block/parallels.c

   1 /*
   2  * Block driver for Parallels disk image format
   3  *
   4  * Copyright (c) 2007 Alex Beregszaszi
   5  * Copyright (c) 2015 Denis V. Lunev <den@openvz.org>
   6  *
   7  * This code was originally based on comparing different disk images created
   8  * by Parallels. Currently it is based on opened OpenVZ sources
   9  * available at
  10  *     http://git.openvz.org/?p=ploop;a=summary
  11  *
  12  * Permission is hereby granted, free of charge, to any person obtaining a copy
  13  * of this software and associated documentation files (the "Software"), to deal
  14  * in the Software without restriction, including without limitation the rights
  15  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
  16  * copies of the Software, and to permit persons to whom the Software is
  17  * furnished to do so, subject to the following conditions:
  18  *
  19  * The above copyright notice and this permission notice shall be included in
  20  * all copies or substantial portions of the Software.
  21  *
  22  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  23  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  24  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
  25  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  26  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  27  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  28  * THE SOFTWARE.
  29  */
  30
  31 #include "qemu/osdep.h"
  32 #include "qapi/error.h"
  33 #include "block/block_int.h"
  34 #include "sysemu/block-backend.h"
  35 #include "qemu/module.h"
  36 #include "qemu/option.h"
  37 #include "qemu/bswap.h"
  38 #include "qemu/bitmap.h"
  39 #include "migration/blocker.h"
  40 #include "parallels.h"
  41
  42 /**************************************************************/
  43
  44 #define HEADER_MAGIC "WithoutFreeSpace"
  45 #define HEADER_MAGIC2 "WithouFreSpacExt"
  46 #define HEADER_VERSION 2
  47 #define HEADER_INUSE_MAGIC  (0x746F6E59)
  48 #define MAX_PARALLELS_IMAGE_FACTOR (1ull << 32)
  49
  50 static QEnumLookup prealloc_mode_lookup = {
  51     .array = (const char *const[]) {
  52         "falloc",
  53         "truncate",
  54     },
  55     .size = PRL_PREALLOC_MODE__MAX
  56 };
  57
  58 #define PARALLELS_OPT_PREALLOC_MODE     "prealloc-mode"
  59 #define PARALLELS_OPT_PREALLOC_SIZE     "prealloc-size"
  60
  61 static QemuOptsList parallels_runtime_opts = {
  62     .name = "parallels",
  63     .head = QTAILQ_HEAD_INITIALIZER(parallels_runtime_opts.head),
  64     .desc = {
  65         {
  66             .name = PARALLELS_OPT_PREALLOC_SIZE,
  67             .type = QEMU_OPT_SIZE,
  68             .help = "Preallocation size on image expansion",
  69             .def_value_str = "128M",
  70         },
  71         {
  72             .name = PARALLELS_OPT_PREALLOC_MODE,
  73             .type = QEMU_OPT_STRING,
  74             .help = "Preallocation mode on image expansion "
  75                     "(allowed values: falloc, truncate)",
  76             .def_value_str = "falloc",
  77         },
  78         { /* end of list */ },
  79     },
  80 };
  81
  82
  83 static int64_t bat2sect(BDRVParallelsState *s, uint32_t idx)
  84 {
  85     return (uint64_t)le32_to_cpu(s->bat_bitmap[idx]) * s->off_multiplier;
  86 }
  87
  88 static uint32_t bat_entry_off(uint32_t idx)
  89 {
  90     return sizeof(ParallelsHeader) + sizeof(uint32_t) * idx;
  91 }
  92
  93 static int64_t seek_to_sector(BDRVParallelsState *s, int64_t sector_num)
  94 {
  95     uint32_t index, offset;
  96
  97     index = sector_num / s->tracks;
  98     offset = sector_num % s->tracks;
  99
 100     /* not allocated */
 101     if ((index >= s->bat_size) || (s->bat_bitmap[index] == 0)) {
 102         return -1;
 103     }
 104     return bat2sect(s, index) + offset;
 105 }
 106
 107 static int cluster_remainder(BDRVParallelsState *s, int64_t sector_num,
 108         int nb_sectors)
 109 {
 110     int ret = s->tracks - sector_num % s->tracks;
 111     return MIN(nb_sectors, ret);
 112 }
 113
 114 static int64_t block_status(BDRVParallelsState *s, int64_t sector_num,
 115                             int nb_sectors, int *pnum)
 116 {
 117     int64_t start_off = -2, prev_end_off = -2;
 118
 119     *pnum = 0;
 120     while (nb_sectors > 0 || start_off == -2) {
 121         int64_t offset = seek_to_sector(s, sector_num);
 122         int to_end;
 123
 124         if (start_off == -2) {
 125             start_off = offset;
 126             prev_end_off = offset;
 127         } else if (offset != prev_end_off) {
 128             break;
 129         }
 130
 131         to_end = cluster_remainder(s, sector_num, nb_sectors);
 132         nb_sectors -= to_end;
 133         sector_num += to_end;
 134         *pnum += to_end;
 135
 136         if (offset > 0) {
 137             prev_end_off += to_end;
 138         }
 139     }
 140     return start_off;
 141 }
 142
 143 static int64_t allocate_clusters(BlockDriverState *bs, int64_t sector_num,
 144                                  int nb_sectors, int *pnum)
 145 {
 146     int ret;
 147     BDRVParallelsState *s = bs->opaque;
 148     int64_t pos, space, idx, to_allocate, i, len;
 149
 150     pos = block_status(s, sector_num, nb_sectors, pnum);
 151     if (pos > 0) {
 152         return pos;
 153     }
 154
 155     idx = sector_num / s->tracks;
 156     to_allocate = DIV_ROUND_UP(sector_num + *pnum, s->tracks) - idx;
 157
 158     /* This function is called only by parallels_co_writev(), which will never
 159      * pass a sector_num at or beyond the end of the image (because the block
 160      * layer never passes such a sector_num to that function). Therefore, idx
 161      * is always below s->bat_size.
 162      * block_status() will limit *pnum so that sector_num + *pnum will not
 163      * exceed the image end. Therefore, idx + to_allocate cannot exceed
 164      * s->bat_size.
 165      * Note that s->bat_size is an unsigned int, therefore idx + to_allocate
 166      * will always fit into a uint32_t. */
 167     assert(idx < s->bat_size && idx + to_allocate <= s->bat_size);
 168
 169     space = to_allocate * s->tracks;
 170     len = bdrv_getlength(bs->file->bs);
 171     if (len < 0) {
 172         return len;
 173     }
 174     if (s->data_end + space > (len >> BDRV_SECTOR_BITS)) {
 175         space += s->prealloc_size;
 176         if (s->prealloc_mode == PRL_PREALLOC_MODE_FALLOCATE) {
 177             ret = bdrv_pwrite_zeroes(bs->file,
 178                                      s->data_end << BDRV_SECTOR_BITS,
 179                                      space << BDRV_SECTOR_BITS, 0);
 180         } else {
 181             ret = bdrv_truncate(bs->file,
 182                                 (s->data_end + space) << BDRV_SECTOR_BITS,
 183                                 PREALLOC_MODE_OFF, NULL);
 184         }
 185         if (ret < 0) {
 186             return ret;
 187         }
 188     }
 189
 190     /* Try to read from backing to fill empty clusters
 191      * FIXME: 1. previous write_zeroes may be redundant
 192      *        2. most of data we read from backing will be rewritten by
 193      *           parallels_co_writev. On aligned-to-cluster write we do not need
 194      *           this read at all.
 195      *        3. it would be good to combine write of data from backing and new
 196      *           data into one write call */
 197     if (bs->backing) {
 198         int64_t nb_cow_sectors = to_allocate * s->tracks;
 199         int64_t nb_cow_bytes = nb_cow_sectors << BDRV_SECTOR_BITS;
 200         QEMUIOVector qiov;
 201         struct iovec iov = {
 202             .iov_len = nb_cow_bytes,
 203             .iov_base = qemu_blockalign(bs, nb_cow_bytes)
 204         };
 205         qemu_iovec_init_external(&qiov, &iov, 1);
 206
 207         ret = bdrv_co_readv(bs->backing, idx * s->tracks, nb_cow_sectors,
 208                             &qiov);
 209         if (ret < 0) {
 210             qemu_vfree(iov.iov_base);
 211             return ret;
 212         }
 213
 214         ret = bdrv_co_writev(bs->file, s->data_end, nb_cow_sectors, &qiov);
 215         qemu_vfree(iov.iov_base);
 216         if (ret < 0) {
 217             return ret;
 218         }
 219     }
 220
 221     for (i = 0; i < to_allocate; i++) {
 222         s->bat_bitmap[idx + i] = cpu_to_le32(s->data_end / s->off_multiplier);
 223         s->data_end += s->tracks;
 224         bitmap_set(s->bat_dirty_bmap,
 225                    bat_entry_off(idx + i) / s->bat_dirty_block, 1);
 226     }
 227
 228     return bat2sect(s, idx) + sector_num % s->tracks;
 229 }
 230
 231
 232 static coroutine_fn int parallels_co_flush_to_os(BlockDriverState *bs)
 233 {
 234     BDRVParallelsState *s = bs->opaque;
 235     unsigned long size = DIV_ROUND_UP(s->header_size, s->bat_dirty_block);
 236     unsigned long bit;
 237
 238     qemu_co_mutex_lock(&s->lock);
 239
 240     bit = find_first_bit(s->bat_dirty_bmap, size);
 241     while (bit < size) {
 242         uint32_t off = bit * s->bat_dirty_block;
 243         uint32_t to_write = s->bat_dirty_block;
 244         int ret;
 245
 246         if (off + to_write > s->header_size) {
 247             to_write = s->header_size - off;
 248         }
 249         ret = bdrv_pwrite(bs->file, off, (uint8_t *)s->header + off,
 250                           to_write);
 251         if (ret < 0) {
 252             qemu_co_mutex_unlock(&s->lock);
 253             return ret;
 254         }
 255         bit = find_next_bit(s->bat_dirty_bmap, size, bit + 1);
 256     }
 257     bitmap_zero(s->bat_dirty_bmap, size);
 258
 259     qemu_co_mutex_unlock(&s->lock);
 260     return 0;
 261 }
 262
 263
 264 static int64_t coroutine_fn parallels_co_get_block_status(BlockDriverState *bs,
 265         int64_t sector_num, int nb_sectors, int *pnum, BlockDriverState **file)
 266 {
 267     BDRVParallelsState *s = bs->opaque;
 268     int64_t offset;
 269
 270     qemu_co_mutex_lock(&s->lock);
 271     offset = block_status(s, sector_num, nb_sectors, pnum);
 272     qemu_co_mutex_unlock(&s->lock);
 273
 274     if (offset < 0) {
 275         return 0;
 276     }
 277
 278     *file = bs->file->bs;
 279     return (offset << BDRV_SECTOR_BITS) |
 280         BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
 281 }
 282
 283 static coroutine_fn int parallels_co_writev(BlockDriverState *bs,
 284         int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
 285 {
 286     BDRVParallelsState *s = bs->opaque;
 287     uint64_t bytes_done = 0;
 288     QEMUIOVector hd_qiov;
 289     int ret = 0;
 290
 291     qemu_iovec_init(&hd_qiov, qiov->niov);
 292
 293     while (nb_sectors > 0) {
 294         int64_t position;
 295         int n, nbytes;
 296
 297         qemu_co_mutex_lock(&s->lock);
 298         position = allocate_clusters(bs, sector_num, nb_sectors, &n);
 299         qemu_co_mutex_unlock(&s->lock);
 300         if (position < 0) {
 301             ret = (int)position;
 302             break;
 303         }
 304
 305         nbytes = n << BDRV_SECTOR_BITS;
 306
 307         qemu_iovec_reset(&hd_qiov);
 308         qemu_iovec_concat(&hd_qiov, qiov, bytes_done, nbytes);
 309
 310         ret = bdrv_co_writev(bs->file, position, n, &hd_qiov);
 311         if (ret < 0) {
 312             break;
 313         }
 314
 315         nb_sectors -= n;
 316         sector_num += n;
 317         bytes_done += nbytes;
 318     }
 319
 320     qemu_iovec_destroy(&hd_qiov);
 321     return ret;
 322 }
 323
 324 static coroutine_fn int parallels_co_readv(BlockDriverState *bs,
 325         int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
 326 {
 327     BDRVParallelsState *s = bs->opaque;
 328     uint64_t bytes_done = 0;
 329     QEMUIOVector hd_qiov;
 330     int ret = 0;
 331
 332     qemu_iovec_init(&hd_qiov, qiov->niov);
 333
 334     while (nb_sectors > 0) {
 335         int64_t position;
 336         int n, nbytes;
 337
 338         qemu_co_mutex_lock(&s->lock);
 339         position = block_status(s, sector_num, nb_sectors, &n);
 340         qemu_co_mutex_unlock(&s->lock);
 341
 342         nbytes = n << BDRV_SECTOR_BITS;
 343
 344         qemu_iovec_reset(&hd_qiov);
 345         qemu_iovec_concat(&hd_qiov, qiov, bytes_done, nbytes);
 346
 347         if (position < 0) {
 348             if (bs->backing) {
 349                 ret = bdrv_co_readv(bs->backing, sector_num, n, &hd_qiov);
 350                 if (ret < 0) {
 351                     break;
 352                 }
 353             } else {
 354                 qemu_iovec_memset(&hd_qiov, 0, 0, nbytes);
 355             }
 356         } else {
 357             ret = bdrv_co_readv(bs->file, position, n, &hd_qiov);
 358             if (ret < 0) {
 359                 break;
 360             }
 361         }
 362
 363         nb_sectors -= n;
 364         sector_num += n;
 365         bytes_done += nbytes;
 366     }
 367
 368     qemu_iovec_destroy(&hd_qiov);
 369     return ret;
 370 }
 371
 372
 373 static int parallels_check(BlockDriverState *bs, BdrvCheckResult *res,
 374                            BdrvCheckMode fix)
 375 {
 376     BDRVParallelsState *s = bs->opaque;
 377     int64_t size, prev_off, high_off;
 378     int ret;
 379     uint32_t i;
 380     bool flush_bat = false;
 381     int cluster_size = s->tracks << BDRV_SECTOR_BITS;
 382
 383     size = bdrv_getlength(bs->file->bs);
 384     if (size < 0) {
 385         res->check_errors++;
 386         return size;
 387     }
 388
 389     if (s->header_unclean) {
 390         fprintf(stderr, "%s image was not closed correctly\n",
 391                 fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR");
 392         res->corruptions++;
 393         if (fix & BDRV_FIX_ERRORS) {
 394             /* parallels_close will do the job right */
 395             res->corruptions_fixed++;
 396             s->header_unclean = false;
 397         }
 398     }
 399
 400     res->bfi.total_clusters = s->bat_size;
 401     res->bfi.compressed_clusters = 0; /* compression is not supported */
 402
 403     high_off = 0;
 404     prev_off = 0;
 405     for (i = 0; i < s->bat_size; i++) {
 406         int64_t off = bat2sect(s, i) << BDRV_SECTOR_BITS;
 407         if (off == 0) {
 408             prev_off = 0;
 409             continue;
 410         }
 411
 412         /* cluster outside the image */
 413         if (off > size) {
 414             fprintf(stderr, "%s cluster %u is outside image\n",
 415                     fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR", i);
 416             res->corruptions++;
 417             if (fix & BDRV_FIX_ERRORS) {
 418                 prev_off = 0;
 419                 s->bat_bitmap[i] = 0;
 420                 res->corruptions_fixed++;
 421                 flush_bat = true;
 422                 continue;
 423             }
 424         }
 425
 426         res->bfi.allocated_clusters++;
 427         if (off > high_off) {
 428             high_off = off;
 429         }
 430
 431         if (prev_off != 0 && (prev_off + cluster_size) != off) {
 432             res->bfi.fragmented_clusters++;
 433         }
 434         prev_off = off;
 435     }
 436
 437     if (flush_bat) {
 438         ret = bdrv_pwrite_sync(bs->file, 0, s->header, s->header_size);
 439         if (ret < 0) {
 440             res->check_errors++;
 441             return ret;
 442         }
 443     }
 444
 445     res->image_end_offset = high_off + cluster_size;
 446     if (size > res->image_end_offset) {
 447         int64_t count;
 448         count = DIV_ROUND_UP(size - res->image_end_offset, cluster_size);
 449         fprintf(stderr, "%s space leaked at the end of the image %" PRId64 "\n",
 450                 fix & BDRV_FIX_LEAKS ? "Repairing" : "ERROR",
 451                 size - res->image_end_offset);
 452         res->leaks += count;
 453         if (fix & BDRV_FIX_LEAKS) {
 454             Error *local_err = NULL;
 455             ret = bdrv_truncate(bs->file, res->image_end_offset,
 456                                 PREALLOC_MODE_OFF, &local_err);
 457             if (ret < 0) {
 458                 error_report_err(local_err);
 459                 res->check_errors++;
 460                 return ret;
 461             }
 462             res->leaks_fixed += count;
 463         }
 464     }
 465
 466     return 0;
 467 }
 468
 469
 470 static int parallels_create(const char *filename, QemuOpts *opts, Error **errp)
 471 {
 472     int64_t total_size, cl_size;
 473     uint8_t tmp[BDRV_SECTOR_SIZE];
 474     Error *local_err = NULL;
 475     BlockBackend *file;
 476     uint32_t bat_entries, bat_sectors;
 477     ParallelsHeader header;
 478     int ret;
 479
 480     total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
 481                           BDRV_SECTOR_SIZE);
 482     cl_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE,
 483                           DEFAULT_CLUSTER_SIZE), BDRV_SECTOR_SIZE);
 484     if (total_size >= MAX_PARALLELS_IMAGE_FACTOR * cl_size) {
 485         error_propagate(errp, local_err);
 486         return -E2BIG;
 487     }
 488
 489     ret = bdrv_create_file(filename, opts, &local_err);
 490     if (ret < 0) {
 491         error_propagate(errp, local_err);
 492         return ret;
 493     }
 494
 495     file = blk_new_open(filename, NULL, NULL,
 496                         BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL,
 497                         &local_err);
 498     if (file == NULL) {
 499         error_propagate(errp, local_err);
 500         return -EIO;
 501     }
 502
 503     blk_set_allow_write_beyond_eof(file, true);
 504
 505     ret = blk_truncate(file, 0, PREALLOC_MODE_OFF, errp);
 506     if (ret < 0) {
 507         goto exit;
 508     }
 509
 510     bat_entries = DIV_ROUND_UP(total_size, cl_size);
 511     bat_sectors = DIV_ROUND_UP(bat_entry_off(bat_entries), cl_size);
 512     bat_sectors = (bat_sectors *  cl_size) >> BDRV_SECTOR_BITS;
 513
 514     memset(&header, 0, sizeof(header));
 515     memcpy(header.magic, HEADER_MAGIC2, sizeof(header.magic));
 516     header.version = cpu_to_le32(HEADER_VERSION);
 517     /* don't care much about geometry, it is not used on image level */
 518     header.heads = cpu_to_le32(HEADS_NUMBER);
 519     header.cylinders = cpu_to_le32(total_size / BDRV_SECTOR_SIZE
 520                                    / HEADS_NUMBER / SEC_IN_CYL);
 521     header.tracks = cpu_to_le32(cl_size >> BDRV_SECTOR_BITS);
 522     header.bat_entries = cpu_to_le32(bat_entries);
 523     header.nb_sectors = cpu_to_le64(DIV_ROUND_UP(total_size, BDRV_SECTOR_SIZE));
 524     header.data_off = cpu_to_le32(bat_sectors);
 525
 526     /* write all the data */
 527     memset(tmp, 0, sizeof(tmp));
 528     memcpy(tmp, &header, sizeof(header));
 529
 530     ret = blk_pwrite(file, 0, tmp, BDRV_SECTOR_SIZE, 0);
 531     if (ret < 0) {
 532         goto exit;
 533     }
 534     ret = blk_pwrite_zeroes(file, BDRV_SECTOR_SIZE,
 535                             (bat_sectors - 1) << BDRV_SECTOR_BITS, 0);
 536     if (ret < 0) {
 537         goto exit;
 538     }
 539     ret = 0;
 540
 541 done:
 542     blk_unref(file);
 543     return ret;
 544
 545 exit:
 546     error_setg_errno(errp, -ret, "Failed to create Parallels image");
 547     goto done;
 548 }
 549
 550
 551 static int parallels_probe(const uint8_t *buf, int buf_size,
 552                            const char *filename)
 553 {
 554     const ParallelsHeader *ph = (const void *)buf;
 555
 556     if (buf_size < sizeof(ParallelsHeader)) {
 557         return 0;
 558     }
 559
 560     if ((!memcmp(ph->magic, HEADER_MAGIC, 16) ||
 561            !memcmp(ph->magic, HEADER_MAGIC2, 16)) &&
 562            (le32_to_cpu(ph->version) == HEADER_VERSION)) {
 563         return 100;
 564     }
 565
 566     return 0;
 567 }
 568
 569 static int parallels_update_header(BlockDriverState *bs)
 570 {
 571     BDRVParallelsState *s = bs->opaque;
 572     unsigned size = MAX(bdrv_opt_mem_align(bs->file->bs),
 573                         sizeof(ParallelsHeader));
 574
 575     if (size > s->header_size) {
 576         size = s->header_size;
 577     }
 578     return bdrv_pwrite_sync(bs->file, 0, s->header, size);
 579 }
 580
 581 static int parallels_open(BlockDriverState *bs, QDict *options, int flags,
 582                           Error **errp)
 583 {
 584     BDRVParallelsState *s = bs->opaque;
 585     ParallelsHeader ph;
 586     int ret, size, i;
 587     QemuOpts *opts = NULL;
 588     Error *local_err = NULL;
 589     char *buf;
 590
 591     bs->file = bdrv_open_child(NULL, options, "file", bs, &child_file,
 592                                false, errp);
 593     if (!bs->file) {
 594         return -EINVAL;
 595     }
 596
 597     ret = bdrv_pread(bs->file, 0, &ph, sizeof(ph));
 598     if (ret < 0) {
 599         goto fail;
 600     }
 601
 602     bs->total_sectors = le64_to_cpu(ph.nb_sectors);
 603
 604     if (le32_to_cpu(ph.version) != HEADER_VERSION) {
 605         goto fail_format;
 606     }
 607     if (!memcmp(ph.magic, HEADER_MAGIC, 16)) {
 608         s->off_multiplier = 1;
 609         bs->total_sectors = 0xffffffff & bs->total_sectors;
 610     } else if (!memcmp(ph.magic, HEADER_MAGIC2, 16)) {
 611         s->off_multiplier = le32_to_cpu(ph.tracks);
 612     } else {
 613         goto fail_format;
 614     }
 615
 616     s->tracks = le32_to_cpu(ph.tracks);
 617     if (s->tracks == 0) {
 618         error_setg(errp, "Invalid image: Zero sectors per track");
 619         ret = -EINVAL;
 620         goto fail;
 621     }
 622     if (s->tracks > INT32_MAX/513) {
 623         error_setg(errp, "Invalid image: Too big cluster");
 624         ret = -EFBIG;
 625         goto fail;
 626     }
 627
 628     s->bat_size = le32_to_cpu(ph.bat_entries);
 629     if (s->bat_size > INT_MAX / sizeof(uint32_t)) {
 630         error_setg(errp, "Catalog too large");
 631         ret = -EFBIG;
 632         goto fail;
 633     }
 634
 635     size = bat_entry_off(s->bat_size);
 636     s->header_size = ROUND_UP(size, bdrv_opt_mem_align(bs->file->bs));
 637     s->header = qemu_try_blockalign(bs->file->bs, s->header_size);
 638     if (s->header == NULL) {
 639         ret = -ENOMEM;
 640         goto fail;
 641     }
 642     s->data_end = le32_to_cpu(ph.data_off);
 643     if (s->data_end == 0) {
 644         s->data_end = ROUND_UP(bat_entry_off(s->bat_size), BDRV_SECTOR_SIZE);
 645     }
 646     if (s->data_end < s->header_size) {
 647         /* there is not enough unused space to fit to block align between BAT
 648            and actual data. We can't avoid read-modify-write... */
 649         s->header_size = size;
 650     }
 651
 652     ret = bdrv_pread(bs->file, 0, s->header, s->header_size);
 653     if (ret < 0) {
 654         goto fail;
 655     }
 656     s->bat_bitmap = (uint32_t *)(s->header + 1);
 657
 658     for (i = 0; i < s->bat_size; i++) {
 659         int64_t off = bat2sect(s, i);
 660         if (off >= s->data_end) {
 661             s->data_end = off + s->tracks;
 662         }
 663     }
 664
 665     if (le32_to_cpu(ph.inuse) == HEADER_INUSE_MAGIC) {
 666         /* Image was not closed correctly. The check is mandatory */
 667         s->header_unclean = true;
 668         if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_CHECK)) {
 669             error_setg(errp, "parallels: Image was not closed correctly; "
 670                        "cannot be opened read/write");
 671             ret = -EACCES;
 672             goto fail;
 673         }
 674     }
 675
 676     opts = qemu_opts_create(&parallels_runtime_opts, NULL, 0, &local_err);
 677     if (local_err != NULL) {
 678         goto fail_options;
 679     }
 680
 681     qemu_opts_absorb_qdict(opts, options, &local_err);
 682     if (local_err != NULL) {
 683         goto fail_options;
 684     }
 685
 686     s->prealloc_size =
 687         qemu_opt_get_size_del(opts, PARALLELS_OPT_PREALLOC_SIZE, 0);
 688     s->prealloc_size = MAX(s->tracks, s->prealloc_size >> BDRV_SECTOR_BITS);
 689     buf = qemu_opt_get_del(opts, PARALLELS_OPT_PREALLOC_MODE);
 690     s->prealloc_mode = qapi_enum_parse(&prealloc_mode_lookup, buf,
 691                                        PRL_PREALLOC_MODE_FALLOCATE,
 692                                        &local_err);
 693     g_free(buf);
 694     if (local_err != NULL) {
 695         goto fail_options;
 696     }
 697
 698     if (!bdrv_has_zero_init(bs->file->bs)) {
 699         s->prealloc_mode = PRL_PREALLOC_MODE_FALLOCATE;
 700     }
 701
 702     if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_INACTIVE)) {
 703         s->header->inuse = cpu_to_le32(HEADER_INUSE_MAGIC);
 704         ret = parallels_update_header(bs);
 705         if (ret < 0) {
 706             goto fail;
 707         }
 708     }
 709
 710     s->bat_dirty_block = 4 * getpagesize();
 711     s->bat_dirty_bmap =
 712         bitmap_new(DIV_ROUND_UP(s->header_size, s->bat_dirty_block));
 713
 714     /* Disable migration until bdrv_invalidate_cache method is added */
 715     error_setg(&s->migration_blocker, "The Parallels format used by node '%s' "
 716                "does not support live migration",
 717                bdrv_get_device_or_node_name(bs));
 718     ret = migrate_add_blocker(s->migration_blocker, &local_err);
 719     if (local_err) {
 720         error_propagate(errp, local_err);
 721         error_free(s->migration_blocker);
 722         goto fail;
 723     }
 724     qemu_co_mutex_init(&s->lock);
 725     return 0;
 726
 727 fail_format:
 728     error_setg(errp, "Image not in Parallels format");
 729     ret = -EINVAL;
 730 fail:
 731     qemu_vfree(s->header);
 732     return ret;
 733
 734 fail_options:
 735     error_propagate(errp, local_err);
 736     ret = -EINVAL;
 737     goto fail;
 738 }
 739
 740
 741 static void parallels_close(BlockDriverState *bs)
 742 {
 743     BDRVParallelsState *s = bs->opaque;
 744
 745     if ((bs->open_flags & BDRV_O_RDWR) && !(bs->open_flags & BDRV_O_INACTIVE)) {
 746         s->header->inuse = 0;
 747         parallels_update_header(bs);
 748         bdrv_truncate(bs->file, s->data_end << BDRV_SECTOR_BITS,
 749                       PREALLOC_MODE_OFF, NULL);
 750     }
 751
 752     g_free(s->bat_dirty_bmap);
 753     qemu_vfree(s->header);
 754
 755     migrate_del_blocker(s->migration_blocker);
 756     error_free(s->migration_blocker);
 757 }
 758
 759 static QemuOptsList parallels_create_opts = {
 760     .name = "parallels-create-opts",
 761     .head = QTAILQ_HEAD_INITIALIZER(parallels_create_opts.head),
 762     .desc = {
 763         {
 764             .name = BLOCK_OPT_SIZE,
 765             .type = QEMU_OPT_SIZE,
 766             .help = "Virtual disk size",
 767         },
 768         {
 769             .name = BLOCK_OPT_CLUSTER_SIZE,
 770             .type = QEMU_OPT_SIZE,
 771             .help = "Parallels image cluster size",
 772             .def_value_str = stringify(DEFAULT_CLUSTER_SIZE),
 773         },
 774         { /* end of list */ }
 775     }
 776 };
 777
 778 static BlockDriver bdrv_parallels = {
 779     .format_name        = "parallels",
 780     .instance_size      = sizeof(BDRVParallelsState),
 781     .bdrv_probe         = parallels_probe,
 782     .bdrv_open          = parallels_open,
 783     .bdrv_close         = parallels_close,
 784     .bdrv_child_perm          = bdrv_format_default_perms,
 785     .bdrv_co_get_block_status = parallels_co_get_block_status,
 786     .bdrv_has_zero_init       = bdrv_has_zero_init_1,
 787     .bdrv_co_flush_to_os      = parallels_co_flush_to_os,
 788     .bdrv_co_readv  = parallels_co_readv,
 789     .bdrv_co_writev = parallels_co_writev,
 790     .supports_backing = true,
 791     .bdrv_create    = parallels_create,
 792     .bdrv_check     = parallels_check,
 793     .create_opts    = &parallels_create_opts,
 794 };
 795
 796 static void bdrv_parallels_init(void)
 797 {
 798     bdrv_register(&bdrv_parallels);
 799 }
 800
 801 block_init(bdrv_parallels_init);