xref: /netbsd-src/sys/kern/vfs_subr.c (revision 5c46dd73a9bcb28b2994504ea090f64066b17a77)
1 /*	$NetBSD: vfs_subr.c,v 1.404 2010/06/06 08:01:31 hannken Exp $	*/
2 
3 /*-
4  * Copyright (c) 1997, 1998, 2004, 2005, 2007, 2008 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, by Charles M. Hannum, and by Andrew Doran.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
21  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
22  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
23  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
24  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
27  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
29  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
30  * POSSIBILITY OF SUCH DAMAGE.
31  */
32 
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. Neither the name of the University nor the names of its contributors
51  *    may be used to endorse or promote products derived from this software
52  *    without specific prior written permission.
53  *
54  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
55  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
56  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
57  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
58  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
59  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
60  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
61  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
62  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
63  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
64  * SUCH DAMAGE.
65  *
66  *	@(#)vfs_subr.c	8.13 (Berkeley) 4/18/94
67  */
68 
69 /*
70  * Note on v_usecount and locking:
71  *
72  * At nearly all points it is known that v_usecount could be zero, the
73  * vnode interlock will be held.
74  *
75  * To change v_usecount away from zero, the interlock must be held.  To
76  * change from a non-zero value to zero, again the interlock must be
77  * held.
78  *
79  * There's a flag bit, VC_XLOCK, embedded in v_usecount.
80  * To raise v_usecount, if the VC_XLOCK bit is set in it, the interlock
81  * must be held.
82  * To modify the VC_XLOCK bit, the interlock must be held.
83  * We always keep the usecount (v_usecount & VC_MASK) non-zero while the
84  * VC_XLOCK bit is set.
85  *
86  * Unless the VC_XLOCK bit is set, changing the usecount from a non-zero
87  * value to a non-zero value can safely be done using atomic operations,
88  * without the interlock held.
89  * Even if the VC_XLOCK bit is set, decreasing the usecount to a non-zero
90  * value can be done using atomic operations, without the interlock held.
91  */
92 
93 #include <sys/cdefs.h>
94 __KERNEL_RCSID(0, "$NetBSD: vfs_subr.c,v 1.404 2010/06/06 08:01:31 hannken Exp $");
95 
96 #include "opt_ddb.h"
97 #include "opt_compat_netbsd.h"
98 #include "opt_compat_43.h"
99 
100 #include <sys/param.h>
101 #include <sys/systm.h>
102 #include <sys/conf.h>
103 #include <sys/dirent.h>
104 #include <sys/proc.h>
105 #include <sys/kernel.h>
106 #include <sys/mount.h>
107 #include <sys/fcntl.h>
108 #include <sys/vnode.h>
109 #include <sys/stat.h>
110 #include <sys/namei.h>
111 #include <sys/ucred.h>
112 #include <sys/buf.h>
113 #include <sys/errno.h>
114 #include <sys/kmem.h>
115 #include <sys/syscallargs.h>
116 #include <sys/device.h>
117 #include <sys/filedesc.h>
118 #include <sys/kauth.h>
119 #include <sys/atomic.h>
120 #include <sys/kthread.h>
121 #include <sys/wapbl.h>
122 #include <sys/module.h>
123 
124 #include <miscfs/genfs/genfs.h>
125 #include <miscfs/specfs/specdev.h>
126 #include <miscfs/syncfs/syncfs.h>
127 
128 #include <uvm/uvm.h>
129 #include <uvm/uvm_readahead.h>
130 #include <uvm/uvm_ddb.h>
131 
132 #include <sys/sysctl.h>
133 
134 const enum vtype iftovt_tab[16] = {
135 	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
136 	VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
137 };
138 const int	vttoif_tab[9] = {
139 	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
140 	S_IFSOCK, S_IFIFO, S_IFMT,
141 };
142 
143 /*
144  * Insq/Remq for the vnode usage lists.
145  */
146 #define	bufinsvn(bp, dp)	LIST_INSERT_HEAD(dp, bp, b_vnbufs)
147 #define	bufremvn(bp) {							\
148 	LIST_REMOVE(bp, b_vnbufs);					\
149 	(bp)->b_vnbufs.le_next = NOLIST;				\
150 }
151 
152 int doforce = 1;		/* 1 => permit forcible unmounting */
153 int prtactive = 0;		/* 1 => print out reclaim of active vnodes */
154 
155 static vnodelst_t vnode_free_list = TAILQ_HEAD_INITIALIZER(vnode_free_list);
156 static vnodelst_t vnode_hold_list = TAILQ_HEAD_INITIALIZER(vnode_hold_list);
157 static vnodelst_t vrele_list = TAILQ_HEAD_INITIALIZER(vrele_list);
158 
159 struct mntlist mountlist =			/* mounted filesystem list */
160     CIRCLEQ_HEAD_INITIALIZER(mountlist);
161 
162 u_int numvnodes;
163 static specificdata_domain_t mount_specificdata_domain;
164 
165 static int vrele_pending;
166 static int vrele_gen;
167 static kmutex_t	vrele_lock;
168 static kcondvar_t vrele_cv;
169 static lwp_t *vrele_lwp;
170 
171 static uint64_t mountgen = 0;
172 static kmutex_t mountgen_lock;
173 
174 kmutex_t mountlist_lock;
175 kmutex_t mntid_lock;
176 kmutex_t mntvnode_lock;
177 kmutex_t vnode_free_list_lock;
178 kmutex_t vfs_list_lock;
179 
180 static pool_cache_t vnode_cache;
181 
182 /*
183  * These define the root filesystem and device.
184  */
185 struct vnode *rootvnode;
186 struct device *root_device;			/* root device */
187 
188 /*
189  * Local declarations.
190  */
191 
192 static void vrele_thread(void *);
193 static void insmntque(vnode_t *, struct mount *);
194 static int getdevvp(dev_t, vnode_t **, enum vtype);
195 static vnode_t *getcleanvnode(void);
196 void vpanic(vnode_t *, const char *);
197 static void vfs_shutdown1(struct lwp *);
198 
199 #ifdef DEBUG
200 void printlockedvnodes(void);
201 #endif
202 
203 #ifdef DIAGNOSTIC
204 void
205 vpanic(vnode_t *vp, const char *msg)
206 {
207 
208 	vprint(NULL, vp);
209 	panic("%s\n", msg);
210 }
211 #else
212 #define	vpanic(vp, msg)	/* nothing */
213 #endif
214 
215 void
216 vn_init1(void)
217 {
218 
219 	vnode_cache = pool_cache_init(sizeof(struct vnode), 0, 0, 0, "vnodepl",
220 	    NULL, IPL_NONE, NULL, NULL, NULL);
221 	KASSERT(vnode_cache != NULL);
222 
223 	/* Create deferred release thread. */
224 	mutex_init(&vrele_lock, MUTEX_DEFAULT, IPL_NONE);
225 	cv_init(&vrele_cv, "vrele");
226 	if (kthread_create(PRI_VM, KTHREAD_MPSAFE, NULL, vrele_thread,
227 	    NULL, &vrele_lwp, "vrele"))
228 		panic("fork vrele");
229 }
230 
231 /*
232  * Initialize the vnode management data structures.
233  */
234 void
235 vntblinit(void)
236 {
237 
238 	mutex_init(&mountgen_lock, MUTEX_DEFAULT, IPL_NONE);
239 	mutex_init(&mountlist_lock, MUTEX_DEFAULT, IPL_NONE);
240 	mutex_init(&mntid_lock, MUTEX_DEFAULT, IPL_NONE);
241 	mutex_init(&mntvnode_lock, MUTEX_DEFAULT, IPL_NONE);
242 	mutex_init(&vnode_free_list_lock, MUTEX_DEFAULT, IPL_NONE);
243 	mutex_init(&vfs_list_lock, MUTEX_DEFAULT, IPL_NONE);
244 
245 	mount_specificdata_domain = specificdata_domain_create();
246 
247 	/* Initialize the filesystem syncer. */
248 	vn_initialize_syncerd();
249 	vn_init1();
250 }
251 
252 int
253 vfs_drainvnodes(long target, struct lwp *l)
254 {
255 
256 	while (numvnodes > target) {
257 		vnode_t *vp;
258 
259 		mutex_enter(&vnode_free_list_lock);
260 		vp = getcleanvnode();
261 		if (vp == NULL)
262 			return EBUSY; /* give up */
263 		ungetnewvnode(vp);
264 	}
265 
266 	return 0;
267 }
268 
269 /*
270  * Lookup a mount point by filesystem identifier.
271  *
272  * XXX Needs to add a reference to the mount point.
273  */
274 struct mount *
275 vfs_getvfs(fsid_t *fsid)
276 {
277 	struct mount *mp;
278 
279 	mutex_enter(&mountlist_lock);
280 	CIRCLEQ_FOREACH(mp, &mountlist, mnt_list) {
281 		if (mp->mnt_stat.f_fsidx.__fsid_val[0] == fsid->__fsid_val[0] &&
282 		    mp->mnt_stat.f_fsidx.__fsid_val[1] == fsid->__fsid_val[1]) {
283 			mutex_exit(&mountlist_lock);
284 			return (mp);
285 		}
286 	}
287 	mutex_exit(&mountlist_lock);
288 	return ((struct mount *)0);
289 }
290 
291 /*
292  * Drop a reference to a mount structure, freeing if the last reference.
293  */
294 void
295 vfs_destroy(struct mount *mp)
296 {
297 
298 	if (__predict_true((int)atomic_dec_uint_nv(&mp->mnt_refcnt) > 0)) {
299 		return;
300 	}
301 
302 	/*
303 	 * Nothing else has visibility of the mount: we can now
304 	 * free the data structures.
305 	 */
306 	KASSERT(mp->mnt_refcnt == 0);
307 	specificdata_fini(mount_specificdata_domain, &mp->mnt_specdataref);
308 	rw_destroy(&mp->mnt_unmounting);
309 	mutex_destroy(&mp->mnt_updating);
310 	mutex_destroy(&mp->mnt_renamelock);
311 	if (mp->mnt_op != NULL) {
312 		vfs_delref(mp->mnt_op);
313 	}
314 	kmem_free(mp, sizeof(*mp));
315 }
316 
317 /*
318  * grab a vnode from freelist and clean it.
319  */
320 vnode_t *
321 getcleanvnode(void)
322 {
323 	vnode_t *vp;
324 	vnodelst_t *listhd;
325 
326 	KASSERT(mutex_owned(&vnode_free_list_lock));
327 
328 retry:
329 	listhd = &vnode_free_list;
330 try_nextlist:
331 	TAILQ_FOREACH(vp, listhd, v_freelist) {
332 		/*
333 		 * It's safe to test v_usecount and v_iflag
334 		 * without holding the interlock here, since
335 		 * these vnodes should never appear on the
336 		 * lists.
337 		 */
338 		if (vp->v_usecount != 0) {
339 			vpanic(vp, "free vnode isn't");
340 		}
341 		if ((vp->v_iflag & VI_CLEAN) != 0) {
342 			vpanic(vp, "clean vnode on freelist");
343 		}
344 		if (vp->v_freelisthd != listhd) {
345 			printf("vnode sez %p, listhd %p\n", vp->v_freelisthd, listhd);
346 			vpanic(vp, "list head mismatch");
347 		}
348 		if (!mutex_tryenter(&vp->v_interlock))
349 			continue;
350 		/*
351 		 * Our lwp might hold the underlying vnode
352 		 * locked, so don't try to reclaim a VI_LAYER
353 		 * node if it's locked.
354 		 */
355 		if ((vp->v_iflag & VI_XLOCK) == 0 &&
356 		    ((vp->v_iflag & VI_LAYER) == 0 || VOP_ISLOCKED(vp) == 0)) {
357 			break;
358 		}
359 		mutex_exit(&vp->v_interlock);
360 	}
361 
362 	if (vp == NULL) {
363 		if (listhd == &vnode_free_list) {
364 			listhd = &vnode_hold_list;
365 			goto try_nextlist;
366 		}
367 		mutex_exit(&vnode_free_list_lock);
368 		return NULL;
369 	}
370 
371 	/* Remove it from the freelist. */
372 	TAILQ_REMOVE(listhd, vp, v_freelist);
373 	vp->v_freelisthd = NULL;
374 	mutex_exit(&vnode_free_list_lock);
375 
376 	if (vp->v_usecount != 0) {
377 		/*
378 		 * was referenced again before we got the interlock
379 		 * Don't return to freelist - the holder of the last
380 		 * reference will destroy it.
381 		 */
382 		mutex_exit(&vp->v_interlock);
383 		mutex_enter(&vnode_free_list_lock);
384 		goto retry;
385 	}
386 
387 	/*
388 	 * The vnode is still associated with a file system, so we must
389 	 * clean it out before reusing it.  We need to add a reference
390 	 * before doing this.  If the vnode gains another reference while
391 	 * being cleaned out then we lose - retry.
392 	 */
393 	atomic_add_int(&vp->v_usecount, 1 + VC_XLOCK);
394 	vclean(vp, DOCLOSE);
395 	KASSERT(vp->v_usecount >= 1 + VC_XLOCK);
396 	atomic_add_int(&vp->v_usecount, -VC_XLOCK);
397 	if (vp->v_usecount == 1) {
398 		/* We're about to dirty it. */
399 		vp->v_iflag &= ~VI_CLEAN;
400 		mutex_exit(&vp->v_interlock);
401 		if (vp->v_type == VBLK || vp->v_type == VCHR) {
402 			spec_node_destroy(vp);
403 		}
404 		vp->v_type = VNON;
405 	} else {
406 		/*
407 		 * Don't return to freelist - the holder of the last
408 		 * reference will destroy it.
409 		 */
410 		vrelel(vp, 0); /* releases vp->v_interlock */
411 		mutex_enter(&vnode_free_list_lock);
412 		goto retry;
413 	}
414 
415 	if (vp->v_data != NULL || vp->v_uobj.uo_npages != 0 ||
416 	    !TAILQ_EMPTY(&vp->v_uobj.memq)) {
417 		vpanic(vp, "cleaned vnode isn't");
418 	}
419 	if (vp->v_numoutput != 0) {
420 		vpanic(vp, "clean vnode has pending I/O's");
421 	}
422 	if ((vp->v_iflag & VI_ONWORKLST) != 0) {
423 		vpanic(vp, "clean vnode on syncer list");
424 	}
425 
426 	return vp;
427 }
428 
429 /*
430  * Mark a mount point as busy, and gain a new reference to it.  Used to
431  * prevent the file system from being unmounted during critical sections.
432  *
433  * => The caller must hold a pre-existing reference to the mount.
434  * => Will fail if the file system is being unmounted, or is unmounted.
435  */
436 int
437 vfs_busy(struct mount *mp, struct mount **nextp)
438 {
439 
440 	KASSERT(mp->mnt_refcnt > 0);
441 
442 	if (__predict_false(!rw_tryenter(&mp->mnt_unmounting, RW_READER))) {
443 		if (nextp != NULL) {
444 			KASSERT(mutex_owned(&mountlist_lock));
445 			*nextp = CIRCLEQ_NEXT(mp, mnt_list);
446 		}
447 		return EBUSY;
448 	}
449 	if (__predict_false((mp->mnt_iflag & IMNT_GONE) != 0)) {
450 		rw_exit(&mp->mnt_unmounting);
451 		if (nextp != NULL) {
452 			KASSERT(mutex_owned(&mountlist_lock));
453 			*nextp = CIRCLEQ_NEXT(mp, mnt_list);
454 		}
455 		return ENOENT;
456 	}
457 	if (nextp != NULL) {
458 		mutex_exit(&mountlist_lock);
459 	}
460 	atomic_inc_uint(&mp->mnt_refcnt);
461 	return 0;
462 }
463 
464 /*
465  * Unbusy a busy filesystem.
466  *
467  * => If keepref is true, preserve reference added by vfs_busy().
468  * => If nextp != NULL, acquire mountlist_lock.
469  */
470 void
471 vfs_unbusy(struct mount *mp, bool keepref, struct mount **nextp)
472 {
473 
474 	KASSERT(mp->mnt_refcnt > 0);
475 
476 	if (nextp != NULL) {
477 		mutex_enter(&mountlist_lock);
478 	}
479 	rw_exit(&mp->mnt_unmounting);
480 	if (!keepref) {
481 		vfs_destroy(mp);
482 	}
483 	if (nextp != NULL) {
484 		KASSERT(mutex_owned(&mountlist_lock));
485 		*nextp = CIRCLEQ_NEXT(mp, mnt_list);
486 	}
487 }
488 
489 struct mount *
490 vfs_mountalloc(struct vfsops *vfsops, struct vnode *vp)
491 {
492 	int error;
493 	struct mount *mp;
494 
495 	mp = kmem_zalloc(sizeof(*mp), KM_SLEEP);
496 	if (mp == NULL)
497 		return NULL;
498 
499 	mp->mnt_op = vfsops;
500 	mp->mnt_refcnt = 1;
501 	TAILQ_INIT(&mp->mnt_vnodelist);
502 	rw_init(&mp->mnt_unmounting);
503 	mutex_init(&mp->mnt_renamelock, MUTEX_DEFAULT, IPL_NONE);
504 	mutex_init(&mp->mnt_updating, MUTEX_DEFAULT, IPL_NONE);
505 	error = vfs_busy(mp, NULL);
506 	KASSERT(error == 0);
507 	mp->mnt_vnodecovered = vp;
508 	mount_initspecific(mp);
509 
510 	mutex_enter(&mountgen_lock);
511 	mp->mnt_gen = mountgen++;
512 	mutex_exit(&mountgen_lock);
513 
514 	return mp;
515 }
516 
517 /*
518  * Lookup a filesystem type, and if found allocate and initialize
519  * a mount structure for it.
520  *
521  * Devname is usually updated by mount(8) after booting.
522  */
523 int
524 vfs_rootmountalloc(const char *fstypename, const char *devname,
525     struct mount **mpp)
526 {
527 	struct vfsops *vfsp = NULL;
528 	struct mount *mp;
529 
530 	mutex_enter(&vfs_list_lock);
531 	LIST_FOREACH(vfsp, &vfs_list, vfs_list)
532 		if (!strncmp(vfsp->vfs_name, fstypename,
533 		    sizeof(mp->mnt_stat.f_fstypename)))
534 			break;
535 	if (vfsp == NULL) {
536 		mutex_exit(&vfs_list_lock);
537 		return (ENODEV);
538 	}
539 	vfsp->vfs_refcount++;
540 	mutex_exit(&vfs_list_lock);
541 
542 	if ((mp = vfs_mountalloc(vfsp, NULL)) == NULL)
543 		return ENOMEM;
544 	mp->mnt_flag = MNT_RDONLY;
545 	(void)strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfs_name,
546 	    sizeof(mp->mnt_stat.f_fstypename));
547 	mp->mnt_stat.f_mntonname[0] = '/';
548 	mp->mnt_stat.f_mntonname[1] = '\0';
549 	mp->mnt_stat.f_mntfromname[sizeof(mp->mnt_stat.f_mntfromname) - 1] =
550 	    '\0';
551 	(void)copystr(devname, mp->mnt_stat.f_mntfromname,
552 	    sizeof(mp->mnt_stat.f_mntfromname) - 1, 0);
553 	*mpp = mp;
554 	return (0);
555 }
556 
557 /*
558  * Routines having to do with the management of the vnode table.
559  */
560 extern int (**dead_vnodeop_p)(void *);
561 
562 /*
563  * Return the next vnode from the free list.
564  */
565 int
566 getnewvnode(enum vtagtype tag, struct mount *mp, int (**vops)(void *),
567 	    vnode_t **vpp)
568 {
569 	struct uvm_object *uobj;
570 	static int toggle;
571 	vnode_t *vp;
572 	int error = 0, tryalloc;
573 
574  try_again:
575 	if (mp != NULL) {
576 		/*
577 		 * Mark filesystem busy while we're creating a
578 		 * vnode.  If unmount is in progress, this will
579 		 * fail.
580 		 */
581 		error = vfs_busy(mp, NULL);
582 		if (error)
583 			return error;
584 	}
585 
586 	/*
587 	 * We must choose whether to allocate a new vnode or recycle an
588 	 * existing one. The criterion for allocating a new one is that
589 	 * the total number of vnodes is less than the number desired or
590 	 * there are no vnodes on either free list. Generally we only
591 	 * want to recycle vnodes that have no buffers associated with
592 	 * them, so we look first on the vnode_free_list. If it is empty,
593 	 * we next consider vnodes with referencing buffers on the
594 	 * vnode_hold_list. The toggle ensures that half the time we
595 	 * will use a buffer from the vnode_hold_list, and half the time
596 	 * we will allocate a new one unless the list has grown to twice
597 	 * the desired size. We are reticent to recycle vnodes from the
598 	 * vnode_hold_list because we will lose the identity of all its
599 	 * referencing buffers.
600 	 */
601 
602 	vp = NULL;
603 
604 	mutex_enter(&vnode_free_list_lock);
605 
606 	toggle ^= 1;
607 	if (numvnodes > 2 * desiredvnodes)
608 		toggle = 0;
609 
610 	tryalloc = numvnodes < desiredvnodes ||
611 	    (TAILQ_FIRST(&vnode_free_list) == NULL &&
612 	     (TAILQ_FIRST(&vnode_hold_list) == NULL || toggle));
613 
614 	if (tryalloc) {
615 		numvnodes++;
616 		mutex_exit(&vnode_free_list_lock);
617 		if ((vp = vnalloc(NULL)) == NULL) {
618 			mutex_enter(&vnode_free_list_lock);
619 			numvnodes--;
620 		} else
621 			vp->v_usecount = 1;
622 	}
623 
624 	if (vp == NULL) {
625 		vp = getcleanvnode();
626 		if (vp == NULL) {
627 			if (mp != NULL) {
628 				vfs_unbusy(mp, false, NULL);
629 			}
630 			if (tryalloc) {
631 				printf("WARNING: unable to allocate new "
632 				    "vnode, retrying...\n");
633 				kpause("newvn", false, hz, NULL);
634 				goto try_again;
635 			}
636 			tablefull("vnode", "increase kern.maxvnodes or NVNODE");
637 			*vpp = 0;
638 			return (ENFILE);
639 		}
640 		vp->v_iflag = 0;
641 		vp->v_vflag = 0;
642 		vp->v_uflag = 0;
643 		vp->v_socket = NULL;
644 	}
645 
646 	KASSERT(vp->v_usecount == 1);
647 	KASSERT(vp->v_freelisthd == NULL);
648 	KASSERT(LIST_EMPTY(&vp->v_nclist));
649 	KASSERT(LIST_EMPTY(&vp->v_dnclist));
650 
651 	vp->v_type = VNON;
652 	vp->v_tag = tag;
653 	vp->v_op = vops;
654 	insmntque(vp, mp);
655 	*vpp = vp;
656 	vp->v_data = 0;
657 
658 	/*
659 	 * initialize uvm_object within vnode.
660 	 */
661 
662 	uobj = &vp->v_uobj;
663 	KASSERT(uobj->pgops == &uvm_vnodeops);
664 	KASSERT(uobj->uo_npages == 0);
665 	KASSERT(TAILQ_FIRST(&uobj->memq) == NULL);
666 	vp->v_size = vp->v_writesize = VSIZENOTSET;
667 
668 	if (mp != NULL) {
669 		if ((mp->mnt_iflag & IMNT_MPSAFE) != 0)
670 			vp->v_vflag |= VV_MPSAFE;
671 		vfs_unbusy(mp, true, NULL);
672 	}
673 
674 	return (0);
675 }
676 
677 /*
678  * This is really just the reverse of getnewvnode(). Needed for
679  * VFS_VGET functions who may need to push back a vnode in case
680  * of a locking race.
681  */
682 void
683 ungetnewvnode(vnode_t *vp)
684 {
685 
686 	KASSERT(vp->v_usecount == 1);
687 	KASSERT(vp->v_data == NULL);
688 	KASSERT(vp->v_freelisthd == NULL);
689 
690 	mutex_enter(&vp->v_interlock);
691 	vp->v_iflag |= VI_CLEAN;
692 	vrelel(vp, 0);
693 }
694 
695 /*
696  * Allocate a new, uninitialized vnode.  If 'mp' is non-NULL, this is a
697  * marker vnode and we are prepared to wait for the allocation.
698  */
699 vnode_t *
700 vnalloc(struct mount *mp)
701 {
702 	vnode_t *vp;
703 
704 	vp = pool_cache_get(vnode_cache, (mp != NULL ? PR_WAITOK : PR_NOWAIT));
705 	if (vp == NULL) {
706 		return NULL;
707 	}
708 
709 	memset(vp, 0, sizeof(*vp));
710 	UVM_OBJ_INIT(&vp->v_uobj, &uvm_vnodeops, 0);
711 	cv_init(&vp->v_cv, "vnode");
712 	/*
713 	 * done by memset() above.
714 	 *	LIST_INIT(&vp->v_nclist);
715 	 *	LIST_INIT(&vp->v_dnclist);
716 	 */
717 
718 	if (mp != NULL) {
719 		vp->v_mount = mp;
720 		vp->v_type = VBAD;
721 		vp->v_iflag = VI_MARKER;
722 	} else {
723 		rw_init(&vp->v_lock.vl_lock);
724 	}
725 
726 	return vp;
727 }
728 
729 /*
730  * Free an unused, unreferenced vnode.
731  */
732 void
733 vnfree(vnode_t *vp)
734 {
735 
736 	KASSERT(vp->v_usecount == 0);
737 
738 	if ((vp->v_iflag & VI_MARKER) == 0) {
739 		rw_destroy(&vp->v_lock.vl_lock);
740 		mutex_enter(&vnode_free_list_lock);
741 		numvnodes--;
742 		mutex_exit(&vnode_free_list_lock);
743 	}
744 
745 	UVM_OBJ_DESTROY(&vp->v_uobj);
746 	cv_destroy(&vp->v_cv);
747 	pool_cache_put(vnode_cache, vp);
748 }
749 
750 /*
751  * Remove a vnode from its freelist.
752  */
753 static inline void
754 vremfree(vnode_t *vp)
755 {
756 
757 	KASSERT(mutex_owned(&vp->v_interlock));
758 	KASSERT(vp->v_usecount == 0);
759 
760 	/*
761 	 * Note that the reference count must not change until
762 	 * the vnode is removed.
763 	 */
764 	mutex_enter(&vnode_free_list_lock);
765 	if (vp->v_holdcnt > 0) {
766 		KASSERT(vp->v_freelisthd == &vnode_hold_list);
767 	} else {
768 		KASSERT(vp->v_freelisthd == &vnode_free_list);
769 	}
770 	TAILQ_REMOVE(vp->v_freelisthd, vp, v_freelist);
771 	vp->v_freelisthd = NULL;
772 	mutex_exit(&vnode_free_list_lock);
773 }
774 
775 /*
776  * Move a vnode from one mount queue to another.
777  */
778 static void
779 insmntque(vnode_t *vp, struct mount *mp)
780 {
781 	struct mount *omp;
782 
783 #ifdef DIAGNOSTIC
784 	if ((mp != NULL) &&
785 	    (mp->mnt_iflag & IMNT_UNMOUNT) &&
786 	    vp->v_tag != VT_VFS) {
787 		panic("insmntque into dying filesystem");
788 	}
789 #endif
790 
791 	mutex_enter(&mntvnode_lock);
792 	/*
793 	 * Delete from old mount point vnode list, if on one.
794 	 */
795 	if ((omp = vp->v_mount) != NULL)
796 		TAILQ_REMOVE(&vp->v_mount->mnt_vnodelist, vp, v_mntvnodes);
797 	/*
798 	 * Insert into list of vnodes for the new mount point, if
799 	 * available.  The caller must take a reference on the mount
800 	 * structure and donate to the vnode.
801 	 */
802 	if ((vp->v_mount = mp) != NULL)
803 		TAILQ_INSERT_TAIL(&mp->mnt_vnodelist, vp, v_mntvnodes);
804 	mutex_exit(&mntvnode_lock);
805 
806 	if (omp != NULL) {
807 		/* Release reference to old mount. */
808 		vfs_destroy(omp);
809 	}
810 }
811 
812 /*
813  * Wait for a vnode (typically with VI_XLOCK set) to be cleaned or
814  * recycled.
815  */
816 void
817 vwait(vnode_t *vp, int flags)
818 {
819 
820 	KASSERT(mutex_owned(&vp->v_interlock));
821 	KASSERT(vp->v_usecount != 0);
822 
823 	while ((vp->v_iflag & flags) != 0)
824 		cv_wait(&vp->v_cv, &vp->v_interlock);
825 }
826 
827 /*
828  * Insert a marker vnode into a mount's vnode list, after the
829  * specified vnode.  mntvnode_lock must be held.
830  */
831 void
832 vmark(vnode_t *mvp, vnode_t *vp)
833 {
834 	struct mount *mp;
835 
836 	mp = mvp->v_mount;
837 
838 	KASSERT(mutex_owned(&mntvnode_lock));
839 	KASSERT((mvp->v_iflag & VI_MARKER) != 0);
840 	KASSERT(vp->v_mount == mp);
841 
842 	TAILQ_INSERT_AFTER(&mp->mnt_vnodelist, vp, mvp, v_mntvnodes);
843 }
844 
845 /*
846  * Remove a marker vnode from a mount's vnode list, and return
847  * a pointer to the next vnode in the list.  mntvnode_lock must
848  * be held.
849  */
850 vnode_t *
851 vunmark(vnode_t *mvp)
852 {
853 	vnode_t *vp;
854 	struct mount *mp;
855 
856 	mp = mvp->v_mount;
857 
858 	KASSERT(mutex_owned(&mntvnode_lock));
859 	KASSERT((mvp->v_iflag & VI_MARKER) != 0);
860 
861 	vp = TAILQ_NEXT(mvp, v_mntvnodes);
862 	TAILQ_REMOVE(&mp->mnt_vnodelist, mvp, v_mntvnodes);
863 
864 	KASSERT(vp == NULL || vp->v_mount == mp);
865 
866 	return vp;
867 }
868 
869 /*
870  * Update outstanding I/O count and do wakeup if requested.
871  */
872 void
873 vwakeup(struct buf *bp)
874 {
875 	struct vnode *vp;
876 
877 	if ((vp = bp->b_vp) == NULL)
878 		return;
879 
880 	KASSERT(bp->b_objlock == &vp->v_interlock);
881 	KASSERT(mutex_owned(bp->b_objlock));
882 
883 	if (--vp->v_numoutput < 0)
884 		panic("vwakeup: neg numoutput, vp %p", vp);
885 	if (vp->v_numoutput == 0)
886 		cv_broadcast(&vp->v_cv);
887 }
888 
889 /*
890  * Flush out and invalidate all buffers associated with a vnode.
891  * Called with the underlying vnode locked, which should prevent new dirty
892  * buffers from being queued.
893  */
894 int
895 vinvalbuf(struct vnode *vp, int flags, kauth_cred_t cred, struct lwp *l,
896 	  bool catch, int slptimeo)
897 {
898 	struct buf *bp, *nbp;
899 	int error;
900 	int flushflags = PGO_ALLPAGES | PGO_FREE | PGO_SYNCIO |
901 	    (flags & V_SAVE ? PGO_CLEANIT | PGO_RECLAIM : 0);
902 
903 	/* XXXUBC this doesn't look at flags or slp* */
904 	mutex_enter(&vp->v_interlock);
905 	error = VOP_PUTPAGES(vp, 0, 0, flushflags);
906 	if (error) {
907 		return error;
908 	}
909 
910 	if (flags & V_SAVE) {
911 		error = VOP_FSYNC(vp, cred, FSYNC_WAIT|FSYNC_RECLAIM, 0, 0);
912 		if (error)
913 		        return (error);
914 		KASSERT(LIST_EMPTY(&vp->v_dirtyblkhd));
915 	}
916 
917 	mutex_enter(&bufcache_lock);
918 restart:
919 	for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
920 		nbp = LIST_NEXT(bp, b_vnbufs);
921 		error = bbusy(bp, catch, slptimeo, NULL);
922 		if (error != 0) {
923 			if (error == EPASSTHROUGH)
924 				goto restart;
925 			mutex_exit(&bufcache_lock);
926 			return (error);
927 		}
928 		brelsel(bp, BC_INVAL | BC_VFLUSH);
929 	}
930 
931 	for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
932 		nbp = LIST_NEXT(bp, b_vnbufs);
933 		error = bbusy(bp, catch, slptimeo, NULL);
934 		if (error != 0) {
935 			if (error == EPASSTHROUGH)
936 				goto restart;
937 			mutex_exit(&bufcache_lock);
938 			return (error);
939 		}
940 		/*
941 		 * XXX Since there are no node locks for NFS, I believe
942 		 * there is a slight chance that a delayed write will
943 		 * occur while sleeping just above, so check for it.
944 		 */
945 		if ((bp->b_oflags & BO_DELWRI) && (flags & V_SAVE)) {
946 #ifdef DEBUG
947 			printf("buffer still DELWRI\n");
948 #endif
949 			bp->b_cflags |= BC_BUSY | BC_VFLUSH;
950 			mutex_exit(&bufcache_lock);
951 			VOP_BWRITE(bp);
952 			mutex_enter(&bufcache_lock);
953 			goto restart;
954 		}
955 		brelsel(bp, BC_INVAL | BC_VFLUSH);
956 	}
957 
958 #ifdef DIAGNOSTIC
959 	if (!LIST_EMPTY(&vp->v_cleanblkhd) || !LIST_EMPTY(&vp->v_dirtyblkhd))
960 		panic("vinvalbuf: flush failed, vp %p", vp);
961 #endif
962 
963 	mutex_exit(&bufcache_lock);
964 
965 	return (0);
966 }
967 
968 /*
969  * Destroy any in core blocks past the truncation length.
970  * Called with the underlying vnode locked, which should prevent new dirty
971  * buffers from being queued.
972  */
973 int
974 vtruncbuf(struct vnode *vp, daddr_t lbn, bool catch, int slptimeo)
975 {
976 	struct buf *bp, *nbp;
977 	int error;
978 	voff_t off;
979 
980 	off = round_page((voff_t)lbn << vp->v_mount->mnt_fs_bshift);
981 	mutex_enter(&vp->v_interlock);
982 	error = VOP_PUTPAGES(vp, off, 0, PGO_FREE | PGO_SYNCIO);
983 	if (error) {
984 		return error;
985 	}
986 
987 	mutex_enter(&bufcache_lock);
988 restart:
989 	for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
990 		nbp = LIST_NEXT(bp, b_vnbufs);
991 		if (bp->b_lblkno < lbn)
992 			continue;
993 		error = bbusy(bp, catch, slptimeo, NULL);
994 		if (error != 0) {
995 			if (error == EPASSTHROUGH)
996 				goto restart;
997 			mutex_exit(&bufcache_lock);
998 			return (error);
999 		}
1000 		brelsel(bp, BC_INVAL | BC_VFLUSH);
1001 	}
1002 
1003 	for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
1004 		nbp = LIST_NEXT(bp, b_vnbufs);
1005 		if (bp->b_lblkno < lbn)
1006 			continue;
1007 		error = bbusy(bp, catch, slptimeo, NULL);
1008 		if (error != 0) {
1009 			if (error == EPASSTHROUGH)
1010 				goto restart;
1011 			mutex_exit(&bufcache_lock);
1012 			return (error);
1013 		}
1014 		brelsel(bp, BC_INVAL | BC_VFLUSH);
1015 	}
1016 	mutex_exit(&bufcache_lock);
1017 
1018 	return (0);
1019 }
1020 
1021 /*
1022  * Flush all dirty buffers from a vnode.
1023  * Called with the underlying vnode locked, which should prevent new dirty
1024  * buffers from being queued.
1025  */
1026 void
1027 vflushbuf(struct vnode *vp, int sync)
1028 {
1029 	struct buf *bp, *nbp;
1030 	int flags = PGO_CLEANIT | PGO_ALLPAGES | (sync ? PGO_SYNCIO : 0);
1031 	bool dirty;
1032 
1033 	mutex_enter(&vp->v_interlock);
1034 	(void) VOP_PUTPAGES(vp, 0, 0, flags);
1035 
1036 loop:
1037 	mutex_enter(&bufcache_lock);
1038 	for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
1039 		nbp = LIST_NEXT(bp, b_vnbufs);
1040 		if ((bp->b_cflags & BC_BUSY))
1041 			continue;
1042 		if ((bp->b_oflags & BO_DELWRI) == 0)
1043 			panic("vflushbuf: not dirty, bp %p", bp);
1044 		bp->b_cflags |= BC_BUSY | BC_VFLUSH;
1045 		mutex_exit(&bufcache_lock);
1046 		/*
1047 		 * Wait for I/O associated with indirect blocks to complete,
1048 		 * since there is no way to quickly wait for them below.
1049 		 */
1050 		if (bp->b_vp == vp || sync == 0)
1051 			(void) bawrite(bp);
1052 		else
1053 			(void) bwrite(bp);
1054 		goto loop;
1055 	}
1056 	mutex_exit(&bufcache_lock);
1057 
1058 	if (sync == 0)
1059 		return;
1060 
1061 	mutex_enter(&vp->v_interlock);
1062 	while (vp->v_numoutput != 0)
1063 		cv_wait(&vp->v_cv, &vp->v_interlock);
1064 	dirty = !LIST_EMPTY(&vp->v_dirtyblkhd);
1065 	mutex_exit(&vp->v_interlock);
1066 
1067 	if (dirty) {
1068 		vprint("vflushbuf: dirty", vp);
1069 		goto loop;
1070 	}
1071 }
1072 
1073 /*
1074  * Create a vnode for a block device.
1075  * Used for root filesystem and swap areas.
1076  * Also used for memory file system special devices.
1077  */
1078 int
1079 bdevvp(dev_t dev, vnode_t **vpp)
1080 {
1081 
1082 	return (getdevvp(dev, vpp, VBLK));
1083 }
1084 
1085 /*
1086  * Create a vnode for a character device.
1087  * Used for kernfs and some console handling.
1088  */
1089 int
1090 cdevvp(dev_t dev, vnode_t **vpp)
1091 {
1092 
1093 	return (getdevvp(dev, vpp, VCHR));
1094 }
1095 
1096 /*
1097  * Associate a buffer with a vnode.  There must already be a hold on
1098  * the vnode.
1099  */
1100 void
1101 bgetvp(struct vnode *vp, struct buf *bp)
1102 {
1103 
1104 	KASSERT(bp->b_vp == NULL);
1105 	KASSERT(bp->b_objlock == &buffer_lock);
1106 	KASSERT(mutex_owned(&vp->v_interlock));
1107 	KASSERT(mutex_owned(&bufcache_lock));
1108 	KASSERT((bp->b_cflags & BC_BUSY) != 0);
1109 	KASSERT(!cv_has_waiters(&bp->b_done));
1110 
1111 	vholdl(vp);
1112 	bp->b_vp = vp;
1113 	if (vp->v_type == VBLK || vp->v_type == VCHR)
1114 		bp->b_dev = vp->v_rdev;
1115 	else
1116 		bp->b_dev = NODEV;
1117 
1118 	/*
1119 	 * Insert onto list for new vnode.
1120 	 */
1121 	bufinsvn(bp, &vp->v_cleanblkhd);
1122 	bp->b_objlock = &vp->v_interlock;
1123 }
1124 
1125 /*
1126  * Disassociate a buffer from a vnode.
1127  */
1128 void
1129 brelvp(struct buf *bp)
1130 {
1131 	struct vnode *vp = bp->b_vp;
1132 
1133 	KASSERT(vp != NULL);
1134 	KASSERT(bp->b_objlock == &vp->v_interlock);
1135 	KASSERT(mutex_owned(&vp->v_interlock));
1136 	KASSERT(mutex_owned(&bufcache_lock));
1137 	KASSERT((bp->b_cflags & BC_BUSY) != 0);
1138 	KASSERT(!cv_has_waiters(&bp->b_done));
1139 
1140 	/*
1141 	 * Delete from old vnode list, if on one.
1142 	 */
1143 	if (LIST_NEXT(bp, b_vnbufs) != NOLIST)
1144 		bufremvn(bp);
1145 
1146 	if (TAILQ_EMPTY(&vp->v_uobj.memq) && (vp->v_iflag & VI_ONWORKLST) &&
1147 	    LIST_FIRST(&vp->v_dirtyblkhd) == NULL) {
1148 		vp->v_iflag &= ~VI_WRMAPDIRTY;
1149 		vn_syncer_remove_from_worklist(vp);
1150 	}
1151 
1152 	bp->b_objlock = &buffer_lock;
1153 	bp->b_vp = NULL;
1154 	holdrelel(vp);
1155 }
1156 
1157 /*
1158  * Reassign a buffer from one vnode list to another.
1159  * The list reassignment must be within the same vnode.
1160  * Used to assign file specific control information
1161  * (indirect blocks) to the list to which they belong.
1162  */
1163 void
1164 reassignbuf(struct buf *bp, struct vnode *vp)
1165 {
1166 	struct buflists *listheadp;
1167 	int delayx;
1168 
1169 	KASSERT(mutex_owned(&bufcache_lock));
1170 	KASSERT(bp->b_objlock == &vp->v_interlock);
1171 	KASSERT(mutex_owned(&vp->v_interlock));
1172 	KASSERT((bp->b_cflags & BC_BUSY) != 0);
1173 
1174 	/*
1175 	 * Delete from old vnode list, if on one.
1176 	 */
1177 	if (LIST_NEXT(bp, b_vnbufs) != NOLIST)
1178 		bufremvn(bp);
1179 
1180 	/*
1181 	 * If dirty, put on list of dirty buffers;
1182 	 * otherwise insert onto list of clean buffers.
1183 	 */
1184 	if ((bp->b_oflags & BO_DELWRI) == 0) {
1185 		listheadp = &vp->v_cleanblkhd;
1186 		if (TAILQ_EMPTY(&vp->v_uobj.memq) &&
1187 		    (vp->v_iflag & VI_ONWORKLST) &&
1188 		    LIST_FIRST(&vp->v_dirtyblkhd) == NULL) {
1189 			vp->v_iflag &= ~VI_WRMAPDIRTY;
1190 			vn_syncer_remove_from_worklist(vp);
1191 		}
1192 	} else {
1193 		listheadp = &vp->v_dirtyblkhd;
1194 		if ((vp->v_iflag & VI_ONWORKLST) == 0) {
1195 			switch (vp->v_type) {
1196 			case VDIR:
1197 				delayx = dirdelay;
1198 				break;
1199 			case VBLK:
1200 				if (vp->v_specmountpoint != NULL) {
1201 					delayx = metadelay;
1202 					break;
1203 				}
1204 				/* fall through */
1205 			default:
1206 				delayx = filedelay;
1207 				break;
1208 			}
1209 			if (!vp->v_mount ||
1210 			    (vp->v_mount->mnt_flag & MNT_ASYNC) == 0)
1211 				vn_syncer_add_to_worklist(vp, delayx);
1212 		}
1213 	}
1214 	bufinsvn(bp, listheadp);
1215 }
1216 
1217 /*
1218  * Create a vnode for a device.
1219  * Used by bdevvp (block device) for root file system etc.,
1220  * and by cdevvp (character device) for console and kernfs.
1221  */
1222 static int
1223 getdevvp(dev_t dev, vnode_t **vpp, enum vtype type)
1224 {
1225 	vnode_t *vp;
1226 	vnode_t *nvp;
1227 	int error;
1228 
1229 	if (dev == NODEV) {
1230 		*vpp = NULL;
1231 		return (0);
1232 	}
1233 	error = getnewvnode(VT_NON, NULL, spec_vnodeop_p, &nvp);
1234 	if (error) {
1235 		*vpp = NULL;
1236 		return (error);
1237 	}
1238 	vp = nvp;
1239 	vp->v_type = type;
1240 	vp->v_vflag |= VV_MPSAFE;
1241 	uvm_vnp_setsize(vp, 0);
1242 	spec_node_init(vp, dev);
1243 	*vpp = vp;
1244 	return (0);
1245 }
1246 
1247 /*
1248  * Try to gain a reference to a vnode, without acquiring its interlock.
1249  * The caller must hold a lock that will prevent the vnode from being
1250  * recycled or freed.
1251  */
1252 bool
1253 vtryget(vnode_t *vp)
1254 {
1255 	u_int use, next;
1256 
1257 	/*
1258 	 * If the vnode is being freed, don't make life any harder
1259 	 * for vclean() by adding another reference without waiting.
1260 	 * This is not strictly necessary, but we'll do it anyway.
1261 	 */
1262 	if (__predict_false((vp->v_iflag & (VI_XLOCK | VI_FREEING)) != 0)) {
1263 		return false;
1264 	}
1265 	for (use = vp->v_usecount;; use = next) {
1266 		if (use == 0 || __predict_false((use & VC_XLOCK) != 0)) {
1267 			/* Need interlock held if first reference. */
1268 			return false;
1269 		}
1270 		next = atomic_cas_uint(&vp->v_usecount, use, use + 1);
1271 		if (__predict_true(next == use)) {
1272 			return true;
1273 		}
1274 	}
1275 }
1276 
1277 /*
1278  * Grab a particular vnode from the free list, increment its
1279  * reference count and lock it. If the vnode lock bit is set the
1280  * vnode is being eliminated in vgone. In that case, we can not
1281  * grab the vnode, so the process is awakened when the transition is
1282  * completed, and an error returned to indicate that the vnode is no
1283  * longer usable (possibly having been changed to a new file system type).
1284  */
1285 int
1286 vget(vnode_t *vp, int flags)
1287 {
1288 	int error = 0;
1289 
1290 	KASSERT((vp->v_iflag & VI_MARKER) == 0);
1291 
1292 	if ((flags & LK_INTERLOCK) == 0)
1293 		mutex_enter(&vp->v_interlock);
1294 
1295 	/*
1296 	 * Before adding a reference, we must remove the vnode
1297 	 * from its freelist.
1298 	 */
1299 	if (vp->v_usecount == 0) {
1300 		vremfree(vp);
1301 		vp->v_usecount = 1;
1302 	} else {
1303 		atomic_inc_uint(&vp->v_usecount);
1304 	}
1305 
1306 	/*
1307 	 * If the vnode is in the process of being cleaned out for
1308 	 * another use, we wait for the cleaning to finish and then
1309 	 * return failure.  Cleaning is determined by checking if
1310 	 * the VI_XLOCK or VI_FREEING flags are set.
1311 	 */
1312 	if ((vp->v_iflag & (VI_XLOCK | VI_FREEING)) != 0) {
1313 		if ((flags & LK_NOWAIT) != 0) {
1314 			vrelel(vp, 0);
1315 			return EBUSY;
1316 		}
1317 		vwait(vp, VI_XLOCK | VI_FREEING);
1318 		vrelel(vp, 0);
1319 		return ENOENT;
1320 	}
1321 
1322 	if ((vp->v_iflag & VI_INACTNOW) != 0) {
1323 		/*
1324 		 * if it's being desactived, wait for it to complete.
1325 		 * Make sure to not return a clean vnode.
1326 		 */
1327 		 if ((flags & LK_NOWAIT) != 0) {
1328 			vrelel(vp, 0);
1329 			return EBUSY;
1330 		}
1331 		vwait(vp, VI_INACTNOW);
1332 		if ((vp->v_iflag & VI_CLEAN) != 0) {
1333 			vrelel(vp, 0);
1334 			return ENOENT;
1335 		}
1336 	}
1337 
1338 	/*
1339 	 * Ok, we got it in good shape.  Just locking left.
1340 	 */
1341 	KASSERT((vp->v_iflag & VI_CLEAN) == 0);
1342 	if (flags & LK_TYPE_MASK) {
1343 		error = vn_lock(vp, flags | LK_INTERLOCK);
1344 		if (error != 0) {
1345 			vrele(vp);
1346 		}
1347 	} else {
1348 		mutex_exit(&vp->v_interlock);
1349 	}
1350 	return error;
1351 }
1352 
1353 /*
1354  * vput(), just unlock and vrele()
1355  */
1356 void
1357 vput(vnode_t *vp)
1358 {
1359 
1360 	KASSERT((vp->v_iflag & VI_MARKER) == 0);
1361 
1362 	VOP_UNLOCK(vp, 0);
1363 	vrele(vp);
1364 }
1365 
1366 /*
1367  * Try to drop reference on a vnode.  Abort if we are releasing the
1368  * last reference.  Note: this _must_ succeed if not the last reference.
1369  */
1370 static inline bool
1371 vtryrele(vnode_t *vp)
1372 {
1373 	u_int use, next;
1374 
1375 	for (use = vp->v_usecount;; use = next) {
1376 		if (use == 1) {
1377 			return false;
1378 		}
1379 		KASSERT((use & VC_MASK) > 1);
1380 		next = atomic_cas_uint(&vp->v_usecount, use, use - 1);
1381 		if (__predict_true(next == use)) {
1382 			return true;
1383 		}
1384 	}
1385 }
1386 
1387 /*
1388  * Vnode release.  If reference count drops to zero, call inactive
1389  * routine and either return to freelist or free to the pool.
1390  */
1391 void
1392 vrelel(vnode_t *vp, int flags)
1393 {
1394 	bool recycle, defer;
1395 	int error;
1396 
1397 	KASSERT(mutex_owned(&vp->v_interlock));
1398 	KASSERT((vp->v_iflag & VI_MARKER) == 0);
1399 	KASSERT(vp->v_freelisthd == NULL);
1400 
1401 	if (__predict_false(vp->v_op == dead_vnodeop_p &&
1402 	    (vp->v_iflag & (VI_CLEAN|VI_XLOCK)) == 0)) {
1403 		vpanic(vp, "dead but not clean");
1404 	}
1405 
1406 	/*
1407 	 * If not the last reference, just drop the reference count
1408 	 * and unlock.
1409 	 */
1410 	if (vtryrele(vp)) {
1411 		vp->v_iflag |= VI_INACTREDO;
1412 		mutex_exit(&vp->v_interlock);
1413 		return;
1414 	}
1415 	if (vp->v_usecount <= 0 || vp->v_writecount != 0) {
1416 		vpanic(vp, "vrelel: bad ref count");
1417 	}
1418 
1419 	KASSERT((vp->v_iflag & VI_XLOCK) == 0);
1420 
1421 	/*
1422 	 * If not clean, deactivate the vnode, but preserve
1423 	 * our reference across the call to VOP_INACTIVE().
1424 	 */
1425  retry:
1426 	if ((vp->v_iflag & VI_CLEAN) == 0) {
1427 		recycle = false;
1428 		vp->v_iflag |= VI_INACTNOW;
1429 
1430 		/*
1431 		 * XXX This ugly block can be largely eliminated if
1432 		 * locking is pushed down into the file systems.
1433 		 *
1434 		 * Defer vnode release to vrele_thread if caller
1435 		 * requests it explicitly.
1436 		 */
1437 		if ((curlwp == uvm.pagedaemon_lwp) ||
1438 		    (flags & VRELEL_ASYNC_RELE) != 0) {
1439 			/* The pagedaemon can't wait around; defer. */
1440 			defer = true;
1441 		} else if (curlwp == vrele_lwp) {
1442 			/*
1443 			 * We have to try harder. But we can't sleep
1444 			 * with VI_INACTNOW as vget() may be waiting on it.
1445 			 */
1446 			vp->v_iflag &= ~(VI_INACTREDO|VI_INACTNOW);
1447 			cv_broadcast(&vp->v_cv);
1448 			error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK |
1449 			    LK_RETRY);
1450 			if (error != 0) {
1451 				/* XXX */
1452 				vpanic(vp, "vrele: unable to lock %p");
1453 			}
1454 			mutex_enter(&vp->v_interlock);
1455 			/*
1456 			 * if we did get another reference while
1457 			 * sleeping, don't try to inactivate it yet.
1458 			 */
1459 			if (__predict_false(vtryrele(vp))) {
1460 				VOP_UNLOCK(vp, 0);
1461 				mutex_exit(&vp->v_interlock);
1462 				return;
1463 			}
1464 			vp->v_iflag |= VI_INACTNOW;
1465 			mutex_exit(&vp->v_interlock);
1466 			defer = false;
1467 		} else if ((vp->v_iflag & VI_LAYER) != 0) {
1468 			/*
1469 			 * Acquiring the stack's lock in vclean() even
1470 			 * for an honest vput/vrele is dangerous because
1471 			 * our caller may hold other vnode locks; defer.
1472 			 */
1473 			defer = true;
1474 		} else {
1475 			/* If we can't acquire the lock, then defer. */
1476 			vp->v_iflag &= ~VI_INACTREDO;
1477 			error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK |
1478 			    LK_NOWAIT);
1479 			if (error != 0) {
1480 				defer = true;
1481 				mutex_enter(&vp->v_interlock);
1482 			} else {
1483 				defer = false;
1484 			}
1485 		}
1486 
1487 		if (defer) {
1488 			/*
1489 			 * Defer reclaim to the kthread; it's not safe to
1490 			 * clean it here.  We donate it our last reference.
1491 			 */
1492 			KASSERT(mutex_owned(&vp->v_interlock));
1493 			KASSERT((vp->v_iflag & VI_INACTPEND) == 0);
1494 			vp->v_iflag &= ~VI_INACTNOW;
1495 			vp->v_iflag |= VI_INACTPEND;
1496 			mutex_enter(&vrele_lock);
1497 			TAILQ_INSERT_TAIL(&vrele_list, vp, v_freelist);
1498 			if (++vrele_pending > (desiredvnodes >> 8))
1499 				cv_signal(&vrele_cv);
1500 			mutex_exit(&vrele_lock);
1501 			cv_broadcast(&vp->v_cv);
1502 			mutex_exit(&vp->v_interlock);
1503 			return;
1504 		}
1505 
1506 #ifdef DIAGNOSTIC
1507 		if ((vp->v_type == VBLK || vp->v_type == VCHR) &&
1508 		    vp->v_specnode != NULL && vp->v_specnode->sn_opencnt != 0) {
1509 			vprint("vrelel: missing VOP_CLOSE()", vp);
1510 		}
1511 #endif
1512 
1513 		/*
1514 		 * The vnode can gain another reference while being
1515 		 * deactivated.  If VOP_INACTIVE() indicates that
1516 		 * the described file has been deleted, then recycle
1517 		 * the vnode irrespective of additional references.
1518 		 * Another thread may be waiting to re-use the on-disk
1519 		 * inode.
1520 		 *
1521 		 * Note that VOP_INACTIVE() will drop the vnode lock.
1522 		 */
1523 		VOP_INACTIVE(vp, &recycle);
1524 		mutex_enter(&vp->v_interlock);
1525 		vp->v_iflag &= ~VI_INACTNOW;
1526 		cv_broadcast(&vp->v_cv);
1527 		if (!recycle) {
1528 			if (vtryrele(vp)) {
1529 				mutex_exit(&vp->v_interlock);
1530 				return;
1531 			}
1532 
1533 			/*
1534 			 * If we grew another reference while
1535 			 * VOP_INACTIVE() was underway, retry.
1536 			 */
1537 			if ((vp->v_iflag & VI_INACTREDO) != 0) {
1538 				goto retry;
1539 			}
1540 		}
1541 
1542 		/* Take care of space accounting. */
1543 		if (vp->v_iflag & VI_EXECMAP) {
1544 			atomic_add_int(&uvmexp.execpages,
1545 			    -vp->v_uobj.uo_npages);
1546 			atomic_add_int(&uvmexp.filepages,
1547 			    vp->v_uobj.uo_npages);
1548 		}
1549 		vp->v_iflag &= ~(VI_TEXT|VI_EXECMAP|VI_WRMAP);
1550 		vp->v_vflag &= ~VV_MAPPED;
1551 
1552 		/*
1553 		 * Recycle the vnode if the file is now unused (unlinked),
1554 		 * otherwise just free it.
1555 		 */
1556 		if (recycle) {
1557 			vclean(vp, DOCLOSE);
1558 		}
1559 		KASSERT(vp->v_usecount > 0);
1560 	}
1561 
1562 	if (atomic_dec_uint_nv(&vp->v_usecount) != 0) {
1563 		/* Gained another reference while being reclaimed. */
1564 		mutex_exit(&vp->v_interlock);
1565 		return;
1566 	}
1567 
1568 	if ((vp->v_iflag & VI_CLEAN) != 0) {
1569 		/*
1570 		 * It's clean so destroy it.  It isn't referenced
1571 		 * anywhere since it has been reclaimed.
1572 		 */
1573 		KASSERT(vp->v_holdcnt == 0);
1574 		KASSERT(vp->v_writecount == 0);
1575 		mutex_exit(&vp->v_interlock);
1576 		insmntque(vp, NULL);
1577 		if (vp->v_type == VBLK || vp->v_type == VCHR) {
1578 			spec_node_destroy(vp);
1579 		}
1580 		vnfree(vp);
1581 	} else {
1582 		/*
1583 		 * Otherwise, put it back onto the freelist.  It
1584 		 * can't be destroyed while still associated with
1585 		 * a file system.
1586 		 */
1587 		mutex_enter(&vnode_free_list_lock);
1588 		if (vp->v_holdcnt > 0) {
1589 			vp->v_freelisthd = &vnode_hold_list;
1590 		} else {
1591 			vp->v_freelisthd = &vnode_free_list;
1592 		}
1593 		TAILQ_INSERT_TAIL(vp->v_freelisthd, vp, v_freelist);
1594 		mutex_exit(&vnode_free_list_lock);
1595 		mutex_exit(&vp->v_interlock);
1596 	}
1597 }
1598 
1599 void
1600 vrele(vnode_t *vp)
1601 {
1602 
1603 	KASSERT((vp->v_iflag & VI_MARKER) == 0);
1604 
1605 	if ((vp->v_iflag & VI_INACTNOW) == 0 && vtryrele(vp)) {
1606 		return;
1607 	}
1608 	mutex_enter(&vp->v_interlock);
1609 	vrelel(vp, 0);
1610 }
1611 
1612 /*
1613  * Asynchronous vnode release, vnode is released in different context.
1614  */
1615 void
1616 vrele_async(vnode_t *vp)
1617 {
1618 
1619 	KASSERT((vp->v_iflag & VI_MARKER) == 0);
1620 
1621 	if ((vp->v_iflag & VI_INACTNOW) == 0 && vtryrele(vp)) {
1622 		return;
1623 	}
1624 
1625 	mutex_enter(&vp->v_interlock);
1626 	vrelel(vp, VRELEL_ASYNC_RELE);
1627 }
1628 
1629 static void
1630 vrele_thread(void *cookie)
1631 {
1632 	vnode_t *vp;
1633 
1634 	for (;;) {
1635 		mutex_enter(&vrele_lock);
1636 		while (TAILQ_EMPTY(&vrele_list)) {
1637 			vrele_gen++;
1638 			cv_broadcast(&vrele_cv);
1639 			cv_timedwait(&vrele_cv, &vrele_lock, hz);
1640 		}
1641 		vp = TAILQ_FIRST(&vrele_list);
1642 		TAILQ_REMOVE(&vrele_list, vp, v_freelist);
1643 		vrele_pending--;
1644 		mutex_exit(&vrele_lock);
1645 
1646 		/*
1647 		 * If not the last reference, then ignore the vnode
1648 		 * and look for more work.
1649 		 */
1650 		mutex_enter(&vp->v_interlock);
1651 		KASSERT((vp->v_iflag & VI_INACTPEND) != 0);
1652 		vp->v_iflag &= ~VI_INACTPEND;
1653 		vrelel(vp, 0);
1654 	}
1655 }
1656 
1657 /*
1658  * Page or buffer structure gets a reference.
1659  * Called with v_interlock held.
1660  */
1661 void
1662 vholdl(vnode_t *vp)
1663 {
1664 
1665 	KASSERT(mutex_owned(&vp->v_interlock));
1666 	KASSERT((vp->v_iflag & VI_MARKER) == 0);
1667 
1668 	if (vp->v_holdcnt++ == 0 && vp->v_usecount == 0) {
1669 		mutex_enter(&vnode_free_list_lock);
1670 		KASSERT(vp->v_freelisthd == &vnode_free_list);
1671 		TAILQ_REMOVE(vp->v_freelisthd, vp, v_freelist);
1672 		vp->v_freelisthd = &vnode_hold_list;
1673 		TAILQ_INSERT_TAIL(vp->v_freelisthd, vp, v_freelist);
1674 		mutex_exit(&vnode_free_list_lock);
1675 	}
1676 }
1677 
1678 /*
1679  * Page or buffer structure frees a reference.
1680  * Called with v_interlock held.
1681  */
1682 void
1683 holdrelel(vnode_t *vp)
1684 {
1685 
1686 	KASSERT(mutex_owned(&vp->v_interlock));
1687 	KASSERT((vp->v_iflag & VI_MARKER) == 0);
1688 
1689 	if (vp->v_holdcnt <= 0) {
1690 		vpanic(vp, "holdrelel: holdcnt vp %p");
1691 	}
1692 
1693 	vp->v_holdcnt--;
1694 	if (vp->v_holdcnt == 0 && vp->v_usecount == 0) {
1695 		mutex_enter(&vnode_free_list_lock);
1696 		KASSERT(vp->v_freelisthd == &vnode_hold_list);
1697 		TAILQ_REMOVE(vp->v_freelisthd, vp, v_freelist);
1698 		vp->v_freelisthd = &vnode_free_list;
1699 		TAILQ_INSERT_TAIL(vp->v_freelisthd, vp, v_freelist);
1700 		mutex_exit(&vnode_free_list_lock);
1701 	}
1702 }
1703 
1704 /*
1705  * Vnode reference, where a reference is already held by some other
1706  * object (for example, a file structure).
1707  */
1708 void
1709 vref(vnode_t *vp)
1710 {
1711 
1712 	KASSERT((vp->v_iflag & VI_MARKER) == 0);
1713 	KASSERT(vp->v_usecount != 0);
1714 
1715 	atomic_inc_uint(&vp->v_usecount);
1716 }
1717 
1718 /*
1719  * Remove any vnodes in the vnode table belonging to mount point mp.
1720  *
1721  * If FORCECLOSE is not specified, there should not be any active ones,
1722  * return error if any are found (nb: this is a user error, not a
1723  * system error). If FORCECLOSE is specified, detach any active vnodes
1724  * that are found.
1725  *
1726  * If WRITECLOSE is set, only flush out regular file vnodes open for
1727  * writing.
1728  *
1729  * SKIPSYSTEM causes any vnodes marked V_SYSTEM to be skipped.
1730  */
1731 #ifdef DEBUG
1732 int busyprt = 0;	/* print out busy vnodes */
1733 struct ctldebug debug1 = { "busyprt", &busyprt };
1734 #endif
1735 
1736 static vnode_t *
1737 vflushnext(vnode_t *mvp, int *when)
1738 {
1739 
1740 	if (hardclock_ticks > *when) {
1741 		mutex_exit(&mntvnode_lock);
1742 		yield();
1743 		mutex_enter(&mntvnode_lock);
1744 		*when = hardclock_ticks + hz / 10;
1745 	}
1746 
1747 	return vunmark(mvp);
1748 }
1749 
1750 int
1751 vflush(struct mount *mp, vnode_t *skipvp, int flags)
1752 {
1753 	vnode_t *vp, *mvp;
1754 	int busy = 0, when = 0, gen;
1755 
1756 	/*
1757 	 * First, flush out any vnode references from vrele_list.
1758 	 */
1759 	mutex_enter(&vrele_lock);
1760 	gen = vrele_gen;
1761 	while (vrele_pending && gen == vrele_gen) {
1762 		cv_broadcast(&vrele_cv);
1763 		cv_wait(&vrele_cv, &vrele_lock);
1764 	}
1765 	mutex_exit(&vrele_lock);
1766 
1767 	/* Allocate a marker vnode. */
1768 	if ((mvp = vnalloc(mp)) == NULL)
1769 		return (ENOMEM);
1770 
1771 	/*
1772 	 * NOTE: not using the TAILQ_FOREACH here since in this loop vgone()
1773 	 * and vclean() are called
1774 	 */
1775 	mutex_enter(&mntvnode_lock);
1776 	for (vp = TAILQ_FIRST(&mp->mnt_vnodelist); vp != NULL;
1777 	    vp = vflushnext(mvp, &when)) {
1778 		vmark(mvp, vp);
1779 		if (vp->v_mount != mp || vismarker(vp))
1780 			continue;
1781 		/*
1782 		 * Skip over a selected vnode.
1783 		 */
1784 		if (vp == skipvp)
1785 			continue;
1786 		mutex_enter(&vp->v_interlock);
1787 		/*
1788 		 * Ignore clean but still referenced vnodes.
1789 		 */
1790 		if ((vp->v_iflag & VI_CLEAN) != 0) {
1791 			mutex_exit(&vp->v_interlock);
1792 			continue;
1793 		}
1794 		/*
1795 		 * Skip over a vnodes marked VSYSTEM.
1796 		 */
1797 		if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
1798 			mutex_exit(&vp->v_interlock);
1799 			continue;
1800 		}
1801 		/*
1802 		 * If WRITECLOSE is set, only flush out regular file
1803 		 * vnodes open for writing.
1804 		 */
1805 		if ((flags & WRITECLOSE) &&
1806 		    (vp->v_writecount == 0 || vp->v_type != VREG)) {
1807 			mutex_exit(&vp->v_interlock);
1808 			continue;
1809 		}
1810 		/*
1811 		 * With v_usecount == 0, all we need to do is clear
1812 		 * out the vnode data structures and we are done.
1813 		 */
1814 		if (vp->v_usecount == 0) {
1815 			mutex_exit(&mntvnode_lock);
1816 			vremfree(vp);
1817 			vp->v_usecount = 1;
1818 			vclean(vp, DOCLOSE);
1819 			vrelel(vp, 0);
1820 			mutex_enter(&mntvnode_lock);
1821 			continue;
1822 		}
1823 		/*
1824 		 * If FORCECLOSE is set, forcibly close the vnode.
1825 		 * For block or character devices, revert to an
1826 		 * anonymous device.  For all other files, just
1827 		 * kill them.
1828 		 */
1829 		if (flags & FORCECLOSE) {
1830 			mutex_exit(&mntvnode_lock);
1831 			atomic_inc_uint(&vp->v_usecount);
1832 			if (vp->v_type != VBLK && vp->v_type != VCHR) {
1833 				vclean(vp, DOCLOSE);
1834 				vrelel(vp, 0);
1835 			} else {
1836 				vclean(vp, 0);
1837 				vp->v_op = spec_vnodeop_p; /* XXXSMP */
1838 				mutex_exit(&vp->v_interlock);
1839 				/*
1840 				 * The vnode isn't clean, but still resides
1841 				 * on the mount list.  Remove it. XXX This
1842 				 * is a bit dodgy.
1843 				 */
1844 				insmntque(vp, NULL);
1845 				vrele(vp);
1846 			}
1847 			mutex_enter(&mntvnode_lock);
1848 			continue;
1849 		}
1850 #ifdef DEBUG
1851 		if (busyprt)
1852 			vprint("vflush: busy vnode", vp);
1853 #endif
1854 		mutex_exit(&vp->v_interlock);
1855 		busy++;
1856 	}
1857 	mutex_exit(&mntvnode_lock);
1858 	vnfree(mvp);
1859 	if (busy)
1860 		return (EBUSY);
1861 	return (0);
1862 }
1863 
1864 /*
1865  * Disassociate the underlying file system from a vnode.
1866  *
1867  * Must be called with the interlock held, and will return with it held.
1868  */
1869 void
1870 vclean(vnode_t *vp, int flags)
1871 {
1872 	lwp_t *l = curlwp;
1873 	bool recycle, active;
1874 	int error;
1875 
1876 	KASSERT(mutex_owned(&vp->v_interlock));
1877 	KASSERT((vp->v_iflag & VI_MARKER) == 0);
1878 	KASSERT(vp->v_usecount != 0);
1879 
1880 	/* If cleaning is already in progress wait until done and return. */
1881 	if (vp->v_iflag & VI_XLOCK) {
1882 		vwait(vp, VI_XLOCK);
1883 		return;
1884 	}
1885 
1886 	/* If already clean, nothing to do. */
1887 	if ((vp->v_iflag & VI_CLEAN) != 0) {
1888 		return;
1889 	}
1890 
1891 	/*
1892 	 * Prevent the vnode from being recycled or brought into use
1893 	 * while we clean it out.
1894 	 */
1895 	vp->v_iflag |= VI_XLOCK;
1896 	if (vp->v_iflag & VI_EXECMAP) {
1897 		atomic_add_int(&uvmexp.execpages, -vp->v_uobj.uo_npages);
1898 		atomic_add_int(&uvmexp.filepages, vp->v_uobj.uo_npages);
1899 	}
1900 	vp->v_iflag &= ~(VI_TEXT|VI_EXECMAP);
1901 	active = (vp->v_usecount > 1);
1902 
1903 	/* XXXAD should not lock vnode under layer */
1904 	VOP_LOCK(vp, LK_EXCLUSIVE | LK_INTERLOCK);
1905 
1906 	/*
1907 	 * Clean out any cached data associated with the vnode.
1908 	 * If purging an active vnode, it must be closed and
1909 	 * deactivated before being reclaimed. Note that the
1910 	 * VOP_INACTIVE will unlock the vnode.
1911 	 */
1912 	if (flags & DOCLOSE) {
1913 		error = vinvalbuf(vp, V_SAVE, NOCRED, l, 0, 0);
1914 		if (error != 0) {
1915 			/* XXX, fix vn_start_write's grab of mp and use that. */
1916 
1917 			if (wapbl_vphaswapbl(vp))
1918 				WAPBL_DISCARD(wapbl_vptomp(vp));
1919 			error = vinvalbuf(vp, 0, NOCRED, l, 0, 0);
1920 		}
1921 		KASSERT(error == 0);
1922 		KASSERT((vp->v_iflag & VI_ONWORKLST) == 0);
1923 		if (active && (vp->v_type == VBLK || vp->v_type == VCHR)) {
1924 			 spec_node_revoke(vp);
1925 		}
1926 	}
1927 	if (active) {
1928 		VOP_INACTIVE(vp, &recycle);
1929 	} else {
1930 		/*
1931 		 * Any other processes trying to obtain this lock must first
1932 		 * wait for VI_XLOCK to clear, then call the new lock operation.
1933 		 */
1934 		VOP_UNLOCK(vp, 0);
1935 	}
1936 
1937 	/* Disassociate the underlying file system from the vnode. */
1938 	if (VOP_RECLAIM(vp)) {
1939 		vpanic(vp, "vclean: cannot reclaim");
1940 	}
1941 
1942 	KASSERT(vp->v_uobj.uo_npages == 0);
1943 	if (vp->v_type == VREG && vp->v_ractx != NULL) {
1944 		uvm_ra_freectx(vp->v_ractx);
1945 		vp->v_ractx = NULL;
1946 	}
1947 	cache_purge(vp);
1948 
1949 	/* Done with purge, notify sleepers of the grim news. */
1950 	mutex_enter(&vp->v_interlock);
1951 	vp->v_op = dead_vnodeop_p;
1952 	vp->v_tag = VT_NON;
1953 	KNOTE(&vp->v_klist, NOTE_REVOKE);
1954 	vp->v_iflag &= ~(VI_XLOCK | VI_FREEING);
1955 	vp->v_vflag &= ~VV_LOCKSWORK;
1956 	if ((flags & DOCLOSE) != 0) {
1957 		vp->v_iflag |= VI_CLEAN;
1958 	}
1959 	cv_broadcast(&vp->v_cv);
1960 
1961 	KASSERT((vp->v_iflag & VI_ONWORKLST) == 0);
1962 }
1963 
1964 /*
1965  * Recycle an unused vnode to the front of the free list.
1966  * Release the passed interlock if the vnode will be recycled.
1967  */
1968 int
1969 vrecycle(vnode_t *vp, kmutex_t *inter_lkp, struct lwp *l)
1970 {
1971 
1972 	KASSERT((vp->v_iflag & VI_MARKER) == 0);
1973 
1974 	mutex_enter(&vp->v_interlock);
1975 	if (vp->v_usecount != 0) {
1976 		mutex_exit(&vp->v_interlock);
1977 		return (0);
1978 	}
1979 	if (inter_lkp)
1980 		mutex_exit(inter_lkp);
1981 	vremfree(vp);
1982 	vp->v_usecount = 1;
1983 	vclean(vp, DOCLOSE);
1984 	vrelel(vp, 0);
1985 	return (1);
1986 }
1987 
1988 /*
1989  * Eliminate all activity associated with a vnode in preparation for
1990  * reuse.  Drops a reference from the vnode.
1991  */
1992 void
1993 vgone(vnode_t *vp)
1994 {
1995 
1996 	mutex_enter(&vp->v_interlock);
1997 	vclean(vp, DOCLOSE);
1998 	vrelel(vp, 0);
1999 }
2000 
2001 /*
2002  * Lookup a vnode by device number.
2003  */
2004 int
2005 vfinddev(dev_t dev, enum vtype type, vnode_t **vpp)
2006 {
2007 	vnode_t *vp;
2008 	int rc = 0;
2009 
2010 	mutex_enter(&device_lock);
2011 	for (vp = specfs_hash[SPECHASH(dev)]; vp; vp = vp->v_specnext) {
2012 		if (dev != vp->v_rdev || type != vp->v_type)
2013 			continue;
2014 		*vpp = vp;
2015 		rc = 1;
2016 		break;
2017 	}
2018 	mutex_exit(&device_lock);
2019 	return (rc);
2020 }
2021 
2022 /*
2023  * Revoke all the vnodes corresponding to the specified minor number
2024  * range (endpoints inclusive) of the specified major.
2025  */
2026 void
2027 vdevgone(int maj, int minl, int minh, enum vtype type)
2028 {
2029 	vnode_t *vp, **vpp;
2030 	dev_t dev;
2031 	int mn;
2032 
2033 	vp = NULL;	/* XXX gcc */
2034 
2035 	mutex_enter(&device_lock);
2036 	for (mn = minl; mn <= minh; mn++) {
2037 		dev = makedev(maj, mn);
2038 		vpp = &specfs_hash[SPECHASH(dev)];
2039 		for (vp = *vpp; vp != NULL;) {
2040 			mutex_enter(&vp->v_interlock);
2041 			if ((vp->v_iflag & VI_CLEAN) != 0 ||
2042 			    dev != vp->v_rdev || type != vp->v_type) {
2043 				mutex_exit(&vp->v_interlock);
2044 				vp = vp->v_specnext;
2045 				continue;
2046 			}
2047 			mutex_exit(&device_lock);
2048 			if (vget(vp, LK_INTERLOCK) == 0) {
2049 				VOP_REVOKE(vp, REVOKEALL);
2050 				vrele(vp);
2051 			}
2052 			mutex_enter(&device_lock);
2053 			vp = *vpp;
2054 		}
2055 	}
2056 	mutex_exit(&device_lock);
2057 }
2058 
2059 /*
2060  * Eliminate all activity associated with the requested vnode
2061  * and with all vnodes aliased to the requested vnode.
2062  */
2063 void
2064 vrevoke(vnode_t *vp)
2065 {
2066 	vnode_t *vq, **vpp;
2067 	enum vtype type;
2068 	dev_t dev;
2069 
2070 	KASSERT(vp->v_usecount > 0);
2071 
2072 	mutex_enter(&vp->v_interlock);
2073 	if ((vp->v_iflag & VI_CLEAN) != 0) {
2074 		mutex_exit(&vp->v_interlock);
2075 		return;
2076 	} else if (vp->v_type != VBLK && vp->v_type != VCHR) {
2077 		atomic_inc_uint(&vp->v_usecount);
2078 		vclean(vp, DOCLOSE);
2079 		vrelel(vp, 0);
2080 		return;
2081 	} else {
2082 		dev = vp->v_rdev;
2083 		type = vp->v_type;
2084 		mutex_exit(&vp->v_interlock);
2085 	}
2086 
2087 	vpp = &specfs_hash[SPECHASH(dev)];
2088 	mutex_enter(&device_lock);
2089 	for (vq = *vpp; vq != NULL;) {
2090 		/* If clean or being cleaned, then ignore it. */
2091 		mutex_enter(&vq->v_interlock);
2092 		if ((vq->v_iflag & (VI_CLEAN | VI_XLOCK)) != 0 ||
2093 		    vq->v_rdev != dev || vq->v_type != type) {
2094 			mutex_exit(&vq->v_interlock);
2095 			vq = vq->v_specnext;
2096 			continue;
2097 		}
2098 		mutex_exit(&device_lock);
2099 		if (vq->v_usecount == 0) {
2100 			vremfree(vq);
2101 			vq->v_usecount = 1;
2102 		} else {
2103 			atomic_inc_uint(&vq->v_usecount);
2104 		}
2105 		vclean(vq, DOCLOSE);
2106 		vrelel(vq, 0);
2107 		mutex_enter(&device_lock);
2108 		vq = *vpp;
2109 	}
2110 	mutex_exit(&device_lock);
2111 }
2112 
2113 /*
2114  * sysctl helper routine to return list of supported fstypes
2115  */
2116 int
2117 sysctl_vfs_generic_fstypes(SYSCTLFN_ARGS)
2118 {
2119 	char bf[sizeof(((struct statvfs *)NULL)->f_fstypename)];
2120 	char *where = oldp;
2121 	struct vfsops *v;
2122 	size_t needed, left, slen;
2123 	int error, first;
2124 
2125 	if (newp != NULL)
2126 		return (EPERM);
2127 	if (namelen != 0)
2128 		return (EINVAL);
2129 
2130 	first = 1;
2131 	error = 0;
2132 	needed = 0;
2133 	left = *oldlenp;
2134 
2135 	sysctl_unlock();
2136 	mutex_enter(&vfs_list_lock);
2137 	LIST_FOREACH(v, &vfs_list, vfs_list) {
2138 		if (where == NULL)
2139 			needed += strlen(v->vfs_name) + 1;
2140 		else {
2141 			memset(bf, 0, sizeof(bf));
2142 			if (first) {
2143 				strncpy(bf, v->vfs_name, sizeof(bf));
2144 				first = 0;
2145 			} else {
2146 				bf[0] = ' ';
2147 				strncpy(bf + 1, v->vfs_name, sizeof(bf) - 1);
2148 			}
2149 			bf[sizeof(bf)-1] = '\0';
2150 			slen = strlen(bf);
2151 			if (left < slen + 1)
2152 				break;
2153 			v->vfs_refcount++;
2154 			mutex_exit(&vfs_list_lock);
2155 			/* +1 to copy out the trailing NUL byte */
2156 			error = copyout(bf, where, slen + 1);
2157 			mutex_enter(&vfs_list_lock);
2158 			v->vfs_refcount--;
2159 			if (error)
2160 				break;
2161 			where += slen;
2162 			needed += slen;
2163 			left -= slen;
2164 		}
2165 	}
2166 	mutex_exit(&vfs_list_lock);
2167 	sysctl_relock();
2168 	*oldlenp = needed;
2169 	return (error);
2170 }
2171 
2172 
2173 int kinfo_vdebug = 1;
2174 int kinfo_vgetfailed;
2175 #define KINFO_VNODESLOP	10
2176 /*
2177  * Dump vnode list (via sysctl).
2178  * Copyout address of vnode followed by vnode.
2179  */
2180 /* ARGSUSED */
2181 int
2182 sysctl_kern_vnode(SYSCTLFN_ARGS)
2183 {
2184 	char *where = oldp;
2185 	size_t *sizep = oldlenp;
2186 	struct mount *mp, *nmp;
2187 	vnode_t *vp, *mvp, vbuf;
2188 	char *bp = where;
2189 	char *ewhere;
2190 	int error;
2191 
2192 	if (namelen != 0)
2193 		return (EOPNOTSUPP);
2194 	if (newp != NULL)
2195 		return (EPERM);
2196 
2197 #define VPTRSZ	sizeof(vnode_t *)
2198 #define VNODESZ	sizeof(vnode_t)
2199 	if (where == NULL) {
2200 		*sizep = (numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ);
2201 		return (0);
2202 	}
2203 	ewhere = where + *sizep;
2204 
2205 	sysctl_unlock();
2206 	mutex_enter(&mountlist_lock);
2207 	for (mp = CIRCLEQ_FIRST(&mountlist); mp != (void *)&mountlist;
2208 	    mp = nmp) {
2209 		if (vfs_busy(mp, &nmp)) {
2210 			continue;
2211 		}
2212 		/* Allocate a marker vnode. */
2213 		mvp = vnalloc(mp);
2214 		/* Should never fail for mp != NULL */
2215 		KASSERT(mvp != NULL);
2216 		mutex_enter(&mntvnode_lock);
2217 		for (vp = TAILQ_FIRST(&mp->mnt_vnodelist); vp;
2218 		    vp = vunmark(mvp)) {
2219 			vmark(mvp, vp);
2220 			/*
2221 			 * Check that the vp is still associated with
2222 			 * this filesystem.  RACE: could have been
2223 			 * recycled onto the same filesystem.
2224 			 */
2225 			if (vp->v_mount != mp || vismarker(vp))
2226 				continue;
2227 			if (bp + VPTRSZ + VNODESZ > ewhere) {
2228 				(void)vunmark(mvp);
2229 				mutex_exit(&mntvnode_lock);
2230 				vnfree(mvp);
2231 				vfs_unbusy(mp, false, NULL);
2232 				sysctl_relock();
2233 				*sizep = bp - where;
2234 				return (ENOMEM);
2235 			}
2236 			memcpy(&vbuf, vp, VNODESZ);
2237 			mutex_exit(&mntvnode_lock);
2238 			if ((error = copyout(&vp, bp, VPTRSZ)) ||
2239 			    (error = copyout(&vbuf, bp + VPTRSZ, VNODESZ))) {
2240 			   	mutex_enter(&mntvnode_lock);
2241 				(void)vunmark(mvp);
2242 				mutex_exit(&mntvnode_lock);
2243 				vnfree(mvp);
2244 				vfs_unbusy(mp, false, NULL);
2245 				sysctl_relock();
2246 				return (error);
2247 			}
2248 			bp += VPTRSZ + VNODESZ;
2249 			mutex_enter(&mntvnode_lock);
2250 		}
2251 		mutex_exit(&mntvnode_lock);
2252 		vnfree(mvp);
2253 		vfs_unbusy(mp, false, &nmp);
2254 	}
2255 	mutex_exit(&mountlist_lock);
2256 	sysctl_relock();
2257 
2258 	*sizep = bp - where;
2259 	return (0);
2260 }
2261 
2262 /*
2263  * Remove clean vnodes from a mountpoint's vnode list.
2264  */
2265 void
2266 vfs_scrubvnlist(struct mount *mp)
2267 {
2268 	vnode_t *vp, *nvp;
2269 
2270  retry:
2271 	mutex_enter(&mntvnode_lock);
2272 	for (vp = TAILQ_FIRST(&mp->mnt_vnodelist); vp; vp = nvp) {
2273 		nvp = TAILQ_NEXT(vp, v_mntvnodes);
2274 		mutex_enter(&vp->v_interlock);
2275 		if ((vp->v_iflag & VI_CLEAN) != 0) {
2276 			TAILQ_REMOVE(&mp->mnt_vnodelist, vp, v_mntvnodes);
2277 			vp->v_mount = NULL;
2278 			mutex_exit(&mntvnode_lock);
2279 			mutex_exit(&vp->v_interlock);
2280 			vfs_destroy(mp);
2281 			goto retry;
2282 		}
2283 		mutex_exit(&vp->v_interlock);
2284 	}
2285 	mutex_exit(&mntvnode_lock);
2286 }
2287 
2288 /*
2289  * Check to see if a filesystem is mounted on a block device.
2290  */
2291 int
2292 vfs_mountedon(vnode_t *vp)
2293 {
2294 	vnode_t *vq;
2295 	int error = 0;
2296 
2297 	if (vp->v_type != VBLK)
2298 		return ENOTBLK;
2299 	if (vp->v_specmountpoint != NULL)
2300 		return (EBUSY);
2301 	mutex_enter(&device_lock);
2302 	for (vq = specfs_hash[SPECHASH(vp->v_rdev)]; vq != NULL;
2303 	    vq = vq->v_specnext) {
2304 		if (vq->v_rdev != vp->v_rdev || vq->v_type != vp->v_type)
2305 			continue;
2306 		if (vq->v_specmountpoint != NULL) {
2307 			error = EBUSY;
2308 			break;
2309 		}
2310 	}
2311 	mutex_exit(&device_lock);
2312 	return (error);
2313 }
2314 
2315 /*
2316  * Unmount all file systems.
2317  * We traverse the list in reverse order under the assumption that doing so
2318  * will avoid needing to worry about dependencies.
2319  */
2320 bool
2321 vfs_unmountall(struct lwp *l)
2322 {
2323 
2324 	printf("unmounting file systems...");
2325 	return vfs_unmountall1(l, true, true);
2326 }
2327 
2328 static void
2329 vfs_unmount_print(struct mount *mp, const char *pfx)
2330 {
2331 
2332 	aprint_verbose("%sunmounted %s on %s type %s\n", pfx,
2333 	    mp->mnt_stat.f_mntfromname, mp->mnt_stat.f_mntonname,
2334 	    mp->mnt_stat.f_fstypename);
2335 }
2336 
2337 bool
2338 vfs_unmount_forceone(struct lwp *l)
2339 {
2340 	struct mount *mp, *nmp;
2341 	int error;
2342 
2343 	nmp = NULL;
2344 
2345 	CIRCLEQ_FOREACH_REVERSE(mp, &mountlist, mnt_list) {
2346 		if (nmp == NULL || mp->mnt_gen > nmp->mnt_gen) {
2347 			nmp = mp;
2348 		}
2349 	}
2350 	if (nmp == NULL) {
2351 		return false;
2352 	}
2353 
2354 #ifdef DEBUG
2355 	printf("\nforcefully unmounting %s (%s)...",
2356 	    nmp->mnt_stat.f_mntonname, nmp->mnt_stat.f_mntfromname);
2357 #endif
2358 	atomic_inc_uint(&nmp->mnt_refcnt);
2359 	if ((error = dounmount(nmp, MNT_FORCE, l)) == 0) {
2360 		vfs_unmount_print(nmp, "forcefully ");
2361 		return true;
2362 	} else {
2363 		vfs_destroy(nmp);
2364 	}
2365 
2366 #ifdef DEBUG
2367 	printf("forceful unmount of %s failed with error %d\n",
2368 	    nmp->mnt_stat.f_mntonname, error);
2369 #endif
2370 
2371 	return false;
2372 }
2373 
2374 bool
2375 vfs_unmountall1(struct lwp *l, bool force, bool verbose)
2376 {
2377 	struct mount *mp, *nmp;
2378 	bool any_error = false, progress = false;
2379 	int error;
2380 
2381 	for (mp = CIRCLEQ_LAST(&mountlist);
2382 	     mp != (void *)&mountlist;
2383 	     mp = nmp) {
2384 		nmp = CIRCLEQ_PREV(mp, mnt_list);
2385 #ifdef DEBUG
2386 		printf("\nunmounting %p %s (%s)...",
2387 		    (void *)mp, mp->mnt_stat.f_mntonname,
2388 		    mp->mnt_stat.f_mntfromname);
2389 #endif
2390 		atomic_inc_uint(&mp->mnt_refcnt);
2391 		if ((error = dounmount(mp, force ? MNT_FORCE : 0, l)) == 0) {
2392 			vfs_unmount_print(mp, "");
2393 			progress = true;
2394 		} else {
2395 			vfs_destroy(mp);
2396 			if (verbose) {
2397 				printf("unmount of %s failed with error %d\n",
2398 				    mp->mnt_stat.f_mntonname, error);
2399 			}
2400 			any_error = true;
2401 		}
2402 	}
2403 	if (verbose) {
2404 		printf(" done\n");
2405 	}
2406 	if (any_error && verbose) {
2407 		printf("WARNING: some file systems would not unmount\n");
2408 	}
2409 	return progress;
2410 }
2411 
2412 /*
2413  * Sync and unmount file systems before shutting down.
2414  */
2415 void
2416 vfs_shutdown(void)
2417 {
2418 	struct lwp *l;
2419 
2420 	/* XXX we're certainly not running in lwp0's context! */
2421 	l = (curlwp == NULL) ? &lwp0 : curlwp;
2422 
2423 	vfs_shutdown1(l);
2424 }
2425 
2426 void
2427 vfs_sync_all(struct lwp *l)
2428 {
2429 	printf("syncing disks... ");
2430 
2431 	/* remove user processes from run queue */
2432 	suspendsched();
2433 	(void) spl0();
2434 
2435 	/* avoid coming back this way again if we panic. */
2436 	doing_shutdown = 1;
2437 
2438 	sys_sync(l, NULL, NULL);
2439 
2440 	/* Wait for sync to finish. */
2441 	if (buf_syncwait() != 0) {
2442 #if defined(DDB) && defined(DEBUG_HALT_BUSY)
2443 		Debugger();
2444 #endif
2445 		printf("giving up\n");
2446 		return;
2447 	} else
2448 		printf("done\n");
2449 }
2450 
2451 static void
2452 vfs_shutdown1(struct lwp *l)
2453 {
2454 
2455 	vfs_sync_all(l);
2456 
2457 	/*
2458 	 * If we've panic'd, don't make the situation potentially
2459 	 * worse by unmounting the file systems.
2460 	 */
2461 	if (panicstr != NULL)
2462 		return;
2463 
2464 	/* Release inodes held by texts before update. */
2465 #ifdef notdef
2466 	vnshutdown();
2467 #endif
2468 	/* Unmount file systems. */
2469 	vfs_unmountall(l);
2470 }
2471 
2472 /*
2473  * Print a list of supported file system types (used by vfs_mountroot)
2474  */
2475 static void
2476 vfs_print_fstypes(void)
2477 {
2478 	struct vfsops *v;
2479 	int cnt = 0;
2480 
2481 	mutex_enter(&vfs_list_lock);
2482 	LIST_FOREACH(v, &vfs_list, vfs_list)
2483 		++cnt;
2484 	mutex_exit(&vfs_list_lock);
2485 
2486 	if (cnt == 0) {
2487 		printf("WARNING: No file system modules have been loaded.\n");
2488 		return;
2489 	}
2490 
2491 	printf("Supported file systems:");
2492 	mutex_enter(&vfs_list_lock);
2493 	LIST_FOREACH(v, &vfs_list, vfs_list) {
2494 		printf(" %s", v->vfs_name);
2495 	}
2496 	mutex_exit(&vfs_list_lock);
2497 	printf("\n");
2498 }
2499 
2500 /*
2501  * Mount the root file system.  If the operator didn't specify a
2502  * file system to use, try all possible file systems until one
2503  * succeeds.
2504  */
2505 int
2506 vfs_mountroot(void)
2507 {
2508 	struct vfsops *v;
2509 	int error = ENODEV;
2510 
2511 	if (root_device == NULL)
2512 		panic("vfs_mountroot: root device unknown");
2513 
2514 	switch (device_class(root_device)) {
2515 	case DV_IFNET:
2516 		if (rootdev != NODEV)
2517 			panic("vfs_mountroot: rootdev set for DV_IFNET "
2518 			    "(0x%llx -> %llu,%llu)",
2519 			    (unsigned long long)rootdev,
2520 			    (unsigned long long)major(rootdev),
2521 			    (unsigned long long)minor(rootdev));
2522 		break;
2523 
2524 	case DV_DISK:
2525 		if (rootdev == NODEV)
2526 			panic("vfs_mountroot: rootdev not set for DV_DISK");
2527 	        if (bdevvp(rootdev, &rootvp))
2528 	                panic("vfs_mountroot: can't get vnode for rootdev");
2529 		error = VOP_OPEN(rootvp, FREAD, FSCRED);
2530 		if (error) {
2531 			printf("vfs_mountroot: can't open root device\n");
2532 			return (error);
2533 		}
2534 		break;
2535 
2536 	case DV_VIRTUAL:
2537 		break;
2538 
2539 	default:
2540 		printf("%s: inappropriate for root file system\n",
2541 		    device_xname(root_device));
2542 		return (ENODEV);
2543 	}
2544 
2545 	/*
2546 	 * If user specified a root fs type, use it.  Make sure the
2547 	 * specified type exists and has a mount_root()
2548 	 */
2549 	if (strcmp(rootfstype, ROOT_FSTYPE_ANY) != 0) {
2550 		v = vfs_getopsbyname(rootfstype);
2551 		error = EFTYPE;
2552 		if (v != NULL) {
2553 			if (v->vfs_mountroot != NULL) {
2554 				error = (v->vfs_mountroot)();
2555 			}
2556 			v->vfs_refcount--;
2557 		}
2558 		goto done;
2559 	}
2560 
2561 	/*
2562 	 * Try each file system currently configured into the kernel.
2563 	 */
2564 	mutex_enter(&vfs_list_lock);
2565 	LIST_FOREACH(v, &vfs_list, vfs_list) {
2566 		if (v->vfs_mountroot == NULL)
2567 			continue;
2568 #ifdef DEBUG
2569 		aprint_normal("mountroot: trying %s...\n", v->vfs_name);
2570 #endif
2571 		v->vfs_refcount++;
2572 		mutex_exit(&vfs_list_lock);
2573 		error = (*v->vfs_mountroot)();
2574 		mutex_enter(&vfs_list_lock);
2575 		v->vfs_refcount--;
2576 		if (!error) {
2577 			aprint_normal("root file system type: %s\n",
2578 			    v->vfs_name);
2579 			break;
2580 		}
2581 	}
2582 	mutex_exit(&vfs_list_lock);
2583 
2584 	if (v == NULL) {
2585 		vfs_print_fstypes();
2586 		printf("no file system for %s", device_xname(root_device));
2587 		if (device_class(root_device) == DV_DISK)
2588 			printf(" (dev 0x%llx)", (unsigned long long)rootdev);
2589 		printf("\n");
2590 		error = EFTYPE;
2591 	}
2592 
2593 done:
2594 	if (error && device_class(root_device) == DV_DISK) {
2595 		VOP_CLOSE(rootvp, FREAD, FSCRED);
2596 		vrele(rootvp);
2597 	}
2598 	if (error == 0) {
2599 		extern struct cwdinfo cwdi0;
2600 
2601 		CIRCLEQ_FIRST(&mountlist)->mnt_flag |= MNT_ROOTFS;
2602 		CIRCLEQ_FIRST(&mountlist)->mnt_op->vfs_refcount++;
2603 
2604 		/*
2605 		 * Get the vnode for '/'.  Set cwdi0.cwdi_cdir to
2606 		 * reference it.
2607 		 */
2608 		error = VFS_ROOT(CIRCLEQ_FIRST(&mountlist), &rootvnode);
2609 		if (error)
2610 			panic("cannot find root vnode, error=%d", error);
2611 		cwdi0.cwdi_cdir = rootvnode;
2612 		vref(cwdi0.cwdi_cdir);
2613 		VOP_UNLOCK(rootvnode, 0);
2614 		cwdi0.cwdi_rdir = NULL;
2615 
2616 		/*
2617 		 * Now that root is mounted, we can fixup initproc's CWD
2618 		 * info.  All other processes are kthreads, which merely
2619 		 * share proc0's CWD info.
2620 		 */
2621 		initproc->p_cwdi->cwdi_cdir = rootvnode;
2622 		vref(initproc->p_cwdi->cwdi_cdir);
2623 		initproc->p_cwdi->cwdi_rdir = NULL;
2624 		/*
2625 		 * Enable loading of modules from the filesystem
2626 		 */
2627 		module_load_vfs_init();
2628 
2629 	}
2630 	return (error);
2631 }
2632 
2633 /*
2634  * Get a new unique fsid
2635  */
2636 void
2637 vfs_getnewfsid(struct mount *mp)
2638 {
2639 	static u_short xxxfs_mntid;
2640 	fsid_t tfsid;
2641 	int mtype;
2642 
2643 	mutex_enter(&mntid_lock);
2644 	mtype = makefstype(mp->mnt_op->vfs_name);
2645 	mp->mnt_stat.f_fsidx.__fsid_val[0] = makedev(mtype, 0);
2646 	mp->mnt_stat.f_fsidx.__fsid_val[1] = mtype;
2647 	mp->mnt_stat.f_fsid = mp->mnt_stat.f_fsidx.__fsid_val[0];
2648 	if (xxxfs_mntid == 0)
2649 		++xxxfs_mntid;
2650 	tfsid.__fsid_val[0] = makedev(mtype & 0xff, xxxfs_mntid);
2651 	tfsid.__fsid_val[1] = mtype;
2652 	if (!CIRCLEQ_EMPTY(&mountlist)) {
2653 		while (vfs_getvfs(&tfsid)) {
2654 			tfsid.__fsid_val[0]++;
2655 			xxxfs_mntid++;
2656 		}
2657 	}
2658 	mp->mnt_stat.f_fsidx.__fsid_val[0] = tfsid.__fsid_val[0];
2659 	mp->mnt_stat.f_fsid = mp->mnt_stat.f_fsidx.__fsid_val[0];
2660 	mutex_exit(&mntid_lock);
2661 }
2662 
2663 /*
2664  * Make a 'unique' number from a mount type name.
2665  */
2666 long
2667 makefstype(const char *type)
2668 {
2669 	long rv;
2670 
2671 	for (rv = 0; *type; type++) {
2672 		rv <<= 2;
2673 		rv ^= *type;
2674 	}
2675 	return rv;
2676 }
2677 
2678 /*
2679  * Set vnode attributes to VNOVAL
2680  */
2681 void
2682 vattr_null(struct vattr *vap)
2683 {
2684 
2685 	memset(vap, 0, sizeof(*vap));
2686 
2687 	vap->va_type = VNON;
2688 
2689 	/*
2690 	 * Assign individually so that it is safe even if size and
2691 	 * sign of each member are varied.
2692 	 */
2693 	vap->va_mode = VNOVAL;
2694 	vap->va_nlink = VNOVAL;
2695 	vap->va_uid = VNOVAL;
2696 	vap->va_gid = VNOVAL;
2697 	vap->va_fsid = VNOVAL;
2698 	vap->va_fileid = VNOVAL;
2699 	vap->va_size = VNOVAL;
2700 	vap->va_blocksize = VNOVAL;
2701 	vap->va_atime.tv_sec =
2702 	    vap->va_mtime.tv_sec =
2703 	    vap->va_ctime.tv_sec =
2704 	    vap->va_birthtime.tv_sec = VNOVAL;
2705 	vap->va_atime.tv_nsec =
2706 	    vap->va_mtime.tv_nsec =
2707 	    vap->va_ctime.tv_nsec =
2708 	    vap->va_birthtime.tv_nsec = VNOVAL;
2709 	vap->va_gen = VNOVAL;
2710 	vap->va_flags = VNOVAL;
2711 	vap->va_rdev = VNOVAL;
2712 	vap->va_bytes = VNOVAL;
2713 }
2714 
2715 #define ARRAY_SIZE(arr) (sizeof(arr) / sizeof(arr[0]))
2716 #define ARRAY_PRINT(idx, arr) \
2717     ((unsigned int)(idx) < ARRAY_SIZE(arr) ? (arr)[(idx)] : "UNKNOWN")
2718 
2719 const char * const vnode_tags[] = { VNODE_TAGS };
2720 const char * const vnode_types[] = { VNODE_TYPES };
2721 const char vnode_flagbits[] = VNODE_FLAGBITS;
2722 
2723 /*
2724  * Print out a description of a vnode.
2725  */
2726 void
2727 vprint(const char *label, struct vnode *vp)
2728 {
2729 	struct vnlock *vl;
2730 	char bf[96];
2731 	int flag;
2732 
2733 	vl = &vp->v_lock;
2734 	flag = vp->v_iflag | vp->v_vflag | vp->v_uflag;
2735 	snprintb(bf, sizeof(bf), vnode_flagbits, flag);
2736 
2737 	if (label != NULL)
2738 		printf("%s: ", label);
2739 	printf("vnode @ %p, flags (%s)\n\ttag %s(%d), type %s(%d), "
2740 	    "usecount %d, writecount %d, holdcount %d\n"
2741 	    "\tfreelisthd %p, mount %p, data %p lock %p recursecnt %d\n",
2742 	    vp, bf, ARRAY_PRINT(vp->v_tag, vnode_tags), vp->v_tag,
2743 	    ARRAY_PRINT(vp->v_type, vnode_types), vp->v_type,
2744 	    vp->v_usecount, vp->v_writecount, vp->v_holdcnt,
2745 	    vp->v_freelisthd, vp->v_mount, vp->v_data, vl, vl->vl_recursecnt);
2746 	if (vp->v_data != NULL) {
2747 		printf("\t");
2748 		VOP_PRINT(vp);
2749 	}
2750 }
2751 
2752 #ifdef DEBUG
2753 /*
2754  * List all of the locked vnodes in the system.
2755  * Called when debugging the kernel.
2756  */
2757 void
2758 printlockedvnodes(void)
2759 {
2760 	struct mount *mp, *nmp;
2761 	struct vnode *vp;
2762 
2763 	printf("Locked vnodes\n");
2764 	mutex_enter(&mountlist_lock);
2765 	for (mp = CIRCLEQ_FIRST(&mountlist); mp != (void *)&mountlist;
2766 	     mp = nmp) {
2767 		if (vfs_busy(mp, &nmp)) {
2768 			continue;
2769 		}
2770 		TAILQ_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
2771 			if (VOP_ISLOCKED(vp))
2772 				vprint(NULL, vp);
2773 		}
2774 		mutex_enter(&mountlist_lock);
2775 		vfs_unbusy(mp, false, &nmp);
2776 	}
2777 	mutex_exit(&mountlist_lock);
2778 }
2779 #endif
2780 
2781 /* Deprecated. Kept for KPI compatibility. */
2782 int
2783 vaccess(enum vtype type, mode_t file_mode, uid_t uid, gid_t gid,
2784     mode_t acc_mode, kauth_cred_t cred)
2785 {
2786 
2787 #ifdef DIAGNOSTIC
2788 	printf("vaccess: deprecated interface used.\n");
2789 #endif /* DIAGNOSTIC */
2790 
2791 	return genfs_can_access(type, file_mode, uid, gid, acc_mode, cred);
2792 }
2793 
2794 /*
2795  * Given a file system name, look up the vfsops for that
2796  * file system, or return NULL if file system isn't present
2797  * in the kernel.
2798  */
2799 struct vfsops *
2800 vfs_getopsbyname(const char *name)
2801 {
2802 	struct vfsops *v;
2803 
2804 	mutex_enter(&vfs_list_lock);
2805 	LIST_FOREACH(v, &vfs_list, vfs_list) {
2806 		if (strcmp(v->vfs_name, name) == 0)
2807 			break;
2808 	}
2809 	if (v != NULL)
2810 		v->vfs_refcount++;
2811 	mutex_exit(&vfs_list_lock);
2812 
2813 	return (v);
2814 }
2815 
2816 void
2817 copy_statvfs_info(struct statvfs *sbp, const struct mount *mp)
2818 {
2819 	const struct statvfs *mbp;
2820 
2821 	if (sbp == (mbp = &mp->mnt_stat))
2822 		return;
2823 
2824 	(void)memcpy(&sbp->f_fsidx, &mbp->f_fsidx, sizeof(sbp->f_fsidx));
2825 	sbp->f_fsid = mbp->f_fsid;
2826 	sbp->f_owner = mbp->f_owner;
2827 	sbp->f_flag = mbp->f_flag;
2828 	sbp->f_syncwrites = mbp->f_syncwrites;
2829 	sbp->f_asyncwrites = mbp->f_asyncwrites;
2830 	sbp->f_syncreads = mbp->f_syncreads;
2831 	sbp->f_asyncreads = mbp->f_asyncreads;
2832 	(void)memcpy(sbp->f_spare, mbp->f_spare, sizeof(mbp->f_spare));
2833 	(void)memcpy(sbp->f_fstypename, mbp->f_fstypename,
2834 	    sizeof(sbp->f_fstypename));
2835 	(void)memcpy(sbp->f_mntonname, mbp->f_mntonname,
2836 	    sizeof(sbp->f_mntonname));
2837 	(void)memcpy(sbp->f_mntfromname, mp->mnt_stat.f_mntfromname,
2838 	    sizeof(sbp->f_mntfromname));
2839 	sbp->f_namemax = mbp->f_namemax;
2840 }
2841 
2842 int
2843 set_statvfs_info(const char *onp, int ukon, const char *fromp, int ukfrom,
2844     const char *vfsname, struct mount *mp, struct lwp *l)
2845 {
2846 	int error;
2847 	size_t size;
2848 	struct statvfs *sfs = &mp->mnt_stat;
2849 	int (*fun)(const void *, void *, size_t, size_t *);
2850 
2851 	(void)strlcpy(mp->mnt_stat.f_fstypename, vfsname,
2852 	    sizeof(mp->mnt_stat.f_fstypename));
2853 
2854 	if (onp) {
2855 		struct cwdinfo *cwdi = l->l_proc->p_cwdi;
2856 		fun = (ukon == UIO_SYSSPACE) ? copystr : copyinstr;
2857 		if (cwdi->cwdi_rdir != NULL) {
2858 			size_t len;
2859 			char *bp;
2860 			char *path = PNBUF_GET();
2861 
2862 			bp = path + MAXPATHLEN;
2863 			*--bp = '\0';
2864 			rw_enter(&cwdi->cwdi_lock, RW_READER);
2865 			error = getcwd_common(cwdi->cwdi_rdir, rootvnode, &bp,
2866 			    path, MAXPATHLEN / 2, 0, l);
2867 			rw_exit(&cwdi->cwdi_lock);
2868 			if (error) {
2869 				PNBUF_PUT(path);
2870 				return error;
2871 			}
2872 
2873 			len = strlen(bp);
2874 			if (len > sizeof(sfs->f_mntonname) - 1)
2875 				len = sizeof(sfs->f_mntonname) - 1;
2876 			(void)strncpy(sfs->f_mntonname, bp, len);
2877 			PNBUF_PUT(path);
2878 
2879 			if (len < sizeof(sfs->f_mntonname) - 1) {
2880 				error = (*fun)(onp, &sfs->f_mntonname[len],
2881 				    sizeof(sfs->f_mntonname) - len - 1, &size);
2882 				if (error)
2883 					return error;
2884 				size += len;
2885 			} else {
2886 				size = len;
2887 			}
2888 		} else {
2889 			error = (*fun)(onp, &sfs->f_mntonname,
2890 			    sizeof(sfs->f_mntonname) - 1, &size);
2891 			if (error)
2892 				return error;
2893 		}
2894 		(void)memset(sfs->f_mntonname + size, 0,
2895 		    sizeof(sfs->f_mntonname) - size);
2896 	}
2897 
2898 	if (fromp) {
2899 		fun = (ukfrom == UIO_SYSSPACE) ? copystr : copyinstr;
2900 		error = (*fun)(fromp, sfs->f_mntfromname,
2901 		    sizeof(sfs->f_mntfromname) - 1, &size);
2902 		if (error)
2903 			return error;
2904 		(void)memset(sfs->f_mntfromname + size, 0,
2905 		    sizeof(sfs->f_mntfromname) - size);
2906 	}
2907 	return 0;
2908 }
2909 
2910 void
2911 vfs_timestamp(struct timespec *ts)
2912 {
2913 
2914 	nanotime(ts);
2915 }
2916 
2917 time_t	rootfstime;			/* recorded root fs time, if known */
2918 void
2919 setrootfstime(time_t t)
2920 {
2921 	rootfstime = t;
2922 }
2923 
2924 /*
2925  * Sham lock manager for vnodes.  This is a temporary measure.
2926  */
2927 int
2928 vlockmgr(struct vnlock *vl, int flags)
2929 {
2930 
2931 	KASSERT((flags & ~(LK_CANRECURSE | LK_NOWAIT | LK_TYPE_MASK)) == 0);
2932 
2933 	switch (flags & LK_TYPE_MASK) {
2934 	case LK_SHARED:
2935 		if (rw_tryenter(&vl->vl_lock, RW_READER)) {
2936 			return 0;
2937 		}
2938 		if ((flags & LK_NOWAIT) != 0) {
2939 			return EBUSY;
2940 		}
2941 		rw_enter(&vl->vl_lock, RW_READER);
2942 		return 0;
2943 
2944 	case LK_EXCLUSIVE:
2945 		if (rw_tryenter(&vl->vl_lock, RW_WRITER)) {
2946 			return 0;
2947 		}
2948 		if ((vl->vl_canrecurse || (flags & LK_CANRECURSE) != 0) &&
2949 		    rw_write_held(&vl->vl_lock)) {
2950 			vl->vl_recursecnt++;
2951 			return 0;
2952 		}
2953 		if ((flags & LK_NOWAIT) != 0) {
2954 			return EBUSY;
2955 		}
2956 		rw_enter(&vl->vl_lock, RW_WRITER);
2957 		return 0;
2958 
2959 	case LK_RELEASE:
2960 		if (vl->vl_recursecnt != 0) {
2961 			KASSERT(rw_write_held(&vl->vl_lock));
2962 			vl->vl_recursecnt--;
2963 			return 0;
2964 		}
2965 		rw_exit(&vl->vl_lock);
2966 		return 0;
2967 
2968 	default:
2969 		panic("vlockmgr: flags %x", flags);
2970 	}
2971 }
2972 
2973 int
2974 vlockstatus(struct vnlock *vl)
2975 {
2976 
2977 	if (rw_write_held(&vl->vl_lock)) {
2978 		return LK_EXCLUSIVE;
2979 	}
2980 	if (rw_read_held(&vl->vl_lock)) {
2981 		return LK_SHARED;
2982 	}
2983 	return 0;
2984 }
2985 
2986 static const uint8_t vttodt_tab[9] = {
2987 	DT_UNKNOWN,	/* VNON  */
2988 	DT_REG,		/* VREG  */
2989 	DT_DIR,		/* VDIR  */
2990 	DT_BLK,		/* VBLK  */
2991 	DT_CHR,		/* VCHR  */
2992 	DT_LNK,		/* VLNK  */
2993 	DT_SOCK,	/* VSUCK */
2994 	DT_FIFO,	/* VFIFO */
2995 	DT_UNKNOWN	/* VBAD  */
2996 };
2997 
2998 uint8_t
2999 vtype2dt(enum vtype vt)
3000 {
3001 
3002 	CTASSERT(VBAD == __arraycount(vttodt_tab) - 1);
3003 	return vttodt_tab[vt];
3004 }
3005 
3006 /*
3007  * mount_specific_key_create --
3008  *	Create a key for subsystem mount-specific data.
3009  */
3010 int
3011 mount_specific_key_create(specificdata_key_t *keyp, specificdata_dtor_t dtor)
3012 {
3013 
3014 	return (specificdata_key_create(mount_specificdata_domain, keyp, dtor));
3015 }
3016 
3017 /*
3018  * mount_specific_key_delete --
3019  *	Delete a key for subsystem mount-specific data.
3020  */
3021 void
3022 mount_specific_key_delete(specificdata_key_t key)
3023 {
3024 
3025 	specificdata_key_delete(mount_specificdata_domain, key);
3026 }
3027 
3028 /*
3029  * mount_initspecific --
3030  *	Initialize a mount's specificdata container.
3031  */
3032 void
3033 mount_initspecific(struct mount *mp)
3034 {
3035 	int error;
3036 
3037 	error = specificdata_init(mount_specificdata_domain,
3038 				  &mp->mnt_specdataref);
3039 	KASSERT(error == 0);
3040 }
3041 
3042 /*
3043  * mount_finispecific --
3044  *	Finalize a mount's specificdata container.
3045  */
3046 void
3047 mount_finispecific(struct mount *mp)
3048 {
3049 
3050 	specificdata_fini(mount_specificdata_domain, &mp->mnt_specdataref);
3051 }
3052 
3053 /*
3054  * mount_getspecific --
3055  *	Return mount-specific data corresponding to the specified key.
3056  */
3057 void *
3058 mount_getspecific(struct mount *mp, specificdata_key_t key)
3059 {
3060 
3061 	return (specificdata_getspecific(mount_specificdata_domain,
3062 					 &mp->mnt_specdataref, key));
3063 }
3064 
3065 /*
3066  * mount_setspecific --
3067  *	Set mount-specific data corresponding to the specified key.
3068  */
3069 void
3070 mount_setspecific(struct mount *mp, specificdata_key_t key, void *data)
3071 {
3072 
3073 	specificdata_setspecific(mount_specificdata_domain,
3074 				 &mp->mnt_specdataref, key, data);
3075 }
3076 
3077 int
3078 VFS_MOUNT(struct mount *mp, const char *a, void *b, size_t *c)
3079 {
3080 	int error;
3081 
3082 	KERNEL_LOCK(1, NULL);
3083 	error = (*(mp->mnt_op->vfs_mount))(mp, a, b, c);
3084 	KERNEL_UNLOCK_ONE(NULL);
3085 
3086 	return error;
3087 }
3088 
3089 int
3090 VFS_START(struct mount *mp, int a)
3091 {
3092 	int error;
3093 
3094 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
3095 		KERNEL_LOCK(1, NULL);
3096 	}
3097 	error = (*(mp->mnt_op->vfs_start))(mp, a);
3098 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
3099 		KERNEL_UNLOCK_ONE(NULL);
3100 	}
3101 
3102 	return error;
3103 }
3104 
3105 int
3106 VFS_UNMOUNT(struct mount *mp, int a)
3107 {
3108 	int error;
3109 
3110 	KERNEL_LOCK(1, NULL);
3111 	error = (*(mp->mnt_op->vfs_unmount))(mp, a);
3112 	KERNEL_UNLOCK_ONE(NULL);
3113 
3114 	return error;
3115 }
3116 
3117 int
3118 VFS_ROOT(struct mount *mp, struct vnode **a)
3119 {
3120 	int error;
3121 
3122 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
3123 		KERNEL_LOCK(1, NULL);
3124 	}
3125 	error = (*(mp->mnt_op->vfs_root))(mp, a);
3126 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
3127 		KERNEL_UNLOCK_ONE(NULL);
3128 	}
3129 
3130 	return error;
3131 }
3132 
3133 int
3134 VFS_QUOTACTL(struct mount *mp, int a, uid_t b, void *c)
3135 {
3136 	int error;
3137 
3138 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
3139 		KERNEL_LOCK(1, NULL);
3140 	}
3141 	error = (*(mp->mnt_op->vfs_quotactl))(mp, a, b, c);
3142 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
3143 		KERNEL_UNLOCK_ONE(NULL);
3144 	}
3145 
3146 	return error;
3147 }
3148 
3149 int
3150 VFS_STATVFS(struct mount *mp, struct statvfs *a)
3151 {
3152 	int error;
3153 
3154 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
3155 		KERNEL_LOCK(1, NULL);
3156 	}
3157 	error = (*(mp->mnt_op->vfs_statvfs))(mp, a);
3158 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
3159 		KERNEL_UNLOCK_ONE(NULL);
3160 	}
3161 
3162 	return error;
3163 }
3164 
3165 int
3166 VFS_SYNC(struct mount *mp, int a, struct kauth_cred *b)
3167 {
3168 	int error;
3169 
3170 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
3171 		KERNEL_LOCK(1, NULL);
3172 	}
3173 	error = (*(mp->mnt_op->vfs_sync))(mp, a, b);
3174 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
3175 		KERNEL_UNLOCK_ONE(NULL);
3176 	}
3177 
3178 	return error;
3179 }
3180 
3181 int
3182 VFS_FHTOVP(struct mount *mp, struct fid *a, struct vnode **b)
3183 {
3184 	int error;
3185 
3186 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
3187 		KERNEL_LOCK(1, NULL);
3188 	}
3189 	error = (*(mp->mnt_op->vfs_fhtovp))(mp, a, b);
3190 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
3191 		KERNEL_UNLOCK_ONE(NULL);
3192 	}
3193 
3194 	return error;
3195 }
3196 
3197 int
3198 VFS_VPTOFH(struct vnode *vp, struct fid *a, size_t *b)
3199 {
3200 	int error;
3201 
3202 	if ((vp->v_vflag & VV_MPSAFE) == 0) {
3203 		KERNEL_LOCK(1, NULL);
3204 	}
3205 	error = (*(vp->v_mount->mnt_op->vfs_vptofh))(vp, a, b);
3206 	if ((vp->v_vflag & VV_MPSAFE) == 0) {
3207 		KERNEL_UNLOCK_ONE(NULL);
3208 	}
3209 
3210 	return error;
3211 }
3212 
3213 int
3214 VFS_SNAPSHOT(struct mount *mp, struct vnode *a, struct timespec *b)
3215 {
3216 	int error;
3217 
3218 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
3219 		KERNEL_LOCK(1, NULL);
3220 	}
3221 	error = (*(mp->mnt_op->vfs_snapshot))(mp, a, b);
3222 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
3223 		KERNEL_UNLOCK_ONE(NULL);
3224 	}
3225 
3226 	return error;
3227 }
3228 
3229 int
3230 VFS_EXTATTRCTL(struct mount *mp, int a, struct vnode *b, int c, const char *d)
3231 {
3232 	int error;
3233 
3234 	KERNEL_LOCK(1, NULL);		/* XXXSMP check ffs */
3235 	error = (*(mp->mnt_op->vfs_extattrctl))(mp, a, b, c, d);
3236 	KERNEL_UNLOCK_ONE(NULL);	/* XXX */
3237 
3238 	return error;
3239 }
3240 
3241 int
3242 VFS_SUSPENDCTL(struct mount *mp, int a)
3243 {
3244 	int error;
3245 
3246 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
3247 		KERNEL_LOCK(1, NULL);
3248 	}
3249 	error = (*(mp->mnt_op->vfs_suspendctl))(mp, a);
3250 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
3251 		KERNEL_UNLOCK_ONE(NULL);
3252 	}
3253 
3254 	return error;
3255 }
3256 
3257 #if defined(DDB) || defined(DEBUGPRINT)
3258 static const char buf_flagbits[] = BUF_FLAGBITS;
3259 
3260 void
3261 vfs_buf_print(struct buf *bp, int full, void (*pr)(const char *, ...))
3262 {
3263 	char bf[1024];
3264 
3265 	(*pr)("  vp %p lblkno 0x%"PRIx64" blkno 0x%"PRIx64" rawblkno 0x%"
3266 	    PRIx64 " dev 0x%x\n",
3267 	    bp->b_vp, bp->b_lblkno, bp->b_blkno, bp->b_rawblkno, bp->b_dev);
3268 
3269 	snprintb(bf, sizeof(bf),
3270 	    buf_flagbits, bp->b_flags | bp->b_oflags | bp->b_cflags);
3271 	(*pr)("  error %d flags 0x%s\n", bp->b_error, bf);
3272 
3273 	(*pr)("  bufsize 0x%lx bcount 0x%lx resid 0x%lx\n",
3274 		  bp->b_bufsize, bp->b_bcount, bp->b_resid);
3275 	(*pr)("  data %p saveaddr %p\n",
3276 		  bp->b_data, bp->b_saveaddr);
3277 	(*pr)("  iodone %p objlock %p\n", bp->b_iodone, bp->b_objlock);
3278 }
3279 
3280 
3281 void
3282 vfs_vnode_print(struct vnode *vp, int full, void (*pr)(const char *, ...))
3283 {
3284 	char bf[256];
3285 
3286 	uvm_object_printit(&vp->v_uobj, full, pr);
3287 	snprintb(bf, sizeof(bf),
3288 	    vnode_flagbits, vp->v_iflag | vp->v_vflag | vp->v_uflag);
3289 	(*pr)("\nVNODE flags %s\n", bf);
3290 	(*pr)("mp %p numoutput %d size 0x%llx writesize 0x%llx\n",
3291 	      vp->v_mount, vp->v_numoutput, vp->v_size, vp->v_writesize);
3292 
3293 	(*pr)("data %p writecount %ld holdcnt %ld\n",
3294 	      vp->v_data, vp->v_writecount, vp->v_holdcnt);
3295 
3296 	(*pr)("tag %s(%d) type %s(%d) mount %p typedata %p\n",
3297 	      ARRAY_PRINT(vp->v_tag, vnode_tags), vp->v_tag,
3298 	      ARRAY_PRINT(vp->v_type, vnode_types), vp->v_type,
3299 	      vp->v_mount, vp->v_mountedhere);
3300 
3301 	(*pr)("v_lock %p\n", &vp->v_lock);
3302 
3303 	if (full) {
3304 		struct buf *bp;
3305 
3306 		(*pr)("clean bufs:\n");
3307 		LIST_FOREACH(bp, &vp->v_cleanblkhd, b_vnbufs) {
3308 			(*pr)(" bp %p\n", bp);
3309 			vfs_buf_print(bp, full, pr);
3310 		}
3311 
3312 		(*pr)("dirty bufs:\n");
3313 		LIST_FOREACH(bp, &vp->v_dirtyblkhd, b_vnbufs) {
3314 			(*pr)(" bp %p\n", bp);
3315 			vfs_buf_print(bp, full, pr);
3316 		}
3317 	}
3318 }
3319 
3320 void
3321 vfs_mount_print(struct mount *mp, int full, void (*pr)(const char *, ...))
3322 {
3323 	char sbuf[256];
3324 
3325 	(*pr)("vnodecovered = %p syncer = %p data = %p\n",
3326 			mp->mnt_vnodecovered,mp->mnt_syncer,mp->mnt_data);
3327 
3328 	(*pr)("fs_bshift %d dev_bshift = %d\n",
3329 			mp->mnt_fs_bshift,mp->mnt_dev_bshift);
3330 
3331 	snprintb(sbuf, sizeof(sbuf), __MNT_FLAG_BITS, mp->mnt_flag);
3332 	(*pr)("flag = %s\n", sbuf);
3333 
3334 	snprintb(sbuf, sizeof(sbuf), __IMNT_FLAG_BITS, mp->mnt_iflag);
3335 	(*pr)("iflag = %s\n", sbuf);
3336 
3337 	(*pr)("refcnt = %d unmounting @ %p updating @ %p\n", mp->mnt_refcnt,
3338 	    &mp->mnt_unmounting, &mp->mnt_updating);
3339 
3340 	(*pr)("statvfs cache:\n");
3341 	(*pr)("\tbsize = %lu\n",mp->mnt_stat.f_bsize);
3342 	(*pr)("\tfrsize = %lu\n",mp->mnt_stat.f_frsize);
3343 	(*pr)("\tiosize = %lu\n",mp->mnt_stat.f_iosize);
3344 
3345 	(*pr)("\tblocks = %"PRIu64"\n",mp->mnt_stat.f_blocks);
3346 	(*pr)("\tbfree = %"PRIu64"\n",mp->mnt_stat.f_bfree);
3347 	(*pr)("\tbavail = %"PRIu64"\n",mp->mnt_stat.f_bavail);
3348 	(*pr)("\tbresvd = %"PRIu64"\n",mp->mnt_stat.f_bresvd);
3349 
3350 	(*pr)("\tfiles = %"PRIu64"\n",mp->mnt_stat.f_files);
3351 	(*pr)("\tffree = %"PRIu64"\n",mp->mnt_stat.f_ffree);
3352 	(*pr)("\tfavail = %"PRIu64"\n",mp->mnt_stat.f_favail);
3353 	(*pr)("\tfresvd = %"PRIu64"\n",mp->mnt_stat.f_fresvd);
3354 
3355 	(*pr)("\tf_fsidx = { 0x%"PRIx32", 0x%"PRIx32" }\n",
3356 			mp->mnt_stat.f_fsidx.__fsid_val[0],
3357 			mp->mnt_stat.f_fsidx.__fsid_val[1]);
3358 
3359 	(*pr)("\towner = %"PRIu32"\n",mp->mnt_stat.f_owner);
3360 	(*pr)("\tnamemax = %lu\n",mp->mnt_stat.f_namemax);
3361 
3362 	snprintb(sbuf, sizeof(sbuf), __MNT_FLAG_BITS, mp->mnt_stat.f_flag);
3363 
3364 	(*pr)("\tflag = %s\n",sbuf);
3365 	(*pr)("\tsyncwrites = %" PRIu64 "\n",mp->mnt_stat.f_syncwrites);
3366 	(*pr)("\tasyncwrites = %" PRIu64 "\n",mp->mnt_stat.f_asyncwrites);
3367 	(*pr)("\tsyncreads = %" PRIu64 "\n",mp->mnt_stat.f_syncreads);
3368 	(*pr)("\tasyncreads = %" PRIu64 "\n",mp->mnt_stat.f_asyncreads);
3369 	(*pr)("\tfstypename = %s\n",mp->mnt_stat.f_fstypename);
3370 	(*pr)("\tmntonname = %s\n",mp->mnt_stat.f_mntonname);
3371 	(*pr)("\tmntfromname = %s\n",mp->mnt_stat.f_mntfromname);
3372 
3373 	{
3374 		int cnt = 0;
3375 		struct vnode *vp;
3376 		(*pr)("locked vnodes =");
3377 		TAILQ_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
3378 			if (VOP_ISLOCKED(vp)) {
3379 				if ((++cnt % 6) == 0) {
3380 					(*pr)(" %p,\n\t", vp);
3381 				} else {
3382 					(*pr)(" %p,", vp);
3383 				}
3384 			}
3385 		}
3386 		(*pr)("\n");
3387 	}
3388 
3389 	if (full) {
3390 		int cnt = 0;
3391 		struct vnode *vp;
3392 		(*pr)("all vnodes =");
3393 		TAILQ_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
3394 			if (!TAILQ_NEXT(vp, v_mntvnodes)) {
3395 				(*pr)(" %p", vp);
3396 			} else if ((++cnt % 6) == 0) {
3397 				(*pr)(" %p,\n\t", vp);
3398 			} else {
3399 				(*pr)(" %p,", vp);
3400 			}
3401 		}
3402 		(*pr)("\n", vp);
3403 	}
3404 }
3405 #endif /* DDB || DEBUGPRINT */
3406 
3407 /*
3408  * Check if a device pointed to by vp is mounted.
3409  *
3410  * Returns:
3411  *   EINVAL	if it's not a disk
3412  *   EBUSY	if it's a disk and mounted
3413  *   0		if it's a disk and not mounted
3414  */
3415 int
3416 rawdev_mounted(struct vnode *vp, struct vnode **bvpp)
3417 {
3418 	struct vnode *bvp;
3419 	dev_t dev;
3420 	int d_type;
3421 
3422 	bvp = NULL;
3423 	dev = vp->v_rdev;
3424 	d_type = D_OTHER;
3425 
3426 	if (iskmemvp(vp))
3427 		return EINVAL;
3428 
3429 	switch (vp->v_type) {
3430 	case VCHR: {
3431 		const struct cdevsw *cdev;
3432 
3433 		cdev = cdevsw_lookup(dev);
3434 		if (cdev != NULL) {
3435 			dev_t blkdev;
3436 
3437 			blkdev = devsw_chr2blk(dev);
3438 			if (blkdev != NODEV) {
3439 				vfinddev(blkdev, VBLK, &bvp);
3440 				if (bvp != NULL)
3441 					d_type = (cdev->d_flag & D_TYPEMASK);
3442 			}
3443 		}
3444 
3445 		break;
3446 		}
3447 
3448 	case VBLK: {
3449 		const struct bdevsw *bdev;
3450 
3451 		bdev = bdevsw_lookup(dev);
3452 		if (bdev != NULL)
3453 			d_type = (bdev->d_flag & D_TYPEMASK);
3454 
3455 		bvp = vp;
3456 
3457 		break;
3458 		}
3459 
3460 	default:
3461 		break;
3462 	}
3463 
3464 	if (d_type != D_DISK)
3465 		return EINVAL;
3466 
3467 	if (bvpp != NULL)
3468 		*bvpp = bvp;
3469 
3470 	/*
3471 	 * XXX: This is bogus. We should be failing the request
3472 	 * XXX: not only if this specific slice is mounted, but
3473 	 * XXX: if it's on a disk with any other mounted slice.
3474 	 */
3475 	if (vfs_mountedon(bvp))
3476 		return EBUSY;
3477 
3478 	return 0;
3479 }
3480