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 int coroutine_fn parallels_co_block_status(BlockDriverState *bs,
 265                                                   bool want_zero,
 266                                                   int64_t offset,
 267                                                   int64_t bytes,
 268                                                   int64_t *pnum,
 269                                                   int64_t *map,
 270                                                   BlockDriverState **file)
 271 {
 272     BDRVParallelsState *s = bs->opaque;
 273     int count;
 274
 275     assert(QEMU_IS_ALIGNED(offset | bytes, BDRV_SECTOR_SIZE));
 276     qemu_co_mutex_lock(&s->lock);
 277     offset = block_status(s, offset >> BDRV_SECTOR_BITS,
 278                           bytes >> BDRV_SECTOR_BITS, &count);
 279     qemu_co_mutex_unlock(&s->lock);
 280
 281     *pnum = count * BDRV_SECTOR_SIZE;
 282     if (offset < 0) {
 283         return 0;
 284     }
 285
 286     *map = offset * BDRV_SECTOR_SIZE;
 287     *file = bs->file->bs;
 288     return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
 289 }
 290
 291 static coroutine_fn int parallels_co_writev(BlockDriverState *bs,
 292         int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
 293 {
 294     BDRVParallelsState *s = bs->opaque;
 295     uint64_t bytes_done = 0;
 296     QEMUIOVector hd_qiov;
 297     int ret = 0;
 298
 299     qemu_iovec_init(&hd_qiov, qiov->niov);
 300
 301     while (nb_sectors > 0) {
 302         int64_t position;
 303         int n, nbytes;
 304
 305         qemu_co_mutex_lock(&s->lock);
 306         position = allocate_clusters(bs, sector_num, nb_sectors, &n);
 307         qemu_co_mutex_unlock(&s->lock);
 308         if (position < 0) {
 309             ret = (int)position;
 310             break;
 311         }
 312
 313         nbytes = n << BDRV_SECTOR_BITS;
 314
 315         qemu_iovec_reset(&hd_qiov);
 316         qemu_iovec_concat(&hd_qiov, qiov, bytes_done, nbytes);
 317
 318         ret = bdrv_co_writev(bs->file, position, n, &hd_qiov);
 319         if (ret < 0) {
 320             break;
 321         }
 322
 323         nb_sectors -= n;
 324         sector_num += n;
 325         bytes_done += nbytes;
 326     }
 327
 328     qemu_iovec_destroy(&hd_qiov);
 329     return ret;
 330 }
 331
 332 static coroutine_fn int parallels_co_readv(BlockDriverState *bs,
 333         int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
 334 {
 335     BDRVParallelsState *s = bs->opaque;
 336     uint64_t bytes_done = 0;
 337     QEMUIOVector hd_qiov;
 338     int ret = 0;
 339
 340     qemu_iovec_init(&hd_qiov, qiov->niov);
 341
 342     while (nb_sectors > 0) {
 343         int64_t position;
 344         int n, nbytes;
 345
 346         qemu_co_mutex_lock(&s->lock);
 347         position = block_status(s, sector_num, nb_sectors, &n);
 348         qemu_co_mutex_unlock(&s->lock);
 349
 350         nbytes = n << BDRV_SECTOR_BITS;
 351
 352         qemu_iovec_reset(&hd_qiov);
 353         qemu_iovec_concat(&hd_qiov, qiov, bytes_done, nbytes);
 354
 355         if (position < 0) {
 356             if (bs->backing) {
 357                 ret = bdrv_co_readv(bs->backing, sector_num, n, &hd_qiov);
 358                 if (ret < 0) {
 359                     break;
 360                 }
 361             } else {
 362                 qemu_iovec_memset(&hd_qiov, 0, 0, nbytes);
 363             }
 364         } else {
 365             ret = bdrv_co_readv(bs->file, position, n, &hd_qiov);
 366             if (ret < 0) {
 367                 break;
 368             }
 369         }
 370
 371         nb_sectors -= n;
 372         sector_num += n;
 373         bytes_done += nbytes;
 374     }
 375
 376     qemu_iovec_destroy(&hd_qiov);
 377     return ret;
 378 }
 379
 380
 381 static int coroutine_fn parallels_co_check(BlockDriverState *bs,
 382                                            BdrvCheckResult *res,
 383                                            BdrvCheckMode fix)
 384 {
 385     BDRVParallelsState *s = bs->opaque;
 386     int64_t size, prev_off, high_off;
 387     int ret;
 388     uint32_t i;
 389     bool flush_bat = false;
 390     int cluster_size = s->tracks << BDRV_SECTOR_BITS;
 391
 392     size = bdrv_getlength(bs->file->bs);
 393     if (size < 0) {
 394         res->check_errors++;
 395         return size;
 396     }
 397
 398     qemu_co_mutex_lock(&s->lock);
 399     if (s->header_unclean) {
 400         fprintf(stderr, "%s image was not closed correctly\n",
 401                 fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR");
 402         res->corruptions++;
 403         if (fix & BDRV_FIX_ERRORS) {
 404             /* parallels_close will do the job right */
 405             res->corruptions_fixed++;
 406             s->header_unclean = false;
 407         }
 408     }
 409
 410     res->bfi.total_clusters = s->bat_size;
 411     res->bfi.compressed_clusters = 0; /* compression is not supported */
 412
 413     high_off = 0;
 414     prev_off = 0;
 415     for (i = 0; i < s->bat_size; i++) {
 416         int64_t off = bat2sect(s, i) << BDRV_SECTOR_BITS;
 417         if (off == 0) {
 418             prev_off = 0;
 419             continue;
 420         }
 421
 422         /* cluster outside the image */
 423         if (off > size) {
 424             fprintf(stderr, "%s cluster %u is outside image\n",
 425                     fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR", i);
 426             res->corruptions++;
 427             if (fix & BDRV_FIX_ERRORS) {
 428                 prev_off = 0;
 429                 s->bat_bitmap[i] = 0;
 430                 res->corruptions_fixed++;
 431                 flush_bat = true;
 432                 continue;
 433             }
 434         }
 435
 436         res->bfi.allocated_clusters++;
 437         if (off > high_off) {
 438             high_off = off;
 439         }
 440
 441         if (prev_off != 0 && (prev_off + cluster_size) != off) {
 442             res->bfi.fragmented_clusters++;
 443         }
 444         prev_off = off;
 445     }
 446
 447     ret = 0;
 448     if (flush_bat) {
 449         ret = bdrv_pwrite_sync(bs->file, 0, s->header, s->header_size);
 450         if (ret < 0) {
 451             res->check_errors++;
 452             goto out;
 453         }
 454     }
 455
 456     res->image_end_offset = high_off + cluster_size;
 457     if (size > res->image_end_offset) {
 458         int64_t count;
 459         count = DIV_ROUND_UP(size - res->image_end_offset, cluster_size);
 460         fprintf(stderr, "%s space leaked at the end of the image %" PRId64 "\n",
 461                 fix & BDRV_FIX_LEAKS ? "Repairing" : "ERROR",
 462                 size - res->image_end_offset);
 463         res->leaks += count;
 464         if (fix & BDRV_FIX_LEAKS) {
 465             Error *local_err = NULL;
 466             ret = bdrv_truncate(bs->file, res->image_end_offset,
 467                                 PREALLOC_MODE_OFF, &local_err);
 468             if (ret < 0) {
 469                 error_report_err(local_err);
 470                 res->check_errors++;
 471                 goto out;
 472             }
 473             res->leaks_fixed += count;
 474         }
 475     }
 476
 477 out:
 478     qemu_co_mutex_unlock(&s->lock);
 479     return ret;
 480 }
 481
 482
 483 static int coroutine_fn parallels_co_create_opts(const char *filename,
 484                                                  QemuOpts *opts,
 485                                                  Error **errp)
 486 {
 487     int64_t total_size, cl_size;
 488     uint8_t tmp[BDRV_SECTOR_SIZE];
 489     Error *local_err = NULL;
 490     BlockBackend *file;
 491     uint32_t bat_entries, bat_sectors;
 492     ParallelsHeader header;
 493     int ret;
 494
 495     total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
 496                           BDRV_SECTOR_SIZE);
 497     cl_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE,
 498                           DEFAULT_CLUSTER_SIZE), BDRV_SECTOR_SIZE);
 499     if (total_size >= MAX_PARALLELS_IMAGE_FACTOR * cl_size) {
 500         error_propagate(errp, local_err);
 501         return -E2BIG;
 502     }
 503
 504     ret = bdrv_create_file(filename, opts, &local_err);
 505     if (ret < 0) {
 506         error_propagate(errp, local_err);
 507         return ret;
 508     }
 509
 510     file = blk_new_open(filename, NULL, NULL,
 511                         BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL,
 512                         &local_err);
 513     if (file == NULL) {
 514         error_propagate(errp, local_err);
 515         return -EIO;
 516     }
 517
 518     blk_set_allow_write_beyond_eof(file, true);
 519
 520     ret = blk_truncate(file, 0, PREALLOC_MODE_OFF, errp);
 521     if (ret < 0) {
 522         goto exit;
 523     }
 524
 525     bat_entries = DIV_ROUND_UP(total_size, cl_size);
 526     bat_sectors = DIV_ROUND_UP(bat_entry_off(bat_entries), cl_size);
 527     bat_sectors = (bat_sectors *  cl_size) >> BDRV_SECTOR_BITS;
 528
 529     memset(&header, 0, sizeof(header));
 530     memcpy(header.magic, HEADER_MAGIC2, sizeof(header.magic));
 531     header.version = cpu_to_le32(HEADER_VERSION);
 532     /* don't care much about geometry, it is not used on image level */
 533     header.heads = cpu_to_le32(HEADS_NUMBER);
 534     header.cylinders = cpu_to_le32(total_size / BDRV_SECTOR_SIZE
 535                                    / HEADS_NUMBER / SEC_IN_CYL);
 536     header.tracks = cpu_to_le32(cl_size >> BDRV_SECTOR_BITS);
 537     header.bat_entries = cpu_to_le32(bat_entries);
 538     header.nb_sectors = cpu_to_le64(DIV_ROUND_UP(total_size, BDRV_SECTOR_SIZE));
 539     header.data_off = cpu_to_le32(bat_sectors);
 540
 541     /* write all the data */
 542     memset(tmp, 0, sizeof(tmp));
 543     memcpy(tmp, &header, sizeof(header));
 544
 545     ret = blk_pwrite(file, 0, tmp, BDRV_SECTOR_SIZE, 0);
 546     if (ret < 0) {
 547         goto exit;
 548     }
 549     ret = blk_pwrite_zeroes(file, BDRV_SECTOR_SIZE,
 550                             (bat_sectors - 1) << BDRV_SECTOR_BITS, 0);
 551     if (ret < 0) {
 552         goto exit;
 553     }
 554     ret = 0;
 555
 556 done:
 557     blk_unref(file);
 558     return ret;
 559
 560 exit:
 561     error_setg_errno(errp, -ret, "Failed to create Parallels image");
 562     goto done;
 563 }
 564
 565
 566 static int parallels_probe(const uint8_t *buf, int buf_size,
 567                            const char *filename)
 568 {
 569     const ParallelsHeader *ph = (const void *)buf;
 570
 571     if (buf_size < sizeof(ParallelsHeader)) {
 572         return 0;
 573     }
 574
 575     if ((!memcmp(ph->magic, HEADER_MAGIC, 16) ||
 576            !memcmp(ph->magic, HEADER_MAGIC2, 16)) &&
 577            (le32_to_cpu(ph->version) == HEADER_VERSION)) {
 578         return 100;
 579     }
 580
 581     return 0;
 582 }
 583
 584 static int parallels_update_header(BlockDriverState *bs)
 585 {
 586     BDRVParallelsState *s = bs->opaque;
 587     unsigned size = MAX(bdrv_opt_mem_align(bs->file->bs),
 588                         sizeof(ParallelsHeader));
 589
 590     if (size > s->header_size) {
 591         size = s->header_size;
 592     }
 593     return bdrv_pwrite_sync(bs->file, 0, s->header, size);
 594 }
 595
 596 static int parallels_open(BlockDriverState *bs, QDict *options, int flags,
 597                           Error **errp)
 598 {
 599     BDRVParallelsState *s = bs->opaque;
 600     ParallelsHeader ph;
 601     int ret, size, i;
 602     QemuOpts *opts = NULL;
 603     Error *local_err = NULL;
 604     char *buf;
 605
 606     bs->file = bdrv_open_child(NULL, options, "file", bs, &child_file,
 607                                false, errp);
 608     if (!bs->file) {
 609         return -EINVAL;
 610     }
 611
 612     ret = bdrv_pread(bs->file, 0, &ph, sizeof(ph));
 613     if (ret < 0) {
 614         goto fail;
 615     }
 616
 617     bs->total_sectors = le64_to_cpu(ph.nb_sectors);
 618
 619     if (le32_to_cpu(ph.version) != HEADER_VERSION) {
 620         goto fail_format;
 621     }
 622     if (!memcmp(ph.magic, HEADER_MAGIC, 16)) {
 623         s->off_multiplier = 1;
 624         bs->total_sectors = 0xffffffff & bs->total_sectors;
 625     } else if (!memcmp(ph.magic, HEADER_MAGIC2, 16)) {
 626         s->off_multiplier = le32_to_cpu(ph.tracks);
 627     } else {
 628         goto fail_format;
 629     }
 630
 631     s->tracks = le32_to_cpu(ph.tracks);
 632     if (s->tracks == 0) {
 633         error_setg(errp, "Invalid image: Zero sectors per track");
 634         ret = -EINVAL;
 635         goto fail;
 636     }
 637     if (s->tracks > INT32_MAX/513) {
 638         error_setg(errp, "Invalid image: Too big cluster");
 639         ret = -EFBIG;
 640         goto fail;
 641     }
 642
 643     s->bat_size = le32_to_cpu(ph.bat_entries);
 644     if (s->bat_size > INT_MAX / sizeof(uint32_t)) {
 645         error_setg(errp, "Catalog too large");
 646         ret = -EFBIG;
 647         goto fail;
 648     }
 649
 650     size = bat_entry_off(s->bat_size);
 651     s->header_size = ROUND_UP(size, bdrv_opt_mem_align(bs->file->bs));
 652     s->header = qemu_try_blockalign(bs->file->bs, s->header_size);
 653     if (s->header == NULL) {
 654         ret = -ENOMEM;
 655         goto fail;
 656     }
 657     s->data_end = le32_to_cpu(ph.data_off);
 658     if (s->data_end == 0) {
 659         s->data_end = ROUND_UP(bat_entry_off(s->bat_size), BDRV_SECTOR_SIZE);
 660     }
 661     if (s->data_end < s->header_size) {
 662         /* there is not enough unused space to fit to block align between BAT
 663            and actual data. We can't avoid read-modify-write... */
 664         s->header_size = size;
 665     }
 666
 667     ret = bdrv_pread(bs->file, 0, s->header, s->header_size);
 668     if (ret < 0) {
 669         goto fail;
 670     }
 671     s->bat_bitmap = (uint32_t *)(s->header + 1);
 672
 673     for (i = 0; i < s->bat_size; i++) {
 674         int64_t off = bat2sect(s, i);
 675         if (off >= s->data_end) {
 676             s->data_end = off + s->tracks;
 677         }
 678     }
 679
 680     if (le32_to_cpu(ph.inuse) == HEADER_INUSE_MAGIC) {
 681         /* Image was not closed correctly. The check is mandatory */
 682         s->header_unclean = true;
 683         if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_CHECK)) {
 684             error_setg(errp, "parallels: Image was not closed correctly; "
 685                        "cannot be opened read/write");
 686             ret = -EACCES;
 687             goto fail;
 688         }
 689     }
 690
 691     opts = qemu_opts_create(&parallels_runtime_opts, NULL, 0, &local_err);
 692     if (local_err != NULL) {
 693         goto fail_options;
 694     }
 695
 696     qemu_opts_absorb_qdict(opts, options, &local_err);
 697     if (local_err != NULL) {
 698         goto fail_options;
 699     }
 700
 701     s->prealloc_size =
 702         qemu_opt_get_size_del(opts, PARALLELS_OPT_PREALLOC_SIZE, 0);
 703     s->prealloc_size = MAX(s->tracks, s->prealloc_size >> BDRV_SECTOR_BITS);
 704     buf = qemu_opt_get_del(opts, PARALLELS_OPT_PREALLOC_MODE);
 705     s->prealloc_mode = qapi_enum_parse(&prealloc_mode_lookup, buf,
 706                                        PRL_PREALLOC_MODE_FALLOCATE,
 707                                        &local_err);
 708     g_free(buf);
 709     if (local_err != NULL) {
 710         goto fail_options;
 711     }
 712
 713     if (!bdrv_has_zero_init(bs->file->bs)) {
 714         s->prealloc_mode = PRL_PREALLOC_MODE_FALLOCATE;
 715     }
 716
 717     if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_INACTIVE)) {
 718         s->header->inuse = cpu_to_le32(HEADER_INUSE_MAGIC);
 719         ret = parallels_update_header(bs);
 720         if (ret < 0) {
 721             goto fail;
 722         }
 723     }
 724
 725     s->bat_dirty_block = 4 * getpagesize();
 726     s->bat_dirty_bmap =
 727         bitmap_new(DIV_ROUND_UP(s->header_size, s->bat_dirty_block));
 728
 729     /* Disable migration until bdrv_invalidate_cache method is added */
 730     error_setg(&s->migration_blocker, "The Parallels format used by node '%s' "
 731                "does not support live migration",
 732                bdrv_get_device_or_node_name(bs));
 733     ret = migrate_add_blocker(s->migration_blocker, &local_err);
 734     if (local_err) {
 735         error_propagate(errp, local_err);
 736         error_free(s->migration_blocker);
 737         goto fail;
 738     }
 739     qemu_co_mutex_init(&s->lock);
 740     return 0;
 741
 742 fail_format:
 743     error_setg(errp, "Image not in Parallels format");
 744     ret = -EINVAL;
 745 fail:
 746     qemu_vfree(s->header);
 747     return ret;
 748
 749 fail_options:
 750     error_propagate(errp, local_err);
 751     ret = -EINVAL;
 752     goto fail;
 753 }
 754
 755
 756 static void parallels_close(BlockDriverState *bs)
 757 {
 758     BDRVParallelsState *s = bs->opaque;
 759
 760     if ((bs->open_flags & BDRV_O_RDWR) && !(bs->open_flags & BDRV_O_INACTIVE)) {
 761         s->header->inuse = 0;
 762         parallels_update_header(bs);
 763         bdrv_truncate(bs->file, s->data_end << BDRV_SECTOR_BITS,
 764                       PREALLOC_MODE_OFF, NULL);
 765     }
 766
 767     g_free(s->bat_dirty_bmap);
 768     qemu_vfree(s->header);
 769
 770     migrate_del_blocker(s->migration_blocker);
 771     error_free(s->migration_blocker);
 772 }
 773
 774 static QemuOptsList parallels_create_opts = {
 775     .name = "parallels-create-opts",
 776     .head = QTAILQ_HEAD_INITIALIZER(parallels_create_opts.head),
 777     .desc = {
 778         {
 779             .name = BLOCK_OPT_SIZE,
 780             .type = QEMU_OPT_SIZE,
 781             .help = "Virtual disk size",
 782         },
 783         {
 784             .name = BLOCK_OPT_CLUSTER_SIZE,
 785             .type = QEMU_OPT_SIZE,
 786             .help = "Parallels image cluster size",
 787             .def_value_str = stringify(DEFAULT_CLUSTER_SIZE),
 788         },
 789         { /* end of list */ }
 790     }
 791 };
 792
 793 static BlockDriver bdrv_parallels = {
 794     .format_name        = "parallels",
 795     .instance_size      = sizeof(BDRVParallelsState),
 796     .bdrv_probe         = parallels_probe,
 797     .bdrv_open          = parallels_open,
 798     .bdrv_close         = parallels_close,
 799     .bdrv_child_perm          = bdrv_format_default_perms,
 800     .bdrv_co_block_status     = parallels_co_block_status,
 801     .bdrv_has_zero_init       = bdrv_has_zero_init_1,
 802     .bdrv_co_flush_to_os      = parallels_co_flush_to_os,
 803     .bdrv_co_readv  = parallels_co_readv,
 804     .bdrv_co_writev = parallels_co_writev,
 805     .supports_backing = true,
 806     .bdrv_co_create_opts = parallels_co_create_opts,
 807     .bdrv_co_check  = parallels_co_check,
 808     .create_opts    = &parallels_create_opts,
 809 };
 810
 811 static void bdrv_parallels_init(void)
 812 {
 813     bdrv_register(&bdrv_parallels);
 814 }
 815
 816 block_init(bdrv_parallels_init);