Home | History | Annotate | Line # | Graph | Download | only in kern
      1 /*	$NetBSD: vfs_subr.c,v 1.458 2017/01/11 12:17:34 joerg 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 1.458 2017/01/11 12:17:34 joerg 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_impl.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_impl) synclist_t;
    556 
    557 static void	vn_syncer_add1(struct vnode *, int);
    558 static void	sysctl_vfs_syncfs_setup(struct sysctllog **);
    559 
    560 /*
    561  * Defines and variables for the syncer process.
    562  */
    563 int syncer_maxdelay = SYNCER_MAXDELAY;	/* maximum delay time */
    564 time_t syncdelay = 30;			/* max time to delay syncing data */
    565 time_t filedelay = 30;			/* time to delay syncing files */
    566 time_t dirdelay  = 15;			/* time to delay syncing directories */
    567 time_t metadelay = 10;			/* time to delay syncing metadata */
    568 time_t lockdelay = 1;			/* time to delay if locking fails */
    569 
    570 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 	vnode_impl_t *vip = VNODE_TO_VIMPL(vp);
    628 
    629 	KASSERT(mutex_owned(&syncer_data_lock));
    630 
    631 	if (vp->v_iflag & VI_ONWORKLST) {
    632 		/*
    633 		 * Remove in order to adjust the position of the vnode.
    634 		 * Note: called from sched_sync(), which will not hold
    635 		 * interlock, therefore we cannot modify v_iflag here.
    636 		 */
    637 		slp = &syncer_workitem_pending[vip->vi_synclist_slot];
    638 		TAILQ_REMOVE(slp, vip, vi_synclist);
    639 	} else {
    640 		KASSERT(mutex_owned(vp->v_interlock));
    641 		vp->v_iflag |= VI_ONWORKLST;
    642 	}
    643 
    644 	vip->vi_synclist_slot = sync_delay_slot(delayx);
    645 
    646 	slp = &syncer_workitem_pending[vip->vi_synclist_slot];
    647 	TAILQ_INSERT_TAIL(slp, vip, vi_synclist);
    648 }
    649 
    650 void
    651 vn_syncer_add_to_worklist(struct vnode *vp, int delayx)
    652 {
    653 
    654 	KASSERT(mutex_owned(vp->v_interlock));
    655 
    656 	mutex_enter(&syncer_data_lock);
    657 	vn_syncer_add1(vp, delayx);
    658 	mutex_exit(&syncer_data_lock);
    659 }
    660 
    661 /*
    662  * Remove an item from the syncer work queue.
    663  */
    664 void
    665 vn_syncer_remove_from_worklist(struct vnode *vp)
    666 {
    667 	synclist_t *slp;
    668 	vnode_impl_t *vip = VNODE_TO_VIMPL(vp);
    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[vip->vi_synclist_slot];
    676 		TAILQ_REMOVE(slp, vip, vi_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 (vcache_tryvget(vp) == 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 = VIMPL_TO_VNODE(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 (VIMPL_TO_VNODE(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 /*
   1062  * Vnode state to string.
   1063  */
   1064 const char *
   1065 vstate_name(enum vnode_state state)
   1066 {
   1067 
   1068 	switch (state) {
   1069 	case VS_MARKER:
   1070 		return "MARKER";
   1071 	case VS_LOADING:
   1072 		return "LOADING";
   1073 	case VS_ACTIVE:
   1074 		return "ACTIVE";
   1075 	case VS_BLOCKED:
   1076 		return "BLOCKED";
   1077 	case VS_RECLAIMING:
   1078 		return "RECLAIMING";
   1079 	case VS_RECLAIMED:
   1080 		return "RECLAIMED";
   1081 	default:
   1082 		return "ILLEGAL";
   1083 	}
   1084 }
   1085 
   1086 /*
   1087  * Print a description of a vnode (common part).
   1088  */
   1089 static void
   1090 vprint_common(struct vnode *vp, const char *prefix,
   1091     void (*pr)(const char *, ...) __printflike(1, 2))
   1092 {
   1093 	int n;
   1094 	char bf[96];
   1095 	const uint8_t *cp;
   1096 	vnode_impl_t *vip;
   1097 	const char * const vnode_tags[] = { VNODE_TAGS };
   1098 	const char * const vnode_types[] = { VNODE_TYPES };
   1099 	const char vnode_flagbits[] = VNODE_FLAGBITS;
   1100 
   1101 #define ARRAY_SIZE(arr) (sizeof(arr) / sizeof(arr[0]))
   1102 #define ARRAY_PRINT(idx, arr) \
   1103     ((unsigned int)(idx) < ARRAY_SIZE(arr) ? (arr)[(idx)] : "UNKNOWN")
   1104 
   1105 	vip = VNODE_TO_VIMPL(vp);
   1106 
   1107 	snprintb(bf, sizeof(bf),
   1108 	    vnode_flagbits, vp->v_iflag | vp->v_vflag | vp->v_uflag);
   1109 
   1110 	(*pr)("vnode %p flags %s\n", vp, bf);
   1111 	(*pr)("%stag %s(%d) type %s(%d) mount %p typedata %p\n", prefix,
   1112 	    ARRAY_PRINT(vp->v_tag, vnode_tags), vp->v_tag,
   1113 	    ARRAY_PRINT(vp->v_type, vnode_types), vp->v_type,
   1114 	    vp->v_mount, vp->v_mountedhere);
   1115 	(*pr)("%susecount %d writecount %d holdcount %d\n", prefix,
   1116 	    vp->v_usecount, vp->v_writecount, vp->v_holdcnt);
   1117 	(*pr)("%ssize %" PRIx64 " writesize %" PRIx64 " numoutput %d\n",
   1118 	    prefix, vp->v_size, vp->v_writesize, vp->v_numoutput);
   1119 	(*pr)("%sdata %p lock %p\n", prefix, vp->v_data, &vip->vi_lock);
   1120 
   1121 	(*pr)("%sstate %s key(%p %zd)", prefix, vstate_name(vip->vi_state),
   1122 	    vip->vi_key.vk_mount, vip->vi_key.vk_key_len);
   1123 	n = vip->vi_key.vk_key_len;
   1124 	cp = vip->vi_key.vk_key;
   1125 	while (n-- > 0)
   1126 		(*pr)(" %02x", *cp++);
   1127 	(*pr)("\n");
   1128 	(*pr)("%slrulisthd %p\n", prefix, vip->vi_lrulisthd);
   1129 
   1130 #undef ARRAY_PRINT
   1131 #undef ARRAY_SIZE
   1132 }
   1133 
   1134 /*
   1135  * Print out a description of a vnode.
   1136  */
   1137 void
   1138 vprint(const char *label, struct vnode *vp)
   1139 {
   1140 
   1141 	if (label != NULL)
   1142 		printf("%s: ", label);
   1143 	vprint_common(vp, "\t", printf);
   1144 	if (vp->v_data != NULL) {
   1145 		printf("\t");
   1146 		VOP_PRINT(vp);
   1147 	}
   1148 }
   1149 
   1150 /* Deprecated. Kept for KPI compatibility. */
   1151 int
   1152 vaccess(enum vtype type, mode_t file_mode, uid_t uid, gid_t gid,
   1153     mode_t acc_mode, kauth_cred_t cred)
   1154 {
   1155 
   1156 #ifdef DIAGNOSTIC
   1157 	printf("vaccess: deprecated interface used.\n");
   1158 #endif /* DIAGNOSTIC */
   1159 
   1160 	return kauth_authorize_vnode(cred, KAUTH_ACCESS_ACTION(acc_mode,
   1161 	    type, file_mode), NULL /* This may panic. */, NULL,
   1162 	    genfs_can_access(type, file_mode, uid, gid, acc_mode, cred));
   1163 }
   1164 
   1165 /*
   1166  * Given a file system name, look up the vfsops for that
   1167  * file system, or return NULL if file system isn't present
   1168  * in the kernel.
   1169  */
   1170 struct vfsops *
   1171 vfs_getopsbyname(const char *name)
   1172 {
   1173 	struct vfsops *v;
   1174 
   1175 	mutex_enter(&vfs_list_lock);
   1176 	LIST_FOREACH(v, &vfs_list, vfs_list) {
   1177 		if (strcmp(v->vfs_name, name) == 0)
   1178 			break;
   1179 	}
   1180 	if (v != NULL)
   1181 		v->vfs_refcount++;
   1182 	mutex_exit(&vfs_list_lock);
   1183 
   1184 	return (v);
   1185 }
   1186 
   1187 void
   1188 copy_statvfs_info(struct statvfs *sbp, const struct mount *mp)
   1189 {
   1190 	const struct statvfs *mbp;
   1191 
   1192 	if (sbp == (mbp = &mp->mnt_stat))
   1193 		return;
   1194 
   1195 	(void)memcpy(&sbp->f_fsidx, &mbp->f_fsidx, sizeof(sbp->f_fsidx));
   1196 	sbp->f_fsid = mbp->f_fsid;
   1197 	sbp->f_owner = mbp->f_owner;
   1198 	sbp->f_flag = mbp->f_flag;
   1199 	sbp->f_syncwrites = mbp->f_syncwrites;
   1200 	sbp->f_asyncwrites = mbp->f_asyncwrites;
   1201 	sbp->f_syncreads = mbp->f_syncreads;
   1202 	sbp->f_asyncreads = mbp->f_asyncreads;
   1203 	(void)memcpy(sbp->f_spare, mbp->f_spare, sizeof(mbp->f_spare));
   1204 	(void)memcpy(sbp->f_fstypename, mbp->f_fstypename,
   1205 	    sizeof(sbp->f_fstypename));
   1206 	(void)memcpy(sbp->f_mntonname, mbp->f_mntonname,
   1207 	    sizeof(sbp->f_mntonname));
   1208 	(void)memcpy(sbp->f_mntfromname, mp->mnt_stat.f_mntfromname,
   1209 	    sizeof(sbp->f_mntfromname));
   1210 	sbp->f_namemax = mbp->f_namemax;
   1211 }
   1212 
   1213 int
   1214 set_statvfs_info(const char *onp, int ukon, const char *fromp, int ukfrom,
   1215     const char *vfsname, struct mount *mp, struct lwp *l)
   1216 {
   1217 	int error;
   1218 	size_t size;
   1219 	struct statvfs *sfs = &mp->mnt_stat;
   1220 	int (*fun)(const void *, void *, size_t, size_t *);
   1221 
   1222 	(void)strlcpy(mp->mnt_stat.f_fstypename, vfsname,
   1223 	    sizeof(mp->mnt_stat.f_fstypename));
   1224 
   1225 	if (onp) {
   1226 		struct cwdinfo *cwdi = l->l_proc->p_cwdi;
   1227 		fun = (ukon == UIO_SYSSPACE) ? copystr : copyinstr;
   1228 		if (cwdi->cwdi_rdir != NULL) {
   1229 			size_t len;
   1230 			char *bp;
   1231 			char *path = PNBUF_GET();
   1232 
   1233 			bp = path + MAXPATHLEN;
   1234 			*--bp = '\0';
   1235 			rw_enter(&cwdi->cwdi_lock, RW_READER);
   1236 			error = getcwd_common(cwdi->cwdi_rdir, rootvnode, &bp,
   1237 			    path, MAXPATHLEN / 2, 0, l);
   1238 			rw_exit(&cwdi->cwdi_lock);
   1239 			if (error) {
   1240 				PNBUF_PUT(path);
   1241 				return error;
   1242 			}
   1243 
   1244 			len = strlen(bp);
   1245 			if (len > sizeof(sfs->f_mntonname) - 1)
   1246 				len = sizeof(sfs->f_mntonname) - 1;
   1247 			(void)strncpy(sfs->f_mntonname, bp, len);
   1248 			PNBUF_PUT(path);
   1249 
   1250 			if (len < sizeof(sfs->f_mntonname) - 1) {
   1251 				error = (*fun)(onp, &sfs->f_mntonname[len],
   1252 				    sizeof(sfs->f_mntonname) - len - 1, &size);
   1253 				if (error)
   1254 					return error;
   1255 				size += len;
   1256 			} else {
   1257 				size = len;
   1258 			}
   1259 		} else {
   1260 			error = (*fun)(onp, &sfs->f_mntonname,
   1261 			    sizeof(sfs->f_mntonname) - 1, &size);
   1262 			if (error)
   1263 				return error;
   1264 		}
   1265 		(void)memset(sfs->f_mntonname + size, 0,
   1266 		    sizeof(sfs->f_mntonname) - size);
   1267 	}
   1268 
   1269 	if (fromp) {
   1270 		fun = (ukfrom == UIO_SYSSPACE) ? copystr : copyinstr;
   1271 		error = (*fun)(fromp, sfs->f_mntfromname,
   1272 		    sizeof(sfs->f_mntfromname) - 1, &size);
   1273 		if (error)
   1274 			return error;
   1275 		(void)memset(sfs->f_mntfromname + size, 0,
   1276 		    sizeof(sfs->f_mntfromname) - size);
   1277 	}
   1278 	return 0;
   1279 }
   1280 
   1281 void
   1282 vfs_timestamp(struct timespec *ts)
   1283 {
   1284 
   1285 	nanotime(ts);
   1286 }
   1287 
   1288 time_t	rootfstime;			/* recorded root fs time, if known */
   1289 void
   1290 setrootfstime(time_t t)
   1291 {
   1292 	rootfstime = t;
   1293 }
   1294 
   1295 static const uint8_t vttodt_tab[ ] = {
   1296 	[VNON]	=	DT_UNKNOWN,
   1297 	[VREG]	=	DT_REG,
   1298 	[VDIR]	=	DT_DIR,
   1299 	[VBLK]	=	DT_BLK,
   1300 	[VCHR]	=	DT_CHR,
   1301 	[VLNK]	=	DT_LNK,
   1302 	[VSOCK]	=	DT_SOCK,
   1303 	[VFIFO]	=	DT_FIFO,
   1304 	[VBAD]	=	DT_UNKNOWN
   1305 };
   1306 
   1307 uint8_t
   1308 vtype2dt(enum vtype vt)
   1309 {
   1310 
   1311 	CTASSERT(VBAD == __arraycount(vttodt_tab) - 1);
   1312 	return vttodt_tab[vt];
   1313 }
   1314 
   1315 int
   1316 VFS_MOUNT(struct mount *mp, const char *a, void *b, size_t *c)
   1317 {
   1318 	int error;
   1319 
   1320 	KERNEL_LOCK(1, NULL);
   1321 	error = (*(mp->mnt_op->vfs_mount))(mp, a, b, c);
   1322 	KERNEL_UNLOCK_ONE(NULL);
   1323 
   1324 	return error;
   1325 }
   1326 
   1327 int
   1328 VFS_START(struct mount *mp, int a)
   1329 {
   1330 	int error;
   1331 
   1332 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
   1333 		KERNEL_LOCK(1, NULL);
   1334 	}
   1335 	error = (*(mp->mnt_op->vfs_start))(mp, a);
   1336 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
   1337 		KERNEL_UNLOCK_ONE(NULL);
   1338 	}
   1339 
   1340 	return error;
   1341 }
   1342 
   1343 int
   1344 VFS_UNMOUNT(struct mount *mp, int a)
   1345 {
   1346 	int error;
   1347 
   1348 	KERNEL_LOCK(1, NULL);
   1349 	error = (*(mp->mnt_op->vfs_unmount))(mp, a);
   1350 	KERNEL_UNLOCK_ONE(NULL);
   1351 
   1352 	return error;
   1353 }
   1354 
   1355 int
   1356 VFS_ROOT(struct mount *mp, struct vnode **a)
   1357 {
   1358 	int error;
   1359 
   1360 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
   1361 		KERNEL_LOCK(1, NULL);
   1362 	}
   1363 	error = (*(mp->mnt_op->vfs_root))(mp, a);
   1364 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
   1365 		KERNEL_UNLOCK_ONE(NULL);
   1366 	}
   1367 
   1368 	return error;
   1369 }
   1370 
   1371 int
   1372 VFS_QUOTACTL(struct mount *mp, struct quotactl_args *args)
   1373 {
   1374 	int error;
   1375 
   1376 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
   1377 		KERNEL_LOCK(1, NULL);
   1378 	}
   1379 	error = (*(mp->mnt_op->vfs_quotactl))(mp, args);
   1380 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
   1381 		KERNEL_UNLOCK_ONE(NULL);
   1382 	}
   1383 
   1384 	return error;
   1385 }
   1386 
   1387 int
   1388 VFS_STATVFS(struct mount *mp, struct statvfs *a)
   1389 {
   1390 	int error;
   1391 
   1392 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
   1393 		KERNEL_LOCK(1, NULL);
   1394 	}
   1395 	error = (*(mp->mnt_op->vfs_statvfs))(mp, a);
   1396 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
   1397 		KERNEL_UNLOCK_ONE(NULL);
   1398 	}
   1399 
   1400 	return error;
   1401 }
   1402 
   1403 int
   1404 VFS_SYNC(struct mount *mp, int a, struct kauth_cred *b)
   1405 {
   1406 	int error;
   1407 
   1408 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
   1409 		KERNEL_LOCK(1, NULL);
   1410 	}
   1411 	error = (*(mp->mnt_op->vfs_sync))(mp, a, b);
   1412 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
   1413 		KERNEL_UNLOCK_ONE(NULL);
   1414 	}
   1415 
   1416 	return error;
   1417 }
   1418 
   1419 int
   1420 VFS_FHTOVP(struct mount *mp, struct fid *a, struct vnode **b)
   1421 {
   1422 	int error;
   1423 
   1424 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
   1425 		KERNEL_LOCK(1, NULL);
   1426 	}
   1427 	error = (*(mp->mnt_op->vfs_fhtovp))(mp, a, b);
   1428 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
   1429 		KERNEL_UNLOCK_ONE(NULL);
   1430 	}
   1431 
   1432 	return error;
   1433 }
   1434 
   1435 int
   1436 VFS_VPTOFH(struct vnode *vp, struct fid *a, size_t *b)
   1437 {
   1438 	int error;
   1439 
   1440 	if ((vp->v_vflag & VV_MPSAFE) == 0) {
   1441 		KERNEL_LOCK(1, NULL);
   1442 	}
   1443 	error = (*(vp->v_mount->mnt_op->vfs_vptofh))(vp, a, b);
   1444 	if ((vp->v_vflag & VV_MPSAFE) == 0) {
   1445 		KERNEL_UNLOCK_ONE(NULL);
   1446 	}
   1447 
   1448 	return error;
   1449 }
   1450 
   1451 int
   1452 VFS_SNAPSHOT(struct mount *mp, struct vnode *a, struct timespec *b)
   1453 {
   1454 	int error;
   1455 
   1456 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
   1457 		KERNEL_LOCK(1, NULL);
   1458 	}
   1459 	error = (*(mp->mnt_op->vfs_snapshot))(mp, a, b);
   1460 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
   1461 		KERNEL_UNLOCK_ONE(NULL);
   1462 	}
   1463 
   1464 	return error;
   1465 }
   1466 
   1467 int
   1468 VFS_EXTATTRCTL(struct mount *mp, int a, struct vnode *b, int c, const char *d)
   1469 {
   1470 	int error;
   1471 
   1472 	KERNEL_LOCK(1, NULL);		/* XXXSMP check ffs */
   1473 	error = (*(mp->mnt_op->vfs_extattrctl))(mp, a, b, c, d);
   1474 	KERNEL_UNLOCK_ONE(NULL);	/* XXX */
   1475 
   1476 	return error;
   1477 }
   1478 
   1479 int
   1480 VFS_SUSPENDCTL(struct mount *mp, int a)
   1481 {
   1482 	int error;
   1483 
   1484 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
   1485 		KERNEL_LOCK(1, NULL);
   1486 	}
   1487 	error = (*(mp->mnt_op->vfs_suspendctl))(mp, a);
   1488 	if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) {
   1489 		KERNEL_UNLOCK_ONE(NULL);
   1490 	}
   1491 
   1492 	return error;
   1493 }
   1494 
   1495 #if defined(DDB) || defined(DEBUGPRINT)
   1496 static const char buf_flagbits[] = BUF_FLAGBITS;
   1497 
   1498 void
   1499 vfs_buf_print(struct buf *bp, int full, void (*pr)(const char *, ...))
   1500 {
   1501 	char bf[1024];
   1502 
   1503 	(*pr)("  vp %p lblkno 0x%"PRIx64" blkno 0x%"PRIx64" rawblkno 0x%"
   1504 	    PRIx64 " dev 0x%x\n",
   1505 	    bp->b_vp, bp->b_lblkno, bp->b_blkno, bp->b_rawblkno, bp->b_dev);
   1506 
   1507 	snprintb(bf, sizeof(bf),
   1508 	    buf_flagbits, bp->b_flags | bp->b_oflags | bp->b_cflags);
   1509 	(*pr)("  error %d flags 0x%s\n", bp->b_error, bf);
   1510 
   1511 	(*pr)("  bufsize 0x%lx bcount 0x%lx resid 0x%lx\n",
   1512 		  bp->b_bufsize, bp->b_bcount, bp->b_resid);
   1513 	(*pr)("  data %p saveaddr %p\n",
   1514 		  bp->b_data, bp->b_saveaddr);
   1515 	(*pr)("  iodone %p objlock %p\n", bp->b_iodone, bp->b_objlock);
   1516 }
   1517 
   1518 void
   1519 vfs_vnode_print(struct vnode *vp, int full, void (*pr)(const char *, ...))
   1520 {
   1521 
   1522 	uvm_object_printit(&vp->v_uobj, full, pr);
   1523 	(*pr)("\n");
   1524 	vprint_common(vp, "", printf);
   1525 	if (full) {
   1526 		struct buf *bp;
   1527 
   1528 		(*pr)("clean bufs:\n");
   1529 		LIST_FOREACH(bp, &vp->v_cleanblkhd, b_vnbufs) {
   1530 			(*pr)(" bp %p\n", bp);
   1531 			vfs_buf_print(bp, full, pr);
   1532 		}
   1533 
   1534 		(*pr)("dirty bufs:\n");
   1535 		LIST_FOREACH(bp, &vp->v_dirtyblkhd, b_vnbufs) {
   1536 			(*pr)(" bp %p\n", bp);
   1537 			vfs_buf_print(bp, full, pr);
   1538 		}
   1539 	}
   1540 }
   1541 
   1542 void
   1543 vfs_vnode_lock_print(void *vlock, int full, void (*pr)(const char *, ...))
   1544 {
   1545 	struct mount *mp;
   1546 	vnode_impl_t *vip;
   1547 
   1548 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
   1549 		TAILQ_FOREACH(vip, &mp->mnt_vnodelist, vi_mntvnodes) {
   1550 			if (&vip->vi_lock != vlock)
   1551 				continue;
   1552 			vfs_vnode_print(VIMPL_TO_VNODE(vip), full, pr);
   1553 		}
   1554 	}
   1555 }
   1556 
   1557 void
   1558 vfs_mount_print(struct mount *mp, int full, void (*pr)(const char *, ...))
   1559 {
   1560 	char sbuf[256];
   1561 
   1562 	(*pr)("vnodecovered = %p data = %p\n",
   1563 			mp->mnt_vnodecovered,mp->mnt_data);
   1564 
   1565 	(*pr)("fs_bshift %d dev_bshift = %d\n",
   1566 			mp->mnt_fs_bshift,mp->mnt_dev_bshift);
   1567 
   1568 	snprintb(sbuf, sizeof(sbuf), __MNT_FLAG_BITS, mp->mnt_flag);
   1569 	(*pr)("flag = %s\n", sbuf);
   1570 
   1571 	snprintb(sbuf, sizeof(sbuf), __IMNT_FLAG_BITS, mp->mnt_iflag);
   1572 	(*pr)("iflag = %s\n", sbuf);
   1573 
   1574 	(*pr)("refcnt = %d unmounting @ %p updating @ %p\n", mp->mnt_refcnt,
   1575 	    &mp->mnt_unmounting, &mp->mnt_updating);
   1576 
   1577 	(*pr)("statvfs cache:\n");
   1578 	(*pr)("\tbsize = %lu\n",mp->mnt_stat.f_bsize);
   1579 	(*pr)("\tfrsize = %lu\n",mp->mnt_stat.f_frsize);
   1580 	(*pr)("\tiosize = %lu\n",mp->mnt_stat.f_iosize);
   1581 
   1582 	(*pr)("\tblocks = %"PRIu64"\n",mp->mnt_stat.f_blocks);
   1583 	(*pr)("\tbfree = %"PRIu64"\n",mp->mnt_stat.f_bfree);
   1584 	(*pr)("\tbavail = %"PRIu64"\n",mp->mnt_stat.f_bavail);
   1585 	(*pr)("\tbresvd = %"PRIu64"\n",mp->mnt_stat.f_bresvd);
   1586 
   1587 	(*pr)("\tfiles = %"PRIu64"\n",mp->mnt_stat.f_files);
   1588 	(*pr)("\tffree = %"PRIu64"\n",mp->mnt_stat.f_ffree);
   1589 	(*pr)("\tfavail = %"PRIu64"\n",mp->mnt_stat.f_favail);
   1590 	(*pr)("\tfresvd = %"PRIu64"\n",mp->mnt_stat.f_fresvd);
   1591 
   1592 	(*pr)("\tf_fsidx = { 0x%"PRIx32", 0x%"PRIx32" }\n",
   1593 			mp->mnt_stat.f_fsidx.__fsid_val[0],
   1594 			mp->mnt_stat.f_fsidx.__fsid_val[1]);
   1595 
   1596 	(*pr)("\towner = %"PRIu32"\n",mp->mnt_stat.f_owner);
   1597 	(*pr)("\tnamemax = %lu\n",mp->mnt_stat.f_namemax);
   1598 
   1599 	snprintb(sbuf, sizeof(sbuf), __MNT_FLAG_BITS, mp->mnt_stat.f_flag);
   1600 
   1601 	(*pr)("\tflag = %s\n",sbuf);
   1602 	(*pr)("\tsyncwrites = %" PRIu64 "\n",mp->mnt_stat.f_syncwrites);
   1603 	(*pr)("\tasyncwrites = %" PRIu64 "\n",mp->mnt_stat.f_asyncwrites);
   1604 	(*pr)("\tsyncreads = %" PRIu64 "\n",mp->mnt_stat.f_syncreads);
   1605 	(*pr)("\tasyncreads = %" PRIu64 "\n",mp->mnt_stat.f_asyncreads);
   1606 	(*pr)("\tfstypename = %s\n",mp->mnt_stat.f_fstypename);
   1607 	(*pr)("\tmntonname = %s\n",mp->mnt_stat.f_mntonname);
   1608 	(*pr)("\tmntfromname = %s\n",mp->mnt_stat.f_mntfromname);
   1609 
   1610 	{
   1611 		int cnt = 0;
   1612 		vnode_t *vp;
   1613 		vnode_impl_t *vip;
   1614 		(*pr)("locked vnodes =");
   1615 		TAILQ_FOREACH(vip, &mp->mnt_vnodelist, vi_mntvnodes) {
   1616 			vp = VIMPL_TO_VNODE(vip);
   1617 			if (VOP_ISLOCKED(vp)) {
   1618 				if ((++cnt % 6) == 0) {
   1619 					(*pr)(" %p,\n\t", vp);
   1620 				} else {
   1621 					(*pr)(" %p,", vp);
   1622 				}
   1623 			}
   1624 		}
   1625 		(*pr)("\n");
   1626 	}
   1627 
   1628 	if (full) {
   1629 		int cnt = 0;
   1630 		vnode_t *vp;
   1631 		vnode_impl_t *vip;
   1632 		(*pr)("all vnodes =");
   1633 		TAILQ_FOREACH(vip, &mp->mnt_vnodelist, vi_mntvnodes) {
   1634 			vp = VIMPL_TO_VNODE(vip);
   1635 			if (!TAILQ_NEXT(vip, vi_mntvnodes)) {
   1636 				(*pr)(" %p", vp);
   1637 			} else if ((++cnt % 6) == 0) {
   1638 				(*pr)(" %p,\n\t", vp);
   1639 			} else {
   1640 				(*pr)(" %p,", vp);
   1641 			}
   1642 		}
   1643 		(*pr)("\n");
   1644 	}
   1645 }
   1646 
   1647 /*
   1648  * List all of the locked vnodes in the system.
   1649  */
   1650 void printlockedvnodes(void);
   1651 
   1652 void
   1653 printlockedvnodes(void)
   1654 {
   1655 	struct mount *mp, *nmp;
   1656 	vnode_t *vp;
   1657 	vnode_impl_t *vip;
   1658 
   1659 	printf("Locked vnodes\n");
   1660 	mutex_enter(&mountlist_lock);
   1661 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
   1662 		if (vfs_busy(mp, &nmp)) {
   1663 			continue;
   1664 		}
   1665 		TAILQ_FOREACH(vip, &mp->mnt_vnodelist, vi_mntvnodes) {
   1666 			vp = VIMPL_TO_VNODE(vip);
   1667 			if (VOP_ISLOCKED(vp))
   1668 				vprint(NULL, vp);
   1669 		}
   1670 		mutex_enter(&mountlist_lock);
   1671 		vfs_unbusy(mp, false, &nmp);
   1672 	}
   1673 	mutex_exit(&mountlist_lock);
   1674 }
   1675 
   1676 #endif /* DDB || DEBUGPRINT */
   1677