n", uuid);
 279 }
 280 #endif
 281
 282 static int vdi_check(BlockDriverState *bs, BdrvCheckResult *res,
 283                      BdrvCheckMode fix)
 284 {
 285     /* TODO: additional checks possible. */
 286     BDRVVdiState *s = (BDRVVdiState *)bs->opaque;
 287     uint32_t blocks_allocated = 0;
 288     uint32_t block;
 289     uint32_t *bmap;
 290     logout("\n");
 291
 292     if (fix) {
 293         return -ENOTSUP;
 294     }
 295
 296     bmap = g_malloc(s->header.blocks_in_image * sizeof(uint32_t));
 297     memset(bmap, 0xff, s->header.blocks_in_image * sizeof(uint32_t));
 298
 299     /* Check block map and value of blocks_allocated. */
 300     for (block = 0; block < s->header.blocks_in_image; block++) {
 301         uint32_t bmap_entry = le32_to_cpu(s->bmap[block]);
 302         if (VDI_IS_ALLOCATED(bmap_entry)) {
 303             if (bmap_entry < s->header.blocks_in_image) {
 304                 blocks_allocated++;
 305                 if (!VDI_IS_ALLOCATED(bmap[bmap_entry])) {
 306                     bmap[bmap_entry] = bmap_entry;
 307                 } else {
 308                     fprintf(stderr, "ERROR: block index %" PRIu32
 309                             " also used by %" PRIu32 "\n", bmap[bmap_entry], bmap_entry);
 310                     res->corruptions++;
 311                 }
 312             } else {
 313                 fprintf(stderr, "ERROR: block index %" PRIu32
 314                         " too large, is %" PRIu32 "\n", block, bmap_entry);
 315                 res->corruptions++;
 316             }
 317         }
 318     }
 319     if (blocks_allocated != s->header.blocks_allocated) {
 320         fprintf(stderr, "ERROR: allocated blocks mismatch, is %" PRIu32
 321                ", should be %" PRIu32 "\n",
 322                blocks_allocated, s->header.blocks_allocated);
 323         res->corruptions++;
 324     }
 325
 326     g_free(bmap);
 327
 328     return 0;
 329 }
 330
 331 static int vdi_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
 332 {
 333     /* TODO: vdi_get_info would be needed for machine snapshots.
 334        vm_state_offset is still missing. */
 335     BDRVVdiState *s = (BDRVVdiState *)bs->opaque;
 336     logout("\n");
 337     bdi->cluster_size = s->block_size;
 338     bdi->vm_state_offset = 0;
 339     bdi->unallocated_blocks_are_zero = true;
 340     return 0;
 341 }
 342
 343 static int vdi_make_empty(BlockDriverState *bs)
 344 {
 345     /* TODO: missing code. */
 346     logout("\n");
 347     /* The return value for missing code must be 0, see block.c. */
 348     return 0;
 349 }
 350
 351 static int vdi_probe(const uint8_t *buf, int buf_size, const char *filename)
 352 {
 353     const VdiHeader *header = (const VdiHeader *)buf;
 354     int result = 0;
 355
 356     logout("\n");
 357
 358     if (buf_size < sizeof(*header)) {
 359         /* Header too small, no VDI. */
 360     } else if (le32_to_cpu(header->signature) == VDI_SIGNATURE) {
 361         result = 100;
 362     }
 363
 364     if (result == 0) {
 365         logout("no vdi image\n");
 366     } else {
 367         logout("%s", header->text);
 368     }
 369
 370     return result;
 371 }
 372
 373 static int vdi_open(BlockDriverState *bs, QDict *options, int flags,
 374                     Error **errp)
 375 {
 376     BDRVVdiState *s = bs->opaque;
 377     VdiHeader header;
 378     size_t bmap_size;
 379     int ret;
 380
 381     logout("\n");
 382
 383     ret = bdrv_read(bs->file, 0, (uint8_t *)&header, 1);
 384     if (ret < 0) {
 385         goto fail;
 386     }
 387
 388     vdi_header_to_cpu(&header);
 389 #if defined(CONFIG_VDI_DEBUG)
 390     vdi_header_print(&header);
 391 #endif
 392
 393     if (header.disk_size > VDI_DISK_SIZE_MAX) {
 394         error_setg(errp, "Unsupported VDI image size (size is 0x%" PRIx64
 395                           ", max supported is 0x%" PRIx64 ")",
 396                           header.disk_size, VDI_DISK_SIZE_MAX);
 397         ret = -ENOTSUP;
 398         goto fail;
 399     }
 400
 401     if (header.disk_size % SECTOR_SIZE != 0) {
 402         /* 'VBoxManage convertfromraw' can create images with odd disk sizes.
 403            We accept them but round the disk size to the next multiple of
 404            SECTOR_SIZE. */
 405         logout("odd disk size %" PRIu64 " B, round up\n", header.disk_size);
 406         header.disk_size += SECTOR_SIZE - 1;
 407         header.disk_size &= ~(SECTOR_SIZE - 1);
 408     }
 409
 410     if (header.signature != VDI_SIGNATURE) {
 411         error_setg(errp, "Image not in VDI format (bad signature %08" PRIx32
 412                    ")", header.signature);
 413         ret = -EINVAL;
 414         goto fail;
 415     } else if (header.version != VDI_VERSION_1_1) {
 416         error_setg(errp, "unsupported VDI image (version %" PRIu32 ".%" PRIu32
 417                    ")", header.version >> 16, header.version & 0xffff);
 418         ret = -ENOTSUP;
 419         goto fail;
 420     } else if (header.offset_bmap % SECTOR_SIZE != 0) {
 421         /* We only support block maps which start on a sector boundary. */
 422         error_setg(errp, "unsupported VDI image (unaligned block map offset "
 423                    "0x%" PRIx32 ")", header.offset_bmap);
 424         ret = -ENOTSUP;
 425         goto fail;
 426     } else if (header.offset_data % SECTOR_SIZE != 0) {
 427         /* We only support data blocks which start on a sector boundary. */
 428         error_setg(errp, "unsupported VDI image (unaligned data offset 0x%"
 429                    PRIx32 ")", header.offset_data);
 430         ret = -ENOTSUP;
 431         goto fail;
 432     } else if (header.sector_size != SECTOR_SIZE) {
 433         error_setg(errp, "unsupported VDI image (sector size %" PRIu32
 434                    " is not %u)", header.sector_size, SECTOR_SIZE);
 435         ret = -ENOTSUP;
 436         goto fail;
 437     } else if (header.block_size != DEFAULT_CLUSTER_SIZE) {
 438         error_setg(errp, "unsupported VDI image (block size %" PRIu32
 439                    " is not %u)", header.block_size, DEFAULT_CLUSTER_SIZE);
 440         ret = -ENOTSUP;
 441         goto fail;
 442     } else if (header.disk_size >
 443                (uint64_t)header.blocks_in_image * header.block_size) {
 444         error_setg(errp, "unsupported VDI image (disk size %" PRIu64 ", "
 445                    "image bitmap has room for %" PRIu64 ")",
 446                    header.disk_size,
 447                    (uint64_t)header.blocks_in_image * header.block_size);
 448         ret = -ENOTSUP;
 449         goto fail;
 450     } else if (!uuid_is_null(header.uuid_link)) {
 451         error_setg(errp, "unsupported VDI image (non-NULL link UUID)");
 452         ret = -ENOTSUP;
 453         goto fail;
 454     } else if (!uuid_is_null(header.uuid_parent)) {
 455         error_setg(errp, "unsupported VDI image (non-NULL parent UUID)");
 456         ret = -ENOTSUP;
 457         goto fail;
 458     } else if (header.blocks_in_image > VDI_BLOCKS_IN_IMAGE_MAX) {
 459         error_setg(errp, "unsupported VDI image "
 460                          "(too many blocks %u, max is %u)",
 461                           header.blocks_in_image, VDI_BLOCKS_IN_IMAGE_MAX);
 462         ret = -ENOTSUP;
 463         goto fail;
 464     }
 465
 466     bs->total_sectors = header.disk_size / SECTOR_SIZE;
 467
 468     s->block_size = header.block_size;
 469     s->block_sectors = header.block_size / SECTOR_SIZE;
 470     s->bmap_sector = header.offset_bmap / SECTOR_SIZE;
 471     s->header = header;
 472
 473     bmap_size = header.blocks_in_image * sizeof(uint32_t);
 474     bmap_size = (bmap_size + SECTOR_SIZE - 1) / SECTOR_SIZE;
 475     s->bmap = g_malloc(bmap_size * SECTOR_SIZE);
 476     ret = bdrv_read(bs->file, s->bmap_sector, (uint8_t *)s->bmap, bmap_size);
 477     if (ret < 0) {
 478         goto fail_free_bmap;
 479     }
 480
 481     /* Disable migration when vdi images are used */
 482     error_set(&s->migration_blocker,
 483               QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED,
 484               "vdi", bs->device_name, "live migration");
 485     migrate_add_blocker(s->migration_blocker);
 486
 487     return 0;
 488
 489  fail_free_bmap:
 490     g_free(s->bmap);
 491
 492  fail:
 493     return ret;
 494 }
 495
 496 static int vdi_reopen_prepare(BDRVReopenState *state,
 497                               BlockReopenQueue *queue, Error **errp)
 498 {
 499     return 0;
 500 }
 501
 502 static int64_t coroutine_fn vdi_co_get_block_status(BlockDriverState *bs,
 503         int64_t sector_num, int nb_sectors, int *pnum)
 504 {
 505     /* TODO: Check for too large sector_num (in bdrv_is_allocated or here). */
 506     BDRVVdiState *s = (BDRVVdiState *)bs->opaque;
 507     size_t bmap_index = sector_num / s->block_sectors;
 508     size_t sector_in_block = sector_num % s->block_sectors;
 509     int n_sectors = s->block_sectors - sector_in_block;
 510     uint32_t bmap_entry = le32_to_cpu(s->bmap[bmap_index]);
 511     uint64_t offset;
 512     int result;
 513
 514     logout("%p, %" PRId64 ", %d, %p\n", bs, sector_num, nb_sectors, pnum);
 515     if (n_sectors > nb_sectors) {
 516         n_sectors = nb_sectors;
 517     }
 518     *pnum = n_sectors;
 519     result = VDI_IS_ALLOCATED(bmap_entry);
 520     if (!result) {
 521         return 0;
 522     }
 523
 524     offset = s->header.offset_data +
 525                               (uint64_t)bmap_entry * s->block_size +
 526                               sector_in_block * SECTOR_SIZE;
 527     return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID | offset;
 528 }
 529
 530 static int vdi_co_read(BlockDriverState *bs,
 531         int64_t sector_num, uint8_t *buf, int nb_sectors)
 532 {
 533     BDRVVdiState *s = bs->opaque;
 534     uint32_t bmap_entry;
 535     uint32_t block_index;
 536     uint32_t sector_in_block;
 537     uint32_t n_sectors;
 538     int ret = 0;
 539
 540     logout("\n");
 541
 542     while (ret >= 0 && nb_sectors > 0) {
 543         block_index = sector_num / s->block_sectors;
 544         sector_in_block = sector_num % s->block_sectors;
 545         n_sectors = s->block_sectors - sector_in_block;
 546         if (n_sectors > nb_sectors) {
 547             n_sectors = nb_sectors;
 548         }
 549
 550         logout("will read %u sectors starting at sector %" PRIu64 "\n",
 551                n_sectors, sector_num);
 552
 553         /* prepare next AIO request */
 554         bmap_entry = le32_to_cpu(s->bmap[block_index]);
 555         if (!VDI_IS_ALLOCATED(bmap_entry)) {
 556             /* Block not allocated, return zeros, no need to wait. */
 557             memset(buf, 0, n_sectors * SECTOR_SIZE);
 558             ret = 0;
 559         } else {
 560             uint64_t offset = s->header.offset_data / SECTOR_SIZE +
 561                               (uint64_t)bmap_entry * s->block_sectors +
 562                               sector_in_block;
 563             ret = bdrv_read(bs->file, offset, buf, n_sectors);
 564         }
 565         logout("%u sectors read\n", n_sectors);
 566
 567         nb_sectors -= n_sectors;
 568         sector_num += n_sectors;
 569         buf += n_sectors * SECTOR_SIZE;
 570     }
 571
 572     return ret;
 573 }
 574
 575 static int vdi_co_write(BlockDriverState *bs,
 576         int64_t sector_num, const uint8_t *buf, int nb_sectors)
 577 {
 578     BDRVVdiState *s = bs->opaque;
 579     uint32_t bmap_entry;
 580     uint32_t block_index;
 581     uint32_t sector_in_block;
 582     uint32_t n_sectors;
 583     uint32_t bmap_first = VDI_UNALLOCATED;
 584     uint32_t bmap_last = VDI_UNALLOCATED;
 585     uint8_t *block = NULL;
 586     int ret = 0;
 587
 588     logout("\n");
 589
 590     while (ret >= 0 && nb_sectors > 0) {
 591         block_index = sector_num / s->block_sectors;
 592         sector_in_block = sector_num % s->block_sectors;
 593         n_sectors = s->block_sectors - sector_in_block;
 594         if (n_sectors > nb_sectors) {
 595             n_sectors = nb_sectors;
 596         }
 597
 598         logout("will write %u sectors starting at sector %" PRIu64 "\n",
 599                n_sectors, sector_num);
 600
 601         /* prepare next AIO request */
 602         bmap_entry = le32_to_cpu(s->bmap[block_index]);
 603         if (!VDI_IS_ALLOCATED(bmap_entry)) {
 604             /* Allocate new block and write to it. */
 605             uint64_t offset;
 606             bmap_entry = s->header.blocks_allocated;
 607             s->bmap[block_index] = cpu_to_le32(bmap_entry);
 608             s->header.blocks_allocated++;
 609             offset = s->header.offset_data / SECTOR_SIZE +
 610                      (uint64_t)bmap_entry * s->block_sectors;
 611             if (block == NULL) {
 612                 block = g_malloc(s->block_size);
 613                 bmap_first = block_index;
 614             }
 615             bmap_last = block_index;
 616             /* Copy data to be written to new block and zero unused parts. */
 617             memset(block, 0, sector_in_block * SECTOR_SIZE);
 618             memcpy(block + sector_in_block * SECTOR_SIZE,
 619                    buf, n_sectors * SECTOR_SIZE);
 620             memset(block + (sector_in_block + n_sectors) * SECTOR_SIZE, 0,
 621                    (s->block_sectors - n_sectors - sector_in_block) * SECTOR_SIZE);
 622             ret = bdrv_write(bs->file, offset, block, s->block_sectors);
 623         } else {
 624             uint64_t offset = s->header.offset_data / SECTOR_SIZE +
 625                               (uint64_t)bmap_entry * s->block_sectors +
 626                               sector_in_block;
 627             ret = bdrv_write(bs->file, offset, buf, n_sectors);
 628         }
 629
 630         nb_sectors -= n_sectors;
 631         sector_num += n_sectors;
 632         buf += n_sectors * SECTOR_SIZE;
 633
 634         logout("%u sectors written\n", n_sectors);
 635     }
 636
 637     logout("finished data write\n");
 638     if (ret < 0) {
 639         return ret;
 640     }
 641
 642     if (block) {
 643         /* One or more new blocks were allocated. */
 644         VdiHeader *header = (VdiHeader *) block;
 645         uint8_t *base;
 646         uint64_t offset;
 647
 648         logout("now writing modified header\n");
 649         assert(VDI_IS_ALLOCATED(bmap_first));
 650         *header = s->header;
 651         vdi_header_to_le(header);
 652         ret = bdrv_write(bs->file, 0, block, 1);
 653         g_free(block);
 654         block = NULL;
 655
 656         if (ret < 0) {
 657             return ret;
 658         }
 659
 660         logout("now writing modified block map entry %u...%u\n",
 661                bmap_first, bmap_last);
 662         /* Write modified sectors from block map. */
 663         bmap_first /= (SECTOR_SIZE / sizeof(uint32_t));
 664         bmap_last /= (SECTOR_SIZE / sizeof(uint32_t));
 665         n_sectors = bmap_last - bmap_first + 1;
 666         offset = s->bmap_sector + bmap_first;
 667         base = ((uint8_t *)&s->bmap[0]) + bmap_first * SECTOR_SIZE;
 668         logout("will write %u block map sectors starting from entry %u\n",
 669                n_sectors, bmap_first);
 670         ret = bdrv_write(bs->file, offset, base, n_sectors);
 671     }
 672
 673     return ret;
 674 }
 675
 676 static int vdi_create(const char *filename, QEMUOptionParameter *options,
 677                       Error **errp)
 678 {
 679     int fd;
 680     int result = 0;
 681     uint64_t bytes = 0;
 682     uint32_t blocks;
 683     size_t block_size = DEFAULT_CLUSTER_SIZE;
 684     uint32_t image_type = VDI_TYPE_DYNAMIC;
 685     VdiHeader header;
 686     size_t i;
 687     size_t bmap_size;
 688
 689     logout("\n");
 690
 691     /* Read out options. */
 692     while (options && options->name) {
 693         if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
 694             bytes = options->value.n;
 695 #if defined(CONFIG_VDI_BLOCK_SIZE)
 696         } else if (!strcmp(options->name, BLOCK_OPT_CLUSTER_SIZE)) {
 697             if (options->value.n) {
 698                 /* TODO: Additional checks (SECTOR_SIZE * 2^n, ...). */
 699                 block_size = options->value.n;
 700             }
 701 #endif
 702 #if defined(CONFIG_VDI_STATIC_IMAGE)
 703         } else if (!strcmp(options->name, BLOCK_OPT_STATIC)) {
 704             if (options->value.n) {
 705                 image_type = VDI_TYPE_STATIC;
 706             }
 707 #endif
 708         }
 709         options++;
 710     }
 711
 712     if (bytes > VDI_DISK_SIZE_MAX) {
 713         result = -ENOTSUP;
 714         error_setg(errp, "Unsupported VDI image size (size is 0x%" PRIx64
 715                           ", max supported is 0x%" PRIx64 ")",
 716                           bytes, VDI_DISK_SIZE_MAX);
 717         goto exit;
 718     }
 719
 720     fd = qemu_open(filename,
 721                    O_WRONLY | O_CREAT | O_TRUNC | O_BINARY | O_LARGEFILE,
 722                    0644);
 723     if (fd < 0) {
 724         result = -errno;
 725         goto exit;
 726     }
 727
 728     /* We need enough blocks to store the given disk size,
 729        so always round up. */
 730     blocks = (bytes + block_size - 1) / block_size;
 731
 732     bmap_size = blocks * sizeof(uint32_t);
 733     bmap_size = ((bmap_size + SECTOR_SIZE - 1) & ~(SECTOR_SIZE -1));
 734
 735     memset(&header, 0, sizeof(header));
 736     pstrcpy(header.text, sizeof(header.text), VDI_TEXT);
 737     header.signature = VDI_SIGNATURE;
 738     header.version = VDI_VERSION_1_1;
 739     header.header_size = 0x180;
 740     header.image_type = image_type;
 741     header.offset_bmap = 0x200;
 742     header.offset_data = 0x200 + bmap_size;
 743     header.sector_size = SECTOR_SIZE;
 744     header.disk_size = bytes;
 745     header.block_size = block_size;
 746     header.blocks_in_image = blocks;
 747     if (image_type == VDI_TYPE_STATIC) {
 748         header.blocks_allocated = blocks;
 749     }
 750     uuid_generate(header.uuid_image);
 751     uuid_generate(header.uuid_last_snap);
 752     /* There is no need to set header.uuid_link or header.uuid_parent here. */
 753 #if defined(CONFIG_VDI_DEBUG)
 754     vdi_header_print(&header);
 755 #endif
 756     vdi_header_to_le(&header);
 757     if (write(fd, &header, sizeof(header)) < 0) {
 758         result = -errno;
 759         goto close_and_exit;
 760     }
 761
 762     if (bmap_size > 0) {
 763         uint32_t *bmap = g_malloc0(bmap_size);
 764         for (i = 0; i < blocks; i++) {
 765             if (image_type == VDI_TYPE_STATIC) {
 766                 bmap[i] = i;
 767             } else {
 768                 bmap[i] = VDI_UNALLOCATED;
 769             }
 770         }
 771         if (write(fd, bmap, bmap_size) < 0) {
 772             result = -errno;
 773             g_free(bmap);
 774             goto close_and_exit;
 775         }
 776         g_free(bmap);
 777     }
 778
 779     if (image_type == VDI_TYPE_STATIC) {
 780         if (ftruncate(fd, sizeof(header) + bmap_size + blocks * block_size)) {
 781             result = -errno;
 782             goto close_and_exit;
 783         }
 784     }
 785
 786 close_and_exit:
 787     if ((close(fd) < 0) && !result) {
 788         result = -errno;
 789     }
 790
 791 exit:
 792     return result;
 793 }
 794
 795 static void vdi_close(BlockDriverState *bs)
 796 {
 797     BDRVVdiState *s = bs->opaque;
 798
 799     g_free(s->bmap);
 800
 801     migrate_del_blocker(s->migration_blocker);
 802     error_free(s->migration_blocker);
 803 }
 804
 805 static QEMUOptionParameter vdi_create_options[] = {
 806     {
 807         .name = BLOCK_OPT_SIZE,
 808         .type = OPT_SIZE,
 809         .help = "Virtual disk size"
 810     },
 811 #if defined(CONFIG_VDI_BLOCK_SIZE)
 812     {
 813         .name = BLOCK_OPT_CLUSTER_SIZE,
 814         .type = OPT_SIZE,
 815         .help = "VDI cluster (block) size",
 816         .value = { .n = DEFAULT_CLUSTER_SIZE },
 817     },
 818 #endif
 819 #if defined(CONFIG_VDI_STATIC_IMAGE)
 820     {
 821         .name = BLOCK_OPT_STATIC,
 822         .type = OPT_FLAG,
 823         .help = "VDI static (pre-allocated) image"
 824     },
 825 #endif
 826     /* TODO: An additional option to set UUID values might be useful. */
 827     { NULL }
 828 };
 829
 830 static BlockDriver bdrv_vdi = {
 831     .format_name = "vdi",
 832     .instance_size = sizeof(BDRVVdiState),
 833     .bdrv_probe = vdi_probe,
 834     .bdrv_open = vdi_open,
 835     .bdrv_close = vdi_close,
 836     .bdrv_reopen_prepare = vdi_reopen_prepare,
 837     .bdrv_create = vdi_create,
 838     .bdrv_has_zero_init = bdrv_has_zero_init_1,
 839     .bdrv_co_get_block_status = vdi_co_get_block_status,
 840     .bdrv_make_empty = vdi_make_empty,
 841
 842     .bdrv_read = vdi_co_read,
 843 #if defined(CONFIG_VDI_WRITE)
 844     .bdrv_write = vdi_co_write,
 845 #endif
 846
 847     .bdrv_get_info = vdi_get_info,
 848
 849     .create_options = vdi_create_options,
 850     .bdrv_check = vdi_check,
 851 };
 852
 853 static void bdrv_vdi_init(void)
 854 {
 855     logout("\n");
 856     bdrv_register(&bdrv_vdi);
 857 }
 858
 859 block_init(bdrv_vdi_init);