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 #ifdef __linux__
  57 #include <linux/fs.h>
  58 #include <sys/ioctl.h>
  59 #ifndef FS_NOCOW_FL
  60 #define FS_NOCOW_FL                     0x00800000 /* Do not cow file */
  61 #endif
  62 #endif
  63
  64 #if defined(CONFIG_UUID)
  65 #include <uuid/uuid.h>
  66 #else
  67 /* TODO: move uuid emulation to some central place in QEMU. */
  68 #include "sysemu/sysemu.h"     /* UUID_FMT */
  69 typedef unsigned char uuid_t[16];
  70 #endif
  71
  72 /* Code configuration options. */
  73
  74 /* Enable debug messages. */
  75 //~ #define CONFIG_VDI_DEBUG
  76
  77 /* Support write operations on VDI images. */
  78 #define CONFIG_VDI_WRITE
  79
  80 /* Support non-standard block (cluster) size. This is untested.
  81  * Maybe it will be needed for very large images.
  82  */
  83 //~ #define CONFIG_VDI_BLOCK_SIZE
  84
  85 /* Support static (fixed, pre-allocated) images. */
  86 #define CONFIG_VDI_STATIC_IMAGE
  87
  88 /* Command line option for static images. */
  89 #define BLOCK_OPT_STATIC "static"
  90
  91 #define KiB     1024
  92 #define MiB     (KiB * KiB)
  93
  94 #define SECTOR_SIZE 512
  95 #define DEFAULT_CLUSTER_SIZE (1 * MiB)
  96
  97 #if defined(CONFIG_VDI_DEBUG)
  98 #define logout(fmt, ...) \
  99                 fprintf(stderr, "vdi\t%-24s" fmt, __func__, ##__VA_ARGS__)
 100 #else
 101 #define logout(fmt, ...) ((void)0)
 102 #endif
 103
 104 /* Image signature. */
 105 #define VDI_SIGNATURE 0xbeda107f
 106
 107 /* Image version. */
 108 #define VDI_VERSION_1_1 0x00010001
 109
 110 /* Image type. */
 111 #define VDI_TYPE_DYNAMIC 1
 112 #define VDI_TYPE_STATIC  2
 113
 114 /* Innotek / SUN images use these strings in header.text:
 115  * "<<< innotek VirtualBox Disk Image >>>\n"
 116  * "<<< Sun xVM VirtualBox Disk Image >>>\n"
 117  * "<<< Sun VirtualBox Disk Image >>>\n"
 118  * The value does not matter, so QEMU created images use a different text.
 119  */
 120 #define VDI_TEXT "<<< QEMU VM Virtual Disk Image >>>\n"
 121
 122 /* A never-allocated block; semantically arbitrary content. */
 123 #define VDI_UNALLOCATED 0xffffffffU
 124
 125 /* A discarded (no longer allocated) block; semantically zero-filled. */
 126 #define VDI_DISCARDED   0xfffffffeU
 127
 128 #define VDI_IS_ALLOCATED(X) ((X) < VDI_DISCARDED)
 129
 130 /* max blocks in image is (0xffffffff / 4) */
 131 #define VDI_BLOCKS_IN_IMAGE_MAX  0x3fffffff
 132 #define VDI_DISK_SIZE_MAX        ((uint64_t)VDI_BLOCKS_IN_IMAGE_MAX * \
 133                                   (uint64_t)DEFAULT_CLUSTER_SIZE)
 134
 135 #if !defined(CONFIG_UUID)
 136 static inline void uuid_generate(uuid_t out)
 137 {
 138     memset(out, 0, sizeof(uuid_t));
 139 }
 140
 141 static inline int uuid_is_null(const uuid_t uu)
 142 {
 143     uuid_t null_uuid = { 0 };
 144     return memcmp(uu, null_uuid, sizeof(uuid_t)) == 0;
 145 }
 146
 147 static inline void uuid_unparse(const uuid_t uu, char *out)
 148 {
 149     snprintf(out, 37, UUID_FMT,
 150             uu[0], uu[1], uu[2], uu[3], uu[4], uu[5], uu[6], uu[7],
 151             uu[8], uu[9], uu[10], uu[11], uu[12], uu[13], uu[14], uu[15]);
 152 }
 153 #endif
 154
 155 typedef struct {
 156     char text[0x40];
 157     uint32_t signature;
 158     uint32_t version;
 159     uint32_t header_size;
 160     uint32_t image_type;
 161     uint32_t image_flags;
 162     char description[256];
 163     uint32_t offset_bmap;
 164     uint32_t offset_data;
 165     uint32_t cylinders;         /* disk geometry, unused here */
 166     uint32_t heads;             /* disk geometry, unused here */
 167     uint32_t sectors;           /* disk geometry, unused here */
 168     uint32_t sector_size;
 169     uint32_t unused1;
 170     uint64_t disk_size;
 171     uint32_t block_size;
 172     uint32_t block_extra;       /* unused here */
 173     uint32_t blocks_in_image;
 174     uint32_t blocks_allocated;
 175     uuid_t uuid_image;
 176     uuid_t uuid_last_snap;
 177     uuid_t uuid_link;
 178     uuid_t uuid_parent;
 179     uint64_t unused2[7];
 180 } QEMU_PACKED VdiHeader;
 181
 182 typedef struct {
 183     /* The block map entries are little endian (even in memory). */
 184     uint32_t *bmap;
 185     /* Size of block (bytes). */
 186     uint32_t block_size;
 187     /* Size of block (sectors). */
 188     uint32_t block_sectors;
 189     /* First sector of block map. */
 190     uint32_t bmap_sector;
 191     /* VDI header (converted to host endianness). */
 192     VdiHeader header;
 193
 194     Error *migration_blocker;
 195 } BDRVVdiState;
 196
 197 /* Change UUID from little endian (IPRT = VirtualBox format) to big endian
 198  * format (network byte order, standard, see RFC 4122) and vice versa.
 199  */
 200 static void uuid_convert(uuid_t uuid)
 201 {
 202     bswap32s((uint32_t *)&uuid[0]);
 203     bswap16s((uint16_t *)&uuid[4]);
 204     bswap16s((uint16_t *)&uuid[6]);
 205 }
 206
 207 static void vdi_header_to_cpu(VdiHeader *header)
 208 {
 209     le32_to_cpus(&header->signature);
 210     le32_to_cpus(&header->version);
 211     le32_to_cpus(&header->header_size);
 212     le32_to_cpus(&header->image_type);
 213     le32_to_cpus(&header->image_flags);
 214     le32_to_cpus(&header->offset_bmap);
 215     le32_to_cpus(&header->offset_data);
 216     le32_to_cpus(&header->cylinders);
 217     le32_to_cpus(&header->heads);
 218     le32_to_cpus(&header->sectors);
 219     le32_to_cpus(&header->sector_size);
 220     le64_to_cpus(&header->disk_size);
 221     le32_to_cpus(&header->block_size);
 222     le32_to_cpus(&header->block_extra);
 223     le32_to_cpus(&header->blocks_in_image);
 224     le32_to_cpus(&header->blocks_allocated);
 225     uuid_convert(header->uuid_image);
 226     uuid_convert(header->uuid_last_snap);
 227     uuid_convert(header->uuid_link);
 228     uuid_convert(header->uuid_parent);
 229 }
 230
 231 static void vdi_header_to_le(VdiHeader *header)
 232 {
 233     cpu_to_le32s(&header->signature);
 234     cpu_to_le32s(&header->version);
 235     cpu_to_le32s(&header->header_size);
 236     cpu_to_le32s(&header->image_type);
 237     cpu_to_le32s(&header->image_flags);
 238     cpu_to_le32s(&header->offset_bmap);
 239     cpu_to_le32s(&header->offset_data);
 240     cpu_to_le32s(&header->cylinders);
 241     cpu_to_le32s(&header->heads);
 242     cpu_to_le32s(&header->sectors);
 243     cpu_to_le32s(&header->sector_size);
 244     cpu_to_le64s(&header->disk_size);
 245     cpu_to_le32s(&header->block_size);
 246     cpu_to_le32s(&header->block_extra);
 247     cpu_to_le32s(&header->blocks_in_image);
 248     cpu_to_le32s(&header->blocks_allocated);
 249     uuid_convert(header->uuid_image);
 250     uuid_convert(header->uuid_last_snap);
 251     uuid_convert(header->uuid_link);
 252     uuid_convert(header->uuid_parent);
 253 }
 254
 255 #if defined(CONFIG_VDI_DEBUG)
 256 static void vdi_header_print(VdiHeader *header)
 257 {
 258     char uuid[37];
 259     logout("text        %s", header->text);
 260     logout("signature   0x%08x\n", header->signature);
 261     logout("header size 0x%04x\n", header->header_size);
 262     logout("image type  0x%04x\n", header->image_type);
 263     logout("image flags 0x%04x\n", header->image_flags);
 264     logout("description %s\n", header->description);
 265     logout("offset bmap 0x%04x\n", header->offset_bmap);
 266     logout("offset data 0x%04x\n", header->offset_data);
 267     logout("cylinders   0x%04x\n", header->cylinders);
 268     logout("heads       0x%04x\n", header->heads);
 269     logout("sectors     0x%04x\n", header->sectors);
 270     logout("sector size 0x%04x\n", header->sector_size);
 271     logout("image size  0x%" PRIx64 " B (%" PRIu64 " MiB)\n",
 272            header->disk_size, header->disk_size / MiB);
 273     logout("block size  0x%04x\n", header->block_size);
 274     logout("block extra 0x%04x\n", header->block_extra);
 275     logout("blocks tot. 0x%04x\n", header->blocks_in_image);
 276     logout("blocks all. 0x%04x\n", header->blocks_allocated);
 277     uuid_unparse(header->uuid_image, uuid);
 278     logout("uuid image  %s\n", uuid);
 279     uuid_unparse(header->uuid_last_snap, uuid);
 280     logout("uuid snap   %s\n", uuid);
 281     uuid_unparse(header->uuid_link, uuid);
 282     logout("uuid link   %s\n", uuid);
 283     uuid_unparse(header->uuid_parent, uuid);
 284     logout("uuid parent %s\n", uuid);
 285 }
 286 #endif
 287
 288 static int vdi_check(BlockDriverState *bs, BdrvCheckResult *res,
 289                      BdrvCheckMode fix)
 290 {
 291     /* TODO: additional checks possible. */
 292     BDRVVdiState *s = (BDRVVdiState *)bs->opaque;
 293     uint32_t blocks_allocated = 0;
 294     uint32_t block;
 295     uint32_t *bmap;
 296     logout("\n");
 297
 298     if (fix) {
 299         return -ENOTSUP;
 300     }
 301
 302     bmap = g_malloc(s->header.blocks_in_image * sizeof(uint32_t));
 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 result = 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         result = 100;
 368     }
 369
 370     if (result == 0) {
 371         logout("no vdi image\n");
 372     } else {
 373         logout("%s", header->text);
 374     }
 375
 376     return result;
 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 += SECTOR_SIZE - 1;
 413         header.disk_size &= ~(SECTOR_SIZE - 1);
 414     }
 415
 416     if (header.signature != VDI_SIGNATURE) {
 417         error_setg(errp, "Image not in VDI format (bad signature %08" PRIx32
 418                    ")", header.signature);
 419         ret = -EINVAL;
 420         goto fail;
 421     } else if (header.version != VDI_VERSION_1_1) {
 422         error_setg(errp, "unsupported VDI image (version %" PRIu32 ".%" PRIu32
 423                    ")", header.version >> 16, header.version & 0xffff);
 424         ret = -ENOTSUP;
 425         goto fail;
 426     } else if (header.offset_bmap % SECTOR_SIZE != 0) {
 427         /* We only support block maps which start on a sector boundary. */
 428         error_setg(errp, "unsupported VDI image (unaligned block map offset "
 429                    "0x%" PRIx32 ")", header.offset_bmap);
 430         ret = -ENOTSUP;
 431         goto fail;
 432     } else if (header.offset_data % SECTOR_SIZE != 0) {
 433         /* We only support data blocks which start on a sector boundary. */
 434         error_setg(errp, "unsupported VDI image (unaligned data offset 0x%"
 435                    PRIx32 ")", header.offset_data);
 436         ret = -ENOTSUP;
 437         goto fail;
 438     } else if (header.sector_size != SECTOR_SIZE) {
 439         error_setg(errp, "unsupported VDI image (sector size %" PRIu32
 440                    " is not %u)", header.sector_size, SECTOR_SIZE);
 441         ret = -ENOTSUP;
 442         goto fail;
 443     } else if (header.block_size != DEFAULT_CLUSTER_SIZE) {
 444         error_setg(errp, "unsupported VDI image (block size %" PRIu32
 445                    " is not %u)", header.block_size, DEFAULT_CLUSTER_SIZE);
 446         ret = -ENOTSUP;
 447         goto fail;
 448     } else if (header.disk_size >
 449                (uint64_t)header.blocks_in_image * header.block_size) {
 450         error_setg(errp, "unsupported VDI image (disk size %" PRIu64 ", "
 451                    "image bitmap has room for %" PRIu64 ")",
 452                    header.disk_size,
 453                    (uint64_t)header.blocks_in_image * header.block_size);
 454         ret = -ENOTSUP;
 455         goto fail;
 456     } else if (!uuid_is_null(header.uuid_link)) {
 457         error_setg(errp, "unsupported VDI image (non-NULL link UUID)");
 458         ret = -ENOTSUP;
 459         goto fail;
 460     } else if (!uuid_is_null(header.uuid_parent)) {
 461         error_setg(errp, "unsupported VDI image (non-NULL parent UUID)");
 462         ret = -ENOTSUP;
 463         goto fail;
 464     } else if (header.blocks_in_image > VDI_BLOCKS_IN_IMAGE_MAX) {
 465         error_setg(errp, "unsupported VDI image "
 466                          "(too many blocks %u, max is %u)",
 467                           header.blocks_in_image, VDI_BLOCKS_IN_IMAGE_MAX);
 468         ret = -ENOTSUP;
 469         goto fail;
 470     }
 471
 472     bs->total_sectors = header.disk_size / SECTOR_SIZE;
 473
 474     s->block_size = header.block_size;
 475     s->block_sectors = header.block_size / SECTOR_SIZE;
 476     s->bmap_sector = header.offset_bmap / SECTOR_SIZE;
 477     s->header = header;
 478
 479     bmap_size = header.blocks_in_image * sizeof(uint32_t);
 480     bmap_size = (bmap_size + SECTOR_SIZE - 1) / SECTOR_SIZE;
 481     s->bmap = g_malloc(bmap_size * SECTOR_SIZE);
 482     ret = bdrv_read(bs->file, s->bmap_sector, (uint8_t *)s->bmap, bmap_size);
 483     if (ret < 0) {
 484         goto fail_free_bmap;
 485     }
 486
 487     /* Disable migration when vdi images are used */
 488     error_set(&s->migration_blocker,
 489               QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED,
 490               "vdi", bs->device_name, "live migration");
 491     migrate_add_blocker(s->migration_blocker);
 492
 493     return 0;
 494
 495  fail_free_bmap:
 496     g_free(s->bmap);
 497
 498  fail:
 499     return ret;
 500 }
 501
 502 static int vdi_reopen_prepare(BDRVReopenState *state,
 503                               BlockReopenQueue *queue, Error **errp)
 504 {
 505     return 0;
 506 }
 507
 508 static int64_t coroutine_fn vdi_co_get_block_status(BlockDriverState *bs,
 509         int64_t sector_num, int nb_sectors, int *pnum)
 510 {
 511     /* TODO: Check for too large sector_num (in bdrv_is_allocated or here). */
 512     BDRVVdiState *s = (BDRVVdiState *)bs->opaque;
 513     size_t bmap_index = sector_num / s->block_sectors;
 514     size_t sector_in_block = sector_num % s->block_sectors;
 515     int n_sectors = s->block_sectors - sector_in_block;
 516     uint32_t bmap_entry = le32_to_cpu(s->bmap[bmap_index]);
 517     uint64_t offset;
 518     int result;
 519
 520     logout("%p, %" PRId64 ", %d, %p\n", bs, sector_num, nb_sectors, pnum);
 521     if (n_sectors > nb_sectors) {
 522         n_sectors = nb_sectors;
 523     }
 524     *pnum = n_sectors;
 525     result = VDI_IS_ALLOCATED(bmap_entry);
 526     if (!result) {
 527         return 0;
 528     }
 529
 530     offset = s->header.offset_data +
 531                               (uint64_t)bmap_entry * s->block_size +
 532                               sector_in_block * SECTOR_SIZE;
 533     return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID | offset;
 534 }
 535
 536 static int vdi_co_read(BlockDriverState *bs,
 537         int64_t sector_num, uint8_t *buf, int nb_sectors)
 538 {
 539     BDRVVdiState *s = bs->opaque;
 540     uint32_t bmap_entry;
 541     uint32_t block_index;
 542     uint32_t sector_in_block;
 543     uint32_t n_sectors;
 544     int ret = 0;
 545
 546     logout("\n");
 547
 548     while (ret >= 0 && nb_sectors > 0) {
 549         block_index = sector_num / s->block_sectors;
 550         sector_in_block = sector_num % s->block_sectors;
 551         n_sectors = s->block_sectors - sector_in_block;
 552         if (n_sectors > nb_sectors) {
 553             n_sectors = nb_sectors;
 554         }
 555
 556         logout("will read %u sectors starting at sector %" PRIu64 "\n",
 557                n_sectors, sector_num);
 558
 559         /* prepare next AIO request */
 560         bmap_entry = le32_to_cpu(s->bmap[block_index]);
 561         if (!VDI_IS_ALLOCATED(bmap_entry)) {
 562             /* Block not allocated, return zeros, no need to wait. */
 563             memset(buf, 0, n_sectors * SECTOR_SIZE);
 564             ret = 0;
 565         } else {
 566             uint64_t offset = s->header.offset_data / SECTOR_SIZE +
 567                               (uint64_t)bmap_entry * s->block_sectors +
 568                               sector_in_block;
 569             ret = bdrv_read(bs->file, offset, buf, n_sectors);
 570         }
 571         logout("%u sectors read\n", n_sectors);
 572
 573         nb_sectors -= n_sectors;
 574         sector_num += n_sectors;
 575         buf += n_sectors * SECTOR_SIZE;
 576     }
 577
 578     return ret;
 579 }
 580
 581 static int vdi_co_write(BlockDriverState *bs,
 582         int64_t sector_num, const uint8_t *buf, int nb_sectors)
 583 {
 584     BDRVVdiState *s = bs->opaque;
 585     uint32_t bmap_entry;
 586     uint32_t block_index;
 587     uint32_t sector_in_block;
 588     uint32_t n_sectors;
 589     uint32_t bmap_first = VDI_UNALLOCATED;
 590     uint32_t bmap_last = VDI_UNALLOCATED;
 591     uint8_t *block = NULL;
 592     int ret = 0;
 593
 594     logout("\n");
 595
 596     while (ret >= 0 && nb_sectors > 0) {
 597         block_index = sector_num / s->block_sectors;
 598         sector_in_block = sector_num % s->block_sectors;
 599         n_sectors = s->block_sectors - sector_in_block;
 600         if (n_sectors > nb_sectors) {
 601             n_sectors = nb_sectors;
 602         }
 603
 604         logout("will write %u sectors starting at sector %" PRIu64 "\n",
 605                n_sectors, sector_num);
 606
 607         /* prepare next AIO request */
 608         bmap_entry = le32_to_cpu(s->bmap[block_index]);
 609         if (!VDI_IS_ALLOCATED(bmap_entry)) {
 610             /* Allocate new block and write to it. */
 611             uint64_t offset;
 612             bmap_entry = s->header.blocks_allocated;
 613             s->bmap[block_index] = cpu_to_le32(bmap_entry);
 614             s->header.blocks_allocated++;
 615             offset = s->header.offset_data / SECTOR_SIZE +
 616                      (uint64_t)bmap_entry * s->block_sectors;
 617             if (block == NULL) {
 618                 block = g_malloc(s->block_size);
 619                 bmap_first = block_index;
 620             }
 621             bmap_last = block_index;
 622             /* Copy data to be written to new block and zero unused parts. */
 623             memset(block, 0, sector_in_block * SECTOR_SIZE);
 624             memcpy(block + sector_in_block * SECTOR_SIZE,
 625                    buf, n_sectors * SECTOR_SIZE);
 626             memset(block + (sector_in_block + n_sectors) * SECTOR_SIZE, 0,
 627                    (s->block_sectors - n_sectors - sector_in_block) * SECTOR_SIZE);
 628             ret = bdrv_write(bs->file, offset, block, s->block_sectors);
 629         } else {
 630             uint64_t offset = s->header.offset_data / SECTOR_SIZE +
 631                               (uint64_t)bmap_entry * s->block_sectors +
 632                               sector_in_block;
 633             ret = bdrv_write(bs->file, offset, buf, n_sectors);
 634         }
 635
 636         nb_sectors -= n_sectors;
 637         sector_num += n_sectors;
 638         buf += n_sectors * SECTOR_SIZE;
 639
 640         logout("%u sectors written\n", n_sectors);
 641     }
 642
 643     logout("finished data write\n");
 644     if (ret < 0) {
 645         return ret;
 646     }
 647
 648     if (block) {
 649         /* One or more new blocks were allocated. */
 650         VdiHeader *header = (VdiHeader *) block;
 651         uint8_t *base;
 652         uint64_t offset;
 653
 654         logout("now writing modified header\n");
 655         assert(VDI_IS_ALLOCATED(bmap_first));
 656         *header = s->header;
 657         vdi_header_to_le(header);
 658         ret = bdrv_write(bs->file, 0, block, 1);
 659         g_free(block);
 660         block = NULL;
 661
 662         if (ret < 0) {
 663             return ret;
 664         }
 665
 666         logout("now writing modified block map entry %u...%u\n",
 667                bmap_first, bmap_last);
 668         /* Write modified sectors from block map. */
 669         bmap_first /= (SECTOR_SIZE / sizeof(uint32_t));
 670         bmap_last /= (SECTOR_SIZE / sizeof(uint32_t));
 671         n_sectors = bmap_last - bmap_first + 1;
 672         offset = s->bmap_sector + bmap_first;
 673         base = ((uint8_t *)&s->bmap[0]) + bmap_first * SECTOR_SIZE;
 674         logout("will write %u block map sectors starting from entry %u\n",
 675                n_sectors, bmap_first);
 676         ret = bdrv_write(bs->file, offset, base, n_sectors);
 677     }
 678
 679     return ret;
 680 }
 681
 682 static int vdi_create(const char *filename, QemuOpts *opts, Error **errp)
 683 {
 684     int fd;
 685     int result = 0;
 686     uint64_t bytes = 0;
 687     uint32_t blocks;
 688     size_t block_size = DEFAULT_CLUSTER_SIZE;
 689     uint32_t image_type = VDI_TYPE_DYNAMIC;
 690     VdiHeader header;
 691     size_t i;
 692     size_t bmap_size;
 693     bool nocow = false;
 694
 695     logout("\n");
 696
 697     /* Read out options. */
 698     bytes = qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0);
 699 #if defined(CONFIG_VDI_BLOCK_SIZE)
 700     /* TODO: Additional checks (SECTOR_SIZE * 2^n, ...). */
 701     block_size = qemu_opt_get_size_del(opts,
 702                                        BLOCK_OPT_CLUSTER_SIZE,
 703                                        DEFAULT_CLUSTER_SIZE);
 704 #endif
 705 #if defined(CONFIG_VDI_STATIC_IMAGE)
 706     if (qemu_opt_get_bool_del(opts, BLOCK_OPT_STATIC, false)) {
 707         image_type = VDI_TYPE_STATIC;
 708     }
 709 #endif
 710     nocow = qemu_opt_get_bool_del(opts, BLOCK_OPT_NOCOW, false);
 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     if (nocow) {
 729 #ifdef __linux__
 730         /* Set NOCOW flag to solve performance issue on fs like btrfs.
 731          * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value will
 732          * be ignored since any failure of this operation should not block the
 733          * left work.
 734          */
 735         int attr;
 736         if (ioctl(fd, FS_IOC_GETFLAGS, &attr) == 0) {
 737             attr |= FS_NOCOW_FL;
 738             ioctl(fd, FS_IOC_SETFLAGS, &attr);
 739         }
 740 #endif
 741     }
 742
 743     /* We need enough blocks to store the given disk size,
 744        so always round up. */
 745     blocks = (bytes + block_size - 1) / block_size;
 746
 747     bmap_size = blocks * sizeof(uint32_t);
 748     bmap_size = ((bmap_size + SECTOR_SIZE - 1) & ~(SECTOR_SIZE -1));
 749
 750     memset(&header, 0, sizeof(header));
 751     pstrcpy(header.text, sizeof(header.text), VDI_TEXT);
 752     header.signature = VDI_SIGNATURE;
 753     header.version = VDI_VERSION_1_1;
 754     header.header_size = 0x180;
 755     header.image_type = image_type;
 756     header.offset_bmap = 0x200;
 757     header.offset_data = 0x200 + bmap_size;
 758     header.sector_size = SECTOR_SIZE;
 759     header.disk_size = bytes;
 760     header.block_size = block_size;
 761     header.blocks_in_image = blocks;
 762     if (image_type == VDI_TYPE_STATIC) {
 763         header.blocks_allocated = blocks;
 764     }
 765     uuid_generate(header.uuid_image);
 766     uuid_generate(header.uuid_last_snap);
 767     /* There is no need to set header.uuid_link or header.uuid_parent here. */
 768 #if defined(CONFIG_VDI_DEBUG)
 769     vdi_header_print(&header);
 770 #endif
 771     vdi_header_to_le(&header);
 772     if (write(fd, &header, sizeof(header)) < 0) {
 773         result = -errno;
 774         goto close_and_exit;
 775     }
 776
 777     if (bmap_size > 0) {
 778         uint32_t *bmap = g_malloc0(bmap_size);
 779         for (i = 0; i < blocks; i++) {
 780             if (image_type == VDI_TYPE_STATIC) {
 781                 bmap[i] = i;
 782             } else {
 783                 bmap[i] = VDI_UNALLOCATED;
 784             }
 785         }
 786         if (write(fd, bmap, bmap_size) < 0) {
 787             result = -errno;
 788             g_free(bmap);
 789             goto close_and_exit;
 790         }
 791         g_free(bmap);
 792     }
 793
 794     if (image_type == VDI_TYPE_STATIC) {
 795         if (ftruncate(fd, sizeof(header) + bmap_size + blocks * block_size)) {
 796             result = -errno;
 797             goto close_and_exit;
 798         }
 799     }
 800
 801 close_and_exit:
 802     if ((close(fd) < 0) && !result) {
 803         result = -errno;
 804     }
 805
 806 exit:
 807     return result;
 808 }
 809
 810 static void vdi_close(BlockDriverState *bs)
 811 {
 812     BDRVVdiState *s = bs->opaque;
 813
 814     g_free(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);