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     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 } BDRVVmdkState;
  79
  80 typedef struct VmdkMetaData {
  81     uint32_t offset;
  82     unsigned int l1_index;
  83     unsigned int l2_index;
  84     unsigned int l2_offset;
  85     int valid;
  86 } VmdkMetaData;
  87
  88 static int vmdk_probe(const uint8_t *buf, int buf_size, const char *filename)
  89 {
  90     uint32_t magic;
  91
  92     if (buf_size < 4)
  93         return 0;
  94     magic = be32_to_cpu(*(uint32_t *)buf);
  95     if (magic == VMDK3_MAGIC ||
  96         magic == VMDK4_MAGIC)
  97         return 100;
  98     else
  99         return 0;
 100 }
 101
 102 #define CHECK_CID 1
 103
 104 #define SECTOR_SIZE 512
 105 #define DESC_SIZE 20*SECTOR_SIZE        // 20 sectors of 512 bytes each
 106 #define HEADER_SIZE 512                         // first sector of 512 bytes
 107
 108 static uint32_t vmdk_read_cid(BlockDriverState *bs, int parent)
 109 {
 110     char desc[DESC_SIZE];
 111     uint32_t cid;
 112     const char *p_name, *cid_str;
 113     size_t cid_str_size;
 114
 115     /* the descriptor offset = 0x200 */
 116     if (bdrv_pread(bs->file, 0x200, desc, DESC_SIZE) != DESC_SIZE)
 117         return 0;
 118
 119     if (parent) {
 120         cid_str = "parentCID";
 121         cid_str_size = sizeof("parentCID");
 122     } else {
 123         cid_str = "CID";
 124         cid_str_size = sizeof("CID");
 125     }
 126
 127     if ((p_name = strstr(desc,cid_str)) != NULL) {
 128         p_name += cid_str_size;
 129         sscanf(p_name,"%x",&cid);
 130     }
 131
 132     return cid;
 133 }
 134
 135 static int vmdk_write_cid(BlockDriverState *bs, uint32_t cid)
 136 {
 137     char desc[DESC_SIZE], tmp_desc[DESC_SIZE];
 138     char *p_name, *tmp_str;
 139
 140     /* the descriptor offset = 0x200 */
 141     if (bdrv_pread(bs->file, 0x200, desc, DESC_SIZE) != DESC_SIZE)
 142         return -1;
 143
 144     tmp_str = strstr(desc,"parentCID");
 145     pstrcpy(tmp_desc, sizeof(tmp_desc), tmp_str);
 146     if ((p_name = strstr(desc,"CID")) != NULL) {
 147         p_name += sizeof("CID");
 148         snprintf(p_name, sizeof(desc) - (p_name - desc), "%x\n", cid);
 149         pstrcat(desc, sizeof(desc), tmp_desc);
 150     }
 151
 152     if (bdrv_pwrite_sync(bs->file, 0x200, desc, DESC_SIZE) < 0)
 153         return -1;
 154     return 0;
 155 }
 156
 157 static int vmdk_is_cid_valid(BlockDriverState *bs)
 158 {
 159 #ifdef CHECK_CID
 160     BDRVVmdkState *s = bs->opaque;
 161     BlockDriverState *p_bs = bs->backing_hd;
 162     uint32_t cur_pcid;
 163
 164     if (p_bs) {
 165         cur_pcid = vmdk_read_cid(p_bs,0);
 166         if (s->parent_cid != cur_pcid)
 167             // CID not valid
 168             return 0;
 169     }
 170 #endif
 171     // CID valid
 172     return 1;
 173 }
 174
 175 static int vmdk_snapshot_create(const char *filename, const char *backing_file)
 176 {
 177     int snp_fd, p_fd;
 178     int ret;
 179     uint32_t p_cid;
 180     char *p_name, *gd_buf, *rgd_buf;
 181     const char *real_filename, *temp_str;
 182     VMDK4Header header;
 183     uint32_t gde_entries, gd_size;
 184     int64_t gd_offset, rgd_offset, capacity, gt_size;
 185     char p_desc[DESC_SIZE], s_desc[DESC_SIZE], hdr[HEADER_SIZE];
 186     static const char desc_template[] =
 187     "# Disk DescriptorFile\n"
 188     "version=1\n"
 189     "CID=%x\n"
 190     "parentCID=%x\n"
 191     "createType=\"monolithicSparse\"\n"
 192     "parentFileNameHint=\"%s\"\n"
 193     "\n"
 194     "# Extent description\n"
 195     "RW %u SPARSE \"%s\"\n"
 196     "\n"
 197     "# The Disk Data Base \n"
 198     "#DDB\n"
 199     "\n";
 200
 201     snp_fd = open(filename, O_RDWR | O_CREAT | O_TRUNC | O_BINARY | O_LARGEFILE, 0644);
 202     if (snp_fd < 0)
 203         return -errno;
 204     p_fd = open(backing_file, O_RDONLY | O_BINARY | O_LARGEFILE);
 205     if (p_fd < 0) {
 206         close(snp_fd);
 207         return -errno;
 208     }
 209
 210     /* read the header */
 211     if (lseek(p_fd, 0x0, SEEK_SET) == -1) {
 212         ret = -errno;
 213         goto fail;
 214     }
 215     if (read(p_fd, hdr, HEADER_SIZE) != HEADER_SIZE) {
 216         ret = -errno;
 217         goto fail;
 218     }
 219
 220     /* write the header */
 221     if (lseek(snp_fd, 0x0, SEEK_SET) == -1) {
 222         ret = -errno;
 223         goto fail;
 224     }
 225     if (write(snp_fd, hdr, HEADER_SIZE) == -1) {
 226         ret = -errno;
 227         goto fail;
 228     }
 229
 230     memset(&header, 0, sizeof(header));
 231     memcpy(&header,&hdr[4], sizeof(header)); // skip the VMDK4_MAGIC
 232
 233     if (ftruncate(snp_fd, header.grain_offset << 9)) {
 234         ret = -errno;
 235         goto fail;
 236     }
 237     /* the descriptor offset = 0x200 */
 238     if (lseek(p_fd, 0x200, SEEK_SET) == -1) {
 239         ret = -errno;
 240         goto fail;
 241     }
 242     if (read(p_fd, p_desc, DESC_SIZE) != DESC_SIZE) {
 243         ret = -errno;
 244         goto fail;
 245     }
 246
 247     if ((p_name = strstr(p_desc,"CID")) != NULL) {
 248         p_name += sizeof("CID");
 249         sscanf(p_name,"%x",&p_cid);
 250     }
 251
 252     real_filename = filename;
 253     if ((temp_str = strrchr(real_filename, '\\')) != NULL)
 254         real_filename = temp_str + 1;
 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
 260     snprintf(s_desc, sizeof(s_desc), desc_template, p_cid, p_cid, backing_file,
 261              (uint32_t)header.capacity, real_filename);
 262
 263     /* write the descriptor */
 264     if (lseek(snp_fd, 0x200, SEEK_SET) == -1) {
 265         ret = -errno;
 266         goto fail;
 267     }
 268     if (write(snp_fd, s_desc, strlen(s_desc)) == -1) {
 269         ret = -errno;
 270         goto fail;
 271     }
 272
 273     gd_offset = header.gd_offset * SECTOR_SIZE;     // offset of GD table
 274     rgd_offset = header.rgd_offset * SECTOR_SIZE;   // offset of RGD table
 275     capacity = header.capacity * SECTOR_SIZE;       // Extent size
 276     /*
 277      * Each GDE span 32M disk, means:
 278      * 512 GTE per GT, each GTE points to grain
 279      */
 280     gt_size = (int64_t)header.num_gtes_per_gte * header.granularity * SECTOR_SIZE;
 281     if (!gt_size) {
 282         ret = -EINVAL;
 283         goto fail;
 284     }
 285     gde_entries = (uint32_t)(capacity / gt_size);  // number of gde/rgde
 286     gd_size = gde_entries * sizeof(uint32_t);
 287
 288     /* write RGD */
 289     rgd_buf = qemu_malloc(gd_size);
 290     if (lseek(p_fd, rgd_offset, SEEK_SET) == -1) {
 291         ret = -errno;
 292         goto fail_rgd;
 293     }
 294     if (read(p_fd, rgd_buf, gd_size) != gd_size) {
 295         ret = -errno;
 296         goto fail_rgd;
 297     }
 298     if (lseek(snp_fd, rgd_offset, SEEK_SET) == -1) {
 299         ret = -errno;
 300         goto fail_rgd;
 301     }
 302     if (write(snp_fd, rgd_buf, gd_size) == -1) {
 303         ret = -errno;
 304         goto fail_rgd;
 305     }
 306
 307     /* write GD */
 308     gd_buf = qemu_malloc(gd_size);
 309     if (lseek(p_fd, gd_offset, SEEK_SET) == -1) {
 310         ret = -errno;
 311         goto fail_gd;
 312     }
 313     if (read(p_fd, gd_buf, gd_size) != gd_size) {
 314         ret = -errno;
 315         goto fail_gd;
 316     }
 317     if (lseek(snp_fd, gd_offset, SEEK_SET) == -1) {
 318         ret = -errno;
 319         goto fail_gd;
 320     }
 321     if (write(snp_fd, gd_buf, gd_size) == -1) {
 322         ret = -errno;
 323         goto fail_gd;
 324     }
 325     ret = 0;
 326
 327 fail_gd:
 328     qemu_free(gd_buf);
 329 fail_rgd:
 330     qemu_free(rgd_buf);
 331 fail:
 332     close(p_fd);
 333     close(snp_fd);
 334     return ret;
 335 }
 336
 337 static int vmdk_parent_open(BlockDriverState *bs)
 338 {
 339     char *p_name;
 340     char desc[DESC_SIZE];
 341
 342     /* the descriptor offset = 0x200 */
 343     if (bdrv_pread(bs->file, 0x200, desc, DESC_SIZE) != DESC_SIZE)
 344         return -1;
 345
 346     if ((p_name = strstr(desc,"parentFileNameHint")) != NULL) {
 347         char *end_name;
 348
 349         p_name += sizeof("parentFileNameHint") + 1;
 350         if ((end_name = strchr(p_name,'\"')) == NULL)
 351             return -1;
 352         if ((end_name - p_name) > sizeof (bs->backing_file) - 1)
 353             return -1;
 354
 355         pstrcpy(bs->backing_file, end_name - p_name + 1, p_name);
 356     }
 357
 358     return 0;
 359 }
 360
 361 static int vmdk_open(BlockDriverState *bs, int flags)
 362 {
 363     BDRVVmdkState *s = bs->opaque;
 364     uint32_t magic;
 365     int l1_size, i;
 366
 367     if (bdrv_pread(bs->file, 0, &magic, sizeof(magic)) != sizeof(magic))
 368         goto fail;
 369
 370     magic = be32_to_cpu(magic);
 371     if (magic == VMDK3_MAGIC) {
 372         VMDK3Header header;
 373
 374         if (bdrv_pread(bs->file, sizeof(magic), &header, sizeof(header)) != sizeof(header))
 375             goto fail;
 376         s->cluster_sectors = le32_to_cpu(header.granularity);
 377         s->l2_size = 1 << 9;
 378         s->l1_size = 1 << 6;
 379         bs->total_sectors = le32_to_cpu(header.disk_sectors);
 380         s->l1_table_offset = le32_to_cpu(header.l1dir_offset) << 9;
 381         s->l1_backup_table_offset = 0;
 382         s->l1_entry_sectors = s->l2_size * s->cluster_sectors;
 383     } else if (magic == VMDK4_MAGIC) {
 384         VMDK4Header header;
 385
 386         if (bdrv_pread(bs->file, sizeof(magic), &header, sizeof(header)) != sizeof(header))
 387             goto fail;
 388         bs->total_sectors = le64_to_cpu(header.capacity);
 389         s->cluster_sectors = le64_to_cpu(header.granularity);
 390         s->l2_size = le32_to_cpu(header.num_gtes_per_gte);
 391         s->l1_entry_sectors = s->l2_size * s->cluster_sectors;
 392         if (s->l1_entry_sectors <= 0)
 393             goto fail;
 394         s->l1_size = (bs->total_sectors + s->l1_entry_sectors - 1)
 395             / s->l1_entry_sectors;
 396         s->l1_table_offset = le64_to_cpu(header.rgd_offset) << 9;
 397         s->l1_backup_table_offset = le64_to_cpu(header.gd_offset) << 9;
 398
 399         // try to open parent images, if exist
 400         if (vmdk_parent_open(bs) != 0)
 401             goto fail;
 402         // write the CID once after the image creation
 403         s->parent_cid = vmdk_read_cid(bs,1);
 404     } else {
 405         goto fail;
 406     }
 407
 408     /* read the L1 table */
 409     l1_size = s->l1_size * sizeof(uint32_t);
 410     s->l1_table = qemu_malloc(l1_size);
 411     if (bdrv_pread(bs->file, s->l1_table_offset, s->l1_table, l1_size) != l1_size)
 412         goto fail;
 413     for(i = 0; i < s->l1_size; i++) {
 414         le32_to_cpus(&s->l1_table[i]);
 415     }
 416
 417     if (s->l1_backup_table_offset) {
 418         s->l1_backup_table = qemu_malloc(l1_size);
 419         if (bdrv_pread(bs->file, s->l1_backup_table_offset, s->l1_backup_table, l1_size) != l1_size)
 420             goto fail;
 421         for(i = 0; i < s->l1_size; i++) {
 422             le32_to_cpus(&s->l1_backup_table[i]);
 423         }
 424     }
 425
 426     s->l2_cache = qemu_malloc(s->l2_size * L2_CACHE_SIZE * sizeof(uint32_t));
 427     return 0;
 428  fail:
 429     qemu_free(s->l1_backup_table);
 430     qemu_free(s->l1_table);
 431     qemu_free(s->l2_cache);
 432     return -1;
 433 }
 434
 435 static uint64_t get_cluster_offset(BlockDriverState *bs, VmdkMetaData *m_data,
 436                                    uint64_t offset, int allocate);
 437
 438 static int get_whole_cluster(BlockDriverState *bs, uint64_t cluster_offset,
 439                              uint64_t offset, int allocate)
 440 {
 441     BDRVVmdkState *s = bs->opaque;
 442     uint8_t  whole_grain[s->cluster_sectors*512];        // 128 sectors * 512 bytes each = grain size 64KB
 443
 444     // we will be here if it's first write on non-exist grain(cluster).
 445     // try to read from parent image, if exist
 446     if (bs->backing_hd) {
 447         int ret;
 448
 449         if (!vmdk_is_cid_valid(bs))
 450             return -1;
 451
 452         ret = bdrv_read(bs->backing_hd, offset >> 9, whole_grain,
 453             s->cluster_sectors);
 454         if (ret < 0) {
 455             return -1;
 456         }
 457
 458         //Write grain only into the active image
 459         ret = bdrv_write(bs->file, cluster_offset, whole_grain,
 460             s->cluster_sectors);
 461         if (ret < 0) {
 462             return -1;
 463         }
 464     }
 465     return 0;
 466 }
 467
 468 static int vmdk_L2update(BlockDriverState *bs, VmdkMetaData *m_data)
 469 {
 470     BDRVVmdkState *s = bs->opaque;
 471
 472     /* update L2 table */
 473     if (bdrv_pwrite_sync(bs->file, ((int64_t)m_data->l2_offset * 512) + (m_data->l2_index * sizeof(m_data->offset)),
 474                     &(m_data->offset), sizeof(m_data->offset)) < 0)
 475         return -1;
 476     /* update backup L2 table */
 477     if (s->l1_backup_table_offset != 0) {
 478         m_data->l2_offset = s->l1_backup_table[m_data->l1_index];
 479         if (bdrv_pwrite_sync(bs->file, ((int64_t)m_data->l2_offset * 512) + (m_data->l2_index * sizeof(m_data->offset)),
 480                         &(m_data->offset), sizeof(m_data->offset)) < 0)
 481             return -1;
 482     }
 483
 484     return 0;
 485 }
 486
 487 static uint64_t get_cluster_offset(BlockDriverState *bs, VmdkMetaData *m_data,
 488                                    uint64_t offset, int allocate)
 489 {
 490     BDRVVmdkState *s = bs->opaque;
 491     unsigned int l1_index, l2_offset, l2_index;
 492     int min_index, i, j;
 493     uint32_t min_count, *l2_table, tmp = 0;
 494     uint64_t cluster_offset;
 495
 496     if (m_data)
 497         m_data->valid = 0;
 498
 499     l1_index = (offset >> 9) / s->l1_entry_sectors;
 500     if (l1_index >= s->l1_size)
 501         return 0;
 502     l2_offset = s->l1_table[l1_index];
 503     if (!l2_offset)
 504         return 0;
 505     for(i = 0; i < L2_CACHE_SIZE; i++) {
 506         if (l2_offset == s->l2_cache_offsets[i]) {
 507             /* increment the hit count */
 508             if (++s->l2_cache_counts[i] == 0xffffffff) {
 509                 for(j = 0; j < L2_CACHE_SIZE; j++) {
 510                     s->l2_cache_counts[j] >>= 1;
 511                 }
 512             }
 513             l2_table = s->l2_cache + (i * s->l2_size);
 514             goto found;
 515         }
 516     }
 517     /* not found: load a new entry in the least used one */
 518     min_index = 0;
 519     min_count = 0xffffffff;
 520     for(i = 0; i < L2_CACHE_SIZE; i++) {
 521         if (s->l2_cache_counts[i] < min_count) {
 522             min_count = s->l2_cache_counts[i];
 523             min_index = i;
 524         }
 525     }
 526     l2_table = s->l2_cache + (min_index * s->l2_size);
 527     if (bdrv_pread(bs->file, (int64_t)l2_offset * 512, l2_table, s->l2_size * sizeof(uint32_t)) !=
 528                                                                         s->l2_size * sizeof(uint32_t))
 529         return 0;
 530
 531     s->l2_cache_offsets[min_index] = l2_offset;
 532     s->l2_cache_counts[min_index] = 1;
 533  found:
 534     l2_index = ((offset >> 9) / s->cluster_sectors) % s->l2_size;
 535     cluster_offset = le32_to_cpu(l2_table[l2_index]);
 536
 537     if (!cluster_offset) {
 538         if (!allocate)
 539             return 0;
 540
 541         // Avoid the L2 tables update for the images that have snapshots.
 542         cluster_offset = bdrv_getlength(bs->file);
 543         bdrv_truncate(bs->file, cluster_offset + (s->cluster_sectors << 9));
 544
 545         cluster_offset >>= 9;
 546         tmp = cpu_to_le32(cluster_offset);
 547         l2_table[l2_index] = tmp;
 548
 549         /* First of all we write grain itself, to avoid race condition
 550          * that may to corrupt the image.
 551          * This problem may occur because of insufficient space on host disk
 552          * or inappropriate VM shutdown.
 553          */
 554         if (get_whole_cluster(bs, cluster_offset, offset, allocate) == -1)
 555             return 0;
 556
 557         if (m_data) {
 558             m_data->offset = tmp;
 559             m_data->l1_index = l1_index;
 560             m_data->l2_index = l2_index;
 561             m_data->l2_offset = l2_offset;
 562             m_data->valid = 1;
 563         }
 564     }
 565     cluster_offset <<= 9;
 566     return cluster_offset;
 567 }
 568
 569 static int vmdk_is_allocated(BlockDriverState *bs, int64_t sector_num,
 570                              int nb_sectors, int *pnum)
 571 {
 572     BDRVVmdkState *s = bs->opaque;
 573     int index_in_cluster, n;
 574     uint64_t cluster_offset;
 575
 576     cluster_offset = get_cluster_offset(bs, NULL, sector_num << 9, 0);
 577     index_in_cluster = sector_num % s->cluster_sectors;
 578     n = s->cluster_sectors - index_in_cluster;
 579     if (n > nb_sectors)
 580         n = nb_sectors;
 581     *pnum = n;
 582     return (cluster_offset != 0);
 583 }
 584
 585 static int vmdk_read(BlockDriverState *bs, int64_t sector_num,
 586                     uint8_t *buf, int nb_sectors)
 587 {
 588     BDRVVmdkState *s = bs->opaque;
 589     int index_in_cluster, n, ret;
 590     uint64_t cluster_offset;
 591
 592     while (nb_sectors > 0) {
 593         cluster_offset = get_cluster_offset(bs, NULL, sector_num << 9, 0);
 594         index_in_cluster = sector_num % s->cluster_sectors;
 595         n = s->cluster_sectors - index_in_cluster;
 596         if (n > nb_sectors)
 597             n = nb_sectors;
 598         if (!cluster_offset) {
 599             // try to read from parent image, if exist
 600             if (bs->backing_hd) {
 601                 if (!vmdk_is_cid_valid(bs))
 602                     return -1;
 603                 ret = bdrv_read(bs->backing_hd, sector_num, buf, n);
 604                 if (ret < 0)
 605                     return -1;
 606             } else {
 607                 memset(buf, 0, 512 * n);
 608             }
 609         } else {
 610             if(bdrv_pread(bs->file, cluster_offset + index_in_cluster * 512, buf, n * 512) != n * 512)
 611                 return -1;
 612         }
 613         nb_sectors -= n;
 614         sector_num += n;
 615         buf += n * 512;
 616     }
 617     return 0;
 618 }
 619
 620 static int vmdk_write(BlockDriverState *bs, int64_t sector_num,
 621                      const uint8_t *buf, int nb_sectors)
 622 {
 623     BDRVVmdkState *s = bs->opaque;
 624     VmdkMetaData m_data;
 625     int index_in_cluster, n;
 626     uint64_t cluster_offset;
 627     static int cid_update = 0;
 628
 629     if (sector_num > bs->total_sectors) {
 630         fprintf(stderr,
 631                 "(VMDK) Wrong offset: sector_num=0x%" PRIx64
 632                 " total_sectors=0x%" PRIx64 "\n",
 633                 sector_num, bs->total_sectors);
 634         return -1;
 635     }
 636
 637     while (nb_sectors > 0) {
 638         index_in_cluster = sector_num & (s->cluster_sectors - 1);
 639         n = s->cluster_sectors - index_in_cluster;
 640         if (n > nb_sectors)
 641             n = nb_sectors;
 642         cluster_offset = get_cluster_offset(bs, &m_data, sector_num << 9, 1);
 643         if (!cluster_offset)
 644             return -1;
 645
 646         if (bdrv_pwrite(bs->file, cluster_offset + index_in_cluster * 512, buf, n * 512) != n * 512)
 647             return -1;
 648         if (m_data.valid) {
 649             /* update L2 tables */
 650             if (vmdk_L2update(bs, &m_data) == -1)
 651                 return -1;
 652         }
 653         nb_sectors -= n;
 654         sector_num += n;
 655         buf += n * 512;
 656
 657         // update CID on the first write every time the virtual disk is opened
 658         if (!cid_update) {
 659             vmdk_write_cid(bs, time(NULL));
 660             cid_update++;
 661         }
 662     }
 663     return 0;
 664 }
 665
 666 static int vmdk_create(const char *filename, QEMUOptionParameter *options)
 667 {
 668     int fd, i;
 669     VMDK4Header header;
 670     uint32_t tmp, magic, grains, gd_size, gt_size, gt_count;
 671     static const char desc_template[] =
 672         "# Disk DescriptorFile\n"
 673         "version=1\n"
 674         "CID=%x\n"
 675         "parentCID=ffffffff\n"
 676         "createType=\"monolithicSparse\"\n"
 677         "\n"
 678         "# Extent description\n"
 679         "RW %" PRId64 " SPARSE \"%s\"\n"
 680         "\n"
 681         "# The Disk Data Base \n"
 682         "#DDB\n"
 683         "\n"
 684         "ddb.virtualHWVersion = \"%d\"\n"
 685         "ddb.geometry.cylinders = \"%" PRId64 "\"\n"
 686         "ddb.geometry.heads = \"16\"\n"
 687         "ddb.geometry.sectors = \"63\"\n"
 688         "ddb.adapterType = \"ide\"\n";
 689     char desc[1024];
 690     const char *real_filename, *temp_str;
 691     int64_t total_size = 0;
 692     const char *backing_file = NULL;
 693     int flags = 0;
 694     int ret;
 695
 696     // Read out options
 697     while (options && options->name) {
 698         if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
 699             total_size = options->value.n / 512;
 700         } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) {
 701             backing_file = options->value.s;
 702         } else if (!strcmp(options->name, BLOCK_OPT_COMPAT6)) {
 703             flags |= options->value.n ? BLOCK_FLAG_COMPAT6: 0;
 704         }
 705         options++;
 706     }
 707
 708     /* XXX: add support for backing file */
 709     if (backing_file) {
 710         return vmdk_snapshot_create(filename, backing_file);
 711     }
 712
 713     fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY | O_LARGEFILE,
 714               0644);
 715     if (fd < 0)
 716         return -errno;
 717     magic = cpu_to_be32(VMDK4_MAGIC);
 718     memset(&header, 0, sizeof(header));
 719     header.version = 1;
 720     header.flags = 3; /* ?? */
 721     header.capacity = total_size;
 722     header.granularity = 128;
 723     header.num_gtes_per_gte = 512;
 724
 725     grains = (total_size + header.granularity - 1) / header.granularity;
 726     gt_size = ((header.num_gtes_per_gte * sizeof(uint32_t)) + 511) >> 9;
 727     gt_count = (grains + header.num_gtes_per_gte - 1) / header.num_gtes_per_gte;
 728     gd_size = (gt_count * sizeof(uint32_t) + 511) >> 9;
 729
 730     header.desc_offset = 1;
 731     header.desc_size = 20;
 732     header.rgd_offset = header.desc_offset + header.desc_size;
 733     header.gd_offset = header.rgd_offset + gd_size + (gt_size * gt_count);
 734     header.grain_offset =
 735        ((header.gd_offset + gd_size + (gt_size * gt_count) +
 736          header.granularity - 1) / header.granularity) *
 737         header.granularity;
 738
 739     /* swap endianness for all header fields */
 740     header.version = cpu_to_le32(header.version);
 741     header.flags = cpu_to_le32(header.flags);
 742     header.capacity = cpu_to_le64(header.capacity);
 743     header.granularity = cpu_to_le64(header.granularity);
 744     header.num_gtes_per_gte = cpu_to_le32(header.num_gtes_per_gte);
 745     header.desc_offset = cpu_to_le64(header.desc_offset);
 746     header.desc_size = cpu_to_le64(header.desc_size);
 747     header.rgd_offset = cpu_to_le64(header.rgd_offset);
 748     header.gd_offset = cpu_to_le64(header.gd_offset);
 749     header.grain_offset = cpu_to_le64(header.grain_offset);
 750
 751     header.check_bytes[0] = 0xa;
 752     header.check_bytes[1] = 0x20;
 753     header.check_bytes[2] = 0xd;
 754     header.check_bytes[3] = 0xa;
 755
 756     /* write all the data */
 757     ret = qemu_write_full(fd, &magic, sizeof(magic));
 758     if (ret != sizeof(magic)) {
 759         ret = -errno;
 760         goto exit;
 761     }
 762     ret = qemu_write_full(fd, &header, sizeof(header));
 763     if (ret != sizeof(header)) {
 764         ret = -errno;
 765         goto exit;
 766     }
 767
 768     ret = ftruncate(fd, le64_to_cpu(header.grain_offset) << 9);
 769     if (ret < 0) {
 770         ret = -errno;
 771         goto exit;
 772     }
 773
 774     /* write grain directory */
 775     lseek(fd, le64_to_cpu(header.rgd_offset) << 9, SEEK_SET);
 776     for (i = 0, tmp = le64_to_cpu(header.rgd_offset) + gd_size;
 777          i < gt_count; i++, tmp += gt_size) {
 778         ret = qemu_write_full(fd, &tmp, sizeof(tmp));
 779         if (ret != sizeof(tmp)) {
 780             ret = -errno;
 781             goto exit;
 782         }
 783     }
 784
 785     /* write backup grain directory */
 786     lseek(fd, le64_to_cpu(header.gd_offset) << 9, SEEK_SET);
 787     for (i = 0, tmp = le64_to_cpu(header.gd_offset) + gd_size;
 788          i < gt_count; i++, tmp += gt_size) {
 789         ret = qemu_write_full(fd, &tmp, sizeof(tmp));
 790         if (ret != sizeof(tmp)) {
 791             ret = -errno;
 792             goto exit;
 793         }
 794     }
 795
 796     /* compose the descriptor */
 797     real_filename = filename;
 798     if ((temp_str = strrchr(real_filename, '\\')) != NULL)
 799         real_filename = temp_str + 1;
 800     if ((temp_str = strrchr(real_filename, '/')) != NULL)
 801         real_filename = temp_str + 1;
 802     if ((temp_str = strrchr(real_filename, ':')) != NULL)
 803         real_filename = temp_str + 1;
 804     snprintf(desc, sizeof(desc), desc_template, (unsigned int)time(NULL),
 805              total_size, real_filename,
 806              (flags & BLOCK_FLAG_COMPAT6 ? 6 : 4),
 807              total_size / (int64_t)(63 * 16));
 808
 809     /* write the descriptor */
 810     lseek(fd, le64_to_cpu(header.desc_offset) << 9, SEEK_SET);
 811     ret = qemu_write_full(fd, desc, strlen(desc));
 812     if (ret != strlen(desc)) {
 813         ret = -errno;
 814         goto exit;
 815     }
 816
 817     ret = 0;
 818 exit:
 819     close(fd);
 820     return ret;
 821 }
 822
 823 static void vmdk_close(BlockDriverState *bs)
 824 {
 825     BDRVVmdkState *s = bs->opaque;
 826
 827     qemu_free(s->l1_table);
 828     qemu_free(s->l2_cache);
 829 }
 830
 831 static int vmdk_flush(BlockDriverState *bs)
 832 {
 833     return bdrv_flush(bs->file);
 834 }
 835
 836
 837 static QEMUOptionParameter vmdk_create_options[] = {
 838     {
 839         .name = BLOCK_OPT_SIZE,
 840         .type = OPT_SIZE,
 841         .help = "Virtual disk size"
 842     },
 843     {
 844         .name = BLOCK_OPT_BACKING_FILE,
 845         .type = OPT_STRING,
 846         .help = "File name of a base image"
 847     },
 848     {
 849         .name = BLOCK_OPT_COMPAT6,
 850         .type = OPT_FLAG,
 851         .help = "VMDK version 6 image"
 852     },
 853     { NULL }
 854 };
 855
 856 static BlockDriver bdrv_vmdk = {
 857     .format_name        = "vmdk",
 858     .instance_size      = sizeof(BDRVVmdkState),
 859     .bdrv_probe         = vmdk_probe,
 860     .bdrv_open      = vmdk_open,
 861     .bdrv_read          = vmdk_read,
 862     .bdrv_write         = vmdk_write,
 863     .bdrv_close         = vmdk_close,
 864     .bdrv_create        = vmdk_create,
 865     .bdrv_flush         = vmdk_flush,
 866     .bdrv_is_allocated  = vmdk_is_allocated,
 867
 868     .create_options = vmdk_create_options,
 869 };
 870
 871 static void bdrv_vmdk_init(void)
 872 {
 873     bdrv_register(&bdrv_vmdk);
 874 }
 875
 876 block_init(bdrv_vmdk_init);