5ebb1ddab637fceb6301027615d5b74816e56d68
[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  * $DragonFly: src/sys/vfs/hammer/hammer_vfsops.c,v 1.74 2008/11/13 02:18:43 dillon Exp $
35  */
36
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/kernel.h>
40 #include <sys/vnode.h>
41 #include <sys/mount.h>
42 #include <sys/malloc.h>
43 #include <sys/nlookup.h>
44 #include <sys/fcntl.h>
45 #include <sys/sysctl.h>
46 #include <sys/buf.h>
47 #include <sys/buf2.h>
48 #include "hammer.h"
49
50 /*
51  * NOTE!  Global statistics may not be MPSAFE so HAMMER never uses them
52  *        in conditionals.
53  */
54 int hammer_supported_version = HAMMER_VOL_VERSION_DEFAULT;
55 int hammer_debug_io;
56 int hammer_debug_general;
57 int hammer_debug_debug = 1;             /* medium-error panics */ 
58 int hammer_debug_inode;
59 int hammer_debug_locks;
60 int hammer_debug_btree;
61 int hammer_debug_tid;
62 int hammer_debug_recover;               /* -1 will disable, +1 will force */
63 int hammer_debug_recover_faults;
64 int hammer_debug_critical;              /* non-zero enter debugger on error */
65 int hammer_cluster_enable = 1;          /* enable read clustering by default */
66 int hammer_live_dedup = 0;
67 int hammer_count_fsyncs;
68 int hammer_count_inodes;
69 int hammer_count_iqueued;
70 int hammer_count_reclaiming;
71 int hammer_count_records;
72 int hammer_count_record_datas;
73 int hammer_count_volumes;
74 int hammer_count_buffers;
75 int hammer_count_nodes;
76 int64_t hammer_count_extra_space_used;
77 int64_t hammer_stats_btree_lookups;
78 int64_t hammer_stats_btree_searches;
79 int64_t hammer_stats_btree_inserts;
80 int64_t hammer_stats_btree_deletes;
81 int64_t hammer_stats_btree_elements;
82 int64_t hammer_stats_btree_splits;
83 int64_t hammer_stats_btree_iterations;
84 int64_t hammer_stats_btree_root_iterations;
85 int64_t hammer_stats_record_iterations;
86
87 int64_t hammer_stats_file_read;
88 int64_t hammer_stats_file_write;
89 int64_t hammer_stats_file_iopsr;
90 int64_t hammer_stats_file_iopsw;
91 int64_t hammer_stats_disk_read;
92 int64_t hammer_stats_disk_write;
93 int64_t hammer_stats_inode_flushes;
94 int64_t hammer_stats_commits;
95 int64_t hammer_stats_undo;
96 int64_t hammer_stats_redo;
97
98 int hammer_count_dirtybufspace;         /* global */
99 int hammer_count_refedbufs;             /* global */
100 int hammer_count_reservations;
101 int hammer_count_io_running_read;
102 int hammer_count_io_running_write;
103 int hammer_count_io_locked;
104 int hammer_limit_dirtybufspace;         /* per-mount */
105 int hammer_limit_running_io;            /* per-mount */
106 int hammer_limit_recs;                  /* as a whole XXX */
107 int hammer_limit_inode_recs = 1024;     /* per inode */
108 int hammer_limit_reclaim = HAMMER_RECLAIM_WAIT;
109 int hammer_limit_redo = 4096 * 1024;    /* per inode */
110 int hammer_autoflush = 2000;            /* auto flush */
111 int hammer_bio_count;
112 int hammer_verify_zone;
113 int hammer_verify_data = 1;
114 int hammer_write_mode;
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, "");
163
164 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_dirtybufspace, CTLFLAG_RW,
165            &hammer_limit_dirtybufspace, 0, "");
166 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_running_io, CTLFLAG_RW,
167            &hammer_limit_running_io, 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_reclaim, CTLFLAG_RW,
173            &hammer_limit_reclaim, 0, "");
174 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_redo, CTLFLAG_RW,
175            &hammer_limit_redo, 0, "");
176
177 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_fsyncs, CTLFLAG_RD,
178            &hammer_count_fsyncs, 0, "");
179 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_inodes, CTLFLAG_RD,
180            &hammer_count_inodes, 0, "");
181 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_iqueued, CTLFLAG_RD,
182            &hammer_count_iqueued, 0, "");
183 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reclaiming, CTLFLAG_RD,
184            &hammer_count_reclaiming, 0, "");
185 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_records, CTLFLAG_RD,
186            &hammer_count_records, 0, "");
187 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_record_datas, CTLFLAG_RD,
188            &hammer_count_record_datas, 0, "");
189 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_volumes, CTLFLAG_RD,
190            &hammer_count_volumes, 0, "");
191 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_buffers, CTLFLAG_RD,
192            &hammer_count_buffers, 0, "");
193 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_nodes, CTLFLAG_RD,
194            &hammer_count_nodes, 0, "");
195 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, count_extra_space_used, CTLFLAG_RD,
196            &hammer_count_extra_space_used, 0, "");
197
198 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_searches, CTLFLAG_RD,
199            &hammer_stats_btree_searches, 0, "");
200 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_lookups, CTLFLAG_RD,
201            &hammer_stats_btree_lookups, 0, "");
202 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_inserts, CTLFLAG_RD,
203            &hammer_stats_btree_inserts, 0, "");
204 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_deletes, CTLFLAG_RD,
205            &hammer_stats_btree_deletes, 0, "");
206 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_elements, CTLFLAG_RD,
207            &hammer_stats_btree_elements, 0, "");
208 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_splits, CTLFLAG_RD,
209            &hammer_stats_btree_splits, 0, "");
210 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_iterations, CTLFLAG_RD,
211            &hammer_stats_btree_iterations, 0, "");
212 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_root_iterations, CTLFLAG_RD,
213            &hammer_stats_btree_root_iterations, 0, "");
214 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_record_iterations, CTLFLAG_RD,
215            &hammer_stats_record_iterations, 0, "");
216
217 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_read, CTLFLAG_RD,
218            &hammer_stats_file_read, 0, "");
219 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_write, CTLFLAG_RD,
220            &hammer_stats_file_write, 0, "");
221 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_iopsr, CTLFLAG_RD,
222            &hammer_stats_file_iopsr, 0, "");
223 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_iopsw, CTLFLAG_RD,
224            &hammer_stats_file_iopsw, 0, "");
225 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_disk_read, CTLFLAG_RD,
226            &hammer_stats_disk_read, 0, "");
227 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_disk_write, CTLFLAG_RD,
228            &hammer_stats_disk_write, 0, "");
229 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_inode_flushes, CTLFLAG_RD,
230            &hammer_stats_inode_flushes, 0, "");
231 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_commits, CTLFLAG_RD,
232            &hammer_stats_commits, 0, "");
233 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_undo, CTLFLAG_RD,
234            &hammer_stats_undo, 0, "");
235 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_redo, CTLFLAG_RD,
236            &hammer_stats_redo, 0, "");
237
238 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, live_dedup_vnode_bcmps, CTLFLAG_RW,
239             &hammer_live_dedup_vnode_bcmps, 0, "");
240 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, live_dedup_device_bcmps, CTLFLAG_RW,
241             &hammer_live_dedup_device_bcmps, 0, "");
242 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, live_dedup_findblk_failures, CTLFLAG_RW,
243             &hammer_live_dedup_findblk_failures, 0, "");
244 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, live_dedup_bmap_saves, CTLFLAG_RW,
245             &hammer_live_dedup_bmap_saves, 0, "");
246
247 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_dirtybufspace, CTLFLAG_RD,
248            &hammer_count_dirtybufspace, 0, "");
249 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_refedbufs, CTLFLAG_RD,
250            &hammer_count_refedbufs, 0, "");
251 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reservations, CTLFLAG_RD,
252            &hammer_count_reservations, 0, "");
253 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_running_read, CTLFLAG_RD,
254            &hammer_count_io_running_read, 0, "");
255 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_locked, CTLFLAG_RD,
256            &hammer_count_io_locked, 0, "");
257 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_running_write, CTLFLAG_RD,
258            &hammer_count_io_running_write, 0, "");
259 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, zone_limit, CTLFLAG_RW,
260            &hammer_zone_limit, 0, "");
261 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, contention_count, CTLFLAG_RW,
262            &hammer_contention_count, 0, "");
263 SYSCTL_INT(_vfs_hammer, OID_AUTO, autoflush, CTLFLAG_RW,
264            &hammer_autoflush, 0, "");
265 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_zone, CTLFLAG_RW,
266            &hammer_verify_zone, 0, "");
267 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_data, CTLFLAG_RW,
268            &hammer_verify_data, 0, "");
269 SYSCTL_INT(_vfs_hammer, OID_AUTO, write_mode, CTLFLAG_RW,
270            &hammer_write_mode, 0, "");
271 SYSCTL_INT(_vfs_hammer, OID_AUTO, yield_check, CTLFLAG_RW,
272            &hammer_yield_check, 0, "");
273 SYSCTL_INT(_vfs_hammer, OID_AUTO, fsync_mode, CTLFLAG_RW,
274            &hammer_fsync_mode, 0, "");
275
276 KTR_INFO_MASTER(hammer);
277
278 /*
279  * VFS ABI
280  */
281 static void     hammer_free_hmp(struct mount *mp);
282
283 static int      hammer_vfs_mount(struct mount *mp, char *path, caddr_t data,
284                                 struct ucred *cred);
285 static int      hammer_vfs_unmount(struct mount *mp, int mntflags);
286 static int      hammer_vfs_root(struct mount *mp, struct vnode **vpp);
287 static int      hammer_vfs_statfs(struct mount *mp, struct statfs *sbp,
288                                 struct ucred *cred);
289 static int      hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp,
290                                 struct ucred *cred);
291 static int      hammer_vfs_sync(struct mount *mp, int waitfor);
292 static int      hammer_vfs_vget(struct mount *mp, struct vnode *dvp,
293                                 ino_t ino, struct vnode **vpp);
294 static int      hammer_vfs_init(struct vfsconf *conf);
295 static int      hammer_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
296                                 struct fid *fhp, struct vnode **vpp);
297 static int      hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp);
298 static int      hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
299                                 int *exflagsp, struct ucred **credanonp);
300
301
302 static struct vfsops hammer_vfsops = {
303         .vfs_mount      = hammer_vfs_mount,
304         .vfs_unmount    = hammer_vfs_unmount,
305         .vfs_root       = hammer_vfs_root,
306         .vfs_statfs     = hammer_vfs_statfs,
307         .vfs_statvfs    = hammer_vfs_statvfs,
308         .vfs_sync       = hammer_vfs_sync,
309         .vfs_vget       = hammer_vfs_vget,
310         .vfs_init       = hammer_vfs_init,
311         .vfs_vptofh     = hammer_vfs_vptofh,
312         .vfs_fhtovp     = hammer_vfs_fhtovp,
313         .vfs_checkexp   = hammer_vfs_checkexp
314 };
315
316 MALLOC_DEFINE(M_HAMMER, "HAMMER-mount", "");
317
318 VFS_SET(hammer_vfsops, hammer, 0);
319 MODULE_VERSION(hammer, 1);
320
321 static int
322 hammer_vfs_init(struct vfsconf *conf)
323 {
324         int n;
325
326         if (hammer_limit_recs == 0) {
327                 hammer_limit_recs = nbuf * 25;
328                 n = kmalloc_limit(M_HAMMER) / 512;
329                 if (hammer_limit_recs > n)
330                         hammer_limit_recs = n;
331         }
332         if (hammer_limit_dirtybufspace == 0) {
333                 hammer_limit_dirtybufspace = hidirtybufspace / 2;
334                 if (hammer_limit_dirtybufspace < 100)
335                         hammer_limit_dirtybufspace = 100;
336         }
337
338         /*
339          * Set reasonable limits to maintain an I/O pipeline.  This is
340          * used by the flush code which explicitly initiates I/O, and
341          * is per-mount.
342          *
343          * The system-driven buffer cache uses vfs.lorunningspace and
344          * vfs.hirunningspace globally.
345          */
346         if (hammer_limit_running_io == 0)
347                 hammer_limit_running_io = hammer_limit_dirtybufspace;
348         if (hammer_limit_running_io > 10 * 1024 * 1024)
349                 hammer_limit_running_io = 10 * 1024 * 1024;
350         return(0);
351 }
352
353 static int
354 hammer_vfs_mount(struct mount *mp, char *mntpt, caddr_t data,
355                  struct ucred *cred)
356 {
357         struct hammer_mount_info info;
358         hammer_mount_t hmp;
359         hammer_volume_t rootvol;
360         struct vnode *rootvp;
361         struct vnode *devvp = NULL;
362         const char *upath;      /* volume name in userspace */
363         char *path;             /* volume name in system space */
364         int error;
365         int i;
366         int master_id;
367         char *next_volume_ptr = NULL;
368
369         /*
370          * Accept hammer_mount_info.  mntpt is NULL for root mounts at boot.
371          */
372         if (mntpt == NULL) {
373                 bzero(&info, sizeof(info));
374                 info.asof = 0;
375                 info.hflags = 0;
376                 info.nvolumes = 1;
377
378                 next_volume_ptr = mp->mnt_stat.f_mntfromname;
379
380                 /* Count number of volumes separated by ':' */
381                 for (char *p = next_volume_ptr; *p != '\0'; ++p) {
382                         if (*p == ':') {
383                                 ++info.nvolumes;
384                         }
385                 }
386
387                 mp->mnt_flag &= ~MNT_RDONLY; /* mount R/W */
388         } else {
389                 if ((error = copyin(data, &info, sizeof(info))) != 0)
390                         return (error);
391         }
392
393         /*
394          * updating or new mount
395          */
396         if (mp->mnt_flag & MNT_UPDATE) {
397                 hmp = (void *)mp->mnt_data;
398                 KKASSERT(hmp != NULL);
399         } else {
400                 if (info.nvolumes <= 0 || info.nvolumes >= 32768)
401                         return (EINVAL);
402                 hmp = NULL;
403         }
404
405         /*
406          * master-id validation.  The master id may not be changed by a
407          * mount update.
408          */
409         if (info.hflags & HMNT_MASTERID) {
410                 if (hmp && hmp->master_id != info.master_id) {
411                         kprintf("hammer: cannot change master id "
412                                 "with mount update\n");
413                         return(EINVAL);
414                 }
415                 master_id = info.master_id;
416                 if (master_id < -1 || master_id >= HAMMER_MAX_MASTERS)
417                         return (EINVAL);
418         } else {
419                 if (hmp)
420                         master_id = hmp->master_id;
421                 else
422                         master_id = 0;
423         }
424
425         /*
426          * Internal mount data structure
427          */
428         if (hmp == NULL) {
429                 hmp = kmalloc(sizeof(*hmp), M_HAMMER, M_WAITOK | M_ZERO);
430                 mp->mnt_data = (qaddr_t)hmp;
431                 hmp->mp = mp;
432                 /*TAILQ_INIT(&hmp->recycle_list);*/
433
434                 /*
435                  * Make sure kmalloc type limits are set appropriately.
436                  *
437                  * Our inode kmalloc group is sized based on maxvnodes
438                  * (controlled by the system, not us).
439                  */
440                 kmalloc_create(&hmp->m_misc, "HAMMER-others");
441                 kmalloc_create(&hmp->m_inodes, "HAMMER-inodes");
442
443                 kmalloc_raise_limit(hmp->m_inodes, 0);  /* unlimited */
444
445                 hmp->root_btree_beg.localization = 0x00000000U;
446                 hmp->root_btree_beg.obj_id = -0x8000000000000000LL;
447                 hmp->root_btree_beg.key = -0x8000000000000000LL;
448                 hmp->root_btree_beg.create_tid = 1;
449                 hmp->root_btree_beg.delete_tid = 1;
450                 hmp->root_btree_beg.rec_type = 0;
451                 hmp->root_btree_beg.obj_type = 0;
452
453                 hmp->root_btree_end.localization = 0xFFFFFFFFU;
454                 hmp->root_btree_end.obj_id = 0x7FFFFFFFFFFFFFFFLL;
455                 hmp->root_btree_end.key = 0x7FFFFFFFFFFFFFFFLL;
456                 hmp->root_btree_end.create_tid = 0xFFFFFFFFFFFFFFFFULL;
457                 hmp->root_btree_end.delete_tid = 0;   /* special case */
458                 hmp->root_btree_end.rec_type = 0xFFFFU;
459                 hmp->root_btree_end.obj_type = 0;
460
461                 hmp->krate.freq = 1;    /* maximum reporting rate (hz) */
462                 hmp->krate.count = -16; /* initial burst */
463
464                 hmp->sync_lock.refs = 1;
465                 hmp->free_lock.refs = 1;
466                 hmp->undo_lock.refs = 1;
467                 hmp->blkmap_lock.refs = 1;
468                 hmp->snapshot_lock.refs = 1;
469                 hmp->volume_lock.refs = 1;
470
471                 TAILQ_INIT(&hmp->delay_list);
472                 TAILQ_INIT(&hmp->flush_group_list);
473                 TAILQ_INIT(&hmp->objid_cache_list);
474                 TAILQ_INIT(&hmp->undo_lru_list);
475                 TAILQ_INIT(&hmp->reclaim_list);
476
477                 RB_INIT(&hmp->rb_dedup_crc_root);
478                 RB_INIT(&hmp->rb_dedup_off_root);       
479                 TAILQ_INIT(&hmp->dedup_lru_list);
480         }
481         hmp->hflags &= ~HMNT_USERFLAGS;
482         hmp->hflags |= info.hflags & HMNT_USERFLAGS;
483
484         hmp->master_id = master_id;
485
486         if (info.asof) {
487                 mp->mnt_flag |= MNT_RDONLY;
488                 hmp->asof = info.asof;
489         } else {
490                 hmp->asof = HAMMER_MAX_TID;
491         }
492
493         hmp->volume_to_remove = -1;
494
495         /*
496          * Re-open read-write if originally read-only, or vise-versa.
497          *
498          * When going from read-only to read-write execute the stage2
499          * recovery if it has not already been run.
500          */
501         if (mp->mnt_flag & MNT_UPDATE) {
502                 lwkt_gettoken(&hmp->fs_token);
503                 error = 0;
504                 if (hmp->ronly && (mp->mnt_kern_flag & MNTK_WANTRDWR)) {
505                         kprintf("HAMMER read-only -> read-write\n");
506                         hmp->ronly = 0;
507                         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
508                                 hammer_adjust_volume_mode, NULL);
509                         rootvol = hammer_get_root_volume(hmp, &error);
510                         if (rootvol) {
511                                 hammer_recover_flush_buffers(hmp, rootvol, 1);
512                                 error = hammer_recover_stage2(hmp, rootvol);
513                                 bcopy(rootvol->ondisk->vol0_blockmap,
514                                       hmp->blockmap,
515                                       sizeof(hmp->blockmap));
516                                 hammer_rel_volume(rootvol, 0);
517                         }
518                         RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
519                                 hammer_reload_inode, NULL);
520                         /* kernel clears MNT_RDONLY */
521                 } else if (hmp->ronly == 0 && (mp->mnt_flag & MNT_RDONLY)) {
522                         kprintf("HAMMER read-write -> read-only\n");
523                         hmp->ronly = 1; /* messy */
524                         RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
525                                 hammer_reload_inode, NULL);
526                         hmp->ronly = 0;
527                         hammer_flusher_sync(hmp);
528                         hammer_flusher_sync(hmp);
529                         hammer_flusher_sync(hmp);
530                         hmp->ronly = 1;
531                         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
532                                 hammer_adjust_volume_mode, NULL);
533                 }
534                 lwkt_reltoken(&hmp->fs_token);
535                 return(error);
536         }
537
538         RB_INIT(&hmp->rb_vols_root);
539         RB_INIT(&hmp->rb_inos_root);
540         RB_INIT(&hmp->rb_redo_root);
541         RB_INIT(&hmp->rb_nods_root);
542         RB_INIT(&hmp->rb_undo_root);
543         RB_INIT(&hmp->rb_resv_root);
544         RB_INIT(&hmp->rb_bufs_root);
545         RB_INIT(&hmp->rb_pfsm_root);
546
547         hmp->ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
548
549         TAILQ_INIT(&hmp->volu_list);
550         TAILQ_INIT(&hmp->undo_list);
551         TAILQ_INIT(&hmp->data_list);
552         TAILQ_INIT(&hmp->meta_list);
553         TAILQ_INIT(&hmp->lose_list);
554         TAILQ_INIT(&hmp->iorun_list);
555
556         lwkt_token_init(&hmp->fs_token, 1, "hammerfs");
557         lwkt_token_init(&hmp->io_token, 1, "hammerio");
558
559         lwkt_gettoken(&hmp->fs_token);
560
561         /*
562          * Load volumes
563          */
564         path = objcache_get(namei_oc, M_WAITOK);
565         hmp->nvolumes = -1;
566         for (i = 0; i < info.nvolumes; ++i) {
567                 if (mntpt == NULL) {
568                         /*
569                          * Root mount.
570                          */
571                         KKASSERT(next_volume_ptr != NULL);
572                         strcpy(path, "");
573                         if (*next_volume_ptr != '/') {
574                                 /* relative path */
575                                 strcpy(path, "/dev/");
576                         }
577                         int k;
578                         for (k = strlen(path); k < MAXPATHLEN-1; ++k) {
579                                 if (*next_volume_ptr == '\0') {
580                                         break;
581                                 } else if (*next_volume_ptr == ':') {
582                                         ++next_volume_ptr;
583                                         break;
584                                 } else {
585                                         path[k] = *next_volume_ptr;
586                                         ++next_volume_ptr;
587                                 }
588                         }
589                         path[k] = '\0';
590
591                         error = 0;
592                         cdev_t dev = kgetdiskbyname(path);
593                         error = bdevvp(dev, &devvp);
594                         if (error) {
595                                 kprintf("hammer_mountroot: can't find devvp\n");
596                         }
597                 } else {
598                         error = copyin(&info.volumes[i], &upath,
599                                        sizeof(char *));
600                         if (error == 0)
601                                 error = copyinstr(upath, path,
602                                                   MAXPATHLEN, NULL);
603                 }
604                 if (error == 0)
605                         error = hammer_install_volume(hmp, path, devvp);
606                 if (error)
607                         break;
608         }
609         objcache_put(namei_oc, path);
610
611         /*
612          * Make sure we found a root volume
613          */
614         if (error == 0 && hmp->rootvol == NULL) {
615                 kprintf("hammer_mount: No root volume found!\n");
616                 error = EINVAL;
617         }
618
619         /*
620          * Check that all required volumes are available
621          */
622         if (error == 0 && hammer_mountcheck_volumes(hmp)) {
623                 kprintf("hammer_mount: Missing volumes, cannot mount!\n");
624                 error = EINVAL;
625         }
626
627         if (error) {
628                 /* called with fs_token held */
629                 hammer_free_hmp(mp);
630                 return (error);
631         }
632
633         /*
634          * No errors, setup enough of the mount point so we can lookup the
635          * root vnode.
636          */
637         mp->mnt_iosize_max = MAXPHYS;
638         mp->mnt_kern_flag |= MNTK_FSMID;
639
640         /*
641          * MPSAFE code.  Note that VOPs and VFSops which are not MPSAFE
642          * will acquire a per-mount token prior to entry and release it
643          * on return, so even if we do not specify it we no longer get
644          * the BGL regardlless of how we are flagged.
645          */
646         mp->mnt_kern_flag |= MNTK_ALL_MPSAFE;
647         /*MNTK_RD_MPSAFE | MNTK_GA_MPSAFE | MNTK_IN_MPSAFE;*/
648
649         /* 
650          * note: f_iosize is used by vnode_pager_haspage() when constructing
651          * its VOP_BMAP call.
652          */
653         mp->mnt_stat.f_iosize = HAMMER_BUFSIZE;
654         mp->mnt_stat.f_bsize = HAMMER_BUFSIZE;
655
656         mp->mnt_vstat.f_frsize = HAMMER_BUFSIZE;
657         mp->mnt_vstat.f_bsize = HAMMER_BUFSIZE;
658
659         mp->mnt_maxsymlinklen = 255;
660         mp->mnt_flag |= MNT_LOCAL;
661
662         vfs_add_vnodeops(mp, &hammer_vnode_vops, &mp->mnt_vn_norm_ops);
663         vfs_add_vnodeops(mp, &hammer_spec_vops, &mp->mnt_vn_spec_ops);
664         vfs_add_vnodeops(mp, &hammer_fifo_vops, &mp->mnt_vn_fifo_ops);
665
666         /*
667          * The root volume's ondisk pointer is only valid if we hold a
668          * reference to it.
669          */
670         rootvol = hammer_get_root_volume(hmp, &error);
671         if (error)
672                 goto failed;
673
674         /*
675          * Perform any necessary UNDO operations.  The recovery code does
676          * call hammer_undo_lookup() so we have to pre-cache the blockmap,
677          * and then re-copy it again after recovery is complete.
678          *
679          * If this is a read-only mount the UNDO information is retained
680          * in memory in the form of dirty buffer cache buffers, and not
681          * written back to the media.
682          */
683         bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
684               sizeof(hmp->blockmap));
685
686         /*
687          * Check filesystem version
688          */
689         hmp->version = rootvol->ondisk->vol_version;
690         if (hmp->version < HAMMER_VOL_VERSION_MIN ||
691             hmp->version > HAMMER_VOL_VERSION_MAX) {
692                 kprintf("HAMMER: mount unsupported fs version %d\n",
693                         hmp->version);
694                 error = ERANGE;
695                 goto done;
696         }
697
698         /*
699          * The undo_rec_limit limits the size of flush groups to avoid
700          * blowing out the UNDO FIFO.  This calculation is typically in
701          * the tens of thousands and is designed primarily when small
702          * HAMMER filesystems are created.
703          */
704         hmp->undo_rec_limit = hammer_undo_max(hmp) / 8192 + 100;
705         if (hammer_debug_general & 0x0001)
706                 kprintf("HAMMER: undo_rec_limit %d\n", hmp->undo_rec_limit);
707
708         /*
709          * NOTE: Recover stage1 not only handles meta-data recovery, it
710          *       also sets hmp->undo_seqno for HAMMER VERSION 4+ filesystems.
711          */
712         error = hammer_recover_stage1(hmp, rootvol);
713         if (error) {
714                 kprintf("Failed to recover HAMMER filesystem on mount\n");
715                 goto done;
716         }
717
718         /*
719          * Finish setup now that we have a good root volume.
720          *
721          * The top 16 bits of fsid.val[1] is a pfs id.
722          */
723         ksnprintf(mp->mnt_stat.f_mntfromname,
724                   sizeof(mp->mnt_stat.f_mntfromname), "%s",
725                   rootvol->ondisk->vol_name);
726         mp->mnt_stat.f_fsid.val[0] =
727                 crc32((char *)&rootvol->ondisk->vol_fsid + 0, 8);
728         mp->mnt_stat.f_fsid.val[1] =
729                 crc32((char *)&rootvol->ondisk->vol_fsid + 8, 8);
730         mp->mnt_stat.f_fsid.val[1] &= 0x0000FFFF;
731
732         mp->mnt_vstat.f_fsid_uuid = rootvol->ondisk->vol_fsid;
733         mp->mnt_vstat.f_fsid = crc32(&mp->mnt_vstat.f_fsid_uuid,
734                                      sizeof(mp->mnt_vstat.f_fsid_uuid));
735
736         /*
737          * Certain often-modified fields in the root volume are cached in
738          * the hammer_mount structure so we do not have to generate lots
739          * of little UNDO structures for them.
740          *
741          * Recopy after recovery.  This also has the side effect of
742          * setting our cached undo FIFO's first_offset, which serves to
743          * placemark the FIFO start for the NEXT flush cycle while the
744          * on-disk first_offset represents the LAST flush cycle.
745          */
746         hmp->next_tid = rootvol->ondisk->vol0_next_tid;
747         hmp->flush_tid1 = hmp->next_tid;
748         hmp->flush_tid2 = hmp->next_tid;
749         bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
750               sizeof(hmp->blockmap));
751         hmp->copy_stat_freebigblocks = rootvol->ondisk->vol0_stat_freebigblocks;
752
753         hammer_flusher_create(hmp);
754
755         /*
756          * Locate the root directory using the root cluster's B-Tree as a
757          * starting point.  The root directory uses an obj_id of 1.
758          *
759          * FUTURE: Leave the root directory cached referenced but unlocked
760          * in hmp->rootvp (need to flush it on unmount).
761          */
762         error = hammer_vfs_vget(mp, NULL, 1, &rootvp);
763         if (error)
764                 goto done;
765         vput(rootvp);
766         /*vn_unlock(hmp->rootvp);*/
767         if (hmp->ronly == 0)
768                 error = hammer_recover_stage2(hmp, rootvol);
769
770         /*
771          * If the stage2 recovery fails be sure to clean out all cached
772          * vnodes before throwing away the mount structure or bad things
773          * will happen.
774          */
775         if (error)
776                 vflush(mp, 0, 0);
777
778 done:
779         hammer_rel_volume(rootvol, 0);
780 failed:
781         /*
782          * Cleanup and return.
783          */
784         if (error) {
785                 /* called with fs_token held */
786                 hammer_free_hmp(mp);
787         } else {
788                 lwkt_reltoken(&hmp->fs_token);
789         }
790         return (error);
791 }
792
793 static int
794 hammer_vfs_unmount(struct mount *mp, int mntflags)
795 {
796         hammer_mount_t hmp = (void *)mp->mnt_data;
797         int flags;
798         int error;
799
800         /*
801          * Clean out the vnodes
802          */
803         lwkt_gettoken(&hmp->fs_token);
804         flags = 0;
805         if (mntflags & MNT_FORCE)
806                 flags |= FORCECLOSE;
807         error = vflush(mp, 0, flags);
808
809         /*
810          * Clean up the internal mount structure and related entities.  This
811          * may issue I/O.
812          */
813         if (error == 0) {
814                 /* called with fs_token held */
815                 hammer_free_hmp(mp);
816         } else {
817                 lwkt_reltoken(&hmp->fs_token);
818         }
819         return(error);
820 }
821
822 /*
823  * Clean up the internal mount structure and disassociate it from the mount.
824  * This may issue I/O.
825  *
826  * Called with fs_token held.
827  */
828 static void
829 hammer_free_hmp(struct mount *mp)
830 {
831         hammer_mount_t hmp = (void *)mp->mnt_data;
832         hammer_flush_group_t flg;
833         int count;
834         int dummy;
835
836         /*
837          * Flush anything dirty.  This won't even run if the
838          * filesystem errored-out.
839          */
840         count = 0;
841         while (hammer_flusher_haswork(hmp)) {
842                 hammer_flusher_sync(hmp);
843                 ++count;
844                 if (count >= 5) {
845                         if (count == 5)
846                                 kprintf("HAMMER: umount flushing.");
847                         else
848                                 kprintf(".");
849                         tsleep(&dummy, 0, "hmrufl", hz);
850                 }
851                 if (count == 30) {
852                         kprintf("giving up\n");
853                         break;
854                 }
855         }
856         if (count >= 5 && count < 30)
857                 kprintf("\n");
858
859         /*
860          * If the mount had a critical error we have to destroy any
861          * remaining inodes before we can finish cleaning up the flusher.
862          */
863         if (hmp->flags & HAMMER_MOUNT_CRITICAL_ERROR) {
864                 RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
865                         hammer_destroy_inode_callback, NULL);
866         }
867
868         /*
869          * There shouldn't be any inodes left now and any left over
870          * flush groups should now be empty.
871          */
872         KKASSERT(RB_EMPTY(&hmp->rb_inos_root));
873         while ((flg = TAILQ_FIRST(&hmp->flush_group_list)) != NULL) {
874                 TAILQ_REMOVE(&hmp->flush_group_list, flg, flush_entry);
875                 KKASSERT(RB_EMPTY(&flg->flush_tree));
876                 if (flg->refs) {
877                         kprintf("HAMMER: Warning, flush_group %p was "
878                                 "not empty on umount!\n", flg);
879                 }
880                 kfree(flg, hmp->m_misc);
881         }
882
883         /*
884          * We can finally destroy the flusher
885          */
886         hammer_flusher_destroy(hmp);
887
888         /*
889          * We may have held recovered buffers due to a read-only mount.
890          * These must be discarded.
891          */
892         if (hmp->ronly)
893                 hammer_recover_flush_buffers(hmp, NULL, -1);
894
895         /*
896          * Unload buffers and then volumes
897          */
898         RB_SCAN(hammer_buf_rb_tree, &hmp->rb_bufs_root, NULL,
899                 hammer_unload_buffer, NULL);
900         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
901                 hammer_unload_volume, NULL);
902
903         mp->mnt_data = NULL;
904         mp->mnt_flag &= ~MNT_LOCAL;
905         hmp->mp = NULL;
906         hammer_destroy_objid_cache(hmp);
907         hammer_destroy_dedup_cache(hmp);
908         if (hmp->dedup_free_cache != NULL) {
909                 kfree(hmp->dedup_free_cache, hmp->m_misc);
910                 hmp->dedup_free_cache = NULL;
911         }
912         kmalloc_destroy(&hmp->m_misc);
913         kmalloc_destroy(&hmp->m_inodes);
914         lwkt_reltoken(&hmp->fs_token);
915         kfree(hmp, M_HAMMER);
916 }
917
918 /*
919  * Report critical errors.  ip may be NULL.
920  */
921 void
922 hammer_critical_error(hammer_mount_t hmp, hammer_inode_t ip,
923                       int error, const char *msg)
924 {
925         hmp->flags |= HAMMER_MOUNT_CRITICAL_ERROR;
926
927         krateprintf(&hmp->krate,
928                     "HAMMER(%s): Critical error inode=%jd error=%d %s\n",
929                     hmp->mp->mnt_stat.f_mntfromname,
930                     (intmax_t)(ip ? ip->obj_id : -1),
931                     error, msg);
932
933         if (hmp->ronly == 0) {
934                 hmp->ronly = 2;         /* special errored read-only mode */
935                 hmp->mp->mnt_flag |= MNT_RDONLY;
936                 kprintf("HAMMER(%s): Forcing read-only mode\n",
937                         hmp->mp->mnt_stat.f_mntfromname);
938         }
939         hmp->error = error;
940         if (hammer_debug_critical)
941                 Debugger("Entering debugger");
942 }
943
944
945 /*
946  * Obtain a vnode for the specified inode number.  An exclusively locked
947  * vnode is returned.
948  */
949 int
950 hammer_vfs_vget(struct mount *mp, struct vnode *dvp,
951                 ino_t ino, struct vnode **vpp)
952 {
953         struct hammer_transaction trans;
954         struct hammer_mount *hmp = (void *)mp->mnt_data;
955         struct hammer_inode *ip;
956         int error;
957         u_int32_t localization;
958
959         lwkt_gettoken(&hmp->fs_token);
960         hammer_simple_transaction(&trans, hmp);
961
962         /*
963          * If a directory vnode is supplied (mainly NFS) then we can acquire
964          * the PFS domain from it.  Otherwise we would only be able to vget
965          * inodes in the root PFS.
966          */
967         if (dvp) {
968                 localization = HAMMER_DEF_LOCALIZATION +
969                                 VTOI(dvp)->obj_localization;
970         } else {
971                 localization = HAMMER_DEF_LOCALIZATION;
972         }
973
974         /*
975          * Lookup the requested HAMMER inode.  The structure must be
976          * left unlocked while we manipulate the related vnode to avoid
977          * a deadlock.
978          */
979         ip = hammer_get_inode(&trans, NULL, ino,
980                               hmp->asof, localization,
981                               0, &error);
982         if (ip == NULL) {
983                 *vpp = NULL;
984         } else {
985                 error = hammer_get_vnode(ip, vpp);
986                 hammer_rel_inode(ip, 0);
987         }
988         hammer_done_transaction(&trans);
989         lwkt_reltoken(&hmp->fs_token);
990         return (error);
991 }
992
993 /*
994  * Return the root vnode for the filesystem.
995  *
996  * HAMMER stores the root vnode in the hammer_mount structure so
997  * getting it is easy.
998  */
999 static int
1000 hammer_vfs_root(struct mount *mp, struct vnode **vpp)
1001 {
1002         int error;
1003
1004         error = hammer_vfs_vget(mp, NULL, 1, vpp);
1005         return (error);
1006 }
1007
1008 static int
1009 hammer_vfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
1010 {
1011         struct hammer_mount *hmp = (void *)mp->mnt_data;
1012         hammer_volume_t volume;
1013         hammer_volume_ondisk_t ondisk;
1014         int error;
1015         int64_t bfree;
1016         int64_t breserved;
1017
1018         lwkt_gettoken(&hmp->fs_token);
1019         volume = hammer_get_root_volume(hmp, &error);
1020         if (error) {
1021                 lwkt_reltoken(&hmp->fs_token);
1022                 return(error);
1023         }
1024         ondisk = volume->ondisk;
1025
1026         /*
1027          * Basic stats
1028          */
1029         _hammer_checkspace(hmp, HAMMER_CHKSPC_WRITE, &breserved);
1030         mp->mnt_stat.f_files = ondisk->vol0_stat_inodes;
1031         bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE;
1032         hammer_rel_volume(volume, 0);
1033
1034         mp->mnt_stat.f_bfree = (bfree - breserved) / HAMMER_BUFSIZE;
1035         mp->mnt_stat.f_bavail = mp->mnt_stat.f_bfree;
1036         if (mp->mnt_stat.f_files < 0)
1037                 mp->mnt_stat.f_files = 0;
1038
1039         *sbp = mp->mnt_stat;
1040         lwkt_reltoken(&hmp->fs_token);
1041         return(0);
1042 }
1043
1044 static int
1045 hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred)
1046 {
1047         struct hammer_mount *hmp = (void *)mp->mnt_data;
1048         hammer_volume_t volume;
1049         hammer_volume_ondisk_t ondisk;
1050         int error;
1051         int64_t bfree;
1052         int64_t breserved;
1053
1054         lwkt_gettoken(&hmp->fs_token);
1055         volume = hammer_get_root_volume(hmp, &error);
1056         if (error) {
1057                 lwkt_reltoken(&hmp->fs_token);
1058                 return(error);
1059         }
1060         ondisk = volume->ondisk;
1061
1062         /*
1063          * Basic stats
1064          */
1065         _hammer_checkspace(hmp, HAMMER_CHKSPC_WRITE, &breserved);
1066         mp->mnt_vstat.f_files = ondisk->vol0_stat_inodes;
1067         bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE;
1068         hammer_rel_volume(volume, 0);
1069
1070         mp->mnt_vstat.f_bfree = (bfree - breserved) / HAMMER_BUFSIZE;
1071         mp->mnt_vstat.f_bavail = mp->mnt_vstat.f_bfree;
1072         if (mp->mnt_vstat.f_files < 0)
1073                 mp->mnt_vstat.f_files = 0;
1074         *sbp = mp->mnt_vstat;
1075         lwkt_reltoken(&hmp->fs_token);
1076         return(0);
1077 }
1078
1079 /*
1080  * Sync the filesystem.  Currently we have to run it twice, the second
1081  * one will advance the undo start index to the end index, so if a crash
1082  * occurs no undos will be run on mount.
1083  *
1084  * We do not sync the filesystem if we are called from a panic.  If we did
1085  * we might end up blowing up a sync that was already in progress.
1086  */
1087 static int
1088 hammer_vfs_sync(struct mount *mp, int waitfor)
1089 {
1090         struct hammer_mount *hmp = (void *)mp->mnt_data;
1091         int error;
1092
1093         lwkt_gettoken(&hmp->fs_token);
1094         if (panicstr == NULL) {
1095                 error = hammer_sync_hmp(hmp, waitfor);
1096         } else {
1097                 error = EIO;
1098         }
1099         lwkt_reltoken(&hmp->fs_token);
1100         return (error);
1101 }
1102
1103 /*
1104  * Convert a vnode to a file handle.
1105  *
1106  * Accesses read-only fields on already-referenced structures so
1107  * no token is needed.
1108  */
1109 static int
1110 hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp)
1111 {
1112         hammer_inode_t ip;
1113
1114         KKASSERT(MAXFIDSZ >= 16);
1115         ip = VTOI(vp);
1116         fhp->fid_len = offsetof(struct fid, fid_data[16]);
1117         fhp->fid_ext = ip->obj_localization >> 16;
1118         bcopy(&ip->obj_id, fhp->fid_data + 0, sizeof(ip->obj_id));
1119         bcopy(&ip->obj_asof, fhp->fid_data + 8, sizeof(ip->obj_asof));
1120         return(0);
1121 }
1122
1123
1124 /*
1125  * Convert a file handle back to a vnode.
1126  *
1127  * Use rootvp to enforce PFS isolation when a PFS is exported via a
1128  * null mount.
1129  */
1130 static int
1131 hammer_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
1132                   struct fid *fhp, struct vnode **vpp)
1133 {
1134         hammer_mount_t hmp = (void *)mp->mnt_data;
1135         struct hammer_transaction trans;
1136         struct hammer_inode *ip;
1137         struct hammer_inode_info info;
1138         int error;
1139         u_int32_t localization;
1140
1141         bcopy(fhp->fid_data + 0, &info.obj_id, sizeof(info.obj_id));
1142         bcopy(fhp->fid_data + 8, &info.obj_asof, sizeof(info.obj_asof));
1143         if (rootvp)
1144                 localization = VTOI(rootvp)->obj_localization;
1145         else
1146                 localization = (u_int32_t)fhp->fid_ext << 16;
1147
1148         lwkt_gettoken(&hmp->fs_token);
1149         hammer_simple_transaction(&trans, hmp);
1150
1151         /*
1152          * Get/allocate the hammer_inode structure.  The structure must be
1153          * unlocked while we manipulate the related vnode to avoid a
1154          * deadlock.
1155          */
1156         ip = hammer_get_inode(&trans, NULL, info.obj_id,
1157                               info.obj_asof, localization, 0, &error);
1158         if (ip) {
1159                 error = hammer_get_vnode(ip, vpp);
1160                 hammer_rel_inode(ip, 0);
1161         } else {
1162                 *vpp = NULL;
1163         }
1164         hammer_done_transaction(&trans);
1165         lwkt_reltoken(&hmp->fs_token);
1166         return (error);
1167 }
1168
1169 static int
1170 hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
1171                     int *exflagsp, struct ucred **credanonp)
1172 {
1173         hammer_mount_t hmp = (void *)mp->mnt_data;
1174         struct netcred *np;
1175         int error;
1176
1177         lwkt_gettoken(&hmp->fs_token);
1178         np = vfs_export_lookup(mp, &hmp->export, nam);
1179         if (np) {
1180                 *exflagsp = np->netc_exflags;
1181                 *credanonp = &np->netc_anon;
1182                 error = 0;
1183         } else {
1184                 error = EACCES;
1185         }
1186         lwkt_reltoken(&hmp->fs_token);
1187         return (error);
1188
1189 }
1190
1191 int
1192 hammer_vfs_export(struct mount *mp, int op, const struct export_args *export)
1193 {
1194         hammer_mount_t hmp = (void *)mp->mnt_data;
1195         int error;
1196
1197         lwkt_gettoken(&hmp->fs_token);
1198
1199         switch(op) {
1200         case MOUNTCTL_SET_EXPORT:
1201                 error = vfs_export(mp, &hmp->export, export);
1202                 break;
1203         default:
1204                 error = EOPNOTSUPP;
1205                 break;
1206         }
1207         lwkt_reltoken(&hmp->fs_token);
1208
1209         return(error);
1210 }
1211