Home | History | Annotate | Line # | Download | only in dev
vnd.c revision 1.103
      1 /*	$NetBSD: vnd.c,v 1.103 2003/10/15 11:28:59 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 1.103 2003/10/15 11:28:59 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 		/*
    567 		 * Just sort by block number
    568 		 */
    569 		s = splbio();
    570 		if (vnx->vx_error != 0) {
    571 			VND_PUTBUF(vnd, nbp);
    572 			goto out;
    573 		}
    574 		vnx->vx_pending++;
    575 		bgetvp(vp, &nbp->vb_buf);
    576 		BUFQ_PUT(&vnd->sc_tab, &nbp->vb_buf);
    577 		vndstart(vnd);
    578 		splx(s);
    579 		bn += sz;
    580 		addr += sz;
    581 	}
    582 
    583 	s = splbio();
    584 
    585 out: /* Arrive here at splbio */
    586 	if ((flags & B_READ) == 0)
    587 		vn_finished_write(mp, 0);
    588 	vnx->vx_flags &= ~VX_BUSY;
    589 	if (vnx->vx_pending == 0) {
    590 		if (vnx->vx_error != 0) {
    591 			bp->b_error = vnx->vx_error;
    592 			bp->b_flags |= B_ERROR;
    593 		}
    594 		VND_PUTXFER(vnd, vnx);
    595 		biodone(bp);
    596 	}
    597 	splx(s);
    598 	return;
    599 
    600  done:
    601 	biodone(bp);
    602 }
    603 
    604 /*
    605  * Feed requests sequentially.
    606  * We do it this way to keep from flooding NFS servers if we are connected
    607  * to an NFS file.  This places the burden on the client rather than the
    608  * server.
    609  */
    610 void
    611 vndstart(vnd)
    612 	struct vnd_softc *vnd;
    613 {
    614 	struct buf	*bp;
    615 
    616 	/*
    617 	 * Dequeue now since lower level strategy routine might
    618 	 * queue using same links
    619 	 */
    620 
    621 	if ((vnd->sc_flags & VNF_BUSY) != 0)
    622 		return;
    623 
    624 	vnd->sc_flags |= VNF_BUSY;
    625 
    626 	while (vnd->sc_active < vnd->sc_maxactive) {
    627 		bp = BUFQ_GET(&vnd->sc_tab);
    628 		if (bp == NULL)
    629 			break;
    630 		vnd->sc_active++;
    631 #ifdef DEBUG
    632 		if (vnddebug & VDB_IO)
    633 			printf("vndstart(%ld): bp %p vp %p blkno 0x%" PRIx64
    634 				" flags %lx addr %p cnt 0x%lx\n",
    635 			    (long) (vnd-vnd_softc), bp, bp->b_vp, bp->b_blkno,
    636 			    bp->b_flags, bp->b_data, bp->b_bcount);
    637 #endif
    638 
    639 		/* Instrumentation. */
    640 		disk_busy(&vnd->sc_dkdev);
    641 
    642 		if ((bp->b_flags & B_READ) == 0)
    643 			bp->b_vp->v_numoutput++;
    644 		VOP_STRATEGY(bp);
    645 	}
    646 	vnd->sc_flags &= ~VNF_BUSY;
    647 }
    648 
    649 void
    650 vndiodone(bp)
    651 	struct buf *bp;
    652 {
    653 	struct vndbuf *vbp = (struct vndbuf *) bp;
    654 	struct vndxfer *vnx = (struct vndxfer *)vbp->vb_xfer;
    655 	struct buf *pbp = vnx->vx_bp;
    656 	struct vnd_softc *vnd = &vnd_softc[vndunit(pbp->b_dev)];
    657 	int s, resid;
    658 
    659 	s = splbio();
    660 #ifdef DEBUG
    661 	if (vnddebug & VDB_IO)
    662 		printf("vndiodone(%ld): vbp %p vp %p blkno 0x%" PRIx64
    663 		       " addr %p cnt 0x%lx\n",
    664 		    (long) (vnd-vnd_softc), vbp, vbp->vb_buf.b_vp,
    665 		    vbp->vb_buf.b_blkno, vbp->vb_buf.b_data,
    666 		    vbp->vb_buf.b_bcount);
    667 #endif
    668 
    669 	resid = vbp->vb_buf.b_bcount - vbp->vb_buf.b_resid;
    670 	pbp->b_resid -= resid;
    671 	disk_unbusy(&vnd->sc_dkdev, resid, (pbp->b_flags & B_READ));
    672 	vnx->vx_pending--;
    673 
    674 	if (vbp->vb_buf.b_error) {
    675 #ifdef DEBUG
    676 		if (vnddebug & VDB_IO)
    677 			printf("vndiodone: vbp %p error %d\n", vbp,
    678 			    vbp->vb_buf.b_error);
    679 #endif
    680 		vnx->vx_error = vbp->vb_buf.b_error;
    681 	}
    682 
    683 	if (vbp->vb_buf.b_vp != NULLVP)
    684 		brelvp(&vbp->vb_buf);
    685 
    686 	VND_PUTBUF(vnd, vbp);
    687 
    688 	/*
    689 	 * Wrap up this transaction if it has run to completion or, in
    690 	 * case of an error, when all auxiliary buffers have returned.
    691 	 */
    692 	if (vnx->vx_error != 0) {
    693 		pbp->b_flags |= B_ERROR;
    694 		pbp->b_error = vnx->vx_error;
    695 		if ((vnx->vx_flags & VX_BUSY) == 0 && vnx->vx_pending == 0) {
    696 
    697 #ifdef DEBUG
    698 			if (vnddebug & VDB_IO)
    699 				printf("vndiodone: pbp %p iodone: error %d\n",
    700 					pbp, vnx->vx_error);
    701 #endif
    702 			VND_PUTXFER(vnd, vnx);
    703 			biodone(pbp);
    704 		}
    705 	} else if (pbp->b_resid == 0) {
    706 
    707 #ifdef DIAGNOSTIC
    708 		if (vnx->vx_pending != 0)
    709 			panic("vndiodone: vnx pending: %d", vnx->vx_pending);
    710 #endif
    711 
    712 		if ((vnx->vx_flags & VX_BUSY) == 0) {
    713 #ifdef DEBUG
    714 			if (vnddebug & VDB_IO)
    715 				printf("vndiodone: pbp %p iodone\n", pbp);
    716 #endif
    717 			VND_PUTXFER(vnd, vnx);
    718 			biodone(pbp);
    719 		}
    720 	}
    721 
    722 	vnd->sc_active--;
    723 	vndstart(vnd);
    724 	splx(s);
    725 }
    726 
    727 /* ARGSUSED */
    728 int
    729 vndread(dev, uio, flags)
    730 	dev_t dev;
    731 	struct uio *uio;
    732 	int flags;
    733 {
    734 	int unit = vndunit(dev);
    735 	struct vnd_softc *sc;
    736 
    737 #ifdef DEBUG
    738 	if (vnddebug & VDB_FOLLOW)
    739 		printf("vndread(0x%x, %p)\n", dev, uio);
    740 #endif
    741 
    742 	if (unit >= numvnd)
    743 		return (ENXIO);
    744 	sc = &vnd_softc[unit];
    745 
    746 	if ((sc->sc_flags & VNF_INITED) == 0)
    747 		return (ENXIO);
    748 
    749 	return (physio(vndstrategy, NULL, dev, B_READ, minphys, uio));
    750 }
    751 
    752 /* ARGSUSED */
    753 int
    754 vndwrite(dev, uio, flags)
    755 	dev_t dev;
    756 	struct uio *uio;
    757 	int flags;
    758 {
    759 	int unit = vndunit(dev);
    760 	struct vnd_softc *sc;
    761 
    762 #ifdef DEBUG
    763 	if (vnddebug & VDB_FOLLOW)
    764 		printf("vndwrite(0x%x, %p)\n", dev, uio);
    765 #endif
    766 
    767 	if (unit >= numvnd)
    768 		return (ENXIO);
    769 	sc = &vnd_softc[unit];
    770 
    771 	if ((sc->sc_flags & VNF_INITED) == 0)
    772 		return (ENXIO);
    773 
    774 	return (physio(vndstrategy, NULL, dev, B_WRITE, minphys, uio));
    775 }
    776 
    777 /* ARGSUSED */
    778 int
    779 vndioctl(dev, cmd, data, flag, p)
    780 	dev_t dev;
    781 	u_long cmd;
    782 	caddr_t data;
    783 	int flag;
    784 	struct proc *p;
    785 {
    786 	int unit = vndunit(dev);
    787 	struct vnd_softc *vnd;
    788 	struct vnd_ioctl *vio;
    789 	struct vattr vattr;
    790 	struct nameidata nd;
    791 	int error, part, pmask;
    792 	size_t geomsize;
    793 	int fflags;
    794 #ifdef __HAVE_OLD_DISKLABEL
    795 	struct disklabel newlabel;
    796 #endif
    797 
    798 #ifdef DEBUG
    799 	if (vnddebug & VDB_FOLLOW)
    800 		printf("vndioctl(0x%x, 0x%lx, %p, 0x%x, %p): unit %d\n",
    801 		    dev, cmd, data, flag, p, unit);
    802 #endif
    803 	if (unit >= numvnd)
    804 		return (ENXIO);
    805 
    806 	vnd = &vnd_softc[unit];
    807 	vio = (struct vnd_ioctl *)data;
    808 
    809 	/* Must be open for writes for these commands... */
    810 	switch (cmd) {
    811 	case VNDIOCSET:
    812 	case VNDIOCCLR:
    813 	case DIOCSDINFO:
    814 	case DIOCWDINFO:
    815 #ifdef __HAVE_OLD_DISKLABEL
    816 	case ODIOCSDINFO:
    817 	case ODIOCWDINFO:
    818 #endif
    819 	case DIOCKLABEL:
    820 	case DIOCWLABEL:
    821 		if ((flag & FWRITE) == 0)
    822 			return (EBADF);
    823 	}
    824 
    825 	/* Must be initialized for these... */
    826 	switch (cmd) {
    827 	case VNDIOCCLR:
    828 	case DIOCGDINFO:
    829 	case DIOCSDINFO:
    830 	case DIOCWDINFO:
    831 	case DIOCGPART:
    832 	case DIOCKLABEL:
    833 	case DIOCWLABEL:
    834 	case DIOCGDEFLABEL:
    835 #ifdef __HAVE_OLD_DISKLABEL
    836 	case ODIOCGDINFO:
    837 	case ODIOCSDINFO:
    838 	case ODIOCWDINFO:
    839 	case ODIOCGDEFLABEL:
    840 #endif
    841 		if ((vnd->sc_flags & VNF_INITED) == 0)
    842 			return (ENXIO);
    843 	}
    844 
    845 	switch (cmd) {
    846 	case VNDIOCSET:
    847 		if (vnd->sc_flags & VNF_INITED)
    848 			return (EBUSY);
    849 
    850 		if ((error = vndlock(vnd)) != 0)
    851 			return (error);
    852 
    853 		fflags = FREAD;
    854 		if ((vio->vnd_flags & VNDIOF_READONLY) == 0)
    855 			fflags |= FWRITE;
    856 		NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, vio->vnd_file, p);
    857 		if ((error = vn_open(&nd, fflags, 0)) != 0)
    858 			goto unlock_and_exit;
    859 		error = VOP_GETATTR(nd.ni_vp, &vattr, p->p_ucred, p);
    860 		VOP_UNLOCK(nd.ni_vp, 0);
    861 		if (!error && nd.ni_vp->v_type != VREG)
    862 			error = EOPNOTSUPP;
    863 		if (error)
    864 			goto close_and_exit;
    865 		vnd->sc_vp = nd.ni_vp;
    866 		vnd->sc_size = btodb(vattr.va_size);	/* note truncation */
    867 
    868 		/*
    869 		 * Use pseudo-geometry specified.  If none was provided,
    870 		 * use "standard" Adaptec fictitious geometry.
    871 		 */
    872 		if (vio->vnd_flags & VNDIOF_HASGEOM) {
    873 
    874 			memcpy(&vnd->sc_geom, &vio->vnd_geom,
    875 			    sizeof(vio->vnd_geom));
    876 
    877 			/*
    878 			 * Sanity-check the sector size.
    879 			 * XXX Don't allow secsize < DEV_BSIZE.  Should
    880 			 * XXX we?
    881 			 */
    882 			if (vnd->sc_geom.vng_secsize < DEV_BSIZE ||
    883 			    (vnd->sc_geom.vng_secsize % DEV_BSIZE) != 0) {
    884 				error = EINVAL;
    885 				goto close_and_exit;
    886 			}
    887 
    888 			/*
    889 			 * Compute the size (in DEV_BSIZE blocks) specified
    890 			 * by the geometry.
    891 			 */
    892 			geomsize = (vnd->sc_geom.vng_nsectors *
    893 			    vnd->sc_geom.vng_ntracks *
    894 			    vnd->sc_geom.vng_ncylinders) *
    895 			    (vnd->sc_geom.vng_secsize / DEV_BSIZE);
    896 
    897 			/*
    898 			 * Sanity-check the size against the specified
    899 			 * geometry.
    900 			 */
    901 			if (vnd->sc_size < geomsize) {
    902 				error = EINVAL;
    903 				goto close_and_exit;
    904 			}
    905 		} else {
    906 			/*
    907 			 * Size must be at least 2048 DEV_BSIZE blocks
    908 			 * (1M) in order to use this geometry.
    909 			 */
    910 			if (vnd->sc_size < (32 * 64)) {
    911 				error = EINVAL;
    912 				goto close_and_exit;
    913 			}
    914 
    915 			vnd->sc_geom.vng_secsize = DEV_BSIZE;
    916 			vnd->sc_geom.vng_nsectors = 32;
    917 			vnd->sc_geom.vng_ntracks = 64;
    918 			vnd->sc_geom.vng_ncylinders = vnd->sc_size / (64 * 32);
    919 		}
    920 
    921 		if (vio->vnd_flags & VNDIOF_READONLY) {
    922 			vnd->sc_flags |= VNF_READONLY;
    923 		}
    924 
    925 		if ((error = vndsetcred(vnd, p->p_ucred)) != 0)
    926 			goto close_and_exit;
    927 		vndthrottle(vnd, vnd->sc_vp);
    928 		vio->vnd_size = dbtob(vnd->sc_size);
    929 		vnd->sc_flags |= VNF_INITED;
    930 #ifdef DEBUG
    931 		if (vnddebug & VDB_INIT)
    932 			printf("vndioctl: SET vp %p size 0x%lx %d/%d/%d/%d\n",
    933 			    vnd->sc_vp, (unsigned long) vnd->sc_size,
    934 			    vnd->sc_geom.vng_secsize,
    935 			    vnd->sc_geom.vng_nsectors,
    936 			    vnd->sc_geom.vng_ntracks,
    937 			    vnd->sc_geom.vng_ncylinders);
    938 #endif
    939 
    940 		/* Attach the disk. */
    941 		memset(vnd->sc_xname, 0, sizeof(vnd->sc_xname)); /* XXX */
    942 		sprintf(vnd->sc_xname, "vnd%d", unit);		/* XXX */
    943 		vnd->sc_dkdev.dk_name = vnd->sc_xname;
    944 		disk_attach(&vnd->sc_dkdev);
    945 
    946 		/* Initialize the xfer and buffer pools. */
    947 		pool_init(&vnd->sc_vxpool, sizeof(struct vndxfer), 0,
    948 		    0, 0, "vndxpl", NULL);
    949 		pool_init(&vnd->sc_vbpool, sizeof(struct vndbuf), 0,
    950 		    0, 0, "vndbpl", NULL);
    951 
    952 		/* Try and read the disklabel. */
    953 		vndgetdisklabel(dev);
    954 
    955 		vndunlock(vnd);
    956 
    957 		break;
    958 
    959 close_and_exit:
    960 		(void) vn_close(nd.ni_vp, fflags, p->p_ucred, p);
    961 unlock_and_exit:
    962 		vndunlock(vnd);
    963 		return (error);
    964 
    965 	case VNDIOCCLR:
    966 		if ((error = vndlock(vnd)) != 0)
    967 			return (error);
    968 
    969 		/*
    970 		 * Don't unconfigure if any other partitions are open
    971 		 * or if both the character and block flavors of this
    972 		 * partition are open.
    973 		 */
    974 		part = DISKPART(dev);
    975 		pmask = (1 << part);
    976 		if (((vnd->sc_dkdev.dk_openmask & ~pmask) ||
    977 		    ((vnd->sc_dkdev.dk_bopenmask & pmask) &&
    978 		    (vnd->sc_dkdev.dk_copenmask & pmask))) &&
    979 			!(vio->vnd_flags & VNDIOF_FORCE)) {
    980 			vndunlock(vnd);
    981 			return (EBUSY);
    982 		}
    983 
    984 		/*
    985 		 * XXX vndclear() might call vndclose() implicitely;
    986 		 * release lock to avoid recursion
    987 		 */
    988 		vndunlock(vnd);
    989 		vndclear(vnd, minor(dev));
    990 #ifdef DEBUG
    991 		if (vnddebug & VDB_INIT)
    992 			printf("vndioctl: CLRed\n");
    993 #endif
    994 
    995 		/* Destroy the xfer and buffer pools. */
    996 		pool_destroy(&vnd->sc_vxpool);
    997 		pool_destroy(&vnd->sc_vbpool);
    998 
    999 		/* Detatch the disk. */
   1000 		disk_detach(&vnd->sc_dkdev);
   1001 
   1002 		break;
   1003 
   1004 	case VNDIOCGET: {
   1005 		struct vnd_user *vnu;
   1006 		struct vattr va;
   1007 
   1008 		vnu = (struct vnd_user *)data;
   1009 
   1010 		if (vnu->vnu_unit == -1)
   1011 			vnu->vnu_unit = unit;
   1012 		if (vnu->vnu_unit >= numvnd)
   1013 			return (ENXIO);
   1014 		if (vnu->vnu_unit < 0)
   1015 			return (EINVAL);
   1016 
   1017 		vnd = &vnd_softc[vnu->vnu_unit];
   1018 
   1019 		if (vnd->sc_flags & VNF_INITED) {
   1020 			error = VOP_GETATTR(vnd->sc_vp, &va, p->p_ucred, p);
   1021 			if (error)
   1022 				return (error);
   1023 			vnu->vnu_dev = va.va_fsid;
   1024 			vnu->vnu_ino = va.va_fileid;
   1025 		}
   1026 		else {
   1027 			/* unused is not an error */
   1028 			vnu->vnu_dev = 0;
   1029 			vnu->vnu_ino = 0;
   1030 		}
   1031 
   1032 		break;
   1033 	}
   1034 
   1035 	case DIOCGDINFO:
   1036 		*(struct disklabel *)data = *(vnd->sc_dkdev.dk_label);
   1037 		break;
   1038 
   1039 #ifdef __HAVE_OLD_DISKLABEL
   1040 	case ODIOCGDINFO:
   1041 		newlabel = *(vnd->sc_dkdev.dk_label);
   1042 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
   1043 			return ENOTTY;
   1044 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
   1045 		break;
   1046 #endif
   1047 
   1048 	case DIOCGPART:
   1049 		((struct partinfo *)data)->disklab = vnd->sc_dkdev.dk_label;
   1050 		((struct partinfo *)data)->part =
   1051 		    &vnd->sc_dkdev.dk_label->d_partitions[DISKPART(dev)];
   1052 		break;
   1053 
   1054 	case DIOCWDINFO:
   1055 	case DIOCSDINFO:
   1056 #ifdef __HAVE_OLD_DISKLABEL
   1057 	case ODIOCWDINFO:
   1058 	case ODIOCSDINFO:
   1059 #endif
   1060 	{
   1061 		struct disklabel *lp;
   1062 
   1063 		if ((error = vndlock(vnd)) != 0)
   1064 			return (error);
   1065 
   1066 		vnd->sc_flags |= VNF_LABELLING;
   1067 
   1068 #ifdef __HAVE_OLD_DISKLABEL
   1069 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
   1070 			memset(&newlabel, 0, sizeof newlabel);
   1071 			memcpy(&newlabel, data, sizeof (struct olddisklabel));
   1072 			lp = &newlabel;
   1073 		} else
   1074 #endif
   1075 		lp = (struct disklabel *)data;
   1076 
   1077 		error = setdisklabel(vnd->sc_dkdev.dk_label,
   1078 		    lp, 0, vnd->sc_dkdev.dk_cpulabel);
   1079 		if (error == 0) {
   1080 			if (cmd == DIOCWDINFO
   1081 #ifdef __HAVE_OLD_DISKLABEL
   1082 			    || cmd == ODIOCWDINFO
   1083 #endif
   1084 			   )
   1085 				error = writedisklabel(VNDLABELDEV(dev),
   1086 				    vndstrategy, vnd->sc_dkdev.dk_label,
   1087 				    vnd->sc_dkdev.dk_cpulabel);
   1088 		}
   1089 
   1090 		vnd->sc_flags &= ~VNF_LABELLING;
   1091 
   1092 		vndunlock(vnd);
   1093 
   1094 		if (error)
   1095 			return (error);
   1096 		break;
   1097 	}
   1098 
   1099 	case DIOCKLABEL:
   1100 		if (*(int *)data != 0)
   1101 			vnd->sc_flags |= VNF_KLABEL;
   1102 		else
   1103 			vnd->sc_flags &= ~VNF_KLABEL;
   1104 		break;
   1105 
   1106 	case DIOCWLABEL:
   1107 		if (*(int *)data != 0)
   1108 			vnd->sc_flags |= VNF_WLABEL;
   1109 		else
   1110 			vnd->sc_flags &= ~VNF_WLABEL;
   1111 		break;
   1112 
   1113 	case DIOCGDEFLABEL:
   1114 		vndgetdefaultlabel(vnd, (struct disklabel *)data);
   1115 		break;
   1116 
   1117 #ifdef __HAVE_OLD_DISKLABEL
   1118 	case ODIOCGDEFLABEL:
   1119 		vndgetdefaultlabel(vnd, &newlabel);
   1120 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
   1121 			return ENOTTY;
   1122 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
   1123 		break;
   1124 #endif
   1125 
   1126 	default:
   1127 		return (ENOTTY);
   1128 	}
   1129 
   1130 	return (0);
   1131 }
   1132 
   1133 /*
   1134  * Duplicate the current processes' credentials.  Since we are called only
   1135  * as the result of a SET ioctl and only root can do that, any future access
   1136  * to this "disk" is essentially as root.  Note that credentials may change
   1137  * if some other uid can write directly to the mapped file (NFS).
   1138  */
   1139 int
   1140 vndsetcred(vnd, cred)
   1141 	struct vnd_softc *vnd;
   1142 	struct ucred *cred;
   1143 {
   1144 	struct uio auio;
   1145 	struct iovec aiov;
   1146 	char *tmpbuf;
   1147 	int error;
   1148 
   1149 	vnd->sc_cred = crdup(cred);
   1150 	tmpbuf = malloc(DEV_BSIZE, M_TEMP, M_WAITOK);
   1151 
   1152 	/* XXX: Horrible kludge to establish credentials for NFS */
   1153 	aiov.iov_base = tmpbuf;
   1154 	aiov.iov_len = min(DEV_BSIZE, dbtob(vnd->sc_size));
   1155 	auio.uio_iov = &aiov;
   1156 	auio.uio_iovcnt = 1;
   1157 	auio.uio_offset = 0;
   1158 	auio.uio_rw = UIO_READ;
   1159 	auio.uio_segflg = UIO_SYSSPACE;
   1160 	auio.uio_resid = aiov.iov_len;
   1161 	vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
   1162 	error = VOP_READ(vnd->sc_vp, &auio, 0, vnd->sc_cred);
   1163 	if (error == 0) {
   1164 		/*
   1165 		 * Because vnd does all IO directly through the vnode
   1166 		 * we need to flush (at least) the buffer from the above
   1167 		 * VOP_READ from the buffer cache to prevent cache
   1168 		 * incoherencies.  Also, be careful to write dirty
   1169 		 * buffers back to stable storage.
   1170 		 */
   1171 		error = vinvalbuf(vnd->sc_vp, V_SAVE, vnd->sc_cred,
   1172 			    curproc, 0, 0);
   1173 	}
   1174 	VOP_UNLOCK(vnd->sc_vp, 0);
   1175 
   1176 	free(tmpbuf, M_TEMP);
   1177 	return (error);
   1178 }
   1179 
   1180 /*
   1181  * Set maxactive based on FS type
   1182  */
   1183 void
   1184 vndthrottle(vnd, vp)
   1185 	struct vnd_softc *vnd;
   1186 	struct vnode *vp;
   1187 {
   1188 #ifdef NFS
   1189 	extern int (**nfsv2_vnodeop_p) __P((void *));
   1190 
   1191 	if (vp->v_op == nfsv2_vnodeop_p)
   1192 		vnd->sc_maxactive = 2;
   1193 	else
   1194 #endif
   1195 		vnd->sc_maxactive = 8;
   1196 
   1197 	if (vnd->sc_maxactive < 1)
   1198 		vnd->sc_maxactive = 1;
   1199 }
   1200 
   1201 #if 0
   1202 void
   1203 vndshutdown()
   1204 {
   1205 	struct vnd_softc *vnd;
   1206 
   1207 	for (vnd = &vnd_softc[0]; vnd < &vnd_softc[numvnd]; vnd++)
   1208 		if (vnd->sc_flags & VNF_INITED)
   1209 			vndclear(vnd);
   1210 }
   1211 #endif
   1212 
   1213 void
   1214 vndclear(vnd, myminor)
   1215 	struct vnd_softc *vnd;
   1216 	int myminor;
   1217 {
   1218 	struct vnode *vp = vnd->sc_vp;
   1219 	struct proc *p = curproc;		/* XXX */
   1220 	int fflags = FREAD;
   1221 	int bmaj, cmaj, i, mn;
   1222 
   1223 #ifdef DEBUG
   1224 	if (vnddebug & VDB_FOLLOW)
   1225 		printf("vndclear(%p): vp %p\n", vnd, vp);
   1226 #endif
   1227 	/* locate the major number */
   1228 	bmaj = bdevsw_lookup_major(&vnd_bdevsw);
   1229 	cmaj = cdevsw_lookup_major(&vnd_cdevsw);
   1230 
   1231 	/* Nuke the vnodes for any open instances */
   1232 	for (i = 0; i < MAXPARTITIONS; i++) {
   1233 		mn = DISKMINOR(vnd->sc_unit, i);
   1234 		vdevgone(bmaj, mn, mn, VBLK);
   1235 		if (mn != myminor) /* XXX avoid to kill own vnode */
   1236 			vdevgone(cmaj, mn, mn, VCHR);
   1237 	}
   1238 
   1239 	if ((vnd->sc_flags & VNF_READONLY) == 0)
   1240 		fflags |= FWRITE;
   1241 	vnd->sc_flags &= ~(VNF_INITED | VNF_READONLY | VNF_VLABEL);
   1242 	if (vp == (struct vnode *)0)
   1243 		panic("vndioctl: null vp");
   1244 	(void) vn_close(vp, fflags, vnd->sc_cred, p);
   1245 	crfree(vnd->sc_cred);
   1246 	vnd->sc_vp = (struct vnode *)0;
   1247 	vnd->sc_cred = (struct ucred *)0;
   1248 	vnd->sc_size = 0;
   1249 }
   1250 
   1251 int
   1252 vndsize(dev)
   1253 	dev_t dev;
   1254 {
   1255 	struct vnd_softc *sc;
   1256 	struct disklabel *lp;
   1257 	int part, unit, omask;
   1258 	int size;
   1259 
   1260 	unit = vndunit(dev);
   1261 	if (unit >= numvnd)
   1262 		return (-1);
   1263 	sc = &vnd_softc[unit];
   1264 
   1265 	if ((sc->sc_flags & VNF_INITED) == 0)
   1266 		return (-1);
   1267 
   1268 	part = DISKPART(dev);
   1269 	omask = sc->sc_dkdev.dk_openmask & (1 << part);
   1270 	lp = sc->sc_dkdev.dk_label;
   1271 
   1272 	if (omask == 0 && vndopen(dev, 0, S_IFBLK, curproc))
   1273 		return (-1);
   1274 
   1275 	if (lp->d_partitions[part].p_fstype != FS_SWAP)
   1276 		size = -1;
   1277 	else
   1278 		size = lp->d_partitions[part].p_size *
   1279 		    (lp->d_secsize / DEV_BSIZE);
   1280 
   1281 	if (omask == 0 && vndclose(dev, 0, S_IFBLK, curproc))
   1282 		return (-1);
   1283 
   1284 	return (size);
   1285 }
   1286 
   1287 int
   1288 vnddump(dev, blkno, va, size)
   1289 	dev_t dev;
   1290 	daddr_t blkno;
   1291 	caddr_t va;
   1292 	size_t size;
   1293 {
   1294 
   1295 	/* Not implemented. */
   1296 	return ENXIO;
   1297 }
   1298 
   1299 void
   1300 vndgetdefaultlabel(sc, lp)
   1301 	struct vnd_softc *sc;
   1302 	struct disklabel *lp;
   1303 {
   1304 	struct vndgeom *vng = &sc->sc_geom;
   1305 	struct partition *pp;
   1306 
   1307 	memset(lp, 0, sizeof(*lp));
   1308 
   1309 	lp->d_secperunit = sc->sc_size / (vng->vng_secsize / DEV_BSIZE);
   1310 	lp->d_secsize = vng->vng_secsize;
   1311 	lp->d_nsectors = vng->vng_nsectors;
   1312 	lp->d_ntracks = vng->vng_ntracks;
   1313 	lp->d_ncylinders = vng->vng_ncylinders;
   1314 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
   1315 
   1316 	strncpy(lp->d_typename, "vnd", sizeof(lp->d_typename));
   1317 	lp->d_type = DTYPE_VND;
   1318 	strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
   1319 	lp->d_rpm = 3600;
   1320 	lp->d_interleave = 1;
   1321 	lp->d_flags = 0;
   1322 
   1323 	pp = &lp->d_partitions[RAW_PART];
   1324 	pp->p_offset = 0;
   1325 	pp->p_size = lp->d_secperunit;
   1326 	pp->p_fstype = FS_UNUSED;
   1327 	lp->d_npartitions = RAW_PART + 1;
   1328 
   1329 	lp->d_magic = DISKMAGIC;
   1330 	lp->d_magic2 = DISKMAGIC;
   1331 	lp->d_checksum = dkcksum(lp);
   1332 }
   1333 
   1334 /*
   1335  * Read the disklabel from a vnd.  If one is not present, create a fake one.
   1336  */
   1337 void
   1338 vndgetdisklabel(dev)
   1339 	dev_t dev;
   1340 {
   1341 	struct vnd_softc *sc = &vnd_softc[vndunit(dev)];
   1342 	const char *errstring;
   1343 	struct disklabel *lp = sc->sc_dkdev.dk_label;
   1344 	struct cpu_disklabel *clp = sc->sc_dkdev.dk_cpulabel;
   1345 	int i;
   1346 
   1347 	memset(clp, 0, sizeof(*clp));
   1348 
   1349 	vndgetdefaultlabel(sc, lp);
   1350 
   1351 	/*
   1352 	 * Call the generic disklabel extraction routine.
   1353 	 */
   1354 	errstring = readdisklabel(VNDLABELDEV(dev), vndstrategy, lp, clp);
   1355 	if (errstring) {
   1356 		/*
   1357 		 * Lack of disklabel is common, but we print the warning
   1358 		 * anyway, since it might contain other useful information.
   1359 		 */
   1360 		printf("%s: %s\n", sc->sc_xname, errstring);
   1361 
   1362 		/*
   1363 		 * For historical reasons, if there's no disklabel
   1364 		 * present, all partitions must be FS_BSDFFS and
   1365 		 * occupy the entire disk.
   1366 		 */
   1367 		for (i = 0; i < MAXPARTITIONS; i++) {
   1368 			/*
   1369 			 * Don't wipe out port specific hack (such as
   1370 			 * dos partition hack of i386 port).
   1371 			 */
   1372 			if (lp->d_partitions[i].p_size != 0)
   1373 				continue;
   1374 
   1375 			lp->d_partitions[i].p_size = lp->d_secperunit;
   1376 			lp->d_partitions[i].p_offset = 0;
   1377 			lp->d_partitions[i].p_fstype = FS_BSDFFS;
   1378 		}
   1379 
   1380 		strncpy(lp->d_packname, "default label",
   1381 		    sizeof(lp->d_packname));
   1382 
   1383 		lp->d_npartitions = MAXPARTITIONS;
   1384 		lp->d_checksum = dkcksum(lp);
   1385 	}
   1386 
   1387 	/* In-core label now valid. */
   1388 	sc->sc_flags |= VNF_VLABEL;
   1389 }
   1390 
   1391 /*
   1392  * Wait interruptibly for an exclusive lock.
   1393  *
   1394  * XXX
   1395  * Several drivers do this; it should be abstracted and made MP-safe.
   1396  */
   1397 static int
   1398 vndlock(sc)
   1399 	struct vnd_softc *sc;
   1400 {
   1401 	int error;
   1402 
   1403 	while ((sc->sc_flags & VNF_LOCKED) != 0) {
   1404 		sc->sc_flags |= VNF_WANTED;
   1405 		if ((error = tsleep(sc, PRIBIO | PCATCH, "vndlck", 0)) != 0)
   1406 			return (error);
   1407 	}
   1408 	sc->sc_flags |= VNF_LOCKED;
   1409 	return (0);
   1410 }
   1411 
   1412 /*
   1413  * Unlock and wake up any waiters.
   1414  */
   1415 static void
   1416 vndunlock(sc)
   1417 	struct vnd_softc *sc;
   1418 {
   1419 
   1420 	sc->sc_flags &= ~VNF_LOCKED;
   1421 	if ((sc->sc_flags & VNF_WANTED) != 0) {
   1422 		sc->sc_flags &= ~VNF_WANTED;
   1423 		wakeup(sc);
   1424 	}
   1425 }
   1426