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/blocker.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 /* Return -ve errno, or 0 on success and write CID into *pcid. */
 246 static int vmdk_read_cid(BlockDriverState *bs, int parent, uint32_t *pcid)
 247 {
 248     char *desc;
 249     uint32_t cid;
 250     const char *p_name, *cid_str;
 251     size_t cid_str_size;
 252     BDRVVmdkState *s = bs->opaque;
 253     int ret;
 254
 255     desc = g_malloc0(DESC_SIZE);
 256     ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE);
 257     if (ret < 0) {
 258         goto out;
 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         ret = -EINVAL;
 273         goto out;
 274     }
 275     p_name += cid_str_size;
 276     if (sscanf(p_name, "%" SCNx32, &cid) != 1) {
 277         ret = -EINVAL;
 278         goto out;
 279     }
 280     *pcid = cid;
 281     ret = 0;
 282
 283 out:
 284     g_free(desc);
 285     return ret;
 286 }
 287
 288 static int vmdk_write_cid(BlockDriverState *bs, uint32_t cid)
 289 {
 290     char *desc, *tmp_desc;
 291     char *p_name, *tmp_str;
 292     BDRVVmdkState *s = bs->opaque;
 293     int ret = 0;
 294
 295     desc = g_malloc0(DESC_SIZE);
 296     tmp_desc = g_malloc0(DESC_SIZE);
 297     ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE);
 298     if (ret < 0) {
 299         goto out;
 300     }
 301
 302     desc[DESC_SIZE - 1] = '\0';
 303     tmp_str = strstr(desc, "parentCID");
 304     if (tmp_str == NULL) {
 305         ret = -EINVAL;
 306         goto out;
 307     }
 308
 309     pstrcpy(tmp_desc, DESC_SIZE, tmp_str);
 310     p_name = strstr(desc, "CID");
 311     if (p_name != NULL) {
 312         p_name += sizeof("CID");
 313         snprintf(p_name, DESC_SIZE - (p_name - desc), "%" PRIx32 "\n", cid);
 314         pstrcat(desc, DESC_SIZE, tmp_desc);
 315     }
 316
 317     ret = bdrv_pwrite_sync(bs->file, s->desc_offset, desc, DESC_SIZE);
 318
 319 out:
 320     g_free(desc);
 321     g_free(tmp_desc);
 322     return ret;
 323 }
 324
 325 static int vmdk_is_cid_valid(BlockDriverState *bs)
 326 {
 327     BDRVVmdkState *s = bs->opaque;
 328     uint32_t cur_pcid;
 329
 330     if (!s->cid_checked && bs->backing) {
 331         BlockDriverState *p_bs = bs->backing->bs;
 332
 333         if (vmdk_read_cid(p_bs, 0, &cur_pcid) != 0) {
 334             /* read failure: report as not valid */
 335             return 0;
 336         }
 337         if (s->parent_cid != cur_pcid) {
 338             /* CID not valid */
 339             return 0;
 340         }
 341     }
 342     s->cid_checked = true;
 343     /* CID valid */
 344     return 1;
 345 }
 346
 347 /* We have nothing to do for VMDK reopen, stubs just return success */
 348 static int vmdk_reopen_prepare(BDRVReopenState *state,
 349                                BlockReopenQueue *queue, Error **errp)
 350 {
 351     assert(state != NULL);
 352     assert(state->bs != NULL);
 353     return 0;
 354 }
 355
 356 static int vmdk_parent_open(BlockDriverState *bs)
 357 {
 358     char *p_name;
 359     char *desc;
 360     BDRVVmdkState *s = bs->opaque;
 361     int ret;
 362
 363     desc = g_malloc0(DESC_SIZE + 1);
 364     ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE);
 365     if (ret < 0) {
 366         goto out;
 367     }
 368     ret = 0;
 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             ret = -EINVAL;
 378             goto out;
 379         }
 380         if ((end_name - p_name) > sizeof(bs->backing_file) - 1) {
 381             ret = -EINVAL;
 382             goto out;
 383         }
 384
 385         pstrcpy(bs->backing_file, end_name - p_name + 1, p_name);
 386     }
 387
 388 out:
 389     g_free(desc);
 390     return ret;
 391 }
 392
 393 /* Create and append extent to the extent array. Return the added VmdkExtent
 394  * address. return NULL if allocation failed. */
 395 static int vmdk_add_extent(BlockDriverState *bs,
 396                            BdrvChild *file, bool flat, int64_t sectors,
 397                            int64_t l1_offset, int64_t l1_backup_offset,
 398                            uint32_t l1_size,
 399                            int l2_size, uint64_t cluster_sectors,
 400                            VmdkExtent **new_extent,
 401                            Error **errp)
 402 {
 403     VmdkExtent *extent;
 404     BDRVVmdkState *s = bs->opaque;
 405     int64_t nb_sectors;
 406
 407     if (cluster_sectors > 0x200000) {
 408         /* 0x200000 * 512Bytes = 1GB for one cluster is unrealistic */
 409         error_setg(errp, "Invalid granularity, image may be corrupt");
 410         return -EFBIG;
 411     }
 412     if (l1_size > 512 * 1024 * 1024) {
 413         /* Although with big capacity and small l1_entry_sectors, we can get a
 414          * big l1_size, we don't want unbounded value to allocate the table.
 415          * Limit it to 512M, which is 16PB for default cluster and L2 table
 416          * size */
 417         error_setg(errp, "L1 size too big");
 418         return -EFBIG;
 419     }
 420
 421     nb_sectors = bdrv_nb_sectors(file->bs);
 422     if (nb_sectors < 0) {
 423         return nb_sectors;
 424     }
 425
 426     s->extents = g_renew(VmdkExtent, s->extents, s->num_extents + 1);
 427     extent = &s->extents[s->num_extents];
 428     s->num_extents++;
 429
 430     memset(extent, 0, sizeof(VmdkExtent));
 431     extent->file = file;
 432     extent->flat = flat;
 433     extent->sectors = sectors;
 434     extent->l1_table_offset = l1_offset;
 435     extent->l1_backup_table_offset = l1_backup_offset;
 436     extent->l1_size = l1_size;
 437     extent->l1_entry_sectors = l2_size * cluster_sectors;
 438     extent->l2_size = l2_size;
 439     extent->cluster_sectors = flat ? sectors : cluster_sectors;
 440     extent->next_cluster_sector = ROUND_UP(nb_sectors, cluster_sectors);
 441
 442     if (s->num_extents > 1) {
 443         extent->end_sector = (*(extent - 1)).end_sector + extent->sectors;
 444     } else {
 445         extent->end_sector = extent->sectors;
 446     }
 447     bs->total_sectors = extent->end_sector;
 448     if (new_extent) {
 449         *new_extent = extent;
 450     }
 451     return 0;
 452 }
 453
 454 static int vmdk_init_tables(BlockDriverState *bs, VmdkExtent *extent,
 455                             Error **errp)
 456 {
 457     int ret;
 458     size_t l1_size;
 459     int i;
 460
 461     /* read the L1 table */
 462     l1_size = extent->l1_size * sizeof(uint32_t);
 463     extent->l1_table = g_try_malloc(l1_size);
 464     if (l1_size && extent->l1_table == NULL) {
 465         return -ENOMEM;
 466     }
 467
 468     ret = bdrv_pread(extent->file,
 469                      extent->l1_table_offset,
 470                      extent->l1_table,
 471                      l1_size);
 472     if (ret < 0) {
 473         error_setg_errno(errp, -ret,
 474                          "Could not read l1 table from extent '%s'",
 475                          extent->file->bs->filename);
 476         goto fail_l1;
 477     }
 478     for (i = 0; i < extent->l1_size; i++) {
 479         le32_to_cpus(&extent->l1_table[i]);
 480     }
 481
 482     if (extent->l1_backup_table_offset) {
 483         extent->l1_backup_table = g_try_malloc(l1_size);
 484         if (l1_size && extent->l1_backup_table == NULL) {
 485             ret = -ENOMEM;
 486             goto fail_l1;
 487         }
 488         ret = bdrv_pread(extent->file,
 489                          extent->l1_backup_table_offset,
 490                          extent->l1_backup_table,
 491                          l1_size);
 492         if (ret < 0) {
 493             error_setg_errno(errp, -ret,
 494                              "Could not read l1 backup table from extent '%s'",
 495                              extent->file->bs->filename);
 496             goto fail_l1b;
 497         }
 498         for (i = 0; i < extent->l1_size; i++) {
 499             le32_to_cpus(&extent->l1_backup_table[i]);
 500         }
 501     }
 502
 503     extent->l2_cache =
 504         g_new(uint32_t, extent->l2_size * L2_CACHE_SIZE);
 505     return 0;
 506  fail_l1b:
 507     g_free(extent->l1_backup_table);
 508  fail_l1:
 509     g_free(extent->l1_table);
 510     return ret;
 511 }
 512
 513 static int vmdk_open_vmfs_sparse(BlockDriverState *bs,
 514                                  BdrvChild *file,
 515                                  int flags, Error **errp)
 516 {
 517     int ret;
 518     uint32_t magic;
 519     VMDK3Header header;
 520     VmdkExtent *extent;
 521
 522     ret = bdrv_pread(file, sizeof(magic), &header, sizeof(header));
 523     if (ret < 0) {
 524         error_setg_errno(errp, -ret,
 525                          "Could not read header from file '%s'",
 526                          file->bs->filename);
 527         return ret;
 528     }
 529     ret = vmdk_add_extent(bs, file, false,
 530                           le32_to_cpu(header.disk_sectors),
 531                           (int64_t)le32_to_cpu(header.l1dir_offset) << 9,
 532                           0,
 533                           le32_to_cpu(header.l1dir_size),
 534                           4096,
 535                           le32_to_cpu(header.granularity),
 536                           &extent,
 537                           errp);
 538     if (ret < 0) {
 539         return ret;
 540     }
 541     ret = vmdk_init_tables(bs, extent, errp);
 542     if (ret) {
 543         /* free extent allocated by vmdk_add_extent */
 544         vmdk_free_last_extent(bs);
 545     }
 546     return ret;
 547 }
 548
 549 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf,
 550                                QDict *options, Error **errp);
 551
 552 static char *vmdk_read_desc(BdrvChild *file, uint64_t desc_offset, Error **errp)
 553 {
 554     int64_t size;
 555     char *buf;
 556     int ret;
 557
 558     size = bdrv_getlength(file->bs);
 559     if (size < 0) {
 560         error_setg_errno(errp, -size, "Could not access file");
 561         return NULL;
 562     }
 563
 564     if (size < 4) {
 565         /* Both descriptor file and sparse image must be much larger than 4
 566          * bytes, also callers of vmdk_read_desc want to compare the first 4
 567          * bytes with VMDK4_MAGIC, let's error out if less is read. */
 568         error_setg(errp, "File is too small, not a valid image");
 569         return NULL;
 570     }
 571
 572     size = MIN(size, (1 << 20) - 1);  /* avoid unbounded allocation */
 573     buf = g_malloc(size + 1);
 574
 575     ret = bdrv_pread(file, desc_offset, buf, size);
 576     if (ret < 0) {
 577         error_setg_errno(errp, -ret, "Could not read from file");
 578         g_free(buf);
 579         return NULL;
 580     }
 581     buf[ret] = 0;
 582
 583     return buf;
 584 }
 585
 586 static int vmdk_open_vmdk4(BlockDriverState *bs,
 587                            BdrvChild *file,
 588                            int flags, QDict *options, Error **errp)
 589 {
 590     int ret;
 591     uint32_t magic;
 592     uint32_t l1_size, l1_entry_sectors;
 593     VMDK4Header header;
 594     VmdkExtent *extent;
 595     BDRVVmdkState *s = bs->opaque;
 596     int64_t l1_backup_offset = 0;
 597     bool compressed;
 598
 599     ret = bdrv_pread(file, sizeof(magic), &header, sizeof(header));
 600     if (ret < 0) {
 601         error_setg_errno(errp, -ret,
 602                          "Could not read header from file '%s'",
 603                          file->bs->filename);
 604         return -EINVAL;
 605     }
 606     if (header.capacity == 0) {
 607         uint64_t desc_offset = le64_to_cpu(header.desc_offset);
 608         if (desc_offset) {
 609             char *buf = vmdk_read_desc(file, desc_offset << 9, errp);
 610             if (!buf) {
 611                 return -EINVAL;
 612             }
 613             ret = vmdk_open_desc_file(bs, flags, buf, options, errp);
 614             g_free(buf);
 615             return ret;
 616         }
 617     }
 618
 619     if (!s->create_type) {
 620         s->create_type = g_strdup("monolithicSparse");
 621     }
 622
 623     if (le64_to_cpu(header.gd_offset) == VMDK4_GD_AT_END) {
 624         /*
 625          * The footer takes precedence over the header, so read it in. The
 626          * footer starts at offset -1024 from the end: One sector for the
 627          * footer, and another one for the end-of-stream marker.
 628          */
 629         struct {
 630             struct {
 631                 uint64_t val;
 632                 uint32_t size;
 633                 uint32_t type;
 634                 uint8_t pad[512 - 16];
 635             } QEMU_PACKED footer_marker;
 636
 637             uint32_t magic;
 638             VMDK4Header header;
 639             uint8_t pad[512 - 4 - sizeof(VMDK4Header)];
 640
 641             struct {
 642                 uint64_t val;
 643                 uint32_t size;
 644                 uint32_t type;
 645                 uint8_t pad[512 - 16];
 646             } QEMU_PACKED eos_marker;
 647         } QEMU_PACKED footer;
 648
 649         ret = bdrv_pread(file,
 650             bs->file->bs->total_sectors * 512 - 1536,
 651             &footer, sizeof(footer));
 652         if (ret < 0) {
 653             error_setg_errno(errp, -ret, "Failed to read footer");
 654             return ret;
 655         }
 656
 657         /* Some sanity checks for the footer */
 658         if (be32_to_cpu(footer.magic) != VMDK4_MAGIC ||
 659             le32_to_cpu(footer.footer_marker.size) != 0  ||
 660             le32_to_cpu(footer.footer_marker.type) != MARKER_FOOTER ||
 661             le64_to_cpu(footer.eos_marker.val) != 0  ||
 662             le32_to_cpu(footer.eos_marker.size) != 0  ||
 663             le32_to_cpu(footer.eos_marker.type) != MARKER_END_OF_STREAM)
 664         {
 665             error_setg(errp, "Invalid footer");
 666             return -EINVAL;
 667         }
 668
 669         header = footer.header;
 670     }
 671
 672     compressed =
 673         le16_to_cpu(header.compressAlgorithm) == VMDK4_COMPRESSION_DEFLATE;
 674     if (le32_to_cpu(header.version) > 3) {
 675         error_setg(errp, "Unsupported VMDK version %" PRIu32,
 676                    le32_to_cpu(header.version));
 677         return -ENOTSUP;
 678     } else if (le32_to_cpu(header.version) == 3 && (flags & BDRV_O_RDWR) &&
 679                !compressed) {
 680         /* VMware KB 2064959 explains that version 3 added support for
 681          * persistent changed block tracking (CBT), and backup software can
 682          * read it as version=1 if it doesn't care about the changed area
 683          * information. So we are safe to enable read only. */
 684         error_setg(errp, "VMDK version 3 must be read only");
 685         return -EINVAL;
 686     }
 687
 688     if (le32_to_cpu(header.num_gtes_per_gt) > 512) {
 689         error_setg(errp, "L2 table size too big");
 690         return -EINVAL;
 691     }
 692
 693     l1_entry_sectors = le32_to_cpu(header.num_gtes_per_gt)
 694                         * le64_to_cpu(header.granularity);
 695     if (l1_entry_sectors == 0) {
 696         error_setg(errp, "L1 entry size is invalid");
 697         return -EINVAL;
 698     }
 699     l1_size = (le64_to_cpu(header.capacity) + l1_entry_sectors - 1)
 700                 / l1_entry_sectors;
 701     if (le32_to_cpu(header.flags) & VMDK4_FLAG_RGD) {
 702         l1_backup_offset = le64_to_cpu(header.rgd_offset) << 9;
 703     }
 704     if (bdrv_nb_sectors(file->bs) < le64_to_cpu(header.grain_offset)) {
 705         error_setg(errp, "File truncated, expecting at least %" PRId64 " bytes",
 706                    (int64_t)(le64_to_cpu(header.grain_offset)
 707                              * BDRV_SECTOR_SIZE));
 708         return -EINVAL;
 709     }
 710
 711     ret = vmdk_add_extent(bs, file, false,
 712                           le64_to_cpu(header.capacity),
 713                           le64_to_cpu(header.gd_offset) << 9,
 714                           l1_backup_offset,
 715                           l1_size,
 716                           le32_to_cpu(header.num_gtes_per_gt),
 717                           le64_to_cpu(header.granularity),
 718                           &extent,
 719                           errp);
 720     if (ret < 0) {
 721         return ret;
 722     }
 723     extent->compressed =
 724         le16_to_cpu(header.compressAlgorithm) == VMDK4_COMPRESSION_DEFLATE;
 725     if (extent->compressed) {
 726         g_free(s->create_type);
 727         s->create_type = g_strdup("streamOptimized");
 728     }
 729     extent->has_marker = le32_to_cpu(header.flags) & VMDK4_FLAG_MARKER;
 730     extent->version = le32_to_cpu(header.version);
 731     extent->has_zero_grain = le32_to_cpu(header.flags) & VMDK4_FLAG_ZERO_GRAIN;
 732     ret = vmdk_init_tables(bs, extent, errp);
 733     if (ret) {
 734         /* free extent allocated by vmdk_add_extent */
 735         vmdk_free_last_extent(bs);
 736     }
 737     return ret;
 738 }
 739
 740 /* find an option value out of descriptor file */
 741 static int vmdk_parse_description(const char *desc, const char *opt_name,
 742         char *buf, int buf_size)
 743 {
 744     char *opt_pos, *opt_end;
 745     const char *end = desc + strlen(desc);
 746
 747     opt_pos = strstr(desc, opt_name);
 748     if (!opt_pos) {
 749         return VMDK_ERROR;
 750     }
 751     /* Skip "=\"" following opt_name */
 752     opt_pos += strlen(opt_name) + 2;
 753     if (opt_pos >= end) {
 754         return VMDK_ERROR;
 755     }
 756     opt_end = opt_pos;
 757     while (opt_end < end && *opt_end != '"') {
 758         opt_end++;
 759     }
 760     if (opt_end == end || buf_size < opt_end - opt_pos + 1) {
 761         return VMDK_ERROR;
 762     }
 763     pstrcpy(buf, opt_end - opt_pos + 1, opt_pos);
 764     return VMDK_OK;
 765 }
 766
 767 /* Open an extent file and append to bs array */
 768 static int vmdk_open_sparse(BlockDriverState *bs, BdrvChild *file, int flags,
 769                             char *buf, QDict *options, Error **errp)
 770 {
 771     uint32_t magic;
 772
 773     magic = ldl_be_p(buf);
 774     switch (magic) {
 775         case VMDK3_MAGIC:
 776             return vmdk_open_vmfs_sparse(bs, file, flags, errp);
 777             break;
 778         case VMDK4_MAGIC:
 779             return vmdk_open_vmdk4(bs, file, flags, options, errp);
 780             break;
 781         default:
 782             error_setg(errp, "Image not in VMDK format");
 783             return -EINVAL;
 784             break;
 785     }
 786 }
 787
 788 static const char *next_line(const char *s)
 789 {
 790     while (*s) {
 791         if (*s == '\n') {
 792             return s + 1;
 793         }
 794         s++;
 795     }
 796     return s;
 797 }
 798
 799 static int vmdk_parse_extents(const char *desc, BlockDriverState *bs,
 800                               const char *desc_file_path, QDict *options,
 801                               Error **errp)
 802 {
 803     int ret;
 804     int matches;
 805     char access[11];
 806     char type[11];
 807     char fname[512];
 808     const char *p, *np;
 809     int64_t sectors = 0;
 810     int64_t flat_offset;
 811     char *extent_path;
 812     BdrvChild *extent_file;
 813     BDRVVmdkState *s = bs->opaque;
 814     VmdkExtent *extent;
 815     char extent_opt_prefix[32];
 816     Error *local_err = NULL;
 817
 818     for (p = desc; *p; p = next_line(p)) {
 819         /* parse extent line in one of below formats:
 820          *
 821          * RW [size in sectors] FLAT "file-name.vmdk" OFFSET
 822          * RW [size in sectors] SPARSE "file-name.vmdk"
 823          * RW [size in sectors] VMFS "file-name.vmdk"
 824          * RW [size in sectors] VMFSSPARSE "file-name.vmdk"
 825          */
 826         flat_offset = -1;
 827         matches = sscanf(p, "%10s %" SCNd64 " %10s \"%511[^\n\r\"]\" %" SCNd64,
 828                          access, &sectors, type, fname, &flat_offset);
 829         if (matches < 4 || strcmp(access, "RW")) {
 830             continue;
 831         } else if (!strcmp(type, "FLAT")) {
 832             if (matches != 5 || flat_offset < 0) {
 833                 goto invalid;
 834             }
 835         } else if (!strcmp(type, "VMFS")) {
 836             if (matches == 4) {
 837                 flat_offset = 0;
 838             } else {
 839                 goto invalid;
 840             }
 841         } else if (matches != 4) {
 842             goto invalid;
 843         }
 844
 845         if (sectors <= 0 ||
 846             (strcmp(type, "FLAT") && strcmp(type, "SPARSE") &&
 847              strcmp(type, "VMFS") && strcmp(type, "VMFSSPARSE")) ||
 848             (strcmp(access, "RW"))) {
 849             continue;
 850         }
 851
 852         if (!path_is_absolute(fname) && !path_has_protocol(fname) &&
 853             !desc_file_path[0])
 854         {
 855             error_setg(errp, "Cannot use relative extent paths with VMDK "
 856                        "descriptor file '%s'", bs->file->bs->filename);
 857             return -EINVAL;
 858         }
 859
 860         extent_path = g_malloc0(PATH_MAX);
 861         path_combine(extent_path, PATH_MAX, desc_file_path, fname);
 862
 863         ret = snprintf(extent_opt_prefix, 32, "extents.%d", s->num_extents);
 864         assert(ret < 32);
 865
 866         extent_file = bdrv_open_child(extent_path, options, extent_opt_prefix,
 867                                       bs, &child_file, false, &local_err);
 868         g_free(extent_path);
 869         if (local_err) {
 870             error_propagate(errp, local_err);
 871             return -EINVAL;
 872         }
 873
 874         /* save to extents array */
 875         if (!strcmp(type, "FLAT") || !strcmp(type, "VMFS")) {
 876             /* FLAT extent */
 877
 878             ret = vmdk_add_extent(bs, extent_file, true, sectors,
 879                             0, 0, 0, 0, 0, &extent, errp);
 880             if (ret < 0) {
 881                 bdrv_unref_child(bs, extent_file);
 882                 return ret;
 883             }
 884             extent->flat_start_offset = flat_offset << 9;
 885         } else if (!strcmp(type, "SPARSE") || !strcmp(type, "VMFSSPARSE")) {
 886             /* SPARSE extent and VMFSSPARSE extent are both "COWD" sparse file*/
 887             char *buf = vmdk_read_desc(extent_file, 0, errp);
 888             if (!buf) {
 889                 ret = -EINVAL;
 890             } else {
 891                 ret = vmdk_open_sparse(bs, extent_file, bs->open_flags, buf,
 892                                        options, errp);
 893             }
 894             g_free(buf);
 895             if (ret) {
 896                 bdrv_unref_child(bs, extent_file);
 897                 return ret;
 898             }
 899             extent = &s->extents[s->num_extents - 1];
 900         } else {
 901             error_setg(errp, "Unsupported extent type '%s'", type);
 902             bdrv_unref_child(bs, extent_file);
 903             return -ENOTSUP;
 904         }
 905         extent->type = g_strdup(type);
 906     }
 907     return 0;
 908
 909 invalid:
 910     np = next_line(p);
 911     assert(np != p);
 912     if (np[-1] == '\n') {
 913         np--;
 914     }
 915     error_setg(errp, "Invalid extent line: %.*s", (int)(np - p), p);
 916     return -EINVAL;
 917 }
 918
 919 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf,
 920                                QDict *options, Error **errp)
 921 {
 922     int ret;
 923     char ct[128];
 924     BDRVVmdkState *s = bs->opaque;
 925
 926     if (vmdk_parse_description(buf, "createType", ct, sizeof(ct))) {
 927         error_setg(errp, "invalid VMDK image descriptor");
 928         ret = -EINVAL;
 929         goto exit;
 930     }
 931     if (strcmp(ct, "monolithicFlat") &&
 932         strcmp(ct, "vmfs") &&
 933         strcmp(ct, "vmfsSparse") &&
 934         strcmp(ct, "twoGbMaxExtentSparse") &&
 935         strcmp(ct, "twoGbMaxExtentFlat")) {
 936         error_setg(errp, "Unsupported image type '%s'", ct);
 937         ret = -ENOTSUP;
 938         goto exit;
 939     }
 940     s->create_type = g_strdup(ct);
 941     s->desc_offset = 0;
 942     ret = vmdk_parse_extents(buf, bs, bs->file->bs->exact_filename, options,
 943                              errp);
 944 exit:
 945     return ret;
 946 }
 947
 948 static int vmdk_open(BlockDriverState *bs, QDict *options, int flags,
 949                      Error **errp)
 950 {
 951     char *buf;
 952     int ret;
 953     BDRVVmdkState *s = bs->opaque;
 954     uint32_t magic;
 955     Error *local_err = NULL;
 956
 957     bs->file = bdrv_open_child(NULL, options, "file", bs, &child_file,
 958                                false, errp);
 959     if (!bs->file) {
 960         return -EINVAL;
 961     }
 962
 963     buf = vmdk_read_desc(bs->file, 0, errp);
 964     if (!buf) {
 965         return -EINVAL;
 966     }
 967
 968     magic = ldl_be_p(buf);
 969     switch (magic) {
 970         case VMDK3_MAGIC:
 971         case VMDK4_MAGIC:
 972             ret = vmdk_open_sparse(bs, bs->file, flags, buf, options,
 973                                    errp);
 974             s->desc_offset = 0x200;
 975             break;
 976         default:
 977             ret = vmdk_open_desc_file(bs, flags, buf, options, errp);
 978             break;
 979     }
 980     if (ret) {
 981         goto fail;
 982     }
 983
 984     /* try to open parent images, if exist */
 985     ret = vmdk_parent_open(bs);
 986     if (ret) {
 987         goto fail;
 988     }
 989     ret = vmdk_read_cid(bs, 0, &s->cid);
 990     if (ret) {
 991         goto fail;
 992     }
 993     ret = vmdk_read_cid(bs, 1, &s->parent_cid);
 994     if (ret) {
 995         goto fail;
 996     }
 997     qemu_co_mutex_init(&s->lock);
 998
 999     /* Disable migration when VMDK images are used */
1000     error_setg(&s->migration_blocker, "The vmdk format used by node '%s' "
1001                "does not support live migration",
1002                bdrv_get_device_or_node_name(bs));
1003     ret = migrate_add_blocker(s->migration_blocker, &local_err);
1004     if (local_err) {
1005         error_propagate(errp, local_err);
1006         error_free(s->migration_blocker);
1007         goto fail;
1008     }
1009
1010     g_free(buf);
1011     return 0;
1012
1013 fail:
1014     g_free(buf);
1015     g_free(s->create_type);
1016     s->create_type = NULL;
1017     vmdk_free_extents(bs);
1018     return ret;
1019 }
1020
1021
1022 static void vmdk_refresh_limits(BlockDriverState *bs, Error **errp)
1023 {
1024     BDRVVmdkState *s = bs->opaque;
1025     int i;
1026
1027     for (i = 0; i < s->num_extents; i++) {
1028         if (!s->extents[i].flat) {
1029             bs->bl.pwrite_zeroes_alignment =
1030                 MAX(bs->bl.pwrite_zeroes_alignment,
1031                     s->extents[i].cluster_sectors << BDRV_SECTOR_BITS);
1032         }
1033     }
1034 }
1035
1036 /**
1037  * get_whole_cluster
1038  *
1039  * Copy backing file's cluster that covers @sector_num, otherwise write zero,
1040  * to the cluster at @cluster_sector_num.
1041  *
1042  * If @skip_start_sector < @skip_end_sector, the relative range
1043  * [@skip_start_sector, @skip_end_sector) is not copied or written, and leave
1044  * it for call to write user data in the request.
1045  */
1046 static int get_whole_cluster(BlockDriverState *bs,
1047                              VmdkExtent *extent,
1048                              uint64_t cluster_offset,
1049                              uint64_t offset,
1050                              uint64_t skip_start_bytes,
1051                              uint64_t skip_end_bytes)
1052 {
1053     int ret = VMDK_OK;
1054     int64_t cluster_bytes;
1055     uint8_t *whole_grain;
1056
1057     /* For COW, align request sector_num to cluster start */
1058     cluster_bytes = extent->cluster_sectors << BDRV_SECTOR_BITS;
1059     offset = QEMU_ALIGN_DOWN(offset, cluster_bytes);
1060     whole_grain = qemu_blockalign(bs, cluster_bytes);
1061
1062     if (!bs->backing) {
1063         memset(whole_grain, 0, skip_start_bytes);
1064         memset(whole_grain + skip_end_bytes, 0, cluster_bytes - skip_end_bytes);
1065     }
1066
1067     assert(skip_end_bytes <= cluster_bytes);
1068     /* we will be here if it's first write on non-exist grain(cluster).
1069      * try to read from parent image, if exist */
1070     if (bs->backing && !vmdk_is_cid_valid(bs)) {
1071         ret = VMDK_ERROR;
1072         goto exit;
1073     }
1074
1075     /* Read backing data before skip range */
1076     if (skip_start_bytes > 0) {
1077         if (bs->backing) {
1078             /* qcow2 emits this on bs->file instead of bs->backing */
1079             BLKDBG_EVENT(extent->file, BLKDBG_COW_READ);
1080             ret = bdrv_pread(bs->backing, offset, whole_grain,
1081                              skip_start_bytes);
1082             if (ret < 0) {
1083                 ret = VMDK_ERROR;
1084                 goto exit;
1085             }
1086         }
1087         BLKDBG_EVENT(extent->file, BLKDBG_COW_WRITE);
1088         ret = bdrv_pwrite(extent->file, cluster_offset, whole_grain,
1089                           skip_start_bytes);
1090         if (ret < 0) {
1091             ret = VMDK_ERROR;
1092             goto exit;
1093         }
1094     }
1095     /* Read backing data after skip range */
1096     if (skip_end_bytes < cluster_bytes) {
1097         if (bs->backing) {
1098             /* qcow2 emits this on bs->file instead of bs->backing */
1099             BLKDBG_EVENT(extent->file, BLKDBG_COW_READ);
1100             ret = bdrv_pread(bs->backing, offset + skip_end_bytes,
1101                              whole_grain + skip_end_bytes,
1102                              cluster_bytes - skip_end_bytes);
1103             if (ret < 0) {
1104                 ret = VMDK_ERROR;
1105                 goto exit;
1106             }
1107         }
1108         BLKDBG_EVENT(extent->file, BLKDBG_COW_WRITE);
1109         ret = bdrv_pwrite(extent->file, cluster_offset + skip_end_bytes,
1110                           whole_grain + skip_end_bytes,
1111                           cluster_bytes - skip_end_bytes);
1112         if (ret < 0) {
1113             ret = VMDK_ERROR;
1114             goto exit;
1115         }
1116     }
1117
1118     ret = VMDK_OK;
1119 exit:
1120     qemu_vfree(whole_grain);
1121     return ret;
1122 }
1123
1124 static int vmdk_L2update(VmdkExtent *extent, VmdkMetaData *m_data,
1125                          uint32_t offset)
1126 {
1127     offset = cpu_to_le32(offset);
1128     /* update L2 table */
1129     BLKDBG_EVENT(extent->file, BLKDBG_L2_UPDATE);
1130     if (bdrv_pwrite_sync(extent->file,
1131                 ((int64_t)m_data->l2_offset * 512)
1132                     + (m_data->l2_index * sizeof(offset)),
1133                 &offset, sizeof(offset)) < 0) {
1134         return VMDK_ERROR;
1135     }
1136     /* update backup L2 table */
1137     if (extent->l1_backup_table_offset != 0) {
1138         m_data->l2_offset = extent->l1_backup_table[m_data->l1_index];
1139         if (bdrv_pwrite_sync(extent->file,
1140                     ((int64_t)m_data->l2_offset * 512)
1141                         + (m_data->l2_index * sizeof(offset)),
1142                     &offset, sizeof(offset)) < 0) {
1143             return VMDK_ERROR;
1144         }
1145     }
1146     if (m_data->l2_cache_entry) {
1147         *m_data->l2_cache_entry = offset;
1148     }
1149
1150     return VMDK_OK;
1151 }
1152
1153 /**
1154  * get_cluster_offset
1155  *
1156  * Look up cluster offset in extent file by sector number, and store in
1157  * @cluster_offset.
1158  *
1159  * For flat extents, the start offset as parsed from the description file is
1160  * returned.
1161  *
1162  * For sparse extents, look up in L1, L2 table. If allocate is true, return an
1163  * offset for a new cluster and update L2 cache. If there is a backing file,
1164  * COW is done before returning; otherwise, zeroes are written to the allocated
1165  * cluster. Both COW and zero writing skips the sector range
1166  * [@skip_start_sector, @skip_end_sector) passed in by caller, because caller
1167  * has new data to write there.
1168  *
1169  * Returns: VMDK_OK if cluster exists and mapped in the image.
1170  *          VMDK_UNALLOC if cluster is not mapped and @allocate is false.
1171  *          VMDK_ERROR if failed.
1172  */
1173 static int get_cluster_offset(BlockDriverState *bs,
1174                               VmdkExtent *extent,
1175                               VmdkMetaData *m_data,
1176                               uint64_t offset,
1177                               bool allocate,
1178                               uint64_t *cluster_offset,
1179                               uint64_t skip_start_bytes,
1180                               uint64_t skip_end_bytes)
1181 {
1182     unsigned int l1_index, l2_offset, l2_index;
1183     int min_index, i, j;
1184     uint32_t min_count, *l2_table;
1185     bool zeroed = false;
1186     int64_t ret;
1187     int64_t cluster_sector;
1188
1189     if (m_data) {
1190         m_data->valid = 0;
1191     }
1192     if (extent->flat) {
1193         *cluster_offset = extent->flat_start_offset;
1194         return VMDK_OK;
1195     }
1196
1197     offset -= (extent->end_sector - extent->sectors) * SECTOR_SIZE;
1198     l1_index = (offset >> 9) / extent->l1_entry_sectors;
1199     if (l1_index >= extent->l1_size) {
1200         return VMDK_ERROR;
1201     }
1202     l2_offset = extent->l1_table[l1_index];
1203     if (!l2_offset) {
1204         return VMDK_UNALLOC;
1205     }
1206     for (i = 0; i < L2_CACHE_SIZE; i++) {
1207         if (l2_offset == extent->l2_cache_offsets[i]) {
1208             /* increment the hit count */
1209             if (++extent->l2_cache_counts[i] == 0xffffffff) {
1210                 for (j = 0; j < L2_CACHE_SIZE; j++) {
1211                     extent->l2_cache_counts[j] >>= 1;
1212                 }
1213             }
1214             l2_table = extent->l2_cache + (i * extent->l2_size);
1215             goto found;
1216         }
1217     }
1218     /* not found: load a new entry in the least used one */
1219     min_index = 0;
1220     min_count = 0xffffffff;
1221     for (i = 0; i < L2_CACHE_SIZE; i++) {
1222         if (extent->l2_cache_counts[i] < min_count) {
1223             min_count = extent->l2_cache_counts[i];
1224             min_index = i;
1225         }
1226     }
1227     l2_table = extent->l2_cache + (min_index * extent->l2_size);
1228     BLKDBG_EVENT(extent->file, BLKDBG_L2_LOAD);
1229     if (bdrv_pread(extent->file,
1230                 (int64_t)l2_offset * 512,
1231                 l2_table,
1232                 extent->l2_size * sizeof(uint32_t)
1233             ) != extent->l2_size * sizeof(uint32_t)) {
1234         return VMDK_ERROR;
1235     }
1236
1237     extent->l2_cache_offsets[min_index] = l2_offset;
1238     extent->l2_cache_counts[min_index] = 1;
1239  found:
1240     l2_index = ((offset >> 9) / extent->cluster_sectors) % extent->l2_size;
1241     cluster_sector = le32_to_cpu(l2_table[l2_index]);
1242
1243     if (extent->has_zero_grain && cluster_sector == VMDK_GTE_ZEROED) {
1244         zeroed = true;
1245     }
1246
1247     if (!cluster_sector || zeroed) {
1248         if (!allocate) {
1249             return zeroed ? VMDK_ZEROED : VMDK_UNALLOC;
1250         }
1251
1252         cluster_sector = extent->next_cluster_sector;
1253         extent->next_cluster_sector += extent->cluster_sectors;
1254
1255         /* First of all we write grain itself, to avoid race condition
1256          * that may to corrupt the image.
1257          * This problem may occur because of insufficient space on host disk
1258          * or inappropriate VM shutdown.
1259          */
1260         ret = get_whole_cluster(bs, extent, cluster_sector * BDRV_SECTOR_SIZE,
1261                                 offset, skip_start_bytes, skip_end_bytes);
1262         if (ret) {
1263             return ret;
1264         }
1265         if (m_data) {
1266             m_data->valid = 1;
1267             m_data->l1_index = l1_index;
1268             m_data->l2_index = l2_index;
1269             m_data->l2_offset = l2_offset;
1270             m_data->l2_cache_entry = &l2_table[l2_index];
1271         }
1272     }
1273     *cluster_offset = cluster_sector << BDRV_SECTOR_BITS;
1274     return VMDK_OK;
1275 }
1276
1277 static VmdkExtent *find_extent(BDRVVmdkState *s,
1278                                 int64_t sector_num, VmdkExtent *start_hint)
1279 {
1280     VmdkExtent *extent = start_hint;
1281
1282     if (!extent) {
1283         extent = &s->extents[0];
1284     }
1285     while (extent < &s->extents[s->num_extents]) {
1286         if (sector_num < extent->end_sector) {
1287             return extent;
1288         }
1289         extent++;
1290     }
1291     return NULL;
1292 }
1293
1294 static inline uint64_t vmdk_find_offset_in_cluster(VmdkExtent *extent,
1295                                                    int64_t offset)
1296 {
1297     uint64_t extent_begin_offset, extent_relative_offset;
1298     uint64_t cluster_size = extent->cluster_sectors * BDRV_SECTOR_SIZE;
1299
1300     extent_begin_offset =
1301         (extent->end_sector - extent->sectors) * BDRV_SECTOR_SIZE;
1302     extent_relative_offset = offset - extent_begin_offset;
1303     return extent_relative_offset % cluster_size;
1304 }
1305
1306 static inline uint64_t vmdk_find_index_in_cluster(VmdkExtent *extent,
1307                                                   int64_t sector_num)
1308 {
1309     uint64_t offset;
1310     offset = vmdk_find_offset_in_cluster(extent, sector_num * BDRV_SECTOR_SIZE);
1311     return offset / BDRV_SECTOR_SIZE;
1312 }
1313
1314 static int64_t coroutine_fn vmdk_co_get_block_status(BlockDriverState *bs,
1315         int64_t sector_num, int nb_sectors, int *pnum, BlockDriverState **file)
1316 {
1317     BDRVVmdkState *s = bs->opaque;
1318     int64_t index_in_cluster, n, ret;
1319     uint64_t offset;
1320     VmdkExtent *extent;
1321
1322     extent = find_extent(s, sector_num, NULL);
1323     if (!extent) {
1324         return 0;
1325     }
1326     qemu_co_mutex_lock(&s->lock);
1327     ret = get_cluster_offset(bs, extent, NULL,
1328                              sector_num * 512, false, &offset,
1329                              0, 0);
1330     qemu_co_mutex_unlock(&s->lock);
1331
1332     index_in_cluster = vmdk_find_index_in_cluster(extent, sector_num);
1333     switch (ret) {
1334     case VMDK_ERROR:
1335         ret = -EIO;
1336         break;
1337     case VMDK_UNALLOC:
1338         ret = 0;
1339         break;
1340     case VMDK_ZEROED:
1341         ret = BDRV_BLOCK_ZERO;
1342         break;
1343     case VMDK_OK:
1344         ret = BDRV_BLOCK_DATA;
1345         if (!extent->compressed) {
1346             ret |= BDRV_BLOCK_OFFSET_VALID;
1347             ret |= (offset + (index_in_cluster << BDRV_SECTOR_BITS))
1348                     & BDRV_BLOCK_OFFSET_MASK;
1349         }
1350         *file = extent->file->bs;
1351         break;
1352     }
1353
1354     n = extent->cluster_sectors - index_in_cluster;
1355     if (n > nb_sectors) {
1356         n = nb_sectors;
1357     }
1358     *pnum = n;
1359     return ret;
1360 }
1361
1362 static int vmdk_write_extent(VmdkExtent *extent, int64_t cluster_offset,
1363                             int64_t offset_in_cluster, QEMUIOVector *qiov,
1364                             uint64_t qiov_offset, uint64_t n_bytes,
1365                             uint64_t offset)
1366 {
1367     int ret;
1368     VmdkGrainMarker *data = NULL;
1369     uLongf buf_len;
1370     QEMUIOVector local_qiov;
1371     struct iovec iov;
1372     int64_t write_offset;
1373     int64_t write_end_sector;
1374
1375     if (extent->compressed) {
1376         void *compressed_data;
1377
1378         if (!extent->has_marker) {
1379             ret = -EINVAL;
1380             goto out;
1381         }
1382         buf_len = (extent->cluster_sectors << 9) * 2;
1383         data = g_malloc(buf_len + sizeof(VmdkGrainMarker));
1384
1385         compressed_data = g_malloc(n_bytes);
1386         qemu_iovec_to_buf(qiov, qiov_offset, compressed_data, n_bytes);
1387         ret = compress(data->data, &buf_len, compressed_data, n_bytes);
1388         g_free(compressed_data);
1389
1390         if (ret != Z_OK || buf_len == 0) {
1391             ret = -EINVAL;
1392             goto out;
1393         }
1394
1395         data->lba = cpu_to_le64(offset >> BDRV_SECTOR_BITS);
1396         data->size = cpu_to_le32(buf_len);
1397
1398         n_bytes = buf_len + sizeof(VmdkGrainMarker);
1399         iov = (struct iovec) {
1400             .iov_base   = data,
1401             .iov_len    = n_bytes,
1402         };
1403         qemu_iovec_init_external(&local_qiov, &iov, 1);
1404
1405         BLKDBG_EVENT(extent->file, BLKDBG_WRITE_COMPRESSED);
1406     } else {
1407         qemu_iovec_init(&local_qiov, qiov->niov);
1408         qemu_iovec_concat(&local_qiov, qiov, qiov_offset, n_bytes);
1409
1410         BLKDBG_EVENT(extent->file, BLKDBG_WRITE_AIO);
1411     }
1412
1413     write_offset = cluster_offset + offset_in_cluster;
1414     ret = bdrv_co_pwritev(extent->file, write_offset, n_bytes,
1415                           &local_qiov, 0);
1416
1417     write_end_sector = DIV_ROUND_UP(write_offset + n_bytes, BDRV_SECTOR_SIZE);
1418
1419     if (extent->compressed) {
1420         extent->next_cluster_sector = write_end_sector;
1421     } else {
1422         extent->next_cluster_sector = MAX(extent->next_cluster_sector,
1423                                           write_end_sector);
1424     }
1425
1426     if (ret < 0) {
1427         goto out;
1428     }
1429     ret = 0;
1430  out:
1431     g_free(data);
1432     if (!extent->compressed) {
1433         qemu_iovec_destroy(&local_qiov);
1434     }
1435     return ret;
1436 }
1437
1438 static int vmdk_read_extent(VmdkExtent *extent, int64_t cluster_offset,
1439                             int64_t offset_in_cluster, QEMUIOVector *qiov,
1440                             int bytes)
1441 {
1442     int ret;
1443     int cluster_bytes, buf_bytes;
1444     uint8_t *cluster_buf, *compressed_data;
1445     uint8_t *uncomp_buf;
1446     uint32_t data_len;
1447     VmdkGrainMarker *marker;
1448     uLongf buf_len;
1449
1450
1451     if (!extent->compressed) {
1452         BLKDBG_EVENT(extent->file, BLKDBG_READ_AIO);
1453         ret = bdrv_co_preadv(extent->file,
1454                              cluster_offset + offset_in_cluster, bytes,
1455                              qiov, 0);
1456         if (ret < 0) {
1457             return ret;
1458         }
1459         return 0;
1460     }
1461     cluster_bytes = extent->cluster_sectors * 512;
1462     /* Read two clusters in case GrainMarker + compressed data > one cluster */
1463     buf_bytes = cluster_bytes * 2;
1464     cluster_buf = g_malloc(buf_bytes);
1465     uncomp_buf = g_malloc(cluster_bytes);
1466     BLKDBG_EVENT(extent->file, BLKDBG_READ_COMPRESSED);
1467     ret = bdrv_pread(extent->file,
1468                 cluster_offset,
1469                 cluster_buf, buf_bytes);
1470     if (ret < 0) {
1471         goto out;
1472     }
1473     compressed_data = cluster_buf;
1474     buf_len = cluster_bytes;
1475     data_len = cluster_bytes;
1476     if (extent->has_marker) {
1477         marker = (VmdkGrainMarker *)cluster_buf;
1478         compressed_data = marker->data;
1479         data_len = le32_to_cpu(marker->size);
1480     }
1481     if (!data_len || data_len > buf_bytes) {
1482         ret = -EINVAL;
1483         goto out;
1484     }
1485     ret = uncompress(uncomp_buf, &buf_len, compressed_data, data_len);
1486     if (ret != Z_OK) {
1487         ret = -EINVAL;
1488         goto out;
1489
1490     }
1491     if (offset_in_cluster < 0 ||
1492             offset_in_cluster + bytes > buf_len) {
1493         ret = -EINVAL;
1494         goto out;
1495     }
1496     qemu_iovec_from_buf(qiov, 0, uncomp_buf + offset_in_cluster, bytes);
1497     ret = 0;
1498
1499  out:
1500     g_free(uncomp_buf);
1501     g_free(cluster_buf);
1502     return ret;
1503 }
1504
1505 static int coroutine_fn
1506 vmdk_co_preadv(BlockDriverState *bs, uint64_t offset, uint64_t bytes,
1507                QEMUIOVector *qiov, int flags)
1508 {
1509     BDRVVmdkState *s = bs->opaque;
1510     int ret;
1511     uint64_t n_bytes, offset_in_cluster;
1512     VmdkExtent *extent = NULL;
1513     QEMUIOVector local_qiov;
1514     uint64_t cluster_offset;
1515     uint64_t bytes_done = 0;
1516
1517     qemu_iovec_init(&local_qiov, qiov->niov);
1518     qemu_co_mutex_lock(&s->lock);
1519
1520     while (bytes > 0) {
1521         extent = find_extent(s, offset >> BDRV_SECTOR_BITS, extent);
1522         if (!extent) {
1523             ret = -EIO;
1524             goto fail;
1525         }
1526         ret = get_cluster_offset(bs, extent, NULL,
1527                                  offset, false, &cluster_offset, 0, 0);
1528         offset_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
1529
1530         n_bytes = MIN(bytes, extent->cluster_sectors * BDRV_SECTOR_SIZE
1531                              - offset_in_cluster);
1532
1533         if (ret != VMDK_OK) {
1534             /* if not allocated, try to read from parent image, if exist */
1535             if (bs->backing && ret != VMDK_ZEROED) {
1536                 if (!vmdk_is_cid_valid(bs)) {
1537                     ret = -EINVAL;
1538                     goto fail;
1539                 }
1540
1541                 qemu_iovec_reset(&local_qiov);
1542                 qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes);
1543
1544                 /* qcow2 emits this on bs->file instead of bs->backing */
1545                 BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
1546                 ret = bdrv_co_preadv(bs->backing, offset, n_bytes,
1547                                      &local_qiov, 0);
1548                 if (ret < 0) {
1549                     goto fail;
1550                 }
1551             } else {
1552                 qemu_iovec_memset(qiov, bytes_done, 0, n_bytes);
1553             }
1554         } else {
1555             qemu_iovec_reset(&local_qiov);
1556             qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes);
1557
1558             ret = vmdk_read_extent(extent, cluster_offset, offset_in_cluster,
1559                                    &local_qiov, n_bytes);
1560             if (ret) {
1561                 goto fail;
1562             }
1563         }
1564         bytes -= n_bytes;
1565         offset += n_bytes;
1566         bytes_done += n_bytes;
1567     }
1568
1569     ret = 0;
1570 fail:
1571     qemu_co_mutex_unlock(&s->lock);
1572     qemu_iovec_destroy(&local_qiov);
1573
1574     return ret;
1575 }
1576
1577 /**
1578  * vmdk_write:
1579  * @zeroed:       buf is ignored (data is zero), use zeroed_grain GTE feature
1580  *                if possible, otherwise return -ENOTSUP.
1581  * @zero_dry_run: used for zeroed == true only, don't update L2 table, just try
1582  *                with each cluster. By dry run we can find if the zero write
1583  *                is possible without modifying image data.
1584  *
1585  * Returns: error code with 0 for success.
1586  */
1587 static int vmdk_pwritev(BlockDriverState *bs, uint64_t offset,
1588                        uint64_t bytes, QEMUIOVector *qiov,
1589                        bool zeroed, bool zero_dry_run)
1590 {
1591     BDRVVmdkState *s = bs->opaque;
1592     VmdkExtent *extent = NULL;
1593     int ret;
1594     int64_t offset_in_cluster, n_bytes;
1595     uint64_t cluster_offset;
1596     uint64_t bytes_done = 0;
1597     VmdkMetaData m_data;
1598
1599     if (DIV_ROUND_UP(offset, BDRV_SECTOR_SIZE) > bs->total_sectors) {
1600         error_report("Wrong offset: offset=0x%" PRIx64
1601                      " total_sectors=0x%" PRIx64,
1602                      offset, bs->total_sectors);
1603         return -EIO;
1604     }
1605
1606     while (bytes > 0) {
1607         extent = find_extent(s, offset >> BDRV_SECTOR_BITS, extent);
1608         if (!extent) {
1609             return -EIO;
1610         }
1611         offset_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
1612         n_bytes = MIN(bytes, extent->cluster_sectors * BDRV_SECTOR_SIZE
1613                              - offset_in_cluster);
1614
1615         ret = get_cluster_offset(bs, extent, &m_data, offset,
1616                                  !(extent->compressed || zeroed),
1617                                  &cluster_offset, offset_in_cluster,
1618                                  offset_in_cluster + n_bytes);
1619         if (extent->compressed) {
1620             if (ret == VMDK_OK) {
1621                 /* Refuse write to allocated cluster for streamOptimized */
1622                 error_report("Could not write to allocated cluster"
1623                               " for streamOptimized");
1624                 return -EIO;
1625             } else {
1626                 /* allocate */
1627                 ret = get_cluster_offset(bs, extent, &m_data, offset,
1628                                          true, &cluster_offset, 0, 0);
1629             }
1630         }
1631         if (ret == VMDK_ERROR) {
1632             return -EINVAL;
1633         }
1634         if (zeroed) {
1635             /* Do zeroed write, buf is ignored */
1636             if (extent->has_zero_grain &&
1637                     offset_in_cluster == 0 &&
1638                     n_bytes >= extent->cluster_sectors * BDRV_SECTOR_SIZE) {
1639                 n_bytes = extent->cluster_sectors * BDRV_SECTOR_SIZE;
1640                 if (!zero_dry_run) {
1641                     /* update L2 tables */
1642                     if (vmdk_L2update(extent, &m_data, VMDK_GTE_ZEROED)
1643                             != VMDK_OK) {
1644                         return -EIO;
1645                     }
1646                 }
1647             } else {
1648                 return -ENOTSUP;
1649             }
1650         } else {
1651             ret = vmdk_write_extent(extent, cluster_offset, offset_in_cluster,
1652                                     qiov, bytes_done, n_bytes, offset);
1653             if (ret) {
1654                 return ret;
1655             }
1656             if (m_data.valid) {
1657                 /* update L2 tables */
1658                 if (vmdk_L2update(extent, &m_data,
1659                                   cluster_offset >> BDRV_SECTOR_BITS)
1660                         != VMDK_OK) {
1661                     return -EIO;
1662                 }
1663             }
1664         }
1665         bytes -= n_bytes;
1666         offset += n_bytes;
1667         bytes_done += n_bytes;
1668
1669         /* update CID on the first write every time the virtual disk is
1670          * opened */
1671         if (!s->cid_updated) {
1672             ret = vmdk_write_cid(bs, g_random_int());
1673             if (ret < 0) {
1674                 return ret;
1675             }
1676             s->cid_updated = true;
1677         }
1678     }
1679     return 0;
1680 }
1681
1682 static int coroutine_fn
1683 vmdk_co_pwritev(BlockDriverState *bs, uint64_t offset, uint64_t bytes,
1684                 QEMUIOVector *qiov, int flags)
1685 {
1686     int ret;
1687     BDRVVmdkState *s = bs->opaque;
1688     qemu_co_mutex_lock(&s->lock);
1689     ret = vmdk_pwritev(bs, offset, bytes, qiov, false, false);
1690     qemu_co_mutex_unlock(&s->lock);
1691     return ret;
1692 }
1693
1694 static int coroutine_fn
1695 vmdk_co_pwritev_compressed(BlockDriverState *bs, uint64_t offset,
1696                            uint64_t bytes, QEMUIOVector *qiov)
1697 {
1698     return vmdk_co_pwritev(bs, offset, bytes, qiov, 0);
1699 }
1700
1701 static int coroutine_fn vmdk_co_pwrite_zeroes(BlockDriverState *bs,
1702                                               int64_t offset,
1703                                               int bytes,
1704                                               BdrvRequestFlags flags)
1705 {
1706     int ret;
1707     BDRVVmdkState *s = bs->opaque;
1708
1709     qemu_co_mutex_lock(&s->lock);
1710     /* write zeroes could fail if sectors not aligned to cluster, test it with
1711      * dry_run == true before really updating image */
1712     ret = vmdk_pwritev(bs, offset, bytes, NULL, true, true);
1713     if (!ret) {
1714         ret = vmdk_pwritev(bs, offset, bytes, NULL, true, false);
1715     }
1716     qemu_co_mutex_unlock(&s->lock);
1717     return ret;
1718 }
1719
1720 static int vmdk_create_extent(const char *filename, int64_t filesize,
1721                               bool flat, bool compress, bool zeroed_grain,
1722                               QemuOpts *opts, Error **errp)
1723 {
1724     int ret, i;
1725     BlockBackend *blk = NULL;
1726     VMDK4Header header;
1727     Error *local_err = NULL;
1728     uint32_t tmp, magic, grains, gd_sectors, gt_size, gt_count;
1729     uint32_t *gd_buf = NULL;
1730     int gd_buf_size;
1731
1732     ret = bdrv_create_file(filename, opts, &local_err);
1733     if (ret < 0) {
1734         error_propagate(errp, local_err);
1735         goto exit;
1736     }
1737
1738     blk = blk_new_open(filename, NULL, NULL,
1739                        BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL,
1740                        &local_err);
1741     if (blk == NULL) {
1742         error_propagate(errp, local_err);
1743         ret = -EIO;
1744         goto exit;
1745     }
1746
1747     blk_set_allow_write_beyond_eof(blk, true);
1748
1749     if (flat) {
1750         ret = blk_truncate(blk, filesize, PREALLOC_MODE_OFF, errp);
1751         goto exit;
1752     }
1753     magic = cpu_to_be32(VMDK4_MAGIC);
1754     memset(&header, 0, sizeof(header));
1755     if (compress) {
1756         header.version = 3;
1757     } else if (zeroed_grain) {
1758         header.version = 2;
1759     } else {
1760         header.version = 1;
1761     }
1762     header.flags = VMDK4_FLAG_RGD | VMDK4_FLAG_NL_DETECT
1763                    | (compress ? VMDK4_FLAG_COMPRESS | VMDK4_FLAG_MARKER : 0)
1764                    | (zeroed_grain ? VMDK4_FLAG_ZERO_GRAIN : 0);
1765     header.compressAlgorithm = compress ? VMDK4_COMPRESSION_DEFLATE : 0;
1766     header.capacity = filesize / BDRV_SECTOR_SIZE;
1767     header.granularity = 128;
1768     header.num_gtes_per_gt = BDRV_SECTOR_SIZE;
1769
1770     grains = DIV_ROUND_UP(filesize / BDRV_SECTOR_SIZE, header.granularity);
1771     gt_size = DIV_ROUND_UP(header.num_gtes_per_gt * sizeof(uint32_t),
1772                            BDRV_SECTOR_SIZE);
1773     gt_count = DIV_ROUND_UP(grains, header.num_gtes_per_gt);
1774     gd_sectors = DIV_ROUND_UP(gt_count * sizeof(uint32_t), BDRV_SECTOR_SIZE);
1775
1776     header.desc_offset = 1;
1777     header.desc_size = 20;
1778     header.rgd_offset = header.desc_offset + header.desc_size;
1779     header.gd_offset = header.rgd_offset + gd_sectors + (gt_size * gt_count);
1780     header.grain_offset =
1781         ROUND_UP(header.gd_offset + gd_sectors + (gt_size * gt_count),
1782                  header.granularity);
1783     /* swap endianness for all header fields */
1784     header.version = cpu_to_le32(header.version);
1785     header.flags = cpu_to_le32(header.flags);
1786     header.capacity = cpu_to_le64(header.capacity);
1787     header.granularity = cpu_to_le64(header.granularity);
1788     header.num_gtes_per_gt = cpu_to_le32(header.num_gtes_per_gt);
1789     header.desc_offset = cpu_to_le64(header.desc_offset);
1790     header.desc_size = cpu_to_le64(header.desc_size);
1791     header.rgd_offset = cpu_to_le64(header.rgd_offset);
1792     header.gd_offset = cpu_to_le64(header.gd_offset);
1793     header.grain_offset = cpu_to_le64(header.grain_offset);
1794     header.compressAlgorithm = cpu_to_le16(header.compressAlgorithm);
1795
1796     header.check_bytes[0] = 0xa;
1797     header.check_bytes[1] = 0x20;
1798     header.check_bytes[2] = 0xd;
1799     header.check_bytes[3] = 0xa;
1800
1801     /* write all the data */
1802     ret = blk_pwrite(blk, 0, &magic, sizeof(magic), 0);
1803     if (ret < 0) {
1804         error_setg(errp, QERR_IO_ERROR);
1805         goto exit;
1806     }
1807     ret = blk_pwrite(blk, sizeof(magic), &header, sizeof(header), 0);
1808     if (ret < 0) {
1809         error_setg(errp, QERR_IO_ERROR);
1810         goto exit;
1811     }
1812
1813     ret = blk_truncate(blk, le64_to_cpu(header.grain_offset) << 9,
1814                        PREALLOC_MODE_OFF, errp);
1815     if (ret < 0) {
1816         goto exit;
1817     }
1818
1819     /* write grain directory */
1820     gd_buf_size = gd_sectors * BDRV_SECTOR_SIZE;
1821     gd_buf = g_malloc0(gd_buf_size);
1822     for (i = 0, tmp = le64_to_cpu(header.rgd_offset) + gd_sectors;
1823          i < gt_count; i++, tmp += gt_size) {
1824         gd_buf[i] = cpu_to_le32(tmp);
1825     }
1826     ret = blk_pwrite(blk, le64_to_cpu(header.rgd_offset) * BDRV_SECTOR_SIZE,
1827                      gd_buf, gd_buf_size, 0);
1828     if (ret < 0) {
1829         error_setg(errp, QERR_IO_ERROR);
1830         goto exit;
1831     }
1832
1833     /* write backup grain directory */
1834     for (i = 0, tmp = le64_to_cpu(header.gd_offset) + gd_sectors;
1835          i < gt_count; i++, tmp += gt_size) {
1836         gd_buf[i] = cpu_to_le32(tmp);
1837     }
1838     ret = blk_pwrite(blk, le64_to_cpu(header.gd_offset) * BDRV_SECTOR_SIZE,
1839                      gd_buf, gd_buf_size, 0);
1840     if (ret < 0) {
1841         error_setg(errp, QERR_IO_ERROR);
1842         goto exit;
1843     }
1844
1845     ret = 0;
1846 exit:
1847     if (blk) {
1848         blk_unref(blk);
1849     }
1850     g_free(gd_buf);
1851     return ret;
1852 }
1853
1854 static int filename_decompose(const char *filename, char *path, char *prefix,
1855                               char *postfix, size_t buf_len, Error **errp)
1856 {
1857     const char *p, *q;
1858
1859     if (filename == NULL || !strlen(filename)) {
1860         error_setg(errp, "No filename provided");
1861         return VMDK_ERROR;
1862     }
1863     p = strrchr(filename, '/');
1864     if (p == NULL) {
1865         p = strrchr(filename, '\\');
1866     }
1867     if (p == NULL) {
1868         p = strrchr(filename, ':');
1869     }
1870     if (p != NULL) {
1871         p++;
1872         if (p - filename >= buf_len) {
1873             return VMDK_ERROR;
1874         }
1875         pstrcpy(path, p - filename + 1, filename);
1876     } else {
1877         p = filename;
1878         path[0] = '\0';
1879     }
1880     q = strrchr(p, '.');
1881     if (q == NULL) {
1882         pstrcpy(prefix, buf_len, p);
1883         postfix[0] = '\0';
1884     } else {
1885         if (q - p >= buf_len) {
1886             return VMDK_ERROR;
1887         }
1888         pstrcpy(prefix, q - p + 1, p);
1889         pstrcpy(postfix, buf_len, q);
1890     }
1891     return VMDK_OK;
1892 }
1893
1894 static int vmdk_create(const char *filename, QemuOpts *opts, Error **errp)
1895 {
1896     int idx = 0;
1897     BlockBackend *new_blk = NULL;
1898     Error *local_err = NULL;
1899     char *desc = NULL;
1900     int64_t total_size = 0, filesize;
1901     char *adapter_type = NULL;
1902     char *backing_file = NULL;
1903     char *hw_version = NULL;
1904     char *fmt = NULL;
1905     int ret = 0;
1906     bool flat, split, compress;
1907     GString *ext_desc_lines;
1908     char *path = g_malloc0(PATH_MAX);
1909     char *prefix = g_malloc0(PATH_MAX);
1910     char *postfix = g_malloc0(PATH_MAX);
1911     char *desc_line = g_malloc0(BUF_SIZE);
1912     char *ext_filename = g_malloc0(PATH_MAX);
1913     char *desc_filename = g_malloc0(PATH_MAX);
1914     const int64_t split_size = 0x80000000;  /* VMDK has constant split size */
1915     const char *desc_extent_line;
1916     char *parent_desc_line = g_malloc0(BUF_SIZE);
1917     uint32_t parent_cid = 0xffffffff;
1918     uint32_t number_heads = 16;
1919     bool zeroed_grain = false;
1920     uint32_t desc_offset = 0, desc_len;
1921     const char desc_template[] =
1922         "# Disk DescriptorFile\n"
1923         "version=1\n"
1924         "CID=%" PRIx32 "\n"
1925         "parentCID=%" PRIx32 "\n"
1926         "createType=\"%s\"\n"
1927         "%s"
1928         "\n"
1929         "# Extent description\n"
1930         "%s"
1931         "\n"
1932         "# The Disk Data Base\n"
1933         "#DDB\n"
1934         "\n"
1935         "ddb.virtualHWVersion = \"%s\"\n"
1936         "ddb.geometry.cylinders = \"%" PRId64 "\"\n"
1937         "ddb.geometry.heads = \"%" PRIu32 "\"\n"
1938         "ddb.geometry.sectors = \"63\"\n"
1939         "ddb.adapterType = \"%s\"\n";
1940
1941     ext_desc_lines = g_string_new(NULL);
1942
1943     if (filename_decompose(filename, path, prefix, postfix, PATH_MAX, errp)) {
1944         ret = -EINVAL;
1945         goto exit;
1946     }
1947     /* Read out options */
1948     total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
1949                           BDRV_SECTOR_SIZE);
1950     adapter_type = qemu_opt_get_del(opts, BLOCK_OPT_ADAPTER_TYPE);
1951     backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
1952     hw_version = qemu_opt_get_del(opts, BLOCK_OPT_HWVERSION);
1953     if (qemu_opt_get_bool_del(opts, BLOCK_OPT_COMPAT6, false)) {
1954         if (strcmp(hw_version, "undefined")) {
1955             error_setg(errp,
1956                        "compat6 cannot be enabled with hwversion set");
1957             ret = -EINVAL;
1958             goto exit;
1959         }
1960         g_free(hw_version);
1961         hw_version = g_strdup("6");
1962     }
1963     if (strcmp(hw_version, "undefined") == 0) {
1964         g_free(hw_version);
1965         hw_version = g_strdup("4");
1966     }
1967     fmt = qemu_opt_get_del(opts, BLOCK_OPT_SUBFMT);
1968     if (qemu_opt_get_bool_del(opts, BLOCK_OPT_ZEROED_GRAIN, false)) {
1969         zeroed_grain = true;
1970     }
1971
1972     if (!adapter_type) {
1973         adapter_type = g_strdup("ide");
1974     } else if (strcmp(adapter_type, "ide") &&
1975                strcmp(adapter_type, "buslogic") &&
1976                strcmp(adapter_type, "lsilogic") &&
1977                strcmp(adapter_type, "legacyESX")) {
1978         error_setg(errp, "Unknown adapter type: '%s'", adapter_type);
1979         ret = -EINVAL;
1980         goto exit;
1981     }
1982     if (strcmp(adapter_type, "ide") != 0) {
1983         /* that's the number of heads with which vmware operates when
1984            creating, exporting, etc. vmdk files with a non-ide adapter type */
1985         number_heads = 255;
1986     }
1987     if (!fmt) {
1988         /* Default format to monolithicSparse */
1989         fmt = g_strdup("monolithicSparse");
1990     } else if (strcmp(fmt, "monolithicFlat") &&
1991                strcmp(fmt, "monolithicSparse") &&
1992                strcmp(fmt, "twoGbMaxExtentSparse") &&
1993                strcmp(fmt, "twoGbMaxExtentFlat") &&
1994                strcmp(fmt, "streamOptimized")) {
1995         error_setg(errp, "Unknown subformat: '%s'", fmt);
1996         ret = -EINVAL;
1997         goto exit;
1998     }
1999     split = !(strcmp(fmt, "twoGbMaxExtentFlat") &&
2000               strcmp(fmt, "twoGbMaxExtentSparse"));
2001     flat = !(strcmp(fmt, "monolithicFlat") &&
2002              strcmp(fmt, "twoGbMaxExtentFlat"));
2003     compress = !strcmp(fmt, "streamOptimized");
2004     if (flat) {
2005         desc_extent_line = "RW %" PRId64 " FLAT \"%s\" 0\n";
2006     } else {
2007         desc_extent_line = "RW %" PRId64 " SPARSE \"%s\"\n";
2008     }
2009     if (flat && backing_file) {
2010         error_setg(errp, "Flat image can't have backing file");
2011         ret = -ENOTSUP;
2012         goto exit;
2013     }
2014     if (flat && zeroed_grain) {
2015         error_setg(errp, "Flat image can't enable zeroed grain");
2016         ret = -ENOTSUP;
2017         goto exit;
2018     }
2019     if (backing_file) {
2020         BlockBackend *blk;
2021         char *full_backing = g_new0(char, PATH_MAX);
2022         bdrv_get_full_backing_filename_from_filename(filename, backing_file,
2023                                                      full_backing, PATH_MAX,
2024                                                      &local_err);
2025         if (local_err) {
2026             g_free(full_backing);
2027             error_propagate(errp, local_err);
2028             ret = -ENOENT;
2029             goto exit;
2030         }
2031
2032         blk = blk_new_open(full_backing, NULL, NULL,
2033                            BDRV_O_NO_BACKING, errp);
2034         g_free(full_backing);
2035         if (blk == NULL) {
2036             ret = -EIO;
2037             goto exit;
2038         }
2039         if (strcmp(blk_bs(blk)->drv->format_name, "vmdk")) {
2040             blk_unref(blk);
2041             ret = -EINVAL;
2042             goto exit;
2043         }
2044         ret = vmdk_read_cid(blk_bs(blk), 0, &parent_cid);
2045         blk_unref(blk);
2046         if (ret) {
2047             goto exit;
2048         }
2049         snprintf(parent_desc_line, BUF_SIZE,
2050                 "parentFileNameHint=\"%s\"", backing_file);
2051     }
2052
2053     /* Create extents */
2054     filesize = total_size;
2055     while (filesize > 0) {
2056         int64_t size = filesize;
2057
2058         if (split && size > split_size) {
2059             size = split_size;
2060         }
2061         if (split) {
2062             snprintf(desc_filename, PATH_MAX, "%s-%c%03d%s",
2063                     prefix, flat ? 'f' : 's', ++idx, postfix);
2064         } else if (flat) {
2065             snprintf(desc_filename, PATH_MAX, "%s-flat%s", prefix, postfix);
2066         } else {
2067             snprintf(desc_filename, PATH_MAX, "%s%s", prefix, postfix);
2068         }
2069         snprintf(ext_filename, PATH_MAX, "%s%s", path, desc_filename);
2070
2071         if (vmdk_create_extent(ext_filename, size,
2072                                flat, compress, zeroed_grain, opts, errp)) {
2073             ret = -EINVAL;
2074             goto exit;
2075         }
2076         filesize -= size;
2077
2078         /* Format description line */
2079         snprintf(desc_line, BUF_SIZE,
2080                     desc_extent_line, size / BDRV_SECTOR_SIZE, desc_filename);
2081         g_string_append(ext_desc_lines, desc_line);
2082     }
2083     /* generate descriptor file */
2084     desc = g_strdup_printf(desc_template,
2085                            g_random_int(),
2086                            parent_cid,
2087                            fmt,
2088                            parent_desc_line,
2089                            ext_desc_lines->str,
2090                            hw_version,
2091                            total_size /
2092                                (int64_t)(63 * number_heads * BDRV_SECTOR_SIZE),
2093                            number_heads,
2094                            adapter_type);
2095     desc_len = strlen(desc);
2096     /* the descriptor offset = 0x200 */
2097     if (!split && !flat) {
2098         desc_offset = 0x200;
2099     } else {
2100         ret = bdrv_create_file(filename, opts, &local_err);
2101         if (ret < 0) {
2102             error_propagate(errp, local_err);
2103             goto exit;
2104         }
2105     }
2106
2107     new_blk = blk_new_open(filename, NULL, NULL,
2108                            BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL,
2109                            &local_err);
2110     if (new_blk == NULL) {
2111         error_propagate(errp, local_err);
2112         ret = -EIO;
2113         goto exit;
2114     }
2115
2116     blk_set_allow_write_beyond_eof(new_blk, true);
2117
2118     ret = blk_pwrite(new_blk, desc_offset, desc, desc_len, 0);
2119     if (ret < 0) {
2120         error_setg_errno(errp, -ret, "Could not write description");
2121         goto exit;
2122     }
2123     /* bdrv_pwrite write padding zeros to align to sector, we don't need that
2124      * for description file */
2125     if (desc_offset == 0) {
2126         ret = blk_truncate(new_blk, desc_len, PREALLOC_MODE_OFF, errp);
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             ret = -EINVAL;
2256             break;
2257         }
2258         ret = get_cluster_offset(bs, extent, NULL,
2259                                  sector_num << BDRV_SECTOR_BITS,
2260                                  false, &cluster_offset, 0, 0);
2261         if (ret == VMDK_ERROR) {
2262             fprintf(stderr,
2263                     "ERROR: could not get cluster_offset for sector %"
2264                     PRId64 "\n", sector_num);
2265             break;
2266         }
2267         if (ret == VMDK_OK) {
2268             int64_t extent_len = bdrv_getlength(extent->file->bs);
2269             if (extent_len < 0) {
2270                 fprintf(stderr,
2271                         "ERROR: could not get extent file length for sector %"
2272                         PRId64 "\n", sector_num);
2273                 ret = extent_len;
2274                 break;
2275             }
2276             if (cluster_offset >= extent_len) {
2277                 fprintf(stderr,
2278                         "ERROR: cluster offset for sector %"
2279                         PRId64 " points after EOF\n", sector_num);
2280                 ret = -EINVAL;
2281                 break;
2282             }
2283         }
2284         sector_num += extent->cluster_sectors;
2285     }
2286
2287     result->corruptions++;
2288     return ret;
2289 }
2290
2291 static ImageInfoSpecific *vmdk_get_specific_info(BlockDriverState *bs)
2292 {
2293     int i;
2294     BDRVVmdkState *s = bs->opaque;
2295     ImageInfoSpecific *spec_info = g_new0(ImageInfoSpecific, 1);
2296     ImageInfoList **next;
2297
2298     *spec_info = (ImageInfoSpecific){
2299         .type = IMAGE_INFO_SPECIFIC_KIND_VMDK,
2300         .u = {
2301             .vmdk.data = g_new0(ImageInfoSpecificVmdk, 1),
2302         },
2303     };
2304
2305     *spec_info->u.vmdk.data = (ImageInfoSpecificVmdk) {
2306         .create_type = g_strdup(s->create_type),
2307         .cid = s->cid,
2308         .parent_cid = s->parent_cid,
2309     };
2310
2311     next = &spec_info->u.vmdk.data->extents;
2312     for (i = 0; i < s->num_extents; i++) {
2313         *next = g_new0(ImageInfoList, 1);
2314         (*next)->value = vmdk_get_extent_info(&s->extents[i]);
2315         (*next)->next = NULL;
2316         next = &(*next)->next;
2317     }
2318
2319     return spec_info;
2320 }
2321
2322 static bool vmdk_extents_type_eq(const VmdkExtent *a, const VmdkExtent *b)
2323 {
2324     return a->flat == b->flat &&
2325            a->compressed == b->compressed &&
2326            (a->flat || a->cluster_sectors == b->cluster_sectors);
2327 }
2328
2329 static int vmdk_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2330 {
2331     int i;
2332     BDRVVmdkState *s = bs->opaque;
2333     assert(s->num_extents);
2334
2335     /* See if we have multiple extents but they have different cases */
2336     for (i = 1; i < s->num_extents; i++) {
2337         if (!vmdk_extents_type_eq(&s->extents[0], &s->extents[i])) {
2338             return -ENOTSUP;
2339         }
2340     }
2341     bdi->needs_compressed_writes = s->extents[0].compressed;
2342     if (!s->extents[0].flat) {
2343         bdi->cluster_size = s->extents[0].cluster_sectors << BDRV_SECTOR_BITS;
2344     }
2345     return 0;
2346 }
2347
2348 static QemuOptsList vmdk_create_opts = {
2349     .name = "vmdk-create-opts",
2350     .head = QTAILQ_HEAD_INITIALIZER(vmdk_create_opts.head),
2351     .desc = {
2352         {
2353             .name = BLOCK_OPT_SIZE,
2354             .type = QEMU_OPT_SIZE,
2355             .help = "Virtual disk size"
2356         },
2357         {
2358             .name = BLOCK_OPT_ADAPTER_TYPE,
2359             .type = QEMU_OPT_STRING,
2360             .help = "Virtual adapter type, can be one of "
2361                     "ide (default), lsilogic, buslogic or legacyESX"
2362         },
2363         {
2364             .name = BLOCK_OPT_BACKING_FILE,
2365             .type = QEMU_OPT_STRING,
2366             .help = "File name of a base image"
2367         },
2368         {
2369             .name = BLOCK_OPT_COMPAT6,
2370             .type = QEMU_OPT_BOOL,
2371             .help = "VMDK version 6 image",
2372             .def_value_str = "off"
2373         },
2374         {
2375             .name = BLOCK_OPT_HWVERSION,
2376             .type = QEMU_OPT_STRING,
2377             .help = "VMDK hardware version",
2378             .def_value_str = "undefined"
2379         },
2380         {
2381             .name = BLOCK_OPT_SUBFMT,
2382             .type = QEMU_OPT_STRING,
2383             .help =
2384                 "VMDK flat extent format, can be one of "
2385                 "{monolithicSparse (default) | monolithicFlat | twoGbMaxExtentSparse | twoGbMaxExtentFlat | streamOptimized} "
2386         },
2387         {
2388             .name = BLOCK_OPT_ZEROED_GRAIN,
2389             .type = QEMU_OPT_BOOL,
2390             .help = "Enable efficient zero writes "
2391                     "using the zeroed-grain GTE feature"
2392         },
2393         { /* end of list */ }
2394     }
2395 };
2396
2397 static BlockDriver bdrv_vmdk = {
2398     .format_name                  = "vmdk",
2399     .instance_size                = sizeof(BDRVVmdkState),
2400     .bdrv_probe                   = vmdk_probe,
2401     .bdrv_open                    = vmdk_open,
2402     .bdrv_check                   = vmdk_check,
2403     .bdrv_reopen_prepare          = vmdk_reopen_prepare,
2404     .bdrv_child_perm              = bdrv_format_default_perms,
2405     .bdrv_co_preadv               = vmdk_co_preadv,
2406     .bdrv_co_pwritev              = vmdk_co_pwritev,
2407     .bdrv_co_pwritev_compressed   = vmdk_co_pwritev_compressed,
2408     .bdrv_co_pwrite_zeroes        = vmdk_co_pwrite_zeroes,
2409     .bdrv_close                   = vmdk_close,
2410     .bdrv_create                  = vmdk_create,
2411     .bdrv_co_flush_to_disk        = vmdk_co_flush,
2412     .bdrv_co_get_block_status     = vmdk_co_get_block_status,
2413     .bdrv_get_allocated_file_size = vmdk_get_allocated_file_size,
2414     .bdrv_has_zero_init           = vmdk_has_zero_init,
2415     .bdrv_get_specific_info       = vmdk_get_specific_info,
2416     .bdrv_refresh_limits          = vmdk_refresh_limits,
2417     .bdrv_get_info                = vmdk_get_info,
2418
2419     .supports_backing             = true,
2420     .create_opts                  = &vmdk_create_opts,
2421 };
2422
2423 static void bdrv_vmdk_init(void)
2424 {
2425     bdrv_register(&bdrv_vmdk);
2426 }
2427
2428 block_init(bdrv_vmdk_init);