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
  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     int is_parent;
  79 } BDRVVmdkState;
  80
  81 typedef struct VmdkMetaData {
  82     uint32_t offset;
  83     unsigned int l1_index;
  84     unsigned int l2_index;
  85     unsigned int l2_offset;
  86     int valid;
  87 } VmdkMetaData;
  88
  89 typedef struct ActiveBDRVState{
  90     BlockDriverState *hd;            // active image handler
  91     uint64_t cluster_offset;         // current write offset
  92 }ActiveBDRVState;
  93
  94 static ActiveBDRVState activeBDRV;
  95
  96 static int vmdk_probe(const uint8_t *buf, int buf_size, const char *filename)
  97 {
  98     uint32_t magic;
  99
 100     if (buf_size < 4)
 101         return 0;
 102     magic = be32_to_cpu(*(uint32_t *)buf);
 103     if (magic == VMDK3_MAGIC ||
 104         magic == VMDK4_MAGIC)
 105         return 100;
 106     else
 107         return 0;
 108 }
 109
 110 #define CHECK_CID 1
 111
 112 #define SECTOR_SIZE 512
 113 #define DESC_SIZE 20*SECTOR_SIZE        // 20 sectors of 512 bytes each
 114 #define HEADER_SIZE 512                         // first sector of 512 bytes
 115
 116 static uint32_t vmdk_read_cid(BlockDriverState *bs, int parent)
 117 {
 118     BDRVVmdkState *s = bs->opaque;
 119     char desc[DESC_SIZE];
 120     uint32_t cid;
 121     char *p_name, *cid_str;
 122     size_t cid_str_size;
 123
 124     /* the descriptor offset = 0x200 */
 125     if (bdrv_pread(s->hd, 0x200, desc, DESC_SIZE) != DESC_SIZE)
 126         return 0;
 127
 128     if (parent) {
 129         cid_str = "parentCID";
 130         cid_str_size = sizeof("parentCID");
 131     } else {
 132         cid_str = "CID";
 133         cid_str_size = sizeof("CID");
 134     }
 135
 136     if ((p_name = strstr(desc,cid_str)) != 0) {
 137         p_name += cid_str_size;
 138         sscanf(p_name,"%x",&cid);
 139     }
 140
 141     return cid;
 142 }
 143
 144 static int vmdk_write_cid(BlockDriverState *bs, uint32_t cid)
 145 {
 146     BDRVVmdkState *s = bs->opaque;
 147     char desc[DESC_SIZE], tmp_desc[DESC_SIZE];
 148     char *p_name, *tmp_str;
 149
 150     /* the descriptor offset = 0x200 */
 151     if (bdrv_pread(s->hd, 0x200, desc, DESC_SIZE) != DESC_SIZE)
 152         return -1;
 153
 154     tmp_str = strstr(desc,"parentCID");
 155     strcpy(tmp_desc, tmp_str);
 156     if ((p_name = strstr(desc,"CID")) != 0) {
 157         p_name += sizeof("CID");
 158         sprintf(p_name,"%x\n",cid);
 159         strcat(desc,tmp_desc);
 160     }
 161
 162     if (bdrv_pwrite(s->hd, 0x200, desc, DESC_SIZE) != DESC_SIZE)
 163         return -1;
 164     return 0;
 165 }
 166
 167 static int vmdk_is_cid_valid(BlockDriverState *bs)
 168 {
 169 #ifdef CHECK_CID
 170     BDRVVmdkState *s = bs->opaque;
 171     BlockDriverState *p_bs = s->hd->backing_hd;
 172     uint32_t cur_pcid;
 173
 174     if (p_bs) {
 175         cur_pcid = vmdk_read_cid(p_bs,0);
 176         if (s->parent_cid != cur_pcid)
 177             // CID not valid
 178             return 0;
 179     }
 180 #endif
 181     // CID valid
 182     return 1;
 183 }
 184
 185 static int vmdk_snapshot_create(const char *filename, const char *backing_file)
 186 {
 187     int snp_fd, p_fd;
 188     uint32_t p_cid;
 189     char *p_name, *gd_buf, *rgd_buf;
 190     const char *real_filename, *temp_str;
 191     VMDK4Header header;
 192     uint32_t gde_entries, gd_size;
 193     int64_t gd_offset, rgd_offset, capacity, gt_size;
 194     char p_desc[DESC_SIZE], s_desc[DESC_SIZE], hdr[HEADER_SIZE];
 195     char *desc_template =
 196     "# Disk DescriptorFile\n"
 197     "version=1\n"
 198     "CID=%x\n"
 199     "parentCID=%x\n"
 200     "createType=\"monolithicSparse\"\n"
 201     "parentFileNameHint=\"%s\"\n"
 202     "\n"
 203     "# Extent description\n"
 204     "RW %lu SPARSE \"%s\"\n"
 205     "\n"
 206     "# The Disk Data Base \n"
 207     "#DDB\n"
 208     "\n";
 209
 210     snp_fd = open(filename, O_RDWR | O_CREAT | O_TRUNC | O_BINARY | O_LARGEFILE, 0644);
 211     if (snp_fd < 0)
 212         return -1;
 213     p_fd = open(backing_file, O_RDONLY | O_BINARY | O_LARGEFILE);
 214     if (p_fd < 0) {
 215         close(snp_fd);
 216         return -1;
 217     }
 218
 219     /* read the header */
 220     if (lseek(p_fd, 0x0, SEEK_SET) == -1)
 221         goto fail;
 222     if (read(p_fd, hdr, HEADER_SIZE) != HEADER_SIZE)
 223         goto fail;
 224
 225     /* write the header */
 226     if (lseek(snp_fd, 0x0, SEEK_SET) == -1)
 227         goto fail;
 228     if (write(snp_fd, hdr, HEADER_SIZE) == -1)
 229         goto fail;
 230
 231     memset(&header, 0, sizeof(header));
 232     memcpy(&header,&hdr[4], sizeof(header)); // skip the VMDK4_MAGIC
 233
 234     ftruncate(snp_fd, header.grain_offset << 9);
 235     /* the descriptor offset = 0x200 */
 236     if (lseek(p_fd, 0x200, SEEK_SET) == -1)
 237         goto fail;
 238     if (read(p_fd, p_desc, DESC_SIZE) != DESC_SIZE)
 239         goto fail;
 240
 241     if ((p_name = strstr(p_desc,"CID")) != 0) {
 242         p_name += sizeof("CID");
 243         sscanf(p_name,"%x",&p_cid);
 244     }
 245
 246     real_filename = filename;
 247     if ((temp_str = strrchr(real_filename, '\\')) != NULL)
 248         real_filename = temp_str + 1;
 249     if ((temp_str = strrchr(real_filename, '/')) != NULL)
 250         real_filename = temp_str + 1;
 251     if ((temp_str = strrchr(real_filename, ':')) != NULL)
 252         real_filename = temp_str + 1;
 253
 254     sprintf(s_desc, desc_template, p_cid, p_cid, backing_file
 255             , (uint32_t)header.capacity, real_filename);
 256
 257     /* write the descriptor */
 258     if (lseek(snp_fd, 0x200, SEEK_SET) == -1)
 259         goto fail;
 260     if (write(snp_fd, s_desc, strlen(s_desc)) == -1)
 261         goto fail;
 262
 263     gd_offset = header.gd_offset * SECTOR_SIZE;     // offset of GD table
 264     rgd_offset = header.rgd_offset * SECTOR_SIZE;   // offset of RGD table
 265     capacity = header.capacity * SECTOR_SIZE;       // Extent size
 266     /*
 267      * Each GDE span 32M disk, means:
 268      * 512 GTE per GT, each GTE points to grain
 269      */
 270     gt_size = (int64_t)header.num_gtes_per_gte * header.granularity * SECTOR_SIZE;
 271     if (!gt_size)
 272         goto fail;
 273     gde_entries = (uint32_t)(capacity / gt_size);  // number of gde/rgde
 274     gd_size = gde_entries * sizeof(uint32_t);
 275
 276     /* write RGD */
 277     rgd_buf = qemu_malloc(gd_size);
 278     if (!rgd_buf)
 279         goto fail;
 280     if (lseek(p_fd, rgd_offset, SEEK_SET) == -1)
 281         goto fail_rgd;
 282     if (read(p_fd, rgd_buf, gd_size) != gd_size)
 283         goto fail_rgd;
 284     if (lseek(snp_fd, rgd_offset, SEEK_SET) == -1)
 285         goto fail_rgd;
 286     if (write(snp_fd, rgd_buf, gd_size) == -1)
 287         goto fail_rgd;
 288     qemu_free(rgd_buf);
 289
 290     /* write GD */
 291     gd_buf = qemu_malloc(gd_size);
 292     if (!gd_buf)
 293         goto fail_rgd;
 294     if (lseek(p_fd, gd_offset, SEEK_SET) == -1)
 295         goto fail_gd;
 296     if (read(p_fd, gd_buf, gd_size) != gd_size)
 297         goto fail_gd;
 298     if (lseek(snp_fd, gd_offset, SEEK_SET) == -1)
 299         goto fail_gd;
 300     if (write(snp_fd, gd_buf, gd_size) == -1)
 301         goto fail_gd;
 302     qemu_free(gd_buf);
 303
 304     close(p_fd);
 305     close(snp_fd);
 306     return 0;
 307
 308     fail_gd:
 309     qemu_free(gd_buf);
 310     fail_rgd:
 311     qemu_free(rgd_buf);
 312     fail:
 313     close(p_fd);
 314     close(snp_fd);
 315     return -1;
 316 }
 317
 318 static void vmdk_parent_close(BlockDriverState *bs)
 319 {
 320     if (bs->backing_hd)
 321         bdrv_close(bs->backing_hd);
 322 }
 323
 324 int parent_open = 0;
 325 static int vmdk_parent_open(BlockDriverState *bs, const char * filename)
 326 {
 327     BDRVVmdkState *s = bs->opaque;
 328     char *p_name;
 329     char desc[DESC_SIZE];
 330     char parent_img_name[1024];
 331
 332     /* the descriptor offset = 0x200 */
 333     if (bdrv_pread(s->hd, 0x200, desc, DESC_SIZE) != DESC_SIZE)
 334         return -1;
 335
 336     if ((p_name = strstr(desc,"parentFileNameHint")) != 0) {
 337         char *end_name;
 338         struct stat file_buf;
 339
 340         p_name += sizeof("parentFileNameHint") + 1;
 341         if ((end_name = strchr(p_name,'\"')) == 0)
 342             return -1;
 343
 344         strncpy(s->hd->backing_file, p_name, end_name - p_name);
 345         if (stat(s->hd->backing_file, &file_buf) != 0) {
 346             path_combine(parent_img_name, sizeof(parent_img_name),
 347                          filename, s->hd->backing_file);
 348         } else {
 349             strcpy(parent_img_name, s->hd->backing_file);
 350         }
 351
 352         s->hd->backing_hd = bdrv_new("");
 353         if (!s->hd->backing_hd) {
 354             failure:
 355             bdrv_close(s->hd);
 356             return -1;
 357         }
 358         parent_open = 1;
 359         if (bdrv_open(s->hd->backing_hd, parent_img_name, BDRV_O_RDONLY) < 0)
 360             goto failure;
 361         parent_open = 0;
 362     }
 363
 364     return 0;
 365 }
 366
 367 static int vmdk_open(BlockDriverState *bs, const char *filename, int flags)
 368 {
 369     BDRVVmdkState *s = bs->opaque;
 370     uint32_t magic;
 371     int l1_size, i, ret;
 372
 373     if (parent_open)
 374         // Parent must be opened as RO.
 375         flags = BDRV_O_RDONLY;
 376     fprintf(stderr, "(VMDK) image open: flags=0x%x filename=%s\n", flags, bs->filename);
 377
 378     ret = bdrv_file_open(&s->hd, filename, flags);
 379     if (ret < 0)
 380         return ret;
 381     if (bdrv_pread(s->hd, 0, &magic, sizeof(magic)) != sizeof(magic))
 382         goto fail;
 383
 384     magic = be32_to_cpu(magic);
 385     if (magic == VMDK3_MAGIC) {
 386         VMDK3Header header;
 387
 388         if (bdrv_pread(s->hd, sizeof(magic), &header, sizeof(header)) != sizeof(header))
 389             goto fail;
 390         s->cluster_sectors = le32_to_cpu(header.granularity);
 391         s->l2_size = 1 << 9;
 392         s->l1_size = 1 << 6;
 393         bs->total_sectors = le32_to_cpu(header.disk_sectors);
 394         s->l1_table_offset = le32_to_cpu(header.l1dir_offset) << 9;
 395         s->l1_backup_table_offset = 0;
 396         s->l1_entry_sectors = s->l2_size * s->cluster_sectors;
 397     } else if (magic == VMDK4_MAGIC) {
 398         VMDK4Header header;
 399
 400         if (bdrv_pread(s->hd, sizeof(magic), &header, sizeof(header)) != sizeof(header))
 401             goto fail;
 402         bs->total_sectors = le64_to_cpu(header.capacity);
 403         s->cluster_sectors = le64_to_cpu(header.granularity);
 404         s->l2_size = le32_to_cpu(header.num_gtes_per_gte);
 405         s->l1_entry_sectors = s->l2_size * s->cluster_sectors;
 406         if (s->l1_entry_sectors <= 0)
 407             goto fail;
 408         s->l1_size = (bs->total_sectors + s->l1_entry_sectors - 1)
 409             / s->l1_entry_sectors;
 410         s->l1_table_offset = le64_to_cpu(header.rgd_offset) << 9;
 411         s->l1_backup_table_offset = le64_to_cpu(header.gd_offset) << 9;
 412
 413         if (parent_open)
 414             s->is_parent = 1;
 415         else
 416             s->is_parent = 0;
 417
 418         // try to open parent images, if exist
 419         if (vmdk_parent_open(bs, filename) != 0)
 420             goto fail;
 421         // write the CID once after the image creation
 422         s->parent_cid = vmdk_read_cid(bs,1);
 423     } else {
 424         goto fail;
 425     }
 426
 427     /* read the L1 table */
 428     l1_size = s->l1_size * sizeof(uint32_t);
 429     s->l1_table = qemu_malloc(l1_size);
 430     if (!s->l1_table)
 431         goto fail;
 432     if (bdrv_pread(s->hd, s->l1_table_offset, s->l1_table, l1_size) != l1_size)
 433         goto fail;
 434     for(i = 0; i < s->l1_size; i++) {
 435         le32_to_cpus(&s->l1_table[i]);
 436     }
 437
 438     if (s->l1_backup_table_offset) {
 439         s->l1_backup_table = qemu_malloc(l1_size);
 440         if (!s->l1_backup_table)
 441             goto fail;
 442         if (bdrv_pread(s->hd, s->l1_backup_table_offset, s->l1_backup_table, l1_size) != l1_size)
 443             goto fail;
 444         for(i = 0; i < s->l1_size; i++) {
 445             le32_to_cpus(&s->l1_backup_table[i]);
 446         }
 447     }
 448
 449     s->l2_cache = qemu_malloc(s->l2_size * L2_CACHE_SIZE * sizeof(uint32_t));
 450     if (!s->l2_cache)
 451         goto fail;
 452     return 0;
 453  fail:
 454     qemu_free(s->l1_backup_table);
 455     qemu_free(s->l1_table);
 456     qemu_free(s->l2_cache);
 457     bdrv_delete(s->hd);
 458     return -1;
 459 }
 460
 461 static uint64_t get_cluster_offset(BlockDriverState *bs, VmdkMetaData *m_data,
 462                                    uint64_t offset, int allocate);
 463
 464 static int get_whole_cluster(BlockDriverState *bs, uint64_t cluster_offset,
 465                              uint64_t offset, int allocate)
 466 {
 467     uint64_t parent_cluster_offset;
 468     BDRVVmdkState *s = bs->opaque;
 469     uint8_t  whole_grain[s->cluster_sectors*512];        // 128 sectors * 512 bytes each = grain size 64KB
 470
 471     // we will be here if it's first write on non-exist grain(cluster).
 472     // try to read from parent image, if exist
 473     if (s->hd->backing_hd) {
 474         BDRVVmdkState *ps = s->hd->backing_hd->opaque;
 475
 476         if (!vmdk_is_cid_valid(bs))
 477             return -1;
 478
 479         parent_cluster_offset = get_cluster_offset(s->hd->backing_hd, NULL, offset, allocate);
 480
 481         if (parent_cluster_offset) {
 482             BDRVVmdkState *act_s = activeBDRV.hd->opaque;
 483
 484             if (bdrv_pread(ps->hd, parent_cluster_offset, whole_grain, ps->cluster_sectors*512) != ps->cluster_sectors*512)
 485                 return -1;
 486
 487             //Write grain only into the active image
 488             if (bdrv_pwrite(act_s->hd, activeBDRV.cluster_offset << 9, whole_grain, sizeof(whole_grain)) != sizeof(whole_grain))
 489                 return -1;
 490         }
 491     }
 492     return 0;
 493 }
 494
 495 static int vmdk_L2update(BlockDriverState *bs, VmdkMetaData *m_data)
 496 {
 497     BDRVVmdkState *s = bs->opaque;
 498
 499     /* update L2 table */
 500     if (bdrv_pwrite(s->hd, ((int64_t)m_data->l2_offset * 512) + (m_data->l2_index * sizeof(m_data->offset)),
 501                     &(m_data->offset), sizeof(m_data->offset)) != sizeof(m_data->offset))
 502         return -1;
 503     /* update backup L2 table */
 504     if (s->l1_backup_table_offset != 0) {
 505         m_data->l2_offset = s->l1_backup_table[m_data->l1_index];
 506         if (bdrv_pwrite(s->hd, ((int64_t)m_data->l2_offset * 512) + (m_data->l2_index * sizeof(m_data->offset)),
 507                         &(m_data->offset), sizeof(m_data->offset)) != sizeof(m_data->offset))
 508             return -1;
 509     }
 510
 511     return 0;
 512 }
 513
 514 static uint64_t get_cluster_offset(BlockDriverState *bs, VmdkMetaData *m_data,
 515                                    uint64_t offset, int allocate)
 516 {
 517     BDRVVmdkState *s = bs->opaque;
 518     unsigned int l1_index, l2_offset, l2_index;
 519     int min_index, i, j;
 520     uint32_t min_count, *l2_table, tmp = 0;
 521     uint64_t cluster_offset;
 522
 523     if (m_data)
 524         m_data->valid = 0;
 525
 526     l1_index = (offset >> 9) / s->l1_entry_sectors;
 527     if (l1_index >= s->l1_size)
 528         return 0;
 529     l2_offset = s->l1_table[l1_index];
 530     if (!l2_offset)
 531         return 0;
 532     for(i = 0; i < L2_CACHE_SIZE; i++) {
 533         if (l2_offset == s->l2_cache_offsets[i]) {
 534             /* increment the hit count */
 535             if (++s->l2_cache_counts[i] == 0xffffffff) {
 536                 for(j = 0; j < L2_CACHE_SIZE; j++) {
 537                     s->l2_cache_counts[j] >>= 1;
 538                 }
 539             }
 540             l2_table = s->l2_cache + (i * s->l2_size);
 541             goto found;
 542         }
 543     }
 544     /* not found: load a new entry in the least used one */
 545     min_index = 0;
 546     min_count = 0xffffffff;
 547     for(i = 0; i < L2_CACHE_SIZE; i++) {
 548         if (s->l2_cache_counts[i] < min_count) {
 549             min_count = s->l2_cache_counts[i];
 550             min_index = i;
 551         }
 552     }
 553     l2_table = s->l2_cache + (min_index * s->l2_size);
 554     if (bdrv_pread(s->hd, (int64_t)l2_offset * 512, l2_table, s->l2_size * sizeof(uint32_t)) !=
 555                                                                         s->l2_size * sizeof(uint32_t))
 556         return 0;
 557
 558     s->l2_cache_offsets[min_index] = l2_offset;
 559     s->l2_cache_counts[min_index] = 1;
 560  found:
 561     l2_index = ((offset >> 9) / s->cluster_sectors) % s->l2_size;
 562     cluster_offset = le32_to_cpu(l2_table[l2_index]);
 563
 564     if (!cluster_offset) {
 565         if (!allocate)
 566             return 0;
 567         // Avoid the L2 tables update for the images that have snapshots.
 568         if (!s->is_parent) {
 569             cluster_offset = bdrv_getlength(s->hd);
 570             bdrv_truncate(s->hd, cluster_offset + (s->cluster_sectors << 9));
 571
 572             cluster_offset >>= 9;
 573             tmp = cpu_to_le32(cluster_offset);
 574             l2_table[l2_index] = tmp;
 575             // Save the active image state
 576             activeBDRV.cluster_offset = cluster_offset;
 577             activeBDRV.hd = bs;
 578         }
 579         /* First of all we write grain itself, to avoid race condition
 580          * that may to corrupt the image.
 581          * This problem may occur because of insufficient space on host disk
 582          * or inappropriate VM shutdown.
 583          */
 584         if (get_whole_cluster(bs, cluster_offset, offset, allocate) == -1)
 585             return 0;
 586
 587         if (m_data) {
 588             m_data->offset = tmp;
 589             m_data->l1_index = l1_index;
 590             m_data->l2_index = l2_index;
 591             m_data->l2_offset = l2_offset;
 592             m_data->valid = 1;
 593         }
 594     }
 595     cluster_offset <<= 9;
 596     return cluster_offset;
 597 }
 598
 599 static int vmdk_is_allocated(BlockDriverState *bs, int64_t sector_num,
 600                              int nb_sectors, int *pnum)
 601 {
 602     BDRVVmdkState *s = bs->opaque;
 603     int index_in_cluster, n;
 604     uint64_t cluster_offset;
 605
 606     cluster_offset = get_cluster_offset(bs, NULL, sector_num << 9, 0);
 607     index_in_cluster = sector_num % s->cluster_sectors;
 608     n = s->cluster_sectors - index_in_cluster;
 609     if (n > nb_sectors)
 610         n = nb_sectors;
 611     *pnum = n;
 612     return (cluster_offset != 0);
 613 }
 614
 615 static int vmdk_read(BlockDriverState *bs, int64_t sector_num,
 616                     uint8_t *buf, int nb_sectors)
 617 {
 618     BDRVVmdkState *s = bs->opaque;
 619     int index_in_cluster, n, ret;
 620     uint64_t cluster_offset;
 621
 622     while (nb_sectors > 0) {
 623         cluster_offset = get_cluster_offset(bs, NULL, sector_num << 9, 0);
 624         index_in_cluster = sector_num % s->cluster_sectors;
 625         n = s->cluster_sectors - index_in_cluster;
 626         if (n > nb_sectors)
 627             n = nb_sectors;
 628         if (!cluster_offset) {
 629             // try to read from parent image, if exist
 630             if (s->hd->backing_hd) {
 631                 if (!vmdk_is_cid_valid(bs))
 632                     return -1;
 633                 ret = bdrv_read(s->hd->backing_hd, sector_num, buf, n);
 634                 if (ret < 0)
 635                     return -1;
 636             } else {
 637                 memset(buf, 0, 512 * n);
 638             }
 639         } else {
 640             if(bdrv_pread(s->hd, cluster_offset + index_in_cluster * 512, buf, n * 512) != n * 512)
 641                 return -1;
 642         }
 643         nb_sectors -= n;
 644         sector_num += n;
 645         buf += n * 512;
 646     }
 647     return 0;
 648 }
 649
 650 static int vmdk_write(BlockDriverState *bs, int64_t sector_num,
 651                      const uint8_t *buf, int nb_sectors)
 652 {
 653     BDRVVmdkState *s = bs->opaque;
 654     VmdkMetaData m_data;
 655     int index_in_cluster, n;
 656     uint64_t cluster_offset;
 657     static int cid_update = 0;
 658
 659     if (sector_num > bs->total_sectors) {
 660         fprintf(stderr,
 661                 "(VMDK) Wrong offset: sector_num=0x%" PRIx64
 662                 " total_sectors=0x%" PRIx64 "\n",
 663                 sector_num, bs->total_sectors);
 664         return -1;
 665     }
 666
 667     while (nb_sectors > 0) {
 668         index_in_cluster = sector_num & (s->cluster_sectors - 1);
 669         n = s->cluster_sectors - index_in_cluster;
 670         if (n > nb_sectors)
 671             n = nb_sectors;
 672         cluster_offset = get_cluster_offset(bs, &m_data, sector_num << 9, 1);
 673         if (!cluster_offset)
 674             return -1;
 675
 676         if (bdrv_pwrite(s->hd, cluster_offset + index_in_cluster * 512, buf, n * 512) != n * 512)
 677             return -1;
 678         if (m_data.valid) {
 679             /* update L2 tables */
 680             if (vmdk_L2update(bs, &m_data) == -1)
 681                 return -1;
 682         }
 683         nb_sectors -= n;
 684         sector_num += n;
 685         buf += n * 512;
 686
 687         // update CID on the first write every time the virtual disk is opened
 688         if (!cid_update) {
 689             vmdk_write_cid(bs, time(NULL));
 690             cid_update++;
 691         }
 692     }
 693     return 0;
 694 }
 695
 696 static int vmdk_create(const char *filename, int64_t total_size,
 697                        const char *backing_file, int flags)
 698 {
 699     int fd, i;
 700     VMDK4Header header;
 701     uint32_t tmp, magic, grains, gd_size, gt_size, gt_count;
 702     char *desc_template =
 703         "# Disk DescriptorFile\n"
 704         "version=1\n"
 705         "CID=%x\n"
 706         "parentCID=ffffffff\n"
 707         "createType=\"monolithicSparse\"\n"
 708         "\n"
 709         "# Extent description\n"
 710         "RW %lu SPARSE \"%s\"\n"
 711         "\n"
 712         "# The Disk Data Base \n"
 713         "#DDB\n"
 714         "\n"
 715         "ddb.virtualHWVersion = \"%d\"\n"
 716         "ddb.geometry.cylinders = \"%lu\"\n"
 717         "ddb.geometry.heads = \"16\"\n"
 718         "ddb.geometry.sectors = \"63\"\n"
 719         "ddb.adapterType = \"ide\"\n";
 720     char desc[1024];
 721     const char *real_filename, *temp_str;
 722
 723     /* XXX: add support for backing file */
 724     if (backing_file) {
 725         return vmdk_snapshot_create(filename, backing_file);
 726     }
 727
 728     fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY | O_LARGEFILE,
 729               0644);
 730     if (fd < 0)
 731         return -1;
 732     magic = cpu_to_be32(VMDK4_MAGIC);
 733     memset(&header, 0, sizeof(header));
 734     header.version = cpu_to_le32(1);
 735     header.flags = cpu_to_le32(3); /* ?? */
 736     header.capacity = cpu_to_le64(total_size);
 737     header.granularity = cpu_to_le64(128);
 738     header.num_gtes_per_gte = cpu_to_le32(512);
 739
 740     grains = (total_size + header.granularity - 1) / header.granularity;
 741     gt_size = ((header.num_gtes_per_gte * sizeof(uint32_t)) + 511) >> 9;
 742     gt_count = (grains + header.num_gtes_per_gte - 1) / header.num_gtes_per_gte;
 743     gd_size = (gt_count * sizeof(uint32_t) + 511) >> 9;
 744
 745     header.desc_offset = 1;
 746     header.desc_size = 20;
 747     header.rgd_offset = header.desc_offset + header.desc_size;
 748     header.gd_offset = header.rgd_offset + gd_size + (gt_size * gt_count);
 749     header.grain_offset =
 750        ((header.gd_offset + gd_size + (gt_size * gt_count) +
 751          header.granularity - 1) / header.granularity) *
 752         header.granularity;
 753
 754     header.desc_offset = cpu_to_le64(header.desc_offset);
 755     header.desc_size = cpu_to_le64(header.desc_size);
 756     header.rgd_offset = cpu_to_le64(header.rgd_offset);
 757     header.gd_offset = cpu_to_le64(header.gd_offset);
 758     header.grain_offset = cpu_to_le64(header.grain_offset);
 759
 760     header.check_bytes[0] = 0xa;
 761     header.check_bytes[1] = 0x20;
 762     header.check_bytes[2] = 0xd;
 763     header.check_bytes[3] = 0xa;
 764
 765     /* write all the data */
 766     write(fd, &magic, sizeof(magic));
 767     write(fd, &header, sizeof(header));
 768
 769     ftruncate(fd, header.grain_offset << 9);
 770
 771     /* write grain directory */
 772     lseek(fd, le64_to_cpu(header.rgd_offset) << 9, SEEK_SET);
 773     for (i = 0, tmp = header.rgd_offset + gd_size;
 774          i < gt_count; i++, tmp += gt_size)
 775         write(fd, &tmp, sizeof(tmp));
 776
 777     /* write backup grain directory */
 778     lseek(fd, le64_to_cpu(header.gd_offset) << 9, SEEK_SET);
 779     for (i = 0, tmp = header.gd_offset + gd_size;
 780          i < gt_count; i++, tmp += gt_size)
 781         write(fd, &tmp, sizeof(tmp));
 782
 783     /* compose the descriptor */
 784     real_filename = filename;
 785     if ((temp_str = strrchr(real_filename, '\\')) != NULL)
 786         real_filename = temp_str + 1;
 787     if ((temp_str = strrchr(real_filename, '/')) != NULL)
 788         real_filename = temp_str + 1;
 789     if ((temp_str = strrchr(real_filename, ':')) != NULL)
 790         real_filename = temp_str + 1;
 791     sprintf(desc, desc_template, time(NULL), (unsigned long)total_size,
 792             real_filename, (flags & BLOCK_FLAG_COMPAT6 ? 6 : 4), total_size / (63 * 16));
 793
 794     /* write the descriptor */
 795     lseek(fd, le64_to_cpu(header.desc_offset) << 9, SEEK_SET);
 796     write(fd, desc, strlen(desc));
 797
 798     close(fd);
 799     return 0;
 800 }
 801
 802 static void vmdk_close(BlockDriverState *bs)
 803 {
 804     BDRVVmdkState *s = bs->opaque;
 805
 806     qemu_free(s->l1_table);
 807     qemu_free(s->l2_cache);
 808     bdrv_delete(s->hd);
 809     // try to close parent image, if exist
 810     vmdk_parent_close(s->hd);
 811 }
 812
 813 static void vmdk_flush(BlockDriverState *bs)
 814 {
 815     BDRVVmdkState *s = bs->opaque;
 816     bdrv_flush(s->hd);
 817 }
 818
 819 BlockDriver bdrv_vmdk = {
 820     "vmdk",
 821     sizeof(BDRVVmdkState),
 822     vmdk_probe,
 823     vmdk_open,
 824     vmdk_read,
 825     vmdk_write,
 826     vmdk_close,
 827     vmdk_create,
 828     vmdk_flush,
 829     vmdk_is_allocated,
 830 };