Home | History | Annotate | Line # | Download | only in dev
vnd.c revision 1.39
      1 /*	$NetBSD: vnd.c,v 1.39 1997/06/08 15:55:34 pk Exp $	*/
      2 
      3 /*
      4  * Copyright (c) 1988 University of Utah.
      5  * Copyright (c) 1990, 1993
      6  *	The Regents of the University of California.  All rights reserved.
      7  *
      8  * This code is derived from software contributed to Berkeley by
      9  * the Systems Programming Group of the University of Utah Computer
     10  * Science Department.
     11  *
     12  * Redistribution and use in source and binary forms, with or without
     13  * modification, are permitted provided that the following conditions
     14  * are met:
     15  * 1. Redistributions of source code must retain the above copyright
     16  *    notice, this list of conditions and the following disclaimer.
     17  * 2. Redistributions in binary form must reproduce the above copyright
     18  *    notice, this list of conditions and the following disclaimer in the
     19  *    documentation and/or other materials provided with the distribution.
     20  * 3. All advertising materials mentioning features or use of this software
     21  *    must display the following acknowledgement:
     22  *	This product includes software developed by the University of
     23  *	California, Berkeley and its contributors.
     24  * 4. Neither the name of the University nor the names of its contributors
     25  *    may be used to endorse or promote products derived from this software
     26  *    without specific prior written permission.
     27  *
     28  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     29  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     30  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     31  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     32  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     33  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     34  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     35  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     36  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     37  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     38  * SUCH DAMAGE.
     39  *
     40  * from: Utah $Hdr: vn.c 1.13 94/04/02$
     41  *
     42  *	@(#)vn.c	8.6 (Berkeley) 4/1/94
     43  */
     44 
     45 /*
     46  * Vnode disk driver.
     47  *
     48  * Block/character interface to a vnode.  Allows one to treat a file
     49  * as a disk (e.g. build a filesystem in it, mount it, etc.).
     50  *
     51  * NOTE 1: This uses the VOP_BMAP/VOP_STRATEGY interface to the vnode
     52  * instead of a simple VOP_RDWR.  We do this to avoid distorting the
     53  * local buffer cache.
     54  *
     55  * NOTE 2: There is a security issue involved with this driver.
     56  * Once mounted all access to the contents of the "mapped" file via
     57  * the special file is controlled by the permissions on the special
     58  * file, the protection of the mapped file is ignored (effectively,
     59  * by using root credentials in all transactions).
     60  *
     61  * NOTE 3: Doesn't interact with leases, should it?
     62  */
     63 
     64 #include <sys/param.h>
     65 #include <sys/systm.h>
     66 #include <sys/namei.h>
     67 #include <sys/proc.h>
     68 #include <sys/errno.h>
     69 #include <sys/buf.h>
     70 #include <sys/malloc.h>
     71 #include <sys/ioctl.h>
     72 #include <sys/disklabel.h>
     73 #include <sys/device.h>
     74 #include <sys/disk.h>
     75 #include <sys/stat.h>
     76 #include <sys/mount.h>
     77 #include <sys/vnode.h>
     78 #include <sys/file.h>
     79 #include <sys/uio.h>
     80 #include <sys/conf.h>
     81 
     82 #include <miscfs/specfs/specdev.h>
     83 
     84 #include <dev/vndioctl.h>
     85 
     86 #ifdef DEBUG
     87 int dovndcluster = 1;
     88 int vnddebug = 0x00;
     89 #define VDB_FOLLOW	0x01
     90 #define VDB_INIT	0x02
     91 #define VDB_IO		0x04
     92 #endif
     93 
     94 #define b_cylin	b_resid
     95 
     96 #define	vndunit(x)	DISKUNIT(x)
     97 
     98 struct vndxfer {
     99 	struct buf	*vx_bp;		/* Pointer to parent buffer */
    100 	int		vx_error;
    101 	int		vx_pending;	/* # of pending aux buffers */
    102 };
    103 
    104 struct vndbuf {
    105 	struct buf	vb_buf;
    106 	struct vndxfer	*vb_xfer;
    107 };
    108 
    109 #define	getvndxfer()	\
    110 	((struct vndxfer *)malloc(sizeof(struct vndxfer), M_DEVBUF, M_WAITOK))
    111 #define putvndxfer(vnx)	\
    112 	free((caddr_t)(vnx), M_DEVBUF)
    113 #define	getvndbuf()	\
    114 	((struct vndbuf *)malloc(sizeof(struct vndbuf), M_DEVBUF, M_WAITOK))
    115 #define putvndbuf(vbp)	\
    116 	free((caddr_t)(vbp), M_DEVBUF)
    117 
    118 struct vnd_softc {
    119 	int		 sc_flags;	/* flags */
    120 	size_t		 sc_size;	/* size of vnd */
    121 	struct vnode	*sc_vp;		/* vnode */
    122 	struct ucred	*sc_cred;	/* credentials */
    123 	int		 sc_maxactive;	/* max # of active requests */
    124 	struct buf	 sc_tab;	/* transfer queue */
    125 	char		 sc_xname[8];	/* XXX external name */
    126 	struct disk	 sc_dkdev;	/* generic disk device info */
    127 };
    128 
    129 /* sc_flags */
    130 #define	VNF_ALIVE	0x01
    131 #define VNF_INITED	0x02
    132 #define VNF_WANTED	0x40
    133 #define VNF_LOCKED	0x80
    134 
    135 struct vnd_softc *vnd_softc;
    136 int numvnd = 0;
    137 
    138 /* called by main() at boot time */
    139 void	vndattach __P((int));
    140 
    141 void	vndclear __P((struct vnd_softc *));
    142 void	vndstart __P((struct vnd_softc *));
    143 int	vndsetcred __P((struct vnd_softc *, struct ucred *));
    144 void	vndthrottle __P((struct vnd_softc *, struct vnode *));
    145 void	vndiodone __P((struct buf *));
    146 void	vndshutdown __P((void));
    147 
    148 static	int vndlock __P((struct vnd_softc *));
    149 static	void vndunlock __P((struct vnd_softc *));
    150 
    151 void
    152 vndattach(num)
    153 	int num;
    154 {
    155 	char *mem;
    156 	register u_long size;
    157 
    158 	if (num <= 0)
    159 		return;
    160 	size = num * sizeof(struct vnd_softc);
    161 	mem = malloc(size, M_DEVBUF, M_NOWAIT);
    162 	if (mem == NULL) {
    163 		printf("WARNING: no memory for vnode disks\n");
    164 		return;
    165 	}
    166 	bzero(mem, size);
    167 	vnd_softc = (struct vnd_softc *)mem;
    168 	numvnd = num;
    169 }
    170 
    171 int
    172 vndopen(dev, flags, mode, p)
    173 	dev_t dev;
    174 	int flags, mode;
    175 	struct proc *p;
    176 {
    177 	int unit = vndunit(dev);
    178 	struct vnd_softc *sc;
    179 	int error = 0, part, pmask;
    180 
    181 	/*
    182 	 * XXX Should support disklabels.
    183 	 */
    184 
    185 #ifdef DEBUG
    186 	if (vnddebug & VDB_FOLLOW)
    187 		printf("vndopen(%x, %x, %x, %p)\n", dev, flags, mode, p);
    188 #endif
    189 	if (unit >= numvnd)
    190 		return (ENXIO);
    191 	sc = &vnd_softc[unit];
    192 
    193 	if ((error = vndlock(sc)) != 0)
    194 		return (error);
    195 
    196 	part = DISKPART(dev);
    197 	pmask = (1 << part);
    198 
    199 	/* Prevent our unit from being unconfigured while open. */
    200 	switch (mode) {
    201 	case S_IFCHR:
    202 		sc->sc_dkdev.dk_copenmask |= pmask;
    203 		break;
    204 
    205 	case S_IFBLK:
    206 		sc->sc_dkdev.dk_bopenmask |= pmask;
    207 		break;
    208 	}
    209 	sc->sc_dkdev.dk_openmask =
    210 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
    211 
    212 	vndunlock(sc);
    213 	return (0);
    214 }
    215 
    216 int
    217 vndclose(dev, flags, mode, p)
    218 	dev_t dev;
    219 	int flags, mode;
    220 	struct proc *p;
    221 {
    222 	int unit = vndunit(dev);
    223 	struct vnd_softc *sc;
    224 	int error = 0, part;
    225 
    226 #ifdef DEBUG
    227 	if (vnddebug & VDB_FOLLOW)
    228 		printf("vndclose(%x, %x, %x, %p)\n", dev, flags, mode, p);
    229 #endif
    230 
    231 	if (unit >= numvnd)
    232 		return (ENXIO);
    233 	sc = &vnd_softc[unit];
    234 
    235 	if ((error = vndlock(sc)) != 0)
    236 		return (error);
    237 
    238 	part = DISKPART(dev);
    239 
    240 	/* ...that much closer to allowing unconfiguration... */
    241 	switch (mode) {
    242 	case S_IFCHR:
    243 		sc->sc_dkdev.dk_copenmask &= ~(1 << part);
    244 		break;
    245 
    246 	case S_IFBLK:
    247 		sc->sc_dkdev.dk_bopenmask &= ~(1 << part);
    248 		break;
    249 	}
    250 	sc->sc_dkdev.dk_openmask =
    251 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
    252 
    253 	vndunlock(sc);
    254 	return (0);
    255 }
    256 
    257 /*
    258  * Break the request into bsize pieces and submit using VOP_BMAP/VOP_STRATEGY.
    259  * Note that this driver can only be used for swapping over NFS on the hp
    260  * since nfs_strategy on the vax cannot handle u-areas and page tables.
    261  */
    262 void
    263 vndstrategy(bp)
    264 	register struct buf *bp;
    265 {
    266 	int unit = vndunit(bp->b_dev);
    267 	struct vnd_softc *vnd = &vnd_softc[unit];
    268 	struct vndbuf *nbp;
    269 	struct vndxfer *vnx;
    270 	int bn, bsize, resid;
    271 	caddr_t addr;
    272 	int sz, flags, error;
    273 
    274 #ifdef DEBUG
    275 	if (vnddebug & VDB_FOLLOW)
    276 		printf("vndstrategy(%p): unit %d\n", bp, unit);
    277 #endif
    278 	if ((vnd->sc_flags & VNF_INITED) == 0) {
    279 		bp->b_error = ENXIO;
    280 		bp->b_flags |= B_ERROR;
    281 		biodone(bp);
    282 		return;
    283 	}
    284 	bn = bp->b_blkno;
    285 	sz = howmany(bp->b_bcount, DEV_BSIZE);
    286 	bp->b_resid = bp->b_bcount;
    287 	if (bn < 0 || bn + sz > vnd->sc_size) {
    288 		if (bn != vnd->sc_size) {
    289 			bp->b_error = EINVAL;
    290 			bp->b_flags |= B_ERROR;
    291 		}
    292 		biodone(bp);
    293 		return;
    294 	}
    295 	bn = dbtob(bn);
    296  	bsize = vnd->sc_vp->v_mount->mnt_stat.f_iosize;
    297 	addr = bp->b_data;
    298 	flags = bp->b_flags | B_CALL;
    299 
    300 	/* Allocate a header for this transfer and link it to the buffer */
    301 	vnx = getvndxfer();
    302 	vnx->vx_error = 0;
    303 	vnx->vx_pending = 0;
    304 	vnx->vx_bp = bp;
    305 
    306 	for (resid = bp->b_resid; resid; resid -= sz) {
    307 		struct vnode *vp;
    308 		daddr_t nbn;
    309 		int off, s, nra;
    310 
    311 		nra = 0;
    312 		VOP_LOCK(vnd->sc_vp);
    313 		error = VOP_BMAP(vnd->sc_vp, bn / bsize, &vp, &nbn, &nra);
    314 		VOP_UNLOCK(vnd->sc_vp);
    315 
    316 #define VND_FILLHOLES
    317 #ifdef VND_FILLHOLES
    318 		if (error == 0 && (long)nbn == -1) {
    319 			int rw = (flags & B_READ) ? UIO_READ : UIO_WRITE;
    320 			sz = resid;
    321 			error = vn_rdwr(rw, vnd->sc_vp, addr, sz,
    322 					bn, UIO_SYSSPACE,
    323 					IO_SYNC | IO_NODELOCKED,
    324 					vnd->sc_cred, &resid, bp->b_proc);
    325 			s = splbio();
    326 			bp->b_resid -= (sz - resid);
    327 			splx(s);
    328 		}
    329 #else
    330 		if (error == 0 && (long)nbn == -1)
    331 			error = EIO;
    332 #endif
    333 
    334 		/*
    335 		 * If there was an error or a hole in the file...punt.
    336 		 * Note that we may have to wait for any operations
    337 		 * that we have already fired off before releasing
    338 		 * the buffer.
    339 		 *
    340 		 * XXX we could deal with holes here but it would be
    341 		 * a hassle (in the write case).
    342 		 */
    343 		if (error || (long)nbn == -1) {
    344 			vnx->vx_error = error;
    345 			s = splbio();
    346 			if (vnx->vx_pending == 0) {
    347 				if (error) {
    348 					bp->b_error = error;
    349 					bp->b_flags |= B_ERROR;
    350 				}
    351 				putvndxfer(vnx);
    352 				biodone(bp);
    353 			}
    354 			splx(s);
    355 			return;
    356 		}
    357 
    358 #ifdef DEBUG
    359 		if (!dovndcluster)
    360 			nra = 0;
    361 #endif
    362 
    363 		if ((off = bn % bsize) != 0)
    364 			sz = bsize - off;
    365 		else
    366 			sz = (1 + nra) * bsize;
    367 		if (resid < sz)
    368 			sz = resid;
    369 #ifdef DEBUG
    370 		if (vnddebug & VDB_IO)
    371 			printf("vndstrategy: vp %p/%p bn %x/%x sz %x\n",
    372 			    vnd->sc_vp, vp, bn, nbn, sz);
    373 #endif
    374 
    375 		nbp = getvndbuf();
    376 		nbp->vb_buf.b_flags = flags;
    377 		nbp->vb_buf.b_bcount = sz;
    378 		nbp->vb_buf.b_bufsize = bp->b_bufsize;
    379 		nbp->vb_buf.b_error = 0;
    380 		if (vp->v_type == VBLK || vp->v_type == VCHR)
    381 			nbp->vb_buf.b_dev = vp->v_rdev;
    382 		else
    383 			nbp->vb_buf.b_dev = NODEV;
    384 		nbp->vb_buf.b_data = addr;
    385 		nbp->vb_buf.b_blkno = nbn + btodb(off);
    386 		nbp->vb_buf.b_proc = bp->b_proc;
    387 		nbp->vb_buf.b_iodone = vndiodone;
    388 		nbp->vb_buf.b_vp = vp;
    389 		nbp->vb_buf.b_rcred = vnd->sc_cred;	/* XXX crdup? */
    390 		nbp->vb_buf.b_wcred = vnd->sc_cred;	/* XXX crdup? */
    391 		if (bp->b_dirtyend == 0) {
    392 			nbp->vb_buf.b_dirtyoff = 0;
    393 			nbp->vb_buf.b_dirtyend = sz;
    394 		} else {
    395 			nbp->vb_buf.b_dirtyoff =
    396 			    max(0, bp->b_dirtyoff - (bp->b_bcount - resid));
    397 			nbp->vb_buf.b_dirtyend =
    398 			    min(sz,
    399 				max(0, bp->b_dirtyend - (bp->b_bcount-resid)));
    400 		}
    401 		if (bp->b_validend == 0) {
    402 			nbp->vb_buf.b_validoff = 0;
    403 			nbp->vb_buf.b_validend = sz;
    404 		} else {
    405 			nbp->vb_buf.b_validoff =
    406 			    max(0, bp->b_validoff - (bp->b_bcount - resid));
    407 			nbp->vb_buf.b_validend =
    408 			    min(sz,
    409 				max(0, bp->b_validend - (bp->b_bcount-resid)));
    410 		}
    411 
    412 		nbp->vb_xfer = vnx;
    413 
    414 		/*
    415 		 * Just sort by block number
    416 		 */
    417 		nbp->vb_buf.b_cylin = nbp->vb_buf.b_blkno;
    418 		s = splbio();
    419 		vnx->vx_pending++;
    420 		disksort(&vnd->sc_tab, &nbp->vb_buf);
    421 		if (vnd->sc_tab.b_active < vnd->sc_maxactive) {
    422 			vnd->sc_tab.b_active++;
    423 			vndstart(vnd);
    424 		}
    425 		splx(s);
    426 		bn += sz;
    427 		addr += sz;
    428 	}
    429 }
    430 
    431 /*
    432  * Feed requests sequentially.
    433  * We do it this way to keep from flooding NFS servers if we are connected
    434  * to an NFS file.  This places the burden on the client rather than the
    435  * server.
    436  */
    437 void
    438 vndstart(vnd)
    439 	register struct vnd_softc *vnd;
    440 {
    441 	register struct buf *bp;
    442 
    443 	/*
    444 	 * Dequeue now since lower level strategy routine might
    445 	 * queue using same links
    446 	 */
    447 	bp = vnd->sc_tab.b_actf;
    448 	vnd->sc_tab.b_actf = bp->b_actf;
    449 #ifdef DEBUG
    450 	if (vnddebug & VDB_IO)
    451 		printf("vndstart(%ld): bp %p vp %p blkno %x addr %p cnt %lx\n",
    452 		    (long) (vnd-vnd_softc), bp, bp->b_vp, bp->b_blkno,
    453 		    bp->b_data, bp->b_bcount);
    454 #endif
    455 
    456 	/* Instrumentation. */
    457 	disk_busy(&vnd->sc_dkdev);
    458 
    459 	if ((bp->b_flags & B_READ) == 0)
    460 		bp->b_vp->v_numoutput++;
    461 	VOP_STRATEGY(bp);
    462 }
    463 
    464 void
    465 vndiodone(bp)
    466 	struct buf *bp;
    467 {
    468 	register struct vndbuf *vbp = (struct vndbuf *) bp;
    469 	register struct vndxfer *vnx = (struct vndxfer *)vbp->vb_xfer;
    470 	register struct buf *pbp = vnx->vx_bp;
    471 	register struct vnd_softc *vnd = &vnd_softc[vndunit(pbp->b_dev)];
    472 	int s, resid;
    473 
    474 	s = splbio();
    475 #ifdef DEBUG
    476 	if (vnddebug & VDB_IO)
    477 		printf("vndiodone(%ld): vbp %p vp %p blkno %x addr %p cnt %lx\n",
    478 		    (long) (vnd-vnd_softc), vbp, vbp->vb_buf.b_vp,
    479 		    vbp->vb_buf.b_blkno, vbp->vb_buf.b_data,
    480 		    vbp->vb_buf.b_bcount);
    481 #endif
    482 
    483 	resid = vbp->vb_buf.b_bcount - vbp->vb_buf.b_resid;
    484 	pbp->b_resid -= resid;
    485 	disk_unbusy(&vnd->sc_dkdev, resid);
    486 	vnx->vx_pending--;
    487 
    488 	if (vbp->vb_buf.b_error) {
    489 #ifdef DEBUG
    490 		if (vnddebug & VDB_IO)
    491 			printf("vndiodone: vbp %p error %d\n", vbp,
    492 			    vbp->vb_buf.b_error);
    493 #endif
    494 		vnx->vx_error = vbp->vb_buf.b_error;
    495 	}
    496 	putvndbuf(vbp);
    497 
    498 	/*
    499 	 * Wrap up this transaction if it has run to completion or, in
    500 	 * case of an error, when all auxiliary buffers have returned.
    501 	 */
    502 	if (pbp->b_resid == 0 || (vnx->vx_error && vnx->vx_pending == 0)) {
    503 
    504 		if (vnx->vx_error != 0) {
    505 			pbp->b_flags |= B_ERROR;
    506 			pbp->b_error = vnx->vx_error;
    507 		}
    508 		putvndxfer(vnx);
    509 #ifdef DEBUG
    510 		if (vnddebug & VDB_IO)
    511 			printf("vndiodone: pbp %p iodone\n", pbp);
    512 #endif
    513 		biodone(pbp);
    514 	}
    515 
    516 	if (vnd->sc_tab.b_actf)
    517 		vndstart(vnd);
    518 	else
    519 		vnd->sc_tab.b_active--;
    520 	splx(s);
    521 }
    522 
    523 /* ARGSUSED */
    524 int
    525 vndread(dev, uio, flags)
    526 	dev_t dev;
    527 	struct uio *uio;
    528 	int flags;
    529 {
    530 	int unit = vndunit(dev);
    531 	struct vnd_softc *sc;
    532 
    533 #ifdef DEBUG
    534 	if (vnddebug & VDB_FOLLOW)
    535 		printf("vndread(%x, %p)\n", dev, uio);
    536 #endif
    537 
    538 	if (unit >= numvnd)
    539 		return (ENXIO);
    540 	sc = &vnd_softc[unit];
    541 
    542 	if ((sc->sc_flags & VNF_INITED) == 0)
    543 		return (ENXIO);
    544 
    545 	return (physio(vndstrategy, NULL, dev, B_READ, minphys, uio));
    546 }
    547 
    548 /* ARGSUSED */
    549 int
    550 vndwrite(dev, uio, flags)
    551 	dev_t dev;
    552 	struct uio *uio;
    553 	int flags;
    554 {
    555 	int unit = vndunit(dev);
    556 	struct vnd_softc *sc;
    557 
    558 #ifdef DEBUG
    559 	if (vnddebug & VDB_FOLLOW)
    560 		printf("vndwrite(%x, %p)\n", dev, uio);
    561 #endif
    562 
    563 	if (unit >= numvnd)
    564 		return (ENXIO);
    565 	sc = &vnd_softc[unit];
    566 
    567 	if ((sc->sc_flags & VNF_INITED) == 0)
    568 		return (ENXIO);
    569 
    570 	return (physio(vndstrategy, NULL, dev, B_WRITE, minphys, uio));
    571 }
    572 
    573 /* ARGSUSED */
    574 int
    575 vndioctl(dev, cmd, data, flag, p)
    576 	dev_t dev;
    577 	u_long cmd;
    578 	caddr_t data;
    579 	int flag;
    580 	struct proc *p;
    581 {
    582 	int unit = vndunit(dev);
    583 	register struct vnd_softc *vnd;
    584 	struct vnd_ioctl *vio;
    585 	struct vattr vattr;
    586 	struct nameidata nd;
    587 	int error, part, pmask;
    588 
    589 #ifdef DEBUG
    590 	if (vnddebug & VDB_FOLLOW)
    591 		printf("vndioctl(%x, %lx, %p, %x, %p): unit %d\n",
    592 		    dev, cmd, data, flag, p, unit);
    593 #endif
    594 	error = suser(p->p_ucred, &p->p_acflag);
    595 	if (error)
    596 		return (error);
    597 	if (unit >= numvnd)
    598 		return (ENXIO);
    599 
    600 	vnd = &vnd_softc[unit];
    601 	vio = (struct vnd_ioctl *)data;
    602 	switch (cmd) {
    603 
    604 	case VNDIOCSET:
    605 		if (vnd->sc_flags & VNF_INITED)
    606 			return (EBUSY);
    607 
    608 		if ((error = vndlock(vnd)) != 0)
    609 			return (error);
    610 
    611 		/*
    612 		 * Always open for read and write.
    613 		 * This is probably bogus, but it lets vn_open()
    614 		 * weed out directories, sockets, etc. so we don't
    615 		 * have to worry about them.
    616 		 */
    617 		NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, vio->vnd_file, p);
    618 		if ((error = vn_open(&nd, FREAD|FWRITE, 0)) != 0) {
    619 			vndunlock(vnd);
    620 			return(error);
    621 		}
    622 		error = VOP_GETATTR(nd.ni_vp, &vattr, p->p_ucred, p);
    623 		if (error) {
    624 			VOP_UNLOCK(nd.ni_vp);
    625 			(void) vn_close(nd.ni_vp, FREAD|FWRITE, p->p_ucred, p);
    626 			vndunlock(vnd);
    627 			return(error);
    628 		}
    629 		VOP_UNLOCK(nd.ni_vp);
    630 		vnd->sc_vp = nd.ni_vp;
    631 		vnd->sc_size = btodb(vattr.va_size);	/* note truncation */
    632 		if ((error = vndsetcred(vnd, p->p_ucred)) != 0) {
    633 			(void) vn_close(nd.ni_vp, FREAD|FWRITE, p->p_ucred, p);
    634 			vndunlock(vnd);
    635 			return(error);
    636 		}
    637 		vndthrottle(vnd, vnd->sc_vp);
    638 		vio->vnd_size = dbtob(vnd->sc_size);
    639 		vnd->sc_flags |= VNF_INITED;
    640 #ifdef DEBUG
    641 		if (vnddebug & VDB_INIT)
    642 			printf("vndioctl: SET vp %p size %lx\n",
    643 			    vnd->sc_vp, (unsigned long) vnd->sc_size);
    644 #endif
    645 
    646 		/* Attach the disk. */
    647 		bzero(vnd->sc_xname, sizeof(vnd->sc_xname));	/* XXX */
    648 		sprintf(vnd->sc_xname, "vnd%d", unit);		/* XXX */
    649 		vnd->sc_dkdev.dk_name = vnd->sc_xname;
    650 		disk_attach(&vnd->sc_dkdev);
    651 
    652 		vndunlock(vnd);
    653 
    654 		break;
    655 
    656 	case VNDIOCCLR:
    657 		if ((vnd->sc_flags & VNF_INITED) == 0)
    658 			return (ENXIO);
    659 
    660 		if ((error = vndlock(vnd)) != 0)
    661 			return (error);
    662 
    663 		/*
    664 		 * Don't unconfigure if any other partitions are open
    665 		 * or if both the character and block flavors of this
    666 		 * partition are open.
    667 		 */
    668 		part = DISKPART(dev);
    669 		pmask = (1 << part);
    670 		if ((vnd->sc_dkdev.dk_openmask & ~pmask) ||
    671 		    ((vnd->sc_dkdev.dk_bopenmask & pmask) &&
    672 		    (vnd->sc_dkdev.dk_copenmask & pmask))) {
    673 			vndunlock(vnd);
    674 			return (EBUSY);
    675 		}
    676 
    677 		vndclear(vnd);
    678 #ifdef DEBUG
    679 		if (vnddebug & VDB_INIT)
    680 			printf("vndioctl: CLRed\n");
    681 #endif
    682 
    683 		/* Detatch the disk. */
    684 		disk_detach(&vnd->sc_dkdev);
    685 
    686 		vndunlock(vnd);
    687 
    688 		break;
    689 
    690 	/*
    691 	 * XXX Should support disklabels.
    692 	 */
    693 
    694 	default:
    695 		return(ENOTTY);
    696 	}
    697 
    698 	return (0);
    699 }
    700 
    701 /*
    702  * Duplicate the current processes' credentials.  Since we are called only
    703  * as the result of a SET ioctl and only root can do that, any future access
    704  * to this "disk" is essentially as root.  Note that credentials may change
    705  * if some other uid can write directly to the mapped file (NFS).
    706  */
    707 int
    708 vndsetcred(vnd, cred)
    709 	register struct vnd_softc *vnd;
    710 	struct ucred *cred;
    711 {
    712 	struct uio auio;
    713 	struct iovec aiov;
    714 	char *tmpbuf;
    715 	int error;
    716 
    717 	vnd->sc_cred = crdup(cred);
    718 	tmpbuf = malloc(DEV_BSIZE, M_TEMP, M_WAITOK);
    719 
    720 	/* XXX: Horrible kludge to establish credentials for NFS */
    721 	aiov.iov_base = tmpbuf;
    722 	aiov.iov_len = min(DEV_BSIZE, dbtob(vnd->sc_size));
    723 	auio.uio_iov = &aiov;
    724 	auio.uio_iovcnt = 1;
    725 	auio.uio_offset = 0;
    726 	auio.uio_rw = UIO_READ;
    727 	auio.uio_segflg = UIO_SYSSPACE;
    728 	auio.uio_resid = aiov.iov_len;
    729 	VOP_LOCK(vnd->sc_vp);
    730 	error = VOP_READ(vnd->sc_vp, &auio, 0, vnd->sc_cred);
    731 	VOP_UNLOCK(vnd->sc_vp);
    732 
    733 	free(tmpbuf, M_TEMP);
    734 	return (error);
    735 }
    736 
    737 /*
    738  * Set maxactive based on FS type
    739  */
    740 void
    741 vndthrottle(vnd, vp)
    742 	register struct vnd_softc *vnd;
    743 	struct vnode *vp;
    744 {
    745 #ifdef NFS
    746 	extern int (**nfsv2_vnodeop_p) __P((void *));
    747 
    748 	if (vp->v_op == nfsv2_vnodeop_p)
    749 		vnd->sc_maxactive = 2;
    750 	else
    751 #endif
    752 		vnd->sc_maxactive = 8;
    753 
    754 	if (vnd->sc_maxactive < 1)
    755 		vnd->sc_maxactive = 1;
    756 }
    757 
    758 void
    759 vndshutdown()
    760 {
    761 	register struct vnd_softc *vnd;
    762 
    763 	for (vnd = &vnd_softc[0]; vnd < &vnd_softc[numvnd]; vnd++)
    764 		if (vnd->sc_flags & VNF_INITED)
    765 			vndclear(vnd);
    766 }
    767 
    768 void
    769 vndclear(vnd)
    770 	register struct vnd_softc *vnd;
    771 {
    772 	register struct vnode *vp = vnd->sc_vp;
    773 	struct proc *p = curproc;		/* XXX */
    774 
    775 #ifdef DEBUG
    776 	if (vnddebug & VDB_FOLLOW)
    777 		printf("vndclear(%p): vp %p\n", vnd, vp);
    778 #endif
    779 	vnd->sc_flags &= ~VNF_INITED;
    780 	if (vp == (struct vnode *)0)
    781 		panic("vndioctl: null vp");
    782 	(void) vn_close(vp, FREAD|FWRITE, vnd->sc_cred, p);
    783 	crfree(vnd->sc_cred);
    784 	vnd->sc_vp = (struct vnode *)0;
    785 	vnd->sc_cred = (struct ucred *)0;
    786 	vnd->sc_size = 0;
    787 }
    788 
    789 int
    790 vndsize(dev)
    791 	dev_t dev;
    792 {
    793 	int unit = vndunit(dev);
    794 	register struct vnd_softc *vnd = &vnd_softc[unit];
    795 
    796 	if (unit >= numvnd || (vnd->sc_flags & VNF_INITED) == 0)
    797 		return(-1);
    798 	return(vnd->sc_size);
    799 }
    800 
    801 int
    802 vnddump(dev, blkno, va, size)
    803 	dev_t dev;
    804 	daddr_t blkno;
    805 	caddr_t va;
    806 	size_t size;
    807 {
    808 
    809 	/* Not implemented. */
    810 	return ENXIO;
    811 }
    812 
    813 /*
    814  * Wait interruptibly for an exclusive lock.
    815  *
    816  * XXX
    817  * Several drivers do this; it should be abstracted and made MP-safe.
    818  */
    819 static int
    820 vndlock(sc)
    821 	struct vnd_softc *sc;
    822 {
    823 	int error;
    824 
    825 	while ((sc->sc_flags & VNF_LOCKED) != 0) {
    826 		sc->sc_flags |= VNF_WANTED;
    827 		if ((error = tsleep(sc, PRIBIO | PCATCH, "vndlck", 0)) != 0)
    828 			return (error);
    829 	}
    830 	sc->sc_flags |= VNF_LOCKED;
    831 	return (0);
    832 }
    833 
    834 /*
    835  * Unlock and wake up any waiters.
    836  */
    837 static void
    838 vndunlock(sc)
    839 	struct vnd_softc *sc;
    840 {
    841 
    842 	sc->sc_flags &= ~VNF_LOCKED;
    843 	if ((sc->sc_flags & VNF_WANTED) != 0) {
    844 		sc->sc_flags &= ~VNF_WANTED;
    845 		wakeup(sc);
    846 	}
    847 }
    848