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     const 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)) != NULL) {
 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     pstrcpy(tmp_desc, sizeof(tmp_desc), tmp_str);
 156     if ((p_name = strstr(desc,"CID")) != NULL) {
 157         p_name += sizeof("CID");
 158         snprintf(p_name, sizeof(desc) - (p_name - desc), "%x\n", cid);
 159         pstrcat(desc, sizeof(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     static const 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 %u 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")) != NULL) {
 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     snprintf(s_desc, sizeof(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 (lseek(p_fd, rgd_offset, SEEK_SET) == -1)
 279         goto fail_rgd;
 280     if (read(p_fd, rgd_buf, gd_size) != gd_size)
 281         goto fail_rgd;
 282     if (lseek(snp_fd, rgd_offset, SEEK_SET) == -1)
 283         goto fail_rgd;
 284     if (write(snp_fd, rgd_buf, gd_size) == -1)
 285         goto fail_rgd;
 286     qemu_free(rgd_buf);
 287
 288     /* write GD */
 289     gd_buf = qemu_malloc(gd_size);
 290     if (lseek(p_fd, gd_offset, SEEK_SET) == -1)
 291         goto fail_gd;
 292     if (read(p_fd, gd_buf, gd_size) != gd_size)
 293         goto fail_gd;
 294     if (lseek(snp_fd, gd_offset, SEEK_SET) == -1)
 295         goto fail_gd;
 296     if (write(snp_fd, gd_buf, gd_size) == -1)
 297         goto fail_gd;
 298     qemu_free(gd_buf);
 299
 300     close(p_fd);
 301     close(snp_fd);
 302     return 0;
 303
 304     fail_gd:
 305     qemu_free(gd_buf);
 306     fail_rgd:
 307     qemu_free(rgd_buf);
 308     fail:
 309     close(p_fd);
 310     close(snp_fd);
 311     return -1;
 312 }
 313
 314 static void vmdk_parent_close(BlockDriverState *bs)
 315 {
 316     if (bs->backing_hd)
 317         bdrv_close(bs->backing_hd);
 318 }
 319
 320 static int parent_open = 0;
 321 static int vmdk_parent_open(BlockDriverState *bs, const char * filename)
 322 {
 323     BDRVVmdkState *s = bs->opaque;
 324     char *p_name;
 325     char desc[DESC_SIZE];
 326     char parent_img_name[1024];
 327
 328     /* the descriptor offset = 0x200 */
 329     if (bdrv_pread(s->hd, 0x200, desc, DESC_SIZE) != DESC_SIZE)
 330         return -1;
 331
 332     if ((p_name = strstr(desc,"parentFileNameHint")) != NULL) {
 333         char *end_name;
 334         struct stat file_buf;
 335
 336         p_name += sizeof("parentFileNameHint") + 1;
 337         if ((end_name = strchr(p_name,'\"')) == NULL)
 338             return -1;
 339         if ((end_name - p_name) > sizeof (s->hd->backing_file) - 1)
 340             return -1;
 341
 342         pstrcpy(s->hd->backing_file, end_name - p_name + 1, p_name);
 343         if (stat(s->hd->backing_file, &file_buf) != 0) {
 344             path_combine(parent_img_name, sizeof(parent_img_name),
 345                          filename, s->hd->backing_file);
 346         } else {
 347             pstrcpy(parent_img_name, sizeof(parent_img_name),
 348                     s->hd->backing_file);
 349         }
 350
 351         s->hd->backing_hd = bdrv_new("");
 352         if (!s->hd->backing_hd) {
 353             failure:
 354             bdrv_close(s->hd);
 355             return -1;
 356         }
 357         parent_open = 1;
 358         if (bdrv_open(s->hd->backing_hd, parent_img_name, BDRV_O_RDONLY) < 0)
 359             goto failure;
 360         parent_open = 0;
 361     }
 362
 363     return 0;
 364 }
 365
 366 static int vmdk_open(BlockDriverState *bs, const char *filename, int flags)
 367 {
 368     BDRVVmdkState *s = bs->opaque;
 369     uint32_t magic;
 370     int l1_size, i, ret;
 371
 372     if (parent_open)
 373         // Parent must be opened as RO.
 374         flags = BDRV_O_RDONLY;
 375
 376     ret = bdrv_file_open(&s->hd, filename, flags);
 377     if (ret < 0)
 378         return ret;
 379     if (bdrv_pread(s->hd, 0, &magic, sizeof(magic)) != sizeof(magic))
 380         goto fail;
 381
 382     magic = be32_to_cpu(magic);
 383     if (magic == VMDK3_MAGIC) {
 384         VMDK3Header header;
 385
 386         if (bdrv_pread(s->hd, sizeof(magic), &header, sizeof(header)) != sizeof(header))
 387             goto fail;
 388         s->cluster_sectors = le32_to_cpu(header.granularity);
 389         s->l2_size = 1 << 9;
 390         s->l1_size = 1 << 6;
 391         bs->total_sectors = le32_to_cpu(header.disk_sectors);
 392         s->l1_table_offset = le32_to_cpu(header.l1dir_offset) << 9;
 393         s->l1_backup_table_offset = 0;
 394         s->l1_entry_sectors = s->l2_size * s->cluster_sectors;
 395     } else if (magic == VMDK4_MAGIC) {
 396         VMDK4Header header;
 397
 398         if (bdrv_pread(s->hd, sizeof(magic), &header, sizeof(header)) != sizeof(header))
 399             goto fail;
 400         bs->total_sectors = le64_to_cpu(header.capacity);
 401         s->cluster_sectors = le64_to_cpu(header.granularity);
 402         s->l2_size = le32_to_cpu(header.num_gtes_per_gte);
 403         s->l1_entry_sectors = s->l2_size * s->cluster_sectors;
 404         if (s->l1_entry_sectors <= 0)
 405             goto fail;
 406         s->l1_size = (bs->total_sectors + s->l1_entry_sectors - 1)
 407             / s->l1_entry_sectors;
 408         s->l1_table_offset = le64_to_cpu(header.rgd_offset) << 9;
 409         s->l1_backup_table_offset = le64_to_cpu(header.gd_offset) << 9;
 410
 411         if (parent_open)
 412             s->is_parent = 1;
 413         else
 414             s->is_parent = 0;
 415
 416         // try to open parent images, if exist
 417         if (vmdk_parent_open(bs, filename) != 0)
 418             goto fail;
 419         // write the CID once after the image creation
 420         s->parent_cid = vmdk_read_cid(bs,1);
 421     } else {
 422         goto fail;
 423     }
 424
 425     /* read the L1 table */
 426     l1_size = s->l1_size * sizeof(uint32_t);
 427     s->l1_table = qemu_malloc(l1_size);
 428     if (bdrv_pread(s->hd, s->l1_table_offset, s->l1_table, l1_size) != l1_size)
 429         goto fail;
 430     for(i = 0; i < s->l1_size; i++) {
 431         le32_to_cpus(&s->l1_table[i]);
 432     }
 433
 434     if (s->l1_backup_table_offset) {
 435         s->l1_backup_table = qemu_malloc(l1_size);
 436         if (bdrv_pread(s->hd, s->l1_backup_table_offset, s->l1_backup_table, l1_size) != l1_size)
 437             goto fail;
 438         for(i = 0; i < s->l1_size; i++) {
 439             le32_to_cpus(&s->l1_backup_table[i]);
 440         }
 441     }
 442
 443     s->l2_cache = qemu_malloc(s->l2_size * L2_CACHE_SIZE * sizeof(uint32_t));
 444     return 0;
 445  fail:
 446     qemu_free(s->l1_backup_table);
 447     qemu_free(s->l1_table);
 448     qemu_free(s->l2_cache);
 449     bdrv_delete(s->hd);
 450     return -1;
 451 }
 452
 453 static uint64_t get_cluster_offset(BlockDriverState *bs, VmdkMetaData *m_data,
 454                                    uint64_t offset, int allocate);
 455
 456 static int get_whole_cluster(BlockDriverState *bs, uint64_t cluster_offset,
 457                              uint64_t offset, int allocate)
 458 {
 459     uint64_t parent_cluster_offset;
 460     BDRVVmdkState *s = bs->opaque;
 461     uint8_t  whole_grain[s->cluster_sectors*512];        // 128 sectors * 512 bytes each = grain size 64KB
 462
 463     // we will be here if it's first write on non-exist grain(cluster).
 464     // try to read from parent image, if exist
 465     if (s->hd->backing_hd) {
 466         BDRVVmdkState *ps = s->hd->backing_hd->opaque;
 467
 468         if (!vmdk_is_cid_valid(bs))
 469             return -1;
 470
 471         parent_cluster_offset = get_cluster_offset(s->hd->backing_hd, NULL, offset, allocate);
 472
 473         if (parent_cluster_offset) {
 474             BDRVVmdkState *act_s = activeBDRV.hd->opaque;
 475
 476             if (bdrv_pread(ps->hd, parent_cluster_offset, whole_grain, ps->cluster_sectors*512) != ps->cluster_sectors*512)
 477                 return -1;
 478
 479             //Write grain only into the active image
 480             if (bdrv_pwrite(act_s->hd, activeBDRV.cluster_offset << 9, whole_grain, sizeof(whole_grain)) != sizeof(whole_grain))
 481                 return -1;
 482         }
 483     }
 484     return 0;
 485 }
 486
 487 static int vmdk_L2update(BlockDriverState *bs, VmdkMetaData *m_data)
 488 {
 489     BDRVVmdkState *s = bs->opaque;
 490
 491     /* update L2 table */
 492     if (bdrv_pwrite(s->hd, ((int64_t)m_data->l2_offset * 512) + (m_data->l2_index * sizeof(m_data->offset)),
 493                     &(m_data->offset), sizeof(m_data->offset)) != sizeof(m_data->offset))
 494         return -1;
 495     /* update backup L2 table */
 496     if (s->l1_backup_table_offset != 0) {
 497         m_data->l2_offset = s->l1_backup_table[m_data->l1_index];
 498         if (bdrv_pwrite(s->hd, ((int64_t)m_data->l2_offset * 512) + (m_data->l2_index * sizeof(m_data->offset)),
 499                         &(m_data->offset), sizeof(m_data->offset)) != sizeof(m_data->offset))
 500             return -1;
 501     }
 502
 503     return 0;
 504 }
 505
 506 static uint64_t get_cluster_offset(BlockDriverState *bs, VmdkMetaData *m_data,
 507                                    uint64_t offset, int allocate)
 508 {
 509     BDRVVmdkState *s = bs->opaque;
 510     unsigned int l1_index, l2_offset, l2_index;
 511     int min_index, i, j;
 512     uint32_t min_count, *l2_table, tmp = 0;
 513     uint64_t cluster_offset;
 514
 515     if (m_data)
 516         m_data->valid = 0;
 517
 518     l1_index = (offset >> 9) / s->l1_entry_sectors;
 519     if (l1_index >= s->l1_size)
 520         return 0;
 521     l2_offset = s->l1_table[l1_index];
 522     if (!l2_offset)
 523         return 0;
 524     for(i = 0; i < L2_CACHE_SIZE; i++) {
 525         if (l2_offset == s->l2_cache_offsets[i]) {
 526             /* increment the hit count */
 527             if (++s->l2_cache_counts[i] == 0xffffffff) {
 528                 for(j = 0; j < L2_CACHE_SIZE; j++) {
 529                     s->l2_cache_counts[j] >>= 1;
 530                 }
 531             }
 532             l2_table = s->l2_cache + (i * s->l2_size);
 533             goto found;
 534         }
 535     }
 536     /* not found: load a new entry in the least used one */
 537     min_index = 0;
 538     min_count = 0xffffffff;
 539     for(i = 0; i < L2_CACHE_SIZE; i++) {
 540         if (s->l2_cache_counts[i] < min_count) {
 541             min_count = s->l2_cache_counts[i];
 542             min_index = i;
 543         }
 544     }
 545     l2_table = s->l2_cache + (min_index * s->l2_size);
 546     if (bdrv_pread(s->hd, (int64_t)l2_offset * 512, l2_table, s->l2_size * sizeof(uint32_t)) !=
 547                                                                         s->l2_size * sizeof(uint32_t))
 548         return 0;
 549
 550     s->l2_cache_offsets[min_index] = l2_offset;
 551     s->l2_cache_counts[min_index] = 1;
 552  found:
 553     l2_index = ((offset >> 9) / s->cluster_sectors) % s->l2_size;
 554     cluster_offset = le32_to_cpu(l2_table[l2_index]);
 555
 556     if (!cluster_offset) {
 557         if (!allocate)
 558             return 0;
 559         // Avoid the L2 tables update for the images that have snapshots.
 560         if (!s->is_parent) {
 561             cluster_offset = bdrv_getlength(s->hd);
 562             bdrv_truncate(s->hd, cluster_offset + (s->cluster_sectors << 9));
 563
 564             cluster_offset >>= 9;
 565             tmp = cpu_to_le32(cluster_offset);
 566             l2_table[l2_index] = tmp;
 567             // Save the active image state
 568             activeBDRV.cluster_offset = cluster_offset;
 569             activeBDRV.hd = bs;
 570         }
 571         /* First of all we write grain itself, to avoid race condition
 572          * that may to corrupt the image.
 573          * This problem may occur because of insufficient space on host disk
 574          * or inappropriate VM shutdown.
 575          */
 576         if (get_whole_cluster(bs, cluster_offset, offset, allocate) == -1)
 577             return 0;
 578
 579         if (m_data) {
 580             m_data->offset = tmp;
 581             m_data->l1_index = l1_index;
 582             m_data->l2_index = l2_index;
 583             m_data->l2_offset = l2_offset;
 584             m_data->valid = 1;
 585         }
 586     }
 587     cluster_offset <<= 9;
 588     return cluster_offset;
 589 }
 590
 591 static int vmdk_is_allocated(BlockDriverState *bs, int64_t sector_num,
 592                              int nb_sectors, int *pnum)
 593 {
 594     BDRVVmdkState *s = bs->opaque;
 595     int index_in_cluster, n;
 596     uint64_t cluster_offset;
 597
 598     cluster_offset = get_cluster_offset(bs, NULL, sector_num << 9, 0);
 599     index_in_cluster = sector_num % s->cluster_sectors;
 600     n = s->cluster_sectors - index_in_cluster;
 601     if (n > nb_sectors)
 602         n = nb_sectors;
 603     *pnum = n;
 604     return (cluster_offset != 0);
 605 }
 606
 607 static int vmdk_read(BlockDriverState *bs, int64_t sector_num,
 608                     uint8_t *buf, int nb_sectors)
 609 {
 610     BDRVVmdkState *s = bs->opaque;
 611     int index_in_cluster, n, ret;
 612     uint64_t cluster_offset;
 613
 614     while (nb_sectors > 0) {
 615         cluster_offset = get_cluster_offset(bs, NULL, sector_num << 9, 0);
 616         index_in_cluster = sector_num % s->cluster_sectors;
 617         n = s->cluster_sectors - index_in_cluster;
 618         if (n > nb_sectors)
 619             n = nb_sectors;
 620         if (!cluster_offset) {
 621             // try to read from parent image, if exist
 622             if (s->hd->backing_hd) {
 623                 if (!vmdk_is_cid_valid(bs))
 624                     return -1;
 625                 ret = bdrv_read(s->hd->backing_hd, sector_num, buf, n);
 626                 if (ret < 0)
 627                     return -1;
 628             } else {
 629                 memset(buf, 0, 512 * n);
 630             }
 631         } else {
 632             if(bdrv_pread(s->hd, cluster_offset + index_in_cluster * 512, buf, n * 512) != n * 512)
 633                 return -1;
 634         }
 635         nb_sectors -= n;
 636         sector_num += n;
 637         buf += n * 512;
 638     }
 639     return 0;
 640 }
 641
 642 static int vmdk_write(BlockDriverState *bs, int64_t sector_num,
 643                      const uint8_t *buf, int nb_sectors)
 644 {
 645     BDRVVmdkState *s = bs->opaque;
 646     VmdkMetaData m_data;
 647     int index_in_cluster, n;
 648     uint64_t cluster_offset;
 649     static int cid_update = 0;
 650
 651     if (sector_num > bs->total_sectors) {
 652         fprintf(stderr,
 653                 "(VMDK) Wrong offset: sector_num=0x%" PRIx64
 654                 " total_sectors=0x%" PRIx64 "\n",
 655                 sector_num, bs->total_sectors);
 656         return -1;
 657     }
 658
 659     while (nb_sectors > 0) {
 660         index_in_cluster = sector_num & (s->cluster_sectors - 1);
 661         n = s->cluster_sectors - index_in_cluster;
 662         if (n > nb_sectors)
 663             n = nb_sectors;
 664         cluster_offset = get_cluster_offset(bs, &m_data, sector_num << 9, 1);
 665         if (!cluster_offset)
 666             return -1;
 667
 668         if (bdrv_pwrite(s->hd, cluster_offset + index_in_cluster * 512, buf, n * 512) != n * 512)
 669             return -1;
 670         if (m_data.valid) {
 671             /* update L2 tables */
 672             if (vmdk_L2update(bs, &m_data) == -1)
 673                 return -1;
 674         }
 675         nb_sectors -= n;
 676         sector_num += n;
 677         buf += n * 512;
 678
 679         // update CID on the first write every time the virtual disk is opened
 680         if (!cid_update) {
 681             vmdk_write_cid(bs, time(NULL));
 682             cid_update++;
 683         }
 684     }
 685     return 0;
 686 }
 687
 688 static int vmdk_create(const char *filename, int64_t total_size,
 689                        const char *backing_file, int flags)
 690 {
 691     int fd, i;
 692     VMDK4Header header;
 693     uint32_t tmp, magic, grains, gd_size, gt_size, gt_count;
 694     static const char desc_template[] =
 695         "# Disk DescriptorFile\n"
 696         "version=1\n"
 697         "CID=%x\n"
 698         "parentCID=ffffffff\n"
 699         "createType=\"monolithicSparse\"\n"
 700         "\n"
 701         "# Extent description\n"
 702         "RW %" PRId64 " SPARSE \"%s\"\n"
 703         "\n"
 704         "# The Disk Data Base \n"
 705         "#DDB\n"
 706         "\n"
 707         "ddb.virtualHWVersion = \"%d\"\n"
 708         "ddb.geometry.cylinders = \"%" PRId64 "\"\n"
 709         "ddb.geometry.heads = \"16\"\n"
 710         "ddb.geometry.sectors = \"63\"\n"
 711         "ddb.adapterType = \"ide\"\n";
 712     char desc[1024];
 713     const char *real_filename, *temp_str;
 714
 715     /* XXX: add support for backing file */
 716     if (backing_file) {
 717         return vmdk_snapshot_create(filename, backing_file);
 718     }
 719
 720     fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY | O_LARGEFILE,
 721               0644);
 722     if (fd < 0)
 723         return -1;
 724     magic = cpu_to_be32(VMDK4_MAGIC);
 725     memset(&header, 0, sizeof(header));
 726     header.version = cpu_to_le32(1);
 727     header.flags = cpu_to_le32(3); /* ?? */
 728     header.capacity = cpu_to_le64(total_size);
 729     header.granularity = cpu_to_le64(128);
 730     header.num_gtes_per_gte = cpu_to_le32(512);
 731
 732     grains = (total_size + header.granularity - 1) / header.granularity;
 733     gt_size = ((header.num_gtes_per_gte * sizeof(uint32_t)) + 511) >> 9;
 734     gt_count = (grains + header.num_gtes_per_gte - 1) / header.num_gtes_per_gte;
 735     gd_size = (gt_count * sizeof(uint32_t) + 511) >> 9;
 736
 737     header.desc_offset = 1;
 738     header.desc_size = 20;
 739     header.rgd_offset = header.desc_offset + header.desc_size;
 740     header.gd_offset = header.rgd_offset + gd_size + (gt_size * gt_count);
 741     header.grain_offset =
 742        ((header.gd_offset + gd_size + (gt_size * gt_count) +
 743          header.granularity - 1) / header.granularity) *
 744         header.granularity;
 745
 746     header.desc_offset = cpu_to_le64(header.desc_offset);
 747     header.desc_size = cpu_to_le64(header.desc_size);
 748     header.rgd_offset = cpu_to_le64(header.rgd_offset);
 749     header.gd_offset = cpu_to_le64(header.gd_offset);
 750     header.grain_offset = cpu_to_le64(header.grain_offset);
 751
 752     header.check_bytes[0] = 0xa;
 753     header.check_bytes[1] = 0x20;
 754     header.check_bytes[2] = 0xd;
 755     header.check_bytes[3] = 0xa;
 756
 757     /* write all the data */
 758     write(fd, &magic, sizeof(magic));
 759     write(fd, &header, sizeof(header));
 760
 761     ftruncate(fd, header.grain_offset << 9);
 762
 763     /* write grain directory */
 764     lseek(fd, le64_to_cpu(header.rgd_offset) << 9, SEEK_SET);
 765     for (i = 0, tmp = header.rgd_offset + gd_size;
 766          i < gt_count; i++, tmp += gt_size)
 767         write(fd, &tmp, sizeof(tmp));
 768
 769     /* write backup grain directory */
 770     lseek(fd, le64_to_cpu(header.gd_offset) << 9, SEEK_SET);
 771     for (i = 0, tmp = header.gd_offset + gd_size;
 772          i < gt_count; i++, tmp += gt_size)
 773         write(fd, &tmp, sizeof(tmp));
 774
 775     /* compose the descriptor */
 776     real_filename = filename;
 777     if ((temp_str = strrchr(real_filename, '\\')) != NULL)
 778         real_filename = temp_str + 1;
 779     if ((temp_str = strrchr(real_filename, '/')) != NULL)
 780         real_filename = temp_str + 1;
 781     if ((temp_str = strrchr(real_filename, ':')) != NULL)
 782         real_filename = temp_str + 1;
 783     snprintf(desc, sizeof(desc), desc_template, (unsigned int)time(NULL),
 784              total_size, real_filename,
 785              (flags & BLOCK_FLAG_COMPAT6 ? 6 : 4),
 786              total_size / (int64_t)(63 * 16));
 787
 788     /* write the descriptor */
 789     lseek(fd, le64_to_cpu(header.desc_offset) << 9, SEEK_SET);
 790     write(fd, desc, strlen(desc));
 791
 792     close(fd);
 793     return 0;
 794 }
 795
 796 static void vmdk_close(BlockDriverState *bs)
 797 {
 798     BDRVVmdkState *s = bs->opaque;
 799
 800     qemu_free(s->l1_table);
 801     qemu_free(s->l2_cache);
 802     // try to close parent image, if exist
 803     vmdk_parent_close(s->hd);
 804     bdrv_delete(s->hd);
 805 }
 806
 807 static void vmdk_flush(BlockDriverState *bs)
 808 {
 809     BDRVVmdkState *s = bs->opaque;
 810     bdrv_flush(s->hd);
 811 }
 812
 813 BlockDriver bdrv_vmdk = {
 814     .format_name        = "vmdk",
 815     .instance_size      = sizeof(BDRVVmdkState),
 816     .bdrv_probe         = vmdk_probe,
 817     .bdrv_open          = vmdk_open,
 818     .bdrv_read          = vmdk_read,
 819     .bdrv_write         = vmdk_write,
 820     .bdrv_close         = vmdk_close,
 821     .bdrv_create        = vmdk_create,
 822     .bdrv_flush         = vmdk_flush,
 823     .bdrv_is_allocated  = vmdk_is_allocated,
 824 };