xref: /netbsd-src/sys/kern/vfs_subr.c (revision bbde328be4e75ea9ad02e9715ea13ca54b797ada)
1 /*	$NetBSD: vfs_subr.c,v 1.400 2010/04/30 10:03:13 pooka 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 1.400 2010/04/30 10:03:13 pooka 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 
123 #include <miscfs/genfs/genfs.h>
124 #include <miscfs/specfs/specdev.h>
125 #include <miscfs/syncfs/syncfs.h>
126 
127 #include <uvm/uvm.h>
128 #include <uvm/uvm_readahead.h>
129 #include <uvm/uvm_ddb.h>
130 
131 #include <sys/sysctl.h>
132 
133 const enum vtype iftovt_tab[16] = {
134 	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
135 	VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
136 };
137 const int	vttoif_tab[9] = {
138 	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
139 	S_IFSOCK, S_IFIFO, S_IFMT,
140 };
141 
142 /*
143  * Insq/Remq for the vnode usage lists.
144  */
145 #define	bufinsvn(bp, dp)	LIST_INSERT_HEAD(dp, bp, b_vnbufs)
146 #define	bufremvn(bp) {							\
147 	LIST_REMOVE(bp, b_vnbufs);					\
148 	(bp)->b_vnbufs.le_next = NOLIST;				\
149 }
150 
151 int doforce = 1;		/* 1 => permit forcible unmounting */
152 int prtactive = 0;		/* 1 => print out reclaim of active vnodes */
153 
154 static vnodelst_t vnode_free_list = TAILQ_HEAD_INITIALIZER(vnode_free_list);
155 static vnodelst_t vnode_hold_list = TAILQ_HEAD_INITIALIZER(vnode_hold_list);
156 static vnodelst_t vrele_list = TAILQ_HEAD_INITIALIZER(vrele_list);
157 
158 struct mntlist mountlist =			/* mounted filesystem list */
159     CIRCLEQ_HEAD_INITIALIZER(mountlist);
160 
161 u_int numvnodes;
162 static specificdata_domain_t mount_specificdata_domain;
163 
164 static int vrele_pending;
165 static int vrele_gen;
166 static kmutex_t	vrele_lock;
167 static kcondvar_t vrele_cv;
168 static lwp_t *vrele_lwp;
169 
170 static uint64_t mountgen = 0;
171 static kmutex_t mountgen_lock;
172 
173 kmutex_t mountlist_lock;
174 kmutex_t mntid_lock;
175 kmutex_t mntvnode_lock;
176 kmutex_t vnode_free_list_lock;
177 kmutex_t vfs_list_lock;
178 
179 static pool_cache_t vnode_cache;
180 
181 /*
182  * These define the root filesystem and device.
183  */
184 struct vnode *rootvnode;
185 struct device *root_device;			/* root device */
186 
187 /*
188  * Local declarations.
189  */
190 
191 static void vrele_thread(void *);
192 static void insmntque(vnode_t *, struct mount *);
193 static int getdevvp(dev_t, vnode_t **, enum vtype);
194 static vnode_t *getcleanvnode(void);
195 void vpanic(vnode_t *, const char *);
196 static void vfs_shutdown1(struct lwp *);
197 
198 #ifdef DEBUG
199 void printlockedvnodes(void);
200 #endif
201 
202 #ifdef DIAGNOSTIC
203 void
204 vpanic(vnode_t *vp, const char *msg)
205 {
206 
207 	vprint(NULL, vp);
208 	panic("%s\n", msg);
209 }
210 #else
211 #define	vpanic(vp, msg)	/* nothing */
212 #endif
213 
214 void
215 vn_init1(void)
216 {
217 
218 	vnode_cache = pool_cache_init(sizeof(struct vnode), 0, 0, 0, "vnodepl",
219 	    NULL, IPL_NONE, NULL, NULL, NULL);
220 	KASSERT(vnode_cache != NULL);
221 
222 	/* Create deferred release thread. */
223 	mutex_init(&vrele_lock, MUTEX_DEFAULT, IPL_NONE);
224 	cv_init(&vrele_cv, "vrele");
225 	if (kthread_create(PRI_VM, KTHREAD_MPSAFE, NULL, vrele_thread,
226 	    NULL, &vrele_lwp, "vrele"))
227 		panic("fork vrele");
228 }
229 
230 /*
231  * Initialize the vnode management data structures.
232  */
233 void
234 vntblinit(void)
235 {
236 
237 	mutex_init(&mountgen_lock, MUTEX_DEFAULT, IPL_NONE);
238 	mutex_init(&mountlist_lock, MUTEX_DEFAULT, IPL_NONE);
239 	mutex_init(&mntid_lock, MUTEX_DEFAULT, IPL_NONE);
240 	mutex_init(&mntvnode_lock, MUTEX_DEFAULT, IPL_NONE);
241 	mutex_init(&vnode_free_list_lock, MUTEX_DEFAULT, IPL_NONE);
242 	mutex_init(&vfs_list_lock, MUTEX_DEFAULT, IPL_NONE);
243 
244 	mount_specificdata_domain = specificdata_domain_create();
245 
246 	/* Initialize the filesystem syncer. */
247 	vn_initialize_syncerd();
248 	vn_init1();
249 }
250 
251 int
252 vfs_drainvnodes(long target, struct lwp *l)
253 {
254 
255 	while (numvnodes > target) {
256 		vnode_t *vp;
257 
258 		mutex_enter(&vnode_free_list_lock);
259 		vp = getcleanvnode();
260 		if (vp == NULL)
261 			return EBUSY; /* give up */
262 		ungetnewvnode(vp);
263 	}
264 
265 	return 0;
266 }
267 
268 /*
269  * Lookup a mount point by filesystem identifier.
270  *
271  * XXX Needs to add a reference to the mount point.
272  */
273 struct mount *
274 vfs_getvfs(fsid_t *fsid)
275 {
276 	struct mount *mp;
277 
278 	mutex_enter(&mountlist_lock);
279 	CIRCLEQ_FOREACH(mp, &mountlist, mnt_list) {
280 		if (mp->mnt_stat.f_fsidx.__fsid_val[0] == fsid->__fsid_val[0] &&
281 		    mp->mnt_stat.f_fsidx.__fsid_val[1] == fsid->__fsid_val[1]) {
282 			mutex_exit(&mountlist_lock);
283 			return (mp);
284 		}
285 	}
286 	mutex_exit(&mountlist_lock);
287 	return ((struct mount *)0);
288 }
289 
290 /*
291  * Drop a reference to a mount structure, freeing if the last reference.
292  */
293 void
294 vfs_destroy(struct mount *mp)
295 {
296 
297 	if (__predict_true((int)atomic_dec_uint_nv(&mp->mnt_refcnt) > 0)) {
298 		return;
299 	}
300 
301 	/*
302 	 * Nothing else has visibility of the mount: we can now
303 	 * free the data structures.
304 	 */
305 	KASSERT(mp->mnt_refcnt == 0);
306 	specificdata_fini(mount_specificdata_domain, &mp->mnt_specdataref);
307 	rw_destroy(&mp->mnt_unmounting);
308 	mutex_destroy(&mp->mnt_updating);
309 	mutex_destroy(&mp->mnt_renamelock);
310 	if (mp->mnt_op != NULL) {
311 		vfs_delref(mp->mnt_op);
312 	}
313 	kmem_free(mp, sizeof(*mp));
314 }
315 
316 /*
317  * grab a vnode from freelist and clean it.
318  */
319 vnode_t *
320 getcleanvnode(void)
321 {
322 	vnode_t *vp;
323 	vnodelst_t *listhd;
324 
325 	KASSERT(mutex_owned(&vnode_free_list_lock));
326 
327 retry:
328 	listhd = &vnode_free_list;
329 try_nextlist:
330 	TAILQ_FOREACH(vp, listhd, v_freelist) {
331 		/*
332 		 * It's safe to test v_usecount and v_iflag
333 		 * without holding the interlock here, since
334 		 * these vnodes should never appear on the
335 		 * lists.
336 		 */
337 		if (vp->v_usecount != 0) {
338 			vpanic(vp, "free vnode isn't");
339 		}
340 		if ((vp->v_iflag & VI_CLEAN) != 0) {
341 			vpanic(vp, "clean vnode on freelist");
342 		}
343 		if (vp->v_freelisthd != listhd) {
344 			printf("vnode sez %p, listhd %p\n", vp->v_freelisthd, listhd);
345 			vpanic(vp, "list head mismatch");
346 		}
347 		if (!mutex_tryenter(&vp->v_interlock))
348 			continue;
349 		/*
350 		 * Our lwp might hold the underlying vnode
351 		 * locked, so don't try to reclaim a VI_LAYER
352 		 * node if it's locked.
353 		 */
354 		if ((vp->v_iflag & VI_XLOCK) == 0 &&
355 		    ((vp->v_iflag & VI_LAYER) == 0 || VOP_ISLOCKED(vp) == 0)) {
356 			break;
357 		}
358 		mutex_exit(&vp->v_interlock);
359 	}
360 
361 	if (vp == NULL) {
362 		if (listhd == &vnode_free_list) {
363 			listhd = &vnode_hold_list;
364 			goto try_nextlist;
365 		}
366 		mutex_exit(&vnode_free_list_lock);
367 		return NULL;
368 	}
369 
370 	/* Remove it from the freelist. */
371 	TAILQ_REMOVE(listhd, vp, v_freelist);
372 	vp->v_freelisthd = NULL;
373 	mutex_exit(&vnode_free_list_lock);
374 
375 	if (vp->v_usecount != 0) {
376 		/*
377 		 * was referenced again before we got the interlock
378 		 * Don't return to freelist - the holder of the last
379 		 * reference will destroy it.
380 		 */
381 		mutex_exit(&vp->v_interlock);
382 		mutex_enter(&vnode_free_list_lock);
383 		goto retry;
384 	}
385 
386 	/*
387 	 * The vnode is still associated with a file system, so we must
388 	 * clean it out before reusing it.  We need to add a reference
389 	 * before doing this.  If the vnode gains another reference while
390 	 * being cleaned out then we lose - retry.
391 	 */
392 	atomic_add_int(&vp->v_usecount, 1 + VC_XLOCK);
393 	vclean(vp, DOCLOSE);
394 	KASSERT(vp->v_usecount >= 1 + VC_XLOCK);
395 	atomic_add_int(&vp->v_usecount, -VC_XLOCK);
396 	if (vp->v_usecount == 1) {
397 		/* We're about to dirty it. */
398 		vp->v_iflag &= ~VI_CLEAN;
399 		mutex_exit(&vp->v_interlock);
400 		if (vp->v_type == VBLK || vp->v_type == VCHR) {
401 			spec_node_destroy(vp);
402 		}
403 		vp->v_type = VNON;
404 	} else {
405 		/*
406 		 * Don't return to freelist - the holder of the last
407 		 * reference will destroy it.
408 		 */
409 		vrelel(vp, 0); /* releases vp->v_interlock */
410 		mutex_enter(&vnode_free_list_lock);
411 		goto retry;
412 	}
413 
414 	if (vp->v_data != NULL || vp->v_uobj.uo_npages != 0 ||
415 	    !TAILQ_EMPTY(&vp->v_uobj.memq)) {
416 		vpanic(vp, "cleaned vnode isn't");
417 	}
418 	if (vp->v_numoutput != 0) {
419 		vpanic(vp, "clean vnode has pending I/O's");
420 	}
421 	if ((vp->v_iflag & VI_ONWORKLST) != 0) {
422 		vpanic(vp, "clean vnode on syncer list");
423 	}
424 
425 	return vp;
426 }
427 
428 /*
429  * Mark a mount point as busy, and gain a new reference to it.  Used to
430  * prevent the file system from being unmounted during critical sections.
431  *
432  * => The caller must hold a pre-existing reference to the mount.
433  * => Will fail if the file system is being unmounted, or is unmounted.
434  */
435 int
436 vfs_busy(struct mount *mp, struct mount **nextp)
437 {
438 
439 	KASSERT(mp->mnt_refcnt > 0);
440 
441 	if (__predict_false(!rw_tryenter(&mp->mnt_unmounting, RW_READER))) {
442 		if (nextp != NULL) {
443 			KASSERT(mutex_owned(&mountlist_lock));
444 			*nextp = CIRCLEQ_NEXT(mp, mnt_list);
445 		}
446 		return EBUSY;
447 	}
448 	if (__predict_false((mp->mnt_iflag & IMNT_GONE) != 0)) {
449 		rw_exit(&mp->mnt_unmounting);
450 		if (nextp != NULL) {
451 			KASSERT(mutex_owned(&mountlist_lock));
452 			*nextp = CIRCLEQ_NEXT(mp, mnt_list);
453 		}
454 		return ENOENT;
455 	}
456 	if (nextp != NULL) {
457 		mutex_exit(&mountlist_lock);
458 	}
459 	atomic_inc_uint(&mp->mnt_refcnt);
460 	return 0;
461 }
462 
463 /*
464  * Unbusy a busy filesystem.
465  *
466  * => If keepref is true, preserve reference added by vfs_busy().
467  * => If nextp != NULL, acquire mountlist_lock.
468  */
469 void
470 vfs_unbusy(struct mount *mp, bool keepref, struct mount **nextp)
471 {
472 
473 	KASSERT(mp->mnt_refcnt > 0);
474 
475 	if (nextp != NULL) {
476 		mutex_enter(&mountlist_lock);
477 	}
478 	rw_exit(&mp->mnt_unmounting);
479 	if (!keepref) {
480 		vfs_destroy(mp);
481 	}
482 	if (nextp != NULL) {
483 		KASSERT(mutex_owned(&mountlist_lock));
484 		*nextp = CIRCLEQ_NEXT(mp, mnt_list);
485 	}
486 }
487 
488 struct mount *
489 vfs_mountalloc(struct vfsops *vfsops, struct vnode *vp)
490 {
491 	int error;
492 	struct mount *mp;
493 
494 	mp = kmem_zalloc(sizeof(*mp), KM_SLEEP);
495 	if (mp == NULL)
496 		return NULL;
497 
498 	mp->mnt_op = vfsops;
499 	mp->mnt_refcnt = 1;
500 	TAILQ_INIT(&mp->mnt_vnodelist);
501 	rw_init(&mp->mnt_unmounting);
502 	mutex_init(&mp->mnt_renamelock, MUTEX_DEFAULT, IPL_NONE);
503 	mutex_init(&mp->mnt_updating, MUTEX_DEFAULT, IPL_NONE);
504 	error = vfs_busy(mp, NULL);
505 	KASSERT(error == 0);
506 	mp->mnt_vnodecovered = vp;
507 	mount_initspecific(mp);
508 
509 	mutex_enter(&mountgen_lock);
510 	mp->mnt_gen = mountgen++;
511 	mutex_exit(&mountgen_lock);
512 
513 	return mp;
514 }
515 
516 /*
517  * Lookup a filesystem type, and if found allocate and initialize
518  * a mount structure for it.
519  *
520  * Devname is usually updated by mount(8) after booting.
521  */
522 int
523 vfs_rootmountalloc(const char *fstypename, const char *devname,
524     struct mount **mpp)
525 {
526 	struct vfsops *vfsp = NULL;
527 	struct mount *mp;
528 
529 	mutex_enter(&vfs_list_lock);
530 	LIST_FOREACH(vfsp, &vfs_list, vfs_list)
531 		if (!strncmp(vfsp->vfs_name, fstypename,
532 		    sizeof(mp->mnt_stat.f_fstypename)))
533 			break;
534 	if (vfsp == NULL) {
535 		mutex_exit(&vfs_list_lock);
536 		return (ENODEV);
537 	}
538 	vfsp->vfs_refcount++;
539 	mutex_exit(&vfs_list_lock);
540 
541 	if ((mp = vfs_mountalloc(vfsp, NULL)) == NULL)
542 		return ENOMEM;
543 	mp->mnt_flag = MNT_RDONLY;
544 	(void)strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfs_name,
545 	    sizeof(mp->mnt_stat.f_fstypename));
546 	mp->mnt_stat.f_mntonname[0] = '/';
547 	mp->mnt_stat.f_mntonname[1] = '\0';
548 	mp->mnt_stat.f_mntfromname[sizeof(mp->mnt_stat.f_mntfromname) - 1] =
549 	    '\0';
550 	(void)copystr(devname, mp->mnt_stat.f_mntfromname,
551 	    sizeof(mp->mnt_stat.f_mntfromname) - 1, 0);
552 	*mpp = mp;
553 	return (0);
554 }
555 
556 /*
557  * Routines having to do with the management of the vnode table.
558  */
559 extern int (**dead_vnodeop_p)(void *);
560 
561 /*
562  * Return the next vnode from the free list.
563  */
564 int
565 getnewvnode(enum vtagtype tag, struct mount *mp, int (**vops)(void *),
566 	    vnode_t **vpp)
567 {
568 	struct uvm_object *uobj;
569 	static int toggle;
570 	vnode_t *vp;
571 	int error = 0, tryalloc;
572 
573  try_again:
574 	if (mp != NULL) {
575 		/*
576 		 * Mark filesystem busy while we're creating a
577 		 * vnode.  If unmount is in progress, this will
578 		 * fail.
579 		 */
580 		error = vfs_busy(mp, NULL);
581 		if (error)
582 			return error;
583 	}
584 
585 	/*
586 	 * We must choose whether to allocate a new vnode or recycle an
587 	 * existing one. The criterion for allocating a new one is that
588 	 * the total number of vnodes is less than the number desired or
589 	 * there are no vnodes on either free list. Generally we only
590 	 * want to recycle vnodes that have no buffers associated with
591 	 * them, so we look first on the vnode_free_list. If it is empty,
592 	 * we next consider vnodes with referencing buffers on the
593 	 * vnode_hold_list. The toggle ensures that half the time we
594 	 * will use a buffer from the vnode_hold_list, and half the time
595 	 * we will allocate a new one unless the list has grown to twice
596 	 * the desired size. We are reticent to recycle vnodes from the
597 	 * vnode_hold_list because we will lose the identity of all its
598 	 * referencing buffers.
599 	 */
600 
601 	vp = NULL;
602 
603 	mutex_enter(&vnode_free_list_lock);
604 
605 	toggle ^= 1;
606 	if (numvnodes > 2 * desiredvnodes)
607 		toggle = 0;
608 
609 	tryalloc = numvnodes < desiredvnodes ||
610 	    (TAILQ_FIRST(&vnode_free_list) == NULL &&
611 	     (TAILQ_FIRST(&vnode_hold_list) == NULL || toggle));
612 
613 	if (tryalloc) {
614 		numvnodes++;
615 		mutex_exit(&vnode_free_list_lock);
616 		if ((vp = vnalloc(NULL)) == NULL) {
617 			mutex_enter(&vnode_free_list_lock);
618 			numvnodes--;
619 		} else
620 			vp->v_usecount = 1;
621 	}
622 
623 	if (vp == NULL) {
624 		vp = getcleanvnode();
625 		if (vp == NULL) {
626 			if (mp != NULL) {
627 				vfs_unbusy(mp, false, NULL);
628 			}
629 			if (tryalloc) {
630 				printf("WARNING: unable to allocate new "
631 				    "vnode, retrying...\n");
632 				kpause("newvn", false, hz, NULL);
633 				goto try_again;
634 			}
635 			tablefull("vnode", "increase kern.maxvnodes or NVNODE");
636 			*vpp = 0;
637 			return (ENFILE);
638 		}
639 		vp->v_iflag = 0;
640 		vp->v_vflag = 0;
641 		vp->v_uflag = 0;
642 		vp->v_socket = NULL;
643 	}
644 
645 	KASSERT(vp->v_usecount == 1);
646 	KASSERT(vp->v_freelisthd == NULL);
647 	KASSERT(LIST_EMPTY(&vp->v_nclist));
648 	KASSERT(LIST_EMPTY(&vp->v_dnclist));
649 
650 	vp->v_type = VNON;
651 	vp->v_vnlock = &vp->v_lock;
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;
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 	if (flags & LK_TYPE_MASK) {
1338 		error = vn_lock(vp, flags | LK_INTERLOCK);
1339 		if (error != 0) {
1340 			vrele(vp);
1341 		}
1342 		return error;
1343 	}
1344 	mutex_exit(&vp->v_interlock);
1345 	return 0;
1346 }
1347 
1348 /*
1349  * vput(), just unlock and vrele()
1350  */
1351 void
1352 vput(vnode_t *vp)
1353 {
1354 
1355 	KASSERT((vp->v_iflag & VI_MARKER) == 0);
1356 
1357 	VOP_UNLOCK(vp, 0);
1358 	vrele(vp);
1359 }
1360 
1361 /*
1362  * Try to drop reference on a vnode.  Abort if we are releasing the
1363  * last reference.  Note: this _must_ succeed if not the last reference.
1364  */
1365 static inline bool
1366 vtryrele(vnode_t *vp)
1367 {
1368 	u_int use, next;
1369 
1370 	for (use = vp->v_usecount;; use = next) {
1371 		if (use == 1) {
1372 			return false;
1373 		}
1374 		KASSERT((use & VC_MASK) > 1);
1375 		next = atomic_cas_uint(&vp->v_usecount, use, use - 1);
1376 		if (__predict_true(next == use)) {
1377 			return true;
1378 		}
1379 	}
1380 }
1381 
1382 /*
1383  * Vnode release.  If reference count drops to zero, call inactive
1384  * routine and either return to freelist or free to the pool.
1385  */
1386 void
1387 vrelel(vnode_t *vp, int flags)
1388 {
1389 	bool recycle, defer;
1390 	int error;
1391 
1392 	KASSERT(mutex_owned(&vp->v_interlock));
1393 	KASSERT((vp->v_iflag & VI_MARKER) == 0);
1394 	KASSERT(vp->v_freelisthd == NULL);
1395 
1396 	if (__predict_false(vp->v_op == dead_vnodeop_p &&
1397 	    (vp->v_iflag & (VI_CLEAN|VI_XLOCK)) == 0)) {
1398 		vpanic(vp, "dead but not clean");
1399 	}
1400 
1401 	/*
1402 	 * If not the last reference, just drop the reference count
1403 	 * and unlock.
1404 	 */
1405 	if (vtryrele(vp)) {
1406 		vp->v_iflag |= VI_INACTREDO;
1407 		mutex_exit(&vp->v_interlock);
1408 		return;
1409 	}
1410 	if (vp->v_usecount <= 0 || vp->v_writecount != 0) {
1411 		vpanic(vp, "vrelel: bad ref count");
1412 	}
1413 
1414 	KASSERT((vp->v_iflag & VI_XLOCK) == 0);
1415 
1416 	/*
1417 	 * If not clean, deactivate the vnode, but preserve
1418 	 * our reference across the call to VOP_INACTIVE().
1419 	 */
1420  retry:
1421 	if ((vp->v_iflag & VI_CLEAN) == 0) {
1422 		recycle = false;
1423 		vp->v_iflag |= VI_INACTNOW;
1424 
1425 		/*
1426 		 * XXX This ugly block can be largely eliminated if
1427 		 * locking is pushed down into the file systems.
1428 		 *
1429 		 * Defer vnode release to vrele_thread if caller
1430 		 * requests it explicitly.
1431 		 */
1432 		if ((curlwp == uvm.pagedaemon_lwp) ||
1433 		    (flags & VRELEL_ASYNC_RELE) != 0) {
1434 			/* The pagedaemon can't wait around; defer. */
1435 			defer = true;
1436 		} else if (curlwp == vrele_lwp) {
1437 			/*
1438 			 * We have to try harder. But we can't sleep
1439 			 * with VI_INACTNOW as vget() may be waiting on it.
1440 			 */
1441 			vp->v_iflag &= ~(VI_INACTREDO|VI_INACTNOW);
1442 			cv_broadcast(&vp->v_cv);
1443 			error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK |
1444 			    LK_RETRY);
1445 			if (error != 0) {
1446 				/* XXX */
1447 				vpanic(vp, "vrele: unable to lock %p");
1448 			}
1449 			mutex_enter(&vp->v_interlock);
1450 			/*
1451 			 * if we did get another reference while
1452 			 * sleeping, don't try to inactivate it yet.
1453 			 */
1454 			if (__predict_false(vtryrele(vp))) {
1455 				VOP_UNLOCK(vp, 0);
1456 				mutex_exit(&vp->v_interlock);
1457 				return;
1458 			}
1459 			vp->v_iflag |= VI_INACTNOW;
1460 			mutex_exit(&vp->v_interlock);
1461 			defer = false;
1462 		} else if ((vp->v_iflag & VI_LAYER) != 0) {
1463 			/*
1464 			 * Acquiring the stack's lock in vclean() even
1465 			 * for an honest vput/vrele is dangerous because
1466 			 * our caller may hold other vnode locks; defer.
1467 			 */
1468 			defer = true;
1469 		} else {
1470 			/* If we can't acquire the lock, then defer. */
1471 			vp->v_iflag &= ~VI_INACTREDO;
1472 			error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK |
1473 			    LK_NOWAIT);
1474 			if (error != 0) {
1475 				defer = true;
1476 				mutex_enter(&vp->v_interlock);
1477 			} else {
1478 				defer = false;
1479 			}
1480 		}
1481 
1482 		if (defer) {
1483 			/*
1484 			 * Defer reclaim to the kthread; it's not safe to
1485 			 * clean it here.  We donate it our last reference.
1486 			 */
1487 			KASSERT(mutex_owned(&vp->v_interlock));
1488 			KASSERT((vp->v_iflag & VI_INACTPEND) == 0);
1489 			vp->v_iflag &= ~VI_INACTNOW;
1490 			vp->v_iflag |= VI_INACTPEND;
1491 			mutex_enter(&vrele_lock);
1492 			TAILQ_INSERT_TAIL(&vrele_list, vp, v_freelist);
1493 			if (++vrele_pending > (desiredvnodes >> 8))
1494 				cv_signal(&vrele_cv);
1495 			mutex_exit(&vrele_lock);
1496 			cv_broadcast(&vp->v_cv);
1497 			mutex_exit(&vp->v_interlock);
1498 			return;
1499 		}
1500 
1501 #ifdef DIAGNOSTIC
1502 		if ((vp->v_type == VBLK || vp->v_type == VCHR) &&
1503 		    vp->v_specnode != NULL && vp->v_specnode->sn_opencnt != 0) {
1504 			vprint("vrelel: missing VOP_CLOSE()", vp);
1505 		}
1506 #endif
1507 
1508 		/*
1509 		 * The vnode can gain another reference while being
1510 		 * deactivated.  If VOP_INACTIVE() indicates that
1511 		 * the described file has been deleted, then recycle
1512 		 * the vnode irrespective of additional references.
1513 		 * Another thread may be waiting to re-use the on-disk
1514 		 * inode.
1515 		 *
1516 		 * Note that VOP_INACTIVE() will drop the vnode lock.
1517 		 */
1518 		VOP_INACTIVE(vp, &recycle);
1519 		mutex_enter(&vp->v_interlock);
1520 		vp->v_iflag &= ~VI_INACTNOW;
1521 		cv_broadcast(&vp->v_cv);
1522 		if (!recycle) {
1523 			if (vtryrele(vp)) {
1524 				mutex_exit(&vp->v_interlock);
1525 				return;
1526 			}
1527 
1528 			/*
1529 			 * If we grew another reference while
1530 			 * VOP_INACTIVE() was underway, retry.
1531 			 */
1532 			if ((vp->v_iflag & VI_INACTREDO) != 0) {
1533 				goto retry;
1534 			}
1535 		}
1536 
1537 		/* Take care of space accounting. */
1538 		if (vp->v_iflag & VI_EXECMAP) {
1539 			atomic_add_int(&uvmexp.execpages,
1540 			    -vp->v_uobj.uo_npages);
1541 			atomic_add_int(&uvmexp.filepages,
1542 			    vp->v_uobj.uo_npages);
1543 		}
1544 		vp->v_iflag &= ~(VI_TEXT|VI_EXECMAP|VI_WRMAP);
1545 		vp->v_vflag &= ~VV_MAPPED;
1546 
1547 		/*
1548 		 * Recycle the vnode if the file is now unused (unlinked),
1549 		 * otherwise just free it.
1550 		 */
1551 		if (recycle) {
1552 			vclean(vp, DOCLOSE);
1553 		}
1554 		KASSERT(vp->v_usecount > 0);
1555 	}
1556 
1557 	if (atomic_dec_uint_nv(&vp->v_usecount) != 0) {
1558 		/* Gained another reference while being reclaimed. */
1559 		mutex_exit(&vp->v_interlock);
1560 		return;
1561 	}
1562 
1563 	if ((vp->v_iflag & VI_CLEAN) != 0) {
1564 		/*
1565 		 * It's clean so destroy it.  It isn't referenced
1566 		 * anywhere since it has been reclaimed.
1567 		 */
1568 		KASSERT(vp->v_holdcnt == 0);
1569 		KASSERT(vp->v_writecount == 0);
1570 		mutex_exit(&vp->v_interlock);
1571 		insmntque(vp, NULL);
1572 		if (vp->v_type == VBLK || vp->v_type == VCHR) {
1573 			spec_node_destroy(vp);
1574 		}
1575 		vnfree(vp);
1576 	} else {
1577 		/*
1578 		 * Otherwise, put it back onto the freelist.  It
1579 		 * can't be destroyed while still associated with
1580 		 * a file system.
1581 		 */
1582 		mutex_enter(&vnode_free_list_lock);
1583 		if (vp->v_holdcnt > 0) {
1584 			vp->v_freelisthd = &vnode_hold_list;
1585 		} else {
1586 			vp->v_freelisthd = &vnode_free_list;
1587 		}
1588 		TAILQ_INSERT_TAIL(vp->v_freelisthd, vp, v_freelist);
1589 		mutex_exit(&vnode_free_list_lock);
1590 		mutex_exit(&vp->v_interlock);
1591 	}
1592 }
1593 
1594 void
1595 vrele(vnode_t *vp)
1596 {
1597 
1598 	KASSERT((vp->v_iflag & VI_MARKER) == 0);
1599 
1600 	if ((vp->v_iflag & VI_INACTNOW) == 0 && vtryrele(vp)) {
1601 		return;
1602 	}
1603 	mutex_enter(&vp->v_interlock);
1604 	vrelel(vp, 0);
1605 }
1606 
1607 /*
1608  * Asynchronous vnode release, vnode is released in different context.
1609  */
1610 void
1611 vrele_async(vnode_t *vp)
1612 {
1613 
1614 	KASSERT((vp->v_iflag & VI_MARKER) == 0);
1615 
1616 	if ((vp->v_iflag & VI_INACTNOW) == 0 && vtryrele(vp)) {
1617 		return;
1618 	}
1619 
1620 	mutex_enter(&vp->v_interlock);
1621 	vrelel(vp, VRELEL_ASYNC_RELE);
1622 }
1623 
1624 static void
1625 vrele_thread(void *cookie)
1626 {
1627 	vnode_t *vp;
1628 
1629 	for (;;) {
1630 		mutex_enter(&vrele_lock);
1631 		while (TAILQ_EMPTY(&vrele_list)) {
1632 			vrele_gen++;
1633 			cv_broadcast(&vrele_cv);
1634 			cv_timedwait(&vrele_cv, &vrele_lock, hz);
1635 		}
1636 		vp = TAILQ_FIRST(&vrele_list);
1637 		TAILQ_REMOVE(&vrele_list, vp, v_freelist);
1638 		vrele_pending--;
1639 		mutex_exit(&vrele_lock);
1640 
1641 		/*
1642 		 * If not the last reference, then ignore the vnode
1643 		 * and look for more work.
1644 		 */
1645 		mutex_enter(&vp->v_interlock);
1646 		KASSERT((vp->v_iflag & VI_INACTPEND) != 0);
1647 		vp->v_iflag &= ~VI_INACTPEND;
1648 		vrelel(vp, 0);
1649 	}
1650 }
1651 
1652 /*
1653  * Page or buffer structure gets a reference.
1654  * Called with v_interlock held.
1655  */
1656 void
1657 vholdl(vnode_t *vp)
1658 {
1659 
1660 	KASSERT(mutex_owned(&vp->v_interlock));
1661 	KASSERT((vp->v_iflag & VI_MARKER) == 0);
1662 
1663 	if (vp->v_holdcnt++ == 0 && vp->v_usecount == 0) {
1664 		mutex_enter(&vnode_free_list_lock);
1665 		KASSERT(vp->v_freelisthd == &vnode_free_list);
1666 		TAILQ_REMOVE(vp->v_freelisthd, vp, v_freelist);
1667 		vp->v_freelisthd = &vnode_hold_list;
1668 		TAILQ_INSERT_TAIL(vp->v_freelisthd, vp, v_freelist);
1669 		mutex_exit(&vnode_free_list_lock);
1670 	}
1671 }
1672 
1673 /*
1674  * Page or buffer structure frees a reference.
1675  * Called with v_interlock held.
1676  */
1677 void
1678 holdrelel(vnode_t *vp)
1679 {
1680 
1681 	KASSERT(mutex_owned(&vp->v_interlock));
1682 	KASSERT((vp->v_iflag & VI_MARKER) == 0);
1683 
1684 	if (vp->v_holdcnt <= 0) {
1685 		vpanic(vp, "holdrelel: holdcnt vp %p");
1686 	}
1687 
1688 	vp->v_holdcnt--;
1689 	if (vp->v_holdcnt == 0 && vp->v_usecount == 0) {
1690 		mutex_enter(&vnode_free_list_lock);
1691 		KASSERT(vp->v_freelisthd == &vnode_hold_list);
1692 		TAILQ_REMOVE(vp->v_freelisthd, vp, v_freelist);
1693 		vp->v_freelisthd = &vnode_free_list;
1694 		TAILQ_INSERT_TAIL(vp->v_freelisthd, vp, v_freelist);
1695 		mutex_exit(&vnode_free_list_lock);
1696 	}
1697 }
1698 
1699 /*
1700  * Vnode reference, where a reference is already held by some other
1701  * object (for example, a file structure).
1702  */
1703 void
1704 vref(vnode_t *vp)
1705 {
1706 
1707 	KASSERT((vp->v_iflag & VI_MARKER) == 0);
1708 	KASSERT(vp->v_usecount != 0);
1709 
1710 	atomic_inc_uint(&vp->v_usecount);
1711 }
1712 
1713 /*
1714  * Remove any vnodes in the vnode table belonging to mount point mp.
1715  *
1716  * If FORCECLOSE is not specified, there should not be any active ones,
1717  * return error if any are found (nb: this is a user error, not a
1718  * system error). If FORCECLOSE is specified, detach any active vnodes
1719  * that are found.
1720  *
1721  * If WRITECLOSE is set, only flush out regular file vnodes open for
1722  * writing.
1723  *
1724  * SKIPSYSTEM causes any vnodes marked V_SYSTEM to be skipped.
1725  */
1726 #ifdef DEBUG
1727 int busyprt = 0;	/* print out busy vnodes */
1728 struct ctldebug debug1 = { "busyprt", &busyprt };
1729 #endif
1730 
1731 static vnode_t *
1732 vflushnext(vnode_t *mvp, int *when)
1733 {
1734 
1735 	if (hardclock_ticks > *when) {
1736 		mutex_exit(&mntvnode_lock);
1737 		yield();
1738 		mutex_enter(&mntvnode_lock);
1739 		*when = hardclock_ticks + hz / 10;
1740 	}
1741 
1742 	return vunmark(mvp);
1743 }
1744 
1745 int
1746 vflush(struct mount *mp, vnode_t *skipvp, int flags)
1747 {
1748 	vnode_t *vp, *mvp;
1749 	int busy = 0, when = 0, gen;
1750 
1751 	/*
1752 	 * First, flush out any vnode references from vrele_list.
1753 	 */
1754 	mutex_enter(&vrele_lock);
1755 	gen = vrele_gen;
1756 	while (vrele_pending && gen == vrele_gen) {
1757 		cv_broadcast(&vrele_cv);
1758 		cv_wait(&vrele_cv, &vrele_lock);
1759 	}
1760 	mutex_exit(&vrele_lock);
1761 
1762 	/* Allocate a marker vnode. */
1763 	if ((mvp = vnalloc(mp)) == NULL)
1764 		return (ENOMEM);
1765 
1766 	/*
1767 	 * NOTE: not using the TAILQ_FOREACH here since in this loop vgone()
1768 	 * and vclean() are called
1769 	 */
1770 	mutex_enter(&mntvnode_lock);
1771 	for (vp = TAILQ_FIRST(&mp->mnt_vnodelist); vp != NULL;
1772 	    vp = vflushnext(mvp, &when)) {
1773 		vmark(mvp, vp);
1774 		if (vp->v_mount != mp || vismarker(vp))
1775 			continue;
1776 		/*
1777 		 * Skip over a selected vnode.
1778 		 */
1779 		if (vp == skipvp)
1780 			continue;
1781 		mutex_enter(&vp->v_interlock);
1782 		/*
1783 		 * Ignore clean but still referenced vnodes.
1784 		 */
1785 		if ((vp->v_iflag & VI_CLEAN) != 0) {
1786 			mutex_exit(&vp->v_interlock);
1787 			continue;
1788 		}
1789 		/*
1790 		 * Skip over a vnodes marked VSYSTEM.
1791 		 */
1792 		if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
1793 			mutex_exit(&vp->v_interlock);
1794 			continue;
1795 		}
1796 		/*
1797 		 * If WRITECLOSE is set, only flush out regular file
1798 		 * vnodes open for writing.
1799 		 */
1800 		if ((flags & WRITECLOSE) &&
1801 		    (vp->v_writecount == 0 || vp->v_type != VREG)) {
1802 			mutex_exit(&vp->v_interlock);
1803 			continue;
1804 		}
1805 		/*
1806 		 * With v_usecount == 0, all we need to do is clear
1807 		 * out the vnode data structures and we are done.
1808 		 */
1809 		if (vp->v_usecount == 0) {
1810 			mutex_exit(&mntvnode_lock);
1811 			vremfree(vp);
1812 			vp->v_usecount = 1;
1813 			vclean(vp, DOCLOSE);
1814 			vrelel(vp, 0);
1815 			mutex_enter(&mntvnode_lock);
1816 			continue;
1817 		}
1818 		/*
1819 		 * If FORCECLOSE is set, forcibly close the vnode.
1820 		 * For block or character devices, revert to an
1821 		 * anonymous device.  For all other files, just
1822 		 * kill them.
1823 		 */
1824 		if (flags & FORCECLOSE) {
1825 			mutex_exit(&mntvnode_lock);
1826 			atomic_inc_uint(&vp->v_usecount);
1827 			if (vp->v_type != VBLK && vp->v_type != VCHR) {
1828 				vclean(vp, DOCLOSE);
1829 				vrelel(vp, 0);
1830 			} else {
1831 				vclean(vp, 0);
1832 				vp->v_op = spec_vnodeop_p; /* XXXSMP */
1833 				mutex_exit(&vp->v_interlock);
1834 				/*
1835 				 * The vnode isn't clean, but still resides
1836 				 * on the mount list.  Remove it. XXX This
1837 				 * is a bit dodgy.
1838 				 */
1839 				insmntque(vp, NULL);
1840 				vrele(vp);
1841 			}
1842 			mutex_enter(&mntvnode_lock);
1843 			continue;
1844 		}
1845 #ifdef DEBUG
1846 		if (busyprt)
1847 			vprint("vflush: busy vnode", vp);
1848 #endif
1849 		mutex_exit(&vp->v_interlock);
1850 		busy++;
1851 	}
1852 	mutex_exit(&mntvnode_lock);
1853 	vnfree(mvp);
1854 	if (busy)
1855 		return (EBUSY);
1856 	return (0);
1857 }
1858 
1859 /*
1860  * Disassociate the underlying file system from a vnode.
1861  *
1862  * Must be called with the interlock held, and will return with it held.
1863  */
1864 void
1865 vclean(vnode_t *vp, int flags)
1866 {
1867 	lwp_t *l = curlwp;
1868 	bool recycle, active;
1869 	int error;
1870 
1871 	KASSERT(mutex_owned(&vp->v_interlock));
1872 	KASSERT((vp->v_iflag & VI_MARKER) == 0);
1873 	KASSERT(vp->v_usecount != 0);
1874 
1875 	/* If cleaning is already in progress wait until done and return. */
1876 	if (vp->v_iflag & VI_XLOCK) {
1877 		vwait(vp, VI_XLOCK);
1878 		return;
1879 	}
1880 
1881 	/* If already clean, nothing to do. */
1882 	if ((vp->v_iflag & VI_CLEAN) != 0) {
1883 		return;
1884 	}
1885 
1886 	/*
1887 	 * Prevent the vnode from being recycled or brought into use
1888 	 * while we clean it out.
1889 	 */
1890 	vp->v_iflag |= VI_XLOCK;
1891 	if (vp->v_iflag & VI_EXECMAP) {
1892 		atomic_add_int(&uvmexp.execpages, -vp->v_uobj.uo_npages);
1893 		atomic_add_int(&uvmexp.filepages, vp->v_uobj.uo_npages);
1894 	}
1895 	vp->v_iflag &= ~(VI_TEXT|VI_EXECMAP);
1896 	active = (vp->v_usecount > 1);
1897 
1898 	/* XXXAD should not lock vnode under layer */
1899 	VOP_LOCK(vp, LK_EXCLUSIVE | LK_INTERLOCK);
1900 
1901 	/*
1902 	 * Clean out any cached data associated with the vnode.
1903 	 * If purging an active vnode, it must be closed and
1904 	 * deactivated before being reclaimed. Note that the
1905 	 * VOP_INACTIVE will unlock the vnode.
1906 	 */
1907 	if (flags & DOCLOSE) {
1908 		error = vinvalbuf(vp, V_SAVE, NOCRED, l, 0, 0);
1909 		if (error != 0) {
1910 			/* XXX, fix vn_start_write's grab of mp and use that. */
1911 
1912 			if (wapbl_vphaswapbl(vp))
1913 				WAPBL_DISCARD(wapbl_vptomp(vp));
1914 			error = vinvalbuf(vp, 0, NOCRED, l, 0, 0);
1915 		}
1916 		KASSERT(error == 0);
1917 		KASSERT((vp->v_iflag & VI_ONWORKLST) == 0);
1918 		if (active && (vp->v_type == VBLK || vp->v_type == VCHR)) {
1919 			 spec_node_revoke(vp);
1920 		}
1921 	}
1922 	if (active) {
1923 		VOP_INACTIVE(vp, &recycle);
1924 	} else {
1925 		/*
1926 		 * Any other processes trying to obtain this lock must first
1927 		 * wait for VI_XLOCK to clear, then call the new lock operation.
1928 		 */
1929 		VOP_UNLOCK(vp, 0);
1930 	}
1931 
1932 	/* Disassociate the underlying file system from the vnode. */
1933 	if (VOP_RECLAIM(vp)) {
1934 		vpanic(vp, "vclean: cannot reclaim");
1935 	}
1936 
1937 	KASSERT(vp->v_uobj.uo_npages == 0);
1938 	if (vp->v_type == VREG && vp->v_ractx != NULL) {
1939 		uvm_ra_freectx(vp->v_ractx);
1940 		vp->v_ractx = NULL;
1941 	}
1942 	cache_purge(vp);
1943 
1944 	/* Done with purge, notify sleepers of the grim news. */
1945 	mutex_enter(&vp->v_interlock);
1946 	vp->v_op = dead_vnodeop_p;
1947 	vp->v_tag = VT_NON;
1948 	vp->v_vnlock = &vp->v_lock;
1949 	KNOTE(&vp->v_klist, NOTE_REVOKE);
1950 	vp->v_iflag &= ~(VI_XLOCK | VI_FREEING);
1951 	vp->v_vflag &= ~VV_LOCKSWORK;
1952 	if ((flags & DOCLOSE) != 0) {
1953 		vp->v_iflag |= VI_CLEAN;
1954 	}
1955 	cv_broadcast(&vp->v_cv);
1956 
1957 	KASSERT((vp->v_iflag & VI_ONWORKLST) == 0);
1958 }
1959 
1960 /*
1961  * Recycle an unused vnode to the front of the free list.
1962  * Release the passed interlock if the vnode will be recycled.
1963  */
1964 int
1965 vrecycle(vnode_t *vp, kmutex_t *inter_lkp, struct lwp *l)
1966 {
1967 
1968 	KASSERT((vp->v_iflag & VI_MARKER) == 0);
1969 
1970 	mutex_enter(&vp->v_interlock);
1971 	if (vp->v_usecount != 0) {
1972 		mutex_exit(&vp->v_interlock);
1973 		return (0);
1974 	}
1975 	if (inter_lkp)
1976 		mutex_exit(inter_lkp);
1977 	vremfree(vp);
1978 	vp->v_usecount = 1;
1979 	vclean(vp, DOCLOSE);
1980 	vrelel(vp, 0);
1981 	return (1);
1982 }
1983 
1984 /*
1985  * Eliminate all activity associated with a vnode in preparation for
1986  * reuse.  Drops a reference from the vnode.
1987  */
1988 void
1989 vgone(vnode_t *vp)
1990 {
1991 
1992 	mutex_enter(&vp->v_interlock);
1993 	vclean(vp, DOCLOSE);
1994 	vrelel(vp, 0);
1995 }
1996 
1997 /*
1998  * Lookup a vnode by device number.
1999  */
2000 int
2001 vfinddev(dev_t dev, enum vtype type, vnode_t **vpp)
2002 {
2003 	vnode_t *vp;
2004 	int rc = 0;
2005 
2006 	mutex_enter(&device_lock);
2007 	for (vp = specfs_hash[SPECHASH(dev)]; vp; vp = vp->v_specnext) {
2008 		if (dev != vp->v_rdev || type != vp->v_type)
2009 			continue;
2010 		*vpp = vp;
2011 		rc = 1;
2012 		break;
2013 	}
2014 	mutex_exit(&device_lock);
2015 	return (rc);
2016 }
2017 
2018 /*
2019  * Revoke all the vnodes corresponding to the specified minor number
2020  * range (endpoints inclusive) of the specified major.
2021  */
2022 void
2023 vdevgone(int maj, int minl, int minh, enum vtype type)
2024 {
2025 	vnode_t *vp, **vpp;
2026 	dev_t dev;
2027 	int mn;
2028 
2029 	vp = NULL;	/* XXX gcc */
2030 
2031 	mutex_enter(&device_lock);
2032 	for (mn = minl; mn <= minh; mn++) {
2033 		dev = makedev(maj, mn);
2034 		vpp = &specfs_hash[SPECHASH(dev)];
2035 		for (vp = *vpp; vp != NULL;) {
2036 			mutex_enter(&vp->v_interlock);
2037 			if ((vp->v_iflag & VI_CLEAN) != 0 ||
2038 			    dev != vp->v_rdev || type != vp->v_type) {
2039 				mutex_exit(&vp->v_interlock);
2040 				vp = vp->v_specnext;
2041 				continue;
2042 			}
2043 			mutex_exit(&device_lock);
2044 			if (vget(vp, LK_INTERLOCK) == 0) {
2045 				VOP_REVOKE(vp, REVOKEALL);
2046 				vrele(vp);
2047 			}
2048 			mutex_enter(&device_lock);
2049 			vp = *vpp;
2050 		}
2051 	}
2052 	mutex_exit(&device_lock);
2053 }
2054 
2055 /*
2056  * Eliminate all activity associated with the requested vnode
2057  * and with all vnodes aliased to the requested vnode.
2058  */
2059 void
2060 vrevoke(vnode_t *vp)
2061 {
2062 	vnode_t *vq, **vpp;
2063 	enum vtype type;
2064 	dev_t dev;
2065 
2066 	KASSERT(vp->v_usecount > 0);
2067 
2068 	mutex_enter(&vp->v_interlock);
2069 	if ((vp->v_iflag & VI_CLEAN) != 0) {
2070 		mutex_exit(&vp->v_interlock);
2071 		return;
2072 	} else if (vp->v_type != VBLK && vp->v_type != VCHR) {
2073 		atomic_inc_uint(&vp->v_usecount);
2074 		vclean(vp, DOCLOSE);
2075 		vrelel(vp, 0);
2076 		return;
2077 	} else {
2078 		dev = vp->v_rdev;
2079 		type = vp->v_type;
2080 		mutex_exit(&vp->v_interlock);
2081 	}
2082 
2083 	vpp = &specfs_hash[SPECHASH(dev)];
2084 	mutex_enter(&device_lock);
2085 	for (vq = *vpp; vq != NULL;) {
2086 		/* If clean or being cleaned, then ignore it. */
2087 		mutex_enter(&vq->v_interlock);
2088 		if ((vq->v_iflag & (VI_CLEAN | VI_XLOCK)) != 0 ||
2089 		    vq->v_rdev != dev || vq->v_type != type) {
2090 			mutex_exit(&vq->v_interlock);
2091 			vq = vq->v_specnext;
2092 			continue;
2093 		}
2094 		mutex_exit(&device_lock);
2095 		if (vq->v_usecount == 0) {
2096 			vremfree(vq);
2097 			vq->v_usecount = 1;
2098 		} else {
2099 			atomic_inc_uint(&vq->v_usecount);
2100 		}
2101 		vclean(vq, DOCLOSE);
2102 		vrelel(vq, 0);
2103 		mutex_enter(&device_lock);
2104 		vq = *vpp;
2105 	}
2106 	mutex_exit(&device_lock);
2107 }
2108 
2109 /*
2110  * sysctl helper routine to return list of supported fstypes
2111  */
2112 int
2113 sysctl_vfs_generic_fstypes(SYSCTLFN_ARGS)
2114 {
2115 	char bf[sizeof(((struct statvfs *)NULL)->f_fstypename)];
2116 	char *where = oldp;
2117 	struct vfsops *v;
2118 	size_t needed, left, slen;
2119 	int error, first;
2120 
2121 	if (newp != NULL)
2122 		return (EPERM);
2123 	if (namelen != 0)
2124 		return (EINVAL);
2125 
2126 	first = 1;
2127 	error = 0;
2128 	needed = 0;
2129 	left = *oldlenp;
2130 
2131 	sysctl_unlock();
2132 	mutex_enter(&vfs_list_lock);
2133 	LIST_FOREACH(v, &vfs_list, vfs_list) {
2134 		if (where == NULL)
2135 			needed += strlen(v->vfs_name) + 1;
2136 		else {
2137 			memset(bf, 0, sizeof(bf));
2138 			if (first) {
2139 				strncpy(bf, v->vfs_name, sizeof(bf));
2140 				first = 0;
2141 			} else {
2142 				bf[0] = ' ';
2143 				strncpy(bf + 1, v->vfs_name, sizeof(bf) - 1);
2144 			}
2145 			bf[sizeof(bf)-1] = '\0';
2146 			slen = strlen(bf);
2147 			if (left < slen + 1)
2148 				break;
2149 			v->vfs_refcount++;
2150 			mutex_exit(&vfs_list_lock);
2151 			/* +1 to copy out the trailing NUL byte */
2152 			error = copyout(bf, where, slen + 1);
2153 			mutex_enter(&vfs_list_lock);
2154 			v->vfs_refcount--;
2155 			if (error)
2156 				break;
2157 			where += slen;
2158 			needed += slen;
2159 			left -= slen;
2160 		}
2161 	}
2162 	mutex_exit(&vfs_list_lock);
2163 	sysctl_relock();
2164 	*oldlenp = needed;
2165 	return (error);
2166 }
2167 
2168 
2169 int kinfo_vdebug = 1;
2170 int kinfo_vgetfailed;
2171 #define KINFO_VNODESLOP	10
2172 /*
2173  * Dump vnode list (via sysctl).
2174  * Copyout address of vnode followed by vnode.
2175  */
2176 /* ARGSUSED */
2177 int
2178 sysctl_kern_vnode(SYSCTLFN_ARGS)
2179 {
2180 	char *where = oldp;
2181 	size_t *sizep = oldlenp;
2182 	struct mount *mp, *nmp;
2183 	vnode_t *vp, *mvp, vbuf;
2184 	char *bp = where;
2185 	char *ewhere;
2186 	int error;
2187 
2188 	if (namelen != 0)
2189 		return (EOPNOTSUPP);
2190 	if (newp != NULL)
2191 		return (EPERM);
2192 
2193 #define VPTRSZ	sizeof(vnode_t *)
2194 #define VNODESZ	sizeof(vnode_t)
2195 	if (where == NULL) {
2196 		*sizep = (numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ);
2197 		return (0);
2198 	}
2199 	ewhere = where + *sizep;
2200 
2201 	sysctl_unlock();
2202 	mutex_enter(&mountlist_lock);
2203 	for (mp = CIRCLEQ_FIRST(&mountlist); mp != (void *)&mountlist;
2204 	    mp = nmp) {
2205 		if (vfs_busy(mp, &nmp)) {
2206 			continue;
2207 		}
2208 		/* Allocate a marker vnode. */
2209 		mvp = vnalloc(mp);
2210 		/* Should never fail for mp != NULL */
2211 		KASSERT(mvp != NULL);
2212 		mutex_enter(&mntvnode_lock);
2213 		for (vp = TAILQ_FIRST(&mp->mnt_vnodelist); vp;
2214 		    vp = vunmark(mvp)) {
2215 			vmark(mvp, vp);
2216 			/*
2217 			 * Check that the vp is still associated with
2218 			 * this filesystem.  RACE: could have been
2219 			 * recycled onto the same filesystem.
2220 			 */
2221 			if (vp->v_mount != mp || vismarker(vp))
2222 				continue;
2223 			if (bp + VPTRSZ + VNODESZ > ewhere) {
2224 				(void)vunmark(mvp);
2225 				mutex_exit(&mntvnode_lock);
2226 				vnfree(mvp);
2227 				vfs_unbusy(mp, false, NULL);
2228 				sysctl_relock();
2229 				*sizep = bp - where;
2230 				return (ENOMEM);
2231 			}
2232 			memcpy(&vbuf, vp, VNODESZ);
2233 			mutex_exit(&mntvnode_lock);
2234 			if ((error = copyout(&vp, bp, VPTRSZ)) ||
2235 			    (error = copyout(&vbuf, bp + VPTRSZ, VNODESZ))) {
2236 			   	mutex_enter(&mntvnode_lock);
2237 				(void)vunmark(mvp);
2238 				mutex_exit(&mntvnode_lock);
2239 				vnfree(mvp);
2240 				vfs_unbusy(mp, false, NULL);
2241 				sysctl_relock();
2242 				return (error);
2243 			}
2244 			bp += VPTRSZ + VNODESZ;
2245 			mutex_enter(&mntvnode_lock);
2246 		}
2247 		mutex_exit(&mntvnode_lock);
2248 		vnfree(mvp);
2249 		vfs_unbusy(mp, false, &nmp);
2250 	}
2251 	mutex_exit(&mountlist_lock);
2252 	sysctl_relock();
2253 
2254 	*sizep = bp - where;
2255 	return (0);
2256 }
2257 
2258 /*
2259  * Remove clean vnodes from a mountpoint's vnode list.
2260  */
2261 void
2262 vfs_scrubvnlist(struct mount *mp)
2263 {
2264 	vnode_t *vp, *nvp;
2265 
2266  retry:
2267 	mutex_enter(&mntvnode_lock);
2268 	for (vp = TAILQ_FIRST(&mp->mnt_vnodelist); vp; vp = nvp) {
2269 		nvp = TAILQ_NEXT(vp, v_mntvnodes);
2270 		mutex_enter(&vp->v_interlock);
2271 		if ((vp->v_iflag & VI_CLEAN) != 0) {
2272 			TAILQ_REMOVE(&mp->mnt_vnodelist, vp, v_mntvnodes);
2273 			vp->v_mount = NULL;
2274 			mutex_exit(&mntvnode_lock);
2275 			mutex_exit(&vp->v_interlock);
2276 			vfs_destroy(mp);
2277 			goto retry;
2278 		}
2279 		mutex_exit(&vp->v_interlock);
2280 	}
2281 	mutex_exit(&mntvnode_lock);
2282 }
2283 
2284 /*
2285  * Check to see if a filesystem is mounted on a block device.
2286  */
2287 int
2288 vfs_mountedon(vnode_t *vp)
2289 {
2290 	vnode_t *vq;
2291 	int error = 0;
2292 
2293 	if (vp->v_type != VBLK)
2294 		return ENOTBLK;
2295 	if (vp->v_specmountpoint != NULL)
2296 		return (EBUSY);
2297 	mutex_enter(&device_lock);
2298 	for (vq = specfs_hash[SPECHASH(vp->v_rdev)]; vq != NULL;
2299 	    vq = vq->v_specnext) {
2300 		if (vq->v_rdev != vp->v_rdev || vq->v_type != vp->v_type)
2301 			continue;
2302 		if (vq->v_specmountpoint != NULL) {
2303 			error = EBUSY;
2304 			break;
2305 		}
2306 	}
2307 	mutex_exit(&device_lock);
2308 	return (error);
2309 }
2310 
2311 /*
2312  * Unmount all file systems.
2313  * We traverse the list in reverse order under the assumption that doing so
2314  * will avoid needing to worry about dependencies.
2315  */
2316 bool
2317 vfs_unmountall(struct lwp *l)
2318 {
2319 
2320 	printf("unmounting file systems...");
2321 	return vfs_unmountall1(l, true, true);
2322 }
2323 
2324 static void
2325 vfs_unmount_print(struct mount *mp, const char *pfx)
2326 {
2327 
2328 	aprint_verbose("%sunmounted %s on %s type %s\n", pfx,
2329 	    mp->mnt_stat.f_mntfromname, mp->mnt_stat.f_mntonname,
2330 	    mp->mnt_stat.f_fstypename);
2331 }
2332 
2333 bool
2334 vfs_unmount_forceone(struct lwp *l)
2335 {
2336 	struct mount *mp, *nmp;
2337 	int error;
2338 
2339 	nmp = NULL;
2340 
2341 	CIRCLEQ_FOREACH_REVERSE(mp, &mountlist, mnt_list) {
2342 		if (nmp == NULL || mp->mnt_gen > nmp->mnt_gen) {
2343 			nmp = mp;
2344 		}
2345 	}
2346 	if (nmp == NULL) {
2347 		return false;
2348 	}
2349 
2350 #ifdef DEBUG
2351 	printf("\nforcefully unmounting %s (%s)...",
2352 	    nmp->mnt_stat.f_mntonname, nmp->mnt_stat.f_mntfromname);
2353 #endif
2354 	atomic_inc_uint(&nmp->mnt_refcnt);
2355 	if ((error = dounmount(nmp, MNT_FORCE, l)) == 0) {
2356 		vfs_unmount_print(nmp, "forcefully ");
2357 		return true;
2358 	} else {
2359 		vfs_destroy(nmp);
2360 	}
2361 
2362 #ifdef DEBUG
2363 	printf("forceful unmount of %s failed with error %d\n",
2364 	    nmp->mnt_stat.f_mntonname, error);
2365 #endif
2366 
2367 	return false;
2368 }
2369 
2370 bool
2371 vfs_unmountall1(struct lwp *l, bool force, bool verbose)
2372 {
2373 	struct mount *mp, *nmp;
2374 	bool any_error = false, progress = false;
2375 	int error;
2376 
2377 	for (mp = CIRCLEQ_LAST(&mountlist);
2378 	     mp != (void *)&mountlist;
2379 	     mp = nmp) {
2380 		nmp = CIRCLEQ_PREV(mp, mnt_list);
2381 #ifdef DEBUG
2382 		printf("\nunmounting %p %s (%s)...",
2383 		    (void *)mp, mp->mnt_stat.f_mntonname,
2384 		    mp->mnt_stat.f_mntfromname);
2385 #endif
2386 		atomic_inc_uint(&mp->mnt_refcnt);
2387 		if ((error = dounmount(mp, force ? MNT_FORCE : 0, l)) == 0) {
2388 			vfs_unmount_print(mp, "");
2389 			progress = true;
2390 		} else {
2391 			vfs_destroy(mp);
2392 			if (verbose) {
2393 				printf("unmount of %s failed with error %d\n",
2394 				    mp->mnt_stat.f_mntonname, error);
2395 			}
2396 			any_error = true;
2397 		}
2398 	}
2399 	if (verbose) {
2400 		printf(" done\n");
2401 	}
2402 	if (any_error && verbose) {
2403 		printf("WARNING: some file systems would not unmount\n");
2404 	}
2405 	return progress;
2406 }
2407 
2408 /*
2409  * Sync and unmount file systems before shutting down.
2410  */
2411 void
2412 vfs_shutdown(void)
2413 {
2414 	struct lwp *l;
2415 
2416 	/* XXX we're certainly not running in lwp0's context! */
2417 	l = (curlwp == NULL) ? &lwp0 : curlwp;
2418 
2419 	vfs_shutdown1(l);
2420 }
2421 
2422 void
2423 vfs_sync_all(struct lwp *l)
2424 {
2425 	printf("syncing disks... ");
2426 
2427 	/* remove user processes from run queue */
2428 	suspendsched();
2429 	(void) spl0();
2430 
2431 	/* avoid coming back this way again if we panic. */
2432 	doing_shutdown = 1;
2433 
2434 	sys_sync(l, NULL, NULL);
2435 
2436 	/* Wait for sync to finish. */
2437 	if (buf_syncwait() != 0) {
2438 #if defined(DDB) && defined(DEBUG_HALT_BUSY)
2439 		Debugger();
2440 #endif
2441 		printf("giving up\n");
2442 		return;
2443 	} else
2444 		printf("done\n");
2445 }
2446 
2447 static void
2448 vfs_shutdown1(struct lwp *l)
2449 {
2450 
2451 	vfs_sync_all(l);
2452 
2453 	/*
2454 	 * If we've panic'd, don't make the situation potentially
2455 	 * worse by unmounting the file systems.
2456 	 */
2457 	if (panicstr != NULL)
2458 		return;
2459 
2460 	/* Release inodes held by texts before update. */
2461 #ifdef notdef
2462 	vnshutdown();
2463 #endif
2464 	/* Unmount file systems. */
2465 	vfs_unmountall(l);
2466 }
2467 
2468 /*
2469  * Print a list of supported file system types (used by vfs_mountroot)
2470  */
2471 static void
2472 vfs_print_fstypes(void)
2473 {
2474 	struct vfsops *v;
2475 	int cnt = 0;
2476 
2477 	mutex_enter(&vfs_list_lock);
2478 	LIST_FOREACH(v, &vfs_list, vfs_list)
2479 		++cnt;
2480 	mutex_exit(&vfs_list_lock);
2481 
2482 	if (cnt == 0) {
2483 		printf("WARNING: No file system modules have been loaded.\n");
2484 		return;
2485 	}
2486 
2487 	printf("Supported file systems:");
2488 	mutex_enter(&vfs_list_lock);
2489 	LIST_FOREACH(v, &vfs_list, vfs_list) {
2490 		printf(" %s", v->vfs_name);
2491 	}
2492 	mutex_exit(&vfs_list_lock);
2493 	printf("\n");
2494 }
2495 
2496 /*
2497  * Mount the root file system.  If the operator didn't specify a
2498  * file system to use, try all possible file systems until one
2499  * succeeds.
2500  */
2501 int
2502 vfs_mountroot(void)
2503 {
2504 	struct vfsops *v;
2505 	int error = ENODEV;
2506 
2507 	if (root_device == NULL)
2508 		panic("vfs_mountroot: root device unknown");
2509 
2510 	switch (device_class(root_device)) {
2511 	case DV_IFNET:
2512 		if (rootdev != NODEV)
2513 			panic("vfs_mountroot: rootdev set for DV_IFNET "
2514 			    "(0x%llx -> %llu,%llu)",
2515 			    (unsigned long long)rootdev,
2516 			    (unsigned long long)major(rootdev),
2517 			    (unsigned long long)minor(rootdev));
2518 		break;
2519 
2520 	case DV_DISK:
2521 		if (rootdev == NODEV)
2522 			panic("vfs_mountroot: rootdev not set for DV_DISK");
2523 	        if (bdevvp(rootdev, &rootvp))
2524 	                panic("vfs_mountroot: can't get vnode for rootdev");
2525 		error = VOP_OPEN(rootvp, FREAD, FSCRED);
2526 		if (error) {
2527 			printf("vfs_mountroot: can't open root device\n");
2528 			return (error);
2529 		}
2530 		break;
2531 
2532 	case DV_VIRTUAL:
2533 		break;
2534 
2535 	default:
2536 		printf("%s: inappropriate for root file system\n",
2537 		    device_xname(root_device));
2538 		return (ENODEV);
2539 	}
2540 
2541 	/*
2542 	 * If user specified a root fs type, use it.  Make sure the
2543 	 * specified type exists and has a mount_root()
2544 	 */
2545 	if (strcmp(rootfstype, ROOT_FSTYPE_ANY) != 0) {
2546 		v = vfs_getopsbyname(rootfstype);
2547 		error = EFTYPE;
2548 		if (v != NULL) {
2549 			if (v->vfs_mountroot != NULL) {
2550 				error = (v->vfs_mountroot)();
2551 			}
2552 			v->vfs_refcount--;
2553 		}
2554 		goto done;
2555 	}
2556 
2557 	/*
2558 	 * Try each file system currently configured into the kernel.
2559 	 */
2560 	mutex_enter(&vfs_list_lock);
2561 	LIST_FOREACH(v, &vfs_list, vfs_list) {
2562 		if (v->vfs_mountroot == NULL)
2563 			continue;
2564 #ifdef DEBUG
2565 		aprint_normal("mountroot: trying %s...\n", v->vfs_name);
2566 #endif
2567 		v->vfs_refcount++;
2568 		mutex_exit(&vfs_list_lock);
2569 		error = (*v->vfs_mountroot)();
2570 		mutex_enter(&vfs_list_lock);
2571 		v->vfs_refcount--;
2572 		if (!error) {
2573 			aprint_normal("root file system type: %s\n",
2574 			    v->vfs_name);
2575 			break;
2576 		}
2577 	}
2578 	mutex_exit(&vfs_list_lock);
2579 
2580 	if (v == NULL) {
2581 		vfs_print_fstypes();
2582 		printf("no file system for %s", device_xname(root_device));
2583 		if (device_class(root_device) == DV_DISK)
2584 			printf(" (dev 0x%llx)", (unsigned long long)rootdev);
2585 		printf("\n");
2586 		error = EFTYPE;
2587 	}
2588 
2589 done:
2590 	if (error && device_class(root_device) == DV_DISK) {
2591 		VOP_CLOSE(rootvp, FREAD, FSCRED);
2592 		vrele(rootvp);
2593 	}
2594 	if (error == 0) {
2595 		extern struct cwdinfo cwdi0;
2596 
2597 		CIRCLEQ_FIRST(&mountlist)->mnt_flag |= MNT_ROOTFS;
2598 		CIRCLEQ_FIRST(&mountlist)->mnt_op->vfs_refcount++;
2599 
2600 		/*
2601 		 * Get the vnode for '/'.  Set cwdi0.cwdi_cdir to
2602 		 * reference it.
2603 		 */
2604 		error = VFS_ROOT(CIRCLEQ_FIRST(&mountlist), &rootvnode);
2605 		if (error)
2606 			panic("cannot find root vnode, error=%d", error);
2607 		cwdi0.cwdi_cdir = rootvnode;
2608 		vref(cwdi0.cwdi_cdir);
2609 		VOP_UNLOCK(rootvnode, 0);
2610 		cwdi0.cwdi_rdir = NULL;
2611 
2612 		/*
2613 		 * Now that root is mounted, we can fixup initproc's CWD
2614 		 * info.  All other processes are kthreads, which merely
2615 		 * share proc0's CWD info.
2616 		 */
2617 		initproc->p_cwdi->cwdi_cdir = rootvnode;
2618 		vref(initproc->p_cwdi->cwdi_cdir);
2619 		initproc->p_cwdi->cwdi_rdir = NULL;
2620 	}
2621 	return (error);
2622 }
2623 
2624 /*
2625  * Get a new unique fsid
2626  */
2627 void
2628 vfs_getnewfsid(struct mount *mp)
2629 {
2630 	static u_short xxxfs_mntid;
2631 	fsid_t tfsid;
2632 	int mtype;
2633 
2634 	mutex_enter(&mntid_lock);
2635 	mtype = makefstype(mp->mnt_op->vfs_name);
2636 	mp->mnt_stat.f_fsidx.__fsid_val[0] = makedev(mtype, 0);
2637 	mp->mnt_stat.f_fsidx.__fsid_val[1] = mtype;
2638 	mp->mnt_stat.f_fsid = mp->mnt_stat.f_fsidx.__fsid_val[0];
2639 	if (xxxfs_mntid == 0)
2640 		++xxxfs_mntid;
2641 	tfsid.__fsid_val[0] = makedev(mtype & 0xff, xxxfs_mntid);
2642 	tfsid.__fsid_val[1] = mtype;
2643 	if (!CIRCLEQ_EMPTY(&mountlist)) {
2644 		while (vfs_getvfs(&tfsid)) {
2645 			tfsid.__fsid_val[0]++;
2646 			xxxfs_mntid++;
2647 		}
2648 	}
2649 	mp->mnt_stat.f_fsidx.__fsid_val[0] = tfsid.__fsid_val[0];
2650 	mp->mnt_stat.f_fsid = mp->mnt_stat.f_fsidx.__fsid_val[0];
2651 	mutex_exit(&mntid_lock);
2652 }
2653 
2654 /*
2655  * Make a 'unique' number from a mount type name.
2656  */
2657 long
2658 makefstype(const char *type)
2659 {
2660 	long rv;
2661 
2662 	for (rv = 0; *type; type++) {
2663 		rv <<= 2;
2664 		rv ^= *type;
2665 	}
2666 	return rv;
2667 }
2668 
2669 /*
2670  * Set vnode attributes to VNOVAL
2671  */
2672 void
2673 vattr_null(struct vattr *vap)
2674 {
2675 
2676 	memset(vap, 0, sizeof(*vap));
2677 
2678 	vap->va_type = VNON;
2679 
2680 	/*
2681 	 * Assign individually so that it is safe even if size and
2682 	 * sign of each member are varied.
2683 	 */
2684 	vap->va_mode = VNOVAL;
2685 	vap->va_nlink = VNOVAL;
2686 	vap->va_uid = VNOVAL;
2687 	vap->va_gid = VNOVAL;
2688 	vap->va_fsid = VNOVAL;
2689 	vap->va_fileid = VNOVAL;
2690 	vap->va_size = VNOVAL;
2691 	vap->va_blocksize = VNOVAL;
2692 	vap->va_atime.tv_sec =
2693 	    vap->va_mtime.tv_sec =
2694 	    vap->va_ctime.tv_sec =
2695 	    vap->va_birthtime.tv_sec = VNOVAL;
2696 	vap->va_atime.tv_nsec =
2697 	    vap->va_mtime.tv_nsec =
2698 	    vap->va_ctime.tv_nsec =
2699 	    vap->va_birthtime.tv_nsec = VNOVAL;
2700 	vap->va_gen = VNOVAL;
2701 	vap->va_flags = VNOVAL;
2702 	vap->va_rdev = VNOVAL;
2703 	vap->va_bytes = VNOVAL;
2704 }
2705 
2706 #define ARRAY_SIZE(arr) (sizeof(arr) / sizeof(arr[0]))
2707 #define ARRAY_PRINT(idx, arr) \
2708     ((unsigned int)(idx) < ARRAY_SIZE(arr) ? (arr)[(idx)] : "UNKNOWN")
2709 
2710 const char * const vnode_tags[] = { VNODE_TAGS };
2711 const char * const vnode_types[] = { VNODE_TYPES };
2712 const char vnode_flagbits[] = VNODE_FLAGBITS;
2713 
2714 /*
2715  * Print out a description of a vnode.
2716  */
2717 void
2718 vprint(const char *label, struct vnode *vp)
2719 {
2720 	struct vnlock *vl;
2721 	char bf[96];
2722 	int flag;
2723 
2724 	vl = (vp->v_vnlock != NULL ? vp->v_vnlock : &vp->v_lock);
2725 	flag = vp->v_iflag | vp->v_vflag | vp->v_uflag;
2726 	snprintb(bf, sizeof(bf), vnode_flagbits, flag);
2727 
2728 	if (label != NULL)
2729 		printf("%s: ", label);
2730 	printf("vnode @ %p, flags (%s)\n\ttag %s(%d), type %s(%d), "
2731 	    "usecount %d, writecount %d, holdcount %d\n"
2732 	    "\tfreelisthd %p, mount %p, data %p lock %p recursecnt %d\n",
2733 	    vp, bf, ARRAY_PRINT(vp->v_tag, vnode_tags), vp->v_tag,
2734 	    ARRAY_PRINT(vp->v_type, vnode_types), vp->v_type,
2735 	    vp->v_usecount, vp->v_writecount, vp->v_holdcnt,
2736 	    vp->v_freelisthd, vp->v_mount, vp->v_data, vl, vl->vl_recursecnt);
2737 	if (vp->v_data != NULL) {
2738 		printf("\t");
2739 		VOP_PRINT(vp);
2740 	}
2741 }
2742 
2743 #ifdef DEBUG
2744 /*
2745  * List all of the locked vnodes in the system.
2746  * Called when debugging the kernel.
2747  */
2748 void
2749 printlockedvnodes(void)
2750 {
2751 	struct mount *mp, *nmp;
2752 	struct vnode *vp;
2753 
2754 	printf("Locked vnodes\n");
2755 	mutex_enter(&mountlist_lock);
2756 	for (mp = CIRCLEQ_FIRST(&mountlist); mp != (void *)&mountlist;
2757 	     mp = nmp) {
2758 		if (vfs_busy(mp, &nmp)) {
2759 			continue;
2760 		}
2761 		TAILQ_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
2762 			if (VOP_ISLOCKED(vp))
2763 				vprint(NULL, vp);
2764 		}
2765 		mutex_enter(&mountlist_lock);
2766 		vfs_unbusy(mp, false, &nmp);
2767 	}
2768 	mutex_exit(&mountlist_lock);
2769 }
2770 #endif
2771 
2772 /* Deprecated. Kept for KPI compatibility. */
2773 int
2774 vaccess(enum vtype type, mode_t file_mode, uid_t uid, gid_t gid,
2775     mode_t acc_mode, kauth_cred_t cred)
2776 {
2777 
2778 #ifdef DIAGNOSTIC
2779 	printf("vaccess: deprecated interface used.\n");
2780 #endif /* DIAGNOSTIC */
2781 
2782 	return genfs_can_access(type, file_mode, uid, gid, acc_mode, cred);
2783 }
2784 
2785 /*
2786  * Given a file system name, look up the vfsops for that
2787  * file system, or return NULL if file system isn't present
2788  * in the kernel.
2789  */
2790 struct vfsops *
2791 vfs_getopsbyname(const char *name)
2792 {
2793 	struct vfsops *v;
2794 
2795 	mutex_enter(&vfs_list_lock);
2796 	LIST_FOREACH(v, &vfs_list, vfs_list) {
2797 		if (strcmp(v->vfs_name, name) == 0)
2798 			break;
2799 	}
2800 	if (v != NULL)
2801 		v->vfs_refcount++;
2802 	mutex_exit(&vfs_list_lock);
2803 
2804 	return (v);
2805 }
2806 
2807 void
2808 copy_statvfs_info(struct statvfs *sbp, const struct mount *mp)
2809 {
2810 	const struct statvfs *mbp;
2811 
2812 	if (sbp == (mbp = &mp->mnt_stat))
2813 		return;
2814 
2815 	(void)memcpy(&sbp->f_fsidx, &mbp->f_fsidx, sizeof(sbp->f_fsidx));
2816 	sbp->f_fsid = mbp->f_fsid;
2817 	sbp->f_owner = mbp->f_owner;
2818 	sbp->f_flag = mbp->f_flag;
2819 	sbp->f_syncwrites = mbp->f_syncwrites;
2820 	sbp->f_asyncwrites = mbp->f_asyncwrites;
2821 	sbp->f_syncreads = mbp->f_syncreads;
2822 	sbp->f_asyncreads = mbp->f_asyncreads;
2823 	(void)memcpy(sbp->f_spare, mbp->f_spare, sizeof(mbp->f_spare));
2824 	(void)memcpy(sbp->f_fstypename, mbp->f_fstypename,
2825 	    sizeof(sbp->f_fstypename));
2826 	(void)memcpy(sbp->f_mntonname, mbp->f_mntonname,
2827 	    sizeof(sbp->f_mntonname));
2828 	(void)memcpy(sbp->f_mntfromname, mp->mnt_stat.f_mntfromname,
2829 	    sizeof(sbp->f_mntfromname));
2830 	sbp->f_namemax = mbp->f_namemax;
2831 }
2832 
2833 int
2834 set_statvfs_info(const char *onp, int ukon, const char *fromp, int ukfrom,
2835     const char *vfsname, struct mount *mp, struct lwp *l)
2836 {
2837 	int error;
2838 	size_t size;
2839 	struct statvfs *sfs = &mp->mnt_stat;
2840 	int (*fun)(const void *, void *, size_t, size_t *);
2841 
2842 	(void)strlcpy(mp->mnt_stat.f_fstypename, vfsname,
2843 	    sizeof(mp->mnt_stat.f_fstypename));
2844 
2845 	if (onp) {
2846 		struct cwdinfo *cwdi = l->l_proc->p_cwdi;
2847 		fun = (ukon == UIO_SYSSPACE) ? copystr : copyinstr;
2848 		if (cwdi->cwdi_rdir != NULL) {
2849 			size_t len;
2850 			char *bp;
2851 			char *path = PNBUF_GET();
2852 
2853 			bp = path + MAXPATHLEN;
2854 			*--bp = '\0';
2855 			rw_enter(&cwdi->cwdi_lock, RW_READER);
2856 			error = getcwd_common(cwdi->cwdi_rdir, rootvnode, &bp,
2857 			    path, MAXPATHLEN / 2, 0, l);
2858 			rw_exit(&cwdi->cwdi_lock);
2859 			if (error) {
2860 				PNBUF_PUT(path);
2861 				return error;
2862 			}
2863 
2864 			len = strlen(bp);
2865 			if (len > sizeof(sfs->f_mntonname) - 1)
2866 				len = sizeof(sfs->f_mntonname) - 1;
2867 			(void)strncpy(sfs->f_mntonname, bp, len);
2868 			PNBUF_PUT(path);
2869 
2870 			if (len < sizeof(sfs->f_mntonname) - 1) {
2871 				error = (*fun)(onp, &sfs->f_mntonname[len],
2872 				    sizeof(sfs->f_mntonname) - len - 1, &size);
2873 				if (error)
2874 					return error;
2875 				size += len;
2876 			} else {
2877 				size = len;
2878 			}
2879 		} else {
2880 			error = (*fun)(onp, &sfs->f_mntonname,
2881 			    sizeof(sfs->f_mntonname) - 1, &size);
2882 			if (error)
2883 				return error;
2884 		}
2885 		(void)memset(sfs->f_mntonname + size, 0,
2886 		    sizeof(sfs->f_mntonname) - size);
2887 	}
2888 
2889 	if (fromp) {
2890 		fun = (ukfrom == UIO_SYSSPACE) ? copystr : copyinstr;
2891 		error = (*fun)(fromp, sfs->f_mntfromname,
2892 		    sizeof(sfs->f_mntfromname) - 1, &size);
2893 		if (error)
2894 			return error;
2895 		(void)memset(sfs->f_mntfromname + size, 0,
2896 		    sizeof(sfs->f_mntfromname) - size);
2897 	}
2898 	return 0;
2899 }
2900 
2901 void
2902 vfs_timestamp(struct timespec *ts)
2903 {
2904 
2905 	nanotime(ts);
2906 }
2907 
2908 time_t	rootfstime;			/* recorded root fs time, if known */
2909 void
2910 setrootfstime(time_t t)
2911 {
2912 	rootfstime = t;
2913 }
2914 
2915 /*
2916  * Sham lock manager for vnodes.  This is a temporary measure.
2917  */
2918 int
2919 vlockmgr(struct vnlock *vl, int flags)
2920 {
2921 
2922 	KASSERT((flags & ~(LK_CANRECURSE | LK_NOWAIT | LK_TYPE_MASK)) == 0);
2923 
2924 	switch (flags & LK_TYPE_MASK) {
2925 	case LK_SHARED:
2926 		if (rw_tryenter(&vl->vl_lock, RW_READER)) {
2927 			return 0;
2928 		}
2929 		if ((flags & LK_NOWAIT) != 0) {
2930 			return EBUSY;
2931 		}
2932 		rw_enter(&vl->vl_lock, RW_READER);
2933 		return 0;
2934 
2935 	case LK_EXCLUSIVE:
2936 		if (rw_tryenter(&vl->vl_lock, RW_WRITER)) {
2937 			return 0;
2938 		}
2939 		if ((vl->vl_canrecurse || (flags & LK_CANRECURSE) != 0) &&
2940 		    rw_write_held(&vl->vl_lock)) {
2941 			vl->vl_recursecnt++;
2942 			return 0;
2943 		}
2944 		if ((flags & LK_NOWAIT) != 0) {
2945 			return EBUSY;
2946 		}
2947 		rw_enter(&vl->vl_lock, RW_WRITER);
2948 		return 0;
2949 
2950 	case LK_RELEASE:
2951 		if (vl->vl_recursecnt != 0) {
2952 			KASSERT(rw_write_held(&vl->vl_lock));
2953 			vl->vl_recursecnt--;
2954 			return 0;
2955 		}
2956 		rw_exit(&vl->vl_lock);
2957 		return 0;
2958 
2959 	default:
2960 		panic("vlockmgr: flags %x", flags);
2961 	}
2962 }
2963 
2964 int
2965 vlockstatus(struct vnlock *vl)
2966 {
2967 
2968 	if (rw_write_held(&vl->vl_lock)) {
2969 		return LK_EXCLUSIVE;
2970 	}
2971 	if (rw_read_held(&vl->vl_lock)) {
2972 		return LK_SHARED;
2973 	}
2974 	return 0;
2975 }
2976 
2977 static const uint8_t vttodt_tab[9] = {
2978 	DT_UNKNOWN,	/* VNON  */
2979 	DT_REG,		/* VREG  */
2980 	DT_DIR,		/* VDIR  */
2981 	DT_BLK,		/* VBLK  */
2982 	DT_CHR,		/* VCHR  */
2983 	DT_LNK,		/* VLNK  */
2984 	DT_SOCK,	/* VSUCK */
2985 	DT_FIFO,	/* VFIFO */
2986 	DT_UNKNOWN	/* VBAD  */
2987 };
2988 
2989 uint8_t
2990 vtype2dt(enum vtype vt)
2991 {
2992 
2993 	CTASSERT(VBAD == __arraycount(vttodt_tab) - 1);
2994 	return vttodt_tab[vt];
2995 }
2996 
2997 /*
2998  * mount_specific_key_create --
2999  *	Create a key for subsystem mount-specific data.
3000  */
3001 int
3002 mount_specific_key_create(specificdata_key_t *keyp, specificdata_dtor_t dtor)
3003 {
3004 
3005 	return (specificdata_key_create(mount_specificdata_domain, keyp, dtor));
3006 }
3007 
3008 /*
3009  * mount_specific_key_delete --
3010  *	Delete a key for subsystem mount-specific data.
3011  */
3012 void
3013 mount_specific_key_delete(specificdata_key_t key)
3014 {
3015 
3016 	specificdata_key_delete(mount_specificdata_domain, key);
3017 }
3018 
3019 /*
3020  * mount_initspecific --
3021  *	Initialize a mount's specificdata container.
3022  */
3023 void
3024 mount_initspecific(struct mount *mp)
3025 {
3026 	int error;
3027 
3028 	error = specificdata_init(mount_specificdata_domain,
3029 				  &mp->mnt_specdataref);
3030 	KASSERT(error == 0);
3031 }
3032 
3033 /*
3034  * mount_finispecific --
3035  *	Finalize a mount's specificdata container.
3036  */
3037 void
3038 mount_finispecific(struct mount *mp)
3039 {
3040 
3041 	specificdata_fini(mount_specificdata_domain, &mp->mnt_specdataref);
3042 }
3043 
3044 /*
3045  * mount_getspecific --
3046  *	Return mount-specific data corresponding to the specified key.
3047  */
3048 void *
3049 mount_getspecific(struct mount *mp, specificdata_key_t key)
3050 {
3051 
3052 	return (specificdata_getspecific(mount_specificdata_domain,
3053 					 &mp->mnt_specdataref, key));
3054 }
3055 
3056 /*
3057  * mount_setspecific --
3058  *	Set mount-specific data corresponding to the specified key.
3059  */
3060 void
3061 mount_setspecific(struct mount *mp, specificdata_key_t key, void *data)
3062 {
3063 
3064 	specificdata_setspecific(mount_specificdata_domain,
3065 				 &mp->mnt_specdataref, key, data);
3066 }
3067 
3068 int
3069 VFS_MOUNT(struct mount *mp, const char *a, void *b, size_t *c)
3070 {
3071 	int error;
3072 
3073 	KERNEL_LOCK(1, NULL);
3074 	error = (*(mp->mnt_op->vfs_mount))(mp, a, b, c);
3075 	KERNEL_UNLOCK_ONE(NULL);
3076 
3077 	return error;
3078 }
3079 
3080 int
3081 VFS_START(struct mount *mp, int a)
3082 {
3083 	int error;
3084 
3085 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
3086 		KERNEL_LOCK(1, NULL);
3087 	}
3088 	error = (*(mp->mnt_op->vfs_start))(mp, a);
3089 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
3090 		KERNEL_UNLOCK_ONE(NULL);
3091 	}
3092 
3093 	return error;
3094 }
3095 
3096 int
3097 VFS_UNMOUNT(struct mount *mp, int a)
3098 {
3099 	int error;
3100 
3101 	KERNEL_LOCK(1, NULL);
3102 	error = (*(mp->mnt_op->vfs_unmount))(mp, a);
3103 	KERNEL_UNLOCK_ONE(NULL);
3104 
3105 	return error;
3106 }
3107 
3108 int
3109 VFS_ROOT(struct mount *mp, struct vnode **a)
3110 {
3111 	int error;
3112 
3113 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
3114 		KERNEL_LOCK(1, NULL);
3115 	}
3116 	error = (*(mp->mnt_op->vfs_root))(mp, a);
3117 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
3118 		KERNEL_UNLOCK_ONE(NULL);
3119 	}
3120 
3121 	return error;
3122 }
3123 
3124 int
3125 VFS_QUOTACTL(struct mount *mp, int a, uid_t b, void *c)
3126 {
3127 	int error;
3128 
3129 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
3130 		KERNEL_LOCK(1, NULL);
3131 	}
3132 	error = (*(mp->mnt_op->vfs_quotactl))(mp, a, b, c);
3133 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
3134 		KERNEL_UNLOCK_ONE(NULL);
3135 	}
3136 
3137 	return error;
3138 }
3139 
3140 int
3141 VFS_STATVFS(struct mount *mp, struct statvfs *a)
3142 {
3143 	int error;
3144 
3145 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
3146 		KERNEL_LOCK(1, NULL);
3147 	}
3148 	error = (*(mp->mnt_op->vfs_statvfs))(mp, a);
3149 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
3150 		KERNEL_UNLOCK_ONE(NULL);
3151 	}
3152 
3153 	return error;
3154 }
3155 
3156 int
3157 VFS_SYNC(struct mount *mp, int a, struct kauth_cred *b)
3158 {
3159 	int error;
3160 
3161 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
3162 		KERNEL_LOCK(1, NULL);
3163 	}
3164 	error = (*(mp->mnt_op->vfs_sync))(mp, a, b);
3165 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
3166 		KERNEL_UNLOCK_ONE(NULL);
3167 	}
3168 
3169 	return error;
3170 }
3171 
3172 int
3173 VFS_FHTOVP(struct mount *mp, struct fid *a, struct vnode **b)
3174 {
3175 	int error;
3176 
3177 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
3178 		KERNEL_LOCK(1, NULL);
3179 	}
3180 	error = (*(mp->mnt_op->vfs_fhtovp))(mp, a, b);
3181 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
3182 		KERNEL_UNLOCK_ONE(NULL);
3183 	}
3184 
3185 	return error;
3186 }
3187 
3188 int
3189 VFS_VPTOFH(struct vnode *vp, struct fid *a, size_t *b)
3190 {
3191 	int error;
3192 
3193 	if ((vp->v_vflag & VV_MPSAFE) == 0) {
3194 		KERNEL_LOCK(1, NULL);
3195 	}
3196 	error = (*(vp->v_mount->mnt_op->vfs_vptofh))(vp, a, b);
3197 	if ((vp->v_vflag & VV_MPSAFE) == 0) {
3198 		KERNEL_UNLOCK_ONE(NULL);
3199 	}
3200 
3201 	return error;
3202 }
3203 
3204 int
3205 VFS_SNAPSHOT(struct mount *mp, struct vnode *a, struct timespec *b)
3206 {
3207 	int error;
3208 
3209 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
3210 		KERNEL_LOCK(1, NULL);
3211 	}
3212 	error = (*(mp->mnt_op->vfs_snapshot))(mp, a, b);
3213 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
3214 		KERNEL_UNLOCK_ONE(NULL);
3215 	}
3216 
3217 	return error;
3218 }
3219 
3220 int
3221 VFS_EXTATTRCTL(struct mount *mp, int a, struct vnode *b, int c, const char *d)
3222 {
3223 	int error;
3224 
3225 	KERNEL_LOCK(1, NULL);		/* XXXSMP check ffs */
3226 	error = (*(mp->mnt_op->vfs_extattrctl))(mp, a, b, c, d);
3227 	KERNEL_UNLOCK_ONE(NULL);	/* XXX */
3228 
3229 	return error;
3230 }
3231 
3232 int
3233 VFS_SUSPENDCTL(struct mount *mp, int a)
3234 {
3235 	int error;
3236 
3237 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
3238 		KERNEL_LOCK(1, NULL);
3239 	}
3240 	error = (*(mp->mnt_op->vfs_suspendctl))(mp, a);
3241 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
3242 		KERNEL_UNLOCK_ONE(NULL);
3243 	}
3244 
3245 	return error;
3246 }
3247 
3248 #if defined(DDB) || defined(DEBUGPRINT)
3249 static const char buf_flagbits[] = BUF_FLAGBITS;
3250 
3251 void
3252 vfs_buf_print(struct buf *bp, int full, void (*pr)(const char *, ...))
3253 {
3254 	char bf[1024];
3255 
3256 	(*pr)("  vp %p lblkno 0x%"PRIx64" blkno 0x%"PRIx64" rawblkno 0x%"
3257 	    PRIx64 " dev 0x%x\n",
3258 	    bp->b_vp, bp->b_lblkno, bp->b_blkno, bp->b_rawblkno, bp->b_dev);
3259 
3260 	snprintb(bf, sizeof(bf),
3261 	    buf_flagbits, bp->b_flags | bp->b_oflags | bp->b_cflags);
3262 	(*pr)("  error %d flags 0x%s\n", bp->b_error, bf);
3263 
3264 	(*pr)("  bufsize 0x%lx bcount 0x%lx resid 0x%lx\n",
3265 		  bp->b_bufsize, bp->b_bcount, bp->b_resid);
3266 	(*pr)("  data %p saveaddr %p\n",
3267 		  bp->b_data, bp->b_saveaddr);
3268 	(*pr)("  iodone %p objlock %p\n", bp->b_iodone, bp->b_objlock);
3269 }
3270 
3271 
3272 void
3273 vfs_vnode_print(struct vnode *vp, int full, void (*pr)(const char *, ...))
3274 {
3275 	char bf[256];
3276 
3277 	uvm_object_printit(&vp->v_uobj, full, pr);
3278 	snprintb(bf, sizeof(bf),
3279 	    vnode_flagbits, vp->v_iflag | vp->v_vflag | vp->v_uflag);
3280 	(*pr)("\nVNODE flags %s\n", bf);
3281 	(*pr)("mp %p numoutput %d size 0x%llx writesize 0x%llx\n",
3282 	      vp->v_mount, vp->v_numoutput, vp->v_size, vp->v_writesize);
3283 
3284 	(*pr)("data %p writecount %ld holdcnt %ld\n",
3285 	      vp->v_data, vp->v_writecount, vp->v_holdcnt);
3286 
3287 	(*pr)("tag %s(%d) type %s(%d) mount %p typedata %p\n",
3288 	      ARRAY_PRINT(vp->v_tag, vnode_tags), vp->v_tag,
3289 	      ARRAY_PRINT(vp->v_type, vnode_types), vp->v_type,
3290 	      vp->v_mount, vp->v_mountedhere);
3291 
3292 	(*pr)("v_lock %p v_vnlock %p\n", &vp->v_lock, vp->v_vnlock);
3293 
3294 	if (full) {
3295 		struct buf *bp;
3296 
3297 		(*pr)("clean bufs:\n");
3298 		LIST_FOREACH(bp, &vp->v_cleanblkhd, b_vnbufs) {
3299 			(*pr)(" bp %p\n", bp);
3300 			vfs_buf_print(bp, full, pr);
3301 		}
3302 
3303 		(*pr)("dirty bufs:\n");
3304 		LIST_FOREACH(bp, &vp->v_dirtyblkhd, b_vnbufs) {
3305 			(*pr)(" bp %p\n", bp);
3306 			vfs_buf_print(bp, full, pr);
3307 		}
3308 	}
3309 }
3310 
3311 void
3312 vfs_mount_print(struct mount *mp, int full, void (*pr)(const char *, ...))
3313 {
3314 	char sbuf[256];
3315 
3316 	(*pr)("vnodecovered = %p syncer = %p data = %p\n",
3317 			mp->mnt_vnodecovered,mp->mnt_syncer,mp->mnt_data);
3318 
3319 	(*pr)("fs_bshift %d dev_bshift = %d\n",
3320 			mp->mnt_fs_bshift,mp->mnt_dev_bshift);
3321 
3322 	snprintb(sbuf, sizeof(sbuf), __MNT_FLAG_BITS, mp->mnt_flag);
3323 	(*pr)("flag = %s\n", sbuf);
3324 
3325 	snprintb(sbuf, sizeof(sbuf), __IMNT_FLAG_BITS, mp->mnt_iflag);
3326 	(*pr)("iflag = %s\n", sbuf);
3327 
3328 	(*pr)("refcnt = %d unmounting @ %p updating @ %p\n", mp->mnt_refcnt,
3329 	    &mp->mnt_unmounting, &mp->mnt_updating);
3330 
3331 	(*pr)("statvfs cache:\n");
3332 	(*pr)("\tbsize = %lu\n",mp->mnt_stat.f_bsize);
3333 	(*pr)("\tfrsize = %lu\n",mp->mnt_stat.f_frsize);
3334 	(*pr)("\tiosize = %lu\n",mp->mnt_stat.f_iosize);
3335 
3336 	(*pr)("\tblocks = %"PRIu64"\n",mp->mnt_stat.f_blocks);
3337 	(*pr)("\tbfree = %"PRIu64"\n",mp->mnt_stat.f_bfree);
3338 	(*pr)("\tbavail = %"PRIu64"\n",mp->mnt_stat.f_bavail);
3339 	(*pr)("\tbresvd = %"PRIu64"\n",mp->mnt_stat.f_bresvd);
3340 
3341 	(*pr)("\tfiles = %"PRIu64"\n",mp->mnt_stat.f_files);
3342 	(*pr)("\tffree = %"PRIu64"\n",mp->mnt_stat.f_ffree);
3343 	(*pr)("\tfavail = %"PRIu64"\n",mp->mnt_stat.f_favail);
3344 	(*pr)("\tfresvd = %"PRIu64"\n",mp->mnt_stat.f_fresvd);
3345 
3346 	(*pr)("\tf_fsidx = { 0x%"PRIx32", 0x%"PRIx32" }\n",
3347 			mp->mnt_stat.f_fsidx.__fsid_val[0],
3348 			mp->mnt_stat.f_fsidx.__fsid_val[1]);
3349 
3350 	(*pr)("\towner = %"PRIu32"\n",mp->mnt_stat.f_owner);
3351 	(*pr)("\tnamemax = %lu\n",mp->mnt_stat.f_namemax);
3352 
3353 	snprintb(sbuf, sizeof(sbuf), __MNT_FLAG_BITS, mp->mnt_stat.f_flag);
3354 
3355 	(*pr)("\tflag = %s\n",sbuf);
3356 	(*pr)("\tsyncwrites = %" PRIu64 "\n",mp->mnt_stat.f_syncwrites);
3357 	(*pr)("\tasyncwrites = %" PRIu64 "\n",mp->mnt_stat.f_asyncwrites);
3358 	(*pr)("\tsyncreads = %" PRIu64 "\n",mp->mnt_stat.f_syncreads);
3359 	(*pr)("\tasyncreads = %" PRIu64 "\n",mp->mnt_stat.f_asyncreads);
3360 	(*pr)("\tfstypename = %s\n",mp->mnt_stat.f_fstypename);
3361 	(*pr)("\tmntonname = %s\n",mp->mnt_stat.f_mntonname);
3362 	(*pr)("\tmntfromname = %s\n",mp->mnt_stat.f_mntfromname);
3363 
3364 	{
3365 		int cnt = 0;
3366 		struct vnode *vp;
3367 		(*pr)("locked vnodes =");
3368 		TAILQ_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
3369 			if (VOP_ISLOCKED(vp)) {
3370 				if ((++cnt % 6) == 0) {
3371 					(*pr)(" %p,\n\t", vp);
3372 				} else {
3373 					(*pr)(" %p,", vp);
3374 				}
3375 			}
3376 		}
3377 		(*pr)("\n");
3378 	}
3379 
3380 	if (full) {
3381 		int cnt = 0;
3382 		struct vnode *vp;
3383 		(*pr)("all vnodes =");
3384 		TAILQ_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
3385 			if (!TAILQ_NEXT(vp, v_mntvnodes)) {
3386 				(*pr)(" %p", vp);
3387 			} else if ((++cnt % 6) == 0) {
3388 				(*pr)(" %p,\n\t", vp);
3389 			} else {
3390 				(*pr)(" %p,", vp);
3391 			}
3392 		}
3393 		(*pr)("\n", vp);
3394 	}
3395 }
3396 #endif /* DDB || DEBUGPRINT */
3397 
3398 /*
3399  * Check if a device pointed to by vp is mounted.
3400  *
3401  * Returns:
3402  *   EINVAL	if it's not a disk
3403  *   EBUSY	if it's a disk and mounted
3404  *   0		if it's a disk and not mounted
3405  */
3406 int
3407 rawdev_mounted(struct vnode *vp, struct vnode **bvpp)
3408 {
3409 	struct vnode *bvp;
3410 	dev_t dev;
3411 	int d_type;
3412 
3413 	bvp = NULL;
3414 	dev = vp->v_rdev;
3415 	d_type = D_OTHER;
3416 
3417 	if (iskmemvp(vp))
3418 		return EINVAL;
3419 
3420 	switch (vp->v_type) {
3421 	case VCHR: {
3422 		const struct cdevsw *cdev;
3423 
3424 		cdev = cdevsw_lookup(dev);
3425 		if (cdev != NULL) {
3426 			dev_t blkdev;
3427 
3428 			blkdev = devsw_chr2blk(dev);
3429 			if (blkdev != NODEV) {
3430 				vfinddev(blkdev, VBLK, &bvp);
3431 				if (bvp != NULL)
3432 					d_type = (cdev->d_flag & D_TYPEMASK);
3433 			}
3434 		}
3435 
3436 		break;
3437 		}
3438 
3439 	case VBLK: {
3440 		const struct bdevsw *bdev;
3441 
3442 		bdev = bdevsw_lookup(dev);
3443 		if (bdev != NULL)
3444 			d_type = (bdev->d_flag & D_TYPEMASK);
3445 
3446 		bvp = vp;
3447 
3448 		break;
3449 		}
3450 
3451 	default:
3452 		break;
3453 	}
3454 
3455 	if (d_type != D_DISK)
3456 		return EINVAL;
3457 
3458 	if (bvpp != NULL)
3459 		*bvpp = bvp;
3460 
3461 	/*
3462 	 * XXX: This is bogus. We should be failing the request
3463 	 * XXX: not only if this specific slice is mounted, but
3464 	 * XXX: if it's on a disk with any other mounted slice.
3465 	 */
3466 	if (vfs_mountedon(bvp))
3467 		return EBUSY;
3468 
3469 	return 0;
3470 }
3471