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/osdep.h"
  27 #include "qapi/error.h"
  28 #include "block/block_int.h"
  29 #include "sysemu/block-backend.h"
  30 #include "qapi/qmp/qerror.h"
  31 #include "qemu/error-report.h"
  32 #include "qemu/module.h"
  33 #include "qemu/bswap.h"
  34 #include "migration/migration.h"
  35 #include "qemu/cutils.h"
  36 #include <zlib.h>
  37
  38 #define VMDK3_MAGIC (('C' << 24) | ('O' << 16) | ('W' << 8) | 'D')
  39 #define VMDK4_MAGIC (('K' << 24) | ('D' << 16) | ('M' << 8) | 'V')
  40 #define VMDK4_COMPRESSION_DEFLATE 1
  41 #define VMDK4_FLAG_NL_DETECT (1 << 0)
  42 #define VMDK4_FLAG_RGD (1 << 1)
  43 /* Zeroed-grain enable bit */
  44 #define VMDK4_FLAG_ZERO_GRAIN   (1 << 2)
  45 #define VMDK4_FLAG_COMPRESS (1 << 16)
  46 #define VMDK4_FLAG_MARKER (1 << 17)
  47 #define VMDK4_GD_AT_END 0xffffffffffffffffULL
  48
  49 #define VMDK_GTE_ZEROED 0x1
  50
  51 /* VMDK internal error codes */
  52 #define VMDK_OK      0
  53 #define VMDK_ERROR   (-1)
  54 /* Cluster not allocated */
  55 #define VMDK_UNALLOC (-2)
  56 #define VMDK_ZEROED  (-3)
  57
  58 #define BLOCK_OPT_ZEROED_GRAIN "zeroed_grain"
  59
  60 typedef struct {
  61     uint32_t version;
  62     uint32_t flags;
  63     uint32_t disk_sectors;
  64     uint32_t granularity;
  65     uint32_t l1dir_offset;
  66     uint32_t l1dir_size;
  67     uint32_t file_sectors;
  68     uint32_t cylinders;
  69     uint32_t heads;
  70     uint32_t sectors_per_track;
  71 } QEMU_PACKED VMDK3Header;
  72
  73 typedef struct {
  74     uint32_t version;
  75     uint32_t flags;
  76     uint64_t capacity;
  77     uint64_t granularity;
  78     uint64_t desc_offset;
  79     uint64_t desc_size;
  80     /* Number of GrainTableEntries per GrainTable */
  81     uint32_t num_gtes_per_gt;
  82     uint64_t rgd_offset;
  83     uint64_t gd_offset;
  84     uint64_t grain_offset;
  85     char filler[1];
  86     char check_bytes[4];
  87     uint16_t compressAlgorithm;
  88 } QEMU_PACKED VMDK4Header;
  89
  90 #define L2_CACHE_SIZE 16
  91
  92 typedef struct VmdkExtent {
  93     BdrvChild *file;
  94     bool flat;
  95     bool compressed;
  96     bool has_marker;
  97     bool has_zero_grain;
  98     int version;
  99     int64_t sectors;
 100     int64_t end_sector;
 101     int64_t flat_start_offset;
 102     int64_t l1_table_offset;
 103     int64_t l1_backup_table_offset;
 104     uint32_t *l1_table;
 105     uint32_t *l1_backup_table;
 106     unsigned int l1_size;
 107     uint32_t l1_entry_sectors;
 108
 109     unsigned int l2_size;
 110     uint32_t *l2_cache;
 111     uint32_t l2_cache_offsets[L2_CACHE_SIZE];
 112     uint32_t l2_cache_counts[L2_CACHE_SIZE];
 113
 114     int64_t cluster_sectors;
 115     int64_t next_cluster_sector;
 116     char *type;
 117 } VmdkExtent;
 118
 119 typedef struct BDRVVmdkState {
 120     CoMutex lock;
 121     uint64_t desc_offset;
 122     bool cid_updated;
 123     bool cid_checked;
 124     uint32_t cid;
 125     uint32_t parent_cid;
 126     int num_extents;
 127     /* Extent array with num_extents entries, ascend ordered by address */
 128     VmdkExtent *extents;
 129     Error *migration_blocker;
 130     char *create_type;
 131 } BDRVVmdkState;
 132
 133 typedef struct VmdkMetaData {
 134     unsigned int l1_index;
 135     unsigned int l2_index;
 136     unsigned int l2_offset;
 137     int valid;
 138     uint32_t *l2_cache_entry;
 139 } VmdkMetaData;
 140
 141 typedef struct VmdkGrainMarker {
 142     uint64_t lba;
 143     uint32_t size;
 144     uint8_t  data[0];
 145 } QEMU_PACKED VmdkGrainMarker;
 146
 147 enum {
 148     MARKER_END_OF_STREAM    = 0,
 149     MARKER_GRAIN_TABLE      = 1,
 150     MARKER_GRAIN_DIRECTORY  = 2,
 151     MARKER_FOOTER           = 3,
 152 };
 153
 154 static int vmdk_probe(const uint8_t *buf, int buf_size, const char *filename)
 155 {
 156     uint32_t magic;
 157
 158     if (buf_size < 4) {
 159         return 0;
 160     }
 161     magic = be32_to_cpu(*(uint32_t *)buf);
 162     if (magic == VMDK3_MAGIC ||
 163         magic == VMDK4_MAGIC) {
 164         return 100;
 165     } else {
 166         const char *p = (const char *)buf;
 167         const char *end = p + buf_size;
 168         while (p < end) {
 169             if (*p == '#') {
 170                 /* skip comment line */
 171                 while (p < end && *p != '\n') {
 172                     p++;
 173                 }
 174                 p++;
 175                 continue;
 176             }
 177             if (*p == ' ') {
 178                 while (p < end && *p == ' ') {
 179                     p++;
 180                 }
 181                 /* skip '\r' if windows line endings used. */
 182                 if (p < end && *p == '\r') {
 183                     p++;
 184                 }
 185                 /* only accept blank lines before 'version=' line */
 186                 if (p == end || *p != '\n') {
 187                     return 0;
 188                 }
 189                 p++;
 190                 continue;
 191             }
 192             if (end - p >= strlen("version=X\n")) {
 193                 if (strncmp("version=1\n", p, strlen("version=1\n")) == 0 ||
 194                     strncmp("version=2\n", p, strlen("version=2\n")) == 0) {
 195                     return 100;
 196                 }
 197             }
 198             if (end - p >= strlen("version=X\r\n")) {
 199                 if (strncmp("version=1\r\n", p, strlen("version=1\r\n")) == 0 ||
 200                     strncmp("version=2\r\n", p, strlen("version=2\r\n")) == 0) {
 201                     return 100;
 202                 }
 203             }
 204             return 0;
 205         }
 206         return 0;
 207     }
 208 }
 209
 210 #define SECTOR_SIZE 512
 211 #define DESC_SIZE (20 * SECTOR_SIZE)    /* 20 sectors of 512 bytes each */
 212 #define BUF_SIZE 4096
 213 #define HEADER_SIZE 512                 /* first sector of 512 bytes */
 214
 215 static void vmdk_free_extents(BlockDriverState *bs)
 216 {
 217     int i;
 218     BDRVVmdkState *s = bs->opaque;
 219     VmdkExtent *e;
 220
 221     for (i = 0; i < s->num_extents; i++) {
 222         e = &s->extents[i];
 223         g_free(e->l1_table);
 224         g_free(e->l2_cache);
 225         g_free(e->l1_backup_table);
 226         g_free(e->type);
 227         if (e->file != bs->file) {
 228             bdrv_unref_child(bs, e->file);
 229         }
 230     }
 231     g_free(s->extents);
 232 }
 233
 234 static void vmdk_free_last_extent(BlockDriverState *bs)
 235 {
 236     BDRVVmdkState *s = bs->opaque;
 237
 238     if (s->num_extents == 0) {
 239         return;
 240     }
 241     s->num_extents--;
 242     s->extents = g_renew(VmdkExtent, s->extents, s->num_extents);
 243 }
 244
 245 static uint32_t vmdk_read_cid(BlockDriverState *bs, int parent)
 246 {
 247     char *desc;
 248     uint32_t cid = 0xffffffff;
 249     const char *p_name, *cid_str;
 250     size_t cid_str_size;
 251     BDRVVmdkState *s = bs->opaque;
 252     int ret;
 253
 254     desc = g_malloc0(DESC_SIZE);
 255     ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE);
 256     if (ret < 0) {
 257         g_free(desc);
 258         return 0;
 259     }
 260
 261     if (parent) {
 262         cid_str = "parentCID";
 263         cid_str_size = sizeof("parentCID");
 264     } else {
 265         cid_str = "CID";
 266         cid_str_size = sizeof("CID");
 267     }
 268
 269     desc[DESC_SIZE - 1] = '\0';
 270     p_name = strstr(desc, cid_str);
 271     if (p_name != NULL) {
 272         p_name += cid_str_size;
 273         sscanf(p_name, "%" SCNx32, &cid);
 274     }
 275
 276     g_free(desc);
 277     return cid;
 278 }
 279
 280 static int vmdk_write_cid(BlockDriverState *bs, uint32_t cid)
 281 {
 282     char *desc, *tmp_desc;
 283     char *p_name, *tmp_str;
 284     BDRVVmdkState *s = bs->opaque;
 285     int ret = 0;
 286
 287     desc = g_malloc0(DESC_SIZE);
 288     tmp_desc = g_malloc0(DESC_SIZE);
 289     ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE);
 290     if (ret < 0) {
 291         goto out;
 292     }
 293
 294     desc[DESC_SIZE - 1] = '\0';
 295     tmp_str = strstr(desc, "parentCID");
 296     if (tmp_str == NULL) {
 297         ret = -EINVAL;
 298         goto out;
 299     }
 300
 301     pstrcpy(tmp_desc, DESC_SIZE, tmp_str);
 302     p_name = strstr(desc, "CID");
 303     if (p_name != NULL) {
 304         p_name += sizeof("CID");
 305         snprintf(p_name, DESC_SIZE - (p_name - desc), "%" PRIx32 "\n", cid);
 306         pstrcat(desc, DESC_SIZE, tmp_desc);
 307     }
 308
 309     ret = bdrv_pwrite_sync(bs->file, s->desc_offset, desc, DESC_SIZE);
 310
 311 out:
 312     g_free(desc);
 313     g_free(tmp_desc);
 314     return ret;
 315 }
 316
 317 static int vmdk_is_cid_valid(BlockDriverState *bs)
 318 {
 319     BDRVVmdkState *s = bs->opaque;
 320     uint32_t cur_pcid;
 321
 322     if (!s->cid_checked && bs->backing) {
 323         BlockDriverState *p_bs = bs->backing->bs;
 324
 325         cur_pcid = vmdk_read_cid(p_bs, 0);
 326         if (s->parent_cid != cur_pcid) {
 327             /* CID not valid */
 328             return 0;
 329         }
 330     }
 331     s->cid_checked = true;
 332     /* CID valid */
 333     return 1;
 334 }
 335
 336 /* We have nothing to do for VMDK reopen, stubs just return success */
 337 static int vmdk_reopen_prepare(BDRVReopenState *state,
 338                                BlockReopenQueue *queue, Error **errp)
 339 {
 340     assert(state != NULL);
 341     assert(state->bs != NULL);
 342     return 0;
 343 }
 344
 345 static int vmdk_parent_open(BlockDriverState *bs)
 346 {
 347     char *p_name;
 348     char *desc;
 349     BDRVVmdkState *s = bs->opaque;
 350     int ret;
 351
 352     desc = g_malloc0(DESC_SIZE + 1);
 353     ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE);
 354     if (ret < 0) {
 355         goto out;
 356     }
 357     ret = 0;
 358
 359     p_name = strstr(desc, "parentFileNameHint");
 360     if (p_name != NULL) {
 361         char *end_name;
 362
 363         p_name += sizeof("parentFileNameHint") + 1;
 364         end_name = strchr(p_name, '\"');
 365         if (end_name == NULL) {
 366             ret = -EINVAL;
 367             goto out;
 368         }
 369         if ((end_name - p_name) > sizeof(bs->backing_file) - 1) {
 370             ret = -EINVAL;
 371             goto out;
 372         }
 373
 374         pstrcpy(bs->backing_file, end_name - p_name + 1, p_name);
 375     }
 376
 377 out:
 378     g_free(desc);
 379     return ret;
 380 }
 381
 382 /* Create and append extent to the extent array. Return the added VmdkExtent
 383  * address. return NULL if allocation failed. */
 384 static int vmdk_add_extent(BlockDriverState *bs,
 385                            BdrvChild *file, bool flat, int64_t sectors,
 386                            int64_t l1_offset, int64_t l1_backup_offset,
 387                            uint32_t l1_size,
 388                            int l2_size, uint64_t cluster_sectors,
 389                            VmdkExtent **new_extent,
 390                            Error **errp)
 391 {
 392     VmdkExtent *extent;
 393     BDRVVmdkState *s = bs->opaque;
 394     int64_t nb_sectors;
 395
 396     if (cluster_sectors > 0x200000) {
 397         /* 0x200000 * 512Bytes = 1GB for one cluster is unrealistic */
 398         error_setg(errp, "Invalid granularity, image may be corrupt");
 399         return -EFBIG;
 400     }
 401     if (l1_size > 512 * 1024 * 1024) {
 402         /* Although with big capacity and small l1_entry_sectors, we can get a
 403          * big l1_size, we don't want unbounded value to allocate the table.
 404          * Limit it to 512M, which is 16PB for default cluster and L2 table
 405          * size */
 406         error_setg(errp, "L1 size too big");
 407         return -EFBIG;
 408     }
 409
 410     nb_sectors = bdrv_nb_sectors(file->bs);
 411     if (nb_sectors < 0) {
 412         return nb_sectors;
 413     }
 414
 415     s->extents = g_renew(VmdkExtent, s->extents, s->num_extents + 1);
 416     extent = &s->extents[s->num_extents];
 417     s->num_extents++;
 418
 419     memset(extent, 0, sizeof(VmdkExtent));
 420     extent->file = file;
 421     extent->flat = flat;
 422     extent->sectors = sectors;
 423     extent->l1_table_offset = l1_offset;
 424     extent->l1_backup_table_offset = l1_backup_offset;
 425     extent->l1_size = l1_size;
 426     extent->l1_entry_sectors = l2_size * cluster_sectors;
 427     extent->l2_size = l2_size;
 428     extent->cluster_sectors = flat ? sectors : cluster_sectors;
 429     extent->next_cluster_sector = ROUND_UP(nb_sectors, cluster_sectors);
 430
 431     if (s->num_extents > 1) {
 432         extent->end_sector = (*(extent - 1)).end_sector + extent->sectors;
 433     } else {
 434         extent->end_sector = extent->sectors;
 435     }
 436     bs->total_sectors = extent->end_sector;
 437     if (new_extent) {
 438         *new_extent = extent;
 439     }
 440     return 0;
 441 }
 442
 443 static int vmdk_init_tables(BlockDriverState *bs, VmdkExtent *extent,
 444                             Error **errp)
 445 {
 446     int ret;
 447     size_t l1_size;
 448     int i;
 449
 450     /* read the L1 table */
 451     l1_size = extent->l1_size * sizeof(uint32_t);
 452     extent->l1_table = g_try_malloc(l1_size);
 453     if (l1_size && extent->l1_table == NULL) {
 454         return -ENOMEM;
 455     }
 456
 457     ret = bdrv_pread(extent->file,
 458                      extent->l1_table_offset,
 459                      extent->l1_table,
 460                      l1_size);
 461     if (ret < 0) {
 462         error_setg_errno(errp, -ret,
 463                          "Could not read l1 table from extent '%s'",
 464                          extent->file->bs->filename);
 465         goto fail_l1;
 466     }
 467     for (i = 0; i < extent->l1_size; i++) {
 468         le32_to_cpus(&extent->l1_table[i]);
 469     }
 470
 471     if (extent->l1_backup_table_offset) {
 472         extent->l1_backup_table = g_try_malloc(l1_size);
 473         if (l1_size && extent->l1_backup_table == NULL) {
 474             ret = -ENOMEM;
 475             goto fail_l1;
 476         }
 477         ret = bdrv_pread(extent->file,
 478                          extent->l1_backup_table_offset,
 479                          extent->l1_backup_table,
 480                          l1_size);
 481         if (ret < 0) {
 482             error_setg_errno(errp, -ret,
 483                              "Could not read l1 backup table from extent '%s'",
 484                              extent->file->bs->filename);
 485             goto fail_l1b;
 486         }
 487         for (i = 0; i < extent->l1_size; i++) {
 488             le32_to_cpus(&extent->l1_backup_table[i]);
 489         }
 490     }
 491
 492     extent->l2_cache =
 493         g_new(uint32_t, extent->l2_size * L2_CACHE_SIZE);
 494     return 0;
 495  fail_l1b:
 496     g_free(extent->l1_backup_table);
 497  fail_l1:
 498     g_free(extent->l1_table);
 499     return ret;
 500 }
 501
 502 static int vmdk_open_vmfs_sparse(BlockDriverState *bs,
 503                                  BdrvChild *file,
 504                                  int flags, Error **errp)
 505 {
 506     int ret;
 507     uint32_t magic;
 508     VMDK3Header header;
 509     VmdkExtent *extent;
 510
 511     ret = bdrv_pread(file, sizeof(magic), &header, sizeof(header));
 512     if (ret < 0) {
 513         error_setg_errno(errp, -ret,
 514                          "Could not read header from file '%s'",
 515                          file->bs->filename);
 516         return ret;
 517     }
 518     ret = vmdk_add_extent(bs, file, false,
 519                           le32_to_cpu(header.disk_sectors),
 520                           (int64_t)le32_to_cpu(header.l1dir_offset) << 9,
 521                           0,
 522                           le32_to_cpu(header.l1dir_size),
 523                           4096,
 524                           le32_to_cpu(header.granularity),
 525                           &extent,
 526                           errp);
 527     if (ret < 0) {
 528         return ret;
 529     }
 530     ret = vmdk_init_tables(bs, extent, errp);
 531     if (ret) {
 532         /* free extent allocated by vmdk_add_extent */
 533         vmdk_free_last_extent(bs);
 534     }
 535     return ret;
 536 }
 537
 538 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf,
 539                                QDict *options, Error **errp);
 540
 541 static char *vmdk_read_desc(BdrvChild *file, uint64_t desc_offset, Error **errp)
 542 {
 543     int64_t size;
 544     char *buf;
 545     int ret;
 546
 547     size = bdrv_getlength(file->bs);
 548     if (size < 0) {
 549         error_setg_errno(errp, -size, "Could not access file");
 550         return NULL;
 551     }
 552
 553     if (size < 4) {
 554         /* Both descriptor file and sparse image must be much larger than 4
 555          * bytes, also callers of vmdk_read_desc want to compare the first 4
 556          * bytes with VMDK4_MAGIC, let's error out if less is read. */
 557         error_setg(errp, "File is too small, not a valid image");
 558         return NULL;
 559     }
 560
 561     size = MIN(size, (1 << 20) - 1);  /* avoid unbounded allocation */
 562     buf = g_malloc(size + 1);
 563
 564     ret = bdrv_pread(file, desc_offset, buf, size);
 565     if (ret < 0) {
 566         error_setg_errno(errp, -ret, "Could not read from file");
 567         g_free(buf);
 568         return NULL;
 569     }
 570     buf[ret] = 0;
 571
 572     return buf;
 573 }
 574
 575 static int vmdk_open_vmdk4(BlockDriverState *bs,
 576                            BdrvChild *file,
 577                            int flags, QDict *options, Error **errp)
 578 {
 579     int ret;
 580     uint32_t magic;
 581     uint32_t l1_size, l1_entry_sectors;
 582     VMDK4Header header;
 583     VmdkExtent *extent;
 584     BDRVVmdkState *s = bs->opaque;
 585     int64_t l1_backup_offset = 0;
 586     bool compressed;
 587
 588     ret = bdrv_pread(file, sizeof(magic), &header, sizeof(header));
 589     if (ret < 0) {
 590         error_setg_errno(errp, -ret,
 591                          "Could not read header from file '%s'",
 592                          file->bs->filename);
 593         return -EINVAL;
 594     }
 595     if (header.capacity == 0) {
 596         uint64_t desc_offset = le64_to_cpu(header.desc_offset);
 597         if (desc_offset) {
 598             char *buf = vmdk_read_desc(file, desc_offset << 9, errp);
 599             if (!buf) {
 600                 return -EINVAL;
 601             }
 602             ret = vmdk_open_desc_file(bs, flags, buf, options, errp);
 603             g_free(buf);
 604             return ret;
 605         }
 606     }
 607
 608     if (!s->create_type) {
 609         s->create_type = g_strdup("monolithicSparse");
 610     }
 611
 612     if (le64_to_cpu(header.gd_offset) == VMDK4_GD_AT_END) {
 613         /*
 614          * The footer takes precedence over the header, so read it in. The
 615          * footer starts at offset -1024 from the end: One sector for the
 616          * footer, and another one for the end-of-stream marker.
 617          */
 618         struct {
 619             struct {
 620                 uint64_t val;
 621                 uint32_t size;
 622                 uint32_t type;
 623                 uint8_t pad[512 - 16];
 624             } QEMU_PACKED footer_marker;
 625
 626             uint32_t magic;
 627             VMDK4Header header;
 628             uint8_t pad[512 - 4 - sizeof(VMDK4Header)];
 629
 630             struct {
 631                 uint64_t val;
 632                 uint32_t size;
 633                 uint32_t type;
 634                 uint8_t pad[512 - 16];
 635             } QEMU_PACKED eos_marker;
 636         } QEMU_PACKED footer;
 637
 638         ret = bdrv_pread(file,
 639             bs->file->bs->total_sectors * 512 - 1536,
 640             &footer, sizeof(footer));
 641         if (ret < 0) {
 642             error_setg_errno(errp, -ret, "Failed to read footer");
 643             return ret;
 644         }
 645
 646         /* Some sanity checks for the footer */
 647         if (be32_to_cpu(footer.magic) != VMDK4_MAGIC ||
 648             le32_to_cpu(footer.footer_marker.size) != 0  ||
 649             le32_to_cpu(footer.footer_marker.type) != MARKER_FOOTER ||
 650             le64_to_cpu(footer.eos_marker.val) != 0  ||
 651             le32_to_cpu(footer.eos_marker.size) != 0  ||
 652             le32_to_cpu(footer.eos_marker.type) != MARKER_END_OF_STREAM)
 653         {
 654             error_setg(errp, "Invalid footer");
 655             return -EINVAL;
 656         }
 657
 658         header = footer.header;
 659     }
 660
 661     compressed =
 662         le16_to_cpu(header.compressAlgorithm) == VMDK4_COMPRESSION_DEFLATE;
 663     if (le32_to_cpu(header.version) > 3) {
 664         error_setg(errp, "Unsupported VMDK version %" PRIu32,
 665                    le32_to_cpu(header.version));
 666         return -ENOTSUP;
 667     } else if (le32_to_cpu(header.version) == 3 && (flags & BDRV_O_RDWR) &&
 668                !compressed) {
 669         /* VMware KB 2064959 explains that version 3 added support for
 670          * persistent changed block tracking (CBT), and backup software can
 671          * read it as version=1 if it doesn't care about the changed area
 672          * information. So we are safe to enable read only. */
 673         error_setg(errp, "VMDK version 3 must be read only");
 674         return -EINVAL;
 675     }
 676
 677     if (le32_to_cpu(header.num_gtes_per_gt) > 512) {
 678         error_setg(errp, "L2 table size too big");
 679         return -EINVAL;
 680     }
 681
 682     l1_entry_sectors = le32_to_cpu(header.num_gtes_per_gt)
 683                         * le64_to_cpu(header.granularity);
 684     if (l1_entry_sectors == 0) {
 685         error_setg(errp, "L1 entry size is invalid");
 686         return -EINVAL;
 687     }
 688     l1_size = (le64_to_cpu(header.capacity) + l1_entry_sectors - 1)
 689                 / l1_entry_sectors;
 690     if (le32_to_cpu(header.flags) & VMDK4_FLAG_RGD) {
 691         l1_backup_offset = le64_to_cpu(header.rgd_offset) << 9;
 692     }
 693     if (bdrv_nb_sectors(file->bs) < le64_to_cpu(header.grain_offset)) {
 694         error_setg(errp, "File truncated, expecting at least %" PRId64 " bytes",
 695                    (int64_t)(le64_to_cpu(header.grain_offset)
 696                              * BDRV_SECTOR_SIZE));
 697         return -EINVAL;
 698     }
 699
 700     ret = vmdk_add_extent(bs, file, false,
 701                           le64_to_cpu(header.capacity),
 702                           le64_to_cpu(header.gd_offset) << 9,
 703                           l1_backup_offset,
 704                           l1_size,
 705                           le32_to_cpu(header.num_gtes_per_gt),
 706                           le64_to_cpu(header.granularity),
 707                           &extent,
 708                           errp);
 709     if (ret < 0) {
 710         return ret;
 711     }
 712     extent->compressed =
 713         le16_to_cpu(header.compressAlgorithm) == VMDK4_COMPRESSION_DEFLATE;
 714     if (extent->compressed) {
 715         g_free(s->create_type);
 716         s->create_type = g_strdup("streamOptimized");
 717     }
 718     extent->has_marker = le32_to_cpu(header.flags) & VMDK4_FLAG_MARKER;
 719     extent->version = le32_to_cpu(header.version);
 720     extent->has_zero_grain = le32_to_cpu(header.flags) & VMDK4_FLAG_ZERO_GRAIN;
 721     ret = vmdk_init_tables(bs, extent, errp);
 722     if (ret) {
 723         /* free extent allocated by vmdk_add_extent */
 724         vmdk_free_last_extent(bs);
 725     }
 726     return ret;
 727 }
 728
 729 /* find an option value out of descriptor file */
 730 static int vmdk_parse_description(const char *desc, const char *opt_name,
 731         char *buf, int buf_size)
 732 {
 733     char *opt_pos, *opt_end;
 734     const char *end = desc + strlen(desc);
 735
 736     opt_pos = strstr(desc, opt_name);
 737     if (!opt_pos) {
 738         return VMDK_ERROR;
 739     }
 740     /* Skip "=\"" following opt_name */
 741     opt_pos += strlen(opt_name) + 2;
 742     if (opt_pos >= end) {
 743         return VMDK_ERROR;
 744     }
 745     opt_end = opt_pos;
 746     while (opt_end < end && *opt_end != '"') {
 747         opt_end++;
 748     }
 749     if (opt_end == end || buf_size < opt_end - opt_pos + 1) {
 750         return VMDK_ERROR;
 751     }
 752     pstrcpy(buf, opt_end - opt_pos + 1, opt_pos);
 753     return VMDK_OK;
 754 }
 755
 756 /* Open an extent file and append to bs array */
 757 static int vmdk_open_sparse(BlockDriverState *bs, BdrvChild *file, int flags,
 758                             char *buf, QDict *options, Error **errp)
 759 {
 760     uint32_t magic;
 761
 762     magic = ldl_be_p(buf);
 763     switch (magic) {
 764         case VMDK3_MAGIC:
 765             return vmdk_open_vmfs_sparse(bs, file, flags, errp);
 766             break;
 767         case VMDK4_MAGIC:
 768             return vmdk_open_vmdk4(bs, file, flags, options, errp);
 769             break;
 770         default:
 771             error_setg(errp, "Image not in VMDK format");
 772             return -EINVAL;
 773             break;
 774     }
 775 }
 776
 777 static const char *next_line(const char *s)
 778 {
 779     while (*s) {
 780         if (*s == '\n') {
 781             return s + 1;
 782         }
 783         s++;
 784     }
 785     return s;
 786 }
 787
 788 static int vmdk_parse_extents(const char *desc, BlockDriverState *bs,
 789                               const char *desc_file_path, QDict *options,
 790                               Error **errp)
 791 {
 792     int ret;
 793     int matches;
 794     char access[11];
 795     char type[11];
 796     char fname[512];
 797     const char *p, *np;
 798     int64_t sectors = 0;
 799     int64_t flat_offset;
 800     char *extent_path;
 801     BdrvChild *extent_file;
 802     BDRVVmdkState *s = bs->opaque;
 803     VmdkExtent *extent;
 804     char extent_opt_prefix[32];
 805     Error *local_err = NULL;
 806
 807     for (p = desc; *p; p = next_line(p)) {
 808         /* parse extent line in one of below formats:
 809          *
 810          * RW [size in sectors] FLAT "file-name.vmdk" OFFSET
 811          * RW [size in sectors] SPARSE "file-name.vmdk"
 812          * RW [size in sectors] VMFS "file-name.vmdk"
 813          * RW [size in sectors] VMFSSPARSE "file-name.vmdk"
 814          */
 815         flat_offset = -1;
 816         matches = sscanf(p, "%10s %" SCNd64 " %10s \"%511[^\n\r\"]\" %" SCNd64,
 817                          access, &sectors, type, fname, &flat_offset);
 818         if (matches < 4 || strcmp(access, "RW")) {
 819             continue;
 820         } else if (!strcmp(type, "FLAT")) {
 821             if (matches != 5 || flat_offset < 0) {
 822                 goto invalid;
 823             }
 824         } else if (!strcmp(type, "VMFS")) {
 825             if (matches == 4) {
 826                 flat_offset = 0;
 827             } else {
 828                 goto invalid;
 829             }
 830         } else if (matches != 4) {
 831             goto invalid;
 832         }
 833
 834         if (sectors <= 0 ||
 835             (strcmp(type, "FLAT") && strcmp(type, "SPARSE") &&
 836              strcmp(type, "VMFS") && strcmp(type, "VMFSSPARSE")) ||
 837             (strcmp(access, "RW"))) {
 838             continue;
 839         }
 840
 841         if (!path_is_absolute(fname) && !path_has_protocol(fname) &&
 842             !desc_file_path[0])
 843         {
 844             error_setg(errp, "Cannot use relative extent paths with VMDK "
 845                        "descriptor file '%s'", bs->file->bs->filename);
 846             return -EINVAL;
 847         }
 848
 849         extent_path = g_malloc0(PATH_MAX);
 850         path_combine(extent_path, PATH_MAX, desc_file_path, fname);
 851
 852         ret = snprintf(extent_opt_prefix, 32, "extents.%d", s->num_extents);
 853         assert(ret < 32);
 854
 855         extent_file = bdrv_open_child(extent_path, options, extent_opt_prefix,
 856                                       bs, &child_file, false, &local_err);
 857         g_free(extent_path);
 858         if (local_err) {
 859             error_propagate(errp, local_err);
 860             return -EINVAL;
 861         }
 862
 863         /* save to extents array */
 864         if (!strcmp(type, "FLAT") || !strcmp(type, "VMFS")) {
 865             /* FLAT extent */
 866
 867             ret = vmdk_add_extent(bs, extent_file, true, sectors,
 868                             0, 0, 0, 0, 0, &extent, errp);
 869             if (ret < 0) {
 870                 bdrv_unref_child(bs, extent_file);
 871                 return ret;
 872             }
 873             extent->flat_start_offset = flat_offset << 9;
 874         } else if (!strcmp(type, "SPARSE") || !strcmp(type, "VMFSSPARSE")) {
 875             /* SPARSE extent and VMFSSPARSE extent are both "COWD" sparse file*/
 876             char *buf = vmdk_read_desc(extent_file, 0, errp);
 877             if (!buf) {
 878                 ret = -EINVAL;
 879             } else {
 880                 ret = vmdk_open_sparse(bs, extent_file, bs->open_flags, buf,
 881                                        options, errp);
 882             }
 883             g_free(buf);
 884             if (ret) {
 885                 bdrv_unref_child(bs, extent_file);
 886                 return ret;
 887             }
 888             extent = &s->extents[s->num_extents - 1];
 889         } else {
 890             error_setg(errp, "Unsupported extent type '%s'", type);
 891             bdrv_unref_child(bs, extent_file);
 892             return -ENOTSUP;
 893         }
 894         extent->type = g_strdup(type);
 895     }
 896     return 0;
 897
 898 invalid:
 899     np = next_line(p);
 900     assert(np != p);
 901     if (np[-1] == '\n') {
 902         np--;
 903     }
 904     error_setg(errp, "Invalid extent line: %.*s", (int)(np - p), p);
 905     return -EINVAL;
 906 }
 907
 908 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf,
 909                                QDict *options, Error **errp)
 910 {
 911     int ret;
 912     char ct[128];
 913     BDRVVmdkState *s = bs->opaque;
 914
 915     if (vmdk_parse_description(buf, "createType", ct, sizeof(ct))) {
 916         error_setg(errp, "invalid VMDK image descriptor");
 917         ret = -EINVAL;
 918         goto exit;
 919     }
 920     if (strcmp(ct, "monolithicFlat") &&
 921         strcmp(ct, "vmfs") &&
 922         strcmp(ct, "vmfsSparse") &&
 923         strcmp(ct, "twoGbMaxExtentSparse") &&
 924         strcmp(ct, "twoGbMaxExtentFlat")) {
 925         error_setg(errp, "Unsupported image type '%s'", ct);
 926         ret = -ENOTSUP;
 927         goto exit;
 928     }
 929     s->create_type = g_strdup(ct);
 930     s->desc_offset = 0;
 931     ret = vmdk_parse_extents(buf, bs, bs->file->bs->exact_filename, options,
 932                              errp);
 933 exit:
 934     return ret;
 935 }
 936
 937 static int vmdk_open(BlockDriverState *bs, QDict *options, int flags,
 938                      Error **errp)
 939 {
 940     char *buf;
 941     int ret;
 942     BDRVVmdkState *s = bs->opaque;
 943     uint32_t magic;
 944
 945     buf = vmdk_read_desc(bs->file, 0, errp);
 946     if (!buf) {
 947         return -EINVAL;
 948     }
 949
 950     magic = ldl_be_p(buf);
 951     switch (magic) {
 952         case VMDK3_MAGIC:
 953         case VMDK4_MAGIC:
 954             ret = vmdk_open_sparse(bs, bs->file, flags, buf, options,
 955                                    errp);
 956             s->desc_offset = 0x200;
 957             break;
 958         default:
 959             ret = vmdk_open_desc_file(bs, flags, buf, options, errp);
 960             break;
 961     }
 962     if (ret) {
 963         goto fail;
 964     }
 965
 966     /* try to open parent images, if exist */
 967     ret = vmdk_parent_open(bs);
 968     if (ret) {
 969         goto fail;
 970     }
 971     s->cid = vmdk_read_cid(bs, 0);
 972     s->parent_cid = vmdk_read_cid(bs, 1);
 973     qemu_co_mutex_init(&s->lock);
 974
 975     /* Disable migration when VMDK images are used */
 976     error_setg(&s->migration_blocker, "The vmdk format used by node '%s' "
 977                "does not support live migration",
 978                bdrv_get_device_or_node_name(bs));
 979     migrate_add_blocker(s->migration_blocker);
 980     g_free(buf);
 981     return 0;
 982
 983 fail:
 984     g_free(buf);
 985     g_free(s->create_type);
 986     s->create_type = NULL;
 987     vmdk_free_extents(bs);
 988     return ret;
 989 }
 990
 991
 992 static void vmdk_refresh_limits(BlockDriverState *bs, Error **errp)
 993 {
 994     BDRVVmdkState *s = bs->opaque;
 995     int i;
 996
 997     for (i = 0; i < s->num_extents; i++) {
 998         if (!s->extents[i].flat) {
 999             bs->bl.pwrite_zeroes_alignment =
1000                 MAX(bs->bl.pwrite_zeroes_alignment,
1001                     s->extents[i].cluster_sectors << BDRV_SECTOR_BITS);
1002         }
1003     }
1004 }
1005
1006 /**
1007  * get_whole_cluster
1008  *
1009  * Copy backing file's cluster that covers @sector_num, otherwise write zero,
1010  * to the cluster at @cluster_sector_num.
1011  *
1012  * If @skip_start_sector < @skip_end_sector, the relative range
1013  * [@skip_start_sector, @skip_end_sector) is not copied or written, and leave
1014  * it for call to write user data in the request.
1015  */
1016 static int get_whole_cluster(BlockDriverState *bs,
1017                              VmdkExtent *extent,
1018                              uint64_t cluster_offset,
1019                              uint64_t offset,
1020                              uint64_t skip_start_bytes,
1021                              uint64_t skip_end_bytes)
1022 {
1023     int ret = VMDK_OK;
1024     int64_t cluster_bytes;
1025     uint8_t *whole_grain;
1026
1027     /* For COW, align request sector_num to cluster start */
1028     cluster_bytes = extent->cluster_sectors << BDRV_SECTOR_BITS;
1029     offset = QEMU_ALIGN_DOWN(offset, cluster_bytes);
1030     whole_grain = qemu_blockalign(bs, cluster_bytes);
1031
1032     if (!bs->backing) {
1033         memset(whole_grain, 0, skip_start_bytes);
1034         memset(whole_grain + skip_end_bytes, 0, cluster_bytes - skip_end_bytes);
1035     }
1036
1037     assert(skip_end_bytes <= cluster_bytes);
1038     /* we will be here if it's first write on non-exist grain(cluster).
1039      * try to read from parent image, if exist */
1040     if (bs->backing && !vmdk_is_cid_valid(bs)) {
1041         ret = VMDK_ERROR;
1042         goto exit;
1043     }
1044
1045     /* Read backing data before skip range */
1046     if (skip_start_bytes > 0) {
1047         if (bs->backing) {
1048             ret = bdrv_pread(bs->backing, offset, whole_grain,
1049                              skip_start_bytes);
1050             if (ret < 0) {
1051                 ret = VMDK_ERROR;
1052                 goto exit;
1053             }
1054         }
1055         ret = bdrv_pwrite(extent->file, cluster_offset, whole_grain,
1056                           skip_start_bytes);
1057         if (ret < 0) {
1058             ret = VMDK_ERROR;
1059             goto exit;
1060         }
1061     }
1062     /* Read backing data after skip range */
1063     if (skip_end_bytes < cluster_bytes) {
1064         if (bs->backing) {
1065             ret = bdrv_pread(bs->backing, offset + skip_end_bytes,
1066                              whole_grain + skip_end_bytes,
1067                              cluster_bytes - skip_end_bytes);
1068             if (ret < 0) {
1069                 ret = VMDK_ERROR;
1070                 goto exit;
1071             }
1072         }
1073         ret = bdrv_pwrite(extent->file, cluster_offset + skip_end_bytes,
1074                           whole_grain + skip_end_bytes,
1075                           cluster_bytes - skip_end_bytes);
1076         if (ret < 0) {
1077             ret = VMDK_ERROR;
1078             goto exit;
1079         }
1080     }
1081
1082     ret = VMDK_OK;
1083 exit:
1084     qemu_vfree(whole_grain);
1085     return ret;
1086 }
1087
1088 static int vmdk_L2update(VmdkExtent *extent, VmdkMetaData *m_data,
1089                          uint32_t offset)
1090 {
1091     offset = cpu_to_le32(offset);
1092     /* update L2 table */
1093     if (bdrv_pwrite_sync(extent->file,
1094                 ((int64_t)m_data->l2_offset * 512)
1095                     + (m_data->l2_index * sizeof(offset)),
1096                 &offset, sizeof(offset)) < 0) {
1097         return VMDK_ERROR;
1098     }
1099     /* update backup L2 table */
1100     if (extent->l1_backup_table_offset != 0) {
1101         m_data->l2_offset = extent->l1_backup_table[m_data->l1_index];
1102         if (bdrv_pwrite_sync(extent->file,
1103                     ((int64_t)m_data->l2_offset * 512)
1104                         + (m_data->l2_index * sizeof(offset)),
1105                     &offset, sizeof(offset)) < 0) {
1106             return VMDK_ERROR;
1107         }
1108     }
1109     if (m_data->l2_cache_entry) {
1110         *m_data->l2_cache_entry = offset;
1111     }
1112
1113     return VMDK_OK;
1114 }
1115
1116 /**
1117  * get_cluster_offset
1118  *
1119  * Look up cluster offset in extent file by sector number, and store in
1120  * @cluster_offset.
1121  *
1122  * For flat extents, the start offset as parsed from the description file is
1123  * returned.
1124  *
1125  * For sparse extents, look up in L1, L2 table. If allocate is true, return an
1126  * offset for a new cluster and update L2 cache. If there is a backing file,
1127  * COW is done before returning; otherwise, zeroes are written to the allocated
1128  * cluster. Both COW and zero writing skips the sector range
1129  * [@skip_start_sector, @skip_end_sector) passed in by caller, because caller
1130  * has new data to write there.
1131  *
1132  * Returns: VMDK_OK if cluster exists and mapped in the image.
1133  *          VMDK_UNALLOC if cluster is not mapped and @allocate is false.
1134  *          VMDK_ERROR if failed.
1135  */
1136 static int get_cluster_offset(BlockDriverState *bs,
1137                               VmdkExtent *extent,
1138                               VmdkMetaData *m_data,
1139                               uint64_t offset,
1140                               bool allocate,
1141                               uint64_t *cluster_offset,
1142                               uint64_t skip_start_bytes,
1143                               uint64_t skip_end_bytes)
1144 {
1145     unsigned int l1_index, l2_offset, l2_index;
1146     int min_index, i, j;
1147     uint32_t min_count, *l2_table;
1148     bool zeroed = false;
1149     int64_t ret;
1150     int64_t cluster_sector;
1151
1152     if (m_data) {
1153         m_data->valid = 0;
1154     }
1155     if (extent->flat) {
1156         *cluster_offset = extent->flat_start_offset;
1157         return VMDK_OK;
1158     }
1159
1160     offset -= (extent->end_sector - extent->sectors) * SECTOR_SIZE;
1161     l1_index = (offset >> 9) / extent->l1_entry_sectors;
1162     if (l1_index >= extent->l1_size) {
1163         return VMDK_ERROR;
1164     }
1165     l2_offset = extent->l1_table[l1_index];
1166     if (!l2_offset) {
1167         return VMDK_UNALLOC;
1168     }
1169     for (i = 0; i < L2_CACHE_SIZE; i++) {
1170         if (l2_offset == extent->l2_cache_offsets[i]) {
1171             /* increment the hit count */
1172             if (++extent->l2_cache_counts[i] == 0xffffffff) {
1173                 for (j = 0; j < L2_CACHE_SIZE; j++) {
1174                     extent->l2_cache_counts[j] >>= 1;
1175                 }
1176             }
1177             l2_table = extent->l2_cache + (i * extent->l2_size);
1178             goto found;
1179         }
1180     }
1181     /* not found: load a new entry in the least used one */
1182     min_index = 0;
1183     min_count = 0xffffffff;
1184     for (i = 0; i < L2_CACHE_SIZE; i++) {
1185         if (extent->l2_cache_counts[i] < min_count) {
1186             min_count = extent->l2_cache_counts[i];
1187             min_index = i;
1188         }
1189     }
1190     l2_table = extent->l2_cache + (min_index * extent->l2_size);
1191     if (bdrv_pread(extent->file,
1192                 (int64_t)l2_offset * 512,
1193                 l2_table,
1194                 extent->l2_size * sizeof(uint32_t)
1195             ) != extent->l2_size * sizeof(uint32_t)) {
1196         return VMDK_ERROR;
1197     }
1198
1199     extent->l2_cache_offsets[min_index] = l2_offset;
1200     extent->l2_cache_counts[min_index] = 1;
1201  found:
1202     l2_index = ((offset >> 9) / extent->cluster_sectors) % extent->l2_size;
1203     cluster_sector = le32_to_cpu(l2_table[l2_index]);
1204
1205     if (extent->has_zero_grain && cluster_sector == VMDK_GTE_ZEROED) {
1206         zeroed = true;
1207     }
1208
1209     if (!cluster_sector || zeroed) {
1210         if (!allocate) {
1211             return zeroed ? VMDK_ZEROED : VMDK_UNALLOC;
1212         }
1213
1214         cluster_sector = extent->next_cluster_sector;
1215         extent->next_cluster_sector += extent->cluster_sectors;
1216
1217         /* First of all we write grain itself, to avoid race condition
1218          * that may to corrupt the image.
1219          * This problem may occur because of insufficient space on host disk
1220          * or inappropriate VM shutdown.
1221          */
1222         ret = get_whole_cluster(bs, extent, cluster_sector * BDRV_SECTOR_SIZE,
1223                                 offset, skip_start_bytes, skip_end_bytes);
1224         if (ret) {
1225             return ret;
1226         }
1227         if (m_data) {
1228             m_data->valid = 1;
1229             m_data->l1_index = l1_index;
1230             m_data->l2_index = l2_index;
1231             m_data->l2_offset = l2_offset;
1232             m_data->l2_cache_entry = &l2_table[l2_index];
1233         }
1234     }
1235     *cluster_offset = cluster_sector << BDRV_SECTOR_BITS;
1236     return VMDK_OK;
1237 }
1238
1239 static VmdkExtent *find_extent(BDRVVmdkState *s,
1240                                 int64_t sector_num, VmdkExtent *start_hint)
1241 {
1242     VmdkExtent *extent = start_hint;
1243
1244     if (!extent) {
1245         extent = &s->extents[0];
1246     }
1247     while (extent < &s->extents[s->num_extents]) {
1248         if (sector_num < extent->end_sector) {
1249             return extent;
1250         }
1251         extent++;
1252     }
1253     return NULL;
1254 }
1255
1256 static inline uint64_t vmdk_find_offset_in_cluster(VmdkExtent *extent,
1257                                                    int64_t offset)
1258 {
1259     uint64_t extent_begin_offset, extent_relative_offset;
1260     uint64_t cluster_size = extent->cluster_sectors * BDRV_SECTOR_SIZE;
1261
1262     extent_begin_offset =
1263         (extent->end_sector - extent->sectors) * BDRV_SECTOR_SIZE;
1264     extent_relative_offset = offset - extent_begin_offset;
1265     return extent_relative_offset % cluster_size;
1266 }
1267
1268 static inline uint64_t vmdk_find_index_in_cluster(VmdkExtent *extent,
1269                                                   int64_t sector_num)
1270 {
1271     uint64_t offset;
1272     offset = vmdk_find_offset_in_cluster(extent, sector_num * BDRV_SECTOR_SIZE);
1273     return offset / BDRV_SECTOR_SIZE;
1274 }
1275
1276 static int64_t coroutine_fn vmdk_co_get_block_status(BlockDriverState *bs,
1277         int64_t sector_num, int nb_sectors, int *pnum, BlockDriverState **file)
1278 {
1279     BDRVVmdkState *s = bs->opaque;
1280     int64_t index_in_cluster, n, ret;
1281     uint64_t offset;
1282     VmdkExtent *extent;
1283
1284     extent = find_extent(s, sector_num, NULL);
1285     if (!extent) {
1286         return 0;
1287     }
1288     qemu_co_mutex_lock(&s->lock);
1289     ret = get_cluster_offset(bs, extent, NULL,
1290                              sector_num * 512, false, &offset,
1291                              0, 0);
1292     qemu_co_mutex_unlock(&s->lock);
1293
1294     index_in_cluster = vmdk_find_index_in_cluster(extent, sector_num);
1295     switch (ret) {
1296     case VMDK_ERROR:
1297         ret = -EIO;
1298         break;
1299     case VMDK_UNALLOC:
1300         ret = 0;
1301         break;
1302     case VMDK_ZEROED:
1303         ret = BDRV_BLOCK_ZERO;
1304         break;
1305     case VMDK_OK:
1306         ret = BDRV_BLOCK_DATA;
1307         if (!extent->compressed) {
1308             ret |= BDRV_BLOCK_OFFSET_VALID;
1309             ret |= (offset + (index_in_cluster << BDRV_SECTOR_BITS))
1310                     & BDRV_BLOCK_OFFSET_MASK;
1311         }
1312         *file = extent->file->bs;
1313         break;
1314     }
1315
1316     n = extent->cluster_sectors - index_in_cluster;
1317     if (n > nb_sectors) {
1318         n = nb_sectors;
1319     }
1320     *pnum = n;
1321     return ret;
1322 }
1323
1324 static int vmdk_write_extent(VmdkExtent *extent, int64_t cluster_offset,
1325                             int64_t offset_in_cluster, QEMUIOVector *qiov,
1326                             uint64_t qiov_offset, uint64_t n_bytes,
1327                             uint64_t offset)
1328 {
1329     int ret;
1330     VmdkGrainMarker *data = NULL;
1331     uLongf buf_len;
1332     QEMUIOVector local_qiov;
1333     struct iovec iov;
1334     int64_t write_offset;
1335     int64_t write_end_sector;
1336
1337     if (extent->compressed) {
1338         void *compressed_data;
1339
1340         if (!extent->has_marker) {
1341             ret = -EINVAL;
1342             goto out;
1343         }
1344         buf_len = (extent->cluster_sectors << 9) * 2;
1345         data = g_malloc(buf_len + sizeof(VmdkGrainMarker));
1346
1347         compressed_data = g_malloc(n_bytes);
1348         qemu_iovec_to_buf(qiov, qiov_offset, compressed_data, n_bytes);
1349         ret = compress(data->data, &buf_len, compressed_data, n_bytes);
1350         g_free(compressed_data);
1351
1352         if (ret != Z_OK || buf_len == 0) {
1353             ret = -EINVAL;
1354             goto out;
1355         }
1356
1357         data->lba = offset >> BDRV_SECTOR_BITS;
1358         data->size = buf_len;
1359
1360         n_bytes = buf_len + sizeof(VmdkGrainMarker);
1361         iov = (struct iovec) {
1362             .iov_base   = data,
1363             .iov_len    = n_bytes,
1364         };
1365         qemu_iovec_init_external(&local_qiov, &iov, 1);
1366     } else {
1367         qemu_iovec_init(&local_qiov, qiov->niov);
1368         qemu_iovec_concat(&local_qiov, qiov, qiov_offset, n_bytes);
1369     }
1370
1371     write_offset = cluster_offset + offset_in_cluster,
1372     ret = bdrv_co_pwritev(extent->file, write_offset, n_bytes,
1373                           &local_qiov, 0);
1374
1375     write_end_sector = DIV_ROUND_UP(write_offset + n_bytes, BDRV_SECTOR_SIZE);
1376
1377     if (extent->compressed) {
1378         extent->next_cluster_sector = write_end_sector;
1379     } else {
1380         extent->next_cluster_sector = MAX(extent->next_cluster_sector,
1381                                           write_end_sector);
1382     }
1383
1384     if (ret < 0) {
1385         goto out;
1386     }
1387     ret = 0;
1388  out:
1389     g_free(data);
1390     if (!extent->compressed) {
1391         qemu_iovec_destroy(&local_qiov);
1392     }
1393     return ret;
1394 }
1395
1396 static int vmdk_read_extent(VmdkExtent *extent, int64_t cluster_offset,
1397                             int64_t offset_in_cluster, QEMUIOVector *qiov,
1398                             int bytes)
1399 {
1400     int ret;
1401     int cluster_bytes, buf_bytes;
1402     uint8_t *cluster_buf, *compressed_data;
1403     uint8_t *uncomp_buf;
1404     uint32_t data_len;
1405     VmdkGrainMarker *marker;
1406     uLongf buf_len;
1407
1408
1409     if (!extent->compressed) {
1410         ret = bdrv_co_preadv(extent->file,
1411                              cluster_offset + offset_in_cluster, bytes,
1412                              qiov, 0);
1413         if (ret < 0) {
1414             return ret;
1415         }
1416         return 0;
1417     }
1418     cluster_bytes = extent->cluster_sectors * 512;
1419     /* Read two clusters in case GrainMarker + compressed data > one cluster */
1420     buf_bytes = cluster_bytes * 2;
1421     cluster_buf = g_malloc(buf_bytes);
1422     uncomp_buf = g_malloc(cluster_bytes);
1423     ret = bdrv_pread(extent->file,
1424                 cluster_offset,
1425                 cluster_buf, buf_bytes);
1426     if (ret < 0) {
1427         goto out;
1428     }
1429     compressed_data = cluster_buf;
1430     buf_len = cluster_bytes;
1431     data_len = cluster_bytes;
1432     if (extent->has_marker) {
1433         marker = (VmdkGrainMarker *)cluster_buf;
1434         compressed_data = marker->data;
1435         data_len = le32_to_cpu(marker->size);
1436     }
1437     if (!data_len || data_len > buf_bytes) {
1438         ret = -EINVAL;
1439         goto out;
1440     }
1441     ret = uncompress(uncomp_buf, &buf_len, compressed_data, data_len);
1442     if (ret != Z_OK) {
1443         ret = -EINVAL;
1444         goto out;
1445
1446     }
1447     if (offset_in_cluster < 0 ||
1448             offset_in_cluster + bytes > buf_len) {
1449         ret = -EINVAL;
1450         goto out;
1451     }
1452     qemu_iovec_from_buf(qiov, 0, uncomp_buf + offset_in_cluster, bytes);
1453     ret = 0;
1454
1455  out:
1456     g_free(uncomp_buf);
1457     g_free(cluster_buf);
1458     return ret;
1459 }
1460
1461 static int coroutine_fn
1462 vmdk_co_preadv(BlockDriverState *bs, uint64_t offset, uint64_t bytes,
1463                QEMUIOVector *qiov, int flags)
1464 {
1465     BDRVVmdkState *s = bs->opaque;
1466     int ret;
1467     uint64_t n_bytes, offset_in_cluster;
1468     VmdkExtent *extent = NULL;
1469     QEMUIOVector local_qiov;
1470     uint64_t cluster_offset;
1471     uint64_t bytes_done = 0;
1472
1473     qemu_iovec_init(&local_qiov, qiov->niov);
1474     qemu_co_mutex_lock(&s->lock);
1475
1476     while (bytes > 0) {
1477         extent = find_extent(s, offset >> BDRV_SECTOR_BITS, extent);
1478         if (!extent) {
1479             ret = -EIO;
1480             goto fail;
1481         }
1482         ret = get_cluster_offset(bs, extent, NULL,
1483                                  offset, false, &cluster_offset, 0, 0);
1484         offset_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
1485
1486         n_bytes = MIN(bytes, extent->cluster_sectors * BDRV_SECTOR_SIZE
1487                              - offset_in_cluster);
1488
1489         if (ret != VMDK_OK) {
1490             /* if not allocated, try to read from parent image, if exist */
1491             if (bs->backing && ret != VMDK_ZEROED) {
1492                 if (!vmdk_is_cid_valid(bs)) {
1493                     ret = -EINVAL;
1494                     goto fail;
1495                 }
1496
1497                 qemu_iovec_reset(&local_qiov);
1498                 qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes);
1499
1500                 ret = bdrv_co_preadv(bs->backing, offset, n_bytes,
1501                                      &local_qiov, 0);
1502                 if (ret < 0) {
1503                     goto fail;
1504                 }
1505             } else {
1506                 qemu_iovec_memset(qiov, bytes_done, 0, n_bytes);
1507             }
1508         } else {
1509             qemu_iovec_reset(&local_qiov);
1510             qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes);
1511
1512             ret = vmdk_read_extent(extent, cluster_offset, offset_in_cluster,
1513                                    &local_qiov, n_bytes);
1514             if (ret) {
1515                 goto fail;
1516             }
1517         }
1518         bytes -= n_bytes;
1519         offset += n_bytes;
1520         bytes_done += n_bytes;
1521     }
1522
1523     ret = 0;
1524 fail:
1525     qemu_co_mutex_unlock(&s->lock);
1526     qemu_iovec_destroy(&local_qiov);
1527
1528     return ret;
1529 }
1530
1531 /**
1532  * vmdk_write:
1533  * @zeroed:       buf is ignored (data is zero), use zeroed_grain GTE feature
1534  *                if possible, otherwise return -ENOTSUP.
1535  * @zero_dry_run: used for zeroed == true only, don't update L2 table, just try
1536  *                with each cluster. By dry run we can find if the zero write
1537  *                is possible without modifying image data.
1538  *
1539  * Returns: error code with 0 for success.
1540  */
1541 static int vmdk_pwritev(BlockDriverState *bs, uint64_t offset,
1542                        uint64_t bytes, QEMUIOVector *qiov,
1543                        bool zeroed, bool zero_dry_run)
1544 {
1545     BDRVVmdkState *s = bs->opaque;
1546     VmdkExtent *extent = NULL;
1547     int ret;
1548     int64_t offset_in_cluster, n_bytes;
1549     uint64_t cluster_offset;
1550     uint64_t bytes_done = 0;
1551     VmdkMetaData m_data;
1552
1553     if (DIV_ROUND_UP(offset, BDRV_SECTOR_SIZE) > bs->total_sectors) {
1554         error_report("Wrong offset: offset=0x%" PRIx64
1555                      " total_sectors=0x%" PRIx64,
1556                      offset, bs->total_sectors);
1557         return -EIO;
1558     }
1559
1560     while (bytes > 0) {
1561         extent = find_extent(s, offset >> BDRV_SECTOR_BITS, extent);
1562         if (!extent) {
1563             return -EIO;
1564         }
1565         offset_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
1566         n_bytes = MIN(bytes, extent->cluster_sectors * BDRV_SECTOR_SIZE
1567                              - offset_in_cluster);
1568
1569         ret = get_cluster_offset(bs, extent, &m_data, offset,
1570                                  !(extent->compressed || zeroed),
1571                                  &cluster_offset, offset_in_cluster,
1572                                  offset_in_cluster + n_bytes);
1573         if (extent->compressed) {
1574             if (ret == VMDK_OK) {
1575                 /* Refuse write to allocated cluster for streamOptimized */
1576                 error_report("Could not write to allocated cluster"
1577                               " for streamOptimized");
1578                 return -EIO;
1579             } else {
1580                 /* allocate */
1581                 ret = get_cluster_offset(bs, extent, &m_data, offset,
1582                                          true, &cluster_offset, 0, 0);
1583             }
1584         }
1585         if (ret == VMDK_ERROR) {
1586             return -EINVAL;
1587         }
1588         if (zeroed) {
1589             /* Do zeroed write, buf is ignored */
1590             if (extent->has_zero_grain &&
1591                     offset_in_cluster == 0 &&
1592                     n_bytes >= extent->cluster_sectors * BDRV_SECTOR_SIZE) {
1593                 n_bytes = extent->cluster_sectors * BDRV_SECTOR_SIZE;
1594                 if (!zero_dry_run) {
1595                     /* update L2 tables */
1596                     if (vmdk_L2update(extent, &m_data, VMDK_GTE_ZEROED)
1597                             != VMDK_OK) {
1598                         return -EIO;
1599                     }
1600                 }
1601             } else {
1602                 return -ENOTSUP;
1603             }
1604         } else {
1605             ret = vmdk_write_extent(extent, cluster_offset, offset_in_cluster,
1606                                     qiov, bytes_done, n_bytes, offset);
1607             if (ret) {
1608                 return ret;
1609             }
1610             if (m_data.valid) {
1611                 /* update L2 tables */
1612                 if (vmdk_L2update(extent, &m_data,
1613                                   cluster_offset >> BDRV_SECTOR_BITS)
1614                         != VMDK_OK) {
1615                     return -EIO;
1616                 }
1617             }
1618         }
1619         bytes -= n_bytes;
1620         offset += n_bytes;
1621         bytes_done += n_bytes;
1622
1623         /* update CID on the first write every time the virtual disk is
1624          * opened */
1625         if (!s->cid_updated) {
1626             ret = vmdk_write_cid(bs, g_random_int());
1627             if (ret < 0) {
1628                 return ret;
1629             }
1630             s->cid_updated = true;
1631         }
1632     }
1633     return 0;
1634 }
1635
1636 static int coroutine_fn
1637 vmdk_co_pwritev(BlockDriverState *bs, uint64_t offset, uint64_t bytes,
1638                 QEMUIOVector *qiov, int flags)
1639 {
1640     int ret;
1641     BDRVVmdkState *s = bs->opaque;
1642     qemu_co_mutex_lock(&s->lock);
1643     ret = vmdk_pwritev(bs, offset, bytes, qiov, false, false);
1644     qemu_co_mutex_unlock(&s->lock);
1645     return ret;
1646 }
1647
1648 typedef struct VmdkWriteCompressedCo {
1649     BlockDriverState *bs;
1650     int64_t sector_num;
1651     const uint8_t *buf;
1652     int nb_sectors;
1653     int ret;
1654 } VmdkWriteCompressedCo;
1655
1656 static void vmdk_co_write_compressed(void *opaque)
1657 {
1658     VmdkWriteCompressedCo *co = opaque;
1659     QEMUIOVector local_qiov;
1660     uint64_t offset = co->sector_num * BDRV_SECTOR_SIZE;
1661     uint64_t bytes = co->nb_sectors * BDRV_SECTOR_SIZE;
1662
1663     struct iovec iov = (struct iovec) {
1664         .iov_base   = (uint8_t*) co->buf,
1665         .iov_len    = bytes,
1666     };
1667     qemu_iovec_init_external(&local_qiov, &iov, 1);
1668
1669     co->ret = vmdk_pwritev(co->bs, offset, bytes, &local_qiov, false, false);
1670 }
1671
1672 static int vmdk_write_compressed(BlockDriverState *bs,
1673                                  int64_t sector_num,
1674                                  const uint8_t *buf,
1675                                  int nb_sectors)
1676 {
1677     BDRVVmdkState *s = bs->opaque;
1678
1679     if (s->num_extents == 1 && s->extents[0].compressed) {
1680         Coroutine *co;
1681         AioContext *aio_context = bdrv_get_aio_context(bs);
1682         VmdkWriteCompressedCo data = {
1683             .bs         = bs,
1684             .sector_num = sector_num,
1685             .buf        = buf,
1686             .nb_sectors = nb_sectors,
1687             .ret        = -EINPROGRESS,
1688         };
1689         co = qemu_coroutine_create(vmdk_co_write_compressed, &data);
1690         qemu_coroutine_enter(co);
1691         while (data.ret == -EINPROGRESS) {
1692             aio_poll(aio_context, true);
1693         }
1694         return data.ret;
1695     } else {
1696         return -ENOTSUP;
1697     }
1698 }
1699
1700 static int coroutine_fn vmdk_co_pwrite_zeroes(BlockDriverState *bs,
1701                                               int64_t offset,
1702                                               int bytes,
1703                                               BdrvRequestFlags flags)
1704 {
1705     int ret;
1706     BDRVVmdkState *s = bs->opaque;
1707
1708     qemu_co_mutex_lock(&s->lock);
1709     /* write zeroes could fail if sectors not aligned to cluster, test it with
1710      * dry_run == true before really updating image */
1711     ret = vmdk_pwritev(bs, offset, bytes, NULL, true, true);
1712     if (!ret) {
1713         ret = vmdk_pwritev(bs, offset, bytes, NULL, true, false);
1714     }
1715     qemu_co_mutex_unlock(&s->lock);
1716     return ret;
1717 }
1718
1719 static int vmdk_create_extent(const char *filename, int64_t filesize,
1720                               bool flat, bool compress, bool zeroed_grain,
1721                               QemuOpts *opts, Error **errp)
1722 {
1723     int ret, i;
1724     BlockBackend *blk = NULL;
1725     VMDK4Header header;
1726     Error *local_err = NULL;
1727     uint32_t tmp, magic, grains, gd_sectors, gt_size, gt_count;
1728     uint32_t *gd_buf = NULL;
1729     int gd_buf_size;
1730
1731     ret = bdrv_create_file(filename, opts, &local_err);
1732     if (ret < 0) {
1733         error_propagate(errp, local_err);
1734         goto exit;
1735     }
1736
1737     blk = blk_new_open(filename, NULL, NULL,
1738                        BDRV_O_RDWR | BDRV_O_PROTOCOL, &local_err);
1739     if (blk == NULL) {
1740         error_propagate(errp, local_err);
1741         ret = -EIO;
1742         goto exit;
1743     }
1744
1745     blk_set_allow_write_beyond_eof(blk, true);
1746
1747     if (flat) {
1748         ret = blk_truncate(blk, filesize);
1749         if (ret < 0) {
1750             error_setg_errno(errp, -ret, "Could not truncate file");
1751         }
1752         goto exit;
1753     }
1754     magic = cpu_to_be32(VMDK4_MAGIC);
1755     memset(&header, 0, sizeof(header));
1756     if (compress) {
1757         header.version = 3;
1758     } else if (zeroed_grain) {
1759         header.version = 2;
1760     } else {
1761         header.version = 1;
1762     }
1763     header.flags = VMDK4_FLAG_RGD | VMDK4_FLAG_NL_DETECT
1764                    | (compress ? VMDK4_FLAG_COMPRESS | VMDK4_FLAG_MARKER : 0)
1765                    | (zeroed_grain ? VMDK4_FLAG_ZERO_GRAIN : 0);
1766     header.compressAlgorithm = compress ? VMDK4_COMPRESSION_DEFLATE : 0;
1767     header.capacity = filesize / BDRV_SECTOR_SIZE;
1768     header.granularity = 128;
1769     header.num_gtes_per_gt = BDRV_SECTOR_SIZE;
1770
1771     grains = DIV_ROUND_UP(filesize / BDRV_SECTOR_SIZE, header.granularity);
1772     gt_size = DIV_ROUND_UP(header.num_gtes_per_gt * sizeof(uint32_t),
1773                            BDRV_SECTOR_SIZE);
1774     gt_count = DIV_ROUND_UP(grains, header.num_gtes_per_gt);
1775     gd_sectors = DIV_ROUND_UP(gt_count * sizeof(uint32_t), BDRV_SECTOR_SIZE);
1776
1777     header.desc_offset = 1;
1778     header.desc_size = 20;
1779     header.rgd_offset = header.desc_offset + header.desc_size;
1780     header.gd_offset = header.rgd_offset + gd_sectors + (gt_size * gt_count);
1781     header.grain_offset =
1782         ROUND_UP(header.gd_offset + gd_sectors + (gt_size * gt_count),
1783                  header.granularity);
1784     /* swap endianness for all header fields */
1785     header.version = cpu_to_le32(header.version);
1786     header.flags = cpu_to_le32(header.flags);
1787     header.capacity = cpu_to_le64(header.capacity);
1788     header.granularity = cpu_to_le64(header.granularity);
1789     header.num_gtes_per_gt = cpu_to_le32(header.num_gtes_per_gt);
1790     header.desc_offset = cpu_to_le64(header.desc_offset);
1791     header.desc_size = cpu_to_le64(header.desc_size);
1792     header.rgd_offset = cpu_to_le64(header.rgd_offset);
1793     header.gd_offset = cpu_to_le64(header.gd_offset);
1794     header.grain_offset = cpu_to_le64(header.grain_offset);
1795     header.compressAlgorithm = cpu_to_le16(header.compressAlgorithm);
1796
1797     header.check_bytes[0] = 0xa;
1798     header.check_bytes[1] = 0x20;
1799     header.check_bytes[2] = 0xd;
1800     header.check_bytes[3] = 0xa;
1801
1802     /* write all the data */
1803     ret = blk_pwrite(blk, 0, &magic, sizeof(magic), 0);
1804     if (ret < 0) {
1805         error_setg(errp, QERR_IO_ERROR);
1806         goto exit;
1807     }
1808     ret = blk_pwrite(blk, sizeof(magic), &header, sizeof(header), 0);
1809     if (ret < 0) {
1810         error_setg(errp, QERR_IO_ERROR);
1811         goto exit;
1812     }
1813
1814     ret = blk_truncate(blk, le64_to_cpu(header.grain_offset) << 9);
1815     if (ret < 0) {
1816         error_setg_errno(errp, -ret, "Could not truncate file");
1817         goto exit;
1818     }
1819
1820     /* write grain directory */
1821     gd_buf_size = gd_sectors * BDRV_SECTOR_SIZE;
1822     gd_buf = g_malloc0(gd_buf_size);
1823     for (i = 0, tmp = le64_to_cpu(header.rgd_offset) + gd_sectors;
1824          i < gt_count; i++, tmp += gt_size) {
1825         gd_buf[i] = cpu_to_le32(tmp);
1826     }
1827     ret = blk_pwrite(blk, le64_to_cpu(header.rgd_offset) * BDRV_SECTOR_SIZE,
1828                      gd_buf, gd_buf_size, 0);
1829     if (ret < 0) {
1830         error_setg(errp, QERR_IO_ERROR);
1831         goto exit;
1832     }
1833
1834     /* write backup grain directory */
1835     for (i = 0, tmp = le64_to_cpu(header.gd_offset) + gd_sectors;
1836          i < gt_count; i++, tmp += gt_size) {
1837         gd_buf[i] = cpu_to_le32(tmp);
1838     }
1839     ret = blk_pwrite(blk, le64_to_cpu(header.gd_offset) * BDRV_SECTOR_SIZE,
1840                      gd_buf, gd_buf_size, 0);
1841     if (ret < 0) {
1842         error_setg(errp, QERR_IO_ERROR);
1843         goto exit;
1844     }
1845
1846     ret = 0;
1847 exit:
1848     if (blk) {
1849         blk_unref(blk);
1850     }
1851     g_free(gd_buf);
1852     return ret;
1853 }
1854
1855 static int filename_decompose(const char *filename, char *path, char *prefix,
1856                               char *postfix, size_t buf_len, Error **errp)
1857 {
1858     const char *p, *q;
1859
1860     if (filename == NULL || !strlen(filename)) {
1861         error_setg(errp, "No filename provided");
1862         return VMDK_ERROR;
1863     }
1864     p = strrchr(filename, '/');
1865     if (p == NULL) {
1866         p = strrchr(filename, '\\');
1867     }
1868     if (p == NULL) {
1869         p = strrchr(filename, ':');
1870     }
1871     if (p != NULL) {
1872         p++;
1873         if (p - filename >= buf_len) {
1874             return VMDK_ERROR;
1875         }
1876         pstrcpy(path, p - filename + 1, filename);
1877     } else {
1878         p = filename;
1879         path[0] = '\0';
1880     }
1881     q = strrchr(p, '.');
1882     if (q == NULL) {
1883         pstrcpy(prefix, buf_len, p);
1884         postfix[0] = '\0';
1885     } else {
1886         if (q - p >= buf_len) {
1887             return VMDK_ERROR;
1888         }
1889         pstrcpy(prefix, q - p + 1, p);
1890         pstrcpy(postfix, buf_len, q);
1891     }
1892     return VMDK_OK;
1893 }
1894
1895 static int vmdk_create(const char *filename, QemuOpts *opts, Error **errp)
1896 {
1897     int idx = 0;
1898     BlockBackend *new_blk = NULL;
1899     Error *local_err = NULL;
1900     char *desc = NULL;
1901     int64_t total_size = 0, filesize;
1902     char *adapter_type = NULL;
1903     char *backing_file = NULL;
1904     char *hw_version = NULL;
1905     char *fmt = NULL;
1906     int ret = 0;
1907     bool flat, split, compress;
1908     GString *ext_desc_lines;
1909     char *path = g_malloc0(PATH_MAX);
1910     char *prefix = g_malloc0(PATH_MAX);
1911     char *postfix = g_malloc0(PATH_MAX);
1912     char *desc_line = g_malloc0(BUF_SIZE);
1913     char *ext_filename = g_malloc0(PATH_MAX);
1914     char *desc_filename = g_malloc0(PATH_MAX);
1915     const int64_t split_size = 0x80000000;  /* VMDK has constant split size */
1916     const char *desc_extent_line;
1917     char *parent_desc_line = g_malloc0(BUF_SIZE);
1918     uint32_t parent_cid = 0xffffffff;
1919     uint32_t number_heads = 16;
1920     bool zeroed_grain = false;
1921     uint32_t desc_offset = 0, desc_len;
1922     const char desc_template[] =
1923         "# Disk DescriptorFile\n"
1924         "version=1\n"
1925         "CID=%" PRIx32 "\n"
1926         "parentCID=%" PRIx32 "\n"
1927         "createType=\"%s\"\n"
1928         "%s"
1929         "\n"
1930         "# Extent description\n"
1931         "%s"
1932         "\n"
1933         "# The Disk Data Base\n"
1934         "#DDB\n"
1935         "\n"
1936         "ddb.virtualHWVersion = \"%s\"\n"
1937         "ddb.geometry.cylinders = \"%" PRId64 "\"\n"
1938         "ddb.geometry.heads = \"%" PRIu32 "\"\n"
1939         "ddb.geometry.sectors = \"63\"\n"
1940         "ddb.adapterType = \"%s\"\n";
1941
1942     ext_desc_lines = g_string_new(NULL);
1943
1944     if (filename_decompose(filename, path, prefix, postfix, PATH_MAX, errp)) {
1945         ret = -EINVAL;
1946         goto exit;
1947     }
1948     /* Read out options */
1949     total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
1950                           BDRV_SECTOR_SIZE);
1951     adapter_type = qemu_opt_get_del(opts, BLOCK_OPT_ADAPTER_TYPE);
1952     backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
1953     hw_version = qemu_opt_get_del(opts, BLOCK_OPT_HWVERSION);
1954     if (qemu_opt_get_bool_del(opts, BLOCK_OPT_COMPAT6, false)) {
1955         if (strcmp(hw_version, "undefined")) {
1956             error_setg(errp,
1957                        "compat6 cannot be enabled with hwversion set");
1958             ret = -EINVAL;
1959             goto exit;
1960         }
1961         g_free(hw_version);
1962         hw_version = g_strdup("6");
1963     }
1964     if (strcmp(hw_version, "undefined") == 0) {
1965         g_free(hw_version);
1966         hw_version = g_strdup("4");
1967     }
1968     fmt = qemu_opt_get_del(opts, BLOCK_OPT_SUBFMT);
1969     if (qemu_opt_get_bool_del(opts, BLOCK_OPT_ZEROED_GRAIN, false)) {
1970         zeroed_grain = true;
1971     }
1972
1973     if (!adapter_type) {
1974         adapter_type = g_strdup("ide");
1975     } else if (strcmp(adapter_type, "ide") &&
1976                strcmp(adapter_type, "buslogic") &&
1977                strcmp(adapter_type, "lsilogic") &&
1978                strcmp(adapter_type, "legacyESX")) {
1979         error_setg(errp, "Unknown adapter type: '%s'", adapter_type);
1980         ret = -EINVAL;
1981         goto exit;
1982     }
1983     if (strcmp(adapter_type, "ide") != 0) {
1984         /* that's the number of heads with which vmware operates when
1985            creating, exporting, etc. vmdk files with a non-ide adapter type */
1986         number_heads = 255;
1987     }
1988     if (!fmt) {
1989         /* Default format to monolithicSparse */
1990         fmt = g_strdup("monolithicSparse");
1991     } else if (strcmp(fmt, "monolithicFlat") &&
1992                strcmp(fmt, "monolithicSparse") &&
1993                strcmp(fmt, "twoGbMaxExtentSparse") &&
1994                strcmp(fmt, "twoGbMaxExtentFlat") &&
1995                strcmp(fmt, "streamOptimized")) {
1996         error_setg(errp, "Unknown subformat: '%s'", fmt);
1997         ret = -EINVAL;
1998         goto exit;
1999     }
2000     split = !(strcmp(fmt, "twoGbMaxExtentFlat") &&
2001               strcmp(fmt, "twoGbMaxExtentSparse"));
2002     flat = !(strcmp(fmt, "monolithicFlat") &&
2003              strcmp(fmt, "twoGbMaxExtentFlat"));
2004     compress = !strcmp(fmt, "streamOptimized");
2005     if (flat) {
2006         desc_extent_line = "RW %" PRId64 " FLAT \"%s\" 0\n";
2007     } else {
2008         desc_extent_line = "RW %" PRId64 " SPARSE \"%s\"\n";
2009     }
2010     if (flat && backing_file) {
2011         error_setg(errp, "Flat image can't have backing file");
2012         ret = -ENOTSUP;
2013         goto exit;
2014     }
2015     if (flat && zeroed_grain) {
2016         error_setg(errp, "Flat image can't enable zeroed grain");
2017         ret = -ENOTSUP;
2018         goto exit;
2019     }
2020     if (backing_file) {
2021         BlockBackend *blk;
2022         char *full_backing = g_new0(char, PATH_MAX);
2023         bdrv_get_full_backing_filename_from_filename(filename, backing_file,
2024                                                      full_backing, PATH_MAX,
2025                                                      &local_err);
2026         if (local_err) {
2027             g_free(full_backing);
2028             error_propagate(errp, local_err);
2029             ret = -ENOENT;
2030             goto exit;
2031         }
2032
2033         blk = blk_new_open(full_backing, NULL, NULL,
2034                            BDRV_O_NO_BACKING, errp);
2035         g_free(full_backing);
2036         if (blk == NULL) {
2037             ret = -EIO;
2038             goto exit;
2039         }
2040         if (strcmp(blk_bs(blk)->drv->format_name, "vmdk")) {
2041             blk_unref(blk);
2042             ret = -EINVAL;
2043             goto exit;
2044         }
2045         parent_cid = vmdk_read_cid(blk_bs(blk), 0);
2046         blk_unref(blk);
2047         snprintf(parent_desc_line, BUF_SIZE,
2048                 "parentFileNameHint=\"%s\"", backing_file);
2049     }
2050
2051     /* Create extents */
2052     filesize = total_size;
2053     while (filesize > 0) {
2054         int64_t size = filesize;
2055
2056         if (split && size > split_size) {
2057             size = split_size;
2058         }
2059         if (split) {
2060             snprintf(desc_filename, PATH_MAX, "%s-%c%03d%s",
2061                     prefix, flat ? 'f' : 's', ++idx, postfix);
2062         } else if (flat) {
2063             snprintf(desc_filename, PATH_MAX, "%s-flat%s", prefix, postfix);
2064         } else {
2065             snprintf(desc_filename, PATH_MAX, "%s%s", prefix, postfix);
2066         }
2067         snprintf(ext_filename, PATH_MAX, "%s%s", path, desc_filename);
2068
2069         if (vmdk_create_extent(ext_filename, size,
2070                                flat, compress, zeroed_grain, opts, errp)) {
2071             ret = -EINVAL;
2072             goto exit;
2073         }
2074         filesize -= size;
2075
2076         /* Format description line */
2077         snprintf(desc_line, BUF_SIZE,
2078                     desc_extent_line, size / BDRV_SECTOR_SIZE, desc_filename);
2079         g_string_append(ext_desc_lines, desc_line);
2080     }
2081     /* generate descriptor file */
2082     desc = g_strdup_printf(desc_template,
2083                            g_random_int(),
2084                            parent_cid,
2085                            fmt,
2086                            parent_desc_line,
2087                            ext_desc_lines->str,
2088                            hw_version,
2089                            total_size /
2090                                (int64_t)(63 * number_heads * BDRV_SECTOR_SIZE),
2091                            number_heads,
2092                            adapter_type);
2093     desc_len = strlen(desc);
2094     /* the descriptor offset = 0x200 */
2095     if (!split && !flat) {
2096         desc_offset = 0x200;
2097     } else {
2098         ret = bdrv_create_file(filename, opts, &local_err);
2099         if (ret < 0) {
2100             error_propagate(errp, local_err);
2101             goto exit;
2102         }
2103     }
2104
2105     new_blk = blk_new_open(filename, NULL, NULL,
2106                            BDRV_O_RDWR | BDRV_O_PROTOCOL, &local_err);
2107     if (new_blk == NULL) {
2108         error_propagate(errp, local_err);
2109         ret = -EIO;
2110         goto exit;
2111     }
2112
2113     blk_set_allow_write_beyond_eof(new_blk, true);
2114
2115     ret = blk_pwrite(new_blk, desc_offset, desc, desc_len, 0);
2116     if (ret < 0) {
2117         error_setg_errno(errp, -ret, "Could not write description");
2118         goto exit;
2119     }
2120     /* bdrv_pwrite write padding zeros to align to sector, we don't need that
2121      * for description file */
2122     if (desc_offset == 0) {
2123         ret = blk_truncate(new_blk, desc_len);
2124         if (ret < 0) {
2125             error_setg_errno(errp, -ret, "Could not truncate file");
2126         }
2127     }
2128 exit:
2129     if (new_blk) {
2130         blk_unref(new_blk);
2131     }
2132     g_free(adapter_type);
2133     g_free(backing_file);
2134     g_free(hw_version);
2135     g_free(fmt);
2136     g_free(desc);
2137     g_free(path);
2138     g_free(prefix);
2139     g_free(postfix);
2140     g_free(desc_line);
2141     g_free(ext_filename);
2142     g_free(desc_filename);
2143     g_free(parent_desc_line);
2144     g_string_free(ext_desc_lines, true);
2145     return ret;
2146 }
2147
2148 static void vmdk_close(BlockDriverState *bs)
2149 {
2150     BDRVVmdkState *s = bs->opaque;
2151
2152     vmdk_free_extents(bs);
2153     g_free(s->create_type);
2154
2155     migrate_del_blocker(s->migration_blocker);
2156     error_free(s->migration_blocker);
2157 }
2158
2159 static coroutine_fn int vmdk_co_flush(BlockDriverState *bs)
2160 {
2161     BDRVVmdkState *s = bs->opaque;
2162     int i, err;
2163     int ret = 0;
2164
2165     for (i = 0; i < s->num_extents; i++) {
2166         err = bdrv_co_flush(s->extents[i].file->bs);
2167         if (err < 0) {
2168             ret = err;
2169         }
2170     }
2171     return ret;
2172 }
2173
2174 static int64_t vmdk_get_allocated_file_size(BlockDriverState *bs)
2175 {
2176     int i;
2177     int64_t ret = 0;
2178     int64_t r;
2179     BDRVVmdkState *s = bs->opaque;
2180
2181     ret = bdrv_get_allocated_file_size(bs->file->bs);
2182     if (ret < 0) {
2183         return ret;
2184     }
2185     for (i = 0; i < s->num_extents; i++) {
2186         if (s->extents[i].file == bs->file) {
2187             continue;
2188         }
2189         r = bdrv_get_allocated_file_size(s->extents[i].file->bs);
2190         if (r < 0) {
2191             return r;
2192         }
2193         ret += r;
2194     }
2195     return ret;
2196 }
2197
2198 static int vmdk_has_zero_init(BlockDriverState *bs)
2199 {
2200     int i;
2201     BDRVVmdkState *s = bs->opaque;
2202
2203     /* If has a flat extent and its underlying storage doesn't have zero init,
2204      * return 0. */
2205     for (i = 0; i < s->num_extents; i++) {
2206         if (s->extents[i].flat) {
2207             if (!bdrv_has_zero_init(s->extents[i].file->bs)) {
2208                 return 0;
2209             }
2210         }
2211     }
2212     return 1;
2213 }
2214
2215 static ImageInfo *vmdk_get_extent_info(VmdkExtent *extent)
2216 {
2217     ImageInfo *info = g_new0(ImageInfo, 1);
2218
2219     *info = (ImageInfo){
2220         .filename         = g_strdup(extent->file->bs->filename),
2221         .format           = g_strdup(extent->type),
2222         .virtual_size     = extent->sectors * BDRV_SECTOR_SIZE,
2223         .compressed       = extent->compressed,
2224         .has_compressed   = extent->compressed,
2225         .cluster_size     = extent->cluster_sectors * BDRV_SECTOR_SIZE,
2226         .has_cluster_size = !extent->flat,
2227     };
2228
2229     return info;
2230 }
2231
2232 static int vmdk_check(BlockDriverState *bs, BdrvCheckResult *result,
2233                       BdrvCheckMode fix)
2234 {
2235     BDRVVmdkState *s = bs->opaque;
2236     VmdkExtent *extent = NULL;
2237     int64_t sector_num = 0;
2238     int64_t total_sectors = bdrv_nb_sectors(bs);
2239     int ret;
2240     uint64_t cluster_offset;
2241
2242     if (fix) {
2243         return -ENOTSUP;
2244     }
2245
2246     for (;;) {
2247         if (sector_num >= total_sectors) {
2248             return 0;
2249         }
2250         extent = find_extent(s, sector_num, extent);
2251         if (!extent) {
2252             fprintf(stderr,
2253                     "ERROR: could not find extent for sector %" PRId64 "\n",
2254                     sector_num);
2255             break;
2256         }
2257         ret = get_cluster_offset(bs, extent, NULL,
2258                                  sector_num << BDRV_SECTOR_BITS,
2259                                  false, &cluster_offset, 0, 0);
2260         if (ret == VMDK_ERROR) {
2261             fprintf(stderr,
2262                     "ERROR: could not get cluster_offset for sector %"
2263                     PRId64 "\n", sector_num);
2264             break;
2265         }
2266         if (ret == VMDK_OK &&
2267             cluster_offset >= bdrv_getlength(extent->file->bs))
2268         {
2269             fprintf(stderr,
2270                     "ERROR: cluster offset for sector %"
2271                     PRId64 " points after EOF\n", sector_num);
2272             break;
2273         }
2274         sector_num += extent->cluster_sectors;
2275     }
2276
2277     result->corruptions++;
2278     return 0;
2279 }
2280
2281 static ImageInfoSpecific *vmdk_get_specific_info(BlockDriverState *bs)
2282 {
2283     int i;
2284     BDRVVmdkState *s = bs->opaque;
2285     ImageInfoSpecific *spec_info = g_new0(ImageInfoSpecific, 1);
2286     ImageInfoList **next;
2287
2288     *spec_info = (ImageInfoSpecific){
2289         .type = IMAGE_INFO_SPECIFIC_KIND_VMDK,
2290         .u = {
2291             .vmdk.data = g_new0(ImageInfoSpecificVmdk, 1),
2292         },
2293     };
2294
2295     *spec_info->u.vmdk.data = (ImageInfoSpecificVmdk) {
2296         .create_type = g_strdup(s->create_type),
2297         .cid = s->cid,
2298         .parent_cid = s->parent_cid,
2299     };
2300
2301     next = &spec_info->u.vmdk.data->extents;
2302     for (i = 0; i < s->num_extents; i++) {
2303         *next = g_new0(ImageInfoList, 1);
2304         (*next)->value = vmdk_get_extent_info(&s->extents[i]);
2305         (*next)->next = NULL;
2306         next = &(*next)->next;
2307     }
2308
2309     return spec_info;
2310 }
2311
2312 static bool vmdk_extents_type_eq(const VmdkExtent *a, const VmdkExtent *b)
2313 {
2314     return a->flat == b->flat &&
2315            a->compressed == b->compressed &&
2316            (a->flat || a->cluster_sectors == b->cluster_sectors);
2317 }
2318
2319 static int vmdk_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2320 {
2321     int i;
2322     BDRVVmdkState *s = bs->opaque;
2323     assert(s->num_extents);
2324
2325     /* See if we have multiple extents but they have different cases */
2326     for (i = 1; i < s->num_extents; i++) {
2327         if (!vmdk_extents_type_eq(&s->extents[0], &s->extents[i])) {
2328             return -ENOTSUP;
2329         }
2330     }
2331     bdi->needs_compressed_writes = s->extents[0].compressed;
2332     if (!s->extents[0].flat) {
2333         bdi->cluster_size = s->extents[0].cluster_sectors << BDRV_SECTOR_BITS;
2334     }
2335     return 0;
2336 }
2337
2338 static QemuOptsList vmdk_create_opts = {
2339     .name = "vmdk-create-opts",
2340     .head = QTAILQ_HEAD_INITIALIZER(vmdk_create_opts.head),
2341     .desc = {
2342         {
2343             .name = BLOCK_OPT_SIZE,
2344             .type = QEMU_OPT_SIZE,
2345             .help = "Virtual disk size"
2346         },
2347         {
2348             .name = BLOCK_OPT_ADAPTER_TYPE,
2349             .type = QEMU_OPT_STRING,
2350             .help = "Virtual adapter type, can be one of "
2351                     "ide (default), lsilogic, buslogic or legacyESX"
2352         },
2353         {
2354             .name = BLOCK_OPT_BACKING_FILE,
2355             .type = QEMU_OPT_STRING,
2356             .help = "File name of a base image"
2357         },
2358         {
2359             .name = BLOCK_OPT_COMPAT6,
2360             .type = QEMU_OPT_BOOL,
2361             .help = "VMDK version 6 image",
2362             .def_value_str = "off"
2363         },
2364         {
2365             .name = BLOCK_OPT_HWVERSION,
2366             .type = QEMU_OPT_STRING,
2367             .help = "VMDK hardware version",
2368             .def_value_str = "undefined"
2369         },
2370         {
2371             .name = BLOCK_OPT_SUBFMT,
2372             .type = QEMU_OPT_STRING,
2373             .help =
2374                 "VMDK flat extent format, can be one of "
2375                 "{monolithicSparse (default) | monolithicFlat | twoGbMaxExtentSparse | twoGbMaxExtentFlat | streamOptimized} "
2376         },
2377         {
2378             .name = BLOCK_OPT_ZEROED_GRAIN,
2379             .type = QEMU_OPT_BOOL,
2380             .help = "Enable efficient zero writes "
2381                     "using the zeroed-grain GTE feature"
2382         },
2383         { /* end of list */ }
2384     }
2385 };
2386
2387 static BlockDriver bdrv_vmdk = {
2388     .format_name                  = "vmdk",
2389     .instance_size                = sizeof(BDRVVmdkState),
2390     .bdrv_probe                   = vmdk_probe,
2391     .bdrv_open                    = vmdk_open,
2392     .bdrv_check                   = vmdk_check,
2393     .bdrv_reopen_prepare          = vmdk_reopen_prepare,
2394     .bdrv_co_preadv               = vmdk_co_preadv,
2395     .bdrv_co_pwritev              = vmdk_co_pwritev,
2396     .bdrv_write_compressed        = vmdk_write_compressed,
2397     .bdrv_co_pwrite_zeroes        = vmdk_co_pwrite_zeroes,
2398     .bdrv_close                   = vmdk_close,
2399     .bdrv_create                  = vmdk_create,
2400     .bdrv_co_flush_to_disk        = vmdk_co_flush,
2401     .bdrv_co_get_block_status     = vmdk_co_get_block_status,
2402     .bdrv_get_allocated_file_size = vmdk_get_allocated_file_size,
2403     .bdrv_has_zero_init           = vmdk_has_zero_init,
2404     .bdrv_get_specific_info       = vmdk_get_specific_info,
2405     .bdrv_refresh_limits          = vmdk_refresh_limits,
2406     .bdrv_get_info                = vmdk_get_info,
2407
2408     .supports_backing             = true,
2409     .create_opts                  = &vmdk_create_opts,
2410 };
2411
2412 static void bdrv_vmdk_init(void)
2413 {
2414     bdrv_register(&bdrv_vmdk);
2415 }
2416
2417 block_init(bdrv_vmdk_init);