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/mountctl.h>
  36
  37 #include "hammer.h"
  38
  39 /*
  40  * NOTE!  Global statistics may not be MPSAFE so HAMMER never uses them
  41  *        in conditionals.
  42  */
  43 int hammer_supported_version = HAMMER_VOL_VERSION_DEFAULT;
  44 int hammer_debug_io;
  45 int hammer_debug_general;
  46 int hammer_debug_inode;
  47 int hammer_debug_locks;
  48 int hammer_debug_btree;
  49 int hammer_debug_tid;
  50 int hammer_debug_recover;               /* -1 will disable, +1 will force */
  51 int hammer_debug_critical;              /* non-zero enter debugger on error */
  52 int hammer_cluster_enable = 2;          /* ena cluster_read, scale x 2 */
  53 int hammer_live_dedup = 0;
  54 int hammer_tdmux_ticks;
  55 int hammer_count_fsyncs;
  56 int hammer_count_inodes;
  57 int hammer_count_iqueued;
  58 int hammer_count_reclaims;
  59 int hammer_count_records;
  60 int hammer_count_record_datas;
  61 int hammer_count_volumes;
  62 int hammer_count_buffers;
  63 int hammer_count_nodes;
  64 int64_t hammer_stats_btree_lookups;
  65 int64_t hammer_stats_btree_searches;
  66 int64_t hammer_stats_btree_inserts;
  67 int64_t hammer_stats_btree_deletes;
  68 int64_t hammer_stats_btree_elements;
  69 int64_t hammer_stats_btree_splits;
  70 int64_t hammer_stats_btree_iterations;
  71 int64_t hammer_stats_btree_root_iterations;
  72 int64_t hammer_stats_record_iterations;
  73
  74 int64_t hammer_stats_file_read;
  75 int64_t hammer_stats_file_write;
  76 int64_t hammer_stats_file_iopsr;
  77 int64_t hammer_stats_file_iopsw;
  78 int64_t hammer_stats_disk_read;
  79 int64_t hammer_stats_disk_write;
  80 int64_t hammer_stats_inode_flushes;
  81 int64_t hammer_stats_commits;
  82 int64_t hammer_stats_undo;
  83 int64_t hammer_stats_redo;
  84
  85 long hammer_count_dirtybufspace;        /* global */
  86 int hammer_count_refedbufs;             /* global */
  87 int hammer_count_reservations;
  88 long hammer_count_io_running_read;
  89 long hammer_count_io_running_write;
  90 int hammer_count_io_locked;
  91 long hammer_limit_dirtybufspace;        /* per-mount */
  92 int hammer_limit_recs;                  /* as a whole XXX */
  93 int hammer_limit_inode_recs = 2048;     /* per inode */
  94 int hammer_limit_reclaims;
  95 int hammer_live_dedup_cache_size = 4096;
  96 int hammer_limit_redo = 4096 * 1024;    /* per inode */
  97 int hammer_autoflush = 500;             /* auto flush (typ on reclaim) */
  98 int hammer_verify_zone;
  99 int hammer_verify_data = 1;
 100 int hammer_double_buffer;
 101 int hammer_btree_full_undo = 1;
 102 int hammer_yield_check = 16;
 103 int hammer_fsync_mode = 3;
 104 int64_t hammer_contention_count;
 105
 106 int hammer_noatime = 1;
 107 TUNABLE_INT("vfs.hammer.noatime", &hammer_noatime);
 108
 109 /*
 110  * Live dedup debug counters (sysctls are writable so that counters
 111  * can be reset from userspace).
 112  */
 113 int64_t hammer_live_dedup_vnode_bcmps = 0;
 114 int64_t hammer_live_dedup_device_bcmps = 0;
 115 int64_t hammer_live_dedup_findblk_failures = 0;
 116 int64_t hammer_live_dedup_bmap_saves = 0;
 117
 118
 119 SYSCTL_NODE(_vfs, OID_AUTO, hammer, CTLFLAG_RW, 0, "HAMMER filesystem");
 120
 121 SYSCTL_INT(_vfs_hammer, OID_AUTO, supported_version, CTLFLAG_RD,
 122            &hammer_supported_version, 0, "");
 123 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_general, CTLFLAG_RW,
 124            &hammer_debug_general, 0, "");
 125 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_io, CTLFLAG_RW,
 126            &hammer_debug_io, 0, "");
 127 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_inode, CTLFLAG_RW,
 128            &hammer_debug_inode, 0, "");
 129 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_locks, CTLFLAG_RW,
 130            &hammer_debug_locks, 0, "");
 131 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_btree, CTLFLAG_RW,
 132            &hammer_debug_btree, 0, "");
 133 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_tid, CTLFLAG_RW,
 134            &hammer_debug_tid, 0, "");
 135 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover, CTLFLAG_RW,
 136            &hammer_debug_recover, 0, "");
 137 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_critical, CTLFLAG_RW,
 138            &hammer_debug_critical, 0, "");
 139 SYSCTL_INT(_vfs_hammer, OID_AUTO, cluster_enable, CTLFLAG_RW,
 140            &hammer_cluster_enable, 0, "");
 141 /*
 142  * 0 - live dedup is disabled
 143  * 1 - dedup cache is populated on reads only
 144  * 2 - dedup cache is populated on both reads and writes
 145  *
 146  * LIVE_DEDUP IS DISABLED PERMANENTLY!  This feature appears to cause
 147  * blockmap corruption over time so we've turned it off permanently.
 148  */
 149 SYSCTL_INT(_vfs_hammer, OID_AUTO, live_dedup, CTLFLAG_RD,
 150            &hammer_live_dedup, 0, "Enable live dedup (experimental)");
 151 SYSCTL_INT(_vfs_hammer, OID_AUTO, tdmux_ticks, CTLFLAG_RW,
 152            &hammer_tdmux_ticks, 0, "Hammer tdmux ticks");
 153
 154 SYSCTL_LONG(_vfs_hammer, OID_AUTO, limit_dirtybufspace, CTLFLAG_RW,
 155            &hammer_limit_dirtybufspace, 0, "");
 156 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_recs, CTLFLAG_RW,
 157            &hammer_limit_recs, 0, "");
 158 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_inode_recs, CTLFLAG_RW,
 159            &hammer_limit_inode_recs, 0, "");
 160 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_reclaims, CTLFLAG_RW,
 161            &hammer_limit_reclaims, 0, "");
 162 SYSCTL_INT(_vfs_hammer, OID_AUTO, live_dedup_cache_size, CTLFLAG_RW,
 163            &hammer_live_dedup_cache_size, 0,
 164            "Number of cache entries");
 165 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_redo, CTLFLAG_RW,
 166            &hammer_limit_redo, 0, "");
 167
 168 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_fsyncs, CTLFLAG_RD,
 169            &hammer_count_fsyncs, 0, "");
 170 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_inodes, CTLFLAG_RD,
 171            &hammer_count_inodes, 0, "");
 172 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_iqueued, CTLFLAG_RD,
 173            &hammer_count_iqueued, 0, "");
 174 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reclaims, CTLFLAG_RD,
 175            &hammer_count_reclaims, 0, "");
 176 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_records, CTLFLAG_RD,
 177            &hammer_count_records, 0, "");
 178 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_record_datas, CTLFLAG_RD,
 179            &hammer_count_record_datas, 0, "");
 180 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_volumes, CTLFLAG_RD,
 181            &hammer_count_volumes, 0, "");
 182 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_buffers, CTLFLAG_RD,
 183            &hammer_count_buffers, 0, "");
 184 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_nodes, CTLFLAG_RD,
 185            &hammer_count_nodes, 0, "");
 186
 187 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_searches, CTLFLAG_RD,
 188            &hammer_stats_btree_searches, 0, "");
 189 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_lookups, CTLFLAG_RD,
 190            &hammer_stats_btree_lookups, 0, "");
 191 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_inserts, CTLFLAG_RD,
 192            &hammer_stats_btree_inserts, 0, "");
 193 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_deletes, CTLFLAG_RD,
 194            &hammer_stats_btree_deletes, 0, "");
 195 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_elements, CTLFLAG_RD,
 196            &hammer_stats_btree_elements, 0, "");
 197 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_splits, CTLFLAG_RD,
 198            &hammer_stats_btree_splits, 0, "");
 199 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_iterations, CTLFLAG_RD,
 200            &hammer_stats_btree_iterations, 0, "");
 201 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_root_iterations, CTLFLAG_RD,
 202            &hammer_stats_btree_root_iterations, 0, "");
 203 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_record_iterations, CTLFLAG_RD,
 204            &hammer_stats_record_iterations, 0, "");
 205
 206 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_read, CTLFLAG_RD,
 207            &hammer_stats_file_read, 0, "");
 208 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_write, CTLFLAG_RD,
 209            &hammer_stats_file_write, 0, "");
 210 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_iopsr, CTLFLAG_RD,
 211            &hammer_stats_file_iopsr, 0, "");
 212 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_iopsw, CTLFLAG_RD,
 213            &hammer_stats_file_iopsw, 0, "");
 214 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_disk_read, CTLFLAG_RD,
 215            &hammer_stats_disk_read, 0, "");
 216 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_disk_write, CTLFLAG_RD,
 217            &hammer_stats_disk_write, 0, "");
 218 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_inode_flushes, CTLFLAG_RD,
 219            &hammer_stats_inode_flushes, 0, "");
 220 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_commits, CTLFLAG_RD,
 221            &hammer_stats_commits, 0, "");
 222 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_undo, CTLFLAG_RD,
 223            &hammer_stats_undo, 0, "");
 224 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_redo, CTLFLAG_RD,
 225            &hammer_stats_redo, 0, "");
 226
 227 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, live_dedup_vnode_bcmps, CTLFLAG_RW,
 228             &hammer_live_dedup_vnode_bcmps, 0,
 229             "successful vnode buffer comparisons");
 230 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, live_dedup_device_bcmps, CTLFLAG_RW,
 231             &hammer_live_dedup_device_bcmps, 0,
 232             "successful device buffer comparisons");
 233 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, live_dedup_findblk_failures, CTLFLAG_RW,
 234             &hammer_live_dedup_findblk_failures, 0,
 235             "block lookup failures for comparison");
 236 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, live_dedup_bmap_saves, CTLFLAG_RW,
 237             &hammer_live_dedup_bmap_saves, 0,
 238             "useful physical block lookups");
 239
 240 SYSCTL_LONG(_vfs_hammer, OID_AUTO, count_dirtybufspace, CTLFLAG_RD,
 241            &hammer_count_dirtybufspace, 0, "");
 242 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_refedbufs, CTLFLAG_RD,
 243            &hammer_count_refedbufs, 0, "");
 244 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reservations, CTLFLAG_RD,
 245            &hammer_count_reservations, 0, "");
 246 SYSCTL_LONG(_vfs_hammer, OID_AUTO, count_io_running_read, CTLFLAG_RD,
 247            &hammer_count_io_running_read, 0, "");
 248 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_locked, CTLFLAG_RD,
 249            &hammer_count_io_locked, 0, "");
 250 SYSCTL_LONG(_vfs_hammer, OID_AUTO, count_io_running_write, CTLFLAG_RD,
 251            &hammer_count_io_running_write, 0, "");
 252 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, contention_count, CTLFLAG_RW,
 253            &hammer_contention_count, 0, "");
 254 SYSCTL_INT(_vfs_hammer, OID_AUTO, autoflush, CTLFLAG_RW,
 255            &hammer_autoflush, 0, "");
 256 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_zone, CTLFLAG_RW,
 257            &hammer_verify_zone, 0, "");
 258 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_data, CTLFLAG_RW,
 259            &hammer_verify_data, 0, "");
 260 SYSCTL_INT(_vfs_hammer, OID_AUTO, double_buffer, CTLFLAG_RW,
 261            &hammer_double_buffer, 0, "");
 262 SYSCTL_INT(_vfs_hammer, OID_AUTO, btree_full_undo, CTLFLAG_RW,
 263            &hammer_btree_full_undo, 0, "");
 264 SYSCTL_INT(_vfs_hammer, OID_AUTO, yield_check, CTLFLAG_RW,
 265            &hammer_yield_check, 0, "");
 266 SYSCTL_INT(_vfs_hammer, OID_AUTO, fsync_mode, CTLFLAG_RW,
 267            &hammer_fsync_mode, 0, "");
 268
 269 /* KTR_INFO_MASTER(hammer); */
 270
 271 /*
 272  * VFS ABI
 273  */
 274 static void     hammer_free_hmp(struct mount *mp);
 275
 276 static int      hammer_vfs_mount(struct mount *mp, char *path, caddr_t data,
 277                                 struct ucred *cred);
 278 static int      hammer_vfs_unmount(struct mount *mp, int mntflags);
 279 static int      hammer_vfs_root(struct mount *mp, struct vnode **vpp);
 280 static int      hammer_vfs_statfs(struct mount *mp, struct statfs *sbp,
 281                                 struct ucred *cred);
 282 static int      hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp,
 283                                 struct ucred *cred);
 284 static int      hammer_vfs_sync(struct mount *mp, int waitfor);
 285 static int      hammer_vfs_vget(struct mount *mp, struct vnode *dvp,
 286                                 ino_t ino, struct vnode **vpp);
 287 static int      hammer_vfs_init(struct vfsconf *conf);
 288 static int      hammer_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
 289                                 struct fid *fhp, struct vnode **vpp);
 290 static int      hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp);
 291 static int      hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
 292                                 int *exflagsp, struct ucred **credanonp);
 293
 294
 295 static struct vfsops hammer_vfsops = {
 296         .vfs_mount      = hammer_vfs_mount,
 297         .vfs_unmount    = hammer_vfs_unmount,
 298         .vfs_root       = hammer_vfs_root,
 299         .vfs_statfs     = hammer_vfs_statfs,
 300         .vfs_statvfs    = hammer_vfs_statvfs,
 301         .vfs_sync       = hammer_vfs_sync,
 302         .vfs_vget       = hammer_vfs_vget,
 303         .vfs_init       = hammer_vfs_init,
 304         .vfs_vptofh     = hammer_vfs_vptofh,
 305         .vfs_fhtovp     = hammer_vfs_fhtovp,
 306         .vfs_checkexp   = hammer_vfs_checkexp
 307 };
 308
 309 MALLOC_DEFINE(M_HAMMER, "HAMMER-mount", "");
 310
 311 VFS_SET(hammer_vfsops, hammer, 0);
 312 MODULE_VERSION(hammer, 1);
 313
 314 static int
 315 hammer_vfs_init(struct vfsconf *conf)
 316 {
 317         long n;
 318
 319         /*
 320          * Wait up to this long for an exclusive deadlock to clear
 321          * before acquiring a new shared lock on the ip.  The deadlock
 322          * may have occured on a b-tree node related to the ip.
 323          */
 324         if (hammer_tdmux_ticks == 0)
 325                 hammer_tdmux_ticks = hz / 5;
 326
 327         /*
 328          * Autosize, but be careful because a hammer filesystem's
 329          * reserve is partially calculated based on dirtybufspace,
 330          * so we simply cannot allow it to get too large.
 331          */
 332         if (hammer_limit_recs == 0) {
 333                 n = nbuf * 25;
 334                 if (n > kmalloc_limit(M_HAMMER) / 512)
 335                         n = kmalloc_limit(M_HAMMER) / 512;
 336                 if (n > 2 * 1024 * 1024)
 337                         n = 2 * 1024 * 1024;
 338                 hammer_limit_recs = (int)n;
 339         }
 340         if (hammer_limit_dirtybufspace == 0) {
 341                 hammer_limit_dirtybufspace = hidirtybufspace / 2;
 342                 if (hammer_limit_dirtybufspace < 1L * 1024 * 1024)
 343                         hammer_limit_dirtybufspace = 1024L * 1024;
 344                 if (hammer_limit_dirtybufspace > 1024L * 1024 * 1024)
 345                         hammer_limit_dirtybufspace = 1024L * 1024 * 1024;
 346         }
 347
 348         /*
 349          * The hammer_inode structure detaches from the vnode on reclaim.
 350          * This limits the number of inodes in this state to prevent a
 351          * memory pool blowout.
 352          */
 353         if (hammer_limit_reclaims == 0) {
 354                 hammer_limit_reclaims = maxvnodes / 10;
 355                 if (hammer_limit_reclaims > HAMMER_LIMIT_RECLAIMS)
 356                         hammer_limit_reclaims = HAMMER_LIMIT_RECLAIMS;
 357         }
 358
 359         return(0);
 360 }
 361
 362 static int
 363 hammer_vfs_mount(struct mount *mp, char *mntpt, caddr_t data,
 364                  struct ucred *cred)
 365 {
 366         struct hammer_mount_info info;
 367         hammer_mount_t hmp;
 368         hammer_volume_t rootvol;
 369         struct vnode *rootvp;
 370         struct vnode *devvp = NULL;
 371         const char *upath;      /* volume name in userspace */
 372         char *path;             /* volume name in system space */
 373         int error;
 374         int i;
 375         int master_id;
 376         int nvolumes;
 377         char *next_volume_ptr = NULL;
 378
 379         if (hammer_noatime) {
 380                 /* Force noatime */
 381                 mp->mnt_flag |= MNT_NOATIME;
 382         }
 383
 384         /*
 385          * Accept hammer_mount_info.  mntpt is NULL for root mounts at boot.
 386          */
 387         if (mntpt == NULL) {
 388                 bzero(&info, sizeof(info));
 389                 info.asof = 0;
 390                 info.hflags = 0;
 391                 info.nvolumes = 1;
 392
 393                 next_volume_ptr = mp->mnt_stat.f_mntfromname;
 394
 395                 /* Count number of volumes separated by ':' */
 396                 for (char *p = next_volume_ptr; *p != '\0'; ++p) {
 397                         if (*p == ':') {
 398                                 ++info.nvolumes;
 399                         }
 400                 }
 401
 402                 mp->mnt_flag &= ~MNT_RDONLY; /* mount R/W */
 403         } else {
 404                 if ((error = copyin(data, &info, sizeof(info))) != 0)
 405                         return (error);
 406         }
 407
 408         /*
 409          * updating or new mount
 410          */
 411         if (mp->mnt_flag & MNT_UPDATE) {
 412                 hmp = (void *)mp->mnt_data;
 413                 KKASSERT(hmp != NULL);
 414         } else {
 415                 if (info.nvolumes <= 0 || info.nvolumes > HAMMER_MAX_VOLUMES)
 416                         return (EINVAL);
 417                 hmp = NULL;
 418         }
 419
 420         /*
 421          * master-id validation.  The master id may not be changed by a
 422          * mount update.
 423          */
 424         if (info.hflags & HMNT_MASTERID || info.hflags & HMNT_NOMIRROR) {
 425                 if (hmp && hmp->master_id != info.master_id) {
 426                         hkprintf("cannot change master id with mount update\n");
 427                         return(EINVAL);
 428                 }
 429                 master_id = info.master_id;
 430                 if (master_id < -1 || master_id >= HAMMER_MAX_MASTERS)
 431                         return (EINVAL);
 432         } else {
 433                 if (hmp)
 434                         master_id = hmp->master_id;
 435                 else
 436                         master_id = 0;
 437         }
 438
 439         /*
 440          * Internal mount data structure
 441          */
 442         if (hmp == NULL) {
 443                 hmp = kmalloc(sizeof(*hmp), M_HAMMER, M_WAITOK | M_ZERO);
 444                 mp->mnt_data = (qaddr_t)hmp;
 445                 hmp->mp = mp;
 446
 447                 /*
 448                  * Make sure kmalloc type limits are set appropriately.
 449                  *
 450                  * Our inode kmalloc group is sized based on maxvnodes
 451                  * (controlled by the system, not us).
 452                  */
 453                 kmalloc_create(&hmp->m_misc, "HAMMER-others");
 454                 kmalloc_create(&hmp->m_inodes, "HAMMER-inodes");
 455
 456                 kmalloc_raise_limit(hmp->m_inodes, 0);  /* unlimited */
 457
 458                 hmp->root_btree_beg.localization =
 459                         HAMMER_MIN_ONDISK_LOCALIZATION;
 460                 hmp->root_btree_beg.obj_id = HAMMER_MIN_OBJID;
 461                 hmp->root_btree_beg.key = HAMMER_MIN_KEY;
 462                 hmp->root_btree_beg.create_tid = 1;
 463                 hmp->root_btree_beg.delete_tid = 1;
 464                 hmp->root_btree_beg.rec_type = HAMMER_MIN_RECTYPE;
 465                 hmp->root_btree_beg.obj_type = 0;
 466                 hmp->root_btree_beg.btype = HAMMER_BTREE_TYPE_NONE;
 467
 468                 hmp->root_btree_end.localization =
 469                         HAMMER_MAX_ONDISK_LOCALIZATION;
 470                 hmp->root_btree_end.obj_id = HAMMER_MAX_OBJID;
 471                 hmp->root_btree_end.key = HAMMER_MAX_KEY;
 472                 hmp->root_btree_end.create_tid = HAMMER_MAX_TID;
 473                 hmp->root_btree_end.delete_tid = 0;   /* special case */
 474                 hmp->root_btree_end.rec_type = HAMMER_MAX_RECTYPE;
 475                 hmp->root_btree_end.obj_type = 0;
 476                 hmp->root_btree_end.btype = HAMMER_BTREE_TYPE_NONE;
 477
 478                 hmp->krate.freq = 1;    /* maximum reporting rate (hz) */
 479                 hmp->krate.count = -16; /* initial burst */
 480                 hmp->kdiag.freq = 1;    /* maximum reporting rate (hz) */
 481                 hmp->kdiag.count = -16; /* initial burst */
 482
 483                 hmp->sync_lock.refs = 1;
 484                 hmp->undo_lock.refs = 1;
 485                 hmp->blkmap_lock.refs = 1;
 486                 hmp->snapshot_lock.refs = 1;
 487                 hmp->volume_lock.refs = 1;
 488
 489                 TAILQ_INIT(&hmp->delay_list);
 490                 TAILQ_INIT(&hmp->flush_group_list);
 491                 TAILQ_INIT(&hmp->objid_cache_list);
 492                 TAILQ_INIT(&hmp->undo_lru_list);
 493                 TAILQ_INIT(&hmp->reclaim_list);
 494
 495                 RB_INIT(&hmp->rb_dedup_crc_root);
 496                 RB_INIT(&hmp->rb_dedup_off_root);
 497                 TAILQ_INIT(&hmp->dedup_lru_list);
 498         }
 499         hmp->hflags &= ~HMNT_USERFLAGS;
 500         hmp->hflags |= info.hflags & HMNT_USERFLAGS;
 501
 502         hmp->master_id = master_id;
 503
 504         if (info.asof) {
 505                 mp->mnt_flag |= MNT_RDONLY;
 506                 hmp->asof = info.asof;
 507         } else {
 508                 hmp->asof = HAMMER_MAX_TID;
 509         }
 510
 511         hmp->volume_to_remove = -1;
 512
 513         /*
 514          * Re-open read-write if originally read-only, or vise-versa.
 515          *
 516          * When going from read-only to read-write execute the stage2
 517          * recovery if it has not already been run.
 518          */
 519         if (mp->mnt_flag & MNT_UPDATE) {
 520                 lwkt_gettoken(&hmp->fs_token);
 521                 error = 0;
 522                 if (hmp->ronly && (mp->mnt_kern_flag & MNTK_WANTRDWR)) {
 523                         hkprintf("read-only -> read-write\n");
 524                         hmp->ronly = 0;
 525                         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
 526                                 hammer_adjust_volume_mode, NULL);
 527                         rootvol = hammer_get_root_volume(hmp, &error);
 528                         if (rootvol) {
 529                                 hammer_recover_flush_buffers(hmp, rootvol, 1);
 530                                 error = hammer_recover_stage2(hmp, rootvol);
 531                                 bcopy(rootvol->ondisk->vol0_blockmap,
 532                                       hmp->blockmap,
 533                                       sizeof(hmp->blockmap));
 534                                 hammer_rel_volume(rootvol, 0);
 535                         }
 536                         RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
 537                                 hammer_reload_inode, NULL);
 538                         /* kernel clears MNT_RDONLY */
 539                 } else if (hmp->ronly == 0 && (mp->mnt_flag & MNT_RDONLY)) {
 540                         hkprintf("read-write -> read-only\n");
 541                         hmp->ronly = 1; /* messy */
 542                         RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
 543                                 hammer_reload_inode, NULL);
 544                         hmp->ronly = 0;
 545                         hammer_flusher_sync(hmp);
 546                         hammer_flusher_sync(hmp);
 547                         hammer_flusher_sync(hmp);
 548                         hmp->ronly = 1;
 549                         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
 550                                 hammer_adjust_volume_mode, NULL);
 551                 }
 552                 lwkt_reltoken(&hmp->fs_token);
 553                 return(error);
 554         }
 555
 556         RB_INIT(&hmp->rb_vols_root);
 557         RB_INIT(&hmp->rb_inos_root);
 558         RB_INIT(&hmp->rb_redo_root);
 559         RB_INIT(&hmp->rb_nods_root);
 560         RB_INIT(&hmp->rb_undo_root);
 561         RB_INIT(&hmp->rb_resv_root);
 562         RB_INIT(&hmp->rb_bufs_root);
 563         RB_INIT(&hmp->rb_pfsm_root);
 564
 565         hmp->ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
 566
 567         RB_INIT(&hmp->volu_root);
 568         RB_INIT(&hmp->undo_root);
 569         RB_INIT(&hmp->data_root);
 570         RB_INIT(&hmp->meta_root);
 571         RB_INIT(&hmp->lose_root);
 572         TAILQ_INIT(&hmp->iorun_list);
 573
 574         lwkt_token_init(&hmp->fs_token, "hammerfs");
 575         lwkt_token_init(&hmp->io_token, "hammerio");
 576
 577         lwkt_gettoken(&hmp->fs_token);
 578
 579         /*
 580          * Load volumes
 581          */
 582         path = objcache_get(namei_oc, M_WAITOK);
 583         hmp->nvolumes = -1;
 584         for (i = 0; i < info.nvolumes; ++i) {
 585                 if (mntpt == NULL) {
 586                         /*
 587                          * Root mount.
 588                          */
 589                         KKASSERT(next_volume_ptr != NULL);
 590                         strcpy(path, "");
 591                         if (*next_volume_ptr != '/') {
 592                                 /* relative path */
 593                                 strcpy(path, "/dev/");
 594                         }
 595                         int k;
 596                         for (k = strlen(path); k < MAXPATHLEN-1; ++k) {
 597                                 if (*next_volume_ptr == '\0') {
 598                                         break;
 599                                 } else if (*next_volume_ptr == ':') {
 600                                         ++next_volume_ptr;
 601                                         break;
 602                                 } else {
 603                                         path[k] = *next_volume_ptr;
 604                                         ++next_volume_ptr;
 605                                 }
 606                         }
 607                         path[k] = '\0';
 608
 609                         error = 0;
 610                         cdev_t dev = kgetdiskbyname(path);
 611                         error = bdevvp(dev, &devvp);
 612                         if (error) {
 613                                 hdkprintf("can't find devvp\n");
 614                         }
 615                 } else {
 616                         error = copyin(&info.volumes[i], &upath,
 617                                        sizeof(char *));
 618                         if (error == 0)
 619                                 error = copyinstr(upath, path,
 620                                                   MAXPATHLEN, NULL);
 621                 }
 622                 if (error == 0)
 623                         error = hammer_install_volume(hmp, path, devvp, NULL);
 624                 if (error)
 625                         break;
 626         }
 627         objcache_put(namei_oc, path);
 628
 629         /*
 630          * Make sure we found a root volume
 631          */
 632         if (hmp->rootvol == NULL) {
 633                 if (error == EBUSY) {
 634                         hdkprintf("The volumes are probably mounted\n");
 635                 } else {
 636                         hdkprintf("No root volume found!\n");
 637                         error = EINVAL;
 638                 }
 639                 goto failed;
 640         }
 641
 642         /*
 643          * Check that all required volumes are available
 644          */
 645         if (error == 0 && hammer_mountcheck_volumes(hmp)) {
 646                 hdkprintf("Missing volumes, cannot mount!\n");
 647                 error = EINVAL;
 648                 goto failed;
 649         }
 650
 651         /*
 652          * Other errors
 653          */
 654         if (error) {
 655                 hdkprintf("Failed to load volumes!\n");
 656                 goto failed;
 657         }
 658
 659         nvolumes = hammer_get_installed_volumes(hmp);
 660         if (hmp->nvolumes != nvolumes) {
 661                 hdkprintf("volume header says %d volumes, but %d installed\n",
 662                         hmp->nvolumes, nvolumes);
 663                 error = EINVAL;
 664                 goto failed;
 665         }
 666
 667         /*
 668          * No errors, setup enough of the mount point so we can lookup the
 669          * root vnode.
 670          */
 671         mp->mnt_iosize_max = MAXPHYS;
 672         mp->mnt_kern_flag |= MNTK_FSMID;
 673         mp->mnt_kern_flag |= MNTK_THR_SYNC;     /* new vsyncscan semantics */
 674
 675         /*
 676          * MPSAFE code.  Note that VOPs and VFSops which are not MPSAFE
 677          * will acquire a per-mount token prior to entry and release it
 678          * on return.
 679          */
 680         mp->mnt_kern_flag |= MNTK_ALL_MPSAFE;
 681
 682         /*
 683          * note: f_iosize is used by vnode_pager_haspage() when constructing
 684          * its VOP_BMAP call.
 685          */
 686         mp->mnt_stat.f_iosize = HAMMER_BUFSIZE;
 687         mp->mnt_stat.f_bsize = HAMMER_BUFSIZE;
 688
 689         mp->mnt_vstat.f_frsize = HAMMER_BUFSIZE;
 690         mp->mnt_vstat.f_bsize = HAMMER_BUFSIZE;
 691
 692         mp->mnt_maxsymlinklen = 255;
 693         mp->mnt_flag |= MNT_LOCAL;
 694
 695         vfs_add_vnodeops(mp, &hammer_vnode_vops, &mp->mnt_vn_norm_ops);
 696         vfs_add_vnodeops(mp, &hammer_spec_vops, &mp->mnt_vn_spec_ops);
 697         vfs_add_vnodeops(mp, &hammer_fifo_vops, &mp->mnt_vn_fifo_ops);
 698
 699         /*
 700          * The root volume's ondisk pointer is only valid if we hold a
 701          * reference to it.
 702          */
 703         rootvol = hammer_get_root_volume(hmp, &error);
 704         if (error)
 705                 goto failed;
 706
 707         /*
 708          * Perform any necessary UNDO operations.  The recovery code does
 709          * call hammer_undo_lookup() so we have to pre-cache the blockmap,
 710          * and then re-copy it again after recovery is complete.
 711          *
 712          * If this is a read-only mount the UNDO information is retained
 713          * in memory in the form of dirty buffer cache buffers, and not
 714          * written back to the media.
 715          */
 716         bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
 717               sizeof(hmp->blockmap));
 718
 719         /*
 720          * Check filesystem version
 721          */
 722         hmp->version = rootvol->ondisk->vol_version;
 723         if (hmp->version < HAMMER_VOL_VERSION_MIN ||
 724             hmp->version > HAMMER_VOL_VERSION_MAX) {
 725                 hkprintf("mount unsupported fs version %d\n", hmp->version);
 726                 error = ERANGE;
 727                 goto done;
 728         }
 729
 730         /*
 731          * The undo_rec_limit limits the size of flush groups to avoid
 732          * blowing out the UNDO FIFO.  This calculation is typically in
 733          * the tens of thousands and is designed primarily when small
 734          * HAMMER filesystems are created.
 735          */
 736         hmp->undo_rec_limit = hammer_undo_max(hmp) / 8192 + 100;
 737         if (hammer_debug_general & 0x0001)
 738                 hkprintf("undo_rec_limit %d\n", hmp->undo_rec_limit);
 739
 740         /*
 741          * NOTE: Recover stage1 not only handles meta-data recovery, it
 742          *       also sets hmp->undo_seqno for HAMMER VERSION 4+ filesystems.
 743          */
 744         error = hammer_recover_stage1(hmp, rootvol);
 745         if (error) {
 746                 kprintf("Failed to recover HAMMER filesystem on mount\n");
 747                 goto done;
 748         }
 749
 750         /*
 751          * Finish setup now that we have a good root volume.
 752          */
 753         ksnprintf(mp->mnt_stat.f_mntfromname,
 754                   sizeof(mp->mnt_stat.f_mntfromname), "%s",
 755                   rootvol->ondisk->vol_label);
 756         mp->mnt_stat.f_fsid.val[0] =
 757                 crc32((char *)&rootvol->ondisk->vol_fsid + 0, 8);
 758         mp->mnt_stat.f_fsid.val[1] =
 759                 crc32((char *)&rootvol->ondisk->vol_fsid + 8, 8);
 760         mp->mnt_stat.f_fsid.val[1] &= HAMMER_LOCALIZE_MASK;
 761
 762         mp->mnt_vstat.f_fsid_uuid = rootvol->ondisk->vol_fsid;
 763         mp->mnt_vstat.f_fsid = crc32(&mp->mnt_vstat.f_fsid_uuid,
 764                                      sizeof(mp->mnt_vstat.f_fsid_uuid));
 765
 766         /*
 767          * Certain often-modified fields in the root volume are cached in
 768          * the hammer_mount structure so we do not have to generate lots
 769          * of little UNDO structures for them.
 770          *
 771          * Recopy after recovery.  This also has the side effect of
 772          * setting our cached undo FIFO's first_offset, which serves to
 773          * placemark the FIFO start for the NEXT flush cycle while the
 774          * on-disk first_offset represents the LAST flush cycle.
 775          */
 776         hmp->next_tid = rootvol->ondisk->vol0_next_tid;
 777         hmp->flush_tid1 = hmp->next_tid;
 778         hmp->flush_tid2 = hmp->next_tid;
 779         bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
 780               sizeof(hmp->blockmap));
 781         hmp->copy_stat_freebigblocks = rootvol->ondisk->vol0_stat_freebigblocks;
 782
 783         hammer_flusher_create(hmp);
 784
 785         /*
 786          * Locate the root directory with an obj_id of 1.
 787          */
 788         error = hammer_vfs_root(mp, &rootvp);
 789         if (error)
 790                 goto done;
 791         vput(rootvp);
 792         if (hmp->ronly == 0)
 793                 error = hammer_recover_stage2(hmp, rootvol);
 794
 795         /*
 796          * If the stage2 recovery fails be sure to clean out all cached
 797          * vnodes before throwing away the mount structure or bad things
 798          * will happen.
 799          */
 800         if (error)
 801                 vflush(mp, 0, 0);
 802
 803 done:
 804         if ((mp->mnt_flag & MNT_UPDATE) == 0) {
 805                 /* New mount */
 806
 807                 /* Populate info for mount point (NULL pad)*/
 808                 bzero(mp->mnt_stat.f_mntonname, MNAMELEN);
 809                 size_t size;
 810                 if (mntpt) {
 811                         copyinstr(mntpt, mp->mnt_stat.f_mntonname,
 812                                                         MNAMELEN -1, &size);
 813                 } else { /* Root mount */
 814                         mp->mnt_stat.f_mntonname[0] = '/';
 815                 }
 816         }
 817         (void)VFS_STATFS(mp, &mp->mnt_stat, cred);
 818         hammer_rel_volume(rootvol, 0);
 819 failed:
 820         /*
 821          * Cleanup and return.
 822          */
 823         if (error) {
 824                 /* called with fs_token held */
 825                 hammer_free_hmp(mp);
 826         } else {
 827                 lwkt_reltoken(&hmp->fs_token);
 828         }
 829         return (error);
 830 }
 831
 832 static int
 833 hammer_vfs_unmount(struct mount *mp, int mntflags)
 834 {
 835         hammer_mount_t hmp = (void *)mp->mnt_data;
 836         int flags;
 837         int error;
 838
 839         /*
 840          * Clean out the vnodes
 841          */
 842         lwkt_gettoken(&hmp->fs_token);
 843         flags = 0;
 844         if (mntflags & MNT_FORCE)
 845                 flags |= FORCECLOSE;
 846         error = vflush(mp, 0, flags);
 847
 848         /*
 849          * Clean up the internal mount structure and related entities.  This
 850          * may issue I/O.
 851          */
 852         if (error == 0) {
 853                 /* called with fs_token held */
 854                 hammer_free_hmp(mp);
 855         } else {
 856                 lwkt_reltoken(&hmp->fs_token);
 857         }
 858         return(error);
 859 }
 860
 861 /*
 862  * Clean up the internal mount structure and disassociate it from the mount.
 863  * This may issue I/O.
 864  *
 865  * Called with fs_token held.
 866  */
 867 static void
 868 hammer_free_hmp(struct mount *mp)
 869 {
 870         hammer_mount_t hmp = (void *)mp->mnt_data;
 871         hammer_flush_group_t flg;
 872
 873         /*
 874          * Flush anything dirty.  This won't even run if the
 875          * filesystem errored-out.
 876          */
 877         hammer_flush_dirty(hmp, 30);
 878
 879         /*
 880          * If the mount had a critical error we have to destroy any
 881          * remaining inodes before we can finish cleaning up the flusher.
 882          */
 883         if (hmp->flags & HAMMER_MOUNT_CRITICAL_ERROR) {
 884                 RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
 885                         hammer_destroy_inode_callback, NULL);
 886         }
 887
 888         /*
 889          * There shouldn't be any inodes left now and any left over
 890          * flush groups should now be empty.
 891          */
 892         KKASSERT(RB_EMPTY(&hmp->rb_inos_root));
 893         while ((flg = TAILQ_FIRST(&hmp->flush_group_list)) != NULL) {
 894                 TAILQ_REMOVE(&hmp->flush_group_list, flg, flush_entry);
 895                 KKASSERT(RB_EMPTY(&flg->flush_tree));
 896                 if (flg->refs) {
 897                         hkprintf("Warning, flush_group %p was "
 898                                 "not empty on umount!\n", flg);
 899                 }
 900                 kfree(flg, hmp->m_misc);
 901         }
 902
 903         /*
 904          * We can finally destroy the flusher
 905          */
 906         hammer_flusher_destroy(hmp);
 907
 908         /*
 909          * We may have held recovered buffers due to a read-only mount.
 910          * These must be discarded.
 911          */
 912         if (hmp->ronly)
 913                 hammer_recover_flush_buffers(hmp, NULL, -1);
 914
 915         /*
 916          * Unload buffers and then volumes
 917          */
 918         RB_SCAN(hammer_buf_rb_tree, &hmp->rb_bufs_root, NULL,
 919                 hammer_unload_buffer, NULL);
 920         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
 921                 hammer_unload_volume, NULL);
 922
 923         mp->mnt_data = NULL;
 924         mp->mnt_flag &= ~MNT_LOCAL;
 925         hmp->mp = NULL;
 926         hammer_destroy_objid_cache(hmp);
 927         hammer_destroy_dedup_cache(hmp);
 928         if (hmp->dedup_free_cache != NULL) {
 929                 kfree(hmp->dedup_free_cache, hmp->m_misc);
 930                 hmp->dedup_free_cache = NULL;
 931         }
 932         kmalloc_destroy(&hmp->m_misc);
 933         kmalloc_destroy(&hmp->m_inodes);
 934         lwkt_reltoken(&hmp->fs_token);
 935         kfree(hmp, M_HAMMER);
 936 }
 937
 938 /*
 939  * Report critical errors.  ip may be NULL.
 940  */
 941 void
 942 hammer_critical_error(hammer_mount_t hmp, hammer_inode_t ip,
 943                       int error, const char *msg)
 944 {
 945         hmp->flags |= HAMMER_MOUNT_CRITICAL_ERROR;
 946
 947         hmkrateprintf(&hmp->krate, hmp,
 948                     "Critical error inode=%jd error=%d %s\n",
 949                     (intmax_t)(ip ? ip->obj_id : -1),
 950                     error, msg);
 951
 952         if (hmp->ronly == 0) {
 953                 hmp->ronly = 2;         /* special errored read-only mode */
 954                 hmp->mp->mnt_flag |= MNT_RDONLY;
 955                 RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
 956                         hammer_adjust_volume_mode, NULL);
 957                 hmkprintf(hmp, "Forcing read-only mode\n");
 958         }
 959         hmp->error = error;
 960         if (hammer_debug_critical)
 961                 Debugger("Entering debugger");
 962 }
 963
 964
 965 /*
 966  * Obtain a vnode for the specified inode number.  An exclusively locked
 967  * vnode is returned.
 968  */
 969 int
 970 hammer_vfs_vget(struct mount *mp, struct vnode *dvp,
 971                 ino_t ino, struct vnode **vpp)
 972 {
 973         struct hammer_transaction trans;
 974         hammer_mount_t hmp = (void *)mp->mnt_data;
 975         hammer_inode_t ip;
 976         int error;
 977         uint32_t localization;
 978
 979         lwkt_gettoken(&hmp->fs_token);
 980         hammer_simple_transaction(&trans, hmp);
 981
 982         /*
 983          * If a directory vnode is supplied (mainly NFS) then we can acquire
 984          * the PFS domain from it.  Otherwise we would only be able to vget
 985          * inodes in the root PFS.
 986          */
 987         if (dvp) {
 988                 localization = HAMMER_DEF_LOCALIZATION |
 989                                 VTOI(dvp)->obj_localization;
 990         } else {
 991                 localization = HAMMER_DEF_LOCALIZATION;
 992         }
 993
 994         /*
 995          * Lookup the requested HAMMER inode.  The structure must be
 996          * left unlocked while we manipulate the related vnode to avoid
 997          * a deadlock.
 998          */
 999         ip = hammer_get_inode(&trans, NULL, ino,
1000                               hmp->asof, localization,
1001                               0, &error);
1002         if (ip == NULL) {
1003                 *vpp = NULL;
1004         } else {
1005                 error = hammer_get_vnode(ip, vpp);
1006                 hammer_rel_inode(ip, 0);
1007         }
1008         hammer_done_transaction(&trans);
1009         lwkt_reltoken(&hmp->fs_token);
1010         return (error);
1011 }
1012
1013 /*
1014  * Return the root vnode for the filesystem.
1015  *
1016  * HAMMER stores the root vnode in the hammer_mount structure so
1017  * getting it is easy.
1018  */
1019 static int
1020 hammer_vfs_root(struct mount *mp, struct vnode **vpp)
1021 {
1022         int error;
1023
1024         error = hammer_vfs_vget(mp, NULL, HAMMER_OBJID_ROOT, vpp);
1025         return (error);
1026 }
1027
1028 static int
1029 hammer_vfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
1030 {
1031         hammer_mount_t hmp = (void *)mp->mnt_data;
1032         hammer_volume_t volume;
1033         hammer_volume_ondisk_t ondisk;
1034         int error;
1035         int64_t bfree;
1036         int64_t breserved;
1037
1038         lwkt_gettoken(&hmp->fs_token);
1039         volume = hammer_get_root_volume(hmp, &error);
1040         if (error) {
1041                 lwkt_reltoken(&hmp->fs_token);
1042                 return(error);
1043         }
1044         ondisk = volume->ondisk;
1045
1046         /*
1047          * Basic stats
1048          */
1049         _hammer_checkspace(hmp, HAMMER_CHKSPC_WRITE, &breserved);
1050         mp->mnt_stat.f_files = ondisk->vol0_stat_inodes;
1051         bfree = ondisk->vol0_stat_freebigblocks * HAMMER_BIGBLOCK_SIZE;
1052         hammer_rel_volume(volume, 0);
1053
1054         if (breserved > bfree)
1055                 breserved = bfree;
1056         mp->mnt_stat.f_bfree = (bfree - breserved) / HAMMER_BUFSIZE;
1057         mp->mnt_stat.f_bavail = mp->mnt_stat.f_bfree;
1058         if (mp->mnt_stat.f_files < 0)
1059                 mp->mnt_stat.f_files = 0;
1060
1061         *sbp = mp->mnt_stat;
1062         lwkt_reltoken(&hmp->fs_token);
1063         return(0);
1064 }
1065
1066 static int
1067 hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred)
1068 {
1069         hammer_mount_t hmp = (void *)mp->mnt_data;
1070         hammer_volume_t volume;
1071         hammer_volume_ondisk_t ondisk;
1072         int error;
1073         int64_t bfree;
1074         int64_t breserved;
1075
1076         lwkt_gettoken(&hmp->fs_token);
1077         volume = hammer_get_root_volume(hmp, &error);
1078         if (error) {
1079                 lwkt_reltoken(&hmp->fs_token);
1080                 return(error);
1081         }
1082         ondisk = volume->ondisk;
1083
1084         /*
1085          * Basic stats
1086          */
1087         _hammer_checkspace(hmp, HAMMER_CHKSPC_WRITE, &breserved);
1088         mp->mnt_vstat.f_files = ondisk->vol0_stat_inodes;
1089         bfree = ondisk->vol0_stat_freebigblocks * HAMMER_BIGBLOCK_SIZE;
1090         hammer_rel_volume(volume, 0);
1091
1092         if (breserved > bfree)
1093                 breserved = bfree;
1094         mp->mnt_vstat.f_bfree = (bfree - breserved) / HAMMER_BUFSIZE;
1095         mp->mnt_vstat.f_bavail = mp->mnt_vstat.f_bfree;
1096         if (mp->mnt_vstat.f_files < 0)
1097                 mp->mnt_vstat.f_files = 0;
1098         *sbp = mp->mnt_vstat;
1099         lwkt_reltoken(&hmp->fs_token);
1100         return(0);
1101 }
1102
1103 /*
1104  * Sync the filesystem.  Currently we have to run it twice, the second
1105  * one will advance the undo start index to the end index, so if a crash
1106  * occurs no undos will be run on mount.
1107  *
1108  * We do not sync the filesystem if we are called from a panic.  If we did
1109  * we might end up blowing up a sync that was already in progress.
1110  */
1111 static int
1112 hammer_vfs_sync(struct mount *mp, int waitfor)
1113 {
1114         hammer_mount_t hmp = (void *)mp->mnt_data;
1115         int error;
1116
1117         lwkt_gettoken(&hmp->fs_token);
1118         if (panicstr == NULL) {
1119                 error = hammer_sync_hmp(hmp, waitfor);
1120         } else {
1121                 error = EIO;
1122         }
1123         lwkt_reltoken(&hmp->fs_token);
1124         return (error);
1125 }
1126
1127 /*
1128  * Convert a vnode to a file handle.
1129  *
1130  * Accesses read-only fields on already-referenced structures so
1131  * no token is needed.
1132  */
1133 static int
1134 hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp)
1135 {
1136         hammer_inode_t ip;
1137
1138         KKASSERT(MAXFIDSZ >= 16);
1139         ip = VTOI(vp);
1140         fhp->fid_len = offsetof(struct fid, fid_data[16]);
1141         fhp->fid_ext = lo_to_pfs(ip->obj_localization);
1142         bcopy(&ip->obj_id, fhp->fid_data + 0, sizeof(ip->obj_id));
1143         bcopy(&ip->obj_asof, fhp->fid_data + 8, sizeof(ip->obj_asof));
1144         return(0);
1145 }
1146
1147
1148 /*
1149  * Convert a file handle back to a vnode.
1150  *
1151  * Use rootvp to enforce PFS isolation when a PFS is exported via a
1152  * null mount.
1153  */
1154 static int
1155 hammer_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
1156                   struct fid *fhp, struct vnode **vpp)
1157 {
1158         hammer_mount_t hmp = (void *)mp->mnt_data;
1159         struct hammer_transaction trans;
1160         hammer_inode_t ip;
1161         struct hammer_inode_info info;
1162         int error;
1163         uint32_t localization;
1164
1165         bcopy(fhp->fid_data + 0, &info.obj_id, sizeof(info.obj_id));
1166         bcopy(fhp->fid_data + 8, &info.obj_asof, sizeof(info.obj_asof));
1167         if (rootvp)
1168                 localization = VTOI(rootvp)->obj_localization;
1169         else
1170                 localization = pfs_to_lo(fhp->fid_ext);
1171
1172         lwkt_gettoken(&hmp->fs_token);
1173         hammer_simple_transaction(&trans, hmp);
1174
1175         /*
1176          * Get/allocate the hammer_inode structure.  The structure must be
1177          * unlocked while we manipulate the related vnode to avoid a
1178          * deadlock.
1179          */
1180         ip = hammer_get_inode(&trans, NULL, info.obj_id,
1181                               info.obj_asof, localization, 0, &error);
1182         if (ip) {
1183                 error = hammer_get_vnode(ip, vpp);
1184                 hammer_rel_inode(ip, 0);
1185         } else {
1186                 *vpp = NULL;
1187         }
1188         hammer_done_transaction(&trans);
1189         lwkt_reltoken(&hmp->fs_token);
1190         return (error);
1191 }
1192
1193 static int
1194 hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
1195                     int *exflagsp, struct ucred **credanonp)
1196 {
1197         hammer_mount_t hmp = (void *)mp->mnt_data;
1198         struct netcred *np;
1199         int error;
1200
1201         lwkt_gettoken(&hmp->fs_token);
1202         np = vfs_export_lookup(mp, &hmp->export, nam);
1203         if (np) {
1204                 *exflagsp = np->netc_exflags;
1205                 *credanonp = &np->netc_anon;
1206                 error = 0;
1207         } else {
1208                 error = EACCES;
1209         }
1210         lwkt_reltoken(&hmp->fs_token);
1211         return (error);
1212
1213 }
1214
1215 int
1216 hammer_vfs_export(struct mount *mp, int op, const struct export_args *export)
1217 {
1218         hammer_mount_t hmp = (void *)mp->mnt_data;
1219         int error;
1220
1221         lwkt_gettoken(&hmp->fs_token);
1222
1223         switch(op) {
1224         case MOUNTCTL_SET_EXPORT:
1225                 error = vfs_export(mp, &hmp->export, export);
1226                 break;
1227         default:
1228                 error = EOPNOTSUPP;
1229                 break;
1230         }
1231         lwkt_reltoken(&hmp->fs_token);
1232
1233         return(error);
1234 }
1235