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             ret = bdrv_pread(bs->backing, offset, whole_grain,
1079                              skip_start_bytes);
1080             if (ret < 0) {
1081                 ret = VMDK_ERROR;
1082                 goto exit;
1083             }
1084         }
1085         ret = bdrv_pwrite(extent->file, cluster_offset, whole_grain,
1086                           skip_start_bytes);
1087         if (ret < 0) {
1088             ret = VMDK_ERROR;
1089             goto exit;
1090         }
1091     }
1092     /* Read backing data after skip range */
1093     if (skip_end_bytes < cluster_bytes) {
1094         if (bs->backing) {
1095             ret = bdrv_pread(bs->backing, offset + skip_end_bytes,
1096                              whole_grain + skip_end_bytes,
1097                              cluster_bytes - skip_end_bytes);
1098             if (ret < 0) {
1099                 ret = VMDK_ERROR;
1100                 goto exit;
1101             }
1102         }
1103         ret = bdrv_pwrite(extent->file, cluster_offset + skip_end_bytes,
1104                           whole_grain + skip_end_bytes,
1105                           cluster_bytes - skip_end_bytes);
1106         if (ret < 0) {
1107             ret = VMDK_ERROR;
1108             goto exit;
1109         }
1110     }
1111
1112     ret = VMDK_OK;
1113 exit:
1114     qemu_vfree(whole_grain);
1115     return ret;
1116 }
1117
1118 static int vmdk_L2update(VmdkExtent *extent, VmdkMetaData *m_data,
1119                          uint32_t offset)
1120 {
1121     offset = cpu_to_le32(offset);
1122     /* update L2 table */
1123     if (bdrv_pwrite_sync(extent->file,
1124                 ((int64_t)m_data->l2_offset * 512)
1125                     + (m_data->l2_index * sizeof(offset)),
1126                 &offset, sizeof(offset)) < 0) {
1127         return VMDK_ERROR;
1128     }
1129     /* update backup L2 table */
1130     if (extent->l1_backup_table_offset != 0) {
1131         m_data->l2_offset = extent->l1_backup_table[m_data->l1_index];
1132         if (bdrv_pwrite_sync(extent->file,
1133                     ((int64_t)m_data->l2_offset * 512)
1134                         + (m_data->l2_index * sizeof(offset)),
1135                     &offset, sizeof(offset)) < 0) {
1136             return VMDK_ERROR;
1137         }
1138     }
1139     if (m_data->l2_cache_entry) {
1140         *m_data->l2_cache_entry = offset;
1141     }
1142
1143     return VMDK_OK;
1144 }
1145
1146 /**
1147  * get_cluster_offset
1148  *
1149  * Look up cluster offset in extent file by sector number, and store in
1150  * @cluster_offset.
1151  *
1152  * For flat extents, the start offset as parsed from the description file is
1153  * returned.
1154  *
1155  * For sparse extents, look up in L1, L2 table. If allocate is true, return an
1156  * offset for a new cluster and update L2 cache. If there is a backing file,
1157  * COW is done before returning; otherwise, zeroes are written to the allocated
1158  * cluster. Both COW and zero writing skips the sector range
1159  * [@skip_start_sector, @skip_end_sector) passed in by caller, because caller
1160  * has new data to write there.
1161  *
1162  * Returns: VMDK_OK if cluster exists and mapped in the image.
1163  *          VMDK_UNALLOC if cluster is not mapped and @allocate is false.
1164  *          VMDK_ERROR if failed.
1165  */
1166 static int get_cluster_offset(BlockDriverState *bs,
1167                               VmdkExtent *extent,
1168                               VmdkMetaData *m_data,
1169                               uint64_t offset,
1170                               bool allocate,
1171                               uint64_t *cluster_offset,
1172                               uint64_t skip_start_bytes,
1173                               uint64_t skip_end_bytes)
1174 {
1175     unsigned int l1_index, l2_offset, l2_index;
1176     int min_index, i, j;
1177     uint32_t min_count, *l2_table;
1178     bool zeroed = false;
1179     int64_t ret;
1180     int64_t cluster_sector;
1181
1182     if (m_data) {
1183         m_data->valid = 0;
1184     }
1185     if (extent->flat) {
1186         *cluster_offset = extent->flat_start_offset;
1187         return VMDK_OK;
1188     }
1189
1190     offset -= (extent->end_sector - extent->sectors) * SECTOR_SIZE;
1191     l1_index = (offset >> 9) / extent->l1_entry_sectors;
1192     if (l1_index >= extent->l1_size) {
1193         return VMDK_ERROR;
1194     }
1195     l2_offset = extent->l1_table[l1_index];
1196     if (!l2_offset) {
1197         return VMDK_UNALLOC;
1198     }
1199     for (i = 0; i < L2_CACHE_SIZE; i++) {
1200         if (l2_offset == extent->l2_cache_offsets[i]) {
1201             /* increment the hit count */
1202             if (++extent->l2_cache_counts[i] == 0xffffffff) {
1203                 for (j = 0; j < L2_CACHE_SIZE; j++) {
1204                     extent->l2_cache_counts[j] >>= 1;
1205                 }
1206             }
1207             l2_table = extent->l2_cache + (i * extent->l2_size);
1208             goto found;
1209         }
1210     }
1211     /* not found: load a new entry in the least used one */
1212     min_index = 0;
1213     min_count = 0xffffffff;
1214     for (i = 0; i < L2_CACHE_SIZE; i++) {
1215         if (extent->l2_cache_counts[i] < min_count) {
1216             min_count = extent->l2_cache_counts[i];
1217             min_index = i;
1218         }
1219     }
1220     l2_table = extent->l2_cache + (min_index * extent->l2_size);
1221     if (bdrv_pread(extent->file,
1222                 (int64_t)l2_offset * 512,
1223                 l2_table,
1224                 extent->l2_size * sizeof(uint32_t)
1225             ) != extent->l2_size * sizeof(uint32_t)) {
1226         return VMDK_ERROR;
1227     }
1228
1229     extent->l2_cache_offsets[min_index] = l2_offset;
1230     extent->l2_cache_counts[min_index] = 1;
1231  found:
1232     l2_index = ((offset >> 9) / extent->cluster_sectors) % extent->l2_size;
1233     cluster_sector = le32_to_cpu(l2_table[l2_index]);
1234
1235     if (extent->has_zero_grain && cluster_sector == VMDK_GTE_ZEROED) {
1236         zeroed = true;
1237     }
1238
1239     if (!cluster_sector || zeroed) {
1240         if (!allocate) {
1241             return zeroed ? VMDK_ZEROED : VMDK_UNALLOC;
1242         }
1243
1244         cluster_sector = extent->next_cluster_sector;
1245         extent->next_cluster_sector += extent->cluster_sectors;
1246
1247         /* First of all we write grain itself, to avoid race condition
1248          * that may to corrupt the image.
1249          * This problem may occur because of insufficient space on host disk
1250          * or inappropriate VM shutdown.
1251          */
1252         ret = get_whole_cluster(bs, extent, cluster_sector * BDRV_SECTOR_SIZE,
1253                                 offset, skip_start_bytes, skip_end_bytes);
1254         if (ret) {
1255             return ret;
1256         }
1257         if (m_data) {
1258             m_data->valid = 1;
1259             m_data->l1_index = l1_index;
1260             m_data->l2_index = l2_index;
1261             m_data->l2_offset = l2_offset;
1262             m_data->l2_cache_entry = &l2_table[l2_index];
1263         }
1264     }
1265     *cluster_offset = cluster_sector << BDRV_SECTOR_BITS;
1266     return VMDK_OK;
1267 }
1268
1269 static VmdkExtent *find_extent(BDRVVmdkState *s,
1270                                 int64_t sector_num, VmdkExtent *start_hint)
1271 {
1272     VmdkExtent *extent = start_hint;
1273
1274     if (!extent) {
1275         extent = &s->extents[0];
1276     }
1277     while (extent < &s->extents[s->num_extents]) {
1278         if (sector_num < extent->end_sector) {
1279             return extent;
1280         }
1281         extent++;
1282     }
1283     return NULL;
1284 }
1285
1286 static inline uint64_t vmdk_find_offset_in_cluster(VmdkExtent *extent,
1287                                                    int64_t offset)
1288 {
1289     uint64_t extent_begin_offset, extent_relative_offset;
1290     uint64_t cluster_size = extent->cluster_sectors * BDRV_SECTOR_SIZE;
1291
1292     extent_begin_offset =
1293         (extent->end_sector - extent->sectors) * BDRV_SECTOR_SIZE;
1294     extent_relative_offset = offset - extent_begin_offset;
1295     return extent_relative_offset % cluster_size;
1296 }
1297
1298 static inline uint64_t vmdk_find_index_in_cluster(VmdkExtent *extent,
1299                                                   int64_t sector_num)
1300 {
1301     uint64_t offset;
1302     offset = vmdk_find_offset_in_cluster(extent, sector_num * BDRV_SECTOR_SIZE);
1303     return offset / BDRV_SECTOR_SIZE;
1304 }
1305
1306 static int64_t coroutine_fn vmdk_co_get_block_status(BlockDriverState *bs,
1307         int64_t sector_num, int nb_sectors, int *pnum, BlockDriverState **file)
1308 {
1309     BDRVVmdkState *s = bs->opaque;
1310     int64_t index_in_cluster, n, ret;
1311     uint64_t offset;
1312     VmdkExtent *extent;
1313
1314     extent = find_extent(s, sector_num, NULL);
1315     if (!extent) {
1316         return 0;
1317     }
1318     qemu_co_mutex_lock(&s->lock);
1319     ret = get_cluster_offset(bs, extent, NULL,
1320                              sector_num * 512, false, &offset,
1321                              0, 0);
1322     qemu_co_mutex_unlock(&s->lock);
1323
1324     index_in_cluster = vmdk_find_index_in_cluster(extent, sector_num);
1325     switch (ret) {
1326     case VMDK_ERROR:
1327         ret = -EIO;
1328         break;
1329     case VMDK_UNALLOC:
1330         ret = 0;
1331         break;
1332     case VMDK_ZEROED:
1333         ret = BDRV_BLOCK_ZERO;
1334         break;
1335     case VMDK_OK:
1336         ret = BDRV_BLOCK_DATA;
1337         if (!extent->compressed) {
1338             ret |= BDRV_BLOCK_OFFSET_VALID;
1339             ret |= (offset + (index_in_cluster << BDRV_SECTOR_BITS))
1340                     & BDRV_BLOCK_OFFSET_MASK;
1341         }
1342         *file = extent->file->bs;
1343         break;
1344     }
1345
1346     n = extent->cluster_sectors - index_in_cluster;
1347     if (n > nb_sectors) {
1348         n = nb_sectors;
1349     }
1350     *pnum = n;
1351     return ret;
1352 }
1353
1354 static int vmdk_write_extent(VmdkExtent *extent, int64_t cluster_offset,
1355                             int64_t offset_in_cluster, QEMUIOVector *qiov,
1356                             uint64_t qiov_offset, uint64_t n_bytes,
1357                             uint64_t offset)
1358 {
1359     int ret;
1360     VmdkGrainMarker *data = NULL;
1361     uLongf buf_len;
1362     QEMUIOVector local_qiov;
1363     struct iovec iov;
1364     int64_t write_offset;
1365     int64_t write_end_sector;
1366
1367     if (extent->compressed) {
1368         void *compressed_data;
1369
1370         if (!extent->has_marker) {
1371             ret = -EINVAL;
1372             goto out;
1373         }
1374         buf_len = (extent->cluster_sectors << 9) * 2;
1375         data = g_malloc(buf_len + sizeof(VmdkGrainMarker));
1376
1377         compressed_data = g_malloc(n_bytes);
1378         qemu_iovec_to_buf(qiov, qiov_offset, compressed_data, n_bytes);
1379         ret = compress(data->data, &buf_len, compressed_data, n_bytes);
1380         g_free(compressed_data);
1381
1382         if (ret != Z_OK || buf_len == 0) {
1383             ret = -EINVAL;
1384             goto out;
1385         }
1386
1387         data->lba = cpu_to_le64(offset >> BDRV_SECTOR_BITS);
1388         data->size = cpu_to_le32(buf_len);
1389
1390         n_bytes = buf_len + sizeof(VmdkGrainMarker);
1391         iov = (struct iovec) {
1392             .iov_base   = data,
1393             .iov_len    = n_bytes,
1394         };
1395         qemu_iovec_init_external(&local_qiov, &iov, 1);
1396     } else {
1397         qemu_iovec_init(&local_qiov, qiov->niov);
1398         qemu_iovec_concat(&local_qiov, qiov, qiov_offset, n_bytes);
1399     }
1400
1401     write_offset = cluster_offset + offset_in_cluster,
1402     ret = bdrv_co_pwritev(extent->file, write_offset, n_bytes,
1403                           &local_qiov, 0);
1404
1405     write_end_sector = DIV_ROUND_UP(write_offset + n_bytes, BDRV_SECTOR_SIZE);
1406
1407     if (extent->compressed) {
1408         extent->next_cluster_sector = write_end_sector;
1409     } else {
1410         extent->next_cluster_sector = MAX(extent->next_cluster_sector,
1411                                           write_end_sector);
1412     }
1413
1414     if (ret < 0) {
1415         goto out;
1416     }
1417     ret = 0;
1418  out:
1419     g_free(data);
1420     if (!extent->compressed) {
1421         qemu_iovec_destroy(&local_qiov);
1422     }
1423     return ret;
1424 }
1425
1426 static int vmdk_read_extent(VmdkExtent *extent, int64_t cluster_offset,
1427                             int64_t offset_in_cluster, QEMUIOVector *qiov,
1428                             int bytes)
1429 {
1430     int ret;
1431     int cluster_bytes, buf_bytes;
1432     uint8_t *cluster_buf, *compressed_data;
1433     uint8_t *uncomp_buf;
1434     uint32_t data_len;
1435     VmdkGrainMarker *marker;
1436     uLongf buf_len;
1437
1438
1439     if (!extent->compressed) {
1440         ret = bdrv_co_preadv(extent->file,
1441                              cluster_offset + offset_in_cluster, bytes,
1442                              qiov, 0);
1443         if (ret < 0) {
1444             return ret;
1445         }
1446         return 0;
1447     }
1448     cluster_bytes = extent->cluster_sectors * 512;
1449     /* Read two clusters in case GrainMarker + compressed data > one cluster */
1450     buf_bytes = cluster_bytes * 2;
1451     cluster_buf = g_malloc(buf_bytes);
1452     uncomp_buf = g_malloc(cluster_bytes);
1453     ret = bdrv_pread(extent->file,
1454                 cluster_offset,
1455                 cluster_buf, buf_bytes);
1456     if (ret < 0) {
1457         goto out;
1458     }
1459     compressed_data = cluster_buf;
1460     buf_len = cluster_bytes;
1461     data_len = cluster_bytes;
1462     if (extent->has_marker) {
1463         marker = (VmdkGrainMarker *)cluster_buf;
1464         compressed_data = marker->data;
1465         data_len = le32_to_cpu(marker->size);
1466     }
1467     if (!data_len || data_len > buf_bytes) {
1468         ret = -EINVAL;
1469         goto out;
1470     }
1471     ret = uncompress(uncomp_buf, &buf_len, compressed_data, data_len);
1472     if (ret != Z_OK) {
1473         ret = -EINVAL;
1474         goto out;
1475
1476     }
1477     if (offset_in_cluster < 0 ||
1478             offset_in_cluster + bytes > buf_len) {
1479         ret = -EINVAL;
1480         goto out;
1481     }
1482     qemu_iovec_from_buf(qiov, 0, uncomp_buf + offset_in_cluster, bytes);
1483     ret = 0;
1484
1485  out:
1486     g_free(uncomp_buf);
1487     g_free(cluster_buf);
1488     return ret;
1489 }
1490
1491 static int coroutine_fn
1492 vmdk_co_preadv(BlockDriverState *bs, uint64_t offset, uint64_t bytes,
1493                QEMUIOVector *qiov, int flags)
1494 {
1495     BDRVVmdkState *s = bs->opaque;
1496     int ret;
1497     uint64_t n_bytes, offset_in_cluster;
1498     VmdkExtent *extent = NULL;
1499     QEMUIOVector local_qiov;
1500     uint64_t cluster_offset;
1501     uint64_t bytes_done = 0;
1502
1503     qemu_iovec_init(&local_qiov, qiov->niov);
1504     qemu_co_mutex_lock(&s->lock);
1505
1506     while (bytes > 0) {
1507         extent = find_extent(s, offset >> BDRV_SECTOR_BITS, extent);
1508         if (!extent) {
1509             ret = -EIO;
1510             goto fail;
1511         }
1512         ret = get_cluster_offset(bs, extent, NULL,
1513                                  offset, false, &cluster_offset, 0, 0);
1514         offset_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
1515
1516         n_bytes = MIN(bytes, extent->cluster_sectors * BDRV_SECTOR_SIZE
1517                              - offset_in_cluster);
1518
1519         if (ret != VMDK_OK) {
1520             /* if not allocated, try to read from parent image, if exist */
1521             if (bs->backing && ret != VMDK_ZEROED) {
1522                 if (!vmdk_is_cid_valid(bs)) {
1523                     ret = -EINVAL;
1524                     goto fail;
1525                 }
1526
1527                 qemu_iovec_reset(&local_qiov);
1528                 qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes);
1529
1530                 ret = bdrv_co_preadv(bs->backing, offset, n_bytes,
1531                                      &local_qiov, 0);
1532                 if (ret < 0) {
1533                     goto fail;
1534                 }
1535             } else {
1536                 qemu_iovec_memset(qiov, bytes_done, 0, n_bytes);
1537             }
1538         } else {
1539             qemu_iovec_reset(&local_qiov);
1540             qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes);
1541
1542             ret = vmdk_read_extent(extent, cluster_offset, offset_in_cluster,
1543                                    &local_qiov, n_bytes);
1544             if (ret) {
1545                 goto fail;
1546             }
1547         }
1548         bytes -= n_bytes;
1549         offset += n_bytes;
1550         bytes_done += n_bytes;
1551     }
1552
1553     ret = 0;
1554 fail:
1555     qemu_co_mutex_unlock(&s->lock);
1556     qemu_iovec_destroy(&local_qiov);
1557
1558     return ret;
1559 }
1560
1561 /**
1562  * vmdk_write:
1563  * @zeroed:       buf is ignored (data is zero), use zeroed_grain GTE feature
1564  *                if possible, otherwise return -ENOTSUP.
1565  * @zero_dry_run: used for zeroed == true only, don't update L2 table, just try
1566  *                with each cluster. By dry run we can find if the zero write
1567  *                is possible without modifying image data.
1568  *
1569  * Returns: error code with 0 for success.
1570  */
1571 static int vmdk_pwritev(BlockDriverState *bs, uint64_t offset,
1572                        uint64_t bytes, QEMUIOVector *qiov,
1573                        bool zeroed, bool zero_dry_run)
1574 {
1575     BDRVVmdkState *s = bs->opaque;
1576     VmdkExtent *extent = NULL;
1577     int ret;
1578     int64_t offset_in_cluster, n_bytes;
1579     uint64_t cluster_offset;
1580     uint64_t bytes_done = 0;
1581     VmdkMetaData m_data;
1582
1583     if (DIV_ROUND_UP(offset, BDRV_SECTOR_SIZE) > bs->total_sectors) {
1584         error_report("Wrong offset: offset=0x%" PRIx64
1585                      " total_sectors=0x%" PRIx64,
1586                      offset, bs->total_sectors);
1587         return -EIO;
1588     }
1589
1590     while (bytes > 0) {
1591         extent = find_extent(s, offset >> BDRV_SECTOR_BITS, extent);
1592         if (!extent) {
1593             return -EIO;
1594         }
1595         offset_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
1596         n_bytes = MIN(bytes, extent->cluster_sectors * BDRV_SECTOR_SIZE
1597                              - offset_in_cluster);
1598
1599         ret = get_cluster_offset(bs, extent, &m_data, offset,
1600                                  !(extent->compressed || zeroed),
1601                                  &cluster_offset, offset_in_cluster,
1602                                  offset_in_cluster + n_bytes);
1603         if (extent->compressed) {
1604             if (ret == VMDK_OK) {
1605                 /* Refuse write to allocated cluster for streamOptimized */
1606                 error_report("Could not write to allocated cluster"
1607                               " for streamOptimized");
1608                 return -EIO;
1609             } else {
1610                 /* allocate */
1611                 ret = get_cluster_offset(bs, extent, &m_data, offset,
1612                                          true, &cluster_offset, 0, 0);
1613             }
1614         }
1615         if (ret == VMDK_ERROR) {
1616             return -EINVAL;
1617         }
1618         if (zeroed) {
1619             /* Do zeroed write, buf is ignored */
1620             if (extent->has_zero_grain &&
1621                     offset_in_cluster == 0 &&
1622                     n_bytes >= extent->cluster_sectors * BDRV_SECTOR_SIZE) {
1623                 n_bytes = extent->cluster_sectors * BDRV_SECTOR_SIZE;
1624                 if (!zero_dry_run) {
1625                     /* update L2 tables */
1626                     if (vmdk_L2update(extent, &m_data, VMDK_GTE_ZEROED)
1627                             != VMDK_OK) {
1628                         return -EIO;
1629                     }
1630                 }
1631             } else {
1632                 return -ENOTSUP;
1633             }
1634         } else {
1635             ret = vmdk_write_extent(extent, cluster_offset, offset_in_cluster,
1636                                     qiov, bytes_done, n_bytes, offset);
1637             if (ret) {
1638                 return ret;
1639             }
1640             if (m_data.valid) {
1641                 /* update L2 tables */
1642                 if (vmdk_L2update(extent, &m_data,
1643                                   cluster_offset >> BDRV_SECTOR_BITS)
1644                         != VMDK_OK) {
1645                     return -EIO;
1646                 }
1647             }
1648         }
1649         bytes -= n_bytes;
1650         offset += n_bytes;
1651         bytes_done += n_bytes;
1652
1653         /* update CID on the first write every time the virtual disk is
1654          * opened */
1655         if (!s->cid_updated) {
1656             ret = vmdk_write_cid(bs, g_random_int());
1657             if (ret < 0) {
1658                 return ret;
1659             }
1660             s->cid_updated = true;
1661         }
1662     }
1663     return 0;
1664 }
1665
1666 static int coroutine_fn
1667 vmdk_co_pwritev(BlockDriverState *bs, uint64_t offset, uint64_t bytes,
1668                 QEMUIOVector *qiov, int flags)
1669 {
1670     int ret;
1671     BDRVVmdkState *s = bs->opaque;
1672     qemu_co_mutex_lock(&s->lock);
1673     ret = vmdk_pwritev(bs, offset, bytes, qiov, false, false);
1674     qemu_co_mutex_unlock(&s->lock);
1675     return ret;
1676 }
1677
1678 static int coroutine_fn
1679 vmdk_co_pwritev_compressed(BlockDriverState *bs, uint64_t offset,
1680                            uint64_t bytes, QEMUIOVector *qiov)
1681 {
1682     return vmdk_co_pwritev(bs, offset, bytes, qiov, 0);
1683 }
1684
1685 static int coroutine_fn vmdk_co_pwrite_zeroes(BlockDriverState *bs,
1686                                               int64_t offset,
1687                                               int bytes,
1688                                               BdrvRequestFlags flags)
1689 {
1690     int ret;
1691     BDRVVmdkState *s = bs->opaque;
1692
1693     qemu_co_mutex_lock(&s->lock);
1694     /* write zeroes could fail if sectors not aligned to cluster, test it with
1695      * dry_run == true before really updating image */
1696     ret = vmdk_pwritev(bs, offset, bytes, NULL, true, true);
1697     if (!ret) {
1698         ret = vmdk_pwritev(bs, offset, bytes, NULL, true, false);
1699     }
1700     qemu_co_mutex_unlock(&s->lock);
1701     return ret;
1702 }
1703
1704 static int vmdk_create_extent(const char *filename, int64_t filesize,
1705                               bool flat, bool compress, bool zeroed_grain,
1706                               QemuOpts *opts, Error **errp)
1707 {
1708     int ret, i;
1709     BlockBackend *blk = NULL;
1710     VMDK4Header header;
1711     Error *local_err = NULL;
1712     uint32_t tmp, magic, grains, gd_sectors, gt_size, gt_count;
1713     uint32_t *gd_buf = NULL;
1714     int gd_buf_size;
1715
1716     ret = bdrv_create_file(filename, opts, &local_err);
1717     if (ret < 0) {
1718         error_propagate(errp, local_err);
1719         goto exit;
1720     }
1721
1722     blk = blk_new_open(filename, NULL, NULL,
1723                        BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL,
1724                        &local_err);
1725     if (blk == NULL) {
1726         error_propagate(errp, local_err);
1727         ret = -EIO;
1728         goto exit;
1729     }
1730
1731     blk_set_allow_write_beyond_eof(blk, true);
1732
1733     if (flat) {
1734         ret = blk_truncate(blk, filesize, PREALLOC_MODE_OFF, errp);
1735         goto exit;
1736     }
1737     magic = cpu_to_be32(VMDK4_MAGIC);
1738     memset(&header, 0, sizeof(header));
1739     if (compress) {
1740         header.version = 3;
1741     } else if (zeroed_grain) {
1742         header.version = 2;
1743     } else {
1744         header.version = 1;
1745     }
1746     header.flags = VMDK4_FLAG_RGD | VMDK4_FLAG_NL_DETECT
1747                    | (compress ? VMDK4_FLAG_COMPRESS | VMDK4_FLAG_MARKER : 0)
1748                    | (zeroed_grain ? VMDK4_FLAG_ZERO_GRAIN : 0);
1749     header.compressAlgorithm = compress ? VMDK4_COMPRESSION_DEFLATE : 0;
1750     header.capacity = filesize / BDRV_SECTOR_SIZE;
1751     header.granularity = 128;
1752     header.num_gtes_per_gt = BDRV_SECTOR_SIZE;
1753
1754     grains = DIV_ROUND_UP(filesize / BDRV_SECTOR_SIZE, header.granularity);
1755     gt_size = DIV_ROUND_UP(header.num_gtes_per_gt * sizeof(uint32_t),
1756                            BDRV_SECTOR_SIZE);
1757     gt_count = DIV_ROUND_UP(grains, header.num_gtes_per_gt);
1758     gd_sectors = DIV_ROUND_UP(gt_count * sizeof(uint32_t), BDRV_SECTOR_SIZE);
1759
1760     header.desc_offset = 1;
1761     header.desc_size = 20;
1762     header.rgd_offset = header.desc_offset + header.desc_size;
1763     header.gd_offset = header.rgd_offset + gd_sectors + (gt_size * gt_count);
1764     header.grain_offset =
1765         ROUND_UP(header.gd_offset + gd_sectors + (gt_size * gt_count),
1766                  header.granularity);
1767     /* swap endianness for all header fields */
1768     header.version = cpu_to_le32(header.version);
1769     header.flags = cpu_to_le32(header.flags);
1770     header.capacity = cpu_to_le64(header.capacity);
1771     header.granularity = cpu_to_le64(header.granularity);
1772     header.num_gtes_per_gt = cpu_to_le32(header.num_gtes_per_gt);
1773     header.desc_offset = cpu_to_le64(header.desc_offset);
1774     header.desc_size = cpu_to_le64(header.desc_size);
1775     header.rgd_offset = cpu_to_le64(header.rgd_offset);
1776     header.gd_offset = cpu_to_le64(header.gd_offset);
1777     header.grain_offset = cpu_to_le64(header.grain_offset);
1778     header.compressAlgorithm = cpu_to_le16(header.compressAlgorithm);
1779
1780     header.check_bytes[0] = 0xa;
1781     header.check_bytes[1] = 0x20;
1782     header.check_bytes[2] = 0xd;
1783     header.check_bytes[3] = 0xa;
1784
1785     /* write all the data */
1786     ret = blk_pwrite(blk, 0, &magic, sizeof(magic), 0);
1787     if (ret < 0) {
1788         error_setg(errp, QERR_IO_ERROR);
1789         goto exit;
1790     }
1791     ret = blk_pwrite(blk, sizeof(magic), &header, sizeof(header), 0);
1792     if (ret < 0) {
1793         error_setg(errp, QERR_IO_ERROR);
1794         goto exit;
1795     }
1796
1797     ret = blk_truncate(blk, le64_to_cpu(header.grain_offset) << 9,
1798                        PREALLOC_MODE_OFF, errp);
1799     if (ret < 0) {
1800         goto exit;
1801     }
1802
1803     /* write grain directory */
1804     gd_buf_size = gd_sectors * BDRV_SECTOR_SIZE;
1805     gd_buf = g_malloc0(gd_buf_size);
1806     for (i = 0, tmp = le64_to_cpu(header.rgd_offset) + gd_sectors;
1807          i < gt_count; i++, tmp += gt_size) {
1808         gd_buf[i] = cpu_to_le32(tmp);
1809     }
1810     ret = blk_pwrite(blk, le64_to_cpu(header.rgd_offset) * BDRV_SECTOR_SIZE,
1811                      gd_buf, gd_buf_size, 0);
1812     if (ret < 0) {
1813         error_setg(errp, QERR_IO_ERROR);
1814         goto exit;
1815     }
1816
1817     /* write backup grain directory */
1818     for (i = 0, tmp = le64_to_cpu(header.gd_offset) + gd_sectors;
1819          i < gt_count; i++, tmp += gt_size) {
1820         gd_buf[i] = cpu_to_le32(tmp);
1821     }
1822     ret = blk_pwrite(blk, le64_to_cpu(header.gd_offset) * BDRV_SECTOR_SIZE,
1823                      gd_buf, gd_buf_size, 0);
1824     if (ret < 0) {
1825         error_setg(errp, QERR_IO_ERROR);
1826         goto exit;
1827     }
1828
1829     ret = 0;
1830 exit:
1831     if (blk) {
1832         blk_unref(blk);
1833     }
1834     g_free(gd_buf);
1835     return ret;
1836 }
1837
1838 static int filename_decompose(const char *filename, char *path, char *prefix,
1839                               char *postfix, size_t buf_len, Error **errp)
1840 {
1841     const char *p, *q;
1842
1843     if (filename == NULL || !strlen(filename)) {
1844         error_setg(errp, "No filename provided");
1845         return VMDK_ERROR;
1846     }
1847     p = strrchr(filename, '/');
1848     if (p == NULL) {
1849         p = strrchr(filename, '\\');
1850     }
1851     if (p == NULL) {
1852         p = strrchr(filename, ':');
1853     }
1854     if (p != NULL) {
1855         p++;
1856         if (p - filename >= buf_len) {
1857             return VMDK_ERROR;
1858         }
1859         pstrcpy(path, p - filename + 1, filename);
1860     } else {
1861         p = filename;
1862         path[0] = '\0';
1863     }
1864     q = strrchr(p, '.');
1865     if (q == NULL) {
1866         pstrcpy(prefix, buf_len, p);
1867         postfix[0] = '\0';
1868     } else {
1869         if (q - p >= buf_len) {
1870             return VMDK_ERROR;
1871         }
1872         pstrcpy(prefix, q - p + 1, p);
1873         pstrcpy(postfix, buf_len, q);
1874     }
1875     return VMDK_OK;
1876 }
1877
1878 static int vmdk_create(const char *filename, QemuOpts *opts, Error **errp)
1879 {
1880     int idx = 0;
1881     BlockBackend *new_blk = NULL;
1882     Error *local_err = NULL;
1883     char *desc = NULL;
1884     int64_t total_size = 0, filesize;
1885     char *adapter_type = NULL;
1886     char *backing_file = NULL;
1887     char *hw_version = NULL;
1888     char *fmt = NULL;
1889     int ret = 0;
1890     bool flat, split, compress;
1891     GString *ext_desc_lines;
1892     char *path = g_malloc0(PATH_MAX);
1893     char *prefix = g_malloc0(PATH_MAX);
1894     char *postfix = g_malloc0(PATH_MAX);
1895     char *desc_line = g_malloc0(BUF_SIZE);
1896     char *ext_filename = g_malloc0(PATH_MAX);
1897     char *desc_filename = g_malloc0(PATH_MAX);
1898     const int64_t split_size = 0x80000000;  /* VMDK has constant split size */
1899     const char *desc_extent_line;
1900     char *parent_desc_line = g_malloc0(BUF_SIZE);
1901     uint32_t parent_cid = 0xffffffff;
1902     uint32_t number_heads = 16;
1903     bool zeroed_grain = false;
1904     uint32_t desc_offset = 0, desc_len;
1905     const char desc_template[] =
1906         "# Disk DescriptorFile\n"
1907         "version=1\n"
1908         "CID=%" PRIx32 "\n"
1909         "parentCID=%" PRIx32 "\n"
1910         "createType=\"%s\"\n"
1911         "%s"
1912         "\n"
1913         "# Extent description\n"
1914         "%s"
1915         "\n"
1916         "# The Disk Data Base\n"
1917         "#DDB\n"
1918         "\n"
1919         "ddb.virtualHWVersion = \"%s\"\n"
1920         "ddb.geometry.cylinders = \"%" PRId64 "\"\n"
1921         "ddb.geometry.heads = \"%" PRIu32 "\"\n"
1922         "ddb.geometry.sectors = \"63\"\n"
1923         "ddb.adapterType = \"%s\"\n";
1924
1925     ext_desc_lines = g_string_new(NULL);
1926
1927     if (filename_decompose(filename, path, prefix, postfix, PATH_MAX, errp)) {
1928         ret = -EINVAL;
1929         goto exit;
1930     }
1931     /* Read out options */
1932     total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
1933                           BDRV_SECTOR_SIZE);
1934     adapter_type = qemu_opt_get_del(opts, BLOCK_OPT_ADAPTER_TYPE);
1935     backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
1936     hw_version = qemu_opt_get_del(opts, BLOCK_OPT_HWVERSION);
1937     if (qemu_opt_get_bool_del(opts, BLOCK_OPT_COMPAT6, false)) {
1938         if (strcmp(hw_version, "undefined")) {
1939             error_setg(errp,
1940                        "compat6 cannot be enabled with hwversion set");
1941             ret = -EINVAL;
1942             goto exit;
1943         }
1944         g_free(hw_version);
1945         hw_version = g_strdup("6");
1946     }
1947     if (strcmp(hw_version, "undefined") == 0) {
1948         g_free(hw_version);
1949         hw_version = g_strdup("4");
1950     }
1951     fmt = qemu_opt_get_del(opts, BLOCK_OPT_SUBFMT);
1952     if (qemu_opt_get_bool_del(opts, BLOCK_OPT_ZEROED_GRAIN, false)) {
1953         zeroed_grain = true;
1954     }
1955
1956     if (!adapter_type) {
1957         adapter_type = g_strdup("ide");
1958     } else if (strcmp(adapter_type, "ide") &&
1959                strcmp(adapter_type, "buslogic") &&
1960                strcmp(adapter_type, "lsilogic") &&
1961                strcmp(adapter_type, "legacyESX")) {
1962         error_setg(errp, "Unknown adapter type: '%s'", adapter_type);
1963         ret = -EINVAL;
1964         goto exit;
1965     }
1966     if (strcmp(adapter_type, "ide") != 0) {
1967         /* that's the number of heads with which vmware operates when
1968            creating, exporting, etc. vmdk files with a non-ide adapter type */
1969         number_heads = 255;
1970     }
1971     if (!fmt) {
1972         /* Default format to monolithicSparse */
1973         fmt = g_strdup("monolithicSparse");
1974     } else if (strcmp(fmt, "monolithicFlat") &&
1975                strcmp(fmt, "monolithicSparse") &&
1976                strcmp(fmt, "twoGbMaxExtentSparse") &&
1977                strcmp(fmt, "twoGbMaxExtentFlat") &&
1978                strcmp(fmt, "streamOptimized")) {
1979         error_setg(errp, "Unknown subformat: '%s'", fmt);
1980         ret = -EINVAL;
1981         goto exit;
1982     }
1983     split = !(strcmp(fmt, "twoGbMaxExtentFlat") &&
1984               strcmp(fmt, "twoGbMaxExtentSparse"));
1985     flat = !(strcmp(fmt, "monolithicFlat") &&
1986              strcmp(fmt, "twoGbMaxExtentFlat"));
1987     compress = !strcmp(fmt, "streamOptimized");
1988     if (flat) {
1989         desc_extent_line = "RW %" PRId64 " FLAT \"%s\" 0\n";
1990     } else {
1991         desc_extent_line = "RW %" PRId64 " SPARSE \"%s\"\n";
1992     }
1993     if (flat && backing_file) {
1994         error_setg(errp, "Flat image can't have backing file");
1995         ret = -ENOTSUP;
1996         goto exit;
1997     }
1998     if (flat && zeroed_grain) {
1999         error_setg(errp, "Flat image can't enable zeroed grain");
2000         ret = -ENOTSUP;
2001         goto exit;
2002     }
2003     if (backing_file) {
2004         BlockBackend *blk;
2005         char *full_backing = g_new0(char, PATH_MAX);
2006         bdrv_get_full_backing_filename_from_filename(filename, backing_file,
2007                                                      full_backing, PATH_MAX,
2008                                                      &local_err);
2009         if (local_err) {
2010             g_free(full_backing);
2011             error_propagate(errp, local_err);
2012             ret = -ENOENT;
2013             goto exit;
2014         }
2015
2016         blk = blk_new_open(full_backing, NULL, NULL,
2017                            BDRV_O_NO_BACKING, errp);
2018         g_free(full_backing);
2019         if (blk == NULL) {
2020             ret = -EIO;
2021             goto exit;
2022         }
2023         if (strcmp(blk_bs(blk)->drv->format_name, "vmdk")) {
2024             blk_unref(blk);
2025             ret = -EINVAL;
2026             goto exit;
2027         }
2028         ret = vmdk_read_cid(blk_bs(blk), 0, &parent_cid);
2029         blk_unref(blk);
2030         if (ret) {
2031             goto exit;
2032         }
2033         snprintf(parent_desc_line, BUF_SIZE,
2034                 "parentFileNameHint=\"%s\"", backing_file);
2035     }
2036
2037     /* Create extents */
2038     filesize = total_size;
2039     while (filesize > 0) {
2040         int64_t size = filesize;
2041
2042         if (split && size > split_size) {
2043             size = split_size;
2044         }
2045         if (split) {
2046             snprintf(desc_filename, PATH_MAX, "%s-%c%03d%s",
2047                     prefix, flat ? 'f' : 's', ++idx, postfix);
2048         } else if (flat) {
2049             snprintf(desc_filename, PATH_MAX, "%s-flat%s", prefix, postfix);
2050         } else {
2051             snprintf(desc_filename, PATH_MAX, "%s%s", prefix, postfix);
2052         }
2053         snprintf(ext_filename, PATH_MAX, "%s%s", path, desc_filename);
2054
2055         if (vmdk_create_extent(ext_filename, size,
2056                                flat, compress, zeroed_grain, opts, errp)) {
2057             ret = -EINVAL;
2058             goto exit;
2059         }
2060         filesize -= size;
2061
2062         /* Format description line */
2063         snprintf(desc_line, BUF_SIZE,
2064                     desc_extent_line, size / BDRV_SECTOR_SIZE, desc_filename);
2065         g_string_append(ext_desc_lines, desc_line);
2066     }
2067     /* generate descriptor file */
2068     desc = g_strdup_printf(desc_template,
2069                            g_random_int(),
2070                            parent_cid,
2071                            fmt,
2072                            parent_desc_line,
2073                            ext_desc_lines->str,
2074                            hw_version,
2075                            total_size /
2076                                (int64_t)(63 * number_heads * BDRV_SECTOR_SIZE),
2077                            number_heads,
2078                            adapter_type);
2079     desc_len = strlen(desc);
2080     /* the descriptor offset = 0x200 */
2081     if (!split && !flat) {
2082         desc_offset = 0x200;
2083     } else {
2084         ret = bdrv_create_file(filename, opts, &local_err);
2085         if (ret < 0) {
2086             error_propagate(errp, local_err);
2087             goto exit;
2088         }
2089     }
2090
2091     new_blk = blk_new_open(filename, NULL, NULL,
2092                            BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL,
2093                            &local_err);
2094     if (new_blk == NULL) {
2095         error_propagate(errp, local_err);
2096         ret = -EIO;
2097         goto exit;
2098     }
2099
2100     blk_set_allow_write_beyond_eof(new_blk, true);
2101
2102     ret = blk_pwrite(new_blk, desc_offset, desc, desc_len, 0);
2103     if (ret < 0) {
2104         error_setg_errno(errp, -ret, "Could not write description");
2105         goto exit;
2106     }
2107     /* bdrv_pwrite write padding zeros to align to sector, we don't need that
2108      * for description file */
2109     if (desc_offset == 0) {
2110         ret = blk_truncate(new_blk, desc_len, PREALLOC_MODE_OFF, errp);
2111     }
2112 exit:
2113     if (new_blk) {
2114         blk_unref(new_blk);
2115     }
2116     g_free(adapter_type);
2117     g_free(backing_file);
2118     g_free(hw_version);
2119     g_free(fmt);
2120     g_free(desc);
2121     g_free(path);
2122     g_free(prefix);
2123     g_free(postfix);
2124     g_free(desc_line);
2125     g_free(ext_filename);
2126     g_free(desc_filename);
2127     g_free(parent_desc_line);
2128     g_string_free(ext_desc_lines, true);
2129     return ret;
2130 }
2131
2132 static void vmdk_close(BlockDriverState *bs)
2133 {
2134     BDRVVmdkState *s = bs->opaque;
2135
2136     vmdk_free_extents(bs);
2137     g_free(s->create_type);
2138
2139     migrate_del_blocker(s->migration_blocker);
2140     error_free(s->migration_blocker);
2141 }
2142
2143 static coroutine_fn int vmdk_co_flush(BlockDriverState *bs)
2144 {
2145     BDRVVmdkState *s = bs->opaque;
2146     int i, err;
2147     int ret = 0;
2148
2149     for (i = 0; i < s->num_extents; i++) {
2150         err = bdrv_co_flush(s->extents[i].file->bs);
2151         if (err < 0) {
2152             ret = err;
2153         }
2154     }
2155     return ret;
2156 }
2157
2158 static int64_t vmdk_get_allocated_file_size(BlockDriverState *bs)
2159 {
2160     int i;
2161     int64_t ret = 0;
2162     int64_t r;
2163     BDRVVmdkState *s = bs->opaque;
2164
2165     ret = bdrv_get_allocated_file_size(bs->file->bs);
2166     if (ret < 0) {
2167         return ret;
2168     }
2169     for (i = 0; i < s->num_extents; i++) {
2170         if (s->extents[i].file == bs->file) {
2171             continue;
2172         }
2173         r = bdrv_get_allocated_file_size(s->extents[i].file->bs);
2174         if (r < 0) {
2175             return r;
2176         }
2177         ret += r;
2178     }
2179     return ret;
2180 }
2181
2182 static int vmdk_has_zero_init(BlockDriverState *bs)
2183 {
2184     int i;
2185     BDRVVmdkState *s = bs->opaque;
2186
2187     /* If has a flat extent and its underlying storage doesn't have zero init,
2188      * return 0. */
2189     for (i = 0; i < s->num_extents; i++) {
2190         if (s->extents[i].flat) {
2191             if (!bdrv_has_zero_init(s->extents[i].file->bs)) {
2192                 return 0;
2193             }
2194         }
2195     }
2196     return 1;
2197 }
2198
2199 static ImageInfo *vmdk_get_extent_info(VmdkExtent *extent)
2200 {
2201     ImageInfo *info = g_new0(ImageInfo, 1);
2202
2203     *info = (ImageInfo){
2204         .filename         = g_strdup(extent->file->bs->filename),
2205         .format           = g_strdup(extent->type),
2206         .virtual_size     = extent->sectors * BDRV_SECTOR_SIZE,
2207         .compressed       = extent->compressed,
2208         .has_compressed   = extent->compressed,
2209         .cluster_size     = extent->cluster_sectors * BDRV_SECTOR_SIZE,
2210         .has_cluster_size = !extent->flat,
2211     };
2212
2213     return info;
2214 }
2215
2216 static int vmdk_check(BlockDriverState *bs, BdrvCheckResult *result,
2217                       BdrvCheckMode fix)
2218 {
2219     BDRVVmdkState *s = bs->opaque;
2220     VmdkExtent *extent = NULL;
2221     int64_t sector_num = 0;
2222     int64_t total_sectors = bdrv_nb_sectors(bs);
2223     int ret;
2224     uint64_t cluster_offset;
2225
2226     if (fix) {
2227         return -ENOTSUP;
2228     }
2229
2230     for (;;) {
2231         if (sector_num >= total_sectors) {
2232             return 0;
2233         }
2234         extent = find_extent(s, sector_num, extent);
2235         if (!extent) {
2236             fprintf(stderr,
2237                     "ERROR: could not find extent for sector %" PRId64 "\n",
2238                     sector_num);
2239             ret = -EINVAL;
2240             break;
2241         }
2242         ret = get_cluster_offset(bs, extent, NULL,
2243                                  sector_num << BDRV_SECTOR_BITS,
2244                                  false, &cluster_offset, 0, 0);
2245         if (ret == VMDK_ERROR) {
2246             fprintf(stderr,
2247                     "ERROR: could not get cluster_offset for sector %"
2248                     PRId64 "\n", sector_num);
2249             break;
2250         }
2251         if (ret == VMDK_OK) {
2252             int64_t extent_len = bdrv_getlength(extent->file->bs);
2253             if (extent_len < 0) {
2254                 fprintf(stderr,
2255                         "ERROR: could not get extent file length for sector %"
2256                         PRId64 "\n", sector_num);
2257                 ret = extent_len;
2258                 break;
2259             }
2260             if (cluster_offset >= extent_len) {
2261                 fprintf(stderr,
2262                         "ERROR: cluster offset for sector %"
2263                         PRId64 " points after EOF\n", sector_num);
2264                 ret = -EINVAL;
2265                 break;
2266             }
2267         }
2268         sector_num += extent->cluster_sectors;
2269     }
2270
2271     result->corruptions++;
2272     return ret;
2273 }
2274
2275 static ImageInfoSpecific *vmdk_get_specific_info(BlockDriverState *bs)
2276 {
2277     int i;
2278     BDRVVmdkState *s = bs->opaque;
2279     ImageInfoSpecific *spec_info = g_new0(ImageInfoSpecific, 1);
2280     ImageInfoList **next;
2281
2282     *spec_info = (ImageInfoSpecific){
2283         .type = IMAGE_INFO_SPECIFIC_KIND_VMDK,
2284         .u = {
2285             .vmdk.data = g_new0(ImageInfoSpecificVmdk, 1),
2286         },
2287     };
2288
2289     *spec_info->u.vmdk.data = (ImageInfoSpecificVmdk) {
2290         .create_type = g_strdup(s->create_type),
2291         .cid = s->cid,
2292         .parent_cid = s->parent_cid,
2293     };
2294
2295     next = &spec_info->u.vmdk.data->extents;
2296     for (i = 0; i < s->num_extents; i++) {
2297         *next = g_new0(ImageInfoList, 1);
2298         (*next)->value = vmdk_get_extent_info(&s->extents[i]);
2299         (*next)->next = NULL;
2300         next = &(*next)->next;
2301     }
2302
2303     return spec_info;
2304 }
2305
2306 static bool vmdk_extents_type_eq(const VmdkExtent *a, const VmdkExtent *b)
2307 {
2308     return a->flat == b->flat &&
2309            a->compressed == b->compressed &&
2310            (a->flat || a->cluster_sectors == b->cluster_sectors);
2311 }
2312
2313 static int vmdk_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2314 {
2315     int i;
2316     BDRVVmdkState *s = bs->opaque;
2317     assert(s->num_extents);
2318
2319     /* See if we have multiple extents but they have different cases */
2320     for (i = 1; i < s->num_extents; i++) {
2321         if (!vmdk_extents_type_eq(&s->extents[0], &s->extents[i])) {
2322             return -ENOTSUP;
2323         }
2324     }
2325     bdi->needs_compressed_writes = s->extents[0].compressed;
2326     if (!s->extents[0].flat) {
2327         bdi->cluster_size = s->extents[0].cluster_sectors << BDRV_SECTOR_BITS;
2328     }
2329     return 0;
2330 }
2331
2332 static QemuOptsList vmdk_create_opts = {
2333     .name = "vmdk-create-opts",
2334     .head = QTAILQ_HEAD_INITIALIZER(vmdk_create_opts.head),
2335     .desc = {
2336         {
2337             .name = BLOCK_OPT_SIZE,
2338             .type = QEMU_OPT_SIZE,
2339             .help = "Virtual disk size"
2340         },
2341         {
2342             .name = BLOCK_OPT_ADAPTER_TYPE,
2343             .type = QEMU_OPT_STRING,
2344             .help = "Virtual adapter type, can be one of "
2345                     "ide (default), lsilogic, buslogic or legacyESX"
2346         },
2347         {
2348             .name = BLOCK_OPT_BACKING_FILE,
2349             .type = QEMU_OPT_STRING,
2350             .help = "File name of a base image"
2351         },
2352         {
2353             .name = BLOCK_OPT_COMPAT6,
2354             .type = QEMU_OPT_BOOL,
2355             .help = "VMDK version 6 image",
2356             .def_value_str = "off"
2357         },
2358         {
2359             .name = BLOCK_OPT_HWVERSION,
2360             .type = QEMU_OPT_STRING,
2361             .help = "VMDK hardware version",
2362             .def_value_str = "undefined"
2363         },
2364         {
2365             .name = BLOCK_OPT_SUBFMT,
2366             .type = QEMU_OPT_STRING,
2367             .help =
2368                 "VMDK flat extent format, can be one of "
2369                 "{monolithicSparse (default) | monolithicFlat | twoGbMaxExtentSparse | twoGbMaxExtentFlat | streamOptimized} "
2370         },
2371         {
2372             .name = BLOCK_OPT_ZEROED_GRAIN,
2373             .type = QEMU_OPT_BOOL,
2374             .help = "Enable efficient zero writes "
2375                     "using the zeroed-grain GTE feature"
2376         },
2377         { /* end of list */ }
2378     }
2379 };
2380
2381 static BlockDriver bdrv_vmdk = {
2382     .format_name                  = "vmdk",
2383     .instance_size                = sizeof(BDRVVmdkState),
2384     .bdrv_probe                   = vmdk_probe,
2385     .bdrv_open                    = vmdk_open,
2386     .bdrv_check                   = vmdk_check,
2387     .bdrv_reopen_prepare          = vmdk_reopen_prepare,
2388     .bdrv_child_perm              = bdrv_format_default_perms,
2389     .bdrv_co_preadv               = vmdk_co_preadv,
2390     .bdrv_co_pwritev              = vmdk_co_pwritev,
2391     .bdrv_co_pwritev_compressed   = vmdk_co_pwritev_compressed,
2392     .bdrv_co_pwrite_zeroes        = vmdk_co_pwrite_zeroes,
2393     .bdrv_close                   = vmdk_close,
2394     .bdrv_create                  = vmdk_create,
2395     .bdrv_co_flush_to_disk        = vmdk_co_flush,
2396     .bdrv_co_get_block_status     = vmdk_co_get_block_status,
2397     .bdrv_get_allocated_file_size = vmdk_get_allocated_file_size,
2398     .bdrv_has_zero_init           = vmdk_has_zero_init,
2399     .bdrv_get_specific_info       = vmdk_get_specific_info,
2400     .bdrv_refresh_limits          = vmdk_refresh_limits,
2401     .bdrv_get_info                = vmdk_get_info,
2402
2403     .supports_backing             = true,
2404     .create_opts                  = &vmdk_create_opts,
2405 };
2406
2407 static void bdrv_vmdk_init(void)
2408 {
2409     bdrv_register(&bdrv_vmdk);
2410 }
2411
2412 block_init(bdrv_vmdk_init);