xref: /dflybsd-src/sys/vfs/hammer/hammer_vfsops.c (revision a7e9bef1725d53a752fb4cd48c0106f88045346c)
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 1.59 2008/07/03 04:24:51 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 int hammer_debug_io;
51 int hammer_debug_general;
52 int hammer_debug_debug;
53 int hammer_debug_inode;
54 int hammer_debug_locks;
55 int hammer_debug_btree;
56 int hammer_debug_tid;
57 int hammer_debug_recover;		/* -1 will disable, +1 will force */
58 int hammer_debug_recover_faults;
59 int hammer_debug_cluster_enable = 1;	/* enable read clustering by default */
60 int hammer_count_inodes;
61 int hammer_count_iqueued;
62 int hammer_count_reclaiming;
63 int hammer_count_records;
64 int hammer_count_record_datas;
65 int hammer_count_volumes;
66 int hammer_count_buffers;
67 int hammer_count_nodes;
68 int64_t hammer_count_extra_space_used;
69 int64_t hammer_stats_btree_lookups;
70 int64_t hammer_stats_btree_searches;
71 int64_t hammer_stats_btree_inserts;
72 int64_t hammer_stats_btree_deletes;
73 int64_t hammer_stats_btree_elements;
74 int64_t hammer_stats_btree_splits;
75 int64_t hammer_stats_btree_iterations;
76 int64_t hammer_stats_record_iterations;
77 int hammer_count_dirtybufspace;		/* global */
78 int hammer_count_refedbufs;		/* global */
79 int hammer_count_reservations;
80 int hammer_count_io_running_read;
81 int hammer_count_io_running_write;
82 int hammer_count_io_locked;
83 int hammer_limit_dirtybufspace;		/* per-mount */
84 int hammer_limit_recs;			/* as a whole XXX */
85 int hammer_limit_iqueued;		/* per-mount */
86 int hammer_bio_count;
87 int hammer_verify_zone;
88 int hammer_write_mode;
89 int64_t hammer_contention_count;
90 int64_t hammer_zone_limit;
91 
92 SYSCTL_NODE(_vfs, OID_AUTO, hammer, CTLFLAG_RW, 0, "HAMMER filesystem");
93 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_general, CTLFLAG_RW,
94 	   &hammer_debug_general, 0, "");
95 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_io, CTLFLAG_RW,
96 	   &hammer_debug_io, 0, "");
97 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_debug, CTLFLAG_RW,
98 	   &hammer_debug_debug, 0, "");
99 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_inode, CTLFLAG_RW,
100 	   &hammer_debug_inode, 0, "");
101 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_locks, CTLFLAG_RW,
102 	   &hammer_debug_locks, 0, "");
103 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_btree, CTLFLAG_RW,
104 	   &hammer_debug_btree, 0, "");
105 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_tid, CTLFLAG_RW,
106 	   &hammer_debug_tid, 0, "");
107 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover, CTLFLAG_RW,
108 	   &hammer_debug_recover, 0, "");
109 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover_faults, CTLFLAG_RW,
110 	   &hammer_debug_recover_faults, 0, "");
111 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_cluster_enable, CTLFLAG_RW,
112 	   &hammer_debug_cluster_enable, 0, "");
113 
114 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_dirtybufspace, CTLFLAG_RW,
115 	   &hammer_limit_dirtybufspace, 0, "");
116 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_recs, CTLFLAG_RW,
117 	   &hammer_limit_recs, 0, "");
118 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_iqueued, CTLFLAG_RW,
119 	   &hammer_limit_iqueued, 0, "");
120 
121 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_inodes, CTLFLAG_RD,
122 	   &hammer_count_inodes, 0, "");
123 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_iqueued, CTLFLAG_RD,
124 	   &hammer_count_iqueued, 0, "");
125 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reclaiming, CTLFLAG_RD,
126 	   &hammer_count_reclaiming, 0, "");
127 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_records, CTLFLAG_RD,
128 	   &hammer_count_records, 0, "");
129 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_record_datas, CTLFLAG_RD,
130 	   &hammer_count_record_datas, 0, "");
131 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_volumes, CTLFLAG_RD,
132 	   &hammer_count_volumes, 0, "");
133 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_buffers, CTLFLAG_RD,
134 	   &hammer_count_buffers, 0, "");
135 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_nodes, CTLFLAG_RD,
136 	   &hammer_count_nodes, 0, "");
137 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, count_extra_space_used, CTLFLAG_RD,
138 	   &hammer_count_extra_space_used, 0, "");
139 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_searches, CTLFLAG_RD,
140 	   &hammer_stats_btree_searches, 0, "");
141 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_lookups, CTLFLAG_RD,
142 	   &hammer_stats_btree_lookups, 0, "");
143 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_inserts, CTLFLAG_RD,
144 	   &hammer_stats_btree_inserts, 0, "");
145 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_deletes, CTLFLAG_RD,
146 	   &hammer_stats_btree_deletes, 0, "");
147 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_elements, CTLFLAG_RD,
148 	   &hammer_stats_btree_elements, 0, "");
149 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_splits, CTLFLAG_RD,
150 	   &hammer_stats_btree_splits, 0, "");
151 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_iterations, CTLFLAG_RD,
152 	   &hammer_stats_btree_iterations, 0, "");
153 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_record_iterations, CTLFLAG_RD,
154 	   &hammer_stats_record_iterations, 0, "");
155 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_dirtybufspace, CTLFLAG_RD,
156 	   &hammer_count_dirtybufspace, 0, "");
157 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_refedbufs, CTLFLAG_RD,
158 	   &hammer_count_refedbufs, 0, "");
159 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reservations, CTLFLAG_RD,
160 	   &hammer_count_reservations, 0, "");
161 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_running_read, CTLFLAG_RD,
162 	   &hammer_count_io_running_read, 0, "");
163 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_locked, CTLFLAG_RD,
164 	   &hammer_count_io_locked, 0, "");
165 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_running_write, CTLFLAG_RD,
166 	   &hammer_count_io_running_write, 0, "");
167 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, zone_limit, CTLFLAG_RW,
168 	   &hammer_zone_limit, 0, "");
169 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, contention_count, CTLFLAG_RW,
170 	   &hammer_contention_count, 0, "");
171 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_zone, CTLFLAG_RW,
172 	   &hammer_verify_zone, 0, "");
173 SYSCTL_INT(_vfs_hammer, OID_AUTO, write_mode, CTLFLAG_RW,
174 	   &hammer_write_mode, 0, "");
175 
176 KTR_INFO_MASTER(hammer);
177 
178 /*
179  * VFS ABI
180  */
181 static void	hammer_free_hmp(struct mount *mp);
182 
183 static int	hammer_vfs_mount(struct mount *mp, char *path, caddr_t data,
184 				struct ucred *cred);
185 static int	hammer_vfs_unmount(struct mount *mp, int mntflags);
186 static int	hammer_vfs_root(struct mount *mp, struct vnode **vpp);
187 static int	hammer_vfs_statfs(struct mount *mp, struct statfs *sbp,
188 				struct ucred *cred);
189 static int	hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp,
190 				struct ucred *cred);
191 static int	hammer_vfs_sync(struct mount *mp, int waitfor);
192 static int	hammer_vfs_vget(struct mount *mp, ino_t ino,
193 				struct vnode **vpp);
194 static int	hammer_vfs_init(struct vfsconf *conf);
195 static int	hammer_vfs_fhtovp(struct mount *mp, struct fid *fhp,
196 				struct vnode **vpp);
197 static int	hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp);
198 static int	hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
199 				int *exflagsp, struct ucred **credanonp);
200 
201 
202 static struct vfsops hammer_vfsops = {
203 	.vfs_mount	= hammer_vfs_mount,
204 	.vfs_unmount	= hammer_vfs_unmount,
205 	.vfs_root 	= hammer_vfs_root,
206 	.vfs_statfs	= hammer_vfs_statfs,
207 	.vfs_statvfs	= hammer_vfs_statvfs,
208 	.vfs_sync	= hammer_vfs_sync,
209 	.vfs_vget	= hammer_vfs_vget,
210 	.vfs_init	= hammer_vfs_init,
211 	.vfs_vptofh	= hammer_vfs_vptofh,
212 	.vfs_fhtovp	= hammer_vfs_fhtovp,
213 	.vfs_checkexp	= hammer_vfs_checkexp
214 };
215 
216 MALLOC_DEFINE(M_HAMMER, "hammer-mount", "hammer mount");
217 
218 VFS_SET(hammer_vfsops, hammer, 0);
219 MODULE_VERSION(hammer, 1);
220 
221 static int
222 hammer_vfs_init(struct vfsconf *conf)
223 {
224 	int n;
225 
226 	if (hammer_limit_recs == 0) {
227 		hammer_limit_recs = nbuf * 25;
228 		n = kmalloc_limit(M_HAMMER) / 512;
229 		if (hammer_limit_recs > n)
230 			hammer_limit_recs = n;
231 	}
232 	if (hammer_limit_dirtybufspace == 0) {
233 		hammer_limit_dirtybufspace = hidirtybufspace / 2;
234 		if (hammer_limit_dirtybufspace < 100)
235 			hammer_limit_dirtybufspace = 100;
236 	}
237 	if (hammer_limit_iqueued == 0)
238 		hammer_limit_iqueued = desiredvnodes / 5;
239 	return(0);
240 }
241 
242 static int
243 hammer_vfs_mount(struct mount *mp, char *mntpt, caddr_t data,
244 		 struct ucred *cred)
245 {
246 	struct hammer_mount_info info;
247 	hammer_mount_t hmp;
248 	hammer_volume_t rootvol;
249 	struct vnode *rootvp;
250 	const char *upath;	/* volume name in userspace */
251 	char *path;		/* volume name in system space */
252 	int error;
253 	int i;
254 
255 	if ((error = copyin(data, &info, sizeof(info))) != 0)
256 		return (error);
257 	if ((mp->mnt_flag & MNT_UPDATE) == 0) {
258 		if (info.nvolumes <= 0 || info.nvolumes >= 32768)
259 			return (EINVAL);
260 	}
261 	if ((info.hflags & HMNT_MASTERID) &&
262 	    (info.masterid < -1 || info.masterid >= HAMMER_MAX_MASTERS)) {
263 			return (EINVAL);
264 	}
265 
266 	/*
267 	 * Interal mount data structure
268 	 */
269 	if (mp->mnt_flag & MNT_UPDATE) {
270 		hmp = (void *)mp->mnt_data;
271 		KKASSERT(hmp != NULL);
272 	} else {
273 		hmp = kmalloc(sizeof(*hmp), M_HAMMER, M_WAITOK | M_ZERO);
274 		mp->mnt_data = (qaddr_t)hmp;
275 		hmp->mp = mp;
276 		hmp->namekey_iterator = mycpu->gd_time_seconds;
277 		/*TAILQ_INIT(&hmp->recycle_list);*/
278 
279 		hmp->root_btree_beg.localization = 0x00000000U;
280 		hmp->root_btree_beg.obj_id = -0x8000000000000000LL;
281 		hmp->root_btree_beg.key = -0x8000000000000000LL;
282 		hmp->root_btree_beg.create_tid = 1;
283 		hmp->root_btree_beg.delete_tid = 1;
284 		hmp->root_btree_beg.rec_type = 0;
285 		hmp->root_btree_beg.obj_type = 0;
286 
287 		hmp->root_btree_end.localization = 0xFFFFFFFFU;
288 		hmp->root_btree_end.obj_id = 0x7FFFFFFFFFFFFFFFLL;
289 		hmp->root_btree_end.key = 0x7FFFFFFFFFFFFFFFLL;
290 		hmp->root_btree_end.create_tid = 0xFFFFFFFFFFFFFFFFULL;
291 		hmp->root_btree_end.delete_tid = 0;   /* special case */
292 		hmp->root_btree_end.rec_type = 0xFFFFU;
293 		hmp->root_btree_end.obj_type = 0;
294 
295 		hmp->sync_lock.refs = 1;
296 		hmp->free_lock.refs = 1;
297 		hmp->undo_lock.refs = 1;
298 		hmp->blkmap_lock.refs = 1;
299 
300 		TAILQ_INIT(&hmp->flush_list);
301 		TAILQ_INIT(&hmp->delay_list);
302 		TAILQ_INIT(&hmp->objid_cache_list);
303 		TAILQ_INIT(&hmp->undo_lru_list);
304 		TAILQ_INIT(&hmp->reclaim_list);
305 	}
306 	hmp->hflags &= ~HMNT_USERFLAGS;
307 	hmp->hflags |= info.hflags & HMNT_USERFLAGS;
308 	if (info.hflags & HMNT_MASTERID)
309 		hmp->masterid = info.masterid;
310 	else
311 		hmp->masterid = -1;
312 	if (info.asof) {
313 		kprintf("ASOF\n");
314 		mp->mnt_flag |= MNT_RDONLY;
315 		hmp->asof = info.asof;
316 	} else {
317 		hmp->asof = HAMMER_MAX_TID;
318 	}
319 
320 	/*
321 	 * Re-open read-write if originally read-only, or vise-versa.
322 	 */
323 	if (mp->mnt_flag & MNT_UPDATE) {
324 		error = 0;
325 		if (hmp->ronly && (mp->mnt_kern_flag & MNTK_WANTRDWR)) {
326 			kprintf("HAMMER read-only -> read-write\n");
327 			hmp->ronly = 0;
328 			RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
329 				hammer_adjust_volume_mode, NULL);
330 			rootvol = hammer_get_root_volume(hmp, &error);
331 			if (rootvol) {
332 				hammer_recover_flush_buffers(hmp, rootvol, 1);
333 				bcopy(rootvol->ondisk->vol0_blockmap,
334 				      hmp->blockmap,
335 				      sizeof(hmp->blockmap));
336 				hammer_rel_volume(rootvol, 0);
337 			}
338 			RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
339 				hammer_reload_inode, NULL);
340 			/* kernel clears MNT_RDONLY */
341 		} else if (hmp->ronly == 0 && (mp->mnt_flag & MNT_RDONLY)) {
342 			kprintf("HAMMER read-write -> read-only\n");
343 			hmp->ronly = 1;	/* messy */
344 			RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
345 				hammer_reload_inode, NULL);
346 			hmp->ronly = 0;
347 			hammer_flusher_sync(hmp);
348 			hammer_flusher_sync(hmp);
349 			hammer_flusher_sync(hmp);
350 			hmp->ronly = 1;
351 			RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
352 				hammer_adjust_volume_mode, NULL);
353 		}
354 		return(error);
355 	}
356 
357 	RB_INIT(&hmp->rb_vols_root);
358 	RB_INIT(&hmp->rb_inos_root);
359 	RB_INIT(&hmp->rb_nods_root);
360 	RB_INIT(&hmp->rb_undo_root);
361 	RB_INIT(&hmp->rb_resv_root);
362 	RB_INIT(&hmp->rb_bufs_root);
363 	RB_INIT(&hmp->rb_pfsm_root);
364 
365 	hmp->ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
366 
367 	TAILQ_INIT(&hmp->volu_list);
368 	TAILQ_INIT(&hmp->undo_list);
369 	TAILQ_INIT(&hmp->data_list);
370 	TAILQ_INIT(&hmp->meta_list);
371 	TAILQ_INIT(&hmp->lose_list);
372 
373 	/*
374 	 * Load volumes
375 	 */
376 	path = objcache_get(namei_oc, M_WAITOK);
377 	hmp->nvolumes = info.nvolumes;
378 	for (i = 0; i < info.nvolumes; ++i) {
379 		error = copyin(&info.volumes[i], &upath, sizeof(char *));
380 		if (error == 0)
381 			error = copyinstr(upath, path, MAXPATHLEN, NULL);
382 		if (error == 0)
383 			error = hammer_install_volume(hmp, path);
384 		if (error)
385 			break;
386 	}
387 	objcache_put(namei_oc, path);
388 
389 	/*
390 	 * Make sure we found a root volume
391 	 */
392 	if (error == 0 && hmp->rootvol == NULL) {
393 		kprintf("hammer_mount: No root volume found!\n");
394 		error = EINVAL;
395 	}
396 	if (error) {
397 		hammer_free_hmp(mp);
398 		return (error);
399 	}
400 
401 	/*
402 	 * No errors, setup enough of the mount point so we can lookup the
403 	 * root vnode.
404 	 */
405 	mp->mnt_iosize_max = MAXPHYS;
406 	mp->mnt_kern_flag |= MNTK_FSMID;
407 
408 	/*
409 	 * note: f_iosize is used by vnode_pager_haspage() when constructing
410 	 * its VOP_BMAP call.
411 	 */
412 	mp->mnt_stat.f_iosize = HAMMER_BUFSIZE;
413 	mp->mnt_stat.f_bsize = HAMMER_BUFSIZE;
414 
415 	mp->mnt_vstat.f_frsize = HAMMER_BUFSIZE;
416 	mp->mnt_vstat.f_bsize = HAMMER_BUFSIZE;
417 
418 	mp->mnt_maxsymlinklen = 255;
419 	mp->mnt_flag |= MNT_LOCAL;
420 
421 	vfs_add_vnodeops(mp, &hammer_vnode_vops, &mp->mnt_vn_norm_ops);
422 	vfs_add_vnodeops(mp, &hammer_spec_vops, &mp->mnt_vn_spec_ops);
423 	vfs_add_vnodeops(mp, &hammer_fifo_vops, &mp->mnt_vn_fifo_ops);
424 
425 	/*
426 	 * The root volume's ondisk pointer is only valid if we hold a
427 	 * reference to it.
428 	 */
429 	rootvol = hammer_get_root_volume(hmp, &error);
430 	if (error)
431 		goto failed;
432 
433 	/*
434 	 * Perform any necessary UNDO operations.  The recovery code does
435 	 * call hammer_undo_lookup() so we have to pre-cache the blockmap,
436 	 * and then re-copy it again after recovery is complete.
437 	 *
438 	 * If this is a read-only mount the UNDO information is retained
439 	 * in memory in the form of dirty buffer cache buffers, and not
440 	 * written back to the media.
441 	 */
442 	bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
443 	      sizeof(hmp->blockmap));
444 
445 	error = hammer_recover(hmp, rootvol);
446 	if (error) {
447 		kprintf("Failed to recover HAMMER filesystem on mount\n");
448 		goto done;
449 	}
450 
451 	/*
452 	 * Finish setup now that we have a good root volume
453 	 */
454 	ksnprintf(mp->mnt_stat.f_mntfromname,
455 		  sizeof(mp->mnt_stat.f_mntfromname), "%s",
456 		  rootvol->ondisk->vol_name);
457 	mp->mnt_stat.f_fsid.val[0] =
458 		crc32((char *)&rootvol->ondisk->vol_fsid + 0, 8);
459 	mp->mnt_stat.f_fsid.val[1] =
460 		crc32((char *)&rootvol->ondisk->vol_fsid + 8, 8);
461 
462 	mp->mnt_vstat.f_fsid_uuid = rootvol->ondisk->vol_fsid;
463 	mp->mnt_vstat.f_fsid = crc32(&mp->mnt_vstat.f_fsid_uuid,
464 				     sizeof(mp->mnt_vstat.f_fsid_uuid));
465 
466 	/*
467 	 * Certain often-modified fields in the root volume are cached in
468 	 * the hammer_mount structure so we do not have to generate lots
469 	 * of little UNDO structures for them.
470 	 *
471 	 * Recopy after recovery.  This also has the side effect of
472 	 * setting our cached undo FIFO's first_offset, which serves to
473 	 * placemark the FIFO start for the NEXT flush cycle while the
474 	 * on-disk first_offset represents the LAST flush cycle.
475 	 */
476 	hmp->next_tid = rootvol->ondisk->vol0_next_tid;
477 	bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
478 	      sizeof(hmp->blockmap));
479 	hmp->copy_stat_freebigblocks = rootvol->ondisk->vol0_stat_freebigblocks;
480 
481 	hammer_flusher_create(hmp);
482 
483 	/*
484 	 * Locate the root directory using the root cluster's B-Tree as a
485 	 * starting point.  The root directory uses an obj_id of 1.
486 	 *
487 	 * FUTURE: Leave the root directory cached referenced but unlocked
488 	 * in hmp->rootvp (need to flush it on unmount).
489 	 */
490 	error = hammer_vfs_vget(mp, 1, &rootvp);
491 	if (error)
492 		goto done;
493 	vput(rootvp);
494 	/*vn_unlock(hmp->rootvp);*/
495 
496 done:
497 	hammer_rel_volume(rootvol, 0);
498 failed:
499 	/*
500 	 * Cleanup and return.
501 	 */
502 	if (error)
503 		hammer_free_hmp(mp);
504 	return (error);
505 }
506 
507 static int
508 hammer_vfs_unmount(struct mount *mp, int mntflags)
509 {
510 #if 0
511 	struct hammer_mount *hmp = (void *)mp->mnt_data;
512 #endif
513 	int flags;
514 	int error;
515 
516 	/*
517 	 * Clean out the vnodes
518 	 */
519 	flags = 0;
520 	if (mntflags & MNT_FORCE)
521 		flags |= FORCECLOSE;
522 	if ((error = vflush(mp, 0, flags)) != 0)
523 		return (error);
524 
525 	/*
526 	 * Clean up the internal mount structure and related entities.  This
527 	 * may issue I/O.
528 	 */
529 	hammer_free_hmp(mp);
530 	return(0);
531 }
532 
533 /*
534  * Clean up the internal mount structure and disassociate it from the mount.
535  * This may issue I/O.
536  */
537 static void
538 hammer_free_hmp(struct mount *mp)
539 {
540 	struct hammer_mount *hmp = (void *)mp->mnt_data;
541 
542 #if 0
543 	/*
544 	 * Clean up the root vnode
545 	 */
546 	if (hmp->rootvp) {
547 		vrele(hmp->rootvp);
548 		hmp->rootvp = NULL;
549 	}
550 #endif
551 	hammer_flusher_sync(hmp);
552 	hammer_flusher_sync(hmp);
553 	hammer_flusher_destroy(hmp);
554 
555 	KKASSERT(RB_EMPTY(&hmp->rb_inos_root));
556 
557 #if 0
558 	/*
559 	 * Unload & flush inodes
560 	 *
561 	 * XXX illegal to call this from here, it can only be done from
562 	 * the flusher.
563 	 */
564 	RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
565 		hammer_unload_inode, (void *)MNT_WAIT);
566 
567 	/*
568 	 * Unload & flush volumes
569 	 */
570 #endif
571 	/*
572 	 * Unload buffers and then volumes
573 	 */
574         RB_SCAN(hammer_buf_rb_tree, &hmp->rb_bufs_root, NULL,
575 		hammer_unload_buffer, NULL);
576 	RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
577 		hammer_unload_volume, NULL);
578 
579 	mp->mnt_data = NULL;
580 	mp->mnt_flag &= ~MNT_LOCAL;
581 	hmp->mp = NULL;
582 	hammer_destroy_objid_cache(hmp);
583 	kfree(hmp, M_HAMMER);
584 }
585 
586 /*
587  * Obtain a vnode for the specified inode number.  An exclusively locked
588  * vnode is returned.
589  */
590 int
591 hammer_vfs_vget(struct mount *mp, ino_t ino, struct vnode **vpp)
592 {
593 	struct hammer_transaction trans;
594 	struct hammer_mount *hmp = (void *)mp->mnt_data;
595 	struct hammer_inode *ip;
596 	int error;
597 
598 	hammer_simple_transaction(&trans, hmp);
599 
600 	/*
601 	 * Lookup the requested HAMMER inode.  The structure must be
602 	 * left unlocked while we manipulate the related vnode to avoid
603 	 * a deadlock.
604 	 */
605 	ip = hammer_get_inode(&trans, NULL, ino,
606 			      hmp->asof, HAMMER_DEF_LOCALIZATION,
607 			      0, &error);
608 	if (ip == NULL) {
609 		*vpp = NULL;
610 		return(error);
611 	}
612 	error = hammer_get_vnode(ip, vpp);
613 	hammer_rel_inode(ip, 0);
614 	hammer_done_transaction(&trans);
615 	return (error);
616 }
617 
618 /*
619  * Return the root vnode for the filesystem.
620  *
621  * HAMMER stores the root vnode in the hammer_mount structure so
622  * getting it is easy.
623  */
624 static int
625 hammer_vfs_root(struct mount *mp, struct vnode **vpp)
626 {
627 #if 0
628 	struct hammer_mount *hmp = (void *)mp->mnt_data;
629 #endif
630 	int error;
631 
632 	error = hammer_vfs_vget(mp, 1, vpp);
633 	return (error);
634 }
635 
636 static int
637 hammer_vfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
638 {
639 	struct hammer_mount *hmp = (void *)mp->mnt_data;
640 	hammer_volume_t volume;
641 	hammer_volume_ondisk_t ondisk;
642 	int error;
643 	int64_t bfree;
644 
645 	volume = hammer_get_root_volume(hmp, &error);
646 	if (error)
647 		return(error);
648 	ondisk = volume->ondisk;
649 
650 	/*
651 	 * Basic stats
652 	 */
653 	mp->mnt_stat.f_files = ondisk->vol0_stat_inodes;
654 	bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE;
655 	hammer_rel_volume(volume, 0);
656 
657 	mp->mnt_stat.f_bfree = bfree / HAMMER_BUFSIZE;
658 	mp->mnt_stat.f_bavail = mp->mnt_stat.f_bfree;
659 	if (mp->mnt_stat.f_files < 0)
660 		mp->mnt_stat.f_files = 0;
661 
662 	*sbp = mp->mnt_stat;
663 	return(0);
664 }
665 
666 static int
667 hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred)
668 {
669 	struct hammer_mount *hmp = (void *)mp->mnt_data;
670 	hammer_volume_t volume;
671 	hammer_volume_ondisk_t ondisk;
672 	int error;
673 	int64_t bfree;
674 
675 	volume = hammer_get_root_volume(hmp, &error);
676 	if (error)
677 		return(error);
678 	ondisk = volume->ondisk;
679 
680 	/*
681 	 * Basic stats
682 	 */
683 	mp->mnt_vstat.f_files = ondisk->vol0_stat_inodes;
684 	bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE;
685 	hammer_rel_volume(volume, 0);
686 
687 	mp->mnt_vstat.f_bfree = bfree / HAMMER_BUFSIZE;
688 	mp->mnt_vstat.f_bavail = mp->mnt_stat.f_bfree;
689 	if (mp->mnt_vstat.f_files < 0)
690 		mp->mnt_vstat.f_files = 0;
691 	*sbp = mp->mnt_vstat;
692 	return(0);
693 }
694 
695 /*
696  * Sync the filesystem.  Currently we have to run it twice, the second
697  * one will advance the undo start index to the end index, so if a crash
698  * occurs no undos will be run on mount.
699  *
700  * We do not sync the filesystem if we are called from a panic.  If we did
701  * we might end up blowing up a sync that was already in progress.
702  */
703 static int
704 hammer_vfs_sync(struct mount *mp, int waitfor)
705 {
706 	struct hammer_mount *hmp = (void *)mp->mnt_data;
707 	int error;
708 
709 	if (panicstr == NULL) {
710 		error = hammer_sync_hmp(hmp, waitfor);
711 		if (error == 0)
712 			error = hammer_sync_hmp(hmp, waitfor);
713 	} else {
714 		error = EIO;
715 	}
716 	return (error);
717 }
718 
719 /*
720  * Convert a vnode to a file handle.
721  */
722 static int
723 hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp)
724 {
725 	hammer_inode_t ip;
726 
727 	KKASSERT(MAXFIDSZ >= 16);
728 	ip = VTOI(vp);
729 	fhp->fid_len = offsetof(struct fid, fid_data[16]);
730 	fhp->fid_reserved = 0;
731 	bcopy(&ip->obj_id, fhp->fid_data + 0, sizeof(ip->obj_id));
732 	bcopy(&ip->obj_asof, fhp->fid_data + 8, sizeof(ip->obj_asof));
733 	return(0);
734 }
735 
736 
737 /*
738  * Convert a file handle back to a vnode.
739  */
740 static int
741 hammer_vfs_fhtovp(struct mount *mp, struct fid *fhp, struct vnode **vpp)
742 {
743 	struct hammer_transaction trans;
744 	struct hammer_inode *ip;
745 	struct hammer_inode_info info;
746 	int error;
747 
748 	bcopy(fhp->fid_data + 0, &info.obj_id, sizeof(info.obj_id));
749 	bcopy(fhp->fid_data + 8, &info.obj_asof, sizeof(info.obj_asof));
750 
751 	hammer_simple_transaction(&trans, (void *)mp->mnt_data);
752 
753 	/*
754 	 * Get/allocate the hammer_inode structure.  The structure must be
755 	 * unlocked while we manipulate the related vnode to avoid a
756 	 * deadlock.
757 	 */
758 	ip = hammer_get_inode(&trans, NULL, info.obj_id,
759 			      info.obj_asof, HAMMER_DEF_LOCALIZATION,
760 			      0, &error);
761 	if (ip == NULL) {
762 		*vpp = NULL;
763 		return(error);
764 	}
765 	error = hammer_get_vnode(ip, vpp);
766 	hammer_rel_inode(ip, 0);
767 	hammer_done_transaction(&trans);
768 	return (error);
769 }
770 
771 static int
772 hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
773 		    int *exflagsp, struct ucred **credanonp)
774 {
775 	hammer_mount_t hmp = (void *)mp->mnt_data;
776 	struct netcred *np;
777 	int error;
778 
779 	np = vfs_export_lookup(mp, &hmp->export, nam);
780 	if (np) {
781 		*exflagsp = np->netc_exflags;
782 		*credanonp = &np->netc_anon;
783 		error = 0;
784 	} else {
785 		error = EACCES;
786 	}
787 	return (error);
788 
789 }
790 
791 int
792 hammer_vfs_export(struct mount *mp, int op, const struct export_args *export)
793 {
794 	hammer_mount_t hmp = (void *)mp->mnt_data;
795 	int error;
796 
797 	switch(op) {
798 	case MOUNTCTL_SET_EXPORT:
799 		error = vfs_export(mp, &hmp->export, export);
800 		break;
801 	default:
802 		error = EOPNOTSUPP;
803 		break;
804 	}
805 	return(error);
806 }
807 
808