Home | History | Annotate | Line # | Graph | Download | only in kern
      1 /*	$NetBSD: vfs_subr.c,v 1.448 2015/08/24 22:50:32 pooka 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 1.448 2015/08/24 22:50:32 pooka 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 #define _VFS_VNODE_PRIVATE	/* for vcache_print(). */
     80 
     81 #include <sys/param.h>
     82 #include <sys/systm.h>
     83 #include <sys/conf.h>
     84 #include <sys/dirent.h>
     85 #include <sys/filedesc.h>
     86 #include <sys/kernel.h>
     87 #include <sys/mount.h>
     88 #include <sys/vnode.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 int prtactive = 0;		/* 1 => print out reclaim of active vnodes */
    123 
    124 extern struct mount *dead_rootmount;
    125 
    126 /*
    127  * Local declarations.
    128  */
    129 
    130 static void vn_initialize_syncerd(void);
    131 
    132 /*
    133  * Initialize the vnode management data structures.
    134  */
    135 void
    136 vntblinit(void)
    137 {
    138 
    139 	vn_initialize_syncerd();
    140 	vfs_mount_sysinit();
    141 	vfs_vnode_sysinit();
    142 }
    143 
    144 /*
    145  * Flush out and invalidate all buffers associated with a vnode.
    146  * Called with the underlying vnode locked, which should prevent new dirty
    147  * buffers from being queued.
    148  */
    149 int
    150 vinvalbuf(struct vnode *vp, int flags, kauth_cred_t cred, struct lwp *l,
    151 	  bool catch_p, int slptimeo)
    152 {
    153 	struct buf *bp, *nbp;
    154 	int error;
    155 	int flushflags = PGO_ALLPAGES | PGO_FREE | PGO_SYNCIO |
    156 	    (flags & V_SAVE ? PGO_CLEANIT | PGO_RECLAIM : 0);
    157 
    158 	/* XXXUBC this doesn't look at flags or slp* */
    159 	mutex_enter(vp->v_interlock);
    160 	error = VOP_PUTPAGES(vp, 0, 0, flushflags);
    161 	if (error) {
    162 		return error;
    163 	}
    164 
    165 	if (flags & V_SAVE) {
    166 		error = VOP_FSYNC(vp, cred, FSYNC_WAIT|FSYNC_RECLAIM, 0, 0);
    167 		if (error)
    168 		        return (error);
    169 		KASSERT(LIST_EMPTY(&vp->v_dirtyblkhd));
    170 	}
    171 
    172 	mutex_enter(&bufcache_lock);
    173 restart:
    174 	for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
    175 		KASSERT(bp->b_vp == vp);
    176 		nbp = LIST_NEXT(bp, b_vnbufs);
    177 		error = bbusy(bp, catch_p, slptimeo, NULL);
    178 		if (error != 0) {
    179 			if (error == EPASSTHROUGH)
    180 				goto restart;
    181 			mutex_exit(&bufcache_lock);
    182 			return (error);
    183 		}
    184 		brelsel(bp, BC_INVAL | BC_VFLUSH);
    185 	}
    186 
    187 	for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
    188 		KASSERT(bp->b_vp == vp);
    189 		nbp = LIST_NEXT(bp, b_vnbufs);
    190 		error = bbusy(bp, catch_p, slptimeo, NULL);
    191 		if (error != 0) {
    192 			if (error == EPASSTHROUGH)
    193 				goto restart;
    194 			mutex_exit(&bufcache_lock);
    195 			return (error);
    196 		}
    197 		/*
    198 		 * XXX Since there are no node locks for NFS, I believe
    199 		 * there is a slight chance that a delayed write will
    200 		 * occur while sleeping just above, so check for it.
    201 		 */
    202 		if ((bp->b_oflags & BO_DELWRI) && (flags & V_SAVE)) {
    203 #ifdef DEBUG
    204 			printf("buffer still DELWRI\n");
    205 #endif
    206 			bp->b_cflags |= BC_BUSY | BC_VFLUSH;
    207 			mutex_exit(&bufcache_lock);
    208 			VOP_BWRITE(bp->b_vp, bp);
    209 			mutex_enter(&bufcache_lock);
    210 			goto restart;
    211 		}
    212 		brelsel(bp, BC_INVAL | BC_VFLUSH);
    213 	}
    214 
    215 #ifdef DIAGNOSTIC
    216 	if (!LIST_EMPTY(&vp->v_cleanblkhd) || !LIST_EMPTY(&vp->v_dirtyblkhd))
    217 		panic("vinvalbuf: flush failed, vp %p", vp);
    218 #endif
    219 
    220 	mutex_exit(&bufcache_lock);
    221 
    222 	return (0);
    223 }
    224 
    225 /*
    226  * Destroy any in core blocks past the truncation length.
    227  * Called with the underlying vnode locked, which should prevent new dirty
    228  * buffers from being queued.
    229  */
    230 int
    231 vtruncbuf(struct vnode *vp, daddr_t lbn, bool catch_p, int slptimeo)
    232 {
    233 	struct buf *bp, *nbp;
    234 	int error;
    235 	voff_t off;
    236 
    237 	off = round_page((voff_t)lbn << vp->v_mount->mnt_fs_bshift);
    238 	mutex_enter(vp->v_interlock);
    239 	error = VOP_PUTPAGES(vp, off, 0, PGO_FREE | PGO_SYNCIO);
    240 	if (error) {
    241 		return error;
    242 	}
    243 
    244 	mutex_enter(&bufcache_lock);
    245 restart:
    246 	for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
    247 		KASSERT(bp->b_vp == vp);
    248 		nbp = LIST_NEXT(bp, b_vnbufs);
    249 		if (bp->b_lblkno < lbn)
    250 			continue;
    251 		error = bbusy(bp, catch_p, slptimeo, NULL);
    252 		if (error != 0) {
    253 			if (error == EPASSTHROUGH)
    254 				goto restart;
    255 			mutex_exit(&bufcache_lock);
    256 			return (error);
    257 		}
    258 		brelsel(bp, BC_INVAL | BC_VFLUSH);
    259 	}
    260 
    261 	for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
    262 		KASSERT(bp->b_vp == vp);
    263 		nbp = LIST_NEXT(bp, b_vnbufs);
    264 		if (bp->b_lblkno < lbn)
    265 			continue;
    266 		error = bbusy(bp, catch_p, slptimeo, NULL);
    267 		if (error != 0) {
    268 			if (error == EPASSTHROUGH)
    269 				goto restart;
    270 			mutex_exit(&bufcache_lock);
    271 			return (error);
    272 		}
    273 		brelsel(bp, BC_INVAL | BC_VFLUSH);
    274 	}
    275 	mutex_exit(&bufcache_lock);
    276 
    277 	return (0);
    278 }
    279 
    280 /*
    281  * Flush all dirty buffers from a vnode.
    282  * Called with the underlying vnode locked, which should prevent new dirty
    283  * buffers from being queued.
    284  */
    285 int
    286 vflushbuf(struct vnode *vp, int flags)
    287 {
    288 	struct buf *bp, *nbp;
    289 	int error, pflags;
    290 	bool dirty, sync;
    291 
    292 	sync = (flags & FSYNC_WAIT) != 0;
    293 	pflags = PGO_CLEANIT | PGO_ALLPAGES |
    294 		(sync ? PGO_SYNCIO : 0) |
    295 		((flags & FSYNC_LAZY) ? PGO_LAZY : 0);
    296 	mutex_enter(vp->v_interlock);
    297 	(void) VOP_PUTPAGES(vp, 0, 0, pflags);
    298 
    299 loop:
    300 	mutex_enter(&bufcache_lock);
    301 	for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
    302 		KASSERT(bp->b_vp == vp);
    303 		nbp = LIST_NEXT(bp, b_vnbufs);
    304 		if ((bp->b_cflags & BC_BUSY))
    305 			continue;
    306 		if ((bp->b_oflags & BO_DELWRI) == 0)
    307 			panic("vflushbuf: not dirty, bp %p", bp);
    308 		bp->b_cflags |= BC_BUSY | BC_VFLUSH;
    309 		mutex_exit(&bufcache_lock);
    310 		/*
    311 		 * Wait for I/O associated with indirect blocks to complete,
    312 		 * since there is no way to quickly wait for them below.
    313 		 */
    314 		if (bp->b_vp == vp || !sync)
    315 			(void) bawrite(bp);
    316 		else {
    317 			error = bwrite(bp);
    318 			if (error)
    319 				return error;
    320 		}
    321 		goto loop;
    322 	}
    323 	mutex_exit(&bufcache_lock);
    324 
    325 	if (!sync)
    326 		return 0;
    327 
    328 	mutex_enter(vp->v_interlock);
    329 	while (vp->v_numoutput != 0)
    330 		cv_wait(&vp->v_cv, vp->v_interlock);
    331 	dirty = !LIST_EMPTY(&vp->v_dirtyblkhd);
    332 	mutex_exit(vp->v_interlock);
    333 
    334 	if (dirty) {
    335 		vprint("vflushbuf: dirty", vp);
    336 		goto loop;
    337 	}
    338 
    339 	return 0;
    340 }
    341 
    342 /*
    343  * Create a vnode for a block device.
    344  * Used for root filesystem and swap areas.
    345  * Also used for memory file system special devices.
    346  */
    347 int
    348 bdevvp(dev_t dev, vnode_t **vpp)
    349 {
    350 	struct vattr va;
    351 
    352 	vattr_null(&va);
    353 	va.va_type = VBLK;
    354 	va.va_rdev = dev;
    355 
    356 	return vcache_new(dead_rootmount, NULL, &va, NOCRED, vpp);
    357 }
    358 
    359 /*
    360  * Create a vnode for a character device.
    361  * Used for kernfs and some console handling.
    362  */
    363 int
    364 cdevvp(dev_t dev, vnode_t **vpp)
    365 {
    366 	struct vattr va;
    367 
    368 	vattr_null(&va);
    369 	va.va_type = VCHR;
    370 	va.va_rdev = dev;
    371 
    372 	return vcache_new(dead_rootmount, NULL, &va, NOCRED, vpp);
    373 }
    374 
    375 /*
    376  * Associate a buffer with a vnode.  There must already be a hold on
    377  * the vnode.
    378  */
    379 void
    380 bgetvp(struct vnode *vp, struct buf *bp)
    381 {
    382 
    383 	KASSERT(bp->b_vp == NULL);
    384 	KASSERT(bp->b_objlock == &buffer_lock);
    385 	KASSERT(mutex_owned(vp->v_interlock));
    386 	KASSERT(mutex_owned(&bufcache_lock));
    387 	KASSERT((bp->b_cflags & BC_BUSY) != 0);
    388 	KASSERT(!cv_has_waiters(&bp->b_done));
    389 
    390 	vholdl(vp);
    391 	bp->b_vp = vp;
    392 	if (vp->v_type == VBLK || vp->v_type == VCHR)
    393 		bp->b_dev = vp->v_rdev;
    394 	else
    395 		bp->b_dev = NODEV;
    396 
    397 	/*
    398 	 * Insert onto list for new vnode.
    399 	 */
    400 	bufinsvn(bp, &vp->v_cleanblkhd);
    401 	bp->b_objlock = vp->v_interlock;
    402 }
    403 
    404 /*
    405  * Disassociate a buffer from a vnode.
    406  */
    407 void
    408 brelvp(struct buf *bp)
    409 {
    410 	struct vnode *vp = bp->b_vp;
    411 
    412 	KASSERT(vp != NULL);
    413 	KASSERT(bp->b_objlock == vp->v_interlock);
    414 	KASSERT(mutex_owned(vp->v_interlock));
    415 	KASSERT(mutex_owned(&bufcache_lock));
    416 	KASSERT((bp->b_cflags & BC_BUSY) != 0);
    417 	KASSERT(!cv_has_waiters(&bp->b_done));
    418 
    419 	/*
    420 	 * Delete from old vnode list, if on one.
    421 	 */
    422 	if (LIST_NEXT(bp, b_vnbufs) != NOLIST)
    423 		bufremvn(bp);
    424 
    425 	if (vp->v_uobj.uo_npages == 0 && (vp->v_iflag & VI_ONWORKLST) &&
    426 	    LIST_FIRST(&vp->v_dirtyblkhd) == NULL) {
    427 		vp->v_iflag &= ~VI_WRMAPDIRTY;
    428 		vn_syncer_remove_from_worklist(vp);
    429 	}
    430 
    431 	bp->b_objlock = &buffer_lock;
    432 	bp->b_vp = NULL;
    433 	holdrelel(vp);
    434 }
    435 
    436 /*
    437  * Reassign a buffer from one vnode list to another.
    438  * The list reassignment must be within the same vnode.
    439  * Used to assign file specific control information
    440  * (indirect blocks) to the list to which they belong.
    441  */
    442 void
    443 reassignbuf(struct buf *bp, struct vnode *vp)
    444 {
    445 	struct buflists *listheadp;
    446 	int delayx;
    447 
    448 	KASSERT(mutex_owned(&bufcache_lock));
    449 	KASSERT(bp->b_objlock == vp->v_interlock);
    450 	KASSERT(mutex_owned(vp->v_interlock));
    451 	KASSERT((bp->b_cflags & BC_BUSY) != 0);
    452 
    453 	/*
    454 	 * Delete from old vnode list, if on one.
    455 	 */
    456 	if (LIST_NEXT(bp, b_vnbufs) != NOLIST)
    457 		bufremvn(bp);
    458 
    459 	/*
    460 	 * If dirty, put on list of dirty buffers;
    461 	 * otherwise insert onto list of clean buffers.
    462 	 */
    463 	if ((bp->b_oflags & BO_DELWRI) == 0) {
    464 		listheadp = &vp->v_cleanblkhd;
    465 		if (vp->v_uobj.uo_npages == 0 &&
    466 		    (vp->v_iflag & VI_ONWORKLST) &&
    467 		    LIST_FIRST(&vp->v_dirtyblkhd) == NULL) {
    468 			vp->v_iflag &= ~VI_WRMAPDIRTY;
    469 			vn_syncer_remove_from_worklist(vp);
    470 		}
    471 	} else {
    472 		listheadp = &vp->v_dirtyblkhd;
    473 		if ((vp->v_iflag & VI_ONWORKLST) == 0) {
    474 			switch (vp->v_type) {
    475 			case VDIR:
    476 				delayx = dirdelay;
    477 				break;
    478 			case VBLK:
    479 				if (spec_node_getmountedfs(vp) != NULL) {
    480 					delayx = metadelay;
    481 					break;
    482 				}
    483 				/* fall through */
    484 			default:
    485 				delayx = filedelay;
    486 				break;
    487 			}
    488 			if (!vp->v_mount ||
    489 			    (vp->v_mount->mnt_flag & MNT_ASYNC) == 0)
    490 				vn_syncer_add_to_worklist(vp, delayx);
    491 		}
    492 	}
    493 	bufinsvn(bp, listheadp);
    494 }
    495 
    496 /*
    497  * Lookup a vnode by device number and return it referenced.
    498  */
    499 int
    500 vfinddev(dev_t dev, enum vtype type, vnode_t **vpp)
    501 {
    502 
    503 	return (spec_node_lookup_by_dev(type, dev, vpp) == 0);
    504 }
    505 
    506 /*
    507  * Revoke all the vnodes corresponding to the specified minor number
    508  * range (endpoints inclusive) of the specified major.
    509  */
    510 void
    511 vdevgone(int maj, int minl, int minh, enum vtype type)
    512 {
    513 	vnode_t *vp;
    514 	dev_t dev;
    515 	int mn;
    516 
    517 	for (mn = minl; mn <= minh; mn++) {
    518 		dev = makedev(maj, mn);
    519 		while (spec_node_lookup_by_dev(type, dev, &vp) == 0) {
    520 			VOP_REVOKE(vp, REVOKEALL);
    521 			vrele(vp);
    522 		}
    523 	}
    524 }
    525 
    526 /*
    527  * The filesystem synchronizer mechanism - syncer.
    528  *
    529  * It is useful to delay writes of file data and filesystem metadata for
    530  * a certain amount of time so that quickly created and deleted files need
    531  * not waste disk bandwidth being created and removed.  To implement this,
    532  * vnodes are appended to a "workitem" queue.
    533  *
    534  * Most pending metadata should not wait for more than ten seconds.  Thus,
    535  * mounted on block devices are delayed only about a half the time that file
    536  * data is delayed.  Similarly, directory updates are more critical, so are
    537  * only delayed about a third the time that file data is delayed.
    538  *
    539  * There are SYNCER_MAXDELAY queues that are processed in a round-robin
    540  * manner at a rate of one each second (driven off the filesystem syner
    541  * thread). The syncer_delayno variable indicates the next queue that is
    542  * to be processed.  Items that need to be processed soon are placed in
    543  * this queue:
    544  *
    545  *	syncer_workitem_pending[syncer_delayno]
    546  *
    547  * A delay of e.g. fifteen seconds is done by placing the request fifteen
    548  * entries later in the queue:
    549  *
    550  *	syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
    551  *
    552  * Flag VI_ONWORKLST indicates that vnode is added into the queue.
    553  */
    554 
    555 #define SYNCER_MAXDELAY		32
    556 
    557 typedef TAILQ_HEAD(synclist, vnode) synclist_t;
    558 
    559 static void	vn_syncer_add1(struct vnode *, int);
    560 static void	sysctl_vfs_syncfs_setup(struct sysctllog **);
    561 
    562 /*
    563  * Defines and variables for the syncer process.
    564  */
    565 int syncer_maxdelay = SYNCER_MAXDELAY;	/* maximum delay time */
    566 time_t syncdelay = 30;			/* max time to delay syncing data */
    567 time_t filedelay = 30;			/* time to delay syncing files */
    568 time_t dirdelay  = 15;			/* time to delay syncing directories */
    569 time_t metadelay = 10;			/* time to delay syncing metadata */
    570 time_t lockdelay = 1;			/* time to delay if locking fails */
    571 
    572 kmutex_t		syncer_mutex;	/* used to freeze syncer, long term */
    573 static kmutex_t		syncer_data_lock; /* short term lock on data structs */
    574 
    575 static int		syncer_delayno = 0;
    576 static long		syncer_last;
    577 static synclist_t *	syncer_workitem_pending;
    578 
    579 static void
    580 vn_initialize_syncerd(void)
    581 {
    582 	int i;
    583 
    584 	syncer_last = SYNCER_MAXDELAY + 2;
    585 
    586 	sysctl_vfs_syncfs_setup(NULL);
    587 
    588 	syncer_workitem_pending =
    589 	    kmem_alloc(syncer_last * sizeof (struct synclist), KM_SLEEP);
    590 
    591 	for (i = 0; i < syncer_last; i++)
    592 		TAILQ_INIT(&syncer_workitem_pending[i]);
    593 
    594 	mutex_init(&syncer_mutex, MUTEX_DEFAULT, IPL_NONE);
    595 	mutex_init(&syncer_data_lock, MUTEX_DEFAULT, IPL_NONE);
    596 }
    597 
    598 /*
    599  * Return delay factor appropriate for the given file system.   For
    600  * WAPBL we use the sync vnode to burst out metadata updates: sync
    601  * those file systems more frequently.
    602  */
    603 static inline int
    604 sync_delay(struct mount *mp)
    605 {
    606 
    607 	return mp->mnt_wapbl != NULL ? metadelay : syncdelay;
    608 }
    609 
    610 /*
    611  * Compute the next slot index from delay.
    612  */
    613 static inline int
    614 sync_delay_slot(int delayx)
    615 {
    616 
    617 	if (delayx > syncer_maxdelay - 2)
    618 		delayx = syncer_maxdelay - 2;
    619 	return (syncer_delayno + delayx) % syncer_last;
    620 }
    621 
    622 /*
    623  * Add an item to the syncer work queue.
    624  */
    625 static void
    626 vn_syncer_add1(struct vnode *vp, int delayx)
    627 {
    628 	synclist_t *slp;
    629 
    630 	KASSERT(mutex_owned(&syncer_data_lock));
    631 
    632 	if (vp->v_iflag & VI_ONWORKLST) {
    633 		/*
    634 		 * Remove in order to adjust the position of the vnode.
    635 		 * Note: called from sched_sync(), which will not hold
    636 		 * interlock, therefore we cannot modify v_iflag here.
    637 		 */
    638 		slp = &syncer_workitem_pending[vp->v_synclist_slot];
    639 		TAILQ_REMOVE(slp, vp, v_synclist);
    640 	} else {
    641 		KASSERT(mutex_owned(vp->v_interlock));
    642 		vp->v_iflag |= VI_ONWORKLST;
    643 	}
    644 
    645 	vp->v_synclist_slot = sync_delay_slot(delayx);
    646 
    647 	slp = &syncer_workitem_pending[vp->v_synclist_slot];
    648 	TAILQ_INSERT_TAIL(slp, vp, v_synclist);
    649 }
    650 
    651 void
    652 vn_syncer_add_to_worklist(struct vnode *vp, int delayx)
    653 {
    654 
    655 	KASSERT(mutex_owned(vp->v_interlock));
    656 
    657 	mutex_enter(&syncer_data_lock);
    658 	vn_syncer_add1(vp, delayx);
    659 	mutex_exit(&syncer_data_lock);
    660 }
    661 
    662 /*
    663  * Remove an item from the syncer work queue.
    664  */
    665 void
    666 vn_syncer_remove_from_worklist(struct vnode *vp)
    667 {
    668 	synclist_t *slp;
    669 
    670 	KASSERT(mutex_owned(vp->v_interlock));
    671 
    672 	mutex_enter(&syncer_data_lock);
    673 	if (vp->v_iflag & VI_ONWORKLST) {
    674 		vp->v_iflag &= ~VI_ONWORKLST;
    675 		slp = &syncer_workitem_pending[vp->v_synclist_slot];
    676 		TAILQ_REMOVE(slp, vp, v_synclist);
    677 	}
    678 	mutex_exit(&syncer_data_lock);
    679 }
    680 
    681 /*
    682  * Add this mount point to the syncer.
    683  */
    684 void
    685 vfs_syncer_add_to_worklist(struct mount *mp)
    686 {
    687 	static int start, incr, next;
    688 	int vdelay;
    689 
    690 	KASSERT(mutex_owned(&mp->mnt_updating));
    691 	KASSERT((mp->mnt_iflag & IMNT_ONWORKLIST) == 0);
    692 
    693 	/*
    694 	 * We attempt to scatter the mount points on the list
    695 	 * so that they will go off at evenly distributed times
    696 	 * even if all the filesystems are mounted at once.
    697 	 */
    698 
    699 	next += incr;
    700 	if (next == 0 || next > syncer_maxdelay) {
    701 		start /= 2;
    702 		incr /= 2;
    703 		if (start == 0) {
    704 			start = syncer_maxdelay / 2;
    705 			incr = syncer_maxdelay;
    706 		}
    707 		next = start;
    708 	}
    709 	mp->mnt_iflag |= IMNT_ONWORKLIST;
    710 	vdelay = sync_delay(mp);
    711 	mp->mnt_synclist_slot = vdelay > 0 ? next % vdelay : 0;
    712 }
    713 
    714 /*
    715  * Remove the mount point from the syncer.
    716  */
    717 void
    718 vfs_syncer_remove_from_worklist(struct mount *mp)
    719 {
    720 
    721 	KASSERT(mutex_owned(&mp->mnt_updating));
    722 	KASSERT((mp->mnt_iflag & IMNT_ONWORKLIST) != 0);
    723 
    724 	mp->mnt_iflag &= ~IMNT_ONWORKLIST;
    725 }
    726 
    727 /*
    728  * Try lazy sync, return true on success.
    729  */
    730 static bool
    731 lazy_sync_vnode(struct vnode *vp)
    732 {
    733 	bool synced;
    734 
    735 	KASSERT(mutex_owned(&syncer_data_lock));
    736 
    737 	synced = false;
    738 	/* We are locking in the wrong direction. */
    739 	if (mutex_tryenter(vp->v_interlock)) {
    740 		mutex_exit(&syncer_data_lock);
    741 		if (vget(vp, LK_NOWAIT, false /* !wait */) == 0) {
    742 			if (vn_lock(vp, LK_EXCLUSIVE | LK_NOWAIT) == 0) {
    743 				synced = true;
    744 				(void) VOP_FSYNC(vp, curlwp->l_cred,
    745 				    FSYNC_LAZY, 0, 0);
    746 				vput(vp);
    747 			} else
    748 				vrele(vp);
    749 		}
    750 		mutex_enter(&syncer_data_lock);
    751 	}
    752 	return synced;
    753 }
    754 
    755 /*
    756  * System filesystem synchronizer daemon.
    757  */
    758 void
    759 sched_sync(void *arg)
    760 {
    761 	synclist_t *slp;
    762 	struct vnode *vp;
    763 	struct mount *mp, *nmp;
    764 	time_t starttime;
    765 	bool synced;
    766 
    767 	for (;;) {
    768 		mutex_enter(&syncer_mutex);
    769 
    770 		starttime = time_second;
    771 
    772 		/*
    773 		 * Sync mounts whose dirty time has expired.
    774 		 */
    775 		mutex_enter(&mountlist_lock);
    776 		for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
    777 			if ((mp->mnt_iflag & IMNT_ONWORKLIST) == 0 ||
    778 			    mp->mnt_synclist_slot != syncer_delayno) {
    779 				nmp = TAILQ_NEXT(mp, mnt_list);
    780 				continue;
    781 			}
    782 			mp->mnt_synclist_slot = sync_delay_slot(sync_delay(mp));
    783 			if (vfs_busy(mp, &nmp))
    784 				continue;
    785 			VFS_SYNC(mp, MNT_LAZY, curlwp->l_cred);
    786 			vfs_unbusy(mp, false, &nmp);
    787 		}
    788 		mutex_exit(&mountlist_lock);
    789 
    790 		mutex_enter(&syncer_data_lock);
    791 
    792 		/*
    793 		 * Push files whose dirty time has expired.
    794 		 */
    795 		slp = &syncer_workitem_pending[syncer_delayno];
    796 		syncer_delayno += 1;
    797 		if (syncer_delayno >= syncer_last)
    798 			syncer_delayno = 0;
    799 
    800 		while ((vp = TAILQ_FIRST(slp)) != NULL) {
    801 			synced = lazy_sync_vnode(vp);
    802 
    803 			/*
    804 			 * XXX The vnode may have been recycled, in which
    805 			 * case it may have a new identity.
    806 			 */
    807 			if (TAILQ_FIRST(slp) == vp) {
    808 				/*
    809 				 * Put us back on the worklist.  The worklist
    810 				 * routine will remove us from our current
    811 				 * position and then add us back in at a later
    812 				 * position.
    813 				 *
    814 				 * Try again sooner rather than later if
    815 				 * we were unable to lock the vnode.  Lock
    816 				 * failure should not prevent us from doing
    817 				 * the sync "soon".
    818 				 *
    819 				 * If we locked it yet arrive here, it's
    820 				 * likely that lazy sync is in progress and
    821 				 * so the vnode still has dirty metadata.
    822 				 * syncdelay is mainly to get this vnode out
    823 				 * of the way so we do not consider it again
    824 				 * "soon" in this loop, so the delay time is
    825 				 * not critical as long as it is not "soon".
    826 				 * While write-back strategy is the file
    827 				 * system's domain, we expect write-back to
    828 				 * occur no later than syncdelay seconds
    829 				 * into the future.
    830 				 */
    831 				vn_syncer_add1(vp,
    832 				    synced ? syncdelay : lockdelay);
    833 			}
    834 		}
    835 		mutex_exit(&syncer_mutex);
    836 
    837 		/*
    838 		 * If it has taken us less than a second to process the
    839 		 * current work, then wait.  Otherwise start right over
    840 		 * again.  We can still lose time if any single round
    841 		 * takes more than two seconds, but it does not really
    842 		 * matter as we are just trying to generally pace the
    843 		 * filesystem activity.
    844 		 */
    845 		if (time_second == starttime) {
    846 			kpause("syncer", false, hz, &syncer_data_lock);
    847 		}
    848 		mutex_exit(&syncer_data_lock);
    849 	}
    850 }
    851 
    852 static void
    853 sysctl_vfs_syncfs_setup(struct sysctllog **clog)
    854 {
    855 	const struct sysctlnode *rnode, *cnode;
    856 
    857 	sysctl_createv(clog, 0, NULL, &rnode,
    858 			CTLFLAG_PERMANENT,
    859 			CTLTYPE_NODE, "sync",
    860 			SYSCTL_DESCR("syncer options"),
    861 			NULL, 0, NULL, 0,
    862 			CTL_VFS, CTL_CREATE, CTL_EOL);
    863 
    864 	sysctl_createv(clog, 0, &rnode, &cnode,
    865 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
    866 			CTLTYPE_QUAD, "delay",
    867 			SYSCTL_DESCR("max time to delay syncing data"),
    868 			NULL, 0, &syncdelay, 0,
    869 			CTL_CREATE, CTL_EOL);
    870 
    871 	sysctl_createv(clog, 0, &rnode, &cnode,
    872 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
    873 			CTLTYPE_QUAD, "filedelay",
    874 			SYSCTL_DESCR("time to delay syncing files"),
    875 			NULL, 0, &filedelay, 0,
    876 			CTL_CREATE, CTL_EOL);
    877 
    878 	sysctl_createv(clog, 0, &rnode, &cnode,
    879 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
    880 			CTLTYPE_QUAD, "dirdelay",
    881 			SYSCTL_DESCR("time to delay syncing directories"),
    882 			NULL, 0, &dirdelay, 0,
    883 			CTL_CREATE, CTL_EOL);
    884 
    885 	sysctl_createv(clog, 0, &rnode, &cnode,
    886 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
    887 			CTLTYPE_QUAD, "metadelay",
    888 			SYSCTL_DESCR("time to delay syncing metadata"),
    889 			NULL, 0, &metadelay, 0,
    890 			CTL_CREATE, CTL_EOL);
    891 }
    892 
    893 /*
    894  * sysctl helper routine to return list of supported fstypes
    895  */
    896 int
    897 sysctl_vfs_generic_fstypes(SYSCTLFN_ARGS)
    898 {
    899 	char bf[sizeof(((struct statvfs *)NULL)->f_fstypename)];
    900 	char *where = oldp;
    901 	struct vfsops *v;
    902 	size_t needed, left, slen;
    903 	int error, first;
    904 
    905 	if (newp != NULL)
    906 		return (EPERM);
    907 	if (namelen != 0)
    908 		return (EINVAL);
    909 
    910 	first = 1;
    911 	error = 0;
    912 	needed = 0;
    913 	left = *oldlenp;
    914 
    915 	sysctl_unlock();
    916 	mutex_enter(&vfs_list_lock);
    917 	LIST_FOREACH(v, &vfs_list, vfs_list) {
    918 		if (where == NULL)
    919 			needed += strlen(v->vfs_name) + 1;
    920 		else {
    921 			memset(bf, 0, sizeof(bf));
    922 			if (first) {
    923 				strncpy(bf, v->vfs_name, sizeof(bf));
    924 				first = 0;
    925 			} else {
    926 				bf[0] = ' ';
    927 				strncpy(bf + 1, v->vfs_name, sizeof(bf) - 1);
    928 			}
    929 			bf[sizeof(bf)-1] = '\0';
    930 			slen = strlen(bf);
    931 			if (left < slen + 1)
    932 				break;
    933 			v->vfs_refcount++;
    934 			mutex_exit(&vfs_list_lock);
    935 			/* +1 to copy out the trailing NUL byte */
    936 			error = copyout(bf, where, slen + 1);
    937 			mutex_enter(&vfs_list_lock);
    938 			v->vfs_refcount--;
    939 			if (error)
    940 				break;
    941 			where += slen;
    942 			needed += slen;
    943 			left -= slen;
    944 		}
    945 	}
    946 	mutex_exit(&vfs_list_lock);
    947 	sysctl_relock();
    948 	*oldlenp = needed;
    949 	return (error);
    950 }
    951 
    952 int kinfo_vdebug = 1;
    953 int kinfo_vgetfailed;
    954 
    955 #define KINFO_VNODESLOP	10
    956 
    957 /*
    958  * Dump vnode list (via sysctl).
    959  * Copyout address of vnode followed by vnode.
    960  */
    961 int
    962 sysctl_kern_vnode(SYSCTLFN_ARGS)
    963 {
    964 	char *where = oldp;
    965 	size_t *sizep = oldlenp;
    966 	struct mount *mp, *nmp;
    967 	vnode_t *vp, vbuf;
    968 	struct vnode_iterator *marker;
    969 	char *bp = where;
    970 	char *ewhere;
    971 	int error;
    972 
    973 	if (namelen != 0)
    974 		return (EOPNOTSUPP);
    975 	if (newp != NULL)
    976 		return (EPERM);
    977 
    978 #define VPTRSZ	sizeof(vnode_t *)
    979 #define VNODESZ	sizeof(vnode_t)
    980 	if (where == NULL) {
    981 		*sizep = (numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ);
    982 		return (0);
    983 	}
    984 	ewhere = where + *sizep;
    985 
    986 	sysctl_unlock();
    987 	mutex_enter(&mountlist_lock);
    988 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
    989 		if (vfs_busy(mp, &nmp)) {
    990 			continue;
    991 		}
    992 		vfs_vnode_iterator_init(mp, &marker);
    993 		while ((vp = vfs_vnode_iterator_next(marker, NULL, NULL))) {
    994 			if (bp + VPTRSZ + VNODESZ > ewhere) {
    995 				vrele(vp);
    996 				vfs_vnode_iterator_destroy(marker);
    997 				vfs_unbusy(mp, false, NULL);
    998 				sysctl_relock();
    999 				*sizep = bp - where;
   1000 				return (ENOMEM);
   1001 			}
   1002 			memcpy(&vbuf, vp, VNODESZ);
   1003 			if ((error = copyout(&vp, bp, VPTRSZ)) ||
   1004 			    (error = copyout(&vbuf, bp + VPTRSZ, VNODESZ))) {
   1005 				vrele(vp);
   1006 				vfs_vnode_iterator_destroy(marker);
   1007 				vfs_unbusy(mp, false, NULL);
   1008 				sysctl_relock();
   1009 				return (error);
   1010 			}
   1011 			vrele(vp);
   1012 			bp += VPTRSZ + VNODESZ;
   1013 		}
   1014 		vfs_vnode_iterator_destroy(marker);
   1015 		vfs_unbusy(mp, false, &nmp);
   1016 	}
   1017 	mutex_exit(&mountlist_lock);
   1018 	sysctl_relock();
   1019 
   1020 	*sizep = bp - where;
   1021 	return (0);
   1022 }
   1023 
   1024 /*
   1025  * Set vnode attributes to VNOVAL
   1026  */
   1027 void
   1028 vattr_null(struct vattr *vap)
   1029 {
   1030 
   1031 	memset(vap, 0, sizeof(*vap));
   1032 
   1033 	vap->va_type = VNON;
   1034 
   1035 	/*
   1036 	 * Assign individually so that it is safe even if size and
   1037 	 * sign of each member are varied.
   1038 	 */
   1039 	vap->va_mode = VNOVAL;
   1040 	vap->va_nlink = VNOVAL;
   1041 	vap->va_uid = VNOVAL;
   1042 	vap->va_gid = VNOVAL;
   1043 	vap->va_fsid = VNOVAL;
   1044 	vap->va_fileid = VNOVAL;
   1045 	vap->va_size = VNOVAL;
   1046 	vap->va_blocksize = VNOVAL;
   1047 	vap->va_atime.tv_sec =
   1048 	    vap->va_mtime.tv_sec =
   1049 	    vap->va_ctime.tv_sec =
   1050 	    vap->va_birthtime.tv_sec = VNOVAL;
   1051 	vap->va_atime.tv_nsec =
   1052 	    vap->va_mtime.tv_nsec =
   1053 	    vap->va_ctime.tv_nsec =
   1054 	    vap->va_birthtime.tv_nsec = VNOVAL;
   1055 	vap->va_gen = VNOVAL;
   1056 	vap->va_flags = VNOVAL;
   1057 	vap->va_rdev = VNOVAL;
   1058 	vap->va_bytes = VNOVAL;
   1059 }
   1060 
   1061 #define ARRAY_SIZE(arr) (sizeof(arr) / sizeof(arr[0]))
   1062 #define ARRAY_PRINT(idx, arr) \
   1063     ((unsigned int)(idx) < ARRAY_SIZE(arr) ? (arr)[(idx)] : "UNKNOWN")
   1064 
   1065 const char * const vnode_tags[] = { VNODE_TAGS };
   1066 const char * const vnode_types[] = { VNODE_TYPES };
   1067 const char vnode_flagbits[] = VNODE_FLAGBITS;
   1068 
   1069 /*
   1070  * Print out a description of a vnode.
   1071  */
   1072 void
   1073 vprint(const char *label, struct vnode *vp)
   1074 {
   1075 	char bf[96];
   1076 	int flag;
   1077 
   1078 	flag = vp->v_iflag | vp->v_vflag | vp->v_uflag;
   1079 	snprintb(bf, sizeof(bf), vnode_flagbits, flag);
   1080 
   1081 	if (label != NULL)
   1082 		printf("%s: ", label);
   1083 	printf("vnode @ %p, flags (%s)\n\ttag %s(%d), type %s(%d), "
   1084 	    "usecount %d, writecount %d, holdcount %d\n"
   1085 	    "\tfreelisthd %p, mount %p, data %p lock %p\n",
   1086 	    vp, bf, ARRAY_PRINT(vp->v_tag, vnode_tags), vp->v_tag,
   1087 	    ARRAY_PRINT(vp->v_type, vnode_types), vp->v_type,
   1088 	    vp->v_usecount, vp->v_writecount, vp->v_holdcnt,
   1089 	    vp->v_freelisthd, vp->v_mount, vp->v_data, &vp->v_lock);
   1090 	vcache_print(vp, "\t", printf);
   1091 	if (vp->v_data != NULL) {
   1092 		printf("\t");
   1093 		VOP_PRINT(vp);
   1094 	}
   1095 }
   1096 
   1097 /* Deprecated. Kept for KPI compatibility. */
   1098 int
   1099 vaccess(enum vtype type, mode_t file_mode, uid_t uid, gid_t gid,
   1100     mode_t acc_mode, kauth_cred_t cred)
   1101 {
   1102 
   1103 #ifdef DIAGNOSTIC
   1104 	printf("vaccess: deprecated interface used.\n");
   1105 #endif /* DIAGNOSTIC */
   1106 
   1107 	return kauth_authorize_vnode(cred, KAUTH_ACCESS_ACTION(acc_mode,
   1108 	    type, file_mode), NULL /* This may panic. */, NULL,
   1109 	    genfs_can_access(type, file_mode, uid, gid, acc_mode, cred));
   1110 }
   1111 
   1112 /*
   1113  * Given a file system name, look up the vfsops for that
   1114  * file system, or return NULL if file system isn't present
   1115  * in the kernel.
   1116  */
   1117 struct vfsops *
   1118 vfs_getopsbyname(const char *name)
   1119 {
   1120 	struct vfsops *v;
   1121 
   1122 	mutex_enter(&vfs_list_lock);
   1123 	LIST_FOREACH(v, &vfs_list, vfs_list) {
   1124 		if (strcmp(v->vfs_name, name) == 0)
   1125 			break;
   1126 	}
   1127 	if (v != NULL)
   1128 		v->vfs_refcount++;
   1129 	mutex_exit(&vfs_list_lock);
   1130 
   1131 	return (v);
   1132 }
   1133 
   1134 void
   1135 copy_statvfs_info(struct statvfs *sbp, const struct mount *mp)
   1136 {
   1137 	const struct statvfs *mbp;
   1138 
   1139 	if (sbp == (mbp = &mp->mnt_stat))
   1140 		return;
   1141 
   1142 	(void)memcpy(&sbp->f_fsidx, &mbp->f_fsidx, sizeof(sbp->f_fsidx));
   1143 	sbp->f_fsid = mbp->f_fsid;
   1144 	sbp->f_owner = mbp->f_owner;
   1145 	sbp->f_flag = mbp->f_flag;
   1146 	sbp->f_syncwrites = mbp->f_syncwrites;
   1147 	sbp->f_asyncwrites = mbp->f_asyncwrites;
   1148 	sbp->f_syncreads = mbp->f_syncreads;
   1149 	sbp->f_asyncreads = mbp->f_asyncreads;
   1150 	(void)memcpy(sbp->f_spare, mbp->f_spare, sizeof(mbp->f_spare));
   1151 	(void)memcpy(sbp->f_fstypename, mbp->f_fstypename,
   1152 	    sizeof(sbp->f_fstypename));
   1153 	(void)memcpy(sbp->f_mntonname, mbp->f_mntonname,
   1154 	    sizeof(sbp->f_mntonname));
   1155 	(void)memcpy(sbp->f_mntfromname, mp->mnt_stat.f_mntfromname,
   1156 	    sizeof(sbp->f_mntfromname));
   1157 	sbp->f_namemax = mbp->f_namemax;
   1158 }
   1159 
   1160 int
   1161 set_statvfs_info(const char *onp, int ukon, const char *fromp, int ukfrom,
   1162     const char *vfsname, struct mount *mp, struct lwp *l)
   1163 {
   1164 	int error;
   1165 	size_t size;
   1166 	struct statvfs *sfs = &mp->mnt_stat;
   1167 	int (*fun)(const void *, void *, size_t, size_t *);
   1168 
   1169 	(void)strlcpy(mp->mnt_stat.f_fstypename, vfsname,
   1170 	    sizeof(mp->mnt_stat.f_fstypename));
   1171 
   1172 	if (onp) {
   1173 		struct cwdinfo *cwdi = l->l_proc->p_cwdi;
   1174 		fun = (ukon == UIO_SYSSPACE) ? copystr : copyinstr;
   1175 		if (cwdi->cwdi_rdir != NULL) {
   1176 			size_t len;
   1177 			char *bp;
   1178 			char *path = PNBUF_GET();
   1179 
   1180 			bp = path + MAXPATHLEN;
   1181 			*--bp = '\0';
   1182 			rw_enter(&cwdi->cwdi_lock, RW_READER);
   1183 			error = getcwd_common(cwdi->cwdi_rdir, rootvnode, &bp,
   1184 			    path, MAXPATHLEN / 2, 0, l);
   1185 			rw_exit(&cwdi->cwdi_lock);
   1186 			if (error) {
   1187 				PNBUF_PUT(path);
   1188 				return error;
   1189 			}
   1190 
   1191 			len = strlen(bp);
   1192 			if (len > sizeof(sfs->f_mntonname) - 1)
   1193 				len = sizeof(sfs->f_mntonname) - 1;
   1194 			(void)strncpy(sfs->f_mntonname, bp, len);
   1195 			PNBUF_PUT(path);
   1196 
   1197 			if (len < sizeof(sfs->f_mntonname) - 1) {
   1198 				error = (*fun)(onp, &sfs->f_mntonname[len],
   1199 				    sizeof(sfs->f_mntonname) - len - 1, &size);
   1200 				if (error)
   1201 					return error;
   1202 				size += len;
   1203 			} else {
   1204 				size = len;
   1205 			}
   1206 		} else {
   1207 			error = (*fun)(onp, &sfs->f_mntonname,
   1208 			    sizeof(sfs->f_mntonname) - 1, &size);
   1209 			if (error)
   1210 				return error;
   1211 		}
   1212 		(void)memset(sfs->f_mntonname + size, 0,
   1213 		    sizeof(sfs->f_mntonname) - size);
   1214 	}
   1215 
   1216 	if (fromp) {
   1217 		fun = (ukfrom == UIO_SYSSPACE) ? copystr : copyinstr;
   1218 		error = (*fun)(fromp, sfs->f_mntfromname,
   1219 		    sizeof(sfs->f_mntfromname) - 1, &size);
   1220 		if (error)
   1221 			return error;
   1222 		(void)memset(sfs->f_mntfromname + size, 0,
   1223 		    sizeof(sfs->f_mntfromname) - size);
   1224 	}
   1225 	return 0;
   1226 }
   1227 
   1228 void
   1229 vfs_timestamp(struct timespec *ts)
   1230 {
   1231 
   1232 	nanotime(ts);
   1233 }
   1234 
   1235 time_t	rootfstime;			/* recorded root fs time, if known */
   1236 void
   1237 setrootfstime(time_t t)
   1238 {
   1239 	rootfstime = t;
   1240 }
   1241 
   1242 static const uint8_t vttodt_tab[ ] = {
   1243 	[VNON]	=	DT_UNKNOWN,
   1244 	[VREG]	=	DT_REG,
   1245 	[VDIR]	=	DT_DIR,
   1246 	[VBLK]	=	DT_BLK,
   1247 	[VCHR]	=	DT_CHR,
   1248 	[VLNK]	=	DT_LNK,
   1249 	[VSOCK]	=	DT_SOCK,
   1250 	[VFIFO]	=	DT_FIFO,
   1251 	[VBAD]	=	DT_UNKNOWN
   1252 };
   1253 
   1254 uint8_t
   1255 vtype2dt(enum vtype vt)
   1256 {
   1257 
   1258 	CTASSERT(VBAD == __arraycount(vttodt_tab) - 1);
   1259 	return vttodt_tab[vt];
   1260 }
   1261 
   1262 int
   1263 VFS_MOUNT(struct mount *mp, const char *a, void *b, size_t *c)
   1264 {
   1265 	int error;
   1266 
   1267 	KERNEL_LOCK(1, NULL);
   1268 	error = (*(mp->mnt_op->vfs_mount))(mp, a, b, c);
   1269 	KERNEL_UNLOCK_ONE(NULL);
   1270 
   1271 	return error;
   1272 }
   1273 
   1274 int
   1275 VFS_START(struct mount *mp, int a)
   1276 {
   1277 	int error;
   1278 
   1279 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
   1280 		KERNEL_LOCK(1, NULL);
   1281 	}
   1282 	error = (*(mp->mnt_op->vfs_start))(mp, a);
   1283 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
   1284 		KERNEL_UNLOCK_ONE(NULL);
   1285 	}
   1286 
   1287 	return error;
   1288 }
   1289 
   1290 int
   1291 VFS_UNMOUNT(struct mount *mp, int a)
   1292 {
   1293 	int error;
   1294 
   1295 	KERNEL_LOCK(1, NULL);
   1296 	error = (*(mp->mnt_op->vfs_unmount))(mp, a);
   1297 	KERNEL_UNLOCK_ONE(NULL);
   1298 
   1299 	return error;
   1300 }
   1301 
   1302 int
   1303 VFS_ROOT(struct mount *mp, struct vnode **a)
   1304 {
   1305 	int error;
   1306 
   1307 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
   1308 		KERNEL_LOCK(1, NULL);
   1309 	}
   1310 	error = (*(mp->mnt_op->vfs_root))(mp, a);
   1311 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
   1312 		KERNEL_UNLOCK_ONE(NULL);
   1313 	}
   1314 
   1315 	return error;
   1316 }
   1317 
   1318 int
   1319 VFS_QUOTACTL(struct mount *mp, struct quotactl_args *args)
   1320 {
   1321 	int error;
   1322 
   1323 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
   1324 		KERNEL_LOCK(1, NULL);
   1325 	}
   1326 	error = (*(mp->mnt_op->vfs_quotactl))(mp, args);
   1327 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
   1328 		KERNEL_UNLOCK_ONE(NULL);
   1329 	}
   1330 
   1331 	return error;
   1332 }
   1333 
   1334 int
   1335 VFS_STATVFS(struct mount *mp, struct statvfs *a)
   1336 {
   1337 	int error;
   1338 
   1339 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
   1340 		KERNEL_LOCK(1, NULL);
   1341 	}
   1342 	error = (*(mp->mnt_op->vfs_statvfs))(mp, a);
   1343 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
   1344 		KERNEL_UNLOCK_ONE(NULL);
   1345 	}
   1346 
   1347 	return error;
   1348 }
   1349 
   1350 int
   1351 VFS_SYNC(struct mount *mp, int a, struct kauth_cred *b)
   1352 {
   1353 	int error;
   1354 
   1355 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
   1356 		KERNEL_LOCK(1, NULL);
   1357 	}
   1358 	error = (*(mp->mnt_op->vfs_sync))(mp, a, b);
   1359 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
   1360 		KERNEL_UNLOCK_ONE(NULL);
   1361 	}
   1362 
   1363 	return error;
   1364 }
   1365 
   1366 int
   1367 VFS_FHTOVP(struct mount *mp, struct fid *a, struct vnode **b)
   1368 {
   1369 	int error;
   1370 
   1371 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
   1372 		KERNEL_LOCK(1, NULL);
   1373 	}
   1374 	error = (*(mp->mnt_op->vfs_fhtovp))(mp, a, b);
   1375 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
   1376 		KERNEL_UNLOCK_ONE(NULL);
   1377 	}
   1378 
   1379 	return error;
   1380 }
   1381 
   1382 int
   1383 VFS_VPTOFH(struct vnode *vp, struct fid *a, size_t *b)
   1384 {
   1385 	int error;
   1386 
   1387 	if ((vp->v_vflag & VV_MPSAFE) == 0) {
   1388 		KERNEL_LOCK(1, NULL);
   1389 	}
   1390 	error = (*(vp->v_mount->mnt_op->vfs_vptofh))(vp, a, b);
   1391 	if ((vp->v_vflag & VV_MPSAFE) == 0) {
   1392 		KERNEL_UNLOCK_ONE(NULL);
   1393 	}
   1394 
   1395 	return error;
   1396 }
   1397 
   1398 int
   1399 VFS_SNAPSHOT(struct mount *mp, struct vnode *a, struct timespec *b)
   1400 {
   1401 	int error;
   1402 
   1403 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
   1404 		KERNEL_LOCK(1, NULL);
   1405 	}
   1406 	error = (*(mp->mnt_op->vfs_snapshot))(mp, a, b);
   1407 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
   1408 		KERNEL_UNLOCK_ONE(NULL);
   1409 	}
   1410 
   1411 	return error;
   1412 }
   1413 
   1414 int
   1415 VFS_EXTATTRCTL(struct mount *mp, int a, struct vnode *b, int c, const char *d)
   1416 {
   1417 	int error;
   1418 
   1419 	KERNEL_LOCK(1, NULL);		/* XXXSMP check ffs */
   1420 	error = (*(mp->mnt_op->vfs_extattrctl))(mp, a, b, c, d);
   1421 	KERNEL_UNLOCK_ONE(NULL);	/* XXX */
   1422 
   1423 	return error;
   1424 }
   1425 
   1426 int
   1427 VFS_SUSPENDCTL(struct mount *mp, int a)
   1428 {
   1429 	int error;
   1430 
   1431 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
   1432 		KERNEL_LOCK(1, NULL);
   1433 	}
   1434 	error = (*(mp->mnt_op->vfs_suspendctl))(mp, a);
   1435 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
   1436 		KERNEL_UNLOCK_ONE(NULL);
   1437 	}
   1438 
   1439 	return error;
   1440 }
   1441 
   1442 #if defined(DDB) || defined(DEBUGPRINT)
   1443 static const char buf_flagbits[] = BUF_FLAGBITS;
   1444 
   1445 void
   1446 vfs_buf_print(struct buf *bp, int full, void (*pr)(const char *, ...))
   1447 {
   1448 	char bf[1024];
   1449 
   1450 	(*pr)("  vp %p lblkno 0x%"PRIx64" blkno 0x%"PRIx64" rawblkno 0x%"
   1451 	    PRIx64 " dev 0x%x\n",
   1452 	    bp->b_vp, bp->b_lblkno, bp->b_blkno, bp->b_rawblkno, bp->b_dev);
   1453 
   1454 	snprintb(bf, sizeof(bf),
   1455 	    buf_flagbits, bp->b_flags | bp->b_oflags | bp->b_cflags);
   1456 	(*pr)("  error %d flags 0x%s\n", bp->b_error, bf);
   1457 
   1458 	(*pr)("  bufsize 0x%lx bcount 0x%lx resid 0x%lx\n",
   1459 		  bp->b_bufsize, bp->b_bcount, bp->b_resid);
   1460 	(*pr)("  data %p saveaddr %p\n",
   1461 		  bp->b_data, bp->b_saveaddr);
   1462 	(*pr)("  iodone %p objlock %p\n", bp->b_iodone, bp->b_objlock);
   1463 }
   1464 
   1465 void
   1466 vfs_vnode_print(struct vnode *vp, int full, void (*pr)(const char *, ...))
   1467 {
   1468 	char bf[256];
   1469 
   1470 	uvm_object_printit(&vp->v_uobj, full, pr);
   1471 	snprintb(bf, sizeof(bf),
   1472 	    vnode_flagbits, vp->v_iflag | vp->v_vflag | vp->v_uflag);
   1473 	(*pr)("\nVNODE flags %s\n", bf);
   1474 	(*pr)("mp %p numoutput %d size 0x%llx writesize 0x%llx\n",
   1475 	      vp->v_mount, vp->v_numoutput, vp->v_size, vp->v_writesize);
   1476 
   1477 	(*pr)("data %p writecount %ld holdcnt %ld\n",
   1478 	      vp->v_data, vp->v_writecount, vp->v_holdcnt);
   1479 
   1480 	(*pr)("tag %s(%d) type %s(%d) mount %p typedata %p\n",
   1481 	      ARRAY_PRINT(vp->v_tag, vnode_tags), vp->v_tag,
   1482 	      ARRAY_PRINT(vp->v_type, vnode_types), vp->v_type,
   1483 	      vp->v_mount, vp->v_mountedhere);
   1484 
   1485 	(*pr)("v_lock %p\n", &vp->v_lock);
   1486 
   1487 	vcache_print(vp, "", pr);
   1488 
   1489 	if (full) {
   1490 		struct buf *bp;
   1491 
   1492 		(*pr)("clean bufs:\n");
   1493 		LIST_FOREACH(bp, &vp->v_cleanblkhd, b_vnbufs) {
   1494 			(*pr)(" bp %p\n", bp);
   1495 			vfs_buf_print(bp, full, pr);
   1496 		}
   1497 
   1498 		(*pr)("dirty bufs:\n");
   1499 		LIST_FOREACH(bp, &vp->v_dirtyblkhd, b_vnbufs) {
   1500 			(*pr)(" bp %p\n", bp);
   1501 			vfs_buf_print(bp, full, pr);
   1502 		}
   1503 	}
   1504 }
   1505 
   1506 void
   1507 vfs_mount_print(struct mount *mp, int full, void (*pr)(const char *, ...))
   1508 {
   1509 	char sbuf[256];
   1510 
   1511 	(*pr)("vnodecovered = %p data = %p\n",
   1512 			mp->mnt_vnodecovered,mp->mnt_data);
   1513 
   1514 	(*pr)("fs_bshift %d dev_bshift = %d\n",
   1515 			mp->mnt_fs_bshift,mp->mnt_dev_bshift);
   1516 
   1517 	snprintb(sbuf, sizeof(sbuf), __MNT_FLAG_BITS, mp->mnt_flag);
   1518 	(*pr)("flag = %s\n", sbuf);
   1519 
   1520 	snprintb(sbuf, sizeof(sbuf), __IMNT_FLAG_BITS, mp->mnt_iflag);
   1521 	(*pr)("iflag = %s\n", sbuf);
   1522 
   1523 	(*pr)("refcnt = %d unmounting @ %p updating @ %p\n", mp->mnt_refcnt,
   1524 	    &mp->mnt_unmounting, &mp->mnt_updating);
   1525 
   1526 	(*pr)("statvfs cache:\n");
   1527 	(*pr)("\tbsize = %lu\n",mp->mnt_stat.f_bsize);
   1528 	(*pr)("\tfrsize = %lu\n",mp->mnt_stat.f_frsize);
   1529 	(*pr)("\tiosize = %lu\n",mp->mnt_stat.f_iosize);
   1530 
   1531 	(*pr)("\tblocks = %"PRIu64"\n",mp->mnt_stat.f_blocks);
   1532 	(*pr)("\tbfree = %"PRIu64"\n",mp->mnt_stat.f_bfree);
   1533 	(*pr)("\tbavail = %"PRIu64"\n",mp->mnt_stat.f_bavail);
   1534 	(*pr)("\tbresvd = %"PRIu64"\n",mp->mnt_stat.f_bresvd);
   1535 
   1536 	(*pr)("\tfiles = %"PRIu64"\n",mp->mnt_stat.f_files);
   1537 	(*pr)("\tffree = %"PRIu64"\n",mp->mnt_stat.f_ffree);
   1538 	(*pr)("\tfavail = %"PRIu64"\n",mp->mnt_stat.f_favail);
   1539 	(*pr)("\tfresvd = %"PRIu64"\n",mp->mnt_stat.f_fresvd);
   1540 
   1541 	(*pr)("\tf_fsidx = { 0x%"PRIx32", 0x%"PRIx32" }\n",
   1542 			mp->mnt_stat.f_fsidx.__fsid_val[0],
   1543 			mp->mnt_stat.f_fsidx.__fsid_val[1]);
   1544 
   1545 	(*pr)("\towner = %"PRIu32"\n",mp->mnt_stat.f_owner);
   1546 	(*pr)("\tnamemax = %lu\n",mp->mnt_stat.f_namemax);
   1547 
   1548 	snprintb(sbuf, sizeof(sbuf), __MNT_FLAG_BITS, mp->mnt_stat.f_flag);
   1549 
   1550 	(*pr)("\tflag = %s\n",sbuf);
   1551 	(*pr)("\tsyncwrites = %" PRIu64 "\n",mp->mnt_stat.f_syncwrites);
   1552 	(*pr)("\tasyncwrites = %" PRIu64 "\n",mp->mnt_stat.f_asyncwrites);
   1553 	(*pr)("\tsyncreads = %" PRIu64 "\n",mp->mnt_stat.f_syncreads);
   1554 	(*pr)("\tasyncreads = %" PRIu64 "\n",mp->mnt_stat.f_asyncreads);
   1555 	(*pr)("\tfstypename = %s\n",mp->mnt_stat.f_fstypename);
   1556 	(*pr)("\tmntonname = %s\n",mp->mnt_stat.f_mntonname);
   1557 	(*pr)("\tmntfromname = %s\n",mp->mnt_stat.f_mntfromname);
   1558 
   1559 	{
   1560 		int cnt = 0;
   1561 		struct vnode *vp;
   1562 		(*pr)("locked vnodes =");
   1563 		TAILQ_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
   1564 			if (VOP_ISLOCKED(vp)) {
   1565 				if ((++cnt % 6) == 0) {
   1566 					(*pr)(" %p,\n\t", vp);
   1567 				} else {
   1568 					(*pr)(" %p,", vp);
   1569 				}
   1570 			}
   1571 		}
   1572 		(*pr)("\n");
   1573 	}
   1574 
   1575 	if (full) {
   1576 		int cnt = 0;
   1577 		struct vnode *vp;
   1578 		(*pr)("all vnodes =");
   1579 		TAILQ_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
   1580 			if (!TAILQ_NEXT(vp, v_mntvnodes)) {
   1581 				(*pr)(" %p", vp);
   1582 			} else if ((++cnt % 6) == 0) {
   1583 				(*pr)(" %p,\n\t", vp);
   1584 			} else {
   1585 				(*pr)(" %p,", vp);
   1586 			}
   1587 		}
   1588 		(*pr)("\n", vp);
   1589 	}
   1590 }
   1591 
   1592 /*
   1593  * List all of the locked vnodes in the system.
   1594  */
   1595 void printlockedvnodes(void);
   1596 
   1597 void
   1598 printlockedvnodes(void)
   1599 {
   1600 	struct mount *mp, *nmp;
   1601 	struct vnode *vp;
   1602 
   1603 	printf("Locked vnodes\n");
   1604 	mutex_enter(&mountlist_lock);
   1605 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
   1606 		if (vfs_busy(mp, &nmp)) {
   1607 			continue;
   1608 		}
   1609 		TAILQ_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
   1610 			if (VOP_ISLOCKED(vp))
   1611 				vprint(NULL, vp);
   1612 		}
   1613 		mutex_enter(&mountlist_lock);
   1614 		vfs_unbusy(mp, false, &nmp);
   1615 	}
   1616 	mutex_exit(&mountlist_lock);
   1617 }
   1618 
   1619 #endif /* DDB || DEBUGPRINT */
   1620