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 #include "qemu-common.h"
  31 #include "block/block_int.h"
  32 #include "qemu/module.h"
  33
  34 /**************************************************************/
  35
  36 #define HEADER_MAGIC "WithoutFreeSpace"
  37 #define HEADER_MAGIC2 "WithouFreSpacExt"
  38 #define HEADER_VERSION 2
  39 #define HEADER_INUSE_MAGIC  (0x746F6E59)
  40
  41 #define DEFAULT_CLUSTER_SIZE 1048576        /* 1 MiB */
  42
  43
  44 // always little-endian
  45 typedef struct ParallelsHeader {
  46     char magic[16]; // "WithoutFreeSpace"
  47     uint32_t version;
  48     uint32_t heads;
  49     uint32_t cylinders;
  50     uint32_t tracks;
  51     uint32_t bat_entries;
  52     uint64_t nb_sectors;
  53     uint32_t inuse;
  54     uint32_t data_off;
  55     char padding[12];
  56 } QEMU_PACKED ParallelsHeader;
  57
  58 typedef struct BDRVParallelsState {
  59     /** Locking is conservative, the lock protects
  60      *   - image file extending (truncate, fallocate)
  61      *   - any access to block allocation table
  62      */
  63     CoMutex lock;
  64
  65     ParallelsHeader *header;
  66     uint32_t header_size;
  67     bool header_unclean;
  68
  69     uint32_t *bat_bitmap;
  70     unsigned int bat_size;
  71
  72     unsigned int tracks;
  73
  74     unsigned int off_multiplier;
  75
  76     bool has_truncate;
  77 } BDRVParallelsState;
  78
  79
  80 static int64_t bat2sect(BDRVParallelsState *s, uint32_t idx)
  81 {
  82     return (uint64_t)le32_to_cpu(s->bat_bitmap[idx]) * s->off_multiplier;
  83 }
  84
  85 static int64_t seek_to_sector(BDRVParallelsState *s, int64_t sector_num)
  86 {
  87     uint32_t index, offset;
  88
  89     index = sector_num / s->tracks;
  90     offset = sector_num % s->tracks;
  91
  92     /* not allocated */
  93     if ((index >= s->bat_size) || (s->bat_bitmap[index] == 0)) {
  94         return -1;
  95     }
  96     return bat2sect(s, index) + offset;
  97 }
  98
  99 static int cluster_remainder(BDRVParallelsState *s, int64_t sector_num,
 100         int nb_sectors)
 101 {
 102     int ret = s->tracks - sector_num % s->tracks;
 103     return MIN(nb_sectors, ret);
 104 }
 105
 106 static int64_t block_status(BDRVParallelsState *s, int64_t sector_num,
 107                             int nb_sectors, int *pnum)
 108 {
 109     int64_t start_off = -2, prev_end_off = -2;
 110
 111     *pnum = 0;
 112     while (nb_sectors > 0 || start_off == -2) {
 113         int64_t offset = seek_to_sector(s, sector_num);
 114         int to_end;
 115
 116         if (start_off == -2) {
 117             start_off = offset;
 118             prev_end_off = offset;
 119         } else if (offset != prev_end_off) {
 120             break;
 121         }
 122
 123         to_end = cluster_remainder(s, sector_num, nb_sectors);
 124         nb_sectors -= to_end;
 125         sector_num += to_end;
 126         *pnum += to_end;
 127
 128         if (offset > 0) {
 129             prev_end_off += to_end;
 130         }
 131     }
 132     return start_off;
 133 }
 134
 135 static int64_t allocate_cluster(BlockDriverState *bs, int64_t sector_num)
 136 {
 137     BDRVParallelsState *s = bs->opaque;
 138     uint32_t idx, offset;
 139     int64_t pos;
 140     int ret;
 141
 142     idx = sector_num / s->tracks;
 143     offset = sector_num % s->tracks;
 144
 145     if (idx >= s->bat_size) {
 146         return -EINVAL;
 147     }
 148     if (s->bat_bitmap[idx] != 0) {
 149         return bat2sect(s, idx) + offset;
 150     }
 151
 152     pos = bdrv_getlength(bs->file) >> BDRV_SECTOR_BITS;
 153     if (s->has_truncate) {
 154         ret = bdrv_truncate(bs->file, (pos + s->tracks) << BDRV_SECTOR_BITS);
 155     } else {
 156         ret = bdrv_write_zeroes(bs->file, pos, s->tracks, 0);
 157     }
 158     if (ret < 0) {
 159         return ret;
 160     }
 161
 162     s->bat_bitmap[idx] = cpu_to_le32(pos / s->off_multiplier);
 163     ret = bdrv_pwrite(bs->file,
 164             sizeof(ParallelsHeader) + idx * sizeof(s->bat_bitmap[idx]),
 165             s->bat_bitmap + idx, sizeof(s->bat_bitmap[idx]));
 166     if (ret < 0) {
 167         s->bat_bitmap[idx] = 0;
 168         return ret;
 169     }
 170     return bat2sect(s, idx) + offset;
 171 }
 172
 173 static int64_t coroutine_fn parallels_co_get_block_status(BlockDriverState *bs,
 174         int64_t sector_num, int nb_sectors, int *pnum)
 175 {
 176     BDRVParallelsState *s = bs->opaque;
 177     int64_t offset;
 178
 179     qemu_co_mutex_lock(&s->lock);
 180     offset = block_status(s, sector_num, nb_sectors, pnum);
 181     qemu_co_mutex_unlock(&s->lock);
 182
 183     if (offset < 0) {
 184         return 0;
 185     }
 186
 187     return (offset << BDRV_SECTOR_BITS) |
 188         BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
 189 }
 190
 191 static coroutine_fn int parallels_co_writev(BlockDriverState *bs,
 192         int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
 193 {
 194     BDRVParallelsState *s = bs->opaque;
 195     uint64_t bytes_done = 0;
 196     QEMUIOVector hd_qiov;
 197     int ret = 0;
 198
 199     qemu_iovec_init(&hd_qiov, qiov->niov);
 200
 201     while (nb_sectors > 0) {
 202         int64_t position;
 203         int n, nbytes;
 204
 205         qemu_co_mutex_lock(&s->lock);
 206         position = allocate_cluster(bs, sector_num);
 207         qemu_co_mutex_unlock(&s->lock);
 208         if (position < 0) {
 209             ret = (int)position;
 210             break;
 211         }
 212
 213         n = cluster_remainder(s, sector_num, nb_sectors);
 214         nbytes = n << BDRV_SECTOR_BITS;
 215
 216         qemu_iovec_reset(&hd_qiov);
 217         qemu_iovec_concat(&hd_qiov, qiov, bytes_done, nbytes);
 218
 219         ret = bdrv_co_writev(bs->file, position, n, &hd_qiov);
 220         if (ret < 0) {
 221             break;
 222         }
 223
 224         nb_sectors -= n;
 225         sector_num += n;
 226         bytes_done += nbytes;
 227     }
 228
 229     qemu_iovec_destroy(&hd_qiov);
 230     return ret;
 231 }
 232
 233 static coroutine_fn int parallels_co_readv(BlockDriverState *bs,
 234         int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
 235 {
 236     BDRVParallelsState *s = bs->opaque;
 237     uint64_t bytes_done = 0;
 238     QEMUIOVector hd_qiov;
 239     int ret = 0;
 240
 241     qemu_iovec_init(&hd_qiov, qiov->niov);
 242
 243     while (nb_sectors > 0) {
 244         int64_t position;
 245         int n, nbytes;
 246
 247         qemu_co_mutex_lock(&s->lock);
 248         position = block_status(s, sector_num, nb_sectors, &n);
 249         qemu_co_mutex_unlock(&s->lock);
 250
 251         nbytes = n << BDRV_SECTOR_BITS;
 252
 253         if (position < 0) {
 254             qemu_iovec_memset(qiov, bytes_done, 0, nbytes);
 255         } else {
 256             qemu_iovec_reset(&hd_qiov);
 257             qemu_iovec_concat(&hd_qiov, qiov, bytes_done, nbytes);
 258
 259             ret = bdrv_co_readv(bs->file, position, n, &hd_qiov);
 260             if (ret < 0) {
 261                 break;
 262             }
 263         }
 264
 265         nb_sectors -= n;
 266         sector_num += n;
 267         bytes_done += nbytes;
 268     }
 269
 270     qemu_iovec_destroy(&hd_qiov);
 271     return ret;
 272 }
 273
 274
 275 static int parallels_check(BlockDriverState *bs, BdrvCheckResult *res,
 276                            BdrvCheckMode fix)
 277 {
 278     BDRVParallelsState *s = bs->opaque;
 279     int64_t size, prev_off, high_off;
 280     int ret;
 281     uint32_t i;
 282     bool flush_bat = false;
 283     int cluster_size = s->tracks << BDRV_SECTOR_BITS;
 284
 285     size = bdrv_getlength(bs->file);
 286     if (size < 0) {
 287         res->check_errors++;
 288         return size;
 289     }
 290
 291     if (s->header_unclean) {
 292         fprintf(stderr, "%s image was not closed correctly\n",
 293                 fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR");
 294         res->corruptions++;
 295         if (fix & BDRV_FIX_ERRORS) {
 296             /* parallels_close will do the job right */
 297             res->corruptions_fixed++;
 298             s->header_unclean = false;
 299         }
 300     }
 301
 302     res->bfi.total_clusters = s->bat_size;
 303     res->bfi.compressed_clusters = 0; /* compression is not supported */
 304
 305     high_off = 0;
 306     prev_off = 0;
 307     for (i = 0; i < s->bat_size; i++) {
 308         int64_t off = bat2sect(s, i) << BDRV_SECTOR_BITS;
 309         if (off == 0) {
 310             prev_off = 0;
 311             continue;
 312         }
 313
 314         /* cluster outside the image */
 315         if (off > size) {
 316             fprintf(stderr, "%s cluster %u is outside image\n",
 317                     fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR", i);
 318             res->corruptions++;
 319             if (fix & BDRV_FIX_ERRORS) {
 320                 prev_off = 0;
 321                 s->bat_bitmap[i] = 0;
 322                 res->corruptions_fixed++;
 323                 flush_bat = true;
 324                 continue;
 325             }
 326         }
 327
 328         res->bfi.allocated_clusters++;
 329         if (off > high_off) {
 330             high_off = off;
 331         }
 332
 333         if (prev_off != 0 && (prev_off + cluster_size) != off) {
 334             res->bfi.fragmented_clusters++;
 335         }
 336         prev_off = off;
 337     }
 338
 339     if (flush_bat) {
 340         ret = bdrv_pwrite_sync(bs->file, 0, s->header, s->header_size);
 341         if (ret < 0) {
 342             res->check_errors++;
 343             return ret;
 344         }
 345     }
 346
 347     res->image_end_offset = high_off + cluster_size;
 348     if (size > res->image_end_offset) {
 349         int64_t count;
 350         count = DIV_ROUND_UP(size - res->image_end_offset, cluster_size);
 351         fprintf(stderr, "%s space leaked at the end of the image %" PRId64 "\n",
 352                 fix & BDRV_FIX_LEAKS ? "Repairing" : "ERROR",
 353                 size - res->image_end_offset);
 354         res->leaks += count;
 355         if (fix & BDRV_FIX_LEAKS) {
 356             ret = bdrv_truncate(bs->file, res->image_end_offset);
 357             if (ret < 0) {
 358                 res->check_errors++;
 359                 return ret;
 360             }
 361             res->leaks_fixed += count;
 362         }
 363     }
 364
 365     return 0;
 366 }
 367
 368
 369 static int parallels_create(const char *filename, QemuOpts *opts, Error **errp)
 370 {
 371     int64_t total_size, cl_size;
 372     uint8_t tmp[BDRV_SECTOR_SIZE];
 373     Error *local_err = NULL;
 374     BlockDriverState *file;
 375     uint32_t bat_entries, bat_sectors;
 376     ParallelsHeader header;
 377     int ret;
 378
 379     total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
 380                           BDRV_SECTOR_SIZE);
 381     cl_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE,
 382                           DEFAULT_CLUSTER_SIZE), BDRV_SECTOR_SIZE);
 383
 384     ret = bdrv_create_file(filename, opts, &local_err);
 385     if (ret < 0) {
 386         error_propagate(errp, local_err);
 387         return ret;
 388     }
 389
 390     file = NULL;
 391     ret = bdrv_open(&file, filename, NULL, NULL,
 392                     BDRV_O_RDWR | BDRV_O_PROTOCOL, NULL, &local_err);
 393     if (ret < 0) {
 394         error_propagate(errp, local_err);
 395         return ret;
 396     }
 397     ret = bdrv_truncate(file, 0);
 398     if (ret < 0) {
 399         goto exit;
 400     }
 401
 402     bat_entries = DIV_ROUND_UP(total_size, cl_size);
 403     bat_sectors = DIV_ROUND_UP(bat_entries * sizeof(uint32_t) +
 404                                sizeof(ParallelsHeader), cl_size);
 405     bat_sectors = (bat_sectors *  cl_size) >> BDRV_SECTOR_BITS;
 406
 407     memset(&header, 0, sizeof(header));
 408     memcpy(header.magic, HEADER_MAGIC2, sizeof(header.magic));
 409     header.version = cpu_to_le32(HEADER_VERSION);
 410     /* don't care much about geometry, it is not used on image level */
 411     header.heads = cpu_to_le32(16);
 412     header.cylinders = cpu_to_le32(total_size / BDRV_SECTOR_SIZE / 16 / 32);
 413     header.tracks = cpu_to_le32(cl_size >> BDRV_SECTOR_BITS);
 414     header.bat_entries = cpu_to_le32(bat_entries);
 415     header.nb_sectors = cpu_to_le64(DIV_ROUND_UP(total_size, BDRV_SECTOR_SIZE));
 416     header.data_off = cpu_to_le32(bat_sectors);
 417
 418     /* write all the data */
 419     memset(tmp, 0, sizeof(tmp));
 420     memcpy(tmp, &header, sizeof(header));
 421
 422     ret = bdrv_pwrite(file, 0, tmp, BDRV_SECTOR_SIZE);
 423     if (ret < 0) {
 424         goto exit;
 425     }
 426     ret = bdrv_write_zeroes(file, 1, bat_sectors - 1, 0);
 427     if (ret < 0) {
 428         goto exit;
 429     }
 430     ret = 0;
 431
 432 done:
 433     bdrv_unref(file);
 434     return ret;
 435
 436 exit:
 437     error_setg_errno(errp, -ret, "Failed to create Parallels image");
 438     goto done;
 439 }
 440
 441
 442 static int parallels_probe(const uint8_t *buf, int buf_size,
 443                            const char *filename)
 444 {
 445     const ParallelsHeader *ph = (const void *)buf;
 446
 447     if (buf_size < sizeof(ParallelsHeader)) {
 448         return 0;
 449     }
 450
 451     if ((!memcmp(ph->magic, HEADER_MAGIC, 16) ||
 452            !memcmp(ph->magic, HEADER_MAGIC2, 16)) &&
 453            (le32_to_cpu(ph->version) == HEADER_VERSION)) {
 454         return 100;
 455     }
 456
 457     return 0;
 458 }
 459
 460 static int parallels_update_header(BlockDriverState *bs)
 461 {
 462     BDRVParallelsState *s = bs->opaque;
 463     unsigned size = MAX(bdrv_opt_mem_align(bs->file), sizeof(ParallelsHeader));
 464
 465     if (size > s->header_size) {
 466         size = s->header_size;
 467     }
 468     return bdrv_pwrite_sync(bs->file, 0, s->header, size);
 469 }
 470
 471 static int parallels_open(BlockDriverState *bs, QDict *options, int flags,
 472                           Error **errp)
 473 {
 474     BDRVParallelsState *s = bs->opaque;
 475     ParallelsHeader ph;
 476     int ret, size;
 477
 478     ret = bdrv_pread(bs->file, 0, &ph, sizeof(ph));
 479     if (ret < 0) {
 480         goto fail;
 481     }
 482
 483     bs->total_sectors = le64_to_cpu(ph.nb_sectors);
 484
 485     if (le32_to_cpu(ph.version) != HEADER_VERSION) {
 486         goto fail_format;
 487     }
 488     if (!memcmp(ph.magic, HEADER_MAGIC, 16)) {
 489         s->off_multiplier = 1;
 490         bs->total_sectors = 0xffffffff & bs->total_sectors;
 491     } else if (!memcmp(ph.magic, HEADER_MAGIC2, 16)) {
 492         s->off_multiplier = le32_to_cpu(ph.tracks);
 493     } else {
 494         goto fail_format;
 495     }
 496
 497     s->tracks = le32_to_cpu(ph.tracks);
 498     if (s->tracks == 0) {
 499         error_setg(errp, "Invalid image: Zero sectors per track");
 500         ret = -EINVAL;
 501         goto fail;
 502     }
 503     if (s->tracks > INT32_MAX/513) {
 504         error_setg(errp, "Invalid image: Too big cluster");
 505         ret = -EFBIG;
 506         goto fail;
 507     }
 508
 509     s->bat_size = le32_to_cpu(ph.bat_entries);
 510     if (s->bat_size > INT_MAX / sizeof(uint32_t)) {
 511         error_setg(errp, "Catalog too large");
 512         ret = -EFBIG;
 513         goto fail;
 514     }
 515
 516     size = sizeof(ParallelsHeader) + sizeof(uint32_t) * s->bat_size;
 517     s->header_size = ROUND_UP(size, bdrv_opt_mem_align(bs->file));
 518     s->header = qemu_try_blockalign(bs->file, s->header_size);
 519     if (s->header == NULL) {
 520         ret = -ENOMEM;
 521         goto fail;
 522     }
 523     if (le32_to_cpu(ph.data_off) < s->header_size) {
 524         /* there is not enough unused space to fit to block align between BAT
 525            and actual data. We can't avoid read-modify-write... */
 526         s->header_size = size;
 527     }
 528
 529     ret = bdrv_pread(bs->file, 0, s->header, s->header_size);
 530     if (ret < 0) {
 531         goto fail;
 532     }
 533     s->bat_bitmap = (uint32_t *)(s->header + 1);
 534
 535     s->has_truncate = bdrv_has_zero_init(bs->file) &&
 536                       bdrv_truncate(bs->file, bdrv_getlength(bs->file)) == 0;
 537
 538     if (le32_to_cpu(ph.inuse) == HEADER_INUSE_MAGIC) {
 539         /* Image was not closed correctly. The check is mandatory */
 540         s->header_unclean = true;
 541         if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_CHECK)) {
 542             error_setg(errp, "parallels: Image was not closed correctly; "
 543                        "cannot be opened read/write");
 544             ret = -EACCES;
 545             goto fail;
 546         }
 547     }
 548
 549     if (flags & BDRV_O_RDWR) {
 550         s->header->inuse = cpu_to_le32(HEADER_INUSE_MAGIC);
 551         ret = parallels_update_header(bs);
 552         if (ret < 0) {
 553             goto fail;
 554         }
 555     }
 556
 557     qemu_co_mutex_init(&s->lock);
 558     return 0;
 559
 560 fail_format:
 561     error_setg(errp, "Image not in Parallels format");
 562     ret = -EINVAL;
 563 fail:
 564     qemu_vfree(s->header);
 565     return ret;
 566 }
 567
 568
 569 static void parallels_close(BlockDriverState *bs)
 570 {
 571     BDRVParallelsState *s = bs->opaque;
 572
 573     if (bs->open_flags & BDRV_O_RDWR) {
 574         s->header->inuse = 0;
 575         parallels_update_header(bs);
 576     }
 577
 578     qemu_vfree(s->header);
 579 }
 580
 581 static QemuOptsList parallels_create_opts = {
 582     .name = "parallels-create-opts",
 583     .head = QTAILQ_HEAD_INITIALIZER(parallels_create_opts.head),
 584     .desc = {
 585         {
 586             .name = BLOCK_OPT_SIZE,
 587             .type = QEMU_OPT_SIZE,
 588             .help = "Virtual disk size",
 589         },
 590         {
 591             .name = BLOCK_OPT_CLUSTER_SIZE,
 592             .type = QEMU_OPT_SIZE,
 593             .help = "Parallels image cluster size",
 594             .def_value_str = stringify(DEFAULT_CLUSTER_SIZE),
 595         },
 596         { /* end of list */ }
 597     }
 598 };
 599
 600 static BlockDriver bdrv_parallels = {
 601     .format_name        = "parallels",
 602     .instance_size      = sizeof(BDRVParallelsState),
 603     .bdrv_probe         = parallels_probe,
 604     .bdrv_open          = parallels_open,
 605     .bdrv_close         = parallels_close,
 606     .bdrv_co_get_block_status = parallels_co_get_block_status,
 607     .bdrv_has_zero_init       = bdrv_has_zero_init_1,
 608     .bdrv_co_readv  = parallels_co_readv,
 609     .bdrv_co_writev = parallels_co_writev,
 610
 611     .bdrv_create    = parallels_create,
 612     .bdrv_check     = parallels_check,
 613     .create_opts    = &parallels_create_opts,
 614 };
 615
 616 static void bdrv_parallels_init(void)
 617 {
 618     bdrv_register(&bdrv_parallels);
 619 }
 620
 621 block_init(bdrv_parallels_init);