62a73c981fd492747113349b61c019a365d68516
[dragonfly.git] / 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_yield_check = 16;
116 int hammer_fsync_mode = 3;
117 int64_t hammer_contention_count;
118 int64_t hammer_zone_limit;
119
120 /*
121  * Live dedup debug counters (sysctls are writable so that counters
122  * can be reset from userspace).
123  */
124 int64_t hammer_live_dedup_vnode_bcmps = 0;
125 int64_t hammer_live_dedup_device_bcmps = 0;
126 int64_t hammer_live_dedup_findblk_failures = 0;
127 int64_t hammer_live_dedup_bmap_saves = 0;
128
129
130 SYSCTL_NODE(_vfs, OID_AUTO, hammer, CTLFLAG_RW, 0, "HAMMER filesystem");
131
132 SYSCTL_INT(_vfs_hammer, OID_AUTO, supported_version, CTLFLAG_RD,
133            &hammer_supported_version, 0, "");
134 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_general, CTLFLAG_RW,
135            &hammer_debug_general, 0, "");
136 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_io, CTLFLAG_RW,
137            &hammer_debug_io, 0, "");
138 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_debug, CTLFLAG_RW,
139            &hammer_debug_debug, 0, "");
140 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_inode, CTLFLAG_RW,
141            &hammer_debug_inode, 0, "");
142 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_locks, CTLFLAG_RW,
143            &hammer_debug_locks, 0, "");
144 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_btree, CTLFLAG_RW,
145            &hammer_debug_btree, 0, "");
146 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_tid, CTLFLAG_RW,
147            &hammer_debug_tid, 0, "");
148 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover, CTLFLAG_RW,
149            &hammer_debug_recover, 0, "");
150 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover_faults, CTLFLAG_RW,
151            &hammer_debug_recover_faults, 0, "");
152 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_critical, CTLFLAG_RW,
153            &hammer_debug_critical, 0, "");
154 SYSCTL_INT(_vfs_hammer, OID_AUTO, cluster_enable, CTLFLAG_RW,
155            &hammer_cluster_enable, 0, "");
156 /*
157  * 0 - live dedup is disabled
158  * 1 - dedup cache is populated on reads only
159  * 2 - dedup cache is populated on both reads and writes
160  */
161 SYSCTL_INT(_vfs_hammer, OID_AUTO, live_dedup, CTLFLAG_RW,
162            &hammer_live_dedup, 0, "Enable live dedup");
163 SYSCTL_INT(_vfs_hammer, OID_AUTO, tdmux_ticks, CTLFLAG_RW,
164            &hammer_tdmux_ticks, 0, "Hammer tdmux ticks");
165
166 SYSCTL_LONG(_vfs_hammer, OID_AUTO, limit_dirtybufspace, CTLFLAG_RW,
167            &hammer_limit_dirtybufspace, 0, "");
168 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_recs, CTLFLAG_RW,
169            &hammer_limit_recs, 0, "");
170 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_inode_recs, CTLFLAG_RW,
171            &hammer_limit_inode_recs, 0, "");
172 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_reclaims, CTLFLAG_RW,
173            &hammer_limit_reclaims, 0, "");
174 SYSCTL_INT(_vfs_hammer, OID_AUTO, live_dedup_cache_size, CTLFLAG_RW,
175            &hammer_live_dedup_cache_size, 0,
176            "Number of cache entries");
177 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_redo, CTLFLAG_RW,
178            &hammer_limit_redo, 0, "");
179
180 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_fsyncs, CTLFLAG_RD,
181            &hammer_count_fsyncs, 0, "");
182 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_inodes, CTLFLAG_RD,
183            &hammer_count_inodes, 0, "");
184 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_iqueued, CTLFLAG_RD,
185            &hammer_count_iqueued, 0, "");
186 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reclaims, CTLFLAG_RD,
187            &hammer_count_reclaims, 0, "");
188 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_records, CTLFLAG_RD,
189            &hammer_count_records, 0, "");
190 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_record_datas, CTLFLAG_RD,
191            &hammer_count_record_datas, 0, "");
192 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_volumes, CTLFLAG_RD,
193            &hammer_count_volumes, 0, "");
194 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_buffers, CTLFLAG_RD,
195            &hammer_count_buffers, 0, "");
196 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_nodes, CTLFLAG_RD,
197            &hammer_count_nodes, 0, "");
198 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, count_extra_space_used, CTLFLAG_RD,
199            &hammer_count_extra_space_used, 0, "");
200
201 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_searches, CTLFLAG_RD,
202            &hammer_stats_btree_searches, 0, "");
203 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_lookups, CTLFLAG_RD,
204            &hammer_stats_btree_lookups, 0, "");
205 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_inserts, CTLFLAG_RD,
206            &hammer_stats_btree_inserts, 0, "");
207 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_deletes, CTLFLAG_RD,
208            &hammer_stats_btree_deletes, 0, "");
209 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_elements, CTLFLAG_RD,
210            &hammer_stats_btree_elements, 0, "");
211 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_splits, CTLFLAG_RD,
212            &hammer_stats_btree_splits, 0, "");
213 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_iterations, CTLFLAG_RD,
214            &hammer_stats_btree_iterations, 0, "");
215 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_root_iterations, CTLFLAG_RD,
216            &hammer_stats_btree_root_iterations, 0, "");
217 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_record_iterations, CTLFLAG_RD,
218            &hammer_stats_record_iterations, 0, "");
219
220 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_read, CTLFLAG_RD,
221            &hammer_stats_file_read, 0, "");
222 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_write, CTLFLAG_RD,
223            &hammer_stats_file_write, 0, "");
224 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_iopsr, CTLFLAG_RD,
225            &hammer_stats_file_iopsr, 0, "");
226 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_iopsw, CTLFLAG_RD,
227            &hammer_stats_file_iopsw, 0, "");
228 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_disk_read, CTLFLAG_RD,
229            &hammer_stats_disk_read, 0, "");
230 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_disk_write, CTLFLAG_RD,
231            &hammer_stats_disk_write, 0, "");
232 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_inode_flushes, CTLFLAG_RD,
233            &hammer_stats_inode_flushes, 0, "");
234 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_commits, CTLFLAG_RD,
235            &hammer_stats_commits, 0, "");
236 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_undo, CTLFLAG_RD,
237            &hammer_stats_undo, 0, "");
238 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_redo, CTLFLAG_RD,
239            &hammer_stats_redo, 0, "");
240
241 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, live_dedup_vnode_bcmps, CTLFLAG_RW,
242             &hammer_live_dedup_vnode_bcmps, 0,
243             "successful vnode buffer comparisons");
244 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, live_dedup_device_bcmps, CTLFLAG_RW,
245             &hammer_live_dedup_device_bcmps, 0,
246             "successful device buffer comparisons");
247 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, live_dedup_findblk_failures, CTLFLAG_RW,
248             &hammer_live_dedup_findblk_failures, 0,
249             "block lookup failures for comparison");
250 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, live_dedup_bmap_saves, CTLFLAG_RW,
251             &hammer_live_dedup_bmap_saves, 0,
252             "useful physical block lookups");
253
254 SYSCTL_LONG(_vfs_hammer, OID_AUTO, count_dirtybufspace, CTLFLAG_RD,
255            &hammer_count_dirtybufspace, 0, "");
256 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_refedbufs, CTLFLAG_RD,
257            &hammer_count_refedbufs, 0, "");
258 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reservations, CTLFLAG_RD,
259            &hammer_count_reservations, 0, "");
260 SYSCTL_LONG(_vfs_hammer, OID_AUTO, count_io_running_read, CTLFLAG_RD,
261            &hammer_count_io_running_read, 0, "");
262 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_locked, CTLFLAG_RD,
263            &hammer_count_io_locked, 0, "");
264 SYSCTL_LONG(_vfs_hammer, OID_AUTO, count_io_running_write, CTLFLAG_RD,
265            &hammer_count_io_running_write, 0, "");
266 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, zone_limit, CTLFLAG_RW,
267            &hammer_zone_limit, 0, "");
268 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, contention_count, CTLFLAG_RW,
269            &hammer_contention_count, 0, "");
270 SYSCTL_INT(_vfs_hammer, OID_AUTO, autoflush, CTLFLAG_RW,
271            &hammer_autoflush, 0, "");
272 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_zone, CTLFLAG_RW,
273            &hammer_verify_zone, 0, "");
274 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_data, CTLFLAG_RW,
275            &hammer_verify_data, 0, "");
276 SYSCTL_INT(_vfs_hammer, OID_AUTO, write_mode, CTLFLAG_RW,
277            &hammer_write_mode, 0, "");
278 SYSCTL_INT(_vfs_hammer, OID_AUTO, double_buffer, CTLFLAG_RW,
279            &hammer_double_buffer, 0, "");
280 SYSCTL_INT(_vfs_hammer, OID_AUTO, yield_check, CTLFLAG_RW,
281            &hammer_yield_check, 0, "");
282 SYSCTL_INT(_vfs_hammer, OID_AUTO, fsync_mode, CTLFLAG_RW,
283            &hammer_fsync_mode, 0, "");
284
285 /* KTR_INFO_MASTER(hammer); */
286
287 /*
288  * VFS ABI
289  */
290 static void     hammer_free_hmp(struct mount *mp);
291
292 static int      hammer_vfs_mount(struct mount *mp, char *path, caddr_t data,
293                                 struct ucred *cred);
294 static int      hammer_vfs_unmount(struct mount *mp, int mntflags);
295 static int      hammer_vfs_root(struct mount *mp, struct vnode **vpp);
296 static int      hammer_vfs_statfs(struct mount *mp, struct statfs *sbp,
297                                 struct ucred *cred);
298 static int      hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp,
299                                 struct ucred *cred);
300 static int      hammer_vfs_sync(struct mount *mp, int waitfor);
301 static int      hammer_vfs_vget(struct mount *mp, struct vnode *dvp,
302                                 ino_t ino, struct vnode **vpp);
303 static int      hammer_vfs_init(struct vfsconf *conf);
304 static int      hammer_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
305                                 struct fid *fhp, struct vnode **vpp);
306 static int      hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp);
307 static int      hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
308                                 int *exflagsp, struct ucred **credanonp);
309
310
311 static struct vfsops hammer_vfsops = {
312         .vfs_mount      = hammer_vfs_mount,
313         .vfs_unmount    = hammer_vfs_unmount,
314         .vfs_root       = hammer_vfs_root,
315         .vfs_statfs     = hammer_vfs_statfs,
316         .vfs_statvfs    = hammer_vfs_statvfs,
317         .vfs_sync       = hammer_vfs_sync,
318         .vfs_vget       = hammer_vfs_vget,
319         .vfs_init       = hammer_vfs_init,
320         .vfs_vptofh     = hammer_vfs_vptofh,
321         .vfs_fhtovp     = hammer_vfs_fhtovp,
322         .vfs_checkexp   = hammer_vfs_checkexp
323 };
324
325 MALLOC_DEFINE(M_HAMMER, "HAMMER-mount", "");
326
327 VFS_SET(hammer_vfsops, hammer, 0);
328 MODULE_VERSION(hammer, 1);
329
330 static int
331 hammer_vfs_init(struct vfsconf *conf)
332 {
333         int n;
334
335         /*
336          * Wait up to this long for an exclusive deadlock to clear
337          * before acquiring a new shared lock on the ip.  The deadlock
338          * may have occured on a b-tree node related to the ip.
339          */
340         if (hammer_tdmux_ticks == 0)
341                 hammer_tdmux_ticks = hz / 5;
342
343         /*
344          * Autosize
345          */
346         if (hammer_limit_recs == 0) {
347                 hammer_limit_recs = nbuf * 25;
348                 n = kmalloc_limit(M_HAMMER) / 512;
349                 if (hammer_limit_recs > n)
350                         hammer_limit_recs = n;
351         }
352         if (hammer_limit_dirtybufspace == 0) {
353                 hammer_limit_dirtybufspace = hidirtybufspace / 2;
354                 if (hammer_limit_dirtybufspace < 100)
355                         hammer_limit_dirtybufspace = 100;
356         }
357
358         /*
359          * The hammer_inode structure detaches from the vnode on reclaim.
360          * This limits the number of inodes in this state to prevent a
361          * memory pool blowout.
362          */
363         if (hammer_limit_reclaims == 0)
364                 hammer_limit_reclaims = desiredvnodes / 10;
365
366         return(0);
367 }
368
369 static int
370 hammer_vfs_mount(struct mount *mp, char *mntpt, caddr_t data,
371                  struct ucred *cred)
372 {
373         struct hammer_mount_info info;
374         hammer_mount_t hmp;
375         hammer_volume_t rootvol;
376         struct vnode *rootvp;
377         struct vnode *devvp = NULL;
378         const char *upath;      /* volume name in userspace */
379         char *path;             /* volume name in system space */
380         int error;
381         int i;
382         int master_id;
383         char *next_volume_ptr = NULL;
384
385         /*
386          * Accept hammer_mount_info.  mntpt is NULL for root mounts at boot.
387          */
388         if (mntpt == NULL) {
389                 bzero(&info, sizeof(info));
390                 info.asof = 0;
391                 info.hflags = 0;
392                 info.nvolumes = 1;
393
394                 next_volume_ptr = mp->mnt_stat.f_mntfromname;
395
396                 /* Count number of volumes separated by ':' */
397                 for (char *p = next_volume_ptr; *p != '\0'; ++p) {
398                         if (*p == ':') {
399                                 ++info.nvolumes;
400                         }
401                 }
402
403                 mp->mnt_flag &= ~MNT_RDONLY; /* mount R/W */
404         } else {
405                 if ((error = copyin(data, &info, sizeof(info))) != 0)
406                         return (error);
407         }
408
409         /*
410          * updating or new mount
411          */
412         if (mp->mnt_flag & MNT_UPDATE) {
413                 hmp = (void *)mp->mnt_data;
414                 KKASSERT(hmp != NULL);
415         } else {
416                 if (info.nvolumes <= 0 || info.nvolumes >= 32768)
417                         return (EINVAL);
418                 hmp = NULL;
419         }
420
421         /*
422          * master-id validation.  The master id may not be changed by a
423          * mount update.
424          */
425         if (info.hflags & HMNT_MASTERID) {
426                 if (hmp && hmp->master_id != info.master_id) {
427                         kprintf("hammer: cannot change master id "
428                                 "with mount update\n");
429                         return(EINVAL);
430                 }
431                 master_id = info.master_id;
432                 if (master_id < -1 || master_id >= HAMMER_MAX_MASTERS)
433                         return (EINVAL);
434         } else {
435                 if (hmp)
436                         master_id = hmp->master_id;
437                 else
438                         master_id = 0;
439         }
440
441         /*
442          * Internal mount data structure
443          */
444         if (hmp == NULL) {
445                 hmp = kmalloc(sizeof(*hmp), M_HAMMER, M_WAITOK | M_ZERO);
446                 mp->mnt_data = (qaddr_t)hmp;
447                 hmp->mp = mp;
448                 /*TAILQ_INIT(&hmp->recycle_list);*/
449
450                 /*
451                  * Make sure kmalloc type limits are set appropriately.
452                  *
453                  * Our inode kmalloc group is sized based on maxvnodes
454                  * (controlled by the system, not us).
455                  */
456                 kmalloc_create(&hmp->m_misc, "HAMMER-others");
457                 kmalloc_create(&hmp->m_inodes, "HAMMER-inodes");
458
459                 kmalloc_raise_limit(hmp->m_inodes, 0);  /* unlimited */
460
461                 hmp->root_btree_beg.localization = 0x00000000U;
462                 hmp->root_btree_beg.obj_id = -0x8000000000000000LL;
463                 hmp->root_btree_beg.key = -0x8000000000000000LL;
464                 hmp->root_btree_beg.create_tid = 1;
465                 hmp->root_btree_beg.delete_tid = 1;
466                 hmp->root_btree_beg.rec_type = 0;
467                 hmp->root_btree_beg.obj_type = 0;
468
469                 hmp->root_btree_end.localization = 0xFFFFFFFFU;
470                 hmp->root_btree_end.obj_id = 0x7FFFFFFFFFFFFFFFLL;
471                 hmp->root_btree_end.key = 0x7FFFFFFFFFFFFFFFLL;
472                 hmp->root_btree_end.create_tid = 0xFFFFFFFFFFFFFFFFULL;
473                 hmp->root_btree_end.delete_tid = 0;   /* special case */
474                 hmp->root_btree_end.rec_type = 0xFFFFU;
475                 hmp->root_btree_end.obj_type = 0;
476
477                 hmp->krate.freq = 1;    /* maximum reporting rate (hz) */
478                 hmp->krate.count = -16; /* initial burst */
479
480                 hmp->sync_lock.refs = 1;
481                 hmp->free_lock.refs = 1;
482                 hmp->undo_lock.refs = 1;
483                 hmp->blkmap_lock.refs = 1;
484                 hmp->snapshot_lock.refs = 1;
485                 hmp->volume_lock.refs = 1;
486
487                 TAILQ_INIT(&hmp->delay_list);
488                 TAILQ_INIT(&hmp->flush_group_list);
489                 TAILQ_INIT(&hmp->objid_cache_list);
490                 TAILQ_INIT(&hmp->undo_lru_list);
491                 TAILQ_INIT(&hmp->reclaim_list);
492
493                 RB_INIT(&hmp->rb_dedup_crc_root);
494                 RB_INIT(&hmp->rb_dedup_off_root);       
495                 TAILQ_INIT(&hmp->dedup_lru_list);
496         }
497         hmp->hflags &= ~HMNT_USERFLAGS;
498         hmp->hflags |= info.hflags & HMNT_USERFLAGS;
499
500         hmp->master_id = master_id;
501
502         if (info.asof) {
503                 mp->mnt_flag |= MNT_RDONLY;
504                 hmp->asof = info.asof;
505         } else {
506                 hmp->asof = HAMMER_MAX_TID;
507         }
508
509         hmp->volume_to_remove = -1;
510
511         /*
512          * Re-open read-write if originally read-only, or vise-versa.
513          *
514          * When going from read-only to read-write execute the stage2
515          * recovery if it has not already been run.
516          */
517         if (mp->mnt_flag & MNT_UPDATE) {
518                 lwkt_gettoken(&hmp->fs_token);
519                 error = 0;
520                 if (hmp->ronly && (mp->mnt_kern_flag & MNTK_WANTRDWR)) {
521                         kprintf("HAMMER read-only -> read-write\n");
522                         hmp->ronly = 0;
523                         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
524                                 hammer_adjust_volume_mode, NULL);
525                         rootvol = hammer_get_root_volume(hmp, &error);
526                         if (rootvol) {
527                                 hammer_recover_flush_buffers(hmp, rootvol, 1);
528                                 error = hammer_recover_stage2(hmp, rootvol);
529                                 bcopy(rootvol->ondisk->vol0_blockmap,
530                                       hmp->blockmap,
531                                       sizeof(hmp->blockmap));
532                                 hammer_rel_volume(rootvol, 0);
533                         }
534                         RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
535                                 hammer_reload_inode, NULL);
536                         /* kernel clears MNT_RDONLY */
537                 } else if (hmp->ronly == 0 && (mp->mnt_flag & MNT_RDONLY)) {
538                         kprintf("HAMMER read-write -> read-only\n");
539                         hmp->ronly = 1; /* messy */
540                         RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
541                                 hammer_reload_inode, NULL);
542                         hmp->ronly = 0;
543                         hammer_flusher_sync(hmp);
544                         hammer_flusher_sync(hmp);
545                         hammer_flusher_sync(hmp);
546                         hmp->ronly = 1;
547                         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
548                                 hammer_adjust_volume_mode, NULL);
549                 }
550                 lwkt_reltoken(&hmp->fs_token);
551                 return(error);
552         }
553
554         RB_INIT(&hmp->rb_vols_root);
555         RB_INIT(&hmp->rb_inos_root);
556         RB_INIT(&hmp->rb_redo_root);
557         RB_INIT(&hmp->rb_nods_root);
558         RB_INIT(&hmp->rb_undo_root);
559         RB_INIT(&hmp->rb_resv_root);
560         RB_INIT(&hmp->rb_bufs_root);
561         RB_INIT(&hmp->rb_pfsm_root);
562
563         hmp->ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
564
565         RB_INIT(&hmp->volu_root);
566         RB_INIT(&hmp->undo_root);
567         RB_INIT(&hmp->data_root);
568         RB_INIT(&hmp->meta_root);
569         RB_INIT(&hmp->lose_root);
570         TAILQ_INIT(&hmp->iorun_list);
571
572         lwkt_token_init(&hmp->fs_token, "hammerfs");
573         lwkt_token_init(&hmp->io_token, "hammerio");
574
575         lwkt_gettoken(&hmp->fs_token);
576
577         /*
578          * Load volumes
579          */
580         path = objcache_get(namei_oc, M_WAITOK);
581         hmp->nvolumes = -1;
582         for (i = 0; i < info.nvolumes; ++i) {
583                 if (mntpt == NULL) {
584                         /*
585                          * Root mount.
586                          */
587                         KKASSERT(next_volume_ptr != NULL);
588                         strcpy(path, "");
589                         if (*next_volume_ptr != '/') {
590                                 /* relative path */
591                                 strcpy(path, "/dev/");
592                         }
593                         int k;
594                         for (k = strlen(path); k < MAXPATHLEN-1; ++k) {
595                                 if (*next_volume_ptr == '\0') {
596                                         break;
597                                 } else if (*next_volume_ptr == ':') {
598                                         ++next_volume_ptr;
599                                         break;
600                                 } else {
601                                         path[k] = *next_volume_ptr;
602                                         ++next_volume_ptr;
603                                 }
604                         }
605                         path[k] = '\0';
606
607                         error = 0;
608                         cdev_t dev = kgetdiskbyname(path);
609                         error = bdevvp(dev, &devvp);
610                         if (error) {
611                                 kprintf("hammer_mountroot: can't find devvp\n");
612                         }
613                 } else {
614                         error = copyin(&info.volumes[i], &upath,
615                                        sizeof(char *));
616                         if (error == 0)
617                                 error = copyinstr(upath, path,
618                                                   MAXPATHLEN, NULL);
619                 }
620                 if (error == 0)
621                         error = hammer_install_volume(hmp, path, devvp);
622                 if (error)
623                         break;
624         }
625         objcache_put(namei_oc, path);
626
627         /*
628          * Make sure we found a root volume
629          */
630         if (error == 0 && hmp->rootvol == NULL) {
631                 kprintf("hammer_mount: No root volume found!\n");
632                 error = EINVAL;
633         }
634
635         /*
636          * Check that all required volumes are available
637          */
638         if (error == 0 && hammer_mountcheck_volumes(hmp)) {
639                 kprintf("hammer_mount: Missing volumes, cannot mount!\n");
640                 error = EINVAL;
641         }
642
643         if (error) {
644                 /* called with fs_token held */
645                 hammer_free_hmp(mp);
646                 return (error);
647         }
648
649         /*
650          * No errors, setup enough of the mount point so we can lookup the
651          * root vnode.
652          */
653         mp->mnt_iosize_max = MAXPHYS;
654         mp->mnt_kern_flag |= MNTK_FSMID;
655
656         /*
657          * MPSAFE code.  Note that VOPs and VFSops which are not MPSAFE
658          * will acquire a per-mount token prior to entry and release it
659          * on return, so even if we do not specify it we no longer get
660          * the BGL regardlless of how we are flagged.
661          */
662         mp->mnt_kern_flag |= MNTK_ALL_MPSAFE;
663         /*MNTK_RD_MPSAFE | MNTK_GA_MPSAFE | MNTK_IN_MPSAFE;*/
664
665         /* 
666          * note: f_iosize is used by vnode_pager_haspage() when constructing
667          * its VOP_BMAP call.
668          */
669         mp->mnt_stat.f_iosize = HAMMER_BUFSIZE;
670         mp->mnt_stat.f_bsize = HAMMER_BUFSIZE;
671
672         mp->mnt_vstat.f_frsize = HAMMER_BUFSIZE;
673         mp->mnt_vstat.f_bsize = HAMMER_BUFSIZE;
674
675         mp->mnt_maxsymlinklen = 255;
676         mp->mnt_flag |= MNT_LOCAL;
677
678         vfs_add_vnodeops(mp, &hammer_vnode_vops, &mp->mnt_vn_norm_ops);
679         vfs_add_vnodeops(mp, &hammer_spec_vops, &mp->mnt_vn_spec_ops);
680         vfs_add_vnodeops(mp, &hammer_fifo_vops, &mp->mnt_vn_fifo_ops);
681
682         /*
683          * The root volume's ondisk pointer is only valid if we hold a
684          * reference to it.
685          */
686         rootvol = hammer_get_root_volume(hmp, &error);
687         if (error)
688                 goto failed;
689
690         /*
691          * Perform any necessary UNDO operations.  The recovery code does
692          * call hammer_undo_lookup() so we have to pre-cache the blockmap,
693          * and then re-copy it again after recovery is complete.
694          *
695          * If this is a read-only mount the UNDO information is retained
696          * in memory in the form of dirty buffer cache buffers, and not
697          * written back to the media.
698          */
699         bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
700               sizeof(hmp->blockmap));
701
702         /*
703          * Check filesystem version
704          */
705         hmp->version = rootvol->ondisk->vol_version;
706         if (hmp->version < HAMMER_VOL_VERSION_MIN ||
707             hmp->version > HAMMER_VOL_VERSION_MAX) {
708                 kprintf("HAMMER: mount unsupported fs version %d\n",
709                         hmp->version);
710                 error = ERANGE;
711                 goto done;
712         }
713
714         /*
715          * The undo_rec_limit limits the size of flush groups to avoid
716          * blowing out the UNDO FIFO.  This calculation is typically in
717          * the tens of thousands and is designed primarily when small
718          * HAMMER filesystems are created.
719          */
720         hmp->undo_rec_limit = hammer_undo_max(hmp) / 8192 + 100;
721         if (hammer_debug_general & 0x0001)
722                 kprintf("HAMMER: undo_rec_limit %d\n", hmp->undo_rec_limit);
723
724         /*
725          * NOTE: Recover stage1 not only handles meta-data recovery, it
726          *       also sets hmp->undo_seqno for HAMMER VERSION 4+ filesystems.
727          */
728         error = hammer_recover_stage1(hmp, rootvol);
729         if (error) {
730                 kprintf("Failed to recover HAMMER filesystem on mount\n");
731                 goto done;
732         }
733
734         /*
735          * Finish setup now that we have a good root volume.
736          *
737          * The top 16 bits of fsid.val[1] is a pfs id.
738          */
739         ksnprintf(mp->mnt_stat.f_mntfromname,
740                   sizeof(mp->mnt_stat.f_mntfromname), "%s",
741                   rootvol->ondisk->vol_name);
742         mp->mnt_stat.f_fsid.val[0] =
743                 crc32((char *)&rootvol->ondisk->vol_fsid + 0, 8);
744         mp->mnt_stat.f_fsid.val[1] =
745                 crc32((char *)&rootvol->ondisk->vol_fsid + 8, 8);
746         mp->mnt_stat.f_fsid.val[1] &= 0x0000FFFF;
747
748         mp->mnt_vstat.f_fsid_uuid = rootvol->ondisk->vol_fsid;
749         mp->mnt_vstat.f_fsid = crc32(&mp->mnt_vstat.f_fsid_uuid,
750                                      sizeof(mp->mnt_vstat.f_fsid_uuid));
751
752         /*
753          * Certain often-modified fields in the root volume are cached in
754          * the hammer_mount structure so we do not have to generate lots
755          * of little UNDO structures for them.
756          *
757          * Recopy after recovery.  This also has the side effect of
758          * setting our cached undo FIFO's first_offset, which serves to
759          * placemark the FIFO start for the NEXT flush cycle while the
760          * on-disk first_offset represents the LAST flush cycle.
761          */
762         hmp->next_tid = rootvol->ondisk->vol0_next_tid;
763         hmp->flush_tid1 = hmp->next_tid;
764         hmp->flush_tid2 = hmp->next_tid;
765         bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
766               sizeof(hmp->blockmap));
767         hmp->copy_stat_freebigblocks = rootvol->ondisk->vol0_stat_freebigblocks;
768
769         hammer_flusher_create(hmp);
770
771         /*
772          * Locate the root directory using the root cluster's B-Tree as a
773          * starting point.  The root directory uses an obj_id of 1.
774          *
775          * FUTURE: Leave the root directory cached referenced but unlocked
776          * in hmp->rootvp (need to flush it on unmount).
777          */
778         error = hammer_vfs_vget(mp, NULL, 1, &rootvp);
779         if (error)
780                 goto done;
781         vput(rootvp);
782         /*vn_unlock(hmp->rootvp);*/
783         if (hmp->ronly == 0)
784                 error = hammer_recover_stage2(hmp, rootvol);
785
786         /*
787          * If the stage2 recovery fails be sure to clean out all cached
788          * vnodes before throwing away the mount structure or bad things
789          * will happen.
790          */
791         if (error)
792                 vflush(mp, 0, 0);
793
794 done:
795         hammer_rel_volume(rootvol, 0);
796 failed:
797         /*
798          * Cleanup and return.
799          */
800         if (error) {
801                 /* called with fs_token held */
802                 hammer_free_hmp(mp);
803         } else {
804                 lwkt_reltoken(&hmp->fs_token);
805         }
806         return (error);
807 }
808
809 static int
810 hammer_vfs_unmount(struct mount *mp, int mntflags)
811 {
812         hammer_mount_t hmp = (void *)mp->mnt_data;
813         int flags;
814         int error;
815
816         /*
817          * Clean out the vnodes
818          */
819         lwkt_gettoken(&hmp->fs_token);
820         flags = 0;
821         if (mntflags & MNT_FORCE)
822                 flags |= FORCECLOSE;
823         error = vflush(mp, 0, flags);
824
825         /*
826          * Clean up the internal mount structure and related entities.  This
827          * may issue I/O.
828          */
829         if (error == 0) {
830                 /* called with fs_token held */
831                 hammer_free_hmp(mp);
832         } else {
833                 lwkt_reltoken(&hmp->fs_token);
834         }
835         return(error);
836 }
837
838 /*
839  * Clean up the internal mount structure and disassociate it from the mount.
840  * This may issue I/O.
841  *
842  * Called with fs_token held.
843  */
844 static void
845 hammer_free_hmp(struct mount *mp)
846 {
847         hammer_mount_t hmp = (void *)mp->mnt_data;
848         hammer_flush_group_t flg;
849         int count;
850         int dummy;
851
852         /*
853          * Flush anything dirty.  This won't even run if the
854          * filesystem errored-out.
855          */
856         count = 0;
857         while (hammer_flusher_haswork(hmp)) {
858                 hammer_flusher_sync(hmp);
859                 ++count;
860                 if (count >= 5) {
861                         if (count == 5)
862                                 kprintf("HAMMER: umount flushing.");
863                         else
864                                 kprintf(".");
865                         tsleep(&dummy, 0, "hmrufl", hz);
866                 }
867                 if (count == 30) {
868                         kprintf("giving up\n");
869                         break;
870                 }
871         }
872         if (count >= 5 && count < 30)
873                 kprintf("\n");
874
875         /*
876          * If the mount had a critical error we have to destroy any
877          * remaining inodes before we can finish cleaning up the flusher.
878          */
879         if (hmp->flags & HAMMER_MOUNT_CRITICAL_ERROR) {
880                 RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
881                         hammer_destroy_inode_callback, NULL);
882         }
883
884         /*
885          * There shouldn't be any inodes left now and any left over
886          * flush groups should now be empty.
887          */
888         KKASSERT(RB_EMPTY(&hmp->rb_inos_root));
889         while ((flg = TAILQ_FIRST(&hmp->flush_group_list)) != NULL) {
890                 TAILQ_REMOVE(&hmp->flush_group_list, flg, flush_entry);
891                 KKASSERT(RB_EMPTY(&flg->flush_tree));
892                 if (flg->refs) {
893                         kprintf("HAMMER: Warning, flush_group %p was "
894                                 "not empty on umount!\n", flg);
895                 }
896                 kfree(flg, hmp->m_misc);
897         }
898
899         /*
900          * We can finally destroy the flusher
901          */
902         hammer_flusher_destroy(hmp);
903
904         /*
905          * We may have held recovered buffers due to a read-only mount.
906          * These must be discarded.
907          */
908         if (hmp->ronly)
909                 hammer_recover_flush_buffers(hmp, NULL, -1);
910
911         /*
912          * Unload buffers and then volumes
913          */
914         RB_SCAN(hammer_buf_rb_tree, &hmp->rb_bufs_root, NULL,
915                 hammer_unload_buffer, NULL);
916         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
917                 hammer_unload_volume, NULL);
918
919         mp->mnt_data = NULL;
920         mp->mnt_flag &= ~MNT_LOCAL;
921         hmp->mp = NULL;
922         hammer_destroy_objid_cache(hmp);
923         hammer_destroy_dedup_cache(hmp);
924         if (hmp->dedup_free_cache != NULL) {
925                 kfree(hmp->dedup_free_cache, hmp->m_misc);
926                 hmp->dedup_free_cache = NULL;
927         }
928         kmalloc_destroy(&hmp->m_misc);
929         kmalloc_destroy(&hmp->m_inodes);
930         lwkt_reltoken(&hmp->fs_token);
931         kfree(hmp, M_HAMMER);
932 }
933
934 /*
935  * Report critical errors.  ip may be NULL.
936  */
937 void
938 hammer_critical_error(hammer_mount_t hmp, hammer_inode_t ip,
939                       int error, const char *msg)
940 {
941         hmp->flags |= HAMMER_MOUNT_CRITICAL_ERROR;
942
943         krateprintf(&hmp->krate,
944                     "HAMMER(%s): Critical error inode=%jd error=%d %s\n",
945                     hmp->mp->mnt_stat.f_mntfromname,
946                     (intmax_t)(ip ? ip->obj_id : -1),
947                     error, msg);
948
949         if (hmp->ronly == 0) {
950                 hmp->ronly = 2;         /* special errored read-only mode */
951                 hmp->mp->mnt_flag |= MNT_RDONLY;
952                 RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
953                         hammer_adjust_volume_mode, NULL);
954                 kprintf("HAMMER(%s): Forcing read-only mode\n",
955                         hmp->mp->mnt_stat.f_mntfromname);
956         }
957         hmp->error = error;
958         if (hammer_debug_critical)
959                 Debugger("Entering debugger");
960 }
961
962
963 /*
964  * Obtain a vnode for the specified inode number.  An exclusively locked
965  * vnode is returned.
966  */
967 int
968 hammer_vfs_vget(struct mount *mp, struct vnode *dvp,
969                 ino_t ino, struct vnode **vpp)
970 {
971         struct hammer_transaction trans;
972         struct hammer_mount *hmp = (void *)mp->mnt_data;
973         struct hammer_inode *ip;
974         int error;
975         u_int32_t localization;
976
977         lwkt_gettoken(&hmp->fs_token);
978         hammer_simple_transaction(&trans, hmp);
979
980         /*
981          * If a directory vnode is supplied (mainly NFS) then we can acquire
982          * the PFS domain from it.  Otherwise we would only be able to vget
983          * inodes in the root PFS.
984          */
985         if (dvp) {
986                 localization = HAMMER_DEF_LOCALIZATION +
987                                 VTOI(dvp)->obj_localization;
988         } else {
989                 localization = HAMMER_DEF_LOCALIZATION;
990         }
991
992         /*
993          * Lookup the requested HAMMER inode.  The structure must be
994          * left unlocked while we manipulate the related vnode to avoid
995          * a deadlock.
996          */
997         ip = hammer_get_inode(&trans, NULL, ino,
998                               hmp->asof, localization,
999                               0, &error);
1000         if (ip == NULL) {
1001                 *vpp = NULL;
1002         } else {
1003                 error = hammer_get_vnode(ip, vpp);
1004                 hammer_rel_inode(ip, 0);
1005         }
1006         hammer_done_transaction(&trans);
1007         lwkt_reltoken(&hmp->fs_token);
1008         return (error);
1009 }
1010
1011 /*
1012  * Return the root vnode for the filesystem.
1013  *
1014  * HAMMER stores the root vnode in the hammer_mount structure so
1015  * getting it is easy.
1016  */
1017 static int
1018 hammer_vfs_root(struct mount *mp, struct vnode **vpp)
1019 {
1020         int error;
1021
1022         error = hammer_vfs_vget(mp, NULL, 1, vpp);
1023         return (error);
1024 }
1025
1026 static int
1027 hammer_vfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
1028 {
1029         struct hammer_mount *hmp = (void *)mp->mnt_data;
1030         hammer_volume_t volume;
1031         hammer_volume_ondisk_t ondisk;
1032         int error;
1033         int64_t bfree;
1034         int64_t breserved;
1035
1036         lwkt_gettoken(&hmp->fs_token);
1037         volume = hammer_get_root_volume(hmp, &error);
1038         if (error) {
1039                 lwkt_reltoken(&hmp->fs_token);
1040                 return(error);
1041         }
1042         ondisk = volume->ondisk;
1043
1044         /*
1045          * Basic stats
1046          */
1047         _hammer_checkspace(hmp, HAMMER_CHKSPC_WRITE, &breserved);
1048         mp->mnt_stat.f_files = ondisk->vol0_stat_inodes;
1049         bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE;
1050         hammer_rel_volume(volume, 0);
1051
1052         mp->mnt_stat.f_bfree = (bfree - breserved) / HAMMER_BUFSIZE;
1053         mp->mnt_stat.f_bavail = mp->mnt_stat.f_bfree;
1054         if (mp->mnt_stat.f_files < 0)
1055                 mp->mnt_stat.f_files = 0;
1056
1057         *sbp = mp->mnt_stat;
1058         lwkt_reltoken(&hmp->fs_token);
1059         return(0);
1060 }
1061
1062 static int
1063 hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred)
1064 {
1065         struct hammer_mount *hmp = (void *)mp->mnt_data;
1066         hammer_volume_t volume;
1067         hammer_volume_ondisk_t ondisk;
1068         int error;
1069         int64_t bfree;
1070         int64_t breserved;
1071
1072         lwkt_gettoken(&hmp->fs_token);
1073         volume = hammer_get_root_volume(hmp, &error);
1074         if (error) {
1075                 lwkt_reltoken(&hmp->fs_token);
1076                 return(error);
1077         }
1078         ondisk = volume->ondisk;
1079
1080         /*
1081          * Basic stats
1082          */
1083         _hammer_checkspace(hmp, HAMMER_CHKSPC_WRITE, &breserved);
1084         mp->mnt_vstat.f_files = ondisk->vol0_stat_inodes;
1085         bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE;
1086         hammer_rel_volume(volume, 0);
1087
1088         mp->mnt_vstat.f_bfree = (bfree - breserved) / HAMMER_BUFSIZE;
1089         mp->mnt_vstat.f_bavail = mp->mnt_vstat.f_bfree;
1090         if (mp->mnt_vstat.f_files < 0)
1091                 mp->mnt_vstat.f_files = 0;
1092         *sbp = mp->mnt_vstat;
1093         lwkt_reltoken(&hmp->fs_token);
1094         return(0);
1095 }
1096
1097 /*
1098  * Sync the filesystem.  Currently we have to run it twice, the second
1099  * one will advance the undo start index to the end index, so if a crash
1100  * occurs no undos will be run on mount.
1101  *
1102  * We do not sync the filesystem if we are called from a panic.  If we did
1103  * we might end up blowing up a sync that was already in progress.
1104  */
1105 static int
1106 hammer_vfs_sync(struct mount *mp, int waitfor)
1107 {
1108         struct hammer_mount *hmp = (void *)mp->mnt_data;
1109         int error;
1110
1111         lwkt_gettoken(&hmp->fs_token);
1112         if (panicstr == NULL) {
1113                 error = hammer_sync_hmp(hmp, waitfor);
1114         } else {
1115                 error = EIO;
1116         }
1117         lwkt_reltoken(&hmp->fs_token);
1118         return (error);
1119 }
1120
1121 /*
1122  * Convert a vnode to a file handle.
1123  *
1124  * Accesses read-only fields on already-referenced structures so
1125  * no token is needed.
1126  */
1127 static int
1128 hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp)
1129 {
1130         hammer_inode_t ip;
1131
1132         KKASSERT(MAXFIDSZ >= 16);
1133         ip = VTOI(vp);
1134         fhp->fid_len = offsetof(struct fid, fid_data[16]);
1135         fhp->fid_ext = ip->obj_localization >> 16;
1136         bcopy(&ip->obj_id, fhp->fid_data + 0, sizeof(ip->obj_id));
1137         bcopy(&ip->obj_asof, fhp->fid_data + 8, sizeof(ip->obj_asof));
1138         return(0);
1139 }
1140
1141
1142 /*
1143  * Convert a file handle back to a vnode.
1144  *
1145  * Use rootvp to enforce PFS isolation when a PFS is exported via a
1146  * null mount.
1147  */
1148 static int
1149 hammer_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
1150                   struct fid *fhp, struct vnode **vpp)
1151 {
1152         hammer_mount_t hmp = (void *)mp->mnt_data;
1153         struct hammer_transaction trans;
1154         struct hammer_inode *ip;
1155         struct hammer_inode_info info;
1156         int error;
1157         u_int32_t localization;
1158
1159         bcopy(fhp->fid_data + 0, &info.obj_id, sizeof(info.obj_id));
1160         bcopy(fhp->fid_data + 8, &info.obj_asof, sizeof(info.obj_asof));
1161         if (rootvp)
1162                 localization = VTOI(rootvp)->obj_localization;
1163         else
1164                 localization = (u_int32_t)fhp->fid_ext << 16;
1165
1166         lwkt_gettoken(&hmp->fs_token);
1167         hammer_simple_transaction(&trans, hmp);
1168
1169         /*
1170          * Get/allocate the hammer_inode structure.  The structure must be
1171          * unlocked while we manipulate the related vnode to avoid a
1172          * deadlock.
1173          */
1174         ip = hammer_get_inode(&trans, NULL, info.obj_id,
1175                               info.obj_asof, localization, 0, &error);
1176         if (ip) {
1177                 error = hammer_get_vnode(ip, vpp);
1178                 hammer_rel_inode(ip, 0);
1179         } else {
1180                 *vpp = NULL;
1181         }
1182         hammer_done_transaction(&trans);
1183         lwkt_reltoken(&hmp->fs_token);
1184         return (error);
1185 }
1186
1187 static int
1188 hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
1189                     int *exflagsp, struct ucred **credanonp)
1190 {
1191         hammer_mount_t hmp = (void *)mp->mnt_data;
1192         struct netcred *np;
1193         int error;
1194
1195         lwkt_gettoken(&hmp->fs_token);
1196         np = vfs_export_lookup(mp, &hmp->export, nam);
1197         if (np) {
1198                 *exflagsp = np->netc_exflags;
1199                 *credanonp = &np->netc_anon;
1200                 error = 0;
1201         } else {
1202                 error = EACCES;
1203         }
1204         lwkt_reltoken(&hmp->fs_token);
1205         return (error);
1206
1207 }
1208
1209 int
1210 hammer_vfs_export(struct mount *mp, int op, const struct export_args *export)
1211 {
1212         hammer_mount_t hmp = (void *)mp->mnt_data;
1213         int error;
1214
1215         lwkt_gettoken(&hmp->fs_token);
1216
1217         switch(op) {
1218         case MOUNTCTL_SET_EXPORT:
1219                 error = vfs_export(mp, &hmp->export, export);
1220                 break;
1221         default:
1222                 error = EOPNOTSUPP;
1223                 break;
1224         }
1225         lwkt_reltoken(&hmp->fs_token);
1226
1227         return(error);
1228 }
1229