xref: /dflybsd-src/sys/vfs/hammer/hammer_vfsops.c (revision f5a07a7a074b65b763bd61fd65e0b29f12032fdd)
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 1.56 2008/06/28 23:50:37 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_stats_btree_lookups;
69 int64_t hammer_stats_btree_searches;
70 int64_t hammer_stats_btree_inserts;
71 int64_t hammer_stats_btree_deletes;
72 int64_t hammer_stats_btree_elements;
73 int64_t hammer_stats_btree_splits;
74 int64_t hammer_stats_btree_iterations;
75 int64_t hammer_stats_record_iterations;
76 int hammer_count_dirtybufspace;		/* global */
77 int hammer_count_refedbufs;		/* global */
78 int hammer_count_reservations;
79 int hammer_count_io_running_read;
80 int hammer_count_io_running_write;
81 int hammer_count_io_locked;
82 int hammer_limit_dirtybufspace;		/* per-mount */
83 int hammer_limit_recs;			/* as a whole XXX */
84 int hammer_limit_iqueued;		/* per-mount */
85 int hammer_bio_count;
86 int hammer_verify_zone;
87 int hammer_write_mode;
88 int64_t hammer_contention_count;
89 int64_t hammer_zone_limit;
90 
91 SYSCTL_NODE(_vfs, OID_AUTO, hammer, CTLFLAG_RW, 0, "HAMMER filesystem");
92 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_general, CTLFLAG_RW,
93 	   &hammer_debug_general, 0, "");
94 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_io, CTLFLAG_RW,
95 	   &hammer_debug_io, 0, "");
96 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_debug, CTLFLAG_RW,
97 	   &hammer_debug_debug, 0, "");
98 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_inode, CTLFLAG_RW,
99 	   &hammer_debug_inode, 0, "");
100 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_locks, CTLFLAG_RW,
101 	   &hammer_debug_locks, 0, "");
102 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_btree, CTLFLAG_RW,
103 	   &hammer_debug_btree, 0, "");
104 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_tid, CTLFLAG_RW,
105 	   &hammer_debug_tid, 0, "");
106 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover, CTLFLAG_RW,
107 	   &hammer_debug_recover, 0, "");
108 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover_faults, CTLFLAG_RW,
109 	   &hammer_debug_recover_faults, 0, "");
110 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_cluster_enable, CTLFLAG_RW,
111 	   &hammer_debug_cluster_enable, 0, "");
112 
113 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_dirtybufspace, CTLFLAG_RW,
114 	   &hammer_limit_dirtybufspace, 0, "");
115 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_recs, CTLFLAG_RW,
116 	   &hammer_limit_recs, 0, "");
117 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_iqueued, CTLFLAG_RW,
118 	   &hammer_limit_iqueued, 0, "");
119 
120 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_inodes, CTLFLAG_RD,
121 	   &hammer_count_inodes, 0, "");
122 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_iqueued, CTLFLAG_RD,
123 	   &hammer_count_iqueued, 0, "");
124 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reclaiming, CTLFLAG_RD,
125 	   &hammer_count_reclaiming, 0, "");
126 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_records, CTLFLAG_RD,
127 	   &hammer_count_records, 0, "");
128 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_record_datas, CTLFLAG_RD,
129 	   &hammer_count_record_datas, 0, "");
130 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_volumes, CTLFLAG_RD,
131 	   &hammer_count_volumes, 0, "");
132 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_buffers, CTLFLAG_RD,
133 	   &hammer_count_buffers, 0, "");
134 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_nodes, CTLFLAG_RD,
135 	   &hammer_count_nodes, 0, "");
136 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_searches, CTLFLAG_RD,
137 	   &hammer_stats_btree_searches, 0, "");
138 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_lookups, CTLFLAG_RD,
139 	   &hammer_stats_btree_lookups, 0, "");
140 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_inserts, CTLFLAG_RD,
141 	   &hammer_stats_btree_inserts, 0, "");
142 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_deletes, CTLFLAG_RD,
143 	   &hammer_stats_btree_deletes, 0, "");
144 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_elements, CTLFLAG_RD,
145 	   &hammer_stats_btree_elements, 0, "");
146 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_splits, CTLFLAG_RD,
147 	   &hammer_stats_btree_splits, 0, "");
148 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_iterations, CTLFLAG_RD,
149 	   &hammer_stats_btree_iterations, 0, "");
150 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_record_iterations, CTLFLAG_RD,
151 	   &hammer_stats_record_iterations, 0, "");
152 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_dirtybufspace, CTLFLAG_RD,
153 	   &hammer_count_dirtybufspace, 0, "");
154 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_refedbufs, CTLFLAG_RD,
155 	   &hammer_count_refedbufs, 0, "");
156 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reservations, CTLFLAG_RD,
157 	   &hammer_count_reservations, 0, "");
158 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_running_read, CTLFLAG_RD,
159 	   &hammer_count_io_running_read, 0, "");
160 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_locked, CTLFLAG_RD,
161 	   &hammer_count_io_locked, 0, "");
162 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_running_write, CTLFLAG_RD,
163 	   &hammer_count_io_running_write, 0, "");
164 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, zone_limit, CTLFLAG_RW,
165 	   &hammer_zone_limit, 0, "");
166 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, contention_count, CTLFLAG_RW,
167 	   &hammer_contention_count, 0, "");
168 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_zone, CTLFLAG_RW,
169 	   &hammer_verify_zone, 0, "");
170 SYSCTL_INT(_vfs_hammer, OID_AUTO, write_mode, CTLFLAG_RW,
171 	   &hammer_write_mode, 0, "");
172 
173 KTR_INFO_MASTER(hammer);
174 
175 /*
176  * VFS ABI
177  */
178 static void	hammer_free_hmp(struct mount *mp);
179 
180 static int	hammer_vfs_mount(struct mount *mp, char *path, caddr_t data,
181 				struct ucred *cred);
182 static int	hammer_vfs_unmount(struct mount *mp, int mntflags);
183 static int	hammer_vfs_root(struct mount *mp, struct vnode **vpp);
184 static int	hammer_vfs_statfs(struct mount *mp, struct statfs *sbp,
185 				struct ucred *cred);
186 static int	hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp,
187 				struct ucred *cred);
188 static int	hammer_vfs_sync(struct mount *mp, int waitfor);
189 static int	hammer_vfs_vget(struct mount *mp, ino_t ino,
190 				struct vnode **vpp);
191 static int	hammer_vfs_init(struct vfsconf *conf);
192 static int	hammer_vfs_fhtovp(struct mount *mp, struct fid *fhp,
193 				struct vnode **vpp);
194 static int	hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp);
195 static int	hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
196 				int *exflagsp, struct ucred **credanonp);
197 
198 
199 static struct vfsops hammer_vfsops = {
200 	.vfs_mount	= hammer_vfs_mount,
201 	.vfs_unmount	= hammer_vfs_unmount,
202 	.vfs_root 	= hammer_vfs_root,
203 	.vfs_statfs	= hammer_vfs_statfs,
204 	.vfs_statvfs	= hammer_vfs_statvfs,
205 	.vfs_sync	= hammer_vfs_sync,
206 	.vfs_vget	= hammer_vfs_vget,
207 	.vfs_init	= hammer_vfs_init,
208 	.vfs_vptofh	= hammer_vfs_vptofh,
209 	.vfs_fhtovp	= hammer_vfs_fhtovp,
210 	.vfs_checkexp	= hammer_vfs_checkexp
211 };
212 
213 MALLOC_DEFINE(M_HAMMER, "hammer-mount", "hammer mount");
214 
215 VFS_SET(hammer_vfsops, hammer, 0);
216 MODULE_VERSION(hammer, 1);
217 
218 static int
219 hammer_vfs_init(struct vfsconf *conf)
220 {
221 	if (hammer_limit_recs == 0)		/* XXX TODO */
222 		hammer_limit_recs = nbuf * 25;
223 	if (hammer_limit_dirtybufspace == 0) {
224 		hammer_limit_dirtybufspace = hidirtybufspace / 2;
225 		if (hammer_limit_dirtybufspace < 100)
226 			hammer_limit_dirtybufspace = 100;
227 	}
228 	if (hammer_limit_iqueued == 0)
229 		hammer_limit_iqueued = desiredvnodes / 5;
230 	return(0);
231 }
232 
233 static int
234 hammer_vfs_mount(struct mount *mp, char *mntpt, caddr_t data,
235 		 struct ucred *cred)
236 {
237 	struct hammer_mount_info info;
238 	hammer_mount_t hmp;
239 	hammer_volume_t rootvol;
240 	struct vnode *rootvp;
241 	const char *upath;	/* volume name in userspace */
242 	char *path;		/* volume name in system space */
243 	int error;
244 	int i;
245 
246 	if ((error = copyin(data, &info, sizeof(info))) != 0)
247 		return (error);
248 	if ((mp->mnt_flag & MNT_UPDATE) == 0) {
249 		if (info.nvolumes <= 0 || info.nvolumes >= 32768)
250 			return (EINVAL);
251 	}
252 	if ((info.hflags & HMNT_MASTERID) &&
253 	    (info.masterid < -1 || info.masterid >= HAMMER_MAX_MASTERS)) {
254 			return (EINVAL);
255 	}
256 
257 	/*
258 	 * Interal mount data structure
259 	 */
260 	if (mp->mnt_flag & MNT_UPDATE) {
261 		hmp = (void *)mp->mnt_data;
262 		KKASSERT(hmp != NULL);
263 	} else {
264 		hmp = kmalloc(sizeof(*hmp), M_HAMMER, M_WAITOK | M_ZERO);
265 		mp->mnt_data = (qaddr_t)hmp;
266 		hmp->mp = mp;
267 		hmp->namekey_iterator = mycpu->gd_time_seconds;
268 		/*TAILQ_INIT(&hmp->recycle_list);*/
269 
270 		hmp->root_btree_beg.localization = 0x00000000U;
271 		hmp->root_btree_beg.obj_id = -0x8000000000000000LL;
272 		hmp->root_btree_beg.key = -0x8000000000000000LL;
273 		hmp->root_btree_beg.create_tid = 1;
274 		hmp->root_btree_beg.delete_tid = 1;
275 		hmp->root_btree_beg.rec_type = 0;
276 		hmp->root_btree_beg.obj_type = 0;
277 
278 		hmp->root_btree_end.localization = 0xFFFFFFFFU;
279 		hmp->root_btree_end.obj_id = 0x7FFFFFFFFFFFFFFFLL;
280 		hmp->root_btree_end.key = 0x7FFFFFFFFFFFFFFFLL;
281 		hmp->root_btree_end.create_tid = 0xFFFFFFFFFFFFFFFFULL;
282 		hmp->root_btree_end.delete_tid = 0;   /* special case */
283 		hmp->root_btree_end.rec_type = 0xFFFFU;
284 		hmp->root_btree_end.obj_type = 0;
285 
286 		hmp->sync_lock.refs = 1;
287 		hmp->free_lock.refs = 1;
288 		hmp->undo_lock.refs = 1;
289 		hmp->blkmap_lock.refs = 1;
290 
291 		TAILQ_INIT(&hmp->flush_list);
292 		TAILQ_INIT(&hmp->delay_list);
293 		TAILQ_INIT(&hmp->objid_cache_list);
294 		TAILQ_INIT(&hmp->undo_lru_list);
295 		TAILQ_INIT(&hmp->reclaim_list);
296 	}
297 	hmp->hflags &= ~HMNT_USERFLAGS;
298 	hmp->hflags |= info.hflags & HMNT_USERFLAGS;
299 	if (info.hflags & HMNT_MASTERID)
300 		hmp->masterid = info.masterid;
301 	else
302 		hmp->masterid = -1;
303 	if (info.asof) {
304 		kprintf("ASOF\n");
305 		mp->mnt_flag |= MNT_RDONLY;
306 		hmp->asof = info.asof;
307 	} else {
308 		hmp->asof = HAMMER_MAX_TID;
309 	}
310 
311 	/*
312 	 * Re-open read-write if originally read-only, or vise-versa.
313 	 */
314 	if (mp->mnt_flag & MNT_UPDATE) {
315 		error = 0;
316 		if (hmp->ronly && (mp->mnt_kern_flag & MNTK_WANTRDWR)) {
317 			kprintf("HAMMER read-only -> read-write\n");
318 			hmp->ronly = 0;
319 			RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
320 				hammer_adjust_volume_mode, NULL);
321 			rootvol = hammer_get_root_volume(hmp, &error);
322 			if (rootvol) {
323 				hammer_recover_flush_buffers(hmp, rootvol, 1);
324 				bcopy(rootvol->ondisk->vol0_blockmap,
325 				      hmp->blockmap,
326 				      sizeof(hmp->blockmap));
327 				hammer_rel_volume(rootvol, 0);
328 			}
329 			RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
330 				hammer_reload_inode, NULL);
331 			/* kernel clears MNT_RDONLY */
332 		} else if (hmp->ronly == 0 && (mp->mnt_flag & MNT_RDONLY)) {
333 			kprintf("HAMMER read-write -> read-only\n");
334 			hmp->ronly = 1;	/* messy */
335 			RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
336 				hammer_reload_inode, NULL);
337 			hmp->ronly = 0;
338 			hammer_flusher_sync(hmp);
339 			hammer_flusher_sync(hmp);
340 			hammer_flusher_sync(hmp);
341 			hmp->ronly = 1;
342 			RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
343 				hammer_adjust_volume_mode, NULL);
344 		}
345 		return(error);
346 	}
347 
348 	RB_INIT(&hmp->rb_vols_root);
349 	RB_INIT(&hmp->rb_inos_root);
350 	RB_INIT(&hmp->rb_nods_root);
351 	RB_INIT(&hmp->rb_undo_root);
352 	RB_INIT(&hmp->rb_resv_root);
353 	RB_INIT(&hmp->rb_bufs_root);
354 
355 	hmp->ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
356 
357 	TAILQ_INIT(&hmp->volu_list);
358 	TAILQ_INIT(&hmp->undo_list);
359 	TAILQ_INIT(&hmp->data_list);
360 	TAILQ_INIT(&hmp->meta_list);
361 	TAILQ_INIT(&hmp->lose_list);
362 
363 	/*
364 	 * Load volumes
365 	 */
366 	path = objcache_get(namei_oc, M_WAITOK);
367 	hmp->nvolumes = info.nvolumes;
368 	for (i = 0; i < info.nvolumes; ++i) {
369 		error = copyin(&info.volumes[i], &upath, sizeof(char *));
370 		if (error == 0)
371 			error = copyinstr(upath, path, MAXPATHLEN, NULL);
372 		if (error == 0)
373 			error = hammer_install_volume(hmp, path);
374 		if (error)
375 			break;
376 	}
377 	objcache_put(namei_oc, path);
378 
379 	/*
380 	 * Make sure we found a root volume
381 	 */
382 	if (error == 0 && hmp->rootvol == NULL) {
383 		kprintf("hammer_mount: No root volume found!\n");
384 		error = EINVAL;
385 	}
386 	if (error) {
387 		hammer_free_hmp(mp);
388 		return (error);
389 	}
390 
391 	/*
392 	 * No errors, setup enough of the mount point so we can lookup the
393 	 * root vnode.
394 	 */
395 	mp->mnt_iosize_max = MAXPHYS;
396 	mp->mnt_kern_flag |= MNTK_FSMID;
397 
398 	/*
399 	 * note: f_iosize is used by vnode_pager_haspage() when constructing
400 	 * its VOP_BMAP call.
401 	 */
402 	mp->mnt_stat.f_iosize = HAMMER_BUFSIZE;
403 	mp->mnt_stat.f_bsize = HAMMER_BUFSIZE;
404 
405 	mp->mnt_vstat.f_frsize = HAMMER_BUFSIZE;
406 	mp->mnt_vstat.f_bsize = HAMMER_BUFSIZE;
407 
408 	mp->mnt_maxsymlinklen = 255;
409 	mp->mnt_flag |= MNT_LOCAL;
410 
411 	vfs_add_vnodeops(mp, &hammer_vnode_vops, &mp->mnt_vn_norm_ops);
412 	vfs_add_vnodeops(mp, &hammer_spec_vops, &mp->mnt_vn_spec_ops);
413 	vfs_add_vnodeops(mp, &hammer_fifo_vops, &mp->mnt_vn_fifo_ops);
414 
415 	/*
416 	 * The root volume's ondisk pointer is only valid if we hold a
417 	 * reference to it.
418 	 */
419 	rootvol = hammer_get_root_volume(hmp, &error);
420 	if (error)
421 		goto failed;
422 
423 	/*
424 	 * Perform any necessary UNDO operations.  The recovery code does
425 	 * call hammer_undo_lookup() so we have to pre-cache the blockmap,
426 	 * and then re-copy it again after recovery is complete.
427 	 *
428 	 * If this is a read-only mount the UNDO information is retained
429 	 * in memory in the form of dirty buffer cache buffers, and not
430 	 * written back to the media.
431 	 */
432 	bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
433 	      sizeof(hmp->blockmap));
434 
435 	error = hammer_recover(hmp, rootvol);
436 	if (error) {
437 		kprintf("Failed to recover HAMMER filesystem on mount\n");
438 		goto done;
439 	}
440 
441 	/*
442 	 * Finish setup now that we have a good root volume
443 	 */
444 	ksnprintf(mp->mnt_stat.f_mntfromname,
445 		  sizeof(mp->mnt_stat.f_mntfromname), "%s",
446 		  rootvol->ondisk->vol_name);
447 	mp->mnt_stat.f_fsid.val[0] =
448 		crc32((char *)&rootvol->ondisk->vol_fsid + 0, 8);
449 	mp->mnt_stat.f_fsid.val[1] =
450 		crc32((char *)&rootvol->ondisk->vol_fsid + 8, 8);
451 
452 	mp->mnt_vstat.f_fsid_uuid = rootvol->ondisk->vol_fsid;
453 	mp->mnt_vstat.f_fsid = crc32(&mp->mnt_vstat.f_fsid_uuid,
454 				     sizeof(mp->mnt_vstat.f_fsid_uuid));
455 
456 	/*
457 	 * Certain often-modified fields in the root volume are cached in
458 	 * the hammer_mount structure so we do not have to generate lots
459 	 * of little UNDO structures for them.
460 	 *
461 	 * Recopy after recovery.  This also has the side effect of
462 	 * setting our cached undo FIFO's first_offset, which serves to
463 	 * placemark the FIFO start for the NEXT flush cycle while the
464 	 * on-disk first_offset represents the LAST flush cycle.
465 	 */
466 	hmp->next_tid = rootvol->ondisk->vol0_next_tid;
467 	bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
468 	      sizeof(hmp->blockmap));
469 	hmp->copy_stat_freebigblocks = rootvol->ondisk->vol0_stat_freebigblocks;
470 
471 	hammer_flusher_create(hmp);
472 
473 	/*
474 	 * Locate the root directory using the root cluster's B-Tree as a
475 	 * starting point.  The root directory uses an obj_id of 1.
476 	 *
477 	 * FUTURE: Leave the root directory cached referenced but unlocked
478 	 * in hmp->rootvp (need to flush it on unmount).
479 	 */
480 	error = hammer_vfs_vget(mp, 1, &rootvp);
481 	if (error)
482 		goto done;
483 	vput(rootvp);
484 	/*vn_unlock(hmp->rootvp);*/
485 
486 done:
487 	hammer_rel_volume(rootvol, 0);
488 failed:
489 	/*
490 	 * Cleanup and return.
491 	 */
492 	if (error)
493 		hammer_free_hmp(mp);
494 	return (error);
495 }
496 
497 static int
498 hammer_vfs_unmount(struct mount *mp, int mntflags)
499 {
500 #if 0
501 	struct hammer_mount *hmp = (void *)mp->mnt_data;
502 #endif
503 	int flags;
504 	int error;
505 
506 	/*
507 	 * Clean out the vnodes
508 	 */
509 	flags = 0;
510 	if (mntflags & MNT_FORCE)
511 		flags |= FORCECLOSE;
512 	if ((error = vflush(mp, 0, flags)) != 0)
513 		return (error);
514 
515 	/*
516 	 * Clean up the internal mount structure and related entities.  This
517 	 * may issue I/O.
518 	 */
519 	hammer_free_hmp(mp);
520 	return(0);
521 }
522 
523 /*
524  * Clean up the internal mount structure and disassociate it from the mount.
525  * This may issue I/O.
526  */
527 static void
528 hammer_free_hmp(struct mount *mp)
529 {
530 	struct hammer_mount *hmp = (void *)mp->mnt_data;
531 
532 #if 0
533 	/*
534 	 * Clean up the root vnode
535 	 */
536 	if (hmp->rootvp) {
537 		vrele(hmp->rootvp);
538 		hmp->rootvp = NULL;
539 	}
540 #endif
541 	hammer_flusher_sync(hmp);
542 	hammer_flusher_sync(hmp);
543 	hammer_flusher_destroy(hmp);
544 
545 	KKASSERT(RB_EMPTY(&hmp->rb_inos_root));
546 
547 #if 0
548 	/*
549 	 * Unload & flush inodes
550 	 *
551 	 * XXX illegal to call this from here, it can only be done from
552 	 * the flusher.
553 	 */
554 	RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
555 		hammer_unload_inode, (void *)MNT_WAIT);
556 
557 	/*
558 	 * Unload & flush volumes
559 	 */
560 #endif
561 	/*
562 	 * Unload buffers and then volumes
563 	 */
564         RB_SCAN(hammer_buf_rb_tree, &hmp->rb_bufs_root, NULL,
565 		hammer_unload_buffer, NULL);
566 	RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
567 		hammer_unload_volume, NULL);
568 
569 	mp->mnt_data = NULL;
570 	mp->mnt_flag &= ~MNT_LOCAL;
571 	hmp->mp = NULL;
572 	hammer_destroy_objid_cache(hmp);
573 	kfree(hmp, M_HAMMER);
574 }
575 
576 /*
577  * Obtain a vnode for the specified inode number.  An exclusively locked
578  * vnode is returned.
579  */
580 int
581 hammer_vfs_vget(struct mount *mp, ino_t ino, struct vnode **vpp)
582 {
583 	struct hammer_transaction trans;
584 	struct hammer_mount *hmp = (void *)mp->mnt_data;
585 	struct hammer_inode *ip;
586 	int error;
587 
588 	hammer_simple_transaction(&trans, hmp);
589 
590 	/*
591 	 * Lookup the requested HAMMER inode.  The structure must be
592 	 * left unlocked while we manipulate the related vnode to avoid
593 	 * a deadlock.
594 	 */
595 	ip = hammer_get_inode(&trans, NULL, ino,
596 			      hmp->asof, HAMMER_DEF_LOCALIZATION,
597 			      0, &error);
598 	if (ip == NULL) {
599 		*vpp = NULL;
600 		return(error);
601 	}
602 	error = hammer_get_vnode(ip, vpp);
603 	hammer_rel_inode(ip, 0);
604 	hammer_done_transaction(&trans);
605 	return (error);
606 }
607 
608 /*
609  * Return the root vnode for the filesystem.
610  *
611  * HAMMER stores the root vnode in the hammer_mount structure so
612  * getting it is easy.
613  */
614 static int
615 hammer_vfs_root(struct mount *mp, struct vnode **vpp)
616 {
617 #if 0
618 	struct hammer_mount *hmp = (void *)mp->mnt_data;
619 #endif
620 	int error;
621 
622 	error = hammer_vfs_vget(mp, 1, vpp);
623 	return (error);
624 }
625 
626 static int
627 hammer_vfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
628 {
629 	struct hammer_mount *hmp = (void *)mp->mnt_data;
630 	hammer_volume_t volume;
631 	hammer_volume_ondisk_t ondisk;
632 	int error;
633 	int64_t bfree;
634 
635 	volume = hammer_get_root_volume(hmp, &error);
636 	if (error)
637 		return(error);
638 	ondisk = volume->ondisk;
639 
640 	/*
641 	 * Basic stats
642 	 */
643 	mp->mnt_stat.f_files = ondisk->vol0_stat_inodes;
644 	bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE;
645 	hammer_rel_volume(volume, 0);
646 
647 	mp->mnt_stat.f_bfree = bfree / HAMMER_BUFSIZE;
648 	mp->mnt_stat.f_bavail = mp->mnt_stat.f_bfree;
649 	if (mp->mnt_stat.f_files < 0)
650 		mp->mnt_stat.f_files = 0;
651 
652 	*sbp = mp->mnt_stat;
653 	return(0);
654 }
655 
656 static int
657 hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred)
658 {
659 	struct hammer_mount *hmp = (void *)mp->mnt_data;
660 	hammer_volume_t volume;
661 	hammer_volume_ondisk_t ondisk;
662 	int error;
663 	int64_t bfree;
664 
665 	volume = hammer_get_root_volume(hmp, &error);
666 	if (error)
667 		return(error);
668 	ondisk = volume->ondisk;
669 
670 	/*
671 	 * Basic stats
672 	 */
673 	mp->mnt_vstat.f_files = ondisk->vol0_stat_inodes;
674 	bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE;
675 	hammer_rel_volume(volume, 0);
676 
677 	mp->mnt_vstat.f_bfree = bfree / HAMMER_BUFSIZE;
678 	mp->mnt_vstat.f_bavail = mp->mnt_stat.f_bfree;
679 	if (mp->mnt_vstat.f_files < 0)
680 		mp->mnt_vstat.f_files = 0;
681 	*sbp = mp->mnt_vstat;
682 	return(0);
683 }
684 
685 /*
686  * Sync the filesystem.  Currently we have to run it twice, the second
687  * one will advance the undo start index to the end index, so if a crash
688  * occurs no undos will be run on mount.
689  *
690  * We do not sync the filesystem if we are called from a panic.  If we did
691  * we might end up blowing up a sync that was already in progress.
692  */
693 static int
694 hammer_vfs_sync(struct mount *mp, int waitfor)
695 {
696 	struct hammer_mount *hmp = (void *)mp->mnt_data;
697 	int error;
698 
699 	if (panicstr == NULL) {
700 		error = hammer_sync_hmp(hmp, waitfor);
701 		if (error == 0)
702 			error = hammer_sync_hmp(hmp, waitfor);
703 	} else {
704 		error = EIO;
705 	}
706 	return (error);
707 }
708 
709 /*
710  * Convert a vnode to a file handle.
711  */
712 static int
713 hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp)
714 {
715 	hammer_inode_t ip;
716 
717 	KKASSERT(MAXFIDSZ >= 16);
718 	ip = VTOI(vp);
719 	fhp->fid_len = offsetof(struct fid, fid_data[16]);
720 	fhp->fid_reserved = 0;
721 	bcopy(&ip->obj_id, fhp->fid_data + 0, sizeof(ip->obj_id));
722 	bcopy(&ip->obj_asof, fhp->fid_data + 8, sizeof(ip->obj_asof));
723 	return(0);
724 }
725 
726 
727 /*
728  * Convert a file handle back to a vnode.
729  */
730 static int
731 hammer_vfs_fhtovp(struct mount *mp, struct fid *fhp, struct vnode **vpp)
732 {
733 	struct hammer_transaction trans;
734 	struct hammer_inode *ip;
735 	struct hammer_inode_info info;
736 	int error;
737 
738 	bcopy(fhp->fid_data + 0, &info.obj_id, sizeof(info.obj_id));
739 	bcopy(fhp->fid_data + 8, &info.obj_asof, sizeof(info.obj_asof));
740 
741 	hammer_simple_transaction(&trans, (void *)mp->mnt_data);
742 
743 	/*
744 	 * Get/allocate the hammer_inode structure.  The structure must be
745 	 * unlocked while we manipulate the related vnode to avoid a
746 	 * deadlock.
747 	 */
748 	ip = hammer_get_inode(&trans, NULL, info.obj_id,
749 			      info.obj_asof, HAMMER_DEF_LOCALIZATION,
750 			      0, &error);
751 	if (ip == NULL) {
752 		*vpp = NULL;
753 		return(error);
754 	}
755 	error = hammer_get_vnode(ip, vpp);
756 	hammer_rel_inode(ip, 0);
757 	hammer_done_transaction(&trans);
758 	return (error);
759 }
760 
761 static int
762 hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
763 		    int *exflagsp, struct ucred **credanonp)
764 {
765 	hammer_mount_t hmp = (void *)mp->mnt_data;
766 	struct netcred *np;
767 	int error;
768 
769 	np = vfs_export_lookup(mp, &hmp->export, nam);
770 	if (np) {
771 		*exflagsp = np->netc_exflags;
772 		*credanonp = &np->netc_anon;
773 		error = 0;
774 	} else {
775 		error = EACCES;
776 	}
777 	return (error);
778 
779 }
780 
781 int
782 hammer_vfs_export(struct mount *mp, int op, const struct export_args *export)
783 {
784 	hammer_mount_t hmp = (void *)mp->mnt_data;
785 	int error;
786 
787 	switch(op) {
788 	case MOUNTCTL_SET_EXPORT:
789 		error = vfs_export(mp, &hmp->export, export);
790 		break;
791 	default:
792 		error = EOPNOTSUPP;
793 		break;
794 	}
795 	return(error);
796 }
797 
798