Home | History | Annotate | Line # | Graph | Download | only in kern
      1 /*	$NetBSD: vfs_subr.c,v 1.447 2015/06/23 10:42:34 hannken 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 1.447 2015/06/23 10:42:34 hannken 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/vnode.h>
     87 #include <sys/stat.h>
     88 #include <sys/sysctl.h>
     89 #include <sys/namei.h>
     90 #include <sys/buf.h>
     91 #include <sys/errno.h>
     92 #include <sys/kmem.h>
     93 #include <sys/syscallargs.h>
     94 #include <sys/kauth.h>
     95 #include <sys/module.h>
     96 
     97 #include <miscfs/genfs/genfs.h>
     98 #include <miscfs/specfs/specdev.h>
     99 #include <uvm/uvm_ddb.h>
    100 
    101 const enum vtype iftovt_tab[16] = {
    102 	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
    103 	VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
    104 };
    105 const int	vttoif_tab[9] = {
    106 	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
    107 	S_IFSOCK, S_IFIFO, S_IFMT,
    108 };
    109 
    110 /*
    111  * Insq/Remq for the vnode usage lists.
    112  */
    113 #define	bufinsvn(bp, dp)	LIST_INSERT_HEAD(dp, bp, b_vnbufs)
    114 #define	bufremvn(bp) {							\
    115 	LIST_REMOVE(bp, b_vnbufs);					\
    116 	(bp)->b_vnbufs.le_next = NOLIST;				\
    117 }
    118 
    119 int doforce = 1;		/* 1 => permit forcible unmounting */
    120 int prtactive = 0;		/* 1 => print out reclaim of active vnodes */
    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, 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, 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) 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 kmutex_t		syncer_mutex;	/* used to freeze syncer, long term */
    571 static kmutex_t		syncer_data_lock; /* short term lock on data structs */
    572 
    573 static int		syncer_delayno = 0;
    574 static long		syncer_last;
    575 static synclist_t *	syncer_workitem_pending;
    576 
    577 static void
    578 vn_initialize_syncerd(void)
    579 {
    580 	int i;
    581 
    582 	syncer_last = SYNCER_MAXDELAY + 2;
    583 
    584 	sysctl_vfs_syncfs_setup(NULL);
    585 
    586 	syncer_workitem_pending =
    587 	    kmem_alloc(syncer_last * sizeof (struct synclist), KM_SLEEP);
    588 
    589 	for (i = 0; i < syncer_last; i++)
    590 		TAILQ_INIT(&syncer_workitem_pending[i]);
    591 
    592 	mutex_init(&syncer_mutex, MUTEX_DEFAULT, IPL_NONE);
    593 	mutex_init(&syncer_data_lock, MUTEX_DEFAULT, IPL_NONE);
    594 }
    595 
    596 /*
    597  * Return delay factor appropriate for the given file system.   For
    598  * WAPBL we use the sync vnode to burst out metadata updates: sync
    599  * those file systems more frequently.
    600  */
    601 static inline int
    602 sync_delay(struct mount *mp)
    603 {
    604 
    605 	return mp->mnt_wapbl != NULL ? metadelay : syncdelay;
    606 }
    607 
    608 /*
    609  * Compute the next slot index from delay.
    610  */
    611 static inline int
    612 sync_delay_slot(int delayx)
    613 {
    614 
    615 	if (delayx > syncer_maxdelay - 2)
    616 		delayx = syncer_maxdelay - 2;
    617 	return (syncer_delayno + delayx) % syncer_last;
    618 }
    619 
    620 /*
    621  * Add an item to the syncer work queue.
    622  */
    623 static void
    624 vn_syncer_add1(struct vnode *vp, int delayx)
    625 {
    626 	synclist_t *slp;
    627 
    628 	KASSERT(mutex_owned(&syncer_data_lock));
    629 
    630 	if (vp->v_iflag & VI_ONWORKLST) {
    631 		/*
    632 		 * Remove in order to adjust the position of the vnode.
    633 		 * Note: called from sched_sync(), which will not hold
    634 		 * interlock, therefore we cannot modify v_iflag here.
    635 		 */
    636 		slp = &syncer_workitem_pending[vp->v_synclist_slot];
    637 		TAILQ_REMOVE(slp, vp, v_synclist);
    638 	} else {
    639 		KASSERT(mutex_owned(vp->v_interlock));
    640 		vp->v_iflag |= VI_ONWORKLST;
    641 	}
    642 
    643 	vp->v_synclist_slot = sync_delay_slot(delayx);
    644 
    645 	slp = &syncer_workitem_pending[vp->v_synclist_slot];
    646 	TAILQ_INSERT_TAIL(slp, vp, v_synclist);
    647 }
    648 
    649 void
    650 vn_syncer_add_to_worklist(struct vnode *vp, int delayx)
    651 {
    652 
    653 	KASSERT(mutex_owned(vp->v_interlock));
    654 
    655 	mutex_enter(&syncer_data_lock);
    656 	vn_syncer_add1(vp, delayx);
    657 	mutex_exit(&syncer_data_lock);
    658 }
    659 
    660 /*
    661  * Remove an item from the syncer work queue.
    662  */
    663 void
    664 vn_syncer_remove_from_worklist(struct vnode *vp)
    665 {
    666 	synclist_t *slp;
    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[vp->v_synclist_slot];
    674 		TAILQ_REMOVE(slp, vp, v_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 (vget(vp, LK_NOWAIT, false /* !wait */) == 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 	synclist_t *slp;
    760 	struct vnode *vp;
    761 	struct mount *mp, *nmp;
    762 	time_t starttime;
    763 	bool synced;
    764 
    765 	for (;;) {
    766 		mutex_enter(&syncer_mutex);
    767 
    768 		starttime = time_second;
    769 
    770 		/*
    771 		 * Sync mounts whose dirty time has expired.
    772 		 */
    773 		mutex_enter(&mountlist_lock);
    774 		for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
    775 			if ((mp->mnt_iflag & IMNT_ONWORKLIST) == 0 ||
    776 			    mp->mnt_synclist_slot != syncer_delayno) {
    777 				nmp = TAILQ_NEXT(mp, mnt_list);
    778 				continue;
    779 			}
    780 			mp->mnt_synclist_slot = sync_delay_slot(sync_delay(mp));
    781 			if (vfs_busy(mp, &nmp))
    782 				continue;
    783 			VFS_SYNC(mp, MNT_LAZY, curlwp->l_cred);
    784 			vfs_unbusy(mp, false, &nmp);
    785 		}
    786 		mutex_exit(&mountlist_lock);
    787 
    788 		mutex_enter(&syncer_data_lock);
    789 
    790 		/*
    791 		 * Push files whose dirty time has expired.
    792 		 */
    793 		slp = &syncer_workitem_pending[syncer_delayno];
    794 		syncer_delayno += 1;
    795 		if (syncer_delayno >= syncer_last)
    796 			syncer_delayno = 0;
    797 
    798 		while ((vp = TAILQ_FIRST(slp)) != NULL) {
    799 			synced = lazy_sync_vnode(vp);
    800 
    801 			/*
    802 			 * XXX The vnode may have been recycled, in which
    803 			 * case it may have a new identity.
    804 			 */
    805 			if (TAILQ_FIRST(slp) == vp) {
    806 				/*
    807 				 * Put us back on the worklist.  The worklist
    808 				 * routine will remove us from our current
    809 				 * position and then add us back in at a later
    810 				 * position.
    811 				 *
    812 				 * Try again sooner rather than later if
    813 				 * we were unable to lock the vnode.  Lock
    814 				 * failure should not prevent us from doing
    815 				 * the sync "soon".
    816 				 *
    817 				 * If we locked it yet arrive here, it's
    818 				 * likely that lazy sync is in progress and
    819 				 * so the vnode still has dirty metadata.
    820 				 * syncdelay is mainly to get this vnode out
    821 				 * of the way so we do not consider it again
    822 				 * "soon" in this loop, so the delay time is
    823 				 * not critical as long as it is not "soon".
    824 				 * While write-back strategy is the file
    825 				 * system's domain, we expect write-back to
    826 				 * occur no later than syncdelay seconds
    827 				 * into the future.
    828 				 */
    829 				vn_syncer_add1(vp,
    830 				    synced ? syncdelay : lockdelay);
    831 			}
    832 		}
    833 		mutex_exit(&syncer_mutex);
    834 
    835 		/*
    836 		 * If it has taken us less than a second to process the
    837 		 * current work, then wait.  Otherwise start right over
    838 		 * again.  We can still lose time if any single round
    839 		 * takes more than two seconds, but it does not really
    840 		 * matter as we are just trying to generally pace the
    841 		 * filesystem activity.
    842 		 */
    843 		if (time_second == starttime) {
    844 			kpause("syncer", false, hz, &syncer_data_lock);
    845 		}
    846 		mutex_exit(&syncer_data_lock);
    847 	}
    848 }
    849 
    850 static void
    851 sysctl_vfs_syncfs_setup(struct sysctllog **clog)
    852 {
    853 	const struct sysctlnode *rnode, *cnode;
    854 
    855 	sysctl_createv(clog, 0, NULL, &rnode,
    856 			CTLFLAG_PERMANENT,
    857 			CTLTYPE_NODE, "sync",
    858 			SYSCTL_DESCR("syncer options"),
    859 			NULL, 0, NULL, 0,
    860 			CTL_VFS, CTL_CREATE, CTL_EOL);
    861 
    862 	sysctl_createv(clog, 0, &rnode, &cnode,
    863 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
    864 			CTLTYPE_QUAD, "delay",
    865 			SYSCTL_DESCR("max time to delay syncing data"),
    866 			NULL, 0, &syncdelay, 0,
    867 			CTL_CREATE, CTL_EOL);
    868 
    869 	sysctl_createv(clog, 0, &rnode, &cnode,
    870 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
    871 			CTLTYPE_QUAD, "filedelay",
    872 			SYSCTL_DESCR("time to delay syncing files"),
    873 			NULL, 0, &filedelay, 0,
    874 			CTL_CREATE, CTL_EOL);
    875 
    876 	sysctl_createv(clog, 0, &rnode, &cnode,
    877 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
    878 			CTLTYPE_QUAD, "dirdelay",
    879 			SYSCTL_DESCR("time to delay syncing directories"),
    880 			NULL, 0, &dirdelay, 0,
    881 			CTL_CREATE, CTL_EOL);
    882 
    883 	sysctl_createv(clog, 0, &rnode, &cnode,
    884 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
    885 			CTLTYPE_QUAD, "metadelay",
    886 			SYSCTL_DESCR("time to delay syncing metadata"),
    887 			NULL, 0, &metadelay, 0,
    888 			CTL_CREATE, CTL_EOL);
    889 }
    890 
    891 /*
    892  * sysctl helper routine to return list of supported fstypes
    893  */
    894 int
    895 sysctl_vfs_generic_fstypes(SYSCTLFN_ARGS)
    896 {
    897 	char bf[sizeof(((struct statvfs *)NULL)->f_fstypename)];
    898 	char *where = oldp;
    899 	struct vfsops *v;
    900 	size_t needed, left, slen;
    901 	int error, first;
    902 
    903 	if (newp != NULL)
    904 		return (EPERM);
    905 	if (namelen != 0)
    906 		return (EINVAL);
    907 
    908 	first = 1;
    909 	error = 0;
    910 	needed = 0;
    911 	left = *oldlenp;
    912 
    913 	sysctl_unlock();
    914 	mutex_enter(&vfs_list_lock);
    915 	LIST_FOREACH(v, &vfs_list, vfs_list) {
    916 		if (where == NULL)
    917 			needed += strlen(v->vfs_name) + 1;
    918 		else {
    919 			memset(bf, 0, sizeof(bf));
    920 			if (first) {
    921 				strncpy(bf, v->vfs_name, sizeof(bf));
    922 				first = 0;
    923 			} else {
    924 				bf[0] = ' ';
    925 				strncpy(bf + 1, v->vfs_name, sizeof(bf) - 1);
    926 			}
    927 			bf[sizeof(bf)-1] = '\0';
    928 			slen = strlen(bf);
    929 			if (left < slen + 1)
    930 				break;
    931 			v->vfs_refcount++;
    932 			mutex_exit(&vfs_list_lock);
    933 			/* +1 to copy out the trailing NUL byte */
    934 			error = copyout(bf, where, slen + 1);
    935 			mutex_enter(&vfs_list_lock);
    936 			v->vfs_refcount--;
    937 			if (error)
    938 				break;
    939 			where += slen;
    940 			needed += slen;
    941 			left -= slen;
    942 		}
    943 	}
    944 	mutex_exit(&vfs_list_lock);
    945 	sysctl_relock();
    946 	*oldlenp = needed;
    947 	return (error);
    948 }
    949 
    950 int kinfo_vdebug = 1;
    951 int kinfo_vgetfailed;
    952 
    953 #define KINFO_VNODESLOP	10
    954 
    955 /*
    956  * Dump vnode list (via sysctl).
    957  * Copyout address of vnode followed by vnode.
    958  */
    959 int
    960 sysctl_kern_vnode(SYSCTLFN_ARGS)
    961 {
    962 	char *where = oldp;
    963 	size_t *sizep = oldlenp;
    964 	struct mount *mp, *nmp;
    965 	vnode_t *vp, vbuf;
    966 	struct vnode_iterator *marker;
    967 	char *bp = where;
    968 	char *ewhere;
    969 	int error;
    970 
    971 	if (namelen != 0)
    972 		return (EOPNOTSUPP);
    973 	if (newp != NULL)
    974 		return (EPERM);
    975 
    976 #define VPTRSZ	sizeof(vnode_t *)
    977 #define VNODESZ	sizeof(vnode_t)
    978 	if (where == NULL) {
    979 		*sizep = (numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ);
    980 		return (0);
    981 	}
    982 	ewhere = where + *sizep;
    983 
    984 	sysctl_unlock();
    985 	mutex_enter(&mountlist_lock);
    986 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
    987 		if (vfs_busy(mp, &nmp)) {
    988 			continue;
    989 		}
    990 		vfs_vnode_iterator_init(mp, &marker);
    991 		while ((vp = vfs_vnode_iterator_next(marker, NULL, NULL))) {
    992 			if (bp + VPTRSZ + VNODESZ > ewhere) {
    993 				vrele(vp);
    994 				vfs_vnode_iterator_destroy(marker);
    995 				vfs_unbusy(mp, false, NULL);
    996 				sysctl_relock();
    997 				*sizep = bp - where;
    998 				return (ENOMEM);
    999 			}
   1000 			memcpy(&vbuf, vp, VNODESZ);
   1001 			if ((error = copyout(&vp, bp, VPTRSZ)) ||
   1002 			    (error = copyout(&vbuf, bp + VPTRSZ, VNODESZ))) {
   1003 				vrele(vp);
   1004 				vfs_vnode_iterator_destroy(marker);
   1005 				vfs_unbusy(mp, false, NULL);
   1006 				sysctl_relock();
   1007 				return (error);
   1008 			}
   1009 			vrele(vp);
   1010 			bp += VPTRSZ + VNODESZ;
   1011 		}
   1012 		vfs_vnode_iterator_destroy(marker);
   1013 		vfs_unbusy(mp, false, &nmp);
   1014 	}
   1015 	mutex_exit(&mountlist_lock);
   1016 	sysctl_relock();
   1017 
   1018 	*sizep = bp - where;
   1019 	return (0);
   1020 }
   1021 
   1022 /*
   1023  * Set vnode attributes to VNOVAL
   1024  */
   1025 void
   1026 vattr_null(struct vattr *vap)
   1027 {
   1028 
   1029 	memset(vap, 0, sizeof(*vap));
   1030 
   1031 	vap->va_type = VNON;
   1032 
   1033 	/*
   1034 	 * Assign individually so that it is safe even if size and
   1035 	 * sign of each member are varied.
   1036 	 */
   1037 	vap->va_mode = VNOVAL;
   1038 	vap->va_nlink = VNOVAL;
   1039 	vap->va_uid = VNOVAL;
   1040 	vap->va_gid = VNOVAL;
   1041 	vap->va_fsid = VNOVAL;
   1042 	vap->va_fileid = VNOVAL;
   1043 	vap->va_size = VNOVAL;
   1044 	vap->va_blocksize = VNOVAL;
   1045 	vap->va_atime.tv_sec =
   1046 	    vap->va_mtime.tv_sec =
   1047 	    vap->va_ctime.tv_sec =
   1048 	    vap->va_birthtime.tv_sec = VNOVAL;
   1049 	vap->va_atime.tv_nsec =
   1050 	    vap->va_mtime.tv_nsec =
   1051 	    vap->va_ctime.tv_nsec =
   1052 	    vap->va_birthtime.tv_nsec = VNOVAL;
   1053 	vap->va_gen = VNOVAL;
   1054 	vap->va_flags = VNOVAL;
   1055 	vap->va_rdev = VNOVAL;
   1056 	vap->va_bytes = VNOVAL;
   1057 }
   1058 
   1059 #define ARRAY_SIZE(arr) (sizeof(arr) / sizeof(arr[0]))
   1060 #define ARRAY_PRINT(idx, arr) \
   1061     ((unsigned int)(idx) < ARRAY_SIZE(arr) ? (arr)[(idx)] : "UNKNOWN")
   1062 
   1063 const char * const vnode_tags[] = { VNODE_TAGS };
   1064 const char * const vnode_types[] = { VNODE_TYPES };
   1065 const char vnode_flagbits[] = VNODE_FLAGBITS;
   1066 
   1067 /*
   1068  * Print out a description of a vnode.
   1069  */
   1070 void
   1071 vprint(const char *label, struct vnode *vp)
   1072 {
   1073 	char bf[96];
   1074 	int flag;
   1075 
   1076 	flag = vp->v_iflag | vp->v_vflag | vp->v_uflag;
   1077 	snprintb(bf, sizeof(bf), vnode_flagbits, flag);
   1078 
   1079 	if (label != NULL)
   1080 		printf("%s: ", label);
   1081 	printf("vnode @ %p, flags (%s)\n\ttag %s(%d), type %s(%d), "
   1082 	    "usecount %d, writecount %d, holdcount %d\n"
   1083 	    "\tfreelisthd %p, mount %p, data %p lock %p\n",
   1084 	    vp, bf, ARRAY_PRINT(vp->v_tag, vnode_tags), vp->v_tag,
   1085 	    ARRAY_PRINT(vp->v_type, vnode_types), vp->v_type,
   1086 	    vp->v_usecount, vp->v_writecount, vp->v_holdcnt,
   1087 	    vp->v_freelisthd, vp->v_mount, vp->v_data, &vp->v_lock);
   1088 	if (vp->v_data != NULL) {
   1089 		printf("\t");
   1090 		VOP_PRINT(vp);
   1091 	}
   1092 }
   1093 
   1094 /* Deprecated. Kept for KPI compatibility. */
   1095 int
   1096 vaccess(enum vtype type, mode_t file_mode, uid_t uid, gid_t gid,
   1097     mode_t acc_mode, kauth_cred_t cred)
   1098 {
   1099 
   1100 #ifdef DIAGNOSTIC
   1101 	printf("vaccess: deprecated interface used.\n");
   1102 #endif /* DIAGNOSTIC */
   1103 
   1104 	return kauth_authorize_vnode(cred, KAUTH_ACCESS_ACTION(acc_mode,
   1105 	    type, file_mode), NULL /* This may panic. */, NULL,
   1106 	    genfs_can_access(type, file_mode, uid, gid, acc_mode, cred));
   1107 }
   1108 
   1109 /*
   1110  * Given a file system name, look up the vfsops for that
   1111  * file system, or return NULL if file system isn't present
   1112  * in the kernel.
   1113  */
   1114 struct vfsops *
   1115 vfs_getopsbyname(const char *name)
   1116 {
   1117 	struct vfsops *v;
   1118 
   1119 	mutex_enter(&vfs_list_lock);
   1120 	LIST_FOREACH(v, &vfs_list, vfs_list) {
   1121 		if (strcmp(v->vfs_name, name) == 0)
   1122 			break;
   1123 	}
   1124 	if (v != NULL)
   1125 		v->vfs_refcount++;
   1126 	mutex_exit(&vfs_list_lock);
   1127 
   1128 	return (v);
   1129 }
   1130 
   1131 void
   1132 copy_statvfs_info(struct statvfs *sbp, const struct mount *mp)
   1133 {
   1134 	const struct statvfs *mbp;
   1135 
   1136 	if (sbp == (mbp = &mp->mnt_stat))
   1137 		return;
   1138 
   1139 	(void)memcpy(&sbp->f_fsidx, &mbp->f_fsidx, sizeof(sbp->f_fsidx));
   1140 	sbp->f_fsid = mbp->f_fsid;
   1141 	sbp->f_owner = mbp->f_owner;
   1142 	sbp->f_flag = mbp->f_flag;
   1143 	sbp->f_syncwrites = mbp->f_syncwrites;
   1144 	sbp->f_asyncwrites = mbp->f_asyncwrites;
   1145 	sbp->f_syncreads = mbp->f_syncreads;
   1146 	sbp->f_asyncreads = mbp->f_asyncreads;
   1147 	(void)memcpy(sbp->f_spare, mbp->f_spare, sizeof(mbp->f_spare));
   1148 	(void)memcpy(sbp->f_fstypename, mbp->f_fstypename,
   1149 	    sizeof(sbp->f_fstypename));
   1150 	(void)memcpy(sbp->f_mntonname, mbp->f_mntonname,
   1151 	    sizeof(sbp->f_mntonname));
   1152 	(void)memcpy(sbp->f_mntfromname, mp->mnt_stat.f_mntfromname,
   1153 	    sizeof(sbp->f_mntfromname));
   1154 	sbp->f_namemax = mbp->f_namemax;
   1155 }
   1156 
   1157 int
   1158 set_statvfs_info(const char *onp, int ukon, const char *fromp, int ukfrom,
   1159     const char *vfsname, struct mount *mp, struct lwp *l)
   1160 {
   1161 	int error;
   1162 	size_t size;
   1163 	struct statvfs *sfs = &mp->mnt_stat;
   1164 	int (*fun)(const void *, void *, size_t, size_t *);
   1165 
   1166 	(void)strlcpy(mp->mnt_stat.f_fstypename, vfsname,
   1167 	    sizeof(mp->mnt_stat.f_fstypename));
   1168 
   1169 	if (onp) {
   1170 		struct cwdinfo *cwdi = l->l_proc->p_cwdi;
   1171 		fun = (ukon == UIO_SYSSPACE) ? copystr : copyinstr;
   1172 		if (cwdi->cwdi_rdir != NULL) {
   1173 			size_t len;
   1174 			char *bp;
   1175 			char *path = PNBUF_GET();
   1176 
   1177 			bp = path + MAXPATHLEN;
   1178 			*--bp = '\0';
   1179 			rw_enter(&cwdi->cwdi_lock, RW_READER);
   1180 			error = getcwd_common(cwdi->cwdi_rdir, rootvnode, &bp,
   1181 			    path, MAXPATHLEN / 2, 0, l);
   1182 			rw_exit(&cwdi->cwdi_lock);
   1183 			if (error) {
   1184 				PNBUF_PUT(path);
   1185 				return error;
   1186 			}
   1187 
   1188 			len = strlen(bp);
   1189 			if (len > sizeof(sfs->f_mntonname) - 1)
   1190 				len = sizeof(sfs->f_mntonname) - 1;
   1191 			(void)strncpy(sfs->f_mntonname, bp, len);
   1192 			PNBUF_PUT(path);
   1193 
   1194 			if (len < sizeof(sfs->f_mntonname) - 1) {
   1195 				error = (*fun)(onp, &sfs->f_mntonname[len],
   1196 				    sizeof(sfs->f_mntonname) - len - 1, &size);
   1197 				if (error)
   1198 					return error;
   1199 				size += len;
   1200 			} else {
   1201 				size = len;
   1202 			}
   1203 		} else {
   1204 			error = (*fun)(onp, &sfs->f_mntonname,
   1205 			    sizeof(sfs->f_mntonname) - 1, &size);
   1206 			if (error)
   1207 				return error;
   1208 		}
   1209 		(void)memset(sfs->f_mntonname + size, 0,
   1210 		    sizeof(sfs->f_mntonname) - size);
   1211 	}
   1212 
   1213 	if (fromp) {
   1214 		fun = (ukfrom == UIO_SYSSPACE) ? copystr : copyinstr;
   1215 		error = (*fun)(fromp, sfs->f_mntfromname,
   1216 		    sizeof(sfs->f_mntfromname) - 1, &size);
   1217 		if (error)
   1218 			return error;
   1219 		(void)memset(sfs->f_mntfromname + size, 0,
   1220 		    sizeof(sfs->f_mntfromname) - size);
   1221 	}
   1222 	return 0;
   1223 }
   1224 
   1225 void
   1226 vfs_timestamp(struct timespec *ts)
   1227 {
   1228 
   1229 	nanotime(ts);
   1230 }
   1231 
   1232 time_t	rootfstime;			/* recorded root fs time, if known */
   1233 void
   1234 setrootfstime(time_t t)
   1235 {
   1236 	rootfstime = t;
   1237 }
   1238 
   1239 static const uint8_t vttodt_tab[ ] = {
   1240 	[VNON]	=	DT_UNKNOWN,
   1241 	[VREG]	=	DT_REG,
   1242 	[VDIR]	=	DT_DIR,
   1243 	[VBLK]	=	DT_BLK,
   1244 	[VCHR]	=	DT_CHR,
   1245 	[VLNK]	=	DT_LNK,
   1246 	[VSOCK]	=	DT_SOCK,
   1247 	[VFIFO]	=	DT_FIFO,
   1248 	[VBAD]	=	DT_UNKNOWN
   1249 };
   1250 
   1251 uint8_t
   1252 vtype2dt(enum vtype vt)
   1253 {
   1254 
   1255 	CTASSERT(VBAD == __arraycount(vttodt_tab) - 1);
   1256 	return vttodt_tab[vt];
   1257 }
   1258 
   1259 int
   1260 VFS_MOUNT(struct mount *mp, const char *a, void *b, size_t *c)
   1261 {
   1262 	int error;
   1263 
   1264 	KERNEL_LOCK(1, NULL);
   1265 	error = (*(mp->mnt_op->vfs_mount))(mp, a, b, c);
   1266 	KERNEL_UNLOCK_ONE(NULL);
   1267 
   1268 	return error;
   1269 }
   1270 
   1271 int
   1272 VFS_START(struct mount *mp, int a)
   1273 {
   1274 	int error;
   1275 
   1276 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
   1277 		KERNEL_LOCK(1, NULL);
   1278 	}
   1279 	error = (*(mp->mnt_op->vfs_start))(mp, a);
   1280 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
   1281 		KERNEL_UNLOCK_ONE(NULL);
   1282 	}
   1283 
   1284 	return error;
   1285 }
   1286 
   1287 int
   1288 VFS_UNMOUNT(struct mount *mp, int a)
   1289 {
   1290 	int error;
   1291 
   1292 	KERNEL_LOCK(1, NULL);
   1293 	error = (*(mp->mnt_op->vfs_unmount))(mp, a);
   1294 	KERNEL_UNLOCK_ONE(NULL);
   1295 
   1296 	return error;
   1297 }
   1298 
   1299 int
   1300 VFS_ROOT(struct mount *mp, struct vnode **a)
   1301 {
   1302 	int error;
   1303 
   1304 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
   1305 		KERNEL_LOCK(1, NULL);
   1306 	}
   1307 	error = (*(mp->mnt_op->vfs_root))(mp, a);
   1308 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
   1309 		KERNEL_UNLOCK_ONE(NULL);
   1310 	}
   1311 
   1312 	return error;
   1313 }
   1314 
   1315 int
   1316 VFS_QUOTACTL(struct mount *mp, struct quotactl_args *args)
   1317 {
   1318 	int error;
   1319 
   1320 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
   1321 		KERNEL_LOCK(1, NULL);
   1322 	}
   1323 	error = (*(mp->mnt_op->vfs_quotactl))(mp, args);
   1324 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
   1325 		KERNEL_UNLOCK_ONE(NULL);
   1326 	}
   1327 
   1328 	return error;
   1329 }
   1330 
   1331 int
   1332 VFS_STATVFS(struct mount *mp, struct statvfs *a)
   1333 {
   1334 	int error;
   1335 
   1336 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
   1337 		KERNEL_LOCK(1, NULL);
   1338 	}
   1339 	error = (*(mp->mnt_op->vfs_statvfs))(mp, a);
   1340 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
   1341 		KERNEL_UNLOCK_ONE(NULL);
   1342 	}
   1343 
   1344 	return error;
   1345 }
   1346 
   1347 int
   1348 VFS_SYNC(struct mount *mp, int a, struct kauth_cred *b)
   1349 {
   1350 	int error;
   1351 
   1352 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
   1353 		KERNEL_LOCK(1, NULL);
   1354 	}
   1355 	error = (*(mp->mnt_op->vfs_sync))(mp, a, b);
   1356 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
   1357 		KERNEL_UNLOCK_ONE(NULL);
   1358 	}
   1359 
   1360 	return error;
   1361 }
   1362 
   1363 int
   1364 VFS_FHTOVP(struct mount *mp, struct fid *a, struct vnode **b)
   1365 {
   1366 	int error;
   1367 
   1368 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
   1369 		KERNEL_LOCK(1, NULL);
   1370 	}
   1371 	error = (*(mp->mnt_op->vfs_fhtovp))(mp, a, b);
   1372 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
   1373 		KERNEL_UNLOCK_ONE(NULL);
   1374 	}
   1375 
   1376 	return error;
   1377 }
   1378 
   1379 int
   1380 VFS_VPTOFH(struct vnode *vp, struct fid *a, size_t *b)
   1381 {
   1382 	int error;
   1383 
   1384 	if ((vp->v_vflag & VV_MPSAFE) == 0) {
   1385 		KERNEL_LOCK(1, NULL);
   1386 	}
   1387 	error = (*(vp->v_mount->mnt_op->vfs_vptofh))(vp, a, b);
   1388 	if ((vp->v_vflag & VV_MPSAFE) == 0) {
   1389 		KERNEL_UNLOCK_ONE(NULL);
   1390 	}
   1391 
   1392 	return error;
   1393 }
   1394 
   1395 int
   1396 VFS_SNAPSHOT(struct mount *mp, struct vnode *a, struct timespec *b)
   1397 {
   1398 	int error;
   1399 
   1400 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
   1401 		KERNEL_LOCK(1, NULL);
   1402 	}
   1403 	error = (*(mp->mnt_op->vfs_snapshot))(mp, a, b);
   1404 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
   1405 		KERNEL_UNLOCK_ONE(NULL);
   1406 	}
   1407 
   1408 	return error;
   1409 }
   1410 
   1411 int
   1412 VFS_EXTATTRCTL(struct mount *mp, int a, struct vnode *b, int c, const char *d)
   1413 {
   1414 	int error;
   1415 
   1416 	KERNEL_LOCK(1, NULL);		/* XXXSMP check ffs */
   1417 	error = (*(mp->mnt_op->vfs_extattrctl))(mp, a, b, c, d);
   1418 	KERNEL_UNLOCK_ONE(NULL);	/* XXX */
   1419 
   1420 	return error;
   1421 }
   1422 
   1423 int
   1424 VFS_SUSPENDCTL(struct mount *mp, int a)
   1425 {
   1426 	int error;
   1427 
   1428 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
   1429 		KERNEL_LOCK(1, NULL);
   1430 	}
   1431 	error = (*(mp->mnt_op->vfs_suspendctl))(mp, a);
   1432 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
   1433 		KERNEL_UNLOCK_ONE(NULL);
   1434 	}
   1435 
   1436 	return error;
   1437 }
   1438 
   1439 #if defined(DDB) || defined(DEBUGPRINT)
   1440 static const char buf_flagbits[] = BUF_FLAGBITS;
   1441 
   1442 void
   1443 vfs_buf_print(struct buf *bp, int full, void (*pr)(const char *, ...))
   1444 {
   1445 	char bf[1024];
   1446 
   1447 	(*pr)("  vp %p lblkno 0x%"PRIx64" blkno 0x%"PRIx64" rawblkno 0x%"
   1448 	    PRIx64 " dev 0x%x\n",
   1449 	    bp->b_vp, bp->b_lblkno, bp->b_blkno, bp->b_rawblkno, bp->b_dev);
   1450 
   1451 	snprintb(bf, sizeof(bf),
   1452 	    buf_flagbits, bp->b_flags | bp->b_oflags | bp->b_cflags);
   1453 	(*pr)("  error %d flags 0x%s\n", bp->b_error, bf);
   1454 
   1455 	(*pr)("  bufsize 0x%lx bcount 0x%lx resid 0x%lx\n",
   1456 		  bp->b_bufsize, bp->b_bcount, bp->b_resid);
   1457 	(*pr)("  data %p saveaddr %p\n",
   1458 		  bp->b_data, bp->b_saveaddr);
   1459 	(*pr)("  iodone %p objlock %p\n", bp->b_iodone, bp->b_objlock);
   1460 }
   1461 
   1462 void
   1463 vfs_vnode_print(struct vnode *vp, int full, void (*pr)(const char *, ...))
   1464 {
   1465 	char bf[256];
   1466 
   1467 	uvm_object_printit(&vp->v_uobj, full, pr);
   1468 	snprintb(bf, sizeof(bf),
   1469 	    vnode_flagbits, vp->v_iflag | vp->v_vflag | vp->v_uflag);
   1470 	(*pr)("\nVNODE flags %s\n", bf);
   1471 	(*pr)("mp %p numoutput %d size 0x%llx writesize 0x%llx\n",
   1472 	      vp->v_mount, vp->v_numoutput, vp->v_size, vp->v_writesize);
   1473 
   1474 	(*pr)("data %p writecount %ld holdcnt %ld\n",
   1475 	      vp->v_data, vp->v_writecount, vp->v_holdcnt);
   1476 
   1477 	(*pr)("tag %s(%d) type %s(%d) mount %p typedata %p\n",
   1478 	      ARRAY_PRINT(vp->v_tag, vnode_tags), vp->v_tag,
   1479 	      ARRAY_PRINT(vp->v_type, vnode_types), vp->v_type,
   1480 	      vp->v_mount, vp->v_mountedhere);
   1481 
   1482 	(*pr)("v_lock %p\n", &vp->v_lock);
   1483 
   1484 	if (full) {
   1485 		struct buf *bp;
   1486 
   1487 		(*pr)("clean bufs:\n");
   1488 		LIST_FOREACH(bp, &vp->v_cleanblkhd, b_vnbufs) {
   1489 			(*pr)(" bp %p\n", bp);
   1490 			vfs_buf_print(bp, full, pr);
   1491 		}
   1492 
   1493 		(*pr)("dirty bufs:\n");
   1494 		LIST_FOREACH(bp, &vp->v_dirtyblkhd, b_vnbufs) {
   1495 			(*pr)(" bp %p\n", bp);
   1496 			vfs_buf_print(bp, full, pr);
   1497 		}
   1498 	}
   1499 }
   1500 
   1501 void
   1502 vfs_mount_print(struct mount *mp, int full, void (*pr)(const char *, ...))
   1503 {
   1504 	char sbuf[256];
   1505 
   1506 	(*pr)("vnodecovered = %p data = %p\n",
   1507 			mp->mnt_vnodecovered,mp->mnt_data);
   1508 
   1509 	(*pr)("fs_bshift %d dev_bshift = %d\n",
   1510 			mp->mnt_fs_bshift,mp->mnt_dev_bshift);
   1511 
   1512 	snprintb(sbuf, sizeof(sbuf), __MNT_FLAG_BITS, mp->mnt_flag);
   1513 	(*pr)("flag = %s\n", sbuf);
   1514 
   1515 	snprintb(sbuf, sizeof(sbuf), __IMNT_FLAG_BITS, mp->mnt_iflag);
   1516 	(*pr)("iflag = %s\n", sbuf);
   1517 
   1518 	(*pr)("refcnt = %d unmounting @ %p updating @ %p\n", mp->mnt_refcnt,
   1519 	    &mp->mnt_unmounting, &mp->mnt_updating);
   1520 
   1521 	(*pr)("statvfs cache:\n");
   1522 	(*pr)("\tbsize = %lu\n",mp->mnt_stat.f_bsize);
   1523 	(*pr)("\tfrsize = %lu\n",mp->mnt_stat.f_frsize);
   1524 	(*pr)("\tiosize = %lu\n",mp->mnt_stat.f_iosize);
   1525 
   1526 	(*pr)("\tblocks = %"PRIu64"\n",mp->mnt_stat.f_blocks);
   1527 	(*pr)("\tbfree = %"PRIu64"\n",mp->mnt_stat.f_bfree);
   1528 	(*pr)("\tbavail = %"PRIu64"\n",mp->mnt_stat.f_bavail);
   1529 	(*pr)("\tbresvd = %"PRIu64"\n",mp->mnt_stat.f_bresvd);
   1530 
   1531 	(*pr)("\tfiles = %"PRIu64"\n",mp->mnt_stat.f_files);
   1532 	(*pr)("\tffree = %"PRIu64"\n",mp->mnt_stat.f_ffree);
   1533 	(*pr)("\tfavail = %"PRIu64"\n",mp->mnt_stat.f_favail);
   1534 	(*pr)("\tfresvd = %"PRIu64"\n",mp->mnt_stat.f_fresvd);
   1535 
   1536 	(*pr)("\tf_fsidx = { 0x%"PRIx32", 0x%"PRIx32" }\n",
   1537 			mp->mnt_stat.f_fsidx.__fsid_val[0],
   1538 			mp->mnt_stat.f_fsidx.__fsid_val[1]);
   1539 
   1540 	(*pr)("\towner = %"PRIu32"\n",mp->mnt_stat.f_owner);
   1541 	(*pr)("\tnamemax = %lu\n",mp->mnt_stat.f_namemax);
   1542 
   1543 	snprintb(sbuf, sizeof(sbuf), __MNT_FLAG_BITS, mp->mnt_stat.f_flag);
   1544 
   1545 	(*pr)("\tflag = %s\n",sbuf);
   1546 	(*pr)("\tsyncwrites = %" PRIu64 "\n",mp->mnt_stat.f_syncwrites);
   1547 	(*pr)("\tasyncwrites = %" PRIu64 "\n",mp->mnt_stat.f_asyncwrites);
   1548 	(*pr)("\tsyncreads = %" PRIu64 "\n",mp->mnt_stat.f_syncreads);
   1549 	(*pr)("\tasyncreads = %" PRIu64 "\n",mp->mnt_stat.f_asyncreads);
   1550 	(*pr)("\tfstypename = %s\n",mp->mnt_stat.f_fstypename);
   1551 	(*pr)("\tmntonname = %s\n",mp->mnt_stat.f_mntonname);
   1552 	(*pr)("\tmntfromname = %s\n",mp->mnt_stat.f_mntfromname);
   1553 
   1554 	{
   1555 		int cnt = 0;
   1556 		struct vnode *vp;
   1557 		(*pr)("locked vnodes =");
   1558 		TAILQ_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
   1559 			if (VOP_ISLOCKED(vp)) {
   1560 				if ((++cnt % 6) == 0) {
   1561 					(*pr)(" %p,\n\t", vp);
   1562 				} else {
   1563 					(*pr)(" %p,", vp);
   1564 				}
   1565 			}
   1566 		}
   1567 		(*pr)("\n");
   1568 	}
   1569 
   1570 	if (full) {
   1571 		int cnt = 0;
   1572 		struct vnode *vp;
   1573 		(*pr)("all vnodes =");
   1574 		TAILQ_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
   1575 			if (!TAILQ_NEXT(vp, v_mntvnodes)) {
   1576 				(*pr)(" %p", vp);
   1577 			} else if ((++cnt % 6) == 0) {
   1578 				(*pr)(" %p,\n\t", vp);
   1579 			} else {
   1580 				(*pr)(" %p,", vp);
   1581 			}
   1582 		}
   1583 		(*pr)("\n", vp);
   1584 	}
   1585 }
   1586 
   1587 /*
   1588  * List all of the locked vnodes in the system.
   1589  */
   1590 void printlockedvnodes(void);
   1591 
   1592 void
   1593 printlockedvnodes(void)
   1594 {
   1595 	struct mount *mp, *nmp;
   1596 	struct vnode *vp;
   1597 
   1598 	printf("Locked vnodes\n");
   1599 	mutex_enter(&mountlist_lock);
   1600 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
   1601 		if (vfs_busy(mp, &nmp)) {
   1602 			continue;
   1603 		}
   1604 		TAILQ_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
   1605 			if (VOP_ISLOCKED(vp))
   1606 				vprint(NULL, vp);
   1607 		}
   1608 		mutex_enter(&mountlist_lock);
   1609 		vfs_unbusy(mp, false, &nmp);
   1610 	}
   1611 	mutex_exit(&mountlist_lock);
   1612 }
   1613 
   1614 #endif /* DDB || DEBUGPRINT */
   1615