sys/vfs/hammer/hammer_volume.c

   1 /*
   2  * Copyright (c) 2009 The DragonFly Project.  All rights reserved.
   3  *
   4  * This code is derived from software contributed to The DragonFly Project
   5  * by Matthew Dillon <dillon@backplane.com> and
   6  * Michael Neumann <mneumann@ntecs.de>
   7  *
   8  * Redistribution and use in source and binary forms, with or without
   9  * modification, are permitted provided that the following conditions
  10  * are met:
  11  *
  12  * 1. Redistributions of source code must retain the above copyright
  13  *    notice, this list of conditions and the following disclaimer.
  14  * 2. Redistributions in binary form must reproduce the above copyright
  15  *    notice, this list of conditions and the following disclaimer in
  16  *    the documentation and/or other materials provided with the
  17  *    distribution.
  18  * 3. Neither the name of The DragonFly Project nor the names of its
  19  *    contributors may be used to endorse or promote products derived
  20  *    from this software without specific, prior written permission.
  21  *
  22  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  23  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  24  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
  25  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
  26  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
  27  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
  28  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
  29  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
  30  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
  31  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
  32  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  33  * SUCH DAMAGE.
  34  *
  35  */
  36
  37 #include "hammer.h"
  38 #include <sys/fcntl.h>
  39 #include <sys/nlookup.h>
  40 #include <sys/buf.h>
  41
  42 static int
  43 hammer_setup_device(struct vnode **devvpp, const char *dev_path, int ronly);
  44
  45 static void
  46 hammer_close_device(struct vnode **devvpp, int ronly);
  47
  48 static int
  49 hammer_format_volume_header(struct hammer_mount *hmp, struct vnode *devvp,
  50         const char *vol_name, int vol_no, int vol_count,
  51         int64_t vol_size, int64_t boot_area_size, int64_t mem_area_size);
  52
  53 static int
  54 hammer_clear_volume_header(struct vnode *devvp);
  55
  56 static uint64_t
  57 hammer_format_freemap(hammer_transaction_t trans, hammer_volume_t volume);
  58
  59 static int
  60 hammer_free_freemap(hammer_transaction_t trans, hammer_volume_t volume);
  61
  62
  63 int
  64 hammer_ioc_volume_add(hammer_transaction_t trans, hammer_inode_t ip,
  65                 struct hammer_ioc_volume *ioc)
  66 {
  67         struct hammer_mount *hmp = trans->hmp;
  68         struct mount *mp = hmp->mp;
  69         hammer_volume_t volume;
  70         int error;
  71
  72         if (mp->mnt_flag & MNT_RDONLY) {
  73                 kprintf("Cannot add volume to read-only HAMMER filesystem\n");
  74                 return (EINVAL);
  75         }
  76
  77         if (hmp->nvolumes + 1 >= HAMMER_MAX_VOLUMES) {
  78                 kprintf("Max number of HAMMER volumes exceeded\n");
  79                 return (EINVAL);
  80         }
  81
  82         if (hammer_lock_ex_try(&hmp->volume_lock) != 0) {
  83                 kprintf("Another volume operation is in progress!\n");
  84                 return (EAGAIN);
  85         }
  86
  87         /*
  88          * Find an unused volume number.
  89          */
  90         int free_vol_no = 0;
  91         while (free_vol_no < HAMMER_MAX_VOLUMES &&
  92                RB_LOOKUP(hammer_vol_rb_tree, &hmp->rb_vols_root, free_vol_no)) {
  93                 ++free_vol_no;
  94         }
  95         if (free_vol_no >= HAMMER_MAX_VOLUMES) {
  96                 kprintf("Max number of HAMMER volumes exceeded\n");
  97                 hammer_unlock(&hmp->volume_lock);
  98                 return (EINVAL);
  99         }
 100
 101         struct vnode *devvp = NULL;
 102         error = hammer_setup_device(&devvp, ioc->device_name, 0);
 103         if (error)
 104                 goto end;
 105         KKASSERT(devvp);
 106         error = hammer_format_volume_header(
 107                 hmp,
 108                 devvp,
 109                 hmp->rootvol->ondisk->vol_name,
 110                 free_vol_no,
 111                 hmp->nvolumes+1,
 112                 ioc->vol_size,
 113                 ioc->boot_area_size,
 114                 ioc->mem_area_size);
 115         hammer_close_device(&devvp, 0);
 116         if (error)
 117                 goto end;
 118
 119         error = hammer_install_volume(hmp, ioc->device_name, NULL);
 120         if (error)
 121                 goto end;
 122
 123         hammer_sync_lock_sh(trans);
 124         hammer_lock_ex(&hmp->blkmap_lock);
 125
 126         ++hmp->nvolumes;
 127
 128         /*
 129          * Set each volumes new value of the vol_count field.
 130          */
 131         for (int vol_no = 0; vol_no < HAMMER_MAX_VOLUMES; ++vol_no) {
 132                 volume = hammer_get_volume(hmp, vol_no, &error);
 133                 if (volume == NULL && error == ENOENT) {
 134                         /*
 135                          * Skip unused volume numbers
 136                          */
 137                         error = 0;
 138                         continue;
 139                 }
 140                 KKASSERT(volume != NULL && error == 0);
 141                 hammer_modify_volume_field(trans, volume, vol_count);
 142                 volume->ondisk->vol_count = hmp->nvolumes;
 143                 hammer_modify_volume_done(volume);
 144
 145                 /*
 146                  * Only changes to the header of the root volume
 147                  * are automatically flushed to disk. For all
 148                  * other volumes that we modify we do it here.
 149                  */
 150                 if (volume != trans->rootvol && volume->io.modified) {
 151                         hammer_crc_set_volume(volume->ondisk);
 152                         hammer_io_flush(&volume->io, 0);
 153                 }
 154
 155                 hammer_rel_volume(volume, 0);
 156         }
 157
 158         volume = hammer_get_volume(hmp, free_vol_no, &error);
 159         KKASSERT(volume != NULL && error == 0);
 160
 161         uint64_t total_free_bigblocks =
 162                 hammer_format_freemap(trans, volume);
 163
 164         /*
 165          * Increase the total number of bigblocks
 166          */
 167         hammer_modify_volume_field(trans, trans->rootvol,
 168                 vol0_stat_bigblocks);
 169         trans->rootvol->ondisk->vol0_stat_bigblocks += total_free_bigblocks;
 170         hammer_modify_volume_done(trans->rootvol);
 171
 172         /*
 173          * Increase the number of free bigblocks
 174          * (including the copy in hmp)
 175          */
 176         hammer_modify_volume_field(trans, trans->rootvol,
 177                 vol0_stat_freebigblocks);
 178         trans->rootvol->ondisk->vol0_stat_freebigblocks += total_free_bigblocks;
 179         hmp->copy_stat_freebigblocks =
 180                 trans->rootvol->ondisk->vol0_stat_freebigblocks;
 181         hammer_modify_volume_done(trans->rootvol);
 182
 183         hammer_rel_volume(volume, 0);
 184
 185         hammer_unlock(&hmp->blkmap_lock);
 186         hammer_sync_unlock(trans);
 187
 188         KKASSERT(error == 0);
 189 end:
 190         hammer_unlock(&hmp->volume_lock);
 191         if (error)
 192                 kprintf("An error occurred: %d\n", error);
 193         return (error);
 194 }
 195
 196
 197 /*
 198  * Remove a volume.
 199  */
 200 int
 201 hammer_ioc_volume_del(hammer_transaction_t trans, hammer_inode_t ip,
 202                 struct hammer_ioc_volume *ioc)
 203 {
 204         struct hammer_mount *hmp = trans->hmp;
 205         struct mount *mp = hmp->mp;
 206         hammer_volume_t volume;
 207         int error = 0;
 208
 209         if (mp->mnt_flag & MNT_RDONLY) {
 210                 kprintf("Cannot del volume from read-only HAMMER filesystem\n");
 211                 return (EINVAL);
 212         }
 213
 214         if (hammer_lock_ex_try(&hmp->volume_lock) != 0) {
 215                 kprintf("Another volume operation is in progress!\n");
 216                 return (EAGAIN);
 217         }
 218
 219         volume = NULL;
 220
 221         /*
 222          * find volume by volname
 223          */
 224         for (int vol_no = 0; vol_no < HAMMER_MAX_VOLUMES; ++vol_no) {
 225                 volume = hammer_get_volume(hmp, vol_no, &error);
 226                 if (volume == NULL && error == ENOENT) {
 227                         /*
 228                          * Skip unused volume numbers
 229                          */
 230                         error = 0;
 231                         continue;
 232                 }
 233                 KKASSERT(volume != NULL && error == 0);
 234                 if (strcmp(volume->vol_name, ioc->device_name) == 0) {
 235                         break;
 236                 }
 237                 hammer_rel_volume(volume, 0);
 238                 volume = NULL;
 239         }
 240
 241         if (volume == NULL) {
 242                 kprintf("Couldn't find volume\n");
 243                 error = EINVAL;
 244                 goto end;
 245         }
 246
 247         if (volume == trans->rootvol) {
 248                 kprintf("Cannot remove root-volume\n");
 249                 hammer_rel_volume(volume, 0);
 250                 error = EINVAL;
 251                 goto end;
 252         }
 253
 254         /*
 255          *
 256          */
 257
 258         hmp->volume_to_remove = volume->vol_no;
 259
 260         struct hammer_ioc_reblock reblock;
 261         bzero(&reblock, sizeof(reblock));
 262
 263         reblock.key_beg.localization = HAMMER_MIN_LOCALIZATION;
 264         reblock.key_beg.obj_id = HAMMER_MIN_OBJID;
 265         reblock.key_end.localization = HAMMER_MAX_LOCALIZATION;
 266         reblock.key_end.obj_id = HAMMER_MAX_OBJID;
 267         reblock.head.flags = HAMMER_IOC_DO_FLAGS;
 268         reblock.free_level = 0;
 269
 270         error = hammer_ioc_reblock(trans, ip, &reblock);
 271
 272         if (reblock.head.flags & HAMMER_IOC_HEAD_INTR) {
 273                 error = EINTR;
 274         }
 275
 276         if (error) {
 277                 if (error == EINTR) {
 278                         kprintf("reblock was interrupted\n");
 279                 } else {
 280                         kprintf("reblock failed: %d\n", error);
 281                 }
 282                 hmp->volume_to_remove = -1;
 283                 hammer_rel_volume(volume, 0);
 284                 goto end;
 285         }
 286
 287         /*
 288          * Sync filesystem
 289          */
 290         int count = 0;
 291         while (hammer_flusher_haswork(hmp)) {
 292                 hammer_flusher_sync(hmp);
 293                 ++count;
 294                 if (count >= 5) {
 295                         if (count == 5)
 296                                 kprintf("HAMMER: flushing.");
 297                         else
 298                                 kprintf(".");
 299                         tsleep(&count, 0, "hmrufl", hz);
 300                 }
 301                 if (count == 30) {
 302                         kprintf("giving up");
 303                         break;
 304                 }
 305         }
 306         kprintf("\n");
 307
 308         hammer_sync_lock_sh(trans);
 309         hammer_lock_ex(&hmp->blkmap_lock);
 310
 311         error = hammer_free_freemap(trans, volume);
 312         if (error) {
 313                 kprintf("Failed to free volume. Volume not empty!\n");
 314                 hmp->volume_to_remove = -1;
 315                 hammer_rel_volume(volume, 0);
 316                 hammer_unlock(&hmp->blkmap_lock);
 317                 hammer_sync_unlock(trans);
 318                 goto end;
 319         }
 320
 321         hmp->volume_to_remove = -1;
 322
 323         hammer_rel_volume(volume, 0);
 324
 325         /*
 326          * Unload buffers
 327          */
 328         RB_SCAN(hammer_buf_rb_tree, &hmp->rb_bufs_root, NULL,
 329                 hammer_unload_buffer, volume);
 330
 331         error = hammer_unload_volume(volume, NULL);
 332         if (error == -1) {
 333                 kprintf("Failed to unload volume\n");
 334                 hammer_unlock(&hmp->blkmap_lock);
 335                 hammer_sync_unlock(trans);
 336                 goto end;
 337         }
 338
 339         volume = NULL;
 340         --hmp->nvolumes;
 341
 342         /*
 343          * Set each volume's new value of the vol_count field.
 344          */
 345         for (int vol_no = 0; vol_no < HAMMER_MAX_VOLUMES; ++vol_no) {
 346                 volume = hammer_get_volume(hmp, vol_no, &error);
 347                 if (volume == NULL && error == ENOENT) {
 348                         /*
 349                          * Skip unused volume numbers
 350                          */
 351                         error = 0;
 352                         continue;
 353                 }
 354
 355                 KKASSERT(volume != NULL && error == 0);
 356                 hammer_modify_volume_field(trans, volume, vol_count);
 357                 volume->ondisk->vol_count = hmp->nvolumes;
 358                 hammer_modify_volume_done(volume);
 359
 360                 /*
 361                  * Only changes to the header of the root volume
 362                  * are automatically flushed to disk. For all
 363                  * other volumes that we modify we do it here.
 364                  */
 365                 if (volume != trans->rootvol && volume->io.modified) {
 366                         hammer_crc_set_volume(volume->ondisk);
 367                         hammer_io_flush(&volume->io, 0);
 368                 }
 369
 370                 hammer_rel_volume(volume, 0);
 371         }
 372
 373         hammer_unlock(&hmp->blkmap_lock);
 374         hammer_sync_unlock(trans);
 375
 376         /*
 377          * Erase the volume header of the removed device.
 378          *
 379          * This is to not accidentally mount the volume again.
 380          */
 381         struct vnode *devvp = NULL;
 382         error = hammer_setup_device(&devvp, ioc->device_name, 0);
 383         if (error) {
 384                 kprintf("Failed to open device: %s\n", ioc->device_name);
 385                 goto end;
 386         }
 387         KKASSERT(devvp);
 388         error = hammer_clear_volume_header(devvp);
 389         if (error) {
 390                 kprintf("Failed to clear volume header of device: %s\n",
 391                         ioc->device_name);
 392                 goto end;
 393         }
 394         hammer_close_device(&devvp, 0);
 395
 396         KKASSERT(error == 0);
 397 end:
 398         hammer_unlock(&hmp->volume_lock);
 399         return (error);
 400 }
 401
 402
 403 /*
 404  * Iterate over all usable L1 entries of the volume and
 405  * the corresponding L2 entries.
 406  */
 407 static int
 408 hammer_iterate_l1l2_entries(hammer_transaction_t trans, hammer_volume_t volume,
 409         int (*callback)(hammer_transaction_t, hammer_volume_t, hammer_buffer_t*,
 410                 struct hammer_blockmap_layer1*, struct hammer_blockmap_layer2*,
 411                 hammer_off_t, hammer_off_t, void*),
 412         void *data)
 413 {
 414         struct hammer_mount *hmp = trans->hmp;
 415         hammer_blockmap_t freemap = &hmp->blockmap[HAMMER_ZONE_FREEMAP_INDEX];
 416         hammer_buffer_t buffer = NULL;
 417         int error = 0;
 418
 419         hammer_off_t phys_off;
 420         hammer_off_t block_off;
 421         hammer_off_t layer1_off;
 422         hammer_off_t layer2_off;
 423         hammer_off_t aligned_buf_end_off;
 424         struct hammer_blockmap_layer1 *layer1;
 425         struct hammer_blockmap_layer2 *layer2;
 426
 427         /*
 428          * Calculate the usable size of the volume, which
 429          * must be aligned at a bigblock (8 MB) boundary.
 430          */
 431         aligned_buf_end_off = (HAMMER_ENCODE_RAW_BUFFER(volume->ondisk->vol_no,
 432                 (volume->ondisk->vol_buf_end - volume->ondisk->vol_buf_beg)
 433                 & ~HAMMER_LARGEBLOCK_MASK64));
 434
 435         /*
 436          * Iterate the volume's address space in chunks of 4 TB, where each
 437          * chunk consists of at least one physically available 8 MB bigblock.
 438          *
 439          * For each chunk we need one L1 entry and one L2 bigblock.
 440          * We use the first bigblock of each chunk as L2 block.
 441          */
 442         for (phys_off = HAMMER_ENCODE_RAW_BUFFER(volume->ondisk->vol_no, 0);
 443              phys_off < aligned_buf_end_off;
 444              phys_off += HAMMER_BLOCKMAP_LAYER2) {
 445                 for (block_off = 0;
 446                      block_off < HAMMER_BLOCKMAP_LAYER2;
 447                      block_off += HAMMER_LARGEBLOCK_SIZE) {
 448                         layer2_off = phys_off +
 449                                 HAMMER_BLOCKMAP_LAYER2_OFFSET(block_off);
 450                         layer2 = hammer_bread(hmp, layer2_off, &error, &buffer);
 451                         if (error)
 452                                 goto end;
 453
 454                         error = callback(trans, volume, &buffer, NULL,
 455                                          layer2, phys_off, block_off, data);
 456                         if (error)
 457                                 goto end;
 458                 }
 459
 460                 layer1_off = freemap->phys_offset +
 461                                 HAMMER_BLOCKMAP_LAYER1_OFFSET(phys_off);
 462                 layer1 = hammer_bread(hmp, layer1_off, &error, &buffer);
 463                 if (error)
 464                         goto end;
 465
 466                 error = callback(trans, volume, &buffer, layer1, NULL,
 467                                  phys_off, 0, data);
 468                 if (error)
 469                         goto end;
 470         }
 471
 472 end:
 473         if (buffer) {
 474                 hammer_rel_buffer(buffer, 0);
 475                 buffer = NULL;
 476         }
 477
 478         return error;
 479 }
 480
 481 struct format_bigblock_stat {
 482         uint64_t total_free_bigblocks;
 483         uint64_t free_bigblocks;
 484 };
 485
 486 static int
 487 format_callback(hammer_transaction_t trans, hammer_volume_t volume,
 488         hammer_buffer_t *bufferp,
 489         struct hammer_blockmap_layer1 *layer1,
 490         struct hammer_blockmap_layer2 *layer2,
 491         hammer_off_t phys_off,
 492         hammer_off_t block_off,
 493         void *data)
 494 {
 495         struct format_bigblock_stat *stat = (struct format_bigblock_stat*)data;
 496
 497         /*
 498          * Calculate the usable size of the volume, which must be aligned
 499          * at a bigblock (8 MB) boundary.
 500          */
 501         hammer_off_t aligned_buf_end_off;
 502         aligned_buf_end_off = (HAMMER_ENCODE_RAW_BUFFER(volume->ondisk->vol_no,
 503                 (volume->ondisk->vol_buf_end - volume->ondisk->vol_buf_beg)
 504                 & ~HAMMER_LARGEBLOCK_MASK64));
 505
 506         if (layer1) {
 507                 KKASSERT(layer1->phys_offset == HAMMER_BLOCKMAP_UNAVAIL);
 508
 509                 hammer_modify_buffer(trans, *bufferp, layer1, sizeof(*layer1));
 510                 bzero(layer1, sizeof(layer1));
 511                 layer1->phys_offset = phys_off;
 512                 layer1->blocks_free = stat->free_bigblocks;
 513                 layer1->layer1_crc = crc32(layer1, HAMMER_LAYER1_CRCSIZE);
 514                 hammer_modify_buffer_done(*bufferp);
 515
 516                 stat->total_free_bigblocks += stat->free_bigblocks;
 517                 stat->free_bigblocks = 0; /* reset */
 518         } else if (layer2) {
 519                 hammer_modify_buffer(trans, *bufferp, layer2, sizeof(*layer2));
 520                 bzero(layer2, sizeof(*layer2));
 521
 522                 if (block_off == 0) {
 523                         /*
 524                          * The first entry represents the L2 bigblock itself.
 525                          */
 526                         layer2->zone = HAMMER_ZONE_FREEMAP_INDEX;
 527                         layer2->append_off = HAMMER_LARGEBLOCK_SIZE;
 528                         layer2->bytes_free = 0;
 529                 } else if (phys_off + block_off < aligned_buf_end_off) {
 530                         /*
 531                          * Available bigblock
 532                          */
 533                         layer2->zone = 0;
 534                         layer2->append_off = 0;
 535                         layer2->bytes_free = HAMMER_LARGEBLOCK_SIZE;
 536                         ++stat->free_bigblocks;
 537                 } else {
 538                         /*
 539                          * Bigblock outside of physically available
 540                          * space
 541                          */
 542                         layer2->zone = HAMMER_ZONE_UNAVAIL_INDEX;
 543                         layer2->append_off = HAMMER_LARGEBLOCK_SIZE;
 544                         layer2->bytes_free = 0;
 545                 }
 546
 547                 layer2->entry_crc = crc32(layer2, HAMMER_LAYER2_CRCSIZE);
 548                 hammer_modify_buffer_done(*bufferp);
 549         } else {
 550                 KKASSERT(0);
 551         }
 552
 553         return 0;
 554 }
 555
 556 static uint64_t
 557 hammer_format_freemap(hammer_transaction_t trans, hammer_volume_t volume)
 558 {
 559         int error = 0;
 560
 561         struct format_bigblock_stat stat;
 562         stat.total_free_bigblocks = 0;
 563         stat.free_bigblocks = 0;
 564
 565         error = hammer_iterate_l1l2_entries(trans, volume, format_callback,
 566                                         (void*)&stat);
 567         KKASSERT(error == 0);
 568
 569         return stat.total_free_bigblocks;
 570 }
 571
 572 static int
 573 free_callback(hammer_transaction_t trans, hammer_volume_t volume __unused,
 574         hammer_buffer_t *bufferp,
 575         struct hammer_blockmap_layer1 *layer1,
 576         struct hammer_blockmap_layer2 *layer2,
 577         hammer_off_t phys_off,
 578         hammer_off_t block_off __unused,
 579         void *data)
 580 {
 581         /*
 582          * No modifications to ondisk structures
 583          */
 584         int testonly = (data != NULL);
 585
 586         if (layer1) {
 587                 if (layer1->phys_offset == HAMMER_BLOCKMAP_UNAVAIL) {
 588                         /*
 589                          * This layer1 entry is already free.
 590                          */
 591                         return 0;
 592                 }
 593
 594                 KKASSERT((int)HAMMER_VOL_DECODE(layer1->phys_offset) ==
 595                         trans->hmp->volume_to_remove);
 596
 597                 if (testonly)
 598                         return 0;
 599
 600                 /*
 601                  * Free the L1 entry
 602                  */
 603                 hammer_modify_buffer(trans, *bufferp, layer1, sizeof(*layer1));
 604                 bzero(layer1, sizeof(layer1));
 605                 layer1->phys_offset = HAMMER_BLOCKMAP_UNAVAIL;
 606                 layer1->layer1_crc = crc32(layer1, HAMMER_LAYER1_CRCSIZE);
 607                 hammer_modify_buffer_done(*bufferp);
 608
 609                 return 0;
 610         } else if (layer2) {
 611                 if (layer2->zone == HAMMER_ZONE_FREEMAP_INDEX ||
 612                     layer2->zone == HAMMER_ZONE_UNAVAIL_INDEX)
 613                         return 0;
 614
 615                 if (layer2->append_off == 0 &&
 616                     layer2->bytes_free == HAMMER_LARGEBLOCK_SIZE)
 617                         return 0;
 618                 /*
 619                  * We found a layer2 entry that is not empty!
 620                  */
 621                 return EBUSY;
 622         } else {
 623                 KKASSERT(0);
 624         }
 625
 626         return EINVAL;
 627 }
 628
 629 static int
 630 hammer_free_freemap(hammer_transaction_t trans, hammer_volume_t volume)
 631 {
 632         int error;
 633         error = hammer_iterate_l1l2_entries(trans, volume, free_callback,
 634                                             (void*)1);
 635         if (error)
 636                 return error;
 637
 638         error = hammer_iterate_l1l2_entries(trans, volume, free_callback, NULL);
 639         return error;
 640 }
 641
 642 /************************************************************************
 643  *                              MISC                                    *
 644  ************************************************************************
 645  */
 646
 647 static int
 648 hammer_setup_device(struct vnode **devvpp, const char *dev_path, int ronly)
 649 {
 650         int error;
 651         struct nlookupdata nd;
 652
 653         /*
 654          * Get the device vnode
 655          */
 656         if (*devvpp == NULL) {
 657                 error = nlookup_init(&nd, dev_path, UIO_SYSSPACE, NLC_FOLLOW);
 658                 if (error == 0)
 659                         error = nlookup(&nd);
 660                 if (error == 0)
 661                         error = cache_vref(&nd.nl_nch, nd.nl_cred, devvpp);
 662                 nlookup_done(&nd);
 663         } else {
 664                 error = 0;
 665         }
 666
 667         if (error == 0) {
 668                 if (vn_isdisk(*devvpp, &error)) {
 669                         error = vfs_mountedon(*devvpp);
 670                 }
 671         }
 672         if (error == 0 && vcount(*devvpp) > 0)
 673                 error = EBUSY;
 674         if (error == 0) {
 675                 vn_lock(*devvpp, LK_EXCLUSIVE | LK_RETRY);
 676                 error = vinvalbuf(*devvpp, V_SAVE, 0, 0);
 677                 if (error == 0) {
 678                         error = VOP_OPEN(*devvpp,
 679                                          (ronly ? FREAD : FREAD|FWRITE),
 680                                          FSCRED, NULL);
 681                 }
 682                 vn_unlock(*devvpp);
 683         }
 684         if (error && *devvpp) {
 685                 vrele(*devvpp);
 686                 *devvpp = NULL;
 687         }
 688         return (error);
 689 }
 690
 691 static void
 692 hammer_close_device(struct vnode **devvpp, int ronly)
 693 {
 694         VOP_CLOSE(*devvpp, (ronly ? FREAD : FREAD|FWRITE));
 695         if (*devvpp) {
 696                 vinvalbuf(*devvpp, ronly ? 0 : V_SAVE, 0, 0);
 697                 vrele(*devvpp);
 698                 *devvpp = NULL;
 699         }
 700 }
 701
 702 static int
 703 hammer_format_volume_header(struct hammer_mount *hmp, struct vnode *devvp,
 704         const char *vol_name, int vol_no, int vol_count,
 705         int64_t vol_size, int64_t boot_area_size, int64_t mem_area_size)
 706 {
 707         struct buf *bp = NULL;
 708         struct hammer_volume_ondisk *ondisk;
 709         int error;
 710
 711         /*
 712          * Extract the volume number from the volume header and do various
 713          * sanity checks.
 714          */
 715         KKASSERT(HAMMER_BUFSIZE >= sizeof(struct hammer_volume_ondisk));
 716         error = bread(devvp, 0LL, HAMMER_BUFSIZE, &bp);
 717         if (error || bp->b_bcount < sizeof(struct hammer_volume_ondisk))
 718                 goto late_failure;
 719
 720         ondisk = (struct hammer_volume_ondisk*) bp->b_data;
 721
 722         /*
 723          * Note that we do NOT allow to use a device that contains
 724          * a valid HAMMER signature. It has to be cleaned up with dd
 725          * before.
 726          */
 727         if (ondisk->vol_signature == HAMMER_FSBUF_VOLUME) {
 728                 kprintf("hammer_volume_add: Formatting of valid HAMMER volume "
 729                         "%s denied. Erase with dd!\n", vol_name);
 730                 error = EFTYPE;
 731                 goto late_failure;
 732         }
 733
 734         bzero(ondisk, sizeof(struct hammer_volume_ondisk));
 735         ksnprintf(ondisk->vol_name, sizeof(ondisk->vol_name), "%s", vol_name);
 736         ondisk->vol_fstype = hmp->rootvol->ondisk->vol_fstype;
 737         ondisk->vol_signature = HAMMER_FSBUF_VOLUME;
 738         ondisk->vol_fsid = hmp->fsid;
 739         ondisk->vol_rootvol = hmp->rootvol->vol_no;
 740         ondisk->vol_no = vol_no;
 741         ondisk->vol_count = vol_count;
 742         ondisk->vol_version = hmp->version;
 743
 744         /*
 745          * Reserve space for (future) header junk, setup our poor-man's
 746          * bigblock allocator.
 747          */
 748         int64_t vol_alloc = HAMMER_BUFSIZE * 16;
 749
 750         ondisk->vol_bot_beg = vol_alloc;
 751         vol_alloc += boot_area_size;
 752         ondisk->vol_mem_beg = vol_alloc;
 753         vol_alloc += mem_area_size;
 754
 755         /*
 756          * The remaining area is the zone 2 buffer allocation area.  These
 757          * buffers
 758          */
 759         ondisk->vol_buf_beg = vol_alloc;
 760         ondisk->vol_buf_end = vol_size & ~(int64_t)HAMMER_BUFMASK;
 761
 762         if (ondisk->vol_buf_end < ondisk->vol_buf_beg) {
 763                 kprintf("volume %d %s is too small to hold the volume header",
 764                      ondisk->vol_no, ondisk->vol_name);
 765                 error = EFTYPE;
 766                 goto late_failure;
 767         }
 768
 769         ondisk->vol_nblocks = (ondisk->vol_buf_end - ondisk->vol_buf_beg) /
 770                               HAMMER_BUFSIZE;
 771         ondisk->vol_blocksize = HAMMER_BUFSIZE;
 772
 773         /*
 774          * Write volume header to disk
 775          */
 776         error = bwrite(bp);
 777         bp = NULL;
 778
 779 late_failure:
 780         if (bp)
 781                 brelse(bp);
 782         return (error);
 783 }
 784
 785 /*
 786  * Invalidates the volume header. Used by volume-del.
 787  */
 788 static int
 789 hammer_clear_volume_header(struct vnode *devvp)
 790 {
 791         struct buf *bp = NULL;
 792         struct hammer_volume_ondisk *ondisk;
 793         int error;
 794
 795         KKASSERT(HAMMER_BUFSIZE >= sizeof(struct hammer_volume_ondisk));
 796         error = bread(devvp, 0LL, HAMMER_BUFSIZE, &bp);
 797         if (error || bp->b_bcount < sizeof(struct hammer_volume_ondisk))
 798                 goto late_failure;
 799
 800         ondisk = (struct hammer_volume_ondisk*) bp->b_data;
 801         bzero(ondisk, sizeof(struct hammer_volume_ondisk));
 802
 803         error = bwrite(bp);
 804         bp = NULL;
 805
 806 late_failure:
 807         if (bp)
 808                 brelse(bp);
 809         return (error);
 810 }