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 "block/qdict.h"
  35 #include "sysemu/block-backend.h"
  36 #include "qemu/module.h"
  37 #include "qemu/option.h"
  38 #include "qapi/qmp/qdict.h"
  39 #include "qapi/qobject-input-visitor.h"
  40 #include "qapi/qapi-visit-block-core.h"
  41 #include "qemu/bswap.h"
  42 #include "qemu/bitmap.h"
  43 #include "migration/blocker.h"
  44 #include "parallels.h"
  45
  46 /**************************************************************/
  47
  48 #define HEADER_MAGIC "WithoutFreeSpace"
  49 #define HEADER_MAGIC2 "WithouFreSpacExt"
  50 #define HEADER_VERSION 2
  51 #define HEADER_INUSE_MAGIC  (0x746F6E59)
  52 #define MAX_PARALLELS_IMAGE_FACTOR (1ull << 32)
  53
  54 static QEnumLookup prealloc_mode_lookup = {
  55     .array = (const char *const[]) {
  56         "falloc",
  57         "truncate",
  58     },
  59     .size = PRL_PREALLOC_MODE__MAX
  60 };
  61
  62 #define PARALLELS_OPT_PREALLOC_MODE     "prealloc-mode"
  63 #define PARALLELS_OPT_PREALLOC_SIZE     "prealloc-size"
  64
  65 static QemuOptsList parallels_runtime_opts = {
  66     .name = "parallels",
  67     .head = QTAILQ_HEAD_INITIALIZER(parallels_runtime_opts.head),
  68     .desc = {
  69         {
  70             .name = PARALLELS_OPT_PREALLOC_SIZE,
  71             .type = QEMU_OPT_SIZE,
  72             .help = "Preallocation size on image expansion",
  73             .def_value_str = "128M",
  74         },
  75         {
  76             .name = PARALLELS_OPT_PREALLOC_MODE,
  77             .type = QEMU_OPT_STRING,
  78             .help = "Preallocation mode on image expansion "
  79                     "(allowed values: falloc, truncate)",
  80             .def_value_str = "falloc",
  81         },
  82         { /* end of list */ },
  83     },
  84 };
  85
  86 static QemuOptsList parallels_create_opts = {
  87     .name = "parallels-create-opts",
  88     .head = QTAILQ_HEAD_INITIALIZER(parallels_create_opts.head),
  89     .desc = {
  90         {
  91             .name = BLOCK_OPT_SIZE,
  92             .type = QEMU_OPT_SIZE,
  93             .help = "Virtual disk size",
  94         },
  95         {
  96             .name = BLOCK_OPT_CLUSTER_SIZE,
  97             .type = QEMU_OPT_SIZE,
  98             .help = "Parallels image cluster size",
  99             .def_value_str = stringify(DEFAULT_CLUSTER_SIZE),
 100         },
 101         { /* end of list */ }
 102     }
 103 };
 104
 105
 106 static int64_t bat2sect(BDRVParallelsState *s, uint32_t idx)
 107 {
 108     return (uint64_t)le32_to_cpu(s->bat_bitmap[idx]) * s->off_multiplier;
 109 }
 110
 111 static uint32_t bat_entry_off(uint32_t idx)
 112 {
 113     return sizeof(ParallelsHeader) + sizeof(uint32_t) * idx;
 114 }
 115
 116 static int64_t seek_to_sector(BDRVParallelsState *s, int64_t sector_num)
 117 {
 118     uint32_t index, offset;
 119
 120     index = sector_num / s->tracks;
 121     offset = sector_num % s->tracks;
 122
 123     /* not allocated */
 124     if ((index >= s->bat_size) || (s->bat_bitmap[index] == 0)) {
 125         return -1;
 126     }
 127     return bat2sect(s, index) + offset;
 128 }
 129
 130 static int cluster_remainder(BDRVParallelsState *s, int64_t sector_num,
 131         int nb_sectors)
 132 {
 133     int ret = s->tracks - sector_num % s->tracks;
 134     return MIN(nb_sectors, ret);
 135 }
 136
 137 static int64_t block_status(BDRVParallelsState *s, int64_t sector_num,
 138                             int nb_sectors, int *pnum)
 139 {
 140     int64_t start_off = -2, prev_end_off = -2;
 141
 142     *pnum = 0;
 143     while (nb_sectors > 0 || start_off == -2) {
 144         int64_t offset = seek_to_sector(s, sector_num);
 145         int to_end;
 146
 147         if (start_off == -2) {
 148             start_off = offset;
 149             prev_end_off = offset;
 150         } else if (offset != prev_end_off) {
 151             break;
 152         }
 153
 154         to_end = cluster_remainder(s, sector_num, nb_sectors);
 155         nb_sectors -= to_end;
 156         sector_num += to_end;
 157         *pnum += to_end;
 158
 159         if (offset > 0) {
 160             prev_end_off += to_end;
 161         }
 162     }
 163     return start_off;
 164 }
 165
 166 static int64_t allocate_clusters(BlockDriverState *bs, int64_t sector_num,
 167                                  int nb_sectors, int *pnum)
 168 {
 169     int ret = 0;
 170     BDRVParallelsState *s = bs->opaque;
 171     int64_t pos, space, idx, to_allocate, i, len;
 172
 173     pos = block_status(s, sector_num, nb_sectors, pnum);
 174     if (pos > 0) {
 175         return pos;
 176     }
 177
 178     idx = sector_num / s->tracks;
 179     to_allocate = DIV_ROUND_UP(sector_num + *pnum, s->tracks) - idx;
 180
 181     /* This function is called only by parallels_co_writev(), which will never
 182      * pass a sector_num at or beyond the end of the image (because the block
 183      * layer never passes such a sector_num to that function). Therefore, idx
 184      * is always below s->bat_size.
 185      * block_status() will limit *pnum so that sector_num + *pnum will not
 186      * exceed the image end. Therefore, idx + to_allocate cannot exceed
 187      * s->bat_size.
 188      * Note that s->bat_size is an unsigned int, therefore idx + to_allocate
 189      * will always fit into a uint32_t. */
 190     assert(idx < s->bat_size && idx + to_allocate <= s->bat_size);
 191
 192     space = to_allocate * s->tracks;
 193     len = bdrv_getlength(bs->file->bs);
 194     if (len < 0) {
 195         return len;
 196     }
 197     if (s->data_end + space > (len >> BDRV_SECTOR_BITS)) {
 198         space += s->prealloc_size;
 199         /*
 200          * We require the expanded size to read back as zero. If the
 201          * user permitted truncation, we try that; but if it fails, we
 202          * force the safer-but-slower fallocate.
 203          */
 204         if (s->prealloc_mode == PRL_PREALLOC_MODE_TRUNCATE) {
 205             ret = bdrv_truncate(bs->file,
 206                                 (s->data_end + space) << BDRV_SECTOR_BITS,
 207                                 false, PREALLOC_MODE_OFF, BDRV_REQ_ZERO_WRITE,
 208                                 NULL);
 209             if (ret == -ENOTSUP) {
 210                 s->prealloc_mode = PRL_PREALLOC_MODE_FALLOCATE;
 211             }
 212         }
 213         if (s->prealloc_mode == PRL_PREALLOC_MODE_FALLOCATE) {
 214             ret = bdrv_pwrite_zeroes(bs->file,
 215                                      s->data_end << BDRV_SECTOR_BITS,
 216                                      space << BDRV_SECTOR_BITS, 0);
 217         }
 218         if (ret < 0) {
 219             return ret;
 220         }
 221     }
 222
 223     /* Try to read from backing to fill empty clusters
 224      * FIXME: 1. previous write_zeroes may be redundant
 225      *        2. most of data we read from backing will be rewritten by
 226      *           parallels_co_writev. On aligned-to-cluster write we do not need
 227      *           this read at all.
 228      *        3. it would be good to combine write of data from backing and new
 229      *           data into one write call */
 230     if (bs->backing) {
 231         int64_t nb_cow_sectors = to_allocate * s->tracks;
 232         int64_t nb_cow_bytes = nb_cow_sectors << BDRV_SECTOR_BITS;
 233         void *buf = qemu_blockalign(bs, nb_cow_bytes);
 234
 235         ret = bdrv_co_pread(bs->backing, idx * s->tracks * BDRV_SECTOR_SIZE,
 236                             nb_cow_bytes, buf, 0);
 237         if (ret < 0) {
 238             qemu_vfree(buf);
 239             return ret;
 240         }
 241
 242         ret = bdrv_co_pwritev(bs->file, s->data_end * BDRV_SECTOR_SIZE,
 243                               nb_cow_bytes, buf, 0);
 244         qemu_vfree(buf);
 245         if (ret < 0) {
 246             return ret;
 247         }
 248     }
 249
 250     for (i = 0; i < to_allocate; i++) {
 251         s->bat_bitmap[idx + i] = cpu_to_le32(s->data_end / s->off_multiplier);
 252         s->data_end += s->tracks;
 253         bitmap_set(s->bat_dirty_bmap,
 254                    bat_entry_off(idx + i) / s->bat_dirty_block, 1);
 255     }
 256
 257     return bat2sect(s, idx) + sector_num % s->tracks;
 258 }
 259
 260
 261 static coroutine_fn int parallels_co_flush_to_os(BlockDriverState *bs)
 262 {
 263     BDRVParallelsState *s = bs->opaque;
 264     unsigned long size = DIV_ROUND_UP(s->header_size, s->bat_dirty_block);
 265     unsigned long bit;
 266
 267     qemu_co_mutex_lock(&s->lock);
 268
 269     bit = find_first_bit(s->bat_dirty_bmap, size);
 270     while (bit < size) {
 271         uint32_t off = bit * s->bat_dirty_block;
 272         uint32_t to_write = s->bat_dirty_block;
 273         int ret;
 274
 275         if (off + to_write > s->header_size) {
 276             to_write = s->header_size - off;
 277         }
 278         ret = bdrv_pwrite(bs->file, off, (uint8_t *)s->header + off,
 279                           to_write);
 280         if (ret < 0) {
 281             qemu_co_mutex_unlock(&s->lock);
 282             return ret;
 283         }
 284         bit = find_next_bit(s->bat_dirty_bmap, size, bit + 1);
 285     }
 286     bitmap_zero(s->bat_dirty_bmap, size);
 287
 288     qemu_co_mutex_unlock(&s->lock);
 289     return 0;
 290 }
 291
 292
 293 static int coroutine_fn parallels_co_block_status(BlockDriverState *bs,
 294                                                   bool want_zero,
 295                                                   int64_t offset,
 296                                                   int64_t bytes,
 297                                                   int64_t *pnum,
 298                                                   int64_t *map,
 299                                                   BlockDriverState **file)
 300 {
 301     BDRVParallelsState *s = bs->opaque;
 302     int count;
 303
 304     assert(QEMU_IS_ALIGNED(offset | bytes, BDRV_SECTOR_SIZE));
 305     qemu_co_mutex_lock(&s->lock);
 306     offset = block_status(s, offset >> BDRV_SECTOR_BITS,
 307                           bytes >> BDRV_SECTOR_BITS, &count);
 308     qemu_co_mutex_unlock(&s->lock);
 309
 310     *pnum = count * BDRV_SECTOR_SIZE;
 311     if (offset < 0) {
 312         return 0;
 313     }
 314
 315     *map = offset * BDRV_SECTOR_SIZE;
 316     *file = bs->file->bs;
 317     return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
 318 }
 319
 320 static coroutine_fn int parallels_co_writev(BlockDriverState *bs,
 321                                             int64_t sector_num, int nb_sectors,
 322                                             QEMUIOVector *qiov, int flags)
 323 {
 324     BDRVParallelsState *s = bs->opaque;
 325     uint64_t bytes_done = 0;
 326     QEMUIOVector hd_qiov;
 327     int ret = 0;
 328
 329     assert(!flags);
 330     qemu_iovec_init(&hd_qiov, qiov->niov);
 331
 332     while (nb_sectors > 0) {
 333         int64_t position;
 334         int n, nbytes;
 335
 336         qemu_co_mutex_lock(&s->lock);
 337         position = allocate_clusters(bs, sector_num, nb_sectors, &n);
 338         qemu_co_mutex_unlock(&s->lock);
 339         if (position < 0) {
 340             ret = (int)position;
 341             break;
 342         }
 343
 344         nbytes = n << BDRV_SECTOR_BITS;
 345
 346         qemu_iovec_reset(&hd_qiov);
 347         qemu_iovec_concat(&hd_qiov, qiov, bytes_done, nbytes);
 348
 349         ret = bdrv_co_pwritev(bs->file, position * BDRV_SECTOR_SIZE, nbytes,
 350                               &hd_qiov, 0);
 351         if (ret < 0) {
 352             break;
 353         }
 354
 355         nb_sectors -= n;
 356         sector_num += n;
 357         bytes_done += nbytes;
 358     }
 359
 360     qemu_iovec_destroy(&hd_qiov);
 361     return ret;
 362 }
 363
 364 static coroutine_fn int parallels_co_readv(BlockDriverState *bs,
 365         int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
 366 {
 367     BDRVParallelsState *s = bs->opaque;
 368     uint64_t bytes_done = 0;
 369     QEMUIOVector hd_qiov;
 370     int ret = 0;
 371
 372     qemu_iovec_init(&hd_qiov, qiov->niov);
 373
 374     while (nb_sectors > 0) {
 375         int64_t position;
 376         int n, nbytes;
 377
 378         qemu_co_mutex_lock(&s->lock);
 379         position = block_status(s, sector_num, nb_sectors, &n);
 380         qemu_co_mutex_unlock(&s->lock);
 381
 382         nbytes = n << BDRV_SECTOR_BITS;
 383
 384         qemu_iovec_reset(&hd_qiov);
 385         qemu_iovec_concat(&hd_qiov, qiov, bytes_done, nbytes);
 386
 387         if (position < 0) {
 388             if (bs->backing) {
 389                 ret = bdrv_co_preadv(bs->backing, sector_num * BDRV_SECTOR_SIZE,
 390                                      nbytes, &hd_qiov, 0);
 391                 if (ret < 0) {
 392                     break;
 393                 }
 394             } else {
 395                 qemu_iovec_memset(&hd_qiov, 0, 0, nbytes);
 396             }
 397         } else {
 398             ret = bdrv_co_preadv(bs->file, position * BDRV_SECTOR_SIZE, nbytes,
 399                                  &hd_qiov, 0);
 400             if (ret < 0) {
 401                 break;
 402             }
 403         }
 404
 405         nb_sectors -= n;
 406         sector_num += n;
 407         bytes_done += nbytes;
 408     }
 409
 410     qemu_iovec_destroy(&hd_qiov);
 411     return ret;
 412 }
 413
 414
 415 static int coroutine_fn parallels_co_check(BlockDriverState *bs,
 416                                            BdrvCheckResult *res,
 417                                            BdrvCheckMode fix)
 418 {
 419     BDRVParallelsState *s = bs->opaque;
 420     int64_t size, prev_off, high_off;
 421     int ret;
 422     uint32_t i;
 423     bool flush_bat = false;
 424     int cluster_size = s->tracks << BDRV_SECTOR_BITS;
 425
 426     size = bdrv_getlength(bs->file->bs);
 427     if (size < 0) {
 428         res->check_errors++;
 429         return size;
 430     }
 431
 432     qemu_co_mutex_lock(&s->lock);
 433     if (s->header_unclean) {
 434         fprintf(stderr, "%s image was not closed correctly\n",
 435                 fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR");
 436         res->corruptions++;
 437         if (fix & BDRV_FIX_ERRORS) {
 438             /* parallels_close will do the job right */
 439             res->corruptions_fixed++;
 440             s->header_unclean = false;
 441         }
 442     }
 443
 444     res->bfi.total_clusters = s->bat_size;
 445     res->bfi.compressed_clusters = 0; /* compression is not supported */
 446
 447     high_off = 0;
 448     prev_off = 0;
 449     for (i = 0; i < s->bat_size; i++) {
 450         int64_t off = bat2sect(s, i) << BDRV_SECTOR_BITS;
 451         if (off == 0) {
 452             prev_off = 0;
 453             continue;
 454         }
 455
 456         /* cluster outside the image */
 457         if (off > size) {
 458             fprintf(stderr, "%s cluster %u is outside image\n",
 459                     fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR", i);
 460             res->corruptions++;
 461             if (fix & BDRV_FIX_ERRORS) {
 462                 prev_off = 0;
 463                 s->bat_bitmap[i] = 0;
 464                 res->corruptions_fixed++;
 465                 flush_bat = true;
 466                 continue;
 467             }
 468         }
 469
 470         res->bfi.allocated_clusters++;
 471         if (off > high_off) {
 472             high_off = off;
 473         }
 474
 475         if (prev_off != 0 && (prev_off + cluster_size) != off) {
 476             res->bfi.fragmented_clusters++;
 477         }
 478         prev_off = off;
 479     }
 480
 481     ret = 0;
 482     if (flush_bat) {
 483         ret = bdrv_pwrite_sync(bs->file, 0, s->header, s->header_size);
 484         if (ret < 0) {
 485             res->check_errors++;
 486             goto out;
 487         }
 488     }
 489
 490     res->image_end_offset = high_off + cluster_size;
 491     if (size > res->image_end_offset) {
 492         int64_t count;
 493         count = DIV_ROUND_UP(size - res->image_end_offset, cluster_size);
 494         fprintf(stderr, "%s space leaked at the end of the image %" PRId64 "\n",
 495                 fix & BDRV_FIX_LEAKS ? "Repairing" : "ERROR",
 496                 size - res->image_end_offset);
 497         res->leaks += count;
 498         if (fix & BDRV_FIX_LEAKS) {
 499             Error *local_err = NULL;
 500
 501             /*
 502              * In order to really repair the image, we must shrink it.
 503              * That means we have to pass exact=true.
 504              */
 505             ret = bdrv_truncate(bs->file, res->image_end_offset, true,
 506                                 PREALLOC_MODE_OFF, 0, &local_err);
 507             if (ret < 0) {
 508                 error_report_err(local_err);
 509                 res->check_errors++;
 510                 goto out;
 511             }
 512             res->leaks_fixed += count;
 513         }
 514     }
 515
 516 out:
 517     qemu_co_mutex_unlock(&s->lock);
 518     return ret;
 519 }
 520
 521
 522 static int coroutine_fn parallels_co_create(BlockdevCreateOptions* opts,
 523                                             Error **errp)
 524 {
 525     BlockdevCreateOptionsParallels *parallels_opts;
 526     BlockDriverState *bs;
 527     BlockBackend *blk;
 528     int64_t total_size, cl_size;
 529     uint32_t bat_entries, bat_sectors;
 530     ParallelsHeader header;
 531     uint8_t tmp[BDRV_SECTOR_SIZE];
 532     int ret;
 533
 534     assert(opts->driver == BLOCKDEV_DRIVER_PARALLELS);
 535     parallels_opts = &opts->u.parallels;
 536
 537     /* Sanity checks */
 538     total_size = parallels_opts->size;
 539
 540     if (parallels_opts->has_cluster_size) {
 541         cl_size = parallels_opts->cluster_size;
 542     } else {
 543         cl_size = DEFAULT_CLUSTER_SIZE;
 544     }
 545
 546     /* XXX What is the real limit here? This is an insanely large maximum. */
 547     if (cl_size >= INT64_MAX / MAX_PARALLELS_IMAGE_FACTOR) {
 548         error_setg(errp, "Cluster size is too large");
 549         return -EINVAL;
 550     }
 551     if (total_size >= MAX_PARALLELS_IMAGE_FACTOR * cl_size) {
 552         error_setg(errp, "Image size is too large for this cluster size");
 553         return -E2BIG;
 554     }
 555
 556     if (!QEMU_IS_ALIGNED(total_size, BDRV_SECTOR_SIZE)) {
 557         error_setg(errp, "Image size must be a multiple of 512 bytes");
 558         return -EINVAL;
 559     }
 560
 561     if (!QEMU_IS_ALIGNED(cl_size, BDRV_SECTOR_SIZE)) {
 562         error_setg(errp, "Cluster size must be a multiple of 512 bytes");
 563         return -EINVAL;
 564     }
 565
 566     /* Create BlockBackend to write to the image */
 567     bs = bdrv_open_blockdev_ref(parallels_opts->file, errp);
 568     if (bs == NULL) {
 569         return -EIO;
 570     }
 571
 572     blk = blk_new_with_bs(bs, BLK_PERM_WRITE | BLK_PERM_RESIZE, BLK_PERM_ALL,
 573                           errp);
 574     if (!blk) {
 575         ret = -EPERM;
 576         goto out;
 577     }
 578     blk_set_allow_write_beyond_eof(blk, true);
 579
 580     /* Create image format */
 581     bat_entries = DIV_ROUND_UP(total_size, cl_size);
 582     bat_sectors = DIV_ROUND_UP(bat_entry_off(bat_entries), cl_size);
 583     bat_sectors = (bat_sectors *  cl_size) >> BDRV_SECTOR_BITS;
 584
 585     memset(&header, 0, sizeof(header));
 586     memcpy(header.magic, HEADER_MAGIC2, sizeof(header.magic));
 587     header.version = cpu_to_le32(HEADER_VERSION);
 588     /* don't care much about geometry, it is not used on image level */
 589     header.heads = cpu_to_le32(HEADS_NUMBER);
 590     header.cylinders = cpu_to_le32(total_size / BDRV_SECTOR_SIZE
 591                                    / HEADS_NUMBER / SEC_IN_CYL);
 592     header.tracks = cpu_to_le32(cl_size >> BDRV_SECTOR_BITS);
 593     header.bat_entries = cpu_to_le32(bat_entries);
 594     header.nb_sectors = cpu_to_le64(DIV_ROUND_UP(total_size, BDRV_SECTOR_SIZE));
 595     header.data_off = cpu_to_le32(bat_sectors);
 596
 597     /* write all the data */
 598     memset(tmp, 0, sizeof(tmp));
 599     memcpy(tmp, &header, sizeof(header));
 600
 601     ret = blk_pwrite(blk, 0, tmp, BDRV_SECTOR_SIZE, 0);
 602     if (ret < 0) {
 603         goto exit;
 604     }
 605     ret = blk_pwrite_zeroes(blk, BDRV_SECTOR_SIZE,
 606                             (bat_sectors - 1) << BDRV_SECTOR_BITS, 0);
 607     if (ret < 0) {
 608         goto exit;
 609     }
 610
 611     ret = 0;
 612 out:
 613     blk_unref(blk);
 614     bdrv_unref(bs);
 615     return ret;
 616
 617 exit:
 618     error_setg_errno(errp, -ret, "Failed to create Parallels image");
 619     goto out;
 620 }
 621
 622 static int coroutine_fn parallels_co_create_opts(BlockDriver *drv,
 623                                                  const char *filename,
 624                                                  QemuOpts *opts,
 625                                                  Error **errp)
 626 {
 627     BlockdevCreateOptions *create_options = NULL;
 628     BlockDriverState *bs = NULL;
 629     QDict *qdict;
 630     Visitor *v;
 631     int ret;
 632
 633     static const QDictRenames opt_renames[] = {
 634         { BLOCK_OPT_CLUSTER_SIZE,       "cluster-size" },
 635         { NULL, NULL },
 636     };
 637
 638     /* Parse options and convert legacy syntax */
 639     qdict = qemu_opts_to_qdict_filtered(opts, NULL, &parallels_create_opts,
 640                                         true);
 641
 642     if (!qdict_rename_keys(qdict, opt_renames, errp)) {
 643         ret = -EINVAL;
 644         goto done;
 645     }
 646
 647     /* Create and open the file (protocol layer) */
 648     ret = bdrv_create_file(filename, opts, errp);
 649     if (ret < 0) {
 650         goto done;
 651     }
 652
 653     bs = bdrv_open(filename, NULL, NULL,
 654                    BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp);
 655     if (bs == NULL) {
 656         ret = -EIO;
 657         goto done;
 658     }
 659
 660     /* Now get the QAPI type BlockdevCreateOptions */
 661     qdict_put_str(qdict, "driver", "parallels");
 662     qdict_put_str(qdict, "file", bs->node_name);
 663
 664     v = qobject_input_visitor_new_flat_confused(qdict, errp);
 665     if (!v) {
 666         ret = -EINVAL;
 667         goto done;
 668     }
 669
 670     visit_type_BlockdevCreateOptions(v, NULL, &create_options, errp);
 671     visit_free(v);
 672     if (!create_options) {
 673         ret = -EINVAL;
 674         goto done;
 675     }
 676
 677     /* Silently round up sizes */
 678     create_options->u.parallels.size =
 679         ROUND_UP(create_options->u.parallels.size, BDRV_SECTOR_SIZE);
 680     create_options->u.parallels.cluster_size =
 681         ROUND_UP(create_options->u.parallels.cluster_size, BDRV_SECTOR_SIZE);
 682
 683     /* Create the Parallels image (format layer) */
 684     ret = parallels_co_create(create_options, errp);
 685     if (ret < 0) {
 686         goto done;
 687     }
 688     ret = 0;
 689
 690 done:
 691     qobject_unref(qdict);
 692     bdrv_unref(bs);
 693     qapi_free_BlockdevCreateOptions(create_options);
 694     return ret;
 695 }
 696
 697
 698 static int parallels_probe(const uint8_t *buf, int buf_size,
 699                            const char *filename)
 700 {
 701     const ParallelsHeader *ph = (const void *)buf;
 702
 703     if (buf_size < sizeof(ParallelsHeader)) {
 704         return 0;
 705     }
 706
 707     if ((!memcmp(ph->magic, HEADER_MAGIC, 16) ||
 708            !memcmp(ph->magic, HEADER_MAGIC2, 16)) &&
 709            (le32_to_cpu(ph->version) == HEADER_VERSION)) {
 710         return 100;
 711     }
 712
 713     return 0;
 714 }
 715
 716 static int parallels_update_header(BlockDriverState *bs)
 717 {
 718     BDRVParallelsState *s = bs->opaque;
 719     unsigned size = MAX(bdrv_opt_mem_align(bs->file->bs),
 720                         sizeof(ParallelsHeader));
 721
 722     if (size > s->header_size) {
 723         size = s->header_size;
 724     }
 725     return bdrv_pwrite_sync(bs->file, 0, s->header, size);
 726 }
 727
 728 static int parallels_open(BlockDriverState *bs, QDict *options, int flags,
 729                           Error **errp)
 730 {
 731     BDRVParallelsState *s = bs->opaque;
 732     ParallelsHeader ph;
 733     int ret, size, i;
 734     QemuOpts *opts = NULL;
 735     Error *local_err = NULL;
 736     char *buf;
 737
 738     bs->file = bdrv_open_child(NULL, options, "file", bs, &child_of_bds,
 739                                BDRV_CHILD_IMAGE, false, errp);
 740     if (!bs->file) {
 741         return -EINVAL;
 742     }
 743
 744     ret = bdrv_pread(bs->file, 0, &ph, sizeof(ph));
 745     if (ret < 0) {
 746         goto fail;
 747     }
 748
 749     bs->total_sectors = le64_to_cpu(ph.nb_sectors);
 750
 751     if (le32_to_cpu(ph.version) != HEADER_VERSION) {
 752         goto fail_format;
 753     }
 754     if (!memcmp(ph.magic, HEADER_MAGIC, 16)) {
 755         s->off_multiplier = 1;
 756         bs->total_sectors = 0xffffffff & bs->total_sectors;
 757     } else if (!memcmp(ph.magic, HEADER_MAGIC2, 16)) {
 758         s->off_multiplier = le32_to_cpu(ph.tracks);
 759     } else {
 760         goto fail_format;
 761     }
 762
 763     s->tracks = le32_to_cpu(ph.tracks);
 764     if (s->tracks == 0) {
 765         error_setg(errp, "Invalid image: Zero sectors per track");
 766         ret = -EINVAL;
 767         goto fail;
 768     }
 769     if (s->tracks > INT32_MAX/513) {
 770         error_setg(errp, "Invalid image: Too big cluster");
 771         ret = -EFBIG;
 772         goto fail;
 773     }
 774
 775     s->bat_size = le32_to_cpu(ph.bat_entries);
 776     if (s->bat_size > INT_MAX / sizeof(uint32_t)) {
 777         error_setg(errp, "Catalog too large");
 778         ret = -EFBIG;
 779         goto fail;
 780     }
 781
 782     size = bat_entry_off(s->bat_size);
 783     s->header_size = ROUND_UP(size, bdrv_opt_mem_align(bs->file->bs));
 784     s->header = qemu_try_blockalign(bs->file->bs, s->header_size);
 785     if (s->header == NULL) {
 786         ret = -ENOMEM;
 787         goto fail;
 788     }
 789     s->data_end = le32_to_cpu(ph.data_off);
 790     if (s->data_end == 0) {
 791         s->data_end = ROUND_UP(bat_entry_off(s->bat_size), BDRV_SECTOR_SIZE);
 792     }
 793     if (s->data_end < s->header_size) {
 794         /* there is not enough unused space to fit to block align between BAT
 795            and actual data. We can't avoid read-modify-write... */
 796         s->header_size = size;
 797     }
 798
 799     ret = bdrv_pread(bs->file, 0, s->header, s->header_size);
 800     if (ret < 0) {
 801         goto fail;
 802     }
 803     s->bat_bitmap = (uint32_t *)(s->header + 1);
 804
 805     for (i = 0; i < s->bat_size; i++) {
 806         int64_t off = bat2sect(s, i);
 807         if (off >= s->data_end) {
 808             s->data_end = off + s->tracks;
 809         }
 810     }
 811
 812     if (le32_to_cpu(ph.inuse) == HEADER_INUSE_MAGIC) {
 813         /* Image was not closed correctly. The check is mandatory */
 814         s->header_unclean = true;
 815         if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_CHECK)) {
 816             error_setg(errp, "parallels: Image was not closed correctly; "
 817                        "cannot be opened read/write");
 818             ret = -EACCES;
 819             goto fail;
 820         }
 821     }
 822
 823     opts = qemu_opts_create(&parallels_runtime_opts, NULL, 0, errp);
 824     if (!opts) {
 825         goto fail_options;
 826     }
 827
 828     if (!qemu_opts_absorb_qdict(opts, options, errp)) {
 829         goto fail_options;
 830     }
 831
 832     s->prealloc_size =
 833         qemu_opt_get_size_del(opts, PARALLELS_OPT_PREALLOC_SIZE, 0);
 834     s->prealloc_size = MAX(s->tracks, s->prealloc_size >> BDRV_SECTOR_BITS);
 835     buf = qemu_opt_get_del(opts, PARALLELS_OPT_PREALLOC_MODE);
 836     /* prealloc_mode can be downgraded later during allocate_clusters */
 837     s->prealloc_mode = qapi_enum_parse(&prealloc_mode_lookup, buf,
 838                                        PRL_PREALLOC_MODE_FALLOCATE,
 839                                        &local_err);
 840     g_free(buf);
 841     if (local_err != NULL) {
 842         error_propagate(errp, local_err);
 843         goto fail_options;
 844     }
 845
 846     if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_INACTIVE)) {
 847         s->header->inuse = cpu_to_le32(HEADER_INUSE_MAGIC);
 848         ret = parallels_update_header(bs);
 849         if (ret < 0) {
 850             goto fail;
 851         }
 852     }
 853
 854     s->bat_dirty_block = 4 * qemu_real_host_page_size;
 855     s->bat_dirty_bmap =
 856         bitmap_new(DIV_ROUND_UP(s->header_size, s->bat_dirty_block));
 857
 858     /* Disable migration until bdrv_invalidate_cache method is added */
 859     error_setg(&s->migration_blocker, "The Parallels format used by node '%s' "
 860                "does not support live migration",
 861                bdrv_get_device_or_node_name(bs));
 862     ret = migrate_add_blocker(s->migration_blocker, errp);
 863     if (ret < 0) {
 864         error_free(s->migration_blocker);
 865         goto fail;
 866     }
 867     qemu_co_mutex_init(&s->lock);
 868     return 0;
 869
 870 fail_format:
 871     error_setg(errp, "Image not in Parallels format");
 872 fail_options:
 873     ret = -EINVAL;
 874 fail:
 875     qemu_vfree(s->header);
 876     return ret;
 877 }
 878
 879
 880 static void parallels_close(BlockDriverState *bs)
 881 {
 882     BDRVParallelsState *s = bs->opaque;
 883
 884     if ((bs->open_flags & BDRV_O_RDWR) && !(bs->open_flags & BDRV_O_INACTIVE)) {
 885         s->header->inuse = 0;
 886         parallels_update_header(bs);
 887
 888         /* errors are ignored, so we might as well pass exact=true */
 889         bdrv_truncate(bs->file, s->data_end << BDRV_SECTOR_BITS, true,
 890                       PREALLOC_MODE_OFF, 0, NULL);
 891     }
 892
 893     g_free(s->bat_dirty_bmap);
 894     qemu_vfree(s->header);
 895
 896     migrate_del_blocker(s->migration_blocker);
 897     error_free(s->migration_blocker);
 898 }
 899
 900 static BlockDriver bdrv_parallels = {
 901     .format_name        = "parallels",
 902     .instance_size      = sizeof(BDRVParallelsState),
 903     .bdrv_probe         = parallels_probe,
 904     .bdrv_open          = parallels_open,
 905     .bdrv_close         = parallels_close,
 906     .bdrv_child_perm          = bdrv_default_perms,
 907     .bdrv_co_block_status     = parallels_co_block_status,
 908     .bdrv_has_zero_init       = bdrv_has_zero_init_1,
 909     .bdrv_co_flush_to_os      = parallels_co_flush_to_os,
 910     .bdrv_co_readv  = parallels_co_readv,
 911     .bdrv_co_writev = parallels_co_writev,
 912     .is_format      = true,
 913     .supports_backing = true,
 914     .bdrv_co_create      = parallels_co_create,
 915     .bdrv_co_create_opts = parallels_co_create_opts,
 916     .bdrv_co_check  = parallels_co_check,
 917     .create_opts    = &parallels_create_opts,
 918 };
 919
 920 static void bdrv_parallels_init(void)
 921 {
 922     bdrv_register(&bdrv_parallels);
 923 }
 924
 925 block_init(bdrv_parallels_init);