Home | History | Annotate | Line # | Download | only in dev
vnd.c revision 1.106
      1 /*	$NetBSD: vnd.c,v 1.106 2004/01/25 18:06:48 hannken 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) 1990, 1993
     41  *	The Regents of the University of California.  All rights reserved.
     42  *
     43  * This code is derived from software contributed to Berkeley by
     44  * the Systems Programming Group of the University of Utah Computer
     45  * Science Department.
     46  *
     47  * Redistribution and use in source and binary forms, with or without
     48  * modification, are permitted provided that the following conditions
     49  * are met:
     50  * 1. Redistributions of source code must retain the above copyright
     51  *    notice, this list of conditions and the following disclaimer.
     52  * 2. Redistributions in binary form must reproduce the above copyright
     53  *    notice, this list of conditions and the following disclaimer in the
     54  *    documentation and/or other materials provided with the distribution.
     55  * 3. Neither the name of the University nor the names of its contributors
     56  *    may be used to endorse or promote products derived from this software
     57  *    without specific prior written permission.
     58  *
     59  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     60  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     61  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     62  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     63  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     64  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     65  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     66  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     67  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     68  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     69  * SUCH DAMAGE.
     70  *
     71  * from: Utah $Hdr: vn.c 1.13 94/04/02$
     72  *
     73  *	@(#)vn.c	8.9 (Berkeley) 5/14/95
     74  */
     75 
     76 /*
     77  * Copyright (c) 1988 University of Utah.
     78  *
     79  * This code is derived from software contributed to Berkeley by
     80  * the Systems Programming Group of the University of Utah Computer
     81  * Science Department.
     82  *
     83  * Redistribution and use in source and binary forms, with or without
     84  * modification, are permitted provided that the following conditions
     85  * are met:
     86  * 1. Redistributions of source code must retain the above copyright
     87  *    notice, this list of conditions and the following disclaimer.
     88  * 2. Redistributions in binary form must reproduce the above copyright
     89  *    notice, this list of conditions and the following disclaimer in the
     90  *    documentation and/or other materials provided with the distribution.
     91  * 3. All advertising materials mentioning features or use of this software
     92  *    must display the following acknowledgement:
     93  *	This product includes software developed by the University of
     94  *	California, Berkeley and its contributors.
     95  * 4. Neither the name of the University nor the names of its contributors
     96  *    may be used to endorse or promote products derived from this software
     97  *    without specific prior written permission.
     98  *
     99  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
    100  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
    101  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
    102  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
    103  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
    104  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
    105  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
    106  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
    107  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
    108  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
    109  * SUCH DAMAGE.
    110  *
    111  * from: Utah $Hdr: vn.c 1.13 94/04/02$
    112  *
    113  *	@(#)vn.c	8.9 (Berkeley) 5/14/95
    114  */
    115 
    116 /*
    117  * Vnode disk driver.
    118  *
    119  * Block/character interface to a vnode.  Allows one to treat a file
    120  * as a disk (e.g. build a filesystem in it, mount it, etc.).
    121  *
    122  * NOTE 1: This uses the VOP_BMAP/VOP_STRATEGY interface to the vnode
    123  * instead of a simple VOP_RDWR.  We do this to avoid distorting the
    124  * local buffer cache.
    125  *
    126  * NOTE 2: There is a security issue involved with this driver.
    127  * Once mounted all access to the contents of the "mapped" file via
    128  * the special file is controlled by the permissions on the special
    129  * file, the protection of the mapped file is ignored (effectively,
    130  * by using root credentials in all transactions).
    131  *
    132  * NOTE 3: Doesn't interact with leases, should it?
    133  */
    134 
    135 #include <sys/cdefs.h>
    136 __KERNEL_RCSID(0, "$NetBSD: vnd.c,v 1.106 2004/01/25 18:06:48 hannken Exp $");
    137 
    138 #if defined(_KERNEL_OPT)
    139 #include "fs_nfs.h"
    140 #endif
    141 
    142 #include <sys/param.h>
    143 #include <sys/systm.h>
    144 #include <sys/namei.h>
    145 #include <sys/proc.h>
    146 #include <sys/errno.h>
    147 #include <sys/buf.h>
    148 #include <sys/malloc.h>
    149 #include <sys/ioctl.h>
    150 #include <sys/disklabel.h>
    151 #include <sys/device.h>
    152 #include <sys/disk.h>
    153 #include <sys/stat.h>
    154 #include <sys/mount.h>
    155 #include <sys/vnode.h>
    156 #include <sys/file.h>
    157 #include <sys/uio.h>
    158 #include <sys/conf.h>
    159 
    160 #include <miscfs/specfs/specdev.h>
    161 
    162 #include <dev/vndvar.h>
    163 
    164 #if defined(VNDDEBUG) && !defined(DEBUG)
    165 #define	DEBUG
    166 #endif
    167 
    168 #ifdef DEBUG
    169 int dovndcluster = 1;
    170 #define	VDB_FOLLOW	0x01
    171 #define	VDB_INIT	0x02
    172 #define	VDB_IO		0x04
    173 #define	VDB_LABEL	0x08
    174 int vnddebug = 0x00;
    175 #endif
    176 
    177 #define	vndunit(x)	DISKUNIT(x)
    178 
    179 struct vndxfer {
    180 	struct buf	*vx_bp;		/* Pointer to parent buffer */
    181 	int		vx_error;
    182 	int		vx_pending;	/* # of pending aux buffers */
    183 	int		vx_flags;
    184 #define VX_BUSY		1
    185 };
    186 
    187 struct vndbuf {
    188 	struct buf	vb_buf;
    189 	struct vndxfer	*vb_xfer;
    190 };
    191 
    192 #define	VND_GETXFER(vnd)	pool_get(&(vnd)->sc_vxpool, PR_NOWAIT)
    193 #define	VND_PUTXFER(vnd, vx)	pool_put(&(vnd)->sc_vxpool, (vx))
    194 
    195 #define	VND_GETBUF(vnd)		pool_get(&(vnd)->sc_vbpool, PR_NOWAIT)
    196 #define	VND_PUTBUF(vnd, vb)	pool_put(&(vnd)->sc_vbpool, (vb))
    197 
    198 struct vnd_softc *vnd_softc;
    199 int numvnd = 0;
    200 
    201 #define	VNDLABELDEV(dev) \
    202 	(MAKEDISKDEV(major((dev)), vndunit((dev)), RAW_PART))
    203 
    204 /* called by main() at boot time (XXX: and the LKM driver) */
    205 void	vndattach __P((int));
    206 int	vnddetach __P((void));
    207 
    208 void	vndclear __P((struct vnd_softc *, int));
    209 void	vndstart __P((struct vnd_softc *));
    210 int	vndsetcred __P((struct vnd_softc *, struct ucred *));
    211 void	vndthrottle __P((struct vnd_softc *, struct vnode *));
    212 void	vndiodone __P((struct buf *));
    213 #if 0
    214 void	vndshutdown __P((void));
    215 #endif
    216 
    217 void	vndgetdefaultlabel __P((struct vnd_softc *, struct disklabel *));
    218 void	vndgetdisklabel __P((dev_t));
    219 
    220 static	int vndlock __P((struct vnd_softc *));
    221 static	void vndunlock __P((struct vnd_softc *));
    222 
    223 dev_type_open(vndopen);
    224 dev_type_close(vndclose);
    225 dev_type_read(vndread);
    226 dev_type_write(vndwrite);
    227 dev_type_ioctl(vndioctl);
    228 dev_type_strategy(vndstrategy);
    229 dev_type_dump(vnddump);
    230 dev_type_size(vndsize);
    231 
    232 const struct bdevsw vnd_bdevsw = {
    233 	vndopen, vndclose, vndstrategy, vndioctl, vnddump, vndsize, D_DISK
    234 };
    235 
    236 const struct cdevsw vnd_cdevsw = {
    237 	vndopen, vndclose, vndread, vndwrite, vndioctl,
    238 	nostop, notty, nopoll, nommap, nokqfilter, D_DISK
    239 };
    240 
    241 int vndattached = 0;
    242 
    243 void
    244 vndattach(num)
    245 	int num;
    246 {
    247 	int i;
    248 	char *mem;
    249 
    250 	if (vndattached)
    251 		return;
    252 	vndattached = 1;
    253 	if (num <= 0)
    254 		return;
    255 	i = num * sizeof(struct vnd_softc);
    256 	mem = malloc(i, M_DEVBUF, M_NOWAIT|M_ZERO);
    257 	if (mem == NULL) {
    258 		printf("WARNING: no memory for vnode disks\n");
    259 		return;
    260 	}
    261 	vnd_softc = (struct vnd_softc *)mem;
    262 	numvnd = num;
    263 
    264 	for (i = 0; i < numvnd; i++) {
    265 		vnd_softc[i].sc_unit = i;
    266 		bufq_alloc(&vnd_softc[i].sc_tab,
    267 		    BUFQ_DISKSORT|BUFQ_SORT_RAWBLOCK);
    268 	}
    269 }
    270 
    271 int
    272 vnddetach()
    273 {
    274 	int i;
    275 
    276 	/* First check we aren't in use. */
    277 	for (i = 0; i < numvnd; i++)
    278 		if (vnd_softc[i].sc_flags & VNF_INITED)
    279 			return (EBUSY);
    280 
    281 	for (i = 0; i < numvnd; i++)
    282 		bufq_free(&vnd_softc[i].sc_tab);
    283 
    284 	free(vnd_softc, M_DEVBUF);
    285 	vndattached = 0;
    286 
    287 	return (0);
    288 }
    289 
    290 int
    291 vndopen(dev, flags, mode, p)
    292 	dev_t dev;
    293 	int flags, mode;
    294 	struct proc *p;
    295 {
    296 	int unit = vndunit(dev);
    297 	struct vnd_softc *sc;
    298 	int error = 0, part, pmask;
    299 	struct disklabel *lp;
    300 
    301 #ifdef DEBUG
    302 	if (vnddebug & VDB_FOLLOW)
    303 		printf("vndopen(0x%x, 0x%x, 0x%x, %p)\n", dev, flags, mode, p);
    304 #endif
    305 	if (unit >= numvnd)
    306 		return (ENXIO);
    307 	sc = &vnd_softc[unit];
    308 
    309 	if ((error = vndlock(sc)) != 0)
    310 		return (error);
    311 
    312 	lp = sc->sc_dkdev.dk_label;
    313 
    314 	part = DISKPART(dev);
    315 	pmask = (1 << part);
    316 
    317 	/*
    318 	 * If we're initialized, check to see if there are any other
    319 	 * open partitions.  If not, then it's safe to update the
    320 	 * in-core disklabel.  Only read the disklabel if it is
    321 	 * not realdy valid.
    322 	 */
    323 	if ((sc->sc_flags & (VNF_INITED|VNF_VLABEL)) == VNF_INITED &&
    324 	    sc->sc_dkdev.dk_openmask == 0)
    325 		vndgetdisklabel(dev);
    326 
    327 	/* Check that the partitions exists. */
    328 	if (part != RAW_PART) {
    329 		if (((sc->sc_flags & VNF_INITED) == 0) ||
    330 		    ((part >= lp->d_npartitions) ||
    331 		     (lp->d_partitions[part].p_fstype == FS_UNUSED))) {
    332 			error = ENXIO;
    333 			goto done;
    334 		}
    335 	}
    336 
    337 	/* Prevent our unit from being unconfigured while open. */
    338 	switch (mode) {
    339 	case S_IFCHR:
    340 		sc->sc_dkdev.dk_copenmask |= pmask;
    341 		break;
    342 
    343 	case S_IFBLK:
    344 		sc->sc_dkdev.dk_bopenmask |= pmask;
    345 		break;
    346 	}
    347 	sc->sc_dkdev.dk_openmask =
    348 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
    349 
    350  done:
    351 	vndunlock(sc);
    352 	return (error);
    353 }
    354 
    355 int
    356 vndclose(dev, flags, mode, p)
    357 	dev_t dev;
    358 	int flags, mode;
    359 	struct proc *p;
    360 {
    361 	int unit = vndunit(dev);
    362 	struct vnd_softc *sc;
    363 	int error = 0, part;
    364 
    365 #ifdef DEBUG
    366 	if (vnddebug & VDB_FOLLOW)
    367 		printf("vndclose(0x%x, 0x%x, 0x%x, %p)\n", dev, flags, mode, p);
    368 #endif
    369 
    370 	if (unit >= numvnd)
    371 		return (ENXIO);
    372 	sc = &vnd_softc[unit];
    373 
    374 	if ((error = vndlock(sc)) != 0)
    375 		return (error);
    376 
    377 	part = DISKPART(dev);
    378 
    379 	/* ...that much closer to allowing unconfiguration... */
    380 	switch (mode) {
    381 	case S_IFCHR:
    382 		sc->sc_dkdev.dk_copenmask &= ~(1 << part);
    383 		break;
    384 
    385 	case S_IFBLK:
    386 		sc->sc_dkdev.dk_bopenmask &= ~(1 << part);
    387 		break;
    388 	}
    389 	sc->sc_dkdev.dk_openmask =
    390 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
    391 
    392 	if (sc->sc_dkdev.dk_openmask == 0) {
    393 		if ((sc->sc_flags & VNF_KLABEL) == 0)
    394 			sc->sc_flags &= ~VNF_VLABEL;
    395 	}
    396 
    397 	vndunlock(sc);
    398 	return (0);
    399 }
    400 
    401 /*
    402  * Break the request into bsize pieces and submit using VOP_BMAP/VOP_STRATEGY.
    403  */
    404 void
    405 vndstrategy(bp)
    406 	struct buf *bp;
    407 {
    408 	int unit = vndunit(bp->b_dev);
    409 	struct vnd_softc *vnd = &vnd_softc[unit];
    410 	struct vndxfer *vnx;
    411 	struct mount *mp;
    412 	int s, bsize, resid;
    413 	off_t bn;
    414 	caddr_t addr;
    415 	int sz, flags, error, wlabel;
    416 	struct disklabel *lp;
    417 	struct partition *pp;
    418 
    419 #ifdef DEBUG
    420 	if (vnddebug & VDB_FOLLOW)
    421 		printf("vndstrategy(%p): unit %d\n", bp, unit);
    422 #endif
    423 	if ((vnd->sc_flags & VNF_INITED) == 0) {
    424 		bp->b_error = ENXIO;
    425 		bp->b_flags |= B_ERROR;
    426 		goto done;
    427 	}
    428 
    429 	/* If it's a nil transfer, wake up the top half now. */
    430 	if (bp->b_bcount == 0)
    431 		goto done;
    432 
    433 	lp = vnd->sc_dkdev.dk_label;
    434 
    435 	/*
    436 	 * The transfer must be a whole number of blocks.
    437 	 */
    438 	if ((bp->b_bcount % lp->d_secsize) != 0) {
    439 		bp->b_error = EINVAL;
    440 		bp->b_flags |= B_ERROR;
    441 		goto done;
    442 	}
    443 
    444 	/*
    445 	 * Do bounds checking and adjust transfer.  If there's an error,
    446 	 * the bounds check will flag that for us.
    447 	 */
    448 	wlabel = vnd->sc_flags & (VNF_WLABEL|VNF_LABELLING);
    449 	if (DISKPART(bp->b_dev) != RAW_PART)
    450 		if (bounds_check_with_label(&vnd->sc_dkdev, bp, wlabel) <= 0)
    451 			goto done;
    452 
    453 	/*
    454 	 * check if we're read-only.
    455 	 */
    456 	if ((vnd->sc_flags & VNF_READONLY) && !(bp->b_flags & B_READ)) {
    457 		bp->b_error = EACCES;
    458 		bp->b_flags |= B_ERROR;
    459 		goto done;
    460 	}
    461 
    462 	bp->b_resid = bp->b_bcount;
    463 
    464 	/*
    465 	 * Put the block number in terms of the logical blocksize
    466 	 * of the "device".
    467 	 */
    468 	bn = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
    469 
    470 	/*
    471 	 * Translate the partition-relative block number to an absolute.
    472 	 */
    473 	if (DISKPART(bp->b_dev) != RAW_PART) {
    474 		pp = &vnd->sc_dkdev.dk_label->d_partitions[DISKPART(bp->b_dev)];
    475 		bn += pp->p_offset;
    476 	}
    477 
    478 	/* ...and convert to a byte offset within the file. */
    479 	bn *= lp->d_secsize;
    480 
    481 	if (vnd->sc_vp->v_mount == NULL) {
    482 		bp->b_error = ENXIO;
    483 		bp->b_flags |= B_ERROR;
    484 		goto done;
    485 	}
    486  	bsize = vnd->sc_vp->v_mount->mnt_stat.f_iosize;
    487 	addr = bp->b_data;
    488 	flags = (bp->b_flags & (B_READ|B_ASYNC)) | B_CALL;
    489 
    490 	/* Allocate a header for this transfer and link it to the buffer */
    491 	s = splbio();
    492 	vnx = VND_GETXFER(vnd);
    493 	splx(s);
    494 	vnx->vx_flags = VX_BUSY;
    495 	vnx->vx_error = 0;
    496 	vnx->vx_pending = 0;
    497 	vnx->vx_bp = bp;
    498 
    499 	if ((flags & B_READ) == 0)
    500 		vn_start_write(vnd->sc_vp, &mp, V_WAIT);
    501 
    502 	for (resid = bp->b_resid; resid; resid -= sz) {
    503 		struct vndbuf *nbp;
    504 		struct vnode *vp;
    505 		daddr_t nbn;
    506 		int off, nra;
    507 
    508 		nra = 0;
    509 		vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY | LK_CANRECURSE);
    510 		error = VOP_BMAP(vnd->sc_vp, bn / bsize, &vp, &nbn, &nra);
    511 		VOP_UNLOCK(vnd->sc_vp, 0);
    512 
    513 		if (error == 0 && (long)nbn == -1)
    514 			error = EIO;
    515 
    516 		/*
    517 		 * If there was an error or a hole in the file...punt.
    518 		 * Note that we may have to wait for any operations
    519 		 * that we have already fired off before releasing
    520 		 * the buffer.
    521 		 *
    522 		 * XXX we could deal with holes here but it would be
    523 		 * a hassle (in the write case).
    524 		 */
    525 		if (error) {
    526 			s = splbio();
    527 			vnx->vx_error = error;
    528 			goto out;
    529 		}
    530 
    531 #ifdef DEBUG
    532 		if (!dovndcluster)
    533 			nra = 0;
    534 #endif
    535 
    536 		if ((off = bn % bsize) != 0)
    537 			sz = bsize - off;
    538 		else
    539 			sz = (1 + nra) * bsize;
    540 		if (resid < sz)
    541 			sz = resid;
    542 #ifdef DEBUG
    543 		if (vnddebug & VDB_IO)
    544 			printf("vndstrategy: vp %p/%p bn 0x%qx/0x%" PRIx64
    545 			       " sz 0x%x\n",
    546 			    vnd->sc_vp, vp, (long long)bn, nbn, sz);
    547 #endif
    548 
    549 		s = splbio();
    550 		nbp = VND_GETBUF(vnd);
    551 		splx(s);
    552 		BUF_INIT(&nbp->vb_buf);
    553 		nbp->vb_buf.b_flags = flags;
    554 		nbp->vb_buf.b_bcount = sz;
    555 		nbp->vb_buf.b_bufsize = round_page((ulong)addr + sz)
    556 		    - trunc_page((ulong) addr);
    557 		nbp->vb_buf.b_error = 0;
    558 		nbp->vb_buf.b_data = addr;
    559 		nbp->vb_buf.b_blkno = nbp->vb_buf.b_rawblkno = nbn + btodb(off);
    560 		nbp->vb_buf.b_proc = bp->b_proc;
    561 		nbp->vb_buf.b_iodone = vndiodone;
    562 		nbp->vb_buf.b_vp = NULLVP;
    563 
    564 		nbp->vb_xfer = vnx;
    565 
    566 		BIO_COPYPRIO(&nbp->vb_buf, bp);
    567 
    568 		/*
    569 		 * Just sort by block number
    570 		 */
    571 		s = splbio();
    572 		if (vnx->vx_error != 0) {
    573 			VND_PUTBUF(vnd, nbp);
    574 			goto out;
    575 		}
    576 		vnx->vx_pending++;
    577 		bgetvp(vp, &nbp->vb_buf);
    578 		BUFQ_PUT(&vnd->sc_tab, &nbp->vb_buf);
    579 		vndstart(vnd);
    580 		splx(s);
    581 		bn += sz;
    582 		addr += sz;
    583 	}
    584 
    585 	s = splbio();
    586 
    587 out: /* Arrive here at splbio */
    588 	if ((flags & B_READ) == 0)
    589 		vn_finished_write(mp, 0);
    590 	vnx->vx_flags &= ~VX_BUSY;
    591 	if (vnx->vx_pending == 0) {
    592 		if (vnx->vx_error != 0) {
    593 			bp->b_error = vnx->vx_error;
    594 			bp->b_flags |= B_ERROR;
    595 		}
    596 		VND_PUTXFER(vnd, vnx);
    597 		biodone(bp);
    598 	}
    599 	splx(s);
    600 	return;
    601 
    602  done:
    603 	biodone(bp);
    604 }
    605 
    606 /*
    607  * Feed requests sequentially.
    608  * We do it this way to keep from flooding NFS servers if we are connected
    609  * to an NFS file.  This places the burden on the client rather than the
    610  * server.
    611  */
    612 void
    613 vndstart(vnd)
    614 	struct vnd_softc *vnd;
    615 {
    616 	struct buf	*bp;
    617 
    618 	/*
    619 	 * Dequeue now since lower level strategy routine might
    620 	 * queue using same links
    621 	 */
    622 
    623 	if ((vnd->sc_flags & VNF_BUSY) != 0)
    624 		return;
    625 
    626 	vnd->sc_flags |= VNF_BUSY;
    627 
    628 	while (vnd->sc_active < vnd->sc_maxactive) {
    629 		bp = BUFQ_GET(&vnd->sc_tab);
    630 		if (bp == NULL)
    631 			break;
    632 		vnd->sc_active++;
    633 #ifdef DEBUG
    634 		if (vnddebug & VDB_IO)
    635 			printf("vndstart(%ld): bp %p vp %p blkno 0x%" PRIx64
    636 				" flags %lx addr %p cnt 0x%lx\n",
    637 			    (long) (vnd-vnd_softc), bp, bp->b_vp, bp->b_blkno,
    638 			    bp->b_flags, bp->b_data, bp->b_bcount);
    639 #endif
    640 
    641 		/* Instrumentation. */
    642 		disk_busy(&vnd->sc_dkdev);
    643 
    644 		if ((bp->b_flags & B_READ) == 0)
    645 			bp->b_vp->v_numoutput++;
    646 		VOP_STRATEGY(bp->b_vp, bp);
    647 	}
    648 	vnd->sc_flags &= ~VNF_BUSY;
    649 }
    650 
    651 void
    652 vndiodone(bp)
    653 	struct buf *bp;
    654 {
    655 	struct vndbuf *vbp = (struct vndbuf *) bp;
    656 	struct vndxfer *vnx = (struct vndxfer *)vbp->vb_xfer;
    657 	struct buf *pbp = vnx->vx_bp;
    658 	struct vnd_softc *vnd = &vnd_softc[vndunit(pbp->b_dev)];
    659 	int s, resid;
    660 
    661 	s = splbio();
    662 #ifdef DEBUG
    663 	if (vnddebug & VDB_IO)
    664 		printf("vndiodone(%ld): vbp %p vp %p blkno 0x%" PRIx64
    665 		       " addr %p cnt 0x%lx\n",
    666 		    (long) (vnd-vnd_softc), vbp, vbp->vb_buf.b_vp,
    667 		    vbp->vb_buf.b_blkno, vbp->vb_buf.b_data,
    668 		    vbp->vb_buf.b_bcount);
    669 #endif
    670 
    671 	resid = vbp->vb_buf.b_bcount - vbp->vb_buf.b_resid;
    672 	pbp->b_resid -= resid;
    673 	disk_unbusy(&vnd->sc_dkdev, resid, (pbp->b_flags & B_READ));
    674 	vnx->vx_pending--;
    675 
    676 	if (vbp->vb_buf.b_error) {
    677 #ifdef DEBUG
    678 		if (vnddebug & VDB_IO)
    679 			printf("vndiodone: vbp %p error %d\n", vbp,
    680 			    vbp->vb_buf.b_error);
    681 #endif
    682 		vnx->vx_error = vbp->vb_buf.b_error;
    683 	}
    684 
    685 	if (vbp->vb_buf.b_vp != NULLVP)
    686 		brelvp(&vbp->vb_buf);
    687 
    688 	VND_PUTBUF(vnd, vbp);
    689 
    690 	/*
    691 	 * Wrap up this transaction if it has run to completion or, in
    692 	 * case of an error, when all auxiliary buffers have returned.
    693 	 */
    694 	if (vnx->vx_error != 0) {
    695 		pbp->b_flags |= B_ERROR;
    696 		pbp->b_error = vnx->vx_error;
    697 		if ((vnx->vx_flags & VX_BUSY) == 0 && vnx->vx_pending == 0) {
    698 
    699 #ifdef DEBUG
    700 			if (vnddebug & VDB_IO)
    701 				printf("vndiodone: pbp %p iodone: error %d\n",
    702 					pbp, vnx->vx_error);
    703 #endif
    704 			VND_PUTXFER(vnd, vnx);
    705 			biodone(pbp);
    706 		}
    707 	} else if (pbp->b_resid == 0) {
    708 
    709 #ifdef DIAGNOSTIC
    710 		if (vnx->vx_pending != 0)
    711 			panic("vndiodone: vnx pending: %d", vnx->vx_pending);
    712 #endif
    713 
    714 		if ((vnx->vx_flags & VX_BUSY) == 0) {
    715 #ifdef DEBUG
    716 			if (vnddebug & VDB_IO)
    717 				printf("vndiodone: pbp %p iodone\n", pbp);
    718 #endif
    719 			VND_PUTXFER(vnd, vnx);
    720 			biodone(pbp);
    721 		}
    722 	}
    723 
    724 	vnd->sc_active--;
    725 	vndstart(vnd);
    726 	splx(s);
    727 }
    728 
    729 /* ARGSUSED */
    730 int
    731 vndread(dev, uio, flags)
    732 	dev_t dev;
    733 	struct uio *uio;
    734 	int flags;
    735 {
    736 	int unit = vndunit(dev);
    737 	struct vnd_softc *sc;
    738 
    739 #ifdef DEBUG
    740 	if (vnddebug & VDB_FOLLOW)
    741 		printf("vndread(0x%x, %p)\n", dev, uio);
    742 #endif
    743 
    744 	if (unit >= numvnd)
    745 		return (ENXIO);
    746 	sc = &vnd_softc[unit];
    747 
    748 	if ((sc->sc_flags & VNF_INITED) == 0)
    749 		return (ENXIO);
    750 
    751 	return (physio(vndstrategy, NULL, dev, B_READ, minphys, uio));
    752 }
    753 
    754 /* ARGSUSED */
    755 int
    756 vndwrite(dev, uio, flags)
    757 	dev_t dev;
    758 	struct uio *uio;
    759 	int flags;
    760 {
    761 	int unit = vndunit(dev);
    762 	struct vnd_softc *sc;
    763 
    764 #ifdef DEBUG
    765 	if (vnddebug & VDB_FOLLOW)
    766 		printf("vndwrite(0x%x, %p)\n", dev, uio);
    767 #endif
    768 
    769 	if (unit >= numvnd)
    770 		return (ENXIO);
    771 	sc = &vnd_softc[unit];
    772 
    773 	if ((sc->sc_flags & VNF_INITED) == 0)
    774 		return (ENXIO);
    775 
    776 	return (physio(vndstrategy, NULL, dev, B_WRITE, minphys, uio));
    777 }
    778 
    779 /* ARGSUSED */
    780 int
    781 vndioctl(dev, cmd, data, flag, p)
    782 	dev_t dev;
    783 	u_long cmd;
    784 	caddr_t data;
    785 	int flag;
    786 	struct proc *p;
    787 {
    788 	int unit = vndunit(dev);
    789 	struct vnd_softc *vnd;
    790 	struct vnd_ioctl *vio;
    791 	struct vattr vattr;
    792 	struct nameidata nd;
    793 	int error, part, pmask;
    794 	size_t geomsize;
    795 	int fflags;
    796 #ifdef __HAVE_OLD_DISKLABEL
    797 	struct disklabel newlabel;
    798 #endif
    799 
    800 #ifdef DEBUG
    801 	if (vnddebug & VDB_FOLLOW)
    802 		printf("vndioctl(0x%x, 0x%lx, %p, 0x%x, %p): unit %d\n",
    803 		    dev, cmd, data, flag, p, unit);
    804 #endif
    805 	if (unit >= numvnd)
    806 		return (ENXIO);
    807 
    808 	vnd = &vnd_softc[unit];
    809 	vio = (struct vnd_ioctl *)data;
    810 
    811 	/* Must be open for writes for these commands... */
    812 	switch (cmd) {
    813 	case VNDIOCSET:
    814 	case VNDIOCCLR:
    815 	case DIOCSDINFO:
    816 	case DIOCWDINFO:
    817 #ifdef __HAVE_OLD_DISKLABEL
    818 	case ODIOCSDINFO:
    819 	case ODIOCWDINFO:
    820 #endif
    821 	case DIOCKLABEL:
    822 	case DIOCWLABEL:
    823 		if ((flag & FWRITE) == 0)
    824 			return (EBADF);
    825 	}
    826 
    827 	/* Must be initialized for these... */
    828 	switch (cmd) {
    829 	case VNDIOCCLR:
    830 	case DIOCGDINFO:
    831 	case DIOCSDINFO:
    832 	case DIOCWDINFO:
    833 	case DIOCGPART:
    834 	case DIOCKLABEL:
    835 	case DIOCWLABEL:
    836 	case DIOCGDEFLABEL:
    837 #ifdef __HAVE_OLD_DISKLABEL
    838 	case ODIOCGDINFO:
    839 	case ODIOCSDINFO:
    840 	case ODIOCWDINFO:
    841 	case ODIOCGDEFLABEL:
    842 #endif
    843 		if ((vnd->sc_flags & VNF_INITED) == 0)
    844 			return (ENXIO);
    845 	}
    846 
    847 	switch (cmd) {
    848 	case VNDIOCSET:
    849 		if (vnd->sc_flags & VNF_INITED)
    850 			return (EBUSY);
    851 
    852 		if ((error = vndlock(vnd)) != 0)
    853 			return (error);
    854 
    855 		fflags = FREAD;
    856 		if ((vio->vnd_flags & VNDIOF_READONLY) == 0)
    857 			fflags |= FWRITE;
    858 		NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, vio->vnd_file, p);
    859 		if ((error = vn_open(&nd, fflags, 0)) != 0)
    860 			goto unlock_and_exit;
    861 		error = VOP_GETATTR(nd.ni_vp, &vattr, p->p_ucred, p);
    862 		VOP_UNLOCK(nd.ni_vp, 0);
    863 		if (!error && nd.ni_vp->v_type != VREG)
    864 			error = EOPNOTSUPP;
    865 		if (error)
    866 			goto close_and_exit;
    867 		vnd->sc_vp = nd.ni_vp;
    868 		vnd->sc_size = btodb(vattr.va_size);	/* note truncation */
    869 
    870 		/*
    871 		 * Use pseudo-geometry specified.  If none was provided,
    872 		 * use "standard" Adaptec fictitious geometry.
    873 		 */
    874 		if (vio->vnd_flags & VNDIOF_HASGEOM) {
    875 
    876 			memcpy(&vnd->sc_geom, &vio->vnd_geom,
    877 			    sizeof(vio->vnd_geom));
    878 
    879 			/*
    880 			 * Sanity-check the sector size.
    881 			 * XXX Don't allow secsize < DEV_BSIZE.  Should
    882 			 * XXX we?
    883 			 */
    884 			if (vnd->sc_geom.vng_secsize < DEV_BSIZE ||
    885 			    (vnd->sc_geom.vng_secsize % DEV_BSIZE) != 0 ||
    886 			    vnd->sc_geom.vng_ncylinders == 0 ||
    887 			    (vnd->sc_geom.vng_ntracks *
    888 			     vnd->sc_geom.vng_nsectors) == 0) {
    889 				error = EINVAL;
    890 				goto close_and_exit;
    891 			}
    892 
    893 			/*
    894 			 * Compute the size (in DEV_BSIZE blocks) specified
    895 			 * by the geometry.
    896 			 */
    897 			geomsize = (vnd->sc_geom.vng_nsectors *
    898 			    vnd->sc_geom.vng_ntracks *
    899 			    vnd->sc_geom.vng_ncylinders) *
    900 			    (vnd->sc_geom.vng_secsize / DEV_BSIZE);
    901 
    902 			/*
    903 			 * Sanity-check the size against the specified
    904 			 * geometry.
    905 			 */
    906 			if (vnd->sc_size < geomsize) {
    907 				error = EINVAL;
    908 				goto close_and_exit;
    909 			}
    910 		} else {
    911 			/*
    912 			 * Size must be at least 2048 DEV_BSIZE blocks
    913 			 * (1M) in order to use this geometry.
    914 			 */
    915 			if (vnd->sc_size < (32 * 64)) {
    916 				error = EINVAL;
    917 				goto close_and_exit;
    918 			}
    919 
    920 			vnd->sc_geom.vng_secsize = DEV_BSIZE;
    921 			vnd->sc_geom.vng_nsectors = 32;
    922 			vnd->sc_geom.vng_ntracks = 64;
    923 			vnd->sc_geom.vng_ncylinders = vnd->sc_size / (64 * 32);
    924 		}
    925 
    926 		if (vio->vnd_flags & VNDIOF_READONLY) {
    927 			vnd->sc_flags |= VNF_READONLY;
    928 		}
    929 
    930 		if ((error = vndsetcred(vnd, p->p_ucred)) != 0)
    931 			goto close_and_exit;
    932 		vndthrottle(vnd, vnd->sc_vp);
    933 		vio->vnd_size = dbtob(vnd->sc_size);
    934 		vnd->sc_flags |= VNF_INITED;
    935 #ifdef DEBUG
    936 		if (vnddebug & VDB_INIT)
    937 			printf("vndioctl: SET vp %p size 0x%lx %d/%d/%d/%d\n",
    938 			    vnd->sc_vp, (unsigned long) vnd->sc_size,
    939 			    vnd->sc_geom.vng_secsize,
    940 			    vnd->sc_geom.vng_nsectors,
    941 			    vnd->sc_geom.vng_ntracks,
    942 			    vnd->sc_geom.vng_ncylinders);
    943 #endif
    944 
    945 		/* Attach the disk. */
    946 		memset(vnd->sc_xname, 0, sizeof(vnd->sc_xname)); /* XXX */
    947 		sprintf(vnd->sc_xname, "vnd%d", unit);		/* XXX */
    948 		vnd->sc_dkdev.dk_name = vnd->sc_xname;
    949 		disk_attach(&vnd->sc_dkdev);
    950 
    951 		/* Initialize the xfer and buffer pools. */
    952 		pool_init(&vnd->sc_vxpool, sizeof(struct vndxfer), 0,
    953 		    0, 0, "vndxpl", NULL);
    954 		pool_init(&vnd->sc_vbpool, sizeof(struct vndbuf), 0,
    955 		    0, 0, "vndbpl", NULL);
    956 
    957 		/* Try and read the disklabel. */
    958 		vndgetdisklabel(dev);
    959 
    960 		vndunlock(vnd);
    961 
    962 		break;
    963 
    964 close_and_exit:
    965 		(void) vn_close(nd.ni_vp, fflags, p->p_ucred, p);
    966 unlock_and_exit:
    967 		vndunlock(vnd);
    968 		return (error);
    969 
    970 	case VNDIOCCLR:
    971 		if ((error = vndlock(vnd)) != 0)
    972 			return (error);
    973 
    974 		/*
    975 		 * Don't unconfigure if any other partitions are open
    976 		 * or if both the character and block flavors of this
    977 		 * partition are open.
    978 		 */
    979 		part = DISKPART(dev);
    980 		pmask = (1 << part);
    981 		if (((vnd->sc_dkdev.dk_openmask & ~pmask) ||
    982 		    ((vnd->sc_dkdev.dk_bopenmask & pmask) &&
    983 		    (vnd->sc_dkdev.dk_copenmask & pmask))) &&
    984 			!(vio->vnd_flags & VNDIOF_FORCE)) {
    985 			vndunlock(vnd);
    986 			return (EBUSY);
    987 		}
    988 
    989 		/*
    990 		 * XXX vndclear() might call vndclose() implicitely;
    991 		 * release lock to avoid recursion
    992 		 */
    993 		vndunlock(vnd);
    994 		vndclear(vnd, minor(dev));
    995 #ifdef DEBUG
    996 		if (vnddebug & VDB_INIT)
    997 			printf("vndioctl: CLRed\n");
    998 #endif
    999 
   1000 		/* Destroy the xfer and buffer pools. */
   1001 		pool_destroy(&vnd->sc_vxpool);
   1002 		pool_destroy(&vnd->sc_vbpool);
   1003 
   1004 		/* Detatch the disk. */
   1005 		disk_detach(&vnd->sc_dkdev);
   1006 
   1007 		break;
   1008 
   1009 	case VNDIOCGET: {
   1010 		struct vnd_user *vnu;
   1011 		struct vattr va;
   1012 
   1013 		vnu = (struct vnd_user *)data;
   1014 
   1015 		if (vnu->vnu_unit == -1)
   1016 			vnu->vnu_unit = unit;
   1017 		if (vnu->vnu_unit >= numvnd)
   1018 			return (ENXIO);
   1019 		if (vnu->vnu_unit < 0)
   1020 			return (EINVAL);
   1021 
   1022 		vnd = &vnd_softc[vnu->vnu_unit];
   1023 
   1024 		if (vnd->sc_flags & VNF_INITED) {
   1025 			error = VOP_GETATTR(vnd->sc_vp, &va, p->p_ucred, p);
   1026 			if (error)
   1027 				return (error);
   1028 			vnu->vnu_dev = va.va_fsid;
   1029 			vnu->vnu_ino = va.va_fileid;
   1030 		}
   1031 		else {
   1032 			/* unused is not an error */
   1033 			vnu->vnu_dev = 0;
   1034 			vnu->vnu_ino = 0;
   1035 		}
   1036 
   1037 		break;
   1038 	}
   1039 
   1040 	case DIOCGDINFO:
   1041 		*(struct disklabel *)data = *(vnd->sc_dkdev.dk_label);
   1042 		break;
   1043 
   1044 #ifdef __HAVE_OLD_DISKLABEL
   1045 	case ODIOCGDINFO:
   1046 		newlabel = *(vnd->sc_dkdev.dk_label);
   1047 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
   1048 			return ENOTTY;
   1049 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
   1050 		break;
   1051 #endif
   1052 
   1053 	case DIOCGPART:
   1054 		((struct partinfo *)data)->disklab = vnd->sc_dkdev.dk_label;
   1055 		((struct partinfo *)data)->part =
   1056 		    &vnd->sc_dkdev.dk_label->d_partitions[DISKPART(dev)];
   1057 		break;
   1058 
   1059 	case DIOCWDINFO:
   1060 	case DIOCSDINFO:
   1061 #ifdef __HAVE_OLD_DISKLABEL
   1062 	case ODIOCWDINFO:
   1063 	case ODIOCSDINFO:
   1064 #endif
   1065 	{
   1066 		struct disklabel *lp;
   1067 
   1068 		if ((error = vndlock(vnd)) != 0)
   1069 			return (error);
   1070 
   1071 		vnd->sc_flags |= VNF_LABELLING;
   1072 
   1073 #ifdef __HAVE_OLD_DISKLABEL
   1074 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
   1075 			memset(&newlabel, 0, sizeof newlabel);
   1076 			memcpy(&newlabel, data, sizeof (struct olddisklabel));
   1077 			lp = &newlabel;
   1078 		} else
   1079 #endif
   1080 		lp = (struct disklabel *)data;
   1081 
   1082 		error = setdisklabel(vnd->sc_dkdev.dk_label,
   1083 		    lp, 0, vnd->sc_dkdev.dk_cpulabel);
   1084 		if (error == 0) {
   1085 			if (cmd == DIOCWDINFO
   1086 #ifdef __HAVE_OLD_DISKLABEL
   1087 			    || cmd == ODIOCWDINFO
   1088 #endif
   1089 			   )
   1090 				error = writedisklabel(VNDLABELDEV(dev),
   1091 				    vndstrategy, vnd->sc_dkdev.dk_label,
   1092 				    vnd->sc_dkdev.dk_cpulabel);
   1093 		}
   1094 
   1095 		vnd->sc_flags &= ~VNF_LABELLING;
   1096 
   1097 		vndunlock(vnd);
   1098 
   1099 		if (error)
   1100 			return (error);
   1101 		break;
   1102 	}
   1103 
   1104 	case DIOCKLABEL:
   1105 		if (*(int *)data != 0)
   1106 			vnd->sc_flags |= VNF_KLABEL;
   1107 		else
   1108 			vnd->sc_flags &= ~VNF_KLABEL;
   1109 		break;
   1110 
   1111 	case DIOCWLABEL:
   1112 		if (*(int *)data != 0)
   1113 			vnd->sc_flags |= VNF_WLABEL;
   1114 		else
   1115 			vnd->sc_flags &= ~VNF_WLABEL;
   1116 		break;
   1117 
   1118 	case DIOCGDEFLABEL:
   1119 		vndgetdefaultlabel(vnd, (struct disklabel *)data);
   1120 		break;
   1121 
   1122 #ifdef __HAVE_OLD_DISKLABEL
   1123 	case ODIOCGDEFLABEL:
   1124 		vndgetdefaultlabel(vnd, &newlabel);
   1125 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
   1126 			return ENOTTY;
   1127 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
   1128 		break;
   1129 #endif
   1130 
   1131 	default:
   1132 		return (ENOTTY);
   1133 	}
   1134 
   1135 	return (0);
   1136 }
   1137 
   1138 /*
   1139  * Duplicate the current processes' credentials.  Since we are called only
   1140  * as the result of a SET ioctl and only root can do that, any future access
   1141  * to this "disk" is essentially as root.  Note that credentials may change
   1142  * if some other uid can write directly to the mapped file (NFS).
   1143  */
   1144 int
   1145 vndsetcred(vnd, cred)
   1146 	struct vnd_softc *vnd;
   1147 	struct ucred *cred;
   1148 {
   1149 	struct uio auio;
   1150 	struct iovec aiov;
   1151 	char *tmpbuf;
   1152 	int error;
   1153 
   1154 	vnd->sc_cred = crdup(cred);
   1155 	tmpbuf = malloc(DEV_BSIZE, M_TEMP, M_WAITOK);
   1156 
   1157 	/* XXX: Horrible kludge to establish credentials for NFS */
   1158 	aiov.iov_base = tmpbuf;
   1159 	aiov.iov_len = min(DEV_BSIZE, dbtob(vnd->sc_size));
   1160 	auio.uio_iov = &aiov;
   1161 	auio.uio_iovcnt = 1;
   1162 	auio.uio_offset = 0;
   1163 	auio.uio_rw = UIO_READ;
   1164 	auio.uio_segflg = UIO_SYSSPACE;
   1165 	auio.uio_resid = aiov.iov_len;
   1166 	vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
   1167 	error = VOP_READ(vnd->sc_vp, &auio, 0, vnd->sc_cred);
   1168 	if (error == 0) {
   1169 		/*
   1170 		 * Because vnd does all IO directly through the vnode
   1171 		 * we need to flush (at least) the buffer from the above
   1172 		 * VOP_READ from the buffer cache to prevent cache
   1173 		 * incoherencies.  Also, be careful to write dirty
   1174 		 * buffers back to stable storage.
   1175 		 */
   1176 		error = vinvalbuf(vnd->sc_vp, V_SAVE, vnd->sc_cred,
   1177 			    curproc, 0, 0);
   1178 	}
   1179 	VOP_UNLOCK(vnd->sc_vp, 0);
   1180 
   1181 	free(tmpbuf, M_TEMP);
   1182 	return (error);
   1183 }
   1184 
   1185 /*
   1186  * Set maxactive based on FS type
   1187  */
   1188 void
   1189 vndthrottle(vnd, vp)
   1190 	struct vnd_softc *vnd;
   1191 	struct vnode *vp;
   1192 {
   1193 #ifdef NFS
   1194 	extern int (**nfsv2_vnodeop_p) __P((void *));
   1195 
   1196 	if (vp->v_op == nfsv2_vnodeop_p)
   1197 		vnd->sc_maxactive = 2;
   1198 	else
   1199 #endif
   1200 		vnd->sc_maxactive = 8;
   1201 
   1202 	if (vnd->sc_maxactive < 1)
   1203 		vnd->sc_maxactive = 1;
   1204 }
   1205 
   1206 #if 0
   1207 void
   1208 vndshutdown()
   1209 {
   1210 	struct vnd_softc *vnd;
   1211 
   1212 	for (vnd = &vnd_softc[0]; vnd < &vnd_softc[numvnd]; vnd++)
   1213 		if (vnd->sc_flags & VNF_INITED)
   1214 			vndclear(vnd);
   1215 }
   1216 #endif
   1217 
   1218 void
   1219 vndclear(vnd, myminor)
   1220 	struct vnd_softc *vnd;
   1221 	int myminor;
   1222 {
   1223 	struct vnode *vp = vnd->sc_vp;
   1224 	struct proc *p = curproc;		/* XXX */
   1225 	int fflags = FREAD;
   1226 	int bmaj, cmaj, i, mn;
   1227 
   1228 #ifdef DEBUG
   1229 	if (vnddebug & VDB_FOLLOW)
   1230 		printf("vndclear(%p): vp %p\n", vnd, vp);
   1231 #endif
   1232 	/* locate the major number */
   1233 	bmaj = bdevsw_lookup_major(&vnd_bdevsw);
   1234 	cmaj = cdevsw_lookup_major(&vnd_cdevsw);
   1235 
   1236 	/* Nuke the vnodes for any open instances */
   1237 	for (i = 0; i < MAXPARTITIONS; i++) {
   1238 		mn = DISKMINOR(vnd->sc_unit, i);
   1239 		vdevgone(bmaj, mn, mn, VBLK);
   1240 		if (mn != myminor) /* XXX avoid to kill own vnode */
   1241 			vdevgone(cmaj, mn, mn, VCHR);
   1242 	}
   1243 
   1244 	if ((vnd->sc_flags & VNF_READONLY) == 0)
   1245 		fflags |= FWRITE;
   1246 	vnd->sc_flags &= ~(VNF_INITED | VNF_READONLY | VNF_VLABEL);
   1247 	if (vp == (struct vnode *)0)
   1248 		panic("vndioctl: null vp");
   1249 	(void) vn_close(vp, fflags, vnd->sc_cred, p);
   1250 	crfree(vnd->sc_cred);
   1251 	vnd->sc_vp = (struct vnode *)0;
   1252 	vnd->sc_cred = (struct ucred *)0;
   1253 	vnd->sc_size = 0;
   1254 }
   1255 
   1256 int
   1257 vndsize(dev)
   1258 	dev_t dev;
   1259 {
   1260 	struct vnd_softc *sc;
   1261 	struct disklabel *lp;
   1262 	int part, unit, omask;
   1263 	int size;
   1264 
   1265 	unit = vndunit(dev);
   1266 	if (unit >= numvnd)
   1267 		return (-1);
   1268 	sc = &vnd_softc[unit];
   1269 
   1270 	if ((sc->sc_flags & VNF_INITED) == 0)
   1271 		return (-1);
   1272 
   1273 	part = DISKPART(dev);
   1274 	omask = sc->sc_dkdev.dk_openmask & (1 << part);
   1275 	lp = sc->sc_dkdev.dk_label;
   1276 
   1277 	if (omask == 0 && vndopen(dev, 0, S_IFBLK, curproc))
   1278 		return (-1);
   1279 
   1280 	if (lp->d_partitions[part].p_fstype != FS_SWAP)
   1281 		size = -1;
   1282 	else
   1283 		size = lp->d_partitions[part].p_size *
   1284 		    (lp->d_secsize / DEV_BSIZE);
   1285 
   1286 	if (omask == 0 && vndclose(dev, 0, S_IFBLK, curproc))
   1287 		return (-1);
   1288 
   1289 	return (size);
   1290 }
   1291 
   1292 int
   1293 vnddump(dev, blkno, va, size)
   1294 	dev_t dev;
   1295 	daddr_t blkno;
   1296 	caddr_t va;
   1297 	size_t size;
   1298 {
   1299 
   1300 	/* Not implemented. */
   1301 	return ENXIO;
   1302 }
   1303 
   1304 void
   1305 vndgetdefaultlabel(sc, lp)
   1306 	struct vnd_softc *sc;
   1307 	struct disklabel *lp;
   1308 {
   1309 	struct vndgeom *vng = &sc->sc_geom;
   1310 	struct partition *pp;
   1311 
   1312 	memset(lp, 0, sizeof(*lp));
   1313 
   1314 	lp->d_secperunit = sc->sc_size / (vng->vng_secsize / DEV_BSIZE);
   1315 	lp->d_secsize = vng->vng_secsize;
   1316 	lp->d_nsectors = vng->vng_nsectors;
   1317 	lp->d_ntracks = vng->vng_ntracks;
   1318 	lp->d_ncylinders = vng->vng_ncylinders;
   1319 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
   1320 
   1321 	strncpy(lp->d_typename, "vnd", sizeof(lp->d_typename));
   1322 	lp->d_type = DTYPE_VND;
   1323 	strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
   1324 	lp->d_rpm = 3600;
   1325 	lp->d_interleave = 1;
   1326 	lp->d_flags = 0;
   1327 
   1328 	pp = &lp->d_partitions[RAW_PART];
   1329 	pp->p_offset = 0;
   1330 	pp->p_size = lp->d_secperunit;
   1331 	pp->p_fstype = FS_UNUSED;
   1332 	lp->d_npartitions = RAW_PART + 1;
   1333 
   1334 	lp->d_magic = DISKMAGIC;
   1335 	lp->d_magic2 = DISKMAGIC;
   1336 	lp->d_checksum = dkcksum(lp);
   1337 }
   1338 
   1339 /*
   1340  * Read the disklabel from a vnd.  If one is not present, create a fake one.
   1341  */
   1342 void
   1343 vndgetdisklabel(dev)
   1344 	dev_t dev;
   1345 {
   1346 	struct vnd_softc *sc = &vnd_softc[vndunit(dev)];
   1347 	const char *errstring;
   1348 	struct disklabel *lp = sc->sc_dkdev.dk_label;
   1349 	struct cpu_disklabel *clp = sc->sc_dkdev.dk_cpulabel;
   1350 	int i;
   1351 
   1352 	memset(clp, 0, sizeof(*clp));
   1353 
   1354 	vndgetdefaultlabel(sc, lp);
   1355 
   1356 	/*
   1357 	 * Call the generic disklabel extraction routine.
   1358 	 */
   1359 	errstring = readdisklabel(VNDLABELDEV(dev), vndstrategy, lp, clp);
   1360 	if (errstring) {
   1361 		/*
   1362 		 * Lack of disklabel is common, but we print the warning
   1363 		 * anyway, since it might contain other useful information.
   1364 		 */
   1365 		printf("%s: %s\n", sc->sc_xname, errstring);
   1366 
   1367 		/*
   1368 		 * For historical reasons, if there's no disklabel
   1369 		 * present, all partitions must be FS_BSDFFS and
   1370 		 * occupy the entire disk.
   1371 		 */
   1372 		for (i = 0; i < MAXPARTITIONS; i++) {
   1373 			/*
   1374 			 * Don't wipe out port specific hack (such as
   1375 			 * dos partition hack of i386 port).
   1376 			 */
   1377 			if (lp->d_partitions[i].p_size != 0)
   1378 				continue;
   1379 
   1380 			lp->d_partitions[i].p_size = lp->d_secperunit;
   1381 			lp->d_partitions[i].p_offset = 0;
   1382 			lp->d_partitions[i].p_fstype = FS_BSDFFS;
   1383 		}
   1384 
   1385 		strncpy(lp->d_packname, "default label",
   1386 		    sizeof(lp->d_packname));
   1387 
   1388 		lp->d_npartitions = MAXPARTITIONS;
   1389 		lp->d_checksum = dkcksum(lp);
   1390 	}
   1391 
   1392 	/* In-core label now valid. */
   1393 	sc->sc_flags |= VNF_VLABEL;
   1394 }
   1395 
   1396 /*
   1397  * Wait interruptibly for an exclusive lock.
   1398  *
   1399  * XXX
   1400  * Several drivers do this; it should be abstracted and made MP-safe.
   1401  */
   1402 static int
   1403 vndlock(sc)
   1404 	struct vnd_softc *sc;
   1405 {
   1406 	int error;
   1407 
   1408 	while ((sc->sc_flags & VNF_LOCKED) != 0) {
   1409 		sc->sc_flags |= VNF_WANTED;
   1410 		if ((error = tsleep(sc, PRIBIO | PCATCH, "vndlck", 0)) != 0)
   1411 			return (error);
   1412 	}
   1413 	sc->sc_flags |= VNF_LOCKED;
   1414 	return (0);
   1415 }
   1416 
   1417 /*
   1418  * Unlock and wake up any waiters.
   1419  */
   1420 static void
   1421 vndunlock(sc)
   1422 	struct vnd_softc *sc;
   1423 {
   1424 
   1425 	sc->sc_flags &= ~VNF_LOCKED;
   1426 	if ((sc->sc_flags & VNF_WANTED) != 0) {
   1427 		sc->sc_flags &= ~VNF_WANTED;
   1428 		wakeup(sc);
   1429 	}
   1430 }
   1431