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 #include <uuid/uuid.h>
  26 #include <libgen.h>
  27 #include "vl.h"
  28 #include "block_int.h"
  29
  30 #define VMDK3_MAGIC (('C' << 24) | ('O' << 16) | ('W' << 8) | 'D')
  31 #define VMDK4_MAGIC (('K' << 24) | ('D' << 16) | ('M' << 8) | 'V')
  32
  33 typedef struct {
  34     uint32_t version;
  35     uint32_t flags;
  36     uint32_t disk_sectors;
  37     uint32_t granularity;
  38     uint32_t l1dir_offset;
  39     uint32_t l1dir_size;
  40     uint32_t file_sectors;
  41     uint32_t cylinders;
  42     uint32_t heads;
  43     uint32_t sectors_per_track;
  44 } VMDK3Header;
  45
  46 typedef struct {
  47     uint32_t version;
  48     uint32_t flags;
  49     int64_t capacity;
  50     int64_t granularity;
  51     int64_t desc_offset;
  52     int64_t desc_size;
  53     int32_t num_gtes_per_gte;
  54     int64_t rgd_offset;
  55     int64_t gd_offset;
  56     int64_t grain_offset;
  57     char filler[1];
  58     char check_bytes[4];
  59 } __attribute__((packed)) VMDK4Header;
  60
  61 #define L2_CACHE_SIZE 16
  62
  63 typedef struct BDRVVmdkState {
  64     int fd;
  65     int64_t l1_table_offset;
  66     int64_t l1_backup_table_offset;
  67     uint32_t *l1_table;
  68     uint32_t *l1_backup_table;
  69     unsigned int l1_size;
  70     uint32_t l1_entry_sectors;
  71
  72     unsigned int l2_size;
  73     uint32_t *l2_cache;
  74     uint32_t l2_cache_offsets[L2_CACHE_SIZE];
  75     uint32_t l2_cache_counts[L2_CACHE_SIZE];
  76
  77     unsigned int cluster_sectors;
  78 } BDRVVmdkState;
  79
  80 static int vmdk_probe(const uint8_t *buf, int buf_size, const char *filename)
  81 {
  82     uint32_t magic;
  83
  84     if (buf_size < 4)
  85         return 0;
  86     magic = be32_to_cpu(*(uint32_t *)buf);
  87     if (magic == VMDK3_MAGIC ||
  88         magic == VMDK4_MAGIC)
  89         return 100;
  90     else
  91         return 0;
  92 }
  93
  94 #define CHECK_CID 1
  95
  96 #define SECTOR_SIZE 512
  97 #define DESC_SIZE 20*SECTOR_SIZE        // 20 sectors of 512 bytes each
  98 #define HEADER_SIZE 512                         // first sector of 512 bytes
  99
 100 static uint32_t vmdk_read_cid(int fd, int parent)
 101 {
 102     char desc[DESC_SIZE];
 103     uint32_t cid;
 104     char *p_name, *cid_str;
 105     size_t cid_str_size;
 106
 107     if (fd) {
 108         /* the descriptor offset = 0x200 */
 109         if (lseek(fd, 0x200, SEEK_SET) == -1)
 110             return 0;
 111         if (read(fd, desc, DESC_SIZE) != DESC_SIZE)
 112             return 0;
 113
 114         if (parent) {
 115             cid_str = "parentCID";
 116             cid_str_size = sizeof("parentCID");
 117         } else {
 118             cid_str = "CID";
 119             cid_str_size = sizeof("CID");
 120         }
 121
 122         if ((p_name = strstr(desc,cid_str)) != 0) {
 123             p_name += cid_str_size;
 124             sscanf(p_name,"%x",&cid);
 125         }
 126
 127         return cid;
 128     }
 129     return 0;
 130 }
 131
 132 static int vmdk_write_cid(BlockDriverState *bs, uint32_t cid)
 133 {
 134     BDRVVmdkState *s = bs->opaque;
 135     char desc[DESC_SIZE], tmp_desc[DESC_SIZE];
 136     char *p_name, *tmp_str;
 137
 138     if (s->fd) {
 139         /* the descriptor offset = 0x200 */
 140         if (lseek(s->fd, 0x200, SEEK_SET) == -1)
 141             return -1;
 142         if (read(s->fd, desc, DESC_SIZE) != DESC_SIZE)
 143             return 0;
 144
 145         tmp_str = strstr(desc,"parentCID");
 146         strcpy(tmp_desc, tmp_str);
 147         if ((p_name = strstr(desc,"CID")) != 0) {
 148             p_name += sizeof("CID");
 149             sprintf(p_name,"%x\n",cid);
 150             strcat(desc,tmp_desc);
 151         }
 152
 153         if (lseek(s->fd, 0x200, SEEK_SET) == -1)
 154             return -1;
 155         if (write(s->fd, &desc, DESC_SIZE) != DESC_SIZE)
 156             return -1;
 157
 158         bs->cid = cid;
 159         return 0;
 160     }
 161     return -1;
 162 }
 163
 164 static int vmdk_is_cid_valid(BlockDriverState *bs)
 165 {
 166 #ifdef CHECK_CID
 167     BlockDriverState *p_bs = bs->bs_par_table;
 168     uint32_t cur_pcid;
 169
 170     if (p_bs) {
 171         BDRVVmdkState *s = p_bs->opaque;
 172
 173         cur_pcid = vmdk_read_cid(s->fd,0);
 174         if (bs->parent_cid != cur_pcid)
 175             // CID not valid
 176             return 0;
 177     }
 178 #endif
 179     // CID valid
 180     return 1;
 181 }
 182
 183 int vmdk_snapshot_create(BlockDriverState *bs)
 184 {
 185     int snp_fd, p_fd;
 186     uint32_t p_cid;
 187     char *p_name, *gd_buf, *rgd_buf;
 188     VMDK4Header header;
 189     uint32_t gde_entries, gd_size;
 190     int64_t gd_offset, rgd_offset, capacity, gt_size;
 191     char p_desc[DESC_SIZE], s_desc[DESC_SIZE], hdr[HEADER_SIZE];
 192     char parent_filename[1024];
 193     char snapshot_filename[1024];
 194     uuid_t name;
 195     char *desc_template =
 196     "# Disk DescriptorFile\n"
 197     "version=1\n"
 198     "CID=%x\n"
 199     "parentCID=%x\n"
 200     "createType=\"monolithicSparse\"\n"
 201     "parentFileNameHint=\"%s\"\n"
 202     "\n"
 203     "# Extent description\n"
 204     "RW %lu SPARSE \"%s\"\n"
 205     "\n"
 206     "# The Disk Data Base \n"
 207     "#DDB\n"
 208     "\n";
 209
 210     strcpy(parent_filename, bs->filename);
 211     uuid_generate(name);     // it should be unique filename
 212     uuid_unparse(name, snapshot_filename);
 213     strcat(snapshot_filename,".vmdk");
 214
 215     snp_fd = open(snapshot_filename, O_RDWR | O_CREAT | O_TRUNC | O_BINARY | O_LARGEFILE, 0644);
 216     if (snp_fd < 0)
 217         return -1;
 218     p_fd = open(parent_filename, O_RDONLY | O_BINARY | O_LARGEFILE);
 219     if (p_fd < 0) {
 220         close(snp_fd);
 221         return -1;
 222     }
 223
 224     if (lseek(p_fd, 0x0, SEEK_SET) == -1)
 225         goto fail;
 226     if (read(p_fd, hdr, HEADER_SIZE) != HEADER_SIZE)
 227         goto fail;
 228
 229     /* write the header */
 230     if (lseek(snp_fd, 0x0, SEEK_SET) == -1)
 231         goto fail;
 232     if (write(snp_fd, hdr, HEADER_SIZE) == -1)
 233         goto fail;
 234
 235     memset(&header, 0, sizeof(header));
 236     memcpy(&header,&hdr[4], sizeof(header)); // skip the VMDK4_MAGIC
 237
 238     ftruncate(snp_fd, header.grain_offset << 9);
 239     /* the descriptor offset = 0x200 */
 240     if (lseek(p_fd, 0x200, SEEK_SET) == -1)
 241         goto fail;
 242     if (read(p_fd, p_desc, DESC_SIZE) != DESC_SIZE)
 243         goto fail;
 244
 245     if ((p_name = strstr(p_desc,"CID")) != 0) {
 246         p_name += sizeof("CID");
 247         sscanf(p_name,"%x",&p_cid);
 248     }
 249     sprintf(s_desc, desc_template, p_cid, p_cid, parent_filename
 250             , (uint32_t)header.capacity, snapshot_filename);
 251
 252     bs->parent_cid = p_cid;
 253
 254     /* write the descriptor */
 255     if (lseek(snp_fd, 0x200, SEEK_SET) == -1)
 256         goto fail;
 257     if (write(snp_fd, s_desc, strlen(s_desc)) == -1)
 258         goto fail;
 259
 260     gd_offset = header.gd_offset * SECTOR_SIZE;     // offset of GD table
 261     rgd_offset = header.rgd_offset * SECTOR_SIZE;   // offset of RGD table
 262     capacity = header.capacity * SECTOR_SIZE;       // Extent size
 263     /*
 264      * Each GDE span 32M disk, means:
 265      * 512 GTE per GT, each GTE points to grain
 266      */
 267     gt_size = (int64_t)header.num_gtes_per_gte * header.granularity * SECTOR_SIZE;
 268     if (!gt_size)
 269         goto fail;
 270     gde_entries = (uint32_t)(capacity / gt_size);  // number of gde/rgde
 271     gd_size = gde_entries * sizeof(uint32_t);
 272
 273     /* write RGD */
 274     rgd_buf = qemu_malloc(gd_size);
 275     if (!rgd_buf)
 276         goto fail;
 277     if (lseek(p_fd, rgd_offset, SEEK_SET) == -1)
 278         goto fail_rgd;
 279     if (read(p_fd, rgd_buf, gd_size) != gd_size)
 280         goto fail_rgd;
 281     if (lseek(snp_fd, rgd_offset, SEEK_SET) == -1)
 282         goto fail_rgd;
 283     if (write(snp_fd, rgd_buf, gd_size) == -1)
 284         goto fail_rgd;
 285     qemu_free(rgd_buf);
 286
 287     /* write GD */
 288     gd_buf = qemu_malloc(gd_size);
 289     if (!gd_buf)
 290         goto fail_rgd;
 291     if (lseek(p_fd, gd_offset, SEEK_SET) == -1)
 292         goto fail_gd;
 293     if (read(p_fd, gd_buf, gd_size) != gd_size)
 294         goto fail_gd;
 295     if (lseek(snp_fd, gd_offset, SEEK_SET) == -1)
 296         goto fail_gd;
 297     if (write(snp_fd, gd_buf, gd_size) == -1)
 298         goto fail_gd;
 299     qemu_free(gd_buf);
 300
 301     close(p_fd);
 302     close(snp_fd);
 303     return 0;
 304
 305     fail_gd:
 306     qemu_free(gd_buf);
 307     fail_rgd:
 308     qemu_free(rgd_buf);
 309     fail:
 310     close(p_fd);
 311     close(snp_fd);
 312     return -1;
 313 }
 314
 315 static void vmdk_parent_close(BlockDriverState *bs)
 316 {
 317     if (bs->bs_par_table)
 318         bdrv_close(bs->bs_par_table);
 319 }
 320
 321
 322 static int vmdk_parent_open(BlockDriverState *bs, int fd, char * dir_name)
 323 {
 324     char *p_name;
 325     char desc[DESC_SIZE];
 326     static int idx=0;
 327     char parent_img_name[1024];
 328
 329     /* the descriptor offset = 0x200 */
 330     if (lseek(fd, 0x200, SEEK_SET) == -1)
 331         return -1;
 332     if (read(fd, desc, DESC_SIZE) != DESC_SIZE)
 333         return -1;
 334
 335     if ((p_name = strstr(desc,"parentFileNameHint")) != 0) {
 336         char *end_name, *tmp_name;
 337         char name[256], buf[128];
 338         int name_size;
 339         struct stat file_buf;
 340
 341         p_name += sizeof("parentFileNameHint") + 1;
 342         if ((end_name = strchr(p_name,'\"')) == 0)
 343             return -1;
 344
 345         bs->parent_img_name = qemu_mallocz(end_name - p_name + 2);
 346         strncpy(bs->parent_img_name, p_name, end_name - p_name);
 347         if (stat(bs->parent_img_name, &file_buf) != 0) {
 348             strcpy(parent_img_name, dir_name);
 349             strcat(parent_img_name, basename(bs->parent_img_name));
 350         } else {
 351             strcpy(parent_img_name, bs->parent_img_name);
 352         }
 353
 354         tmp_name = strstr(bs->device_name,"_QEMU");
 355         name_size = tmp_name ? (tmp_name - bs->device_name) : sizeof(bs->device_name);
 356         strncpy(name,bs->device_name,name_size);
 357         sprintf(buf, "_QEMU_%d", ++idx);
 358
 359         bs->bs_par_table = bdrv_new(strcat(name, buf));
 360         if (!bs->bs_par_table) {
 361             failure:
 362             bdrv_close(bs);
 363             return -1;
 364         }
 365
 366         if (bdrv_open(bs->bs_par_table, parent_img_name, 0) < 0)
 367             goto failure;
 368     }
 369
 370     return 0;
 371 }
 372
 373 static int vmdk_open(BlockDriverState *bs, const char *filename)
 374 {
 375     BDRVVmdkState *s = bs->opaque;
 376     int fd, i;
 377     uint32_t magic;
 378     int l1_size;
 379
 380     fd = open(filename, O_RDWR | O_BINARY | O_LARGEFILE);
 381     if (fd < 0) {
 382         fd = open(filename, O_RDONLY | O_BINARY | O_LARGEFILE);
 383         if (fd < 0)
 384             return -1;
 385         bs->read_only = 1;
 386     }
 387     if (read(fd, &magic, sizeof(magic)) != sizeof(magic))
 388         goto fail;
 389     magic = be32_to_cpu(magic);
 390     if (magic == VMDK3_MAGIC) {
 391         VMDK3Header header;
 392         if (read(fd, &header, sizeof(header)) !=
 393             sizeof(header))
 394             goto fail;
 395         s->cluster_sectors = le32_to_cpu(header.granularity);
 396         s->l2_size = 1 << 9;
 397         s->l1_size = 1 << 6;
 398         bs->total_sectors = le32_to_cpu(header.disk_sectors);
 399         s->l1_table_offset = le32_to_cpu(header.l1dir_offset) << 9;
 400         s->l1_backup_table_offset = 0;
 401         s->l1_entry_sectors = s->l2_size * s->cluster_sectors;
 402     } else if (magic == VMDK4_MAGIC) {
 403         char dir_name[1024];
 404         VMDK4Header header;
 405
 406         if (read(fd, &header, sizeof(header)) != sizeof(header))
 407             goto fail;
 408         bs->total_sectors = le64_to_cpu(header.capacity);
 409         s->cluster_sectors = le64_to_cpu(header.granularity);
 410         s->l2_size = le32_to_cpu(header.num_gtes_per_gte);
 411         s->l1_entry_sectors = s->l2_size * s->cluster_sectors;
 412         if (s->l1_entry_sectors <= 0)
 413             goto fail;
 414         s->l1_size = (bs->total_sectors + s->l1_entry_sectors - 1)
 415             / s->l1_entry_sectors;
 416         s->l1_table_offset = le64_to_cpu(header.rgd_offset) << 9;
 417         s->l1_backup_table_offset = le64_to_cpu(header.gd_offset) << 9;
 418
 419         // try to open parent images, if exist
 420         strcpy(dir_name, dirname(filename));
 421         strcat(dir_name,"/");
 422         if (vmdk_parent_open(bs, fd, dir_name) != 0)
 423             goto fail;
 424         // write the CID once after the image creation
 425         bs->cid = vmdk_read_cid(fd,0);
 426         bs->parent_cid = vmdk_read_cid(fd,1);
 427     } else {
 428         goto fail;
 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 (lseek(fd, s->l1_table_offset, SEEK_SET) == -1)
 436         goto fail;
 437     if (read(fd, s->l1_table, l1_size) != l1_size)
 438         goto fail;
 439     for(i = 0; i < s->l1_size; i++) {
 440         le32_to_cpus(&s->l1_table[i]);
 441     }
 442
 443     if (s->l1_backup_table_offset) {
 444         s->l1_backup_table = qemu_malloc(l1_size);
 445         if (!s->l1_backup_table)
 446             goto fail;
 447         if (lseek(fd, s->l1_backup_table_offset, SEEK_SET) == -1)
 448             goto fail;
 449         if (read(fd, s->l1_backup_table, l1_size) != l1_size)
 450             goto fail;
 451         for(i = 0; i < s->l1_size; i++) {
 452             le32_to_cpus(&s->l1_backup_table[i]);
 453         }
 454     }
 455
 456     s->l2_cache = qemu_malloc(s->l2_size * L2_CACHE_SIZE * sizeof(uint32_t));
 457     if (!s->l2_cache)
 458         goto fail;
 459     s->fd = fd;
 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     close(fd);
 466     return -1;
 467 }
 468
 469 static uint64_t get_cluster_offset(BlockDriverState *bs, uint64_t offset, int allocate);
 470
 471 static int get_whole_cluster(BlockDriverState *bs, uint64_t cluster_offset,
 472                              uint64_t offset, int allocate)
 473 {
 474     int ret;
 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 (bs->bs_par_table) {
 482         BDRVVmdkState *ps = bs->bs_par_table->opaque;
 483
 484         if (!vmdk_is_cid_valid(bs))
 485             return -1;
 486
 487         parent_cluster_offset = get_cluster_offset(bs->bs_par_table, offset, allocate);
 488
 489         lseek(ps->fd, parent_cluster_offset, SEEK_SET);
 490         ret = read(ps->fd, whole_grain, ps->cluster_sectors*512);
 491         if (ret != ps->cluster_sectors*512)
 492             return -1;
 493
 494         lseek(s->fd, cluster_offset << 9, SEEK_SET);
 495         ret = write(s->fd, whole_grain, sizeof(whole_grain));
 496         if (ret != sizeof(whole_grain))
 497             return -1;
 498     }
 499     return 0;
 500 }
 501
 502 static uint64_t get_cluster_offset(BlockDriverState *bs,
 503                                    uint64_t offset, int allocate)
 504 {
 505     BDRVVmdkState *s = bs->opaque;
 506     unsigned int l1_index, l2_offset, l2_index;
 507     int min_index, i, j;
 508     uint32_t min_count, *l2_table, tmp;
 509     uint64_t cluster_offset;
 510
 511     l1_index = (offset >> 9) / s->l1_entry_sectors;
 512     if (l1_index >= s->l1_size)
 513         return 0;
 514     l2_offset = s->l1_table[l1_index];
 515     if (!l2_offset)
 516         return 0;
 517     for(i = 0; i < L2_CACHE_SIZE; i++) {
 518         if (l2_offset == s->l2_cache_offsets[i]) {
 519             /* increment the hit count */
 520             if (++s->l2_cache_counts[i] == 0xffffffff) {
 521                 for(j = 0; j < L2_CACHE_SIZE; j++) {
 522                     s->l2_cache_counts[j] >>= 1;
 523                 }
 524             }
 525             l2_table = s->l2_cache + (i * s->l2_size);
 526             goto found;
 527         }
 528     }
 529     /* not found: load a new entry in the least used one */
 530     min_index = 0;
 531     min_count = 0xffffffff;
 532     for(i = 0; i < L2_CACHE_SIZE; i++) {
 533         if (s->l2_cache_counts[i] < min_count) {
 534             min_count = s->l2_cache_counts[i];
 535             min_index = i;
 536         }
 537     }
 538     l2_table = s->l2_cache + (min_index * s->l2_size);
 539     lseek(s->fd, (int64_t)l2_offset * 512, SEEK_SET);
 540     if (read(s->fd, l2_table, s->l2_size * sizeof(uint32_t)) !=
 541         s->l2_size * sizeof(uint32_t))
 542         return 0;
 543     s->l2_cache_offsets[min_index] = l2_offset;
 544     s->l2_cache_counts[min_index] = 1;
 545  found:
 546     l2_index = ((offset >> 9) / s->cluster_sectors) % s->l2_size;
 547     cluster_offset = le32_to_cpu(l2_table[l2_index]);
 548     if (!cluster_offset) {
 549         if (!allocate)
 550             return 0;
 551         cluster_offset = lseek(s->fd, 0, SEEK_END);
 552         ftruncate(s->fd, cluster_offset + (s->cluster_sectors << 9));
 553         cluster_offset >>= 9;
 554         /* update L2 table */
 555         tmp = cpu_to_le32(cluster_offset);
 556         l2_table[l2_index] = tmp;
 557         lseek(s->fd, ((int64_t)l2_offset * 512) + (l2_index * sizeof(tmp)), SEEK_SET);
 558         if (write(s->fd, &tmp, sizeof(tmp)) != sizeof(tmp))
 559             return 0;
 560         /* update backup L2 table */
 561         if (s->l1_backup_table_offset != 0) {
 562             l2_offset = s->l1_backup_table[l1_index];
 563             lseek(s->fd, ((int64_t)l2_offset * 512) + (l2_index * sizeof(tmp)), SEEK_SET);
 564             if (write(s->fd, &tmp, sizeof(tmp)) != sizeof(tmp))
 565                 return 0;
 566         }
 567
 568         if (get_whole_cluster(bs, cluster_offset, offset, allocate) == -1)
 569             return 0;
 570     }
 571     cluster_offset <<= 9;
 572     return cluster_offset;
 573 }
 574
 575 static int vmdk_is_allocated(BlockDriverState *bs, int64_t sector_num,
 576                              int nb_sectors, int *pnum)
 577 {
 578     BDRVVmdkState *s = bs->opaque;
 579     int index_in_cluster, n;
 580     uint64_t cluster_offset;
 581
 582     cluster_offset = get_cluster_offset(bs, sector_num << 9, 0);
 583     index_in_cluster = sector_num % s->cluster_sectors;
 584     n = s->cluster_sectors - index_in_cluster;
 585     if (n > nb_sectors)
 586         n = nb_sectors;
 587     *pnum = n;
 588     return (cluster_offset != 0);
 589 }
 590
 591 static int vmdk_read(BlockDriverState *bs, int64_t sector_num,
 592                     uint8_t *buf, int nb_sectors)
 593 {
 594     BDRVVmdkState *s = bs->opaque;
 595     int ret, index_in_cluster, n;
 596     uint64_t cluster_offset;
 597
 598     while (nb_sectors > 0) {
 599         cluster_offset = get_cluster_offset(bs, sector_num << 9, 0);
 600         index_in_cluster = sector_num % s->cluster_sectors;
 601         n = s->cluster_sectors - index_in_cluster;
 602         if (n > nb_sectors)
 603             n = nb_sectors;
 604         if (!cluster_offset) {
 605             // try to read from parent image, if exist
 606             if (bs->bs_par_table) {
 607                 if (!vmdk_is_cid_valid(bs))
 608                     return -1;
 609                 if (vmdk_read(bs->bs_par_table, sector_num, buf, nb_sectors) == -1)
 610                     return -1;
 611             } else {
 612                 memset(buf, 0, 512 * n);
 613             }
 614         } else {
 615             lseek(s->fd, cluster_offset + index_in_cluster * 512, SEEK_SET);
 616             ret = read(s->fd, buf, n * 512);
 617             if (ret != n * 512)
 618                 return -1;
 619         }
 620         nb_sectors -= n;
 621         sector_num += n;
 622         buf += n * 512;
 623     }
 624     return 0;
 625 }
 626
 627 static int vmdk_write(BlockDriverState *bs, int64_t sector_num,
 628                      const uint8_t *buf, int nb_sectors)
 629 {
 630     BDRVVmdkState *s = bs->opaque;
 631     int ret, index_in_cluster, n;
 632     uint64_t cluster_offset;
 633     static int cid_update = 0;
 634
 635     while (nb_sectors > 0) {
 636         index_in_cluster = sector_num & (s->cluster_sectors - 1);
 637         n = s->cluster_sectors - index_in_cluster;
 638         if (n > nb_sectors)
 639             n = nb_sectors;
 640         cluster_offset = get_cluster_offset(bs, sector_num << 9, 1);
 641         if (!cluster_offset)
 642             return -1;
 643         lseek(s->fd, cluster_offset + index_in_cluster * 512, SEEK_SET);
 644         ret = write(s->fd, buf, n * 512);
 645         if (ret != n * 512)
 646             return -1;
 647         nb_sectors -= n;
 648         sector_num += n;
 649         buf += n * 512;
 650
 651         // update CID on the first write every time the virtual disk is opened
 652         if (!cid_update) {
 653             vmdk_write_cid(bs, time(NULL));
 654             cid_update++;
 655         }
 656     }
 657     return 0;
 658 }
 659
 660 static int vmdk_create(const char *filename, int64_t total_size,
 661                        const char *backing_file, int flags)
 662 {
 663     int fd, i;
 664     VMDK4Header header;
 665     uint32_t tmp, magic, grains, gd_size, gt_size, gt_count;
 666     char *desc_template =
 667         "# Disk DescriptorFile\n"
 668         "version=1\n"
 669         "CID=%x\n"
 670         "parentCID=ffffffff\n"
 671         "createType=\"monolithicSparse\"\n"
 672         "\n"
 673         "# Extent description\n"
 674         "RW %lu SPARSE \"%s\"\n"
 675         "\n"
 676         "# The Disk Data Base \n"
 677         "#DDB\n"
 678         "\n"
 679         "ddb.virtualHWVersion = \"4\"\n"
 680         "ddb.geometry.cylinders = \"%lu\"\n"
 681         "ddb.geometry.heads = \"16\"\n"
 682         "ddb.geometry.sectors = \"63\"\n"
 683         "ddb.adapterType = \"ide\"\n";
 684     char desc[1024];
 685     const char *real_filename, *temp_str;
 686
 687     /* XXX: add support for backing file */
 688
 689     fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY | O_LARGEFILE,
 690               0644);
 691     if (fd < 0)
 692         return -1;
 693     magic = cpu_to_be32(VMDK4_MAGIC);
 694     memset(&header, 0, sizeof(header));
 695     header.version = cpu_to_le32(1);
 696     header.flags = cpu_to_le32(3); /* ?? */
 697     header.capacity = cpu_to_le64(total_size);
 698     header.granularity = cpu_to_le64(128);
 699     header.num_gtes_per_gte = cpu_to_le32(512);
 700
 701     grains = (total_size + header.granularity - 1) / header.granularity;
 702     gt_size = ((header.num_gtes_per_gte * sizeof(uint32_t)) + 511) >> 9;
 703     gt_count = (grains + header.num_gtes_per_gte - 1) / header.num_gtes_per_gte;
 704     gd_size = (gt_count * sizeof(uint32_t) + 511) >> 9;
 705
 706     header.desc_offset = 1;
 707     header.desc_size = 20;
 708     header.rgd_offset = header.desc_offset + header.desc_size;
 709     header.gd_offset = header.rgd_offset + gd_size + (gt_size * gt_count);
 710     header.grain_offset =
 711        ((header.gd_offset + gd_size + (gt_size * gt_count) +
 712          header.granularity - 1) / header.granularity) *
 713         header.granularity;
 714
 715     header.desc_offset = cpu_to_le64(header.desc_offset);
 716     header.desc_size = cpu_to_le64(header.desc_size);
 717     header.rgd_offset = cpu_to_le64(header.rgd_offset);
 718     header.gd_offset = cpu_to_le64(header.gd_offset);
 719     header.grain_offset = cpu_to_le64(header.grain_offset);
 720
 721     header.check_bytes[0] = 0xa;
 722     header.check_bytes[1] = 0x20;
 723     header.check_bytes[2] = 0xd;
 724     header.check_bytes[3] = 0xa;
 725
 726     /* write all the data */
 727     write(fd, &magic, sizeof(magic));
 728     write(fd, &header, sizeof(header));
 729
 730     ftruncate(fd, header.grain_offset << 9);
 731
 732     /* write grain directory */
 733     lseek(fd, le64_to_cpu(header.rgd_offset) << 9, SEEK_SET);
 734     for (i = 0, tmp = header.rgd_offset + gd_size;
 735          i < gt_count; i++, tmp += gt_size)
 736         write(fd, &tmp, sizeof(tmp));
 737
 738     /* write backup grain directory */
 739     lseek(fd, le64_to_cpu(header.gd_offset) << 9, SEEK_SET);
 740     for (i = 0, tmp = header.gd_offset + gd_size;
 741          i < gt_count; i++, tmp += gt_size)
 742         write(fd, &tmp, sizeof(tmp));
 743
 744     /* compose the descriptor */
 745     real_filename = filename;
 746     if ((temp_str = strrchr(real_filename, '\\')) != NULL)
 747         real_filename = temp_str + 1;
 748     if ((temp_str = strrchr(real_filename, '/')) != NULL)
 749         real_filename = temp_str + 1;
 750     if ((temp_str = strrchr(real_filename, ':')) != NULL)
 751         real_filename = temp_str + 1;
 752     sprintf(desc, desc_template, time(NULL), (unsigned long)total_size,
 753             real_filename, total_size / (63 * 16));
 754
 755     /* write the descriptor */
 756     lseek(fd, le64_to_cpu(header.desc_offset) << 9, SEEK_SET);
 757     write(fd, desc, strlen(desc));
 758
 759     close(fd);
 760     return 0;
 761 }
 762
 763 static void vmdk_close(BlockDriverState *bs)
 764 {
 765     BDRVVmdkState *s = bs->opaque;
 766
 767     qemu_free(s->l1_table);
 768     qemu_free(s->l2_cache);
 769     close(s->fd);
 770     // try to close parent image, if exist
 771     vmdk_parent_close(bs);
 772 }
 773
 774 static void vmdk_flush(BlockDriverState *bs)
 775 {
 776     BDRVVmdkState *s = bs->opaque;
 777     fsync(s->fd);
 778 }
 779
 780 BlockDriver bdrv_vmdk = {
 781     "vmdk",
 782     sizeof(BDRVVmdkState),
 783     vmdk_probe,
 784     vmdk_open,
 785     vmdk_read,
 786     vmdk_write,
 787     vmdk_close,
 788     vmdk_create,
 789     vmdk_flush,
 790     vmdk_is_allocated,
 791 };