xref: /freebsd/sys/kern/vfs_subr.c (revision 830940567b49bb0c08dfaed40418999e76616909)
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 
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 #include <sys/bio.h>
49 #include <sys/buf.h>
50 #include <sys/condvar.h>
51 #include <sys/conf.h>
52 #include <sys/dirent.h>
53 #include <sys/event.h>
54 #include <sys/eventhandler.h>
55 #include <sys/extattr.h>
56 #include <sys/file.h>
57 #include <sys/fcntl.h>
58 #include <sys/jail.h>
59 #include <sys/kdb.h>
60 #include <sys/kernel.h>
61 #include <sys/kthread.h>
62 #include <sys/lockf.h>
63 #include <sys/malloc.h>
64 #include <sys/mount.h>
65 #include <sys/namei.h>
66 #include <sys/priv.h>
67 #include <sys/reboot.h>
68 #include <sys/sleepqueue.h>
69 #include <sys/stat.h>
70 #include <sys/sysctl.h>
71 #include <sys/syslog.h>
72 #include <sys/vmmeter.h>
73 #include <sys/vnode.h>
74 
75 #include <machine/stdarg.h>
76 
77 #include <security/mac/mac_framework.h>
78 
79 #include <vm/vm.h>
80 #include <vm/vm_object.h>
81 #include <vm/vm_extern.h>
82 #include <vm/pmap.h>
83 #include <vm/vm_map.h>
84 #include <vm/vm_page.h>
85 #include <vm/vm_kern.h>
86 #include <vm/uma.h>
87 
88 #ifdef DDB
89 #include <ddb/ddb.h>
90 #endif
91 
92 #define	WI_MPSAFEQ	0
93 #define	WI_GIANTQ	1
94 
95 static MALLOC_DEFINE(M_NETADDR, "subr_export_host", "Export host address structure");
96 
97 static void	delmntque(struct vnode *vp);
98 static int	flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo,
99 		    int slpflag, int slptimeo);
100 static void	syncer_shutdown(void *arg, int howto);
101 static int	vtryrecycle(struct vnode *vp);
102 static void	vbusy(struct vnode *vp);
103 static void	vinactive(struct vnode *, struct thread *);
104 static void	v_incr_usecount(struct vnode *);
105 static void	v_decr_usecount(struct vnode *);
106 static void	v_decr_useonly(struct vnode *);
107 static void	v_upgrade_usecount(struct vnode *);
108 static void	vfree(struct vnode *);
109 static void	vnlru_free(int);
110 static void	vgonel(struct vnode *);
111 static void	vfs_knllock(void *arg);
112 static void	vfs_knlunlock(void *arg);
113 static void	vfs_knl_assert_locked(void *arg);
114 static void	vfs_knl_assert_unlocked(void *arg);
115 static void	destroy_vpollinfo(struct vpollinfo *vi);
116 
117 /*
118  * Number of vnodes in existence.  Increased whenever getnewvnode()
119  * allocates a new vnode, decreased on vdestroy() called on VI_DOOMed
120  * vnode.
121  */
122 static unsigned long	numvnodes;
123 
124 SYSCTL_LONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, "");
125 
126 /*
127  * Conversion tables for conversion from vnode types to inode formats
128  * and back.
129  */
130 enum vtype iftovt_tab[16] = {
131 	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
132 	VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
133 };
134 int vttoif_tab[10] = {
135 	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
136 	S_IFSOCK, S_IFIFO, S_IFMT, S_IFMT
137 };
138 
139 /*
140  * List of vnodes that are ready for recycling.
141  */
142 static TAILQ_HEAD(freelst, vnode) vnode_free_list;
143 
144 /*
145  * Free vnode target.  Free vnodes may simply be files which have been stat'd
146  * but not read.  This is somewhat common, and a small cache of such files
147  * should be kept to avoid recreation costs.
148  */
149 static u_long wantfreevnodes;
150 SYSCTL_LONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
151 /* Number of vnodes in the free list. */
152 static u_long freevnodes;
153 SYSCTL_LONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, "");
154 
155 /*
156  * Various variables used for debugging the new implementation of
157  * reassignbuf().
158  * XXX these are probably of (very) limited utility now.
159  */
160 static int reassignbufcalls;
161 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0, "");
162 
163 /*
164  * Cache for the mount type id assigned to NFS.  This is used for
165  * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c.
166  */
167 int	nfs_mount_type = -1;
168 
169 /* To keep more than one thread at a time from running vfs_getnewfsid */
170 static struct mtx mntid_mtx;
171 
172 /*
173  * Lock for any access to the following:
174  *	vnode_free_list
175  *	numvnodes
176  *	freevnodes
177  */
178 static struct mtx vnode_free_list_mtx;
179 
180 /* Publicly exported FS */
181 struct nfs_public nfs_pub;
182 
183 /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
184 static uma_zone_t vnode_zone;
185 static uma_zone_t vnodepoll_zone;
186 
187 /* Set to 1 to print out reclaim of active vnodes */
188 int	prtactive;
189 
190 /*
191  * The workitem queue.
192  *
193  * It is useful to delay writes of file data and filesystem metadata
194  * for tens of seconds so that quickly created and deleted files need
195  * not waste disk bandwidth being created and removed. To realize this,
196  * we append vnodes to a "workitem" queue. When running with a soft
197  * updates implementation, most pending metadata dependencies should
198  * not wait for more than a few seconds. Thus, mounted on block devices
199  * are delayed only about a half the time that file data is delayed.
200  * Similarly, directory updates are more critical, so are only delayed
201  * about a third the time that file data is delayed. Thus, there are
202  * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
203  * one each second (driven off the filesystem syncer process). The
204  * syncer_delayno variable indicates the next queue that is to be processed.
205  * Items that need to be processed soon are placed in this queue:
206  *
207  *	syncer_workitem_pending[syncer_delayno]
208  *
209  * A delay of fifteen seconds is done by placing the request fifteen
210  * entries later in the queue:
211  *
212  *	syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
213  *
214  */
215 static int syncer_delayno;
216 static long syncer_mask;
217 LIST_HEAD(synclist, bufobj);
218 static struct synclist *syncer_workitem_pending[2];
219 /*
220  * The sync_mtx protects:
221  *	bo->bo_synclist
222  *	sync_vnode_count
223  *	syncer_delayno
224  *	syncer_state
225  *	syncer_workitem_pending
226  *	syncer_worklist_len
227  *	rushjob
228  */
229 static struct mtx sync_mtx;
230 static struct cv sync_wakeup;
231 
232 #define SYNCER_MAXDELAY		32
233 static int syncer_maxdelay = SYNCER_MAXDELAY;	/* maximum delay time */
234 static int syncdelay = 30;		/* max time to delay syncing data */
235 static int filedelay = 30;		/* time to delay syncing files */
236 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0, "");
237 static int dirdelay = 29;		/* time to delay syncing directories */
238 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0, "");
239 static int metadelay = 28;		/* time to delay syncing metadata */
240 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0, "");
241 static int rushjob;		/* number of slots to run ASAP */
242 static int stat_rush_requests;	/* number of times I/O speeded up */
243 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, "");
244 
245 /*
246  * When shutting down the syncer, run it at four times normal speed.
247  */
248 #define SYNCER_SHUTDOWN_SPEEDUP		4
249 static int sync_vnode_count;
250 static int syncer_worklist_len;
251 static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY }
252     syncer_state;
253 
254 /*
255  * Number of vnodes we want to exist at any one time.  This is mostly used
256  * to size hash tables in vnode-related code.  It is normally not used in
257  * getnewvnode(), as wantfreevnodes is normally nonzero.)
258  *
259  * XXX desiredvnodes is historical cruft and should not exist.
260  */
261 int desiredvnodes;
262 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
263     &desiredvnodes, 0, "Maximum number of vnodes");
264 SYSCTL_INT(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
265     &wantfreevnodes, 0, "Minimum number of vnodes (legacy)");
266 static int vnlru_nowhere;
267 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW,
268     &vnlru_nowhere, 0, "Number of times the vnlru process ran without success");
269 
270 /*
271  * Macros to control when a vnode is freed and recycled.  All require
272  * the vnode interlock.
273  */
274 #define VCANRECYCLE(vp) (((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
275 #define VSHOULDFREE(vp) (!((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
276 #define VSHOULDBUSY(vp) (((vp)->v_iflag & VI_FREE) && (vp)->v_holdcnt)
277 
278 
279 /*
280  * Initialize the vnode management data structures.
281  */
282 #ifndef	MAXVNODES_MAX
283 #define	MAXVNODES_MAX	100000
284 #endif
285 static void
286 vntblinit(void *dummy __unused)
287 {
288 
289 	/*
290 	 * Desiredvnodes is a function of the physical memory size and
291 	 * the kernel's heap size.  Specifically, desiredvnodes scales
292 	 * in proportion to the physical memory size until two fifths
293 	 * of the kernel's heap size is consumed by vnodes and vm
294 	 * objects.
295 	 */
296 	desiredvnodes = min(maxproc + cnt.v_page_count / 4, 2 * vm_kmem_size /
297 	    (5 * (sizeof(struct vm_object) + sizeof(struct vnode))));
298 	if (desiredvnodes > MAXVNODES_MAX) {
299 		if (bootverbose)
300 			printf("Reducing kern.maxvnodes %d -> %d\n",
301 			    desiredvnodes, MAXVNODES_MAX);
302 		desiredvnodes = MAXVNODES_MAX;
303 	}
304 	wantfreevnodes = desiredvnodes / 4;
305 	mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF);
306 	TAILQ_INIT(&vnode_free_list);
307 	mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF);
308 	vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL,
309 	    NULL, NULL, UMA_ALIGN_PTR, 0);
310 	vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo),
311 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
312 	/*
313 	 * Initialize the filesystem syncer.
314 	 */
315 	syncer_workitem_pending[WI_MPSAFEQ] = hashinit(syncer_maxdelay, M_VNODE,
316 	    &syncer_mask);
317 	syncer_workitem_pending[WI_GIANTQ] = hashinit(syncer_maxdelay, M_VNODE,
318 	    &syncer_mask);
319 	syncer_maxdelay = syncer_mask + 1;
320 	mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF);
321 	cv_init(&sync_wakeup, "syncer");
322 }
323 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL);
324 
325 
326 /*
327  * Mark a mount point as busy. Used to synchronize access and to delay
328  * unmounting. Eventually, mountlist_mtx is not released on failure.
329  */
330 int
331 vfs_busy(struct mount *mp, int flags)
332 {
333 
334 	MPASS((flags & ~MBF_MASK) == 0);
335 	CTR3(KTR_VFS, "%s: mp %p with flags %d", __func__, mp, flags);
336 
337 	MNT_ILOCK(mp);
338 	MNT_REF(mp);
339 	/*
340 	 * If mount point is currenly being unmounted, sleep until the
341 	 * mount point fate is decided.  If thread doing the unmounting fails,
342 	 * it will clear MNTK_UNMOUNT flag before waking us up, indicating
343 	 * that this mount point has survived the unmount attempt and vfs_busy
344 	 * should retry.  Otherwise the unmounter thread will set MNTK_REFEXPIRE
345 	 * flag in addition to MNTK_UNMOUNT, indicating that mount point is
346 	 * about to be really destroyed.  vfs_busy needs to release its
347 	 * reference on the mount point in this case and return with ENOENT,
348 	 * telling the caller that mount mount it tried to busy is no longer
349 	 * valid.
350 	 */
351 	while (mp->mnt_kern_flag & MNTK_UNMOUNT) {
352 		if (flags & MBF_NOWAIT || mp->mnt_kern_flag & MNTK_REFEXPIRE) {
353 			MNT_REL(mp);
354 			MNT_IUNLOCK(mp);
355 			CTR1(KTR_VFS, "%s: failed busying before sleeping",
356 			    __func__);
357 			return (ENOENT);
358 		}
359 		if (flags & MBF_MNTLSTLOCK)
360 			mtx_unlock(&mountlist_mtx);
361 		mp->mnt_kern_flag |= MNTK_MWAIT;
362 		msleep(mp, MNT_MTX(mp), PVFS, "vfs_busy", 0);
363 		if (flags & MBF_MNTLSTLOCK)
364 			mtx_lock(&mountlist_mtx);
365 	}
366 	if (flags & MBF_MNTLSTLOCK)
367 		mtx_unlock(&mountlist_mtx);
368 	mp->mnt_lockref++;
369 	MNT_IUNLOCK(mp);
370 	return (0);
371 }
372 
373 /*
374  * Free a busy filesystem.
375  */
376 void
377 vfs_unbusy(struct mount *mp)
378 {
379 
380 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
381 	MNT_ILOCK(mp);
382 	MNT_REL(mp);
383 	KASSERT(mp->mnt_lockref > 0, ("negative mnt_lockref"));
384 	mp->mnt_lockref--;
385 	if (mp->mnt_lockref == 0 && (mp->mnt_kern_flag & MNTK_DRAINING) != 0) {
386 		MPASS(mp->mnt_kern_flag & MNTK_UNMOUNT);
387 		CTR1(KTR_VFS, "%s: waking up waiters", __func__);
388 		mp->mnt_kern_flag &= ~MNTK_DRAINING;
389 		wakeup(&mp->mnt_lockref);
390 	}
391 	MNT_IUNLOCK(mp);
392 }
393 
394 /*
395  * Lookup a mount point by filesystem identifier.
396  */
397 struct mount *
398 vfs_getvfs(fsid_t *fsid)
399 {
400 	struct mount *mp;
401 
402 	CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
403 	mtx_lock(&mountlist_mtx);
404 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
405 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
406 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
407 			vfs_ref(mp);
408 			mtx_unlock(&mountlist_mtx);
409 			return (mp);
410 		}
411 	}
412 	mtx_unlock(&mountlist_mtx);
413 	CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
414 	return ((struct mount *) 0);
415 }
416 
417 /*
418  * Lookup a mount point by filesystem identifier, busying it before
419  * returning.
420  */
421 struct mount *
422 vfs_busyfs(fsid_t *fsid)
423 {
424 	struct mount *mp;
425 	int error;
426 
427 	CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
428 	mtx_lock(&mountlist_mtx);
429 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
430 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
431 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
432 			error = vfs_busy(mp, MBF_MNTLSTLOCK);
433 			if (error) {
434 				mtx_unlock(&mountlist_mtx);
435 				return (NULL);
436 			}
437 			return (mp);
438 		}
439 	}
440 	CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
441 	mtx_unlock(&mountlist_mtx);
442 	return ((struct mount *) 0);
443 }
444 
445 /*
446  * Check if a user can access privileged mount options.
447  */
448 int
449 vfs_suser(struct mount *mp, struct thread *td)
450 {
451 	int error;
452 
453 	/*
454 	 * If the thread is jailed, but this is not a jail-friendly file
455 	 * system, deny immediately.
456 	 */
457 	if (!(mp->mnt_vfc->vfc_flags & VFCF_JAIL) && jailed(td->td_ucred))
458 		return (EPERM);
459 
460 	/*
461 	 * If the file system was mounted outside the jail of the calling
462 	 * thread, deny immediately.
463 	 */
464 	if (prison_check(td->td_ucred, mp->mnt_cred) != 0)
465 		return (EPERM);
466 
467 	/*
468 	 * If file system supports delegated administration, we don't check
469 	 * for the PRIV_VFS_MOUNT_OWNER privilege - it will be better verified
470 	 * by the file system itself.
471 	 * If this is not the user that did original mount, we check for
472 	 * the PRIV_VFS_MOUNT_OWNER privilege.
473 	 */
474 	if (!(mp->mnt_vfc->vfc_flags & VFCF_DELEGADMIN) &&
475 	    mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) {
476 		if ((error = priv_check(td, PRIV_VFS_MOUNT_OWNER)) != 0)
477 			return (error);
478 	}
479 	return (0);
480 }
481 
482 /*
483  * Get a new unique fsid.  Try to make its val[0] unique, since this value
484  * will be used to create fake device numbers for stat().  Also try (but
485  * not so hard) make its val[0] unique mod 2^16, since some emulators only
486  * support 16-bit device numbers.  We end up with unique val[0]'s for the
487  * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
488  *
489  * Keep in mind that several mounts may be running in parallel.  Starting
490  * the search one past where the previous search terminated is both a
491  * micro-optimization and a defense against returning the same fsid to
492  * different mounts.
493  */
494 void
495 vfs_getnewfsid(struct mount *mp)
496 {
497 	static u_int16_t mntid_base;
498 	struct mount *nmp;
499 	fsid_t tfsid;
500 	int mtype;
501 
502 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
503 	mtx_lock(&mntid_mtx);
504 	mtype = mp->mnt_vfc->vfc_typenum;
505 	tfsid.val[1] = mtype;
506 	mtype = (mtype & 0xFF) << 24;
507 	for (;;) {
508 		tfsid.val[0] = makedev(255,
509 		    mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
510 		mntid_base++;
511 		if ((nmp = vfs_getvfs(&tfsid)) == NULL)
512 			break;
513 		vfs_rel(nmp);
514 	}
515 	mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
516 	mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
517 	mtx_unlock(&mntid_mtx);
518 }
519 
520 /*
521  * Knob to control the precision of file timestamps:
522  *
523  *   0 = seconds only; nanoseconds zeroed.
524  *   1 = seconds and nanoseconds, accurate within 1/HZ.
525  *   2 = seconds and nanoseconds, truncated to microseconds.
526  * >=3 = seconds and nanoseconds, maximum precision.
527  */
528 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
529 
530 static int timestamp_precision = TSP_SEC;
531 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
532     &timestamp_precision, 0, "");
533 
534 /*
535  * Get a current timestamp.
536  */
537 void
538 vfs_timestamp(struct timespec *tsp)
539 {
540 	struct timeval tv;
541 
542 	switch (timestamp_precision) {
543 	case TSP_SEC:
544 		tsp->tv_sec = time_second;
545 		tsp->tv_nsec = 0;
546 		break;
547 	case TSP_HZ:
548 		getnanotime(tsp);
549 		break;
550 	case TSP_USEC:
551 		microtime(&tv);
552 		TIMEVAL_TO_TIMESPEC(&tv, tsp);
553 		break;
554 	case TSP_NSEC:
555 	default:
556 		nanotime(tsp);
557 		break;
558 	}
559 }
560 
561 /*
562  * Set vnode attributes to VNOVAL
563  */
564 void
565 vattr_null(struct vattr *vap)
566 {
567 
568 	vap->va_type = VNON;
569 	vap->va_size = VNOVAL;
570 	vap->va_bytes = VNOVAL;
571 	vap->va_mode = VNOVAL;
572 	vap->va_nlink = VNOVAL;
573 	vap->va_uid = VNOVAL;
574 	vap->va_gid = VNOVAL;
575 	vap->va_fsid = VNOVAL;
576 	vap->va_fileid = VNOVAL;
577 	vap->va_blocksize = VNOVAL;
578 	vap->va_rdev = VNOVAL;
579 	vap->va_atime.tv_sec = VNOVAL;
580 	vap->va_atime.tv_nsec = VNOVAL;
581 	vap->va_mtime.tv_sec = VNOVAL;
582 	vap->va_mtime.tv_nsec = VNOVAL;
583 	vap->va_ctime.tv_sec = VNOVAL;
584 	vap->va_ctime.tv_nsec = VNOVAL;
585 	vap->va_birthtime.tv_sec = VNOVAL;
586 	vap->va_birthtime.tv_nsec = VNOVAL;
587 	vap->va_flags = VNOVAL;
588 	vap->va_gen = VNOVAL;
589 	vap->va_vaflags = 0;
590 }
591 
592 /*
593  * This routine is called when we have too many vnodes.  It attempts
594  * to free <count> vnodes and will potentially free vnodes that still
595  * have VM backing store (VM backing store is typically the cause
596  * of a vnode blowout so we want to do this).  Therefore, this operation
597  * is not considered cheap.
598  *
599  * A number of conditions may prevent a vnode from being reclaimed.
600  * the buffer cache may have references on the vnode, a directory
601  * vnode may still have references due to the namei cache representing
602  * underlying files, or the vnode may be in active use.   It is not
603  * desireable to reuse such vnodes.  These conditions may cause the
604  * number of vnodes to reach some minimum value regardless of what
605  * you set kern.maxvnodes to.  Do not set kern.maxvnodes too low.
606  */
607 static int
608 vlrureclaim(struct mount *mp)
609 {
610 	struct vnode *vp;
611 	int done;
612 	int trigger;
613 	int usevnodes;
614 	int count;
615 
616 	/*
617 	 * Calculate the trigger point, don't allow user
618 	 * screwups to blow us up.   This prevents us from
619 	 * recycling vnodes with lots of resident pages.  We
620 	 * aren't trying to free memory, we are trying to
621 	 * free vnodes.
622 	 */
623 	usevnodes = desiredvnodes;
624 	if (usevnodes <= 0)
625 		usevnodes = 1;
626 	trigger = cnt.v_page_count * 2 / usevnodes;
627 	done = 0;
628 	vn_start_write(NULL, &mp, V_WAIT);
629 	MNT_ILOCK(mp);
630 	count = mp->mnt_nvnodelistsize / 10 + 1;
631 	while (count != 0) {
632 		vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
633 		while (vp != NULL && vp->v_type == VMARKER)
634 			vp = TAILQ_NEXT(vp, v_nmntvnodes);
635 		if (vp == NULL)
636 			break;
637 		TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
638 		TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
639 		--count;
640 		if (!VI_TRYLOCK(vp))
641 			goto next_iter;
642 		/*
643 		 * If it's been deconstructed already, it's still
644 		 * referenced, or it exceeds the trigger, skip it.
645 		 */
646 		if (vp->v_usecount || !LIST_EMPTY(&(vp)->v_cache_src) ||
647 		    (vp->v_iflag & VI_DOOMED) != 0 || (vp->v_object != NULL &&
648 		    vp->v_object->resident_page_count > trigger)) {
649 			VI_UNLOCK(vp);
650 			goto next_iter;
651 		}
652 		MNT_IUNLOCK(mp);
653 		vholdl(vp);
654 		if (VOP_LOCK(vp, LK_INTERLOCK|LK_EXCLUSIVE|LK_NOWAIT)) {
655 			vdrop(vp);
656 			goto next_iter_mntunlocked;
657 		}
658 		VI_LOCK(vp);
659 		/*
660 		 * v_usecount may have been bumped after VOP_LOCK() dropped
661 		 * the vnode interlock and before it was locked again.
662 		 *
663 		 * It is not necessary to recheck VI_DOOMED because it can
664 		 * only be set by another thread that holds both the vnode
665 		 * lock and vnode interlock.  If another thread has the
666 		 * vnode lock before we get to VOP_LOCK() and obtains the
667 		 * vnode interlock after VOP_LOCK() drops the vnode
668 		 * interlock, the other thread will be unable to drop the
669 		 * vnode lock before our VOP_LOCK() call fails.
670 		 */
671 		if (vp->v_usecount || !LIST_EMPTY(&(vp)->v_cache_src) ||
672 		    (vp->v_object != NULL &&
673 		    vp->v_object->resident_page_count > trigger)) {
674 			VOP_UNLOCK(vp, LK_INTERLOCK);
675 			goto next_iter_mntunlocked;
676 		}
677 		KASSERT((vp->v_iflag & VI_DOOMED) == 0,
678 		    ("VI_DOOMED unexpectedly detected in vlrureclaim()"));
679 		vgonel(vp);
680 		VOP_UNLOCK(vp, 0);
681 		vdropl(vp);
682 		done++;
683 next_iter_mntunlocked:
684 		if ((count % 256) != 0)
685 			goto relock_mnt;
686 		goto yield;
687 next_iter:
688 		if ((count % 256) != 0)
689 			continue;
690 		MNT_IUNLOCK(mp);
691 yield:
692 		uio_yield();
693 relock_mnt:
694 		MNT_ILOCK(mp);
695 	}
696 	MNT_IUNLOCK(mp);
697 	vn_finished_write(mp);
698 	return done;
699 }
700 
701 /*
702  * Attempt to keep the free list at wantfreevnodes length.
703  */
704 static void
705 vnlru_free(int count)
706 {
707 	struct vnode *vp;
708 	int vfslocked;
709 
710 	mtx_assert(&vnode_free_list_mtx, MA_OWNED);
711 	for (; count > 0; count--) {
712 		vp = TAILQ_FIRST(&vnode_free_list);
713 		/*
714 		 * The list can be modified while the free_list_mtx
715 		 * has been dropped and vp could be NULL here.
716 		 */
717 		if (!vp)
718 			break;
719 		VNASSERT(vp->v_op != NULL, vp,
720 		    ("vnlru_free: vnode already reclaimed."));
721 		TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
722 		/*
723 		 * Don't recycle if we can't get the interlock.
724 		 */
725 		if (!VI_TRYLOCK(vp)) {
726 			TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
727 			continue;
728 		}
729 		VNASSERT(VCANRECYCLE(vp), vp,
730 		    ("vp inconsistent on freelist"));
731 		freevnodes--;
732 		vp->v_iflag &= ~VI_FREE;
733 		vholdl(vp);
734 		mtx_unlock(&vnode_free_list_mtx);
735 		VI_UNLOCK(vp);
736 		vfslocked = VFS_LOCK_GIANT(vp->v_mount);
737 		vtryrecycle(vp);
738 		VFS_UNLOCK_GIANT(vfslocked);
739 		/*
740 		 * If the recycled succeeded this vdrop will actually free
741 		 * the vnode.  If not it will simply place it back on
742 		 * the free list.
743 		 */
744 		vdrop(vp);
745 		mtx_lock(&vnode_free_list_mtx);
746 	}
747 }
748 /*
749  * Attempt to recycle vnodes in a context that is always safe to block.
750  * Calling vlrurecycle() from the bowels of filesystem code has some
751  * interesting deadlock problems.
752  */
753 static struct proc *vnlruproc;
754 static int vnlruproc_sig;
755 
756 static void
757 vnlru_proc(void)
758 {
759 	struct mount *mp, *nmp;
760 	int done, vfslocked;
761 	struct proc *p = vnlruproc;
762 
763 	EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
764 	    SHUTDOWN_PRI_FIRST);
765 
766 	for (;;) {
767 		kproc_suspend_check(p);
768 		mtx_lock(&vnode_free_list_mtx);
769 		if (freevnodes > wantfreevnodes)
770 			vnlru_free(freevnodes - wantfreevnodes);
771 		if (numvnodes <= desiredvnodes * 9 / 10) {
772 			vnlruproc_sig = 0;
773 			wakeup(&vnlruproc_sig);
774 			msleep(vnlruproc, &vnode_free_list_mtx,
775 			    PVFS|PDROP, "vlruwt", hz);
776 			continue;
777 		}
778 		mtx_unlock(&vnode_free_list_mtx);
779 		done = 0;
780 		mtx_lock(&mountlist_mtx);
781 		for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
782 			if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) {
783 				nmp = TAILQ_NEXT(mp, mnt_list);
784 				continue;
785 			}
786 			vfslocked = VFS_LOCK_GIANT(mp);
787 			done += vlrureclaim(mp);
788 			VFS_UNLOCK_GIANT(vfslocked);
789 			mtx_lock(&mountlist_mtx);
790 			nmp = TAILQ_NEXT(mp, mnt_list);
791 			vfs_unbusy(mp);
792 		}
793 		mtx_unlock(&mountlist_mtx);
794 		if (done == 0) {
795 			EVENTHANDLER_INVOKE(vfs_lowvnodes, desiredvnodes / 10);
796 #if 0
797 			/* These messages are temporary debugging aids */
798 			if (vnlru_nowhere < 5)
799 				printf("vnlru process getting nowhere..\n");
800 			else if (vnlru_nowhere == 5)
801 				printf("vnlru process messages stopped.\n");
802 #endif
803 			vnlru_nowhere++;
804 			tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
805 		} else
806 			uio_yield();
807 	}
808 }
809 
810 static struct kproc_desc vnlru_kp = {
811 	"vnlru",
812 	vnlru_proc,
813 	&vnlruproc
814 };
815 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start,
816     &vnlru_kp);
817 
818 /*
819  * Routines having to do with the management of the vnode table.
820  */
821 
822 void
823 vdestroy(struct vnode *vp)
824 {
825 	struct bufobj *bo;
826 
827 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
828 	mtx_lock(&vnode_free_list_mtx);
829 	numvnodes--;
830 	mtx_unlock(&vnode_free_list_mtx);
831 	bo = &vp->v_bufobj;
832 	VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
833 	    ("cleaned vnode still on the free list."));
834 	VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't"));
835 	VNASSERT(vp->v_holdcnt == 0, vp, ("Non-zero hold count"));
836 	VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count"));
837 	VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count"));
838 	VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's"));
839 	VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0"));
840 	VNASSERT(bo->bo_clean.bv_root == NULL, vp, ("cleanblkroot not NULL"));
841 	VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0"));
842 	VNASSERT(bo->bo_dirty.bv_root == NULL, vp, ("dirtyblkroot not NULL"));
843 	VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst"));
844 	VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src"));
845 	VNASSERT(vp->v_cache_dd == NULL, vp, ("vp has namecache for .."));
846 	VI_UNLOCK(vp);
847 #ifdef MAC
848 	mac_vnode_destroy(vp);
849 #endif
850 	if (vp->v_pollinfo != NULL)
851 		destroy_vpollinfo(vp->v_pollinfo);
852 #ifdef INVARIANTS
853 	/* XXX Elsewhere we can detect an already freed vnode via NULL v_op. */
854 	vp->v_op = NULL;
855 #endif
856 	lockdestroy(vp->v_vnlock);
857 	mtx_destroy(&vp->v_interlock);
858 	mtx_destroy(BO_MTX(bo));
859 	uma_zfree(vnode_zone, vp);
860 }
861 
862 /*
863  * Try to recycle a freed vnode.  We abort if anyone picks up a reference
864  * before we actually vgone().  This function must be called with the vnode
865  * held to prevent the vnode from being returned to the free list midway
866  * through vgone().
867  */
868 static int
869 vtryrecycle(struct vnode *vp)
870 {
871 	struct mount *vnmp;
872 
873 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
874 	VNASSERT(vp->v_holdcnt, vp,
875 	    ("vtryrecycle: Recycling vp %p without a reference.", vp));
876 	/*
877 	 * This vnode may found and locked via some other list, if so we
878 	 * can't recycle it yet.
879 	 */
880 	if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
881 		CTR2(KTR_VFS,
882 		    "%s: impossible to recycle, vp %p lock is already held",
883 		    __func__, vp);
884 		return (EWOULDBLOCK);
885 	}
886 	/*
887 	 * Don't recycle if its filesystem is being suspended.
888 	 */
889 	if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) {
890 		VOP_UNLOCK(vp, 0);
891 		CTR2(KTR_VFS,
892 		    "%s: impossible to recycle, cannot start the write for %p",
893 		    __func__, vp);
894 		return (EBUSY);
895 	}
896 	/*
897 	 * If we got this far, we need to acquire the interlock and see if
898 	 * anyone picked up this vnode from another list.  If not, we will
899 	 * mark it with DOOMED via vgonel() so that anyone who does find it
900 	 * will skip over it.
901 	 */
902 	VI_LOCK(vp);
903 	if (vp->v_usecount) {
904 		VOP_UNLOCK(vp, LK_INTERLOCK);
905 		vn_finished_write(vnmp);
906 		CTR2(KTR_VFS,
907 		    "%s: impossible to recycle, %p is already referenced",
908 		    __func__, vp);
909 		return (EBUSY);
910 	}
911 	if ((vp->v_iflag & VI_DOOMED) == 0)
912 		vgonel(vp);
913 	VOP_UNLOCK(vp, LK_INTERLOCK);
914 	vn_finished_write(vnmp);
915 	return (0);
916 }
917 
918 /*
919  * Return the next vnode from the free list.
920  */
921 int
922 getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops,
923     struct vnode **vpp)
924 {
925 	struct vnode *vp = NULL;
926 	struct bufobj *bo;
927 
928 	CTR3(KTR_VFS, "%s: mp %p with tag %s", __func__, mp, tag);
929 	mtx_lock(&vnode_free_list_mtx);
930 	/*
931 	 * Lend our context to reclaim vnodes if they've exceeded the max.
932 	 */
933 	if (freevnodes > wantfreevnodes)
934 		vnlru_free(1);
935 	/*
936 	 * Wait for available vnodes.
937 	 */
938 	if (numvnodes > desiredvnodes) {
939 		if (mp != NULL && (mp->mnt_kern_flag & MNTK_SUSPEND)) {
940 			/*
941 			 * File system is beeing suspended, we cannot risk a
942 			 * deadlock here, so allocate new vnode anyway.
943 			 */
944 			if (freevnodes > wantfreevnodes)
945 				vnlru_free(freevnodes - wantfreevnodes);
946 			goto alloc;
947 		}
948 		if (vnlruproc_sig == 0) {
949 			vnlruproc_sig = 1;	/* avoid unnecessary wakeups */
950 			wakeup(vnlruproc);
951 		}
952 		msleep(&vnlruproc_sig, &vnode_free_list_mtx, PVFS,
953 		    "vlruwk", hz);
954 #if 0	/* XXX Not all VFS_VGET/ffs_vget callers check returns. */
955 		if (numvnodes > desiredvnodes) {
956 			mtx_unlock(&vnode_free_list_mtx);
957 			return (ENFILE);
958 		}
959 #endif
960 	}
961 alloc:
962 	numvnodes++;
963 	mtx_unlock(&vnode_free_list_mtx);
964 	vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO);
965 	/*
966 	 * Setup locks.
967 	 */
968 	vp->v_vnlock = &vp->v_lock;
969 	mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
970 	/*
971 	 * By default, don't allow shared locks unless filesystems
972 	 * opt-in.
973 	 */
974 	lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOSHARE);
975 	/*
976 	 * Initialize bufobj.
977 	 */
978 	bo = &vp->v_bufobj;
979 	bo->__bo_vnode = vp;
980 	mtx_init(BO_MTX(bo), "bufobj interlock", NULL, MTX_DEF);
981 	bo->bo_ops = &buf_ops_bio;
982 	bo->bo_private = vp;
983 	TAILQ_INIT(&bo->bo_clean.bv_hd);
984 	TAILQ_INIT(&bo->bo_dirty.bv_hd);
985 	/*
986 	 * Initialize namecache.
987 	 */
988 	LIST_INIT(&vp->v_cache_src);
989 	TAILQ_INIT(&vp->v_cache_dst);
990 	/*
991 	 * Finalize various vnode identity bits.
992 	 */
993 	vp->v_type = VNON;
994 	vp->v_tag = tag;
995 	vp->v_op = vops;
996 	v_incr_usecount(vp);
997 	vp->v_data = 0;
998 #ifdef MAC
999 	mac_vnode_init(vp);
1000 	if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0)
1001 		mac_vnode_associate_singlelabel(mp, vp);
1002 	else if (mp == NULL && vops != &dead_vnodeops)
1003 		printf("NULL mp in getnewvnode()\n");
1004 #endif
1005 	if (mp != NULL) {
1006 		bo->bo_bsize = mp->mnt_stat.f_iosize;
1007 		if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0)
1008 			vp->v_vflag |= VV_NOKNOTE;
1009 	}
1010 
1011 	*vpp = vp;
1012 	return (0);
1013 }
1014 
1015 /*
1016  * Delete from old mount point vnode list, if on one.
1017  */
1018 static void
1019 delmntque(struct vnode *vp)
1020 {
1021 	struct mount *mp;
1022 
1023 	mp = vp->v_mount;
1024 	if (mp == NULL)
1025 		return;
1026 	MNT_ILOCK(mp);
1027 	vp->v_mount = NULL;
1028 	VNASSERT(mp->mnt_nvnodelistsize > 0, vp,
1029 		("bad mount point vnode list size"));
1030 	TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1031 	mp->mnt_nvnodelistsize--;
1032 	MNT_REL(mp);
1033 	MNT_IUNLOCK(mp);
1034 }
1035 
1036 static void
1037 insmntque_stddtr(struct vnode *vp, void *dtr_arg)
1038 {
1039 
1040 	vp->v_data = NULL;
1041 	vp->v_op = &dead_vnodeops;
1042 	/* XXX non mp-safe fs may still call insmntque with vnode
1043 	   unlocked */
1044 	if (!VOP_ISLOCKED(vp))
1045 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1046 	vgone(vp);
1047 	vput(vp);
1048 }
1049 
1050 /*
1051  * Insert into list of vnodes for the new mount point, if available.
1052  */
1053 int
1054 insmntque1(struct vnode *vp, struct mount *mp,
1055 	void (*dtr)(struct vnode *, void *), void *dtr_arg)
1056 {
1057 	int locked;
1058 
1059 	KASSERT(vp->v_mount == NULL,
1060 		("insmntque: vnode already on per mount vnode list"));
1061 	VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)"));
1062 #ifdef DEBUG_VFS_LOCKS
1063 	if (!VFS_NEEDSGIANT(mp))
1064 		ASSERT_VOP_ELOCKED(vp,
1065 		    "insmntque: mp-safe fs and non-locked vp");
1066 #endif
1067 	MNT_ILOCK(mp);
1068 	if ((mp->mnt_kern_flag & MNTK_NOINSMNTQ) != 0 &&
1069 	    ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0 ||
1070 	     mp->mnt_nvnodelistsize == 0)) {
1071 		locked = VOP_ISLOCKED(vp);
1072 		if (!locked || (locked == LK_EXCLUSIVE &&
1073 		     (vp->v_vflag & VV_FORCEINSMQ) == 0)) {
1074 			MNT_IUNLOCK(mp);
1075 			if (dtr != NULL)
1076 				dtr(vp, dtr_arg);
1077 			return (EBUSY);
1078 		}
1079 	}
1080 	vp->v_mount = mp;
1081 	MNT_REF(mp);
1082 	TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1083 	VNASSERT(mp->mnt_nvnodelistsize >= 0, vp,
1084 		("neg mount point vnode list size"));
1085 	mp->mnt_nvnodelistsize++;
1086 	MNT_IUNLOCK(mp);
1087 	return (0);
1088 }
1089 
1090 int
1091 insmntque(struct vnode *vp, struct mount *mp)
1092 {
1093 
1094 	return (insmntque1(vp, mp, insmntque_stddtr, NULL));
1095 }
1096 
1097 /*
1098  * Flush out and invalidate all buffers associated with a bufobj
1099  * Called with the underlying object locked.
1100  */
1101 int
1102 bufobj_invalbuf(struct bufobj *bo, int flags, int slpflag, int slptimeo)
1103 {
1104 	int error;
1105 
1106 	BO_LOCK(bo);
1107 	if (flags & V_SAVE) {
1108 		error = bufobj_wwait(bo, slpflag, slptimeo);
1109 		if (error) {
1110 			BO_UNLOCK(bo);
1111 			return (error);
1112 		}
1113 		if (bo->bo_dirty.bv_cnt > 0) {
1114 			BO_UNLOCK(bo);
1115 			if ((error = BO_SYNC(bo, MNT_WAIT)) != 0)
1116 				return (error);
1117 			/*
1118 			 * XXX We could save a lock/unlock if this was only
1119 			 * enabled under INVARIANTS
1120 			 */
1121 			BO_LOCK(bo);
1122 			if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)
1123 				panic("vinvalbuf: dirty bufs");
1124 		}
1125 	}
1126 	/*
1127 	 * If you alter this loop please notice that interlock is dropped and
1128 	 * reacquired in flushbuflist.  Special care is needed to ensure that
1129 	 * no race conditions occur from this.
1130 	 */
1131 	do {
1132 		error = flushbuflist(&bo->bo_clean,
1133 		    flags, bo, slpflag, slptimeo);
1134 		if (error == 0)
1135 			error = flushbuflist(&bo->bo_dirty,
1136 			    flags, bo, slpflag, slptimeo);
1137 		if (error != 0 && error != EAGAIN) {
1138 			BO_UNLOCK(bo);
1139 			return (error);
1140 		}
1141 	} while (error != 0);
1142 
1143 	/*
1144 	 * Wait for I/O to complete.  XXX needs cleaning up.  The vnode can
1145 	 * have write I/O in-progress but if there is a VM object then the
1146 	 * VM object can also have read-I/O in-progress.
1147 	 */
1148 	do {
1149 		bufobj_wwait(bo, 0, 0);
1150 		BO_UNLOCK(bo);
1151 		if (bo->bo_object != NULL) {
1152 			VM_OBJECT_LOCK(bo->bo_object);
1153 			vm_object_pip_wait(bo->bo_object, "bovlbx");
1154 			VM_OBJECT_UNLOCK(bo->bo_object);
1155 		}
1156 		BO_LOCK(bo);
1157 	} while (bo->bo_numoutput > 0);
1158 	BO_UNLOCK(bo);
1159 
1160 	/*
1161 	 * Destroy the copy in the VM cache, too.
1162 	 */
1163 	if (bo->bo_object != NULL && (flags & (V_ALT | V_NORMAL)) == 0) {
1164 		VM_OBJECT_LOCK(bo->bo_object);
1165 		vm_object_page_remove(bo->bo_object, 0, 0,
1166 			(flags & V_SAVE) ? TRUE : FALSE);
1167 		VM_OBJECT_UNLOCK(bo->bo_object);
1168 	}
1169 
1170 #ifdef INVARIANTS
1171 	BO_LOCK(bo);
1172 	if ((flags & (V_ALT | V_NORMAL)) == 0 &&
1173 	    (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0))
1174 		panic("vinvalbuf: flush failed");
1175 	BO_UNLOCK(bo);
1176 #endif
1177 	return (0);
1178 }
1179 
1180 /*
1181  * Flush out and invalidate all buffers associated with a vnode.
1182  * Called with the underlying object locked.
1183  */
1184 int
1185 vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo)
1186 {
1187 
1188 	CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
1189 	ASSERT_VOP_LOCKED(vp, "vinvalbuf");
1190 	return (bufobj_invalbuf(&vp->v_bufobj, flags, slpflag, slptimeo));
1191 }
1192 
1193 /*
1194  * Flush out buffers on the specified list.
1195  *
1196  */
1197 static int
1198 flushbuflist( struct bufv *bufv, int flags, struct bufobj *bo, int slpflag,
1199     int slptimeo)
1200 {
1201 	struct buf *bp, *nbp;
1202 	int retval, error;
1203 	daddr_t lblkno;
1204 	b_xflags_t xflags;
1205 
1206 	ASSERT_BO_LOCKED(bo);
1207 
1208 	retval = 0;
1209 	TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) {
1210 		if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) ||
1211 		    ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) {
1212 			continue;
1213 		}
1214 		lblkno = 0;
1215 		xflags = 0;
1216 		if (nbp != NULL) {
1217 			lblkno = nbp->b_lblkno;
1218 			xflags = nbp->b_xflags &
1219 				(BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN);
1220 		}
1221 		retval = EAGAIN;
1222 		error = BUF_TIMELOCK(bp,
1223 		    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_MTX(bo),
1224 		    "flushbuf", slpflag, slptimeo);
1225 		if (error) {
1226 			BO_LOCK(bo);
1227 			return (error != ENOLCK ? error : EAGAIN);
1228 		}
1229 		KASSERT(bp->b_bufobj == bo,
1230 		    ("bp %p wrong b_bufobj %p should be %p",
1231 		    bp, bp->b_bufobj, bo));
1232 		if (bp->b_bufobj != bo) {	/* XXX: necessary ? */
1233 			BUF_UNLOCK(bp);
1234 			BO_LOCK(bo);
1235 			return (EAGAIN);
1236 		}
1237 		/*
1238 		 * XXX Since there are no node locks for NFS, I
1239 		 * believe there is a slight chance that a delayed
1240 		 * write will occur while sleeping just above, so
1241 		 * check for it.
1242 		 */
1243 		if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
1244 		    (flags & V_SAVE)) {
1245 			bremfree(bp);
1246 			bp->b_flags |= B_ASYNC;
1247 			bwrite(bp);
1248 			BO_LOCK(bo);
1249 			return (EAGAIN);	/* XXX: why not loop ? */
1250 		}
1251 		bremfree(bp);
1252 		bp->b_flags |= (B_INVAL | B_RELBUF);
1253 		bp->b_flags &= ~B_ASYNC;
1254 		brelse(bp);
1255 		BO_LOCK(bo);
1256 		if (nbp != NULL &&
1257 		    (nbp->b_bufobj != bo ||
1258 		     nbp->b_lblkno != lblkno ||
1259 		     (nbp->b_xflags &
1260 		      (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN)) != xflags))
1261 			break;			/* nbp invalid */
1262 	}
1263 	return (retval);
1264 }
1265 
1266 /*
1267  * Truncate a file's buffer and pages to a specified length.  This
1268  * is in lieu of the old vinvalbuf mechanism, which performed unneeded
1269  * sync activity.
1270  */
1271 int
1272 vtruncbuf(struct vnode *vp, struct ucred *cred, struct thread *td,
1273     off_t length, int blksize)
1274 {
1275 	struct buf *bp, *nbp;
1276 	int anyfreed;
1277 	int trunclbn;
1278 	struct bufobj *bo;
1279 
1280 	CTR5(KTR_VFS, "%s: vp %p with cred %p and block %d:%ju", __func__,
1281 	    vp, cred, blksize, (uintmax_t)length);
1282 
1283 	/*
1284 	 * Round up to the *next* lbn.
1285 	 */
1286 	trunclbn = (length + blksize - 1) / blksize;
1287 
1288 	ASSERT_VOP_LOCKED(vp, "vtruncbuf");
1289 restart:
1290 	bo = &vp->v_bufobj;
1291 	BO_LOCK(bo);
1292 	anyfreed = 1;
1293 	for (;anyfreed;) {
1294 		anyfreed = 0;
1295 		TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) {
1296 			if (bp->b_lblkno < trunclbn)
1297 				continue;
1298 			if (BUF_LOCK(bp,
1299 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1300 			    BO_MTX(bo)) == ENOLCK)
1301 				goto restart;
1302 
1303 			bremfree(bp);
1304 			bp->b_flags |= (B_INVAL | B_RELBUF);
1305 			bp->b_flags &= ~B_ASYNC;
1306 			brelse(bp);
1307 			anyfreed = 1;
1308 
1309 			if (nbp != NULL &&
1310 			    (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1311 			    (nbp->b_vp != vp) ||
1312 			    (nbp->b_flags & B_DELWRI))) {
1313 				goto restart;
1314 			}
1315 			BO_LOCK(bo);
1316 		}
1317 
1318 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1319 			if (bp->b_lblkno < trunclbn)
1320 				continue;
1321 			if (BUF_LOCK(bp,
1322 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1323 			    BO_MTX(bo)) == ENOLCK)
1324 				goto restart;
1325 			bremfree(bp);
1326 			bp->b_flags |= (B_INVAL | B_RELBUF);
1327 			bp->b_flags &= ~B_ASYNC;
1328 			brelse(bp);
1329 			anyfreed = 1;
1330 			if (nbp != NULL &&
1331 			    (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1332 			    (nbp->b_vp != vp) ||
1333 			    (nbp->b_flags & B_DELWRI) == 0)) {
1334 				goto restart;
1335 			}
1336 			BO_LOCK(bo);
1337 		}
1338 	}
1339 
1340 	if (length > 0) {
1341 restartsync:
1342 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1343 			if (bp->b_lblkno > 0)
1344 				continue;
1345 			/*
1346 			 * Since we hold the vnode lock this should only
1347 			 * fail if we're racing with the buf daemon.
1348 			 */
1349 			if (BUF_LOCK(bp,
1350 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1351 			    BO_MTX(bo)) == ENOLCK) {
1352 				goto restart;
1353 			}
1354 			VNASSERT((bp->b_flags & B_DELWRI), vp,
1355 			    ("buf(%p) on dirty queue without DELWRI", bp));
1356 
1357 			bremfree(bp);
1358 			bawrite(bp);
1359 			BO_LOCK(bo);
1360 			goto restartsync;
1361 		}
1362 	}
1363 
1364 	bufobj_wwait(bo, 0, 0);
1365 	BO_UNLOCK(bo);
1366 	vnode_pager_setsize(vp, length);
1367 
1368 	return (0);
1369 }
1370 
1371 /*
1372  * buf_splay() - splay tree core for the clean/dirty list of buffers in
1373  * 		 a vnode.
1374  *
1375  *	NOTE: We have to deal with the special case of a background bitmap
1376  *	buffer, a situation where two buffers will have the same logical
1377  *	block offset.  We want (1) only the foreground buffer to be accessed
1378  *	in a lookup and (2) must differentiate between the foreground and
1379  *	background buffer in the splay tree algorithm because the splay
1380  *	tree cannot normally handle multiple entities with the same 'index'.
1381  *	We accomplish this by adding differentiating flags to the splay tree's
1382  *	numerical domain.
1383  */
1384 static
1385 struct buf *
1386 buf_splay(daddr_t lblkno, b_xflags_t xflags, struct buf *root)
1387 {
1388 	struct buf dummy;
1389 	struct buf *lefttreemax, *righttreemin, *y;
1390 
1391 	if (root == NULL)
1392 		return (NULL);
1393 	lefttreemax = righttreemin = &dummy;
1394 	for (;;) {
1395 		if (lblkno < root->b_lblkno ||
1396 		    (lblkno == root->b_lblkno &&
1397 		    (xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1398 			if ((y = root->b_left) == NULL)
1399 				break;
1400 			if (lblkno < y->b_lblkno) {
1401 				/* Rotate right. */
1402 				root->b_left = y->b_right;
1403 				y->b_right = root;
1404 				root = y;
1405 				if ((y = root->b_left) == NULL)
1406 					break;
1407 			}
1408 			/* Link into the new root's right tree. */
1409 			righttreemin->b_left = root;
1410 			righttreemin = root;
1411 		} else if (lblkno > root->b_lblkno ||
1412 		    (lblkno == root->b_lblkno &&
1413 		    (xflags & BX_BKGRDMARKER) > (root->b_xflags & BX_BKGRDMARKER))) {
1414 			if ((y = root->b_right) == NULL)
1415 				break;
1416 			if (lblkno > y->b_lblkno) {
1417 				/* Rotate left. */
1418 				root->b_right = y->b_left;
1419 				y->b_left = root;
1420 				root = y;
1421 				if ((y = root->b_right) == NULL)
1422 					break;
1423 			}
1424 			/* Link into the new root's left tree. */
1425 			lefttreemax->b_right = root;
1426 			lefttreemax = root;
1427 		} else {
1428 			break;
1429 		}
1430 		root = y;
1431 	}
1432 	/* Assemble the new root. */
1433 	lefttreemax->b_right = root->b_left;
1434 	righttreemin->b_left = root->b_right;
1435 	root->b_left = dummy.b_right;
1436 	root->b_right = dummy.b_left;
1437 	return (root);
1438 }
1439 
1440 static void
1441 buf_vlist_remove(struct buf *bp)
1442 {
1443 	struct buf *root;
1444 	struct bufv *bv;
1445 
1446 	KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp));
1447 	ASSERT_BO_LOCKED(bp->b_bufobj);
1448 	KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) !=
1449 	    (BX_VNDIRTY|BX_VNCLEAN),
1450 	    ("buf_vlist_remove: Buf %p is on two lists", bp));
1451 	if (bp->b_xflags & BX_VNDIRTY)
1452 		bv = &bp->b_bufobj->bo_dirty;
1453 	else
1454 		bv = &bp->b_bufobj->bo_clean;
1455 	if (bp != bv->bv_root) {
1456 		root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1457 		KASSERT(root == bp, ("splay lookup failed in remove"));
1458 	}
1459 	if (bp->b_left == NULL) {
1460 		root = bp->b_right;
1461 	} else {
1462 		root = buf_splay(bp->b_lblkno, bp->b_xflags, bp->b_left);
1463 		root->b_right = bp->b_right;
1464 	}
1465 	bv->bv_root = root;
1466 	TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs);
1467 	bv->bv_cnt--;
1468 	bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1469 }
1470 
1471 /*
1472  * Add the buffer to the sorted clean or dirty block list using a
1473  * splay tree algorithm.
1474  *
1475  * NOTE: xflags is passed as a constant, optimizing this inline function!
1476  */
1477 static void
1478 buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags)
1479 {
1480 	struct buf *root;
1481 	struct bufv *bv;
1482 
1483 	ASSERT_BO_LOCKED(bo);
1484 	KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0,
1485 	    ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags));
1486 	bp->b_xflags |= xflags;
1487 	if (xflags & BX_VNDIRTY)
1488 		bv = &bo->bo_dirty;
1489 	else
1490 		bv = &bo->bo_clean;
1491 
1492 	root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1493 	if (root == NULL) {
1494 		bp->b_left = NULL;
1495 		bp->b_right = NULL;
1496 		TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs);
1497 	} else if (bp->b_lblkno < root->b_lblkno ||
1498 	    (bp->b_lblkno == root->b_lblkno &&
1499 	    (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1500 		bp->b_left = root->b_left;
1501 		bp->b_right = root;
1502 		root->b_left = NULL;
1503 		TAILQ_INSERT_BEFORE(root, bp, b_bobufs);
1504 	} else {
1505 		bp->b_right = root->b_right;
1506 		bp->b_left = root;
1507 		root->b_right = NULL;
1508 		TAILQ_INSERT_AFTER(&bv->bv_hd, root, bp, b_bobufs);
1509 	}
1510 	bv->bv_cnt++;
1511 	bv->bv_root = bp;
1512 }
1513 
1514 /*
1515  * Lookup a buffer using the splay tree.  Note that we specifically avoid
1516  * shadow buffers used in background bitmap writes.
1517  *
1518  * This code isn't quite efficient as it could be because we are maintaining
1519  * two sorted lists and do not know which list the block resides in.
1520  *
1521  * During a "make buildworld" the desired buffer is found at one of
1522  * the roots more than 60% of the time.  Thus, checking both roots
1523  * before performing either splay eliminates unnecessary splays on the
1524  * first tree splayed.
1525  */
1526 struct buf *
1527 gbincore(struct bufobj *bo, daddr_t lblkno)
1528 {
1529 	struct buf *bp;
1530 
1531 	ASSERT_BO_LOCKED(bo);
1532 	if ((bp = bo->bo_clean.bv_root) != NULL &&
1533 	    bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1534 		return (bp);
1535 	if ((bp = bo->bo_dirty.bv_root) != NULL &&
1536 	    bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1537 		return (bp);
1538 	if ((bp = bo->bo_clean.bv_root) != NULL) {
1539 		bo->bo_clean.bv_root = bp = buf_splay(lblkno, 0, bp);
1540 		if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1541 			return (bp);
1542 	}
1543 	if ((bp = bo->bo_dirty.bv_root) != NULL) {
1544 		bo->bo_dirty.bv_root = bp = buf_splay(lblkno, 0, bp);
1545 		if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1546 			return (bp);
1547 	}
1548 	return (NULL);
1549 }
1550 
1551 /*
1552  * Associate a buffer with a vnode.
1553  */
1554 void
1555 bgetvp(struct vnode *vp, struct buf *bp)
1556 {
1557 	struct bufobj *bo;
1558 
1559 	bo = &vp->v_bufobj;
1560 	ASSERT_BO_LOCKED(bo);
1561 	VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free"));
1562 
1563 	CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags);
1564 	VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp,
1565 	    ("bgetvp: bp already attached! %p", bp));
1566 
1567 	vhold(vp);
1568 	if (VFS_NEEDSGIANT(vp->v_mount) || bo->bo_flag & BO_NEEDSGIANT)
1569 		bp->b_flags |= B_NEEDSGIANT;
1570 	bp->b_vp = vp;
1571 	bp->b_bufobj = bo;
1572 	/*
1573 	 * Insert onto list for new vnode.
1574 	 */
1575 	buf_vlist_add(bp, bo, BX_VNCLEAN);
1576 }
1577 
1578 /*
1579  * Disassociate a buffer from a vnode.
1580  */
1581 void
1582 brelvp(struct buf *bp)
1583 {
1584 	struct bufobj *bo;
1585 	struct vnode *vp;
1586 
1587 	CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
1588 	KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1589 
1590 	/*
1591 	 * Delete from old vnode list, if on one.
1592 	 */
1593 	vp = bp->b_vp;		/* XXX */
1594 	bo = bp->b_bufobj;
1595 	BO_LOCK(bo);
1596 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1597 		buf_vlist_remove(bp);
1598 	else
1599 		panic("brelvp: Buffer %p not on queue.", bp);
1600 	if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1601 		bo->bo_flag &= ~BO_ONWORKLST;
1602 		mtx_lock(&sync_mtx);
1603 		LIST_REMOVE(bo, bo_synclist);
1604 		syncer_worklist_len--;
1605 		mtx_unlock(&sync_mtx);
1606 	}
1607 	bp->b_flags &= ~B_NEEDSGIANT;
1608 	bp->b_vp = NULL;
1609 	bp->b_bufobj = NULL;
1610 	BO_UNLOCK(bo);
1611 	vdrop(vp);
1612 }
1613 
1614 /*
1615  * Add an item to the syncer work queue.
1616  */
1617 static void
1618 vn_syncer_add_to_worklist(struct bufobj *bo, int delay)
1619 {
1620 	int queue, slot;
1621 
1622 	ASSERT_BO_LOCKED(bo);
1623 
1624 	mtx_lock(&sync_mtx);
1625 	if (bo->bo_flag & BO_ONWORKLST)
1626 		LIST_REMOVE(bo, bo_synclist);
1627 	else {
1628 		bo->bo_flag |= BO_ONWORKLST;
1629 		syncer_worklist_len++;
1630 	}
1631 
1632 	if (delay > syncer_maxdelay - 2)
1633 		delay = syncer_maxdelay - 2;
1634 	slot = (syncer_delayno + delay) & syncer_mask;
1635 
1636 	queue = VFS_NEEDSGIANT(bo->__bo_vnode->v_mount) ? WI_GIANTQ :
1637 	    WI_MPSAFEQ;
1638 	LIST_INSERT_HEAD(&syncer_workitem_pending[queue][slot], bo,
1639 	    bo_synclist);
1640 	mtx_unlock(&sync_mtx);
1641 }
1642 
1643 static int
1644 sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS)
1645 {
1646 	int error, len;
1647 
1648 	mtx_lock(&sync_mtx);
1649 	len = syncer_worklist_len - sync_vnode_count;
1650 	mtx_unlock(&sync_mtx);
1651 	error = SYSCTL_OUT(req, &len, sizeof(len));
1652 	return (error);
1653 }
1654 
1655 SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0,
1656     sysctl_vfs_worklist_len, "I", "Syncer thread worklist length");
1657 
1658 static struct proc *updateproc;
1659 static void sched_sync(void);
1660 static struct kproc_desc up_kp = {
1661 	"syncer",
1662 	sched_sync,
1663 	&updateproc
1664 };
1665 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp);
1666 
1667 static int
1668 sync_vnode(struct synclist *slp, struct bufobj **bo, struct thread *td)
1669 {
1670 	struct vnode *vp;
1671 	struct mount *mp;
1672 
1673 	*bo = LIST_FIRST(slp);
1674 	if (*bo == NULL)
1675 		return (0);
1676 	vp = (*bo)->__bo_vnode;	/* XXX */
1677 	if (VOP_ISLOCKED(vp) != 0 || VI_TRYLOCK(vp) == 0)
1678 		return (1);
1679 	/*
1680 	 * We use vhold in case the vnode does not
1681 	 * successfully sync.  vhold prevents the vnode from
1682 	 * going away when we unlock the sync_mtx so that
1683 	 * we can acquire the vnode interlock.
1684 	 */
1685 	vholdl(vp);
1686 	mtx_unlock(&sync_mtx);
1687 	VI_UNLOCK(vp);
1688 	if (vn_start_write(vp, &mp, V_NOWAIT) != 0) {
1689 		vdrop(vp);
1690 		mtx_lock(&sync_mtx);
1691 		return (*bo == LIST_FIRST(slp));
1692 	}
1693 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1694 	(void) VOP_FSYNC(vp, MNT_LAZY, td);
1695 	VOP_UNLOCK(vp, 0);
1696 	vn_finished_write(mp);
1697 	BO_LOCK(*bo);
1698 	if (((*bo)->bo_flag & BO_ONWORKLST) != 0) {
1699 		/*
1700 		 * Put us back on the worklist.  The worklist
1701 		 * routine will remove us from our current
1702 		 * position and then add us back in at a later
1703 		 * position.
1704 		 */
1705 		vn_syncer_add_to_worklist(*bo, syncdelay);
1706 	}
1707 	BO_UNLOCK(*bo);
1708 	vdrop(vp);
1709 	mtx_lock(&sync_mtx);
1710 	return (0);
1711 }
1712 
1713 /*
1714  * System filesystem synchronizer daemon.
1715  */
1716 static void
1717 sched_sync(void)
1718 {
1719 	struct synclist *gnext, *next;
1720 	struct synclist *gslp, *slp;
1721 	struct bufobj *bo;
1722 	long starttime;
1723 	struct thread *td = curthread;
1724 	int last_work_seen;
1725 	int net_worklist_len;
1726 	int syncer_final_iter;
1727 	int first_printf;
1728 	int error;
1729 
1730 	last_work_seen = 0;
1731 	syncer_final_iter = 0;
1732 	first_printf = 1;
1733 	syncer_state = SYNCER_RUNNING;
1734 	starttime = time_uptime;
1735 	td->td_pflags |= TDP_NORUNNINGBUF;
1736 
1737 	EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc,
1738 	    SHUTDOWN_PRI_LAST);
1739 
1740 	mtx_lock(&sync_mtx);
1741 	for (;;) {
1742 		if (syncer_state == SYNCER_FINAL_DELAY &&
1743 		    syncer_final_iter == 0) {
1744 			mtx_unlock(&sync_mtx);
1745 			kproc_suspend_check(td->td_proc);
1746 			mtx_lock(&sync_mtx);
1747 		}
1748 		net_worklist_len = syncer_worklist_len - sync_vnode_count;
1749 		if (syncer_state != SYNCER_RUNNING &&
1750 		    starttime != time_uptime) {
1751 			if (first_printf) {
1752 				printf("\nSyncing disks, vnodes remaining...");
1753 				first_printf = 0;
1754 			}
1755 			printf("%d ", net_worklist_len);
1756 		}
1757 		starttime = time_uptime;
1758 
1759 		/*
1760 		 * Push files whose dirty time has expired.  Be careful
1761 		 * of interrupt race on slp queue.
1762 		 *
1763 		 * Skip over empty worklist slots when shutting down.
1764 		 */
1765 		do {
1766 			slp = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno];
1767 			gslp = &syncer_workitem_pending[WI_GIANTQ][syncer_delayno];
1768 			syncer_delayno += 1;
1769 			if (syncer_delayno == syncer_maxdelay)
1770 				syncer_delayno = 0;
1771 			next = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno];
1772 			gnext = &syncer_workitem_pending[WI_GIANTQ][syncer_delayno];
1773 			/*
1774 			 * If the worklist has wrapped since the
1775 			 * it was emptied of all but syncer vnodes,
1776 			 * switch to the FINAL_DELAY state and run
1777 			 * for one more second.
1778 			 */
1779 			if (syncer_state == SYNCER_SHUTTING_DOWN &&
1780 			    net_worklist_len == 0 &&
1781 			    last_work_seen == syncer_delayno) {
1782 				syncer_state = SYNCER_FINAL_DELAY;
1783 				syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP;
1784 			}
1785 		} while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) &&
1786 		    LIST_EMPTY(gslp) && syncer_worklist_len > 0);
1787 
1788 		/*
1789 		 * Keep track of the last time there was anything
1790 		 * on the worklist other than syncer vnodes.
1791 		 * Return to the SHUTTING_DOWN state if any
1792 		 * new work appears.
1793 		 */
1794 		if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING)
1795 			last_work_seen = syncer_delayno;
1796 		if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY)
1797 			syncer_state = SYNCER_SHUTTING_DOWN;
1798 		while (!LIST_EMPTY(slp)) {
1799 			error = sync_vnode(slp, &bo, td);
1800 			if (error == 1) {
1801 				LIST_REMOVE(bo, bo_synclist);
1802 				LIST_INSERT_HEAD(next, bo, bo_synclist);
1803 				continue;
1804 			}
1805 		}
1806 		if (!LIST_EMPTY(gslp)) {
1807 			mtx_unlock(&sync_mtx);
1808 			mtx_lock(&Giant);
1809 			mtx_lock(&sync_mtx);
1810 			while (!LIST_EMPTY(gslp)) {
1811 				error = sync_vnode(gslp, &bo, td);
1812 				if (error == 1) {
1813 					LIST_REMOVE(bo, bo_synclist);
1814 					LIST_INSERT_HEAD(gnext, bo,
1815 					    bo_synclist);
1816 					continue;
1817 				}
1818 			}
1819 			mtx_unlock(&Giant);
1820 		}
1821 		if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0)
1822 			syncer_final_iter--;
1823 		/*
1824 		 * The variable rushjob allows the kernel to speed up the
1825 		 * processing of the filesystem syncer process. A rushjob
1826 		 * value of N tells the filesystem syncer to process the next
1827 		 * N seconds worth of work on its queue ASAP. Currently rushjob
1828 		 * is used by the soft update code to speed up the filesystem
1829 		 * syncer process when the incore state is getting so far
1830 		 * ahead of the disk that the kernel memory pool is being
1831 		 * threatened with exhaustion.
1832 		 */
1833 		if (rushjob > 0) {
1834 			rushjob -= 1;
1835 			continue;
1836 		}
1837 		/*
1838 		 * Just sleep for a short period of time between
1839 		 * iterations when shutting down to allow some I/O
1840 		 * to happen.
1841 		 *
1842 		 * If it has taken us less than a second to process the
1843 		 * current work, then wait. Otherwise start right over
1844 		 * again. We can still lose time if any single round
1845 		 * takes more than two seconds, but it does not really
1846 		 * matter as we are just trying to generally pace the
1847 		 * filesystem activity.
1848 		 */
1849 		if (syncer_state != SYNCER_RUNNING)
1850 			cv_timedwait(&sync_wakeup, &sync_mtx,
1851 			    hz / SYNCER_SHUTDOWN_SPEEDUP);
1852 		else if (time_uptime == starttime)
1853 			cv_timedwait(&sync_wakeup, &sync_mtx, hz);
1854 	}
1855 }
1856 
1857 /*
1858  * Request the syncer daemon to speed up its work.
1859  * We never push it to speed up more than half of its
1860  * normal turn time, otherwise it could take over the cpu.
1861  */
1862 int
1863 speedup_syncer(void)
1864 {
1865 	int ret = 0;
1866 
1867 	mtx_lock(&sync_mtx);
1868 	if (rushjob < syncdelay / 2) {
1869 		rushjob += 1;
1870 		stat_rush_requests += 1;
1871 		ret = 1;
1872 	}
1873 	mtx_unlock(&sync_mtx);
1874 	cv_broadcast(&sync_wakeup);
1875 	return (ret);
1876 }
1877 
1878 /*
1879  * Tell the syncer to speed up its work and run though its work
1880  * list several times, then tell it to shut down.
1881  */
1882 static void
1883 syncer_shutdown(void *arg, int howto)
1884 {
1885 
1886 	if (howto & RB_NOSYNC)
1887 		return;
1888 	mtx_lock(&sync_mtx);
1889 	syncer_state = SYNCER_SHUTTING_DOWN;
1890 	rushjob = 0;
1891 	mtx_unlock(&sync_mtx);
1892 	cv_broadcast(&sync_wakeup);
1893 	kproc_shutdown(arg, howto);
1894 }
1895 
1896 /*
1897  * Reassign a buffer from one vnode to another.
1898  * Used to assign file specific control information
1899  * (indirect blocks) to the vnode to which they belong.
1900  */
1901 void
1902 reassignbuf(struct buf *bp)
1903 {
1904 	struct vnode *vp;
1905 	struct bufobj *bo;
1906 	int delay;
1907 #ifdef INVARIANTS
1908 	struct bufv *bv;
1909 #endif
1910 
1911 	vp = bp->b_vp;
1912 	bo = bp->b_bufobj;
1913 	++reassignbufcalls;
1914 
1915 	CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X",
1916 	    bp, bp->b_vp, bp->b_flags);
1917 	/*
1918 	 * B_PAGING flagged buffers cannot be reassigned because their vp
1919 	 * is not fully linked in.
1920 	 */
1921 	if (bp->b_flags & B_PAGING)
1922 		panic("cannot reassign paging buffer");
1923 
1924 	/*
1925 	 * Delete from old vnode list, if on one.
1926 	 */
1927 	BO_LOCK(bo);
1928 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1929 		buf_vlist_remove(bp);
1930 	else
1931 		panic("reassignbuf: Buffer %p not on queue.", bp);
1932 	/*
1933 	 * If dirty, put on list of dirty buffers; otherwise insert onto list
1934 	 * of clean buffers.
1935 	 */
1936 	if (bp->b_flags & B_DELWRI) {
1937 		if ((bo->bo_flag & BO_ONWORKLST) == 0) {
1938 			switch (vp->v_type) {
1939 			case VDIR:
1940 				delay = dirdelay;
1941 				break;
1942 			case VCHR:
1943 				delay = metadelay;
1944 				break;
1945 			default:
1946 				delay = filedelay;
1947 			}
1948 			vn_syncer_add_to_worklist(bo, delay);
1949 		}
1950 		buf_vlist_add(bp, bo, BX_VNDIRTY);
1951 	} else {
1952 		buf_vlist_add(bp, bo, BX_VNCLEAN);
1953 
1954 		if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1955 			mtx_lock(&sync_mtx);
1956 			LIST_REMOVE(bo, bo_synclist);
1957 			syncer_worklist_len--;
1958 			mtx_unlock(&sync_mtx);
1959 			bo->bo_flag &= ~BO_ONWORKLST;
1960 		}
1961 	}
1962 #ifdef INVARIANTS
1963 	bv = &bo->bo_clean;
1964 	bp = TAILQ_FIRST(&bv->bv_hd);
1965 	KASSERT(bp == NULL || bp->b_bufobj == bo,
1966 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1967 	bp = TAILQ_LAST(&bv->bv_hd, buflists);
1968 	KASSERT(bp == NULL || bp->b_bufobj == bo,
1969 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1970 	bv = &bo->bo_dirty;
1971 	bp = TAILQ_FIRST(&bv->bv_hd);
1972 	KASSERT(bp == NULL || bp->b_bufobj == bo,
1973 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1974 	bp = TAILQ_LAST(&bv->bv_hd, buflists);
1975 	KASSERT(bp == NULL || bp->b_bufobj == bo,
1976 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1977 #endif
1978 	BO_UNLOCK(bo);
1979 }
1980 
1981 /*
1982  * Increment the use and hold counts on the vnode, taking care to reference
1983  * the driver's usecount if this is a chardev.  The vholdl() will remove
1984  * the vnode from the free list if it is presently free.  Requires the
1985  * vnode interlock and returns with it held.
1986  */
1987 static void
1988 v_incr_usecount(struct vnode *vp)
1989 {
1990 
1991 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
1992 	vp->v_usecount++;
1993 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
1994 		dev_lock();
1995 		vp->v_rdev->si_usecount++;
1996 		dev_unlock();
1997 	}
1998 	vholdl(vp);
1999 }
2000 
2001 /*
2002  * Turn a holdcnt into a use+holdcnt such that only one call to
2003  * v_decr_usecount is needed.
2004  */
2005 static void
2006 v_upgrade_usecount(struct vnode *vp)
2007 {
2008 
2009 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2010 	vp->v_usecount++;
2011 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2012 		dev_lock();
2013 		vp->v_rdev->si_usecount++;
2014 		dev_unlock();
2015 	}
2016 }
2017 
2018 /*
2019  * Decrement the vnode use and hold count along with the driver's usecount
2020  * if this is a chardev.  The vdropl() below releases the vnode interlock
2021  * as it may free the vnode.
2022  */
2023 static void
2024 v_decr_usecount(struct vnode *vp)
2025 {
2026 
2027 	ASSERT_VI_LOCKED(vp, __FUNCTION__);
2028 	VNASSERT(vp->v_usecount > 0, vp,
2029 	    ("v_decr_usecount: negative usecount"));
2030 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2031 	vp->v_usecount--;
2032 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2033 		dev_lock();
2034 		vp->v_rdev->si_usecount--;
2035 		dev_unlock();
2036 	}
2037 	vdropl(vp);
2038 }
2039 
2040 /*
2041  * Decrement only the use count and driver use count.  This is intended to
2042  * be paired with a follow on vdropl() to release the remaining hold count.
2043  * In this way we may vgone() a vnode with a 0 usecount without risk of
2044  * having it end up on a free list because the hold count is kept above 0.
2045  */
2046 static void
2047 v_decr_useonly(struct vnode *vp)
2048 {
2049 
2050 	ASSERT_VI_LOCKED(vp, __FUNCTION__);
2051 	VNASSERT(vp->v_usecount > 0, vp,
2052 	    ("v_decr_useonly: negative usecount"));
2053 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2054 	vp->v_usecount--;
2055 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2056 		dev_lock();
2057 		vp->v_rdev->si_usecount--;
2058 		dev_unlock();
2059 	}
2060 }
2061 
2062 /*
2063  * Grab a particular vnode from the free list, increment its
2064  * reference count and lock it.  VI_DOOMED is set if the vnode
2065  * is being destroyed.  Only callers who specify LK_RETRY will
2066  * see doomed vnodes.  If inactive processing was delayed in
2067  * vput try to do it here.
2068  */
2069 int
2070 vget(struct vnode *vp, int flags, struct thread *td)
2071 {
2072 	int error;
2073 
2074 	error = 0;
2075 	VFS_ASSERT_GIANT(vp->v_mount);
2076 	VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
2077 	    ("vget: invalid lock operation"));
2078 	CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
2079 
2080 	if ((flags & LK_INTERLOCK) == 0)
2081 		VI_LOCK(vp);
2082 	vholdl(vp);
2083 	if ((error = vn_lock(vp, flags | LK_INTERLOCK)) != 0) {
2084 		vdrop(vp);
2085 		CTR2(KTR_VFS, "%s: impossible to lock vnode %p", __func__,
2086 		    vp);
2087 		return (error);
2088 	}
2089 	if (vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0)
2090 		panic("vget: vn_lock failed to return ENOENT\n");
2091 	VI_LOCK(vp);
2092 	/* Upgrade our holdcnt to a usecount. */
2093 	v_upgrade_usecount(vp);
2094 	/*
2095  	 * We don't guarantee that any particular close will
2096 	 * trigger inactive processing so just make a best effort
2097 	 * here at preventing a reference to a removed file.  If
2098 	 * we don't succeed no harm is done.
2099 	 */
2100 	if (vp->v_iflag & VI_OWEINACT) {
2101 		if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE &&
2102 		    (flags & LK_NOWAIT) == 0)
2103 			vinactive(vp, td);
2104 		vp->v_iflag &= ~VI_OWEINACT;
2105 	}
2106 	VI_UNLOCK(vp);
2107 	return (0);
2108 }
2109 
2110 /*
2111  * Increase the reference count of a vnode.
2112  */
2113 void
2114 vref(struct vnode *vp)
2115 {
2116 
2117 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2118 	VI_LOCK(vp);
2119 	v_incr_usecount(vp);
2120 	VI_UNLOCK(vp);
2121 }
2122 
2123 /*
2124  * Return reference count of a vnode.
2125  *
2126  * The results of this call are only guaranteed when some mechanism other
2127  * than the VI lock is used to stop other processes from gaining references
2128  * to the vnode.  This may be the case if the caller holds the only reference.
2129  * This is also useful when stale data is acceptable as race conditions may
2130  * be accounted for by some other means.
2131  */
2132 int
2133 vrefcnt(struct vnode *vp)
2134 {
2135 	int usecnt;
2136 
2137 	VI_LOCK(vp);
2138 	usecnt = vp->v_usecount;
2139 	VI_UNLOCK(vp);
2140 
2141 	return (usecnt);
2142 }
2143 
2144 
2145 /*
2146  * Vnode put/release.
2147  * If count drops to zero, call inactive routine and return to freelist.
2148  */
2149 void
2150 vrele(struct vnode *vp)
2151 {
2152 	struct thread *td = curthread;	/* XXX */
2153 
2154 	KASSERT(vp != NULL, ("vrele: null vp"));
2155 	VFS_ASSERT_GIANT(vp->v_mount);
2156 
2157 	VI_LOCK(vp);
2158 
2159 	/* Skip this v_writecount check if we're going to panic below. */
2160 	VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp,
2161 	    ("vrele: missed vn_close"));
2162 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2163 
2164 	if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2165 	    vp->v_usecount == 1)) {
2166 		v_decr_usecount(vp);
2167 		return;
2168 	}
2169 	if (vp->v_usecount != 1) {
2170 #ifdef DIAGNOSTIC
2171 		vprint("vrele: negative ref count", vp);
2172 #endif
2173 		VI_UNLOCK(vp);
2174 		panic("vrele: negative ref cnt");
2175 	}
2176 	CTR2(KTR_VFS, "%s: return vnode %p to the freelist", __func__, vp);
2177 	/*
2178 	 * We want to hold the vnode until the inactive finishes to
2179 	 * prevent vgone() races.  We drop the use count here and the
2180 	 * hold count below when we're done.
2181 	 */
2182 	v_decr_useonly(vp);
2183 	/*
2184 	 * We must call VOP_INACTIVE with the node locked. Mark
2185 	 * as VI_DOINGINACT to avoid recursion.
2186 	 */
2187 	vp->v_iflag |= VI_OWEINACT;
2188 	if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK) == 0) {
2189 		VI_LOCK(vp);
2190 		if (vp->v_usecount > 0)
2191 			vp->v_iflag &= ~VI_OWEINACT;
2192 		if (vp->v_iflag & VI_OWEINACT)
2193 			vinactive(vp, td);
2194 		VOP_UNLOCK(vp, 0);
2195 	} else {
2196 		VI_LOCK(vp);
2197 		if (vp->v_usecount > 0)
2198 			vp->v_iflag &= ~VI_OWEINACT;
2199 	}
2200 	vdropl(vp);
2201 }
2202 
2203 /*
2204  * Release an already locked vnode.  This give the same effects as
2205  * unlock+vrele(), but takes less time and avoids releasing and
2206  * re-aquiring the lock (as vrele() acquires the lock internally.)
2207  */
2208 void
2209 vput(struct vnode *vp)
2210 {
2211 	struct thread *td = curthread;	/* XXX */
2212 	int error;
2213 
2214 	KASSERT(vp != NULL, ("vput: null vp"));
2215 	ASSERT_VOP_LOCKED(vp, "vput");
2216 	VFS_ASSERT_GIANT(vp->v_mount);
2217 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2218 	VI_LOCK(vp);
2219 	/* Skip this v_writecount check if we're going to panic below. */
2220 	VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp,
2221 	    ("vput: missed vn_close"));
2222 	error = 0;
2223 
2224 	if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2225 	    vp->v_usecount == 1)) {
2226 		VOP_UNLOCK(vp, 0);
2227 		v_decr_usecount(vp);
2228 		return;
2229 	}
2230 
2231 	if (vp->v_usecount != 1) {
2232 #ifdef DIAGNOSTIC
2233 		vprint("vput: negative ref count", vp);
2234 #endif
2235 		panic("vput: negative ref cnt");
2236 	}
2237 	CTR2(KTR_VFS, "%s: return to freelist the vnode %p", __func__, vp);
2238 	/*
2239 	 * We want to hold the vnode until the inactive finishes to
2240 	 * prevent vgone() races.  We drop the use count here and the
2241 	 * hold count below when we're done.
2242 	 */
2243 	v_decr_useonly(vp);
2244 	vp->v_iflag |= VI_OWEINACT;
2245 	if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
2246 		error = VOP_LOCK(vp, LK_UPGRADE|LK_INTERLOCK|LK_NOWAIT);
2247 		VI_LOCK(vp);
2248 		if (error) {
2249 			if (vp->v_usecount > 0)
2250 				vp->v_iflag &= ~VI_OWEINACT;
2251 			goto done;
2252 		}
2253 	}
2254 	if (vp->v_usecount > 0)
2255 		vp->v_iflag &= ~VI_OWEINACT;
2256 	if (vp->v_iflag & VI_OWEINACT)
2257 		vinactive(vp, td);
2258 	VOP_UNLOCK(vp, 0);
2259 done:
2260 	vdropl(vp);
2261 }
2262 
2263 /*
2264  * Somebody doesn't want the vnode recycled.
2265  */
2266 void
2267 vhold(struct vnode *vp)
2268 {
2269 
2270 	VI_LOCK(vp);
2271 	vholdl(vp);
2272 	VI_UNLOCK(vp);
2273 }
2274 
2275 void
2276 vholdl(struct vnode *vp)
2277 {
2278 
2279 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2280 	vp->v_holdcnt++;
2281 	if (VSHOULDBUSY(vp))
2282 		vbusy(vp);
2283 }
2284 
2285 /*
2286  * Note that there is one less who cares about this vnode.  vdrop() is the
2287  * opposite of vhold().
2288  */
2289 void
2290 vdrop(struct vnode *vp)
2291 {
2292 
2293 	VI_LOCK(vp);
2294 	vdropl(vp);
2295 }
2296 
2297 /*
2298  * Drop the hold count of the vnode.  If this is the last reference to
2299  * the vnode we will free it if it has been vgone'd otherwise it is
2300  * placed on the free list.
2301  */
2302 void
2303 vdropl(struct vnode *vp)
2304 {
2305 
2306 	ASSERT_VI_LOCKED(vp, "vdropl");
2307 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2308 	if (vp->v_holdcnt <= 0)
2309 		panic("vdrop: holdcnt %d", vp->v_holdcnt);
2310 	vp->v_holdcnt--;
2311 	if (vp->v_holdcnt == 0) {
2312 		if (vp->v_iflag & VI_DOOMED) {
2313 			CTR2(KTR_VFS, "%s: destroying the vnode %p", __func__,
2314 			    vp);
2315 			vdestroy(vp);
2316 			return;
2317 		} else
2318 			vfree(vp);
2319 	}
2320 	VI_UNLOCK(vp);
2321 }
2322 
2323 /*
2324  * Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT
2325  * flags.  DOINGINACT prevents us from recursing in calls to vinactive.
2326  * OWEINACT tracks whether a vnode missed a call to inactive due to a
2327  * failed lock upgrade.
2328  */
2329 static void
2330 vinactive(struct vnode *vp, struct thread *td)
2331 {
2332 
2333 	ASSERT_VOP_ELOCKED(vp, "vinactive");
2334 	ASSERT_VI_LOCKED(vp, "vinactive");
2335 	VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp,
2336 	    ("vinactive: recursed on VI_DOINGINACT"));
2337 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2338 	vp->v_iflag |= VI_DOINGINACT;
2339 	vp->v_iflag &= ~VI_OWEINACT;
2340 	VI_UNLOCK(vp);
2341 	VOP_INACTIVE(vp, td);
2342 	VI_LOCK(vp);
2343 	VNASSERT(vp->v_iflag & VI_DOINGINACT, vp,
2344 	    ("vinactive: lost VI_DOINGINACT"));
2345 	vp->v_iflag &= ~VI_DOINGINACT;
2346 }
2347 
2348 /*
2349  * Remove any vnodes in the vnode table belonging to mount point mp.
2350  *
2351  * If FORCECLOSE is not specified, there should not be any active ones,
2352  * return error if any are found (nb: this is a user error, not a
2353  * system error). If FORCECLOSE is specified, detach any active vnodes
2354  * that are found.
2355  *
2356  * If WRITECLOSE is set, only flush out regular file vnodes open for
2357  * writing.
2358  *
2359  * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
2360  *
2361  * `rootrefs' specifies the base reference count for the root vnode
2362  * of this filesystem. The root vnode is considered busy if its
2363  * v_usecount exceeds this value. On a successful return, vflush(, td)
2364  * will call vrele() on the root vnode exactly rootrefs times.
2365  * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
2366  * be zero.
2367  */
2368 #ifdef DIAGNOSTIC
2369 static int busyprt = 0;		/* print out busy vnodes */
2370 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
2371 #endif
2372 
2373 int
2374 vflush( struct mount *mp, int rootrefs, int flags, struct thread *td)
2375 {
2376 	struct vnode *vp, *mvp, *rootvp = NULL;
2377 	struct vattr vattr;
2378 	int busy = 0, error;
2379 
2380 	CTR4(KTR_VFS, "%s: mp %p with rootrefs %d and flags %d", __func__, mp,
2381 	    rootrefs, flags);
2382 	if (rootrefs > 0) {
2383 		KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
2384 		    ("vflush: bad args"));
2385 		/*
2386 		 * Get the filesystem root vnode. We can vput() it
2387 		 * immediately, since with rootrefs > 0, it won't go away.
2388 		 */
2389 		if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp)) != 0) {
2390 			CTR2(KTR_VFS, "%s: vfs_root lookup failed with %d",
2391 			    __func__, error);
2392 			return (error);
2393 		}
2394 		vput(rootvp);
2395 
2396 	}
2397 	MNT_ILOCK(mp);
2398 loop:
2399 	MNT_VNODE_FOREACH(vp, mp, mvp) {
2400 
2401 		VI_LOCK(vp);
2402 		vholdl(vp);
2403 		MNT_IUNLOCK(mp);
2404 		error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE);
2405 		if (error) {
2406 			vdrop(vp);
2407 			MNT_ILOCK(mp);
2408 			MNT_VNODE_FOREACH_ABORT_ILOCKED(mp, mvp);
2409 			goto loop;
2410 		}
2411 		/*
2412 		 * Skip over a vnodes marked VV_SYSTEM.
2413 		 */
2414 		if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
2415 			VOP_UNLOCK(vp, 0);
2416 			vdrop(vp);
2417 			MNT_ILOCK(mp);
2418 			continue;
2419 		}
2420 		/*
2421 		 * If WRITECLOSE is set, flush out unlinked but still open
2422 		 * files (even if open only for reading) and regular file
2423 		 * vnodes open for writing.
2424 		 */
2425 		if (flags & WRITECLOSE) {
2426 			error = VOP_GETATTR(vp, &vattr, td->td_ucred);
2427 			VI_LOCK(vp);
2428 
2429 			if ((vp->v_type == VNON ||
2430 			    (error == 0 && vattr.va_nlink > 0)) &&
2431 			    (vp->v_writecount == 0 || vp->v_type != VREG)) {
2432 				VOP_UNLOCK(vp, 0);
2433 				vdropl(vp);
2434 				MNT_ILOCK(mp);
2435 				continue;
2436 			}
2437 		} else
2438 			VI_LOCK(vp);
2439 		/*
2440 		 * With v_usecount == 0, all we need to do is clear out the
2441 		 * vnode data structures and we are done.
2442 		 *
2443 		 * If FORCECLOSE is set, forcibly close the vnode.
2444 		 */
2445 		if (vp->v_usecount == 0 || (flags & FORCECLOSE)) {
2446 			VNASSERT(vp->v_usecount == 0 ||
2447 			    (vp->v_type != VCHR && vp->v_type != VBLK), vp,
2448 			    ("device VNODE %p is FORCECLOSED", vp));
2449 			vgonel(vp);
2450 		} else {
2451 			busy++;
2452 #ifdef DIAGNOSTIC
2453 			if (busyprt)
2454 				vprint("vflush: busy vnode", vp);
2455 #endif
2456 		}
2457 		VOP_UNLOCK(vp, 0);
2458 		vdropl(vp);
2459 		MNT_ILOCK(mp);
2460 	}
2461 	MNT_IUNLOCK(mp);
2462 	if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
2463 		/*
2464 		 * If just the root vnode is busy, and if its refcount
2465 		 * is equal to `rootrefs', then go ahead and kill it.
2466 		 */
2467 		VI_LOCK(rootvp);
2468 		KASSERT(busy > 0, ("vflush: not busy"));
2469 		VNASSERT(rootvp->v_usecount >= rootrefs, rootvp,
2470 		    ("vflush: usecount %d < rootrefs %d",
2471 		     rootvp->v_usecount, rootrefs));
2472 		if (busy == 1 && rootvp->v_usecount == rootrefs) {
2473 			VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK);
2474 			vgone(rootvp);
2475 			VOP_UNLOCK(rootvp, 0);
2476 			busy = 0;
2477 		} else
2478 			VI_UNLOCK(rootvp);
2479 	}
2480 	if (busy) {
2481 		CTR2(KTR_VFS, "%s: failing as %d vnodes are busy", __func__,
2482 		    busy);
2483 		return (EBUSY);
2484 	}
2485 	for (; rootrefs > 0; rootrefs--)
2486 		vrele(rootvp);
2487 	return (0);
2488 }
2489 
2490 /*
2491  * Recycle an unused vnode to the front of the free list.
2492  */
2493 int
2494 vrecycle(struct vnode *vp, struct thread *td)
2495 {
2496 	int recycled;
2497 
2498 	ASSERT_VOP_ELOCKED(vp, "vrecycle");
2499 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2500 	recycled = 0;
2501 	VI_LOCK(vp);
2502 	if (vp->v_usecount == 0) {
2503 		recycled = 1;
2504 		vgonel(vp);
2505 	}
2506 	VI_UNLOCK(vp);
2507 	return (recycled);
2508 }
2509 
2510 /*
2511  * Eliminate all activity associated with a vnode
2512  * in preparation for reuse.
2513  */
2514 void
2515 vgone(struct vnode *vp)
2516 {
2517 	VI_LOCK(vp);
2518 	vgonel(vp);
2519 	VI_UNLOCK(vp);
2520 }
2521 
2522 /*
2523  * vgone, with the vp interlock held.
2524  */
2525 void
2526 vgonel(struct vnode *vp)
2527 {
2528 	struct thread *td;
2529 	int oweinact;
2530 	int active;
2531 	struct mount *mp;
2532 
2533 	ASSERT_VOP_ELOCKED(vp, "vgonel");
2534 	ASSERT_VI_LOCKED(vp, "vgonel");
2535 	VNASSERT(vp->v_holdcnt, vp,
2536 	    ("vgonel: vp %p has no reference.", vp));
2537 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2538 	td = curthread;
2539 
2540 	/*
2541 	 * Don't vgonel if we're already doomed.
2542 	 */
2543 	if (vp->v_iflag & VI_DOOMED)
2544 		return;
2545 	vp->v_iflag |= VI_DOOMED;
2546 	/*
2547 	 * Check to see if the vnode is in use.  If so, we have to call
2548 	 * VOP_CLOSE() and VOP_INACTIVE().
2549 	 */
2550 	active = vp->v_usecount;
2551 	oweinact = (vp->v_iflag & VI_OWEINACT);
2552 	VI_UNLOCK(vp);
2553 	/*
2554 	 * Clean out any buffers associated with the vnode.
2555 	 * If the flush fails, just toss the buffers.
2556 	 */
2557 	mp = NULL;
2558 	if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd))
2559 		(void) vn_start_secondary_write(vp, &mp, V_WAIT);
2560 	if (vinvalbuf(vp, V_SAVE, 0, 0) != 0)
2561 		vinvalbuf(vp, 0, 0, 0);
2562 
2563 	/*
2564 	 * If purging an active vnode, it must be closed and
2565 	 * deactivated before being reclaimed.
2566 	 */
2567 	if (active)
2568 		VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
2569 	if (oweinact || active) {
2570 		VI_LOCK(vp);
2571 		if ((vp->v_iflag & VI_DOINGINACT) == 0)
2572 			vinactive(vp, td);
2573 		VI_UNLOCK(vp);
2574 	}
2575 	/*
2576 	 * Reclaim the vnode.
2577 	 */
2578 	if (VOP_RECLAIM(vp, td))
2579 		panic("vgone: cannot reclaim");
2580 	if (mp != NULL)
2581 		vn_finished_secondary_write(mp);
2582 	VNASSERT(vp->v_object == NULL, vp,
2583 	    ("vop_reclaim left v_object vp=%p, tag=%s", vp, vp->v_tag));
2584 	/*
2585 	 * Clear the advisory locks and wake up waiting threads.
2586 	 */
2587 	lf_purgelocks(vp, &(vp->v_lockf));
2588 	/*
2589 	 * Delete from old mount point vnode list.
2590 	 */
2591 	delmntque(vp);
2592 	cache_purge(vp);
2593 	/*
2594 	 * Done with purge, reset to the standard lock and invalidate
2595 	 * the vnode.
2596 	 */
2597 	VI_LOCK(vp);
2598 	vp->v_vnlock = &vp->v_lock;
2599 	vp->v_op = &dead_vnodeops;
2600 	vp->v_tag = "none";
2601 	vp->v_type = VBAD;
2602 }
2603 
2604 /*
2605  * Calculate the total number of references to a special device.
2606  */
2607 int
2608 vcount(struct vnode *vp)
2609 {
2610 	int count;
2611 
2612 	dev_lock();
2613 	count = vp->v_rdev->si_usecount;
2614 	dev_unlock();
2615 	return (count);
2616 }
2617 
2618 /*
2619  * Same as above, but using the struct cdev *as argument
2620  */
2621 int
2622 count_dev(struct cdev *dev)
2623 {
2624 	int count;
2625 
2626 	dev_lock();
2627 	count = dev->si_usecount;
2628 	dev_unlock();
2629 	return(count);
2630 }
2631 
2632 /*
2633  * Print out a description of a vnode.
2634  */
2635 static char *typename[] =
2636 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD",
2637  "VMARKER"};
2638 
2639 void
2640 vn_printf(struct vnode *vp, const char *fmt, ...)
2641 {
2642 	va_list ap;
2643 	char buf[256], buf2[16];
2644 	u_long flags;
2645 
2646 	va_start(ap, fmt);
2647 	vprintf(fmt, ap);
2648 	va_end(ap);
2649 	printf("%p: ", (void *)vp);
2650 	printf("tag %s, type %s\n", vp->v_tag, typename[vp->v_type]);
2651 	printf("    usecount %d, writecount %d, refcount %d mountedhere %p\n",
2652 	    vp->v_usecount, vp->v_writecount, vp->v_holdcnt, vp->v_mountedhere);
2653 	buf[0] = '\0';
2654 	buf[1] = '\0';
2655 	if (vp->v_vflag & VV_ROOT)
2656 		strlcat(buf, "|VV_ROOT", sizeof(buf));
2657 	if (vp->v_vflag & VV_ISTTY)
2658 		strlcat(buf, "|VV_ISTTY", sizeof(buf));
2659 	if (vp->v_vflag & VV_NOSYNC)
2660 		strlcat(buf, "|VV_NOSYNC", sizeof(buf));
2661 	if (vp->v_vflag & VV_CACHEDLABEL)
2662 		strlcat(buf, "|VV_CACHEDLABEL", sizeof(buf));
2663 	if (vp->v_vflag & VV_TEXT)
2664 		strlcat(buf, "|VV_TEXT", sizeof(buf));
2665 	if (vp->v_vflag & VV_COPYONWRITE)
2666 		strlcat(buf, "|VV_COPYONWRITE", sizeof(buf));
2667 	if (vp->v_vflag & VV_SYSTEM)
2668 		strlcat(buf, "|VV_SYSTEM", sizeof(buf));
2669 	if (vp->v_vflag & VV_PROCDEP)
2670 		strlcat(buf, "|VV_PROCDEP", sizeof(buf));
2671 	if (vp->v_vflag & VV_NOKNOTE)
2672 		strlcat(buf, "|VV_NOKNOTE", sizeof(buf));
2673 	if (vp->v_vflag & VV_DELETED)
2674 		strlcat(buf, "|VV_DELETED", sizeof(buf));
2675 	if (vp->v_vflag & VV_MD)
2676 		strlcat(buf, "|VV_MD", sizeof(buf));
2677 	flags = vp->v_vflag & ~(VV_ROOT | VV_ISTTY | VV_NOSYNC |
2678 	    VV_CACHEDLABEL | VV_TEXT | VV_COPYONWRITE | VV_SYSTEM | VV_PROCDEP |
2679 	    VV_NOKNOTE | VV_DELETED | VV_MD);
2680 	if (flags != 0) {
2681 		snprintf(buf2, sizeof(buf2), "|VV(0x%lx)", flags);
2682 		strlcat(buf, buf2, sizeof(buf));
2683 	}
2684 	if (vp->v_iflag & VI_MOUNT)
2685 		strlcat(buf, "|VI_MOUNT", sizeof(buf));
2686 	if (vp->v_iflag & VI_AGE)
2687 		strlcat(buf, "|VI_AGE", sizeof(buf));
2688 	if (vp->v_iflag & VI_DOOMED)
2689 		strlcat(buf, "|VI_DOOMED", sizeof(buf));
2690 	if (vp->v_iflag & VI_FREE)
2691 		strlcat(buf, "|VI_FREE", sizeof(buf));
2692 	if (vp->v_iflag & VI_OBJDIRTY)
2693 		strlcat(buf, "|VI_OBJDIRTY", sizeof(buf));
2694 	if (vp->v_iflag & VI_DOINGINACT)
2695 		strlcat(buf, "|VI_DOINGINACT", sizeof(buf));
2696 	if (vp->v_iflag & VI_OWEINACT)
2697 		strlcat(buf, "|VI_OWEINACT", sizeof(buf));
2698 	flags = vp->v_iflag & ~(VI_MOUNT | VI_AGE | VI_DOOMED | VI_FREE |
2699 	    VI_OBJDIRTY | VI_DOINGINACT | VI_OWEINACT);
2700 	if (flags != 0) {
2701 		snprintf(buf2, sizeof(buf2), "|VI(0x%lx)", flags);
2702 		strlcat(buf, buf2, sizeof(buf));
2703 	}
2704 	printf("    flags (%s)\n", buf + 1);
2705 	if (mtx_owned(VI_MTX(vp)))
2706 		printf(" VI_LOCKed");
2707 	if (vp->v_object != NULL)
2708 		printf("    v_object %p ref %d pages %d\n",
2709 		    vp->v_object, vp->v_object->ref_count,
2710 		    vp->v_object->resident_page_count);
2711 	printf("    ");
2712 	lockmgr_printinfo(vp->v_vnlock);
2713 	if (vp->v_data != NULL)
2714 		VOP_PRINT(vp);
2715 }
2716 
2717 #ifdef DDB
2718 /*
2719  * List all of the locked vnodes in the system.
2720  * Called when debugging the kernel.
2721  */
2722 DB_SHOW_COMMAND(lockedvnods, lockedvnodes)
2723 {
2724 	struct mount *mp, *nmp;
2725 	struct vnode *vp;
2726 
2727 	/*
2728 	 * Note: because this is DDB, we can't obey the locking semantics
2729 	 * for these structures, which means we could catch an inconsistent
2730 	 * state and dereference a nasty pointer.  Not much to be done
2731 	 * about that.
2732 	 */
2733 	db_printf("Locked vnodes\n");
2734 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2735 		nmp = TAILQ_NEXT(mp, mnt_list);
2736 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2737 			if (vp->v_type != VMARKER &&
2738 			    VOP_ISLOCKED(vp))
2739 				vprint("", vp);
2740 		}
2741 		nmp = TAILQ_NEXT(mp, mnt_list);
2742 	}
2743 }
2744 
2745 /*
2746  * Show details about the given vnode.
2747  */
2748 DB_SHOW_COMMAND(vnode, db_show_vnode)
2749 {
2750 	struct vnode *vp;
2751 
2752 	if (!have_addr)
2753 		return;
2754 	vp = (struct vnode *)addr;
2755 	vn_printf(vp, "vnode ");
2756 }
2757 
2758 /*
2759  * Show details about the given mount point.
2760  */
2761 DB_SHOW_COMMAND(mount, db_show_mount)
2762 {
2763 	struct mount *mp;
2764 	struct statfs *sp;
2765 	struct vnode *vp;
2766 	char buf[512];
2767 	u_int flags;
2768 
2769 	if (!have_addr) {
2770 		/* No address given, print short info about all mount points. */
2771 		TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2772 			db_printf("%p %s on %s (%s)\n", mp,
2773 			    mp->mnt_stat.f_mntfromname,
2774 			    mp->mnt_stat.f_mntonname,
2775 			    mp->mnt_stat.f_fstypename);
2776 			if (db_pager_quit)
2777 				break;
2778 		}
2779 		db_printf("\nMore info: show mount <addr>\n");
2780 		return;
2781 	}
2782 
2783 	mp = (struct mount *)addr;
2784 	db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname,
2785 	    mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename);
2786 
2787 	buf[0] = '\0';
2788 	flags = mp->mnt_flag;
2789 #define	MNT_FLAG(flag)	do {						\
2790 	if (flags & (flag)) {						\
2791 		if (buf[0] != '\0')					\
2792 			strlcat(buf, ", ", sizeof(buf));		\
2793 		strlcat(buf, (#flag) + 4, sizeof(buf));			\
2794 		flags &= ~(flag);					\
2795 	}								\
2796 } while (0)
2797 	MNT_FLAG(MNT_RDONLY);
2798 	MNT_FLAG(MNT_SYNCHRONOUS);
2799 	MNT_FLAG(MNT_NOEXEC);
2800 	MNT_FLAG(MNT_NOSUID);
2801 	MNT_FLAG(MNT_UNION);
2802 	MNT_FLAG(MNT_ASYNC);
2803 	MNT_FLAG(MNT_SUIDDIR);
2804 	MNT_FLAG(MNT_SOFTDEP);
2805 	MNT_FLAG(MNT_NOSYMFOLLOW);
2806 	MNT_FLAG(MNT_GJOURNAL);
2807 	MNT_FLAG(MNT_MULTILABEL);
2808 	MNT_FLAG(MNT_ACLS);
2809 	MNT_FLAG(MNT_NOATIME);
2810 	MNT_FLAG(MNT_NOCLUSTERR);
2811 	MNT_FLAG(MNT_NOCLUSTERW);
2812 	MNT_FLAG(MNT_EXRDONLY);
2813 	MNT_FLAG(MNT_EXPORTED);
2814 	MNT_FLAG(MNT_DEFEXPORTED);
2815 	MNT_FLAG(MNT_EXPORTANON);
2816 	MNT_FLAG(MNT_EXKERB);
2817 	MNT_FLAG(MNT_EXPUBLIC);
2818 	MNT_FLAG(MNT_LOCAL);
2819 	MNT_FLAG(MNT_QUOTA);
2820 	MNT_FLAG(MNT_ROOTFS);
2821 	MNT_FLAG(MNT_USER);
2822 	MNT_FLAG(MNT_IGNORE);
2823 	MNT_FLAG(MNT_UPDATE);
2824 	MNT_FLAG(MNT_DELEXPORT);
2825 	MNT_FLAG(MNT_RELOAD);
2826 	MNT_FLAG(MNT_FORCE);
2827 	MNT_FLAG(MNT_SNAPSHOT);
2828 	MNT_FLAG(MNT_BYFSID);
2829 #undef MNT_FLAG
2830 	if (flags != 0) {
2831 		if (buf[0] != '\0')
2832 			strlcat(buf, ", ", sizeof(buf));
2833 		snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
2834 		    "0x%08x", flags);
2835 	}
2836 	db_printf("    mnt_flag = %s\n", buf);
2837 
2838 	buf[0] = '\0';
2839 	flags = mp->mnt_kern_flag;
2840 #define	MNT_KERN_FLAG(flag)	do {					\
2841 	if (flags & (flag)) {						\
2842 		if (buf[0] != '\0')					\
2843 			strlcat(buf, ", ", sizeof(buf));		\
2844 		strlcat(buf, (#flag) + 5, sizeof(buf));			\
2845 		flags &= ~(flag);					\
2846 	}								\
2847 } while (0)
2848 	MNT_KERN_FLAG(MNTK_UNMOUNTF);
2849 	MNT_KERN_FLAG(MNTK_ASYNC);
2850 	MNT_KERN_FLAG(MNTK_SOFTDEP);
2851 	MNT_KERN_FLAG(MNTK_NOINSMNTQ);
2852 	MNT_KERN_FLAG(MNTK_UNMOUNT);
2853 	MNT_KERN_FLAG(MNTK_MWAIT);
2854 	MNT_KERN_FLAG(MNTK_SUSPEND);
2855 	MNT_KERN_FLAG(MNTK_SUSPEND2);
2856 	MNT_KERN_FLAG(MNTK_SUSPENDED);
2857 	MNT_KERN_FLAG(MNTK_MPSAFE);
2858 	MNT_KERN_FLAG(MNTK_NOKNOTE);
2859 	MNT_KERN_FLAG(MNTK_LOOKUP_SHARED);
2860 #undef MNT_KERN_FLAG
2861 	if (flags != 0) {
2862 		if (buf[0] != '\0')
2863 			strlcat(buf, ", ", sizeof(buf));
2864 		snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
2865 		    "0x%08x", flags);
2866 	}
2867 	db_printf("    mnt_kern_flag = %s\n", buf);
2868 
2869 	sp = &mp->mnt_stat;
2870 	db_printf("    mnt_stat = { version=%u type=%u flags=0x%016jx "
2871 	    "bsize=%ju iosize=%ju blocks=%ju bfree=%ju bavail=%jd files=%ju "
2872 	    "ffree=%jd syncwrites=%ju asyncwrites=%ju syncreads=%ju "
2873 	    "asyncreads=%ju namemax=%u owner=%u fsid=[%d, %d] }\n",
2874 	    (u_int)sp->f_version, (u_int)sp->f_type, (uintmax_t)sp->f_flags,
2875 	    (uintmax_t)sp->f_bsize, (uintmax_t)sp->f_iosize,
2876 	    (uintmax_t)sp->f_blocks, (uintmax_t)sp->f_bfree,
2877 	    (intmax_t)sp->f_bavail, (uintmax_t)sp->f_files,
2878 	    (intmax_t)sp->f_ffree, (uintmax_t)sp->f_syncwrites,
2879 	    (uintmax_t)sp->f_asyncwrites, (uintmax_t)sp->f_syncreads,
2880 	    (uintmax_t)sp->f_asyncreads, (u_int)sp->f_namemax,
2881 	    (u_int)sp->f_owner, (int)sp->f_fsid.val[0], (int)sp->f_fsid.val[1]);
2882 
2883 	db_printf("    mnt_cred = { uid=%u ruid=%u",
2884 	    (u_int)mp->mnt_cred->cr_uid, (u_int)mp->mnt_cred->cr_ruid);
2885 	if (jailed(mp->mnt_cred))
2886 		db_printf(", jail=%d", mp->mnt_cred->cr_prison->pr_id);
2887 	db_printf(" }\n");
2888 	db_printf("    mnt_ref = %d\n", mp->mnt_ref);
2889 	db_printf("    mnt_gen = %d\n", mp->mnt_gen);
2890 	db_printf("    mnt_nvnodelistsize = %d\n", mp->mnt_nvnodelistsize);
2891 	db_printf("    mnt_writeopcount = %d\n", mp->mnt_writeopcount);
2892 	db_printf("    mnt_noasync = %u\n", mp->mnt_noasync);
2893 	db_printf("    mnt_maxsymlinklen = %d\n", mp->mnt_maxsymlinklen);
2894 	db_printf("    mnt_iosize_max = %d\n", mp->mnt_iosize_max);
2895 	db_printf("    mnt_hashseed = %u\n", mp->mnt_hashseed);
2896 	db_printf("    mnt_secondary_writes = %d\n", mp->mnt_secondary_writes);
2897 	db_printf("    mnt_secondary_accwrites = %d\n",
2898 	    mp->mnt_secondary_accwrites);
2899 	db_printf("    mnt_gjprovider = %s\n",
2900 	    mp->mnt_gjprovider != NULL ? mp->mnt_gjprovider : "NULL");
2901 	db_printf("\n");
2902 
2903 	TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2904 		if (vp->v_type != VMARKER) {
2905 			vn_printf(vp, "vnode ");
2906 			if (db_pager_quit)
2907 				break;
2908 		}
2909 	}
2910 }
2911 #endif	/* DDB */
2912 
2913 /*
2914  * Fill in a struct xvfsconf based on a struct vfsconf.
2915  */
2916 static void
2917 vfsconf2x(struct vfsconf *vfsp, struct xvfsconf *xvfsp)
2918 {
2919 
2920 	strcpy(xvfsp->vfc_name, vfsp->vfc_name);
2921 	xvfsp->vfc_typenum = vfsp->vfc_typenum;
2922 	xvfsp->vfc_refcount = vfsp->vfc_refcount;
2923 	xvfsp->vfc_flags = vfsp->vfc_flags;
2924 	/*
2925 	 * These are unused in userland, we keep them
2926 	 * to not break binary compatibility.
2927 	 */
2928 	xvfsp->vfc_vfsops = NULL;
2929 	xvfsp->vfc_next = NULL;
2930 }
2931 
2932 /*
2933  * Top level filesystem related information gathering.
2934  */
2935 static int
2936 sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS)
2937 {
2938 	struct vfsconf *vfsp;
2939 	struct xvfsconf xvfsp;
2940 	int error;
2941 
2942 	error = 0;
2943 	TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
2944 		bzero(&xvfsp, sizeof(xvfsp));
2945 		vfsconf2x(vfsp, &xvfsp);
2946 		error = SYSCTL_OUT(req, &xvfsp, sizeof xvfsp);
2947 		if (error)
2948 			break;
2949 	}
2950 	return (error);
2951 }
2952 
2953 SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLFLAG_RD, NULL, 0, sysctl_vfs_conflist,
2954     "S,xvfsconf", "List of all configured filesystems");
2955 
2956 #ifndef BURN_BRIDGES
2957 static int	sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
2958 
2959 static int
2960 vfs_sysctl(SYSCTL_HANDLER_ARGS)
2961 {
2962 	int *name = (int *)arg1 - 1;	/* XXX */
2963 	u_int namelen = arg2 + 1;	/* XXX */
2964 	struct vfsconf *vfsp;
2965 	struct xvfsconf xvfsp;
2966 
2967 	printf("WARNING: userland calling deprecated sysctl, "
2968 	    "please rebuild world\n");
2969 
2970 #if 1 || defined(COMPAT_PRELITE2)
2971 	/* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
2972 	if (namelen == 1)
2973 		return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
2974 #endif
2975 
2976 	switch (name[1]) {
2977 	case VFS_MAXTYPENUM:
2978 		if (namelen != 2)
2979 			return (ENOTDIR);
2980 		return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
2981 	case VFS_CONF:
2982 		if (namelen != 3)
2983 			return (ENOTDIR);	/* overloaded */
2984 		TAILQ_FOREACH(vfsp, &vfsconf, vfc_list)
2985 			if (vfsp->vfc_typenum == name[2])
2986 				break;
2987 		if (vfsp == NULL)
2988 			return (EOPNOTSUPP);
2989 		bzero(&xvfsp, sizeof(xvfsp));
2990 		vfsconf2x(vfsp, &xvfsp);
2991 		return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
2992 	}
2993 	return (EOPNOTSUPP);
2994 }
2995 
2996 static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP,
2997 	vfs_sysctl, "Generic filesystem");
2998 
2999 #if 1 || defined(COMPAT_PRELITE2)
3000 
3001 static int
3002 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
3003 {
3004 	int error;
3005 	struct vfsconf *vfsp;
3006 	struct ovfsconf ovfs;
3007 
3008 	TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3009 		bzero(&ovfs, sizeof(ovfs));
3010 		ovfs.vfc_vfsops = vfsp->vfc_vfsops;	/* XXX used as flag */
3011 		strcpy(ovfs.vfc_name, vfsp->vfc_name);
3012 		ovfs.vfc_index = vfsp->vfc_typenum;
3013 		ovfs.vfc_refcount = vfsp->vfc_refcount;
3014 		ovfs.vfc_flags = vfsp->vfc_flags;
3015 		error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
3016 		if (error)
3017 			return error;
3018 	}
3019 	return 0;
3020 }
3021 
3022 #endif /* 1 || COMPAT_PRELITE2 */
3023 #endif /* !BURN_BRIDGES */
3024 
3025 #define KINFO_VNODESLOP		10
3026 #ifdef notyet
3027 /*
3028  * Dump vnode list (via sysctl).
3029  */
3030 /* ARGSUSED */
3031 static int
3032 sysctl_vnode(SYSCTL_HANDLER_ARGS)
3033 {
3034 	struct xvnode *xvn;
3035 	struct mount *mp;
3036 	struct vnode *vp;
3037 	int error, len, n;
3038 
3039 	/*
3040 	 * Stale numvnodes access is not fatal here.
3041 	 */
3042 	req->lock = 0;
3043 	len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn;
3044 	if (!req->oldptr)
3045 		/* Make an estimate */
3046 		return (SYSCTL_OUT(req, 0, len));
3047 
3048 	error = sysctl_wire_old_buffer(req, 0);
3049 	if (error != 0)
3050 		return (error);
3051 	xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK);
3052 	n = 0;
3053 	mtx_lock(&mountlist_mtx);
3054 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3055 		if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK))
3056 			continue;
3057 		MNT_ILOCK(mp);
3058 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3059 			if (n == len)
3060 				break;
3061 			vref(vp);
3062 			xvn[n].xv_size = sizeof *xvn;
3063 			xvn[n].xv_vnode = vp;
3064 			xvn[n].xv_id = 0;	/* XXX compat */
3065 #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field
3066 			XV_COPY(usecount);
3067 			XV_COPY(writecount);
3068 			XV_COPY(holdcnt);
3069 			XV_COPY(mount);
3070 			XV_COPY(numoutput);
3071 			XV_COPY(type);
3072 #undef XV_COPY
3073 			xvn[n].xv_flag = vp->v_vflag;
3074 
3075 			switch (vp->v_type) {
3076 			case VREG:
3077 			case VDIR:
3078 			case VLNK:
3079 				break;
3080 			case VBLK:
3081 			case VCHR:
3082 				if (vp->v_rdev == NULL) {
3083 					vrele(vp);
3084 					continue;
3085 				}
3086 				xvn[n].xv_dev = dev2udev(vp->v_rdev);
3087 				break;
3088 			case VSOCK:
3089 				xvn[n].xv_socket = vp->v_socket;
3090 				break;
3091 			case VFIFO:
3092 				xvn[n].xv_fifo = vp->v_fifoinfo;
3093 				break;
3094 			case VNON:
3095 			case VBAD:
3096 			default:
3097 				/* shouldn't happen? */
3098 				vrele(vp);
3099 				continue;
3100 			}
3101 			vrele(vp);
3102 			++n;
3103 		}
3104 		MNT_IUNLOCK(mp);
3105 		mtx_lock(&mountlist_mtx);
3106 		vfs_unbusy(mp);
3107 		if (n == len)
3108 			break;
3109 	}
3110 	mtx_unlock(&mountlist_mtx);
3111 
3112 	error = SYSCTL_OUT(req, xvn, n * sizeof *xvn);
3113 	free(xvn, M_TEMP);
3114 	return (error);
3115 }
3116 
3117 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
3118 	0, 0, sysctl_vnode, "S,xvnode", "");
3119 #endif
3120 
3121 /*
3122  * Unmount all filesystems. The list is traversed in reverse order
3123  * of mounting to avoid dependencies.
3124  */
3125 void
3126 vfs_unmountall(void)
3127 {
3128 	struct mount *mp;
3129 	struct thread *td;
3130 	int error;
3131 
3132 	KASSERT(curthread != NULL, ("vfs_unmountall: NULL curthread"));
3133 	CTR1(KTR_VFS, "%s: unmounting all filesystems", __func__);
3134 	td = curthread;
3135 
3136 	/*
3137 	 * Since this only runs when rebooting, it is not interlocked.
3138 	 */
3139 	while(!TAILQ_EMPTY(&mountlist)) {
3140 		mp = TAILQ_LAST(&mountlist, mntlist);
3141 		error = dounmount(mp, MNT_FORCE, td);
3142 		if (error) {
3143 			TAILQ_REMOVE(&mountlist, mp, mnt_list);
3144 			/*
3145 			 * XXX: Due to the way in which we mount the root
3146 			 * file system off of devfs, devfs will generate a
3147 			 * "busy" warning when we try to unmount it before
3148 			 * the root.  Don't print a warning as a result in
3149 			 * order to avoid false positive errors that may
3150 			 * cause needless upset.
3151 			 */
3152 			if (strcmp(mp->mnt_vfc->vfc_name, "devfs") != 0) {
3153 				printf("unmount of %s failed (",
3154 				    mp->mnt_stat.f_mntonname);
3155 				if (error == EBUSY)
3156 					printf("BUSY)\n");
3157 				else
3158 					printf("%d)\n", error);
3159 			}
3160 		} else {
3161 			/* The unmount has removed mp from the mountlist */
3162 		}
3163 	}
3164 }
3165 
3166 /*
3167  * perform msync on all vnodes under a mount point
3168  * the mount point must be locked.
3169  */
3170 void
3171 vfs_msync(struct mount *mp, int flags)
3172 {
3173 	struct vnode *vp, *mvp;
3174 	struct vm_object *obj;
3175 
3176 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
3177 	MNT_ILOCK(mp);
3178 	MNT_VNODE_FOREACH(vp, mp, mvp) {
3179 		VI_LOCK(vp);
3180 		if ((vp->v_iflag & VI_OBJDIRTY) &&
3181 		    (flags == MNT_WAIT || VOP_ISLOCKED(vp) == 0)) {
3182 			MNT_IUNLOCK(mp);
3183 			if (!vget(vp,
3184 			    LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
3185 			    curthread)) {
3186 				if (vp->v_vflag & VV_NOSYNC) {	/* unlinked */
3187 					vput(vp);
3188 					MNT_ILOCK(mp);
3189 					continue;
3190 				}
3191 
3192 				obj = vp->v_object;
3193 				if (obj != NULL) {
3194 					VM_OBJECT_LOCK(obj);
3195 					vm_object_page_clean(obj, 0, 0,
3196 					    flags == MNT_WAIT ?
3197 					    OBJPC_SYNC : OBJPC_NOSYNC);
3198 					VM_OBJECT_UNLOCK(obj);
3199 				}
3200 				vput(vp);
3201 			}
3202 			MNT_ILOCK(mp);
3203 		} else
3204 			VI_UNLOCK(vp);
3205 	}
3206 	MNT_IUNLOCK(mp);
3207 }
3208 
3209 /*
3210  * Mark a vnode as free, putting it up for recycling.
3211  */
3212 static void
3213 vfree(struct vnode *vp)
3214 {
3215 
3216 	ASSERT_VI_LOCKED(vp, "vfree");
3217 	mtx_lock(&vnode_free_list_mtx);
3218 	VNASSERT(vp->v_op != NULL, vp, ("vfree: vnode already reclaimed."));
3219 	VNASSERT((vp->v_iflag & VI_FREE) == 0, vp, ("vnode already free"));
3220 	VNASSERT(VSHOULDFREE(vp), vp, ("vfree: freeing when we shouldn't"));
3221 	VNASSERT((vp->v_iflag & VI_DOOMED) == 0, vp,
3222 	    ("vfree: Freeing doomed vnode"));
3223 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
3224 	if (vp->v_iflag & VI_AGE) {
3225 		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
3226 	} else {
3227 		TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
3228 	}
3229 	freevnodes++;
3230 	vp->v_iflag &= ~VI_AGE;
3231 	vp->v_iflag |= VI_FREE;
3232 	mtx_unlock(&vnode_free_list_mtx);
3233 }
3234 
3235 /*
3236  * Opposite of vfree() - mark a vnode as in use.
3237  */
3238 static void
3239 vbusy(struct vnode *vp)
3240 {
3241 	ASSERT_VI_LOCKED(vp, "vbusy");
3242 	VNASSERT((vp->v_iflag & VI_FREE) != 0, vp, ("vnode not free"));
3243 	VNASSERT(vp->v_op != NULL, vp, ("vbusy: vnode already reclaimed."));
3244 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
3245 
3246 	mtx_lock(&vnode_free_list_mtx);
3247 	TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
3248 	freevnodes--;
3249 	vp->v_iflag &= ~(VI_FREE|VI_AGE);
3250 	mtx_unlock(&vnode_free_list_mtx);
3251 }
3252 
3253 static void
3254 destroy_vpollinfo(struct vpollinfo *vi)
3255 {
3256 	knlist_destroy(&vi->vpi_selinfo.si_note);
3257 	mtx_destroy(&vi->vpi_lock);
3258 	uma_zfree(vnodepoll_zone, vi);
3259 }
3260 
3261 /*
3262  * Initalize per-vnode helper structure to hold poll-related state.
3263  */
3264 void
3265 v_addpollinfo(struct vnode *vp)
3266 {
3267 	struct vpollinfo *vi;
3268 
3269 	if (vp->v_pollinfo != NULL)
3270 		return;
3271 	vi = uma_zalloc(vnodepoll_zone, M_WAITOK);
3272 	mtx_init(&vi->vpi_lock, "vnode pollinfo", NULL, MTX_DEF);
3273 	knlist_init(&vi->vpi_selinfo.si_note, vp, vfs_knllock,
3274 	    vfs_knlunlock, vfs_knl_assert_locked, vfs_knl_assert_unlocked);
3275 	VI_LOCK(vp);
3276 	if (vp->v_pollinfo != NULL) {
3277 		VI_UNLOCK(vp);
3278 		destroy_vpollinfo(vi);
3279 		return;
3280 	}
3281 	vp->v_pollinfo = vi;
3282 	VI_UNLOCK(vp);
3283 }
3284 
3285 /*
3286  * Record a process's interest in events which might happen to
3287  * a vnode.  Because poll uses the historic select-style interface
3288  * internally, this routine serves as both the ``check for any
3289  * pending events'' and the ``record my interest in future events''
3290  * functions.  (These are done together, while the lock is held,
3291  * to avoid race conditions.)
3292  */
3293 int
3294 vn_pollrecord(struct vnode *vp, struct thread *td, int events)
3295 {
3296 
3297 	v_addpollinfo(vp);
3298 	mtx_lock(&vp->v_pollinfo->vpi_lock);
3299 	if (vp->v_pollinfo->vpi_revents & events) {
3300 		/*
3301 		 * This leaves events we are not interested
3302 		 * in available for the other process which
3303 		 * which presumably had requested them
3304 		 * (otherwise they would never have been
3305 		 * recorded).
3306 		 */
3307 		events &= vp->v_pollinfo->vpi_revents;
3308 		vp->v_pollinfo->vpi_revents &= ~events;
3309 
3310 		mtx_unlock(&vp->v_pollinfo->vpi_lock);
3311 		return (events);
3312 	}
3313 	vp->v_pollinfo->vpi_events |= events;
3314 	selrecord(td, &vp->v_pollinfo->vpi_selinfo);
3315 	mtx_unlock(&vp->v_pollinfo->vpi_lock);
3316 	return (0);
3317 }
3318 
3319 /*
3320  * Routine to create and manage a filesystem syncer vnode.
3321  */
3322 #define sync_close ((int (*)(struct  vop_close_args *))nullop)
3323 static int	sync_fsync(struct  vop_fsync_args *);
3324 static int	sync_inactive(struct  vop_inactive_args *);
3325 static int	sync_reclaim(struct  vop_reclaim_args *);
3326 
3327 static struct vop_vector sync_vnodeops = {
3328 	.vop_bypass =	VOP_EOPNOTSUPP,
3329 	.vop_close =	sync_close,		/* close */
3330 	.vop_fsync =	sync_fsync,		/* fsync */
3331 	.vop_inactive =	sync_inactive,	/* inactive */
3332 	.vop_reclaim =	sync_reclaim,	/* reclaim */
3333 	.vop_lock1 =	vop_stdlock,	/* lock */
3334 	.vop_unlock =	vop_stdunlock,	/* unlock */
3335 	.vop_islocked =	vop_stdislocked,	/* islocked */
3336 };
3337 
3338 /*
3339  * Create a new filesystem syncer vnode for the specified mount point.
3340  */
3341 int
3342 vfs_allocate_syncvnode(struct mount *mp)
3343 {
3344 	struct vnode *vp;
3345 	struct bufobj *bo;
3346 	static long start, incr, next;
3347 	int error;
3348 
3349 	/* Allocate a new vnode */
3350 	if ((error = getnewvnode("syncer", mp, &sync_vnodeops, &vp)) != 0) {
3351 		mp->mnt_syncer = NULL;
3352 		return (error);
3353 	}
3354 	vp->v_type = VNON;
3355 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3356 	vp->v_vflag |= VV_FORCEINSMQ;
3357 	error = insmntque(vp, mp);
3358 	if (error != 0)
3359 		panic("vfs_allocate_syncvnode: insmntque failed");
3360 	vp->v_vflag &= ~VV_FORCEINSMQ;
3361 	VOP_UNLOCK(vp, 0);
3362 	/*
3363 	 * Place the vnode onto the syncer worklist. We attempt to
3364 	 * scatter them about on the list so that they will go off
3365 	 * at evenly distributed times even if all the filesystems
3366 	 * are mounted at once.
3367 	 */
3368 	next += incr;
3369 	if (next == 0 || next > syncer_maxdelay) {
3370 		start /= 2;
3371 		incr /= 2;
3372 		if (start == 0) {
3373 			start = syncer_maxdelay / 2;
3374 			incr = syncer_maxdelay;
3375 		}
3376 		next = start;
3377 	}
3378 	bo = &vp->v_bufobj;
3379 	BO_LOCK(bo);
3380 	vn_syncer_add_to_worklist(bo, syncdelay > 0 ? next % syncdelay : 0);
3381 	/* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */
3382 	mtx_lock(&sync_mtx);
3383 	sync_vnode_count++;
3384 	mtx_unlock(&sync_mtx);
3385 	BO_UNLOCK(bo);
3386 	mp->mnt_syncer = vp;
3387 	return (0);
3388 }
3389 
3390 /*
3391  * Do a lazy sync of the filesystem.
3392  */
3393 static int
3394 sync_fsync(struct vop_fsync_args *ap)
3395 {
3396 	struct vnode *syncvp = ap->a_vp;
3397 	struct mount *mp = syncvp->v_mount;
3398 	int error;
3399 	struct bufobj *bo;
3400 
3401 	/*
3402 	 * We only need to do something if this is a lazy evaluation.
3403 	 */
3404 	if (ap->a_waitfor != MNT_LAZY)
3405 		return (0);
3406 
3407 	/*
3408 	 * Move ourselves to the back of the sync list.
3409 	 */
3410 	bo = &syncvp->v_bufobj;
3411 	BO_LOCK(bo);
3412 	vn_syncer_add_to_worklist(bo, syncdelay);
3413 	BO_UNLOCK(bo);
3414 
3415 	/*
3416 	 * Walk the list of vnodes pushing all that are dirty and
3417 	 * not already on the sync list.
3418 	 */
3419 	mtx_lock(&mountlist_mtx);
3420 	if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK) != 0) {
3421 		mtx_unlock(&mountlist_mtx);
3422 		return (0);
3423 	}
3424 	if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
3425 		vfs_unbusy(mp);
3426 		return (0);
3427 	}
3428 	MNT_ILOCK(mp);
3429 	mp->mnt_noasync++;
3430 	mp->mnt_kern_flag &= ~MNTK_ASYNC;
3431 	MNT_IUNLOCK(mp);
3432 	vfs_msync(mp, MNT_NOWAIT);
3433 	error = VFS_SYNC(mp, MNT_LAZY);
3434 	MNT_ILOCK(mp);
3435 	mp->mnt_noasync--;
3436 	if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0)
3437 		mp->mnt_kern_flag |= MNTK_ASYNC;
3438 	MNT_IUNLOCK(mp);
3439 	vn_finished_write(mp);
3440 	vfs_unbusy(mp);
3441 	return (error);
3442 }
3443 
3444 /*
3445  * The syncer vnode is no referenced.
3446  */
3447 static int
3448 sync_inactive(struct vop_inactive_args *ap)
3449 {
3450 
3451 	vgone(ap->a_vp);
3452 	return (0);
3453 }
3454 
3455 /*
3456  * The syncer vnode is no longer needed and is being decommissioned.
3457  *
3458  * Modifications to the worklist must be protected by sync_mtx.
3459  */
3460 static int
3461 sync_reclaim(struct vop_reclaim_args *ap)
3462 {
3463 	struct vnode *vp = ap->a_vp;
3464 	struct bufobj *bo;
3465 
3466 	bo = &vp->v_bufobj;
3467 	BO_LOCK(bo);
3468 	vp->v_mount->mnt_syncer = NULL;
3469 	if (bo->bo_flag & BO_ONWORKLST) {
3470 		mtx_lock(&sync_mtx);
3471 		LIST_REMOVE(bo, bo_synclist);
3472 		syncer_worklist_len--;
3473 		sync_vnode_count--;
3474 		mtx_unlock(&sync_mtx);
3475 		bo->bo_flag &= ~BO_ONWORKLST;
3476 	}
3477 	BO_UNLOCK(bo);
3478 
3479 	return (0);
3480 }
3481 
3482 /*
3483  * Check if vnode represents a disk device
3484  */
3485 int
3486 vn_isdisk(struct vnode *vp, int *errp)
3487 {
3488 	int error;
3489 
3490 	error = 0;
3491 	dev_lock();
3492 	if (vp->v_type != VCHR)
3493 		error = ENOTBLK;
3494 	else if (vp->v_rdev == NULL)
3495 		error = ENXIO;
3496 	else if (vp->v_rdev->si_devsw == NULL)
3497 		error = ENXIO;
3498 	else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK))
3499 		error = ENOTBLK;
3500 	dev_unlock();
3501 	if (errp != NULL)
3502 		*errp = error;
3503 	return (error == 0);
3504 }
3505 
3506 /*
3507  * Common filesystem object access control check routine.  Accepts a
3508  * vnode's type, "mode", uid and gid, requested access mode, credentials,
3509  * and optional call-by-reference privused argument allowing vaccess()
3510  * to indicate to the caller whether privilege was used to satisfy the
3511  * request (obsoleted).  Returns 0 on success, or an errno on failure.
3512  *
3513  * The ifdef'd CAPABILITIES version is here for reference, but is not
3514  * actually used.
3515  */
3516 int
3517 vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid,
3518     accmode_t accmode, struct ucred *cred, int *privused)
3519 {
3520 	accmode_t dac_granted;
3521 	accmode_t priv_granted;
3522 
3523 	/*
3524 	 * Look for a normal, non-privileged way to access the file/directory
3525 	 * as requested.  If it exists, go with that.
3526 	 */
3527 
3528 	if (privused != NULL)
3529 		*privused = 0;
3530 
3531 	dac_granted = 0;
3532 
3533 	/* Check the owner. */
3534 	if (cred->cr_uid == file_uid) {
3535 		dac_granted |= VADMIN;
3536 		if (file_mode & S_IXUSR)
3537 			dac_granted |= VEXEC;
3538 		if (file_mode & S_IRUSR)
3539 			dac_granted |= VREAD;
3540 		if (file_mode & S_IWUSR)
3541 			dac_granted |= (VWRITE | VAPPEND);
3542 
3543 		if ((accmode & dac_granted) == accmode)
3544 			return (0);
3545 
3546 		goto privcheck;
3547 	}
3548 
3549 	/* Otherwise, check the groups (first match) */
3550 	if (groupmember(file_gid, cred)) {
3551 		if (file_mode & S_IXGRP)
3552 			dac_granted |= VEXEC;
3553 		if (file_mode & S_IRGRP)
3554 			dac_granted |= VREAD;
3555 		if (file_mode & S_IWGRP)
3556 			dac_granted |= (VWRITE | VAPPEND);
3557 
3558 		if ((accmode & dac_granted) == accmode)
3559 			return (0);
3560 
3561 		goto privcheck;
3562 	}
3563 
3564 	/* Otherwise, check everyone else. */
3565 	if (file_mode & S_IXOTH)
3566 		dac_granted |= VEXEC;
3567 	if (file_mode & S_IROTH)
3568 		dac_granted |= VREAD;
3569 	if (file_mode & S_IWOTH)
3570 		dac_granted |= (VWRITE | VAPPEND);
3571 	if ((accmode & dac_granted) == accmode)
3572 		return (0);
3573 
3574 privcheck:
3575 	/*
3576 	 * Build a privilege mask to determine if the set of privileges
3577 	 * satisfies the requirements when combined with the granted mask
3578 	 * from above.  For each privilege, if the privilege is required,
3579 	 * bitwise or the request type onto the priv_granted mask.
3580 	 */
3581 	priv_granted = 0;
3582 
3583 	if (type == VDIR) {
3584 		/*
3585 		 * For directories, use PRIV_VFS_LOOKUP to satisfy VEXEC
3586 		 * requests, instead of PRIV_VFS_EXEC.
3587 		 */
3588 		if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3589 		    !priv_check_cred(cred, PRIV_VFS_LOOKUP, 0))
3590 			priv_granted |= VEXEC;
3591 	} else {
3592 		if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3593 		    !priv_check_cred(cred, PRIV_VFS_EXEC, 0))
3594 			priv_granted |= VEXEC;
3595 	}
3596 
3597 	if ((accmode & VREAD) && ((dac_granted & VREAD) == 0) &&
3598 	    !priv_check_cred(cred, PRIV_VFS_READ, 0))
3599 		priv_granted |= VREAD;
3600 
3601 	if ((accmode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
3602 	    !priv_check_cred(cred, PRIV_VFS_WRITE, 0))
3603 		priv_granted |= (VWRITE | VAPPEND);
3604 
3605 	if ((accmode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
3606 	    !priv_check_cred(cred, PRIV_VFS_ADMIN, 0))
3607 		priv_granted |= VADMIN;
3608 
3609 	if ((accmode & (priv_granted | dac_granted)) == accmode) {
3610 		/* XXX audit: privilege used */
3611 		if (privused != NULL)
3612 			*privused = 1;
3613 		return (0);
3614 	}
3615 
3616 	return ((accmode & VADMIN) ? EPERM : EACCES);
3617 }
3618 
3619 /*
3620  * Credential check based on process requesting service, and per-attribute
3621  * permissions.
3622  */
3623 int
3624 extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred,
3625     struct thread *td, accmode_t accmode)
3626 {
3627 
3628 	/*
3629 	 * Kernel-invoked always succeeds.
3630 	 */
3631 	if (cred == NOCRED)
3632 		return (0);
3633 
3634 	/*
3635 	 * Do not allow privileged processes in jail to directly manipulate
3636 	 * system attributes.
3637 	 */
3638 	switch (attrnamespace) {
3639 	case EXTATTR_NAMESPACE_SYSTEM:
3640 		/* Potentially should be: return (EPERM); */
3641 		return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM, 0));
3642 	case EXTATTR_NAMESPACE_USER:
3643 		return (VOP_ACCESS(vp, accmode, cred, td));
3644 	default:
3645 		return (EPERM);
3646 	}
3647 }
3648 
3649 #ifdef DEBUG_VFS_LOCKS
3650 /*
3651  * This only exists to supress warnings from unlocked specfs accesses.  It is
3652  * no longer ok to have an unlocked VFS.
3653  */
3654 #define	IGNORE_LOCK(vp) (panicstr != NULL || (vp) == NULL ||		\
3655 	(vp)->v_type == VCHR ||	(vp)->v_type == VBAD)
3656 
3657 int vfs_badlock_ddb = 1;	/* Drop into debugger on violation. */
3658 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0, "");
3659 
3660 int vfs_badlock_mutex = 1;	/* Check for interlock across VOPs. */
3661 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex, 0, "");
3662 
3663 int vfs_badlock_print = 1;	/* Print lock violations. */
3664 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print, 0, "");
3665 
3666 #ifdef KDB
3667 int vfs_badlock_backtrace = 1;	/* Print backtrace at lock violations. */
3668 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW, &vfs_badlock_backtrace, 0, "");
3669 #endif
3670 
3671 static void
3672 vfs_badlock(const char *msg, const char *str, struct vnode *vp)
3673 {
3674 
3675 #ifdef KDB
3676 	if (vfs_badlock_backtrace)
3677 		kdb_backtrace();
3678 #endif
3679 	if (vfs_badlock_print)
3680 		printf("%s: %p %s\n", str, (void *)vp, msg);
3681 	if (vfs_badlock_ddb)
3682 		kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
3683 }
3684 
3685 void
3686 assert_vi_locked(struct vnode *vp, const char *str)
3687 {
3688 
3689 	if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp)))
3690 		vfs_badlock("interlock is not locked but should be", str, vp);
3691 }
3692 
3693 void
3694 assert_vi_unlocked(struct vnode *vp, const char *str)
3695 {
3696 
3697 	if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp)))
3698 		vfs_badlock("interlock is locked but should not be", str, vp);
3699 }
3700 
3701 void
3702 assert_vop_locked(struct vnode *vp, const char *str)
3703 {
3704 
3705 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == 0)
3706 		vfs_badlock("is not locked but should be", str, vp);
3707 }
3708 
3709 void
3710 assert_vop_unlocked(struct vnode *vp, const char *str)
3711 {
3712 
3713 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == LK_EXCLUSIVE)
3714 		vfs_badlock("is locked but should not be", str, vp);
3715 }
3716 
3717 void
3718 assert_vop_elocked(struct vnode *vp, const char *str)
3719 {
3720 
3721 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
3722 		vfs_badlock("is not exclusive locked but should be", str, vp);
3723 }
3724 
3725 #if 0
3726 void
3727 assert_vop_elocked_other(struct vnode *vp, const char *str)
3728 {
3729 
3730 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLOTHER)
3731 		vfs_badlock("is not exclusive locked by another thread",
3732 		    str, vp);
3733 }
3734 
3735 void
3736 assert_vop_slocked(struct vnode *vp, const char *str)
3737 {
3738 
3739 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_SHARED)
3740 		vfs_badlock("is not locked shared but should be", str, vp);
3741 }
3742 #endif /* 0 */
3743 #endif /* DEBUG_VFS_LOCKS */
3744 
3745 void
3746 vop_rename_pre(void *ap)
3747 {
3748 	struct vop_rename_args *a = ap;
3749 
3750 #ifdef DEBUG_VFS_LOCKS
3751 	if (a->a_tvp)
3752 		ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME");
3753 	ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME");
3754 	ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME");
3755 	ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME");
3756 
3757 	/* Check the source (from). */
3758 	if (a->a_tdvp != a->a_fdvp && a->a_tvp != a->a_fdvp)
3759 		ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked");
3760 	if (a->a_tvp != a->a_fvp)
3761 		ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: fvp locked");
3762 
3763 	/* Check the target. */
3764 	if (a->a_tvp)
3765 		ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked");
3766 	ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked");
3767 #endif
3768 	if (a->a_tdvp != a->a_fdvp)
3769 		vhold(a->a_fdvp);
3770 	if (a->a_tvp != a->a_fvp)
3771 		vhold(a->a_fvp);
3772 	vhold(a->a_tdvp);
3773 	if (a->a_tvp)
3774 		vhold(a->a_tvp);
3775 }
3776 
3777 void
3778 vop_strategy_pre(void *ap)
3779 {
3780 #ifdef DEBUG_VFS_LOCKS
3781 	struct vop_strategy_args *a;
3782 	struct buf *bp;
3783 
3784 	a = ap;
3785 	bp = a->a_bp;
3786 
3787 	/*
3788 	 * Cluster ops lock their component buffers but not the IO container.
3789 	 */
3790 	if ((bp->b_flags & B_CLUSTER) != 0)
3791 		return;
3792 
3793 	if (!BUF_ISLOCKED(bp)) {
3794 		if (vfs_badlock_print)
3795 			printf(
3796 			    "VOP_STRATEGY: bp is not locked but should be\n");
3797 		if (vfs_badlock_ddb)
3798 			kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
3799 	}
3800 #endif
3801 }
3802 
3803 void
3804 vop_lookup_pre(void *ap)
3805 {
3806 #ifdef DEBUG_VFS_LOCKS
3807 	struct vop_lookup_args *a;
3808 	struct vnode *dvp;
3809 
3810 	a = ap;
3811 	dvp = a->a_dvp;
3812 	ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
3813 	ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
3814 #endif
3815 }
3816 
3817 void
3818 vop_lookup_post(void *ap, int rc)
3819 {
3820 #ifdef DEBUG_VFS_LOCKS
3821 	struct vop_lookup_args *a;
3822 	struct vnode *dvp;
3823 	struct vnode *vp;
3824 
3825 	a = ap;
3826 	dvp = a->a_dvp;
3827 	vp = *(a->a_vpp);
3828 
3829 	ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
3830 	ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
3831 
3832 	if (!rc)
3833 		ASSERT_VOP_LOCKED(vp, "VOP_LOOKUP (child)");
3834 #endif
3835 }
3836 
3837 void
3838 vop_lock_pre(void *ap)
3839 {
3840 #ifdef DEBUG_VFS_LOCKS
3841 	struct vop_lock1_args *a = ap;
3842 
3843 	if ((a->a_flags & LK_INTERLOCK) == 0)
3844 		ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
3845 	else
3846 		ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK");
3847 #endif
3848 }
3849 
3850 void
3851 vop_lock_post(void *ap, int rc)
3852 {
3853 #ifdef DEBUG_VFS_LOCKS
3854 	struct vop_lock1_args *a = ap;
3855 
3856 	ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
3857 	if (rc == 0)
3858 		ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK");
3859 #endif
3860 }
3861 
3862 void
3863 vop_unlock_pre(void *ap)
3864 {
3865 #ifdef DEBUG_VFS_LOCKS
3866 	struct vop_unlock_args *a = ap;
3867 
3868 	if (a->a_flags & LK_INTERLOCK)
3869 		ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK");
3870 	ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK");
3871 #endif
3872 }
3873 
3874 void
3875 vop_unlock_post(void *ap, int rc)
3876 {
3877 #ifdef DEBUG_VFS_LOCKS
3878 	struct vop_unlock_args *a = ap;
3879 
3880 	if (a->a_flags & LK_INTERLOCK)
3881 		ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK");
3882 #endif
3883 }
3884 
3885 void
3886 vop_create_post(void *ap, int rc)
3887 {
3888 	struct vop_create_args *a = ap;
3889 
3890 	if (!rc)
3891 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3892 }
3893 
3894 void
3895 vop_link_post(void *ap, int rc)
3896 {
3897 	struct vop_link_args *a = ap;
3898 
3899 	if (!rc) {
3900 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_LINK);
3901 		VFS_KNOTE_LOCKED(a->a_tdvp, NOTE_WRITE);
3902 	}
3903 }
3904 
3905 void
3906 vop_mkdir_post(void *ap, int rc)
3907 {
3908 	struct vop_mkdir_args *a = ap;
3909 
3910 	if (!rc)
3911 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
3912 }
3913 
3914 void
3915 vop_mknod_post(void *ap, int rc)
3916 {
3917 	struct vop_mknod_args *a = ap;
3918 
3919 	if (!rc)
3920 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3921 }
3922 
3923 void
3924 vop_remove_post(void *ap, int rc)
3925 {
3926 	struct vop_remove_args *a = ap;
3927 
3928 	if (!rc) {
3929 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3930 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
3931 	}
3932 }
3933 
3934 void
3935 vop_rename_post(void *ap, int rc)
3936 {
3937 	struct vop_rename_args *a = ap;
3938 
3939 	if (!rc) {
3940 		VFS_KNOTE_UNLOCKED(a->a_fdvp, NOTE_WRITE);
3941 		VFS_KNOTE_UNLOCKED(a->a_tdvp, NOTE_WRITE);
3942 		VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME);
3943 		if (a->a_tvp)
3944 			VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE);
3945 	}
3946 	if (a->a_tdvp != a->a_fdvp)
3947 		vdrop(a->a_fdvp);
3948 	if (a->a_tvp != a->a_fvp)
3949 		vdrop(a->a_fvp);
3950 	vdrop(a->a_tdvp);
3951 	if (a->a_tvp)
3952 		vdrop(a->a_tvp);
3953 }
3954 
3955 void
3956 vop_rmdir_post(void *ap, int rc)
3957 {
3958 	struct vop_rmdir_args *a = ap;
3959 
3960 	if (!rc) {
3961 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
3962 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
3963 	}
3964 }
3965 
3966 void
3967 vop_setattr_post(void *ap, int rc)
3968 {
3969 	struct vop_setattr_args *a = ap;
3970 
3971 	if (!rc)
3972 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
3973 }
3974 
3975 void
3976 vop_symlink_post(void *ap, int rc)
3977 {
3978 	struct vop_symlink_args *a = ap;
3979 
3980 	if (!rc)
3981 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3982 }
3983 
3984 static struct knlist fs_knlist;
3985 
3986 static void
3987 vfs_event_init(void *arg)
3988 {
3989 	knlist_init_mtx(&fs_knlist, NULL);
3990 }
3991 /* XXX - correct order? */
3992 SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL);
3993 
3994 void
3995 vfs_event_signal(fsid_t *fsid, u_int32_t event, intptr_t data __unused)
3996 {
3997 
3998 	KNOTE_UNLOCKED(&fs_knlist, event);
3999 }
4000 
4001 static int	filt_fsattach(struct knote *kn);
4002 static void	filt_fsdetach(struct knote *kn);
4003 static int	filt_fsevent(struct knote *kn, long hint);
4004 
4005 struct filterops fs_filtops =
4006 	{ 0, filt_fsattach, filt_fsdetach, filt_fsevent };
4007 
4008 static int
4009 filt_fsattach(struct knote *kn)
4010 {
4011 
4012 	kn->kn_flags |= EV_CLEAR;
4013 	knlist_add(&fs_knlist, kn, 0);
4014 	return (0);
4015 }
4016 
4017 static void
4018 filt_fsdetach(struct knote *kn)
4019 {
4020 
4021 	knlist_remove(&fs_knlist, kn, 0);
4022 }
4023 
4024 static int
4025 filt_fsevent(struct knote *kn, long hint)
4026 {
4027 
4028 	kn->kn_fflags |= hint;
4029 	return (kn->kn_fflags != 0);
4030 }
4031 
4032 static int
4033 sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS)
4034 {
4035 	struct vfsidctl vc;
4036 	int error;
4037 	struct mount *mp;
4038 
4039 	error = SYSCTL_IN(req, &vc, sizeof(vc));
4040 	if (error)
4041 		return (error);
4042 	if (vc.vc_vers != VFS_CTL_VERS1)
4043 		return (EINVAL);
4044 	mp = vfs_getvfs(&vc.vc_fsid);
4045 	if (mp == NULL)
4046 		return (ENOENT);
4047 	/* ensure that a specific sysctl goes to the right filesystem. */
4048 	if (strcmp(vc.vc_fstypename, "*") != 0 &&
4049 	    strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) {
4050 		vfs_rel(mp);
4051 		return (EINVAL);
4052 	}
4053 	VCTLTOREQ(&vc, req);
4054 	error = VFS_SYSCTL(mp, vc.vc_op, req);
4055 	vfs_rel(mp);
4056 	return (error);
4057 }
4058 
4059 SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLFLAG_WR, NULL, 0, sysctl_vfs_ctl, "",
4060     "Sysctl by fsid");
4061 
4062 /*
4063  * Function to initialize a va_filerev field sensibly.
4064  * XXX: Wouldn't a random number make a lot more sense ??
4065  */
4066 u_quad_t
4067 init_va_filerev(void)
4068 {
4069 	struct bintime bt;
4070 
4071 	getbinuptime(&bt);
4072 	return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL));
4073 }
4074 
4075 static int	filt_vfsread(struct knote *kn, long hint);
4076 static int	filt_vfswrite(struct knote *kn, long hint);
4077 static int	filt_vfsvnode(struct knote *kn, long hint);
4078 static void	filt_vfsdetach(struct knote *kn);
4079 static struct filterops vfsread_filtops =
4080 	{ 1, NULL, filt_vfsdetach, filt_vfsread };
4081 static struct filterops vfswrite_filtops =
4082 	{ 1, NULL, filt_vfsdetach, filt_vfswrite };
4083 static struct filterops vfsvnode_filtops =
4084 	{ 1, NULL, filt_vfsdetach, filt_vfsvnode };
4085 
4086 static void
4087 vfs_knllock(void *arg)
4088 {
4089 	struct vnode *vp = arg;
4090 
4091 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4092 }
4093 
4094 static void
4095 vfs_knlunlock(void *arg)
4096 {
4097 	struct vnode *vp = arg;
4098 
4099 	VOP_UNLOCK(vp, 0);
4100 }
4101 
4102 static void
4103 vfs_knl_assert_locked(void *arg)
4104 {
4105 #ifdef DEBUG_VFS_LOCKS
4106 	struct vnode *vp = arg;
4107 
4108 	ASSERT_VOP_LOCKED(vp, "vfs_knl_assert_locked");
4109 #endif
4110 }
4111 
4112 static void
4113 vfs_knl_assert_unlocked(void *arg)
4114 {
4115 #ifdef DEBUG_VFS_LOCKS
4116 	struct vnode *vp = arg;
4117 
4118 	ASSERT_VOP_UNLOCKED(vp, "vfs_knl_assert_unlocked");
4119 #endif
4120 }
4121 
4122 int
4123 vfs_kqfilter(struct vop_kqfilter_args *ap)
4124 {
4125 	struct vnode *vp = ap->a_vp;
4126 	struct knote *kn = ap->a_kn;
4127 	struct knlist *knl;
4128 
4129 	switch (kn->kn_filter) {
4130 	case EVFILT_READ:
4131 		kn->kn_fop = &vfsread_filtops;
4132 		break;
4133 	case EVFILT_WRITE:
4134 		kn->kn_fop = &vfswrite_filtops;
4135 		break;
4136 	case EVFILT_VNODE:
4137 		kn->kn_fop = &vfsvnode_filtops;
4138 		break;
4139 	default:
4140 		return (EINVAL);
4141 	}
4142 
4143 	kn->kn_hook = (caddr_t)vp;
4144 
4145 	v_addpollinfo(vp);
4146 	if (vp->v_pollinfo == NULL)
4147 		return (ENOMEM);
4148 	knl = &vp->v_pollinfo->vpi_selinfo.si_note;
4149 	knlist_add(knl, kn, 0);
4150 
4151 	return (0);
4152 }
4153 
4154 /*
4155  * Detach knote from vnode
4156  */
4157 static void
4158 filt_vfsdetach(struct knote *kn)
4159 {
4160 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4161 
4162 	KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo"));
4163 	knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0);
4164 }
4165 
4166 /*ARGSUSED*/
4167 static int
4168 filt_vfsread(struct knote *kn, long hint)
4169 {
4170 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4171 	struct vattr va;
4172 	int res;
4173 
4174 	/*
4175 	 * filesystem is gone, so set the EOF flag and schedule
4176 	 * the knote for deletion.
4177 	 */
4178 	if (hint == NOTE_REVOKE) {
4179 		VI_LOCK(vp);
4180 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4181 		VI_UNLOCK(vp);
4182 		return (1);
4183 	}
4184 
4185 	if (VOP_GETATTR(vp, &va, curthread->td_ucred))
4186 		return (0);
4187 
4188 	VI_LOCK(vp);
4189 	kn->kn_data = va.va_size - kn->kn_fp->f_offset;
4190 	res = (kn->kn_data != 0);
4191 	VI_UNLOCK(vp);
4192 	return (res);
4193 }
4194 
4195 /*ARGSUSED*/
4196 static int
4197 filt_vfswrite(struct knote *kn, long hint)
4198 {
4199 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4200 
4201 	VI_LOCK(vp);
4202 
4203 	/*
4204 	 * filesystem is gone, so set the EOF flag and schedule
4205 	 * the knote for deletion.
4206 	 */
4207 	if (hint == NOTE_REVOKE)
4208 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4209 
4210 	kn->kn_data = 0;
4211 	VI_UNLOCK(vp);
4212 	return (1);
4213 }
4214 
4215 static int
4216 filt_vfsvnode(struct knote *kn, long hint)
4217 {
4218 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4219 	int res;
4220 
4221 	VI_LOCK(vp);
4222 	if (kn->kn_sfflags & hint)
4223 		kn->kn_fflags |= hint;
4224 	if (hint == NOTE_REVOKE) {
4225 		kn->kn_flags |= EV_EOF;
4226 		VI_UNLOCK(vp);
4227 		return (1);
4228 	}
4229 	res = (kn->kn_fflags != 0);
4230 	VI_UNLOCK(vp);
4231 	return (res);
4232 }
4233 
4234 int
4235 vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off)
4236 {
4237 	int error;
4238 
4239 	if (dp->d_reclen > ap->a_uio->uio_resid)
4240 		return (ENAMETOOLONG);
4241 	error = uiomove(dp, dp->d_reclen, ap->a_uio);
4242 	if (error) {
4243 		if (ap->a_ncookies != NULL) {
4244 			if (ap->a_cookies != NULL)
4245 				free(ap->a_cookies, M_TEMP);
4246 			ap->a_cookies = NULL;
4247 			*ap->a_ncookies = 0;
4248 		}
4249 		return (error);
4250 	}
4251 	if (ap->a_ncookies == NULL)
4252 		return (0);
4253 
4254 	KASSERT(ap->a_cookies,
4255 	    ("NULL ap->a_cookies value with non-NULL ap->a_ncookies!"));
4256 
4257 	*ap->a_cookies = realloc(*ap->a_cookies,
4258 	    (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO);
4259 	(*ap->a_cookies)[*ap->a_ncookies] = off;
4260 	return (0);
4261 }
4262 
4263 /*
4264  * Mark for update the access time of the file if the filesystem
4265  * supports VOP_MARKATIME.  This functionality is used by execve and
4266  * mmap, so we want to avoid the I/O implied by directly setting
4267  * va_atime for the sake of efficiency.
4268  */
4269 void
4270 vfs_mark_atime(struct vnode *vp, struct ucred *cred)
4271 {
4272 
4273 	if ((vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
4274 		(void)VOP_MARKATIME(vp);
4275 }
4276 
4277 /*
4278  * The purpose of this routine is to remove granularity from accmode_t,
4279  * reducing it into standard unix access bits - VEXEC, VREAD, VWRITE,
4280  * VADMIN and VAPPEND.
4281  *
4282  * If it returns 0, the caller is supposed to continue with the usual
4283  * access checks using 'accmode' as modified by this routine.  If it
4284  * returns nonzero value, the caller is supposed to return that value
4285  * as errno.
4286  *
4287  * Note that after this routine runs, accmode may be zero.
4288  */
4289 int
4290 vfs_unixify_accmode(accmode_t *accmode)
4291 {
4292 	/*
4293 	 * There is no way to specify explicit "deny" rule using
4294 	 * file mode or POSIX.1e ACLs.
4295 	 */
4296 	if (*accmode & VEXPLICIT_DENY) {
4297 		*accmode = 0;
4298 		return (0);
4299 	}
4300 
4301 	/*
4302 	 * None of these can be translated into usual access bits.
4303 	 * Also, the common case for NFSv4 ACLs is to not contain
4304 	 * either of these bits. Caller should check for VWRITE
4305 	 * on the containing directory instead.
4306 	 */
4307 	if (*accmode & (VDELETE_CHILD | VDELETE))
4308 		return (EPERM);
4309 
4310 	if (*accmode & VADMIN_PERMS) {
4311 		*accmode &= ~VADMIN_PERMS;
4312 		*accmode |= VADMIN;
4313 	}
4314 
4315 	/*
4316 	 * There is no way to deny VREAD_ATTRIBUTES, VREAD_ACL
4317 	 * or VSYNCHRONIZE using file mode or POSIX.1e ACL.
4318 	 */
4319 	*accmode &= ~(VSTAT_PERMS | VSYNCHRONIZE);
4320 
4321 	return (0);
4322 }
4323