sys/vfs/hammer/hammer_vfsops.c

   1 /*
   2  * Copyright (c) 2007-2008 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>
   6  *
   7  * Redistribution and use in source and binary forms, with or without
   8  * modification, are permitted provided that the following conditions
   9  * are met:
  10  *
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in
  15  *    the documentation and/or other materials provided with the
  16  *    distribution.
  17  * 3. Neither the name of The DragonFly Project nor the names of its
  18  *    contributors may be used to endorse or promote products derived
  19  *    from this software without specific, prior written permission.
  20  *
  21  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  22  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  23  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
  24  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
  25  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
  26  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
  27  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
  28  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
  29  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
  30  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
  31  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  32  * SUCH DAMAGE.
  33  *
  34  * $DragonFly: src/sys/vfs/hammer/hammer_vfsops.c,v 1.41 2008/06/09 04:19:10 dillon Exp $
  35  */
  36
  37 #include <sys/param.h>
  38 #include <sys/systm.h>
  39 #include <sys/kernel.h>
  40 #include <sys/vnode.h>
  41 #include <sys/mount.h>
  42 #include <sys/malloc.h>
  43 #include <sys/nlookup.h>
  44 #include <sys/fcntl.h>
  45 #include <sys/sysctl.h>
  46 #include <sys/buf.h>
  47 #include <sys/buf2.h>
  48 #include "hammer.h"
  49
  50 int hammer_debug_io;
  51 int hammer_debug_general;
  52 int hammer_debug_debug;
  53 int hammer_debug_inode;
  54 int hammer_debug_locks;
  55 int hammer_debug_btree;
  56 int hammer_debug_tid;
  57 int hammer_debug_recover;       /* -1 will disable, +1 will force */
  58 int hammer_debug_recover_faults;
  59 int hammer_count_inodes;
  60 int hammer_count_reclaiming;
  61 int hammer_count_records;
  62 int hammer_count_record_datas;
  63 int hammer_count_volumes;
  64 int hammer_count_buffers;
  65 int hammer_count_nodes;
  66 int hammer_count_dirtybufs;             /* global */
  67 int hammer_count_reservations;
  68 int hammer_stats_btree_iterations;
  69 int hammer_stats_record_iterations;
  70 int hammer_limit_dirtybufs;             /* per-mount */
  71 int hammer_limit_irecs;                 /* per-inode */
  72 int hammer_limit_recs;                  /* as a whole XXX */
  73 int hammer_bio_count;
  74 int64_t hammer_contention_count;
  75 int64_t hammer_zone_limit;
  76
  77 SYSCTL_NODE(_vfs, OID_AUTO, hammer, CTLFLAG_RW, 0, "HAMMER filesystem");
  78 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_general, CTLFLAG_RW,
  79            &hammer_debug_general, 0, "");
  80 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_io, CTLFLAG_RW,
  81            &hammer_debug_io, 0, "");
  82 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_debug, CTLFLAG_RW,
  83            &hammer_debug_debug, 0, "");
  84 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_inode, CTLFLAG_RW,
  85            &hammer_debug_inode, 0, "");
  86 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_locks, CTLFLAG_RW,
  87            &hammer_debug_locks, 0, "");
  88 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_btree, CTLFLAG_RW,
  89            &hammer_debug_btree, 0, "");
  90 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_tid, CTLFLAG_RW,
  91            &hammer_debug_tid, 0, "");
  92 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover, CTLFLAG_RW,
  93            &hammer_debug_recover, 0, "");
  94 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover_faults, CTLFLAG_RW,
  95            &hammer_debug_recover_faults, 0, "");
  96
  97 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_dirtybufs, CTLFLAG_RW,
  98            &hammer_limit_dirtybufs, 0, "");
  99 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_irecs, CTLFLAG_RW,
 100            &hammer_limit_irecs, 0, "");
 101 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_recs, CTLFLAG_RW,
 102            &hammer_limit_recs, 0, "");
 103
 104 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_inodes, CTLFLAG_RD,
 105            &hammer_count_inodes, 0, "");
 106 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reclaiming, CTLFLAG_RD,
 107            &hammer_count_reclaiming, 0, "");
 108 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_records, CTLFLAG_RD,
 109            &hammer_count_records, 0, "");
 110 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_record_datas, CTLFLAG_RD,
 111            &hammer_count_record_datas, 0, "");
 112 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_volumes, CTLFLAG_RD,
 113            &hammer_count_volumes, 0, "");
 114 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_buffers, CTLFLAG_RD,
 115            &hammer_count_buffers, 0, "");
 116 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_nodes, CTLFLAG_RD,
 117            &hammer_count_nodes, 0, "");
 118 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_dirtybufs, CTLFLAG_RD,
 119            &hammer_count_dirtybufs, 0, "");
 120 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reservations, CTLFLAG_RD,
 121            &hammer_count_reservations, 0, "");
 122 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, zone_limit, CTLFLAG_RW,
 123            &hammer_zone_limit, 0, "");
 124 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, contention_count, CTLFLAG_RW,
 125            &hammer_contention_count, 0, "");
 126
 127 /*
 128  * VFS ABI
 129  */
 130 static void     hammer_free_hmp(struct mount *mp);
 131
 132 static int      hammer_vfs_mount(struct mount *mp, char *path, caddr_t data,
 133                                 struct ucred *cred);
 134 static int      hammer_vfs_unmount(struct mount *mp, int mntflags);
 135 static int      hammer_vfs_root(struct mount *mp, struct vnode **vpp);
 136 static int      hammer_vfs_statfs(struct mount *mp, struct statfs *sbp,
 137                                 struct ucred *cred);
 138 static int      hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp,
 139                                 struct ucred *cred);
 140 static int      hammer_vfs_sync(struct mount *mp, int waitfor);
 141 static int      hammer_vfs_vget(struct mount *mp, ino_t ino,
 142                                 struct vnode **vpp);
 143 static int      hammer_vfs_init(struct vfsconf *conf);
 144 static int      hammer_vfs_fhtovp(struct mount *mp, struct fid *fhp,
 145                                 struct vnode **vpp);
 146 static int      hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp);
 147 static int      hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
 148                                 int *exflagsp, struct ucred **credanonp);
 149
 150
 151 static struct vfsops hammer_vfsops = {
 152         .vfs_mount      = hammer_vfs_mount,
 153         .vfs_unmount    = hammer_vfs_unmount,
 154         .vfs_root       = hammer_vfs_root,
 155         .vfs_statfs     = hammer_vfs_statfs,
 156         .vfs_statvfs    = hammer_vfs_statvfs,
 157         .vfs_sync       = hammer_vfs_sync,
 158         .vfs_vget       = hammer_vfs_vget,
 159         .vfs_init       = hammer_vfs_init,
 160         .vfs_vptofh     = hammer_vfs_vptofh,
 161         .vfs_fhtovp     = hammer_vfs_fhtovp,
 162         .vfs_checkexp   = hammer_vfs_checkexp
 163 };
 164
 165 MALLOC_DEFINE(M_HAMMER, "hammer-mount", "hammer mount");
 166
 167 VFS_SET(hammer_vfsops, hammer, 0);
 168 MODULE_VERSION(hammer, 1);
 169
 170 static int
 171 hammer_vfs_init(struct vfsconf *conf)
 172 {
 173         if (hammer_limit_irecs == 0)
 174                 hammer_limit_irecs = nbuf;
 175         if (hammer_limit_recs == 0)             /* XXX TODO */
 176                 hammer_limit_recs = hammer_limit_irecs * 4;
 177         if (hammer_limit_dirtybufs == 0) {
 178                 hammer_limit_dirtybufs = hidirtybuffers / 2;
 179                 if (hammer_limit_dirtybufs < 100)
 180                         hammer_limit_dirtybufs = 100;
 181         }
 182         return(0);
 183 }
 184
 185 static int
 186 hammer_vfs_mount(struct mount *mp, char *mntpt, caddr_t data,
 187                  struct ucred *cred)
 188 {
 189         struct hammer_mount_info info;
 190         hammer_mount_t hmp;
 191         hammer_volume_t rootvol;
 192         struct vnode *rootvp;
 193         const char *upath;      /* volume name in userspace */
 194         char *path;             /* volume name in system space */
 195         int error;
 196         int i;
 197
 198         if ((error = copyin(data, &info, sizeof(info))) != 0)
 199                 return (error);
 200         if ((mp->mnt_flag & MNT_UPDATE) == 0) {
 201                 if (info.nvolumes <= 0 || info.nvolumes >= 32768)
 202                         return (EINVAL);
 203         }
 204
 205         /*
 206          * Interal mount data structure
 207          */
 208         if (mp->mnt_flag & MNT_UPDATE) {
 209                 hmp = (void *)mp->mnt_data;
 210                 KKASSERT(hmp != NULL);
 211         } else {
 212                 hmp = kmalloc(sizeof(*hmp), M_HAMMER, M_WAITOK | M_ZERO);
 213                 mp->mnt_data = (qaddr_t)hmp;
 214                 hmp->mp = mp;
 215                 hmp->zbuf = kmalloc(HAMMER_BUFSIZE, M_HAMMER, M_WAITOK|M_ZERO);
 216                 hmp->namekey_iterator = mycpu->gd_time_seconds;
 217                 /*TAILQ_INIT(&hmp->recycle_list);*/
 218
 219                 hmp->root_btree_beg.localization = HAMMER_MIN_LOCALIZATION;
 220                 hmp->root_btree_beg.obj_id = -0x8000000000000000LL;
 221                 hmp->root_btree_beg.key = -0x8000000000000000LL;
 222                 hmp->root_btree_beg.create_tid = 1;
 223                 hmp->root_btree_beg.delete_tid = 1;
 224                 hmp->root_btree_beg.rec_type = 0;
 225                 hmp->root_btree_beg.obj_type = 0;
 226
 227                 hmp->root_btree_end.localization = HAMMER_MAX_LOCALIZATION;
 228                 hmp->root_btree_end.obj_id = 0x7FFFFFFFFFFFFFFFLL;
 229                 hmp->root_btree_end.key = 0x7FFFFFFFFFFFFFFFLL;
 230                 hmp->root_btree_end.create_tid = 0xFFFFFFFFFFFFFFFFULL;
 231                 hmp->root_btree_end.delete_tid = 0;   /* special case */
 232                 hmp->root_btree_end.rec_type = 0xFFFFU;
 233                 hmp->root_btree_end.obj_type = 0;
 234                 lockinit(&hmp->blockmap_lock, "blkmap", 0, 0);
 235
 236                 hmp->sync_lock.refs = 1;
 237                 hmp->free_lock.refs = 1;
 238
 239                 TAILQ_INIT(&hmp->flush_list);
 240                 TAILQ_INIT(&hmp->objid_cache_list);
 241                 TAILQ_INIT(&hmp->undo_lru_list);
 242
 243                 /*
 244                  * Set default zone limits.  This value can be reduced
 245                  * further by the zone limit specified in the root volume.
 246                  *
 247                  * The sysctl can force a small zone limit for debugging
 248                  * purposes.
 249                  */
 250                 for (i = 0; i < HAMMER_MAX_ZONES; ++i) {
 251                         hmp->zone_limits[i] =
 252                                 HAMMER_ZONE_ENCODE(i, HAMMER_ZONE_LIMIT);
 253
 254                         if (hammer_zone_limit) {
 255                                 hmp->zone_limits[i] =
 256                                     HAMMER_ZONE_ENCODE(i, hammer_zone_limit);
 257                         }
 258                         hammer_init_holes(hmp, &hmp->holes[i]);
 259                 }
 260         }
 261         hmp->hflags &= ~HMNT_USERFLAGS;
 262         hmp->hflags |= info.hflags & HMNT_USERFLAGS;
 263         if (info.asof) {
 264                 kprintf("ASOF\n");
 265                 mp->mnt_flag |= MNT_RDONLY;
 266                 hmp->asof = info.asof;
 267         } else {
 268                 hmp->asof = HAMMER_MAX_TID;
 269         }
 270
 271         /*
 272          * Re-open read-write if originally read-only, or vise-versa.
 273          */
 274         if (mp->mnt_flag & MNT_UPDATE) {
 275                 error = 0;
 276                 if (hmp->ronly && (mp->mnt_kern_flag & MNTK_WANTRDWR)) {
 277                         kprintf("HAMMER read-only -> read-write\n");
 278                         hmp->ronly = 0;
 279                         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
 280                                 hammer_adjust_volume_mode, NULL);
 281                         rootvol = hammer_get_root_volume(hmp, &error);
 282                         if (rootvol) {
 283                                 hammer_recover_flush_buffers(hmp, rootvol);
 284                                 bcopy(rootvol->ondisk->vol0_blockmap,
 285                                       hmp->blockmap,
 286                                       sizeof(hmp->blockmap));
 287                                 hammer_rel_volume(rootvol, 0);
 288                         }
 289                         RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
 290                                 hammer_reload_inode, NULL);
 291                         /* kernel clears MNT_RDONLY */
 292                 } else if (hmp->ronly == 0 && (mp->mnt_flag & MNT_RDONLY)) {
 293                         kprintf("HAMMER read-write -> read-only\n");
 294                         hmp->ronly = 1; /* messy */
 295                         RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
 296                                 hammer_reload_inode, NULL);
 297                         hmp->ronly = 0;
 298                         hammer_flusher_sync(hmp);
 299                         hammer_flusher_sync(hmp);
 300                         hammer_flusher_sync(hmp);
 301                         hmp->ronly = 1;
 302                         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
 303                                 hammer_adjust_volume_mode, NULL);
 304                 }
 305                 return(error);
 306         }
 307
 308         RB_INIT(&hmp->rb_vols_root);
 309         RB_INIT(&hmp->rb_inos_root);
 310         RB_INIT(&hmp->rb_nods_root);
 311         RB_INIT(&hmp->rb_undo_root);
 312         RB_INIT(&hmp->rb_resv_root);
 313         RB_INIT(&hmp->rb_bufs_root);
 314
 315         hmp->ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
 316
 317         TAILQ_INIT(&hmp->volu_list);
 318         TAILQ_INIT(&hmp->undo_list);
 319         TAILQ_INIT(&hmp->data_list);
 320         TAILQ_INIT(&hmp->meta_list);
 321         TAILQ_INIT(&hmp->lose_list);
 322
 323         /*
 324          * Load volumes
 325          */
 326         path = objcache_get(namei_oc, M_WAITOK);
 327         hmp->nvolumes = info.nvolumes;
 328         for (i = 0; i < info.nvolumes; ++i) {
 329                 error = copyin(&info.volumes[i], &upath, sizeof(char *));
 330                 if (error == 0)
 331                         error = copyinstr(upath, path, MAXPATHLEN, NULL);
 332                 if (error == 0)
 333                         error = hammer_install_volume(hmp, path);
 334                 if (error)
 335                         break;
 336         }
 337         objcache_put(namei_oc, path);
 338
 339         /*
 340          * Make sure we found a root volume
 341          */
 342         if (error == 0 && hmp->rootvol == NULL) {
 343                 kprintf("hammer_mount: No root volume found!\n");
 344                 error = EINVAL;
 345         }
 346         if (error) {
 347                 hammer_free_hmp(mp);
 348                 return (error);
 349         }
 350
 351         /*
 352          * No errors, setup enough of the mount point so we can lookup the
 353          * root vnode.
 354          */
 355         mp->mnt_iosize_max = MAXPHYS;
 356         mp->mnt_kern_flag |= MNTK_FSMID;
 357
 358         /*
 359          * note: f_iosize is used by vnode_pager_haspage() when constructing
 360          * its VOP_BMAP call.
 361          */
 362         mp->mnt_stat.f_iosize = HAMMER_BUFSIZE;
 363         mp->mnt_stat.f_bsize = HAMMER_BUFSIZE;
 364
 365         mp->mnt_vstat.f_frsize = HAMMER_BUFSIZE;
 366         mp->mnt_vstat.f_bsize = HAMMER_BUFSIZE;
 367
 368         mp->mnt_maxsymlinklen = 255;
 369         mp->mnt_flag |= MNT_LOCAL;
 370
 371         vfs_add_vnodeops(mp, &hammer_vnode_vops, &mp->mnt_vn_norm_ops);
 372         vfs_add_vnodeops(mp, &hammer_spec_vops, &mp->mnt_vn_spec_ops);
 373         vfs_add_vnodeops(mp, &hammer_fifo_vops, &mp->mnt_vn_fifo_ops);
 374
 375         /*
 376          * The root volume's ondisk pointer is only valid if we hold a
 377          * reference to it.
 378          */
 379         rootvol = hammer_get_root_volume(hmp, &error);
 380         if (error)
 381                 goto failed;
 382
 383         /*
 384          * Perform any necessary UNDO operations.  The recovery code does
 385          * call hammer_undo_lookup() so we have to pre-cache the blockmap,
 386          * and then re-copy it again after recovery is complete.
 387          *
 388          * If this is a read-only mount the UNDO information is retained
 389          * in memory in the form of dirty buffer cache buffers, and not
 390          * written back to the media.
 391          */
 392         bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
 393               sizeof(hmp->blockmap));
 394
 395         error = hammer_recover(hmp, rootvol);
 396         if (error) {
 397                 kprintf("Failed to recover HAMMER filesystem on mount\n");
 398                 goto done;
 399         }
 400
 401         /*
 402          * Finish setup now that we have a good root volume
 403          */
 404         ksnprintf(mp->mnt_stat.f_mntfromname,
 405                   sizeof(mp->mnt_stat.f_mntfromname), "%s",
 406                   rootvol->ondisk->vol_name);
 407         mp->mnt_stat.f_fsid.val[0] =
 408                 crc32((char *)&rootvol->ondisk->vol_fsid + 0, 8);
 409         mp->mnt_stat.f_fsid.val[1] =
 410                 crc32((char *)&rootvol->ondisk->vol_fsid + 8, 8);
 411
 412         mp->mnt_vstat.f_fsid_uuid = rootvol->ondisk->vol_fsid;
 413         mp->mnt_vstat.f_fsid = crc32(&mp->mnt_vstat.f_fsid_uuid,
 414                                      sizeof(mp->mnt_vstat.f_fsid_uuid));
 415
 416         /*
 417          * Certain often-modified fields in the root volume are cached in
 418          * the hammer_mount structure so we do not have to generate lots
 419          * of little UNDO structures for them.
 420          *
 421          * Recopy after recovery.  This also has the side effect of
 422          * setting our cached undo FIFO's first_offset, which serves to
 423          * placemark the FIFO start for the NEXT flush cycle while the
 424          * on-disk first_offset represents the LAST flush cycle.
 425          */
 426         hmp->next_tid = rootvol->ondisk->vol0_next_tid;
 427         bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
 428               sizeof(hmp->blockmap));
 429         hmp->copy_stat_freebigblocks = rootvol->ondisk->vol0_stat_freebigblocks;
 430
 431         /*
 432          * Use the zone limit set by newfs_hammer, or the zone limit set by
 433          * sysctl (for debugging), whichever is smaller.
 434          */
 435         if (rootvol->ondisk->vol0_zone_limit) {
 436                 hammer_off_t vol0_zone_limit;
 437
 438                 vol0_zone_limit = rootvol->ondisk->vol0_zone_limit;
 439                 for (i = 0; i < HAMMER_MAX_ZONES; ++i) {
 440                         if (hmp->zone_limits[i] > vol0_zone_limit)
 441                                 hmp->zone_limits[i] = vol0_zone_limit;
 442                 }
 443         }
 444
 445         hammer_flusher_create(hmp);
 446
 447         /*
 448          * Locate the root directory using the root cluster's B-Tree as a
 449          * starting point.  The root directory uses an obj_id of 1.
 450          *
 451          * FUTURE: Leave the root directory cached referenced but unlocked
 452          * in hmp->rootvp (need to flush it on unmount).
 453          */
 454         error = hammer_vfs_vget(mp, 1, &rootvp);
 455         if (error)
 456                 goto done;
 457         vput(rootvp);
 458         /*vn_unlock(hmp->rootvp);*/
 459
 460 done:
 461         hammer_rel_volume(rootvol, 0);
 462 failed:
 463         /*
 464          * Cleanup and return.
 465          */
 466         if (error)
 467                 hammer_free_hmp(mp);
 468         return (error);
 469 }
 470
 471 static int
 472 hammer_vfs_unmount(struct mount *mp, int mntflags)
 473 {
 474 #if 0
 475         struct hammer_mount *hmp = (void *)mp->mnt_data;
 476 #endif
 477         int flags;
 478         int error;
 479
 480         /*
 481          * Clean out the vnodes
 482          */
 483         flags = 0;
 484         if (mntflags & MNT_FORCE)
 485                 flags |= FORCECLOSE;
 486         if ((error = vflush(mp, 0, flags)) != 0)
 487                 return (error);
 488
 489         /*
 490          * Clean up the internal mount structure and related entities.  This
 491          * may issue I/O.
 492          */
 493         hammer_free_hmp(mp);
 494         return(0);
 495 }
 496
 497 /*
 498  * Clean up the internal mount structure and disassociate it from the mount.
 499  * This may issue I/O.
 500  */
 501 static void
 502 hammer_free_hmp(struct mount *mp)
 503 {
 504         struct hammer_mount *hmp = (void *)mp->mnt_data;
 505         int i;
 506
 507 #if 0
 508         /*
 509          * Clean up the root vnode
 510          */
 511         if (hmp->rootvp) {
 512                 vrele(hmp->rootvp);
 513                 hmp->rootvp = NULL;
 514         }
 515 #endif
 516         hammer_flusher_sync(hmp);
 517         hammer_flusher_sync(hmp);
 518         hammer_flusher_destroy(hmp);
 519
 520         KKASSERT(RB_EMPTY(&hmp->rb_inos_root));
 521
 522 #if 0
 523         /*
 524          * Unload & flush inodes
 525          *
 526          * XXX illegal to call this from here, it can only be done from
 527          * the flusher.
 528          */
 529         RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
 530                 hammer_unload_inode, (void *)MNT_WAIT);
 531
 532         /*
 533          * Unload & flush volumes
 534          */
 535 #endif
 536         /*
 537          * Unload buffers and then volumes
 538          */
 539         RB_SCAN(hammer_buf_rb_tree, &hmp->rb_bufs_root, NULL,
 540                 hammer_unload_buffer, NULL);
 541         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
 542                 hammer_unload_volume, NULL);
 543
 544         mp->mnt_data = NULL;
 545         mp->mnt_flag &= ~MNT_LOCAL;
 546         hmp->mp = NULL;
 547         hammer_destroy_objid_cache(hmp);
 548         kfree(hmp->zbuf, M_HAMMER);
 549         lockuninit(&hmp->blockmap_lock);
 550
 551         for (i = 0; i < HAMMER_MAX_ZONES; ++i)
 552                 hammer_free_holes(hmp, &hmp->holes[i]);
 553
 554         kfree(hmp, M_HAMMER);
 555 }
 556
 557 /*
 558  * Obtain a vnode for the specified inode number.  An exclusively locked
 559  * vnode is returned.
 560  */
 561 int
 562 hammer_vfs_vget(struct mount *mp, ino_t ino, struct vnode **vpp)
 563 {
 564         struct hammer_transaction trans;
 565         struct hammer_mount *hmp = (void *)mp->mnt_data;
 566         struct hammer_inode *ip;
 567         int error;
 568
 569         hammer_simple_transaction(&trans, hmp);
 570
 571         /*
 572          * Lookup the requested HAMMER inode.  The structure must be
 573          * left unlocked while we manipulate the related vnode to avoid
 574          * a deadlock.
 575          */
 576         ip = hammer_get_inode(&trans, NULL, ino, hmp->asof, 0, &error);
 577         if (ip == NULL) {
 578                 *vpp = NULL;
 579                 return(error);
 580         }
 581         error = hammer_get_vnode(ip, vpp);
 582         hammer_rel_inode(ip, 0);
 583         hammer_done_transaction(&trans);
 584         return (error);
 585 }
 586
 587 /*
 588  * Return the root vnode for the filesystem.
 589  *
 590  * HAMMER stores the root vnode in the hammer_mount structure so
 591  * getting it is easy.
 592  */
 593 static int
 594 hammer_vfs_root(struct mount *mp, struct vnode **vpp)
 595 {
 596 #if 0
 597         struct hammer_mount *hmp = (void *)mp->mnt_data;
 598 #endif
 599         int error;
 600
 601         error = hammer_vfs_vget(mp, 1, vpp);
 602         return (error);
 603 }
 604
 605 static int
 606 hammer_vfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
 607 {
 608         struct hammer_mount *hmp = (void *)mp->mnt_data;
 609         hammer_volume_t volume;
 610         hammer_volume_ondisk_t ondisk;
 611         int error;
 612         int64_t bfree;
 613
 614         volume = hammer_get_root_volume(hmp, &error);
 615         if (error)
 616                 return(error);
 617         ondisk = volume->ondisk;
 618
 619         /*
 620          * Basic stats
 621          */
 622         mp->mnt_stat.f_files = ondisk->vol0_stat_inodes;
 623         bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE;
 624         hammer_rel_volume(volume, 0);
 625
 626         mp->mnt_stat.f_bfree = bfree / HAMMER_BUFSIZE;
 627         mp->mnt_stat.f_bavail = mp->mnt_stat.f_bfree;
 628         if (mp->mnt_stat.f_files < 0)
 629                 mp->mnt_stat.f_files = 0;
 630
 631         *sbp = mp->mnt_stat;
 632         return(0);
 633 }
 634
 635 static int
 636 hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred)
 637 {
 638         struct hammer_mount *hmp = (void *)mp->mnt_data;
 639         hammer_volume_t volume;
 640         hammer_volume_ondisk_t ondisk;
 641         int error;
 642         int64_t bfree;
 643
 644         volume = hammer_get_root_volume(hmp, &error);
 645         if (error)
 646                 return(error);
 647         ondisk = volume->ondisk;
 648
 649         /*
 650          * Basic stats
 651          */
 652         mp->mnt_vstat.f_files = ondisk->vol0_stat_inodes;
 653         bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE;
 654         hammer_rel_volume(volume, 0);
 655
 656         mp->mnt_vstat.f_bfree = bfree / HAMMER_BUFSIZE;
 657         mp->mnt_vstat.f_bavail = mp->mnt_stat.f_bfree;
 658         if (mp->mnt_vstat.f_files < 0)
 659                 mp->mnt_vstat.f_files = 0;
 660         *sbp = mp->mnt_vstat;
 661         return(0);
 662 }
 663
 664 /*
 665  * Sync the filesystem.  Currently we have to run it twice, the second
 666  * one will advance the undo start index to the end index, so if a crash
 667  * occurs no undos will be run on mount.
 668  *
 669  * We do not sync the filesystem if we are called from a panic.  If we did
 670  * we might end up blowing up a sync that was already in progress.
 671  */
 672 static int
 673 hammer_vfs_sync(struct mount *mp, int waitfor)
 674 {
 675         struct hammer_mount *hmp = (void *)mp->mnt_data;
 676         int error;
 677
 678         if (panicstr == NULL) {
 679                 error = hammer_sync_hmp(hmp, waitfor);
 680                 if (error == 0)
 681                         error = hammer_sync_hmp(hmp, waitfor);
 682         } else {
 683                 error = EIO;
 684         }
 685         return (error);
 686 }
 687
 688 /*
 689  * Convert a vnode to a file handle.
 690  */
 691 static int
 692 hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp)
 693 {
 694         hammer_inode_t ip;
 695
 696         KKASSERT(MAXFIDSZ >= 16);
 697         ip = VTOI(vp);
 698         fhp->fid_len = offsetof(struct fid, fid_data[16]);
 699         fhp->fid_reserved = 0;
 700         bcopy(&ip->obj_id, fhp->fid_data + 0, sizeof(ip->obj_id));
 701         bcopy(&ip->obj_asof, fhp->fid_data + 8, sizeof(ip->obj_asof));
 702         return(0);
 703 }
 704
 705
 706 /*
 707  * Convert a file handle back to a vnode.
 708  */
 709 static int
 710 hammer_vfs_fhtovp(struct mount *mp, struct fid *fhp, struct vnode **vpp)
 711 {
 712         struct hammer_transaction trans;
 713         struct hammer_inode *ip;
 714         struct hammer_inode_info info;
 715         int error;
 716
 717         bcopy(fhp->fid_data + 0, &info.obj_id, sizeof(info.obj_id));
 718         bcopy(fhp->fid_data + 8, &info.obj_asof, sizeof(info.obj_asof));
 719
 720         hammer_simple_transaction(&trans, (void *)mp->mnt_data);
 721
 722         /*
 723          * Get/allocate the hammer_inode structure.  The structure must be
 724          * unlocked while we manipulate the related vnode to avoid a
 725          * deadlock.
 726          */
 727         ip = hammer_get_inode(&trans, NULL, info.obj_id, info.obj_asof,
 728                               0, &error);
 729         if (ip == NULL) {
 730                 *vpp = NULL;
 731                 return(error);
 732         }
 733         error = hammer_get_vnode(ip, vpp);
 734         hammer_rel_inode(ip, 0);
 735         hammer_done_transaction(&trans);
 736         return (error);
 737 }
 738
 739 static int
 740 hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
 741                     int *exflagsp, struct ucred **credanonp)
 742 {
 743         hammer_mount_t hmp = (void *)mp->mnt_data;
 744         struct netcred *np;
 745         int error;
 746
 747         np = vfs_export_lookup(mp, &hmp->export, nam);
 748         if (np) {
 749                 *exflagsp = np->netc_exflags;
 750                 *credanonp = &np->netc_anon;
 751                 error = 0;
 752         } else {
 753                 error = EACCES;
 754         }
 755         return (error);
 756
 757 }
 758
 759 int
 760 hammer_vfs_export(struct mount *mp, int op, const struct export_args *export)
 761 {
 762         hammer_mount_t hmp = (void *)mp->mnt_data;
 763         int error;
 764
 765         switch(op) {
 766         case MOUNTCTL_SET_EXPORT:
 767                 error = vfs_export(mp, &hmp->export, export);
 768                 break;
 769         default:
 770                 error = EOPNOTSUPP;
 771                 break;
 772         }
 773         return(error);
 774 }
 775