xref: /freebsd/sys/kern/vfs_subr.c (revision 0bc7cf6fde4083c1a0cdf29b26075699028909e0)
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_ddb.h"
45 #include "opt_watchdog.h"
46 
47 #include <sys/param.h>
48 #include <sys/systm.h>
49 #include <sys/bio.h>
50 #include <sys/buf.h>
51 #include <sys/condvar.h>
52 #include <sys/conf.h>
53 #include <sys/dirent.h>
54 #include <sys/event.h>
55 #include <sys/eventhandler.h>
56 #include <sys/extattr.h>
57 #include <sys/file.h>
58 #include <sys/fcntl.h>
59 #include <sys/jail.h>
60 #include <sys/kdb.h>
61 #include <sys/kernel.h>
62 #include <sys/kthread.h>
63 #include <sys/lockf.h>
64 #include <sys/malloc.h>
65 #include <sys/mount.h>
66 #include <sys/namei.h>
67 #include <sys/priv.h>
68 #include <sys/reboot.h>
69 #include <sys/sched.h>
70 #include <sys/sleepqueue.h>
71 #include <sys/stat.h>
72 #include <sys/sysctl.h>
73 #include <sys/syslog.h>
74 #include <sys/vmmeter.h>
75 #include <sys/vnode.h>
76 #ifdef SW_WATCHDOG
77 #include <sys/watchdog.h>
78 #endif
79 
80 #include <machine/stdarg.h>
81 
82 #include <security/mac/mac_framework.h>
83 
84 #include <vm/vm.h>
85 #include <vm/vm_object.h>
86 #include <vm/vm_extern.h>
87 #include <vm/pmap.h>
88 #include <vm/vm_map.h>
89 #include <vm/vm_page.h>
90 #include <vm/vm_kern.h>
91 #include <vm/uma.h>
92 
93 #ifdef DDB
94 #include <ddb/ddb.h>
95 #endif
96 
97 #define	WI_MPSAFEQ	0
98 #define	WI_GIANTQ	1
99 
100 static void	delmntque(struct vnode *vp);
101 static int	flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo,
102 		    int slpflag, int slptimeo);
103 static void	syncer_shutdown(void *arg, int howto);
104 static int	vtryrecycle(struct vnode *vp);
105 static void	vbusy(struct vnode *vp);
106 static void	vinactive(struct vnode *, struct thread *);
107 static void	v_incr_usecount(struct vnode *);
108 static void	v_decr_usecount(struct vnode *);
109 static void	v_decr_useonly(struct vnode *);
110 static void	v_upgrade_usecount(struct vnode *);
111 static void	vfree(struct vnode *);
112 static void	vnlru_free(int);
113 static void	vgonel(struct vnode *);
114 static void	vfs_knllock(void *arg);
115 static void	vfs_knlunlock(void *arg);
116 static void	vfs_knl_assert_locked(void *arg);
117 static void	vfs_knl_assert_unlocked(void *arg);
118 static void	destroy_vpollinfo(struct vpollinfo *vi);
119 
120 /*
121  * Number of vnodes in existence.  Increased whenever getnewvnode()
122  * allocates a new vnode, decreased on vdestroy() called on VI_DOOMed
123  * vnode.
124  */
125 static unsigned long	numvnodes;
126 
127 SYSCTL_ULONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0,
128     "Number of vnodes in existence");
129 
130 /*
131  * Conversion tables for conversion from vnode types to inode formats
132  * and back.
133  */
134 enum vtype iftovt_tab[16] = {
135 	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
136 	VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
137 };
138 int vttoif_tab[10] = {
139 	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
140 	S_IFSOCK, S_IFIFO, S_IFMT, S_IFMT
141 };
142 
143 /*
144  * List of vnodes that are ready for recycling.
145  */
146 static TAILQ_HEAD(freelst, vnode) vnode_free_list;
147 
148 /*
149  * Free vnode target.  Free vnodes may simply be files which have been stat'd
150  * but not read.  This is somewhat common, and a small cache of such files
151  * should be kept to avoid recreation costs.
152  */
153 static u_long wantfreevnodes;
154 SYSCTL_ULONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
155 /* Number of vnodes in the free list. */
156 static u_long freevnodes;
157 SYSCTL_ULONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0,
158     "Number of vnodes in the free list");
159 
160 static int vlru_allow_cache_src;
161 SYSCTL_INT(_vfs, OID_AUTO, vlru_allow_cache_src, CTLFLAG_RW,
162     &vlru_allow_cache_src, 0, "Allow vlru to reclaim source vnode");
163 
164 /*
165  * Various variables used for debugging the new implementation of
166  * reassignbuf().
167  * XXX these are probably of (very) limited utility now.
168  */
169 static int reassignbufcalls;
170 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0,
171     "Number of calls to reassignbuf");
172 
173 /*
174  * Cache for the mount type id assigned to NFS.  This is used for
175  * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c.
176  */
177 int	nfs_mount_type = -1;
178 
179 /* To keep more than one thread at a time from running vfs_getnewfsid */
180 static struct mtx mntid_mtx;
181 
182 /*
183  * Lock for any access to the following:
184  *	vnode_free_list
185  *	numvnodes
186  *	freevnodes
187  */
188 static struct mtx vnode_free_list_mtx;
189 
190 /* Publicly exported FS */
191 struct nfs_public nfs_pub;
192 
193 /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
194 static uma_zone_t vnode_zone;
195 static uma_zone_t vnodepoll_zone;
196 
197 /*
198  * The workitem queue.
199  *
200  * It is useful to delay writes of file data and filesystem metadata
201  * for tens of seconds so that quickly created and deleted files need
202  * not waste disk bandwidth being created and removed. To realize this,
203  * we append vnodes to a "workitem" queue. When running with a soft
204  * updates implementation, most pending metadata dependencies should
205  * not wait for more than a few seconds. Thus, mounted on block devices
206  * are delayed only about a half the time that file data is delayed.
207  * Similarly, directory updates are more critical, so are only delayed
208  * about a third the time that file data is delayed. Thus, there are
209  * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
210  * one each second (driven off the filesystem syncer process). The
211  * syncer_delayno variable indicates the next queue that is to be processed.
212  * Items that need to be processed soon are placed in this queue:
213  *
214  *	syncer_workitem_pending[syncer_delayno]
215  *
216  * A delay of fifteen seconds is done by placing the request fifteen
217  * entries later in the queue:
218  *
219  *	syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
220  *
221  */
222 static int syncer_delayno;
223 static long syncer_mask;
224 LIST_HEAD(synclist, bufobj);
225 static struct synclist *syncer_workitem_pending[2];
226 /*
227  * The sync_mtx protects:
228  *	bo->bo_synclist
229  *	sync_vnode_count
230  *	syncer_delayno
231  *	syncer_state
232  *	syncer_workitem_pending
233  *	syncer_worklist_len
234  *	rushjob
235  */
236 static struct mtx sync_mtx;
237 static struct cv sync_wakeup;
238 
239 #define SYNCER_MAXDELAY		32
240 static int syncer_maxdelay = SYNCER_MAXDELAY;	/* maximum delay time */
241 static int syncdelay = 30;		/* max time to delay syncing data */
242 static int filedelay = 30;		/* time to delay syncing files */
243 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0,
244     "Time to delay syncing files (in seconds)");
245 static int dirdelay = 29;		/* time to delay syncing directories */
246 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0,
247     "Time to delay syncing directories (in seconds)");
248 static int metadelay = 28;		/* time to delay syncing metadata */
249 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0,
250     "Time to delay syncing metadata (in seconds)");
251 static int rushjob;		/* number of slots to run ASAP */
252 static int stat_rush_requests;	/* number of times I/O speeded up */
253 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0,
254     "Number of times I/O speeded up (rush requests)");
255 
256 /*
257  * When shutting down the syncer, run it at four times normal speed.
258  */
259 #define SYNCER_SHUTDOWN_SPEEDUP		4
260 static int sync_vnode_count;
261 static int syncer_worklist_len;
262 static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY }
263     syncer_state;
264 
265 /*
266  * Number of vnodes we want to exist at any one time.  This is mostly used
267  * to size hash tables in vnode-related code.  It is normally not used in
268  * getnewvnode(), as wantfreevnodes is normally nonzero.)
269  *
270  * XXX desiredvnodes is historical cruft and should not exist.
271  */
272 int desiredvnodes;
273 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
274     &desiredvnodes, 0, "Maximum number of vnodes");
275 SYSCTL_ULONG(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
276     &wantfreevnodes, 0, "Minimum number of vnodes (legacy)");
277 static int vnlru_nowhere;
278 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW,
279     &vnlru_nowhere, 0, "Number of times the vnlru process ran without success");
280 
281 /*
282  * Macros to control when a vnode is freed and recycled.  All require
283  * the vnode interlock.
284  */
285 #define VCANRECYCLE(vp) (((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
286 #define VSHOULDFREE(vp) (!((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
287 #define VSHOULDBUSY(vp) (((vp)->v_iflag & VI_FREE) && (vp)->v_holdcnt)
288 
289 
290 /*
291  * Initialize the vnode management data structures.
292  *
293  * Reevaluate the following cap on the number of vnodes after the physical
294  * memory size exceeds 512GB.  In the limit, as the physical memory size
295  * grows, the ratio of physical pages to vnodes approaches sixteen to one.
296  */
297 #ifndef	MAXVNODES_MAX
298 #define	MAXVNODES_MAX	(512 * (1024 * 1024 * 1024 / (int)PAGE_SIZE / 16))
299 #endif
300 static void
301 vntblinit(void *dummy __unused)
302 {
303 	int physvnodes, virtvnodes;
304 
305 	/*
306 	 * Desiredvnodes is a function of the physical memory size and the
307 	 * kernel's heap size.  Generally speaking, it scales with the
308 	 * physical memory size.  The ratio of desiredvnodes to physical pages
309 	 * is one to four until desiredvnodes exceeds 98,304.  Thereafter, the
310 	 * marginal ratio of desiredvnodes to physical pages is one to
311 	 * sixteen.  However, desiredvnodes is limited by the kernel's heap
312 	 * size.  The memory required by desiredvnodes vnodes and vm objects
313 	 * may not exceed one seventh of the kernel's heap size.
314 	 */
315 	physvnodes = maxproc + cnt.v_page_count / 16 + 3 * min(98304 * 4,
316 	    cnt.v_page_count) / 16;
317 	virtvnodes = vm_kmem_size / (7 * (sizeof(struct vm_object) +
318 	    sizeof(struct vnode)));
319 	desiredvnodes = min(physvnodes, virtvnodes);
320 	if (desiredvnodes > MAXVNODES_MAX) {
321 		if (bootverbose)
322 			printf("Reducing kern.maxvnodes %d -> %d\n",
323 			    desiredvnodes, MAXVNODES_MAX);
324 		desiredvnodes = MAXVNODES_MAX;
325 	}
326 	wantfreevnodes = desiredvnodes / 4;
327 	mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF);
328 	TAILQ_INIT(&vnode_free_list);
329 	mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF);
330 	vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL,
331 	    NULL, NULL, UMA_ALIGN_PTR, 0);
332 	vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo),
333 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
334 	/*
335 	 * Initialize the filesystem syncer.
336 	 */
337 	syncer_workitem_pending[WI_MPSAFEQ] = hashinit(syncer_maxdelay, M_VNODE,
338 	    &syncer_mask);
339 	syncer_workitem_pending[WI_GIANTQ] = hashinit(syncer_maxdelay, M_VNODE,
340 	    &syncer_mask);
341 	syncer_maxdelay = syncer_mask + 1;
342 	mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF);
343 	cv_init(&sync_wakeup, "syncer");
344 }
345 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL);
346 
347 
348 /*
349  * Mark a mount point as busy. Used to synchronize access and to delay
350  * unmounting. Eventually, mountlist_mtx is not released on failure.
351  *
352  * vfs_busy() is a custom lock, it can block the caller.
353  * vfs_busy() only sleeps if the unmount is active on the mount point.
354  * For a mountpoint mp, vfs_busy-enforced lock is before lock of any
355  * vnode belonging to mp.
356  *
357  * Lookup uses vfs_busy() to traverse mount points.
358  * root fs			var fs
359  * / vnode lock		A	/ vnode lock (/var)		D
360  * /var vnode lock	B	/log vnode lock(/var/log)	E
361  * vfs_busy lock	C	vfs_busy lock			F
362  *
363  * Within each file system, the lock order is C->A->B and F->D->E.
364  *
365  * When traversing across mounts, the system follows that lock order:
366  *
367  *        C->A->B
368  *              |
369  *              +->F->D->E
370  *
371  * The lookup() process for namei("/var") illustrates the process:
372  *  VOP_LOOKUP() obtains B while A is held
373  *  vfs_busy() obtains a shared lock on F while A and B are held
374  *  vput() releases lock on B
375  *  vput() releases lock on A
376  *  VFS_ROOT() obtains lock on D while shared lock on F is held
377  *  vfs_unbusy() releases shared lock on F
378  *  vn_lock() obtains lock on deadfs vnode vp_crossmp instead of A.
379  *    Attempt to lock A (instead of vp_crossmp) while D is held would
380  *    violate the global order, causing deadlocks.
381  *
382  * dounmount() locks B while F is drained.
383  */
384 int
385 vfs_busy(struct mount *mp, int flags)
386 {
387 
388 	MPASS((flags & ~MBF_MASK) == 0);
389 	CTR3(KTR_VFS, "%s: mp %p with flags %d", __func__, mp, flags);
390 
391 	MNT_ILOCK(mp);
392 	MNT_REF(mp);
393 	/*
394 	 * If mount point is currenly being unmounted, sleep until the
395 	 * mount point fate is decided.  If thread doing the unmounting fails,
396 	 * it will clear MNTK_UNMOUNT flag before waking us up, indicating
397 	 * that this mount point has survived the unmount attempt and vfs_busy
398 	 * should retry.  Otherwise the unmounter thread will set MNTK_REFEXPIRE
399 	 * flag in addition to MNTK_UNMOUNT, indicating that mount point is
400 	 * about to be really destroyed.  vfs_busy needs to release its
401 	 * reference on the mount point in this case and return with ENOENT,
402 	 * telling the caller that mount mount it tried to busy is no longer
403 	 * valid.
404 	 */
405 	while (mp->mnt_kern_flag & MNTK_UNMOUNT) {
406 		if (flags & MBF_NOWAIT || mp->mnt_kern_flag & MNTK_REFEXPIRE) {
407 			MNT_REL(mp);
408 			MNT_IUNLOCK(mp);
409 			CTR1(KTR_VFS, "%s: failed busying before sleeping",
410 			    __func__);
411 			return (ENOENT);
412 		}
413 		if (flags & MBF_MNTLSTLOCK)
414 			mtx_unlock(&mountlist_mtx);
415 		mp->mnt_kern_flag |= MNTK_MWAIT;
416 		msleep(mp, MNT_MTX(mp), PVFS | PDROP, "vfs_busy", 0);
417 		if (flags & MBF_MNTLSTLOCK)
418 			mtx_lock(&mountlist_mtx);
419 		MNT_ILOCK(mp);
420 	}
421 	if (flags & MBF_MNTLSTLOCK)
422 		mtx_unlock(&mountlist_mtx);
423 	mp->mnt_lockref++;
424 	MNT_IUNLOCK(mp);
425 	return (0);
426 }
427 
428 /*
429  * Free a busy filesystem.
430  */
431 void
432 vfs_unbusy(struct mount *mp)
433 {
434 
435 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
436 	MNT_ILOCK(mp);
437 	MNT_REL(mp);
438 	KASSERT(mp->mnt_lockref > 0, ("negative mnt_lockref"));
439 	mp->mnt_lockref--;
440 	if (mp->mnt_lockref == 0 && (mp->mnt_kern_flag & MNTK_DRAINING) != 0) {
441 		MPASS(mp->mnt_kern_flag & MNTK_UNMOUNT);
442 		CTR1(KTR_VFS, "%s: waking up waiters", __func__);
443 		mp->mnt_kern_flag &= ~MNTK_DRAINING;
444 		wakeup(&mp->mnt_lockref);
445 	}
446 	MNT_IUNLOCK(mp);
447 }
448 
449 /*
450  * Lookup a mount point by filesystem identifier.
451  */
452 struct mount *
453 vfs_getvfs(fsid_t *fsid)
454 {
455 	struct mount *mp;
456 
457 	CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
458 	mtx_lock(&mountlist_mtx);
459 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
460 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
461 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
462 			vfs_ref(mp);
463 			mtx_unlock(&mountlist_mtx);
464 			return (mp);
465 		}
466 	}
467 	mtx_unlock(&mountlist_mtx);
468 	CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
469 	return ((struct mount *) 0);
470 }
471 
472 /*
473  * Lookup a mount point by filesystem identifier, busying it before
474  * returning.
475  */
476 struct mount *
477 vfs_busyfs(fsid_t *fsid)
478 {
479 	struct mount *mp;
480 	int error;
481 
482 	CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
483 	mtx_lock(&mountlist_mtx);
484 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
485 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
486 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
487 			error = vfs_busy(mp, MBF_MNTLSTLOCK);
488 			if (error) {
489 				mtx_unlock(&mountlist_mtx);
490 				return (NULL);
491 			}
492 			return (mp);
493 		}
494 	}
495 	CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
496 	mtx_unlock(&mountlist_mtx);
497 	return ((struct mount *) 0);
498 }
499 
500 /*
501  * Check if a user can access privileged mount options.
502  */
503 int
504 vfs_suser(struct mount *mp, struct thread *td)
505 {
506 	int error;
507 
508 	/*
509 	 * If the thread is jailed, but this is not a jail-friendly file
510 	 * system, deny immediately.
511 	 */
512 	if (!(mp->mnt_vfc->vfc_flags & VFCF_JAIL) && jailed(td->td_ucred))
513 		return (EPERM);
514 
515 	/*
516 	 * If the file system was mounted outside the jail of the calling
517 	 * thread, deny immediately.
518 	 */
519 	if (prison_check(td->td_ucred, mp->mnt_cred) != 0)
520 		return (EPERM);
521 
522 	/*
523 	 * If file system supports delegated administration, we don't check
524 	 * for the PRIV_VFS_MOUNT_OWNER privilege - it will be better verified
525 	 * by the file system itself.
526 	 * If this is not the user that did original mount, we check for
527 	 * the PRIV_VFS_MOUNT_OWNER privilege.
528 	 */
529 	if (!(mp->mnt_vfc->vfc_flags & VFCF_DELEGADMIN) &&
530 	    mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) {
531 		if ((error = priv_check(td, PRIV_VFS_MOUNT_OWNER)) != 0)
532 			return (error);
533 	}
534 	return (0);
535 }
536 
537 /*
538  * Get a new unique fsid.  Try to make its val[0] unique, since this value
539  * will be used to create fake device numbers for stat().  Also try (but
540  * not so hard) make its val[0] unique mod 2^16, since some emulators only
541  * support 16-bit device numbers.  We end up with unique val[0]'s for the
542  * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
543  *
544  * Keep in mind that several mounts may be running in parallel.  Starting
545  * the search one past where the previous search terminated is both a
546  * micro-optimization and a defense against returning the same fsid to
547  * different mounts.
548  */
549 void
550 vfs_getnewfsid(struct mount *mp)
551 {
552 	static uint16_t mntid_base;
553 	struct mount *nmp;
554 	fsid_t tfsid;
555 	int mtype;
556 
557 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
558 	mtx_lock(&mntid_mtx);
559 	mtype = mp->mnt_vfc->vfc_typenum;
560 	tfsid.val[1] = mtype;
561 	mtype = (mtype & 0xFF) << 24;
562 	for (;;) {
563 		tfsid.val[0] = makedev(255,
564 		    mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
565 		mntid_base++;
566 		if ((nmp = vfs_getvfs(&tfsid)) == NULL)
567 			break;
568 		vfs_rel(nmp);
569 	}
570 	mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
571 	mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
572 	mtx_unlock(&mntid_mtx);
573 }
574 
575 /*
576  * Knob to control the precision of file timestamps:
577  *
578  *   0 = seconds only; nanoseconds zeroed.
579  *   1 = seconds and nanoseconds, accurate within 1/HZ.
580  *   2 = seconds and nanoseconds, truncated to microseconds.
581  * >=3 = seconds and nanoseconds, maximum precision.
582  */
583 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
584 
585 static int timestamp_precision = TSP_SEC;
586 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
587     &timestamp_precision, 0, "File timestamp precision (0: seconds, "
588     "1: sec + ns accurate to 1/HZ, 2: sec + ns truncated to ms, "
589     "3+: sec + ns (max. precision))");
590 
591 /*
592  * Get a current timestamp.
593  */
594 void
595 vfs_timestamp(struct timespec *tsp)
596 {
597 	struct timeval tv;
598 
599 	switch (timestamp_precision) {
600 	case TSP_SEC:
601 		tsp->tv_sec = time_second;
602 		tsp->tv_nsec = 0;
603 		break;
604 	case TSP_HZ:
605 		getnanotime(tsp);
606 		break;
607 	case TSP_USEC:
608 		microtime(&tv);
609 		TIMEVAL_TO_TIMESPEC(&tv, tsp);
610 		break;
611 	case TSP_NSEC:
612 	default:
613 		nanotime(tsp);
614 		break;
615 	}
616 }
617 
618 /*
619  * Set vnode attributes to VNOVAL
620  */
621 void
622 vattr_null(struct vattr *vap)
623 {
624 
625 	vap->va_type = VNON;
626 	vap->va_size = VNOVAL;
627 	vap->va_bytes = VNOVAL;
628 	vap->va_mode = VNOVAL;
629 	vap->va_nlink = VNOVAL;
630 	vap->va_uid = VNOVAL;
631 	vap->va_gid = VNOVAL;
632 	vap->va_fsid = VNOVAL;
633 	vap->va_fileid = VNOVAL;
634 	vap->va_blocksize = VNOVAL;
635 	vap->va_rdev = VNOVAL;
636 	vap->va_atime.tv_sec = VNOVAL;
637 	vap->va_atime.tv_nsec = VNOVAL;
638 	vap->va_mtime.tv_sec = VNOVAL;
639 	vap->va_mtime.tv_nsec = VNOVAL;
640 	vap->va_ctime.tv_sec = VNOVAL;
641 	vap->va_ctime.tv_nsec = VNOVAL;
642 	vap->va_birthtime.tv_sec = VNOVAL;
643 	vap->va_birthtime.tv_nsec = VNOVAL;
644 	vap->va_flags = VNOVAL;
645 	vap->va_gen = VNOVAL;
646 	vap->va_vaflags = 0;
647 }
648 
649 /*
650  * This routine is called when we have too many vnodes.  It attempts
651  * to free <count> vnodes and will potentially free vnodes that still
652  * have VM backing store (VM backing store is typically the cause
653  * of a vnode blowout so we want to do this).  Therefore, this operation
654  * is not considered cheap.
655  *
656  * A number of conditions may prevent a vnode from being reclaimed.
657  * the buffer cache may have references on the vnode, a directory
658  * vnode may still have references due to the namei cache representing
659  * underlying files, or the vnode may be in active use.   It is not
660  * desireable to reuse such vnodes.  These conditions may cause the
661  * number of vnodes to reach some minimum value regardless of what
662  * you set kern.maxvnodes to.  Do not set kern.maxvnodes too low.
663  */
664 static int
665 vlrureclaim(struct mount *mp)
666 {
667 	struct vnode *vp;
668 	int done;
669 	int trigger;
670 	int usevnodes;
671 	int count;
672 
673 	/*
674 	 * Calculate the trigger point, don't allow user
675 	 * screwups to blow us up.   This prevents us from
676 	 * recycling vnodes with lots of resident pages.  We
677 	 * aren't trying to free memory, we are trying to
678 	 * free vnodes.
679 	 */
680 	usevnodes = desiredvnodes;
681 	if (usevnodes <= 0)
682 		usevnodes = 1;
683 	trigger = cnt.v_page_count * 2 / usevnodes;
684 	done = 0;
685 	vn_start_write(NULL, &mp, V_WAIT);
686 	MNT_ILOCK(mp);
687 	count = mp->mnt_nvnodelistsize / 10 + 1;
688 	while (count != 0) {
689 		vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
690 		while (vp != NULL && vp->v_type == VMARKER)
691 			vp = TAILQ_NEXT(vp, v_nmntvnodes);
692 		if (vp == NULL)
693 			break;
694 		TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
695 		TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
696 		--count;
697 		if (!VI_TRYLOCK(vp))
698 			goto next_iter;
699 		/*
700 		 * If it's been deconstructed already, it's still
701 		 * referenced, or it exceeds the trigger, skip it.
702 		 */
703 		if (vp->v_usecount ||
704 		    (!vlru_allow_cache_src &&
705 			!LIST_EMPTY(&(vp)->v_cache_src)) ||
706 		    (vp->v_iflag & VI_DOOMED) != 0 || (vp->v_object != NULL &&
707 		    vp->v_object->resident_page_count > trigger)) {
708 			VI_UNLOCK(vp);
709 			goto next_iter;
710 		}
711 		MNT_IUNLOCK(mp);
712 		vholdl(vp);
713 		if (VOP_LOCK(vp, LK_INTERLOCK|LK_EXCLUSIVE|LK_NOWAIT)) {
714 			vdrop(vp);
715 			goto next_iter_mntunlocked;
716 		}
717 		VI_LOCK(vp);
718 		/*
719 		 * v_usecount may have been bumped after VOP_LOCK() dropped
720 		 * the vnode interlock and before it was locked again.
721 		 *
722 		 * It is not necessary to recheck VI_DOOMED because it can
723 		 * only be set by another thread that holds both the vnode
724 		 * lock and vnode interlock.  If another thread has the
725 		 * vnode lock before we get to VOP_LOCK() and obtains the
726 		 * vnode interlock after VOP_LOCK() drops the vnode
727 		 * interlock, the other thread will be unable to drop the
728 		 * vnode lock before our VOP_LOCK() call fails.
729 		 */
730 		if (vp->v_usecount ||
731 		    (!vlru_allow_cache_src &&
732 			!LIST_EMPTY(&(vp)->v_cache_src)) ||
733 		    (vp->v_object != NULL &&
734 		    vp->v_object->resident_page_count > trigger)) {
735 			VOP_UNLOCK(vp, LK_INTERLOCK);
736 			goto next_iter_mntunlocked;
737 		}
738 		KASSERT((vp->v_iflag & VI_DOOMED) == 0,
739 		    ("VI_DOOMED unexpectedly detected in vlrureclaim()"));
740 		vgonel(vp);
741 		VOP_UNLOCK(vp, 0);
742 		vdropl(vp);
743 		done++;
744 next_iter_mntunlocked:
745 		if (!should_yield())
746 			goto relock_mnt;
747 		goto yield;
748 next_iter:
749 		if (!should_yield())
750 			continue;
751 		MNT_IUNLOCK(mp);
752 yield:
753 		kern_yield(PRI_UNCHANGED);
754 relock_mnt:
755 		MNT_ILOCK(mp);
756 	}
757 	MNT_IUNLOCK(mp);
758 	vn_finished_write(mp);
759 	return done;
760 }
761 
762 /*
763  * Attempt to keep the free list at wantfreevnodes length.
764  */
765 static void
766 vnlru_free(int count)
767 {
768 	struct vnode *vp;
769 	int vfslocked;
770 
771 	mtx_assert(&vnode_free_list_mtx, MA_OWNED);
772 	for (; count > 0; count--) {
773 		vp = TAILQ_FIRST(&vnode_free_list);
774 		/*
775 		 * The list can be modified while the free_list_mtx
776 		 * has been dropped and vp could be NULL here.
777 		 */
778 		if (!vp)
779 			break;
780 		VNASSERT(vp->v_op != NULL, vp,
781 		    ("vnlru_free: vnode already reclaimed."));
782 		TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
783 		/*
784 		 * Don't recycle if we can't get the interlock.
785 		 */
786 		if (!VI_TRYLOCK(vp)) {
787 			TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
788 			continue;
789 		}
790 		VNASSERT(VCANRECYCLE(vp), vp,
791 		    ("vp inconsistent on freelist"));
792 		freevnodes--;
793 		vp->v_iflag &= ~VI_FREE;
794 		vholdl(vp);
795 		mtx_unlock(&vnode_free_list_mtx);
796 		VI_UNLOCK(vp);
797 		vfslocked = VFS_LOCK_GIANT(vp->v_mount);
798 		vtryrecycle(vp);
799 		VFS_UNLOCK_GIANT(vfslocked);
800 		/*
801 		 * If the recycled succeeded this vdrop will actually free
802 		 * the vnode.  If not it will simply place it back on
803 		 * the free list.
804 		 */
805 		vdrop(vp);
806 		mtx_lock(&vnode_free_list_mtx);
807 	}
808 }
809 /*
810  * Attempt to recycle vnodes in a context that is always safe to block.
811  * Calling vlrurecycle() from the bowels of filesystem code has some
812  * interesting deadlock problems.
813  */
814 static struct proc *vnlruproc;
815 static int vnlruproc_sig;
816 
817 static void
818 vnlru_proc(void)
819 {
820 	struct mount *mp, *nmp;
821 	int done, vfslocked;
822 	struct proc *p = vnlruproc;
823 
824 	EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
825 	    SHUTDOWN_PRI_FIRST);
826 
827 	for (;;) {
828 		kproc_suspend_check(p);
829 		mtx_lock(&vnode_free_list_mtx);
830 		if (freevnodes > wantfreevnodes)
831 			vnlru_free(freevnodes - wantfreevnodes);
832 		if (numvnodes <= desiredvnodes * 9 / 10) {
833 			vnlruproc_sig = 0;
834 			wakeup(&vnlruproc_sig);
835 			msleep(vnlruproc, &vnode_free_list_mtx,
836 			    PVFS|PDROP, "vlruwt", hz);
837 			continue;
838 		}
839 		mtx_unlock(&vnode_free_list_mtx);
840 		done = 0;
841 		mtx_lock(&mountlist_mtx);
842 		for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
843 			if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) {
844 				nmp = TAILQ_NEXT(mp, mnt_list);
845 				continue;
846 			}
847 			vfslocked = VFS_LOCK_GIANT(mp);
848 			done += vlrureclaim(mp);
849 			VFS_UNLOCK_GIANT(vfslocked);
850 			mtx_lock(&mountlist_mtx);
851 			nmp = TAILQ_NEXT(mp, mnt_list);
852 			vfs_unbusy(mp);
853 		}
854 		mtx_unlock(&mountlist_mtx);
855 		if (done == 0) {
856 #if 0
857 			/* These messages are temporary debugging aids */
858 			if (vnlru_nowhere < 5)
859 				printf("vnlru process getting nowhere..\n");
860 			else if (vnlru_nowhere == 5)
861 				printf("vnlru process messages stopped.\n");
862 #endif
863 			vnlru_nowhere++;
864 			tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
865 		} else
866 			kern_yield(PRI_UNCHANGED);
867 	}
868 }
869 
870 static struct kproc_desc vnlru_kp = {
871 	"vnlru",
872 	vnlru_proc,
873 	&vnlruproc
874 };
875 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start,
876     &vnlru_kp);
877 
878 /*
879  * Routines having to do with the management of the vnode table.
880  */
881 
882 void
883 vdestroy(struct vnode *vp)
884 {
885 	struct bufobj *bo;
886 
887 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
888 	mtx_lock(&vnode_free_list_mtx);
889 	numvnodes--;
890 	mtx_unlock(&vnode_free_list_mtx);
891 	bo = &vp->v_bufobj;
892 	VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
893 	    ("cleaned vnode still on the free list."));
894 	VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't"));
895 	VNASSERT(vp->v_holdcnt == 0, vp, ("Non-zero hold count"));
896 	VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count"));
897 	VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count"));
898 	VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's"));
899 	VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0"));
900 	VNASSERT(bo->bo_clean.bv_root == NULL, vp, ("cleanblkroot not NULL"));
901 	VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0"));
902 	VNASSERT(bo->bo_dirty.bv_root == NULL, vp, ("dirtyblkroot not NULL"));
903 	VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst"));
904 	VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src"));
905 	VNASSERT(vp->v_cache_dd == NULL, vp, ("vp has namecache for .."));
906 	VI_UNLOCK(vp);
907 #ifdef MAC
908 	mac_vnode_destroy(vp);
909 #endif
910 	if (vp->v_pollinfo != NULL)
911 		destroy_vpollinfo(vp->v_pollinfo);
912 #ifdef INVARIANTS
913 	/* XXX Elsewhere we can detect an already freed vnode via NULL v_op. */
914 	vp->v_op = NULL;
915 #endif
916 	lockdestroy(vp->v_vnlock);
917 	mtx_destroy(&vp->v_interlock);
918 	mtx_destroy(BO_MTX(bo));
919 	uma_zfree(vnode_zone, vp);
920 }
921 
922 /*
923  * Try to recycle a freed vnode.  We abort if anyone picks up a reference
924  * before we actually vgone().  This function must be called with the vnode
925  * held to prevent the vnode from being returned to the free list midway
926  * through vgone().
927  */
928 static int
929 vtryrecycle(struct vnode *vp)
930 {
931 	struct mount *vnmp;
932 
933 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
934 	VNASSERT(vp->v_holdcnt, vp,
935 	    ("vtryrecycle: Recycling vp %p without a reference.", vp));
936 	/*
937 	 * This vnode may found and locked via some other list, if so we
938 	 * can't recycle it yet.
939 	 */
940 	if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
941 		CTR2(KTR_VFS,
942 		    "%s: impossible to recycle, vp %p lock is already held",
943 		    __func__, vp);
944 		return (EWOULDBLOCK);
945 	}
946 	/*
947 	 * Don't recycle if its filesystem is being suspended.
948 	 */
949 	if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) {
950 		VOP_UNLOCK(vp, 0);
951 		CTR2(KTR_VFS,
952 		    "%s: impossible to recycle, cannot start the write for %p",
953 		    __func__, vp);
954 		return (EBUSY);
955 	}
956 	/*
957 	 * If we got this far, we need to acquire the interlock and see if
958 	 * anyone picked up this vnode from another list.  If not, we will
959 	 * mark it with DOOMED via vgonel() so that anyone who does find it
960 	 * will skip over it.
961 	 */
962 	VI_LOCK(vp);
963 	if (vp->v_usecount) {
964 		VOP_UNLOCK(vp, LK_INTERLOCK);
965 		vn_finished_write(vnmp);
966 		CTR2(KTR_VFS,
967 		    "%s: impossible to recycle, %p is already referenced",
968 		    __func__, vp);
969 		return (EBUSY);
970 	}
971 	if ((vp->v_iflag & VI_DOOMED) == 0)
972 		vgonel(vp);
973 	VOP_UNLOCK(vp, LK_INTERLOCK);
974 	vn_finished_write(vnmp);
975 	return (0);
976 }
977 
978 /*
979  * Return the next vnode from the free list.
980  */
981 int
982 getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops,
983     struct vnode **vpp)
984 {
985 	struct vnode *vp = NULL;
986 	struct bufobj *bo;
987 
988 	CTR3(KTR_VFS, "%s: mp %p with tag %s", __func__, mp, tag);
989 	mtx_lock(&vnode_free_list_mtx);
990 	/*
991 	 * Lend our context to reclaim vnodes if they've exceeded the max.
992 	 */
993 	if (freevnodes > wantfreevnodes)
994 		vnlru_free(1);
995 	/*
996 	 * Wait for available vnodes.
997 	 */
998 	if (numvnodes > desiredvnodes) {
999 		if (mp != NULL && (mp->mnt_kern_flag & MNTK_SUSPEND)) {
1000 			/*
1001 			 * File system is beeing suspended, we cannot risk a
1002 			 * deadlock here, so allocate new vnode anyway.
1003 			 */
1004 			if (freevnodes > wantfreevnodes)
1005 				vnlru_free(freevnodes - wantfreevnodes);
1006 			goto alloc;
1007 		}
1008 		if (vnlruproc_sig == 0) {
1009 			vnlruproc_sig = 1;	/* avoid unnecessary wakeups */
1010 			wakeup(vnlruproc);
1011 		}
1012 		msleep(&vnlruproc_sig, &vnode_free_list_mtx, PVFS,
1013 		    "vlruwk", hz);
1014 #if 0	/* XXX Not all VFS_VGET/ffs_vget callers check returns. */
1015 		if (numvnodes > desiredvnodes) {
1016 			mtx_unlock(&vnode_free_list_mtx);
1017 			return (ENFILE);
1018 		}
1019 #endif
1020 	}
1021 alloc:
1022 	numvnodes++;
1023 	mtx_unlock(&vnode_free_list_mtx);
1024 	vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO);
1025 	/*
1026 	 * Setup locks.
1027 	 */
1028 	vp->v_vnlock = &vp->v_lock;
1029 	mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
1030 	/*
1031 	 * By default, don't allow shared locks unless filesystems
1032 	 * opt-in.
1033 	 */
1034 	lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOSHARE);
1035 	/*
1036 	 * Initialize bufobj.
1037 	 */
1038 	bo = &vp->v_bufobj;
1039 	bo->__bo_vnode = vp;
1040 	mtx_init(BO_MTX(bo), "bufobj interlock", NULL, MTX_DEF);
1041 	bo->bo_ops = &buf_ops_bio;
1042 	bo->bo_private = vp;
1043 	TAILQ_INIT(&bo->bo_clean.bv_hd);
1044 	TAILQ_INIT(&bo->bo_dirty.bv_hd);
1045 	/*
1046 	 * Initialize namecache.
1047 	 */
1048 	LIST_INIT(&vp->v_cache_src);
1049 	TAILQ_INIT(&vp->v_cache_dst);
1050 	/*
1051 	 * Finalize various vnode identity bits.
1052 	 */
1053 	vp->v_type = VNON;
1054 	vp->v_tag = tag;
1055 	vp->v_op = vops;
1056 	v_incr_usecount(vp);
1057 	vp->v_data = NULL;
1058 #ifdef MAC
1059 	mac_vnode_init(vp);
1060 	if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0)
1061 		mac_vnode_associate_singlelabel(mp, vp);
1062 	else if (mp == NULL && vops != &dead_vnodeops)
1063 		printf("NULL mp in getnewvnode()\n");
1064 #endif
1065 	if (mp != NULL) {
1066 		bo->bo_bsize = mp->mnt_stat.f_iosize;
1067 		if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0)
1068 			vp->v_vflag |= VV_NOKNOTE;
1069 	}
1070 
1071 	*vpp = vp;
1072 	return (0);
1073 }
1074 
1075 /*
1076  * Delete from old mount point vnode list, if on one.
1077  */
1078 static void
1079 delmntque(struct vnode *vp)
1080 {
1081 	struct mount *mp;
1082 
1083 	mp = vp->v_mount;
1084 	if (mp == NULL)
1085 		return;
1086 	MNT_ILOCK(mp);
1087 	vp->v_mount = NULL;
1088 	VNASSERT(mp->mnt_nvnodelistsize > 0, vp,
1089 		("bad mount point vnode list size"));
1090 	TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1091 	mp->mnt_nvnodelistsize--;
1092 	MNT_REL(mp);
1093 	MNT_IUNLOCK(mp);
1094 }
1095 
1096 static void
1097 insmntque_stddtr(struct vnode *vp, void *dtr_arg)
1098 {
1099 
1100 	vp->v_data = NULL;
1101 	vp->v_op = &dead_vnodeops;
1102 	/* XXX non mp-safe fs may still call insmntque with vnode
1103 	   unlocked */
1104 	if (!VOP_ISLOCKED(vp))
1105 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1106 	vgone(vp);
1107 	vput(vp);
1108 }
1109 
1110 /*
1111  * Insert into list of vnodes for the new mount point, if available.
1112  */
1113 int
1114 insmntque1(struct vnode *vp, struct mount *mp,
1115 	void (*dtr)(struct vnode *, void *), void *dtr_arg)
1116 {
1117 	int locked;
1118 
1119 	KASSERT(vp->v_mount == NULL,
1120 		("insmntque: vnode already on per mount vnode list"));
1121 	VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)"));
1122 #ifdef DEBUG_VFS_LOCKS
1123 	if (!VFS_NEEDSGIANT(mp))
1124 		ASSERT_VOP_ELOCKED(vp,
1125 		    "insmntque: mp-safe fs and non-locked vp");
1126 #endif
1127 	MNT_ILOCK(mp);
1128 	if ((mp->mnt_kern_flag & MNTK_NOINSMNTQ) != 0 &&
1129 	    ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0 ||
1130 	     mp->mnt_nvnodelistsize == 0)) {
1131 		locked = VOP_ISLOCKED(vp);
1132 		if (!locked || (locked == LK_EXCLUSIVE &&
1133 		     (vp->v_vflag & VV_FORCEINSMQ) == 0)) {
1134 			MNT_IUNLOCK(mp);
1135 			if (dtr != NULL)
1136 				dtr(vp, dtr_arg);
1137 			return (EBUSY);
1138 		}
1139 	}
1140 	vp->v_mount = mp;
1141 	MNT_REF(mp);
1142 	TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1143 	VNASSERT(mp->mnt_nvnodelistsize >= 0, vp,
1144 		("neg mount point vnode list size"));
1145 	mp->mnt_nvnodelistsize++;
1146 	MNT_IUNLOCK(mp);
1147 	return (0);
1148 }
1149 
1150 int
1151 insmntque(struct vnode *vp, struct mount *mp)
1152 {
1153 
1154 	return (insmntque1(vp, mp, insmntque_stddtr, NULL));
1155 }
1156 
1157 /*
1158  * Flush out and invalidate all buffers associated with a bufobj
1159  * Called with the underlying object locked.
1160  */
1161 int
1162 bufobj_invalbuf(struct bufobj *bo, int flags, int slpflag, int slptimeo)
1163 {
1164 	int error;
1165 
1166 	BO_LOCK(bo);
1167 	if (flags & V_SAVE) {
1168 		error = bufobj_wwait(bo, slpflag, slptimeo);
1169 		if (error) {
1170 			BO_UNLOCK(bo);
1171 			return (error);
1172 		}
1173 		if (bo->bo_dirty.bv_cnt > 0) {
1174 			BO_UNLOCK(bo);
1175 			if ((error = BO_SYNC(bo, MNT_WAIT)) != 0)
1176 				return (error);
1177 			/*
1178 			 * XXX We could save a lock/unlock if this was only
1179 			 * enabled under INVARIANTS
1180 			 */
1181 			BO_LOCK(bo);
1182 			if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)
1183 				panic("vinvalbuf: dirty bufs");
1184 		}
1185 	}
1186 	/*
1187 	 * If you alter this loop please notice that interlock is dropped and
1188 	 * reacquired in flushbuflist.  Special care is needed to ensure that
1189 	 * no race conditions occur from this.
1190 	 */
1191 	do {
1192 		error = flushbuflist(&bo->bo_clean,
1193 		    flags, bo, slpflag, slptimeo);
1194 		if (error == 0)
1195 			error = flushbuflist(&bo->bo_dirty,
1196 			    flags, bo, slpflag, slptimeo);
1197 		if (error != 0 && error != EAGAIN) {
1198 			BO_UNLOCK(bo);
1199 			return (error);
1200 		}
1201 	} while (error != 0);
1202 
1203 	/*
1204 	 * Wait for I/O to complete.  XXX needs cleaning up.  The vnode can
1205 	 * have write I/O in-progress but if there is a VM object then the
1206 	 * VM object can also have read-I/O in-progress.
1207 	 */
1208 	do {
1209 		bufobj_wwait(bo, 0, 0);
1210 		BO_UNLOCK(bo);
1211 		if (bo->bo_object != NULL) {
1212 			VM_OBJECT_LOCK(bo->bo_object);
1213 			vm_object_pip_wait(bo->bo_object, "bovlbx");
1214 			VM_OBJECT_UNLOCK(bo->bo_object);
1215 		}
1216 		BO_LOCK(bo);
1217 	} while (bo->bo_numoutput > 0);
1218 	BO_UNLOCK(bo);
1219 
1220 	/*
1221 	 * Destroy the copy in the VM cache, too.
1222 	 */
1223 	if (bo->bo_object != NULL && (flags & (V_ALT | V_NORMAL)) == 0) {
1224 		VM_OBJECT_LOCK(bo->bo_object);
1225 		vm_object_page_remove(bo->bo_object, 0, 0, (flags & V_SAVE) ?
1226 		    OBJPR_CLEANONLY : 0);
1227 		VM_OBJECT_UNLOCK(bo->bo_object);
1228 	}
1229 
1230 #ifdef INVARIANTS
1231 	BO_LOCK(bo);
1232 	if ((flags & (V_ALT | V_NORMAL)) == 0 &&
1233 	    (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0))
1234 		panic("vinvalbuf: flush failed");
1235 	BO_UNLOCK(bo);
1236 #endif
1237 	return (0);
1238 }
1239 
1240 /*
1241  * Flush out and invalidate all buffers associated with a vnode.
1242  * Called with the underlying object locked.
1243  */
1244 int
1245 vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo)
1246 {
1247 
1248 	CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
1249 	ASSERT_VOP_LOCKED(vp, "vinvalbuf");
1250 	return (bufobj_invalbuf(&vp->v_bufobj, flags, slpflag, slptimeo));
1251 }
1252 
1253 /*
1254  * Flush out buffers on the specified list.
1255  *
1256  */
1257 static int
1258 flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo, int slpflag,
1259     int slptimeo)
1260 {
1261 	struct buf *bp, *nbp;
1262 	int retval, error;
1263 	daddr_t lblkno;
1264 	b_xflags_t xflags;
1265 
1266 	ASSERT_BO_LOCKED(bo);
1267 
1268 	retval = 0;
1269 	TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) {
1270 		if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) ||
1271 		    ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) {
1272 			continue;
1273 		}
1274 		lblkno = 0;
1275 		xflags = 0;
1276 		if (nbp != NULL) {
1277 			lblkno = nbp->b_lblkno;
1278 			xflags = nbp->b_xflags &
1279 				(BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN);
1280 		}
1281 		retval = EAGAIN;
1282 		error = BUF_TIMELOCK(bp,
1283 		    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_MTX(bo),
1284 		    "flushbuf", slpflag, slptimeo);
1285 		if (error) {
1286 			BO_LOCK(bo);
1287 			return (error != ENOLCK ? error : EAGAIN);
1288 		}
1289 		KASSERT(bp->b_bufobj == bo,
1290 		    ("bp %p wrong b_bufobj %p should be %p",
1291 		    bp, bp->b_bufobj, bo));
1292 		if (bp->b_bufobj != bo) {	/* XXX: necessary ? */
1293 			BUF_UNLOCK(bp);
1294 			BO_LOCK(bo);
1295 			return (EAGAIN);
1296 		}
1297 		/*
1298 		 * XXX Since there are no node locks for NFS, I
1299 		 * believe there is a slight chance that a delayed
1300 		 * write will occur while sleeping just above, so
1301 		 * check for it.
1302 		 */
1303 		if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
1304 		    (flags & V_SAVE)) {
1305 			BO_LOCK(bo);
1306 			bremfree(bp);
1307 			BO_UNLOCK(bo);
1308 			bp->b_flags |= B_ASYNC;
1309 			bwrite(bp);
1310 			BO_LOCK(bo);
1311 			return (EAGAIN);	/* XXX: why not loop ? */
1312 		}
1313 		BO_LOCK(bo);
1314 		bremfree(bp);
1315 		BO_UNLOCK(bo);
1316 		bp->b_flags |= (B_INVAL | B_RELBUF);
1317 		bp->b_flags &= ~B_ASYNC;
1318 		brelse(bp);
1319 		BO_LOCK(bo);
1320 		if (nbp != NULL &&
1321 		    (nbp->b_bufobj != bo ||
1322 		     nbp->b_lblkno != lblkno ||
1323 		     (nbp->b_xflags &
1324 		      (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN)) != xflags))
1325 			break;			/* nbp invalid */
1326 	}
1327 	return (retval);
1328 }
1329 
1330 /*
1331  * Truncate a file's buffer and pages to a specified length.  This
1332  * is in lieu of the old vinvalbuf mechanism, which performed unneeded
1333  * sync activity.
1334  */
1335 int
1336 vtruncbuf(struct vnode *vp, struct ucred *cred, struct thread *td,
1337     off_t length, int blksize)
1338 {
1339 	struct buf *bp, *nbp;
1340 	int anyfreed;
1341 	int trunclbn;
1342 	struct bufobj *bo;
1343 
1344 	CTR5(KTR_VFS, "%s: vp %p with cred %p and block %d:%ju", __func__,
1345 	    vp, cred, blksize, (uintmax_t)length);
1346 
1347 	/*
1348 	 * Round up to the *next* lbn.
1349 	 */
1350 	trunclbn = (length + blksize - 1) / blksize;
1351 
1352 	ASSERT_VOP_LOCKED(vp, "vtruncbuf");
1353 restart:
1354 	bo = &vp->v_bufobj;
1355 	BO_LOCK(bo);
1356 	anyfreed = 1;
1357 	for (;anyfreed;) {
1358 		anyfreed = 0;
1359 		TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) {
1360 			if (bp->b_lblkno < trunclbn)
1361 				continue;
1362 			if (BUF_LOCK(bp,
1363 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1364 			    BO_MTX(bo)) == ENOLCK)
1365 				goto restart;
1366 
1367 			BO_LOCK(bo);
1368 			bremfree(bp);
1369 			BO_UNLOCK(bo);
1370 			bp->b_flags |= (B_INVAL | B_RELBUF);
1371 			bp->b_flags &= ~B_ASYNC;
1372 			brelse(bp);
1373 			anyfreed = 1;
1374 
1375 			BO_LOCK(bo);
1376 			if (nbp != NULL &&
1377 			    (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1378 			    (nbp->b_vp != vp) ||
1379 			    (nbp->b_flags & B_DELWRI))) {
1380 				BO_UNLOCK(bo);
1381 				goto restart;
1382 			}
1383 		}
1384 
1385 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1386 			if (bp->b_lblkno < trunclbn)
1387 				continue;
1388 			if (BUF_LOCK(bp,
1389 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1390 			    BO_MTX(bo)) == ENOLCK)
1391 				goto restart;
1392 			BO_LOCK(bo);
1393 			bremfree(bp);
1394 			BO_UNLOCK(bo);
1395 			bp->b_flags |= (B_INVAL | B_RELBUF);
1396 			bp->b_flags &= ~B_ASYNC;
1397 			brelse(bp);
1398 			anyfreed = 1;
1399 
1400 			BO_LOCK(bo);
1401 			if (nbp != NULL &&
1402 			    (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1403 			    (nbp->b_vp != vp) ||
1404 			    (nbp->b_flags & B_DELWRI) == 0)) {
1405 				BO_UNLOCK(bo);
1406 				goto restart;
1407 			}
1408 		}
1409 	}
1410 
1411 	if (length > 0) {
1412 restartsync:
1413 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1414 			if (bp->b_lblkno > 0)
1415 				continue;
1416 			/*
1417 			 * Since we hold the vnode lock this should only
1418 			 * fail if we're racing with the buf daemon.
1419 			 */
1420 			if (BUF_LOCK(bp,
1421 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1422 			    BO_MTX(bo)) == ENOLCK) {
1423 				goto restart;
1424 			}
1425 			VNASSERT((bp->b_flags & B_DELWRI), vp,
1426 			    ("buf(%p) on dirty queue without DELWRI", bp));
1427 
1428 			BO_LOCK(bo);
1429 			bremfree(bp);
1430 			BO_UNLOCK(bo);
1431 			bawrite(bp);
1432 			BO_LOCK(bo);
1433 			goto restartsync;
1434 		}
1435 	}
1436 
1437 	bufobj_wwait(bo, 0, 0);
1438 	BO_UNLOCK(bo);
1439 	vnode_pager_setsize(vp, length);
1440 
1441 	return (0);
1442 }
1443 
1444 /*
1445  * buf_splay() - splay tree core for the clean/dirty list of buffers in
1446  *		 a vnode.
1447  *
1448  *	NOTE: We have to deal with the special case of a background bitmap
1449  *	buffer, a situation where two buffers will have the same logical
1450  *	block offset.  We want (1) only the foreground buffer to be accessed
1451  *	in a lookup and (2) must differentiate between the foreground and
1452  *	background buffer in the splay tree algorithm because the splay
1453  *	tree cannot normally handle multiple entities with the same 'index'.
1454  *	We accomplish this by adding differentiating flags to the splay tree's
1455  *	numerical domain.
1456  */
1457 static
1458 struct buf *
1459 buf_splay(daddr_t lblkno, b_xflags_t xflags, struct buf *root)
1460 {
1461 	struct buf dummy;
1462 	struct buf *lefttreemax, *righttreemin, *y;
1463 
1464 	if (root == NULL)
1465 		return (NULL);
1466 	lefttreemax = righttreemin = &dummy;
1467 	for (;;) {
1468 		if (lblkno < root->b_lblkno ||
1469 		    (lblkno == root->b_lblkno &&
1470 		    (xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1471 			if ((y = root->b_left) == NULL)
1472 				break;
1473 			if (lblkno < y->b_lblkno) {
1474 				/* Rotate right. */
1475 				root->b_left = y->b_right;
1476 				y->b_right = root;
1477 				root = y;
1478 				if ((y = root->b_left) == NULL)
1479 					break;
1480 			}
1481 			/* Link into the new root's right tree. */
1482 			righttreemin->b_left = root;
1483 			righttreemin = root;
1484 		} else if (lblkno > root->b_lblkno ||
1485 		    (lblkno == root->b_lblkno &&
1486 		    (xflags & BX_BKGRDMARKER) > (root->b_xflags & BX_BKGRDMARKER))) {
1487 			if ((y = root->b_right) == NULL)
1488 				break;
1489 			if (lblkno > y->b_lblkno) {
1490 				/* Rotate left. */
1491 				root->b_right = y->b_left;
1492 				y->b_left = root;
1493 				root = y;
1494 				if ((y = root->b_right) == NULL)
1495 					break;
1496 			}
1497 			/* Link into the new root's left tree. */
1498 			lefttreemax->b_right = root;
1499 			lefttreemax = root;
1500 		} else {
1501 			break;
1502 		}
1503 		root = y;
1504 	}
1505 	/* Assemble the new root. */
1506 	lefttreemax->b_right = root->b_left;
1507 	righttreemin->b_left = root->b_right;
1508 	root->b_left = dummy.b_right;
1509 	root->b_right = dummy.b_left;
1510 	return (root);
1511 }
1512 
1513 static void
1514 buf_vlist_remove(struct buf *bp)
1515 {
1516 	struct buf *root;
1517 	struct bufv *bv;
1518 
1519 	KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp));
1520 	ASSERT_BO_LOCKED(bp->b_bufobj);
1521 	KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) !=
1522 	    (BX_VNDIRTY|BX_VNCLEAN),
1523 	    ("buf_vlist_remove: Buf %p is on two lists", bp));
1524 	if (bp->b_xflags & BX_VNDIRTY)
1525 		bv = &bp->b_bufobj->bo_dirty;
1526 	else
1527 		bv = &bp->b_bufobj->bo_clean;
1528 	if (bp != bv->bv_root) {
1529 		root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1530 		KASSERT(root == bp, ("splay lookup failed in remove"));
1531 	}
1532 	if (bp->b_left == NULL) {
1533 		root = bp->b_right;
1534 	} else {
1535 		root = buf_splay(bp->b_lblkno, bp->b_xflags, bp->b_left);
1536 		root->b_right = bp->b_right;
1537 	}
1538 	bv->bv_root = root;
1539 	TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs);
1540 	bv->bv_cnt--;
1541 	bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1542 }
1543 
1544 /*
1545  * Add the buffer to the sorted clean or dirty block list using a
1546  * splay tree algorithm.
1547  *
1548  * NOTE: xflags is passed as a constant, optimizing this inline function!
1549  */
1550 static void
1551 buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags)
1552 {
1553 	struct buf *root;
1554 	struct bufv *bv;
1555 
1556 	ASSERT_BO_LOCKED(bo);
1557 	KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0,
1558 	    ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags));
1559 	bp->b_xflags |= xflags;
1560 	if (xflags & BX_VNDIRTY)
1561 		bv = &bo->bo_dirty;
1562 	else
1563 		bv = &bo->bo_clean;
1564 
1565 	root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1566 	if (root == NULL) {
1567 		bp->b_left = NULL;
1568 		bp->b_right = NULL;
1569 		TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs);
1570 	} else if (bp->b_lblkno < root->b_lblkno ||
1571 	    (bp->b_lblkno == root->b_lblkno &&
1572 	    (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1573 		bp->b_left = root->b_left;
1574 		bp->b_right = root;
1575 		root->b_left = NULL;
1576 		TAILQ_INSERT_BEFORE(root, bp, b_bobufs);
1577 	} else {
1578 		bp->b_right = root->b_right;
1579 		bp->b_left = root;
1580 		root->b_right = NULL;
1581 		TAILQ_INSERT_AFTER(&bv->bv_hd, root, bp, b_bobufs);
1582 	}
1583 	bv->bv_cnt++;
1584 	bv->bv_root = bp;
1585 }
1586 
1587 /*
1588  * Lookup a buffer using the splay tree.  Note that we specifically avoid
1589  * shadow buffers used in background bitmap writes.
1590  *
1591  * This code isn't quite efficient as it could be because we are maintaining
1592  * two sorted lists and do not know which list the block resides in.
1593  *
1594  * During a "make buildworld" the desired buffer is found at one of
1595  * the roots more than 60% of the time.  Thus, checking both roots
1596  * before performing either splay eliminates unnecessary splays on the
1597  * first tree splayed.
1598  */
1599 struct buf *
1600 gbincore(struct bufobj *bo, daddr_t lblkno)
1601 {
1602 	struct buf *bp;
1603 
1604 	ASSERT_BO_LOCKED(bo);
1605 	if ((bp = bo->bo_clean.bv_root) != NULL &&
1606 	    bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1607 		return (bp);
1608 	if ((bp = bo->bo_dirty.bv_root) != NULL &&
1609 	    bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1610 		return (bp);
1611 	if ((bp = bo->bo_clean.bv_root) != NULL) {
1612 		bo->bo_clean.bv_root = bp = buf_splay(lblkno, 0, bp);
1613 		if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1614 			return (bp);
1615 	}
1616 	if ((bp = bo->bo_dirty.bv_root) != NULL) {
1617 		bo->bo_dirty.bv_root = bp = buf_splay(lblkno, 0, bp);
1618 		if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1619 			return (bp);
1620 	}
1621 	return (NULL);
1622 }
1623 
1624 /*
1625  * Associate a buffer with a vnode.
1626  */
1627 void
1628 bgetvp(struct vnode *vp, struct buf *bp)
1629 {
1630 	struct bufobj *bo;
1631 
1632 	bo = &vp->v_bufobj;
1633 	ASSERT_BO_LOCKED(bo);
1634 	VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free"));
1635 
1636 	CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags);
1637 	VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp,
1638 	    ("bgetvp: bp already attached! %p", bp));
1639 
1640 	vhold(vp);
1641 	if (VFS_NEEDSGIANT(vp->v_mount) || bo->bo_flag & BO_NEEDSGIANT)
1642 		bp->b_flags |= B_NEEDSGIANT;
1643 	bp->b_vp = vp;
1644 	bp->b_bufobj = bo;
1645 	/*
1646 	 * Insert onto list for new vnode.
1647 	 */
1648 	buf_vlist_add(bp, bo, BX_VNCLEAN);
1649 }
1650 
1651 /*
1652  * Disassociate a buffer from a vnode.
1653  */
1654 void
1655 brelvp(struct buf *bp)
1656 {
1657 	struct bufobj *bo;
1658 	struct vnode *vp;
1659 
1660 	CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
1661 	KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1662 
1663 	/*
1664 	 * Delete from old vnode list, if on one.
1665 	 */
1666 	vp = bp->b_vp;		/* XXX */
1667 	bo = bp->b_bufobj;
1668 	BO_LOCK(bo);
1669 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1670 		buf_vlist_remove(bp);
1671 	else
1672 		panic("brelvp: Buffer %p not on queue.", bp);
1673 	if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1674 		bo->bo_flag &= ~BO_ONWORKLST;
1675 		mtx_lock(&sync_mtx);
1676 		LIST_REMOVE(bo, bo_synclist);
1677 		syncer_worklist_len--;
1678 		mtx_unlock(&sync_mtx);
1679 	}
1680 	bp->b_flags &= ~B_NEEDSGIANT;
1681 	bp->b_vp = NULL;
1682 	bp->b_bufobj = NULL;
1683 	BO_UNLOCK(bo);
1684 	vdrop(vp);
1685 }
1686 
1687 /*
1688  * Add an item to the syncer work queue.
1689  */
1690 static void
1691 vn_syncer_add_to_worklist(struct bufobj *bo, int delay)
1692 {
1693 	int queue, slot;
1694 
1695 	ASSERT_BO_LOCKED(bo);
1696 
1697 	mtx_lock(&sync_mtx);
1698 	if (bo->bo_flag & BO_ONWORKLST)
1699 		LIST_REMOVE(bo, bo_synclist);
1700 	else {
1701 		bo->bo_flag |= BO_ONWORKLST;
1702 		syncer_worklist_len++;
1703 	}
1704 
1705 	if (delay > syncer_maxdelay - 2)
1706 		delay = syncer_maxdelay - 2;
1707 	slot = (syncer_delayno + delay) & syncer_mask;
1708 
1709 	queue = VFS_NEEDSGIANT(bo->__bo_vnode->v_mount) ? WI_GIANTQ :
1710 	    WI_MPSAFEQ;
1711 	LIST_INSERT_HEAD(&syncer_workitem_pending[queue][slot], bo,
1712 	    bo_synclist);
1713 	mtx_unlock(&sync_mtx);
1714 }
1715 
1716 static int
1717 sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS)
1718 {
1719 	int error, len;
1720 
1721 	mtx_lock(&sync_mtx);
1722 	len = syncer_worklist_len - sync_vnode_count;
1723 	mtx_unlock(&sync_mtx);
1724 	error = SYSCTL_OUT(req, &len, sizeof(len));
1725 	return (error);
1726 }
1727 
1728 SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0,
1729     sysctl_vfs_worklist_len, "I", "Syncer thread worklist length");
1730 
1731 static struct proc *updateproc;
1732 static void sched_sync(void);
1733 static struct kproc_desc up_kp = {
1734 	"syncer",
1735 	sched_sync,
1736 	&updateproc
1737 };
1738 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp);
1739 
1740 static int
1741 sync_vnode(struct synclist *slp, struct bufobj **bo, struct thread *td)
1742 {
1743 	struct vnode *vp;
1744 	struct mount *mp;
1745 
1746 	*bo = LIST_FIRST(slp);
1747 	if (*bo == NULL)
1748 		return (0);
1749 	vp = (*bo)->__bo_vnode;	/* XXX */
1750 	if (VOP_ISLOCKED(vp) != 0 || VI_TRYLOCK(vp) == 0)
1751 		return (1);
1752 	/*
1753 	 * We use vhold in case the vnode does not
1754 	 * successfully sync.  vhold prevents the vnode from
1755 	 * going away when we unlock the sync_mtx so that
1756 	 * we can acquire the vnode interlock.
1757 	 */
1758 	vholdl(vp);
1759 	mtx_unlock(&sync_mtx);
1760 	VI_UNLOCK(vp);
1761 	if (vn_start_write(vp, &mp, V_NOWAIT) != 0) {
1762 		vdrop(vp);
1763 		mtx_lock(&sync_mtx);
1764 		return (*bo == LIST_FIRST(slp));
1765 	}
1766 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1767 	(void) VOP_FSYNC(vp, MNT_LAZY, td);
1768 	VOP_UNLOCK(vp, 0);
1769 	vn_finished_write(mp);
1770 	BO_LOCK(*bo);
1771 	if (((*bo)->bo_flag & BO_ONWORKLST) != 0) {
1772 		/*
1773 		 * Put us back on the worklist.  The worklist
1774 		 * routine will remove us from our current
1775 		 * position and then add us back in at a later
1776 		 * position.
1777 		 */
1778 		vn_syncer_add_to_worklist(*bo, syncdelay);
1779 	}
1780 	BO_UNLOCK(*bo);
1781 	vdrop(vp);
1782 	mtx_lock(&sync_mtx);
1783 	return (0);
1784 }
1785 
1786 /*
1787  * System filesystem synchronizer daemon.
1788  */
1789 static void
1790 sched_sync(void)
1791 {
1792 	struct synclist *gnext, *next;
1793 	struct synclist *gslp, *slp;
1794 	struct bufobj *bo;
1795 	long starttime;
1796 	struct thread *td = curthread;
1797 	int last_work_seen;
1798 	int net_worklist_len;
1799 	int syncer_final_iter;
1800 	int first_printf;
1801 	int error;
1802 
1803 	last_work_seen = 0;
1804 	syncer_final_iter = 0;
1805 	first_printf = 1;
1806 	syncer_state = SYNCER_RUNNING;
1807 	starttime = time_uptime;
1808 	td->td_pflags |= TDP_NORUNNINGBUF;
1809 
1810 	EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc,
1811 	    SHUTDOWN_PRI_LAST);
1812 
1813 	mtx_lock(&sync_mtx);
1814 	for (;;) {
1815 		if (syncer_state == SYNCER_FINAL_DELAY &&
1816 		    syncer_final_iter == 0) {
1817 			mtx_unlock(&sync_mtx);
1818 			kproc_suspend_check(td->td_proc);
1819 			mtx_lock(&sync_mtx);
1820 		}
1821 		net_worklist_len = syncer_worklist_len - sync_vnode_count;
1822 		if (syncer_state != SYNCER_RUNNING &&
1823 		    starttime != time_uptime) {
1824 			if (first_printf) {
1825 				printf("\nSyncing disks, vnodes remaining...");
1826 				first_printf = 0;
1827 			}
1828 			printf("%d ", net_worklist_len);
1829 		}
1830 		starttime = time_uptime;
1831 
1832 		/*
1833 		 * Push files whose dirty time has expired.  Be careful
1834 		 * of interrupt race on slp queue.
1835 		 *
1836 		 * Skip over empty worklist slots when shutting down.
1837 		 */
1838 		do {
1839 			slp = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno];
1840 			gslp = &syncer_workitem_pending[WI_GIANTQ][syncer_delayno];
1841 			syncer_delayno += 1;
1842 			if (syncer_delayno == syncer_maxdelay)
1843 				syncer_delayno = 0;
1844 			next = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno];
1845 			gnext = &syncer_workitem_pending[WI_GIANTQ][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 		    LIST_EMPTY(gslp) && 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 #ifdef SW_WATCHDOG
1879 			if (first_printf == 0)
1880 				wdog_kern_pat(WD_LASTVAL);
1881 #endif
1882 		}
1883 		if (!LIST_EMPTY(gslp)) {
1884 			mtx_unlock(&sync_mtx);
1885 			mtx_lock(&Giant);
1886 			mtx_lock(&sync_mtx);
1887 			while (!LIST_EMPTY(gslp)) {
1888 				error = sync_vnode(gslp, &bo, td);
1889 				if (error == 1) {
1890 					LIST_REMOVE(bo, bo_synclist);
1891 					LIST_INSERT_HEAD(gnext, bo,
1892 					    bo_synclist);
1893 					continue;
1894 				}
1895 			}
1896 			mtx_unlock(&Giant);
1897 		}
1898 		if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0)
1899 			syncer_final_iter--;
1900 		/*
1901 		 * The variable rushjob allows the kernel to speed up the
1902 		 * processing of the filesystem syncer process. A rushjob
1903 		 * value of N tells the filesystem syncer to process the next
1904 		 * N seconds worth of work on its queue ASAP. Currently rushjob
1905 		 * is used by the soft update code to speed up the filesystem
1906 		 * syncer process when the incore state is getting so far
1907 		 * ahead of the disk that the kernel memory pool is being
1908 		 * threatened with exhaustion.
1909 		 */
1910 		if (rushjob > 0) {
1911 			rushjob -= 1;
1912 			continue;
1913 		}
1914 		/*
1915 		 * Just sleep for a short period of time between
1916 		 * iterations when shutting down to allow some I/O
1917 		 * to happen.
1918 		 *
1919 		 * If it has taken us less than a second to process the
1920 		 * current work, then wait. Otherwise start right over
1921 		 * again. We can still lose time if any single round
1922 		 * takes more than two seconds, but it does not really
1923 		 * matter as we are just trying to generally pace the
1924 		 * filesystem activity.
1925 		 */
1926 		if (syncer_state != SYNCER_RUNNING ||
1927 		    time_uptime == starttime) {
1928 			thread_lock(td);
1929 			sched_prio(td, PPAUSE);
1930 			thread_unlock(td);
1931 		}
1932 		if (syncer_state != SYNCER_RUNNING)
1933 			cv_timedwait(&sync_wakeup, &sync_mtx,
1934 			    hz / SYNCER_SHUTDOWN_SPEEDUP);
1935 		else if (time_uptime == starttime)
1936 			cv_timedwait(&sync_wakeup, &sync_mtx, hz);
1937 	}
1938 }
1939 
1940 /*
1941  * Request the syncer daemon to speed up its work.
1942  * We never push it to speed up more than half of its
1943  * normal turn time, otherwise it could take over the cpu.
1944  */
1945 int
1946 speedup_syncer(void)
1947 {
1948 	int ret = 0;
1949 
1950 	mtx_lock(&sync_mtx);
1951 	if (rushjob < syncdelay / 2) {
1952 		rushjob += 1;
1953 		stat_rush_requests += 1;
1954 		ret = 1;
1955 	}
1956 	mtx_unlock(&sync_mtx);
1957 	cv_broadcast(&sync_wakeup);
1958 	return (ret);
1959 }
1960 
1961 /*
1962  * Tell the syncer to speed up its work and run though its work
1963  * list several times, then tell it to shut down.
1964  */
1965 static void
1966 syncer_shutdown(void *arg, int howto)
1967 {
1968 
1969 	if (howto & RB_NOSYNC)
1970 		return;
1971 	mtx_lock(&sync_mtx);
1972 	syncer_state = SYNCER_SHUTTING_DOWN;
1973 	rushjob = 0;
1974 	mtx_unlock(&sync_mtx);
1975 	cv_broadcast(&sync_wakeup);
1976 	kproc_shutdown(arg, howto);
1977 }
1978 
1979 /*
1980  * Reassign a buffer from one vnode to another.
1981  * Used to assign file specific control information
1982  * (indirect blocks) to the vnode to which they belong.
1983  */
1984 void
1985 reassignbuf(struct buf *bp)
1986 {
1987 	struct vnode *vp;
1988 	struct bufobj *bo;
1989 	int delay;
1990 #ifdef INVARIANTS
1991 	struct bufv *bv;
1992 #endif
1993 
1994 	vp = bp->b_vp;
1995 	bo = bp->b_bufobj;
1996 	++reassignbufcalls;
1997 
1998 	CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X",
1999 	    bp, bp->b_vp, bp->b_flags);
2000 	/*
2001 	 * B_PAGING flagged buffers cannot be reassigned because their vp
2002 	 * is not fully linked in.
2003 	 */
2004 	if (bp->b_flags & B_PAGING)
2005 		panic("cannot reassign paging buffer");
2006 
2007 	/*
2008 	 * Delete from old vnode list, if on one.
2009 	 */
2010 	BO_LOCK(bo);
2011 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
2012 		buf_vlist_remove(bp);
2013 	else
2014 		panic("reassignbuf: Buffer %p not on queue.", bp);
2015 	/*
2016 	 * If dirty, put on list of dirty buffers; otherwise insert onto list
2017 	 * of clean buffers.
2018 	 */
2019 	if (bp->b_flags & B_DELWRI) {
2020 		if ((bo->bo_flag & BO_ONWORKLST) == 0) {
2021 			switch (vp->v_type) {
2022 			case VDIR:
2023 				delay = dirdelay;
2024 				break;
2025 			case VCHR:
2026 				delay = metadelay;
2027 				break;
2028 			default:
2029 				delay = filedelay;
2030 			}
2031 			vn_syncer_add_to_worklist(bo, delay);
2032 		}
2033 		buf_vlist_add(bp, bo, BX_VNDIRTY);
2034 	} else {
2035 		buf_vlist_add(bp, bo, BX_VNCLEAN);
2036 
2037 		if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
2038 			mtx_lock(&sync_mtx);
2039 			LIST_REMOVE(bo, bo_synclist);
2040 			syncer_worklist_len--;
2041 			mtx_unlock(&sync_mtx);
2042 			bo->bo_flag &= ~BO_ONWORKLST;
2043 		}
2044 	}
2045 #ifdef INVARIANTS
2046 	bv = &bo->bo_clean;
2047 	bp = TAILQ_FIRST(&bv->bv_hd);
2048 	KASSERT(bp == NULL || bp->b_bufobj == bo,
2049 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2050 	bp = TAILQ_LAST(&bv->bv_hd, buflists);
2051 	KASSERT(bp == NULL || bp->b_bufobj == bo,
2052 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2053 	bv = &bo->bo_dirty;
2054 	bp = TAILQ_FIRST(&bv->bv_hd);
2055 	KASSERT(bp == NULL || bp->b_bufobj == bo,
2056 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2057 	bp = TAILQ_LAST(&bv->bv_hd, buflists);
2058 	KASSERT(bp == NULL || bp->b_bufobj == bo,
2059 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2060 #endif
2061 	BO_UNLOCK(bo);
2062 }
2063 
2064 /*
2065  * Increment the use and hold counts on the vnode, taking care to reference
2066  * the driver's usecount if this is a chardev.  The vholdl() will remove
2067  * the vnode from the free list if it is presently free.  Requires the
2068  * vnode interlock and returns with it held.
2069  */
2070 static void
2071 v_incr_usecount(struct vnode *vp)
2072 {
2073 
2074 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2075 	vp->v_usecount++;
2076 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2077 		dev_lock();
2078 		vp->v_rdev->si_usecount++;
2079 		dev_unlock();
2080 	}
2081 	vholdl(vp);
2082 }
2083 
2084 /*
2085  * Turn a holdcnt into a use+holdcnt such that only one call to
2086  * v_decr_usecount is needed.
2087  */
2088 static void
2089 v_upgrade_usecount(struct vnode *vp)
2090 {
2091 
2092 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2093 	vp->v_usecount++;
2094 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2095 		dev_lock();
2096 		vp->v_rdev->si_usecount++;
2097 		dev_unlock();
2098 	}
2099 }
2100 
2101 /*
2102  * Decrement the vnode use and hold count along with the driver's usecount
2103  * if this is a chardev.  The vdropl() below releases the vnode interlock
2104  * as it may free the vnode.
2105  */
2106 static void
2107 v_decr_usecount(struct vnode *vp)
2108 {
2109 
2110 	ASSERT_VI_LOCKED(vp, __FUNCTION__);
2111 	VNASSERT(vp->v_usecount > 0, vp,
2112 	    ("v_decr_usecount: negative usecount"));
2113 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2114 	vp->v_usecount--;
2115 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2116 		dev_lock();
2117 		vp->v_rdev->si_usecount--;
2118 		dev_unlock();
2119 	}
2120 	vdropl(vp);
2121 }
2122 
2123 /*
2124  * Decrement only the use count and driver use count.  This is intended to
2125  * be paired with a follow on vdropl() to release the remaining hold count.
2126  * In this way we may vgone() a vnode with a 0 usecount without risk of
2127  * having it end up on a free list because the hold count is kept above 0.
2128  */
2129 static void
2130 v_decr_useonly(struct vnode *vp)
2131 {
2132 
2133 	ASSERT_VI_LOCKED(vp, __FUNCTION__);
2134 	VNASSERT(vp->v_usecount > 0, vp,
2135 	    ("v_decr_useonly: negative usecount"));
2136 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2137 	vp->v_usecount--;
2138 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2139 		dev_lock();
2140 		vp->v_rdev->si_usecount--;
2141 		dev_unlock();
2142 	}
2143 }
2144 
2145 /*
2146  * Grab a particular vnode from the free list, increment its
2147  * reference count and lock it.  VI_DOOMED is set if the vnode
2148  * is being destroyed.  Only callers who specify LK_RETRY will
2149  * see doomed vnodes.  If inactive processing was delayed in
2150  * vput try to do it here.
2151  */
2152 int
2153 vget(struct vnode *vp, int flags, struct thread *td)
2154 {
2155 	int error;
2156 
2157 	error = 0;
2158 	VFS_ASSERT_GIANT(vp->v_mount);
2159 	VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
2160 	    ("vget: invalid lock operation"));
2161 	CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
2162 
2163 	if ((flags & LK_INTERLOCK) == 0)
2164 		VI_LOCK(vp);
2165 	vholdl(vp);
2166 	if ((error = vn_lock(vp, flags | LK_INTERLOCK)) != 0) {
2167 		vdrop(vp);
2168 		CTR2(KTR_VFS, "%s: impossible to lock vnode %p", __func__,
2169 		    vp);
2170 		return (error);
2171 	}
2172 	if (vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0)
2173 		panic("vget: vn_lock failed to return ENOENT\n");
2174 	VI_LOCK(vp);
2175 	/* Upgrade our holdcnt to a usecount. */
2176 	v_upgrade_usecount(vp);
2177 	/*
2178 	 * We don't guarantee that any particular close will
2179 	 * trigger inactive processing so just make a best effort
2180 	 * here at preventing a reference to a removed file.  If
2181 	 * we don't succeed no harm is done.
2182 	 */
2183 	if (vp->v_iflag & VI_OWEINACT) {
2184 		if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE &&
2185 		    (flags & LK_NOWAIT) == 0)
2186 			vinactive(vp, td);
2187 		vp->v_iflag &= ~VI_OWEINACT;
2188 	}
2189 	VI_UNLOCK(vp);
2190 	return (0);
2191 }
2192 
2193 /*
2194  * Increase the reference count of a vnode.
2195  */
2196 void
2197 vref(struct vnode *vp)
2198 {
2199 
2200 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2201 	VI_LOCK(vp);
2202 	v_incr_usecount(vp);
2203 	VI_UNLOCK(vp);
2204 }
2205 
2206 /*
2207  * Return reference count of a vnode.
2208  *
2209  * The results of this call are only guaranteed when some mechanism other
2210  * than the VI lock is used to stop other processes from gaining references
2211  * to the vnode.  This may be the case if the caller holds the only reference.
2212  * This is also useful when stale data is acceptable as race conditions may
2213  * be accounted for by some other means.
2214  */
2215 int
2216 vrefcnt(struct vnode *vp)
2217 {
2218 	int usecnt;
2219 
2220 	VI_LOCK(vp);
2221 	usecnt = vp->v_usecount;
2222 	VI_UNLOCK(vp);
2223 
2224 	return (usecnt);
2225 }
2226 
2227 #define	VPUTX_VRELE	1
2228 #define	VPUTX_VPUT	2
2229 #define	VPUTX_VUNREF	3
2230 
2231 static void
2232 vputx(struct vnode *vp, int func)
2233 {
2234 	int error;
2235 
2236 	KASSERT(vp != NULL, ("vputx: null vp"));
2237 	if (func == VPUTX_VUNREF)
2238 		ASSERT_VOP_LOCKED(vp, "vunref");
2239 	else if (func == VPUTX_VPUT)
2240 		ASSERT_VOP_LOCKED(vp, "vput");
2241 	else
2242 		KASSERT(func == VPUTX_VRELE, ("vputx: wrong func"));
2243 	VFS_ASSERT_GIANT(vp->v_mount);
2244 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2245 	VI_LOCK(vp);
2246 
2247 	/* Skip this v_writecount check if we're going to panic below. */
2248 	VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp,
2249 	    ("vputx: missed vn_close"));
2250 	error = 0;
2251 
2252 	if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2253 	    vp->v_usecount == 1)) {
2254 		if (func == VPUTX_VPUT)
2255 			VOP_UNLOCK(vp, 0);
2256 		v_decr_usecount(vp);
2257 		return;
2258 	}
2259 
2260 	if (vp->v_usecount != 1) {
2261 		vprint("vputx: negative ref count", vp);
2262 		panic("vputx: negative ref cnt");
2263 	}
2264 	CTR2(KTR_VFS, "%s: return vnode %p to the freelist", __func__, vp);
2265 	/*
2266 	 * We want to hold the vnode until the inactive finishes to
2267 	 * prevent vgone() races.  We drop the use count here and the
2268 	 * hold count below when we're done.
2269 	 */
2270 	v_decr_useonly(vp);
2271 	/*
2272 	 * We must call VOP_INACTIVE with the node locked. Mark
2273 	 * as VI_DOINGINACT to avoid recursion.
2274 	 */
2275 	vp->v_iflag |= VI_OWEINACT;
2276 	switch (func) {
2277 	case VPUTX_VRELE:
2278 		error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK);
2279 		VI_LOCK(vp);
2280 		break;
2281 	case VPUTX_VPUT:
2282 		if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
2283 			error = VOP_LOCK(vp, LK_UPGRADE | LK_INTERLOCK |
2284 			    LK_NOWAIT);
2285 			VI_LOCK(vp);
2286 		}
2287 		break;
2288 	case VPUTX_VUNREF:
2289 		if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
2290 			error = EBUSY;
2291 		break;
2292 	}
2293 	if (vp->v_usecount > 0)
2294 		vp->v_iflag &= ~VI_OWEINACT;
2295 	if (error == 0) {
2296 		if (vp->v_iflag & VI_OWEINACT)
2297 			vinactive(vp, curthread);
2298 		if (func != VPUTX_VUNREF)
2299 			VOP_UNLOCK(vp, 0);
2300 	}
2301 	vdropl(vp);
2302 }
2303 
2304 /*
2305  * Vnode put/release.
2306  * If count drops to zero, call inactive routine and return to freelist.
2307  */
2308 void
2309 vrele(struct vnode *vp)
2310 {
2311 
2312 	vputx(vp, VPUTX_VRELE);
2313 }
2314 
2315 /*
2316  * Release an already locked vnode.  This give the same effects as
2317  * unlock+vrele(), but takes less time and avoids releasing and
2318  * re-aquiring the lock (as vrele() acquires the lock internally.)
2319  */
2320 void
2321 vput(struct vnode *vp)
2322 {
2323 
2324 	vputx(vp, VPUTX_VPUT);
2325 }
2326 
2327 /*
2328  * Release an exclusively locked vnode. Do not unlock the vnode lock.
2329  */
2330 void
2331 vunref(struct vnode *vp)
2332 {
2333 
2334 	vputx(vp, VPUTX_VUNREF);
2335 }
2336 
2337 /*
2338  * Somebody doesn't want the vnode recycled.
2339  */
2340 void
2341 vhold(struct vnode *vp)
2342 {
2343 
2344 	VI_LOCK(vp);
2345 	vholdl(vp);
2346 	VI_UNLOCK(vp);
2347 }
2348 
2349 void
2350 vholdl(struct vnode *vp)
2351 {
2352 
2353 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2354 	vp->v_holdcnt++;
2355 	if (VSHOULDBUSY(vp))
2356 		vbusy(vp);
2357 }
2358 
2359 /*
2360  * Note that there is one less who cares about this vnode.  vdrop() is the
2361  * opposite of vhold().
2362  */
2363 void
2364 vdrop(struct vnode *vp)
2365 {
2366 
2367 	VI_LOCK(vp);
2368 	vdropl(vp);
2369 }
2370 
2371 /*
2372  * Drop the hold count of the vnode.  If this is the last reference to
2373  * the vnode we will free it if it has been vgone'd otherwise it is
2374  * placed on the free list.
2375  */
2376 void
2377 vdropl(struct vnode *vp)
2378 {
2379 
2380 	ASSERT_VI_LOCKED(vp, "vdropl");
2381 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2382 	if (vp->v_holdcnt <= 0)
2383 		panic("vdrop: holdcnt %d", vp->v_holdcnt);
2384 	vp->v_holdcnt--;
2385 	if (vp->v_holdcnt == 0) {
2386 		if (vp->v_iflag & VI_DOOMED) {
2387 			CTR2(KTR_VFS, "%s: destroying the vnode %p", __func__,
2388 			    vp);
2389 			vdestroy(vp);
2390 			return;
2391 		} else
2392 			vfree(vp);
2393 	}
2394 	VI_UNLOCK(vp);
2395 }
2396 
2397 /*
2398  * Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT
2399  * flags.  DOINGINACT prevents us from recursing in calls to vinactive.
2400  * OWEINACT tracks whether a vnode missed a call to inactive due to a
2401  * failed lock upgrade.
2402  */
2403 static void
2404 vinactive(struct vnode *vp, struct thread *td)
2405 {
2406 
2407 	ASSERT_VOP_ELOCKED(vp, "vinactive");
2408 	ASSERT_VI_LOCKED(vp, "vinactive");
2409 	VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp,
2410 	    ("vinactive: recursed on VI_DOINGINACT"));
2411 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2412 	vp->v_iflag |= VI_DOINGINACT;
2413 	vp->v_iflag &= ~VI_OWEINACT;
2414 	VI_UNLOCK(vp);
2415 	VOP_INACTIVE(vp, td);
2416 	VI_LOCK(vp);
2417 	VNASSERT(vp->v_iflag & VI_DOINGINACT, vp,
2418 	    ("vinactive: lost VI_DOINGINACT"));
2419 	vp->v_iflag &= ~VI_DOINGINACT;
2420 }
2421 
2422 /*
2423  * Remove any vnodes in the vnode table belonging to mount point mp.
2424  *
2425  * If FORCECLOSE is not specified, there should not be any active ones,
2426  * return error if any are found (nb: this is a user error, not a
2427  * system error). If FORCECLOSE is specified, detach any active vnodes
2428  * that are found.
2429  *
2430  * If WRITECLOSE is set, only flush out regular file vnodes open for
2431  * writing.
2432  *
2433  * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
2434  *
2435  * `rootrefs' specifies the base reference count for the root vnode
2436  * of this filesystem. The root vnode is considered busy if its
2437  * v_usecount exceeds this value. On a successful return, vflush(, td)
2438  * will call vrele() on the root vnode exactly rootrefs times.
2439  * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
2440  * be zero.
2441  */
2442 #ifdef DIAGNOSTIC
2443 static int busyprt = 0;		/* print out busy vnodes */
2444 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "Print out busy vnodes");
2445 #endif
2446 
2447 int
2448 vflush(struct mount *mp, int rootrefs, int flags, struct thread *td)
2449 {
2450 	struct vnode *vp, *mvp, *rootvp = NULL;
2451 	struct vattr vattr;
2452 	int busy = 0, error;
2453 
2454 	CTR4(KTR_VFS, "%s: mp %p with rootrefs %d and flags %d", __func__, mp,
2455 	    rootrefs, flags);
2456 	if (rootrefs > 0) {
2457 		KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
2458 		    ("vflush: bad args"));
2459 		/*
2460 		 * Get the filesystem root vnode. We can vput() it
2461 		 * immediately, since with rootrefs > 0, it won't go away.
2462 		 */
2463 		if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp)) != 0) {
2464 			CTR2(KTR_VFS, "%s: vfs_root lookup failed with %d",
2465 			    __func__, error);
2466 			return (error);
2467 		}
2468 		vput(rootvp);
2469 	}
2470 	MNT_ILOCK(mp);
2471 loop:
2472 	MNT_VNODE_FOREACH(vp, mp, mvp) {
2473 		VI_LOCK(vp);
2474 		vholdl(vp);
2475 		MNT_IUNLOCK(mp);
2476 		error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE);
2477 		if (error) {
2478 			vdrop(vp);
2479 			MNT_ILOCK(mp);
2480 			MNT_VNODE_FOREACH_ABORT_ILOCKED(mp, mvp);
2481 			goto loop;
2482 		}
2483 		/*
2484 		 * Skip over a vnodes marked VV_SYSTEM.
2485 		 */
2486 		if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
2487 			VOP_UNLOCK(vp, 0);
2488 			vdrop(vp);
2489 			MNT_ILOCK(mp);
2490 			continue;
2491 		}
2492 		/*
2493 		 * If WRITECLOSE is set, flush out unlinked but still open
2494 		 * files (even if open only for reading) and regular file
2495 		 * vnodes open for writing.
2496 		 */
2497 		if (flags & WRITECLOSE) {
2498 			error = VOP_GETATTR(vp, &vattr, td->td_ucred);
2499 			VI_LOCK(vp);
2500 
2501 			if ((vp->v_type == VNON ||
2502 			    (error == 0 && vattr.va_nlink > 0)) &&
2503 			    (vp->v_writecount == 0 || vp->v_type != VREG)) {
2504 				VOP_UNLOCK(vp, 0);
2505 				vdropl(vp);
2506 				MNT_ILOCK(mp);
2507 				continue;
2508 			}
2509 		} else
2510 			VI_LOCK(vp);
2511 		/*
2512 		 * With v_usecount == 0, all we need to do is clear out the
2513 		 * vnode data structures and we are done.
2514 		 *
2515 		 * If FORCECLOSE is set, forcibly close the vnode.
2516 		 */
2517 		if (vp->v_usecount == 0 || (flags & FORCECLOSE)) {
2518 			VNASSERT(vp->v_usecount == 0 ||
2519 			    (vp->v_type != VCHR && vp->v_type != VBLK), vp,
2520 			    ("device VNODE %p is FORCECLOSED", vp));
2521 			vgonel(vp);
2522 		} else {
2523 			busy++;
2524 #ifdef DIAGNOSTIC
2525 			if (busyprt)
2526 				vprint("vflush: busy vnode", vp);
2527 #endif
2528 		}
2529 		VOP_UNLOCK(vp, 0);
2530 		vdropl(vp);
2531 		MNT_ILOCK(mp);
2532 	}
2533 	MNT_IUNLOCK(mp);
2534 	if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
2535 		/*
2536 		 * If just the root vnode is busy, and if its refcount
2537 		 * is equal to `rootrefs', then go ahead and kill it.
2538 		 */
2539 		VI_LOCK(rootvp);
2540 		KASSERT(busy > 0, ("vflush: not busy"));
2541 		VNASSERT(rootvp->v_usecount >= rootrefs, rootvp,
2542 		    ("vflush: usecount %d < rootrefs %d",
2543 		     rootvp->v_usecount, rootrefs));
2544 		if (busy == 1 && rootvp->v_usecount == rootrefs) {
2545 			VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK);
2546 			vgone(rootvp);
2547 			VOP_UNLOCK(rootvp, 0);
2548 			busy = 0;
2549 		} else
2550 			VI_UNLOCK(rootvp);
2551 	}
2552 	if (busy) {
2553 		CTR2(KTR_VFS, "%s: failing as %d vnodes are busy", __func__,
2554 		    busy);
2555 		return (EBUSY);
2556 	}
2557 	for (; rootrefs > 0; rootrefs--)
2558 		vrele(rootvp);
2559 	return (0);
2560 }
2561 
2562 /*
2563  * Recycle an unused vnode to the front of the free list.
2564  */
2565 int
2566 vrecycle(struct vnode *vp, struct thread *td)
2567 {
2568 	int recycled;
2569 
2570 	ASSERT_VOP_ELOCKED(vp, "vrecycle");
2571 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2572 	recycled = 0;
2573 	VI_LOCK(vp);
2574 	if (vp->v_usecount == 0) {
2575 		recycled = 1;
2576 		vgonel(vp);
2577 	}
2578 	VI_UNLOCK(vp);
2579 	return (recycled);
2580 }
2581 
2582 /*
2583  * Eliminate all activity associated with a vnode
2584  * in preparation for reuse.
2585  */
2586 void
2587 vgone(struct vnode *vp)
2588 {
2589 	VI_LOCK(vp);
2590 	vgonel(vp);
2591 	VI_UNLOCK(vp);
2592 }
2593 
2594 /*
2595  * vgone, with the vp interlock held.
2596  */
2597 void
2598 vgonel(struct vnode *vp)
2599 {
2600 	struct thread *td;
2601 	int oweinact;
2602 	int active;
2603 	struct mount *mp;
2604 
2605 	ASSERT_VOP_ELOCKED(vp, "vgonel");
2606 	ASSERT_VI_LOCKED(vp, "vgonel");
2607 	VNASSERT(vp->v_holdcnt, vp,
2608 	    ("vgonel: vp %p has no reference.", vp));
2609 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2610 	td = curthread;
2611 
2612 	/*
2613 	 * Don't vgonel if we're already doomed.
2614 	 */
2615 	if (vp->v_iflag & VI_DOOMED)
2616 		return;
2617 	vp->v_iflag |= VI_DOOMED;
2618 	/*
2619 	 * Check to see if the vnode is in use.  If so, we have to call
2620 	 * VOP_CLOSE() and VOP_INACTIVE().
2621 	 */
2622 	active = vp->v_usecount;
2623 	oweinact = (vp->v_iflag & VI_OWEINACT);
2624 	VI_UNLOCK(vp);
2625 	/*
2626 	 * Clean out any buffers associated with the vnode.
2627 	 * If the flush fails, just toss the buffers.
2628 	 */
2629 	mp = NULL;
2630 	if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd))
2631 		(void) vn_start_secondary_write(vp, &mp, V_WAIT);
2632 	if (vinvalbuf(vp, V_SAVE, 0, 0) != 0)
2633 		vinvalbuf(vp, 0, 0, 0);
2634 
2635 	/*
2636 	 * If purging an active vnode, it must be closed and
2637 	 * deactivated before being reclaimed.
2638 	 */
2639 	if (active)
2640 		VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
2641 	if (oweinact || active) {
2642 		VI_LOCK(vp);
2643 		if ((vp->v_iflag & VI_DOINGINACT) == 0)
2644 			vinactive(vp, td);
2645 		VI_UNLOCK(vp);
2646 	}
2647 	/*
2648 	 * Reclaim the vnode.
2649 	 */
2650 	if (VOP_RECLAIM(vp, td))
2651 		panic("vgone: cannot reclaim");
2652 	if (mp != NULL)
2653 		vn_finished_secondary_write(mp);
2654 	VNASSERT(vp->v_object == NULL, vp,
2655 	    ("vop_reclaim left v_object vp=%p, tag=%s", vp, vp->v_tag));
2656 	/*
2657 	 * Clear the advisory locks and wake up waiting threads.
2658 	 */
2659 	(void)VOP_ADVLOCKPURGE(vp);
2660 	/*
2661 	 * Delete from old mount point vnode list.
2662 	 */
2663 	delmntque(vp);
2664 	cache_purge(vp);
2665 	/*
2666 	 * Done with purge, reset to the standard lock and invalidate
2667 	 * the vnode.
2668 	 */
2669 	VI_LOCK(vp);
2670 	vp->v_vnlock = &vp->v_lock;
2671 	vp->v_op = &dead_vnodeops;
2672 	vp->v_tag = "none";
2673 	vp->v_type = VBAD;
2674 }
2675 
2676 /*
2677  * Calculate the total number of references to a special device.
2678  */
2679 int
2680 vcount(struct vnode *vp)
2681 {
2682 	int count;
2683 
2684 	dev_lock();
2685 	count = vp->v_rdev->si_usecount;
2686 	dev_unlock();
2687 	return (count);
2688 }
2689 
2690 /*
2691  * Same as above, but using the struct cdev *as argument
2692  */
2693 int
2694 count_dev(struct cdev *dev)
2695 {
2696 	int count;
2697 
2698 	dev_lock();
2699 	count = dev->si_usecount;
2700 	dev_unlock();
2701 	return(count);
2702 }
2703 
2704 /*
2705  * Print out a description of a vnode.
2706  */
2707 static char *typename[] =
2708 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD",
2709  "VMARKER"};
2710 
2711 void
2712 vn_printf(struct vnode *vp, const char *fmt, ...)
2713 {
2714 	va_list ap;
2715 	char buf[256], buf2[16];
2716 	u_long flags;
2717 
2718 	va_start(ap, fmt);
2719 	vprintf(fmt, ap);
2720 	va_end(ap);
2721 	printf("%p: ", (void *)vp);
2722 	printf("tag %s, type %s\n", vp->v_tag, typename[vp->v_type]);
2723 	printf("    usecount %d, writecount %d, refcount %d mountedhere %p\n",
2724 	    vp->v_usecount, vp->v_writecount, vp->v_holdcnt, vp->v_mountedhere);
2725 	buf[0] = '\0';
2726 	buf[1] = '\0';
2727 	if (vp->v_vflag & VV_ROOT)
2728 		strlcat(buf, "|VV_ROOT", sizeof(buf));
2729 	if (vp->v_vflag & VV_ISTTY)
2730 		strlcat(buf, "|VV_ISTTY", sizeof(buf));
2731 	if (vp->v_vflag & VV_NOSYNC)
2732 		strlcat(buf, "|VV_NOSYNC", sizeof(buf));
2733 	if (vp->v_vflag & VV_CACHEDLABEL)
2734 		strlcat(buf, "|VV_CACHEDLABEL", sizeof(buf));
2735 	if (vp->v_vflag & VV_TEXT)
2736 		strlcat(buf, "|VV_TEXT", sizeof(buf));
2737 	if (vp->v_vflag & VV_COPYONWRITE)
2738 		strlcat(buf, "|VV_COPYONWRITE", sizeof(buf));
2739 	if (vp->v_vflag & VV_SYSTEM)
2740 		strlcat(buf, "|VV_SYSTEM", sizeof(buf));
2741 	if (vp->v_vflag & VV_PROCDEP)
2742 		strlcat(buf, "|VV_PROCDEP", sizeof(buf));
2743 	if (vp->v_vflag & VV_NOKNOTE)
2744 		strlcat(buf, "|VV_NOKNOTE", sizeof(buf));
2745 	if (vp->v_vflag & VV_DELETED)
2746 		strlcat(buf, "|VV_DELETED", sizeof(buf));
2747 	if (vp->v_vflag & VV_MD)
2748 		strlcat(buf, "|VV_MD", sizeof(buf));
2749 	flags = vp->v_vflag & ~(VV_ROOT | VV_ISTTY | VV_NOSYNC |
2750 	    VV_CACHEDLABEL | VV_TEXT | VV_COPYONWRITE | VV_SYSTEM | VV_PROCDEP |
2751 	    VV_NOKNOTE | VV_DELETED | VV_MD);
2752 	if (flags != 0) {
2753 		snprintf(buf2, sizeof(buf2), "|VV(0x%lx)", flags);
2754 		strlcat(buf, buf2, sizeof(buf));
2755 	}
2756 	if (vp->v_iflag & VI_MOUNT)
2757 		strlcat(buf, "|VI_MOUNT", sizeof(buf));
2758 	if (vp->v_iflag & VI_AGE)
2759 		strlcat(buf, "|VI_AGE", sizeof(buf));
2760 	if (vp->v_iflag & VI_DOOMED)
2761 		strlcat(buf, "|VI_DOOMED", sizeof(buf));
2762 	if (vp->v_iflag & VI_FREE)
2763 		strlcat(buf, "|VI_FREE", sizeof(buf));
2764 	if (vp->v_iflag & VI_DOINGINACT)
2765 		strlcat(buf, "|VI_DOINGINACT", sizeof(buf));
2766 	if (vp->v_iflag & VI_OWEINACT)
2767 		strlcat(buf, "|VI_OWEINACT", sizeof(buf));
2768 	flags = vp->v_iflag & ~(VI_MOUNT | VI_AGE | VI_DOOMED | VI_FREE |
2769 	    VI_DOINGINACT | VI_OWEINACT);
2770 	if (flags != 0) {
2771 		snprintf(buf2, sizeof(buf2), "|VI(0x%lx)", flags);
2772 		strlcat(buf, buf2, sizeof(buf));
2773 	}
2774 	printf("    flags (%s)\n", buf + 1);
2775 	if (mtx_owned(VI_MTX(vp)))
2776 		printf(" VI_LOCKed");
2777 	if (vp->v_object != NULL)
2778 		printf("    v_object %p ref %d pages %d\n",
2779 		    vp->v_object, vp->v_object->ref_count,
2780 		    vp->v_object->resident_page_count);
2781 	printf("    ");
2782 	lockmgr_printinfo(vp->v_vnlock);
2783 	if (vp->v_data != NULL)
2784 		VOP_PRINT(vp);
2785 }
2786 
2787 #ifdef DDB
2788 /*
2789  * List all of the locked vnodes in the system.
2790  * Called when debugging the kernel.
2791  */
2792 DB_SHOW_COMMAND(lockedvnods, lockedvnodes)
2793 {
2794 	struct mount *mp, *nmp;
2795 	struct vnode *vp;
2796 
2797 	/*
2798 	 * Note: because this is DDB, we can't obey the locking semantics
2799 	 * for these structures, which means we could catch an inconsistent
2800 	 * state and dereference a nasty pointer.  Not much to be done
2801 	 * about that.
2802 	 */
2803 	db_printf("Locked vnodes\n");
2804 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2805 		nmp = TAILQ_NEXT(mp, mnt_list);
2806 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2807 			if (vp->v_type != VMARKER &&
2808 			    VOP_ISLOCKED(vp))
2809 				vprint("", vp);
2810 		}
2811 		nmp = TAILQ_NEXT(mp, mnt_list);
2812 	}
2813 }
2814 
2815 /*
2816  * Show details about the given vnode.
2817  */
2818 DB_SHOW_COMMAND(vnode, db_show_vnode)
2819 {
2820 	struct vnode *vp;
2821 
2822 	if (!have_addr)
2823 		return;
2824 	vp = (struct vnode *)addr;
2825 	vn_printf(vp, "vnode ");
2826 }
2827 
2828 /*
2829  * Show details about the given mount point.
2830  */
2831 DB_SHOW_COMMAND(mount, db_show_mount)
2832 {
2833 	struct mount *mp;
2834 	struct vfsopt *opt;
2835 	struct statfs *sp;
2836 	struct vnode *vp;
2837 	char buf[512];
2838 	u_int flags;
2839 
2840 	if (!have_addr) {
2841 		/* No address given, print short info about all mount points. */
2842 		TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2843 			db_printf("%p %s on %s (%s)\n", mp,
2844 			    mp->mnt_stat.f_mntfromname,
2845 			    mp->mnt_stat.f_mntonname,
2846 			    mp->mnt_stat.f_fstypename);
2847 			if (db_pager_quit)
2848 				break;
2849 		}
2850 		db_printf("\nMore info: show mount <addr>\n");
2851 		return;
2852 	}
2853 
2854 	mp = (struct mount *)addr;
2855 	db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname,
2856 	    mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename);
2857 
2858 	buf[0] = '\0';
2859 	flags = mp->mnt_flag;
2860 #define	MNT_FLAG(flag)	do {						\
2861 	if (flags & (flag)) {						\
2862 		if (buf[0] != '\0')					\
2863 			strlcat(buf, ", ", sizeof(buf));		\
2864 		strlcat(buf, (#flag) + 4, sizeof(buf));			\
2865 		flags &= ~(flag);					\
2866 	}								\
2867 } while (0)
2868 	MNT_FLAG(MNT_RDONLY);
2869 	MNT_FLAG(MNT_SYNCHRONOUS);
2870 	MNT_FLAG(MNT_NOEXEC);
2871 	MNT_FLAG(MNT_NOSUID);
2872 	MNT_FLAG(MNT_UNION);
2873 	MNT_FLAG(MNT_ASYNC);
2874 	MNT_FLAG(MNT_SUIDDIR);
2875 	MNT_FLAG(MNT_SOFTDEP);
2876 	MNT_FLAG(MNT_SUJ);
2877 	MNT_FLAG(MNT_NOSYMFOLLOW);
2878 	MNT_FLAG(MNT_GJOURNAL);
2879 	MNT_FLAG(MNT_MULTILABEL);
2880 	MNT_FLAG(MNT_ACLS);
2881 	MNT_FLAG(MNT_NOATIME);
2882 	MNT_FLAG(MNT_NOCLUSTERR);
2883 	MNT_FLAG(MNT_NOCLUSTERW);
2884 	MNT_FLAG(MNT_NFS4ACLS);
2885 	MNT_FLAG(MNT_EXRDONLY);
2886 	MNT_FLAG(MNT_EXPORTED);
2887 	MNT_FLAG(MNT_DEFEXPORTED);
2888 	MNT_FLAG(MNT_EXPORTANON);
2889 	MNT_FLAG(MNT_EXKERB);
2890 	MNT_FLAG(MNT_EXPUBLIC);
2891 	MNT_FLAG(MNT_LOCAL);
2892 	MNT_FLAG(MNT_QUOTA);
2893 	MNT_FLAG(MNT_ROOTFS);
2894 	MNT_FLAG(MNT_USER);
2895 	MNT_FLAG(MNT_IGNORE);
2896 	MNT_FLAG(MNT_UPDATE);
2897 	MNT_FLAG(MNT_DELEXPORT);
2898 	MNT_FLAG(MNT_RELOAD);
2899 	MNT_FLAG(MNT_FORCE);
2900 	MNT_FLAG(MNT_SNAPSHOT);
2901 	MNT_FLAG(MNT_BYFSID);
2902 #undef MNT_FLAG
2903 	if (flags != 0) {
2904 		if (buf[0] != '\0')
2905 			strlcat(buf, ", ", sizeof(buf));
2906 		snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
2907 		    "0x%08x", flags);
2908 	}
2909 	db_printf("    mnt_flag = %s\n", buf);
2910 
2911 	buf[0] = '\0';
2912 	flags = mp->mnt_kern_flag;
2913 #define	MNT_KERN_FLAG(flag)	do {					\
2914 	if (flags & (flag)) {						\
2915 		if (buf[0] != '\0')					\
2916 			strlcat(buf, ", ", sizeof(buf));		\
2917 		strlcat(buf, (#flag) + 5, sizeof(buf));			\
2918 		flags &= ~(flag);					\
2919 	}								\
2920 } while (0)
2921 	MNT_KERN_FLAG(MNTK_UNMOUNTF);
2922 	MNT_KERN_FLAG(MNTK_ASYNC);
2923 	MNT_KERN_FLAG(MNTK_SOFTDEP);
2924 	MNT_KERN_FLAG(MNTK_NOINSMNTQ);
2925 	MNT_KERN_FLAG(MNTK_DRAINING);
2926 	MNT_KERN_FLAG(MNTK_REFEXPIRE);
2927 	MNT_KERN_FLAG(MNTK_EXTENDED_SHARED);
2928 	MNT_KERN_FLAG(MNTK_SHARED_WRITES);
2929 	MNT_KERN_FLAG(MNTK_UNMOUNT);
2930 	MNT_KERN_FLAG(MNTK_MWAIT);
2931 	MNT_KERN_FLAG(MNTK_SUSPEND);
2932 	MNT_KERN_FLAG(MNTK_SUSPEND2);
2933 	MNT_KERN_FLAG(MNTK_SUSPENDED);
2934 	MNT_KERN_FLAG(MNTK_MPSAFE);
2935 	MNT_KERN_FLAG(MNTK_LOOKUP_SHARED);
2936 	MNT_KERN_FLAG(MNTK_NOKNOTE);
2937 #undef MNT_KERN_FLAG
2938 	if (flags != 0) {
2939 		if (buf[0] != '\0')
2940 			strlcat(buf, ", ", sizeof(buf));
2941 		snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
2942 		    "0x%08x", flags);
2943 	}
2944 	db_printf("    mnt_kern_flag = %s\n", buf);
2945 
2946 	db_printf("    mnt_opt = ");
2947 	opt = TAILQ_FIRST(mp->mnt_opt);
2948 	if (opt != NULL) {
2949 		db_printf("%s", opt->name);
2950 		opt = TAILQ_NEXT(opt, link);
2951 		while (opt != NULL) {
2952 			db_printf(", %s", opt->name);
2953 			opt = TAILQ_NEXT(opt, link);
2954 		}
2955 	}
2956 	db_printf("\n");
2957 
2958 	sp = &mp->mnt_stat;
2959 	db_printf("    mnt_stat = { version=%u type=%u flags=0x%016jx "
2960 	    "bsize=%ju iosize=%ju blocks=%ju bfree=%ju bavail=%jd files=%ju "
2961 	    "ffree=%jd syncwrites=%ju asyncwrites=%ju syncreads=%ju "
2962 	    "asyncreads=%ju namemax=%u owner=%u fsid=[%d, %d] }\n",
2963 	    (u_int)sp->f_version, (u_int)sp->f_type, (uintmax_t)sp->f_flags,
2964 	    (uintmax_t)sp->f_bsize, (uintmax_t)sp->f_iosize,
2965 	    (uintmax_t)sp->f_blocks, (uintmax_t)sp->f_bfree,
2966 	    (intmax_t)sp->f_bavail, (uintmax_t)sp->f_files,
2967 	    (intmax_t)sp->f_ffree, (uintmax_t)sp->f_syncwrites,
2968 	    (uintmax_t)sp->f_asyncwrites, (uintmax_t)sp->f_syncreads,
2969 	    (uintmax_t)sp->f_asyncreads, (u_int)sp->f_namemax,
2970 	    (u_int)sp->f_owner, (int)sp->f_fsid.val[0], (int)sp->f_fsid.val[1]);
2971 
2972 	db_printf("    mnt_cred = { uid=%u ruid=%u",
2973 	    (u_int)mp->mnt_cred->cr_uid, (u_int)mp->mnt_cred->cr_ruid);
2974 	if (jailed(mp->mnt_cred))
2975 		db_printf(", jail=%d", mp->mnt_cred->cr_prison->pr_id);
2976 	db_printf(" }\n");
2977 	db_printf("    mnt_ref = %d\n", mp->mnt_ref);
2978 	db_printf("    mnt_gen = %d\n", mp->mnt_gen);
2979 	db_printf("    mnt_nvnodelistsize = %d\n", mp->mnt_nvnodelistsize);
2980 	db_printf("    mnt_writeopcount = %d\n", mp->mnt_writeopcount);
2981 	db_printf("    mnt_noasync = %u\n", mp->mnt_noasync);
2982 	db_printf("    mnt_maxsymlinklen = %d\n", mp->mnt_maxsymlinklen);
2983 	db_printf("    mnt_iosize_max = %d\n", mp->mnt_iosize_max);
2984 	db_printf("    mnt_hashseed = %u\n", mp->mnt_hashseed);
2985 	db_printf("    mnt_secondary_writes = %d\n", mp->mnt_secondary_writes);
2986 	db_printf("    mnt_secondary_accwrites = %d\n",
2987 	    mp->mnt_secondary_accwrites);
2988 	db_printf("    mnt_gjprovider = %s\n",
2989 	    mp->mnt_gjprovider != NULL ? mp->mnt_gjprovider : "NULL");
2990 	db_printf("\n");
2991 
2992 	TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2993 		if (vp->v_type != VMARKER) {
2994 			vn_printf(vp, "vnode ");
2995 			if (db_pager_quit)
2996 				break;
2997 		}
2998 	}
2999 }
3000 #endif	/* DDB */
3001 
3002 /*
3003  * Fill in a struct xvfsconf based on a struct vfsconf.
3004  */
3005 static void
3006 vfsconf2x(struct vfsconf *vfsp, struct xvfsconf *xvfsp)
3007 {
3008 
3009 	strcpy(xvfsp->vfc_name, vfsp->vfc_name);
3010 	xvfsp->vfc_typenum = vfsp->vfc_typenum;
3011 	xvfsp->vfc_refcount = vfsp->vfc_refcount;
3012 	xvfsp->vfc_flags = vfsp->vfc_flags;
3013 	/*
3014 	 * These are unused in userland, we keep them
3015 	 * to not break binary compatibility.
3016 	 */
3017 	xvfsp->vfc_vfsops = NULL;
3018 	xvfsp->vfc_next = NULL;
3019 }
3020 
3021 /*
3022  * Top level filesystem related information gathering.
3023  */
3024 static int
3025 sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS)
3026 {
3027 	struct vfsconf *vfsp;
3028 	struct xvfsconf xvfsp;
3029 	int error;
3030 
3031 	error = 0;
3032 	TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3033 		bzero(&xvfsp, sizeof(xvfsp));
3034 		vfsconf2x(vfsp, &xvfsp);
3035 		error = SYSCTL_OUT(req, &xvfsp, sizeof xvfsp);
3036 		if (error)
3037 			break;
3038 	}
3039 	return (error);
3040 }
3041 
3042 SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLTYPE_OPAQUE | CTLFLAG_RD,
3043     NULL, 0, sysctl_vfs_conflist,
3044     "S,xvfsconf", "List of all configured filesystems");
3045 
3046 #ifndef BURN_BRIDGES
3047 static int	sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
3048 
3049 static int
3050 vfs_sysctl(SYSCTL_HANDLER_ARGS)
3051 {
3052 	int *name = (int *)arg1 - 1;	/* XXX */
3053 	u_int namelen = arg2 + 1;	/* XXX */
3054 	struct vfsconf *vfsp;
3055 	struct xvfsconf xvfsp;
3056 
3057 	log(LOG_WARNING, "userland calling deprecated sysctl, "
3058 	    "please rebuild world\n");
3059 
3060 #if 1 || defined(COMPAT_PRELITE2)
3061 	/* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
3062 	if (namelen == 1)
3063 		return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
3064 #endif
3065 
3066 	switch (name[1]) {
3067 	case VFS_MAXTYPENUM:
3068 		if (namelen != 2)
3069 			return (ENOTDIR);
3070 		return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
3071 	case VFS_CONF:
3072 		if (namelen != 3)
3073 			return (ENOTDIR);	/* overloaded */
3074 		TAILQ_FOREACH(vfsp, &vfsconf, vfc_list)
3075 			if (vfsp->vfc_typenum == name[2])
3076 				break;
3077 		if (vfsp == NULL)
3078 			return (EOPNOTSUPP);
3079 		bzero(&xvfsp, sizeof(xvfsp));
3080 		vfsconf2x(vfsp, &xvfsp);
3081 		return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
3082 	}
3083 	return (EOPNOTSUPP);
3084 }
3085 
3086 static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP,
3087     vfs_sysctl, "Generic filesystem");
3088 
3089 #if 1 || defined(COMPAT_PRELITE2)
3090 
3091 static int
3092 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
3093 {
3094 	int error;
3095 	struct vfsconf *vfsp;
3096 	struct ovfsconf ovfs;
3097 
3098 	TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3099 		bzero(&ovfs, sizeof(ovfs));
3100 		ovfs.vfc_vfsops = vfsp->vfc_vfsops;	/* XXX used as flag */
3101 		strcpy(ovfs.vfc_name, vfsp->vfc_name);
3102 		ovfs.vfc_index = vfsp->vfc_typenum;
3103 		ovfs.vfc_refcount = vfsp->vfc_refcount;
3104 		ovfs.vfc_flags = vfsp->vfc_flags;
3105 		error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
3106 		if (error)
3107 			return error;
3108 	}
3109 	return 0;
3110 }
3111 
3112 #endif /* 1 || COMPAT_PRELITE2 */
3113 #endif /* !BURN_BRIDGES */
3114 
3115 #define KINFO_VNODESLOP		10
3116 #ifdef notyet
3117 /*
3118  * Dump vnode list (via sysctl).
3119  */
3120 /* ARGSUSED */
3121 static int
3122 sysctl_vnode(SYSCTL_HANDLER_ARGS)
3123 {
3124 	struct xvnode *xvn;
3125 	struct mount *mp;
3126 	struct vnode *vp;
3127 	int error, len, n;
3128 
3129 	/*
3130 	 * Stale numvnodes access is not fatal here.
3131 	 */
3132 	req->lock = 0;
3133 	len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn;
3134 	if (!req->oldptr)
3135 		/* Make an estimate */
3136 		return (SYSCTL_OUT(req, 0, len));
3137 
3138 	error = sysctl_wire_old_buffer(req, 0);
3139 	if (error != 0)
3140 		return (error);
3141 	xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK);
3142 	n = 0;
3143 	mtx_lock(&mountlist_mtx);
3144 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3145 		if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK))
3146 			continue;
3147 		MNT_ILOCK(mp);
3148 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3149 			if (n == len)
3150 				break;
3151 			vref(vp);
3152 			xvn[n].xv_size = sizeof *xvn;
3153 			xvn[n].xv_vnode = vp;
3154 			xvn[n].xv_id = 0;	/* XXX compat */
3155 #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field
3156 			XV_COPY(usecount);
3157 			XV_COPY(writecount);
3158 			XV_COPY(holdcnt);
3159 			XV_COPY(mount);
3160 			XV_COPY(numoutput);
3161 			XV_COPY(type);
3162 #undef XV_COPY
3163 			xvn[n].xv_flag = vp->v_vflag;
3164 
3165 			switch (vp->v_type) {
3166 			case VREG:
3167 			case VDIR:
3168 			case VLNK:
3169 				break;
3170 			case VBLK:
3171 			case VCHR:
3172 				if (vp->v_rdev == NULL) {
3173 					vrele(vp);
3174 					continue;
3175 				}
3176 				xvn[n].xv_dev = dev2udev(vp->v_rdev);
3177 				break;
3178 			case VSOCK:
3179 				xvn[n].xv_socket = vp->v_socket;
3180 				break;
3181 			case VFIFO:
3182 				xvn[n].xv_fifo = vp->v_fifoinfo;
3183 				break;
3184 			case VNON:
3185 			case VBAD:
3186 			default:
3187 				/* shouldn't happen? */
3188 				vrele(vp);
3189 				continue;
3190 			}
3191 			vrele(vp);
3192 			++n;
3193 		}
3194 		MNT_IUNLOCK(mp);
3195 		mtx_lock(&mountlist_mtx);
3196 		vfs_unbusy(mp);
3197 		if (n == len)
3198 			break;
3199 	}
3200 	mtx_unlock(&mountlist_mtx);
3201 
3202 	error = SYSCTL_OUT(req, xvn, n * sizeof *xvn);
3203 	free(xvn, M_TEMP);
3204 	return (error);
3205 }
3206 
3207 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
3208     0, 0, sysctl_vnode, "S,xvnode", "");
3209 #endif
3210 
3211 /*
3212  * Unmount all filesystems. The list is traversed in reverse order
3213  * of mounting to avoid dependencies.
3214  */
3215 void
3216 vfs_unmountall(void)
3217 {
3218 	struct mount *mp;
3219 	struct thread *td;
3220 	int error;
3221 
3222 	KASSERT(curthread != NULL, ("vfs_unmountall: NULL curthread"));
3223 	CTR1(KTR_VFS, "%s: unmounting all filesystems", __func__);
3224 	td = curthread;
3225 
3226 	/*
3227 	 * Since this only runs when rebooting, it is not interlocked.
3228 	 */
3229 	while(!TAILQ_EMPTY(&mountlist)) {
3230 		mp = TAILQ_LAST(&mountlist, mntlist);
3231 		error = dounmount(mp, MNT_FORCE, td);
3232 		if (error) {
3233 			TAILQ_REMOVE(&mountlist, mp, mnt_list);
3234 			/*
3235 			 * XXX: Due to the way in which we mount the root
3236 			 * file system off of devfs, devfs will generate a
3237 			 * "busy" warning when we try to unmount it before
3238 			 * the root.  Don't print a warning as a result in
3239 			 * order to avoid false positive errors that may
3240 			 * cause needless upset.
3241 			 */
3242 			if (strcmp(mp->mnt_vfc->vfc_name, "devfs") != 0) {
3243 				printf("unmount of %s failed (",
3244 				    mp->mnt_stat.f_mntonname);
3245 				if (error == EBUSY)
3246 					printf("BUSY)\n");
3247 				else
3248 					printf("%d)\n", error);
3249 			}
3250 		} else {
3251 			/* The unmount has removed mp from the mountlist */
3252 		}
3253 	}
3254 }
3255 
3256 /*
3257  * perform msync on all vnodes under a mount point
3258  * the mount point must be locked.
3259  */
3260 void
3261 vfs_msync(struct mount *mp, int flags)
3262 {
3263 	struct vnode *vp, *mvp;
3264 	struct vm_object *obj;
3265 
3266 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
3267 	MNT_ILOCK(mp);
3268 	MNT_VNODE_FOREACH(vp, mp, mvp) {
3269 		VI_LOCK(vp);
3270 		obj = vp->v_object;
3271 		if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0 &&
3272 		    (flags == MNT_WAIT || VOP_ISLOCKED(vp) == 0)) {
3273 			MNT_IUNLOCK(mp);
3274 			if (!vget(vp,
3275 			    LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
3276 			    curthread)) {
3277 				if (vp->v_vflag & VV_NOSYNC) {	/* unlinked */
3278 					vput(vp);
3279 					MNT_ILOCK(mp);
3280 					continue;
3281 				}
3282 
3283 				obj = vp->v_object;
3284 				if (obj != NULL) {
3285 					VM_OBJECT_LOCK(obj);
3286 					vm_object_page_clean(obj, 0, 0,
3287 					    flags == MNT_WAIT ?
3288 					    OBJPC_SYNC : OBJPC_NOSYNC);
3289 					VM_OBJECT_UNLOCK(obj);
3290 				}
3291 				vput(vp);
3292 			}
3293 			MNT_ILOCK(mp);
3294 		} else
3295 			VI_UNLOCK(vp);
3296 	}
3297 	MNT_IUNLOCK(mp);
3298 }
3299 
3300 /*
3301  * Mark a vnode as free, putting it up for recycling.
3302  */
3303 static void
3304 vfree(struct vnode *vp)
3305 {
3306 
3307 	ASSERT_VI_LOCKED(vp, "vfree");
3308 	mtx_lock(&vnode_free_list_mtx);
3309 	VNASSERT(vp->v_op != NULL, vp, ("vfree: vnode already reclaimed."));
3310 	VNASSERT((vp->v_iflag & VI_FREE) == 0, vp, ("vnode already free"));
3311 	VNASSERT(VSHOULDFREE(vp), vp, ("vfree: freeing when we shouldn't"));
3312 	VNASSERT((vp->v_iflag & VI_DOOMED) == 0, vp,
3313 	    ("vfree: Freeing doomed vnode"));
3314 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
3315 	if (vp->v_iflag & VI_AGE) {
3316 		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
3317 	} else {
3318 		TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
3319 	}
3320 	freevnodes++;
3321 	vp->v_iflag &= ~VI_AGE;
3322 	vp->v_iflag |= VI_FREE;
3323 	mtx_unlock(&vnode_free_list_mtx);
3324 }
3325 
3326 /*
3327  * Opposite of vfree() - mark a vnode as in use.
3328  */
3329 static void
3330 vbusy(struct vnode *vp)
3331 {
3332 	ASSERT_VI_LOCKED(vp, "vbusy");
3333 	VNASSERT((vp->v_iflag & VI_FREE) != 0, vp, ("vnode not free"));
3334 	VNASSERT(vp->v_op != NULL, vp, ("vbusy: vnode already reclaimed."));
3335 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
3336 
3337 	mtx_lock(&vnode_free_list_mtx);
3338 	TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
3339 	freevnodes--;
3340 	vp->v_iflag &= ~(VI_FREE|VI_AGE);
3341 	mtx_unlock(&vnode_free_list_mtx);
3342 }
3343 
3344 static void
3345 destroy_vpollinfo(struct vpollinfo *vi)
3346 {
3347 	seldrain(&vi->vpi_selinfo);
3348 	knlist_destroy(&vi->vpi_selinfo.si_note);
3349 	mtx_destroy(&vi->vpi_lock);
3350 	uma_zfree(vnodepoll_zone, vi);
3351 }
3352 
3353 /*
3354  * Initalize per-vnode helper structure to hold poll-related state.
3355  */
3356 void
3357 v_addpollinfo(struct vnode *vp)
3358 {
3359 	struct vpollinfo *vi;
3360 
3361 	if (vp->v_pollinfo != NULL)
3362 		return;
3363 	vi = uma_zalloc(vnodepoll_zone, M_WAITOK);
3364 	mtx_init(&vi->vpi_lock, "vnode pollinfo", NULL, MTX_DEF);
3365 	knlist_init(&vi->vpi_selinfo.si_note, vp, vfs_knllock,
3366 	    vfs_knlunlock, vfs_knl_assert_locked, vfs_knl_assert_unlocked);
3367 	VI_LOCK(vp);
3368 	if (vp->v_pollinfo != NULL) {
3369 		VI_UNLOCK(vp);
3370 		destroy_vpollinfo(vi);
3371 		return;
3372 	}
3373 	vp->v_pollinfo = vi;
3374 	VI_UNLOCK(vp);
3375 }
3376 
3377 /*
3378  * Record a process's interest in events which might happen to
3379  * a vnode.  Because poll uses the historic select-style interface
3380  * internally, this routine serves as both the ``check for any
3381  * pending events'' and the ``record my interest in future events''
3382  * functions.  (These are done together, while the lock is held,
3383  * to avoid race conditions.)
3384  */
3385 int
3386 vn_pollrecord(struct vnode *vp, struct thread *td, int events)
3387 {
3388 
3389 	v_addpollinfo(vp);
3390 	mtx_lock(&vp->v_pollinfo->vpi_lock);
3391 	if (vp->v_pollinfo->vpi_revents & events) {
3392 		/*
3393 		 * This leaves events we are not interested
3394 		 * in available for the other process which
3395 		 * which presumably had requested them
3396 		 * (otherwise they would never have been
3397 		 * recorded).
3398 		 */
3399 		events &= vp->v_pollinfo->vpi_revents;
3400 		vp->v_pollinfo->vpi_revents &= ~events;
3401 
3402 		mtx_unlock(&vp->v_pollinfo->vpi_lock);
3403 		return (events);
3404 	}
3405 	vp->v_pollinfo->vpi_events |= events;
3406 	selrecord(td, &vp->v_pollinfo->vpi_selinfo);
3407 	mtx_unlock(&vp->v_pollinfo->vpi_lock);
3408 	return (0);
3409 }
3410 
3411 /*
3412  * Routine to create and manage a filesystem syncer vnode.
3413  */
3414 #define sync_close ((int (*)(struct  vop_close_args *))nullop)
3415 static int	sync_fsync(struct  vop_fsync_args *);
3416 static int	sync_inactive(struct  vop_inactive_args *);
3417 static int	sync_reclaim(struct  vop_reclaim_args *);
3418 
3419 static struct vop_vector sync_vnodeops = {
3420 	.vop_bypass =	VOP_EOPNOTSUPP,
3421 	.vop_close =	sync_close,		/* close */
3422 	.vop_fsync =	sync_fsync,		/* fsync */
3423 	.vop_inactive =	sync_inactive,	/* inactive */
3424 	.vop_reclaim =	sync_reclaim,	/* reclaim */
3425 	.vop_lock1 =	vop_stdlock,	/* lock */
3426 	.vop_unlock =	vop_stdunlock,	/* unlock */
3427 	.vop_islocked =	vop_stdislocked,	/* islocked */
3428 };
3429 
3430 /*
3431  * Create a new filesystem syncer vnode for the specified mount point.
3432  */
3433 void
3434 vfs_allocate_syncvnode(struct mount *mp)
3435 {
3436 	struct vnode *vp;
3437 	struct bufobj *bo;
3438 	static long start, incr, next;
3439 	int error;
3440 
3441 	/* Allocate a new vnode */
3442 	error = getnewvnode("syncer", mp, &sync_vnodeops, &vp);
3443 	if (error != 0)
3444 		panic("vfs_allocate_syncvnode: getnewvnode() failed");
3445 	vp->v_type = VNON;
3446 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3447 	vp->v_vflag |= VV_FORCEINSMQ;
3448 	error = insmntque(vp, mp);
3449 	if (error != 0)
3450 		panic("vfs_allocate_syncvnode: insmntque() failed");
3451 	vp->v_vflag &= ~VV_FORCEINSMQ;
3452 	VOP_UNLOCK(vp, 0);
3453 	/*
3454 	 * Place the vnode onto the syncer worklist. We attempt to
3455 	 * scatter them about on the list so that they will go off
3456 	 * at evenly distributed times even if all the filesystems
3457 	 * are mounted at once.
3458 	 */
3459 	next += incr;
3460 	if (next == 0 || next > syncer_maxdelay) {
3461 		start /= 2;
3462 		incr /= 2;
3463 		if (start == 0) {
3464 			start = syncer_maxdelay / 2;
3465 			incr = syncer_maxdelay;
3466 		}
3467 		next = start;
3468 	}
3469 	bo = &vp->v_bufobj;
3470 	BO_LOCK(bo);
3471 	vn_syncer_add_to_worklist(bo, syncdelay > 0 ? next % syncdelay : 0);
3472 	/* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */
3473 	mtx_lock(&sync_mtx);
3474 	sync_vnode_count++;
3475 	if (mp->mnt_syncer == NULL) {
3476 		mp->mnt_syncer = vp;
3477 		vp = NULL;
3478 	}
3479 	mtx_unlock(&sync_mtx);
3480 	BO_UNLOCK(bo);
3481 	if (vp != NULL) {
3482 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3483 		vgone(vp);
3484 		vput(vp);
3485 	}
3486 }
3487 
3488 void
3489 vfs_deallocate_syncvnode(struct mount *mp)
3490 {
3491 	struct vnode *vp;
3492 
3493 	mtx_lock(&sync_mtx);
3494 	vp = mp->mnt_syncer;
3495 	if (vp != NULL)
3496 		mp->mnt_syncer = NULL;
3497 	mtx_unlock(&sync_mtx);
3498 	if (vp != NULL)
3499 		vrele(vp);
3500 }
3501 
3502 /*
3503  * Do a lazy sync of the filesystem.
3504  */
3505 static int
3506 sync_fsync(struct vop_fsync_args *ap)
3507 {
3508 	struct vnode *syncvp = ap->a_vp;
3509 	struct mount *mp = syncvp->v_mount;
3510 	int error;
3511 	struct bufobj *bo;
3512 
3513 	/*
3514 	 * We only need to do something if this is a lazy evaluation.
3515 	 */
3516 	if (ap->a_waitfor != MNT_LAZY)
3517 		return (0);
3518 
3519 	/*
3520 	 * Move ourselves to the back of the sync list.
3521 	 */
3522 	bo = &syncvp->v_bufobj;
3523 	BO_LOCK(bo);
3524 	vn_syncer_add_to_worklist(bo, syncdelay);
3525 	BO_UNLOCK(bo);
3526 
3527 	/*
3528 	 * Walk the list of vnodes pushing all that are dirty and
3529 	 * not already on the sync list.
3530 	 */
3531 	mtx_lock(&mountlist_mtx);
3532 	if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK) != 0) {
3533 		mtx_unlock(&mountlist_mtx);
3534 		return (0);
3535 	}
3536 	if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
3537 		vfs_unbusy(mp);
3538 		return (0);
3539 	}
3540 	MNT_ILOCK(mp);
3541 	mp->mnt_noasync++;
3542 	mp->mnt_kern_flag &= ~MNTK_ASYNC;
3543 	MNT_IUNLOCK(mp);
3544 	vfs_msync(mp, MNT_NOWAIT);
3545 	error = VFS_SYNC(mp, MNT_LAZY);
3546 	MNT_ILOCK(mp);
3547 	mp->mnt_noasync--;
3548 	if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0)
3549 		mp->mnt_kern_flag |= MNTK_ASYNC;
3550 	MNT_IUNLOCK(mp);
3551 	vn_finished_write(mp);
3552 	vfs_unbusy(mp);
3553 	return (error);
3554 }
3555 
3556 /*
3557  * The syncer vnode is no referenced.
3558  */
3559 static int
3560 sync_inactive(struct vop_inactive_args *ap)
3561 {
3562 
3563 	vgone(ap->a_vp);
3564 	return (0);
3565 }
3566 
3567 /*
3568  * The syncer vnode is no longer needed and is being decommissioned.
3569  *
3570  * Modifications to the worklist must be protected by sync_mtx.
3571  */
3572 static int
3573 sync_reclaim(struct vop_reclaim_args *ap)
3574 {
3575 	struct vnode *vp = ap->a_vp;
3576 	struct bufobj *bo;
3577 
3578 	bo = &vp->v_bufobj;
3579 	BO_LOCK(bo);
3580 	mtx_lock(&sync_mtx);
3581 	if (vp->v_mount->mnt_syncer == vp)
3582 		vp->v_mount->mnt_syncer = NULL;
3583 	if (bo->bo_flag & BO_ONWORKLST) {
3584 		LIST_REMOVE(bo, bo_synclist);
3585 		syncer_worklist_len--;
3586 		sync_vnode_count--;
3587 		bo->bo_flag &= ~BO_ONWORKLST;
3588 	}
3589 	mtx_unlock(&sync_mtx);
3590 	BO_UNLOCK(bo);
3591 
3592 	return (0);
3593 }
3594 
3595 /*
3596  * Check if vnode represents a disk device
3597  */
3598 int
3599 vn_isdisk(struct vnode *vp, int *errp)
3600 {
3601 	int error;
3602 
3603 	error = 0;
3604 	dev_lock();
3605 	if (vp->v_type != VCHR)
3606 		error = ENOTBLK;
3607 	else if (vp->v_rdev == NULL)
3608 		error = ENXIO;
3609 	else if (vp->v_rdev->si_devsw == NULL)
3610 		error = ENXIO;
3611 	else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK))
3612 		error = ENOTBLK;
3613 	dev_unlock();
3614 	if (errp != NULL)
3615 		*errp = error;
3616 	return (error == 0);
3617 }
3618 
3619 /*
3620  * Common filesystem object access control check routine.  Accepts a
3621  * vnode's type, "mode", uid and gid, requested access mode, credentials,
3622  * and optional call-by-reference privused argument allowing vaccess()
3623  * to indicate to the caller whether privilege was used to satisfy the
3624  * request (obsoleted).  Returns 0 on success, or an errno on failure.
3625  */
3626 int
3627 vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid,
3628     accmode_t accmode, struct ucred *cred, int *privused)
3629 {
3630 	accmode_t dac_granted;
3631 	accmode_t priv_granted;
3632 
3633 	KASSERT((accmode & ~(VEXEC | VWRITE | VREAD | VADMIN | VAPPEND)) == 0,
3634 	    ("invalid bit in accmode"));
3635 	KASSERT((accmode & VAPPEND) == 0 || (accmode & VWRITE),
3636 	    ("VAPPEND without VWRITE"));
3637 
3638 	/*
3639 	 * Look for a normal, non-privileged way to access the file/directory
3640 	 * as requested.  If it exists, go with that.
3641 	 */
3642 
3643 	if (privused != NULL)
3644 		*privused = 0;
3645 
3646 	dac_granted = 0;
3647 
3648 	/* Check the owner. */
3649 	if (cred->cr_uid == file_uid) {
3650 		dac_granted |= VADMIN;
3651 		if (file_mode & S_IXUSR)
3652 			dac_granted |= VEXEC;
3653 		if (file_mode & S_IRUSR)
3654 			dac_granted |= VREAD;
3655 		if (file_mode & S_IWUSR)
3656 			dac_granted |= (VWRITE | VAPPEND);
3657 
3658 		if ((accmode & dac_granted) == accmode)
3659 			return (0);
3660 
3661 		goto privcheck;
3662 	}
3663 
3664 	/* Otherwise, check the groups (first match) */
3665 	if (groupmember(file_gid, cred)) {
3666 		if (file_mode & S_IXGRP)
3667 			dac_granted |= VEXEC;
3668 		if (file_mode & S_IRGRP)
3669 			dac_granted |= VREAD;
3670 		if (file_mode & S_IWGRP)
3671 			dac_granted |= (VWRITE | VAPPEND);
3672 
3673 		if ((accmode & dac_granted) == accmode)
3674 			return (0);
3675 
3676 		goto privcheck;
3677 	}
3678 
3679 	/* Otherwise, check everyone else. */
3680 	if (file_mode & S_IXOTH)
3681 		dac_granted |= VEXEC;
3682 	if (file_mode & S_IROTH)
3683 		dac_granted |= VREAD;
3684 	if (file_mode & S_IWOTH)
3685 		dac_granted |= (VWRITE | VAPPEND);
3686 	if ((accmode & dac_granted) == accmode)
3687 		return (0);
3688 
3689 privcheck:
3690 	/*
3691 	 * Build a privilege mask to determine if the set of privileges
3692 	 * satisfies the requirements when combined with the granted mask
3693 	 * from above.  For each privilege, if the privilege is required,
3694 	 * bitwise or the request type onto the priv_granted mask.
3695 	 */
3696 	priv_granted = 0;
3697 
3698 	if (type == VDIR) {
3699 		/*
3700 		 * For directories, use PRIV_VFS_LOOKUP to satisfy VEXEC
3701 		 * requests, instead of PRIV_VFS_EXEC.
3702 		 */
3703 		if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3704 		    !priv_check_cred(cred, PRIV_VFS_LOOKUP, 0))
3705 			priv_granted |= VEXEC;
3706 	} else {
3707 		/*
3708 		 * Ensure that at least one execute bit is on. Otherwise,
3709 		 * a privileged user will always succeed, and we don't want
3710 		 * this to happen unless the file really is executable.
3711 		 */
3712 		if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3713 		    (file_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) != 0 &&
3714 		    !priv_check_cred(cred, PRIV_VFS_EXEC, 0))
3715 			priv_granted |= VEXEC;
3716 	}
3717 
3718 	if ((accmode & VREAD) && ((dac_granted & VREAD) == 0) &&
3719 	    !priv_check_cred(cred, PRIV_VFS_READ, 0))
3720 		priv_granted |= VREAD;
3721 
3722 	if ((accmode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
3723 	    !priv_check_cred(cred, PRIV_VFS_WRITE, 0))
3724 		priv_granted |= (VWRITE | VAPPEND);
3725 
3726 	if ((accmode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
3727 	    !priv_check_cred(cred, PRIV_VFS_ADMIN, 0))
3728 		priv_granted |= VADMIN;
3729 
3730 	if ((accmode & (priv_granted | dac_granted)) == accmode) {
3731 		/* XXX audit: privilege used */
3732 		if (privused != NULL)
3733 			*privused = 1;
3734 		return (0);
3735 	}
3736 
3737 	return ((accmode & VADMIN) ? EPERM : EACCES);
3738 }
3739 
3740 /*
3741  * Credential check based on process requesting service, and per-attribute
3742  * permissions.
3743  */
3744 int
3745 extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred,
3746     struct thread *td, accmode_t accmode)
3747 {
3748 
3749 	/*
3750 	 * Kernel-invoked always succeeds.
3751 	 */
3752 	if (cred == NOCRED)
3753 		return (0);
3754 
3755 	/*
3756 	 * Do not allow privileged processes in jail to directly manipulate
3757 	 * system attributes.
3758 	 */
3759 	switch (attrnamespace) {
3760 	case EXTATTR_NAMESPACE_SYSTEM:
3761 		/* Potentially should be: return (EPERM); */
3762 		return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM, 0));
3763 	case EXTATTR_NAMESPACE_USER:
3764 		return (VOP_ACCESS(vp, accmode, cred, td));
3765 	default:
3766 		return (EPERM);
3767 	}
3768 }
3769 
3770 #ifdef DEBUG_VFS_LOCKS
3771 /*
3772  * This only exists to supress warnings from unlocked specfs accesses.  It is
3773  * no longer ok to have an unlocked VFS.
3774  */
3775 #define	IGNORE_LOCK(vp) (panicstr != NULL || (vp) == NULL ||		\
3776 	(vp)->v_type == VCHR ||	(vp)->v_type == VBAD)
3777 
3778 int vfs_badlock_ddb = 1;	/* Drop into debugger on violation. */
3779 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0,
3780     "Drop into debugger on lock violation");
3781 
3782 int vfs_badlock_mutex = 1;	/* Check for interlock across VOPs. */
3783 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex,
3784     0, "Check for interlock across VOPs");
3785 
3786 int vfs_badlock_print = 1;	/* Print lock violations. */
3787 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print,
3788     0, "Print lock violations");
3789 
3790 #ifdef KDB
3791 int vfs_badlock_backtrace = 1;	/* Print backtrace at lock violations. */
3792 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW,
3793     &vfs_badlock_backtrace, 0, "Print backtrace at lock violations");
3794 #endif
3795 
3796 static void
3797 vfs_badlock(const char *msg, const char *str, struct vnode *vp)
3798 {
3799 
3800 #ifdef KDB
3801 	if (vfs_badlock_backtrace)
3802 		kdb_backtrace();
3803 #endif
3804 	if (vfs_badlock_print)
3805 		printf("%s: %p %s\n", str, (void *)vp, msg);
3806 	if (vfs_badlock_ddb)
3807 		kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
3808 }
3809 
3810 void
3811 assert_vi_locked(struct vnode *vp, const char *str)
3812 {
3813 
3814 	if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp)))
3815 		vfs_badlock("interlock is not locked but should be", str, vp);
3816 }
3817 
3818 void
3819 assert_vi_unlocked(struct vnode *vp, const char *str)
3820 {
3821 
3822 	if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp)))
3823 		vfs_badlock("interlock is locked but should not be", str, vp);
3824 }
3825 
3826 void
3827 assert_vop_locked(struct vnode *vp, const char *str)
3828 {
3829 
3830 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == 0)
3831 		vfs_badlock("is not locked but should be", str, vp);
3832 }
3833 
3834 void
3835 assert_vop_unlocked(struct vnode *vp, const char *str)
3836 {
3837 
3838 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == LK_EXCLUSIVE)
3839 		vfs_badlock("is locked but should not be", str, vp);
3840 }
3841 
3842 void
3843 assert_vop_elocked(struct vnode *vp, const char *str)
3844 {
3845 
3846 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
3847 		vfs_badlock("is not exclusive locked but should be", str, vp);
3848 }
3849 
3850 #if 0
3851 void
3852 assert_vop_elocked_other(struct vnode *vp, const char *str)
3853 {
3854 
3855 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLOTHER)
3856 		vfs_badlock("is not exclusive locked by another thread",
3857 		    str, vp);
3858 }
3859 
3860 void
3861 assert_vop_slocked(struct vnode *vp, const char *str)
3862 {
3863 
3864 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_SHARED)
3865 		vfs_badlock("is not locked shared but should be", str, vp);
3866 }
3867 #endif /* 0 */
3868 #endif /* DEBUG_VFS_LOCKS */
3869 
3870 void
3871 vop_rename_fail(struct vop_rename_args *ap)
3872 {
3873 
3874 	if (ap->a_tvp != NULL)
3875 		vput(ap->a_tvp);
3876 	if (ap->a_tdvp == ap->a_tvp)
3877 		vrele(ap->a_tdvp);
3878 	else
3879 		vput(ap->a_tdvp);
3880 	vrele(ap->a_fdvp);
3881 	vrele(ap->a_fvp);
3882 }
3883 
3884 void
3885 vop_rename_pre(void *ap)
3886 {
3887 	struct vop_rename_args *a = ap;
3888 
3889 #ifdef DEBUG_VFS_LOCKS
3890 	if (a->a_tvp)
3891 		ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME");
3892 	ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME");
3893 	ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME");
3894 	ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME");
3895 
3896 	/* Check the source (from). */
3897 	if (a->a_tdvp->v_vnlock != a->a_fdvp->v_vnlock &&
3898 	    (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fdvp->v_vnlock))
3899 		ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked");
3900 	if (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fvp->v_vnlock)
3901 		ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: fvp locked");
3902 
3903 	/* Check the target. */
3904 	if (a->a_tvp)
3905 		ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked");
3906 	ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked");
3907 #endif
3908 	if (a->a_tdvp != a->a_fdvp)
3909 		vhold(a->a_fdvp);
3910 	if (a->a_tvp != a->a_fvp)
3911 		vhold(a->a_fvp);
3912 	vhold(a->a_tdvp);
3913 	if (a->a_tvp)
3914 		vhold(a->a_tvp);
3915 }
3916 
3917 void
3918 vop_strategy_pre(void *ap)
3919 {
3920 #ifdef DEBUG_VFS_LOCKS
3921 	struct vop_strategy_args *a;
3922 	struct buf *bp;
3923 
3924 	a = ap;
3925 	bp = a->a_bp;
3926 
3927 	/*
3928 	 * Cluster ops lock their component buffers but not the IO container.
3929 	 */
3930 	if ((bp->b_flags & B_CLUSTER) != 0)
3931 		return;
3932 
3933 	if (panicstr == NULL && !BUF_ISLOCKED(bp)) {
3934 		if (vfs_badlock_print)
3935 			printf(
3936 			    "VOP_STRATEGY: bp is not locked but should be\n");
3937 		if (vfs_badlock_ddb)
3938 			kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
3939 	}
3940 #endif
3941 }
3942 
3943 void
3944 vop_lookup_pre(void *ap)
3945 {
3946 #ifdef DEBUG_VFS_LOCKS
3947 	struct vop_lookup_args *a;
3948 	struct vnode *dvp;
3949 
3950 	a = ap;
3951 	dvp = a->a_dvp;
3952 	ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
3953 	ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
3954 #endif
3955 }
3956 
3957 void
3958 vop_lookup_post(void *ap, int rc)
3959 {
3960 #ifdef DEBUG_VFS_LOCKS
3961 	struct vop_lookup_args *a;
3962 	struct vnode *dvp;
3963 	struct vnode *vp;
3964 
3965 	a = ap;
3966 	dvp = a->a_dvp;
3967 	vp = *(a->a_vpp);
3968 
3969 	ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
3970 	ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
3971 
3972 	if (!rc)
3973 		ASSERT_VOP_LOCKED(vp, "VOP_LOOKUP (child)");
3974 #endif
3975 }
3976 
3977 void
3978 vop_lock_pre(void *ap)
3979 {
3980 #ifdef DEBUG_VFS_LOCKS
3981 	struct vop_lock1_args *a = ap;
3982 
3983 	if ((a->a_flags & LK_INTERLOCK) == 0)
3984 		ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
3985 	else
3986 		ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK");
3987 #endif
3988 }
3989 
3990 void
3991 vop_lock_post(void *ap, int rc)
3992 {
3993 #ifdef DEBUG_VFS_LOCKS
3994 	struct vop_lock1_args *a = ap;
3995 
3996 	ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
3997 	if (rc == 0)
3998 		ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK");
3999 #endif
4000 }
4001 
4002 void
4003 vop_unlock_pre(void *ap)
4004 {
4005 #ifdef DEBUG_VFS_LOCKS
4006 	struct vop_unlock_args *a = ap;
4007 
4008 	if (a->a_flags & LK_INTERLOCK)
4009 		ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK");
4010 	ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK");
4011 #endif
4012 }
4013 
4014 void
4015 vop_unlock_post(void *ap, int rc)
4016 {
4017 #ifdef DEBUG_VFS_LOCKS
4018 	struct vop_unlock_args *a = ap;
4019 
4020 	if (a->a_flags & LK_INTERLOCK)
4021 		ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK");
4022 #endif
4023 }
4024 
4025 void
4026 vop_create_post(void *ap, int rc)
4027 {
4028 	struct vop_create_args *a = ap;
4029 
4030 	if (!rc)
4031 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4032 }
4033 
4034 void
4035 vop_link_post(void *ap, int rc)
4036 {
4037 	struct vop_link_args *a = ap;
4038 
4039 	if (!rc) {
4040 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_LINK);
4041 		VFS_KNOTE_LOCKED(a->a_tdvp, NOTE_WRITE);
4042 	}
4043 }
4044 
4045 void
4046 vop_mkdir_post(void *ap, int rc)
4047 {
4048 	struct vop_mkdir_args *a = ap;
4049 
4050 	if (!rc)
4051 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
4052 }
4053 
4054 void
4055 vop_mknod_post(void *ap, int rc)
4056 {
4057 	struct vop_mknod_args *a = ap;
4058 
4059 	if (!rc)
4060 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4061 }
4062 
4063 void
4064 vop_remove_post(void *ap, int rc)
4065 {
4066 	struct vop_remove_args *a = ap;
4067 
4068 	if (!rc) {
4069 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4070 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
4071 	}
4072 }
4073 
4074 void
4075 vop_rename_post(void *ap, int rc)
4076 {
4077 	struct vop_rename_args *a = ap;
4078 
4079 	if (!rc) {
4080 		VFS_KNOTE_UNLOCKED(a->a_fdvp, NOTE_WRITE);
4081 		VFS_KNOTE_UNLOCKED(a->a_tdvp, NOTE_WRITE);
4082 		VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME);
4083 		if (a->a_tvp)
4084 			VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE);
4085 	}
4086 	if (a->a_tdvp != a->a_fdvp)
4087 		vdrop(a->a_fdvp);
4088 	if (a->a_tvp != a->a_fvp)
4089 		vdrop(a->a_fvp);
4090 	vdrop(a->a_tdvp);
4091 	if (a->a_tvp)
4092 		vdrop(a->a_tvp);
4093 }
4094 
4095 void
4096 vop_rmdir_post(void *ap, int rc)
4097 {
4098 	struct vop_rmdir_args *a = ap;
4099 
4100 	if (!rc) {
4101 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
4102 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
4103 	}
4104 }
4105 
4106 void
4107 vop_setattr_post(void *ap, int rc)
4108 {
4109 	struct vop_setattr_args *a = ap;
4110 
4111 	if (!rc)
4112 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4113 }
4114 
4115 void
4116 vop_symlink_post(void *ap, int rc)
4117 {
4118 	struct vop_symlink_args *a = ap;
4119 
4120 	if (!rc)
4121 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4122 }
4123 
4124 static struct knlist fs_knlist;
4125 
4126 static void
4127 vfs_event_init(void *arg)
4128 {
4129 	knlist_init_mtx(&fs_knlist, NULL);
4130 }
4131 /* XXX - correct order? */
4132 SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL);
4133 
4134 void
4135 vfs_event_signal(fsid_t *fsid, uint32_t event, intptr_t data __unused)
4136 {
4137 
4138 	KNOTE_UNLOCKED(&fs_knlist, event);
4139 }
4140 
4141 static int	filt_fsattach(struct knote *kn);
4142 static void	filt_fsdetach(struct knote *kn);
4143 static int	filt_fsevent(struct knote *kn, long hint);
4144 
4145 struct filterops fs_filtops = {
4146 	.f_isfd = 0,
4147 	.f_attach = filt_fsattach,
4148 	.f_detach = filt_fsdetach,
4149 	.f_event = filt_fsevent
4150 };
4151 
4152 static int
4153 filt_fsattach(struct knote *kn)
4154 {
4155 
4156 	kn->kn_flags |= EV_CLEAR;
4157 	knlist_add(&fs_knlist, kn, 0);
4158 	return (0);
4159 }
4160 
4161 static void
4162 filt_fsdetach(struct knote *kn)
4163 {
4164 
4165 	knlist_remove(&fs_knlist, kn, 0);
4166 }
4167 
4168 static int
4169 filt_fsevent(struct knote *kn, long hint)
4170 {
4171 
4172 	kn->kn_fflags |= hint;
4173 	return (kn->kn_fflags != 0);
4174 }
4175 
4176 static int
4177 sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS)
4178 {
4179 	struct vfsidctl vc;
4180 	int error;
4181 	struct mount *mp;
4182 
4183 	error = SYSCTL_IN(req, &vc, sizeof(vc));
4184 	if (error)
4185 		return (error);
4186 	if (vc.vc_vers != VFS_CTL_VERS1)
4187 		return (EINVAL);
4188 	mp = vfs_getvfs(&vc.vc_fsid);
4189 	if (mp == NULL)
4190 		return (ENOENT);
4191 	/* ensure that a specific sysctl goes to the right filesystem. */
4192 	if (strcmp(vc.vc_fstypename, "*") != 0 &&
4193 	    strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) {
4194 		vfs_rel(mp);
4195 		return (EINVAL);
4196 	}
4197 	VCTLTOREQ(&vc, req);
4198 	error = VFS_SYSCTL(mp, vc.vc_op, req);
4199 	vfs_rel(mp);
4200 	return (error);
4201 }
4202 
4203 SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLTYPE_OPAQUE | CTLFLAG_WR,
4204     NULL, 0, sysctl_vfs_ctl, "",
4205     "Sysctl by fsid");
4206 
4207 /*
4208  * Function to initialize a va_filerev field sensibly.
4209  * XXX: Wouldn't a random number make a lot more sense ??
4210  */
4211 u_quad_t
4212 init_va_filerev(void)
4213 {
4214 	struct bintime bt;
4215 
4216 	getbinuptime(&bt);
4217 	return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL));
4218 }
4219 
4220 static int	filt_vfsread(struct knote *kn, long hint);
4221 static int	filt_vfswrite(struct knote *kn, long hint);
4222 static int	filt_vfsvnode(struct knote *kn, long hint);
4223 static void	filt_vfsdetach(struct knote *kn);
4224 static struct filterops vfsread_filtops = {
4225 	.f_isfd = 1,
4226 	.f_detach = filt_vfsdetach,
4227 	.f_event = filt_vfsread
4228 };
4229 static struct filterops vfswrite_filtops = {
4230 	.f_isfd = 1,
4231 	.f_detach = filt_vfsdetach,
4232 	.f_event = filt_vfswrite
4233 };
4234 static struct filterops vfsvnode_filtops = {
4235 	.f_isfd = 1,
4236 	.f_detach = filt_vfsdetach,
4237 	.f_event = filt_vfsvnode
4238 };
4239 
4240 static void
4241 vfs_knllock(void *arg)
4242 {
4243 	struct vnode *vp = arg;
4244 
4245 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4246 }
4247 
4248 static void
4249 vfs_knlunlock(void *arg)
4250 {
4251 	struct vnode *vp = arg;
4252 
4253 	VOP_UNLOCK(vp, 0);
4254 }
4255 
4256 static void
4257 vfs_knl_assert_locked(void *arg)
4258 {
4259 #ifdef DEBUG_VFS_LOCKS
4260 	struct vnode *vp = arg;
4261 
4262 	ASSERT_VOP_LOCKED(vp, "vfs_knl_assert_locked");
4263 #endif
4264 }
4265 
4266 static void
4267 vfs_knl_assert_unlocked(void *arg)
4268 {
4269 #ifdef DEBUG_VFS_LOCKS
4270 	struct vnode *vp = arg;
4271 
4272 	ASSERT_VOP_UNLOCKED(vp, "vfs_knl_assert_unlocked");
4273 #endif
4274 }
4275 
4276 int
4277 vfs_kqfilter(struct vop_kqfilter_args *ap)
4278 {
4279 	struct vnode *vp = ap->a_vp;
4280 	struct knote *kn = ap->a_kn;
4281 	struct knlist *knl;
4282 
4283 	switch (kn->kn_filter) {
4284 	case EVFILT_READ:
4285 		kn->kn_fop = &vfsread_filtops;
4286 		break;
4287 	case EVFILT_WRITE:
4288 		kn->kn_fop = &vfswrite_filtops;
4289 		break;
4290 	case EVFILT_VNODE:
4291 		kn->kn_fop = &vfsvnode_filtops;
4292 		break;
4293 	default:
4294 		return (EINVAL);
4295 	}
4296 
4297 	kn->kn_hook = (caddr_t)vp;
4298 
4299 	v_addpollinfo(vp);
4300 	if (vp->v_pollinfo == NULL)
4301 		return (ENOMEM);
4302 	knl = &vp->v_pollinfo->vpi_selinfo.si_note;
4303 	knlist_add(knl, kn, 0);
4304 
4305 	return (0);
4306 }
4307 
4308 /*
4309  * Detach knote from vnode
4310  */
4311 static void
4312 filt_vfsdetach(struct knote *kn)
4313 {
4314 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4315 
4316 	KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo"));
4317 	knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0);
4318 }
4319 
4320 /*ARGSUSED*/
4321 static int
4322 filt_vfsread(struct knote *kn, long hint)
4323 {
4324 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4325 	struct vattr va;
4326 	int res;
4327 
4328 	/*
4329 	 * filesystem is gone, so set the EOF flag and schedule
4330 	 * the knote for deletion.
4331 	 */
4332 	if (hint == NOTE_REVOKE) {
4333 		VI_LOCK(vp);
4334 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4335 		VI_UNLOCK(vp);
4336 		return (1);
4337 	}
4338 
4339 	if (VOP_GETATTR(vp, &va, curthread->td_ucred))
4340 		return (0);
4341 
4342 	VI_LOCK(vp);
4343 	kn->kn_data = va.va_size - kn->kn_fp->f_offset;
4344 	res = (kn->kn_data != 0);
4345 	VI_UNLOCK(vp);
4346 	return (res);
4347 }
4348 
4349 /*ARGSUSED*/
4350 static int
4351 filt_vfswrite(struct knote *kn, long hint)
4352 {
4353 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4354 
4355 	VI_LOCK(vp);
4356 
4357 	/*
4358 	 * filesystem is gone, so set the EOF flag and schedule
4359 	 * the knote for deletion.
4360 	 */
4361 	if (hint == NOTE_REVOKE)
4362 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4363 
4364 	kn->kn_data = 0;
4365 	VI_UNLOCK(vp);
4366 	return (1);
4367 }
4368 
4369 static int
4370 filt_vfsvnode(struct knote *kn, long hint)
4371 {
4372 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4373 	int res;
4374 
4375 	VI_LOCK(vp);
4376 	if (kn->kn_sfflags & hint)
4377 		kn->kn_fflags |= hint;
4378 	if (hint == NOTE_REVOKE) {
4379 		kn->kn_flags |= EV_EOF;
4380 		VI_UNLOCK(vp);
4381 		return (1);
4382 	}
4383 	res = (kn->kn_fflags != 0);
4384 	VI_UNLOCK(vp);
4385 	return (res);
4386 }
4387 
4388 int
4389 vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off)
4390 {
4391 	int error;
4392 
4393 	if (dp->d_reclen > ap->a_uio->uio_resid)
4394 		return (ENAMETOOLONG);
4395 	error = uiomove(dp, dp->d_reclen, ap->a_uio);
4396 	if (error) {
4397 		if (ap->a_ncookies != NULL) {
4398 			if (ap->a_cookies != NULL)
4399 				free(ap->a_cookies, M_TEMP);
4400 			ap->a_cookies = NULL;
4401 			*ap->a_ncookies = 0;
4402 		}
4403 		return (error);
4404 	}
4405 	if (ap->a_ncookies == NULL)
4406 		return (0);
4407 
4408 	KASSERT(ap->a_cookies,
4409 	    ("NULL ap->a_cookies value with non-NULL ap->a_ncookies!"));
4410 
4411 	*ap->a_cookies = realloc(*ap->a_cookies,
4412 	    (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO);
4413 	(*ap->a_cookies)[*ap->a_ncookies] = off;
4414 	return (0);
4415 }
4416 
4417 /*
4418  * Mark for update the access time of the file if the filesystem
4419  * supports VOP_MARKATIME.  This functionality is used by execve and
4420  * mmap, so we want to avoid the I/O implied by directly setting
4421  * va_atime for the sake of efficiency.
4422  */
4423 void
4424 vfs_mark_atime(struct vnode *vp, struct ucred *cred)
4425 {
4426 	struct mount *mp;
4427 
4428 	mp = vp->v_mount;
4429 	VFS_ASSERT_GIANT(mp);
4430 	ASSERT_VOP_LOCKED(vp, "vfs_mark_atime");
4431 	if (mp != NULL && (mp->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
4432 		(void)VOP_MARKATIME(vp);
4433 }
4434 
4435 /*
4436  * The purpose of this routine is to remove granularity from accmode_t,
4437  * reducing it into standard unix access bits - VEXEC, VREAD, VWRITE,
4438  * VADMIN and VAPPEND.
4439  *
4440  * If it returns 0, the caller is supposed to continue with the usual
4441  * access checks using 'accmode' as modified by this routine.  If it
4442  * returns nonzero value, the caller is supposed to return that value
4443  * as errno.
4444  *
4445  * Note that after this routine runs, accmode may be zero.
4446  */
4447 int
4448 vfs_unixify_accmode(accmode_t *accmode)
4449 {
4450 	/*
4451 	 * There is no way to specify explicit "deny" rule using
4452 	 * file mode or POSIX.1e ACLs.
4453 	 */
4454 	if (*accmode & VEXPLICIT_DENY) {
4455 		*accmode = 0;
4456 		return (0);
4457 	}
4458 
4459 	/*
4460 	 * None of these can be translated into usual access bits.
4461 	 * Also, the common case for NFSv4 ACLs is to not contain
4462 	 * either of these bits. Caller should check for VWRITE
4463 	 * on the containing directory instead.
4464 	 */
4465 	if (*accmode & (VDELETE_CHILD | VDELETE))
4466 		return (EPERM);
4467 
4468 	if (*accmode & VADMIN_PERMS) {
4469 		*accmode &= ~VADMIN_PERMS;
4470 		*accmode |= VADMIN;
4471 	}
4472 
4473 	/*
4474 	 * There is no way to deny VREAD_ATTRIBUTES, VREAD_ACL
4475 	 * or VSYNCHRONIZE using file mode or POSIX.1e ACL.
4476 	 */
4477 	*accmode &= ~(VSTAT_PERMS | VSYNCHRONIZE);
4478 
4479 	return (0);
4480 }
4481