xref: /netbsd-src/sys/kern/vfs_subr.c (revision 53d1339bf7f9c7367b35a9e1ebe693f9b047a47b)
1 /*	$NetBSD: vfs_subr.c,v 1.490 2021/02/04 21:07:06 jdolecek Exp $	*/
2 
3 /*-
4  * Copyright (c) 1997, 1998, 2004, 2005, 2007, 2008, 2019, 2020
5  *     The NetBSD Foundation, Inc.
6  * All rights reserved.
7  *
8  * This code is derived from software contributed to The NetBSD Foundation
9  * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility,
10  * NASA Ames Research Center, by Charles M. Hannum, by Andrew Doran,
11  * by Marshall Kirk McKusick and Greg Ganger at the University of Michigan.
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
16  * 1. Redistributions of source code must retain the above copyright
17  *    notice, this list of conditions and the following disclaimer.
18  * 2. Redistributions in binary form must reproduce the above copyright
19  *    notice, this list of conditions and the following disclaimer in the
20  *    documentation and/or other materials provided with the distribution.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
23  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
24  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
25  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
26  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
27  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
28  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
29  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
30  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
31  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
32  * POSSIBILITY OF SUCH DAMAGE.
33  */
34 
35 /*
36  * Copyright (c) 1989, 1993
37  *	The Regents of the University of California.  All rights reserved.
38  * (c) UNIX System Laboratories, Inc.
39  * All or some portions of this file are derived from material licensed
40  * to the University of California by American Telephone and Telegraph
41  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
42  * the permission of UNIX System Laboratories, Inc.
43  *
44  * Redistribution and use in source and binary forms, with or without
45  * modification, are permitted provided that the following conditions
46  * are met:
47  * 1. Redistributions of source code must retain the above copyright
48  *    notice, this list of conditions and the following disclaimer.
49  * 2. Redistributions in binary form must reproduce the above copyright
50  *    notice, this list of conditions and the following disclaimer in the
51  *    documentation and/or other materials provided with the distribution.
52  * 3. Neither the name of the University nor the names of its contributors
53  *    may be used to endorse or promote products derived from this software
54  *    without specific prior written permission.
55  *
56  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
57  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
58  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
59  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
60  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
61  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
62  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
63  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
64  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
65  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
66  * SUCH DAMAGE.
67  *
68  *	@(#)vfs_subr.c	8.13 (Berkeley) 4/18/94
69  */
70 
71 #include <sys/cdefs.h>
72 __KERNEL_RCSID(0, "$NetBSD: vfs_subr.c,v 1.490 2021/02/04 21:07:06 jdolecek Exp $");
73 
74 #ifdef _KERNEL_OPT
75 #include "opt_ddb.h"
76 #include "opt_compat_netbsd.h"
77 #include "opt_compat_43.h"
78 #endif
79 
80 #include <sys/param.h>
81 #include <sys/systm.h>
82 #include <sys/conf.h>
83 #include <sys/dirent.h>
84 #include <sys/filedesc.h>
85 #include <sys/kernel.h>
86 #include <sys/mount.h>
87 #include <sys/fstrans.h>
88 #include <sys/vnode_impl.h>
89 #include <sys/stat.h>
90 #include <sys/sysctl.h>
91 #include <sys/namei.h>
92 #include <sys/buf.h>
93 #include <sys/errno.h>
94 #include <sys/kmem.h>
95 #include <sys/syscallargs.h>
96 #include <sys/kauth.h>
97 #include <sys/module.h>
98 
99 #include <miscfs/genfs/genfs.h>
100 #include <miscfs/specfs/specdev.h>
101 #include <uvm/uvm_ddb.h>
102 
103 const enum vtype iftovt_tab[16] = {
104 	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
105 	VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
106 };
107 const int	vttoif_tab[9] = {
108 	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
109 	S_IFSOCK, S_IFIFO, S_IFMT,
110 };
111 
112 /*
113  * Insq/Remq for the vnode usage lists.
114  */
115 #define	bufinsvn(bp, dp)	LIST_INSERT_HEAD(dp, bp, b_vnbufs)
116 #define	bufremvn(bp) {							\
117 	LIST_REMOVE(bp, b_vnbufs);					\
118 	(bp)->b_vnbufs.le_next = NOLIST;				\
119 }
120 
121 int doforce = 1;		/* 1 => permit forcible unmounting */
122 
123 extern struct mount *dead_rootmount;
124 
125 /*
126  * Local declarations.
127  */
128 
129 static void vn_initialize_syncerd(void);
130 
131 /*
132  * Initialize the vnode management data structures.
133  */
134 void
135 vntblinit(void)
136 {
137 
138 	vn_initialize_syncerd();
139 	vfs_mount_sysinit();
140 	vfs_vnode_sysinit();
141 }
142 
143 /*
144  * Flush out and invalidate all buffers associated with a vnode.
145  * Called with the underlying vnode locked, which should prevent new dirty
146  * buffers from being queued.
147  */
148 int
149 vinvalbuf(struct vnode *vp, int flags, kauth_cred_t cred, struct lwp *l,
150 	  bool catch_p, int slptimeo)
151 {
152 	struct buf *bp, *nbp;
153 	int error;
154 	int flushflags = PGO_ALLPAGES | PGO_FREE | PGO_SYNCIO |
155 	    (flags & V_SAVE ? PGO_CLEANIT | PGO_RECLAIM : 0);
156 
157 	/* XXXUBC this doesn't look at flags or slp* */
158 	rw_enter(vp->v_uobj.vmobjlock, RW_WRITER);
159 	error = VOP_PUTPAGES(vp, 0, 0, flushflags);
160 	if (error) {
161 		return error;
162 	}
163 
164 	if (flags & V_SAVE) {
165 		error = VOP_FSYNC(vp, cred, FSYNC_WAIT|FSYNC_RECLAIM, 0, 0);
166 		if (error)
167 		        return (error);
168 		KASSERT(LIST_EMPTY(&vp->v_dirtyblkhd));
169 	}
170 
171 	mutex_enter(&bufcache_lock);
172 restart:
173 	for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
174 		KASSERT(bp->b_vp == vp);
175 		nbp = LIST_NEXT(bp, b_vnbufs);
176 		error = bbusy(bp, catch_p, slptimeo, NULL);
177 		if (error != 0) {
178 			if (error == EPASSTHROUGH)
179 				goto restart;
180 			mutex_exit(&bufcache_lock);
181 			return (error);
182 		}
183 		brelsel(bp, BC_INVAL | BC_VFLUSH);
184 	}
185 
186 	for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
187 		KASSERT(bp->b_vp == vp);
188 		nbp = LIST_NEXT(bp, b_vnbufs);
189 		error = bbusy(bp, catch_p, slptimeo, NULL);
190 		if (error != 0) {
191 			if (error == EPASSTHROUGH)
192 				goto restart;
193 			mutex_exit(&bufcache_lock);
194 			return (error);
195 		}
196 		/*
197 		 * XXX Since there are no node locks for NFS, I believe
198 		 * there is a slight chance that a delayed write will
199 		 * occur while sleeping just above, so check for it.
200 		 */
201 		if ((bp->b_oflags & BO_DELWRI) && (flags & V_SAVE)) {
202 #ifdef DEBUG
203 			printf("buffer still DELWRI\n");
204 #endif
205 			bp->b_cflags |= BC_BUSY | BC_VFLUSH;
206 			mutex_exit(&bufcache_lock);
207 			VOP_BWRITE(bp->b_vp, bp);
208 			mutex_enter(&bufcache_lock);
209 			goto restart;
210 		}
211 		brelsel(bp, BC_INVAL | BC_VFLUSH);
212 	}
213 
214 #ifdef DIAGNOSTIC
215 	if (!LIST_EMPTY(&vp->v_cleanblkhd) || !LIST_EMPTY(&vp->v_dirtyblkhd))
216 		panic("vinvalbuf: flush failed, vp %p", vp);
217 #endif
218 
219 	mutex_exit(&bufcache_lock);
220 
221 	return (0);
222 }
223 
224 /*
225  * Destroy any in core blocks past the truncation length.
226  * Called with the underlying vnode locked, which should prevent new dirty
227  * buffers from being queued.
228  */
229 int
230 vtruncbuf(struct vnode *vp, daddr_t lbn, bool catch_p, int slptimeo)
231 {
232 	struct buf *bp, *nbp;
233 	int error;
234 	voff_t off;
235 
236 	off = round_page((voff_t)lbn << vp->v_mount->mnt_fs_bshift);
237 	rw_enter(vp->v_uobj.vmobjlock, RW_WRITER);
238 	error = VOP_PUTPAGES(vp, off, 0, PGO_FREE | PGO_SYNCIO);
239 	if (error) {
240 		return error;
241 	}
242 
243 	mutex_enter(&bufcache_lock);
244 restart:
245 	for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
246 		KASSERT(bp->b_vp == vp);
247 		nbp = LIST_NEXT(bp, b_vnbufs);
248 		if (bp->b_lblkno < lbn)
249 			continue;
250 		error = bbusy(bp, catch_p, slptimeo, NULL);
251 		if (error != 0) {
252 			if (error == EPASSTHROUGH)
253 				goto restart;
254 			mutex_exit(&bufcache_lock);
255 			return (error);
256 		}
257 		brelsel(bp, BC_INVAL | BC_VFLUSH);
258 	}
259 
260 	for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
261 		KASSERT(bp->b_vp == vp);
262 		nbp = LIST_NEXT(bp, b_vnbufs);
263 		if (bp->b_lblkno < lbn)
264 			continue;
265 		error = bbusy(bp, catch_p, slptimeo, NULL);
266 		if (error != 0) {
267 			if (error == EPASSTHROUGH)
268 				goto restart;
269 			mutex_exit(&bufcache_lock);
270 			return (error);
271 		}
272 		brelsel(bp, BC_INVAL | BC_VFLUSH);
273 	}
274 	mutex_exit(&bufcache_lock);
275 
276 	return (0);
277 }
278 
279 /*
280  * Flush all dirty buffers from a vnode.
281  * Called with the underlying vnode locked, which should prevent new dirty
282  * buffers from being queued.
283  */
284 int
285 vflushbuf(struct vnode *vp, int flags)
286 {
287 	struct buf *bp, *nbp;
288 	int error, pflags;
289 	bool dirty, sync;
290 
291 	sync = (flags & FSYNC_WAIT) != 0;
292 	pflags = PGO_CLEANIT | PGO_ALLPAGES |
293 		(sync ? PGO_SYNCIO : 0) |
294 		((flags & FSYNC_LAZY) ? PGO_LAZY : 0);
295 	rw_enter(vp->v_uobj.vmobjlock, RW_WRITER);
296 	(void) VOP_PUTPAGES(vp, 0, 0, pflags);
297 
298 loop:
299 	mutex_enter(&bufcache_lock);
300 	for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
301 		KASSERT(bp->b_vp == vp);
302 		nbp = LIST_NEXT(bp, b_vnbufs);
303 		if ((bp->b_cflags & BC_BUSY))
304 			continue;
305 		if ((bp->b_oflags & BO_DELWRI) == 0)
306 			panic("vflushbuf: not dirty, bp %p", bp);
307 		bp->b_cflags |= BC_BUSY | BC_VFLUSH;
308 		mutex_exit(&bufcache_lock);
309 		/*
310 		 * Wait for I/O associated with indirect blocks to complete,
311 		 * since there is no way to quickly wait for them below.
312 		 */
313 		if (bp->b_vp == vp || !sync)
314 			(void) bawrite(bp);
315 		else {
316 			error = bwrite(bp);
317 			if (error)
318 				return error;
319 		}
320 		goto loop;
321 	}
322 	mutex_exit(&bufcache_lock);
323 
324 	if (!sync)
325 		return 0;
326 
327 	mutex_enter(vp->v_interlock);
328 	while (vp->v_numoutput != 0)
329 		cv_wait(&vp->v_cv, vp->v_interlock);
330 	dirty = !LIST_EMPTY(&vp->v_dirtyblkhd);
331 	mutex_exit(vp->v_interlock);
332 
333 	if (dirty) {
334 		vprint("vflushbuf: dirty", vp);
335 		goto loop;
336 	}
337 
338 	return 0;
339 }
340 
341 /*
342  * Create a vnode for a block device.
343  * Used for root filesystem and swap areas.
344  * Also used for memory file system special devices.
345  */
346 int
347 bdevvp(dev_t dev, vnode_t **vpp)
348 {
349 	struct vattr va;
350 
351 	vattr_null(&va);
352 	va.va_type = VBLK;
353 	va.va_rdev = dev;
354 
355 	return vcache_new(dead_rootmount, NULL, &va, NOCRED, NULL, vpp);
356 }
357 
358 /*
359  * Create a vnode for a character device.
360  * Used for kernfs and some console handling.
361  */
362 int
363 cdevvp(dev_t dev, vnode_t **vpp)
364 {
365 	struct vattr va;
366 
367 	vattr_null(&va);
368 	va.va_type = VCHR;
369 	va.va_rdev = dev;
370 
371 	return vcache_new(dead_rootmount, NULL, &va, NOCRED, NULL, vpp);
372 }
373 
374 /*
375  * Associate a buffer with a vnode.  There must already be a hold on
376  * the vnode.
377  */
378 void
379 bgetvp(struct vnode *vp, struct buf *bp)
380 {
381 
382 	KASSERT(bp->b_vp == NULL);
383 	KASSERT(bp->b_objlock == &buffer_lock);
384 	KASSERT(mutex_owned(vp->v_interlock));
385 	KASSERT(mutex_owned(&bufcache_lock));
386 	KASSERT((bp->b_cflags & BC_BUSY) != 0);
387 	KASSERT(!cv_has_waiters(&bp->b_done));
388 
389 	vholdl(vp);
390 	bp->b_vp = vp;
391 	if (vp->v_type == VBLK || vp->v_type == VCHR)
392 		bp->b_dev = vp->v_rdev;
393 	else
394 		bp->b_dev = NODEV;
395 
396 	/*
397 	 * Insert onto list for new vnode.
398 	 */
399 	bufinsvn(bp, &vp->v_cleanblkhd);
400 	bp->b_objlock = vp->v_interlock;
401 }
402 
403 /*
404  * Disassociate a buffer from a vnode.
405  */
406 void
407 brelvp(struct buf *bp)
408 {
409 	struct vnode *vp = bp->b_vp;
410 
411 	KASSERT(vp != NULL);
412 	KASSERT(bp->b_objlock == vp->v_interlock);
413 	KASSERT(mutex_owned(vp->v_interlock));
414 	KASSERT(mutex_owned(&bufcache_lock));
415 	KASSERT((bp->b_cflags & BC_BUSY) != 0);
416 	KASSERT(!cv_has_waiters(&bp->b_done));
417 
418 	/*
419 	 * Delete from old vnode list, if on one.
420 	 */
421 	if (LIST_NEXT(bp, b_vnbufs) != NOLIST)
422 		bufremvn(bp);
423 
424 	if ((vp->v_iflag & (VI_ONWORKLST | VI_PAGES)) == VI_ONWORKLST &&
425 	    LIST_FIRST(&vp->v_dirtyblkhd) == NULL)
426 		vn_syncer_remove_from_worklist(vp);
427 
428 	bp->b_objlock = &buffer_lock;
429 	bp->b_vp = NULL;
430 	holdrelel(vp);
431 }
432 
433 /*
434  * Reassign a buffer from one vnode list to another.
435  * The list reassignment must be within the same vnode.
436  * Used to assign file specific control information
437  * (indirect blocks) to the list to which they belong.
438  */
439 void
440 reassignbuf(struct buf *bp, struct vnode *vp)
441 {
442 	struct buflists *listheadp;
443 	int delayx;
444 
445 	KASSERT(mutex_owned(&bufcache_lock));
446 	KASSERT(bp->b_objlock == vp->v_interlock);
447 	KASSERT(mutex_owned(vp->v_interlock));
448 	KASSERT((bp->b_cflags & BC_BUSY) != 0);
449 
450 	/*
451 	 * Delete from old vnode list, if on one.
452 	 */
453 	if (LIST_NEXT(bp, b_vnbufs) != NOLIST)
454 		bufremvn(bp);
455 
456 	/*
457 	 * If dirty, put on list of dirty buffers;
458 	 * otherwise insert onto list of clean buffers.
459 	 */
460 	if ((bp->b_oflags & BO_DELWRI) == 0) {
461 		listheadp = &vp->v_cleanblkhd;
462 		if ((vp->v_iflag & (VI_ONWORKLST | VI_PAGES)) ==
463 		    VI_ONWORKLST &&
464 		    LIST_FIRST(&vp->v_dirtyblkhd) == NULL)
465 			vn_syncer_remove_from_worklist(vp);
466 	} else {
467 		listheadp = &vp->v_dirtyblkhd;
468 		if ((vp->v_iflag & VI_ONWORKLST) == 0) {
469 			switch (vp->v_type) {
470 			case VDIR:
471 				delayx = dirdelay;
472 				break;
473 			case VBLK:
474 				if (spec_node_getmountedfs(vp) != NULL) {
475 					delayx = metadelay;
476 					break;
477 				}
478 				/* fall through */
479 			default:
480 				delayx = filedelay;
481 				break;
482 			}
483 			if (!vp->v_mount ||
484 			    (vp->v_mount->mnt_flag & MNT_ASYNC) == 0)
485 				vn_syncer_add_to_worklist(vp, delayx);
486 		}
487 	}
488 	bufinsvn(bp, listheadp);
489 }
490 
491 /*
492  * Lookup a vnode by device number and return it referenced.
493  */
494 int
495 vfinddev(dev_t dev, enum vtype type, vnode_t **vpp)
496 {
497 
498 	return (spec_node_lookup_by_dev(type, dev, vpp) == 0);
499 }
500 
501 /*
502  * Revoke all the vnodes corresponding to the specified minor number
503  * range (endpoints inclusive) of the specified major.
504  */
505 void
506 vdevgone(int maj, int minl, int minh, enum vtype type)
507 {
508 	vnode_t *vp;
509 	dev_t dev;
510 	int mn;
511 
512 	for (mn = minl; mn <= minh; mn++) {
513 		dev = makedev(maj, mn);
514 		while (spec_node_lookup_by_dev(type, dev, &vp) == 0) {
515 			VOP_REVOKE(vp, REVOKEALL);
516 			vrele(vp);
517 		}
518 	}
519 }
520 
521 /*
522  * The filesystem synchronizer mechanism - syncer.
523  *
524  * It is useful to delay writes of file data and filesystem metadata for
525  * a certain amount of time so that quickly created and deleted files need
526  * not waste disk bandwidth being created and removed.  To implement this,
527  * vnodes are appended to a "workitem" queue.
528  *
529  * Most pending metadata should not wait for more than ten seconds.  Thus,
530  * mounted on block devices are delayed only about a half the time that file
531  * data is delayed.  Similarly, directory updates are more critical, so are
532  * only delayed about a third the time that file data is delayed.
533  *
534  * There are SYNCER_MAXDELAY queues that are processed in a round-robin
535  * manner at a rate of one each second (driven off the filesystem syner
536  * thread). The syncer_delayno variable indicates the next queue that is
537  * to be processed.  Items that need to be processed soon are placed in
538  * this queue:
539  *
540  *	syncer_workitem_pending[syncer_delayno]
541  *
542  * A delay of e.g. fifteen seconds is done by placing the request fifteen
543  * entries later in the queue:
544  *
545  *	syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
546  *
547  * Flag VI_ONWORKLST indicates that vnode is added into the queue.
548  */
549 
550 #define SYNCER_MAXDELAY		32
551 
552 typedef TAILQ_HEAD(synclist, vnode_impl) synclist_t;
553 
554 static void	vn_syncer_add1(struct vnode *, int);
555 static void	sysctl_vfs_syncfs_setup(struct sysctllog **);
556 
557 /*
558  * Defines and variables for the syncer process.
559  */
560 int syncer_maxdelay = SYNCER_MAXDELAY;	/* maximum delay time */
561 time_t syncdelay = 30;			/* max time to delay syncing data */
562 time_t filedelay = 30;			/* time to delay syncing files */
563 time_t dirdelay  = 15;			/* time to delay syncing directories */
564 time_t metadelay = 10;			/* time to delay syncing metadata */
565 time_t lockdelay = 1;			/* time to delay if locking fails */
566 
567 static kmutex_t		syncer_data_lock; /* short term lock on data structs */
568 
569 static int		syncer_delayno = 0;
570 static long		syncer_last;
571 static synclist_t *	syncer_workitem_pending;
572 
573 static void
574 vn_initialize_syncerd(void)
575 {
576 	int i;
577 
578 	syncer_last = SYNCER_MAXDELAY + 2;
579 
580 	sysctl_vfs_syncfs_setup(NULL);
581 
582 	syncer_workitem_pending =
583 	    kmem_alloc(syncer_last * sizeof (struct synclist), KM_SLEEP);
584 
585 	for (i = 0; i < syncer_last; i++)
586 		TAILQ_INIT(&syncer_workitem_pending[i]);
587 
588 	mutex_init(&syncer_data_lock, MUTEX_DEFAULT, IPL_NONE);
589 }
590 
591 /*
592  * Return delay factor appropriate for the given file system.   For
593  * WAPBL we use the sync vnode to burst out metadata updates: sync
594  * those file systems more frequently.
595  */
596 static inline int
597 sync_delay(struct mount *mp)
598 {
599 
600 	return mp->mnt_wapbl != NULL ? metadelay : syncdelay;
601 }
602 
603 /*
604  * Compute the next slot index from delay.
605  */
606 static inline int
607 sync_delay_slot(int delayx)
608 {
609 
610 	if (delayx > syncer_maxdelay - 2)
611 		delayx = syncer_maxdelay - 2;
612 	return (syncer_delayno + delayx) % syncer_last;
613 }
614 
615 /*
616  * Add an item to the syncer work queue.
617  */
618 static void
619 vn_syncer_add1(struct vnode *vp, int delayx)
620 {
621 	synclist_t *slp;
622 	vnode_impl_t *vip = VNODE_TO_VIMPL(vp);
623 
624 	KASSERT(mutex_owned(&syncer_data_lock));
625 
626 	if (vp->v_iflag & VI_ONWORKLST) {
627 		/*
628 		 * Remove in order to adjust the position of the vnode.
629 		 * Note: called from sched_sync(), which will not hold
630 		 * interlock, therefore we cannot modify v_iflag here.
631 		 */
632 		slp = &syncer_workitem_pending[vip->vi_synclist_slot];
633 		TAILQ_REMOVE(slp, vip, vi_synclist);
634 	} else {
635 		KASSERT(mutex_owned(vp->v_interlock));
636 		vp->v_iflag |= VI_ONWORKLST;
637 	}
638 
639 	vip->vi_synclist_slot = sync_delay_slot(delayx);
640 
641 	slp = &syncer_workitem_pending[vip->vi_synclist_slot];
642 	TAILQ_INSERT_TAIL(slp, vip, vi_synclist);
643 }
644 
645 void
646 vn_syncer_add_to_worklist(struct vnode *vp, int delayx)
647 {
648 
649 	KASSERT(mutex_owned(vp->v_interlock));
650 
651 	mutex_enter(&syncer_data_lock);
652 	vn_syncer_add1(vp, delayx);
653 	mutex_exit(&syncer_data_lock);
654 }
655 
656 /*
657  * Remove an item from the syncer work queue.
658  */
659 void
660 vn_syncer_remove_from_worklist(struct vnode *vp)
661 {
662 	synclist_t *slp;
663 	vnode_impl_t *vip = VNODE_TO_VIMPL(vp);
664 
665 	KASSERT(mutex_owned(vp->v_interlock));
666 
667 	if (vp->v_iflag & VI_ONWORKLST) {
668 		mutex_enter(&syncer_data_lock);
669 		vp->v_iflag &= ~VI_ONWORKLST;
670 		slp = &syncer_workitem_pending[vip->vi_synclist_slot];
671 		TAILQ_REMOVE(slp, vip, vi_synclist);
672 		mutex_exit(&syncer_data_lock);
673 	}
674 }
675 
676 /*
677  * Add this mount point to the syncer.
678  */
679 void
680 vfs_syncer_add_to_worklist(struct mount *mp)
681 {
682 	static int start, incr, next;
683 	int vdelay;
684 
685 	KASSERT(mutex_owned(mp->mnt_updating));
686 	KASSERT((mp->mnt_iflag & IMNT_ONWORKLIST) == 0);
687 
688 	/*
689 	 * We attempt to scatter the mount points on the list
690 	 * so that they will go off at evenly distributed times
691 	 * even if all the filesystems are mounted at once.
692 	 */
693 
694 	next += incr;
695 	if (next == 0 || next > syncer_maxdelay) {
696 		start /= 2;
697 		incr /= 2;
698 		if (start == 0) {
699 			start = syncer_maxdelay / 2;
700 			incr = syncer_maxdelay;
701 		}
702 		next = start;
703 	}
704 	mp->mnt_iflag |= IMNT_ONWORKLIST;
705 	vdelay = sync_delay(mp);
706 	mp->mnt_synclist_slot = vdelay > 0 ? next % vdelay : 0;
707 }
708 
709 /*
710  * Remove the mount point from the syncer.
711  */
712 void
713 vfs_syncer_remove_from_worklist(struct mount *mp)
714 {
715 
716 	KASSERT(mutex_owned(mp->mnt_updating));
717 	KASSERT((mp->mnt_iflag & IMNT_ONWORKLIST) != 0);
718 
719 	mp->mnt_iflag &= ~IMNT_ONWORKLIST;
720 }
721 
722 /*
723  * Try lazy sync, return true on success.
724  */
725 static bool
726 lazy_sync_vnode(struct vnode *vp)
727 {
728 	bool synced;
729 
730 	KASSERT(mutex_owned(&syncer_data_lock));
731 
732 	synced = false;
733 	if (vcache_tryvget(vp) == 0) {
734 		mutex_exit(&syncer_data_lock);
735 		if (vn_lock(vp, LK_EXCLUSIVE | LK_NOWAIT) == 0) {
736 			synced = true;
737 			(void) VOP_FSYNC(vp, curlwp->l_cred,
738 			    FSYNC_LAZY, 0, 0);
739 			vput(vp);
740 		} else
741 			vrele(vp);
742 		mutex_enter(&syncer_data_lock);
743 	}
744 	return synced;
745 }
746 
747 /*
748  * System filesystem synchronizer daemon.
749  */
750 void
751 sched_sync(void *arg)
752 {
753 	mount_iterator_t *iter;
754 	synclist_t *slp;
755 	struct vnode_impl *vi;
756 	struct vnode *vp;
757 	struct mount *mp;
758 	time_t starttime;
759 	bool synced;
760 
761 	for (;;) {
762 		starttime = time_second;
763 
764 		/*
765 		 * Sync mounts whose dirty time has expired.
766 		 */
767 		mountlist_iterator_init(&iter);
768 		while ((mp = mountlist_iterator_trynext(iter)) != NULL) {
769 			if ((mp->mnt_iflag & IMNT_ONWORKLIST) == 0 ||
770 			    mp->mnt_synclist_slot != syncer_delayno) {
771 				continue;
772 			}
773 			mp->mnt_synclist_slot = sync_delay_slot(sync_delay(mp));
774 			VFS_SYNC(mp, MNT_LAZY, curlwp->l_cred);
775 		}
776 		mountlist_iterator_destroy(iter);
777 
778 		mutex_enter(&syncer_data_lock);
779 
780 		/*
781 		 * Push files whose dirty time has expired.
782 		 */
783 		slp = &syncer_workitem_pending[syncer_delayno];
784 		syncer_delayno += 1;
785 		if (syncer_delayno >= syncer_last)
786 			syncer_delayno = 0;
787 
788 		while ((vi = TAILQ_FIRST(slp)) != NULL) {
789 			vp = VIMPL_TO_VNODE(vi);
790 			synced = lazy_sync_vnode(vp);
791 
792 			/*
793 			 * XXX The vnode may have been recycled, in which
794 			 * case it may have a new identity.
795 			 */
796 			vi = TAILQ_FIRST(slp);
797 			if (vi != NULL && VIMPL_TO_VNODE(vi) == vp) {
798 				/*
799 				 * Put us back on the worklist.  The worklist
800 				 * routine will remove us from our current
801 				 * position and then add us back in at a later
802 				 * position.
803 				 *
804 				 * Try again sooner rather than later if
805 				 * we were unable to lock the vnode.  Lock
806 				 * failure should not prevent us from doing
807 				 * the sync "soon".
808 				 *
809 				 * If we locked it yet arrive here, it's
810 				 * likely that lazy sync is in progress and
811 				 * so the vnode still has dirty metadata.
812 				 * syncdelay is mainly to get this vnode out
813 				 * of the way so we do not consider it again
814 				 * "soon" in this loop, so the delay time is
815 				 * not critical as long as it is not "soon".
816 				 * While write-back strategy is the file
817 				 * system's domain, we expect write-back to
818 				 * occur no later than syncdelay seconds
819 				 * into the future.
820 				 */
821 				vn_syncer_add1(vp,
822 				    synced ? syncdelay : lockdelay);
823 			}
824 		}
825 
826 		/*
827 		 * If it has taken us less than a second to process the
828 		 * current work, then wait.  Otherwise start right over
829 		 * again.  We can still lose time if any single round
830 		 * takes more than two seconds, but it does not really
831 		 * matter as we are just trying to generally pace the
832 		 * filesystem activity.
833 		 */
834 		if (time_second == starttime) {
835 			kpause("syncer", false, hz, &syncer_data_lock);
836 		}
837 		mutex_exit(&syncer_data_lock);
838 	}
839 }
840 
841 static void
842 sysctl_vfs_syncfs_setup(struct sysctllog **clog)
843 {
844 	const struct sysctlnode *rnode, *cnode;
845 
846 	sysctl_createv(clog, 0, NULL, &rnode,
847 			CTLFLAG_PERMANENT,
848 			CTLTYPE_NODE, "sync",
849 			SYSCTL_DESCR("syncer options"),
850 			NULL, 0, NULL, 0,
851 			CTL_VFS, CTL_CREATE, CTL_EOL);
852 
853 	sysctl_createv(clog, 0, &rnode, &cnode,
854 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
855 			CTLTYPE_QUAD, "delay",
856 			SYSCTL_DESCR("max time to delay syncing data"),
857 			NULL, 0, &syncdelay, 0,
858 			CTL_CREATE, CTL_EOL);
859 
860 	sysctl_createv(clog, 0, &rnode, &cnode,
861 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
862 			CTLTYPE_QUAD, "filedelay",
863 			SYSCTL_DESCR("time to delay syncing files"),
864 			NULL, 0, &filedelay, 0,
865 			CTL_CREATE, CTL_EOL);
866 
867 	sysctl_createv(clog, 0, &rnode, &cnode,
868 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
869 			CTLTYPE_QUAD, "dirdelay",
870 			SYSCTL_DESCR("time to delay syncing directories"),
871 			NULL, 0, &dirdelay, 0,
872 			CTL_CREATE, CTL_EOL);
873 
874 	sysctl_createv(clog, 0, &rnode, &cnode,
875 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
876 			CTLTYPE_QUAD, "metadelay",
877 			SYSCTL_DESCR("time to delay syncing metadata"),
878 			NULL, 0, &metadelay, 0,
879 			CTL_CREATE, CTL_EOL);
880 }
881 
882 /*
883  * sysctl helper routine to return list of supported fstypes
884  */
885 int
886 sysctl_vfs_generic_fstypes(SYSCTLFN_ARGS)
887 {
888 	char bf[sizeof(((struct statvfs *)NULL)->f_fstypename)];
889 	char *where = oldp;
890 	struct vfsops *v;
891 	size_t needed, left, slen;
892 	int error, first;
893 
894 	if (newp != NULL)
895 		return (EPERM);
896 	if (namelen != 0)
897 		return (EINVAL);
898 
899 	first = 1;
900 	error = 0;
901 	needed = 0;
902 	left = *oldlenp;
903 
904 	sysctl_unlock();
905 	mutex_enter(&vfs_list_lock);
906 	LIST_FOREACH(v, &vfs_list, vfs_list) {
907 		if (where == NULL)
908 			needed += strlen(v->vfs_name) + 1;
909 		else {
910 			memset(bf, 0, sizeof(bf));
911 			if (first) {
912 				strncpy(bf, v->vfs_name, sizeof(bf));
913 				first = 0;
914 			} else {
915 				bf[0] = ' ';
916 				strncpy(bf + 1, v->vfs_name, sizeof(bf) - 1);
917 			}
918 			bf[sizeof(bf)-1] = '\0';
919 			slen = strlen(bf);
920 			if (left < slen + 1)
921 				break;
922 			v->vfs_refcount++;
923 			mutex_exit(&vfs_list_lock);
924 			/* +1 to copy out the trailing NUL byte */
925 			error = copyout(bf, where, slen + 1);
926 			mutex_enter(&vfs_list_lock);
927 			v->vfs_refcount--;
928 			if (error)
929 				break;
930 			where += slen;
931 			needed += slen;
932 			left -= slen;
933 		}
934 	}
935 	mutex_exit(&vfs_list_lock);
936 	sysctl_relock();
937 	*oldlenp = needed;
938 	return (error);
939 }
940 
941 int kinfo_vdebug = 1;
942 int kinfo_vgetfailed;
943 
944 #define KINFO_VNODESLOP	10
945 
946 /*
947  * Dump vnode list (via sysctl).
948  * Copyout address of vnode followed by vnode.
949  */
950 int
951 sysctl_kern_vnode(SYSCTLFN_ARGS)
952 {
953 	char *where = oldp;
954 	size_t *sizep = oldlenp;
955 	struct mount *mp;
956 	vnode_t *vp, vbuf;
957 	mount_iterator_t *iter;
958 	struct vnode_iterator *marker;
959 	char *bp = where;
960 	char *ewhere;
961 	int error;
962 
963 	if (namelen != 0)
964 		return (EOPNOTSUPP);
965 	if (newp != NULL)
966 		return (EPERM);
967 
968 #define VPTRSZ	sizeof(vnode_t *)
969 #define VNODESZ	sizeof(vnode_t)
970 	if (where == NULL) {
971 		*sizep = (numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ);
972 		return (0);
973 	}
974 	ewhere = where + *sizep;
975 
976 	sysctl_unlock();
977 	mountlist_iterator_init(&iter);
978 	while ((mp = mountlist_iterator_next(iter)) != NULL) {
979 		vfs_vnode_iterator_init(mp, &marker);
980 		while ((vp = vfs_vnode_iterator_next(marker, NULL, NULL))) {
981 			if (bp + VPTRSZ + VNODESZ > ewhere) {
982 				vrele(vp);
983 				vfs_vnode_iterator_destroy(marker);
984 				mountlist_iterator_destroy(iter);
985 				sysctl_relock();
986 				*sizep = bp - where;
987 				return (ENOMEM);
988 			}
989 			memcpy(&vbuf, vp, VNODESZ);
990 			if ((error = copyout(&vp, bp, VPTRSZ)) ||
991 			    (error = copyout(&vbuf, bp + VPTRSZ, VNODESZ))) {
992 				vrele(vp);
993 				vfs_vnode_iterator_destroy(marker);
994 				mountlist_iterator_destroy(iter);
995 				sysctl_relock();
996 				return (error);
997 			}
998 			vrele(vp);
999 			bp += VPTRSZ + VNODESZ;
1000 		}
1001 		vfs_vnode_iterator_destroy(marker);
1002 	}
1003 	mountlist_iterator_destroy(iter);
1004 	sysctl_relock();
1005 
1006 	*sizep = bp - where;
1007 	return (0);
1008 }
1009 
1010 /*
1011  * Set vnode attributes to VNOVAL
1012  */
1013 void
1014 vattr_null(struct vattr *vap)
1015 {
1016 
1017 	memset(vap, 0, sizeof(*vap));
1018 
1019 	vap->va_type = VNON;
1020 
1021 	/*
1022 	 * Assign individually so that it is safe even if size and
1023 	 * sign of each member are varied.
1024 	 */
1025 	vap->va_mode = VNOVAL;
1026 	vap->va_nlink = VNOVAL;
1027 	vap->va_uid = VNOVAL;
1028 	vap->va_gid = VNOVAL;
1029 	vap->va_fsid = VNOVAL;
1030 	vap->va_fileid = VNOVAL;
1031 	vap->va_size = VNOVAL;
1032 	vap->va_blocksize = VNOVAL;
1033 	vap->va_atime.tv_sec =
1034 	    vap->va_mtime.tv_sec =
1035 	    vap->va_ctime.tv_sec =
1036 	    vap->va_birthtime.tv_sec = VNOVAL;
1037 	vap->va_atime.tv_nsec =
1038 	    vap->va_mtime.tv_nsec =
1039 	    vap->va_ctime.tv_nsec =
1040 	    vap->va_birthtime.tv_nsec = VNOVAL;
1041 	vap->va_gen = VNOVAL;
1042 	vap->va_flags = VNOVAL;
1043 	vap->va_rdev = VNOVAL;
1044 	vap->va_bytes = VNOVAL;
1045 }
1046 
1047 /*
1048  * Vnode state to string.
1049  */
1050 const char *
1051 vstate_name(enum vnode_state state)
1052 {
1053 
1054 	switch (state) {
1055 	case VS_ACTIVE:
1056 		return "ACTIVE";
1057 	case VS_MARKER:
1058 		return "MARKER";
1059 	case VS_LOADING:
1060 		return "LOADING";
1061 	case VS_LOADED:
1062 		return "LOADED";
1063 	case VS_BLOCKED:
1064 		return "BLOCKED";
1065 	case VS_RECLAIMING:
1066 		return "RECLAIMING";
1067 	case VS_RECLAIMED:
1068 		return "RECLAIMED";
1069 	default:
1070 		return "ILLEGAL";
1071 	}
1072 }
1073 
1074 /*
1075  * Print a description of a vnode (common part).
1076  */
1077 static void
1078 vprint_common(struct vnode *vp, const char *prefix,
1079     void (*pr)(const char *, ...) __printflike(1, 2))
1080 {
1081 	int n;
1082 	char bf[96];
1083 	const uint8_t *cp;
1084 	vnode_impl_t *vip;
1085 	const char * const vnode_tags[] = { VNODE_TAGS };
1086 	const char * const vnode_types[] = { VNODE_TYPES };
1087 	const char vnode_flagbits[] = VNODE_FLAGBITS;
1088 
1089 #define ARRAY_SIZE(arr) (sizeof(arr) / sizeof(arr[0]))
1090 #define ARRAY_PRINT(idx, arr) \
1091     ((unsigned int)(idx) < ARRAY_SIZE(arr) ? (arr)[(idx)] : "UNKNOWN")
1092 
1093 	vip = VNODE_TO_VIMPL(vp);
1094 
1095 	snprintb(bf, sizeof(bf),
1096 	    vnode_flagbits, vp->v_iflag | vp->v_vflag | vp->v_uflag);
1097 
1098 	(*pr)("vnode %p flags %s\n", vp, bf);
1099 	(*pr)("%stag %s(%d) type %s(%d) mount %p typedata %p\n", prefix,
1100 	    ARRAY_PRINT(vp->v_tag, vnode_tags), vp->v_tag,
1101 	    ARRAY_PRINT(vp->v_type, vnode_types), vp->v_type,
1102 	    vp->v_mount, vp->v_mountedhere);
1103 	(*pr)("%susecount %d writecount %d holdcount %d\n", prefix,
1104 	    vrefcnt(vp), vp->v_writecount, vp->v_holdcnt);
1105 	(*pr)("%ssize %" PRIx64 " writesize %" PRIx64 " numoutput %d\n",
1106 	    prefix, vp->v_size, vp->v_writesize, vp->v_numoutput);
1107 	(*pr)("%sdata %p lock %p\n", prefix, vp->v_data, &vip->vi_lock);
1108 
1109 	(*pr)("%sstate %s key(%p %zd)", prefix, vstate_name(vip->vi_state),
1110 	    vip->vi_key.vk_mount, vip->vi_key.vk_key_len);
1111 	n = vip->vi_key.vk_key_len;
1112 	cp = vip->vi_key.vk_key;
1113 	while (n-- > 0)
1114 		(*pr)(" %02x", *cp++);
1115 	(*pr)("\n");
1116 	(*pr)("%slrulisthd %p\n", prefix, vip->vi_lrulisthd);
1117 
1118 #undef ARRAY_PRINT
1119 #undef ARRAY_SIZE
1120 }
1121 
1122 /*
1123  * Print out a description of a vnode.
1124  */
1125 void
1126 vprint(const char *label, struct vnode *vp)
1127 {
1128 
1129 	if (label != NULL)
1130 		printf("%s: ", label);
1131 	vprint_common(vp, "\t", printf);
1132 	if (vp->v_data != NULL) {
1133 		printf("\t");
1134 		VOP_PRINT(vp);
1135 	}
1136 }
1137 
1138 /*
1139  * Given a file system name, look up the vfsops for that
1140  * file system, or return NULL if file system isn't present
1141  * in the kernel.
1142  */
1143 struct vfsops *
1144 vfs_getopsbyname(const char *name)
1145 {
1146 	struct vfsops *v;
1147 
1148 	mutex_enter(&vfs_list_lock);
1149 	LIST_FOREACH(v, &vfs_list, vfs_list) {
1150 		if (strcmp(v->vfs_name, name) == 0)
1151 			break;
1152 	}
1153 	if (v != NULL)
1154 		v->vfs_refcount++;
1155 	mutex_exit(&vfs_list_lock);
1156 
1157 	return (v);
1158 }
1159 
1160 void
1161 copy_statvfs_info(struct statvfs *sbp, const struct mount *mp)
1162 {
1163 	const struct statvfs *mbp;
1164 
1165 	if (sbp == (mbp = &mp->mnt_stat))
1166 		return;
1167 
1168 	(void)memcpy(&sbp->f_fsidx, &mbp->f_fsidx, sizeof(sbp->f_fsidx));
1169 	sbp->f_fsid = mbp->f_fsid;
1170 	sbp->f_owner = mbp->f_owner;
1171 	sbp->f_flag = mbp->f_flag;
1172 	sbp->f_syncwrites = mbp->f_syncwrites;
1173 	sbp->f_asyncwrites = mbp->f_asyncwrites;
1174 	sbp->f_syncreads = mbp->f_syncreads;
1175 	sbp->f_asyncreads = mbp->f_asyncreads;
1176 	(void)memcpy(sbp->f_spare, mbp->f_spare, sizeof(mbp->f_spare));
1177 	(void)memcpy(sbp->f_fstypename, mbp->f_fstypename,
1178 	    sizeof(sbp->f_fstypename));
1179 	(void)memcpy(sbp->f_mntonname, mbp->f_mntonname,
1180 	    sizeof(sbp->f_mntonname));
1181 	(void)memcpy(sbp->f_mntfromname, mp->mnt_stat.f_mntfromname,
1182 	    sizeof(sbp->f_mntfromname));
1183 	(void)memcpy(sbp->f_mntfromlabel, mp->mnt_stat.f_mntfromlabel,
1184 	    sizeof(sbp->f_mntfromlabel));
1185 	sbp->f_namemax = mbp->f_namemax;
1186 }
1187 
1188 int
1189 set_statvfs_info(const char *onp, int ukon, const char *fromp, int ukfrom,
1190     const char *vfsname, struct mount *mp, struct lwp *l)
1191 {
1192 	int error;
1193 	size_t size;
1194 	struct statvfs *sfs = &mp->mnt_stat;
1195 	int (*fun)(const void *, void *, size_t, size_t *);
1196 
1197 	(void)strlcpy(mp->mnt_stat.f_fstypename, vfsname,
1198 	    sizeof(mp->mnt_stat.f_fstypename));
1199 
1200 	if (onp) {
1201 		struct cwdinfo *cwdi = l->l_proc->p_cwdi;
1202 		fun = (ukon == UIO_SYSSPACE) ? copystr : copyinstr;
1203 		if (cwdi->cwdi_rdir != NULL) {
1204 			size_t len;
1205 			char *bp;
1206 			char *path = PNBUF_GET();
1207 
1208 			bp = path + MAXPATHLEN;
1209 			*--bp = '\0';
1210 			rw_enter(&cwdi->cwdi_lock, RW_READER);
1211 			error = getcwd_common(cwdi->cwdi_rdir, rootvnode, &bp,
1212 			    path, MAXPATHLEN / 2, 0, l);
1213 			rw_exit(&cwdi->cwdi_lock);
1214 			if (error) {
1215 				PNBUF_PUT(path);
1216 				return error;
1217 			}
1218 
1219 			len = strlen(bp);
1220 			if (len > sizeof(sfs->f_mntonname) - 1)
1221 				len = sizeof(sfs->f_mntonname) - 1;
1222 			(void)strncpy(sfs->f_mntonname, bp, len);
1223 			PNBUF_PUT(path);
1224 
1225 			if (len < sizeof(sfs->f_mntonname) - 1) {
1226 				error = (*fun)(onp, &sfs->f_mntonname[len],
1227 				    sizeof(sfs->f_mntonname) - len - 1, &size);
1228 				if (error)
1229 					return error;
1230 				size += len;
1231 			} else {
1232 				size = len;
1233 			}
1234 		} else {
1235 			error = (*fun)(onp, &sfs->f_mntonname,
1236 			    sizeof(sfs->f_mntonname) - 1, &size);
1237 			if (error)
1238 				return error;
1239 		}
1240 		(void)memset(sfs->f_mntonname + size, 0,
1241 		    sizeof(sfs->f_mntonname) - size);
1242 	}
1243 
1244 	if (fromp) {
1245 		fun = (ukfrom == UIO_SYSSPACE) ? copystr : copyinstr;
1246 		error = (*fun)(fromp, sfs->f_mntfromname,
1247 		    sizeof(sfs->f_mntfromname) - 1, &size);
1248 		if (error)
1249 			return error;
1250 		(void)memset(sfs->f_mntfromname + size, 0,
1251 		    sizeof(sfs->f_mntfromname) - size);
1252 	}
1253 	return 0;
1254 }
1255 
1256 /*
1257  * Knob to control the precision of file timestamps:
1258  *
1259  *   0 = seconds only; nanoseconds zeroed.
1260  *   1 = seconds and nanoseconds, accurate within 1/HZ.
1261  *   2 = seconds and nanoseconds, truncated to microseconds.
1262  * >=3 = seconds and nanoseconds, maximum precision.
1263  */
1264 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
1265 
1266 int vfs_timestamp_precision __read_mostly = TSP_NSEC;
1267 
1268 void
1269 vfs_timestamp(struct timespec *tsp)
1270 {
1271 	struct timeval tv;
1272 
1273 	switch (vfs_timestamp_precision) {
1274 	case TSP_SEC:
1275 		tsp->tv_sec = time_second;
1276 		tsp->tv_nsec = 0;
1277 		break;
1278 	case TSP_HZ:
1279 		getnanotime(tsp);
1280 		break;
1281 	case TSP_USEC:
1282 		microtime(&tv);
1283 		TIMEVAL_TO_TIMESPEC(&tv, tsp);
1284 		break;
1285 	case TSP_NSEC:
1286 	default:
1287 		nanotime(tsp);
1288 		break;
1289 	}
1290 }
1291 
1292 /*
1293  * The purpose of this routine is to remove granularity from accmode_t,
1294  * reducing it into standard unix access bits - VEXEC, VREAD, VWRITE,
1295  * VADMIN and VAPPEND.
1296  *
1297  * If it returns 0, the caller is supposed to continue with the usual
1298  * access checks using 'accmode' as modified by this routine.  If it
1299  * returns nonzero value, the caller is supposed to return that value
1300  * as errno.
1301  *
1302  * Note that after this routine runs, accmode may be zero.
1303  */
1304 int
1305 vfs_unixify_accmode(accmode_t *accmode)
1306 {
1307 	/*
1308 	 * There is no way to specify explicit "deny" rule using
1309 	 * file mode or POSIX.1e ACLs.
1310 	 */
1311 	if (*accmode & VEXPLICIT_DENY) {
1312 		*accmode = 0;
1313 		return (0);
1314 	}
1315 
1316 	/*
1317 	 * None of these can be translated into usual access bits.
1318 	 * Also, the common case for NFSv4 ACLs is to not contain
1319 	 * either of these bits. Caller should check for VWRITE
1320 	 * on the containing directory instead.
1321 	 */
1322 	if (*accmode & (VDELETE_CHILD | VDELETE))
1323 		return (EPERM);
1324 
1325 	if (*accmode & VADMIN_PERMS) {
1326 		*accmode &= ~VADMIN_PERMS;
1327 		*accmode |= VADMIN;
1328 	}
1329 
1330 	/*
1331 	 * There is no way to deny VREAD_ATTRIBUTES, VREAD_ACL
1332 	 * or VSYNCHRONIZE using file mode or POSIX.1e ACL.
1333 	 */
1334 	*accmode &= ~(VSTAT_PERMS | VSYNCHRONIZE);
1335 
1336 	return (0);
1337 }
1338 
1339 time_t	rootfstime;			/* recorded root fs time, if known */
1340 void
1341 setrootfstime(time_t t)
1342 {
1343 	rootfstime = t;
1344 }
1345 
1346 static const uint8_t vttodt_tab[ ] = {
1347 	[VNON]	=	DT_UNKNOWN,
1348 	[VREG]	=	DT_REG,
1349 	[VDIR]	=	DT_DIR,
1350 	[VBLK]	=	DT_BLK,
1351 	[VCHR]	=	DT_CHR,
1352 	[VLNK]	=	DT_LNK,
1353 	[VSOCK]	=	DT_SOCK,
1354 	[VFIFO]	=	DT_FIFO,
1355 	[VBAD]	=	DT_UNKNOWN
1356 };
1357 
1358 uint8_t
1359 vtype2dt(enum vtype vt)
1360 {
1361 
1362 	CTASSERT(VBAD == __arraycount(vttodt_tab) - 1);
1363 	return vttodt_tab[vt];
1364 }
1365 
1366 int
1367 VFS_MOUNT(struct mount *mp, const char *a, void *b, size_t *c)
1368 {
1369 	int error;
1370 
1371 	KERNEL_LOCK(1, NULL);
1372 	error = (*(mp->mnt_op->vfs_mount))(mp, a, b, c);
1373 	KERNEL_UNLOCK_ONE(NULL);
1374 
1375 	return error;
1376 }
1377 
1378 int
1379 VFS_START(struct mount *mp, int a)
1380 {
1381 	int error;
1382 
1383 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
1384 		KERNEL_LOCK(1, NULL);
1385 	}
1386 	error = (*(mp->mnt_op->vfs_start))(mp, a);
1387 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
1388 		KERNEL_UNLOCK_ONE(NULL);
1389 	}
1390 
1391 	return error;
1392 }
1393 
1394 int
1395 VFS_UNMOUNT(struct mount *mp, int a)
1396 {
1397 	int error;
1398 
1399 	KERNEL_LOCK(1, NULL);
1400 	error = (*(mp->mnt_op->vfs_unmount))(mp, a);
1401 	KERNEL_UNLOCK_ONE(NULL);
1402 
1403 	return error;
1404 }
1405 
1406 int
1407 VFS_ROOT(struct mount *mp, int lktype, struct vnode **a)
1408 {
1409 	int error;
1410 
1411 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
1412 		KERNEL_LOCK(1, NULL);
1413 	}
1414 	error = (*(mp->mnt_op->vfs_root))(mp, lktype, a);
1415 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
1416 		KERNEL_UNLOCK_ONE(NULL);
1417 	}
1418 
1419 	return error;
1420 }
1421 
1422 int
1423 VFS_QUOTACTL(struct mount *mp, struct quotactl_args *args)
1424 {
1425 	int error;
1426 
1427 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
1428 		KERNEL_LOCK(1, NULL);
1429 	}
1430 	error = (*(mp->mnt_op->vfs_quotactl))(mp, args);
1431 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
1432 		KERNEL_UNLOCK_ONE(NULL);
1433 	}
1434 
1435 	return error;
1436 }
1437 
1438 int
1439 VFS_STATVFS(struct mount *mp, struct statvfs *a)
1440 {
1441 	int error;
1442 
1443 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
1444 		KERNEL_LOCK(1, NULL);
1445 	}
1446 	error = (*(mp->mnt_op->vfs_statvfs))(mp, a);
1447 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
1448 		KERNEL_UNLOCK_ONE(NULL);
1449 	}
1450 
1451 	return error;
1452 }
1453 
1454 int
1455 VFS_SYNC(struct mount *mp, int a, struct kauth_cred *b)
1456 {
1457 	int error;
1458 
1459 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
1460 		KERNEL_LOCK(1, NULL);
1461 	}
1462 	error = (*(mp->mnt_op->vfs_sync))(mp, a, b);
1463 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
1464 		KERNEL_UNLOCK_ONE(NULL);
1465 	}
1466 
1467 	return error;
1468 }
1469 
1470 int
1471 VFS_FHTOVP(struct mount *mp, struct fid *a, int b, struct vnode **c)
1472 {
1473 	int error;
1474 
1475 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
1476 		KERNEL_LOCK(1, NULL);
1477 	}
1478 	error = (*(mp->mnt_op->vfs_fhtovp))(mp, a, b, c);
1479 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
1480 		KERNEL_UNLOCK_ONE(NULL);
1481 	}
1482 
1483 	return error;
1484 }
1485 
1486 int
1487 VFS_VPTOFH(struct vnode *vp, struct fid *a, size_t *b)
1488 {
1489 	int error;
1490 
1491 	if ((vp->v_vflag & VV_MPSAFE) == 0) {
1492 		KERNEL_LOCK(1, NULL);
1493 	}
1494 	error = (*(vp->v_mount->mnt_op->vfs_vptofh))(vp, a, b);
1495 	if ((vp->v_vflag & VV_MPSAFE) == 0) {
1496 		KERNEL_UNLOCK_ONE(NULL);
1497 	}
1498 
1499 	return error;
1500 }
1501 
1502 int
1503 VFS_SNAPSHOT(struct mount *mp, struct vnode *a, struct timespec *b)
1504 {
1505 	int error;
1506 
1507 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
1508 		KERNEL_LOCK(1, NULL);
1509 	}
1510 	error = (*(mp->mnt_op->vfs_snapshot))(mp, a, b);
1511 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
1512 		KERNEL_UNLOCK_ONE(NULL);
1513 	}
1514 
1515 	return error;
1516 }
1517 
1518 int
1519 VFS_EXTATTRCTL(struct mount *mp, int a, struct vnode *b, int c, const char *d)
1520 {
1521 	int error;
1522 
1523 	KERNEL_LOCK(1, NULL);		/* XXXSMP check ffs */
1524 	error = (*(mp->mnt_op->vfs_extattrctl))(mp, a, b, c, d);
1525 	KERNEL_UNLOCK_ONE(NULL);	/* XXX */
1526 
1527 	return error;
1528 }
1529 
1530 int
1531 VFS_SUSPENDCTL(struct mount *mp, int a)
1532 {
1533 	int error;
1534 
1535 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
1536 		KERNEL_LOCK(1, NULL);
1537 	}
1538 	error = (*(mp->mnt_op->vfs_suspendctl))(mp, a);
1539 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
1540 		KERNEL_UNLOCK_ONE(NULL);
1541 	}
1542 
1543 	return error;
1544 }
1545 
1546 #if defined(DDB) || defined(DEBUGPRINT)
1547 static const char buf_flagbits[] = BUF_FLAGBITS;
1548 
1549 void
1550 vfs_buf_print(struct buf *bp, int full, void (*pr)(const char *, ...))
1551 {
1552 	char bf[1024];
1553 
1554 	(*pr)("  vp %p lblkno 0x%"PRIx64" blkno 0x%"PRIx64" rawblkno 0x%"
1555 	    PRIx64 " dev 0x%x\n",
1556 	    bp->b_vp, bp->b_lblkno, bp->b_blkno, bp->b_rawblkno, bp->b_dev);
1557 
1558 	snprintb(bf, sizeof(bf),
1559 	    buf_flagbits, bp->b_flags | bp->b_oflags | bp->b_cflags);
1560 	(*pr)("  error %d flags %s\n", bp->b_error, bf);
1561 
1562 	(*pr)("  bufsize 0x%lx bcount 0x%lx resid 0x%lx\n",
1563 		  bp->b_bufsize, bp->b_bcount, bp->b_resid);
1564 	(*pr)("  data %p saveaddr %p\n",
1565 		  bp->b_data, bp->b_saveaddr);
1566 	(*pr)("  iodone %p objlock %p\n", bp->b_iodone, bp->b_objlock);
1567 }
1568 
1569 void
1570 vfs_vnode_print(struct vnode *vp, int full, void (*pr)(const char *, ...))
1571 {
1572 
1573 	uvm_object_printit(&vp->v_uobj, full, pr);
1574 	(*pr)("\n");
1575 	vprint_common(vp, "", pr);
1576 	if (full) {
1577 		struct buf *bp;
1578 
1579 		(*pr)("clean bufs:\n");
1580 		LIST_FOREACH(bp, &vp->v_cleanblkhd, b_vnbufs) {
1581 			(*pr)(" bp %p\n", bp);
1582 			vfs_buf_print(bp, full, pr);
1583 		}
1584 
1585 		(*pr)("dirty bufs:\n");
1586 		LIST_FOREACH(bp, &vp->v_dirtyblkhd, b_vnbufs) {
1587 			(*pr)(" bp %p\n", bp);
1588 			vfs_buf_print(bp, full, pr);
1589 		}
1590 	}
1591 }
1592 
1593 void
1594 vfs_vnode_lock_print(void *vlock, int full, void (*pr)(const char *, ...))
1595 {
1596 	struct mount *mp;
1597 	vnode_impl_t *vip;
1598 
1599 	for (mp = _mountlist_next(NULL); mp; mp = _mountlist_next(mp)) {
1600 		TAILQ_FOREACH(vip, &mp->mnt_vnodelist, vi_mntvnodes) {
1601 			if (&vip->vi_lock == vlock ||
1602 			    VIMPL_TO_VNODE(vip)->v_interlock == vlock)
1603 				vfs_vnode_print(VIMPL_TO_VNODE(vip), full, pr);
1604 		}
1605 	}
1606 }
1607 
1608 void
1609 vfs_mount_print_all(int full, void (*pr)(const char *, ...))
1610 {
1611 	struct mount *mp;
1612 	for (mp = _mountlist_next(NULL); mp; mp = _mountlist_next(mp))
1613 		vfs_mount_print(mp, full, pr);
1614 }
1615 
1616 void
1617 vfs_mount_print(struct mount *mp, int full, void (*pr)(const char *, ...))
1618 {
1619 	char sbuf[256];
1620 
1621 	(*pr)("vnodecovered = %p data = %p\n",
1622 			mp->mnt_vnodecovered,mp->mnt_data);
1623 
1624 	(*pr)("fs_bshift %d dev_bshift = %d\n",
1625 			mp->mnt_fs_bshift,mp->mnt_dev_bshift);
1626 
1627 	snprintb(sbuf, sizeof(sbuf), __MNT_FLAG_BITS, mp->mnt_flag);
1628 	(*pr)("flag = %s\n", sbuf);
1629 
1630 	snprintb(sbuf, sizeof(sbuf), __IMNT_FLAG_BITS, mp->mnt_iflag);
1631 	(*pr)("iflag = %s\n", sbuf);
1632 
1633 	(*pr)("refcnt = %d updating @ %p\n", mp->mnt_refcnt, mp->mnt_updating);
1634 
1635 	(*pr)("statvfs cache:\n");
1636 	(*pr)("\tbsize = %lu\n",mp->mnt_stat.f_bsize);
1637 	(*pr)("\tfrsize = %lu\n",mp->mnt_stat.f_frsize);
1638 	(*pr)("\tiosize = %lu\n",mp->mnt_stat.f_iosize);
1639 
1640 	(*pr)("\tblocks = %"PRIu64"\n",mp->mnt_stat.f_blocks);
1641 	(*pr)("\tbfree = %"PRIu64"\n",mp->mnt_stat.f_bfree);
1642 	(*pr)("\tbavail = %"PRIu64"\n",mp->mnt_stat.f_bavail);
1643 	(*pr)("\tbresvd = %"PRIu64"\n",mp->mnt_stat.f_bresvd);
1644 
1645 	(*pr)("\tfiles = %"PRIu64"\n",mp->mnt_stat.f_files);
1646 	(*pr)("\tffree = %"PRIu64"\n",mp->mnt_stat.f_ffree);
1647 	(*pr)("\tfavail = %"PRIu64"\n",mp->mnt_stat.f_favail);
1648 	(*pr)("\tfresvd = %"PRIu64"\n",mp->mnt_stat.f_fresvd);
1649 
1650 	(*pr)("\tf_fsidx = { 0x%"PRIx32", 0x%"PRIx32" }\n",
1651 			mp->mnt_stat.f_fsidx.__fsid_val[0],
1652 			mp->mnt_stat.f_fsidx.__fsid_val[1]);
1653 
1654 	(*pr)("\towner = %"PRIu32"\n",mp->mnt_stat.f_owner);
1655 	(*pr)("\tnamemax = %lu\n",mp->mnt_stat.f_namemax);
1656 
1657 	snprintb(sbuf, sizeof(sbuf), __MNT_FLAG_BITS, mp->mnt_stat.f_flag);
1658 
1659 	(*pr)("\tflag = %s\n",sbuf);
1660 	(*pr)("\tsyncwrites = %" PRIu64 "\n",mp->mnt_stat.f_syncwrites);
1661 	(*pr)("\tasyncwrites = %" PRIu64 "\n",mp->mnt_stat.f_asyncwrites);
1662 	(*pr)("\tsyncreads = %" PRIu64 "\n",mp->mnt_stat.f_syncreads);
1663 	(*pr)("\tasyncreads = %" PRIu64 "\n",mp->mnt_stat.f_asyncreads);
1664 	(*pr)("\tfstypename = %s\n",mp->mnt_stat.f_fstypename);
1665 	(*pr)("\tmntonname = %s\n",mp->mnt_stat.f_mntonname);
1666 	(*pr)("\tmntfromname = %s\n",mp->mnt_stat.f_mntfromname);
1667 
1668 	{
1669 		int cnt = 0;
1670 		vnode_t *vp;
1671 		vnode_impl_t *vip;
1672 		(*pr)("locked vnodes =");
1673 		TAILQ_FOREACH(vip, &mp->mnt_vnodelist, vi_mntvnodes) {
1674 			vp = VIMPL_TO_VNODE(vip);
1675 			if (VOP_ISLOCKED(vp)) {
1676 				if ((++cnt % 6) == 0) {
1677 					(*pr)(" %p,\n\t", vp);
1678 				} else {
1679 					(*pr)(" %p,", vp);
1680 				}
1681 			}
1682 		}
1683 		(*pr)("\n");
1684 	}
1685 
1686 	if (full) {
1687 		int cnt = 0;
1688 		vnode_t *vp;
1689 		vnode_impl_t *vip;
1690 		(*pr)("all vnodes =");
1691 		TAILQ_FOREACH(vip, &mp->mnt_vnodelist, vi_mntvnodes) {
1692 			vp = VIMPL_TO_VNODE(vip);
1693 			if (!TAILQ_NEXT(vip, vi_mntvnodes)) {
1694 				(*pr)(" %p", vp);
1695 			} else if ((++cnt % 6) == 0) {
1696 				(*pr)(" %p,\n\t", vp);
1697 			} else {
1698 				(*pr)(" %p,", vp);
1699 			}
1700 		}
1701 		(*pr)("\n");
1702 	}
1703 }
1704 
1705 /*
1706  * List all of the locked vnodes in the system.
1707  */
1708 void printlockedvnodes(void);
1709 
1710 void
1711 printlockedvnodes(void)
1712 {
1713 	struct mount *mp;
1714 	vnode_t *vp;
1715 	vnode_impl_t *vip;
1716 
1717 	printf("Locked vnodes\n");
1718 	for (mp = _mountlist_next(NULL); mp; mp = _mountlist_next(mp)) {
1719 		TAILQ_FOREACH(vip, &mp->mnt_vnodelist, vi_mntvnodes) {
1720 			vp = VIMPL_TO_VNODE(vip);
1721 			if (VOP_ISLOCKED(vp))
1722 				vprint(NULL, vp);
1723 		}
1724 	}
1725 }
1726 
1727 #endif /* DDB || DEBUGPRINT */
1728