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/block_int.h"
  28 #include "qemu/module.h"
  29 #include "migration/migration.h"
  30 #include <zlib.h>
  31
  32 #define VMDK3_MAGIC (('C' << 24) | ('O' << 16) | ('W' << 8) | 'D')
  33 #define VMDK4_MAGIC (('K' << 24) | ('D' << 16) | ('M' << 8) | 'V')
  34 #define VMDK4_COMPRESSION_DEFLATE 1
  35 #define VMDK4_FLAG_NL_DETECT (1 << 0)
  36 #define VMDK4_FLAG_RGD (1 << 1)
  37 /* Zeroed-grain enable bit */
  38 #define VMDK4_FLAG_ZERO_GRAIN   (1 << 2)
  39 #define VMDK4_FLAG_COMPRESS (1 << 16)
  40 #define VMDK4_FLAG_MARKER (1 << 17)
  41 #define VMDK4_GD_AT_END 0xffffffffffffffffULL
  42
  43 #define VMDK_GTE_ZEROED 0x1
  44
  45 /* VMDK internal error codes */
  46 #define VMDK_OK      0
  47 #define VMDK_ERROR   (-1)
  48 /* Cluster not allocated */
  49 #define VMDK_UNALLOC (-2)
  50 #define VMDK_ZEROED  (-3)
  51
  52 #define BLOCK_OPT_ZEROED_GRAIN "zeroed_grain"
  53
  54 typedef struct {
  55     uint32_t version;
  56     uint32_t flags;
  57     uint32_t disk_sectors;
  58     uint32_t granularity;
  59     uint32_t l1dir_offset;
  60     uint32_t l1dir_size;
  61     uint32_t file_sectors;
  62     uint32_t cylinders;
  63     uint32_t heads;
  64     uint32_t sectors_per_track;
  65 } QEMU_PACKED VMDK3Header;
  66
  67 typedef struct {
  68     uint32_t version;
  69     uint32_t flags;
  70     uint64_t capacity;
  71     uint64_t granularity;
  72     uint64_t desc_offset;
  73     uint64_t desc_size;
  74     /* Number of GrainTableEntries per GrainTable */
  75     uint32_t num_gtes_per_gt;
  76     uint64_t rgd_offset;
  77     uint64_t gd_offset;
  78     uint64_t grain_offset;
  79     char filler[1];
  80     char check_bytes[4];
  81     uint16_t compressAlgorithm;
  82 } QEMU_PACKED VMDK4Header;
  83
  84 #define L2_CACHE_SIZE 16
  85
  86 typedef struct VmdkExtent {
  87     BlockDriverState *file;
  88     bool flat;
  89     bool compressed;
  90     bool has_marker;
  91     bool has_zero_grain;
  92     int version;
  93     int64_t sectors;
  94     int64_t end_sector;
  95     int64_t flat_start_offset;
  96     int64_t l1_table_offset;
  97     int64_t l1_backup_table_offset;
  98     uint32_t *l1_table;
  99     uint32_t *l1_backup_table;
 100     unsigned int l1_size;
 101     uint32_t l1_entry_sectors;
 102
 103     unsigned int l2_size;
 104     uint32_t *l2_cache;
 105     uint32_t l2_cache_offsets[L2_CACHE_SIZE];
 106     uint32_t l2_cache_counts[L2_CACHE_SIZE];
 107
 108     int64_t cluster_sectors;
 109     int64_t next_cluster_sector;
 110     char *type;
 111 } VmdkExtent;
 112
 113 typedef struct BDRVVmdkState {
 114     CoMutex lock;
 115     uint64_t desc_offset;
 116     bool cid_updated;
 117     bool cid_checked;
 118     uint32_t cid;
 119     uint32_t parent_cid;
 120     int num_extents;
 121     /* Extent array with num_extents entries, ascend ordered by address */
 122     VmdkExtent *extents;
 123     Error *migration_blocker;
 124     char *create_type;
 125 } BDRVVmdkState;
 126
 127 typedef struct VmdkMetaData {
 128     unsigned int l1_index;
 129     unsigned int l2_index;
 130     unsigned int l2_offset;
 131     int valid;
 132     uint32_t *l2_cache_entry;
 133 } VmdkMetaData;
 134
 135 typedef struct VmdkGrainMarker {
 136     uint64_t lba;
 137     uint32_t size;
 138     uint8_t  data[0];
 139 } QEMU_PACKED VmdkGrainMarker;
 140
 141 enum {
 142     MARKER_END_OF_STREAM    = 0,
 143     MARKER_GRAIN_TABLE      = 1,
 144     MARKER_GRAIN_DIRECTORY  = 2,
 145     MARKER_FOOTER           = 3,
 146 };
 147
 148 static int vmdk_probe(const uint8_t *buf, int buf_size, const char *filename)
 149 {
 150     uint32_t magic;
 151
 152     if (buf_size < 4) {
 153         return 0;
 154     }
 155     magic = be32_to_cpu(*(uint32_t *)buf);
 156     if (magic == VMDK3_MAGIC ||
 157         magic == VMDK4_MAGIC) {
 158         return 100;
 159     } else {
 160         const char *p = (const char *)buf;
 161         const char *end = p + buf_size;
 162         while (p < end) {
 163             if (*p == '#') {
 164                 /* skip comment line */
 165                 while (p < end && *p != '\n') {
 166                     p++;
 167                 }
 168                 p++;
 169                 continue;
 170             }
 171             if (*p == ' ') {
 172                 while (p < end && *p == ' ') {
 173                     p++;
 174                 }
 175                 /* skip '\r' if windows line endings used. */
 176                 if (p < end && *p == '\r') {
 177                     p++;
 178                 }
 179                 /* only accept blank lines before 'version=' line */
 180                 if (p == end || *p != '\n') {
 181                     return 0;
 182                 }
 183                 p++;
 184                 continue;
 185             }
 186             if (end - p >= strlen("version=X\n")) {
 187                 if (strncmp("version=1\n", p, strlen("version=1\n")) == 0 ||
 188                     strncmp("version=2\n", p, strlen("version=2\n")) == 0) {
 189                     return 100;
 190                 }
 191             }
 192             if (end - p >= strlen("version=X\r\n")) {
 193                 if (strncmp("version=1\r\n", p, strlen("version=1\r\n")) == 0 ||
 194                     strncmp("version=2\r\n", p, strlen("version=2\r\n")) == 0) {
 195                     return 100;
 196                 }
 197             }
 198             return 0;
 199         }
 200         return 0;
 201     }
 202 }
 203
 204 #define SECTOR_SIZE 512
 205 #define DESC_SIZE (20 * SECTOR_SIZE)    /* 20 sectors of 512 bytes each */
 206 #define BUF_SIZE 4096
 207 #define HEADER_SIZE 512                 /* first sector of 512 bytes */
 208
 209 static void vmdk_free_extents(BlockDriverState *bs)
 210 {
 211     int i;
 212     BDRVVmdkState *s = bs->opaque;
 213     VmdkExtent *e;
 214
 215     for (i = 0; i < s->num_extents; i++) {
 216         e = &s->extents[i];
 217         g_free(e->l1_table);
 218         g_free(e->l2_cache);
 219         g_free(e->l1_backup_table);
 220         g_free(e->type);
 221         if (e->file != bs->file) {
 222             bdrv_unref(e->file);
 223         }
 224     }
 225     g_free(s->extents);
 226 }
 227
 228 static void vmdk_free_last_extent(BlockDriverState *bs)
 229 {
 230     BDRVVmdkState *s = bs->opaque;
 231
 232     if (s->num_extents == 0) {
 233         return;
 234     }
 235     s->num_extents--;
 236     s->extents = g_renew(VmdkExtent, s->extents, s->num_extents);
 237 }
 238
 239 static uint32_t vmdk_read_cid(BlockDriverState *bs, int parent)
 240 {
 241     char desc[DESC_SIZE];
 242     uint32_t cid = 0xffffffff;
 243     const char *p_name, *cid_str;
 244     size_t cid_str_size;
 245     BDRVVmdkState *s = bs->opaque;
 246     int ret;
 247
 248     ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE);
 249     if (ret < 0) {
 250         return 0;
 251     }
 252
 253     if (parent) {
 254         cid_str = "parentCID";
 255         cid_str_size = sizeof("parentCID");
 256     } else {
 257         cid_str = "CID";
 258         cid_str_size = sizeof("CID");
 259     }
 260
 261     desc[DESC_SIZE - 1] = '\0';
 262     p_name = strstr(desc, cid_str);
 263     if (p_name != NULL) {
 264         p_name += cid_str_size;
 265         sscanf(p_name, "%" SCNx32, &cid);
 266     }
 267
 268     return cid;
 269 }
 270
 271 static int vmdk_write_cid(BlockDriverState *bs, uint32_t cid)
 272 {
 273     char desc[DESC_SIZE], tmp_desc[DESC_SIZE];
 274     char *p_name, *tmp_str;
 275     BDRVVmdkState *s = bs->opaque;
 276     int ret;
 277
 278     ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE);
 279     if (ret < 0) {
 280         return ret;
 281     }
 282
 283     desc[DESC_SIZE - 1] = '\0';
 284     tmp_str = strstr(desc, "parentCID");
 285     if (tmp_str == NULL) {
 286         return -EINVAL;
 287     }
 288
 289     pstrcpy(tmp_desc, sizeof(tmp_desc), tmp_str);
 290     p_name = strstr(desc, "CID");
 291     if (p_name != NULL) {
 292         p_name += sizeof("CID");
 293         snprintf(p_name, sizeof(desc) - (p_name - desc), "%" PRIx32 "\n", cid);
 294         pstrcat(desc, sizeof(desc), tmp_desc);
 295     }
 296
 297     ret = bdrv_pwrite_sync(bs->file, s->desc_offset, desc, DESC_SIZE);
 298     if (ret < 0) {
 299         return ret;
 300     }
 301
 302     return 0;
 303 }
 304
 305 static int vmdk_is_cid_valid(BlockDriverState *bs)
 306 {
 307     BDRVVmdkState *s = bs->opaque;
 308     BlockDriverState *p_bs = bs->backing_hd;
 309     uint32_t cur_pcid;
 310
 311     if (!s->cid_checked && p_bs) {
 312         cur_pcid = vmdk_read_cid(p_bs, 0);
 313         if (s->parent_cid != cur_pcid) {
 314             /* CID not valid */
 315             return 0;
 316         }
 317     }
 318     s->cid_checked = true;
 319     /* CID valid */
 320     return 1;
 321 }
 322
 323 /* Queue extents, if any, for reopen() */
 324 static int vmdk_reopen_prepare(BDRVReopenState *state,
 325                                BlockReopenQueue *queue, Error **errp)
 326 {
 327     BDRVVmdkState *s;
 328     int ret = -1;
 329     int i;
 330     VmdkExtent *e;
 331
 332     assert(state != NULL);
 333     assert(state->bs != NULL);
 334
 335     if (queue == NULL) {
 336         error_setg(errp, "No reopen queue for VMDK extents");
 337         goto exit;
 338     }
 339
 340     s = state->bs->opaque;
 341
 342     assert(s != NULL);
 343
 344     for (i = 0; i < s->num_extents; i++) {
 345         e = &s->extents[i];
 346         if (e->file != state->bs->file) {
 347             bdrv_reopen_queue(queue, e->file, state->flags);
 348         }
 349     }
 350     ret = 0;
 351
 352 exit:
 353     return ret;
 354 }
 355
 356 static int vmdk_parent_open(BlockDriverState *bs)
 357 {
 358     char *p_name;
 359     char desc[DESC_SIZE + 1];
 360     BDRVVmdkState *s = bs->opaque;
 361     int ret;
 362
 363     desc[DESC_SIZE] = '\0';
 364     ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE);
 365     if (ret < 0) {
 366         return ret;
 367     }
 368
 369     p_name = strstr(desc, "parentFileNameHint");
 370     if (p_name != NULL) {
 371         char *end_name;
 372
 373         p_name += sizeof("parentFileNameHint") + 1;
 374         end_name = strchr(p_name, '\"');
 375         if (end_name == NULL) {
 376             return -EINVAL;
 377         }
 378         if ((end_name - p_name) > sizeof(bs->backing_file) - 1) {
 379             return -EINVAL;
 380         }
 381
 382         pstrcpy(bs->backing_file, end_name - p_name + 1, p_name);
 383     }
 384
 385     return 0;
 386 }
 387
 388 /* Create and append extent to the extent array. Return the added VmdkExtent
 389  * address. return NULL if allocation failed. */
 390 static int vmdk_add_extent(BlockDriverState *bs,
 391                            BlockDriverState *file, bool flat, int64_t sectors,
 392                            int64_t l1_offset, int64_t l1_backup_offset,
 393                            uint32_t l1_size,
 394                            int l2_size, uint64_t cluster_sectors,
 395                            VmdkExtent **new_extent,
 396                            Error **errp)
 397 {
 398     VmdkExtent *extent;
 399     BDRVVmdkState *s = bs->opaque;
 400     int64_t nb_sectors;
 401
 402     if (cluster_sectors > 0x200000) {
 403         /* 0x200000 * 512Bytes = 1GB for one cluster is unrealistic */
 404         error_setg(errp, "Invalid granularity, image may be corrupt");
 405         return -EFBIG;
 406     }
 407     if (l1_size > 512 * 1024 * 1024) {
 408         /* Although with big capacity and small l1_entry_sectors, we can get a
 409          * big l1_size, we don't want unbounded value to allocate the table.
 410          * Limit it to 512M, which is 16PB for default cluster and L2 table
 411          * size */
 412         error_setg(errp, "L1 size too big");
 413         return -EFBIG;
 414     }
 415
 416     nb_sectors = bdrv_nb_sectors(file);
 417     if (nb_sectors < 0) {
 418         return nb_sectors;
 419     }
 420
 421     s->extents = g_renew(VmdkExtent, s->extents, s->num_extents + 1);
 422     extent = &s->extents[s->num_extents];
 423     s->num_extents++;
 424
 425     memset(extent, 0, sizeof(VmdkExtent));
 426     extent->file = file;
 427     extent->flat = flat;
 428     extent->sectors = sectors;
 429     extent->l1_table_offset = l1_offset;
 430     extent->l1_backup_table_offset = l1_backup_offset;
 431     extent->l1_size = l1_size;
 432     extent->l1_entry_sectors = l2_size * cluster_sectors;
 433     extent->l2_size = l2_size;
 434     extent->cluster_sectors = flat ? sectors : cluster_sectors;
 435     extent->next_cluster_sector = ROUND_UP(nb_sectors, cluster_sectors);
 436
 437     if (s->num_extents > 1) {
 438         extent->end_sector = (*(extent - 1)).end_sector + extent->sectors;
 439     } else {
 440         extent->end_sector = extent->sectors;
 441     }
 442     bs->total_sectors = extent->end_sector;
 443     if (new_extent) {
 444         *new_extent = extent;
 445     }
 446     return 0;
 447 }
 448
 449 static int vmdk_init_tables(BlockDriverState *bs, VmdkExtent *extent,
 450                             Error **errp)
 451 {
 452     int ret;
 453     int l1_size, i;
 454
 455     /* read the L1 table */
 456     l1_size = extent->l1_size * sizeof(uint32_t);
 457     extent->l1_table = g_try_malloc(l1_size);
 458     if (l1_size && extent->l1_table == NULL) {
 459         return -ENOMEM;
 460     }
 461
 462     ret = bdrv_pread(extent->file,
 463                      extent->l1_table_offset,
 464                      extent->l1_table,
 465                      l1_size);
 466     if (ret < 0) {
 467         error_setg_errno(errp, -ret,
 468                          "Could not read l1 table from extent '%s'",
 469                          extent->file->filename);
 470         goto fail_l1;
 471     }
 472     for (i = 0; i < extent->l1_size; i++) {
 473         le32_to_cpus(&extent->l1_table[i]);
 474     }
 475
 476     if (extent->l1_backup_table_offset) {
 477         extent->l1_backup_table = g_try_malloc(l1_size);
 478         if (l1_size && extent->l1_backup_table == NULL) {
 479             ret = -ENOMEM;
 480             goto fail_l1;
 481         }
 482         ret = bdrv_pread(extent->file,
 483                          extent->l1_backup_table_offset,
 484                          extent->l1_backup_table,
 485                          l1_size);
 486         if (ret < 0) {
 487             error_setg_errno(errp, -ret,
 488                              "Could not read l1 backup table from extent '%s'",
 489                              extent->file->filename);
 490             goto fail_l1b;
 491         }
 492         for (i = 0; i < extent->l1_size; i++) {
 493             le32_to_cpus(&extent->l1_backup_table[i]);
 494         }
 495     }
 496
 497     extent->l2_cache =
 498         g_new(uint32_t, extent->l2_size * L2_CACHE_SIZE);
 499     return 0;
 500  fail_l1b:
 501     g_free(extent->l1_backup_table);
 502  fail_l1:
 503     g_free(extent->l1_table);
 504     return ret;
 505 }
 506
 507 static int vmdk_open_vmfs_sparse(BlockDriverState *bs,
 508                                  BlockDriverState *file,
 509                                  int flags, Error **errp)
 510 {
 511     int ret;
 512     uint32_t magic;
 513     VMDK3Header header;
 514     VmdkExtent *extent;
 515
 516     ret = bdrv_pread(file, sizeof(magic), &header, sizeof(header));
 517     if (ret < 0) {
 518         error_setg_errno(errp, -ret,
 519                          "Could not read header from file '%s'",
 520                          file->filename);
 521         return ret;
 522     }
 523     ret = vmdk_add_extent(bs, file, false,
 524                           le32_to_cpu(header.disk_sectors),
 525                           le32_to_cpu(header.l1dir_offset) << 9,
 526                           0,
 527                           le32_to_cpu(header.l1dir_size),
 528                           4096,
 529                           le32_to_cpu(header.granularity),
 530                           &extent,
 531                           errp);
 532     if (ret < 0) {
 533         return ret;
 534     }
 535     ret = vmdk_init_tables(bs, extent, errp);
 536     if (ret) {
 537         /* free extent allocated by vmdk_add_extent */
 538         vmdk_free_last_extent(bs);
 539     }
 540     return ret;
 541 }
 542
 543 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf,
 544                                Error **errp);
 545
 546 static char *vmdk_read_desc(BlockDriverState *file, uint64_t desc_offset,
 547                             Error **errp)
 548 {
 549     int64_t size;
 550     char *buf;
 551     int ret;
 552
 553     size = bdrv_getlength(file);
 554     if (size < 0) {
 555         error_setg_errno(errp, -size, "Could not access file");
 556         return NULL;
 557     }
 558
 559     size = MIN(size, 1 << 20);  /* avoid unbounded allocation */
 560     buf = g_malloc0(size + 1);
 561
 562     ret = bdrv_pread(file, desc_offset, buf, size);
 563     if (ret < 0) {
 564         error_setg_errno(errp, -ret, "Could not read from file");
 565         g_free(buf);
 566         return NULL;
 567     }
 568
 569     return buf;
 570 }
 571
 572 static int vmdk_open_vmdk4(BlockDriverState *bs,
 573                            BlockDriverState *file,
 574                            int flags, Error **errp)
 575 {
 576     int ret;
 577     uint32_t magic;
 578     uint32_t l1_size, l1_entry_sectors;
 579     VMDK4Header header;
 580     VmdkExtent *extent;
 581     BDRVVmdkState *s = bs->opaque;
 582     int64_t l1_backup_offset = 0;
 583
 584     ret = bdrv_pread(file, sizeof(magic), &header, sizeof(header));
 585     if (ret < 0) {
 586         error_setg_errno(errp, -ret,
 587                          "Could not read header from file '%s'",
 588                          file->filename);
 589         return -EINVAL;
 590     }
 591     if (header.capacity == 0) {
 592         uint64_t desc_offset = le64_to_cpu(header.desc_offset);
 593         if (desc_offset) {
 594             char *buf = vmdk_read_desc(file, desc_offset << 9, errp);
 595             if (!buf) {
 596                 return -EINVAL;
 597             }
 598             ret = vmdk_open_desc_file(bs, flags, buf, errp);
 599             g_free(buf);
 600             return ret;
 601         }
 602     }
 603
 604     if (!s->create_type) {
 605         s->create_type = g_strdup("monolithicSparse");
 606     }
 607
 608     if (le64_to_cpu(header.gd_offset) == VMDK4_GD_AT_END) {
 609         /*
 610          * The footer takes precedence over the header, so read it in. The
 611          * footer starts at offset -1024 from the end: One sector for the
 612          * footer, and another one for the end-of-stream marker.
 613          */
 614         struct {
 615             struct {
 616                 uint64_t val;
 617                 uint32_t size;
 618                 uint32_t type;
 619                 uint8_t pad[512 - 16];
 620             } QEMU_PACKED footer_marker;
 621
 622             uint32_t magic;
 623             VMDK4Header header;
 624             uint8_t pad[512 - 4 - sizeof(VMDK4Header)];
 625
 626             struct {
 627                 uint64_t val;
 628                 uint32_t size;
 629                 uint32_t type;
 630                 uint8_t pad[512 - 16];
 631             } QEMU_PACKED eos_marker;
 632         } QEMU_PACKED footer;
 633
 634         ret = bdrv_pread(file,
 635             bs->file->total_sectors * 512 - 1536,
 636             &footer, sizeof(footer));
 637         if (ret < 0) {
 638             return ret;
 639         }
 640
 641         /* Some sanity checks for the footer */
 642         if (be32_to_cpu(footer.magic) != VMDK4_MAGIC ||
 643             le32_to_cpu(footer.footer_marker.size) != 0  ||
 644             le32_to_cpu(footer.footer_marker.type) != MARKER_FOOTER ||
 645             le64_to_cpu(footer.eos_marker.val) != 0  ||
 646             le32_to_cpu(footer.eos_marker.size) != 0  ||
 647             le32_to_cpu(footer.eos_marker.type) != MARKER_END_OF_STREAM)
 648         {
 649             return -EINVAL;
 650         }
 651
 652         header = footer.header;
 653     }
 654
 655     if (le32_to_cpu(header.version) > 3) {
 656         char buf[64];
 657         snprintf(buf, sizeof(buf), "VMDK version %" PRId32,
 658                  le32_to_cpu(header.version));
 659         error_set(errp, QERR_UNKNOWN_BLOCK_FORMAT_FEATURE,
 660                   bdrv_get_device_name(bs), "vmdk", buf);
 661         return -ENOTSUP;
 662     } else if (le32_to_cpu(header.version) == 3 && (flags & BDRV_O_RDWR)) {
 663         /* VMware KB 2064959 explains that version 3 added support for
 664          * persistent changed block tracking (CBT), and backup software can
 665          * read it as version=1 if it doesn't care about the changed area
 666          * information. So we are safe to enable read only. */
 667         error_setg(errp, "VMDK version 3 must be read only");
 668         return -EINVAL;
 669     }
 670
 671     if (le32_to_cpu(header.num_gtes_per_gt) > 512) {
 672         error_setg(errp, "L2 table size too big");
 673         return -EINVAL;
 674     }
 675
 676     l1_entry_sectors = le32_to_cpu(header.num_gtes_per_gt)
 677                         * le64_to_cpu(header.granularity);
 678     if (l1_entry_sectors == 0) {
 679         return -EINVAL;
 680     }
 681     l1_size = (le64_to_cpu(header.capacity) + l1_entry_sectors - 1)
 682                 / l1_entry_sectors;
 683     if (le32_to_cpu(header.flags) & VMDK4_FLAG_RGD) {
 684         l1_backup_offset = le64_to_cpu(header.rgd_offset) << 9;
 685     }
 686     if (bdrv_nb_sectors(file) < le64_to_cpu(header.grain_offset)) {
 687         error_setg(errp, "File truncated, expecting at least %" PRId64 " bytes",
 688                    (int64_t)(le64_to_cpu(header.grain_offset)
 689                              * BDRV_SECTOR_SIZE));
 690         return -EINVAL;
 691     }
 692
 693     ret = vmdk_add_extent(bs, file, false,
 694                           le64_to_cpu(header.capacity),
 695                           le64_to_cpu(header.gd_offset) << 9,
 696                           l1_backup_offset,
 697                           l1_size,
 698                           le32_to_cpu(header.num_gtes_per_gt),
 699                           le64_to_cpu(header.granularity),
 700                           &extent,
 701                           errp);
 702     if (ret < 0) {
 703         return ret;
 704     }
 705     extent->compressed =
 706         le16_to_cpu(header.compressAlgorithm) == VMDK4_COMPRESSION_DEFLATE;
 707     if (extent->compressed) {
 708         g_free(s->create_type);
 709         s->create_type = g_strdup("streamOptimized");
 710     }
 711     extent->has_marker = le32_to_cpu(header.flags) & VMDK4_FLAG_MARKER;
 712     extent->version = le32_to_cpu(header.version);
 713     extent->has_zero_grain = le32_to_cpu(header.flags) & VMDK4_FLAG_ZERO_GRAIN;
 714     ret = vmdk_init_tables(bs, extent, errp);
 715     if (ret) {
 716         /* free extent allocated by vmdk_add_extent */
 717         vmdk_free_last_extent(bs);
 718     }
 719     return ret;
 720 }
 721
 722 /* find an option value out of descriptor file */
 723 static int vmdk_parse_description(const char *desc, const char *opt_name,
 724         char *buf, int buf_size)
 725 {
 726     char *opt_pos, *opt_end;
 727     const char *end = desc + strlen(desc);
 728
 729     opt_pos = strstr(desc, opt_name);
 730     if (!opt_pos) {
 731         return VMDK_ERROR;
 732     }
 733     /* Skip "=\"" following opt_name */
 734     opt_pos += strlen(opt_name) + 2;
 735     if (opt_pos >= end) {
 736         return VMDK_ERROR;
 737     }
 738     opt_end = opt_pos;
 739     while (opt_end < end && *opt_end != '"') {
 740         opt_end++;
 741     }
 742     if (opt_end == end || buf_size < opt_end - opt_pos + 1) {
 743         return VMDK_ERROR;
 744     }
 745     pstrcpy(buf, opt_end - opt_pos + 1, opt_pos);
 746     return VMDK_OK;
 747 }
 748
 749 /* Open an extent file and append to bs array */
 750 static int vmdk_open_sparse(BlockDriverState *bs,
 751                             BlockDriverState *file, int flags,
 752                             char *buf, Error **errp)
 753 {
 754     uint32_t magic;
 755
 756     magic = ldl_be_p(buf);
 757     switch (magic) {
 758         case VMDK3_MAGIC:
 759             return vmdk_open_vmfs_sparse(bs, file, flags, errp);
 760             break;
 761         case VMDK4_MAGIC:
 762             return vmdk_open_vmdk4(bs, file, flags, errp);
 763             break;
 764         default:
 765             error_setg(errp, "Image not in VMDK format");
 766             return -EINVAL;
 767             break;
 768     }
 769 }
 770
 771 static int vmdk_parse_extents(const char *desc, BlockDriverState *bs,
 772                               const char *desc_file_path, Error **errp)
 773 {
 774     int ret;
 775     char access[11];
 776     char type[11];
 777     char fname[512];
 778     const char *p = desc;
 779     int64_t sectors = 0;
 780     int64_t flat_offset;
 781     char extent_path[PATH_MAX];
 782     BlockDriverState *extent_file;
 783     BDRVVmdkState *s = bs->opaque;
 784     VmdkExtent *extent;
 785
 786     while (*p) {
 787         /* parse extent line:
 788          * RW [size in sectors] FLAT "file-name.vmdk" OFFSET
 789          * or
 790          * RW [size in sectors] SPARSE "file-name.vmdk"
 791          */
 792         flat_offset = -1;
 793         ret = sscanf(p, "%10s %" SCNd64 " %10s \"%511[^\n\r\"]\" %" SCNd64,
 794                 access, &sectors, type, fname, &flat_offset);
 795         if (ret < 4 || strcmp(access, "RW")) {
 796             goto next_line;
 797         } else if (!strcmp(type, "FLAT")) {
 798             if (ret != 5 || flat_offset < 0) {
 799                 error_setg(errp, "Invalid extent lines: \n%s", p);
 800                 return -EINVAL;
 801             }
 802         } else if (!strcmp(type, "VMFS")) {
 803             if (ret == 4) {
 804                 flat_offset = 0;
 805             } else {
 806                 error_setg(errp, "Invalid extent lines:\n%s", p);
 807                 return -EINVAL;
 808             }
 809         } else if (ret != 4) {
 810             error_setg(errp, "Invalid extent lines:\n%s", p);
 811             return -EINVAL;
 812         }
 813
 814         if (sectors <= 0 ||
 815             (strcmp(type, "FLAT") && strcmp(type, "SPARSE") &&
 816              strcmp(type, "VMFS") && strcmp(type, "VMFSSPARSE")) ||
 817             (strcmp(access, "RW"))) {
 818             goto next_line;
 819         }
 820
 821         path_combine(extent_path, sizeof(extent_path),
 822                 desc_file_path, fname);
 823         extent_file = NULL;
 824         ret = bdrv_open(&extent_file, extent_path, NULL, NULL,
 825                         bs->open_flags | BDRV_O_PROTOCOL, NULL, errp);
 826         if (ret) {
 827             return ret;
 828         }
 829
 830         /* save to extents array */
 831         if (!strcmp(type, "FLAT") || !strcmp(type, "VMFS")) {
 832             /* FLAT extent */
 833
 834             ret = vmdk_add_extent(bs, extent_file, true, sectors,
 835                             0, 0, 0, 0, 0, &extent, errp);
 836             if (ret < 0) {
 837                 bdrv_unref(extent_file);
 838                 return ret;
 839             }
 840             extent->flat_start_offset = flat_offset << 9;
 841         } else if (!strcmp(type, "SPARSE") || !strcmp(type, "VMFSSPARSE")) {
 842             /* SPARSE extent and VMFSSPARSE extent are both "COWD" sparse file*/
 843             char *buf = vmdk_read_desc(extent_file, 0, errp);
 844             if (!buf) {
 845                 ret = -EINVAL;
 846             } else {
 847                 ret = vmdk_open_sparse(bs, extent_file, bs->open_flags, buf, errp);
 848             }
 849             g_free(buf);
 850             if (ret) {
 851                 bdrv_unref(extent_file);
 852                 return ret;
 853             }
 854             extent = &s->extents[s->num_extents - 1];
 855         } else {
 856             error_setg(errp, "Unsupported extent type '%s'", type);
 857             bdrv_unref(extent_file);
 858             return -ENOTSUP;
 859         }
 860         extent->type = g_strdup(type);
 861 next_line:
 862         /* move to next line */
 863         while (*p) {
 864             if (*p == '\n') {
 865                 p++;
 866                 break;
 867             }
 868             p++;
 869         }
 870     }
 871     return 0;
 872 }
 873
 874 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf,
 875                                Error **errp)
 876 {
 877     int ret;
 878     char ct[128];
 879     BDRVVmdkState *s = bs->opaque;
 880
 881     if (vmdk_parse_description(buf, "createType", ct, sizeof(ct))) {
 882         error_setg(errp, "invalid VMDK image descriptor");
 883         ret = -EINVAL;
 884         goto exit;
 885     }
 886     if (strcmp(ct, "monolithicFlat") &&
 887         strcmp(ct, "vmfs") &&
 888         strcmp(ct, "vmfsSparse") &&
 889         strcmp(ct, "twoGbMaxExtentSparse") &&
 890         strcmp(ct, "twoGbMaxExtentFlat")) {
 891         error_setg(errp, "Unsupported image type '%s'", ct);
 892         ret = -ENOTSUP;
 893         goto exit;
 894     }
 895     s->create_type = g_strdup(ct);
 896     s->desc_offset = 0;
 897     ret = vmdk_parse_extents(buf, bs, bs->file->filename, errp);
 898 exit:
 899     return ret;
 900 }
 901
 902 static int vmdk_open(BlockDriverState *bs, QDict *options, int flags,
 903                      Error **errp)
 904 {
 905     char *buf = NULL;
 906     int ret;
 907     BDRVVmdkState *s = bs->opaque;
 908     uint32_t magic;
 909
 910     buf = vmdk_read_desc(bs->file, 0, errp);
 911     if (!buf) {
 912         return -EINVAL;
 913     }
 914
 915     magic = ldl_be_p(buf);
 916     switch (magic) {
 917         case VMDK3_MAGIC:
 918         case VMDK4_MAGIC:
 919             ret = vmdk_open_sparse(bs, bs->file, flags, buf, errp);
 920             s->desc_offset = 0x200;
 921             break;
 922         default:
 923             ret = vmdk_open_desc_file(bs, flags, buf, errp);
 924             break;
 925     }
 926     if (ret) {
 927         goto fail;
 928     }
 929
 930     /* try to open parent images, if exist */
 931     ret = vmdk_parent_open(bs);
 932     if (ret) {
 933         goto fail;
 934     }
 935     s->cid = vmdk_read_cid(bs, 0);
 936     s->parent_cid = vmdk_read_cid(bs, 1);
 937     qemu_co_mutex_init(&s->lock);
 938
 939     /* Disable migration when VMDK images are used */
 940     error_set(&s->migration_blocker,
 941               QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED,
 942               "vmdk", bdrv_get_device_name(bs), "live migration");
 943     migrate_add_blocker(s->migration_blocker);
 944     g_free(buf);
 945     return 0;
 946
 947 fail:
 948     g_free(buf);
 949     g_free(s->create_type);
 950     s->create_type = NULL;
 951     vmdk_free_extents(bs);
 952     return ret;
 953 }
 954
 955
 956 static void vmdk_refresh_limits(BlockDriverState *bs, Error **errp)
 957 {
 958     BDRVVmdkState *s = bs->opaque;
 959     int i;
 960
 961     for (i = 0; i < s->num_extents; i++) {
 962         if (!s->extents[i].flat) {
 963             bs->bl.write_zeroes_alignment =
 964                 MAX(bs->bl.write_zeroes_alignment,
 965                     s->extents[i].cluster_sectors);
 966         }
 967     }
 968 }
 969
 970 /**
 971  * get_whole_cluster
 972  *
 973  * Copy backing file's cluster that covers @sector_num, otherwise write zero,
 974  * to the cluster at @cluster_sector_num.
 975  *
 976  * If @skip_start_sector < @skip_end_sector, the relative range
 977  * [@skip_start_sector, @skip_end_sector) is not copied or written, and leave
 978  * it for call to write user data in the request.
 979  */
 980 static int get_whole_cluster(BlockDriverState *bs,
 981                              VmdkExtent *extent,
 982                              uint64_t cluster_sector_num,
 983                              uint64_t sector_num,
 984                              uint64_t skip_start_sector,
 985                              uint64_t skip_end_sector)
 986 {
 987     int ret = VMDK_OK;
 988     int64_t cluster_bytes;
 989     uint8_t *whole_grain;
 990
 991     /* For COW, align request sector_num to cluster start */
 992     sector_num = QEMU_ALIGN_DOWN(sector_num, extent->cluster_sectors);
 993     cluster_bytes = extent->cluster_sectors << BDRV_SECTOR_BITS;
 994     whole_grain = qemu_blockalign(bs, cluster_bytes);
 995
 996     if (!bs->backing_hd) {
 997         memset(whole_grain, 0,  skip_start_sector << BDRV_SECTOR_BITS);
 998         memset(whole_grain + (skip_end_sector << BDRV_SECTOR_BITS), 0,
 999                cluster_bytes - (skip_end_sector << BDRV_SECTOR_BITS));
1000     }
1001
1002     assert(skip_end_sector <= extent->cluster_sectors);
1003     /* we will be here if it's first write on non-exist grain(cluster).
1004      * try to read from parent image, if exist */
1005     if (bs->backing_hd && !vmdk_is_cid_valid(bs)) {
1006         ret = VMDK_ERROR;
1007         goto exit;
1008     }
1009
1010     /* Read backing data before skip range */
1011     if (skip_start_sector > 0) {
1012         if (bs->backing_hd) {
1013             ret = bdrv_read(bs->backing_hd, sector_num,
1014                             whole_grain, skip_start_sector);
1015             if (ret < 0) {
1016                 ret = VMDK_ERROR;
1017                 goto exit;
1018             }
1019         }
1020         ret = bdrv_write(extent->file, cluster_sector_num, whole_grain,
1021                          skip_start_sector);
1022         if (ret < 0) {
1023             ret = VMDK_ERROR;
1024             goto exit;
1025         }
1026     }
1027     /* Read backing data after skip range */
1028     if (skip_end_sector < extent->cluster_sectors) {
1029         if (bs->backing_hd) {
1030             ret = bdrv_read(bs->backing_hd, sector_num + skip_end_sector,
1031                             whole_grain + (skip_end_sector << BDRV_SECTOR_BITS),
1032                             extent->cluster_sectors - skip_end_sector);
1033             if (ret < 0) {
1034                 ret = VMDK_ERROR;
1035                 goto exit;
1036             }
1037         }
1038         ret = bdrv_write(extent->file, cluster_sector_num + skip_end_sector,
1039                          whole_grain + (skip_end_sector << BDRV_SECTOR_BITS),
1040                          extent->cluster_sectors - skip_end_sector);
1041         if (ret < 0) {
1042             ret = VMDK_ERROR;
1043             goto exit;
1044         }
1045     }
1046
1047 exit:
1048     qemu_vfree(whole_grain);
1049     return ret;
1050 }
1051
1052 static int vmdk_L2update(VmdkExtent *extent, VmdkMetaData *m_data,
1053                          uint32_t offset)
1054 {
1055     offset = cpu_to_le32(offset);
1056     /* update L2 table */
1057     if (bdrv_pwrite_sync(
1058                 extent->file,
1059                 ((int64_t)m_data->l2_offset * 512)
1060                     + (m_data->l2_index * sizeof(offset)),
1061                 &offset, sizeof(offset)) < 0) {
1062         return VMDK_ERROR;
1063     }
1064     /* update backup L2 table */
1065     if (extent->l1_backup_table_offset != 0) {
1066         m_data->l2_offset = extent->l1_backup_table[m_data->l1_index];
1067         if (bdrv_pwrite_sync(
1068                     extent->file,
1069                     ((int64_t)m_data->l2_offset * 512)
1070                         + (m_data->l2_index * sizeof(offset)),
1071                     &offset, sizeof(offset)) < 0) {
1072             return VMDK_ERROR;
1073         }
1074     }
1075     if (m_data->l2_cache_entry) {
1076         *m_data->l2_cache_entry = offset;
1077     }
1078
1079     return VMDK_OK;
1080 }
1081
1082 /**
1083  * get_cluster_offset
1084  *
1085  * Look up cluster offset in extent file by sector number, and store in
1086  * @cluster_offset.
1087  *
1088  * For flat extents, the start offset as parsed from the description file is
1089  * returned.
1090  *
1091  * For sparse extents, look up in L1, L2 table. If allocate is true, return an
1092  * offset for a new cluster and update L2 cache. If there is a backing file,
1093  * COW is done before returning; otherwise, zeroes are written to the allocated
1094  * cluster. Both COW and zero writing skips the sector range
1095  * [@skip_start_sector, @skip_end_sector) passed in by caller, because caller
1096  * has new data to write there.
1097  *
1098  * Returns: VMDK_OK if cluster exists and mapped in the image.
1099  *          VMDK_UNALLOC if cluster is not mapped and @allocate is false.
1100  *          VMDK_ERROR if failed.
1101  */
1102 static int get_cluster_offset(BlockDriverState *bs,
1103                               VmdkExtent *extent,
1104                               VmdkMetaData *m_data,
1105                               uint64_t offset,
1106                               bool allocate,
1107                               uint64_t *cluster_offset,
1108                               uint64_t skip_start_sector,
1109                               uint64_t skip_end_sector)
1110 {
1111     unsigned int l1_index, l2_offset, l2_index;
1112     int min_index, i, j;
1113     uint32_t min_count, *l2_table;
1114     bool zeroed = false;
1115     int64_t ret;
1116     int64_t cluster_sector;
1117
1118     if (m_data) {
1119         m_data->valid = 0;
1120     }
1121     if (extent->flat) {
1122         *cluster_offset = extent->flat_start_offset;
1123         return VMDK_OK;
1124     }
1125
1126     offset -= (extent->end_sector - extent->sectors) * SECTOR_SIZE;
1127     l1_index = (offset >> 9) / extent->l1_entry_sectors;
1128     if (l1_index >= extent->l1_size) {
1129         return VMDK_ERROR;
1130     }
1131     l2_offset = extent->l1_table[l1_index];
1132     if (!l2_offset) {
1133         return VMDK_UNALLOC;
1134     }
1135     for (i = 0; i < L2_CACHE_SIZE; i++) {
1136         if (l2_offset == extent->l2_cache_offsets[i]) {
1137             /* increment the hit count */
1138             if (++extent->l2_cache_counts[i] == 0xffffffff) {
1139                 for (j = 0; j < L2_CACHE_SIZE; j++) {
1140                     extent->l2_cache_counts[j] >>= 1;
1141                 }
1142             }
1143             l2_table = extent->l2_cache + (i * extent->l2_size);
1144             goto found;
1145         }
1146     }
1147     /* not found: load a new entry in the least used one */
1148     min_index = 0;
1149     min_count = 0xffffffff;
1150     for (i = 0; i < L2_CACHE_SIZE; i++) {
1151         if (extent->l2_cache_counts[i] < min_count) {
1152             min_count = extent->l2_cache_counts[i];
1153             min_index = i;
1154         }
1155     }
1156     l2_table = extent->l2_cache + (min_index * extent->l2_size);
1157     if (bdrv_pread(
1158                 extent->file,
1159                 (int64_t)l2_offset * 512,
1160                 l2_table,
1161                 extent->l2_size * sizeof(uint32_t)
1162             ) != extent->l2_size * sizeof(uint32_t)) {
1163         return VMDK_ERROR;
1164     }
1165
1166     extent->l2_cache_offsets[min_index] = l2_offset;
1167     extent->l2_cache_counts[min_index] = 1;
1168  found:
1169     l2_index = ((offset >> 9) / extent->cluster_sectors) % extent->l2_size;
1170     cluster_sector = le32_to_cpu(l2_table[l2_index]);
1171
1172     if (m_data) {
1173         m_data->valid = 1;
1174         m_data->l1_index = l1_index;
1175         m_data->l2_index = l2_index;
1176         m_data->l2_offset = l2_offset;
1177         m_data->l2_cache_entry = &l2_table[l2_index];
1178     }
1179     if (extent->has_zero_grain && cluster_sector == VMDK_GTE_ZEROED) {
1180         zeroed = true;
1181     }
1182
1183     if (!cluster_sector || zeroed) {
1184         if (!allocate) {
1185             return zeroed ? VMDK_ZEROED : VMDK_UNALLOC;
1186         }
1187
1188         cluster_sector = extent->next_cluster_sector;
1189         extent->next_cluster_sector += extent->cluster_sectors;
1190
1191         /* First of all we write grain itself, to avoid race condition
1192          * that may to corrupt the image.
1193          * This problem may occur because of insufficient space on host disk
1194          * or inappropriate VM shutdown.
1195          */
1196         ret = get_whole_cluster(bs, extent,
1197                                 cluster_sector,
1198                                 offset >> BDRV_SECTOR_BITS,
1199                                 skip_start_sector, skip_end_sector);
1200         if (ret) {
1201             return ret;
1202         }
1203     }
1204     *cluster_offset = cluster_sector << BDRV_SECTOR_BITS;
1205     return VMDK_OK;
1206 }
1207
1208 static VmdkExtent *find_extent(BDRVVmdkState *s,
1209                                 int64_t sector_num, VmdkExtent *start_hint)
1210 {
1211     VmdkExtent *extent = start_hint;
1212
1213     if (!extent) {
1214         extent = &s->extents[0];
1215     }
1216     while (extent < &s->extents[s->num_extents]) {
1217         if (sector_num < extent->end_sector) {
1218             return extent;
1219         }
1220         extent++;
1221     }
1222     return NULL;
1223 }
1224
1225 static int64_t coroutine_fn vmdk_co_get_block_status(BlockDriverState *bs,
1226         int64_t sector_num, int nb_sectors, int *pnum)
1227 {
1228     BDRVVmdkState *s = bs->opaque;
1229     int64_t index_in_cluster, n, ret;
1230     uint64_t offset;
1231     VmdkExtent *extent;
1232
1233     extent = find_extent(s, sector_num, NULL);
1234     if (!extent) {
1235         return 0;
1236     }
1237     qemu_co_mutex_lock(&s->lock);
1238     ret = get_cluster_offset(bs, extent, NULL,
1239                              sector_num * 512, false, &offset,
1240                              0, 0);
1241     qemu_co_mutex_unlock(&s->lock);
1242
1243     switch (ret) {
1244     case VMDK_ERROR:
1245         ret = -EIO;
1246         break;
1247     case VMDK_UNALLOC:
1248         ret = 0;
1249         break;
1250     case VMDK_ZEROED:
1251         ret = BDRV_BLOCK_ZERO;
1252         break;
1253     case VMDK_OK:
1254         ret = BDRV_BLOCK_DATA;
1255         if (extent->file == bs->file && !extent->compressed) {
1256             ret |= BDRV_BLOCK_OFFSET_VALID | offset;
1257         }
1258
1259         break;
1260     }
1261
1262     index_in_cluster = sector_num % extent->cluster_sectors;
1263     n = extent->cluster_sectors - index_in_cluster;
1264     if (n > nb_sectors) {
1265         n = nb_sectors;
1266     }
1267     *pnum = n;
1268     return ret;
1269 }
1270
1271 static int vmdk_write_extent(VmdkExtent *extent, int64_t cluster_offset,
1272                             int64_t offset_in_cluster, const uint8_t *buf,
1273                             int nb_sectors, int64_t sector_num)
1274 {
1275     int ret;
1276     VmdkGrainMarker *data = NULL;
1277     uLongf buf_len;
1278     const uint8_t *write_buf = buf;
1279     int write_len = nb_sectors * 512;
1280
1281     if (extent->compressed) {
1282         if (!extent->has_marker) {
1283             ret = -EINVAL;
1284             goto out;
1285         }
1286         buf_len = (extent->cluster_sectors << 9) * 2;
1287         data = g_malloc(buf_len + sizeof(VmdkGrainMarker));
1288         if (compress(data->data, &buf_len, buf, nb_sectors << 9) != Z_OK ||
1289                 buf_len == 0) {
1290             ret = -EINVAL;
1291             goto out;
1292         }
1293         data->lba = sector_num;
1294         data->size = buf_len;
1295         write_buf = (uint8_t *)data;
1296         write_len = buf_len + sizeof(VmdkGrainMarker);
1297     }
1298     ret = bdrv_pwrite(extent->file,
1299                         cluster_offset + offset_in_cluster,
1300                         write_buf,
1301                         write_len);
1302     if (ret != write_len) {
1303         ret = ret < 0 ? ret : -EIO;
1304         goto out;
1305     }
1306     ret = 0;
1307  out:
1308     g_free(data);
1309     return ret;
1310 }
1311
1312 static int vmdk_read_extent(VmdkExtent *extent, int64_t cluster_offset,
1313                             int64_t offset_in_cluster, uint8_t *buf,
1314                             int nb_sectors)
1315 {
1316     int ret;
1317     int cluster_bytes, buf_bytes;
1318     uint8_t *cluster_buf, *compressed_data;
1319     uint8_t *uncomp_buf;
1320     uint32_t data_len;
1321     VmdkGrainMarker *marker;
1322     uLongf buf_len;
1323
1324
1325     if (!extent->compressed) {
1326         ret = bdrv_pread(extent->file,
1327                           cluster_offset + offset_in_cluster,
1328                           buf, nb_sectors * 512);
1329         if (ret == nb_sectors * 512) {
1330             return 0;
1331         } else {
1332             return -EIO;
1333         }
1334     }
1335     cluster_bytes = extent->cluster_sectors * 512;
1336     /* Read two clusters in case GrainMarker + compressed data > one cluster */
1337     buf_bytes = cluster_bytes * 2;
1338     cluster_buf = g_malloc(buf_bytes);
1339     uncomp_buf = g_malloc(cluster_bytes);
1340     ret = bdrv_pread(extent->file,
1341                 cluster_offset,
1342                 cluster_buf, buf_bytes);
1343     if (ret < 0) {
1344         goto out;
1345     }
1346     compressed_data = cluster_buf;
1347     buf_len = cluster_bytes;
1348     data_len = cluster_bytes;
1349     if (extent->has_marker) {
1350         marker = (VmdkGrainMarker *)cluster_buf;
1351         compressed_data = marker->data;
1352         data_len = le32_to_cpu(marker->size);
1353     }
1354     if (!data_len || data_len > buf_bytes) {
1355         ret = -EINVAL;
1356         goto out;
1357     }
1358     ret = uncompress(uncomp_buf, &buf_len, compressed_data, data_len);
1359     if (ret != Z_OK) {
1360         ret = -EINVAL;
1361         goto out;
1362
1363     }
1364     if (offset_in_cluster < 0 ||
1365             offset_in_cluster + nb_sectors * 512 > buf_len) {
1366         ret = -EINVAL;
1367         goto out;
1368     }
1369     memcpy(buf, uncomp_buf + offset_in_cluster, nb_sectors * 512);
1370     ret = 0;
1371
1372  out:
1373     g_free(uncomp_buf);
1374     g_free(cluster_buf);
1375     return ret;
1376 }
1377
1378 static int vmdk_read(BlockDriverState *bs, int64_t sector_num,
1379                     uint8_t *buf, int nb_sectors)
1380 {
1381     BDRVVmdkState *s = bs->opaque;
1382     int ret;
1383     uint64_t n, index_in_cluster;
1384     uint64_t extent_begin_sector, extent_relative_sector_num;
1385     VmdkExtent *extent = NULL;
1386     uint64_t cluster_offset;
1387
1388     while (nb_sectors > 0) {
1389         extent = find_extent(s, sector_num, extent);
1390         if (!extent) {
1391             return -EIO;
1392         }
1393         ret = get_cluster_offset(bs, extent, NULL,
1394                                  sector_num << 9, false, &cluster_offset,
1395                                  0, 0);
1396         extent_begin_sector = extent->end_sector - extent->sectors;
1397         extent_relative_sector_num = sector_num - extent_begin_sector;
1398         index_in_cluster = extent_relative_sector_num % extent->cluster_sectors;
1399         n = extent->cluster_sectors - index_in_cluster;
1400         if (n > nb_sectors) {
1401             n = nb_sectors;
1402         }
1403         if (ret != VMDK_OK) {
1404             /* if not allocated, try to read from parent image, if exist */
1405             if (bs->backing_hd && ret != VMDK_ZEROED) {
1406                 if (!vmdk_is_cid_valid(bs)) {
1407                     return -EINVAL;
1408                 }
1409                 ret = bdrv_read(bs->backing_hd, sector_num, buf, n);
1410                 if (ret < 0) {
1411                     return ret;
1412                 }
1413             } else {
1414                 memset(buf, 0, 512 * n);
1415             }
1416         } else {
1417             ret = vmdk_read_extent(extent,
1418                             cluster_offset, index_in_cluster * 512,
1419                             buf, n);
1420             if (ret) {
1421                 return ret;
1422             }
1423         }
1424         nb_sectors -= n;
1425         sector_num += n;
1426         buf += n * 512;
1427     }
1428     return 0;
1429 }
1430
1431 static coroutine_fn int vmdk_co_read(BlockDriverState *bs, int64_t sector_num,
1432                                      uint8_t *buf, int nb_sectors)
1433 {
1434     int ret;
1435     BDRVVmdkState *s = bs->opaque;
1436     qemu_co_mutex_lock(&s->lock);
1437     ret = vmdk_read(bs, sector_num, buf, nb_sectors);
1438     qemu_co_mutex_unlock(&s->lock);
1439     return ret;
1440 }
1441
1442 /**
1443  * vmdk_write:
1444  * @zeroed:       buf is ignored (data is zero), use zeroed_grain GTE feature
1445  *                if possible, otherwise return -ENOTSUP.
1446  * @zero_dry_run: used for zeroed == true only, don't update L2 table, just try
1447  *                with each cluster. By dry run we can find if the zero write
1448  *                is possible without modifying image data.
1449  *
1450  * Returns: error code with 0 for success.
1451  */
1452 static int vmdk_write(BlockDriverState *bs, int64_t sector_num,
1453                       const uint8_t *buf, int nb_sectors,
1454                       bool zeroed, bool zero_dry_run)
1455 {
1456     BDRVVmdkState *s = bs->opaque;
1457     VmdkExtent *extent = NULL;
1458     int ret;
1459     int64_t index_in_cluster, n;
1460     uint64_t extent_begin_sector, extent_relative_sector_num;
1461     uint64_t cluster_offset;
1462     VmdkMetaData m_data;
1463
1464     if (sector_num > bs->total_sectors) {
1465         error_report("Wrong offset: sector_num=0x%" PRIx64
1466                 " total_sectors=0x%" PRIx64 "\n",
1467                 sector_num, bs->total_sectors);
1468         return -EIO;
1469     }
1470
1471     while (nb_sectors > 0) {
1472         extent = find_extent(s, sector_num, extent);
1473         if (!extent) {
1474             return -EIO;
1475         }
1476         extent_begin_sector = extent->end_sector - extent->sectors;
1477         extent_relative_sector_num = sector_num - extent_begin_sector;
1478         index_in_cluster = extent_relative_sector_num % extent->cluster_sectors;
1479         n = extent->cluster_sectors - index_in_cluster;
1480         if (n > nb_sectors) {
1481             n = nb_sectors;
1482         }
1483         ret = get_cluster_offset(bs, extent, &m_data, sector_num << 9,
1484                                  !(extent->compressed || zeroed),
1485                                  &cluster_offset,
1486                                  index_in_cluster, index_in_cluster + n);
1487         if (extent->compressed) {
1488             if (ret == VMDK_OK) {
1489                 /* Refuse write to allocated cluster for streamOptimized */
1490                 error_report("Could not write to allocated cluster"
1491                               " for streamOptimized");
1492                 return -EIO;
1493             } else {
1494                 /* allocate */
1495                 ret = get_cluster_offset(bs, extent, &m_data, sector_num << 9,
1496                                          true, &cluster_offset, 0, 0);
1497             }
1498         }
1499         if (ret == VMDK_ERROR) {
1500             return -EINVAL;
1501         }
1502         if (zeroed) {
1503             /* Do zeroed write, buf is ignored */
1504             if (extent->has_zero_grain &&
1505                     index_in_cluster == 0 &&
1506                     n >= extent->cluster_sectors) {
1507                 n = extent->cluster_sectors;
1508                 if (!zero_dry_run) {
1509                     /* update L2 tables */
1510                     if (vmdk_L2update(extent, &m_data, VMDK_GTE_ZEROED)
1511                             != VMDK_OK) {
1512                         return -EIO;
1513                     }
1514                 }
1515             } else {
1516                 return -ENOTSUP;
1517             }
1518         } else {
1519             ret = vmdk_write_extent(extent,
1520                             cluster_offset, index_in_cluster * 512,
1521                             buf, n, sector_num);
1522             if (ret) {
1523                 return ret;
1524             }
1525             if (m_data.valid) {
1526                 /* update L2 tables */
1527                 if (vmdk_L2update(extent, &m_data,
1528                                   cluster_offset >> BDRV_SECTOR_BITS)
1529                         != VMDK_OK) {
1530                     return -EIO;
1531                 }
1532             }
1533         }
1534         nb_sectors -= n;
1535         sector_num += n;
1536         buf += n * 512;
1537
1538         /* update CID on the first write every time the virtual disk is
1539          * opened */
1540         if (!s->cid_updated) {
1541             ret = vmdk_write_cid(bs, time(NULL));
1542             if (ret < 0) {
1543                 return ret;
1544             }
1545             s->cid_updated = true;
1546         }
1547     }
1548     return 0;
1549 }
1550
1551 static coroutine_fn int vmdk_co_write(BlockDriverState *bs, int64_t sector_num,
1552                                       const uint8_t *buf, int nb_sectors)
1553 {
1554     int ret;
1555     BDRVVmdkState *s = bs->opaque;
1556     qemu_co_mutex_lock(&s->lock);
1557     ret = vmdk_write(bs, sector_num, buf, nb_sectors, false, false);
1558     qemu_co_mutex_unlock(&s->lock);
1559     return ret;
1560 }
1561
1562 static int vmdk_write_compressed(BlockDriverState *bs,
1563                                  int64_t sector_num,
1564                                  const uint8_t *buf,
1565                                  int nb_sectors)
1566 {
1567     BDRVVmdkState *s = bs->opaque;
1568     if (s->num_extents == 1 && s->extents[0].compressed) {
1569         return vmdk_write(bs, sector_num, buf, nb_sectors, false, false);
1570     } else {
1571         return -ENOTSUP;
1572     }
1573 }
1574
1575 static int coroutine_fn vmdk_co_write_zeroes(BlockDriverState *bs,
1576                                              int64_t sector_num,
1577                                              int nb_sectors,
1578                                              BdrvRequestFlags flags)
1579 {
1580     int ret;
1581     BDRVVmdkState *s = bs->opaque;
1582     qemu_co_mutex_lock(&s->lock);
1583     /* write zeroes could fail if sectors not aligned to cluster, test it with
1584      * dry_run == true before really updating image */
1585     ret = vmdk_write(bs, sector_num, NULL, nb_sectors, true, true);
1586     if (!ret) {
1587         ret = vmdk_write(bs, sector_num, NULL, nb_sectors, true, false);
1588     }
1589     qemu_co_mutex_unlock(&s->lock);
1590     return ret;
1591 }
1592
1593 static int vmdk_create_extent(const char *filename, int64_t filesize,
1594                               bool flat, bool compress, bool zeroed_grain,
1595                               QemuOpts *opts, Error **errp)
1596 {
1597     int ret, i;
1598     BlockDriverState *bs = NULL;
1599     VMDK4Header header;
1600     Error *local_err = NULL;
1601     uint32_t tmp, magic, grains, gd_sectors, gt_size, gt_count;
1602     uint32_t *gd_buf = NULL;
1603     int gd_buf_size;
1604
1605     ret = bdrv_create_file(filename, opts, &local_err);
1606     if (ret < 0) {
1607         error_propagate(errp, local_err);
1608         goto exit;
1609     }
1610
1611     assert(bs == NULL);
1612     ret = bdrv_open(&bs, filename, NULL, NULL, BDRV_O_RDWR | BDRV_O_PROTOCOL,
1613                     NULL, &local_err);
1614     if (ret < 0) {
1615         error_propagate(errp, local_err);
1616         goto exit;
1617     }
1618
1619     if (flat) {
1620         ret = bdrv_truncate(bs, filesize);
1621         if (ret < 0) {
1622             error_setg_errno(errp, -ret, "Could not truncate file");
1623         }
1624         goto exit;
1625     }
1626     magic = cpu_to_be32(VMDK4_MAGIC);
1627     memset(&header, 0, sizeof(header));
1628     header.version = zeroed_grain ? 2 : 1;
1629     header.flags = VMDK4_FLAG_RGD | VMDK4_FLAG_NL_DETECT
1630                    | (compress ? VMDK4_FLAG_COMPRESS | VMDK4_FLAG_MARKER : 0)
1631                    | (zeroed_grain ? VMDK4_FLAG_ZERO_GRAIN : 0);
1632     header.compressAlgorithm = compress ? VMDK4_COMPRESSION_DEFLATE : 0;
1633     header.capacity = filesize / BDRV_SECTOR_SIZE;
1634     header.granularity = 128;
1635     header.num_gtes_per_gt = BDRV_SECTOR_SIZE;
1636
1637     grains = DIV_ROUND_UP(filesize / BDRV_SECTOR_SIZE, header.granularity);
1638     gt_size = DIV_ROUND_UP(header.num_gtes_per_gt * sizeof(uint32_t),
1639                            BDRV_SECTOR_SIZE);
1640     gt_count = DIV_ROUND_UP(grains, header.num_gtes_per_gt);
1641     gd_sectors = DIV_ROUND_UP(gt_count * sizeof(uint32_t), BDRV_SECTOR_SIZE);
1642
1643     header.desc_offset = 1;
1644     header.desc_size = 20;
1645     header.rgd_offset = header.desc_offset + header.desc_size;
1646     header.gd_offset = header.rgd_offset + gd_sectors + (gt_size * gt_count);
1647     header.grain_offset =
1648         ROUND_UP(header.gd_offset + gd_sectors + (gt_size * gt_count),
1649                  header.granularity);
1650     /* swap endianness for all header fields */
1651     header.version = cpu_to_le32(header.version);
1652     header.flags = cpu_to_le32(header.flags);
1653     header.capacity = cpu_to_le64(header.capacity);
1654     header.granularity = cpu_to_le64(header.granularity);
1655     header.num_gtes_per_gt = cpu_to_le32(header.num_gtes_per_gt);
1656     header.desc_offset = cpu_to_le64(header.desc_offset);
1657     header.desc_size = cpu_to_le64(header.desc_size);
1658     header.rgd_offset = cpu_to_le64(header.rgd_offset);
1659     header.gd_offset = cpu_to_le64(header.gd_offset);
1660     header.grain_offset = cpu_to_le64(header.grain_offset);
1661     header.compressAlgorithm = cpu_to_le16(header.compressAlgorithm);
1662
1663     header.check_bytes[0] = 0xa;
1664     header.check_bytes[1] = 0x20;
1665     header.check_bytes[2] = 0xd;
1666     header.check_bytes[3] = 0xa;
1667
1668     /* write all the data */
1669     ret = bdrv_pwrite(bs, 0, &magic, sizeof(magic));
1670     if (ret < 0) {
1671         error_set(errp, QERR_IO_ERROR);
1672         goto exit;
1673     }
1674     ret = bdrv_pwrite(bs, sizeof(magic), &header, sizeof(header));
1675     if (ret < 0) {
1676         error_set(errp, QERR_IO_ERROR);
1677         goto exit;
1678     }
1679
1680     ret = bdrv_truncate(bs, le64_to_cpu(header.grain_offset) << 9);
1681     if (ret < 0) {
1682         error_setg_errno(errp, -ret, "Could not truncate file");
1683         goto exit;
1684     }
1685
1686     /* write grain directory */
1687     gd_buf_size = gd_sectors * BDRV_SECTOR_SIZE;
1688     gd_buf = g_malloc0(gd_buf_size);
1689     for (i = 0, tmp = le64_to_cpu(header.rgd_offset) + gd_sectors;
1690          i < gt_count; i++, tmp += gt_size) {
1691         gd_buf[i] = cpu_to_le32(tmp);
1692     }
1693     ret = bdrv_pwrite(bs, le64_to_cpu(header.rgd_offset) * BDRV_SECTOR_SIZE,
1694                       gd_buf, gd_buf_size);
1695     if (ret < 0) {
1696         error_set(errp, QERR_IO_ERROR);
1697         goto exit;
1698     }
1699
1700     /* write backup grain directory */
1701     for (i = 0, tmp = le64_to_cpu(header.gd_offset) + gd_sectors;
1702          i < gt_count; i++, tmp += gt_size) {
1703         gd_buf[i] = cpu_to_le32(tmp);
1704     }
1705     ret = bdrv_pwrite(bs, le64_to_cpu(header.gd_offset) * BDRV_SECTOR_SIZE,
1706                       gd_buf, gd_buf_size);
1707     if (ret < 0) {
1708         error_set(errp, QERR_IO_ERROR);
1709         goto exit;
1710     }
1711
1712     ret = 0;
1713 exit:
1714     if (bs) {
1715         bdrv_unref(bs);
1716     }
1717     g_free(gd_buf);
1718     return ret;
1719 }
1720
1721 static int filename_decompose(const char *filename, char *path, char *prefix,
1722                               char *postfix, size_t buf_len, Error **errp)
1723 {
1724     const char *p, *q;
1725
1726     if (filename == NULL || !strlen(filename)) {
1727         error_setg(errp, "No filename provided");
1728         return VMDK_ERROR;
1729     }
1730     p = strrchr(filename, '/');
1731     if (p == NULL) {
1732         p = strrchr(filename, '\\');
1733     }
1734     if (p == NULL) {
1735         p = strrchr(filename, ':');
1736     }
1737     if (p != NULL) {
1738         p++;
1739         if (p - filename >= buf_len) {
1740             return VMDK_ERROR;
1741         }
1742         pstrcpy(path, p - filename + 1, filename);
1743     } else {
1744         p = filename;
1745         path[0] = '\0';
1746     }
1747     q = strrchr(p, '.');
1748     if (q == NULL) {
1749         pstrcpy(prefix, buf_len, p);
1750         postfix[0] = '\0';
1751     } else {
1752         if (q - p >= buf_len) {
1753             return VMDK_ERROR;
1754         }
1755         pstrcpy(prefix, q - p + 1, p);
1756         pstrcpy(postfix, buf_len, q);
1757     }
1758     return VMDK_OK;
1759 }
1760
1761 static int vmdk_create(const char *filename, QemuOpts *opts, Error **errp)
1762 {
1763     int idx = 0;
1764     BlockDriverState *new_bs = NULL;
1765     Error *local_err = NULL;
1766     char *desc = NULL;
1767     int64_t total_size = 0, filesize;
1768     char *adapter_type = NULL;
1769     char *backing_file = NULL;
1770     char *fmt = NULL;
1771     int flags = 0;
1772     int ret = 0;
1773     bool flat, split, compress;
1774     GString *ext_desc_lines;
1775     char path[PATH_MAX], prefix[PATH_MAX], postfix[PATH_MAX];
1776     const int64_t split_size = 0x80000000;  /* VMDK has constant split size */
1777     const char *desc_extent_line;
1778     char parent_desc_line[BUF_SIZE] = "";
1779     uint32_t parent_cid = 0xffffffff;
1780     uint32_t number_heads = 16;
1781     bool zeroed_grain = false;
1782     uint32_t desc_offset = 0, desc_len;
1783     const char desc_template[] =
1784         "# Disk DescriptorFile\n"
1785         "version=1\n"
1786         "CID=%" PRIx32 "\n"
1787         "parentCID=%" PRIx32 "\n"
1788         "createType=\"%s\"\n"
1789         "%s"
1790         "\n"
1791         "# Extent description\n"
1792         "%s"
1793         "\n"
1794         "# The Disk Data Base\n"
1795         "#DDB\n"
1796         "\n"
1797         "ddb.virtualHWVersion = \"%d\"\n"
1798         "ddb.geometry.cylinders = \"%" PRId64 "\"\n"
1799         "ddb.geometry.heads = \"%" PRIu32 "\"\n"
1800         "ddb.geometry.sectors = \"63\"\n"
1801         "ddb.adapterType = \"%s\"\n";
1802
1803     ext_desc_lines = g_string_new(NULL);
1804
1805     if (filename_decompose(filename, path, prefix, postfix, PATH_MAX, errp)) {
1806         ret = -EINVAL;
1807         goto exit;
1808     }
1809     /* Read out options */
1810     total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
1811                           BDRV_SECTOR_SIZE);
1812     adapter_type = qemu_opt_get_del(opts, BLOCK_OPT_ADAPTER_TYPE);
1813     backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
1814     if (qemu_opt_get_bool_del(opts, BLOCK_OPT_COMPAT6, false)) {
1815         flags |= BLOCK_FLAG_COMPAT6;
1816     }
1817     fmt = qemu_opt_get_del(opts, BLOCK_OPT_SUBFMT);
1818     if (qemu_opt_get_bool_del(opts, BLOCK_OPT_ZEROED_GRAIN, false)) {
1819         zeroed_grain = true;
1820     }
1821
1822     if (!adapter_type) {
1823         adapter_type = g_strdup("ide");
1824     } else if (strcmp(adapter_type, "ide") &&
1825                strcmp(adapter_type, "buslogic") &&
1826                strcmp(adapter_type, "lsilogic") &&
1827                strcmp(adapter_type, "legacyESX")) {
1828         error_setg(errp, "Unknown adapter type: '%s'", adapter_type);
1829         ret = -EINVAL;
1830         goto exit;
1831     }
1832     if (strcmp(adapter_type, "ide") != 0) {
1833         /* that's the number of heads with which vmware operates when
1834            creating, exporting, etc. vmdk files with a non-ide adapter type */
1835         number_heads = 255;
1836     }
1837     if (!fmt) {
1838         /* Default format to monolithicSparse */
1839         fmt = g_strdup("monolithicSparse");
1840     } else if (strcmp(fmt, "monolithicFlat") &&
1841                strcmp(fmt, "monolithicSparse") &&
1842                strcmp(fmt, "twoGbMaxExtentSparse") &&
1843                strcmp(fmt, "twoGbMaxExtentFlat") &&
1844                strcmp(fmt, "streamOptimized")) {
1845         error_setg(errp, "Unknown subformat: '%s'", fmt);
1846         ret = -EINVAL;
1847         goto exit;
1848     }
1849     split = !(strcmp(fmt, "twoGbMaxExtentFlat") &&
1850               strcmp(fmt, "twoGbMaxExtentSparse"));
1851     flat = !(strcmp(fmt, "monolithicFlat") &&
1852              strcmp(fmt, "twoGbMaxExtentFlat"));
1853     compress = !strcmp(fmt, "streamOptimized");
1854     if (flat) {
1855         desc_extent_line = "RW %" PRId64 " FLAT \"%s\" 0\n";
1856     } else {
1857         desc_extent_line = "RW %" PRId64 " SPARSE \"%s\"\n";
1858     }
1859     if (flat && backing_file) {
1860         error_setg(errp, "Flat image can't have backing file");
1861         ret = -ENOTSUP;
1862         goto exit;
1863     }
1864     if (flat && zeroed_grain) {
1865         error_setg(errp, "Flat image can't enable zeroed grain");
1866         ret = -ENOTSUP;
1867         goto exit;
1868     }
1869     if (backing_file) {
1870         BlockDriverState *bs = NULL;
1871         ret = bdrv_open(&bs, backing_file, NULL, NULL, BDRV_O_NO_BACKING, NULL,
1872                         errp);
1873         if (ret != 0) {
1874             goto exit;
1875         }
1876         if (strcmp(bs->drv->format_name, "vmdk")) {
1877             bdrv_unref(bs);
1878             ret = -EINVAL;
1879             goto exit;
1880         }
1881         parent_cid = vmdk_read_cid(bs, 0);
1882         bdrv_unref(bs);
1883         snprintf(parent_desc_line, sizeof(parent_desc_line),
1884                 "parentFileNameHint=\"%s\"", backing_file);
1885     }
1886
1887     /* Create extents */
1888     filesize = total_size;
1889     while (filesize > 0) {
1890         char desc_line[BUF_SIZE];
1891         char ext_filename[PATH_MAX];
1892         char desc_filename[PATH_MAX];
1893         int64_t size = filesize;
1894
1895         if (split && size > split_size) {
1896             size = split_size;
1897         }
1898         if (split) {
1899             snprintf(desc_filename, sizeof(desc_filename), "%s-%c%03d%s",
1900                     prefix, flat ? 'f' : 's', ++idx, postfix);
1901         } else if (flat) {
1902             snprintf(desc_filename, sizeof(desc_filename), "%s-flat%s",
1903                     prefix, postfix);
1904         } else {
1905             snprintf(desc_filename, sizeof(desc_filename), "%s%s",
1906                     prefix, postfix);
1907         }
1908         snprintf(ext_filename, sizeof(ext_filename), "%s%s",
1909                 path, desc_filename);
1910
1911         if (vmdk_create_extent(ext_filename, size,
1912                                flat, compress, zeroed_grain, opts, errp)) {
1913             ret = -EINVAL;
1914             goto exit;
1915         }
1916         filesize -= size;
1917
1918         /* Format description line */
1919         snprintf(desc_line, sizeof(desc_line),
1920                     desc_extent_line, size / BDRV_SECTOR_SIZE, desc_filename);
1921         g_string_append(ext_desc_lines, desc_line);
1922     }
1923     /* generate descriptor file */
1924     desc = g_strdup_printf(desc_template,
1925                            (uint32_t)time(NULL),
1926                            parent_cid,
1927                            fmt,
1928                            parent_desc_line,
1929                            ext_desc_lines->str,
1930                            (flags & BLOCK_FLAG_COMPAT6 ? 6 : 4),
1931                            total_size /
1932                                (int64_t)(63 * number_heads * BDRV_SECTOR_SIZE),
1933                            number_heads,
1934                            adapter_type);
1935     desc_len = strlen(desc);
1936     /* the descriptor offset = 0x200 */
1937     if (!split && !flat) {
1938         desc_offset = 0x200;
1939     } else {
1940         ret = bdrv_create_file(filename, opts, &local_err);
1941         if (ret < 0) {
1942             error_propagate(errp, local_err);
1943             goto exit;
1944         }
1945     }
1946     assert(new_bs == NULL);
1947     ret = bdrv_open(&new_bs, filename, NULL, NULL,
1948                     BDRV_O_RDWR | BDRV_O_PROTOCOL, NULL, &local_err);
1949     if (ret < 0) {
1950         error_propagate(errp, local_err);
1951         goto exit;
1952     }
1953     ret = bdrv_pwrite(new_bs, desc_offset, desc, desc_len);
1954     if (ret < 0) {
1955         error_setg_errno(errp, -ret, "Could not write description");
1956         goto exit;
1957     }
1958     /* bdrv_pwrite write padding zeros to align to sector, we don't need that
1959      * for description file */
1960     if (desc_offset == 0) {
1961         ret = bdrv_truncate(new_bs, desc_len);
1962         if (ret < 0) {
1963             error_setg_errno(errp, -ret, "Could not truncate file");
1964         }
1965     }
1966 exit:
1967     if (new_bs) {
1968         bdrv_unref(new_bs);
1969     }
1970     g_free(adapter_type);
1971     g_free(backing_file);
1972     g_free(fmt);
1973     g_free(desc);
1974     g_string_free(ext_desc_lines, true);
1975     return ret;
1976 }
1977
1978 static void vmdk_close(BlockDriverState *bs)
1979 {
1980     BDRVVmdkState *s = bs->opaque;
1981
1982     vmdk_free_extents(bs);
1983     g_free(s->create_type);
1984
1985     migrate_del_blocker(s->migration_blocker);
1986     error_free(s->migration_blocker);
1987 }
1988
1989 static coroutine_fn int vmdk_co_flush(BlockDriverState *bs)
1990 {
1991     BDRVVmdkState *s = bs->opaque;
1992     int i, err;
1993     int ret = 0;
1994
1995     for (i = 0; i < s->num_extents; i++) {
1996         err = bdrv_co_flush(s->extents[i].file);
1997         if (err < 0) {
1998             ret = err;
1999         }
2000     }
2001     return ret;
2002 }
2003
2004 static int64_t vmdk_get_allocated_file_size(BlockDriverState *bs)
2005 {
2006     int i;
2007     int64_t ret = 0;
2008     int64_t r;
2009     BDRVVmdkState *s = bs->opaque;
2010
2011     ret = bdrv_get_allocated_file_size(bs->file);
2012     if (ret < 0) {
2013         return ret;
2014     }
2015     for (i = 0; i < s->num_extents; i++) {
2016         if (s->extents[i].file == bs->file) {
2017             continue;
2018         }
2019         r = bdrv_get_allocated_file_size(s->extents[i].file);
2020         if (r < 0) {
2021             return r;
2022         }
2023         ret += r;
2024     }
2025     return ret;
2026 }
2027
2028 static int vmdk_has_zero_init(BlockDriverState *bs)
2029 {
2030     int i;
2031     BDRVVmdkState *s = bs->opaque;
2032
2033     /* If has a flat extent and its underlying storage doesn't have zero init,
2034      * return 0. */
2035     for (i = 0; i < s->num_extents; i++) {
2036         if (s->extents[i].flat) {
2037             if (!bdrv_has_zero_init(s->extents[i].file)) {
2038                 return 0;
2039             }
2040         }
2041     }
2042     return 1;
2043 }
2044
2045 static ImageInfo *vmdk_get_extent_info(VmdkExtent *extent)
2046 {
2047     ImageInfo *info = g_new0(ImageInfo, 1);
2048
2049     *info = (ImageInfo){
2050         .filename         = g_strdup(extent->file->filename),
2051         .format           = g_strdup(extent->type),
2052         .virtual_size     = extent->sectors * BDRV_SECTOR_SIZE,
2053         .compressed       = extent->compressed,
2054         .has_compressed   = extent->compressed,
2055         .cluster_size     = extent->cluster_sectors * BDRV_SECTOR_SIZE,
2056         .has_cluster_size = !extent->flat,
2057     };
2058
2059     return info;
2060 }
2061
2062 static int vmdk_check(BlockDriverState *bs, BdrvCheckResult *result,
2063                       BdrvCheckMode fix)
2064 {
2065     BDRVVmdkState *s = bs->opaque;
2066     VmdkExtent *extent = NULL;
2067     int64_t sector_num = 0;
2068     int64_t total_sectors = bdrv_nb_sectors(bs);
2069     int ret;
2070     uint64_t cluster_offset;
2071
2072     if (fix) {
2073         return -ENOTSUP;
2074     }
2075
2076     for (;;) {
2077         if (sector_num >= total_sectors) {
2078             return 0;
2079         }
2080         extent = find_extent(s, sector_num, extent);
2081         if (!extent) {
2082             fprintf(stderr,
2083                     "ERROR: could not find extent for sector %" PRId64 "\n",
2084                     sector_num);
2085             break;
2086         }
2087         ret = get_cluster_offset(bs, extent, NULL,
2088                                  sector_num << BDRV_SECTOR_BITS,
2089                                  false, &cluster_offset, 0, 0);
2090         if (ret == VMDK_ERROR) {
2091             fprintf(stderr,
2092                     "ERROR: could not get cluster_offset for sector %"
2093                     PRId64 "\n", sector_num);
2094             break;
2095         }
2096         if (ret == VMDK_OK && cluster_offset >= bdrv_getlength(extent->file)) {
2097             fprintf(stderr,
2098                     "ERROR: cluster offset for sector %"
2099                     PRId64 " points after EOF\n", sector_num);
2100             break;
2101         }
2102         sector_num += extent->cluster_sectors;
2103     }
2104
2105     result->corruptions++;
2106     return 0;
2107 }
2108
2109 static ImageInfoSpecific *vmdk_get_specific_info(BlockDriverState *bs)
2110 {
2111     int i;
2112     BDRVVmdkState *s = bs->opaque;
2113     ImageInfoSpecific *spec_info = g_new0(ImageInfoSpecific, 1);
2114     ImageInfoList **next;
2115
2116     *spec_info = (ImageInfoSpecific){
2117         .kind = IMAGE_INFO_SPECIFIC_KIND_VMDK,
2118         {
2119             .vmdk = g_new0(ImageInfoSpecificVmdk, 1),
2120         },
2121     };
2122
2123     *spec_info->vmdk = (ImageInfoSpecificVmdk) {
2124         .create_type = g_strdup(s->create_type),
2125         .cid = s->cid,
2126         .parent_cid = s->parent_cid,
2127     };
2128
2129     next = &spec_info->vmdk->extents;
2130     for (i = 0; i < s->num_extents; i++) {
2131         *next = g_new0(ImageInfoList, 1);
2132         (*next)->value = vmdk_get_extent_info(&s->extents[i]);
2133         (*next)->next = NULL;
2134         next = &(*next)->next;
2135     }
2136
2137     return spec_info;
2138 }
2139
2140 static bool vmdk_extents_type_eq(const VmdkExtent *a, const VmdkExtent *b)
2141 {
2142     return a->flat == b->flat &&
2143            a->compressed == b->compressed &&
2144            (a->flat || a->cluster_sectors == b->cluster_sectors);
2145 }
2146
2147 static int vmdk_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2148 {
2149     int i;
2150     BDRVVmdkState *s = bs->opaque;
2151     assert(s->num_extents);
2152
2153     /* See if we have multiple extents but they have different cases */
2154     for (i = 1; i < s->num_extents; i++) {
2155         if (!vmdk_extents_type_eq(&s->extents[0], &s->extents[i])) {
2156             return -ENOTSUP;
2157         }
2158     }
2159     bdi->needs_compressed_writes = s->extents[0].compressed;
2160     if (!s->extents[0].flat) {
2161         bdi->cluster_size = s->extents[0].cluster_sectors << BDRV_SECTOR_BITS;
2162     }
2163     return 0;
2164 }
2165
2166 static void vmdk_detach_aio_context(BlockDriverState *bs)
2167 {
2168     BDRVVmdkState *s = bs->opaque;
2169     int i;
2170
2171     for (i = 0; i < s->num_extents; i++) {
2172         bdrv_detach_aio_context(s->extents[i].file);
2173     }
2174 }
2175
2176 static void vmdk_attach_aio_context(BlockDriverState *bs,
2177                                     AioContext *new_context)
2178 {
2179     BDRVVmdkState *s = bs->opaque;
2180     int i;
2181
2182     for (i = 0; i < s->num_extents; i++) {
2183         bdrv_attach_aio_context(s->extents[i].file, new_context);
2184     }
2185 }
2186
2187 static QemuOptsList vmdk_create_opts = {
2188     .name = "vmdk-create-opts",
2189     .head = QTAILQ_HEAD_INITIALIZER(vmdk_create_opts.head),
2190     .desc = {
2191         {
2192             .name = BLOCK_OPT_SIZE,
2193             .type = QEMU_OPT_SIZE,
2194             .help = "Virtual disk size"
2195         },
2196         {
2197             .name = BLOCK_OPT_ADAPTER_TYPE,
2198             .type = QEMU_OPT_STRING,
2199             .help = "Virtual adapter type, can be one of "
2200                     "ide (default), lsilogic, buslogic or legacyESX"
2201         },
2202         {
2203             .name = BLOCK_OPT_BACKING_FILE,
2204             .type = QEMU_OPT_STRING,
2205             .help = "File name of a base image"
2206         },
2207         {
2208             .name = BLOCK_OPT_COMPAT6,
2209             .type = QEMU_OPT_BOOL,
2210             .help = "VMDK version 6 image",
2211             .def_value_str = "off"
2212         },
2213         {
2214             .name = BLOCK_OPT_SUBFMT,
2215             .type = QEMU_OPT_STRING,
2216             .help =
2217                 "VMDK flat extent format, can be one of "
2218                 "{monolithicSparse (default) | monolithicFlat | twoGbMaxExtentSparse | twoGbMaxExtentFlat | streamOptimized} "
2219         },
2220         {
2221             .name = BLOCK_OPT_ZEROED_GRAIN,
2222             .type = QEMU_OPT_BOOL,
2223             .help = "Enable efficient zero writes "
2224                     "using the zeroed-grain GTE feature"
2225         },
2226         { /* end of list */ }
2227     }
2228 };
2229
2230 static BlockDriver bdrv_vmdk = {
2231     .format_name                  = "vmdk",
2232     .instance_size                = sizeof(BDRVVmdkState),
2233     .bdrv_probe                   = vmdk_probe,
2234     .bdrv_open                    = vmdk_open,
2235     .bdrv_check                   = vmdk_check,
2236     .bdrv_reopen_prepare          = vmdk_reopen_prepare,
2237     .bdrv_read                    = vmdk_co_read,
2238     .bdrv_write                   = vmdk_co_write,
2239     .bdrv_write_compressed        = vmdk_write_compressed,
2240     .bdrv_co_write_zeroes         = vmdk_co_write_zeroes,
2241     .bdrv_close                   = vmdk_close,
2242     .bdrv_create                  = vmdk_create,
2243     .bdrv_co_flush_to_disk        = vmdk_co_flush,
2244     .bdrv_co_get_block_status     = vmdk_co_get_block_status,
2245     .bdrv_get_allocated_file_size = vmdk_get_allocated_file_size,
2246     .bdrv_has_zero_init           = vmdk_has_zero_init,
2247     .bdrv_get_specific_info       = vmdk_get_specific_info,
2248     .bdrv_refresh_limits          = vmdk_refresh_limits,
2249     .bdrv_get_info                = vmdk_get_info,
2250     .bdrv_detach_aio_context      = vmdk_detach_aio_context,
2251     .bdrv_attach_aio_context      = vmdk_attach_aio_context,
2252
2253     .supports_backing             = true,
2254     .create_opts                  = &vmdk_create_opts,
2255 };
2256
2257 static void bdrv_vmdk_init(void)
2258 {
2259     bdrv_register(&bdrv_vmdk);
2260 }
2261
2262 block_init(bdrv_vmdk_init);