Home | History | Annotate | Line # | Download | only in dev
vnd.c revision 1.101
      1 /*	$NetBSD: vnd.c,v 1.101 2003/06/29 22:30:03 fvdl Exp $	*/
      2 
      3 /*-
      4  * Copyright (c) 1996, 1997, 1998 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.
      9  *
     10  * Redistribution and use in source and binary forms, with or without
     11  * modification, are permitted provided that the following conditions
     12  * are met:
     13  * 1. Redistributions of source code must retain the above copyright
     14  *    notice, this list of conditions and the following disclaimer.
     15  * 2. Redistributions in binary form must reproduce the above copyright
     16  *    notice, this list of conditions and the following disclaimer in the
     17  *    documentation and/or other materials provided with the distribution.
     18  * 3. All advertising materials mentioning features or use of this software
     19  *    must display the following acknowledgement:
     20  *	This product includes software developed by the NetBSD
     21  *	Foundation, Inc. and its contributors.
     22  * 4. Neither the name of The NetBSD Foundation nor the names of its
     23  *    contributors may be used to endorse or promote products derived
     24  *    from this software without specific prior written permission.
     25  *
     26  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     27  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     28  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     29  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     30  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     31  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     32  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     33  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     34  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     35  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     36  * POSSIBILITY OF SUCH DAMAGE.
     37  */
     38 
     39 /*
     40  * Copyright (c) 1988 University of Utah.
     41  * Copyright (c) 1990, 1993
     42  *	The Regents of the University of California.  All rights reserved.
     43  *
     44  * This code is derived from software contributed to Berkeley by
     45  * the Systems Programming Group of the University of Utah Computer
     46  * Science Department.
     47  *
     48  * Redistribution and use in source and binary forms, with or without
     49  * modification, are permitted provided that the following conditions
     50  * are met:
     51  * 1. Redistributions of source code must retain the above copyright
     52  *    notice, this list of conditions and the following disclaimer.
     53  * 2. Redistributions in binary form must reproduce the above copyright
     54  *    notice, this list of conditions and the following disclaimer in the
     55  *    documentation and/or other materials provided with the distribution.
     56  * 3. All advertising materials mentioning features or use of this software
     57  *    must display the following acknowledgement:
     58  *	This product includes software developed by the University of
     59  *	California, Berkeley and its contributors.
     60  * 4. Neither the name of the University nor the names of its contributors
     61  *    may be used to endorse or promote products derived from this software
     62  *    without specific prior written permission.
     63  *
     64  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     65  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     66  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     67  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     68  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     69  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     70  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     71  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     72  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     73  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     74  * SUCH DAMAGE.
     75  *
     76  * from: Utah $Hdr: vn.c 1.13 94/04/02$
     77  *
     78  *	@(#)vn.c	8.9 (Berkeley) 5/14/95
     79  */
     80 
     81 /*
     82  * Vnode disk driver.
     83  *
     84  * Block/character interface to a vnode.  Allows one to treat a file
     85  * as a disk (e.g. build a filesystem in it, mount it, etc.).
     86  *
     87  * NOTE 1: This uses the VOP_BMAP/VOP_STRATEGY interface to the vnode
     88  * instead of a simple VOP_RDWR.  We do this to avoid distorting the
     89  * local buffer cache.
     90  *
     91  * NOTE 2: There is a security issue involved with this driver.
     92  * Once mounted all access to the contents of the "mapped" file via
     93  * the special file is controlled by the permissions on the special
     94  * file, the protection of the mapped file is ignored (effectively,
     95  * by using root credentials in all transactions).
     96  *
     97  * NOTE 3: Doesn't interact with leases, should it?
     98  */
     99 
    100 #include <sys/cdefs.h>
    101 __KERNEL_RCSID(0, "$NetBSD: vnd.c,v 1.101 2003/06/29 22:30:03 fvdl Exp $");
    102 
    103 #if defined(_KERNEL_OPT)
    104 #include "fs_nfs.h"
    105 #endif
    106 
    107 #include <sys/param.h>
    108 #include <sys/systm.h>
    109 #include <sys/namei.h>
    110 #include <sys/proc.h>
    111 #include <sys/errno.h>
    112 #include <sys/buf.h>
    113 #include <sys/malloc.h>
    114 #include <sys/ioctl.h>
    115 #include <sys/disklabel.h>
    116 #include <sys/device.h>
    117 #include <sys/disk.h>
    118 #include <sys/stat.h>
    119 #include <sys/mount.h>
    120 #include <sys/vnode.h>
    121 #include <sys/file.h>
    122 #include <sys/uio.h>
    123 #include <sys/conf.h>
    124 
    125 #include <miscfs/specfs/specdev.h>
    126 
    127 #include <dev/vndvar.h>
    128 
    129 #if defined(VNDDEBUG) && !defined(DEBUG)
    130 #define	DEBUG
    131 #endif
    132 
    133 #ifdef DEBUG
    134 int dovndcluster = 1;
    135 #define	VDB_FOLLOW	0x01
    136 #define	VDB_INIT	0x02
    137 #define	VDB_IO		0x04
    138 #define	VDB_LABEL	0x08
    139 int vnddebug = 0x00;
    140 #endif
    141 
    142 #define	vndunit(x)	DISKUNIT(x)
    143 
    144 struct vndxfer {
    145 	struct buf	*vx_bp;		/* Pointer to parent buffer */
    146 	int		vx_error;
    147 	int		vx_pending;	/* # of pending aux buffers */
    148 	int		vx_flags;
    149 #define VX_BUSY		1
    150 };
    151 
    152 struct vndbuf {
    153 	struct buf	vb_buf;
    154 	struct vndxfer	*vb_xfer;
    155 };
    156 
    157 #define	VND_GETXFER(vnd)	pool_get(&(vnd)->sc_vxpool, PR_NOWAIT)
    158 #define	VND_PUTXFER(vnd, vx)	pool_put(&(vnd)->sc_vxpool, (vx))
    159 
    160 #define	VND_GETBUF(vnd)		pool_get(&(vnd)->sc_vbpool, PR_NOWAIT)
    161 #define	VND_PUTBUF(vnd, vb)	pool_put(&(vnd)->sc_vbpool, (vb))
    162 
    163 struct vnd_softc *vnd_softc;
    164 int numvnd = 0;
    165 
    166 #define	VNDLABELDEV(dev) \
    167 	(MAKEDISKDEV(major((dev)), vndunit((dev)), RAW_PART))
    168 
    169 /* called by main() at boot time (XXX: and the LKM driver) */
    170 void	vndattach __P((int));
    171 int	vnddetach __P((void));
    172 
    173 void	vndclear __P((struct vnd_softc *, int));
    174 void	vndstart __P((struct vnd_softc *));
    175 int	vndsetcred __P((struct vnd_softc *, struct ucred *));
    176 void	vndthrottle __P((struct vnd_softc *, struct vnode *));
    177 void	vndiodone __P((struct buf *));
    178 #if 0
    179 void	vndshutdown __P((void));
    180 #endif
    181 
    182 void	vndgetdefaultlabel __P((struct vnd_softc *, struct disklabel *));
    183 void	vndgetdisklabel __P((dev_t));
    184 
    185 static	int vndlock __P((struct vnd_softc *));
    186 static	void vndunlock __P((struct vnd_softc *));
    187 
    188 dev_type_open(vndopen);
    189 dev_type_close(vndclose);
    190 dev_type_read(vndread);
    191 dev_type_write(vndwrite);
    192 dev_type_ioctl(vndioctl);
    193 dev_type_strategy(vndstrategy);
    194 dev_type_dump(vnddump);
    195 dev_type_size(vndsize);
    196 
    197 const struct bdevsw vnd_bdevsw = {
    198 	vndopen, vndclose, vndstrategy, vndioctl, vnddump, vndsize, D_DISK
    199 };
    200 
    201 const struct cdevsw vnd_cdevsw = {
    202 	vndopen, vndclose, vndread, vndwrite, vndioctl,
    203 	nostop, notty, nopoll, nommap, nokqfilter, D_DISK
    204 };
    205 
    206 int vndattached = 0;
    207 
    208 void
    209 vndattach(num)
    210 	int num;
    211 {
    212 	int i;
    213 	char *mem;
    214 
    215 	if (vndattached)
    216 		return;
    217 	vndattached = 1;
    218 	if (num <= 0)
    219 		return;
    220 	i = num * sizeof(struct vnd_softc);
    221 	mem = malloc(i, M_DEVBUF, M_NOWAIT|M_ZERO);
    222 	if (mem == NULL) {
    223 		printf("WARNING: no memory for vnode disks\n");
    224 		return;
    225 	}
    226 	vnd_softc = (struct vnd_softc *)mem;
    227 	numvnd = num;
    228 
    229 	for (i = 0; i < numvnd; i++) {
    230 		vnd_softc[i].sc_unit = i;
    231 		bufq_alloc(&vnd_softc[i].sc_tab,
    232 		    BUFQ_DISKSORT|BUFQ_SORT_RAWBLOCK);
    233 	}
    234 }
    235 
    236 int
    237 vnddetach()
    238 {
    239 	int i;
    240 
    241 	/* First check we aren't in use. */
    242 	for (i = 0; i < numvnd; i++)
    243 		if (vnd_softc[i].sc_flags & VNF_INITED)
    244 			return (EBUSY);
    245 
    246 	for (i = 0; i < numvnd; i++)
    247 		bufq_free(&vnd_softc[i].sc_tab);
    248 
    249 	free(vnd_softc, M_DEVBUF);
    250 	vndattached = 0;
    251 
    252 	return (0);
    253 }
    254 
    255 int
    256 vndopen(dev, flags, mode, p)
    257 	dev_t dev;
    258 	int flags, mode;
    259 	struct proc *p;
    260 {
    261 	int unit = vndunit(dev);
    262 	struct vnd_softc *sc;
    263 	int error = 0, part, pmask;
    264 	struct disklabel *lp;
    265 
    266 #ifdef DEBUG
    267 	if (vnddebug & VDB_FOLLOW)
    268 		printf("vndopen(0x%x, 0x%x, 0x%x, %p)\n", dev, flags, mode, p);
    269 #endif
    270 	if (unit >= numvnd)
    271 		return (ENXIO);
    272 	sc = &vnd_softc[unit];
    273 
    274 	if ((error = vndlock(sc)) != 0)
    275 		return (error);
    276 
    277 	lp = sc->sc_dkdev.dk_label;
    278 
    279 	part = DISKPART(dev);
    280 	pmask = (1 << part);
    281 
    282 	/*
    283 	 * If we're initialized, check to see if there are any other
    284 	 * open partitions.  If not, then it's safe to update the
    285 	 * in-core disklabel.  Only read the disklabel if it is
    286 	 * not realdy valid.
    287 	 */
    288 	if ((sc->sc_flags & (VNF_INITED|VNF_VLABEL)) == VNF_INITED &&
    289 	    sc->sc_dkdev.dk_openmask == 0)
    290 		vndgetdisklabel(dev);
    291 
    292 	/* Check that the partitions exists. */
    293 	if (part != RAW_PART) {
    294 		if (((sc->sc_flags & VNF_INITED) == 0) ||
    295 		    ((part >= lp->d_npartitions) ||
    296 		     (lp->d_partitions[part].p_fstype == FS_UNUSED))) {
    297 			error = ENXIO;
    298 			goto done;
    299 		}
    300 	}
    301 
    302 	/* Prevent our unit from being unconfigured while open. */
    303 	switch (mode) {
    304 	case S_IFCHR:
    305 		sc->sc_dkdev.dk_copenmask |= pmask;
    306 		break;
    307 
    308 	case S_IFBLK:
    309 		sc->sc_dkdev.dk_bopenmask |= pmask;
    310 		break;
    311 	}
    312 	sc->sc_dkdev.dk_openmask =
    313 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
    314 
    315  done:
    316 	vndunlock(sc);
    317 	return (error);
    318 }
    319 
    320 int
    321 vndclose(dev, flags, mode, p)
    322 	dev_t dev;
    323 	int flags, mode;
    324 	struct proc *p;
    325 {
    326 	int unit = vndunit(dev);
    327 	struct vnd_softc *sc;
    328 	int error = 0, part;
    329 
    330 #ifdef DEBUG
    331 	if (vnddebug & VDB_FOLLOW)
    332 		printf("vndclose(0x%x, 0x%x, 0x%x, %p)\n", dev, flags, mode, p);
    333 #endif
    334 
    335 	if (unit >= numvnd)
    336 		return (ENXIO);
    337 	sc = &vnd_softc[unit];
    338 
    339 	if ((error = vndlock(sc)) != 0)
    340 		return (error);
    341 
    342 	part = DISKPART(dev);
    343 
    344 	/* ...that much closer to allowing unconfiguration... */
    345 	switch (mode) {
    346 	case S_IFCHR:
    347 		sc->sc_dkdev.dk_copenmask &= ~(1 << part);
    348 		break;
    349 
    350 	case S_IFBLK:
    351 		sc->sc_dkdev.dk_bopenmask &= ~(1 << part);
    352 		break;
    353 	}
    354 	sc->sc_dkdev.dk_openmask =
    355 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
    356 
    357 	if (sc->sc_dkdev.dk_openmask == 0) {
    358 		if ((sc->sc_flags & VNF_KLABEL) == 0)
    359 			sc->sc_flags &= ~VNF_VLABEL;
    360 	}
    361 
    362 	vndunlock(sc);
    363 	return (0);
    364 }
    365 
    366 /*
    367  * Break the request into bsize pieces and submit using VOP_BMAP/VOP_STRATEGY.
    368  */
    369 void
    370 vndstrategy(bp)
    371 	struct buf *bp;
    372 {
    373 	int unit = vndunit(bp->b_dev);
    374 	struct vnd_softc *vnd = &vnd_softc[unit];
    375 	struct vndxfer *vnx;
    376 	int s, bsize, resid;
    377 	off_t bn;
    378 	caddr_t addr;
    379 	int sz, flags, error, wlabel;
    380 	struct disklabel *lp;
    381 	struct partition *pp;
    382 
    383 #ifdef DEBUG
    384 	if (vnddebug & VDB_FOLLOW)
    385 		printf("vndstrategy(%p): unit %d\n", bp, unit);
    386 #endif
    387 	if ((vnd->sc_flags & VNF_INITED) == 0) {
    388 		bp->b_error = ENXIO;
    389 		bp->b_flags |= B_ERROR;
    390 		goto done;
    391 	}
    392 
    393 	/* If it's a nil transfer, wake up the top half now. */
    394 	if (bp->b_bcount == 0)
    395 		goto done;
    396 
    397 	lp = vnd->sc_dkdev.dk_label;
    398 
    399 	/*
    400 	 * The transfer must be a whole number of blocks.
    401 	 */
    402 	if ((bp->b_bcount % lp->d_secsize) != 0) {
    403 		bp->b_error = EINVAL;
    404 		bp->b_flags |= B_ERROR;
    405 		goto done;
    406 	}
    407 
    408 	/*
    409 	 * Do bounds checking and adjust transfer.  If there's an error,
    410 	 * the bounds check will flag that for us.
    411 	 */
    412 	wlabel = vnd->sc_flags & (VNF_WLABEL|VNF_LABELLING);
    413 	if (DISKPART(bp->b_dev) != RAW_PART)
    414 		if (bounds_check_with_label(&vnd->sc_dkdev, bp, wlabel) <= 0)
    415 			goto done;
    416 
    417 	/*
    418 	 * check if we're read-only.
    419 	 */
    420 	if ((vnd->sc_flags & VNF_READONLY) && !(bp->b_flags & B_READ)) {
    421 		bp->b_error = EACCES;
    422 		bp->b_flags |= B_ERROR;
    423 		goto done;
    424 	}
    425 
    426 	bp->b_resid = bp->b_bcount;
    427 
    428 	/*
    429 	 * Put the block number in terms of the logical blocksize
    430 	 * of the "device".
    431 	 */
    432 	bn = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
    433 
    434 	/*
    435 	 * Translate the partition-relative block number to an absolute.
    436 	 */
    437 	if (DISKPART(bp->b_dev) != RAW_PART) {
    438 		pp = &vnd->sc_dkdev.dk_label->d_partitions[DISKPART(bp->b_dev)];
    439 		bn += pp->p_offset;
    440 	}
    441 
    442 	/* ...and convert to a byte offset within the file. */
    443 	bn *= lp->d_secsize;
    444 
    445 	if (vnd->sc_vp->v_mount == NULL) {
    446 		bp->b_error = ENXIO;
    447 		bp->b_flags |= B_ERROR;
    448 		goto done;
    449 	}
    450  	bsize = vnd->sc_vp->v_mount->mnt_stat.f_iosize;
    451 	addr = bp->b_data;
    452 	flags = (bp->b_flags & (B_READ|B_ASYNC)) | B_CALL;
    453 
    454 	/* Allocate a header for this transfer and link it to the buffer */
    455 	s = splbio();
    456 	vnx = VND_GETXFER(vnd);
    457 	splx(s);
    458 	vnx->vx_flags = VX_BUSY;
    459 	vnx->vx_error = 0;
    460 	vnx->vx_pending = 0;
    461 	vnx->vx_bp = bp;
    462 
    463 	for (resid = bp->b_resid; resid; resid -= sz) {
    464 		struct vndbuf *nbp;
    465 		struct vnode *vp;
    466 		daddr_t nbn;
    467 		int off, nra;
    468 
    469 		nra = 0;
    470 		vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY | LK_CANRECURSE);
    471 		error = VOP_BMAP(vnd->sc_vp, bn / bsize, &vp, &nbn, &nra);
    472 		VOP_UNLOCK(vnd->sc_vp, 0);
    473 
    474 		if (error == 0 && (long)nbn == -1)
    475 			error = EIO;
    476 
    477 		/*
    478 		 * If there was an error or a hole in the file...punt.
    479 		 * Note that we may have to wait for any operations
    480 		 * that we have already fired off before releasing
    481 		 * the buffer.
    482 		 *
    483 		 * XXX we could deal with holes here but it would be
    484 		 * a hassle (in the write case).
    485 		 */
    486 		if (error) {
    487 			s = splbio();
    488 			vnx->vx_error = error;
    489 			goto out;
    490 		}
    491 
    492 #ifdef DEBUG
    493 		if (!dovndcluster)
    494 			nra = 0;
    495 #endif
    496 
    497 		if ((off = bn % bsize) != 0)
    498 			sz = bsize - off;
    499 		else
    500 			sz = (1 + nra) * bsize;
    501 		if (resid < sz)
    502 			sz = resid;
    503 #ifdef DEBUG
    504 		if (vnddebug & VDB_IO)
    505 			printf("vndstrategy: vp %p/%p bn 0x%qx/0x%" PRIx64
    506 			       " sz 0x%x\n",
    507 			    vnd->sc_vp, vp, (long long)bn, nbn, sz);
    508 #endif
    509 
    510 		s = splbio();
    511 		nbp = VND_GETBUF(vnd);
    512 		splx(s);
    513 		BUF_INIT(&nbp->vb_buf);
    514 		nbp->vb_buf.b_flags = flags;
    515 		nbp->vb_buf.b_bcount = sz;
    516 		nbp->vb_buf.b_bufsize = round_page((ulong)addr + sz)
    517 		    - trunc_page((ulong) addr);
    518 		nbp->vb_buf.b_error = 0;
    519 		nbp->vb_buf.b_data = addr;
    520 		nbp->vb_buf.b_blkno = nbp->vb_buf.b_rawblkno = nbn + btodb(off);
    521 		nbp->vb_buf.b_proc = bp->b_proc;
    522 		nbp->vb_buf.b_iodone = vndiodone;
    523 		nbp->vb_buf.b_vp = NULLVP;
    524 
    525 		nbp->vb_xfer = vnx;
    526 
    527 		/*
    528 		 * Just sort by block number
    529 		 */
    530 		s = splbio();
    531 		if (vnx->vx_error != 0) {
    532 			VND_PUTBUF(vnd, nbp);
    533 			goto out;
    534 		}
    535 		vnx->vx_pending++;
    536 		bgetvp(vp, &nbp->vb_buf);
    537 		BUFQ_PUT(&vnd->sc_tab, &nbp->vb_buf);
    538 		vndstart(vnd);
    539 		splx(s);
    540 		bn += sz;
    541 		addr += sz;
    542 	}
    543 
    544 	s = splbio();
    545 
    546 out: /* Arrive here at splbio */
    547 	vnx->vx_flags &= ~VX_BUSY;
    548 	if (vnx->vx_pending == 0) {
    549 		if (vnx->vx_error != 0) {
    550 			bp->b_error = vnx->vx_error;
    551 			bp->b_flags |= B_ERROR;
    552 		}
    553 		VND_PUTXFER(vnd, vnx);
    554 		biodone(bp);
    555 	}
    556 	splx(s);
    557 	return;
    558 
    559  done:
    560 	biodone(bp);
    561 }
    562 
    563 /*
    564  * Feed requests sequentially.
    565  * We do it this way to keep from flooding NFS servers if we are connected
    566  * to an NFS file.  This places the burden on the client rather than the
    567  * server.
    568  */
    569 void
    570 vndstart(vnd)
    571 	struct vnd_softc *vnd;
    572 {
    573 	struct buf	*bp;
    574 
    575 	/*
    576 	 * Dequeue now since lower level strategy routine might
    577 	 * queue using same links
    578 	 */
    579 
    580 	if ((vnd->sc_flags & VNF_BUSY) != 0)
    581 		return;
    582 
    583 	vnd->sc_flags |= VNF_BUSY;
    584 
    585 	while (vnd->sc_active < vnd->sc_maxactive) {
    586 		bp = BUFQ_GET(&vnd->sc_tab);
    587 		if (bp == NULL)
    588 			break;
    589 		vnd->sc_active++;
    590 #ifdef DEBUG
    591 		if (vnddebug & VDB_IO)
    592 			printf("vndstart(%ld): bp %p vp %p blkno 0x%" PRIx64
    593 				" flags %lx addr %p cnt 0x%lx\n",
    594 			    (long) (vnd-vnd_softc), bp, bp->b_vp, bp->b_blkno,
    595 			    bp->b_flags, bp->b_data, bp->b_bcount);
    596 #endif
    597 
    598 		/* Instrumentation. */
    599 		disk_busy(&vnd->sc_dkdev);
    600 
    601 		if ((bp->b_flags & B_READ) == 0)
    602 			bp->b_vp->v_numoutput++;
    603 		VOP_STRATEGY(bp);
    604 	}
    605 	vnd->sc_flags &= ~VNF_BUSY;
    606 }
    607 
    608 void
    609 vndiodone(bp)
    610 	struct buf *bp;
    611 {
    612 	struct vndbuf *vbp = (struct vndbuf *) bp;
    613 	struct vndxfer *vnx = (struct vndxfer *)vbp->vb_xfer;
    614 	struct buf *pbp = vnx->vx_bp;
    615 	struct vnd_softc *vnd = &vnd_softc[vndunit(pbp->b_dev)];
    616 	int s, resid;
    617 
    618 	s = splbio();
    619 #ifdef DEBUG
    620 	if (vnddebug & VDB_IO)
    621 		printf("vndiodone(%ld): vbp %p vp %p blkno 0x%" PRIx64
    622 		       " addr %p cnt 0x%lx\n",
    623 		    (long) (vnd-vnd_softc), vbp, vbp->vb_buf.b_vp,
    624 		    vbp->vb_buf.b_blkno, vbp->vb_buf.b_data,
    625 		    vbp->vb_buf.b_bcount);
    626 #endif
    627 
    628 	resid = vbp->vb_buf.b_bcount - vbp->vb_buf.b_resid;
    629 	pbp->b_resid -= resid;
    630 	disk_unbusy(&vnd->sc_dkdev, resid, (pbp->b_flags & B_READ));
    631 	vnx->vx_pending--;
    632 
    633 	if (vbp->vb_buf.b_error) {
    634 #ifdef DEBUG
    635 		if (vnddebug & VDB_IO)
    636 			printf("vndiodone: vbp %p error %d\n", vbp,
    637 			    vbp->vb_buf.b_error);
    638 #endif
    639 		vnx->vx_error = vbp->vb_buf.b_error;
    640 	}
    641 
    642 	if (vbp->vb_buf.b_vp != NULLVP)
    643 		brelvp(&vbp->vb_buf);
    644 
    645 	VND_PUTBUF(vnd, vbp);
    646 
    647 	/*
    648 	 * Wrap up this transaction if it has run to completion or, in
    649 	 * case of an error, when all auxiliary buffers have returned.
    650 	 */
    651 	if (vnx->vx_error != 0) {
    652 		pbp->b_flags |= B_ERROR;
    653 		pbp->b_error = vnx->vx_error;
    654 		if ((vnx->vx_flags & VX_BUSY) == 0 && vnx->vx_pending == 0) {
    655 
    656 #ifdef DEBUG
    657 			if (vnddebug & VDB_IO)
    658 				printf("vndiodone: pbp %p iodone: error %d\n",
    659 					pbp, vnx->vx_error);
    660 #endif
    661 			VND_PUTXFER(vnd, vnx);
    662 			biodone(pbp);
    663 		}
    664 	} else if (pbp->b_resid == 0) {
    665 
    666 #ifdef DIAGNOSTIC
    667 		if (vnx->vx_pending != 0)
    668 			panic("vndiodone: vnx pending: %d", vnx->vx_pending);
    669 #endif
    670 
    671 		if ((vnx->vx_flags & VX_BUSY) == 0) {
    672 #ifdef DEBUG
    673 			if (vnddebug & VDB_IO)
    674 				printf("vndiodone: pbp %p iodone\n", pbp);
    675 #endif
    676 			VND_PUTXFER(vnd, vnx);
    677 			biodone(pbp);
    678 		}
    679 	}
    680 
    681 	vnd->sc_active--;
    682 	vndstart(vnd);
    683 	splx(s);
    684 }
    685 
    686 /* ARGSUSED */
    687 int
    688 vndread(dev, uio, flags)
    689 	dev_t dev;
    690 	struct uio *uio;
    691 	int flags;
    692 {
    693 	int unit = vndunit(dev);
    694 	struct vnd_softc *sc;
    695 
    696 #ifdef DEBUG
    697 	if (vnddebug & VDB_FOLLOW)
    698 		printf("vndread(0x%x, %p)\n", dev, uio);
    699 #endif
    700 
    701 	if (unit >= numvnd)
    702 		return (ENXIO);
    703 	sc = &vnd_softc[unit];
    704 
    705 	if ((sc->sc_flags & VNF_INITED) == 0)
    706 		return (ENXIO);
    707 
    708 	return (physio(vndstrategy, NULL, dev, B_READ, minphys, uio));
    709 }
    710 
    711 /* ARGSUSED */
    712 int
    713 vndwrite(dev, uio, flags)
    714 	dev_t dev;
    715 	struct uio *uio;
    716 	int flags;
    717 {
    718 	int unit = vndunit(dev);
    719 	struct vnd_softc *sc;
    720 
    721 #ifdef DEBUG
    722 	if (vnddebug & VDB_FOLLOW)
    723 		printf("vndwrite(0x%x, %p)\n", dev, uio);
    724 #endif
    725 
    726 	if (unit >= numvnd)
    727 		return (ENXIO);
    728 	sc = &vnd_softc[unit];
    729 
    730 	if ((sc->sc_flags & VNF_INITED) == 0)
    731 		return (ENXIO);
    732 
    733 	return (physio(vndstrategy, NULL, dev, B_WRITE, minphys, uio));
    734 }
    735 
    736 /* ARGSUSED */
    737 int
    738 vndioctl(dev, cmd, data, flag, p)
    739 	dev_t dev;
    740 	u_long cmd;
    741 	caddr_t data;
    742 	int flag;
    743 	struct proc *p;
    744 {
    745 	int unit = vndunit(dev);
    746 	struct vnd_softc *vnd;
    747 	struct vnd_ioctl *vio;
    748 	struct vattr vattr;
    749 	struct nameidata nd;
    750 	int error, part, pmask;
    751 	size_t geomsize;
    752 	int fflags;
    753 #ifdef __HAVE_OLD_DISKLABEL
    754 	struct disklabel newlabel;
    755 #endif
    756 
    757 #ifdef DEBUG
    758 	if (vnddebug & VDB_FOLLOW)
    759 		printf("vndioctl(0x%x, 0x%lx, %p, 0x%x, %p): unit %d\n",
    760 		    dev, cmd, data, flag, p, unit);
    761 #endif
    762 	if (unit >= numvnd)
    763 		return (ENXIO);
    764 
    765 	vnd = &vnd_softc[unit];
    766 	vio = (struct vnd_ioctl *)data;
    767 
    768 	/* Must be open for writes for these commands... */
    769 	switch (cmd) {
    770 	case VNDIOCSET:
    771 	case VNDIOCCLR:
    772 	case DIOCSDINFO:
    773 	case DIOCWDINFO:
    774 #ifdef __HAVE_OLD_DISKLABEL
    775 	case ODIOCSDINFO:
    776 	case ODIOCWDINFO:
    777 #endif
    778 	case DIOCKLABEL:
    779 	case DIOCWLABEL:
    780 		if ((flag & FWRITE) == 0)
    781 			return (EBADF);
    782 	}
    783 
    784 	/* Must be initialized for these... */
    785 	switch (cmd) {
    786 	case VNDIOCCLR:
    787 	case DIOCGDINFO:
    788 	case DIOCSDINFO:
    789 	case DIOCWDINFO:
    790 	case DIOCGPART:
    791 	case DIOCKLABEL:
    792 	case DIOCWLABEL:
    793 	case DIOCGDEFLABEL:
    794 #ifdef __HAVE_OLD_DISKLABEL
    795 	case ODIOCGDINFO:
    796 	case ODIOCSDINFO:
    797 	case ODIOCWDINFO:
    798 	case ODIOCGDEFLABEL:
    799 #endif
    800 		if ((vnd->sc_flags & VNF_INITED) == 0)
    801 			return (ENXIO);
    802 	}
    803 
    804 	switch (cmd) {
    805 	case VNDIOCSET:
    806 		if (vnd->sc_flags & VNF_INITED)
    807 			return (EBUSY);
    808 
    809 		if ((error = vndlock(vnd)) != 0)
    810 			return (error);
    811 
    812 		fflags = FREAD;
    813 		if ((vio->vnd_flags & VNDIOF_READONLY) == 0)
    814 			fflags |= FWRITE;
    815 		NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, vio->vnd_file, p);
    816 		if ((error = vn_open(&nd, fflags, 0)) != 0)
    817 			goto unlock_and_exit;
    818 		error = VOP_GETATTR(nd.ni_vp, &vattr, p->p_ucred, p);
    819 		VOP_UNLOCK(nd.ni_vp, 0);
    820 		if (!error && nd.ni_vp->v_type != VREG)
    821 			error = EOPNOTSUPP;
    822 		if (error)
    823 			goto close_and_exit;
    824 		vnd->sc_vp = nd.ni_vp;
    825 		vnd->sc_size = btodb(vattr.va_size);	/* note truncation */
    826 
    827 		/*
    828 		 * Use pseudo-geometry specified.  If none was provided,
    829 		 * use "standard" Adaptec fictitious geometry.
    830 		 */
    831 		if (vio->vnd_flags & VNDIOF_HASGEOM) {
    832 
    833 			memcpy(&vnd->sc_geom, &vio->vnd_geom,
    834 			    sizeof(vio->vnd_geom));
    835 
    836 			/*
    837 			 * Sanity-check the sector size.
    838 			 * XXX Don't allow secsize < DEV_BSIZE.  Should
    839 			 * XXX we?
    840 			 */
    841 			if (vnd->sc_geom.vng_secsize < DEV_BSIZE ||
    842 			    (vnd->sc_geom.vng_secsize % DEV_BSIZE) != 0) {
    843 				error = EINVAL;
    844 				goto close_and_exit;
    845 			}
    846 
    847 			/*
    848 			 * Compute the size (in DEV_BSIZE blocks) specified
    849 			 * by the geometry.
    850 			 */
    851 			geomsize = (vnd->sc_geom.vng_nsectors *
    852 			    vnd->sc_geom.vng_ntracks *
    853 			    vnd->sc_geom.vng_ncylinders) *
    854 			    (vnd->sc_geom.vng_secsize / DEV_BSIZE);
    855 
    856 			/*
    857 			 * Sanity-check the size against the specified
    858 			 * geometry.
    859 			 */
    860 			if (vnd->sc_size < geomsize) {
    861 				error = EINVAL;
    862 				goto close_and_exit;
    863 			}
    864 		} else {
    865 			/*
    866 			 * Size must be at least 2048 DEV_BSIZE blocks
    867 			 * (1M) in order to use this geometry.
    868 			 */
    869 			if (vnd->sc_size < (32 * 64)) {
    870 				error = EINVAL;
    871 				goto close_and_exit;
    872 			}
    873 
    874 			vnd->sc_geom.vng_secsize = DEV_BSIZE;
    875 			vnd->sc_geom.vng_nsectors = 32;
    876 			vnd->sc_geom.vng_ntracks = 64;
    877 			vnd->sc_geom.vng_ncylinders = vnd->sc_size / (64 * 32);
    878 		}
    879 
    880 		if (vio->vnd_flags & VNDIOF_READONLY) {
    881 			vnd->sc_flags |= VNF_READONLY;
    882 		}
    883 
    884 		if ((error = vndsetcred(vnd, p->p_ucred)) != 0)
    885 			goto close_and_exit;
    886 		vndthrottle(vnd, vnd->sc_vp);
    887 		vio->vnd_size = dbtob(vnd->sc_size);
    888 		vnd->sc_flags |= VNF_INITED;
    889 #ifdef DEBUG
    890 		if (vnddebug & VDB_INIT)
    891 			printf("vndioctl: SET vp %p size 0x%lx %d/%d/%d/%d\n",
    892 			    vnd->sc_vp, (unsigned long) vnd->sc_size,
    893 			    vnd->sc_geom.vng_secsize,
    894 			    vnd->sc_geom.vng_nsectors,
    895 			    vnd->sc_geom.vng_ntracks,
    896 			    vnd->sc_geom.vng_ncylinders);
    897 #endif
    898 
    899 		/* Attach the disk. */
    900 		memset(vnd->sc_xname, 0, sizeof(vnd->sc_xname)); /* XXX */
    901 		sprintf(vnd->sc_xname, "vnd%d", unit);		/* XXX */
    902 		vnd->sc_dkdev.dk_name = vnd->sc_xname;
    903 		disk_attach(&vnd->sc_dkdev);
    904 
    905 		/* Initialize the xfer and buffer pools. */
    906 		pool_init(&vnd->sc_vxpool, sizeof(struct vndxfer), 0,
    907 		    0, 0, "vndxpl", NULL);
    908 		pool_init(&vnd->sc_vbpool, sizeof(struct vndbuf), 0,
    909 		    0, 0, "vndbpl", NULL);
    910 
    911 		/* Try and read the disklabel. */
    912 		vndgetdisklabel(dev);
    913 
    914 		vndunlock(vnd);
    915 
    916 		break;
    917 
    918 close_and_exit:
    919 		(void) vn_close(nd.ni_vp, fflags, p->p_ucred, p);
    920 unlock_and_exit:
    921 		vndunlock(vnd);
    922 		return (error);
    923 
    924 	case VNDIOCCLR:
    925 		if ((error = vndlock(vnd)) != 0)
    926 			return (error);
    927 
    928 		/*
    929 		 * Don't unconfigure if any other partitions are open
    930 		 * or if both the character and block flavors of this
    931 		 * partition are open.
    932 		 */
    933 		part = DISKPART(dev);
    934 		pmask = (1 << part);
    935 		if (((vnd->sc_dkdev.dk_openmask & ~pmask) ||
    936 		    ((vnd->sc_dkdev.dk_bopenmask & pmask) &&
    937 		    (vnd->sc_dkdev.dk_copenmask & pmask))) &&
    938 			!(vio->vnd_flags & VNDIOF_FORCE)) {
    939 			vndunlock(vnd);
    940 			return (EBUSY);
    941 		}
    942 
    943 		/*
    944 		 * XXX vndclear() might call vndclose() implicitely;
    945 		 * release lock to avoid recursion
    946 		 */
    947 		vndunlock(vnd);
    948 		vndclear(vnd, minor(dev));
    949 #ifdef DEBUG
    950 		if (vnddebug & VDB_INIT)
    951 			printf("vndioctl: CLRed\n");
    952 #endif
    953 
    954 		/* Destroy the xfer and buffer pools. */
    955 		pool_destroy(&vnd->sc_vxpool);
    956 		pool_destroy(&vnd->sc_vbpool);
    957 
    958 		/* Detatch the disk. */
    959 		disk_detach(&vnd->sc_dkdev);
    960 
    961 		break;
    962 
    963 	case VNDIOCGET: {
    964 		struct vnd_user *vnu;
    965 		struct vattr va;
    966 
    967 		vnu = (struct vnd_user *)data;
    968 
    969 		if (vnu->vnu_unit == -1)
    970 			vnu->vnu_unit = unit;
    971 		if (vnu->vnu_unit >= numvnd)
    972 			return (ENXIO);
    973 		if (vnu->vnu_unit < 0)
    974 			return (EINVAL);
    975 
    976 		vnd = &vnd_softc[vnu->vnu_unit];
    977 
    978 		if (vnd->sc_flags & VNF_INITED) {
    979 			error = VOP_GETATTR(vnd->sc_vp, &va, p->p_ucred, p);
    980 			if (error)
    981 				return (error);
    982 			vnu->vnu_dev = va.va_fsid;
    983 			vnu->vnu_ino = va.va_fileid;
    984 		}
    985 		else {
    986 			/* unused is not an error */
    987 			vnu->vnu_dev = 0;
    988 			vnu->vnu_ino = 0;
    989 		}
    990 
    991 		break;
    992 	}
    993 
    994 	case DIOCGDINFO:
    995 		*(struct disklabel *)data = *(vnd->sc_dkdev.dk_label);
    996 		break;
    997 
    998 #ifdef __HAVE_OLD_DISKLABEL
    999 	case ODIOCGDINFO:
   1000 		newlabel = *(vnd->sc_dkdev.dk_label);
   1001 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
   1002 			return ENOTTY;
   1003 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
   1004 		break;
   1005 #endif
   1006 
   1007 	case DIOCGPART:
   1008 		((struct partinfo *)data)->disklab = vnd->sc_dkdev.dk_label;
   1009 		((struct partinfo *)data)->part =
   1010 		    &vnd->sc_dkdev.dk_label->d_partitions[DISKPART(dev)];
   1011 		break;
   1012 
   1013 	case DIOCWDINFO:
   1014 	case DIOCSDINFO:
   1015 #ifdef __HAVE_OLD_DISKLABEL
   1016 	case ODIOCWDINFO:
   1017 	case ODIOCSDINFO:
   1018 #endif
   1019 	{
   1020 		struct disklabel *lp;
   1021 
   1022 		if ((error = vndlock(vnd)) != 0)
   1023 			return (error);
   1024 
   1025 		vnd->sc_flags |= VNF_LABELLING;
   1026 
   1027 #ifdef __HAVE_OLD_DISKLABEL
   1028 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
   1029 			memset(&newlabel, 0, sizeof newlabel);
   1030 			memcpy(&newlabel, data, sizeof (struct olddisklabel));
   1031 			lp = &newlabel;
   1032 		} else
   1033 #endif
   1034 		lp = (struct disklabel *)data;
   1035 
   1036 		error = setdisklabel(vnd->sc_dkdev.dk_label,
   1037 		    lp, 0, vnd->sc_dkdev.dk_cpulabel);
   1038 		if (error == 0) {
   1039 			if (cmd == DIOCWDINFO
   1040 #ifdef __HAVE_OLD_DISKLABEL
   1041 			    || cmd == ODIOCWDINFO
   1042 #endif
   1043 			   )
   1044 				error = writedisklabel(VNDLABELDEV(dev),
   1045 				    vndstrategy, vnd->sc_dkdev.dk_label,
   1046 				    vnd->sc_dkdev.dk_cpulabel);
   1047 		}
   1048 
   1049 		vnd->sc_flags &= ~VNF_LABELLING;
   1050 
   1051 		vndunlock(vnd);
   1052 
   1053 		if (error)
   1054 			return (error);
   1055 		break;
   1056 	}
   1057 
   1058 	case DIOCKLABEL:
   1059 		if (*(int *)data != 0)
   1060 			vnd->sc_flags |= VNF_KLABEL;
   1061 		else
   1062 			vnd->sc_flags &= ~VNF_KLABEL;
   1063 		break;
   1064 
   1065 	case DIOCWLABEL:
   1066 		if (*(int *)data != 0)
   1067 			vnd->sc_flags |= VNF_WLABEL;
   1068 		else
   1069 			vnd->sc_flags &= ~VNF_WLABEL;
   1070 		break;
   1071 
   1072 	case DIOCGDEFLABEL:
   1073 		vndgetdefaultlabel(vnd, (struct disklabel *)data);
   1074 		break;
   1075 
   1076 #ifdef __HAVE_OLD_DISKLABEL
   1077 	case ODIOCGDEFLABEL:
   1078 		vndgetdefaultlabel(vnd, &newlabel);
   1079 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
   1080 			return ENOTTY;
   1081 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
   1082 		break;
   1083 #endif
   1084 
   1085 	default:
   1086 		return (ENOTTY);
   1087 	}
   1088 
   1089 	return (0);
   1090 }
   1091 
   1092 /*
   1093  * Duplicate the current processes' credentials.  Since we are called only
   1094  * as the result of a SET ioctl and only root can do that, any future access
   1095  * to this "disk" is essentially as root.  Note that credentials may change
   1096  * if some other uid can write directly to the mapped file (NFS).
   1097  */
   1098 int
   1099 vndsetcred(vnd, cred)
   1100 	struct vnd_softc *vnd;
   1101 	struct ucred *cred;
   1102 {
   1103 	struct uio auio;
   1104 	struct iovec aiov;
   1105 	char *tmpbuf;
   1106 	int error;
   1107 
   1108 	vnd->sc_cred = crdup(cred);
   1109 	tmpbuf = malloc(DEV_BSIZE, M_TEMP, M_WAITOK);
   1110 
   1111 	/* XXX: Horrible kludge to establish credentials for NFS */
   1112 	aiov.iov_base = tmpbuf;
   1113 	aiov.iov_len = min(DEV_BSIZE, dbtob(vnd->sc_size));
   1114 	auio.uio_iov = &aiov;
   1115 	auio.uio_iovcnt = 1;
   1116 	auio.uio_offset = 0;
   1117 	auio.uio_rw = UIO_READ;
   1118 	auio.uio_segflg = UIO_SYSSPACE;
   1119 	auio.uio_resid = aiov.iov_len;
   1120 	vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
   1121 	error = VOP_READ(vnd->sc_vp, &auio, 0, vnd->sc_cred);
   1122 	if (error == 0) {
   1123 		/*
   1124 		 * Because vnd does all IO directly through the vnode
   1125 		 * we need to flush (at least) the buffer from the above
   1126 		 * VOP_READ from the buffer cache to prevent cache
   1127 		 * incoherencies.  Also, be careful to write dirty
   1128 		 * buffers back to stable storage.
   1129 		 */
   1130 		error = vinvalbuf(vnd->sc_vp, V_SAVE, vnd->sc_cred,
   1131 			    curproc, 0, 0);
   1132 	}
   1133 	VOP_UNLOCK(vnd->sc_vp, 0);
   1134 
   1135 	free(tmpbuf, M_TEMP);
   1136 	return (error);
   1137 }
   1138 
   1139 /*
   1140  * Set maxactive based on FS type
   1141  */
   1142 void
   1143 vndthrottle(vnd, vp)
   1144 	struct vnd_softc *vnd;
   1145 	struct vnode *vp;
   1146 {
   1147 #ifdef NFS
   1148 	extern int (**nfsv2_vnodeop_p) __P((void *));
   1149 
   1150 	if (vp->v_op == nfsv2_vnodeop_p)
   1151 		vnd->sc_maxactive = 2;
   1152 	else
   1153 #endif
   1154 		vnd->sc_maxactive = 8;
   1155 
   1156 	if (vnd->sc_maxactive < 1)
   1157 		vnd->sc_maxactive = 1;
   1158 }
   1159 
   1160 #if 0
   1161 void
   1162 vndshutdown()
   1163 {
   1164 	struct vnd_softc *vnd;
   1165 
   1166 	for (vnd = &vnd_softc[0]; vnd < &vnd_softc[numvnd]; vnd++)
   1167 		if (vnd->sc_flags & VNF_INITED)
   1168 			vndclear(vnd);
   1169 }
   1170 #endif
   1171 
   1172 void
   1173 vndclear(vnd, myminor)
   1174 	struct vnd_softc *vnd;
   1175 	int myminor;
   1176 {
   1177 	struct vnode *vp = vnd->sc_vp;
   1178 	struct proc *p = curproc;		/* XXX */
   1179 	int fflags = FREAD;
   1180 	int bmaj, cmaj, i, mn;
   1181 
   1182 #ifdef DEBUG
   1183 	if (vnddebug & VDB_FOLLOW)
   1184 		printf("vndclear(%p): vp %p\n", vnd, vp);
   1185 #endif
   1186 	/* locate the major number */
   1187 	bmaj = bdevsw_lookup_major(&vnd_bdevsw);
   1188 	cmaj = cdevsw_lookup_major(&vnd_cdevsw);
   1189 
   1190 	/* Nuke the vnodes for any open instances */
   1191 	for (i = 0; i < MAXPARTITIONS; i++) {
   1192 		mn = DISKMINOR(vnd->sc_unit, i);
   1193 		vdevgone(bmaj, mn, mn, VBLK);
   1194 		if (mn != myminor) /* XXX avoid to kill own vnode */
   1195 			vdevgone(cmaj, mn, mn, VCHR);
   1196 	}
   1197 
   1198 	if ((vnd->sc_flags & VNF_READONLY) == 0)
   1199 		fflags |= FWRITE;
   1200 	vnd->sc_flags &= ~(VNF_INITED | VNF_READONLY | VNF_VLABEL);
   1201 	if (vp == (struct vnode *)0)
   1202 		panic("vndioctl: null vp");
   1203 	(void) vn_close(vp, fflags, vnd->sc_cred, p);
   1204 	crfree(vnd->sc_cred);
   1205 	vnd->sc_vp = (struct vnode *)0;
   1206 	vnd->sc_cred = (struct ucred *)0;
   1207 	vnd->sc_size = 0;
   1208 }
   1209 
   1210 int
   1211 vndsize(dev)
   1212 	dev_t dev;
   1213 {
   1214 	struct vnd_softc *sc;
   1215 	struct disklabel *lp;
   1216 	int part, unit, omask;
   1217 	int size;
   1218 
   1219 	unit = vndunit(dev);
   1220 	if (unit >= numvnd)
   1221 		return (-1);
   1222 	sc = &vnd_softc[unit];
   1223 
   1224 	if ((sc->sc_flags & VNF_INITED) == 0)
   1225 		return (-1);
   1226 
   1227 	part = DISKPART(dev);
   1228 	omask = sc->sc_dkdev.dk_openmask & (1 << part);
   1229 	lp = sc->sc_dkdev.dk_label;
   1230 
   1231 	if (omask == 0 && vndopen(dev, 0, S_IFBLK, curproc))
   1232 		return (-1);
   1233 
   1234 	if (lp->d_partitions[part].p_fstype != FS_SWAP)
   1235 		size = -1;
   1236 	else
   1237 		size = lp->d_partitions[part].p_size *
   1238 		    (lp->d_secsize / DEV_BSIZE);
   1239 
   1240 	if (omask == 0 && vndclose(dev, 0, S_IFBLK, curproc))
   1241 		return (-1);
   1242 
   1243 	return (size);
   1244 }
   1245 
   1246 int
   1247 vnddump(dev, blkno, va, size)
   1248 	dev_t dev;
   1249 	daddr_t blkno;
   1250 	caddr_t va;
   1251 	size_t size;
   1252 {
   1253 
   1254 	/* Not implemented. */
   1255 	return ENXIO;
   1256 }
   1257 
   1258 void
   1259 vndgetdefaultlabel(sc, lp)
   1260 	struct vnd_softc *sc;
   1261 	struct disklabel *lp;
   1262 {
   1263 	struct vndgeom *vng = &sc->sc_geom;
   1264 	struct partition *pp;
   1265 
   1266 	memset(lp, 0, sizeof(*lp));
   1267 
   1268 	lp->d_secperunit = sc->sc_size / (vng->vng_secsize / DEV_BSIZE);
   1269 	lp->d_secsize = vng->vng_secsize;
   1270 	lp->d_nsectors = vng->vng_nsectors;
   1271 	lp->d_ntracks = vng->vng_ntracks;
   1272 	lp->d_ncylinders = vng->vng_ncylinders;
   1273 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
   1274 
   1275 	strncpy(lp->d_typename, "vnd", sizeof(lp->d_typename));
   1276 	lp->d_type = DTYPE_VND;
   1277 	strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
   1278 	lp->d_rpm = 3600;
   1279 	lp->d_interleave = 1;
   1280 	lp->d_flags = 0;
   1281 
   1282 	pp = &lp->d_partitions[RAW_PART];
   1283 	pp->p_offset = 0;
   1284 	pp->p_size = lp->d_secperunit;
   1285 	pp->p_fstype = FS_UNUSED;
   1286 	lp->d_npartitions = RAW_PART + 1;
   1287 
   1288 	lp->d_magic = DISKMAGIC;
   1289 	lp->d_magic2 = DISKMAGIC;
   1290 	lp->d_checksum = dkcksum(lp);
   1291 }
   1292 
   1293 /*
   1294  * Read the disklabel from a vnd.  If one is not present, create a fake one.
   1295  */
   1296 void
   1297 vndgetdisklabel(dev)
   1298 	dev_t dev;
   1299 {
   1300 	struct vnd_softc *sc = &vnd_softc[vndunit(dev)];
   1301 	const char *errstring;
   1302 	struct disklabel *lp = sc->sc_dkdev.dk_label;
   1303 	struct cpu_disklabel *clp = sc->sc_dkdev.dk_cpulabel;
   1304 	int i;
   1305 
   1306 	memset(clp, 0, sizeof(*clp));
   1307 
   1308 	vndgetdefaultlabel(sc, lp);
   1309 
   1310 	/*
   1311 	 * Call the generic disklabel extraction routine.
   1312 	 */
   1313 	errstring = readdisklabel(VNDLABELDEV(dev), vndstrategy, lp, clp);
   1314 	if (errstring) {
   1315 		/*
   1316 		 * Lack of disklabel is common, but we print the warning
   1317 		 * anyway, since it might contain other useful information.
   1318 		 */
   1319 		printf("%s: %s\n", sc->sc_xname, errstring);
   1320 
   1321 		/*
   1322 		 * For historical reasons, if there's no disklabel
   1323 		 * present, all partitions must be FS_BSDFFS and
   1324 		 * occupy the entire disk.
   1325 		 */
   1326 		for (i = 0; i < MAXPARTITIONS; i++) {
   1327 			/*
   1328 			 * Don't wipe out port specific hack (such as
   1329 			 * dos partition hack of i386 port).
   1330 			 */
   1331 			if (lp->d_partitions[i].p_size != 0)
   1332 				continue;
   1333 
   1334 			lp->d_partitions[i].p_size = lp->d_secperunit;
   1335 			lp->d_partitions[i].p_offset = 0;
   1336 			lp->d_partitions[i].p_fstype = FS_BSDFFS;
   1337 		}
   1338 
   1339 		strncpy(lp->d_packname, "default label",
   1340 		    sizeof(lp->d_packname));
   1341 
   1342 		lp->d_npartitions = MAXPARTITIONS;
   1343 		lp->d_checksum = dkcksum(lp);
   1344 	}
   1345 
   1346 	/* In-core label now valid. */
   1347 	sc->sc_flags |= VNF_VLABEL;
   1348 }
   1349 
   1350 /*
   1351  * Wait interruptibly for an exclusive lock.
   1352  *
   1353  * XXX
   1354  * Several drivers do this; it should be abstracted and made MP-safe.
   1355  */
   1356 static int
   1357 vndlock(sc)
   1358 	struct vnd_softc *sc;
   1359 {
   1360 	int error;
   1361 
   1362 	while ((sc->sc_flags & VNF_LOCKED) != 0) {
   1363 		sc->sc_flags |= VNF_WANTED;
   1364 		if ((error = tsleep(sc, PRIBIO | PCATCH, "vndlck", 0)) != 0)
   1365 			return (error);
   1366 	}
   1367 	sc->sc_flags |= VNF_LOCKED;
   1368 	return (0);
   1369 }
   1370 
   1371 /*
   1372  * Unlock and wake up any waiters.
   1373  */
   1374 static void
   1375 vndunlock(sc)
   1376 	struct vnd_softc *sc;
   1377 {
   1378 
   1379 	sc->sc_flags &= ~VNF_LOCKED;
   1380 	if ((sc->sc_flags & VNF_WANTED) != 0) {
   1381 		sc->sc_flags &= ~VNF_WANTED;
   1382 		wakeup(sc);
   1383 	}
   1384 }
   1385