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
  35 #include <sys/param.h>
  36 #include <sys/systm.h>
  37 #include <sys/kernel.h>
  38 #include <sys/vnode.h>
  39 #include <sys/mount.h>
  40 #include <sys/malloc.h>
  41 #include <sys/nlookup.h>
  42 #include <sys/fcntl.h>
  43 #include <sys/sysctl.h>
  44 #include <sys/buf.h>
  45 #include <sys/buf2.h>
  46 #include "hammer.h"
  47
  48 /*
  49  * NOTE!  Global statistics may not be MPSAFE so HAMMER never uses them
  50  *        in conditionals.
  51  */
  52 int hammer_supported_version = HAMMER_VOL_VERSION_DEFAULT;
  53 int hammer_debug_io;
  54 int hammer_debug_general;
  55 int hammer_debug_debug = 1;             /* medium-error panics */
  56 int hammer_debug_inode;
  57 int hammer_debug_locks;
  58 int hammer_debug_btree;
  59 int hammer_debug_tid;
  60 int hammer_debug_recover;               /* -1 will disable, +1 will force */
  61 int hammer_debug_recover_faults;
  62 int hammer_debug_critical;              /* non-zero enter debugger on error */
  63 int hammer_cluster_enable = 1;          /* enable read clustering by default */
  64 int hammer_live_dedup = 0;
  65 int hammer_tdmux_ticks;
  66 int hammer_count_fsyncs;
  67 int hammer_count_inodes;
  68 int hammer_count_iqueued;
  69 int hammer_count_reclaims;
  70 int hammer_count_records;
  71 int hammer_count_record_datas;
  72 int hammer_count_volumes;
  73 int hammer_count_buffers;
  74 int hammer_count_nodes;
  75 int64_t hammer_count_extra_space_used;
  76 int64_t hammer_stats_btree_lookups;
  77 int64_t hammer_stats_btree_searches;
  78 int64_t hammer_stats_btree_inserts;
  79 int64_t hammer_stats_btree_deletes;
  80 int64_t hammer_stats_btree_elements;
  81 int64_t hammer_stats_btree_splits;
  82 int64_t hammer_stats_btree_iterations;
  83 int64_t hammer_stats_btree_root_iterations;
  84 int64_t hammer_stats_record_iterations;
  85
  86 int64_t hammer_stats_file_read;
  87 int64_t hammer_stats_file_write;
  88 int64_t hammer_stats_file_iopsr;
  89 int64_t hammer_stats_file_iopsw;
  90 int64_t hammer_stats_disk_read;
  91 int64_t hammer_stats_disk_write;
  92 int64_t hammer_stats_inode_flushes;
  93 int64_t hammer_stats_commits;
  94 int64_t hammer_stats_undo;
  95 int64_t hammer_stats_redo;
  96
  97 long hammer_count_dirtybufspace;        /* global */
  98 int hammer_count_refedbufs;             /* global */
  99 int hammer_count_reservations;
 100 long hammer_count_io_running_read;
 101 long hammer_count_io_running_write;
 102 int hammer_count_io_locked;
 103 long hammer_limit_dirtybufspace;        /* per-mount */
 104 int hammer_limit_recs;                  /* as a whole XXX */
 105 int hammer_limit_inode_recs = 2048;     /* per inode */
 106 int hammer_limit_reclaims;
 107 int hammer_live_dedup_cache_size = DEDUP_CACHE_SIZE;
 108 int hammer_limit_redo = 4096 * 1024;    /* per inode */
 109 int hammer_autoflush = 500;             /* auto flush (typ on reclaim) */
 110 int hammer_bio_count;
 111 int hammer_verify_zone;
 112 int hammer_verify_data = 1;
 113 int hammer_write_mode;
 114 int hammer_double_buffer;
 115 int hammer_btree_full_undo = 1;
 116 int hammer_yield_check = 16;
 117 int hammer_fsync_mode = 3;
 118 int64_t hammer_contention_count;
 119 int64_t hammer_zone_limit;
 120
 121 /*
 122  * Live dedup debug counters (sysctls are writable so that counters
 123  * can be reset from userspace).
 124  */
 125 int64_t hammer_live_dedup_vnode_bcmps = 0;
 126 int64_t hammer_live_dedup_device_bcmps = 0;
 127 int64_t hammer_live_dedup_findblk_failures = 0;
 128 int64_t hammer_live_dedup_bmap_saves = 0;
 129
 130
 131 SYSCTL_NODE(_vfs, OID_AUTO, hammer, CTLFLAG_RW, 0, "HAMMER filesystem");
 132
 133 SYSCTL_INT(_vfs_hammer, OID_AUTO, supported_version, CTLFLAG_RD,
 134            &hammer_supported_version, 0, "");
 135 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_general, CTLFLAG_RW,
 136            &hammer_debug_general, 0, "");
 137 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_io, CTLFLAG_RW,
 138            &hammer_debug_io, 0, "");
 139 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_debug, CTLFLAG_RW,
 140            &hammer_debug_debug, 0, "");
 141 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_inode, CTLFLAG_RW,
 142            &hammer_debug_inode, 0, "");
 143 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_locks, CTLFLAG_RW,
 144            &hammer_debug_locks, 0, "");
 145 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_btree, CTLFLAG_RW,
 146            &hammer_debug_btree, 0, "");
 147 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_tid, CTLFLAG_RW,
 148            &hammer_debug_tid, 0, "");
 149 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover, CTLFLAG_RW,
 150            &hammer_debug_recover, 0, "");
 151 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover_faults, CTLFLAG_RW,
 152            &hammer_debug_recover_faults, 0, "");
 153 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_critical, CTLFLAG_RW,
 154            &hammer_debug_critical, 0, "");
 155 SYSCTL_INT(_vfs_hammer, OID_AUTO, cluster_enable, CTLFLAG_RW,
 156            &hammer_cluster_enable, 0, "");
 157 /*
 158  * 0 - live dedup is disabled
 159  * 1 - dedup cache is populated on reads only
 160  * 2 - dedup cache is populated on both reads and writes
 161  *
 162  * LIVE_DEDUP IS DISABLED PERMANENTLY!  This feature appears to cause
 163  * blockmap corruption over time so we've turned it off permanently.
 164  */
 165 SYSCTL_INT(_vfs_hammer, OID_AUTO, live_dedup, CTLFLAG_RD,
 166            &hammer_live_dedup, 0, "Enable live dedup (experimental)");
 167 SYSCTL_INT(_vfs_hammer, OID_AUTO, tdmux_ticks, CTLFLAG_RW,
 168            &hammer_tdmux_ticks, 0, "Hammer tdmux ticks");
 169
 170 SYSCTL_LONG(_vfs_hammer, OID_AUTO, limit_dirtybufspace, CTLFLAG_RW,
 171            &hammer_limit_dirtybufspace, 0, "");
 172 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_recs, CTLFLAG_RW,
 173            &hammer_limit_recs, 0, "");
 174 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_inode_recs, CTLFLAG_RW,
 175            &hammer_limit_inode_recs, 0, "");
 176 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_reclaims, CTLFLAG_RW,
 177            &hammer_limit_reclaims, 0, "");
 178 SYSCTL_INT(_vfs_hammer, OID_AUTO, live_dedup_cache_size, CTLFLAG_RW,
 179            &hammer_live_dedup_cache_size, 0,
 180            "Number of cache entries");
 181 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_redo, CTLFLAG_RW,
 182            &hammer_limit_redo, 0, "");
 183
 184 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_fsyncs, CTLFLAG_RD,
 185            &hammer_count_fsyncs, 0, "");
 186 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_inodes, CTLFLAG_RD,
 187            &hammer_count_inodes, 0, "");
 188 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_iqueued, CTLFLAG_RD,
 189            &hammer_count_iqueued, 0, "");
 190 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reclaims, CTLFLAG_RD,
 191            &hammer_count_reclaims, 0, "");
 192 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_records, CTLFLAG_RD,
 193            &hammer_count_records, 0, "");
 194 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_record_datas, CTLFLAG_RD,
 195            &hammer_count_record_datas, 0, "");
 196 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_volumes, CTLFLAG_RD,
 197            &hammer_count_volumes, 0, "");
 198 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_buffers, CTLFLAG_RD,
 199            &hammer_count_buffers, 0, "");
 200 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_nodes, CTLFLAG_RD,
 201            &hammer_count_nodes, 0, "");
 202 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, count_extra_space_used, CTLFLAG_RD,
 203            &hammer_count_extra_space_used, 0, "");
 204
 205 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_searches, CTLFLAG_RD,
 206            &hammer_stats_btree_searches, 0, "");
 207 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_lookups, CTLFLAG_RD,
 208            &hammer_stats_btree_lookups, 0, "");
 209 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_inserts, CTLFLAG_RD,
 210            &hammer_stats_btree_inserts, 0, "");
 211 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_deletes, CTLFLAG_RD,
 212            &hammer_stats_btree_deletes, 0, "");
 213 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_elements, CTLFLAG_RD,
 214            &hammer_stats_btree_elements, 0, "");
 215 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_splits, CTLFLAG_RD,
 216            &hammer_stats_btree_splits, 0, "");
 217 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_iterations, CTLFLAG_RD,
 218            &hammer_stats_btree_iterations, 0, "");
 219 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_root_iterations, CTLFLAG_RD,
 220            &hammer_stats_btree_root_iterations, 0, "");
 221 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_record_iterations, CTLFLAG_RD,
 222            &hammer_stats_record_iterations, 0, "");
 223
 224 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_read, CTLFLAG_RD,
 225            &hammer_stats_file_read, 0, "");
 226 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_write, CTLFLAG_RD,
 227            &hammer_stats_file_write, 0, "");
 228 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_iopsr, CTLFLAG_RD,
 229            &hammer_stats_file_iopsr, 0, "");
 230 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_iopsw, CTLFLAG_RD,
 231            &hammer_stats_file_iopsw, 0, "");
 232 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_disk_read, CTLFLAG_RD,
 233            &hammer_stats_disk_read, 0, "");
 234 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_disk_write, CTLFLAG_RD,
 235            &hammer_stats_disk_write, 0, "");
 236 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_inode_flushes, CTLFLAG_RD,
 237            &hammer_stats_inode_flushes, 0, "");
 238 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_commits, CTLFLAG_RD,
 239            &hammer_stats_commits, 0, "");
 240 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_undo, CTLFLAG_RD,
 241            &hammer_stats_undo, 0, "");
 242 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_redo, CTLFLAG_RD,
 243            &hammer_stats_redo, 0, "");
 244
 245 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, live_dedup_vnode_bcmps, CTLFLAG_RW,
 246             &hammer_live_dedup_vnode_bcmps, 0,
 247             "successful vnode buffer comparisons");
 248 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, live_dedup_device_bcmps, CTLFLAG_RW,
 249             &hammer_live_dedup_device_bcmps, 0,
 250             "successful device buffer comparisons");
 251 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, live_dedup_findblk_failures, CTLFLAG_RW,
 252             &hammer_live_dedup_findblk_failures, 0,
 253             "block lookup failures for comparison");
 254 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, live_dedup_bmap_saves, CTLFLAG_RW,
 255             &hammer_live_dedup_bmap_saves, 0,
 256             "useful physical block lookups");
 257
 258 SYSCTL_LONG(_vfs_hammer, OID_AUTO, count_dirtybufspace, CTLFLAG_RD,
 259            &hammer_count_dirtybufspace, 0, "");
 260 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_refedbufs, CTLFLAG_RD,
 261            &hammer_count_refedbufs, 0, "");
 262 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reservations, CTLFLAG_RD,
 263            &hammer_count_reservations, 0, "");
 264 SYSCTL_LONG(_vfs_hammer, OID_AUTO, count_io_running_read, CTLFLAG_RD,
 265            &hammer_count_io_running_read, 0, "");
 266 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_locked, CTLFLAG_RD,
 267            &hammer_count_io_locked, 0, "");
 268 SYSCTL_LONG(_vfs_hammer, OID_AUTO, count_io_running_write, CTLFLAG_RD,
 269            &hammer_count_io_running_write, 0, "");
 270 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, zone_limit, CTLFLAG_RW,
 271            &hammer_zone_limit, 0, "");
 272 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, contention_count, CTLFLAG_RW,
 273            &hammer_contention_count, 0, "");
 274 SYSCTL_INT(_vfs_hammer, OID_AUTO, autoflush, CTLFLAG_RW,
 275            &hammer_autoflush, 0, "");
 276 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_zone, CTLFLAG_RW,
 277            &hammer_verify_zone, 0, "");
 278 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_data, CTLFLAG_RW,
 279            &hammer_verify_data, 0, "");
 280 SYSCTL_INT(_vfs_hammer, OID_AUTO, write_mode, CTLFLAG_RW,
 281            &hammer_write_mode, 0, "");
 282 SYSCTL_INT(_vfs_hammer, OID_AUTO, double_buffer, CTLFLAG_RW,
 283            &hammer_double_buffer, 0, "");
 284 SYSCTL_INT(_vfs_hammer, OID_AUTO, btree_full_undo, CTLFLAG_RW,
 285            &hammer_btree_full_undo, 0, "");
 286 SYSCTL_INT(_vfs_hammer, OID_AUTO, yield_check, CTLFLAG_RW,
 287            &hammer_yield_check, 0, "");
 288 SYSCTL_INT(_vfs_hammer, OID_AUTO, fsync_mode, CTLFLAG_RW,
 289            &hammer_fsync_mode, 0, "");
 290
 291 /* KTR_INFO_MASTER(hammer); */
 292
 293 /*
 294  * VFS ABI
 295  */
 296 static void     hammer_free_hmp(struct mount *mp);
 297
 298 static int      hammer_vfs_mount(struct mount *mp, char *path, caddr_t data,
 299                                 struct ucred *cred);
 300 static int      hammer_vfs_unmount(struct mount *mp, int mntflags);
 301 static int      hammer_vfs_root(struct mount *mp, struct vnode **vpp);
 302 static int      hammer_vfs_statfs(struct mount *mp, struct statfs *sbp,
 303                                 struct ucred *cred);
 304 static int      hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp,
 305                                 struct ucred *cred);
 306 static int      hammer_vfs_sync(struct mount *mp, int waitfor);
 307 static int      hammer_vfs_vget(struct mount *mp, struct vnode *dvp,
 308                                 ino_t ino, struct vnode **vpp);
 309 static int      hammer_vfs_init(struct vfsconf *conf);
 310 static int      hammer_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
 311                                 struct fid *fhp, struct vnode **vpp);
 312 static int      hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp);
 313 static int      hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
 314                                 int *exflagsp, struct ucred **credanonp);
 315
 316
 317 static struct vfsops hammer_vfsops = {
 318         .vfs_mount      = hammer_vfs_mount,
 319         .vfs_unmount    = hammer_vfs_unmount,
 320         .vfs_root       = hammer_vfs_root,
 321         .vfs_statfs     = hammer_vfs_statfs,
 322         .vfs_statvfs    = hammer_vfs_statvfs,
 323         .vfs_sync       = hammer_vfs_sync,
 324         .vfs_vget       = hammer_vfs_vget,
 325         .vfs_init       = hammer_vfs_init,
 326         .vfs_vptofh     = hammer_vfs_vptofh,
 327         .vfs_fhtovp     = hammer_vfs_fhtovp,
 328         .vfs_checkexp   = hammer_vfs_checkexp
 329 };
 330
 331 MALLOC_DEFINE(M_HAMMER, "HAMMER-mount", "");
 332
 333 VFS_SET(hammer_vfsops, hammer, 0);
 334 MODULE_VERSION(hammer, 1);
 335
 336 static int
 337 hammer_vfs_init(struct vfsconf *conf)
 338 {
 339         long n;
 340
 341         /*
 342          * Wait up to this long for an exclusive deadlock to clear
 343          * before acquiring a new shared lock on the ip.  The deadlock
 344          * may have occured on a b-tree node related to the ip.
 345          */
 346         if (hammer_tdmux_ticks == 0)
 347                 hammer_tdmux_ticks = hz / 5;
 348
 349         /*
 350          * Autosize, but be careful because a hammer filesystem's
 351          * reserve is partially calculated based on dirtybufspace,
 352          * so we simply cannot allow it to get too large.
 353          */
 354         if (hammer_limit_recs == 0) {
 355                 n = nbuf * 25;
 356                 if (n > kmalloc_limit(M_HAMMER) / 512)
 357                         n = kmalloc_limit(M_HAMMER) / 512;
 358                 if (n > 2 * 1024 * 1024)
 359                         n = 2 * 1024 * 1024;
 360                 hammer_limit_recs = (int)n;
 361         }
 362         if (hammer_limit_dirtybufspace == 0) {
 363                 hammer_limit_dirtybufspace = hidirtybufspace / 2;
 364                 if (hammer_limit_dirtybufspace < 1L * 1024 * 1024)
 365                         hammer_limit_dirtybufspace = 1024L * 1024;
 366                 if (hammer_limit_dirtybufspace > 1024L * 1024 * 1024)
 367                         hammer_limit_dirtybufspace = 1024L * 1024 * 1024;
 368         }
 369
 370         /*
 371          * The hammer_inode structure detaches from the vnode on reclaim.
 372          * This limits the number of inodes in this state to prevent a
 373          * memory pool blowout.
 374          */
 375         if (hammer_limit_reclaims == 0)
 376                 hammer_limit_reclaims = desiredvnodes / 10;
 377
 378         return(0);
 379 }
 380
 381 static int
 382 hammer_vfs_mount(struct mount *mp, char *mntpt, caddr_t data,
 383                  struct ucred *cred)
 384 {
 385         struct hammer_mount_info info;
 386         hammer_mount_t hmp;
 387         hammer_volume_t rootvol;
 388         struct vnode *rootvp;
 389         struct vnode *devvp = NULL;
 390         const char *upath;      /* volume name in userspace */
 391         char *path;             /* volume name in system space */
 392         int error;
 393         int i;
 394         int master_id;
 395         char *next_volume_ptr = NULL;
 396
 397         /*
 398          * Accept hammer_mount_info.  mntpt is NULL for root mounts at boot.
 399          */
 400         if (mntpt == NULL) {
 401                 bzero(&info, sizeof(info));
 402                 info.asof = 0;
 403                 info.hflags = 0;
 404                 info.nvolumes = 1;
 405
 406                 next_volume_ptr = mp->mnt_stat.f_mntfromname;
 407
 408                 /* Count number of volumes separated by ':' */
 409                 for (char *p = next_volume_ptr; *p != '\0'; ++p) {
 410                         if (*p == ':') {
 411                                 ++info.nvolumes;
 412                         }
 413                 }
 414
 415                 mp->mnt_flag &= ~MNT_RDONLY; /* mount R/W */
 416         } else {
 417                 if ((error = copyin(data, &info, sizeof(info))) != 0)
 418                         return (error);
 419         }
 420
 421         /*
 422          * updating or new mount
 423          */
 424         if (mp->mnt_flag & MNT_UPDATE) {
 425                 hmp = (void *)mp->mnt_data;
 426                 KKASSERT(hmp != NULL);
 427         } else {
 428                 if (info.nvolumes <= 0 || info.nvolumes > HAMMER_MAX_VOLUMES)
 429                         return (EINVAL);
 430                 hmp = NULL;
 431         }
 432
 433         /*
 434          * master-id validation.  The master id may not be changed by a
 435          * mount update.
 436          */
 437         if (info.hflags & HMNT_MASTERID) {
 438                 if (hmp && hmp->master_id != info.master_id) {
 439                         kprintf("hammer: cannot change master id "
 440                                 "with mount update\n");
 441                         return(EINVAL);
 442                 }
 443                 master_id = info.master_id;
 444                 if (master_id < -1 || master_id >= HAMMER_MAX_MASTERS)
 445                         return (EINVAL);
 446         } else {
 447                 if (hmp)
 448                         master_id = hmp->master_id;
 449                 else
 450                         master_id = 0;
 451         }
 452
 453         /*
 454          * Internal mount data structure
 455          */
 456         if (hmp == NULL) {
 457                 hmp = kmalloc(sizeof(*hmp), M_HAMMER, M_WAITOK | M_ZERO);
 458                 mp->mnt_data = (qaddr_t)hmp;
 459                 hmp->mp = mp;
 460                 /*TAILQ_INIT(&hmp->recycle_list);*/
 461
 462                 /*
 463                  * Make sure kmalloc type limits are set appropriately.
 464                  *
 465                  * Our inode kmalloc group is sized based on maxvnodes
 466                  * (controlled by the system, not us).
 467                  */
 468                 kmalloc_create(&hmp->m_misc, "HAMMER-others");
 469                 kmalloc_create(&hmp->m_inodes, "HAMMER-inodes");
 470
 471                 kmalloc_raise_limit(hmp->m_inodes, 0);  /* unlimited */
 472
 473                 hmp->root_btree_beg.localization = 0x00000000U;
 474                 hmp->root_btree_beg.obj_id = -0x8000000000000000LL;
 475                 hmp->root_btree_beg.key = -0x8000000000000000LL;
 476                 hmp->root_btree_beg.create_tid = 1;
 477                 hmp->root_btree_beg.delete_tid = 1;
 478                 hmp->root_btree_beg.rec_type = 0;
 479                 hmp->root_btree_beg.obj_type = 0;
 480
 481                 hmp->root_btree_end.localization = 0xFFFFFFFFU;
 482                 hmp->root_btree_end.obj_id = 0x7FFFFFFFFFFFFFFFLL;
 483                 hmp->root_btree_end.key = 0x7FFFFFFFFFFFFFFFLL;
 484                 hmp->root_btree_end.create_tid = 0xFFFFFFFFFFFFFFFFULL;
 485                 hmp->root_btree_end.delete_tid = 0;   /* special case */
 486                 hmp->root_btree_end.rec_type = 0xFFFFU;
 487                 hmp->root_btree_end.obj_type = 0;
 488
 489                 hmp->krate.freq = 1;    /* maximum reporting rate (hz) */
 490                 hmp->krate.count = -16; /* initial burst */
 491
 492                 hmp->sync_lock.refs = 1;
 493                 hmp->free_lock.refs = 1;
 494                 hmp->undo_lock.refs = 1;
 495                 hmp->blkmap_lock.refs = 1;
 496                 hmp->snapshot_lock.refs = 1;
 497                 hmp->volume_lock.refs = 1;
 498
 499                 TAILQ_INIT(&hmp->delay_list);
 500                 TAILQ_INIT(&hmp->flush_group_list);
 501                 TAILQ_INIT(&hmp->objid_cache_list);
 502                 TAILQ_INIT(&hmp->undo_lru_list);
 503                 TAILQ_INIT(&hmp->reclaim_list);
 504
 505                 RB_INIT(&hmp->rb_dedup_crc_root);
 506                 RB_INIT(&hmp->rb_dedup_off_root);
 507                 TAILQ_INIT(&hmp->dedup_lru_list);
 508         }
 509         hmp->hflags &= ~HMNT_USERFLAGS;
 510         hmp->hflags |= info.hflags & HMNT_USERFLAGS;
 511
 512         hmp->master_id = master_id;
 513
 514         if (info.asof) {
 515                 mp->mnt_flag |= MNT_RDONLY;
 516                 hmp->asof = info.asof;
 517         } else {
 518                 hmp->asof = HAMMER_MAX_TID;
 519         }
 520
 521         hmp->volume_to_remove = -1;
 522
 523         /*
 524          * Re-open read-write if originally read-only, or vise-versa.
 525          *
 526          * When going from read-only to read-write execute the stage2
 527          * recovery if it has not already been run.
 528          */
 529         if (mp->mnt_flag & MNT_UPDATE) {
 530                 lwkt_gettoken(&hmp->fs_token);
 531                 error = 0;
 532                 if (hmp->ronly && (mp->mnt_kern_flag & MNTK_WANTRDWR)) {
 533                         kprintf("HAMMER read-only -> read-write\n");
 534                         hmp->ronly = 0;
 535                         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
 536                                 hammer_adjust_volume_mode, NULL);
 537                         rootvol = hammer_get_root_volume(hmp, &error);
 538                         if (rootvol) {
 539                                 hammer_recover_flush_buffers(hmp, rootvol, 1);
 540                                 error = hammer_recover_stage2(hmp, rootvol);
 541                                 bcopy(rootvol->ondisk->vol0_blockmap,
 542                                       hmp->blockmap,
 543                                       sizeof(hmp->blockmap));
 544                                 hammer_rel_volume(rootvol, 0);
 545                         }
 546                         RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
 547                                 hammer_reload_inode, NULL);
 548                         /* kernel clears MNT_RDONLY */
 549                 } else if (hmp->ronly == 0 && (mp->mnt_flag & MNT_RDONLY)) {
 550                         kprintf("HAMMER read-write -> read-only\n");
 551                         hmp->ronly = 1; /* messy */
 552                         RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
 553                                 hammer_reload_inode, NULL);
 554                         hmp->ronly = 0;
 555                         hammer_flusher_sync(hmp);
 556                         hammer_flusher_sync(hmp);
 557                         hammer_flusher_sync(hmp);
 558                         hmp->ronly = 1;
 559                         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
 560                                 hammer_adjust_volume_mode, NULL);
 561                 }
 562                 lwkt_reltoken(&hmp->fs_token);
 563                 return(error);
 564         }
 565
 566         RB_INIT(&hmp->rb_vols_root);
 567         RB_INIT(&hmp->rb_inos_root);
 568         RB_INIT(&hmp->rb_redo_root);
 569         RB_INIT(&hmp->rb_nods_root);
 570         RB_INIT(&hmp->rb_undo_root);
 571         RB_INIT(&hmp->rb_resv_root);
 572         RB_INIT(&hmp->rb_bufs_root);
 573         RB_INIT(&hmp->rb_pfsm_root);
 574
 575         hmp->ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
 576
 577         RB_INIT(&hmp->volu_root);
 578         RB_INIT(&hmp->undo_root);
 579         RB_INIT(&hmp->data_root);
 580         RB_INIT(&hmp->meta_root);
 581         RB_INIT(&hmp->lose_root);
 582         TAILQ_INIT(&hmp->iorun_list);
 583
 584         lwkt_token_init(&hmp->fs_token, "hammerfs");
 585         lwkt_token_init(&hmp->io_token, "hammerio");
 586
 587         lwkt_gettoken(&hmp->fs_token);
 588
 589         /*
 590          * Load volumes
 591          */
 592         path = objcache_get(namei_oc, M_WAITOK);
 593         hmp->nvolumes = -1;
 594         for (i = 0; i < info.nvolumes; ++i) {
 595                 if (mntpt == NULL) {
 596                         /*
 597                          * Root mount.
 598                          */
 599                         KKASSERT(next_volume_ptr != NULL);
 600                         strcpy(path, "");
 601                         if (*next_volume_ptr != '/') {
 602                                 /* relative path */
 603                                 strcpy(path, "/dev/");
 604                         }
 605                         int k;
 606                         for (k = strlen(path); k < MAXPATHLEN-1; ++k) {
 607                                 if (*next_volume_ptr == '\0') {
 608                                         break;
 609                                 } else if (*next_volume_ptr == ':') {
 610                                         ++next_volume_ptr;
 611                                         break;
 612                                 } else {
 613                                         path[k] = *next_volume_ptr;
 614                                         ++next_volume_ptr;
 615                                 }
 616                         }
 617                         path[k] = '\0';
 618
 619                         error = 0;
 620                         cdev_t dev = kgetdiskbyname(path);
 621                         error = bdevvp(dev, &devvp);
 622                         if (error) {
 623                                 kprintf("hammer_mountroot: can't find devvp\n");
 624                         }
 625                 } else {
 626                         error = copyin(&info.volumes[i], &upath,
 627                                        sizeof(char *));
 628                         if (error == 0)
 629                                 error = copyinstr(upath, path,
 630                                                   MAXPATHLEN, NULL);
 631                 }
 632                 if (error == 0)
 633                         error = hammer_install_volume(hmp, path, devvp);
 634                 if (error)
 635                         break;
 636         }
 637         objcache_put(namei_oc, path);
 638
 639         /*
 640          * Make sure we found a root volume
 641          */
 642         if (error == 0 && hmp->rootvol == NULL) {
 643                 kprintf("hammer_mount: No root volume found!\n");
 644                 error = EINVAL;
 645         }
 646
 647         /*
 648          * Check that all required volumes are available
 649          */
 650         if (error == 0 && hammer_mountcheck_volumes(hmp)) {
 651                 kprintf("hammer_mount: Missing volumes, cannot mount!\n");
 652                 error = EINVAL;
 653         }
 654
 655         if (error) {
 656                 /* called with fs_token held */
 657                 hammer_free_hmp(mp);
 658                 return (error);
 659         }
 660
 661         /*
 662          * No errors, setup enough of the mount point so we can lookup the
 663          * root vnode.
 664          */
 665         mp->mnt_iosize_max = MAXPHYS;
 666         mp->mnt_kern_flag |= MNTK_FSMID;
 667         mp->mnt_kern_flag |= MNTK_THR_SYNC;     /* new vsyncscan semantics */
 668
 669         /*
 670          * MPSAFE code.  Note that VOPs and VFSops which are not MPSAFE
 671          * will acquire a per-mount token prior to entry and release it
 672          * on return, so even if we do not specify it we no longer get
 673          * the BGL regardlless of how we are flagged.
 674          */
 675         mp->mnt_kern_flag |= MNTK_ALL_MPSAFE;
 676         /*MNTK_RD_MPSAFE | MNTK_GA_MPSAFE | MNTK_IN_MPSAFE;*/
 677
 678         /*
 679          * note: f_iosize is used by vnode_pager_haspage() when constructing
 680          * its VOP_BMAP call.
 681          */
 682         mp->mnt_stat.f_iosize = HAMMER_BUFSIZE;
 683         mp->mnt_stat.f_bsize = HAMMER_BUFSIZE;
 684
 685         mp->mnt_vstat.f_frsize = HAMMER_BUFSIZE;
 686         mp->mnt_vstat.f_bsize = HAMMER_BUFSIZE;
 687
 688         mp->mnt_maxsymlinklen = 255;
 689         mp->mnt_flag |= MNT_LOCAL;
 690
 691         vfs_add_vnodeops(mp, &hammer_vnode_vops, &mp->mnt_vn_norm_ops);
 692         vfs_add_vnodeops(mp, &hammer_spec_vops, &mp->mnt_vn_spec_ops);
 693         vfs_add_vnodeops(mp, &hammer_fifo_vops, &mp->mnt_vn_fifo_ops);
 694
 695         /*
 696          * The root volume's ondisk pointer is only valid if we hold a
 697          * reference to it.
 698          */
 699         rootvol = hammer_get_root_volume(hmp, &error);
 700         if (error)
 701                 goto failed;
 702
 703         /*
 704          * Perform any necessary UNDO operations.  The recovery code does
 705          * call hammer_undo_lookup() so we have to pre-cache the blockmap,
 706          * and then re-copy it again after recovery is complete.
 707          *
 708          * If this is a read-only mount the UNDO information is retained
 709          * in memory in the form of dirty buffer cache buffers, and not
 710          * written back to the media.
 711          */
 712         bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
 713               sizeof(hmp->blockmap));
 714
 715         /*
 716          * Check filesystem version
 717          */
 718         hmp->version = rootvol->ondisk->vol_version;
 719         if (hmp->version < HAMMER_VOL_VERSION_MIN ||
 720             hmp->version > HAMMER_VOL_VERSION_MAX) {
 721                 kprintf("HAMMER: mount unsupported fs version %d\n",
 722                         hmp->version);
 723                 error = ERANGE;
 724                 goto done;
 725         }
 726
 727         /*
 728          * The undo_rec_limit limits the size of flush groups to avoid
 729          * blowing out the UNDO FIFO.  This calculation is typically in
 730          * the tens of thousands and is designed primarily when small
 731          * HAMMER filesystems are created.
 732          */
 733         hmp->undo_rec_limit = hammer_undo_max(hmp) / 8192 + 100;
 734         if (hammer_debug_general & 0x0001)
 735                 kprintf("HAMMER: undo_rec_limit %d\n", hmp->undo_rec_limit);
 736
 737         /*
 738          * NOTE: Recover stage1 not only handles meta-data recovery, it
 739          *       also sets hmp->undo_seqno for HAMMER VERSION 4+ filesystems.
 740          */
 741         error = hammer_recover_stage1(hmp, rootvol);
 742         if (error) {
 743                 kprintf("Failed to recover HAMMER filesystem on mount\n");
 744                 goto done;
 745         }
 746
 747         /*
 748          * Finish setup now that we have a good root volume.
 749          *
 750          * The top 16 bits of fsid.val[1] is a pfs id.
 751          */
 752         ksnprintf(mp->mnt_stat.f_mntfromname,
 753                   sizeof(mp->mnt_stat.f_mntfromname), "%s",
 754                   rootvol->ondisk->vol_name);
 755         mp->mnt_stat.f_fsid.val[0] =
 756                 crc32((char *)&rootvol->ondisk->vol_fsid + 0, 8);
 757         mp->mnt_stat.f_fsid.val[1] =
 758                 crc32((char *)&rootvol->ondisk->vol_fsid + 8, 8);
 759         mp->mnt_stat.f_fsid.val[1] &= 0x0000FFFF;
 760
 761         mp->mnt_vstat.f_fsid_uuid = rootvol->ondisk->vol_fsid;
 762         mp->mnt_vstat.f_fsid = crc32(&mp->mnt_vstat.f_fsid_uuid,
 763                                      sizeof(mp->mnt_vstat.f_fsid_uuid));
 764
 765         /*
 766          * Certain often-modified fields in the root volume are cached in
 767          * the hammer_mount structure so we do not have to generate lots
 768          * of little UNDO structures for them.
 769          *
 770          * Recopy after recovery.  This also has the side effect of
 771          * setting our cached undo FIFO's first_offset, which serves to
 772          * placemark the FIFO start for the NEXT flush cycle while the
 773          * on-disk first_offset represents the LAST flush cycle.
 774          */
 775         hmp->next_tid = rootvol->ondisk->vol0_next_tid;
 776         hmp->flush_tid1 = hmp->next_tid;
 777         hmp->flush_tid2 = hmp->next_tid;
 778         bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
 779               sizeof(hmp->blockmap));
 780         hmp->copy_stat_freebigblocks = rootvol->ondisk->vol0_stat_freebigblocks;
 781
 782         hammer_flusher_create(hmp);
 783
 784         /*
 785          * Locate the root directory using the root cluster's B-Tree as a
 786          * starting point.  The root directory uses an obj_id of 1.
 787          *
 788          * FUTURE: Leave the root directory cached referenced but unlocked
 789          * in hmp->rootvp (need to flush it on unmount).
 790          */
 791         error = hammer_vfs_vget(mp, NULL, 1, &rootvp);
 792         if (error)
 793                 goto done;
 794         vput(rootvp);
 795         /*vn_unlock(hmp->rootvp);*/
 796         if (hmp->ronly == 0)
 797                 error = hammer_recover_stage2(hmp, rootvol);
 798
 799         /*
 800          * If the stage2 recovery fails be sure to clean out all cached
 801          * vnodes before throwing away the mount structure or bad things
 802          * will happen.
 803          */
 804         if (error)
 805                 vflush(mp, 0, 0);
 806
 807 done:
 808         if ((mp->mnt_flag & MNT_UPDATE) == 0) {
 809                 /* New mount */
 810
 811                 /* Populate info for mount point (NULL pad)*/
 812                 bzero(mp->mnt_stat.f_mntonname, MNAMELEN);
 813                 size_t size;
 814                 if (mntpt) {
 815                         copyinstr(mntpt, mp->mnt_stat.f_mntonname,
 816                                                         MNAMELEN -1, &size);
 817                 } else { /* Root mount */
 818                         mp->mnt_stat.f_mntonname[0] = '/';
 819                 }
 820         }
 821         (void)VFS_STATFS(mp, &mp->mnt_stat, cred);
 822         hammer_rel_volume(rootvol, 0);
 823 failed:
 824         /*
 825          * Cleanup and return.
 826          */
 827         if (error) {
 828                 /* called with fs_token held */
 829                 hammer_free_hmp(mp);
 830         } else {
 831                 lwkt_reltoken(&hmp->fs_token);
 832         }
 833         return (error);
 834 }
 835
 836 static int
 837 hammer_vfs_unmount(struct mount *mp, int mntflags)
 838 {
 839         hammer_mount_t hmp = (void *)mp->mnt_data;
 840         int flags;
 841         int error;
 842
 843         /*
 844          * Clean out the vnodes
 845          */
 846         lwkt_gettoken(&hmp->fs_token);
 847         flags = 0;
 848         if (mntflags & MNT_FORCE)
 849                 flags |= FORCECLOSE;
 850         error = vflush(mp, 0, flags);
 851
 852         /*
 853          * Clean up the internal mount structure and related entities.  This
 854          * may issue I/O.
 855          */
 856         if (error == 0) {
 857                 /* called with fs_token held */
 858                 hammer_free_hmp(mp);
 859         } else {
 860                 lwkt_reltoken(&hmp->fs_token);
 861         }
 862         return(error);
 863 }
 864
 865 /*
 866  * Clean up the internal mount structure and disassociate it from the mount.
 867  * This may issue I/O.
 868  *
 869  * Called with fs_token held.
 870  */
 871 static void
 872 hammer_free_hmp(struct mount *mp)
 873 {
 874         hammer_mount_t hmp = (void *)mp->mnt_data;
 875         hammer_flush_group_t flg;
 876         int count;
 877         int dummy;
 878
 879         /*
 880          * Flush anything dirty.  This won't even run if the
 881          * filesystem errored-out.
 882          */
 883         count = 0;
 884         while (hammer_flusher_haswork(hmp)) {
 885                 hammer_flusher_sync(hmp);
 886                 ++count;
 887                 if (count >= 5) {
 888                         if (count == 5)
 889                                 kprintf("HAMMER: umount flushing.");
 890                         else
 891                                 kprintf(".");
 892                         tsleep(&dummy, 0, "hmrufl", hz);
 893                 }
 894                 if (count == 30) {
 895                         kprintf("giving up\n");
 896                         break;
 897                 }
 898         }
 899         if (count >= 5 && count < 30)
 900                 kprintf("\n");
 901
 902         /*
 903          * If the mount had a critical error we have to destroy any
 904          * remaining inodes before we can finish cleaning up the flusher.
 905          */
 906         if (hmp->flags & HAMMER_MOUNT_CRITICAL_ERROR) {
 907                 RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
 908                         hammer_destroy_inode_callback, NULL);
 909         }
 910
 911         /*
 912          * There shouldn't be any inodes left now and any left over
 913          * flush groups should now be empty.
 914          */
 915         KKASSERT(RB_EMPTY(&hmp->rb_inos_root));
 916         while ((flg = TAILQ_FIRST(&hmp->flush_group_list)) != NULL) {
 917                 TAILQ_REMOVE(&hmp->flush_group_list, flg, flush_entry);
 918                 KKASSERT(RB_EMPTY(&flg->flush_tree));
 919                 if (flg->refs) {
 920                         kprintf("HAMMER: Warning, flush_group %p was "
 921                                 "not empty on umount!\n", flg);
 922                 }
 923                 kfree(flg, hmp->m_misc);
 924         }
 925
 926         /*
 927          * We can finally destroy the flusher
 928          */
 929         hammer_flusher_destroy(hmp);
 930
 931         /*
 932          * We may have held recovered buffers due to a read-only mount.
 933          * These must be discarded.
 934          */
 935         if (hmp->ronly)
 936                 hammer_recover_flush_buffers(hmp, NULL, -1);
 937
 938         /*
 939          * Unload buffers and then volumes
 940          */
 941         RB_SCAN(hammer_buf_rb_tree, &hmp->rb_bufs_root, NULL,
 942                 hammer_unload_buffer, NULL);
 943         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
 944                 hammer_unload_volume, NULL);
 945
 946         mp->mnt_data = NULL;
 947         mp->mnt_flag &= ~MNT_LOCAL;
 948         hmp->mp = NULL;
 949         hammer_destroy_objid_cache(hmp);
 950         hammer_destroy_dedup_cache(hmp);
 951         if (hmp->dedup_free_cache != NULL) {
 952                 kfree(hmp->dedup_free_cache, hmp->m_misc);
 953                 hmp->dedup_free_cache = NULL;
 954         }
 955         kmalloc_destroy(&hmp->m_misc);
 956         kmalloc_destroy(&hmp->m_inodes);
 957         lwkt_reltoken(&hmp->fs_token);
 958         kfree(hmp, M_HAMMER);
 959 }
 960
 961 /*
 962  * Report critical errors.  ip may be NULL.
 963  */
 964 void
 965 hammer_critical_error(hammer_mount_t hmp, hammer_inode_t ip,
 966                       int error, const char *msg)
 967 {
 968         hmp->flags |= HAMMER_MOUNT_CRITICAL_ERROR;
 969
 970         krateprintf(&hmp->krate,
 971                     "HAMMER(%s): Critical error inode=%jd error=%d %s\n",
 972                     hmp->mp->mnt_stat.f_mntfromname,
 973                     (intmax_t)(ip ? ip->obj_id : -1),
 974                     error, msg);
 975
 976         if (hmp->ronly == 0) {
 977                 hmp->ronly = 2;         /* special errored read-only mode */
 978                 hmp->mp->mnt_flag |= MNT_RDONLY;
 979                 RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
 980                         hammer_adjust_volume_mode, NULL);
 981                 kprintf("HAMMER(%s): Forcing read-only mode\n",
 982                         hmp->mp->mnt_stat.f_mntfromname);
 983         }
 984         hmp->error = error;
 985         if (hammer_debug_critical)
 986                 Debugger("Entering debugger");
 987 }
 988
 989
 990 /*
 991  * Obtain a vnode for the specified inode number.  An exclusively locked
 992  * vnode is returned.
 993  */
 994 int
 995 hammer_vfs_vget(struct mount *mp, struct vnode *dvp,
 996                 ino_t ino, struct vnode **vpp)
 997 {
 998         struct hammer_transaction trans;
 999         struct hammer_mount *hmp = (void *)mp->mnt_data;
1000         struct hammer_inode *ip;
1001         int error;
1002         u_int32_t localization;
1003
1004         lwkt_gettoken(&hmp->fs_token);
1005         hammer_simple_transaction(&trans, hmp);
1006
1007         /*
1008          * If a directory vnode is supplied (mainly NFS) then we can acquire
1009          * the PFS domain from it.  Otherwise we would only be able to vget
1010          * inodes in the root PFS.
1011          */
1012         if (dvp) {
1013                 localization = HAMMER_DEF_LOCALIZATION +
1014                                 VTOI(dvp)->obj_localization;
1015         } else {
1016                 localization = HAMMER_DEF_LOCALIZATION;
1017         }
1018
1019         /*
1020          * Lookup the requested HAMMER inode.  The structure must be
1021          * left unlocked while we manipulate the related vnode to avoid
1022          * a deadlock.
1023          */
1024         ip = hammer_get_inode(&trans, NULL, ino,
1025                               hmp->asof, localization,
1026                               0, &error);
1027         if (ip == NULL) {
1028                 *vpp = NULL;
1029         } else {
1030                 error = hammer_get_vnode(ip, vpp);
1031                 hammer_rel_inode(ip, 0);
1032         }
1033         hammer_done_transaction(&trans);
1034         lwkt_reltoken(&hmp->fs_token);
1035         return (error);
1036 }
1037
1038 /*
1039  * Return the root vnode for the filesystem.
1040  *
1041  * HAMMER stores the root vnode in the hammer_mount structure so
1042  * getting it is easy.
1043  */
1044 static int
1045 hammer_vfs_root(struct mount *mp, struct vnode **vpp)
1046 {
1047         int error;
1048
1049         error = hammer_vfs_vget(mp, NULL, 1, vpp);
1050         return (error);
1051 }
1052
1053 static int
1054 hammer_vfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
1055 {
1056         struct hammer_mount *hmp = (void *)mp->mnt_data;
1057         hammer_volume_t volume;
1058         hammer_volume_ondisk_t ondisk;
1059         int error;
1060         int64_t bfree;
1061         int64_t breserved;
1062
1063         lwkt_gettoken(&hmp->fs_token);
1064         volume = hammer_get_root_volume(hmp, &error);
1065         if (error) {
1066                 lwkt_reltoken(&hmp->fs_token);
1067                 return(error);
1068         }
1069         ondisk = volume->ondisk;
1070
1071         /*
1072          * Basic stats
1073          */
1074         _hammer_checkspace(hmp, HAMMER_CHKSPC_WRITE, &breserved);
1075         mp->mnt_stat.f_files = ondisk->vol0_stat_inodes;
1076         bfree = ondisk->vol0_stat_freebigblocks * HAMMER_BIGBLOCK_SIZE;
1077         hammer_rel_volume(volume, 0);
1078
1079         mp->mnt_stat.f_bfree = (bfree - breserved) / HAMMER_BUFSIZE;
1080         mp->mnt_stat.f_bavail = mp->mnt_stat.f_bfree;
1081         if (mp->mnt_stat.f_files < 0)
1082                 mp->mnt_stat.f_files = 0;
1083
1084         *sbp = mp->mnt_stat;
1085         lwkt_reltoken(&hmp->fs_token);
1086         return(0);
1087 }
1088
1089 static int
1090 hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred)
1091 {
1092         struct hammer_mount *hmp = (void *)mp->mnt_data;
1093         hammer_volume_t volume;
1094         hammer_volume_ondisk_t ondisk;
1095         int error;
1096         int64_t bfree;
1097         int64_t breserved;
1098
1099         lwkt_gettoken(&hmp->fs_token);
1100         volume = hammer_get_root_volume(hmp, &error);
1101         if (error) {
1102                 lwkt_reltoken(&hmp->fs_token);
1103                 return(error);
1104         }
1105         ondisk = volume->ondisk;
1106
1107         /*
1108          * Basic stats
1109          */
1110         _hammer_checkspace(hmp, HAMMER_CHKSPC_WRITE, &breserved);
1111         mp->mnt_vstat.f_files = ondisk->vol0_stat_inodes;
1112         bfree = ondisk->vol0_stat_freebigblocks * HAMMER_BIGBLOCK_SIZE;
1113         hammer_rel_volume(volume, 0);
1114
1115         mp->mnt_vstat.f_bfree = (bfree - breserved) / HAMMER_BUFSIZE;
1116         mp->mnt_vstat.f_bavail = mp->mnt_vstat.f_bfree;
1117         if (mp->mnt_vstat.f_files < 0)
1118                 mp->mnt_vstat.f_files = 0;
1119         *sbp = mp->mnt_vstat;
1120         lwkt_reltoken(&hmp->fs_token);
1121         return(0);
1122 }
1123
1124 /*
1125  * Sync the filesystem.  Currently we have to run it twice, the second
1126  * one will advance the undo start index to the end index, so if a crash
1127  * occurs no undos will be run on mount.
1128  *
1129  * We do not sync the filesystem if we are called from a panic.  If we did
1130  * we might end up blowing up a sync that was already in progress.
1131  */
1132 static int
1133 hammer_vfs_sync(struct mount *mp, int waitfor)
1134 {
1135         struct hammer_mount *hmp = (void *)mp->mnt_data;
1136         int error;
1137
1138         lwkt_gettoken(&hmp->fs_token);
1139         if (panicstr == NULL) {
1140                 error = hammer_sync_hmp(hmp, waitfor);
1141         } else {
1142                 error = EIO;
1143         }
1144         lwkt_reltoken(&hmp->fs_token);
1145         return (error);
1146 }
1147
1148 /*
1149  * Convert a vnode to a file handle.
1150  *
1151  * Accesses read-only fields on already-referenced structures so
1152  * no token is needed.
1153  */
1154 static int
1155 hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp)
1156 {
1157         hammer_inode_t ip;
1158
1159         KKASSERT(MAXFIDSZ >= 16);
1160         ip = VTOI(vp);
1161         fhp->fid_len = offsetof(struct fid, fid_data[16]);
1162         fhp->fid_ext = ip->obj_localization >> 16;
1163         bcopy(&ip->obj_id, fhp->fid_data + 0, sizeof(ip->obj_id));
1164         bcopy(&ip->obj_asof, fhp->fid_data + 8, sizeof(ip->obj_asof));
1165         return(0);
1166 }
1167
1168
1169 /*
1170  * Convert a file handle back to a vnode.
1171  *
1172  * Use rootvp to enforce PFS isolation when a PFS is exported via a
1173  * null mount.
1174  */
1175 static int
1176 hammer_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
1177                   struct fid *fhp, struct vnode **vpp)
1178 {
1179         hammer_mount_t hmp = (void *)mp->mnt_data;
1180         struct hammer_transaction trans;
1181         struct hammer_inode *ip;
1182         struct hammer_inode_info info;
1183         int error;
1184         u_int32_t localization;
1185
1186         bcopy(fhp->fid_data + 0, &info.obj_id, sizeof(info.obj_id));
1187         bcopy(fhp->fid_data + 8, &info.obj_asof, sizeof(info.obj_asof));
1188         if (rootvp)
1189                 localization = VTOI(rootvp)->obj_localization;
1190         else
1191                 localization = (u_int32_t)fhp->fid_ext << 16;
1192
1193         lwkt_gettoken(&hmp->fs_token);
1194         hammer_simple_transaction(&trans, hmp);
1195
1196         /*
1197          * Get/allocate the hammer_inode structure.  The structure must be
1198          * unlocked while we manipulate the related vnode to avoid a
1199          * deadlock.
1200          */
1201         ip = hammer_get_inode(&trans, NULL, info.obj_id,
1202                               info.obj_asof, localization, 0, &error);
1203         if (ip) {
1204                 error = hammer_get_vnode(ip, vpp);
1205                 hammer_rel_inode(ip, 0);
1206         } else {
1207                 *vpp = NULL;
1208         }
1209         hammer_done_transaction(&trans);
1210         lwkt_reltoken(&hmp->fs_token);
1211         return (error);
1212 }
1213
1214 static int
1215 hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
1216                     int *exflagsp, struct ucred **credanonp)
1217 {
1218         hammer_mount_t hmp = (void *)mp->mnt_data;
1219         struct netcred *np;
1220         int error;
1221
1222         lwkt_gettoken(&hmp->fs_token);
1223         np = vfs_export_lookup(mp, &hmp->export, nam);
1224         if (np) {
1225                 *exflagsp = np->netc_exflags;
1226                 *credanonp = &np->netc_anon;
1227                 error = 0;
1228         } else {
1229                 error = EACCES;
1230         }
1231         lwkt_reltoken(&hmp->fs_token);
1232         return (error);
1233
1234 }
1235
1236 int
1237 hammer_vfs_export(struct mount *mp, int op, const struct export_args *export)
1238 {
1239         hammer_mount_t hmp = (void *)mp->mnt_data;
1240         int error;
1241
1242         lwkt_gettoken(&hmp->fs_token);
1243
1244         switch(op) {
1245         case MOUNTCTL_SET_EXPORT:
1246                 error = vfs_export(mp, &hmp->export, export);
1247                 break;
1248         default:
1249                 error = EOPNOTSUPP;
1250                 break;
1251         }
1252         lwkt_reltoken(&hmp->fs_token);
1253
1254         return(error);
1255 }
1256