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 = bs->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     int ret;
 191     uint32_t p_cid;
 192     char *p_name, *gd_buf, *rgd_buf;
 193     const char *real_filename, *temp_str;
 194     VMDK4Header header;
 195     uint32_t gde_entries, gd_size;
 196     int64_t gd_offset, rgd_offset, capacity, gt_size;
 197     char p_desc[DESC_SIZE], s_desc[DESC_SIZE], hdr[HEADER_SIZE];
 198     static const char desc_template[] =
 199     "# Disk DescriptorFile\n"
 200     "version=1\n"
 201     "CID=%x\n"
 202     "parentCID=%x\n"
 203     "createType=\"monolithicSparse\"\n"
 204     "parentFileNameHint=\"%s\"\n"
 205     "\n"
 206     "# Extent description\n"
 207     "RW %u SPARSE \"%s\"\n"
 208     "\n"
 209     "# The Disk Data Base \n"
 210     "#DDB\n"
 211     "\n";
 212
 213     snp_fd = open(filename, O_RDWR | O_CREAT | O_TRUNC | O_BINARY | O_LARGEFILE, 0644);
 214     if (snp_fd < 0)
 215         return -errno;
 216     p_fd = open(backing_file, O_RDONLY | O_BINARY | O_LARGEFILE);
 217     if (p_fd < 0) {
 218         close(snp_fd);
 219         return -errno;
 220     }
 221
 222     /* read the header */
 223     if (lseek(p_fd, 0x0, SEEK_SET) == -1) {
 224         ret = -errno;
 225         goto fail;
 226     }
 227     if (read(p_fd, hdr, HEADER_SIZE) != HEADER_SIZE) {
 228         ret = -errno;
 229         goto fail;
 230     }
 231
 232     /* write the header */
 233     if (lseek(snp_fd, 0x0, SEEK_SET) == -1) {
 234         ret = -errno;
 235         goto fail;
 236     }
 237     if (write(snp_fd, hdr, HEADER_SIZE) == -1) {
 238         ret = -errno;
 239         goto fail;
 240     }
 241
 242     memset(&header, 0, sizeof(header));
 243     memcpy(&header,&hdr[4], sizeof(header)); // skip the VMDK4_MAGIC
 244
 245     if (ftruncate(snp_fd, header.grain_offset << 9)) {
 246         ret = -errno;
 247         goto fail;
 248     }
 249     /* the descriptor offset = 0x200 */
 250     if (lseek(p_fd, 0x200, SEEK_SET) == -1) {
 251         ret = -errno;
 252         goto fail;
 253     }
 254     if (read(p_fd, p_desc, DESC_SIZE) != DESC_SIZE) {
 255         ret = -errno;
 256         goto fail;
 257     }
 258
 259     if ((p_name = strstr(p_desc,"CID")) != NULL) {
 260         p_name += sizeof("CID");
 261         sscanf(p_name,"%x",&p_cid);
 262     }
 263
 264     real_filename = filename;
 265     if ((temp_str = strrchr(real_filename, '\\')) != NULL)
 266         real_filename = temp_str + 1;
 267     if ((temp_str = strrchr(real_filename, '/')) != NULL)
 268         real_filename = temp_str + 1;
 269     if ((temp_str = strrchr(real_filename, ':')) != NULL)
 270         real_filename = temp_str + 1;
 271
 272     snprintf(s_desc, sizeof(s_desc), desc_template, p_cid, p_cid, backing_file,
 273              (uint32_t)header.capacity, real_filename);
 274
 275     /* write the descriptor */
 276     if (lseek(snp_fd, 0x200, SEEK_SET) == -1) {
 277         ret = -errno;
 278         goto fail;
 279     }
 280     if (write(snp_fd, s_desc, strlen(s_desc)) == -1) {
 281         ret = -errno;
 282         goto fail;
 283     }
 284
 285     gd_offset = header.gd_offset * SECTOR_SIZE;     // offset of GD table
 286     rgd_offset = header.rgd_offset * SECTOR_SIZE;   // offset of RGD table
 287     capacity = header.capacity * SECTOR_SIZE;       // Extent size
 288     /*
 289      * Each GDE span 32M disk, means:
 290      * 512 GTE per GT, each GTE points to grain
 291      */
 292     gt_size = (int64_t)header.num_gtes_per_gte * header.granularity * SECTOR_SIZE;
 293     if (!gt_size) {
 294         ret = -EINVAL;
 295         goto fail;
 296     }
 297     gde_entries = (uint32_t)(capacity / gt_size);  // number of gde/rgde
 298     gd_size = gde_entries * sizeof(uint32_t);
 299
 300     /* write RGD */
 301     rgd_buf = qemu_malloc(gd_size);
 302     if (lseek(p_fd, rgd_offset, SEEK_SET) == -1) {
 303         ret = -errno;
 304         goto fail_rgd;
 305     }
 306     if (read(p_fd, rgd_buf, gd_size) != gd_size) {
 307         ret = -errno;
 308         goto fail_rgd;
 309     }
 310     if (lseek(snp_fd, rgd_offset, SEEK_SET) == -1) {
 311         ret = -errno;
 312         goto fail_rgd;
 313     }
 314     if (write(snp_fd, rgd_buf, gd_size) == -1) {
 315         ret = -errno;
 316         goto fail_rgd;
 317     }
 318
 319     /* write GD */
 320     gd_buf = qemu_malloc(gd_size);
 321     if (lseek(p_fd, gd_offset, SEEK_SET) == -1) {
 322         ret = -errno;
 323         goto fail_gd;
 324     }
 325     if (read(p_fd, gd_buf, gd_size) != gd_size) {
 326         ret = -errno;
 327         goto fail_gd;
 328     }
 329     if (lseek(snp_fd, gd_offset, SEEK_SET) == -1) {
 330         ret = -errno;
 331         goto fail_gd;
 332     }
 333     if (write(snp_fd, gd_buf, gd_size) == -1) {
 334         ret = -errno;
 335         goto fail_gd;
 336     }
 337     qemu_free(gd_buf);
 338     qemu_free(rgd_buf);
 339
 340     close(p_fd);
 341     close(snp_fd);
 342     return 0;
 343
 344     fail_gd:
 345     qemu_free(gd_buf);
 346     fail_rgd:
 347     qemu_free(rgd_buf);
 348     fail:
 349     close(p_fd);
 350     close(snp_fd);
 351     return ret;
 352 }
 353
 354 static void vmdk_parent_close(BlockDriverState *bs)
 355 {
 356     if (bs->backing_hd)
 357         bdrv_close(bs->backing_hd);
 358 }
 359
 360 static int parent_open = 0;
 361 static int vmdk_parent_open(BlockDriverState *bs, const char * filename)
 362 {
 363     BDRVVmdkState *s = bs->opaque;
 364     char *p_name;
 365     char desc[DESC_SIZE];
 366     char parent_img_name[1024];
 367
 368     /* the descriptor offset = 0x200 */
 369     if (bdrv_pread(s->hd, 0x200, desc, DESC_SIZE) != DESC_SIZE)
 370         return -1;
 371
 372     if ((p_name = strstr(desc,"parentFileNameHint")) != NULL) {
 373         char *end_name;
 374         struct stat file_buf;
 375
 376         p_name += sizeof("parentFileNameHint") + 1;
 377         if ((end_name = strchr(p_name,'\"')) == NULL)
 378             return -1;
 379         if ((end_name - p_name) > sizeof (bs->backing_file) - 1)
 380             return -1;
 381
 382         pstrcpy(bs->backing_file, end_name - p_name + 1, p_name);
 383         if (stat(bs->backing_file, &file_buf) != 0) {
 384             path_combine(parent_img_name, sizeof(parent_img_name),
 385                          filename, bs->backing_file);
 386         } else {
 387             pstrcpy(parent_img_name, sizeof(parent_img_name),
 388                     bs->backing_file);
 389         }
 390
 391         bs->backing_hd = bdrv_new("");
 392         if (!bs->backing_hd) {
 393             failure:
 394             bdrv_close(s->hd);
 395             return -1;
 396         }
 397         parent_open = 1;
 398         if (bdrv_open(bs->backing_hd, parent_img_name, 0) < 0)
 399             goto failure;
 400         parent_open = 0;
 401     }
 402
 403     return 0;
 404 }
 405
 406 static int vmdk_open(BlockDriverState *bs, const char *filename, int flags)
 407 {
 408     BDRVVmdkState *s = bs->opaque;
 409     uint32_t magic;
 410     int l1_size, i, ret;
 411
 412     if (parent_open) {
 413         /* Parent must be opened as RO, no RDWR. */
 414         flags = 0;
 415     }
 416
 417     ret = bdrv_file_open(&s->hd, filename, flags);
 418     if (ret < 0)
 419         return ret;
 420     if (bdrv_pread(s->hd, 0, &magic, sizeof(magic)) != sizeof(magic))
 421         goto fail;
 422
 423     magic = be32_to_cpu(magic);
 424     if (magic == VMDK3_MAGIC) {
 425         VMDK3Header header;
 426
 427         if (bdrv_pread(s->hd, sizeof(magic), &header, sizeof(header)) != sizeof(header))
 428             goto fail;
 429         s->cluster_sectors = le32_to_cpu(header.granularity);
 430         s->l2_size = 1 << 9;
 431         s->l1_size = 1 << 6;
 432         bs->total_sectors = le32_to_cpu(header.disk_sectors);
 433         s->l1_table_offset = le32_to_cpu(header.l1dir_offset) << 9;
 434         s->l1_backup_table_offset = 0;
 435         s->l1_entry_sectors = s->l2_size * s->cluster_sectors;
 436     } else if (magic == VMDK4_MAGIC) {
 437         VMDK4Header header;
 438
 439         if (bdrv_pread(s->hd, sizeof(magic), &header, sizeof(header)) != sizeof(header))
 440             goto fail;
 441         bs->total_sectors = le64_to_cpu(header.capacity);
 442         s->cluster_sectors = le64_to_cpu(header.granularity);
 443         s->l2_size = le32_to_cpu(header.num_gtes_per_gte);
 444         s->l1_entry_sectors = s->l2_size * s->cluster_sectors;
 445         if (s->l1_entry_sectors <= 0)
 446             goto fail;
 447         s->l1_size = (bs->total_sectors + s->l1_entry_sectors - 1)
 448             / s->l1_entry_sectors;
 449         s->l1_table_offset = le64_to_cpu(header.rgd_offset) << 9;
 450         s->l1_backup_table_offset = le64_to_cpu(header.gd_offset) << 9;
 451
 452         if (parent_open)
 453             s->is_parent = 1;
 454         else
 455             s->is_parent = 0;
 456
 457         // try to open parent images, if exist
 458         if (vmdk_parent_open(bs, filename) != 0)
 459             goto fail;
 460         // write the CID once after the image creation
 461         s->parent_cid = vmdk_read_cid(bs,1);
 462     } else {
 463         goto fail;
 464     }
 465
 466     /* read the L1 table */
 467     l1_size = s->l1_size * sizeof(uint32_t);
 468     s->l1_table = qemu_malloc(l1_size);
 469     if (bdrv_pread(s->hd, s->l1_table_offset, s->l1_table, l1_size) != l1_size)
 470         goto fail;
 471     for(i = 0; i < s->l1_size; i++) {
 472         le32_to_cpus(&s->l1_table[i]);
 473     }
 474
 475     if (s->l1_backup_table_offset) {
 476         s->l1_backup_table = qemu_malloc(l1_size);
 477         if (bdrv_pread(s->hd, s->l1_backup_table_offset, s->l1_backup_table, l1_size) != l1_size)
 478             goto fail;
 479         for(i = 0; i < s->l1_size; i++) {
 480             le32_to_cpus(&s->l1_backup_table[i]);
 481         }
 482     }
 483
 484     s->l2_cache = qemu_malloc(s->l2_size * L2_CACHE_SIZE * sizeof(uint32_t));
 485     return 0;
 486  fail:
 487     qemu_free(s->l1_backup_table);
 488     qemu_free(s->l1_table);
 489     qemu_free(s->l2_cache);
 490     bdrv_delete(s->hd);
 491     return -1;
 492 }
 493
 494 static uint64_t get_cluster_offset(BlockDriverState *bs, VmdkMetaData *m_data,
 495                                    uint64_t offset, int allocate);
 496
 497 static int get_whole_cluster(BlockDriverState *bs, uint64_t cluster_offset,
 498                              uint64_t offset, int allocate)
 499 {
 500     uint64_t parent_cluster_offset;
 501     BDRVVmdkState *s = bs->opaque;
 502     uint8_t  whole_grain[s->cluster_sectors*512];        // 128 sectors * 512 bytes each = grain size 64KB
 503
 504     // we will be here if it's first write on non-exist grain(cluster).
 505     // try to read from parent image, if exist
 506     if (bs->backing_hd) {
 507         BDRVVmdkState *ps = bs->backing_hd->opaque;
 508
 509         if (!vmdk_is_cid_valid(bs))
 510             return -1;
 511
 512         parent_cluster_offset = get_cluster_offset(bs->backing_hd, NULL,
 513             offset, allocate);
 514
 515         if (parent_cluster_offset) {
 516             BDRVVmdkState *act_s = activeBDRV.hd->opaque;
 517
 518             if (bdrv_pread(ps->hd, parent_cluster_offset, whole_grain, ps->cluster_sectors*512) != ps->cluster_sectors*512)
 519                 return -1;
 520
 521             //Write grain only into the active image
 522             if (bdrv_pwrite(act_s->hd, activeBDRV.cluster_offset << 9, whole_grain, sizeof(whole_grain)) != sizeof(whole_grain))
 523                 return -1;
 524         }
 525     }
 526     return 0;
 527 }
 528
 529 static int vmdk_L2update(BlockDriverState *bs, VmdkMetaData *m_data)
 530 {
 531     BDRVVmdkState *s = bs->opaque;
 532
 533     /* update L2 table */
 534     if (bdrv_pwrite(s->hd, ((int64_t)m_data->l2_offset * 512) + (m_data->l2_index * sizeof(m_data->offset)),
 535                     &(m_data->offset), sizeof(m_data->offset)) != sizeof(m_data->offset))
 536         return -1;
 537     /* update backup L2 table */
 538     if (s->l1_backup_table_offset != 0) {
 539         m_data->l2_offset = s->l1_backup_table[m_data->l1_index];
 540         if (bdrv_pwrite(s->hd, ((int64_t)m_data->l2_offset * 512) + (m_data->l2_index * sizeof(m_data->offset)),
 541                         &(m_data->offset), sizeof(m_data->offset)) != sizeof(m_data->offset))
 542             return -1;
 543     }
 544
 545     return 0;
 546 }
 547
 548 static uint64_t get_cluster_offset(BlockDriverState *bs, VmdkMetaData *m_data,
 549                                    uint64_t offset, int allocate)
 550 {
 551     BDRVVmdkState *s = bs->opaque;
 552     unsigned int l1_index, l2_offset, l2_index;
 553     int min_index, i, j;
 554     uint32_t min_count, *l2_table, tmp = 0;
 555     uint64_t cluster_offset;
 556
 557     if (m_data)
 558         m_data->valid = 0;
 559
 560     l1_index = (offset >> 9) / s->l1_entry_sectors;
 561     if (l1_index >= s->l1_size)
 562         return 0;
 563     l2_offset = s->l1_table[l1_index];
 564     if (!l2_offset)
 565         return 0;
 566     for(i = 0; i < L2_CACHE_SIZE; i++) {
 567         if (l2_offset == s->l2_cache_offsets[i]) {
 568             /* increment the hit count */
 569             if (++s->l2_cache_counts[i] == 0xffffffff) {
 570                 for(j = 0; j < L2_CACHE_SIZE; j++) {
 571                     s->l2_cache_counts[j] >>= 1;
 572                 }
 573             }
 574             l2_table = s->l2_cache + (i * s->l2_size);
 575             goto found;
 576         }
 577     }
 578     /* not found: load a new entry in the least used one */
 579     min_index = 0;
 580     min_count = 0xffffffff;
 581     for(i = 0; i < L2_CACHE_SIZE; i++) {
 582         if (s->l2_cache_counts[i] < min_count) {
 583             min_count = s->l2_cache_counts[i];
 584             min_index = i;
 585         }
 586     }
 587     l2_table = s->l2_cache + (min_index * s->l2_size);
 588     if (bdrv_pread(s->hd, (int64_t)l2_offset * 512, l2_table, s->l2_size * sizeof(uint32_t)) !=
 589                                                                         s->l2_size * sizeof(uint32_t))
 590         return 0;
 591
 592     s->l2_cache_offsets[min_index] = l2_offset;
 593     s->l2_cache_counts[min_index] = 1;
 594  found:
 595     l2_index = ((offset >> 9) / s->cluster_sectors) % s->l2_size;
 596     cluster_offset = le32_to_cpu(l2_table[l2_index]);
 597
 598     if (!cluster_offset) {
 599         if (!allocate)
 600             return 0;
 601         // Avoid the L2 tables update for the images that have snapshots.
 602         if (!s->is_parent) {
 603             cluster_offset = bdrv_getlength(s->hd);
 604             bdrv_truncate(s->hd, cluster_offset + (s->cluster_sectors << 9));
 605
 606             cluster_offset >>= 9;
 607             tmp = cpu_to_le32(cluster_offset);
 608             l2_table[l2_index] = tmp;
 609             // Save the active image state
 610             activeBDRV.cluster_offset = cluster_offset;
 611             activeBDRV.hd = bs;
 612         }
 613         /* First of all we write grain itself, to avoid race condition
 614          * that may to corrupt the image.
 615          * This problem may occur because of insufficient space on host disk
 616          * or inappropriate VM shutdown.
 617          */
 618         if (get_whole_cluster(bs, cluster_offset, offset, allocate) == -1)
 619             return 0;
 620
 621         if (m_data) {
 622             m_data->offset = tmp;
 623             m_data->l1_index = l1_index;
 624             m_data->l2_index = l2_index;
 625             m_data->l2_offset = l2_offset;
 626             m_data->valid = 1;
 627         }
 628     }
 629     cluster_offset <<= 9;
 630     return cluster_offset;
 631 }
 632
 633 static int vmdk_is_allocated(BlockDriverState *bs, int64_t sector_num,
 634                              int nb_sectors, int *pnum)
 635 {
 636     BDRVVmdkState *s = bs->opaque;
 637     int index_in_cluster, n;
 638     uint64_t cluster_offset;
 639
 640     cluster_offset = get_cluster_offset(bs, NULL, sector_num << 9, 0);
 641     index_in_cluster = sector_num % s->cluster_sectors;
 642     n = s->cluster_sectors - index_in_cluster;
 643     if (n > nb_sectors)
 644         n = nb_sectors;
 645     *pnum = n;
 646     return (cluster_offset != 0);
 647 }
 648
 649 static int vmdk_read(BlockDriverState *bs, int64_t sector_num,
 650                     uint8_t *buf, int nb_sectors)
 651 {
 652     BDRVVmdkState *s = bs->opaque;
 653     int index_in_cluster, n, ret;
 654     uint64_t cluster_offset;
 655
 656     while (nb_sectors > 0) {
 657         cluster_offset = get_cluster_offset(bs, NULL, sector_num << 9, 0);
 658         index_in_cluster = sector_num % s->cluster_sectors;
 659         n = s->cluster_sectors - index_in_cluster;
 660         if (n > nb_sectors)
 661             n = nb_sectors;
 662         if (!cluster_offset) {
 663             // try to read from parent image, if exist
 664             if (bs->backing_hd) {
 665                 if (!vmdk_is_cid_valid(bs))
 666                     return -1;
 667                 ret = bdrv_read(bs->backing_hd, sector_num, buf, n);
 668                 if (ret < 0)
 669                     return -1;
 670             } else {
 671                 memset(buf, 0, 512 * n);
 672             }
 673         } else {
 674             if(bdrv_pread(s->hd, cluster_offset + index_in_cluster * 512, buf, n * 512) != n * 512)
 675                 return -1;
 676         }
 677         nb_sectors -= n;
 678         sector_num += n;
 679         buf += n * 512;
 680     }
 681     return 0;
 682 }
 683
 684 static int vmdk_write(BlockDriverState *bs, int64_t sector_num,
 685                      const uint8_t *buf, int nb_sectors)
 686 {
 687     BDRVVmdkState *s = bs->opaque;
 688     VmdkMetaData m_data;
 689     int index_in_cluster, n;
 690     uint64_t cluster_offset;
 691     static int cid_update = 0;
 692
 693     if (sector_num > bs->total_sectors) {
 694         fprintf(stderr,
 695                 "(VMDK) Wrong offset: sector_num=0x%" PRIx64
 696                 " total_sectors=0x%" PRIx64 "\n",
 697                 sector_num, bs->total_sectors);
 698         return -1;
 699     }
 700
 701     while (nb_sectors > 0) {
 702         index_in_cluster = sector_num & (s->cluster_sectors - 1);
 703         n = s->cluster_sectors - index_in_cluster;
 704         if (n > nb_sectors)
 705             n = nb_sectors;
 706         cluster_offset = get_cluster_offset(bs, &m_data, sector_num << 9, 1);
 707         if (!cluster_offset)
 708             return -1;
 709
 710         if (bdrv_pwrite(s->hd, cluster_offset + index_in_cluster * 512, buf, n * 512) != n * 512)
 711             return -1;
 712         if (m_data.valid) {
 713             /* update L2 tables */
 714             if (vmdk_L2update(bs, &m_data) == -1)
 715                 return -1;
 716         }
 717         nb_sectors -= n;
 718         sector_num += n;
 719         buf += n * 512;
 720
 721         // update CID on the first write every time the virtual disk is opened
 722         if (!cid_update) {
 723             vmdk_write_cid(bs, time(NULL));
 724             cid_update++;
 725         }
 726     }
 727     return 0;
 728 }
 729
 730 static int vmdk_create(const char *filename, QEMUOptionParameter *options)
 731 {
 732     int fd, i;
 733     VMDK4Header header;
 734     uint32_t tmp, magic, grains, gd_size, gt_size, gt_count;
 735     static const char desc_template[] =
 736         "# Disk DescriptorFile\n"
 737         "version=1\n"
 738         "CID=%x\n"
 739         "parentCID=ffffffff\n"
 740         "createType=\"monolithicSparse\"\n"
 741         "\n"
 742         "# Extent description\n"
 743         "RW %" PRId64 " SPARSE \"%s\"\n"
 744         "\n"
 745         "# The Disk Data Base \n"
 746         "#DDB\n"
 747         "\n"
 748         "ddb.virtualHWVersion = \"%d\"\n"
 749         "ddb.geometry.cylinders = \"%" PRId64 "\"\n"
 750         "ddb.geometry.heads = \"16\"\n"
 751         "ddb.geometry.sectors = \"63\"\n"
 752         "ddb.adapterType = \"ide\"\n";
 753     char desc[1024];
 754     const char *real_filename, *temp_str;
 755     int64_t total_size = 0;
 756     const char *backing_file = NULL;
 757     int flags = 0;
 758     int ret;
 759
 760     // Read out options
 761     while (options && options->name) {
 762         if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
 763             total_size = options->value.n / 512;
 764         } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) {
 765             backing_file = options->value.s;
 766         } else if (!strcmp(options->name, BLOCK_OPT_COMPAT6)) {
 767             flags |= options->value.n ? BLOCK_FLAG_COMPAT6: 0;
 768         }
 769         options++;
 770     }
 771
 772     /* XXX: add support for backing file */
 773     if (backing_file) {
 774         return vmdk_snapshot_create(filename, backing_file);
 775     }
 776
 777     fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY | O_LARGEFILE,
 778               0644);
 779     if (fd < 0)
 780         return -errno;
 781     magic = cpu_to_be32(VMDK4_MAGIC);
 782     memset(&header, 0, sizeof(header));
 783     header.version = cpu_to_le32(1);
 784     header.flags = cpu_to_le32(3); /* ?? */
 785     header.capacity = cpu_to_le64(total_size);
 786     header.granularity = cpu_to_le64(128);
 787     header.num_gtes_per_gte = cpu_to_le32(512);
 788
 789     grains = (total_size + header.granularity - 1) / header.granularity;
 790     gt_size = ((header.num_gtes_per_gte * sizeof(uint32_t)) + 511) >> 9;
 791     gt_count = (grains + header.num_gtes_per_gte - 1) / header.num_gtes_per_gte;
 792     gd_size = (gt_count * sizeof(uint32_t) + 511) >> 9;
 793
 794     header.desc_offset = 1;
 795     header.desc_size = 20;
 796     header.rgd_offset = header.desc_offset + header.desc_size;
 797     header.gd_offset = header.rgd_offset + gd_size + (gt_size * gt_count);
 798     header.grain_offset =
 799        ((header.gd_offset + gd_size + (gt_size * gt_count) +
 800          header.granularity - 1) / header.granularity) *
 801         header.granularity;
 802
 803     header.desc_offset = cpu_to_le64(header.desc_offset);
 804     header.desc_size = cpu_to_le64(header.desc_size);
 805     header.rgd_offset = cpu_to_le64(header.rgd_offset);
 806     header.gd_offset = cpu_to_le64(header.gd_offset);
 807     header.grain_offset = cpu_to_le64(header.grain_offset);
 808
 809     header.check_bytes[0] = 0xa;
 810     header.check_bytes[1] = 0x20;
 811     header.check_bytes[2] = 0xd;
 812     header.check_bytes[3] = 0xa;
 813
 814     /* write all the data */
 815     ret = qemu_write_full(fd, &magic, sizeof(magic));
 816     if (ret != sizeof(magic)) {
 817         ret = -errno;
 818         goto exit;
 819     }
 820     ret = qemu_write_full(fd, &header, sizeof(header));
 821     if (ret != sizeof(header)) {
 822         ret = -errno;
 823         goto exit;
 824     }
 825
 826     ret = ftruncate(fd, header.grain_offset << 9);
 827     if (ret < 0) {
 828         ret = -errno;
 829         goto exit;
 830     }
 831
 832     /* write grain directory */
 833     lseek(fd, le64_to_cpu(header.rgd_offset) << 9, SEEK_SET);
 834     for (i = 0, tmp = header.rgd_offset + gd_size;
 835          i < gt_count; i++, tmp += gt_size) {
 836         ret = qemu_write_full(fd, &tmp, sizeof(tmp));
 837         if (ret != sizeof(tmp)) {
 838             ret = -errno;
 839             goto exit;
 840         }
 841     }
 842
 843     /* write backup grain directory */
 844     lseek(fd, le64_to_cpu(header.gd_offset) << 9, SEEK_SET);
 845     for (i = 0, tmp = header.gd_offset + gd_size;
 846          i < gt_count; i++, tmp += gt_size) {
 847         ret = qemu_write_full(fd, &tmp, sizeof(tmp));
 848         if (ret != sizeof(tmp)) {
 849             ret = -errno;
 850             goto exit;
 851         }
 852     }
 853
 854     /* compose the descriptor */
 855     real_filename = filename;
 856     if ((temp_str = strrchr(real_filename, '\\')) != NULL)
 857         real_filename = temp_str + 1;
 858     if ((temp_str = strrchr(real_filename, '/')) != NULL)
 859         real_filename = temp_str + 1;
 860     if ((temp_str = strrchr(real_filename, ':')) != NULL)
 861         real_filename = temp_str + 1;
 862     snprintf(desc, sizeof(desc), desc_template, (unsigned int)time(NULL),
 863              total_size, real_filename,
 864              (flags & BLOCK_FLAG_COMPAT6 ? 6 : 4),
 865              total_size / (int64_t)(63 * 16));
 866
 867     /* write the descriptor */
 868     lseek(fd, le64_to_cpu(header.desc_offset) << 9, SEEK_SET);
 869     ret = qemu_write_full(fd, desc, strlen(desc));
 870     if (ret != strlen(desc)) {
 871         ret = -errno;
 872         goto exit;
 873     }
 874
 875     ret = 0;
 876 exit:
 877     close(fd);
 878     return ret;
 879 }
 880
 881 static void vmdk_close(BlockDriverState *bs)
 882 {
 883     BDRVVmdkState *s = bs->opaque;
 884
 885     qemu_free(s->l1_table);
 886     qemu_free(s->l2_cache);
 887     // try to close parent image, if exist
 888     vmdk_parent_close(s->hd);
 889     bdrv_delete(s->hd);
 890 }
 891
 892 static void vmdk_flush(BlockDriverState *bs)
 893 {
 894     BDRVVmdkState *s = bs->opaque;
 895     bdrv_flush(s->hd);
 896 }
 897
 898
 899 static QEMUOptionParameter vmdk_create_options[] = {
 900     {
 901         .name = BLOCK_OPT_SIZE,
 902         .type = OPT_SIZE,
 903         .help = "Virtual disk size"
 904     },
 905     {
 906         .name = BLOCK_OPT_BACKING_FILE,
 907         .type = OPT_STRING,
 908         .help = "File name of a base image"
 909     },
 910     {
 911         .name = BLOCK_OPT_COMPAT6,
 912         .type = OPT_FLAG,
 913         .help = "VMDK version 6 image"
 914     },
 915     { NULL }
 916 };
 917
 918 static BlockDriver bdrv_vmdk = {
 919     .format_name        = "vmdk",
 920     .instance_size      = sizeof(BDRVVmdkState),
 921     .bdrv_probe         = vmdk_probe,
 922     .bdrv_open          = vmdk_open,
 923     .bdrv_read          = vmdk_read,
 924     .bdrv_write         = vmdk_write,
 925     .bdrv_close         = vmdk_close,
 926     .bdrv_create        = vmdk_create,
 927     .bdrv_flush         = vmdk_flush,
 928     .bdrv_is_allocated  = vmdk_is_allocated,
 929
 930     .create_options = vmdk_create_options,
 931 };
 932
 933 static void bdrv_vmdk_init(void)
 934 {
 935     bdrv_register(&bdrv_vmdk);
 936 }
 937
 938 block_init(bdrv_vmdk_init);