xref: /netbsd-src/sys/kern/vfs_subr.c (revision c0179c282a5968435315a82f4128c61372c68fc3)
1 /*	$NetBSD: vfs_subr.c,v 1.276 2006/11/17 17:05:18 hannken Exp $	*/
2 
3 /*-
4  * Copyright (c) 1997, 1998, 2004, 2005 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility,
9  * NASA Ames Research Center.
10  * This code is derived from software contributed to The NetBSD Foundation
11  * by Charles M. Hannum.
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
16  * 1. Redistributions of source code must retain the above copyright
17  *    notice, this list of conditions and the following disclaimer.
18  * 2. Redistributions in binary form must reproduce the above copyright
19  *    notice, this list of conditions and the following disclaimer in the
20  *    documentation and/or other materials provided with the distribution.
21  * 3. All advertising materials mentioning features or use of this software
22  *    must display the following acknowledgement:
23  *	This product includes software developed by the NetBSD
24  *	Foundation, Inc. and its contributors.
25  * 4. Neither the name of The NetBSD Foundation nor the names of its
26  *    contributors may be used to endorse or promote products derived
27  *    from this software without specific prior written permission.
28  *
29  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
30  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
31  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
32  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
33  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
34  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
35  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
36  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
37  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
38  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
39  * POSSIBILITY OF SUCH DAMAGE.
40  */
41 
42 /*
43  * Copyright (c) 1989, 1993
44  *	The Regents of the University of California.  All rights reserved.
45  * (c) UNIX System Laboratories, Inc.
46  * All or some portions of this file are derived from material licensed
47  * to the University of California by American Telephone and Telegraph
48  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
49  * the permission of UNIX System Laboratories, Inc.
50  *
51  * Redistribution and use in source and binary forms, with or without
52  * modification, are permitted provided that the following conditions
53  * are met:
54  * 1. Redistributions of source code must retain the above copyright
55  *    notice, this list of conditions and the following disclaimer.
56  * 2. Redistributions in binary form must reproduce the above copyright
57  *    notice, this list of conditions and the following disclaimer in the
58  *    documentation and/or other materials provided with the distribution.
59  * 3. Neither the name of the University nor the names of its contributors
60  *    may be used to endorse or promote products derived from this software
61  *    without specific prior written permission.
62  *
63  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
64  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
65  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
66  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
67  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
68  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
69  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
70  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
71  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
72  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
73  * SUCH DAMAGE.
74  *
75  *	@(#)vfs_subr.c	8.13 (Berkeley) 4/18/94
76  */
77 
78 /*
79  * External virtual filesystem routines
80  */
81 
82 #include <sys/cdefs.h>
83 __KERNEL_RCSID(0, "$NetBSD: vfs_subr.c,v 1.276 2006/11/17 17:05:18 hannken Exp $");
84 
85 #include "opt_inet.h"
86 #include "opt_ddb.h"
87 #include "opt_compat_netbsd.h"
88 #include "opt_compat_43.h"
89 
90 #include <sys/param.h>
91 #include <sys/systm.h>
92 #include <sys/proc.h>
93 #include <sys/kernel.h>
94 #include <sys/mount.h>
95 #include <sys/fcntl.h>
96 #include <sys/vnode.h>
97 #include <sys/stat.h>
98 #include <sys/namei.h>
99 #include <sys/ucred.h>
100 #include <sys/buf.h>
101 #include <sys/errno.h>
102 #include <sys/malloc.h>
103 #include <sys/domain.h>
104 #include <sys/mbuf.h>
105 #include <sys/sa.h>
106 #include <sys/syscallargs.h>
107 #include <sys/device.h>
108 #include <sys/filedesc.h>
109 #include <sys/kauth.h>
110 
111 #include <miscfs/specfs/specdev.h>
112 #include <miscfs/genfs/genfs.h>
113 #include <miscfs/syncfs/syncfs.h>
114 
115 #include <uvm/uvm.h>
116 #include <uvm/uvm_readahead.h>
117 #include <uvm/uvm_ddb.h>
118 
119 #include <sys/sysctl.h>
120 
121 const enum vtype iftovt_tab[16] = {
122 	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
123 	VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
124 };
125 const int	vttoif_tab[9] = {
126 	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
127 	S_IFSOCK, S_IFIFO, S_IFMT,
128 };
129 
130 int doforce = 1;		/* 1 => permit forcible unmounting */
131 int prtactive = 0;		/* 1 => print out reclaim of active vnodes */
132 
133 extern int dovfsusermount;	/* 1 => permit any user to mount filesystems */
134 extern int vfs_magiclinks;	/* 1 => expand "magic" symlinks */
135 
136 /*
137  * Insq/Remq for the vnode usage lists.
138  */
139 #define	bufinsvn(bp, dp)	LIST_INSERT_HEAD(dp, bp, b_vnbufs)
140 #define	bufremvn(bp) {							\
141 	LIST_REMOVE(bp, b_vnbufs);					\
142 	(bp)->b_vnbufs.le_next = NOLIST;				\
143 }
144 /* TAILQ_HEAD(freelst, vnode) vnode_free_list =	vnode free list (in vnode.h) */
145 struct freelst vnode_free_list = TAILQ_HEAD_INITIALIZER(vnode_free_list);
146 struct freelst vnode_hold_list = TAILQ_HEAD_INITIALIZER(vnode_hold_list);
147 
148 struct mntlist mountlist =			/* mounted filesystem list */
149     CIRCLEQ_HEAD_INITIALIZER(mountlist);
150 struct vfs_list_head vfs_list =			/* vfs list */
151     LIST_HEAD_INITIALIZER(vfs_list);
152 
153 struct simplelock mountlist_slock = SIMPLELOCK_INITIALIZER;
154 static struct simplelock mntid_slock = SIMPLELOCK_INITIALIZER;
155 struct simplelock mntvnode_slock = SIMPLELOCK_INITIALIZER;
156 struct simplelock vnode_free_list_slock = SIMPLELOCK_INITIALIZER;
157 struct simplelock spechash_slock = SIMPLELOCK_INITIALIZER;
158 
159 /* XXX - gross; single global lock to protect v_numoutput */
160 struct simplelock global_v_numoutput_slock = SIMPLELOCK_INITIALIZER;
161 
162 /*
163  * These define the root filesystem and device.
164  */
165 struct mount *rootfs;
166 struct vnode *rootvnode;
167 struct device *root_device;			/* root device */
168 
169 POOL_INIT(vnode_pool, sizeof(struct vnode), 0, 0, 0, "vnodepl",
170     &pool_allocator_nointr);
171 
172 MALLOC_DEFINE(M_VNODE, "vnodes", "Dynamically allocated vnodes");
173 
174 /*
175  * Local declarations.
176  */
177 
178 static specificdata_domain_t mount_specificdata_domain;
179 
180 static void insmntque(struct vnode *, struct mount *);
181 static int getdevvp(dev_t, struct vnode **, enum vtype);
182 static void vclean(struct vnode *, int, struct lwp *);
183 static struct vnode *getcleanvnode(struct lwp *);
184 
185 #ifdef DEBUG
186 void printlockedvnodes(void);
187 #endif
188 
189 /*
190  * Initialize the vnode management data structures.
191  */
192 void
193 vntblinit(void)
194 {
195 
196 	mount_specificdata_domain = specificdata_domain_create();
197 
198 	/*
199 	 * Initialize the filesystem syncer.
200 	 */
201 	vn_initialize_syncerd();
202 }
203 
204 int
205 vfs_drainvnodes(long target, struct lwp *l)
206 {
207 
208 	simple_lock(&vnode_free_list_slock);
209 	while (numvnodes > target) {
210 		struct vnode *vp;
211 
212 		vp = getcleanvnode(l);
213 		if (vp == NULL)
214 			return EBUSY; /* give up */
215 		pool_put(&vnode_pool, vp);
216 		simple_lock(&vnode_free_list_slock);
217 		numvnodes--;
218 	}
219 	simple_unlock(&vnode_free_list_slock);
220 
221 	return 0;
222 }
223 
224 /*
225  * grab a vnode from freelist and clean it.
226  */
227 struct vnode *
228 getcleanvnode(struct lwp *l)
229 {
230 	struct vnode *vp;
231 	struct mount *mp;
232 	struct freelst *listhd;
233 
234 	LOCK_ASSERT(simple_lock_held(&vnode_free_list_slock));
235 
236 	listhd = &vnode_free_list;
237 try_nextlist:
238 	TAILQ_FOREACH(vp, listhd, v_freelist) {
239 		if (!simple_lock_try(&vp->v_interlock))
240 			continue;
241 		/*
242 		 * as our lwp might hold the underlying vnode locked,
243 		 * don't try to reclaim the VLAYER vnode if it's locked.
244 		 */
245 		if ((vp->v_flag & VXLOCK) == 0 &&
246 		    ((vp->v_flag & VLAYER) == 0 || VOP_ISLOCKED(vp) == 0)) {
247 			if (vn_start_write(vp, &mp, V_NOWAIT) == 0)
248 				break;
249 		}
250 		mp = NULL;
251 		simple_unlock(&vp->v_interlock);
252 	}
253 
254 	if (vp == NULLVP) {
255 		if (listhd == &vnode_free_list) {
256 			listhd = &vnode_hold_list;
257 			goto try_nextlist;
258 		}
259 		simple_unlock(&vnode_free_list_slock);
260 		return NULLVP;
261 	}
262 
263 	if (vp->v_usecount)
264 		panic("free vnode isn't, vp %p", vp);
265 	TAILQ_REMOVE(listhd, vp, v_freelist);
266 	/* see comment on why 0xdeadb is set at end of vgone (below) */
267 	vp->v_freelist.tqe_prev = (struct vnode **)0xdeadb;
268 	simple_unlock(&vnode_free_list_slock);
269 	vp->v_lease = NULL;
270 
271 	if (vp->v_type != VBAD)
272 		vgonel(vp, l);
273 	else
274 		simple_unlock(&vp->v_interlock);
275 	vn_finished_write(mp, 0);
276 #ifdef DIAGNOSTIC
277 	if (vp->v_data || vp->v_uobj.uo_npages ||
278 	    TAILQ_FIRST(&vp->v_uobj.memq))
279 		panic("cleaned vnode isn't, vp %p", vp);
280 	if (vp->v_numoutput)
281 		panic("clean vnode has pending I/O's, vp %p", vp);
282 #endif
283 	KASSERT((vp->v_flag & VONWORKLST) == 0);
284 
285 	return vp;
286 }
287 
288 /*
289  * Mark a mount point as busy. Used to synchronize access and to delay
290  * unmounting. Interlock is not released on failure.
291  */
292 int
293 vfs_busy(struct mount *mp, int flags, struct simplelock *interlkp)
294 {
295 	int lkflags;
296 
297 	while (mp->mnt_iflag & IMNT_UNMOUNT) {
298 		int gone, n;
299 
300 		if (flags & LK_NOWAIT)
301 			return (ENOENT);
302 		if ((flags & LK_RECURSEFAIL) && mp->mnt_unmounter != NULL
303 		    && mp->mnt_unmounter == curlwp)
304 			return (EDEADLK);
305 		if (interlkp)
306 			simple_unlock(interlkp);
307 		/*
308 		 * Since all busy locks are shared except the exclusive
309 		 * lock granted when unmounting, the only place that a
310 		 * wakeup needs to be done is at the release of the
311 		 * exclusive lock at the end of dounmount.
312 		 */
313 		simple_lock(&mp->mnt_slock);
314 		mp->mnt_wcnt++;
315 		ltsleep((caddr_t)mp, PVFS, "vfs_busy", 0, &mp->mnt_slock);
316 		n = --mp->mnt_wcnt;
317 		simple_unlock(&mp->mnt_slock);
318 		gone = mp->mnt_iflag & IMNT_GONE;
319 
320 		if (n == 0)
321 			wakeup(&mp->mnt_wcnt);
322 		if (interlkp)
323 			simple_lock(interlkp);
324 		if (gone)
325 			return (ENOENT);
326 	}
327 	lkflags = LK_SHARED;
328 	if (interlkp)
329 		lkflags |= LK_INTERLOCK;
330 	if (lockmgr(&mp->mnt_lock, lkflags, interlkp))
331 		panic("vfs_busy: unexpected lock failure");
332 	return (0);
333 }
334 
335 /*
336  * Free a busy filesystem.
337  */
338 void
339 vfs_unbusy(struct mount *mp)
340 {
341 
342 	lockmgr(&mp->mnt_lock, LK_RELEASE, NULL);
343 }
344 
345 /*
346  * Lookup a filesystem type, and if found allocate and initialize
347  * a mount structure for it.
348  *
349  * Devname is usually updated by mount(8) after booting.
350  */
351 int
352 vfs_rootmountalloc(const char *fstypename, const char *devname,
353     struct mount **mpp)
354 {
355 	struct vfsops *vfsp = NULL;
356 	struct mount *mp;
357 
358 	LIST_FOREACH(vfsp, &vfs_list, vfs_list)
359 		if (!strncmp(vfsp->vfs_name, fstypename, MFSNAMELEN))
360 			break;
361 
362 	if (vfsp == NULL)
363 		return (ENODEV);
364 	mp = malloc((u_long)sizeof(struct mount), M_MOUNT, M_WAITOK);
365 	memset((char *)mp, 0, (u_long)sizeof(struct mount));
366 	lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, 0);
367 	simple_lock_init(&mp->mnt_slock);
368 	(void)vfs_busy(mp, LK_NOWAIT, 0);
369 	TAILQ_INIT(&mp->mnt_vnodelist);
370 	mp->mnt_op = vfsp;
371 	mp->mnt_flag = MNT_RDONLY;
372 	mp->mnt_vnodecovered = NULLVP;
373 	mp->mnt_leaf = mp;
374 	vfsp->vfs_refcount++;
375 	strncpy(mp->mnt_stat.f_fstypename, vfsp->vfs_name, MFSNAMELEN);
376 	mp->mnt_stat.f_mntonname[0] = '/';
377 	(void) copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 0);
378 	mount_initspecific(mp);
379 	*mpp = mp;
380 	return (0);
381 }
382 
383 /*
384  * Lookup a mount point by filesystem identifier.
385  */
386 struct mount *
387 vfs_getvfs(fsid_t *fsid)
388 {
389 	struct mount *mp;
390 
391 	simple_lock(&mountlist_slock);
392 	CIRCLEQ_FOREACH(mp, &mountlist, mnt_list) {
393 		if (mp->mnt_stat.f_fsidx.__fsid_val[0] == fsid->__fsid_val[0] &&
394 		    mp->mnt_stat.f_fsidx.__fsid_val[1] == fsid->__fsid_val[1]) {
395 			simple_unlock(&mountlist_slock);
396 			return (mp);
397 		}
398 	}
399 	simple_unlock(&mountlist_slock);
400 	return ((struct mount *)0);
401 }
402 
403 /*
404  * Get a new unique fsid
405  */
406 void
407 vfs_getnewfsid(struct mount *mp)
408 {
409 	static u_short xxxfs_mntid;
410 	fsid_t tfsid;
411 	int mtype;
412 
413 	simple_lock(&mntid_slock);
414 	mtype = makefstype(mp->mnt_op->vfs_name);
415 	mp->mnt_stat.f_fsidx.__fsid_val[0] = makedev(mtype, 0);
416 	mp->mnt_stat.f_fsidx.__fsid_val[1] = mtype;
417 	mp->mnt_stat.f_fsid = mp->mnt_stat.f_fsidx.__fsid_val[0];
418 	if (xxxfs_mntid == 0)
419 		++xxxfs_mntid;
420 	tfsid.__fsid_val[0] = makedev(mtype & 0xff, xxxfs_mntid);
421 	tfsid.__fsid_val[1] = mtype;
422 	if (!CIRCLEQ_EMPTY(&mountlist)) {
423 		while (vfs_getvfs(&tfsid)) {
424 			tfsid.__fsid_val[0]++;
425 			xxxfs_mntid++;
426 		}
427 	}
428 	mp->mnt_stat.f_fsidx.__fsid_val[0] = tfsid.__fsid_val[0];
429 	mp->mnt_stat.f_fsid = mp->mnt_stat.f_fsidx.__fsid_val[0];
430 	simple_unlock(&mntid_slock);
431 }
432 
433 /*
434  * Make a 'unique' number from a mount type name.
435  */
436 long
437 makefstype(const char *type)
438 {
439 	long rv;
440 
441 	for (rv = 0; *type; type++) {
442 		rv <<= 2;
443 		rv ^= *type;
444 	}
445 	return rv;
446 }
447 
448 
449 /*
450  * Set vnode attributes to VNOVAL
451  */
452 void
453 vattr_null(struct vattr *vap)
454 {
455 
456 	vap->va_type = VNON;
457 
458 	/*
459 	 * Assign individually so that it is safe even if size and
460 	 * sign of each member are varied.
461 	 */
462 	vap->va_mode = VNOVAL;
463 	vap->va_nlink = VNOVAL;
464 	vap->va_uid = VNOVAL;
465 	vap->va_gid = VNOVAL;
466 	vap->va_fsid = VNOVAL;
467 	vap->va_fileid = VNOVAL;
468 	vap->va_size = VNOVAL;
469 	vap->va_blocksize = VNOVAL;
470 	vap->va_atime.tv_sec =
471 	    vap->va_mtime.tv_sec =
472 	    vap->va_ctime.tv_sec =
473 	    vap->va_birthtime.tv_sec = VNOVAL;
474 	vap->va_atime.tv_nsec =
475 	    vap->va_mtime.tv_nsec =
476 	    vap->va_ctime.tv_nsec =
477 	    vap->va_birthtime.tv_nsec = VNOVAL;
478 	vap->va_gen = VNOVAL;
479 	vap->va_flags = VNOVAL;
480 	vap->va_rdev = VNOVAL;
481 	vap->va_bytes = VNOVAL;
482 	vap->va_vaflags = 0;
483 }
484 
485 /*
486  * Routines having to do with the management of the vnode table.
487  */
488 extern int (**dead_vnodeop_p)(void *);
489 long numvnodes;
490 
491 /*
492  * Return the next vnode from the free list.
493  */
494 int
495 getnewvnode(enum vtagtype tag, struct mount *mp, int (**vops)(void *),
496     struct vnode **vpp)
497 {
498 	extern struct uvm_pagerops uvm_vnodeops;
499 	struct uvm_object *uobj;
500 	struct lwp *l = curlwp;		/* XXX */
501 	static int toggle;
502 	struct vnode *vp;
503 	int error = 0, tryalloc;
504 
505  try_again:
506 	if (mp) {
507 		/*
508 		 * Mark filesystem busy while we're creating a vnode.
509 		 * If unmount is in progress, this will wait; if the
510 		 * unmount succeeds (only if umount -f), this will
511 		 * return an error.  If the unmount fails, we'll keep
512 		 * going afterwards.
513 		 * (This puts the per-mount vnode list logically under
514 		 * the protection of the vfs_busy lock).
515 		 */
516 		error = vfs_busy(mp, LK_RECURSEFAIL, 0);
517 		if (error && error != EDEADLK)
518 			return error;
519 	}
520 
521 	/*
522 	 * We must choose whether to allocate a new vnode or recycle an
523 	 * existing one. The criterion for allocating a new one is that
524 	 * the total number of vnodes is less than the number desired or
525 	 * there are no vnodes on either free list. Generally we only
526 	 * want to recycle vnodes that have no buffers associated with
527 	 * them, so we look first on the vnode_free_list. If it is empty,
528 	 * we next consider vnodes with referencing buffers on the
529 	 * vnode_hold_list. The toggle ensures that half the time we
530 	 * will use a buffer from the vnode_hold_list, and half the time
531 	 * we will allocate a new one unless the list has grown to twice
532 	 * the desired size. We are reticent to recycle vnodes from the
533 	 * vnode_hold_list because we will lose the identity of all its
534 	 * referencing buffers.
535 	 */
536 
537 	vp = NULL;
538 
539 	simple_lock(&vnode_free_list_slock);
540 
541 	toggle ^= 1;
542 	if (numvnodes > 2 * desiredvnodes)
543 		toggle = 0;
544 
545 	tryalloc = numvnodes < desiredvnodes ||
546 	    (TAILQ_FIRST(&vnode_free_list) == NULL &&
547 	     (TAILQ_FIRST(&vnode_hold_list) == NULL || toggle));
548 
549 	if (tryalloc &&
550 	    (vp = pool_get(&vnode_pool, PR_NOWAIT)) != NULL) {
551 		numvnodes++;
552 		simple_unlock(&vnode_free_list_slock);
553 		memset(vp, 0, sizeof(*vp));
554 		UVM_OBJ_INIT(&vp->v_uobj, &uvm_vnodeops, 1);
555 		/*
556 		 * done by memset() above.
557 		 *	LIST_INIT(&vp->v_nclist);
558 		 *	LIST_INIT(&vp->v_dnclist);
559 		 */
560 	} else {
561 		vp = getcleanvnode(l);
562 		/*
563 		 * Unless this is a bad time of the month, at most
564 		 * the first NCPUS items on the free list are
565 		 * locked, so this is close enough to being empty.
566 		 */
567 		if (vp == NULLVP) {
568 			if (mp && error != EDEADLK)
569 				vfs_unbusy(mp);
570 			if (tryalloc) {
571 				printf("WARNING: unable to allocate new "
572 				    "vnode, retrying...\n");
573 				(void) tsleep(&lbolt, PRIBIO, "newvn", hz);
574 				goto try_again;
575 			}
576 			tablefull("vnode", "increase kern.maxvnodes or NVNODE");
577 			*vpp = 0;
578 			return (ENFILE);
579 		}
580 		vp->v_usecount = 1;
581 		vp->v_flag = 0;
582 		vp->v_socket = NULL;
583 	}
584 	vp->v_type = VNON;
585 	vp->v_vnlock = &vp->v_lock;
586 	lockinit(vp->v_vnlock, PVFS, "vnlock", 0, 0);
587 	KASSERT(LIST_EMPTY(&vp->v_nclist));
588 	KASSERT(LIST_EMPTY(&vp->v_dnclist));
589 	vp->v_tag = tag;
590 	vp->v_op = vops;
591 	insmntque(vp, mp);
592 	*vpp = vp;
593 	vp->v_data = 0;
594 	simple_lock_init(&vp->v_interlock);
595 
596 	/*
597 	 * initialize uvm_object within vnode.
598 	 */
599 
600 	uobj = &vp->v_uobj;
601 	KASSERT(uobj->pgops == &uvm_vnodeops);
602 	KASSERT(uobj->uo_npages == 0);
603 	KASSERT(TAILQ_FIRST(&uobj->memq) == NULL);
604 	vp->v_size = VSIZENOTSET;
605 
606 	if (mp && error != EDEADLK)
607 		vfs_unbusy(mp);
608 	return (0);
609 }
610 
611 /*
612  * This is really just the reverse of getnewvnode(). Needed for
613  * VFS_VGET functions who may need to push back a vnode in case
614  * of a locking race.
615  */
616 void
617 ungetnewvnode(struct vnode *vp)
618 {
619 #ifdef DIAGNOSTIC
620 	if (vp->v_usecount != 1)
621 		panic("ungetnewvnode: busy vnode");
622 #endif
623 	vp->v_usecount--;
624 	insmntque(vp, NULL);
625 	vp->v_type = VBAD;
626 
627 	simple_lock(&vp->v_interlock);
628 	/*
629 	 * Insert at head of LRU list
630 	 */
631 	simple_lock(&vnode_free_list_slock);
632 	if (vp->v_holdcnt > 0)
633 		TAILQ_INSERT_HEAD(&vnode_hold_list, vp, v_freelist);
634 	else
635 		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
636 	simple_unlock(&vnode_free_list_slock);
637 	simple_unlock(&vp->v_interlock);
638 }
639 
640 /*
641  * Move a vnode from one mount queue to another.
642  */
643 static void
644 insmntque(struct vnode *vp, struct mount *mp)
645 {
646 
647 #ifdef DIAGNOSTIC
648 	if ((mp != NULL) &&
649 	    (mp->mnt_iflag & IMNT_UNMOUNT) &&
650 	    !(mp->mnt_flag & MNT_SOFTDEP) &&
651 	    vp->v_tag != VT_VFS) {
652 		panic("insmntque into dying filesystem");
653 	}
654 #endif
655 
656 	simple_lock(&mntvnode_slock);
657 	/*
658 	 * Delete from old mount point vnode list, if on one.
659 	 */
660 	if (vp->v_mount != NULL)
661 		TAILQ_REMOVE(&vp->v_mount->mnt_vnodelist, vp, v_mntvnodes);
662 	/*
663 	 * Insert into list of vnodes for the new mount point, if available.
664 	 */
665 	if ((vp->v_mount = mp) != NULL) {
666 		if (TAILQ_EMPTY(&mp->mnt_vnodelist)) {
667 			TAILQ_INSERT_HEAD(&mp->mnt_vnodelist, vp, v_mntvnodes);
668 		} else {
669 			TAILQ_INSERT_TAIL(&mp->mnt_vnodelist, vp, v_mntvnodes);
670 		}
671 	}
672 	simple_unlock(&mntvnode_slock);
673 }
674 
675 /*
676  * Update outstanding I/O count and do wakeup if requested.
677  */
678 void
679 vwakeup(struct buf *bp)
680 {
681 	struct vnode *vp;
682 
683 	if ((vp = bp->b_vp) != NULL) {
684 		/* XXX global lock hack
685 		 * can't use v_interlock here since this is called
686 		 * in interrupt context from biodone().
687 		 */
688 		simple_lock(&global_v_numoutput_slock);
689 		if (--vp->v_numoutput < 0)
690 			panic("vwakeup: neg numoutput, vp %p", vp);
691 		if ((vp->v_flag & VBWAIT) && vp->v_numoutput <= 0) {
692 			vp->v_flag &= ~VBWAIT;
693 			wakeup((caddr_t)&vp->v_numoutput);
694 		}
695 		simple_unlock(&global_v_numoutput_slock);
696 	}
697 }
698 
699 /*
700  * Flush out and invalidate all buffers associated with a vnode.
701  * Called with the underlying vnode locked, which should prevent new dirty
702  * buffers from being queued.
703  */
704 int
705 vinvalbuf(struct vnode *vp, int flags, kauth_cred_t cred, struct lwp *l,
706     int slpflag, int slptimeo)
707 {
708 	struct buf *bp, *nbp;
709 	int s, error;
710 	int flushflags = PGO_ALLPAGES | PGO_FREE | PGO_SYNCIO |
711 		(flags & V_SAVE ? PGO_CLEANIT : 0);
712 
713 	/* XXXUBC this doesn't look at flags or slp* */
714 	simple_lock(&vp->v_interlock);
715 	error = VOP_PUTPAGES(vp, 0, 0, flushflags);
716 	if (error) {
717 		return error;
718 	}
719 
720 	if (flags & V_SAVE) {
721 		error = VOP_FSYNC(vp, cred, FSYNC_WAIT|FSYNC_RECLAIM, 0, 0, l);
722 		if (error)
723 		        return (error);
724 #ifdef DIAGNOSTIC
725 		s = splbio();
726 		if (vp->v_numoutput > 0 || !LIST_EMPTY(&vp->v_dirtyblkhd))
727 		        panic("vinvalbuf: dirty bufs, vp %p", vp);
728 		splx(s);
729 #endif
730 	}
731 
732 	s = splbio();
733 
734 restart:
735 	for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
736 		nbp = LIST_NEXT(bp, b_vnbufs);
737 		simple_lock(&bp->b_interlock);
738 		if (bp->b_flags & B_BUSY) {
739 			bp->b_flags |= B_WANTED;
740 			error = ltsleep((caddr_t)bp,
741 				    slpflag | (PRIBIO + 1) | PNORELOCK,
742 				    "vinvalbuf", slptimeo, &bp->b_interlock);
743 			if (error) {
744 				splx(s);
745 				return (error);
746 			}
747 			goto restart;
748 		}
749 		bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH;
750 		simple_unlock(&bp->b_interlock);
751 		brelse(bp);
752 	}
753 
754 	for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
755 		nbp = LIST_NEXT(bp, b_vnbufs);
756 		simple_lock(&bp->b_interlock);
757 		if (bp->b_flags & B_BUSY) {
758 			bp->b_flags |= B_WANTED;
759 			error = ltsleep((caddr_t)bp,
760 				    slpflag | (PRIBIO + 1) | PNORELOCK,
761 				    "vinvalbuf", slptimeo, &bp->b_interlock);
762 			if (error) {
763 				splx(s);
764 				return (error);
765 			}
766 			goto restart;
767 		}
768 		/*
769 		 * XXX Since there are no node locks for NFS, I believe
770 		 * there is a slight chance that a delayed write will
771 		 * occur while sleeping just above, so check for it.
772 		 */
773 		if ((bp->b_flags & B_DELWRI) && (flags & V_SAVE)) {
774 #ifdef DEBUG
775 			printf("buffer still DELWRI\n");
776 #endif
777 			bp->b_flags |= B_BUSY | B_VFLUSH;
778 			simple_unlock(&bp->b_interlock);
779 			VOP_BWRITE(bp);
780 			goto restart;
781 		}
782 		bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH;
783 		simple_unlock(&bp->b_interlock);
784 		brelse(bp);
785 	}
786 
787 #ifdef DIAGNOSTIC
788 	if (!LIST_EMPTY(&vp->v_cleanblkhd) || !LIST_EMPTY(&vp->v_dirtyblkhd))
789 		panic("vinvalbuf: flush failed, vp %p", vp);
790 #endif
791 
792 	splx(s);
793 
794 	return (0);
795 }
796 
797 /*
798  * Destroy any in core blocks past the truncation length.
799  * Called with the underlying vnode locked, which should prevent new dirty
800  * buffers from being queued.
801  */
802 int
803 vtruncbuf(struct vnode *vp, daddr_t lbn, int slpflag, int slptimeo)
804 {
805 	struct buf *bp, *nbp;
806 	int s, error;
807 	voff_t off;
808 
809 	off = round_page((voff_t)lbn << vp->v_mount->mnt_fs_bshift);
810 	simple_lock(&vp->v_interlock);
811 	error = VOP_PUTPAGES(vp, off, 0, PGO_FREE | PGO_SYNCIO);
812 	if (error) {
813 		return error;
814 	}
815 
816 	s = splbio();
817 
818 restart:
819 	for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
820 		nbp = LIST_NEXT(bp, b_vnbufs);
821 		if (bp->b_lblkno < lbn)
822 			continue;
823 		simple_lock(&bp->b_interlock);
824 		if (bp->b_flags & B_BUSY) {
825 			bp->b_flags |= B_WANTED;
826 			error = ltsleep(bp, slpflag | (PRIBIO + 1) | PNORELOCK,
827 			    "vtruncbuf", slptimeo, &bp->b_interlock);
828 			if (error) {
829 				splx(s);
830 				return (error);
831 			}
832 			goto restart;
833 		}
834 		bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH;
835 		simple_unlock(&bp->b_interlock);
836 		brelse(bp);
837 	}
838 
839 	for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
840 		nbp = LIST_NEXT(bp, b_vnbufs);
841 		if (bp->b_lblkno < lbn)
842 			continue;
843 		simple_lock(&bp->b_interlock);
844 		if (bp->b_flags & B_BUSY) {
845 			bp->b_flags |= B_WANTED;
846 			error = ltsleep(bp, slpflag | (PRIBIO + 1) | PNORELOCK,
847 			    "vtruncbuf", slptimeo, &bp->b_interlock);
848 			if (error) {
849 				splx(s);
850 				return (error);
851 			}
852 			goto restart;
853 		}
854 		bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH;
855 		simple_unlock(&bp->b_interlock);
856 		brelse(bp);
857 	}
858 
859 	splx(s);
860 
861 	return (0);
862 }
863 
864 void
865 vflushbuf(struct vnode *vp, int sync)
866 {
867 	struct buf *bp, *nbp;
868 	int flags = PGO_CLEANIT | PGO_ALLPAGES | (sync ? PGO_SYNCIO : 0);
869 	int s;
870 
871 	simple_lock(&vp->v_interlock);
872 	(void) VOP_PUTPAGES(vp, 0, 0, flags);
873 
874 loop:
875 	s = splbio();
876 	for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
877 		nbp = LIST_NEXT(bp, b_vnbufs);
878 		simple_lock(&bp->b_interlock);
879 		if ((bp->b_flags & B_BUSY)) {
880 			simple_unlock(&bp->b_interlock);
881 			continue;
882 		}
883 		if ((bp->b_flags & B_DELWRI) == 0)
884 			panic("vflushbuf: not dirty, bp %p", bp);
885 		bp->b_flags |= B_BUSY | B_VFLUSH;
886 		simple_unlock(&bp->b_interlock);
887 		splx(s);
888 		/*
889 		 * Wait for I/O associated with indirect blocks to complete,
890 		 * since there is no way to quickly wait for them below.
891 		 */
892 		if (bp->b_vp == vp || sync == 0)
893 			(void) bawrite(bp);
894 		else
895 			(void) bwrite(bp);
896 		goto loop;
897 	}
898 	if (sync == 0) {
899 		splx(s);
900 		return;
901 	}
902 	simple_lock(&global_v_numoutput_slock);
903 	while (vp->v_numoutput) {
904 		vp->v_flag |= VBWAIT;
905 		ltsleep((caddr_t)&vp->v_numoutput, PRIBIO + 1, "vflushbuf", 0,
906 			&global_v_numoutput_slock);
907 	}
908 	simple_unlock(&global_v_numoutput_slock);
909 	splx(s);
910 	if (!LIST_EMPTY(&vp->v_dirtyblkhd)) {
911 		vprint("vflushbuf: dirty", vp);
912 		goto loop;
913 	}
914 }
915 
916 /*
917  * Associate a buffer with a vnode.
918  */
919 void
920 bgetvp(struct vnode *vp, struct buf *bp)
921 {
922 	int s;
923 
924 	if (bp->b_vp)
925 		panic("bgetvp: not free, bp %p", bp);
926 	VHOLD(vp);
927 	s = splbio();
928 	bp->b_vp = vp;
929 	if (vp->v_type == VBLK || vp->v_type == VCHR)
930 		bp->b_dev = vp->v_rdev;
931 	else
932 		bp->b_dev = NODEV;
933 	/*
934 	 * Insert onto list for new vnode.
935 	 */
936 	bufinsvn(bp, &vp->v_cleanblkhd);
937 	splx(s);
938 }
939 
940 /*
941  * Disassociate a buffer from a vnode.
942  */
943 void
944 brelvp(struct buf *bp)
945 {
946 	struct vnode *vp;
947 	int s;
948 
949 	if (bp->b_vp == NULL)
950 		panic("brelvp: vp NULL, bp %p", bp);
951 
952 	s = splbio();
953 	vp = bp->b_vp;
954 	/*
955 	 * Delete from old vnode list, if on one.
956 	 */
957 	if (LIST_NEXT(bp, b_vnbufs) != NOLIST)
958 		bufremvn(bp);
959 
960 	if (TAILQ_EMPTY(&vp->v_uobj.memq) && (vp->v_flag & VONWORKLST) &&
961 	    LIST_FIRST(&vp->v_dirtyblkhd) == NULL) {
962 		vp->v_flag &= ~VWRITEMAPDIRTY;
963 		vn_syncer_remove_from_worklist(vp);
964 	}
965 
966 	bp->b_vp = NULL;
967 	HOLDRELE(vp);
968 	splx(s);
969 }
970 
971 /*
972  * Reassign a buffer from one vnode to another.
973  * Used to assign file specific control information
974  * (indirect blocks) to the vnode to which they belong.
975  *
976  * This function must be called at splbio().
977  */
978 void
979 reassignbuf(struct buf *bp, struct vnode *newvp)
980 {
981 	struct buflists *listheadp;
982 	int delayx;
983 
984 	/*
985 	 * Delete from old vnode list, if on one.
986 	 */
987 	if (LIST_NEXT(bp, b_vnbufs) != NOLIST)
988 		bufremvn(bp);
989 	/*
990 	 * If dirty, put on list of dirty buffers;
991 	 * otherwise insert onto list of clean buffers.
992 	 */
993 	if ((bp->b_flags & B_DELWRI) == 0) {
994 		listheadp = &newvp->v_cleanblkhd;
995 		if (TAILQ_EMPTY(&newvp->v_uobj.memq) &&
996 		    (newvp->v_flag & VONWORKLST) &&
997 		    LIST_FIRST(&newvp->v_dirtyblkhd) == NULL) {
998 			newvp->v_flag &= ~VWRITEMAPDIRTY;
999 			vn_syncer_remove_from_worklist(newvp);
1000 		}
1001 	} else {
1002 		listheadp = &newvp->v_dirtyblkhd;
1003 		if ((newvp->v_flag & VONWORKLST) == 0) {
1004 			switch (newvp->v_type) {
1005 			case VDIR:
1006 				delayx = dirdelay;
1007 				break;
1008 			case VBLK:
1009 				if (newvp->v_specmountpoint != NULL) {
1010 					delayx = metadelay;
1011 					break;
1012 				}
1013 				/* fall through */
1014 			default:
1015 				delayx = filedelay;
1016 				break;
1017 			}
1018 			if (!newvp->v_mount ||
1019 			    (newvp->v_mount->mnt_flag & MNT_ASYNC) == 0)
1020 				vn_syncer_add_to_worklist(newvp, delayx);
1021 		}
1022 	}
1023 	bufinsvn(bp, listheadp);
1024 }
1025 
1026 /*
1027  * Create a vnode for a block device.
1028  * Used for root filesystem and swap areas.
1029  * Also used for memory file system special devices.
1030  */
1031 int
1032 bdevvp(dev_t dev, struct vnode **vpp)
1033 {
1034 
1035 	return (getdevvp(dev, vpp, VBLK));
1036 }
1037 
1038 /*
1039  * Create a vnode for a character device.
1040  * Used for kernfs and some console handling.
1041  */
1042 int
1043 cdevvp(dev_t dev, struct vnode **vpp)
1044 {
1045 
1046 	return (getdevvp(dev, vpp, VCHR));
1047 }
1048 
1049 /*
1050  * Create a vnode for a device.
1051  * Used by bdevvp (block device) for root file system etc.,
1052  * and by cdevvp (character device) for console and kernfs.
1053  */
1054 static int
1055 getdevvp(dev_t dev, struct vnode **vpp, enum vtype type)
1056 {
1057 	struct vnode *vp;
1058 	struct vnode *nvp;
1059 	int error;
1060 
1061 	if (dev == NODEV) {
1062 		*vpp = NULLVP;
1063 		return (0);
1064 	}
1065 	error = getnewvnode(VT_NON, NULL, spec_vnodeop_p, &nvp);
1066 	if (error) {
1067 		*vpp = NULLVP;
1068 		return (error);
1069 	}
1070 	vp = nvp;
1071 	vp->v_type = type;
1072 	if ((nvp = checkalias(vp, dev, NULL)) != 0) {
1073 		vput(vp);
1074 		vp = nvp;
1075 	}
1076 	*vpp = vp;
1077 	return (0);
1078 }
1079 
1080 /*
1081  * Check to see if the new vnode represents a special device
1082  * for which we already have a vnode (either because of
1083  * bdevvp() or because of a different vnode representing
1084  * the same block device). If such an alias exists, deallocate
1085  * the existing contents and return the aliased vnode. The
1086  * caller is responsible for filling it with its new contents.
1087  */
1088 struct vnode *
1089 checkalias(struct vnode *nvp, dev_t nvp_rdev, struct mount *mp)
1090 {
1091 	struct lwp *l = curlwp;		/* XXX */
1092 	struct vnode *vp;
1093 	struct vnode **vpp;
1094 
1095 	if (nvp->v_type != VBLK && nvp->v_type != VCHR)
1096 		return (NULLVP);
1097 
1098 	vpp = &speclisth[SPECHASH(nvp_rdev)];
1099 loop:
1100 	simple_lock(&spechash_slock);
1101 	for (vp = *vpp; vp; vp = vp->v_specnext) {
1102 		if (nvp_rdev != vp->v_rdev || nvp->v_type != vp->v_type)
1103 			continue;
1104 		/*
1105 		 * Alias, but not in use, so flush it out.
1106 		 */
1107 		simple_lock(&vp->v_interlock);
1108 		simple_unlock(&spechash_slock);
1109 		if (vp->v_usecount == 0) {
1110 			vgonel(vp, l);
1111 			goto loop;
1112 		}
1113 		/*
1114 		 * What we're interested to know here is if someone else has
1115 		 * removed this vnode from the device hash list while we were
1116 		 * waiting.  This can only happen if vclean() did it, and
1117 		 * this requires the vnode to be locked.
1118 		 */
1119 		if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK))
1120 			goto loop;
1121 		if (vp->v_specinfo == NULL) {
1122 			vput(vp);
1123 			goto loop;
1124 		}
1125 		simple_lock(&spechash_slock);
1126 		break;
1127 	}
1128 	if (vp == NULL || vp->v_tag != VT_NON || vp->v_type != VBLK) {
1129 		MALLOC(nvp->v_specinfo, struct specinfo *,
1130 			sizeof(struct specinfo), M_VNODE, M_NOWAIT);
1131 		/* XXX Erg. */
1132 		if (nvp->v_specinfo == NULL) {
1133 			simple_unlock(&spechash_slock);
1134 			uvm_wait("checkalias");
1135 			goto loop;
1136 		}
1137 
1138 		nvp->v_rdev = nvp_rdev;
1139 		nvp->v_hashchain = vpp;
1140 		nvp->v_specnext = *vpp;
1141 		nvp->v_specmountpoint = NULL;
1142 		simple_unlock(&spechash_slock);
1143 		nvp->v_speclockf = NULL;
1144 		simple_lock_init(&nvp->v_spec_cow_slock);
1145 		SLIST_INIT(&nvp->v_spec_cow_head);
1146 		nvp->v_spec_cow_req = 0;
1147 		nvp->v_spec_cow_count = 0;
1148 
1149 		*vpp = nvp;
1150 		if (vp != NULLVP) {
1151 			nvp->v_flag |= VALIASED;
1152 			vp->v_flag |= VALIASED;
1153 			vput(vp);
1154 		}
1155 		return (NULLVP);
1156 	}
1157 	simple_unlock(&spechash_slock);
1158 	VOP_UNLOCK(vp, 0);
1159 	simple_lock(&vp->v_interlock);
1160 	vclean(vp, 0, l);
1161 	vp->v_op = nvp->v_op;
1162 	vp->v_tag = nvp->v_tag;
1163 	vp->v_vnlock = &vp->v_lock;
1164 	lockinit(vp->v_vnlock, PVFS, "vnlock", 0, 0);
1165 	nvp->v_type = VNON;
1166 	insmntque(vp, mp);
1167 	return (vp);
1168 }
1169 
1170 /*
1171  * Grab a particular vnode from the free list, increment its
1172  * reference count and lock it. If the vnode lock bit is set the
1173  * vnode is being eliminated in vgone. In that case, we can not
1174  * grab the vnode, so the process is awakened when the transition is
1175  * completed, and an error returned to indicate that the vnode is no
1176  * longer usable (possibly having been changed to a new file system type).
1177  */
1178 int
1179 vget(struct vnode *vp, int flags)
1180 {
1181 	int error;
1182 
1183 	/*
1184 	 * If the vnode is in the process of being cleaned out for
1185 	 * another use, we wait for the cleaning to finish and then
1186 	 * return failure. Cleaning is determined by checking that
1187 	 * the VXLOCK flag is set.
1188 	 */
1189 
1190 	if ((flags & LK_INTERLOCK) == 0)
1191 		simple_lock(&vp->v_interlock);
1192 	if ((vp->v_flag & (VXLOCK | VFREEING)) != 0) {
1193 		if (flags & LK_NOWAIT) {
1194 			simple_unlock(&vp->v_interlock);
1195 			return EBUSY;
1196 		}
1197 		vp->v_flag |= VXWANT;
1198 		ltsleep(vp, PINOD|PNORELOCK, "vget", 0, &vp->v_interlock);
1199 		return (ENOENT);
1200 	}
1201 	if (vp->v_usecount == 0) {
1202 		simple_lock(&vnode_free_list_slock);
1203 		if (vp->v_holdcnt > 0)
1204 			TAILQ_REMOVE(&vnode_hold_list, vp, v_freelist);
1205 		else
1206 			TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
1207 		simple_unlock(&vnode_free_list_slock);
1208 	}
1209 	vp->v_usecount++;
1210 #ifdef DIAGNOSTIC
1211 	if (vp->v_usecount == 0) {
1212 		vprint("vget", vp);
1213 		panic("vget: usecount overflow, vp %p", vp);
1214 	}
1215 #endif
1216 	if (flags & LK_TYPE_MASK) {
1217 		if ((error = vn_lock(vp, flags | LK_INTERLOCK))) {
1218 			vrele(vp);
1219 		}
1220 		return (error);
1221 	}
1222 	simple_unlock(&vp->v_interlock);
1223 	return (0);
1224 }
1225 
1226 /*
1227  * vput(), just unlock and vrele()
1228  */
1229 void
1230 vput(struct vnode *vp)
1231 {
1232 	struct lwp *l = curlwp;		/* XXX */
1233 
1234 #ifdef DIAGNOSTIC
1235 	if (vp == NULL)
1236 		panic("vput: null vp");
1237 #endif
1238 	simple_lock(&vp->v_interlock);
1239 	vp->v_usecount--;
1240 	if (vp->v_usecount > 0) {
1241 		simple_unlock(&vp->v_interlock);
1242 		VOP_UNLOCK(vp, 0);
1243 		return;
1244 	}
1245 #ifdef DIAGNOSTIC
1246 	if (vp->v_usecount < 0 || vp->v_writecount != 0) {
1247 		vprint("vput: bad ref count", vp);
1248 		panic("vput: ref cnt");
1249 	}
1250 #endif
1251 	/*
1252 	 * Insert at tail of LRU list.
1253 	 */
1254 	simple_lock(&vnode_free_list_slock);
1255 	if (vp->v_holdcnt > 0)
1256 		TAILQ_INSERT_TAIL(&vnode_hold_list, vp, v_freelist);
1257 	else
1258 		TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
1259 	simple_unlock(&vnode_free_list_slock);
1260 	if (vp->v_flag & VEXECMAP) {
1261 		uvmexp.execpages -= vp->v_uobj.uo_npages;
1262 		uvmexp.filepages += vp->v_uobj.uo_npages;
1263 	}
1264 	vp->v_flag &= ~(VTEXT|VEXECMAP|VWRITEMAP|VMAPPED);
1265 	simple_unlock(&vp->v_interlock);
1266 	VOP_INACTIVE(vp, l);
1267 }
1268 
1269 /*
1270  * Vnode release.
1271  * If count drops to zero, call inactive routine and return to freelist.
1272  */
1273 void
1274 vrele(struct vnode *vp)
1275 {
1276 	struct lwp *l = curlwp;		/* XXX */
1277 
1278 #ifdef DIAGNOSTIC
1279 	if (vp == NULL)
1280 		panic("vrele: null vp");
1281 #endif
1282 	simple_lock(&vp->v_interlock);
1283 	vp->v_usecount--;
1284 	if (vp->v_usecount > 0) {
1285 		simple_unlock(&vp->v_interlock);
1286 		return;
1287 	}
1288 #ifdef DIAGNOSTIC
1289 	if (vp->v_usecount < 0 || vp->v_writecount != 0) {
1290 		vprint("vrele: bad ref count", vp);
1291 		panic("vrele: ref cnt vp %p", vp);
1292 	}
1293 #endif
1294 	/*
1295 	 * Insert at tail of LRU list.
1296 	 */
1297 	simple_lock(&vnode_free_list_slock);
1298 	if (vp->v_holdcnt > 0)
1299 		TAILQ_INSERT_TAIL(&vnode_hold_list, vp, v_freelist);
1300 	else
1301 		TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
1302 	simple_unlock(&vnode_free_list_slock);
1303 	if (vp->v_flag & VEXECMAP) {
1304 		uvmexp.execpages -= vp->v_uobj.uo_npages;
1305 		uvmexp.filepages += vp->v_uobj.uo_npages;
1306 	}
1307 	vp->v_flag &= ~(VTEXT|VEXECMAP|VWRITEMAP|VMAPPED);
1308 	if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK) == 0)
1309 		VOP_INACTIVE(vp, l);
1310 }
1311 
1312 /*
1313  * Page or buffer structure gets a reference.
1314  * Called with v_interlock held.
1315  */
1316 void
1317 vholdl(struct vnode *vp)
1318 {
1319 
1320 	/*
1321 	 * If it is on the freelist and the hold count is currently
1322 	 * zero, move it to the hold list. The test of the back
1323 	 * pointer and the use reference count of zero is because
1324 	 * it will be removed from a free list by getnewvnode,
1325 	 * but will not have its reference count incremented until
1326 	 * after calling vgone. If the reference count were
1327 	 * incremented first, vgone would (incorrectly) try to
1328 	 * close the previous instance of the underlying object.
1329 	 * So, the back pointer is explicitly set to `0xdeadb' in
1330 	 * getnewvnode after removing it from a freelist to ensure
1331 	 * that we do not try to move it here.
1332 	 */
1333 	if ((vp->v_freelist.tqe_prev != (struct vnode **)0xdeadb) &&
1334 	    vp->v_holdcnt == 0 && vp->v_usecount == 0) {
1335 		simple_lock(&vnode_free_list_slock);
1336 		TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
1337 		TAILQ_INSERT_TAIL(&vnode_hold_list, vp, v_freelist);
1338 		simple_unlock(&vnode_free_list_slock);
1339 	}
1340 	vp->v_holdcnt++;
1341 }
1342 
1343 /*
1344  * Page or buffer structure frees a reference.
1345  * Called with v_interlock held.
1346  */
1347 void
1348 holdrelel(struct vnode *vp)
1349 {
1350 
1351 	if (vp->v_holdcnt <= 0)
1352 		panic("holdrelel: holdcnt vp %p", vp);
1353 	vp->v_holdcnt--;
1354 
1355 	/*
1356 	 * If it is on the holdlist and the hold count drops to
1357 	 * zero, move it to the free list. The test of the back
1358 	 * pointer and the use reference count of zero is because
1359 	 * it will be removed from a free list by getnewvnode,
1360 	 * but will not have its reference count incremented until
1361 	 * after calling vgone. If the reference count were
1362 	 * incremented first, vgone would (incorrectly) try to
1363 	 * close the previous instance of the underlying object.
1364 	 * So, the back pointer is explicitly set to `0xdeadb' in
1365 	 * getnewvnode after removing it from a freelist to ensure
1366 	 * that we do not try to move it here.
1367 	 */
1368 
1369 	if ((vp->v_freelist.tqe_prev != (struct vnode **)0xdeadb) &&
1370 	    vp->v_holdcnt == 0 && vp->v_usecount == 0) {
1371 		simple_lock(&vnode_free_list_slock);
1372 		TAILQ_REMOVE(&vnode_hold_list, vp, v_freelist);
1373 		TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
1374 		simple_unlock(&vnode_free_list_slock);
1375 	}
1376 }
1377 
1378 /*
1379  * Vnode reference.
1380  */
1381 void
1382 vref(struct vnode *vp)
1383 {
1384 
1385 	simple_lock(&vp->v_interlock);
1386 	if (vp->v_usecount <= 0)
1387 		panic("vref used where vget required, vp %p", vp);
1388 	vp->v_usecount++;
1389 #ifdef DIAGNOSTIC
1390 	if (vp->v_usecount == 0) {
1391 		vprint("vref", vp);
1392 		panic("vref: usecount overflow, vp %p", vp);
1393 	}
1394 #endif
1395 	simple_unlock(&vp->v_interlock);
1396 }
1397 
1398 /*
1399  * Remove any vnodes in the vnode table belonging to mount point mp.
1400  *
1401  * If FORCECLOSE is not specified, there should not be any active ones,
1402  * return error if any are found (nb: this is a user error, not a
1403  * system error). If FORCECLOSE is specified, detach any active vnodes
1404  * that are found.
1405  *
1406  * If WRITECLOSE is set, only flush out regular file vnodes open for
1407  * writing.
1408  *
1409  * SKIPSYSTEM causes any vnodes marked V_SYSTEM to be skipped.
1410  */
1411 #ifdef DEBUG
1412 int busyprt = 0;	/* print out busy vnodes */
1413 struct ctldebug debug1 = { "busyprt", &busyprt };
1414 #endif
1415 
1416 int
1417 vflush(struct mount *mp, struct vnode *skipvp, int flags)
1418 {
1419 	struct lwp *l = curlwp;		/* XXX */
1420 	struct vnode *vp, *nvp;
1421 	int busy = 0;
1422 
1423 	simple_lock(&mntvnode_slock);
1424 loop:
1425 	/*
1426 	 * NOTE: not using the TAILQ_FOREACH here since in this loop vgone()
1427 	 * and vclean() are called
1428 	 */
1429 	for (vp = TAILQ_FIRST(&mp->mnt_vnodelist); vp; vp = nvp) {
1430 		if (vp->v_mount != mp)
1431 			goto loop;
1432 		nvp = TAILQ_NEXT(vp, v_mntvnodes);
1433 		/*
1434 		 * Skip over a selected vnode.
1435 		 */
1436 		if (vp == skipvp)
1437 			continue;
1438 		simple_lock(&vp->v_interlock);
1439 		/*
1440 		 * Skip over a vnodes marked VSYSTEM.
1441 		 */
1442 		if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) {
1443 			simple_unlock(&vp->v_interlock);
1444 			continue;
1445 		}
1446 		/*
1447 		 * If WRITECLOSE is set, only flush out regular file
1448 		 * vnodes open for writing.
1449 		 */
1450 		if ((flags & WRITECLOSE) &&
1451 		    (vp->v_writecount == 0 || vp->v_type != VREG)) {
1452 			simple_unlock(&vp->v_interlock);
1453 			continue;
1454 		}
1455 		/*
1456 		 * With v_usecount == 0, all we need to do is clear
1457 		 * out the vnode data structures and we are done.
1458 		 */
1459 		if (vp->v_usecount == 0) {
1460 			simple_unlock(&mntvnode_slock);
1461 			vgonel(vp, l);
1462 			simple_lock(&mntvnode_slock);
1463 			continue;
1464 		}
1465 		/*
1466 		 * If FORCECLOSE is set, forcibly close the vnode.
1467 		 * For block or character devices, revert to an
1468 		 * anonymous device. For all other files, just kill them.
1469 		 */
1470 		if (flags & FORCECLOSE) {
1471 			simple_unlock(&mntvnode_slock);
1472 			if (vp->v_type != VBLK && vp->v_type != VCHR) {
1473 				vgonel(vp, l);
1474 			} else {
1475 				vclean(vp, 0, l);
1476 				vp->v_op = spec_vnodeop_p;
1477 				insmntque(vp, (struct mount *)0);
1478 			}
1479 			simple_lock(&mntvnode_slock);
1480 			continue;
1481 		}
1482 #ifdef DEBUG
1483 		if (busyprt)
1484 			vprint("vflush: busy vnode", vp);
1485 #endif
1486 		simple_unlock(&vp->v_interlock);
1487 		busy++;
1488 	}
1489 	simple_unlock(&mntvnode_slock);
1490 	if (busy)
1491 		return (EBUSY);
1492 	return (0);
1493 }
1494 
1495 /*
1496  * Disassociate the underlying file system from a vnode.
1497  */
1498 static void
1499 vclean(struct vnode *vp, int flags, struct lwp *l)
1500 {
1501 	struct mount *mp;
1502 	int active;
1503 
1504 	LOCK_ASSERT(simple_lock_held(&vp->v_interlock));
1505 
1506 	/*
1507 	 * Check to see if the vnode is in use.
1508 	 * If so we have to reference it before we clean it out
1509 	 * so that its count cannot fall to zero and generate a
1510 	 * race against ourselves to recycle it.
1511 	 */
1512 
1513 	if ((active = vp->v_usecount) != 0) {
1514 		vp->v_usecount++;
1515 #ifdef DIAGNOSTIC
1516 		if (vp->v_usecount == 0) {
1517 			vprint("vclean", vp);
1518 			panic("vclean: usecount overflow");
1519 		}
1520 #endif
1521 	}
1522 
1523 	/*
1524 	 * Prevent the vnode from being recycled or
1525 	 * brought into use while we clean it out.
1526 	 */
1527 	if (vp->v_flag & VXLOCK)
1528 		panic("vclean: deadlock, vp %p", vp);
1529 	vp->v_flag |= VXLOCK;
1530 	if (vp->v_flag & VEXECMAP) {
1531 		uvmexp.execpages -= vp->v_uobj.uo_npages;
1532 		uvmexp.filepages += vp->v_uobj.uo_npages;
1533 	}
1534 	vp->v_flag &= ~(VTEXT|VEXECMAP);
1535 
1536 	/*
1537 	 * Even if the count is zero, the VOP_INACTIVE routine may still
1538 	 * have the object locked while it cleans it out. The VOP_LOCK
1539 	 * ensures that the VOP_INACTIVE routine is done with its work.
1540 	 * For active vnodes, it ensures that no other activity can
1541 	 * occur while the underlying object is being cleaned out.
1542 	 */
1543 	VOP_LOCK(vp, LK_DRAIN | LK_INTERLOCK);
1544 
1545 	/*
1546 	 * Clean out any cached data associated with the vnode.
1547 	 * If special device, remove it from special device alias list.
1548 	 * if it is on one.
1549 	 */
1550 	if (flags & DOCLOSE) {
1551 		int error;
1552 		struct vnode *vq, *vx;
1553 
1554 		vn_start_write(vp, &mp, V_WAIT | V_LOWER);
1555 		error = vinvalbuf(vp, V_SAVE, NOCRED, l, 0, 0);
1556 		vn_finished_write(mp, V_LOWER);
1557 		if (error)
1558 			error = vinvalbuf(vp, 0, NOCRED, l, 0, 0);
1559 		KASSERT(error == 0);
1560 		KASSERT((vp->v_flag & VONWORKLST) == 0);
1561 
1562 		if (active)
1563 			VOP_CLOSE(vp, FNONBLOCK, NOCRED, NULL);
1564 
1565 		if ((vp->v_type == VBLK || vp->v_type == VCHR) &&
1566 		    vp->v_specinfo != 0) {
1567 			simple_lock(&spechash_slock);
1568 			if (vp->v_hashchain != NULL) {
1569 				if (*vp->v_hashchain == vp) {
1570 					*vp->v_hashchain = vp->v_specnext;
1571 				} else {
1572 					for (vq = *vp->v_hashchain; vq;
1573 					     vq = vq->v_specnext) {
1574 						if (vq->v_specnext != vp)
1575 							continue;
1576 						vq->v_specnext = vp->v_specnext;
1577 						break;
1578 					}
1579 					if (vq == NULL)
1580 						panic("missing bdev");
1581 				}
1582 				if (vp->v_flag & VALIASED) {
1583 					vx = NULL;
1584 						for (vq = *vp->v_hashchain; vq;
1585 						     vq = vq->v_specnext) {
1586 						if (vq->v_rdev != vp->v_rdev ||
1587 						    vq->v_type != vp->v_type)
1588 							continue;
1589 						if (vx)
1590 							break;
1591 						vx = vq;
1592 					}
1593 					if (vx == NULL)
1594 						panic("missing alias");
1595 					if (vq == NULL)
1596 						vx->v_flag &= ~VALIASED;
1597 					vp->v_flag &= ~VALIASED;
1598 				}
1599 			}
1600 			simple_unlock(&spechash_slock);
1601 			FREE(vp->v_specinfo, M_VNODE);
1602 			vp->v_specinfo = NULL;
1603 		}
1604 	}
1605 	LOCK_ASSERT(!simple_lock_held(&vp->v_interlock));
1606 
1607 	/*
1608 	 * If purging an active vnode, it must be closed and
1609 	 * deactivated before being reclaimed. Note that the
1610 	 * VOP_INACTIVE will unlock the vnode.
1611 	 */
1612 	if (active) {
1613 		VOP_INACTIVE(vp, l);
1614 	} else {
1615 		/*
1616 		 * Any other processes trying to obtain this lock must first
1617 		 * wait for VXLOCK to clear, then call the new lock operation.
1618 		 */
1619 		VOP_UNLOCK(vp, 0);
1620 	}
1621 	/*
1622 	 * Reclaim the vnode.
1623 	 */
1624 	if (VOP_RECLAIM(vp, l))
1625 		panic("vclean: cannot reclaim, vp %p", vp);
1626 	if (active) {
1627 		/*
1628 		 * Inline copy of vrele() since VOP_INACTIVE
1629 		 * has already been called.
1630 		 */
1631 		simple_lock(&vp->v_interlock);
1632 		if (--vp->v_usecount <= 0) {
1633 #ifdef DIAGNOSTIC
1634 			if (vp->v_usecount < 0 || vp->v_writecount != 0) {
1635 				vprint("vclean: bad ref count", vp);
1636 				panic("vclean: ref cnt");
1637 			}
1638 #endif
1639 			/*
1640 			 * Insert at tail of LRU list.
1641 			 */
1642 
1643 			simple_unlock(&vp->v_interlock);
1644 			simple_lock(&vnode_free_list_slock);
1645 #ifdef DIAGNOSTIC
1646 			if (vp->v_holdcnt > 0)
1647 				panic("vclean: not clean, vp %p", vp);
1648 #endif
1649 			TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
1650 			simple_unlock(&vnode_free_list_slock);
1651 		} else
1652 			simple_unlock(&vp->v_interlock);
1653 	}
1654 
1655 	KASSERT(vp->v_uobj.uo_npages == 0);
1656 	if (vp->v_type == VREG && vp->v_ractx != NULL) {
1657 		uvm_ra_freectx(vp->v_ractx);
1658 		vp->v_ractx = NULL;
1659 	}
1660 	cache_purge(vp);
1661 
1662 	/*
1663 	 * Done with purge, notify sleepers of the grim news.
1664 	 */
1665 	vp->v_op = dead_vnodeop_p;
1666 	vp->v_tag = VT_NON;
1667 	simple_lock(&vp->v_interlock);
1668 	VN_KNOTE(vp, NOTE_REVOKE);	/* FreeBSD has this in vn_pollgone() */
1669 	vp->v_flag &= ~(VXLOCK|VLOCKSWORK);
1670 	if (vp->v_flag & VXWANT) {
1671 		vp->v_flag &= ~VXWANT;
1672 		simple_unlock(&vp->v_interlock);
1673 		wakeup((caddr_t)vp);
1674 	} else
1675 		simple_unlock(&vp->v_interlock);
1676 }
1677 
1678 /*
1679  * Recycle an unused vnode to the front of the free list.
1680  * Release the passed interlock if the vnode will be recycled.
1681  */
1682 int
1683 vrecycle(struct vnode *vp, struct simplelock *inter_lkp, struct lwp *l)
1684 {
1685 
1686 	simple_lock(&vp->v_interlock);
1687 	if (vp->v_usecount == 0) {
1688 		if (inter_lkp)
1689 			simple_unlock(inter_lkp);
1690 		vgonel(vp, l);
1691 		return (1);
1692 	}
1693 	simple_unlock(&vp->v_interlock);
1694 	return (0);
1695 }
1696 
1697 /*
1698  * Eliminate all activity associated with a vnode
1699  * in preparation for reuse.
1700  */
1701 void
1702 vgone(struct vnode *vp)
1703 {
1704 	struct lwp *l = curlwp;		/* XXX */
1705 
1706 	simple_lock(&vp->v_interlock);
1707 	vgonel(vp, l);
1708 }
1709 
1710 /*
1711  * vgone, with the vp interlock held.
1712  */
1713 void
1714 vgonel(struct vnode *vp, struct lwp *l)
1715 {
1716 
1717 	LOCK_ASSERT(simple_lock_held(&vp->v_interlock));
1718 
1719 	/*
1720 	 * If a vgone (or vclean) is already in progress,
1721 	 * wait until it is done and return.
1722 	 */
1723 
1724 	if (vp->v_flag & VXLOCK) {
1725 		vp->v_flag |= VXWANT;
1726 		ltsleep(vp, PINOD | PNORELOCK, "vgone", 0, &vp->v_interlock);
1727 		return;
1728 	}
1729 
1730 	/*
1731 	 * Clean out the filesystem specific data.
1732 	 */
1733 
1734 	vclean(vp, DOCLOSE, l);
1735 	KASSERT((vp->v_flag & VONWORKLST) == 0);
1736 
1737 	/*
1738 	 * Delete from old mount point vnode list, if on one.
1739 	 */
1740 
1741 	if (vp->v_mount != NULL)
1742 		insmntque(vp, (struct mount *)0);
1743 
1744 	/*
1745 	 * The test of the back pointer and the reference count of
1746 	 * zero is because it will be removed from the free list by
1747 	 * getcleanvnode, but will not have its reference count
1748 	 * incremented until after calling vgone. If the reference
1749 	 * count were incremented first, vgone would (incorrectly)
1750 	 * try to close the previous instance of the underlying object.
1751 	 * So, the back pointer is explicitly set to `0xdeadb' in
1752 	 * getnewvnode after removing it from the freelist to ensure
1753 	 * that we do not try to move it here.
1754 	 */
1755 
1756 	vp->v_type = VBAD;
1757 	if (vp->v_usecount == 0) {
1758 		boolean_t dofree;
1759 
1760 		simple_lock(&vnode_free_list_slock);
1761 		if (vp->v_holdcnt > 0)
1762 			panic("vgonel: not clean, vp %p", vp);
1763 		/*
1764 		 * if it isn't on the freelist, we're called by getcleanvnode
1765 		 * and vnode is being re-used.  otherwise, we'll free it.
1766 		 */
1767 		dofree = vp->v_freelist.tqe_prev != (struct vnode **)0xdeadb;
1768 		if (dofree) {
1769 			TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
1770 			numvnodes--;
1771 		}
1772 		simple_unlock(&vnode_free_list_slock);
1773 		if (dofree)
1774 			pool_put(&vnode_pool, vp);
1775 	}
1776 }
1777 
1778 /*
1779  * Lookup a vnode by device number.
1780  */
1781 int
1782 vfinddev(dev_t dev, enum vtype type, struct vnode **vpp)
1783 {
1784 	struct vnode *vp;
1785 	int rc = 0;
1786 
1787 	simple_lock(&spechash_slock);
1788 	for (vp = speclisth[SPECHASH(dev)]; vp; vp = vp->v_specnext) {
1789 		if (dev != vp->v_rdev || type != vp->v_type)
1790 			continue;
1791 		*vpp = vp;
1792 		rc = 1;
1793 		break;
1794 	}
1795 	simple_unlock(&spechash_slock);
1796 	return (rc);
1797 }
1798 
1799 /*
1800  * Revoke all the vnodes corresponding to the specified minor number
1801  * range (endpoints inclusive) of the specified major.
1802  */
1803 void
1804 vdevgone(int maj, int minl, int minh, enum vtype type)
1805 {
1806 	struct vnode *vp;
1807 	int mn;
1808 
1809 	vp = NULL;	/* XXX gcc */
1810 
1811 	for (mn = minl; mn <= minh; mn++)
1812 		if (vfinddev(makedev(maj, mn), type, &vp))
1813 			VOP_REVOKE(vp, REVOKEALL);
1814 }
1815 
1816 /*
1817  * Calculate the total number of references to a special device.
1818  */
1819 int
1820 vcount(struct vnode *vp)
1821 {
1822 	struct vnode *vq, *vnext;
1823 	int count;
1824 
1825 loop:
1826 	if ((vp->v_flag & VALIASED) == 0)
1827 		return (vp->v_usecount);
1828 	simple_lock(&spechash_slock);
1829 	for (count = 0, vq = *vp->v_hashchain; vq; vq = vnext) {
1830 		vnext = vq->v_specnext;
1831 		if (vq->v_rdev != vp->v_rdev || vq->v_type != vp->v_type)
1832 			continue;
1833 		/*
1834 		 * Alias, but not in use, so flush it out.
1835 		 */
1836 		if (vq->v_usecount == 0 && vq != vp &&
1837 		    (vq->v_flag & VXLOCK) == 0) {
1838 			simple_unlock(&spechash_slock);
1839 			vgone(vq);
1840 			goto loop;
1841 		}
1842 		count += vq->v_usecount;
1843 	}
1844 	simple_unlock(&spechash_slock);
1845 	return (count);
1846 }
1847 
1848 #define ARRAY_SIZE(arr) (sizeof(arr) / sizeof(arr[0]))
1849 #define ARRAY_PRINT(idx, arr) \
1850     ((idx) > 0 && (idx) < ARRAY_SIZE(arr) ? (arr)[(idx)] : "UNKNOWN")
1851 
1852 const char * const vnode_tags[] = { VNODE_TAGS };
1853 const char * const vnode_types[] = { VNODE_TYPES };
1854 const char vnode_flagbits[] = VNODE_FLAGBITS;
1855 
1856 /*
1857  * Print out a description of a vnode.
1858  */
1859 void
1860 vprint(const char *label, struct vnode *vp)
1861 {
1862 	char bf[96];
1863 
1864 	if (label != NULL)
1865 		printf("%s: ", label);
1866 	printf("tag %s(%d) type %s(%d), usecount %d, writecount %ld, "
1867 	    "refcount %ld,", ARRAY_PRINT(vp->v_tag, vnode_tags), vp->v_tag,
1868 	    ARRAY_PRINT(vp->v_type, vnode_types), vp->v_type,
1869 	    vp->v_usecount, vp->v_writecount, vp->v_holdcnt);
1870 	bitmask_snprintf(vp->v_flag, vnode_flagbits, bf, sizeof(bf));
1871 	if (bf[0] != '\0')
1872 		printf(" flags (%s)", &bf[1]);
1873 	if (vp->v_data == NULL) {
1874 		printf("\n");
1875 	} else {
1876 		printf("\n\t");
1877 		VOP_PRINT(vp);
1878 	}
1879 }
1880 
1881 #ifdef DEBUG
1882 /*
1883  * List all of the locked vnodes in the system.
1884  * Called when debugging the kernel.
1885  */
1886 void
1887 printlockedvnodes(void)
1888 {
1889 	struct mount *mp, *nmp;
1890 	struct vnode *vp;
1891 
1892 	printf("Locked vnodes\n");
1893 	simple_lock(&mountlist_slock);
1894 	for (mp = CIRCLEQ_FIRST(&mountlist); mp != (void *)&mountlist;
1895 	     mp = nmp) {
1896 		if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock)) {
1897 			nmp = CIRCLEQ_NEXT(mp, mnt_list);
1898 			continue;
1899 		}
1900 		TAILQ_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
1901 			if (VOP_ISLOCKED(vp))
1902 				vprint(NULL, vp);
1903 		}
1904 		simple_lock(&mountlist_slock);
1905 		nmp = CIRCLEQ_NEXT(mp, mnt_list);
1906 		vfs_unbusy(mp);
1907 	}
1908 	simple_unlock(&mountlist_slock);
1909 }
1910 #endif
1911 
1912 /*
1913  * sysctl helper routine to return list of supported fstypes
1914  */
1915 static int
1916 sysctl_vfs_generic_fstypes(SYSCTLFN_ARGS)
1917 {
1918 	char bf[MFSNAMELEN];
1919 	char *where = oldp;
1920 	struct vfsops *v;
1921 	size_t needed, left, slen;
1922 	int error, first;
1923 
1924 	if (newp != NULL)
1925 		return (EPERM);
1926 	if (namelen != 0)
1927 		return (EINVAL);
1928 
1929 	first = 1;
1930 	error = 0;
1931 	needed = 0;
1932 	left = *oldlenp;
1933 
1934 	LIST_FOREACH(v, &vfs_list, vfs_list) {
1935 		if (where == NULL)
1936 			needed += strlen(v->vfs_name) + 1;
1937 		else {
1938 			memset(bf, 0, sizeof(bf));
1939 			if (first) {
1940 				strncpy(bf, v->vfs_name, sizeof(bf));
1941 				first = 0;
1942 			} else {
1943 				bf[0] = ' ';
1944 				strncpy(bf + 1, v->vfs_name, sizeof(bf) - 1);
1945 			}
1946 			bf[sizeof(bf)-1] = '\0';
1947 			slen = strlen(bf);
1948 			if (left < slen + 1)
1949 				break;
1950 			/* +1 to copy out the trailing NUL byte */
1951 			error = copyout(bf, where, slen + 1);
1952 			if (error)
1953 				break;
1954 			where += slen;
1955 			needed += slen;
1956 			left -= slen;
1957 		}
1958 	}
1959 	*oldlenp = needed;
1960 	return (error);
1961 }
1962 
1963 /*
1964  * Top level filesystem related information gathering.
1965  */
1966 SYSCTL_SETUP(sysctl_vfs_setup, "sysctl vfs subtree setup")
1967 {
1968 	sysctl_createv(clog, 0, NULL, NULL,
1969 		       CTLFLAG_PERMANENT,
1970 		       CTLTYPE_NODE, "vfs", NULL,
1971 		       NULL, 0, NULL, 0,
1972 		       CTL_VFS, CTL_EOL);
1973 	sysctl_createv(clog, 0, NULL, NULL,
1974 		       CTLFLAG_PERMANENT,
1975 		       CTLTYPE_NODE, "generic",
1976 		       SYSCTL_DESCR("Non-specific vfs related information"),
1977 		       NULL, 0, NULL, 0,
1978 		       CTL_VFS, VFS_GENERIC, CTL_EOL);
1979 	sysctl_createv(clog, 0, NULL, NULL,
1980 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1981 		       CTLTYPE_INT, "usermount",
1982 		       SYSCTL_DESCR("Whether unprivileged users may mount "
1983 				    "filesystems"),
1984 		       NULL, 0, &dovfsusermount, 0,
1985 		       CTL_VFS, VFS_GENERIC, VFS_USERMOUNT, CTL_EOL);
1986 	sysctl_createv(clog, 0, NULL, NULL,
1987 		       CTLFLAG_PERMANENT,
1988 		       CTLTYPE_STRING, "fstypes",
1989 		       SYSCTL_DESCR("List of file systems present"),
1990 		       sysctl_vfs_generic_fstypes, 0, NULL, 0,
1991 		       CTL_VFS, VFS_GENERIC, CTL_CREATE, CTL_EOL);
1992 	sysctl_createv(clog, 0, NULL, NULL,
1993 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1994 		       CTLTYPE_INT, "magiclinks",
1995 		       SYSCTL_DESCR("Whether \"magic\" symlinks are expanded"),
1996 		       NULL, 0, &vfs_magiclinks, 0,
1997 		       CTL_VFS, VFS_GENERIC, VFS_MAGICLINKS, CTL_EOL);
1998 }
1999 
2000 
2001 int kinfo_vdebug = 1;
2002 int kinfo_vgetfailed;
2003 #define KINFO_VNODESLOP	10
2004 /*
2005  * Dump vnode list (via sysctl).
2006  * Copyout address of vnode followed by vnode.
2007  */
2008 /* ARGSUSED */
2009 int
2010 sysctl_kern_vnode(SYSCTLFN_ARGS)
2011 {
2012 	char *where = oldp;
2013 	size_t *sizep = oldlenp;
2014 	struct mount *mp, *nmp;
2015 	struct vnode *vp;
2016 	char *bp = where, *savebp;
2017 	char *ewhere;
2018 	int error;
2019 
2020 	if (namelen != 0)
2021 		return (EOPNOTSUPP);
2022 	if (newp != NULL)
2023 		return (EPERM);
2024 
2025 #define VPTRSZ	sizeof(struct vnode *)
2026 #define VNODESZ	sizeof(struct vnode)
2027 	if (where == NULL) {
2028 		*sizep = (numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ);
2029 		return (0);
2030 	}
2031 	ewhere = where + *sizep;
2032 
2033 	simple_lock(&mountlist_slock);
2034 	for (mp = CIRCLEQ_FIRST(&mountlist); mp != (void *)&mountlist;
2035 	     mp = nmp) {
2036 		if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock)) {
2037 			nmp = CIRCLEQ_NEXT(mp, mnt_list);
2038 			continue;
2039 		}
2040 		savebp = bp;
2041 again:
2042 		simple_lock(&mntvnode_slock);
2043 		TAILQ_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
2044 			/*
2045 			 * Check that the vp is still associated with
2046 			 * this filesystem.  RACE: could have been
2047 			 * recycled onto the same filesystem.
2048 			 */
2049 			if (vp->v_mount != mp) {
2050 				simple_unlock(&mntvnode_slock);
2051 				if (kinfo_vdebug)
2052 					printf("kinfo: vp changed\n");
2053 				bp = savebp;
2054 				goto again;
2055 			}
2056 			if (bp + VPTRSZ + VNODESZ > ewhere) {
2057 				simple_unlock(&mntvnode_slock);
2058 				*sizep = bp - where;
2059 				return (ENOMEM);
2060 			}
2061 			simple_unlock(&mntvnode_slock);
2062 			if ((error = copyout((caddr_t)&vp, bp, VPTRSZ)) ||
2063 			   (error = copyout((caddr_t)vp, bp + VPTRSZ, VNODESZ)))
2064 				return (error);
2065 			bp += VPTRSZ + VNODESZ;
2066 			simple_lock(&mntvnode_slock);
2067 		}
2068 		simple_unlock(&mntvnode_slock);
2069 		simple_lock(&mountlist_slock);
2070 		nmp = CIRCLEQ_NEXT(mp, mnt_list);
2071 		vfs_unbusy(mp);
2072 	}
2073 	simple_unlock(&mountlist_slock);
2074 
2075 	*sizep = bp - where;
2076 	return (0);
2077 }
2078 
2079 /*
2080  * Check to see if a filesystem is mounted on a block device.
2081  */
2082 int
2083 vfs_mountedon(struct vnode *vp)
2084 {
2085 	struct vnode *vq;
2086 	int error = 0;
2087 
2088 	if (vp->v_type != VBLK)
2089 		return ENOTBLK;
2090 	if (vp->v_specmountpoint != NULL)
2091 		return (EBUSY);
2092 	if (vp->v_flag & VALIASED) {
2093 		simple_lock(&spechash_slock);
2094 		for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) {
2095 			if (vq->v_rdev != vp->v_rdev ||
2096 			    vq->v_type != vp->v_type)
2097 				continue;
2098 			if (vq->v_specmountpoint != NULL) {
2099 				error = EBUSY;
2100 				break;
2101 			}
2102 		}
2103 		simple_unlock(&spechash_slock);
2104 	}
2105 	return (error);
2106 }
2107 
2108 /*
2109  * Do the usual access checking.
2110  * file_mode, uid and gid are from the vnode in question,
2111  * while acc_mode and cred are from the VOP_ACCESS parameter list
2112  */
2113 int
2114 vaccess(enum vtype type, mode_t file_mode, uid_t uid, gid_t gid,
2115     mode_t acc_mode, kauth_cred_t cred)
2116 {
2117 	mode_t mask;
2118 	int error, ismember;
2119 
2120 	/*
2121 	 * Super-user always gets read/write access, but execute access depends
2122 	 * on at least one execute bit being set.
2123 	 */
2124 	if (kauth_cred_geteuid(cred) == 0) {
2125 		if ((acc_mode & VEXEC) && type != VDIR &&
2126 		    (file_mode & (S_IXUSR|S_IXGRP|S_IXOTH)) == 0)
2127 			return (EACCES);
2128 		return (0);
2129 	}
2130 
2131 	mask = 0;
2132 
2133 	/* Otherwise, check the owner. */
2134 	if (kauth_cred_geteuid(cred) == uid) {
2135 		if (acc_mode & VEXEC)
2136 			mask |= S_IXUSR;
2137 		if (acc_mode & VREAD)
2138 			mask |= S_IRUSR;
2139 		if (acc_mode & VWRITE)
2140 			mask |= S_IWUSR;
2141 		return ((file_mode & mask) == mask ? 0 : EACCES);
2142 	}
2143 
2144 	/* Otherwise, check the groups. */
2145 	error = kauth_cred_ismember_gid(cred, gid, &ismember);
2146 	if (error)
2147 		return (error);
2148 	if (kauth_cred_getegid(cred) == gid || ismember) {
2149 		if (acc_mode & VEXEC)
2150 			mask |= S_IXGRP;
2151 		if (acc_mode & VREAD)
2152 			mask |= S_IRGRP;
2153 		if (acc_mode & VWRITE)
2154 			mask |= S_IWGRP;
2155 		return ((file_mode & mask) == mask ? 0 : EACCES);
2156 	}
2157 
2158 	/* Otherwise, check everyone else. */
2159 	if (acc_mode & VEXEC)
2160 		mask |= S_IXOTH;
2161 	if (acc_mode & VREAD)
2162 		mask |= S_IROTH;
2163 	if (acc_mode & VWRITE)
2164 		mask |= S_IWOTH;
2165 	return ((file_mode & mask) == mask ? 0 : EACCES);
2166 }
2167 
2168 /*
2169  * Unmount all file systems.
2170  * We traverse the list in reverse order under the assumption that doing so
2171  * will avoid needing to worry about dependencies.
2172  */
2173 void
2174 vfs_unmountall(struct lwp *l)
2175 {
2176 	struct mount *mp, *nmp;
2177 	int allerror, error;
2178 
2179 	printf("unmounting file systems...");
2180 	for (allerror = 0,
2181 	     mp = mountlist.cqh_last; mp != (void *)&mountlist; mp = nmp) {
2182 		nmp = mp->mnt_list.cqe_prev;
2183 #ifdef DEBUG
2184 		printf("\nunmounting %s (%s)...",
2185 		    mp->mnt_stat.f_mntonname, mp->mnt_stat.f_mntfromname);
2186 #endif
2187 		/*
2188 		 * XXX Freeze syncer.  Must do this before locking the
2189 		 * mount point.  See dounmount() for details.
2190 		 */
2191 		lockmgr(&syncer_lock, LK_EXCLUSIVE, NULL);
2192 		if (vfs_busy(mp, 0, 0)) {
2193 			lockmgr(&syncer_lock, LK_RELEASE, NULL);
2194 			continue;
2195 		}
2196 		if ((error = dounmount(mp, MNT_FORCE, l)) != 0) {
2197 			printf("unmount of %s failed with error %d\n",
2198 			    mp->mnt_stat.f_mntonname, error);
2199 			allerror = 1;
2200 		}
2201 	}
2202 	printf(" done\n");
2203 	if (allerror)
2204 		printf("WARNING: some file systems would not unmount\n");
2205 }
2206 
2207 extern struct simplelock bqueue_slock; /* XXX */
2208 
2209 /*
2210  * Sync and unmount file systems before shutting down.
2211  */
2212 void
2213 vfs_shutdown(void)
2214 {
2215 	struct lwp *l;
2216 
2217 	/* XXX we're certainly not running in lwp0's context! */
2218 	l = curlwp;
2219 	if (l == NULL)
2220 		l = &lwp0;
2221 
2222 	printf("syncing disks... ");
2223 
2224 	/* remove user process from run queue */
2225 	suspendsched();
2226 	(void) spl0();
2227 
2228 	/* avoid coming back this way again if we panic. */
2229 	doing_shutdown = 1;
2230 
2231 	sys_sync(l, NULL, NULL);
2232 
2233 	/* Wait for sync to finish. */
2234 	if (buf_syncwait() != 0) {
2235 #if defined(DDB) && defined(DEBUG_HALT_BUSY)
2236 		Debugger();
2237 #endif
2238 		printf("giving up\n");
2239 		return;
2240 	} else
2241 		printf("done\n");
2242 
2243 	/*
2244 	 * If we've panic'd, don't make the situation potentially
2245 	 * worse by unmounting the file systems.
2246 	 */
2247 	if (panicstr != NULL)
2248 		return;
2249 
2250 	/* Release inodes held by texts before update. */
2251 #ifdef notdef
2252 	vnshutdown();
2253 #endif
2254 	/* Unmount file systems. */
2255 	vfs_unmountall(l);
2256 }
2257 
2258 /*
2259  * Mount the root file system.  If the operator didn't specify a
2260  * file system to use, try all possible file systems until one
2261  * succeeds.
2262  */
2263 int
2264 vfs_mountroot(void)
2265 {
2266 	struct vfsops *v;
2267 	int error = ENODEV;
2268 
2269 	if (root_device == NULL)
2270 		panic("vfs_mountroot: root device unknown");
2271 
2272 	switch (device_class(root_device)) {
2273 	case DV_IFNET:
2274 		if (rootdev != NODEV)
2275 			panic("vfs_mountroot: rootdev set for DV_IFNET "
2276 			    "(0x%08x -> %d,%d)", rootdev,
2277 			    major(rootdev), minor(rootdev));
2278 		break;
2279 
2280 	case DV_DISK:
2281 		if (rootdev == NODEV)
2282 			panic("vfs_mountroot: rootdev not set for DV_DISK");
2283 	        if (bdevvp(rootdev, &rootvp))
2284 	                panic("vfs_mountroot: can't get vnode for rootdev");
2285 		error = VOP_OPEN(rootvp, FREAD, FSCRED, curlwp);
2286 		if (error) {
2287 			printf("vfs_mountroot: can't open root device\n");
2288 			return (error);
2289 		}
2290 		break;
2291 
2292 	default:
2293 		printf("%s: inappropriate for root file system\n",
2294 		    root_device->dv_xname);
2295 		return (ENODEV);
2296 	}
2297 
2298 	/*
2299 	 * If user specified a file system, use it.
2300 	 */
2301 	if (mountroot != NULL) {
2302 		error = (*mountroot)();
2303 		goto done;
2304 	}
2305 
2306 	/*
2307 	 * Try each file system currently configured into the kernel.
2308 	 */
2309 	LIST_FOREACH(v, &vfs_list, vfs_list) {
2310 		if (v->vfs_mountroot == NULL)
2311 			continue;
2312 #ifdef DEBUG
2313 		aprint_normal("mountroot: trying %s...\n", v->vfs_name);
2314 #endif
2315 		error = (*v->vfs_mountroot)();
2316 		if (!error) {
2317 			aprint_normal("root file system type: %s\n",
2318 			    v->vfs_name);
2319 			break;
2320 		}
2321 	}
2322 
2323 	if (v == NULL) {
2324 		printf("no file system for %s", root_device->dv_xname);
2325 		if (device_class(root_device) == DV_DISK)
2326 			printf(" (dev 0x%x)", rootdev);
2327 		printf("\n");
2328 		error = EFTYPE;
2329 	}
2330 
2331 done:
2332 	if (error && device_class(root_device) == DV_DISK) {
2333 		VOP_CLOSE(rootvp, FREAD, FSCRED, curlwp);
2334 		vrele(rootvp);
2335 	}
2336 	return (error);
2337 }
2338 
2339 /*
2340  * Given a file system name, look up the vfsops for that
2341  * file system, or return NULL if file system isn't present
2342  * in the kernel.
2343  */
2344 struct vfsops *
2345 vfs_getopsbyname(const char *name)
2346 {
2347 	struct vfsops *v;
2348 
2349 	LIST_FOREACH(v, &vfs_list, vfs_list) {
2350 		if (strcmp(v->vfs_name, name) == 0)
2351 			break;
2352 	}
2353 
2354 	return (v);
2355 }
2356 
2357 /*
2358  * Establish a file system and initialize it.
2359  */
2360 int
2361 vfs_attach(struct vfsops *vfs)
2362 {
2363 	struct vfsops *v;
2364 	int error = 0;
2365 
2366 
2367 	/*
2368 	 * Make sure this file system doesn't already exist.
2369 	 */
2370 	LIST_FOREACH(v, &vfs_list, vfs_list) {
2371 		if (strcmp(vfs->vfs_name, v->vfs_name) == 0) {
2372 			error = EEXIST;
2373 			goto out;
2374 		}
2375 	}
2376 
2377 	/*
2378 	 * Initialize the vnode operations for this file system.
2379 	 */
2380 	vfs_opv_init(vfs->vfs_opv_descs);
2381 
2382 	/*
2383 	 * Now initialize the file system itself.
2384 	 */
2385 	(*vfs->vfs_init)();
2386 
2387 	/*
2388 	 * ...and link it into the kernel's list.
2389 	 */
2390 	LIST_INSERT_HEAD(&vfs_list, vfs, vfs_list);
2391 
2392 	/*
2393 	 * Sanity: make sure the reference count is 0.
2394 	 */
2395 	vfs->vfs_refcount = 0;
2396 
2397  out:
2398 	return (error);
2399 }
2400 
2401 /*
2402  * Remove a file system from the kernel.
2403  */
2404 int
2405 vfs_detach(struct vfsops *vfs)
2406 {
2407 	struct vfsops *v;
2408 
2409 	/*
2410 	 * Make sure no one is using the filesystem.
2411 	 */
2412 	if (vfs->vfs_refcount != 0)
2413 		return (EBUSY);
2414 
2415 	/*
2416 	 * ...and remove it from the kernel's list.
2417 	 */
2418 	LIST_FOREACH(v, &vfs_list, vfs_list) {
2419 		if (v == vfs) {
2420 			LIST_REMOVE(v, vfs_list);
2421 			break;
2422 		}
2423 	}
2424 
2425 	if (v == NULL)
2426 		return (ESRCH);
2427 
2428 	/*
2429 	 * Now run the file system-specific cleanups.
2430 	 */
2431 	(*vfs->vfs_done)();
2432 
2433 	/*
2434 	 * Free the vnode operations vector.
2435 	 */
2436 	vfs_opv_free(vfs->vfs_opv_descs);
2437 	return (0);
2438 }
2439 
2440 void
2441 vfs_reinit(void)
2442 {
2443 	struct vfsops *vfs;
2444 
2445 	LIST_FOREACH(vfs, &vfs_list, vfs_list) {
2446 		if (vfs->vfs_reinit) {
2447 			(*vfs->vfs_reinit)();
2448 		}
2449 	}
2450 }
2451 
2452 /*
2453  * Request a filesystem to suspend write operations.
2454  */
2455 int
2456 vfs_write_suspend(struct mount *mp, int slpflag, int slptimeo)
2457 {
2458 	struct lwp *l = curlwp;	/* XXX */
2459 	int error;
2460 
2461 	while ((mp->mnt_iflag & IMNT_SUSPEND)) {
2462 		if (slptimeo < 0)
2463 			return EWOULDBLOCK;
2464 		error = tsleep(&mp->mnt_flag, slpflag, "suspwt1", slptimeo);
2465 		if (error)
2466 			return error;
2467 	}
2468 	mp->mnt_iflag |= IMNT_SUSPEND;
2469 
2470 	simple_lock(&mp->mnt_slock);
2471 	if (mp->mnt_writeopcountupper > 0)
2472 		ltsleep(&mp->mnt_writeopcountupper, PUSER - 1, "suspwt",
2473 			0, &mp->mnt_slock);
2474 	simple_unlock(&mp->mnt_slock);
2475 
2476 	error = VFS_SYNC(mp, MNT_WAIT, l->l_cred, l);
2477 	if (error) {
2478 		vfs_write_resume(mp);
2479 		return error;
2480 	}
2481 	mp->mnt_iflag |= IMNT_SUSPENDLOW;
2482 
2483 	simple_lock(&mp->mnt_slock);
2484 	if (mp->mnt_writeopcountlower > 0)
2485 		ltsleep(&mp->mnt_writeopcountlower, PUSER - 1, "suspwt",
2486 			0, &mp->mnt_slock);
2487 	mp->mnt_iflag |= IMNT_SUSPENDED;
2488 	simple_unlock(&mp->mnt_slock);
2489 
2490 	return 0;
2491 }
2492 
2493 /*
2494  * Request a filesystem to resume write operations.
2495  */
2496 void
2497 vfs_write_resume(struct mount *mp)
2498 {
2499 
2500 	if ((mp->mnt_iflag & IMNT_SUSPEND) == 0)
2501 		return;
2502 	mp->mnt_iflag &= ~(IMNT_SUSPEND | IMNT_SUSPENDLOW | IMNT_SUSPENDED);
2503 	wakeup(&mp->mnt_flag);
2504 }
2505 
2506 void
2507 copy_statvfs_info(struct statvfs *sbp, const struct mount *mp)
2508 {
2509 	const struct statvfs *mbp;
2510 
2511 	if (sbp == (mbp = &mp->mnt_stat))
2512 		return;
2513 
2514 	(void)memcpy(&sbp->f_fsidx, &mbp->f_fsidx, sizeof(sbp->f_fsidx));
2515 	sbp->f_fsid = mbp->f_fsid;
2516 	sbp->f_owner = mbp->f_owner;
2517 	sbp->f_flag = mbp->f_flag;
2518 	sbp->f_syncwrites = mbp->f_syncwrites;
2519 	sbp->f_asyncwrites = mbp->f_asyncwrites;
2520 	sbp->f_syncreads = mbp->f_syncreads;
2521 	sbp->f_asyncreads = mbp->f_asyncreads;
2522 	(void)memcpy(sbp->f_spare, mbp->f_spare, sizeof(mbp->f_spare));
2523 	(void)memcpy(sbp->f_fstypename, mbp->f_fstypename,
2524 	    sizeof(sbp->f_fstypename));
2525 	(void)memcpy(sbp->f_mntonname, mbp->f_mntonname,
2526 	    sizeof(sbp->f_mntonname));
2527 	(void)memcpy(sbp->f_mntfromname, mp->mnt_stat.f_mntfromname,
2528 	    sizeof(sbp->f_mntfromname));
2529 	sbp->f_namemax = mbp->f_namemax;
2530 }
2531 
2532 int
2533 set_statvfs_info(const char *onp, int ukon, const char *fromp, int ukfrom,
2534     struct mount *mp, struct lwp *l)
2535 {
2536 	int error;
2537 	size_t size;
2538 	struct statvfs *sfs = &mp->mnt_stat;
2539 	int (*fun)(const void *, void *, size_t, size_t *);
2540 
2541 	(void)strncpy(mp->mnt_stat.f_fstypename, mp->mnt_op->vfs_name,
2542 	    sizeof(mp->mnt_stat.f_fstypename));
2543 
2544 	if (onp) {
2545 		struct cwdinfo *cwdi = l->l_proc->p_cwdi;
2546 		fun = (ukon == UIO_SYSSPACE) ? copystr : copyinstr;
2547 		if (cwdi->cwdi_rdir != NULL) {
2548 			size_t len;
2549 			char *bp;
2550 			char *path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
2551 
2552 			if (!path) /* XXX can't happen with M_WAITOK */
2553 				return ENOMEM;
2554 
2555 			bp = path + MAXPATHLEN;
2556 			*--bp = '\0';
2557 			error = getcwd_common(cwdi->cwdi_rdir, rootvnode, &bp,
2558 			    path, MAXPATHLEN / 2, 0, l);
2559 			if (error) {
2560 				free(path, M_TEMP);
2561 				return error;
2562 			}
2563 
2564 			len = strlen(bp);
2565 			if (len > sizeof(sfs->f_mntonname) - 1)
2566 				len = sizeof(sfs->f_mntonname) - 1;
2567 			(void)strncpy(sfs->f_mntonname, bp, len);
2568 			free(path, M_TEMP);
2569 
2570 			if (len < sizeof(sfs->f_mntonname) - 1) {
2571 				error = (*fun)(onp, &sfs->f_mntonname[len],
2572 				    sizeof(sfs->f_mntonname) - len - 1, &size);
2573 				if (error)
2574 					return error;
2575 				size += len;
2576 			} else {
2577 				size = len;
2578 			}
2579 		} else {
2580 			error = (*fun)(onp, &sfs->f_mntonname,
2581 			    sizeof(sfs->f_mntonname) - 1, &size);
2582 			if (error)
2583 				return error;
2584 		}
2585 		(void)memset(sfs->f_mntonname + size, 0,
2586 		    sizeof(sfs->f_mntonname) - size);
2587 	}
2588 
2589 	if (fromp) {
2590 		fun = (ukfrom == UIO_SYSSPACE) ? copystr : copyinstr;
2591 		error = (*fun)(fromp, sfs->f_mntfromname,
2592 		    sizeof(sfs->f_mntfromname) - 1, &size);
2593 		if (error)
2594 			return error;
2595 		(void)memset(sfs->f_mntfromname + size, 0,
2596 		    sizeof(sfs->f_mntfromname) - size);
2597 	}
2598 	return 0;
2599 }
2600 
2601 void
2602 vfs_timestamp(struct timespec *ts)
2603 {
2604 
2605 	nanotime(ts);
2606 }
2607 
2608 /*
2609  * mount_specific_key_create --
2610  *	Create a key for subsystem mount-specific data.
2611  */
2612 int
2613 mount_specific_key_create(specificdata_key_t *keyp, specificdata_dtor_t dtor)
2614 {
2615 
2616 	return (specificdata_key_create(mount_specificdata_domain, keyp, dtor));
2617 }
2618 
2619 /*
2620  * mount_specific_key_delete --
2621  *	Delete a key for subsystem mount-specific data.
2622  */
2623 void
2624 mount_specific_key_delete(specificdata_key_t key)
2625 {
2626 
2627 	specificdata_key_delete(mount_specificdata_domain, key);
2628 }
2629 
2630 /*
2631  * mount_initspecific --
2632  *	Initialize a mount's specificdata container.
2633  */
2634 void
2635 mount_initspecific(struct mount *mp)
2636 {
2637 	int error;
2638 
2639 	error = specificdata_init(mount_specificdata_domain,
2640 				  &mp->mnt_specdataref);
2641 	KASSERT(error == 0);
2642 }
2643 
2644 /*
2645  * mount_finispecific --
2646  *	Finalize a mount's specificdata container.
2647  */
2648 void
2649 mount_finispecific(struct mount *mp)
2650 {
2651 
2652 	specificdata_fini(mount_specificdata_domain, &mp->mnt_specdataref);
2653 }
2654 
2655 /*
2656  * mount_getspecific --
2657  *	Return mount-specific data corresponding to the specified key.
2658  */
2659 void *
2660 mount_getspecific(struct mount *mp, specificdata_key_t key)
2661 {
2662 
2663 	return (specificdata_getspecific(mount_specificdata_domain,
2664 					 &mp->mnt_specdataref, key));
2665 }
2666 
2667 /*
2668  * mount_setspecific --
2669  *	Set mount-specific data corresponding to the specified key.
2670  */
2671 void
2672 mount_setspecific(struct mount *mp, specificdata_key_t key, void *data)
2673 {
2674 
2675 	specificdata_setspecific(mount_specificdata_domain,
2676 				 &mp->mnt_specdataref, key, data);
2677 }
2678 
2679 #ifdef DDB
2680 static const char buf_flagbits[] = BUF_FLAGBITS;
2681 
2682 void
2683 vfs_buf_print(struct buf *bp, int full, void (*pr)(const char *, ...))
2684 {
2685 	char bf[1024];
2686 
2687 	(*pr)("  vp %p lblkno 0x%"PRIx64" blkno 0x%"PRIx64" rawblkno 0x%"
2688 	    PRIx64 " dev 0x%x\n",
2689 	    bp->b_vp, bp->b_lblkno, bp->b_blkno, bp->b_rawblkno, bp->b_dev);
2690 
2691 	bitmask_snprintf(bp->b_flags, buf_flagbits, bf, sizeof(bf));
2692 	(*pr)("  error %d flags 0x%s\n", bp->b_error, bf);
2693 
2694 	(*pr)("  bufsize 0x%lx bcount 0x%lx resid 0x%lx\n",
2695 		  bp->b_bufsize, bp->b_bcount, bp->b_resid);
2696 	(*pr)("  data %p saveaddr %p dep %p\n",
2697 		  bp->b_data, bp->b_saveaddr, LIST_FIRST(&bp->b_dep));
2698 	(*pr)("  iodone %p\n", bp->b_iodone);
2699 }
2700 
2701 
2702 void
2703 vfs_vnode_print(struct vnode *vp, int full, void (*pr)(const char *, ...))
2704 {
2705 	char bf[256];
2706 
2707 	uvm_object_printit(&vp->v_uobj, full, pr);
2708 	bitmask_snprintf(vp->v_flag, vnode_flagbits, bf, sizeof(bf));
2709 	(*pr)("\nVNODE flags %s\n", bf);
2710 	(*pr)("mp %p numoutput %d size 0x%llx\n",
2711 	      vp->v_mount, vp->v_numoutput, vp->v_size);
2712 
2713 	(*pr)("data %p usecount %d writecount %ld holdcnt %ld numoutput %d\n",
2714 	      vp->v_data, vp->v_usecount, vp->v_writecount,
2715 	      vp->v_holdcnt, vp->v_numoutput);
2716 
2717 	(*pr)("tag %s(%d) type %s(%d) mount %p typedata %p\n",
2718 	      ARRAY_PRINT(vp->v_tag, vnode_tags), vp->v_tag,
2719 	      ARRAY_PRINT(vp->v_type, vnode_types), vp->v_type,
2720 	      vp->v_mount, vp->v_mountedhere);
2721 
2722 	if (full) {
2723 		struct buf *bp;
2724 
2725 		(*pr)("clean bufs:\n");
2726 		LIST_FOREACH(bp, &vp->v_cleanblkhd, b_vnbufs) {
2727 			(*pr)(" bp %p\n", bp);
2728 			vfs_buf_print(bp, full, pr);
2729 		}
2730 
2731 		(*pr)("dirty bufs:\n");
2732 		LIST_FOREACH(bp, &vp->v_dirtyblkhd, b_vnbufs) {
2733 			(*pr)(" bp %p\n", bp);
2734 			vfs_buf_print(bp, full, pr);
2735 		}
2736 	}
2737 }
2738 
2739 void
2740 vfs_mount_print(struct mount *mp, int full, void (*pr)(const char *, ...))
2741 {
2742 	char sbuf[256];
2743 
2744 	(*pr)("vnodecovered = %p syncer = %p data = %p\n",
2745 			mp->mnt_vnodecovered,mp->mnt_syncer,mp->mnt_data);
2746 
2747 	(*pr)("fs_bshift %d dev_bshift = %d\n",
2748 			mp->mnt_fs_bshift,mp->mnt_dev_bshift);
2749 
2750 	bitmask_snprintf(mp->mnt_flag, __MNT_FLAG_BITS, sbuf, sizeof(sbuf));
2751 	(*pr)("flag = %s\n", sbuf);
2752 
2753 	bitmask_snprintf(mp->mnt_iflag, __IMNT_FLAG_BITS, sbuf, sizeof(sbuf));
2754 	(*pr)("iflag = %s\n", sbuf);
2755 
2756 	/* XXX use lockmgr_printinfo */
2757 	if (mp->mnt_lock.lk_sharecount)
2758 		(*pr)(" lock type %s: SHARED (count %d)", mp->mnt_lock.lk_wmesg,
2759 		    mp->mnt_lock.lk_sharecount);
2760 	else if (mp->mnt_lock.lk_flags & LK_HAVE_EXCL) {
2761 		(*pr)(" lock type %s: EXCL (count %d) by ",
2762 		    mp->mnt_lock.lk_wmesg, mp->mnt_lock.lk_exclusivecount);
2763 		if (mp->mnt_lock.lk_flags & LK_SPIN)
2764 			(*pr)("processor %lu", mp->mnt_lock.lk_cpu);
2765 		else
2766 			(*pr)("pid %d.%d", mp->mnt_lock.lk_lockholder,
2767 			    mp->mnt_lock.lk_locklwp);
2768 	} else
2769 		(*pr)(" not locked");
2770 	if ((mp->mnt_lock.lk_flags & LK_SPIN) == 0 && mp->mnt_lock.lk_waitcount > 0)
2771 		(*pr)(" with %d pending", mp->mnt_lock.lk_waitcount);
2772 
2773 	(*pr)("\n");
2774 
2775 	if (mp->mnt_unmounter) {
2776 		(*pr)("unmounter pid = %d ",mp->mnt_unmounter->l_proc);
2777 	}
2778 	(*pr)("wcnt = %d, writeopcountupper = %d, writeopcountupper = %d\n",
2779 		mp->mnt_wcnt,mp->mnt_writeopcountupper,mp->mnt_writeopcountlower);
2780 
2781 	(*pr)("statvfs cache:\n");
2782 	(*pr)("\tbsize = %lu\n",mp->mnt_stat.f_bsize);
2783 	(*pr)("\tfrsize = %lu\n",mp->mnt_stat.f_frsize);
2784 	(*pr)("\tiosize = %lu\n",mp->mnt_stat.f_iosize);
2785 
2786 	(*pr)("\tblocks = %"PRIu64"\n",mp->mnt_stat.f_blocks);
2787 	(*pr)("\tbfree = %"PRIu64"\n",mp->mnt_stat.f_bfree);
2788 	(*pr)("\tbavail = %"PRIu64"\n",mp->mnt_stat.f_bavail);
2789 	(*pr)("\tbresvd = %"PRIu64"\n",mp->mnt_stat.f_bresvd);
2790 
2791 	(*pr)("\tfiles = %"PRIu64"\n",mp->mnt_stat.f_files);
2792 	(*pr)("\tffree = %"PRIu64"\n",mp->mnt_stat.f_ffree);
2793 	(*pr)("\tfavail = %"PRIu64"\n",mp->mnt_stat.f_favail);
2794 	(*pr)("\tfresvd = %"PRIu64"\n",mp->mnt_stat.f_fresvd);
2795 
2796 	(*pr)("\tf_fsidx = { 0x%"PRIx32", 0x%"PRIx32" }\n",
2797 			mp->mnt_stat.f_fsidx.__fsid_val[0],
2798 			mp->mnt_stat.f_fsidx.__fsid_val[1]);
2799 
2800 	(*pr)("\towner = %"PRIu32"\n",mp->mnt_stat.f_owner);
2801 	(*pr)("\tnamemax = %lu\n",mp->mnt_stat.f_namemax);
2802 
2803 	bitmask_snprintf(mp->mnt_stat.f_flag, __MNT_FLAG_BITS, sbuf,
2804 	    sizeof(sbuf));
2805 	(*pr)("\tflag = %s\n",sbuf);
2806 	(*pr)("\tsyncwrites = %" PRIu64 "\n",mp->mnt_stat.f_syncwrites);
2807 	(*pr)("\tasyncwrites = %" PRIu64 "\n",mp->mnt_stat.f_asyncwrites);
2808 	(*pr)("\tsyncreads = %" PRIu64 "\n",mp->mnt_stat.f_syncreads);
2809 	(*pr)("\tasyncreads = %" PRIu64 "\n",mp->mnt_stat.f_asyncreads);
2810 	(*pr)("\tfstypename = %s\n",mp->mnt_stat.f_fstypename);
2811 	(*pr)("\tmntonname = %s\n",mp->mnt_stat.f_mntonname);
2812 	(*pr)("\tmntfromname = %s\n",mp->mnt_stat.f_mntfromname);
2813 
2814 	{
2815 		int cnt = 0;
2816 		struct vnode *vp;
2817 		(*pr)("locked vnodes =");
2818 		/* XXX would take mountlist lock, except ddb may not have context */
2819 		TAILQ_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
2820 			if (VOP_ISLOCKED(vp)) {
2821 				if ((++cnt % 6) == 0) {
2822 					(*pr)(" %p,\n\t", vp);
2823 				} else {
2824 					(*pr)(" %p,", vp);
2825 				}
2826 			}
2827 		}
2828 		(*pr)("\n");
2829 	}
2830 
2831 	if (full) {
2832 		int cnt = 0;
2833 		struct vnode *vp;
2834 		(*pr)("all vnodes =");
2835 		/* XXX would take mountlist lock, except ddb may not have context */
2836 		TAILQ_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
2837 			if (!TAILQ_NEXT(vp, v_mntvnodes)) {
2838 				(*pr)(" %p", vp);
2839 			} else if ((++cnt % 6) == 0) {
2840 				(*pr)(" %p,\n\t", vp);
2841 			} else {
2842 				(*pr)(" %p,", vp);
2843 			}
2844 		}
2845 		(*pr)("\n", vp);
2846 	}
2847 }
2848 #endif /* DDB */
2849