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;
 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         if (s->prealloc_mode == PRL_PREALLOC_MODE_FALLOCATE) {
 200             ret = bdrv_pwrite_zeroes(bs->file,
 201                                      s->data_end << BDRV_SECTOR_BITS,
 202                                      space << BDRV_SECTOR_BITS, 0);
 203         } else {
 204             ret = bdrv_truncate(bs->file,
 205                                 (s->data_end + space) << BDRV_SECTOR_BITS,
 206                                 PREALLOC_MODE_OFF, NULL);
 207         }
 208         if (ret < 0) {
 209             return ret;
 210         }
 211     }
 212
 213     /* Try to read from backing to fill empty clusters
 214      * FIXME: 1. previous write_zeroes may be redundant
 215      *        2. most of data we read from backing will be rewritten by
 216      *           parallels_co_writev. On aligned-to-cluster write we do not need
 217      *           this read at all.
 218      *        3. it would be good to combine write of data from backing and new
 219      *           data into one write call */
 220     if (bs->backing) {
 221         int64_t nb_cow_sectors = to_allocate * s->tracks;
 222         int64_t nb_cow_bytes = nb_cow_sectors << BDRV_SECTOR_BITS;
 223         QEMUIOVector qiov;
 224         struct iovec iov = {
 225             .iov_len = nb_cow_bytes,
 226             .iov_base = qemu_blockalign(bs, nb_cow_bytes)
 227         };
 228         qemu_iovec_init_external(&qiov, &iov, 1);
 229
 230         ret = bdrv_co_preadv(bs->backing, idx * s->tracks * BDRV_SECTOR_SIZE,
 231                              nb_cow_bytes, &qiov, 0);
 232         if (ret < 0) {
 233             qemu_vfree(iov.iov_base);
 234             return ret;
 235         }
 236
 237         ret = bdrv_co_pwritev(bs->file, s->data_end * BDRV_SECTOR_SIZE,
 238                               nb_cow_bytes, &qiov, 0);
 239         qemu_vfree(iov.iov_base);
 240         if (ret < 0) {
 241             return ret;
 242         }
 243     }
 244
 245     for (i = 0; i < to_allocate; i++) {
 246         s->bat_bitmap[idx + i] = cpu_to_le32(s->data_end / s->off_multiplier);
 247         s->data_end += s->tracks;
 248         bitmap_set(s->bat_dirty_bmap,
 249                    bat_entry_off(idx + i) / s->bat_dirty_block, 1);
 250     }
 251
 252     return bat2sect(s, idx) + sector_num % s->tracks;
 253 }
 254
 255
 256 static coroutine_fn int parallels_co_flush_to_os(BlockDriverState *bs)
 257 {
 258     BDRVParallelsState *s = bs->opaque;
 259     unsigned long size = DIV_ROUND_UP(s->header_size, s->bat_dirty_block);
 260     unsigned long bit;
 261
 262     qemu_co_mutex_lock(&s->lock);
 263
 264     bit = find_first_bit(s->bat_dirty_bmap, size);
 265     while (bit < size) {
 266         uint32_t off = bit * s->bat_dirty_block;
 267         uint32_t to_write = s->bat_dirty_block;
 268         int ret;
 269
 270         if (off + to_write > s->header_size) {
 271             to_write = s->header_size - off;
 272         }
 273         ret = bdrv_pwrite(bs->file, off, (uint8_t *)s->header + off,
 274                           to_write);
 275         if (ret < 0) {
 276             qemu_co_mutex_unlock(&s->lock);
 277             return ret;
 278         }
 279         bit = find_next_bit(s->bat_dirty_bmap, size, bit + 1);
 280     }
 281     bitmap_zero(s->bat_dirty_bmap, size);
 282
 283     qemu_co_mutex_unlock(&s->lock);
 284     return 0;
 285 }
 286
 287
 288 static int coroutine_fn parallels_co_block_status(BlockDriverState *bs,
 289                                                   bool want_zero,
 290                                                   int64_t offset,
 291                                                   int64_t bytes,
 292                                                   int64_t *pnum,
 293                                                   int64_t *map,
 294                                                   BlockDriverState **file)
 295 {
 296     BDRVParallelsState *s = bs->opaque;
 297     int count;
 298
 299     assert(QEMU_IS_ALIGNED(offset | bytes, BDRV_SECTOR_SIZE));
 300     qemu_co_mutex_lock(&s->lock);
 301     offset = block_status(s, offset >> BDRV_SECTOR_BITS,
 302                           bytes >> BDRV_SECTOR_BITS, &count);
 303     qemu_co_mutex_unlock(&s->lock);
 304
 305     *pnum = count * BDRV_SECTOR_SIZE;
 306     if (offset < 0) {
 307         return 0;
 308     }
 309
 310     *map = offset * BDRV_SECTOR_SIZE;
 311     *file = bs->file->bs;
 312     return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
 313 }
 314
 315 static coroutine_fn int parallels_co_writev(BlockDriverState *bs,
 316                                             int64_t sector_num, int nb_sectors,
 317                                             QEMUIOVector *qiov, int flags)
 318 {
 319     BDRVParallelsState *s = bs->opaque;
 320     uint64_t bytes_done = 0;
 321     QEMUIOVector hd_qiov;
 322     int ret = 0;
 323
 324     assert(!flags);
 325     qemu_iovec_init(&hd_qiov, qiov->niov);
 326
 327     while (nb_sectors > 0) {
 328         int64_t position;
 329         int n, nbytes;
 330
 331         qemu_co_mutex_lock(&s->lock);
 332         position = allocate_clusters(bs, sector_num, nb_sectors, &n);
 333         qemu_co_mutex_unlock(&s->lock);
 334         if (position < 0) {
 335             ret = (int)position;
 336             break;
 337         }
 338
 339         nbytes = n << BDRV_SECTOR_BITS;
 340
 341         qemu_iovec_reset(&hd_qiov);
 342         qemu_iovec_concat(&hd_qiov, qiov, bytes_done, nbytes);
 343
 344         ret = bdrv_co_pwritev(bs->file, position * BDRV_SECTOR_SIZE, nbytes,
 345                               &hd_qiov, 0);
 346         if (ret < 0) {
 347             break;
 348         }
 349
 350         nb_sectors -= n;
 351         sector_num += n;
 352         bytes_done += nbytes;
 353     }
 354
 355     qemu_iovec_destroy(&hd_qiov);
 356     return ret;
 357 }
 358
 359 static coroutine_fn int parallels_co_readv(BlockDriverState *bs,
 360         int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
 361 {
 362     BDRVParallelsState *s = bs->opaque;
 363     uint64_t bytes_done = 0;
 364     QEMUIOVector hd_qiov;
 365     int ret = 0;
 366
 367     qemu_iovec_init(&hd_qiov, qiov->niov);
 368
 369     while (nb_sectors > 0) {
 370         int64_t position;
 371         int n, nbytes;
 372
 373         qemu_co_mutex_lock(&s->lock);
 374         position = block_status(s, sector_num, nb_sectors, &n);
 375         qemu_co_mutex_unlock(&s->lock);
 376
 377         nbytes = n << BDRV_SECTOR_BITS;
 378
 379         qemu_iovec_reset(&hd_qiov);
 380         qemu_iovec_concat(&hd_qiov, qiov, bytes_done, nbytes);
 381
 382         if (position < 0) {
 383             if (bs->backing) {
 384                 ret = bdrv_co_preadv(bs->backing, sector_num * BDRV_SECTOR_SIZE,
 385                                      nbytes, &hd_qiov, 0);
 386                 if (ret < 0) {
 387                     break;
 388                 }
 389             } else {
 390                 qemu_iovec_memset(&hd_qiov, 0, 0, nbytes);
 391             }
 392         } else {
 393             ret = bdrv_co_preadv(bs->file, position * BDRV_SECTOR_SIZE, nbytes,
 394                                  &hd_qiov, 0);
 395             if (ret < 0) {
 396                 break;
 397             }
 398         }
 399
 400         nb_sectors -= n;
 401         sector_num += n;
 402         bytes_done += nbytes;
 403     }
 404
 405     qemu_iovec_destroy(&hd_qiov);
 406     return ret;
 407 }
 408
 409
 410 static int coroutine_fn parallels_co_check(BlockDriverState *bs,
 411                                            BdrvCheckResult *res,
 412                                            BdrvCheckMode fix)
 413 {
 414     BDRVParallelsState *s = bs->opaque;
 415     int64_t size, prev_off, high_off;
 416     int ret;
 417     uint32_t i;
 418     bool flush_bat = false;
 419     int cluster_size = s->tracks << BDRV_SECTOR_BITS;
 420
 421     size = bdrv_getlength(bs->file->bs);
 422     if (size < 0) {
 423         res->check_errors++;
 424         return size;
 425     }
 426
 427     qemu_co_mutex_lock(&s->lock);
 428     if (s->header_unclean) {
 429         fprintf(stderr, "%s image was not closed correctly\n",
 430                 fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR");
 431         res->corruptions++;
 432         if (fix & BDRV_FIX_ERRORS) {
 433             /* parallels_close will do the job right */
 434             res->corruptions_fixed++;
 435             s->header_unclean = false;
 436         }
 437     }
 438
 439     res->bfi.total_clusters = s->bat_size;
 440     res->bfi.compressed_clusters = 0; /* compression is not supported */
 441
 442     high_off = 0;
 443     prev_off = 0;
 444     for (i = 0; i < s->bat_size; i++) {
 445         int64_t off = bat2sect(s, i) << BDRV_SECTOR_BITS;
 446         if (off == 0) {
 447             prev_off = 0;
 448             continue;
 449         }
 450
 451         /* cluster outside the image */
 452         if (off > size) {
 453             fprintf(stderr, "%s cluster %u is outside image\n",
 454                     fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR", i);
 455             res->corruptions++;
 456             if (fix & BDRV_FIX_ERRORS) {
 457                 prev_off = 0;
 458                 s->bat_bitmap[i] = 0;
 459                 res->corruptions_fixed++;
 460                 flush_bat = true;
 461                 continue;
 462             }
 463         }
 464
 465         res->bfi.allocated_clusters++;
 466         if (off > high_off) {
 467             high_off = off;
 468         }
 469
 470         if (prev_off != 0 && (prev_off + cluster_size) != off) {
 471             res->bfi.fragmented_clusters++;
 472         }
 473         prev_off = off;
 474     }
 475
 476     ret = 0;
 477     if (flush_bat) {
 478         ret = bdrv_pwrite_sync(bs->file, 0, s->header, s->header_size);
 479         if (ret < 0) {
 480             res->check_errors++;
 481             goto out;
 482         }
 483     }
 484
 485     res->image_end_offset = high_off + cluster_size;
 486     if (size > res->image_end_offset) {
 487         int64_t count;
 488         count = DIV_ROUND_UP(size - res->image_end_offset, cluster_size);
 489         fprintf(stderr, "%s space leaked at the end of the image %" PRId64 "\n",
 490                 fix & BDRV_FIX_LEAKS ? "Repairing" : "ERROR",
 491                 size - res->image_end_offset);
 492         res->leaks += count;
 493         if (fix & BDRV_FIX_LEAKS) {
 494             Error *local_err = NULL;
 495             ret = bdrv_truncate(bs->file, res->image_end_offset,
 496                                 PREALLOC_MODE_OFF, &local_err);
 497             if (ret < 0) {
 498                 error_report_err(local_err);
 499                 res->check_errors++;
 500                 goto out;
 501             }
 502             res->leaks_fixed += count;
 503         }
 504     }
 505
 506 out:
 507     qemu_co_mutex_unlock(&s->lock);
 508     return ret;
 509 }
 510
 511
 512 static int coroutine_fn parallels_co_create(BlockdevCreateOptions* opts,
 513                                             Error **errp)
 514 {
 515     BlockdevCreateOptionsParallels *parallels_opts;
 516     BlockDriverState *bs;
 517     BlockBackend *blk;
 518     int64_t total_size, cl_size;
 519     uint32_t bat_entries, bat_sectors;
 520     ParallelsHeader header;
 521     uint8_t tmp[BDRV_SECTOR_SIZE];
 522     int ret;
 523
 524     assert(opts->driver == BLOCKDEV_DRIVER_PARALLELS);
 525     parallels_opts = &opts->u.parallels;
 526
 527     /* Sanity checks */
 528     total_size = parallels_opts->size;
 529
 530     if (parallels_opts->has_cluster_size) {
 531         cl_size = parallels_opts->cluster_size;
 532     } else {
 533         cl_size = DEFAULT_CLUSTER_SIZE;
 534     }
 535
 536     /* XXX What is the real limit here? This is an insanely large maximum. */
 537     if (cl_size >= INT64_MAX / MAX_PARALLELS_IMAGE_FACTOR) {
 538         error_setg(errp, "Cluster size is too large");
 539         return -EINVAL;
 540     }
 541     if (total_size >= MAX_PARALLELS_IMAGE_FACTOR * cl_size) {
 542         error_setg(errp, "Image size is too large for this cluster size");
 543         return -E2BIG;
 544     }
 545
 546     if (!QEMU_IS_ALIGNED(total_size, BDRV_SECTOR_SIZE)) {
 547         error_setg(errp, "Image size must be a multiple of 512 bytes");
 548         return -EINVAL;
 549     }
 550
 551     if (!QEMU_IS_ALIGNED(cl_size, BDRV_SECTOR_SIZE)) {
 552         error_setg(errp, "Cluster size must be a multiple of 512 bytes");
 553         return -EINVAL;
 554     }
 555
 556     /* Create BlockBackend to write to the image */
 557     bs = bdrv_open_blockdev_ref(parallels_opts->file, errp);
 558     if (bs == NULL) {
 559         return -EIO;
 560     }
 561
 562     blk = blk_new(BLK_PERM_WRITE | BLK_PERM_RESIZE, BLK_PERM_ALL);
 563     ret = blk_insert_bs(blk, bs, errp);
 564     if (ret < 0) {
 565         goto out;
 566     }
 567     blk_set_allow_write_beyond_eof(blk, true);
 568
 569     /* Create image format */
 570     ret = blk_truncate(blk, 0, PREALLOC_MODE_OFF, errp);
 571     if (ret < 0) {
 572         goto out;
 573     }
 574
 575     bat_entries = DIV_ROUND_UP(total_size, cl_size);
 576     bat_sectors = DIV_ROUND_UP(bat_entry_off(bat_entries), cl_size);
 577     bat_sectors = (bat_sectors *  cl_size) >> BDRV_SECTOR_BITS;
 578
 579     memset(&header, 0, sizeof(header));
 580     memcpy(header.magic, HEADER_MAGIC2, sizeof(header.magic));
 581     header.version = cpu_to_le32(HEADER_VERSION);
 582     /* don't care much about geometry, it is not used on image level */
 583     header.heads = cpu_to_le32(HEADS_NUMBER);
 584     header.cylinders = cpu_to_le32(total_size / BDRV_SECTOR_SIZE
 585                                    / HEADS_NUMBER / SEC_IN_CYL);
 586     header.tracks = cpu_to_le32(cl_size >> BDRV_SECTOR_BITS);
 587     header.bat_entries = cpu_to_le32(bat_entries);
 588     header.nb_sectors = cpu_to_le64(DIV_ROUND_UP(total_size, BDRV_SECTOR_SIZE));
 589     header.data_off = cpu_to_le32(bat_sectors);
 590
 591     /* write all the data */
 592     memset(tmp, 0, sizeof(tmp));
 593     memcpy(tmp, &header, sizeof(header));
 594
 595     ret = blk_pwrite(blk, 0, tmp, BDRV_SECTOR_SIZE, 0);
 596     if (ret < 0) {
 597         goto exit;
 598     }
 599     ret = blk_pwrite_zeroes(blk, BDRV_SECTOR_SIZE,
 600                             (bat_sectors - 1) << BDRV_SECTOR_BITS, 0);
 601     if (ret < 0) {
 602         goto exit;
 603     }
 604
 605     ret = 0;
 606 out:
 607     blk_unref(blk);
 608     bdrv_unref(bs);
 609     return ret;
 610
 611 exit:
 612     error_setg_errno(errp, -ret, "Failed to create Parallels image");
 613     goto out;
 614 }
 615
 616 static int coroutine_fn parallels_co_create_opts(const char *filename,
 617                                                  QemuOpts *opts,
 618                                                  Error **errp)
 619 {
 620     BlockdevCreateOptions *create_options = NULL;
 621     Error *local_err = NULL;
 622     BlockDriverState *bs = NULL;
 623     QDict *qdict;
 624     Visitor *v;
 625     int ret;
 626
 627     static const QDictRenames opt_renames[] = {
 628         { BLOCK_OPT_CLUSTER_SIZE,       "cluster-size" },
 629         { NULL, NULL },
 630     };
 631
 632     /* Parse options and convert legacy syntax */
 633     qdict = qemu_opts_to_qdict_filtered(opts, NULL, &parallels_create_opts,
 634                                         true);
 635
 636     if (!qdict_rename_keys(qdict, opt_renames, errp)) {
 637         ret = -EINVAL;
 638         goto done;
 639     }
 640
 641     /* Create and open the file (protocol layer) */
 642     ret = bdrv_create_file(filename, opts, &local_err);
 643     if (ret < 0) {
 644         error_propagate(errp, local_err);
 645         goto done;
 646     }
 647
 648     bs = bdrv_open(filename, NULL, NULL,
 649                    BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp);
 650     if (bs == NULL) {
 651         ret = -EIO;
 652         goto done;
 653     }
 654
 655     /* Now get the QAPI type BlockdevCreateOptions */
 656     qdict_put_str(qdict, "driver", "parallels");
 657     qdict_put_str(qdict, "file", bs->node_name);
 658
 659     v = qobject_input_visitor_new_flat_confused(qdict, errp);
 660     if (!v) {
 661         ret = -EINVAL;
 662         goto done;
 663     }
 664
 665     visit_type_BlockdevCreateOptions(v, NULL, &create_options, &local_err);
 666     visit_free(v);
 667
 668     if (local_err) {
 669         error_propagate(errp, local_err);
 670         ret = -EINVAL;
 671         goto done;
 672     }
 673
 674     /* Silently round up sizes */
 675     create_options->u.parallels.size =
 676         ROUND_UP(create_options->u.parallels.size, BDRV_SECTOR_SIZE);
 677     create_options->u.parallels.cluster_size =
 678         ROUND_UP(create_options->u.parallels.cluster_size, BDRV_SECTOR_SIZE);
 679
 680     /* Create the Parallels image (format layer) */
 681     ret = parallels_co_create(create_options, errp);
 682     if (ret < 0) {
 683         goto done;
 684     }
 685     ret = 0;
 686
 687 done:
 688     qobject_unref(qdict);
 689     bdrv_unref(bs);
 690     qapi_free_BlockdevCreateOptions(create_options);
 691     return ret;
 692 }
 693
 694
 695 static int parallels_probe(const uint8_t *buf, int buf_size,
 696                            const char *filename)
 697 {
 698     const ParallelsHeader *ph = (const void *)buf;
 699
 700     if (buf_size < sizeof(ParallelsHeader)) {
 701         return 0;
 702     }
 703
 704     if ((!memcmp(ph->magic, HEADER_MAGIC, 16) ||
 705            !memcmp(ph->magic, HEADER_MAGIC2, 16)) &&
 706            (le32_to_cpu(ph->version) == HEADER_VERSION)) {
 707         return 100;
 708     }
 709
 710     return 0;
 711 }
 712
 713 static int parallels_update_header(BlockDriverState *bs)
 714 {
 715     BDRVParallelsState *s = bs->opaque;
 716     unsigned size = MAX(bdrv_opt_mem_align(bs->file->bs),
 717                         sizeof(ParallelsHeader));
 718
 719     if (size > s->header_size) {
 720         size = s->header_size;
 721     }
 722     return bdrv_pwrite_sync(bs->file, 0, s->header, size);
 723 }
 724
 725 static int parallels_open(BlockDriverState *bs, QDict *options, int flags,
 726                           Error **errp)
 727 {
 728     BDRVParallelsState *s = bs->opaque;
 729     ParallelsHeader ph;
 730     int ret, size, i;
 731     QemuOpts *opts = NULL;
 732     Error *local_err = NULL;
 733     char *buf;
 734
 735     bs->file = bdrv_open_child(NULL, options, "file", bs, &child_file,
 736                                false, errp);
 737     if (!bs->file) {
 738         return -EINVAL;
 739     }
 740
 741     ret = bdrv_pread(bs->file, 0, &ph, sizeof(ph));
 742     if (ret < 0) {
 743         goto fail;
 744     }
 745
 746     bs->total_sectors = le64_to_cpu(ph.nb_sectors);
 747
 748     if (le32_to_cpu(ph.version) != HEADER_VERSION) {
 749         goto fail_format;
 750     }
 751     if (!memcmp(ph.magic, HEADER_MAGIC, 16)) {
 752         s->off_multiplier = 1;
 753         bs->total_sectors = 0xffffffff & bs->total_sectors;
 754     } else if (!memcmp(ph.magic, HEADER_MAGIC2, 16)) {
 755         s->off_multiplier = le32_to_cpu(ph.tracks);
 756     } else {
 757         goto fail_format;
 758     }
 759
 760     s->tracks = le32_to_cpu(ph.tracks);
 761     if (s->tracks == 0) {
 762         error_setg(errp, "Invalid image: Zero sectors per track");
 763         ret = -EINVAL;
 764         goto fail;
 765     }
 766     if (s->tracks > INT32_MAX/513) {
 767         error_setg(errp, "Invalid image: Too big cluster");
 768         ret = -EFBIG;
 769         goto fail;
 770     }
 771
 772     s->bat_size = le32_to_cpu(ph.bat_entries);
 773     if (s->bat_size > INT_MAX / sizeof(uint32_t)) {
 774         error_setg(errp, "Catalog too large");
 775         ret = -EFBIG;
 776         goto fail;
 777     }
 778
 779     size = bat_entry_off(s->bat_size);
 780     s->header_size = ROUND_UP(size, bdrv_opt_mem_align(bs->file->bs));
 781     s->header = qemu_try_blockalign(bs->file->bs, s->header_size);
 782     if (s->header == NULL) {
 783         ret = -ENOMEM;
 784         goto fail;
 785     }
 786     s->data_end = le32_to_cpu(ph.data_off);
 787     if (s->data_end == 0) {
 788         s->data_end = ROUND_UP(bat_entry_off(s->bat_size), BDRV_SECTOR_SIZE);
 789     }
 790     if (s->data_end < s->header_size) {
 791         /* there is not enough unused space to fit to block align between BAT
 792            and actual data. We can't avoid read-modify-write... */
 793         s->header_size = size;
 794     }
 795
 796     ret = bdrv_pread(bs->file, 0, s->header, s->header_size);
 797     if (ret < 0) {
 798         goto fail;
 799     }
 800     s->bat_bitmap = (uint32_t *)(s->header + 1);
 801
 802     for (i = 0; i < s->bat_size; i++) {
 803         int64_t off = bat2sect(s, i);
 804         if (off >= s->data_end) {
 805             s->data_end = off + s->tracks;
 806         }
 807     }
 808
 809     if (le32_to_cpu(ph.inuse) == HEADER_INUSE_MAGIC) {
 810         /* Image was not closed correctly. The check is mandatory */
 811         s->header_unclean = true;
 812         if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_CHECK)) {
 813             error_setg(errp, "parallels: Image was not closed correctly; "
 814                        "cannot be opened read/write");
 815             ret = -EACCES;
 816             goto fail;
 817         }
 818     }
 819
 820     opts = qemu_opts_create(&parallels_runtime_opts, NULL, 0, &local_err);
 821     if (local_err != NULL) {
 822         goto fail_options;
 823     }
 824
 825     qemu_opts_absorb_qdict(opts, options, &local_err);
 826     if (local_err != NULL) {
 827         goto fail_options;
 828     }
 829
 830     s->prealloc_size =
 831         qemu_opt_get_size_del(opts, PARALLELS_OPT_PREALLOC_SIZE, 0);
 832     s->prealloc_size = MAX(s->tracks, s->prealloc_size >> BDRV_SECTOR_BITS);
 833     buf = qemu_opt_get_del(opts, PARALLELS_OPT_PREALLOC_MODE);
 834     s->prealloc_mode = qapi_enum_parse(&prealloc_mode_lookup, buf,
 835                                        PRL_PREALLOC_MODE_FALLOCATE,
 836                                        &local_err);
 837     g_free(buf);
 838     if (local_err != NULL) {
 839         goto fail_options;
 840     }
 841
 842     if (!bdrv_has_zero_init(bs->file->bs)) {
 843         s->prealloc_mode = PRL_PREALLOC_MODE_FALLOCATE;
 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 * getpagesize();
 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, &local_err);
 863     if (local_err) {
 864         error_propagate(errp, local_err);
 865         error_free(s->migration_blocker);
 866         goto fail;
 867     }
 868     qemu_co_mutex_init(&s->lock);
 869     return 0;
 870
 871 fail_format:
 872     error_setg(errp, "Image not in Parallels format");
 873     ret = -EINVAL;
 874 fail:
 875     qemu_vfree(s->header);
 876     return ret;
 877
 878 fail_options:
 879     error_propagate(errp, local_err);
 880     ret = -EINVAL;
 881     goto fail;
 882 }
 883
 884
 885 static void parallels_close(BlockDriverState *bs)
 886 {
 887     BDRVParallelsState *s = bs->opaque;
 888
 889     if ((bs->open_flags & BDRV_O_RDWR) && !(bs->open_flags & BDRV_O_INACTIVE)) {
 890         s->header->inuse = 0;
 891         parallels_update_header(bs);
 892         bdrv_truncate(bs->file, s->data_end << BDRV_SECTOR_BITS,
 893                       PREALLOC_MODE_OFF, NULL);
 894     }
 895
 896     g_free(s->bat_dirty_bmap);
 897     qemu_vfree(s->header);
 898
 899     migrate_del_blocker(s->migration_blocker);
 900     error_free(s->migration_blocker);
 901 }
 902
 903 static BlockDriver bdrv_parallels = {
 904     .format_name        = "parallels",
 905     .instance_size      = sizeof(BDRVParallelsState),
 906     .bdrv_probe         = parallels_probe,
 907     .bdrv_open          = parallels_open,
 908     .bdrv_close         = parallels_close,
 909     .bdrv_child_perm          = bdrv_format_default_perms,
 910     .bdrv_co_block_status     = parallels_co_block_status,
 911     .bdrv_has_zero_init       = bdrv_has_zero_init_1,
 912     .bdrv_co_flush_to_os      = parallels_co_flush_to_os,
 913     .bdrv_co_readv  = parallels_co_readv,
 914     .bdrv_co_writev = parallels_co_writev,
 915     .supports_backing = true,
 916     .bdrv_co_create      = parallels_co_create,
 917     .bdrv_co_create_opts = parallels_co_create_opts,
 918     .bdrv_co_check  = parallels_co_check,
 919     .create_opts    = &parallels_create_opts,
 920 };
 921
 922 static void bdrv_parallels_init(void)
 923 {
 924     bdrv_register(&bdrv_parallels);
 925 }
 926
 927 block_init(bdrv_parallels_init);