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