block-vmdk.c

   1 /*
   2  * Block driver for the VMDK format
   3  *
   4  * Copyright (c) 2004 Fabrice Bellard
   5  * Copyright (c) 2005 Filip Navara
   6  *
   7  * Permission is hereby granted, free of charge, to any person obtaining a copy
   8  * of this software and associated documentation files (the "Software"), to deal
   9  * in the Software without restriction, including without limitation the rights
  10  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
  11  * copies of the Software, and to permit persons to whom the Software is
  12  * furnished to do so, subject to the following conditions:
  13  *
  14  * The above copyright notice and this permission notice shall be included in
  15  * all copies or substantial portions of the Software.
  16  *
  17  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  18  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  19  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
  20  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  21  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  22  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  23  * THE SOFTWARE.
  24  */
  25
  26 #include "vl.h"
  27 #include "block_int.h"
  28
  29 #define VMDK3_MAGIC (('C' << 24) | ('O' << 16) | ('W' << 8) | 'D')
  30 #define VMDK4_MAGIC (('K' << 24) | ('D' << 16) | ('M' << 8) | 'V')
  31
  32 typedef struct {
  33     uint32_t version;
  34     uint32_t flags;
  35     uint32_t disk_sectors;
  36     uint32_t granularity;
  37     uint32_t l1dir_offset;
  38     uint32_t l1dir_size;
  39     uint32_t file_sectors;
  40     uint32_t cylinders;
  41     uint32_t heads;
  42     uint32_t sectors_per_track;
  43 } VMDK3Header;
  44
  45 typedef struct {
  46     uint32_t version;
  47     uint32_t flags;
  48     int64_t capacity;
  49     int64_t granularity;
  50     int64_t desc_offset;
  51     int64_t desc_size;
  52     int32_t num_gtes_per_gte;
  53     int64_t rgd_offset;
  54     int64_t gd_offset;
  55     int64_t grain_offset;
  56     char filler[1];
  57     char check_bytes[4];
  58 } __attribute__((packed)) VMDK4Header;
  59
  60 #define L2_CACHE_SIZE 16
  61
  62 typedef struct BDRVVmdkState {
  63     BlockDriverState *hd;
  64     int64_t l1_table_offset;
  65     int64_t l1_backup_table_offset;
  66     uint32_t *l1_table;
  67     uint32_t *l1_backup_table;
  68     unsigned int l1_size;
  69     uint32_t l1_entry_sectors;
  70
  71     unsigned int l2_size;
  72     uint32_t *l2_cache;
  73     uint32_t l2_cache_offsets[L2_CACHE_SIZE];
  74     uint32_t l2_cache_counts[L2_CACHE_SIZE];
  75
  76     unsigned int cluster_sectors;
  77     uint32_t parent_cid;
  78     int is_parent;
  79     DiskIOStatistics io;
  80 } BDRVVmdkState;
  81
  82 typedef struct VmdkMetaData {
  83     uint32_t offset;
  84     unsigned int l1_index;
  85     unsigned int l2_index;
  86     unsigned int l2_offset;
  87     int valid;
  88 } VmdkMetaData;
  89
  90 typedef struct ActiveBDRVState{
  91     BlockDriverState *hd;            // active image handler
  92     uint64_t cluster_offset;         // current write offset
  93 }ActiveBDRVState;
  94
  95 static ActiveBDRVState activeBDRV;
  96
  97 DiskIOStatistics vmdk_io_statistics(BlockDriverState *bs)
  98 {
  99     BDRVVmdkState *s = bs->opaque;
 100     // return disk I/O counters
 101     return s->io;
 102 }
 103
 104 static int vmdk_probe(const uint8_t *buf, int buf_size, const char *filename)
 105 {
 106     uint32_t magic;
 107
 108     if (buf_size < 4)
 109         return 0;
 110     magic = be32_to_cpu(*(uint32_t *)buf);
 111     if (magic == VMDK3_MAGIC ||
 112         magic == VMDK4_MAGIC)
 113         return 100;
 114     else
 115         return 0;
 116 }
 117
 118 #define CHECK_CID 1
 119
 120 #define SECTOR_SIZE 512
 121 #define DESC_SIZE 20*SECTOR_SIZE        // 20 sectors of 512 bytes each
 122 #define HEADER_SIZE 512                         // first sector of 512 bytes
 123
 124 static uint32_t vmdk_read_cid(BlockDriverState *bs, int parent)
 125 {
 126     BDRVVmdkState *s = bs->opaque;
 127     char desc[DESC_SIZE];
 128     uint32_t cid;
 129     char *p_name, *cid_str;
 130     size_t cid_str_size;
 131
 132     /* the descriptor offset = 0x200 */
 133     if (bdrv_pread(s->hd, 0x200, desc, DESC_SIZE) != DESC_SIZE)
 134         return 0;
 135
 136     if (parent) {
 137         cid_str = "parentCID";
 138         cid_str_size = sizeof("parentCID");
 139     } else {
 140         cid_str = "CID";
 141         cid_str_size = sizeof("CID");
 142     }
 143
 144     if ((p_name = strstr(desc,cid_str)) != 0) {
 145         p_name += cid_str_size;
 146         sscanf(p_name,"%x",&cid);
 147     }
 148
 149     return cid;
 150 }
 151
 152 static int vmdk_write_cid(BlockDriverState *bs, uint32_t cid)
 153 {
 154     BDRVVmdkState *s = bs->opaque;
 155     char desc[DESC_SIZE], tmp_desc[DESC_SIZE];
 156     char *p_name, *tmp_str;
 157
 158     /* the descriptor offset = 0x200 */
 159     if (bdrv_pread(s->hd, 0x200, desc, DESC_SIZE) != DESC_SIZE)
 160         return -1;
 161
 162     tmp_str = strstr(desc,"parentCID");
 163     strcpy(tmp_desc, tmp_str);
 164     if ((p_name = strstr(desc,"CID")) != 0) {
 165         p_name += sizeof("CID");
 166         sprintf(p_name,"%x\n",cid);
 167         strcat(desc,tmp_desc);
 168     }
 169
 170     if (bdrv_pwrite(s->hd, 0x200, desc, DESC_SIZE) != DESC_SIZE)
 171         return -1;
 172     return 0;
 173 }
 174
 175 static int vmdk_is_cid_valid(BlockDriverState *bs)
 176 {
 177 #ifdef CHECK_CID
 178     BDRVVmdkState *s = bs->opaque;
 179     BlockDriverState *p_bs = s->hd->backing_hd;
 180     uint32_t cur_pcid;
 181
 182     if (p_bs) {
 183         cur_pcid = vmdk_read_cid(p_bs,0);
 184         if (s->parent_cid != cur_pcid)
 185             // CID not valid
 186             return 0;
 187     }
 188 #endif
 189     // CID valid
 190     return 1;
 191 }
 192
 193 static int vmdk_snapshot_create(const char *filename, const char *backing_file)
 194 {
 195     int snp_fd, p_fd;
 196     uint32_t p_cid;
 197     char *p_name, *gd_buf, *rgd_buf;
 198     const char *real_filename, *temp_str;
 199     VMDK4Header header;
 200     uint32_t gde_entries, gd_size;
 201     int64_t gd_offset, rgd_offset, capacity, gt_size;
 202     char p_desc[DESC_SIZE], s_desc[DESC_SIZE], hdr[HEADER_SIZE];
 203     char *desc_template =
 204     "# Disk DescriptorFile\n"
 205     "version=1\n"
 206     "CID=%x\n"
 207     "parentCID=%x\n"
 208     "createType=\"monolithicSparse\"\n"
 209     "parentFileNameHint=\"%s\"\n"
 210     "\n"
 211     "# Extent description\n"
 212     "RW %lu SPARSE \"%s\"\n"
 213     "\n"
 214     "# The Disk Data Base \n"
 215     "#DDB\n"
 216     "\n";
 217
 218     snp_fd = open(filename, O_RDWR | O_CREAT | O_TRUNC | O_BINARY | O_LARGEFILE, 0644);
 219     if (snp_fd < 0)
 220         return -1;
 221     p_fd = open(backing_file, O_RDONLY | O_BINARY | O_LARGEFILE);
 222     if (p_fd < 0) {
 223         close(snp_fd);
 224         return -1;
 225     }
 226
 227     /* read the header */
 228     if (lseek(p_fd, 0x0, SEEK_SET) == -1)
 229         goto fail;
 230     if (read(p_fd, hdr, HEADER_SIZE) != HEADER_SIZE)
 231         goto fail;
 232
 233     /* write the header */
 234     if (lseek(snp_fd, 0x0, SEEK_SET) == -1)
 235         goto fail;
 236     if (write(snp_fd, hdr, HEADER_SIZE) == -1)
 237         goto fail;
 238
 239     memset(&header, 0, sizeof(header));
 240     memcpy(&header,&hdr[4], sizeof(header)); // skip the VMDK4_MAGIC
 241
 242     ftruncate(snp_fd, header.grain_offset << 9);
 243     /* the descriptor offset = 0x200 */
 244     if (lseek(p_fd, 0x200, SEEK_SET) == -1)
 245         goto fail;
 246     if (read(p_fd, p_desc, DESC_SIZE) != DESC_SIZE)
 247         goto fail;
 248
 249     if ((p_name = strstr(p_desc,"CID")) != 0) {
 250         p_name += sizeof("CID");
 251         sscanf(p_name,"%x",&p_cid);
 252     }
 253
 254     real_filename = filename;
 255     if ((temp_str = strrchr(real_filename, '\\')) != NULL)
 256         real_filename = temp_str + 1;
 257     if ((temp_str = strrchr(real_filename, '/')) != NULL)
 258         real_filename = temp_str + 1;
 259     if ((temp_str = strrchr(real_filename, ':')) != NULL)
 260         real_filename = temp_str + 1;
 261
 262     sprintf(s_desc, desc_template, p_cid, p_cid, backing_file
 263             , (uint32_t)header.capacity, real_filename);
 264
 265     /* write the descriptor */
 266     if (lseek(snp_fd, 0x200, SEEK_SET) == -1)
 267         goto fail;
 268     if (write(snp_fd, s_desc, strlen(s_desc)) == -1)
 269         goto fail;
 270
 271     gd_offset = header.gd_offset * SECTOR_SIZE;     // offset of GD table
 272     rgd_offset = header.rgd_offset * SECTOR_SIZE;   // offset of RGD table
 273     capacity = header.capacity * SECTOR_SIZE;       // Extent size
 274     /*
 275      * Each GDE span 32M disk, means:
 276      * 512 GTE per GT, each GTE points to grain
 277      */
 278     gt_size = (int64_t)header.num_gtes_per_gte * header.granularity * SECTOR_SIZE;
 279     if (!gt_size)
 280         goto fail;
 281     gde_entries = (uint32_t)(capacity / gt_size);  // number of gde/rgde
 282     gd_size = gde_entries * sizeof(uint32_t);
 283
 284     /* write RGD */
 285     rgd_buf = qemu_malloc(gd_size);
 286     if (!rgd_buf)
 287         goto fail;
 288     if (lseek(p_fd, rgd_offset, SEEK_SET) == -1)
 289         goto fail_rgd;
 290     if (read(p_fd, rgd_buf, gd_size) != gd_size)
 291         goto fail_rgd;
 292     if (lseek(snp_fd, rgd_offset, SEEK_SET) == -1)
 293         goto fail_rgd;
 294     if (write(snp_fd, rgd_buf, gd_size) == -1)
 295         goto fail_rgd;
 296     qemu_free(rgd_buf);
 297
 298     /* write GD */
 299     gd_buf = qemu_malloc(gd_size);
 300     if (!gd_buf)
 301         goto fail_rgd;
 302     if (lseek(p_fd, gd_offset, SEEK_SET) == -1)
 303         goto fail_gd;
 304     if (read(p_fd, gd_buf, gd_size) != gd_size)
 305         goto fail_gd;
 306     if (lseek(snp_fd, gd_offset, SEEK_SET) == -1)
 307         goto fail_gd;
 308     if (write(snp_fd, gd_buf, gd_size) == -1)
 309         goto fail_gd;
 310     qemu_free(gd_buf);
 311
 312     close(p_fd);
 313     close(snp_fd);
 314     return 0;
 315
 316     fail_gd:
 317     qemu_free(gd_buf);
 318     fail_rgd:
 319     qemu_free(rgd_buf);
 320     fail:
 321     close(p_fd);
 322     close(snp_fd);
 323     return -1;
 324 }
 325
 326 static void vmdk_parent_close(BlockDriverState *bs)
 327 {
 328     if (bs->backing_hd)
 329         bdrv_close(bs->backing_hd);
 330 }
 331
 332 int parent_open = 0;
 333 static int vmdk_parent_open(BlockDriverState *bs, const char * filename)
 334 {
 335     BDRVVmdkState *s = bs->opaque;
 336     char *p_name;
 337     char desc[DESC_SIZE];
 338     char parent_img_name[1024];
 339
 340     /* the descriptor offset = 0x200 */
 341     if (bdrv_pread(s->hd, 0x200, desc, DESC_SIZE) != DESC_SIZE)
 342         return -1;
 343
 344     if ((p_name = strstr(desc,"parentFileNameHint")) != 0) {
 345         char *end_name;
 346         struct stat file_buf;
 347
 348         p_name += sizeof("parentFileNameHint") + 1;
 349         if ((end_name = strchr(p_name,'\"')) == 0)
 350             return -1;
 351
 352         strncpy(s->hd->backing_file, p_name, end_name - p_name);
 353         if (stat(s->hd->backing_file, &file_buf) != 0) {
 354             path_combine(parent_img_name, sizeof(parent_img_name),
 355                          filename, s->hd->backing_file);
 356         } else {
 357             strcpy(parent_img_name, s->hd->backing_file);
 358         }
 359
 360         s->hd->backing_hd = bdrv_new("");
 361         if (!s->hd->backing_hd) {
 362             failure:
 363             bdrv_close(s->hd);
 364             return -1;
 365         }
 366         parent_open = 1;
 367         if (bdrv_open(s->hd->backing_hd, parent_img_name, BDRV_O_RDONLY) < 0)
 368             goto failure;
 369         parent_open = 0;
 370     }
 371
 372     return 0;
 373 }
 374
 375 static int vmdk_open(BlockDriverState *bs, const char *filename, int flags)
 376 {
 377     BDRVVmdkState *s = bs->opaque;
 378     uint32_t magic;
 379     int l1_size, i, ret;
 380
 381     if (parent_open)
 382         // Parent must be opened as RO.
 383         flags = BDRV_O_RDONLY;
 384     fprintf(stderr, "(VMDK) image open: flags=0x%x filename=%s\n", flags, bs->filename);
 385
 386     ret = bdrv_file_open(&s->hd, filename, flags);
 387     if (ret < 0)
 388         return ret;
 389     if (bdrv_pread(s->hd, 0, &magic, sizeof(magic)) != sizeof(magic))
 390         goto fail;
 391
 392     magic = be32_to_cpu(magic);
 393     if (magic == VMDK3_MAGIC) {
 394         VMDK3Header header;
 395
 396         if (bdrv_pread(s->hd, sizeof(magic), &header, sizeof(header)) != sizeof(header))
 397             goto fail;
 398         s->cluster_sectors = le32_to_cpu(header.granularity);
 399         s->l2_size = 1 << 9;
 400         s->l1_size = 1 << 6;
 401         bs->total_sectors = le32_to_cpu(header.disk_sectors);
 402         s->l1_table_offset = le32_to_cpu(header.l1dir_offset) << 9;
 403         s->l1_backup_table_offset = 0;
 404         s->l1_entry_sectors = s->l2_size * s->cluster_sectors;
 405     } else if (magic == VMDK4_MAGIC) {
 406         VMDK4Header header;
 407
 408         if (bdrv_pread(s->hd, sizeof(magic), &header, sizeof(header)) != sizeof(header))
 409             goto fail;
 410         bs->total_sectors = le64_to_cpu(header.capacity);
 411         s->cluster_sectors = le64_to_cpu(header.granularity);
 412         s->l2_size = le32_to_cpu(header.num_gtes_per_gte);
 413         s->l1_entry_sectors = s->l2_size * s->cluster_sectors;
 414         if (s->l1_entry_sectors <= 0)
 415             goto fail;
 416         s->l1_size = (bs->total_sectors + s->l1_entry_sectors - 1)
 417             / s->l1_entry_sectors;
 418         s->l1_table_offset = le64_to_cpu(header.rgd_offset) << 9;
 419         s->l1_backup_table_offset = le64_to_cpu(header.gd_offset) << 9;
 420
 421         if (parent_open)
 422             s->is_parent = 1;
 423         else
 424             s->is_parent = 0;
 425
 426         // try to open parent images, if exist
 427         if (vmdk_parent_open(bs, filename) != 0)
 428             goto fail;
 429         // write the CID once after the image creation
 430         s->parent_cid = vmdk_read_cid(bs,1);
 431     } else {
 432         goto fail;
 433     }
 434
 435     /* read the L1 table */
 436     l1_size = s->l1_size * sizeof(uint32_t);
 437     s->l1_table = qemu_malloc(l1_size);
 438     if (!s->l1_table)
 439         goto fail;
 440     if (bdrv_pread(s->hd, s->l1_table_offset, s->l1_table, l1_size) != l1_size)
 441         goto fail;
 442     for(i = 0; i < s->l1_size; i++) {
 443         le32_to_cpus(&s->l1_table[i]);
 444     }
 445
 446     if (s->l1_backup_table_offset) {
 447         s->l1_backup_table = qemu_malloc(l1_size);
 448         if (!s->l1_backup_table)
 449             goto fail;
 450         if (bdrv_pread(s->hd, s->l1_backup_table_offset, s->l1_backup_table, l1_size) != l1_size)
 451             goto fail;
 452         for(i = 0; i < s->l1_size; i++) {
 453             le32_to_cpus(&s->l1_backup_table[i]);
 454         }
 455     }
 456
 457     s->l2_cache = qemu_malloc(s->l2_size * L2_CACHE_SIZE * sizeof(uint32_t));
 458     if (!s->l2_cache)
 459         goto fail;
 460     return 0;
 461  fail:
 462     qemu_free(s->l1_backup_table);
 463     qemu_free(s->l1_table);
 464     qemu_free(s->l2_cache);
 465     bdrv_delete(s->hd);
 466     return -1;
 467 }
 468
 469 static uint64_t get_cluster_offset(BlockDriverState *bs, VmdkMetaData *m_data,
 470                                    uint64_t offset, int allocate);
 471
 472 static int get_whole_cluster(BlockDriverState *bs, uint64_t cluster_offset,
 473                              uint64_t offset, int allocate)
 474 {
 475     uint64_t parent_cluster_offset;
 476     BDRVVmdkState *s = bs->opaque;
 477     uint8_t  whole_grain[s->cluster_sectors*512];        // 128 sectors * 512 bytes each = grain size 64KB
 478
 479     // we will be here if it's first write on non-exist grain(cluster).
 480     // try to read from parent image, if exist
 481     if (s->hd->backing_hd) {
 482         BDRVVmdkState *ps = s->hd->backing_hd->opaque;
 483
 484         if (!vmdk_is_cid_valid(bs))
 485             return -1;
 486
 487         parent_cluster_offset = get_cluster_offset(s->hd->backing_hd, NULL, offset, allocate);
 488
 489         if (parent_cluster_offset) {
 490             BDRVVmdkState *act_s = activeBDRV.hd->opaque;
 491
 492             if (bdrv_pread(ps->hd, parent_cluster_offset, whole_grain, ps->cluster_sectors*512) != ps->cluster_sectors*512)
 493                 return -1;
 494
 495             //Write grain only into the active image
 496             if (bdrv_pwrite(act_s->hd, activeBDRV.cluster_offset << 9, whole_grain, sizeof(whole_grain)) != sizeof(whole_grain))
 497                 return -1;
 498         }
 499     }
 500     return 0;
 501 }
 502
 503 static int vmdk_L2update(BlockDriverState *bs, VmdkMetaData *m_data)
 504 {
 505     BDRVVmdkState *s = bs->opaque;
 506
 507     /* update L2 table */
 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     /* update backup L2 table */
 512     if (s->l1_backup_table_offset != 0) {
 513         m_data->l2_offset = s->l1_backup_table[m_data->l1_index];
 514         if (bdrv_pwrite(s->hd, ((int64_t)m_data->l2_offset * 512) + (m_data->l2_index * sizeof(m_data->offset)),
 515                         &(m_data->offset), sizeof(m_data->offset)) != sizeof(m_data->offset))
 516             return -1;
 517     }
 518
 519     return 0;
 520 }
 521
 522 static uint64_t get_cluster_offset(BlockDriverState *bs, VmdkMetaData *m_data,
 523                                    uint64_t offset, int allocate)
 524 {
 525     BDRVVmdkState *s = bs->opaque;
 526     unsigned int l1_index, l2_offset, l2_index;
 527     int min_index, i, j;
 528     uint32_t min_count, *l2_table, tmp = 0;
 529     uint64_t cluster_offset;
 530
 531     if (m_data)
 532         m_data->valid = 0;
 533
 534     l1_index = (offset >> 9) / s->l1_entry_sectors;
 535     if (l1_index >= s->l1_size)
 536         return 0;
 537     l2_offset = s->l1_table[l1_index];
 538     if (!l2_offset)
 539         return 0;
 540     for(i = 0; i < L2_CACHE_SIZE; i++) {
 541         if (l2_offset == s->l2_cache_offsets[i]) {
 542             /* increment the hit count */
 543             if (++s->l2_cache_counts[i] == 0xffffffff) {
 544                 for(j = 0; j < L2_CACHE_SIZE; j++) {
 545                     s->l2_cache_counts[j] >>= 1;
 546                 }
 547             }
 548             l2_table = s->l2_cache + (i * s->l2_size);
 549             goto found;
 550         }
 551     }
 552     /* not found: load a new entry in the least used one */
 553     min_index = 0;
 554     min_count = 0xffffffff;
 555     for(i = 0; i < L2_CACHE_SIZE; i++) {
 556         if (s->l2_cache_counts[i] < min_count) {
 557             min_count = s->l2_cache_counts[i];
 558             min_index = i;
 559         }
 560     }
 561     l2_table = s->l2_cache + (min_index * s->l2_size);
 562     if (bdrv_pread(s->hd, (int64_t)l2_offset * 512, l2_table, s->l2_size * sizeof(uint32_t)) !=
 563                                                                         s->l2_size * sizeof(uint32_t))
 564         return 0;
 565
 566     s->l2_cache_offsets[min_index] = l2_offset;
 567     s->l2_cache_counts[min_index] = 1;
 568  found:
 569     l2_index = ((offset >> 9) / s->cluster_sectors) % s->l2_size;
 570     cluster_offset = le32_to_cpu(l2_table[l2_index]);
 571
 572     if (!cluster_offset) {
 573         if (!allocate)
 574             return 0;
 575         // Avoid the L2 tables update for the images that have snapshots.
 576         if (!s->is_parent) {
 577             cluster_offset = bdrv_getlength(s->hd);
 578             bdrv_truncate(s->hd, cluster_offset + (s->cluster_sectors << 9));
 579
 580             cluster_offset >>= 9;
 581             tmp = cpu_to_le32(cluster_offset);
 582             l2_table[l2_index] = tmp;
 583             // Save the active image state
 584             activeBDRV.cluster_offset = cluster_offset;
 585             activeBDRV.hd = bs;
 586         }
 587         /* First of all we write grain itself, to avoid race condition
 588          * that may to corrupt the image.
 589          * This problem may occur because of insufficient space on host disk
 590          * or inappropriate VM shutdown.
 591          */
 592         if (get_whole_cluster(bs, cluster_offset, offset, allocate) == -1)
 593             return 0;
 594
 595         if (m_data) {
 596             m_data->offset = tmp;
 597             m_data->l1_index = l1_index;
 598             m_data->l2_index = l2_index;
 599             m_data->l2_offset = l2_offset;
 600             m_data->valid = 1;
 601         }
 602     }
 603     cluster_offset <<= 9;
 604     return cluster_offset;
 605 }
 606
 607 static int vmdk_is_allocated(BlockDriverState *bs, int64_t sector_num,
 608                              int nb_sectors, int *pnum)
 609 {
 610     BDRVVmdkState *s = bs->opaque;
 611     int index_in_cluster, n;
 612     uint64_t cluster_offset;
 613
 614     cluster_offset = get_cluster_offset(bs, NULL, sector_num << 9, 0);
 615     index_in_cluster = sector_num % s->cluster_sectors;
 616     n = s->cluster_sectors - index_in_cluster;
 617     if (n > nb_sectors)
 618         n = nb_sectors;
 619     *pnum = n;
 620     return (cluster_offset != 0);
 621 }
 622
 623 static int vmdk_read(BlockDriverState *bs, int64_t sector_num,
 624                     uint8_t *buf, int nb_sectors)
 625 {
 626     BDRVVmdkState *s = bs->opaque;
 627     int index_in_cluster, n, ret;
 628     uint64_t cluster_offset;
 629
 630     while (nb_sectors > 0) {
 631         cluster_offset = get_cluster_offset(bs, NULL, sector_num << 9, 0);
 632         index_in_cluster = sector_num % s->cluster_sectors;
 633         n = s->cluster_sectors - index_in_cluster;
 634         if (n > nb_sectors)
 635             n = nb_sectors;
 636         if (!cluster_offset) {
 637             // try to read from parent image, if exist
 638             if (s->hd->backing_hd) {
 639                 if (!vmdk_is_cid_valid(bs))
 640                     return -1;
 641                 ret = bdrv_read(s->hd->backing_hd, sector_num, buf, n);
 642                 if (ret < 0)
 643                     return -1;
 644             } else {
 645                 memset(buf, 0, 512 * n);
 646             }
 647         } else {
 648             if(bdrv_pread(s->hd, cluster_offset + index_in_cluster * 512, buf, n * 512) != n * 512)
 649                 return -1;
 650         }
 651         nb_sectors -= n;
 652         sector_num += n;
 653         buf += n * 512;
 654         s->io.read_byte_counter += (uint64_t)(n*512);
 655     }
 656     return 0;
 657 }
 658
 659 static int vmdk_write(BlockDriverState *bs, int64_t sector_num,
 660                      const uint8_t *buf, int nb_sectors)
 661 {
 662     BDRVVmdkState *s = bs->opaque;
 663     VmdkMetaData m_data;
 664     int index_in_cluster, n;
 665     uint64_t cluster_offset;
 666     static int cid_update = 0;
 667
 668     if (sector_num > bs->total_sectors) {
 669         fprintf(stderr,
 670                 "(VMDK) Wrong offset: sector_num=0x%" PRIx64
 671                 " total_sectors=0x%" PRIx64 "\n",
 672                 sector_num, bs->total_sectors);
 673         return -1;
 674     }
 675
 676     while (nb_sectors > 0) {
 677         index_in_cluster = sector_num & (s->cluster_sectors - 1);
 678         n = s->cluster_sectors - index_in_cluster;
 679         if (n > nb_sectors)
 680             n = nb_sectors;
 681         cluster_offset = get_cluster_offset(bs, &m_data, sector_num << 9, 1);
 682         if (!cluster_offset)
 683             return -1;
 684
 685         if (bdrv_pwrite(s->hd, cluster_offset + index_in_cluster * 512, buf, n * 512) != n * 512)
 686             return -1;
 687         if (m_data.valid) {
 688             /* update L2 tables */
 689             if (vmdk_L2update(bs, &m_data) == -1)
 690                 return -1;
 691         }
 692         nb_sectors -= n;
 693         sector_num += n;
 694         buf += n * 512;
 695         s->io.write_byte_counter += (uint64_t)(n*512);
 696
 697         // update CID on the first write every time the virtual disk is opened
 698         if (!cid_update) {
 699             vmdk_write_cid(bs, time(NULL));
 700             cid_update++;
 701         }
 702     }
 703     return 0;
 704 }
 705
 706 static int vmdk_create(const char *filename, int64_t total_size,
 707                        const char *backing_file, int flags)
 708 {
 709     int fd, i;
 710     VMDK4Header header;
 711     uint32_t tmp, magic, grains, gd_size, gt_size, gt_count;
 712     char *desc_template =
 713         "# Disk DescriptorFile\n"
 714         "version=1\n"
 715         "CID=%x\n"
 716         "parentCID=ffffffff\n"
 717         "createType=\"monolithicSparse\"\n"
 718         "\n"
 719         "# Extent description\n"
 720         "RW %lu SPARSE \"%s\"\n"
 721         "\n"
 722         "# The Disk Data Base \n"
 723         "#DDB\n"
 724         "\n"
 725         "ddb.virtualHWVersion = \"%d\"\n"
 726         "ddb.geometry.cylinders = \"%lu\"\n"
 727         "ddb.geometry.heads = \"16\"\n"
 728         "ddb.geometry.sectors = \"63\"\n"
 729         "ddb.adapterType = \"ide\"\n";
 730     char desc[1024];
 731     const char *real_filename, *temp_str;
 732
 733     /* XXX: add support for backing file */
 734     if (backing_file) {
 735         return vmdk_snapshot_create(filename, backing_file);
 736     }
 737
 738     fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY | O_LARGEFILE,
 739               0644);
 740     if (fd < 0)
 741         return -1;
 742     magic = cpu_to_be32(VMDK4_MAGIC);
 743     memset(&header, 0, sizeof(header));
 744     header.version = cpu_to_le32(1);
 745     header.flags = cpu_to_le32(3); /* ?? */
 746     header.capacity = cpu_to_le64(total_size);
 747     header.granularity = cpu_to_le64(128);
 748     header.num_gtes_per_gte = cpu_to_le32(512);
 749
 750     grains = (total_size + header.granularity - 1) / header.granularity;
 751     gt_size = ((header.num_gtes_per_gte * sizeof(uint32_t)) + 511) >> 9;
 752     gt_count = (grains + header.num_gtes_per_gte - 1) / header.num_gtes_per_gte;
 753     gd_size = (gt_count * sizeof(uint32_t) + 511) >> 9;
 754
 755     header.desc_offset = 1;
 756     header.desc_size = 20;
 757     header.rgd_offset = header.desc_offset + header.desc_size;
 758     header.gd_offset = header.rgd_offset + gd_size + (gt_size * gt_count);
 759     header.grain_offset =
 760        ((header.gd_offset + gd_size + (gt_size * gt_count) +
 761          header.granularity - 1) / header.granularity) *
 762         header.granularity;
 763
 764     header.desc_offset = cpu_to_le64(header.desc_offset);
 765     header.desc_size = cpu_to_le64(header.desc_size);
 766     header.rgd_offset = cpu_to_le64(header.rgd_offset);
 767     header.gd_offset = cpu_to_le64(header.gd_offset);
 768     header.grain_offset = cpu_to_le64(header.grain_offset);
 769
 770     header.check_bytes[0] = 0xa;
 771     header.check_bytes[1] = 0x20;
 772     header.check_bytes[2] = 0xd;
 773     header.check_bytes[3] = 0xa;
 774
 775     /* write all the data */
 776     write(fd, &magic, sizeof(magic));
 777     write(fd, &header, sizeof(header));
 778
 779     ftruncate(fd, header.grain_offset << 9);
 780
 781     /* write grain directory */
 782     lseek(fd, le64_to_cpu(header.rgd_offset) << 9, SEEK_SET);
 783     for (i = 0, tmp = header.rgd_offset + gd_size;
 784          i < gt_count; i++, tmp += gt_size)
 785         write(fd, &tmp, sizeof(tmp));
 786
 787     /* write backup grain directory */
 788     lseek(fd, le64_to_cpu(header.gd_offset) << 9, SEEK_SET);
 789     for (i = 0, tmp = header.gd_offset + gd_size;
 790          i < gt_count; i++, tmp += gt_size)
 791         write(fd, &tmp, sizeof(tmp));
 792
 793     /* compose the descriptor */
 794     real_filename = filename;
 795     if ((temp_str = strrchr(real_filename, '\\')) != NULL)
 796         real_filename = temp_str + 1;
 797     if ((temp_str = strrchr(real_filename, '/')) != NULL)
 798         real_filename = temp_str + 1;
 799     if ((temp_str = strrchr(real_filename, ':')) != NULL)
 800         real_filename = temp_str + 1;
 801     sprintf(desc, desc_template, time(NULL), (unsigned long)total_size,
 802             real_filename, (flags & BLOCK_FLAG_COMPAT6 ? 6 : 4), total_size / (63 * 16));
 803
 804     /* write the descriptor */
 805     lseek(fd, le64_to_cpu(header.desc_offset) << 9, SEEK_SET);
 806     write(fd, desc, strlen(desc));
 807
 808     close(fd);
 809     return 0;
 810 }
 811
 812 static void vmdk_close(BlockDriverState *bs)
 813 {
 814     BDRVVmdkState *s = bs->opaque;
 815
 816     qemu_free(s->l1_table);
 817     qemu_free(s->l2_cache);
 818     bdrv_delete(s->hd);
 819     // try to close parent image, if exist
 820     vmdk_parent_close(s->hd);
 821 }
 822
 823 static void vmdk_flush(BlockDriverState *bs)
 824 {
 825     BDRVVmdkState *s = bs->opaque;
 826     bdrv_flush(s->hd);
 827 }
 828
 829 BlockDriver bdrv_vmdk = {
 830     "vmdk",
 831     sizeof(BDRVVmdkState),
 832     vmdk_probe,
 833     vmdk_open,
 834     vmdk_read,
 835     vmdk_write,
 836     vmdk_close,
 837     vmdk_create,
 838     vmdk_flush,
 839     vmdk_is_allocated,
 840 };