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 VmdkExtent {
  64     BlockDriverState *file;
  65     bool flat;
  66     int64_t sectors;
  67     int64_t end_sector;
  68     int64_t l1_table_offset;
  69     int64_t l1_backup_table_offset;
  70     uint32_t *l1_table;
  71     uint32_t *l1_backup_table;
  72     unsigned int l1_size;
  73     uint32_t l1_entry_sectors;
  74
  75     unsigned int l2_size;
  76     uint32_t *l2_cache;
  77     uint32_t l2_cache_offsets[L2_CACHE_SIZE];
  78     uint32_t l2_cache_counts[L2_CACHE_SIZE];
  79
  80     unsigned int cluster_sectors;
  81 } VmdkExtent;
  82
  83 typedef struct BDRVVmdkState {
  84     uint32_t parent_cid;
  85     int num_extents;
  86     /* Extent array with num_extents entries, ascend ordered by address */
  87     VmdkExtent *extents;
  88 } BDRVVmdkState;
  89
  90 typedef struct VmdkMetaData {
  91     uint32_t offset;
  92     unsigned int l1_index;
  93     unsigned int l2_index;
  94     unsigned int l2_offset;
  95     int valid;
  96 } VmdkMetaData;
  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         const char *p = (const char *)buf;
 110         const char *end = p + buf_size;
 111         while (p < end) {
 112             if (*p == '#') {
 113                 /* skip comment line */
 114                 while (p < end && *p != '\n') {
 115                     p++;
 116                 }
 117                 p++;
 118                 continue;
 119             }
 120             if (*p == ' ') {
 121                 while (p < end && *p == ' ') {
 122                     p++;
 123                 }
 124                 /* skip '\r' if windows line endings used. */
 125                 if (p < end && *p == '\r') {
 126                     p++;
 127                 }
 128                 /* only accept blank lines before 'version=' line */
 129                 if (p == end || *p != '\n') {
 130                     return 0;
 131                 }
 132                 p++;
 133                 continue;
 134             }
 135             if (end - p >= strlen("version=X\n")) {
 136                 if (strncmp("version=1\n", p, strlen("version=1\n")) == 0 ||
 137                     strncmp("version=2\n", p, strlen("version=2\n")) == 0) {
 138                     return 100;
 139                 }
 140             }
 141             if (end - p >= strlen("version=X\r\n")) {
 142                 if (strncmp("version=1\r\n", p, strlen("version=1\r\n")) == 0 ||
 143                     strncmp("version=2\r\n", p, strlen("version=2\r\n")) == 0) {
 144                     return 100;
 145                 }
 146             }
 147             return 0;
 148         }
 149         return 0;
 150     }
 151 }
 152
 153 #define CHECK_CID 1
 154
 155 #define SECTOR_SIZE 512
 156 #define DESC_SIZE 20*SECTOR_SIZE        // 20 sectors of 512 bytes each
 157 #define HEADER_SIZE 512                         // first sector of 512 bytes
 158
 159 static void vmdk_free_extents(BlockDriverState *bs)
 160 {
 161     int i;
 162     BDRVVmdkState *s = bs->opaque;
 163
 164     for (i = 0; i < s->num_extents; i++) {
 165         qemu_free(s->extents[i].l1_table);
 166         qemu_free(s->extents[i].l2_cache);
 167         qemu_free(s->extents[i].l1_backup_table);
 168     }
 169     qemu_free(s->extents);
 170 }
 171
 172 static uint32_t vmdk_read_cid(BlockDriverState *bs, int parent)
 173 {
 174     char desc[DESC_SIZE];
 175     uint32_t cid;
 176     const char *p_name, *cid_str;
 177     size_t cid_str_size;
 178
 179     /* the descriptor offset = 0x200 */
 180     if (bdrv_pread(bs->file, 0x200, desc, DESC_SIZE) != DESC_SIZE)
 181         return 0;
 182
 183     if (parent) {
 184         cid_str = "parentCID";
 185         cid_str_size = sizeof("parentCID");
 186     } else {
 187         cid_str = "CID";
 188         cid_str_size = sizeof("CID");
 189     }
 190
 191     if ((p_name = strstr(desc,cid_str)) != NULL) {
 192         p_name += cid_str_size;
 193         sscanf(p_name,"%x",&cid);
 194     }
 195
 196     return cid;
 197 }
 198
 199 static int vmdk_write_cid(BlockDriverState *bs, uint32_t cid)
 200 {
 201     char desc[DESC_SIZE], tmp_desc[DESC_SIZE];
 202     char *p_name, *tmp_str;
 203
 204     /* the descriptor offset = 0x200 */
 205     if (bdrv_pread(bs->file, 0x200, desc, DESC_SIZE) != DESC_SIZE)
 206         return -1;
 207
 208     tmp_str = strstr(desc,"parentCID");
 209     pstrcpy(tmp_desc, sizeof(tmp_desc), tmp_str);
 210     if ((p_name = strstr(desc,"CID")) != NULL) {
 211         p_name += sizeof("CID");
 212         snprintf(p_name, sizeof(desc) - (p_name - desc), "%x\n", cid);
 213         pstrcat(desc, sizeof(desc), tmp_desc);
 214     }
 215
 216     if (bdrv_pwrite_sync(bs->file, 0x200, desc, DESC_SIZE) < 0)
 217         return -1;
 218     return 0;
 219 }
 220
 221 static int vmdk_is_cid_valid(BlockDriverState *bs)
 222 {
 223 #ifdef CHECK_CID
 224     BDRVVmdkState *s = bs->opaque;
 225     BlockDriverState *p_bs = bs->backing_hd;
 226     uint32_t cur_pcid;
 227
 228     if (p_bs) {
 229         cur_pcid = vmdk_read_cid(p_bs,0);
 230         if (s->parent_cid != cur_pcid)
 231             // CID not valid
 232             return 0;
 233     }
 234 #endif
 235     // CID valid
 236     return 1;
 237 }
 238
 239 static int vmdk_snapshot_create(const char *filename, const char *backing_file)
 240 {
 241     int snp_fd, p_fd;
 242     int ret;
 243     uint32_t p_cid;
 244     char *p_name, *gd_buf, *rgd_buf;
 245     const char *real_filename, *temp_str;
 246     VMDK4Header header;
 247     uint32_t gde_entries, gd_size;
 248     int64_t gd_offset, rgd_offset, capacity, gt_size;
 249     char p_desc[DESC_SIZE], s_desc[DESC_SIZE], hdr[HEADER_SIZE];
 250     static const char desc_template[] =
 251     "# Disk DescriptorFile\n"
 252     "version=1\n"
 253     "CID=%x\n"
 254     "parentCID=%x\n"
 255     "createType=\"monolithicSparse\"\n"
 256     "parentFileNameHint=\"%s\"\n"
 257     "\n"
 258     "# Extent description\n"
 259     "RW %u SPARSE \"%s\"\n"
 260     "\n"
 261     "# The Disk Data Base \n"
 262     "#DDB\n"
 263     "\n";
 264
 265     snp_fd = open(filename, O_RDWR | O_CREAT | O_TRUNC | O_BINARY | O_LARGEFILE, 0644);
 266     if (snp_fd < 0)
 267         return -errno;
 268     p_fd = open(backing_file, O_RDONLY | O_BINARY | O_LARGEFILE);
 269     if (p_fd < 0) {
 270         close(snp_fd);
 271         return -errno;
 272     }
 273
 274     /* read the header */
 275     if (lseek(p_fd, 0x0, SEEK_SET) == -1) {
 276         ret = -errno;
 277         goto fail;
 278     }
 279     if (read(p_fd, hdr, HEADER_SIZE) != HEADER_SIZE) {
 280         ret = -errno;
 281         goto fail;
 282     }
 283
 284     /* write the header */
 285     if (lseek(snp_fd, 0x0, SEEK_SET) == -1) {
 286         ret = -errno;
 287         goto fail;
 288     }
 289     if (write(snp_fd, hdr, HEADER_SIZE) == -1) {
 290         ret = -errno;
 291         goto fail;
 292     }
 293
 294     memset(&header, 0, sizeof(header));
 295     memcpy(&header,&hdr[4], sizeof(header)); // skip the VMDK4_MAGIC
 296
 297     if (ftruncate(snp_fd, header.grain_offset << 9)) {
 298         ret = -errno;
 299         goto fail;
 300     }
 301     /* the descriptor offset = 0x200 */
 302     if (lseek(p_fd, 0x200, SEEK_SET) == -1) {
 303         ret = -errno;
 304         goto fail;
 305     }
 306     if (read(p_fd, p_desc, DESC_SIZE) != DESC_SIZE) {
 307         ret = -errno;
 308         goto fail;
 309     }
 310
 311     if ((p_name = strstr(p_desc,"CID")) != NULL) {
 312         p_name += sizeof("CID");
 313         sscanf(p_name,"%x",&p_cid);
 314     }
 315
 316     real_filename = filename;
 317     if ((temp_str = strrchr(real_filename, '\\')) != NULL)
 318         real_filename = temp_str + 1;
 319     if ((temp_str = strrchr(real_filename, '/')) != NULL)
 320         real_filename = temp_str + 1;
 321     if ((temp_str = strrchr(real_filename, ':')) != NULL)
 322         real_filename = temp_str + 1;
 323
 324     snprintf(s_desc, sizeof(s_desc), desc_template, p_cid, p_cid, backing_file,
 325              (uint32_t)header.capacity, real_filename);
 326
 327     /* write the descriptor */
 328     if (lseek(snp_fd, 0x200, SEEK_SET) == -1) {
 329         ret = -errno;
 330         goto fail;
 331     }
 332     if (write(snp_fd, s_desc, strlen(s_desc)) == -1) {
 333         ret = -errno;
 334         goto fail;
 335     }
 336
 337     gd_offset = header.gd_offset * SECTOR_SIZE;     // offset of GD table
 338     rgd_offset = header.rgd_offset * SECTOR_SIZE;   // offset of RGD table
 339     capacity = header.capacity * SECTOR_SIZE;       // Extent size
 340     /*
 341      * Each GDE span 32M disk, means:
 342      * 512 GTE per GT, each GTE points to grain
 343      */
 344     gt_size = (int64_t)header.num_gtes_per_gte * header.granularity * SECTOR_SIZE;
 345     if (!gt_size) {
 346         ret = -EINVAL;
 347         goto fail;
 348     }
 349     gde_entries = (uint32_t)(capacity / gt_size);  // number of gde/rgde
 350     gd_size = gde_entries * sizeof(uint32_t);
 351
 352     /* write RGD */
 353     rgd_buf = qemu_malloc(gd_size);
 354     if (lseek(p_fd, rgd_offset, SEEK_SET) == -1) {
 355         ret = -errno;
 356         goto fail_rgd;
 357     }
 358     if (read(p_fd, rgd_buf, gd_size) != gd_size) {
 359         ret = -errno;
 360         goto fail_rgd;
 361     }
 362     if (lseek(snp_fd, rgd_offset, SEEK_SET) == -1) {
 363         ret = -errno;
 364         goto fail_rgd;
 365     }
 366     if (write(snp_fd, rgd_buf, gd_size) == -1) {
 367         ret = -errno;
 368         goto fail_rgd;
 369     }
 370
 371     /* write GD */
 372     gd_buf = qemu_malloc(gd_size);
 373     if (lseek(p_fd, gd_offset, SEEK_SET) == -1) {
 374         ret = -errno;
 375         goto fail_gd;
 376     }
 377     if (read(p_fd, gd_buf, gd_size) != gd_size) {
 378         ret = -errno;
 379         goto fail_gd;
 380     }
 381     if (lseek(snp_fd, gd_offset, SEEK_SET) == -1) {
 382         ret = -errno;
 383         goto fail_gd;
 384     }
 385     if (write(snp_fd, gd_buf, gd_size) == -1) {
 386         ret = -errno;
 387         goto fail_gd;
 388     }
 389     ret = 0;
 390
 391 fail_gd:
 392     qemu_free(gd_buf);
 393 fail_rgd:
 394     qemu_free(rgd_buf);
 395 fail:
 396     close(p_fd);
 397     close(snp_fd);
 398     return ret;
 399 }
 400
 401 static int vmdk_parent_open(BlockDriverState *bs)
 402 {
 403     char *p_name;
 404     char desc[DESC_SIZE];
 405
 406     /* the descriptor offset = 0x200 */
 407     if (bdrv_pread(bs->file, 0x200, desc, DESC_SIZE) != DESC_SIZE)
 408         return -1;
 409
 410     if ((p_name = strstr(desc,"parentFileNameHint")) != NULL) {
 411         char *end_name;
 412
 413         p_name += sizeof("parentFileNameHint") + 1;
 414         if ((end_name = strchr(p_name,'\"')) == NULL)
 415             return -1;
 416         if ((end_name - p_name) > sizeof (bs->backing_file) - 1)
 417             return -1;
 418
 419         pstrcpy(bs->backing_file, end_name - p_name + 1, p_name);
 420     }
 421
 422     return 0;
 423 }
 424
 425 /* Create and append extent to the extent array. Return the added VmdkExtent
 426  * address. return NULL if allocation failed. */
 427 static VmdkExtent *vmdk_add_extent(BlockDriverState *bs,
 428                            BlockDriverState *file, bool flat, int64_t sectors,
 429                            int64_t l1_offset, int64_t l1_backup_offset,
 430                            uint32_t l1_size,
 431                            int l2_size, unsigned int cluster_sectors)
 432 {
 433     VmdkExtent *extent;
 434     BDRVVmdkState *s = bs->opaque;
 435
 436     s->extents = qemu_realloc(s->extents,
 437                               (s->num_extents + 1) * sizeof(VmdkExtent));
 438     extent = &s->extents[s->num_extents];
 439     s->num_extents++;
 440
 441     memset(extent, 0, sizeof(VmdkExtent));
 442     extent->file = file;
 443     extent->flat = flat;
 444     extent->sectors = sectors;
 445     extent->l1_table_offset = l1_offset;
 446     extent->l1_backup_table_offset = l1_backup_offset;
 447     extent->l1_size = l1_size;
 448     extent->l1_entry_sectors = l2_size * cluster_sectors;
 449     extent->l2_size = l2_size;
 450     extent->cluster_sectors = cluster_sectors;
 451
 452     if (s->num_extents > 1) {
 453         extent->end_sector = (*(extent - 1)).end_sector + extent->sectors;
 454     } else {
 455         extent->end_sector = extent->sectors;
 456     }
 457     bs->total_sectors = extent->end_sector;
 458     return extent;
 459 }
 460
 461
 462 static int vmdk_open(BlockDriverState *bs, int flags)
 463 {
 464     BDRVVmdkState *s = bs->opaque;
 465     uint32_t magic;
 466     int i;
 467     uint32_t l1_size, l1_entry_sectors;
 468     VmdkExtent *extent = NULL;
 469
 470     if (bdrv_pread(bs->file, 0, &magic, sizeof(magic)) != sizeof(magic))
 471         goto fail;
 472
 473     magic = be32_to_cpu(magic);
 474     if (magic == VMDK3_MAGIC) {
 475         VMDK3Header header;
 476         if (bdrv_pread(bs->file, sizeof(magic), &header, sizeof(header))
 477                 != sizeof(header)) {
 478             goto fail;
 479         }
 480         extent = vmdk_add_extent(bs, bs->file, false,
 481                               le32_to_cpu(header.disk_sectors),
 482                               le32_to_cpu(header.l1dir_offset) << 9, 0,
 483                               1 << 6, 1 << 9, le32_to_cpu(header.granularity));
 484     } else if (magic == VMDK4_MAGIC) {
 485         VMDK4Header header;
 486         if (bdrv_pread(bs->file, sizeof(magic), &header, sizeof(header))
 487                 != sizeof(header)) {
 488             goto fail;
 489         }
 490         l1_entry_sectors = le32_to_cpu(header.num_gtes_per_gte)
 491                             * le64_to_cpu(header.granularity);
 492         l1_size = (le64_to_cpu(header.capacity) + l1_entry_sectors - 1)
 493                     / l1_entry_sectors;
 494         extent = vmdk_add_extent(bs, bs->file, false,
 495                               le64_to_cpu(header.capacity),
 496                               le64_to_cpu(header.gd_offset) << 9,
 497                               le64_to_cpu(header.rgd_offset) << 9,
 498                               l1_size,
 499                               le32_to_cpu(header.num_gtes_per_gte),
 500                               le64_to_cpu(header.granularity));
 501         if (extent->l1_entry_sectors <= 0) {
 502             goto fail;
 503         }
 504         // try to open parent images, if exist
 505         if (vmdk_parent_open(bs) != 0)
 506             goto fail;
 507         // write the CID once after the image creation
 508         s->parent_cid = vmdk_read_cid(bs,1);
 509     } else {
 510         goto fail;
 511     }
 512
 513     /* read the L1 table */
 514     l1_size = extent->l1_size * sizeof(uint32_t);
 515     extent->l1_table = qemu_malloc(l1_size);
 516     if (bdrv_pread(bs->file,
 517             extent->l1_table_offset,
 518             extent->l1_table,
 519             l1_size)
 520         != l1_size) {
 521         goto fail;
 522     }
 523     for (i = 0; i < extent->l1_size; i++) {
 524         le32_to_cpus(&extent->l1_table[i]);
 525     }
 526
 527     if (extent->l1_backup_table_offset) {
 528         extent->l1_backup_table = qemu_malloc(l1_size);
 529         if (bdrv_pread(bs->file,
 530                     extent->l1_backup_table_offset,
 531                     extent->l1_backup_table,
 532                     l1_size)
 533                 != l1_size) {
 534             goto fail;
 535         }
 536         for (i = 0; i < extent->l1_size; i++) {
 537             le32_to_cpus(&extent->l1_backup_table[i]);
 538         }
 539     }
 540
 541     extent->l2_cache =
 542         qemu_malloc(extent->l2_size * L2_CACHE_SIZE * sizeof(uint32_t));
 543     return 0;
 544  fail:
 545     vmdk_free_extents(bs);
 546     return -1;
 547 }
 548
 549 static int get_whole_cluster(BlockDriverState *bs,
 550                 VmdkExtent *extent,
 551                 uint64_t cluster_offset,
 552                 uint64_t offset,
 553                 bool allocate)
 554 {
 555     /* 128 sectors * 512 bytes each = grain size 64KB */
 556     uint8_t  whole_grain[extent->cluster_sectors * 512];
 557
 558     /* we will be here if it's first write on non-exist grain(cluster).
 559      * try to read from parent image, if exist */
 560     if (bs->backing_hd) {
 561         int ret;
 562
 563         if (!vmdk_is_cid_valid(bs))
 564             return -1;
 565
 566         /* floor offset to cluster */
 567         offset -= offset % (extent->cluster_sectors * 512);
 568         ret = bdrv_read(bs->backing_hd, offset >> 9, whole_grain,
 569                 extent->cluster_sectors);
 570         if (ret < 0) {
 571             return -1;
 572         }
 573
 574         /* Write grain only into the active image */
 575         ret = bdrv_write(extent->file, cluster_offset, whole_grain,
 576                 extent->cluster_sectors);
 577         if (ret < 0) {
 578             return -1;
 579         }
 580     }
 581     return 0;
 582 }
 583
 584 static int vmdk_L2update(VmdkExtent *extent, VmdkMetaData *m_data)
 585 {
 586     /* update L2 table */
 587     if (bdrv_pwrite_sync(
 588                 extent->file,
 589                 ((int64_t)m_data->l2_offset * 512)
 590                     + (m_data->l2_index * sizeof(m_data->offset)),
 591                 &(m_data->offset),
 592                 sizeof(m_data->offset)
 593             ) < 0) {
 594         return -1;
 595     }
 596     /* update backup L2 table */
 597     if (extent->l1_backup_table_offset != 0) {
 598         m_data->l2_offset = extent->l1_backup_table[m_data->l1_index];
 599         if (bdrv_pwrite_sync(
 600                     extent->file,
 601                     ((int64_t)m_data->l2_offset * 512)
 602                         + (m_data->l2_index * sizeof(m_data->offset)),
 603                     &(m_data->offset), sizeof(m_data->offset)
 604                 ) < 0) {
 605             return -1;
 606         }
 607     }
 608
 609     return 0;
 610 }
 611
 612 static uint64_t get_cluster_offset(BlockDriverState *bs,
 613                                     VmdkExtent *extent,
 614                                     VmdkMetaData *m_data,
 615                                     uint64_t offset, int allocate)
 616 {
 617     unsigned int l1_index, l2_offset, l2_index;
 618     int min_index, i, j;
 619     uint32_t min_count, *l2_table, tmp = 0;
 620     uint64_t cluster_offset;
 621
 622     if (m_data)
 623         m_data->valid = 0;
 624
 625     l1_index = (offset >> 9) / extent->l1_entry_sectors;
 626     if (l1_index >= extent->l1_size) {
 627         return 0;
 628     }
 629     l2_offset = extent->l1_table[l1_index];
 630     if (!l2_offset) {
 631         return 0;
 632     }
 633     for(i = 0; i < L2_CACHE_SIZE; i++) {
 634         if (l2_offset == extent->l2_cache_offsets[i]) {
 635             /* increment the hit count */
 636             if (++extent->l2_cache_counts[i] == 0xffffffff) {
 637                 for(j = 0; j < L2_CACHE_SIZE; j++) {
 638                     extent->l2_cache_counts[j] >>= 1;
 639                 }
 640             }
 641             l2_table = extent->l2_cache + (i * extent->l2_size);
 642             goto found;
 643         }
 644     }
 645     /* not found: load a new entry in the least used one */
 646     min_index = 0;
 647     min_count = 0xffffffff;
 648     for(i = 0; i < L2_CACHE_SIZE; i++) {
 649         if (extent->l2_cache_counts[i] < min_count) {
 650             min_count = extent->l2_cache_counts[i];
 651             min_index = i;
 652         }
 653     }
 654     l2_table = extent->l2_cache + (min_index * extent->l2_size);
 655     if (bdrv_pread(
 656                 extent->file,
 657                 (int64_t)l2_offset * 512,
 658                 l2_table,
 659                 extent->l2_size * sizeof(uint32_t)
 660             ) != extent->l2_size * sizeof(uint32_t)) {
 661         return 0;
 662     }
 663
 664     extent->l2_cache_offsets[min_index] = l2_offset;
 665     extent->l2_cache_counts[min_index] = 1;
 666  found:
 667     l2_index = ((offset >> 9) / extent->cluster_sectors) % extent->l2_size;
 668     cluster_offset = le32_to_cpu(l2_table[l2_index]);
 669
 670     if (!cluster_offset) {
 671         if (!allocate)
 672             return 0;
 673
 674         // Avoid the L2 tables update for the images that have snapshots.
 675         cluster_offset = bdrv_getlength(extent->file);
 676         bdrv_truncate(
 677             extent->file,
 678             cluster_offset + (extent->cluster_sectors << 9)
 679         );
 680
 681         cluster_offset >>= 9;
 682         tmp = cpu_to_le32(cluster_offset);
 683         l2_table[l2_index] = tmp;
 684
 685         /* First of all we write grain itself, to avoid race condition
 686          * that may to corrupt the image.
 687          * This problem may occur because of insufficient space on host disk
 688          * or inappropriate VM shutdown.
 689          */
 690         if (get_whole_cluster(
 691                 bs, extent, cluster_offset, offset, allocate) == -1)
 692             return 0;
 693
 694         if (m_data) {
 695             m_data->offset = tmp;
 696             m_data->l1_index = l1_index;
 697             m_data->l2_index = l2_index;
 698             m_data->l2_offset = l2_offset;
 699             m_data->valid = 1;
 700         }
 701     }
 702     cluster_offset <<= 9;
 703     return cluster_offset;
 704 }
 705
 706 static VmdkExtent *find_extent(BDRVVmdkState *s,
 707                                 int64_t sector_num, VmdkExtent *start_hint)
 708 {
 709     VmdkExtent *extent = start_hint;
 710
 711     if (!extent) {
 712         extent = &s->extents[0];
 713     }
 714     while (extent < &s->extents[s->num_extents]) {
 715         if (sector_num < extent->end_sector) {
 716             return extent;
 717         }
 718         extent++;
 719     }
 720     return NULL;
 721 }
 722
 723 static int vmdk_is_allocated(BlockDriverState *bs, int64_t sector_num,
 724                              int nb_sectors, int *pnum)
 725 {
 726     BDRVVmdkState *s = bs->opaque;
 727
 728     int64_t index_in_cluster, n, ret;
 729     uint64_t offset;
 730     VmdkExtent *extent;
 731
 732     extent = find_extent(s, sector_num, NULL);
 733     if (!extent) {
 734         return 0;
 735     }
 736     if (extent->flat) {
 737         n = extent->end_sector - sector_num;
 738         ret = 1;
 739     } else {
 740         offset = get_cluster_offset(bs, extent, NULL, sector_num * 512, 0);
 741         index_in_cluster = sector_num % extent->cluster_sectors;
 742         n = extent->cluster_sectors - index_in_cluster;
 743         ret = offset ? 1 : 0;
 744     }
 745     if (n > nb_sectors)
 746         n = nb_sectors;
 747     *pnum = n;
 748     return ret;
 749 }
 750
 751 static int vmdk_read(BlockDriverState *bs, int64_t sector_num,
 752                     uint8_t *buf, int nb_sectors)
 753 {
 754     BDRVVmdkState *s = bs->opaque;
 755     int ret;
 756     uint64_t n, index_in_cluster;
 757     VmdkExtent *extent = NULL;
 758     uint64_t cluster_offset;
 759
 760     while (nb_sectors > 0) {
 761         extent = find_extent(s, sector_num, extent);
 762         if (!extent) {
 763             return -EIO;
 764         }
 765         cluster_offset = get_cluster_offset(
 766                             bs, extent, NULL, sector_num << 9, 0);
 767         index_in_cluster = sector_num % extent->cluster_sectors;
 768         n = extent->cluster_sectors - index_in_cluster;
 769         if (n > nb_sectors)
 770             n = nb_sectors;
 771         if (!cluster_offset) {
 772             // try to read from parent image, if exist
 773             if (bs->backing_hd) {
 774                 if (!vmdk_is_cid_valid(bs))
 775                     return -1;
 776                 ret = bdrv_read(bs->backing_hd, sector_num, buf, n);
 777                 if (ret < 0)
 778                     return -1;
 779             } else {
 780                 memset(buf, 0, 512 * n);
 781             }
 782         } else {
 783             if(bdrv_pread(bs->file, cluster_offset + index_in_cluster * 512, buf, n * 512) != n * 512)
 784                 return -1;
 785         }
 786         nb_sectors -= n;
 787         sector_num += n;
 788         buf += n * 512;
 789     }
 790     return 0;
 791 }
 792
 793 static int vmdk_write(BlockDriverState *bs, int64_t sector_num,
 794                      const uint8_t *buf, int nb_sectors)
 795 {
 796     BDRVVmdkState *s = bs->opaque;
 797     VmdkExtent *extent = NULL;
 798     int n;
 799     int64_t index_in_cluster;
 800     uint64_t cluster_offset;
 801     static int cid_update = 0;
 802     VmdkMetaData m_data;
 803
 804     if (sector_num > bs->total_sectors) {
 805         fprintf(stderr,
 806                 "(VMDK) Wrong offset: sector_num=0x%" PRIx64
 807                 " total_sectors=0x%" PRIx64 "\n",
 808                 sector_num, bs->total_sectors);
 809         return -1;
 810     }
 811
 812     while (nb_sectors > 0) {
 813         extent = find_extent(s, sector_num, extent);
 814         if (!extent) {
 815             return -EIO;
 816         }
 817         cluster_offset = get_cluster_offset(
 818                                 bs,
 819                                 extent,
 820                                 &m_data,
 821                                 sector_num << 9, 1);
 822         if (!cluster_offset) {
 823             return -1;
 824         }
 825         index_in_cluster = sector_num % extent->cluster_sectors;
 826         n = extent->cluster_sectors - index_in_cluster;
 827         if (n > nb_sectors) {
 828             n = nb_sectors;
 829         }
 830
 831         if (bdrv_pwrite(bs->file,
 832                         cluster_offset + index_in_cluster * 512,
 833                         buf, n * 512)
 834                 != n * 512) {
 835             return -1;
 836         }
 837         if (m_data.valid) {
 838             /* update L2 tables */
 839             if (vmdk_L2update(extent, &m_data) == -1) {
 840                 return -1;
 841             }
 842         }
 843         nb_sectors -= n;
 844         sector_num += n;
 845         buf += n * 512;
 846
 847         // update CID on the first write every time the virtual disk is opened
 848         if (!cid_update) {
 849             vmdk_write_cid(bs, time(NULL));
 850             cid_update++;
 851         }
 852     }
 853     return 0;
 854 }
 855
 856 static int vmdk_create(const char *filename, QEMUOptionParameter *options)
 857 {
 858     int fd, i;
 859     VMDK4Header header;
 860     uint32_t tmp, magic, grains, gd_size, gt_size, gt_count;
 861     static const char desc_template[] =
 862         "# Disk DescriptorFile\n"
 863         "version=1\n"
 864         "CID=%x\n"
 865         "parentCID=ffffffff\n"
 866         "createType=\"monolithicSparse\"\n"
 867         "\n"
 868         "# Extent description\n"
 869         "RW %" PRId64 " SPARSE \"%s\"\n"
 870         "\n"
 871         "# The Disk Data Base \n"
 872         "#DDB\n"
 873         "\n"
 874         "ddb.virtualHWVersion = \"%d\"\n"
 875         "ddb.geometry.cylinders = \"%" PRId64 "\"\n"
 876         "ddb.geometry.heads = \"16\"\n"
 877         "ddb.geometry.sectors = \"63\"\n"
 878         "ddb.adapterType = \"ide\"\n";
 879     char desc[1024];
 880     const char *real_filename, *temp_str;
 881     int64_t total_size = 0;
 882     const char *backing_file = NULL;
 883     int flags = 0;
 884     int ret;
 885
 886     // Read out options
 887     while (options && options->name) {
 888         if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
 889             total_size = options->value.n / 512;
 890         } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) {
 891             backing_file = options->value.s;
 892         } else if (!strcmp(options->name, BLOCK_OPT_COMPAT6)) {
 893             flags |= options->value.n ? BLOCK_FLAG_COMPAT6: 0;
 894         }
 895         options++;
 896     }
 897
 898     /* XXX: add support for backing file */
 899     if (backing_file) {
 900         return vmdk_snapshot_create(filename, backing_file);
 901     }
 902
 903     fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY | O_LARGEFILE,
 904               0644);
 905     if (fd < 0)
 906         return -errno;
 907     magic = cpu_to_be32(VMDK4_MAGIC);
 908     memset(&header, 0, sizeof(header));
 909     header.version = 1;
 910     header.flags = 3; /* ?? */
 911     header.capacity = total_size;
 912     header.granularity = 128;
 913     header.num_gtes_per_gte = 512;
 914
 915     grains = (total_size + header.granularity - 1) / header.granularity;
 916     gt_size = ((header.num_gtes_per_gte * sizeof(uint32_t)) + 511) >> 9;
 917     gt_count = (grains + header.num_gtes_per_gte - 1) / header.num_gtes_per_gte;
 918     gd_size = (gt_count * sizeof(uint32_t) + 511) >> 9;
 919
 920     header.desc_offset = 1;
 921     header.desc_size = 20;
 922     header.rgd_offset = header.desc_offset + header.desc_size;
 923     header.gd_offset = header.rgd_offset + gd_size + (gt_size * gt_count);
 924     header.grain_offset =
 925        ((header.gd_offset + gd_size + (gt_size * gt_count) +
 926          header.granularity - 1) / header.granularity) *
 927         header.granularity;
 928
 929     /* swap endianness for all header fields */
 930     header.version = cpu_to_le32(header.version);
 931     header.flags = cpu_to_le32(header.flags);
 932     header.capacity = cpu_to_le64(header.capacity);
 933     header.granularity = cpu_to_le64(header.granularity);
 934     header.num_gtes_per_gte = cpu_to_le32(header.num_gtes_per_gte);
 935     header.desc_offset = cpu_to_le64(header.desc_offset);
 936     header.desc_size = cpu_to_le64(header.desc_size);
 937     header.rgd_offset = cpu_to_le64(header.rgd_offset);
 938     header.gd_offset = cpu_to_le64(header.gd_offset);
 939     header.grain_offset = cpu_to_le64(header.grain_offset);
 940
 941     header.check_bytes[0] = 0xa;
 942     header.check_bytes[1] = 0x20;
 943     header.check_bytes[2] = 0xd;
 944     header.check_bytes[3] = 0xa;
 945
 946     /* write all the data */
 947     ret = qemu_write_full(fd, &magic, sizeof(magic));
 948     if (ret != sizeof(magic)) {
 949         ret = -errno;
 950         goto exit;
 951     }
 952     ret = qemu_write_full(fd, &header, sizeof(header));
 953     if (ret != sizeof(header)) {
 954         ret = -errno;
 955         goto exit;
 956     }
 957
 958     ret = ftruncate(fd, le64_to_cpu(header.grain_offset) << 9);
 959     if (ret < 0) {
 960         ret = -errno;
 961         goto exit;
 962     }
 963
 964     /* write grain directory */
 965     lseek(fd, le64_to_cpu(header.rgd_offset) << 9, SEEK_SET);
 966     for (i = 0, tmp = le64_to_cpu(header.rgd_offset) + gd_size;
 967          i < gt_count; i++, tmp += gt_size) {
 968         ret = qemu_write_full(fd, &tmp, sizeof(tmp));
 969         if (ret != sizeof(tmp)) {
 970             ret = -errno;
 971             goto exit;
 972         }
 973     }
 974
 975     /* write backup grain directory */
 976     lseek(fd, le64_to_cpu(header.gd_offset) << 9, SEEK_SET);
 977     for (i = 0, tmp = le64_to_cpu(header.gd_offset) + gd_size;
 978          i < gt_count; i++, tmp += gt_size) {
 979         ret = qemu_write_full(fd, &tmp, sizeof(tmp));
 980         if (ret != sizeof(tmp)) {
 981             ret = -errno;
 982             goto exit;
 983         }
 984     }
 985
 986     /* compose the descriptor */
 987     real_filename = filename;
 988     if ((temp_str = strrchr(real_filename, '\\')) != NULL)
 989         real_filename = temp_str + 1;
 990     if ((temp_str = strrchr(real_filename, '/')) != NULL)
 991         real_filename = temp_str + 1;
 992     if ((temp_str = strrchr(real_filename, ':')) != NULL)
 993         real_filename = temp_str + 1;
 994     snprintf(desc, sizeof(desc), desc_template, (unsigned int)time(NULL),
 995              total_size, real_filename,
 996              (flags & BLOCK_FLAG_COMPAT6 ? 6 : 4),
 997              total_size / (int64_t)(63 * 16));
 998
 999     /* write the descriptor */
1000     lseek(fd, le64_to_cpu(header.desc_offset) << 9, SEEK_SET);
1001     ret = qemu_write_full(fd, desc, strlen(desc));
1002     if (ret != strlen(desc)) {
1003         ret = -errno;
1004         goto exit;
1005     }
1006
1007     ret = 0;
1008 exit:
1009     close(fd);
1010     return ret;
1011 }
1012
1013 static void vmdk_close(BlockDriverState *bs)
1014 {
1015     vmdk_free_extents(bs);
1016 }
1017
1018 static int vmdk_flush(BlockDriverState *bs)
1019 {
1020     return bdrv_flush(bs->file);
1021 }
1022
1023
1024 static QEMUOptionParameter vmdk_create_options[] = {
1025     {
1026         .name = BLOCK_OPT_SIZE,
1027         .type = OPT_SIZE,
1028         .help = "Virtual disk size"
1029     },
1030     {
1031         .name = BLOCK_OPT_BACKING_FILE,
1032         .type = OPT_STRING,
1033         .help = "File name of a base image"
1034     },
1035     {
1036         .name = BLOCK_OPT_COMPAT6,
1037         .type = OPT_FLAG,
1038         .help = "VMDK version 6 image"
1039     },
1040     { NULL }
1041 };
1042
1043 static BlockDriver bdrv_vmdk = {
1044     .format_name        = "vmdk",
1045     .instance_size      = sizeof(BDRVVmdkState),
1046     .bdrv_probe         = vmdk_probe,
1047     .bdrv_open      = vmdk_open,
1048     .bdrv_read          = vmdk_read,
1049     .bdrv_write         = vmdk_write,
1050     .bdrv_close         = vmdk_close,
1051     .bdrv_create        = vmdk_create,
1052     .bdrv_flush         = vmdk_flush,
1053     .bdrv_is_allocated  = vmdk_is_allocated,
1054
1055     .create_options = vmdk_create_options,
1056 };
1057
1058 static void bdrv_vmdk_init(void)
1059 {
1060     bdrv_register(&bdrv_vmdk);
1061 }
1062
1063 block_init(bdrv_vmdk_init);