xref: /freebsd/sys/kern/vfs_subr.c (revision 0e1497aefd602cea581d2380d22e67dfdcac6b4e)
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     "Number of vnodes in existence");
126 
127 /*
128  * Conversion tables for conversion from vnode types to inode formats
129  * and back.
130  */
131 enum vtype iftovt_tab[16] = {
132 	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
133 	VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
134 };
135 int vttoif_tab[10] = {
136 	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
137 	S_IFSOCK, S_IFIFO, S_IFMT, S_IFMT
138 };
139 
140 /*
141  * List of vnodes that are ready for recycling.
142  */
143 static TAILQ_HEAD(freelst, vnode) vnode_free_list;
144 
145 /*
146  * Free vnode target.  Free vnodes may simply be files which have been stat'd
147  * but not read.  This is somewhat common, and a small cache of such files
148  * should be kept to avoid recreation costs.
149  */
150 static u_long wantfreevnodes;
151 SYSCTL_LONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
152 /* Number of vnodes in the free list. */
153 static u_long freevnodes;
154 SYSCTL_LONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0,
155     "Number of vnodes in the free list");
156 
157 static int vlru_allow_cache_src;
158 SYSCTL_INT(_vfs, OID_AUTO, vlru_allow_cache_src, CTLFLAG_RW,
159     &vlru_allow_cache_src, 0, "Allow vlru to reclaim source vnode");
160 
161 /*
162  * Various variables used for debugging the new implementation of
163  * reassignbuf().
164  * XXX these are probably of (very) limited utility now.
165  */
166 static int reassignbufcalls;
167 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0,
168     "Number of calls to reassignbuf");
169 
170 /*
171  * Cache for the mount type id assigned to NFS.  This is used for
172  * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c.
173  */
174 int	nfs_mount_type = -1;
175 
176 /* To keep more than one thread at a time from running vfs_getnewfsid */
177 static struct mtx mntid_mtx;
178 
179 /*
180  * Lock for any access to the following:
181  *	vnode_free_list
182  *	numvnodes
183  *	freevnodes
184  */
185 static struct mtx vnode_free_list_mtx;
186 
187 /* Publicly exported FS */
188 struct nfs_public nfs_pub;
189 
190 /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
191 static uma_zone_t vnode_zone;
192 static uma_zone_t vnodepoll_zone;
193 
194 /*
195  * The workitem queue.
196  *
197  * It is useful to delay writes of file data and filesystem metadata
198  * for tens of seconds so that quickly created and deleted files need
199  * not waste disk bandwidth being created and removed. To realize this,
200  * we append vnodes to a "workitem" queue. When running with a soft
201  * updates implementation, most pending metadata dependencies should
202  * not wait for more than a few seconds. Thus, mounted on block devices
203  * are delayed only about a half the time that file data is delayed.
204  * Similarly, directory updates are more critical, so are only delayed
205  * about a third the time that file data is delayed. Thus, there are
206  * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
207  * one each second (driven off the filesystem syncer process). The
208  * syncer_delayno variable indicates the next queue that is to be processed.
209  * Items that need to be processed soon are placed in this queue:
210  *
211  *	syncer_workitem_pending[syncer_delayno]
212  *
213  * A delay of fifteen seconds is done by placing the request fifteen
214  * entries later in the queue:
215  *
216  *	syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
217  *
218  */
219 static int syncer_delayno;
220 static long syncer_mask;
221 LIST_HEAD(synclist, bufobj);
222 static struct synclist *syncer_workitem_pending[2];
223 /*
224  * The sync_mtx protects:
225  *	bo->bo_synclist
226  *	sync_vnode_count
227  *	syncer_delayno
228  *	syncer_state
229  *	syncer_workitem_pending
230  *	syncer_worklist_len
231  *	rushjob
232  */
233 static struct mtx sync_mtx;
234 static struct cv sync_wakeup;
235 
236 #define SYNCER_MAXDELAY		32
237 static int syncer_maxdelay = SYNCER_MAXDELAY;	/* maximum delay time */
238 static int syncdelay = 30;		/* max time to delay syncing data */
239 static int filedelay = 30;		/* time to delay syncing files */
240 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0,
241     "Time to delay syncing files (in seconds)");
242 static int dirdelay = 29;		/* time to delay syncing directories */
243 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0,
244     "Time to delay syncing directories (in seconds)");
245 static int metadelay = 28;		/* time to delay syncing metadata */
246 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0,
247     "Time to delay syncing metadata (in seconds)");
248 static int rushjob;		/* number of slots to run ASAP */
249 static int stat_rush_requests;	/* number of times I/O speeded up */
250 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0,
251     "Number of times I/O speeded up (rush requests)");
252 
253 /*
254  * When shutting down the syncer, run it at four times normal speed.
255  */
256 #define SYNCER_SHUTDOWN_SPEEDUP		4
257 static int sync_vnode_count;
258 static int syncer_worklist_len;
259 static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY }
260     syncer_state;
261 
262 /*
263  * Number of vnodes we want to exist at any one time.  This is mostly used
264  * to size hash tables in vnode-related code.  It is normally not used in
265  * getnewvnode(), as wantfreevnodes is normally nonzero.)
266  *
267  * XXX desiredvnodes is historical cruft and should not exist.
268  */
269 int desiredvnodes;
270 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
271     &desiredvnodes, 0, "Maximum number of vnodes");
272 SYSCTL_INT(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
273     &wantfreevnodes, 0, "Minimum number of vnodes (legacy)");
274 static int vnlru_nowhere;
275 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW,
276     &vnlru_nowhere, 0, "Number of times the vnlru process ran without success");
277 
278 /*
279  * Macros to control when a vnode is freed and recycled.  All require
280  * the vnode interlock.
281  */
282 #define VCANRECYCLE(vp) (((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
283 #define VSHOULDFREE(vp) (!((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
284 #define VSHOULDBUSY(vp) (((vp)->v_iflag & VI_FREE) && (vp)->v_holdcnt)
285 
286 
287 /*
288  * Initialize the vnode management data structures.
289  *
290  * Reevaluate the following cap on the number of vnodes after the physical
291  * memory size exceeds 512GB.  In the limit, as the physical memory size
292  * grows, the ratio of physical pages to vnodes approaches sixteen to one.
293  */
294 #ifndef	MAXVNODES_MAX
295 #define	MAXVNODES_MAX	(512 * (1024 * 1024 * 1024 / (int)PAGE_SIZE / 16))
296 #endif
297 static void
298 vntblinit(void *dummy __unused)
299 {
300 	int physvnodes, virtvnodes;
301 
302 	/*
303 	 * Desiredvnodes is a function of the physical memory size and the
304 	 * kernel's heap size.  Generally speaking, it scales with the
305 	 * physical memory size.  The ratio of desiredvnodes to physical pages
306 	 * is one to four until desiredvnodes exceeds 98,304.  Thereafter, the
307 	 * marginal ratio of desiredvnodes to physical pages is one to
308 	 * sixteen.  However, desiredvnodes is limited by the kernel's heap
309 	 * size.  The memory required by desiredvnodes vnodes and vm objects
310 	 * may not exceed one seventh of the kernel's heap size.
311 	 */
312 	physvnodes = maxproc + cnt.v_page_count / 16 + 3 * min(98304 * 4,
313 	    cnt.v_page_count) / 16;
314 	virtvnodes = vm_kmem_size / (7 * (sizeof(struct vm_object) +
315 	    sizeof(struct vnode)));
316 	desiredvnodes = min(physvnodes, virtvnodes);
317 	if (desiredvnodes > MAXVNODES_MAX) {
318 		if (bootverbose)
319 			printf("Reducing kern.maxvnodes %d -> %d\n",
320 			    desiredvnodes, MAXVNODES_MAX);
321 		desiredvnodes = MAXVNODES_MAX;
322 	}
323 	wantfreevnodes = desiredvnodes / 4;
324 	mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF);
325 	TAILQ_INIT(&vnode_free_list);
326 	mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF);
327 	vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL,
328 	    NULL, NULL, UMA_ALIGN_PTR, 0);
329 	vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo),
330 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
331 	/*
332 	 * Initialize the filesystem syncer.
333 	 */
334 	syncer_workitem_pending[WI_MPSAFEQ] = hashinit(syncer_maxdelay, M_VNODE,
335 	    &syncer_mask);
336 	syncer_workitem_pending[WI_GIANTQ] = hashinit(syncer_maxdelay, M_VNODE,
337 	    &syncer_mask);
338 	syncer_maxdelay = syncer_mask + 1;
339 	mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF);
340 	cv_init(&sync_wakeup, "syncer");
341 }
342 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL);
343 
344 
345 /*
346  * Mark a mount point as busy. Used to synchronize access and to delay
347  * unmounting. Eventually, mountlist_mtx is not released on failure.
348  */
349 int
350 vfs_busy(struct mount *mp, int flags)
351 {
352 
353 	MPASS((flags & ~MBF_MASK) == 0);
354 	CTR3(KTR_VFS, "%s: mp %p with flags %d", __func__, mp, flags);
355 
356 	MNT_ILOCK(mp);
357 	MNT_REF(mp);
358 	/*
359 	 * If mount point is currenly being unmounted, sleep until the
360 	 * mount point fate is decided.  If thread doing the unmounting fails,
361 	 * it will clear MNTK_UNMOUNT flag before waking us up, indicating
362 	 * that this mount point has survived the unmount attempt and vfs_busy
363 	 * should retry.  Otherwise the unmounter thread will set MNTK_REFEXPIRE
364 	 * flag in addition to MNTK_UNMOUNT, indicating that mount point is
365 	 * about to be really destroyed.  vfs_busy needs to release its
366 	 * reference on the mount point in this case and return with ENOENT,
367 	 * telling the caller that mount mount it tried to busy is no longer
368 	 * valid.
369 	 */
370 	while (mp->mnt_kern_flag & MNTK_UNMOUNT) {
371 		if (flags & MBF_NOWAIT || mp->mnt_kern_flag & MNTK_REFEXPIRE) {
372 			MNT_REL(mp);
373 			MNT_IUNLOCK(mp);
374 			CTR1(KTR_VFS, "%s: failed busying before sleeping",
375 			    __func__);
376 			return (ENOENT);
377 		}
378 		if (flags & MBF_MNTLSTLOCK)
379 			mtx_unlock(&mountlist_mtx);
380 		mp->mnt_kern_flag |= MNTK_MWAIT;
381 		msleep(mp, MNT_MTX(mp), PVFS, "vfs_busy", 0);
382 		if (flags & MBF_MNTLSTLOCK)
383 			mtx_lock(&mountlist_mtx);
384 	}
385 	if (flags & MBF_MNTLSTLOCK)
386 		mtx_unlock(&mountlist_mtx);
387 	mp->mnt_lockref++;
388 	MNT_IUNLOCK(mp);
389 	return (0);
390 }
391 
392 /*
393  * Free a busy filesystem.
394  */
395 void
396 vfs_unbusy(struct mount *mp)
397 {
398 
399 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
400 	MNT_ILOCK(mp);
401 	MNT_REL(mp);
402 	KASSERT(mp->mnt_lockref > 0, ("negative mnt_lockref"));
403 	mp->mnt_lockref--;
404 	if (mp->mnt_lockref == 0 && (mp->mnt_kern_flag & MNTK_DRAINING) != 0) {
405 		MPASS(mp->mnt_kern_flag & MNTK_UNMOUNT);
406 		CTR1(KTR_VFS, "%s: waking up waiters", __func__);
407 		mp->mnt_kern_flag &= ~MNTK_DRAINING;
408 		wakeup(&mp->mnt_lockref);
409 	}
410 	MNT_IUNLOCK(mp);
411 }
412 
413 /*
414  * Lookup a mount point by filesystem identifier.
415  */
416 struct mount *
417 vfs_getvfs(fsid_t *fsid)
418 {
419 	struct mount *mp;
420 
421 	CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
422 	mtx_lock(&mountlist_mtx);
423 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
424 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
425 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
426 			vfs_ref(mp);
427 			mtx_unlock(&mountlist_mtx);
428 			return (mp);
429 		}
430 	}
431 	mtx_unlock(&mountlist_mtx);
432 	CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
433 	return ((struct mount *) 0);
434 }
435 
436 /*
437  * Lookup a mount point by filesystem identifier, busying it before
438  * returning.
439  */
440 struct mount *
441 vfs_busyfs(fsid_t *fsid)
442 {
443 	struct mount *mp;
444 	int error;
445 
446 	CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
447 	mtx_lock(&mountlist_mtx);
448 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
449 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
450 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
451 			error = vfs_busy(mp, MBF_MNTLSTLOCK);
452 			if (error) {
453 				mtx_unlock(&mountlist_mtx);
454 				return (NULL);
455 			}
456 			return (mp);
457 		}
458 	}
459 	CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
460 	mtx_unlock(&mountlist_mtx);
461 	return ((struct mount *) 0);
462 }
463 
464 /*
465  * Check if a user can access privileged mount options.
466  */
467 int
468 vfs_suser(struct mount *mp, struct thread *td)
469 {
470 	int error;
471 
472 	/*
473 	 * If the thread is jailed, but this is not a jail-friendly file
474 	 * system, deny immediately.
475 	 */
476 	if (!(mp->mnt_vfc->vfc_flags & VFCF_JAIL) && jailed(td->td_ucred))
477 		return (EPERM);
478 
479 	/*
480 	 * If the file system was mounted outside the jail of the calling
481 	 * thread, deny immediately.
482 	 */
483 	if (prison_check(td->td_ucred, mp->mnt_cred) != 0)
484 		return (EPERM);
485 
486 	/*
487 	 * If file system supports delegated administration, we don't check
488 	 * for the PRIV_VFS_MOUNT_OWNER privilege - it will be better verified
489 	 * by the file system itself.
490 	 * If this is not the user that did original mount, we check for
491 	 * the PRIV_VFS_MOUNT_OWNER privilege.
492 	 */
493 	if (!(mp->mnt_vfc->vfc_flags & VFCF_DELEGADMIN) &&
494 	    mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) {
495 		if ((error = priv_check(td, PRIV_VFS_MOUNT_OWNER)) != 0)
496 			return (error);
497 	}
498 	return (0);
499 }
500 
501 /*
502  * Get a new unique fsid.  Try to make its val[0] unique, since this value
503  * will be used to create fake device numbers for stat().  Also try (but
504  * not so hard) make its val[0] unique mod 2^16, since some emulators only
505  * support 16-bit device numbers.  We end up with unique val[0]'s for the
506  * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
507  *
508  * Keep in mind that several mounts may be running in parallel.  Starting
509  * the search one past where the previous search terminated is both a
510  * micro-optimization and a defense against returning the same fsid to
511  * different mounts.
512  */
513 void
514 vfs_getnewfsid(struct mount *mp)
515 {
516 	static uint16_t mntid_base;
517 	struct mount *nmp;
518 	fsid_t tfsid;
519 	int mtype;
520 
521 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
522 	mtx_lock(&mntid_mtx);
523 	mtype = mp->mnt_vfc->vfc_typenum;
524 	tfsid.val[1] = mtype;
525 	mtype = (mtype & 0xFF) << 24;
526 	for (;;) {
527 		tfsid.val[0] = makedev(255,
528 		    mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
529 		mntid_base++;
530 		if ((nmp = vfs_getvfs(&tfsid)) == NULL)
531 			break;
532 		vfs_rel(nmp);
533 	}
534 	mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
535 	mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
536 	mtx_unlock(&mntid_mtx);
537 }
538 
539 /*
540  * Knob to control the precision of file timestamps:
541  *
542  *   0 = seconds only; nanoseconds zeroed.
543  *   1 = seconds and nanoseconds, accurate within 1/HZ.
544  *   2 = seconds and nanoseconds, truncated to microseconds.
545  * >=3 = seconds and nanoseconds, maximum precision.
546  */
547 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
548 
549 static int timestamp_precision = TSP_SEC;
550 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
551     &timestamp_precision, 0, "File timestamp precision (0: seconds, "
552     "1: sec + ns accurate to 1/HZ, 2: sec + ns truncated to ms, "
553     "3+: sec + ns (max. precision))");
554 
555 /*
556  * Get a current timestamp.
557  */
558 void
559 vfs_timestamp(struct timespec *tsp)
560 {
561 	struct timeval tv;
562 
563 	switch (timestamp_precision) {
564 	case TSP_SEC:
565 		tsp->tv_sec = time_second;
566 		tsp->tv_nsec = 0;
567 		break;
568 	case TSP_HZ:
569 		getnanotime(tsp);
570 		break;
571 	case TSP_USEC:
572 		microtime(&tv);
573 		TIMEVAL_TO_TIMESPEC(&tv, tsp);
574 		break;
575 	case TSP_NSEC:
576 	default:
577 		nanotime(tsp);
578 		break;
579 	}
580 }
581 
582 /*
583  * Set vnode attributes to VNOVAL
584  */
585 void
586 vattr_null(struct vattr *vap)
587 {
588 
589 	vap->va_type = VNON;
590 	vap->va_size = VNOVAL;
591 	vap->va_bytes = VNOVAL;
592 	vap->va_mode = VNOVAL;
593 	vap->va_nlink = VNOVAL;
594 	vap->va_uid = VNOVAL;
595 	vap->va_gid = VNOVAL;
596 	vap->va_fsid = VNOVAL;
597 	vap->va_fileid = VNOVAL;
598 	vap->va_blocksize = VNOVAL;
599 	vap->va_rdev = VNOVAL;
600 	vap->va_atime.tv_sec = VNOVAL;
601 	vap->va_atime.tv_nsec = VNOVAL;
602 	vap->va_mtime.tv_sec = VNOVAL;
603 	vap->va_mtime.tv_nsec = VNOVAL;
604 	vap->va_ctime.tv_sec = VNOVAL;
605 	vap->va_ctime.tv_nsec = VNOVAL;
606 	vap->va_birthtime.tv_sec = VNOVAL;
607 	vap->va_birthtime.tv_nsec = VNOVAL;
608 	vap->va_flags = VNOVAL;
609 	vap->va_gen = VNOVAL;
610 	vap->va_vaflags = 0;
611 }
612 
613 /*
614  * This routine is called when we have too many vnodes.  It attempts
615  * to free <count> vnodes and will potentially free vnodes that still
616  * have VM backing store (VM backing store is typically the cause
617  * of a vnode blowout so we want to do this).  Therefore, this operation
618  * is not considered cheap.
619  *
620  * A number of conditions may prevent a vnode from being reclaimed.
621  * the buffer cache may have references on the vnode, a directory
622  * vnode may still have references due to the namei cache representing
623  * underlying files, or the vnode may be in active use.   It is not
624  * desireable to reuse such vnodes.  These conditions may cause the
625  * number of vnodes to reach some minimum value regardless of what
626  * you set kern.maxvnodes to.  Do not set kern.maxvnodes too low.
627  */
628 static int
629 vlrureclaim(struct mount *mp)
630 {
631 	struct vnode *vp;
632 	int done;
633 	int trigger;
634 	int usevnodes;
635 	int count;
636 
637 	/*
638 	 * Calculate the trigger point, don't allow user
639 	 * screwups to blow us up.   This prevents us from
640 	 * recycling vnodes with lots of resident pages.  We
641 	 * aren't trying to free memory, we are trying to
642 	 * free vnodes.
643 	 */
644 	usevnodes = desiredvnodes;
645 	if (usevnodes <= 0)
646 		usevnodes = 1;
647 	trigger = cnt.v_page_count * 2 / usevnodes;
648 	done = 0;
649 	vn_start_write(NULL, &mp, V_WAIT);
650 	MNT_ILOCK(mp);
651 	count = mp->mnt_nvnodelistsize / 10 + 1;
652 	while (count != 0) {
653 		vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
654 		while (vp != NULL && vp->v_type == VMARKER)
655 			vp = TAILQ_NEXT(vp, v_nmntvnodes);
656 		if (vp == NULL)
657 			break;
658 		TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
659 		TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
660 		--count;
661 		if (!VI_TRYLOCK(vp))
662 			goto next_iter;
663 		/*
664 		 * If it's been deconstructed already, it's still
665 		 * referenced, or it exceeds the trigger, skip it.
666 		 */
667 		if (vp->v_usecount ||
668 		    (!vlru_allow_cache_src &&
669 			!LIST_EMPTY(&(vp)->v_cache_src)) ||
670 		    (vp->v_iflag & VI_DOOMED) != 0 || (vp->v_object != NULL &&
671 		    vp->v_object->resident_page_count > trigger)) {
672 			VI_UNLOCK(vp);
673 			goto next_iter;
674 		}
675 		MNT_IUNLOCK(mp);
676 		vholdl(vp);
677 		if (VOP_LOCK(vp, LK_INTERLOCK|LK_EXCLUSIVE|LK_NOWAIT)) {
678 			vdrop(vp);
679 			goto next_iter_mntunlocked;
680 		}
681 		VI_LOCK(vp);
682 		/*
683 		 * v_usecount may have been bumped after VOP_LOCK() dropped
684 		 * the vnode interlock and before it was locked again.
685 		 *
686 		 * It is not necessary to recheck VI_DOOMED because it can
687 		 * only be set by another thread that holds both the vnode
688 		 * lock and vnode interlock.  If another thread has the
689 		 * vnode lock before we get to VOP_LOCK() and obtains the
690 		 * vnode interlock after VOP_LOCK() drops the vnode
691 		 * interlock, the other thread will be unable to drop the
692 		 * vnode lock before our VOP_LOCK() call fails.
693 		 */
694 		if (vp->v_usecount ||
695 		    (!vlru_allow_cache_src &&
696 			!LIST_EMPTY(&(vp)->v_cache_src)) ||
697 		    (vp->v_object != NULL &&
698 		    vp->v_object->resident_page_count > trigger)) {
699 			VOP_UNLOCK(vp, LK_INTERLOCK);
700 			goto next_iter_mntunlocked;
701 		}
702 		KASSERT((vp->v_iflag & VI_DOOMED) == 0,
703 		    ("VI_DOOMED unexpectedly detected in vlrureclaim()"));
704 		vgonel(vp);
705 		VOP_UNLOCK(vp, 0);
706 		vdropl(vp);
707 		done++;
708 next_iter_mntunlocked:
709 		if ((count % 256) != 0)
710 			goto relock_mnt;
711 		goto yield;
712 next_iter:
713 		if ((count % 256) != 0)
714 			continue;
715 		MNT_IUNLOCK(mp);
716 yield:
717 		uio_yield();
718 relock_mnt:
719 		MNT_ILOCK(mp);
720 	}
721 	MNT_IUNLOCK(mp);
722 	vn_finished_write(mp);
723 	return done;
724 }
725 
726 /*
727  * Attempt to keep the free list at wantfreevnodes length.
728  */
729 static void
730 vnlru_free(int count)
731 {
732 	struct vnode *vp;
733 	int vfslocked;
734 
735 	mtx_assert(&vnode_free_list_mtx, MA_OWNED);
736 	for (; count > 0; count--) {
737 		vp = TAILQ_FIRST(&vnode_free_list);
738 		/*
739 		 * The list can be modified while the free_list_mtx
740 		 * has been dropped and vp could be NULL here.
741 		 */
742 		if (!vp)
743 			break;
744 		VNASSERT(vp->v_op != NULL, vp,
745 		    ("vnlru_free: vnode already reclaimed."));
746 		TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
747 		/*
748 		 * Don't recycle if we can't get the interlock.
749 		 */
750 		if (!VI_TRYLOCK(vp)) {
751 			TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
752 			continue;
753 		}
754 		VNASSERT(VCANRECYCLE(vp), vp,
755 		    ("vp inconsistent on freelist"));
756 		freevnodes--;
757 		vp->v_iflag &= ~VI_FREE;
758 		vholdl(vp);
759 		mtx_unlock(&vnode_free_list_mtx);
760 		VI_UNLOCK(vp);
761 		vfslocked = VFS_LOCK_GIANT(vp->v_mount);
762 		vtryrecycle(vp);
763 		VFS_UNLOCK_GIANT(vfslocked);
764 		/*
765 		 * If the recycled succeeded this vdrop will actually free
766 		 * the vnode.  If not it will simply place it back on
767 		 * the free list.
768 		 */
769 		vdrop(vp);
770 		mtx_lock(&vnode_free_list_mtx);
771 	}
772 }
773 /*
774  * Attempt to recycle vnodes in a context that is always safe to block.
775  * Calling vlrurecycle() from the bowels of filesystem code has some
776  * interesting deadlock problems.
777  */
778 static struct proc *vnlruproc;
779 static int vnlruproc_sig;
780 
781 static void
782 vnlru_proc(void)
783 {
784 	struct mount *mp, *nmp;
785 	int done, vfslocked;
786 	struct proc *p = vnlruproc;
787 
788 	EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
789 	    SHUTDOWN_PRI_FIRST);
790 
791 	for (;;) {
792 		kproc_suspend_check(p);
793 		mtx_lock(&vnode_free_list_mtx);
794 		if (freevnodes > wantfreevnodes)
795 			vnlru_free(freevnodes - wantfreevnodes);
796 		if (numvnodes <= desiredvnodes * 9 / 10) {
797 			vnlruproc_sig = 0;
798 			wakeup(&vnlruproc_sig);
799 			msleep(vnlruproc, &vnode_free_list_mtx,
800 			    PVFS|PDROP, "vlruwt", hz);
801 			continue;
802 		}
803 		mtx_unlock(&vnode_free_list_mtx);
804 		done = 0;
805 		mtx_lock(&mountlist_mtx);
806 		for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
807 			if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) {
808 				nmp = TAILQ_NEXT(mp, mnt_list);
809 				continue;
810 			}
811 			vfslocked = VFS_LOCK_GIANT(mp);
812 			done += vlrureclaim(mp);
813 			VFS_UNLOCK_GIANT(vfslocked);
814 			mtx_lock(&mountlist_mtx);
815 			nmp = TAILQ_NEXT(mp, mnt_list);
816 			vfs_unbusy(mp);
817 		}
818 		mtx_unlock(&mountlist_mtx);
819 		if (done == 0) {
820 #if 0
821 			/* These messages are temporary debugging aids */
822 			if (vnlru_nowhere < 5)
823 				printf("vnlru process getting nowhere..\n");
824 			else if (vnlru_nowhere == 5)
825 				printf("vnlru process messages stopped.\n");
826 #endif
827 			vnlru_nowhere++;
828 			tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
829 		} else
830 			uio_yield();
831 	}
832 }
833 
834 static struct kproc_desc vnlru_kp = {
835 	"vnlru",
836 	vnlru_proc,
837 	&vnlruproc
838 };
839 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start,
840     &vnlru_kp);
841 
842 /*
843  * Routines having to do with the management of the vnode table.
844  */
845 
846 void
847 vdestroy(struct vnode *vp)
848 {
849 	struct bufobj *bo;
850 
851 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
852 	mtx_lock(&vnode_free_list_mtx);
853 	numvnodes--;
854 	mtx_unlock(&vnode_free_list_mtx);
855 	bo = &vp->v_bufobj;
856 	VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
857 	    ("cleaned vnode still on the free list."));
858 	VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't"));
859 	VNASSERT(vp->v_holdcnt == 0, vp, ("Non-zero hold count"));
860 	VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count"));
861 	VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count"));
862 	VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's"));
863 	VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0"));
864 	VNASSERT(bo->bo_clean.bv_root == NULL, vp, ("cleanblkroot not NULL"));
865 	VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0"));
866 	VNASSERT(bo->bo_dirty.bv_root == NULL, vp, ("dirtyblkroot not NULL"));
867 	VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst"));
868 	VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src"));
869 	VNASSERT(vp->v_cache_dd == NULL, vp, ("vp has namecache for .."));
870 	VI_UNLOCK(vp);
871 #ifdef MAC
872 	mac_vnode_destroy(vp);
873 #endif
874 	if (vp->v_pollinfo != NULL)
875 		destroy_vpollinfo(vp->v_pollinfo);
876 #ifdef INVARIANTS
877 	/* XXX Elsewhere we can detect an already freed vnode via NULL v_op. */
878 	vp->v_op = NULL;
879 #endif
880 	lockdestroy(vp->v_vnlock);
881 	mtx_destroy(&vp->v_interlock);
882 	mtx_destroy(BO_MTX(bo));
883 	uma_zfree(vnode_zone, vp);
884 }
885 
886 /*
887  * Try to recycle a freed vnode.  We abort if anyone picks up a reference
888  * before we actually vgone().  This function must be called with the vnode
889  * held to prevent the vnode from being returned to the free list midway
890  * through vgone().
891  */
892 static int
893 vtryrecycle(struct vnode *vp)
894 {
895 	struct mount *vnmp;
896 
897 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
898 	VNASSERT(vp->v_holdcnt, vp,
899 	    ("vtryrecycle: Recycling vp %p without a reference.", vp));
900 	/*
901 	 * This vnode may found and locked via some other list, if so we
902 	 * can't recycle it yet.
903 	 */
904 	if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
905 		CTR2(KTR_VFS,
906 		    "%s: impossible to recycle, vp %p lock is already held",
907 		    __func__, vp);
908 		return (EWOULDBLOCK);
909 	}
910 	/*
911 	 * Don't recycle if its filesystem is being suspended.
912 	 */
913 	if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) {
914 		VOP_UNLOCK(vp, 0);
915 		CTR2(KTR_VFS,
916 		    "%s: impossible to recycle, cannot start the write for %p",
917 		    __func__, vp);
918 		return (EBUSY);
919 	}
920 	/*
921 	 * If we got this far, we need to acquire the interlock and see if
922 	 * anyone picked up this vnode from another list.  If not, we will
923 	 * mark it with DOOMED via vgonel() so that anyone who does find it
924 	 * will skip over it.
925 	 */
926 	VI_LOCK(vp);
927 	if (vp->v_usecount) {
928 		VOP_UNLOCK(vp, LK_INTERLOCK);
929 		vn_finished_write(vnmp);
930 		CTR2(KTR_VFS,
931 		    "%s: impossible to recycle, %p is already referenced",
932 		    __func__, vp);
933 		return (EBUSY);
934 	}
935 	if ((vp->v_iflag & VI_DOOMED) == 0)
936 		vgonel(vp);
937 	VOP_UNLOCK(vp, LK_INTERLOCK);
938 	vn_finished_write(vnmp);
939 	return (0);
940 }
941 
942 /*
943  * Return the next vnode from the free list.
944  */
945 int
946 getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops,
947     struct vnode **vpp)
948 {
949 	struct vnode *vp = NULL;
950 	struct bufobj *bo;
951 
952 	CTR3(KTR_VFS, "%s: mp %p with tag %s", __func__, mp, tag);
953 	mtx_lock(&vnode_free_list_mtx);
954 	/*
955 	 * Lend our context to reclaim vnodes if they've exceeded the max.
956 	 */
957 	if (freevnodes > wantfreevnodes)
958 		vnlru_free(1);
959 	/*
960 	 * Wait for available vnodes.
961 	 */
962 	if (numvnodes > desiredvnodes) {
963 		if (mp != NULL && (mp->mnt_kern_flag & MNTK_SUSPEND)) {
964 			/*
965 			 * File system is beeing suspended, we cannot risk a
966 			 * deadlock here, so allocate new vnode anyway.
967 			 */
968 			if (freevnodes > wantfreevnodes)
969 				vnlru_free(freevnodes - wantfreevnodes);
970 			goto alloc;
971 		}
972 		if (vnlruproc_sig == 0) {
973 			vnlruproc_sig = 1;	/* avoid unnecessary wakeups */
974 			wakeup(vnlruproc);
975 		}
976 		msleep(&vnlruproc_sig, &vnode_free_list_mtx, PVFS,
977 		    "vlruwk", hz);
978 #if 0	/* XXX Not all VFS_VGET/ffs_vget callers check returns. */
979 		if (numvnodes > desiredvnodes) {
980 			mtx_unlock(&vnode_free_list_mtx);
981 			return (ENFILE);
982 		}
983 #endif
984 	}
985 alloc:
986 	numvnodes++;
987 	mtx_unlock(&vnode_free_list_mtx);
988 	vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO);
989 	/*
990 	 * Setup locks.
991 	 */
992 	vp->v_vnlock = &vp->v_lock;
993 	mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
994 	/*
995 	 * By default, don't allow shared locks unless filesystems
996 	 * opt-in.
997 	 */
998 	lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOSHARE);
999 	/*
1000 	 * Initialize bufobj.
1001 	 */
1002 	bo = &vp->v_bufobj;
1003 	bo->__bo_vnode = vp;
1004 	mtx_init(BO_MTX(bo), "bufobj interlock", NULL, MTX_DEF);
1005 	bo->bo_ops = &buf_ops_bio;
1006 	bo->bo_private = vp;
1007 	TAILQ_INIT(&bo->bo_clean.bv_hd);
1008 	TAILQ_INIT(&bo->bo_dirty.bv_hd);
1009 	/*
1010 	 * Initialize namecache.
1011 	 */
1012 	LIST_INIT(&vp->v_cache_src);
1013 	TAILQ_INIT(&vp->v_cache_dst);
1014 	/*
1015 	 * Finalize various vnode identity bits.
1016 	 */
1017 	vp->v_type = VNON;
1018 	vp->v_tag = tag;
1019 	vp->v_op = vops;
1020 	v_incr_usecount(vp);
1021 	vp->v_data = 0;
1022 #ifdef MAC
1023 	mac_vnode_init(vp);
1024 	if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0)
1025 		mac_vnode_associate_singlelabel(mp, vp);
1026 	else if (mp == NULL && vops != &dead_vnodeops)
1027 		printf("NULL mp in getnewvnode()\n");
1028 #endif
1029 	if (mp != NULL) {
1030 		bo->bo_bsize = mp->mnt_stat.f_iosize;
1031 		if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0)
1032 			vp->v_vflag |= VV_NOKNOTE;
1033 	}
1034 
1035 	*vpp = vp;
1036 	return (0);
1037 }
1038 
1039 /*
1040  * Delete from old mount point vnode list, if on one.
1041  */
1042 static void
1043 delmntque(struct vnode *vp)
1044 {
1045 	struct mount *mp;
1046 
1047 	mp = vp->v_mount;
1048 	if (mp == NULL)
1049 		return;
1050 	MNT_ILOCK(mp);
1051 	vp->v_mount = NULL;
1052 	VNASSERT(mp->mnt_nvnodelistsize > 0, vp,
1053 		("bad mount point vnode list size"));
1054 	TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1055 	mp->mnt_nvnodelistsize--;
1056 	MNT_REL(mp);
1057 	MNT_IUNLOCK(mp);
1058 }
1059 
1060 static void
1061 insmntque_stddtr(struct vnode *vp, void *dtr_arg)
1062 {
1063 
1064 	vp->v_data = NULL;
1065 	vp->v_op = &dead_vnodeops;
1066 	/* XXX non mp-safe fs may still call insmntque with vnode
1067 	   unlocked */
1068 	if (!VOP_ISLOCKED(vp))
1069 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1070 	vgone(vp);
1071 	vput(vp);
1072 }
1073 
1074 /*
1075  * Insert into list of vnodes for the new mount point, if available.
1076  */
1077 int
1078 insmntque1(struct vnode *vp, struct mount *mp,
1079 	void (*dtr)(struct vnode *, void *), void *dtr_arg)
1080 {
1081 	int locked;
1082 
1083 	KASSERT(vp->v_mount == NULL,
1084 		("insmntque: vnode already on per mount vnode list"));
1085 	VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)"));
1086 #ifdef DEBUG_VFS_LOCKS
1087 	if (!VFS_NEEDSGIANT(mp))
1088 		ASSERT_VOP_ELOCKED(vp,
1089 		    "insmntque: mp-safe fs and non-locked vp");
1090 #endif
1091 	MNT_ILOCK(mp);
1092 	if ((mp->mnt_kern_flag & MNTK_NOINSMNTQ) != 0 &&
1093 	    ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0 ||
1094 	     mp->mnt_nvnodelistsize == 0)) {
1095 		locked = VOP_ISLOCKED(vp);
1096 		if (!locked || (locked == LK_EXCLUSIVE &&
1097 		     (vp->v_vflag & VV_FORCEINSMQ) == 0)) {
1098 			MNT_IUNLOCK(mp);
1099 			if (dtr != NULL)
1100 				dtr(vp, dtr_arg);
1101 			return (EBUSY);
1102 		}
1103 	}
1104 	vp->v_mount = mp;
1105 	MNT_REF(mp);
1106 	TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1107 	VNASSERT(mp->mnt_nvnodelistsize >= 0, vp,
1108 		("neg mount point vnode list size"));
1109 	mp->mnt_nvnodelistsize++;
1110 	MNT_IUNLOCK(mp);
1111 	return (0);
1112 }
1113 
1114 int
1115 insmntque(struct vnode *vp, struct mount *mp)
1116 {
1117 
1118 	return (insmntque1(vp, mp, insmntque_stddtr, NULL));
1119 }
1120 
1121 /*
1122  * Flush out and invalidate all buffers associated with a bufobj
1123  * Called with the underlying object locked.
1124  */
1125 int
1126 bufobj_invalbuf(struct bufobj *bo, int flags, int slpflag, int slptimeo)
1127 {
1128 	int error;
1129 
1130 	BO_LOCK(bo);
1131 	if (flags & V_SAVE) {
1132 		error = bufobj_wwait(bo, slpflag, slptimeo);
1133 		if (error) {
1134 			BO_UNLOCK(bo);
1135 			return (error);
1136 		}
1137 		if (bo->bo_dirty.bv_cnt > 0) {
1138 			BO_UNLOCK(bo);
1139 			if ((error = BO_SYNC(bo, MNT_WAIT)) != 0)
1140 				return (error);
1141 			/*
1142 			 * XXX We could save a lock/unlock if this was only
1143 			 * enabled under INVARIANTS
1144 			 */
1145 			BO_LOCK(bo);
1146 			if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)
1147 				panic("vinvalbuf: dirty bufs");
1148 		}
1149 	}
1150 	/*
1151 	 * If you alter this loop please notice that interlock is dropped and
1152 	 * reacquired in flushbuflist.  Special care is needed to ensure that
1153 	 * no race conditions occur from this.
1154 	 */
1155 	do {
1156 		error = flushbuflist(&bo->bo_clean,
1157 		    flags, bo, slpflag, slptimeo);
1158 		if (error == 0)
1159 			error = flushbuflist(&bo->bo_dirty,
1160 			    flags, bo, slpflag, slptimeo);
1161 		if (error != 0 && error != EAGAIN) {
1162 			BO_UNLOCK(bo);
1163 			return (error);
1164 		}
1165 	} while (error != 0);
1166 
1167 	/*
1168 	 * Wait for I/O to complete.  XXX needs cleaning up.  The vnode can
1169 	 * have write I/O in-progress but if there is a VM object then the
1170 	 * VM object can also have read-I/O in-progress.
1171 	 */
1172 	do {
1173 		bufobj_wwait(bo, 0, 0);
1174 		BO_UNLOCK(bo);
1175 		if (bo->bo_object != NULL) {
1176 			VM_OBJECT_LOCK(bo->bo_object);
1177 			vm_object_pip_wait(bo->bo_object, "bovlbx");
1178 			VM_OBJECT_UNLOCK(bo->bo_object);
1179 		}
1180 		BO_LOCK(bo);
1181 	} while (bo->bo_numoutput > 0);
1182 	BO_UNLOCK(bo);
1183 
1184 	/*
1185 	 * Destroy the copy in the VM cache, too.
1186 	 */
1187 	if (bo->bo_object != NULL && (flags & (V_ALT | V_NORMAL)) == 0) {
1188 		VM_OBJECT_LOCK(bo->bo_object);
1189 		vm_object_page_remove(bo->bo_object, 0, 0,
1190 			(flags & V_SAVE) ? TRUE : FALSE);
1191 		VM_OBJECT_UNLOCK(bo->bo_object);
1192 	}
1193 
1194 #ifdef INVARIANTS
1195 	BO_LOCK(bo);
1196 	if ((flags & (V_ALT | V_NORMAL)) == 0 &&
1197 	    (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0))
1198 		panic("vinvalbuf: flush failed");
1199 	BO_UNLOCK(bo);
1200 #endif
1201 	return (0);
1202 }
1203 
1204 /*
1205  * Flush out and invalidate all buffers associated with a vnode.
1206  * Called with the underlying object locked.
1207  */
1208 int
1209 vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo)
1210 {
1211 
1212 	CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
1213 	ASSERT_VOP_LOCKED(vp, "vinvalbuf");
1214 	return (bufobj_invalbuf(&vp->v_bufobj, flags, slpflag, slptimeo));
1215 }
1216 
1217 /*
1218  * Flush out buffers on the specified list.
1219  *
1220  */
1221 static int
1222 flushbuflist( struct bufv *bufv, int flags, struct bufobj *bo, int slpflag,
1223     int slptimeo)
1224 {
1225 	struct buf *bp, *nbp;
1226 	int retval, error;
1227 	daddr_t lblkno;
1228 	b_xflags_t xflags;
1229 
1230 	ASSERT_BO_LOCKED(bo);
1231 
1232 	retval = 0;
1233 	TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) {
1234 		if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) ||
1235 		    ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) {
1236 			continue;
1237 		}
1238 		lblkno = 0;
1239 		xflags = 0;
1240 		if (nbp != NULL) {
1241 			lblkno = nbp->b_lblkno;
1242 			xflags = nbp->b_xflags &
1243 				(BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN);
1244 		}
1245 		retval = EAGAIN;
1246 		error = BUF_TIMELOCK(bp,
1247 		    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_MTX(bo),
1248 		    "flushbuf", slpflag, slptimeo);
1249 		if (error) {
1250 			BO_LOCK(bo);
1251 			return (error != ENOLCK ? error : EAGAIN);
1252 		}
1253 		KASSERT(bp->b_bufobj == bo,
1254 		    ("bp %p wrong b_bufobj %p should be %p",
1255 		    bp, bp->b_bufobj, bo));
1256 		if (bp->b_bufobj != bo) {	/* XXX: necessary ? */
1257 			BUF_UNLOCK(bp);
1258 			BO_LOCK(bo);
1259 			return (EAGAIN);
1260 		}
1261 		/*
1262 		 * XXX Since there are no node locks for NFS, I
1263 		 * believe there is a slight chance that a delayed
1264 		 * write will occur while sleeping just above, so
1265 		 * check for it.
1266 		 */
1267 		if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
1268 		    (flags & V_SAVE)) {
1269 			BO_LOCK(bo);
1270 			bremfree(bp);
1271 			BO_UNLOCK(bo);
1272 			bp->b_flags |= B_ASYNC;
1273 			bwrite(bp);
1274 			BO_LOCK(bo);
1275 			return (EAGAIN);	/* XXX: why not loop ? */
1276 		}
1277 		BO_LOCK(bo);
1278 		bremfree(bp);
1279 		BO_UNLOCK(bo);
1280 		bp->b_flags |= (B_INVAL | B_RELBUF);
1281 		bp->b_flags &= ~B_ASYNC;
1282 		brelse(bp);
1283 		BO_LOCK(bo);
1284 		if (nbp != NULL &&
1285 		    (nbp->b_bufobj != bo ||
1286 		     nbp->b_lblkno != lblkno ||
1287 		     (nbp->b_xflags &
1288 		      (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN)) != xflags))
1289 			break;			/* nbp invalid */
1290 	}
1291 	return (retval);
1292 }
1293 
1294 /*
1295  * Truncate a file's buffer and pages to a specified length.  This
1296  * is in lieu of the old vinvalbuf mechanism, which performed unneeded
1297  * sync activity.
1298  */
1299 int
1300 vtruncbuf(struct vnode *vp, struct ucred *cred, struct thread *td,
1301     off_t length, int blksize)
1302 {
1303 	struct buf *bp, *nbp;
1304 	int anyfreed;
1305 	int trunclbn;
1306 	struct bufobj *bo;
1307 
1308 	CTR5(KTR_VFS, "%s: vp %p with cred %p and block %d:%ju", __func__,
1309 	    vp, cred, blksize, (uintmax_t)length);
1310 
1311 	/*
1312 	 * Round up to the *next* lbn.
1313 	 */
1314 	trunclbn = (length + blksize - 1) / blksize;
1315 
1316 	ASSERT_VOP_LOCKED(vp, "vtruncbuf");
1317 restart:
1318 	bo = &vp->v_bufobj;
1319 	BO_LOCK(bo);
1320 	anyfreed = 1;
1321 	for (;anyfreed;) {
1322 		anyfreed = 0;
1323 		TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) {
1324 			if (bp->b_lblkno < trunclbn)
1325 				continue;
1326 			if (BUF_LOCK(bp,
1327 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1328 			    BO_MTX(bo)) == ENOLCK)
1329 				goto restart;
1330 
1331 			BO_LOCK(bo);
1332 			bremfree(bp);
1333 			BO_UNLOCK(bo);
1334 			bp->b_flags |= (B_INVAL | B_RELBUF);
1335 			bp->b_flags &= ~B_ASYNC;
1336 			brelse(bp);
1337 			anyfreed = 1;
1338 
1339 			if (nbp != NULL &&
1340 			    (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1341 			    (nbp->b_vp != vp) ||
1342 			    (nbp->b_flags & B_DELWRI))) {
1343 				goto restart;
1344 			}
1345 			BO_LOCK(bo);
1346 		}
1347 
1348 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1349 			if (bp->b_lblkno < trunclbn)
1350 				continue;
1351 			if (BUF_LOCK(bp,
1352 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1353 			    BO_MTX(bo)) == ENOLCK)
1354 				goto restart;
1355 			BO_LOCK(bo);
1356 			bremfree(bp);
1357 			BO_UNLOCK(bo);
1358 			bp->b_flags |= (B_INVAL | B_RELBUF);
1359 			bp->b_flags &= ~B_ASYNC;
1360 			brelse(bp);
1361 			anyfreed = 1;
1362 			if (nbp != NULL &&
1363 			    (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1364 			    (nbp->b_vp != vp) ||
1365 			    (nbp->b_flags & B_DELWRI) == 0)) {
1366 				goto restart;
1367 			}
1368 			BO_LOCK(bo);
1369 		}
1370 	}
1371 
1372 	if (length > 0) {
1373 restartsync:
1374 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1375 			if (bp->b_lblkno > 0)
1376 				continue;
1377 			/*
1378 			 * Since we hold the vnode lock this should only
1379 			 * fail if we're racing with the buf daemon.
1380 			 */
1381 			if (BUF_LOCK(bp,
1382 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1383 			    BO_MTX(bo)) == ENOLCK) {
1384 				goto restart;
1385 			}
1386 			VNASSERT((bp->b_flags & B_DELWRI), vp,
1387 			    ("buf(%p) on dirty queue without DELWRI", bp));
1388 
1389 			BO_LOCK(bo);
1390 			bremfree(bp);
1391 			BO_UNLOCK(bo);
1392 			bawrite(bp);
1393 			BO_LOCK(bo);
1394 			goto restartsync;
1395 		}
1396 	}
1397 
1398 	bufobj_wwait(bo, 0, 0);
1399 	BO_UNLOCK(bo);
1400 	vnode_pager_setsize(vp, length);
1401 
1402 	return (0);
1403 }
1404 
1405 /*
1406  * buf_splay() - splay tree core for the clean/dirty list of buffers in
1407  *		 a vnode.
1408  *
1409  *	NOTE: We have to deal with the special case of a background bitmap
1410  *	buffer, a situation where two buffers will have the same logical
1411  *	block offset.  We want (1) only the foreground buffer to be accessed
1412  *	in a lookup and (2) must differentiate between the foreground and
1413  *	background buffer in the splay tree algorithm because the splay
1414  *	tree cannot normally handle multiple entities with the same 'index'.
1415  *	We accomplish this by adding differentiating flags to the splay tree's
1416  *	numerical domain.
1417  */
1418 static
1419 struct buf *
1420 buf_splay(daddr_t lblkno, b_xflags_t xflags, struct buf *root)
1421 {
1422 	struct buf dummy;
1423 	struct buf *lefttreemax, *righttreemin, *y;
1424 
1425 	if (root == NULL)
1426 		return (NULL);
1427 	lefttreemax = righttreemin = &dummy;
1428 	for (;;) {
1429 		if (lblkno < root->b_lblkno ||
1430 		    (lblkno == root->b_lblkno &&
1431 		    (xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1432 			if ((y = root->b_left) == NULL)
1433 				break;
1434 			if (lblkno < y->b_lblkno) {
1435 				/* Rotate right. */
1436 				root->b_left = y->b_right;
1437 				y->b_right = root;
1438 				root = y;
1439 				if ((y = root->b_left) == NULL)
1440 					break;
1441 			}
1442 			/* Link into the new root's right tree. */
1443 			righttreemin->b_left = root;
1444 			righttreemin = root;
1445 		} else if (lblkno > root->b_lblkno ||
1446 		    (lblkno == root->b_lblkno &&
1447 		    (xflags & BX_BKGRDMARKER) > (root->b_xflags & BX_BKGRDMARKER))) {
1448 			if ((y = root->b_right) == NULL)
1449 				break;
1450 			if (lblkno > y->b_lblkno) {
1451 				/* Rotate left. */
1452 				root->b_right = y->b_left;
1453 				y->b_left = root;
1454 				root = y;
1455 				if ((y = root->b_right) == NULL)
1456 					break;
1457 			}
1458 			/* Link into the new root's left tree. */
1459 			lefttreemax->b_right = root;
1460 			lefttreemax = root;
1461 		} else {
1462 			break;
1463 		}
1464 		root = y;
1465 	}
1466 	/* Assemble the new root. */
1467 	lefttreemax->b_right = root->b_left;
1468 	righttreemin->b_left = root->b_right;
1469 	root->b_left = dummy.b_right;
1470 	root->b_right = dummy.b_left;
1471 	return (root);
1472 }
1473 
1474 static void
1475 buf_vlist_remove(struct buf *bp)
1476 {
1477 	struct buf *root;
1478 	struct bufv *bv;
1479 
1480 	KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp));
1481 	ASSERT_BO_LOCKED(bp->b_bufobj);
1482 	KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) !=
1483 	    (BX_VNDIRTY|BX_VNCLEAN),
1484 	    ("buf_vlist_remove: Buf %p is on two lists", bp));
1485 	if (bp->b_xflags & BX_VNDIRTY)
1486 		bv = &bp->b_bufobj->bo_dirty;
1487 	else
1488 		bv = &bp->b_bufobj->bo_clean;
1489 	if (bp != bv->bv_root) {
1490 		root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1491 		KASSERT(root == bp, ("splay lookup failed in remove"));
1492 	}
1493 	if (bp->b_left == NULL) {
1494 		root = bp->b_right;
1495 	} else {
1496 		root = buf_splay(bp->b_lblkno, bp->b_xflags, bp->b_left);
1497 		root->b_right = bp->b_right;
1498 	}
1499 	bv->bv_root = root;
1500 	TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs);
1501 	bv->bv_cnt--;
1502 	bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1503 }
1504 
1505 /*
1506  * Add the buffer to the sorted clean or dirty block list using a
1507  * splay tree algorithm.
1508  *
1509  * NOTE: xflags is passed as a constant, optimizing this inline function!
1510  */
1511 static void
1512 buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags)
1513 {
1514 	struct buf *root;
1515 	struct bufv *bv;
1516 
1517 	ASSERT_BO_LOCKED(bo);
1518 	KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0,
1519 	    ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags));
1520 	bp->b_xflags |= xflags;
1521 	if (xflags & BX_VNDIRTY)
1522 		bv = &bo->bo_dirty;
1523 	else
1524 		bv = &bo->bo_clean;
1525 
1526 	root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1527 	if (root == NULL) {
1528 		bp->b_left = NULL;
1529 		bp->b_right = NULL;
1530 		TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs);
1531 	} else if (bp->b_lblkno < root->b_lblkno ||
1532 	    (bp->b_lblkno == root->b_lblkno &&
1533 	    (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1534 		bp->b_left = root->b_left;
1535 		bp->b_right = root;
1536 		root->b_left = NULL;
1537 		TAILQ_INSERT_BEFORE(root, bp, b_bobufs);
1538 	} else {
1539 		bp->b_right = root->b_right;
1540 		bp->b_left = root;
1541 		root->b_right = NULL;
1542 		TAILQ_INSERT_AFTER(&bv->bv_hd, root, bp, b_bobufs);
1543 	}
1544 	bv->bv_cnt++;
1545 	bv->bv_root = bp;
1546 }
1547 
1548 /*
1549  * Lookup a buffer using the splay tree.  Note that we specifically avoid
1550  * shadow buffers used in background bitmap writes.
1551  *
1552  * This code isn't quite efficient as it could be because we are maintaining
1553  * two sorted lists and do not know which list the block resides in.
1554  *
1555  * During a "make buildworld" the desired buffer is found at one of
1556  * the roots more than 60% of the time.  Thus, checking both roots
1557  * before performing either splay eliminates unnecessary splays on the
1558  * first tree splayed.
1559  */
1560 struct buf *
1561 gbincore(struct bufobj *bo, daddr_t lblkno)
1562 {
1563 	struct buf *bp;
1564 
1565 	ASSERT_BO_LOCKED(bo);
1566 	if ((bp = bo->bo_clean.bv_root) != NULL &&
1567 	    bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1568 		return (bp);
1569 	if ((bp = bo->bo_dirty.bv_root) != NULL &&
1570 	    bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1571 		return (bp);
1572 	if ((bp = bo->bo_clean.bv_root) != NULL) {
1573 		bo->bo_clean.bv_root = bp = buf_splay(lblkno, 0, bp);
1574 		if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1575 			return (bp);
1576 	}
1577 	if ((bp = bo->bo_dirty.bv_root) != NULL) {
1578 		bo->bo_dirty.bv_root = bp = buf_splay(lblkno, 0, bp);
1579 		if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1580 			return (bp);
1581 	}
1582 	return (NULL);
1583 }
1584 
1585 /*
1586  * Associate a buffer with a vnode.
1587  */
1588 void
1589 bgetvp(struct vnode *vp, struct buf *bp)
1590 {
1591 	struct bufobj *bo;
1592 
1593 	bo = &vp->v_bufobj;
1594 	ASSERT_BO_LOCKED(bo);
1595 	VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free"));
1596 
1597 	CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags);
1598 	VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp,
1599 	    ("bgetvp: bp already attached! %p", bp));
1600 
1601 	vhold(vp);
1602 	if (VFS_NEEDSGIANT(vp->v_mount) || bo->bo_flag & BO_NEEDSGIANT)
1603 		bp->b_flags |= B_NEEDSGIANT;
1604 	bp->b_vp = vp;
1605 	bp->b_bufobj = bo;
1606 	/*
1607 	 * Insert onto list for new vnode.
1608 	 */
1609 	buf_vlist_add(bp, bo, BX_VNCLEAN);
1610 }
1611 
1612 /*
1613  * Disassociate a buffer from a vnode.
1614  */
1615 void
1616 brelvp(struct buf *bp)
1617 {
1618 	struct bufobj *bo;
1619 	struct vnode *vp;
1620 
1621 	CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
1622 	KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1623 
1624 	/*
1625 	 * Delete from old vnode list, if on one.
1626 	 */
1627 	vp = bp->b_vp;		/* XXX */
1628 	bo = bp->b_bufobj;
1629 	BO_LOCK(bo);
1630 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1631 		buf_vlist_remove(bp);
1632 	else
1633 		panic("brelvp: Buffer %p not on queue.", bp);
1634 	if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1635 		bo->bo_flag &= ~BO_ONWORKLST;
1636 		mtx_lock(&sync_mtx);
1637 		LIST_REMOVE(bo, bo_synclist);
1638 		syncer_worklist_len--;
1639 		mtx_unlock(&sync_mtx);
1640 	}
1641 	bp->b_flags &= ~B_NEEDSGIANT;
1642 	bp->b_vp = NULL;
1643 	bp->b_bufobj = NULL;
1644 	BO_UNLOCK(bo);
1645 	vdrop(vp);
1646 }
1647 
1648 /*
1649  * Add an item to the syncer work queue.
1650  */
1651 static void
1652 vn_syncer_add_to_worklist(struct bufobj *bo, int delay)
1653 {
1654 	int queue, slot;
1655 
1656 	ASSERT_BO_LOCKED(bo);
1657 
1658 	mtx_lock(&sync_mtx);
1659 	if (bo->bo_flag & BO_ONWORKLST)
1660 		LIST_REMOVE(bo, bo_synclist);
1661 	else {
1662 		bo->bo_flag |= BO_ONWORKLST;
1663 		syncer_worklist_len++;
1664 	}
1665 
1666 	if (delay > syncer_maxdelay - 2)
1667 		delay = syncer_maxdelay - 2;
1668 	slot = (syncer_delayno + delay) & syncer_mask;
1669 
1670 	queue = VFS_NEEDSGIANT(bo->__bo_vnode->v_mount) ? WI_GIANTQ :
1671 	    WI_MPSAFEQ;
1672 	LIST_INSERT_HEAD(&syncer_workitem_pending[queue][slot], bo,
1673 	    bo_synclist);
1674 	mtx_unlock(&sync_mtx);
1675 }
1676 
1677 static int
1678 sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS)
1679 {
1680 	int error, len;
1681 
1682 	mtx_lock(&sync_mtx);
1683 	len = syncer_worklist_len - sync_vnode_count;
1684 	mtx_unlock(&sync_mtx);
1685 	error = SYSCTL_OUT(req, &len, sizeof(len));
1686 	return (error);
1687 }
1688 
1689 SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0,
1690     sysctl_vfs_worklist_len, "I", "Syncer thread worklist length");
1691 
1692 static struct proc *updateproc;
1693 static void sched_sync(void);
1694 static struct kproc_desc up_kp = {
1695 	"syncer",
1696 	sched_sync,
1697 	&updateproc
1698 };
1699 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp);
1700 
1701 static int
1702 sync_vnode(struct synclist *slp, struct bufobj **bo, struct thread *td)
1703 {
1704 	struct vnode *vp;
1705 	struct mount *mp;
1706 
1707 	*bo = LIST_FIRST(slp);
1708 	if (*bo == NULL)
1709 		return (0);
1710 	vp = (*bo)->__bo_vnode;	/* XXX */
1711 	if (VOP_ISLOCKED(vp) != 0 || VI_TRYLOCK(vp) == 0)
1712 		return (1);
1713 	/*
1714 	 * We use vhold in case the vnode does not
1715 	 * successfully sync.  vhold prevents the vnode from
1716 	 * going away when we unlock the sync_mtx so that
1717 	 * we can acquire the vnode interlock.
1718 	 */
1719 	vholdl(vp);
1720 	mtx_unlock(&sync_mtx);
1721 	VI_UNLOCK(vp);
1722 	if (vn_start_write(vp, &mp, V_NOWAIT) != 0) {
1723 		vdrop(vp);
1724 		mtx_lock(&sync_mtx);
1725 		return (*bo == LIST_FIRST(slp));
1726 	}
1727 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1728 	(void) VOP_FSYNC(vp, MNT_LAZY, td);
1729 	VOP_UNLOCK(vp, 0);
1730 	vn_finished_write(mp);
1731 	BO_LOCK(*bo);
1732 	if (((*bo)->bo_flag & BO_ONWORKLST) != 0) {
1733 		/*
1734 		 * Put us back on the worklist.  The worklist
1735 		 * routine will remove us from our current
1736 		 * position and then add us back in at a later
1737 		 * position.
1738 		 */
1739 		vn_syncer_add_to_worklist(*bo, syncdelay);
1740 	}
1741 	BO_UNLOCK(*bo);
1742 	vdrop(vp);
1743 	mtx_lock(&sync_mtx);
1744 	return (0);
1745 }
1746 
1747 /*
1748  * System filesystem synchronizer daemon.
1749  */
1750 static void
1751 sched_sync(void)
1752 {
1753 	struct synclist *gnext, *next;
1754 	struct synclist *gslp, *slp;
1755 	struct bufobj *bo;
1756 	long starttime;
1757 	struct thread *td = curthread;
1758 	int last_work_seen;
1759 	int net_worklist_len;
1760 	int syncer_final_iter;
1761 	int first_printf;
1762 	int error;
1763 
1764 	last_work_seen = 0;
1765 	syncer_final_iter = 0;
1766 	first_printf = 1;
1767 	syncer_state = SYNCER_RUNNING;
1768 	starttime = time_uptime;
1769 	td->td_pflags |= TDP_NORUNNINGBUF;
1770 
1771 	EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc,
1772 	    SHUTDOWN_PRI_LAST);
1773 
1774 	mtx_lock(&sync_mtx);
1775 	for (;;) {
1776 		if (syncer_state == SYNCER_FINAL_DELAY &&
1777 		    syncer_final_iter == 0) {
1778 			mtx_unlock(&sync_mtx);
1779 			kproc_suspend_check(td->td_proc);
1780 			mtx_lock(&sync_mtx);
1781 		}
1782 		net_worklist_len = syncer_worklist_len - sync_vnode_count;
1783 		if (syncer_state != SYNCER_RUNNING &&
1784 		    starttime != time_uptime) {
1785 			if (first_printf) {
1786 				printf("\nSyncing disks, vnodes remaining...");
1787 				first_printf = 0;
1788 			}
1789 			printf("%d ", net_worklist_len);
1790 		}
1791 		starttime = time_uptime;
1792 
1793 		/*
1794 		 * Push files whose dirty time has expired.  Be careful
1795 		 * of interrupt race on slp queue.
1796 		 *
1797 		 * Skip over empty worklist slots when shutting down.
1798 		 */
1799 		do {
1800 			slp = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno];
1801 			gslp = &syncer_workitem_pending[WI_GIANTQ][syncer_delayno];
1802 			syncer_delayno += 1;
1803 			if (syncer_delayno == syncer_maxdelay)
1804 				syncer_delayno = 0;
1805 			next = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno];
1806 			gnext = &syncer_workitem_pending[WI_GIANTQ][syncer_delayno];
1807 			/*
1808 			 * If the worklist has wrapped since the
1809 			 * it was emptied of all but syncer vnodes,
1810 			 * switch to the FINAL_DELAY state and run
1811 			 * for one more second.
1812 			 */
1813 			if (syncer_state == SYNCER_SHUTTING_DOWN &&
1814 			    net_worklist_len == 0 &&
1815 			    last_work_seen == syncer_delayno) {
1816 				syncer_state = SYNCER_FINAL_DELAY;
1817 				syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP;
1818 			}
1819 		} while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) &&
1820 		    LIST_EMPTY(gslp) && syncer_worklist_len > 0);
1821 
1822 		/*
1823 		 * Keep track of the last time there was anything
1824 		 * on the worklist other than syncer vnodes.
1825 		 * Return to the SHUTTING_DOWN state if any
1826 		 * new work appears.
1827 		 */
1828 		if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING)
1829 			last_work_seen = syncer_delayno;
1830 		if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY)
1831 			syncer_state = SYNCER_SHUTTING_DOWN;
1832 		while (!LIST_EMPTY(slp)) {
1833 			error = sync_vnode(slp, &bo, td);
1834 			if (error == 1) {
1835 				LIST_REMOVE(bo, bo_synclist);
1836 				LIST_INSERT_HEAD(next, bo, bo_synclist);
1837 				continue;
1838 			}
1839 		}
1840 		if (!LIST_EMPTY(gslp)) {
1841 			mtx_unlock(&sync_mtx);
1842 			mtx_lock(&Giant);
1843 			mtx_lock(&sync_mtx);
1844 			while (!LIST_EMPTY(gslp)) {
1845 				error = sync_vnode(gslp, &bo, td);
1846 				if (error == 1) {
1847 					LIST_REMOVE(bo, bo_synclist);
1848 					LIST_INSERT_HEAD(gnext, bo,
1849 					    bo_synclist);
1850 					continue;
1851 				}
1852 			}
1853 			mtx_unlock(&Giant);
1854 		}
1855 		if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0)
1856 			syncer_final_iter--;
1857 		/*
1858 		 * The variable rushjob allows the kernel to speed up the
1859 		 * processing of the filesystem syncer process. A rushjob
1860 		 * value of N tells the filesystem syncer to process the next
1861 		 * N seconds worth of work on its queue ASAP. Currently rushjob
1862 		 * is used by the soft update code to speed up the filesystem
1863 		 * syncer process when the incore state is getting so far
1864 		 * ahead of the disk that the kernel memory pool is being
1865 		 * threatened with exhaustion.
1866 		 */
1867 		if (rushjob > 0) {
1868 			rushjob -= 1;
1869 			continue;
1870 		}
1871 		/*
1872 		 * Just sleep for a short period of time between
1873 		 * iterations when shutting down to allow some I/O
1874 		 * to happen.
1875 		 *
1876 		 * If it has taken us less than a second to process the
1877 		 * current work, then wait. Otherwise start right over
1878 		 * again. We can still lose time if any single round
1879 		 * takes more than two seconds, but it does not really
1880 		 * matter as we are just trying to generally pace the
1881 		 * filesystem activity.
1882 		 */
1883 		if (syncer_state != SYNCER_RUNNING)
1884 			cv_timedwait(&sync_wakeup, &sync_mtx,
1885 			    hz / SYNCER_SHUTDOWN_SPEEDUP);
1886 		else if (time_uptime == starttime)
1887 			cv_timedwait(&sync_wakeup, &sync_mtx, hz);
1888 	}
1889 }
1890 
1891 /*
1892  * Request the syncer daemon to speed up its work.
1893  * We never push it to speed up more than half of its
1894  * normal turn time, otherwise it could take over the cpu.
1895  */
1896 int
1897 speedup_syncer(void)
1898 {
1899 	int ret = 0;
1900 
1901 	mtx_lock(&sync_mtx);
1902 	if (rushjob < syncdelay / 2) {
1903 		rushjob += 1;
1904 		stat_rush_requests += 1;
1905 		ret = 1;
1906 	}
1907 	mtx_unlock(&sync_mtx);
1908 	cv_broadcast(&sync_wakeup);
1909 	return (ret);
1910 }
1911 
1912 /*
1913  * Tell the syncer to speed up its work and run though its work
1914  * list several times, then tell it to shut down.
1915  */
1916 static void
1917 syncer_shutdown(void *arg, int howto)
1918 {
1919 
1920 	if (howto & RB_NOSYNC)
1921 		return;
1922 	mtx_lock(&sync_mtx);
1923 	syncer_state = SYNCER_SHUTTING_DOWN;
1924 	rushjob = 0;
1925 	mtx_unlock(&sync_mtx);
1926 	cv_broadcast(&sync_wakeup);
1927 	kproc_shutdown(arg, howto);
1928 }
1929 
1930 /*
1931  * Reassign a buffer from one vnode to another.
1932  * Used to assign file specific control information
1933  * (indirect blocks) to the vnode to which they belong.
1934  */
1935 void
1936 reassignbuf(struct buf *bp)
1937 {
1938 	struct vnode *vp;
1939 	struct bufobj *bo;
1940 	int delay;
1941 #ifdef INVARIANTS
1942 	struct bufv *bv;
1943 #endif
1944 
1945 	vp = bp->b_vp;
1946 	bo = bp->b_bufobj;
1947 	++reassignbufcalls;
1948 
1949 	CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X",
1950 	    bp, bp->b_vp, bp->b_flags);
1951 	/*
1952 	 * B_PAGING flagged buffers cannot be reassigned because their vp
1953 	 * is not fully linked in.
1954 	 */
1955 	if (bp->b_flags & B_PAGING)
1956 		panic("cannot reassign paging buffer");
1957 
1958 	/*
1959 	 * Delete from old vnode list, if on one.
1960 	 */
1961 	BO_LOCK(bo);
1962 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1963 		buf_vlist_remove(bp);
1964 	else
1965 		panic("reassignbuf: Buffer %p not on queue.", bp);
1966 	/*
1967 	 * If dirty, put on list of dirty buffers; otherwise insert onto list
1968 	 * of clean buffers.
1969 	 */
1970 	if (bp->b_flags & B_DELWRI) {
1971 		if ((bo->bo_flag & BO_ONWORKLST) == 0) {
1972 			switch (vp->v_type) {
1973 			case VDIR:
1974 				delay = dirdelay;
1975 				break;
1976 			case VCHR:
1977 				delay = metadelay;
1978 				break;
1979 			default:
1980 				delay = filedelay;
1981 			}
1982 			vn_syncer_add_to_worklist(bo, delay);
1983 		}
1984 		buf_vlist_add(bp, bo, BX_VNDIRTY);
1985 	} else {
1986 		buf_vlist_add(bp, bo, BX_VNCLEAN);
1987 
1988 		if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1989 			mtx_lock(&sync_mtx);
1990 			LIST_REMOVE(bo, bo_synclist);
1991 			syncer_worklist_len--;
1992 			mtx_unlock(&sync_mtx);
1993 			bo->bo_flag &= ~BO_ONWORKLST;
1994 		}
1995 	}
1996 #ifdef INVARIANTS
1997 	bv = &bo->bo_clean;
1998 	bp = TAILQ_FIRST(&bv->bv_hd);
1999 	KASSERT(bp == NULL || bp->b_bufobj == bo,
2000 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2001 	bp = TAILQ_LAST(&bv->bv_hd, buflists);
2002 	KASSERT(bp == NULL || bp->b_bufobj == bo,
2003 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2004 	bv = &bo->bo_dirty;
2005 	bp = TAILQ_FIRST(&bv->bv_hd);
2006 	KASSERT(bp == NULL || bp->b_bufobj == bo,
2007 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2008 	bp = TAILQ_LAST(&bv->bv_hd, buflists);
2009 	KASSERT(bp == NULL || bp->b_bufobj == bo,
2010 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2011 #endif
2012 	BO_UNLOCK(bo);
2013 }
2014 
2015 /*
2016  * Increment the use and hold counts on the vnode, taking care to reference
2017  * the driver's usecount if this is a chardev.  The vholdl() will remove
2018  * the vnode from the free list if it is presently free.  Requires the
2019  * vnode interlock and returns with it held.
2020  */
2021 static void
2022 v_incr_usecount(struct vnode *vp)
2023 {
2024 
2025 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2026 	vp->v_usecount++;
2027 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2028 		dev_lock();
2029 		vp->v_rdev->si_usecount++;
2030 		dev_unlock();
2031 	}
2032 	vholdl(vp);
2033 }
2034 
2035 /*
2036  * Turn a holdcnt into a use+holdcnt such that only one call to
2037  * v_decr_usecount is needed.
2038  */
2039 static void
2040 v_upgrade_usecount(struct vnode *vp)
2041 {
2042 
2043 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2044 	vp->v_usecount++;
2045 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2046 		dev_lock();
2047 		vp->v_rdev->si_usecount++;
2048 		dev_unlock();
2049 	}
2050 }
2051 
2052 /*
2053  * Decrement the vnode use and hold count along with the driver's usecount
2054  * if this is a chardev.  The vdropl() below releases the vnode interlock
2055  * as it may free the vnode.
2056  */
2057 static void
2058 v_decr_usecount(struct vnode *vp)
2059 {
2060 
2061 	ASSERT_VI_LOCKED(vp, __FUNCTION__);
2062 	VNASSERT(vp->v_usecount > 0, vp,
2063 	    ("v_decr_usecount: negative usecount"));
2064 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2065 	vp->v_usecount--;
2066 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2067 		dev_lock();
2068 		vp->v_rdev->si_usecount--;
2069 		dev_unlock();
2070 	}
2071 	vdropl(vp);
2072 }
2073 
2074 /*
2075  * Decrement only the use count and driver use count.  This is intended to
2076  * be paired with a follow on vdropl() to release the remaining hold count.
2077  * In this way we may vgone() a vnode with a 0 usecount without risk of
2078  * having it end up on a free list because the hold count is kept above 0.
2079  */
2080 static void
2081 v_decr_useonly(struct vnode *vp)
2082 {
2083 
2084 	ASSERT_VI_LOCKED(vp, __FUNCTION__);
2085 	VNASSERT(vp->v_usecount > 0, vp,
2086 	    ("v_decr_useonly: negative usecount"));
2087 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2088 	vp->v_usecount--;
2089 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2090 		dev_lock();
2091 		vp->v_rdev->si_usecount--;
2092 		dev_unlock();
2093 	}
2094 }
2095 
2096 /*
2097  * Grab a particular vnode from the free list, increment its
2098  * reference count and lock it.  VI_DOOMED is set if the vnode
2099  * is being destroyed.  Only callers who specify LK_RETRY will
2100  * see doomed vnodes.  If inactive processing was delayed in
2101  * vput try to do it here.
2102  */
2103 int
2104 vget(struct vnode *vp, int flags, struct thread *td)
2105 {
2106 	int error;
2107 
2108 	error = 0;
2109 	VFS_ASSERT_GIANT(vp->v_mount);
2110 	VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
2111 	    ("vget: invalid lock operation"));
2112 	CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
2113 
2114 	if ((flags & LK_INTERLOCK) == 0)
2115 		VI_LOCK(vp);
2116 	vholdl(vp);
2117 	if ((error = vn_lock(vp, flags | LK_INTERLOCK)) != 0) {
2118 		vdrop(vp);
2119 		CTR2(KTR_VFS, "%s: impossible to lock vnode %p", __func__,
2120 		    vp);
2121 		return (error);
2122 	}
2123 	if (vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0)
2124 		panic("vget: vn_lock failed to return ENOENT\n");
2125 	VI_LOCK(vp);
2126 	/* Upgrade our holdcnt to a usecount. */
2127 	v_upgrade_usecount(vp);
2128 	/*
2129 	 * We don't guarantee that any particular close will
2130 	 * trigger inactive processing so just make a best effort
2131 	 * here at preventing a reference to a removed file.  If
2132 	 * we don't succeed no harm is done.
2133 	 */
2134 	if (vp->v_iflag & VI_OWEINACT) {
2135 		if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE &&
2136 		    (flags & LK_NOWAIT) == 0)
2137 			vinactive(vp, td);
2138 		vp->v_iflag &= ~VI_OWEINACT;
2139 	}
2140 	VI_UNLOCK(vp);
2141 	return (0);
2142 }
2143 
2144 /*
2145  * Increase the reference count of a vnode.
2146  */
2147 void
2148 vref(struct vnode *vp)
2149 {
2150 
2151 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2152 	VI_LOCK(vp);
2153 	v_incr_usecount(vp);
2154 	VI_UNLOCK(vp);
2155 }
2156 
2157 /*
2158  * Return reference count of a vnode.
2159  *
2160  * The results of this call are only guaranteed when some mechanism other
2161  * than the VI lock is used to stop other processes from gaining references
2162  * to the vnode.  This may be the case if the caller holds the only reference.
2163  * This is also useful when stale data is acceptable as race conditions may
2164  * be accounted for by some other means.
2165  */
2166 int
2167 vrefcnt(struct vnode *vp)
2168 {
2169 	int usecnt;
2170 
2171 	VI_LOCK(vp);
2172 	usecnt = vp->v_usecount;
2173 	VI_UNLOCK(vp);
2174 
2175 	return (usecnt);
2176 }
2177 
2178 #define	VPUTX_VRELE	1
2179 #define	VPUTX_VPUT	2
2180 #define	VPUTX_VUNREF	3
2181 
2182 static void
2183 vputx(struct vnode *vp, int func)
2184 {
2185 	int error;
2186 
2187 	KASSERT(vp != NULL, ("vputx: null vp"));
2188 	if (func == VPUTX_VUNREF)
2189 		ASSERT_VOP_LOCKED(vp, "vunref");
2190 	else if (func == VPUTX_VPUT)
2191 		ASSERT_VOP_LOCKED(vp, "vput");
2192 	else
2193 		KASSERT(func == VPUTX_VRELE, ("vputx: wrong func"));
2194 	VFS_ASSERT_GIANT(vp->v_mount);
2195 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2196 	VI_LOCK(vp);
2197 
2198 	/* Skip this v_writecount check if we're going to panic below. */
2199 	VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp,
2200 	    ("vputx: missed vn_close"));
2201 	error = 0;
2202 
2203 	if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2204 	    vp->v_usecount == 1)) {
2205 		if (func == VPUTX_VPUT)
2206 			VOP_UNLOCK(vp, 0);
2207 		v_decr_usecount(vp);
2208 		return;
2209 	}
2210 
2211 	if (vp->v_usecount != 1) {
2212 		vprint("vputx: negative ref count", vp);
2213 		panic("vputx: negative ref cnt");
2214 	}
2215 	CTR2(KTR_VFS, "%s: return vnode %p to the freelist", __func__, vp);
2216 	/*
2217 	 * We want to hold the vnode until the inactive finishes to
2218 	 * prevent vgone() races.  We drop the use count here and the
2219 	 * hold count below when we're done.
2220 	 */
2221 	v_decr_useonly(vp);
2222 	/*
2223 	 * We must call VOP_INACTIVE with the node locked. Mark
2224 	 * as VI_DOINGINACT to avoid recursion.
2225 	 */
2226 	vp->v_iflag |= VI_OWEINACT;
2227 	switch (func) {
2228 	case VPUTX_VRELE:
2229 		error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK);
2230 		VI_LOCK(vp);
2231 		break;
2232 	case VPUTX_VPUT:
2233 		if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
2234 			error = VOP_LOCK(vp, LK_UPGRADE | LK_INTERLOCK |
2235 			    LK_NOWAIT);
2236 			VI_LOCK(vp);
2237 		}
2238 		break;
2239 	case VPUTX_VUNREF:
2240 		if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
2241 			error = EBUSY;
2242 		break;
2243 	}
2244 	if (vp->v_usecount > 0)
2245 		vp->v_iflag &= ~VI_OWEINACT;
2246 	if (error == 0) {
2247 		if (vp->v_iflag & VI_OWEINACT)
2248 			vinactive(vp, curthread);
2249 		if (func != VPUTX_VUNREF)
2250 			VOP_UNLOCK(vp, 0);
2251 	}
2252 	vdropl(vp);
2253 }
2254 
2255 /*
2256  * Vnode put/release.
2257  * If count drops to zero, call inactive routine and return to freelist.
2258  */
2259 void
2260 vrele(struct vnode *vp)
2261 {
2262 
2263 	vputx(vp, VPUTX_VRELE);
2264 }
2265 
2266 /*
2267  * Release an already locked vnode.  This give the same effects as
2268  * unlock+vrele(), but takes less time and avoids releasing and
2269  * re-aquiring the lock (as vrele() acquires the lock internally.)
2270  */
2271 void
2272 vput(struct vnode *vp)
2273 {
2274 
2275 	vputx(vp, VPUTX_VPUT);
2276 }
2277 
2278 /*
2279  * Release an exclusively locked vnode. Do not unlock the vnode lock.
2280  */
2281 void
2282 vunref(struct vnode *vp)
2283 {
2284 
2285 	vputx(vp, VPUTX_VUNREF);
2286 }
2287 
2288 /*
2289  * Somebody doesn't want the vnode recycled.
2290  */
2291 void
2292 vhold(struct vnode *vp)
2293 {
2294 
2295 	VI_LOCK(vp);
2296 	vholdl(vp);
2297 	VI_UNLOCK(vp);
2298 }
2299 
2300 void
2301 vholdl(struct vnode *vp)
2302 {
2303 
2304 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2305 	vp->v_holdcnt++;
2306 	if (VSHOULDBUSY(vp))
2307 		vbusy(vp);
2308 }
2309 
2310 /*
2311  * Note that there is one less who cares about this vnode.  vdrop() is the
2312  * opposite of vhold().
2313  */
2314 void
2315 vdrop(struct vnode *vp)
2316 {
2317 
2318 	VI_LOCK(vp);
2319 	vdropl(vp);
2320 }
2321 
2322 /*
2323  * Drop the hold count of the vnode.  If this is the last reference to
2324  * the vnode we will free it if it has been vgone'd otherwise it is
2325  * placed on the free list.
2326  */
2327 void
2328 vdropl(struct vnode *vp)
2329 {
2330 
2331 	ASSERT_VI_LOCKED(vp, "vdropl");
2332 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2333 	if (vp->v_holdcnt <= 0)
2334 		panic("vdrop: holdcnt %d", vp->v_holdcnt);
2335 	vp->v_holdcnt--;
2336 	if (vp->v_holdcnt == 0) {
2337 		if (vp->v_iflag & VI_DOOMED) {
2338 			CTR2(KTR_VFS, "%s: destroying the vnode %p", __func__,
2339 			    vp);
2340 			vdestroy(vp);
2341 			return;
2342 		} else
2343 			vfree(vp);
2344 	}
2345 	VI_UNLOCK(vp);
2346 }
2347 
2348 /*
2349  * Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT
2350  * flags.  DOINGINACT prevents us from recursing in calls to vinactive.
2351  * OWEINACT tracks whether a vnode missed a call to inactive due to a
2352  * failed lock upgrade.
2353  */
2354 static void
2355 vinactive(struct vnode *vp, struct thread *td)
2356 {
2357 
2358 	ASSERT_VOP_ELOCKED(vp, "vinactive");
2359 	ASSERT_VI_LOCKED(vp, "vinactive");
2360 	VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp,
2361 	    ("vinactive: recursed on VI_DOINGINACT"));
2362 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2363 	vp->v_iflag |= VI_DOINGINACT;
2364 	vp->v_iflag &= ~VI_OWEINACT;
2365 	VI_UNLOCK(vp);
2366 	VOP_INACTIVE(vp, td);
2367 	VI_LOCK(vp);
2368 	VNASSERT(vp->v_iflag & VI_DOINGINACT, vp,
2369 	    ("vinactive: lost VI_DOINGINACT"));
2370 	vp->v_iflag &= ~VI_DOINGINACT;
2371 }
2372 
2373 /*
2374  * Remove any vnodes in the vnode table belonging to mount point mp.
2375  *
2376  * If FORCECLOSE is not specified, there should not be any active ones,
2377  * return error if any are found (nb: this is a user error, not a
2378  * system error). If FORCECLOSE is specified, detach any active vnodes
2379  * that are found.
2380  *
2381  * If WRITECLOSE is set, only flush out regular file vnodes open for
2382  * writing.
2383  *
2384  * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
2385  *
2386  * `rootrefs' specifies the base reference count for the root vnode
2387  * of this filesystem. The root vnode is considered busy if its
2388  * v_usecount exceeds this value. On a successful return, vflush(, td)
2389  * will call vrele() on the root vnode exactly rootrefs times.
2390  * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
2391  * be zero.
2392  */
2393 #ifdef DIAGNOSTIC
2394 static int busyprt = 0;		/* print out busy vnodes */
2395 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "Print out busy vnodes");
2396 #endif
2397 
2398 int
2399 vflush(struct mount *mp, int rootrefs, int flags, struct thread *td)
2400 {
2401 	struct vnode *vp, *mvp, *rootvp = NULL;
2402 	struct vattr vattr;
2403 	int busy = 0, error;
2404 
2405 	CTR4(KTR_VFS, "%s: mp %p with rootrefs %d and flags %d", __func__, mp,
2406 	    rootrefs, flags);
2407 	if (rootrefs > 0) {
2408 		KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
2409 		    ("vflush: bad args"));
2410 		/*
2411 		 * Get the filesystem root vnode. We can vput() it
2412 		 * immediately, since with rootrefs > 0, it won't go away.
2413 		 */
2414 		if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp)) != 0) {
2415 			CTR2(KTR_VFS, "%s: vfs_root lookup failed with %d",
2416 			    __func__, error);
2417 			return (error);
2418 		}
2419 		vput(rootvp);
2420 	}
2421 	MNT_ILOCK(mp);
2422 loop:
2423 	MNT_VNODE_FOREACH(vp, mp, mvp) {
2424 		VI_LOCK(vp);
2425 		vholdl(vp);
2426 		MNT_IUNLOCK(mp);
2427 		error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE);
2428 		if (error) {
2429 			vdrop(vp);
2430 			MNT_ILOCK(mp);
2431 			MNT_VNODE_FOREACH_ABORT_ILOCKED(mp, mvp);
2432 			goto loop;
2433 		}
2434 		/*
2435 		 * Skip over a vnodes marked VV_SYSTEM.
2436 		 */
2437 		if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
2438 			VOP_UNLOCK(vp, 0);
2439 			vdrop(vp);
2440 			MNT_ILOCK(mp);
2441 			continue;
2442 		}
2443 		/*
2444 		 * If WRITECLOSE is set, flush out unlinked but still open
2445 		 * files (even if open only for reading) and regular file
2446 		 * vnodes open for writing.
2447 		 */
2448 		if (flags & WRITECLOSE) {
2449 			error = VOP_GETATTR(vp, &vattr, td->td_ucred);
2450 			VI_LOCK(vp);
2451 
2452 			if ((vp->v_type == VNON ||
2453 			    (error == 0 && vattr.va_nlink > 0)) &&
2454 			    (vp->v_writecount == 0 || vp->v_type != VREG)) {
2455 				VOP_UNLOCK(vp, 0);
2456 				vdropl(vp);
2457 				MNT_ILOCK(mp);
2458 				continue;
2459 			}
2460 		} else
2461 			VI_LOCK(vp);
2462 		/*
2463 		 * With v_usecount == 0, all we need to do is clear out the
2464 		 * vnode data structures and we are done.
2465 		 *
2466 		 * If FORCECLOSE is set, forcibly close the vnode.
2467 		 */
2468 		if (vp->v_usecount == 0 || (flags & FORCECLOSE)) {
2469 			VNASSERT(vp->v_usecount == 0 ||
2470 			    (vp->v_type != VCHR && vp->v_type != VBLK), vp,
2471 			    ("device VNODE %p is FORCECLOSED", vp));
2472 			vgonel(vp);
2473 		} else {
2474 			busy++;
2475 #ifdef DIAGNOSTIC
2476 			if (busyprt)
2477 				vprint("vflush: busy vnode", vp);
2478 #endif
2479 		}
2480 		VOP_UNLOCK(vp, 0);
2481 		vdropl(vp);
2482 		MNT_ILOCK(mp);
2483 	}
2484 	MNT_IUNLOCK(mp);
2485 	if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
2486 		/*
2487 		 * If just the root vnode is busy, and if its refcount
2488 		 * is equal to `rootrefs', then go ahead and kill it.
2489 		 */
2490 		VI_LOCK(rootvp);
2491 		KASSERT(busy > 0, ("vflush: not busy"));
2492 		VNASSERT(rootvp->v_usecount >= rootrefs, rootvp,
2493 		    ("vflush: usecount %d < rootrefs %d",
2494 		     rootvp->v_usecount, rootrefs));
2495 		if (busy == 1 && rootvp->v_usecount == rootrefs) {
2496 			VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK);
2497 			vgone(rootvp);
2498 			VOP_UNLOCK(rootvp, 0);
2499 			busy = 0;
2500 		} else
2501 			VI_UNLOCK(rootvp);
2502 	}
2503 	if (busy) {
2504 		CTR2(KTR_VFS, "%s: failing as %d vnodes are busy", __func__,
2505 		    busy);
2506 		return (EBUSY);
2507 	}
2508 	for (; rootrefs > 0; rootrefs--)
2509 		vrele(rootvp);
2510 	return (0);
2511 }
2512 
2513 /*
2514  * Recycle an unused vnode to the front of the free list.
2515  */
2516 int
2517 vrecycle(struct vnode *vp, struct thread *td)
2518 {
2519 	int recycled;
2520 
2521 	ASSERT_VOP_ELOCKED(vp, "vrecycle");
2522 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2523 	recycled = 0;
2524 	VI_LOCK(vp);
2525 	if (vp->v_usecount == 0) {
2526 		recycled = 1;
2527 		vgonel(vp);
2528 	}
2529 	VI_UNLOCK(vp);
2530 	return (recycled);
2531 }
2532 
2533 /*
2534  * Eliminate all activity associated with a vnode
2535  * in preparation for reuse.
2536  */
2537 void
2538 vgone(struct vnode *vp)
2539 {
2540 	VI_LOCK(vp);
2541 	vgonel(vp);
2542 	VI_UNLOCK(vp);
2543 }
2544 
2545 /*
2546  * vgone, with the vp interlock held.
2547  */
2548 void
2549 vgonel(struct vnode *vp)
2550 {
2551 	struct thread *td;
2552 	int oweinact;
2553 	int active;
2554 	struct mount *mp;
2555 
2556 	ASSERT_VOP_ELOCKED(vp, "vgonel");
2557 	ASSERT_VI_LOCKED(vp, "vgonel");
2558 	VNASSERT(vp->v_holdcnt, vp,
2559 	    ("vgonel: vp %p has no reference.", vp));
2560 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2561 	td = curthread;
2562 
2563 	/*
2564 	 * Don't vgonel if we're already doomed.
2565 	 */
2566 	if (vp->v_iflag & VI_DOOMED)
2567 		return;
2568 	vp->v_iflag |= VI_DOOMED;
2569 	/*
2570 	 * Check to see if the vnode is in use.  If so, we have to call
2571 	 * VOP_CLOSE() and VOP_INACTIVE().
2572 	 */
2573 	active = vp->v_usecount;
2574 	oweinact = (vp->v_iflag & VI_OWEINACT);
2575 	VI_UNLOCK(vp);
2576 	/*
2577 	 * Clean out any buffers associated with the vnode.
2578 	 * If the flush fails, just toss the buffers.
2579 	 */
2580 	mp = NULL;
2581 	if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd))
2582 		(void) vn_start_secondary_write(vp, &mp, V_WAIT);
2583 	if (vinvalbuf(vp, V_SAVE, 0, 0) != 0)
2584 		vinvalbuf(vp, 0, 0, 0);
2585 
2586 	/*
2587 	 * If purging an active vnode, it must be closed and
2588 	 * deactivated before being reclaimed.
2589 	 */
2590 	if (active)
2591 		VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
2592 	if (oweinact || active) {
2593 		VI_LOCK(vp);
2594 		if ((vp->v_iflag & VI_DOINGINACT) == 0)
2595 			vinactive(vp, td);
2596 		VI_UNLOCK(vp);
2597 	}
2598 	/*
2599 	 * Reclaim the vnode.
2600 	 */
2601 	if (VOP_RECLAIM(vp, td))
2602 		panic("vgone: cannot reclaim");
2603 	if (mp != NULL)
2604 		vn_finished_secondary_write(mp);
2605 	VNASSERT(vp->v_object == NULL, vp,
2606 	    ("vop_reclaim left v_object vp=%p, tag=%s", vp, vp->v_tag));
2607 	/*
2608 	 * Clear the advisory locks and wake up waiting threads.
2609 	 */
2610 	(void)VOP_ADVLOCKPURGE(vp);
2611 	/*
2612 	 * Delete from old mount point vnode list.
2613 	 */
2614 	delmntque(vp);
2615 	cache_purge(vp);
2616 	/*
2617 	 * Done with purge, reset to the standard lock and invalidate
2618 	 * the vnode.
2619 	 */
2620 	VI_LOCK(vp);
2621 	vp->v_vnlock = &vp->v_lock;
2622 	vp->v_op = &dead_vnodeops;
2623 	vp->v_tag = "none";
2624 	vp->v_type = VBAD;
2625 }
2626 
2627 /*
2628  * Calculate the total number of references to a special device.
2629  */
2630 int
2631 vcount(struct vnode *vp)
2632 {
2633 	int count;
2634 
2635 	dev_lock();
2636 	count = vp->v_rdev->si_usecount;
2637 	dev_unlock();
2638 	return (count);
2639 }
2640 
2641 /*
2642  * Same as above, but using the struct cdev *as argument
2643  */
2644 int
2645 count_dev(struct cdev *dev)
2646 {
2647 	int count;
2648 
2649 	dev_lock();
2650 	count = dev->si_usecount;
2651 	dev_unlock();
2652 	return(count);
2653 }
2654 
2655 /*
2656  * Print out a description of a vnode.
2657  */
2658 static char *typename[] =
2659 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD",
2660  "VMARKER"};
2661 
2662 void
2663 vn_printf(struct vnode *vp, const char *fmt, ...)
2664 {
2665 	va_list ap;
2666 	char buf[256], buf2[16];
2667 	u_long flags;
2668 
2669 	va_start(ap, fmt);
2670 	vprintf(fmt, ap);
2671 	va_end(ap);
2672 	printf("%p: ", (void *)vp);
2673 	printf("tag %s, type %s\n", vp->v_tag, typename[vp->v_type]);
2674 	printf("    usecount %d, writecount %d, refcount %d mountedhere %p\n",
2675 	    vp->v_usecount, vp->v_writecount, vp->v_holdcnt, vp->v_mountedhere);
2676 	buf[0] = '\0';
2677 	buf[1] = '\0';
2678 	if (vp->v_vflag & VV_ROOT)
2679 		strlcat(buf, "|VV_ROOT", sizeof(buf));
2680 	if (vp->v_vflag & VV_ISTTY)
2681 		strlcat(buf, "|VV_ISTTY", sizeof(buf));
2682 	if (vp->v_vflag & VV_NOSYNC)
2683 		strlcat(buf, "|VV_NOSYNC", sizeof(buf));
2684 	if (vp->v_vflag & VV_CACHEDLABEL)
2685 		strlcat(buf, "|VV_CACHEDLABEL", sizeof(buf));
2686 	if (vp->v_vflag & VV_TEXT)
2687 		strlcat(buf, "|VV_TEXT", sizeof(buf));
2688 	if (vp->v_vflag & VV_COPYONWRITE)
2689 		strlcat(buf, "|VV_COPYONWRITE", sizeof(buf));
2690 	if (vp->v_vflag & VV_SYSTEM)
2691 		strlcat(buf, "|VV_SYSTEM", sizeof(buf));
2692 	if (vp->v_vflag & VV_PROCDEP)
2693 		strlcat(buf, "|VV_PROCDEP", sizeof(buf));
2694 	if (vp->v_vflag & VV_NOKNOTE)
2695 		strlcat(buf, "|VV_NOKNOTE", sizeof(buf));
2696 	if (vp->v_vflag & VV_DELETED)
2697 		strlcat(buf, "|VV_DELETED", sizeof(buf));
2698 	if (vp->v_vflag & VV_MD)
2699 		strlcat(buf, "|VV_MD", sizeof(buf));
2700 	flags = vp->v_vflag & ~(VV_ROOT | VV_ISTTY | VV_NOSYNC |
2701 	    VV_CACHEDLABEL | VV_TEXT | VV_COPYONWRITE | VV_SYSTEM | VV_PROCDEP |
2702 	    VV_NOKNOTE | VV_DELETED | VV_MD);
2703 	if (flags != 0) {
2704 		snprintf(buf2, sizeof(buf2), "|VV(0x%lx)", flags);
2705 		strlcat(buf, buf2, sizeof(buf));
2706 	}
2707 	if (vp->v_iflag & VI_MOUNT)
2708 		strlcat(buf, "|VI_MOUNT", sizeof(buf));
2709 	if (vp->v_iflag & VI_AGE)
2710 		strlcat(buf, "|VI_AGE", sizeof(buf));
2711 	if (vp->v_iflag & VI_DOOMED)
2712 		strlcat(buf, "|VI_DOOMED", sizeof(buf));
2713 	if (vp->v_iflag & VI_FREE)
2714 		strlcat(buf, "|VI_FREE", sizeof(buf));
2715 	if (vp->v_iflag & VI_DOINGINACT)
2716 		strlcat(buf, "|VI_DOINGINACT", sizeof(buf));
2717 	if (vp->v_iflag & VI_OWEINACT)
2718 		strlcat(buf, "|VI_OWEINACT", sizeof(buf));
2719 	flags = vp->v_iflag & ~(VI_MOUNT | VI_AGE | VI_DOOMED | VI_FREE |
2720 	    VI_DOINGINACT | VI_OWEINACT);
2721 	if (flags != 0) {
2722 		snprintf(buf2, sizeof(buf2), "|VI(0x%lx)", flags);
2723 		strlcat(buf, buf2, sizeof(buf));
2724 	}
2725 	printf("    flags (%s)\n", buf + 1);
2726 	if (mtx_owned(VI_MTX(vp)))
2727 		printf(" VI_LOCKed");
2728 	if (vp->v_object != NULL)
2729 		printf("    v_object %p ref %d pages %d\n",
2730 		    vp->v_object, vp->v_object->ref_count,
2731 		    vp->v_object->resident_page_count);
2732 	printf("    ");
2733 	lockmgr_printinfo(vp->v_vnlock);
2734 	if (vp->v_data != NULL)
2735 		VOP_PRINT(vp);
2736 }
2737 
2738 #ifdef DDB
2739 /*
2740  * List all of the locked vnodes in the system.
2741  * Called when debugging the kernel.
2742  */
2743 DB_SHOW_COMMAND(lockedvnods, lockedvnodes)
2744 {
2745 	struct mount *mp, *nmp;
2746 	struct vnode *vp;
2747 
2748 	/*
2749 	 * Note: because this is DDB, we can't obey the locking semantics
2750 	 * for these structures, which means we could catch an inconsistent
2751 	 * state and dereference a nasty pointer.  Not much to be done
2752 	 * about that.
2753 	 */
2754 	db_printf("Locked vnodes\n");
2755 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2756 		nmp = TAILQ_NEXT(mp, mnt_list);
2757 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2758 			if (vp->v_type != VMARKER &&
2759 			    VOP_ISLOCKED(vp))
2760 				vprint("", vp);
2761 		}
2762 		nmp = TAILQ_NEXT(mp, mnt_list);
2763 	}
2764 }
2765 
2766 /*
2767  * Show details about the given vnode.
2768  */
2769 DB_SHOW_COMMAND(vnode, db_show_vnode)
2770 {
2771 	struct vnode *vp;
2772 
2773 	if (!have_addr)
2774 		return;
2775 	vp = (struct vnode *)addr;
2776 	vn_printf(vp, "vnode ");
2777 }
2778 
2779 /*
2780  * Show details about the given mount point.
2781  */
2782 DB_SHOW_COMMAND(mount, db_show_mount)
2783 {
2784 	struct mount *mp;
2785 	struct vfsopt *opt;
2786 	struct statfs *sp;
2787 	struct vnode *vp;
2788 	char buf[512];
2789 	u_int flags;
2790 
2791 	if (!have_addr) {
2792 		/* No address given, print short info about all mount points. */
2793 		TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2794 			db_printf("%p %s on %s (%s)\n", mp,
2795 			    mp->mnt_stat.f_mntfromname,
2796 			    mp->mnt_stat.f_mntonname,
2797 			    mp->mnt_stat.f_fstypename);
2798 			if (db_pager_quit)
2799 				break;
2800 		}
2801 		db_printf("\nMore info: show mount <addr>\n");
2802 		return;
2803 	}
2804 
2805 	mp = (struct mount *)addr;
2806 	db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname,
2807 	    mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename);
2808 
2809 	buf[0] = '\0';
2810 	flags = mp->mnt_flag;
2811 #define	MNT_FLAG(flag)	do {						\
2812 	if (flags & (flag)) {						\
2813 		if (buf[0] != '\0')					\
2814 			strlcat(buf, ", ", sizeof(buf));		\
2815 		strlcat(buf, (#flag) + 4, sizeof(buf));			\
2816 		flags &= ~(flag);					\
2817 	}								\
2818 } while (0)
2819 	MNT_FLAG(MNT_RDONLY);
2820 	MNT_FLAG(MNT_SYNCHRONOUS);
2821 	MNT_FLAG(MNT_NOEXEC);
2822 	MNT_FLAG(MNT_NOSUID);
2823 	MNT_FLAG(MNT_UNION);
2824 	MNT_FLAG(MNT_ASYNC);
2825 	MNT_FLAG(MNT_SUIDDIR);
2826 	MNT_FLAG(MNT_SOFTDEP);
2827 	MNT_FLAG(MNT_NOSYMFOLLOW);
2828 	MNT_FLAG(MNT_GJOURNAL);
2829 	MNT_FLAG(MNT_MULTILABEL);
2830 	MNT_FLAG(MNT_ACLS);
2831 	MNT_FLAG(MNT_NOATIME);
2832 	MNT_FLAG(MNT_NOCLUSTERR);
2833 	MNT_FLAG(MNT_NOCLUSTERW);
2834 	MNT_FLAG(MNT_NFS4ACLS);
2835 	MNT_FLAG(MNT_EXRDONLY);
2836 	MNT_FLAG(MNT_EXPORTED);
2837 	MNT_FLAG(MNT_DEFEXPORTED);
2838 	MNT_FLAG(MNT_EXPORTANON);
2839 	MNT_FLAG(MNT_EXKERB);
2840 	MNT_FLAG(MNT_EXPUBLIC);
2841 	MNT_FLAG(MNT_LOCAL);
2842 	MNT_FLAG(MNT_QUOTA);
2843 	MNT_FLAG(MNT_ROOTFS);
2844 	MNT_FLAG(MNT_USER);
2845 	MNT_FLAG(MNT_IGNORE);
2846 	MNT_FLAG(MNT_UPDATE);
2847 	MNT_FLAG(MNT_DELEXPORT);
2848 	MNT_FLAG(MNT_RELOAD);
2849 	MNT_FLAG(MNT_FORCE);
2850 	MNT_FLAG(MNT_SNAPSHOT);
2851 	MNT_FLAG(MNT_BYFSID);
2852 	MNT_FLAG(MNT_SOFTDEP);
2853 #undef MNT_FLAG
2854 	if (flags != 0) {
2855 		if (buf[0] != '\0')
2856 			strlcat(buf, ", ", sizeof(buf));
2857 		snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
2858 		    "0x%08x", flags);
2859 	}
2860 	db_printf("    mnt_flag = %s\n", buf);
2861 
2862 	buf[0] = '\0';
2863 	flags = mp->mnt_kern_flag;
2864 #define	MNT_KERN_FLAG(flag)	do {					\
2865 	if (flags & (flag)) {						\
2866 		if (buf[0] != '\0')					\
2867 			strlcat(buf, ", ", sizeof(buf));		\
2868 		strlcat(buf, (#flag) + 5, sizeof(buf));			\
2869 		flags &= ~(flag);					\
2870 	}								\
2871 } while (0)
2872 	MNT_KERN_FLAG(MNTK_UNMOUNTF);
2873 	MNT_KERN_FLAG(MNTK_ASYNC);
2874 	MNT_KERN_FLAG(MNTK_SOFTDEP);
2875 	MNT_KERN_FLAG(MNTK_NOINSMNTQ);
2876 	MNT_KERN_FLAG(MNTK_DRAINING);
2877 	MNT_KERN_FLAG(MNTK_REFEXPIRE);
2878 	MNT_KERN_FLAG(MNTK_EXTENDED_SHARED);
2879 	MNT_KERN_FLAG(MNTK_SHARED_WRITES);
2880 	MNT_KERN_FLAG(MNTK_UNMOUNT);
2881 	MNT_KERN_FLAG(MNTK_MWAIT);
2882 	MNT_KERN_FLAG(MNTK_SUSPEND);
2883 	MNT_KERN_FLAG(MNTK_SUSPEND2);
2884 	MNT_KERN_FLAG(MNTK_SUSPENDED);
2885 	MNT_KERN_FLAG(MNTK_MPSAFE);
2886 	MNT_KERN_FLAG(MNTK_LOOKUP_SHARED);
2887 	MNT_KERN_FLAG(MNTK_NOKNOTE);
2888 #undef MNT_KERN_FLAG
2889 	if (flags != 0) {
2890 		if (buf[0] != '\0')
2891 			strlcat(buf, ", ", sizeof(buf));
2892 		snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
2893 		    "0x%08x", flags);
2894 	}
2895 	db_printf("    mnt_kern_flag = %s\n", buf);
2896 
2897 	db_printf("    mnt_opt = ");
2898 	opt = TAILQ_FIRST(mp->mnt_opt);
2899 	if (opt != NULL) {
2900 		db_printf("%s", opt->name);
2901 		opt = TAILQ_NEXT(opt, link);
2902 		while (opt != NULL) {
2903 			db_printf(", %s", opt->name);
2904 			opt = TAILQ_NEXT(opt, link);
2905 		}
2906 	}
2907 	db_printf("\n");
2908 
2909 	sp = &mp->mnt_stat;
2910 	db_printf("    mnt_stat = { version=%u type=%u flags=0x%016jx "
2911 	    "bsize=%ju iosize=%ju blocks=%ju bfree=%ju bavail=%jd files=%ju "
2912 	    "ffree=%jd syncwrites=%ju asyncwrites=%ju syncreads=%ju "
2913 	    "asyncreads=%ju namemax=%u owner=%u fsid=[%d, %d] }\n",
2914 	    (u_int)sp->f_version, (u_int)sp->f_type, (uintmax_t)sp->f_flags,
2915 	    (uintmax_t)sp->f_bsize, (uintmax_t)sp->f_iosize,
2916 	    (uintmax_t)sp->f_blocks, (uintmax_t)sp->f_bfree,
2917 	    (intmax_t)sp->f_bavail, (uintmax_t)sp->f_files,
2918 	    (intmax_t)sp->f_ffree, (uintmax_t)sp->f_syncwrites,
2919 	    (uintmax_t)sp->f_asyncwrites, (uintmax_t)sp->f_syncreads,
2920 	    (uintmax_t)sp->f_asyncreads, (u_int)sp->f_namemax,
2921 	    (u_int)sp->f_owner, (int)sp->f_fsid.val[0], (int)sp->f_fsid.val[1]);
2922 
2923 	db_printf("    mnt_cred = { uid=%u ruid=%u",
2924 	    (u_int)mp->mnt_cred->cr_uid, (u_int)mp->mnt_cred->cr_ruid);
2925 	if (jailed(mp->mnt_cred))
2926 		db_printf(", jail=%d", mp->mnt_cred->cr_prison->pr_id);
2927 	db_printf(" }\n");
2928 	db_printf("    mnt_ref = %d\n", mp->mnt_ref);
2929 	db_printf("    mnt_gen = %d\n", mp->mnt_gen);
2930 	db_printf("    mnt_nvnodelistsize = %d\n", mp->mnt_nvnodelistsize);
2931 	db_printf("    mnt_writeopcount = %d\n", mp->mnt_writeopcount);
2932 	db_printf("    mnt_noasync = %u\n", mp->mnt_noasync);
2933 	db_printf("    mnt_maxsymlinklen = %d\n", mp->mnt_maxsymlinklen);
2934 	db_printf("    mnt_iosize_max = %d\n", mp->mnt_iosize_max);
2935 	db_printf("    mnt_hashseed = %u\n", mp->mnt_hashseed);
2936 	db_printf("    mnt_secondary_writes = %d\n", mp->mnt_secondary_writes);
2937 	db_printf("    mnt_secondary_accwrites = %d\n",
2938 	    mp->mnt_secondary_accwrites);
2939 	db_printf("    mnt_gjprovider = %s\n",
2940 	    mp->mnt_gjprovider != NULL ? mp->mnt_gjprovider : "NULL");
2941 	db_printf("\n");
2942 
2943 	TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2944 		if (vp->v_type != VMARKER) {
2945 			vn_printf(vp, "vnode ");
2946 			if (db_pager_quit)
2947 				break;
2948 		}
2949 	}
2950 }
2951 #endif	/* DDB */
2952 
2953 /*
2954  * Fill in a struct xvfsconf based on a struct vfsconf.
2955  */
2956 static void
2957 vfsconf2x(struct vfsconf *vfsp, struct xvfsconf *xvfsp)
2958 {
2959 
2960 	strcpy(xvfsp->vfc_name, vfsp->vfc_name);
2961 	xvfsp->vfc_typenum = vfsp->vfc_typenum;
2962 	xvfsp->vfc_refcount = vfsp->vfc_refcount;
2963 	xvfsp->vfc_flags = vfsp->vfc_flags;
2964 	/*
2965 	 * These are unused in userland, we keep them
2966 	 * to not break binary compatibility.
2967 	 */
2968 	xvfsp->vfc_vfsops = NULL;
2969 	xvfsp->vfc_next = NULL;
2970 }
2971 
2972 /*
2973  * Top level filesystem related information gathering.
2974  */
2975 static int
2976 sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS)
2977 {
2978 	struct vfsconf *vfsp;
2979 	struct xvfsconf xvfsp;
2980 	int error;
2981 
2982 	error = 0;
2983 	TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
2984 		bzero(&xvfsp, sizeof(xvfsp));
2985 		vfsconf2x(vfsp, &xvfsp);
2986 		error = SYSCTL_OUT(req, &xvfsp, sizeof xvfsp);
2987 		if (error)
2988 			break;
2989 	}
2990 	return (error);
2991 }
2992 
2993 SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLFLAG_RD, NULL, 0, sysctl_vfs_conflist,
2994     "S,xvfsconf", "List of all configured filesystems");
2995 
2996 #ifndef BURN_BRIDGES
2997 static int	sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
2998 
2999 static int
3000 vfs_sysctl(SYSCTL_HANDLER_ARGS)
3001 {
3002 	int *name = (int *)arg1 - 1;	/* XXX */
3003 	u_int namelen = arg2 + 1;	/* XXX */
3004 	struct vfsconf *vfsp;
3005 	struct xvfsconf xvfsp;
3006 
3007 	printf("WARNING: userland calling deprecated sysctl, "
3008 	    "please rebuild world\n");
3009 
3010 #if 1 || defined(COMPAT_PRELITE2)
3011 	/* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
3012 	if (namelen == 1)
3013 		return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
3014 #endif
3015 
3016 	switch (name[1]) {
3017 	case VFS_MAXTYPENUM:
3018 		if (namelen != 2)
3019 			return (ENOTDIR);
3020 		return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
3021 	case VFS_CONF:
3022 		if (namelen != 3)
3023 			return (ENOTDIR);	/* overloaded */
3024 		TAILQ_FOREACH(vfsp, &vfsconf, vfc_list)
3025 			if (vfsp->vfc_typenum == name[2])
3026 				break;
3027 		if (vfsp == NULL)
3028 			return (EOPNOTSUPP);
3029 		bzero(&xvfsp, sizeof(xvfsp));
3030 		vfsconf2x(vfsp, &xvfsp);
3031 		return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
3032 	}
3033 	return (EOPNOTSUPP);
3034 }
3035 
3036 static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP,
3037     vfs_sysctl, "Generic filesystem");
3038 
3039 #if 1 || defined(COMPAT_PRELITE2)
3040 
3041 static int
3042 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
3043 {
3044 	int error;
3045 	struct vfsconf *vfsp;
3046 	struct ovfsconf ovfs;
3047 
3048 	TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3049 		bzero(&ovfs, sizeof(ovfs));
3050 		ovfs.vfc_vfsops = vfsp->vfc_vfsops;	/* XXX used as flag */
3051 		strcpy(ovfs.vfc_name, vfsp->vfc_name);
3052 		ovfs.vfc_index = vfsp->vfc_typenum;
3053 		ovfs.vfc_refcount = vfsp->vfc_refcount;
3054 		ovfs.vfc_flags = vfsp->vfc_flags;
3055 		error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
3056 		if (error)
3057 			return error;
3058 	}
3059 	return 0;
3060 }
3061 
3062 #endif /* 1 || COMPAT_PRELITE2 */
3063 #endif /* !BURN_BRIDGES */
3064 
3065 #define KINFO_VNODESLOP		10
3066 #ifdef notyet
3067 /*
3068  * Dump vnode list (via sysctl).
3069  */
3070 /* ARGSUSED */
3071 static int
3072 sysctl_vnode(SYSCTL_HANDLER_ARGS)
3073 {
3074 	struct xvnode *xvn;
3075 	struct mount *mp;
3076 	struct vnode *vp;
3077 	int error, len, n;
3078 
3079 	/*
3080 	 * Stale numvnodes access is not fatal here.
3081 	 */
3082 	req->lock = 0;
3083 	len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn;
3084 	if (!req->oldptr)
3085 		/* Make an estimate */
3086 		return (SYSCTL_OUT(req, 0, len));
3087 
3088 	error = sysctl_wire_old_buffer(req, 0);
3089 	if (error != 0)
3090 		return (error);
3091 	xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK);
3092 	n = 0;
3093 	mtx_lock(&mountlist_mtx);
3094 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3095 		if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK))
3096 			continue;
3097 		MNT_ILOCK(mp);
3098 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3099 			if (n == len)
3100 				break;
3101 			vref(vp);
3102 			xvn[n].xv_size = sizeof *xvn;
3103 			xvn[n].xv_vnode = vp;
3104 			xvn[n].xv_id = 0;	/* XXX compat */
3105 #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field
3106 			XV_COPY(usecount);
3107 			XV_COPY(writecount);
3108 			XV_COPY(holdcnt);
3109 			XV_COPY(mount);
3110 			XV_COPY(numoutput);
3111 			XV_COPY(type);
3112 #undef XV_COPY
3113 			xvn[n].xv_flag = vp->v_vflag;
3114 
3115 			switch (vp->v_type) {
3116 			case VREG:
3117 			case VDIR:
3118 			case VLNK:
3119 				break;
3120 			case VBLK:
3121 			case VCHR:
3122 				if (vp->v_rdev == NULL) {
3123 					vrele(vp);
3124 					continue;
3125 				}
3126 				xvn[n].xv_dev = dev2udev(vp->v_rdev);
3127 				break;
3128 			case VSOCK:
3129 				xvn[n].xv_socket = vp->v_socket;
3130 				break;
3131 			case VFIFO:
3132 				xvn[n].xv_fifo = vp->v_fifoinfo;
3133 				break;
3134 			case VNON:
3135 			case VBAD:
3136 			default:
3137 				/* shouldn't happen? */
3138 				vrele(vp);
3139 				continue;
3140 			}
3141 			vrele(vp);
3142 			++n;
3143 		}
3144 		MNT_IUNLOCK(mp);
3145 		mtx_lock(&mountlist_mtx);
3146 		vfs_unbusy(mp);
3147 		if (n == len)
3148 			break;
3149 	}
3150 	mtx_unlock(&mountlist_mtx);
3151 
3152 	error = SYSCTL_OUT(req, xvn, n * sizeof *xvn);
3153 	free(xvn, M_TEMP);
3154 	return (error);
3155 }
3156 
3157 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
3158     0, 0, sysctl_vnode, "S,xvnode", "");
3159 #endif
3160 
3161 /*
3162  * Unmount all filesystems. The list is traversed in reverse order
3163  * of mounting to avoid dependencies.
3164  */
3165 void
3166 vfs_unmountall(void)
3167 {
3168 	struct mount *mp;
3169 	struct thread *td;
3170 	int error;
3171 
3172 	KASSERT(curthread != NULL, ("vfs_unmountall: NULL curthread"));
3173 	CTR1(KTR_VFS, "%s: unmounting all filesystems", __func__);
3174 	td = curthread;
3175 
3176 	/*
3177 	 * Since this only runs when rebooting, it is not interlocked.
3178 	 */
3179 	while(!TAILQ_EMPTY(&mountlist)) {
3180 		mp = TAILQ_LAST(&mountlist, mntlist);
3181 		error = dounmount(mp, MNT_FORCE, td);
3182 		if (error) {
3183 			TAILQ_REMOVE(&mountlist, mp, mnt_list);
3184 			/*
3185 			 * XXX: Due to the way in which we mount the root
3186 			 * file system off of devfs, devfs will generate a
3187 			 * "busy" warning when we try to unmount it before
3188 			 * the root.  Don't print a warning as a result in
3189 			 * order to avoid false positive errors that may
3190 			 * cause needless upset.
3191 			 */
3192 			if (strcmp(mp->mnt_vfc->vfc_name, "devfs") != 0) {
3193 				printf("unmount of %s failed (",
3194 				    mp->mnt_stat.f_mntonname);
3195 				if (error == EBUSY)
3196 					printf("BUSY)\n");
3197 				else
3198 					printf("%d)\n", error);
3199 			}
3200 		} else {
3201 			/* The unmount has removed mp from the mountlist */
3202 		}
3203 	}
3204 }
3205 
3206 /*
3207  * perform msync on all vnodes under a mount point
3208  * the mount point must be locked.
3209  */
3210 void
3211 vfs_msync(struct mount *mp, int flags)
3212 {
3213 	struct vnode *vp, *mvp;
3214 	struct vm_object *obj;
3215 
3216 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
3217 	MNT_ILOCK(mp);
3218 	MNT_VNODE_FOREACH(vp, mp, mvp) {
3219 		VI_LOCK(vp);
3220 		obj = vp->v_object;
3221 		if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0 &&
3222 		    (flags == MNT_WAIT || VOP_ISLOCKED(vp) == 0)) {
3223 			MNT_IUNLOCK(mp);
3224 			if (!vget(vp,
3225 			    LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
3226 			    curthread)) {
3227 				if (vp->v_vflag & VV_NOSYNC) {	/* unlinked */
3228 					vput(vp);
3229 					MNT_ILOCK(mp);
3230 					continue;
3231 				}
3232 
3233 				obj = vp->v_object;
3234 				if (obj != NULL) {
3235 					VM_OBJECT_LOCK(obj);
3236 					vm_object_page_clean(obj, 0, 0,
3237 					    flags == MNT_WAIT ?
3238 					    OBJPC_SYNC : OBJPC_NOSYNC);
3239 					VM_OBJECT_UNLOCK(obj);
3240 				}
3241 				vput(vp);
3242 			}
3243 			MNT_ILOCK(mp);
3244 		} else
3245 			VI_UNLOCK(vp);
3246 	}
3247 	MNT_IUNLOCK(mp);
3248 }
3249 
3250 /*
3251  * Mark a vnode as free, putting it up for recycling.
3252  */
3253 static void
3254 vfree(struct vnode *vp)
3255 {
3256 
3257 	ASSERT_VI_LOCKED(vp, "vfree");
3258 	mtx_lock(&vnode_free_list_mtx);
3259 	VNASSERT(vp->v_op != NULL, vp, ("vfree: vnode already reclaimed."));
3260 	VNASSERT((vp->v_iflag & VI_FREE) == 0, vp, ("vnode already free"));
3261 	VNASSERT(VSHOULDFREE(vp), vp, ("vfree: freeing when we shouldn't"));
3262 	VNASSERT((vp->v_iflag & VI_DOOMED) == 0, vp,
3263 	    ("vfree: Freeing doomed vnode"));
3264 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
3265 	if (vp->v_iflag & VI_AGE) {
3266 		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
3267 	} else {
3268 		TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
3269 	}
3270 	freevnodes++;
3271 	vp->v_iflag &= ~VI_AGE;
3272 	vp->v_iflag |= VI_FREE;
3273 	mtx_unlock(&vnode_free_list_mtx);
3274 }
3275 
3276 /*
3277  * Opposite of vfree() - mark a vnode as in use.
3278  */
3279 static void
3280 vbusy(struct vnode *vp)
3281 {
3282 	ASSERT_VI_LOCKED(vp, "vbusy");
3283 	VNASSERT((vp->v_iflag & VI_FREE) != 0, vp, ("vnode not free"));
3284 	VNASSERT(vp->v_op != NULL, vp, ("vbusy: vnode already reclaimed."));
3285 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
3286 
3287 	mtx_lock(&vnode_free_list_mtx);
3288 	TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
3289 	freevnodes--;
3290 	vp->v_iflag &= ~(VI_FREE|VI_AGE);
3291 	mtx_unlock(&vnode_free_list_mtx);
3292 }
3293 
3294 static void
3295 destroy_vpollinfo(struct vpollinfo *vi)
3296 {
3297 	knlist_destroy(&vi->vpi_selinfo.si_note);
3298 	mtx_destroy(&vi->vpi_lock);
3299 	uma_zfree(vnodepoll_zone, vi);
3300 }
3301 
3302 /*
3303  * Initalize per-vnode helper structure to hold poll-related state.
3304  */
3305 void
3306 v_addpollinfo(struct vnode *vp)
3307 {
3308 	struct vpollinfo *vi;
3309 
3310 	if (vp->v_pollinfo != NULL)
3311 		return;
3312 	vi = uma_zalloc(vnodepoll_zone, M_WAITOK);
3313 	mtx_init(&vi->vpi_lock, "vnode pollinfo", NULL, MTX_DEF);
3314 	knlist_init(&vi->vpi_selinfo.si_note, vp, vfs_knllock,
3315 	    vfs_knlunlock, vfs_knl_assert_locked, vfs_knl_assert_unlocked);
3316 	VI_LOCK(vp);
3317 	if (vp->v_pollinfo != NULL) {
3318 		VI_UNLOCK(vp);
3319 		destroy_vpollinfo(vi);
3320 		return;
3321 	}
3322 	vp->v_pollinfo = vi;
3323 	VI_UNLOCK(vp);
3324 }
3325 
3326 /*
3327  * Record a process's interest in events which might happen to
3328  * a vnode.  Because poll uses the historic select-style interface
3329  * internally, this routine serves as both the ``check for any
3330  * pending events'' and the ``record my interest in future events''
3331  * functions.  (These are done together, while the lock is held,
3332  * to avoid race conditions.)
3333  */
3334 int
3335 vn_pollrecord(struct vnode *vp, struct thread *td, int events)
3336 {
3337 
3338 	v_addpollinfo(vp);
3339 	mtx_lock(&vp->v_pollinfo->vpi_lock);
3340 	if (vp->v_pollinfo->vpi_revents & events) {
3341 		/*
3342 		 * This leaves events we are not interested
3343 		 * in available for the other process which
3344 		 * which presumably had requested them
3345 		 * (otherwise they would never have been
3346 		 * recorded).
3347 		 */
3348 		events &= vp->v_pollinfo->vpi_revents;
3349 		vp->v_pollinfo->vpi_revents &= ~events;
3350 
3351 		mtx_unlock(&vp->v_pollinfo->vpi_lock);
3352 		return (events);
3353 	}
3354 	vp->v_pollinfo->vpi_events |= events;
3355 	selrecord(td, &vp->v_pollinfo->vpi_selinfo);
3356 	mtx_unlock(&vp->v_pollinfo->vpi_lock);
3357 	return (0);
3358 }
3359 
3360 /*
3361  * Routine to create and manage a filesystem syncer vnode.
3362  */
3363 #define sync_close ((int (*)(struct  vop_close_args *))nullop)
3364 static int	sync_fsync(struct  vop_fsync_args *);
3365 static int	sync_inactive(struct  vop_inactive_args *);
3366 static int	sync_reclaim(struct  vop_reclaim_args *);
3367 
3368 static struct vop_vector sync_vnodeops = {
3369 	.vop_bypass =	VOP_EOPNOTSUPP,
3370 	.vop_close =	sync_close,		/* close */
3371 	.vop_fsync =	sync_fsync,		/* fsync */
3372 	.vop_inactive =	sync_inactive,	/* inactive */
3373 	.vop_reclaim =	sync_reclaim,	/* reclaim */
3374 	.vop_lock1 =	vop_stdlock,	/* lock */
3375 	.vop_unlock =	vop_stdunlock,	/* unlock */
3376 	.vop_islocked =	vop_stdislocked,	/* islocked */
3377 };
3378 
3379 /*
3380  * Create a new filesystem syncer vnode for the specified mount point.
3381  */
3382 void
3383 vfs_allocate_syncvnode(struct mount *mp)
3384 {
3385 	struct vnode *vp;
3386 	struct bufobj *bo;
3387 	static long start, incr, next;
3388 	int error;
3389 
3390 	/* Allocate a new vnode */
3391 	error = getnewvnode("syncer", mp, &sync_vnodeops, &vp);
3392 	if (error != 0)
3393 		panic("vfs_allocate_syncvnode: getnewvnode() failed");
3394 	vp->v_type = VNON;
3395 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3396 	vp->v_vflag |= VV_FORCEINSMQ;
3397 	error = insmntque(vp, mp);
3398 	if (error != 0)
3399 		panic("vfs_allocate_syncvnode: insmntque() failed");
3400 	vp->v_vflag &= ~VV_FORCEINSMQ;
3401 	VOP_UNLOCK(vp, 0);
3402 	/*
3403 	 * Place the vnode onto the syncer worklist. We attempt to
3404 	 * scatter them about on the list so that they will go off
3405 	 * at evenly distributed times even if all the filesystems
3406 	 * are mounted at once.
3407 	 */
3408 	next += incr;
3409 	if (next == 0 || next > syncer_maxdelay) {
3410 		start /= 2;
3411 		incr /= 2;
3412 		if (start == 0) {
3413 			start = syncer_maxdelay / 2;
3414 			incr = syncer_maxdelay;
3415 		}
3416 		next = start;
3417 	}
3418 	bo = &vp->v_bufobj;
3419 	BO_LOCK(bo);
3420 	vn_syncer_add_to_worklist(bo, syncdelay > 0 ? next % syncdelay : 0);
3421 	/* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */
3422 	mtx_lock(&sync_mtx);
3423 	sync_vnode_count++;
3424 	if (mp->mnt_syncer == NULL) {
3425 		mp->mnt_syncer = vp;
3426 		vp = NULL;
3427 	}
3428 	mtx_unlock(&sync_mtx);
3429 	BO_UNLOCK(bo);
3430 	if (vp != NULL) {
3431 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3432 		vgone(vp);
3433 		vput(vp);
3434 	}
3435 }
3436 
3437 void
3438 vfs_deallocate_syncvnode(struct mount *mp)
3439 {
3440 	struct vnode *vp;
3441 
3442 	mtx_lock(&sync_mtx);
3443 	vp = mp->mnt_syncer;
3444 	if (vp != NULL)
3445 		mp->mnt_syncer = NULL;
3446 	mtx_unlock(&sync_mtx);
3447 	if (vp != NULL)
3448 		vrele(vp);
3449 }
3450 
3451 /*
3452  * Do a lazy sync of the filesystem.
3453  */
3454 static int
3455 sync_fsync(struct vop_fsync_args *ap)
3456 {
3457 	struct vnode *syncvp = ap->a_vp;
3458 	struct mount *mp = syncvp->v_mount;
3459 	int error;
3460 	struct bufobj *bo;
3461 
3462 	/*
3463 	 * We only need to do something if this is a lazy evaluation.
3464 	 */
3465 	if (ap->a_waitfor != MNT_LAZY)
3466 		return (0);
3467 
3468 	/*
3469 	 * Move ourselves to the back of the sync list.
3470 	 */
3471 	bo = &syncvp->v_bufobj;
3472 	BO_LOCK(bo);
3473 	vn_syncer_add_to_worklist(bo, syncdelay);
3474 	BO_UNLOCK(bo);
3475 
3476 	/*
3477 	 * Walk the list of vnodes pushing all that are dirty and
3478 	 * not already on the sync list.
3479 	 */
3480 	mtx_lock(&mountlist_mtx);
3481 	if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK) != 0) {
3482 		mtx_unlock(&mountlist_mtx);
3483 		return (0);
3484 	}
3485 	if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
3486 		vfs_unbusy(mp);
3487 		return (0);
3488 	}
3489 	MNT_ILOCK(mp);
3490 	mp->mnt_noasync++;
3491 	mp->mnt_kern_flag &= ~MNTK_ASYNC;
3492 	MNT_IUNLOCK(mp);
3493 	vfs_msync(mp, MNT_NOWAIT);
3494 	error = VFS_SYNC(mp, MNT_LAZY);
3495 	MNT_ILOCK(mp);
3496 	mp->mnt_noasync--;
3497 	if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0)
3498 		mp->mnt_kern_flag |= MNTK_ASYNC;
3499 	MNT_IUNLOCK(mp);
3500 	vn_finished_write(mp);
3501 	vfs_unbusy(mp);
3502 	return (error);
3503 }
3504 
3505 /*
3506  * The syncer vnode is no referenced.
3507  */
3508 static int
3509 sync_inactive(struct vop_inactive_args *ap)
3510 {
3511 
3512 	vgone(ap->a_vp);
3513 	return (0);
3514 }
3515 
3516 /*
3517  * The syncer vnode is no longer needed and is being decommissioned.
3518  *
3519  * Modifications to the worklist must be protected by sync_mtx.
3520  */
3521 static int
3522 sync_reclaim(struct vop_reclaim_args *ap)
3523 {
3524 	struct vnode *vp = ap->a_vp;
3525 	struct bufobj *bo;
3526 
3527 	bo = &vp->v_bufobj;
3528 	BO_LOCK(bo);
3529 	mtx_lock(&sync_mtx);
3530 	if (vp->v_mount->mnt_syncer == vp)
3531 		vp->v_mount->mnt_syncer = NULL;
3532 	if (bo->bo_flag & BO_ONWORKLST) {
3533 		LIST_REMOVE(bo, bo_synclist);
3534 		syncer_worklist_len--;
3535 		sync_vnode_count--;
3536 		bo->bo_flag &= ~BO_ONWORKLST;
3537 	}
3538 	mtx_unlock(&sync_mtx);
3539 	BO_UNLOCK(bo);
3540 
3541 	return (0);
3542 }
3543 
3544 /*
3545  * Check if vnode represents a disk device
3546  */
3547 int
3548 vn_isdisk(struct vnode *vp, int *errp)
3549 {
3550 	int error;
3551 
3552 	error = 0;
3553 	dev_lock();
3554 	if (vp->v_type != VCHR)
3555 		error = ENOTBLK;
3556 	else if (vp->v_rdev == NULL)
3557 		error = ENXIO;
3558 	else if (vp->v_rdev->si_devsw == NULL)
3559 		error = ENXIO;
3560 	else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK))
3561 		error = ENOTBLK;
3562 	dev_unlock();
3563 	if (errp != NULL)
3564 		*errp = error;
3565 	return (error == 0);
3566 }
3567 
3568 /*
3569  * Common filesystem object access control check routine.  Accepts a
3570  * vnode's type, "mode", uid and gid, requested access mode, credentials,
3571  * and optional call-by-reference privused argument allowing vaccess()
3572  * to indicate to the caller whether privilege was used to satisfy the
3573  * request (obsoleted).  Returns 0 on success, or an errno on failure.
3574  *
3575  * The ifdef'd CAPABILITIES version is here for reference, but is not
3576  * actually used.
3577  */
3578 int
3579 vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid,
3580     accmode_t accmode, struct ucred *cred, int *privused)
3581 {
3582 	accmode_t dac_granted;
3583 	accmode_t priv_granted;
3584 
3585 	KASSERT((accmode & ~(VEXEC | VWRITE | VREAD | VADMIN | VAPPEND)) == 0,
3586 	    ("invalid bit in accmode"));
3587 	KASSERT((accmode & VAPPEND) == 0 || (accmode & VWRITE),
3588 	    ("VAPPEND without VWRITE"));
3589 
3590 	/*
3591 	 * Look for a normal, non-privileged way to access the file/directory
3592 	 * as requested.  If it exists, go with that.
3593 	 */
3594 
3595 	if (privused != NULL)
3596 		*privused = 0;
3597 
3598 	dac_granted = 0;
3599 
3600 	/* Check the owner. */
3601 	if (cred->cr_uid == file_uid) {
3602 		dac_granted |= VADMIN;
3603 		if (file_mode & S_IXUSR)
3604 			dac_granted |= VEXEC;
3605 		if (file_mode & S_IRUSR)
3606 			dac_granted |= VREAD;
3607 		if (file_mode & S_IWUSR)
3608 			dac_granted |= (VWRITE | VAPPEND);
3609 
3610 		if ((accmode & dac_granted) == accmode)
3611 			return (0);
3612 
3613 		goto privcheck;
3614 	}
3615 
3616 	/* Otherwise, check the groups (first match) */
3617 	if (groupmember(file_gid, cred)) {
3618 		if (file_mode & S_IXGRP)
3619 			dac_granted |= VEXEC;
3620 		if (file_mode & S_IRGRP)
3621 			dac_granted |= VREAD;
3622 		if (file_mode & S_IWGRP)
3623 			dac_granted |= (VWRITE | VAPPEND);
3624 
3625 		if ((accmode & dac_granted) == accmode)
3626 			return (0);
3627 
3628 		goto privcheck;
3629 	}
3630 
3631 	/* Otherwise, check everyone else. */
3632 	if (file_mode & S_IXOTH)
3633 		dac_granted |= VEXEC;
3634 	if (file_mode & S_IROTH)
3635 		dac_granted |= VREAD;
3636 	if (file_mode & S_IWOTH)
3637 		dac_granted |= (VWRITE | VAPPEND);
3638 	if ((accmode & dac_granted) == accmode)
3639 		return (0);
3640 
3641 privcheck:
3642 	/*
3643 	 * Build a privilege mask to determine if the set of privileges
3644 	 * satisfies the requirements when combined with the granted mask
3645 	 * from above.  For each privilege, if the privilege is required,
3646 	 * bitwise or the request type onto the priv_granted mask.
3647 	 */
3648 	priv_granted = 0;
3649 
3650 	if (type == VDIR) {
3651 		/*
3652 		 * For directories, use PRIV_VFS_LOOKUP to satisfy VEXEC
3653 		 * requests, instead of PRIV_VFS_EXEC.
3654 		 */
3655 		if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3656 		    !priv_check_cred(cred, PRIV_VFS_LOOKUP, 0))
3657 			priv_granted |= VEXEC;
3658 	} else {
3659 		/*
3660 		 * Ensure that at least one execute bit is on. Otherwise,
3661 		 * a privileged user will always succeed, and we don't want
3662 		 * this to happen unless the file really is executable.
3663 		 */
3664 		if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3665 		    (file_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) != 0 &&
3666 		    !priv_check_cred(cred, PRIV_VFS_EXEC, 0))
3667 			priv_granted |= VEXEC;
3668 	}
3669 
3670 	if ((accmode & VREAD) && ((dac_granted & VREAD) == 0) &&
3671 	    !priv_check_cred(cred, PRIV_VFS_READ, 0))
3672 		priv_granted |= VREAD;
3673 
3674 	if ((accmode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
3675 	    !priv_check_cred(cred, PRIV_VFS_WRITE, 0))
3676 		priv_granted |= (VWRITE | VAPPEND);
3677 
3678 	if ((accmode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
3679 	    !priv_check_cred(cred, PRIV_VFS_ADMIN, 0))
3680 		priv_granted |= VADMIN;
3681 
3682 	if ((accmode & (priv_granted | dac_granted)) == accmode) {
3683 		/* XXX audit: privilege used */
3684 		if (privused != NULL)
3685 			*privused = 1;
3686 		return (0);
3687 	}
3688 
3689 	return ((accmode & VADMIN) ? EPERM : EACCES);
3690 }
3691 
3692 /*
3693  * Credential check based on process requesting service, and per-attribute
3694  * permissions.
3695  */
3696 int
3697 extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred,
3698     struct thread *td, accmode_t accmode)
3699 {
3700 
3701 	/*
3702 	 * Kernel-invoked always succeeds.
3703 	 */
3704 	if (cred == NOCRED)
3705 		return (0);
3706 
3707 	/*
3708 	 * Do not allow privileged processes in jail to directly manipulate
3709 	 * system attributes.
3710 	 */
3711 	switch (attrnamespace) {
3712 	case EXTATTR_NAMESPACE_SYSTEM:
3713 		/* Potentially should be: return (EPERM); */
3714 		return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM, 0));
3715 	case EXTATTR_NAMESPACE_USER:
3716 		return (VOP_ACCESS(vp, accmode, cred, td));
3717 	default:
3718 		return (EPERM);
3719 	}
3720 }
3721 
3722 #ifdef DEBUG_VFS_LOCKS
3723 /*
3724  * This only exists to supress warnings from unlocked specfs accesses.  It is
3725  * no longer ok to have an unlocked VFS.
3726  */
3727 #define	IGNORE_LOCK(vp) (panicstr != NULL || (vp) == NULL ||		\
3728 	(vp)->v_type == VCHR ||	(vp)->v_type == VBAD)
3729 
3730 int vfs_badlock_ddb = 1;	/* Drop into debugger on violation. */
3731 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0,
3732     "Drop into debugger on lock violation");
3733 
3734 int vfs_badlock_mutex = 1;	/* Check for interlock across VOPs. */
3735 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex,
3736     0, "Check for interlock across VOPs");
3737 
3738 int vfs_badlock_print = 1;	/* Print lock violations. */
3739 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print,
3740     0, "Print lock violations");
3741 
3742 #ifdef KDB
3743 int vfs_badlock_backtrace = 1;	/* Print backtrace at lock violations. */
3744 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW,
3745     &vfs_badlock_backtrace, 0, "Print backtrace at lock violations");
3746 #endif
3747 
3748 static void
3749 vfs_badlock(const char *msg, const char *str, struct vnode *vp)
3750 {
3751 
3752 #ifdef KDB
3753 	if (vfs_badlock_backtrace)
3754 		kdb_backtrace();
3755 #endif
3756 	if (vfs_badlock_print)
3757 		printf("%s: %p %s\n", str, (void *)vp, msg);
3758 	if (vfs_badlock_ddb)
3759 		kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
3760 }
3761 
3762 void
3763 assert_vi_locked(struct vnode *vp, const char *str)
3764 {
3765 
3766 	if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp)))
3767 		vfs_badlock("interlock is not locked but should be", str, vp);
3768 }
3769 
3770 void
3771 assert_vi_unlocked(struct vnode *vp, const char *str)
3772 {
3773 
3774 	if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp)))
3775 		vfs_badlock("interlock is locked but should not be", str, vp);
3776 }
3777 
3778 void
3779 assert_vop_locked(struct vnode *vp, const char *str)
3780 {
3781 
3782 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == 0)
3783 		vfs_badlock("is not locked but should be", str, vp);
3784 }
3785 
3786 void
3787 assert_vop_unlocked(struct vnode *vp, const char *str)
3788 {
3789 
3790 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == LK_EXCLUSIVE)
3791 		vfs_badlock("is locked but should not be", str, vp);
3792 }
3793 
3794 void
3795 assert_vop_elocked(struct vnode *vp, const char *str)
3796 {
3797 
3798 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
3799 		vfs_badlock("is not exclusive locked but should be", str, vp);
3800 }
3801 
3802 #if 0
3803 void
3804 assert_vop_elocked_other(struct vnode *vp, const char *str)
3805 {
3806 
3807 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLOTHER)
3808 		vfs_badlock("is not exclusive locked by another thread",
3809 		    str, vp);
3810 }
3811 
3812 void
3813 assert_vop_slocked(struct vnode *vp, const char *str)
3814 {
3815 
3816 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_SHARED)
3817 		vfs_badlock("is not locked shared but should be", str, vp);
3818 }
3819 #endif /* 0 */
3820 #endif /* DEBUG_VFS_LOCKS */
3821 
3822 void
3823 vop_rename_fail(struct vop_rename_args *ap)
3824 {
3825 
3826 	if (ap->a_tvp != NULL)
3827 		vput(ap->a_tvp);
3828 	if (ap->a_tdvp == ap->a_tvp)
3829 		vrele(ap->a_tdvp);
3830 	else
3831 		vput(ap->a_tdvp);
3832 	vrele(ap->a_fdvp);
3833 	vrele(ap->a_fvp);
3834 }
3835 
3836 void
3837 vop_rename_pre(void *ap)
3838 {
3839 	struct vop_rename_args *a = ap;
3840 
3841 #ifdef DEBUG_VFS_LOCKS
3842 	if (a->a_tvp)
3843 		ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME");
3844 	ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME");
3845 	ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME");
3846 	ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME");
3847 
3848 	/* Check the source (from). */
3849 	if (a->a_tdvp->v_vnlock != a->a_fdvp->v_vnlock &&
3850 	    (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fdvp->v_vnlock))
3851 		ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked");
3852 	if (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fvp->v_vnlock)
3853 		ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: fvp locked");
3854 
3855 	/* Check the target. */
3856 	if (a->a_tvp)
3857 		ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked");
3858 	ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked");
3859 #endif
3860 	if (a->a_tdvp != a->a_fdvp)
3861 		vhold(a->a_fdvp);
3862 	if (a->a_tvp != a->a_fvp)
3863 		vhold(a->a_fvp);
3864 	vhold(a->a_tdvp);
3865 	if (a->a_tvp)
3866 		vhold(a->a_tvp);
3867 }
3868 
3869 void
3870 vop_strategy_pre(void *ap)
3871 {
3872 #ifdef DEBUG_VFS_LOCKS
3873 	struct vop_strategy_args *a;
3874 	struct buf *bp;
3875 
3876 	a = ap;
3877 	bp = a->a_bp;
3878 
3879 	/*
3880 	 * Cluster ops lock their component buffers but not the IO container.
3881 	 */
3882 	if ((bp->b_flags & B_CLUSTER) != 0)
3883 		return;
3884 
3885 	if (!BUF_ISLOCKED(bp)) {
3886 		if (vfs_badlock_print)
3887 			printf(
3888 			    "VOP_STRATEGY: bp is not locked but should be\n");
3889 		if (vfs_badlock_ddb)
3890 			kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
3891 	}
3892 #endif
3893 }
3894 
3895 void
3896 vop_lookup_pre(void *ap)
3897 {
3898 #ifdef DEBUG_VFS_LOCKS
3899 	struct vop_lookup_args *a;
3900 	struct vnode *dvp;
3901 
3902 	a = ap;
3903 	dvp = a->a_dvp;
3904 	ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
3905 	ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
3906 #endif
3907 }
3908 
3909 void
3910 vop_lookup_post(void *ap, int rc)
3911 {
3912 #ifdef DEBUG_VFS_LOCKS
3913 	struct vop_lookup_args *a;
3914 	struct vnode *dvp;
3915 	struct vnode *vp;
3916 
3917 	a = ap;
3918 	dvp = a->a_dvp;
3919 	vp = *(a->a_vpp);
3920 
3921 	ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
3922 	ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
3923 
3924 	if (!rc)
3925 		ASSERT_VOP_LOCKED(vp, "VOP_LOOKUP (child)");
3926 #endif
3927 }
3928 
3929 void
3930 vop_lock_pre(void *ap)
3931 {
3932 #ifdef DEBUG_VFS_LOCKS
3933 	struct vop_lock1_args *a = ap;
3934 
3935 	if ((a->a_flags & LK_INTERLOCK) == 0)
3936 		ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
3937 	else
3938 		ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK");
3939 #endif
3940 }
3941 
3942 void
3943 vop_lock_post(void *ap, int rc)
3944 {
3945 #ifdef DEBUG_VFS_LOCKS
3946 	struct vop_lock1_args *a = ap;
3947 
3948 	ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
3949 	if (rc == 0)
3950 		ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK");
3951 #endif
3952 }
3953 
3954 void
3955 vop_unlock_pre(void *ap)
3956 {
3957 #ifdef DEBUG_VFS_LOCKS
3958 	struct vop_unlock_args *a = ap;
3959 
3960 	if (a->a_flags & LK_INTERLOCK)
3961 		ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK");
3962 	ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK");
3963 #endif
3964 }
3965 
3966 void
3967 vop_unlock_post(void *ap, int rc)
3968 {
3969 #ifdef DEBUG_VFS_LOCKS
3970 	struct vop_unlock_args *a = ap;
3971 
3972 	if (a->a_flags & LK_INTERLOCK)
3973 		ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK");
3974 #endif
3975 }
3976 
3977 void
3978 vop_create_post(void *ap, int rc)
3979 {
3980 	struct vop_create_args *a = ap;
3981 
3982 	if (!rc)
3983 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3984 }
3985 
3986 void
3987 vop_link_post(void *ap, int rc)
3988 {
3989 	struct vop_link_args *a = ap;
3990 
3991 	if (!rc) {
3992 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_LINK);
3993 		VFS_KNOTE_LOCKED(a->a_tdvp, NOTE_WRITE);
3994 	}
3995 }
3996 
3997 void
3998 vop_mkdir_post(void *ap, int rc)
3999 {
4000 	struct vop_mkdir_args *a = ap;
4001 
4002 	if (!rc)
4003 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
4004 }
4005 
4006 void
4007 vop_mknod_post(void *ap, int rc)
4008 {
4009 	struct vop_mknod_args *a = ap;
4010 
4011 	if (!rc)
4012 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4013 }
4014 
4015 void
4016 vop_remove_post(void *ap, int rc)
4017 {
4018 	struct vop_remove_args *a = ap;
4019 
4020 	if (!rc) {
4021 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4022 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
4023 	}
4024 }
4025 
4026 void
4027 vop_rename_post(void *ap, int rc)
4028 {
4029 	struct vop_rename_args *a = ap;
4030 
4031 	if (!rc) {
4032 		VFS_KNOTE_UNLOCKED(a->a_fdvp, NOTE_WRITE);
4033 		VFS_KNOTE_UNLOCKED(a->a_tdvp, NOTE_WRITE);
4034 		VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME);
4035 		if (a->a_tvp)
4036 			VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE);
4037 	}
4038 	if (a->a_tdvp != a->a_fdvp)
4039 		vdrop(a->a_fdvp);
4040 	if (a->a_tvp != a->a_fvp)
4041 		vdrop(a->a_fvp);
4042 	vdrop(a->a_tdvp);
4043 	if (a->a_tvp)
4044 		vdrop(a->a_tvp);
4045 }
4046 
4047 void
4048 vop_rmdir_post(void *ap, int rc)
4049 {
4050 	struct vop_rmdir_args *a = ap;
4051 
4052 	if (!rc) {
4053 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
4054 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
4055 	}
4056 }
4057 
4058 void
4059 vop_setattr_post(void *ap, int rc)
4060 {
4061 	struct vop_setattr_args *a = ap;
4062 
4063 	if (!rc)
4064 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4065 }
4066 
4067 void
4068 vop_symlink_post(void *ap, int rc)
4069 {
4070 	struct vop_symlink_args *a = ap;
4071 
4072 	if (!rc)
4073 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4074 }
4075 
4076 static struct knlist fs_knlist;
4077 
4078 static void
4079 vfs_event_init(void *arg)
4080 {
4081 	knlist_init_mtx(&fs_knlist, NULL);
4082 }
4083 /* XXX - correct order? */
4084 SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL);
4085 
4086 void
4087 vfs_event_signal(fsid_t *fsid, uint32_t event, intptr_t data __unused)
4088 {
4089 
4090 	KNOTE_UNLOCKED(&fs_knlist, event);
4091 }
4092 
4093 static int	filt_fsattach(struct knote *kn);
4094 static void	filt_fsdetach(struct knote *kn);
4095 static int	filt_fsevent(struct knote *kn, long hint);
4096 
4097 struct filterops fs_filtops = {
4098 	.f_isfd = 0,
4099 	.f_attach = filt_fsattach,
4100 	.f_detach = filt_fsdetach,
4101 	.f_event = filt_fsevent
4102 };
4103 
4104 static int
4105 filt_fsattach(struct knote *kn)
4106 {
4107 
4108 	kn->kn_flags |= EV_CLEAR;
4109 	knlist_add(&fs_knlist, kn, 0);
4110 	return (0);
4111 }
4112 
4113 static void
4114 filt_fsdetach(struct knote *kn)
4115 {
4116 
4117 	knlist_remove(&fs_knlist, kn, 0);
4118 }
4119 
4120 static int
4121 filt_fsevent(struct knote *kn, long hint)
4122 {
4123 
4124 	kn->kn_fflags |= hint;
4125 	return (kn->kn_fflags != 0);
4126 }
4127 
4128 static int
4129 sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS)
4130 {
4131 	struct vfsidctl vc;
4132 	int error;
4133 	struct mount *mp;
4134 
4135 	error = SYSCTL_IN(req, &vc, sizeof(vc));
4136 	if (error)
4137 		return (error);
4138 	if (vc.vc_vers != VFS_CTL_VERS1)
4139 		return (EINVAL);
4140 	mp = vfs_getvfs(&vc.vc_fsid);
4141 	if (mp == NULL)
4142 		return (ENOENT);
4143 	/* ensure that a specific sysctl goes to the right filesystem. */
4144 	if (strcmp(vc.vc_fstypename, "*") != 0 &&
4145 	    strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) {
4146 		vfs_rel(mp);
4147 		return (EINVAL);
4148 	}
4149 	VCTLTOREQ(&vc, req);
4150 	error = VFS_SYSCTL(mp, vc.vc_op, req);
4151 	vfs_rel(mp);
4152 	return (error);
4153 }
4154 
4155 SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLFLAG_WR, NULL, 0, sysctl_vfs_ctl, "",
4156     "Sysctl by fsid");
4157 
4158 /*
4159  * Function to initialize a va_filerev field sensibly.
4160  * XXX: Wouldn't a random number make a lot more sense ??
4161  */
4162 u_quad_t
4163 init_va_filerev(void)
4164 {
4165 	struct bintime bt;
4166 
4167 	getbinuptime(&bt);
4168 	return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL));
4169 }
4170 
4171 static int	filt_vfsread(struct knote *kn, long hint);
4172 static int	filt_vfswrite(struct knote *kn, long hint);
4173 static int	filt_vfsvnode(struct knote *kn, long hint);
4174 static void	filt_vfsdetach(struct knote *kn);
4175 static struct filterops vfsread_filtops = {
4176 	.f_isfd = 1,
4177 	.f_detach = filt_vfsdetach,
4178 	.f_event = filt_vfsread
4179 };
4180 static struct filterops vfswrite_filtops = {
4181 	.f_isfd = 1,
4182 	.f_detach = filt_vfsdetach,
4183 	.f_event = filt_vfswrite
4184 };
4185 static struct filterops vfsvnode_filtops = {
4186 	.f_isfd = 1,
4187 	.f_detach = filt_vfsdetach,
4188 	.f_event = filt_vfsvnode
4189 };
4190 
4191 static void
4192 vfs_knllock(void *arg)
4193 {
4194 	struct vnode *vp = arg;
4195 
4196 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4197 }
4198 
4199 static void
4200 vfs_knlunlock(void *arg)
4201 {
4202 	struct vnode *vp = arg;
4203 
4204 	VOP_UNLOCK(vp, 0);
4205 }
4206 
4207 static void
4208 vfs_knl_assert_locked(void *arg)
4209 {
4210 #ifdef DEBUG_VFS_LOCKS
4211 	struct vnode *vp = arg;
4212 
4213 	ASSERT_VOP_LOCKED(vp, "vfs_knl_assert_locked");
4214 #endif
4215 }
4216 
4217 static void
4218 vfs_knl_assert_unlocked(void *arg)
4219 {
4220 #ifdef DEBUG_VFS_LOCKS
4221 	struct vnode *vp = arg;
4222 
4223 	ASSERT_VOP_UNLOCKED(vp, "vfs_knl_assert_unlocked");
4224 #endif
4225 }
4226 
4227 int
4228 vfs_kqfilter(struct vop_kqfilter_args *ap)
4229 {
4230 	struct vnode *vp = ap->a_vp;
4231 	struct knote *kn = ap->a_kn;
4232 	struct knlist *knl;
4233 
4234 	switch (kn->kn_filter) {
4235 	case EVFILT_READ:
4236 		kn->kn_fop = &vfsread_filtops;
4237 		break;
4238 	case EVFILT_WRITE:
4239 		kn->kn_fop = &vfswrite_filtops;
4240 		break;
4241 	case EVFILT_VNODE:
4242 		kn->kn_fop = &vfsvnode_filtops;
4243 		break;
4244 	default:
4245 		return (EINVAL);
4246 	}
4247 
4248 	kn->kn_hook = (caddr_t)vp;
4249 
4250 	v_addpollinfo(vp);
4251 	if (vp->v_pollinfo == NULL)
4252 		return (ENOMEM);
4253 	knl = &vp->v_pollinfo->vpi_selinfo.si_note;
4254 	knlist_add(knl, kn, 0);
4255 
4256 	return (0);
4257 }
4258 
4259 /*
4260  * Detach knote from vnode
4261  */
4262 static void
4263 filt_vfsdetach(struct knote *kn)
4264 {
4265 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4266 
4267 	KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo"));
4268 	knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0);
4269 }
4270 
4271 /*ARGSUSED*/
4272 static int
4273 filt_vfsread(struct knote *kn, long hint)
4274 {
4275 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4276 	struct vattr va;
4277 	int res;
4278 
4279 	/*
4280 	 * filesystem is gone, so set the EOF flag and schedule
4281 	 * the knote for deletion.
4282 	 */
4283 	if (hint == NOTE_REVOKE) {
4284 		VI_LOCK(vp);
4285 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4286 		VI_UNLOCK(vp);
4287 		return (1);
4288 	}
4289 
4290 	if (VOP_GETATTR(vp, &va, curthread->td_ucred))
4291 		return (0);
4292 
4293 	VI_LOCK(vp);
4294 	kn->kn_data = va.va_size - kn->kn_fp->f_offset;
4295 	res = (kn->kn_data != 0);
4296 	VI_UNLOCK(vp);
4297 	return (res);
4298 }
4299 
4300 /*ARGSUSED*/
4301 static int
4302 filt_vfswrite(struct knote *kn, long hint)
4303 {
4304 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4305 
4306 	VI_LOCK(vp);
4307 
4308 	/*
4309 	 * filesystem is gone, so set the EOF flag and schedule
4310 	 * the knote for deletion.
4311 	 */
4312 	if (hint == NOTE_REVOKE)
4313 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4314 
4315 	kn->kn_data = 0;
4316 	VI_UNLOCK(vp);
4317 	return (1);
4318 }
4319 
4320 static int
4321 filt_vfsvnode(struct knote *kn, long hint)
4322 {
4323 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4324 	int res;
4325 
4326 	VI_LOCK(vp);
4327 	if (kn->kn_sfflags & hint)
4328 		kn->kn_fflags |= hint;
4329 	if (hint == NOTE_REVOKE) {
4330 		kn->kn_flags |= EV_EOF;
4331 		VI_UNLOCK(vp);
4332 		return (1);
4333 	}
4334 	res = (kn->kn_fflags != 0);
4335 	VI_UNLOCK(vp);
4336 	return (res);
4337 }
4338 
4339 int
4340 vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off)
4341 {
4342 	int error;
4343 
4344 	if (dp->d_reclen > ap->a_uio->uio_resid)
4345 		return (ENAMETOOLONG);
4346 	error = uiomove(dp, dp->d_reclen, ap->a_uio);
4347 	if (error) {
4348 		if (ap->a_ncookies != NULL) {
4349 			if (ap->a_cookies != NULL)
4350 				free(ap->a_cookies, M_TEMP);
4351 			ap->a_cookies = NULL;
4352 			*ap->a_ncookies = 0;
4353 		}
4354 		return (error);
4355 	}
4356 	if (ap->a_ncookies == NULL)
4357 		return (0);
4358 
4359 	KASSERT(ap->a_cookies,
4360 	    ("NULL ap->a_cookies value with non-NULL ap->a_ncookies!"));
4361 
4362 	*ap->a_cookies = realloc(*ap->a_cookies,
4363 	    (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO);
4364 	(*ap->a_cookies)[*ap->a_ncookies] = off;
4365 	return (0);
4366 }
4367 
4368 /*
4369  * Mark for update the access time of the file if the filesystem
4370  * supports VOP_MARKATIME.  This functionality is used by execve and
4371  * mmap, so we want to avoid the I/O implied by directly setting
4372  * va_atime for the sake of efficiency.
4373  */
4374 void
4375 vfs_mark_atime(struct vnode *vp, struct ucred *cred)
4376 {
4377 	struct mount *mp;
4378 
4379 	mp = vp->v_mount;
4380 	VFS_ASSERT_GIANT(mp);
4381 	ASSERT_VOP_LOCKED(vp, "vfs_mark_atime");
4382 	if (mp != NULL && (mp->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
4383 		(void)VOP_MARKATIME(vp);
4384 }
4385 
4386 /*
4387  * The purpose of this routine is to remove granularity from accmode_t,
4388  * reducing it into standard unix access bits - VEXEC, VREAD, VWRITE,
4389  * VADMIN and VAPPEND.
4390  *
4391  * If it returns 0, the caller is supposed to continue with the usual
4392  * access checks using 'accmode' as modified by this routine.  If it
4393  * returns nonzero value, the caller is supposed to return that value
4394  * as errno.
4395  *
4396  * Note that after this routine runs, accmode may be zero.
4397  */
4398 int
4399 vfs_unixify_accmode(accmode_t *accmode)
4400 {
4401 	/*
4402 	 * There is no way to specify explicit "deny" rule using
4403 	 * file mode or POSIX.1e ACLs.
4404 	 */
4405 	if (*accmode & VEXPLICIT_DENY) {
4406 		*accmode = 0;
4407 		return (0);
4408 	}
4409 
4410 	/*
4411 	 * None of these can be translated into usual access bits.
4412 	 * Also, the common case for NFSv4 ACLs is to not contain
4413 	 * either of these bits. Caller should check for VWRITE
4414 	 * on the containing directory instead.
4415 	 */
4416 	if (*accmode & (VDELETE_CHILD | VDELETE))
4417 		return (EPERM);
4418 
4419 	if (*accmode & VADMIN_PERMS) {
4420 		*accmode &= ~VADMIN_PERMS;
4421 		*accmode |= VADMIN;
4422 	}
4423 
4424 	/*
4425 	 * There is no way to deny VREAD_ATTRIBUTES, VREAD_ACL
4426 	 * or VSYNCHRONIZE using file mode or POSIX.1e ACL.
4427 	 */
4428 	*accmode &= ~(VSTAT_PERMS | VSYNCHRONIZE);
4429 
4430 	return (0);
4431 }
4432