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 #include "module.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     BlockDriverState *hd;
  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     uint32_t parent_cid;
  79     int is_parent;
  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
  98 static int vmdk_probe(const uint8_t *buf, int buf_size, const char *filename)
  99 {
 100     uint32_t magic;
 101
 102     if (buf_size < 4)
 103         return 0;
 104     magic = be32_to_cpu(*(uint32_t *)buf);
 105     if (magic == VMDK3_MAGIC ||
 106         magic == VMDK4_MAGIC)
 107         return 100;
 108     else
 109         return 0;
 110 }
 111
 112 #define CHECK_CID 1
 113
 114 #define SECTOR_SIZE 512
 115 #define DESC_SIZE 20*SECTOR_SIZE        // 20 sectors of 512 bytes each
 116 #define HEADER_SIZE 512                         // first sector of 512 bytes
 117
 118 static uint32_t vmdk_read_cid(BlockDriverState *bs, int parent)
 119 {
 120     BDRVVmdkState *s = bs->opaque;
 121     char desc[DESC_SIZE];
 122     uint32_t cid;
 123     const char *p_name, *cid_str;
 124     size_t cid_str_size;
 125
 126     /* the descriptor offset = 0x200 */
 127     if (bdrv_pread(s->hd, 0x200, desc, DESC_SIZE) != DESC_SIZE)
 128         return 0;
 129
 130     if (parent) {
 131         cid_str = "parentCID";
 132         cid_str_size = sizeof("parentCID");
 133     } else {
 134         cid_str = "CID";
 135         cid_str_size = sizeof("CID");
 136     }
 137
 138     if ((p_name = strstr(desc,cid_str)) != NULL) {
 139         p_name += cid_str_size;
 140         sscanf(p_name,"%x",&cid);
 141     }
 142
 143     return cid;
 144 }
 145
 146 static int vmdk_write_cid(BlockDriverState *bs, uint32_t cid)
 147 {
 148     BDRVVmdkState *s = bs->opaque;
 149     char desc[DESC_SIZE], tmp_desc[DESC_SIZE];
 150     char *p_name, *tmp_str;
 151
 152     /* the descriptor offset = 0x200 */
 153     if (bdrv_pread(s->hd, 0x200, desc, DESC_SIZE) != DESC_SIZE)
 154         return -1;
 155
 156     tmp_str = strstr(desc,"parentCID");
 157     pstrcpy(tmp_desc, sizeof(tmp_desc), tmp_str);
 158     if ((p_name = strstr(desc,"CID")) != NULL) {
 159         p_name += sizeof("CID");
 160         snprintf(p_name, sizeof(desc) - (p_name - desc), "%x\n", cid);
 161         pstrcat(desc, sizeof(desc), tmp_desc);
 162     }
 163
 164     if (bdrv_pwrite(s->hd, 0x200, desc, DESC_SIZE) != DESC_SIZE)
 165         return -1;
 166     return 0;
 167 }
 168
 169 static int vmdk_is_cid_valid(BlockDriverState *bs)
 170 {
 171 #ifdef CHECK_CID
 172     BDRVVmdkState *s = bs->opaque;
 173     BlockDriverState *p_bs = s->hd->backing_hd;
 174     uint32_t cur_pcid;
 175
 176     if (p_bs) {
 177         cur_pcid = vmdk_read_cid(p_bs,0);
 178         if (s->parent_cid != cur_pcid)
 179             // CID not valid
 180             return 0;
 181     }
 182 #endif
 183     // CID valid
 184     return 1;
 185 }
 186
 187 static int vmdk_snapshot_create(const char *filename, const char *backing_file)
 188 {
 189     int snp_fd, p_fd;
 190     uint32_t p_cid;
 191     char *p_name, *gd_buf, *rgd_buf;
 192     const char *real_filename, *temp_str;
 193     VMDK4Header header;
 194     uint32_t gde_entries, gd_size;
 195     int64_t gd_offset, rgd_offset, capacity, gt_size;
 196     char p_desc[DESC_SIZE], s_desc[DESC_SIZE], hdr[HEADER_SIZE];
 197     static const char desc_template[] =
 198     "# Disk DescriptorFile\n"
 199     "version=1\n"
 200     "CID=%x\n"
 201     "parentCID=%x\n"
 202     "createType=\"monolithicSparse\"\n"
 203     "parentFileNameHint=\"%s\"\n"
 204     "\n"
 205     "# Extent description\n"
 206     "RW %u SPARSE \"%s\"\n"
 207     "\n"
 208     "# The Disk Data Base \n"
 209     "#DDB\n"
 210     "\n";
 211
 212     snp_fd = open(filename, O_RDWR | O_CREAT | O_TRUNC | O_BINARY | O_LARGEFILE, 0644);
 213     if (snp_fd < 0)
 214         return -1;
 215     p_fd = open(backing_file, O_RDONLY | O_BINARY | O_LARGEFILE);
 216     if (p_fd < 0) {
 217         close(snp_fd);
 218         return -1;
 219     }
 220
 221     /* read the header */
 222     if (lseek(p_fd, 0x0, SEEK_SET) == -1)
 223         goto fail;
 224     if (read(p_fd, hdr, HEADER_SIZE) != HEADER_SIZE)
 225         goto fail;
 226
 227     /* write the header */
 228     if (lseek(snp_fd, 0x0, SEEK_SET) == -1)
 229         goto fail;
 230     if (write(snp_fd, hdr, HEADER_SIZE) == -1)
 231         goto fail;
 232
 233     memset(&header, 0, sizeof(header));
 234     memcpy(&header,&hdr[4], sizeof(header)); // skip the VMDK4_MAGIC
 235
 236     ftruncate(snp_fd, header.grain_offset << 9);
 237     /* the descriptor offset = 0x200 */
 238     if (lseek(p_fd, 0x200, SEEK_SET) == -1)
 239         goto fail;
 240     if (read(p_fd, p_desc, DESC_SIZE) != DESC_SIZE)
 241         goto fail;
 242
 243     if ((p_name = strstr(p_desc,"CID")) != NULL) {
 244         p_name += sizeof("CID");
 245         sscanf(p_name,"%x",&p_cid);
 246     }
 247
 248     real_filename = filename;
 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     if ((temp_str = strrchr(real_filename, ':')) != NULL)
 254         real_filename = temp_str + 1;
 255
 256     snprintf(s_desc, sizeof(s_desc), desc_template, p_cid, p_cid, backing_file,
 257              (uint32_t)header.capacity, real_filename);
 258
 259     /* write the descriptor */
 260     if (lseek(snp_fd, 0x200, SEEK_SET) == -1)
 261         goto fail;
 262     if (write(snp_fd, s_desc, strlen(s_desc)) == -1)
 263         goto fail;
 264
 265     gd_offset = header.gd_offset * SECTOR_SIZE;     // offset of GD table
 266     rgd_offset = header.rgd_offset * SECTOR_SIZE;   // offset of RGD table
 267     capacity = header.capacity * SECTOR_SIZE;       // Extent size
 268     /*
 269      * Each GDE span 32M disk, means:
 270      * 512 GTE per GT, each GTE points to grain
 271      */
 272     gt_size = (int64_t)header.num_gtes_per_gte * header.granularity * SECTOR_SIZE;
 273     if (!gt_size)
 274         goto fail;
 275     gde_entries = (uint32_t)(capacity / gt_size);  // number of gde/rgde
 276     gd_size = gde_entries * sizeof(uint32_t);
 277
 278     /* write RGD */
 279     rgd_buf = qemu_malloc(gd_size);
 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 (lseek(p_fd, gd_offset, SEEK_SET) == -1)
 293         goto fail_gd;
 294     if (read(p_fd, gd_buf, gd_size) != gd_size)
 295         goto fail_gd;
 296     if (lseek(snp_fd, gd_offset, SEEK_SET) == -1)
 297         goto fail_gd;
 298     if (write(snp_fd, gd_buf, gd_size) == -1)
 299         goto fail_gd;
 300     qemu_free(gd_buf);
 301
 302     close(p_fd);
 303     close(snp_fd);
 304     return 0;
 305
 306     fail_gd:
 307     qemu_free(gd_buf);
 308     fail_rgd:
 309     qemu_free(rgd_buf);
 310     fail:
 311     close(p_fd);
 312     close(snp_fd);
 313     return -1;
 314 }
 315
 316 static void vmdk_parent_close(BlockDriverState *bs)
 317 {
 318     if (bs->backing_hd)
 319         bdrv_close(bs->backing_hd);
 320 }
 321
 322 static int parent_open = 0;
 323 static int vmdk_parent_open(BlockDriverState *bs, const char * filename)
 324 {
 325     BDRVVmdkState *s = bs->opaque;
 326     char *p_name;
 327     char desc[DESC_SIZE];
 328     char parent_img_name[1024];
 329
 330     /* the descriptor offset = 0x200 */
 331     if (bdrv_pread(s->hd, 0x200, desc, DESC_SIZE) != DESC_SIZE)
 332         return -1;
 333
 334     if ((p_name = strstr(desc,"parentFileNameHint")) != NULL) {
 335         char *end_name;
 336         struct stat file_buf;
 337
 338         p_name += sizeof("parentFileNameHint") + 1;
 339         if ((end_name = strchr(p_name,'\"')) == NULL)
 340             return -1;
 341         if ((end_name - p_name) > sizeof (s->hd->backing_file) - 1)
 342             return -1;
 343
 344         pstrcpy(s->hd->backing_file, end_name - p_name + 1, p_name);
 345         if (stat(s->hd->backing_file, &file_buf) != 0) {
 346             path_combine(parent_img_name, sizeof(parent_img_name),
 347                          filename, s->hd->backing_file);
 348         } else {
 349             pstrcpy(parent_img_name, sizeof(parent_img_name),
 350                     s->hd->backing_file);
 351         }
 352
 353         s->hd->backing_hd = bdrv_new("");
 354         if (!s->hd->backing_hd) {
 355             failure:
 356             bdrv_close(s->hd);
 357             return -1;
 358         }
 359         parent_open = 1;
 360         if (bdrv_open(s->hd->backing_hd, parent_img_name, BDRV_O_RDONLY) < 0)
 361             goto failure;
 362         parent_open = 0;
 363     }
 364
 365     return 0;
 366 }
 367
 368 static int vmdk_open(BlockDriverState *bs, const char *filename, int flags)
 369 {
 370     BDRVVmdkState *s = bs->opaque;
 371     uint32_t magic;
 372     int l1_size, i, ret;
 373
 374     if (parent_open)
 375         // Parent must be opened as RO.
 376         flags = BDRV_O_RDONLY;
 377
 378     ret = bdrv_file_open(&s->hd, filename, flags);
 379     if (ret < 0)
 380         return ret;
 381     if (bdrv_pread(s->hd, 0, &magic, sizeof(magic)) != sizeof(magic))
 382         goto fail;
 383
 384     magic = be32_to_cpu(magic);
 385     if (magic == VMDK3_MAGIC) {
 386         VMDK3Header header;
 387
 388         if (bdrv_pread(s->hd, sizeof(magic), &header, sizeof(header)) != sizeof(header))
 389             goto fail;
 390         s->cluster_sectors = le32_to_cpu(header.granularity);
 391         s->l2_size = 1 << 9;
 392         s->l1_size = 1 << 6;
 393         bs->total_sectors = le32_to_cpu(header.disk_sectors);
 394         s->l1_table_offset = le32_to_cpu(header.l1dir_offset) << 9;
 395         s->l1_backup_table_offset = 0;
 396         s->l1_entry_sectors = s->l2_size * s->cluster_sectors;
 397     } else if (magic == VMDK4_MAGIC) {
 398         VMDK4Header header;
 399
 400         if (bdrv_pread(s->hd, sizeof(magic), &header, sizeof(header)) != sizeof(header))
 401             goto fail;
 402         bs->total_sectors = le64_to_cpu(header.capacity);
 403         s->cluster_sectors = le64_to_cpu(header.granularity);
 404         s->l2_size = le32_to_cpu(header.num_gtes_per_gte);
 405         s->l1_entry_sectors = s->l2_size * s->cluster_sectors;
 406         if (s->l1_entry_sectors <= 0)
 407             goto fail;
 408         s->l1_size = (bs->total_sectors + s->l1_entry_sectors - 1)
 409             / s->l1_entry_sectors;
 410         s->l1_table_offset = le64_to_cpu(header.rgd_offset) << 9;
 411         s->l1_backup_table_offset = le64_to_cpu(header.gd_offset) << 9;
 412
 413         if (parent_open)
 414             s->is_parent = 1;
 415         else
 416             s->is_parent = 0;
 417
 418         // try to open parent images, if exist
 419         if (vmdk_parent_open(bs, filename) != 0)
 420             goto fail;
 421         // write the CID once after the image creation
 422         s->parent_cid = vmdk_read_cid(bs,1);
 423     } else {
 424         goto fail;
 425     }
 426
 427     /* read the L1 table */
 428     l1_size = s->l1_size * sizeof(uint32_t);
 429     s->l1_table = qemu_malloc(l1_size);
 430     if (bdrv_pread(s->hd, s->l1_table_offset, s->l1_table, l1_size) != l1_size)
 431         goto fail;
 432     for(i = 0; i < s->l1_size; i++) {
 433         le32_to_cpus(&s->l1_table[i]);
 434     }
 435
 436     if (s->l1_backup_table_offset) {
 437         s->l1_backup_table = qemu_malloc(l1_size);
 438         if (bdrv_pread(s->hd, s->l1_backup_table_offset, s->l1_backup_table, l1_size) != l1_size)
 439             goto fail;
 440         for(i = 0; i < s->l1_size; i++) {
 441             le32_to_cpus(&s->l1_backup_table[i]);
 442         }
 443     }
 444
 445     s->l2_cache = qemu_malloc(s->l2_size * L2_CACHE_SIZE * sizeof(uint32_t));
 446     return 0;
 447  fail:
 448     qemu_free(s->l1_backup_table);
 449     qemu_free(s->l1_table);
 450     qemu_free(s->l2_cache);
 451     bdrv_delete(s->hd);
 452     return -1;
 453 }
 454
 455 static uint64_t get_cluster_offset(BlockDriverState *bs, VmdkMetaData *m_data,
 456                                    uint64_t offset, int allocate);
 457
 458 static int get_whole_cluster(BlockDriverState *bs, uint64_t cluster_offset,
 459                              uint64_t offset, int allocate)
 460 {
 461     uint64_t parent_cluster_offset;
 462     BDRVVmdkState *s = bs->opaque;
 463     uint8_t  whole_grain[s->cluster_sectors*512];        // 128 sectors * 512 bytes each = grain size 64KB
 464
 465     // we will be here if it's first write on non-exist grain(cluster).
 466     // try to read from parent image, if exist
 467     if (s->hd->backing_hd) {
 468         BDRVVmdkState *ps = s->hd->backing_hd->opaque;
 469
 470         if (!vmdk_is_cid_valid(bs))
 471             return -1;
 472
 473         parent_cluster_offset = get_cluster_offset(s->hd->backing_hd, NULL, offset, allocate);
 474
 475         if (parent_cluster_offset) {
 476             BDRVVmdkState *act_s = activeBDRV.hd->opaque;
 477
 478             if (bdrv_pread(ps->hd, parent_cluster_offset, whole_grain, ps->cluster_sectors*512) != ps->cluster_sectors*512)
 479                 return -1;
 480
 481             //Write grain only into the active image
 482             if (bdrv_pwrite(act_s->hd, activeBDRV.cluster_offset << 9, whole_grain, sizeof(whole_grain)) != sizeof(whole_grain))
 483                 return -1;
 484         }
 485     }
 486     return 0;
 487 }
 488
 489 static int vmdk_L2update(BlockDriverState *bs, VmdkMetaData *m_data)
 490 {
 491     BDRVVmdkState *s = bs->opaque;
 492
 493     /* update L2 table */
 494     if (bdrv_pwrite(s->hd, ((int64_t)m_data->l2_offset * 512) + (m_data->l2_index * sizeof(m_data->offset)),
 495                     &(m_data->offset), sizeof(m_data->offset)) != sizeof(m_data->offset))
 496         return -1;
 497     /* update backup L2 table */
 498     if (s->l1_backup_table_offset != 0) {
 499         m_data->l2_offset = s->l1_backup_table[m_data->l1_index];
 500         if (bdrv_pwrite(s->hd, ((int64_t)m_data->l2_offset * 512) + (m_data->l2_index * sizeof(m_data->offset)),
 501                         &(m_data->offset), sizeof(m_data->offset)) != sizeof(m_data->offset))
 502             return -1;
 503     }
 504
 505     return 0;
 506 }
 507
 508 static uint64_t get_cluster_offset(BlockDriverState *bs, VmdkMetaData *m_data,
 509                                    uint64_t offset, int allocate)
 510 {
 511     BDRVVmdkState *s = bs->opaque;
 512     unsigned int l1_index, l2_offset, l2_index;
 513     int min_index, i, j;
 514     uint32_t min_count, *l2_table, tmp = 0;
 515     uint64_t cluster_offset;
 516
 517     if (m_data)
 518         m_data->valid = 0;
 519
 520     l1_index = (offset >> 9) / s->l1_entry_sectors;
 521     if (l1_index >= s->l1_size)
 522         return 0;
 523     l2_offset = s->l1_table[l1_index];
 524     if (!l2_offset)
 525         return 0;
 526     for(i = 0; i < L2_CACHE_SIZE; i++) {
 527         if (l2_offset == s->l2_cache_offsets[i]) {
 528             /* increment the hit count */
 529             if (++s->l2_cache_counts[i] == 0xffffffff) {
 530                 for(j = 0; j < L2_CACHE_SIZE; j++) {
 531                     s->l2_cache_counts[j] >>= 1;
 532                 }
 533             }
 534             l2_table = s->l2_cache + (i * s->l2_size);
 535             goto found;
 536         }
 537     }
 538     /* not found: load a new entry in the least used one */
 539     min_index = 0;
 540     min_count = 0xffffffff;
 541     for(i = 0; i < L2_CACHE_SIZE; i++) {
 542         if (s->l2_cache_counts[i] < min_count) {
 543             min_count = s->l2_cache_counts[i];
 544             min_index = i;
 545         }
 546     }
 547     l2_table = s->l2_cache + (min_index * s->l2_size);
 548     if (bdrv_pread(s->hd, (int64_t)l2_offset * 512, l2_table, s->l2_size * sizeof(uint32_t)) !=
 549                                                                         s->l2_size * sizeof(uint32_t))
 550         return 0;
 551
 552     s->l2_cache_offsets[min_index] = l2_offset;
 553     s->l2_cache_counts[min_index] = 1;
 554  found:
 555     l2_index = ((offset >> 9) / s->cluster_sectors) % s->l2_size;
 556     cluster_offset = le32_to_cpu(l2_table[l2_index]);
 557
 558     if (!cluster_offset) {
 559         if (!allocate)
 560             return 0;
 561         // Avoid the L2 tables update for the images that have snapshots.
 562         if (!s->is_parent) {
 563             cluster_offset = bdrv_getlength(s->hd);
 564             bdrv_truncate(s->hd, cluster_offset + (s->cluster_sectors << 9));
 565
 566             cluster_offset >>= 9;
 567             tmp = cpu_to_le32(cluster_offset);
 568             l2_table[l2_index] = tmp;
 569             // Save the active image state
 570             activeBDRV.cluster_offset = cluster_offset;
 571             activeBDRV.hd = bs;
 572         }
 573         /* First of all we write grain itself, to avoid race condition
 574          * that may to corrupt the image.
 575          * This problem may occur because of insufficient space on host disk
 576          * or inappropriate VM shutdown.
 577          */
 578         if (get_whole_cluster(bs, cluster_offset, offset, allocate) == -1)
 579             return 0;
 580
 581         if (m_data) {
 582             m_data->offset = tmp;
 583             m_data->l1_index = l1_index;
 584             m_data->l2_index = l2_index;
 585             m_data->l2_offset = l2_offset;
 586             m_data->valid = 1;
 587         }
 588     }
 589     cluster_offset <<= 9;
 590     return cluster_offset;
 591 }
 592
 593 static int vmdk_is_allocated(BlockDriverState *bs, int64_t sector_num,
 594                              int nb_sectors, int *pnum)
 595 {
 596     BDRVVmdkState *s = bs->opaque;
 597     int index_in_cluster, n;
 598     uint64_t cluster_offset;
 599
 600     cluster_offset = get_cluster_offset(bs, NULL, sector_num << 9, 0);
 601     index_in_cluster = sector_num % s->cluster_sectors;
 602     n = s->cluster_sectors - index_in_cluster;
 603     if (n > nb_sectors)
 604         n = nb_sectors;
 605     *pnum = n;
 606     return (cluster_offset != 0);
 607 }
 608
 609 static int vmdk_read(BlockDriverState *bs, int64_t sector_num,
 610                     uint8_t *buf, int nb_sectors)
 611 {
 612     BDRVVmdkState *s = bs->opaque;
 613     int index_in_cluster, n, ret;
 614     uint64_t cluster_offset;
 615
 616     while (nb_sectors > 0) {
 617         cluster_offset = get_cluster_offset(bs, NULL, sector_num << 9, 0);
 618         index_in_cluster = sector_num % s->cluster_sectors;
 619         n = s->cluster_sectors - index_in_cluster;
 620         if (n > nb_sectors)
 621             n = nb_sectors;
 622         if (!cluster_offset) {
 623             // try to read from parent image, if exist
 624             if (s->hd->backing_hd) {
 625                 if (!vmdk_is_cid_valid(bs))
 626                     return -1;
 627                 ret = bdrv_read(s->hd->backing_hd, sector_num, buf, n);
 628                 if (ret < 0)
 629                     return -1;
 630             } else {
 631                 memset(buf, 0, 512 * n);
 632             }
 633         } else {
 634             if(bdrv_pread(s->hd, cluster_offset + index_in_cluster * 512, buf, n * 512) != n * 512)
 635                 return -1;
 636         }
 637         nb_sectors -= n;
 638         sector_num += n;
 639         buf += n * 512;
 640     }
 641     return 0;
 642 }
 643
 644 static int vmdk_write(BlockDriverState *bs, int64_t sector_num,
 645                      const uint8_t *buf, int nb_sectors)
 646 {
 647     BDRVVmdkState *s = bs->opaque;
 648     VmdkMetaData m_data;
 649     int index_in_cluster, n;
 650     uint64_t cluster_offset;
 651     static int cid_update = 0;
 652
 653     if (sector_num > bs->total_sectors) {
 654         fprintf(stderr,
 655                 "(VMDK) Wrong offset: sector_num=0x%" PRIx64
 656                 " total_sectors=0x%" PRIx64 "\n",
 657                 sector_num, bs->total_sectors);
 658         return -1;
 659     }
 660
 661     while (nb_sectors > 0) {
 662         index_in_cluster = sector_num & (s->cluster_sectors - 1);
 663         n = s->cluster_sectors - index_in_cluster;
 664         if (n > nb_sectors)
 665             n = nb_sectors;
 666         cluster_offset = get_cluster_offset(bs, &m_data, sector_num << 9, 1);
 667         if (!cluster_offset)
 668             return -1;
 669
 670         if (bdrv_pwrite(s->hd, cluster_offset + index_in_cluster * 512, buf, n * 512) != n * 512)
 671             return -1;
 672         if (m_data.valid) {
 673             /* update L2 tables */
 674             if (vmdk_L2update(bs, &m_data) == -1)
 675                 return -1;
 676         }
 677         nb_sectors -= n;
 678         sector_num += n;
 679         buf += n * 512;
 680
 681         // update CID on the first write every time the virtual disk is opened
 682         if (!cid_update) {
 683             vmdk_write_cid(bs, time(NULL));
 684             cid_update++;
 685         }
 686     }
 687     return 0;
 688 }
 689
 690 static int vmdk_create(const char *filename, int64_t total_size,
 691                        const char *backing_file, int flags)
 692 {
 693     int fd, i;
 694     VMDK4Header header;
 695     uint32_t tmp, magic, grains, gd_size, gt_size, gt_count;
 696     static const char desc_template[] =
 697         "# Disk DescriptorFile\n"
 698         "version=1\n"
 699         "CID=%x\n"
 700         "parentCID=ffffffff\n"
 701         "createType=\"monolithicSparse\"\n"
 702         "\n"
 703         "# Extent description\n"
 704         "RW %" PRId64 " SPARSE \"%s\"\n"
 705         "\n"
 706         "# The Disk Data Base \n"
 707         "#DDB\n"
 708         "\n"
 709         "ddb.virtualHWVersion = \"%d\"\n"
 710         "ddb.geometry.cylinders = \"%" PRId64 "\"\n"
 711         "ddb.geometry.heads = \"16\"\n"
 712         "ddb.geometry.sectors = \"63\"\n"
 713         "ddb.adapterType = \"ide\"\n";
 714     char desc[1024];
 715     const char *real_filename, *temp_str;
 716
 717     /* XXX: add support for backing file */
 718     if (backing_file) {
 719         return vmdk_snapshot_create(filename, backing_file);
 720     }
 721
 722     fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY | O_LARGEFILE,
 723               0644);
 724     if (fd < 0)
 725         return -1;
 726     magic = cpu_to_be32(VMDK4_MAGIC);
 727     memset(&header, 0, sizeof(header));
 728     header.version = cpu_to_le32(1);
 729     header.flags = cpu_to_le32(3); /* ?? */
 730     header.capacity = cpu_to_le64(total_size);
 731     header.granularity = cpu_to_le64(128);
 732     header.num_gtes_per_gte = cpu_to_le32(512);
 733
 734     grains = (total_size + header.granularity - 1) / header.granularity;
 735     gt_size = ((header.num_gtes_per_gte * sizeof(uint32_t)) + 511) >> 9;
 736     gt_count = (grains + header.num_gtes_per_gte - 1) / header.num_gtes_per_gte;
 737     gd_size = (gt_count * sizeof(uint32_t) + 511) >> 9;
 738
 739     header.desc_offset = 1;
 740     header.desc_size = 20;
 741     header.rgd_offset = header.desc_offset + header.desc_size;
 742     header.gd_offset = header.rgd_offset + gd_size + (gt_size * gt_count);
 743     header.grain_offset =
 744        ((header.gd_offset + gd_size + (gt_size * gt_count) +
 745          header.granularity - 1) / header.granularity) *
 746         header.granularity;
 747
 748     header.desc_offset = cpu_to_le64(header.desc_offset);
 749     header.desc_size = cpu_to_le64(header.desc_size);
 750     header.rgd_offset = cpu_to_le64(header.rgd_offset);
 751     header.gd_offset = cpu_to_le64(header.gd_offset);
 752     header.grain_offset = cpu_to_le64(header.grain_offset);
 753
 754     header.check_bytes[0] = 0xa;
 755     header.check_bytes[1] = 0x20;
 756     header.check_bytes[2] = 0xd;
 757     header.check_bytes[3] = 0xa;
 758
 759     /* write all the data */
 760     write(fd, &magic, sizeof(magic));
 761     write(fd, &header, sizeof(header));
 762
 763     ftruncate(fd, header.grain_offset << 9);
 764
 765     /* write grain directory */
 766     lseek(fd, le64_to_cpu(header.rgd_offset) << 9, SEEK_SET);
 767     for (i = 0, tmp = header.rgd_offset + gd_size;
 768          i < gt_count; i++, tmp += gt_size)
 769         write(fd, &tmp, sizeof(tmp));
 770
 771     /* write backup grain directory */
 772     lseek(fd, le64_to_cpu(header.gd_offset) << 9, SEEK_SET);
 773     for (i = 0, tmp = header.gd_offset + gd_size;
 774          i < gt_count; i++, tmp += gt_size)
 775         write(fd, &tmp, sizeof(tmp));
 776
 777     /* compose the descriptor */
 778     real_filename = filename;
 779     if ((temp_str = strrchr(real_filename, '\\')) != NULL)
 780         real_filename = temp_str + 1;
 781     if ((temp_str = strrchr(real_filename, '/')) != NULL)
 782         real_filename = temp_str + 1;
 783     if ((temp_str = strrchr(real_filename, ':')) != NULL)
 784         real_filename = temp_str + 1;
 785     snprintf(desc, sizeof(desc), desc_template, (unsigned int)time(NULL),
 786              total_size, real_filename,
 787              (flags & BLOCK_FLAG_COMPAT6 ? 6 : 4),
 788              total_size / (int64_t)(63 * 16));
 789
 790     /* write the descriptor */
 791     lseek(fd, le64_to_cpu(header.desc_offset) << 9, SEEK_SET);
 792     write(fd, desc, strlen(desc));
 793
 794     close(fd);
 795     return 0;
 796 }
 797
 798 static void vmdk_close(BlockDriverState *bs)
 799 {
 800     BDRVVmdkState *s = bs->opaque;
 801
 802     qemu_free(s->l1_table);
 803     qemu_free(s->l2_cache);
 804     // try to close parent image, if exist
 805     vmdk_parent_close(s->hd);
 806     bdrv_delete(s->hd);
 807 }
 808
 809 static void vmdk_flush(BlockDriverState *bs)
 810 {
 811     BDRVVmdkState *s = bs->opaque;
 812     bdrv_flush(s->hd);
 813 }
 814
 815 static BlockDriver bdrv_vmdk = {
 816     .format_name        = "vmdk",
 817     .instance_size      = sizeof(BDRVVmdkState),
 818     .bdrv_probe         = vmdk_probe,
 819     .bdrv_open          = vmdk_open,
 820     .bdrv_read          = vmdk_read,
 821     .bdrv_write         = vmdk_write,
 822     .bdrv_close         = vmdk_close,
 823     .bdrv_create        = vmdk_create,
 824     .bdrv_flush         = vmdk_flush,
 825     .bdrv_is_allocated  = vmdk_is_allocated,
 826 };
 827
 828 static void bdrv_vmdk_init(void)
 829 {
 830     bdrv_register(&bdrv_vmdk);
 831 }
 832
 833 block_init(bdrv_vmdk_init);