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 # if defined(CONFIG_VDI_DEBUG)
 141 static inline void uuid_unparse(const uuid_t uu, char *out)
 142 {
 143     snprintf(out, 37, UUID_FMT,
 144             uu[0], uu[1], uu[2], uu[3], uu[4], uu[5], uu[6], uu[7],
 145             uu[8], uu[9], uu[10], uu[11], uu[12], uu[13], uu[14], uu[15]);
 146 }
 147 # endif
 148 #endif
 149
 150 typedef struct {
 151     char text[0x40];
 152     uint32_t signature;
 153     uint32_t version;
 154     uint32_t header_size;
 155     uint32_t image_type;
 156     uint32_t image_flags;
 157     char description[256];
 158     uint32_t offset_bmap;
 159     uint32_t offset_data;
 160     uint32_t cylinders;         /* disk geometry, unused here */
 161     uint32_t heads;             /* disk geometry, unused here */
 162     uint32_t sectors;           /* disk geometry, unused here */
 163     uint32_t sector_size;
 164     uint32_t unused1;
 165     uint64_t disk_size;
 166     uint32_t block_size;
 167     uint32_t block_extra;       /* unused here */
 168     uint32_t blocks_in_image;
 169     uint32_t blocks_allocated;
 170     uuid_t uuid_image;
 171     uuid_t uuid_last_snap;
 172     uuid_t uuid_link;
 173     uuid_t uuid_parent;
 174     uint64_t unused2[7];
 175 } QEMU_PACKED VdiHeader;
 176
 177 typedef struct {
 178     /* The block map entries are little endian (even in memory). */
 179     uint32_t *bmap;
 180     /* Size of block (bytes). */
 181     uint32_t block_size;
 182     /* Size of block (sectors). */
 183     uint32_t block_sectors;
 184     /* First sector of block map. */
 185     uint32_t bmap_sector;
 186     /* VDI header (converted to host endianness). */
 187     VdiHeader header;
 188
 189     Error *migration_blocker;
 190 } BDRVVdiState;
 191
 192 /* Change UUID from little endian (IPRT = VirtualBox format) to big endian
 193  * format (network byte order, standard, see RFC 4122) and vice versa.
 194  */
 195 static void uuid_convert(uuid_t uuid)
 196 {
 197     bswap32s((uint32_t *)&uuid[0]);
 198     bswap16s((uint16_t *)&uuid[4]);
 199     bswap16s((uint16_t *)&uuid[6]);
 200 }
 201
 202 static void vdi_header_to_cpu(VdiHeader *header)
 203 {
 204     le32_to_cpus(&header->signature);
 205     le32_to_cpus(&header->version);
 206     le32_to_cpus(&header->header_size);
 207     le32_to_cpus(&header->image_type);
 208     le32_to_cpus(&header->image_flags);
 209     le32_to_cpus(&header->offset_bmap);
 210     le32_to_cpus(&header->offset_data);
 211     le32_to_cpus(&header->cylinders);
 212     le32_to_cpus(&header->heads);
 213     le32_to_cpus(&header->sectors);
 214     le32_to_cpus(&header->sector_size);
 215     le64_to_cpus(&header->disk_size);
 216     le32_to_cpus(&header->block_size);
 217     le32_to_cpus(&header->block_extra);
 218     le32_to_cpus(&header->blocks_in_image);
 219     le32_to_cpus(&header->blocks_allocated);
 220     uuid_convert(header->uuid_image);
 221     uuid_convert(header->uuid_last_snap);
 222     uuid_convert(header->uuid_link);
 223     uuid_convert(header->uuid_parent);
 224 }
 225
 226 static void vdi_header_to_le(VdiHeader *header)
 227 {
 228     cpu_to_le32s(&header->signature);
 229     cpu_to_le32s(&header->version);
 230     cpu_to_le32s(&header->header_size);
 231     cpu_to_le32s(&header->image_type);
 232     cpu_to_le32s(&header->image_flags);
 233     cpu_to_le32s(&header->offset_bmap);
 234     cpu_to_le32s(&header->offset_data);
 235     cpu_to_le32s(&header->cylinders);
 236     cpu_to_le32s(&header->heads);
 237     cpu_to_le32s(&header->sectors);
 238     cpu_to_le32s(&header->sector_size);
 239     cpu_to_le64s(&header->disk_size);
 240     cpu_to_le32s(&header->block_size);
 241     cpu_to_le32s(&header->block_extra);
 242     cpu_to_le32s(&header->blocks_in_image);
 243     cpu_to_le32s(&header->blocks_allocated);
 244     uuid_convert(header->uuid_image);
 245     uuid_convert(header->uuid_last_snap);
 246     uuid_convert(header->uuid_link);
 247     uuid_convert(header->uuid_parent);
 248 }
 249
 250 #if defined(CONFIG_VDI_DEBUG)
 251 static void vdi_header_print(VdiHeader *header)
 252 {
 253     char uuid[37];
 254     logout("text        %s", header->text);
 255     logout("signature   0x%08x\n", header->signature);
 256     logout("header size 0x%04x\n", header->header_size);
 257     logout("image type  0x%04x\n", header->image_type);
 258     logout("image flags 0x%04x\n", header->image_flags);
 259     logout("description %s\n", header->description);
 260     logout("offset bmap 0x%04x\n", header->offset_bmap);
 261     logout("offset data 0x%04x\n", header->offset_data);
 262     logout("cylinders   0x%04x\n", header->cylinders);
 263     logout("heads       0x%04x\n", header->heads);
 264     logout("sectors     0x%04x\n", header->sectors);
 265     logout("sector size 0x%04x\n", header->sector_size);
 266     logout("image size  0x%" PRIx64 " B (%" PRIu64 " MiB)\n",
 267            header->disk_size, header->disk_size / MiB);
 268     logout("block size  0x%04x\n", header->block_size);
 269     logout("block extra 0x%04x\n", header->block_extra);
 270     logout("blocks tot. 0x%04x\n", header->blocks_in_image);
 271     logout("blocks all. 0x%04x\n", header->blocks_allocated);
 272     uuid_unparse(header->uuid_image, uuid);
 273     logout("uuid image  %s\n", uuid);
 274     uuid_unparse(header->uuid_last_snap, uuid);
 275     logout("uuid snap   %s\n", uuid);
 276     uuid_unparse(header->uuid_link, uuid);
 277     logout("uuid link   %s\n", uuid);
 278     uuid_unparse(header->uuid_parent, uuid);
 279     logout("uuid parent %s\n", uuid);
 280 }
 281 #endif
 282
 283 static int vdi_check(BlockDriverState *bs, BdrvCheckResult *res,
 284                      BdrvCheckMode fix)
 285 {
 286     /* TODO: additional checks possible. */
 287     BDRVVdiState *s = (BDRVVdiState *)bs->opaque;
 288     uint32_t blocks_allocated = 0;
 289     uint32_t block;
 290     uint32_t *bmap;
 291     logout("\n");
 292
 293     if (fix) {
 294         return -ENOTSUP;
 295     }
 296
 297     bmap = g_try_new(uint32_t, s->header.blocks_in_image);
 298     if (s->header.blocks_in_image && bmap == NULL) {
 299         res->check_errors++;
 300         return -ENOMEM;
 301     }
 302
 303     memset(bmap, 0xff, s->header.blocks_in_image * sizeof(uint32_t));
 304
 305     /* Check block map and value of blocks_allocated. */
 306     for (block = 0; block < s->header.blocks_in_image; block++) {
 307         uint32_t bmap_entry = le32_to_cpu(s->bmap[block]);
 308         if (VDI_IS_ALLOCATED(bmap_entry)) {
 309             if (bmap_entry < s->header.blocks_in_image) {
 310                 blocks_allocated++;
 311                 if (!VDI_IS_ALLOCATED(bmap[bmap_entry])) {
 312                     bmap[bmap_entry] = bmap_entry;
 313                 } else {
 314                     fprintf(stderr, "ERROR: block index %" PRIu32
 315                             " also used by %" PRIu32 "\n", bmap[bmap_entry], bmap_entry);
 316                     res->corruptions++;
 317                 }
 318             } else {
 319                 fprintf(stderr, "ERROR: block index %" PRIu32
 320                         " too large, is %" PRIu32 "\n", block, bmap_entry);
 321                 res->corruptions++;
 322             }
 323         }
 324     }
 325     if (blocks_allocated != s->header.blocks_allocated) {
 326         fprintf(stderr, "ERROR: allocated blocks mismatch, is %" PRIu32
 327                ", should be %" PRIu32 "\n",
 328                blocks_allocated, s->header.blocks_allocated);
 329         res->corruptions++;
 330     }
 331
 332     g_free(bmap);
 333
 334     return 0;
 335 }
 336
 337 static int vdi_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
 338 {
 339     /* TODO: vdi_get_info would be needed for machine snapshots.
 340        vm_state_offset is still missing. */
 341     BDRVVdiState *s = (BDRVVdiState *)bs->opaque;
 342     logout("\n");
 343     bdi->cluster_size = s->block_size;
 344     bdi->vm_state_offset = 0;
 345     bdi->unallocated_blocks_are_zero = true;
 346     return 0;
 347 }
 348
 349 static int vdi_make_empty(BlockDriverState *bs)
 350 {
 351     /* TODO: missing code. */
 352     logout("\n");
 353     /* The return value for missing code must be 0, see block.c. */
 354     return 0;
 355 }
 356
 357 static int vdi_probe(const uint8_t *buf, int buf_size, const char *filename)
 358 {
 359     const VdiHeader *header = (const VdiHeader *)buf;
 360     int ret = 0;
 361
 362     logout("\n");
 363
 364     if (buf_size < sizeof(*header)) {
 365         /* Header too small, no VDI. */
 366     } else if (le32_to_cpu(header->signature) == VDI_SIGNATURE) {
 367         ret = 100;
 368     }
 369
 370     if (ret == 0) {
 371         logout("no vdi image\n");
 372     } else {
 373         logout("%s", header->text);
 374     }
 375
 376     return ret;
 377 }
 378
 379 static int vdi_open(BlockDriverState *bs, QDict *options, int flags,
 380                     Error **errp)
 381 {
 382     BDRVVdiState *s = bs->opaque;
 383     VdiHeader header;
 384     size_t bmap_size;
 385     int ret;
 386
 387     logout("\n");
 388
 389     ret = bdrv_read(bs->file, 0, (uint8_t *)&header, 1);
 390     if (ret < 0) {
 391         goto fail;
 392     }
 393
 394     vdi_header_to_cpu(&header);
 395 #if defined(CONFIG_VDI_DEBUG)
 396     vdi_header_print(&header);
 397 #endif
 398
 399     if (header.disk_size > VDI_DISK_SIZE_MAX) {
 400         error_setg(errp, "Unsupported VDI image size (size is 0x%" PRIx64
 401                           ", max supported is 0x%" PRIx64 ")",
 402                           header.disk_size, VDI_DISK_SIZE_MAX);
 403         ret = -ENOTSUP;
 404         goto fail;
 405     }
 406
 407     if (header.disk_size % SECTOR_SIZE != 0) {
 408         /* 'VBoxManage convertfromraw' can create images with odd disk sizes.
 409            We accept them but round the disk size to the next multiple of
 410            SECTOR_SIZE. */
 411         logout("odd disk size %" PRIu64 " B, round up\n", header.disk_size);
 412         header.disk_size = ROUND_UP(header.disk_size, SECTOR_SIZE);
 413     }
 414
 415     if (header.signature != VDI_SIGNATURE) {
 416         error_setg(errp, "Image not in VDI format (bad signature %08" PRIx32
 417                    ")", header.signature);
 418         ret = -EINVAL;
 419         goto fail;
 420     } else if (header.version != VDI_VERSION_1_1) {
 421         error_setg(errp, "unsupported VDI image (version %" PRIu32 ".%" PRIu32
 422                    ")", header.version >> 16, header.version & 0xffff);
 423         ret = -ENOTSUP;
 424         goto fail;
 425     } else if (header.offset_bmap % SECTOR_SIZE != 0) {
 426         /* We only support block maps which start on a sector boundary. */
 427         error_setg(errp, "unsupported VDI image (unaligned block map offset "
 428                    "0x%" PRIx32 ")", header.offset_bmap);
 429         ret = -ENOTSUP;
 430         goto fail;
 431     } else if (header.offset_data % SECTOR_SIZE != 0) {
 432         /* We only support data blocks which start on a sector boundary. */
 433         error_setg(errp, "unsupported VDI image (unaligned data offset 0x%"
 434                    PRIx32 ")", header.offset_data);
 435         ret = -ENOTSUP;
 436         goto fail;
 437     } else if (header.sector_size != SECTOR_SIZE) {
 438         error_setg(errp, "unsupported VDI image (sector size %" PRIu32
 439                    " is not %u)", header.sector_size, SECTOR_SIZE);
 440         ret = -ENOTSUP;
 441         goto fail;
 442     } else if (header.block_size != DEFAULT_CLUSTER_SIZE) {
 443         error_setg(errp, "unsupported VDI image (block size %" PRIu32
 444                    " is not %u)", header.block_size, DEFAULT_CLUSTER_SIZE);
 445         ret = -ENOTSUP;
 446         goto fail;
 447     } else if (header.disk_size >
 448                (uint64_t)header.blocks_in_image * header.block_size) {
 449         error_setg(errp, "unsupported VDI image (disk size %" PRIu64 ", "
 450                    "image bitmap has room for %" PRIu64 ")",
 451                    header.disk_size,
 452                    (uint64_t)header.blocks_in_image * header.block_size);
 453         ret = -ENOTSUP;
 454         goto fail;
 455     } else if (!uuid_is_null(header.uuid_link)) {
 456         error_setg(errp, "unsupported VDI image (non-NULL link UUID)");
 457         ret = -ENOTSUP;
 458         goto fail;
 459     } else if (!uuid_is_null(header.uuid_parent)) {
 460         error_setg(errp, "unsupported VDI image (non-NULL parent UUID)");
 461         ret = -ENOTSUP;
 462         goto fail;
 463     } else if (header.blocks_in_image > VDI_BLOCKS_IN_IMAGE_MAX) {
 464         error_setg(errp, "unsupported VDI image "
 465                          "(too many blocks %u, max is %u)",
 466                           header.blocks_in_image, VDI_BLOCKS_IN_IMAGE_MAX);
 467         ret = -ENOTSUP;
 468         goto fail;
 469     }
 470
 471     bs->total_sectors = header.disk_size / SECTOR_SIZE;
 472
 473     s->block_size = header.block_size;
 474     s->block_sectors = header.block_size / SECTOR_SIZE;
 475     s->bmap_sector = header.offset_bmap / SECTOR_SIZE;
 476     s->header = header;
 477
 478     bmap_size = header.blocks_in_image * sizeof(uint32_t);
 479     bmap_size = DIV_ROUND_UP(bmap_size, SECTOR_SIZE);
 480     s->bmap = qemu_try_blockalign(bs->file, bmap_size * SECTOR_SIZE);
 481     if (s->bmap == NULL) {
 482         ret = -ENOMEM;
 483         goto fail;
 484     }
 485
 486     ret = bdrv_read(bs->file, s->bmap_sector, (uint8_t *)s->bmap, bmap_size);
 487     if (ret < 0) {
 488         goto fail_free_bmap;
 489     }
 490
 491     /* Disable migration when vdi images are used */
 492     error_set(&s->migration_blocker,
 493               QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED,
 494               "vdi", bdrv_get_device_name(bs), "live migration");
 495     migrate_add_blocker(s->migration_blocker);
 496
 497     return 0;
 498
 499  fail_free_bmap:
 500     qemu_vfree(s->bmap);
 501
 502  fail:
 503     return ret;
 504 }
 505
 506 static int vdi_reopen_prepare(BDRVReopenState *state,
 507                               BlockReopenQueue *queue, Error **errp)
 508 {
 509     return 0;
 510 }
 511
 512 static int64_t coroutine_fn vdi_co_get_block_status(BlockDriverState *bs,
 513         int64_t sector_num, int nb_sectors, int *pnum)
 514 {
 515     /* TODO: Check for too large sector_num (in bdrv_is_allocated or here). */
 516     BDRVVdiState *s = (BDRVVdiState *)bs->opaque;
 517     size_t bmap_index = sector_num / s->block_sectors;
 518     size_t sector_in_block = sector_num % s->block_sectors;
 519     int n_sectors = s->block_sectors - sector_in_block;
 520     uint32_t bmap_entry = le32_to_cpu(s->bmap[bmap_index]);
 521     uint64_t offset;
 522     int result;
 523
 524     logout("%p, %" PRId64 ", %d, %p\n", bs, sector_num, nb_sectors, pnum);
 525     if (n_sectors > nb_sectors) {
 526         n_sectors = nb_sectors;
 527     }
 528     *pnum = n_sectors;
 529     result = VDI_IS_ALLOCATED(bmap_entry);
 530     if (!result) {
 531         return 0;
 532     }
 533
 534     offset = s->header.offset_data +
 535                               (uint64_t)bmap_entry * s->block_size +
 536                               sector_in_block * SECTOR_SIZE;
 537     return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID | offset;
 538 }
 539
 540 static int vdi_co_read(BlockDriverState *bs,
 541         int64_t sector_num, uint8_t *buf, int nb_sectors)
 542 {
 543     BDRVVdiState *s = bs->opaque;
 544     uint32_t bmap_entry;
 545     uint32_t block_index;
 546     uint32_t sector_in_block;
 547     uint32_t n_sectors;
 548     int ret = 0;
 549
 550     logout("\n");
 551
 552     while (ret >= 0 && nb_sectors > 0) {
 553         block_index = sector_num / s->block_sectors;
 554         sector_in_block = sector_num % s->block_sectors;
 555         n_sectors = s->block_sectors - sector_in_block;
 556         if (n_sectors > nb_sectors) {
 557             n_sectors = nb_sectors;
 558         }
 559
 560         logout("will read %u sectors starting at sector %" PRIu64 "\n",
 561                n_sectors, sector_num);
 562
 563         /* prepare next AIO request */
 564         bmap_entry = le32_to_cpu(s->bmap[block_index]);
 565         if (!VDI_IS_ALLOCATED(bmap_entry)) {
 566             /* Block not allocated, return zeros, no need to wait. */
 567             memset(buf, 0, n_sectors * SECTOR_SIZE);
 568             ret = 0;
 569         } else {
 570             uint64_t offset = s->header.offset_data / SECTOR_SIZE +
 571                               (uint64_t)bmap_entry * s->block_sectors +
 572                               sector_in_block;
 573             ret = bdrv_read(bs->file, offset, buf, n_sectors);
 574         }
 575         logout("%u sectors read\n", n_sectors);
 576
 577         nb_sectors -= n_sectors;
 578         sector_num += n_sectors;
 579         buf += n_sectors * SECTOR_SIZE;
 580     }
 581
 582     return ret;
 583 }
 584
 585 static int vdi_co_write(BlockDriverState *bs,
 586         int64_t sector_num, const uint8_t *buf, int nb_sectors)
 587 {
 588     BDRVVdiState *s = bs->opaque;
 589     uint32_t bmap_entry;
 590     uint32_t block_index;
 591     uint32_t sector_in_block;
 592     uint32_t n_sectors;
 593     uint32_t bmap_first = VDI_UNALLOCATED;
 594     uint32_t bmap_last = VDI_UNALLOCATED;
 595     uint8_t *block = NULL;
 596     int ret = 0;
 597
 598     logout("\n");
 599
 600     while (ret >= 0 && nb_sectors > 0) {
 601         block_index = sector_num / s->block_sectors;
 602         sector_in_block = sector_num % s->block_sectors;
 603         n_sectors = s->block_sectors - sector_in_block;
 604         if (n_sectors > nb_sectors) {
 605             n_sectors = nb_sectors;
 606         }
 607
 608         logout("will write %u sectors starting at sector %" PRIu64 "\n",
 609                n_sectors, sector_num);
 610
 611         /* prepare next AIO request */
 612         bmap_entry = le32_to_cpu(s->bmap[block_index]);
 613         if (!VDI_IS_ALLOCATED(bmap_entry)) {
 614             /* Allocate new block and write to it. */
 615             uint64_t offset;
 616             bmap_entry = s->header.blocks_allocated;
 617             s->bmap[block_index] = cpu_to_le32(bmap_entry);
 618             s->header.blocks_allocated++;
 619             offset = s->header.offset_data / SECTOR_SIZE +
 620                      (uint64_t)bmap_entry * s->block_sectors;
 621             if (block == NULL) {
 622                 block = g_malloc(s->block_size);
 623                 bmap_first = block_index;
 624             }
 625             bmap_last = block_index;
 626             /* Copy data to be written to new block and zero unused parts. */
 627             memset(block, 0, sector_in_block * SECTOR_SIZE);
 628             memcpy(block + sector_in_block * SECTOR_SIZE,
 629                    buf, n_sectors * SECTOR_SIZE);
 630             memset(block + (sector_in_block + n_sectors) * SECTOR_SIZE, 0,
 631                    (s->block_sectors - n_sectors - sector_in_block) * SECTOR_SIZE);
 632             ret = bdrv_write(bs->file, offset, block, s->block_sectors);
 633         } else {
 634             uint64_t offset = s->header.offset_data / SECTOR_SIZE +
 635                               (uint64_t)bmap_entry * s->block_sectors +
 636                               sector_in_block;
 637             ret = bdrv_write(bs->file, offset, buf, n_sectors);
 638         }
 639
 640         nb_sectors -= n_sectors;
 641         sector_num += n_sectors;
 642         buf += n_sectors * SECTOR_SIZE;
 643
 644         logout("%u sectors written\n", n_sectors);
 645     }
 646
 647     logout("finished data write\n");
 648     if (ret < 0) {
 649         return ret;
 650     }
 651
 652     if (block) {
 653         /* One or more new blocks were allocated. */
 654         VdiHeader *header = (VdiHeader *) block;
 655         uint8_t *base;
 656         uint64_t offset;
 657
 658         logout("now writing modified header\n");
 659         assert(VDI_IS_ALLOCATED(bmap_first));
 660         *header = s->header;
 661         vdi_header_to_le(header);
 662         ret = bdrv_write(bs->file, 0, block, 1);
 663         g_free(block);
 664         block = NULL;
 665
 666         if (ret < 0) {
 667             return ret;
 668         }
 669
 670         logout("now writing modified block map entry %u...%u\n",
 671                bmap_first, bmap_last);
 672         /* Write modified sectors from block map. */
 673         bmap_first /= (SECTOR_SIZE / sizeof(uint32_t));
 674         bmap_last /= (SECTOR_SIZE / sizeof(uint32_t));
 675         n_sectors = bmap_last - bmap_first + 1;
 676         offset = s->bmap_sector + bmap_first;
 677         base = ((uint8_t *)&s->bmap[0]) + bmap_first * SECTOR_SIZE;
 678         logout("will write %u block map sectors starting from entry %u\n",
 679                n_sectors, bmap_first);
 680         ret = bdrv_write(bs->file, offset, base, n_sectors);
 681     }
 682
 683     return ret;
 684 }
 685
 686 static int vdi_create(const char *filename, QemuOpts *opts, Error **errp)
 687 {
 688     int ret = 0;
 689     uint64_t bytes = 0;
 690     uint32_t blocks;
 691     size_t block_size = DEFAULT_CLUSTER_SIZE;
 692     uint32_t image_type = VDI_TYPE_DYNAMIC;
 693     VdiHeader header;
 694     size_t i;
 695     size_t bmap_size;
 696     int64_t offset = 0;
 697     Error *local_err = NULL;
 698     BlockDriverState *bs = NULL;
 699     uint32_t *bmap = NULL;
 700
 701     logout("\n");
 702
 703     /* Read out options. */
 704     bytes = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
 705                      BDRV_SECTOR_SIZE);
 706 #if defined(CONFIG_VDI_BLOCK_SIZE)
 707     /* TODO: Additional checks (SECTOR_SIZE * 2^n, ...). */
 708     block_size = qemu_opt_get_size_del(opts,
 709                                        BLOCK_OPT_CLUSTER_SIZE,
 710                                        DEFAULT_CLUSTER_SIZE);
 711 #endif
 712 #if defined(CONFIG_VDI_STATIC_IMAGE)
 713     if (qemu_opt_get_bool_del(opts, BLOCK_OPT_STATIC, false)) {
 714         image_type = VDI_TYPE_STATIC;
 715     }
 716 #endif
 717
 718     if (bytes > VDI_DISK_SIZE_MAX) {
 719         ret = -ENOTSUP;
 720         error_setg(errp, "Unsupported VDI image size (size is 0x%" PRIx64
 721                           ", max supported is 0x%" PRIx64 ")",
 722                           bytes, VDI_DISK_SIZE_MAX);
 723         goto exit;
 724     }
 725
 726     ret = bdrv_create_file(filename, opts, &local_err);
 727     if (ret < 0) {
 728         error_propagate(errp, local_err);
 729         goto exit;
 730     }
 731     ret = bdrv_open(&bs, filename, NULL, NULL, BDRV_O_RDWR | BDRV_O_PROTOCOL,
 732                     NULL, &local_err);
 733     if (ret < 0) {
 734         error_propagate(errp, local_err);
 735         goto exit;
 736     }
 737
 738     /* We need enough blocks to store the given disk size,
 739        so always round up. */
 740     blocks = DIV_ROUND_UP(bytes, block_size);
 741
 742     bmap_size = blocks * sizeof(uint32_t);
 743     bmap_size = ROUND_UP(bmap_size, SECTOR_SIZE);
 744
 745     memset(&header, 0, sizeof(header));
 746     pstrcpy(header.text, sizeof(header.text), VDI_TEXT);
 747     header.signature = VDI_SIGNATURE;
 748     header.version = VDI_VERSION_1_1;
 749     header.header_size = 0x180;
 750     header.image_type = image_type;
 751     header.offset_bmap = 0x200;
 752     header.offset_data = 0x200 + bmap_size;
 753     header.sector_size = SECTOR_SIZE;
 754     header.disk_size = bytes;
 755     header.block_size = block_size;
 756     header.blocks_in_image = blocks;
 757     if (image_type == VDI_TYPE_STATIC) {
 758         header.blocks_allocated = blocks;
 759     }
 760     uuid_generate(header.uuid_image);
 761     uuid_generate(header.uuid_last_snap);
 762     /* There is no need to set header.uuid_link or header.uuid_parent here. */
 763 #if defined(CONFIG_VDI_DEBUG)
 764     vdi_header_print(&header);
 765 #endif
 766     vdi_header_to_le(&header);
 767     ret = bdrv_pwrite_sync(bs, offset, &header, sizeof(header));
 768     if (ret < 0) {
 769         error_setg(errp, "Error writing header to %s", filename);
 770         goto exit;
 771     }
 772     offset += sizeof(header);
 773
 774     if (bmap_size > 0) {
 775         bmap = g_try_malloc0(bmap_size);
 776         if (bmap == NULL) {
 777             ret = -ENOMEM;
 778             error_setg(errp, "Could not allocate bmap");
 779             goto exit;
 780         }
 781         for (i = 0; i < blocks; i++) {
 782             if (image_type == VDI_TYPE_STATIC) {
 783                 bmap[i] = i;
 784             } else {
 785                 bmap[i] = VDI_UNALLOCATED;
 786             }
 787         }
 788         ret = bdrv_pwrite_sync(bs, offset, bmap, bmap_size);
 789         if (ret < 0) {
 790             error_setg(errp, "Error writing bmap to %s", filename);
 791             goto exit;
 792         }
 793         offset += bmap_size;
 794     }
 795
 796     if (image_type == VDI_TYPE_STATIC) {
 797         ret = bdrv_truncate(bs, offset + blocks * block_size);
 798         if (ret < 0) {
 799             error_setg(errp, "Failed to statically allocate %s", filename);
 800             goto exit;
 801         }
 802     }
 803
 804 exit:
 805     bdrv_unref(bs);
 806     g_free(bmap);
 807     return ret;
 808 }
 809
 810 static void vdi_close(BlockDriverState *bs)
 811 {
 812     BDRVVdiState *s = bs->opaque;
 813
 814     qemu_vfree(s->bmap);
 815
 816     migrate_del_blocker(s->migration_blocker);
 817     error_free(s->migration_blocker);
 818 }
 819
 820 static QemuOptsList vdi_create_opts = {
 821     .name = "vdi-create-opts",
 822     .head = QTAILQ_HEAD_INITIALIZER(vdi_create_opts.head),
 823     .desc = {
 824         {
 825             .name = BLOCK_OPT_SIZE,
 826             .type = QEMU_OPT_SIZE,
 827             .help = "Virtual disk size"
 828         },
 829 #if defined(CONFIG_VDI_BLOCK_SIZE)
 830         {
 831             .name = BLOCK_OPT_CLUSTER_SIZE,
 832             .type = QEMU_OPT_SIZE,
 833             .help = "VDI cluster (block) size",
 834             .def_value_str = stringify(DEFAULT_CLUSTER_SIZE)
 835         },
 836 #endif
 837 #if defined(CONFIG_VDI_STATIC_IMAGE)
 838         {
 839             .name = BLOCK_OPT_STATIC,
 840             .type = QEMU_OPT_BOOL,
 841             .help = "VDI static (pre-allocated) image",
 842             .def_value_str = "off"
 843         },
 844 #endif
 845         {
 846             .name = BLOCK_OPT_NOCOW,
 847             .type = QEMU_OPT_BOOL,
 848             .help = "Turn off copy-on-write (valid only on btrfs)"
 849         },
 850         /* TODO: An additional option to set UUID values might be useful. */
 851         { /* end of list */ }
 852     }
 853 };
 854
 855 static BlockDriver bdrv_vdi = {
 856     .format_name = "vdi",
 857     .instance_size = sizeof(BDRVVdiState),
 858     .bdrv_probe = vdi_probe,
 859     .bdrv_open = vdi_open,
 860     .bdrv_close = vdi_close,
 861     .bdrv_reopen_prepare = vdi_reopen_prepare,
 862     .bdrv_create = vdi_create,
 863     .bdrv_has_zero_init = bdrv_has_zero_init_1,
 864     .bdrv_co_get_block_status = vdi_co_get_block_status,
 865     .bdrv_make_empty = vdi_make_empty,
 866
 867     .bdrv_read = vdi_co_read,
 868 #if defined(CONFIG_VDI_WRITE)
 869     .bdrv_write = vdi_co_write,
 870 #endif
 871
 872     .bdrv_get_info = vdi_get_info,
 873
 874     .create_opts = &vdi_create_opts,
 875     .bdrv_check = vdi_check,
 876 };
 877
 878 static void bdrv_vdi_init(void)
 879 {
 880     logout("\n");
 881     bdrv_register(&bdrv_vdi);
 882 }
 883
 884 block_init(bdrv_vdi_init);