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 "vl.h"
  27 #include "block_int.h"
  28
  29 #define VMDK3_MAGIC (('C' << 24) | ('O' << 16) | ('W' << 8) | 'D')
  30 #define VMDK4_MAGIC (('K' << 24) | ('D' << 16) | ('M' << 8) | 'V')
  31
  32 typedef struct {
  33     uint32_t version;
  34     uint32_t flags;
  35     uint32_t disk_sectors;
  36     uint32_t granularity;
  37     uint32_t l1dir_offset;
  38     uint32_t l1dir_size;
  39     uint32_t file_sectors;
  40     uint32_t cylinders;
  41     uint32_t heads;
  42     uint32_t sectors_per_track;
  43 } VMDK3Header;
  44
  45 typedef struct {
  46     uint32_t version;
  47     uint32_t flags;
  48     int64_t capacity;
  49     int64_t granularity;
  50     int64_t desc_offset;
  51     int64_t desc_size;
  52     int32_t num_gtes_per_gte;
  53     int64_t rgd_offset;
  54     int64_t gd_offset;
  55     int64_t grain_offset;
  56     char filler[1];
  57     char check_bytes[4];
  58 } __attribute__((packed)) VMDK4Header;
  59
  60 #define L2_CACHE_SIZE 16
  61
  62 typedef struct BDRVVmdkState {
  63     BlockDriverState *hd;
  64     int64_t l1_table_offset;
  65     int64_t l1_backup_table_offset;
  66     uint32_t *l1_table;
  67     uint32_t *l1_backup_table;
  68     unsigned int l1_size;
  69     uint32_t l1_entry_sectors;
  70
  71     unsigned int l2_size;
  72     uint32_t *l2_cache;
  73     uint32_t l2_cache_offsets[L2_CACHE_SIZE];
  74     uint32_t l2_cache_counts[L2_CACHE_SIZE];
  75
  76     unsigned int cluster_sectors;
  77     uint32_t parent_cid;
  78 } BDRVVmdkState;
  79
  80 static int vmdk_probe(const uint8_t *buf, int buf_size, const char *filename)
  81 {
  82     uint32_t magic;
  83
  84     if (buf_size < 4)
  85         return 0;
  86     magic = be32_to_cpu(*(uint32_t *)buf);
  87     if (magic == VMDK3_MAGIC ||
  88         magic == VMDK4_MAGIC)
  89         return 100;
  90     else
  91         return 0;
  92 }
  93
  94 #define CHECK_CID 1
  95
  96 #define SECTOR_SIZE 512
  97 #define DESC_SIZE 20*SECTOR_SIZE        // 20 sectors of 512 bytes each
  98 #define HEADER_SIZE 512                         // first sector of 512 bytes
  99
 100 static uint32_t vmdk_read_cid(BlockDriverState *bs, int parent)
 101 {
 102     BDRVVmdkState *s = bs->opaque;
 103     char desc[DESC_SIZE];
 104     uint32_t cid;
 105     char *p_name, *cid_str;
 106     size_t cid_str_size;
 107
 108     /* the descriptor offset = 0x200 */
 109     if (bdrv_pread(s->hd, 0x200, desc, DESC_SIZE) != DESC_SIZE)
 110         return 0;
 111
 112     if (parent) {
 113         cid_str = "parentCID";
 114         cid_str_size = sizeof("parentCID");
 115     } else {
 116         cid_str = "CID";
 117         cid_str_size = sizeof("CID");
 118     }
 119
 120     if ((p_name = strstr(desc,cid_str)) != 0) {
 121         p_name += cid_str_size;
 122         sscanf(p_name,"%x",&cid);
 123     }
 124
 125     return cid;
 126 }
 127
 128 static int vmdk_write_cid(BlockDriverState *bs, uint32_t cid)
 129 {
 130     BDRVVmdkState *s = bs->opaque;
 131     char desc[DESC_SIZE], tmp_desc[DESC_SIZE];
 132     char *p_name, *tmp_str;
 133
 134     /* the descriptor offset = 0x200 */
 135     if (bdrv_pread(s->hd, 0x200, desc, DESC_SIZE) != DESC_SIZE)
 136         return -1;
 137
 138     tmp_str = strstr(desc,"parentCID");
 139     strcpy(tmp_desc, tmp_str);
 140     if ((p_name = strstr(desc,"CID")) != 0) {
 141         p_name += sizeof("CID");
 142         sprintf(p_name,"%x\n",cid);
 143         strcat(desc,tmp_desc);
 144     }
 145
 146     if (bdrv_pwrite(s->hd, 0x200, desc, DESC_SIZE) != DESC_SIZE)
 147         return -1;
 148     return 0;
 149 }
 150
 151 static int vmdk_is_cid_valid(BlockDriverState *bs)
 152 {
 153 #ifdef CHECK_CID
 154     BDRVVmdkState *s = bs->opaque;
 155     BlockDriverState *p_bs = s->hd->backing_hd;
 156     uint32_t cur_pcid;
 157
 158     if (p_bs) {
 159         cur_pcid = vmdk_read_cid(p_bs,0);
 160         if (s->parent_cid != cur_pcid)
 161             // CID not valid
 162             return 0;
 163     }
 164 #endif
 165     // CID valid
 166     return 1;
 167 }
 168
 169 static int vmdk_snapshot_create(const char *filename, const char *backing_file)
 170 {
 171     int snp_fd, p_fd;
 172     uint32_t p_cid;
 173     char *p_name, *gd_buf, *rgd_buf;
 174     const char *real_filename, *temp_str;
 175     VMDK4Header header;
 176     uint32_t gde_entries, gd_size;
 177     int64_t gd_offset, rgd_offset, capacity, gt_size;
 178     char p_desc[DESC_SIZE], s_desc[DESC_SIZE], hdr[HEADER_SIZE];
 179     char *desc_template =
 180     "# Disk DescriptorFile\n"
 181     "version=1\n"
 182     "CID=%x\n"
 183     "parentCID=%x\n"
 184     "createType=\"monolithicSparse\"\n"
 185     "parentFileNameHint=\"%s\"\n"
 186     "\n"
 187     "# Extent description\n"
 188     "RW %lu SPARSE \"%s\"\n"
 189     "\n"
 190     "# The Disk Data Base \n"
 191     "#DDB\n"
 192     "\n";
 193
 194     snp_fd = open(filename, O_RDWR | O_CREAT | O_TRUNC | O_BINARY | O_LARGEFILE, 0644);
 195     if (snp_fd < 0)
 196         return -1;
 197     p_fd = open(backing_file, O_RDONLY | O_BINARY | O_LARGEFILE);
 198     if (p_fd < 0) {
 199         close(snp_fd);
 200         return -1;
 201     }
 202
 203     /* read the header */
 204     if (lseek(p_fd, 0x0, SEEK_SET) == -1)
 205         goto fail;
 206     if (read(p_fd, hdr, HEADER_SIZE) != HEADER_SIZE)
 207         goto fail;
 208
 209     /* write the header */
 210     if (lseek(snp_fd, 0x0, SEEK_SET) == -1)
 211         goto fail;
 212     if (write(snp_fd, hdr, HEADER_SIZE) == -1)
 213         goto fail;
 214
 215     memset(&header, 0, sizeof(header));
 216     memcpy(&header,&hdr[4], sizeof(header)); // skip the VMDK4_MAGIC
 217
 218     ftruncate(snp_fd, header.grain_offset << 9);
 219     /* the descriptor offset = 0x200 */
 220     if (lseek(p_fd, 0x200, SEEK_SET) == -1)
 221         goto fail;
 222     if (read(p_fd, p_desc, DESC_SIZE) != DESC_SIZE)
 223         goto fail;
 224
 225     if ((p_name = strstr(p_desc,"CID")) != 0) {
 226         p_name += sizeof("CID");
 227         sscanf(p_name,"%x",&p_cid);
 228     }
 229
 230     real_filename = filename;
 231     if ((temp_str = strrchr(real_filename, '\\')) != NULL)
 232         real_filename = temp_str + 1;
 233     if ((temp_str = strrchr(real_filename, '/')) != NULL)
 234         real_filename = temp_str + 1;
 235     if ((temp_str = strrchr(real_filename, ':')) != NULL)
 236         real_filename = temp_str + 1;
 237
 238     sprintf(s_desc, desc_template, p_cid, p_cid, backing_file
 239             , (uint32_t)header.capacity, real_filename);
 240
 241     /* write the descriptor */
 242     if (lseek(snp_fd, 0x200, SEEK_SET) == -1)
 243         goto fail;
 244     if (write(snp_fd, s_desc, strlen(s_desc)) == -1)
 245         goto fail;
 246
 247     gd_offset = header.gd_offset * SECTOR_SIZE;     // offset of GD table
 248     rgd_offset = header.rgd_offset * SECTOR_SIZE;   // offset of RGD table
 249     capacity = header.capacity * SECTOR_SIZE;       // Extent size
 250     /*
 251      * Each GDE span 32M disk, means:
 252      * 512 GTE per GT, each GTE points to grain
 253      */
 254     gt_size = (int64_t)header.num_gtes_per_gte * header.granularity * SECTOR_SIZE;
 255     if (!gt_size)
 256         goto fail;
 257     gde_entries = (uint32_t)(capacity / gt_size);  // number of gde/rgde
 258     gd_size = gde_entries * sizeof(uint32_t);
 259
 260     /* write RGD */
 261     rgd_buf = qemu_malloc(gd_size);
 262     if (!rgd_buf)
 263         goto fail;
 264     if (lseek(p_fd, rgd_offset, SEEK_SET) == -1)
 265         goto fail_rgd;
 266     if (read(p_fd, rgd_buf, gd_size) != gd_size)
 267         goto fail_rgd;
 268     if (lseek(snp_fd, rgd_offset, SEEK_SET) == -1)
 269         goto fail_rgd;
 270     if (write(snp_fd, rgd_buf, gd_size) == -1)
 271         goto fail_rgd;
 272     qemu_free(rgd_buf);
 273
 274     /* write GD */
 275     gd_buf = qemu_malloc(gd_size);
 276     if (!gd_buf)
 277         goto fail_rgd;
 278     if (lseek(p_fd, gd_offset, SEEK_SET) == -1)
 279         goto fail_gd;
 280     if (read(p_fd, gd_buf, gd_size) != gd_size)
 281         goto fail_gd;
 282     if (lseek(snp_fd, gd_offset, SEEK_SET) == -1)
 283         goto fail_gd;
 284     if (write(snp_fd, gd_buf, gd_size) == -1)
 285         goto fail_gd;
 286     qemu_free(gd_buf);
 287
 288     close(p_fd);
 289     close(snp_fd);
 290     return 0;
 291
 292     fail_gd:
 293     qemu_free(gd_buf);
 294     fail_rgd:
 295     qemu_free(rgd_buf);
 296     fail:
 297     close(p_fd);
 298     close(snp_fd);
 299     return -1;
 300 }
 301
 302 static void vmdk_parent_close(BlockDriverState *bs)
 303 {
 304     if (bs->backing_hd)
 305         bdrv_close(bs->backing_hd);
 306 }
 307
 308
 309 static int vmdk_parent_open(BlockDriverState *bs, const char * filename)
 310 {
 311     BDRVVmdkState *s = bs->opaque;
 312     char *p_name;
 313     char desc[DESC_SIZE];
 314     char parent_img_name[1024];
 315
 316     /* the descriptor offset = 0x200 */
 317     if (bdrv_pread(s->hd, 0x200, desc, DESC_SIZE) != DESC_SIZE)
 318         return -1;
 319
 320     if ((p_name = strstr(desc,"parentFileNameHint")) != 0) {
 321         char *end_name;
 322         struct stat file_buf;
 323
 324         p_name += sizeof("parentFileNameHint") + 1;
 325         if ((end_name = strchr(p_name,'\"')) == 0)
 326             return -1;
 327
 328         strncpy(s->hd->backing_file, p_name, end_name - p_name);
 329         if (stat(s->hd->backing_file, &file_buf) != 0) {
 330             path_combine(parent_img_name, sizeof(parent_img_name),
 331                          filename, s->hd->backing_file);
 332         } else {
 333             strcpy(parent_img_name, s->hd->backing_file);
 334         }
 335
 336         s->hd->backing_hd = bdrv_new("");
 337         if (!s->hd->backing_hd) {
 338             failure:
 339             bdrv_close(s->hd);
 340             return -1;
 341         }
 342         if (bdrv_open(s->hd->backing_hd, parent_img_name, 0) < 0)
 343             goto failure;
 344     }
 345
 346     return 0;
 347 }
 348
 349 static int vmdk_open(BlockDriverState *bs, const char *filename, int flags)
 350 {
 351     BDRVVmdkState *s = bs->opaque;
 352     uint32_t magic;
 353     int l1_size, i, ret;
 354
 355     ret = bdrv_file_open(&s->hd, filename, flags);
 356     if (ret < 0)
 357         return ret;
 358     if (bdrv_pread(s->hd, 0, &magic, sizeof(magic)) != sizeof(magic))
 359         goto fail;
 360
 361     magic = be32_to_cpu(magic);
 362     if (magic == VMDK3_MAGIC) {
 363         VMDK3Header header;
 364
 365         if (bdrv_pread(s->hd, sizeof(magic), &header, sizeof(header)) != sizeof(header))
 366             goto fail;
 367         s->cluster_sectors = le32_to_cpu(header.granularity);
 368         s->l2_size = 1 << 9;
 369         s->l1_size = 1 << 6;
 370         bs->total_sectors = le32_to_cpu(header.disk_sectors);
 371         s->l1_table_offset = le32_to_cpu(header.l1dir_offset) << 9;
 372         s->l1_backup_table_offset = 0;
 373         s->l1_entry_sectors = s->l2_size * s->cluster_sectors;
 374     } else if (magic == VMDK4_MAGIC) {
 375         VMDK4Header header;
 376
 377         if (bdrv_pread(s->hd, sizeof(magic), &header, sizeof(header)) != sizeof(header))
 378             goto fail;
 379         bs->total_sectors = le64_to_cpu(header.capacity);
 380         s->cluster_sectors = le64_to_cpu(header.granularity);
 381         s->l2_size = le32_to_cpu(header.num_gtes_per_gte);
 382         s->l1_entry_sectors = s->l2_size * s->cluster_sectors;
 383         if (s->l1_entry_sectors <= 0)
 384             goto fail;
 385         s->l1_size = (bs->total_sectors + s->l1_entry_sectors - 1)
 386             / s->l1_entry_sectors;
 387         s->l1_table_offset = le64_to_cpu(header.rgd_offset) << 9;
 388         s->l1_backup_table_offset = le64_to_cpu(header.gd_offset) << 9;
 389
 390         // try to open parent images, if exist
 391         if (vmdk_parent_open(bs, filename) != 0)
 392             goto fail;
 393         // write the CID once after the image creation
 394         s->parent_cid = vmdk_read_cid(bs,1);
 395     } else {
 396         goto fail;
 397     }
 398
 399     /* read the L1 table */
 400     l1_size = s->l1_size * sizeof(uint32_t);
 401     s->l1_table = qemu_malloc(l1_size);
 402     if (!s->l1_table)
 403         goto fail;
 404     if (bdrv_pread(s->hd, s->l1_table_offset, s->l1_table, l1_size) != l1_size)
 405         goto fail;
 406     for(i = 0; i < s->l1_size; i++) {
 407         le32_to_cpus(&s->l1_table[i]);
 408     }
 409
 410     if (s->l1_backup_table_offset) {
 411         s->l1_backup_table = qemu_malloc(l1_size);
 412         if (!s->l1_backup_table)
 413             goto fail;
 414         if (bdrv_pread(s->hd, s->l1_backup_table_offset, s->l1_backup_table, l1_size) != l1_size)
 415             goto fail;
 416         for(i = 0; i < s->l1_size; i++) {
 417             le32_to_cpus(&s->l1_backup_table[i]);
 418         }
 419     }
 420
 421     s->l2_cache = qemu_malloc(s->l2_size * L2_CACHE_SIZE * sizeof(uint32_t));
 422     if (!s->l2_cache)
 423         goto fail;
 424     return 0;
 425  fail:
 426     qemu_free(s->l1_backup_table);
 427     qemu_free(s->l1_table);
 428     qemu_free(s->l2_cache);
 429     bdrv_delete(s->hd);
 430     return -1;
 431 }
 432
 433 static uint64_t get_cluster_offset(BlockDriverState *bs, uint64_t offset, int allocate);
 434
 435 static int get_whole_cluster(BlockDriverState *bs, uint64_t cluster_offset,
 436                              uint64_t offset, int allocate)
 437 {
 438     uint64_t parent_cluster_offset;
 439     BDRVVmdkState *s = bs->opaque;
 440     uint8_t  whole_grain[s->cluster_sectors*512];        // 128 sectors * 512 bytes each = grain size 64KB
 441
 442     // we will be here if it's first write on non-exist grain(cluster).
 443     // try to read from parent image, if exist
 444     if (s->hd->backing_hd) {
 445         BDRVVmdkState *ps = s->hd->backing_hd->opaque;
 446
 447         if (!vmdk_is_cid_valid(bs))
 448             return -1;
 449         parent_cluster_offset = get_cluster_offset(s->hd->backing_hd, offset, allocate);
 450         if (bdrv_pread(ps->hd, parent_cluster_offset, whole_grain, ps->cluster_sectors*512) !=
 451                                                                             ps->cluster_sectors*512)
 452             return -1;
 453
 454         if (bdrv_pwrite(s->hd, cluster_offset << 9, whole_grain, sizeof(whole_grain)) !=
 455                                                                             sizeof(whole_grain))
 456             return -1;
 457     }
 458     return 0;
 459 }
 460
 461 static uint64_t get_cluster_offset(BlockDriverState *bs,
 462                                    uint64_t offset, int allocate)
 463 {
 464     BDRVVmdkState *s = bs->opaque;
 465     unsigned int l1_index, l2_offset, l2_index;
 466     int min_index, i, j;
 467     uint32_t min_count, *l2_table, tmp;
 468     uint64_t cluster_offset;
 469
 470     l1_index = (offset >> 9) / s->l1_entry_sectors;
 471     if (l1_index >= s->l1_size)
 472         return 0;
 473     l2_offset = s->l1_table[l1_index];
 474     if (!l2_offset)
 475         return 0;
 476     for(i = 0; i < L2_CACHE_SIZE; i++) {
 477         if (l2_offset == s->l2_cache_offsets[i]) {
 478             /* increment the hit count */
 479             if (++s->l2_cache_counts[i] == 0xffffffff) {
 480                 for(j = 0; j < L2_CACHE_SIZE; j++) {
 481                     s->l2_cache_counts[j] >>= 1;
 482                 }
 483             }
 484             l2_table = s->l2_cache + (i * s->l2_size);
 485             goto found;
 486         }
 487     }
 488     /* not found: load a new entry in the least used one */
 489     min_index = 0;
 490     min_count = 0xffffffff;
 491     for(i = 0; i < L2_CACHE_SIZE; i++) {
 492         if (s->l2_cache_counts[i] < min_count) {
 493             min_count = s->l2_cache_counts[i];
 494             min_index = i;
 495         }
 496     }
 497     l2_table = s->l2_cache + (min_index * s->l2_size);
 498     if (bdrv_pread(s->hd, (int64_t)l2_offset * 512, l2_table, s->l2_size * sizeof(uint32_t)) !=
 499                                                                         s->l2_size * sizeof(uint32_t))
 500         return 0;
 501
 502     s->l2_cache_offsets[min_index] = l2_offset;
 503     s->l2_cache_counts[min_index] = 1;
 504  found:
 505     l2_index = ((offset >> 9) / s->cluster_sectors) % s->l2_size;
 506     cluster_offset = le32_to_cpu(l2_table[l2_index]);
 507     if (!cluster_offset) {
 508         struct stat file_buf;
 509
 510         if (!allocate)
 511             return 0;
 512         stat(s->hd->filename, &file_buf);
 513         cluster_offset = file_buf.st_size;
 514         bdrv_truncate(s->hd, cluster_offset + (s->cluster_sectors << 9));
 515
 516         cluster_offset >>= 9;
 517         /* update L2 table */
 518         tmp = cpu_to_le32(cluster_offset);
 519         l2_table[l2_index] = tmp;
 520         if (bdrv_pwrite(s->hd, ((int64_t)l2_offset * 512) + (l2_index * sizeof(tmp)),
 521                         &tmp, sizeof(tmp)) != sizeof(tmp))
 522             return 0;
 523         /* update backup L2 table */
 524         if (s->l1_backup_table_offset != 0) {
 525             l2_offset = s->l1_backup_table[l1_index];
 526             if (bdrv_pwrite(s->hd, ((int64_t)l2_offset * 512) + (l2_index * sizeof(tmp)),
 527                             &tmp, sizeof(tmp)) != sizeof(tmp))
 528                 return 0;
 529         }
 530
 531         if (get_whole_cluster(bs, cluster_offset, offset, allocate) == -1)
 532             return 0;
 533     }
 534     cluster_offset <<= 9;
 535     return cluster_offset;
 536 }
 537
 538 static int vmdk_is_allocated(BlockDriverState *bs, int64_t sector_num,
 539                              int nb_sectors, int *pnum)
 540 {
 541     BDRVVmdkState *s = bs->opaque;
 542     int index_in_cluster, n;
 543     uint64_t cluster_offset;
 544
 545     cluster_offset = get_cluster_offset(bs, sector_num << 9, 0);
 546     index_in_cluster = sector_num % s->cluster_sectors;
 547     n = s->cluster_sectors - index_in_cluster;
 548     if (n > nb_sectors)
 549         n = nb_sectors;
 550     *pnum = n;
 551     return (cluster_offset != 0);
 552 }
 553
 554 static int vmdk_read(BlockDriverState *bs, int64_t sector_num,
 555                     uint8_t *buf, int nb_sectors)
 556 {
 557     BDRVVmdkState *s = bs->opaque;
 558     int index_in_cluster, n, ret;
 559     uint64_t cluster_offset;
 560
 561     while (nb_sectors > 0) {
 562         cluster_offset = get_cluster_offset(bs, sector_num << 9, 0);
 563         index_in_cluster = sector_num % s->cluster_sectors;
 564         n = s->cluster_sectors - index_in_cluster;
 565         if (n > nb_sectors)
 566             n = nb_sectors;
 567         if (!cluster_offset) {
 568             // try to read from parent image, if exist
 569             if (s->hd->backing_hd) {
 570                 if (!vmdk_is_cid_valid(bs))
 571                     return -1;
 572                 ret = bdrv_read(s->hd->backing_hd, sector_num, buf, n);
 573                 if (ret < 0)
 574                     return -1;
 575             } else {
 576                 memset(buf, 0, 512 * n);
 577             }
 578         } else {
 579             if(bdrv_pread(s->hd, cluster_offset + index_in_cluster * 512, buf, n * 512) != n * 512)
 580                 return -1;
 581         }
 582         nb_sectors -= n;
 583         sector_num += n;
 584         buf += n * 512;
 585     }
 586     return 0;
 587 }
 588
 589 static int vmdk_write(BlockDriverState *bs, int64_t sector_num,
 590                      const uint8_t *buf, int nb_sectors)
 591 {
 592     BDRVVmdkState *s = bs->opaque;
 593     int index_in_cluster, n;
 594     uint64_t cluster_offset;
 595     static int cid_update = 0;
 596
 597     while (nb_sectors > 0) {
 598         index_in_cluster = sector_num & (s->cluster_sectors - 1);
 599         n = s->cluster_sectors - index_in_cluster;
 600         if (n > nb_sectors)
 601             n = nb_sectors;
 602         cluster_offset = get_cluster_offset(bs, sector_num << 9, 1);
 603         if (!cluster_offset)
 604             return -1;
 605         if (bdrv_pwrite(s->hd, cluster_offset + index_in_cluster * 512, buf, n * 512) != n * 512)
 606             return -1;
 607         nb_sectors -= n;
 608         sector_num += n;
 609         buf += n * 512;
 610
 611         // update CID on the first write every time the virtual disk is opened
 612         if (!cid_update) {
 613             vmdk_write_cid(bs, time(NULL));
 614             cid_update++;
 615         }
 616     }
 617     return 0;
 618 }
 619
 620 static int vmdk_create(const char *filename, int64_t total_size,
 621                        const char *backing_file, int flags)
 622 {
 623     int fd, i;
 624     VMDK4Header header;
 625     uint32_t tmp, magic, grains, gd_size, gt_size, gt_count;
 626     char *desc_template =
 627         "# Disk DescriptorFile\n"
 628         "version=1\n"
 629         "CID=%x\n"
 630         "parentCID=ffffffff\n"
 631         "createType=\"monolithicSparse\"\n"
 632         "\n"
 633         "# Extent description\n"
 634         "RW %lu SPARSE \"%s\"\n"
 635         "\n"
 636         "# The Disk Data Base \n"
 637         "#DDB\n"
 638         "\n"
 639         "ddb.virtualHWVersion = \"4\"\n"
 640         "ddb.geometry.cylinders = \"%lu\"\n"
 641         "ddb.geometry.heads = \"16\"\n"
 642         "ddb.geometry.sectors = \"63\"\n"
 643         "ddb.adapterType = \"ide\"\n";
 644     char desc[1024];
 645     const char *real_filename, *temp_str;
 646
 647     /* XXX: add support for backing file */
 648     if (backing_file) {
 649         return vmdk_snapshot_create(filename, backing_file);
 650     }
 651
 652     fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY | O_LARGEFILE,
 653               0644);
 654     if (fd < 0)
 655         return -1;
 656     magic = cpu_to_be32(VMDK4_MAGIC);
 657     memset(&header, 0, sizeof(header));
 658     header.version = cpu_to_le32(1);
 659     header.flags = cpu_to_le32(3); /* ?? */
 660     header.capacity = cpu_to_le64(total_size);
 661     header.granularity = cpu_to_le64(128);
 662     header.num_gtes_per_gte = cpu_to_le32(512);
 663
 664     grains = (total_size + header.granularity - 1) / header.granularity;
 665     gt_size = ((header.num_gtes_per_gte * sizeof(uint32_t)) + 511) >> 9;
 666     gt_count = (grains + header.num_gtes_per_gte - 1) / header.num_gtes_per_gte;
 667     gd_size = (gt_count * sizeof(uint32_t) + 511) >> 9;
 668
 669     header.desc_offset = 1;
 670     header.desc_size = 20;
 671     header.rgd_offset = header.desc_offset + header.desc_size;
 672     header.gd_offset = header.rgd_offset + gd_size + (gt_size * gt_count);
 673     header.grain_offset =
 674        ((header.gd_offset + gd_size + (gt_size * gt_count) +
 675          header.granularity - 1) / header.granularity) *
 676         header.granularity;
 677
 678     header.desc_offset = cpu_to_le64(header.desc_offset);
 679     header.desc_size = cpu_to_le64(header.desc_size);
 680     header.rgd_offset = cpu_to_le64(header.rgd_offset);
 681     header.gd_offset = cpu_to_le64(header.gd_offset);
 682     header.grain_offset = cpu_to_le64(header.grain_offset);
 683
 684     header.check_bytes[0] = 0xa;
 685     header.check_bytes[1] = 0x20;
 686     header.check_bytes[2] = 0xd;
 687     header.check_bytes[3] = 0xa;
 688
 689     /* write all the data */
 690     write(fd, &magic, sizeof(magic));
 691     write(fd, &header, sizeof(header));
 692
 693     ftruncate(fd, header.grain_offset << 9);
 694
 695     /* write grain directory */
 696     lseek(fd, le64_to_cpu(header.rgd_offset) << 9, SEEK_SET);
 697     for (i = 0, tmp = header.rgd_offset + gd_size;
 698          i < gt_count; i++, tmp += gt_size)
 699         write(fd, &tmp, sizeof(tmp));
 700
 701     /* write backup grain directory */
 702     lseek(fd, le64_to_cpu(header.gd_offset) << 9, SEEK_SET);
 703     for (i = 0, tmp = header.gd_offset + gd_size;
 704          i < gt_count; i++, tmp += gt_size)
 705         write(fd, &tmp, sizeof(tmp));
 706
 707     /* compose the descriptor */
 708     real_filename = filename;
 709     if ((temp_str = strrchr(real_filename, '\\')) != NULL)
 710         real_filename = temp_str + 1;
 711     if ((temp_str = strrchr(real_filename, '/')) != NULL)
 712         real_filename = temp_str + 1;
 713     if ((temp_str = strrchr(real_filename, ':')) != NULL)
 714         real_filename = temp_str + 1;
 715     sprintf(desc, desc_template, time(NULL), (unsigned long)total_size,
 716             real_filename, total_size / (63 * 16));
 717
 718     /* write the descriptor */
 719     lseek(fd, le64_to_cpu(header.desc_offset) << 9, SEEK_SET);
 720     write(fd, desc, strlen(desc));
 721
 722     close(fd);
 723     return 0;
 724 }
 725
 726 static void vmdk_close(BlockDriverState *bs)
 727 {
 728     BDRVVmdkState *s = bs->opaque;
 729
 730     qemu_free(s->l1_table);
 731     qemu_free(s->l2_cache);
 732     bdrv_delete(s->hd);
 733     // try to close parent image, if exist
 734     vmdk_parent_close(s->hd);
 735 }
 736
 737 static void vmdk_flush(BlockDriverState *bs)
 738 {
 739     BDRVVmdkState *s = bs->opaque;
 740     bdrv_flush(s->hd);
 741 }
 742
 743 BlockDriver bdrv_vmdk = {
 744     "vmdk",
 745     sizeof(BDRVVmdkState),
 746     vmdk_probe,
 747     vmdk_open,
 748     vmdk_read,
 749     vmdk_write,
 750     vmdk_close,
 751     vmdk_create,
 752     vmdk_flush,
 753     vmdk_is_allocated,
 754 };