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 "qemu/error-report.h"
  33 #include "qapi/error.h"
  34 #include "block/block_int.h"
  35 #include "block/qdict.h"
  36 #include "sysemu/block-backend.h"
  37 #include "qemu/module.h"
  38 #include "qemu/option.h"
  39 #include "qapi/qmp/qdict.h"
  40 #include "qapi/qobject-input-visitor.h"
  41 #include "qapi/qapi-visit-block-core.h"
  42 #include "qemu/bswap.h"
  43 #include "qemu/bitmap.h"
  44 #include "qemu/memalign.h"
  45 #include "migration/blocker.h"
  46 #include "parallels.h"
  47
  48 /**************************************************************/
  49
  50 #define HEADER_MAGIC "WithoutFreeSpace"
  51 #define HEADER_MAGIC2 "WithouFreSpacExt"
  52 #define HEADER_VERSION 2
  53 #define HEADER_INUSE_MAGIC  (0x746F6E59)
  54 #define MAX_PARALLELS_IMAGE_FACTOR (1ull << 32)
  55
  56 static QEnumLookup prealloc_mode_lookup = {
  57     .array = (const char *const[]) {
  58         "falloc",
  59         "truncate",
  60     },
  61     .size = PRL_PREALLOC_MODE__MAX
  62 };
  63
  64 #define PARALLELS_OPT_PREALLOC_MODE     "prealloc-mode"
  65 #define PARALLELS_OPT_PREALLOC_SIZE     "prealloc-size"
  66
  67 static QemuOptsList parallels_runtime_opts = {
  68     .name = "parallels",
  69     .head = QTAILQ_HEAD_INITIALIZER(parallels_runtime_opts.head),
  70     .desc = {
  71         {
  72             .name = PARALLELS_OPT_PREALLOC_SIZE,
  73             .type = QEMU_OPT_SIZE,
  74             .help = "Preallocation size on image expansion",
  75             .def_value_str = "128M",
  76         },
  77         {
  78             .name = PARALLELS_OPT_PREALLOC_MODE,
  79             .type = QEMU_OPT_STRING,
  80             .help = "Preallocation mode on image expansion "
  81                     "(allowed values: falloc, truncate)",
  82             .def_value_str = "falloc",
  83         },
  84         { /* end of list */ },
  85     },
  86 };
  87
  88 static QemuOptsList parallels_create_opts = {
  89     .name = "parallels-create-opts",
  90     .head = QTAILQ_HEAD_INITIALIZER(parallels_create_opts.head),
  91     .desc = {
  92         {
  93             .name = BLOCK_OPT_SIZE,
  94             .type = QEMU_OPT_SIZE,
  95             .help = "Virtual disk size",
  96         },
  97         {
  98             .name = BLOCK_OPT_CLUSTER_SIZE,
  99             .type = QEMU_OPT_SIZE,
 100             .help = "Parallels image cluster size",
 101             .def_value_str = stringify(DEFAULT_CLUSTER_SIZE),
 102         },
 103         { /* end of list */ }
 104     }
 105 };
 106
 107
 108 static int64_t bat2sect(BDRVParallelsState *s, uint32_t idx)
 109 {
 110     return (uint64_t)le32_to_cpu(s->bat_bitmap[idx]) * s->off_multiplier;
 111 }
 112
 113 static uint32_t bat_entry_off(uint32_t idx)
 114 {
 115     return sizeof(ParallelsHeader) + sizeof(uint32_t) * idx;
 116 }
 117
 118 static int64_t seek_to_sector(BDRVParallelsState *s, int64_t sector_num)
 119 {
 120     uint32_t index, offset;
 121
 122     index = sector_num / s->tracks;
 123     offset = sector_num % s->tracks;
 124
 125     /* not allocated */
 126     if ((index >= s->bat_size) || (s->bat_bitmap[index] == 0)) {
 127         return -1;
 128     }
 129     return bat2sect(s, index) + offset;
 130 }
 131
 132 static int cluster_remainder(BDRVParallelsState *s, int64_t sector_num,
 133         int nb_sectors)
 134 {
 135     int ret = s->tracks - sector_num % s->tracks;
 136     return MIN(nb_sectors, ret);
 137 }
 138
 139 static int64_t block_status(BDRVParallelsState *s, int64_t sector_num,
 140                             int nb_sectors, int *pnum)
 141 {
 142     int64_t start_off = -2, prev_end_off = -2;
 143
 144     *pnum = 0;
 145     while (nb_sectors > 0 || start_off == -2) {
 146         int64_t offset = seek_to_sector(s, sector_num);
 147         int to_end;
 148
 149         if (start_off == -2) {
 150             start_off = offset;
 151             prev_end_off = offset;
 152         } else if (offset != prev_end_off) {
 153             break;
 154         }
 155
 156         to_end = cluster_remainder(s, sector_num, nb_sectors);
 157         nb_sectors -= to_end;
 158         sector_num += to_end;
 159         *pnum += to_end;
 160
 161         if (offset > 0) {
 162             prev_end_off += to_end;
 163         }
 164     }
 165     return start_off;
 166 }
 167
 168 static void parallels_set_bat_entry(BDRVParallelsState *s,
 169                                     uint32_t index, uint32_t offset)
 170 {
 171     s->bat_bitmap[index] = cpu_to_le32(offset);
 172     bitmap_set(s->bat_dirty_bmap, bat_entry_off(index) / s->bat_dirty_block, 1);
 173 }
 174
 175 static int64_t coroutine_fn GRAPH_RDLOCK
 176 allocate_clusters(BlockDriverState *bs, int64_t sector_num,
 177                   int nb_sectors, int *pnum)
 178 {
 179     int ret = 0;
 180     BDRVParallelsState *s = bs->opaque;
 181     int64_t pos, space, idx, to_allocate, i, len;
 182
 183     pos = block_status(s, sector_num, nb_sectors, pnum);
 184     if (pos > 0) {
 185         return pos;
 186     }
 187
 188     idx = sector_num / s->tracks;
 189     to_allocate = DIV_ROUND_UP(sector_num + *pnum, s->tracks) - idx;
 190
 191     /* This function is called only by parallels_co_writev(), which will never
 192      * pass a sector_num at or beyond the end of the image (because the block
 193      * layer never passes such a sector_num to that function). Therefore, idx
 194      * is always below s->bat_size.
 195      * block_status() will limit *pnum so that sector_num + *pnum will not
 196      * exceed the image end. Therefore, idx + to_allocate cannot exceed
 197      * s->bat_size.
 198      * Note that s->bat_size is an unsigned int, therefore idx + to_allocate
 199      * will always fit into a uint32_t. */
 200     assert(idx < s->bat_size && idx + to_allocate <= s->bat_size);
 201
 202     space = to_allocate * s->tracks;
 203     len = bdrv_getlength(bs->file->bs);
 204     if (len < 0) {
 205         return len;
 206     }
 207     if (s->data_end + space > (len >> BDRV_SECTOR_BITS)) {
 208         space += s->prealloc_size;
 209         /*
 210          * We require the expanded size to read back as zero. If the
 211          * user permitted truncation, we try that; but if it fails, we
 212          * force the safer-but-slower fallocate.
 213          */
 214         if (s->prealloc_mode == PRL_PREALLOC_MODE_TRUNCATE) {
 215             ret = bdrv_co_truncate(bs->file,
 216                                    (s->data_end + space) << BDRV_SECTOR_BITS,
 217                                    false, PREALLOC_MODE_OFF,
 218                                    BDRV_REQ_ZERO_WRITE, NULL);
 219             if (ret == -ENOTSUP) {
 220                 s->prealloc_mode = PRL_PREALLOC_MODE_FALLOCATE;
 221             }
 222         }
 223         if (s->prealloc_mode == PRL_PREALLOC_MODE_FALLOCATE) {
 224             ret = bdrv_co_pwrite_zeroes(bs->file,
 225                                         s->data_end << BDRV_SECTOR_BITS,
 226                                         space << BDRV_SECTOR_BITS, 0);
 227         }
 228         if (ret < 0) {
 229             return ret;
 230         }
 231     }
 232
 233     /* Try to read from backing to fill empty clusters
 234      * FIXME: 1. previous write_zeroes may be redundant
 235      *        2. most of data we read from backing will be rewritten by
 236      *           parallels_co_writev. On aligned-to-cluster write we do not need
 237      *           this read at all.
 238      *        3. it would be good to combine write of data from backing and new
 239      *           data into one write call */
 240     if (bs->backing) {
 241         int64_t nb_cow_sectors = to_allocate * s->tracks;
 242         int64_t nb_cow_bytes = nb_cow_sectors << BDRV_SECTOR_BITS;
 243         void *buf = qemu_blockalign(bs, nb_cow_bytes);
 244
 245         ret = bdrv_co_pread(bs->backing, idx * s->tracks * BDRV_SECTOR_SIZE,
 246                             nb_cow_bytes, buf, 0);
 247         if (ret < 0) {
 248             qemu_vfree(buf);
 249             return ret;
 250         }
 251
 252         ret = bdrv_co_pwrite(bs->file, s->data_end * BDRV_SECTOR_SIZE,
 253                              nb_cow_bytes, buf, 0);
 254         qemu_vfree(buf);
 255         if (ret < 0) {
 256             return ret;
 257         }
 258     }
 259
 260     for (i = 0; i < to_allocate; i++) {
 261         parallels_set_bat_entry(s, idx + i, s->data_end / s->off_multiplier);
 262         s->data_end += s->tracks;
 263     }
 264
 265     return bat2sect(s, idx) + sector_num % s->tracks;
 266 }
 267
 268
 269 static int coroutine_fn GRAPH_RDLOCK
 270 parallels_co_flush_to_os(BlockDriverState *bs)
 271 {
 272     BDRVParallelsState *s = bs->opaque;
 273     unsigned long size = DIV_ROUND_UP(s->header_size, s->bat_dirty_block);
 274     unsigned long bit;
 275
 276     qemu_co_mutex_lock(&s->lock);
 277
 278     bit = find_first_bit(s->bat_dirty_bmap, size);
 279     while (bit < size) {
 280         uint32_t off = bit * s->bat_dirty_block;
 281         uint32_t to_write = s->bat_dirty_block;
 282         int ret;
 283
 284         if (off + to_write > s->header_size) {
 285             to_write = s->header_size - off;
 286         }
 287         ret = bdrv_co_pwrite(bs->file, off, to_write,
 288                              (uint8_t *)s->header + off, 0);
 289         if (ret < 0) {
 290             qemu_co_mutex_unlock(&s->lock);
 291             return ret;
 292         }
 293         bit = find_next_bit(s->bat_dirty_bmap, size, bit + 1);
 294     }
 295     bitmap_zero(s->bat_dirty_bmap, size);
 296
 297     qemu_co_mutex_unlock(&s->lock);
 298     return 0;
 299 }
 300
 301
 302 static int coroutine_fn parallels_co_block_status(BlockDriverState *bs,
 303                                                   bool want_zero,
 304                                                   int64_t offset,
 305                                                   int64_t bytes,
 306                                                   int64_t *pnum,
 307                                                   int64_t *map,
 308                                                   BlockDriverState **file)
 309 {
 310     BDRVParallelsState *s = bs->opaque;
 311     int count;
 312
 313     assert(QEMU_IS_ALIGNED(offset | bytes, BDRV_SECTOR_SIZE));
 314     qemu_co_mutex_lock(&s->lock);
 315     offset = block_status(s, offset >> BDRV_SECTOR_BITS,
 316                           bytes >> BDRV_SECTOR_BITS, &count);
 317     qemu_co_mutex_unlock(&s->lock);
 318
 319     *pnum = count * BDRV_SECTOR_SIZE;
 320     if (offset < 0) {
 321         return 0;
 322     }
 323
 324     *map = offset * BDRV_SECTOR_SIZE;
 325     *file = bs->file->bs;
 326     return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
 327 }
 328
 329 static int coroutine_fn GRAPH_RDLOCK
 330 parallels_co_writev(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
 331                     QEMUIOVector *qiov, int flags)
 332 {
 333     BDRVParallelsState *s = bs->opaque;
 334     uint64_t bytes_done = 0;
 335     QEMUIOVector hd_qiov;
 336     int ret = 0;
 337
 338     qemu_iovec_init(&hd_qiov, qiov->niov);
 339
 340     while (nb_sectors > 0) {
 341         int64_t position;
 342         int n, nbytes;
 343
 344         qemu_co_mutex_lock(&s->lock);
 345         position = allocate_clusters(bs, sector_num, nb_sectors, &n);
 346         qemu_co_mutex_unlock(&s->lock);
 347         if (position < 0) {
 348             ret = (int)position;
 349             break;
 350         }
 351
 352         nbytes = n << BDRV_SECTOR_BITS;
 353
 354         qemu_iovec_reset(&hd_qiov);
 355         qemu_iovec_concat(&hd_qiov, qiov, bytes_done, nbytes);
 356
 357         ret = bdrv_co_pwritev(bs->file, position * BDRV_SECTOR_SIZE, nbytes,
 358                               &hd_qiov, 0);
 359         if (ret < 0) {
 360             break;
 361         }
 362
 363         nb_sectors -= n;
 364         sector_num += n;
 365         bytes_done += nbytes;
 366     }
 367
 368     qemu_iovec_destroy(&hd_qiov);
 369     return ret;
 370 }
 371
 372 static int coroutine_fn GRAPH_RDLOCK
 373 parallels_co_readv(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
 374                    QEMUIOVector *qiov)
 375 {
 376     BDRVParallelsState *s = bs->opaque;
 377     uint64_t bytes_done = 0;
 378     QEMUIOVector hd_qiov;
 379     int ret = 0;
 380
 381     qemu_iovec_init(&hd_qiov, qiov->niov);
 382
 383     while (nb_sectors > 0) {
 384         int64_t position;
 385         int n, nbytes;
 386
 387         qemu_co_mutex_lock(&s->lock);
 388         position = block_status(s, sector_num, nb_sectors, &n);
 389         qemu_co_mutex_unlock(&s->lock);
 390
 391         nbytes = n << BDRV_SECTOR_BITS;
 392
 393         qemu_iovec_reset(&hd_qiov);
 394         qemu_iovec_concat(&hd_qiov, qiov, bytes_done, nbytes);
 395
 396         if (position < 0) {
 397             if (bs->backing) {
 398                 ret = bdrv_co_preadv(bs->backing, sector_num * BDRV_SECTOR_SIZE,
 399                                      nbytes, &hd_qiov, 0);
 400                 if (ret < 0) {
 401                     break;
 402                 }
 403             } else {
 404                 qemu_iovec_memset(&hd_qiov, 0, 0, nbytes);
 405             }
 406         } else {
 407             ret = bdrv_co_preadv(bs->file, position * BDRV_SECTOR_SIZE, nbytes,
 408                                  &hd_qiov, 0);
 409             if (ret < 0) {
 410                 break;
 411             }
 412         }
 413
 414         nb_sectors -= n;
 415         sector_num += n;
 416         bytes_done += nbytes;
 417     }
 418
 419     qemu_iovec_destroy(&hd_qiov);
 420     return ret;
 421 }
 422
 423 static void parallels_check_unclean(BlockDriverState *bs,
 424                                     BdrvCheckResult *res,
 425                                     BdrvCheckMode fix)
 426 {
 427     BDRVParallelsState *s = bs->opaque;
 428
 429     if (!s->header_unclean) {
 430         return;
 431     }
 432
 433     fprintf(stderr, "%s image was not closed correctly\n",
 434             fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR");
 435     res->corruptions++;
 436     if (fix & BDRV_FIX_ERRORS) {
 437         /* parallels_close will do the job right */
 438         res->corruptions_fixed++;
 439         s->header_unclean = false;
 440     }
 441 }
 442
 443 static int coroutine_fn GRAPH_RDLOCK
 444 parallels_check_outside_image(BlockDriverState *bs, BdrvCheckResult *res,
 445                               BdrvCheckMode fix)
 446 {
 447     BDRVParallelsState *s = bs->opaque;
 448     uint32_t i;
 449     int64_t off, high_off, size;
 450
 451     size = bdrv_getlength(bs->file->bs);
 452     if (size < 0) {
 453         res->check_errors++;
 454         return size;
 455     }
 456
 457     high_off = 0;
 458     for (i = 0; i < s->bat_size; i++) {
 459         off = bat2sect(s, i) << BDRV_SECTOR_BITS;
 460         if (off + s->cluster_size > size) {
 461             fprintf(stderr, "%s cluster %u is outside image\n",
 462                     fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR", i);
 463             res->corruptions++;
 464             if (fix & BDRV_FIX_ERRORS) {
 465                 parallels_set_bat_entry(s, i, 0);
 466                 res->corruptions_fixed++;
 467             }
 468             continue;
 469         }
 470         if (high_off < off) {
 471             high_off = off;
 472         }
 473     }
 474
 475     if (high_off == 0) {
 476         res->image_end_offset = s->data_end << BDRV_SECTOR_BITS;
 477     } else {
 478         res->image_end_offset = high_off + s->cluster_size;
 479         s->data_end = res->image_end_offset >> BDRV_SECTOR_BITS;
 480     }
 481
 482     return 0;
 483 }
 484
 485 static int coroutine_fn GRAPH_RDLOCK
 486 parallels_check_leak(BlockDriverState *bs, BdrvCheckResult *res,
 487                      BdrvCheckMode fix)
 488 {
 489     BDRVParallelsState *s = bs->opaque;
 490     int64_t size;
 491     int ret;
 492
 493     size = bdrv_getlength(bs->file->bs);
 494     if (size < 0) {
 495         res->check_errors++;
 496         return size;
 497     }
 498
 499     if (size > res->image_end_offset) {
 500         int64_t count;
 501         count = DIV_ROUND_UP(size - res->image_end_offset, s->cluster_size);
 502         fprintf(stderr, "%s space leaked at the end of the image %" PRId64 "\n",
 503                 fix & BDRV_FIX_LEAKS ? "Repairing" : "ERROR",
 504                 size - res->image_end_offset);
 505         res->leaks += count;
 506         if (fix & BDRV_FIX_LEAKS) {
 507             Error *local_err = NULL;
 508
 509             /*
 510              * In order to really repair the image, we must shrink it.
 511              * That means we have to pass exact=true.
 512              */
 513             ret = bdrv_co_truncate(bs->file, res->image_end_offset, true,
 514                                    PREALLOC_MODE_OFF, 0, &local_err);
 515             if (ret < 0) {
 516                 error_report_err(local_err);
 517                 res->check_errors++;
 518                 return ret;
 519             }
 520             res->leaks_fixed += count;
 521         }
 522     }
 523
 524     return 0;
 525 }
 526
 527 static void parallels_collect_statistics(BlockDriverState *bs,
 528                                          BdrvCheckResult *res,
 529                                          BdrvCheckMode fix)
 530 {
 531     BDRVParallelsState *s = bs->opaque;
 532     int64_t off, prev_off;
 533     uint32_t i;
 534
 535     res->bfi.total_clusters = s->bat_size;
 536     res->bfi.compressed_clusters = 0; /* compression is not supported */
 537
 538     prev_off = 0;
 539     for (i = 0; i < s->bat_size; i++) {
 540         off = bat2sect(s, i) << BDRV_SECTOR_BITS;
 541         /*
 542          * If BDRV_FIX_ERRORS is not set, out-of-image BAT entries were not
 543          * fixed. Skip not allocated and out-of-image BAT entries.
 544          */
 545         if (off == 0 || off + s->cluster_size > res->image_end_offset) {
 546             prev_off = 0;
 547             continue;
 548         }
 549
 550         if (prev_off != 0 && (prev_off + s->cluster_size) != off) {
 551             res->bfi.fragmented_clusters++;
 552         }
 553         prev_off = off;
 554         res->bfi.allocated_clusters++;
 555     }
 556 }
 557
 558 static int coroutine_fn GRAPH_RDLOCK
 559 parallels_co_check(BlockDriverState *bs, BdrvCheckResult *res,
 560                    BdrvCheckMode fix)
 561 {
 562     BDRVParallelsState *s = bs->opaque;
 563     int ret;
 564
 565     WITH_QEMU_LOCK_GUARD(&s->lock) {
 566         parallels_check_unclean(bs, res, fix);
 567
 568         ret = parallels_check_outside_image(bs, res, fix);
 569         if (ret < 0) {
 570             return ret;
 571         }
 572
 573         ret = parallels_check_leak(bs, res, fix);
 574         if (ret < 0) {
 575             return ret;
 576         }
 577
 578         parallels_collect_statistics(bs, res, fix);
 579     }
 580
 581     ret = bdrv_co_flush(bs);
 582     if (ret < 0) {
 583         res->check_errors++;
 584     }
 585
 586     return ret;
 587 }
 588
 589
 590 static int coroutine_fn GRAPH_UNLOCKED
 591 parallels_co_create(BlockdevCreateOptions* opts, Error **errp)
 592 {
 593     BlockdevCreateOptionsParallels *parallels_opts;
 594     BlockDriverState *bs;
 595     BlockBackend *blk;
 596     int64_t total_size, cl_size;
 597     uint32_t bat_entries, bat_sectors;
 598     ParallelsHeader header;
 599     uint8_t tmp[BDRV_SECTOR_SIZE];
 600     int ret;
 601
 602     assert(opts->driver == BLOCKDEV_DRIVER_PARALLELS);
 603     parallels_opts = &opts->u.parallels;
 604
 605     /* Sanity checks */
 606     total_size = parallels_opts->size;
 607
 608     if (parallels_opts->has_cluster_size) {
 609         cl_size = parallels_opts->cluster_size;
 610     } else {
 611         cl_size = DEFAULT_CLUSTER_SIZE;
 612     }
 613
 614     /* XXX What is the real limit here? This is an insanely large maximum. */
 615     if (cl_size >= INT64_MAX / MAX_PARALLELS_IMAGE_FACTOR) {
 616         error_setg(errp, "Cluster size is too large");
 617         return -EINVAL;
 618     }
 619     if (total_size >= MAX_PARALLELS_IMAGE_FACTOR * cl_size) {
 620         error_setg(errp, "Image size is too large for this cluster size");
 621         return -E2BIG;
 622     }
 623
 624     if (!QEMU_IS_ALIGNED(total_size, BDRV_SECTOR_SIZE)) {
 625         error_setg(errp, "Image size must be a multiple of 512 bytes");
 626         return -EINVAL;
 627     }
 628
 629     if (!QEMU_IS_ALIGNED(cl_size, BDRV_SECTOR_SIZE)) {
 630         error_setg(errp, "Cluster size must be a multiple of 512 bytes");
 631         return -EINVAL;
 632     }
 633
 634     /* Create BlockBackend to write to the image */
 635     bs = bdrv_co_open_blockdev_ref(parallels_opts->file, errp);
 636     if (bs == NULL) {
 637         return -EIO;
 638     }
 639
 640     blk = blk_co_new_with_bs(bs, BLK_PERM_WRITE | BLK_PERM_RESIZE, BLK_PERM_ALL,
 641                              errp);
 642     if (!blk) {
 643         ret = -EPERM;
 644         goto out;
 645     }
 646     blk_set_allow_write_beyond_eof(blk, true);
 647
 648     /* Create image format */
 649     bat_entries = DIV_ROUND_UP(total_size, cl_size);
 650     bat_sectors = DIV_ROUND_UP(bat_entry_off(bat_entries), cl_size);
 651     bat_sectors = (bat_sectors *  cl_size) >> BDRV_SECTOR_BITS;
 652
 653     memset(&header, 0, sizeof(header));
 654     memcpy(header.magic, HEADER_MAGIC2, sizeof(header.magic));
 655     header.version = cpu_to_le32(HEADER_VERSION);
 656     /* don't care much about geometry, it is not used on image level */
 657     header.heads = cpu_to_le32(HEADS_NUMBER);
 658     header.cylinders = cpu_to_le32(total_size / BDRV_SECTOR_SIZE
 659                                    / HEADS_NUMBER / SEC_IN_CYL);
 660     header.tracks = cpu_to_le32(cl_size >> BDRV_SECTOR_BITS);
 661     header.bat_entries = cpu_to_le32(bat_entries);
 662     header.nb_sectors = cpu_to_le64(DIV_ROUND_UP(total_size, BDRV_SECTOR_SIZE));
 663     header.data_off = cpu_to_le32(bat_sectors);
 664
 665     /* write all the data */
 666     memset(tmp, 0, sizeof(tmp));
 667     memcpy(tmp, &header, sizeof(header));
 668
 669     ret = blk_co_pwrite(blk, 0, BDRV_SECTOR_SIZE, tmp, 0);
 670     if (ret < 0) {
 671         goto exit;
 672     }
 673     ret = blk_co_pwrite_zeroes(blk, BDRV_SECTOR_SIZE,
 674                                (bat_sectors - 1) << BDRV_SECTOR_BITS, 0);
 675     if (ret < 0) {
 676         goto exit;
 677     }
 678
 679     ret = 0;
 680 out:
 681     blk_co_unref(blk);
 682     bdrv_co_unref(bs);
 683     return ret;
 684
 685 exit:
 686     error_setg_errno(errp, -ret, "Failed to create Parallels image");
 687     goto out;
 688 }
 689
 690 static int coroutine_fn GRAPH_UNLOCKED
 691 parallels_co_create_opts(BlockDriver *drv, const char *filename,
 692                          QemuOpts *opts, Error **errp)
 693 {
 694     BlockdevCreateOptions *create_options = NULL;
 695     BlockDriverState *bs = NULL;
 696     QDict *qdict;
 697     Visitor *v;
 698     int ret;
 699
 700     static const QDictRenames opt_renames[] = {
 701         { BLOCK_OPT_CLUSTER_SIZE,       "cluster-size" },
 702         { NULL, NULL },
 703     };
 704
 705     /* Parse options and convert legacy syntax */
 706     qdict = qemu_opts_to_qdict_filtered(opts, NULL, &parallels_create_opts,
 707                                         true);
 708
 709     if (!qdict_rename_keys(qdict, opt_renames, errp)) {
 710         ret = -EINVAL;
 711         goto done;
 712     }
 713
 714     /* Create and open the file (protocol layer) */
 715     ret = bdrv_co_create_file(filename, opts, errp);
 716     if (ret < 0) {
 717         goto done;
 718     }
 719
 720     bs = bdrv_co_open(filename, NULL, NULL,
 721                       BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp);
 722     if (bs == NULL) {
 723         ret = -EIO;
 724         goto done;
 725     }
 726
 727     /* Now get the QAPI type BlockdevCreateOptions */
 728     qdict_put_str(qdict, "driver", "parallels");
 729     qdict_put_str(qdict, "file", bs->node_name);
 730
 731     v = qobject_input_visitor_new_flat_confused(qdict, errp);
 732     if (!v) {
 733         ret = -EINVAL;
 734         goto done;
 735     }
 736
 737     visit_type_BlockdevCreateOptions(v, NULL, &create_options, errp);
 738     visit_free(v);
 739     if (!create_options) {
 740         ret = -EINVAL;
 741         goto done;
 742     }
 743
 744     /* Silently round up sizes */
 745     create_options->u.parallels.size =
 746         ROUND_UP(create_options->u.parallels.size, BDRV_SECTOR_SIZE);
 747     create_options->u.parallels.cluster_size =
 748         ROUND_UP(create_options->u.parallels.cluster_size, BDRV_SECTOR_SIZE);
 749
 750     /* Create the Parallels image (format layer) */
 751     ret = parallels_co_create(create_options, errp);
 752     if (ret < 0) {
 753         goto done;
 754     }
 755     ret = 0;
 756
 757 done:
 758     qobject_unref(qdict);
 759     bdrv_co_unref(bs);
 760     qapi_free_BlockdevCreateOptions(create_options);
 761     return ret;
 762 }
 763
 764
 765 static int parallels_probe(const uint8_t *buf, int buf_size,
 766                            const char *filename)
 767 {
 768     const ParallelsHeader *ph = (const void *)buf;
 769
 770     if (buf_size < sizeof(ParallelsHeader)) {
 771         return 0;
 772     }
 773
 774     if ((!memcmp(ph->magic, HEADER_MAGIC, 16) ||
 775            !memcmp(ph->magic, HEADER_MAGIC2, 16)) &&
 776            (le32_to_cpu(ph->version) == HEADER_VERSION)) {
 777         return 100;
 778     }
 779
 780     return 0;
 781 }
 782
 783 static int parallels_update_header(BlockDriverState *bs)
 784 {
 785     BDRVParallelsState *s = bs->opaque;
 786     unsigned size = MAX(bdrv_opt_mem_align(bs->file->bs),
 787                         sizeof(ParallelsHeader));
 788
 789     if (size > s->header_size) {
 790         size = s->header_size;
 791     }
 792     return bdrv_pwrite_sync(bs->file, 0, size, s->header, 0);
 793 }
 794
 795 static int parallels_open(BlockDriverState *bs, QDict *options, int flags,
 796                           Error **errp)
 797 {
 798     BDRVParallelsState *s = bs->opaque;
 799     ParallelsHeader ph;
 800     int ret, size, i;
 801     int64_t file_nb_sectors;
 802     QemuOpts *opts = NULL;
 803     Error *local_err = NULL;
 804     char *buf;
 805
 806     ret = bdrv_open_file_child(NULL, options, "file", bs, errp);
 807     if (ret < 0) {
 808         return ret;
 809     }
 810
 811     file_nb_sectors = bdrv_nb_sectors(bs->file->bs);
 812     if (file_nb_sectors < 0) {
 813         return -EINVAL;
 814     }
 815
 816     ret = bdrv_pread(bs->file, 0, sizeof(ph), &ph, 0);
 817     if (ret < 0) {
 818         goto fail;
 819     }
 820
 821     bs->total_sectors = le64_to_cpu(ph.nb_sectors);
 822
 823     if (le32_to_cpu(ph.version) != HEADER_VERSION) {
 824         goto fail_format;
 825     }
 826     if (!memcmp(ph.magic, HEADER_MAGIC, 16)) {
 827         s->off_multiplier = 1;
 828         bs->total_sectors = 0xffffffff & bs->total_sectors;
 829     } else if (!memcmp(ph.magic, HEADER_MAGIC2, 16)) {
 830         s->off_multiplier = le32_to_cpu(ph.tracks);
 831     } else {
 832         goto fail_format;
 833     }
 834
 835     s->tracks = le32_to_cpu(ph.tracks);
 836     if (s->tracks == 0) {
 837         error_setg(errp, "Invalid image: Zero sectors per track");
 838         ret = -EINVAL;
 839         goto fail;
 840     }
 841     if (s->tracks > INT32_MAX/513) {
 842         error_setg(errp, "Invalid image: Too big cluster");
 843         ret = -EFBIG;
 844         goto fail;
 845     }
 846     s->cluster_size = s->tracks << BDRV_SECTOR_BITS;
 847
 848     s->bat_size = le32_to_cpu(ph.bat_entries);
 849     if (s->bat_size > INT_MAX / sizeof(uint32_t)) {
 850         error_setg(errp, "Catalog too large");
 851         ret = -EFBIG;
 852         goto fail;
 853     }
 854
 855     size = bat_entry_off(s->bat_size);
 856     s->header_size = ROUND_UP(size, bdrv_opt_mem_align(bs->file->bs));
 857     s->header = qemu_try_blockalign(bs->file->bs, s->header_size);
 858     if (s->header == NULL) {
 859         ret = -ENOMEM;
 860         goto fail;
 861     }
 862     s->data_end = le32_to_cpu(ph.data_off);
 863     if (s->data_end == 0) {
 864         s->data_end = ROUND_UP(bat_entry_off(s->bat_size), BDRV_SECTOR_SIZE);
 865     }
 866     if (s->data_end < s->header_size) {
 867         /* there is not enough unused space to fit to block align between BAT
 868            and actual data. We can't avoid read-modify-write... */
 869         s->header_size = size;
 870     }
 871
 872     ret = bdrv_pread(bs->file, 0, s->header_size, s->header, 0);
 873     if (ret < 0) {
 874         goto fail;
 875     }
 876     s->bat_bitmap = (uint32_t *)(s->header + 1);
 877
 878     for (i = 0; i < s->bat_size; i++) {
 879         int64_t off = bat2sect(s, i);
 880         if (off >= file_nb_sectors) {
 881             if (flags & BDRV_O_CHECK) {
 882                 continue;
 883             }
 884             error_setg(errp, "parallels: Offset %" PRIi64 " in BAT[%d] entry "
 885                        "is larger than file size (%" PRIi64 ")",
 886                        off << BDRV_SECTOR_BITS, i,
 887                        file_nb_sectors << BDRV_SECTOR_BITS);
 888             ret = -EINVAL;
 889             goto fail;
 890         }
 891         if (off >= s->data_end) {
 892             s->data_end = off + s->tracks;
 893         }
 894     }
 895
 896     if (le32_to_cpu(ph.inuse) == HEADER_INUSE_MAGIC) {
 897         /* Image was not closed correctly. The check is mandatory */
 898         s->header_unclean = true;
 899         if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_CHECK)) {
 900             error_setg(errp, "parallels: Image was not closed correctly; "
 901                        "cannot be opened read/write");
 902             ret = -EACCES;
 903             goto fail;
 904         }
 905     }
 906
 907     opts = qemu_opts_create(&parallels_runtime_opts, NULL, 0, errp);
 908     if (!opts) {
 909         goto fail_options;
 910     }
 911
 912     if (!qemu_opts_absorb_qdict(opts, options, errp)) {
 913         goto fail_options;
 914     }
 915
 916     s->prealloc_size =
 917         qemu_opt_get_size_del(opts, PARALLELS_OPT_PREALLOC_SIZE, 0);
 918     s->prealloc_size = MAX(s->tracks, s->prealloc_size >> BDRV_SECTOR_BITS);
 919     buf = qemu_opt_get_del(opts, PARALLELS_OPT_PREALLOC_MODE);
 920     /* prealloc_mode can be downgraded later during allocate_clusters */
 921     s->prealloc_mode = qapi_enum_parse(&prealloc_mode_lookup, buf,
 922                                        PRL_PREALLOC_MODE_FALLOCATE,
 923                                        &local_err);
 924     g_free(buf);
 925     if (local_err != NULL) {
 926         error_propagate(errp, local_err);
 927         goto fail_options;
 928     }
 929
 930     if (ph.ext_off) {
 931         if (flags & BDRV_O_RDWR) {
 932             /*
 933              * It's unsafe to open image RW if there is an extension (as we
 934              * don't support it). But parallels driver in QEMU historically
 935              * ignores the extension, so print warning and don't care.
 936              */
 937             warn_report("Format Extension ignored in RW mode");
 938         } else {
 939             ret = parallels_read_format_extension(
 940                     bs, le64_to_cpu(ph.ext_off) << BDRV_SECTOR_BITS, errp);
 941             if (ret < 0) {
 942                 goto fail;
 943             }
 944         }
 945     }
 946
 947     if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_INACTIVE)) {
 948         s->header->inuse = cpu_to_le32(HEADER_INUSE_MAGIC);
 949         ret = parallels_update_header(bs);
 950         if (ret < 0) {
 951             goto fail;
 952         }
 953     }
 954
 955     s->bat_dirty_block = 4 * qemu_real_host_page_size();
 956     s->bat_dirty_bmap =
 957         bitmap_new(DIV_ROUND_UP(s->header_size, s->bat_dirty_block));
 958
 959     /* Disable migration until bdrv_activate method is added */
 960     error_setg(&s->migration_blocker, "The Parallels format used by node '%s' "
 961                "does not support live migration",
 962                bdrv_get_device_or_node_name(bs));
 963     ret = migrate_add_blocker(s->migration_blocker, errp);
 964     if (ret < 0) {
 965         error_free(s->migration_blocker);
 966         goto fail;
 967     }
 968     qemu_co_mutex_init(&s->lock);
 969     return 0;
 970
 971 fail_format:
 972     error_setg(errp, "Image not in Parallels format");
 973 fail_options:
 974     ret = -EINVAL;
 975 fail:
 976     qemu_vfree(s->header);
 977     return ret;
 978 }
 979
 980
 981 static void parallels_close(BlockDriverState *bs)
 982 {
 983     BDRVParallelsState *s = bs->opaque;
 984
 985     if ((bs->open_flags & BDRV_O_RDWR) && !(bs->open_flags & BDRV_O_INACTIVE)) {
 986         s->header->inuse = 0;
 987         parallels_update_header(bs);
 988
 989         /* errors are ignored, so we might as well pass exact=true */
 990         bdrv_truncate(bs->file, s->data_end << BDRV_SECTOR_BITS, true,
 991                       PREALLOC_MODE_OFF, 0, NULL);
 992     }
 993
 994     g_free(s->bat_dirty_bmap);
 995     qemu_vfree(s->header);
 996
 997     migrate_del_blocker(s->migration_blocker);
 998     error_free(s->migration_blocker);
 999 }
1000
1001 static BlockDriver bdrv_parallels = {
1002     .format_name        = "parallels",
1003     .instance_size      = sizeof(BDRVParallelsState),
1004     .bdrv_probe         = parallels_probe,
1005     .bdrv_open          = parallels_open,
1006     .bdrv_close         = parallels_close,
1007     .bdrv_child_perm          = bdrv_default_perms,
1008     .bdrv_co_block_status     = parallels_co_block_status,
1009     .bdrv_has_zero_init       = bdrv_has_zero_init_1,
1010     .bdrv_co_flush_to_os      = parallels_co_flush_to_os,
1011     .bdrv_co_readv  = parallels_co_readv,
1012     .bdrv_co_writev = parallels_co_writev,
1013     .is_format      = true,
1014     .supports_backing = true,
1015     .bdrv_co_create      = parallels_co_create,
1016     .bdrv_co_create_opts = parallels_co_create_opts,
1017     .bdrv_co_check  = parallels_co_check,
1018     .create_opts    = &parallels_create_opts,
1019 };
1020
1021 static void bdrv_parallels_init(void)
1022 {
1023     bdrv_register(&bdrv_parallels);
1024 }
1025
1026 block_init(bdrv_parallels_init);