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
  97 static int vmdk_probe(const uint8_t *buf, int buf_size, const char *filename)
  98 {
  99     uint32_t magic;
 100
 101     if (buf_size < 4)
 102         return 0;
 103     magic = be32_to_cpu(*(uint32_t *)buf);
 104     if (magic == VMDK3_MAGIC ||
 105         magic == VMDK4_MAGIC)
 106         return 100;
 107     else
 108         return 0;
 109 }
 110
 111 #define CHECK_CID 1
 112
 113 #define SECTOR_SIZE 512
 114 #define DESC_SIZE 20*SECTOR_SIZE        // 20 sectors of 512 bytes each
 115 #define HEADER_SIZE 512                         // first sector of 512 bytes
 116
 117 static uint32_t vmdk_read_cid(BlockDriverState *bs, int parent)
 118 {
 119     BDRVVmdkState *s = bs->opaque;
 120     char desc[DESC_SIZE];
 121     uint32_t cid;
 122     char *p_name, *cid_str;
 123     size_t cid_str_size;
 124
 125     /* the descriptor offset = 0x200 */
 126     if (bdrv_pread(s->hd, 0x200, desc, DESC_SIZE) != DESC_SIZE)
 127         return 0;
 128
 129     if (parent) {
 130         cid_str = "parentCID";
 131         cid_str_size = sizeof("parentCID");
 132     } else {
 133         cid_str = "CID";
 134         cid_str_size = sizeof("CID");
 135     }
 136
 137     if ((p_name = strstr(desc,cid_str)) != 0) {
 138         p_name += cid_str_size;
 139         sscanf(p_name,"%x",&cid);
 140     }
 141
 142     return cid;
 143 }
 144
 145 static int vmdk_write_cid(BlockDriverState *bs, uint32_t cid)
 146 {
 147     BDRVVmdkState *s = bs->opaque;
 148     char desc[DESC_SIZE], tmp_desc[DESC_SIZE];
 149     char *p_name, *tmp_str;
 150
 151     /* the descriptor offset = 0x200 */
 152     if (bdrv_pread(s->hd, 0x200, desc, DESC_SIZE) != DESC_SIZE)
 153         return -1;
 154
 155     tmp_str = strstr(desc,"parentCID");
 156     strcpy(tmp_desc, tmp_str);
 157     if ((p_name = strstr(desc,"CID")) != 0) {
 158         p_name += sizeof("CID");
 159         sprintf(p_name,"%x\n",cid);
 160         strcat(desc,tmp_desc);
 161     }
 162
 163     if (bdrv_pwrite(s->hd, 0x200, desc, DESC_SIZE) != DESC_SIZE)
 164         return -1;
 165     return 0;
 166 }
 167
 168 static int vmdk_is_cid_valid(BlockDriverState *bs)
 169 {
 170 #ifdef CHECK_CID
 171     BDRVVmdkState *s = bs->opaque;
 172     BlockDriverState *p_bs = s->hd->backing_hd;
 173     uint32_t cur_pcid;
 174
 175     if (p_bs) {
 176         cur_pcid = vmdk_read_cid(p_bs,0);
 177         if (s->parent_cid != cur_pcid)
 178             // CID not valid
 179             return 0;
 180     }
 181 #endif
 182     // CID valid
 183     return 1;
 184 }
 185
 186 static int vmdk_snapshot_create(const char *filename, const char *backing_file)
 187 {
 188     int snp_fd, p_fd;
 189     uint32_t p_cid;
 190     char *p_name, *gd_buf, *rgd_buf;
 191     const char *real_filename, *temp_str;
 192     VMDK4Header header;
 193     uint32_t gde_entries, gd_size;
 194     int64_t gd_offset, rgd_offset, capacity, gt_size;
 195     char p_desc[DESC_SIZE], s_desc[DESC_SIZE], hdr[HEADER_SIZE];
 196     char *desc_template =
 197     "# Disk DescriptorFile\n"
 198     "version=1\n"
 199     "CID=%x\n"
 200     "parentCID=%x\n"
 201     "createType=\"monolithicSparse\"\n"
 202     "parentFileNameHint=\"%s\"\n"
 203     "\n"
 204     "# Extent description\n"
 205     "RW %lu SPARSE \"%s\"\n"
 206     "\n"
 207     "# The Disk Data Base \n"
 208     "#DDB\n"
 209     "\n";
 210
 211     snp_fd = open(filename, O_RDWR | O_CREAT | O_TRUNC | O_BINARY | O_LARGEFILE, 0644);
 212     if (snp_fd < 0)
 213         return -1;
 214     p_fd = open(backing_file, O_RDONLY | O_BINARY | O_LARGEFILE);
 215     if (p_fd < 0) {
 216         close(snp_fd);
 217         return -1;
 218     }
 219
 220     /* read the header */
 221     if (lseek(p_fd, 0x0, SEEK_SET) == -1)
 222         goto fail;
 223     if (read(p_fd, hdr, HEADER_SIZE) != HEADER_SIZE)
 224         goto fail;
 225
 226     /* write the header */
 227     if (lseek(snp_fd, 0x0, SEEK_SET) == -1)
 228         goto fail;
 229     if (write(snp_fd, hdr, HEADER_SIZE) == -1)
 230         goto fail;
 231
 232     memset(&header, 0, sizeof(header));
 233     memcpy(&header,&hdr[4], sizeof(header)); // skip the VMDK4_MAGIC
 234
 235     ftruncate(snp_fd, header.grain_offset << 9);
 236     /* the descriptor offset = 0x200 */
 237     if (lseek(p_fd, 0x200, SEEK_SET) == -1)
 238         goto fail;
 239     if (read(p_fd, p_desc, DESC_SIZE) != DESC_SIZE)
 240         goto fail;
 241
 242     if ((p_name = strstr(p_desc,"CID")) != 0) {
 243         p_name += sizeof("CID");
 244         sscanf(p_name,"%x",&p_cid);
 245     }
 246
 247     real_filename = filename;
 248     if ((temp_str = strrchr(real_filename, '\\')) != NULL)
 249         real_filename = temp_str + 1;
 250     if ((temp_str = strrchr(real_filename, '/')) != NULL)
 251         real_filename = temp_str + 1;
 252     if ((temp_str = strrchr(real_filename, ':')) != NULL)
 253         real_filename = temp_str + 1;
 254
 255     sprintf(s_desc, desc_template, p_cid, p_cid, backing_file
 256             , (uint32_t)header.capacity, real_filename);
 257
 258     /* write the descriptor */
 259     if (lseek(snp_fd, 0x200, SEEK_SET) == -1)
 260         goto fail;
 261     if (write(snp_fd, s_desc, strlen(s_desc)) == -1)
 262         goto fail;
 263
 264     gd_offset = header.gd_offset * SECTOR_SIZE;     // offset of GD table
 265     rgd_offset = header.rgd_offset * SECTOR_SIZE;   // offset of RGD table
 266     capacity = header.capacity * SECTOR_SIZE;       // Extent size
 267     /*
 268      * Each GDE span 32M disk, means:
 269      * 512 GTE per GT, each GTE points to grain
 270      */
 271     gt_size = (int64_t)header.num_gtes_per_gte * header.granularity * SECTOR_SIZE;
 272     if (!gt_size)
 273         goto fail;
 274     gde_entries = (uint32_t)(capacity / gt_size);  // number of gde/rgde
 275     gd_size = gde_entries * sizeof(uint32_t);
 276
 277     /* write RGD */
 278     rgd_buf = qemu_malloc(gd_size);
 279     if (!rgd_buf)
 280         goto fail;
 281     if (lseek(p_fd, rgd_offset, SEEK_SET) == -1)
 282         goto fail_rgd;
 283     if (read(p_fd, rgd_buf, gd_size) != gd_size)
 284         goto fail_rgd;
 285     if (lseek(snp_fd, rgd_offset, SEEK_SET) == -1)
 286         goto fail_rgd;
 287     if (write(snp_fd, rgd_buf, gd_size) == -1)
 288         goto fail_rgd;
 289     qemu_free(rgd_buf);
 290
 291     /* write GD */
 292     gd_buf = qemu_malloc(gd_size);
 293     if (!gd_buf)
 294         goto fail_rgd;
 295     if (lseek(p_fd, gd_offset, SEEK_SET) == -1)
 296         goto fail_gd;
 297     if (read(p_fd, gd_buf, gd_size) != gd_size)
 298         goto fail_gd;
 299     if (lseek(snp_fd, gd_offset, SEEK_SET) == -1)
 300         goto fail_gd;
 301     if (write(snp_fd, gd_buf, gd_size) == -1)
 302         goto fail_gd;
 303     qemu_free(gd_buf);
 304
 305     close(p_fd);
 306     close(snp_fd);
 307     return 0;
 308
 309     fail_gd:
 310     qemu_free(gd_buf);
 311     fail_rgd:
 312     qemu_free(rgd_buf);
 313     fail:
 314     close(p_fd);
 315     close(snp_fd);
 316     return -1;
 317 }
 318
 319 static void vmdk_parent_close(BlockDriverState *bs)
 320 {
 321     if (bs->backing_hd)
 322         bdrv_close(bs->backing_hd);
 323 }
 324
 325 int parent_open = 0;
 326 static int vmdk_parent_open(BlockDriverState *bs, const char * filename)
 327 {
 328     BDRVVmdkState *s = bs->opaque;
 329     char *p_name;
 330     char desc[DESC_SIZE];
 331     char parent_img_name[1024];
 332
 333     /* the descriptor offset = 0x200 */
 334     if (bdrv_pread(s->hd, 0x200, desc, DESC_SIZE) != DESC_SIZE)
 335         return -1;
 336
 337     if ((p_name = strstr(desc,"parentFileNameHint")) != 0) {
 338         char *end_name;
 339         struct stat file_buf;
 340
 341         p_name += sizeof("parentFileNameHint") + 1;
 342         if ((end_name = strchr(p_name,'\"')) == 0)
 343             return -1;
 344         if ((end_name - p_name) > sizeof (s->hd->backing_file) - 1)
 345             return -1;
 346
 347         strncpy(s->hd->backing_file, p_name, end_name - p_name);
 348         if (stat(s->hd->backing_file, &file_buf) != 0) {
 349             path_combine(parent_img_name, sizeof(parent_img_name),
 350                          filename, s->hd->backing_file);
 351         } else {
 352             strcpy(parent_img_name, 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     fprintf(stderr, "(VMDK) image open: flags=0x%x filename=%s\n", flags, bs->filename);
 380
 381     ret = bdrv_file_open(&s->hd, filename, flags);
 382     if (ret < 0)
 383         return ret;
 384     if (bdrv_pread(s->hd, 0, &magic, sizeof(magic)) != sizeof(magic))
 385         goto fail;
 386
 387     magic = be32_to_cpu(magic);
 388     if (magic == VMDK3_MAGIC) {
 389         VMDK3Header header;
 390
 391         if (bdrv_pread(s->hd, sizeof(magic), &header, sizeof(header)) != sizeof(header))
 392             goto fail;
 393         s->cluster_sectors = le32_to_cpu(header.granularity);
 394         s->l2_size = 1 << 9;
 395         s->l1_size = 1 << 6;
 396         bs->total_sectors = le32_to_cpu(header.disk_sectors);
 397         s->l1_table_offset = le32_to_cpu(header.l1dir_offset) << 9;
 398         s->l1_backup_table_offset = 0;
 399         s->l1_entry_sectors = s->l2_size * s->cluster_sectors;
 400     } else if (magic == VMDK4_MAGIC) {
 401         VMDK4Header header;
 402
 403         if (bdrv_pread(s->hd, sizeof(magic), &header, sizeof(header)) != sizeof(header))
 404             goto fail;
 405         bs->total_sectors = le64_to_cpu(header.capacity);
 406         s->cluster_sectors = le64_to_cpu(header.granularity);
 407         s->l2_size = le32_to_cpu(header.num_gtes_per_gte);
 408         s->l1_entry_sectors = s->l2_size * s->cluster_sectors;
 409         if (s->l1_entry_sectors <= 0)
 410             goto fail;
 411         s->l1_size = (bs->total_sectors + s->l1_entry_sectors - 1)
 412             / s->l1_entry_sectors;
 413         s->l1_table_offset = le64_to_cpu(header.rgd_offset) << 9;
 414         s->l1_backup_table_offset = le64_to_cpu(header.gd_offset) << 9;
 415
 416         if (parent_open)
 417             s->is_parent = 1;
 418         else
 419             s->is_parent = 0;
 420
 421         // try to open parent images, if exist
 422         if (vmdk_parent_open(bs, filename) != 0)
 423             goto fail;
 424         // write the CID once after the image creation
 425         s->parent_cid = vmdk_read_cid(bs,1);
 426     } else {
 427         goto fail;
 428     }
 429
 430     /* read the L1 table */
 431     l1_size = s->l1_size * sizeof(uint32_t);
 432     s->l1_table = qemu_malloc(l1_size);
 433     if (!s->l1_table)
 434         goto fail;
 435     if (bdrv_pread(s->hd, s->l1_table_offset, s->l1_table, l1_size) != l1_size)
 436         goto fail;
 437     for(i = 0; i < s->l1_size; i++) {
 438         le32_to_cpus(&s->l1_table[i]);
 439     }
 440
 441     if (s->l1_backup_table_offset) {
 442         s->l1_backup_table = qemu_malloc(l1_size);
 443         if (!s->l1_backup_table)
 444             goto fail;
 445         if (bdrv_pread(s->hd, s->l1_backup_table_offset, s->l1_backup_table, l1_size) != l1_size)
 446             goto fail;
 447         for(i = 0; i < s->l1_size; i++) {
 448             le32_to_cpus(&s->l1_backup_table[i]);
 449         }
 450     }
 451
 452     s->l2_cache = qemu_malloc(s->l2_size * L2_CACHE_SIZE * sizeof(uint32_t));
 453     if (!s->l2_cache)
 454         goto fail;
 455     return 0;
 456  fail:
 457     qemu_free(s->l1_backup_table);
 458     qemu_free(s->l1_table);
 459     qemu_free(s->l2_cache);
 460     bdrv_delete(s->hd);
 461     return -1;
 462 }
 463
 464 static uint64_t get_cluster_offset(BlockDriverState *bs, VmdkMetaData *m_data,
 465                                    uint64_t offset, int allocate);
 466
 467 static int get_whole_cluster(BlockDriverState *bs, uint64_t cluster_offset,
 468                              uint64_t offset, int allocate)
 469 {
 470     uint64_t parent_cluster_offset;
 471     BDRVVmdkState *s = bs->opaque;
 472     uint8_t  whole_grain[s->cluster_sectors*512];        // 128 sectors * 512 bytes each = grain size 64KB
 473
 474     // we will be here if it's first write on non-exist grain(cluster).
 475     // try to read from parent image, if exist
 476     if (s->hd->backing_hd) {
 477         BDRVVmdkState *ps = s->hd->backing_hd->opaque;
 478
 479         if (!vmdk_is_cid_valid(bs))
 480             return -1;
 481
 482         parent_cluster_offset = get_cluster_offset(s->hd->backing_hd, NULL, offset, allocate);
 483
 484         if (parent_cluster_offset) {
 485             BDRVVmdkState *act_s = activeBDRV.hd->opaque;
 486
 487             if (bdrv_pread(ps->hd, parent_cluster_offset, whole_grain, ps->cluster_sectors*512) != ps->cluster_sectors*512)
 488                 return -1;
 489
 490             //Write grain only into the active image
 491             if (bdrv_pwrite(act_s->hd, activeBDRV.cluster_offset << 9, whole_grain, sizeof(whole_grain)) != sizeof(whole_grain))
 492                 return -1;
 493         }
 494     }
 495     return 0;
 496 }
 497
 498 static int vmdk_L2update(BlockDriverState *bs, VmdkMetaData *m_data)
 499 {
 500     BDRVVmdkState *s = bs->opaque;
 501
 502     /* update L2 table */
 503     if (bdrv_pwrite(s->hd, ((int64_t)m_data->l2_offset * 512) + (m_data->l2_index * sizeof(m_data->offset)),
 504                     &(m_data->offset), sizeof(m_data->offset)) != sizeof(m_data->offset))
 505         return -1;
 506     /* update backup L2 table */
 507     if (s->l1_backup_table_offset != 0) {
 508         m_data->l2_offset = s->l1_backup_table[m_data->l1_index];
 509         if (bdrv_pwrite(s->hd, ((int64_t)m_data->l2_offset * 512) + (m_data->l2_index * sizeof(m_data->offset)),
 510                         &(m_data->offset), sizeof(m_data->offset)) != sizeof(m_data->offset))
 511             return -1;
 512     }
 513
 514     return 0;
 515 }
 516
 517 static uint64_t get_cluster_offset(BlockDriverState *bs, VmdkMetaData *m_data,
 518                                    uint64_t offset, int allocate)
 519 {
 520     BDRVVmdkState *s = bs->opaque;
 521     unsigned int l1_index, l2_offset, l2_index;
 522     int min_index, i, j;
 523     uint32_t min_count, *l2_table, tmp = 0;
 524     uint64_t cluster_offset;
 525
 526     if (m_data)
 527         m_data->valid = 0;
 528
 529     l1_index = (offset >> 9) / s->l1_entry_sectors;
 530     if (l1_index >= s->l1_size)
 531         return 0;
 532     l2_offset = s->l1_table[l1_index];
 533     if (!l2_offset)
 534         return 0;
 535     for(i = 0; i < L2_CACHE_SIZE; i++) {
 536         if (l2_offset == s->l2_cache_offsets[i]) {
 537             /* increment the hit count */
 538             if (++s->l2_cache_counts[i] == 0xffffffff) {
 539                 for(j = 0; j < L2_CACHE_SIZE; j++) {
 540                     s->l2_cache_counts[j] >>= 1;
 541                 }
 542             }
 543             l2_table = s->l2_cache + (i * s->l2_size);
 544             goto found;
 545         }
 546     }
 547     /* not found: load a new entry in the least used one */
 548     min_index = 0;
 549     min_count = 0xffffffff;
 550     for(i = 0; i < L2_CACHE_SIZE; i++) {
 551         if (s->l2_cache_counts[i] < min_count) {
 552             min_count = s->l2_cache_counts[i];
 553             min_index = i;
 554         }
 555     }
 556     l2_table = s->l2_cache + (min_index * s->l2_size);
 557     if (bdrv_pread(s->hd, (int64_t)l2_offset * 512, l2_table, s->l2_size * sizeof(uint32_t)) !=
 558                                                                         s->l2_size * sizeof(uint32_t))
 559         return 0;
 560
 561     s->l2_cache_offsets[min_index] = l2_offset;
 562     s->l2_cache_counts[min_index] = 1;
 563  found:
 564     l2_index = ((offset >> 9) / s->cluster_sectors) % s->l2_size;
 565     cluster_offset = le32_to_cpu(l2_table[l2_index]);
 566
 567     if (!cluster_offset) {
 568         if (!allocate)
 569             return 0;
 570         // Avoid the L2 tables update for the images that have snapshots.
 571         if (!s->is_parent) {
 572             cluster_offset = bdrv_getlength(s->hd);
 573             bdrv_truncate(s->hd, cluster_offset + (s->cluster_sectors << 9));
 574
 575             cluster_offset >>= 9;
 576             tmp = cpu_to_le32(cluster_offset);
 577             l2_table[l2_index] = tmp;
 578             // Save the active image state
 579             activeBDRV.cluster_offset = cluster_offset;
 580             activeBDRV.hd = bs;
 581         }
 582         /* First of all we write grain itself, to avoid race condition
 583          * that may to corrupt the image.
 584          * This problem may occur because of insufficient space on host disk
 585          * or inappropriate VM shutdown.
 586          */
 587         if (get_whole_cluster(bs, cluster_offset, offset, allocate) == -1)
 588             return 0;
 589
 590         if (m_data) {
 591             m_data->offset = tmp;
 592             m_data->l1_index = l1_index;
 593             m_data->l2_index = l2_index;
 594             m_data->l2_offset = l2_offset;
 595             m_data->valid = 1;
 596         }
 597     }
 598     cluster_offset <<= 9;
 599     return cluster_offset;
 600 }
 601
 602 static int vmdk_is_allocated(BlockDriverState *bs, int64_t sector_num,
 603                              int nb_sectors, int *pnum)
 604 {
 605     BDRVVmdkState *s = bs->opaque;
 606     int index_in_cluster, n;
 607     uint64_t cluster_offset;
 608
 609     cluster_offset = get_cluster_offset(bs, NULL, sector_num << 9, 0);
 610     index_in_cluster = sector_num % s->cluster_sectors;
 611     n = s->cluster_sectors - index_in_cluster;
 612     if (n > nb_sectors)
 613         n = nb_sectors;
 614     *pnum = n;
 615     return (cluster_offset != 0);
 616 }
 617
 618 static int vmdk_read(BlockDriverState *bs, int64_t sector_num,
 619                     uint8_t *buf, int nb_sectors)
 620 {
 621     BDRVVmdkState *s = bs->opaque;
 622     int index_in_cluster, n, ret;
 623     uint64_t cluster_offset;
 624
 625     while (nb_sectors > 0) {
 626         cluster_offset = get_cluster_offset(bs, NULL, sector_num << 9, 0);
 627         index_in_cluster = sector_num % s->cluster_sectors;
 628         n = s->cluster_sectors - index_in_cluster;
 629         if (n > nb_sectors)
 630             n = nb_sectors;
 631         if (!cluster_offset) {
 632             // try to read from parent image, if exist
 633             if (s->hd->backing_hd) {
 634                 if (!vmdk_is_cid_valid(bs))
 635                     return -1;
 636                 ret = bdrv_read(s->hd->backing_hd, sector_num, buf, n);
 637                 if (ret < 0)
 638                     return -1;
 639             } else {
 640                 memset(buf, 0, 512 * n);
 641             }
 642         } else {
 643             if(bdrv_pread(s->hd, cluster_offset + index_in_cluster * 512, buf, n * 512) != n * 512)
 644                 return -1;
 645         }
 646         nb_sectors -= n;
 647         sector_num += n;
 648         buf += n * 512;
 649     }
 650     return 0;
 651 }
 652
 653 static int vmdk_write(BlockDriverState *bs, int64_t sector_num,
 654                      const uint8_t *buf, int nb_sectors)
 655 {
 656     BDRVVmdkState *s = bs->opaque;
 657     VmdkMetaData m_data;
 658     int index_in_cluster, n;
 659     uint64_t cluster_offset;
 660     static int cid_update = 0;
 661
 662     if (sector_num > bs->total_sectors) {
 663         fprintf(stderr,
 664                 "(VMDK) Wrong offset: sector_num=0x%" PRIx64
 665                 " total_sectors=0x%" PRIx64 "\n",
 666                 sector_num, bs->total_sectors);
 667         return -1;
 668     }
 669
 670     while (nb_sectors > 0) {
 671         index_in_cluster = sector_num & (s->cluster_sectors - 1);
 672         n = s->cluster_sectors - index_in_cluster;
 673         if (n > nb_sectors)
 674             n = nb_sectors;
 675         cluster_offset = get_cluster_offset(bs, &m_data, sector_num << 9, 1);
 676         if (!cluster_offset)
 677             return -1;
 678
 679         if (bdrv_pwrite(s->hd, cluster_offset + index_in_cluster * 512, buf, n * 512) != n * 512)
 680             return -1;
 681         if (m_data.valid) {
 682             /* update L2 tables */
 683             if (vmdk_L2update(bs, &m_data) == -1)
 684                 return -1;
 685         }
 686         nb_sectors -= n;
 687         sector_num += n;
 688         buf += n * 512;
 689
 690         // update CID on the first write every time the virtual disk is opened
 691         if (!cid_update) {
 692             vmdk_write_cid(bs, time(NULL));
 693             cid_update++;
 694         }
 695     }
 696     return 0;
 697 }
 698
 699 static int vmdk_create(const char *filename, int64_t total_size,
 700                        const char *backing_file, int flags)
 701 {
 702     int fd, i;
 703     VMDK4Header header;
 704     uint32_t tmp, magic, grains, gd_size, gt_size, gt_count;
 705     char *desc_template =
 706         "# Disk DescriptorFile\n"
 707         "version=1\n"
 708         "CID=%x\n"
 709         "parentCID=ffffffff\n"
 710         "createType=\"monolithicSparse\"\n"
 711         "\n"
 712         "# Extent description\n"
 713         "RW %lu SPARSE \"%s\"\n"
 714         "\n"
 715         "# The Disk Data Base \n"
 716         "#DDB\n"
 717         "\n"
 718         "ddb.virtualHWVersion = \"%d\"\n"
 719         "ddb.geometry.cylinders = \"%lu\"\n"
 720         "ddb.geometry.heads = \"16\"\n"
 721         "ddb.geometry.sectors = \"63\"\n"
 722         "ddb.adapterType = \"ide\"\n";
 723     char desc[1024];
 724     const char *real_filename, *temp_str;
 725
 726     /* XXX: add support for backing file */
 727     if (backing_file) {
 728         return vmdk_snapshot_create(filename, backing_file);
 729     }
 730
 731     fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY | O_LARGEFILE,
 732               0644);
 733     if (fd < 0)
 734         return -1;
 735     magic = cpu_to_be32(VMDK4_MAGIC);
 736     memset(&header, 0, sizeof(header));
 737     header.version = cpu_to_le32(1);
 738     header.flags = cpu_to_le32(3); /* ?? */
 739     header.capacity = cpu_to_le64(total_size);
 740     header.granularity = cpu_to_le64(128);
 741     header.num_gtes_per_gte = cpu_to_le32(512);
 742
 743     grains = (total_size + header.granularity - 1) / header.granularity;
 744     gt_size = ((header.num_gtes_per_gte * sizeof(uint32_t)) + 511) >> 9;
 745     gt_count = (grains + header.num_gtes_per_gte - 1) / header.num_gtes_per_gte;
 746     gd_size = (gt_count * sizeof(uint32_t) + 511) >> 9;
 747
 748     header.desc_offset = 1;
 749     header.desc_size = 20;
 750     header.rgd_offset = header.desc_offset + header.desc_size;
 751     header.gd_offset = header.rgd_offset + gd_size + (gt_size * gt_count);
 752     header.grain_offset =
 753        ((header.gd_offset + gd_size + (gt_size * gt_count) +
 754          header.granularity - 1) / header.granularity) *
 755         header.granularity;
 756
 757     header.desc_offset = cpu_to_le64(header.desc_offset);
 758     header.desc_size = cpu_to_le64(header.desc_size);
 759     header.rgd_offset = cpu_to_le64(header.rgd_offset);
 760     header.gd_offset = cpu_to_le64(header.gd_offset);
 761     header.grain_offset = cpu_to_le64(header.grain_offset);
 762
 763     header.check_bytes[0] = 0xa;
 764     header.check_bytes[1] = 0x20;
 765     header.check_bytes[2] = 0xd;
 766     header.check_bytes[3] = 0xa;
 767
 768     /* write all the data */
 769     write(fd, &magic, sizeof(magic));
 770     write(fd, &header, sizeof(header));
 771
 772     ftruncate(fd, header.grain_offset << 9);
 773
 774     /* write grain directory */
 775     lseek(fd, le64_to_cpu(header.rgd_offset) << 9, SEEK_SET);
 776     for (i = 0, tmp = header.rgd_offset + gd_size;
 777          i < gt_count; i++, tmp += gt_size)
 778         write(fd, &tmp, sizeof(tmp));
 779
 780     /* write backup grain directory */
 781     lseek(fd, le64_to_cpu(header.gd_offset) << 9, SEEK_SET);
 782     for (i = 0, tmp = header.gd_offset + gd_size;
 783          i < gt_count; i++, tmp += gt_size)
 784         write(fd, &tmp, sizeof(tmp));
 785
 786     /* compose the descriptor */
 787     real_filename = filename;
 788     if ((temp_str = strrchr(real_filename, '\\')) != NULL)
 789         real_filename = temp_str + 1;
 790     if ((temp_str = strrchr(real_filename, '/')) != NULL)
 791         real_filename = temp_str + 1;
 792     if ((temp_str = strrchr(real_filename, ':')) != NULL)
 793         real_filename = temp_str + 1;
 794     sprintf(desc, desc_template, time(NULL), (unsigned long)total_size,
 795             real_filename, (flags & BLOCK_FLAG_COMPAT6 ? 6 : 4), total_size / (63 * 16));
 796
 797     /* write the descriptor */
 798     lseek(fd, le64_to_cpu(header.desc_offset) << 9, SEEK_SET);
 799     write(fd, desc, strlen(desc));
 800
 801     close(fd);
 802     return 0;
 803 }
 804
 805 static void vmdk_close(BlockDriverState *bs)
 806 {
 807     BDRVVmdkState *s = bs->opaque;
 808
 809     qemu_free(s->l1_table);
 810     qemu_free(s->l2_cache);
 811     bdrv_delete(s->hd);
 812     // try to close parent image, if exist
 813     vmdk_parent_close(s->hd);
 814 }
 815
 816 static void vmdk_flush(BlockDriverState *bs)
 817 {
 818     BDRVVmdkState *s = bs->opaque;
 819     bdrv_flush(s->hd);
 820 }
 821
 822 BlockDriver bdrv_vmdk = {
 823     "vmdk",
 824     sizeof(BDRVVmdkState),
 825     vmdk_probe,
 826     vmdk_open,
 827     vmdk_read,
 828     vmdk_write,
 829     vmdk_close,
 830     vmdk_create,
 831     vmdk_flush,
 832     vmdk_is_allocated,
 833 };