xref: /openbsd/sys/kern/vfs_subr.c (revision e5dd7070)
1 /*	$OpenBSD: vfs_subr.c,v 1.301 2020/03/27 07:58:17 anton Exp $	*/
2 /*	$NetBSD: vfs_subr.c,v 1.53 1996/04/22 01:39:13 christos Exp $	*/
3 
4 /*
5  * Copyright (c) 1989, 1993
6  *	The Regents of the University of California.  All rights reserved.
7  * (c) UNIX System Laboratories, Inc.
8  * All or some portions of this file are derived from material licensed
9  * to the University of California by American Telephone and Telegraph
10  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
11  * the permission of UNIX System Laboratories, Inc.
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. Neither the name of the University nor the names of its contributors
22  *    may be used to endorse or promote products derived from this software
23  *    without specific prior written permission.
24  *
25  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35  * SUCH DAMAGE.
36  *
37  *	@(#)vfs_subr.c	8.13 (Berkeley) 4/18/94
38  */
39 
40 /*
41  * External virtual filesystem routines
42  */
43 
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/proc.h>
47 #include <sys/sysctl.h>
48 #include <sys/mount.h>
49 #include <sys/time.h>
50 #include <sys/fcntl.h>
51 #include <sys/kernel.h>
52 #include <sys/conf.h>
53 #include <sys/vnode.h>
54 #include <sys/lock.h>
55 #include <sys/lockf.h>
56 #include <sys/stat.h>
57 #include <sys/acct.h>
58 #include <sys/namei.h>
59 #include <sys/ucred.h>
60 #include <sys/buf.h>
61 #include <sys/errno.h>
62 #include <sys/malloc.h>
63 #include <sys/mbuf.h>
64 #include <sys/syscallargs.h>
65 #include <sys/pool.h>
66 #include <sys/tree.h>
67 #include <sys/specdev.h>
68 #include <sys/atomic.h>
69 
70 #include <netinet/in.h>
71 
72 #include <uvm/uvm_extern.h>
73 #include <uvm/uvm_vnode.h>
74 
75 #include "softraid.h"
76 
77 void sr_quiesce(void);
78 
79 enum vtype iftovt_tab[16] = {
80 	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
81 	VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
82 };
83 
84 int	vttoif_tab[9] = {
85 	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
86 	S_IFSOCK, S_IFIFO, S_IFMT,
87 };
88 
89 int prtactive = 0;		/* 1 => print out reclaim of active vnodes */
90 int suid_clear = 1;		/* 1 => clear SUID / SGID on owner change */
91 
92 /*
93  * Insq/Remq for the vnode usage lists.
94  */
95 #define	bufinsvn(bp, dp)	LIST_INSERT_HEAD(dp, bp, b_vnbufs)
96 #define	bufremvn(bp) {							\
97 	LIST_REMOVE(bp, b_vnbufs);					\
98 	LIST_NEXT(bp, b_vnbufs) = NOLIST;				\
99 }
100 
101 struct freelst vnode_hold_list;	/* list of vnodes referencing buffers */
102 struct freelst vnode_free_list;	/* vnode free list */
103 
104 struct mntlist mountlist;	/* mounted filesystem list */
105 
106 void	vclean(struct vnode *, int, struct proc *);
107 
108 void insmntque(struct vnode *, struct mount *);
109 int getdevvp(dev_t, struct vnode **, enum vtype);
110 
111 int vfs_hang_addrlist(struct mount *, struct netexport *,
112 				  struct export_args *);
113 int vfs_free_netcred(struct radix_node *, void *, u_int);
114 void vfs_free_addrlist(struct netexport *);
115 void vputonfreelist(struct vnode *);
116 
117 int vflush_vnode(struct vnode *, void *);
118 int maxvnodes;
119 
120 void vfs_unmountall(void);
121 
122 #ifdef DEBUG
123 void printlockedvnodes(void);
124 #endif
125 
126 struct pool vnode_pool;
127 struct pool uvm_vnode_pool;
128 
129 static inline int rb_buf_compare(const struct buf *b1, const struct buf *b2);
130 RBT_GENERATE(buf_rb_bufs, buf, b_rbbufs, rb_buf_compare);
131 
132 static inline int
133 rb_buf_compare(const struct buf *b1, const struct buf *b2)
134 {
135 	if (b1->b_lblkno < b2->b_lblkno)
136 		return(-1);
137 	if (b1->b_lblkno > b2->b_lblkno)
138 		return(1);
139 	return(0);
140 }
141 
142 /*
143  * Initialize the vnode management data structures.
144  */
145 void
146 vntblinit(void)
147 {
148 	/* buffer cache may need a vnode for each buffer */
149 	maxvnodes = 2 * initialvnodes;
150 	pool_init(&vnode_pool, sizeof(struct vnode), 0, IPL_NONE,
151 	    PR_WAITOK, "vnodes", NULL);
152 	pool_init(&uvm_vnode_pool, sizeof(struct uvm_vnode), 0, IPL_NONE,
153 	    PR_WAITOK, "uvmvnodes", NULL);
154 	TAILQ_INIT(&vnode_hold_list);
155 	TAILQ_INIT(&vnode_free_list);
156 	TAILQ_INIT(&mountlist);
157 	/*
158 	 * Initialize the filesystem syncer.
159 	 */
160 	vn_initialize_syncerd();
161 
162 #ifdef NFSSERVER
163 	rn_init(sizeof(struct sockaddr_in));
164 #endif /* NFSSERVER */
165 }
166 
167 /*
168  * Allocate a mount point.
169  *
170  * The returned mount point is marked as busy.
171  */
172 struct mount *
173 vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp)
174 {
175 	struct mount *mp;
176 
177 	mp = malloc(sizeof(*mp), M_MOUNT, M_WAITOK|M_ZERO);
178 	rw_init_flags(&mp->mnt_lock, "vfslock", RWL_IS_VNODE);
179 	(void)vfs_busy(mp, VB_READ|VB_NOWAIT);
180 
181 	TAILQ_INIT(&mp->mnt_vnodelist);
182 	mp->mnt_vnodecovered = vp;
183 
184 	atomic_inc_int(&vfsp->vfc_refcount);
185 	mp->mnt_vfc = vfsp;
186 	mp->mnt_op = vfsp->vfc_vfsops;
187 	mp->mnt_flag = vfsp->vfc_flags;
188 	strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
189 
190 	return (mp);
191 }
192 
193 /*
194  * Release a mount point.
195  */
196 void
197 vfs_mount_free(struct mount *mp)
198 {
199 	atomic_dec_int(&mp->mnt_vfc->vfc_refcount);
200 	free(mp, M_MOUNT, sizeof(*mp));
201 }
202 
203 /*
204  * Mark a mount point as busy. Used to synchronize access and to delay
205  * unmounting.
206  *
207  * Default behaviour is to attempt getting a READ lock and in case of an
208  * ongoing unmount, to wait for it to finish and then return failure.
209  */
210 int
211 vfs_busy(struct mount *mp, int flags)
212 {
213 	int rwflags = 0;
214 
215 	if (flags & VB_WRITE)
216 		rwflags |= RW_WRITE;
217 	else
218 		rwflags |= RW_READ;
219 
220 	if (flags & VB_WAIT)
221 		rwflags |= RW_SLEEPFAIL;
222 	else
223 		rwflags |= RW_NOSLEEP;
224 
225 #ifdef WITNESS
226 	if (flags & VB_DUPOK)
227 		rwflags |= RW_DUPOK;
228 #endif
229 
230 	if (rw_enter(&mp->mnt_lock, rwflags))
231 		return (EBUSY);
232 
233 	return (0);
234 }
235 
236 /*
237  * Free a busy file system
238  */
239 void
240 vfs_unbusy(struct mount *mp)
241 {
242 	rw_exit(&mp->mnt_lock);
243 }
244 
245 int
246 vfs_isbusy(struct mount *mp)
247 {
248 	if (RWLOCK_OWNER(&mp->mnt_lock) > 0)
249 		return (1);
250 	else
251 		return (0);
252 }
253 
254 /*
255  * Lookup a filesystem type, and if found allocate and initialize
256  * a mount structure for it.
257  *
258  * Devname is usually updated by mount(8) after booting.
259  */
260 int
261 vfs_rootmountalloc(char *fstypename, char *devname, struct mount **mpp)
262 {
263 	struct vfsconf *vfsp;
264 	struct mount *mp;
265 
266 	vfsp = vfs_byname(fstypename);
267 	if (vfsp == NULL)
268 		return (ENODEV);
269 	mp = vfs_mount_alloc(NULLVP, vfsp);
270 	mp->mnt_flag |= MNT_RDONLY;
271 	mp->mnt_stat.f_mntonname[0] = '/';
272 	copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN, 0);
273 	copystr(devname, mp->mnt_stat.f_mntfromspec, MNAMELEN, 0);
274 	*mpp = mp;
275 	return (0);
276  }
277 
278 /*
279  * Lookup a mount point by filesystem identifier.
280  */
281 struct mount *
282 vfs_getvfs(fsid_t *fsid)
283 {
284 	struct mount *mp;
285 
286 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
287 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
288 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
289 			return (mp);
290 		}
291 	}
292 
293 	return (NULL);
294 }
295 
296 
297 /*
298  * Get a new unique fsid
299  */
300 void
301 vfs_getnewfsid(struct mount *mp)
302 {
303 	static u_short xxxfs_mntid;
304 
305 	fsid_t tfsid;
306 	int mtype;
307 
308 	mtype = mp->mnt_vfc->vfc_typenum;
309 	mp->mnt_stat.f_fsid.val[0] = makedev(nblkdev + mtype, 0);
310 	mp->mnt_stat.f_fsid.val[1] = mtype;
311 	if (xxxfs_mntid == 0)
312 		++xxxfs_mntid;
313 	tfsid.val[0] = makedev(nblkdev + mtype, xxxfs_mntid);
314 	tfsid.val[1] = mtype;
315 	if (!TAILQ_EMPTY(&mountlist)) {
316 		while (vfs_getvfs(&tfsid)) {
317 			tfsid.val[0]++;
318 			xxxfs_mntid++;
319 		}
320 	}
321 	mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
322 }
323 
324 /*
325  * Set vnode attributes to VNOVAL
326  */
327 void
328 vattr_null(struct vattr *vap)
329 {
330 
331 	vap->va_type = VNON;
332 	/*
333 	 * Don't get fancy: u_quad_t = u_int = VNOVAL leaves the u_quad_t
334 	 * with 2^31-1 instead of 2^64-1.  Just write'm out and let
335 	 * the compiler do its job.
336 	 */
337 	vap->va_mode = VNOVAL;
338 	vap->va_nlink = VNOVAL;
339 	vap->va_uid = VNOVAL;
340 	vap->va_gid = VNOVAL;
341 	vap->va_fsid = VNOVAL;
342 	vap->va_fileid = VNOVAL;
343 	vap->va_size = VNOVAL;
344 	vap->va_blocksize = VNOVAL;
345 	vap->va_atime.tv_sec = VNOVAL;
346 	vap->va_atime.tv_nsec = VNOVAL;
347 	vap->va_mtime.tv_sec = VNOVAL;
348 	vap->va_mtime.tv_nsec = VNOVAL;
349 	vap->va_ctime.tv_sec = VNOVAL;
350 	vap->va_ctime.tv_nsec = VNOVAL;
351 	vap->va_gen = VNOVAL;
352 	vap->va_flags = VNOVAL;
353 	vap->va_rdev = VNOVAL;
354 	vap->va_bytes = VNOVAL;
355 	vap->va_filerev = VNOVAL;
356 	vap->va_vaflags = 0;
357 }
358 
359 /*
360  * Routines having to do with the management of the vnode table.
361  */
362 long numvnodes;
363 
364 /*
365  * Return the next vnode from the free list.
366  */
367 int
368 getnewvnode(enum vtagtype tag, struct mount *mp, const struct vops *vops,
369     struct vnode **vpp)
370 {
371 	struct proc *p = curproc;
372 	struct freelst *listhd;
373 	static int toggle;
374 	struct vnode *vp;
375 	int s;
376 
377 	/*
378 	 * allow maxvnodes to increase if the buffer cache itself
379 	 * is big enough to justify it. (we don't shrink it ever)
380 	 */
381 	maxvnodes = maxvnodes < bcstats.numbufs ? bcstats.numbufs
382 	    : maxvnodes;
383 
384 	/*
385 	 * We must choose whether to allocate a new vnode or recycle an
386 	 * existing one. The criterion for allocating a new one is that
387 	 * the total number of vnodes is less than the number desired or
388 	 * there are no vnodes on either free list. Generally we only
389 	 * want to recycle vnodes that have no buffers associated with
390 	 * them, so we look first on the vnode_free_list. If it is empty,
391 	 * we next consider vnodes with referencing buffers on the
392 	 * vnode_hold_list. The toggle ensures that half the time we
393 	 * will use a buffer from the vnode_hold_list, and half the time
394 	 * we will allocate a new one unless the list has grown to twice
395 	 * the desired size. We are reticent to recycle vnodes from the
396 	 * vnode_hold_list because we will lose the identity of all its
397 	 * referencing buffers.
398 	 */
399 	toggle ^= 1;
400 	if (numvnodes / 2 > maxvnodes)
401 		toggle = 0;
402 
403 	s = splbio();
404 	if ((numvnodes < maxvnodes) ||
405 	    ((TAILQ_FIRST(listhd = &vnode_free_list) == NULL) &&
406 	    ((TAILQ_FIRST(listhd = &vnode_hold_list) == NULL) || toggle))) {
407 		splx(s);
408 		vp = pool_get(&vnode_pool, PR_WAITOK | PR_ZERO);
409 		vp->v_uvm = pool_get(&uvm_vnode_pool, PR_WAITOK | PR_ZERO);
410 		vp->v_uvm->u_vnode = vp;
411 		RBT_INIT(buf_rb_bufs, &vp->v_bufs_tree);
412 		cache_tree_init(&vp->v_nc_tree);
413 		TAILQ_INIT(&vp->v_cache_dst);
414 		numvnodes++;
415 	} else {
416 		TAILQ_FOREACH(vp, listhd, v_freelist) {
417 			if (VOP_ISLOCKED(vp) == 0)
418 				break;
419 		}
420 		/*
421 		 * Unless this is a bad time of the month, at most
422 		 * the first NCPUS items on the free list are
423 		 * locked, so this is close enough to being empty.
424 		 */
425 		if (vp == NULL) {
426 			splx(s);
427 			tablefull("vnode");
428 			*vpp = 0;
429 			return (ENFILE);
430 		}
431 
432 #ifdef DIAGNOSTIC
433 		if (vp->v_usecount) {
434 			vprint("free vnode", vp);
435 			panic("free vnode isn't");
436 		}
437 #endif
438 
439 		TAILQ_REMOVE(listhd, vp, v_freelist);
440 		vp->v_bioflag &= ~VBIOONFREELIST;
441 		splx(s);
442 
443 		if (vp->v_type != VBAD)
444 			vgonel(vp, p);
445 #ifdef DIAGNOSTIC
446 		if (vp->v_data) {
447 			vprint("cleaned vnode", vp);
448 			panic("cleaned vnode isn't");
449 		}
450 		s = splbio();
451 		if (vp->v_numoutput)
452 			panic("Clean vnode has pending I/O's");
453 		splx(s);
454 #endif
455 		vp->v_flag = 0;
456 		vp->v_socket = 0;
457 	}
458 	cache_purge(vp);
459 	vp->v_type = VNON;
460 	vp->v_tag = tag;
461 	vp->v_op = vops;
462 	insmntque(vp, mp);
463 	*vpp = vp;
464 	vp->v_usecount = 1;
465 	vp->v_data = 0;
466 	return (0);
467 }
468 
469 /*
470  * Move a vnode from one mount queue to another.
471  */
472 void
473 insmntque(struct vnode *vp, struct mount *mp)
474 {
475 	/*
476 	 * Delete from old mount point vnode list, if on one.
477 	 */
478 	if (vp->v_mount != NULL)
479 		TAILQ_REMOVE(&vp->v_mount->mnt_vnodelist, vp, v_mntvnodes);
480 	/*
481 	 * Insert into list of vnodes for the new mount point, if available.
482 	 */
483 	if ((vp->v_mount = mp) != NULL)
484 		TAILQ_INSERT_TAIL(&mp->mnt_vnodelist, vp, v_mntvnodes);
485 }
486 
487 /*
488  * Create a vnode for a block device.
489  * Used for root filesystem, argdev, and swap areas.
490  * Also used for memory file system special devices.
491  */
492 int
493 bdevvp(dev_t dev, struct vnode **vpp)
494 {
495 	return (getdevvp(dev, vpp, VBLK));
496 }
497 
498 /*
499  * Create a vnode for a character device.
500  * Used for console handling.
501  */
502 int
503 cdevvp(dev_t dev, struct vnode **vpp)
504 {
505 	return (getdevvp(dev, vpp, VCHR));
506 }
507 
508 /*
509  * Create a vnode for a device.
510  * Used by bdevvp (block device) for root file system etc.,
511  * and by cdevvp (character device) for console.
512  */
513 int
514 getdevvp(dev_t dev, struct vnode **vpp, enum vtype type)
515 {
516 	struct vnode *vp;
517 	struct vnode *nvp;
518 	int error;
519 
520 	if (dev == NODEV) {
521 		*vpp = NULLVP;
522 		return (0);
523 	}
524 	error = getnewvnode(VT_NON, NULL, &spec_vops, &nvp);
525 	if (error) {
526 		*vpp = NULLVP;
527 		return (error);
528 	}
529 	vp = nvp;
530 	vp->v_type = type;
531 	if ((nvp = checkalias(vp, dev, NULL)) != 0) {
532 		vput(vp);
533 		vp = nvp;
534 	}
535 	if (vp->v_type == VCHR && cdevsw[major(vp->v_rdev)].d_type == D_TTY)
536 		vp->v_flag |= VISTTY;
537 	*vpp = vp;
538 	return (0);
539 }
540 
541 /*
542  * Check to see if the new vnode represents a special device
543  * for which we already have a vnode (either because of
544  * bdevvp() or because of a different vnode representing
545  * the same block device). If such an alias exists, deallocate
546  * the existing contents and return the aliased vnode. The
547  * caller is responsible for filling it with its new contents.
548  */
549 struct vnode *
550 checkalias(struct vnode *nvp, dev_t nvp_rdev, struct mount *mp)
551 {
552 	struct proc *p = curproc;
553 	struct vnode *vp;
554 	struct vnodechain *vchain;
555 
556 	if (nvp->v_type != VBLK && nvp->v_type != VCHR)
557 		return (NULLVP);
558 
559 	vchain = &speclisth[SPECHASH(nvp_rdev)];
560 loop:
561 	SLIST_FOREACH(vp, vchain, v_specnext) {
562 		if (nvp_rdev != vp->v_rdev || nvp->v_type != vp->v_type) {
563 			continue;
564 		}
565 		/*
566 		 * Alias, but not in use, so flush it out.
567 		 */
568 		if (vp->v_usecount == 0) {
569 			vgonel(vp, p);
570 			goto loop;
571 		}
572 		if (vget(vp, LK_EXCLUSIVE)) {
573 			goto loop;
574 		}
575 		break;
576 	}
577 
578 	/*
579 	 * Common case is actually in the if statement
580 	 */
581 	if (vp == NULL || !(vp->v_tag == VT_NON && vp->v_type == VBLK)) {
582 		nvp->v_specinfo = malloc(sizeof(struct specinfo), M_VNODE,
583 			M_WAITOK);
584 		nvp->v_rdev = nvp_rdev;
585 		nvp->v_hashchain = vchain;
586 		nvp->v_specmountpoint = NULL;
587 		nvp->v_speclockf = NULL;
588 		nvp->v_specbitmap = NULL;
589 		if (nvp->v_type == VCHR &&
590 		    (cdevsw[major(nvp_rdev)].d_flags & D_CLONE) &&
591 		    (minor(nvp_rdev) >> CLONE_SHIFT == 0)) {
592 			if (vp != NULLVP)
593 				nvp->v_specbitmap = vp->v_specbitmap;
594 			else
595 				nvp->v_specbitmap = malloc(CLONE_MAPSZ,
596 				    M_VNODE, M_WAITOK | M_ZERO);
597 		}
598 		SLIST_INSERT_HEAD(vchain, nvp, v_specnext);
599 		if (vp != NULLVP) {
600 			nvp->v_flag |= VALIASED;
601 			vp->v_flag |= VALIASED;
602 			vput(vp);
603 		}
604 		return (NULLVP);
605 	}
606 
607 	/*
608 	 * This code is the uncommon case. It is called in case
609 	 * we found an alias that was VT_NON && vtype of VBLK
610 	 * This means we found a block device that was created
611 	 * using bdevvp.
612 	 * An example of such a vnode is the root partition device vnode
613 	 * created in ffs_mountroot.
614 	 *
615 	 * The vnodes created by bdevvp should not be aliased (why?).
616 	 */
617 
618 	VOP_UNLOCK(vp);
619 	vclean(vp, 0, p);
620 	vp->v_op = nvp->v_op;
621 	vp->v_tag = nvp->v_tag;
622 	nvp->v_type = VNON;
623 	insmntque(vp, mp);
624 	return (vp);
625 }
626 
627 /*
628  * Grab a particular vnode from the free list, increment its
629  * reference count and lock it. If the vnode lock bit is set,
630  * the vnode is being eliminated in vgone. In that case, we
631  * cannot grab it, so the process is awakened when the
632  * transition is completed, and an error code is returned to
633  * indicate that the vnode is no longer usable, possibly
634  * having been changed to a new file system type.
635  */
636 int
637 vget(struct vnode *vp, int flags)
638 {
639 	int error, s, onfreelist;
640 
641 	/*
642 	 * If the vnode is in the process of being cleaned out for
643 	 * another use, we wait for the cleaning to finish and then
644 	 * return failure. Cleaning is determined by checking that
645 	 * the VXLOCK flag is set.
646 	 */
647 
648 	if (vp->v_flag & VXLOCK) {
649 		if (flags & LK_NOWAIT) {
650 			return (EBUSY);
651 		}
652 
653 		vp->v_flag |= VXWANT;
654 		tsleep_nsec(vp, PINOD, "vget", INFSLP);
655 		return (ENOENT);
656 	}
657 
658 	onfreelist = vp->v_bioflag & VBIOONFREELIST;
659 	if (vp->v_usecount == 0 && onfreelist) {
660 		s = splbio();
661 		if (vp->v_holdcnt > 0)
662 			TAILQ_REMOVE(&vnode_hold_list, vp, v_freelist);
663 		else
664 			TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
665 		vp->v_bioflag &= ~VBIOONFREELIST;
666 		splx(s);
667 	}
668 
669 	vp->v_usecount++;
670 	if (flags & LK_TYPE_MASK) {
671 		if ((error = vn_lock(vp, flags)) != 0) {
672 			vp->v_usecount--;
673 			if (vp->v_usecount == 0 && onfreelist)
674 				vputonfreelist(vp);
675 		}
676 		return (error);
677 	}
678 
679 	return (0);
680 }
681 
682 
683 /* Vnode reference. */
684 void
685 vref(struct vnode *vp)
686 {
687 #ifdef DIAGNOSTIC
688 	if (vp->v_usecount == 0)
689 		panic("vref used where vget required");
690 	if (vp->v_type == VNON)
691 		panic("vref on a VNON vnode");
692 #endif
693 	vp->v_usecount++;
694 }
695 
696 void
697 vputonfreelist(struct vnode *vp)
698 {
699 	int s;
700 	struct freelst *lst;
701 
702 	s = splbio();
703 #ifdef DIAGNOSTIC
704 	if (vp->v_usecount != 0)
705 		panic("Use count is not zero!");
706 
707 	/*
708 	 * If the hold count is still positive, one or many threads could still
709 	 * be waiting on the vnode lock inside uvn_io().
710 	 */
711 	if (vp->v_holdcnt == 0 && vp->v_lockcount != 0)
712 		panic("%s: lock count is not zero", __func__);
713 
714 	if (vp->v_bioflag & VBIOONFREELIST) {
715 		vprint("vnode already on free list: ", vp);
716 		panic("vnode already on free list");
717 	}
718 #endif
719 
720 	vp->v_bioflag |= VBIOONFREELIST;
721 	vp->v_bioflag &= ~VBIOERROR;
722 
723 	if (vp->v_holdcnt > 0)
724 		lst = &vnode_hold_list;
725 	else
726 		lst = &vnode_free_list;
727 
728 	if (vp->v_type == VBAD)
729 		TAILQ_INSERT_HEAD(lst, vp, v_freelist);
730 	else
731 		TAILQ_INSERT_TAIL(lst, vp, v_freelist);
732 
733 	splx(s);
734 }
735 
736 /*
737  * vput(), just unlock and vrele()
738  */
739 void
740 vput(struct vnode *vp)
741 {
742 	struct proc *p = curproc;
743 
744 #ifdef DIAGNOSTIC
745 	if (vp == NULL)
746 		panic("vput: null vp");
747 #endif
748 
749 #ifdef DIAGNOSTIC
750 	if (vp->v_usecount == 0) {
751 		vprint("vput: bad ref count", vp);
752 		panic("vput: ref cnt");
753 	}
754 #endif
755 	vp->v_usecount--;
756 	KASSERT(vp->v_usecount > 0 || vp->v_uvcount == 0);
757 	if (vp->v_usecount > 0) {
758 		VOP_UNLOCK(vp);
759 		return;
760 	}
761 
762 #ifdef DIAGNOSTIC
763 	if (vp->v_writecount != 0) {
764 		vprint("vput: bad writecount", vp);
765 		panic("vput: v_writecount != 0");
766 	}
767 #endif
768 
769 	VOP_INACTIVE(vp, p);
770 
771 	if (vp->v_usecount == 0 && !(vp->v_bioflag & VBIOONFREELIST))
772 		vputonfreelist(vp);
773 }
774 
775 /*
776  * Vnode release - use for active VNODES.
777  * If count drops to zero, call inactive routine and return to freelist.
778  * Returns 0 if it did not sleep.
779  */
780 int
781 vrele(struct vnode *vp)
782 {
783 	struct proc *p = curproc;
784 
785 #ifdef DIAGNOSTIC
786 	if (vp == NULL)
787 		panic("vrele: null vp");
788 #endif
789 #ifdef DIAGNOSTIC
790 	if (vp->v_usecount == 0) {
791 		vprint("vrele: bad ref count", vp);
792 		panic("vrele: ref cnt");
793 	}
794 #endif
795 	vp->v_usecount--;
796 	if (vp->v_usecount > 0) {
797 		return (0);
798 	}
799 
800 #ifdef DIAGNOSTIC
801 	if (vp->v_writecount != 0) {
802 		vprint("vrele: bad writecount", vp);
803 		panic("vrele: v_writecount != 0");
804 	}
805 #endif
806 
807 	if (vn_lock(vp, LK_EXCLUSIVE)) {
808 #ifdef DIAGNOSTIC
809 		vprint("vrele: cannot lock", vp);
810 #endif
811 		return (1);
812 	}
813 
814 	VOP_INACTIVE(vp, p);
815 
816 	if (vp->v_usecount == 0 && !(vp->v_bioflag & VBIOONFREELIST))
817 		vputonfreelist(vp);
818 	return (1);
819 }
820 
821 /* Page or buffer structure gets a reference. */
822 void
823 vhold(struct vnode *vp)
824 {
825 	/*
826 	 * If it is on the freelist and the hold count is currently
827 	 * zero, move it to the hold list.
828 	 */
829 	if ((vp->v_bioflag & VBIOONFREELIST) &&
830 	    vp->v_holdcnt == 0 && vp->v_usecount == 0) {
831 		TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
832 		TAILQ_INSERT_TAIL(&vnode_hold_list, vp, v_freelist);
833 	}
834 	vp->v_holdcnt++;
835 }
836 
837 /* Lose interest in a vnode. */
838 void
839 vdrop(struct vnode *vp)
840 {
841 #ifdef DIAGNOSTIC
842 	if (vp->v_holdcnt == 0)
843 		panic("vdrop: zero holdcnt");
844 #endif
845 
846 	vp->v_holdcnt--;
847 
848 	/*
849 	 * If it is on the holdlist and the hold count drops to
850 	 * zero, move it to the free list.
851 	 */
852 	if ((vp->v_bioflag & VBIOONFREELIST) &&
853 	    vp->v_holdcnt == 0 && vp->v_usecount == 0) {
854 		TAILQ_REMOVE(&vnode_hold_list, vp, v_freelist);
855 		TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
856 	}
857 }
858 
859 /*
860  * Remove any vnodes in the vnode table belonging to mount point mp.
861  *
862  * If MNT_NOFORCE is specified, there should not be any active ones,
863  * return error if any are found (nb: this is a user error, not a
864  * system error). If MNT_FORCE is specified, detach any active vnodes
865  * that are found.
866  */
867 #ifdef DEBUG
868 int busyprt = 0;	/* print out busy vnodes */
869 struct ctldebug debug1 = { "busyprt", &busyprt };
870 #endif
871 
872 int
873 vfs_mount_foreach_vnode(struct mount *mp,
874     int (*func)(struct vnode *, void *), void *arg) {
875 	struct vnode *vp, *nvp;
876 	int error = 0;
877 
878 loop:
879 	TAILQ_FOREACH_SAFE(vp , &mp->mnt_vnodelist, v_mntvnodes, nvp) {
880 		if (vp->v_mount != mp)
881 			goto loop;
882 
883 		error = func(vp, arg);
884 
885 		if (error != 0)
886 			break;
887 	}
888 
889 	return (error);
890 }
891 
892 struct vflush_args {
893 	struct vnode *skipvp;
894 	int busy;
895 	int flags;
896 };
897 
898 int
899 vflush_vnode(struct vnode *vp, void *arg)
900 {
901 	struct vflush_args *va = arg;
902 	struct proc *p = curproc;
903 
904 	if (vp == va->skipvp) {
905 		return (0);
906 	}
907 
908 	if ((va->flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) {
909 		return (0);
910 	}
911 
912 	/*
913 	 * If WRITECLOSE is set, only flush out regular file
914 	 * vnodes open for writing.
915 	 */
916 	if ((va->flags & WRITECLOSE) &&
917 	    (vp->v_writecount == 0 || vp->v_type != VREG)) {
918 		return (0);
919 	}
920 
921 	/*
922 	 * With v_usecount == 0, all we need to do is clear
923 	 * out the vnode data structures and we are done.
924 	 */
925 	if (vp->v_usecount == 0) {
926 		vgonel(vp, p);
927 		return (0);
928 	}
929 
930 	/*
931 	 * If FORCECLOSE is set, forcibly close the vnode.
932 	 * For block or character devices, revert to an
933 	 * anonymous device. For all other files, just kill them.
934 	 */
935 	if (va->flags & FORCECLOSE) {
936 		if (vp->v_type != VBLK && vp->v_type != VCHR) {
937 			vgonel(vp, p);
938 		} else {
939 			vclean(vp, 0, p);
940 			vp->v_op = &spec_vops;
941 			insmntque(vp, NULL);
942 		}
943 		return (0);
944 	}
945 
946 	/*
947 	 * If set, this is allowed to ignore vnodes which don't
948 	 * have changes pending to disk.
949 	 * XXX Might be nice to check per-fs "inode" flags, but
950 	 * generally the filesystem is sync'd already, right?
951 	 */
952 	if ((va->flags & IGNORECLEAN) &&
953 	    LIST_EMPTY(&vp->v_dirtyblkhd))
954 		return (0);
955 
956 #ifdef DEBUG
957 	if (busyprt)
958 		vprint("vflush: busy vnode", vp);
959 #endif
960 	va->busy++;
961 	return (0);
962 }
963 
964 int
965 vflush(struct mount *mp, struct vnode *skipvp, int flags)
966 {
967 	struct vflush_args va;
968 	va.skipvp = skipvp;
969 	va.busy = 0;
970 	va.flags = flags;
971 
972 	vfs_mount_foreach_vnode(mp, vflush_vnode, &va);
973 
974 	if (va.busy)
975 		return (EBUSY);
976 	return (0);
977 }
978 
979 /*
980  * Disassociate the underlying file system from a vnode.
981  */
982 void
983 vclean(struct vnode *vp, int flags, struct proc *p)
984 {
985 	int active;
986 
987 	/*
988 	 * Check to see if the vnode is in use.
989 	 * If so we have to reference it before we clean it out
990 	 * so that its count cannot fall to zero and generate a
991 	 * race against ourselves to recycle it.
992 	 */
993 	if ((active = vp->v_usecount) != 0)
994 		vp->v_usecount++;
995 
996 	/*
997 	 * Prevent the vnode from being recycled or
998 	 * brought into use while we clean it out.
999 	 */
1000 	if (vp->v_flag & VXLOCK)
1001 		panic("vclean: deadlock");
1002 	vp->v_flag |= VXLOCK;
1003 
1004 	if (vp->v_lockcount > 0) {
1005 		/*
1006 		 * Ensure that any thread currently waiting on the same lock has
1007 		 * observed that the vnode is about to be exclusively locked
1008 		 * before continuing.
1009 		 */
1010 		tsleep_nsec(&vp->v_lockcount, PINOD, "vop_lock", INFSLP);
1011 		KASSERT(vp->v_lockcount == 0);
1012 	}
1013 
1014 	/*
1015 	 * Even if the count is zero, the VOP_INACTIVE routine may still
1016 	 * have the object locked while it cleans it out. The VOP_LOCK
1017 	 * ensures that the VOP_INACTIVE routine is done with its work.
1018 	 * For active vnodes, it ensures that no other activity can
1019 	 * occur while the underlying object is being cleaned out.
1020 	 */
1021 	VOP_LOCK(vp, LK_DRAIN | LK_EXCLUSIVE);
1022 
1023 	/*
1024 	 * Clean out any VM data associated with the vnode.
1025 	 */
1026 	uvm_vnp_terminate(vp);
1027 	/*
1028 	 * Clean out any buffers associated with the vnode.
1029 	 */
1030 	if (flags & DOCLOSE)
1031 		vinvalbuf(vp, V_SAVE, NOCRED, p, 0, INFSLP);
1032 	/*
1033 	 * If purging an active vnode, it must be closed and
1034 	 * deactivated before being reclaimed. Note that the
1035 	 * VOP_INACTIVE will unlock the vnode
1036 	 */
1037 	if (active) {
1038 		if (flags & DOCLOSE)
1039 			VOP_CLOSE(vp, FNONBLOCK, NOCRED, p);
1040 		VOP_INACTIVE(vp, p);
1041 	} else {
1042 		/*
1043 		 * Any other processes trying to obtain this lock must first
1044 		 * wait for VXLOCK to clear, then call the new lock operation.
1045 		 */
1046 		VOP_UNLOCK(vp);
1047 	}
1048 
1049 	/*
1050 	 * Reclaim the vnode.
1051 	 */
1052 	if (VOP_RECLAIM(vp, p))
1053 		panic("vclean: cannot reclaim");
1054 	if (active) {
1055 		vp->v_usecount--;
1056 		if (vp->v_usecount == 0) {
1057 			if (vp->v_holdcnt > 0)
1058 				panic("vclean: not clean");
1059 			vputonfreelist(vp);
1060 		}
1061 	}
1062 	cache_purge(vp);
1063 
1064 	/*
1065 	 * Done with purge, notify sleepers of the grim news.
1066 	 */
1067 	vp->v_op = &dead_vops;
1068 	VN_KNOTE(vp, NOTE_REVOKE);
1069 	vp->v_tag = VT_NON;
1070 	vp->v_flag &= ~VXLOCK;
1071 #ifdef VFSLCKDEBUG
1072 	vp->v_flag &= ~VLOCKSWORK;
1073 #endif
1074 	if (vp->v_flag & VXWANT) {
1075 		vp->v_flag &= ~VXWANT;
1076 		wakeup(vp);
1077 	}
1078 }
1079 
1080 /*
1081  * Recycle an unused vnode to the front of the free list.
1082  */
1083 int
1084 vrecycle(struct vnode *vp, struct proc *p)
1085 {
1086 	if (vp->v_usecount == 0) {
1087 		vgonel(vp, p);
1088 		return (1);
1089 	}
1090 	return (0);
1091 }
1092 
1093 /*
1094  * Eliminate all activity associated with a vnode
1095  * in preparation for reuse.
1096  */
1097 void
1098 vgone(struct vnode *vp)
1099 {
1100 	struct proc *p = curproc;
1101 	vgonel(vp, p);
1102 }
1103 
1104 /*
1105  * vgone, with struct proc.
1106  */
1107 void
1108 vgonel(struct vnode *vp, struct proc *p)
1109 {
1110 	struct vnode *vq;
1111 	struct vnode *vx;
1112 
1113 	KASSERT(vp->v_uvcount == 0);
1114 
1115 	/*
1116 	 * If a vgone (or vclean) is already in progress,
1117 	 * wait until it is done and return.
1118 	 */
1119 	if (vp->v_flag & VXLOCK) {
1120 		vp->v_flag |= VXWANT;
1121 		tsleep_nsec(vp, PINOD, "vgone", INFSLP);
1122 		return;
1123 	}
1124 
1125 	/*
1126 	 * Clean out the filesystem specific data.
1127 	 */
1128 	vclean(vp, DOCLOSE, p);
1129 	/*
1130 	 * Delete from old mount point vnode list, if on one.
1131 	 */
1132 	if (vp->v_mount != NULL)
1133 		insmntque(vp, NULL);
1134 	/*
1135 	 * If special device, remove it from special device alias list
1136 	 * if it is on one.
1137 	 */
1138 	if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_specinfo != 0) {
1139 		if ((vp->v_flag & VALIASED) == 0 && vp->v_type == VCHR &&
1140 		    (cdevsw[major(vp->v_rdev)].d_flags & D_CLONE) &&
1141 		    (minor(vp->v_rdev) >> CLONE_SHIFT == 0)) {
1142 			free(vp->v_specbitmap, M_VNODE, CLONE_MAPSZ);
1143 		}
1144 		SLIST_REMOVE(vp->v_hashchain, vp, vnode, v_specnext);
1145 		if (vp->v_flag & VALIASED) {
1146 			vx = NULL;
1147 			SLIST_FOREACH(vq, vp->v_hashchain, v_specnext) {
1148 				if (vq->v_rdev != vp->v_rdev ||
1149 				    vq->v_type != vp->v_type)
1150 					continue;
1151 				if (vx)
1152 					break;
1153 				vx = vq;
1154 			}
1155 			if (vx == NULL)
1156 				panic("missing alias");
1157 			if (vq == NULL)
1158 				vx->v_flag &= ~VALIASED;
1159 			vp->v_flag &= ~VALIASED;
1160 		}
1161 		lf_purgelocks(&vp->v_speclockf);
1162 		free(vp->v_specinfo, M_VNODE, sizeof(struct specinfo));
1163 		vp->v_specinfo = NULL;
1164 	}
1165 	/*
1166 	 * If it is on the freelist and not already at the head,
1167 	 * move it to the head of the list.
1168 	 */
1169 	vp->v_type = VBAD;
1170 
1171 	/*
1172 	 * Move onto the free list, unless we were called from
1173 	 * getnewvnode and we're not on any free list
1174 	 */
1175 	if (vp->v_usecount == 0 &&
1176 	    (vp->v_bioflag & VBIOONFREELIST)) {
1177 		int s;
1178 
1179 		s = splbio();
1180 
1181 		if (vp->v_holdcnt > 0)
1182 			panic("vgonel: not clean");
1183 
1184 		if (TAILQ_FIRST(&vnode_free_list) != vp) {
1185 			TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
1186 			TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
1187 		}
1188 		splx(s);
1189 	}
1190 }
1191 
1192 /*
1193  * Lookup a vnode by device number.
1194  */
1195 int
1196 vfinddev(dev_t dev, enum vtype type, struct vnode **vpp)
1197 {
1198 	struct vnode *vp;
1199 	int rc =0;
1200 
1201 	SLIST_FOREACH(vp, &speclisth[SPECHASH(dev)], v_specnext) {
1202 		if (dev != vp->v_rdev || type != vp->v_type)
1203 			continue;
1204 		*vpp = vp;
1205 		rc = 1;
1206 		break;
1207 	}
1208 	return (rc);
1209 }
1210 
1211 /*
1212  * Revoke all the vnodes corresponding to the specified minor number
1213  * range (endpoints inclusive) of the specified major.
1214  */
1215 void
1216 vdevgone(int maj, int minl, int minh, enum vtype type)
1217 {
1218 	struct vnode *vp;
1219 	int mn;
1220 
1221 	for (mn = minl; mn <= minh; mn++)
1222 		if (vfinddev(makedev(maj, mn), type, &vp))
1223 			VOP_REVOKE(vp, REVOKEALL);
1224 }
1225 
1226 /*
1227  * Calculate the total number of references to a special device.
1228  */
1229 int
1230 vcount(struct vnode *vp)
1231 {
1232 	struct vnode *vq;
1233 	int count;
1234 
1235 loop:
1236 	if ((vp->v_flag & VALIASED) == 0)
1237 		return (vp->v_usecount);
1238 	count = 0;
1239 	SLIST_FOREACH(vq, vp->v_hashchain, v_specnext) {
1240 		if (vq->v_rdev != vp->v_rdev || vq->v_type != vp->v_type)
1241 			continue;
1242 		/*
1243 		 * Alias, but not in use, so flush it out.
1244 		 */
1245 		if (vq->v_usecount == 0 && vq != vp) {
1246 			vgone(vq);
1247 			goto loop;
1248 		}
1249 		count += vq->v_usecount;
1250 	}
1251 	return (count);
1252 }
1253 
1254 #if defined(DEBUG) || defined(DIAGNOSTIC)
1255 /*
1256  * Print out a description of a vnode.
1257  */
1258 static char *typename[] =
1259    { "VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD" };
1260 
1261 void
1262 vprint(char *label, struct vnode *vp)
1263 {
1264 	char buf[64];
1265 
1266 	if (label != NULL)
1267 		printf("%s: ", label);
1268 	printf("%p, type %s, use %u, write %u, hold %u,",
1269 		vp, typename[vp->v_type], vp->v_usecount, vp->v_writecount,
1270 		vp->v_holdcnt);
1271 	buf[0] = '\0';
1272 	if (vp->v_flag & VROOT)
1273 		strlcat(buf, "|VROOT", sizeof buf);
1274 	if (vp->v_flag & VTEXT)
1275 		strlcat(buf, "|VTEXT", sizeof buf);
1276 	if (vp->v_flag & VSYSTEM)
1277 		strlcat(buf, "|VSYSTEM", sizeof buf);
1278 	if (vp->v_flag & VXLOCK)
1279 		strlcat(buf, "|VXLOCK", sizeof buf);
1280 	if (vp->v_flag & VXWANT)
1281 		strlcat(buf, "|VXWANT", sizeof buf);
1282 	if (vp->v_bioflag & VBIOWAIT)
1283 		strlcat(buf, "|VBIOWAIT", sizeof buf);
1284 	if (vp->v_bioflag & VBIOONFREELIST)
1285 		strlcat(buf, "|VBIOONFREELIST", sizeof buf);
1286 	if (vp->v_bioflag & VBIOONSYNCLIST)
1287 		strlcat(buf, "|VBIOONSYNCLIST", sizeof buf);
1288 	if (vp->v_flag & VALIASED)
1289 		strlcat(buf, "|VALIASED", sizeof buf);
1290 	if (buf[0] != '\0')
1291 		printf(" flags (%s)", &buf[1]);
1292 	if (vp->v_data == NULL) {
1293 		printf("\n");
1294 	} else {
1295 		printf("\n\t");
1296 		VOP_PRINT(vp);
1297 	}
1298 }
1299 #endif /* DEBUG || DIAGNOSTIC */
1300 
1301 #ifdef DEBUG
1302 /*
1303  * List all of the locked vnodes in the system.
1304  * Called when debugging the kernel.
1305  */
1306 void
1307 printlockedvnodes(void)
1308 {
1309 	struct mount *mp;
1310 	struct vnode *vp;
1311 
1312 	printf("Locked vnodes\n");
1313 
1314 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
1315 		if (vfs_busy(mp, VB_READ|VB_NOWAIT))
1316 			continue;
1317 		TAILQ_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
1318 			if (VOP_ISLOCKED(vp))
1319 				vprint(NULL, vp);
1320 		}
1321 		vfs_unbusy(mp);
1322 	}
1323 
1324 }
1325 #endif
1326 
1327 /*
1328  * Top level filesystem related information gathering.
1329  */
1330 int
1331 vfs_sysctl(int *name, u_int namelen, void *oldp, size_t *oldlenp, void *newp,
1332     size_t newlen, struct proc *p)
1333 {
1334 	struct vfsconf *vfsp, *tmpvfsp;
1335 	int ret;
1336 
1337 	/* all sysctl names at this level are at least name and field */
1338 	if (namelen < 2)
1339 		return (ENOTDIR);		/* overloaded */
1340 
1341 	if (name[0] != VFS_GENERIC) {
1342 		vfsp = vfs_bytypenum(name[0]);
1343 		if (vfsp == NULL || vfsp->vfc_vfsops->vfs_sysctl == NULL)
1344 			return (EOPNOTSUPP);
1345 
1346 		return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1,
1347 		    oldp, oldlenp, newp, newlen, p));
1348 	}
1349 
1350 	switch (name[1]) {
1351 	case VFS_MAXTYPENUM:
1352 		return (sysctl_rdint(oldp, oldlenp, newp, maxvfsconf));
1353 
1354 	case VFS_CONF:
1355 		if (namelen < 3)
1356 			return (ENOTDIR);	/* overloaded */
1357 
1358 		vfsp = vfs_bytypenum(name[2]);
1359 		if (vfsp == NULL)
1360 			return (EOPNOTSUPP);
1361 
1362 		/* Make a copy, clear out kernel pointers */
1363 		tmpvfsp = malloc(sizeof(*tmpvfsp), M_TEMP, M_WAITOK|M_ZERO);
1364 		memcpy(tmpvfsp, vfsp, sizeof(*tmpvfsp));
1365 		tmpvfsp->vfc_vfsops = NULL;
1366 
1367 		ret = sysctl_rdstruct(oldp, oldlenp, newp, tmpvfsp,
1368 		    sizeof(struct vfsconf));
1369 
1370 		free(tmpvfsp, M_TEMP, sizeof(*tmpvfsp));
1371 		return (ret);
1372 	case VFS_BCACHESTAT:	/* buffer cache statistics */
1373 		ret = sysctl_rdstruct(oldp, oldlenp, newp, &bcstats,
1374 		    sizeof(struct bcachestats));
1375 		return(ret);
1376 	}
1377 	return (EOPNOTSUPP);
1378 }
1379 
1380 /*
1381  * Check to see if a filesystem is mounted on a block device.
1382  */
1383 int
1384 vfs_mountedon(struct vnode *vp)
1385 {
1386 	struct vnode *vq;
1387 	int error = 0;
1388 
1389 	if (vp->v_specmountpoint != NULL)
1390 		return (EBUSY);
1391 	if (vp->v_flag & VALIASED) {
1392 		SLIST_FOREACH(vq, vp->v_hashchain, v_specnext) {
1393 			if (vq->v_rdev != vp->v_rdev ||
1394 			    vq->v_type != vp->v_type)
1395 				continue;
1396 			if (vq->v_specmountpoint != NULL) {
1397 				error = EBUSY;
1398 				break;
1399 			}
1400 		}
1401 	}
1402 	return (error);
1403 }
1404 
1405 #ifdef NFSSERVER
1406 /*
1407  * Build hash lists of net addresses and hang them off the mount point.
1408  * Called by vfs_export() to set up the lists of export addresses.
1409  */
1410 int
1411 vfs_hang_addrlist(struct mount *mp, struct netexport *nep,
1412     struct export_args *argp)
1413 {
1414 	struct netcred *np;
1415 	struct radix_node_head *rnh;
1416 	int nplen, i;
1417 	struct radix_node *rn;
1418 	struct sockaddr *saddr, *smask = 0;
1419 	int error;
1420 
1421 	if (argp->ex_addrlen == 0) {
1422 		if (mp->mnt_flag & MNT_DEFEXPORTED)
1423 			return (EPERM);
1424 		np = &nep->ne_defexported;
1425 		/* fill in the kernel's ucred from userspace's xucred */
1426 		if ((error = crfromxucred(&np->netc_anon, &argp->ex_anon)))
1427 			return (error);
1428 		mp->mnt_flag |= MNT_DEFEXPORTED;
1429 		goto finish;
1430 	}
1431 	if (argp->ex_addrlen > MLEN || argp->ex_masklen > MLEN ||
1432 	    argp->ex_addrlen < 0 || argp->ex_masklen < 0)
1433 		return (EINVAL);
1434 	nplen = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen;
1435 	np = (struct netcred *)malloc(nplen, M_NETADDR, M_WAITOK|M_ZERO);
1436 	np->netc_len = nplen;
1437 	saddr = (struct sockaddr *)(np + 1);
1438 	error = copyin(argp->ex_addr, saddr, argp->ex_addrlen);
1439 	if (error)
1440 		goto out;
1441 	if (saddr->sa_len > argp->ex_addrlen)
1442 		saddr->sa_len = argp->ex_addrlen;
1443 	if (argp->ex_masklen) {
1444 		smask = (struct sockaddr *)((caddr_t)saddr + argp->ex_addrlen);
1445 		error = copyin(argp->ex_mask, smask, argp->ex_masklen);
1446 		if (error)
1447 			goto out;
1448 		if (smask->sa_len > argp->ex_masklen)
1449 			smask->sa_len = argp->ex_masklen;
1450 	}
1451 	/* fill in the kernel's ucred from userspace's xucred */
1452 	if ((error = crfromxucred(&np->netc_anon, &argp->ex_anon)))
1453 		goto out;
1454 	i = saddr->sa_family;
1455 	switch (i) {
1456 	case AF_INET:
1457 		if ((rnh = nep->ne_rtable_inet) == NULL) {
1458 			if (!rn_inithead((void **)&nep->ne_rtable_inet,
1459 			    offsetof(struct sockaddr_in, sin_addr))) {
1460 				error = ENOBUFS;
1461 				goto out;
1462 			}
1463 			rnh = nep->ne_rtable_inet;
1464 		}
1465 		break;
1466 	default:
1467 		error = EINVAL;
1468 		goto out;
1469 	}
1470 	rn = rn_addroute(saddr, smask, rnh, np->netc_rnodes, 0);
1471 	if (rn == 0 || np != (struct netcred *)rn) { /* already exists */
1472 		error = EPERM;
1473 		goto out;
1474 	}
1475 finish:
1476 	np->netc_exflags = argp->ex_flags;
1477 	return (0);
1478 out:
1479 	free(np, M_NETADDR, np->netc_len);
1480 	return (error);
1481 }
1482 
1483 int
1484 vfs_free_netcred(struct radix_node *rn, void *w, u_int id)
1485 {
1486 	struct radix_node_head *rnh = (struct radix_node_head *)w;
1487 	struct netcred * np = (struct netcred *)rn;
1488 
1489 	rn_delete(rn->rn_key, rn->rn_mask, rnh, NULL);
1490 	free(np, M_NETADDR, np->netc_len);
1491 	return (0);
1492 }
1493 
1494 /*
1495  * Free the net address hash lists that are hanging off the mount points.
1496  */
1497 void
1498 vfs_free_addrlist(struct netexport *nep)
1499 {
1500 	struct radix_node_head *rnh;
1501 
1502 	if ((rnh = nep->ne_rtable_inet) != NULL) {
1503 		rn_walktree(rnh, vfs_free_netcred, rnh);
1504 		free(rnh, M_RTABLE, sizeof(*rnh));
1505 		nep->ne_rtable_inet = NULL;
1506 	}
1507 }
1508 #endif /* NFSSERVER */
1509 
1510 int
1511 vfs_export(struct mount *mp, struct netexport *nep, struct export_args *argp)
1512 {
1513 #ifdef NFSSERVER
1514 	int error;
1515 
1516 	if (argp->ex_flags & MNT_DELEXPORT) {
1517 		vfs_free_addrlist(nep);
1518 		mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED);
1519 	}
1520 	if (argp->ex_flags & MNT_EXPORTED) {
1521 		if ((error = vfs_hang_addrlist(mp, nep, argp)) != 0)
1522 			return (error);
1523 		mp->mnt_flag |= MNT_EXPORTED;
1524 	}
1525 	return (0);
1526 #else
1527 	return (ENOTSUP);
1528 #endif /* NFSSERVER */
1529 }
1530 
1531 struct netcred *
1532 vfs_export_lookup(struct mount *mp, struct netexport *nep, struct mbuf *nam)
1533 {
1534 #ifdef NFSSERVER
1535 	struct netcred *np;
1536 	struct radix_node_head *rnh;
1537 	struct sockaddr *saddr;
1538 
1539 	np = NULL;
1540 	if (mp->mnt_flag & MNT_EXPORTED) {
1541 		/*
1542 		 * Lookup in the export list first.
1543 		 */
1544 		if (nam != NULL) {
1545 			saddr = mtod(nam, struct sockaddr *);
1546 			switch(saddr->sa_family) {
1547 			case AF_INET:
1548 				rnh = nep->ne_rtable_inet;
1549 				break;
1550 			default:
1551 				rnh = NULL;
1552 				break;
1553 			}
1554 			if (rnh != NULL)
1555 				np = (struct netcred *)rn_match(saddr, rnh);
1556 		}
1557 		/*
1558 		 * If no address match, use the default if it exists.
1559 		 */
1560 		if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED)
1561 			np = &nep->ne_defexported;
1562 	}
1563 	return (np);
1564 #else
1565 	return (NULL);
1566 #endif /* NFSSERVER */
1567 }
1568 
1569 /*
1570  * Do the usual access checking.
1571  * file_mode, uid and gid are from the vnode in question,
1572  * while acc_mode and cred are from the VOP_ACCESS parameter list
1573  */
1574 int
1575 vaccess(enum vtype type, mode_t file_mode, uid_t uid, gid_t gid,
1576     mode_t acc_mode, struct ucred *cred)
1577 {
1578 	mode_t mask;
1579 
1580 	/* User id 0 always gets read/write access. */
1581 	if (cred->cr_uid == 0) {
1582 		/* For VEXEC, at least one of the execute bits must be set. */
1583 		if ((acc_mode & VEXEC) && type != VDIR &&
1584 		    (file_mode & (S_IXUSR|S_IXGRP|S_IXOTH)) == 0)
1585 			return EACCES;
1586 		return 0;
1587 	}
1588 
1589 	mask = 0;
1590 
1591 	/* Otherwise, check the owner. */
1592 	if (cred->cr_uid == uid) {
1593 		if (acc_mode & VEXEC)
1594 			mask |= S_IXUSR;
1595 		if (acc_mode & VREAD)
1596 			mask |= S_IRUSR;
1597 		if (acc_mode & VWRITE)
1598 			mask |= S_IWUSR;
1599 		return (file_mode & mask) == mask ? 0 : EACCES;
1600 	}
1601 
1602 	/* Otherwise, check the groups. */
1603 	if (groupmember(gid, cred)) {
1604 		if (acc_mode & VEXEC)
1605 			mask |= S_IXGRP;
1606 		if (acc_mode & VREAD)
1607 			mask |= S_IRGRP;
1608 		if (acc_mode & VWRITE)
1609 			mask |= S_IWGRP;
1610 		return (file_mode & mask) == mask ? 0 : EACCES;
1611 	}
1612 
1613 	/* Otherwise, check everyone else. */
1614 	if (acc_mode & VEXEC)
1615 		mask |= S_IXOTH;
1616 	if (acc_mode & VREAD)
1617 		mask |= S_IROTH;
1618 	if (acc_mode & VWRITE)
1619 		mask |= S_IWOTH;
1620 	return (file_mode & mask) == mask ? 0 : EACCES;
1621 }
1622 
1623 int
1624 vnoperm(struct vnode *vp)
1625 {
1626 	if (vp->v_flag & VROOT || vp->v_mount == NULL)
1627 		return 0;
1628 
1629 	return (vp->v_mount->mnt_flag & MNT_NOPERM);
1630 }
1631 
1632 struct rwlock vfs_stall_lock = RWLOCK_INITIALIZER("vfs_stall");
1633 unsigned int vfs_stalling = 0;
1634 
1635 int
1636 vfs_stall(struct proc *p, int stall)
1637 {
1638 	struct mount *mp;
1639 	int allerror = 0, error;
1640 
1641 	if (stall) {
1642 		atomic_inc_int(&vfs_stalling);
1643 		rw_enter_write(&vfs_stall_lock);
1644 	}
1645 
1646 	/*
1647 	 * The loop variable mp is protected by vfs_busy() so that it cannot
1648 	 * be unmounted while VFS_SYNC() sleeps.  Traverse forward to keep the
1649 	 * lock order consistent with dounmount().
1650 	 */
1651 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
1652 		if (stall) {
1653 			error = vfs_busy(mp, VB_WRITE|VB_WAIT|VB_DUPOK);
1654 			if (error) {
1655 				printf("%s: busy\n", mp->mnt_stat.f_mntonname);
1656 				allerror = error;
1657 				continue;
1658 			}
1659 			uvm_vnp_sync(mp);
1660 			error = VFS_SYNC(mp, MNT_WAIT, stall, p->p_ucred, p);
1661 			if (error) {
1662 				printf("%s: failed to sync\n",
1663 				    mp->mnt_stat.f_mntonname);
1664 				vfs_unbusy(mp);
1665 				allerror = error;
1666 				continue;
1667 			}
1668 			mp->mnt_flag |= MNT_STALLED;
1669 		} else {
1670 			if (mp->mnt_flag & MNT_STALLED) {
1671 				vfs_unbusy(mp);
1672 				mp->mnt_flag &= ~MNT_STALLED;
1673 			}
1674 		}
1675 	}
1676 
1677 	if (!stall) {
1678 		rw_exit_write(&vfs_stall_lock);
1679 		atomic_dec_int(&vfs_stalling);
1680 	}
1681 
1682 	return (allerror);
1683 }
1684 
1685 void
1686 vfs_stall_barrier(void)
1687 {
1688 	if (__predict_false(vfs_stalling)) {
1689 		rw_enter_read(&vfs_stall_lock);
1690 		rw_exit_read(&vfs_stall_lock);
1691 	}
1692 }
1693 
1694 /*
1695  * Unmount all file systems.
1696  * We traverse the list in reverse order under the assumption that doing so
1697  * will avoid needing to worry about dependencies.
1698  */
1699 void
1700 vfs_unmountall(void)
1701 {
1702 	struct mount *mp, *nmp;
1703 	int allerror, error, again = 1;
1704 
1705  retry:
1706 	allerror = 0;
1707 	TAILQ_FOREACH_REVERSE_SAFE(mp, &mountlist, mntlist, mnt_list, nmp) {
1708 		if (vfs_busy(mp, VB_WRITE|VB_NOWAIT))
1709 			continue;
1710 		/* XXX Here is a race, the next pointer is not locked. */
1711 		if ((error = dounmount(mp, MNT_FORCE, curproc)) != 0) {
1712 			printf("unmount of %s failed with error %d\n",
1713 			    mp->mnt_stat.f_mntonname, error);
1714 			allerror = 1;
1715 		}
1716 	}
1717 
1718 	if (allerror) {
1719 		printf("WARNING: some file systems would not unmount\n");
1720 		if (again) {
1721 			printf("retrying\n");
1722 			again = 0;
1723 			goto retry;
1724 		}
1725 	}
1726 }
1727 
1728 /*
1729  * Sync and unmount file systems before shutting down.
1730  */
1731 void
1732 vfs_shutdown(struct proc *p)
1733 {
1734 #ifdef ACCOUNTING
1735 	acct_shutdown();
1736 #endif
1737 
1738 	printf("syncing disks...");
1739 
1740 	if (panicstr == 0) {
1741 		/* Sync before unmount, in case we hang on something. */
1742 		sys_sync(p, NULL, NULL);
1743 		vfs_unmountall();
1744 	}
1745 
1746 #if NSOFTRAID > 0
1747 	sr_quiesce();
1748 #endif
1749 
1750 	if (vfs_syncwait(p, 1))
1751 		printf(" giving up\n");
1752 	else
1753 		printf(" done\n");
1754 }
1755 
1756 /*
1757  * perform sync() operation and wait for buffers to flush.
1758  */
1759 int
1760 vfs_syncwait(struct proc *p, int verbose)
1761 {
1762 	struct buf *bp;
1763 	int iter, nbusy, dcount, s;
1764 #ifdef MULTIPROCESSOR
1765 	int hold_count;
1766 #endif
1767 
1768 	sys_sync(p, NULL, NULL);
1769 
1770 	/* Wait for sync to finish. */
1771 	dcount = 10000;
1772 	for (iter = 0; iter < 20; iter++) {
1773 		nbusy = 0;
1774 		LIST_FOREACH(bp, &bufhead, b_list) {
1775 			if ((bp->b_flags & (B_BUSY|B_INVAL|B_READ)) == B_BUSY)
1776 				nbusy++;
1777 			/*
1778 			 * With soft updates, some buffers that are
1779 			 * written will be remarked as dirty until other
1780 			 * buffers are written.
1781 			 */
1782 			if (bp->b_flags & B_DELWRI) {
1783 				s = splbio();
1784 				bremfree(bp);
1785 				buf_acquire(bp);
1786 				splx(s);
1787 				nbusy++;
1788 				bawrite(bp);
1789 				if (dcount-- <= 0) {
1790 					if (verbose)
1791 						printf("softdep ");
1792 					return 1;
1793 				}
1794 			}
1795 		}
1796 		if (nbusy == 0)
1797 			break;
1798 		if (verbose)
1799 			printf("%d ", nbusy);
1800 #ifdef MULTIPROCESSOR
1801 		if (_kernel_lock_held())
1802 			hold_count = __mp_release_all(&kernel_lock);
1803 		else
1804 			hold_count = 0;
1805 #endif
1806 		DELAY(40000 * iter);
1807 #ifdef MULTIPROCESSOR
1808 		if (hold_count)
1809 			__mp_acquire_count(&kernel_lock, hold_count);
1810 #endif
1811 	}
1812 
1813 	return nbusy;
1814 }
1815 
1816 /*
1817  * posix file system related system variables.
1818  */
1819 int
1820 fs_posix_sysctl(int *name, u_int namelen, void *oldp, size_t *oldlenp,
1821     void *newp, size_t newlen, struct proc *p)
1822 {
1823 	/* all sysctl names at this level are terminal */
1824 	if (namelen != 1)
1825 		return (ENOTDIR);
1826 
1827 	switch (name[0]) {
1828 	case FS_POSIX_SETUID:
1829 		if (newp && securelevel > 0)
1830 			return (EPERM);
1831 		return(sysctl_int(oldp, oldlenp, newp, newlen, &suid_clear));
1832 	default:
1833 		return (EOPNOTSUPP);
1834 	}
1835 	/* NOTREACHED */
1836 }
1837 
1838 /*
1839  * file system related system variables.
1840  */
1841 int
1842 fs_sysctl(int *name, u_int namelen, void *oldp, size_t *oldlenp, void *newp,
1843     size_t newlen, struct proc *p)
1844 {
1845 	sysctlfn *fn;
1846 
1847 	switch (name[0]) {
1848 	case FS_POSIX:
1849 		fn = fs_posix_sysctl;
1850 		break;
1851 	default:
1852 		return (EOPNOTSUPP);
1853 	}
1854 	return (*fn)(name + 1, namelen - 1, oldp, oldlenp, newp, newlen, p);
1855 }
1856 
1857 
1858 /*
1859  * Routines dealing with vnodes and buffers
1860  */
1861 
1862 /*
1863  * Wait for all outstanding I/Os to complete
1864  *
1865  * Manipulates v_numoutput. Must be called at splbio()
1866  */
1867 int
1868 vwaitforio(struct vnode *vp, int slpflag, char *wmesg, uint64_t timeo)
1869 {
1870 	int error = 0;
1871 
1872 	splassert(IPL_BIO);
1873 
1874 	while (vp->v_numoutput) {
1875 		vp->v_bioflag |= VBIOWAIT;
1876 		error = tsleep_nsec(&vp->v_numoutput,
1877 		    slpflag | (PRIBIO + 1), wmesg, timeo);
1878 		if (error)
1879 			break;
1880 	}
1881 
1882 	return (error);
1883 }
1884 
1885 /*
1886  * Update outstanding I/O count and do wakeup if requested.
1887  *
1888  * Manipulates v_numoutput. Must be called at splbio()
1889  */
1890 void
1891 vwakeup(struct vnode *vp)
1892 {
1893 	splassert(IPL_BIO);
1894 
1895 	if (vp != NULL) {
1896 		if (vp->v_numoutput-- == 0)
1897 			panic("vwakeup: neg numoutput");
1898 		if ((vp->v_bioflag & VBIOWAIT) && vp->v_numoutput == 0) {
1899 			vp->v_bioflag &= ~VBIOWAIT;
1900 			wakeup(&vp->v_numoutput);
1901 		}
1902 	}
1903 }
1904 
1905 /*
1906  * Flush out and invalidate all buffers associated with a vnode.
1907  * Called with the underlying object locked.
1908  */
1909 int
1910 vinvalbuf(struct vnode *vp, int flags, struct ucred *cred, struct proc *p,
1911     int slpflag, uint64_t slptimeo)
1912 {
1913 	struct buf *bp;
1914 	struct buf *nbp, *blist;
1915 	int s, error;
1916 
1917 #ifdef VFSLCKDEBUG
1918 	if ((vp->v_flag & VLOCKSWORK) && !VOP_ISLOCKED(vp))
1919 		panic("%s: vp isn't locked, vp %p", __func__, vp);
1920 #endif
1921 
1922 	if (flags & V_SAVE) {
1923 		s = splbio();
1924 		vwaitforio(vp, 0, "vinvalbuf", INFSLP);
1925 		if (!LIST_EMPTY(&vp->v_dirtyblkhd)) {
1926 			splx(s);
1927 			if ((error = VOP_FSYNC(vp, cred, MNT_WAIT, p)) != 0)
1928 				return (error);
1929 			s = splbio();
1930 			if (vp->v_numoutput > 0 ||
1931 			    !LIST_EMPTY(&vp->v_dirtyblkhd))
1932 				panic("%s: dirty bufs, vp %p", __func__, vp);
1933 		}
1934 		splx(s);
1935 	}
1936 loop:
1937 	s = splbio();
1938 	for (;;) {
1939 		int count = 0;
1940 		if ((blist = LIST_FIRST(&vp->v_cleanblkhd)) &&
1941 		    (flags & V_SAVEMETA))
1942 			while (blist && blist->b_lblkno < 0)
1943 				blist = LIST_NEXT(blist, b_vnbufs);
1944 		if (blist == NULL &&
1945 		    (blist = LIST_FIRST(&vp->v_dirtyblkhd)) &&
1946 		    (flags & V_SAVEMETA))
1947 			while (blist && blist->b_lblkno < 0)
1948 				blist = LIST_NEXT(blist, b_vnbufs);
1949 		if (!blist)
1950 			break;
1951 
1952 		for (bp = blist; bp; bp = nbp) {
1953 			nbp = LIST_NEXT(bp, b_vnbufs);
1954 			if (flags & V_SAVEMETA && bp->b_lblkno < 0)
1955 				continue;
1956 			if (bp->b_flags & B_BUSY) {
1957 				bp->b_flags |= B_WANTED;
1958 				error = tsleep_nsec(bp, slpflag | (PRIBIO + 1),
1959 				    "vinvalbuf", slptimeo);
1960 				if (error) {
1961 					splx(s);
1962 					return (error);
1963 				}
1964 				break;
1965 			}
1966 			bremfree(bp);
1967 			/*
1968 			 * XXX Since there are no node locks for NFS, I believe
1969 			 * there is a slight chance that a delayed write will
1970 			 * occur while sleeping just above, so check for it.
1971 			 */
1972 			if ((bp->b_flags & B_DELWRI) && (flags & V_SAVE)) {
1973 				buf_acquire(bp);
1974 				splx(s);
1975 				(void) VOP_BWRITE(bp);
1976 				goto loop;
1977 			}
1978 			buf_acquire_nomap(bp);
1979 			bp->b_flags |= B_INVAL;
1980 			brelse(bp);
1981 			count++;
1982 			/*
1983 			 * XXX Temporary workaround XXX
1984 			 *
1985 			 * If this is a gigantisch vnode and we are
1986 			 * trashing a ton of buffers, drop the lock
1987 			 * and yield every so often. The longer term
1988 			 * fix is to add a separate list for these
1989 			 * invalid buffers so we don't have to do the
1990 			 * work to free these here.
1991 			 */
1992 			if (count > 100) {
1993 				splx(s);
1994 				sched_pause(yield);
1995 				goto loop;
1996 			}
1997 		}
1998 	}
1999 	if (!(flags & V_SAVEMETA) &&
2000 	    (!LIST_EMPTY(&vp->v_dirtyblkhd) || !LIST_EMPTY(&vp->v_cleanblkhd)))
2001 		panic("%s: flush failed, vp %p", __func__, vp);
2002 	splx(s);
2003 	return (0);
2004 }
2005 
2006 void
2007 vflushbuf(struct vnode *vp, int sync)
2008 {
2009 	struct buf *bp, *nbp;
2010 	int s;
2011 
2012 loop:
2013 	s = splbio();
2014 	LIST_FOREACH_SAFE(bp, &vp->v_dirtyblkhd, b_vnbufs, nbp) {
2015 		if ((bp->b_flags & B_BUSY))
2016 			continue;
2017 		if ((bp->b_flags & B_DELWRI) == 0)
2018 			panic("vflushbuf: not dirty");
2019 		bremfree(bp);
2020 		buf_acquire(bp);
2021 		splx(s);
2022 		/*
2023 		 * Wait for I/O associated with indirect blocks to complete,
2024 		 * since there is no way to quickly wait for them below.
2025 		 */
2026 		if (bp->b_vp == vp || sync == 0)
2027 			(void) bawrite(bp);
2028 		else
2029 			(void) bwrite(bp);
2030 		goto loop;
2031 	}
2032 	if (sync == 0) {
2033 		splx(s);
2034 		return;
2035 	}
2036 	vwaitforio(vp, 0, "vflushbuf", INFSLP);
2037 	if (!LIST_EMPTY(&vp->v_dirtyblkhd)) {
2038 		splx(s);
2039 #ifdef DIAGNOSTIC
2040 		vprint("vflushbuf: dirty", vp);
2041 #endif
2042 		goto loop;
2043 	}
2044 	splx(s);
2045 }
2046 
2047 /*
2048  * Associate a buffer with a vnode.
2049  *
2050  * Manipulates buffer vnode queues. Must be called at splbio().
2051  */
2052 void
2053 bgetvp(struct vnode *vp, struct buf *bp)
2054 {
2055 	splassert(IPL_BIO);
2056 
2057 
2058 	if (bp->b_vp)
2059 		panic("bgetvp: not free");
2060 	vhold(vp);
2061 	bp->b_vp = vp;
2062 	if (vp->v_type == VBLK || vp->v_type == VCHR)
2063 		bp->b_dev = vp->v_rdev;
2064 	else
2065 		bp->b_dev = NODEV;
2066 	/*
2067 	 * Insert onto list for new vnode.
2068 	 */
2069 	bufinsvn(bp, &vp->v_cleanblkhd);
2070 }
2071 
2072 /*
2073  * Disassociate a buffer from a vnode.
2074  *
2075  * Manipulates vnode buffer queues. Must be called at splbio().
2076  */
2077 void
2078 brelvp(struct buf *bp)
2079 {
2080 	struct vnode *vp;
2081 
2082 	splassert(IPL_BIO);
2083 
2084 	if ((vp = bp->b_vp) == (struct vnode *) 0)
2085 		panic("brelvp: NULL");
2086 	/*
2087 	 * Delete from old vnode list, if on one.
2088 	 */
2089 	if (LIST_NEXT(bp, b_vnbufs) != NOLIST)
2090 		bufremvn(bp);
2091 	if ((vp->v_bioflag & VBIOONSYNCLIST) &&
2092 	    LIST_EMPTY(&vp->v_dirtyblkhd)) {
2093 		vp->v_bioflag &= ~VBIOONSYNCLIST;
2094 		LIST_REMOVE(vp, v_synclist);
2095 	}
2096 	bp->b_vp = NULL;
2097 
2098 	vdrop(vp);
2099 }
2100 
2101 /*
2102  * Replaces the current vnode associated with the buffer, if any,
2103  * with a new vnode.
2104  *
2105  * If an output I/O is pending on the buffer, the old vnode
2106  * I/O count is adjusted.
2107  *
2108  * Ignores vnode buffer queues. Must be called at splbio().
2109  */
2110 void
2111 buf_replacevnode(struct buf *bp, struct vnode *newvp)
2112 {
2113 	struct vnode *oldvp = bp->b_vp;
2114 
2115 	splassert(IPL_BIO);
2116 
2117 	if (oldvp)
2118 		brelvp(bp);
2119 
2120 	if ((bp->b_flags & (B_READ | B_DONE)) == 0) {
2121 		newvp->v_numoutput++;	/* put it on swapdev */
2122 		vwakeup(oldvp);
2123 	}
2124 
2125 	bgetvp(newvp, bp);
2126 	bufremvn(bp);
2127 }
2128 
2129 /*
2130  * Used to assign buffers to the appropriate clean or dirty list on
2131  * the vnode and to add newly dirty vnodes to the appropriate
2132  * filesystem syncer list.
2133  *
2134  * Manipulates vnode buffer queues. Must be called at splbio().
2135  */
2136 void
2137 reassignbuf(struct buf *bp)
2138 {
2139 	struct buflists *listheadp;
2140 	int delay;
2141 	struct vnode *vp = bp->b_vp;
2142 
2143 	splassert(IPL_BIO);
2144 
2145 	/*
2146 	 * Delete from old vnode list, if on one.
2147 	 */
2148 	if (LIST_NEXT(bp, b_vnbufs) != NOLIST)
2149 		bufremvn(bp);
2150 
2151 	/*
2152 	 * If dirty, put on list of dirty buffers;
2153 	 * otherwise insert onto list of clean buffers.
2154 	 */
2155 	if ((bp->b_flags & B_DELWRI) == 0) {
2156 		listheadp = &vp->v_cleanblkhd;
2157 		if ((vp->v_bioflag & VBIOONSYNCLIST) &&
2158 		    LIST_EMPTY(&vp->v_dirtyblkhd)) {
2159 			vp->v_bioflag &= ~VBIOONSYNCLIST;
2160 			LIST_REMOVE(vp, v_synclist);
2161 		}
2162 	} else {
2163 		listheadp = &vp->v_dirtyblkhd;
2164 		if ((vp->v_bioflag & VBIOONSYNCLIST) == 0) {
2165 			switch (vp->v_type) {
2166 			case VDIR:
2167 				delay = syncdelay / 2;
2168 				break;
2169 			case VBLK:
2170 				if (vp->v_specmountpoint != NULL) {
2171 					delay = syncdelay / 3;
2172 					break;
2173 				}
2174 				/* FALLTHROUGH */
2175 			default:
2176 				delay = syncdelay;
2177 			}
2178 			vn_syncer_add_to_worklist(vp, delay);
2179 		}
2180 	}
2181 	bufinsvn(bp, listheadp);
2182 }
2183 
2184 /*
2185  * Check if vnode represents a disk device
2186  */
2187 int
2188 vn_isdisk(struct vnode *vp, int *errp)
2189 {
2190 	if (vp->v_type != VBLK && vp->v_type != VCHR)
2191 		return (0);
2192 
2193 	return (1);
2194 }
2195 
2196 #ifdef DDB
2197 #include <machine/db_machdep.h>
2198 #include <ddb/db_interface.h>
2199 
2200 void
2201 vfs_buf_print(void *b, int full,
2202     int (*pr)(const char *, ...) __attribute__((__format__(__kprintf__,1,2))))
2203 {
2204 	struct buf *bp = b;
2205 
2206 	(*pr)("  vp %p lblkno 0x%llx blkno 0x%llx dev 0x%x\n"
2207 	      "  proc %p error %d flags %lb\n",
2208 	    bp->b_vp, (int64_t)bp->b_lblkno, (int64_t)bp->b_blkno, bp->b_dev,
2209 	    bp->b_proc, bp->b_error, bp->b_flags, B_BITS);
2210 
2211 	(*pr)("  bufsize 0x%lx bcount 0x%lx resid 0x%lx\n"
2212 	      "  data %p saveaddr %p dep %p iodone %p\n",
2213 	    bp->b_bufsize, bp->b_bcount, (long)bp->b_resid,
2214 	    bp->b_data, bp->b_saveaddr,
2215 	    LIST_FIRST(&bp->b_dep), bp->b_iodone);
2216 
2217 	(*pr)("  dirty {off 0x%x end 0x%x} valid {off 0x%x end 0x%x}\n",
2218 	    bp->b_dirtyoff, bp->b_dirtyend, bp->b_validoff, bp->b_validend);
2219 
2220 #ifdef FFS_SOFTUPDATES
2221 	if (full)
2222 		softdep_print(bp, full, pr);
2223 #endif
2224 }
2225 
2226 const char *vtypes[] = { VTYPE_NAMES };
2227 const char *vtags[] = { VTAG_NAMES };
2228 
2229 void
2230 vfs_vnode_print(void *v, int full,
2231     int (*pr)(const char *, ...) __attribute__((__format__(__kprintf__,1,2))))
2232 {
2233 	struct vnode *vp = v;
2234 
2235 	(*pr)("tag %s(%d) type %s(%d) mount %p typedata %p\n",
2236 	      (u_int)vp->v_tag >= nitems(vtags)? "<unk>":vtags[vp->v_tag],
2237 	      vp->v_tag,
2238 	      (u_int)vp->v_type >= nitems(vtypes)? "<unk>":vtypes[vp->v_type],
2239 	      vp->v_type, vp->v_mount, vp->v_mountedhere);
2240 
2241 	(*pr)("data %p usecount %d writecount %d holdcnt %d numoutput %d\n",
2242 	      vp->v_data, vp->v_usecount, vp->v_writecount,
2243 	      vp->v_holdcnt, vp->v_numoutput);
2244 
2245 	/* uvm_object_printit(&vp->v_uobj, full, pr); */
2246 
2247 	if (full) {
2248 		struct buf *bp;
2249 
2250 		(*pr)("clean bufs:\n");
2251 		LIST_FOREACH(bp, &vp->v_cleanblkhd, b_vnbufs) {
2252 			(*pr)(" bp %p\n", bp);
2253 			vfs_buf_print(bp, full, pr);
2254 		}
2255 
2256 		(*pr)("dirty bufs:\n");
2257 		LIST_FOREACH(bp, &vp->v_dirtyblkhd, b_vnbufs) {
2258 			(*pr)(" bp %p\n", bp);
2259 			vfs_buf_print(bp, full, pr);
2260 		}
2261 	}
2262 }
2263 
2264 void
2265 vfs_mount_print(struct mount *mp, int full,
2266     int (*pr)(const char *, ...) __attribute__((__format__(__kprintf__,1,2))))
2267 {
2268 	struct vfsconf *vfc = mp->mnt_vfc;
2269 	struct vnode *vp;
2270 	int cnt;
2271 
2272 	(*pr)("flags %b\nvnodecovered %p syncer %p data %p\n",
2273 	    mp->mnt_flag, MNT_BITS,
2274 	    mp->mnt_vnodecovered, mp->mnt_syncer, mp->mnt_data);
2275 
2276 	(*pr)("vfsconf: ops %p name \"%s\" num %d ref %u flags 0x%x\n",
2277 	    vfc->vfc_vfsops, vfc->vfc_name, vfc->vfc_typenum,
2278 	    vfc->vfc_refcount, vfc->vfc_flags);
2279 
2280 	(*pr)("statvfs cache: bsize %x iosize %x\n"
2281 	    "blocks %llu free %llu avail %lld\n",
2282 	    mp->mnt_stat.f_bsize, mp->mnt_stat.f_iosize, mp->mnt_stat.f_blocks,
2283 	    mp->mnt_stat.f_bfree, mp->mnt_stat.f_bavail);
2284 
2285 	(*pr)("  files %llu ffiles %llu favail %lld\n", mp->mnt_stat.f_files,
2286 	    mp->mnt_stat.f_ffree, mp->mnt_stat.f_favail);
2287 
2288 	(*pr)("  f_fsidx {0x%x, 0x%x} owner %u ctime 0x%llx\n",
2289 	    mp->mnt_stat.f_fsid.val[0], mp->mnt_stat.f_fsid.val[1],
2290 	    mp->mnt_stat.f_owner, mp->mnt_stat.f_ctime);
2291 
2292 	(*pr)("  syncwrites %llu asyncwrites = %llu\n",
2293 	    mp->mnt_stat.f_syncwrites, mp->mnt_stat.f_asyncwrites);
2294 
2295 	(*pr)("  syncreads %llu asyncreads = %llu\n",
2296 	    mp->mnt_stat.f_syncreads, mp->mnt_stat.f_asyncreads);
2297 
2298 	(*pr)("  fstype \"%s\" mnton \"%s\" mntfrom \"%s\" mntspec \"%s\"\n",
2299 	    mp->mnt_stat.f_fstypename, mp->mnt_stat.f_mntonname,
2300 	    mp->mnt_stat.f_mntfromname, mp->mnt_stat.f_mntfromspec);
2301 
2302 	(*pr)("locked vnodes:");
2303 	/* XXX would take mountlist lock, except ddb has no context */
2304 	cnt = 0;
2305 	TAILQ_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
2306 		if (VOP_ISLOCKED(vp)) {
2307 			if (cnt == 0)
2308 				(*pr)("\n  %p", vp);
2309 			else if ((cnt % (72 / (sizeof(void *) * 2 + 4))) == 0)
2310 				(*pr)(",\n  %p", vp);
2311 			else
2312 				(*pr)(", %p", vp);
2313 			cnt++;
2314 		}
2315 	}
2316 	(*pr)("\n");
2317 
2318 	if (full) {
2319 		(*pr)("all vnodes:");
2320 		/* XXX would take mountlist lock, except ddb has no context */
2321 		cnt = 0;
2322 		TAILQ_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
2323 			if (cnt == 0)
2324 				(*pr)("\n  %p", vp);
2325 			else if ((cnt % (72 / (sizeof(void *) * 2 + 4))) == 0)
2326 				(*pr)(",\n  %p", vp);
2327 			else
2328 				(*pr)(", %p", vp);
2329 			cnt++;
2330 		}
2331 		(*pr)("\n");
2332 	}
2333 }
2334 #endif /* DDB */
2335 
2336 void
2337 copy_statfs_info(struct statfs *sbp, const struct mount *mp)
2338 {
2339 	const struct statfs *mbp;
2340 
2341 	strncpy(sbp->f_fstypename, mp->mnt_vfc->vfc_name, MFSNAMELEN);
2342 
2343 	if (sbp == (mbp = &mp->mnt_stat))
2344 		return;
2345 
2346 	sbp->f_fsid = mbp->f_fsid;
2347 	sbp->f_owner = mbp->f_owner;
2348 	sbp->f_flags = mbp->f_flags;
2349 	sbp->f_syncwrites = mbp->f_syncwrites;
2350 	sbp->f_asyncwrites = mbp->f_asyncwrites;
2351 	sbp->f_syncreads = mbp->f_syncreads;
2352 	sbp->f_asyncreads = mbp->f_asyncreads;
2353 	sbp->f_namemax = mbp->f_namemax;
2354 	memcpy(sbp->f_mntonname, mp->mnt_stat.f_mntonname, MNAMELEN);
2355 	memcpy(sbp->f_mntfromname, mp->mnt_stat.f_mntfromname, MNAMELEN);
2356 	memcpy(sbp->f_mntfromspec, mp->mnt_stat.f_mntfromspec, MNAMELEN);
2357 	memcpy(&sbp->mount_info, &mp->mnt_stat.mount_info,
2358 	    sizeof(union mount_info));
2359 }
2360