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