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)) != 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     pstrcpy(tmp_desc, sizeof(tmp_desc), tmp_str);
 156     if ((p_name = strstr(desc,"CID")) != 0) {
 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")) != 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     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 (!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 static 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         if ((end_name - p_name) > sizeof (s->hd->backing_file) - 1)
 344             return -1;
 345
 346         pstrcpy(s->hd->backing_file, end_name - p_name + 1, p_name);
 347         if (stat(s->hd->backing_file, &file_buf) != 0) {
 348             path_combine(parent_img_name, sizeof(parent_img_name),
 349                          filename, s->hd->backing_file);
 350         } else {
 351             pstrcpy(parent_img_name, sizeof(parent_img_name),
 352                     s->hd->backing_file);
 353         }
 354
 355         s->hd->backing_hd = bdrv_new("");
 356         if (!s->hd->backing_hd) {
 357             failure:
 358             bdrv_close(s->hd);
 359             return -1;
 360         }
 361         parent_open = 1;
 362         if (bdrv_open(s->hd->backing_hd, parent_img_name, BDRV_O_RDONLY) < 0)
 363             goto failure;
 364         parent_open = 0;
 365     }
 366
 367     return 0;
 368 }
 369
 370 static int vmdk_open(BlockDriverState *bs, const char *filename, int flags)
 371 {
 372     BDRVVmdkState *s = bs->opaque;
 373     uint32_t magic;
 374     int l1_size, i, ret;
 375
 376     if (parent_open)
 377         // Parent must be opened as RO.
 378         flags = BDRV_O_RDONLY;
 379
 380     ret = bdrv_file_open(&s->hd, filename, flags);
 381     if (ret < 0)
 382         return ret;
 383     if (bdrv_pread(s->hd, 0, &magic, sizeof(magic)) != sizeof(magic))
 384         goto fail;
 385
 386     magic = be32_to_cpu(magic);
 387     if (magic == VMDK3_MAGIC) {
 388         VMDK3Header header;
 389
 390         if (bdrv_pread(s->hd, sizeof(magic), &header, sizeof(header)) != sizeof(header))
 391             goto fail;
 392         s->cluster_sectors = le32_to_cpu(header.granularity);
 393         s->l2_size = 1 << 9;
 394         s->l1_size = 1 << 6;
 395         bs->total_sectors = le32_to_cpu(header.disk_sectors);
 396         s->l1_table_offset = le32_to_cpu(header.l1dir_offset) << 9;
 397         s->l1_backup_table_offset = 0;
 398         s->l1_entry_sectors = s->l2_size * s->cluster_sectors;
 399     } else if (magic == VMDK4_MAGIC) {
 400         VMDK4Header header;
 401
 402         if (bdrv_pread(s->hd, sizeof(magic), &header, sizeof(header)) != sizeof(header))
 403             goto fail;
 404         bs->total_sectors = le64_to_cpu(header.capacity);
 405         s->cluster_sectors = le64_to_cpu(header.granularity);
 406         s->l2_size = le32_to_cpu(header.num_gtes_per_gte);
 407         s->l1_entry_sectors = s->l2_size * s->cluster_sectors;
 408         if (s->l1_entry_sectors <= 0)
 409             goto fail;
 410         s->l1_size = (bs->total_sectors + s->l1_entry_sectors - 1)
 411             / s->l1_entry_sectors;
 412         s->l1_table_offset = le64_to_cpu(header.rgd_offset) << 9;
 413         s->l1_backup_table_offset = le64_to_cpu(header.gd_offset) << 9;
 414
 415         if (parent_open)
 416             s->is_parent = 1;
 417         else
 418             s->is_parent = 0;
 419
 420         // try to open parent images, if exist
 421         if (vmdk_parent_open(bs, filename) != 0)
 422             goto fail;
 423         // write the CID once after the image creation
 424         s->parent_cid = vmdk_read_cid(bs,1);
 425     } else {
 426         goto fail;
 427     }
 428
 429     /* read the L1 table */
 430     l1_size = s->l1_size * sizeof(uint32_t);
 431     s->l1_table = qemu_malloc(l1_size);
 432     if (!s->l1_table)
 433         goto fail;
 434     if (bdrv_pread(s->hd, s->l1_table_offset, s->l1_table, l1_size) != l1_size)
 435         goto fail;
 436     for(i = 0; i < s->l1_size; i++) {
 437         le32_to_cpus(&s->l1_table[i]);
 438     }
 439
 440     if (s->l1_backup_table_offset) {
 441         s->l1_backup_table = qemu_malloc(l1_size);
 442         if (!s->l1_backup_table)
 443             goto fail;
 444         if (bdrv_pread(s->hd, s->l1_backup_table_offset, s->l1_backup_table, l1_size) != l1_size)
 445             goto fail;
 446         for(i = 0; i < s->l1_size; i++) {
 447             le32_to_cpus(&s->l1_backup_table[i]);
 448         }
 449     }
 450
 451     s->l2_cache = qemu_malloc(s->l2_size * L2_CACHE_SIZE * sizeof(uint32_t));
 452     if (!s->l2_cache)
 453         goto fail;
 454     return 0;
 455  fail:
 456     qemu_free(s->l1_backup_table);
 457     qemu_free(s->l1_table);
 458     qemu_free(s->l2_cache);
 459     bdrv_delete(s->hd);
 460     return -1;
 461 }
 462
 463 static uint64_t get_cluster_offset(BlockDriverState *bs, VmdkMetaData *m_data,
 464                                    uint64_t offset, int allocate);
 465
 466 static int get_whole_cluster(BlockDriverState *bs, uint64_t cluster_offset,
 467                              uint64_t offset, int allocate)
 468 {
 469     uint64_t parent_cluster_offset;
 470     BDRVVmdkState *s = bs->opaque;
 471     uint8_t  whole_grain[s->cluster_sectors*512];        // 128 sectors * 512 bytes each = grain size 64KB
 472
 473     // we will be here if it's first write on non-exist grain(cluster).
 474     // try to read from parent image, if exist
 475     if (s->hd->backing_hd) {
 476         BDRVVmdkState *ps = s->hd->backing_hd->opaque;
 477
 478         if (!vmdk_is_cid_valid(bs))
 479             return -1;
 480
 481         parent_cluster_offset = get_cluster_offset(s->hd->backing_hd, NULL, offset, allocate);
 482
 483         if (parent_cluster_offset) {
 484             BDRVVmdkState *act_s = activeBDRV.hd->opaque;
 485
 486             if (bdrv_pread(ps->hd, parent_cluster_offset, whole_grain, ps->cluster_sectors*512) != ps->cluster_sectors*512)
 487                 return -1;
 488
 489             //Write grain only into the active image
 490             if (bdrv_pwrite(act_s->hd, activeBDRV.cluster_offset << 9, whole_grain, sizeof(whole_grain)) != sizeof(whole_grain))
 491                 return -1;
 492         }
 493     }
 494     return 0;
 495 }
 496
 497 static int vmdk_L2update(BlockDriverState *bs, VmdkMetaData *m_data)
 498 {
 499     BDRVVmdkState *s = bs->opaque;
 500
 501     /* update L2 table */
 502     if (bdrv_pwrite(s->hd, ((int64_t)m_data->l2_offset * 512) + (m_data->l2_index * sizeof(m_data->offset)),
 503                     &(m_data->offset), sizeof(m_data->offset)) != sizeof(m_data->offset))
 504         return -1;
 505     /* update backup L2 table */
 506     if (s->l1_backup_table_offset != 0) {
 507         m_data->l2_offset = s->l1_backup_table[m_data->l1_index];
 508         if (bdrv_pwrite(s->hd, ((int64_t)m_data->l2_offset * 512) + (m_data->l2_index * sizeof(m_data->offset)),
 509                         &(m_data->offset), sizeof(m_data->offset)) != sizeof(m_data->offset))
 510             return -1;
 511     }
 512
 513     return 0;
 514 }
 515
 516 static uint64_t get_cluster_offset(BlockDriverState *bs, VmdkMetaData *m_data,
 517                                    uint64_t offset, int allocate)
 518 {
 519     BDRVVmdkState *s = bs->opaque;
 520     unsigned int l1_index, l2_offset, l2_index;
 521     int min_index, i, j;
 522     uint32_t min_count, *l2_table, tmp = 0;
 523     uint64_t cluster_offset;
 524
 525     if (m_data)
 526         m_data->valid = 0;
 527
 528     l1_index = (offset >> 9) / s->l1_entry_sectors;
 529     if (l1_index >= s->l1_size)
 530         return 0;
 531     l2_offset = s->l1_table[l1_index];
 532     if (!l2_offset)
 533         return 0;
 534     for(i = 0; i < L2_CACHE_SIZE; i++) {
 535         if (l2_offset == s->l2_cache_offsets[i]) {
 536             /* increment the hit count */
 537             if (++s->l2_cache_counts[i] == 0xffffffff) {
 538                 for(j = 0; j < L2_CACHE_SIZE; j++) {
 539                     s->l2_cache_counts[j] >>= 1;
 540                 }
 541             }
 542             l2_table = s->l2_cache + (i * s->l2_size);
 543             goto found;
 544         }
 545     }
 546     /* not found: load a new entry in the least used one */
 547     min_index = 0;
 548     min_count = 0xffffffff;
 549     for(i = 0; i < L2_CACHE_SIZE; i++) {
 550         if (s->l2_cache_counts[i] < min_count) {
 551             min_count = s->l2_cache_counts[i];
 552             min_index = i;
 553         }
 554     }
 555     l2_table = s->l2_cache + (min_index * s->l2_size);
 556     if (bdrv_pread(s->hd, (int64_t)l2_offset * 512, l2_table, s->l2_size * sizeof(uint32_t)) !=
 557                                                                         s->l2_size * sizeof(uint32_t))
 558         return 0;
 559
 560     s->l2_cache_offsets[min_index] = l2_offset;
 561     s->l2_cache_counts[min_index] = 1;
 562  found:
 563     l2_index = ((offset >> 9) / s->cluster_sectors) % s->l2_size;
 564     cluster_offset = le32_to_cpu(l2_table[l2_index]);
 565
 566     if (!cluster_offset) {
 567         if (!allocate)
 568             return 0;
 569         // Avoid the L2 tables update for the images that have snapshots.
 570         if (!s->is_parent) {
 571             cluster_offset = bdrv_getlength(s->hd);
 572             bdrv_truncate(s->hd, cluster_offset + (s->cluster_sectors << 9));
 573
 574             cluster_offset >>= 9;
 575             tmp = cpu_to_le32(cluster_offset);
 576             l2_table[l2_index] = tmp;
 577             // Save the active image state
 578             activeBDRV.cluster_offset = cluster_offset;
 579             activeBDRV.hd = bs;
 580         }
 581         /* First of all we write grain itself, to avoid race condition
 582          * that may to corrupt the image.
 583          * This problem may occur because of insufficient space on host disk
 584          * or inappropriate VM shutdown.
 585          */
 586         if (get_whole_cluster(bs, cluster_offset, offset, allocate) == -1)
 587             return 0;
 588
 589         if (m_data) {
 590             m_data->offset = tmp;
 591             m_data->l1_index = l1_index;
 592             m_data->l2_index = l2_index;
 593             m_data->l2_offset = l2_offset;
 594             m_data->valid = 1;
 595         }
 596     }
 597     cluster_offset <<= 9;
 598     return cluster_offset;
 599 }
 600
 601 static int vmdk_is_allocated(BlockDriverState *bs, int64_t sector_num,
 602                              int nb_sectors, int *pnum)
 603 {
 604     BDRVVmdkState *s = bs->opaque;
 605     int index_in_cluster, n;
 606     uint64_t cluster_offset;
 607
 608     cluster_offset = get_cluster_offset(bs, NULL, sector_num << 9, 0);
 609     index_in_cluster = sector_num % s->cluster_sectors;
 610     n = s->cluster_sectors - index_in_cluster;
 611     if (n > nb_sectors)
 612         n = nb_sectors;
 613     *pnum = n;
 614     return (cluster_offset != 0);
 615 }
 616
 617 static int vmdk_read(BlockDriverState *bs, int64_t sector_num,
 618                     uint8_t *buf, int nb_sectors)
 619 {
 620     BDRVVmdkState *s = bs->opaque;
 621     int index_in_cluster, n, ret;
 622     uint64_t cluster_offset;
 623
 624     while (nb_sectors > 0) {
 625         cluster_offset = get_cluster_offset(bs, NULL, sector_num << 9, 0);
 626         index_in_cluster = sector_num % s->cluster_sectors;
 627         n = s->cluster_sectors - index_in_cluster;
 628         if (n > nb_sectors)
 629             n = nb_sectors;
 630         if (!cluster_offset) {
 631             // try to read from parent image, if exist
 632             if (s->hd->backing_hd) {
 633                 if (!vmdk_is_cid_valid(bs))
 634                     return -1;
 635                 ret = bdrv_read(s->hd->backing_hd, sector_num, buf, n);
 636                 if (ret < 0)
 637                     return -1;
 638             } else {
 639                 memset(buf, 0, 512 * n);
 640             }
 641         } else {
 642             if(bdrv_pread(s->hd, cluster_offset + index_in_cluster * 512, buf, n * 512) != n * 512)
 643                 return -1;
 644         }
 645         nb_sectors -= n;
 646         sector_num += n;
 647         buf += n * 512;
 648     }
 649     return 0;
 650 }
 651
 652 static int vmdk_write(BlockDriverState *bs, int64_t sector_num,
 653                      const uint8_t *buf, int nb_sectors)
 654 {
 655     BDRVVmdkState *s = bs->opaque;
 656     VmdkMetaData m_data;
 657     int index_in_cluster, n;
 658     uint64_t cluster_offset;
 659     static int cid_update = 0;
 660
 661     if (sector_num > bs->total_sectors) {
 662         fprintf(stderr,
 663                 "(VMDK) Wrong offset: sector_num=0x%" PRIx64
 664                 " total_sectors=0x%" PRIx64 "\n",
 665                 sector_num, bs->total_sectors);
 666         return -1;
 667     }
 668
 669     while (nb_sectors > 0) {
 670         index_in_cluster = sector_num & (s->cluster_sectors - 1);
 671         n = s->cluster_sectors - index_in_cluster;
 672         if (n > nb_sectors)
 673             n = nb_sectors;
 674         cluster_offset = get_cluster_offset(bs, &m_data, sector_num << 9, 1);
 675         if (!cluster_offset)
 676             return -1;
 677
 678         if (bdrv_pwrite(s->hd, cluster_offset + index_in_cluster * 512, buf, n * 512) != n * 512)
 679             return -1;
 680         if (m_data.valid) {
 681             /* update L2 tables */
 682             if (vmdk_L2update(bs, &m_data) == -1)
 683                 return -1;
 684         }
 685         nb_sectors -= n;
 686         sector_num += n;
 687         buf += n * 512;
 688
 689         // update CID on the first write every time the virtual disk is opened
 690         if (!cid_update) {
 691             vmdk_write_cid(bs, time(NULL));
 692             cid_update++;
 693         }
 694     }
 695     return 0;
 696 }
 697
 698 static int vmdk_create(const char *filename, int64_t total_size,
 699                        const char *backing_file, int flags)
 700 {
 701     int fd, i;
 702     VMDK4Header header;
 703     uint32_t tmp, magic, grains, gd_size, gt_size, gt_count;
 704     static const char desc_template[] =
 705         "# Disk DescriptorFile\n"
 706         "version=1\n"
 707         "CID=%x\n"
 708         "parentCID=ffffffff\n"
 709         "createType=\"monolithicSparse\"\n"
 710         "\n"
 711         "# Extent description\n"
 712         "RW %" PRId64 " SPARSE \"%s\"\n"
 713         "\n"
 714         "# The Disk Data Base \n"
 715         "#DDB\n"
 716         "\n"
 717         "ddb.virtualHWVersion = \"%d\"\n"
 718         "ddb.geometry.cylinders = \"%" PRId64 "\"\n"
 719         "ddb.geometry.heads = \"16\"\n"
 720         "ddb.geometry.sectors = \"63\"\n"
 721         "ddb.adapterType = \"ide\"\n";
 722     char desc[1024];
 723     const char *real_filename, *temp_str;
 724
 725     /* XXX: add support for backing file */
 726     if (backing_file) {
 727         return vmdk_snapshot_create(filename, backing_file);
 728     }
 729
 730     fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY | O_LARGEFILE,
 731               0644);
 732     if (fd < 0)
 733         return -1;
 734     magic = cpu_to_be32(VMDK4_MAGIC);
 735     memset(&header, 0, sizeof(header));
 736     header.version = cpu_to_le32(1);
 737     header.flags = cpu_to_le32(3); /* ?? */
 738     header.capacity = cpu_to_le64(total_size);
 739     header.granularity = cpu_to_le64(128);
 740     header.num_gtes_per_gte = cpu_to_le32(512);
 741
 742     grains = (total_size + header.granularity - 1) / header.granularity;
 743     gt_size = ((header.num_gtes_per_gte * sizeof(uint32_t)) + 511) >> 9;
 744     gt_count = (grains + header.num_gtes_per_gte - 1) / header.num_gtes_per_gte;
 745     gd_size = (gt_count * sizeof(uint32_t) + 511) >> 9;
 746
 747     header.desc_offset = 1;
 748     header.desc_size = 20;
 749     header.rgd_offset = header.desc_offset + header.desc_size;
 750     header.gd_offset = header.rgd_offset + gd_size + (gt_size * gt_count);
 751     header.grain_offset =
 752        ((header.gd_offset + gd_size + (gt_size * gt_count) +
 753          header.granularity - 1) / header.granularity) *
 754         header.granularity;
 755
 756     header.desc_offset = cpu_to_le64(header.desc_offset);
 757     header.desc_size = cpu_to_le64(header.desc_size);
 758     header.rgd_offset = cpu_to_le64(header.rgd_offset);
 759     header.gd_offset = cpu_to_le64(header.gd_offset);
 760     header.grain_offset = cpu_to_le64(header.grain_offset);
 761
 762     header.check_bytes[0] = 0xa;
 763     header.check_bytes[1] = 0x20;
 764     header.check_bytes[2] = 0xd;
 765     header.check_bytes[3] = 0xa;
 766
 767     /* write all the data */
 768     write(fd, &magic, sizeof(magic));
 769     write(fd, &header, sizeof(header));
 770
 771     ftruncate(fd, header.grain_offset << 9);
 772
 773     /* write grain directory */
 774     lseek(fd, le64_to_cpu(header.rgd_offset) << 9, SEEK_SET);
 775     for (i = 0, tmp = header.rgd_offset + gd_size;
 776          i < gt_count; i++, tmp += gt_size)
 777         write(fd, &tmp, sizeof(tmp));
 778
 779     /* write backup grain directory */
 780     lseek(fd, le64_to_cpu(header.gd_offset) << 9, SEEK_SET);
 781     for (i = 0, tmp = header.gd_offset + gd_size;
 782          i < gt_count; i++, tmp += gt_size)
 783         write(fd, &tmp, sizeof(tmp));
 784
 785     /* compose the descriptor */
 786     real_filename = filename;
 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     if ((temp_str = strrchr(real_filename, ':')) != NULL)
 792         real_filename = temp_str + 1;
 793     snprintf(desc, sizeof(desc), desc_template, (unsigned int)time(NULL),
 794              total_size, real_filename,
 795              (flags & BLOCK_FLAG_COMPAT6 ? 6 : 4),
 796              total_size / (int64_t)(63 * 16));
 797
 798     /* write the descriptor */
 799     lseek(fd, le64_to_cpu(header.desc_offset) << 9, SEEK_SET);
 800     write(fd, desc, strlen(desc));
 801
 802     close(fd);
 803     return 0;
 804 }
 805
 806 static void vmdk_close(BlockDriverState *bs)
 807 {
 808     BDRVVmdkState *s = bs->opaque;
 809
 810     qemu_free(s->l1_table);
 811     qemu_free(s->l2_cache);
 812     // try to close parent image, if exist
 813     vmdk_parent_close(s->hd);
 814     bdrv_delete(s->hd);
 815 }
 816
 817 static void vmdk_flush(BlockDriverState *bs)
 818 {
 819     BDRVVmdkState *s = bs->opaque;
 820     bdrv_flush(s->hd);
 821 }
 822
 823 BlockDriver bdrv_vmdk = {
 824     "vmdk",
 825     sizeof(BDRVVmdkState),
 826     vmdk_probe,
 827     vmdk_open,
 828     vmdk_read,
 829     vmdk_write,
 830     vmdk_close,
 831     vmdk_create,
 832     vmdk_flush,
 833     vmdk_is_allocated,
 834 };