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 1.50 2008/06/20 05:38:26 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 int64_t hammer_stats_btree_lookups;
  69 int64_t hammer_stats_btree_searches;
  70 int64_t hammer_stats_btree_inserts;
  71 int64_t hammer_stats_btree_deletes;
  72 int64_t hammer_stats_btree_elements;
  73 int64_t hammer_stats_btree_splits;
  74 int64_t hammer_stats_btree_iterations;
  75 int64_t hammer_stats_record_iterations;
  76 int hammer_count_dirtybufs;             /* global */
  77 int hammer_count_refedbufs;             /* global */
  78 int hammer_count_reservations;
  79 int hammer_count_io_running_read;
  80 int hammer_count_io_running_write;
  81 int hammer_count_io_locked;
  82 int hammer_limit_dirtybufs;             /* per-mount */
  83 int hammer_limit_recs;                  /* as a whole XXX */
  84 int hammer_limit_iqueued;               /* per-mount */
  85 int hammer_bio_count;
  86 int hammer_verify_zone;
  87 int hammer_write_mode;
  88 int64_t hammer_contention_count;
  89 int64_t hammer_zone_limit;
  90
  91 SYSCTL_NODE(_vfs, OID_AUTO, hammer, CTLFLAG_RW, 0, "HAMMER filesystem");
  92 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_general, CTLFLAG_RW,
  93            &hammer_debug_general, 0, "");
  94 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_io, CTLFLAG_RW,
  95            &hammer_debug_io, 0, "");
  96 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_debug, CTLFLAG_RW,
  97            &hammer_debug_debug, 0, "");
  98 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_inode, CTLFLAG_RW,
  99            &hammer_debug_inode, 0, "");
 100 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_locks, CTLFLAG_RW,
 101            &hammer_debug_locks, 0, "");
 102 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_btree, CTLFLAG_RW,
 103            &hammer_debug_btree, 0, "");
 104 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_tid, CTLFLAG_RW,
 105            &hammer_debug_tid, 0, "");
 106 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover, CTLFLAG_RW,
 107            &hammer_debug_recover, 0, "");
 108 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover_faults, CTLFLAG_RW,
 109            &hammer_debug_recover_faults, 0, "");
 110 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_cluster_enable, CTLFLAG_RW,
 111            &hammer_debug_cluster_enable, 0, "");
 112
 113 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_dirtybufs, CTLFLAG_RW,
 114            &hammer_limit_dirtybufs, 0, "");
 115 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_recs, CTLFLAG_RW,
 116            &hammer_limit_recs, 0, "");
 117 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_iqueued, CTLFLAG_RW,
 118            &hammer_limit_iqueued, 0, "");
 119
 120 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_inodes, CTLFLAG_RD,
 121            &hammer_count_inodes, 0, "");
 122 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_iqueued, CTLFLAG_RD,
 123            &hammer_count_iqueued, 0, "");
 124 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reclaiming, CTLFLAG_RD,
 125            &hammer_count_reclaiming, 0, "");
 126 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_records, CTLFLAG_RD,
 127            &hammer_count_records, 0, "");
 128 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_record_datas, CTLFLAG_RD,
 129            &hammer_count_record_datas, 0, "");
 130 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_volumes, CTLFLAG_RD,
 131            &hammer_count_volumes, 0, "");
 132 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_buffers, CTLFLAG_RD,
 133            &hammer_count_buffers, 0, "");
 134 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_nodes, CTLFLAG_RD,
 135            &hammer_count_nodes, 0, "");
 136 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_searches, CTLFLAG_RD,
 137            &hammer_stats_btree_searches, 0, "");
 138 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_lookups, CTLFLAG_RD,
 139            &hammer_stats_btree_lookups, 0, "");
 140 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_inserts, CTLFLAG_RD,
 141            &hammer_stats_btree_inserts, 0, "");
 142 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_deletes, CTLFLAG_RD,
 143            &hammer_stats_btree_deletes, 0, "");
 144 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_elements, CTLFLAG_RD,
 145            &hammer_stats_btree_elements, 0, "");
 146 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_splits, CTLFLAG_RD,
 147            &hammer_stats_btree_splits, 0, "");
 148 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_iterations, CTLFLAG_RD,
 149            &hammer_stats_btree_iterations, 0, "");
 150 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_record_iterations, CTLFLAG_RD,
 151            &hammer_stats_record_iterations, 0, "");
 152 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_dirtybufs, CTLFLAG_RD,
 153            &hammer_count_dirtybufs, 0, "");
 154 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_refedbufs, CTLFLAG_RD,
 155            &hammer_count_refedbufs, 0, "");
 156 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reservations, CTLFLAG_RD,
 157            &hammer_count_reservations, 0, "");
 158 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_running_read, CTLFLAG_RD,
 159            &hammer_count_io_running_read, 0, "");
 160 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_locked, CTLFLAG_RD,
 161            &hammer_count_io_locked, 0, "");
 162 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_running_write, CTLFLAG_RD,
 163            &hammer_count_io_running_write, 0, "");
 164 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, zone_limit, CTLFLAG_RW,
 165            &hammer_zone_limit, 0, "");
 166 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, contention_count, CTLFLAG_RW,
 167            &hammer_contention_count, 0, "");
 168 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_zone, CTLFLAG_RW,
 169            &hammer_verify_zone, 0, "");
 170 SYSCTL_INT(_vfs_hammer, OID_AUTO, write_mode, CTLFLAG_RW,
 171            &hammer_write_mode, 0, "");
 172
 173 KTR_INFO_MASTER(hammer);
 174
 175 /*
 176  * VFS ABI
 177  */
 178 static void     hammer_free_hmp(struct mount *mp);
 179
 180 static int      hammer_vfs_mount(struct mount *mp, char *path, caddr_t data,
 181                                 struct ucred *cred);
 182 static int      hammer_vfs_unmount(struct mount *mp, int mntflags);
 183 static int      hammer_vfs_root(struct mount *mp, struct vnode **vpp);
 184 static int      hammer_vfs_statfs(struct mount *mp, struct statfs *sbp,
 185                                 struct ucred *cred);
 186 static int      hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp,
 187                                 struct ucred *cred);
 188 static int      hammer_vfs_sync(struct mount *mp, int waitfor);
 189 static int      hammer_vfs_vget(struct mount *mp, ino_t ino,
 190                                 struct vnode **vpp);
 191 static int      hammer_vfs_init(struct vfsconf *conf);
 192 static int      hammer_vfs_fhtovp(struct mount *mp, struct fid *fhp,
 193                                 struct vnode **vpp);
 194 static int      hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp);
 195 static int      hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
 196                                 int *exflagsp, struct ucred **credanonp);
 197
 198
 199 static struct vfsops hammer_vfsops = {
 200         .vfs_mount      = hammer_vfs_mount,
 201         .vfs_unmount    = hammer_vfs_unmount,
 202         .vfs_root       = hammer_vfs_root,
 203         .vfs_statfs     = hammer_vfs_statfs,
 204         .vfs_statvfs    = hammer_vfs_statvfs,
 205         .vfs_sync       = hammer_vfs_sync,
 206         .vfs_vget       = hammer_vfs_vget,
 207         .vfs_init       = hammer_vfs_init,
 208         .vfs_vptofh     = hammer_vfs_vptofh,
 209         .vfs_fhtovp     = hammer_vfs_fhtovp,
 210         .vfs_checkexp   = hammer_vfs_checkexp
 211 };
 212
 213 MALLOC_DEFINE(M_HAMMER, "hammer-mount", "hammer mount");
 214
 215 VFS_SET(hammer_vfsops, hammer, 0);
 216 MODULE_VERSION(hammer, 1);
 217
 218 static int
 219 hammer_vfs_init(struct vfsconf *conf)
 220 {
 221         if (hammer_limit_recs == 0)             /* XXX TODO */
 222                 hammer_limit_recs = nbuf * 25;
 223         if (hammer_limit_dirtybufs == 0) {
 224                 hammer_limit_dirtybufs = hidirtybuffers / 2;
 225                 if (hammer_limit_dirtybufs < 100)
 226                         hammer_limit_dirtybufs = 100;
 227         }
 228         if (hammer_limit_iqueued == 0)
 229                 hammer_limit_iqueued = desiredvnodes / 5;
 230         return(0);
 231 }
 232
 233 static int
 234 hammer_vfs_mount(struct mount *mp, char *mntpt, caddr_t data,
 235                  struct ucred *cred)
 236 {
 237         struct hammer_mount_info info;
 238         hammer_mount_t hmp;
 239         hammer_volume_t rootvol;
 240         struct vnode *rootvp;
 241         const char *upath;      /* volume name in userspace */
 242         char *path;             /* volume name in system space */
 243         int error;
 244         int i;
 245
 246         if ((error = copyin(data, &info, sizeof(info))) != 0)
 247                 return (error);
 248         if ((mp->mnt_flag & MNT_UPDATE) == 0) {
 249                 if (info.nvolumes <= 0 || info.nvolumes >= 32768)
 250                         return (EINVAL);
 251         }
 252
 253         /*
 254          * Interal mount data structure
 255          */
 256         if (mp->mnt_flag & MNT_UPDATE) {
 257                 hmp = (void *)mp->mnt_data;
 258                 KKASSERT(hmp != NULL);
 259         } else {
 260                 hmp = kmalloc(sizeof(*hmp), M_HAMMER, M_WAITOK | M_ZERO);
 261                 mp->mnt_data = (qaddr_t)hmp;
 262                 hmp->mp = mp;
 263                 hmp->namekey_iterator = mycpu->gd_time_seconds;
 264                 /*TAILQ_INIT(&hmp->recycle_list);*/
 265
 266                 hmp->root_btree_beg.localization = HAMMER_MIN_LOCALIZATION;
 267                 hmp->root_btree_beg.obj_id = -0x8000000000000000LL;
 268                 hmp->root_btree_beg.key = -0x8000000000000000LL;
 269                 hmp->root_btree_beg.create_tid = 1;
 270                 hmp->root_btree_beg.delete_tid = 1;
 271                 hmp->root_btree_beg.rec_type = 0;
 272                 hmp->root_btree_beg.obj_type = 0;
 273
 274                 hmp->root_btree_end.localization = HAMMER_MAX_LOCALIZATION;
 275                 hmp->root_btree_end.obj_id = 0x7FFFFFFFFFFFFFFFLL;
 276                 hmp->root_btree_end.key = 0x7FFFFFFFFFFFFFFFLL;
 277                 hmp->root_btree_end.create_tid = 0xFFFFFFFFFFFFFFFFULL;
 278                 hmp->root_btree_end.delete_tid = 0;   /* special case */
 279                 hmp->root_btree_end.rec_type = 0xFFFFU;
 280                 hmp->root_btree_end.obj_type = 0;
 281
 282                 hmp->sync_lock.refs = 1;
 283                 hmp->free_lock.refs = 1;
 284                 hmp->undo_lock.refs = 1;
 285                 hmp->blkmap_lock.refs = 1;
 286
 287                 TAILQ_INIT(&hmp->flush_list);
 288                 TAILQ_INIT(&hmp->delay_list);
 289                 TAILQ_INIT(&hmp->objid_cache_list);
 290                 TAILQ_INIT(&hmp->undo_lru_list);
 291                 TAILQ_INIT(&hmp->reclaim_list);
 292         }
 293         hmp->hflags &= ~HMNT_USERFLAGS;
 294         hmp->hflags |= info.hflags & HMNT_USERFLAGS;
 295         if (info.asof) {
 296                 kprintf("ASOF\n");
 297                 mp->mnt_flag |= MNT_RDONLY;
 298                 hmp->asof = info.asof;
 299         } else {
 300                 hmp->asof = HAMMER_MAX_TID;
 301         }
 302
 303         /*
 304          * Re-open read-write if originally read-only, or vise-versa.
 305          */
 306         if (mp->mnt_flag & MNT_UPDATE) {
 307                 error = 0;
 308                 if (hmp->ronly && (mp->mnt_kern_flag & MNTK_WANTRDWR)) {
 309                         kprintf("HAMMER read-only -> read-write\n");
 310                         hmp->ronly = 0;
 311                         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
 312                                 hammer_adjust_volume_mode, NULL);
 313                         rootvol = hammer_get_root_volume(hmp, &error);
 314                         if (rootvol) {
 315                                 hammer_recover_flush_buffers(hmp, rootvol);
 316                                 bcopy(rootvol->ondisk->vol0_blockmap,
 317                                       hmp->blockmap,
 318                                       sizeof(hmp->blockmap));
 319                                 hammer_rel_volume(rootvol, 0);
 320                         }
 321                         RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
 322                                 hammer_reload_inode, NULL);
 323                         /* kernel clears MNT_RDONLY */
 324                 } else if (hmp->ronly == 0 && (mp->mnt_flag & MNT_RDONLY)) {
 325                         kprintf("HAMMER read-write -> read-only\n");
 326                         hmp->ronly = 1; /* messy */
 327                         RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
 328                                 hammer_reload_inode, NULL);
 329                         hmp->ronly = 0;
 330                         hammer_flusher_sync(hmp);
 331                         hammer_flusher_sync(hmp);
 332                         hammer_flusher_sync(hmp);
 333                         hmp->ronly = 1;
 334                         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
 335                                 hammer_adjust_volume_mode, NULL);
 336                 }
 337                 return(error);
 338         }
 339
 340         RB_INIT(&hmp->rb_vols_root);
 341         RB_INIT(&hmp->rb_inos_root);
 342         RB_INIT(&hmp->rb_nods_root);
 343         RB_INIT(&hmp->rb_undo_root);
 344         RB_INIT(&hmp->rb_resv_root);
 345         RB_INIT(&hmp->rb_bufs_root);
 346
 347         hmp->ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
 348
 349         TAILQ_INIT(&hmp->volu_list);
 350         TAILQ_INIT(&hmp->undo_list);
 351         TAILQ_INIT(&hmp->data_list);
 352         TAILQ_INIT(&hmp->meta_list);
 353         TAILQ_INIT(&hmp->lose_list);
 354
 355         /*
 356          * Load volumes
 357          */
 358         path = objcache_get(namei_oc, M_WAITOK);
 359         hmp->nvolumes = info.nvolumes;
 360         for (i = 0; i < info.nvolumes; ++i) {
 361                 error = copyin(&info.volumes[i], &upath, sizeof(char *));
 362                 if (error == 0)
 363                         error = copyinstr(upath, path, MAXPATHLEN, NULL);
 364                 if (error == 0)
 365                         error = hammer_install_volume(hmp, path);
 366                 if (error)
 367                         break;
 368         }
 369         objcache_put(namei_oc, path);
 370
 371         /*
 372          * Make sure we found a root volume
 373          */
 374         if (error == 0 && hmp->rootvol == NULL) {
 375                 kprintf("hammer_mount: No root volume found!\n");
 376                 error = EINVAL;
 377         }
 378         if (error) {
 379                 hammer_free_hmp(mp);
 380                 return (error);
 381         }
 382
 383         /*
 384          * No errors, setup enough of the mount point so we can lookup the
 385          * root vnode.
 386          */
 387         mp->mnt_iosize_max = MAXPHYS;
 388         mp->mnt_kern_flag |= MNTK_FSMID;
 389
 390         /*
 391          * note: f_iosize is used by vnode_pager_haspage() when constructing
 392          * its VOP_BMAP call.
 393          */
 394         mp->mnt_stat.f_iosize = HAMMER_BUFSIZE;
 395         mp->mnt_stat.f_bsize = HAMMER_BUFSIZE;
 396
 397         mp->mnt_vstat.f_frsize = HAMMER_BUFSIZE;
 398         mp->mnt_vstat.f_bsize = HAMMER_BUFSIZE;
 399
 400         mp->mnt_maxsymlinklen = 255;
 401         mp->mnt_flag |= MNT_LOCAL;
 402
 403         vfs_add_vnodeops(mp, &hammer_vnode_vops, &mp->mnt_vn_norm_ops);
 404         vfs_add_vnodeops(mp, &hammer_spec_vops, &mp->mnt_vn_spec_ops);
 405         vfs_add_vnodeops(mp, &hammer_fifo_vops, &mp->mnt_vn_fifo_ops);
 406
 407         /*
 408          * The root volume's ondisk pointer is only valid if we hold a
 409          * reference to it.
 410          */
 411         rootvol = hammer_get_root_volume(hmp, &error);
 412         if (error)
 413                 goto failed;
 414
 415         /*
 416          * Perform any necessary UNDO operations.  The recovery code does
 417          * call hammer_undo_lookup() so we have to pre-cache the blockmap,
 418          * and then re-copy it again after recovery is complete.
 419          *
 420          * If this is a read-only mount the UNDO information is retained
 421          * in memory in the form of dirty buffer cache buffers, and not
 422          * written back to the media.
 423          */
 424         bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
 425               sizeof(hmp->blockmap));
 426
 427         error = hammer_recover(hmp, rootvol);
 428         if (error) {
 429                 kprintf("Failed to recover HAMMER filesystem on mount\n");
 430                 goto done;
 431         }
 432
 433         /*
 434          * Finish setup now that we have a good root volume
 435          */
 436         ksnprintf(mp->mnt_stat.f_mntfromname,
 437                   sizeof(mp->mnt_stat.f_mntfromname), "%s",
 438                   rootvol->ondisk->vol_name);
 439         mp->mnt_stat.f_fsid.val[0] =
 440                 crc32((char *)&rootvol->ondisk->vol_fsid + 0, 8);
 441         mp->mnt_stat.f_fsid.val[1] =
 442                 crc32((char *)&rootvol->ondisk->vol_fsid + 8, 8);
 443
 444         mp->mnt_vstat.f_fsid_uuid = rootvol->ondisk->vol_fsid;
 445         mp->mnt_vstat.f_fsid = crc32(&mp->mnt_vstat.f_fsid_uuid,
 446                                      sizeof(mp->mnt_vstat.f_fsid_uuid));
 447
 448         /*
 449          * Certain often-modified fields in the root volume are cached in
 450          * the hammer_mount structure so we do not have to generate lots
 451          * of little UNDO structures for them.
 452          *
 453          * Recopy after recovery.  This also has the side effect of
 454          * setting our cached undo FIFO's first_offset, which serves to
 455          * placemark the FIFO start for the NEXT flush cycle while the
 456          * on-disk first_offset represents the LAST flush cycle.
 457          */
 458         hmp->next_tid = rootvol->ondisk->vol0_next_tid;
 459         bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
 460               sizeof(hmp->blockmap));
 461         hmp->copy_stat_freebigblocks = rootvol->ondisk->vol0_stat_freebigblocks;
 462
 463         hammer_flusher_create(hmp);
 464
 465         /*
 466          * Locate the root directory using the root cluster's B-Tree as a
 467          * starting point.  The root directory uses an obj_id of 1.
 468          *
 469          * FUTURE: Leave the root directory cached referenced but unlocked
 470          * in hmp->rootvp (need to flush it on unmount).
 471          */
 472         error = hammer_vfs_vget(mp, 1, &rootvp);
 473         if (error)
 474                 goto done;
 475         vput(rootvp);
 476         /*vn_unlock(hmp->rootvp);*/
 477
 478 done:
 479         hammer_rel_volume(rootvol, 0);
 480 failed:
 481         /*
 482          * Cleanup and return.
 483          */
 484         if (error)
 485                 hammer_free_hmp(mp);
 486         return (error);
 487 }
 488
 489 static int
 490 hammer_vfs_unmount(struct mount *mp, int mntflags)
 491 {
 492 #if 0
 493         struct hammer_mount *hmp = (void *)mp->mnt_data;
 494 #endif
 495         int flags;
 496         int error;
 497
 498         /*
 499          * Clean out the vnodes
 500          */
 501         flags = 0;
 502         if (mntflags & MNT_FORCE)
 503                 flags |= FORCECLOSE;
 504         if ((error = vflush(mp, 0, flags)) != 0)
 505                 return (error);
 506
 507         /*
 508          * Clean up the internal mount structure and related entities.  This
 509          * may issue I/O.
 510          */
 511         hammer_free_hmp(mp);
 512         return(0);
 513 }
 514
 515 /*
 516  * Clean up the internal mount structure and disassociate it from the mount.
 517  * This may issue I/O.
 518  */
 519 static void
 520 hammer_free_hmp(struct mount *mp)
 521 {
 522         struct hammer_mount *hmp = (void *)mp->mnt_data;
 523
 524 #if 0
 525         /*
 526          * Clean up the root vnode
 527          */
 528         if (hmp->rootvp) {
 529                 vrele(hmp->rootvp);
 530                 hmp->rootvp = NULL;
 531         }
 532 #endif
 533         hammer_flusher_sync(hmp);
 534         hammer_flusher_sync(hmp);
 535         hammer_flusher_destroy(hmp);
 536
 537         KKASSERT(RB_EMPTY(&hmp->rb_inos_root));
 538
 539 #if 0
 540         /*
 541          * Unload & flush inodes
 542          *
 543          * XXX illegal to call this from here, it can only be done from
 544          * the flusher.
 545          */
 546         RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
 547                 hammer_unload_inode, (void *)MNT_WAIT);
 548
 549         /*
 550          * Unload & flush volumes
 551          */
 552 #endif
 553         /*
 554          * Unload buffers and then volumes
 555          */
 556         RB_SCAN(hammer_buf_rb_tree, &hmp->rb_bufs_root, NULL,
 557                 hammer_unload_buffer, NULL);
 558         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
 559                 hammer_unload_volume, NULL);
 560
 561         mp->mnt_data = NULL;
 562         mp->mnt_flag &= ~MNT_LOCAL;
 563         hmp->mp = NULL;
 564         hammer_destroy_objid_cache(hmp);
 565         kfree(hmp, M_HAMMER);
 566 }
 567
 568 /*
 569  * Obtain a vnode for the specified inode number.  An exclusively locked
 570  * vnode is returned.
 571  */
 572 int
 573 hammer_vfs_vget(struct mount *mp, ino_t ino, struct vnode **vpp)
 574 {
 575         struct hammer_transaction trans;
 576         struct hammer_mount *hmp = (void *)mp->mnt_data;
 577         struct hammer_inode *ip;
 578         int error;
 579
 580         hammer_simple_transaction(&trans, hmp);
 581
 582         /*
 583          * Lookup the requested HAMMER inode.  The structure must be
 584          * left unlocked while we manipulate the related vnode to avoid
 585          * a deadlock.
 586          */
 587         ip = hammer_get_inode(&trans, NULL, ino, hmp->asof, 0, &error);
 588         if (ip == NULL) {
 589                 *vpp = NULL;
 590                 return(error);
 591         }
 592         error = hammer_get_vnode(ip, vpp);
 593         hammer_rel_inode(ip, 0);
 594         hammer_done_transaction(&trans);
 595         return (error);
 596 }
 597
 598 /*
 599  * Return the root vnode for the filesystem.
 600  *
 601  * HAMMER stores the root vnode in the hammer_mount structure so
 602  * getting it is easy.
 603  */
 604 static int
 605 hammer_vfs_root(struct mount *mp, struct vnode **vpp)
 606 {
 607 #if 0
 608         struct hammer_mount *hmp = (void *)mp->mnt_data;
 609 #endif
 610         int error;
 611
 612         error = hammer_vfs_vget(mp, 1, vpp);
 613         return (error);
 614 }
 615
 616 static int
 617 hammer_vfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
 618 {
 619         struct hammer_mount *hmp = (void *)mp->mnt_data;
 620         hammer_volume_t volume;
 621         hammer_volume_ondisk_t ondisk;
 622         int error;
 623         int64_t bfree;
 624
 625         volume = hammer_get_root_volume(hmp, &error);
 626         if (error)
 627                 return(error);
 628         ondisk = volume->ondisk;
 629
 630         /*
 631          * Basic stats
 632          */
 633         mp->mnt_stat.f_files = ondisk->vol0_stat_inodes;
 634         bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE;
 635         hammer_rel_volume(volume, 0);
 636
 637         mp->mnt_stat.f_bfree = bfree / HAMMER_BUFSIZE;
 638         mp->mnt_stat.f_bavail = mp->mnt_stat.f_bfree;
 639         if (mp->mnt_stat.f_files < 0)
 640                 mp->mnt_stat.f_files = 0;
 641
 642         *sbp = mp->mnt_stat;
 643         return(0);
 644 }
 645
 646 static int
 647 hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred)
 648 {
 649         struct hammer_mount *hmp = (void *)mp->mnt_data;
 650         hammer_volume_t volume;
 651         hammer_volume_ondisk_t ondisk;
 652         int error;
 653         int64_t bfree;
 654
 655         volume = hammer_get_root_volume(hmp, &error);
 656         if (error)
 657                 return(error);
 658         ondisk = volume->ondisk;
 659
 660         /*
 661          * Basic stats
 662          */
 663         mp->mnt_vstat.f_files = ondisk->vol0_stat_inodes;
 664         bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE;
 665         hammer_rel_volume(volume, 0);
 666
 667         mp->mnt_vstat.f_bfree = bfree / HAMMER_BUFSIZE;
 668         mp->mnt_vstat.f_bavail = mp->mnt_stat.f_bfree;
 669         if (mp->mnt_vstat.f_files < 0)
 670                 mp->mnt_vstat.f_files = 0;
 671         *sbp = mp->mnt_vstat;
 672         return(0);
 673 }
 674
 675 /*
 676  * Sync the filesystem.  Currently we have to run it twice, the second
 677  * one will advance the undo start index to the end index, so if a crash
 678  * occurs no undos will be run on mount.
 679  *
 680  * We do not sync the filesystem if we are called from a panic.  If we did
 681  * we might end up blowing up a sync that was already in progress.
 682  */
 683 static int
 684 hammer_vfs_sync(struct mount *mp, int waitfor)
 685 {
 686         struct hammer_mount *hmp = (void *)mp->mnt_data;
 687         int error;
 688
 689         if (panicstr == NULL) {
 690                 error = hammer_sync_hmp(hmp, waitfor);
 691                 if (error == 0)
 692                         error = hammer_sync_hmp(hmp, waitfor);
 693         } else {
 694                 error = EIO;
 695         }
 696         return (error);
 697 }
 698
 699 /*
 700  * Convert a vnode to a file handle.
 701  */
 702 static int
 703 hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp)
 704 {
 705         hammer_inode_t ip;
 706
 707         KKASSERT(MAXFIDSZ >= 16);
 708         ip = VTOI(vp);
 709         fhp->fid_len = offsetof(struct fid, fid_data[16]);
 710         fhp->fid_reserved = 0;
 711         bcopy(&ip->obj_id, fhp->fid_data + 0, sizeof(ip->obj_id));
 712         bcopy(&ip->obj_asof, fhp->fid_data + 8, sizeof(ip->obj_asof));
 713         return(0);
 714 }
 715
 716
 717 /*
 718  * Convert a file handle back to a vnode.
 719  */
 720 static int
 721 hammer_vfs_fhtovp(struct mount *mp, struct fid *fhp, struct vnode **vpp)
 722 {
 723         struct hammer_transaction trans;
 724         struct hammer_inode *ip;
 725         struct hammer_inode_info info;
 726         int error;
 727
 728         bcopy(fhp->fid_data + 0, &info.obj_id, sizeof(info.obj_id));
 729         bcopy(fhp->fid_data + 8, &info.obj_asof, sizeof(info.obj_asof));
 730
 731         hammer_simple_transaction(&trans, (void *)mp->mnt_data);
 732
 733         /*
 734          * Get/allocate the hammer_inode structure.  The structure must be
 735          * unlocked while we manipulate the related vnode to avoid a
 736          * deadlock.
 737          */
 738         ip = hammer_get_inode(&trans, NULL, info.obj_id, info.obj_asof,
 739                               0, &error);
 740         if (ip == NULL) {
 741                 *vpp = NULL;
 742                 return(error);
 743         }
 744         error = hammer_get_vnode(ip, vpp);
 745         hammer_rel_inode(ip, 0);
 746         hammer_done_transaction(&trans);
 747         return (error);
 748 }
 749
 750 static int
 751 hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
 752                     int *exflagsp, struct ucred **credanonp)
 753 {
 754         hammer_mount_t hmp = (void *)mp->mnt_data;
 755         struct netcred *np;
 756         int error;
 757
 758         np = vfs_export_lookup(mp, &hmp->export, nam);
 759         if (np) {
 760                 *exflagsp = np->netc_exflags;
 761                 *credanonp = &np->netc_anon;
 762                 error = 0;
 763         } else {
 764                 error = EACCES;
 765         }
 766         return (error);
 767
 768 }
 769
 770 int
 771 hammer_vfs_export(struct mount *mp, int op, const struct export_args *export)
 772 {
 773         hammer_mount_t hmp = (void *)mp->mnt_data;
 774         int error;
 775
 776         switch(op) {
 777         case MOUNTCTL_SET_EXPORT:
 778                 error = vfs_export(mp, &hmp->export, export);
 779                 break;
 780         default:
 781                 error = EOPNOTSUPP;
 782                 break;
 783         }
 784         return(error);
 785 }
 786