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