xref: /dflybsd-src/sys/kern/vfs_mount.c (revision 17ea22213f86a5c5966c1e6bf8e95f022ebb92b9)
1 /*
2  * Copyright (c) 2004 The DragonFly Project.  All rights reserved.
3  *
4  * This code is derived from software contributed to The DragonFly Project
5  * by Matthew Dillon <dillon@backplane.com>
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  *
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in
15  *    the documentation and/or other materials provided with the
16  *    distribution.
17  * 3. Neither the name of The DragonFly Project nor the names of its
18  *    contributors may be used to endorse or promote products derived
19  *    from this software without specific, prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
25  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
27  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
29  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  * Copyright (c) 1989, 1993
35  *	The Regents of the University of California.  All rights reserved.
36  * (c) UNIX System Laboratories, Inc.
37  * All or some portions of this file are derived from material licensed
38  * to the University of California by American Telephone and Telegraph
39  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
40  * the permission of UNIX System Laboratories, Inc.
41  *
42  * Redistribution and use in source and binary forms, with or without
43  * modification, are permitted provided that the following conditions
44  * are met:
45  * 1. Redistributions of source code must retain the above copyright
46  *    notice, this list of conditions and the following disclaimer.
47  * 2. Redistributions in binary form must reproduce the above copyright
48  *    notice, this list of conditions and the following disclaimer in the
49  *    documentation and/or other materials provided with the distribution.
50  * 3. All advertising materials mentioning features or use of this software
51  *    must display the following acknowledgement:
52  *	This product includes software developed by the University of
53  *	California, Berkeley and its contributors.
54  * 4. Neither the name of the University nor the names of its contributors
55  *    may be used to endorse or promote products derived from this software
56  *    without specific prior written permission.
57  *
58  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
59  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
60  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
61  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
62  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
63  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
64  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
65  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
66  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
67  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
68  * SUCH DAMAGE.
69  *
70  * $DragonFly: src/sys/kern/vfs_mount.c,v 1.4 2004/12/29 02:40:02 dillon Exp $
71  */
72 
73 /*
74  * External virtual filesystem routines
75  */
76 #include "opt_ddb.h"
77 
78 #include <sys/param.h>
79 #include <sys/systm.h>
80 #include <sys/kernel.h>
81 #include <sys/malloc.h>
82 #include <sys/mount.h>
83 #include <sys/proc.h>
84 #include <sys/vnode.h>
85 #include <sys/buf.h>
86 #include <sys/eventhandler.h>
87 #include <sys/kthread.h>
88 #include <sys/sysctl.h>
89 
90 #include <machine/limits.h>
91 
92 #include <sys/buf2.h>
93 #include <sys/thread2.h>
94 
95 #include <vm/vm.h>
96 #include <vm/vm_object.h>
97 
98 static int vnlru_nowhere = 0;
99 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW,
100 	    &vnlru_nowhere, 0,
101 	    "Number of times the vnlru process ran without success");
102 
103 
104 static struct lwkt_token mntid_token;
105 
106 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist); /* mounted fs */
107 struct lwkt_token mountlist_token;
108 struct lwkt_token mntvnode_token;
109 
110 
111 /*
112  * Called from vfsinit()
113  */
114 void
115 vfs_mount_init(void)
116 {
117 	lwkt_token_init(&mountlist_token);
118 	lwkt_token_init(&mntvnode_token);
119 	lwkt_token_init(&mntid_token);
120 }
121 
122 /*
123  * Allocate a new vnode and associate it with a tag, mount point, and
124  * operations vector.
125  *
126  * A VX locked and refd vnode is returned.  The caller should setup the
127  * remaining fields and vx_put() or, if he wishes to leave a vref,
128  * vx_unlock() the vnode.
129  */
130 int
131 getnewvnode(enum vtagtype tag, struct mount *mp,
132 		struct vnode **vpp, int lktimeout, int lkflags)
133 {
134 	struct vnode *vp;
135 
136 	KKASSERT(mp != NULL);
137 
138 	vp = allocvnode(lktimeout, lkflags);
139 	vp->v_tag = tag;
140 	vp->v_data = NULL;
141 
142 	/*
143 	 * By default the vnode is assigned the mount point's normal
144 	 * operations vector.
145 	 */
146 	vp->v_ops = &mp->mnt_vn_use_ops;
147 
148 	/*
149 	 * Placing the vnode on the mount point's queue makes it visible.
150 	 * VNON prevents it from being messed with, however.
151 	 */
152 	insmntque(vp, mp);
153 	vfs_object_create(vp, curthread);
154 
155 	/*
156 	 * A VX locked & refd vnode is returned.
157 	 */
158 	*vpp = vp;
159 	return (0);
160 }
161 
162 /*
163  * This function creates vnodes with special operations vectors.  The
164  * mount point is optional.
165  *
166  * This routine is being phased out.
167  */
168 int
169 getspecialvnode(enum vtagtype tag, struct mount *mp,
170 		struct vop_ops **ops_pp,
171 		struct vnode **vpp, int lktimeout, int lkflags)
172 {
173 	struct vnode *vp;
174 
175 	vp = allocvnode(lktimeout, lkflags);
176 	vp->v_tag = tag;
177 	vp->v_data = NULL;
178 	vp->v_ops = ops_pp;
179 
180 	/*
181 	 * Placing the vnode on the mount point's queue makes it visible.
182 	 * VNON prevents it from being messed with, however.
183 	 */
184 	insmntque(vp, mp);
185 	vfs_object_create(vp, curthread);
186 
187 	/*
188 	 * A VX locked & refd vnode is returned.
189 	 */
190 	*vpp = vp;
191 	return (0);
192 }
193 
194 /*
195  * Mark a mount point as busy. Used to synchronize access and to delay
196  * unmounting. Interlock is not released on failure.
197  */
198 int
199 vfs_busy(struct mount *mp, int flags,
200 	lwkt_tokref_t interlkp, struct thread *td)
201 {
202 	int lkflags;
203 
204 	if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
205 		if (flags & LK_NOWAIT)
206 			return (ENOENT);
207 		mp->mnt_kern_flag |= MNTK_MWAIT;
208 		/*
209 		 * Since all busy locks are shared except the exclusive
210 		 * lock granted when unmounting, the only place that a
211 		 * wakeup needs to be done is at the release of the
212 		 * exclusive lock at the end of dounmount.
213 		 *
214 		 * note: interlkp is a serializer and thus can be safely
215 		 * held through any sleep
216 		 */
217 		tsleep((caddr_t)mp, 0, "vfs_busy", 0);
218 		return (ENOENT);
219 	}
220 	lkflags = LK_SHARED | LK_NOPAUSE;
221 	if (interlkp)
222 		lkflags |= LK_INTERLOCK;
223 	if (lockmgr(&mp->mnt_lock, lkflags, interlkp, td))
224 		panic("vfs_busy: unexpected lock failure");
225 	return (0);
226 }
227 
228 /*
229  * Free a busy filesystem.
230  */
231 void
232 vfs_unbusy(struct mount *mp, struct thread *td)
233 {
234 	lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td);
235 }
236 
237 /*
238  * Lookup a filesystem type, and if found allocate and initialize
239  * a mount structure for it.
240  *
241  * Devname is usually updated by mount(8) after booting.
242  */
243 int
244 vfs_rootmountalloc(char *fstypename, char *devname, struct mount **mpp)
245 {
246 	struct thread *td = curthread;	/* XXX */
247 	struct vfsconf *vfsp;
248 	struct mount *mp;
249 
250 	if (fstypename == NULL)
251 		return (ENODEV);
252 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
253 		if (!strcmp(vfsp->vfc_name, fstypename))
254 			break;
255 	}
256 	if (vfsp == NULL)
257 		return (ENODEV);
258 	mp = malloc(sizeof(struct mount), M_MOUNT, M_WAITOK);
259 	bzero((char *)mp, (u_long)sizeof(struct mount));
260 	lockinit(&mp->mnt_lock, 0, "vfslock", VLKTIMEOUT, LK_NOPAUSE);
261 	vfs_busy(mp, LK_NOWAIT, NULL, td);
262 	TAILQ_INIT(&mp->mnt_nvnodelist);
263 	TAILQ_INIT(&mp->mnt_reservedvnlist);
264 	TAILQ_INIT(&mp->mnt_jlist);
265 	mp->mnt_nvnodelistsize = 0;
266 	mp->mnt_vfc = vfsp;
267 	mp->mnt_op = vfsp->vfc_vfsops;
268 	mp->mnt_flag = MNT_RDONLY;
269 	mp->mnt_vnodecovered = NULLVP;
270 	vfsp->vfc_refcount++;
271 	mp->mnt_iosize_max = DFLTPHYS;
272 	mp->mnt_stat.f_type = vfsp->vfc_typenum;
273 	mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
274 	strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
275 	mp->mnt_stat.f_mntonname[0] = '/';
276 	mp->mnt_stat.f_mntonname[1] = 0;
277 	(void) copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 0);
278 	*mpp = mp;
279 	return (0);
280 }
281 
282 /*
283  * Lookup a mount point by filesystem identifier.
284  */
285 struct mount *
286 vfs_getvfs(fsid_t *fsid)
287 {
288 	struct mount *mp;
289 	lwkt_tokref ilock;
290 
291 	lwkt_gettoken(&ilock, &mountlist_token);
292 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
293 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
294 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
295 			break;
296 	    }
297 	}
298 	lwkt_reltoken(&ilock);
299 	return (mp);
300 }
301 
302 /*
303  * Get a new unique fsid.  Try to make its val[0] unique, since this value
304  * will be used to create fake device numbers for stat().  Also try (but
305  * not so hard) make its val[0] unique mod 2^16, since some emulators only
306  * support 16-bit device numbers.  We end up with unique val[0]'s for the
307  * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
308  *
309  * Keep in mind that several mounts may be running in parallel.  Starting
310  * the search one past where the previous search terminated is both a
311  * micro-optimization and a defense against returning the same fsid to
312  * different mounts.
313  */
314 void
315 vfs_getnewfsid(struct mount *mp)
316 {
317 	static u_int16_t mntid_base;
318 	lwkt_tokref ilock;
319 	fsid_t tfsid;
320 	int mtype;
321 
322 	lwkt_gettoken(&ilock, &mntid_token);
323 	mtype = mp->mnt_vfc->vfc_typenum;
324 	tfsid.val[1] = mtype;
325 	mtype = (mtype & 0xFF) << 24;
326 	for (;;) {
327 		tfsid.val[0] = makeudev(255,
328 		    mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
329 		mntid_base++;
330 		if (vfs_getvfs(&tfsid) == NULL)
331 			break;
332 	}
333 	mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
334 	mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
335 	lwkt_reltoken(&ilock);
336 }
337 
338 /*
339  * This routine is called when we have too many vnodes.  It attempts
340  * to free <count> vnodes and will potentially free vnodes that still
341  * have VM backing store (VM backing store is typically the cause
342  * of a vnode blowout so we want to do this).  Therefore, this operation
343  * is not considered cheap.
344  *
345  * A number of conditions may prevent a vnode from being reclaimed.
346  * the buffer cache may have references on the vnode, a directory
347  * vnode may still have references due to the namei cache representing
348  * underlying files, or the vnode may be in active use.   It is not
349  * desireable to reuse such vnodes.  These conditions may cause the
350  * number of vnodes to reach some minimum value regardless of what
351  * you set kern.maxvnodes to.  Do not set kern.maxvnodes too low.
352  */
353 
354 /*
355  * Return 0 if the vnode is not already on the free list, return 1 if the
356  * vnode, with some additional work could possibly be placed on the free list.
357  */
358 static __inline int
359 vmightfree(struct vnode *vp, int use_count, int page_count)
360 {
361 	if (vp->v_flag & VFREE)
362 		return (0);
363 	if (vp->v_usecount != use_count || vp->v_holdcnt)
364 		return (0);
365 	if (vp->v_object && vp->v_object->resident_page_count >= page_count)
366 		return (0);
367 	return (1);
368 }
369 
370 
371 static int
372 vlrureclaim(struct mount *mp)
373 {
374 	struct vnode *vp;
375 	lwkt_tokref ilock;
376 	int done;
377 	int trigger;
378 	int usevnodes;
379 	int count;
380 
381 	/*
382 	 * Calculate the trigger point, don't allow user
383 	 * screwups to blow us up.   This prevents us from
384 	 * recycling vnodes with lots of resident pages.  We
385 	 * aren't trying to free memory, we are trying to
386 	 * free vnodes.
387 	 */
388 	usevnodes = desiredvnodes;
389 	if (usevnodes <= 0)
390 		usevnodes = 1;
391 	trigger = vmstats.v_page_count * 2 / usevnodes;
392 
393 	done = 0;
394 	lwkt_gettoken(&ilock, &mntvnode_token);
395 	count = mp->mnt_nvnodelistsize / 10 + 1;
396 	while (count && (vp = TAILQ_FIRST(&mp->mnt_nvnodelist)) != NULL) {
397 		/*
398 		 * __VNODESCAN__
399 		 *
400 		 * The VP will stick around while we hold mntvnode_token,
401 		 * at least until we block, so we can safely do an initial
402 		 * check, and then must check again after we lock the vnode.
403 		 */
404 		if (vp->v_type == VNON ||	/* XXX */
405 		    vp->v_type == VBAD ||	/* XXX */
406 		    !vmightfree(vp, 0, trigger)	/* critical path opt */
407 		) {
408 			TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
409 			TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist,vp, v_nmntvnodes);
410 			--count;
411 			continue;
412 		}
413 
414 		/*
415 		 * VX get the candidate vnode.  If the VX get fails the
416 		 * vnode might still be on the mountlist.  Our loop depends
417 		 * on us at least cycling the vnode to the end of the
418 		 * mountlist.
419 		 */
420 		if (vx_get_nonblock(vp) != 0) {
421 			if (vp->v_mount == mp) {
422 				TAILQ_REMOVE(&mp->mnt_nvnodelist,
423 						vp, v_nmntvnodes);
424 				TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist,
425 						vp, v_nmntvnodes);
426 			}
427 			--count;
428 			continue;
429 		}
430 
431 		/*
432 		 * Since we blocked locking the vp, make sure it is still
433 		 * a candidate for reclamation.  That is, it has not already
434 		 * been reclaimed and only has our VX reference associated
435 		 * with it.
436 		 */
437 		if (vp->v_type == VNON ||	/* XXX */
438 		    vp->v_type == VBAD ||	/* XXX */
439 		    (vp->v_flag & VRECLAIMED) ||
440 		    vp->v_mount != mp ||
441 		    !vmightfree(vp, 1, trigger)	/* critical path opt */
442 		) {
443 			if (vp->v_mount == mp) {
444 				TAILQ_REMOVE(&mp->mnt_nvnodelist,
445 						vp, v_nmntvnodes);
446 				TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist,
447 						vp, v_nmntvnodes);
448 			}
449 			--count;
450 			vx_put(vp);
451 			continue;
452 		}
453 
454 		/*
455 		 * All right, we are good, move the vp to the end of the
456 		 * mountlist and clean it out.  The vget will have returned
457 		 * an error if the vnode was destroyed (VRECLAIMED set), so we
458 		 * do not have to check again.  The vput() will move the
459 		 * vnode to the free list if the vgone() was successful.
460 		 */
461 		KKASSERT(vp->v_mount == mp);
462 		TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
463 		TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist,vp, v_nmntvnodes);
464 		vgone(vp);
465 		vx_put(vp);
466 		++done;
467 		--count;
468 	}
469 	lwkt_reltoken(&ilock);
470 	return (done);
471 }
472 
473 /*
474  * Attempt to recycle vnodes in a context that is always safe to block.
475  * Calling vlrurecycle() from the bowels of file system code has some
476  * interesting deadlock problems.
477  */
478 static struct thread *vnlruthread;
479 static int vnlruproc_sig;
480 
481 void
482 vnlru_proc_wait(void)
483 {
484 	if (vnlruproc_sig == 0) {
485 		vnlruproc_sig = 1;      /* avoid unnecessary wakeups */
486 		wakeup(vnlruthread);
487 	}
488 	tsleep(&vnlruproc_sig, 0, "vlruwk", hz);
489 }
490 
491 static void
492 vnlru_proc(void)
493 {
494 	struct mount *mp, *nmp;
495 	lwkt_tokref ilock;
496 	int s;
497 	int done;
498 	struct thread *td = curthread;
499 
500 	EVENTHANDLER_REGISTER(shutdown_pre_sync, shutdown_kproc, td,
501 	    SHUTDOWN_PRI_FIRST);
502 
503 	s = splbio();
504 	for (;;) {
505 		kproc_suspend_loop();
506 		if (numvnodes - freevnodes <= desiredvnodes * 9 / 10) {
507 			vnlruproc_sig = 0;
508 			wakeup(&vnlruproc_sig);
509 			tsleep(td, 0, "vlruwt", hz);
510 			continue;
511 		}
512 		done = 0;
513 		cache_cleanneg(0);
514 		lwkt_gettoken(&ilock, &mountlist_token);
515 		for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
516 			if (vfs_busy(mp, LK_NOWAIT, &ilock, td)) {
517 				nmp = TAILQ_NEXT(mp, mnt_list);
518 				continue;
519 			}
520 			done += vlrureclaim(mp);
521 			lwkt_gettokref(&ilock);
522 			nmp = TAILQ_NEXT(mp, mnt_list);
523 			vfs_unbusy(mp, td);
524 		}
525 		lwkt_reltoken(&ilock);
526 		if (done == 0) {
527 			++vnlru_nowhere;
528 			tsleep(td, 0, "vlrup", hz * 3);
529 			if (vnlru_nowhere % 10 == 0)
530 				printf("vnlru_proc: vnode recycler stopped working!\n");
531 		} else {
532 			vnlru_nowhere = 0;
533 		}
534 	}
535 	splx(s);
536 }
537 
538 static struct kproc_desc vnlru_kp = {
539 	"vnlru",
540 	vnlru_proc,
541 	&vnlruthread
542 };
543 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &vnlru_kp)
544 
545 /*
546  * Move a vnode from one mount queue to another.
547  */
548 void
549 insmntque(struct vnode *vp, struct mount *mp)
550 {
551 	lwkt_tokref ilock;
552 
553 	lwkt_gettoken(&ilock, &mntvnode_token);
554 	/*
555 	 * Delete from old mount point vnode list, if on one.
556 	 */
557 	if (vp->v_mount != NULL) {
558 		KASSERT(vp->v_mount->mnt_nvnodelistsize > 0,
559 			("bad mount point vnode list size"));
560 		TAILQ_REMOVE(&vp->v_mount->mnt_nvnodelist, vp, v_nmntvnodes);
561 		vp->v_mount->mnt_nvnodelistsize--;
562 	}
563 	/*
564 	 * Insert into list of vnodes for the new mount point, if available.
565 	 */
566 	if ((vp->v_mount = mp) == NULL) {
567 		lwkt_reltoken(&ilock);
568 		return;
569 	}
570 	TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
571 	mp->mnt_nvnodelistsize++;
572 	lwkt_reltoken(&ilock);
573 }
574 
575 
576 /*
577  * Scan the vnodes under a mount point.  The first function is called
578  * with just the mountlist token held (no vnode lock).  The second
579  * function is called with the vnode VX locked.
580  */
581 int
582 vmntvnodescan(
583     struct mount *mp,
584     int flags,
585     int (*fastfunc)(struct mount *mp, struct vnode *vp, void *data),
586     int (*slowfunc)(struct mount *mp, struct vnode *vp, void *data),
587     void *data
588 ) {
589 	lwkt_tokref ilock;
590 	struct vnode *pvp;
591 	struct vnode *vp;
592 	int r = 0;
593 
594 	/*
595 	 * Scan the vnodes on the mount's vnode list.  Use a placemarker
596 	 */
597 	pvp = allocvnode_placemarker();
598 
599 	lwkt_gettoken(&ilock, &mntvnode_token);
600 	TAILQ_INSERT_HEAD(&mp->mnt_nvnodelist, pvp, v_nmntvnodes);
601 
602 	while ((vp = TAILQ_NEXT(pvp, v_nmntvnodes)) != NULL) {
603 		/*
604 		 * Move the placemarker and skip other placemarkers we
605 		 * encounter.  The nothing can get in our way so the
606 		 * mount point on the vp must be valid.
607 		 */
608 		TAILQ_REMOVE(&mp->mnt_nvnodelist, pvp, v_nmntvnodes);
609 		TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, pvp, v_nmntvnodes);
610 		if (vp->v_flag & VPLACEMARKER)	/* another procs placemarker */
611 			continue;
612 		if (vp->v_type == VNON)		/* visible but not ready */
613 			continue;
614 		KKASSERT(vp->v_mount == mp);
615 
616 		/*
617 		 * Quick test.  A negative return continues the loop without
618 		 * calling the slow test.  0 continues onto the slow test.
619 		 * A positive number aborts the loop.
620 		 */
621 		if (fastfunc) {
622 			if ((r = fastfunc(mp, vp, data)) < 0)
623 				continue;
624 			if (r)
625 				break;
626 		}
627 
628 		/*
629 		 * Get a vxlock on the vnode, retry if it has moved or isn't
630 		 * in the mountlist where we expect it.
631 		 */
632 		if (slowfunc) {
633 			int error;
634 
635 			switch(flags) {
636 			case VMSC_GETVP:
637 				error = vget(vp, LK_EXCLUSIVE, curthread);
638 				break;
639 			case VMSC_GETVP|VMSC_NOWAIT:
640 				error = vget(vp, LK_EXCLUSIVE|LK_NOWAIT,
641 						curthread);
642 				break;
643 			case VMSC_GETVX:
644 				error = vx_get(vp);
645 				break;
646 			case VMSC_REFVP:
647 				vref(vp);
648 				/* fall through */
649 			default:
650 				error = 0;
651 				break;
652 			}
653 			if (error)
654 				continue;
655 			if (TAILQ_PREV(pvp, vnodelst, v_nmntvnodes) != vp)
656 				goto skip;
657 			if (vp->v_type == VNON)
658 				goto skip;
659 			r = slowfunc(mp, vp, data);
660 skip:
661 			switch(flags) {
662 			case VMSC_GETVP:
663 			case VMSC_GETVP|VMSC_NOWAIT:
664 				vput(vp);
665 				break;
666 			case VMSC_GETVX:
667 				vx_put(vp);
668 				break;
669 			case VMSC_REFVP:
670 				vrele(vp);
671 				/* fall through */
672 			default:
673 				break;
674 			}
675 			if (r != 0)
676 				break;
677 		}
678 	}
679 	TAILQ_REMOVE(&mp->mnt_nvnodelist, pvp, v_nmntvnodes);
680 	freevnode_placemarker(pvp);
681 	lwkt_reltoken(&ilock);
682 	return(r);
683 }
684 
685 /*
686  * Remove any vnodes in the vnode table belonging to mount point mp.
687  *
688  * If FORCECLOSE is not specified, there should not be any active ones,
689  * return error if any are found (nb: this is a user error, not a
690  * system error). If FORCECLOSE is specified, detach any active vnodes
691  * that are found.
692  *
693  * If WRITECLOSE is set, only flush out regular file vnodes open for
694  * writing.
695  *
696  * SKIPSYSTEM causes any vnodes marked VSYSTEM to be skipped.
697  *
698  * `rootrefs' specifies the base reference count for the root vnode
699  * of this filesystem. The root vnode is considered busy if its
700  * v_usecount exceeds this value. On a successful return, vflush()
701  * will call vrele() on the root vnode exactly rootrefs times.
702  * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
703  * be zero.
704  */
705 #ifdef DIAGNOSTIC
706 static int busyprt = 0;		/* print out busy vnodes */
707 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
708 #endif
709 
710 static int vflush_scan(struct mount *mp, struct vnode *vp, void *data);
711 
712 struct vflush_info {
713 	int flags;
714 	int busy;
715 	thread_t td;
716 };
717 
718 int
719 vflush(struct mount *mp, int rootrefs, int flags)
720 {
721 	struct thread *td = curthread;	/* XXX */
722 	struct vnode *rootvp = NULL;
723 	int error;
724 	struct vflush_info vflush_info;
725 
726 	if (rootrefs > 0) {
727 		KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
728 		    ("vflush: bad args"));
729 		/*
730 		 * Get the filesystem root vnode. We can vput() it
731 		 * immediately, since with rootrefs > 0, it won't go away.
732 		 */
733 		if ((error = VFS_ROOT(mp, &rootvp)) != 0)
734 			return (error);
735 		vput(rootvp);
736 	}
737 
738 	vflush_info.busy = 0;
739 	vflush_info.flags = flags;
740 	vflush_info.td = td;
741 	vmntvnodescan(mp, VMSC_GETVX, NULL, vflush_scan, &vflush_info);
742 
743 	if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
744 		/*
745 		 * If just the root vnode is busy, and if its refcount
746 		 * is equal to `rootrefs', then go ahead and kill it.
747 		 */
748 		KASSERT(vflush_info.busy > 0, ("vflush: not busy"));
749 		KASSERT(rootvp->v_usecount >= rootrefs, ("vflush: rootrefs"));
750 		if (vflush_info.busy == 1 && rootvp->v_usecount == rootrefs) {
751 			if (vx_lock(rootvp) == 0) {
752 				vgone(rootvp);
753 				vx_unlock(rootvp);
754 				vflush_info.busy = 0;
755 			}
756 		}
757 	}
758 	if (vflush_info.busy)
759 		return (EBUSY);
760 	for (; rootrefs > 0; rootrefs--)
761 		vrele(rootvp);
762 	return (0);
763 }
764 
765 /*
766  * The scan callback is made with an VX locked vnode.
767  */
768 static int
769 vflush_scan(struct mount *mp, struct vnode *vp, void *data)
770 {
771 	struct vflush_info *info = data;
772 	struct vattr vattr;
773 
774 	/*
775 	 * Skip over a vnodes marked VSYSTEM.
776 	 */
777 	if ((info->flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) {
778 		return(0);
779 	}
780 
781 	/*
782 	 * If WRITECLOSE is set, flush out unlinked but still open
783 	 * files (even if open only for reading) and regular file
784 	 * vnodes open for writing.
785 	 */
786 	if ((info->flags & WRITECLOSE) &&
787 	    (vp->v_type == VNON ||
788 	    (VOP_GETATTR(vp, &vattr, info->td) == 0 &&
789 	    vattr.va_nlink > 0)) &&
790 	    (vp->v_writecount == 0 || vp->v_type != VREG)) {
791 		return(0);
792 	}
793 
794 	/*
795 	 * With v_usecount == 0, all we need to do is clear out the
796 	 * vnode data structures and we are done.
797 	 */
798 	if (vp->v_usecount == 1) {
799 		vgone(vp);
800 		return(0);
801 	}
802 
803 	/*
804 	 * If FORCECLOSE is set, forcibly close the vnode. For block
805 	 * or character devices, revert to an anonymous device. For
806 	 * all other files, just kill them.
807 	 */
808 	if (info->flags & FORCECLOSE) {
809 		if (vp->v_type != VBLK && vp->v_type != VCHR) {
810 			vgone(vp);
811 		} else {
812 			vclean(vp, 0, info->td);
813 			vp->v_ops = &spec_vnode_vops;
814 			insmntque(vp, NULL);
815 		}
816 		return(0);
817 	}
818 #ifdef DIAGNOSTIC
819 	if (busyprt)
820 		vprint("vflush: busy vnode", vp);
821 #endif
822 	++info->busy;
823 	return(0);
824 }
825 
826