xref: /dflybsd-src/sys/vfs/hammer/hammer_vfsops.c (revision 28637087e34e261812af71c17a10136d03b207b6)
1 /*
2  * Copyright (c) 2007-2008 The DragonFly Project.  All rights reserved.
3  *
4  * This code is derived from software contributed to The DragonFly Project
5  * by Matthew Dillon <dillon@backplane.com>
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  *
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in
15  *    the documentation and/or other materials provided with the
16  *    distribution.
17  * 3. Neither the name of The DragonFly Project nor the names of its
18  *    contributors may be used to endorse or promote products derived
19  *    from this software without specific, prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
25  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
27  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
29  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  */
34 
35 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/kernel.h>
38 #include <sys/vnode.h>
39 #include <sys/mount.h>
40 #include <sys/malloc.h>
41 #include <sys/nlookup.h>
42 #include <sys/fcntl.h>
43 #include <sys/sysctl.h>
44 #include <sys/buf.h>
45 #include <sys/buf2.h>
46 #include "hammer.h"
47 
48 /*
49  * NOTE!  Global statistics may not be MPSAFE so HAMMER never uses them
50  *	  in conditionals.
51  */
52 int hammer_supported_version = HAMMER_VOL_VERSION_DEFAULT;
53 int hammer_debug_io;
54 int hammer_debug_general;
55 int hammer_debug_debug = 1;		/* medium-error panics */
56 int hammer_debug_inode;
57 int hammer_debug_locks;
58 int hammer_debug_btree;
59 int hammer_debug_tid;
60 int hammer_debug_recover;		/* -1 will disable, +1 will force */
61 int hammer_debug_recover_faults;
62 int hammer_debug_critical;		/* non-zero enter debugger on error */
63 int hammer_cluster_enable = 1;		/* enable read clustering by default */
64 int hammer_live_dedup = 0;
65 int hammer_tdmux_ticks;
66 int hammer_count_fsyncs;
67 int hammer_count_inodes;
68 int hammer_count_iqueued;
69 int hammer_count_reclaims;
70 int hammer_count_records;
71 int hammer_count_record_datas;
72 int hammer_count_volumes;
73 int hammer_count_buffers;
74 int hammer_count_nodes;
75 int64_t hammer_count_extra_space_used;
76 int64_t hammer_stats_btree_lookups;
77 int64_t hammer_stats_btree_searches;
78 int64_t hammer_stats_btree_inserts;
79 int64_t hammer_stats_btree_deletes;
80 int64_t hammer_stats_btree_elements;
81 int64_t hammer_stats_btree_splits;
82 int64_t hammer_stats_btree_iterations;
83 int64_t hammer_stats_btree_root_iterations;
84 int64_t hammer_stats_record_iterations;
85 
86 int64_t hammer_stats_file_read;
87 int64_t hammer_stats_file_write;
88 int64_t hammer_stats_file_iopsr;
89 int64_t hammer_stats_file_iopsw;
90 int64_t hammer_stats_disk_read;
91 int64_t hammer_stats_disk_write;
92 int64_t hammer_stats_inode_flushes;
93 int64_t hammer_stats_commits;
94 int64_t hammer_stats_undo;
95 int64_t hammer_stats_redo;
96 
97 long hammer_count_dirtybufspace;	/* global */
98 int hammer_count_refedbufs;		/* global */
99 int hammer_count_reservations;
100 long hammer_count_io_running_read;
101 long hammer_count_io_running_write;
102 int hammer_count_io_locked;
103 long hammer_limit_dirtybufspace;	/* per-mount */
104 int hammer_limit_recs;			/* as a whole XXX */
105 int hammer_limit_inode_recs = 2048;	/* per inode */
106 int hammer_limit_reclaims;
107 int hammer_live_dedup_cache_size = DEDUP_CACHE_SIZE;
108 int hammer_limit_redo = 4096 * 1024;	/* per inode */
109 int hammer_autoflush = 500;		/* auto flush (typ on reclaim) */
110 int hammer_bio_count;
111 int hammer_verify_zone;
112 int hammer_verify_data = 1;
113 int hammer_write_mode;
114 int hammer_double_buffer;
115 int hammer_btree_full_undo = 1;
116 int hammer_yield_check = 16;
117 int hammer_fsync_mode = 3;
118 int64_t hammer_contention_count;
119 int64_t hammer_zone_limit;
120 
121 /*
122  * Live dedup debug counters (sysctls are writable so that counters
123  * can be reset from userspace).
124  */
125 int64_t hammer_live_dedup_vnode_bcmps = 0;
126 int64_t hammer_live_dedup_device_bcmps = 0;
127 int64_t hammer_live_dedup_findblk_failures = 0;
128 int64_t hammer_live_dedup_bmap_saves = 0;
129 
130 
131 SYSCTL_NODE(_vfs, OID_AUTO, hammer, CTLFLAG_RW, 0, "HAMMER filesystem");
132 
133 SYSCTL_INT(_vfs_hammer, OID_AUTO, supported_version, CTLFLAG_RD,
134 	   &hammer_supported_version, 0, "");
135 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_general, CTLFLAG_RW,
136 	   &hammer_debug_general, 0, "");
137 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_io, CTLFLAG_RW,
138 	   &hammer_debug_io, 0, "");
139 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_debug, CTLFLAG_RW,
140 	   &hammer_debug_debug, 0, "");
141 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_inode, CTLFLAG_RW,
142 	   &hammer_debug_inode, 0, "");
143 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_locks, CTLFLAG_RW,
144 	   &hammer_debug_locks, 0, "");
145 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_btree, CTLFLAG_RW,
146 	   &hammer_debug_btree, 0, "");
147 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_tid, CTLFLAG_RW,
148 	   &hammer_debug_tid, 0, "");
149 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover, CTLFLAG_RW,
150 	   &hammer_debug_recover, 0, "");
151 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover_faults, CTLFLAG_RW,
152 	   &hammer_debug_recover_faults, 0, "");
153 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_critical, CTLFLAG_RW,
154 	   &hammer_debug_critical, 0, "");
155 SYSCTL_INT(_vfs_hammer, OID_AUTO, cluster_enable, CTLFLAG_RW,
156 	   &hammer_cluster_enable, 0, "");
157 /*
158  * 0 - live dedup is disabled
159  * 1 - dedup cache is populated on reads only
160  * 2 - dedup cache is populated on both reads and writes
161  *
162  * LIVE_DEDUP IS DISABLED PERMANENTLY!  This feature appears to cause
163  * blockmap corruption over time so we've turned it off permanently.
164  */
165 SYSCTL_INT(_vfs_hammer, OID_AUTO, live_dedup, CTLFLAG_RD,
166 	   &hammer_live_dedup, 0, "Enable live dedup (experimental)");
167 SYSCTL_INT(_vfs_hammer, OID_AUTO, tdmux_ticks, CTLFLAG_RW,
168 	   &hammer_tdmux_ticks, 0, "Hammer tdmux ticks");
169 
170 SYSCTL_LONG(_vfs_hammer, OID_AUTO, limit_dirtybufspace, CTLFLAG_RW,
171 	   &hammer_limit_dirtybufspace, 0, "");
172 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_recs, CTLFLAG_RW,
173 	   &hammer_limit_recs, 0, "");
174 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_inode_recs, CTLFLAG_RW,
175 	   &hammer_limit_inode_recs, 0, "");
176 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_reclaims, CTLFLAG_RW,
177 	   &hammer_limit_reclaims, 0, "");
178 SYSCTL_INT(_vfs_hammer, OID_AUTO, live_dedup_cache_size, CTLFLAG_RW,
179 	   &hammer_live_dedup_cache_size, 0,
180 	   "Number of cache entries");
181 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_redo, CTLFLAG_RW,
182 	   &hammer_limit_redo, 0, "");
183 
184 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_fsyncs, CTLFLAG_RD,
185 	   &hammer_count_fsyncs, 0, "");
186 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_inodes, CTLFLAG_RD,
187 	   &hammer_count_inodes, 0, "");
188 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_iqueued, CTLFLAG_RD,
189 	   &hammer_count_iqueued, 0, "");
190 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reclaims, CTLFLAG_RD,
191 	   &hammer_count_reclaims, 0, "");
192 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_records, CTLFLAG_RD,
193 	   &hammer_count_records, 0, "");
194 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_record_datas, CTLFLAG_RD,
195 	   &hammer_count_record_datas, 0, "");
196 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_volumes, CTLFLAG_RD,
197 	   &hammer_count_volumes, 0, "");
198 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_buffers, CTLFLAG_RD,
199 	   &hammer_count_buffers, 0, "");
200 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_nodes, CTLFLAG_RD,
201 	   &hammer_count_nodes, 0, "");
202 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, count_extra_space_used, CTLFLAG_RD,
203 	   &hammer_count_extra_space_used, 0, "");
204 
205 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_searches, CTLFLAG_RD,
206 	   &hammer_stats_btree_searches, 0, "");
207 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_lookups, CTLFLAG_RD,
208 	   &hammer_stats_btree_lookups, 0, "");
209 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_inserts, CTLFLAG_RD,
210 	   &hammer_stats_btree_inserts, 0, "");
211 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_deletes, CTLFLAG_RD,
212 	   &hammer_stats_btree_deletes, 0, "");
213 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_elements, CTLFLAG_RD,
214 	   &hammer_stats_btree_elements, 0, "");
215 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_splits, CTLFLAG_RD,
216 	   &hammer_stats_btree_splits, 0, "");
217 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_iterations, CTLFLAG_RD,
218 	   &hammer_stats_btree_iterations, 0, "");
219 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_root_iterations, CTLFLAG_RD,
220 	   &hammer_stats_btree_root_iterations, 0, "");
221 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_record_iterations, CTLFLAG_RD,
222 	   &hammer_stats_record_iterations, 0, "");
223 
224 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_read, CTLFLAG_RD,
225 	   &hammer_stats_file_read, 0, "");
226 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_write, CTLFLAG_RD,
227 	   &hammer_stats_file_write, 0, "");
228 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_iopsr, CTLFLAG_RD,
229 	   &hammer_stats_file_iopsr, 0, "");
230 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_iopsw, CTLFLAG_RD,
231 	   &hammer_stats_file_iopsw, 0, "");
232 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_disk_read, CTLFLAG_RD,
233 	   &hammer_stats_disk_read, 0, "");
234 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_disk_write, CTLFLAG_RD,
235 	   &hammer_stats_disk_write, 0, "");
236 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_inode_flushes, CTLFLAG_RD,
237 	   &hammer_stats_inode_flushes, 0, "");
238 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_commits, CTLFLAG_RD,
239 	   &hammer_stats_commits, 0, "");
240 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_undo, CTLFLAG_RD,
241 	   &hammer_stats_undo, 0, "");
242 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_redo, CTLFLAG_RD,
243 	   &hammer_stats_redo, 0, "");
244 
245 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, live_dedup_vnode_bcmps, CTLFLAG_RW,
246 	    &hammer_live_dedup_vnode_bcmps, 0,
247 	    "successful vnode buffer comparisons");
248 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, live_dedup_device_bcmps, CTLFLAG_RW,
249 	    &hammer_live_dedup_device_bcmps, 0,
250 	    "successful device buffer comparisons");
251 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, live_dedup_findblk_failures, CTLFLAG_RW,
252 	    &hammer_live_dedup_findblk_failures, 0,
253 	    "block lookup failures for comparison");
254 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, live_dedup_bmap_saves, CTLFLAG_RW,
255 	    &hammer_live_dedup_bmap_saves, 0,
256 	    "useful physical block lookups");
257 
258 SYSCTL_LONG(_vfs_hammer, OID_AUTO, count_dirtybufspace, CTLFLAG_RD,
259 	   &hammer_count_dirtybufspace, 0, "");
260 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_refedbufs, CTLFLAG_RD,
261 	   &hammer_count_refedbufs, 0, "");
262 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reservations, CTLFLAG_RD,
263 	   &hammer_count_reservations, 0, "");
264 SYSCTL_LONG(_vfs_hammer, OID_AUTO, count_io_running_read, CTLFLAG_RD,
265 	   &hammer_count_io_running_read, 0, "");
266 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_locked, CTLFLAG_RD,
267 	   &hammer_count_io_locked, 0, "");
268 SYSCTL_LONG(_vfs_hammer, OID_AUTO, count_io_running_write, CTLFLAG_RD,
269 	   &hammer_count_io_running_write, 0, "");
270 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, zone_limit, CTLFLAG_RW,
271 	   &hammer_zone_limit, 0, "");
272 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, contention_count, CTLFLAG_RW,
273 	   &hammer_contention_count, 0, "");
274 SYSCTL_INT(_vfs_hammer, OID_AUTO, autoflush, CTLFLAG_RW,
275 	   &hammer_autoflush, 0, "");
276 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_zone, CTLFLAG_RW,
277 	   &hammer_verify_zone, 0, "");
278 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_data, CTLFLAG_RW,
279 	   &hammer_verify_data, 0, "");
280 SYSCTL_INT(_vfs_hammer, OID_AUTO, write_mode, CTLFLAG_RW,
281 	   &hammer_write_mode, 0, "");
282 SYSCTL_INT(_vfs_hammer, OID_AUTO, double_buffer, CTLFLAG_RW,
283 	   &hammer_double_buffer, 0, "");
284 SYSCTL_INT(_vfs_hammer, OID_AUTO, btree_full_undo, CTLFLAG_RW,
285 	   &hammer_btree_full_undo, 0, "");
286 SYSCTL_INT(_vfs_hammer, OID_AUTO, yield_check, CTLFLAG_RW,
287 	   &hammer_yield_check, 0, "");
288 SYSCTL_INT(_vfs_hammer, OID_AUTO, fsync_mode, CTLFLAG_RW,
289 	   &hammer_fsync_mode, 0, "");
290 
291 /* KTR_INFO_MASTER(hammer); */
292 
293 /*
294  * VFS ABI
295  */
296 static void	hammer_free_hmp(struct mount *mp);
297 
298 static int	hammer_vfs_mount(struct mount *mp, char *path, caddr_t data,
299 				struct ucred *cred);
300 static int	hammer_vfs_unmount(struct mount *mp, int mntflags);
301 static int	hammer_vfs_root(struct mount *mp, struct vnode **vpp);
302 static int	hammer_vfs_statfs(struct mount *mp, struct statfs *sbp,
303 				struct ucred *cred);
304 static int	hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp,
305 				struct ucred *cred);
306 static int	hammer_vfs_sync(struct mount *mp, int waitfor);
307 static int	hammer_vfs_vget(struct mount *mp, struct vnode *dvp,
308 				ino_t ino, struct vnode **vpp);
309 static int	hammer_vfs_init(struct vfsconf *conf);
310 static int	hammer_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
311 				struct fid *fhp, struct vnode **vpp);
312 static int	hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp);
313 static int	hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
314 				int *exflagsp, struct ucred **credanonp);
315 
316 
317 static struct vfsops hammer_vfsops = {
318 	.vfs_mount	= hammer_vfs_mount,
319 	.vfs_unmount	= hammer_vfs_unmount,
320 	.vfs_root	= hammer_vfs_root,
321 	.vfs_statfs	= hammer_vfs_statfs,
322 	.vfs_statvfs	= hammer_vfs_statvfs,
323 	.vfs_sync	= hammer_vfs_sync,
324 	.vfs_vget	= hammer_vfs_vget,
325 	.vfs_init	= hammer_vfs_init,
326 	.vfs_vptofh	= hammer_vfs_vptofh,
327 	.vfs_fhtovp	= hammer_vfs_fhtovp,
328 	.vfs_checkexp	= hammer_vfs_checkexp
329 };
330 
331 MALLOC_DEFINE(M_HAMMER, "HAMMER-mount", "");
332 
333 VFS_SET(hammer_vfsops, hammer, 0);
334 MODULE_VERSION(hammer, 1);
335 
336 static int
337 hammer_vfs_init(struct vfsconf *conf)
338 {
339 	long n;
340 
341 	/*
342 	 * Wait up to this long for an exclusive deadlock to clear
343 	 * before acquiring a new shared lock on the ip.  The deadlock
344 	 * may have occured on a b-tree node related to the ip.
345 	 */
346 	if (hammer_tdmux_ticks == 0)
347 		hammer_tdmux_ticks = hz / 5;
348 
349 	/*
350 	 * Autosize, but be careful because a hammer filesystem's
351 	 * reserve is partially calculated based on dirtybufspace,
352 	 * so we simply cannot allow it to get too large.
353 	 */
354 	if (hammer_limit_recs == 0) {
355 		n = nbuf * 25;
356 		if (n > kmalloc_limit(M_HAMMER) / 512)
357 			n = kmalloc_limit(M_HAMMER) / 512;
358 		if (n > 2 * 1024 * 1024)
359 			n = 2 * 1024 * 1024;
360 		hammer_limit_recs = (int)n;
361 	}
362 	if (hammer_limit_dirtybufspace == 0) {
363 		hammer_limit_dirtybufspace = hidirtybufspace / 2;
364 		if (hammer_limit_dirtybufspace < 1L * 1024 * 1024)
365 			hammer_limit_dirtybufspace = 1024L * 1024;
366 		if (hammer_limit_dirtybufspace > 1024L * 1024 * 1024)
367 			hammer_limit_dirtybufspace = 1024L * 1024 * 1024;
368 	}
369 
370 	/*
371 	 * The hammer_inode structure detaches from the vnode on reclaim.
372 	 * This limits the number of inodes in this state to prevent a
373 	 * memory pool blowout.
374 	 */
375 	if (hammer_limit_reclaims == 0)
376 		hammer_limit_reclaims = desiredvnodes / 10;
377 
378 	return(0);
379 }
380 
381 static int
382 hammer_vfs_mount(struct mount *mp, char *mntpt, caddr_t data,
383 		 struct ucred *cred)
384 {
385 	struct hammer_mount_info info;
386 	hammer_mount_t hmp;
387 	hammer_volume_t rootvol;
388 	struct vnode *rootvp;
389 	struct vnode *devvp = NULL;
390 	const char *upath;	/* volume name in userspace */
391 	char *path;		/* volume name in system space */
392 	int error;
393 	int i;
394 	int master_id;
395 	char *next_volume_ptr = NULL;
396 
397 	/*
398 	 * Accept hammer_mount_info.  mntpt is NULL for root mounts at boot.
399 	 */
400 	if (mntpt == NULL) {
401 		bzero(&info, sizeof(info));
402 		info.asof = 0;
403 		info.hflags = 0;
404 		info.nvolumes = 1;
405 
406 		next_volume_ptr = mp->mnt_stat.f_mntfromname;
407 
408 		/* Count number of volumes separated by ':' */
409 		for (char *p = next_volume_ptr; *p != '\0'; ++p) {
410 			if (*p == ':') {
411 				++info.nvolumes;
412 			}
413 		}
414 
415 		mp->mnt_flag &= ~MNT_RDONLY; /* mount R/W */
416 	} else {
417 		if ((error = copyin(data, &info, sizeof(info))) != 0)
418 			return (error);
419 	}
420 
421 	/*
422 	 * updating or new mount
423 	 */
424 	if (mp->mnt_flag & MNT_UPDATE) {
425 		hmp = (void *)mp->mnt_data;
426 		KKASSERT(hmp != NULL);
427 	} else {
428 		if (info.nvolumes <= 0 || info.nvolumes > HAMMER_MAX_VOLUMES)
429 			return (EINVAL);
430 		hmp = NULL;
431 	}
432 
433 	/*
434 	 * master-id validation.  The master id may not be changed by a
435 	 * mount update.
436 	 */
437 	if (info.hflags & HMNT_MASTERID) {
438 		if (hmp && hmp->master_id != info.master_id) {
439 			kprintf("hammer: cannot change master id "
440 				"with mount update\n");
441 			return(EINVAL);
442 		}
443 		master_id = info.master_id;
444 		if (master_id < -1 || master_id >= HAMMER_MAX_MASTERS)
445 			return (EINVAL);
446 	} else {
447 		if (hmp)
448 			master_id = hmp->master_id;
449 		else
450 			master_id = 0;
451 	}
452 
453 	/*
454 	 * Internal mount data structure
455 	 */
456 	if (hmp == NULL) {
457 		hmp = kmalloc(sizeof(*hmp), M_HAMMER, M_WAITOK | M_ZERO);
458 		mp->mnt_data = (qaddr_t)hmp;
459 		hmp->mp = mp;
460 		/*TAILQ_INIT(&hmp->recycle_list);*/
461 
462 		/*
463 		 * Make sure kmalloc type limits are set appropriately.
464 		 *
465 		 * Our inode kmalloc group is sized based on maxvnodes
466 		 * (controlled by the system, not us).
467 		 */
468 		kmalloc_create(&hmp->m_misc, "HAMMER-others");
469 		kmalloc_create(&hmp->m_inodes, "HAMMER-inodes");
470 
471 		kmalloc_raise_limit(hmp->m_inodes, 0);	/* unlimited */
472 
473 		hmp->root_btree_beg.localization = 0x00000000U;
474 		hmp->root_btree_beg.obj_id = -0x8000000000000000LL;
475 		hmp->root_btree_beg.key = -0x8000000000000000LL;
476 		hmp->root_btree_beg.create_tid = 1;
477 		hmp->root_btree_beg.delete_tid = 1;
478 		hmp->root_btree_beg.rec_type = 0;
479 		hmp->root_btree_beg.obj_type = 0;
480 		hmp->root_btree_beg.btype = HAMMER_BTREE_TYPE_NONE;
481 
482 		hmp->root_btree_end.localization = 0xFFFFFFFFU;
483 		hmp->root_btree_end.obj_id = 0x7FFFFFFFFFFFFFFFLL;
484 		hmp->root_btree_end.key = 0x7FFFFFFFFFFFFFFFLL;
485 		hmp->root_btree_end.create_tid = 0xFFFFFFFFFFFFFFFFULL;
486 		hmp->root_btree_end.delete_tid = 0;   /* special case */
487 		hmp->root_btree_end.rec_type = 0xFFFFU;
488 		hmp->root_btree_end.obj_type = 0;
489 		hmp->root_btree_end.btype = HAMMER_BTREE_TYPE_NONE;
490 
491 		hmp->krate.freq = 1;	/* maximum reporting rate (hz) */
492 		hmp->krate.count = -16;	/* initial burst */
493 		hmp->kdiag.freq = 1;	/* maximum reporting rate (hz) */
494 		hmp->kdiag.count = -16;	/* initial burst */
495 
496 		hmp->sync_lock.refs = 1;
497 		hmp->free_lock.refs = 1;
498 		hmp->undo_lock.refs = 1;
499 		hmp->blkmap_lock.refs = 1;
500 		hmp->snapshot_lock.refs = 1;
501 		hmp->volume_lock.refs = 1;
502 
503 		TAILQ_INIT(&hmp->delay_list);
504 		TAILQ_INIT(&hmp->flush_group_list);
505 		TAILQ_INIT(&hmp->objid_cache_list);
506 		TAILQ_INIT(&hmp->undo_lru_list);
507 		TAILQ_INIT(&hmp->reclaim_list);
508 
509 		RB_INIT(&hmp->rb_dedup_crc_root);
510 		RB_INIT(&hmp->rb_dedup_off_root);
511 		TAILQ_INIT(&hmp->dedup_lru_list);
512 	}
513 	hmp->hflags &= ~HMNT_USERFLAGS;
514 	hmp->hflags |= info.hflags & HMNT_USERFLAGS;
515 
516 	hmp->master_id = master_id;
517 
518 	if (info.asof) {
519 		mp->mnt_flag |= MNT_RDONLY;
520 		hmp->asof = info.asof;
521 	} else {
522 		hmp->asof = HAMMER_MAX_TID;
523 	}
524 
525 	hmp->volume_to_remove = -1;
526 
527 	/*
528 	 * Re-open read-write if originally read-only, or vise-versa.
529 	 *
530 	 * When going from read-only to read-write execute the stage2
531 	 * recovery if it has not already been run.
532 	 */
533 	if (mp->mnt_flag & MNT_UPDATE) {
534 		lwkt_gettoken(&hmp->fs_token);
535 		error = 0;
536 		if (hmp->ronly && (mp->mnt_kern_flag & MNTK_WANTRDWR)) {
537 			kprintf("HAMMER read-only -> read-write\n");
538 			hmp->ronly = 0;
539 			RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
540 				hammer_adjust_volume_mode, NULL);
541 			rootvol = hammer_get_root_volume(hmp, &error);
542 			if (rootvol) {
543 				hammer_recover_flush_buffers(hmp, rootvol, 1);
544 				error = hammer_recover_stage2(hmp, rootvol);
545 				bcopy(rootvol->ondisk->vol0_blockmap,
546 				      hmp->blockmap,
547 				      sizeof(hmp->blockmap));
548 				hammer_rel_volume(rootvol, 0);
549 			}
550 			RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
551 				hammer_reload_inode, NULL);
552 			/* kernel clears MNT_RDONLY */
553 		} else if (hmp->ronly == 0 && (mp->mnt_flag & MNT_RDONLY)) {
554 			kprintf("HAMMER read-write -> read-only\n");
555 			hmp->ronly = 1;	/* messy */
556 			RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
557 				hammer_reload_inode, NULL);
558 			hmp->ronly = 0;
559 			hammer_flusher_sync(hmp);
560 			hammer_flusher_sync(hmp);
561 			hammer_flusher_sync(hmp);
562 			hmp->ronly = 1;
563 			RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
564 				hammer_adjust_volume_mode, NULL);
565 		}
566 		lwkt_reltoken(&hmp->fs_token);
567 		return(error);
568 	}
569 
570 	RB_INIT(&hmp->rb_vols_root);
571 	RB_INIT(&hmp->rb_inos_root);
572 	RB_INIT(&hmp->rb_redo_root);
573 	RB_INIT(&hmp->rb_nods_root);
574 	RB_INIT(&hmp->rb_undo_root);
575 	RB_INIT(&hmp->rb_resv_root);
576 	RB_INIT(&hmp->rb_bufs_root);
577 	RB_INIT(&hmp->rb_pfsm_root);
578 
579 	hmp->ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
580 
581 	RB_INIT(&hmp->volu_root);
582 	RB_INIT(&hmp->undo_root);
583 	RB_INIT(&hmp->data_root);
584 	RB_INIT(&hmp->meta_root);
585 	RB_INIT(&hmp->lose_root);
586 	TAILQ_INIT(&hmp->iorun_list);
587 
588 	lwkt_token_init(&hmp->fs_token, "hammerfs");
589 	lwkt_token_init(&hmp->io_token, "hammerio");
590 
591 	lwkt_gettoken(&hmp->fs_token);
592 
593 	/*
594 	 * Load volumes
595 	 */
596 	path = objcache_get(namei_oc, M_WAITOK);
597 	hmp->nvolumes = -1;
598 	for (i = 0; i < info.nvolumes; ++i) {
599 		if (mntpt == NULL) {
600 			/*
601 			 * Root mount.
602 			 */
603 			KKASSERT(next_volume_ptr != NULL);
604 			strcpy(path, "");
605 			if (*next_volume_ptr != '/') {
606 				/* relative path */
607 				strcpy(path, "/dev/");
608 			}
609 			int k;
610 			for (k = strlen(path); k < MAXPATHLEN-1; ++k) {
611 				if (*next_volume_ptr == '\0') {
612 					break;
613 				} else if (*next_volume_ptr == ':') {
614 					++next_volume_ptr;
615 					break;
616 				} else {
617 					path[k] = *next_volume_ptr;
618 					++next_volume_ptr;
619 				}
620 			}
621 			path[k] = '\0';
622 
623 			error = 0;
624 			cdev_t dev = kgetdiskbyname(path);
625 			error = bdevvp(dev, &devvp);
626 			if (error) {
627 				kprintf("hammer_mountroot: can't find devvp\n");
628 			}
629 		} else {
630 			error = copyin(&info.volumes[i], &upath,
631 				       sizeof(char *));
632 			if (error == 0)
633 				error = copyinstr(upath, path,
634 						  MAXPATHLEN, NULL);
635 		}
636 		if (error == 0)
637 			error = hammer_install_volume(hmp, path, devvp);
638 		if (error)
639 			break;
640 	}
641 	objcache_put(namei_oc, path);
642 
643 	/*
644 	 * Make sure we found a root volume
645 	 */
646 	if (error == 0 && hmp->rootvol == NULL) {
647 		kprintf("hammer_mount: No root volume found!\n");
648 		error = EINVAL;
649 	}
650 
651 	/*
652 	 * Check that all required volumes are available
653 	 */
654 	if (error == 0 && hammer_mountcheck_volumes(hmp)) {
655 		kprintf("hammer_mount: Missing volumes, cannot mount!\n");
656 		error = EINVAL;
657 	}
658 
659 	if (error) {
660 		/* called with fs_token held */
661 		hammer_free_hmp(mp);
662 		return (error);
663 	}
664 
665 	/*
666 	 * No errors, setup enough of the mount point so we can lookup the
667 	 * root vnode.
668 	 */
669 	mp->mnt_iosize_max = MAXPHYS;
670 	mp->mnt_kern_flag |= MNTK_FSMID;
671 	mp->mnt_kern_flag |= MNTK_THR_SYNC;	/* new vsyncscan semantics */
672 
673 	/*
674 	 * MPSAFE code.  Note that VOPs and VFSops which are not MPSAFE
675 	 * will acquire a per-mount token prior to entry and release it
676 	 * on return.
677 	 */
678 	mp->mnt_kern_flag |= MNTK_ALL_MPSAFE;
679 	/*MNTK_RD_MPSAFE | MNTK_GA_MPSAFE | MNTK_IN_MPSAFE;*/
680 
681 	/*
682 	 * note: f_iosize is used by vnode_pager_haspage() when constructing
683 	 * its VOP_BMAP call.
684 	 */
685 	mp->mnt_stat.f_iosize = HAMMER_BUFSIZE;
686 	mp->mnt_stat.f_bsize = HAMMER_BUFSIZE;
687 
688 	mp->mnt_vstat.f_frsize = HAMMER_BUFSIZE;
689 	mp->mnt_vstat.f_bsize = HAMMER_BUFSIZE;
690 
691 	mp->mnt_maxsymlinklen = 255;
692 	mp->mnt_flag |= MNT_LOCAL;
693 
694 	vfs_add_vnodeops(mp, &hammer_vnode_vops, &mp->mnt_vn_norm_ops);
695 	vfs_add_vnodeops(mp, &hammer_spec_vops, &mp->mnt_vn_spec_ops);
696 	vfs_add_vnodeops(mp, &hammer_fifo_vops, &mp->mnt_vn_fifo_ops);
697 
698 	/*
699 	 * The root volume's ondisk pointer is only valid if we hold a
700 	 * reference to it.
701 	 */
702 	rootvol = hammer_get_root_volume(hmp, &error);
703 	if (error)
704 		goto failed;
705 
706 	/*
707 	 * Perform any necessary UNDO operations.  The recovery code does
708 	 * call hammer_undo_lookup() so we have to pre-cache the blockmap,
709 	 * and then re-copy it again after recovery is complete.
710 	 *
711 	 * If this is a read-only mount the UNDO information is retained
712 	 * in memory in the form of dirty buffer cache buffers, and not
713 	 * written back to the media.
714 	 */
715 	bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
716 	      sizeof(hmp->blockmap));
717 
718 	/*
719 	 * Check filesystem version
720 	 */
721 	hmp->version = rootvol->ondisk->vol_version;
722 	if (hmp->version < HAMMER_VOL_VERSION_MIN ||
723 	    hmp->version > HAMMER_VOL_VERSION_MAX) {
724 		kprintf("HAMMER: mount unsupported fs version %d\n",
725 			hmp->version);
726 		error = ERANGE;
727 		goto done;
728 	}
729 
730 	/*
731 	 * The undo_rec_limit limits the size of flush groups to avoid
732 	 * blowing out the UNDO FIFO.  This calculation is typically in
733 	 * the tens of thousands and is designed primarily when small
734 	 * HAMMER filesystems are created.
735 	 */
736 	hmp->undo_rec_limit = hammer_undo_max(hmp) / 8192 + 100;
737 	if (hammer_debug_general & 0x0001)
738 		kprintf("HAMMER: undo_rec_limit %d\n", hmp->undo_rec_limit);
739 
740 	/*
741 	 * NOTE: Recover stage1 not only handles meta-data recovery, it
742 	 * 	 also sets hmp->undo_seqno for HAMMER VERSION 4+ filesystems.
743 	 */
744 	error = hammer_recover_stage1(hmp, rootvol);
745 	if (error) {
746 		kprintf("Failed to recover HAMMER filesystem on mount\n");
747 		goto done;
748 	}
749 
750 	/*
751 	 * Finish setup now that we have a good root volume.
752 	 *
753 	 * The top 16 bits of fsid.val[1] is a pfs id.
754 	 */
755 	ksnprintf(mp->mnt_stat.f_mntfromname,
756 		  sizeof(mp->mnt_stat.f_mntfromname), "%s",
757 		  rootvol->ondisk->vol_name);
758 	mp->mnt_stat.f_fsid.val[0] =
759 		crc32((char *)&rootvol->ondisk->vol_fsid + 0, 8);
760 	mp->mnt_stat.f_fsid.val[1] =
761 		crc32((char *)&rootvol->ondisk->vol_fsid + 8, 8);
762 	mp->mnt_stat.f_fsid.val[1] &= 0x0000FFFF;
763 
764 	mp->mnt_vstat.f_fsid_uuid = rootvol->ondisk->vol_fsid;
765 	mp->mnt_vstat.f_fsid = crc32(&mp->mnt_vstat.f_fsid_uuid,
766 				     sizeof(mp->mnt_vstat.f_fsid_uuid));
767 
768 	/*
769 	 * Certain often-modified fields in the root volume are cached in
770 	 * the hammer_mount structure so we do not have to generate lots
771 	 * of little UNDO structures for them.
772 	 *
773 	 * Recopy after recovery.  This also has the side effect of
774 	 * setting our cached undo FIFO's first_offset, which serves to
775 	 * placemark the FIFO start for the NEXT flush cycle while the
776 	 * on-disk first_offset represents the LAST flush cycle.
777 	 */
778 	hmp->next_tid = rootvol->ondisk->vol0_next_tid;
779 	hmp->flush_tid1 = hmp->next_tid;
780 	hmp->flush_tid2 = hmp->next_tid;
781 	bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
782 	      sizeof(hmp->blockmap));
783 	hmp->copy_stat_freebigblocks = rootvol->ondisk->vol0_stat_freebigblocks;
784 
785 	hammer_flusher_create(hmp);
786 
787 	/*
788 	 * Locate the root directory using the root cluster's B-Tree as a
789 	 * starting point.  The root directory uses an obj_id of 1.
790 	 *
791 	 * FUTURE: Leave the root directory cached referenced but unlocked
792 	 * in hmp->rootvp (need to flush it on unmount).
793 	 */
794 	error = hammer_vfs_vget(mp, NULL, 1, &rootvp);
795 	if (error)
796 		goto done;
797 	vput(rootvp);
798 	/*vn_unlock(hmp->rootvp);*/
799 	if (hmp->ronly == 0)
800 		error = hammer_recover_stage2(hmp, rootvol);
801 
802 	/*
803 	 * If the stage2 recovery fails be sure to clean out all cached
804 	 * vnodes before throwing away the mount structure or bad things
805 	 * will happen.
806 	 */
807 	if (error)
808 		vflush(mp, 0, 0);
809 
810 done:
811 	if ((mp->mnt_flag & MNT_UPDATE) == 0) {
812 		/* New mount */
813 
814 		/* Populate info for mount point (NULL pad)*/
815 		bzero(mp->mnt_stat.f_mntonname, MNAMELEN);
816 		size_t size;
817 		if (mntpt) {
818 			copyinstr(mntpt, mp->mnt_stat.f_mntonname,
819 							MNAMELEN -1, &size);
820 		} else { /* Root mount */
821 			mp->mnt_stat.f_mntonname[0] = '/';
822 		}
823 	}
824 	(void)VFS_STATFS(mp, &mp->mnt_stat, cred);
825 	hammer_rel_volume(rootvol, 0);
826 failed:
827 	/*
828 	 * Cleanup and return.
829 	 */
830 	if (error) {
831 		/* called with fs_token held */
832 		hammer_free_hmp(mp);
833 	} else {
834 		lwkt_reltoken(&hmp->fs_token);
835 	}
836 	return (error);
837 }
838 
839 static int
840 hammer_vfs_unmount(struct mount *mp, int mntflags)
841 {
842 	hammer_mount_t hmp = (void *)mp->mnt_data;
843 	int flags;
844 	int error;
845 
846 	/*
847 	 * Clean out the vnodes
848 	 */
849 	lwkt_gettoken(&hmp->fs_token);
850 	flags = 0;
851 	if (mntflags & MNT_FORCE)
852 		flags |= FORCECLOSE;
853 	error = vflush(mp, 0, flags);
854 
855 	/*
856 	 * Clean up the internal mount structure and related entities.  This
857 	 * may issue I/O.
858 	 */
859 	if (error == 0) {
860 		/* called with fs_token held */
861 		hammer_free_hmp(mp);
862 	} else {
863 		lwkt_reltoken(&hmp->fs_token);
864 	}
865 	return(error);
866 }
867 
868 /*
869  * Clean up the internal mount structure and disassociate it from the mount.
870  * This may issue I/O.
871  *
872  * Called with fs_token held.
873  */
874 static void
875 hammer_free_hmp(struct mount *mp)
876 {
877 	hammer_mount_t hmp = (void *)mp->mnt_data;
878 	hammer_flush_group_t flg;
879 	int count;
880 	int dummy;
881 
882 	/*
883 	 * Flush anything dirty.  This won't even run if the
884 	 * filesystem errored-out.
885 	 */
886 	count = 0;
887 	while (hammer_flusher_haswork(hmp)) {
888 		hammer_flusher_sync(hmp);
889 		++count;
890 		if (count >= 5) {
891 			if (count == 5)
892 				kprintf("HAMMER: umount flushing.");
893 			else
894 				kprintf(".");
895 			tsleep(&dummy, 0, "hmrufl", hz);
896 		}
897 		if (count == 30) {
898 			kprintf("giving up\n");
899 			break;
900 		}
901 	}
902 	if (count >= 5 && count < 30)
903 		kprintf("\n");
904 
905 	/*
906 	 * If the mount had a critical error we have to destroy any
907 	 * remaining inodes before we can finish cleaning up the flusher.
908 	 */
909 	if (hmp->flags & HAMMER_MOUNT_CRITICAL_ERROR) {
910 		RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
911 			hammer_destroy_inode_callback, NULL);
912 	}
913 
914 	/*
915 	 * There shouldn't be any inodes left now and any left over
916 	 * flush groups should now be empty.
917 	 */
918 	KKASSERT(RB_EMPTY(&hmp->rb_inos_root));
919 	while ((flg = TAILQ_FIRST(&hmp->flush_group_list)) != NULL) {
920 		TAILQ_REMOVE(&hmp->flush_group_list, flg, flush_entry);
921 		KKASSERT(RB_EMPTY(&flg->flush_tree));
922 		if (flg->refs) {
923 			kprintf("HAMMER: Warning, flush_group %p was "
924 				"not empty on umount!\n", flg);
925 		}
926 		kfree(flg, hmp->m_misc);
927 	}
928 
929 	/*
930 	 * We can finally destroy the flusher
931 	 */
932 	hammer_flusher_destroy(hmp);
933 
934 	/*
935 	 * We may have held recovered buffers due to a read-only mount.
936 	 * These must be discarded.
937 	 */
938 	if (hmp->ronly)
939 		hammer_recover_flush_buffers(hmp, NULL, -1);
940 
941 	/*
942 	 * Unload buffers and then volumes
943 	 */
944         RB_SCAN(hammer_buf_rb_tree, &hmp->rb_bufs_root, NULL,
945 		hammer_unload_buffer, NULL);
946 	RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
947 		hammer_unload_volume, NULL);
948 
949 	mp->mnt_data = NULL;
950 	mp->mnt_flag &= ~MNT_LOCAL;
951 	hmp->mp = NULL;
952 	hammer_destroy_objid_cache(hmp);
953 	hammer_destroy_dedup_cache(hmp);
954 	if (hmp->dedup_free_cache != NULL) {
955 		kfree(hmp->dedup_free_cache, hmp->m_misc);
956 		hmp->dedup_free_cache = NULL;
957 	}
958 	kmalloc_destroy(&hmp->m_misc);
959 	kmalloc_destroy(&hmp->m_inodes);
960 	lwkt_reltoken(&hmp->fs_token);
961 	kfree(hmp, M_HAMMER);
962 }
963 
964 /*
965  * Report critical errors.  ip may be NULL.
966  */
967 void
968 hammer_critical_error(hammer_mount_t hmp, hammer_inode_t ip,
969 		      int error, const char *msg)
970 {
971 	hmp->flags |= HAMMER_MOUNT_CRITICAL_ERROR;
972 
973 	krateprintf(&hmp->krate,
974 		    "HAMMER(%s): Critical error inode=%jd error=%d %s\n",
975 		    hmp->mp->mnt_stat.f_mntfromname,
976 		    (intmax_t)(ip ? ip->obj_id : -1),
977 		    error, msg);
978 
979 	if (hmp->ronly == 0) {
980 		hmp->ronly = 2;		/* special errored read-only mode */
981 		hmp->mp->mnt_flag |= MNT_RDONLY;
982 		RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
983 			hammer_adjust_volume_mode, NULL);
984 		kprintf("HAMMER(%s): Forcing read-only mode\n",
985 			hmp->mp->mnt_stat.f_mntfromname);
986 	}
987 	hmp->error = error;
988 	if (hammer_debug_critical)
989 		Debugger("Entering debugger");
990 }
991 
992 
993 /*
994  * Obtain a vnode for the specified inode number.  An exclusively locked
995  * vnode is returned.
996  */
997 int
998 hammer_vfs_vget(struct mount *mp, struct vnode *dvp,
999 		ino_t ino, struct vnode **vpp)
1000 {
1001 	struct hammer_transaction trans;
1002 	struct hammer_mount *hmp = (void *)mp->mnt_data;
1003 	struct hammer_inode *ip;
1004 	int error;
1005 	u_int32_t localization;
1006 
1007 	lwkt_gettoken(&hmp->fs_token);
1008 	hammer_simple_transaction(&trans, hmp);
1009 
1010 	/*
1011 	 * If a directory vnode is supplied (mainly NFS) then we can acquire
1012 	 * the PFS domain from it.  Otherwise we would only be able to vget
1013 	 * inodes in the root PFS.
1014 	 */
1015 	if (dvp) {
1016 		localization = HAMMER_DEF_LOCALIZATION +
1017 				VTOI(dvp)->obj_localization;
1018 	} else {
1019 		localization = HAMMER_DEF_LOCALIZATION;
1020 	}
1021 
1022 	/*
1023 	 * Lookup the requested HAMMER inode.  The structure must be
1024 	 * left unlocked while we manipulate the related vnode to avoid
1025 	 * a deadlock.
1026 	 */
1027 	ip = hammer_get_inode(&trans, NULL, ino,
1028 			      hmp->asof, localization,
1029 			      0, &error);
1030 	if (ip == NULL) {
1031 		*vpp = NULL;
1032 	} else {
1033 		error = hammer_get_vnode(ip, vpp);
1034 		hammer_rel_inode(ip, 0);
1035 	}
1036 	hammer_done_transaction(&trans);
1037 	lwkt_reltoken(&hmp->fs_token);
1038 	return (error);
1039 }
1040 
1041 /*
1042  * Return the root vnode for the filesystem.
1043  *
1044  * HAMMER stores the root vnode in the hammer_mount structure so
1045  * getting it is easy.
1046  */
1047 static int
1048 hammer_vfs_root(struct mount *mp, struct vnode **vpp)
1049 {
1050 	int error;
1051 
1052 	error = hammer_vfs_vget(mp, NULL, 1, vpp);
1053 	return (error);
1054 }
1055 
1056 static int
1057 hammer_vfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
1058 {
1059 	struct hammer_mount *hmp = (void *)mp->mnt_data;
1060 	hammer_volume_t volume;
1061 	hammer_volume_ondisk_t ondisk;
1062 	int error;
1063 	int64_t bfree;
1064 	int64_t breserved;
1065 
1066 	lwkt_gettoken(&hmp->fs_token);
1067 	volume = hammer_get_root_volume(hmp, &error);
1068 	if (error) {
1069 		lwkt_reltoken(&hmp->fs_token);
1070 		return(error);
1071 	}
1072 	ondisk = volume->ondisk;
1073 
1074 	/*
1075 	 * Basic stats
1076 	 */
1077 	_hammer_checkspace(hmp, HAMMER_CHKSPC_WRITE, &breserved);
1078 	mp->mnt_stat.f_files = ondisk->vol0_stat_inodes;
1079 	bfree = ondisk->vol0_stat_freebigblocks * HAMMER_BIGBLOCK_SIZE;
1080 	hammer_rel_volume(volume, 0);
1081 
1082 	mp->mnt_stat.f_bfree = (bfree - breserved) / HAMMER_BUFSIZE;
1083 	mp->mnt_stat.f_bavail = mp->mnt_stat.f_bfree;
1084 	if (mp->mnt_stat.f_files < 0)
1085 		mp->mnt_stat.f_files = 0;
1086 
1087 	*sbp = mp->mnt_stat;
1088 	lwkt_reltoken(&hmp->fs_token);
1089 	return(0);
1090 }
1091 
1092 static int
1093 hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred)
1094 {
1095 	struct hammer_mount *hmp = (void *)mp->mnt_data;
1096 	hammer_volume_t volume;
1097 	hammer_volume_ondisk_t ondisk;
1098 	int error;
1099 	int64_t bfree;
1100 	int64_t breserved;
1101 
1102 	lwkt_gettoken(&hmp->fs_token);
1103 	volume = hammer_get_root_volume(hmp, &error);
1104 	if (error) {
1105 		lwkt_reltoken(&hmp->fs_token);
1106 		return(error);
1107 	}
1108 	ondisk = volume->ondisk;
1109 
1110 	/*
1111 	 * Basic stats
1112 	 */
1113 	_hammer_checkspace(hmp, HAMMER_CHKSPC_WRITE, &breserved);
1114 	mp->mnt_vstat.f_files = ondisk->vol0_stat_inodes;
1115 	bfree = ondisk->vol0_stat_freebigblocks * HAMMER_BIGBLOCK_SIZE;
1116 	hammer_rel_volume(volume, 0);
1117 
1118 	mp->mnt_vstat.f_bfree = (bfree - breserved) / HAMMER_BUFSIZE;
1119 	mp->mnt_vstat.f_bavail = mp->mnt_vstat.f_bfree;
1120 	if (mp->mnt_vstat.f_files < 0)
1121 		mp->mnt_vstat.f_files = 0;
1122 	*sbp = mp->mnt_vstat;
1123 	lwkt_reltoken(&hmp->fs_token);
1124 	return(0);
1125 }
1126 
1127 /*
1128  * Sync the filesystem.  Currently we have to run it twice, the second
1129  * one will advance the undo start index to the end index, so if a crash
1130  * occurs no undos will be run on mount.
1131  *
1132  * We do not sync the filesystem if we are called from a panic.  If we did
1133  * we might end up blowing up a sync that was already in progress.
1134  */
1135 static int
1136 hammer_vfs_sync(struct mount *mp, int waitfor)
1137 {
1138 	struct hammer_mount *hmp = (void *)mp->mnt_data;
1139 	int error;
1140 
1141 	lwkt_gettoken(&hmp->fs_token);
1142 	if (panicstr == NULL) {
1143 		error = hammer_sync_hmp(hmp, waitfor);
1144 	} else {
1145 		error = EIO;
1146 	}
1147 	lwkt_reltoken(&hmp->fs_token);
1148 	return (error);
1149 }
1150 
1151 /*
1152  * Convert a vnode to a file handle.
1153  *
1154  * Accesses read-only fields on already-referenced structures so
1155  * no token is needed.
1156  */
1157 static int
1158 hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp)
1159 {
1160 	hammer_inode_t ip;
1161 
1162 	KKASSERT(MAXFIDSZ >= 16);
1163 	ip = VTOI(vp);
1164 	fhp->fid_len = offsetof(struct fid, fid_data[16]);
1165 	fhp->fid_ext = ip->obj_localization >> 16;
1166 	bcopy(&ip->obj_id, fhp->fid_data + 0, sizeof(ip->obj_id));
1167 	bcopy(&ip->obj_asof, fhp->fid_data + 8, sizeof(ip->obj_asof));
1168 	return(0);
1169 }
1170 
1171 
1172 /*
1173  * Convert a file handle back to a vnode.
1174  *
1175  * Use rootvp to enforce PFS isolation when a PFS is exported via a
1176  * null mount.
1177  */
1178 static int
1179 hammer_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
1180 		  struct fid *fhp, struct vnode **vpp)
1181 {
1182 	hammer_mount_t hmp = (void *)mp->mnt_data;
1183 	struct hammer_transaction trans;
1184 	struct hammer_inode *ip;
1185 	struct hammer_inode_info info;
1186 	int error;
1187 	u_int32_t localization;
1188 
1189 	bcopy(fhp->fid_data + 0, &info.obj_id, sizeof(info.obj_id));
1190 	bcopy(fhp->fid_data + 8, &info.obj_asof, sizeof(info.obj_asof));
1191 	if (rootvp)
1192 		localization = VTOI(rootvp)->obj_localization;
1193 	else
1194 		localization = (u_int32_t)fhp->fid_ext << 16;
1195 
1196 	lwkt_gettoken(&hmp->fs_token);
1197 	hammer_simple_transaction(&trans, hmp);
1198 
1199 	/*
1200 	 * Get/allocate the hammer_inode structure.  The structure must be
1201 	 * unlocked while we manipulate the related vnode to avoid a
1202 	 * deadlock.
1203 	 */
1204 	ip = hammer_get_inode(&trans, NULL, info.obj_id,
1205 			      info.obj_asof, localization, 0, &error);
1206 	if (ip) {
1207 		error = hammer_get_vnode(ip, vpp);
1208 		hammer_rel_inode(ip, 0);
1209 	} else {
1210 		*vpp = NULL;
1211 	}
1212 	hammer_done_transaction(&trans);
1213 	lwkt_reltoken(&hmp->fs_token);
1214 	return (error);
1215 }
1216 
1217 static int
1218 hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
1219 		    int *exflagsp, struct ucred **credanonp)
1220 {
1221 	hammer_mount_t hmp = (void *)mp->mnt_data;
1222 	struct netcred *np;
1223 	int error;
1224 
1225 	lwkt_gettoken(&hmp->fs_token);
1226 	np = vfs_export_lookup(mp, &hmp->export, nam);
1227 	if (np) {
1228 		*exflagsp = np->netc_exflags;
1229 		*credanonp = &np->netc_anon;
1230 		error = 0;
1231 	} else {
1232 		error = EACCES;
1233 	}
1234 	lwkt_reltoken(&hmp->fs_token);
1235 	return (error);
1236 
1237 }
1238 
1239 int
1240 hammer_vfs_export(struct mount *mp, int op, const struct export_args *export)
1241 {
1242 	hammer_mount_t hmp = (void *)mp->mnt_data;
1243 	int error;
1244 
1245 	lwkt_gettoken(&hmp->fs_token);
1246 
1247 	switch(op) {
1248 	case MOUNTCTL_SET_EXPORT:
1249 		error = vfs_export(mp, &hmp->export, export);
1250 		break;
1251 	default:
1252 		error = EOPNOTSUPP;
1253 		break;
1254 	}
1255 	lwkt_reltoken(&hmp->fs_token);
1256 
1257 	return(error);
1258 }
1259 
1260