xref: /netbsd-src/sys/kern/vfs_subr.c (revision 404ee5b9334f618040b6cdef96a0ff35a6fc4636)
1 /*	$NetBSD: vfs_subr.c,v 1.474 2019/11/16 10:05:44 maxv Exp $	*/
2 
3 /*-
4  * Copyright (c) 1997, 1998, 2004, 2005, 2007, 2008 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility,
9  * NASA Ames Research Center, by Charles M. Hannum, by Andrew Doran,
10  * by Marshall Kirk McKusick and Greg Ganger at the University of Michigan.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
22  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
23  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
24  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
25  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
26  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
27  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
28  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
29  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
30  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
31  * POSSIBILITY OF SUCH DAMAGE.
32  */
33 
34 /*
35  * Copyright (c) 1989, 1993
36  *	The Regents of the University of California.  All rights reserved.
37  * (c) UNIX System Laboratories, Inc.
38  * All or some portions of this file are derived from material licensed
39  * to the University of California by American Telephone and Telegraph
40  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
41  * the permission of UNIX System Laboratories, Inc.
42  *
43  * Redistribution and use in source and binary forms, with or without
44  * modification, are permitted provided that the following conditions
45  * are met:
46  * 1. Redistributions of source code must retain the above copyright
47  *    notice, this list of conditions and the following disclaimer.
48  * 2. Redistributions in binary form must reproduce the above copyright
49  *    notice, this list of conditions and the following disclaimer in the
50  *    documentation and/or other materials provided with the distribution.
51  * 3. Neither the name of the University nor the names of its contributors
52  *    may be used to endorse or promote products derived from this software
53  *    without specific prior written permission.
54  *
55  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
56  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
57  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
58  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
59  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
60  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
61  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
62  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
63  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
64  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
65  * SUCH DAMAGE.
66  *
67  *	@(#)vfs_subr.c	8.13 (Berkeley) 4/18/94
68  */
69 
70 #include <sys/cdefs.h>
71 __KERNEL_RCSID(0, "$NetBSD: vfs_subr.c,v 1.474 2019/11/16 10:05:44 maxv Exp $");
72 
73 #ifdef _KERNEL_OPT
74 #include "opt_ddb.h"
75 #include "opt_compat_netbsd.h"
76 #include "opt_compat_43.h"
77 #endif
78 
79 #include <sys/param.h>
80 #include <sys/systm.h>
81 #include <sys/conf.h>
82 #include <sys/dirent.h>
83 #include <sys/filedesc.h>
84 #include <sys/kernel.h>
85 #include <sys/mount.h>
86 #include <sys/fstrans.h>
87 #include <sys/vnode_impl.h>
88 #include <sys/stat.h>
89 #include <sys/sysctl.h>
90 #include <sys/namei.h>
91 #include <sys/buf.h>
92 #include <sys/errno.h>
93 #include <sys/kmem.h>
94 #include <sys/syscallargs.h>
95 #include <sys/kauth.h>
96 #include <sys/module.h>
97 
98 #include <miscfs/genfs/genfs.h>
99 #include <miscfs/specfs/specdev.h>
100 #include <uvm/uvm_ddb.h>
101 
102 const enum vtype iftovt_tab[16] = {
103 	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
104 	VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
105 };
106 const int	vttoif_tab[9] = {
107 	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
108 	S_IFSOCK, S_IFIFO, S_IFMT,
109 };
110 
111 /*
112  * Insq/Remq for the vnode usage lists.
113  */
114 #define	bufinsvn(bp, dp)	LIST_INSERT_HEAD(dp, bp, b_vnbufs)
115 #define	bufremvn(bp) {							\
116 	LIST_REMOVE(bp, b_vnbufs);					\
117 	(bp)->b_vnbufs.le_next = NOLIST;				\
118 }
119 
120 int doforce = 1;		/* 1 => permit forcible unmounting */
121 
122 extern struct mount *dead_rootmount;
123 
124 /*
125  * Local declarations.
126  */
127 
128 static void vn_initialize_syncerd(void);
129 
130 /*
131  * Initialize the vnode management data structures.
132  */
133 void
134 vntblinit(void)
135 {
136 
137 	vn_initialize_syncerd();
138 	vfs_mount_sysinit();
139 	vfs_vnode_sysinit();
140 }
141 
142 /*
143  * Flush out and invalidate all buffers associated with a vnode.
144  * Called with the underlying vnode locked, which should prevent new dirty
145  * buffers from being queued.
146  */
147 int
148 vinvalbuf(struct vnode *vp, int flags, kauth_cred_t cred, struct lwp *l,
149 	  bool catch_p, int slptimeo)
150 {
151 	struct buf *bp, *nbp;
152 	int error;
153 	int flushflags = PGO_ALLPAGES | PGO_FREE | PGO_SYNCIO |
154 	    (flags & V_SAVE ? PGO_CLEANIT | PGO_RECLAIM : 0);
155 
156 	/* XXXUBC this doesn't look at flags or slp* */
157 	mutex_enter(vp->v_interlock);
158 	error = VOP_PUTPAGES(vp, 0, 0, flushflags);
159 	if (error) {
160 		return error;
161 	}
162 
163 	if (flags & V_SAVE) {
164 		error = VOP_FSYNC(vp, cred, FSYNC_WAIT|FSYNC_RECLAIM, 0, 0);
165 		if (error)
166 		        return (error);
167 		KASSERT(LIST_EMPTY(&vp->v_dirtyblkhd));
168 	}
169 
170 	mutex_enter(&bufcache_lock);
171 restart:
172 	for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
173 		KASSERT(bp->b_vp == vp);
174 		nbp = LIST_NEXT(bp, b_vnbufs);
175 		error = bbusy(bp, catch_p, slptimeo, NULL);
176 		if (error != 0) {
177 			if (error == EPASSTHROUGH)
178 				goto restart;
179 			mutex_exit(&bufcache_lock);
180 			return (error);
181 		}
182 		brelsel(bp, BC_INVAL | BC_VFLUSH);
183 	}
184 
185 	for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
186 		KASSERT(bp->b_vp == vp);
187 		nbp = LIST_NEXT(bp, b_vnbufs);
188 		error = bbusy(bp, catch_p, slptimeo, NULL);
189 		if (error != 0) {
190 			if (error == EPASSTHROUGH)
191 				goto restart;
192 			mutex_exit(&bufcache_lock);
193 			return (error);
194 		}
195 		/*
196 		 * XXX Since there are no node locks for NFS, I believe
197 		 * there is a slight chance that a delayed write will
198 		 * occur while sleeping just above, so check for it.
199 		 */
200 		if ((bp->b_oflags & BO_DELWRI) && (flags & V_SAVE)) {
201 #ifdef DEBUG
202 			printf("buffer still DELWRI\n");
203 #endif
204 			bp->b_cflags |= BC_BUSY | BC_VFLUSH;
205 			mutex_exit(&bufcache_lock);
206 			VOP_BWRITE(bp->b_vp, bp);
207 			mutex_enter(&bufcache_lock);
208 			goto restart;
209 		}
210 		brelsel(bp, BC_INVAL | BC_VFLUSH);
211 	}
212 
213 #ifdef DIAGNOSTIC
214 	if (!LIST_EMPTY(&vp->v_cleanblkhd) || !LIST_EMPTY(&vp->v_dirtyblkhd))
215 		panic("vinvalbuf: flush failed, vp %p", vp);
216 #endif
217 
218 	mutex_exit(&bufcache_lock);
219 
220 	return (0);
221 }
222 
223 /*
224  * Destroy any in core blocks past the truncation length.
225  * Called with the underlying vnode locked, which should prevent new dirty
226  * buffers from being queued.
227  */
228 int
229 vtruncbuf(struct vnode *vp, daddr_t lbn, bool catch_p, int slptimeo)
230 {
231 	struct buf *bp, *nbp;
232 	int error;
233 	voff_t off;
234 
235 	off = round_page((voff_t)lbn << vp->v_mount->mnt_fs_bshift);
236 	mutex_enter(vp->v_interlock);
237 	error = VOP_PUTPAGES(vp, off, 0, PGO_FREE | PGO_SYNCIO);
238 	if (error) {
239 		return error;
240 	}
241 
242 	mutex_enter(&bufcache_lock);
243 restart:
244 	for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
245 		KASSERT(bp->b_vp == vp);
246 		nbp = LIST_NEXT(bp, b_vnbufs);
247 		if (bp->b_lblkno < lbn)
248 			continue;
249 		error = bbusy(bp, catch_p, slptimeo, NULL);
250 		if (error != 0) {
251 			if (error == EPASSTHROUGH)
252 				goto restart;
253 			mutex_exit(&bufcache_lock);
254 			return (error);
255 		}
256 		brelsel(bp, BC_INVAL | BC_VFLUSH);
257 	}
258 
259 	for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
260 		KASSERT(bp->b_vp == vp);
261 		nbp = LIST_NEXT(bp, b_vnbufs);
262 		if (bp->b_lblkno < lbn)
263 			continue;
264 		error = bbusy(bp, catch_p, slptimeo, NULL);
265 		if (error != 0) {
266 			if (error == EPASSTHROUGH)
267 				goto restart;
268 			mutex_exit(&bufcache_lock);
269 			return (error);
270 		}
271 		brelsel(bp, BC_INVAL | BC_VFLUSH);
272 	}
273 	mutex_exit(&bufcache_lock);
274 
275 	return (0);
276 }
277 
278 /*
279  * Flush all dirty buffers from a vnode.
280  * Called with the underlying vnode locked, which should prevent new dirty
281  * buffers from being queued.
282  */
283 int
284 vflushbuf(struct vnode *vp, int flags)
285 {
286 	struct buf *bp, *nbp;
287 	int error, pflags;
288 	bool dirty, sync;
289 
290 	sync = (flags & FSYNC_WAIT) != 0;
291 	pflags = PGO_CLEANIT | PGO_ALLPAGES |
292 		(sync ? PGO_SYNCIO : 0) |
293 		((flags & FSYNC_LAZY) ? PGO_LAZY : 0);
294 	mutex_enter(vp->v_interlock);
295 	(void) VOP_PUTPAGES(vp, 0, 0, pflags);
296 
297 loop:
298 	mutex_enter(&bufcache_lock);
299 	for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
300 		KASSERT(bp->b_vp == vp);
301 		nbp = LIST_NEXT(bp, b_vnbufs);
302 		if ((bp->b_cflags & BC_BUSY))
303 			continue;
304 		if ((bp->b_oflags & BO_DELWRI) == 0)
305 			panic("vflushbuf: not dirty, bp %p", bp);
306 		bp->b_cflags |= BC_BUSY | BC_VFLUSH;
307 		mutex_exit(&bufcache_lock);
308 		/*
309 		 * Wait for I/O associated with indirect blocks to complete,
310 		 * since there is no way to quickly wait for them below.
311 		 */
312 		if (bp->b_vp == vp || !sync)
313 			(void) bawrite(bp);
314 		else {
315 			error = bwrite(bp);
316 			if (error)
317 				return error;
318 		}
319 		goto loop;
320 	}
321 	mutex_exit(&bufcache_lock);
322 
323 	if (!sync)
324 		return 0;
325 
326 	mutex_enter(vp->v_interlock);
327 	while (vp->v_numoutput != 0)
328 		cv_wait(&vp->v_cv, vp->v_interlock);
329 	dirty = !LIST_EMPTY(&vp->v_dirtyblkhd);
330 	mutex_exit(vp->v_interlock);
331 
332 	if (dirty) {
333 		vprint("vflushbuf: dirty", vp);
334 		goto loop;
335 	}
336 
337 	return 0;
338 }
339 
340 /*
341  * Create a vnode for a block device.
342  * Used for root filesystem and swap areas.
343  * Also used for memory file system special devices.
344  */
345 int
346 bdevvp(dev_t dev, vnode_t **vpp)
347 {
348 	struct vattr va;
349 
350 	vattr_null(&va);
351 	va.va_type = VBLK;
352 	va.va_rdev = dev;
353 
354 	return vcache_new(dead_rootmount, NULL, &va, NOCRED, NULL, vpp);
355 }
356 
357 /*
358  * Create a vnode for a character device.
359  * Used for kernfs and some console handling.
360  */
361 int
362 cdevvp(dev_t dev, vnode_t **vpp)
363 {
364 	struct vattr va;
365 
366 	vattr_null(&va);
367 	va.va_type = VCHR;
368 	va.va_rdev = dev;
369 
370 	return vcache_new(dead_rootmount, NULL, &va, NOCRED, NULL, vpp);
371 }
372 
373 /*
374  * Associate a buffer with a vnode.  There must already be a hold on
375  * the vnode.
376  */
377 void
378 bgetvp(struct vnode *vp, struct buf *bp)
379 {
380 
381 	KASSERT(bp->b_vp == NULL);
382 	KASSERT(bp->b_objlock == &buffer_lock);
383 	KASSERT(mutex_owned(vp->v_interlock));
384 	KASSERT(mutex_owned(&bufcache_lock));
385 	KASSERT((bp->b_cflags & BC_BUSY) != 0);
386 	KASSERT(!cv_has_waiters(&bp->b_done));
387 
388 	vholdl(vp);
389 	bp->b_vp = vp;
390 	if (vp->v_type == VBLK || vp->v_type == VCHR)
391 		bp->b_dev = vp->v_rdev;
392 	else
393 		bp->b_dev = NODEV;
394 
395 	/*
396 	 * Insert onto list for new vnode.
397 	 */
398 	bufinsvn(bp, &vp->v_cleanblkhd);
399 	bp->b_objlock = vp->v_interlock;
400 }
401 
402 /*
403  * Disassociate a buffer from a vnode.
404  */
405 void
406 brelvp(struct buf *bp)
407 {
408 	struct vnode *vp = bp->b_vp;
409 
410 	KASSERT(vp != NULL);
411 	KASSERT(bp->b_objlock == vp->v_interlock);
412 	KASSERT(mutex_owned(vp->v_interlock));
413 	KASSERT(mutex_owned(&bufcache_lock));
414 	KASSERT((bp->b_cflags & BC_BUSY) != 0);
415 	KASSERT(!cv_has_waiters(&bp->b_done));
416 
417 	/*
418 	 * Delete from old vnode list, if on one.
419 	 */
420 	if (LIST_NEXT(bp, b_vnbufs) != NOLIST)
421 		bufremvn(bp);
422 
423 	if (vp->v_uobj.uo_npages == 0 && (vp->v_iflag & VI_ONWORKLST) &&
424 	    LIST_FIRST(&vp->v_dirtyblkhd) == NULL) {
425 		vp->v_iflag &= ~VI_WRMAPDIRTY;
426 		vn_syncer_remove_from_worklist(vp);
427 	}
428 
429 	bp->b_objlock = &buffer_lock;
430 	bp->b_vp = NULL;
431 	holdrelel(vp);
432 }
433 
434 /*
435  * Reassign a buffer from one vnode list to another.
436  * The list reassignment must be within the same vnode.
437  * Used to assign file specific control information
438  * (indirect blocks) to the list to which they belong.
439  */
440 void
441 reassignbuf(struct buf *bp, struct vnode *vp)
442 {
443 	struct buflists *listheadp;
444 	int delayx;
445 
446 	KASSERT(mutex_owned(&bufcache_lock));
447 	KASSERT(bp->b_objlock == vp->v_interlock);
448 	KASSERT(mutex_owned(vp->v_interlock));
449 	KASSERT((bp->b_cflags & BC_BUSY) != 0);
450 
451 	/*
452 	 * Delete from old vnode list, if on one.
453 	 */
454 	if (LIST_NEXT(bp, b_vnbufs) != NOLIST)
455 		bufremvn(bp);
456 
457 	/*
458 	 * If dirty, put on list of dirty buffers;
459 	 * otherwise insert onto list of clean buffers.
460 	 */
461 	if ((bp->b_oflags & BO_DELWRI) == 0) {
462 		listheadp = &vp->v_cleanblkhd;
463 		if (vp->v_uobj.uo_npages == 0 &&
464 		    (vp->v_iflag & VI_ONWORKLST) &&
465 		    LIST_FIRST(&vp->v_dirtyblkhd) == NULL) {
466 			vp->v_iflag &= ~VI_WRMAPDIRTY;
467 			vn_syncer_remove_from_worklist(vp);
468 		}
469 	} else {
470 		listheadp = &vp->v_dirtyblkhd;
471 		if ((vp->v_iflag & VI_ONWORKLST) == 0) {
472 			switch (vp->v_type) {
473 			case VDIR:
474 				delayx = dirdelay;
475 				break;
476 			case VBLK:
477 				if (spec_node_getmountedfs(vp) != NULL) {
478 					delayx = metadelay;
479 					break;
480 				}
481 				/* fall through */
482 			default:
483 				delayx = filedelay;
484 				break;
485 			}
486 			if (!vp->v_mount ||
487 			    (vp->v_mount->mnt_flag & MNT_ASYNC) == 0)
488 				vn_syncer_add_to_worklist(vp, delayx);
489 		}
490 	}
491 	bufinsvn(bp, listheadp);
492 }
493 
494 /*
495  * Lookup a vnode by device number and return it referenced.
496  */
497 int
498 vfinddev(dev_t dev, enum vtype type, vnode_t **vpp)
499 {
500 
501 	return (spec_node_lookup_by_dev(type, dev, vpp) == 0);
502 }
503 
504 /*
505  * Revoke all the vnodes corresponding to the specified minor number
506  * range (endpoints inclusive) of the specified major.
507  */
508 void
509 vdevgone(int maj, int minl, int minh, enum vtype type)
510 {
511 	vnode_t *vp;
512 	dev_t dev;
513 	int mn;
514 
515 	for (mn = minl; mn <= minh; mn++) {
516 		dev = makedev(maj, mn);
517 		while (spec_node_lookup_by_dev(type, dev, &vp) == 0) {
518 			VOP_REVOKE(vp, REVOKEALL);
519 			vrele(vp);
520 		}
521 	}
522 }
523 
524 /*
525  * The filesystem synchronizer mechanism - syncer.
526  *
527  * It is useful to delay writes of file data and filesystem metadata for
528  * a certain amount of time so that quickly created and deleted files need
529  * not waste disk bandwidth being created and removed.  To implement this,
530  * vnodes are appended to a "workitem" queue.
531  *
532  * Most pending metadata should not wait for more than ten seconds.  Thus,
533  * mounted on block devices are delayed only about a half the time that file
534  * data is delayed.  Similarly, directory updates are more critical, so are
535  * only delayed about a third the time that file data is delayed.
536  *
537  * There are SYNCER_MAXDELAY queues that are processed in a round-robin
538  * manner at a rate of one each second (driven off the filesystem syner
539  * thread). The syncer_delayno variable indicates the next queue that is
540  * to be processed.  Items that need to be processed soon are placed in
541  * this queue:
542  *
543  *	syncer_workitem_pending[syncer_delayno]
544  *
545  * A delay of e.g. fifteen seconds is done by placing the request fifteen
546  * entries later in the queue:
547  *
548  *	syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
549  *
550  * Flag VI_ONWORKLST indicates that vnode is added into the queue.
551  */
552 
553 #define SYNCER_MAXDELAY		32
554 
555 typedef TAILQ_HEAD(synclist, vnode_impl) synclist_t;
556 
557 static void	vn_syncer_add1(struct vnode *, int);
558 static void	sysctl_vfs_syncfs_setup(struct sysctllog **);
559 
560 /*
561  * Defines and variables for the syncer process.
562  */
563 int syncer_maxdelay = SYNCER_MAXDELAY;	/* maximum delay time */
564 time_t syncdelay = 30;			/* max time to delay syncing data */
565 time_t filedelay = 30;			/* time to delay syncing files */
566 time_t dirdelay  = 15;			/* time to delay syncing directories */
567 time_t metadelay = 10;			/* time to delay syncing metadata */
568 time_t lockdelay = 1;			/* time to delay if locking fails */
569 
570 static kmutex_t		syncer_data_lock; /* short term lock on data structs */
571 
572 static int		syncer_delayno = 0;
573 static long		syncer_last;
574 static synclist_t *	syncer_workitem_pending;
575 
576 static void
577 vn_initialize_syncerd(void)
578 {
579 	int i;
580 
581 	syncer_last = SYNCER_MAXDELAY + 2;
582 
583 	sysctl_vfs_syncfs_setup(NULL);
584 
585 	syncer_workitem_pending =
586 	    kmem_alloc(syncer_last * sizeof (struct synclist), KM_SLEEP);
587 
588 	for (i = 0; i < syncer_last; i++)
589 		TAILQ_INIT(&syncer_workitem_pending[i]);
590 
591 	mutex_init(&syncer_data_lock, MUTEX_DEFAULT, IPL_NONE);
592 }
593 
594 /*
595  * Return delay factor appropriate for the given file system.   For
596  * WAPBL we use the sync vnode to burst out metadata updates: sync
597  * those file systems more frequently.
598  */
599 static inline int
600 sync_delay(struct mount *mp)
601 {
602 
603 	return mp->mnt_wapbl != NULL ? metadelay : syncdelay;
604 }
605 
606 /*
607  * Compute the next slot index from delay.
608  */
609 static inline int
610 sync_delay_slot(int delayx)
611 {
612 
613 	if (delayx > syncer_maxdelay - 2)
614 		delayx = syncer_maxdelay - 2;
615 	return (syncer_delayno + delayx) % syncer_last;
616 }
617 
618 /*
619  * Add an item to the syncer work queue.
620  */
621 static void
622 vn_syncer_add1(struct vnode *vp, int delayx)
623 {
624 	synclist_t *slp;
625 	vnode_impl_t *vip = VNODE_TO_VIMPL(vp);
626 
627 	KASSERT(mutex_owned(&syncer_data_lock));
628 
629 	if (vp->v_iflag & VI_ONWORKLST) {
630 		/*
631 		 * Remove in order to adjust the position of the vnode.
632 		 * Note: called from sched_sync(), which will not hold
633 		 * interlock, therefore we cannot modify v_iflag here.
634 		 */
635 		slp = &syncer_workitem_pending[vip->vi_synclist_slot];
636 		TAILQ_REMOVE(slp, vip, vi_synclist);
637 	} else {
638 		KASSERT(mutex_owned(vp->v_interlock));
639 		vp->v_iflag |= VI_ONWORKLST;
640 	}
641 
642 	vip->vi_synclist_slot = sync_delay_slot(delayx);
643 
644 	slp = &syncer_workitem_pending[vip->vi_synclist_slot];
645 	TAILQ_INSERT_TAIL(slp, vip, vi_synclist);
646 }
647 
648 void
649 vn_syncer_add_to_worklist(struct vnode *vp, int delayx)
650 {
651 
652 	KASSERT(mutex_owned(vp->v_interlock));
653 
654 	mutex_enter(&syncer_data_lock);
655 	vn_syncer_add1(vp, delayx);
656 	mutex_exit(&syncer_data_lock);
657 }
658 
659 /*
660  * Remove an item from the syncer work queue.
661  */
662 void
663 vn_syncer_remove_from_worklist(struct vnode *vp)
664 {
665 	synclist_t *slp;
666 	vnode_impl_t *vip = VNODE_TO_VIMPL(vp);
667 
668 	KASSERT(mutex_owned(vp->v_interlock));
669 
670 	mutex_enter(&syncer_data_lock);
671 	if (vp->v_iflag & VI_ONWORKLST) {
672 		vp->v_iflag &= ~VI_ONWORKLST;
673 		slp = &syncer_workitem_pending[vip->vi_synclist_slot];
674 		TAILQ_REMOVE(slp, vip, vi_synclist);
675 	}
676 	mutex_exit(&syncer_data_lock);
677 }
678 
679 /*
680  * Add this mount point to the syncer.
681  */
682 void
683 vfs_syncer_add_to_worklist(struct mount *mp)
684 {
685 	static int start, incr, next;
686 	int vdelay;
687 
688 	KASSERT(mutex_owned(&mp->mnt_updating));
689 	KASSERT((mp->mnt_iflag & IMNT_ONWORKLIST) == 0);
690 
691 	/*
692 	 * We attempt to scatter the mount points on the list
693 	 * so that they will go off at evenly distributed times
694 	 * even if all the filesystems are mounted at once.
695 	 */
696 
697 	next += incr;
698 	if (next == 0 || next > syncer_maxdelay) {
699 		start /= 2;
700 		incr /= 2;
701 		if (start == 0) {
702 			start = syncer_maxdelay / 2;
703 			incr = syncer_maxdelay;
704 		}
705 		next = start;
706 	}
707 	mp->mnt_iflag |= IMNT_ONWORKLIST;
708 	vdelay = sync_delay(mp);
709 	mp->mnt_synclist_slot = vdelay > 0 ? next % vdelay : 0;
710 }
711 
712 /*
713  * Remove the mount point from the syncer.
714  */
715 void
716 vfs_syncer_remove_from_worklist(struct mount *mp)
717 {
718 
719 	KASSERT(mutex_owned(&mp->mnt_updating));
720 	KASSERT((mp->mnt_iflag & IMNT_ONWORKLIST) != 0);
721 
722 	mp->mnt_iflag &= ~IMNT_ONWORKLIST;
723 }
724 
725 /*
726  * Try lazy sync, return true on success.
727  */
728 static bool
729 lazy_sync_vnode(struct vnode *vp)
730 {
731 	bool synced;
732 
733 	KASSERT(mutex_owned(&syncer_data_lock));
734 
735 	synced = false;
736 	/* We are locking in the wrong direction. */
737 	if (mutex_tryenter(vp->v_interlock)) {
738 		mutex_exit(&syncer_data_lock);
739 		if (vcache_tryvget(vp) == 0) {
740 			if (vn_lock(vp, LK_EXCLUSIVE | LK_NOWAIT) == 0) {
741 				synced = true;
742 				(void) VOP_FSYNC(vp, curlwp->l_cred,
743 				    FSYNC_LAZY, 0, 0);
744 				vput(vp);
745 			} else
746 				vrele(vp);
747 		}
748 		mutex_enter(&syncer_data_lock);
749 	}
750 	return synced;
751 }
752 
753 /*
754  * System filesystem synchronizer daemon.
755  */
756 void
757 sched_sync(void *arg)
758 {
759 	mount_iterator_t *iter;
760 	synclist_t *slp;
761 	struct vnode_impl *vi;
762 	struct vnode *vp;
763 	struct mount *mp;
764 	time_t starttime;
765 	bool synced;
766 
767 	for (;;) {
768 		starttime = time_second;
769 
770 		/*
771 		 * Sync mounts whose dirty time has expired.
772 		 */
773 		mountlist_iterator_init(&iter);
774 		while ((mp = mountlist_iterator_trynext(iter)) != NULL) {
775 			if ((mp->mnt_iflag & IMNT_ONWORKLIST) == 0 ||
776 			    mp->mnt_synclist_slot != syncer_delayno) {
777 				continue;
778 			}
779 			mp->mnt_synclist_slot = sync_delay_slot(sync_delay(mp));
780 			VFS_SYNC(mp, MNT_LAZY, curlwp->l_cred);
781 		}
782 		mountlist_iterator_destroy(iter);
783 
784 		mutex_enter(&syncer_data_lock);
785 
786 		/*
787 		 * Push files whose dirty time has expired.
788 		 */
789 		slp = &syncer_workitem_pending[syncer_delayno];
790 		syncer_delayno += 1;
791 		if (syncer_delayno >= syncer_last)
792 			syncer_delayno = 0;
793 
794 		while ((vi = TAILQ_FIRST(slp)) != NULL) {
795 			vp = VIMPL_TO_VNODE(vi);
796 			synced = lazy_sync_vnode(vp);
797 
798 			/*
799 			 * XXX The vnode may have been recycled, in which
800 			 * case it may have a new identity.
801 			 */
802 			vi = TAILQ_FIRST(slp);
803 			if (vi != NULL && VIMPL_TO_VNODE(vi) == vp) {
804 				/*
805 				 * Put us back on the worklist.  The worklist
806 				 * routine will remove us from our current
807 				 * position and then add us back in at a later
808 				 * position.
809 				 *
810 				 * Try again sooner rather than later if
811 				 * we were unable to lock the vnode.  Lock
812 				 * failure should not prevent us from doing
813 				 * the sync "soon".
814 				 *
815 				 * If we locked it yet arrive here, it's
816 				 * likely that lazy sync is in progress and
817 				 * so the vnode still has dirty metadata.
818 				 * syncdelay is mainly to get this vnode out
819 				 * of the way so we do not consider it again
820 				 * "soon" in this loop, so the delay time is
821 				 * not critical as long as it is not "soon".
822 				 * While write-back strategy is the file
823 				 * system's domain, we expect write-back to
824 				 * occur no later than syncdelay seconds
825 				 * into the future.
826 				 */
827 				vn_syncer_add1(vp,
828 				    synced ? syncdelay : lockdelay);
829 			}
830 		}
831 
832 		/*
833 		 * If it has taken us less than a second to process the
834 		 * current work, then wait.  Otherwise start right over
835 		 * again.  We can still lose time if any single round
836 		 * takes more than two seconds, but it does not really
837 		 * matter as we are just trying to generally pace the
838 		 * filesystem activity.
839 		 */
840 		if (time_second == starttime) {
841 			kpause("syncer", false, hz, &syncer_data_lock);
842 		}
843 		mutex_exit(&syncer_data_lock);
844 	}
845 }
846 
847 static void
848 sysctl_vfs_syncfs_setup(struct sysctllog **clog)
849 {
850 	const struct sysctlnode *rnode, *cnode;
851 
852 	sysctl_createv(clog, 0, NULL, &rnode,
853 			CTLFLAG_PERMANENT,
854 			CTLTYPE_NODE, "sync",
855 			SYSCTL_DESCR("syncer options"),
856 			NULL, 0, NULL, 0,
857 			CTL_VFS, CTL_CREATE, CTL_EOL);
858 
859 	sysctl_createv(clog, 0, &rnode, &cnode,
860 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
861 			CTLTYPE_QUAD, "delay",
862 			SYSCTL_DESCR("max time to delay syncing data"),
863 			NULL, 0, &syncdelay, 0,
864 			CTL_CREATE, CTL_EOL);
865 
866 	sysctl_createv(clog, 0, &rnode, &cnode,
867 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
868 			CTLTYPE_QUAD, "filedelay",
869 			SYSCTL_DESCR("time to delay syncing files"),
870 			NULL, 0, &filedelay, 0,
871 			CTL_CREATE, CTL_EOL);
872 
873 	sysctl_createv(clog, 0, &rnode, &cnode,
874 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
875 			CTLTYPE_QUAD, "dirdelay",
876 			SYSCTL_DESCR("time to delay syncing directories"),
877 			NULL, 0, &dirdelay, 0,
878 			CTL_CREATE, CTL_EOL);
879 
880 	sysctl_createv(clog, 0, &rnode, &cnode,
881 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
882 			CTLTYPE_QUAD, "metadelay",
883 			SYSCTL_DESCR("time to delay syncing metadata"),
884 			NULL, 0, &metadelay, 0,
885 			CTL_CREATE, CTL_EOL);
886 }
887 
888 /*
889  * sysctl helper routine to return list of supported fstypes
890  */
891 int
892 sysctl_vfs_generic_fstypes(SYSCTLFN_ARGS)
893 {
894 	char bf[sizeof(((struct statvfs *)NULL)->f_fstypename)];
895 	char *where = oldp;
896 	struct vfsops *v;
897 	size_t needed, left, slen;
898 	int error, first;
899 
900 	if (newp != NULL)
901 		return (EPERM);
902 	if (namelen != 0)
903 		return (EINVAL);
904 
905 	first = 1;
906 	error = 0;
907 	needed = 0;
908 	left = *oldlenp;
909 
910 	sysctl_unlock();
911 	mutex_enter(&vfs_list_lock);
912 	LIST_FOREACH(v, &vfs_list, vfs_list) {
913 		if (where == NULL)
914 			needed += strlen(v->vfs_name) + 1;
915 		else {
916 			memset(bf, 0, sizeof(bf));
917 			if (first) {
918 				strncpy(bf, v->vfs_name, sizeof(bf));
919 				first = 0;
920 			} else {
921 				bf[0] = ' ';
922 				strncpy(bf + 1, v->vfs_name, sizeof(bf) - 1);
923 			}
924 			bf[sizeof(bf)-1] = '\0';
925 			slen = strlen(bf);
926 			if (left < slen + 1)
927 				break;
928 			v->vfs_refcount++;
929 			mutex_exit(&vfs_list_lock);
930 			/* +1 to copy out the trailing NUL byte */
931 			error = copyout(bf, where, slen + 1);
932 			mutex_enter(&vfs_list_lock);
933 			v->vfs_refcount--;
934 			if (error)
935 				break;
936 			where += slen;
937 			needed += slen;
938 			left -= slen;
939 		}
940 	}
941 	mutex_exit(&vfs_list_lock);
942 	sysctl_relock();
943 	*oldlenp = needed;
944 	return (error);
945 }
946 
947 int kinfo_vdebug = 1;
948 int kinfo_vgetfailed;
949 
950 #define KINFO_VNODESLOP	10
951 
952 /*
953  * Dump vnode list (via sysctl).
954  * Copyout address of vnode followed by vnode.
955  */
956 int
957 sysctl_kern_vnode(SYSCTLFN_ARGS)
958 {
959 	char *where = oldp;
960 	size_t *sizep = oldlenp;
961 	struct mount *mp;
962 	vnode_t *vp, vbuf;
963 	mount_iterator_t *iter;
964 	struct vnode_iterator *marker;
965 	char *bp = where;
966 	char *ewhere;
967 	int error;
968 
969 	if (namelen != 0)
970 		return (EOPNOTSUPP);
971 	if (newp != NULL)
972 		return (EPERM);
973 
974 #define VPTRSZ	sizeof(vnode_t *)
975 #define VNODESZ	sizeof(vnode_t)
976 	if (where == NULL) {
977 		*sizep = (numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ);
978 		return (0);
979 	}
980 	ewhere = where + *sizep;
981 
982 	sysctl_unlock();
983 	mountlist_iterator_init(&iter);
984 	while ((mp = mountlist_iterator_next(iter)) != NULL) {
985 		vfs_vnode_iterator_init(mp, &marker);
986 		while ((vp = vfs_vnode_iterator_next(marker, NULL, NULL))) {
987 			if (bp + VPTRSZ + VNODESZ > ewhere) {
988 				vrele(vp);
989 				vfs_vnode_iterator_destroy(marker);
990 				mountlist_iterator_destroy(iter);
991 				sysctl_relock();
992 				*sizep = bp - where;
993 				return (ENOMEM);
994 			}
995 			memcpy(&vbuf, vp, VNODESZ);
996 			if ((error = copyout(&vp, bp, VPTRSZ)) ||
997 			    (error = copyout(&vbuf, bp + VPTRSZ, VNODESZ))) {
998 				vrele(vp);
999 				vfs_vnode_iterator_destroy(marker);
1000 				mountlist_iterator_destroy(iter);
1001 				sysctl_relock();
1002 				return (error);
1003 			}
1004 			vrele(vp);
1005 			bp += VPTRSZ + VNODESZ;
1006 		}
1007 		vfs_vnode_iterator_destroy(marker);
1008 	}
1009 	mountlist_iterator_destroy(iter);
1010 	sysctl_relock();
1011 
1012 	*sizep = bp - where;
1013 	return (0);
1014 }
1015 
1016 /*
1017  * Set vnode attributes to VNOVAL
1018  */
1019 void
1020 vattr_null(struct vattr *vap)
1021 {
1022 
1023 	memset(vap, 0, sizeof(*vap));
1024 
1025 	vap->va_type = VNON;
1026 
1027 	/*
1028 	 * Assign individually so that it is safe even if size and
1029 	 * sign of each member are varied.
1030 	 */
1031 	vap->va_mode = VNOVAL;
1032 	vap->va_nlink = VNOVAL;
1033 	vap->va_uid = VNOVAL;
1034 	vap->va_gid = VNOVAL;
1035 	vap->va_fsid = VNOVAL;
1036 	vap->va_fileid = VNOVAL;
1037 	vap->va_size = VNOVAL;
1038 	vap->va_blocksize = VNOVAL;
1039 	vap->va_atime.tv_sec =
1040 	    vap->va_mtime.tv_sec =
1041 	    vap->va_ctime.tv_sec =
1042 	    vap->va_birthtime.tv_sec = VNOVAL;
1043 	vap->va_atime.tv_nsec =
1044 	    vap->va_mtime.tv_nsec =
1045 	    vap->va_ctime.tv_nsec =
1046 	    vap->va_birthtime.tv_nsec = VNOVAL;
1047 	vap->va_gen = VNOVAL;
1048 	vap->va_flags = VNOVAL;
1049 	vap->va_rdev = VNOVAL;
1050 	vap->va_bytes = VNOVAL;
1051 }
1052 
1053 /*
1054  * Vnode state to string.
1055  */
1056 const char *
1057 vstate_name(enum vnode_state state)
1058 {
1059 
1060 	switch (state) {
1061 	case VS_ACTIVE:
1062 		return "ACTIVE";
1063 	case VS_MARKER:
1064 		return "MARKER";
1065 	case VS_LOADING:
1066 		return "LOADING";
1067 	case VS_LOADED:
1068 		return "LOADED";
1069 	case VS_BLOCKED:
1070 		return "BLOCKED";
1071 	case VS_RECLAIMING:
1072 		return "RECLAIMING";
1073 	case VS_RECLAIMED:
1074 		return "RECLAIMED";
1075 	default:
1076 		return "ILLEGAL";
1077 	}
1078 }
1079 
1080 /*
1081  * Print a description of a vnode (common part).
1082  */
1083 static void
1084 vprint_common(struct vnode *vp, const char *prefix,
1085     void (*pr)(const char *, ...) __printflike(1, 2))
1086 {
1087 	int n;
1088 	char bf[96];
1089 	const uint8_t *cp;
1090 	vnode_impl_t *vip;
1091 	const char * const vnode_tags[] = { VNODE_TAGS };
1092 	const char * const vnode_types[] = { VNODE_TYPES };
1093 	const char vnode_flagbits[] = VNODE_FLAGBITS;
1094 
1095 #define ARRAY_SIZE(arr) (sizeof(arr) / sizeof(arr[0]))
1096 #define ARRAY_PRINT(idx, arr) \
1097     ((unsigned int)(idx) < ARRAY_SIZE(arr) ? (arr)[(idx)] : "UNKNOWN")
1098 
1099 	vip = VNODE_TO_VIMPL(vp);
1100 
1101 	snprintb(bf, sizeof(bf),
1102 	    vnode_flagbits, vp->v_iflag | vp->v_vflag | vp->v_uflag);
1103 
1104 	(*pr)("vnode %p flags %s\n", vp, bf);
1105 	(*pr)("%stag %s(%d) type %s(%d) mount %p typedata %p\n", prefix,
1106 	    ARRAY_PRINT(vp->v_tag, vnode_tags), vp->v_tag,
1107 	    ARRAY_PRINT(vp->v_type, vnode_types), vp->v_type,
1108 	    vp->v_mount, vp->v_mountedhere);
1109 	(*pr)("%susecount %d writecount %d holdcount %d\n", prefix,
1110 	    vp->v_usecount, vp->v_writecount, vp->v_holdcnt);
1111 	(*pr)("%ssize %" PRIx64 " writesize %" PRIx64 " numoutput %d\n",
1112 	    prefix, vp->v_size, vp->v_writesize, vp->v_numoutput);
1113 	(*pr)("%sdata %p lock %p\n", prefix, vp->v_data, &vip->vi_lock);
1114 
1115 	(*pr)("%sstate %s key(%p %zd)", prefix, vstate_name(vip->vi_state),
1116 	    vip->vi_key.vk_mount, vip->vi_key.vk_key_len);
1117 	n = vip->vi_key.vk_key_len;
1118 	cp = vip->vi_key.vk_key;
1119 	while (n-- > 0)
1120 		(*pr)(" %02x", *cp++);
1121 	(*pr)("\n");
1122 	(*pr)("%slrulisthd %p\n", prefix, vip->vi_lrulisthd);
1123 
1124 #undef ARRAY_PRINT
1125 #undef ARRAY_SIZE
1126 }
1127 
1128 /*
1129  * Print out a description of a vnode.
1130  */
1131 void
1132 vprint(const char *label, struct vnode *vp)
1133 {
1134 
1135 	if (label != NULL)
1136 		printf("%s: ", label);
1137 	vprint_common(vp, "\t", printf);
1138 	if (vp->v_data != NULL) {
1139 		printf("\t");
1140 		VOP_PRINT(vp);
1141 	}
1142 }
1143 
1144 /* Deprecated. Kept for KPI compatibility. */
1145 int
1146 vaccess(enum vtype type, mode_t file_mode, uid_t uid, gid_t gid,
1147     mode_t acc_mode, kauth_cred_t cred)
1148 {
1149 
1150 #ifdef DIAGNOSTIC
1151 	printf("vaccess: deprecated interface used.\n");
1152 #endif /* DIAGNOSTIC */
1153 
1154 	return kauth_authorize_vnode(cred, KAUTH_ACCESS_ACTION(acc_mode,
1155 	    type, file_mode), NULL /* This may panic. */, NULL,
1156 	    genfs_can_access(type, file_mode, uid, gid, acc_mode, cred));
1157 }
1158 
1159 /*
1160  * Given a file system name, look up the vfsops for that
1161  * file system, or return NULL if file system isn't present
1162  * in the kernel.
1163  */
1164 struct vfsops *
1165 vfs_getopsbyname(const char *name)
1166 {
1167 	struct vfsops *v;
1168 
1169 	mutex_enter(&vfs_list_lock);
1170 	LIST_FOREACH(v, &vfs_list, vfs_list) {
1171 		if (strcmp(v->vfs_name, name) == 0)
1172 			break;
1173 	}
1174 	if (v != NULL)
1175 		v->vfs_refcount++;
1176 	mutex_exit(&vfs_list_lock);
1177 
1178 	return (v);
1179 }
1180 
1181 void
1182 copy_statvfs_info(struct statvfs *sbp, const struct mount *mp)
1183 {
1184 	const struct statvfs *mbp;
1185 
1186 	if (sbp == (mbp = &mp->mnt_stat))
1187 		return;
1188 
1189 	(void)memcpy(&sbp->f_fsidx, &mbp->f_fsidx, sizeof(sbp->f_fsidx));
1190 	sbp->f_fsid = mbp->f_fsid;
1191 	sbp->f_owner = mbp->f_owner;
1192 	sbp->f_flag = mbp->f_flag;
1193 	sbp->f_syncwrites = mbp->f_syncwrites;
1194 	sbp->f_asyncwrites = mbp->f_asyncwrites;
1195 	sbp->f_syncreads = mbp->f_syncreads;
1196 	sbp->f_asyncreads = mbp->f_asyncreads;
1197 	(void)memcpy(sbp->f_spare, mbp->f_spare, sizeof(mbp->f_spare));
1198 	(void)memcpy(sbp->f_fstypename, mbp->f_fstypename,
1199 	    sizeof(sbp->f_fstypename));
1200 	(void)memcpy(sbp->f_mntonname, mbp->f_mntonname,
1201 	    sizeof(sbp->f_mntonname));
1202 	(void)memcpy(sbp->f_mntfromname, mp->mnt_stat.f_mntfromname,
1203 	    sizeof(sbp->f_mntfromname));
1204 	(void)memcpy(sbp->f_mntfromlabel, mp->mnt_stat.f_mntfromlabel,
1205 	    sizeof(sbp->f_mntfromlabel));
1206 	sbp->f_namemax = mbp->f_namemax;
1207 }
1208 
1209 int
1210 set_statvfs_info(const char *onp, int ukon, const char *fromp, int ukfrom,
1211     const char *vfsname, struct mount *mp, struct lwp *l)
1212 {
1213 	int error;
1214 	size_t size;
1215 	struct statvfs *sfs = &mp->mnt_stat;
1216 	int (*fun)(const void *, void *, size_t, size_t *);
1217 
1218 	(void)strlcpy(mp->mnt_stat.f_fstypename, vfsname,
1219 	    sizeof(mp->mnt_stat.f_fstypename));
1220 
1221 	if (onp) {
1222 		struct cwdinfo *cwdi = l->l_proc->p_cwdi;
1223 		fun = (ukon == UIO_SYSSPACE) ? copystr : copyinstr;
1224 		if (cwdi->cwdi_rdir != NULL) {
1225 			size_t len;
1226 			char *bp;
1227 			char *path = PNBUF_GET();
1228 
1229 			bp = path + MAXPATHLEN;
1230 			*--bp = '\0';
1231 			rw_enter(&cwdi->cwdi_lock, RW_READER);
1232 			error = getcwd_common(cwdi->cwdi_rdir, rootvnode, &bp,
1233 			    path, MAXPATHLEN / 2, 0, l);
1234 			rw_exit(&cwdi->cwdi_lock);
1235 			if (error) {
1236 				PNBUF_PUT(path);
1237 				return error;
1238 			}
1239 
1240 			len = strlen(bp);
1241 			if (len > sizeof(sfs->f_mntonname) - 1)
1242 				len = sizeof(sfs->f_mntonname) - 1;
1243 			(void)strncpy(sfs->f_mntonname, bp, len);
1244 			PNBUF_PUT(path);
1245 
1246 			if (len < sizeof(sfs->f_mntonname) - 1) {
1247 				error = (*fun)(onp, &sfs->f_mntonname[len],
1248 				    sizeof(sfs->f_mntonname) - len - 1, &size);
1249 				if (error)
1250 					return error;
1251 				size += len;
1252 			} else {
1253 				size = len;
1254 			}
1255 		} else {
1256 			error = (*fun)(onp, &sfs->f_mntonname,
1257 			    sizeof(sfs->f_mntonname) - 1, &size);
1258 			if (error)
1259 				return error;
1260 		}
1261 		(void)memset(sfs->f_mntonname + size, 0,
1262 		    sizeof(sfs->f_mntonname) - size);
1263 	}
1264 
1265 	if (fromp) {
1266 		fun = (ukfrom == UIO_SYSSPACE) ? copystr : copyinstr;
1267 		error = (*fun)(fromp, sfs->f_mntfromname,
1268 		    sizeof(sfs->f_mntfromname) - 1, &size);
1269 		if (error)
1270 			return error;
1271 		(void)memset(sfs->f_mntfromname + size, 0,
1272 		    sizeof(sfs->f_mntfromname) - size);
1273 	}
1274 	return 0;
1275 }
1276 
1277 void
1278 vfs_timestamp(struct timespec *ts)
1279 {
1280 
1281 	nanotime(ts);
1282 }
1283 
1284 time_t	rootfstime;			/* recorded root fs time, if known */
1285 void
1286 setrootfstime(time_t t)
1287 {
1288 	rootfstime = t;
1289 }
1290 
1291 static const uint8_t vttodt_tab[ ] = {
1292 	[VNON]	=	DT_UNKNOWN,
1293 	[VREG]	=	DT_REG,
1294 	[VDIR]	=	DT_DIR,
1295 	[VBLK]	=	DT_BLK,
1296 	[VCHR]	=	DT_CHR,
1297 	[VLNK]	=	DT_LNK,
1298 	[VSOCK]	=	DT_SOCK,
1299 	[VFIFO]	=	DT_FIFO,
1300 	[VBAD]	=	DT_UNKNOWN
1301 };
1302 
1303 uint8_t
1304 vtype2dt(enum vtype vt)
1305 {
1306 
1307 	CTASSERT(VBAD == __arraycount(vttodt_tab) - 1);
1308 	return vttodt_tab[vt];
1309 }
1310 
1311 int
1312 VFS_MOUNT(struct mount *mp, const char *a, void *b, size_t *c)
1313 {
1314 	int error;
1315 
1316 	KERNEL_LOCK(1, NULL);
1317 	error = (*(mp->mnt_op->vfs_mount))(mp, a, b, c);
1318 	KERNEL_UNLOCK_ONE(NULL);
1319 
1320 	return error;
1321 }
1322 
1323 int
1324 VFS_START(struct mount *mp, int a)
1325 {
1326 	int error;
1327 
1328 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
1329 		KERNEL_LOCK(1, NULL);
1330 	}
1331 	error = (*(mp->mnt_op->vfs_start))(mp, a);
1332 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
1333 		KERNEL_UNLOCK_ONE(NULL);
1334 	}
1335 
1336 	return error;
1337 }
1338 
1339 int
1340 VFS_UNMOUNT(struct mount *mp, int a)
1341 {
1342 	int error;
1343 
1344 	KERNEL_LOCK(1, NULL);
1345 	error = (*(mp->mnt_op->vfs_unmount))(mp, a);
1346 	KERNEL_UNLOCK_ONE(NULL);
1347 
1348 	return error;
1349 }
1350 
1351 int
1352 VFS_ROOT(struct mount *mp, struct vnode **a)
1353 {
1354 	int error;
1355 
1356 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
1357 		KERNEL_LOCK(1, NULL);
1358 	}
1359 	error = (*(mp->mnt_op->vfs_root))(mp, a);
1360 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
1361 		KERNEL_UNLOCK_ONE(NULL);
1362 	}
1363 
1364 	return error;
1365 }
1366 
1367 int
1368 VFS_QUOTACTL(struct mount *mp, struct quotactl_args *args)
1369 {
1370 	int error;
1371 
1372 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
1373 		KERNEL_LOCK(1, NULL);
1374 	}
1375 	error = (*(mp->mnt_op->vfs_quotactl))(mp, args);
1376 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
1377 		KERNEL_UNLOCK_ONE(NULL);
1378 	}
1379 
1380 	return error;
1381 }
1382 
1383 int
1384 VFS_STATVFS(struct mount *mp, struct statvfs *a)
1385 {
1386 	int error;
1387 
1388 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
1389 		KERNEL_LOCK(1, NULL);
1390 	}
1391 	error = (*(mp->mnt_op->vfs_statvfs))(mp, a);
1392 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
1393 		KERNEL_UNLOCK_ONE(NULL);
1394 	}
1395 
1396 	return error;
1397 }
1398 
1399 int
1400 VFS_SYNC(struct mount *mp, int a, struct kauth_cred *b)
1401 {
1402 	int error;
1403 
1404 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
1405 		KERNEL_LOCK(1, NULL);
1406 	}
1407 	error = (*(mp->mnt_op->vfs_sync))(mp, a, b);
1408 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
1409 		KERNEL_UNLOCK_ONE(NULL);
1410 	}
1411 
1412 	return error;
1413 }
1414 
1415 int
1416 VFS_FHTOVP(struct mount *mp, struct fid *a, struct vnode **b)
1417 {
1418 	int error;
1419 
1420 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
1421 		KERNEL_LOCK(1, NULL);
1422 	}
1423 	error = (*(mp->mnt_op->vfs_fhtovp))(mp, a, b);
1424 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
1425 		KERNEL_UNLOCK_ONE(NULL);
1426 	}
1427 
1428 	return error;
1429 }
1430 
1431 int
1432 VFS_VPTOFH(struct vnode *vp, struct fid *a, size_t *b)
1433 {
1434 	int error;
1435 
1436 	if ((vp->v_vflag & VV_MPSAFE) == 0) {
1437 		KERNEL_LOCK(1, NULL);
1438 	}
1439 	error = (*(vp->v_mount->mnt_op->vfs_vptofh))(vp, a, b);
1440 	if ((vp->v_vflag & VV_MPSAFE) == 0) {
1441 		KERNEL_UNLOCK_ONE(NULL);
1442 	}
1443 
1444 	return error;
1445 }
1446 
1447 int
1448 VFS_SNAPSHOT(struct mount *mp, struct vnode *a, struct timespec *b)
1449 {
1450 	int error;
1451 
1452 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
1453 		KERNEL_LOCK(1, NULL);
1454 	}
1455 	error = (*(mp->mnt_op->vfs_snapshot))(mp, a, b);
1456 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
1457 		KERNEL_UNLOCK_ONE(NULL);
1458 	}
1459 
1460 	return error;
1461 }
1462 
1463 int
1464 VFS_EXTATTRCTL(struct mount *mp, int a, struct vnode *b, int c, const char *d)
1465 {
1466 	int error;
1467 
1468 	KERNEL_LOCK(1, NULL);		/* XXXSMP check ffs */
1469 	error = (*(mp->mnt_op->vfs_extattrctl))(mp, a, b, c, d);
1470 	KERNEL_UNLOCK_ONE(NULL);	/* XXX */
1471 
1472 	return error;
1473 }
1474 
1475 int
1476 VFS_SUSPENDCTL(struct mount *mp, int a)
1477 {
1478 	int error;
1479 
1480 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
1481 		KERNEL_LOCK(1, NULL);
1482 	}
1483 	error = (*(mp->mnt_op->vfs_suspendctl))(mp, a);
1484 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
1485 		KERNEL_UNLOCK_ONE(NULL);
1486 	}
1487 
1488 	return error;
1489 }
1490 
1491 #if defined(DDB) || defined(DEBUGPRINT)
1492 static const char buf_flagbits[] = BUF_FLAGBITS;
1493 
1494 void
1495 vfs_buf_print(struct buf *bp, int full, void (*pr)(const char *, ...))
1496 {
1497 	char bf[1024];
1498 
1499 	(*pr)("  vp %p lblkno 0x%"PRIx64" blkno 0x%"PRIx64" rawblkno 0x%"
1500 	    PRIx64 " dev 0x%x\n",
1501 	    bp->b_vp, bp->b_lblkno, bp->b_blkno, bp->b_rawblkno, bp->b_dev);
1502 
1503 	snprintb(bf, sizeof(bf),
1504 	    buf_flagbits, bp->b_flags | bp->b_oflags | bp->b_cflags);
1505 	(*pr)("  error %d flags %s\n", bp->b_error, bf);
1506 
1507 	(*pr)("  bufsize 0x%lx bcount 0x%lx resid 0x%lx\n",
1508 		  bp->b_bufsize, bp->b_bcount, bp->b_resid);
1509 	(*pr)("  data %p saveaddr %p\n",
1510 		  bp->b_data, bp->b_saveaddr);
1511 	(*pr)("  iodone %p objlock %p\n", bp->b_iodone, bp->b_objlock);
1512 }
1513 
1514 void
1515 vfs_vnode_print(struct vnode *vp, int full, void (*pr)(const char *, ...))
1516 {
1517 
1518 	uvm_object_printit(&vp->v_uobj, full, pr);
1519 	(*pr)("\n");
1520 	vprint_common(vp, "", printf);
1521 	if (full) {
1522 		struct buf *bp;
1523 
1524 		(*pr)("clean bufs:\n");
1525 		LIST_FOREACH(bp, &vp->v_cleanblkhd, b_vnbufs) {
1526 			(*pr)(" bp %p\n", bp);
1527 			vfs_buf_print(bp, full, pr);
1528 		}
1529 
1530 		(*pr)("dirty bufs:\n");
1531 		LIST_FOREACH(bp, &vp->v_dirtyblkhd, b_vnbufs) {
1532 			(*pr)(" bp %p\n", bp);
1533 			vfs_buf_print(bp, full, pr);
1534 		}
1535 	}
1536 }
1537 
1538 void
1539 vfs_vnode_lock_print(void *vlock, int full, void (*pr)(const char *, ...))
1540 {
1541 	struct mount *mp;
1542 	vnode_impl_t *vip;
1543 
1544 	for (mp = _mountlist_next(NULL); mp; mp = _mountlist_next(mp)) {
1545 		TAILQ_FOREACH(vip, &mp->mnt_vnodelist, vi_mntvnodes) {
1546 			if (&vip->vi_lock != vlock)
1547 				continue;
1548 			vfs_vnode_print(VIMPL_TO_VNODE(vip), full, pr);
1549 		}
1550 	}
1551 }
1552 
1553 void
1554 vfs_mount_print(struct mount *mp, int full, void (*pr)(const char *, ...))
1555 {
1556 	char sbuf[256];
1557 
1558 	(*pr)("vnodecovered = %p data = %p\n",
1559 			mp->mnt_vnodecovered,mp->mnt_data);
1560 
1561 	(*pr)("fs_bshift %d dev_bshift = %d\n",
1562 			mp->mnt_fs_bshift,mp->mnt_dev_bshift);
1563 
1564 	snprintb(sbuf, sizeof(sbuf), __MNT_FLAG_BITS, mp->mnt_flag);
1565 	(*pr)("flag = %s\n", sbuf);
1566 
1567 	snprintb(sbuf, sizeof(sbuf), __IMNT_FLAG_BITS, mp->mnt_iflag);
1568 	(*pr)("iflag = %s\n", sbuf);
1569 
1570 	(*pr)("refcnt = %d updating @ %p\n", mp->mnt_refcnt, &mp->mnt_updating);
1571 
1572 	(*pr)("statvfs cache:\n");
1573 	(*pr)("\tbsize = %lu\n",mp->mnt_stat.f_bsize);
1574 	(*pr)("\tfrsize = %lu\n",mp->mnt_stat.f_frsize);
1575 	(*pr)("\tiosize = %lu\n",mp->mnt_stat.f_iosize);
1576 
1577 	(*pr)("\tblocks = %"PRIu64"\n",mp->mnt_stat.f_blocks);
1578 	(*pr)("\tbfree = %"PRIu64"\n",mp->mnt_stat.f_bfree);
1579 	(*pr)("\tbavail = %"PRIu64"\n",mp->mnt_stat.f_bavail);
1580 	(*pr)("\tbresvd = %"PRIu64"\n",mp->mnt_stat.f_bresvd);
1581 
1582 	(*pr)("\tfiles = %"PRIu64"\n",mp->mnt_stat.f_files);
1583 	(*pr)("\tffree = %"PRIu64"\n",mp->mnt_stat.f_ffree);
1584 	(*pr)("\tfavail = %"PRIu64"\n",mp->mnt_stat.f_favail);
1585 	(*pr)("\tfresvd = %"PRIu64"\n",mp->mnt_stat.f_fresvd);
1586 
1587 	(*pr)("\tf_fsidx = { 0x%"PRIx32", 0x%"PRIx32" }\n",
1588 			mp->mnt_stat.f_fsidx.__fsid_val[0],
1589 			mp->mnt_stat.f_fsidx.__fsid_val[1]);
1590 
1591 	(*pr)("\towner = %"PRIu32"\n",mp->mnt_stat.f_owner);
1592 	(*pr)("\tnamemax = %lu\n",mp->mnt_stat.f_namemax);
1593 
1594 	snprintb(sbuf, sizeof(sbuf), __MNT_FLAG_BITS, mp->mnt_stat.f_flag);
1595 
1596 	(*pr)("\tflag = %s\n",sbuf);
1597 	(*pr)("\tsyncwrites = %" PRIu64 "\n",mp->mnt_stat.f_syncwrites);
1598 	(*pr)("\tasyncwrites = %" PRIu64 "\n",mp->mnt_stat.f_asyncwrites);
1599 	(*pr)("\tsyncreads = %" PRIu64 "\n",mp->mnt_stat.f_syncreads);
1600 	(*pr)("\tasyncreads = %" PRIu64 "\n",mp->mnt_stat.f_asyncreads);
1601 	(*pr)("\tfstypename = %s\n",mp->mnt_stat.f_fstypename);
1602 	(*pr)("\tmntonname = %s\n",mp->mnt_stat.f_mntonname);
1603 	(*pr)("\tmntfromname = %s\n",mp->mnt_stat.f_mntfromname);
1604 
1605 	{
1606 		int cnt = 0;
1607 		vnode_t *vp;
1608 		vnode_impl_t *vip;
1609 		(*pr)("locked vnodes =");
1610 		TAILQ_FOREACH(vip, &mp->mnt_vnodelist, vi_mntvnodes) {
1611 			vp = VIMPL_TO_VNODE(vip);
1612 			if (VOP_ISLOCKED(vp)) {
1613 				if ((++cnt % 6) == 0) {
1614 					(*pr)(" %p,\n\t", vp);
1615 				} else {
1616 					(*pr)(" %p,", vp);
1617 				}
1618 			}
1619 		}
1620 		(*pr)("\n");
1621 	}
1622 
1623 	if (full) {
1624 		int cnt = 0;
1625 		vnode_t *vp;
1626 		vnode_impl_t *vip;
1627 		(*pr)("all vnodes =");
1628 		TAILQ_FOREACH(vip, &mp->mnt_vnodelist, vi_mntvnodes) {
1629 			vp = VIMPL_TO_VNODE(vip);
1630 			if (!TAILQ_NEXT(vip, vi_mntvnodes)) {
1631 				(*pr)(" %p", vp);
1632 			} else if ((++cnt % 6) == 0) {
1633 				(*pr)(" %p,\n\t", vp);
1634 			} else {
1635 				(*pr)(" %p,", vp);
1636 			}
1637 		}
1638 		(*pr)("\n");
1639 	}
1640 }
1641 
1642 /*
1643  * List all of the locked vnodes in the system.
1644  */
1645 void printlockedvnodes(void);
1646 
1647 void
1648 printlockedvnodes(void)
1649 {
1650 	struct mount *mp;
1651 	vnode_t *vp;
1652 	vnode_impl_t *vip;
1653 
1654 	printf("Locked vnodes\n");
1655 	for (mp = _mountlist_next(NULL); mp; mp = _mountlist_next(mp)) {
1656 		TAILQ_FOREACH(vip, &mp->mnt_vnodelist, vi_mntvnodes) {
1657 			vp = VIMPL_TO_VNODE(vip);
1658 			if (VOP_ISLOCKED(vp))
1659 				vprint(NULL, vp);
1660 		}
1661 	}
1662 }
1663 
1664 #endif /* DDB || DEBUGPRINT */
1665