xref: /netbsd-src/sys/kern/vfs_subr.c (revision 1423e65b26cfa3a30fb7aaea6d57132588c43746)
1 /*	$NetBSD: vfs_subr.c,v 1.407 2010/06/24 13:03:12 hannken Exp $	*/
2 
3 /*-
4  * Copyright (c) 1997, 1998, 2004, 2005, 2007, 2008 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility,
9  * NASA Ames Research Center, by Charles M. Hannum, and by Andrew Doran.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
21  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
22  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
23  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
24  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
27  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
29  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
30  * POSSIBILITY OF SUCH DAMAGE.
31  */
32 
33 /*
34  * Copyright (c) 1989, 1993
35  *	The Regents of the University of California.  All rights reserved.
36  * (c) UNIX System Laboratories, Inc.
37  * All or some portions of this file are derived from material licensed
38  * to the University of California by American Telephone and Telegraph
39  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
40  * the permission of UNIX System Laboratories, Inc.
41  *
42  * Redistribution and use in source and binary forms, with or without
43  * modification, are permitted provided that the following conditions
44  * are met:
45  * 1. Redistributions of source code must retain the above copyright
46  *    notice, this list of conditions and the following disclaimer.
47  * 2. Redistributions in binary form must reproduce the above copyright
48  *    notice, this list of conditions and the following disclaimer in the
49  *    documentation and/or other materials provided with the distribution.
50  * 3. Neither the name of the University nor the names of its contributors
51  *    may be used to endorse or promote products derived from this software
52  *    without specific prior written permission.
53  *
54  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
55  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
56  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
57  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
58  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
59  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
60  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
61  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
62  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
63  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
64  * SUCH DAMAGE.
65  *
66  *	@(#)vfs_subr.c	8.13 (Berkeley) 4/18/94
67  */
68 
69 /*
70  * Note on v_usecount and locking:
71  *
72  * At nearly all points it is known that v_usecount could be zero, the
73  * vnode interlock will be held.
74  *
75  * To change v_usecount away from zero, the interlock must be held.  To
76  * change from a non-zero value to zero, again the interlock must be
77  * held.
78  *
79  * There's a flag bit, VC_XLOCK, embedded in v_usecount.
80  * To raise v_usecount, if the VC_XLOCK bit is set in it, the interlock
81  * must be held.
82  * To modify the VC_XLOCK bit, the interlock must be held.
83  * We always keep the usecount (v_usecount & VC_MASK) non-zero while the
84  * VC_XLOCK bit is set.
85  *
86  * Unless the VC_XLOCK bit is set, changing the usecount from a non-zero
87  * value to a non-zero value can safely be done using atomic operations,
88  * without the interlock held.
89  * Even if the VC_XLOCK bit is set, decreasing the usecount to a non-zero
90  * value can be done using atomic operations, without the interlock held.
91  */
92 
93 #include <sys/cdefs.h>
94 __KERNEL_RCSID(0, "$NetBSD: vfs_subr.c,v 1.407 2010/06/24 13:03:12 hannken Exp $");
95 
96 #include "opt_ddb.h"
97 #include "opt_compat_netbsd.h"
98 #include "opt_compat_43.h"
99 
100 #include <sys/param.h>
101 #include <sys/systm.h>
102 #include <sys/conf.h>
103 #include <sys/dirent.h>
104 #include <sys/proc.h>
105 #include <sys/kernel.h>
106 #include <sys/mount.h>
107 #include <sys/fcntl.h>
108 #include <sys/vnode.h>
109 #include <sys/stat.h>
110 #include <sys/namei.h>
111 #include <sys/ucred.h>
112 #include <sys/buf.h>
113 #include <sys/errno.h>
114 #include <sys/kmem.h>
115 #include <sys/syscallargs.h>
116 #include <sys/device.h>
117 #include <sys/filedesc.h>
118 #include <sys/kauth.h>
119 #include <sys/atomic.h>
120 #include <sys/kthread.h>
121 #include <sys/wapbl.h>
122 #include <sys/module.h>
123 
124 #include <miscfs/genfs/genfs.h>
125 #include <miscfs/specfs/specdev.h>
126 #include <miscfs/syncfs/syncfs.h>
127 
128 #include <uvm/uvm.h>
129 #include <uvm/uvm_readahead.h>
130 #include <uvm/uvm_ddb.h>
131 
132 #include <sys/sysctl.h>
133 
134 const enum vtype iftovt_tab[16] = {
135 	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
136 	VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
137 };
138 const int	vttoif_tab[9] = {
139 	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
140 	S_IFSOCK, S_IFIFO, S_IFMT,
141 };
142 
143 /*
144  * Insq/Remq for the vnode usage lists.
145  */
146 #define	bufinsvn(bp, dp)	LIST_INSERT_HEAD(dp, bp, b_vnbufs)
147 #define	bufremvn(bp) {							\
148 	LIST_REMOVE(bp, b_vnbufs);					\
149 	(bp)->b_vnbufs.le_next = NOLIST;				\
150 }
151 
152 int doforce = 1;		/* 1 => permit forcible unmounting */
153 int prtactive = 0;		/* 1 => print out reclaim of active vnodes */
154 
155 static vnodelst_t vnode_free_list = TAILQ_HEAD_INITIALIZER(vnode_free_list);
156 static vnodelst_t vnode_hold_list = TAILQ_HEAD_INITIALIZER(vnode_hold_list);
157 static vnodelst_t vrele_list = TAILQ_HEAD_INITIALIZER(vrele_list);
158 
159 struct mntlist mountlist =			/* mounted filesystem list */
160     CIRCLEQ_HEAD_INITIALIZER(mountlist);
161 
162 u_int numvnodes;
163 static specificdata_domain_t mount_specificdata_domain;
164 
165 static int vrele_pending;
166 static int vrele_gen;
167 static kmutex_t	vrele_lock;
168 static kcondvar_t vrele_cv;
169 static lwp_t *vrele_lwp;
170 
171 static uint64_t mountgen = 0;
172 static kmutex_t mountgen_lock;
173 
174 kmutex_t mountlist_lock;
175 kmutex_t mntid_lock;
176 kmutex_t mntvnode_lock;
177 kmutex_t vnode_free_list_lock;
178 kmutex_t vfs_list_lock;
179 
180 static pool_cache_t vnode_cache;
181 
182 /*
183  * These define the root filesystem and device.
184  */
185 struct vnode *rootvnode;
186 struct device *root_device;			/* root device */
187 
188 /*
189  * Local declarations.
190  */
191 
192 static void vrele_thread(void *);
193 static void insmntque(vnode_t *, struct mount *);
194 static int getdevvp(dev_t, vnode_t **, enum vtype);
195 static vnode_t *getcleanvnode(void);
196 void vpanic(vnode_t *, const char *);
197 static void vfs_shutdown1(struct lwp *);
198 
199 #ifdef DEBUG
200 void printlockedvnodes(void);
201 #endif
202 
203 #ifdef DIAGNOSTIC
204 void
205 vpanic(vnode_t *vp, const char *msg)
206 {
207 
208 	vprint(NULL, vp);
209 	panic("%s\n", msg);
210 }
211 #else
212 #define	vpanic(vp, msg)	/* nothing */
213 #endif
214 
215 void
216 vn_init1(void)
217 {
218 
219 	vnode_cache = pool_cache_init(sizeof(struct vnode), 0, 0, 0, "vnodepl",
220 	    NULL, IPL_NONE, NULL, NULL, NULL);
221 	KASSERT(vnode_cache != NULL);
222 
223 	/* Create deferred release thread. */
224 	mutex_init(&vrele_lock, MUTEX_DEFAULT, IPL_NONE);
225 	cv_init(&vrele_cv, "vrele");
226 	if (kthread_create(PRI_VM, KTHREAD_MPSAFE, NULL, vrele_thread,
227 	    NULL, &vrele_lwp, "vrele"))
228 		panic("fork vrele");
229 }
230 
231 /*
232  * Initialize the vnode management data structures.
233  */
234 void
235 vntblinit(void)
236 {
237 
238 	mutex_init(&mountgen_lock, MUTEX_DEFAULT, IPL_NONE);
239 	mutex_init(&mountlist_lock, MUTEX_DEFAULT, IPL_NONE);
240 	mutex_init(&mntid_lock, MUTEX_DEFAULT, IPL_NONE);
241 	mutex_init(&mntvnode_lock, MUTEX_DEFAULT, IPL_NONE);
242 	mutex_init(&vnode_free_list_lock, MUTEX_DEFAULT, IPL_NONE);
243 	mutex_init(&vfs_list_lock, MUTEX_DEFAULT, IPL_NONE);
244 
245 	mount_specificdata_domain = specificdata_domain_create();
246 
247 	/* Initialize the filesystem syncer. */
248 	vn_initialize_syncerd();
249 	vn_init1();
250 }
251 
252 int
253 vfs_drainvnodes(long target, struct lwp *l)
254 {
255 
256 	while (numvnodes > target) {
257 		vnode_t *vp;
258 
259 		mutex_enter(&vnode_free_list_lock);
260 		vp = getcleanvnode();
261 		if (vp == NULL)
262 			return EBUSY; /* give up */
263 		ungetnewvnode(vp);
264 	}
265 
266 	return 0;
267 }
268 
269 /*
270  * Lookup a mount point by filesystem identifier.
271  *
272  * XXX Needs to add a reference to the mount point.
273  */
274 struct mount *
275 vfs_getvfs(fsid_t *fsid)
276 {
277 	struct mount *mp;
278 
279 	mutex_enter(&mountlist_lock);
280 	CIRCLEQ_FOREACH(mp, &mountlist, mnt_list) {
281 		if (mp->mnt_stat.f_fsidx.__fsid_val[0] == fsid->__fsid_val[0] &&
282 		    mp->mnt_stat.f_fsidx.__fsid_val[1] == fsid->__fsid_val[1]) {
283 			mutex_exit(&mountlist_lock);
284 			return (mp);
285 		}
286 	}
287 	mutex_exit(&mountlist_lock);
288 	return ((struct mount *)0);
289 }
290 
291 /*
292  * Drop a reference to a mount structure, freeing if the last reference.
293  */
294 void
295 vfs_destroy(struct mount *mp)
296 {
297 
298 	if (__predict_true((int)atomic_dec_uint_nv(&mp->mnt_refcnt) > 0)) {
299 		return;
300 	}
301 
302 	/*
303 	 * Nothing else has visibility of the mount: we can now
304 	 * free the data structures.
305 	 */
306 	KASSERT(mp->mnt_refcnt == 0);
307 	specificdata_fini(mount_specificdata_domain, &mp->mnt_specdataref);
308 	rw_destroy(&mp->mnt_unmounting);
309 	mutex_destroy(&mp->mnt_updating);
310 	mutex_destroy(&mp->mnt_renamelock);
311 	if (mp->mnt_op != NULL) {
312 		vfs_delref(mp->mnt_op);
313 	}
314 	kmem_free(mp, sizeof(*mp));
315 }
316 
317 /*
318  * grab a vnode from freelist and clean it.
319  */
320 vnode_t *
321 getcleanvnode(void)
322 {
323 	vnode_t *vp;
324 	vnodelst_t *listhd;
325 
326 	KASSERT(mutex_owned(&vnode_free_list_lock));
327 
328 retry:
329 	listhd = &vnode_free_list;
330 try_nextlist:
331 	TAILQ_FOREACH(vp, listhd, v_freelist) {
332 		/*
333 		 * It's safe to test v_usecount and v_iflag
334 		 * without holding the interlock here, since
335 		 * these vnodes should never appear on the
336 		 * lists.
337 		 */
338 		if (vp->v_usecount != 0) {
339 			vpanic(vp, "free vnode isn't");
340 		}
341 		if ((vp->v_iflag & VI_CLEAN) != 0) {
342 			vpanic(vp, "clean vnode on freelist");
343 		}
344 		if (vp->v_freelisthd != listhd) {
345 			printf("vnode sez %p, listhd %p\n", vp->v_freelisthd, listhd);
346 			vpanic(vp, "list head mismatch");
347 		}
348 		if (!mutex_tryenter(&vp->v_interlock))
349 			continue;
350 		if ((vp->v_iflag & VI_XLOCK) == 0)
351 			break;
352 		mutex_exit(&vp->v_interlock);
353 	}
354 
355 	if (vp == NULL) {
356 		if (listhd == &vnode_free_list) {
357 			listhd = &vnode_hold_list;
358 			goto try_nextlist;
359 		}
360 		mutex_exit(&vnode_free_list_lock);
361 		return NULL;
362 	}
363 
364 	/* Remove it from the freelist. */
365 	TAILQ_REMOVE(listhd, vp, v_freelist);
366 	vp->v_freelisthd = NULL;
367 	mutex_exit(&vnode_free_list_lock);
368 
369 	if (vp->v_usecount != 0) {
370 		/*
371 		 * was referenced again before we got the interlock
372 		 * Don't return to freelist - the holder of the last
373 		 * reference will destroy it.
374 		 */
375 		mutex_exit(&vp->v_interlock);
376 		mutex_enter(&vnode_free_list_lock);
377 		goto retry;
378 	}
379 
380 	/*
381 	 * The vnode is still associated with a file system, so we must
382 	 * clean it out before reusing it.  We need to add a reference
383 	 * before doing this.  If the vnode gains another reference while
384 	 * being cleaned out then we lose - retry.
385 	 */
386 	atomic_add_int(&vp->v_usecount, 1 + VC_XLOCK);
387 	vclean(vp, DOCLOSE);
388 	KASSERT(vp->v_usecount >= 1 + VC_XLOCK);
389 	atomic_add_int(&vp->v_usecount, -VC_XLOCK);
390 	if (vp->v_usecount == 1) {
391 		/* We're about to dirty it. */
392 		vp->v_iflag &= ~VI_CLEAN;
393 		mutex_exit(&vp->v_interlock);
394 		if (vp->v_type == VBLK || vp->v_type == VCHR) {
395 			spec_node_destroy(vp);
396 		}
397 		vp->v_type = VNON;
398 	} else {
399 		/*
400 		 * Don't return to freelist - the holder of the last
401 		 * reference will destroy it.
402 		 */
403 		vrelel(vp, 0); /* releases vp->v_interlock */
404 		mutex_enter(&vnode_free_list_lock);
405 		goto retry;
406 	}
407 
408 	if (vp->v_data != NULL || vp->v_uobj.uo_npages != 0 ||
409 	    !TAILQ_EMPTY(&vp->v_uobj.memq)) {
410 		vpanic(vp, "cleaned vnode isn't");
411 	}
412 	if (vp->v_numoutput != 0) {
413 		vpanic(vp, "clean vnode has pending I/O's");
414 	}
415 	if ((vp->v_iflag & VI_ONWORKLST) != 0) {
416 		vpanic(vp, "clean vnode on syncer list");
417 	}
418 
419 	return vp;
420 }
421 
422 /*
423  * Mark a mount point as busy, and gain a new reference to it.  Used to
424  * prevent the file system from being unmounted during critical sections.
425  *
426  * => The caller must hold a pre-existing reference to the mount.
427  * => Will fail if the file system is being unmounted, or is unmounted.
428  */
429 int
430 vfs_busy(struct mount *mp, struct mount **nextp)
431 {
432 
433 	KASSERT(mp->mnt_refcnt > 0);
434 
435 	if (__predict_false(!rw_tryenter(&mp->mnt_unmounting, RW_READER))) {
436 		if (nextp != NULL) {
437 			KASSERT(mutex_owned(&mountlist_lock));
438 			*nextp = CIRCLEQ_NEXT(mp, mnt_list);
439 		}
440 		return EBUSY;
441 	}
442 	if (__predict_false((mp->mnt_iflag & IMNT_GONE) != 0)) {
443 		rw_exit(&mp->mnt_unmounting);
444 		if (nextp != NULL) {
445 			KASSERT(mutex_owned(&mountlist_lock));
446 			*nextp = CIRCLEQ_NEXT(mp, mnt_list);
447 		}
448 		return ENOENT;
449 	}
450 	if (nextp != NULL) {
451 		mutex_exit(&mountlist_lock);
452 	}
453 	atomic_inc_uint(&mp->mnt_refcnt);
454 	return 0;
455 }
456 
457 /*
458  * Unbusy a busy filesystem.
459  *
460  * => If keepref is true, preserve reference added by vfs_busy().
461  * => If nextp != NULL, acquire mountlist_lock.
462  */
463 void
464 vfs_unbusy(struct mount *mp, bool keepref, struct mount **nextp)
465 {
466 
467 	KASSERT(mp->mnt_refcnt > 0);
468 
469 	if (nextp != NULL) {
470 		mutex_enter(&mountlist_lock);
471 	}
472 	rw_exit(&mp->mnt_unmounting);
473 	if (!keepref) {
474 		vfs_destroy(mp);
475 	}
476 	if (nextp != NULL) {
477 		KASSERT(mutex_owned(&mountlist_lock));
478 		*nextp = CIRCLEQ_NEXT(mp, mnt_list);
479 	}
480 }
481 
482 struct mount *
483 vfs_mountalloc(struct vfsops *vfsops, struct vnode *vp)
484 {
485 	int error;
486 	struct mount *mp;
487 
488 	mp = kmem_zalloc(sizeof(*mp), KM_SLEEP);
489 	if (mp == NULL)
490 		return NULL;
491 
492 	mp->mnt_op = vfsops;
493 	mp->mnt_refcnt = 1;
494 	TAILQ_INIT(&mp->mnt_vnodelist);
495 	rw_init(&mp->mnt_unmounting);
496 	mutex_init(&mp->mnt_renamelock, MUTEX_DEFAULT, IPL_NONE);
497 	mutex_init(&mp->mnt_updating, MUTEX_DEFAULT, IPL_NONE);
498 	error = vfs_busy(mp, NULL);
499 	KASSERT(error == 0);
500 	mp->mnt_vnodecovered = vp;
501 	mount_initspecific(mp);
502 
503 	mutex_enter(&mountgen_lock);
504 	mp->mnt_gen = mountgen++;
505 	mutex_exit(&mountgen_lock);
506 
507 	return mp;
508 }
509 
510 /*
511  * Lookup a filesystem type, and if found allocate and initialize
512  * a mount structure for it.
513  *
514  * Devname is usually updated by mount(8) after booting.
515  */
516 int
517 vfs_rootmountalloc(const char *fstypename, const char *devname,
518     struct mount **mpp)
519 {
520 	struct vfsops *vfsp = NULL;
521 	struct mount *mp;
522 
523 	mutex_enter(&vfs_list_lock);
524 	LIST_FOREACH(vfsp, &vfs_list, vfs_list)
525 		if (!strncmp(vfsp->vfs_name, fstypename,
526 		    sizeof(mp->mnt_stat.f_fstypename)))
527 			break;
528 	if (vfsp == NULL) {
529 		mutex_exit(&vfs_list_lock);
530 		return (ENODEV);
531 	}
532 	vfsp->vfs_refcount++;
533 	mutex_exit(&vfs_list_lock);
534 
535 	if ((mp = vfs_mountalloc(vfsp, NULL)) == NULL)
536 		return ENOMEM;
537 	mp->mnt_flag = MNT_RDONLY;
538 	(void)strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfs_name,
539 	    sizeof(mp->mnt_stat.f_fstypename));
540 	mp->mnt_stat.f_mntonname[0] = '/';
541 	mp->mnt_stat.f_mntonname[1] = '\0';
542 	mp->mnt_stat.f_mntfromname[sizeof(mp->mnt_stat.f_mntfromname) - 1] =
543 	    '\0';
544 	(void)copystr(devname, mp->mnt_stat.f_mntfromname,
545 	    sizeof(mp->mnt_stat.f_mntfromname) - 1, 0);
546 	*mpp = mp;
547 	return (0);
548 }
549 
550 /*
551  * Routines having to do with the management of the vnode table.
552  */
553 extern int (**dead_vnodeop_p)(void *);
554 
555 /*
556  * Return the next vnode from the free list.
557  */
558 int
559 getnewvnode(enum vtagtype tag, struct mount *mp, int (**vops)(void *),
560 	    vnode_t **vpp)
561 {
562 	struct uvm_object *uobj;
563 	static int toggle;
564 	vnode_t *vp;
565 	int error = 0, tryalloc;
566 
567  try_again:
568 	if (mp != NULL) {
569 		/*
570 		 * Mark filesystem busy while we're creating a
571 		 * vnode.  If unmount is in progress, this will
572 		 * fail.
573 		 */
574 		error = vfs_busy(mp, NULL);
575 		if (error)
576 			return error;
577 	}
578 
579 	/*
580 	 * We must choose whether to allocate a new vnode or recycle an
581 	 * existing one. The criterion for allocating a new one is that
582 	 * the total number of vnodes is less than the number desired or
583 	 * there are no vnodes on either free list. Generally we only
584 	 * want to recycle vnodes that have no buffers associated with
585 	 * them, so we look first on the vnode_free_list. If it is empty,
586 	 * we next consider vnodes with referencing buffers on the
587 	 * vnode_hold_list. The toggle ensures that half the time we
588 	 * will use a buffer from the vnode_hold_list, and half the time
589 	 * we will allocate a new one unless the list has grown to twice
590 	 * the desired size. We are reticent to recycle vnodes from the
591 	 * vnode_hold_list because we will lose the identity of all its
592 	 * referencing buffers.
593 	 */
594 
595 	vp = NULL;
596 
597 	mutex_enter(&vnode_free_list_lock);
598 
599 	toggle ^= 1;
600 	if (numvnodes > 2 * desiredvnodes)
601 		toggle = 0;
602 
603 	tryalloc = numvnodes < desiredvnodes ||
604 	    (TAILQ_FIRST(&vnode_free_list) == NULL &&
605 	     (TAILQ_FIRST(&vnode_hold_list) == NULL || toggle));
606 
607 	if (tryalloc) {
608 		numvnodes++;
609 		mutex_exit(&vnode_free_list_lock);
610 		if ((vp = vnalloc(NULL)) == NULL) {
611 			mutex_enter(&vnode_free_list_lock);
612 			numvnodes--;
613 		} else
614 			vp->v_usecount = 1;
615 	}
616 
617 	if (vp == NULL) {
618 		vp = getcleanvnode();
619 		if (vp == NULL) {
620 			if (mp != NULL) {
621 				vfs_unbusy(mp, false, NULL);
622 			}
623 			if (tryalloc) {
624 				printf("WARNING: unable to allocate new "
625 				    "vnode, retrying...\n");
626 				kpause("newvn", false, hz, NULL);
627 				goto try_again;
628 			}
629 			tablefull("vnode", "increase kern.maxvnodes or NVNODE");
630 			*vpp = 0;
631 			return (ENFILE);
632 		}
633 		vp->v_iflag = 0;
634 		vp->v_vflag = 0;
635 		vp->v_uflag = 0;
636 		vp->v_socket = NULL;
637 	}
638 
639 	KASSERT(vp->v_usecount == 1);
640 	KASSERT(vp->v_freelisthd == NULL);
641 	KASSERT(LIST_EMPTY(&vp->v_nclist));
642 	KASSERT(LIST_EMPTY(&vp->v_dnclist));
643 
644 	vp->v_type = VNON;
645 	vp->v_tag = tag;
646 	vp->v_op = vops;
647 	insmntque(vp, mp);
648 	*vpp = vp;
649 	vp->v_data = 0;
650 
651 	/*
652 	 * initialize uvm_object within vnode.
653 	 */
654 
655 	uobj = &vp->v_uobj;
656 	KASSERT(uobj->pgops == &uvm_vnodeops);
657 	KASSERT(uobj->uo_npages == 0);
658 	KASSERT(TAILQ_FIRST(&uobj->memq) == NULL);
659 	vp->v_size = vp->v_writesize = VSIZENOTSET;
660 
661 	if (mp != NULL) {
662 		if ((mp->mnt_iflag & IMNT_MPSAFE) != 0)
663 			vp->v_vflag |= VV_MPSAFE;
664 		vfs_unbusy(mp, true, NULL);
665 	}
666 
667 	return (0);
668 }
669 
670 /*
671  * This is really just the reverse of getnewvnode(). Needed for
672  * VFS_VGET functions who may need to push back a vnode in case
673  * of a locking race.
674  */
675 void
676 ungetnewvnode(vnode_t *vp)
677 {
678 
679 	KASSERT(vp->v_usecount == 1);
680 	KASSERT(vp->v_data == NULL);
681 	KASSERT(vp->v_freelisthd == NULL);
682 
683 	mutex_enter(&vp->v_interlock);
684 	vp->v_iflag |= VI_CLEAN;
685 	vrelel(vp, 0);
686 }
687 
688 /*
689  * Allocate a new, uninitialized vnode.  If 'mp' is non-NULL, this is a
690  * marker vnode and we are prepared to wait for the allocation.
691  */
692 vnode_t *
693 vnalloc(struct mount *mp)
694 {
695 	vnode_t *vp;
696 
697 	vp = pool_cache_get(vnode_cache, (mp != NULL ? PR_WAITOK : PR_NOWAIT));
698 	if (vp == NULL) {
699 		return NULL;
700 	}
701 
702 	memset(vp, 0, sizeof(*vp));
703 	UVM_OBJ_INIT(&vp->v_uobj, &uvm_vnodeops, 0);
704 	cv_init(&vp->v_cv, "vnode");
705 	/*
706 	 * done by memset() above.
707 	 *	LIST_INIT(&vp->v_nclist);
708 	 *	LIST_INIT(&vp->v_dnclist);
709 	 */
710 
711 	if (mp != NULL) {
712 		vp->v_mount = mp;
713 		vp->v_type = VBAD;
714 		vp->v_iflag = VI_MARKER;
715 	} else {
716 		rw_init(&vp->v_lock.vl_lock);
717 	}
718 
719 	return vp;
720 }
721 
722 /*
723  * Free an unused, unreferenced vnode.
724  */
725 void
726 vnfree(vnode_t *vp)
727 {
728 
729 	KASSERT(vp->v_usecount == 0);
730 
731 	if ((vp->v_iflag & VI_MARKER) == 0) {
732 		rw_destroy(&vp->v_lock.vl_lock);
733 		mutex_enter(&vnode_free_list_lock);
734 		numvnodes--;
735 		mutex_exit(&vnode_free_list_lock);
736 	}
737 
738 	UVM_OBJ_DESTROY(&vp->v_uobj);
739 	cv_destroy(&vp->v_cv);
740 	pool_cache_put(vnode_cache, vp);
741 }
742 
743 /*
744  * Remove a vnode from its freelist.
745  */
746 static inline void
747 vremfree(vnode_t *vp)
748 {
749 
750 	KASSERT(mutex_owned(&vp->v_interlock));
751 	KASSERT(vp->v_usecount == 0);
752 
753 	/*
754 	 * Note that the reference count must not change until
755 	 * the vnode is removed.
756 	 */
757 	mutex_enter(&vnode_free_list_lock);
758 	if (vp->v_holdcnt > 0) {
759 		KASSERT(vp->v_freelisthd == &vnode_hold_list);
760 	} else {
761 		KASSERT(vp->v_freelisthd == &vnode_free_list);
762 	}
763 	TAILQ_REMOVE(vp->v_freelisthd, vp, v_freelist);
764 	vp->v_freelisthd = NULL;
765 	mutex_exit(&vnode_free_list_lock);
766 }
767 
768 /*
769  * Move a vnode from one mount queue to another.
770  */
771 static void
772 insmntque(vnode_t *vp, struct mount *mp)
773 {
774 	struct mount *omp;
775 
776 #ifdef DIAGNOSTIC
777 	if ((mp != NULL) &&
778 	    (mp->mnt_iflag & IMNT_UNMOUNT) &&
779 	    vp->v_tag != VT_VFS) {
780 		panic("insmntque into dying filesystem");
781 	}
782 #endif
783 
784 	mutex_enter(&mntvnode_lock);
785 	/*
786 	 * Delete from old mount point vnode list, if on one.
787 	 */
788 	if ((omp = vp->v_mount) != NULL)
789 		TAILQ_REMOVE(&vp->v_mount->mnt_vnodelist, vp, v_mntvnodes);
790 	/*
791 	 * Insert into list of vnodes for the new mount point, if
792 	 * available.  The caller must take a reference on the mount
793 	 * structure and donate to the vnode.
794 	 */
795 	if ((vp->v_mount = mp) != NULL)
796 		TAILQ_INSERT_TAIL(&mp->mnt_vnodelist, vp, v_mntvnodes);
797 	mutex_exit(&mntvnode_lock);
798 
799 	if (omp != NULL) {
800 		/* Release reference to old mount. */
801 		vfs_destroy(omp);
802 	}
803 }
804 
805 /*
806  * Wait for a vnode (typically with VI_XLOCK set) to be cleaned or
807  * recycled.
808  */
809 void
810 vwait(vnode_t *vp, int flags)
811 {
812 
813 	KASSERT(mutex_owned(&vp->v_interlock));
814 	KASSERT(vp->v_usecount != 0);
815 
816 	while ((vp->v_iflag & flags) != 0)
817 		cv_wait(&vp->v_cv, &vp->v_interlock);
818 }
819 
820 /*
821  * Insert a marker vnode into a mount's vnode list, after the
822  * specified vnode.  mntvnode_lock must be held.
823  */
824 void
825 vmark(vnode_t *mvp, vnode_t *vp)
826 {
827 	struct mount *mp;
828 
829 	mp = mvp->v_mount;
830 
831 	KASSERT(mutex_owned(&mntvnode_lock));
832 	KASSERT((mvp->v_iflag & VI_MARKER) != 0);
833 	KASSERT(vp->v_mount == mp);
834 
835 	TAILQ_INSERT_AFTER(&mp->mnt_vnodelist, vp, mvp, v_mntvnodes);
836 }
837 
838 /*
839  * Remove a marker vnode from a mount's vnode list, and return
840  * a pointer to the next vnode in the list.  mntvnode_lock must
841  * be held.
842  */
843 vnode_t *
844 vunmark(vnode_t *mvp)
845 {
846 	vnode_t *vp;
847 	struct mount *mp;
848 
849 	mp = mvp->v_mount;
850 
851 	KASSERT(mutex_owned(&mntvnode_lock));
852 	KASSERT((mvp->v_iflag & VI_MARKER) != 0);
853 
854 	vp = TAILQ_NEXT(mvp, v_mntvnodes);
855 	TAILQ_REMOVE(&mp->mnt_vnodelist, mvp, v_mntvnodes);
856 
857 	KASSERT(vp == NULL || vp->v_mount == mp);
858 
859 	return vp;
860 }
861 
862 /*
863  * Update outstanding I/O count and do wakeup if requested.
864  */
865 void
866 vwakeup(struct buf *bp)
867 {
868 	struct vnode *vp;
869 
870 	if ((vp = bp->b_vp) == NULL)
871 		return;
872 
873 	KASSERT(bp->b_objlock == &vp->v_interlock);
874 	KASSERT(mutex_owned(bp->b_objlock));
875 
876 	if (--vp->v_numoutput < 0)
877 		panic("vwakeup: neg numoutput, vp %p", vp);
878 	if (vp->v_numoutput == 0)
879 		cv_broadcast(&vp->v_cv);
880 }
881 
882 /*
883  * Flush out and invalidate all buffers associated with a vnode.
884  * Called with the underlying vnode locked, which should prevent new dirty
885  * buffers from being queued.
886  */
887 int
888 vinvalbuf(struct vnode *vp, int flags, kauth_cred_t cred, struct lwp *l,
889 	  bool catch, int slptimeo)
890 {
891 	struct buf *bp, *nbp;
892 	int error;
893 	int flushflags = PGO_ALLPAGES | PGO_FREE | PGO_SYNCIO |
894 	    (flags & V_SAVE ? PGO_CLEANIT | PGO_RECLAIM : 0);
895 
896 	/* XXXUBC this doesn't look at flags or slp* */
897 	mutex_enter(&vp->v_interlock);
898 	error = VOP_PUTPAGES(vp, 0, 0, flushflags);
899 	if (error) {
900 		return error;
901 	}
902 
903 	if (flags & V_SAVE) {
904 		error = VOP_FSYNC(vp, cred, FSYNC_WAIT|FSYNC_RECLAIM, 0, 0);
905 		if (error)
906 		        return (error);
907 		KASSERT(LIST_EMPTY(&vp->v_dirtyblkhd));
908 	}
909 
910 	mutex_enter(&bufcache_lock);
911 restart:
912 	for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
913 		nbp = LIST_NEXT(bp, b_vnbufs);
914 		error = bbusy(bp, catch, slptimeo, NULL);
915 		if (error != 0) {
916 			if (error == EPASSTHROUGH)
917 				goto restart;
918 			mutex_exit(&bufcache_lock);
919 			return (error);
920 		}
921 		brelsel(bp, BC_INVAL | BC_VFLUSH);
922 	}
923 
924 	for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
925 		nbp = LIST_NEXT(bp, b_vnbufs);
926 		error = bbusy(bp, catch, slptimeo, NULL);
927 		if (error != 0) {
928 			if (error == EPASSTHROUGH)
929 				goto restart;
930 			mutex_exit(&bufcache_lock);
931 			return (error);
932 		}
933 		/*
934 		 * XXX Since there are no node locks for NFS, I believe
935 		 * there is a slight chance that a delayed write will
936 		 * occur while sleeping just above, so check for it.
937 		 */
938 		if ((bp->b_oflags & BO_DELWRI) && (flags & V_SAVE)) {
939 #ifdef DEBUG
940 			printf("buffer still DELWRI\n");
941 #endif
942 			bp->b_cflags |= BC_BUSY | BC_VFLUSH;
943 			mutex_exit(&bufcache_lock);
944 			VOP_BWRITE(bp);
945 			mutex_enter(&bufcache_lock);
946 			goto restart;
947 		}
948 		brelsel(bp, BC_INVAL | BC_VFLUSH);
949 	}
950 
951 #ifdef DIAGNOSTIC
952 	if (!LIST_EMPTY(&vp->v_cleanblkhd) || !LIST_EMPTY(&vp->v_dirtyblkhd))
953 		panic("vinvalbuf: flush failed, vp %p", vp);
954 #endif
955 
956 	mutex_exit(&bufcache_lock);
957 
958 	return (0);
959 }
960 
961 /*
962  * Destroy any in core blocks past the truncation length.
963  * Called with the underlying vnode locked, which should prevent new dirty
964  * buffers from being queued.
965  */
966 int
967 vtruncbuf(struct vnode *vp, daddr_t lbn, bool catch, int slptimeo)
968 {
969 	struct buf *bp, *nbp;
970 	int error;
971 	voff_t off;
972 
973 	off = round_page((voff_t)lbn << vp->v_mount->mnt_fs_bshift);
974 	mutex_enter(&vp->v_interlock);
975 	error = VOP_PUTPAGES(vp, off, 0, PGO_FREE | PGO_SYNCIO);
976 	if (error) {
977 		return error;
978 	}
979 
980 	mutex_enter(&bufcache_lock);
981 restart:
982 	for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
983 		nbp = LIST_NEXT(bp, b_vnbufs);
984 		if (bp->b_lblkno < lbn)
985 			continue;
986 		error = bbusy(bp, catch, slptimeo, NULL);
987 		if (error != 0) {
988 			if (error == EPASSTHROUGH)
989 				goto restart;
990 			mutex_exit(&bufcache_lock);
991 			return (error);
992 		}
993 		brelsel(bp, BC_INVAL | BC_VFLUSH);
994 	}
995 
996 	for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
997 		nbp = LIST_NEXT(bp, b_vnbufs);
998 		if (bp->b_lblkno < lbn)
999 			continue;
1000 		error = bbusy(bp, catch, slptimeo, NULL);
1001 		if (error != 0) {
1002 			if (error == EPASSTHROUGH)
1003 				goto restart;
1004 			mutex_exit(&bufcache_lock);
1005 			return (error);
1006 		}
1007 		brelsel(bp, BC_INVAL | BC_VFLUSH);
1008 	}
1009 	mutex_exit(&bufcache_lock);
1010 
1011 	return (0);
1012 }
1013 
1014 /*
1015  * Flush all dirty buffers from a vnode.
1016  * Called with the underlying vnode locked, which should prevent new dirty
1017  * buffers from being queued.
1018  */
1019 void
1020 vflushbuf(struct vnode *vp, int sync)
1021 {
1022 	struct buf *bp, *nbp;
1023 	int flags = PGO_CLEANIT | PGO_ALLPAGES | (sync ? PGO_SYNCIO : 0);
1024 	bool dirty;
1025 
1026 	mutex_enter(&vp->v_interlock);
1027 	(void) VOP_PUTPAGES(vp, 0, 0, flags);
1028 
1029 loop:
1030 	mutex_enter(&bufcache_lock);
1031 	for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
1032 		nbp = LIST_NEXT(bp, b_vnbufs);
1033 		if ((bp->b_cflags & BC_BUSY))
1034 			continue;
1035 		if ((bp->b_oflags & BO_DELWRI) == 0)
1036 			panic("vflushbuf: not dirty, bp %p", bp);
1037 		bp->b_cflags |= BC_BUSY | BC_VFLUSH;
1038 		mutex_exit(&bufcache_lock);
1039 		/*
1040 		 * Wait for I/O associated with indirect blocks to complete,
1041 		 * since there is no way to quickly wait for them below.
1042 		 */
1043 		if (bp->b_vp == vp || sync == 0)
1044 			(void) bawrite(bp);
1045 		else
1046 			(void) bwrite(bp);
1047 		goto loop;
1048 	}
1049 	mutex_exit(&bufcache_lock);
1050 
1051 	if (sync == 0)
1052 		return;
1053 
1054 	mutex_enter(&vp->v_interlock);
1055 	while (vp->v_numoutput != 0)
1056 		cv_wait(&vp->v_cv, &vp->v_interlock);
1057 	dirty = !LIST_EMPTY(&vp->v_dirtyblkhd);
1058 	mutex_exit(&vp->v_interlock);
1059 
1060 	if (dirty) {
1061 		vprint("vflushbuf: dirty", vp);
1062 		goto loop;
1063 	}
1064 }
1065 
1066 /*
1067  * Create a vnode for a block device.
1068  * Used for root filesystem and swap areas.
1069  * Also used for memory file system special devices.
1070  */
1071 int
1072 bdevvp(dev_t dev, vnode_t **vpp)
1073 {
1074 
1075 	return (getdevvp(dev, vpp, VBLK));
1076 }
1077 
1078 /*
1079  * Create a vnode for a character device.
1080  * Used for kernfs and some console handling.
1081  */
1082 int
1083 cdevvp(dev_t dev, vnode_t **vpp)
1084 {
1085 
1086 	return (getdevvp(dev, vpp, VCHR));
1087 }
1088 
1089 /*
1090  * Associate a buffer with a vnode.  There must already be a hold on
1091  * the vnode.
1092  */
1093 void
1094 bgetvp(struct vnode *vp, struct buf *bp)
1095 {
1096 
1097 	KASSERT(bp->b_vp == NULL);
1098 	KASSERT(bp->b_objlock == &buffer_lock);
1099 	KASSERT(mutex_owned(&vp->v_interlock));
1100 	KASSERT(mutex_owned(&bufcache_lock));
1101 	KASSERT((bp->b_cflags & BC_BUSY) != 0);
1102 	KASSERT(!cv_has_waiters(&bp->b_done));
1103 
1104 	vholdl(vp);
1105 	bp->b_vp = vp;
1106 	if (vp->v_type == VBLK || vp->v_type == VCHR)
1107 		bp->b_dev = vp->v_rdev;
1108 	else
1109 		bp->b_dev = NODEV;
1110 
1111 	/*
1112 	 * Insert onto list for new vnode.
1113 	 */
1114 	bufinsvn(bp, &vp->v_cleanblkhd);
1115 	bp->b_objlock = &vp->v_interlock;
1116 }
1117 
1118 /*
1119  * Disassociate a buffer from a vnode.
1120  */
1121 void
1122 brelvp(struct buf *bp)
1123 {
1124 	struct vnode *vp = bp->b_vp;
1125 
1126 	KASSERT(vp != NULL);
1127 	KASSERT(bp->b_objlock == &vp->v_interlock);
1128 	KASSERT(mutex_owned(&vp->v_interlock));
1129 	KASSERT(mutex_owned(&bufcache_lock));
1130 	KASSERT((bp->b_cflags & BC_BUSY) != 0);
1131 	KASSERT(!cv_has_waiters(&bp->b_done));
1132 
1133 	/*
1134 	 * Delete from old vnode list, if on one.
1135 	 */
1136 	if (LIST_NEXT(bp, b_vnbufs) != NOLIST)
1137 		bufremvn(bp);
1138 
1139 	if (TAILQ_EMPTY(&vp->v_uobj.memq) && (vp->v_iflag & VI_ONWORKLST) &&
1140 	    LIST_FIRST(&vp->v_dirtyblkhd) == NULL) {
1141 		vp->v_iflag &= ~VI_WRMAPDIRTY;
1142 		vn_syncer_remove_from_worklist(vp);
1143 	}
1144 
1145 	bp->b_objlock = &buffer_lock;
1146 	bp->b_vp = NULL;
1147 	holdrelel(vp);
1148 }
1149 
1150 /*
1151  * Reassign a buffer from one vnode list to another.
1152  * The list reassignment must be within the same vnode.
1153  * Used to assign file specific control information
1154  * (indirect blocks) to the list to which they belong.
1155  */
1156 void
1157 reassignbuf(struct buf *bp, struct vnode *vp)
1158 {
1159 	struct buflists *listheadp;
1160 	int delayx;
1161 
1162 	KASSERT(mutex_owned(&bufcache_lock));
1163 	KASSERT(bp->b_objlock == &vp->v_interlock);
1164 	KASSERT(mutex_owned(&vp->v_interlock));
1165 	KASSERT((bp->b_cflags & BC_BUSY) != 0);
1166 
1167 	/*
1168 	 * Delete from old vnode list, if on one.
1169 	 */
1170 	if (LIST_NEXT(bp, b_vnbufs) != NOLIST)
1171 		bufremvn(bp);
1172 
1173 	/*
1174 	 * If dirty, put on list of dirty buffers;
1175 	 * otherwise insert onto list of clean buffers.
1176 	 */
1177 	if ((bp->b_oflags & BO_DELWRI) == 0) {
1178 		listheadp = &vp->v_cleanblkhd;
1179 		if (TAILQ_EMPTY(&vp->v_uobj.memq) &&
1180 		    (vp->v_iflag & VI_ONWORKLST) &&
1181 		    LIST_FIRST(&vp->v_dirtyblkhd) == NULL) {
1182 			vp->v_iflag &= ~VI_WRMAPDIRTY;
1183 			vn_syncer_remove_from_worklist(vp);
1184 		}
1185 	} else {
1186 		listheadp = &vp->v_dirtyblkhd;
1187 		if ((vp->v_iflag & VI_ONWORKLST) == 0) {
1188 			switch (vp->v_type) {
1189 			case VDIR:
1190 				delayx = dirdelay;
1191 				break;
1192 			case VBLK:
1193 				if (vp->v_specmountpoint != NULL) {
1194 					delayx = metadelay;
1195 					break;
1196 				}
1197 				/* fall through */
1198 			default:
1199 				delayx = filedelay;
1200 				break;
1201 			}
1202 			if (!vp->v_mount ||
1203 			    (vp->v_mount->mnt_flag & MNT_ASYNC) == 0)
1204 				vn_syncer_add_to_worklist(vp, delayx);
1205 		}
1206 	}
1207 	bufinsvn(bp, listheadp);
1208 }
1209 
1210 /*
1211  * Create a vnode for a device.
1212  * Used by bdevvp (block device) for root file system etc.,
1213  * and by cdevvp (character device) for console and kernfs.
1214  */
1215 static int
1216 getdevvp(dev_t dev, vnode_t **vpp, enum vtype type)
1217 {
1218 	vnode_t *vp;
1219 	vnode_t *nvp;
1220 	int error;
1221 
1222 	if (dev == NODEV) {
1223 		*vpp = NULL;
1224 		return (0);
1225 	}
1226 	error = getnewvnode(VT_NON, NULL, spec_vnodeop_p, &nvp);
1227 	if (error) {
1228 		*vpp = NULL;
1229 		return (error);
1230 	}
1231 	vp = nvp;
1232 	vp->v_type = type;
1233 	vp->v_vflag |= VV_MPSAFE;
1234 	uvm_vnp_setsize(vp, 0);
1235 	spec_node_init(vp, dev);
1236 	*vpp = vp;
1237 	return (0);
1238 }
1239 
1240 /*
1241  * Try to gain a reference to a vnode, without acquiring its interlock.
1242  * The caller must hold a lock that will prevent the vnode from being
1243  * recycled or freed.
1244  */
1245 bool
1246 vtryget(vnode_t *vp)
1247 {
1248 	u_int use, next;
1249 
1250 	/*
1251 	 * If the vnode is being freed, don't make life any harder
1252 	 * for vclean() by adding another reference without waiting.
1253 	 * This is not strictly necessary, but we'll do it anyway.
1254 	 */
1255 	if (__predict_false((vp->v_iflag & (VI_XLOCK | VI_FREEING)) != 0)) {
1256 		return false;
1257 	}
1258 	for (use = vp->v_usecount;; use = next) {
1259 		if (use == 0 || __predict_false((use & VC_XLOCK) != 0)) {
1260 			/* Need interlock held if first reference. */
1261 			return false;
1262 		}
1263 		next = atomic_cas_uint(&vp->v_usecount, use, use + 1);
1264 		if (__predict_true(next == use)) {
1265 			return true;
1266 		}
1267 	}
1268 }
1269 
1270 /*
1271  * Grab a particular vnode from the free list, increment its
1272  * reference count and lock it. If the vnode lock bit is set the
1273  * vnode is being eliminated in vgone. In that case, we can not
1274  * grab the vnode, so the process is awakened when the transition is
1275  * completed, and an error returned to indicate that the vnode is no
1276  * longer usable (possibly having been changed to a new file system type).
1277  */
1278 int
1279 vget(vnode_t *vp, int flags)
1280 {
1281 	int error = 0;
1282 
1283 	KASSERT((vp->v_iflag & VI_MARKER) == 0);
1284 
1285 	if ((flags & LK_INTERLOCK) == 0)
1286 		mutex_enter(&vp->v_interlock);
1287 
1288 	/*
1289 	 * Before adding a reference, we must remove the vnode
1290 	 * from its freelist.
1291 	 */
1292 	if (vp->v_usecount == 0) {
1293 		vremfree(vp);
1294 		vp->v_usecount = 1;
1295 	} else {
1296 		atomic_inc_uint(&vp->v_usecount);
1297 	}
1298 
1299 	/*
1300 	 * If the vnode is in the process of being cleaned out for
1301 	 * another use, we wait for the cleaning to finish and then
1302 	 * return failure.  Cleaning is determined by checking if
1303 	 * the VI_XLOCK or VI_FREEING flags are set.
1304 	 */
1305 	if ((vp->v_iflag & (VI_XLOCK | VI_FREEING)) != 0) {
1306 		if ((flags & LK_NOWAIT) != 0) {
1307 			vrelel(vp, 0);
1308 			return EBUSY;
1309 		}
1310 		vwait(vp, VI_XLOCK | VI_FREEING);
1311 		vrelel(vp, 0);
1312 		return ENOENT;
1313 	}
1314 
1315 	if ((vp->v_iflag & VI_INACTNOW) != 0) {
1316 		/*
1317 		 * if it's being desactived, wait for it to complete.
1318 		 * Make sure to not return a clean vnode.
1319 		 */
1320 		 if ((flags & LK_NOWAIT) != 0) {
1321 			vrelel(vp, 0);
1322 			return EBUSY;
1323 		}
1324 		vwait(vp, VI_INACTNOW);
1325 		if ((vp->v_iflag & VI_CLEAN) != 0) {
1326 			vrelel(vp, 0);
1327 			return ENOENT;
1328 		}
1329 	}
1330 
1331 	/*
1332 	 * Ok, we got it in good shape.  Just locking left.
1333 	 */
1334 	KASSERT((vp->v_iflag & VI_CLEAN) == 0);
1335 	if (flags & LK_TYPE_MASK) {
1336 		error = vn_lock(vp, flags | LK_INTERLOCK);
1337 		if (error != 0) {
1338 			vrele(vp);
1339 		}
1340 	} else {
1341 		mutex_exit(&vp->v_interlock);
1342 	}
1343 	return error;
1344 }
1345 
1346 /*
1347  * vput(), just unlock and vrele()
1348  */
1349 void
1350 vput(vnode_t *vp)
1351 {
1352 
1353 	KASSERT((vp->v_iflag & VI_MARKER) == 0);
1354 
1355 	VOP_UNLOCK(vp);
1356 	vrele(vp);
1357 }
1358 
1359 /*
1360  * Try to drop reference on a vnode.  Abort if we are releasing the
1361  * last reference.  Note: this _must_ succeed if not the last reference.
1362  */
1363 static inline bool
1364 vtryrele(vnode_t *vp)
1365 {
1366 	u_int use, next;
1367 
1368 	for (use = vp->v_usecount;; use = next) {
1369 		if (use == 1) {
1370 			return false;
1371 		}
1372 		KASSERT((use & VC_MASK) > 1);
1373 		next = atomic_cas_uint(&vp->v_usecount, use, use - 1);
1374 		if (__predict_true(next == use)) {
1375 			return true;
1376 		}
1377 	}
1378 }
1379 
1380 /*
1381  * Vnode release.  If reference count drops to zero, call inactive
1382  * routine and either return to freelist or free to the pool.
1383  */
1384 void
1385 vrelel(vnode_t *vp, int flags)
1386 {
1387 	bool recycle, defer;
1388 	int error;
1389 
1390 	KASSERT(mutex_owned(&vp->v_interlock));
1391 	KASSERT((vp->v_iflag & VI_MARKER) == 0);
1392 	KASSERT(vp->v_freelisthd == NULL);
1393 
1394 	if (__predict_false(vp->v_op == dead_vnodeop_p &&
1395 	    (vp->v_iflag & (VI_CLEAN|VI_XLOCK)) == 0)) {
1396 		vpanic(vp, "dead but not clean");
1397 	}
1398 
1399 	/*
1400 	 * If not the last reference, just drop the reference count
1401 	 * and unlock.
1402 	 */
1403 	if (vtryrele(vp)) {
1404 		vp->v_iflag |= VI_INACTREDO;
1405 		mutex_exit(&vp->v_interlock);
1406 		return;
1407 	}
1408 	if (vp->v_usecount <= 0 || vp->v_writecount != 0) {
1409 		vpanic(vp, "vrelel: bad ref count");
1410 	}
1411 
1412 	KASSERT((vp->v_iflag & VI_XLOCK) == 0);
1413 
1414 	/*
1415 	 * If not clean, deactivate the vnode, but preserve
1416 	 * our reference across the call to VOP_INACTIVE().
1417 	 */
1418  retry:
1419 	if ((vp->v_iflag & VI_CLEAN) == 0) {
1420 		recycle = false;
1421 		vp->v_iflag |= VI_INACTNOW;
1422 
1423 		/*
1424 		 * XXX This ugly block can be largely eliminated if
1425 		 * locking is pushed down into the file systems.
1426 		 *
1427 		 * Defer vnode release to vrele_thread if caller
1428 		 * requests it explicitly.
1429 		 */
1430 		if ((curlwp == uvm.pagedaemon_lwp) ||
1431 		    (flags & VRELEL_ASYNC_RELE) != 0) {
1432 			/* The pagedaemon can't wait around; defer. */
1433 			defer = true;
1434 		} else if (curlwp == vrele_lwp) {
1435 			/*
1436 			 * We have to try harder. But we can't sleep
1437 			 * with VI_INACTNOW as vget() may be waiting on it.
1438 			 */
1439 			vp->v_iflag &= ~(VI_INACTREDO|VI_INACTNOW);
1440 			cv_broadcast(&vp->v_cv);
1441 			error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK |
1442 			    LK_RETRY);
1443 			if (error != 0) {
1444 				/* XXX */
1445 				vpanic(vp, "vrele: unable to lock %p");
1446 			}
1447 			mutex_enter(&vp->v_interlock);
1448 			/*
1449 			 * if we did get another reference while
1450 			 * sleeping, don't try to inactivate it yet.
1451 			 */
1452 			if (__predict_false(vtryrele(vp))) {
1453 				VOP_UNLOCK(vp);
1454 				mutex_exit(&vp->v_interlock);
1455 				return;
1456 			}
1457 			vp->v_iflag |= VI_INACTNOW;
1458 			mutex_exit(&vp->v_interlock);
1459 			defer = false;
1460 		} else if ((vp->v_iflag & VI_LAYER) != 0) {
1461 			/*
1462 			 * Acquiring the stack's lock in vclean() even
1463 			 * for an honest vput/vrele is dangerous because
1464 			 * our caller may hold other vnode locks; defer.
1465 			 */
1466 			defer = true;
1467 		} else {
1468 			/* If we can't acquire the lock, then defer. */
1469 			vp->v_iflag &= ~VI_INACTREDO;
1470 			error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK |
1471 			    LK_NOWAIT);
1472 			if (error != 0) {
1473 				defer = true;
1474 				mutex_enter(&vp->v_interlock);
1475 			} else {
1476 				defer = false;
1477 			}
1478 		}
1479 
1480 		if (defer) {
1481 			/*
1482 			 * Defer reclaim to the kthread; it's not safe to
1483 			 * clean it here.  We donate it our last reference.
1484 			 */
1485 			KASSERT(mutex_owned(&vp->v_interlock));
1486 			KASSERT((vp->v_iflag & VI_INACTPEND) == 0);
1487 			vp->v_iflag &= ~VI_INACTNOW;
1488 			vp->v_iflag |= VI_INACTPEND;
1489 			mutex_enter(&vrele_lock);
1490 			TAILQ_INSERT_TAIL(&vrele_list, vp, v_freelist);
1491 			if (++vrele_pending > (desiredvnodes >> 8))
1492 				cv_signal(&vrele_cv);
1493 			mutex_exit(&vrele_lock);
1494 			cv_broadcast(&vp->v_cv);
1495 			mutex_exit(&vp->v_interlock);
1496 			return;
1497 		}
1498 
1499 #ifdef DIAGNOSTIC
1500 		if ((vp->v_type == VBLK || vp->v_type == VCHR) &&
1501 		    vp->v_specnode != NULL && vp->v_specnode->sn_opencnt != 0) {
1502 			vprint("vrelel: missing VOP_CLOSE()", vp);
1503 		}
1504 #endif
1505 
1506 		/*
1507 		 * The vnode can gain another reference while being
1508 		 * deactivated.  If VOP_INACTIVE() indicates that
1509 		 * the described file has been deleted, then recycle
1510 		 * the vnode irrespective of additional references.
1511 		 * Another thread may be waiting to re-use the on-disk
1512 		 * inode.
1513 		 *
1514 		 * Note that VOP_INACTIVE() will drop the vnode lock.
1515 		 */
1516 		VOP_INACTIVE(vp, &recycle);
1517 		mutex_enter(&vp->v_interlock);
1518 		vp->v_iflag &= ~VI_INACTNOW;
1519 		cv_broadcast(&vp->v_cv);
1520 		if (!recycle) {
1521 			if (vtryrele(vp)) {
1522 				mutex_exit(&vp->v_interlock);
1523 				return;
1524 			}
1525 
1526 			/*
1527 			 * If we grew another reference while
1528 			 * VOP_INACTIVE() was underway, retry.
1529 			 */
1530 			if ((vp->v_iflag & VI_INACTREDO) != 0) {
1531 				goto retry;
1532 			}
1533 		}
1534 
1535 		/* Take care of space accounting. */
1536 		if (vp->v_iflag & VI_EXECMAP) {
1537 			atomic_add_int(&uvmexp.execpages,
1538 			    -vp->v_uobj.uo_npages);
1539 			atomic_add_int(&uvmexp.filepages,
1540 			    vp->v_uobj.uo_npages);
1541 		}
1542 		vp->v_iflag &= ~(VI_TEXT|VI_EXECMAP|VI_WRMAP);
1543 		vp->v_vflag &= ~VV_MAPPED;
1544 
1545 		/*
1546 		 * Recycle the vnode if the file is now unused (unlinked),
1547 		 * otherwise just free it.
1548 		 */
1549 		if (recycle) {
1550 			vclean(vp, DOCLOSE);
1551 		}
1552 		KASSERT(vp->v_usecount > 0);
1553 	}
1554 
1555 	if (atomic_dec_uint_nv(&vp->v_usecount) != 0) {
1556 		/* Gained another reference while being reclaimed. */
1557 		mutex_exit(&vp->v_interlock);
1558 		return;
1559 	}
1560 
1561 	if ((vp->v_iflag & VI_CLEAN) != 0) {
1562 		/*
1563 		 * It's clean so destroy it.  It isn't referenced
1564 		 * anywhere since it has been reclaimed.
1565 		 */
1566 		KASSERT(vp->v_holdcnt == 0);
1567 		KASSERT(vp->v_writecount == 0);
1568 		mutex_exit(&vp->v_interlock);
1569 		insmntque(vp, NULL);
1570 		if (vp->v_type == VBLK || vp->v_type == VCHR) {
1571 			spec_node_destroy(vp);
1572 		}
1573 		vnfree(vp);
1574 	} else {
1575 		/*
1576 		 * Otherwise, put it back onto the freelist.  It
1577 		 * can't be destroyed while still associated with
1578 		 * a file system.
1579 		 */
1580 		mutex_enter(&vnode_free_list_lock);
1581 		if (vp->v_holdcnt > 0) {
1582 			vp->v_freelisthd = &vnode_hold_list;
1583 		} else {
1584 			vp->v_freelisthd = &vnode_free_list;
1585 		}
1586 		TAILQ_INSERT_TAIL(vp->v_freelisthd, vp, v_freelist);
1587 		mutex_exit(&vnode_free_list_lock);
1588 		mutex_exit(&vp->v_interlock);
1589 	}
1590 }
1591 
1592 void
1593 vrele(vnode_t *vp)
1594 {
1595 
1596 	KASSERT((vp->v_iflag & VI_MARKER) == 0);
1597 
1598 	if ((vp->v_iflag & VI_INACTNOW) == 0 && vtryrele(vp)) {
1599 		return;
1600 	}
1601 	mutex_enter(&vp->v_interlock);
1602 	vrelel(vp, 0);
1603 }
1604 
1605 /*
1606  * Asynchronous vnode release, vnode is released in different context.
1607  */
1608 void
1609 vrele_async(vnode_t *vp)
1610 {
1611 
1612 	KASSERT((vp->v_iflag & VI_MARKER) == 0);
1613 
1614 	if ((vp->v_iflag & VI_INACTNOW) == 0 && vtryrele(vp)) {
1615 		return;
1616 	}
1617 
1618 	mutex_enter(&vp->v_interlock);
1619 	vrelel(vp, VRELEL_ASYNC_RELE);
1620 }
1621 
1622 static void
1623 vrele_thread(void *cookie)
1624 {
1625 	vnode_t *vp;
1626 
1627 	for (;;) {
1628 		mutex_enter(&vrele_lock);
1629 		while (TAILQ_EMPTY(&vrele_list)) {
1630 			vrele_gen++;
1631 			cv_broadcast(&vrele_cv);
1632 			cv_timedwait(&vrele_cv, &vrele_lock, hz);
1633 		}
1634 		vp = TAILQ_FIRST(&vrele_list);
1635 		TAILQ_REMOVE(&vrele_list, vp, v_freelist);
1636 		vrele_pending--;
1637 		mutex_exit(&vrele_lock);
1638 
1639 		/*
1640 		 * If not the last reference, then ignore the vnode
1641 		 * and look for more work.
1642 		 */
1643 		mutex_enter(&vp->v_interlock);
1644 		KASSERT((vp->v_iflag & VI_INACTPEND) != 0);
1645 		vp->v_iflag &= ~VI_INACTPEND;
1646 		vrelel(vp, 0);
1647 	}
1648 }
1649 
1650 /*
1651  * Page or buffer structure gets a reference.
1652  * Called with v_interlock held.
1653  */
1654 void
1655 vholdl(vnode_t *vp)
1656 {
1657 
1658 	KASSERT(mutex_owned(&vp->v_interlock));
1659 	KASSERT((vp->v_iflag & VI_MARKER) == 0);
1660 
1661 	if (vp->v_holdcnt++ == 0 && vp->v_usecount == 0) {
1662 		mutex_enter(&vnode_free_list_lock);
1663 		KASSERT(vp->v_freelisthd == &vnode_free_list);
1664 		TAILQ_REMOVE(vp->v_freelisthd, vp, v_freelist);
1665 		vp->v_freelisthd = &vnode_hold_list;
1666 		TAILQ_INSERT_TAIL(vp->v_freelisthd, vp, v_freelist);
1667 		mutex_exit(&vnode_free_list_lock);
1668 	}
1669 }
1670 
1671 /*
1672  * Page or buffer structure frees a reference.
1673  * Called with v_interlock held.
1674  */
1675 void
1676 holdrelel(vnode_t *vp)
1677 {
1678 
1679 	KASSERT(mutex_owned(&vp->v_interlock));
1680 	KASSERT((vp->v_iflag & VI_MARKER) == 0);
1681 
1682 	if (vp->v_holdcnt <= 0) {
1683 		vpanic(vp, "holdrelel: holdcnt vp %p");
1684 	}
1685 
1686 	vp->v_holdcnt--;
1687 	if (vp->v_holdcnt == 0 && vp->v_usecount == 0) {
1688 		mutex_enter(&vnode_free_list_lock);
1689 		KASSERT(vp->v_freelisthd == &vnode_hold_list);
1690 		TAILQ_REMOVE(vp->v_freelisthd, vp, v_freelist);
1691 		vp->v_freelisthd = &vnode_free_list;
1692 		TAILQ_INSERT_TAIL(vp->v_freelisthd, vp, v_freelist);
1693 		mutex_exit(&vnode_free_list_lock);
1694 	}
1695 }
1696 
1697 /*
1698  * Vnode reference, where a reference is already held by some other
1699  * object (for example, a file structure).
1700  */
1701 void
1702 vref(vnode_t *vp)
1703 {
1704 
1705 	KASSERT((vp->v_iflag & VI_MARKER) == 0);
1706 	KASSERT(vp->v_usecount != 0);
1707 
1708 	atomic_inc_uint(&vp->v_usecount);
1709 }
1710 
1711 /*
1712  * Remove any vnodes in the vnode table belonging to mount point mp.
1713  *
1714  * If FORCECLOSE is not specified, there should not be any active ones,
1715  * return error if any are found (nb: this is a user error, not a
1716  * system error). If FORCECLOSE is specified, detach any active vnodes
1717  * that are found.
1718  *
1719  * If WRITECLOSE is set, only flush out regular file vnodes open for
1720  * writing.
1721  *
1722  * SKIPSYSTEM causes any vnodes marked V_SYSTEM to be skipped.
1723  */
1724 #ifdef DEBUG
1725 int busyprt = 0;	/* print out busy vnodes */
1726 struct ctldebug debug1 = { "busyprt", &busyprt };
1727 #endif
1728 
1729 static vnode_t *
1730 vflushnext(vnode_t *mvp, int *when)
1731 {
1732 
1733 	if (hardclock_ticks > *when) {
1734 		mutex_exit(&mntvnode_lock);
1735 		yield();
1736 		mutex_enter(&mntvnode_lock);
1737 		*when = hardclock_ticks + hz / 10;
1738 	}
1739 
1740 	return vunmark(mvp);
1741 }
1742 
1743 int
1744 vflush(struct mount *mp, vnode_t *skipvp, int flags)
1745 {
1746 	vnode_t *vp, *mvp;
1747 	int busy = 0, when = 0, gen;
1748 
1749 	/*
1750 	 * First, flush out any vnode references from vrele_list.
1751 	 */
1752 	mutex_enter(&vrele_lock);
1753 	gen = vrele_gen;
1754 	while (vrele_pending && gen == vrele_gen) {
1755 		cv_broadcast(&vrele_cv);
1756 		cv_wait(&vrele_cv, &vrele_lock);
1757 	}
1758 	mutex_exit(&vrele_lock);
1759 
1760 	/* Allocate a marker vnode. */
1761 	if ((mvp = vnalloc(mp)) == NULL)
1762 		return (ENOMEM);
1763 
1764 	/*
1765 	 * NOTE: not using the TAILQ_FOREACH here since in this loop vgone()
1766 	 * and vclean() are called
1767 	 */
1768 	mutex_enter(&mntvnode_lock);
1769 	for (vp = TAILQ_FIRST(&mp->mnt_vnodelist); vp != NULL;
1770 	    vp = vflushnext(mvp, &when)) {
1771 		vmark(mvp, vp);
1772 		if (vp->v_mount != mp || vismarker(vp))
1773 			continue;
1774 		/*
1775 		 * Skip over a selected vnode.
1776 		 */
1777 		if (vp == skipvp)
1778 			continue;
1779 		mutex_enter(&vp->v_interlock);
1780 		/*
1781 		 * Ignore clean but still referenced vnodes.
1782 		 */
1783 		if ((vp->v_iflag & VI_CLEAN) != 0) {
1784 			mutex_exit(&vp->v_interlock);
1785 			continue;
1786 		}
1787 		/*
1788 		 * Skip over a vnodes marked VSYSTEM.
1789 		 */
1790 		if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
1791 			mutex_exit(&vp->v_interlock);
1792 			continue;
1793 		}
1794 		/*
1795 		 * If WRITECLOSE is set, only flush out regular file
1796 		 * vnodes open for writing.
1797 		 */
1798 		if ((flags & WRITECLOSE) &&
1799 		    (vp->v_writecount == 0 || vp->v_type != VREG)) {
1800 			mutex_exit(&vp->v_interlock);
1801 			continue;
1802 		}
1803 		/*
1804 		 * With v_usecount == 0, all we need to do is clear
1805 		 * out the vnode data structures and we are done.
1806 		 */
1807 		if (vp->v_usecount == 0) {
1808 			mutex_exit(&mntvnode_lock);
1809 			vremfree(vp);
1810 			vp->v_usecount = 1;
1811 			vclean(vp, DOCLOSE);
1812 			vrelel(vp, 0);
1813 			mutex_enter(&mntvnode_lock);
1814 			continue;
1815 		}
1816 		/*
1817 		 * If FORCECLOSE is set, forcibly close the vnode.
1818 		 * For block or character devices, revert to an
1819 		 * anonymous device.  For all other files, just
1820 		 * kill them.
1821 		 */
1822 		if (flags & FORCECLOSE) {
1823 			mutex_exit(&mntvnode_lock);
1824 			atomic_inc_uint(&vp->v_usecount);
1825 			if (vp->v_type != VBLK && vp->v_type != VCHR) {
1826 				vclean(vp, DOCLOSE);
1827 				vrelel(vp, 0);
1828 			} else {
1829 				vclean(vp, 0);
1830 				vp->v_op = spec_vnodeop_p; /* XXXSMP */
1831 				mutex_exit(&vp->v_interlock);
1832 				/*
1833 				 * The vnode isn't clean, but still resides
1834 				 * on the mount list.  Remove it. XXX This
1835 				 * is a bit dodgy.
1836 				 */
1837 				insmntque(vp, NULL);
1838 				vrele(vp);
1839 			}
1840 			mutex_enter(&mntvnode_lock);
1841 			continue;
1842 		}
1843 #ifdef DEBUG
1844 		if (busyprt)
1845 			vprint("vflush: busy vnode", vp);
1846 #endif
1847 		mutex_exit(&vp->v_interlock);
1848 		busy++;
1849 	}
1850 	mutex_exit(&mntvnode_lock);
1851 	vnfree(mvp);
1852 	if (busy)
1853 		return (EBUSY);
1854 	return (0);
1855 }
1856 
1857 /*
1858  * Disassociate the underlying file system from a vnode.
1859  *
1860  * Must be called with the interlock held, and will return with it held.
1861  */
1862 void
1863 vclean(vnode_t *vp, int flags)
1864 {
1865 	lwp_t *l = curlwp;
1866 	bool recycle, active;
1867 	int error;
1868 
1869 	KASSERT(mutex_owned(&vp->v_interlock));
1870 	KASSERT((vp->v_iflag & VI_MARKER) == 0);
1871 	KASSERT(vp->v_usecount != 0);
1872 
1873 	/* If cleaning is already in progress wait until done and return. */
1874 	if (vp->v_iflag & VI_XLOCK) {
1875 		vwait(vp, VI_XLOCK);
1876 		return;
1877 	}
1878 
1879 	/* If already clean, nothing to do. */
1880 	if ((vp->v_iflag & VI_CLEAN) != 0) {
1881 		return;
1882 	}
1883 
1884 	/*
1885 	 * Prevent the vnode from being recycled or brought into use
1886 	 * while we clean it out.
1887 	 */
1888 	vp->v_iflag |= VI_XLOCK;
1889 	if (vp->v_iflag & VI_EXECMAP) {
1890 		atomic_add_int(&uvmexp.execpages, -vp->v_uobj.uo_npages);
1891 		atomic_add_int(&uvmexp.filepages, vp->v_uobj.uo_npages);
1892 	}
1893 	vp->v_iflag &= ~(VI_TEXT|VI_EXECMAP);
1894 	active = (vp->v_usecount > 1);
1895 
1896 	/* XXXAD should not lock vnode under layer */
1897 	mutex_exit(&vp->v_interlock);
1898 	VOP_LOCK(vp, LK_EXCLUSIVE);
1899 
1900 	/*
1901 	 * Clean out any cached data associated with the vnode.
1902 	 * If purging an active vnode, it must be closed and
1903 	 * deactivated before being reclaimed. Note that the
1904 	 * VOP_INACTIVE will unlock the vnode.
1905 	 */
1906 	if (flags & DOCLOSE) {
1907 		error = vinvalbuf(vp, V_SAVE, NOCRED, l, 0, 0);
1908 		if (error != 0) {
1909 			/* XXX, fix vn_start_write's grab of mp and use that. */
1910 
1911 			if (wapbl_vphaswapbl(vp))
1912 				WAPBL_DISCARD(wapbl_vptomp(vp));
1913 			error = vinvalbuf(vp, 0, NOCRED, l, 0, 0);
1914 		}
1915 		KASSERT(error == 0);
1916 		KASSERT((vp->v_iflag & VI_ONWORKLST) == 0);
1917 		if (active && (vp->v_type == VBLK || vp->v_type == VCHR)) {
1918 			 spec_node_revoke(vp);
1919 		}
1920 	}
1921 	if (active) {
1922 		VOP_INACTIVE(vp, &recycle);
1923 	} else {
1924 		/*
1925 		 * Any other processes trying to obtain this lock must first
1926 		 * wait for VI_XLOCK to clear, then call the new lock operation.
1927 		 */
1928 		VOP_UNLOCK(vp);
1929 	}
1930 
1931 	/* Disassociate the underlying file system from the vnode. */
1932 	if (VOP_RECLAIM(vp)) {
1933 		vpanic(vp, "vclean: cannot reclaim");
1934 	}
1935 
1936 	KASSERT(vp->v_uobj.uo_npages == 0);
1937 	if (vp->v_type == VREG && vp->v_ractx != NULL) {
1938 		uvm_ra_freectx(vp->v_ractx);
1939 		vp->v_ractx = NULL;
1940 	}
1941 	cache_purge(vp);
1942 
1943 	/* Done with purge, notify sleepers of the grim news. */
1944 	mutex_enter(&vp->v_interlock);
1945 	vp->v_op = dead_vnodeop_p;
1946 	vp->v_tag = VT_NON;
1947 	KNOTE(&vp->v_klist, NOTE_REVOKE);
1948 	vp->v_iflag &= ~(VI_XLOCK | VI_FREEING);
1949 	vp->v_vflag &= ~VV_LOCKSWORK;
1950 	if ((flags & DOCLOSE) != 0) {
1951 		vp->v_iflag |= VI_CLEAN;
1952 	}
1953 	cv_broadcast(&vp->v_cv);
1954 
1955 	KASSERT((vp->v_iflag & VI_ONWORKLST) == 0);
1956 }
1957 
1958 /*
1959  * Recycle an unused vnode to the front of the free list.
1960  * Release the passed interlock if the vnode will be recycled.
1961  */
1962 int
1963 vrecycle(vnode_t *vp, kmutex_t *inter_lkp, struct lwp *l)
1964 {
1965 
1966 	KASSERT((vp->v_iflag & VI_MARKER) == 0);
1967 
1968 	mutex_enter(&vp->v_interlock);
1969 	if (vp->v_usecount != 0) {
1970 		mutex_exit(&vp->v_interlock);
1971 		return (0);
1972 	}
1973 	if (inter_lkp)
1974 		mutex_exit(inter_lkp);
1975 	vremfree(vp);
1976 	vp->v_usecount = 1;
1977 	vclean(vp, DOCLOSE);
1978 	vrelel(vp, 0);
1979 	return (1);
1980 }
1981 
1982 /*
1983  * Eliminate all activity associated with a vnode in preparation for
1984  * reuse.  Drops a reference from the vnode.
1985  */
1986 void
1987 vgone(vnode_t *vp)
1988 {
1989 
1990 	mutex_enter(&vp->v_interlock);
1991 	vclean(vp, DOCLOSE);
1992 	vrelel(vp, 0);
1993 }
1994 
1995 /*
1996  * Lookup a vnode by device number.
1997  */
1998 int
1999 vfinddev(dev_t dev, enum vtype type, vnode_t **vpp)
2000 {
2001 	vnode_t *vp;
2002 	int rc = 0;
2003 
2004 	mutex_enter(&device_lock);
2005 	for (vp = specfs_hash[SPECHASH(dev)]; vp; vp = vp->v_specnext) {
2006 		if (dev != vp->v_rdev || type != vp->v_type)
2007 			continue;
2008 		*vpp = vp;
2009 		rc = 1;
2010 		break;
2011 	}
2012 	mutex_exit(&device_lock);
2013 	return (rc);
2014 }
2015 
2016 /*
2017  * Revoke all the vnodes corresponding to the specified minor number
2018  * range (endpoints inclusive) of the specified major.
2019  */
2020 void
2021 vdevgone(int maj, int minl, int minh, enum vtype type)
2022 {
2023 	vnode_t *vp, **vpp;
2024 	dev_t dev;
2025 	int mn;
2026 
2027 	vp = NULL;	/* XXX gcc */
2028 
2029 	mutex_enter(&device_lock);
2030 	for (mn = minl; mn <= minh; mn++) {
2031 		dev = makedev(maj, mn);
2032 		vpp = &specfs_hash[SPECHASH(dev)];
2033 		for (vp = *vpp; vp != NULL;) {
2034 			mutex_enter(&vp->v_interlock);
2035 			if ((vp->v_iflag & VI_CLEAN) != 0 ||
2036 			    dev != vp->v_rdev || type != vp->v_type) {
2037 				mutex_exit(&vp->v_interlock);
2038 				vp = vp->v_specnext;
2039 				continue;
2040 			}
2041 			mutex_exit(&device_lock);
2042 			if (vget(vp, LK_INTERLOCK) == 0) {
2043 				VOP_REVOKE(vp, REVOKEALL);
2044 				vrele(vp);
2045 			}
2046 			mutex_enter(&device_lock);
2047 			vp = *vpp;
2048 		}
2049 	}
2050 	mutex_exit(&device_lock);
2051 }
2052 
2053 /*
2054  * Eliminate all activity associated with the requested vnode
2055  * and with all vnodes aliased to the requested vnode.
2056  */
2057 void
2058 vrevoke(vnode_t *vp)
2059 {
2060 	vnode_t *vq, **vpp;
2061 	enum vtype type;
2062 	dev_t dev;
2063 
2064 	KASSERT(vp->v_usecount > 0);
2065 
2066 	mutex_enter(&vp->v_interlock);
2067 	if ((vp->v_iflag & VI_CLEAN) != 0) {
2068 		mutex_exit(&vp->v_interlock);
2069 		return;
2070 	} else if (vp->v_type != VBLK && vp->v_type != VCHR) {
2071 		atomic_inc_uint(&vp->v_usecount);
2072 		vclean(vp, DOCLOSE);
2073 		vrelel(vp, 0);
2074 		return;
2075 	} else {
2076 		dev = vp->v_rdev;
2077 		type = vp->v_type;
2078 		mutex_exit(&vp->v_interlock);
2079 	}
2080 
2081 	vpp = &specfs_hash[SPECHASH(dev)];
2082 	mutex_enter(&device_lock);
2083 	for (vq = *vpp; vq != NULL;) {
2084 		/* If clean or being cleaned, then ignore it. */
2085 		mutex_enter(&vq->v_interlock);
2086 		if ((vq->v_iflag & (VI_CLEAN | VI_XLOCK)) != 0 ||
2087 		    vq->v_rdev != dev || vq->v_type != type) {
2088 			mutex_exit(&vq->v_interlock);
2089 			vq = vq->v_specnext;
2090 			continue;
2091 		}
2092 		mutex_exit(&device_lock);
2093 		if (vq->v_usecount == 0) {
2094 			vremfree(vq);
2095 			vq->v_usecount = 1;
2096 		} else {
2097 			atomic_inc_uint(&vq->v_usecount);
2098 		}
2099 		vclean(vq, DOCLOSE);
2100 		vrelel(vq, 0);
2101 		mutex_enter(&device_lock);
2102 		vq = *vpp;
2103 	}
2104 	mutex_exit(&device_lock);
2105 }
2106 
2107 /*
2108  * sysctl helper routine to return list of supported fstypes
2109  */
2110 int
2111 sysctl_vfs_generic_fstypes(SYSCTLFN_ARGS)
2112 {
2113 	char bf[sizeof(((struct statvfs *)NULL)->f_fstypename)];
2114 	char *where = oldp;
2115 	struct vfsops *v;
2116 	size_t needed, left, slen;
2117 	int error, first;
2118 
2119 	if (newp != NULL)
2120 		return (EPERM);
2121 	if (namelen != 0)
2122 		return (EINVAL);
2123 
2124 	first = 1;
2125 	error = 0;
2126 	needed = 0;
2127 	left = *oldlenp;
2128 
2129 	sysctl_unlock();
2130 	mutex_enter(&vfs_list_lock);
2131 	LIST_FOREACH(v, &vfs_list, vfs_list) {
2132 		if (where == NULL)
2133 			needed += strlen(v->vfs_name) + 1;
2134 		else {
2135 			memset(bf, 0, sizeof(bf));
2136 			if (first) {
2137 				strncpy(bf, v->vfs_name, sizeof(bf));
2138 				first = 0;
2139 			} else {
2140 				bf[0] = ' ';
2141 				strncpy(bf + 1, v->vfs_name, sizeof(bf) - 1);
2142 			}
2143 			bf[sizeof(bf)-1] = '\0';
2144 			slen = strlen(bf);
2145 			if (left < slen + 1)
2146 				break;
2147 			v->vfs_refcount++;
2148 			mutex_exit(&vfs_list_lock);
2149 			/* +1 to copy out the trailing NUL byte */
2150 			error = copyout(bf, where, slen + 1);
2151 			mutex_enter(&vfs_list_lock);
2152 			v->vfs_refcount--;
2153 			if (error)
2154 				break;
2155 			where += slen;
2156 			needed += slen;
2157 			left -= slen;
2158 		}
2159 	}
2160 	mutex_exit(&vfs_list_lock);
2161 	sysctl_relock();
2162 	*oldlenp = needed;
2163 	return (error);
2164 }
2165 
2166 
2167 int kinfo_vdebug = 1;
2168 int kinfo_vgetfailed;
2169 #define KINFO_VNODESLOP	10
2170 /*
2171  * Dump vnode list (via sysctl).
2172  * Copyout address of vnode followed by vnode.
2173  */
2174 /* ARGSUSED */
2175 int
2176 sysctl_kern_vnode(SYSCTLFN_ARGS)
2177 {
2178 	char *where = oldp;
2179 	size_t *sizep = oldlenp;
2180 	struct mount *mp, *nmp;
2181 	vnode_t *vp, *mvp, vbuf;
2182 	char *bp = where;
2183 	char *ewhere;
2184 	int error;
2185 
2186 	if (namelen != 0)
2187 		return (EOPNOTSUPP);
2188 	if (newp != NULL)
2189 		return (EPERM);
2190 
2191 #define VPTRSZ	sizeof(vnode_t *)
2192 #define VNODESZ	sizeof(vnode_t)
2193 	if (where == NULL) {
2194 		*sizep = (numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ);
2195 		return (0);
2196 	}
2197 	ewhere = where + *sizep;
2198 
2199 	sysctl_unlock();
2200 	mutex_enter(&mountlist_lock);
2201 	for (mp = CIRCLEQ_FIRST(&mountlist); mp != (void *)&mountlist;
2202 	    mp = nmp) {
2203 		if (vfs_busy(mp, &nmp)) {
2204 			continue;
2205 		}
2206 		/* Allocate a marker vnode. */
2207 		mvp = vnalloc(mp);
2208 		/* Should never fail for mp != NULL */
2209 		KASSERT(mvp != NULL);
2210 		mutex_enter(&mntvnode_lock);
2211 		for (vp = TAILQ_FIRST(&mp->mnt_vnodelist); vp;
2212 		    vp = vunmark(mvp)) {
2213 			vmark(mvp, vp);
2214 			/*
2215 			 * Check that the vp is still associated with
2216 			 * this filesystem.  RACE: could have been
2217 			 * recycled onto the same filesystem.
2218 			 */
2219 			if (vp->v_mount != mp || vismarker(vp))
2220 				continue;
2221 			if (bp + VPTRSZ + VNODESZ > ewhere) {
2222 				(void)vunmark(mvp);
2223 				mutex_exit(&mntvnode_lock);
2224 				vnfree(mvp);
2225 				vfs_unbusy(mp, false, NULL);
2226 				sysctl_relock();
2227 				*sizep = bp - where;
2228 				return (ENOMEM);
2229 			}
2230 			memcpy(&vbuf, vp, VNODESZ);
2231 			mutex_exit(&mntvnode_lock);
2232 			if ((error = copyout(&vp, bp, VPTRSZ)) ||
2233 			    (error = copyout(&vbuf, bp + VPTRSZ, VNODESZ))) {
2234 			   	mutex_enter(&mntvnode_lock);
2235 				(void)vunmark(mvp);
2236 				mutex_exit(&mntvnode_lock);
2237 				vnfree(mvp);
2238 				vfs_unbusy(mp, false, NULL);
2239 				sysctl_relock();
2240 				return (error);
2241 			}
2242 			bp += VPTRSZ + VNODESZ;
2243 			mutex_enter(&mntvnode_lock);
2244 		}
2245 		mutex_exit(&mntvnode_lock);
2246 		vnfree(mvp);
2247 		vfs_unbusy(mp, false, &nmp);
2248 	}
2249 	mutex_exit(&mountlist_lock);
2250 	sysctl_relock();
2251 
2252 	*sizep = bp - where;
2253 	return (0);
2254 }
2255 
2256 /*
2257  * Remove clean vnodes from a mountpoint's vnode list.
2258  */
2259 void
2260 vfs_scrubvnlist(struct mount *mp)
2261 {
2262 	vnode_t *vp, *nvp;
2263 
2264  retry:
2265 	mutex_enter(&mntvnode_lock);
2266 	for (vp = TAILQ_FIRST(&mp->mnt_vnodelist); vp; vp = nvp) {
2267 		nvp = TAILQ_NEXT(vp, v_mntvnodes);
2268 		mutex_enter(&vp->v_interlock);
2269 		if ((vp->v_iflag & VI_CLEAN) != 0) {
2270 			TAILQ_REMOVE(&mp->mnt_vnodelist, vp, v_mntvnodes);
2271 			vp->v_mount = NULL;
2272 			mutex_exit(&mntvnode_lock);
2273 			mutex_exit(&vp->v_interlock);
2274 			vfs_destroy(mp);
2275 			goto retry;
2276 		}
2277 		mutex_exit(&vp->v_interlock);
2278 	}
2279 	mutex_exit(&mntvnode_lock);
2280 }
2281 
2282 /*
2283  * Check to see if a filesystem is mounted on a block device.
2284  */
2285 int
2286 vfs_mountedon(vnode_t *vp)
2287 {
2288 	vnode_t *vq;
2289 	int error = 0;
2290 
2291 	if (vp->v_type != VBLK)
2292 		return ENOTBLK;
2293 	if (vp->v_specmountpoint != NULL)
2294 		return (EBUSY);
2295 	mutex_enter(&device_lock);
2296 	for (vq = specfs_hash[SPECHASH(vp->v_rdev)]; vq != NULL;
2297 	    vq = vq->v_specnext) {
2298 		if (vq->v_rdev != vp->v_rdev || vq->v_type != vp->v_type)
2299 			continue;
2300 		if (vq->v_specmountpoint != NULL) {
2301 			error = EBUSY;
2302 			break;
2303 		}
2304 	}
2305 	mutex_exit(&device_lock);
2306 	return (error);
2307 }
2308 
2309 /*
2310  * Unmount all file systems.
2311  * We traverse the list in reverse order under the assumption that doing so
2312  * will avoid needing to worry about dependencies.
2313  */
2314 bool
2315 vfs_unmountall(struct lwp *l)
2316 {
2317 
2318 	printf("unmounting file systems...");
2319 	return vfs_unmountall1(l, true, true);
2320 }
2321 
2322 static void
2323 vfs_unmount_print(struct mount *mp, const char *pfx)
2324 {
2325 
2326 	aprint_verbose("%sunmounted %s on %s type %s\n", pfx,
2327 	    mp->mnt_stat.f_mntfromname, mp->mnt_stat.f_mntonname,
2328 	    mp->mnt_stat.f_fstypename);
2329 }
2330 
2331 bool
2332 vfs_unmount_forceone(struct lwp *l)
2333 {
2334 	struct mount *mp, *nmp;
2335 	int error;
2336 
2337 	nmp = NULL;
2338 
2339 	CIRCLEQ_FOREACH_REVERSE(mp, &mountlist, mnt_list) {
2340 		if (nmp == NULL || mp->mnt_gen > nmp->mnt_gen) {
2341 			nmp = mp;
2342 		}
2343 	}
2344 	if (nmp == NULL) {
2345 		return false;
2346 	}
2347 
2348 #ifdef DEBUG
2349 	printf("\nforcefully unmounting %s (%s)...",
2350 	    nmp->mnt_stat.f_mntonname, nmp->mnt_stat.f_mntfromname);
2351 #endif
2352 	atomic_inc_uint(&nmp->mnt_refcnt);
2353 	if ((error = dounmount(nmp, MNT_FORCE, l)) == 0) {
2354 		vfs_unmount_print(nmp, "forcefully ");
2355 		return true;
2356 	} else {
2357 		vfs_destroy(nmp);
2358 	}
2359 
2360 #ifdef DEBUG
2361 	printf("forceful unmount of %s failed with error %d\n",
2362 	    nmp->mnt_stat.f_mntonname, error);
2363 #endif
2364 
2365 	return false;
2366 }
2367 
2368 bool
2369 vfs_unmountall1(struct lwp *l, bool force, bool verbose)
2370 {
2371 	struct mount *mp, *nmp;
2372 	bool any_error = false, progress = false;
2373 	int error;
2374 
2375 	for (mp = CIRCLEQ_LAST(&mountlist);
2376 	     mp != (void *)&mountlist;
2377 	     mp = nmp) {
2378 		nmp = CIRCLEQ_PREV(mp, mnt_list);
2379 #ifdef DEBUG
2380 		printf("\nunmounting %p %s (%s)...",
2381 		    (void *)mp, mp->mnt_stat.f_mntonname,
2382 		    mp->mnt_stat.f_mntfromname);
2383 #endif
2384 		atomic_inc_uint(&mp->mnt_refcnt);
2385 		if ((error = dounmount(mp, force ? MNT_FORCE : 0, l)) == 0) {
2386 			vfs_unmount_print(mp, "");
2387 			progress = true;
2388 		} else {
2389 			vfs_destroy(mp);
2390 			if (verbose) {
2391 				printf("unmount of %s failed with error %d\n",
2392 				    mp->mnt_stat.f_mntonname, error);
2393 			}
2394 			any_error = true;
2395 		}
2396 	}
2397 	if (verbose) {
2398 		printf(" done\n");
2399 	}
2400 	if (any_error && verbose) {
2401 		printf("WARNING: some file systems would not unmount\n");
2402 	}
2403 	return progress;
2404 }
2405 
2406 /*
2407  * Sync and unmount file systems before shutting down.
2408  */
2409 void
2410 vfs_shutdown(void)
2411 {
2412 	struct lwp *l;
2413 
2414 	/* XXX we're certainly not running in lwp0's context! */
2415 	l = (curlwp == NULL) ? &lwp0 : curlwp;
2416 
2417 	vfs_shutdown1(l);
2418 }
2419 
2420 void
2421 vfs_sync_all(struct lwp *l)
2422 {
2423 	printf("syncing disks... ");
2424 
2425 	/* remove user processes from run queue */
2426 	suspendsched();
2427 	(void) spl0();
2428 
2429 	/* avoid coming back this way again if we panic. */
2430 	doing_shutdown = 1;
2431 
2432 	sys_sync(l, NULL, NULL);
2433 
2434 	/* Wait for sync to finish. */
2435 	if (buf_syncwait() != 0) {
2436 #if defined(DDB) && defined(DEBUG_HALT_BUSY)
2437 		Debugger();
2438 #endif
2439 		printf("giving up\n");
2440 		return;
2441 	} else
2442 		printf("done\n");
2443 }
2444 
2445 static void
2446 vfs_shutdown1(struct lwp *l)
2447 {
2448 
2449 	vfs_sync_all(l);
2450 
2451 	/*
2452 	 * If we've panic'd, don't make the situation potentially
2453 	 * worse by unmounting the file systems.
2454 	 */
2455 	if (panicstr != NULL)
2456 		return;
2457 
2458 	/* Release inodes held by texts before update. */
2459 #ifdef notdef
2460 	vnshutdown();
2461 #endif
2462 	/* Unmount file systems. */
2463 	vfs_unmountall(l);
2464 }
2465 
2466 /*
2467  * Print a list of supported file system types (used by vfs_mountroot)
2468  */
2469 static void
2470 vfs_print_fstypes(void)
2471 {
2472 	struct vfsops *v;
2473 	int cnt = 0;
2474 
2475 	mutex_enter(&vfs_list_lock);
2476 	LIST_FOREACH(v, &vfs_list, vfs_list)
2477 		++cnt;
2478 	mutex_exit(&vfs_list_lock);
2479 
2480 	if (cnt == 0) {
2481 		printf("WARNING: No file system modules have been loaded.\n");
2482 		return;
2483 	}
2484 
2485 	printf("Supported file systems:");
2486 	mutex_enter(&vfs_list_lock);
2487 	LIST_FOREACH(v, &vfs_list, vfs_list) {
2488 		printf(" %s", v->vfs_name);
2489 	}
2490 	mutex_exit(&vfs_list_lock);
2491 	printf("\n");
2492 }
2493 
2494 /*
2495  * Mount the root file system.  If the operator didn't specify a
2496  * file system to use, try all possible file systems until one
2497  * succeeds.
2498  */
2499 int
2500 vfs_mountroot(void)
2501 {
2502 	struct vfsops *v;
2503 	int error = ENODEV;
2504 
2505 	if (root_device == NULL)
2506 		panic("vfs_mountroot: root device unknown");
2507 
2508 	switch (device_class(root_device)) {
2509 	case DV_IFNET:
2510 		if (rootdev != NODEV)
2511 			panic("vfs_mountroot: rootdev set for DV_IFNET "
2512 			    "(0x%llx -> %llu,%llu)",
2513 			    (unsigned long long)rootdev,
2514 			    (unsigned long long)major(rootdev),
2515 			    (unsigned long long)minor(rootdev));
2516 		break;
2517 
2518 	case DV_DISK:
2519 		if (rootdev == NODEV)
2520 			panic("vfs_mountroot: rootdev not set for DV_DISK");
2521 	        if (bdevvp(rootdev, &rootvp))
2522 	                panic("vfs_mountroot: can't get vnode for rootdev");
2523 		error = VOP_OPEN(rootvp, FREAD, FSCRED);
2524 		if (error) {
2525 			printf("vfs_mountroot: can't open root device\n");
2526 			return (error);
2527 		}
2528 		break;
2529 
2530 	case DV_VIRTUAL:
2531 		break;
2532 
2533 	default:
2534 		printf("%s: inappropriate for root file system\n",
2535 		    device_xname(root_device));
2536 		return (ENODEV);
2537 	}
2538 
2539 	/*
2540 	 * If user specified a root fs type, use it.  Make sure the
2541 	 * specified type exists and has a mount_root()
2542 	 */
2543 	if (strcmp(rootfstype, ROOT_FSTYPE_ANY) != 0) {
2544 		v = vfs_getopsbyname(rootfstype);
2545 		error = EFTYPE;
2546 		if (v != NULL) {
2547 			if (v->vfs_mountroot != NULL) {
2548 				error = (v->vfs_mountroot)();
2549 			}
2550 			v->vfs_refcount--;
2551 		}
2552 		goto done;
2553 	}
2554 
2555 	/*
2556 	 * Try each file system currently configured into the kernel.
2557 	 */
2558 	mutex_enter(&vfs_list_lock);
2559 	LIST_FOREACH(v, &vfs_list, vfs_list) {
2560 		if (v->vfs_mountroot == NULL)
2561 			continue;
2562 #ifdef DEBUG
2563 		aprint_normal("mountroot: trying %s...\n", v->vfs_name);
2564 #endif
2565 		v->vfs_refcount++;
2566 		mutex_exit(&vfs_list_lock);
2567 		error = (*v->vfs_mountroot)();
2568 		mutex_enter(&vfs_list_lock);
2569 		v->vfs_refcount--;
2570 		if (!error) {
2571 			aprint_normal("root file system type: %s\n",
2572 			    v->vfs_name);
2573 			break;
2574 		}
2575 	}
2576 	mutex_exit(&vfs_list_lock);
2577 
2578 	if (v == NULL) {
2579 		vfs_print_fstypes();
2580 		printf("no file system for %s", device_xname(root_device));
2581 		if (device_class(root_device) == DV_DISK)
2582 			printf(" (dev 0x%llx)", (unsigned long long)rootdev);
2583 		printf("\n");
2584 		error = EFTYPE;
2585 	}
2586 
2587 done:
2588 	if (error && device_class(root_device) == DV_DISK) {
2589 		VOP_CLOSE(rootvp, FREAD, FSCRED);
2590 		vrele(rootvp);
2591 	}
2592 	if (error == 0) {
2593 		extern struct cwdinfo cwdi0;
2594 
2595 		CIRCLEQ_FIRST(&mountlist)->mnt_flag |= MNT_ROOTFS;
2596 		CIRCLEQ_FIRST(&mountlist)->mnt_op->vfs_refcount++;
2597 
2598 		/*
2599 		 * Get the vnode for '/'.  Set cwdi0.cwdi_cdir to
2600 		 * reference it.
2601 		 */
2602 		error = VFS_ROOT(CIRCLEQ_FIRST(&mountlist), &rootvnode);
2603 		if (error)
2604 			panic("cannot find root vnode, error=%d", error);
2605 		cwdi0.cwdi_cdir = rootvnode;
2606 		vref(cwdi0.cwdi_cdir);
2607 		VOP_UNLOCK(rootvnode);
2608 		cwdi0.cwdi_rdir = NULL;
2609 
2610 		/*
2611 		 * Now that root is mounted, we can fixup initproc's CWD
2612 		 * info.  All other processes are kthreads, which merely
2613 		 * share proc0's CWD info.
2614 		 */
2615 		initproc->p_cwdi->cwdi_cdir = rootvnode;
2616 		vref(initproc->p_cwdi->cwdi_cdir);
2617 		initproc->p_cwdi->cwdi_rdir = NULL;
2618 		/*
2619 		 * Enable loading of modules from the filesystem
2620 		 */
2621 		module_load_vfs_init();
2622 
2623 	}
2624 	return (error);
2625 }
2626 
2627 /*
2628  * Get a new unique fsid
2629  */
2630 void
2631 vfs_getnewfsid(struct mount *mp)
2632 {
2633 	static u_short xxxfs_mntid;
2634 	fsid_t tfsid;
2635 	int mtype;
2636 
2637 	mutex_enter(&mntid_lock);
2638 	mtype = makefstype(mp->mnt_op->vfs_name);
2639 	mp->mnt_stat.f_fsidx.__fsid_val[0] = makedev(mtype, 0);
2640 	mp->mnt_stat.f_fsidx.__fsid_val[1] = mtype;
2641 	mp->mnt_stat.f_fsid = mp->mnt_stat.f_fsidx.__fsid_val[0];
2642 	if (xxxfs_mntid == 0)
2643 		++xxxfs_mntid;
2644 	tfsid.__fsid_val[0] = makedev(mtype & 0xff, xxxfs_mntid);
2645 	tfsid.__fsid_val[1] = mtype;
2646 	if (!CIRCLEQ_EMPTY(&mountlist)) {
2647 		while (vfs_getvfs(&tfsid)) {
2648 			tfsid.__fsid_val[0]++;
2649 			xxxfs_mntid++;
2650 		}
2651 	}
2652 	mp->mnt_stat.f_fsidx.__fsid_val[0] = tfsid.__fsid_val[0];
2653 	mp->mnt_stat.f_fsid = mp->mnt_stat.f_fsidx.__fsid_val[0];
2654 	mutex_exit(&mntid_lock);
2655 }
2656 
2657 /*
2658  * Make a 'unique' number from a mount type name.
2659  */
2660 long
2661 makefstype(const char *type)
2662 {
2663 	long rv;
2664 
2665 	for (rv = 0; *type; type++) {
2666 		rv <<= 2;
2667 		rv ^= *type;
2668 	}
2669 	return rv;
2670 }
2671 
2672 /*
2673  * Set vnode attributes to VNOVAL
2674  */
2675 void
2676 vattr_null(struct vattr *vap)
2677 {
2678 
2679 	memset(vap, 0, sizeof(*vap));
2680 
2681 	vap->va_type = VNON;
2682 
2683 	/*
2684 	 * Assign individually so that it is safe even if size and
2685 	 * sign of each member are varied.
2686 	 */
2687 	vap->va_mode = VNOVAL;
2688 	vap->va_nlink = VNOVAL;
2689 	vap->va_uid = VNOVAL;
2690 	vap->va_gid = VNOVAL;
2691 	vap->va_fsid = VNOVAL;
2692 	vap->va_fileid = VNOVAL;
2693 	vap->va_size = VNOVAL;
2694 	vap->va_blocksize = VNOVAL;
2695 	vap->va_atime.tv_sec =
2696 	    vap->va_mtime.tv_sec =
2697 	    vap->va_ctime.tv_sec =
2698 	    vap->va_birthtime.tv_sec = VNOVAL;
2699 	vap->va_atime.tv_nsec =
2700 	    vap->va_mtime.tv_nsec =
2701 	    vap->va_ctime.tv_nsec =
2702 	    vap->va_birthtime.tv_nsec = VNOVAL;
2703 	vap->va_gen = VNOVAL;
2704 	vap->va_flags = VNOVAL;
2705 	vap->va_rdev = VNOVAL;
2706 	vap->va_bytes = VNOVAL;
2707 }
2708 
2709 #define ARRAY_SIZE(arr) (sizeof(arr) / sizeof(arr[0]))
2710 #define ARRAY_PRINT(idx, arr) \
2711     ((unsigned int)(idx) < ARRAY_SIZE(arr) ? (arr)[(idx)] : "UNKNOWN")
2712 
2713 const char * const vnode_tags[] = { VNODE_TAGS };
2714 const char * const vnode_types[] = { VNODE_TYPES };
2715 const char vnode_flagbits[] = VNODE_FLAGBITS;
2716 
2717 /*
2718  * Print out a description of a vnode.
2719  */
2720 void
2721 vprint(const char *label, struct vnode *vp)
2722 {
2723 	struct vnlock *vl;
2724 	char bf[96];
2725 	int flag;
2726 
2727 	vl = &vp->v_lock;
2728 	flag = vp->v_iflag | vp->v_vflag | vp->v_uflag;
2729 	snprintb(bf, sizeof(bf), vnode_flagbits, flag);
2730 
2731 	if (label != NULL)
2732 		printf("%s: ", label);
2733 	printf("vnode @ %p, flags (%s)\n\ttag %s(%d), type %s(%d), "
2734 	    "usecount %d, writecount %d, holdcount %d\n"
2735 	    "\tfreelisthd %p, mount %p, data %p lock %p\n",
2736 	    vp, bf, ARRAY_PRINT(vp->v_tag, vnode_tags), vp->v_tag,
2737 	    ARRAY_PRINT(vp->v_type, vnode_types), vp->v_type,
2738 	    vp->v_usecount, vp->v_writecount, vp->v_holdcnt,
2739 	    vp->v_freelisthd, vp->v_mount, vp->v_data, vl);
2740 	if (vp->v_data != NULL) {
2741 		printf("\t");
2742 		VOP_PRINT(vp);
2743 	}
2744 }
2745 
2746 #ifdef DEBUG
2747 /*
2748  * List all of the locked vnodes in the system.
2749  * Called when debugging the kernel.
2750  */
2751 void
2752 printlockedvnodes(void)
2753 {
2754 	struct mount *mp, *nmp;
2755 	struct vnode *vp;
2756 
2757 	printf("Locked vnodes\n");
2758 	mutex_enter(&mountlist_lock);
2759 	for (mp = CIRCLEQ_FIRST(&mountlist); mp != (void *)&mountlist;
2760 	     mp = nmp) {
2761 		if (vfs_busy(mp, &nmp)) {
2762 			continue;
2763 		}
2764 		TAILQ_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
2765 			if (VOP_ISLOCKED(vp))
2766 				vprint(NULL, vp);
2767 		}
2768 		mutex_enter(&mountlist_lock);
2769 		vfs_unbusy(mp, false, &nmp);
2770 	}
2771 	mutex_exit(&mountlist_lock);
2772 }
2773 #endif
2774 
2775 /* Deprecated. Kept for KPI compatibility. */
2776 int
2777 vaccess(enum vtype type, mode_t file_mode, uid_t uid, gid_t gid,
2778     mode_t acc_mode, kauth_cred_t cred)
2779 {
2780 
2781 #ifdef DIAGNOSTIC
2782 	printf("vaccess: deprecated interface used.\n");
2783 #endif /* DIAGNOSTIC */
2784 
2785 	return genfs_can_access(type, file_mode, uid, gid, acc_mode, cred);
2786 }
2787 
2788 /*
2789  * Given a file system name, look up the vfsops for that
2790  * file system, or return NULL if file system isn't present
2791  * in the kernel.
2792  */
2793 struct vfsops *
2794 vfs_getopsbyname(const char *name)
2795 {
2796 	struct vfsops *v;
2797 
2798 	mutex_enter(&vfs_list_lock);
2799 	LIST_FOREACH(v, &vfs_list, vfs_list) {
2800 		if (strcmp(v->vfs_name, name) == 0)
2801 			break;
2802 	}
2803 	if (v != NULL)
2804 		v->vfs_refcount++;
2805 	mutex_exit(&vfs_list_lock);
2806 
2807 	return (v);
2808 }
2809 
2810 void
2811 copy_statvfs_info(struct statvfs *sbp, const struct mount *mp)
2812 {
2813 	const struct statvfs *mbp;
2814 
2815 	if (sbp == (mbp = &mp->mnt_stat))
2816 		return;
2817 
2818 	(void)memcpy(&sbp->f_fsidx, &mbp->f_fsidx, sizeof(sbp->f_fsidx));
2819 	sbp->f_fsid = mbp->f_fsid;
2820 	sbp->f_owner = mbp->f_owner;
2821 	sbp->f_flag = mbp->f_flag;
2822 	sbp->f_syncwrites = mbp->f_syncwrites;
2823 	sbp->f_asyncwrites = mbp->f_asyncwrites;
2824 	sbp->f_syncreads = mbp->f_syncreads;
2825 	sbp->f_asyncreads = mbp->f_asyncreads;
2826 	(void)memcpy(sbp->f_spare, mbp->f_spare, sizeof(mbp->f_spare));
2827 	(void)memcpy(sbp->f_fstypename, mbp->f_fstypename,
2828 	    sizeof(sbp->f_fstypename));
2829 	(void)memcpy(sbp->f_mntonname, mbp->f_mntonname,
2830 	    sizeof(sbp->f_mntonname));
2831 	(void)memcpy(sbp->f_mntfromname, mp->mnt_stat.f_mntfromname,
2832 	    sizeof(sbp->f_mntfromname));
2833 	sbp->f_namemax = mbp->f_namemax;
2834 }
2835 
2836 int
2837 set_statvfs_info(const char *onp, int ukon, const char *fromp, int ukfrom,
2838     const char *vfsname, struct mount *mp, struct lwp *l)
2839 {
2840 	int error;
2841 	size_t size;
2842 	struct statvfs *sfs = &mp->mnt_stat;
2843 	int (*fun)(const void *, void *, size_t, size_t *);
2844 
2845 	(void)strlcpy(mp->mnt_stat.f_fstypename, vfsname,
2846 	    sizeof(mp->mnt_stat.f_fstypename));
2847 
2848 	if (onp) {
2849 		struct cwdinfo *cwdi = l->l_proc->p_cwdi;
2850 		fun = (ukon == UIO_SYSSPACE) ? copystr : copyinstr;
2851 		if (cwdi->cwdi_rdir != NULL) {
2852 			size_t len;
2853 			char *bp;
2854 			char *path = PNBUF_GET();
2855 
2856 			bp = path + MAXPATHLEN;
2857 			*--bp = '\0';
2858 			rw_enter(&cwdi->cwdi_lock, RW_READER);
2859 			error = getcwd_common(cwdi->cwdi_rdir, rootvnode, &bp,
2860 			    path, MAXPATHLEN / 2, 0, l);
2861 			rw_exit(&cwdi->cwdi_lock);
2862 			if (error) {
2863 				PNBUF_PUT(path);
2864 				return error;
2865 			}
2866 
2867 			len = strlen(bp);
2868 			if (len > sizeof(sfs->f_mntonname) - 1)
2869 				len = sizeof(sfs->f_mntonname) - 1;
2870 			(void)strncpy(sfs->f_mntonname, bp, len);
2871 			PNBUF_PUT(path);
2872 
2873 			if (len < sizeof(sfs->f_mntonname) - 1) {
2874 				error = (*fun)(onp, &sfs->f_mntonname[len],
2875 				    sizeof(sfs->f_mntonname) - len - 1, &size);
2876 				if (error)
2877 					return error;
2878 				size += len;
2879 			} else {
2880 				size = len;
2881 			}
2882 		} else {
2883 			error = (*fun)(onp, &sfs->f_mntonname,
2884 			    sizeof(sfs->f_mntonname) - 1, &size);
2885 			if (error)
2886 				return error;
2887 		}
2888 		(void)memset(sfs->f_mntonname + size, 0,
2889 		    sizeof(sfs->f_mntonname) - size);
2890 	}
2891 
2892 	if (fromp) {
2893 		fun = (ukfrom == UIO_SYSSPACE) ? copystr : copyinstr;
2894 		error = (*fun)(fromp, sfs->f_mntfromname,
2895 		    sizeof(sfs->f_mntfromname) - 1, &size);
2896 		if (error)
2897 			return error;
2898 		(void)memset(sfs->f_mntfromname + size, 0,
2899 		    sizeof(sfs->f_mntfromname) - size);
2900 	}
2901 	return 0;
2902 }
2903 
2904 void
2905 vfs_timestamp(struct timespec *ts)
2906 {
2907 
2908 	nanotime(ts);
2909 }
2910 
2911 time_t	rootfstime;			/* recorded root fs time, if known */
2912 void
2913 setrootfstime(time_t t)
2914 {
2915 	rootfstime = t;
2916 }
2917 
2918 /*
2919  * Sham lock manager for vnodes.  This is a temporary measure.
2920  */
2921 int
2922 vlockmgr(struct vnlock *vl, int flags)
2923 {
2924 
2925 	KASSERT((flags & ~(LK_NOWAIT | LK_TYPE_MASK)) == 0);
2926 
2927 	switch (flags & (LK_NOWAIT | LK_TYPE_MASK)) {
2928 	case LK_SHARED:
2929 		rw_enter(&vl->vl_lock, RW_READER);
2930 		return 0;
2931 
2932 	case LK_SHARED | LK_NOWAIT:
2933 		return rw_tryenter(&vl->vl_lock, RW_READER) ? 0 : EBUSY;
2934 
2935 	case LK_EXCLUSIVE:
2936 		rw_enter(&vl->vl_lock, RW_WRITER);
2937 		return 0;
2938 
2939 	case LK_EXCLUSIVE | LK_NOWAIT:
2940 		return rw_tryenter(&vl->vl_lock, RW_WRITER) ? 0 : EBUSY;
2941 
2942 	case LK_RELEASE:
2943 		rw_exit(&vl->vl_lock);
2944 		return 0;
2945 
2946 	default:
2947 		panic("vlockmgr: flags %x", flags);
2948 	}
2949 }
2950 
2951 int
2952 vlockstatus(struct vnlock *vl)
2953 {
2954 
2955 	if (rw_write_held(&vl->vl_lock)) {
2956 		return LK_EXCLUSIVE;
2957 	}
2958 	if (rw_read_held(&vl->vl_lock)) {
2959 		return LK_SHARED;
2960 	}
2961 	return 0;
2962 }
2963 
2964 static const uint8_t vttodt_tab[9] = {
2965 	DT_UNKNOWN,	/* VNON  */
2966 	DT_REG,		/* VREG  */
2967 	DT_DIR,		/* VDIR  */
2968 	DT_BLK,		/* VBLK  */
2969 	DT_CHR,		/* VCHR  */
2970 	DT_LNK,		/* VLNK  */
2971 	DT_SOCK,	/* VSUCK */
2972 	DT_FIFO,	/* VFIFO */
2973 	DT_UNKNOWN	/* VBAD  */
2974 };
2975 
2976 uint8_t
2977 vtype2dt(enum vtype vt)
2978 {
2979 
2980 	CTASSERT(VBAD == __arraycount(vttodt_tab) - 1);
2981 	return vttodt_tab[vt];
2982 }
2983 
2984 /*
2985  * mount_specific_key_create --
2986  *	Create a key for subsystem mount-specific data.
2987  */
2988 int
2989 mount_specific_key_create(specificdata_key_t *keyp, specificdata_dtor_t dtor)
2990 {
2991 
2992 	return (specificdata_key_create(mount_specificdata_domain, keyp, dtor));
2993 }
2994 
2995 /*
2996  * mount_specific_key_delete --
2997  *	Delete a key for subsystem mount-specific data.
2998  */
2999 void
3000 mount_specific_key_delete(specificdata_key_t key)
3001 {
3002 
3003 	specificdata_key_delete(mount_specificdata_domain, key);
3004 }
3005 
3006 /*
3007  * mount_initspecific --
3008  *	Initialize a mount's specificdata container.
3009  */
3010 void
3011 mount_initspecific(struct mount *mp)
3012 {
3013 	int error;
3014 
3015 	error = specificdata_init(mount_specificdata_domain,
3016 				  &mp->mnt_specdataref);
3017 	KASSERT(error == 0);
3018 }
3019 
3020 /*
3021  * mount_finispecific --
3022  *	Finalize a mount's specificdata container.
3023  */
3024 void
3025 mount_finispecific(struct mount *mp)
3026 {
3027 
3028 	specificdata_fini(mount_specificdata_domain, &mp->mnt_specdataref);
3029 }
3030 
3031 /*
3032  * mount_getspecific --
3033  *	Return mount-specific data corresponding to the specified key.
3034  */
3035 void *
3036 mount_getspecific(struct mount *mp, specificdata_key_t key)
3037 {
3038 
3039 	return (specificdata_getspecific(mount_specificdata_domain,
3040 					 &mp->mnt_specdataref, key));
3041 }
3042 
3043 /*
3044  * mount_setspecific --
3045  *	Set mount-specific data corresponding to the specified key.
3046  */
3047 void
3048 mount_setspecific(struct mount *mp, specificdata_key_t key, void *data)
3049 {
3050 
3051 	specificdata_setspecific(mount_specificdata_domain,
3052 				 &mp->mnt_specdataref, key, data);
3053 }
3054 
3055 int
3056 VFS_MOUNT(struct mount *mp, const char *a, void *b, size_t *c)
3057 {
3058 	int error;
3059 
3060 	KERNEL_LOCK(1, NULL);
3061 	error = (*(mp->mnt_op->vfs_mount))(mp, a, b, c);
3062 	KERNEL_UNLOCK_ONE(NULL);
3063 
3064 	return error;
3065 }
3066 
3067 int
3068 VFS_START(struct mount *mp, int a)
3069 {
3070 	int error;
3071 
3072 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
3073 		KERNEL_LOCK(1, NULL);
3074 	}
3075 	error = (*(mp->mnt_op->vfs_start))(mp, a);
3076 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
3077 		KERNEL_UNLOCK_ONE(NULL);
3078 	}
3079 
3080 	return error;
3081 }
3082 
3083 int
3084 VFS_UNMOUNT(struct mount *mp, int a)
3085 {
3086 	int error;
3087 
3088 	KERNEL_LOCK(1, NULL);
3089 	error = (*(mp->mnt_op->vfs_unmount))(mp, a);
3090 	KERNEL_UNLOCK_ONE(NULL);
3091 
3092 	return error;
3093 }
3094 
3095 int
3096 VFS_ROOT(struct mount *mp, struct vnode **a)
3097 {
3098 	int error;
3099 
3100 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
3101 		KERNEL_LOCK(1, NULL);
3102 	}
3103 	error = (*(mp->mnt_op->vfs_root))(mp, a);
3104 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
3105 		KERNEL_UNLOCK_ONE(NULL);
3106 	}
3107 
3108 	return error;
3109 }
3110 
3111 int
3112 VFS_QUOTACTL(struct mount *mp, int a, uid_t b, void *c)
3113 {
3114 	int error;
3115 
3116 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
3117 		KERNEL_LOCK(1, NULL);
3118 	}
3119 	error = (*(mp->mnt_op->vfs_quotactl))(mp, a, b, c);
3120 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
3121 		KERNEL_UNLOCK_ONE(NULL);
3122 	}
3123 
3124 	return error;
3125 }
3126 
3127 int
3128 VFS_STATVFS(struct mount *mp, struct statvfs *a)
3129 {
3130 	int error;
3131 
3132 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
3133 		KERNEL_LOCK(1, NULL);
3134 	}
3135 	error = (*(mp->mnt_op->vfs_statvfs))(mp, a);
3136 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
3137 		KERNEL_UNLOCK_ONE(NULL);
3138 	}
3139 
3140 	return error;
3141 }
3142 
3143 int
3144 VFS_SYNC(struct mount *mp, int a, struct kauth_cred *b)
3145 {
3146 	int error;
3147 
3148 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
3149 		KERNEL_LOCK(1, NULL);
3150 	}
3151 	error = (*(mp->mnt_op->vfs_sync))(mp, a, b);
3152 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
3153 		KERNEL_UNLOCK_ONE(NULL);
3154 	}
3155 
3156 	return error;
3157 }
3158 
3159 int
3160 VFS_FHTOVP(struct mount *mp, struct fid *a, struct vnode **b)
3161 {
3162 	int error;
3163 
3164 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
3165 		KERNEL_LOCK(1, NULL);
3166 	}
3167 	error = (*(mp->mnt_op->vfs_fhtovp))(mp, a, b);
3168 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
3169 		KERNEL_UNLOCK_ONE(NULL);
3170 	}
3171 
3172 	return error;
3173 }
3174 
3175 int
3176 VFS_VPTOFH(struct vnode *vp, struct fid *a, size_t *b)
3177 {
3178 	int error;
3179 
3180 	if ((vp->v_vflag & VV_MPSAFE) == 0) {
3181 		KERNEL_LOCK(1, NULL);
3182 	}
3183 	error = (*(vp->v_mount->mnt_op->vfs_vptofh))(vp, a, b);
3184 	if ((vp->v_vflag & VV_MPSAFE) == 0) {
3185 		KERNEL_UNLOCK_ONE(NULL);
3186 	}
3187 
3188 	return error;
3189 }
3190 
3191 int
3192 VFS_SNAPSHOT(struct mount *mp, struct vnode *a, struct timespec *b)
3193 {
3194 	int error;
3195 
3196 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
3197 		KERNEL_LOCK(1, NULL);
3198 	}
3199 	error = (*(mp->mnt_op->vfs_snapshot))(mp, a, b);
3200 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
3201 		KERNEL_UNLOCK_ONE(NULL);
3202 	}
3203 
3204 	return error;
3205 }
3206 
3207 int
3208 VFS_EXTATTRCTL(struct mount *mp, int a, struct vnode *b, int c, const char *d)
3209 {
3210 	int error;
3211 
3212 	KERNEL_LOCK(1, NULL);		/* XXXSMP check ffs */
3213 	error = (*(mp->mnt_op->vfs_extattrctl))(mp, a, b, c, d);
3214 	KERNEL_UNLOCK_ONE(NULL);	/* XXX */
3215 
3216 	return error;
3217 }
3218 
3219 int
3220 VFS_SUSPENDCTL(struct mount *mp, int a)
3221 {
3222 	int error;
3223 
3224 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
3225 		KERNEL_LOCK(1, NULL);
3226 	}
3227 	error = (*(mp->mnt_op->vfs_suspendctl))(mp, a);
3228 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
3229 		KERNEL_UNLOCK_ONE(NULL);
3230 	}
3231 
3232 	return error;
3233 }
3234 
3235 #if defined(DDB) || defined(DEBUGPRINT)
3236 static const char buf_flagbits[] = BUF_FLAGBITS;
3237 
3238 void
3239 vfs_buf_print(struct buf *bp, int full, void (*pr)(const char *, ...))
3240 {
3241 	char bf[1024];
3242 
3243 	(*pr)("  vp %p lblkno 0x%"PRIx64" blkno 0x%"PRIx64" rawblkno 0x%"
3244 	    PRIx64 " dev 0x%x\n",
3245 	    bp->b_vp, bp->b_lblkno, bp->b_blkno, bp->b_rawblkno, bp->b_dev);
3246 
3247 	snprintb(bf, sizeof(bf),
3248 	    buf_flagbits, bp->b_flags | bp->b_oflags | bp->b_cflags);
3249 	(*pr)("  error %d flags 0x%s\n", bp->b_error, bf);
3250 
3251 	(*pr)("  bufsize 0x%lx bcount 0x%lx resid 0x%lx\n",
3252 		  bp->b_bufsize, bp->b_bcount, bp->b_resid);
3253 	(*pr)("  data %p saveaddr %p\n",
3254 		  bp->b_data, bp->b_saveaddr);
3255 	(*pr)("  iodone %p objlock %p\n", bp->b_iodone, bp->b_objlock);
3256 }
3257 
3258 
3259 void
3260 vfs_vnode_print(struct vnode *vp, int full, void (*pr)(const char *, ...))
3261 {
3262 	char bf[256];
3263 
3264 	uvm_object_printit(&vp->v_uobj, full, pr);
3265 	snprintb(bf, sizeof(bf),
3266 	    vnode_flagbits, vp->v_iflag | vp->v_vflag | vp->v_uflag);
3267 	(*pr)("\nVNODE flags %s\n", bf);
3268 	(*pr)("mp %p numoutput %d size 0x%llx writesize 0x%llx\n",
3269 	      vp->v_mount, vp->v_numoutput, vp->v_size, vp->v_writesize);
3270 
3271 	(*pr)("data %p writecount %ld holdcnt %ld\n",
3272 	      vp->v_data, vp->v_writecount, vp->v_holdcnt);
3273 
3274 	(*pr)("tag %s(%d) type %s(%d) mount %p typedata %p\n",
3275 	      ARRAY_PRINT(vp->v_tag, vnode_tags), vp->v_tag,
3276 	      ARRAY_PRINT(vp->v_type, vnode_types), vp->v_type,
3277 	      vp->v_mount, vp->v_mountedhere);
3278 
3279 	(*pr)("v_lock %p\n", &vp->v_lock);
3280 
3281 	if (full) {
3282 		struct buf *bp;
3283 
3284 		(*pr)("clean bufs:\n");
3285 		LIST_FOREACH(bp, &vp->v_cleanblkhd, b_vnbufs) {
3286 			(*pr)(" bp %p\n", bp);
3287 			vfs_buf_print(bp, full, pr);
3288 		}
3289 
3290 		(*pr)("dirty bufs:\n");
3291 		LIST_FOREACH(bp, &vp->v_dirtyblkhd, b_vnbufs) {
3292 			(*pr)(" bp %p\n", bp);
3293 			vfs_buf_print(bp, full, pr);
3294 		}
3295 	}
3296 }
3297 
3298 void
3299 vfs_mount_print(struct mount *mp, int full, void (*pr)(const char *, ...))
3300 {
3301 	char sbuf[256];
3302 
3303 	(*pr)("vnodecovered = %p syncer = %p data = %p\n",
3304 			mp->mnt_vnodecovered,mp->mnt_syncer,mp->mnt_data);
3305 
3306 	(*pr)("fs_bshift %d dev_bshift = %d\n",
3307 			mp->mnt_fs_bshift,mp->mnt_dev_bshift);
3308 
3309 	snprintb(sbuf, sizeof(sbuf), __MNT_FLAG_BITS, mp->mnt_flag);
3310 	(*pr)("flag = %s\n", sbuf);
3311 
3312 	snprintb(sbuf, sizeof(sbuf), __IMNT_FLAG_BITS, mp->mnt_iflag);
3313 	(*pr)("iflag = %s\n", sbuf);
3314 
3315 	(*pr)("refcnt = %d unmounting @ %p updating @ %p\n", mp->mnt_refcnt,
3316 	    &mp->mnt_unmounting, &mp->mnt_updating);
3317 
3318 	(*pr)("statvfs cache:\n");
3319 	(*pr)("\tbsize = %lu\n",mp->mnt_stat.f_bsize);
3320 	(*pr)("\tfrsize = %lu\n",mp->mnt_stat.f_frsize);
3321 	(*pr)("\tiosize = %lu\n",mp->mnt_stat.f_iosize);
3322 
3323 	(*pr)("\tblocks = %"PRIu64"\n",mp->mnt_stat.f_blocks);
3324 	(*pr)("\tbfree = %"PRIu64"\n",mp->mnt_stat.f_bfree);
3325 	(*pr)("\tbavail = %"PRIu64"\n",mp->mnt_stat.f_bavail);
3326 	(*pr)("\tbresvd = %"PRIu64"\n",mp->mnt_stat.f_bresvd);
3327 
3328 	(*pr)("\tfiles = %"PRIu64"\n",mp->mnt_stat.f_files);
3329 	(*pr)("\tffree = %"PRIu64"\n",mp->mnt_stat.f_ffree);
3330 	(*pr)("\tfavail = %"PRIu64"\n",mp->mnt_stat.f_favail);
3331 	(*pr)("\tfresvd = %"PRIu64"\n",mp->mnt_stat.f_fresvd);
3332 
3333 	(*pr)("\tf_fsidx = { 0x%"PRIx32", 0x%"PRIx32" }\n",
3334 			mp->mnt_stat.f_fsidx.__fsid_val[0],
3335 			mp->mnt_stat.f_fsidx.__fsid_val[1]);
3336 
3337 	(*pr)("\towner = %"PRIu32"\n",mp->mnt_stat.f_owner);
3338 	(*pr)("\tnamemax = %lu\n",mp->mnt_stat.f_namemax);
3339 
3340 	snprintb(sbuf, sizeof(sbuf), __MNT_FLAG_BITS, mp->mnt_stat.f_flag);
3341 
3342 	(*pr)("\tflag = %s\n",sbuf);
3343 	(*pr)("\tsyncwrites = %" PRIu64 "\n",mp->mnt_stat.f_syncwrites);
3344 	(*pr)("\tasyncwrites = %" PRIu64 "\n",mp->mnt_stat.f_asyncwrites);
3345 	(*pr)("\tsyncreads = %" PRIu64 "\n",mp->mnt_stat.f_syncreads);
3346 	(*pr)("\tasyncreads = %" PRIu64 "\n",mp->mnt_stat.f_asyncreads);
3347 	(*pr)("\tfstypename = %s\n",mp->mnt_stat.f_fstypename);
3348 	(*pr)("\tmntonname = %s\n",mp->mnt_stat.f_mntonname);
3349 	(*pr)("\tmntfromname = %s\n",mp->mnt_stat.f_mntfromname);
3350 
3351 	{
3352 		int cnt = 0;
3353 		struct vnode *vp;
3354 		(*pr)("locked vnodes =");
3355 		TAILQ_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
3356 			if (VOP_ISLOCKED(vp)) {
3357 				if ((++cnt % 6) == 0) {
3358 					(*pr)(" %p,\n\t", vp);
3359 				} else {
3360 					(*pr)(" %p,", vp);
3361 				}
3362 			}
3363 		}
3364 		(*pr)("\n");
3365 	}
3366 
3367 	if (full) {
3368 		int cnt = 0;
3369 		struct vnode *vp;
3370 		(*pr)("all vnodes =");
3371 		TAILQ_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
3372 			if (!TAILQ_NEXT(vp, v_mntvnodes)) {
3373 				(*pr)(" %p", vp);
3374 			} else if ((++cnt % 6) == 0) {
3375 				(*pr)(" %p,\n\t", vp);
3376 			} else {
3377 				(*pr)(" %p,", vp);
3378 			}
3379 		}
3380 		(*pr)("\n", vp);
3381 	}
3382 }
3383 #endif /* DDB || DEBUGPRINT */
3384 
3385 /*
3386  * Check if a device pointed to by vp is mounted.
3387  *
3388  * Returns:
3389  *   EINVAL	if it's not a disk
3390  *   EBUSY	if it's a disk and mounted
3391  *   0		if it's a disk and not mounted
3392  */
3393 int
3394 rawdev_mounted(struct vnode *vp, struct vnode **bvpp)
3395 {
3396 	struct vnode *bvp;
3397 	dev_t dev;
3398 	int d_type;
3399 
3400 	bvp = NULL;
3401 	dev = vp->v_rdev;
3402 	d_type = D_OTHER;
3403 
3404 	if (iskmemvp(vp))
3405 		return EINVAL;
3406 
3407 	switch (vp->v_type) {
3408 	case VCHR: {
3409 		const struct cdevsw *cdev;
3410 
3411 		cdev = cdevsw_lookup(dev);
3412 		if (cdev != NULL) {
3413 			dev_t blkdev;
3414 
3415 			blkdev = devsw_chr2blk(dev);
3416 			if (blkdev != NODEV) {
3417 				vfinddev(blkdev, VBLK, &bvp);
3418 				if (bvp != NULL)
3419 					d_type = (cdev->d_flag & D_TYPEMASK);
3420 			}
3421 		}
3422 
3423 		break;
3424 		}
3425 
3426 	case VBLK: {
3427 		const struct bdevsw *bdev;
3428 
3429 		bdev = bdevsw_lookup(dev);
3430 		if (bdev != NULL)
3431 			d_type = (bdev->d_flag & D_TYPEMASK);
3432 
3433 		bvp = vp;
3434 
3435 		break;
3436 		}
3437 
3438 	default:
3439 		break;
3440 	}
3441 
3442 	if (d_type != D_DISK)
3443 		return EINVAL;
3444 
3445 	if (bvpp != NULL)
3446 		*bvpp = bvp;
3447 
3448 	/*
3449 	 * XXX: This is bogus. We should be failing the request
3450 	 * XXX: not only if this specific slice is mounted, but
3451 	 * XXX: if it's on a disk with any other mounted slice.
3452 	 */
3453 	if (vfs_mountedon(bvp))
3454 		return EBUSY;
3455 
3456 	return 0;
3457 }
3458