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 #include <sys/buf2.h>
  43
  44 static int
  45 hammer_setup_device(struct vnode **devvpp, const char *dev_path, int ronly);
  46
  47 static void
  48 hammer_close_device(struct vnode **devvpp, int ronly);
  49
  50 static int
  51 hammer_format_volume_header(struct hammer_mount *hmp, struct vnode *devvp,
  52         const char *vol_name, int vol_no, int vol_count,
  53         int64_t vol_size, int64_t boot_area_size, int64_t mem_area_size);
  54
  55 static int
  56 hammer_clear_volume_header(struct vnode *devvp);
  57
  58 struct bigblock_stat {
  59         uint64_t total_bigblocks;
  60         uint64_t total_free_bigblocks;
  61         uint64_t counter;
  62 };
  63
  64 static int
  65 hammer_format_freemap(hammer_transaction_t trans, hammer_volume_t volume,
  66         struct bigblock_stat *stat);
  67
  68 static int
  69 hammer_free_freemap(hammer_transaction_t trans, hammer_volume_t volume,
  70         struct bigblock_stat *stat);
  71
  72 int
  73 hammer_ioc_volume_add(hammer_transaction_t trans, hammer_inode_t ip,
  74                 struct hammer_ioc_volume *ioc)
  75 {
  76         struct hammer_mount *hmp = trans->hmp;
  77         struct mount *mp = hmp->mp;
  78         hammer_volume_t volume;
  79         int error;
  80
  81         if (mp->mnt_flag & MNT_RDONLY) {
  82                 kprintf("Cannot add volume to read-only HAMMER filesystem\n");
  83                 return (EINVAL);
  84         }
  85
  86         if (hmp->nvolumes + 1 >= HAMMER_MAX_VOLUMES) {
  87                 kprintf("Max number of HAMMER volumes exceeded\n");
  88                 return (EINVAL);
  89         }
  90
  91         if (hammer_lock_ex_try(&hmp->volume_lock) != 0) {
  92                 kprintf("Another volume operation is in progress!\n");
  93                 return (EAGAIN);
  94         }
  95
  96         /*
  97          * Find an unused volume number.
  98          */
  99         int free_vol_no = 0;
 100         while (free_vol_no < HAMMER_MAX_VOLUMES &&
 101                RB_LOOKUP(hammer_vol_rb_tree, &hmp->rb_vols_root, free_vol_no)) {
 102                 ++free_vol_no;
 103         }
 104         if (free_vol_no >= HAMMER_MAX_VOLUMES) {
 105                 kprintf("Max number of HAMMER volumes exceeded\n");
 106                 hammer_unlock(&hmp->volume_lock);
 107                 return (EINVAL);
 108         }
 109
 110         struct vnode *devvp = NULL;
 111         error = hammer_setup_device(&devvp, ioc->device_name, 0);
 112         if (error)
 113                 goto end;
 114         KKASSERT(devvp);
 115         error = hammer_format_volume_header(
 116                 hmp,
 117                 devvp,
 118                 hmp->rootvol->ondisk->vol_name,
 119                 free_vol_no,
 120                 hmp->nvolumes+1,
 121                 ioc->vol_size,
 122                 ioc->boot_area_size,
 123                 ioc->mem_area_size);
 124         hammer_close_device(&devvp, 0);
 125         if (error)
 126                 goto end;
 127
 128         error = hammer_install_volume(hmp, ioc->device_name, NULL);
 129         if (error)
 130                 goto end;
 131
 132         hammer_sync_lock_sh(trans);
 133         hammer_lock_ex(&hmp->blkmap_lock);
 134
 135         ++hmp->nvolumes;
 136
 137         /*
 138          * Set each volumes new value of the vol_count field.
 139          */
 140         for (int vol_no = 0; vol_no < HAMMER_MAX_VOLUMES; ++vol_no) {
 141                 volume = hammer_get_volume(hmp, vol_no, &error);
 142                 if (volume == NULL && error == ENOENT) {
 143                         /*
 144                          * Skip unused volume numbers
 145                          */
 146                         error = 0;
 147                         continue;
 148                 }
 149                 KKASSERT(volume != NULL && error == 0);
 150                 hammer_modify_volume_field(trans, volume, vol_count);
 151                 volume->ondisk->vol_count = hmp->nvolumes;
 152                 hammer_modify_volume_done(volume);
 153
 154                 /*
 155                  * Only changes to the header of the root volume
 156                  * are automatically flushed to disk. For all
 157                  * other volumes that we modify we do it here.
 158                  *
 159                  * No interlock is needed, volume buffers are not
 160                  * messed with by bioops.
 161                  */
 162                 if (volume != trans->rootvol && volume->io.modified) {
 163                         hammer_crc_set_volume(volume->ondisk);
 164                         hammer_io_flush(&volume->io, 0);
 165                 }
 166
 167                 hammer_rel_volume(volume, 0);
 168         }
 169
 170         volume = hammer_get_volume(hmp, free_vol_no, &error);
 171         KKASSERT(volume != NULL && error == 0);
 172
 173         struct bigblock_stat stat;
 174         error = hammer_format_freemap(trans, volume, &stat);
 175         KKASSERT(error == 0);
 176
 177         /*
 178          * Increase the total number of bigblocks and update stat/vstat totals.
 179          */
 180         hammer_modify_volume_field(trans, trans->rootvol,
 181                 vol0_stat_bigblocks);
 182         trans->rootvol->ondisk->vol0_stat_bigblocks += stat.total_bigblocks;
 183         hammer_modify_volume_done(trans->rootvol);
 184         mp->mnt_stat.f_blocks += trans->rootvol->ondisk->vol0_stat_bigblocks *
 185             (HAMMER_LARGEBLOCK_SIZE / HAMMER_BUFSIZE);
 186         mp->mnt_vstat.f_blocks += trans->rootvol->ondisk->vol0_stat_bigblocks *
 187             (HAMMER_LARGEBLOCK_SIZE / HAMMER_BUFSIZE);
 188
 189         /*
 190          * Increase the number of free bigblocks
 191          * (including the copy in hmp)
 192          */
 193         hammer_modify_volume_field(trans, trans->rootvol,
 194                 vol0_stat_freebigblocks);
 195         trans->rootvol->ondisk->vol0_stat_freebigblocks += stat.total_free_bigblocks;
 196         hmp->copy_stat_freebigblocks =
 197                 trans->rootvol->ondisk->vol0_stat_freebigblocks;
 198         hammer_modify_volume_done(trans->rootvol);
 199
 200         hammer_rel_volume(volume, 0);
 201
 202         hammer_unlock(&hmp->blkmap_lock);
 203         hammer_sync_unlock(trans);
 204
 205         KKASSERT(error == 0);
 206 end:
 207         hammer_unlock(&hmp->volume_lock);
 208         if (error)
 209                 kprintf("An error occurred: %d\n", error);
 210         return (error);
 211 }
 212
 213
 214 /*
 215  * Remove a volume.
 216  */
 217 int
 218 hammer_ioc_volume_del(hammer_transaction_t trans, hammer_inode_t ip,
 219                 struct hammer_ioc_volume *ioc)
 220 {
 221         struct hammer_mount *hmp = trans->hmp;
 222         struct mount *mp = hmp->mp;
 223         hammer_volume_t volume;
 224         int error = 0;
 225
 226         if (mp->mnt_flag & MNT_RDONLY) {
 227                 kprintf("Cannot del volume from read-only HAMMER filesystem\n");
 228                 return (EINVAL);
 229         }
 230
 231         if (hammer_lock_ex_try(&hmp->volume_lock) != 0) {
 232                 kprintf("Another volume operation is in progress!\n");
 233                 return (EAGAIN);
 234         }
 235
 236         volume = NULL;
 237
 238         /*
 239          * find volume by volname
 240          */
 241         for (int vol_no = 0; vol_no < HAMMER_MAX_VOLUMES; ++vol_no) {
 242                 volume = hammer_get_volume(hmp, vol_no, &error);
 243                 if (volume == NULL && error == ENOENT) {
 244                         /*
 245                          * Skip unused volume numbers
 246                          */
 247                         error = 0;
 248                         continue;
 249                 }
 250                 KKASSERT(volume != NULL && error == 0);
 251                 if (strcmp(volume->vol_name, ioc->device_name) == 0) {
 252                         break;
 253                 }
 254                 hammer_rel_volume(volume, 0);
 255                 volume = NULL;
 256         }
 257
 258         if (volume == NULL) {
 259                 kprintf("Couldn't find volume\n");
 260                 error = EINVAL;
 261                 goto end;
 262         }
 263
 264         if (volume == trans->rootvol) {
 265                 kprintf("Cannot remove root-volume\n");
 266                 hammer_rel_volume(volume, 0);
 267                 error = EINVAL;
 268                 goto end;
 269         }
 270
 271         /*
 272          *
 273          */
 274
 275         hmp->volume_to_remove = volume->vol_no;
 276
 277         struct hammer_ioc_reblock reblock;
 278         bzero(&reblock, sizeof(reblock));
 279
 280         reblock.key_beg.localization = HAMMER_MIN_LOCALIZATION;
 281         reblock.key_beg.obj_id = HAMMER_MIN_OBJID;
 282         reblock.key_end.localization = HAMMER_MAX_LOCALIZATION;
 283         reblock.key_end.obj_id = HAMMER_MAX_OBJID;
 284         reblock.head.flags = HAMMER_IOC_DO_FLAGS;
 285         reblock.free_level = 0;
 286
 287         error = hammer_ioc_reblock(trans, ip, &reblock);
 288
 289         if (reblock.head.flags & HAMMER_IOC_HEAD_INTR) {
 290                 error = EINTR;
 291         }
 292
 293         if (error) {
 294                 if (error == EINTR) {
 295                         kprintf("reblock was interrupted\n");
 296                 } else {
 297                         kprintf("reblock failed: %d\n", error);
 298                 }
 299                 hmp->volume_to_remove = -1;
 300                 hammer_rel_volume(volume, 0);
 301                 goto end;
 302         }
 303
 304         /*
 305          * Sync filesystem
 306          */
 307         int count = 0;
 308         while (hammer_flusher_haswork(hmp)) {
 309                 hammer_flusher_sync(hmp);
 310                 ++count;
 311                 if (count >= 5) {
 312                         if (count == 5)
 313                                 kprintf("HAMMER: flushing.");
 314                         else
 315                                 kprintf(".");
 316                         tsleep(&count, 0, "hmrufl", hz);
 317                 }
 318                 if (count == 30) {
 319                         kprintf("giving up");
 320                         break;
 321                 }
 322         }
 323         kprintf("\n");
 324
 325         hammer_sync_lock_sh(trans);
 326         hammer_lock_ex(&hmp->blkmap_lock);
 327
 328         /*
 329          * We use stat later to update rootvol's bigblock stats
 330          */
 331         struct bigblock_stat stat;
 332         error = hammer_free_freemap(trans, volume, &stat);
 333         if (error) {
 334                 kprintf("Failed to free volume. Volume not empty!\n");
 335                 hmp->volume_to_remove = -1;
 336                 hammer_rel_volume(volume, 0);
 337                 hammer_unlock(&hmp->blkmap_lock);
 338                 hammer_sync_unlock(trans);
 339                 goto end;
 340         }
 341
 342         hmp->volume_to_remove = -1;
 343
 344         hammer_rel_volume(volume, 0);
 345
 346         /*
 347          * Unload buffers
 348          */
 349         RB_SCAN(hammer_buf_rb_tree, &hmp->rb_bufs_root, NULL,
 350                 hammer_unload_buffer, volume);
 351
 352         error = hammer_unload_volume(volume, NULL);
 353         if (error == -1) {
 354                 kprintf("Failed to unload volume\n");
 355                 hammer_unlock(&hmp->blkmap_lock);
 356                 hammer_sync_unlock(trans);
 357                 goto end;
 358         }
 359
 360         volume = NULL;
 361         --hmp->nvolumes;
 362
 363         /*
 364          * Set each volume's new value of the vol_count field.
 365          */
 366         for (int vol_no = 0; vol_no < HAMMER_MAX_VOLUMES; ++vol_no) {
 367                 volume = hammer_get_volume(hmp, vol_no, &error);
 368                 if (volume == NULL && error == ENOENT) {
 369                         /*
 370                          * Skip unused volume numbers
 371                          */
 372                         error = 0;
 373                         continue;
 374                 }
 375
 376                 KKASSERT(volume != NULL && error == 0);
 377                 hammer_modify_volume_field(trans, volume, vol_count);
 378                 volume->ondisk->vol_count = hmp->nvolumes;
 379                 hammer_modify_volume_done(volume);
 380
 381                 /*
 382                  * Only changes to the header of the root volume
 383                  * are automatically flushed to disk. For all
 384                  * other volumes that we modify we do it here.
 385                  *
 386                  * No interlock is needed, volume buffers are not
 387                  * messed with by bioops.
 388                  */
 389                 if (volume != trans->rootvol && volume->io.modified) {
 390                         hammer_crc_set_volume(volume->ondisk);
 391                         hammer_io_flush(&volume->io, 0);
 392                 }
 393
 394                 hammer_rel_volume(volume, 0);
 395         }
 396
 397         /*
 398          * Update the total number of bigblocks
 399          */
 400         hammer_modify_volume_field(trans, trans->rootvol,
 401                 vol0_stat_bigblocks);
 402         trans->rootvol->ondisk->vol0_stat_bigblocks -= stat.total_bigblocks;
 403         hammer_modify_volume_done(trans->rootvol);
 404
 405         /*
 406          * Update the number of free bigblocks
 407          * (including the copy in hmp)
 408          */
 409         hammer_modify_volume_field(trans, trans->rootvol,
 410                 vol0_stat_freebigblocks);
 411         trans->rootvol->ondisk->vol0_stat_freebigblocks -= stat.total_free_bigblocks;
 412         hmp->copy_stat_freebigblocks =
 413                 trans->rootvol->ondisk->vol0_stat_freebigblocks;
 414         hammer_modify_volume_done(trans->rootvol);
 415
 416
 417         hammer_unlock(&hmp->blkmap_lock);
 418         hammer_sync_unlock(trans);
 419
 420         /*
 421          * Erase the volume header of the removed device.
 422          *
 423          * This is to not accidentally mount the volume again.
 424          */
 425         struct vnode *devvp = NULL;
 426         error = hammer_setup_device(&devvp, ioc->device_name, 0);
 427         if (error) {
 428                 kprintf("Failed to open device: %s\n", ioc->device_name);
 429                 goto end;
 430         }
 431         KKASSERT(devvp);
 432         error = hammer_clear_volume_header(devvp);
 433         if (error) {
 434                 kprintf("Failed to clear volume header of device: %s\n",
 435                         ioc->device_name);
 436                 goto end;
 437         }
 438         hammer_close_device(&devvp, 0);
 439
 440         KKASSERT(error == 0);
 441 end:
 442         hammer_unlock(&hmp->volume_lock);
 443         return (error);
 444 }
 445
 446
 447 int
 448 hammer_ioc_volume_list(hammer_transaction_t trans, hammer_inode_t ip,
 449     struct hammer_ioc_volume_list *ioc)
 450 {
 451         struct hammer_mount *hmp = trans->hmp;
 452         hammer_volume_t volume;
 453         int error = 0;
 454         int i, cnt, len;
 455
 456         for (i = 0, cnt = 0; i < HAMMER_MAX_VOLUMES && cnt < ioc->nvols; i++) {
 457                 volume = hammer_get_volume(hmp, i, &error);
 458                 if (volume == NULL && error == ENOENT) {
 459                         error = 0;
 460                         continue;
 461                 }
 462                 KKASSERT(volume != NULL && error == 0);
 463
 464                 len = strlen(volume->vol_name) + 1;
 465                 KKASSERT(len <= MAXPATHLEN);
 466
 467                 error = copyout(volume->vol_name, ioc->vols[cnt].device_name,
 468                                 len);
 469                 if (error) {
 470                         hammer_rel_volume(volume, 0);
 471                         return (error);
 472                 }
 473                 cnt++;
 474                 hammer_rel_volume(volume, 0);
 475         }
 476         ioc->nvols = cnt;
 477
 478         return (error);
 479 }
 480
 481 /*
 482  * Iterate over all usable L1 entries of the volume and
 483  * the corresponding L2 entries.
 484  */
 485 static int
 486 hammer_iterate_l1l2_entries(hammer_transaction_t trans, hammer_volume_t volume,
 487         int (*callback)(hammer_transaction_t, hammer_volume_t, hammer_buffer_t*,
 488                 struct hammer_blockmap_layer1*, struct hammer_blockmap_layer2*,
 489                 hammer_off_t, hammer_off_t, void*),
 490         void *data)
 491 {
 492         struct hammer_mount *hmp = trans->hmp;
 493         hammer_blockmap_t freemap = &hmp->blockmap[HAMMER_ZONE_FREEMAP_INDEX];
 494         hammer_buffer_t buffer = NULL;
 495         int error = 0;
 496
 497         hammer_off_t phys_off;
 498         hammer_off_t block_off;
 499         hammer_off_t layer1_off;
 500         hammer_off_t layer2_off;
 501         hammer_off_t aligned_buf_end_off;
 502         struct hammer_blockmap_layer1 *layer1;
 503         struct hammer_blockmap_layer2 *layer2;
 504
 505         /*
 506          * Calculate the usable size of the volume, which
 507          * must be aligned at a bigblock (8 MB) boundary.
 508          */
 509         aligned_buf_end_off = (HAMMER_ENCODE_RAW_BUFFER(volume->ondisk->vol_no,
 510                 (volume->ondisk->vol_buf_end - volume->ondisk->vol_buf_beg)
 511                 & ~HAMMER_LARGEBLOCK_MASK64));
 512
 513         /*
 514          * Iterate the volume's address space in chunks of 4 TB, where each
 515          * chunk consists of at least one physically available 8 MB bigblock.
 516          *
 517          * For each chunk we need one L1 entry and one L2 bigblock.
 518          * We use the first bigblock of each chunk as L2 block.
 519          */
 520         for (phys_off = HAMMER_ENCODE_RAW_BUFFER(volume->ondisk->vol_no, 0);
 521              phys_off < aligned_buf_end_off;
 522              phys_off += HAMMER_BLOCKMAP_LAYER2) {
 523                 for (block_off = 0;
 524                      block_off < HAMMER_BLOCKMAP_LAYER2;
 525                      block_off += HAMMER_LARGEBLOCK_SIZE) {
 526                         layer2_off = phys_off +
 527                                 HAMMER_BLOCKMAP_LAYER2_OFFSET(block_off);
 528                         layer2 = hammer_bread(hmp, layer2_off, &error, &buffer);
 529                         if (error)
 530                                 goto end;
 531
 532                         error = callback(trans, volume, &buffer, NULL,
 533                                          layer2, phys_off, block_off, data);
 534                         if (error)
 535                                 goto end;
 536                 }
 537
 538                 layer1_off = freemap->phys_offset +
 539                                 HAMMER_BLOCKMAP_LAYER1_OFFSET(phys_off);
 540                 layer1 = hammer_bread(hmp, layer1_off, &error, &buffer);
 541                 if (error)
 542                         goto end;
 543
 544                 error = callback(trans, volume, &buffer, layer1, NULL,
 545                                  phys_off, 0, data);
 546                 if (error)
 547                         goto end;
 548         }
 549
 550 end:
 551         if (buffer) {
 552                 hammer_rel_buffer(buffer, 0);
 553                 buffer = NULL;
 554         }
 555
 556         return error;
 557 }
 558
 559
 560 static int
 561 format_callback(hammer_transaction_t trans, hammer_volume_t volume,
 562         hammer_buffer_t *bufferp,
 563         struct hammer_blockmap_layer1 *layer1,
 564         struct hammer_blockmap_layer2 *layer2,
 565         hammer_off_t phys_off,
 566         hammer_off_t block_off,
 567         void *data)
 568 {
 569         struct bigblock_stat *stat = (struct bigblock_stat*)data;
 570
 571         /*
 572          * Calculate the usable size of the volume, which must be aligned
 573          * at a bigblock (8 MB) boundary.
 574          */
 575         hammer_off_t aligned_buf_end_off;
 576         aligned_buf_end_off = (HAMMER_ENCODE_RAW_BUFFER(volume->ondisk->vol_no,
 577                 (volume->ondisk->vol_buf_end - volume->ondisk->vol_buf_beg)
 578                 & ~HAMMER_LARGEBLOCK_MASK64));
 579
 580         if (layer1) {
 581                 KKASSERT(layer1->phys_offset == HAMMER_BLOCKMAP_UNAVAIL);
 582
 583                 hammer_modify_buffer(trans, *bufferp, layer1, sizeof(*layer1));
 584                 bzero(layer1, sizeof(*layer1));
 585                 layer1->phys_offset = phys_off;
 586                 layer1->blocks_free = stat->counter;
 587                 layer1->layer1_crc = crc32(layer1, HAMMER_LAYER1_CRCSIZE);
 588                 hammer_modify_buffer_done(*bufferp);
 589
 590                 stat->total_free_bigblocks += stat->counter;
 591                 stat->counter = 0; /* reset */
 592         } else if (layer2) {
 593                 hammer_modify_buffer(trans, *bufferp, layer2, sizeof(*layer2));
 594                 bzero(layer2, sizeof(*layer2));
 595
 596                 if (block_off == 0) {
 597                         /*
 598                          * The first entry represents the L2 bigblock itself.
 599                          */
 600                         layer2->zone = HAMMER_ZONE_FREEMAP_INDEX;
 601                         layer2->append_off = HAMMER_LARGEBLOCK_SIZE;
 602                         layer2->bytes_free = 0;
 603                         ++stat->total_bigblocks;
 604                 } else if (phys_off + block_off < aligned_buf_end_off) {
 605                         /*
 606                          * Available bigblock
 607                          */
 608                         layer2->zone = 0;
 609                         layer2->append_off = 0;
 610                         layer2->bytes_free = HAMMER_LARGEBLOCK_SIZE;
 611                         ++stat->total_bigblocks;
 612                         ++stat->counter;
 613                 } else {
 614                         /*
 615                          * Bigblock outside of physically available
 616                          * space
 617                          */
 618                         layer2->zone = HAMMER_ZONE_UNAVAIL_INDEX;
 619                         layer2->append_off = HAMMER_LARGEBLOCK_SIZE;
 620                         layer2->bytes_free = 0;
 621                 }
 622
 623                 layer2->entry_crc = crc32(layer2, HAMMER_LAYER2_CRCSIZE);
 624                 hammer_modify_buffer_done(*bufferp);
 625         } else {
 626                 KKASSERT(0);
 627         }
 628
 629         return 0;
 630 }
 631
 632 static int
 633 hammer_format_freemap(hammer_transaction_t trans, hammer_volume_t volume,
 634         struct bigblock_stat *stat)
 635 {
 636         stat->total_bigblocks = 0;
 637         stat->total_free_bigblocks = 0;
 638         stat->counter = 0;
 639         return hammer_iterate_l1l2_entries(trans, volume, format_callback, stat);
 640 }
 641
 642 static int
 643 free_callback(hammer_transaction_t trans, hammer_volume_t volume __unused,
 644         hammer_buffer_t *bufferp,
 645         struct hammer_blockmap_layer1 *layer1,
 646         struct hammer_blockmap_layer2 *layer2,
 647         hammer_off_t phys_off,
 648         hammer_off_t block_off __unused,
 649         void *data)
 650 {
 651         struct bigblock_stat *stat = (struct bigblock_stat*)data;
 652
 653         /*
 654          * No modifications to ondisk structures
 655          */
 656         int testonly = (stat == NULL);
 657
 658         if (layer1) {
 659                 if (layer1->phys_offset == HAMMER_BLOCKMAP_UNAVAIL) {
 660                         /*
 661                          * This layer1 entry is already free.
 662                          */
 663                         return 0;
 664                 }
 665
 666                 KKASSERT((int)HAMMER_VOL_DECODE(layer1->phys_offset) ==
 667                         trans->hmp->volume_to_remove);
 668
 669                 if (testonly)
 670                         return 0;
 671
 672                 /*
 673                  * Free the L1 entry
 674                  */
 675                 hammer_modify_buffer(trans, *bufferp, layer1, sizeof(*layer1));
 676                 bzero(layer1, sizeof(*layer1));
 677                 layer1->phys_offset = HAMMER_BLOCKMAP_UNAVAIL;
 678                 layer1->layer1_crc = crc32(layer1, HAMMER_LAYER1_CRCSIZE);
 679                 hammer_modify_buffer_done(*bufferp);
 680
 681                 return 0;
 682         } else if (layer2) {
 683                 if (layer2->zone == HAMMER_ZONE_UNAVAIL_INDEX) {
 684                         return 0;
 685                 }
 686
 687                 if (layer2->zone == HAMMER_ZONE_FREEMAP_INDEX) {
 688                         if (stat) {
 689                                 ++stat->total_bigblocks;
 690                         }
 691                         return 0;
 692                 }
 693
 694                 if (layer2->append_off == 0 &&
 695                     layer2->bytes_free == HAMMER_LARGEBLOCK_SIZE) {
 696                         if (stat) {
 697                                 ++stat->total_bigblocks;
 698                                 ++stat->total_free_bigblocks;
 699                         }
 700                         return 0;
 701                 }
 702
 703                 /*
 704                  * We found a layer2 entry that is not empty!
 705                  */
 706                 return EBUSY;
 707         } else {
 708                 KKASSERT(0);
 709         }
 710
 711         return EINVAL;
 712 }
 713
 714 static int
 715 hammer_free_freemap(hammer_transaction_t trans, hammer_volume_t volume,
 716         struct bigblock_stat *stat)
 717 {
 718         int error;
 719
 720         stat->total_bigblocks = 0;
 721         stat->total_free_bigblocks = 0;
 722         stat->counter = 0;
 723
 724         error = hammer_iterate_l1l2_entries(trans, volume, free_callback, NULL);
 725         if (error)
 726                 return error;
 727
 728         error = hammer_iterate_l1l2_entries(trans, volume, free_callback, stat);
 729         return error;
 730 }
 731
 732 /************************************************************************
 733  *                              MISC                                    *
 734  ************************************************************************
 735  */
 736
 737 static int
 738 hammer_setup_device(struct vnode **devvpp, const char *dev_path, int ronly)
 739 {
 740         int error;
 741         struct nlookupdata nd;
 742
 743         /*
 744          * Get the device vnode
 745          */
 746         if (*devvpp == NULL) {
 747                 error = nlookup_init(&nd, dev_path, UIO_SYSSPACE, NLC_FOLLOW);
 748                 if (error == 0)
 749                         error = nlookup(&nd);
 750                 if (error == 0)
 751                         error = cache_vref(&nd.nl_nch, nd.nl_cred, devvpp);
 752                 nlookup_done(&nd);
 753         } else {
 754                 error = 0;
 755         }
 756
 757         if (error == 0) {
 758                 if (vn_isdisk(*devvpp, &error)) {
 759                         error = vfs_mountedon(*devvpp);
 760                 }
 761         }
 762         if (error == 0 && vcount(*devvpp) > 0)
 763                 error = EBUSY;
 764         if (error == 0) {
 765                 vn_lock(*devvpp, LK_EXCLUSIVE | LK_RETRY);
 766                 error = vinvalbuf(*devvpp, V_SAVE, 0, 0);
 767                 if (error == 0) {
 768                         error = VOP_OPEN(*devvpp,
 769                                          (ronly ? FREAD : FREAD|FWRITE),
 770                                          FSCRED, NULL);
 771                 }
 772                 vn_unlock(*devvpp);
 773         }
 774         if (error && *devvpp) {
 775                 vrele(*devvpp);
 776                 *devvpp = NULL;
 777         }
 778         return (error);
 779 }
 780
 781 static void
 782 hammer_close_device(struct vnode **devvpp, int ronly)
 783 {
 784         if (*devvpp) {
 785                 vn_lock(*devvpp, LK_EXCLUSIVE | LK_RETRY);
 786                 vinvalbuf(*devvpp, ronly ? 0 : V_SAVE, 0, 0);
 787                 VOP_CLOSE(*devvpp, (ronly ? FREAD : FREAD|FWRITE), NULL);
 788                 vn_unlock(*devvpp);
 789                 vrele(*devvpp);
 790                 *devvpp = NULL;
 791         }
 792 }
 793
 794 static int
 795 hammer_format_volume_header(struct hammer_mount *hmp, struct vnode *devvp,
 796         const char *vol_name, int vol_no, int vol_count,
 797         int64_t vol_size, int64_t boot_area_size, int64_t mem_area_size)
 798 {
 799         struct buf *bp = NULL;
 800         struct hammer_volume_ondisk *ondisk;
 801         int error;
 802
 803         /*
 804          * Extract the volume number from the volume header and do various
 805          * sanity checks.
 806          */
 807         KKASSERT(HAMMER_BUFSIZE >= sizeof(struct hammer_volume_ondisk));
 808         error = bread(devvp, 0LL, HAMMER_BUFSIZE, &bp);
 809         if (error || bp->b_bcount < sizeof(struct hammer_volume_ondisk))
 810                 goto late_failure;
 811
 812         ondisk = (struct hammer_volume_ondisk*) bp->b_data;
 813
 814         /*
 815          * Note that we do NOT allow to use a device that contains
 816          * a valid HAMMER signature. It has to be cleaned up with dd
 817          * before.
 818          */
 819         if (ondisk->vol_signature == HAMMER_FSBUF_VOLUME) {
 820                 kprintf("hammer_volume_add: Formatting of valid HAMMER volume "
 821                         "%s denied. Erase with dd!\n", vol_name);
 822                 error = EFTYPE;
 823                 goto late_failure;
 824         }
 825
 826         bzero(ondisk, sizeof(struct hammer_volume_ondisk));
 827         ksnprintf(ondisk->vol_name, sizeof(ondisk->vol_name), "%s", vol_name);
 828         ondisk->vol_fstype = hmp->rootvol->ondisk->vol_fstype;
 829         ondisk->vol_signature = HAMMER_FSBUF_VOLUME;
 830         ondisk->vol_fsid = hmp->fsid;
 831         ondisk->vol_rootvol = hmp->rootvol->vol_no;
 832         ondisk->vol_no = vol_no;
 833         ondisk->vol_count = vol_count;
 834         ondisk->vol_version = hmp->version;
 835
 836         /*
 837          * Reserve space for (future) header junk, setup our poor-man's
 838          * bigblock allocator.
 839          */
 840         int64_t vol_alloc = HAMMER_BUFSIZE * 16;
 841
 842         ondisk->vol_bot_beg = vol_alloc;
 843         vol_alloc += boot_area_size;
 844         ondisk->vol_mem_beg = vol_alloc;
 845         vol_alloc += mem_area_size;
 846
 847         /*
 848          * The remaining area is the zone 2 buffer allocation area.  These
 849          * buffers
 850          */
 851         ondisk->vol_buf_beg = vol_alloc;
 852         ondisk->vol_buf_end = vol_size & ~(int64_t)HAMMER_BUFMASK;
 853
 854         if (ondisk->vol_buf_end < ondisk->vol_buf_beg) {
 855                 kprintf("volume %d %s is too small to hold the volume header",
 856                      ondisk->vol_no, ondisk->vol_name);
 857                 error = EFTYPE;
 858                 goto late_failure;
 859         }
 860
 861         ondisk->vol_nblocks = (ondisk->vol_buf_end - ondisk->vol_buf_beg) /
 862                               HAMMER_BUFSIZE;
 863         ondisk->vol_blocksize = HAMMER_BUFSIZE;
 864
 865         /*
 866          * Write volume header to disk
 867          */
 868         error = bwrite(bp);
 869         bp = NULL;
 870
 871 late_failure:
 872         if (bp)
 873                 brelse(bp);
 874         return (error);
 875 }
 876
 877 /*
 878  * Invalidates the volume header. Used by volume-del.
 879  */
 880 static int
 881 hammer_clear_volume_header(struct vnode *devvp)
 882 {
 883         struct buf *bp = NULL;
 884         struct hammer_volume_ondisk *ondisk;
 885         int error;
 886
 887         KKASSERT(HAMMER_BUFSIZE >= sizeof(struct hammer_volume_ondisk));
 888         error = bread(devvp, 0LL, HAMMER_BUFSIZE, &bp);
 889         if (error || bp->b_bcount < sizeof(struct hammer_volume_ondisk))
 890                 goto late_failure;
 891
 892         ondisk = (struct hammer_volume_ondisk*) bp->b_data;
 893         bzero(ondisk, sizeof(struct hammer_volume_ondisk));
 894
 895         error = bwrite(bp);
 896         bp = NULL;
 897
 898 late_failure:
 899         if (bp)
 900                 brelse(bp);
 901         return (error);
 902 }