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