xref: /freebsd/sys/kern/vfs_subr.c (revision f95c930e6701aa675377bc5871ea490dd565eeba)
1 /*-
2  * Copyright (c) 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 4. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)vfs_subr.c	8.31 (Berkeley) 5/26/95
35  */
36 
37 /*
38  * External virtual filesystem routines
39  */
40 
41 #include <sys/cdefs.h>
42 __FBSDID("$FreeBSD$");
43 
44 #include "opt_compat.h"
45 #include "opt_ddb.h"
46 #include "opt_watchdog.h"
47 
48 #include <sys/param.h>
49 #include <sys/systm.h>
50 #include <sys/bio.h>
51 #include <sys/buf.h>
52 #include <sys/condvar.h>
53 #include <sys/conf.h>
54 #include <sys/dirent.h>
55 #include <sys/event.h>
56 #include <sys/eventhandler.h>
57 #include <sys/extattr.h>
58 #include <sys/file.h>
59 #include <sys/fcntl.h>
60 #include <sys/jail.h>
61 #include <sys/kdb.h>
62 #include <sys/kernel.h>
63 #include <sys/kthread.h>
64 #include <sys/lockf.h>
65 #include <sys/malloc.h>
66 #include <sys/mount.h>
67 #include <sys/namei.h>
68 #include <sys/pctrie.h>
69 #include <sys/priv.h>
70 #include <sys/reboot.h>
71 #include <sys/rwlock.h>
72 #include <sys/sched.h>
73 #include <sys/sleepqueue.h>
74 #include <sys/smp.h>
75 #include <sys/stat.h>
76 #include <sys/sysctl.h>
77 #include <sys/syslog.h>
78 #include <sys/vmmeter.h>
79 #include <sys/vnode.h>
80 #include <sys/watchdog.h>
81 
82 #include <machine/stdarg.h>
83 
84 #include <security/mac/mac_framework.h>
85 
86 #include <vm/vm.h>
87 #include <vm/vm_object.h>
88 #include <vm/vm_extern.h>
89 #include <vm/pmap.h>
90 #include <vm/vm_map.h>
91 #include <vm/vm_page.h>
92 #include <vm/vm_kern.h>
93 #include <vm/uma.h>
94 
95 #ifdef DDB
96 #include <ddb/ddb.h>
97 #endif
98 
99 static void	delmntque(struct vnode *vp);
100 static int	flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo,
101 		    int slpflag, int slptimeo);
102 static void	syncer_shutdown(void *arg, int howto);
103 static int	vtryrecycle(struct vnode *vp);
104 static void	v_incr_usecount(struct vnode *);
105 static void	v_decr_usecount(struct vnode *);
106 static void	v_decr_useonly(struct vnode *);
107 static void	v_upgrade_usecount(struct vnode *);
108 static void	vnlru_free(int);
109 static void	vgonel(struct vnode *);
110 static void	vfs_knllock(void *arg);
111 static void	vfs_knlunlock(void *arg);
112 static void	vfs_knl_assert_locked(void *arg);
113 static void	vfs_knl_assert_unlocked(void *arg);
114 static void	destroy_vpollinfo(struct vpollinfo *vi);
115 
116 /*
117  * Number of vnodes in existence.  Increased whenever getnewvnode()
118  * allocates a new vnode, decreased in vdropl() for VI_DOOMED vnode.
119  */
120 static unsigned long	numvnodes;
121 
122 SYSCTL_ULONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0,
123     "Number of vnodes in existence");
124 
125 static u_long vnodes_created;
126 SYSCTL_ULONG(_vfs, OID_AUTO, vnodes_created, CTLFLAG_RD, &vnodes_created,
127     0, "Number of vnodes created by getnewvnode");
128 
129 /*
130  * Conversion tables for conversion from vnode types to inode formats
131  * and back.
132  */
133 enum vtype iftovt_tab[16] = {
134 	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
135 	VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
136 };
137 int vttoif_tab[10] = {
138 	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
139 	S_IFSOCK, S_IFIFO, S_IFMT, S_IFMT
140 };
141 
142 /*
143  * List of vnodes that are ready for recycling.
144  */
145 static TAILQ_HEAD(freelst, vnode) vnode_free_list;
146 
147 /*
148  * Free vnode target.  Free vnodes may simply be files which have been stat'd
149  * but not read.  This is somewhat common, and a small cache of such files
150  * should be kept to avoid recreation costs.
151  */
152 static u_long wantfreevnodes;
153 SYSCTL_ULONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
154 /* Number of vnodes in the free list. */
155 static u_long freevnodes;
156 SYSCTL_ULONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0,
157     "Number of vnodes in the free list");
158 
159 static int vlru_allow_cache_src;
160 SYSCTL_INT(_vfs, OID_AUTO, vlru_allow_cache_src, CTLFLAG_RW,
161     &vlru_allow_cache_src, 0, "Allow vlru to reclaim source vnode");
162 
163 static u_long recycles_count;
164 SYSCTL_ULONG(_vfs, OID_AUTO, recycles, CTLFLAG_RD, &recycles_count, 0,
165     "Number of vnodes recycled to avoid exceding kern.maxvnodes");
166 
167 /*
168  * Various variables used for debugging the new implementation of
169  * reassignbuf().
170  * XXX these are probably of (very) limited utility now.
171  */
172 static int reassignbufcalls;
173 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0,
174     "Number of calls to reassignbuf");
175 
176 /*
177  * Cache for the mount type id assigned to NFS.  This is used for
178  * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c.
179  */
180 int	nfs_mount_type = -1;
181 
182 /* To keep more than one thread at a time from running vfs_getnewfsid */
183 static struct mtx mntid_mtx;
184 
185 /*
186  * Lock for any access to the following:
187  *	vnode_free_list
188  *	numvnodes
189  *	freevnodes
190  */
191 static struct mtx vnode_free_list_mtx;
192 
193 /* Publicly exported FS */
194 struct nfs_public nfs_pub;
195 
196 static uma_zone_t buf_trie_zone;
197 
198 /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
199 static uma_zone_t vnode_zone;
200 static uma_zone_t vnodepoll_zone;
201 
202 /*
203  * The workitem queue.
204  *
205  * It is useful to delay writes of file data and filesystem metadata
206  * for tens of seconds so that quickly created and deleted files need
207  * not waste disk bandwidth being created and removed. To realize this,
208  * we append vnodes to a "workitem" queue. When running with a soft
209  * updates implementation, most pending metadata dependencies should
210  * not wait for more than a few seconds. Thus, mounted on block devices
211  * are delayed only about a half the time that file data is delayed.
212  * Similarly, directory updates are more critical, so are only delayed
213  * about a third the time that file data is delayed. Thus, there are
214  * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
215  * one each second (driven off the filesystem syncer process). The
216  * syncer_delayno variable indicates the next queue that is to be processed.
217  * Items that need to be processed soon are placed in this queue:
218  *
219  *	syncer_workitem_pending[syncer_delayno]
220  *
221  * A delay of fifteen seconds is done by placing the request fifteen
222  * entries later in the queue:
223  *
224  *	syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
225  *
226  */
227 static int syncer_delayno;
228 static long syncer_mask;
229 LIST_HEAD(synclist, bufobj);
230 static struct synclist *syncer_workitem_pending;
231 /*
232  * The sync_mtx protects:
233  *	bo->bo_synclist
234  *	sync_vnode_count
235  *	syncer_delayno
236  *	syncer_state
237  *	syncer_workitem_pending
238  *	syncer_worklist_len
239  *	rushjob
240  */
241 static struct mtx sync_mtx;
242 static struct cv sync_wakeup;
243 
244 #define SYNCER_MAXDELAY		32
245 static int syncer_maxdelay = SYNCER_MAXDELAY;	/* maximum delay time */
246 static int syncdelay = 30;		/* max time to delay syncing data */
247 static int filedelay = 30;		/* time to delay syncing files */
248 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0,
249     "Time to delay syncing files (in seconds)");
250 static int dirdelay = 29;		/* time to delay syncing directories */
251 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0,
252     "Time to delay syncing directories (in seconds)");
253 static int metadelay = 28;		/* time to delay syncing metadata */
254 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0,
255     "Time to delay syncing metadata (in seconds)");
256 static int rushjob;		/* number of slots to run ASAP */
257 static int stat_rush_requests;	/* number of times I/O speeded up */
258 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0,
259     "Number of times I/O speeded up (rush requests)");
260 
261 /*
262  * When shutting down the syncer, run it at four times normal speed.
263  */
264 #define SYNCER_SHUTDOWN_SPEEDUP		4
265 static int sync_vnode_count;
266 static int syncer_worklist_len;
267 static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY }
268     syncer_state;
269 
270 /*
271  * Number of vnodes we want to exist at any one time.  This is mostly used
272  * to size hash tables in vnode-related code.  It is normally not used in
273  * getnewvnode(), as wantfreevnodes is normally nonzero.)
274  *
275  * XXX desiredvnodes is historical cruft and should not exist.
276  */
277 int desiredvnodes;
278 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
279     &desiredvnodes, 0, "Maximum number of vnodes");
280 SYSCTL_ULONG(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
281     &wantfreevnodes, 0, "Minimum number of vnodes (legacy)");
282 static int vnlru_nowhere;
283 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW,
284     &vnlru_nowhere, 0, "Number of times the vnlru process ran without success");
285 
286 /* Shift count for (uintptr_t)vp to initialize vp->v_hash. */
287 static int vnsz2log;
288 
289 /*
290  * Support for the bufobj clean & dirty pctrie.
291  */
292 static void *
293 buf_trie_alloc(struct pctrie *ptree)
294 {
295 
296 	return uma_zalloc(buf_trie_zone, M_NOWAIT);
297 }
298 
299 static void
300 buf_trie_free(struct pctrie *ptree, void *node)
301 {
302 
303 	uma_zfree(buf_trie_zone, node);
304 }
305 PCTRIE_DEFINE(BUF, buf, b_lblkno, buf_trie_alloc, buf_trie_free);
306 
307 /*
308  * Initialize the vnode management data structures.
309  *
310  * Reevaluate the following cap on the number of vnodes after the physical
311  * memory size exceeds 512GB.  In the limit, as the physical memory size
312  * grows, the ratio of physical pages to vnodes approaches sixteen to one.
313  */
314 #ifndef	MAXVNODES_MAX
315 #define	MAXVNODES_MAX	(512 * (1024 * 1024 * 1024 / (int)PAGE_SIZE / 16))
316 #endif
317 static void
318 vntblinit(void *dummy __unused)
319 {
320 	u_int i;
321 	int physvnodes, virtvnodes;
322 
323 	/*
324 	 * Desiredvnodes is a function of the physical memory size and the
325 	 * kernel's heap size.  Generally speaking, it scales with the
326 	 * physical memory size.  The ratio of desiredvnodes to physical pages
327 	 * is one to four until desiredvnodes exceeds 98,304.  Thereafter, the
328 	 * marginal ratio of desiredvnodes to physical pages is one to
329 	 * sixteen.  However, desiredvnodes is limited by the kernel's heap
330 	 * size.  The memory required by desiredvnodes vnodes and vm objects
331 	 * may not exceed one seventh of the kernel's heap size.
332 	 */
333 	physvnodes = maxproc + vm_cnt.v_page_count / 16 + 3 * min(98304 * 4,
334 	    vm_cnt.v_page_count) / 16;
335 	virtvnodes = vm_kmem_size / (7 * (sizeof(struct vm_object) +
336 	    sizeof(struct vnode)));
337 	desiredvnodes = min(physvnodes, virtvnodes);
338 	if (desiredvnodes > MAXVNODES_MAX) {
339 		if (bootverbose)
340 			printf("Reducing kern.maxvnodes %d -> %d\n",
341 			    desiredvnodes, MAXVNODES_MAX);
342 		desiredvnodes = MAXVNODES_MAX;
343 	}
344 	wantfreevnodes = desiredvnodes / 4;
345 	mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF);
346 	TAILQ_INIT(&vnode_free_list);
347 	mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF);
348 	vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL,
349 	    NULL, NULL, UMA_ALIGN_PTR, 0);
350 	vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo),
351 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
352 	/*
353 	 * Preallocate enough nodes to support one-per buf so that
354 	 * we can not fail an insert.  reassignbuf() callers can not
355 	 * tolerate the insertion failure.
356 	 */
357 	buf_trie_zone = uma_zcreate("BUF TRIE", pctrie_node_size(),
358 	    NULL, NULL, pctrie_zone_init, NULL, UMA_ALIGN_PTR,
359 	    UMA_ZONE_NOFREE | UMA_ZONE_VM);
360 	uma_prealloc(buf_trie_zone, nbuf);
361 	/*
362 	 * Initialize the filesystem syncer.
363 	 */
364 	syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE,
365 	    &syncer_mask);
366 	syncer_maxdelay = syncer_mask + 1;
367 	mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF);
368 	cv_init(&sync_wakeup, "syncer");
369 	for (i = 1; i <= sizeof(struct vnode); i <<= 1)
370 		vnsz2log++;
371 	vnsz2log--;
372 }
373 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL);
374 
375 
376 /*
377  * Mark a mount point as busy. Used to synchronize access and to delay
378  * unmounting. Eventually, mountlist_mtx is not released on failure.
379  *
380  * vfs_busy() is a custom lock, it can block the caller.
381  * vfs_busy() only sleeps if the unmount is active on the mount point.
382  * For a mountpoint mp, vfs_busy-enforced lock is before lock of any
383  * vnode belonging to mp.
384  *
385  * Lookup uses vfs_busy() to traverse mount points.
386  * root fs			var fs
387  * / vnode lock		A	/ vnode lock (/var)		D
388  * /var vnode lock	B	/log vnode lock(/var/log)	E
389  * vfs_busy lock	C	vfs_busy lock			F
390  *
391  * Within each file system, the lock order is C->A->B and F->D->E.
392  *
393  * When traversing across mounts, the system follows that lock order:
394  *
395  *        C->A->B
396  *              |
397  *              +->F->D->E
398  *
399  * The lookup() process for namei("/var") illustrates the process:
400  *  VOP_LOOKUP() obtains B while A is held
401  *  vfs_busy() obtains a shared lock on F while A and B are held
402  *  vput() releases lock on B
403  *  vput() releases lock on A
404  *  VFS_ROOT() obtains lock on D while shared lock on F is held
405  *  vfs_unbusy() releases shared lock on F
406  *  vn_lock() obtains lock on deadfs vnode vp_crossmp instead of A.
407  *    Attempt to lock A (instead of vp_crossmp) while D is held would
408  *    violate the global order, causing deadlocks.
409  *
410  * dounmount() locks B while F is drained.
411  */
412 int
413 vfs_busy(struct mount *mp, int flags)
414 {
415 
416 	MPASS((flags & ~MBF_MASK) == 0);
417 	CTR3(KTR_VFS, "%s: mp %p with flags %d", __func__, mp, flags);
418 
419 	MNT_ILOCK(mp);
420 	MNT_REF(mp);
421 	/*
422 	 * If mount point is currenly being unmounted, sleep until the
423 	 * mount point fate is decided.  If thread doing the unmounting fails,
424 	 * it will clear MNTK_UNMOUNT flag before waking us up, indicating
425 	 * that this mount point has survived the unmount attempt and vfs_busy
426 	 * should retry.  Otherwise the unmounter thread will set MNTK_REFEXPIRE
427 	 * flag in addition to MNTK_UNMOUNT, indicating that mount point is
428 	 * about to be really destroyed.  vfs_busy needs to release its
429 	 * reference on the mount point in this case and return with ENOENT,
430 	 * telling the caller that mount mount it tried to busy is no longer
431 	 * valid.
432 	 */
433 	while (mp->mnt_kern_flag & MNTK_UNMOUNT) {
434 		if (flags & MBF_NOWAIT || mp->mnt_kern_flag & MNTK_REFEXPIRE) {
435 			MNT_REL(mp);
436 			MNT_IUNLOCK(mp);
437 			CTR1(KTR_VFS, "%s: failed busying before sleeping",
438 			    __func__);
439 			return (ENOENT);
440 		}
441 		if (flags & MBF_MNTLSTLOCK)
442 			mtx_unlock(&mountlist_mtx);
443 		mp->mnt_kern_flag |= MNTK_MWAIT;
444 		msleep(mp, MNT_MTX(mp), PVFS | PDROP, "vfs_busy", 0);
445 		if (flags & MBF_MNTLSTLOCK)
446 			mtx_lock(&mountlist_mtx);
447 		MNT_ILOCK(mp);
448 	}
449 	if (flags & MBF_MNTLSTLOCK)
450 		mtx_unlock(&mountlist_mtx);
451 	mp->mnt_lockref++;
452 	MNT_IUNLOCK(mp);
453 	return (0);
454 }
455 
456 /*
457  * Free a busy filesystem.
458  */
459 void
460 vfs_unbusy(struct mount *mp)
461 {
462 
463 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
464 	MNT_ILOCK(mp);
465 	MNT_REL(mp);
466 	KASSERT(mp->mnt_lockref > 0, ("negative mnt_lockref"));
467 	mp->mnt_lockref--;
468 	if (mp->mnt_lockref == 0 && (mp->mnt_kern_flag & MNTK_DRAINING) != 0) {
469 		MPASS(mp->mnt_kern_flag & MNTK_UNMOUNT);
470 		CTR1(KTR_VFS, "%s: waking up waiters", __func__);
471 		mp->mnt_kern_flag &= ~MNTK_DRAINING;
472 		wakeup(&mp->mnt_lockref);
473 	}
474 	MNT_IUNLOCK(mp);
475 }
476 
477 /*
478  * Lookup a mount point by filesystem identifier.
479  */
480 struct mount *
481 vfs_getvfs(fsid_t *fsid)
482 {
483 	struct mount *mp;
484 
485 	CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
486 	mtx_lock(&mountlist_mtx);
487 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
488 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
489 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
490 			vfs_ref(mp);
491 			mtx_unlock(&mountlist_mtx);
492 			return (mp);
493 		}
494 	}
495 	mtx_unlock(&mountlist_mtx);
496 	CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
497 	return ((struct mount *) 0);
498 }
499 
500 /*
501  * Lookup a mount point by filesystem identifier, busying it before
502  * returning.
503  *
504  * To avoid congestion on mountlist_mtx, implement simple direct-mapped
505  * cache for popular filesystem identifiers.  The cache is lockess, using
506  * the fact that struct mount's are never freed.  In worst case we may
507  * get pointer to unmounted or even different filesystem, so we have to
508  * check what we got, and go slow way if so.
509  */
510 struct mount *
511 vfs_busyfs(fsid_t *fsid)
512 {
513 #define	FSID_CACHE_SIZE	256
514 	typedef struct mount * volatile vmp_t;
515 	static vmp_t cache[FSID_CACHE_SIZE];
516 	struct mount *mp;
517 	int error;
518 	uint32_t hash;
519 
520 	CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
521 	hash = fsid->val[0] ^ fsid->val[1];
522 	hash = (hash >> 16 ^ hash) & (FSID_CACHE_SIZE - 1);
523 	mp = cache[hash];
524 	if (mp == NULL ||
525 	    mp->mnt_stat.f_fsid.val[0] != fsid->val[0] ||
526 	    mp->mnt_stat.f_fsid.val[1] != fsid->val[1])
527 		goto slow;
528 	if (vfs_busy(mp, 0) != 0) {
529 		cache[hash] = NULL;
530 		goto slow;
531 	}
532 	if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
533 	    mp->mnt_stat.f_fsid.val[1] == fsid->val[1])
534 		return (mp);
535 	else
536 	    vfs_unbusy(mp);
537 
538 slow:
539 	mtx_lock(&mountlist_mtx);
540 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
541 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
542 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
543 			error = vfs_busy(mp, MBF_MNTLSTLOCK);
544 			if (error) {
545 				cache[hash] = NULL;
546 				mtx_unlock(&mountlist_mtx);
547 				return (NULL);
548 			}
549 			cache[hash] = mp;
550 			return (mp);
551 		}
552 	}
553 	CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
554 	mtx_unlock(&mountlist_mtx);
555 	return ((struct mount *) 0);
556 }
557 
558 /*
559  * Check if a user can access privileged mount options.
560  */
561 int
562 vfs_suser(struct mount *mp, struct thread *td)
563 {
564 	int error;
565 
566 	/*
567 	 * If the thread is jailed, but this is not a jail-friendly file
568 	 * system, deny immediately.
569 	 */
570 	if (!(mp->mnt_vfc->vfc_flags & VFCF_JAIL) && jailed(td->td_ucred))
571 		return (EPERM);
572 
573 	/*
574 	 * If the file system was mounted outside the jail of the calling
575 	 * thread, deny immediately.
576 	 */
577 	if (prison_check(td->td_ucred, mp->mnt_cred) != 0)
578 		return (EPERM);
579 
580 	/*
581 	 * If file system supports delegated administration, we don't check
582 	 * for the PRIV_VFS_MOUNT_OWNER privilege - it will be better verified
583 	 * by the file system itself.
584 	 * If this is not the user that did original mount, we check for
585 	 * the PRIV_VFS_MOUNT_OWNER privilege.
586 	 */
587 	if (!(mp->mnt_vfc->vfc_flags & VFCF_DELEGADMIN) &&
588 	    mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) {
589 		if ((error = priv_check(td, PRIV_VFS_MOUNT_OWNER)) != 0)
590 			return (error);
591 	}
592 	return (0);
593 }
594 
595 /*
596  * Get a new unique fsid.  Try to make its val[0] unique, since this value
597  * will be used to create fake device numbers for stat().  Also try (but
598  * not so hard) make its val[0] unique mod 2^16, since some emulators only
599  * support 16-bit device numbers.  We end up with unique val[0]'s for the
600  * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
601  *
602  * Keep in mind that several mounts may be running in parallel.  Starting
603  * the search one past where the previous search terminated is both a
604  * micro-optimization and a defense against returning the same fsid to
605  * different mounts.
606  */
607 void
608 vfs_getnewfsid(struct mount *mp)
609 {
610 	static uint16_t mntid_base;
611 	struct mount *nmp;
612 	fsid_t tfsid;
613 	int mtype;
614 
615 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
616 	mtx_lock(&mntid_mtx);
617 	mtype = mp->mnt_vfc->vfc_typenum;
618 	tfsid.val[1] = mtype;
619 	mtype = (mtype & 0xFF) << 24;
620 	for (;;) {
621 		tfsid.val[0] = makedev(255,
622 		    mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
623 		mntid_base++;
624 		if ((nmp = vfs_getvfs(&tfsid)) == NULL)
625 			break;
626 		vfs_rel(nmp);
627 	}
628 	mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
629 	mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
630 	mtx_unlock(&mntid_mtx);
631 }
632 
633 /*
634  * Knob to control the precision of file timestamps:
635  *
636  *   0 = seconds only; nanoseconds zeroed.
637  *   1 = seconds and nanoseconds, accurate within 1/HZ.
638  *   2 = seconds and nanoseconds, truncated to microseconds.
639  * >=3 = seconds and nanoseconds, maximum precision.
640  */
641 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
642 
643 static int timestamp_precision = TSP_USEC;
644 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
645     &timestamp_precision, 0, "File timestamp precision (0: seconds, "
646     "1: sec + ns accurate to 1/HZ, 2: sec + ns truncated to ms, "
647     "3+: sec + ns (max. precision))");
648 
649 /*
650  * Get a current timestamp.
651  */
652 void
653 vfs_timestamp(struct timespec *tsp)
654 {
655 	struct timeval tv;
656 
657 	switch (timestamp_precision) {
658 	case TSP_SEC:
659 		tsp->tv_sec = time_second;
660 		tsp->tv_nsec = 0;
661 		break;
662 	case TSP_HZ:
663 		getnanotime(tsp);
664 		break;
665 	case TSP_USEC:
666 		microtime(&tv);
667 		TIMEVAL_TO_TIMESPEC(&tv, tsp);
668 		break;
669 	case TSP_NSEC:
670 	default:
671 		nanotime(tsp);
672 		break;
673 	}
674 }
675 
676 /*
677  * Set vnode attributes to VNOVAL
678  */
679 void
680 vattr_null(struct vattr *vap)
681 {
682 
683 	vap->va_type = VNON;
684 	vap->va_size = VNOVAL;
685 	vap->va_bytes = VNOVAL;
686 	vap->va_mode = VNOVAL;
687 	vap->va_nlink = VNOVAL;
688 	vap->va_uid = VNOVAL;
689 	vap->va_gid = VNOVAL;
690 	vap->va_fsid = VNOVAL;
691 	vap->va_fileid = VNOVAL;
692 	vap->va_blocksize = VNOVAL;
693 	vap->va_rdev = VNOVAL;
694 	vap->va_atime.tv_sec = VNOVAL;
695 	vap->va_atime.tv_nsec = VNOVAL;
696 	vap->va_mtime.tv_sec = VNOVAL;
697 	vap->va_mtime.tv_nsec = VNOVAL;
698 	vap->va_ctime.tv_sec = VNOVAL;
699 	vap->va_ctime.tv_nsec = VNOVAL;
700 	vap->va_birthtime.tv_sec = VNOVAL;
701 	vap->va_birthtime.tv_nsec = VNOVAL;
702 	vap->va_flags = VNOVAL;
703 	vap->va_gen = VNOVAL;
704 	vap->va_vaflags = 0;
705 }
706 
707 /*
708  * This routine is called when we have too many vnodes.  It attempts
709  * to free <count> vnodes and will potentially free vnodes that still
710  * have VM backing store (VM backing store is typically the cause
711  * of a vnode blowout so we want to do this).  Therefore, this operation
712  * is not considered cheap.
713  *
714  * A number of conditions may prevent a vnode from being reclaimed.
715  * the buffer cache may have references on the vnode, a directory
716  * vnode may still have references due to the namei cache representing
717  * underlying files, or the vnode may be in active use.   It is not
718  * desireable to reuse such vnodes.  These conditions may cause the
719  * number of vnodes to reach some minimum value regardless of what
720  * you set kern.maxvnodes to.  Do not set kern.maxvnodes too low.
721  */
722 static int
723 vlrureclaim(struct mount *mp)
724 {
725 	struct vnode *vp;
726 	int done;
727 	int trigger;
728 	int usevnodes;
729 	int count;
730 
731 	/*
732 	 * Calculate the trigger point, don't allow user
733 	 * screwups to blow us up.   This prevents us from
734 	 * recycling vnodes with lots of resident pages.  We
735 	 * aren't trying to free memory, we are trying to
736 	 * free vnodes.
737 	 */
738 	usevnodes = desiredvnodes;
739 	if (usevnodes <= 0)
740 		usevnodes = 1;
741 	trigger = vm_cnt.v_page_count * 2 / usevnodes;
742 	done = 0;
743 	vn_start_write(NULL, &mp, V_WAIT);
744 	MNT_ILOCK(mp);
745 	count = mp->mnt_nvnodelistsize / 10 + 1;
746 	while (count != 0) {
747 		vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
748 		while (vp != NULL && vp->v_type == VMARKER)
749 			vp = TAILQ_NEXT(vp, v_nmntvnodes);
750 		if (vp == NULL)
751 			break;
752 		TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
753 		TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
754 		--count;
755 		if (!VI_TRYLOCK(vp))
756 			goto next_iter;
757 		/*
758 		 * If it's been deconstructed already, it's still
759 		 * referenced, or it exceeds the trigger, skip it.
760 		 */
761 		if (vp->v_usecount ||
762 		    (!vlru_allow_cache_src &&
763 			!LIST_EMPTY(&(vp)->v_cache_src)) ||
764 		    (vp->v_iflag & VI_DOOMED) != 0 || (vp->v_object != NULL &&
765 		    vp->v_object->resident_page_count > trigger)) {
766 			VI_UNLOCK(vp);
767 			goto next_iter;
768 		}
769 		MNT_IUNLOCK(mp);
770 		vholdl(vp);
771 		if (VOP_LOCK(vp, LK_INTERLOCK|LK_EXCLUSIVE|LK_NOWAIT)) {
772 			vdrop(vp);
773 			goto next_iter_mntunlocked;
774 		}
775 		VI_LOCK(vp);
776 		/*
777 		 * v_usecount may have been bumped after VOP_LOCK() dropped
778 		 * the vnode interlock and before it was locked again.
779 		 *
780 		 * It is not necessary to recheck VI_DOOMED because it can
781 		 * only be set by another thread that holds both the vnode
782 		 * lock and vnode interlock.  If another thread has the
783 		 * vnode lock before we get to VOP_LOCK() and obtains the
784 		 * vnode interlock after VOP_LOCK() drops the vnode
785 		 * interlock, the other thread will be unable to drop the
786 		 * vnode lock before our VOP_LOCK() call fails.
787 		 */
788 		if (vp->v_usecount ||
789 		    (!vlru_allow_cache_src &&
790 			!LIST_EMPTY(&(vp)->v_cache_src)) ||
791 		    (vp->v_object != NULL &&
792 		    vp->v_object->resident_page_count > trigger)) {
793 			VOP_UNLOCK(vp, LK_INTERLOCK);
794 			vdrop(vp);
795 			goto next_iter_mntunlocked;
796 		}
797 		KASSERT((vp->v_iflag & VI_DOOMED) == 0,
798 		    ("VI_DOOMED unexpectedly detected in vlrureclaim()"));
799 		atomic_add_long(&recycles_count, 1);
800 		vgonel(vp);
801 		VOP_UNLOCK(vp, 0);
802 		vdropl(vp);
803 		done++;
804 next_iter_mntunlocked:
805 		if (!should_yield())
806 			goto relock_mnt;
807 		goto yield;
808 next_iter:
809 		if (!should_yield())
810 			continue;
811 		MNT_IUNLOCK(mp);
812 yield:
813 		kern_yield(PRI_USER);
814 relock_mnt:
815 		MNT_ILOCK(mp);
816 	}
817 	MNT_IUNLOCK(mp);
818 	vn_finished_write(mp);
819 	return done;
820 }
821 
822 /*
823  * Attempt to keep the free list at wantfreevnodes length.
824  */
825 static void
826 vnlru_free(int count)
827 {
828 	struct vnode *vp;
829 
830 	mtx_assert(&vnode_free_list_mtx, MA_OWNED);
831 	for (; count > 0; count--) {
832 		vp = TAILQ_FIRST(&vnode_free_list);
833 		/*
834 		 * The list can be modified while the free_list_mtx
835 		 * has been dropped and vp could be NULL here.
836 		 */
837 		if (!vp)
838 			break;
839 		VNASSERT(vp->v_op != NULL, vp,
840 		    ("vnlru_free: vnode already reclaimed."));
841 		KASSERT((vp->v_iflag & VI_FREE) != 0,
842 		    ("Removing vnode not on freelist"));
843 		KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
844 		    ("Mangling active vnode"));
845 		TAILQ_REMOVE(&vnode_free_list, vp, v_actfreelist);
846 		/*
847 		 * Don't recycle if we can't get the interlock.
848 		 */
849 		if (!VI_TRYLOCK(vp)) {
850 			TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_actfreelist);
851 			continue;
852 		}
853 		VNASSERT((vp->v_iflag & VI_FREE) != 0 && vp->v_holdcnt == 0,
854 		    vp, ("vp inconsistent on freelist"));
855 
856 		/*
857 		 * The clear of VI_FREE prevents activation of the
858 		 * vnode.  There is no sense in putting the vnode on
859 		 * the mount point active list, only to remove it
860 		 * later during recycling.  Inline the relevant part
861 		 * of vholdl(), to avoid triggering assertions or
862 		 * activating.
863 		 */
864 		freevnodes--;
865 		vp->v_iflag &= ~VI_FREE;
866 		vp->v_holdcnt++;
867 
868 		mtx_unlock(&vnode_free_list_mtx);
869 		VI_UNLOCK(vp);
870 		vtryrecycle(vp);
871 		/*
872 		 * If the recycled succeeded this vdrop will actually free
873 		 * the vnode.  If not it will simply place it back on
874 		 * the free list.
875 		 */
876 		vdrop(vp);
877 		mtx_lock(&vnode_free_list_mtx);
878 	}
879 }
880 /*
881  * Attempt to recycle vnodes in a context that is always safe to block.
882  * Calling vlrurecycle() from the bowels of filesystem code has some
883  * interesting deadlock problems.
884  */
885 static struct proc *vnlruproc;
886 static int vnlruproc_sig;
887 
888 static void
889 vnlru_proc(void)
890 {
891 	struct mount *mp, *nmp;
892 	int done;
893 	struct proc *p = vnlruproc;
894 
895 	EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
896 	    SHUTDOWN_PRI_FIRST);
897 
898 	for (;;) {
899 		kproc_suspend_check(p);
900 		mtx_lock(&vnode_free_list_mtx);
901 		if (freevnodes > wantfreevnodes)
902 			vnlru_free(freevnodes - wantfreevnodes);
903 		if (numvnodes <= desiredvnodes * 9 / 10) {
904 			vnlruproc_sig = 0;
905 			wakeup(&vnlruproc_sig);
906 			msleep(vnlruproc, &vnode_free_list_mtx,
907 			    PVFS|PDROP, "vlruwt", hz);
908 			continue;
909 		}
910 		mtx_unlock(&vnode_free_list_mtx);
911 		done = 0;
912 		mtx_lock(&mountlist_mtx);
913 		for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
914 			if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) {
915 				nmp = TAILQ_NEXT(mp, mnt_list);
916 				continue;
917 			}
918 			done += vlrureclaim(mp);
919 			mtx_lock(&mountlist_mtx);
920 			nmp = TAILQ_NEXT(mp, mnt_list);
921 			vfs_unbusy(mp);
922 		}
923 		mtx_unlock(&mountlist_mtx);
924 		if (done == 0) {
925 #if 0
926 			/* These messages are temporary debugging aids */
927 			if (vnlru_nowhere < 5)
928 				printf("vnlru process getting nowhere..\n");
929 			else if (vnlru_nowhere == 5)
930 				printf("vnlru process messages stopped.\n");
931 #endif
932 			vnlru_nowhere++;
933 			tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
934 		} else
935 			kern_yield(PRI_USER);
936 	}
937 }
938 
939 static struct kproc_desc vnlru_kp = {
940 	"vnlru",
941 	vnlru_proc,
942 	&vnlruproc
943 };
944 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start,
945     &vnlru_kp);
946 
947 /*
948  * Routines having to do with the management of the vnode table.
949  */
950 
951 /*
952  * Try to recycle a freed vnode.  We abort if anyone picks up a reference
953  * before we actually vgone().  This function must be called with the vnode
954  * held to prevent the vnode from being returned to the free list midway
955  * through vgone().
956  */
957 static int
958 vtryrecycle(struct vnode *vp)
959 {
960 	struct mount *vnmp;
961 
962 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
963 	VNASSERT(vp->v_holdcnt, vp,
964 	    ("vtryrecycle: Recycling vp %p without a reference.", vp));
965 	/*
966 	 * This vnode may found and locked via some other list, if so we
967 	 * can't recycle it yet.
968 	 */
969 	if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
970 		CTR2(KTR_VFS,
971 		    "%s: impossible to recycle, vp %p lock is already held",
972 		    __func__, vp);
973 		return (EWOULDBLOCK);
974 	}
975 	/*
976 	 * Don't recycle if its filesystem is being suspended.
977 	 */
978 	if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) {
979 		VOP_UNLOCK(vp, 0);
980 		CTR2(KTR_VFS,
981 		    "%s: impossible to recycle, cannot start the write for %p",
982 		    __func__, vp);
983 		return (EBUSY);
984 	}
985 	/*
986 	 * If we got this far, we need to acquire the interlock and see if
987 	 * anyone picked up this vnode from another list.  If not, we will
988 	 * mark it with DOOMED via vgonel() so that anyone who does find it
989 	 * will skip over it.
990 	 */
991 	VI_LOCK(vp);
992 	if (vp->v_usecount) {
993 		VOP_UNLOCK(vp, LK_INTERLOCK);
994 		vn_finished_write(vnmp);
995 		CTR2(KTR_VFS,
996 		    "%s: impossible to recycle, %p is already referenced",
997 		    __func__, vp);
998 		return (EBUSY);
999 	}
1000 	if ((vp->v_iflag & VI_DOOMED) == 0) {
1001 		atomic_add_long(&recycles_count, 1);
1002 		vgonel(vp);
1003 	}
1004 	VOP_UNLOCK(vp, LK_INTERLOCK);
1005 	vn_finished_write(vnmp);
1006 	return (0);
1007 }
1008 
1009 /*
1010  * Wait for available vnodes.
1011  */
1012 static int
1013 getnewvnode_wait(int suspended)
1014 {
1015 
1016 	mtx_assert(&vnode_free_list_mtx, MA_OWNED);
1017 	if (numvnodes > desiredvnodes) {
1018 		if (suspended) {
1019 			/*
1020 			 * File system is beeing suspended, we cannot risk a
1021 			 * deadlock here, so allocate new vnode anyway.
1022 			 */
1023 			if (freevnodes > wantfreevnodes)
1024 				vnlru_free(freevnodes - wantfreevnodes);
1025 			return (0);
1026 		}
1027 		if (vnlruproc_sig == 0) {
1028 			vnlruproc_sig = 1;	/* avoid unnecessary wakeups */
1029 			wakeup(vnlruproc);
1030 		}
1031 		msleep(&vnlruproc_sig, &vnode_free_list_mtx, PVFS,
1032 		    "vlruwk", hz);
1033 	}
1034 	return (numvnodes > desiredvnodes ? ENFILE : 0);
1035 }
1036 
1037 void
1038 getnewvnode_reserve(u_int count)
1039 {
1040 	struct thread *td;
1041 
1042 	td = curthread;
1043 	/* First try to be quick and racy. */
1044 	if (atomic_fetchadd_long(&numvnodes, count) + count <= desiredvnodes) {
1045 		td->td_vp_reserv += count;
1046 		return;
1047 	} else
1048 		atomic_subtract_long(&numvnodes, count);
1049 
1050 	mtx_lock(&vnode_free_list_mtx);
1051 	while (count > 0) {
1052 		if (getnewvnode_wait(0) == 0) {
1053 			count--;
1054 			td->td_vp_reserv++;
1055 			atomic_add_long(&numvnodes, 1);
1056 		}
1057 	}
1058 	mtx_unlock(&vnode_free_list_mtx);
1059 }
1060 
1061 void
1062 getnewvnode_drop_reserve(void)
1063 {
1064 	struct thread *td;
1065 
1066 	td = curthread;
1067 	atomic_subtract_long(&numvnodes, td->td_vp_reserv);
1068 	td->td_vp_reserv = 0;
1069 }
1070 
1071 /*
1072  * Return the next vnode from the free list.
1073  */
1074 int
1075 getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops,
1076     struct vnode **vpp)
1077 {
1078 	struct vnode *vp;
1079 	struct bufobj *bo;
1080 	struct thread *td;
1081 	int error;
1082 
1083 	CTR3(KTR_VFS, "%s: mp %p with tag %s", __func__, mp, tag);
1084 	vp = NULL;
1085 	td = curthread;
1086 	if (td->td_vp_reserv > 0) {
1087 		td->td_vp_reserv -= 1;
1088 		goto alloc;
1089 	}
1090 	mtx_lock(&vnode_free_list_mtx);
1091 	/*
1092 	 * Lend our context to reclaim vnodes if they've exceeded the max.
1093 	 */
1094 	if (freevnodes > wantfreevnodes)
1095 		vnlru_free(1);
1096 	error = getnewvnode_wait(mp != NULL && (mp->mnt_kern_flag &
1097 	    MNTK_SUSPEND));
1098 #if 0	/* XXX Not all VFS_VGET/ffs_vget callers check returns. */
1099 	if (error != 0) {
1100 		mtx_unlock(&vnode_free_list_mtx);
1101 		return (error);
1102 	}
1103 #endif
1104 	atomic_add_long(&numvnodes, 1);
1105 	mtx_unlock(&vnode_free_list_mtx);
1106 alloc:
1107 	atomic_add_long(&vnodes_created, 1);
1108 	vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO);
1109 	/*
1110 	 * Setup locks.
1111 	 */
1112 	vp->v_vnlock = &vp->v_lock;
1113 	mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
1114 	/*
1115 	 * By default, don't allow shared locks unless filesystems
1116 	 * opt-in.
1117 	 */
1118 	lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOSHARE | LK_IS_VNODE);
1119 	/*
1120 	 * Initialize bufobj.
1121 	 */
1122 	bo = &vp->v_bufobj;
1123 	bo->__bo_vnode = vp;
1124 	rw_init(BO_LOCKPTR(bo), "bufobj interlock");
1125 	bo->bo_ops = &buf_ops_bio;
1126 	bo->bo_private = vp;
1127 	TAILQ_INIT(&bo->bo_clean.bv_hd);
1128 	TAILQ_INIT(&bo->bo_dirty.bv_hd);
1129 	/*
1130 	 * Initialize namecache.
1131 	 */
1132 	LIST_INIT(&vp->v_cache_src);
1133 	TAILQ_INIT(&vp->v_cache_dst);
1134 	/*
1135 	 * Finalize various vnode identity bits.
1136 	 */
1137 	vp->v_type = VNON;
1138 	vp->v_tag = tag;
1139 	vp->v_op = vops;
1140 	v_incr_usecount(vp);
1141 	vp->v_data = NULL;
1142 #ifdef MAC
1143 	mac_vnode_init(vp);
1144 	if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0)
1145 		mac_vnode_associate_singlelabel(mp, vp);
1146 	else if (mp == NULL && vops != &dead_vnodeops)
1147 		printf("NULL mp in getnewvnode()\n");
1148 #endif
1149 	if (mp != NULL) {
1150 		bo->bo_bsize = mp->mnt_stat.f_iosize;
1151 		if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0)
1152 			vp->v_vflag |= VV_NOKNOTE;
1153 	}
1154 	rangelock_init(&vp->v_rl);
1155 
1156 	/*
1157 	 * For the filesystems which do not use vfs_hash_insert(),
1158 	 * still initialize v_hash to have vfs_hash_index() useful.
1159 	 * E.g., nullfs uses vfs_hash_index() on the lower vnode for
1160 	 * its own hashing.
1161 	 */
1162 	vp->v_hash = (uintptr_t)vp >> vnsz2log;
1163 
1164 	*vpp = vp;
1165 	return (0);
1166 }
1167 
1168 /*
1169  * Delete from old mount point vnode list, if on one.
1170  */
1171 static void
1172 delmntque(struct vnode *vp)
1173 {
1174 	struct mount *mp;
1175 	int active;
1176 
1177 	mp = vp->v_mount;
1178 	if (mp == NULL)
1179 		return;
1180 	MNT_ILOCK(mp);
1181 	VI_LOCK(vp);
1182 	KASSERT(mp->mnt_activevnodelistsize <= mp->mnt_nvnodelistsize,
1183 	    ("Active vnode list size %d > Vnode list size %d",
1184 	     mp->mnt_activevnodelistsize, mp->mnt_nvnodelistsize));
1185 	active = vp->v_iflag & VI_ACTIVE;
1186 	vp->v_iflag &= ~VI_ACTIVE;
1187 	if (active) {
1188 		mtx_lock(&vnode_free_list_mtx);
1189 		TAILQ_REMOVE(&mp->mnt_activevnodelist, vp, v_actfreelist);
1190 		mp->mnt_activevnodelistsize--;
1191 		mtx_unlock(&vnode_free_list_mtx);
1192 	}
1193 	vp->v_mount = NULL;
1194 	VI_UNLOCK(vp);
1195 	VNASSERT(mp->mnt_nvnodelistsize > 0, vp,
1196 		("bad mount point vnode list size"));
1197 	TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1198 	mp->mnt_nvnodelistsize--;
1199 	MNT_REL(mp);
1200 	MNT_IUNLOCK(mp);
1201 }
1202 
1203 static void
1204 insmntque_stddtr(struct vnode *vp, void *dtr_arg)
1205 {
1206 
1207 	vp->v_data = NULL;
1208 	vp->v_op = &dead_vnodeops;
1209 	vgone(vp);
1210 	vput(vp);
1211 }
1212 
1213 /*
1214  * Insert into list of vnodes for the new mount point, if available.
1215  */
1216 int
1217 insmntque1(struct vnode *vp, struct mount *mp,
1218 	void (*dtr)(struct vnode *, void *), void *dtr_arg)
1219 {
1220 
1221 	KASSERT(vp->v_mount == NULL,
1222 		("insmntque: vnode already on per mount vnode list"));
1223 	VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)"));
1224 	ASSERT_VOP_ELOCKED(vp, "insmntque: non-locked vp");
1225 
1226 	/*
1227 	 * We acquire the vnode interlock early to ensure that the
1228 	 * vnode cannot be recycled by another process releasing a
1229 	 * holdcnt on it before we get it on both the vnode list
1230 	 * and the active vnode list. The mount mutex protects only
1231 	 * manipulation of the vnode list and the vnode freelist
1232 	 * mutex protects only manipulation of the active vnode list.
1233 	 * Hence the need to hold the vnode interlock throughout.
1234 	 */
1235 	MNT_ILOCK(mp);
1236 	VI_LOCK(vp);
1237 	if (((mp->mnt_kern_flag & MNTK_NOINSMNTQ) != 0 &&
1238 	    ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0 ||
1239 	    mp->mnt_nvnodelistsize == 0)) &&
1240 	    (vp->v_vflag & VV_FORCEINSMQ) == 0) {
1241 		VI_UNLOCK(vp);
1242 		MNT_IUNLOCK(mp);
1243 		if (dtr != NULL)
1244 			dtr(vp, dtr_arg);
1245 		return (EBUSY);
1246 	}
1247 	vp->v_mount = mp;
1248 	MNT_REF(mp);
1249 	TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1250 	VNASSERT(mp->mnt_nvnodelistsize >= 0, vp,
1251 		("neg mount point vnode list size"));
1252 	mp->mnt_nvnodelistsize++;
1253 	KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
1254 	    ("Activating already active vnode"));
1255 	vp->v_iflag |= VI_ACTIVE;
1256 	mtx_lock(&vnode_free_list_mtx);
1257 	TAILQ_INSERT_HEAD(&mp->mnt_activevnodelist, vp, v_actfreelist);
1258 	mp->mnt_activevnodelistsize++;
1259 	mtx_unlock(&vnode_free_list_mtx);
1260 	VI_UNLOCK(vp);
1261 	MNT_IUNLOCK(mp);
1262 	return (0);
1263 }
1264 
1265 int
1266 insmntque(struct vnode *vp, struct mount *mp)
1267 {
1268 
1269 	return (insmntque1(vp, mp, insmntque_stddtr, NULL));
1270 }
1271 
1272 /*
1273  * Flush out and invalidate all buffers associated with a bufobj
1274  * Called with the underlying object locked.
1275  */
1276 int
1277 bufobj_invalbuf(struct bufobj *bo, int flags, int slpflag, int slptimeo)
1278 {
1279 	int error;
1280 
1281 	BO_LOCK(bo);
1282 	if (flags & V_SAVE) {
1283 		error = bufobj_wwait(bo, slpflag, slptimeo);
1284 		if (error) {
1285 			BO_UNLOCK(bo);
1286 			return (error);
1287 		}
1288 		if (bo->bo_dirty.bv_cnt > 0) {
1289 			BO_UNLOCK(bo);
1290 			if ((error = BO_SYNC(bo, MNT_WAIT)) != 0)
1291 				return (error);
1292 			/*
1293 			 * XXX We could save a lock/unlock if this was only
1294 			 * enabled under INVARIANTS
1295 			 */
1296 			BO_LOCK(bo);
1297 			if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)
1298 				panic("vinvalbuf: dirty bufs");
1299 		}
1300 	}
1301 	/*
1302 	 * If you alter this loop please notice that interlock is dropped and
1303 	 * reacquired in flushbuflist.  Special care is needed to ensure that
1304 	 * no race conditions occur from this.
1305 	 */
1306 	do {
1307 		error = flushbuflist(&bo->bo_clean,
1308 		    flags, bo, slpflag, slptimeo);
1309 		if (error == 0 && !(flags & V_CLEANONLY))
1310 			error = flushbuflist(&bo->bo_dirty,
1311 			    flags, bo, slpflag, slptimeo);
1312 		if (error != 0 && error != EAGAIN) {
1313 			BO_UNLOCK(bo);
1314 			return (error);
1315 		}
1316 	} while (error != 0);
1317 
1318 	/*
1319 	 * Wait for I/O to complete.  XXX needs cleaning up.  The vnode can
1320 	 * have write I/O in-progress but if there is a VM object then the
1321 	 * VM object can also have read-I/O in-progress.
1322 	 */
1323 	do {
1324 		bufobj_wwait(bo, 0, 0);
1325 		BO_UNLOCK(bo);
1326 		if (bo->bo_object != NULL) {
1327 			VM_OBJECT_WLOCK(bo->bo_object);
1328 			vm_object_pip_wait(bo->bo_object, "bovlbx");
1329 			VM_OBJECT_WUNLOCK(bo->bo_object);
1330 		}
1331 		BO_LOCK(bo);
1332 	} while (bo->bo_numoutput > 0);
1333 	BO_UNLOCK(bo);
1334 
1335 	/*
1336 	 * Destroy the copy in the VM cache, too.
1337 	 */
1338 	if (bo->bo_object != NULL &&
1339 	    (flags & (V_ALT | V_NORMAL | V_CLEANONLY)) == 0) {
1340 		VM_OBJECT_WLOCK(bo->bo_object);
1341 		vm_object_page_remove(bo->bo_object, 0, 0, (flags & V_SAVE) ?
1342 		    OBJPR_CLEANONLY : 0);
1343 		VM_OBJECT_WUNLOCK(bo->bo_object);
1344 	}
1345 
1346 #ifdef INVARIANTS
1347 	BO_LOCK(bo);
1348 	if ((flags & (V_ALT | V_NORMAL | V_CLEANONLY)) == 0 &&
1349 	    (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0))
1350 		panic("vinvalbuf: flush failed");
1351 	BO_UNLOCK(bo);
1352 #endif
1353 	return (0);
1354 }
1355 
1356 /*
1357  * Flush out and invalidate all buffers associated with a vnode.
1358  * Called with the underlying object locked.
1359  */
1360 int
1361 vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo)
1362 {
1363 
1364 	CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
1365 	ASSERT_VOP_LOCKED(vp, "vinvalbuf");
1366 	if (vp->v_object != NULL && vp->v_object->handle != vp)
1367 		return (0);
1368 	return (bufobj_invalbuf(&vp->v_bufobj, flags, slpflag, slptimeo));
1369 }
1370 
1371 /*
1372  * Flush out buffers on the specified list.
1373  *
1374  */
1375 static int
1376 flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo, int slpflag,
1377     int slptimeo)
1378 {
1379 	struct buf *bp, *nbp;
1380 	int retval, error;
1381 	daddr_t lblkno;
1382 	b_xflags_t xflags;
1383 
1384 	ASSERT_BO_WLOCKED(bo);
1385 
1386 	retval = 0;
1387 	TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) {
1388 		if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) ||
1389 		    ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) {
1390 			continue;
1391 		}
1392 		lblkno = 0;
1393 		xflags = 0;
1394 		if (nbp != NULL) {
1395 			lblkno = nbp->b_lblkno;
1396 			xflags = nbp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN);
1397 		}
1398 		retval = EAGAIN;
1399 		error = BUF_TIMELOCK(bp,
1400 		    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_LOCKPTR(bo),
1401 		    "flushbuf", slpflag, slptimeo);
1402 		if (error) {
1403 			BO_LOCK(bo);
1404 			return (error != ENOLCK ? error : EAGAIN);
1405 		}
1406 		KASSERT(bp->b_bufobj == bo,
1407 		    ("bp %p wrong b_bufobj %p should be %p",
1408 		    bp, bp->b_bufobj, bo));
1409 		if (bp->b_bufobj != bo) {	/* XXX: necessary ? */
1410 			BUF_UNLOCK(bp);
1411 			BO_LOCK(bo);
1412 			return (EAGAIN);
1413 		}
1414 		/*
1415 		 * XXX Since there are no node locks for NFS, I
1416 		 * believe there is a slight chance that a delayed
1417 		 * write will occur while sleeping just above, so
1418 		 * check for it.
1419 		 */
1420 		if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
1421 		    (flags & V_SAVE)) {
1422 			bremfree(bp);
1423 			bp->b_flags |= B_ASYNC;
1424 			bwrite(bp);
1425 			BO_LOCK(bo);
1426 			return (EAGAIN);	/* XXX: why not loop ? */
1427 		}
1428 		bremfree(bp);
1429 		bp->b_flags |= (B_INVAL | B_RELBUF);
1430 		bp->b_flags &= ~B_ASYNC;
1431 		brelse(bp);
1432 		BO_LOCK(bo);
1433 		if (nbp != NULL &&
1434 		    (nbp->b_bufobj != bo ||
1435 		     nbp->b_lblkno != lblkno ||
1436 		     (nbp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) != xflags))
1437 			break;			/* nbp invalid */
1438 	}
1439 	return (retval);
1440 }
1441 
1442 /*
1443  * Truncate a file's buffer and pages to a specified length.  This
1444  * is in lieu of the old vinvalbuf mechanism, which performed unneeded
1445  * sync activity.
1446  */
1447 int
1448 vtruncbuf(struct vnode *vp, struct ucred *cred, off_t length, int blksize)
1449 {
1450 	struct buf *bp, *nbp;
1451 	int anyfreed;
1452 	int trunclbn;
1453 	struct bufobj *bo;
1454 
1455 	CTR5(KTR_VFS, "%s: vp %p with cred %p and block %d:%ju", __func__,
1456 	    vp, cred, blksize, (uintmax_t)length);
1457 
1458 	/*
1459 	 * Round up to the *next* lbn.
1460 	 */
1461 	trunclbn = (length + blksize - 1) / blksize;
1462 
1463 	ASSERT_VOP_LOCKED(vp, "vtruncbuf");
1464 restart:
1465 	bo = &vp->v_bufobj;
1466 	BO_LOCK(bo);
1467 	anyfreed = 1;
1468 	for (;anyfreed;) {
1469 		anyfreed = 0;
1470 		TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) {
1471 			if (bp->b_lblkno < trunclbn)
1472 				continue;
1473 			if (BUF_LOCK(bp,
1474 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1475 			    BO_LOCKPTR(bo)) == ENOLCK)
1476 				goto restart;
1477 
1478 			bremfree(bp);
1479 			bp->b_flags |= (B_INVAL | B_RELBUF);
1480 			bp->b_flags &= ~B_ASYNC;
1481 			brelse(bp);
1482 			anyfreed = 1;
1483 
1484 			BO_LOCK(bo);
1485 			if (nbp != NULL &&
1486 			    (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1487 			    (nbp->b_vp != vp) ||
1488 			    (nbp->b_flags & B_DELWRI))) {
1489 				BO_UNLOCK(bo);
1490 				goto restart;
1491 			}
1492 		}
1493 
1494 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1495 			if (bp->b_lblkno < trunclbn)
1496 				continue;
1497 			if (BUF_LOCK(bp,
1498 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1499 			    BO_LOCKPTR(bo)) == ENOLCK)
1500 				goto restart;
1501 			bremfree(bp);
1502 			bp->b_flags |= (B_INVAL | B_RELBUF);
1503 			bp->b_flags &= ~B_ASYNC;
1504 			brelse(bp);
1505 			anyfreed = 1;
1506 
1507 			BO_LOCK(bo);
1508 			if (nbp != NULL &&
1509 			    (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1510 			    (nbp->b_vp != vp) ||
1511 			    (nbp->b_flags & B_DELWRI) == 0)) {
1512 				BO_UNLOCK(bo);
1513 				goto restart;
1514 			}
1515 		}
1516 	}
1517 
1518 	if (length > 0) {
1519 restartsync:
1520 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1521 			if (bp->b_lblkno > 0)
1522 				continue;
1523 			/*
1524 			 * Since we hold the vnode lock this should only
1525 			 * fail if we're racing with the buf daemon.
1526 			 */
1527 			if (BUF_LOCK(bp,
1528 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1529 			    BO_LOCKPTR(bo)) == ENOLCK) {
1530 				goto restart;
1531 			}
1532 			VNASSERT((bp->b_flags & B_DELWRI), vp,
1533 			    ("buf(%p) on dirty queue without DELWRI", bp));
1534 
1535 			bremfree(bp);
1536 			bawrite(bp);
1537 			BO_LOCK(bo);
1538 			goto restartsync;
1539 		}
1540 	}
1541 
1542 	bufobj_wwait(bo, 0, 0);
1543 	BO_UNLOCK(bo);
1544 	vnode_pager_setsize(vp, length);
1545 
1546 	return (0);
1547 }
1548 
1549 static void
1550 buf_vlist_remove(struct buf *bp)
1551 {
1552 	struct bufv *bv;
1553 
1554 	KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp));
1555 	ASSERT_BO_WLOCKED(bp->b_bufobj);
1556 	KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) !=
1557 	    (BX_VNDIRTY|BX_VNCLEAN),
1558 	    ("buf_vlist_remove: Buf %p is on two lists", bp));
1559 	if (bp->b_xflags & BX_VNDIRTY)
1560 		bv = &bp->b_bufobj->bo_dirty;
1561 	else
1562 		bv = &bp->b_bufobj->bo_clean;
1563 	BUF_PCTRIE_REMOVE(&bv->bv_root, bp->b_lblkno);
1564 	TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs);
1565 	bv->bv_cnt--;
1566 	bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1567 }
1568 
1569 /*
1570  * Add the buffer to the sorted clean or dirty block list.
1571  *
1572  * NOTE: xflags is passed as a constant, optimizing this inline function!
1573  */
1574 static void
1575 buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags)
1576 {
1577 	struct bufv *bv;
1578 	struct buf *n;
1579 	int error;
1580 
1581 	ASSERT_BO_WLOCKED(bo);
1582 	KASSERT((bo->bo_flag & BO_DEAD) == 0, ("dead bo %p", bo));
1583 	KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0,
1584 	    ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags));
1585 	bp->b_xflags |= xflags;
1586 	if (xflags & BX_VNDIRTY)
1587 		bv = &bo->bo_dirty;
1588 	else
1589 		bv = &bo->bo_clean;
1590 
1591 	/*
1592 	 * Keep the list ordered.  Optimize empty list insertion.  Assume
1593 	 * we tend to grow at the tail so lookup_le should usually be cheaper
1594 	 * than _ge.
1595 	 */
1596 	if (bv->bv_cnt == 0 ||
1597 	    bp->b_lblkno > TAILQ_LAST(&bv->bv_hd, buflists)->b_lblkno)
1598 		TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs);
1599 	else if ((n = BUF_PCTRIE_LOOKUP_LE(&bv->bv_root, bp->b_lblkno)) == NULL)
1600 		TAILQ_INSERT_HEAD(&bv->bv_hd, bp, b_bobufs);
1601 	else
1602 		TAILQ_INSERT_AFTER(&bv->bv_hd, n, bp, b_bobufs);
1603 	error = BUF_PCTRIE_INSERT(&bv->bv_root, bp);
1604 	if (error)
1605 		panic("buf_vlist_add:  Preallocated nodes insufficient.");
1606 	bv->bv_cnt++;
1607 }
1608 
1609 /*
1610  * Lookup a buffer using the splay tree.  Note that we specifically avoid
1611  * shadow buffers used in background bitmap writes.
1612  *
1613  * This code isn't quite efficient as it could be because we are maintaining
1614  * two sorted lists and do not know which list the block resides in.
1615  *
1616  * During a "make buildworld" the desired buffer is found at one of
1617  * the roots more than 60% of the time.  Thus, checking both roots
1618  * before performing either splay eliminates unnecessary splays on the
1619  * first tree splayed.
1620  */
1621 struct buf *
1622 gbincore(struct bufobj *bo, daddr_t lblkno)
1623 {
1624 	struct buf *bp;
1625 
1626 	ASSERT_BO_LOCKED(bo);
1627 	bp = BUF_PCTRIE_LOOKUP(&bo->bo_clean.bv_root, lblkno);
1628 	if (bp != NULL)
1629 		return (bp);
1630 	return BUF_PCTRIE_LOOKUP(&bo->bo_dirty.bv_root, lblkno);
1631 }
1632 
1633 /*
1634  * Associate a buffer with a vnode.
1635  */
1636 void
1637 bgetvp(struct vnode *vp, struct buf *bp)
1638 {
1639 	struct bufobj *bo;
1640 
1641 	bo = &vp->v_bufobj;
1642 	ASSERT_BO_WLOCKED(bo);
1643 	VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free"));
1644 
1645 	CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags);
1646 	VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp,
1647 	    ("bgetvp: bp already attached! %p", bp));
1648 
1649 	vhold(vp);
1650 	bp->b_vp = vp;
1651 	bp->b_bufobj = bo;
1652 	/*
1653 	 * Insert onto list for new vnode.
1654 	 */
1655 	buf_vlist_add(bp, bo, BX_VNCLEAN);
1656 }
1657 
1658 /*
1659  * Disassociate a buffer from a vnode.
1660  */
1661 void
1662 brelvp(struct buf *bp)
1663 {
1664 	struct bufobj *bo;
1665 	struct vnode *vp;
1666 
1667 	CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
1668 	KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1669 
1670 	/*
1671 	 * Delete from old vnode list, if on one.
1672 	 */
1673 	vp = bp->b_vp;		/* XXX */
1674 	bo = bp->b_bufobj;
1675 	BO_LOCK(bo);
1676 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1677 		buf_vlist_remove(bp);
1678 	else
1679 		panic("brelvp: Buffer %p not on queue.", bp);
1680 	if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1681 		bo->bo_flag &= ~BO_ONWORKLST;
1682 		mtx_lock(&sync_mtx);
1683 		LIST_REMOVE(bo, bo_synclist);
1684 		syncer_worklist_len--;
1685 		mtx_unlock(&sync_mtx);
1686 	}
1687 	bp->b_vp = NULL;
1688 	bp->b_bufobj = NULL;
1689 	BO_UNLOCK(bo);
1690 	vdrop(vp);
1691 }
1692 
1693 /*
1694  * Add an item to the syncer work queue.
1695  */
1696 static void
1697 vn_syncer_add_to_worklist(struct bufobj *bo, int delay)
1698 {
1699 	int slot;
1700 
1701 	ASSERT_BO_WLOCKED(bo);
1702 
1703 	mtx_lock(&sync_mtx);
1704 	if (bo->bo_flag & BO_ONWORKLST)
1705 		LIST_REMOVE(bo, bo_synclist);
1706 	else {
1707 		bo->bo_flag |= BO_ONWORKLST;
1708 		syncer_worklist_len++;
1709 	}
1710 
1711 	if (delay > syncer_maxdelay - 2)
1712 		delay = syncer_maxdelay - 2;
1713 	slot = (syncer_delayno + delay) & syncer_mask;
1714 
1715 	LIST_INSERT_HEAD(&syncer_workitem_pending[slot], bo, bo_synclist);
1716 	mtx_unlock(&sync_mtx);
1717 }
1718 
1719 static int
1720 sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS)
1721 {
1722 	int error, len;
1723 
1724 	mtx_lock(&sync_mtx);
1725 	len = syncer_worklist_len - sync_vnode_count;
1726 	mtx_unlock(&sync_mtx);
1727 	error = SYSCTL_OUT(req, &len, sizeof(len));
1728 	return (error);
1729 }
1730 
1731 SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0,
1732     sysctl_vfs_worklist_len, "I", "Syncer thread worklist length");
1733 
1734 static struct proc *updateproc;
1735 static void sched_sync(void);
1736 static struct kproc_desc up_kp = {
1737 	"syncer",
1738 	sched_sync,
1739 	&updateproc
1740 };
1741 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp);
1742 
1743 static int
1744 sync_vnode(struct synclist *slp, struct bufobj **bo, struct thread *td)
1745 {
1746 	struct vnode *vp;
1747 	struct mount *mp;
1748 
1749 	*bo = LIST_FIRST(slp);
1750 	if (*bo == NULL)
1751 		return (0);
1752 	vp = (*bo)->__bo_vnode;	/* XXX */
1753 	if (VOP_ISLOCKED(vp) != 0 || VI_TRYLOCK(vp) == 0)
1754 		return (1);
1755 	/*
1756 	 * We use vhold in case the vnode does not
1757 	 * successfully sync.  vhold prevents the vnode from
1758 	 * going away when we unlock the sync_mtx so that
1759 	 * we can acquire the vnode interlock.
1760 	 */
1761 	vholdl(vp);
1762 	mtx_unlock(&sync_mtx);
1763 	VI_UNLOCK(vp);
1764 	if (vn_start_write(vp, &mp, V_NOWAIT) != 0) {
1765 		vdrop(vp);
1766 		mtx_lock(&sync_mtx);
1767 		return (*bo == LIST_FIRST(slp));
1768 	}
1769 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1770 	(void) VOP_FSYNC(vp, MNT_LAZY, td);
1771 	VOP_UNLOCK(vp, 0);
1772 	vn_finished_write(mp);
1773 	BO_LOCK(*bo);
1774 	if (((*bo)->bo_flag & BO_ONWORKLST) != 0) {
1775 		/*
1776 		 * Put us back on the worklist.  The worklist
1777 		 * routine will remove us from our current
1778 		 * position and then add us back in at a later
1779 		 * position.
1780 		 */
1781 		vn_syncer_add_to_worklist(*bo, syncdelay);
1782 	}
1783 	BO_UNLOCK(*bo);
1784 	vdrop(vp);
1785 	mtx_lock(&sync_mtx);
1786 	return (0);
1787 }
1788 
1789 static int first_printf = 1;
1790 
1791 /*
1792  * System filesystem synchronizer daemon.
1793  */
1794 static void
1795 sched_sync(void)
1796 {
1797 	struct synclist *next, *slp;
1798 	struct bufobj *bo;
1799 	long starttime;
1800 	struct thread *td = curthread;
1801 	int last_work_seen;
1802 	int net_worklist_len;
1803 	int syncer_final_iter;
1804 	int error;
1805 
1806 	last_work_seen = 0;
1807 	syncer_final_iter = 0;
1808 	syncer_state = SYNCER_RUNNING;
1809 	starttime = time_uptime;
1810 	td->td_pflags |= TDP_NORUNNINGBUF;
1811 
1812 	EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc,
1813 	    SHUTDOWN_PRI_LAST);
1814 
1815 	mtx_lock(&sync_mtx);
1816 	for (;;) {
1817 		if (syncer_state == SYNCER_FINAL_DELAY &&
1818 		    syncer_final_iter == 0) {
1819 			mtx_unlock(&sync_mtx);
1820 			kproc_suspend_check(td->td_proc);
1821 			mtx_lock(&sync_mtx);
1822 		}
1823 		net_worklist_len = syncer_worklist_len - sync_vnode_count;
1824 		if (syncer_state != SYNCER_RUNNING &&
1825 		    starttime != time_uptime) {
1826 			if (first_printf) {
1827 				printf("\nSyncing disks, vnodes remaining...");
1828 				first_printf = 0;
1829 			}
1830 			printf("%d ", net_worklist_len);
1831 		}
1832 		starttime = time_uptime;
1833 
1834 		/*
1835 		 * Push files whose dirty time has expired.  Be careful
1836 		 * of interrupt race on slp queue.
1837 		 *
1838 		 * Skip over empty worklist slots when shutting down.
1839 		 */
1840 		do {
1841 			slp = &syncer_workitem_pending[syncer_delayno];
1842 			syncer_delayno += 1;
1843 			if (syncer_delayno == syncer_maxdelay)
1844 				syncer_delayno = 0;
1845 			next = &syncer_workitem_pending[syncer_delayno];
1846 			/*
1847 			 * If the worklist has wrapped since the
1848 			 * it was emptied of all but syncer vnodes,
1849 			 * switch to the FINAL_DELAY state and run
1850 			 * for one more second.
1851 			 */
1852 			if (syncer_state == SYNCER_SHUTTING_DOWN &&
1853 			    net_worklist_len == 0 &&
1854 			    last_work_seen == syncer_delayno) {
1855 				syncer_state = SYNCER_FINAL_DELAY;
1856 				syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP;
1857 			}
1858 		} while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) &&
1859 		    syncer_worklist_len > 0);
1860 
1861 		/*
1862 		 * Keep track of the last time there was anything
1863 		 * on the worklist other than syncer vnodes.
1864 		 * Return to the SHUTTING_DOWN state if any
1865 		 * new work appears.
1866 		 */
1867 		if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING)
1868 			last_work_seen = syncer_delayno;
1869 		if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY)
1870 			syncer_state = SYNCER_SHUTTING_DOWN;
1871 		while (!LIST_EMPTY(slp)) {
1872 			error = sync_vnode(slp, &bo, td);
1873 			if (error == 1) {
1874 				LIST_REMOVE(bo, bo_synclist);
1875 				LIST_INSERT_HEAD(next, bo, bo_synclist);
1876 				continue;
1877 			}
1878 
1879 			if (first_printf == 0) {
1880 				/*
1881 				 * Drop the sync mutex, because some watchdog
1882 				 * drivers need to sleep while patting
1883 				 */
1884 				mtx_unlock(&sync_mtx);
1885 				wdog_kern_pat(WD_LASTVAL);
1886 				mtx_lock(&sync_mtx);
1887 			}
1888 
1889 		}
1890 		if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0)
1891 			syncer_final_iter--;
1892 		/*
1893 		 * The variable rushjob allows the kernel to speed up the
1894 		 * processing of the filesystem syncer process. A rushjob
1895 		 * value of N tells the filesystem syncer to process the next
1896 		 * N seconds worth of work on its queue ASAP. Currently rushjob
1897 		 * is used by the soft update code to speed up the filesystem
1898 		 * syncer process when the incore state is getting so far
1899 		 * ahead of the disk that the kernel memory pool is being
1900 		 * threatened with exhaustion.
1901 		 */
1902 		if (rushjob > 0) {
1903 			rushjob -= 1;
1904 			continue;
1905 		}
1906 		/*
1907 		 * Just sleep for a short period of time between
1908 		 * iterations when shutting down to allow some I/O
1909 		 * to happen.
1910 		 *
1911 		 * If it has taken us less than a second to process the
1912 		 * current work, then wait. Otherwise start right over
1913 		 * again. We can still lose time if any single round
1914 		 * takes more than two seconds, but it does not really
1915 		 * matter as we are just trying to generally pace the
1916 		 * filesystem activity.
1917 		 */
1918 		if (syncer_state != SYNCER_RUNNING ||
1919 		    time_uptime == starttime) {
1920 			thread_lock(td);
1921 			sched_prio(td, PPAUSE);
1922 			thread_unlock(td);
1923 		}
1924 		if (syncer_state != SYNCER_RUNNING)
1925 			cv_timedwait(&sync_wakeup, &sync_mtx,
1926 			    hz / SYNCER_SHUTDOWN_SPEEDUP);
1927 		else if (time_uptime == starttime)
1928 			cv_timedwait(&sync_wakeup, &sync_mtx, hz);
1929 	}
1930 }
1931 
1932 /*
1933  * Request the syncer daemon to speed up its work.
1934  * We never push it to speed up more than half of its
1935  * normal turn time, otherwise it could take over the cpu.
1936  */
1937 int
1938 speedup_syncer(void)
1939 {
1940 	int ret = 0;
1941 
1942 	mtx_lock(&sync_mtx);
1943 	if (rushjob < syncdelay / 2) {
1944 		rushjob += 1;
1945 		stat_rush_requests += 1;
1946 		ret = 1;
1947 	}
1948 	mtx_unlock(&sync_mtx);
1949 	cv_broadcast(&sync_wakeup);
1950 	return (ret);
1951 }
1952 
1953 /*
1954  * Tell the syncer to speed up its work and run though its work
1955  * list several times, then tell it to shut down.
1956  */
1957 static void
1958 syncer_shutdown(void *arg, int howto)
1959 {
1960 
1961 	if (howto & RB_NOSYNC)
1962 		return;
1963 	mtx_lock(&sync_mtx);
1964 	syncer_state = SYNCER_SHUTTING_DOWN;
1965 	rushjob = 0;
1966 	mtx_unlock(&sync_mtx);
1967 	cv_broadcast(&sync_wakeup);
1968 	kproc_shutdown(arg, howto);
1969 }
1970 
1971 void
1972 syncer_suspend(void)
1973 {
1974 
1975 	syncer_shutdown(updateproc, 0);
1976 }
1977 
1978 void
1979 syncer_resume(void)
1980 {
1981 
1982 	mtx_lock(&sync_mtx);
1983 	first_printf = 1;
1984 	syncer_state = SYNCER_RUNNING;
1985 	mtx_unlock(&sync_mtx);
1986 	cv_broadcast(&sync_wakeup);
1987 	kproc_resume(updateproc);
1988 }
1989 
1990 /*
1991  * Reassign a buffer from one vnode to another.
1992  * Used to assign file specific control information
1993  * (indirect blocks) to the vnode to which they belong.
1994  */
1995 void
1996 reassignbuf(struct buf *bp)
1997 {
1998 	struct vnode *vp;
1999 	struct bufobj *bo;
2000 	int delay;
2001 #ifdef INVARIANTS
2002 	struct bufv *bv;
2003 #endif
2004 
2005 	vp = bp->b_vp;
2006 	bo = bp->b_bufobj;
2007 	++reassignbufcalls;
2008 
2009 	CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X",
2010 	    bp, bp->b_vp, bp->b_flags);
2011 	/*
2012 	 * B_PAGING flagged buffers cannot be reassigned because their vp
2013 	 * is not fully linked in.
2014 	 */
2015 	if (bp->b_flags & B_PAGING)
2016 		panic("cannot reassign paging buffer");
2017 
2018 	/*
2019 	 * Delete from old vnode list, if on one.
2020 	 */
2021 	BO_LOCK(bo);
2022 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
2023 		buf_vlist_remove(bp);
2024 	else
2025 		panic("reassignbuf: Buffer %p not on queue.", bp);
2026 	/*
2027 	 * If dirty, put on list of dirty buffers; otherwise insert onto list
2028 	 * of clean buffers.
2029 	 */
2030 	if (bp->b_flags & B_DELWRI) {
2031 		if ((bo->bo_flag & BO_ONWORKLST) == 0) {
2032 			switch (vp->v_type) {
2033 			case VDIR:
2034 				delay = dirdelay;
2035 				break;
2036 			case VCHR:
2037 				delay = metadelay;
2038 				break;
2039 			default:
2040 				delay = filedelay;
2041 			}
2042 			vn_syncer_add_to_worklist(bo, delay);
2043 		}
2044 		buf_vlist_add(bp, bo, BX_VNDIRTY);
2045 	} else {
2046 		buf_vlist_add(bp, bo, BX_VNCLEAN);
2047 
2048 		if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
2049 			mtx_lock(&sync_mtx);
2050 			LIST_REMOVE(bo, bo_synclist);
2051 			syncer_worklist_len--;
2052 			mtx_unlock(&sync_mtx);
2053 			bo->bo_flag &= ~BO_ONWORKLST;
2054 		}
2055 	}
2056 #ifdef INVARIANTS
2057 	bv = &bo->bo_clean;
2058 	bp = TAILQ_FIRST(&bv->bv_hd);
2059 	KASSERT(bp == NULL || bp->b_bufobj == bo,
2060 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2061 	bp = TAILQ_LAST(&bv->bv_hd, buflists);
2062 	KASSERT(bp == NULL || bp->b_bufobj == bo,
2063 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2064 	bv = &bo->bo_dirty;
2065 	bp = TAILQ_FIRST(&bv->bv_hd);
2066 	KASSERT(bp == NULL || bp->b_bufobj == bo,
2067 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2068 	bp = TAILQ_LAST(&bv->bv_hd, buflists);
2069 	KASSERT(bp == NULL || bp->b_bufobj == bo,
2070 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2071 #endif
2072 	BO_UNLOCK(bo);
2073 }
2074 
2075 /*
2076  * Increment the use and hold counts on the vnode, taking care to reference
2077  * the driver's usecount if this is a chardev.  The vholdl() will remove
2078  * the vnode from the free list if it is presently free.  Requires the
2079  * vnode interlock and returns with it held.
2080  */
2081 static void
2082 v_incr_usecount(struct vnode *vp)
2083 {
2084 
2085 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2086 	vholdl(vp);
2087 	vp->v_usecount++;
2088 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2089 		dev_lock();
2090 		vp->v_rdev->si_usecount++;
2091 		dev_unlock();
2092 	}
2093 }
2094 
2095 /*
2096  * Turn a holdcnt into a use+holdcnt such that only one call to
2097  * v_decr_usecount is needed.
2098  */
2099 static void
2100 v_upgrade_usecount(struct vnode *vp)
2101 {
2102 
2103 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2104 	vp->v_usecount++;
2105 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2106 		dev_lock();
2107 		vp->v_rdev->si_usecount++;
2108 		dev_unlock();
2109 	}
2110 }
2111 
2112 /*
2113  * Decrement the vnode use and hold count along with the driver's usecount
2114  * if this is a chardev.  The vdropl() below releases the vnode interlock
2115  * as it may free the vnode.
2116  */
2117 static void
2118 v_decr_usecount(struct vnode *vp)
2119 {
2120 
2121 	ASSERT_VI_LOCKED(vp, __FUNCTION__);
2122 	VNASSERT(vp->v_usecount > 0, vp,
2123 	    ("v_decr_usecount: negative usecount"));
2124 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2125 	vp->v_usecount--;
2126 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2127 		dev_lock();
2128 		vp->v_rdev->si_usecount--;
2129 		dev_unlock();
2130 	}
2131 	vdropl(vp);
2132 }
2133 
2134 /*
2135  * Decrement only the use count and driver use count.  This is intended to
2136  * be paired with a follow on vdropl() to release the remaining hold count.
2137  * In this way we may vgone() a vnode with a 0 usecount without risk of
2138  * having it end up on a free list because the hold count is kept above 0.
2139  */
2140 static void
2141 v_decr_useonly(struct vnode *vp)
2142 {
2143 
2144 	ASSERT_VI_LOCKED(vp, __FUNCTION__);
2145 	VNASSERT(vp->v_usecount > 0, vp,
2146 	    ("v_decr_useonly: negative usecount"));
2147 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2148 	vp->v_usecount--;
2149 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2150 		dev_lock();
2151 		vp->v_rdev->si_usecount--;
2152 		dev_unlock();
2153 	}
2154 }
2155 
2156 /*
2157  * Grab a particular vnode from the free list, increment its
2158  * reference count and lock it.  VI_DOOMED is set if the vnode
2159  * is being destroyed.  Only callers who specify LK_RETRY will
2160  * see doomed vnodes.  If inactive processing was delayed in
2161  * vput try to do it here.
2162  */
2163 int
2164 vget(struct vnode *vp, int flags, struct thread *td)
2165 {
2166 	int error;
2167 
2168 	error = 0;
2169 	VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
2170 	    ("vget: invalid lock operation"));
2171 	CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
2172 
2173 	if ((flags & LK_INTERLOCK) == 0)
2174 		VI_LOCK(vp);
2175 	vholdl(vp);
2176 	if ((error = vn_lock(vp, flags | LK_INTERLOCK)) != 0) {
2177 		vdrop(vp);
2178 		CTR2(KTR_VFS, "%s: impossible to lock vnode %p", __func__,
2179 		    vp);
2180 		return (error);
2181 	}
2182 	if (vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0)
2183 		panic("vget: vn_lock failed to return ENOENT\n");
2184 	VI_LOCK(vp);
2185 	/* Upgrade our holdcnt to a usecount. */
2186 	v_upgrade_usecount(vp);
2187 	/*
2188 	 * We don't guarantee that any particular close will
2189 	 * trigger inactive processing so just make a best effort
2190 	 * here at preventing a reference to a removed file.  If
2191 	 * we don't succeed no harm is done.
2192 	 */
2193 	if (vp->v_iflag & VI_OWEINACT) {
2194 		if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE &&
2195 		    (flags & LK_NOWAIT) == 0)
2196 			vinactive(vp, td);
2197 		vp->v_iflag &= ~VI_OWEINACT;
2198 	}
2199 	VI_UNLOCK(vp);
2200 	return (0);
2201 }
2202 
2203 /*
2204  * Increase the reference count of a vnode.
2205  */
2206 void
2207 vref(struct vnode *vp)
2208 {
2209 
2210 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2211 	VI_LOCK(vp);
2212 	v_incr_usecount(vp);
2213 	VI_UNLOCK(vp);
2214 }
2215 
2216 /*
2217  * Return reference count of a vnode.
2218  *
2219  * The results of this call are only guaranteed when some mechanism other
2220  * than the VI lock is used to stop other processes from gaining references
2221  * to the vnode.  This may be the case if the caller holds the only reference.
2222  * This is also useful when stale data is acceptable as race conditions may
2223  * be accounted for by some other means.
2224  */
2225 int
2226 vrefcnt(struct vnode *vp)
2227 {
2228 	int usecnt;
2229 
2230 	VI_LOCK(vp);
2231 	usecnt = vp->v_usecount;
2232 	VI_UNLOCK(vp);
2233 
2234 	return (usecnt);
2235 }
2236 
2237 #define	VPUTX_VRELE	1
2238 #define	VPUTX_VPUT	2
2239 #define	VPUTX_VUNREF	3
2240 
2241 static void
2242 vputx(struct vnode *vp, int func)
2243 {
2244 	int error;
2245 
2246 	KASSERT(vp != NULL, ("vputx: null vp"));
2247 	if (func == VPUTX_VUNREF)
2248 		ASSERT_VOP_LOCKED(vp, "vunref");
2249 	else if (func == VPUTX_VPUT)
2250 		ASSERT_VOP_LOCKED(vp, "vput");
2251 	else
2252 		KASSERT(func == VPUTX_VRELE, ("vputx: wrong func"));
2253 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2254 	VI_LOCK(vp);
2255 
2256 	/* Skip this v_writecount check if we're going to panic below. */
2257 	VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp,
2258 	    ("vputx: missed vn_close"));
2259 	error = 0;
2260 
2261 	if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2262 	    vp->v_usecount == 1)) {
2263 		if (func == VPUTX_VPUT)
2264 			VOP_UNLOCK(vp, 0);
2265 		v_decr_usecount(vp);
2266 		return;
2267 	}
2268 
2269 	if (vp->v_usecount != 1) {
2270 		vprint("vputx: negative ref count", vp);
2271 		panic("vputx: negative ref cnt");
2272 	}
2273 	CTR2(KTR_VFS, "%s: return vnode %p to the freelist", __func__, vp);
2274 	/*
2275 	 * We want to hold the vnode until the inactive finishes to
2276 	 * prevent vgone() races.  We drop the use count here and the
2277 	 * hold count below when we're done.
2278 	 */
2279 	v_decr_useonly(vp);
2280 	/*
2281 	 * We must call VOP_INACTIVE with the node locked. Mark
2282 	 * as VI_DOINGINACT to avoid recursion.
2283 	 */
2284 	vp->v_iflag |= VI_OWEINACT;
2285 	switch (func) {
2286 	case VPUTX_VRELE:
2287 		error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK);
2288 		VI_LOCK(vp);
2289 		break;
2290 	case VPUTX_VPUT:
2291 		if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
2292 			error = VOP_LOCK(vp, LK_UPGRADE | LK_INTERLOCK |
2293 			    LK_NOWAIT);
2294 			VI_LOCK(vp);
2295 		}
2296 		break;
2297 	case VPUTX_VUNREF:
2298 		if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
2299 			error = VOP_LOCK(vp, LK_TRYUPGRADE | LK_INTERLOCK);
2300 			VI_LOCK(vp);
2301 		}
2302 		break;
2303 	}
2304 	if (vp->v_usecount > 0)
2305 		vp->v_iflag &= ~VI_OWEINACT;
2306 	if (error == 0) {
2307 		if (vp->v_iflag & VI_OWEINACT)
2308 			vinactive(vp, curthread);
2309 		if (func != VPUTX_VUNREF)
2310 			VOP_UNLOCK(vp, 0);
2311 	}
2312 	vdropl(vp);
2313 }
2314 
2315 /*
2316  * Vnode put/release.
2317  * If count drops to zero, call inactive routine and return to freelist.
2318  */
2319 void
2320 vrele(struct vnode *vp)
2321 {
2322 
2323 	vputx(vp, VPUTX_VRELE);
2324 }
2325 
2326 /*
2327  * Release an already locked vnode.  This give the same effects as
2328  * unlock+vrele(), but takes less time and avoids releasing and
2329  * re-aquiring the lock (as vrele() acquires the lock internally.)
2330  */
2331 void
2332 vput(struct vnode *vp)
2333 {
2334 
2335 	vputx(vp, VPUTX_VPUT);
2336 }
2337 
2338 /*
2339  * Release an exclusively locked vnode. Do not unlock the vnode lock.
2340  */
2341 void
2342 vunref(struct vnode *vp)
2343 {
2344 
2345 	vputx(vp, VPUTX_VUNREF);
2346 }
2347 
2348 /*
2349  * Somebody doesn't want the vnode recycled.
2350  */
2351 void
2352 vhold(struct vnode *vp)
2353 {
2354 
2355 	VI_LOCK(vp);
2356 	vholdl(vp);
2357 	VI_UNLOCK(vp);
2358 }
2359 
2360 /*
2361  * Increase the hold count and activate if this is the first reference.
2362  */
2363 void
2364 vholdl(struct vnode *vp)
2365 {
2366 	struct mount *mp;
2367 
2368 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2369 #ifdef INVARIANTS
2370 	/* getnewvnode() calls v_incr_usecount() without holding interlock. */
2371 	if (vp->v_type != VNON || vp->v_data != NULL) {
2372 		ASSERT_VI_LOCKED(vp, "vholdl");
2373 		VNASSERT(vp->v_holdcnt > 0 || (vp->v_iflag & VI_FREE) != 0,
2374 		    vp, ("vholdl: free vnode is held"));
2375 	}
2376 #endif
2377 	vp->v_holdcnt++;
2378 	if ((vp->v_iflag & VI_FREE) == 0)
2379 		return;
2380 	VNASSERT(vp->v_holdcnt == 1, vp, ("vholdl: wrong hold count"));
2381 	VNASSERT(vp->v_op != NULL, vp, ("vholdl: vnode already reclaimed."));
2382 	/*
2383 	 * Remove a vnode from the free list, mark it as in use,
2384 	 * and put it on the active list.
2385 	 */
2386 	mtx_lock(&vnode_free_list_mtx);
2387 	TAILQ_REMOVE(&vnode_free_list, vp, v_actfreelist);
2388 	freevnodes--;
2389 	vp->v_iflag &= ~(VI_FREE|VI_AGE);
2390 	KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
2391 	    ("Activating already active vnode"));
2392 	vp->v_iflag |= VI_ACTIVE;
2393 	mp = vp->v_mount;
2394 	TAILQ_INSERT_HEAD(&mp->mnt_activevnodelist, vp, v_actfreelist);
2395 	mp->mnt_activevnodelistsize++;
2396 	mtx_unlock(&vnode_free_list_mtx);
2397 }
2398 
2399 /*
2400  * Note that there is one less who cares about this vnode.
2401  * vdrop() is the opposite of vhold().
2402  */
2403 void
2404 vdrop(struct vnode *vp)
2405 {
2406 
2407 	VI_LOCK(vp);
2408 	vdropl(vp);
2409 }
2410 
2411 /*
2412  * Drop the hold count of the vnode.  If this is the last reference to
2413  * the vnode we place it on the free list unless it has been vgone'd
2414  * (marked VI_DOOMED) in which case we will free it.
2415  */
2416 void
2417 vdropl(struct vnode *vp)
2418 {
2419 	struct bufobj *bo;
2420 	struct mount *mp;
2421 	int active;
2422 
2423 	ASSERT_VI_LOCKED(vp, "vdropl");
2424 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2425 	if (vp->v_holdcnt <= 0)
2426 		panic("vdrop: holdcnt %d", vp->v_holdcnt);
2427 	vp->v_holdcnt--;
2428 	VNASSERT(vp->v_holdcnt >= vp->v_usecount, vp,
2429 	    ("hold count less than use count"));
2430 	if (vp->v_holdcnt > 0) {
2431 		VI_UNLOCK(vp);
2432 		return;
2433 	}
2434 	if ((vp->v_iflag & VI_DOOMED) == 0) {
2435 		/*
2436 		 * Mark a vnode as free: remove it from its active list
2437 		 * and put it up for recycling on the freelist.
2438 		 */
2439 		VNASSERT(vp->v_op != NULL, vp,
2440 		    ("vdropl: vnode already reclaimed."));
2441 		VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
2442 		    ("vnode already free"));
2443 		VNASSERT(vp->v_holdcnt == 0, vp,
2444 		    ("vdropl: freeing when we shouldn't"));
2445 		active = vp->v_iflag & VI_ACTIVE;
2446 		vp->v_iflag &= ~VI_ACTIVE;
2447 		mp = vp->v_mount;
2448 		mtx_lock(&vnode_free_list_mtx);
2449 		if (active) {
2450 			TAILQ_REMOVE(&mp->mnt_activevnodelist, vp,
2451 			    v_actfreelist);
2452 			mp->mnt_activevnodelistsize--;
2453 		}
2454 		if (vp->v_iflag & VI_AGE) {
2455 			TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_actfreelist);
2456 		} else {
2457 			TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_actfreelist);
2458 		}
2459 		freevnodes++;
2460 		vp->v_iflag &= ~VI_AGE;
2461 		vp->v_iflag |= VI_FREE;
2462 		mtx_unlock(&vnode_free_list_mtx);
2463 		VI_UNLOCK(vp);
2464 		return;
2465 	}
2466 	/*
2467 	 * The vnode has been marked for destruction, so free it.
2468 	 */
2469 	CTR2(KTR_VFS, "%s: destroying the vnode %p", __func__, vp);
2470 	atomic_subtract_long(&numvnodes, 1);
2471 	bo = &vp->v_bufobj;
2472 	VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
2473 	    ("cleaned vnode still on the free list."));
2474 	VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't"));
2475 	VNASSERT(vp->v_holdcnt == 0, vp, ("Non-zero hold count"));
2476 	VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count"));
2477 	VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count"));
2478 	VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's"));
2479 	VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0"));
2480 	VNASSERT(pctrie_is_empty(&bo->bo_clean.bv_root), vp,
2481 	    ("clean blk trie not empty"));
2482 	VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0"));
2483 	VNASSERT(pctrie_is_empty(&bo->bo_dirty.bv_root), vp,
2484 	    ("dirty blk trie not empty"));
2485 	VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst"));
2486 	VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src"));
2487 	VNASSERT(vp->v_cache_dd == NULL, vp, ("vp has namecache for .."));
2488 	VI_UNLOCK(vp);
2489 #ifdef MAC
2490 	mac_vnode_destroy(vp);
2491 #endif
2492 	if (vp->v_pollinfo != NULL)
2493 		destroy_vpollinfo(vp->v_pollinfo);
2494 #ifdef INVARIANTS
2495 	/* XXX Elsewhere we detect an already freed vnode via NULL v_op. */
2496 	vp->v_op = NULL;
2497 #endif
2498 	rangelock_destroy(&vp->v_rl);
2499 	lockdestroy(vp->v_vnlock);
2500 	mtx_destroy(&vp->v_interlock);
2501 	rw_destroy(BO_LOCKPTR(bo));
2502 	uma_zfree(vnode_zone, vp);
2503 }
2504 
2505 /*
2506  * Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT
2507  * flags.  DOINGINACT prevents us from recursing in calls to vinactive.
2508  * OWEINACT tracks whether a vnode missed a call to inactive due to a
2509  * failed lock upgrade.
2510  */
2511 void
2512 vinactive(struct vnode *vp, struct thread *td)
2513 {
2514 	struct vm_object *obj;
2515 
2516 	ASSERT_VOP_ELOCKED(vp, "vinactive");
2517 	ASSERT_VI_LOCKED(vp, "vinactive");
2518 	VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp,
2519 	    ("vinactive: recursed on VI_DOINGINACT"));
2520 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2521 	vp->v_iflag |= VI_DOINGINACT;
2522 	vp->v_iflag &= ~VI_OWEINACT;
2523 	VI_UNLOCK(vp);
2524 	/*
2525 	 * Before moving off the active list, we must be sure that any
2526 	 * modified pages are on the vnode's dirty list since these will
2527 	 * no longer be checked once the vnode is on the inactive list.
2528 	 * Because the vnode vm object keeps a hold reference on the vnode
2529 	 * if there is at least one resident non-cached page, the vnode
2530 	 * cannot leave the active list without the page cleanup done.
2531 	 */
2532 	obj = vp->v_object;
2533 	if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0) {
2534 		VM_OBJECT_WLOCK(obj);
2535 		vm_object_page_clean(obj, 0, 0, OBJPC_NOSYNC);
2536 		VM_OBJECT_WUNLOCK(obj);
2537 	}
2538 	VOP_INACTIVE(vp, td);
2539 	VI_LOCK(vp);
2540 	VNASSERT(vp->v_iflag & VI_DOINGINACT, vp,
2541 	    ("vinactive: lost VI_DOINGINACT"));
2542 	vp->v_iflag &= ~VI_DOINGINACT;
2543 }
2544 
2545 /*
2546  * Remove any vnodes in the vnode table belonging to mount point mp.
2547  *
2548  * If FORCECLOSE is not specified, there should not be any active ones,
2549  * return error if any are found (nb: this is a user error, not a
2550  * system error). If FORCECLOSE is specified, detach any active vnodes
2551  * that are found.
2552  *
2553  * If WRITECLOSE is set, only flush out regular file vnodes open for
2554  * writing.
2555  *
2556  * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
2557  *
2558  * `rootrefs' specifies the base reference count for the root vnode
2559  * of this filesystem. The root vnode is considered busy if its
2560  * v_usecount exceeds this value. On a successful return, vflush(, td)
2561  * will call vrele() on the root vnode exactly rootrefs times.
2562  * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
2563  * be zero.
2564  */
2565 #ifdef DIAGNOSTIC
2566 static int busyprt = 0;		/* print out busy vnodes */
2567 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "Print out busy vnodes");
2568 #endif
2569 
2570 int
2571 vflush(struct mount *mp, int rootrefs, int flags, struct thread *td)
2572 {
2573 	struct vnode *vp, *mvp, *rootvp = NULL;
2574 	struct vattr vattr;
2575 	int busy = 0, error;
2576 
2577 	CTR4(KTR_VFS, "%s: mp %p with rootrefs %d and flags %d", __func__, mp,
2578 	    rootrefs, flags);
2579 	if (rootrefs > 0) {
2580 		KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
2581 		    ("vflush: bad args"));
2582 		/*
2583 		 * Get the filesystem root vnode. We can vput() it
2584 		 * immediately, since with rootrefs > 0, it won't go away.
2585 		 */
2586 		if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp)) != 0) {
2587 			CTR2(KTR_VFS, "%s: vfs_root lookup failed with %d",
2588 			    __func__, error);
2589 			return (error);
2590 		}
2591 		vput(rootvp);
2592 	}
2593 loop:
2594 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
2595 		vholdl(vp);
2596 		error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE);
2597 		if (error) {
2598 			vdrop(vp);
2599 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
2600 			goto loop;
2601 		}
2602 		/*
2603 		 * Skip over a vnodes marked VV_SYSTEM.
2604 		 */
2605 		if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
2606 			VOP_UNLOCK(vp, 0);
2607 			vdrop(vp);
2608 			continue;
2609 		}
2610 		/*
2611 		 * If WRITECLOSE is set, flush out unlinked but still open
2612 		 * files (even if open only for reading) and regular file
2613 		 * vnodes open for writing.
2614 		 */
2615 		if (flags & WRITECLOSE) {
2616 			if (vp->v_object != NULL) {
2617 				VM_OBJECT_WLOCK(vp->v_object);
2618 				vm_object_page_clean(vp->v_object, 0, 0, 0);
2619 				VM_OBJECT_WUNLOCK(vp->v_object);
2620 			}
2621 			error = VOP_FSYNC(vp, MNT_WAIT, td);
2622 			if (error != 0) {
2623 				VOP_UNLOCK(vp, 0);
2624 				vdrop(vp);
2625 				MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
2626 				return (error);
2627 			}
2628 			error = VOP_GETATTR(vp, &vattr, td->td_ucred);
2629 			VI_LOCK(vp);
2630 
2631 			if ((vp->v_type == VNON ||
2632 			    (error == 0 && vattr.va_nlink > 0)) &&
2633 			    (vp->v_writecount == 0 || vp->v_type != VREG)) {
2634 				VOP_UNLOCK(vp, 0);
2635 				vdropl(vp);
2636 				continue;
2637 			}
2638 		} else
2639 			VI_LOCK(vp);
2640 		/*
2641 		 * With v_usecount == 0, all we need to do is clear out the
2642 		 * vnode data structures and we are done.
2643 		 *
2644 		 * If FORCECLOSE is set, forcibly close the vnode.
2645 		 */
2646 		if (vp->v_usecount == 0 || (flags & FORCECLOSE)) {
2647 			VNASSERT(vp->v_usecount == 0 ||
2648 			    vp->v_op != &devfs_specops ||
2649 			    (vp->v_type != VCHR && vp->v_type != VBLK), vp,
2650 			    ("device VNODE %p is FORCECLOSED", vp));
2651 			vgonel(vp);
2652 		} else {
2653 			busy++;
2654 #ifdef DIAGNOSTIC
2655 			if (busyprt)
2656 				vprint("vflush: busy vnode", vp);
2657 #endif
2658 		}
2659 		VOP_UNLOCK(vp, 0);
2660 		vdropl(vp);
2661 	}
2662 	if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
2663 		/*
2664 		 * If just the root vnode is busy, and if its refcount
2665 		 * is equal to `rootrefs', then go ahead and kill it.
2666 		 */
2667 		VI_LOCK(rootvp);
2668 		KASSERT(busy > 0, ("vflush: not busy"));
2669 		VNASSERT(rootvp->v_usecount >= rootrefs, rootvp,
2670 		    ("vflush: usecount %d < rootrefs %d",
2671 		     rootvp->v_usecount, rootrefs));
2672 		if (busy == 1 && rootvp->v_usecount == rootrefs) {
2673 			VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK);
2674 			vgone(rootvp);
2675 			VOP_UNLOCK(rootvp, 0);
2676 			busy = 0;
2677 		} else
2678 			VI_UNLOCK(rootvp);
2679 	}
2680 	if (busy) {
2681 		CTR2(KTR_VFS, "%s: failing as %d vnodes are busy", __func__,
2682 		    busy);
2683 		return (EBUSY);
2684 	}
2685 	for (; rootrefs > 0; rootrefs--)
2686 		vrele(rootvp);
2687 	return (0);
2688 }
2689 
2690 /*
2691  * Recycle an unused vnode to the front of the free list.
2692  */
2693 int
2694 vrecycle(struct vnode *vp)
2695 {
2696 	int recycled;
2697 
2698 	ASSERT_VOP_ELOCKED(vp, "vrecycle");
2699 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2700 	recycled = 0;
2701 	VI_LOCK(vp);
2702 	if (vp->v_usecount == 0) {
2703 		recycled = 1;
2704 		vgonel(vp);
2705 	}
2706 	VI_UNLOCK(vp);
2707 	return (recycled);
2708 }
2709 
2710 /*
2711  * Eliminate all activity associated with a vnode
2712  * in preparation for reuse.
2713  */
2714 void
2715 vgone(struct vnode *vp)
2716 {
2717 	VI_LOCK(vp);
2718 	vgonel(vp);
2719 	VI_UNLOCK(vp);
2720 }
2721 
2722 static void
2723 notify_lowervp_vfs_dummy(struct mount *mp __unused,
2724     struct vnode *lowervp __unused)
2725 {
2726 }
2727 
2728 /*
2729  * Notify upper mounts about reclaimed or unlinked vnode.
2730  */
2731 void
2732 vfs_notify_upper(struct vnode *vp, int event)
2733 {
2734 	static struct vfsops vgonel_vfsops = {
2735 		.vfs_reclaim_lowervp = notify_lowervp_vfs_dummy,
2736 		.vfs_unlink_lowervp = notify_lowervp_vfs_dummy,
2737 	};
2738 	struct mount *mp, *ump, *mmp;
2739 
2740 	mp = vp->v_mount;
2741 	if (mp == NULL)
2742 		return;
2743 
2744 	MNT_ILOCK(mp);
2745 	if (TAILQ_EMPTY(&mp->mnt_uppers))
2746 		goto unlock;
2747 	MNT_IUNLOCK(mp);
2748 	mmp = malloc(sizeof(struct mount), M_TEMP, M_WAITOK | M_ZERO);
2749 	mmp->mnt_op = &vgonel_vfsops;
2750 	mmp->mnt_kern_flag |= MNTK_MARKER;
2751 	MNT_ILOCK(mp);
2752 	mp->mnt_kern_flag |= MNTK_VGONE_UPPER;
2753 	for (ump = TAILQ_FIRST(&mp->mnt_uppers); ump != NULL;) {
2754 		if ((ump->mnt_kern_flag & MNTK_MARKER) != 0) {
2755 			ump = TAILQ_NEXT(ump, mnt_upper_link);
2756 			continue;
2757 		}
2758 		TAILQ_INSERT_AFTER(&mp->mnt_uppers, ump, mmp, mnt_upper_link);
2759 		MNT_IUNLOCK(mp);
2760 		switch (event) {
2761 		case VFS_NOTIFY_UPPER_RECLAIM:
2762 			VFS_RECLAIM_LOWERVP(ump, vp);
2763 			break;
2764 		case VFS_NOTIFY_UPPER_UNLINK:
2765 			VFS_UNLINK_LOWERVP(ump, vp);
2766 			break;
2767 		default:
2768 			KASSERT(0, ("invalid event %d", event));
2769 			break;
2770 		}
2771 		MNT_ILOCK(mp);
2772 		ump = TAILQ_NEXT(mmp, mnt_upper_link);
2773 		TAILQ_REMOVE(&mp->mnt_uppers, mmp, mnt_upper_link);
2774 	}
2775 	free(mmp, M_TEMP);
2776 	mp->mnt_kern_flag &= ~MNTK_VGONE_UPPER;
2777 	if ((mp->mnt_kern_flag & MNTK_VGONE_WAITER) != 0) {
2778 		mp->mnt_kern_flag &= ~MNTK_VGONE_WAITER;
2779 		wakeup(&mp->mnt_uppers);
2780 	}
2781 unlock:
2782 	MNT_IUNLOCK(mp);
2783 }
2784 
2785 /*
2786  * vgone, with the vp interlock held.
2787  */
2788 void
2789 vgonel(struct vnode *vp)
2790 {
2791 	struct thread *td;
2792 	int oweinact;
2793 	int active;
2794 	struct mount *mp;
2795 
2796 	ASSERT_VOP_ELOCKED(vp, "vgonel");
2797 	ASSERT_VI_LOCKED(vp, "vgonel");
2798 	VNASSERT(vp->v_holdcnt, vp,
2799 	    ("vgonel: vp %p has no reference.", vp));
2800 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2801 	td = curthread;
2802 
2803 	/*
2804 	 * Don't vgonel if we're already doomed.
2805 	 */
2806 	if (vp->v_iflag & VI_DOOMED)
2807 		return;
2808 	vp->v_iflag |= VI_DOOMED;
2809 
2810 	/*
2811 	 * Check to see if the vnode is in use.  If so, we have to call
2812 	 * VOP_CLOSE() and VOP_INACTIVE().
2813 	 */
2814 	active = vp->v_usecount;
2815 	oweinact = (vp->v_iflag & VI_OWEINACT);
2816 	VI_UNLOCK(vp);
2817 	vfs_notify_upper(vp, VFS_NOTIFY_UPPER_RECLAIM);
2818 
2819 	/*
2820 	 * If purging an active vnode, it must be closed and
2821 	 * deactivated before being reclaimed.
2822 	 */
2823 	if (active)
2824 		VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
2825 	if (oweinact || active) {
2826 		VI_LOCK(vp);
2827 		if ((vp->v_iflag & VI_DOINGINACT) == 0)
2828 			vinactive(vp, td);
2829 		VI_UNLOCK(vp);
2830 	}
2831 	if (vp->v_type == VSOCK)
2832 		vfs_unp_reclaim(vp);
2833 
2834 	/*
2835 	 * Clean out any buffers associated with the vnode.
2836 	 * If the flush fails, just toss the buffers.
2837 	 */
2838 	mp = NULL;
2839 	if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd))
2840 		(void) vn_start_secondary_write(vp, &mp, V_WAIT);
2841 	if (vinvalbuf(vp, V_SAVE, 0, 0) != 0) {
2842 		while (vinvalbuf(vp, 0, 0, 0) != 0)
2843 			;
2844 	}
2845 #ifdef INVARIANTS
2846 	BO_LOCK(&vp->v_bufobj);
2847 	KASSERT(TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd) &&
2848 	    vp->v_bufobj.bo_dirty.bv_cnt == 0 &&
2849 	    TAILQ_EMPTY(&vp->v_bufobj.bo_clean.bv_hd) &&
2850 	    vp->v_bufobj.bo_clean.bv_cnt == 0,
2851 	    ("vp %p bufobj not invalidated", vp));
2852 	vp->v_bufobj.bo_flag |= BO_DEAD;
2853 	BO_UNLOCK(&vp->v_bufobj);
2854 #endif
2855 
2856 	/*
2857 	 * Reclaim the vnode.
2858 	 */
2859 	if (VOP_RECLAIM(vp, td))
2860 		panic("vgone: cannot reclaim");
2861 	if (mp != NULL)
2862 		vn_finished_secondary_write(mp);
2863 	VNASSERT(vp->v_object == NULL, vp,
2864 	    ("vop_reclaim left v_object vp=%p, tag=%s", vp, vp->v_tag));
2865 	/*
2866 	 * Clear the advisory locks and wake up waiting threads.
2867 	 */
2868 	(void)VOP_ADVLOCKPURGE(vp);
2869 	/*
2870 	 * Delete from old mount point vnode list.
2871 	 */
2872 	delmntque(vp);
2873 	cache_purge(vp);
2874 	/*
2875 	 * Done with purge, reset to the standard lock and invalidate
2876 	 * the vnode.
2877 	 */
2878 	VI_LOCK(vp);
2879 	vp->v_vnlock = &vp->v_lock;
2880 	vp->v_op = &dead_vnodeops;
2881 	vp->v_tag = "none";
2882 	vp->v_type = VBAD;
2883 }
2884 
2885 /*
2886  * Calculate the total number of references to a special device.
2887  */
2888 int
2889 vcount(struct vnode *vp)
2890 {
2891 	int count;
2892 
2893 	dev_lock();
2894 	count = vp->v_rdev->si_usecount;
2895 	dev_unlock();
2896 	return (count);
2897 }
2898 
2899 /*
2900  * Same as above, but using the struct cdev *as argument
2901  */
2902 int
2903 count_dev(struct cdev *dev)
2904 {
2905 	int count;
2906 
2907 	dev_lock();
2908 	count = dev->si_usecount;
2909 	dev_unlock();
2910 	return(count);
2911 }
2912 
2913 /*
2914  * Print out a description of a vnode.
2915  */
2916 static char *typename[] =
2917 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD",
2918  "VMARKER"};
2919 
2920 void
2921 vn_printf(struct vnode *vp, const char *fmt, ...)
2922 {
2923 	va_list ap;
2924 	char buf[256], buf2[16];
2925 	u_long flags;
2926 
2927 	va_start(ap, fmt);
2928 	vprintf(fmt, ap);
2929 	va_end(ap);
2930 	printf("%p: ", (void *)vp);
2931 	printf("tag %s, type %s\n", vp->v_tag, typename[vp->v_type]);
2932 	printf("    usecount %d, writecount %d, refcount %d mountedhere %p\n",
2933 	    vp->v_usecount, vp->v_writecount, vp->v_holdcnt, vp->v_mountedhere);
2934 	buf[0] = '\0';
2935 	buf[1] = '\0';
2936 	if (vp->v_vflag & VV_ROOT)
2937 		strlcat(buf, "|VV_ROOT", sizeof(buf));
2938 	if (vp->v_vflag & VV_ISTTY)
2939 		strlcat(buf, "|VV_ISTTY", sizeof(buf));
2940 	if (vp->v_vflag & VV_NOSYNC)
2941 		strlcat(buf, "|VV_NOSYNC", sizeof(buf));
2942 	if (vp->v_vflag & VV_ETERNALDEV)
2943 		strlcat(buf, "|VV_ETERNALDEV", sizeof(buf));
2944 	if (vp->v_vflag & VV_CACHEDLABEL)
2945 		strlcat(buf, "|VV_CACHEDLABEL", sizeof(buf));
2946 	if (vp->v_vflag & VV_TEXT)
2947 		strlcat(buf, "|VV_TEXT", sizeof(buf));
2948 	if (vp->v_vflag & VV_COPYONWRITE)
2949 		strlcat(buf, "|VV_COPYONWRITE", sizeof(buf));
2950 	if (vp->v_vflag & VV_SYSTEM)
2951 		strlcat(buf, "|VV_SYSTEM", sizeof(buf));
2952 	if (vp->v_vflag & VV_PROCDEP)
2953 		strlcat(buf, "|VV_PROCDEP", sizeof(buf));
2954 	if (vp->v_vflag & VV_NOKNOTE)
2955 		strlcat(buf, "|VV_NOKNOTE", sizeof(buf));
2956 	if (vp->v_vflag & VV_DELETED)
2957 		strlcat(buf, "|VV_DELETED", sizeof(buf));
2958 	if (vp->v_vflag & VV_MD)
2959 		strlcat(buf, "|VV_MD", sizeof(buf));
2960 	if (vp->v_vflag & VV_FORCEINSMQ)
2961 		strlcat(buf, "|VV_FORCEINSMQ", sizeof(buf));
2962 	flags = vp->v_vflag & ~(VV_ROOT | VV_ISTTY | VV_NOSYNC | VV_ETERNALDEV |
2963 	    VV_CACHEDLABEL | VV_TEXT | VV_COPYONWRITE | VV_SYSTEM | VV_PROCDEP |
2964 	    VV_NOKNOTE | VV_DELETED | VV_MD | VV_FORCEINSMQ);
2965 	if (flags != 0) {
2966 		snprintf(buf2, sizeof(buf2), "|VV(0x%lx)", flags);
2967 		strlcat(buf, buf2, sizeof(buf));
2968 	}
2969 	if (vp->v_iflag & VI_MOUNT)
2970 		strlcat(buf, "|VI_MOUNT", sizeof(buf));
2971 	if (vp->v_iflag & VI_AGE)
2972 		strlcat(buf, "|VI_AGE", sizeof(buf));
2973 	if (vp->v_iflag & VI_DOOMED)
2974 		strlcat(buf, "|VI_DOOMED", sizeof(buf));
2975 	if (vp->v_iflag & VI_FREE)
2976 		strlcat(buf, "|VI_FREE", sizeof(buf));
2977 	if (vp->v_iflag & VI_ACTIVE)
2978 		strlcat(buf, "|VI_ACTIVE", sizeof(buf));
2979 	if (vp->v_iflag & VI_DOINGINACT)
2980 		strlcat(buf, "|VI_DOINGINACT", sizeof(buf));
2981 	if (vp->v_iflag & VI_OWEINACT)
2982 		strlcat(buf, "|VI_OWEINACT", sizeof(buf));
2983 	flags = vp->v_iflag & ~(VI_MOUNT | VI_AGE | VI_DOOMED | VI_FREE |
2984 	    VI_ACTIVE | VI_DOINGINACT | VI_OWEINACT);
2985 	if (flags != 0) {
2986 		snprintf(buf2, sizeof(buf2), "|VI(0x%lx)", flags);
2987 		strlcat(buf, buf2, sizeof(buf));
2988 	}
2989 	printf("    flags (%s)\n", buf + 1);
2990 	if (mtx_owned(VI_MTX(vp)))
2991 		printf(" VI_LOCKed");
2992 	if (vp->v_object != NULL)
2993 		printf("    v_object %p ref %d pages %d "
2994 		    "cleanbuf %d dirtybuf %d\n",
2995 		    vp->v_object, vp->v_object->ref_count,
2996 		    vp->v_object->resident_page_count,
2997 		    vp->v_bufobj.bo_dirty.bv_cnt,
2998 		    vp->v_bufobj.bo_clean.bv_cnt);
2999 	printf("    ");
3000 	lockmgr_printinfo(vp->v_vnlock);
3001 	if (vp->v_data != NULL)
3002 		VOP_PRINT(vp);
3003 }
3004 
3005 #ifdef DDB
3006 /*
3007  * List all of the locked vnodes in the system.
3008  * Called when debugging the kernel.
3009  */
3010 DB_SHOW_COMMAND(lockedvnods, lockedvnodes)
3011 {
3012 	struct mount *mp;
3013 	struct vnode *vp;
3014 
3015 	/*
3016 	 * Note: because this is DDB, we can't obey the locking semantics
3017 	 * for these structures, which means we could catch an inconsistent
3018 	 * state and dereference a nasty pointer.  Not much to be done
3019 	 * about that.
3020 	 */
3021 	db_printf("Locked vnodes\n");
3022 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3023 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3024 			if (vp->v_type != VMARKER && VOP_ISLOCKED(vp))
3025 				vprint("", vp);
3026 		}
3027 	}
3028 }
3029 
3030 /*
3031  * Show details about the given vnode.
3032  */
3033 DB_SHOW_COMMAND(vnode, db_show_vnode)
3034 {
3035 	struct vnode *vp;
3036 
3037 	if (!have_addr)
3038 		return;
3039 	vp = (struct vnode *)addr;
3040 	vn_printf(vp, "vnode ");
3041 }
3042 
3043 /*
3044  * Show details about the given mount point.
3045  */
3046 DB_SHOW_COMMAND(mount, db_show_mount)
3047 {
3048 	struct mount *mp;
3049 	struct vfsopt *opt;
3050 	struct statfs *sp;
3051 	struct vnode *vp;
3052 	char buf[512];
3053 	uint64_t mflags;
3054 	u_int flags;
3055 
3056 	if (!have_addr) {
3057 		/* No address given, print short info about all mount points. */
3058 		TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3059 			db_printf("%p %s on %s (%s)\n", mp,
3060 			    mp->mnt_stat.f_mntfromname,
3061 			    mp->mnt_stat.f_mntonname,
3062 			    mp->mnt_stat.f_fstypename);
3063 			if (db_pager_quit)
3064 				break;
3065 		}
3066 		db_printf("\nMore info: show mount <addr>\n");
3067 		return;
3068 	}
3069 
3070 	mp = (struct mount *)addr;
3071 	db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname,
3072 	    mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename);
3073 
3074 	buf[0] = '\0';
3075 	mflags = mp->mnt_flag;
3076 #define	MNT_FLAG(flag)	do {						\
3077 	if (mflags & (flag)) {						\
3078 		if (buf[0] != '\0')					\
3079 			strlcat(buf, ", ", sizeof(buf));		\
3080 		strlcat(buf, (#flag) + 4, sizeof(buf));			\
3081 		mflags &= ~(flag);					\
3082 	}								\
3083 } while (0)
3084 	MNT_FLAG(MNT_RDONLY);
3085 	MNT_FLAG(MNT_SYNCHRONOUS);
3086 	MNT_FLAG(MNT_NOEXEC);
3087 	MNT_FLAG(MNT_NOSUID);
3088 	MNT_FLAG(MNT_NFS4ACLS);
3089 	MNT_FLAG(MNT_UNION);
3090 	MNT_FLAG(MNT_ASYNC);
3091 	MNT_FLAG(MNT_SUIDDIR);
3092 	MNT_FLAG(MNT_SOFTDEP);
3093 	MNT_FLAG(MNT_NOSYMFOLLOW);
3094 	MNT_FLAG(MNT_GJOURNAL);
3095 	MNT_FLAG(MNT_MULTILABEL);
3096 	MNT_FLAG(MNT_ACLS);
3097 	MNT_FLAG(MNT_NOATIME);
3098 	MNT_FLAG(MNT_NOCLUSTERR);
3099 	MNT_FLAG(MNT_NOCLUSTERW);
3100 	MNT_FLAG(MNT_SUJ);
3101 	MNT_FLAG(MNT_EXRDONLY);
3102 	MNT_FLAG(MNT_EXPORTED);
3103 	MNT_FLAG(MNT_DEFEXPORTED);
3104 	MNT_FLAG(MNT_EXPORTANON);
3105 	MNT_FLAG(MNT_EXKERB);
3106 	MNT_FLAG(MNT_EXPUBLIC);
3107 	MNT_FLAG(MNT_LOCAL);
3108 	MNT_FLAG(MNT_QUOTA);
3109 	MNT_FLAG(MNT_ROOTFS);
3110 	MNT_FLAG(MNT_USER);
3111 	MNT_FLAG(MNT_IGNORE);
3112 	MNT_FLAG(MNT_UPDATE);
3113 	MNT_FLAG(MNT_DELEXPORT);
3114 	MNT_FLAG(MNT_RELOAD);
3115 	MNT_FLAG(MNT_FORCE);
3116 	MNT_FLAG(MNT_SNAPSHOT);
3117 	MNT_FLAG(MNT_BYFSID);
3118 #undef MNT_FLAG
3119 	if (mflags != 0) {
3120 		if (buf[0] != '\0')
3121 			strlcat(buf, ", ", sizeof(buf));
3122 		snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
3123 		    "0x%016jx", mflags);
3124 	}
3125 	db_printf("    mnt_flag = %s\n", buf);
3126 
3127 	buf[0] = '\0';
3128 	flags = mp->mnt_kern_flag;
3129 #define	MNT_KERN_FLAG(flag)	do {					\
3130 	if (flags & (flag)) {						\
3131 		if (buf[0] != '\0')					\
3132 			strlcat(buf, ", ", sizeof(buf));		\
3133 		strlcat(buf, (#flag) + 5, sizeof(buf));			\
3134 		flags &= ~(flag);					\
3135 	}								\
3136 } while (0)
3137 	MNT_KERN_FLAG(MNTK_UNMOUNTF);
3138 	MNT_KERN_FLAG(MNTK_ASYNC);
3139 	MNT_KERN_FLAG(MNTK_SOFTDEP);
3140 	MNT_KERN_FLAG(MNTK_NOINSMNTQ);
3141 	MNT_KERN_FLAG(MNTK_DRAINING);
3142 	MNT_KERN_FLAG(MNTK_REFEXPIRE);
3143 	MNT_KERN_FLAG(MNTK_EXTENDED_SHARED);
3144 	MNT_KERN_FLAG(MNTK_SHARED_WRITES);
3145 	MNT_KERN_FLAG(MNTK_NO_IOPF);
3146 	MNT_KERN_FLAG(MNTK_VGONE_UPPER);
3147 	MNT_KERN_FLAG(MNTK_VGONE_WAITER);
3148 	MNT_KERN_FLAG(MNTK_LOOKUP_EXCL_DOTDOT);
3149 	MNT_KERN_FLAG(MNTK_MARKER);
3150 	MNT_KERN_FLAG(MNTK_USES_BCACHE);
3151 	MNT_KERN_FLAG(MNTK_NOASYNC);
3152 	MNT_KERN_FLAG(MNTK_UNMOUNT);
3153 	MNT_KERN_FLAG(MNTK_MWAIT);
3154 	MNT_KERN_FLAG(MNTK_SUSPEND);
3155 	MNT_KERN_FLAG(MNTK_SUSPEND2);
3156 	MNT_KERN_FLAG(MNTK_SUSPENDED);
3157 	MNT_KERN_FLAG(MNTK_LOOKUP_SHARED);
3158 	MNT_KERN_FLAG(MNTK_NOKNOTE);
3159 #undef MNT_KERN_FLAG
3160 	if (flags != 0) {
3161 		if (buf[0] != '\0')
3162 			strlcat(buf, ", ", sizeof(buf));
3163 		snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
3164 		    "0x%08x", flags);
3165 	}
3166 	db_printf("    mnt_kern_flag = %s\n", buf);
3167 
3168 	db_printf("    mnt_opt = ");
3169 	opt = TAILQ_FIRST(mp->mnt_opt);
3170 	if (opt != NULL) {
3171 		db_printf("%s", opt->name);
3172 		opt = TAILQ_NEXT(opt, link);
3173 		while (opt != NULL) {
3174 			db_printf(", %s", opt->name);
3175 			opt = TAILQ_NEXT(opt, link);
3176 		}
3177 	}
3178 	db_printf("\n");
3179 
3180 	sp = &mp->mnt_stat;
3181 	db_printf("    mnt_stat = { version=%u type=%u flags=0x%016jx "
3182 	    "bsize=%ju iosize=%ju blocks=%ju bfree=%ju bavail=%jd files=%ju "
3183 	    "ffree=%jd syncwrites=%ju asyncwrites=%ju syncreads=%ju "
3184 	    "asyncreads=%ju namemax=%u owner=%u fsid=[%d, %d] }\n",
3185 	    (u_int)sp->f_version, (u_int)sp->f_type, (uintmax_t)sp->f_flags,
3186 	    (uintmax_t)sp->f_bsize, (uintmax_t)sp->f_iosize,
3187 	    (uintmax_t)sp->f_blocks, (uintmax_t)sp->f_bfree,
3188 	    (intmax_t)sp->f_bavail, (uintmax_t)sp->f_files,
3189 	    (intmax_t)sp->f_ffree, (uintmax_t)sp->f_syncwrites,
3190 	    (uintmax_t)sp->f_asyncwrites, (uintmax_t)sp->f_syncreads,
3191 	    (uintmax_t)sp->f_asyncreads, (u_int)sp->f_namemax,
3192 	    (u_int)sp->f_owner, (int)sp->f_fsid.val[0], (int)sp->f_fsid.val[1]);
3193 
3194 	db_printf("    mnt_cred = { uid=%u ruid=%u",
3195 	    (u_int)mp->mnt_cred->cr_uid, (u_int)mp->mnt_cred->cr_ruid);
3196 	if (jailed(mp->mnt_cred))
3197 		db_printf(", jail=%d", mp->mnt_cred->cr_prison->pr_id);
3198 	db_printf(" }\n");
3199 	db_printf("    mnt_ref = %d\n", mp->mnt_ref);
3200 	db_printf("    mnt_gen = %d\n", mp->mnt_gen);
3201 	db_printf("    mnt_nvnodelistsize = %d\n", mp->mnt_nvnodelistsize);
3202 	db_printf("    mnt_activevnodelistsize = %d\n",
3203 	    mp->mnt_activevnodelistsize);
3204 	db_printf("    mnt_writeopcount = %d\n", mp->mnt_writeopcount);
3205 	db_printf("    mnt_maxsymlinklen = %d\n", mp->mnt_maxsymlinklen);
3206 	db_printf("    mnt_iosize_max = %d\n", mp->mnt_iosize_max);
3207 	db_printf("    mnt_hashseed = %u\n", mp->mnt_hashseed);
3208 	db_printf("    mnt_lockref = %d\n", mp->mnt_lockref);
3209 	db_printf("    mnt_secondary_writes = %d\n", mp->mnt_secondary_writes);
3210 	db_printf("    mnt_secondary_accwrites = %d\n",
3211 	    mp->mnt_secondary_accwrites);
3212 	db_printf("    mnt_gjprovider = %s\n",
3213 	    mp->mnt_gjprovider != NULL ? mp->mnt_gjprovider : "NULL");
3214 
3215 	db_printf("\n\nList of active vnodes\n");
3216 	TAILQ_FOREACH(vp, &mp->mnt_activevnodelist, v_actfreelist) {
3217 		if (vp->v_type != VMARKER) {
3218 			vn_printf(vp, "vnode ");
3219 			if (db_pager_quit)
3220 				break;
3221 		}
3222 	}
3223 	db_printf("\n\nList of inactive vnodes\n");
3224 	TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3225 		if (vp->v_type != VMARKER && (vp->v_iflag & VI_ACTIVE) == 0) {
3226 			vn_printf(vp, "vnode ");
3227 			if (db_pager_quit)
3228 				break;
3229 		}
3230 	}
3231 }
3232 #endif	/* DDB */
3233 
3234 /*
3235  * Fill in a struct xvfsconf based on a struct vfsconf.
3236  */
3237 static int
3238 vfsconf2x(struct sysctl_req *req, struct vfsconf *vfsp)
3239 {
3240 	struct xvfsconf xvfsp;
3241 
3242 	bzero(&xvfsp, sizeof(xvfsp));
3243 	strcpy(xvfsp.vfc_name, vfsp->vfc_name);
3244 	xvfsp.vfc_typenum = vfsp->vfc_typenum;
3245 	xvfsp.vfc_refcount = vfsp->vfc_refcount;
3246 	xvfsp.vfc_flags = vfsp->vfc_flags;
3247 	/*
3248 	 * These are unused in userland, we keep them
3249 	 * to not break binary compatibility.
3250 	 */
3251 	xvfsp.vfc_vfsops = NULL;
3252 	xvfsp.vfc_next = NULL;
3253 	return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
3254 }
3255 
3256 #ifdef COMPAT_FREEBSD32
3257 struct xvfsconf32 {
3258 	uint32_t	vfc_vfsops;
3259 	char		vfc_name[MFSNAMELEN];
3260 	int32_t		vfc_typenum;
3261 	int32_t		vfc_refcount;
3262 	int32_t		vfc_flags;
3263 	uint32_t	vfc_next;
3264 };
3265 
3266 static int
3267 vfsconf2x32(struct sysctl_req *req, struct vfsconf *vfsp)
3268 {
3269 	struct xvfsconf32 xvfsp;
3270 
3271 	strcpy(xvfsp.vfc_name, vfsp->vfc_name);
3272 	xvfsp.vfc_typenum = vfsp->vfc_typenum;
3273 	xvfsp.vfc_refcount = vfsp->vfc_refcount;
3274 	xvfsp.vfc_flags = vfsp->vfc_flags;
3275 	xvfsp.vfc_vfsops = 0;
3276 	xvfsp.vfc_next = 0;
3277 	return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
3278 }
3279 #endif
3280 
3281 /*
3282  * Top level filesystem related information gathering.
3283  */
3284 static int
3285 sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS)
3286 {
3287 	struct vfsconf *vfsp;
3288 	int error;
3289 
3290 	error = 0;
3291 	vfsconf_slock();
3292 	TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3293 #ifdef COMPAT_FREEBSD32
3294 		if (req->flags & SCTL_MASK32)
3295 			error = vfsconf2x32(req, vfsp);
3296 		else
3297 #endif
3298 			error = vfsconf2x(req, vfsp);
3299 		if (error)
3300 			break;
3301 	}
3302 	vfsconf_sunlock();
3303 	return (error);
3304 }
3305 
3306 SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLTYPE_OPAQUE | CTLFLAG_RD |
3307     CTLFLAG_MPSAFE, NULL, 0, sysctl_vfs_conflist,
3308     "S,xvfsconf", "List of all configured filesystems");
3309 
3310 #ifndef BURN_BRIDGES
3311 static int	sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
3312 
3313 static int
3314 vfs_sysctl(SYSCTL_HANDLER_ARGS)
3315 {
3316 	int *name = (int *)arg1 - 1;	/* XXX */
3317 	u_int namelen = arg2 + 1;	/* XXX */
3318 	struct vfsconf *vfsp;
3319 
3320 	log(LOG_WARNING, "userland calling deprecated sysctl, "
3321 	    "please rebuild world\n");
3322 
3323 #if 1 || defined(COMPAT_PRELITE2)
3324 	/* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
3325 	if (namelen == 1)
3326 		return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
3327 #endif
3328 
3329 	switch (name[1]) {
3330 	case VFS_MAXTYPENUM:
3331 		if (namelen != 2)
3332 			return (ENOTDIR);
3333 		return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
3334 	case VFS_CONF:
3335 		if (namelen != 3)
3336 			return (ENOTDIR);	/* overloaded */
3337 		vfsconf_slock();
3338 		TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3339 			if (vfsp->vfc_typenum == name[2])
3340 				break;
3341 		}
3342 		vfsconf_sunlock();
3343 		if (vfsp == NULL)
3344 			return (EOPNOTSUPP);
3345 #ifdef COMPAT_FREEBSD32
3346 		if (req->flags & SCTL_MASK32)
3347 			return (vfsconf2x32(req, vfsp));
3348 		else
3349 #endif
3350 			return (vfsconf2x(req, vfsp));
3351 	}
3352 	return (EOPNOTSUPP);
3353 }
3354 
3355 static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP |
3356     CTLFLAG_MPSAFE, vfs_sysctl,
3357     "Generic filesystem");
3358 
3359 #if 1 || defined(COMPAT_PRELITE2)
3360 
3361 static int
3362 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
3363 {
3364 	int error;
3365 	struct vfsconf *vfsp;
3366 	struct ovfsconf ovfs;
3367 
3368 	vfsconf_slock();
3369 	TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3370 		bzero(&ovfs, sizeof(ovfs));
3371 		ovfs.vfc_vfsops = vfsp->vfc_vfsops;	/* XXX used as flag */
3372 		strcpy(ovfs.vfc_name, vfsp->vfc_name);
3373 		ovfs.vfc_index = vfsp->vfc_typenum;
3374 		ovfs.vfc_refcount = vfsp->vfc_refcount;
3375 		ovfs.vfc_flags = vfsp->vfc_flags;
3376 		error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
3377 		if (error != 0) {
3378 			vfsconf_sunlock();
3379 			return (error);
3380 		}
3381 	}
3382 	vfsconf_sunlock();
3383 	return (0);
3384 }
3385 
3386 #endif /* 1 || COMPAT_PRELITE2 */
3387 #endif /* !BURN_BRIDGES */
3388 
3389 #define KINFO_VNODESLOP		10
3390 #ifdef notyet
3391 /*
3392  * Dump vnode list (via sysctl).
3393  */
3394 /* ARGSUSED */
3395 static int
3396 sysctl_vnode(SYSCTL_HANDLER_ARGS)
3397 {
3398 	struct xvnode *xvn;
3399 	struct mount *mp;
3400 	struct vnode *vp;
3401 	int error, len, n;
3402 
3403 	/*
3404 	 * Stale numvnodes access is not fatal here.
3405 	 */
3406 	req->lock = 0;
3407 	len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn;
3408 	if (!req->oldptr)
3409 		/* Make an estimate */
3410 		return (SYSCTL_OUT(req, 0, len));
3411 
3412 	error = sysctl_wire_old_buffer(req, 0);
3413 	if (error != 0)
3414 		return (error);
3415 	xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK);
3416 	n = 0;
3417 	mtx_lock(&mountlist_mtx);
3418 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3419 		if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK))
3420 			continue;
3421 		MNT_ILOCK(mp);
3422 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3423 			if (n == len)
3424 				break;
3425 			vref(vp);
3426 			xvn[n].xv_size = sizeof *xvn;
3427 			xvn[n].xv_vnode = vp;
3428 			xvn[n].xv_id = 0;	/* XXX compat */
3429 #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field
3430 			XV_COPY(usecount);
3431 			XV_COPY(writecount);
3432 			XV_COPY(holdcnt);
3433 			XV_COPY(mount);
3434 			XV_COPY(numoutput);
3435 			XV_COPY(type);
3436 #undef XV_COPY
3437 			xvn[n].xv_flag = vp->v_vflag;
3438 
3439 			switch (vp->v_type) {
3440 			case VREG:
3441 			case VDIR:
3442 			case VLNK:
3443 				break;
3444 			case VBLK:
3445 			case VCHR:
3446 				if (vp->v_rdev == NULL) {
3447 					vrele(vp);
3448 					continue;
3449 				}
3450 				xvn[n].xv_dev = dev2udev(vp->v_rdev);
3451 				break;
3452 			case VSOCK:
3453 				xvn[n].xv_socket = vp->v_socket;
3454 				break;
3455 			case VFIFO:
3456 				xvn[n].xv_fifo = vp->v_fifoinfo;
3457 				break;
3458 			case VNON:
3459 			case VBAD:
3460 			default:
3461 				/* shouldn't happen? */
3462 				vrele(vp);
3463 				continue;
3464 			}
3465 			vrele(vp);
3466 			++n;
3467 		}
3468 		MNT_IUNLOCK(mp);
3469 		mtx_lock(&mountlist_mtx);
3470 		vfs_unbusy(mp);
3471 		if (n == len)
3472 			break;
3473 	}
3474 	mtx_unlock(&mountlist_mtx);
3475 
3476 	error = SYSCTL_OUT(req, xvn, n * sizeof *xvn);
3477 	free(xvn, M_TEMP);
3478 	return (error);
3479 }
3480 
3481 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE | CTLFLAG_RD |
3482     CTLFLAG_MPSAFE, 0, 0, sysctl_vnode, "S,xvnode",
3483     "");
3484 #endif
3485 
3486 /*
3487  * Unmount all filesystems. The list is traversed in reverse order
3488  * of mounting to avoid dependencies.
3489  */
3490 void
3491 vfs_unmountall(void)
3492 {
3493 	struct mount *mp;
3494 	struct thread *td;
3495 	int error;
3496 
3497 	CTR1(KTR_VFS, "%s: unmounting all filesystems", __func__);
3498 	td = curthread;
3499 
3500 	/*
3501 	 * Since this only runs when rebooting, it is not interlocked.
3502 	 */
3503 	while(!TAILQ_EMPTY(&mountlist)) {
3504 		mp = TAILQ_LAST(&mountlist, mntlist);
3505 		error = dounmount(mp, MNT_FORCE, td);
3506 		if (error) {
3507 			TAILQ_REMOVE(&mountlist, mp, mnt_list);
3508 			/*
3509 			 * XXX: Due to the way in which we mount the root
3510 			 * file system off of devfs, devfs will generate a
3511 			 * "busy" warning when we try to unmount it before
3512 			 * the root.  Don't print a warning as a result in
3513 			 * order to avoid false positive errors that may
3514 			 * cause needless upset.
3515 			 */
3516 			if (strcmp(mp->mnt_vfc->vfc_name, "devfs") != 0) {
3517 				printf("unmount of %s failed (",
3518 				    mp->mnt_stat.f_mntonname);
3519 				if (error == EBUSY)
3520 					printf("BUSY)\n");
3521 				else
3522 					printf("%d)\n", error);
3523 			}
3524 		} else {
3525 			/* The unmount has removed mp from the mountlist */
3526 		}
3527 	}
3528 }
3529 
3530 /*
3531  * perform msync on all vnodes under a mount point
3532  * the mount point must be locked.
3533  */
3534 void
3535 vfs_msync(struct mount *mp, int flags)
3536 {
3537 	struct vnode *vp, *mvp;
3538 	struct vm_object *obj;
3539 
3540 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
3541 	MNT_VNODE_FOREACH_ACTIVE(vp, mp, mvp) {
3542 		obj = vp->v_object;
3543 		if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0 &&
3544 		    (flags == MNT_WAIT || VOP_ISLOCKED(vp) == 0)) {
3545 			if (!vget(vp,
3546 			    LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
3547 			    curthread)) {
3548 				if (vp->v_vflag & VV_NOSYNC) {	/* unlinked */
3549 					vput(vp);
3550 					continue;
3551 				}
3552 
3553 				obj = vp->v_object;
3554 				if (obj != NULL) {
3555 					VM_OBJECT_WLOCK(obj);
3556 					vm_object_page_clean(obj, 0, 0,
3557 					    flags == MNT_WAIT ?
3558 					    OBJPC_SYNC : OBJPC_NOSYNC);
3559 					VM_OBJECT_WUNLOCK(obj);
3560 				}
3561 				vput(vp);
3562 			}
3563 		} else
3564 			VI_UNLOCK(vp);
3565 	}
3566 }
3567 
3568 static void
3569 destroy_vpollinfo_free(struct vpollinfo *vi)
3570 {
3571 
3572 	knlist_destroy(&vi->vpi_selinfo.si_note);
3573 	mtx_destroy(&vi->vpi_lock);
3574 	uma_zfree(vnodepoll_zone, vi);
3575 }
3576 
3577 static void
3578 destroy_vpollinfo(struct vpollinfo *vi)
3579 {
3580 
3581 	knlist_clear(&vi->vpi_selinfo.si_note, 1);
3582 	seldrain(&vi->vpi_selinfo);
3583 	destroy_vpollinfo_free(vi);
3584 }
3585 
3586 /*
3587  * Initalize per-vnode helper structure to hold poll-related state.
3588  */
3589 void
3590 v_addpollinfo(struct vnode *vp)
3591 {
3592 	struct vpollinfo *vi;
3593 
3594 	if (vp->v_pollinfo != NULL)
3595 		return;
3596 	vi = uma_zalloc(vnodepoll_zone, M_WAITOK);
3597 	mtx_init(&vi->vpi_lock, "vnode pollinfo", NULL, MTX_DEF);
3598 	knlist_init(&vi->vpi_selinfo.si_note, vp, vfs_knllock,
3599 	    vfs_knlunlock, vfs_knl_assert_locked, vfs_knl_assert_unlocked);
3600 	VI_LOCK(vp);
3601 	if (vp->v_pollinfo != NULL) {
3602 		VI_UNLOCK(vp);
3603 		destroy_vpollinfo_free(vi);
3604 		return;
3605 	}
3606 	vp->v_pollinfo = vi;
3607 	VI_UNLOCK(vp);
3608 }
3609 
3610 /*
3611  * Record a process's interest in events which might happen to
3612  * a vnode.  Because poll uses the historic select-style interface
3613  * internally, this routine serves as both the ``check for any
3614  * pending events'' and the ``record my interest in future events''
3615  * functions.  (These are done together, while the lock is held,
3616  * to avoid race conditions.)
3617  */
3618 int
3619 vn_pollrecord(struct vnode *vp, struct thread *td, int events)
3620 {
3621 
3622 	v_addpollinfo(vp);
3623 	mtx_lock(&vp->v_pollinfo->vpi_lock);
3624 	if (vp->v_pollinfo->vpi_revents & events) {
3625 		/*
3626 		 * This leaves events we are not interested
3627 		 * in available for the other process which
3628 		 * which presumably had requested them
3629 		 * (otherwise they would never have been
3630 		 * recorded).
3631 		 */
3632 		events &= vp->v_pollinfo->vpi_revents;
3633 		vp->v_pollinfo->vpi_revents &= ~events;
3634 
3635 		mtx_unlock(&vp->v_pollinfo->vpi_lock);
3636 		return (events);
3637 	}
3638 	vp->v_pollinfo->vpi_events |= events;
3639 	selrecord(td, &vp->v_pollinfo->vpi_selinfo);
3640 	mtx_unlock(&vp->v_pollinfo->vpi_lock);
3641 	return (0);
3642 }
3643 
3644 /*
3645  * Routine to create and manage a filesystem syncer vnode.
3646  */
3647 #define sync_close ((int (*)(struct  vop_close_args *))nullop)
3648 static int	sync_fsync(struct  vop_fsync_args *);
3649 static int	sync_inactive(struct  vop_inactive_args *);
3650 static int	sync_reclaim(struct  vop_reclaim_args *);
3651 
3652 static struct vop_vector sync_vnodeops = {
3653 	.vop_bypass =	VOP_EOPNOTSUPP,
3654 	.vop_close =	sync_close,		/* close */
3655 	.vop_fsync =	sync_fsync,		/* fsync */
3656 	.vop_inactive =	sync_inactive,	/* inactive */
3657 	.vop_reclaim =	sync_reclaim,	/* reclaim */
3658 	.vop_lock1 =	vop_stdlock,	/* lock */
3659 	.vop_unlock =	vop_stdunlock,	/* unlock */
3660 	.vop_islocked =	vop_stdislocked,	/* islocked */
3661 };
3662 
3663 /*
3664  * Create a new filesystem syncer vnode for the specified mount point.
3665  */
3666 void
3667 vfs_allocate_syncvnode(struct mount *mp)
3668 {
3669 	struct vnode *vp;
3670 	struct bufobj *bo;
3671 	static long start, incr, next;
3672 	int error;
3673 
3674 	/* Allocate a new vnode */
3675 	error = getnewvnode("syncer", mp, &sync_vnodeops, &vp);
3676 	if (error != 0)
3677 		panic("vfs_allocate_syncvnode: getnewvnode() failed");
3678 	vp->v_type = VNON;
3679 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3680 	vp->v_vflag |= VV_FORCEINSMQ;
3681 	error = insmntque(vp, mp);
3682 	if (error != 0)
3683 		panic("vfs_allocate_syncvnode: insmntque() failed");
3684 	vp->v_vflag &= ~VV_FORCEINSMQ;
3685 	VOP_UNLOCK(vp, 0);
3686 	/*
3687 	 * Place the vnode onto the syncer worklist. We attempt to
3688 	 * scatter them about on the list so that they will go off
3689 	 * at evenly distributed times even if all the filesystems
3690 	 * are mounted at once.
3691 	 */
3692 	next += incr;
3693 	if (next == 0 || next > syncer_maxdelay) {
3694 		start /= 2;
3695 		incr /= 2;
3696 		if (start == 0) {
3697 			start = syncer_maxdelay / 2;
3698 			incr = syncer_maxdelay;
3699 		}
3700 		next = start;
3701 	}
3702 	bo = &vp->v_bufobj;
3703 	BO_LOCK(bo);
3704 	vn_syncer_add_to_worklist(bo, syncdelay > 0 ? next % syncdelay : 0);
3705 	/* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */
3706 	mtx_lock(&sync_mtx);
3707 	sync_vnode_count++;
3708 	if (mp->mnt_syncer == NULL) {
3709 		mp->mnt_syncer = vp;
3710 		vp = NULL;
3711 	}
3712 	mtx_unlock(&sync_mtx);
3713 	BO_UNLOCK(bo);
3714 	if (vp != NULL) {
3715 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3716 		vgone(vp);
3717 		vput(vp);
3718 	}
3719 }
3720 
3721 void
3722 vfs_deallocate_syncvnode(struct mount *mp)
3723 {
3724 	struct vnode *vp;
3725 
3726 	mtx_lock(&sync_mtx);
3727 	vp = mp->mnt_syncer;
3728 	if (vp != NULL)
3729 		mp->mnt_syncer = NULL;
3730 	mtx_unlock(&sync_mtx);
3731 	if (vp != NULL)
3732 		vrele(vp);
3733 }
3734 
3735 /*
3736  * Do a lazy sync of the filesystem.
3737  */
3738 static int
3739 sync_fsync(struct vop_fsync_args *ap)
3740 {
3741 	struct vnode *syncvp = ap->a_vp;
3742 	struct mount *mp = syncvp->v_mount;
3743 	int error, save;
3744 	struct bufobj *bo;
3745 
3746 	/*
3747 	 * We only need to do something if this is a lazy evaluation.
3748 	 */
3749 	if (ap->a_waitfor != MNT_LAZY)
3750 		return (0);
3751 
3752 	/*
3753 	 * Move ourselves to the back of the sync list.
3754 	 */
3755 	bo = &syncvp->v_bufobj;
3756 	BO_LOCK(bo);
3757 	vn_syncer_add_to_worklist(bo, syncdelay);
3758 	BO_UNLOCK(bo);
3759 
3760 	/*
3761 	 * Walk the list of vnodes pushing all that are dirty and
3762 	 * not already on the sync list.
3763 	 */
3764 	if (vfs_busy(mp, MBF_NOWAIT) != 0)
3765 		return (0);
3766 	if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
3767 		vfs_unbusy(mp);
3768 		return (0);
3769 	}
3770 	save = curthread_pflags_set(TDP_SYNCIO);
3771 	vfs_msync(mp, MNT_NOWAIT);
3772 	error = VFS_SYNC(mp, MNT_LAZY);
3773 	curthread_pflags_restore(save);
3774 	vn_finished_write(mp);
3775 	vfs_unbusy(mp);
3776 	return (error);
3777 }
3778 
3779 /*
3780  * The syncer vnode is no referenced.
3781  */
3782 static int
3783 sync_inactive(struct vop_inactive_args *ap)
3784 {
3785 
3786 	vgone(ap->a_vp);
3787 	return (0);
3788 }
3789 
3790 /*
3791  * The syncer vnode is no longer needed and is being decommissioned.
3792  *
3793  * Modifications to the worklist must be protected by sync_mtx.
3794  */
3795 static int
3796 sync_reclaim(struct vop_reclaim_args *ap)
3797 {
3798 	struct vnode *vp = ap->a_vp;
3799 	struct bufobj *bo;
3800 
3801 	bo = &vp->v_bufobj;
3802 	BO_LOCK(bo);
3803 	mtx_lock(&sync_mtx);
3804 	if (vp->v_mount->mnt_syncer == vp)
3805 		vp->v_mount->mnt_syncer = NULL;
3806 	if (bo->bo_flag & BO_ONWORKLST) {
3807 		LIST_REMOVE(bo, bo_synclist);
3808 		syncer_worklist_len--;
3809 		sync_vnode_count--;
3810 		bo->bo_flag &= ~BO_ONWORKLST;
3811 	}
3812 	mtx_unlock(&sync_mtx);
3813 	BO_UNLOCK(bo);
3814 
3815 	return (0);
3816 }
3817 
3818 /*
3819  * Check if vnode represents a disk device
3820  */
3821 int
3822 vn_isdisk(struct vnode *vp, int *errp)
3823 {
3824 	int error;
3825 
3826 	if (vp->v_type != VCHR) {
3827 		error = ENOTBLK;
3828 		goto out;
3829 	}
3830 	error = 0;
3831 	dev_lock();
3832 	if (vp->v_rdev == NULL)
3833 		error = ENXIO;
3834 	else if (vp->v_rdev->si_devsw == NULL)
3835 		error = ENXIO;
3836 	else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK))
3837 		error = ENOTBLK;
3838 	dev_unlock();
3839 out:
3840 	if (errp != NULL)
3841 		*errp = error;
3842 	return (error == 0);
3843 }
3844 
3845 /*
3846  * Common filesystem object access control check routine.  Accepts a
3847  * vnode's type, "mode", uid and gid, requested access mode, credentials,
3848  * and optional call-by-reference privused argument allowing vaccess()
3849  * to indicate to the caller whether privilege was used to satisfy the
3850  * request (obsoleted).  Returns 0 on success, or an errno on failure.
3851  */
3852 int
3853 vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid,
3854     accmode_t accmode, struct ucred *cred, int *privused)
3855 {
3856 	accmode_t dac_granted;
3857 	accmode_t priv_granted;
3858 
3859 	KASSERT((accmode & ~(VEXEC | VWRITE | VREAD | VADMIN | VAPPEND)) == 0,
3860 	    ("invalid bit in accmode"));
3861 	KASSERT((accmode & VAPPEND) == 0 || (accmode & VWRITE),
3862 	    ("VAPPEND without VWRITE"));
3863 
3864 	/*
3865 	 * Look for a normal, non-privileged way to access the file/directory
3866 	 * as requested.  If it exists, go with that.
3867 	 */
3868 
3869 	if (privused != NULL)
3870 		*privused = 0;
3871 
3872 	dac_granted = 0;
3873 
3874 	/* Check the owner. */
3875 	if (cred->cr_uid == file_uid) {
3876 		dac_granted |= VADMIN;
3877 		if (file_mode & S_IXUSR)
3878 			dac_granted |= VEXEC;
3879 		if (file_mode & S_IRUSR)
3880 			dac_granted |= VREAD;
3881 		if (file_mode & S_IWUSR)
3882 			dac_granted |= (VWRITE | VAPPEND);
3883 
3884 		if ((accmode & dac_granted) == accmode)
3885 			return (0);
3886 
3887 		goto privcheck;
3888 	}
3889 
3890 	/* Otherwise, check the groups (first match) */
3891 	if (groupmember(file_gid, cred)) {
3892 		if (file_mode & S_IXGRP)
3893 			dac_granted |= VEXEC;
3894 		if (file_mode & S_IRGRP)
3895 			dac_granted |= VREAD;
3896 		if (file_mode & S_IWGRP)
3897 			dac_granted |= (VWRITE | VAPPEND);
3898 
3899 		if ((accmode & dac_granted) == accmode)
3900 			return (0);
3901 
3902 		goto privcheck;
3903 	}
3904 
3905 	/* Otherwise, check everyone else. */
3906 	if (file_mode & S_IXOTH)
3907 		dac_granted |= VEXEC;
3908 	if (file_mode & S_IROTH)
3909 		dac_granted |= VREAD;
3910 	if (file_mode & S_IWOTH)
3911 		dac_granted |= (VWRITE | VAPPEND);
3912 	if ((accmode & dac_granted) == accmode)
3913 		return (0);
3914 
3915 privcheck:
3916 	/*
3917 	 * Build a privilege mask to determine if the set of privileges
3918 	 * satisfies the requirements when combined with the granted mask
3919 	 * from above.  For each privilege, if the privilege is required,
3920 	 * bitwise or the request type onto the priv_granted mask.
3921 	 */
3922 	priv_granted = 0;
3923 
3924 	if (type == VDIR) {
3925 		/*
3926 		 * For directories, use PRIV_VFS_LOOKUP to satisfy VEXEC
3927 		 * requests, instead of PRIV_VFS_EXEC.
3928 		 */
3929 		if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3930 		    !priv_check_cred(cred, PRIV_VFS_LOOKUP, 0))
3931 			priv_granted |= VEXEC;
3932 	} else {
3933 		/*
3934 		 * Ensure that at least one execute bit is on. Otherwise,
3935 		 * a privileged user will always succeed, and we don't want
3936 		 * this to happen unless the file really is executable.
3937 		 */
3938 		if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3939 		    (file_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) != 0 &&
3940 		    !priv_check_cred(cred, PRIV_VFS_EXEC, 0))
3941 			priv_granted |= VEXEC;
3942 	}
3943 
3944 	if ((accmode & VREAD) && ((dac_granted & VREAD) == 0) &&
3945 	    !priv_check_cred(cred, PRIV_VFS_READ, 0))
3946 		priv_granted |= VREAD;
3947 
3948 	if ((accmode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
3949 	    !priv_check_cred(cred, PRIV_VFS_WRITE, 0))
3950 		priv_granted |= (VWRITE | VAPPEND);
3951 
3952 	if ((accmode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
3953 	    !priv_check_cred(cred, PRIV_VFS_ADMIN, 0))
3954 		priv_granted |= VADMIN;
3955 
3956 	if ((accmode & (priv_granted | dac_granted)) == accmode) {
3957 		/* XXX audit: privilege used */
3958 		if (privused != NULL)
3959 			*privused = 1;
3960 		return (0);
3961 	}
3962 
3963 	return ((accmode & VADMIN) ? EPERM : EACCES);
3964 }
3965 
3966 /*
3967  * Credential check based on process requesting service, and per-attribute
3968  * permissions.
3969  */
3970 int
3971 extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred,
3972     struct thread *td, accmode_t accmode)
3973 {
3974 
3975 	/*
3976 	 * Kernel-invoked always succeeds.
3977 	 */
3978 	if (cred == NOCRED)
3979 		return (0);
3980 
3981 	/*
3982 	 * Do not allow privileged processes in jail to directly manipulate
3983 	 * system attributes.
3984 	 */
3985 	switch (attrnamespace) {
3986 	case EXTATTR_NAMESPACE_SYSTEM:
3987 		/* Potentially should be: return (EPERM); */
3988 		return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM, 0));
3989 	case EXTATTR_NAMESPACE_USER:
3990 		return (VOP_ACCESS(vp, accmode, cred, td));
3991 	default:
3992 		return (EPERM);
3993 	}
3994 }
3995 
3996 #ifdef DEBUG_VFS_LOCKS
3997 /*
3998  * This only exists to supress warnings from unlocked specfs accesses.  It is
3999  * no longer ok to have an unlocked VFS.
4000  */
4001 #define	IGNORE_LOCK(vp) (panicstr != NULL || (vp) == NULL ||		\
4002 	(vp)->v_type == VCHR ||	(vp)->v_type == VBAD)
4003 
4004 int vfs_badlock_ddb = 1;	/* Drop into debugger on violation. */
4005 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0,
4006     "Drop into debugger on lock violation");
4007 
4008 int vfs_badlock_mutex = 1;	/* Check for interlock across VOPs. */
4009 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex,
4010     0, "Check for interlock across VOPs");
4011 
4012 int vfs_badlock_print = 1;	/* Print lock violations. */
4013 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print,
4014     0, "Print lock violations");
4015 
4016 #ifdef KDB
4017 int vfs_badlock_backtrace = 1;	/* Print backtrace at lock violations. */
4018 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW,
4019     &vfs_badlock_backtrace, 0, "Print backtrace at lock violations");
4020 #endif
4021 
4022 static void
4023 vfs_badlock(const char *msg, const char *str, struct vnode *vp)
4024 {
4025 
4026 #ifdef KDB
4027 	if (vfs_badlock_backtrace)
4028 		kdb_backtrace();
4029 #endif
4030 	if (vfs_badlock_print)
4031 		printf("%s: %p %s\n", str, (void *)vp, msg);
4032 	if (vfs_badlock_ddb)
4033 		kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
4034 }
4035 
4036 void
4037 assert_vi_locked(struct vnode *vp, const char *str)
4038 {
4039 
4040 	if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp)))
4041 		vfs_badlock("interlock is not locked but should be", str, vp);
4042 }
4043 
4044 void
4045 assert_vi_unlocked(struct vnode *vp, const char *str)
4046 {
4047 
4048 	if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp)))
4049 		vfs_badlock("interlock is locked but should not be", str, vp);
4050 }
4051 
4052 void
4053 assert_vop_locked(struct vnode *vp, const char *str)
4054 {
4055 	int locked;
4056 
4057 	if (!IGNORE_LOCK(vp)) {
4058 		locked = VOP_ISLOCKED(vp);
4059 		if (locked == 0 || locked == LK_EXCLOTHER)
4060 			vfs_badlock("is not locked but should be", str, vp);
4061 	}
4062 }
4063 
4064 void
4065 assert_vop_unlocked(struct vnode *vp, const char *str)
4066 {
4067 
4068 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == LK_EXCLUSIVE)
4069 		vfs_badlock("is locked but should not be", str, vp);
4070 }
4071 
4072 void
4073 assert_vop_elocked(struct vnode *vp, const char *str)
4074 {
4075 
4076 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
4077 		vfs_badlock("is not exclusive locked but should be", str, vp);
4078 }
4079 
4080 #if 0
4081 void
4082 assert_vop_elocked_other(struct vnode *vp, const char *str)
4083 {
4084 
4085 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLOTHER)
4086 		vfs_badlock("is not exclusive locked by another thread",
4087 		    str, vp);
4088 }
4089 
4090 void
4091 assert_vop_slocked(struct vnode *vp, const char *str)
4092 {
4093 
4094 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_SHARED)
4095 		vfs_badlock("is not locked shared but should be", str, vp);
4096 }
4097 #endif /* 0 */
4098 #endif /* DEBUG_VFS_LOCKS */
4099 
4100 void
4101 vop_rename_fail(struct vop_rename_args *ap)
4102 {
4103 
4104 	if (ap->a_tvp != NULL)
4105 		vput(ap->a_tvp);
4106 	if (ap->a_tdvp == ap->a_tvp)
4107 		vrele(ap->a_tdvp);
4108 	else
4109 		vput(ap->a_tdvp);
4110 	vrele(ap->a_fdvp);
4111 	vrele(ap->a_fvp);
4112 }
4113 
4114 void
4115 vop_rename_pre(void *ap)
4116 {
4117 	struct vop_rename_args *a = ap;
4118 
4119 #ifdef DEBUG_VFS_LOCKS
4120 	if (a->a_tvp)
4121 		ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME");
4122 	ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME");
4123 	ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME");
4124 	ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME");
4125 
4126 	/* Check the source (from). */
4127 	if (a->a_tdvp->v_vnlock != a->a_fdvp->v_vnlock &&
4128 	    (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fdvp->v_vnlock))
4129 		ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked");
4130 	if (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fvp->v_vnlock)
4131 		ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: fvp locked");
4132 
4133 	/* Check the target. */
4134 	if (a->a_tvp)
4135 		ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked");
4136 	ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked");
4137 #endif
4138 	if (a->a_tdvp != a->a_fdvp)
4139 		vhold(a->a_fdvp);
4140 	if (a->a_tvp != a->a_fvp)
4141 		vhold(a->a_fvp);
4142 	vhold(a->a_tdvp);
4143 	if (a->a_tvp)
4144 		vhold(a->a_tvp);
4145 }
4146 
4147 void
4148 vop_strategy_pre(void *ap)
4149 {
4150 #ifdef DEBUG_VFS_LOCKS
4151 	struct vop_strategy_args *a;
4152 	struct buf *bp;
4153 
4154 	a = ap;
4155 	bp = a->a_bp;
4156 
4157 	/*
4158 	 * Cluster ops lock their component buffers but not the IO container.
4159 	 */
4160 	if ((bp->b_flags & B_CLUSTER) != 0)
4161 		return;
4162 
4163 	if (panicstr == NULL && !BUF_ISLOCKED(bp)) {
4164 		if (vfs_badlock_print)
4165 			printf(
4166 			    "VOP_STRATEGY: bp is not locked but should be\n");
4167 		if (vfs_badlock_ddb)
4168 			kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
4169 	}
4170 #endif
4171 }
4172 
4173 void
4174 vop_lock_pre(void *ap)
4175 {
4176 #ifdef DEBUG_VFS_LOCKS
4177 	struct vop_lock1_args *a = ap;
4178 
4179 	if ((a->a_flags & LK_INTERLOCK) == 0)
4180 		ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
4181 	else
4182 		ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK");
4183 #endif
4184 }
4185 
4186 void
4187 vop_lock_post(void *ap, int rc)
4188 {
4189 #ifdef DEBUG_VFS_LOCKS
4190 	struct vop_lock1_args *a = ap;
4191 
4192 	ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
4193 	if (rc == 0 && (a->a_flags & LK_EXCLOTHER) == 0)
4194 		ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK");
4195 #endif
4196 }
4197 
4198 void
4199 vop_unlock_pre(void *ap)
4200 {
4201 #ifdef DEBUG_VFS_LOCKS
4202 	struct vop_unlock_args *a = ap;
4203 
4204 	if (a->a_flags & LK_INTERLOCK)
4205 		ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK");
4206 	ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK");
4207 #endif
4208 }
4209 
4210 void
4211 vop_unlock_post(void *ap, int rc)
4212 {
4213 #ifdef DEBUG_VFS_LOCKS
4214 	struct vop_unlock_args *a = ap;
4215 
4216 	if (a->a_flags & LK_INTERLOCK)
4217 		ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK");
4218 #endif
4219 }
4220 
4221 void
4222 vop_create_post(void *ap, int rc)
4223 {
4224 	struct vop_create_args *a = ap;
4225 
4226 	if (!rc)
4227 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4228 }
4229 
4230 void
4231 vop_deleteextattr_post(void *ap, int rc)
4232 {
4233 	struct vop_deleteextattr_args *a = ap;
4234 
4235 	if (!rc)
4236 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4237 }
4238 
4239 void
4240 vop_link_post(void *ap, int rc)
4241 {
4242 	struct vop_link_args *a = ap;
4243 
4244 	if (!rc) {
4245 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_LINK);
4246 		VFS_KNOTE_LOCKED(a->a_tdvp, NOTE_WRITE);
4247 	}
4248 }
4249 
4250 void
4251 vop_mkdir_post(void *ap, int rc)
4252 {
4253 	struct vop_mkdir_args *a = ap;
4254 
4255 	if (!rc)
4256 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
4257 }
4258 
4259 void
4260 vop_mknod_post(void *ap, int rc)
4261 {
4262 	struct vop_mknod_args *a = ap;
4263 
4264 	if (!rc)
4265 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4266 }
4267 
4268 void
4269 vop_remove_post(void *ap, int rc)
4270 {
4271 	struct vop_remove_args *a = ap;
4272 
4273 	if (!rc) {
4274 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4275 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
4276 	}
4277 }
4278 
4279 void
4280 vop_rename_post(void *ap, int rc)
4281 {
4282 	struct vop_rename_args *a = ap;
4283 
4284 	if (!rc) {
4285 		VFS_KNOTE_UNLOCKED(a->a_fdvp, NOTE_WRITE);
4286 		VFS_KNOTE_UNLOCKED(a->a_tdvp, NOTE_WRITE);
4287 		VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME);
4288 		if (a->a_tvp)
4289 			VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE);
4290 	}
4291 	if (a->a_tdvp != a->a_fdvp)
4292 		vdrop(a->a_fdvp);
4293 	if (a->a_tvp != a->a_fvp)
4294 		vdrop(a->a_fvp);
4295 	vdrop(a->a_tdvp);
4296 	if (a->a_tvp)
4297 		vdrop(a->a_tvp);
4298 }
4299 
4300 void
4301 vop_rmdir_post(void *ap, int rc)
4302 {
4303 	struct vop_rmdir_args *a = ap;
4304 
4305 	if (!rc) {
4306 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
4307 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
4308 	}
4309 }
4310 
4311 void
4312 vop_setattr_post(void *ap, int rc)
4313 {
4314 	struct vop_setattr_args *a = ap;
4315 
4316 	if (!rc)
4317 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4318 }
4319 
4320 void
4321 vop_setextattr_post(void *ap, int rc)
4322 {
4323 	struct vop_setextattr_args *a = ap;
4324 
4325 	if (!rc)
4326 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4327 }
4328 
4329 void
4330 vop_symlink_post(void *ap, int rc)
4331 {
4332 	struct vop_symlink_args *a = ap;
4333 
4334 	if (!rc)
4335 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4336 }
4337 
4338 static struct knlist fs_knlist;
4339 
4340 static void
4341 vfs_event_init(void *arg)
4342 {
4343 	knlist_init_mtx(&fs_knlist, NULL);
4344 }
4345 /* XXX - correct order? */
4346 SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL);
4347 
4348 void
4349 vfs_event_signal(fsid_t *fsid, uint32_t event, intptr_t data __unused)
4350 {
4351 
4352 	KNOTE_UNLOCKED(&fs_knlist, event);
4353 }
4354 
4355 static int	filt_fsattach(struct knote *kn);
4356 static void	filt_fsdetach(struct knote *kn);
4357 static int	filt_fsevent(struct knote *kn, long hint);
4358 
4359 struct filterops fs_filtops = {
4360 	.f_isfd = 0,
4361 	.f_attach = filt_fsattach,
4362 	.f_detach = filt_fsdetach,
4363 	.f_event = filt_fsevent
4364 };
4365 
4366 static int
4367 filt_fsattach(struct knote *kn)
4368 {
4369 
4370 	kn->kn_flags |= EV_CLEAR;
4371 	knlist_add(&fs_knlist, kn, 0);
4372 	return (0);
4373 }
4374 
4375 static void
4376 filt_fsdetach(struct knote *kn)
4377 {
4378 
4379 	knlist_remove(&fs_knlist, kn, 0);
4380 }
4381 
4382 static int
4383 filt_fsevent(struct knote *kn, long hint)
4384 {
4385 
4386 	kn->kn_fflags |= hint;
4387 	return (kn->kn_fflags != 0);
4388 }
4389 
4390 static int
4391 sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS)
4392 {
4393 	struct vfsidctl vc;
4394 	int error;
4395 	struct mount *mp;
4396 
4397 	error = SYSCTL_IN(req, &vc, sizeof(vc));
4398 	if (error)
4399 		return (error);
4400 	if (vc.vc_vers != VFS_CTL_VERS1)
4401 		return (EINVAL);
4402 	mp = vfs_getvfs(&vc.vc_fsid);
4403 	if (mp == NULL)
4404 		return (ENOENT);
4405 	/* ensure that a specific sysctl goes to the right filesystem. */
4406 	if (strcmp(vc.vc_fstypename, "*") != 0 &&
4407 	    strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) {
4408 		vfs_rel(mp);
4409 		return (EINVAL);
4410 	}
4411 	VCTLTOREQ(&vc, req);
4412 	error = VFS_SYSCTL(mp, vc.vc_op, req);
4413 	vfs_rel(mp);
4414 	return (error);
4415 }
4416 
4417 SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLTYPE_OPAQUE | CTLFLAG_WR,
4418     NULL, 0, sysctl_vfs_ctl, "",
4419     "Sysctl by fsid");
4420 
4421 /*
4422  * Function to initialize a va_filerev field sensibly.
4423  * XXX: Wouldn't a random number make a lot more sense ??
4424  */
4425 u_quad_t
4426 init_va_filerev(void)
4427 {
4428 	struct bintime bt;
4429 
4430 	getbinuptime(&bt);
4431 	return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL));
4432 }
4433 
4434 static int	filt_vfsread(struct knote *kn, long hint);
4435 static int	filt_vfswrite(struct knote *kn, long hint);
4436 static int	filt_vfsvnode(struct knote *kn, long hint);
4437 static void	filt_vfsdetach(struct knote *kn);
4438 static struct filterops vfsread_filtops = {
4439 	.f_isfd = 1,
4440 	.f_detach = filt_vfsdetach,
4441 	.f_event = filt_vfsread
4442 };
4443 static struct filterops vfswrite_filtops = {
4444 	.f_isfd = 1,
4445 	.f_detach = filt_vfsdetach,
4446 	.f_event = filt_vfswrite
4447 };
4448 static struct filterops vfsvnode_filtops = {
4449 	.f_isfd = 1,
4450 	.f_detach = filt_vfsdetach,
4451 	.f_event = filt_vfsvnode
4452 };
4453 
4454 static void
4455 vfs_knllock(void *arg)
4456 {
4457 	struct vnode *vp = arg;
4458 
4459 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4460 }
4461 
4462 static void
4463 vfs_knlunlock(void *arg)
4464 {
4465 	struct vnode *vp = arg;
4466 
4467 	VOP_UNLOCK(vp, 0);
4468 }
4469 
4470 static void
4471 vfs_knl_assert_locked(void *arg)
4472 {
4473 #ifdef DEBUG_VFS_LOCKS
4474 	struct vnode *vp = arg;
4475 
4476 	ASSERT_VOP_LOCKED(vp, "vfs_knl_assert_locked");
4477 #endif
4478 }
4479 
4480 static void
4481 vfs_knl_assert_unlocked(void *arg)
4482 {
4483 #ifdef DEBUG_VFS_LOCKS
4484 	struct vnode *vp = arg;
4485 
4486 	ASSERT_VOP_UNLOCKED(vp, "vfs_knl_assert_unlocked");
4487 #endif
4488 }
4489 
4490 int
4491 vfs_kqfilter(struct vop_kqfilter_args *ap)
4492 {
4493 	struct vnode *vp = ap->a_vp;
4494 	struct knote *kn = ap->a_kn;
4495 	struct knlist *knl;
4496 
4497 	switch (kn->kn_filter) {
4498 	case EVFILT_READ:
4499 		kn->kn_fop = &vfsread_filtops;
4500 		break;
4501 	case EVFILT_WRITE:
4502 		kn->kn_fop = &vfswrite_filtops;
4503 		break;
4504 	case EVFILT_VNODE:
4505 		kn->kn_fop = &vfsvnode_filtops;
4506 		break;
4507 	default:
4508 		return (EINVAL);
4509 	}
4510 
4511 	kn->kn_hook = (caddr_t)vp;
4512 
4513 	v_addpollinfo(vp);
4514 	if (vp->v_pollinfo == NULL)
4515 		return (ENOMEM);
4516 	knl = &vp->v_pollinfo->vpi_selinfo.si_note;
4517 	vhold(vp);
4518 	knlist_add(knl, kn, 0);
4519 
4520 	return (0);
4521 }
4522 
4523 /*
4524  * Detach knote from vnode
4525  */
4526 static void
4527 filt_vfsdetach(struct knote *kn)
4528 {
4529 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4530 
4531 	KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo"));
4532 	knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0);
4533 	vdrop(vp);
4534 }
4535 
4536 /*ARGSUSED*/
4537 static int
4538 filt_vfsread(struct knote *kn, long hint)
4539 {
4540 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4541 	struct vattr va;
4542 	int res;
4543 
4544 	/*
4545 	 * filesystem is gone, so set the EOF flag and schedule
4546 	 * the knote for deletion.
4547 	 */
4548 	if (hint == NOTE_REVOKE) {
4549 		VI_LOCK(vp);
4550 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4551 		VI_UNLOCK(vp);
4552 		return (1);
4553 	}
4554 
4555 	if (VOP_GETATTR(vp, &va, curthread->td_ucred))
4556 		return (0);
4557 
4558 	VI_LOCK(vp);
4559 	kn->kn_data = va.va_size - kn->kn_fp->f_offset;
4560 	res = (kn->kn_data != 0);
4561 	VI_UNLOCK(vp);
4562 	return (res);
4563 }
4564 
4565 /*ARGSUSED*/
4566 static int
4567 filt_vfswrite(struct knote *kn, long hint)
4568 {
4569 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4570 
4571 	VI_LOCK(vp);
4572 
4573 	/*
4574 	 * filesystem is gone, so set the EOF flag and schedule
4575 	 * the knote for deletion.
4576 	 */
4577 	if (hint == NOTE_REVOKE)
4578 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4579 
4580 	kn->kn_data = 0;
4581 	VI_UNLOCK(vp);
4582 	return (1);
4583 }
4584 
4585 static int
4586 filt_vfsvnode(struct knote *kn, long hint)
4587 {
4588 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4589 	int res;
4590 
4591 	VI_LOCK(vp);
4592 	if (kn->kn_sfflags & hint)
4593 		kn->kn_fflags |= hint;
4594 	if (hint == NOTE_REVOKE) {
4595 		kn->kn_flags |= EV_EOF;
4596 		VI_UNLOCK(vp);
4597 		return (1);
4598 	}
4599 	res = (kn->kn_fflags != 0);
4600 	VI_UNLOCK(vp);
4601 	return (res);
4602 }
4603 
4604 int
4605 vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off)
4606 {
4607 	int error;
4608 
4609 	if (dp->d_reclen > ap->a_uio->uio_resid)
4610 		return (ENAMETOOLONG);
4611 	error = uiomove(dp, dp->d_reclen, ap->a_uio);
4612 	if (error) {
4613 		if (ap->a_ncookies != NULL) {
4614 			if (ap->a_cookies != NULL)
4615 				free(ap->a_cookies, M_TEMP);
4616 			ap->a_cookies = NULL;
4617 			*ap->a_ncookies = 0;
4618 		}
4619 		return (error);
4620 	}
4621 	if (ap->a_ncookies == NULL)
4622 		return (0);
4623 
4624 	KASSERT(ap->a_cookies,
4625 	    ("NULL ap->a_cookies value with non-NULL ap->a_ncookies!"));
4626 
4627 	*ap->a_cookies = realloc(*ap->a_cookies,
4628 	    (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO);
4629 	(*ap->a_cookies)[*ap->a_ncookies] = off;
4630 	return (0);
4631 }
4632 
4633 /*
4634  * Mark for update the access time of the file if the filesystem
4635  * supports VOP_MARKATIME.  This functionality is used by execve and
4636  * mmap, so we want to avoid the I/O implied by directly setting
4637  * va_atime for the sake of efficiency.
4638  */
4639 void
4640 vfs_mark_atime(struct vnode *vp, struct ucred *cred)
4641 {
4642 	struct mount *mp;
4643 
4644 	mp = vp->v_mount;
4645 	ASSERT_VOP_LOCKED(vp, "vfs_mark_atime");
4646 	if (mp != NULL && (mp->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
4647 		(void)VOP_MARKATIME(vp);
4648 }
4649 
4650 /*
4651  * The purpose of this routine is to remove granularity from accmode_t,
4652  * reducing it into standard unix access bits - VEXEC, VREAD, VWRITE,
4653  * VADMIN and VAPPEND.
4654  *
4655  * If it returns 0, the caller is supposed to continue with the usual
4656  * access checks using 'accmode' as modified by this routine.  If it
4657  * returns nonzero value, the caller is supposed to return that value
4658  * as errno.
4659  *
4660  * Note that after this routine runs, accmode may be zero.
4661  */
4662 int
4663 vfs_unixify_accmode(accmode_t *accmode)
4664 {
4665 	/*
4666 	 * There is no way to specify explicit "deny" rule using
4667 	 * file mode or POSIX.1e ACLs.
4668 	 */
4669 	if (*accmode & VEXPLICIT_DENY) {
4670 		*accmode = 0;
4671 		return (0);
4672 	}
4673 
4674 	/*
4675 	 * None of these can be translated into usual access bits.
4676 	 * Also, the common case for NFSv4 ACLs is to not contain
4677 	 * either of these bits. Caller should check for VWRITE
4678 	 * on the containing directory instead.
4679 	 */
4680 	if (*accmode & (VDELETE_CHILD | VDELETE))
4681 		return (EPERM);
4682 
4683 	if (*accmode & VADMIN_PERMS) {
4684 		*accmode &= ~VADMIN_PERMS;
4685 		*accmode |= VADMIN;
4686 	}
4687 
4688 	/*
4689 	 * There is no way to deny VREAD_ATTRIBUTES, VREAD_ACL
4690 	 * or VSYNCHRONIZE using file mode or POSIX.1e ACL.
4691 	 */
4692 	*accmode &= ~(VSTAT_PERMS | VSYNCHRONIZE);
4693 
4694 	return (0);
4695 }
4696 
4697 /*
4698  * These are helper functions for filesystems to traverse all
4699  * their vnodes.  See MNT_VNODE_FOREACH_ALL() in sys/mount.h.
4700  *
4701  * This interface replaces MNT_VNODE_FOREACH.
4702  */
4703 
4704 MALLOC_DEFINE(M_VNODE_MARKER, "vnodemarker", "vnode marker");
4705 
4706 struct vnode *
4707 __mnt_vnode_next_all(struct vnode **mvp, struct mount *mp)
4708 {
4709 	struct vnode *vp;
4710 
4711 	if (should_yield())
4712 		kern_yield(PRI_USER);
4713 	MNT_ILOCK(mp);
4714 	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4715 	vp = TAILQ_NEXT(*mvp, v_nmntvnodes);
4716 	while (vp != NULL && (vp->v_type == VMARKER ||
4717 	    (vp->v_iflag & VI_DOOMED) != 0))
4718 		vp = TAILQ_NEXT(vp, v_nmntvnodes);
4719 
4720 	/* Check if we are done */
4721 	if (vp == NULL) {
4722 		__mnt_vnode_markerfree_all(mvp, mp);
4723 		/* MNT_IUNLOCK(mp); -- done in above function */
4724 		mtx_assert(MNT_MTX(mp), MA_NOTOWNED);
4725 		return (NULL);
4726 	}
4727 	TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
4728 	TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
4729 	VI_LOCK(vp);
4730 	MNT_IUNLOCK(mp);
4731 	return (vp);
4732 }
4733 
4734 struct vnode *
4735 __mnt_vnode_first_all(struct vnode **mvp, struct mount *mp)
4736 {
4737 	struct vnode *vp;
4738 
4739 	*mvp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO);
4740 	MNT_ILOCK(mp);
4741 	MNT_REF(mp);
4742 	(*mvp)->v_type = VMARKER;
4743 
4744 	vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
4745 	while (vp != NULL && (vp->v_type == VMARKER ||
4746 	    (vp->v_iflag & VI_DOOMED) != 0))
4747 		vp = TAILQ_NEXT(vp, v_nmntvnodes);
4748 
4749 	/* Check if we are done */
4750 	if (vp == NULL) {
4751 		MNT_REL(mp);
4752 		MNT_IUNLOCK(mp);
4753 		free(*mvp, M_VNODE_MARKER);
4754 		*mvp = NULL;
4755 		return (NULL);
4756 	}
4757 	(*mvp)->v_mount = mp;
4758 	TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
4759 	VI_LOCK(vp);
4760 	MNT_IUNLOCK(mp);
4761 	return (vp);
4762 }
4763 
4764 
4765 void
4766 __mnt_vnode_markerfree_all(struct vnode **mvp, struct mount *mp)
4767 {
4768 
4769 	if (*mvp == NULL) {
4770 		MNT_IUNLOCK(mp);
4771 		return;
4772 	}
4773 
4774 	mtx_assert(MNT_MTX(mp), MA_OWNED);
4775 
4776 	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4777 	TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
4778 	MNT_REL(mp);
4779 	MNT_IUNLOCK(mp);
4780 	free(*mvp, M_VNODE_MARKER);
4781 	*mvp = NULL;
4782 }
4783 
4784 /*
4785  * These are helper functions for filesystems to traverse their
4786  * active vnodes.  See MNT_VNODE_FOREACH_ACTIVE() in sys/mount.h
4787  */
4788 static void
4789 mnt_vnode_markerfree_active(struct vnode **mvp, struct mount *mp)
4790 {
4791 
4792 	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4793 
4794 	MNT_ILOCK(mp);
4795 	MNT_REL(mp);
4796 	MNT_IUNLOCK(mp);
4797 	free(*mvp, M_VNODE_MARKER);
4798 	*mvp = NULL;
4799 }
4800 
4801 static struct vnode *
4802 mnt_vnode_next_active(struct vnode **mvp, struct mount *mp)
4803 {
4804 	struct vnode *vp, *nvp;
4805 
4806 	mtx_assert(&vnode_free_list_mtx, MA_OWNED);
4807 	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4808 restart:
4809 	vp = TAILQ_NEXT(*mvp, v_actfreelist);
4810 	TAILQ_REMOVE(&mp->mnt_activevnodelist, *mvp, v_actfreelist);
4811 	while (vp != NULL) {
4812 		if (vp->v_type == VMARKER) {
4813 			vp = TAILQ_NEXT(vp, v_actfreelist);
4814 			continue;
4815 		}
4816 		if (!VI_TRYLOCK(vp)) {
4817 			if (mp_ncpus == 1 || should_yield()) {
4818 				TAILQ_INSERT_BEFORE(vp, *mvp, v_actfreelist);
4819 				mtx_unlock(&vnode_free_list_mtx);
4820 				pause("vnacti", 1);
4821 				mtx_lock(&vnode_free_list_mtx);
4822 				goto restart;
4823 			}
4824 			continue;
4825 		}
4826 		KASSERT(vp->v_type != VMARKER, ("locked marker %p", vp));
4827 		KASSERT(vp->v_mount == mp || vp->v_mount == NULL,
4828 		    ("alien vnode on the active list %p %p", vp, mp));
4829 		if (vp->v_mount == mp && (vp->v_iflag & VI_DOOMED) == 0)
4830 			break;
4831 		nvp = TAILQ_NEXT(vp, v_actfreelist);
4832 		VI_UNLOCK(vp);
4833 		vp = nvp;
4834 	}
4835 
4836 	/* Check if we are done */
4837 	if (vp == NULL) {
4838 		mtx_unlock(&vnode_free_list_mtx);
4839 		mnt_vnode_markerfree_active(mvp, mp);
4840 		return (NULL);
4841 	}
4842 	TAILQ_INSERT_AFTER(&mp->mnt_activevnodelist, vp, *mvp, v_actfreelist);
4843 	mtx_unlock(&vnode_free_list_mtx);
4844 	ASSERT_VI_LOCKED(vp, "active iter");
4845 	KASSERT((vp->v_iflag & VI_ACTIVE) != 0, ("Non-active vp %p", vp));
4846 	return (vp);
4847 }
4848 
4849 struct vnode *
4850 __mnt_vnode_next_active(struct vnode **mvp, struct mount *mp)
4851 {
4852 
4853 	if (should_yield())
4854 		kern_yield(PRI_USER);
4855 	mtx_lock(&vnode_free_list_mtx);
4856 	return (mnt_vnode_next_active(mvp, mp));
4857 }
4858 
4859 struct vnode *
4860 __mnt_vnode_first_active(struct vnode **mvp, struct mount *mp)
4861 {
4862 	struct vnode *vp;
4863 
4864 	*mvp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO);
4865 	MNT_ILOCK(mp);
4866 	MNT_REF(mp);
4867 	MNT_IUNLOCK(mp);
4868 	(*mvp)->v_type = VMARKER;
4869 	(*mvp)->v_mount = mp;
4870 
4871 	mtx_lock(&vnode_free_list_mtx);
4872 	vp = TAILQ_FIRST(&mp->mnt_activevnodelist);
4873 	if (vp == NULL) {
4874 		mtx_unlock(&vnode_free_list_mtx);
4875 		mnt_vnode_markerfree_active(mvp, mp);
4876 		return (NULL);
4877 	}
4878 	TAILQ_INSERT_BEFORE(vp, *mvp, v_actfreelist);
4879 	return (mnt_vnode_next_active(mvp, mp));
4880 }
4881 
4882 void
4883 __mnt_vnode_markerfree_active(struct vnode **mvp, struct mount *mp)
4884 {
4885 
4886 	if (*mvp == NULL)
4887 		return;
4888 
4889 	mtx_lock(&vnode_free_list_mtx);
4890 	TAILQ_REMOVE(&mp->mnt_activevnodelist, *mvp, v_actfreelist);
4891 	mtx_unlock(&vnode_free_list_mtx);
4892 	mnt_vnode_markerfree_active(mvp, mp);
4893 }
4894