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 #include <glib.h>
  32
  33 #define VMDK3_MAGIC (('C' << 24) | ('O' << 16) | ('W' << 8) | 'D')
  34 #define VMDK4_MAGIC (('K' << 24) | ('D' << 16) | ('M' << 8) | 'V')
  35 #define VMDK4_COMPRESSION_DEFLATE 1
  36 #define VMDK4_FLAG_NL_DETECT (1 << 0)
  37 #define VMDK4_FLAG_RGD (1 << 1)
  38 /* Zeroed-grain enable bit */
  39 #define VMDK4_FLAG_ZERO_GRAIN   (1 << 2)
  40 #define VMDK4_FLAG_COMPRESS (1 << 16)
  41 #define VMDK4_FLAG_MARKER (1 << 17)
  42 #define VMDK4_GD_AT_END 0xffffffffffffffffULL
  43
  44 #define VMDK_GTE_ZEROED 0x1
  45
  46 /* VMDK internal error codes */
  47 #define VMDK_OK      0
  48 #define VMDK_ERROR   (-1)
  49 /* Cluster not allocated */
  50 #define VMDK_UNALLOC (-2)
  51 #define VMDK_ZEROED  (-3)
  52
  53 #define BLOCK_OPT_ZEROED_GRAIN "zeroed_grain"
  54
  55 typedef struct {
  56     uint32_t version;
  57     uint32_t flags;
  58     uint32_t disk_sectors;
  59     uint32_t granularity;
  60     uint32_t l1dir_offset;
  61     uint32_t l1dir_size;
  62     uint32_t file_sectors;
  63     uint32_t cylinders;
  64     uint32_t heads;
  65     uint32_t sectors_per_track;
  66 } QEMU_PACKED VMDK3Header;
  67
  68 typedef struct {
  69     uint32_t version;
  70     uint32_t flags;
  71     uint64_t capacity;
  72     uint64_t granularity;
  73     uint64_t desc_offset;
  74     uint64_t desc_size;
  75     /* Number of GrainTableEntries per GrainTable */
  76     uint32_t num_gtes_per_gt;
  77     uint64_t rgd_offset;
  78     uint64_t gd_offset;
  79     uint64_t grain_offset;
  80     char filler[1];
  81     char check_bytes[4];
  82     uint16_t compressAlgorithm;
  83 } QEMU_PACKED VMDK4Header;
  84
  85 #define L2_CACHE_SIZE 16
  86
  87 typedef struct VmdkExtent {
  88     BlockDriverState *file;
  89     bool flat;
  90     bool compressed;
  91     bool has_marker;
  92     bool has_zero_grain;
  93     int version;
  94     int64_t sectors;
  95     int64_t end_sector;
  96     int64_t flat_start_offset;
  97     int64_t l1_table_offset;
  98     int64_t l1_backup_table_offset;
  99     uint32_t *l1_table;
 100     uint32_t *l1_backup_table;
 101     unsigned int l1_size;
 102     uint32_t l1_entry_sectors;
 103
 104     unsigned int l2_size;
 105     uint32_t *l2_cache;
 106     uint32_t l2_cache_offsets[L2_CACHE_SIZE];
 107     uint32_t l2_cache_counts[L2_CACHE_SIZE];
 108
 109     int64_t cluster_sectors;
 110     int64_t next_cluster_sector;
 111     char *type;
 112 } VmdkExtent;
 113
 114 typedef struct BDRVVmdkState {
 115     CoMutex lock;
 116     uint64_t desc_offset;
 117     bool cid_updated;
 118     bool cid_checked;
 119     uint32_t cid;
 120     uint32_t parent_cid;
 121     int num_extents;
 122     /* Extent array with num_extents entries, ascend ordered by address */
 123     VmdkExtent *extents;
 124     Error *migration_blocker;
 125     char *create_type;
 126 } BDRVVmdkState;
 127
 128 typedef struct VmdkMetaData {
 129     unsigned int l1_index;
 130     unsigned int l2_index;
 131     unsigned int l2_offset;
 132     int valid;
 133     uint32_t *l2_cache_entry;
 134 } VmdkMetaData;
 135
 136 typedef struct VmdkGrainMarker {
 137     uint64_t lba;
 138     uint32_t size;
 139     uint8_t  data[0];
 140 } QEMU_PACKED VmdkGrainMarker;
 141
 142 enum {
 143     MARKER_END_OF_STREAM    = 0,
 144     MARKER_GRAIN_TABLE      = 1,
 145     MARKER_GRAIN_DIRECTORY  = 2,
 146     MARKER_FOOTER           = 3,
 147 };
 148
 149 static int vmdk_probe(const uint8_t *buf, int buf_size, const char *filename)
 150 {
 151     uint32_t magic;
 152
 153     if (buf_size < 4) {
 154         return 0;
 155     }
 156     magic = be32_to_cpu(*(uint32_t *)buf);
 157     if (magic == VMDK3_MAGIC ||
 158         magic == VMDK4_MAGIC) {
 159         return 100;
 160     } else {
 161         const char *p = (const char *)buf;
 162         const char *end = p + buf_size;
 163         while (p < end) {
 164             if (*p == '#') {
 165                 /* skip comment line */
 166                 while (p < end && *p != '\n') {
 167                     p++;
 168                 }
 169                 p++;
 170                 continue;
 171             }
 172             if (*p == ' ') {
 173                 while (p < end && *p == ' ') {
 174                     p++;
 175                 }
 176                 /* skip '\r' if windows line endings used. */
 177                 if (p < end && *p == '\r') {
 178                     p++;
 179                 }
 180                 /* only accept blank lines before 'version=' line */
 181                 if (p == end || *p != '\n') {
 182                     return 0;
 183                 }
 184                 p++;
 185                 continue;
 186             }
 187             if (end - p >= strlen("version=X\n")) {
 188                 if (strncmp("version=1\n", p, strlen("version=1\n")) == 0 ||
 189                     strncmp("version=2\n", p, strlen("version=2\n")) == 0) {
 190                     return 100;
 191                 }
 192             }
 193             if (end - p >= strlen("version=X\r\n")) {
 194                 if (strncmp("version=1\r\n", p, strlen("version=1\r\n")) == 0 ||
 195                     strncmp("version=2\r\n", p, strlen("version=2\r\n")) == 0) {
 196                     return 100;
 197                 }
 198             }
 199             return 0;
 200         }
 201         return 0;
 202     }
 203 }
 204
 205 #define SECTOR_SIZE 512
 206 #define DESC_SIZE (20 * SECTOR_SIZE)    /* 20 sectors of 512 bytes each */
 207 #define BUF_SIZE 4096
 208 #define HEADER_SIZE 512                 /* first sector of 512 bytes */
 209
 210 static void vmdk_free_extents(BlockDriverState *bs)
 211 {
 212     int i;
 213     BDRVVmdkState *s = bs->opaque;
 214     VmdkExtent *e;
 215
 216     for (i = 0; i < s->num_extents; i++) {
 217         e = &s->extents[i];
 218         g_free(e->l1_table);
 219         g_free(e->l2_cache);
 220         g_free(e->l1_backup_table);
 221         g_free(e->type);
 222         if (e->file != bs->file) {
 223             bdrv_unref(e->file);
 224         }
 225     }
 226     g_free(s->extents);
 227 }
 228
 229 static void vmdk_free_last_extent(BlockDriverState *bs)
 230 {
 231     BDRVVmdkState *s = bs->opaque;
 232
 233     if (s->num_extents == 0) {
 234         return;
 235     }
 236     s->num_extents--;
 237     s->extents = g_renew(VmdkExtent, s->extents, s->num_extents);
 238 }
 239
 240 static uint32_t vmdk_read_cid(BlockDriverState *bs, int parent)
 241 {
 242     char desc[DESC_SIZE];
 243     uint32_t cid = 0xffffffff;
 244     const char *p_name, *cid_str;
 245     size_t cid_str_size;
 246     BDRVVmdkState *s = bs->opaque;
 247     int ret;
 248
 249     ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE);
 250     if (ret < 0) {
 251         return 0;
 252     }
 253
 254     if (parent) {
 255         cid_str = "parentCID";
 256         cid_str_size = sizeof("parentCID");
 257     } else {
 258         cid_str = "CID";
 259         cid_str_size = sizeof("CID");
 260     }
 261
 262     desc[DESC_SIZE - 1] = '\0';
 263     p_name = strstr(desc, cid_str);
 264     if (p_name != NULL) {
 265         p_name += cid_str_size;
 266         sscanf(p_name, "%" SCNx32, &cid);
 267     }
 268
 269     return cid;
 270 }
 271
 272 static int vmdk_write_cid(BlockDriverState *bs, uint32_t cid)
 273 {
 274     char desc[DESC_SIZE], tmp_desc[DESC_SIZE];
 275     char *p_name, *tmp_str;
 276     BDRVVmdkState *s = bs->opaque;
 277     int ret;
 278
 279     ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE);
 280     if (ret < 0) {
 281         return ret;
 282     }
 283
 284     desc[DESC_SIZE - 1] = '\0';
 285     tmp_str = strstr(desc, "parentCID");
 286     if (tmp_str == NULL) {
 287         return -EINVAL;
 288     }
 289
 290     pstrcpy(tmp_desc, sizeof(tmp_desc), tmp_str);
 291     p_name = strstr(desc, "CID");
 292     if (p_name != NULL) {
 293         p_name += sizeof("CID");
 294         snprintf(p_name, sizeof(desc) - (p_name - desc), "%" PRIx32 "\n", cid);
 295         pstrcat(desc, sizeof(desc), tmp_desc);
 296     }
 297
 298     ret = bdrv_pwrite_sync(bs->file, s->desc_offset, desc, DESC_SIZE);
 299     if (ret < 0) {
 300         return ret;
 301     }
 302
 303     return 0;
 304 }
 305
 306 static int vmdk_is_cid_valid(BlockDriverState *bs)
 307 {
 308     BDRVVmdkState *s = bs->opaque;
 309     BlockDriverState *p_bs = bs->backing_hd;
 310     uint32_t cur_pcid;
 311
 312     if (!s->cid_checked && p_bs) {
 313         cur_pcid = vmdk_read_cid(p_bs, 0);
 314         if (s->parent_cid != cur_pcid) {
 315             /* CID not valid */
 316             return 0;
 317         }
 318     }
 319     s->cid_checked = true;
 320     /* CID valid */
 321     return 1;
 322 }
 323
 324 /* Queue extents, if any, for reopen() */
 325 static int vmdk_reopen_prepare(BDRVReopenState *state,
 326                                BlockReopenQueue *queue, Error **errp)
 327 {
 328     BDRVVmdkState *s;
 329     int ret = -1;
 330     int i;
 331     VmdkExtent *e;
 332
 333     assert(state != NULL);
 334     assert(state->bs != NULL);
 335
 336     if (queue == NULL) {
 337         error_setg(errp, "No reopen queue for VMDK extents");
 338         goto exit;
 339     }
 340
 341     s = state->bs->opaque;
 342
 343     assert(s != NULL);
 344
 345     for (i = 0; i < s->num_extents; i++) {
 346         e = &s->extents[i];
 347         if (e->file != state->bs->file) {
 348             bdrv_reopen_queue(queue, e->file, state->flags);
 349         }
 350     }
 351     ret = 0;
 352
 353 exit:
 354     return ret;
 355 }
 356
 357 static int vmdk_parent_open(BlockDriverState *bs)
 358 {
 359     char *p_name;
 360     char desc[DESC_SIZE + 1];
 361     BDRVVmdkState *s = bs->opaque;
 362     int ret;
 363
 364     desc[DESC_SIZE] = '\0';
 365     ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE);
 366     if (ret < 0) {
 367         return ret;
 368     }
 369
 370     p_name = strstr(desc, "parentFileNameHint");
 371     if (p_name != NULL) {
 372         char *end_name;
 373
 374         p_name += sizeof("parentFileNameHint") + 1;
 375         end_name = strchr(p_name, '\"');
 376         if (end_name == NULL) {
 377             return -EINVAL;
 378         }
 379         if ((end_name - p_name) > sizeof(bs->backing_file) - 1) {
 380             return -EINVAL;
 381         }
 382
 383         pstrcpy(bs->backing_file, end_name - p_name + 1, p_name);
 384     }
 385
 386     return 0;
 387 }
 388
 389 /* Create and append extent to the extent array. Return the added VmdkExtent
 390  * address. return NULL if allocation failed. */
 391 static int vmdk_add_extent(BlockDriverState *bs,
 392                            BlockDriverState *file, bool flat, int64_t sectors,
 393                            int64_t l1_offset, int64_t l1_backup_offset,
 394                            uint32_t l1_size,
 395                            int l2_size, uint64_t cluster_sectors,
 396                            VmdkExtent **new_extent,
 397                            Error **errp)
 398 {
 399     VmdkExtent *extent;
 400     BDRVVmdkState *s = bs->opaque;
 401     int64_t nb_sectors;
 402
 403     if (cluster_sectors > 0x200000) {
 404         /* 0x200000 * 512Bytes = 1GB for one cluster is unrealistic */
 405         error_setg(errp, "Invalid granularity, image may be corrupt");
 406         return -EFBIG;
 407     }
 408     if (l1_size > 512 * 1024 * 1024) {
 409         /* Although with big capacity and small l1_entry_sectors, we can get a
 410          * big l1_size, we don't want unbounded value to allocate the table.
 411          * Limit it to 512M, which is 16PB for default cluster and L2 table
 412          * size */
 413         error_setg(errp, "L1 size too big");
 414         return -EFBIG;
 415     }
 416
 417     nb_sectors = bdrv_nb_sectors(file);
 418     if (nb_sectors < 0) {
 419         return nb_sectors;
 420     }
 421
 422     s->extents = g_renew(VmdkExtent, s->extents, s->num_extents + 1);
 423     extent = &s->extents[s->num_extents];
 424     s->num_extents++;
 425
 426     memset(extent, 0, sizeof(VmdkExtent));
 427     extent->file = file;
 428     extent->flat = flat;
 429     extent->sectors = sectors;
 430     extent->l1_table_offset = l1_offset;
 431     extent->l1_backup_table_offset = l1_backup_offset;
 432     extent->l1_size = l1_size;
 433     extent->l1_entry_sectors = l2_size * cluster_sectors;
 434     extent->l2_size = l2_size;
 435     extent->cluster_sectors = flat ? sectors : cluster_sectors;
 436     extent->next_cluster_sector = ROUND_UP(nb_sectors, cluster_sectors);
 437
 438     if (s->num_extents > 1) {
 439         extent->end_sector = (*(extent - 1)).end_sector + extent->sectors;
 440     } else {
 441         extent->end_sector = extent->sectors;
 442     }
 443     bs->total_sectors = extent->end_sector;
 444     if (new_extent) {
 445         *new_extent = extent;
 446     }
 447     return 0;
 448 }
 449
 450 static int vmdk_init_tables(BlockDriverState *bs, VmdkExtent *extent,
 451                             Error **errp)
 452 {
 453     int ret;
 454     int l1_size, i;
 455
 456     /* read the L1 table */
 457     l1_size = extent->l1_size * sizeof(uint32_t);
 458     extent->l1_table = g_try_malloc(l1_size);
 459     if (l1_size && extent->l1_table == NULL) {
 460         return -ENOMEM;
 461     }
 462
 463     ret = bdrv_pread(extent->file,
 464                      extent->l1_table_offset,
 465                      extent->l1_table,
 466                      l1_size);
 467     if (ret < 0) {
 468         error_setg_errno(errp, -ret,
 469                          "Could not read l1 table from extent '%s'",
 470                          extent->file->filename);
 471         goto fail_l1;
 472     }
 473     for (i = 0; i < extent->l1_size; i++) {
 474         le32_to_cpus(&extent->l1_table[i]);
 475     }
 476
 477     if (extent->l1_backup_table_offset) {
 478         extent->l1_backup_table = g_try_malloc(l1_size);
 479         if (l1_size && extent->l1_backup_table == NULL) {
 480             ret = -ENOMEM;
 481             goto fail_l1;
 482         }
 483         ret = bdrv_pread(extent->file,
 484                          extent->l1_backup_table_offset,
 485                          extent->l1_backup_table,
 486                          l1_size);
 487         if (ret < 0) {
 488             error_setg_errno(errp, -ret,
 489                              "Could not read l1 backup table from extent '%s'",
 490                              extent->file->filename);
 491             goto fail_l1b;
 492         }
 493         for (i = 0; i < extent->l1_size; i++) {
 494             le32_to_cpus(&extent->l1_backup_table[i]);
 495         }
 496     }
 497
 498     extent->l2_cache =
 499         g_new(uint32_t, extent->l2_size * L2_CACHE_SIZE);
 500     return 0;
 501  fail_l1b:
 502     g_free(extent->l1_backup_table);
 503  fail_l1:
 504     g_free(extent->l1_table);
 505     return ret;
 506 }
 507
 508 static int vmdk_open_vmfs_sparse(BlockDriverState *bs,
 509                                  BlockDriverState *file,
 510                                  int flags, Error **errp)
 511 {
 512     int ret;
 513     uint32_t magic;
 514     VMDK3Header header;
 515     VmdkExtent *extent;
 516
 517     ret = bdrv_pread(file, sizeof(magic), &header, sizeof(header));
 518     if (ret < 0) {
 519         error_setg_errno(errp, -ret,
 520                          "Could not read header from file '%s'",
 521                          file->filename);
 522         return ret;
 523     }
 524     ret = vmdk_add_extent(bs, file, false,
 525                           le32_to_cpu(header.disk_sectors),
 526                           le32_to_cpu(header.l1dir_offset) << 9,
 527                           0,
 528                           le32_to_cpu(header.l1dir_size),
 529                           4096,
 530                           le32_to_cpu(header.granularity),
 531                           &extent,
 532                           errp);
 533     if (ret < 0) {
 534         return ret;
 535     }
 536     ret = vmdk_init_tables(bs, extent, errp);
 537     if (ret) {
 538         /* free extent allocated by vmdk_add_extent */
 539         vmdk_free_last_extent(bs);
 540     }
 541     return ret;
 542 }
 543
 544 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf,
 545                                Error **errp);
 546
 547 static char *vmdk_read_desc(BlockDriverState *file, uint64_t desc_offset,
 548                             Error **errp)
 549 {
 550     int64_t size;
 551     char *buf;
 552     int ret;
 553
 554     size = bdrv_getlength(file);
 555     if (size < 0) {
 556         error_setg_errno(errp, -size, "Could not access file");
 557         return NULL;
 558     }
 559
 560     if (size < 4) {
 561         /* Both descriptor file and sparse image must be much larger than 4
 562          * bytes, also callers of vmdk_read_desc want to compare the first 4
 563          * bytes with VMDK4_MAGIC, let's error out if less is read. */
 564         error_setg(errp, "File is too small, not a valid image");
 565         return NULL;
 566     }
 567
 568     size = MIN(size, (1 << 20) - 1);  /* avoid unbounded allocation */
 569     buf = g_malloc(size + 1);
 570
 571     ret = bdrv_pread(file, desc_offset, buf, size);
 572     if (ret < 0) {
 573         error_setg_errno(errp, -ret, "Could not read from file");
 574         g_free(buf);
 575         return NULL;
 576     }
 577     buf[ret] = 0;
 578
 579     return buf;
 580 }
 581
 582 static int vmdk_open_vmdk4(BlockDriverState *bs,
 583                            BlockDriverState *file,
 584                            int flags, Error **errp)
 585 {
 586     int ret;
 587     uint32_t magic;
 588     uint32_t l1_size, l1_entry_sectors;
 589     VMDK4Header header;
 590     VmdkExtent *extent;
 591     BDRVVmdkState *s = bs->opaque;
 592     int64_t l1_backup_offset = 0;
 593
 594     ret = bdrv_pread(file, sizeof(magic), &header, sizeof(header));
 595     if (ret < 0) {
 596         error_setg_errno(errp, -ret,
 597                          "Could not read header from file '%s'",
 598                          file->filename);
 599         return -EINVAL;
 600     }
 601     if (header.capacity == 0) {
 602         uint64_t desc_offset = le64_to_cpu(header.desc_offset);
 603         if (desc_offset) {
 604             char *buf = vmdk_read_desc(file, desc_offset << 9, errp);
 605             if (!buf) {
 606                 return -EINVAL;
 607             }
 608             ret = vmdk_open_desc_file(bs, flags, buf, errp);
 609             g_free(buf);
 610             return ret;
 611         }
 612     }
 613
 614     if (!s->create_type) {
 615         s->create_type = g_strdup("monolithicSparse");
 616     }
 617
 618     if (le64_to_cpu(header.gd_offset) == VMDK4_GD_AT_END) {
 619         /*
 620          * The footer takes precedence over the header, so read it in. The
 621          * footer starts at offset -1024 from the end: One sector for the
 622          * footer, and another one for the end-of-stream marker.
 623          */
 624         struct {
 625             struct {
 626                 uint64_t val;
 627                 uint32_t size;
 628                 uint32_t type;
 629                 uint8_t pad[512 - 16];
 630             } QEMU_PACKED footer_marker;
 631
 632             uint32_t magic;
 633             VMDK4Header header;
 634             uint8_t pad[512 - 4 - sizeof(VMDK4Header)];
 635
 636             struct {
 637                 uint64_t val;
 638                 uint32_t size;
 639                 uint32_t type;
 640                 uint8_t pad[512 - 16];
 641             } QEMU_PACKED eos_marker;
 642         } QEMU_PACKED footer;
 643
 644         ret = bdrv_pread(file,
 645             bs->file->total_sectors * 512 - 1536,
 646             &footer, sizeof(footer));
 647         if (ret < 0) {
 648             error_setg_errno(errp, -ret, "Failed to read footer");
 649             return ret;
 650         }
 651
 652         /* Some sanity checks for the footer */
 653         if (be32_to_cpu(footer.magic) != VMDK4_MAGIC ||
 654             le32_to_cpu(footer.footer_marker.size) != 0  ||
 655             le32_to_cpu(footer.footer_marker.type) != MARKER_FOOTER ||
 656             le64_to_cpu(footer.eos_marker.val) != 0  ||
 657             le32_to_cpu(footer.eos_marker.size) != 0  ||
 658             le32_to_cpu(footer.eos_marker.type) != MARKER_END_OF_STREAM)
 659         {
 660             error_setg(errp, "Invalid footer");
 661             return -EINVAL;
 662         }
 663
 664         header = footer.header;
 665     }
 666
 667     if (le32_to_cpu(header.version) > 3) {
 668         char buf[64];
 669         snprintf(buf, sizeof(buf), "VMDK version %" PRId32,
 670                  le32_to_cpu(header.version));
 671         error_set(errp, QERR_UNKNOWN_BLOCK_FORMAT_FEATURE,
 672                   bdrv_get_device_name(bs), "vmdk", buf);
 673         return -ENOTSUP;
 674     } else if (le32_to_cpu(header.version) == 3 && (flags & BDRV_O_RDWR)) {
 675         /* VMware KB 2064959 explains that version 3 added support for
 676          * persistent changed block tracking (CBT), and backup software can
 677          * read it as version=1 if it doesn't care about the changed area
 678          * information. So we are safe to enable read only. */
 679         error_setg(errp, "VMDK version 3 must be read only");
 680         return -EINVAL;
 681     }
 682
 683     if (le32_to_cpu(header.num_gtes_per_gt) > 512) {
 684         error_setg(errp, "L2 table size too big");
 685         return -EINVAL;
 686     }
 687
 688     l1_entry_sectors = le32_to_cpu(header.num_gtes_per_gt)
 689                         * le64_to_cpu(header.granularity);
 690     if (l1_entry_sectors == 0) {
 691         error_setg(errp, "L1 entry size is invalid");
 692         return -EINVAL;
 693     }
 694     l1_size = (le64_to_cpu(header.capacity) + l1_entry_sectors - 1)
 695                 / l1_entry_sectors;
 696     if (le32_to_cpu(header.flags) & VMDK4_FLAG_RGD) {
 697         l1_backup_offset = le64_to_cpu(header.rgd_offset) << 9;
 698     }
 699     if (bdrv_nb_sectors(file) < le64_to_cpu(header.grain_offset)) {
 700         error_setg(errp, "File truncated, expecting at least %" PRId64 " bytes",
 701                    (int64_t)(le64_to_cpu(header.grain_offset)
 702                              * BDRV_SECTOR_SIZE));
 703         return -EINVAL;
 704     }
 705
 706     ret = vmdk_add_extent(bs, file, false,
 707                           le64_to_cpu(header.capacity),
 708                           le64_to_cpu(header.gd_offset) << 9,
 709                           l1_backup_offset,
 710                           l1_size,
 711                           le32_to_cpu(header.num_gtes_per_gt),
 712                           le64_to_cpu(header.granularity),
 713                           &extent,
 714                           errp);
 715     if (ret < 0) {
 716         return ret;
 717     }
 718     extent->compressed =
 719         le16_to_cpu(header.compressAlgorithm) == VMDK4_COMPRESSION_DEFLATE;
 720     if (extent->compressed) {
 721         g_free(s->create_type);
 722         s->create_type = g_strdup("streamOptimized");
 723     }
 724     extent->has_marker = le32_to_cpu(header.flags) & VMDK4_FLAG_MARKER;
 725     extent->version = le32_to_cpu(header.version);
 726     extent->has_zero_grain = le32_to_cpu(header.flags) & VMDK4_FLAG_ZERO_GRAIN;
 727     ret = vmdk_init_tables(bs, extent, errp);
 728     if (ret) {
 729         /* free extent allocated by vmdk_add_extent */
 730         vmdk_free_last_extent(bs);
 731     }
 732     return ret;
 733 }
 734
 735 /* find an option value out of descriptor file */
 736 static int vmdk_parse_description(const char *desc, const char *opt_name,
 737         char *buf, int buf_size)
 738 {
 739     char *opt_pos, *opt_end;
 740     const char *end = desc + strlen(desc);
 741
 742     opt_pos = strstr(desc, opt_name);
 743     if (!opt_pos) {
 744         return VMDK_ERROR;
 745     }
 746     /* Skip "=\"" following opt_name */
 747     opt_pos += strlen(opt_name) + 2;
 748     if (opt_pos >= end) {
 749         return VMDK_ERROR;
 750     }
 751     opt_end = opt_pos;
 752     while (opt_end < end && *opt_end != '"') {
 753         opt_end++;
 754     }
 755     if (opt_end == end || buf_size < opt_end - opt_pos + 1) {
 756         return VMDK_ERROR;
 757     }
 758     pstrcpy(buf, opt_end - opt_pos + 1, opt_pos);
 759     return VMDK_OK;
 760 }
 761
 762 /* Open an extent file and append to bs array */
 763 static int vmdk_open_sparse(BlockDriverState *bs,
 764                             BlockDriverState *file, int flags,
 765                             char *buf, Error **errp)
 766 {
 767     uint32_t magic;
 768
 769     magic = ldl_be_p(buf);
 770     switch (magic) {
 771         case VMDK3_MAGIC:
 772             return vmdk_open_vmfs_sparse(bs, file, flags, errp);
 773             break;
 774         case VMDK4_MAGIC:
 775             return vmdk_open_vmdk4(bs, file, flags, errp);
 776             break;
 777         default:
 778             error_setg(errp, "Image not in VMDK format");
 779             return -EINVAL;
 780             break;
 781     }
 782 }
 783
 784 static int vmdk_parse_extents(const char *desc, BlockDriverState *bs,
 785                               const char *desc_file_path, Error **errp)
 786 {
 787     int ret;
 788     char access[11];
 789     char type[11];
 790     char fname[512];
 791     const char *p = desc;
 792     int64_t sectors = 0;
 793     int64_t flat_offset;
 794     char extent_path[PATH_MAX];
 795     BlockDriverState *extent_file;
 796     BDRVVmdkState *s = bs->opaque;
 797     VmdkExtent *extent;
 798
 799     while (*p) {
 800         /* parse extent line in one of below formats:
 801          *
 802          * RW [size in sectors] FLAT "file-name.vmdk" OFFSET
 803          * RW [size in sectors] SPARSE "file-name.vmdk"
 804          * RW [size in sectors] VMFS "file-name.vmdk"
 805          * RW [size in sectors] VMFSSPARSE "file-name.vmdk"
 806          */
 807         flat_offset = -1;
 808         ret = sscanf(p, "%10s %" SCNd64 " %10s \"%511[^\n\r\"]\" %" SCNd64,
 809                 access, &sectors, type, fname, &flat_offset);
 810         if (ret < 4 || strcmp(access, "RW")) {
 811             goto next_line;
 812         } else if (!strcmp(type, "FLAT")) {
 813             if (ret != 5 || flat_offset < 0) {
 814                 error_setg(errp, "Invalid extent lines: \n%s", p);
 815                 return -EINVAL;
 816             }
 817         } else if (!strcmp(type, "VMFS")) {
 818             if (ret == 4) {
 819                 flat_offset = 0;
 820             } else {
 821                 error_setg(errp, "Invalid extent lines:\n%s", p);
 822                 return -EINVAL;
 823             }
 824         } else if (ret != 4) {
 825             error_setg(errp, "Invalid extent lines:\n%s", p);
 826             return -EINVAL;
 827         }
 828
 829         if (sectors <= 0 ||
 830             (strcmp(type, "FLAT") && strcmp(type, "SPARSE") &&
 831              strcmp(type, "VMFS") && strcmp(type, "VMFSSPARSE")) ||
 832             (strcmp(access, "RW"))) {
 833             goto next_line;
 834         }
 835
 836         if (!path_is_absolute(fname) && !path_has_protocol(fname) &&
 837             !desc_file_path[0])
 838         {
 839             error_setg(errp, "Cannot use relative extent paths with VMDK "
 840                        "descriptor file '%s'", bs->file->filename);
 841             return -EINVAL;
 842         }
 843
 844         path_combine(extent_path, sizeof(extent_path),
 845                 desc_file_path, fname);
 846         extent_file = NULL;
 847         ret = bdrv_open(&extent_file, extent_path, NULL, NULL,
 848                         bs->open_flags | BDRV_O_PROTOCOL, NULL, errp);
 849         if (ret) {
 850             return ret;
 851         }
 852
 853         /* save to extents array */
 854         if (!strcmp(type, "FLAT") || !strcmp(type, "VMFS")) {
 855             /* FLAT extent */
 856
 857             ret = vmdk_add_extent(bs, extent_file, true, sectors,
 858                             0, 0, 0, 0, 0, &extent, errp);
 859             if (ret < 0) {
 860                 bdrv_unref(extent_file);
 861                 return ret;
 862             }
 863             extent->flat_start_offset = flat_offset << 9;
 864         } else if (!strcmp(type, "SPARSE") || !strcmp(type, "VMFSSPARSE")) {
 865             /* SPARSE extent and VMFSSPARSE extent are both "COWD" sparse file*/
 866             char *buf = vmdk_read_desc(extent_file, 0, errp);
 867             if (!buf) {
 868                 ret = -EINVAL;
 869             } else {
 870                 ret = vmdk_open_sparse(bs, extent_file, bs->open_flags, buf, errp);
 871             }
 872             g_free(buf);
 873             if (ret) {
 874                 bdrv_unref(extent_file);
 875                 return ret;
 876             }
 877             extent = &s->extents[s->num_extents - 1];
 878         } else {
 879             error_setg(errp, "Unsupported extent type '%s'", type);
 880             bdrv_unref(extent_file);
 881             return -ENOTSUP;
 882         }
 883         extent->type = g_strdup(type);
 884 next_line:
 885         /* move to next line */
 886         while (*p) {
 887             if (*p == '\n') {
 888                 p++;
 889                 break;
 890             }
 891             p++;
 892         }
 893     }
 894     return 0;
 895 }
 896
 897 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf,
 898                                Error **errp)
 899 {
 900     int ret;
 901     char ct[128];
 902     BDRVVmdkState *s = bs->opaque;
 903
 904     if (vmdk_parse_description(buf, "createType", ct, sizeof(ct))) {
 905         error_setg(errp, "invalid VMDK image descriptor");
 906         ret = -EINVAL;
 907         goto exit;
 908     }
 909     if (strcmp(ct, "monolithicFlat") &&
 910         strcmp(ct, "vmfs") &&
 911         strcmp(ct, "vmfsSparse") &&
 912         strcmp(ct, "twoGbMaxExtentSparse") &&
 913         strcmp(ct, "twoGbMaxExtentFlat")) {
 914         error_setg(errp, "Unsupported image type '%s'", ct);
 915         ret = -ENOTSUP;
 916         goto exit;
 917     }
 918     s->create_type = g_strdup(ct);
 919     s->desc_offset = 0;
 920     ret = vmdk_parse_extents(buf, bs, bs->file->exact_filename, errp);
 921 exit:
 922     return ret;
 923 }
 924
 925 static int vmdk_open(BlockDriverState *bs, QDict *options, int flags,
 926                      Error **errp)
 927 {
 928     char *buf;
 929     int ret;
 930     BDRVVmdkState *s = bs->opaque;
 931     uint32_t magic;
 932
 933     buf = vmdk_read_desc(bs->file, 0, errp);
 934     if (!buf) {
 935         return -EINVAL;
 936     }
 937
 938     magic = ldl_be_p(buf);
 939     switch (magic) {
 940         case VMDK3_MAGIC:
 941         case VMDK4_MAGIC:
 942             ret = vmdk_open_sparse(bs, bs->file, flags, buf, errp);
 943             s->desc_offset = 0x200;
 944             break;
 945         default:
 946             ret = vmdk_open_desc_file(bs, flags, buf, errp);
 947             break;
 948     }
 949     if (ret) {
 950         goto fail;
 951     }
 952
 953     /* try to open parent images, if exist */
 954     ret = vmdk_parent_open(bs);
 955     if (ret) {
 956         goto fail;
 957     }
 958     s->cid = vmdk_read_cid(bs, 0);
 959     s->parent_cid = vmdk_read_cid(bs, 1);
 960     qemu_co_mutex_init(&s->lock);
 961
 962     /* Disable migration when VMDK images are used */
 963     error_set(&s->migration_blocker,
 964               QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED,
 965               "vmdk", bdrv_get_device_name(bs), "live migration");
 966     migrate_add_blocker(s->migration_blocker);
 967     g_free(buf);
 968     return 0;
 969
 970 fail:
 971     g_free(buf);
 972     g_free(s->create_type);
 973     s->create_type = NULL;
 974     vmdk_free_extents(bs);
 975     return ret;
 976 }
 977
 978
 979 static void vmdk_refresh_limits(BlockDriverState *bs, Error **errp)
 980 {
 981     BDRVVmdkState *s = bs->opaque;
 982     int i;
 983
 984     for (i = 0; i < s->num_extents; i++) {
 985         if (!s->extents[i].flat) {
 986             bs->bl.write_zeroes_alignment =
 987                 MAX(bs->bl.write_zeroes_alignment,
 988                     s->extents[i].cluster_sectors);
 989         }
 990     }
 991 }
 992
 993 /**
 994  * get_whole_cluster
 995  *
 996  * Copy backing file's cluster that covers @sector_num, otherwise write zero,
 997  * to the cluster at @cluster_sector_num.
 998  *
 999  * If @skip_start_sector < @skip_end_sector, the relative range
1000  * [@skip_start_sector, @skip_end_sector) is not copied or written, and leave
1001  * it for call to write user data in the request.
1002  */
1003 static int get_whole_cluster(BlockDriverState *bs,
1004                              VmdkExtent *extent,
1005                              uint64_t cluster_sector_num,
1006                              uint64_t sector_num,
1007                              uint64_t skip_start_sector,
1008                              uint64_t skip_end_sector)
1009 {
1010     int ret = VMDK_OK;
1011     int64_t cluster_bytes;
1012     uint8_t *whole_grain;
1013
1014     /* For COW, align request sector_num to cluster start */
1015     sector_num = QEMU_ALIGN_DOWN(sector_num, extent->cluster_sectors);
1016     cluster_bytes = extent->cluster_sectors << BDRV_SECTOR_BITS;
1017     whole_grain = qemu_blockalign(bs, cluster_bytes);
1018
1019     if (!bs->backing_hd) {
1020         memset(whole_grain, 0,  skip_start_sector << BDRV_SECTOR_BITS);
1021         memset(whole_grain + (skip_end_sector << BDRV_SECTOR_BITS), 0,
1022                cluster_bytes - (skip_end_sector << BDRV_SECTOR_BITS));
1023     }
1024
1025     assert(skip_end_sector <= extent->cluster_sectors);
1026     /* we will be here if it's first write on non-exist grain(cluster).
1027      * try to read from parent image, if exist */
1028     if (bs->backing_hd && !vmdk_is_cid_valid(bs)) {
1029         ret = VMDK_ERROR;
1030         goto exit;
1031     }
1032
1033     /* Read backing data before skip range */
1034     if (skip_start_sector > 0) {
1035         if (bs->backing_hd) {
1036             ret = bdrv_read(bs->backing_hd, sector_num,
1037                             whole_grain, skip_start_sector);
1038             if (ret < 0) {
1039                 ret = VMDK_ERROR;
1040                 goto exit;
1041             }
1042         }
1043         ret = bdrv_write(extent->file, cluster_sector_num, whole_grain,
1044                          skip_start_sector);
1045         if (ret < 0) {
1046             ret = VMDK_ERROR;
1047             goto exit;
1048         }
1049     }
1050     /* Read backing data after skip range */
1051     if (skip_end_sector < extent->cluster_sectors) {
1052         if (bs->backing_hd) {
1053             ret = bdrv_read(bs->backing_hd, sector_num + skip_end_sector,
1054                             whole_grain + (skip_end_sector << BDRV_SECTOR_BITS),
1055                             extent->cluster_sectors - skip_end_sector);
1056             if (ret < 0) {
1057                 ret = VMDK_ERROR;
1058                 goto exit;
1059             }
1060         }
1061         ret = bdrv_write(extent->file, cluster_sector_num + skip_end_sector,
1062                          whole_grain + (skip_end_sector << BDRV_SECTOR_BITS),
1063                          extent->cluster_sectors - skip_end_sector);
1064         if (ret < 0) {
1065             ret = VMDK_ERROR;
1066             goto exit;
1067         }
1068     }
1069
1070 exit:
1071     qemu_vfree(whole_grain);
1072     return ret;
1073 }
1074
1075 static int vmdk_L2update(VmdkExtent *extent, VmdkMetaData *m_data,
1076                          uint32_t offset)
1077 {
1078     offset = cpu_to_le32(offset);
1079     /* update L2 table */
1080     if (bdrv_pwrite_sync(
1081                 extent->file,
1082                 ((int64_t)m_data->l2_offset * 512)
1083                     + (m_data->l2_index * sizeof(offset)),
1084                 &offset, sizeof(offset)) < 0) {
1085         return VMDK_ERROR;
1086     }
1087     /* update backup L2 table */
1088     if (extent->l1_backup_table_offset != 0) {
1089         m_data->l2_offset = extent->l1_backup_table[m_data->l1_index];
1090         if (bdrv_pwrite_sync(
1091                     extent->file,
1092                     ((int64_t)m_data->l2_offset * 512)
1093                         + (m_data->l2_index * sizeof(offset)),
1094                     &offset, sizeof(offset)) < 0) {
1095             return VMDK_ERROR;
1096         }
1097     }
1098     if (m_data->l2_cache_entry) {
1099         *m_data->l2_cache_entry = offset;
1100     }
1101
1102     return VMDK_OK;
1103 }
1104
1105 /**
1106  * get_cluster_offset
1107  *
1108  * Look up cluster offset in extent file by sector number, and store in
1109  * @cluster_offset.
1110  *
1111  * For flat extents, the start offset as parsed from the description file is
1112  * returned.
1113  *
1114  * For sparse extents, look up in L1, L2 table. If allocate is true, return an
1115  * offset for a new cluster and update L2 cache. If there is a backing file,
1116  * COW is done before returning; otherwise, zeroes are written to the allocated
1117  * cluster. Both COW and zero writing skips the sector range
1118  * [@skip_start_sector, @skip_end_sector) passed in by caller, because caller
1119  * has new data to write there.
1120  *
1121  * Returns: VMDK_OK if cluster exists and mapped in the image.
1122  *          VMDK_UNALLOC if cluster is not mapped and @allocate is false.
1123  *          VMDK_ERROR if failed.
1124  */
1125 static int get_cluster_offset(BlockDriverState *bs,
1126                               VmdkExtent *extent,
1127                               VmdkMetaData *m_data,
1128                               uint64_t offset,
1129                               bool allocate,
1130                               uint64_t *cluster_offset,
1131                               uint64_t skip_start_sector,
1132                               uint64_t skip_end_sector)
1133 {
1134     unsigned int l1_index, l2_offset, l2_index;
1135     int min_index, i, j;
1136     uint32_t min_count, *l2_table;
1137     bool zeroed = false;
1138     int64_t ret;
1139     int64_t cluster_sector;
1140
1141     if (m_data) {
1142         m_data->valid = 0;
1143     }
1144     if (extent->flat) {
1145         *cluster_offset = extent->flat_start_offset;
1146         return VMDK_OK;
1147     }
1148
1149     offset -= (extent->end_sector - extent->sectors) * SECTOR_SIZE;
1150     l1_index = (offset >> 9) / extent->l1_entry_sectors;
1151     if (l1_index >= extent->l1_size) {
1152         return VMDK_ERROR;
1153     }
1154     l2_offset = extent->l1_table[l1_index];
1155     if (!l2_offset) {
1156         return VMDK_UNALLOC;
1157     }
1158     for (i = 0; i < L2_CACHE_SIZE; i++) {
1159         if (l2_offset == extent->l2_cache_offsets[i]) {
1160             /* increment the hit count */
1161             if (++extent->l2_cache_counts[i] == 0xffffffff) {
1162                 for (j = 0; j < L2_CACHE_SIZE; j++) {
1163                     extent->l2_cache_counts[j] >>= 1;
1164                 }
1165             }
1166             l2_table = extent->l2_cache + (i * extent->l2_size);
1167             goto found;
1168         }
1169     }
1170     /* not found: load a new entry in the least used one */
1171     min_index = 0;
1172     min_count = 0xffffffff;
1173     for (i = 0; i < L2_CACHE_SIZE; i++) {
1174         if (extent->l2_cache_counts[i] < min_count) {
1175             min_count = extent->l2_cache_counts[i];
1176             min_index = i;
1177         }
1178     }
1179     l2_table = extent->l2_cache + (min_index * extent->l2_size);
1180     if (bdrv_pread(
1181                 extent->file,
1182                 (int64_t)l2_offset * 512,
1183                 l2_table,
1184                 extent->l2_size * sizeof(uint32_t)
1185             ) != extent->l2_size * sizeof(uint32_t)) {
1186         return VMDK_ERROR;
1187     }
1188
1189     extent->l2_cache_offsets[min_index] = l2_offset;
1190     extent->l2_cache_counts[min_index] = 1;
1191  found:
1192     l2_index = ((offset >> 9) / extent->cluster_sectors) % extent->l2_size;
1193     cluster_sector = le32_to_cpu(l2_table[l2_index]);
1194
1195     if (m_data) {
1196         m_data->valid = 1;
1197         m_data->l1_index = l1_index;
1198         m_data->l2_index = l2_index;
1199         m_data->l2_offset = l2_offset;
1200         m_data->l2_cache_entry = &l2_table[l2_index];
1201     }
1202     if (extent->has_zero_grain && cluster_sector == VMDK_GTE_ZEROED) {
1203         zeroed = true;
1204     }
1205
1206     if (!cluster_sector || zeroed) {
1207         if (!allocate) {
1208             return zeroed ? VMDK_ZEROED : VMDK_UNALLOC;
1209         }
1210
1211         cluster_sector = extent->next_cluster_sector;
1212         extent->next_cluster_sector += extent->cluster_sectors;
1213
1214         /* First of all we write grain itself, to avoid race condition
1215          * that may to corrupt the image.
1216          * This problem may occur because of insufficient space on host disk
1217          * or inappropriate VM shutdown.
1218          */
1219         ret = get_whole_cluster(bs, extent,
1220                                 cluster_sector,
1221                                 offset >> BDRV_SECTOR_BITS,
1222                                 skip_start_sector, skip_end_sector);
1223         if (ret) {
1224             return ret;
1225         }
1226     }
1227     *cluster_offset = cluster_sector << BDRV_SECTOR_BITS;
1228     return VMDK_OK;
1229 }
1230
1231 static VmdkExtent *find_extent(BDRVVmdkState *s,
1232                                 int64_t sector_num, VmdkExtent *start_hint)
1233 {
1234     VmdkExtent *extent = start_hint;
1235
1236     if (!extent) {
1237         extent = &s->extents[0];
1238     }
1239     while (extent < &s->extents[s->num_extents]) {
1240         if (sector_num < extent->end_sector) {
1241             return extent;
1242         }
1243         extent++;
1244     }
1245     return NULL;
1246 }
1247
1248 static int64_t coroutine_fn vmdk_co_get_block_status(BlockDriverState *bs,
1249         int64_t sector_num, int nb_sectors, int *pnum)
1250 {
1251     BDRVVmdkState *s = bs->opaque;
1252     int64_t index_in_cluster, n, ret;
1253     uint64_t offset;
1254     VmdkExtent *extent;
1255
1256     extent = find_extent(s, sector_num, NULL);
1257     if (!extent) {
1258         return 0;
1259     }
1260     qemu_co_mutex_lock(&s->lock);
1261     ret = get_cluster_offset(bs, extent, NULL,
1262                              sector_num * 512, false, &offset,
1263                              0, 0);
1264     qemu_co_mutex_unlock(&s->lock);
1265
1266     switch (ret) {
1267     case VMDK_ERROR:
1268         ret = -EIO;
1269         break;
1270     case VMDK_UNALLOC:
1271         ret = 0;
1272         break;
1273     case VMDK_ZEROED:
1274         ret = BDRV_BLOCK_ZERO;
1275         break;
1276     case VMDK_OK:
1277         ret = BDRV_BLOCK_DATA;
1278         if (extent->file == bs->file && !extent->compressed) {
1279             ret |= BDRV_BLOCK_OFFSET_VALID | offset;
1280         }
1281
1282         break;
1283     }
1284
1285     index_in_cluster = sector_num % extent->cluster_sectors;
1286     n = extent->cluster_sectors - index_in_cluster;
1287     if (n > nb_sectors) {
1288         n = nb_sectors;
1289     }
1290     *pnum = n;
1291     return ret;
1292 }
1293
1294 static int vmdk_write_extent(VmdkExtent *extent, int64_t cluster_offset,
1295                             int64_t offset_in_cluster, const uint8_t *buf,
1296                             int nb_sectors, int64_t sector_num)
1297 {
1298     int ret;
1299     VmdkGrainMarker *data = NULL;
1300     uLongf buf_len;
1301     const uint8_t *write_buf = buf;
1302     int write_len = nb_sectors * 512;
1303
1304     if (extent->compressed) {
1305         if (!extent->has_marker) {
1306             ret = -EINVAL;
1307             goto out;
1308         }
1309         buf_len = (extent->cluster_sectors << 9) * 2;
1310         data = g_malloc(buf_len + sizeof(VmdkGrainMarker));
1311         if (compress(data->data, &buf_len, buf, nb_sectors << 9) != Z_OK ||
1312                 buf_len == 0) {
1313             ret = -EINVAL;
1314             goto out;
1315         }
1316         data->lba = sector_num;
1317         data->size = buf_len;
1318         write_buf = (uint8_t *)data;
1319         write_len = buf_len + sizeof(VmdkGrainMarker);
1320     }
1321     ret = bdrv_pwrite(extent->file,
1322                         cluster_offset + offset_in_cluster,
1323                         write_buf,
1324                         write_len);
1325     if (ret != write_len) {
1326         ret = ret < 0 ? ret : -EIO;
1327         goto out;
1328     }
1329     ret = 0;
1330  out:
1331     g_free(data);
1332     return ret;
1333 }
1334
1335 static int vmdk_read_extent(VmdkExtent *extent, int64_t cluster_offset,
1336                             int64_t offset_in_cluster, uint8_t *buf,
1337                             int nb_sectors)
1338 {
1339     int ret;
1340     int cluster_bytes, buf_bytes;
1341     uint8_t *cluster_buf, *compressed_data;
1342     uint8_t *uncomp_buf;
1343     uint32_t data_len;
1344     VmdkGrainMarker *marker;
1345     uLongf buf_len;
1346
1347
1348     if (!extent->compressed) {
1349         ret = bdrv_pread(extent->file,
1350                           cluster_offset + offset_in_cluster,
1351                           buf, nb_sectors * 512);
1352         if (ret == nb_sectors * 512) {
1353             return 0;
1354         } else {
1355             return -EIO;
1356         }
1357     }
1358     cluster_bytes = extent->cluster_sectors * 512;
1359     /* Read two clusters in case GrainMarker + compressed data > one cluster */
1360     buf_bytes = cluster_bytes * 2;
1361     cluster_buf = g_malloc(buf_bytes);
1362     uncomp_buf = g_malloc(cluster_bytes);
1363     ret = bdrv_pread(extent->file,
1364                 cluster_offset,
1365                 cluster_buf, buf_bytes);
1366     if (ret < 0) {
1367         goto out;
1368     }
1369     compressed_data = cluster_buf;
1370     buf_len = cluster_bytes;
1371     data_len = cluster_bytes;
1372     if (extent->has_marker) {
1373         marker = (VmdkGrainMarker *)cluster_buf;
1374         compressed_data = marker->data;
1375         data_len = le32_to_cpu(marker->size);
1376     }
1377     if (!data_len || data_len > buf_bytes) {
1378         ret = -EINVAL;
1379         goto out;
1380     }
1381     ret = uncompress(uncomp_buf, &buf_len, compressed_data, data_len);
1382     if (ret != Z_OK) {
1383         ret = -EINVAL;
1384         goto out;
1385
1386     }
1387     if (offset_in_cluster < 0 ||
1388             offset_in_cluster + nb_sectors * 512 > buf_len) {
1389         ret = -EINVAL;
1390         goto out;
1391     }
1392     memcpy(buf, uncomp_buf + offset_in_cluster, nb_sectors * 512);
1393     ret = 0;
1394
1395  out:
1396     g_free(uncomp_buf);
1397     g_free(cluster_buf);
1398     return ret;
1399 }
1400
1401 static int vmdk_read(BlockDriverState *bs, int64_t sector_num,
1402                     uint8_t *buf, int nb_sectors)
1403 {
1404     BDRVVmdkState *s = bs->opaque;
1405     int ret;
1406     uint64_t n, index_in_cluster;
1407     uint64_t extent_begin_sector, extent_relative_sector_num;
1408     VmdkExtent *extent = NULL;
1409     uint64_t cluster_offset;
1410
1411     while (nb_sectors > 0) {
1412         extent = find_extent(s, sector_num, extent);
1413         if (!extent) {
1414             return -EIO;
1415         }
1416         ret = get_cluster_offset(bs, extent, NULL,
1417                                  sector_num << 9, false, &cluster_offset,
1418                                  0, 0);
1419         extent_begin_sector = extent->end_sector - extent->sectors;
1420         extent_relative_sector_num = sector_num - extent_begin_sector;
1421         index_in_cluster = extent_relative_sector_num % extent->cluster_sectors;
1422         n = extent->cluster_sectors - index_in_cluster;
1423         if (n > nb_sectors) {
1424             n = nb_sectors;
1425         }
1426         if (ret != VMDK_OK) {
1427             /* if not allocated, try to read from parent image, if exist */
1428             if (bs->backing_hd && ret != VMDK_ZEROED) {
1429                 if (!vmdk_is_cid_valid(bs)) {
1430                     return -EINVAL;
1431                 }
1432                 ret = bdrv_read(bs->backing_hd, sector_num, buf, n);
1433                 if (ret < 0) {
1434                     return ret;
1435                 }
1436             } else {
1437                 memset(buf, 0, 512 * n);
1438             }
1439         } else {
1440             ret = vmdk_read_extent(extent,
1441                             cluster_offset, index_in_cluster * 512,
1442                             buf, n);
1443             if (ret) {
1444                 return ret;
1445             }
1446         }
1447         nb_sectors -= n;
1448         sector_num += n;
1449         buf += n * 512;
1450     }
1451     return 0;
1452 }
1453
1454 static coroutine_fn int vmdk_co_read(BlockDriverState *bs, int64_t sector_num,
1455                                      uint8_t *buf, int nb_sectors)
1456 {
1457     int ret;
1458     BDRVVmdkState *s = bs->opaque;
1459     qemu_co_mutex_lock(&s->lock);
1460     ret = vmdk_read(bs, sector_num, buf, nb_sectors);
1461     qemu_co_mutex_unlock(&s->lock);
1462     return ret;
1463 }
1464
1465 /**
1466  * vmdk_write:
1467  * @zeroed:       buf is ignored (data is zero), use zeroed_grain GTE feature
1468  *                if possible, otherwise return -ENOTSUP.
1469  * @zero_dry_run: used for zeroed == true only, don't update L2 table, just try
1470  *                with each cluster. By dry run we can find if the zero write
1471  *                is possible without modifying image data.
1472  *
1473  * Returns: error code with 0 for success.
1474  */
1475 static int vmdk_write(BlockDriverState *bs, int64_t sector_num,
1476                       const uint8_t *buf, int nb_sectors,
1477                       bool zeroed, bool zero_dry_run)
1478 {
1479     BDRVVmdkState *s = bs->opaque;
1480     VmdkExtent *extent = NULL;
1481     int ret;
1482     int64_t index_in_cluster, n;
1483     uint64_t extent_begin_sector, extent_relative_sector_num;
1484     uint64_t cluster_offset;
1485     VmdkMetaData m_data;
1486
1487     if (sector_num > bs->total_sectors) {
1488         error_report("Wrong offset: sector_num=0x%" PRIx64
1489                 " total_sectors=0x%" PRIx64 "\n",
1490                 sector_num, bs->total_sectors);
1491         return -EIO;
1492     }
1493
1494     while (nb_sectors > 0) {
1495         extent = find_extent(s, sector_num, extent);
1496         if (!extent) {
1497             return -EIO;
1498         }
1499         extent_begin_sector = extent->end_sector - extent->sectors;
1500         extent_relative_sector_num = sector_num - extent_begin_sector;
1501         index_in_cluster = extent_relative_sector_num % extent->cluster_sectors;
1502         n = extent->cluster_sectors - index_in_cluster;
1503         if (n > nb_sectors) {
1504             n = nb_sectors;
1505         }
1506         ret = get_cluster_offset(bs, extent, &m_data, sector_num << 9,
1507                                  !(extent->compressed || zeroed),
1508                                  &cluster_offset,
1509                                  index_in_cluster, index_in_cluster + n);
1510         if (extent->compressed) {
1511             if (ret == VMDK_OK) {
1512                 /* Refuse write to allocated cluster for streamOptimized */
1513                 error_report("Could not write to allocated cluster"
1514                               " for streamOptimized");
1515                 return -EIO;
1516             } else {
1517                 /* allocate */
1518                 ret = get_cluster_offset(bs, extent, &m_data, sector_num << 9,
1519                                          true, &cluster_offset, 0, 0);
1520             }
1521         }
1522         if (ret == VMDK_ERROR) {
1523             return -EINVAL;
1524         }
1525         if (zeroed) {
1526             /* Do zeroed write, buf is ignored */
1527             if (extent->has_zero_grain &&
1528                     index_in_cluster == 0 &&
1529                     n >= extent->cluster_sectors) {
1530                 n = extent->cluster_sectors;
1531                 if (!zero_dry_run) {
1532                     /* update L2 tables */
1533                     if (vmdk_L2update(extent, &m_data, VMDK_GTE_ZEROED)
1534                             != VMDK_OK) {
1535                         return -EIO;
1536                     }
1537                 }
1538             } else {
1539                 return -ENOTSUP;
1540             }
1541         } else {
1542             ret = vmdk_write_extent(extent,
1543                             cluster_offset, index_in_cluster * 512,
1544                             buf, n, sector_num);
1545             if (ret) {
1546                 return ret;
1547             }
1548             if (m_data.valid) {
1549                 /* update L2 tables */
1550                 if (vmdk_L2update(extent, &m_data,
1551                                   cluster_offset >> BDRV_SECTOR_BITS)
1552                         != VMDK_OK) {
1553                     return -EIO;
1554                 }
1555             }
1556         }
1557         nb_sectors -= n;
1558         sector_num += n;
1559         buf += n * 512;
1560
1561         /* update CID on the first write every time the virtual disk is
1562          * opened */
1563         if (!s->cid_updated) {
1564             ret = vmdk_write_cid(bs, g_random_int());
1565             if (ret < 0) {
1566                 return ret;
1567             }
1568             s->cid_updated = true;
1569         }
1570     }
1571     return 0;
1572 }
1573
1574 static coroutine_fn int vmdk_co_write(BlockDriverState *bs, int64_t sector_num,
1575                                       const uint8_t *buf, int nb_sectors)
1576 {
1577     int ret;
1578     BDRVVmdkState *s = bs->opaque;
1579     qemu_co_mutex_lock(&s->lock);
1580     ret = vmdk_write(bs, sector_num, buf, nb_sectors, false, false);
1581     qemu_co_mutex_unlock(&s->lock);
1582     return ret;
1583 }
1584
1585 static int vmdk_write_compressed(BlockDriverState *bs,
1586                                  int64_t sector_num,
1587                                  const uint8_t *buf,
1588                                  int nb_sectors)
1589 {
1590     BDRVVmdkState *s = bs->opaque;
1591     if (s->num_extents == 1 && s->extents[0].compressed) {
1592         return vmdk_write(bs, sector_num, buf, nb_sectors, false, false);
1593     } else {
1594         return -ENOTSUP;
1595     }
1596 }
1597
1598 static int coroutine_fn vmdk_co_write_zeroes(BlockDriverState *bs,
1599                                              int64_t sector_num,
1600                                              int nb_sectors,
1601                                              BdrvRequestFlags flags)
1602 {
1603     int ret;
1604     BDRVVmdkState *s = bs->opaque;
1605     qemu_co_mutex_lock(&s->lock);
1606     /* write zeroes could fail if sectors not aligned to cluster, test it with
1607      * dry_run == true before really updating image */
1608     ret = vmdk_write(bs, sector_num, NULL, nb_sectors, true, true);
1609     if (!ret) {
1610         ret = vmdk_write(bs, sector_num, NULL, nb_sectors, true, false);
1611     }
1612     qemu_co_mutex_unlock(&s->lock);
1613     return ret;
1614 }
1615
1616 static int vmdk_create_extent(const char *filename, int64_t filesize,
1617                               bool flat, bool compress, bool zeroed_grain,
1618                               QemuOpts *opts, Error **errp)
1619 {
1620     int ret, i;
1621     BlockDriverState *bs = NULL;
1622     VMDK4Header header;
1623     Error *local_err = NULL;
1624     uint32_t tmp, magic, grains, gd_sectors, gt_size, gt_count;
1625     uint32_t *gd_buf = NULL;
1626     int gd_buf_size;
1627
1628     ret = bdrv_create_file(filename, opts, &local_err);
1629     if (ret < 0) {
1630         error_propagate(errp, local_err);
1631         goto exit;
1632     }
1633
1634     assert(bs == NULL);
1635     ret = bdrv_open(&bs, filename, NULL, NULL, BDRV_O_RDWR | BDRV_O_PROTOCOL,
1636                     NULL, &local_err);
1637     if (ret < 0) {
1638         error_propagate(errp, local_err);
1639         goto exit;
1640     }
1641
1642     if (flat) {
1643         ret = bdrv_truncate(bs, filesize);
1644         if (ret < 0) {
1645             error_setg_errno(errp, -ret, "Could not truncate file");
1646         }
1647         goto exit;
1648     }
1649     magic = cpu_to_be32(VMDK4_MAGIC);
1650     memset(&header, 0, sizeof(header));
1651     header.version = zeroed_grain ? 2 : 1;
1652     header.flags = VMDK4_FLAG_RGD | VMDK4_FLAG_NL_DETECT
1653                    | (compress ? VMDK4_FLAG_COMPRESS | VMDK4_FLAG_MARKER : 0)
1654                    | (zeroed_grain ? VMDK4_FLAG_ZERO_GRAIN : 0);
1655     header.compressAlgorithm = compress ? VMDK4_COMPRESSION_DEFLATE : 0;
1656     header.capacity = filesize / BDRV_SECTOR_SIZE;
1657     header.granularity = 128;
1658     header.num_gtes_per_gt = BDRV_SECTOR_SIZE;
1659
1660     grains = DIV_ROUND_UP(filesize / BDRV_SECTOR_SIZE, header.granularity);
1661     gt_size = DIV_ROUND_UP(header.num_gtes_per_gt * sizeof(uint32_t),
1662                            BDRV_SECTOR_SIZE);
1663     gt_count = DIV_ROUND_UP(grains, header.num_gtes_per_gt);
1664     gd_sectors = DIV_ROUND_UP(gt_count * sizeof(uint32_t), BDRV_SECTOR_SIZE);
1665
1666     header.desc_offset = 1;
1667     header.desc_size = 20;
1668     header.rgd_offset = header.desc_offset + header.desc_size;
1669     header.gd_offset = header.rgd_offset + gd_sectors + (gt_size * gt_count);
1670     header.grain_offset =
1671         ROUND_UP(header.gd_offset + gd_sectors + (gt_size * gt_count),
1672                  header.granularity);
1673     /* swap endianness for all header fields */
1674     header.version = cpu_to_le32(header.version);
1675     header.flags = cpu_to_le32(header.flags);
1676     header.capacity = cpu_to_le64(header.capacity);
1677     header.granularity = cpu_to_le64(header.granularity);
1678     header.num_gtes_per_gt = cpu_to_le32(header.num_gtes_per_gt);
1679     header.desc_offset = cpu_to_le64(header.desc_offset);
1680     header.desc_size = cpu_to_le64(header.desc_size);
1681     header.rgd_offset = cpu_to_le64(header.rgd_offset);
1682     header.gd_offset = cpu_to_le64(header.gd_offset);
1683     header.grain_offset = cpu_to_le64(header.grain_offset);
1684     header.compressAlgorithm = cpu_to_le16(header.compressAlgorithm);
1685
1686     header.check_bytes[0] = 0xa;
1687     header.check_bytes[1] = 0x20;
1688     header.check_bytes[2] = 0xd;
1689     header.check_bytes[3] = 0xa;
1690
1691     /* write all the data */
1692     ret = bdrv_pwrite(bs, 0, &magic, sizeof(magic));
1693     if (ret < 0) {
1694         error_set(errp, QERR_IO_ERROR);
1695         goto exit;
1696     }
1697     ret = bdrv_pwrite(bs, sizeof(magic), &header, sizeof(header));
1698     if (ret < 0) {
1699         error_set(errp, QERR_IO_ERROR);
1700         goto exit;
1701     }
1702
1703     ret = bdrv_truncate(bs, le64_to_cpu(header.grain_offset) << 9);
1704     if (ret < 0) {
1705         error_setg_errno(errp, -ret, "Could not truncate file");
1706         goto exit;
1707     }
1708
1709     /* write grain directory */
1710     gd_buf_size = gd_sectors * BDRV_SECTOR_SIZE;
1711     gd_buf = g_malloc0(gd_buf_size);
1712     for (i = 0, tmp = le64_to_cpu(header.rgd_offset) + gd_sectors;
1713          i < gt_count; i++, tmp += gt_size) {
1714         gd_buf[i] = cpu_to_le32(tmp);
1715     }
1716     ret = bdrv_pwrite(bs, le64_to_cpu(header.rgd_offset) * BDRV_SECTOR_SIZE,
1717                       gd_buf, gd_buf_size);
1718     if (ret < 0) {
1719         error_set(errp, QERR_IO_ERROR);
1720         goto exit;
1721     }
1722
1723     /* write backup grain directory */
1724     for (i = 0, tmp = le64_to_cpu(header.gd_offset) + gd_sectors;
1725          i < gt_count; i++, tmp += gt_size) {
1726         gd_buf[i] = cpu_to_le32(tmp);
1727     }
1728     ret = bdrv_pwrite(bs, le64_to_cpu(header.gd_offset) * BDRV_SECTOR_SIZE,
1729                       gd_buf, gd_buf_size);
1730     if (ret < 0) {
1731         error_set(errp, QERR_IO_ERROR);
1732         goto exit;
1733     }
1734
1735     ret = 0;
1736 exit:
1737     if (bs) {
1738         bdrv_unref(bs);
1739     }
1740     g_free(gd_buf);
1741     return ret;
1742 }
1743
1744 static int filename_decompose(const char *filename, char *path, char *prefix,
1745                               char *postfix, size_t buf_len, Error **errp)
1746 {
1747     const char *p, *q;
1748
1749     if (filename == NULL || !strlen(filename)) {
1750         error_setg(errp, "No filename provided");
1751         return VMDK_ERROR;
1752     }
1753     p = strrchr(filename, '/');
1754     if (p == NULL) {
1755         p = strrchr(filename, '\\');
1756     }
1757     if (p == NULL) {
1758         p = strrchr(filename, ':');
1759     }
1760     if (p != NULL) {
1761         p++;
1762         if (p - filename >= buf_len) {
1763             return VMDK_ERROR;
1764         }
1765         pstrcpy(path, p - filename + 1, filename);
1766     } else {
1767         p = filename;
1768         path[0] = '\0';
1769     }
1770     q = strrchr(p, '.');
1771     if (q == NULL) {
1772         pstrcpy(prefix, buf_len, p);
1773         postfix[0] = '\0';
1774     } else {
1775         if (q - p >= buf_len) {
1776             return VMDK_ERROR;
1777         }
1778         pstrcpy(prefix, q - p + 1, p);
1779         pstrcpy(postfix, buf_len, q);
1780     }
1781     return VMDK_OK;
1782 }
1783
1784 static int vmdk_create(const char *filename, QemuOpts *opts, Error **errp)
1785 {
1786     int idx = 0;
1787     BlockDriverState *new_bs = NULL;
1788     Error *local_err = NULL;
1789     char *desc = NULL;
1790     int64_t total_size = 0, filesize;
1791     char *adapter_type = NULL;
1792     char *backing_file = NULL;
1793     char *fmt = NULL;
1794     int flags = 0;
1795     int ret = 0;
1796     bool flat, split, compress;
1797     GString *ext_desc_lines;
1798     char path[PATH_MAX], prefix[PATH_MAX], postfix[PATH_MAX];
1799     const int64_t split_size = 0x80000000;  /* VMDK has constant split size */
1800     const char *desc_extent_line;
1801     char parent_desc_line[BUF_SIZE] = "";
1802     uint32_t parent_cid = 0xffffffff;
1803     uint32_t number_heads = 16;
1804     bool zeroed_grain = false;
1805     uint32_t desc_offset = 0, desc_len;
1806     const char desc_template[] =
1807         "# Disk DescriptorFile\n"
1808         "version=1\n"
1809         "CID=%" PRIx32 "\n"
1810         "parentCID=%" PRIx32 "\n"
1811         "createType=\"%s\"\n"
1812         "%s"
1813         "\n"
1814         "# Extent description\n"
1815         "%s"
1816         "\n"
1817         "# The Disk Data Base\n"
1818         "#DDB\n"
1819         "\n"
1820         "ddb.virtualHWVersion = \"%d\"\n"
1821         "ddb.geometry.cylinders = \"%" PRId64 "\"\n"
1822         "ddb.geometry.heads = \"%" PRIu32 "\"\n"
1823         "ddb.geometry.sectors = \"63\"\n"
1824         "ddb.adapterType = \"%s\"\n";
1825
1826     ext_desc_lines = g_string_new(NULL);
1827
1828     if (filename_decompose(filename, path, prefix, postfix, PATH_MAX, errp)) {
1829         ret = -EINVAL;
1830         goto exit;
1831     }
1832     /* Read out options */
1833     total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
1834                           BDRV_SECTOR_SIZE);
1835     adapter_type = qemu_opt_get_del(opts, BLOCK_OPT_ADAPTER_TYPE);
1836     backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
1837     if (qemu_opt_get_bool_del(opts, BLOCK_OPT_COMPAT6, false)) {
1838         flags |= BLOCK_FLAG_COMPAT6;
1839     }
1840     fmt = qemu_opt_get_del(opts, BLOCK_OPT_SUBFMT);
1841     if (qemu_opt_get_bool_del(opts, BLOCK_OPT_ZEROED_GRAIN, false)) {
1842         zeroed_grain = true;
1843     }
1844
1845     if (!adapter_type) {
1846         adapter_type = g_strdup("ide");
1847     } else if (strcmp(adapter_type, "ide") &&
1848                strcmp(adapter_type, "buslogic") &&
1849                strcmp(adapter_type, "lsilogic") &&
1850                strcmp(adapter_type, "legacyESX")) {
1851         error_setg(errp, "Unknown adapter type: '%s'", adapter_type);
1852         ret = -EINVAL;
1853         goto exit;
1854     }
1855     if (strcmp(adapter_type, "ide") != 0) {
1856         /* that's the number of heads with which vmware operates when
1857            creating, exporting, etc. vmdk files with a non-ide adapter type */
1858         number_heads = 255;
1859     }
1860     if (!fmt) {
1861         /* Default format to monolithicSparse */
1862         fmt = g_strdup("monolithicSparse");
1863     } else if (strcmp(fmt, "monolithicFlat") &&
1864                strcmp(fmt, "monolithicSparse") &&
1865                strcmp(fmt, "twoGbMaxExtentSparse") &&
1866                strcmp(fmt, "twoGbMaxExtentFlat") &&
1867                strcmp(fmt, "streamOptimized")) {
1868         error_setg(errp, "Unknown subformat: '%s'", fmt);
1869         ret = -EINVAL;
1870         goto exit;
1871     }
1872     split = !(strcmp(fmt, "twoGbMaxExtentFlat") &&
1873               strcmp(fmt, "twoGbMaxExtentSparse"));
1874     flat = !(strcmp(fmt, "monolithicFlat") &&
1875              strcmp(fmt, "twoGbMaxExtentFlat"));
1876     compress = !strcmp(fmt, "streamOptimized");
1877     if (flat) {
1878         desc_extent_line = "RW %" PRId64 " FLAT \"%s\" 0\n";
1879     } else {
1880         desc_extent_line = "RW %" PRId64 " SPARSE \"%s\"\n";
1881     }
1882     if (flat && backing_file) {
1883         error_setg(errp, "Flat image can't have backing file");
1884         ret = -ENOTSUP;
1885         goto exit;
1886     }
1887     if (flat && zeroed_grain) {
1888         error_setg(errp, "Flat image can't enable zeroed grain");
1889         ret = -ENOTSUP;
1890         goto exit;
1891     }
1892     if (backing_file) {
1893         BlockDriverState *bs = NULL;
1894         char *full_backing = g_new0(char, PATH_MAX);
1895         bdrv_get_full_backing_filename_from_filename(filename, backing_file,
1896                                                      full_backing, PATH_MAX,
1897                                                      &local_err);
1898         if (local_err) {
1899             g_free(full_backing);
1900             error_propagate(errp, local_err);
1901             ret = -ENOENT;
1902             goto exit;
1903         }
1904         ret = bdrv_open(&bs, full_backing, NULL, NULL, BDRV_O_NO_BACKING, NULL,
1905                         errp);
1906         g_free(full_backing);
1907         if (ret != 0) {
1908             goto exit;
1909         }
1910         if (strcmp(bs->drv->format_name, "vmdk")) {
1911             bdrv_unref(bs);
1912             ret = -EINVAL;
1913             goto exit;
1914         }
1915         parent_cid = vmdk_read_cid(bs, 0);
1916         bdrv_unref(bs);
1917         snprintf(parent_desc_line, sizeof(parent_desc_line),
1918                 "parentFileNameHint=\"%s\"", backing_file);
1919     }
1920
1921     /* Create extents */
1922     filesize = total_size;
1923     while (filesize > 0) {
1924         char desc_line[BUF_SIZE];
1925         char ext_filename[PATH_MAX];
1926         char desc_filename[PATH_MAX];
1927         int64_t size = filesize;
1928
1929         if (split && size > split_size) {
1930             size = split_size;
1931         }
1932         if (split) {
1933             snprintf(desc_filename, sizeof(desc_filename), "%s-%c%03d%s",
1934                     prefix, flat ? 'f' : 's', ++idx, postfix);
1935         } else if (flat) {
1936             snprintf(desc_filename, sizeof(desc_filename), "%s-flat%s",
1937                     prefix, postfix);
1938         } else {
1939             snprintf(desc_filename, sizeof(desc_filename), "%s%s",
1940                     prefix, postfix);
1941         }
1942         snprintf(ext_filename, sizeof(ext_filename), "%s%s",
1943                 path, desc_filename);
1944
1945         if (vmdk_create_extent(ext_filename, size,
1946                                flat, compress, zeroed_grain, opts, errp)) {
1947             ret = -EINVAL;
1948             goto exit;
1949         }
1950         filesize -= size;
1951
1952         /* Format description line */
1953         snprintf(desc_line, sizeof(desc_line),
1954                     desc_extent_line, size / BDRV_SECTOR_SIZE, desc_filename);
1955         g_string_append(ext_desc_lines, desc_line);
1956     }
1957     /* generate descriptor file */
1958     desc = g_strdup_printf(desc_template,
1959                            g_random_int(),
1960                            parent_cid,
1961                            fmt,
1962                            parent_desc_line,
1963                            ext_desc_lines->str,
1964                            (flags & BLOCK_FLAG_COMPAT6 ? 6 : 4),
1965                            total_size /
1966                                (int64_t)(63 * number_heads * BDRV_SECTOR_SIZE),
1967                            number_heads,
1968                            adapter_type);
1969     desc_len = strlen(desc);
1970     /* the descriptor offset = 0x200 */
1971     if (!split && !flat) {
1972         desc_offset = 0x200;
1973     } else {
1974         ret = bdrv_create_file(filename, opts, &local_err);
1975         if (ret < 0) {
1976             error_propagate(errp, local_err);
1977             goto exit;
1978         }
1979     }
1980     assert(new_bs == NULL);
1981     ret = bdrv_open(&new_bs, filename, NULL, NULL,
1982                     BDRV_O_RDWR | BDRV_O_PROTOCOL, NULL, &local_err);
1983     if (ret < 0) {
1984         error_propagate(errp, local_err);
1985         goto exit;
1986     }
1987     ret = bdrv_pwrite(new_bs, desc_offset, desc, desc_len);
1988     if (ret < 0) {
1989         error_setg_errno(errp, -ret, "Could not write description");
1990         goto exit;
1991     }
1992     /* bdrv_pwrite write padding zeros to align to sector, we don't need that
1993      * for description file */
1994     if (desc_offset == 0) {
1995         ret = bdrv_truncate(new_bs, desc_len);
1996         if (ret < 0) {
1997             error_setg_errno(errp, -ret, "Could not truncate file");
1998         }
1999     }
2000 exit:
2001     if (new_bs) {
2002         bdrv_unref(new_bs);
2003     }
2004     g_free(adapter_type);
2005     g_free(backing_file);
2006     g_free(fmt);
2007     g_free(desc);
2008     g_string_free(ext_desc_lines, true);
2009     return ret;
2010 }
2011
2012 static void vmdk_close(BlockDriverState *bs)
2013 {
2014     BDRVVmdkState *s = bs->opaque;
2015
2016     vmdk_free_extents(bs);
2017     g_free(s->create_type);
2018
2019     migrate_del_blocker(s->migration_blocker);
2020     error_free(s->migration_blocker);
2021 }
2022
2023 static coroutine_fn int vmdk_co_flush(BlockDriverState *bs)
2024 {
2025     BDRVVmdkState *s = bs->opaque;
2026     int i, err;
2027     int ret = 0;
2028
2029     for (i = 0; i < s->num_extents; i++) {
2030         err = bdrv_co_flush(s->extents[i].file);
2031         if (err < 0) {
2032             ret = err;
2033         }
2034     }
2035     return ret;
2036 }
2037
2038 static int64_t vmdk_get_allocated_file_size(BlockDriverState *bs)
2039 {
2040     int i;
2041     int64_t ret = 0;
2042     int64_t r;
2043     BDRVVmdkState *s = bs->opaque;
2044
2045     ret = bdrv_get_allocated_file_size(bs->file);
2046     if (ret < 0) {
2047         return ret;
2048     }
2049     for (i = 0; i < s->num_extents; i++) {
2050         if (s->extents[i].file == bs->file) {
2051             continue;
2052         }
2053         r = bdrv_get_allocated_file_size(s->extents[i].file);
2054         if (r < 0) {
2055             return r;
2056         }
2057         ret += r;
2058     }
2059     return ret;
2060 }
2061
2062 static int vmdk_has_zero_init(BlockDriverState *bs)
2063 {
2064     int i;
2065     BDRVVmdkState *s = bs->opaque;
2066
2067     /* If has a flat extent and its underlying storage doesn't have zero init,
2068      * return 0. */
2069     for (i = 0; i < s->num_extents; i++) {
2070         if (s->extents[i].flat) {
2071             if (!bdrv_has_zero_init(s->extents[i].file)) {
2072                 return 0;
2073             }
2074         }
2075     }
2076     return 1;
2077 }
2078
2079 static ImageInfo *vmdk_get_extent_info(VmdkExtent *extent)
2080 {
2081     ImageInfo *info = g_new0(ImageInfo, 1);
2082
2083     *info = (ImageInfo){
2084         .filename         = g_strdup(extent->file->filename),
2085         .format           = g_strdup(extent->type),
2086         .virtual_size     = extent->sectors * BDRV_SECTOR_SIZE,
2087         .compressed       = extent->compressed,
2088         .has_compressed   = extent->compressed,
2089         .cluster_size     = extent->cluster_sectors * BDRV_SECTOR_SIZE,
2090         .has_cluster_size = !extent->flat,
2091     };
2092
2093     return info;
2094 }
2095
2096 static int vmdk_check(BlockDriverState *bs, BdrvCheckResult *result,
2097                       BdrvCheckMode fix)
2098 {
2099     BDRVVmdkState *s = bs->opaque;
2100     VmdkExtent *extent = NULL;
2101     int64_t sector_num = 0;
2102     int64_t total_sectors = bdrv_nb_sectors(bs);
2103     int ret;
2104     uint64_t cluster_offset;
2105
2106     if (fix) {
2107         return -ENOTSUP;
2108     }
2109
2110     for (;;) {
2111         if (sector_num >= total_sectors) {
2112             return 0;
2113         }
2114         extent = find_extent(s, sector_num, extent);
2115         if (!extent) {
2116             fprintf(stderr,
2117                     "ERROR: could not find extent for sector %" PRId64 "\n",
2118                     sector_num);
2119             break;
2120         }
2121         ret = get_cluster_offset(bs, extent, NULL,
2122                                  sector_num << BDRV_SECTOR_BITS,
2123                                  false, &cluster_offset, 0, 0);
2124         if (ret == VMDK_ERROR) {
2125             fprintf(stderr,
2126                     "ERROR: could not get cluster_offset for sector %"
2127                     PRId64 "\n", sector_num);
2128             break;
2129         }
2130         if (ret == VMDK_OK && cluster_offset >= bdrv_getlength(extent->file)) {
2131             fprintf(stderr,
2132                     "ERROR: cluster offset for sector %"
2133                     PRId64 " points after EOF\n", sector_num);
2134             break;
2135         }
2136         sector_num += extent->cluster_sectors;
2137     }
2138
2139     result->corruptions++;
2140     return 0;
2141 }
2142
2143 static ImageInfoSpecific *vmdk_get_specific_info(BlockDriverState *bs)
2144 {
2145     int i;
2146     BDRVVmdkState *s = bs->opaque;
2147     ImageInfoSpecific *spec_info = g_new0(ImageInfoSpecific, 1);
2148     ImageInfoList **next;
2149
2150     *spec_info = (ImageInfoSpecific){
2151         .kind = IMAGE_INFO_SPECIFIC_KIND_VMDK,
2152         {
2153             .vmdk = g_new0(ImageInfoSpecificVmdk, 1),
2154         },
2155     };
2156
2157     *spec_info->vmdk = (ImageInfoSpecificVmdk) {
2158         .create_type = g_strdup(s->create_type),
2159         .cid = s->cid,
2160         .parent_cid = s->parent_cid,
2161     };
2162
2163     next = &spec_info->vmdk->extents;
2164     for (i = 0; i < s->num_extents; i++) {
2165         *next = g_new0(ImageInfoList, 1);
2166         (*next)->value = vmdk_get_extent_info(&s->extents[i]);
2167         (*next)->next = NULL;
2168         next = &(*next)->next;
2169     }
2170
2171     return spec_info;
2172 }
2173
2174 static bool vmdk_extents_type_eq(const VmdkExtent *a, const VmdkExtent *b)
2175 {
2176     return a->flat == b->flat &&
2177            a->compressed == b->compressed &&
2178            (a->flat || a->cluster_sectors == b->cluster_sectors);
2179 }
2180
2181 static int vmdk_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2182 {
2183     int i;
2184     BDRVVmdkState *s = bs->opaque;
2185     assert(s->num_extents);
2186
2187     /* See if we have multiple extents but they have different cases */
2188     for (i = 1; i < s->num_extents; i++) {
2189         if (!vmdk_extents_type_eq(&s->extents[0], &s->extents[i])) {
2190             return -ENOTSUP;
2191         }
2192     }
2193     bdi->needs_compressed_writes = s->extents[0].compressed;
2194     if (!s->extents[0].flat) {
2195         bdi->cluster_size = s->extents[0].cluster_sectors << BDRV_SECTOR_BITS;
2196     }
2197     return 0;
2198 }
2199
2200 static void vmdk_detach_aio_context(BlockDriverState *bs)
2201 {
2202     BDRVVmdkState *s = bs->opaque;
2203     int i;
2204
2205     for (i = 0; i < s->num_extents; i++) {
2206         bdrv_detach_aio_context(s->extents[i].file);
2207     }
2208 }
2209
2210 static void vmdk_attach_aio_context(BlockDriverState *bs,
2211                                     AioContext *new_context)
2212 {
2213     BDRVVmdkState *s = bs->opaque;
2214     int i;
2215
2216     for (i = 0; i < s->num_extents; i++) {
2217         bdrv_attach_aio_context(s->extents[i].file, new_context);
2218     }
2219 }
2220
2221 static QemuOptsList vmdk_create_opts = {
2222     .name = "vmdk-create-opts",
2223     .head = QTAILQ_HEAD_INITIALIZER(vmdk_create_opts.head),
2224     .desc = {
2225         {
2226             .name = BLOCK_OPT_SIZE,
2227             .type = QEMU_OPT_SIZE,
2228             .help = "Virtual disk size"
2229         },
2230         {
2231             .name = BLOCK_OPT_ADAPTER_TYPE,
2232             .type = QEMU_OPT_STRING,
2233             .help = "Virtual adapter type, can be one of "
2234                     "ide (default), lsilogic, buslogic or legacyESX"
2235         },
2236         {
2237             .name = BLOCK_OPT_BACKING_FILE,
2238             .type = QEMU_OPT_STRING,
2239             .help = "File name of a base image"
2240         },
2241         {
2242             .name = BLOCK_OPT_COMPAT6,
2243             .type = QEMU_OPT_BOOL,
2244             .help = "VMDK version 6 image",
2245             .def_value_str = "off"
2246         },
2247         {
2248             .name = BLOCK_OPT_SUBFMT,
2249             .type = QEMU_OPT_STRING,
2250             .help =
2251                 "VMDK flat extent format, can be one of "
2252                 "{monolithicSparse (default) | monolithicFlat | twoGbMaxExtentSparse | twoGbMaxExtentFlat | streamOptimized} "
2253         },
2254         {
2255             .name = BLOCK_OPT_ZEROED_GRAIN,
2256             .type = QEMU_OPT_BOOL,
2257             .help = "Enable efficient zero writes "
2258                     "using the zeroed-grain GTE feature"
2259         },
2260         { /* end of list */ }
2261     }
2262 };
2263
2264 static BlockDriver bdrv_vmdk = {
2265     .format_name                  = "vmdk",
2266     .instance_size                = sizeof(BDRVVmdkState),
2267     .bdrv_probe                   = vmdk_probe,
2268     .bdrv_open                    = vmdk_open,
2269     .bdrv_check                   = vmdk_check,
2270     .bdrv_reopen_prepare          = vmdk_reopen_prepare,
2271     .bdrv_read                    = vmdk_co_read,
2272     .bdrv_write                   = vmdk_co_write,
2273     .bdrv_write_compressed        = vmdk_write_compressed,
2274     .bdrv_co_write_zeroes         = vmdk_co_write_zeroes,
2275     .bdrv_close                   = vmdk_close,
2276     .bdrv_create                  = vmdk_create,
2277     .bdrv_co_flush_to_disk        = vmdk_co_flush,
2278     .bdrv_co_get_block_status     = vmdk_co_get_block_status,
2279     .bdrv_get_allocated_file_size = vmdk_get_allocated_file_size,
2280     .bdrv_has_zero_init           = vmdk_has_zero_init,
2281     .bdrv_get_specific_info       = vmdk_get_specific_info,
2282     .bdrv_refresh_limits          = vmdk_refresh_limits,
2283     .bdrv_get_info                = vmdk_get_info,
2284     .bdrv_detach_aio_context      = vmdk_detach_aio_context,
2285     .bdrv_attach_aio_context      = vmdk_attach_aio_context,
2286
2287     .supports_backing             = true,
2288     .create_opts                  = &vmdk_create_opts,
2289 };
2290
2291 static void bdrv_vmdk_init(void)
2292 {
2293     bdrv_register(&bdrv_vmdk);
2294 }
2295
2296 block_init(bdrv_vmdk_init);