xref: /dflybsd-src/sys/vfs/hammer/hammer_vfsops.c (revision 8d378610e3b5687c707bc8aad4e11a3a96bea2fc)
1 /*
2  * Copyright (c) 2007-2008 The DragonFly Project.  All rights reserved.
3  *
4  * This code is derived from software contributed to The DragonFly Project
5  * by Matthew Dillon <dillon@backplane.com>
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  *
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in
15  *    the documentation and/or other materials provided with the
16  *    distribution.
17  * 3. Neither the name of The DragonFly Project nor the names of its
18  *    contributors may be used to endorse or promote products derived
19  *    from this software without specific, prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
25  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
27  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
29  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  */
34 
35 #include <sys/mountctl.h>
36 
37 #include "hammer.h"
38 
39 /*
40  * NOTE!  Global statistics may not be MPSAFE so HAMMER never uses them
41  *	  in conditionals.
42  */
43 int hammer_supported_version = HAMMER_VOL_VERSION_DEFAULT;
44 int hammer_debug_io;
45 int hammer_debug_general;
46 int hammer_debug_inode;
47 int hammer_debug_locks;
48 int hammer_debug_btree;
49 int hammer_debug_tid;
50 int hammer_debug_recover;		/* -1 will disable, +1 will force */
51 int hammer_debug_critical;		/* non-zero enter debugger on error */
52 int hammer_cluster_enable = 2;		/* ena cluster_read, scale x 2 */
53 int hammer_live_dedup = 0;
54 int hammer_tdmux_ticks;
55 int hammer_count_fsyncs;
56 int hammer_count_inodes;
57 int hammer_count_iqueued;
58 int hammer_count_reclaims;
59 int hammer_count_records;
60 int hammer_count_record_datas;
61 int hammer_count_volumes;
62 int hammer_count_buffers;
63 int hammer_count_nodes;
64 int64_t hammer_stats_btree_lookups;
65 int64_t hammer_stats_btree_searches;
66 int64_t hammer_stats_btree_inserts;
67 int64_t hammer_stats_btree_deletes;
68 int64_t hammer_stats_btree_elements;
69 int64_t hammer_stats_btree_splits;
70 int64_t hammer_stats_btree_iterations;
71 int64_t hammer_stats_btree_root_iterations;
72 int64_t hammer_stats_record_iterations;
73 
74 int64_t hammer_stats_file_read;
75 int64_t hammer_stats_file_write;
76 int64_t hammer_stats_file_iopsr;
77 int64_t hammer_stats_file_iopsw;
78 int64_t hammer_stats_disk_read;
79 int64_t hammer_stats_disk_write;
80 int64_t hammer_stats_inode_flushes;
81 int64_t hammer_stats_commits;
82 int64_t hammer_stats_undo;
83 int64_t hammer_stats_redo;
84 
85 long hammer_count_dirtybufspace;	/* global */
86 int hammer_count_refedbufs;		/* global */
87 int hammer_count_reservations;
88 long hammer_count_io_running_read;
89 long hammer_count_io_running_write;
90 int hammer_count_io_locked;
91 long hammer_limit_dirtybufspace;	/* per-mount */
92 int hammer_limit_recs;			/* as a whole XXX */
93 int hammer_limit_inode_recs = 2048;	/* per inode */
94 int hammer_limit_reclaims;
95 int hammer_live_dedup_cache_size = 4096;
96 int hammer_limit_redo = 4096 * 1024;	/* per inode */
97 int hammer_autoflush = 500;		/* auto flush (typ on reclaim) */
98 int hammer_verify_zone;
99 int hammer_verify_data = 1;
100 int hammer_double_buffer;
101 int hammer_btree_full_undo = 1;
102 int hammer_yield_check = 16;
103 int hammer_fsync_mode = 3;
104 int64_t hammer_contention_count;
105 
106 int hammer_noatime = 1;
107 TUNABLE_INT("vfs.hammer.noatime", &hammer_noatime);
108 
109 /*
110  * Live dedup debug counters (sysctls are writable so that counters
111  * can be reset from userspace).
112  */
113 int64_t hammer_live_dedup_vnode_bcmps = 0;
114 int64_t hammer_live_dedup_device_bcmps = 0;
115 int64_t hammer_live_dedup_findblk_failures = 0;
116 int64_t hammer_live_dedup_bmap_saves = 0;
117 
118 
119 SYSCTL_NODE(_vfs, OID_AUTO, hammer, CTLFLAG_RW, 0, "HAMMER filesystem");
120 
121 SYSCTL_INT(_vfs_hammer, OID_AUTO, supported_version, CTLFLAG_RD,
122 	   &hammer_supported_version, 0, "");
123 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_general, CTLFLAG_RW,
124 	   &hammer_debug_general, 0, "");
125 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_io, CTLFLAG_RW,
126 	   &hammer_debug_io, 0, "");
127 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_inode, CTLFLAG_RW,
128 	   &hammer_debug_inode, 0, "");
129 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_locks, CTLFLAG_RW,
130 	   &hammer_debug_locks, 0, "");
131 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_btree, CTLFLAG_RW,
132 	   &hammer_debug_btree, 0, "");
133 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_tid, CTLFLAG_RW,
134 	   &hammer_debug_tid, 0, "");
135 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover, CTLFLAG_RW,
136 	   &hammer_debug_recover, 0, "");
137 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_critical, CTLFLAG_RW,
138 	   &hammer_debug_critical, 0, "");
139 SYSCTL_INT(_vfs_hammer, OID_AUTO, cluster_enable, CTLFLAG_RW,
140 	   &hammer_cluster_enable, 0, "");
141 /*
142  * 0 - live dedup is disabled
143  * 1 - dedup cache is populated on reads only
144  * 2 - dedup cache is populated on both reads and writes
145  *
146  * LIVE_DEDUP IS DISABLED PERMANENTLY!  This feature appears to cause
147  * blockmap corruption over time so we've turned it off permanently.
148  */
149 SYSCTL_INT(_vfs_hammer, OID_AUTO, live_dedup, CTLFLAG_RD,
150 	   &hammer_live_dedup, 0, "Enable live dedup (experimental)");
151 SYSCTL_INT(_vfs_hammer, OID_AUTO, tdmux_ticks, CTLFLAG_RW,
152 	   &hammer_tdmux_ticks, 0, "Hammer tdmux ticks");
153 
154 SYSCTL_LONG(_vfs_hammer, OID_AUTO, limit_dirtybufspace, CTLFLAG_RW,
155 	   &hammer_limit_dirtybufspace, 0, "");
156 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_recs, CTLFLAG_RW,
157 	   &hammer_limit_recs, 0, "");
158 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_inode_recs, CTLFLAG_RW,
159 	   &hammer_limit_inode_recs, 0, "");
160 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_reclaims, CTLFLAG_RW,
161 	   &hammer_limit_reclaims, 0, "");
162 SYSCTL_INT(_vfs_hammer, OID_AUTO, live_dedup_cache_size, CTLFLAG_RW,
163 	   &hammer_live_dedup_cache_size, 0,
164 	   "Number of cache entries");
165 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_redo, CTLFLAG_RW,
166 	   &hammer_limit_redo, 0, "");
167 
168 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_fsyncs, CTLFLAG_RD,
169 	   &hammer_count_fsyncs, 0, "");
170 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_inodes, CTLFLAG_RD,
171 	   &hammer_count_inodes, 0, "");
172 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_iqueued, CTLFLAG_RD,
173 	   &hammer_count_iqueued, 0, "");
174 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reclaims, CTLFLAG_RD,
175 	   &hammer_count_reclaims, 0, "");
176 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_records, CTLFLAG_RD,
177 	   &hammer_count_records, 0, "");
178 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_record_datas, CTLFLAG_RD,
179 	   &hammer_count_record_datas, 0, "");
180 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_volumes, CTLFLAG_RD,
181 	   &hammer_count_volumes, 0, "");
182 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_buffers, CTLFLAG_RD,
183 	   &hammer_count_buffers, 0, "");
184 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_nodes, CTLFLAG_RD,
185 	   &hammer_count_nodes, 0, "");
186 
187 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_searches, CTLFLAG_RD,
188 	   &hammer_stats_btree_searches, 0, "");
189 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_lookups, CTLFLAG_RD,
190 	   &hammer_stats_btree_lookups, 0, "");
191 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_inserts, CTLFLAG_RD,
192 	   &hammer_stats_btree_inserts, 0, "");
193 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_deletes, CTLFLAG_RD,
194 	   &hammer_stats_btree_deletes, 0, "");
195 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_elements, CTLFLAG_RD,
196 	   &hammer_stats_btree_elements, 0, "");
197 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_splits, CTLFLAG_RD,
198 	   &hammer_stats_btree_splits, 0, "");
199 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_iterations, CTLFLAG_RD,
200 	   &hammer_stats_btree_iterations, 0, "");
201 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_root_iterations, CTLFLAG_RD,
202 	   &hammer_stats_btree_root_iterations, 0, "");
203 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_record_iterations, CTLFLAG_RD,
204 	   &hammer_stats_record_iterations, 0, "");
205 
206 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_read, CTLFLAG_RD,
207 	   &hammer_stats_file_read, 0, "");
208 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_write, CTLFLAG_RD,
209 	   &hammer_stats_file_write, 0, "");
210 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_iopsr, CTLFLAG_RD,
211 	   &hammer_stats_file_iopsr, 0, "");
212 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_iopsw, CTLFLAG_RD,
213 	   &hammer_stats_file_iopsw, 0, "");
214 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_disk_read, CTLFLAG_RD,
215 	   &hammer_stats_disk_read, 0, "");
216 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_disk_write, CTLFLAG_RD,
217 	   &hammer_stats_disk_write, 0, "");
218 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_inode_flushes, CTLFLAG_RD,
219 	   &hammer_stats_inode_flushes, 0, "");
220 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_commits, CTLFLAG_RD,
221 	   &hammer_stats_commits, 0, "");
222 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_undo, CTLFLAG_RD,
223 	   &hammer_stats_undo, 0, "");
224 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_redo, CTLFLAG_RD,
225 	   &hammer_stats_redo, 0, "");
226 
227 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, live_dedup_vnode_bcmps, CTLFLAG_RW,
228 	    &hammer_live_dedup_vnode_bcmps, 0,
229 	    "successful vnode buffer comparisons");
230 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, live_dedup_device_bcmps, CTLFLAG_RW,
231 	    &hammer_live_dedup_device_bcmps, 0,
232 	    "successful device buffer comparisons");
233 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, live_dedup_findblk_failures, CTLFLAG_RW,
234 	    &hammer_live_dedup_findblk_failures, 0,
235 	    "block lookup failures for comparison");
236 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, live_dedup_bmap_saves, CTLFLAG_RW,
237 	    &hammer_live_dedup_bmap_saves, 0,
238 	    "useful physical block lookups");
239 
240 SYSCTL_LONG(_vfs_hammer, OID_AUTO, count_dirtybufspace, CTLFLAG_RD,
241 	   &hammer_count_dirtybufspace, 0, "");
242 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_refedbufs, CTLFLAG_RD,
243 	   &hammer_count_refedbufs, 0, "");
244 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reservations, CTLFLAG_RD,
245 	   &hammer_count_reservations, 0, "");
246 SYSCTL_LONG(_vfs_hammer, OID_AUTO, count_io_running_read, CTLFLAG_RD,
247 	   &hammer_count_io_running_read, 0, "");
248 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_locked, CTLFLAG_RD,
249 	   &hammer_count_io_locked, 0, "");
250 SYSCTL_LONG(_vfs_hammer, OID_AUTO, count_io_running_write, CTLFLAG_RD,
251 	   &hammer_count_io_running_write, 0, "");
252 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, contention_count, CTLFLAG_RW,
253 	   &hammer_contention_count, 0, "");
254 SYSCTL_INT(_vfs_hammer, OID_AUTO, autoflush, CTLFLAG_RW,
255 	   &hammer_autoflush, 0, "");
256 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_zone, CTLFLAG_RW,
257 	   &hammer_verify_zone, 0, "");
258 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_data, CTLFLAG_RW,
259 	   &hammer_verify_data, 0, "");
260 SYSCTL_INT(_vfs_hammer, OID_AUTO, double_buffer, CTLFLAG_RW,
261 	   &hammer_double_buffer, 0, "");
262 SYSCTL_INT(_vfs_hammer, OID_AUTO, btree_full_undo, CTLFLAG_RW,
263 	   &hammer_btree_full_undo, 0, "");
264 SYSCTL_INT(_vfs_hammer, OID_AUTO, yield_check, CTLFLAG_RW,
265 	   &hammer_yield_check, 0, "");
266 SYSCTL_INT(_vfs_hammer, OID_AUTO, fsync_mode, CTLFLAG_RW,
267 	   &hammer_fsync_mode, 0, "");
268 
269 /* KTR_INFO_MASTER(hammer); */
270 
271 /*
272  * VFS ABI
273  */
274 static void	hammer_free_hmp(struct mount *mp);
275 
276 static int	hammer_vfs_mount(struct mount *mp, char *path, caddr_t data,
277 				struct ucred *cred);
278 static int	hammer_vfs_unmount(struct mount *mp, int mntflags);
279 static int	hammer_vfs_root(struct mount *mp, struct vnode **vpp);
280 static int	hammer_vfs_statfs(struct mount *mp, struct statfs *sbp,
281 				struct ucred *cred);
282 static int	hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp,
283 				struct ucred *cred);
284 static int	hammer_vfs_sync(struct mount *mp, int waitfor);
285 static int	hammer_vfs_vget(struct mount *mp, struct vnode *dvp,
286 				ino_t ino, struct vnode **vpp);
287 static int	hammer_vfs_init(struct vfsconf *conf);
288 static int	hammer_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
289 				struct fid *fhp, struct vnode **vpp);
290 static int	hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp);
291 static int	hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
292 				int *exflagsp, struct ucred **credanonp);
293 
294 
295 static struct vfsops hammer_vfsops = {
296 	.vfs_mount	= hammer_vfs_mount,
297 	.vfs_unmount	= hammer_vfs_unmount,
298 	.vfs_root	= hammer_vfs_root,
299 	.vfs_statfs	= hammer_vfs_statfs,
300 	.vfs_statvfs	= hammer_vfs_statvfs,
301 	.vfs_sync	= hammer_vfs_sync,
302 	.vfs_vget	= hammer_vfs_vget,
303 	.vfs_init	= hammer_vfs_init,
304 	.vfs_vptofh	= hammer_vfs_vptofh,
305 	.vfs_fhtovp	= hammer_vfs_fhtovp,
306 	.vfs_checkexp	= hammer_vfs_checkexp
307 };
308 
309 MALLOC_DEFINE(M_HAMMER, "HAMMER-mount", "");
310 
311 VFS_SET(hammer_vfsops, hammer, 0);
312 MODULE_VERSION(hammer, 1);
313 
314 static int
315 hammer_vfs_init(struct vfsconf *conf)
316 {
317 	long n;
318 
319 	/*
320 	 * Wait up to this long for an exclusive deadlock to clear
321 	 * before acquiring a new shared lock on the ip.  The deadlock
322 	 * may have occured on a b-tree node related to the ip.
323 	 */
324 	if (hammer_tdmux_ticks == 0)
325 		hammer_tdmux_ticks = hz / 5;
326 
327 	/*
328 	 * Autosize, but be careful because a hammer filesystem's
329 	 * reserve is partially calculated based on dirtybufspace,
330 	 * so we simply cannot allow it to get too large.
331 	 */
332 	if (hammer_limit_recs == 0) {
333 		n = nbuf * 25;
334 		if (n > kmalloc_limit(M_HAMMER) / 512)
335 			n = kmalloc_limit(M_HAMMER) / 512;
336 		if (n > 2 * 1024 * 1024)
337 			n = 2 * 1024 * 1024;
338 		hammer_limit_recs = (int)n;
339 	}
340 	if (hammer_limit_dirtybufspace == 0) {
341 		hammer_limit_dirtybufspace = hidirtybufspace / 2;
342 		if (hammer_limit_dirtybufspace < 1L * 1024 * 1024)
343 			hammer_limit_dirtybufspace = 1024L * 1024;
344 		if (hammer_limit_dirtybufspace > 1024L * 1024 * 1024)
345 			hammer_limit_dirtybufspace = 1024L * 1024 * 1024;
346 	}
347 
348 	/*
349 	 * The hammer_inode structure detaches from the vnode on reclaim.
350 	 * This limits the number of inodes in this state to prevent a
351 	 * memory pool blowout.
352 	 */
353 	if (hammer_limit_reclaims == 0)
354 		hammer_limit_reclaims = maxvnodes / 10;
355 
356 	return(0);
357 }
358 
359 static int
360 hammer_vfs_mount(struct mount *mp, char *mntpt, caddr_t data,
361 		 struct ucred *cred)
362 {
363 	struct hammer_mount_info info;
364 	hammer_mount_t hmp;
365 	hammer_volume_t rootvol;
366 	struct vnode *rootvp;
367 	struct vnode *devvp = NULL;
368 	const char *upath;	/* volume name in userspace */
369 	char *path;		/* volume name in system space */
370 	int error;
371 	int i;
372 	int master_id;
373 	int nvolumes;
374 	char *next_volume_ptr = NULL;
375 
376 	if (hammer_noatime) {
377 		/* Force noatime */
378 		mp->mnt_flag |= MNT_NOATIME;
379 	}
380 
381 	/*
382 	 * Accept hammer_mount_info.  mntpt is NULL for root mounts at boot.
383 	 */
384 	if (mntpt == NULL) {
385 		bzero(&info, sizeof(info));
386 		info.asof = 0;
387 		info.hflags = 0;
388 		info.nvolumes = 1;
389 
390 		next_volume_ptr = mp->mnt_stat.f_mntfromname;
391 
392 		/* Count number of volumes separated by ':' */
393 		for (char *p = next_volume_ptr; *p != '\0'; ++p) {
394 			if (*p == ':') {
395 				++info.nvolumes;
396 			}
397 		}
398 
399 		mp->mnt_flag &= ~MNT_RDONLY; /* mount R/W */
400 	} else {
401 		if ((error = copyin(data, &info, sizeof(info))) != 0)
402 			return (error);
403 	}
404 
405 	/*
406 	 * updating or new mount
407 	 */
408 	if (mp->mnt_flag & MNT_UPDATE) {
409 		hmp = (void *)mp->mnt_data;
410 		KKASSERT(hmp != NULL);
411 	} else {
412 		if (info.nvolumes <= 0 || info.nvolumes > HAMMER_MAX_VOLUMES)
413 			return (EINVAL);
414 		hmp = NULL;
415 	}
416 
417 	/*
418 	 * master-id validation.  The master id may not be changed by a
419 	 * mount update.
420 	 */
421 	if (info.hflags & HMNT_MASTERID || info.hflags & HMNT_NOMIRROR) {
422 		if (hmp && hmp->master_id != info.master_id) {
423 			hkprintf("cannot change master id with mount update\n");
424 			return(EINVAL);
425 		}
426 		master_id = info.master_id;
427 		if (master_id < -1 || master_id >= HAMMER_MAX_MASTERS)
428 			return (EINVAL);
429 	} else {
430 		if (hmp)
431 			master_id = hmp->master_id;
432 		else
433 			master_id = 0;
434 	}
435 
436 	/*
437 	 * Internal mount data structure
438 	 */
439 	if (hmp == NULL) {
440 		hmp = kmalloc(sizeof(*hmp), M_HAMMER, M_WAITOK | M_ZERO);
441 		mp->mnt_data = (qaddr_t)hmp;
442 		hmp->mp = mp;
443 
444 		/*
445 		 * Make sure kmalloc type limits are set appropriately.
446 		 *
447 		 * Our inode kmalloc group is sized based on maxvnodes
448 		 * (controlled by the system, not us).
449 		 */
450 		kmalloc_create(&hmp->m_misc, "HAMMER-others");
451 		kmalloc_create(&hmp->m_inodes, "HAMMER-inodes");
452 
453 		kmalloc_raise_limit(hmp->m_inodes, 0);	/* unlimited */
454 
455 		hmp->root_btree_beg.localization =
456 			HAMMER_MIN_ONDISK_LOCALIZATION;
457 		hmp->root_btree_beg.obj_id = HAMMER_MIN_OBJID;
458 		hmp->root_btree_beg.key = HAMMER_MIN_KEY;
459 		hmp->root_btree_beg.create_tid = 1;
460 		hmp->root_btree_beg.delete_tid = 1;
461 		hmp->root_btree_beg.rec_type = HAMMER_MIN_RECTYPE;
462 		hmp->root_btree_beg.obj_type = 0;
463 		hmp->root_btree_beg.btype = HAMMER_BTREE_TYPE_NONE;
464 
465 		hmp->root_btree_end.localization =
466 			HAMMER_MAX_ONDISK_LOCALIZATION;
467 		hmp->root_btree_end.obj_id = HAMMER_MAX_OBJID;
468 		hmp->root_btree_end.key = HAMMER_MAX_KEY;
469 		hmp->root_btree_end.create_tid = HAMMER_MAX_TID;
470 		hmp->root_btree_end.delete_tid = 0;   /* special case */
471 		hmp->root_btree_end.rec_type = HAMMER_MAX_RECTYPE;
472 		hmp->root_btree_end.obj_type = 0;
473 		hmp->root_btree_end.btype = HAMMER_BTREE_TYPE_NONE;
474 
475 		hmp->krate.freq = 1;	/* maximum reporting rate (hz) */
476 		hmp->krate.count = -16;	/* initial burst */
477 		hmp->kdiag.freq = 1;	/* maximum reporting rate (hz) */
478 		hmp->kdiag.count = -16;	/* initial burst */
479 
480 		hmp->sync_lock.refs = 1;
481 		hmp->undo_lock.refs = 1;
482 		hmp->blkmap_lock.refs = 1;
483 		hmp->snapshot_lock.refs = 1;
484 		hmp->volume_lock.refs = 1;
485 
486 		TAILQ_INIT(&hmp->delay_list);
487 		TAILQ_INIT(&hmp->flush_group_list);
488 		TAILQ_INIT(&hmp->objid_cache_list);
489 		TAILQ_INIT(&hmp->undo_lru_list);
490 		TAILQ_INIT(&hmp->reclaim_list);
491 
492 		RB_INIT(&hmp->rb_dedup_crc_root);
493 		RB_INIT(&hmp->rb_dedup_off_root);
494 		TAILQ_INIT(&hmp->dedup_lru_list);
495 	}
496 	hmp->hflags &= ~HMNT_USERFLAGS;
497 	hmp->hflags |= info.hflags & HMNT_USERFLAGS;
498 
499 	hmp->master_id = master_id;
500 
501 	if (info.asof) {
502 		mp->mnt_flag |= MNT_RDONLY;
503 		hmp->asof = info.asof;
504 	} else {
505 		hmp->asof = HAMMER_MAX_TID;
506 	}
507 
508 	hmp->volume_to_remove = -1;
509 
510 	/*
511 	 * Re-open read-write if originally read-only, or vise-versa.
512 	 *
513 	 * When going from read-only to read-write execute the stage2
514 	 * recovery if it has not already been run.
515 	 */
516 	if (mp->mnt_flag & MNT_UPDATE) {
517 		lwkt_gettoken(&hmp->fs_token);
518 		error = 0;
519 		if (hmp->ronly && (mp->mnt_kern_flag & MNTK_WANTRDWR)) {
520 			hkprintf("read-only -> read-write\n");
521 			hmp->ronly = 0;
522 			RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
523 				hammer_adjust_volume_mode, NULL);
524 			rootvol = hammer_get_root_volume(hmp, &error);
525 			if (rootvol) {
526 				hammer_recover_flush_buffers(hmp, rootvol, 1);
527 				error = hammer_recover_stage2(hmp, rootvol);
528 				bcopy(rootvol->ondisk->vol0_blockmap,
529 				      hmp->blockmap,
530 				      sizeof(hmp->blockmap));
531 				hammer_rel_volume(rootvol, 0);
532 			}
533 			RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
534 				hammer_reload_inode, NULL);
535 			/* kernel clears MNT_RDONLY */
536 		} else if (hmp->ronly == 0 && (mp->mnt_flag & MNT_RDONLY)) {
537 			hkprintf("read-write -> read-only\n");
538 			hmp->ronly = 1;	/* messy */
539 			RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
540 				hammer_reload_inode, NULL);
541 			hmp->ronly = 0;
542 			hammer_flusher_sync(hmp);
543 			hammer_flusher_sync(hmp);
544 			hammer_flusher_sync(hmp);
545 			hmp->ronly = 1;
546 			RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
547 				hammer_adjust_volume_mode, NULL);
548 		}
549 		lwkt_reltoken(&hmp->fs_token);
550 		return(error);
551 	}
552 
553 	RB_INIT(&hmp->rb_vols_root);
554 	RB_INIT(&hmp->rb_inos_root);
555 	RB_INIT(&hmp->rb_redo_root);
556 	RB_INIT(&hmp->rb_nods_root);
557 	RB_INIT(&hmp->rb_undo_root);
558 	RB_INIT(&hmp->rb_resv_root);
559 	RB_INIT(&hmp->rb_bufs_root);
560 	RB_INIT(&hmp->rb_pfsm_root);
561 
562 	hmp->ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
563 
564 	RB_INIT(&hmp->volu_root);
565 	RB_INIT(&hmp->undo_root);
566 	RB_INIT(&hmp->data_root);
567 	RB_INIT(&hmp->meta_root);
568 	RB_INIT(&hmp->lose_root);
569 	TAILQ_INIT(&hmp->iorun_list);
570 
571 	lwkt_token_init(&hmp->fs_token, "hammerfs");
572 	lwkt_token_init(&hmp->io_token, "hammerio");
573 
574 	lwkt_gettoken(&hmp->fs_token);
575 
576 	/*
577 	 * Load volumes
578 	 */
579 	path = objcache_get(namei_oc, M_WAITOK);
580 	hmp->nvolumes = -1;
581 	for (i = 0; i < info.nvolumes; ++i) {
582 		if (mntpt == NULL) {
583 			/*
584 			 * Root mount.
585 			 */
586 			KKASSERT(next_volume_ptr != NULL);
587 			strcpy(path, "");
588 			if (*next_volume_ptr != '/') {
589 				/* relative path */
590 				strcpy(path, "/dev/");
591 			}
592 			int k;
593 			for (k = strlen(path); k < MAXPATHLEN-1; ++k) {
594 				if (*next_volume_ptr == '\0') {
595 					break;
596 				} else if (*next_volume_ptr == ':') {
597 					++next_volume_ptr;
598 					break;
599 				} else {
600 					path[k] = *next_volume_ptr;
601 					++next_volume_ptr;
602 				}
603 			}
604 			path[k] = '\0';
605 
606 			error = 0;
607 			cdev_t dev = kgetdiskbyname(path);
608 			error = bdevvp(dev, &devvp);
609 			if (error) {
610 				hdkprintf("can't find devvp\n");
611 			}
612 		} else {
613 			error = copyin(&info.volumes[i], &upath,
614 				       sizeof(char *));
615 			if (error == 0)
616 				error = copyinstr(upath, path,
617 						  MAXPATHLEN, NULL);
618 		}
619 		if (error == 0)
620 			error = hammer_install_volume(hmp, path, devvp, NULL);
621 		if (error)
622 			break;
623 	}
624 	objcache_put(namei_oc, path);
625 
626 	/*
627 	 * Make sure we found a root volume
628 	 */
629 	if (hmp->rootvol == NULL) {
630 		if (error == EBUSY) {
631 			hdkprintf("The volumes are probably mounted\n");
632 		} else {
633 			hdkprintf("No root volume found!\n");
634 			error = EINVAL;
635 		}
636 		goto failed;
637 	}
638 
639 	/*
640 	 * Check that all required volumes are available
641 	 */
642 	if (error == 0 && hammer_mountcheck_volumes(hmp)) {
643 		hdkprintf("Missing volumes, cannot mount!\n");
644 		error = EINVAL;
645 		goto failed;
646 	}
647 
648 	/*
649 	 * Other errors
650 	 */
651 	if (error) {
652 		hdkprintf("Failed to load volumes!\n");
653 		goto failed;
654 	}
655 
656 	nvolumes = hammer_get_installed_volumes(hmp);
657 	if (hmp->nvolumes != nvolumes) {
658 		hdkprintf("volume header says %d volumes, but %d installed\n",
659 			hmp->nvolumes, nvolumes);
660 		error = EINVAL;
661 		goto failed;
662 	}
663 
664 	/*
665 	 * No errors, setup enough of the mount point so we can lookup the
666 	 * root vnode.
667 	 */
668 	mp->mnt_iosize_max = MAXPHYS;
669 	mp->mnt_kern_flag |= MNTK_FSMID;
670 	mp->mnt_kern_flag |= MNTK_THR_SYNC;	/* new vsyncscan semantics */
671 
672 	/*
673 	 * MPSAFE code.  Note that VOPs and VFSops which are not MPSAFE
674 	 * will acquire a per-mount token prior to entry and release it
675 	 * on return.
676 	 */
677 	mp->mnt_kern_flag |= MNTK_ALL_MPSAFE;
678 
679 	/*
680 	 * note: f_iosize is used by vnode_pager_haspage() when constructing
681 	 * its VOP_BMAP call.
682 	 */
683 	mp->mnt_stat.f_iosize = HAMMER_BUFSIZE;
684 	mp->mnt_stat.f_bsize = HAMMER_BUFSIZE;
685 
686 	mp->mnt_vstat.f_frsize = HAMMER_BUFSIZE;
687 	mp->mnt_vstat.f_bsize = HAMMER_BUFSIZE;
688 
689 	mp->mnt_maxsymlinklen = 255;
690 	mp->mnt_flag |= MNT_LOCAL;
691 
692 	vfs_add_vnodeops(mp, &hammer_vnode_vops, &mp->mnt_vn_norm_ops);
693 	vfs_add_vnodeops(mp, &hammer_spec_vops, &mp->mnt_vn_spec_ops);
694 	vfs_add_vnodeops(mp, &hammer_fifo_vops, &mp->mnt_vn_fifo_ops);
695 
696 	/*
697 	 * The root volume's ondisk pointer is only valid if we hold a
698 	 * reference to it.
699 	 */
700 	rootvol = hammer_get_root_volume(hmp, &error);
701 	if (error)
702 		goto failed;
703 
704 	/*
705 	 * Perform any necessary UNDO operations.  The recovery code does
706 	 * call hammer_undo_lookup() so we have to pre-cache the blockmap,
707 	 * and then re-copy it again after recovery is complete.
708 	 *
709 	 * If this is a read-only mount the UNDO information is retained
710 	 * in memory in the form of dirty buffer cache buffers, and not
711 	 * written back to the media.
712 	 */
713 	bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
714 	      sizeof(hmp->blockmap));
715 
716 	/*
717 	 * Check filesystem version
718 	 */
719 	hmp->version = rootvol->ondisk->vol_version;
720 	if (hmp->version < HAMMER_VOL_VERSION_MIN ||
721 	    hmp->version > HAMMER_VOL_VERSION_MAX) {
722 		hkprintf("mount unsupported fs version %d\n", hmp->version);
723 		error = ERANGE;
724 		goto done;
725 	}
726 
727 	/*
728 	 * The undo_rec_limit limits the size of flush groups to avoid
729 	 * blowing out the UNDO FIFO.  This calculation is typically in
730 	 * the tens of thousands and is designed primarily when small
731 	 * HAMMER filesystems are created.
732 	 */
733 	hmp->undo_rec_limit = hammer_undo_max(hmp) / 8192 + 100;
734 	if (hammer_debug_general & 0x0001)
735 		hkprintf("undo_rec_limit %d\n", hmp->undo_rec_limit);
736 
737 	/*
738 	 * NOTE: Recover stage1 not only handles meta-data recovery, it
739 	 * 	 also sets hmp->undo_seqno for HAMMER VERSION 4+ filesystems.
740 	 */
741 	error = hammer_recover_stage1(hmp, rootvol);
742 	if (error) {
743 		kprintf("Failed to recover HAMMER filesystem on mount\n");
744 		goto done;
745 	}
746 
747 	/*
748 	 * Finish setup now that we have a good root volume.
749 	 */
750 	ksnprintf(mp->mnt_stat.f_mntfromname,
751 		  sizeof(mp->mnt_stat.f_mntfromname), "%s",
752 		  rootvol->ondisk->vol_label);
753 	mp->mnt_stat.f_fsid.val[0] =
754 		crc32((char *)&rootvol->ondisk->vol_fsid + 0, 8);
755 	mp->mnt_stat.f_fsid.val[1] =
756 		crc32((char *)&rootvol->ondisk->vol_fsid + 8, 8);
757 	mp->mnt_stat.f_fsid.val[1] &= HAMMER_LOCALIZE_MASK;
758 
759 	mp->mnt_vstat.f_fsid_uuid = rootvol->ondisk->vol_fsid;
760 	mp->mnt_vstat.f_fsid = crc32(&mp->mnt_vstat.f_fsid_uuid,
761 				     sizeof(mp->mnt_vstat.f_fsid_uuid));
762 
763 	/*
764 	 * Certain often-modified fields in the root volume are cached in
765 	 * the hammer_mount structure so we do not have to generate lots
766 	 * of little UNDO structures for them.
767 	 *
768 	 * Recopy after recovery.  This also has the side effect of
769 	 * setting our cached undo FIFO's first_offset, which serves to
770 	 * placemark the FIFO start for the NEXT flush cycle while the
771 	 * on-disk first_offset represents the LAST flush cycle.
772 	 */
773 	hmp->next_tid = rootvol->ondisk->vol0_next_tid;
774 	hmp->flush_tid1 = hmp->next_tid;
775 	hmp->flush_tid2 = hmp->next_tid;
776 	bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
777 	      sizeof(hmp->blockmap));
778 	hmp->copy_stat_freebigblocks = rootvol->ondisk->vol0_stat_freebigblocks;
779 
780 	hammer_flusher_create(hmp);
781 
782 	/*
783 	 * Locate the root directory with an obj_id of 1.
784 	 */
785 	error = hammer_vfs_root(mp, &rootvp);
786 	if (error)
787 		goto done;
788 	vput(rootvp);
789 	if (hmp->ronly == 0)
790 		error = hammer_recover_stage2(hmp, rootvol);
791 
792 	/*
793 	 * If the stage2 recovery fails be sure to clean out all cached
794 	 * vnodes before throwing away the mount structure or bad things
795 	 * will happen.
796 	 */
797 	if (error)
798 		vflush(mp, 0, 0);
799 
800 done:
801 	if ((mp->mnt_flag & MNT_UPDATE) == 0) {
802 		/* New mount */
803 
804 		/* Populate info for mount point (NULL pad)*/
805 		bzero(mp->mnt_stat.f_mntonname, MNAMELEN);
806 		size_t size;
807 		if (mntpt) {
808 			copyinstr(mntpt, mp->mnt_stat.f_mntonname,
809 							MNAMELEN -1, &size);
810 		} else { /* Root mount */
811 			mp->mnt_stat.f_mntonname[0] = '/';
812 		}
813 	}
814 	(void)VFS_STATFS(mp, &mp->mnt_stat, cred);
815 	hammer_rel_volume(rootvol, 0);
816 failed:
817 	/*
818 	 * Cleanup and return.
819 	 */
820 	if (error) {
821 		/* called with fs_token held */
822 		hammer_free_hmp(mp);
823 	} else {
824 		lwkt_reltoken(&hmp->fs_token);
825 	}
826 	return (error);
827 }
828 
829 static int
830 hammer_vfs_unmount(struct mount *mp, int mntflags)
831 {
832 	hammer_mount_t hmp = (void *)mp->mnt_data;
833 	int flags;
834 	int error;
835 
836 	/*
837 	 * Clean out the vnodes
838 	 */
839 	lwkt_gettoken(&hmp->fs_token);
840 	flags = 0;
841 	if (mntflags & MNT_FORCE)
842 		flags |= FORCECLOSE;
843 	error = vflush(mp, 0, flags);
844 
845 	/*
846 	 * Clean up the internal mount structure and related entities.  This
847 	 * may issue I/O.
848 	 */
849 	if (error == 0) {
850 		/* called with fs_token held */
851 		hammer_free_hmp(mp);
852 	} else {
853 		lwkt_reltoken(&hmp->fs_token);
854 	}
855 	return(error);
856 }
857 
858 /*
859  * Clean up the internal mount structure and disassociate it from the mount.
860  * This may issue I/O.
861  *
862  * Called with fs_token held.
863  */
864 static void
865 hammer_free_hmp(struct mount *mp)
866 {
867 	hammer_mount_t hmp = (void *)mp->mnt_data;
868 	hammer_flush_group_t flg;
869 
870 	/*
871 	 * Flush anything dirty.  This won't even run if the
872 	 * filesystem errored-out.
873 	 */
874 	hammer_flush_dirty(hmp, 30);
875 
876 	/*
877 	 * If the mount had a critical error we have to destroy any
878 	 * remaining inodes before we can finish cleaning up the flusher.
879 	 */
880 	if (hmp->flags & HAMMER_MOUNT_CRITICAL_ERROR) {
881 		RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
882 			hammer_destroy_inode_callback, NULL);
883 	}
884 
885 	/*
886 	 * There shouldn't be any inodes left now and any left over
887 	 * flush groups should now be empty.
888 	 */
889 	KKASSERT(RB_EMPTY(&hmp->rb_inos_root));
890 	while ((flg = TAILQ_FIRST(&hmp->flush_group_list)) != NULL) {
891 		TAILQ_REMOVE(&hmp->flush_group_list, flg, flush_entry);
892 		KKASSERT(RB_EMPTY(&flg->flush_tree));
893 		if (flg->refs) {
894 			hkprintf("Warning, flush_group %p was "
895 				"not empty on umount!\n", flg);
896 		}
897 		kfree(flg, hmp->m_misc);
898 	}
899 
900 	/*
901 	 * We can finally destroy the flusher
902 	 */
903 	hammer_flusher_destroy(hmp);
904 
905 	/*
906 	 * We may have held recovered buffers due to a read-only mount.
907 	 * These must be discarded.
908 	 */
909 	if (hmp->ronly)
910 		hammer_recover_flush_buffers(hmp, NULL, -1);
911 
912 	/*
913 	 * Unload buffers and then volumes
914 	 */
915         RB_SCAN(hammer_buf_rb_tree, &hmp->rb_bufs_root, NULL,
916 		hammer_unload_buffer, NULL);
917 	RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
918 		hammer_unload_volume, NULL);
919 
920 	mp->mnt_data = NULL;
921 	mp->mnt_flag &= ~MNT_LOCAL;
922 	hmp->mp = NULL;
923 	hammer_destroy_objid_cache(hmp);
924 	hammer_destroy_dedup_cache(hmp);
925 	if (hmp->dedup_free_cache != NULL) {
926 		kfree(hmp->dedup_free_cache, hmp->m_misc);
927 		hmp->dedup_free_cache = NULL;
928 	}
929 	kmalloc_destroy(&hmp->m_misc);
930 	kmalloc_destroy(&hmp->m_inodes);
931 	lwkt_reltoken(&hmp->fs_token);
932 	kfree(hmp, M_HAMMER);
933 }
934 
935 /*
936  * Report critical errors.  ip may be NULL.
937  */
938 void
939 hammer_critical_error(hammer_mount_t hmp, hammer_inode_t ip,
940 		      int error, const char *msg)
941 {
942 	hmp->flags |= HAMMER_MOUNT_CRITICAL_ERROR;
943 
944 	hmkrateprintf(&hmp->krate, hmp,
945 		    "Critical error inode=%jd error=%d %s\n",
946 		    (intmax_t)(ip ? ip->obj_id : -1),
947 		    error, msg);
948 
949 	if (hmp->ronly == 0) {
950 		hmp->ronly = 2;		/* special errored read-only mode */
951 		hmp->mp->mnt_flag |= MNT_RDONLY;
952 		RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
953 			hammer_adjust_volume_mode, NULL);
954 		hmkprintf(hmp, "Forcing read-only mode\n");
955 	}
956 	hmp->error = error;
957 	if (hammer_debug_critical)
958 		Debugger("Entering debugger");
959 }
960 
961 
962 /*
963  * Obtain a vnode for the specified inode number.  An exclusively locked
964  * vnode is returned.
965  */
966 int
967 hammer_vfs_vget(struct mount *mp, struct vnode *dvp,
968 		ino_t ino, struct vnode **vpp)
969 {
970 	struct hammer_transaction trans;
971 	hammer_mount_t hmp = (void *)mp->mnt_data;
972 	hammer_inode_t ip;
973 	int error;
974 	uint32_t localization;
975 
976 	lwkt_gettoken(&hmp->fs_token);
977 	hammer_simple_transaction(&trans, hmp);
978 
979 	/*
980 	 * If a directory vnode is supplied (mainly NFS) then we can acquire
981 	 * the PFS domain from it.  Otherwise we would only be able to vget
982 	 * inodes in the root PFS.
983 	 */
984 	if (dvp) {
985 		localization = HAMMER_DEF_LOCALIZATION |
986 				VTOI(dvp)->obj_localization;
987 	} else {
988 		localization = HAMMER_DEF_LOCALIZATION;
989 	}
990 
991 	/*
992 	 * Lookup the requested HAMMER inode.  The structure must be
993 	 * left unlocked while we manipulate the related vnode to avoid
994 	 * a deadlock.
995 	 */
996 	ip = hammer_get_inode(&trans, NULL, ino,
997 			      hmp->asof, localization,
998 			      0, &error);
999 	if (ip == NULL) {
1000 		*vpp = NULL;
1001 	} else {
1002 		error = hammer_get_vnode(ip, vpp);
1003 		hammer_rel_inode(ip, 0);
1004 	}
1005 	hammer_done_transaction(&trans);
1006 	lwkt_reltoken(&hmp->fs_token);
1007 	return (error);
1008 }
1009 
1010 /*
1011  * Return the root vnode for the filesystem.
1012  *
1013  * HAMMER stores the root vnode in the hammer_mount structure so
1014  * getting it is easy.
1015  */
1016 static int
1017 hammer_vfs_root(struct mount *mp, struct vnode **vpp)
1018 {
1019 	int error;
1020 
1021 	error = hammer_vfs_vget(mp, NULL, HAMMER_OBJID_ROOT, vpp);
1022 	return (error);
1023 }
1024 
1025 static int
1026 hammer_vfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
1027 {
1028 	hammer_mount_t hmp = (void *)mp->mnt_data;
1029 	hammer_volume_t volume;
1030 	hammer_volume_ondisk_t ondisk;
1031 	int error;
1032 	int64_t bfree;
1033 	int64_t breserved;
1034 
1035 	lwkt_gettoken(&hmp->fs_token);
1036 	volume = hammer_get_root_volume(hmp, &error);
1037 	if (error) {
1038 		lwkt_reltoken(&hmp->fs_token);
1039 		return(error);
1040 	}
1041 	ondisk = volume->ondisk;
1042 
1043 	/*
1044 	 * Basic stats
1045 	 */
1046 	_hammer_checkspace(hmp, HAMMER_CHKSPC_WRITE, &breserved);
1047 	mp->mnt_stat.f_files = ondisk->vol0_stat_inodes;
1048 	bfree = ondisk->vol0_stat_freebigblocks * HAMMER_BIGBLOCK_SIZE;
1049 	hammer_rel_volume(volume, 0);
1050 
1051 	if (breserved > bfree)
1052 		breserved = bfree;
1053 	mp->mnt_stat.f_bfree = (bfree - breserved) / HAMMER_BUFSIZE;
1054 	mp->mnt_stat.f_bavail = mp->mnt_stat.f_bfree;
1055 	if (mp->mnt_stat.f_files < 0)
1056 		mp->mnt_stat.f_files = 0;
1057 
1058 	*sbp = mp->mnt_stat;
1059 	lwkt_reltoken(&hmp->fs_token);
1060 	return(0);
1061 }
1062 
1063 static int
1064 hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred)
1065 {
1066 	hammer_mount_t hmp = (void *)mp->mnt_data;
1067 	hammer_volume_t volume;
1068 	hammer_volume_ondisk_t ondisk;
1069 	int error;
1070 	int64_t bfree;
1071 	int64_t breserved;
1072 
1073 	lwkt_gettoken(&hmp->fs_token);
1074 	volume = hammer_get_root_volume(hmp, &error);
1075 	if (error) {
1076 		lwkt_reltoken(&hmp->fs_token);
1077 		return(error);
1078 	}
1079 	ondisk = volume->ondisk;
1080 
1081 	/*
1082 	 * Basic stats
1083 	 */
1084 	_hammer_checkspace(hmp, HAMMER_CHKSPC_WRITE, &breserved);
1085 	mp->mnt_vstat.f_files = ondisk->vol0_stat_inodes;
1086 	bfree = ondisk->vol0_stat_freebigblocks * HAMMER_BIGBLOCK_SIZE;
1087 	hammer_rel_volume(volume, 0);
1088 
1089 	if (breserved > bfree)
1090 		breserved = bfree;
1091 	mp->mnt_vstat.f_bfree = (bfree - breserved) / HAMMER_BUFSIZE;
1092 	mp->mnt_vstat.f_bavail = mp->mnt_vstat.f_bfree;
1093 	if (mp->mnt_vstat.f_files < 0)
1094 		mp->mnt_vstat.f_files = 0;
1095 	*sbp = mp->mnt_vstat;
1096 	lwkt_reltoken(&hmp->fs_token);
1097 	return(0);
1098 }
1099 
1100 /*
1101  * Sync the filesystem.  Currently we have to run it twice, the second
1102  * one will advance the undo start index to the end index, so if a crash
1103  * occurs no undos will be run on mount.
1104  *
1105  * We do not sync the filesystem if we are called from a panic.  If we did
1106  * we might end up blowing up a sync that was already in progress.
1107  */
1108 static int
1109 hammer_vfs_sync(struct mount *mp, int waitfor)
1110 {
1111 	hammer_mount_t hmp = (void *)mp->mnt_data;
1112 	int error;
1113 
1114 	lwkt_gettoken(&hmp->fs_token);
1115 	if (panicstr == NULL) {
1116 		error = hammer_sync_hmp(hmp, waitfor);
1117 	} else {
1118 		error = EIO;
1119 	}
1120 	lwkt_reltoken(&hmp->fs_token);
1121 	return (error);
1122 }
1123 
1124 /*
1125  * Convert a vnode to a file handle.
1126  *
1127  * Accesses read-only fields on already-referenced structures so
1128  * no token is needed.
1129  */
1130 static int
1131 hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp)
1132 {
1133 	hammer_inode_t ip;
1134 
1135 	KKASSERT(MAXFIDSZ >= 16);
1136 	ip = VTOI(vp);
1137 	fhp->fid_len = offsetof(struct fid, fid_data[16]);
1138 	fhp->fid_ext = lo_to_pfs(ip->obj_localization);
1139 	bcopy(&ip->obj_id, fhp->fid_data + 0, sizeof(ip->obj_id));
1140 	bcopy(&ip->obj_asof, fhp->fid_data + 8, sizeof(ip->obj_asof));
1141 	return(0);
1142 }
1143 
1144 
1145 /*
1146  * Convert a file handle back to a vnode.
1147  *
1148  * Use rootvp to enforce PFS isolation when a PFS is exported via a
1149  * null mount.
1150  */
1151 static int
1152 hammer_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
1153 		  struct fid *fhp, struct vnode **vpp)
1154 {
1155 	hammer_mount_t hmp = (void *)mp->mnt_data;
1156 	struct hammer_transaction trans;
1157 	hammer_inode_t ip;
1158 	struct hammer_inode_info info;
1159 	int error;
1160 	uint32_t localization;
1161 
1162 	bcopy(fhp->fid_data + 0, &info.obj_id, sizeof(info.obj_id));
1163 	bcopy(fhp->fid_data + 8, &info.obj_asof, sizeof(info.obj_asof));
1164 	if (rootvp)
1165 		localization = VTOI(rootvp)->obj_localization;
1166 	else
1167 		localization = pfs_to_lo(fhp->fid_ext);
1168 
1169 	lwkt_gettoken(&hmp->fs_token);
1170 	hammer_simple_transaction(&trans, hmp);
1171 
1172 	/*
1173 	 * Get/allocate the hammer_inode structure.  The structure must be
1174 	 * unlocked while we manipulate the related vnode to avoid a
1175 	 * deadlock.
1176 	 */
1177 	ip = hammer_get_inode(&trans, NULL, info.obj_id,
1178 			      info.obj_asof, localization, 0, &error);
1179 	if (ip) {
1180 		error = hammer_get_vnode(ip, vpp);
1181 		hammer_rel_inode(ip, 0);
1182 	} else {
1183 		*vpp = NULL;
1184 	}
1185 	hammer_done_transaction(&trans);
1186 	lwkt_reltoken(&hmp->fs_token);
1187 	return (error);
1188 }
1189 
1190 static int
1191 hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
1192 		    int *exflagsp, struct ucred **credanonp)
1193 {
1194 	hammer_mount_t hmp = (void *)mp->mnt_data;
1195 	struct netcred *np;
1196 	int error;
1197 
1198 	lwkt_gettoken(&hmp->fs_token);
1199 	np = vfs_export_lookup(mp, &hmp->export, nam);
1200 	if (np) {
1201 		*exflagsp = np->netc_exflags;
1202 		*credanonp = &np->netc_anon;
1203 		error = 0;
1204 	} else {
1205 		error = EACCES;
1206 	}
1207 	lwkt_reltoken(&hmp->fs_token);
1208 	return (error);
1209 
1210 }
1211 
1212 int
1213 hammer_vfs_export(struct mount *mp, int op, const struct export_args *export)
1214 {
1215 	hammer_mount_t hmp = (void *)mp->mnt_data;
1216 	int error;
1217 
1218 	lwkt_gettoken(&hmp->fs_token);
1219 
1220 	switch(op) {
1221 	case MOUNTCTL_SET_EXPORT:
1222 		error = vfs_export(mp, &hmp->export, export);
1223 		break;
1224 	default:
1225 		error = EOPNOTSUPP;
1226 		break;
1227 	}
1228 	lwkt_reltoken(&hmp->fs_token);
1229 
1230 	return(error);
1231 }
1232 
1233