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 1.58 2008/07/02 21:57:54 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_dirtybufspace;         /* 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_dirtybufspace;         /* 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_dirtybufspace, CTLFLAG_RW,
 114            &hammer_limit_dirtybufspace, 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_dirtybufspace, CTLFLAG_RD,
 153            &hammer_count_dirtybufspace, 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         int n;
 222
 223         if (hammer_limit_recs == 0) {
 224                 hammer_limit_recs = nbuf * 25;
 225                 n = kmalloc_limit(M_HAMMER) / 512;
 226                 if (hammer_limit_recs > n)
 227                         hammer_limit_recs = n;
 228         }
 229         if (hammer_limit_dirtybufspace == 0) {
 230                 hammer_limit_dirtybufspace = hidirtybufspace / 2;
 231                 if (hammer_limit_dirtybufspace < 100)
 232                         hammer_limit_dirtybufspace = 100;
 233         }
 234         if (hammer_limit_iqueued == 0)
 235                 hammer_limit_iqueued = desiredvnodes / 5;
 236         return(0);
 237 }
 238
 239 static int
 240 hammer_vfs_mount(struct mount *mp, char *mntpt, caddr_t data,
 241                  struct ucred *cred)
 242 {
 243         struct hammer_mount_info info;
 244         hammer_mount_t hmp;
 245         hammer_volume_t rootvol;
 246         struct vnode *rootvp;
 247         const char *upath;      /* volume name in userspace */
 248         char *path;             /* volume name in system space */
 249         int error;
 250         int i;
 251
 252         if ((error = copyin(data, &info, sizeof(info))) != 0)
 253                 return (error);
 254         if ((mp->mnt_flag & MNT_UPDATE) == 0) {
 255                 if (info.nvolumes <= 0 || info.nvolumes >= 32768)
 256                         return (EINVAL);
 257         }
 258         if ((info.hflags & HMNT_MASTERID) &&
 259             (info.masterid < -1 || info.masterid >= HAMMER_MAX_MASTERS)) {
 260                         return (EINVAL);
 261         }
 262
 263         /*
 264          * Interal mount data structure
 265          */
 266         if (mp->mnt_flag & MNT_UPDATE) {
 267                 hmp = (void *)mp->mnt_data;
 268                 KKASSERT(hmp != NULL);
 269         } else {
 270                 hmp = kmalloc(sizeof(*hmp), M_HAMMER, M_WAITOK | M_ZERO);
 271                 mp->mnt_data = (qaddr_t)hmp;
 272                 hmp->mp = mp;
 273                 hmp->namekey_iterator = mycpu->gd_time_seconds;
 274                 /*TAILQ_INIT(&hmp->recycle_list);*/
 275
 276                 hmp->root_btree_beg.localization = 0x00000000U;
 277                 hmp->root_btree_beg.obj_id = -0x8000000000000000LL;
 278                 hmp->root_btree_beg.key = -0x8000000000000000LL;
 279                 hmp->root_btree_beg.create_tid = 1;
 280                 hmp->root_btree_beg.delete_tid = 1;
 281                 hmp->root_btree_beg.rec_type = 0;
 282                 hmp->root_btree_beg.obj_type = 0;
 283
 284                 hmp->root_btree_end.localization = 0xFFFFFFFFU;
 285                 hmp->root_btree_end.obj_id = 0x7FFFFFFFFFFFFFFFLL;
 286                 hmp->root_btree_end.key = 0x7FFFFFFFFFFFFFFFLL;
 287                 hmp->root_btree_end.create_tid = 0xFFFFFFFFFFFFFFFFULL;
 288                 hmp->root_btree_end.delete_tid = 0;   /* special case */
 289                 hmp->root_btree_end.rec_type = 0xFFFFU;
 290                 hmp->root_btree_end.obj_type = 0;
 291
 292                 hmp->sync_lock.refs = 1;
 293                 hmp->free_lock.refs = 1;
 294                 hmp->undo_lock.refs = 1;
 295                 hmp->blkmap_lock.refs = 1;
 296
 297                 TAILQ_INIT(&hmp->flush_list);
 298                 TAILQ_INIT(&hmp->delay_list);
 299                 TAILQ_INIT(&hmp->objid_cache_list);
 300                 TAILQ_INIT(&hmp->undo_lru_list);
 301                 TAILQ_INIT(&hmp->reclaim_list);
 302         }
 303         hmp->hflags &= ~HMNT_USERFLAGS;
 304         hmp->hflags |= info.hflags & HMNT_USERFLAGS;
 305         if (info.hflags & HMNT_MASTERID)
 306                 hmp->masterid = info.masterid;
 307         else
 308                 hmp->masterid = -1;
 309         if (info.asof) {
 310                 kprintf("ASOF\n");
 311                 mp->mnt_flag |= MNT_RDONLY;
 312                 hmp->asof = info.asof;
 313         } else {
 314                 hmp->asof = HAMMER_MAX_TID;
 315         }
 316
 317         /*
 318          * Re-open read-write if originally read-only, or vise-versa.
 319          */
 320         if (mp->mnt_flag & MNT_UPDATE) {
 321                 error = 0;
 322                 if (hmp->ronly && (mp->mnt_kern_flag & MNTK_WANTRDWR)) {
 323                         kprintf("HAMMER read-only -> read-write\n");
 324                         hmp->ronly = 0;
 325                         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
 326                                 hammer_adjust_volume_mode, NULL);
 327                         rootvol = hammer_get_root_volume(hmp, &error);
 328                         if (rootvol) {
 329                                 hammer_recover_flush_buffers(hmp, rootvol, 1);
 330                                 bcopy(rootvol->ondisk->vol0_blockmap,
 331                                       hmp->blockmap,
 332                                       sizeof(hmp->blockmap));
 333                                 hammer_rel_volume(rootvol, 0);
 334                         }
 335                         RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
 336                                 hammer_reload_inode, NULL);
 337                         /* kernel clears MNT_RDONLY */
 338                 } else if (hmp->ronly == 0 && (mp->mnt_flag & MNT_RDONLY)) {
 339                         kprintf("HAMMER read-write -> read-only\n");
 340                         hmp->ronly = 1; /* messy */
 341                         RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
 342                                 hammer_reload_inode, NULL);
 343                         hmp->ronly = 0;
 344                         hammer_flusher_sync(hmp);
 345                         hammer_flusher_sync(hmp);
 346                         hammer_flusher_sync(hmp);
 347                         hmp->ronly = 1;
 348                         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
 349                                 hammer_adjust_volume_mode, NULL);
 350                 }
 351                 return(error);
 352         }
 353
 354         RB_INIT(&hmp->rb_vols_root);
 355         RB_INIT(&hmp->rb_inos_root);
 356         RB_INIT(&hmp->rb_nods_root);
 357         RB_INIT(&hmp->rb_undo_root);
 358         RB_INIT(&hmp->rb_resv_root);
 359         RB_INIT(&hmp->rb_bufs_root);
 360         RB_INIT(&hmp->rb_pfsm_root);
 361
 362         hmp->ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
 363
 364         TAILQ_INIT(&hmp->volu_list);
 365         TAILQ_INIT(&hmp->undo_list);
 366         TAILQ_INIT(&hmp->data_list);
 367         TAILQ_INIT(&hmp->meta_list);
 368         TAILQ_INIT(&hmp->lose_list);
 369
 370         /*
 371          * Load volumes
 372          */
 373         path = objcache_get(namei_oc, M_WAITOK);
 374         hmp->nvolumes = info.nvolumes;
 375         for (i = 0; i < info.nvolumes; ++i) {
 376                 error = copyin(&info.volumes[i], &upath, sizeof(char *));
 377                 if (error == 0)
 378                         error = copyinstr(upath, path, MAXPATHLEN, NULL);
 379                 if (error == 0)
 380                         error = hammer_install_volume(hmp, path);
 381                 if (error)
 382                         break;
 383         }
 384         objcache_put(namei_oc, path);
 385
 386         /*
 387          * Make sure we found a root volume
 388          */
 389         if (error == 0 && hmp->rootvol == NULL) {
 390                 kprintf("hammer_mount: No root volume found!\n");
 391                 error = EINVAL;
 392         }
 393         if (error) {
 394                 hammer_free_hmp(mp);
 395                 return (error);
 396         }
 397
 398         /*
 399          * No errors, setup enough of the mount point so we can lookup the
 400          * root vnode.
 401          */
 402         mp->mnt_iosize_max = MAXPHYS;
 403         mp->mnt_kern_flag |= MNTK_FSMID;
 404
 405         /*
 406          * note: f_iosize is used by vnode_pager_haspage() when constructing
 407          * its VOP_BMAP call.
 408          */
 409         mp->mnt_stat.f_iosize = HAMMER_BUFSIZE;
 410         mp->mnt_stat.f_bsize = HAMMER_BUFSIZE;
 411
 412         mp->mnt_vstat.f_frsize = HAMMER_BUFSIZE;
 413         mp->mnt_vstat.f_bsize = HAMMER_BUFSIZE;
 414
 415         mp->mnt_maxsymlinklen = 255;
 416         mp->mnt_flag |= MNT_LOCAL;
 417
 418         vfs_add_vnodeops(mp, &hammer_vnode_vops, &mp->mnt_vn_norm_ops);
 419         vfs_add_vnodeops(mp, &hammer_spec_vops, &mp->mnt_vn_spec_ops);
 420         vfs_add_vnodeops(mp, &hammer_fifo_vops, &mp->mnt_vn_fifo_ops);
 421
 422         /*
 423          * The root volume's ondisk pointer is only valid if we hold a
 424          * reference to it.
 425          */
 426         rootvol = hammer_get_root_volume(hmp, &error);
 427         if (error)
 428                 goto failed;
 429
 430         /*
 431          * Perform any necessary UNDO operations.  The recovery code does
 432          * call hammer_undo_lookup() so we have to pre-cache the blockmap,
 433          * and then re-copy it again after recovery is complete.
 434          *
 435          * If this is a read-only mount the UNDO information is retained
 436          * in memory in the form of dirty buffer cache buffers, and not
 437          * written back to the media.
 438          */
 439         bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
 440               sizeof(hmp->blockmap));
 441
 442         error = hammer_recover(hmp, rootvol);
 443         if (error) {
 444                 kprintf("Failed to recover HAMMER filesystem on mount\n");
 445                 goto done;
 446         }
 447
 448         /*
 449          * Finish setup now that we have a good root volume
 450          */
 451         ksnprintf(mp->mnt_stat.f_mntfromname,
 452                   sizeof(mp->mnt_stat.f_mntfromname), "%s",
 453                   rootvol->ondisk->vol_name);
 454         mp->mnt_stat.f_fsid.val[0] =
 455                 crc32((char *)&rootvol->ondisk->vol_fsid + 0, 8);
 456         mp->mnt_stat.f_fsid.val[1] =
 457                 crc32((char *)&rootvol->ondisk->vol_fsid + 8, 8);
 458
 459         mp->mnt_vstat.f_fsid_uuid = rootvol->ondisk->vol_fsid;
 460         mp->mnt_vstat.f_fsid = crc32(&mp->mnt_vstat.f_fsid_uuid,
 461                                      sizeof(mp->mnt_vstat.f_fsid_uuid));
 462
 463         /*
 464          * Certain often-modified fields in the root volume are cached in
 465          * the hammer_mount structure so we do not have to generate lots
 466          * of little UNDO structures for them.
 467          *
 468          * Recopy after recovery.  This also has the side effect of
 469          * setting our cached undo FIFO's first_offset, which serves to
 470          * placemark the FIFO start for the NEXT flush cycle while the
 471          * on-disk first_offset represents the LAST flush cycle.
 472          */
 473         hmp->next_tid = rootvol->ondisk->vol0_next_tid;
 474         bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
 475               sizeof(hmp->blockmap));
 476         hmp->copy_stat_freebigblocks = rootvol->ondisk->vol0_stat_freebigblocks;
 477
 478         hammer_flusher_create(hmp);
 479
 480         /*
 481          * Locate the root directory using the root cluster's B-Tree as a
 482          * starting point.  The root directory uses an obj_id of 1.
 483          *
 484          * FUTURE: Leave the root directory cached referenced but unlocked
 485          * in hmp->rootvp (need to flush it on unmount).
 486          */
 487         error = hammer_vfs_vget(mp, 1, &rootvp);
 488         if (error)
 489                 goto done;
 490         vput(rootvp);
 491         /*vn_unlock(hmp->rootvp);*/
 492
 493 done:
 494         hammer_rel_volume(rootvol, 0);
 495 failed:
 496         /*
 497          * Cleanup and return.
 498          */
 499         if (error)
 500                 hammer_free_hmp(mp);
 501         return (error);
 502 }
 503
 504 static int
 505 hammer_vfs_unmount(struct mount *mp, int mntflags)
 506 {
 507 #if 0
 508         struct hammer_mount *hmp = (void *)mp->mnt_data;
 509 #endif
 510         int flags;
 511         int error;
 512
 513         /*
 514          * Clean out the vnodes
 515          */
 516         flags = 0;
 517         if (mntflags & MNT_FORCE)
 518                 flags |= FORCECLOSE;
 519         if ((error = vflush(mp, 0, flags)) != 0)
 520                 return (error);
 521
 522         /*
 523          * Clean up the internal mount structure and related entities.  This
 524          * may issue I/O.
 525          */
 526         hammer_free_hmp(mp);
 527         return(0);
 528 }
 529
 530 /*
 531  * Clean up the internal mount structure and disassociate it from the mount.
 532  * This may issue I/O.
 533  */
 534 static void
 535 hammer_free_hmp(struct mount *mp)
 536 {
 537         struct hammer_mount *hmp = (void *)mp->mnt_data;
 538
 539 #if 0
 540         /*
 541          * Clean up the root vnode
 542          */
 543         if (hmp->rootvp) {
 544                 vrele(hmp->rootvp);
 545                 hmp->rootvp = NULL;
 546         }
 547 #endif
 548         hammer_flusher_sync(hmp);
 549         hammer_flusher_sync(hmp);
 550         hammer_flusher_destroy(hmp);
 551
 552         KKASSERT(RB_EMPTY(&hmp->rb_inos_root));
 553
 554 #if 0
 555         /*
 556          * Unload & flush inodes
 557          *
 558          * XXX illegal to call this from here, it can only be done from
 559          * the flusher.
 560          */
 561         RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
 562                 hammer_unload_inode, (void *)MNT_WAIT);
 563
 564         /*
 565          * Unload & flush volumes
 566          */
 567 #endif
 568         /*
 569          * Unload buffers and then volumes
 570          */
 571         RB_SCAN(hammer_buf_rb_tree, &hmp->rb_bufs_root, NULL,
 572                 hammer_unload_buffer, NULL);
 573         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
 574                 hammer_unload_volume, NULL);
 575
 576         mp->mnt_data = NULL;
 577         mp->mnt_flag &= ~MNT_LOCAL;
 578         hmp->mp = NULL;
 579         hammer_destroy_objid_cache(hmp);
 580         kfree(hmp, M_HAMMER);
 581 }
 582
 583 /*
 584  * Obtain a vnode for the specified inode number.  An exclusively locked
 585  * vnode is returned.
 586  */
 587 int
 588 hammer_vfs_vget(struct mount *mp, ino_t ino, struct vnode **vpp)
 589 {
 590         struct hammer_transaction trans;
 591         struct hammer_mount *hmp = (void *)mp->mnt_data;
 592         struct hammer_inode *ip;
 593         int error;
 594
 595         hammer_simple_transaction(&trans, hmp);
 596
 597         /*
 598          * Lookup the requested HAMMER inode.  The structure must be
 599          * left unlocked while we manipulate the related vnode to avoid
 600          * a deadlock.
 601          */
 602         ip = hammer_get_inode(&trans, NULL, ino,
 603                               hmp->asof, HAMMER_DEF_LOCALIZATION,
 604                               0, &error);
 605         if (ip == NULL) {
 606                 *vpp = NULL;
 607                 return(error);
 608         }
 609         error = hammer_get_vnode(ip, vpp);
 610         hammer_rel_inode(ip, 0);
 611         hammer_done_transaction(&trans);
 612         return (error);
 613 }
 614
 615 /*
 616  * Return the root vnode for the filesystem.
 617  *
 618  * HAMMER stores the root vnode in the hammer_mount structure so
 619  * getting it is easy.
 620  */
 621 static int
 622 hammer_vfs_root(struct mount *mp, struct vnode **vpp)
 623 {
 624 #if 0
 625         struct hammer_mount *hmp = (void *)mp->mnt_data;
 626 #endif
 627         int error;
 628
 629         error = hammer_vfs_vget(mp, 1, vpp);
 630         return (error);
 631 }
 632
 633 static int
 634 hammer_vfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
 635 {
 636         struct hammer_mount *hmp = (void *)mp->mnt_data;
 637         hammer_volume_t volume;
 638         hammer_volume_ondisk_t ondisk;
 639         int error;
 640         int64_t bfree;
 641
 642         volume = hammer_get_root_volume(hmp, &error);
 643         if (error)
 644                 return(error);
 645         ondisk = volume->ondisk;
 646
 647         /*
 648          * Basic stats
 649          */
 650         mp->mnt_stat.f_files = ondisk->vol0_stat_inodes;
 651         bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE;
 652         hammer_rel_volume(volume, 0);
 653
 654         mp->mnt_stat.f_bfree = bfree / HAMMER_BUFSIZE;
 655         mp->mnt_stat.f_bavail = mp->mnt_stat.f_bfree;
 656         if (mp->mnt_stat.f_files < 0)
 657                 mp->mnt_stat.f_files = 0;
 658
 659         *sbp = mp->mnt_stat;
 660         return(0);
 661 }
 662
 663 static int
 664 hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred)
 665 {
 666         struct hammer_mount *hmp = (void *)mp->mnt_data;
 667         hammer_volume_t volume;
 668         hammer_volume_ondisk_t ondisk;
 669         int error;
 670         int64_t bfree;
 671
 672         volume = hammer_get_root_volume(hmp, &error);
 673         if (error)
 674                 return(error);
 675         ondisk = volume->ondisk;
 676
 677         /*
 678          * Basic stats
 679          */
 680         mp->mnt_vstat.f_files = ondisk->vol0_stat_inodes;
 681         bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE;
 682         hammer_rel_volume(volume, 0);
 683
 684         mp->mnt_vstat.f_bfree = bfree / HAMMER_BUFSIZE;
 685         mp->mnt_vstat.f_bavail = mp->mnt_stat.f_bfree;
 686         if (mp->mnt_vstat.f_files < 0)
 687                 mp->mnt_vstat.f_files = 0;
 688         *sbp = mp->mnt_vstat;
 689         return(0);
 690 }
 691
 692 /*
 693  * Sync the filesystem.  Currently we have to run it twice, the second
 694  * one will advance the undo start index to the end index, so if a crash
 695  * occurs no undos will be run on mount.
 696  *
 697  * We do not sync the filesystem if we are called from a panic.  If we did
 698  * we might end up blowing up a sync that was already in progress.
 699  */
 700 static int
 701 hammer_vfs_sync(struct mount *mp, int waitfor)
 702 {
 703         struct hammer_mount *hmp = (void *)mp->mnt_data;
 704         int error;
 705
 706         if (panicstr == NULL) {
 707                 error = hammer_sync_hmp(hmp, waitfor);
 708                 if (error == 0)
 709                         error = hammer_sync_hmp(hmp, waitfor);
 710         } else {
 711                 error = EIO;
 712         }
 713         return (error);
 714 }
 715
 716 /*
 717  * Convert a vnode to a file handle.
 718  */
 719 static int
 720 hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp)
 721 {
 722         hammer_inode_t ip;
 723
 724         KKASSERT(MAXFIDSZ >= 16);
 725         ip = VTOI(vp);
 726         fhp->fid_len = offsetof(struct fid, fid_data[16]);
 727         fhp->fid_reserved = 0;
 728         bcopy(&ip->obj_id, fhp->fid_data + 0, sizeof(ip->obj_id));
 729         bcopy(&ip->obj_asof, fhp->fid_data + 8, sizeof(ip->obj_asof));
 730         return(0);
 731 }
 732
 733
 734 /*
 735  * Convert a file handle back to a vnode.
 736  */
 737 static int
 738 hammer_vfs_fhtovp(struct mount *mp, struct fid *fhp, struct vnode **vpp)
 739 {
 740         struct hammer_transaction trans;
 741         struct hammer_inode *ip;
 742         struct hammer_inode_info info;
 743         int error;
 744
 745         bcopy(fhp->fid_data + 0, &info.obj_id, sizeof(info.obj_id));
 746         bcopy(fhp->fid_data + 8, &info.obj_asof, sizeof(info.obj_asof));
 747
 748         hammer_simple_transaction(&trans, (void *)mp->mnt_data);
 749
 750         /*
 751          * Get/allocate the hammer_inode structure.  The structure must be
 752          * unlocked while we manipulate the related vnode to avoid a
 753          * deadlock.
 754          */
 755         ip = hammer_get_inode(&trans, NULL, info.obj_id,
 756                               info.obj_asof, HAMMER_DEF_LOCALIZATION,
 757                               0, &error);
 758         if (ip == NULL) {
 759                 *vpp = NULL;
 760                 return(error);
 761         }
 762         error = hammer_get_vnode(ip, vpp);
 763         hammer_rel_inode(ip, 0);
 764         hammer_done_transaction(&trans);
 765         return (error);
 766 }
 767
 768 static int
 769 hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
 770                     int *exflagsp, struct ucred **credanonp)
 771 {
 772         hammer_mount_t hmp = (void *)mp->mnt_data;
 773         struct netcred *np;
 774         int error;
 775
 776         np = vfs_export_lookup(mp, &hmp->export, nam);
 777         if (np) {
 778                 *exflagsp = np->netc_exflags;
 779                 *credanonp = &np->netc_anon;
 780                 error = 0;
 781         } else {
 782                 error = EACCES;
 783         }
 784         return (error);
 785
 786 }
 787
 788 int
 789 hammer_vfs_export(struct mount *mp, int op, const struct export_args *export)
 790 {
 791         hammer_mount_t hmp = (void *)mp->mnt_data;
 792         int error;
 793
 794         switch(op) {
 795         case MOUNTCTL_SET_EXPORT:
 796                 error = vfs_export(mp, &hmp->export, export);
 797                 break;
 798         default:
 799                 error = EOPNOTSUPP;
 800                 break;
 801         }
 802         return(error);
 803 }
 804