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
  40 /**************************************************************/
  41
  42 #define HEADER_MAGIC "WithoutFreeSpace"
  43 #define HEADER_MAGIC2 "WithouFreSpacExt"
  44 #define HEADER_VERSION 2
  45 #define HEADER_INUSE_MAGIC  (0x746F6E59)
  46 #define MAX_PARALLELS_IMAGE_FACTOR (1ull << 32)
  47
  48 #define DEFAULT_CLUSTER_SIZE 1048576        /* 1 MiB */
  49
  50
  51 // always little-endian
  52 typedef struct ParallelsHeader {
  53     char magic[16]; // "WithoutFreeSpace"
  54     uint32_t version;
  55     uint32_t heads;
  56     uint32_t cylinders;
  57     uint32_t tracks;
  58     uint32_t bat_entries;
  59     uint64_t nb_sectors;
  60     uint32_t inuse;
  61     uint32_t data_off;
  62     char padding[12];
  63 } QEMU_PACKED ParallelsHeader;
  64
  65
  66 typedef enum ParallelsPreallocMode {
  67     PRL_PREALLOC_MODE_FALLOCATE = 0,
  68     PRL_PREALLOC_MODE_TRUNCATE = 1,
  69     PRL_PREALLOC_MODE__MAX = 2,
  70 } ParallelsPreallocMode;
  71
  72 static QEnumLookup prealloc_mode_lookup = {
  73     .array = (const char *const[]) {
  74         "falloc",
  75         "truncate",
  76     },
  77     .size = PRL_PREALLOC_MODE__MAX
  78 };
  79
  80 typedef struct BDRVParallelsState {
  81     /** Locking is conservative, the lock protects
  82      *   - image file extending (truncate, fallocate)
  83      *   - any access to block allocation table
  84      */
  85     CoMutex lock;
  86
  87     ParallelsHeader *header;
  88     uint32_t header_size;
  89     bool header_unclean;
  90
  91     unsigned long *bat_dirty_bmap;
  92     unsigned int  bat_dirty_block;
  93
  94     uint32_t *bat_bitmap;
  95     unsigned int bat_size;
  96
  97     int64_t  data_end;
  98     uint64_t prealloc_size;
  99     ParallelsPreallocMode prealloc_mode;
 100
 101     unsigned int tracks;
 102
 103     unsigned int off_multiplier;
 104     Error *migration_blocker;
 105 } BDRVParallelsState;
 106
 107
 108 #define PARALLELS_OPT_PREALLOC_MODE     "prealloc-mode"
 109 #define PARALLELS_OPT_PREALLOC_SIZE     "prealloc-size"
 110
 111 static QemuOptsList parallels_runtime_opts = {
 112     .name = "parallels",
 113     .head = QTAILQ_HEAD_INITIALIZER(parallels_runtime_opts.head),
 114     .desc = {
 115         {
 116             .name = PARALLELS_OPT_PREALLOC_SIZE,
 117             .type = QEMU_OPT_SIZE,
 118             .help = "Preallocation size on image expansion",
 119             .def_value_str = "128M",
 120         },
 121         {
 122             .name = PARALLELS_OPT_PREALLOC_MODE,
 123             .type = QEMU_OPT_STRING,
 124             .help = "Preallocation mode on image expansion "
 125                     "(allowed values: falloc, truncate)",
 126             .def_value_str = "falloc",
 127         },
 128         { /* end of list */ },
 129     },
 130 };
 131
 132
 133 static int64_t bat2sect(BDRVParallelsState *s, uint32_t idx)
 134 {
 135     return (uint64_t)le32_to_cpu(s->bat_bitmap[idx]) * s->off_multiplier;
 136 }
 137
 138 static uint32_t bat_entry_off(uint32_t idx)
 139 {
 140     return sizeof(ParallelsHeader) + sizeof(uint32_t) * idx;
 141 }
 142
 143 static int64_t seek_to_sector(BDRVParallelsState *s, int64_t sector_num)
 144 {
 145     uint32_t index, offset;
 146
 147     index = sector_num / s->tracks;
 148     offset = sector_num % s->tracks;
 149
 150     /* not allocated */
 151     if ((index >= s->bat_size) || (s->bat_bitmap[index] == 0)) {
 152         return -1;
 153     }
 154     return bat2sect(s, index) + offset;
 155 }
 156
 157 static int cluster_remainder(BDRVParallelsState *s, int64_t sector_num,
 158         int nb_sectors)
 159 {
 160     int ret = s->tracks - sector_num % s->tracks;
 161     return MIN(nb_sectors, ret);
 162 }
 163
 164 static int64_t block_status(BDRVParallelsState *s, int64_t sector_num,
 165                             int nb_sectors, int *pnum)
 166 {
 167     int64_t start_off = -2, prev_end_off = -2;
 168
 169     *pnum = 0;
 170     while (nb_sectors > 0 || start_off == -2) {
 171         int64_t offset = seek_to_sector(s, sector_num);
 172         int to_end;
 173
 174         if (start_off == -2) {
 175             start_off = offset;
 176             prev_end_off = offset;
 177         } else if (offset != prev_end_off) {
 178             break;
 179         }
 180
 181         to_end = cluster_remainder(s, sector_num, nb_sectors);
 182         nb_sectors -= to_end;
 183         sector_num += to_end;
 184         *pnum += to_end;
 185
 186         if (offset > 0) {
 187             prev_end_off += to_end;
 188         }
 189     }
 190     return start_off;
 191 }
 192
 193 static int64_t allocate_clusters(BlockDriverState *bs, int64_t sector_num,
 194                                  int nb_sectors, int *pnum)
 195 {
 196     BDRVParallelsState *s = bs->opaque;
 197     int64_t pos, space, idx, to_allocate, i, len;
 198
 199     pos = block_status(s, sector_num, nb_sectors, pnum);
 200     if (pos > 0) {
 201         return pos;
 202     }
 203
 204     idx = sector_num / s->tracks;
 205     to_allocate = DIV_ROUND_UP(sector_num + *pnum, s->tracks) - idx;
 206
 207     /* This function is called only by parallels_co_writev(), which will never
 208      * pass a sector_num at or beyond the end of the image (because the block
 209      * layer never passes such a sector_num to that function). Therefore, idx
 210      * is always below s->bat_size.
 211      * block_status() will limit *pnum so that sector_num + *pnum will not
 212      * exceed the image end. Therefore, idx + to_allocate cannot exceed
 213      * s->bat_size.
 214      * Note that s->bat_size is an unsigned int, therefore idx + to_allocate
 215      * will always fit into a uint32_t. */
 216     assert(idx < s->bat_size && idx + to_allocate <= s->bat_size);
 217
 218     space = to_allocate * s->tracks;
 219     len = bdrv_getlength(bs->file->bs);
 220     if (len < 0) {
 221         return len;
 222     }
 223     if (s->data_end + space > (len >> BDRV_SECTOR_BITS)) {
 224         int ret;
 225         space += s->prealloc_size;
 226         if (s->prealloc_mode == PRL_PREALLOC_MODE_FALLOCATE) {
 227             ret = bdrv_pwrite_zeroes(bs->file,
 228                                      s->data_end << BDRV_SECTOR_BITS,
 229                                      space << BDRV_SECTOR_BITS, 0);
 230         } else {
 231             ret = bdrv_truncate(bs->file,
 232                                 (s->data_end + space) << BDRV_SECTOR_BITS,
 233                                 PREALLOC_MODE_OFF, NULL);
 234         }
 235         if (ret < 0) {
 236             return ret;
 237         }
 238     }
 239
 240     for (i = 0; i < to_allocate; i++) {
 241         s->bat_bitmap[idx + i] = cpu_to_le32(s->data_end / s->off_multiplier);
 242         s->data_end += s->tracks;
 243         bitmap_set(s->bat_dirty_bmap,
 244                    bat_entry_off(idx + i) / s->bat_dirty_block, 1);
 245     }
 246
 247     return bat2sect(s, idx) + sector_num % s->tracks;
 248 }
 249
 250
 251 static coroutine_fn int parallels_co_flush_to_os(BlockDriverState *bs)
 252 {
 253     BDRVParallelsState *s = bs->opaque;
 254     unsigned long size = DIV_ROUND_UP(s->header_size, s->bat_dirty_block);
 255     unsigned long bit;
 256
 257     qemu_co_mutex_lock(&s->lock);
 258
 259     bit = find_first_bit(s->bat_dirty_bmap, size);
 260     while (bit < size) {
 261         uint32_t off = bit * s->bat_dirty_block;
 262         uint32_t to_write = s->bat_dirty_block;
 263         int ret;
 264
 265         if (off + to_write > s->header_size) {
 266             to_write = s->header_size - off;
 267         }
 268         ret = bdrv_pwrite(bs->file, off, (uint8_t *)s->header + off,
 269                           to_write);
 270         if (ret < 0) {
 271             qemu_co_mutex_unlock(&s->lock);
 272             return ret;
 273         }
 274         bit = find_next_bit(s->bat_dirty_bmap, size, bit + 1);
 275     }
 276     bitmap_zero(s->bat_dirty_bmap, size);
 277
 278     qemu_co_mutex_unlock(&s->lock);
 279     return 0;
 280 }
 281
 282
 283 static int64_t coroutine_fn parallels_co_get_block_status(BlockDriverState *bs,
 284         int64_t sector_num, int nb_sectors, int *pnum, BlockDriverState **file)
 285 {
 286     BDRVParallelsState *s = bs->opaque;
 287     int64_t offset;
 288
 289     qemu_co_mutex_lock(&s->lock);
 290     offset = block_status(s, sector_num, nb_sectors, pnum);
 291     qemu_co_mutex_unlock(&s->lock);
 292
 293     if (offset < 0) {
 294         return 0;
 295     }
 296
 297     *file = bs->file->bs;
 298     return (offset << BDRV_SECTOR_BITS) |
 299         BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
 300 }
 301
 302 static coroutine_fn int parallels_co_writev(BlockDriverState *bs,
 303         int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
 304 {
 305     BDRVParallelsState *s = bs->opaque;
 306     uint64_t bytes_done = 0;
 307     QEMUIOVector hd_qiov;
 308     int ret = 0;
 309
 310     qemu_iovec_init(&hd_qiov, qiov->niov);
 311
 312     while (nb_sectors > 0) {
 313         int64_t position;
 314         int n, nbytes;
 315
 316         qemu_co_mutex_lock(&s->lock);
 317         position = allocate_clusters(bs, sector_num, nb_sectors, &n);
 318         qemu_co_mutex_unlock(&s->lock);
 319         if (position < 0) {
 320             ret = (int)position;
 321             break;
 322         }
 323
 324         nbytes = n << BDRV_SECTOR_BITS;
 325
 326         qemu_iovec_reset(&hd_qiov);
 327         qemu_iovec_concat(&hd_qiov, qiov, bytes_done, nbytes);
 328
 329         ret = bdrv_co_writev(bs->file, position, n, &hd_qiov);
 330         if (ret < 0) {
 331             break;
 332         }
 333
 334         nb_sectors -= n;
 335         sector_num += n;
 336         bytes_done += nbytes;
 337     }
 338
 339     qemu_iovec_destroy(&hd_qiov);
 340     return ret;
 341 }
 342
 343 static coroutine_fn int parallels_co_readv(BlockDriverState *bs,
 344         int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
 345 {
 346     BDRVParallelsState *s = bs->opaque;
 347     uint64_t bytes_done = 0;
 348     QEMUIOVector hd_qiov;
 349     int ret = 0;
 350
 351     qemu_iovec_init(&hd_qiov, qiov->niov);
 352
 353     while (nb_sectors > 0) {
 354         int64_t position;
 355         int n, nbytes;
 356
 357         qemu_co_mutex_lock(&s->lock);
 358         position = block_status(s, sector_num, nb_sectors, &n);
 359         qemu_co_mutex_unlock(&s->lock);
 360
 361         nbytes = n << BDRV_SECTOR_BITS;
 362
 363         if (position < 0) {
 364             qemu_iovec_memset(qiov, bytes_done, 0, nbytes);
 365         } else {
 366             qemu_iovec_reset(&hd_qiov);
 367             qemu_iovec_concat(&hd_qiov, qiov, bytes_done, nbytes);
 368
 369             ret = bdrv_co_readv(bs->file, position, n, &hd_qiov);
 370             if (ret < 0) {
 371                 break;
 372             }
 373         }
 374
 375         nb_sectors -= n;
 376         sector_num += n;
 377         bytes_done += nbytes;
 378     }
 379
 380     qemu_iovec_destroy(&hd_qiov);
 381     return ret;
 382 }
 383
 384
 385 static int parallels_check(BlockDriverState *bs, BdrvCheckResult *res,
 386                            BdrvCheckMode fix)
 387 {
 388     BDRVParallelsState *s = bs->opaque;
 389     int64_t size, prev_off, high_off;
 390     int ret;
 391     uint32_t i;
 392     bool flush_bat = false;
 393     int cluster_size = s->tracks << BDRV_SECTOR_BITS;
 394
 395     size = bdrv_getlength(bs->file->bs);
 396     if (size < 0) {
 397         res->check_errors++;
 398         return size;
 399     }
 400
 401     if (s->header_unclean) {
 402         fprintf(stderr, "%s image was not closed correctly\n",
 403                 fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR");
 404         res->corruptions++;
 405         if (fix & BDRV_FIX_ERRORS) {
 406             /* parallels_close will do the job right */
 407             res->corruptions_fixed++;
 408             s->header_unclean = false;
 409         }
 410     }
 411
 412     res->bfi.total_clusters = s->bat_size;
 413     res->bfi.compressed_clusters = 0; /* compression is not supported */
 414
 415     high_off = 0;
 416     prev_off = 0;
 417     for (i = 0; i < s->bat_size; i++) {
 418         int64_t off = bat2sect(s, i) << BDRV_SECTOR_BITS;
 419         if (off == 0) {
 420             prev_off = 0;
 421             continue;
 422         }
 423
 424         /* cluster outside the image */
 425         if (off > size) {
 426             fprintf(stderr, "%s cluster %u is outside image\n",
 427                     fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR", i);
 428             res->corruptions++;
 429             if (fix & BDRV_FIX_ERRORS) {
 430                 prev_off = 0;
 431                 s->bat_bitmap[i] = 0;
 432                 res->corruptions_fixed++;
 433                 flush_bat = true;
 434                 continue;
 435             }
 436         }
 437
 438         res->bfi.allocated_clusters++;
 439         if (off > high_off) {
 440             high_off = off;
 441         }
 442
 443         if (prev_off != 0 && (prev_off + cluster_size) != off) {
 444             res->bfi.fragmented_clusters++;
 445         }
 446         prev_off = off;
 447     }
 448
 449     if (flush_bat) {
 450         ret = bdrv_pwrite_sync(bs->file, 0, s->header, s->header_size);
 451         if (ret < 0) {
 452             res->check_errors++;
 453             return ret;
 454         }
 455     }
 456
 457     res->image_end_offset = high_off + cluster_size;
 458     if (size > res->image_end_offset) {
 459         int64_t count;
 460         count = DIV_ROUND_UP(size - res->image_end_offset, cluster_size);
 461         fprintf(stderr, "%s space leaked at the end of the image %" PRId64 "\n",
 462                 fix & BDRV_FIX_LEAKS ? "Repairing" : "ERROR",
 463                 size - res->image_end_offset);
 464         res->leaks += count;
 465         if (fix & BDRV_FIX_LEAKS) {
 466             Error *local_err = NULL;
 467             ret = bdrv_truncate(bs->file, res->image_end_offset,
 468                                 PREALLOC_MODE_OFF, &local_err);
 469             if (ret < 0) {
 470                 error_report_err(local_err);
 471                 res->check_errors++;
 472                 return ret;
 473             }
 474             res->leaks_fixed += count;
 475         }
 476     }
 477
 478     return 0;
 479 }
 480
 481
 482 static int parallels_create(const char *filename, QemuOpts *opts, Error **errp)
 483 {
 484     int64_t total_size, cl_size;
 485     uint8_t tmp[BDRV_SECTOR_SIZE];
 486     Error *local_err = NULL;
 487     BlockBackend *file;
 488     uint32_t bat_entries, bat_sectors;
 489     ParallelsHeader header;
 490     int ret;
 491
 492     total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
 493                           BDRV_SECTOR_SIZE);
 494     cl_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE,
 495                           DEFAULT_CLUSTER_SIZE), BDRV_SECTOR_SIZE);
 496     if (total_size >= MAX_PARALLELS_IMAGE_FACTOR * cl_size) {
 497         error_propagate(errp, local_err);
 498         return -E2BIG;
 499     }
 500
 501     ret = bdrv_create_file(filename, opts, &local_err);
 502     if (ret < 0) {
 503         error_propagate(errp, local_err);
 504         return ret;
 505     }
 506
 507     file = blk_new_open(filename, NULL, NULL,
 508                         BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL,
 509                         &local_err);
 510     if (file == NULL) {
 511         error_propagate(errp, local_err);
 512         return -EIO;
 513     }
 514
 515     blk_set_allow_write_beyond_eof(file, true);
 516
 517     ret = blk_truncate(file, 0, PREALLOC_MODE_OFF, errp);
 518     if (ret < 0) {
 519         goto exit;
 520     }
 521
 522     bat_entries = DIV_ROUND_UP(total_size, cl_size);
 523     bat_sectors = DIV_ROUND_UP(bat_entry_off(bat_entries), cl_size);
 524     bat_sectors = (bat_sectors *  cl_size) >> BDRV_SECTOR_BITS;
 525
 526     memset(&header, 0, sizeof(header));
 527     memcpy(header.magic, HEADER_MAGIC2, sizeof(header.magic));
 528     header.version = cpu_to_le32(HEADER_VERSION);
 529     /* don't care much about geometry, it is not used on image level */
 530     header.heads = cpu_to_le32(16);
 531     header.cylinders = cpu_to_le32(total_size / BDRV_SECTOR_SIZE / 16 / 32);
 532     header.tracks = cpu_to_le32(cl_size >> BDRV_SECTOR_BITS);
 533     header.bat_entries = cpu_to_le32(bat_entries);
 534     header.nb_sectors = cpu_to_le64(DIV_ROUND_UP(total_size, BDRV_SECTOR_SIZE));
 535     header.data_off = cpu_to_le32(bat_sectors);
 536
 537     /* write all the data */
 538     memset(tmp, 0, sizeof(tmp));
 539     memcpy(tmp, &header, sizeof(header));
 540
 541     ret = blk_pwrite(file, 0, tmp, BDRV_SECTOR_SIZE, 0);
 542     if (ret < 0) {
 543         goto exit;
 544     }
 545     ret = blk_pwrite_zeroes(file, BDRV_SECTOR_SIZE,
 546                             (bat_sectors - 1) << BDRV_SECTOR_BITS, 0);
 547     if (ret < 0) {
 548         goto exit;
 549     }
 550     ret = 0;
 551
 552 done:
 553     blk_unref(file);
 554     return ret;
 555
 556 exit:
 557     error_setg_errno(errp, -ret, "Failed to create Parallels image");
 558     goto done;
 559 }
 560
 561
 562 static int parallels_probe(const uint8_t *buf, int buf_size,
 563                            const char *filename)
 564 {
 565     const ParallelsHeader *ph = (const void *)buf;
 566
 567     if (buf_size < sizeof(ParallelsHeader)) {
 568         return 0;
 569     }
 570
 571     if ((!memcmp(ph->magic, HEADER_MAGIC, 16) ||
 572            !memcmp(ph->magic, HEADER_MAGIC2, 16)) &&
 573            (le32_to_cpu(ph->version) == HEADER_VERSION)) {
 574         return 100;
 575     }
 576
 577     return 0;
 578 }
 579
 580 static int parallels_update_header(BlockDriverState *bs)
 581 {
 582     BDRVParallelsState *s = bs->opaque;
 583     unsigned size = MAX(bdrv_opt_mem_align(bs->file->bs),
 584                         sizeof(ParallelsHeader));
 585
 586     if (size > s->header_size) {
 587         size = s->header_size;
 588     }
 589     return bdrv_pwrite_sync(bs->file, 0, s->header, size);
 590 }
 591
 592 static int parallels_open(BlockDriverState *bs, QDict *options, int flags,
 593                           Error **errp)
 594 {
 595     BDRVParallelsState *s = bs->opaque;
 596     ParallelsHeader ph;
 597     int ret, size, i;
 598     QemuOpts *opts = NULL;
 599     Error *local_err = NULL;
 600     char *buf;
 601
 602     bs->file = bdrv_open_child(NULL, options, "file", bs, &child_file,
 603                                false, errp);
 604     if (!bs->file) {
 605         return -EINVAL;
 606     }
 607
 608     ret = bdrv_pread(bs->file, 0, &ph, sizeof(ph));
 609     if (ret < 0) {
 610         goto fail;
 611     }
 612
 613     bs->total_sectors = le64_to_cpu(ph.nb_sectors);
 614
 615     if (le32_to_cpu(ph.version) != HEADER_VERSION) {
 616         goto fail_format;
 617     }
 618     if (!memcmp(ph.magic, HEADER_MAGIC, 16)) {
 619         s->off_multiplier = 1;
 620         bs->total_sectors = 0xffffffff & bs->total_sectors;
 621     } else if (!memcmp(ph.magic, HEADER_MAGIC2, 16)) {
 622         s->off_multiplier = le32_to_cpu(ph.tracks);
 623     } else {
 624         goto fail_format;
 625     }
 626
 627     s->tracks = le32_to_cpu(ph.tracks);
 628     if (s->tracks == 0) {
 629         error_setg(errp, "Invalid image: Zero sectors per track");
 630         ret = -EINVAL;
 631         goto fail;
 632     }
 633     if (s->tracks > INT32_MAX/513) {
 634         error_setg(errp, "Invalid image: Too big cluster");
 635         ret = -EFBIG;
 636         goto fail;
 637     }
 638
 639     s->bat_size = le32_to_cpu(ph.bat_entries);
 640     if (s->bat_size > INT_MAX / sizeof(uint32_t)) {
 641         error_setg(errp, "Catalog too large");
 642         ret = -EFBIG;
 643         goto fail;
 644     }
 645
 646     size = bat_entry_off(s->bat_size);
 647     s->header_size = ROUND_UP(size, bdrv_opt_mem_align(bs->file->bs));
 648     s->header = qemu_try_blockalign(bs->file->bs, s->header_size);
 649     if (s->header == NULL) {
 650         ret = -ENOMEM;
 651         goto fail;
 652     }
 653     s->data_end = le32_to_cpu(ph.data_off);
 654     if (s->data_end == 0) {
 655         s->data_end = ROUND_UP(bat_entry_off(s->bat_size), BDRV_SECTOR_SIZE);
 656     }
 657     if (s->data_end < s->header_size) {
 658         /* there is not enough unused space to fit to block align between BAT
 659            and actual data. We can't avoid read-modify-write... */
 660         s->header_size = size;
 661     }
 662
 663     ret = bdrv_pread(bs->file, 0, s->header, s->header_size);
 664     if (ret < 0) {
 665         goto fail;
 666     }
 667     s->bat_bitmap = (uint32_t *)(s->header + 1);
 668
 669     for (i = 0; i < s->bat_size; i++) {
 670         int64_t off = bat2sect(s, i);
 671         if (off >= s->data_end) {
 672             s->data_end = off + s->tracks;
 673         }
 674     }
 675
 676     if (le32_to_cpu(ph.inuse) == HEADER_INUSE_MAGIC) {
 677         /* Image was not closed correctly. The check is mandatory */
 678         s->header_unclean = true;
 679         if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_CHECK)) {
 680             error_setg(errp, "parallels: Image was not closed correctly; "
 681                        "cannot be opened read/write");
 682             ret = -EACCES;
 683             goto fail;
 684         }
 685     }
 686
 687     opts = qemu_opts_create(&parallels_runtime_opts, NULL, 0, &local_err);
 688     if (local_err != NULL) {
 689         goto fail_options;
 690     }
 691
 692     qemu_opts_absorb_qdict(opts, options, &local_err);
 693     if (local_err != NULL) {
 694         goto fail_options;
 695     }
 696
 697     s->prealloc_size =
 698         qemu_opt_get_size_del(opts, PARALLELS_OPT_PREALLOC_SIZE, 0);
 699     s->prealloc_size = MAX(s->tracks, s->prealloc_size >> BDRV_SECTOR_BITS);
 700     buf = qemu_opt_get_del(opts, PARALLELS_OPT_PREALLOC_MODE);
 701     s->prealloc_mode = qapi_enum_parse(&prealloc_mode_lookup, buf,
 702                                        PRL_PREALLOC_MODE_FALLOCATE,
 703                                        &local_err);
 704     g_free(buf);
 705     if (local_err != NULL) {
 706         goto fail_options;
 707     }
 708
 709     if (!bdrv_has_zero_init(bs->file->bs)) {
 710         s->prealloc_mode = PRL_PREALLOC_MODE_FALLOCATE;
 711     }
 712
 713     if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_INACTIVE)) {
 714         s->header->inuse = cpu_to_le32(HEADER_INUSE_MAGIC);
 715         ret = parallels_update_header(bs);
 716         if (ret < 0) {
 717             goto fail;
 718         }
 719     }
 720
 721     s->bat_dirty_block = 4 * getpagesize();
 722     s->bat_dirty_bmap =
 723         bitmap_new(DIV_ROUND_UP(s->header_size, s->bat_dirty_block));
 724
 725     /* Disable migration until bdrv_invalidate_cache method is added */
 726     error_setg(&s->migration_blocker, "The Parallels format used by node '%s' "
 727                "does not support live migration",
 728                bdrv_get_device_or_node_name(bs));
 729     ret = migrate_add_blocker(s->migration_blocker, &local_err);
 730     if (local_err) {
 731         error_propagate(errp, local_err);
 732         error_free(s->migration_blocker);
 733         goto fail;
 734     }
 735     qemu_co_mutex_init(&s->lock);
 736     return 0;
 737
 738 fail_format:
 739     error_setg(errp, "Image not in Parallels format");
 740     ret = -EINVAL;
 741 fail:
 742     qemu_vfree(s->header);
 743     return ret;
 744
 745 fail_options:
 746     error_propagate(errp, local_err);
 747     ret = -EINVAL;
 748     goto fail;
 749 }
 750
 751
 752 static void parallels_close(BlockDriverState *bs)
 753 {
 754     BDRVParallelsState *s = bs->opaque;
 755
 756     if ((bs->open_flags & BDRV_O_RDWR) && !(bs->open_flags & BDRV_O_INACTIVE)) {
 757         s->header->inuse = 0;
 758         parallels_update_header(bs);
 759         bdrv_truncate(bs->file, s->data_end << BDRV_SECTOR_BITS,
 760                       PREALLOC_MODE_OFF, NULL);
 761     }
 762
 763     g_free(s->bat_dirty_bmap);
 764     qemu_vfree(s->header);
 765
 766     migrate_del_blocker(s->migration_blocker);
 767     error_free(s->migration_blocker);
 768 }
 769
 770 static QemuOptsList parallels_create_opts = {
 771     .name = "parallels-create-opts",
 772     .head = QTAILQ_HEAD_INITIALIZER(parallels_create_opts.head),
 773     .desc = {
 774         {
 775             .name = BLOCK_OPT_SIZE,
 776             .type = QEMU_OPT_SIZE,
 777             .help = "Virtual disk size",
 778         },
 779         {
 780             .name = BLOCK_OPT_CLUSTER_SIZE,
 781             .type = QEMU_OPT_SIZE,
 782             .help = "Parallels image cluster size",
 783             .def_value_str = stringify(DEFAULT_CLUSTER_SIZE),
 784         },
 785         { /* end of list */ }
 786     }
 787 };
 788
 789 static BlockDriver bdrv_parallels = {
 790     .format_name        = "parallels",
 791     .instance_size      = sizeof(BDRVParallelsState),
 792     .bdrv_probe         = parallels_probe,
 793     .bdrv_open          = parallels_open,
 794     .bdrv_close         = parallels_close,
 795     .bdrv_child_perm          = bdrv_format_default_perms,
 796     .bdrv_co_get_block_status = parallels_co_get_block_status,
 797     .bdrv_has_zero_init       = bdrv_has_zero_init_1,
 798     .bdrv_co_flush_to_os      = parallels_co_flush_to_os,
 799     .bdrv_co_readv  = parallels_co_readv,
 800     .bdrv_co_writev = parallels_co_writev,
 801
 802     .bdrv_create    = parallels_create,
 803     .bdrv_check     = parallels_check,
 804     .create_opts    = &parallels_create_opts,
 805 };
 806
 807 static void bdrv_parallels_init(void)
 808 {
 809     bdrv_register(&bdrv_parallels);
 810 }
 811
 812 block_init(bdrv_parallels_init);