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                                 false, 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         void *buf = qemu_blockalign(bs, nb_cow_bytes);
 224
 225         ret = bdrv_co_pread(bs->backing, idx * s->tracks * BDRV_SECTOR_SIZE,
 226                             nb_cow_bytes, buf, 0);
 227         if (ret < 0) {
 228             qemu_vfree(buf);
 229             return ret;
 230         }
 231
 232         ret = bdrv_co_pwritev(bs->file, s->data_end * BDRV_SECTOR_SIZE,
 233                               nb_cow_bytes, buf, 0);
 234         qemu_vfree(buf);
 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 int coroutine_fn parallels_co_block_status(BlockDriverState *bs,
 284                                                   bool want_zero,
 285                                                   int64_t offset,
 286                                                   int64_t bytes,
 287                                                   int64_t *pnum,
 288                                                   int64_t *map,
 289                                                   BlockDriverState **file)
 290 {
 291     BDRVParallelsState *s = bs->opaque;
 292     int count;
 293
 294     assert(QEMU_IS_ALIGNED(offset | bytes, BDRV_SECTOR_SIZE));
 295     qemu_co_mutex_lock(&s->lock);
 296     offset = block_status(s, offset >> BDRV_SECTOR_BITS,
 297                           bytes >> BDRV_SECTOR_BITS, &count);
 298     qemu_co_mutex_unlock(&s->lock);
 299
 300     *pnum = count * BDRV_SECTOR_SIZE;
 301     if (offset < 0) {
 302         return 0;
 303     }
 304
 305     *map = offset * BDRV_SECTOR_SIZE;
 306     *file = bs->file->bs;
 307     return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
 308 }
 309
 310 static coroutine_fn int parallels_co_writev(BlockDriverState *bs,
 311                                             int64_t sector_num, int nb_sectors,
 312                                             QEMUIOVector *qiov, int flags)
 313 {
 314     BDRVParallelsState *s = bs->opaque;
 315     uint64_t bytes_done = 0;
 316     QEMUIOVector hd_qiov;
 317     int ret = 0;
 318
 319     assert(!flags);
 320     qemu_iovec_init(&hd_qiov, qiov->niov);
 321
 322     while (nb_sectors > 0) {
 323         int64_t position;
 324         int n, nbytes;
 325
 326         qemu_co_mutex_lock(&s->lock);
 327         position = allocate_clusters(bs, sector_num, nb_sectors, &n);
 328         qemu_co_mutex_unlock(&s->lock);
 329         if (position < 0) {
 330             ret = (int)position;
 331             break;
 332         }
 333
 334         nbytes = n << BDRV_SECTOR_BITS;
 335
 336         qemu_iovec_reset(&hd_qiov);
 337         qemu_iovec_concat(&hd_qiov, qiov, bytes_done, nbytes);
 338
 339         ret = bdrv_co_pwritev(bs->file, position * BDRV_SECTOR_SIZE, nbytes,
 340                               &hd_qiov, 0);
 341         if (ret < 0) {
 342             break;
 343         }
 344
 345         nb_sectors -= n;
 346         sector_num += n;
 347         bytes_done += nbytes;
 348     }
 349
 350     qemu_iovec_destroy(&hd_qiov);
 351     return ret;
 352 }
 353
 354 static coroutine_fn int parallels_co_readv(BlockDriverState *bs,
 355         int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
 356 {
 357     BDRVParallelsState *s = bs->opaque;
 358     uint64_t bytes_done = 0;
 359     QEMUIOVector hd_qiov;
 360     int ret = 0;
 361
 362     qemu_iovec_init(&hd_qiov, qiov->niov);
 363
 364     while (nb_sectors > 0) {
 365         int64_t position;
 366         int n, nbytes;
 367
 368         qemu_co_mutex_lock(&s->lock);
 369         position = block_status(s, sector_num, nb_sectors, &n);
 370         qemu_co_mutex_unlock(&s->lock);
 371
 372         nbytes = n << BDRV_SECTOR_BITS;
 373
 374         qemu_iovec_reset(&hd_qiov);
 375         qemu_iovec_concat(&hd_qiov, qiov, bytes_done, nbytes);
 376
 377         if (position < 0) {
 378             if (bs->backing) {
 379                 ret = bdrv_co_preadv(bs->backing, sector_num * BDRV_SECTOR_SIZE,
 380                                      nbytes, &hd_qiov, 0);
 381                 if (ret < 0) {
 382                     break;
 383                 }
 384             } else {
 385                 qemu_iovec_memset(&hd_qiov, 0, 0, nbytes);
 386             }
 387         } else {
 388             ret = bdrv_co_preadv(bs->file, position * BDRV_SECTOR_SIZE, nbytes,
 389                                  &hd_qiov, 0);
 390             if (ret < 0) {
 391                 break;
 392             }
 393         }
 394
 395         nb_sectors -= n;
 396         sector_num += n;
 397         bytes_done += nbytes;
 398     }
 399
 400     qemu_iovec_destroy(&hd_qiov);
 401     return ret;
 402 }
 403
 404
 405 static int coroutine_fn parallels_co_check(BlockDriverState *bs,
 406                                            BdrvCheckResult *res,
 407                                            BdrvCheckMode fix)
 408 {
 409     BDRVParallelsState *s = bs->opaque;
 410     int64_t size, prev_off, high_off;
 411     int ret;
 412     uint32_t i;
 413     bool flush_bat = false;
 414     int cluster_size = s->tracks << BDRV_SECTOR_BITS;
 415
 416     size = bdrv_getlength(bs->file->bs);
 417     if (size < 0) {
 418         res->check_errors++;
 419         return size;
 420     }
 421
 422     qemu_co_mutex_lock(&s->lock);
 423     if (s->header_unclean) {
 424         fprintf(stderr, "%s image was not closed correctly\n",
 425                 fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR");
 426         res->corruptions++;
 427         if (fix & BDRV_FIX_ERRORS) {
 428             /* parallels_close will do the job right */
 429             res->corruptions_fixed++;
 430             s->header_unclean = false;
 431         }
 432     }
 433
 434     res->bfi.total_clusters = s->bat_size;
 435     res->bfi.compressed_clusters = 0; /* compression is not supported */
 436
 437     high_off = 0;
 438     prev_off = 0;
 439     for (i = 0; i < s->bat_size; i++) {
 440         int64_t off = bat2sect(s, i) << BDRV_SECTOR_BITS;
 441         if (off == 0) {
 442             prev_off = 0;
 443             continue;
 444         }
 445
 446         /* cluster outside the image */
 447         if (off > size) {
 448             fprintf(stderr, "%s cluster %u is outside image\n",
 449                     fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR", i);
 450             res->corruptions++;
 451             if (fix & BDRV_FIX_ERRORS) {
 452                 prev_off = 0;
 453                 s->bat_bitmap[i] = 0;
 454                 res->corruptions_fixed++;
 455                 flush_bat = true;
 456                 continue;
 457             }
 458         }
 459
 460         res->bfi.allocated_clusters++;
 461         if (off > high_off) {
 462             high_off = off;
 463         }
 464
 465         if (prev_off != 0 && (prev_off + cluster_size) != off) {
 466             res->bfi.fragmented_clusters++;
 467         }
 468         prev_off = off;
 469     }
 470
 471     ret = 0;
 472     if (flush_bat) {
 473         ret = bdrv_pwrite_sync(bs->file, 0, s->header, s->header_size);
 474         if (ret < 0) {
 475             res->check_errors++;
 476             goto out;
 477         }
 478     }
 479
 480     res->image_end_offset = high_off + cluster_size;
 481     if (size > res->image_end_offset) {
 482         int64_t count;
 483         count = DIV_ROUND_UP(size - res->image_end_offset, cluster_size);
 484         fprintf(stderr, "%s space leaked at the end of the image %" PRId64 "\n",
 485                 fix & BDRV_FIX_LEAKS ? "Repairing" : "ERROR",
 486                 size - res->image_end_offset);
 487         res->leaks += count;
 488         if (fix & BDRV_FIX_LEAKS) {
 489             Error *local_err = NULL;
 490
 491             /*
 492              * In order to really repair the image, we must shrink it.
 493              * That means we have to pass exact=true.
 494              */
 495             ret = bdrv_truncate(bs->file, res->image_end_offset, true,
 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(bdrv_get_aio_context(bs),
 563                   BLK_PERM_WRITE | BLK_PERM_RESIZE, BLK_PERM_ALL);
 564     ret = blk_insert_bs(blk, bs, errp);
 565     if (ret < 0) {
 566         goto out;
 567     }
 568     blk_set_allow_write_beyond_eof(blk, true);
 569
 570     /* Create image format */
 571     bat_entries = DIV_ROUND_UP(total_size, cl_size);
 572     bat_sectors = DIV_ROUND_UP(bat_entry_off(bat_entries), cl_size);
 573     bat_sectors = (bat_sectors *  cl_size) >> BDRV_SECTOR_BITS;
 574
 575     memset(&header, 0, sizeof(header));
 576     memcpy(header.magic, HEADER_MAGIC2, sizeof(header.magic));
 577     header.version = cpu_to_le32(HEADER_VERSION);
 578     /* don't care much about geometry, it is not used on image level */
 579     header.heads = cpu_to_le32(HEADS_NUMBER);
 580     header.cylinders = cpu_to_le32(total_size / BDRV_SECTOR_SIZE
 581                                    / HEADS_NUMBER / SEC_IN_CYL);
 582     header.tracks = cpu_to_le32(cl_size >> BDRV_SECTOR_BITS);
 583     header.bat_entries = cpu_to_le32(bat_entries);
 584     header.nb_sectors = cpu_to_le64(DIV_ROUND_UP(total_size, BDRV_SECTOR_SIZE));
 585     header.data_off = cpu_to_le32(bat_sectors);
 586
 587     /* write all the data */
 588     memset(tmp, 0, sizeof(tmp));
 589     memcpy(tmp, &header, sizeof(header));
 590
 591     ret = blk_pwrite(blk, 0, tmp, BDRV_SECTOR_SIZE, 0);
 592     if (ret < 0) {
 593         goto exit;
 594     }
 595     ret = blk_pwrite_zeroes(blk, BDRV_SECTOR_SIZE,
 596                             (bat_sectors - 1) << BDRV_SECTOR_BITS, 0);
 597     if (ret < 0) {
 598         goto exit;
 599     }
 600
 601     ret = 0;
 602 out:
 603     blk_unref(blk);
 604     bdrv_unref(bs);
 605     return ret;
 606
 607 exit:
 608     error_setg_errno(errp, -ret, "Failed to create Parallels image");
 609     goto out;
 610 }
 611
 612 static int coroutine_fn parallels_co_create_opts(const char *filename,
 613                                                  QemuOpts *opts,
 614                                                  Error **errp)
 615 {
 616     BlockdevCreateOptions *create_options = NULL;
 617     Error *local_err = NULL;
 618     BlockDriverState *bs = NULL;
 619     QDict *qdict;
 620     Visitor *v;
 621     int ret;
 622
 623     static const QDictRenames opt_renames[] = {
 624         { BLOCK_OPT_CLUSTER_SIZE,       "cluster-size" },
 625         { NULL, NULL },
 626     };
 627
 628     /* Parse options and convert legacy syntax */
 629     qdict = qemu_opts_to_qdict_filtered(opts, NULL, &parallels_create_opts,
 630                                         true);
 631
 632     if (!qdict_rename_keys(qdict, opt_renames, errp)) {
 633         ret = -EINVAL;
 634         goto done;
 635     }
 636
 637     /* Create and open the file (protocol layer) */
 638     ret = bdrv_create_file(filename, opts, &local_err);
 639     if (ret < 0) {
 640         error_propagate(errp, local_err);
 641         goto done;
 642     }
 643
 644     bs = bdrv_open(filename, NULL, NULL,
 645                    BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp);
 646     if (bs == NULL) {
 647         ret = -EIO;
 648         goto done;
 649     }
 650
 651     /* Now get the QAPI type BlockdevCreateOptions */
 652     qdict_put_str(qdict, "driver", "parallels");
 653     qdict_put_str(qdict, "file", bs->node_name);
 654
 655     v = qobject_input_visitor_new_flat_confused(qdict, errp);
 656     if (!v) {
 657         ret = -EINVAL;
 658         goto done;
 659     }
 660
 661     visit_type_BlockdevCreateOptions(v, NULL, &create_options, &local_err);
 662     visit_free(v);
 663
 664     if (local_err) {
 665         error_propagate(errp, local_err);
 666         ret = -EINVAL;
 667         goto done;
 668     }
 669
 670     /* Silently round up sizes */
 671     create_options->u.parallels.size =
 672         ROUND_UP(create_options->u.parallels.size, BDRV_SECTOR_SIZE);
 673     create_options->u.parallels.cluster_size =
 674         ROUND_UP(create_options->u.parallels.cluster_size, BDRV_SECTOR_SIZE);
 675
 676     /* Create the Parallels image (format layer) */
 677     ret = parallels_co_create(create_options, errp);
 678     if (ret < 0) {
 679         goto done;
 680     }
 681     ret = 0;
 682
 683 done:
 684     qobject_unref(qdict);
 685     bdrv_unref(bs);
 686     qapi_free_BlockdevCreateOptions(create_options);
 687     return ret;
 688 }
 689
 690
 691 static int parallels_probe(const uint8_t *buf, int buf_size,
 692                            const char *filename)
 693 {
 694     const ParallelsHeader *ph = (const void *)buf;
 695
 696     if (buf_size < sizeof(ParallelsHeader)) {
 697         return 0;
 698     }
 699
 700     if ((!memcmp(ph->magic, HEADER_MAGIC, 16) ||
 701            !memcmp(ph->magic, HEADER_MAGIC2, 16)) &&
 702            (le32_to_cpu(ph->version) == HEADER_VERSION)) {
 703         return 100;
 704     }
 705
 706     return 0;
 707 }
 708
 709 static int parallels_update_header(BlockDriverState *bs)
 710 {
 711     BDRVParallelsState *s = bs->opaque;
 712     unsigned size = MAX(bdrv_opt_mem_align(bs->file->bs),
 713                         sizeof(ParallelsHeader));
 714
 715     if (size > s->header_size) {
 716         size = s->header_size;
 717     }
 718     return bdrv_pwrite_sync(bs->file, 0, s->header, size);
 719 }
 720
 721 static int parallels_open(BlockDriverState *bs, QDict *options, int flags,
 722                           Error **errp)
 723 {
 724     BDRVParallelsState *s = bs->opaque;
 725     ParallelsHeader ph;
 726     int ret, size, i;
 727     QemuOpts *opts = NULL;
 728     Error *local_err = NULL;
 729     char *buf;
 730
 731     bs->file = bdrv_open_child(NULL, options, "file", bs, &child_file,
 732                                false, errp);
 733     if (!bs->file) {
 734         return -EINVAL;
 735     }
 736
 737     ret = bdrv_pread(bs->file, 0, &ph, sizeof(ph));
 738     if (ret < 0) {
 739         goto fail;
 740     }
 741
 742     bs->total_sectors = le64_to_cpu(ph.nb_sectors);
 743
 744     if (le32_to_cpu(ph.version) != HEADER_VERSION) {
 745         goto fail_format;
 746     }
 747     if (!memcmp(ph.magic, HEADER_MAGIC, 16)) {
 748         s->off_multiplier = 1;
 749         bs->total_sectors = 0xffffffff & bs->total_sectors;
 750     } else if (!memcmp(ph.magic, HEADER_MAGIC2, 16)) {
 751         s->off_multiplier = le32_to_cpu(ph.tracks);
 752     } else {
 753         goto fail_format;
 754     }
 755
 756     s->tracks = le32_to_cpu(ph.tracks);
 757     if (s->tracks == 0) {
 758         error_setg(errp, "Invalid image: Zero sectors per track");
 759         ret = -EINVAL;
 760         goto fail;
 761     }
 762     if (s->tracks > INT32_MAX/513) {
 763         error_setg(errp, "Invalid image: Too big cluster");
 764         ret = -EFBIG;
 765         goto fail;
 766     }
 767
 768     s->bat_size = le32_to_cpu(ph.bat_entries);
 769     if (s->bat_size > INT_MAX / sizeof(uint32_t)) {
 770         error_setg(errp, "Catalog too large");
 771         ret = -EFBIG;
 772         goto fail;
 773     }
 774
 775     size = bat_entry_off(s->bat_size);
 776     s->header_size = ROUND_UP(size, bdrv_opt_mem_align(bs->file->bs));
 777     s->header = qemu_try_blockalign(bs->file->bs, s->header_size);
 778     if (s->header == NULL) {
 779         ret = -ENOMEM;
 780         goto fail;
 781     }
 782     s->data_end = le32_to_cpu(ph.data_off);
 783     if (s->data_end == 0) {
 784         s->data_end = ROUND_UP(bat_entry_off(s->bat_size), BDRV_SECTOR_SIZE);
 785     }
 786     if (s->data_end < s->header_size) {
 787         /* there is not enough unused space to fit to block align between BAT
 788            and actual data. We can't avoid read-modify-write... */
 789         s->header_size = size;
 790     }
 791
 792     ret = bdrv_pread(bs->file, 0, s->header, s->header_size);
 793     if (ret < 0) {
 794         goto fail;
 795     }
 796     s->bat_bitmap = (uint32_t *)(s->header + 1);
 797
 798     for (i = 0; i < s->bat_size; i++) {
 799         int64_t off = bat2sect(s, i);
 800         if (off >= s->data_end) {
 801             s->data_end = off + s->tracks;
 802         }
 803     }
 804
 805     if (le32_to_cpu(ph.inuse) == HEADER_INUSE_MAGIC) {
 806         /* Image was not closed correctly. The check is mandatory */
 807         s->header_unclean = true;
 808         if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_CHECK)) {
 809             error_setg(errp, "parallels: Image was not closed correctly; "
 810                        "cannot be opened read/write");
 811             ret = -EACCES;
 812             goto fail;
 813         }
 814     }
 815
 816     opts = qemu_opts_create(&parallels_runtime_opts, NULL, 0, &local_err);
 817     if (local_err != NULL) {
 818         goto fail_options;
 819     }
 820
 821     qemu_opts_absorb_qdict(opts, options, &local_err);
 822     if (local_err != NULL) {
 823         goto fail_options;
 824     }
 825
 826     s->prealloc_size =
 827         qemu_opt_get_size_del(opts, PARALLELS_OPT_PREALLOC_SIZE, 0);
 828     s->prealloc_size = MAX(s->tracks, s->prealloc_size >> BDRV_SECTOR_BITS);
 829     buf = qemu_opt_get_del(opts, PARALLELS_OPT_PREALLOC_MODE);
 830     s->prealloc_mode = qapi_enum_parse(&prealloc_mode_lookup, buf,
 831                                        PRL_PREALLOC_MODE_FALLOCATE,
 832                                        &local_err);
 833     g_free(buf);
 834     if (local_err != NULL) {
 835         goto fail_options;
 836     }
 837
 838     if (!bdrv_has_zero_init_truncate(bs->file->bs)) {
 839         s->prealloc_mode = PRL_PREALLOC_MODE_FALLOCATE;
 840     }
 841
 842     if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_INACTIVE)) {
 843         s->header->inuse = cpu_to_le32(HEADER_INUSE_MAGIC);
 844         ret = parallels_update_header(bs);
 845         if (ret < 0) {
 846             goto fail;
 847         }
 848     }
 849
 850     s->bat_dirty_block = 4 * qemu_real_host_page_size;
 851     s->bat_dirty_bmap =
 852         bitmap_new(DIV_ROUND_UP(s->header_size, s->bat_dirty_block));
 853
 854     /* Disable migration until bdrv_invalidate_cache method is added */
 855     error_setg(&s->migration_blocker, "The Parallels format used by node '%s' "
 856                "does not support live migration",
 857                bdrv_get_device_or_node_name(bs));
 858     ret = migrate_add_blocker(s->migration_blocker, &local_err);
 859     if (local_err) {
 860         error_propagate(errp, local_err);
 861         error_free(s->migration_blocker);
 862         goto fail;
 863     }
 864     qemu_co_mutex_init(&s->lock);
 865     return 0;
 866
 867 fail_format:
 868     error_setg(errp, "Image not in Parallels format");
 869     ret = -EINVAL;
 870 fail:
 871     qemu_vfree(s->header);
 872     return ret;
 873
 874 fail_options:
 875     error_propagate(errp, local_err);
 876     ret = -EINVAL;
 877     goto fail;
 878 }
 879
 880
 881 static void parallels_close(BlockDriverState *bs)
 882 {
 883     BDRVParallelsState *s = bs->opaque;
 884
 885     if ((bs->open_flags & BDRV_O_RDWR) && !(bs->open_flags & BDRV_O_INACTIVE)) {
 886         s->header->inuse = 0;
 887         parallels_update_header(bs);
 888
 889         /* errors are ignored, so we might as well pass exact=true */
 890         bdrv_truncate(bs->file, s->data_end << BDRV_SECTOR_BITS, true,
 891                       PREALLOC_MODE_OFF, NULL);
 892     }
 893
 894     g_free(s->bat_dirty_bmap);
 895     qemu_vfree(s->header);
 896
 897     migrate_del_blocker(s->migration_blocker);
 898     error_free(s->migration_blocker);
 899 }
 900
 901 static BlockDriver bdrv_parallels = {
 902     .format_name        = "parallels",
 903     .instance_size      = sizeof(BDRVParallelsState),
 904     .bdrv_probe         = parallels_probe,
 905     .bdrv_open          = parallels_open,
 906     .bdrv_close         = parallels_close,
 907     .bdrv_child_perm          = bdrv_format_default_perms,
 908     .bdrv_co_block_status     = parallels_co_block_status,
 909     .bdrv_has_zero_init       = bdrv_has_zero_init_1,
 910     .bdrv_co_flush_to_os      = parallels_co_flush_to_os,
 911     .bdrv_co_readv  = parallels_co_readv,
 912     .bdrv_co_writev = parallels_co_writev,
 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);