xref: /netbsd-src/sys/kern/vfs_subr.c (revision 4b71a66d0f279143147d63ebfcfd8a59499a3684)
1 /*	$NetBSD: vfs_subr.c,v 1.345 2008/05/27 17:49:07 ad 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  * External virtual filesystem routines.
71  *
72  * This file contains vfs subroutines which are heavily dependant on
73  * the kernel and are not suitable for standalone use.  Examples include
74  * routines involved vnode and mountpoint management.
75  */
76 
77 #include <sys/cdefs.h>
78 __KERNEL_RCSID(0, "$NetBSD: vfs_subr.c,v 1.345 2008/05/27 17:49:07 ad Exp $");
79 
80 #include "opt_ddb.h"
81 #include "opt_compat_netbsd.h"
82 #include "opt_compat_43.h"
83 
84 #include <sys/param.h>
85 #include <sys/systm.h>
86 #include <sys/proc.h>
87 #include <sys/kernel.h>
88 #include <sys/mount.h>
89 #include <sys/fcntl.h>
90 #include <sys/vnode.h>
91 #include <sys/stat.h>
92 #include <sys/namei.h>
93 #include <sys/ucred.h>
94 #include <sys/buf.h>
95 #include <sys/errno.h>
96 #include <sys/malloc.h>
97 #include <sys/syscallargs.h>
98 #include <sys/device.h>
99 #include <sys/filedesc.h>
100 #include <sys/kauth.h>
101 #include <sys/atomic.h>
102 #include <sys/kthread.h>
103 
104 #include <miscfs/specfs/specdev.h>
105 #include <miscfs/syncfs/syncfs.h>
106 
107 #include <uvm/uvm.h>
108 #include <uvm/uvm_readahead.h>
109 #include <uvm/uvm_ddb.h>
110 
111 #include <sys/sysctl.h>
112 
113 extern int dovfsusermount;	/* 1 => permit any user to mount filesystems */
114 extern int vfs_magiclinks;	/* 1 => expand "magic" symlinks */
115 
116 static vnodelst_t vnode_free_list = TAILQ_HEAD_INITIALIZER(vnode_free_list);
117 static vnodelst_t vnode_hold_list = TAILQ_HEAD_INITIALIZER(vnode_hold_list);
118 static vnodelst_t vrele_list = TAILQ_HEAD_INITIALIZER(vrele_list);
119 
120 static int vrele_pending;
121 static kmutex_t	vrele_lock;
122 static kcondvar_t vrele_cv;
123 static lwp_t *vrele_lwp;
124 
125 static pool_cache_t vnode_cache;
126 
127 MALLOC_DEFINE(M_VNODE, "vnodes", "Dynamically allocated vnodes");
128 
129 /*
130  * Local declarations.
131  */
132 
133 static void vrele_thread(void *);
134 static void insmntque(vnode_t *, struct mount *);
135 static int getdevvp(dev_t, vnode_t **, enum vtype);
136 static vnode_t *getcleanvnode(void);;
137 void vpanic(vnode_t *, const char *);
138 
139 #ifdef DIAGNOSTIC
140 void
141 vpanic(vnode_t *vp, const char *msg)
142 {
143 
144 	vprint(NULL, vp);
145 	panic("%s\n", msg);
146 }
147 #else
148 #define	vpanic(vp, msg)	/* nothing */
149 #endif
150 
151 void
152 vn_init1(void)
153 {
154 
155 	vnode_cache = pool_cache_init(sizeof(struct vnode), 0, 0, 0, "vnodepl",
156 	    NULL, IPL_NONE, NULL, NULL, NULL);
157 	KASSERT(vnode_cache != NULL);
158 
159 	/* Create deferred release thread. */
160 	mutex_init(&vrele_lock, MUTEX_DEFAULT, IPL_NONE);
161 	cv_init(&vrele_cv, "vrele");
162 	if (kthread_create(PRI_VM, KTHREAD_MPSAFE, NULL, vrele_thread,
163 	    NULL, &vrele_lwp, "vrele"))
164 		panic("fork vrele");
165 }
166 
167 int
168 vfs_drainvnodes(long target, struct lwp *l)
169 {
170 
171 	while (numvnodes > target) {
172 		vnode_t *vp;
173 
174 		mutex_enter(&vnode_free_list_lock);
175 		vp = getcleanvnode();
176 		if (vp == NULL)
177 			return EBUSY; /* give up */
178 		ungetnewvnode(vp);
179 	}
180 
181 	return 0;
182 }
183 
184 /*
185  * grab a vnode from freelist and clean it.
186  */
187 vnode_t *
188 getcleanvnode(void)
189 {
190 	vnode_t *vp;
191 	vnodelst_t *listhd;
192 
193 	KASSERT(mutex_owned(&vnode_free_list_lock));
194 
195 retry:
196 	listhd = &vnode_free_list;
197 try_nextlist:
198 	TAILQ_FOREACH(vp, listhd, v_freelist) {
199 		/*
200 		 * It's safe to test v_usecount and v_iflag
201 		 * without holding the interlock here, since
202 		 * these vnodes should never appear on the
203 		 * lists.
204 		 */
205 		if (vp->v_usecount != 0) {
206 			vpanic(vp, "free vnode isn't");
207 		}
208 		if ((vp->v_iflag & VI_CLEAN) != 0) {
209 			vpanic(vp, "clean vnode on freelist");
210 		}
211 		if (vp->v_freelisthd != listhd) {
212 			printf("vnode sez %p, listhd %p\n", vp->v_freelisthd, listhd);
213 			vpanic(vp, "list head mismatch");
214 		}
215 		if (!mutex_tryenter(&vp->v_interlock))
216 			continue;
217 		/*
218 		 * Our lwp might hold the underlying vnode
219 		 * locked, so don't try to reclaim a VI_LAYER
220 		 * node if it's locked.
221 		 */
222 		if ((vp->v_iflag & VI_XLOCK) == 0 &&
223 		    ((vp->v_iflag & VI_LAYER) == 0 || VOP_ISLOCKED(vp) == 0)) {
224 			break;
225 		}
226 		mutex_exit(&vp->v_interlock);
227 	}
228 
229 	if (vp == NULL) {
230 		if (listhd == &vnode_free_list) {
231 			listhd = &vnode_hold_list;
232 			goto try_nextlist;
233 		}
234 		mutex_exit(&vnode_free_list_lock);
235 		return NULL;
236 	}
237 
238 	/* Remove it from the freelist. */
239 	TAILQ_REMOVE(listhd, vp, v_freelist);
240 	vp->v_freelisthd = NULL;
241 	mutex_exit(&vnode_free_list_lock);
242 
243 	/*
244 	 * The vnode is still associated with a file system, so we must
245 	 * clean it out before reusing it.  We need to add a reference
246 	 * before doing this.  If the vnode gains another reference while
247 	 * being cleaned out then we lose - retry.
248 	 */
249 	vp->v_usecount++;
250 	vclean(vp, DOCLOSE);
251 	if (vp->v_usecount == 1) {
252 		/* We're about to dirty it. */
253 		vp->v_iflag &= ~VI_CLEAN;
254 		mutex_exit(&vp->v_interlock);
255 		if (vp->v_type == VBLK || vp->v_type == VCHR) {
256 			spec_node_destroy(vp);
257 		}
258 		vp->v_type = VNON;
259 	} else {
260 		/*
261 		 * Don't return to freelist - the holder of the last
262 		 * reference will destroy it.
263 		 */
264 		KASSERT(vp->v_usecount > 1);
265 		vp->v_usecount--;
266 		mutex_exit(&vp->v_interlock);
267 		mutex_enter(&vnode_free_list_lock);
268 		goto retry;
269 	}
270 
271 	if (vp->v_data != NULL || vp->v_uobj.uo_npages != 0 ||
272 	    !TAILQ_EMPTY(&vp->v_uobj.memq)) {
273 		vpanic(vp, "cleaned vnode isn't");
274 	}
275 	if (vp->v_numoutput != 0) {
276 		vpanic(vp, "clean vnode has pending I/O's");
277 	}
278 	if ((vp->v_iflag & VI_ONWORKLST) != 0) {
279 		vpanic(vp, "clean vnode on syncer list");
280 	}
281 
282 	return vp;
283 }
284 
285 /*
286  * Mark a mount point as busy, and gain a new reference to it.  Used to
287  * prevent the file system from being unmounted during critical sections.
288  *
289  * => The caller must hold a pre-existing reference to the mount.
290  * => Will fail if the file system is being unmounted, or is unmounted.
291  */
292 int
293 vfs_busy(struct mount *mp, struct mount **nextp)
294 {
295 
296 	KASSERT(mp->mnt_refcnt > 0);
297 
298 	if (__predict_false(!rw_tryenter(&mp->mnt_unmounting, RW_READER))) {
299 		if (nextp != NULL) {
300 			KASSERT(mutex_owned(&mountlist_lock));
301 			*nextp = CIRCLEQ_NEXT(mp, mnt_list);
302 		}
303 		return EBUSY;
304 	}
305 	if (__predict_false((mp->mnt_iflag & IMNT_GONE) != 0)) {
306 		rw_exit(&mp->mnt_unmounting);
307 		if (nextp != NULL) {
308 			KASSERT(mutex_owned(&mountlist_lock));
309 			*nextp = CIRCLEQ_NEXT(mp, mnt_list);
310 		}
311 		return ENOENT;
312 	}
313 	if (nextp != NULL) {
314 		mutex_exit(&mountlist_lock);
315 	}
316 	atomic_inc_uint(&mp->mnt_refcnt);
317 	return 0;
318 }
319 
320 /*
321  * Unbusy a busy filesystem.
322  *
323  * => If keepref is true, preserve reference added by vfs_busy().
324  * => If nextp != NULL, acquire mountlist_lock.
325  */
326 void
327 vfs_unbusy(struct mount *mp, bool keepref, struct mount **nextp)
328 {
329 
330 	KASSERT(mp->mnt_refcnt > 0);
331 
332 	if (nextp != NULL) {
333 		mutex_enter(&mountlist_lock);
334 	}
335 	rw_exit(&mp->mnt_unmounting);
336 	if (!keepref) {
337 		vfs_destroy(mp);
338 	}
339 	if (nextp != NULL) {
340 		KASSERT(mutex_owned(&mountlist_lock));
341 		*nextp = CIRCLEQ_NEXT(mp, mnt_list);
342 	}
343 }
344 
345 /*
346  * Lookup a filesystem type, and if found allocate and initialize
347  * a mount structure for it.
348  *
349  * Devname is usually updated by mount(8) after booting.
350  */
351 int
352 vfs_rootmountalloc(const char *fstypename, const char *devname,
353     struct mount **mpp)
354 {
355 	struct vfsops *vfsp = NULL;
356 	struct mount *mp;
357 
358 	mutex_enter(&vfs_list_lock);
359 	LIST_FOREACH(vfsp, &vfs_list, vfs_list)
360 		if (!strncmp(vfsp->vfs_name, fstypename,
361 		    sizeof(mp->mnt_stat.f_fstypename)))
362 			break;
363 	if (vfsp == NULL) {
364 		mutex_exit(&vfs_list_lock);
365 		return (ENODEV);
366 	}
367 	vfsp->vfs_refcount++;
368 	mutex_exit(&vfs_list_lock);
369 
370 	mp = kmem_zalloc(sizeof(*mp), KM_SLEEP);
371 	if (mp == NULL)
372 		return ENOMEM;
373 	mp->mnt_refcnt = 1;
374 	rw_init(&mp->mnt_unmounting);
375 	mutex_init(&mp->mnt_updating, MUTEX_DEFAULT, IPL_NONE);
376 	mutex_init(&mp->mnt_renamelock, MUTEX_DEFAULT, IPL_NONE);
377 	(void)vfs_busy(mp, NULL);
378 	TAILQ_INIT(&mp->mnt_vnodelist);
379 	mp->mnt_op = vfsp;
380 	mp->mnt_flag = MNT_RDONLY;
381 	mp->mnt_vnodecovered = NULL;
382 	(void)strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfs_name,
383 	    sizeof(mp->mnt_stat.f_fstypename));
384 	mp->mnt_stat.f_mntonname[0] = '/';
385 	mp->mnt_stat.f_mntonname[1] = '\0';
386 	mp->mnt_stat.f_mntfromname[sizeof(mp->mnt_stat.f_mntfromname) - 1] =
387 	    '\0';
388 	(void)copystr(devname, mp->mnt_stat.f_mntfromname,
389 	    sizeof(mp->mnt_stat.f_mntfromname) - 1, 0);
390 	mount_initspecific(mp);
391 	*mpp = mp;
392 	return (0);
393 }
394 
395 /*
396  * Routines having to do with the management of the vnode table.
397  */
398 extern int (**dead_vnodeop_p)(void *);
399 
400 /*
401  * Return the next vnode from the free list.
402  */
403 int
404 getnewvnode(enum vtagtype tag, struct mount *mp, int (**vops)(void *),
405 	    vnode_t **vpp)
406 {
407 	struct uvm_object *uobj;
408 	static int toggle;
409 	vnode_t *vp;
410 	int error = 0, tryalloc;
411 
412  try_again:
413 	if (mp != NULL) {
414 		/*
415 		 * Mark filesystem busy while we're creating a
416 		 * vnode.  If unmount is in progress, this will
417 		 * fail.
418 		 */
419 		error = vfs_busy(mp, NULL);
420 		if (error)
421 			return error;
422 	}
423 
424 	/*
425 	 * We must choose whether to allocate a new vnode or recycle an
426 	 * existing one. The criterion for allocating a new one is that
427 	 * the total number of vnodes is less than the number desired or
428 	 * there are no vnodes on either free list. Generally we only
429 	 * want to recycle vnodes that have no buffers associated with
430 	 * them, so we look first on the vnode_free_list. If it is empty,
431 	 * we next consider vnodes with referencing buffers on the
432 	 * vnode_hold_list. The toggle ensures that half the time we
433 	 * will use a buffer from the vnode_hold_list, and half the time
434 	 * we will allocate a new one unless the list has grown to twice
435 	 * the desired size. We are reticent to recycle vnodes from the
436 	 * vnode_hold_list because we will lose the identity of all its
437 	 * referencing buffers.
438 	 */
439 
440 	vp = NULL;
441 
442 	mutex_enter(&vnode_free_list_lock);
443 
444 	toggle ^= 1;
445 	if (numvnodes > 2 * desiredvnodes)
446 		toggle = 0;
447 
448 	tryalloc = numvnodes < desiredvnodes ||
449 	    (TAILQ_FIRST(&vnode_free_list) == NULL &&
450 	     (TAILQ_FIRST(&vnode_hold_list) == NULL || toggle));
451 
452 	if (tryalloc) {
453 		numvnodes++;
454 		mutex_exit(&vnode_free_list_lock);
455 		if ((vp = vnalloc(NULL)) == NULL) {
456 			mutex_enter(&vnode_free_list_lock);
457 			numvnodes--;
458 		} else
459 			vp->v_usecount = 1;
460 	}
461 
462 	if (vp == NULL) {
463 		vp = getcleanvnode();
464 		if (vp == NULL) {
465 			if (mp != NULL) {
466 				vfs_unbusy(mp, false, NULL);
467 			}
468 			if (tryalloc) {
469 				printf("WARNING: unable to allocate new "
470 				    "vnode, retrying...\n");
471 				kpause("newvn", false, hz, NULL);
472 				goto try_again;
473 			}
474 			tablefull("vnode", "increase kern.maxvnodes or NVNODE");
475 			*vpp = 0;
476 			return (ENFILE);
477 		}
478 		vp->v_iflag = 0;
479 		vp->v_vflag = 0;
480 		vp->v_uflag = 0;
481 		vp->v_socket = NULL;
482 	}
483 
484 	KASSERT(vp->v_usecount == 1);
485 	KASSERT(vp->v_freelisthd == NULL);
486 	KASSERT(LIST_EMPTY(&vp->v_nclist));
487 	KASSERT(LIST_EMPTY(&vp->v_dnclist));
488 
489 	vp->v_type = VNON;
490 	vp->v_vnlock = &vp->v_lock;
491 	vp->v_tag = tag;
492 	vp->v_op = vops;
493 	insmntque(vp, mp);
494 	*vpp = vp;
495 	vp->v_data = 0;
496 
497 	/*
498 	 * initialize uvm_object within vnode.
499 	 */
500 
501 	uobj = &vp->v_uobj;
502 	KASSERT(uobj->pgops == &uvm_vnodeops);
503 	KASSERT(uobj->uo_npages == 0);
504 	KASSERT(TAILQ_FIRST(&uobj->memq) == NULL);
505 	vp->v_size = vp->v_writesize = VSIZENOTSET;
506 
507 	if (mp != NULL) {
508 		if ((mp->mnt_iflag & IMNT_MPSAFE) != 0)
509 			vp->v_vflag |= VV_MPSAFE;
510 		vfs_unbusy(mp, true, NULL);
511 	}
512 
513 	return (0);
514 }
515 
516 /*
517  * This is really just the reverse of getnewvnode(). Needed for
518  * VFS_VGET functions who may need to push back a vnode in case
519  * of a locking race.
520  */
521 void
522 ungetnewvnode(vnode_t *vp)
523 {
524 
525 	KASSERT(vp->v_usecount == 1);
526 	KASSERT(vp->v_data == NULL);
527 	KASSERT(vp->v_freelisthd == NULL);
528 
529 	mutex_enter(&vp->v_interlock);
530 	vp->v_iflag |= VI_CLEAN;
531 	vrelel(vp, 0);
532 }
533 
534 /*
535  * Allocate a new, uninitialized vnode.  If 'mp' is non-NULL, this is a
536  * marker vnode and we are prepared to wait for the allocation.
537  */
538 vnode_t *
539 vnalloc(struct mount *mp)
540 {
541 	vnode_t *vp;
542 
543 	vp = pool_cache_get(vnode_cache, (mp != NULL ? PR_WAITOK : PR_NOWAIT));
544 	if (vp == NULL) {
545 		return NULL;
546 	}
547 
548 	memset(vp, 0, sizeof(*vp));
549 	UVM_OBJ_INIT(&vp->v_uobj, &uvm_vnodeops, 0);
550 	cv_init(&vp->v_cv, "vnode");
551 	/*
552 	 * done by memset() above.
553 	 *	LIST_INIT(&vp->v_nclist);
554 	 *	LIST_INIT(&vp->v_dnclist);
555 	 */
556 
557 	if (mp != NULL) {
558 		vp->v_mount = mp;
559 		vp->v_type = VBAD;
560 		vp->v_iflag = VI_MARKER;
561 	} else {
562 		rw_init(&vp->v_lock.vl_lock);
563 	}
564 
565 	return vp;
566 }
567 
568 /*
569  * Free an unused, unreferenced vnode.
570  */
571 void
572 vnfree(vnode_t *vp)
573 {
574 
575 	KASSERT(vp->v_usecount == 0);
576 
577 	if ((vp->v_iflag & VI_MARKER) == 0) {
578 		rw_destroy(&vp->v_lock.vl_lock);
579 		mutex_enter(&vnode_free_list_lock);
580 		numvnodes--;
581 		mutex_exit(&vnode_free_list_lock);
582 	}
583 
584 	UVM_OBJ_DESTROY(&vp->v_uobj);
585 	cv_destroy(&vp->v_cv);
586 	pool_cache_put(vnode_cache, vp);
587 }
588 
589 /*
590  * Remove a vnode from its freelist.
591  */
592 static inline void
593 vremfree(vnode_t *vp)
594 {
595 
596 	KASSERT(mutex_owned(&vp->v_interlock));
597 	KASSERT(vp->v_usecount == 0);
598 
599 	/*
600 	 * Note that the reference count must not change until
601 	 * the vnode is removed.
602 	 */
603 	mutex_enter(&vnode_free_list_lock);
604 	if (vp->v_holdcnt > 0) {
605 		KASSERT(vp->v_freelisthd == &vnode_hold_list);
606 	} else {
607 		KASSERT(vp->v_freelisthd == &vnode_free_list);
608 	}
609 	TAILQ_REMOVE(vp->v_freelisthd, vp, v_freelist);
610 	vp->v_freelisthd = NULL;
611 	mutex_exit(&vnode_free_list_lock);
612 }
613 
614 /*
615  * Move a vnode from one mount queue to another.
616  */
617 static void
618 insmntque(vnode_t *vp, struct mount *mp)
619 {
620 	struct mount *omp;
621 
622 #ifdef DIAGNOSTIC
623 	if ((mp != NULL) &&
624 	    (mp->mnt_iflag & IMNT_UNMOUNT) &&
625 	    !(mp->mnt_flag & MNT_SOFTDEP) &&
626 	    vp->v_tag != VT_VFS) {
627 		panic("insmntque into dying filesystem");
628 	}
629 #endif
630 
631 	mutex_enter(&mntvnode_lock);
632 	/*
633 	 * Delete from old mount point vnode list, if on one.
634 	 */
635 	if ((omp = vp->v_mount) != NULL)
636 		TAILQ_REMOVE(&vp->v_mount->mnt_vnodelist, vp, v_mntvnodes);
637 	/*
638 	 * Insert into list of vnodes for the new mount point, if
639 	 * available.  The caller must take a reference on the mount
640 	 * structure and donate to the vnode.
641 	 */
642 	if ((vp->v_mount = mp) != NULL)
643 		TAILQ_INSERT_TAIL(&mp->mnt_vnodelist, vp, v_mntvnodes);
644 	mutex_exit(&mntvnode_lock);
645 
646 	if (omp != NULL) {
647 		/* Release reference to old mount. */
648 		vfs_destroy(omp);
649 	}
650 }
651 
652 /*
653  * Create a vnode for a block device.
654  * Used for root filesystem and swap areas.
655  * Also used for memory file system special devices.
656  */
657 int
658 bdevvp(dev_t dev, vnode_t **vpp)
659 {
660 
661 	return (getdevvp(dev, vpp, VBLK));
662 }
663 
664 /*
665  * Create a vnode for a character device.
666  * Used for kernfs and some console handling.
667  */
668 int
669 cdevvp(dev_t dev, vnode_t **vpp)
670 {
671 
672 	return (getdevvp(dev, vpp, VCHR));
673 }
674 
675 /*
676  * Create a vnode for a device.
677  * Used by bdevvp (block device) for root file system etc.,
678  * and by cdevvp (character device) for console and kernfs.
679  */
680 static int
681 getdevvp(dev_t dev, vnode_t **vpp, enum vtype type)
682 {
683 	vnode_t *vp;
684 	vnode_t *nvp;
685 	int error;
686 
687 	if (dev == NODEV) {
688 		*vpp = NULL;
689 		return (0);
690 	}
691 	error = getnewvnode(VT_NON, NULL, spec_vnodeop_p, &nvp);
692 	if (error) {
693 		*vpp = NULL;
694 		return (error);
695 	}
696 	vp = nvp;
697 	vp->v_type = type;
698 	vp->v_vflag |= VV_MPSAFE;
699 	uvm_vnp_setsize(vp, 0);
700 	spec_node_init(vp, dev);
701 	*vpp = vp;
702 	return (0);
703 }
704 
705 /*
706  * Grab a particular vnode from the free list, increment its
707  * reference count and lock it. If the vnode lock bit is set the
708  * vnode is being eliminated in vgone. In that case, we can not
709  * grab the vnode, so the process is awakened when the transition is
710  * completed, and an error returned to indicate that the vnode is no
711  * longer usable (possibly having been changed to a new file system type).
712  */
713 int
714 vget(vnode_t *vp, int flags)
715 {
716 	int error;
717 
718 	KASSERT((vp->v_iflag & VI_MARKER) == 0);
719 
720 	if ((flags & LK_INTERLOCK) == 0)
721 		mutex_enter(&vp->v_interlock);
722 
723 	/*
724 	 * Before adding a reference, we must remove the vnode
725 	 * from its freelist.
726 	 */
727 	if (vp->v_usecount == 0) {
728 		vremfree(vp);
729 	}
730 	if (++vp->v_usecount == 0) {
731 		vpanic(vp, "vget: usecount overflow");
732 	}
733 
734 	/*
735 	 * If the vnode is in the process of being cleaned out for
736 	 * another use, we wait for the cleaning to finish and then
737 	 * return failure.  Cleaning is determined by checking if
738 	 * the VI_XLOCK or VI_FREEING flags are set.
739 	 */
740 	if ((vp->v_iflag & (VI_XLOCK | VI_FREEING)) != 0) {
741 		if ((flags & LK_NOWAIT) != 0) {
742 			vrelel(vp, 0);
743 			return EBUSY;
744 		}
745 		vwait(vp, VI_XLOCK | VI_FREEING);
746 		vrelel(vp, 0);
747 		return ENOENT;
748 	}
749 	if (flags & LK_TYPE_MASK) {
750 		error = vn_lock(vp, flags | LK_INTERLOCK);
751 		if (error != 0) {
752 			vrele(vp);
753 		}
754 		return error;
755 	}
756 	mutex_exit(&vp->v_interlock);
757 	return 0;
758 }
759 
760 /*
761  * vput(), just unlock and vrele()
762  */
763 void
764 vput(vnode_t *vp)
765 {
766 
767 	KASSERT((vp->v_iflag & VI_MARKER) == 0);
768 
769 	VOP_UNLOCK(vp, 0);
770 	vrele(vp);
771 }
772 
773 /*
774  * Vnode release.  If reference count drops to zero, call inactive
775  * routine and either return to freelist or free to the pool.
776  */
777 void
778 vrelel(vnode_t *vp, int flags)
779 {
780 	bool recycle, defer;
781 	int error;
782 
783 	KASSERT(mutex_owned(&vp->v_interlock));
784 	KASSERT((vp->v_iflag & VI_MARKER) == 0);
785 	KASSERT(vp->v_freelisthd == NULL);
786 
787 	if (vp->v_op == dead_vnodeop_p && (vp->v_iflag & VI_CLEAN) == 0) {
788 		vpanic(vp, "dead but not clean");
789 	}
790 
791 	/*
792 	 * If not the last reference, just drop the reference count
793 	 * and unlock.
794 	 */
795 	if (vp->v_usecount > 1) {
796 		vp->v_usecount--;
797 		vp->v_iflag |= VI_INACTREDO;
798 		mutex_exit(&vp->v_interlock);
799 		return;
800 	}
801 	if (vp->v_usecount <= 0 || vp->v_writecount != 0) {
802 		vpanic(vp, "vput: bad ref count");
803 	}
804 
805 	/*
806 	 * If not clean, deactivate the vnode, but preserve
807 	 * our reference across the call to VOP_INACTIVE().
808 	 */
809  retry:
810 	if ((vp->v_iflag & VI_CLEAN) == 0) {
811 		recycle = false;
812 		/*
813 		 * XXX This ugly block can be largely eliminated if
814 		 * locking is pushed down into the file systems.
815 		 */
816 		if (curlwp == uvm.pagedaemon_lwp) {
817 			/* The pagedaemon can't wait around; defer. */
818 			defer = true;
819 		} else if (curlwp == vrele_lwp) {
820 			/* We have to try harder. */
821 			vp->v_iflag &= ~VI_INACTREDO;
822 			error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK |
823 			    LK_RETRY);
824 			if (error != 0) {
825 				/* XXX */
826 				vpanic(vp, "vrele: unable to lock %p");
827 			}
828 			defer = false;
829 		} else if ((vp->v_iflag & VI_LAYER) != 0) {
830 			/*
831 			 * Acquiring the stack's lock in vclean() even
832 			 * for an honest vput/vrele is dangerous because
833 			 * our caller may hold other vnode locks; defer.
834 			 */
835 			defer = true;
836 		} else {
837 			/* If we can't acquire the lock, then defer. */
838 			vp->v_iflag &= ~VI_INACTREDO;
839 			error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK |
840 			    LK_NOWAIT);
841 			if (error != 0) {
842 				defer = true;
843 				mutex_enter(&vp->v_interlock);
844 			} else {
845 				defer = false;
846 			}
847 		}
848 
849 		if (defer) {
850 			/*
851 			 * Defer reclaim to the kthread; it's not safe to
852 			 * clean it here.  We donate it our last reference.
853 			 */
854 			KASSERT(mutex_owned(&vp->v_interlock));
855 			KASSERT((vp->v_iflag & VI_INACTPEND) == 0);
856 			vp->v_iflag |= VI_INACTPEND;
857 			mutex_enter(&vrele_lock);
858 			TAILQ_INSERT_TAIL(&vrele_list, vp, v_freelist);
859 			if (++vrele_pending > (desiredvnodes >> 8))
860 				cv_signal(&vrele_cv);
861 			mutex_exit(&vrele_lock);
862 			mutex_exit(&vp->v_interlock);
863 			return;
864 		}
865 
866 #ifdef DIAGNOSTIC
867 		if ((vp->v_type == VBLK || vp->v_type == VCHR) &&
868 		    vp->v_specnode != NULL && vp->v_specnode->sn_opencnt != 0) {
869 			vprint("vrelel: missing VOP_CLOSE()", vp);
870 		}
871 #endif
872 
873 		/*
874 		 * The vnode can gain another reference while being
875 		 * deactivated.  If VOP_INACTIVE() indicates that
876 		 * the described file has been deleted, then recycle
877 		 * the vnode irrespective of additional references.
878 		 * Another thread may be waiting to re-use the on-disk
879 		 * inode.
880 		 *
881 		 * Note that VOP_INACTIVE() will drop the vnode lock.
882 		 */
883 		VOP_INACTIVE(vp, &recycle);
884 		mutex_enter(&vp->v_interlock);
885 		if (!recycle) {
886 			if (vp->v_usecount > 1) {
887 				vp->v_usecount--;
888 				mutex_exit(&vp->v_interlock);
889 				return;
890 			}
891 
892 			/*
893 			 * If we grew another reference while
894 			 * VOP_INACTIVE() was underway, retry.
895 			 */
896 			if ((vp->v_iflag & VI_INACTREDO) != 0) {
897 				goto retry;
898 			}
899 		}
900 
901 		/* Take care of space accounting. */
902 		if (vp->v_iflag & VI_EXECMAP) {
903 			atomic_add_int(&uvmexp.execpages,
904 			    -vp->v_uobj.uo_npages);
905 			atomic_add_int(&uvmexp.filepages,
906 			    vp->v_uobj.uo_npages);
907 		}
908 		vp->v_iflag &= ~(VI_TEXT|VI_EXECMAP|VI_WRMAP|VI_MAPPED);
909 		vp->v_vflag &= ~VV_MAPPED;
910 
911 		/*
912 		 * Recycle the vnode if the file is now unused (unlinked),
913 		 * otherwise just free it.
914 		 */
915 		if (recycle) {
916 			vclean(vp, DOCLOSE);
917 		}
918 		KASSERT(vp->v_usecount > 0);
919 	}
920 
921 	if (--vp->v_usecount != 0) {
922 		/* Gained another reference while being reclaimed. */
923 		mutex_exit(&vp->v_interlock);
924 		return;
925 	}
926 
927 	if ((vp->v_iflag & VI_CLEAN) != 0) {
928 		/*
929 		 * It's clean so destroy it.  It isn't referenced
930 		 * anywhere since it has been reclaimed.
931 		 */
932 		KASSERT(vp->v_holdcnt == 0);
933 		KASSERT(vp->v_writecount == 0);
934 		mutex_exit(&vp->v_interlock);
935 		insmntque(vp, NULL);
936 		if (vp->v_type == VBLK || vp->v_type == VCHR) {
937 			spec_node_destroy(vp);
938 		}
939 		vnfree(vp);
940 	} else {
941 		/*
942 		 * Otherwise, put it back onto the freelist.  It
943 		 * can't be destroyed while still associated with
944 		 * a file system.
945 		 */
946 		mutex_enter(&vnode_free_list_lock);
947 		if (vp->v_holdcnt > 0) {
948 			vp->v_freelisthd = &vnode_hold_list;
949 		} else {
950 			vp->v_freelisthd = &vnode_free_list;
951 		}
952 		TAILQ_INSERT_TAIL(vp->v_freelisthd, vp, v_freelist);
953 		mutex_exit(&vnode_free_list_lock);
954 		mutex_exit(&vp->v_interlock);
955 	}
956 }
957 
958 void
959 vrele(vnode_t *vp)
960 {
961 
962 	KASSERT((vp->v_iflag & VI_MARKER) == 0);
963 
964 	mutex_enter(&vp->v_interlock);
965 	vrelel(vp, 0);
966 }
967 
968 static void
969 vrele_thread(void *cookie)
970 {
971 	vnode_t *vp;
972 
973 	for (;;) {
974 		mutex_enter(&vrele_lock);
975 		while (TAILQ_EMPTY(&vrele_list)) {
976 			cv_timedwait(&vrele_cv, &vrele_lock, hz);
977 		}
978 		vp = TAILQ_FIRST(&vrele_list);
979 		TAILQ_REMOVE(&vrele_list, vp, v_freelist);
980 		vrele_pending--;
981 		mutex_exit(&vrele_lock);
982 
983 		/*
984 		 * If not the last reference, then ignore the vnode
985 		 * and look for more work.
986 		 */
987 		mutex_enter(&vp->v_interlock);
988 		KASSERT((vp->v_iflag & VI_INACTPEND) != 0);
989 		vp->v_iflag &= ~VI_INACTPEND;
990 		if (vp->v_usecount > 1) {
991 			vp->v_usecount--;
992 			mutex_exit(&vp->v_interlock);
993 			continue;
994 		}
995 		vrelel(vp, 0);
996 	}
997 }
998 
999 /*
1000  * Page or buffer structure gets a reference.
1001  * Called with v_interlock held.
1002  */
1003 void
1004 vholdl(vnode_t *vp)
1005 {
1006 
1007 	KASSERT(mutex_owned(&vp->v_interlock));
1008 	KASSERT((vp->v_iflag & VI_MARKER) == 0);
1009 
1010 	if (vp->v_holdcnt++ == 0 && vp->v_usecount == 0) {
1011 		mutex_enter(&vnode_free_list_lock);
1012 		KASSERT(vp->v_freelisthd == &vnode_free_list);
1013 		TAILQ_REMOVE(vp->v_freelisthd, vp, v_freelist);
1014 		vp->v_freelisthd = &vnode_hold_list;
1015 		TAILQ_INSERT_TAIL(vp->v_freelisthd, vp, v_freelist);
1016 		mutex_exit(&vnode_free_list_lock);
1017 	}
1018 }
1019 
1020 /*
1021  * Page or buffer structure frees a reference.
1022  * Called with v_interlock held.
1023  */
1024 void
1025 holdrelel(vnode_t *vp)
1026 {
1027 
1028 	KASSERT(mutex_owned(&vp->v_interlock));
1029 	KASSERT((vp->v_iflag & VI_MARKER) == 0);
1030 
1031 	if (vp->v_holdcnt <= 0) {
1032 		vpanic(vp, "holdrelel: holdcnt vp %p");
1033 	}
1034 
1035 	vp->v_holdcnt--;
1036 	if (vp->v_holdcnt == 0 && vp->v_usecount == 0) {
1037 		mutex_enter(&vnode_free_list_lock);
1038 		KASSERT(vp->v_freelisthd == &vnode_hold_list);
1039 		TAILQ_REMOVE(vp->v_freelisthd, vp, v_freelist);
1040 		vp->v_freelisthd = &vnode_free_list;
1041 		TAILQ_INSERT_TAIL(vp->v_freelisthd, vp, v_freelist);
1042 		mutex_exit(&vnode_free_list_lock);
1043 	}
1044 }
1045 
1046 /*
1047  * Vnode reference, where a reference is already held by some other
1048  * object (for example, a file structure).
1049  */
1050 void
1051 vref(vnode_t *vp)
1052 {
1053 
1054 	KASSERT((vp->v_iflag & VI_MARKER) == 0);
1055 
1056 	mutex_enter(&vp->v_interlock);
1057 	if (vp->v_usecount <= 0) {
1058 		vpanic(vp, "vref used where vget required");
1059 	}
1060 	if (++vp->v_usecount == 0) {
1061 		vpanic(vp, "vref: usecount overflow");
1062 	}
1063 	mutex_exit(&vp->v_interlock);
1064 }
1065 
1066 /*
1067  * Remove any vnodes in the vnode table belonging to mount point mp.
1068  *
1069  * If FORCECLOSE is not specified, there should not be any active ones,
1070  * return error if any are found (nb: this is a user error, not a
1071  * system error). If FORCECLOSE is specified, detach any active vnodes
1072  * that are found.
1073  *
1074  * If WRITECLOSE is set, only flush out regular file vnodes open for
1075  * writing.
1076  *
1077  * SKIPSYSTEM causes any vnodes marked V_SYSTEM to be skipped.
1078  */
1079 #ifdef DEBUG
1080 int busyprt = 0;	/* print out busy vnodes */
1081 struct ctldebug debug1 = { "busyprt", &busyprt };
1082 #endif
1083 
1084 static vnode_t *
1085 vflushnext(vnode_t *mvp, int *when)
1086 {
1087 
1088 	if (hardclock_ticks > *when) {
1089 		mutex_exit(&mntvnode_lock);
1090 		yield();
1091 		mutex_enter(&mntvnode_lock);
1092 		*when = hardclock_ticks + hz / 10;
1093 	}
1094 
1095 	return vunmark(mvp);
1096 }
1097 
1098 int
1099 vflush(struct mount *mp, vnode_t *skipvp, int flags)
1100 {
1101 	vnode_t *vp, *mvp;
1102 	int busy = 0, when = 0;
1103 
1104 	/* Allocate a marker vnode. */
1105 	if ((mvp = vnalloc(mp)) == NULL)
1106 		return (ENOMEM);
1107 
1108 	mutex_enter(&mntvnode_lock);
1109 	/*
1110 	 * NOTE: not using the TAILQ_FOREACH here since in this loop vgone()
1111 	 * and vclean() are called
1112 	 */
1113 	for (vp = TAILQ_FIRST(&mp->mnt_vnodelist); vp != NULL;
1114 	    vp = vflushnext(mvp, &when)) {
1115 		vmark(mvp, vp);
1116 		if (vp->v_mount != mp || vismarker(vp))
1117 			continue;
1118 		/*
1119 		 * Skip over a selected vnode.
1120 		 */
1121 		if (vp == skipvp)
1122 			continue;
1123 		mutex_enter(&vp->v_interlock);
1124 		/*
1125 		 * Ignore clean but still referenced vnodes.
1126 		 */
1127 		if ((vp->v_iflag & VI_CLEAN) != 0) {
1128 			mutex_exit(&vp->v_interlock);
1129 			continue;
1130 		}
1131 		/*
1132 		 * Skip over a vnodes marked VSYSTEM.
1133 		 */
1134 		if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
1135 			mutex_exit(&vp->v_interlock);
1136 			continue;
1137 		}
1138 		/*
1139 		 * If WRITECLOSE is set, only flush out regular file
1140 		 * vnodes open for writing.
1141 		 */
1142 		if ((flags & WRITECLOSE) &&
1143 		    (vp->v_writecount == 0 || vp->v_type != VREG)) {
1144 			mutex_exit(&vp->v_interlock);
1145 			continue;
1146 		}
1147 		/*
1148 		 * With v_usecount == 0, all we need to do is clear
1149 		 * out the vnode data structures and we are done.
1150 		 */
1151 		if (vp->v_usecount == 0) {
1152 			mutex_exit(&mntvnode_lock);
1153 			vremfree(vp);
1154 			vp->v_usecount++;
1155 			vclean(vp, DOCLOSE);
1156 			vrelel(vp, 0);
1157 			mutex_enter(&mntvnode_lock);
1158 			continue;
1159 		}
1160 		/*
1161 		 * If FORCECLOSE is set, forcibly close the vnode.
1162 		 * For block or character devices, revert to an
1163 		 * anonymous device.  For all other files, just
1164 		 * kill them.
1165 		 */
1166 		if (flags & FORCECLOSE) {
1167 			mutex_exit(&mntvnode_lock);
1168 			vp->v_usecount++;
1169 			if (vp->v_type != VBLK && vp->v_type != VCHR) {
1170 				vclean(vp, DOCLOSE);
1171 				vrelel(vp, 0);
1172 			} else {
1173 				vclean(vp, 0);
1174 				vp->v_op = spec_vnodeop_p; /* XXXSMP */
1175 				mutex_exit(&vp->v_interlock);
1176 				/*
1177 				 * The vnode isn't clean, but still resides
1178 				 * on the mount list.  Remove it. XXX This
1179 				 * is a bit dodgy.
1180 				 */
1181 				insmntque(vp, NULL);
1182 				vrele(vp);
1183 			}
1184 			mutex_enter(&mntvnode_lock);
1185 			continue;
1186 		}
1187 #ifdef DEBUG
1188 		if (busyprt)
1189 			vprint("vflush: busy vnode", vp);
1190 #endif
1191 		mutex_exit(&vp->v_interlock);
1192 		busy++;
1193 	}
1194 	mutex_exit(&mntvnode_lock);
1195 	vnfree(mvp);
1196 	if (busy)
1197 		return (EBUSY);
1198 	return (0);
1199 }
1200 
1201 /*
1202  * Disassociate the underlying file system from a vnode.
1203  *
1204  * Must be called with the interlock held, and will return with it held.
1205  */
1206 void
1207 vclean(vnode_t *vp, int flags)
1208 {
1209 	lwp_t *l = curlwp;
1210 	bool recycle, active;
1211 	int error;
1212 
1213 	KASSERT(mutex_owned(&vp->v_interlock));
1214 	KASSERT((vp->v_iflag & VI_MARKER) == 0);
1215 	KASSERT(vp->v_usecount != 0);
1216 
1217 	/* If cleaning is already in progress wait until done and return. */
1218 	if (vp->v_iflag & VI_XLOCK) {
1219 		vwait(vp, VI_XLOCK);
1220 		return;
1221 	}
1222 
1223 	/* If already clean, nothing to do. */
1224 	if ((vp->v_iflag & VI_CLEAN) != 0) {
1225 		return;
1226 	}
1227 
1228 	/*
1229 	 * Prevent the vnode from being recycled or brought into use
1230 	 * while we clean it out.
1231 	 */
1232 	vp->v_iflag |= VI_XLOCK;
1233 	if (vp->v_iflag & VI_EXECMAP) {
1234 		atomic_add_int(&uvmexp.execpages, -vp->v_uobj.uo_npages);
1235 		atomic_add_int(&uvmexp.filepages, vp->v_uobj.uo_npages);
1236 	}
1237 	vp->v_iflag &= ~(VI_TEXT|VI_EXECMAP);
1238 	active = (vp->v_usecount > 1);
1239 
1240 	/* XXXAD should not lock vnode under layer */
1241 	VOP_LOCK(vp, LK_EXCLUSIVE | LK_INTERLOCK);
1242 
1243 	/*
1244 	 * Clean out any cached data associated with the vnode.
1245 	 * If purging an active vnode, it must be closed and
1246 	 * deactivated before being reclaimed. Note that the
1247 	 * VOP_INACTIVE will unlock the vnode.
1248 	 */
1249 	if (flags & DOCLOSE) {
1250 		error = vinvalbuf(vp, V_SAVE, NOCRED, l, 0, 0);
1251 		if (error != 0)
1252 			error = vinvalbuf(vp, 0, NOCRED, l, 0, 0);
1253 		KASSERT(error == 0);
1254 		KASSERT((vp->v_iflag & VI_ONWORKLST) == 0);
1255 		if (active && (vp->v_type == VBLK || vp->v_type == VCHR)) {
1256 			 spec_node_revoke(vp);
1257 		}
1258 	}
1259 	if (active) {
1260 		VOP_INACTIVE(vp, &recycle);
1261 	} else {
1262 		/*
1263 		 * Any other processes trying to obtain this lock must first
1264 		 * wait for VI_XLOCK to clear, then call the new lock operation.
1265 		 */
1266 		VOP_UNLOCK(vp, 0);
1267 	}
1268 
1269 	/* Disassociate the underlying file system from the vnode. */
1270 	if (VOP_RECLAIM(vp)) {
1271 		vpanic(vp, "vclean: cannot reclaim");
1272 	}
1273 
1274 	KASSERT(vp->v_uobj.uo_npages == 0);
1275 	if (vp->v_type == VREG && vp->v_ractx != NULL) {
1276 		uvm_ra_freectx(vp->v_ractx);
1277 		vp->v_ractx = NULL;
1278 	}
1279 	cache_purge(vp);
1280 
1281 	/* Done with purge, notify sleepers of the grim news. */
1282 	vp->v_op = dead_vnodeop_p;
1283 	vp->v_tag = VT_NON;
1284 	mutex_enter(&vp->v_interlock);
1285 	vp->v_vnlock = &vp->v_lock;
1286 	KNOTE(&vp->v_klist, NOTE_REVOKE);
1287 	vp->v_iflag &= ~(VI_XLOCK | VI_FREEING);
1288 	vp->v_vflag &= ~VV_LOCKSWORK;
1289 	if ((flags & DOCLOSE) != 0) {
1290 		vp->v_iflag |= VI_CLEAN;
1291 	}
1292 	cv_broadcast(&vp->v_cv);
1293 
1294 	KASSERT((vp->v_iflag & VI_ONWORKLST) == 0);
1295 }
1296 
1297 /*
1298  * Recycle an unused vnode to the front of the free list.
1299  * Release the passed interlock if the vnode will be recycled.
1300  */
1301 int
1302 vrecycle(vnode_t *vp, kmutex_t *inter_lkp, struct lwp *l)
1303 {
1304 
1305 	KASSERT((vp->v_iflag & VI_MARKER) == 0);
1306 
1307 	mutex_enter(&vp->v_interlock);
1308 	if (vp->v_usecount != 0) {
1309 		mutex_exit(&vp->v_interlock);
1310 		return (0);
1311 	}
1312 	if (inter_lkp)
1313 		mutex_exit(inter_lkp);
1314 	vremfree(vp);
1315 	vp->v_usecount++;
1316 	vclean(vp, DOCLOSE);
1317 	vrelel(vp, 0);
1318 	return (1);
1319 }
1320 
1321 /*
1322  * Eliminate all activity associated with a vnode in preparation for
1323  * reuse.  Drops a reference from the vnode.
1324  */
1325 void
1326 vgone(vnode_t *vp)
1327 {
1328 
1329 	mutex_enter(&vp->v_interlock);
1330 	vclean(vp, DOCLOSE);
1331 	vrelel(vp, 0);
1332 }
1333 
1334 /*
1335  * Lookup a vnode by device number.
1336  */
1337 int
1338 vfinddev(dev_t dev, enum vtype type, vnode_t **vpp)
1339 {
1340 	vnode_t *vp;
1341 	int rc = 0;
1342 
1343 	mutex_enter(&specfs_lock);
1344 	for (vp = specfs_hash[SPECHASH(dev)]; vp; vp = vp->v_specnext) {
1345 		if (dev != vp->v_rdev || type != vp->v_type)
1346 			continue;
1347 		*vpp = vp;
1348 		rc = 1;
1349 		break;
1350 	}
1351 	mutex_exit(&specfs_lock);
1352 	return (rc);
1353 }
1354 
1355 /*
1356  * Revoke all the vnodes corresponding to the specified minor number
1357  * range (endpoints inclusive) of the specified major.
1358  */
1359 void
1360 vdevgone(int maj, int minl, int minh, enum vtype type)
1361 {
1362 	vnode_t *vp, **vpp;
1363 	dev_t dev;
1364 	int mn;
1365 
1366 	vp = NULL;	/* XXX gcc */
1367 
1368 	mutex_enter(&specfs_lock);
1369 	for (mn = minl; mn <= minh; mn++) {
1370 		dev = makedev(maj, mn);
1371 		vpp = &specfs_hash[SPECHASH(dev)];
1372 		for (vp = *vpp; vp != NULL;) {
1373 			mutex_enter(&vp->v_interlock);
1374 			if ((vp->v_iflag & VI_CLEAN) != 0 ||
1375 			    dev != vp->v_rdev || type != vp->v_type) {
1376 				mutex_exit(&vp->v_interlock);
1377 				vp = vp->v_specnext;
1378 				continue;
1379 			}
1380 			mutex_exit(&specfs_lock);
1381 			if (vget(vp, LK_INTERLOCK) == 0) {
1382 				VOP_REVOKE(vp, REVOKEALL);
1383 				vrele(vp);
1384 			}
1385 			mutex_enter(&specfs_lock);
1386 			vp = *vpp;
1387 		}
1388 	}
1389 	mutex_exit(&specfs_lock);
1390 }
1391 
1392 /*
1393  * Calculate the total number of references to a special device.
1394  */
1395 int
1396 vcount(vnode_t *vp)
1397 {
1398 	int count;
1399 
1400 	mutex_enter(&specfs_lock);
1401 	mutex_enter(&vp->v_interlock);
1402 	if (vp->v_specnode == NULL) {
1403 		count = vp->v_usecount - ((vp->v_iflag & VI_INACTPEND) != 0);
1404 		mutex_exit(&vp->v_interlock);
1405 		mutex_exit(&specfs_lock);
1406 		return (count);
1407 	}
1408 	mutex_exit(&vp->v_interlock);
1409 	count = vp->v_specnode->sn_dev->sd_opencnt;
1410 	mutex_exit(&specfs_lock);
1411 	return (count);
1412 }
1413 
1414 /*
1415  * Eliminate all activity associated with the requested vnode
1416  * and with all vnodes aliased to the requested vnode.
1417  */
1418 void
1419 vrevoke(vnode_t *vp)
1420 {
1421 	vnode_t *vq, **vpp;
1422 	enum vtype type;
1423 	dev_t dev;
1424 
1425 	KASSERT(vp->v_usecount > 0);
1426 
1427 	mutex_enter(&vp->v_interlock);
1428 	if ((vp->v_iflag & VI_CLEAN) != 0) {
1429 		mutex_exit(&vp->v_interlock);
1430 		return;
1431 	} else {
1432 		dev = vp->v_rdev;
1433 		type = vp->v_type;
1434 		mutex_exit(&vp->v_interlock);
1435 	}
1436 
1437 	vpp = &specfs_hash[SPECHASH(dev)];
1438 	mutex_enter(&specfs_lock);
1439 	for (vq = *vpp; vq != NULL;) {
1440 		/* If clean or being cleaned, then ignore it. */
1441 		mutex_enter(&vq->v_interlock);
1442 		if ((vq->v_iflag & (VI_CLEAN | VI_XLOCK)) != 0 ||
1443 		    vq->v_rdev != dev || vq->v_type != type) {
1444 			mutex_exit(&vq->v_interlock);
1445 			vq = vq->v_specnext;
1446 			continue;
1447 		}
1448 		mutex_exit(&specfs_lock);
1449 		if (vq->v_usecount == 0) {
1450 			vremfree(vq);
1451 		}
1452 		vq->v_usecount++;
1453 		vclean(vq, DOCLOSE);
1454 		vrelel(vq, 0);
1455 		mutex_enter(&specfs_lock);
1456 		vq = *vpp;
1457 	}
1458 	mutex_exit(&specfs_lock);
1459 }
1460 
1461 /*
1462  * sysctl helper routine to return list of supported fstypes
1463  */
1464 static int
1465 sysctl_vfs_generic_fstypes(SYSCTLFN_ARGS)
1466 {
1467 	char bf[sizeof(((struct statvfs *)NULL)->f_fstypename)];
1468 	char *where = oldp;
1469 	struct vfsops *v;
1470 	size_t needed, left, slen;
1471 	int error, first;
1472 
1473 	if (newp != NULL)
1474 		return (EPERM);
1475 	if (namelen != 0)
1476 		return (EINVAL);
1477 
1478 	first = 1;
1479 	error = 0;
1480 	needed = 0;
1481 	left = *oldlenp;
1482 
1483 	sysctl_unlock();
1484 	mutex_enter(&vfs_list_lock);
1485 	LIST_FOREACH(v, &vfs_list, vfs_list) {
1486 		if (where == NULL)
1487 			needed += strlen(v->vfs_name) + 1;
1488 		else {
1489 			memset(bf, 0, sizeof(bf));
1490 			if (first) {
1491 				strncpy(bf, v->vfs_name, sizeof(bf));
1492 				first = 0;
1493 			} else {
1494 				bf[0] = ' ';
1495 				strncpy(bf + 1, v->vfs_name, sizeof(bf) - 1);
1496 			}
1497 			bf[sizeof(bf)-1] = '\0';
1498 			slen = strlen(bf);
1499 			if (left < slen + 1)
1500 				break;
1501 			/* +1 to copy out the trailing NUL byte */
1502 			v->vfs_refcount++;
1503 			mutex_exit(&vfs_list_lock);
1504 			error = copyout(bf, where, slen + 1);
1505 			mutex_enter(&vfs_list_lock);
1506 			v->vfs_refcount--;
1507 			if (error)
1508 				break;
1509 			where += slen;
1510 			needed += slen;
1511 			left -= slen;
1512 		}
1513 	}
1514 	mutex_exit(&vfs_list_lock);
1515 	sysctl_relock();
1516 	*oldlenp = needed;
1517 	return (error);
1518 }
1519 
1520 /*
1521  * Top level filesystem related information gathering.
1522  */
1523 SYSCTL_SETUP(sysctl_vfs_setup, "sysctl vfs subtree setup")
1524 {
1525 	sysctl_createv(clog, 0, NULL, NULL,
1526 		       CTLFLAG_PERMANENT,
1527 		       CTLTYPE_NODE, "vfs", NULL,
1528 		       NULL, 0, NULL, 0,
1529 		       CTL_VFS, CTL_EOL);
1530 	sysctl_createv(clog, 0, NULL, NULL,
1531 		       CTLFLAG_PERMANENT,
1532 		       CTLTYPE_NODE, "generic",
1533 		       SYSCTL_DESCR("Non-specific vfs related information"),
1534 		       NULL, 0, NULL, 0,
1535 		       CTL_VFS, VFS_GENERIC, CTL_EOL);
1536 	sysctl_createv(clog, 0, NULL, NULL,
1537 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1538 		       CTLTYPE_INT, "usermount",
1539 		       SYSCTL_DESCR("Whether unprivileged users may mount "
1540 				    "filesystems"),
1541 		       NULL, 0, &dovfsusermount, 0,
1542 		       CTL_VFS, VFS_GENERIC, VFS_USERMOUNT, CTL_EOL);
1543 	sysctl_createv(clog, 0, NULL, NULL,
1544 		       CTLFLAG_PERMANENT,
1545 		       CTLTYPE_STRING, "fstypes",
1546 		       SYSCTL_DESCR("List of file systems present"),
1547 		       sysctl_vfs_generic_fstypes, 0, NULL, 0,
1548 		       CTL_VFS, VFS_GENERIC, CTL_CREATE, CTL_EOL);
1549 	sysctl_createv(clog, 0, NULL, NULL,
1550 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1551 		       CTLTYPE_INT, "magiclinks",
1552 		       SYSCTL_DESCR("Whether \"magic\" symlinks are expanded"),
1553 		       NULL, 0, &vfs_magiclinks, 0,
1554 		       CTL_VFS, VFS_GENERIC, VFS_MAGICLINKS, CTL_EOL);
1555 }
1556 
1557 
1558 int kinfo_vdebug = 1;
1559 int kinfo_vgetfailed;
1560 #define KINFO_VNODESLOP	10
1561 /*
1562  * Dump vnode list (via sysctl).
1563  * Copyout address of vnode followed by vnode.
1564  */
1565 /* ARGSUSED */
1566 int
1567 sysctl_kern_vnode(SYSCTLFN_ARGS)
1568 {
1569 	char *where = oldp;
1570 	size_t *sizep = oldlenp;
1571 	struct mount *mp, *nmp;
1572 	vnode_t *vp, *mvp, vbuf;
1573 	char *bp = where, *savebp;
1574 	char *ewhere;
1575 	int error;
1576 
1577 	if (namelen != 0)
1578 		return (EOPNOTSUPP);
1579 	if (newp != NULL)
1580 		return (EPERM);
1581 
1582 #define VPTRSZ	sizeof(vnode_t *)
1583 #define VNODESZ	sizeof(vnode_t)
1584 	if (where == NULL) {
1585 		*sizep = (numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ);
1586 		return (0);
1587 	}
1588 	ewhere = where + *sizep;
1589 
1590 	sysctl_unlock();
1591 	mutex_enter(&mountlist_lock);
1592 	for (mp = CIRCLEQ_FIRST(&mountlist); mp != (void *)&mountlist;
1593 	     mp = nmp) {
1594 		if (vfs_busy(mp, &nmp)) {
1595 			continue;
1596 		}
1597 		savebp = bp;
1598 		/* Allocate a marker vnode. */
1599 		if ((mvp = vnalloc(mp)) == NULL) {
1600 			sysctl_relock();
1601 			return (ENOMEM);
1602 		}
1603 		mutex_enter(&mntvnode_lock);
1604 		for (vp = TAILQ_FIRST(&mp->mnt_vnodelist); vp; vp = vunmark(mvp)) {
1605 			vmark(mvp, vp);
1606 			/*
1607 			 * Check that the vp is still associated with
1608 			 * this filesystem.  RACE: could have been
1609 			 * recycled onto the same filesystem.
1610 			 */
1611 			if (vp->v_mount != mp || vismarker(vp))
1612 				continue;
1613 			if (bp + VPTRSZ + VNODESZ > ewhere) {
1614 				(void)vunmark(mvp);
1615 				mutex_exit(&mntvnode_lock);
1616 				vnfree(mvp);
1617 				sysctl_relock();
1618 				*sizep = bp - where;
1619 				return (ENOMEM);
1620 			}
1621 			memcpy(&vbuf, vp, VNODESZ);
1622 			mutex_exit(&mntvnode_lock);
1623 			if ((error = copyout(vp, bp, VPTRSZ)) ||
1624 			   (error = copyout(&vbuf, bp + VPTRSZ, VNODESZ))) {
1625 			   	mutex_enter(&mntvnode_lock);
1626 				(void)vunmark(mvp);
1627 				mutex_exit(&mntvnode_lock);
1628 				vnfree(mvp);
1629 				sysctl_relock();
1630 				return (error);
1631 			}
1632 			bp += VPTRSZ + VNODESZ;
1633 			mutex_enter(&mntvnode_lock);
1634 		}
1635 		mutex_exit(&mntvnode_lock);
1636 		vnfree(mvp);
1637 		vfs_unbusy(mp, false, &nmp);
1638 	}
1639 	mutex_exit(&mountlist_lock);
1640 	sysctl_relock();
1641 
1642 	*sizep = bp - where;
1643 	return (0);
1644 }
1645 
1646 /*
1647  * Remove clean vnodes from a mountpoint's vnode list.
1648  */
1649 void
1650 vfs_scrubvnlist(struct mount *mp)
1651 {
1652 	vnode_t *vp, *nvp;
1653 
1654  retry:
1655 	mutex_enter(&mntvnode_lock);
1656 	for (vp = TAILQ_FIRST(&mp->mnt_vnodelist); vp; vp = nvp) {
1657 		nvp = TAILQ_NEXT(vp, v_mntvnodes);
1658 		mutex_enter(&vp->v_interlock);
1659 		if ((vp->v_iflag & VI_CLEAN) != 0) {
1660 			TAILQ_REMOVE(&mp->mnt_vnodelist, vp, v_mntvnodes);
1661 			vp->v_mount = NULL;
1662 			mutex_exit(&mntvnode_lock);
1663 			mutex_exit(&vp->v_interlock);
1664 			vfs_destroy(mp);
1665 			goto retry;
1666 		}
1667 		mutex_exit(&vp->v_interlock);
1668 	}
1669 	mutex_exit(&mntvnode_lock);
1670 }
1671 
1672 /*
1673  * Check to see if a filesystem is mounted on a block device.
1674  */
1675 int
1676 vfs_mountedon(vnode_t *vp)
1677 {
1678 	vnode_t *vq;
1679 	int error = 0;
1680 
1681 	if (vp->v_type != VBLK)
1682 		return ENOTBLK;
1683 	if (vp->v_specmountpoint != NULL)
1684 		return (EBUSY);
1685 	mutex_enter(&specfs_lock);
1686 	for (vq = specfs_hash[SPECHASH(vp->v_rdev)]; vq != NULL;
1687 	    vq = vq->v_specnext) {
1688 		if (vq->v_rdev != vp->v_rdev || vq->v_type != vp->v_type)
1689 			continue;
1690 		if (vq->v_specmountpoint != NULL) {
1691 			error = EBUSY;
1692 			break;
1693 		}
1694 	}
1695 	mutex_exit(&specfs_lock);
1696 	return (error);
1697 }
1698 
1699 /*
1700  * Unmount all file systems.
1701  * We traverse the list in reverse order under the assumption that doing so
1702  * will avoid needing to worry about dependencies.
1703  */
1704 void
1705 vfs_unmountall(struct lwp *l)
1706 {
1707 	struct mount *mp, *nmp;
1708 	int allerror, error;
1709 
1710 	printf("unmounting file systems...");
1711 	for (allerror = 0, mp = CIRCLEQ_LAST(&mountlist);
1712 	     !CIRCLEQ_EMPTY(&mountlist);
1713 	     mp = nmp) {
1714 		nmp = CIRCLEQ_PREV(mp, mnt_list);
1715 #ifdef DEBUG
1716 		printf("\nunmounting %s (%s)...",
1717 		    mp->mnt_stat.f_mntonname, mp->mnt_stat.f_mntfromname);
1718 #endif
1719 		atomic_inc_uint(&mp->mnt_refcnt);
1720 		if ((error = dounmount(mp, MNT_FORCE, l)) != 0) {
1721 			printf("unmount of %s failed with error %d\n",
1722 			    mp->mnt_stat.f_mntonname, error);
1723 			allerror = 1;
1724 		}
1725 	}
1726 	printf(" done\n");
1727 	if (allerror)
1728 		printf("WARNING: some file systems would not unmount\n");
1729 }
1730 
1731 /*
1732  * Sync and unmount file systems before shutting down.
1733  */
1734 void
1735 vfs_shutdown(void)
1736 {
1737 	struct lwp *l;
1738 
1739 	/* XXX we're certainly not running in lwp0's context! */
1740 	l = curlwp;
1741 	if (l == NULL)
1742 		l = &lwp0;
1743 
1744 	printf("syncing disks... ");
1745 
1746 	/* remove user processes from run queue */
1747 	suspendsched();
1748 	(void) spl0();
1749 
1750 	/* avoid coming back this way again if we panic. */
1751 	doing_shutdown = 1;
1752 
1753 	sys_sync(l, NULL, NULL);
1754 
1755 	/* Wait for sync to finish. */
1756 	if (buf_syncwait() != 0) {
1757 #if defined(DDB) && defined(DEBUG_HALT_BUSY)
1758 		Debugger();
1759 #endif
1760 		printf("giving up\n");
1761 		return;
1762 	} else
1763 		printf("done\n");
1764 
1765 	/*
1766 	 * If we've panic'd, don't make the situation potentially
1767 	 * worse by unmounting the file systems.
1768 	 */
1769 	if (panicstr != NULL)
1770 		return;
1771 
1772 	/* Release inodes held by texts before update. */
1773 #ifdef notdef
1774 	vnshutdown();
1775 #endif
1776 	/* Unmount file systems. */
1777 	vfs_unmountall(l);
1778 }
1779 
1780 /*
1781  * Mount the root file system.  If the operator didn't specify a
1782  * file system to use, try all possible file systems until one
1783  * succeeds.
1784  */
1785 int
1786 vfs_mountroot(void)
1787 {
1788 	struct vfsops *v;
1789 	int error = ENODEV;
1790 
1791 	if (root_device == NULL)
1792 		panic("vfs_mountroot: root device unknown");
1793 
1794 	switch (device_class(root_device)) {
1795 	case DV_IFNET:
1796 		if (rootdev != NODEV)
1797 			panic("vfs_mountroot: rootdev set for DV_IFNET "
1798 			    "(0x%08x -> %d,%d)", rootdev,
1799 			    major(rootdev), minor(rootdev));
1800 		break;
1801 
1802 	case DV_DISK:
1803 		if (rootdev == NODEV)
1804 			panic("vfs_mountroot: rootdev not set for DV_DISK");
1805 	        if (bdevvp(rootdev, &rootvp))
1806 	                panic("vfs_mountroot: can't get vnode for rootdev");
1807 		error = VOP_OPEN(rootvp, FREAD, FSCRED);
1808 		if (error) {
1809 			printf("vfs_mountroot: can't open root device\n");
1810 			return (error);
1811 		}
1812 		break;
1813 
1814 	default:
1815 		printf("%s: inappropriate for root file system\n",
1816 		    device_xname(root_device));
1817 		return (ENODEV);
1818 	}
1819 
1820 	/*
1821 	 * If user specified a file system, use it.
1822 	 */
1823 	if (mountroot != NULL) {
1824 		error = (*mountroot)();
1825 		goto done;
1826 	}
1827 
1828 	/*
1829 	 * Try each file system currently configured into the kernel.
1830 	 */
1831 	mutex_enter(&vfs_list_lock);
1832 	LIST_FOREACH(v, &vfs_list, vfs_list) {
1833 		if (v->vfs_mountroot == NULL)
1834 			continue;
1835 #ifdef DEBUG
1836 		aprint_normal("mountroot: trying %s...\n", v->vfs_name);
1837 #endif
1838 		v->vfs_refcount++;
1839 		mutex_exit(&vfs_list_lock);
1840 		error = (*v->vfs_mountroot)();
1841 		mutex_enter(&vfs_list_lock);
1842 		v->vfs_refcount--;
1843 		if (!error) {
1844 			aprint_normal("root file system type: %s\n",
1845 			    v->vfs_name);
1846 			break;
1847 		}
1848 	}
1849 	mutex_exit(&vfs_list_lock);
1850 
1851 	if (v == NULL) {
1852 		printf("no file system for %s", device_xname(root_device));
1853 		if (device_class(root_device) == DV_DISK)
1854 			printf(" (dev 0x%x)", rootdev);
1855 		printf("\n");
1856 		error = EFTYPE;
1857 	}
1858 
1859 done:
1860 	if (error && device_class(root_device) == DV_DISK) {
1861 		VOP_CLOSE(rootvp, FREAD, FSCRED);
1862 		vrele(rootvp);
1863 	}
1864 	return (error);
1865 }
1866 
1867 /*
1868  * Sham lock manager for vnodes.  This is a temporary measure.
1869  */
1870 int
1871 vlockmgr(struct vnlock *vl, int flags)
1872 {
1873 
1874 	KASSERT((flags & ~(LK_CANRECURSE | LK_NOWAIT | LK_TYPE_MASK)) == 0);
1875 
1876 	switch (flags & LK_TYPE_MASK) {
1877 	case LK_SHARED:
1878 		if (rw_tryenter(&vl->vl_lock, RW_READER)) {
1879 			return 0;
1880 		}
1881 		if ((flags & LK_NOWAIT) != 0) {
1882 			return EBUSY;
1883 		}
1884 		rw_enter(&vl->vl_lock, RW_READER);
1885 		return 0;
1886 
1887 	case LK_EXCLUSIVE:
1888 		if (rw_tryenter(&vl->vl_lock, RW_WRITER)) {
1889 			return 0;
1890 		}
1891 		if ((vl->vl_canrecurse || (flags & LK_CANRECURSE) != 0) &&
1892 		    rw_write_held(&vl->vl_lock)) {
1893 			vl->vl_recursecnt++;
1894 			return 0;
1895 		}
1896 		if ((flags & LK_NOWAIT) != 0) {
1897 			return EBUSY;
1898 		}
1899 		rw_enter(&vl->vl_lock, RW_WRITER);
1900 		return 0;
1901 
1902 	case LK_RELEASE:
1903 		if (vl->vl_recursecnt != 0) {
1904 			KASSERT(rw_write_held(&vl->vl_lock));
1905 			vl->vl_recursecnt--;
1906 			return 0;
1907 		}
1908 		rw_exit(&vl->vl_lock);
1909 		return 0;
1910 
1911 	default:
1912 		panic("vlockmgr: flags %x", flags);
1913 	}
1914 }
1915 
1916 int
1917 vlockstatus(struct vnlock *vl)
1918 {
1919 
1920 	if (rw_write_held(&vl->vl_lock)) {
1921 		return LK_EXCLUSIVE;
1922 	}
1923 	if (rw_read_held(&vl->vl_lock)) {
1924 		return LK_SHARED;
1925 	}
1926 	return 0;
1927 }
1928