block/vmdk.c

   1 /*
   2  * Block driver for the VMDK format
   3  *
   4  * Copyright (c) 2004 Fabrice Bellard
   5  * Copyright (c) 2005 Filip Navara
   6  *
   7  * Permission is hereby granted, free of charge, to any person obtaining a copy
   8  * of this software and associated documentation files (the "Software"), to deal
   9  * in the Software without restriction, including without limitation the rights
  10  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
  11  * copies of the Software, and to permit persons to whom the Software is
  12  * furnished to do so, subject to the following conditions:
  13  *
  14  * The above copyright notice and this permission notice shall be included in
  15  * all copies or substantial portions of the Software.
  16  *
  17  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  18  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  19  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
  20  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  21  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  22  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  23  * THE SOFTWARE.
  24  */
  25
  26 #include "qemu-common.h"
  27 #include "block_int.h"
  28 #include "module.h"
  29
  30 #define VMDK3_MAGIC (('C' << 24) | ('O' << 16) | ('W' << 8) | 'D')
  31 #define VMDK4_MAGIC (('K' << 24) | ('D' << 16) | ('M' << 8) | 'V')
  32
  33 typedef struct {
  34     uint32_t version;
  35     uint32_t flags;
  36     uint32_t disk_sectors;
  37     uint32_t granularity;
  38     uint32_t l1dir_offset;
  39     uint32_t l1dir_size;
  40     uint32_t file_sectors;
  41     uint32_t cylinders;
  42     uint32_t heads;
  43     uint32_t sectors_per_track;
  44 } VMDK3Header;
  45
  46 typedef struct {
  47     uint32_t version;
  48     uint32_t flags;
  49     int64_t capacity;
  50     int64_t granularity;
  51     int64_t desc_offset;
  52     int64_t desc_size;
  53     int32_t num_gtes_per_gte;
  54     int64_t rgd_offset;
  55     int64_t gd_offset;
  56     int64_t grain_offset;
  57     char filler[1];
  58     char check_bytes[4];
  59 } QEMU_PACKED VMDK4Header;
  60
  61 #define L2_CACHE_SIZE 16
  62
  63 typedef struct VmdkExtent {
  64     BlockDriverState *file;
  65     bool flat;
  66     int64_t sectors;
  67     int64_t end_sector;
  68     int64_t flat_start_offset;
  69     int64_t l1_table_offset;
  70     int64_t l1_backup_table_offset;
  71     uint32_t *l1_table;
  72     uint32_t *l1_backup_table;
  73     unsigned int l1_size;
  74     uint32_t l1_entry_sectors;
  75
  76     unsigned int l2_size;
  77     uint32_t *l2_cache;
  78     uint32_t l2_cache_offsets[L2_CACHE_SIZE];
  79     uint32_t l2_cache_counts[L2_CACHE_SIZE];
  80
  81     unsigned int cluster_sectors;
  82 } VmdkExtent;
  83
  84 typedef struct BDRVVmdkState {
  85     int desc_offset;
  86     bool cid_updated;
  87     uint32_t parent_cid;
  88     int num_extents;
  89     /* Extent array with num_extents entries, ascend ordered by address */
  90     VmdkExtent *extents;
  91 } BDRVVmdkState;
  92
  93 typedef struct VmdkMetaData {
  94     uint32_t offset;
  95     unsigned int l1_index;
  96     unsigned int l2_index;
  97     unsigned int l2_offset;
  98     int valid;
  99 } VmdkMetaData;
 100
 101 static int vmdk_probe(const uint8_t *buf, int buf_size, const char *filename)
 102 {
 103     uint32_t magic;
 104
 105     if (buf_size < 4) {
 106         return 0;
 107     }
 108     magic = be32_to_cpu(*(uint32_t *)buf);
 109     if (magic == VMDK3_MAGIC ||
 110         magic == VMDK4_MAGIC) {
 111         return 100;
 112     } else {
 113         const char *p = (const char *)buf;
 114         const char *end = p + buf_size;
 115         while (p < end) {
 116             if (*p == '#') {
 117                 /* skip comment line */
 118                 while (p < end && *p != '\n') {
 119                     p++;
 120                 }
 121                 p++;
 122                 continue;
 123             }
 124             if (*p == ' ') {
 125                 while (p < end && *p == ' ') {
 126                     p++;
 127                 }
 128                 /* skip '\r' if windows line endings used. */
 129                 if (p < end && *p == '\r') {
 130                     p++;
 131                 }
 132                 /* only accept blank lines before 'version=' line */
 133                 if (p == end || *p != '\n') {
 134                     return 0;
 135                 }
 136                 p++;
 137                 continue;
 138             }
 139             if (end - p >= strlen("version=X\n")) {
 140                 if (strncmp("version=1\n", p, strlen("version=1\n")) == 0 ||
 141                     strncmp("version=2\n", p, strlen("version=2\n")) == 0) {
 142                     return 100;
 143                 }
 144             }
 145             if (end - p >= strlen("version=X\r\n")) {
 146                 if (strncmp("version=1\r\n", p, strlen("version=1\r\n")) == 0 ||
 147                     strncmp("version=2\r\n", p, strlen("version=2\r\n")) == 0) {
 148                     return 100;
 149                 }
 150             }
 151             return 0;
 152         }
 153         return 0;
 154     }
 155 }
 156
 157 #define CHECK_CID 1
 158
 159 #define SECTOR_SIZE 512
 160 #define DESC_SIZE (20 * SECTOR_SIZE)    /* 20 sectors of 512 bytes each */
 161 #define BUF_SIZE 4096
 162 #define HEADER_SIZE 512                 /* first sector of 512 bytes */
 163
 164 static void vmdk_free_extents(BlockDriverState *bs)
 165 {
 166     int i;
 167     BDRVVmdkState *s = bs->opaque;
 168
 169     for (i = 0; i < s->num_extents; i++) {
 170         g_free(s->extents[i].l1_table);
 171         g_free(s->extents[i].l2_cache);
 172         g_free(s->extents[i].l1_backup_table);
 173     }
 174     g_free(s->extents);
 175 }
 176
 177 static void vmdk_free_last_extent(BlockDriverState *bs)
 178 {
 179     BDRVVmdkState *s = bs->opaque;
 180
 181     if (s->num_extents == 0) {
 182         return;
 183     }
 184     s->num_extents--;
 185     s->extents = g_realloc(s->extents, s->num_extents * sizeof(VmdkExtent));
 186 }
 187
 188 static uint32_t vmdk_read_cid(BlockDriverState *bs, int parent)
 189 {
 190     char desc[DESC_SIZE];
 191     uint32_t cid;
 192     const char *p_name, *cid_str;
 193     size_t cid_str_size;
 194     BDRVVmdkState *s = bs->opaque;
 195
 196     if (bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE) != DESC_SIZE) {
 197         return 0;
 198     }
 199
 200     if (parent) {
 201         cid_str = "parentCID";
 202         cid_str_size = sizeof("parentCID");
 203     } else {
 204         cid_str = "CID";
 205         cid_str_size = sizeof("CID");
 206     }
 207
 208     p_name = strstr(desc, cid_str);
 209     if (p_name != NULL) {
 210         p_name += cid_str_size;
 211         sscanf(p_name, "%x", &cid);
 212     }
 213
 214     return cid;
 215 }
 216
 217 static int vmdk_write_cid(BlockDriverState *bs, uint32_t cid)
 218 {
 219     char desc[DESC_SIZE], tmp_desc[DESC_SIZE];
 220     char *p_name, *tmp_str;
 221     BDRVVmdkState *s = bs->opaque;
 222
 223     memset(desc, 0, sizeof(desc));
 224     if (bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE) != DESC_SIZE) {
 225         return -EIO;
 226     }
 227
 228     tmp_str = strstr(desc, "parentCID");
 229     pstrcpy(tmp_desc, sizeof(tmp_desc), tmp_str);
 230     p_name = strstr(desc, "CID");
 231     if (p_name != NULL) {
 232         p_name += sizeof("CID");
 233         snprintf(p_name, sizeof(desc) - (p_name - desc), "%x\n", cid);
 234         pstrcat(desc, sizeof(desc), tmp_desc);
 235     }
 236
 237     if (bdrv_pwrite_sync(bs->file, s->desc_offset, desc, DESC_SIZE) < 0) {
 238         return -EIO;
 239     }
 240     return 0;
 241 }
 242
 243 static int vmdk_is_cid_valid(BlockDriverState *bs)
 244 {
 245 #ifdef CHECK_CID
 246     BDRVVmdkState *s = bs->opaque;
 247     BlockDriverState *p_bs = bs->backing_hd;
 248     uint32_t cur_pcid;
 249
 250     if (p_bs) {
 251         cur_pcid = vmdk_read_cid(p_bs, 0);
 252         if (s->parent_cid != cur_pcid) {
 253             /* CID not valid */
 254             return 0;
 255         }
 256     }
 257 #endif
 258     /* CID valid */
 259     return 1;
 260 }
 261
 262 static int vmdk_parent_open(BlockDriverState *bs)
 263 {
 264     char *p_name;
 265     char desc[DESC_SIZE + 1];
 266     BDRVVmdkState *s = bs->opaque;
 267
 268     desc[DESC_SIZE] = '\0';
 269     if (bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE) != DESC_SIZE) {
 270         return -1;
 271     }
 272
 273     p_name = strstr(desc, "parentFileNameHint");
 274     if (p_name != NULL) {
 275         char *end_name;
 276
 277         p_name += sizeof("parentFileNameHint") + 1;
 278         end_name = strchr(p_name, '\"');
 279         if (end_name == NULL) {
 280             return -1;
 281         }
 282         if ((end_name - p_name) > sizeof(bs->backing_file) - 1) {
 283             return -1;
 284         }
 285
 286         pstrcpy(bs->backing_file, end_name - p_name + 1, p_name);
 287     }
 288
 289     return 0;
 290 }
 291
 292 /* Create and append extent to the extent array. Return the added VmdkExtent
 293  * address. return NULL if allocation failed. */
 294 static VmdkExtent *vmdk_add_extent(BlockDriverState *bs,
 295                            BlockDriverState *file, bool flat, int64_t sectors,
 296                            int64_t l1_offset, int64_t l1_backup_offset,
 297                            uint32_t l1_size,
 298                            int l2_size, unsigned int cluster_sectors)
 299 {
 300     VmdkExtent *extent;
 301     BDRVVmdkState *s = bs->opaque;
 302
 303     s->extents = g_realloc(s->extents,
 304                               (s->num_extents + 1) * sizeof(VmdkExtent));
 305     extent = &s->extents[s->num_extents];
 306     s->num_extents++;
 307
 308     memset(extent, 0, sizeof(VmdkExtent));
 309     extent->file = file;
 310     extent->flat = flat;
 311     extent->sectors = sectors;
 312     extent->l1_table_offset = l1_offset;
 313     extent->l1_backup_table_offset = l1_backup_offset;
 314     extent->l1_size = l1_size;
 315     extent->l1_entry_sectors = l2_size * cluster_sectors;
 316     extent->l2_size = l2_size;
 317     extent->cluster_sectors = cluster_sectors;
 318
 319     if (s->num_extents > 1) {
 320         extent->end_sector = (*(extent - 1)).end_sector + extent->sectors;
 321     } else {
 322         extent->end_sector = extent->sectors;
 323     }
 324     bs->total_sectors = extent->end_sector;
 325     return extent;
 326 }
 327
 328 static int vmdk_init_tables(BlockDriverState *bs, VmdkExtent *extent)
 329 {
 330     int ret;
 331     int l1_size, i;
 332
 333     /* read the L1 table */
 334     l1_size = extent->l1_size * sizeof(uint32_t);
 335     extent->l1_table = g_malloc(l1_size);
 336     ret = bdrv_pread(extent->file,
 337                     extent->l1_table_offset,
 338                     extent->l1_table,
 339                     l1_size);
 340     if (ret < 0) {
 341         goto fail_l1;
 342     }
 343     for (i = 0; i < extent->l1_size; i++) {
 344         le32_to_cpus(&extent->l1_table[i]);
 345     }
 346
 347     if (extent->l1_backup_table_offset) {
 348         extent->l1_backup_table = g_malloc(l1_size);
 349         ret = bdrv_pread(extent->file,
 350                         extent->l1_backup_table_offset,
 351                         extent->l1_backup_table,
 352                         l1_size);
 353         if (ret < 0) {
 354             goto fail_l1b;
 355         }
 356         for (i = 0; i < extent->l1_size; i++) {
 357             le32_to_cpus(&extent->l1_backup_table[i]);
 358         }
 359     }
 360
 361     extent->l2_cache =
 362         g_malloc(extent->l2_size * L2_CACHE_SIZE * sizeof(uint32_t));
 363     return 0;
 364  fail_l1b:
 365     g_free(extent->l1_backup_table);
 366  fail_l1:
 367     g_free(extent->l1_table);
 368     return ret;
 369 }
 370
 371 static int vmdk_open_vmdk3(BlockDriverState *bs,
 372                            BlockDriverState *file,
 373                            int flags)
 374 {
 375     int ret;
 376     uint32_t magic;
 377     VMDK3Header header;
 378     VmdkExtent *extent;
 379
 380     ret = bdrv_pread(file, sizeof(magic), &header, sizeof(header));
 381     if (ret < 0) {
 382         return ret;
 383     }
 384     extent = vmdk_add_extent(bs,
 385                              bs->file, false,
 386                              le32_to_cpu(header.disk_sectors),
 387                              le32_to_cpu(header.l1dir_offset) << 9,
 388                              0, 1 << 6, 1 << 9,
 389                              le32_to_cpu(header.granularity));
 390     ret = vmdk_init_tables(bs, extent);
 391     if (ret) {
 392         /* free extent allocated by vmdk_add_extent */
 393         vmdk_free_last_extent(bs);
 394     }
 395     return ret;
 396 }
 397
 398 static int vmdk_open_vmdk4(BlockDriverState *bs,
 399                            BlockDriverState *file,
 400                            int flags)
 401 {
 402     int ret;
 403     uint32_t magic;
 404     uint32_t l1_size, l1_entry_sectors;
 405     VMDK4Header header;
 406     VmdkExtent *extent;
 407
 408     ret = bdrv_pread(file, sizeof(magic), &header, sizeof(header));
 409     if (ret < 0) {
 410         return ret;
 411     }
 412     l1_entry_sectors = le32_to_cpu(header.num_gtes_per_gte)
 413                         * le64_to_cpu(header.granularity);
 414     if (l1_entry_sectors <= 0) {
 415         return -EINVAL;
 416     }
 417     l1_size = (le64_to_cpu(header.capacity) + l1_entry_sectors - 1)
 418                 / l1_entry_sectors;
 419     extent = vmdk_add_extent(bs, file, false,
 420                           le64_to_cpu(header.capacity),
 421                           le64_to_cpu(header.gd_offset) << 9,
 422                           le64_to_cpu(header.rgd_offset) << 9,
 423                           l1_size,
 424                           le32_to_cpu(header.num_gtes_per_gte),
 425                           le64_to_cpu(header.granularity));
 426     ret = vmdk_init_tables(bs, extent);
 427     if (ret) {
 428         /* free extent allocated by vmdk_add_extent */
 429         vmdk_free_last_extent(bs);
 430     }
 431     return ret;
 432 }
 433
 434 /* find an option value out of descriptor file */
 435 static int vmdk_parse_description(const char *desc, const char *opt_name,
 436         char *buf, int buf_size)
 437 {
 438     char *opt_pos, *opt_end;
 439     const char *end = desc + strlen(desc);
 440
 441     opt_pos = strstr(desc, opt_name);
 442     if (!opt_pos) {
 443         return -1;
 444     }
 445     /* Skip "=\"" following opt_name */
 446     opt_pos += strlen(opt_name) + 2;
 447     if (opt_pos >= end) {
 448         return -1;
 449     }
 450     opt_end = opt_pos;
 451     while (opt_end < end && *opt_end != '"') {
 452         opt_end++;
 453     }
 454     if (opt_end == end || buf_size < opt_end - opt_pos + 1) {
 455         return -1;
 456     }
 457     pstrcpy(buf, opt_end - opt_pos + 1, opt_pos);
 458     return 0;
 459 }
 460
 461 /* Open an extent file and append to bs array */
 462 static int vmdk_open_sparse(BlockDriverState *bs,
 463                             BlockDriverState *file,
 464                             int flags)
 465 {
 466     uint32_t magic;
 467
 468     if (bdrv_pread(file, 0, &magic, sizeof(magic)) != sizeof(magic)) {
 469         return -EIO;
 470     }
 471
 472     magic = be32_to_cpu(magic);
 473     switch (magic) {
 474         case VMDK3_MAGIC:
 475             return vmdk_open_vmdk3(bs, file, flags);
 476             break;
 477         case VMDK4_MAGIC:
 478             return vmdk_open_vmdk4(bs, file, flags);
 479             break;
 480         default:
 481             return -EINVAL;
 482             break;
 483     }
 484 }
 485
 486 static int vmdk_parse_extents(const char *desc, BlockDriverState *bs,
 487         const char *desc_file_path)
 488 {
 489     int ret;
 490     char access[11];
 491     char type[11];
 492     char fname[512];
 493     const char *p = desc;
 494     int64_t sectors = 0;
 495     int64_t flat_offset;
 496     char extent_path[PATH_MAX];
 497     BlockDriverState *extent_file;
 498
 499     while (*p) {
 500         /* parse extent line:
 501          * RW [size in sectors] FLAT "file-name.vmdk" OFFSET
 502          * or
 503          * RW [size in sectors] SPARSE "file-name.vmdk"
 504          */
 505         flat_offset = -1;
 506         ret = sscanf(p, "%10s %" SCNd64 " %10s %511s %" SCNd64,
 507                 access, &sectors, type, fname, &flat_offset);
 508         if (ret < 4 || strcmp(access, "RW")) {
 509             goto next_line;
 510         } else if (!strcmp(type, "FLAT")) {
 511             if (ret != 5 || flat_offset < 0) {
 512                 return -EINVAL;
 513             }
 514         } else if (ret != 4) {
 515             return -EINVAL;
 516         }
 517
 518         /* trim the quotation marks around */
 519         if (fname[0] == '"') {
 520             memmove(fname, fname + 1, strlen(fname));
 521             if (strlen(fname) <= 1 || fname[strlen(fname) - 1] != '"') {
 522                 return -EINVAL;
 523             }
 524             fname[strlen(fname) - 1] = '\0';
 525         }
 526         if (sectors <= 0 ||
 527             (strcmp(type, "FLAT") && strcmp(type, "SPARSE")) ||
 528             (strcmp(access, "RW"))) {
 529             goto next_line;
 530         }
 531
 532         path_combine(extent_path, sizeof(extent_path),
 533                 desc_file_path, fname);
 534         ret = bdrv_file_open(&extent_file, extent_path, bs->open_flags);
 535         if (ret) {
 536             return ret;
 537         }
 538
 539         /* save to extents array */
 540         if (!strcmp(type, "FLAT")) {
 541             /* FLAT extent */
 542             VmdkExtent *extent;
 543
 544             extent = vmdk_add_extent(bs, extent_file, true, sectors,
 545                             0, 0, 0, 0, sectors);
 546             extent->flat_start_offset = flat_offset;
 547         } else if (!strcmp(type, "SPARSE")) {
 548             /* SPARSE extent */
 549             ret = vmdk_open_sparse(bs, extent_file, bs->open_flags);
 550             if (ret) {
 551                 bdrv_delete(extent_file);
 552                 return ret;
 553             }
 554         } else {
 555             fprintf(stderr,
 556                 "VMDK: Not supported extent type \"%s\""".\n", type);
 557             return -ENOTSUP;
 558         }
 559 next_line:
 560         /* move to next line */
 561         while (*p && *p != '\n') {
 562             p++;
 563         }
 564         p++;
 565     }
 566     return 0;
 567 }
 568
 569 static int vmdk_open_desc_file(BlockDriverState *bs, int flags)
 570 {
 571     int ret;
 572     char buf[2048];
 573     char ct[128];
 574     BDRVVmdkState *s = bs->opaque;
 575
 576     ret = bdrv_pread(bs->file, 0, buf, sizeof(buf));
 577     if (ret < 0) {
 578         return ret;
 579     }
 580     buf[2047] = '\0';
 581     if (vmdk_parse_description(buf, "createType", ct, sizeof(ct))) {
 582         return -EINVAL;
 583     }
 584     if (strcmp(ct, "monolithicFlat") &&
 585         strcmp(ct, "twoGbMaxExtentSparse") &&
 586         strcmp(ct, "twoGbMaxExtentFlat")) {
 587         fprintf(stderr,
 588                 "VMDK: Not supported image type \"%s\""".\n", ct);
 589         return -ENOTSUP;
 590     }
 591     s->desc_offset = 0;
 592     ret = vmdk_parse_extents(buf, bs, bs->file->filename);
 593     if (ret) {
 594         return ret;
 595     }
 596
 597     /* try to open parent images, if exist */
 598     if (vmdk_parent_open(bs)) {
 599         g_free(s->extents);
 600         return -EINVAL;
 601     }
 602     s->parent_cid = vmdk_read_cid(bs, 1);
 603     return 0;
 604 }
 605
 606 static int vmdk_open(BlockDriverState *bs, int flags)
 607 {
 608     int ret;
 609     BDRVVmdkState *s = bs->opaque;
 610
 611     if (vmdk_open_sparse(bs, bs->file, flags) == 0) {
 612         s->desc_offset = 0x200;
 613         /* try to open parent images, if exist */
 614         ret = vmdk_parent_open(bs);
 615         if (ret) {
 616             vmdk_free_extents(bs);
 617             return ret;
 618         }
 619         s->parent_cid = vmdk_read_cid(bs, 1);
 620         return 0;
 621     } else {
 622         return vmdk_open_desc_file(bs, flags);
 623     }
 624 }
 625
 626 static int get_whole_cluster(BlockDriverState *bs,
 627                 VmdkExtent *extent,
 628                 uint64_t cluster_offset,
 629                 uint64_t offset,
 630                 bool allocate)
 631 {
 632     /* 128 sectors * 512 bytes each = grain size 64KB */
 633     uint8_t  whole_grain[extent->cluster_sectors * 512];
 634
 635     /* we will be here if it's first write on non-exist grain(cluster).
 636      * try to read from parent image, if exist */
 637     if (bs->backing_hd) {
 638         int ret;
 639
 640         if (!vmdk_is_cid_valid(bs)) {
 641             return -1;
 642         }
 643
 644         /* floor offset to cluster */
 645         offset -= offset % (extent->cluster_sectors * 512);
 646         ret = bdrv_read(bs->backing_hd, offset >> 9, whole_grain,
 647                 extent->cluster_sectors);
 648         if (ret < 0) {
 649             return -1;
 650         }
 651
 652         /* Write grain only into the active image */
 653         ret = bdrv_write(extent->file, cluster_offset, whole_grain,
 654                 extent->cluster_sectors);
 655         if (ret < 0) {
 656             return -1;
 657         }
 658     }
 659     return 0;
 660 }
 661
 662 static int vmdk_L2update(VmdkExtent *extent, VmdkMetaData *m_data)
 663 {
 664     /* update L2 table */
 665     if (bdrv_pwrite_sync(
 666                 extent->file,
 667                 ((int64_t)m_data->l2_offset * 512)
 668                     + (m_data->l2_index * sizeof(m_data->offset)),
 669                 &(m_data->offset),
 670                 sizeof(m_data->offset)
 671             ) < 0) {
 672         return -1;
 673     }
 674     /* update backup L2 table */
 675     if (extent->l1_backup_table_offset != 0) {
 676         m_data->l2_offset = extent->l1_backup_table[m_data->l1_index];
 677         if (bdrv_pwrite_sync(
 678                     extent->file,
 679                     ((int64_t)m_data->l2_offset * 512)
 680                         + (m_data->l2_index * sizeof(m_data->offset)),
 681                     &(m_data->offset), sizeof(m_data->offset)
 682                 ) < 0) {
 683             return -1;
 684         }
 685     }
 686
 687     return 0;
 688 }
 689
 690 static int get_cluster_offset(BlockDriverState *bs,
 691                                     VmdkExtent *extent,
 692                                     VmdkMetaData *m_data,
 693                                     uint64_t offset,
 694                                     int allocate,
 695                                     uint64_t *cluster_offset)
 696 {
 697     unsigned int l1_index, l2_offset, l2_index;
 698     int min_index, i, j;
 699     uint32_t min_count, *l2_table, tmp = 0;
 700
 701     if (m_data) {
 702         m_data->valid = 0;
 703     }
 704     if (extent->flat) {
 705         *cluster_offset = extent->flat_start_offset;
 706         return 0;
 707     }
 708
 709     offset -= (extent->end_sector - extent->sectors) * SECTOR_SIZE;
 710     l1_index = (offset >> 9) / extent->l1_entry_sectors;
 711     if (l1_index >= extent->l1_size) {
 712         return -1;
 713     }
 714     l2_offset = extent->l1_table[l1_index];
 715     if (!l2_offset) {
 716         return -1;
 717     }
 718     for (i = 0; i < L2_CACHE_SIZE; i++) {
 719         if (l2_offset == extent->l2_cache_offsets[i]) {
 720             /* increment the hit count */
 721             if (++extent->l2_cache_counts[i] == 0xffffffff) {
 722                 for (j = 0; j < L2_CACHE_SIZE; j++) {
 723                     extent->l2_cache_counts[j] >>= 1;
 724                 }
 725             }
 726             l2_table = extent->l2_cache + (i * extent->l2_size);
 727             goto found;
 728         }
 729     }
 730     /* not found: load a new entry in the least used one */
 731     min_index = 0;
 732     min_count = 0xffffffff;
 733     for (i = 0; i < L2_CACHE_SIZE; i++) {
 734         if (extent->l2_cache_counts[i] < min_count) {
 735             min_count = extent->l2_cache_counts[i];
 736             min_index = i;
 737         }
 738     }
 739     l2_table = extent->l2_cache + (min_index * extent->l2_size);
 740     if (bdrv_pread(
 741                 extent->file,
 742                 (int64_t)l2_offset * 512,
 743                 l2_table,
 744                 extent->l2_size * sizeof(uint32_t)
 745             ) != extent->l2_size * sizeof(uint32_t)) {
 746         return -1;
 747     }
 748
 749     extent->l2_cache_offsets[min_index] = l2_offset;
 750     extent->l2_cache_counts[min_index] = 1;
 751  found:
 752     l2_index = ((offset >> 9) / extent->cluster_sectors) % extent->l2_size;
 753     *cluster_offset = le32_to_cpu(l2_table[l2_index]);
 754
 755     if (!*cluster_offset) {
 756         if (!allocate) {
 757             return -1;
 758         }
 759
 760         /* Avoid the L2 tables update for the images that have snapshots. */
 761         *cluster_offset = bdrv_getlength(extent->file);
 762         bdrv_truncate(
 763             extent->file,
 764             *cluster_offset + (extent->cluster_sectors << 9)
 765         );
 766
 767         *cluster_offset >>= 9;
 768         tmp = cpu_to_le32(*cluster_offset);
 769         l2_table[l2_index] = tmp;
 770
 771         /* First of all we write grain itself, to avoid race condition
 772          * that may to corrupt the image.
 773          * This problem may occur because of insufficient space on host disk
 774          * or inappropriate VM shutdown.
 775          */
 776         if (get_whole_cluster(
 777                 bs, extent, *cluster_offset, offset, allocate) == -1) {
 778             return -1;
 779         }
 780
 781         if (m_data) {
 782             m_data->offset = tmp;
 783             m_data->l1_index = l1_index;
 784             m_data->l2_index = l2_index;
 785             m_data->l2_offset = l2_offset;
 786             m_data->valid = 1;
 787         }
 788     }
 789     *cluster_offset <<= 9;
 790     return 0;
 791 }
 792
 793 static VmdkExtent *find_extent(BDRVVmdkState *s,
 794                                 int64_t sector_num, VmdkExtent *start_hint)
 795 {
 796     VmdkExtent *extent = start_hint;
 797
 798     if (!extent) {
 799         extent = &s->extents[0];
 800     }
 801     while (extent < &s->extents[s->num_extents]) {
 802         if (sector_num < extent->end_sector) {
 803             return extent;
 804         }
 805         extent++;
 806     }
 807     return NULL;
 808 }
 809
 810 static int vmdk_is_allocated(BlockDriverState *bs, int64_t sector_num,
 811                              int nb_sectors, int *pnum)
 812 {
 813     BDRVVmdkState *s = bs->opaque;
 814     int64_t index_in_cluster, n, ret;
 815     uint64_t offset;
 816     VmdkExtent *extent;
 817
 818     extent = find_extent(s, sector_num, NULL);
 819     if (!extent) {
 820         return 0;
 821     }
 822     ret = get_cluster_offset(bs, extent, NULL,
 823                             sector_num * 512, 0, &offset);
 824     /* get_cluster_offset returning 0 means success */
 825     ret = !ret;
 826
 827     index_in_cluster = sector_num % extent->cluster_sectors;
 828     n = extent->cluster_sectors - index_in_cluster;
 829     if (n > nb_sectors) {
 830         n = nb_sectors;
 831     }
 832     *pnum = n;
 833     return ret;
 834 }
 835
 836 static int vmdk_write_extent(VmdkExtent *extent, int64_t cluster_offset,
 837                             int64_t offset_in_cluster, const uint8_t *buf,
 838                             int nb_sectors, int64_t sector_num)
 839 {
 840     int ret;
 841     const uint8_t *write_buf = buf;
 842     int write_len = nb_sectors * 512;
 843
 844     ret = bdrv_pwrite(extent->file,
 845                         cluster_offset + offset_in_cluster,
 846                         write_buf,
 847                         write_len);
 848     if (ret != write_len) {
 849         ret = ret < 0 ? ret : -EIO;
 850         goto out;
 851     }
 852     ret = 0;
 853  out:
 854     return ret;
 855 }
 856
 857 static int vmdk_read_extent(VmdkExtent *extent, int64_t cluster_offset,
 858                             int64_t offset_in_cluster, uint8_t *buf,
 859                             int nb_sectors)
 860 {
 861     int ret;
 862
 863     ret = bdrv_pread(extent->file,
 864                       cluster_offset + offset_in_cluster,
 865                       buf, nb_sectors * 512);
 866     if (ret == nb_sectors * 512) {
 867         return 0;
 868     } else {
 869         return -EIO;
 870     }
 871 }
 872
 873 static int vmdk_read(BlockDriverState *bs, int64_t sector_num,
 874                     uint8_t *buf, int nb_sectors)
 875 {
 876     BDRVVmdkState *s = bs->opaque;
 877     int ret;
 878     uint64_t n, index_in_cluster;
 879     VmdkExtent *extent = NULL;
 880     uint64_t cluster_offset;
 881
 882     while (nb_sectors > 0) {
 883         extent = find_extent(s, sector_num, extent);
 884         if (!extent) {
 885             return -EIO;
 886         }
 887         ret = get_cluster_offset(
 888                             bs, extent, NULL,
 889                             sector_num << 9, 0, &cluster_offset);
 890         index_in_cluster = sector_num % extent->cluster_sectors;
 891         n = extent->cluster_sectors - index_in_cluster;
 892         if (n > nb_sectors) {
 893             n = nb_sectors;
 894         }
 895         if (ret) {
 896             /* if not allocated, try to read from parent image, if exist */
 897             if (bs->backing_hd) {
 898                 if (!vmdk_is_cid_valid(bs)) {
 899                     return -EINVAL;
 900                 }
 901                 ret = bdrv_read(bs->backing_hd, sector_num, buf, n);
 902                 if (ret < 0) {
 903                     return ret;
 904                 }
 905             } else {
 906                 memset(buf, 0, 512 * n);
 907             }
 908         } else {
 909             ret = vmdk_read_extent(extent,
 910                             cluster_offset, index_in_cluster * 512,
 911                             buf, n);
 912             if (ret) {
 913                 return ret;
 914             }
 915         }
 916         nb_sectors -= n;
 917         sector_num += n;
 918         buf += n * 512;
 919     }
 920     return 0;
 921 }
 922
 923 static int vmdk_write(BlockDriverState *bs, int64_t sector_num,
 924                      const uint8_t *buf, int nb_sectors)
 925 {
 926     BDRVVmdkState *s = bs->opaque;
 927     VmdkExtent *extent = NULL;
 928     int n, ret;
 929     int64_t index_in_cluster;
 930     uint64_t cluster_offset;
 931     VmdkMetaData m_data;
 932
 933     if (sector_num > bs->total_sectors) {
 934         fprintf(stderr,
 935                 "(VMDK) Wrong offset: sector_num=0x%" PRIx64
 936                 " total_sectors=0x%" PRIx64 "\n",
 937                 sector_num, bs->total_sectors);
 938         return -EIO;
 939     }
 940
 941     while (nb_sectors > 0) {
 942         extent = find_extent(s, sector_num, extent);
 943         if (!extent) {
 944             return -EIO;
 945         }
 946         ret = get_cluster_offset(
 947                                 bs,
 948                                 extent,
 949                                 &m_data,
 950                                 sector_num << 9, 1,
 951                                 &cluster_offset);
 952         if (ret) {
 953             return -EINVAL;
 954         }
 955         index_in_cluster = sector_num % extent->cluster_sectors;
 956         n = extent->cluster_sectors - index_in_cluster;
 957         if (n > nb_sectors) {
 958             n = nb_sectors;
 959         }
 960
 961         ret = vmdk_write_extent(extent,
 962                         cluster_offset, index_in_cluster * 512,
 963                         buf, n, sector_num);
 964         if (ret) {
 965             return ret;
 966         }
 967         if (m_data.valid) {
 968             /* update L2 tables */
 969             if (vmdk_L2update(extent, &m_data) == -1) {
 970                 return -EIO;
 971             }
 972         }
 973         nb_sectors -= n;
 974         sector_num += n;
 975         buf += n * 512;
 976
 977         /* update CID on the first write every time the virtual disk is
 978          * opened */
 979         if (!s->cid_updated) {
 980             vmdk_write_cid(bs, time(NULL));
 981             s->cid_updated = true;
 982         }
 983     }
 984     return 0;
 985 }
 986
 987
 988 static int vmdk_create_extent(const char *filename, int64_t filesize, bool flat)
 989 {
 990     int ret, i;
 991     int fd = 0;
 992     VMDK4Header header;
 993     uint32_t tmp, magic, grains, gd_size, gt_size, gt_count;
 994
 995     fd = open(
 996         filename,
 997         O_WRONLY | O_CREAT | O_TRUNC | O_BINARY | O_LARGEFILE,
 998         0644);
 999     if (fd < 0) {
1000         return -errno;
1001     }
1002     if (flat) {
1003         ret = ftruncate(fd, filesize);
1004         if (ret < 0) {
1005             ret = -errno;
1006         }
1007         goto exit;
1008     }
1009     magic = cpu_to_be32(VMDK4_MAGIC);
1010     memset(&header, 0, sizeof(header));
1011     header.version = 1;
1012     header.flags = 3; /* ?? */
1013     header.capacity = filesize / 512;
1014     header.granularity = 128;
1015     header.num_gtes_per_gte = 512;
1016
1017     grains = (filesize / 512 + header.granularity - 1) / header.granularity;
1018     gt_size = ((header.num_gtes_per_gte * sizeof(uint32_t)) + 511) >> 9;
1019     gt_count =
1020         (grains + header.num_gtes_per_gte - 1) / header.num_gtes_per_gte;
1021     gd_size = (gt_count * sizeof(uint32_t) + 511) >> 9;
1022
1023     header.desc_offset = 1;
1024     header.desc_size = 20;
1025     header.rgd_offset = header.desc_offset + header.desc_size;
1026     header.gd_offset = header.rgd_offset + gd_size + (gt_size * gt_count);
1027     header.grain_offset =
1028        ((header.gd_offset + gd_size + (gt_size * gt_count) +
1029          header.granularity - 1) / header.granularity) *
1030         header.granularity;
1031     /* swap endianness for all header fields */
1032     header.version = cpu_to_le32(header.version);
1033     header.flags = cpu_to_le32(header.flags);
1034     header.capacity = cpu_to_le64(header.capacity);
1035     header.granularity = cpu_to_le64(header.granularity);
1036     header.num_gtes_per_gte = cpu_to_le32(header.num_gtes_per_gte);
1037     header.desc_offset = cpu_to_le64(header.desc_offset);
1038     header.desc_size = cpu_to_le64(header.desc_size);
1039     header.rgd_offset = cpu_to_le64(header.rgd_offset);
1040     header.gd_offset = cpu_to_le64(header.gd_offset);
1041     header.grain_offset = cpu_to_le64(header.grain_offset);
1042
1043     header.check_bytes[0] = 0xa;
1044     header.check_bytes[1] = 0x20;
1045     header.check_bytes[2] = 0xd;
1046     header.check_bytes[3] = 0xa;
1047
1048     /* write all the data */
1049     ret = qemu_write_full(fd, &magic, sizeof(magic));
1050     if (ret != sizeof(magic)) {
1051         ret = -errno;
1052         goto exit;
1053     }
1054     ret = qemu_write_full(fd, &header, sizeof(header));
1055     if (ret != sizeof(header)) {
1056         ret = -errno;
1057         goto exit;
1058     }
1059
1060     ret = ftruncate(fd, le64_to_cpu(header.grain_offset) << 9);
1061     if (ret < 0) {
1062         ret = -errno;
1063         goto exit;
1064     }
1065
1066     /* write grain directory */
1067     lseek(fd, le64_to_cpu(header.rgd_offset) << 9, SEEK_SET);
1068     for (i = 0, tmp = le64_to_cpu(header.rgd_offset) + gd_size;
1069          i < gt_count; i++, tmp += gt_size) {
1070         ret = qemu_write_full(fd, &tmp, sizeof(tmp));
1071         if (ret != sizeof(tmp)) {
1072             ret = -errno;
1073             goto exit;
1074         }
1075     }
1076
1077     /* write backup grain directory */
1078     lseek(fd, le64_to_cpu(header.gd_offset) << 9, SEEK_SET);
1079     for (i = 0, tmp = le64_to_cpu(header.gd_offset) + gd_size;
1080          i < gt_count; i++, tmp += gt_size) {
1081         ret = qemu_write_full(fd, &tmp, sizeof(tmp));
1082         if (ret != sizeof(tmp)) {
1083             ret = -errno;
1084             goto exit;
1085         }
1086     }
1087
1088     ret = 0;
1089  exit:
1090     close(fd);
1091     return ret;
1092 }
1093
1094 static int filename_decompose(const char *filename, char *path, char *prefix,
1095         char *postfix, size_t buf_len)
1096 {
1097     const char *p, *q;
1098
1099     if (filename == NULL || !strlen(filename)) {
1100         fprintf(stderr, "Vmdk: no filename provided.\n");
1101         return -1;
1102     }
1103     p = strrchr(filename, '/');
1104     if (p == NULL) {
1105         p = strrchr(filename, '\\');
1106     }
1107     if (p == NULL) {
1108         p = strrchr(filename, ':');
1109     }
1110     if (p != NULL) {
1111         p++;
1112         if (p - filename >= buf_len) {
1113             return -1;
1114         }
1115         pstrcpy(path, p - filename + 1, filename);
1116     } else {
1117         p = filename;
1118         path[0] = '\0';
1119     }
1120     q = strrchr(p, '.');
1121     if (q == NULL) {
1122         pstrcpy(prefix, buf_len, p);
1123         postfix[0] = '\0';
1124     } else {
1125         if (q - p >= buf_len) {
1126             return -1;
1127         }
1128         pstrcpy(prefix, q - p + 1, p);
1129         pstrcpy(postfix, buf_len, q);
1130     }
1131     return 0;
1132 }
1133
1134 static int relative_path(char *dest, int dest_size,
1135         const char *base, const char *target)
1136 {
1137     int i = 0;
1138     int n = 0;
1139     const char *p, *q;
1140 #ifdef _WIN32
1141     const char *sep = "\\";
1142 #else
1143     const char *sep = "/";
1144 #endif
1145
1146     if (!(dest && base && target)) {
1147         return -1;
1148     }
1149     if (path_is_absolute(target)) {
1150         dest[dest_size - 1] = '\0';
1151         strncpy(dest, target, dest_size - 1);
1152         return 0;
1153     }
1154     while (base[i] == target[i]) {
1155         i++;
1156     }
1157     p = &base[i];
1158     q = &target[i];
1159     while (*p) {
1160         if (*p == *sep) {
1161             n++;
1162         }
1163         p++;
1164     }
1165     dest[0] = '\0';
1166     for (; n; n--) {
1167         pstrcat(dest, dest_size, "..");
1168         pstrcat(dest, dest_size, sep);
1169     }
1170     pstrcat(dest, dest_size, q);
1171     return 0;
1172 }
1173
1174 static int vmdk_create(const char *filename, QEMUOptionParameter *options)
1175 {
1176     int fd, idx = 0;
1177     char desc[BUF_SIZE];
1178     int64_t total_size = 0, filesize;
1179     const char *backing_file = NULL;
1180     const char *fmt = NULL;
1181     int flags = 0;
1182     int ret = 0;
1183     bool flat, split;
1184     char ext_desc_lines[BUF_SIZE] = "";
1185     char path[PATH_MAX], prefix[PATH_MAX], postfix[PATH_MAX];
1186     const int64_t split_size = 0x80000000;  /* VMDK has constant split size */
1187     const char *desc_extent_line;
1188     char parent_desc_line[BUF_SIZE] = "";
1189     uint32_t parent_cid = 0xffffffff;
1190     const char desc_template[] =
1191         "# Disk DescriptorFile\n"
1192         "version=1\n"
1193         "CID=%x\n"
1194         "parentCID=%x\n"
1195         "createType=\"%s\"\n"
1196         "%s"
1197         "\n"
1198         "# Extent description\n"
1199         "%s"
1200         "\n"
1201         "# The Disk Data Base\n"
1202         "#DDB\n"
1203         "\n"
1204         "ddb.virtualHWVersion = \"%d\"\n"
1205         "ddb.geometry.cylinders = \"%" PRId64 "\"\n"
1206         "ddb.geometry.heads = \"16\"\n"
1207         "ddb.geometry.sectors = \"63\"\n"
1208         "ddb.adapterType = \"ide\"\n";
1209
1210     if (filename_decompose(filename, path, prefix, postfix, PATH_MAX)) {
1211         return -EINVAL;
1212     }
1213     /* Read out options */
1214     while (options && options->name) {
1215         if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
1216             total_size = options->value.n;
1217         } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) {
1218             backing_file = options->value.s;
1219         } else if (!strcmp(options->name, BLOCK_OPT_COMPAT6)) {
1220             flags |= options->value.n ? BLOCK_FLAG_COMPAT6 : 0;
1221         } else if (!strcmp(options->name, BLOCK_OPT_SUBFMT)) {
1222             fmt = options->value.s;
1223         }
1224         options++;
1225     }
1226     if (!fmt) {
1227         /* Default format to monolithicSparse */
1228         fmt = "monolithicSparse";
1229     } else if (strcmp(fmt, "monolithicFlat") &&
1230                strcmp(fmt, "monolithicSparse") &&
1231                strcmp(fmt, "twoGbMaxExtentSparse") &&
1232                strcmp(fmt, "twoGbMaxExtentFlat")) {
1233         fprintf(stderr, "VMDK: Unknown subformat: %s\n", fmt);
1234         return -EINVAL;
1235     }
1236     split = !(strcmp(fmt, "twoGbMaxExtentFlat") &&
1237               strcmp(fmt, "twoGbMaxExtentSparse"));
1238     flat = !(strcmp(fmt, "monolithicFlat") &&
1239              strcmp(fmt, "twoGbMaxExtentFlat"));
1240     if (flat) {
1241         desc_extent_line = "RW %lld FLAT \"%s\" 0\n";
1242     } else {
1243         desc_extent_line = "RW %lld SPARSE \"%s\"\n";
1244     }
1245     if (flat && backing_file) {
1246         /* not supporting backing file for flat image */
1247         return -ENOTSUP;
1248     }
1249     if (backing_file) {
1250         char parent_filename[PATH_MAX];
1251         BlockDriverState *bs = bdrv_new("");
1252         ret = bdrv_open(bs, backing_file, 0, NULL);
1253         if (ret != 0) {
1254             bdrv_delete(bs);
1255             return ret;
1256         }
1257         if (strcmp(bs->drv->format_name, "vmdk")) {
1258             bdrv_delete(bs);
1259             return -EINVAL;
1260         }
1261         filesize = bdrv_getlength(bs);
1262         parent_cid = vmdk_read_cid(bs, 0);
1263         bdrv_delete(bs);
1264         relative_path(parent_filename, sizeof(parent_filename),
1265                       filename, backing_file);
1266         snprintf(parent_desc_line, sizeof(parent_desc_line),
1267                 "parentFileNameHint=\"%s\"", parent_filename);
1268     }
1269
1270     /* Create extents */
1271     filesize = total_size;
1272     while (filesize > 0) {
1273         char desc_line[BUF_SIZE];
1274         char ext_filename[PATH_MAX];
1275         char desc_filename[PATH_MAX];
1276         int64_t size = filesize;
1277
1278         if (split && size > split_size) {
1279             size = split_size;
1280         }
1281         if (split) {
1282             snprintf(desc_filename, sizeof(desc_filename), "%s-%c%03d%s",
1283                     prefix, flat ? 'f' : 's', ++idx, postfix);
1284         } else if (flat) {
1285             snprintf(desc_filename, sizeof(desc_filename), "%s-flat%s",
1286                     prefix, postfix);
1287         } else {
1288             snprintf(desc_filename, sizeof(desc_filename), "%s%s",
1289                     prefix, postfix);
1290         }
1291         snprintf(ext_filename, sizeof(ext_filename), "%s%s",
1292                 path, desc_filename);
1293
1294         if (vmdk_create_extent(ext_filename, size, flat)) {
1295             return -EINVAL;
1296         }
1297         filesize -= size;
1298
1299         /* Format description line */
1300         snprintf(desc_line, sizeof(desc_line),
1301                     desc_extent_line, size / 512, desc_filename);
1302         pstrcat(ext_desc_lines, sizeof(ext_desc_lines), desc_line);
1303     }
1304     /* generate descriptor file */
1305     snprintf(desc, sizeof(desc), desc_template,
1306             (unsigned int)time(NULL),
1307             parent_cid,
1308             fmt,
1309             parent_desc_line,
1310             ext_desc_lines,
1311             (flags & BLOCK_FLAG_COMPAT6 ? 6 : 4),
1312             total_size / (int64_t)(63 * 16 * 512));
1313     if (split || flat) {
1314         fd = open(
1315                 filename,
1316                 O_WRONLY | O_CREAT | O_TRUNC | O_BINARY | O_LARGEFILE,
1317                 0644);
1318     } else {
1319         fd = open(
1320                 filename,
1321                 O_WRONLY | O_BINARY | O_LARGEFILE,
1322                 0644);
1323     }
1324     if (fd < 0) {
1325         return -errno;
1326     }
1327     /* the descriptor offset = 0x200 */
1328     if (!split && !flat && 0x200 != lseek(fd, 0x200, SEEK_SET)) {
1329         ret = -errno;
1330         goto exit;
1331     }
1332     ret = qemu_write_full(fd, desc, strlen(desc));
1333     if (ret != strlen(desc)) {
1334         ret = -errno;
1335         goto exit;
1336     }
1337     ret = 0;
1338 exit:
1339     close(fd);
1340     return ret;
1341 }
1342
1343 static void vmdk_close(BlockDriverState *bs)
1344 {
1345     vmdk_free_extents(bs);
1346 }
1347
1348 static int vmdk_flush(BlockDriverState *bs)
1349 {
1350     int i, ret, err;
1351     BDRVVmdkState *s = bs->opaque;
1352
1353     ret = bdrv_flush(bs->file);
1354     for (i = 0; i < s->num_extents; i++) {
1355         err = bdrv_flush(s->extents[i].file);
1356         if (err < 0) {
1357             ret = err;
1358         }
1359     }
1360     return ret;
1361 }
1362
1363 static int64_t vmdk_get_allocated_file_size(BlockDriverState *bs)
1364 {
1365     int i;
1366     int64_t ret = 0;
1367     int64_t r;
1368     BDRVVmdkState *s = bs->opaque;
1369
1370     ret = bdrv_get_allocated_file_size(bs->file);
1371     if (ret < 0) {
1372         return ret;
1373     }
1374     for (i = 0; i < s->num_extents; i++) {
1375         if (s->extents[i].file == bs->file) {
1376             continue;
1377         }
1378         r = bdrv_get_allocated_file_size(s->extents[i].file);
1379         if (r < 0) {
1380             return r;
1381         }
1382         ret += r;
1383     }
1384     return ret;
1385 }
1386
1387 static QEMUOptionParameter vmdk_create_options[] = {
1388     {
1389         .name = BLOCK_OPT_SIZE,
1390         .type = OPT_SIZE,
1391         .help = "Virtual disk size"
1392     },
1393     {
1394         .name = BLOCK_OPT_BACKING_FILE,
1395         .type = OPT_STRING,
1396         .help = "File name of a base image"
1397     },
1398     {
1399         .name = BLOCK_OPT_COMPAT6,
1400         .type = OPT_FLAG,
1401         .help = "VMDK version 6 image"
1402     },
1403     {
1404         .name = BLOCK_OPT_SUBFMT,
1405         .type = OPT_STRING,
1406         .help =
1407             "VMDK flat extent format, can be one of "
1408             "{monolithicSparse (default) | monolithicFlat | twoGbMaxExtentSparse | twoGbMaxExtentFlat} "
1409     },
1410     { NULL }
1411 };
1412
1413 static BlockDriver bdrv_vmdk = {
1414     .format_name    = "vmdk",
1415     .instance_size  = sizeof(BDRVVmdkState),
1416     .bdrv_probe     = vmdk_probe,
1417     .bdrv_open      = vmdk_open,
1418     .bdrv_read      = vmdk_read,
1419     .bdrv_write     = vmdk_write,
1420     .bdrv_close     = vmdk_close,
1421     .bdrv_create    = vmdk_create,
1422     .bdrv_flush     = vmdk_flush,
1423     .bdrv_is_allocated  = vmdk_is_allocated,
1424     .bdrv_get_allocated_file_size  = vmdk_get_allocated_file_size,
1425
1426     .create_options = vmdk_create_options,
1427 };
1428
1429 static void bdrv_vmdk_init(void)
1430 {
1431     bdrv_register(&bdrv_vmdk);
1432 }
1433
1434 block_init(bdrv_vmdk_init);