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 1.47 2008/06/13 00:25:33 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_debug_cluster_enable = 1;    /* enable read clustering by default */
  60 int hammer_count_inodes;
  61 int hammer_count_iqueued;
  62 int hammer_count_reclaiming;
  63 int hammer_count_records;
  64 int hammer_count_record_datas;
  65 int hammer_count_volumes;
  66 int hammer_count_buffers;
  67 int hammer_count_nodes;
  68 int hammer_count_dirtybufs;             /* global */
  69 int hammer_count_refedbufs;             /* global */
  70 int hammer_count_reservations;
  71 int hammer_count_io_running_read;
  72 int hammer_count_io_running_write;
  73 int hammer_count_io_locked;
  74 int hammer_stats_btree_iterations;
  75 int hammer_stats_record_iterations;
  76 int hammer_limit_dirtybufs;             /* per-mount */
  77 int hammer_limit_irecs;                 /* per-inode */
  78 int hammer_limit_recs;                  /* as a whole XXX */
  79 int hammer_limit_iqueued;               /* per-mount */
  80 int hammer_bio_count;
  81 int64_t hammer_contention_count;
  82 int64_t hammer_zone_limit;
  83
  84 SYSCTL_NODE(_vfs, OID_AUTO, hammer, CTLFLAG_RW, 0, "HAMMER filesystem");
  85 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_general, CTLFLAG_RW,
  86            &hammer_debug_general, 0, "");
  87 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_io, CTLFLAG_RW,
  88            &hammer_debug_io, 0, "");
  89 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_debug, CTLFLAG_RW,
  90            &hammer_debug_debug, 0, "");
  91 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_inode, CTLFLAG_RW,
  92            &hammer_debug_inode, 0, "");
  93 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_locks, CTLFLAG_RW,
  94            &hammer_debug_locks, 0, "");
  95 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_btree, CTLFLAG_RW,
  96            &hammer_debug_btree, 0, "");
  97 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_tid, CTLFLAG_RW,
  98            &hammer_debug_tid, 0, "");
  99 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover, CTLFLAG_RW,
 100            &hammer_debug_recover, 0, "");
 101 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover_faults, CTLFLAG_RW,
 102            &hammer_debug_recover_faults, 0, "");
 103 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_cluster_enable, CTLFLAG_RW,
 104            &hammer_debug_cluster_enable, 0, "");
 105
 106 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_dirtybufs, CTLFLAG_RW,
 107            &hammer_limit_dirtybufs, 0, "");
 108 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_irecs, CTLFLAG_RW,
 109            &hammer_limit_irecs, 0, "");
 110 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_recs, CTLFLAG_RW,
 111            &hammer_limit_recs, 0, "");
 112 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_iqueued, CTLFLAG_RW,
 113            &hammer_limit_iqueued, 0, "");
 114
 115 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_inodes, CTLFLAG_RD,
 116            &hammer_count_inodes, 0, "");
 117 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_iqueued, CTLFLAG_RD,
 118            &hammer_count_iqueued, 0, "");
 119 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reclaiming, CTLFLAG_RD,
 120            &hammer_count_reclaiming, 0, "");
 121 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_records, CTLFLAG_RD,
 122            &hammer_count_records, 0, "");
 123 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_record_datas, CTLFLAG_RD,
 124            &hammer_count_record_datas, 0, "");
 125 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_volumes, CTLFLAG_RD,
 126            &hammer_count_volumes, 0, "");
 127 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_buffers, CTLFLAG_RD,
 128            &hammer_count_buffers, 0, "");
 129 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_nodes, CTLFLAG_RD,
 130            &hammer_count_nodes, 0, "");
 131 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_dirtybufs, CTLFLAG_RD,
 132            &hammer_count_dirtybufs, 0, "");
 133 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_refedbufs, CTLFLAG_RD,
 134            &hammer_count_refedbufs, 0, "");
 135 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reservations, CTLFLAG_RD,
 136            &hammer_count_reservations, 0, "");
 137 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_running_read, CTLFLAG_RD,
 138            &hammer_count_io_running_read, 0, "");
 139 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_locked, CTLFLAG_RD,
 140            &hammer_count_io_locked, 0, "");
 141 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_running_write, CTLFLAG_RD,
 142            &hammer_count_io_running_write, 0, "");
 143 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, zone_limit, CTLFLAG_RW,
 144            &hammer_zone_limit, 0, "");
 145 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, contention_count, CTLFLAG_RW,
 146            &hammer_contention_count, 0, "");
 147
 148 /*
 149  * VFS ABI
 150  */
 151 static void     hammer_free_hmp(struct mount *mp);
 152
 153 static int      hammer_vfs_mount(struct mount *mp, char *path, caddr_t data,
 154                                 struct ucred *cred);
 155 static int      hammer_vfs_unmount(struct mount *mp, int mntflags);
 156 static int      hammer_vfs_root(struct mount *mp, struct vnode **vpp);
 157 static int      hammer_vfs_statfs(struct mount *mp, struct statfs *sbp,
 158                                 struct ucred *cred);
 159 static int      hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp,
 160                                 struct ucred *cred);
 161 static int      hammer_vfs_sync(struct mount *mp, int waitfor);
 162 static int      hammer_vfs_vget(struct mount *mp, ino_t ino,
 163                                 struct vnode **vpp);
 164 static int      hammer_vfs_init(struct vfsconf *conf);
 165 static int      hammer_vfs_fhtovp(struct mount *mp, struct fid *fhp,
 166                                 struct vnode **vpp);
 167 static int      hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp);
 168 static int      hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
 169                                 int *exflagsp, struct ucred **credanonp);
 170
 171
 172 static struct vfsops hammer_vfsops = {
 173         .vfs_mount      = hammer_vfs_mount,
 174         .vfs_unmount    = hammer_vfs_unmount,
 175         .vfs_root       = hammer_vfs_root,
 176         .vfs_statfs     = hammer_vfs_statfs,
 177         .vfs_statvfs    = hammer_vfs_statvfs,
 178         .vfs_sync       = hammer_vfs_sync,
 179         .vfs_vget       = hammer_vfs_vget,
 180         .vfs_init       = hammer_vfs_init,
 181         .vfs_vptofh     = hammer_vfs_vptofh,
 182         .vfs_fhtovp     = hammer_vfs_fhtovp,
 183         .vfs_checkexp   = hammer_vfs_checkexp
 184 };
 185
 186 MALLOC_DEFINE(M_HAMMER, "hammer-mount", "hammer mount");
 187
 188 VFS_SET(hammer_vfsops, hammer, 0);
 189 MODULE_VERSION(hammer, 1);
 190
 191 static int
 192 hammer_vfs_init(struct vfsconf *conf)
 193 {
 194         if (hammer_limit_irecs == 0)
 195                 hammer_limit_irecs = nbuf * 8;
 196         if (hammer_limit_recs == 0)             /* XXX TODO */
 197                 hammer_limit_recs = nbuf * 25;
 198         if (hammer_limit_dirtybufs == 0) {
 199                 hammer_limit_dirtybufs = hidirtybuffers / 2;
 200                 if (hammer_limit_dirtybufs < 100)
 201                         hammer_limit_dirtybufs = 100;
 202         }
 203         if (hammer_limit_iqueued == 0)
 204                 hammer_limit_iqueued = desiredvnodes / 5;
 205         return(0);
 206 }
 207
 208 static int
 209 hammer_vfs_mount(struct mount *mp, char *mntpt, caddr_t data,
 210                  struct ucred *cred)
 211 {
 212         struct hammer_mount_info info;
 213         hammer_mount_t hmp;
 214         hammer_volume_t rootvol;
 215         struct vnode *rootvp;
 216         const char *upath;      /* volume name in userspace */
 217         char *path;             /* volume name in system space */
 218         int error;
 219         int i;
 220
 221         if ((error = copyin(data, &info, sizeof(info))) != 0)
 222                 return (error);
 223         if ((mp->mnt_flag & MNT_UPDATE) == 0) {
 224                 if (info.nvolumes <= 0 || info.nvolumes >= 32768)
 225                         return (EINVAL);
 226         }
 227
 228         /*
 229          * Interal mount data structure
 230          */
 231         if (mp->mnt_flag & MNT_UPDATE) {
 232                 hmp = (void *)mp->mnt_data;
 233                 KKASSERT(hmp != NULL);
 234         } else {
 235                 hmp = kmalloc(sizeof(*hmp), M_HAMMER, M_WAITOK | M_ZERO);
 236                 mp->mnt_data = (qaddr_t)hmp;
 237                 hmp->mp = mp;
 238                 hmp->zbuf = kmalloc(HAMMER_BUFSIZE, M_HAMMER, M_WAITOK|M_ZERO);
 239                 hmp->namekey_iterator = mycpu->gd_time_seconds;
 240                 /*TAILQ_INIT(&hmp->recycle_list);*/
 241
 242                 hmp->root_btree_beg.localization = HAMMER_MIN_LOCALIZATION;
 243                 hmp->root_btree_beg.obj_id = -0x8000000000000000LL;
 244                 hmp->root_btree_beg.key = -0x8000000000000000LL;
 245                 hmp->root_btree_beg.create_tid = 1;
 246                 hmp->root_btree_beg.delete_tid = 1;
 247                 hmp->root_btree_beg.rec_type = 0;
 248                 hmp->root_btree_beg.obj_type = 0;
 249
 250                 hmp->root_btree_end.localization = HAMMER_MAX_LOCALIZATION;
 251                 hmp->root_btree_end.obj_id = 0x7FFFFFFFFFFFFFFFLL;
 252                 hmp->root_btree_end.key = 0x7FFFFFFFFFFFFFFFLL;
 253                 hmp->root_btree_end.create_tid = 0xFFFFFFFFFFFFFFFFULL;
 254                 hmp->root_btree_end.delete_tid = 0;   /* special case */
 255                 hmp->root_btree_end.rec_type = 0xFFFFU;
 256                 hmp->root_btree_end.obj_type = 0;
 257
 258                 hmp->sync_lock.refs = 1;
 259                 hmp->free_lock.refs = 1;
 260                 hmp->undo_lock.refs = 1;
 261                 hmp->blkmap_lock.refs = 1;
 262
 263                 TAILQ_INIT(&hmp->flush_list);
 264                 TAILQ_INIT(&hmp->delay_list);
 265                 TAILQ_INIT(&hmp->objid_cache_list);
 266                 TAILQ_INIT(&hmp->undo_lru_list);
 267                 TAILQ_INIT(&hmp->reclaim_list);
 268
 269                 /*
 270                  * Set default zone limits.  This value can be reduced
 271                  * further by the zone limit specified in the root volume.
 272                  *
 273                  * The sysctl can force a small zone limit for debugging
 274                  * purposes.
 275                  */
 276                 for (i = 0; i < HAMMER_MAX_ZONES; ++i) {
 277                         hmp->zone_limits[i] =
 278                                 HAMMER_ZONE_ENCODE(i, HAMMER_ZONE_LIMIT);
 279
 280                         if (hammer_zone_limit) {
 281                                 hmp->zone_limits[i] =
 282                                     HAMMER_ZONE_ENCODE(i, hammer_zone_limit);
 283                         }
 284                         hammer_init_holes(hmp, &hmp->holes[i]);
 285                 }
 286         }
 287         hmp->hflags &= ~HMNT_USERFLAGS;
 288         hmp->hflags |= info.hflags & HMNT_USERFLAGS;
 289         if (info.asof) {
 290                 kprintf("ASOF\n");
 291                 mp->mnt_flag |= MNT_RDONLY;
 292                 hmp->asof = info.asof;
 293         } else {
 294                 hmp->asof = HAMMER_MAX_TID;
 295         }
 296
 297         /*
 298          * Re-open read-write if originally read-only, or vise-versa.
 299          */
 300         if (mp->mnt_flag & MNT_UPDATE) {
 301                 error = 0;
 302                 if (hmp->ronly && (mp->mnt_kern_flag & MNTK_WANTRDWR)) {
 303                         kprintf("HAMMER read-only -> read-write\n");
 304                         hmp->ronly = 0;
 305                         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
 306                                 hammer_adjust_volume_mode, NULL);
 307                         rootvol = hammer_get_root_volume(hmp, &error);
 308                         if (rootvol) {
 309                                 hammer_recover_flush_buffers(hmp, rootvol);
 310                                 bcopy(rootvol->ondisk->vol0_blockmap,
 311                                       hmp->blockmap,
 312                                       sizeof(hmp->blockmap));
 313                                 hammer_rel_volume(rootvol, 0);
 314                         }
 315                         RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
 316                                 hammer_reload_inode, NULL);
 317                         /* kernel clears MNT_RDONLY */
 318                 } else if (hmp->ronly == 0 && (mp->mnt_flag & MNT_RDONLY)) {
 319                         kprintf("HAMMER read-write -> read-only\n");
 320                         hmp->ronly = 1; /* messy */
 321                         RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
 322                                 hammer_reload_inode, NULL);
 323                         hmp->ronly = 0;
 324                         hammer_flusher_sync(hmp);
 325                         hammer_flusher_sync(hmp);
 326                         hammer_flusher_sync(hmp);
 327                         hmp->ronly = 1;
 328                         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
 329                                 hammer_adjust_volume_mode, NULL);
 330                 }
 331                 return(error);
 332         }
 333
 334         RB_INIT(&hmp->rb_vols_root);
 335         RB_INIT(&hmp->rb_inos_root);
 336         RB_INIT(&hmp->rb_nods_root);
 337         RB_INIT(&hmp->rb_undo_root);
 338         RB_INIT(&hmp->rb_resv_root);
 339         RB_INIT(&hmp->rb_bufs_root);
 340
 341         hmp->ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
 342
 343         TAILQ_INIT(&hmp->volu_list);
 344         TAILQ_INIT(&hmp->undo_list);
 345         TAILQ_INIT(&hmp->data_list);
 346         TAILQ_INIT(&hmp->meta_list);
 347         TAILQ_INIT(&hmp->lose_list);
 348
 349         /*
 350          * Load volumes
 351          */
 352         path = objcache_get(namei_oc, M_WAITOK);
 353         hmp->nvolumes = info.nvolumes;
 354         for (i = 0; i < info.nvolumes; ++i) {
 355                 error = copyin(&info.volumes[i], &upath, sizeof(char *));
 356                 if (error == 0)
 357                         error = copyinstr(upath, path, MAXPATHLEN, NULL);
 358                 if (error == 0)
 359                         error = hammer_install_volume(hmp, path);
 360                 if (error)
 361                         break;
 362         }
 363         objcache_put(namei_oc, path);
 364
 365         /*
 366          * Make sure we found a root volume
 367          */
 368         if (error == 0 && hmp->rootvol == NULL) {
 369                 kprintf("hammer_mount: No root volume found!\n");
 370                 error = EINVAL;
 371         }
 372         if (error) {
 373                 hammer_free_hmp(mp);
 374                 return (error);
 375         }
 376
 377         /*
 378          * No errors, setup enough of the mount point so we can lookup the
 379          * root vnode.
 380          */
 381         mp->mnt_iosize_max = MAXPHYS;
 382         mp->mnt_kern_flag |= MNTK_FSMID;
 383
 384         /*
 385          * note: f_iosize is used by vnode_pager_haspage() when constructing
 386          * its VOP_BMAP call.
 387          */
 388         mp->mnt_stat.f_iosize = HAMMER_BUFSIZE;
 389         mp->mnt_stat.f_bsize = HAMMER_BUFSIZE;
 390
 391         mp->mnt_vstat.f_frsize = HAMMER_BUFSIZE;
 392         mp->mnt_vstat.f_bsize = HAMMER_BUFSIZE;
 393
 394         mp->mnt_maxsymlinklen = 255;
 395         mp->mnt_flag |= MNT_LOCAL;
 396
 397         vfs_add_vnodeops(mp, &hammer_vnode_vops, &mp->mnt_vn_norm_ops);
 398         vfs_add_vnodeops(mp, &hammer_spec_vops, &mp->mnt_vn_spec_ops);
 399         vfs_add_vnodeops(mp, &hammer_fifo_vops, &mp->mnt_vn_fifo_ops);
 400
 401         /*
 402          * The root volume's ondisk pointer is only valid if we hold a
 403          * reference to it.
 404          */
 405         rootvol = hammer_get_root_volume(hmp, &error);
 406         if (error)
 407                 goto failed;
 408
 409         /*
 410          * Perform any necessary UNDO operations.  The recovery code does
 411          * call hammer_undo_lookup() so we have to pre-cache the blockmap,
 412          * and then re-copy it again after recovery is complete.
 413          *
 414          * If this is a read-only mount the UNDO information is retained
 415          * in memory in the form of dirty buffer cache buffers, and not
 416          * written back to the media.
 417          */
 418         bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
 419               sizeof(hmp->blockmap));
 420
 421         error = hammer_recover(hmp, rootvol);
 422         if (error) {
 423                 kprintf("Failed to recover HAMMER filesystem on mount\n");
 424                 goto done;
 425         }
 426
 427         /*
 428          * Finish setup now that we have a good root volume
 429          */
 430         ksnprintf(mp->mnt_stat.f_mntfromname,
 431                   sizeof(mp->mnt_stat.f_mntfromname), "%s",
 432                   rootvol->ondisk->vol_name);
 433         mp->mnt_stat.f_fsid.val[0] =
 434                 crc32((char *)&rootvol->ondisk->vol_fsid + 0, 8);
 435         mp->mnt_stat.f_fsid.val[1] =
 436                 crc32((char *)&rootvol->ondisk->vol_fsid + 8, 8);
 437
 438         mp->mnt_vstat.f_fsid_uuid = rootvol->ondisk->vol_fsid;
 439         mp->mnt_vstat.f_fsid = crc32(&mp->mnt_vstat.f_fsid_uuid,
 440                                      sizeof(mp->mnt_vstat.f_fsid_uuid));
 441
 442         /*
 443          * Certain often-modified fields in the root volume are cached in
 444          * the hammer_mount structure so we do not have to generate lots
 445          * of little UNDO structures for them.
 446          *
 447          * Recopy after recovery.  This also has the side effect of
 448          * setting our cached undo FIFO's first_offset, which serves to
 449          * placemark the FIFO start for the NEXT flush cycle while the
 450          * on-disk first_offset represents the LAST flush cycle.
 451          */
 452         hmp->next_tid = rootvol->ondisk->vol0_next_tid;
 453         bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
 454               sizeof(hmp->blockmap));
 455         hmp->copy_stat_freebigblocks = rootvol->ondisk->vol0_stat_freebigblocks;
 456
 457         /*
 458          * Use the zone limit set by newfs_hammer, or the zone limit set by
 459          * sysctl (for debugging), whichever is smaller.
 460          */
 461         if (rootvol->ondisk->vol0_zone_limit) {
 462                 hammer_off_t vol0_zone_limit;
 463
 464                 vol0_zone_limit = rootvol->ondisk->vol0_zone_limit;
 465                 for (i = 0; i < HAMMER_MAX_ZONES; ++i) {
 466                         if (hmp->zone_limits[i] > vol0_zone_limit)
 467                                 hmp->zone_limits[i] = vol0_zone_limit;
 468                 }
 469         }
 470
 471         hammer_flusher_create(hmp);
 472
 473         /*
 474          * Locate the root directory using the root cluster's B-Tree as a
 475          * starting point.  The root directory uses an obj_id of 1.
 476          *
 477          * FUTURE: Leave the root directory cached referenced but unlocked
 478          * in hmp->rootvp (need to flush it on unmount).
 479          */
 480         error = hammer_vfs_vget(mp, 1, &rootvp);
 481         if (error)
 482                 goto done;
 483         vput(rootvp);
 484         /*vn_unlock(hmp->rootvp);*/
 485
 486 done:
 487         hammer_rel_volume(rootvol, 0);
 488 failed:
 489         /*
 490          * Cleanup and return.
 491          */
 492         if (error)
 493                 hammer_free_hmp(mp);
 494         return (error);
 495 }
 496
 497 static int
 498 hammer_vfs_unmount(struct mount *mp, int mntflags)
 499 {
 500 #if 0
 501         struct hammer_mount *hmp = (void *)mp->mnt_data;
 502 #endif
 503         int flags;
 504         int error;
 505
 506         /*
 507          * Clean out the vnodes
 508          */
 509         flags = 0;
 510         if (mntflags & MNT_FORCE)
 511                 flags |= FORCECLOSE;
 512         if ((error = vflush(mp, 0, flags)) != 0)
 513                 return (error);
 514
 515         /*
 516          * Clean up the internal mount structure and related entities.  This
 517          * may issue I/O.
 518          */
 519         hammer_free_hmp(mp);
 520         return(0);
 521 }
 522
 523 /*
 524  * Clean up the internal mount structure and disassociate it from the mount.
 525  * This may issue I/O.
 526  */
 527 static void
 528 hammer_free_hmp(struct mount *mp)
 529 {
 530         struct hammer_mount *hmp = (void *)mp->mnt_data;
 531         int i;
 532
 533 #if 0
 534         /*
 535          * Clean up the root vnode
 536          */
 537         if (hmp->rootvp) {
 538                 vrele(hmp->rootvp);
 539                 hmp->rootvp = NULL;
 540         }
 541 #endif
 542         kprintf("X1");
 543         hammer_flusher_sync(hmp);
 544         kprintf("X2");
 545         hammer_flusher_sync(hmp);
 546         kprintf("X3");
 547         hammer_flusher_destroy(hmp);
 548         kprintf("X4");
 549
 550         KKASSERT(RB_EMPTY(&hmp->rb_inos_root));
 551
 552 #if 0
 553         /*
 554          * Unload & flush inodes
 555          *
 556          * XXX illegal to call this from here, it can only be done from
 557          * the flusher.
 558          */
 559         RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
 560                 hammer_unload_inode, (void *)MNT_WAIT);
 561
 562         /*
 563          * Unload & flush volumes
 564          */
 565 #endif
 566         /*
 567          * Unload buffers and then volumes
 568          */
 569         RB_SCAN(hammer_buf_rb_tree, &hmp->rb_bufs_root, NULL,
 570                 hammer_unload_buffer, NULL);
 571         kprintf("X5");
 572         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
 573                 hammer_unload_volume, NULL);
 574         kprintf("X6");
 575
 576         mp->mnt_data = NULL;
 577         mp->mnt_flag &= ~MNT_LOCAL;
 578         hmp->mp = NULL;
 579         kprintf("X7");
 580         hammer_destroy_objid_cache(hmp);
 581         kprintf("X8");
 582         kfree(hmp->zbuf, M_HAMMER);
 583         kprintf("X9");
 584         kprintf("X10");
 585
 586         for (i = 0; i < HAMMER_MAX_ZONES; ++i)
 587                 hammer_free_holes(hmp, &hmp->holes[i]);
 588         kprintf("X11");
 589
 590         kfree(hmp, M_HAMMER);
 591         kprintf("X12");
 592 }
 593
 594 /*
 595  * Obtain a vnode for the specified inode number.  An exclusively locked
 596  * vnode is returned.
 597  */
 598 int
 599 hammer_vfs_vget(struct mount *mp, ino_t ino, struct vnode **vpp)
 600 {
 601         struct hammer_transaction trans;
 602         struct hammer_mount *hmp = (void *)mp->mnt_data;
 603         struct hammer_inode *ip;
 604         int error;
 605
 606         hammer_simple_transaction(&trans, hmp);
 607
 608         /*
 609          * Lookup the requested HAMMER inode.  The structure must be
 610          * left unlocked while we manipulate the related vnode to avoid
 611          * a deadlock.
 612          */
 613         ip = hammer_get_inode(&trans, NULL, ino, hmp->asof, 0, &error);
 614         if (ip == NULL) {
 615                 *vpp = NULL;
 616                 return(error);
 617         }
 618         error = hammer_get_vnode(ip, vpp);
 619         hammer_rel_inode(ip, 0);
 620         hammer_done_transaction(&trans);
 621         return (error);
 622 }
 623
 624 /*
 625  * Return the root vnode for the filesystem.
 626  *
 627  * HAMMER stores the root vnode in the hammer_mount structure so
 628  * getting it is easy.
 629  */
 630 static int
 631 hammer_vfs_root(struct mount *mp, struct vnode **vpp)
 632 {
 633 #if 0
 634         struct hammer_mount *hmp = (void *)mp->mnt_data;
 635 #endif
 636         int error;
 637
 638         error = hammer_vfs_vget(mp, 1, vpp);
 639         return (error);
 640 }
 641
 642 static int
 643 hammer_vfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
 644 {
 645         struct hammer_mount *hmp = (void *)mp->mnt_data;
 646         hammer_volume_t volume;
 647         hammer_volume_ondisk_t ondisk;
 648         int error;
 649         int64_t bfree;
 650
 651         volume = hammer_get_root_volume(hmp, &error);
 652         if (error)
 653                 return(error);
 654         ondisk = volume->ondisk;
 655
 656         /*
 657          * Basic stats
 658          */
 659         mp->mnt_stat.f_files = ondisk->vol0_stat_inodes;
 660         bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE;
 661         hammer_rel_volume(volume, 0);
 662
 663         mp->mnt_stat.f_bfree = bfree / HAMMER_BUFSIZE;
 664         mp->mnt_stat.f_bavail = mp->mnt_stat.f_bfree;
 665         if (mp->mnt_stat.f_files < 0)
 666                 mp->mnt_stat.f_files = 0;
 667
 668         *sbp = mp->mnt_stat;
 669         return(0);
 670 }
 671
 672 static int
 673 hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred)
 674 {
 675         struct hammer_mount *hmp = (void *)mp->mnt_data;
 676         hammer_volume_t volume;
 677         hammer_volume_ondisk_t ondisk;
 678         int error;
 679         int64_t bfree;
 680
 681         volume = hammer_get_root_volume(hmp, &error);
 682         if (error)
 683                 return(error);
 684         ondisk = volume->ondisk;
 685
 686         /*
 687          * Basic stats
 688          */
 689         mp->mnt_vstat.f_files = ondisk->vol0_stat_inodes;
 690         bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE;
 691         hammer_rel_volume(volume, 0);
 692
 693         mp->mnt_vstat.f_bfree = bfree / HAMMER_BUFSIZE;
 694         mp->mnt_vstat.f_bavail = mp->mnt_stat.f_bfree;
 695         if (mp->mnt_vstat.f_files < 0)
 696                 mp->mnt_vstat.f_files = 0;
 697         *sbp = mp->mnt_vstat;
 698         return(0);
 699 }
 700
 701 /*
 702  * Sync the filesystem.  Currently we have to run it twice, the second
 703  * one will advance the undo start index to the end index, so if a crash
 704  * occurs no undos will be run on mount.
 705  *
 706  * We do not sync the filesystem if we are called from a panic.  If we did
 707  * we might end up blowing up a sync that was already in progress.
 708  */
 709 static int
 710 hammer_vfs_sync(struct mount *mp, int waitfor)
 711 {
 712         struct hammer_mount *hmp = (void *)mp->mnt_data;
 713         int error;
 714
 715         if (panicstr == NULL) {
 716                 error = hammer_sync_hmp(hmp, waitfor);
 717                 if (error == 0)
 718                         error = hammer_sync_hmp(hmp, waitfor);
 719         } else {
 720                 error = EIO;
 721         }
 722         return (error);
 723 }
 724
 725 /*
 726  * Convert a vnode to a file handle.
 727  */
 728 static int
 729 hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp)
 730 {
 731         hammer_inode_t ip;
 732
 733         KKASSERT(MAXFIDSZ >= 16);
 734         ip = VTOI(vp);
 735         fhp->fid_len = offsetof(struct fid, fid_data[16]);
 736         fhp->fid_reserved = 0;
 737         bcopy(&ip->obj_id, fhp->fid_data + 0, sizeof(ip->obj_id));
 738         bcopy(&ip->obj_asof, fhp->fid_data + 8, sizeof(ip->obj_asof));
 739         return(0);
 740 }
 741
 742
 743 /*
 744  * Convert a file handle back to a vnode.
 745  */
 746 static int
 747 hammer_vfs_fhtovp(struct mount *mp, struct fid *fhp, struct vnode **vpp)
 748 {
 749         struct hammer_transaction trans;
 750         struct hammer_inode *ip;
 751         struct hammer_inode_info info;
 752         int error;
 753
 754         bcopy(fhp->fid_data + 0, &info.obj_id, sizeof(info.obj_id));
 755         bcopy(fhp->fid_data + 8, &info.obj_asof, sizeof(info.obj_asof));
 756
 757         hammer_simple_transaction(&trans, (void *)mp->mnt_data);
 758
 759         /*
 760          * Get/allocate the hammer_inode structure.  The structure must be
 761          * unlocked while we manipulate the related vnode to avoid a
 762          * deadlock.
 763          */
 764         ip = hammer_get_inode(&trans, NULL, info.obj_id, info.obj_asof,
 765                               0, &error);
 766         if (ip == NULL) {
 767                 *vpp = NULL;
 768                 return(error);
 769         }
 770         error = hammer_get_vnode(ip, vpp);
 771         hammer_rel_inode(ip, 0);
 772         hammer_done_transaction(&trans);
 773         return (error);
 774 }
 775
 776 static int
 777 hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
 778                     int *exflagsp, struct ucred **credanonp)
 779 {
 780         hammer_mount_t hmp = (void *)mp->mnt_data;
 781         struct netcred *np;
 782         int error;
 783
 784         np = vfs_export_lookup(mp, &hmp->export, nam);
 785         if (np) {
 786                 *exflagsp = np->netc_exflags;
 787                 *credanonp = &np->netc_anon;
 788                 error = 0;
 789         } else {
 790                 error = EACCES;
 791         }
 792         return (error);
 793
 794 }
 795
 796 int
 797 hammer_vfs_export(struct mount *mp, int op, const struct export_args *export)
 798 {
 799         hammer_mount_t hmp = (void *)mp->mnt_data;
 800         int error;
 801
 802         switch(op) {
 803         case MOUNTCTL_SET_EXPORT:
 804                 error = vfs_export(mp, &hmp->export, export);
 805                 break;
 806         default:
 807                 error = EOPNOTSUPP;
 808                 break;
 809         }
 810         return(error);
 811 }
 812