Home | History | Annotate | Line # | Download | only in dev
vnd.c revision 1.38
      1 /*	$NetBSD: vnd.c,v 1.38 1997/05/26 20:28:38 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 #ifdef VND_FILLHOLES
    317 		if (error == 0 && (long)nbn == -1) {
    318 			int rw = (flags & B_READ) ? UIO_READ : UIO_WRITE;
    319 			sz = resid;
    320 			error = vn_rdwr(rw, vnd->sc_vp, addr, sz,
    321 					bn, UIO_SYSSPACE,
    322 					IO_SYNC | IO_NODELOCKED,
    323 					vnd->sc_cred, &resid, bp->b_proc);
    324 			bp->b_resid -= (sz - resid);
    325 		}
    326 #else
    327 		if (error == 0 && (long)nbn == -1)
    328 			error = EIO;
    329 #endif
    330 
    331 		/*
    332 		 * If there was an error or a hole in the file...punt.
    333 		 * Note that we may have to wait for any operations
    334 		 * that we have already fired off before releasing
    335 		 * the buffer.
    336 		 *
    337 		 * XXX we could deal with holes here but it would be
    338 		 * a hassle (in the write case).
    339 		 */
    340 		if (error || (long)nbn == -1) {
    341 			vnx->vx_error = error;
    342 			if (vnx->vx_pending == 0) {
    343 				if (error) {
    344 					bp->b_error = error;
    345 					bp->b_flags |= B_ERROR;
    346 				}
    347 				putvndxfer(vnx);
    348 				biodone(bp);
    349 			}
    350 			return;
    351 		}
    352 
    353 #ifdef DEBUG
    354 		if (!dovndcluster)
    355 			nra = 0;
    356 #endif
    357 
    358 		if ((off = bn % bsize) != 0)
    359 			sz = bsize - off;
    360 		else
    361 			sz = (1 + nra) * bsize;
    362 		if (resid < sz)
    363 			sz = resid;
    364 #ifdef DEBUG
    365 		if (vnddebug & VDB_IO)
    366 			printf("vndstrategy: vp %p/%p bn %x/%x sz %x\n",
    367 			    vnd->sc_vp, vp, bn, nbn, sz);
    368 #endif
    369 
    370 		nbp = getvndbuf();
    371 		nbp->vb_buf.b_flags = flags;
    372 		nbp->vb_buf.b_bcount = sz;
    373 		nbp->vb_buf.b_bufsize = bp->b_bufsize;
    374 		nbp->vb_buf.b_error = 0;
    375 		if (vp->v_type == VBLK || vp->v_type == VCHR)
    376 			nbp->vb_buf.b_dev = vp->v_rdev;
    377 		else
    378 			nbp->vb_buf.b_dev = NODEV;
    379 		nbp->vb_buf.b_data = addr;
    380 		nbp->vb_buf.b_blkno = nbn + btodb(off);
    381 		nbp->vb_buf.b_proc = bp->b_proc;
    382 		nbp->vb_buf.b_iodone = vndiodone;
    383 		nbp->vb_buf.b_vp = vp;
    384 		nbp->vb_buf.b_rcred = vnd->sc_cred;	/* XXX crdup? */
    385 		nbp->vb_buf.b_wcred = vnd->sc_cred;	/* XXX crdup? */
    386 		if (bp->b_dirtyend == 0) {
    387 			nbp->vb_buf.b_dirtyoff = 0;
    388 			nbp->vb_buf.b_dirtyend = sz;
    389 		} else {
    390 			nbp->vb_buf.b_dirtyoff =
    391 			    max(0, bp->b_dirtyoff - (bp->b_bcount - resid));
    392 			nbp->vb_buf.b_dirtyend =
    393 			    min(sz,
    394 				max(0, bp->b_dirtyend - (bp->b_bcount-resid)));
    395 		}
    396 		if (bp->b_validend == 0) {
    397 			nbp->vb_buf.b_validoff = 0;
    398 			nbp->vb_buf.b_validend = sz;
    399 		} else {
    400 			nbp->vb_buf.b_validoff =
    401 			    max(0, bp->b_validoff - (bp->b_bcount - resid));
    402 			nbp->vb_buf.b_validend =
    403 			    min(sz,
    404 				max(0, bp->b_validend - (bp->b_bcount-resid)));
    405 		}
    406 
    407 		nbp->vb_xfer = vnx;
    408 		vnx->vx_pending++;
    409 
    410 		/*
    411 		 * Just sort by block number
    412 		 */
    413 		nbp->vb_buf.b_cylin = nbp->vb_buf.b_blkno;
    414 		s = splbio();
    415 		disksort(&vnd->sc_tab, &nbp->vb_buf);
    416 		if (vnd->sc_tab.b_active < vnd->sc_maxactive) {
    417 			vnd->sc_tab.b_active++;
    418 			vndstart(vnd);
    419 		}
    420 		splx(s);
    421 		bn += sz;
    422 		addr += sz;
    423 	}
    424 }
    425 
    426 /*
    427  * Feed requests sequentially.
    428  * We do it this way to keep from flooding NFS servers if we are connected
    429  * to an NFS file.  This places the burden on the client rather than the
    430  * server.
    431  */
    432 void
    433 vndstart(vnd)
    434 	register struct vnd_softc *vnd;
    435 {
    436 	register struct buf *bp;
    437 
    438 	/*
    439 	 * Dequeue now since lower level strategy routine might
    440 	 * queue using same links
    441 	 */
    442 	bp = vnd->sc_tab.b_actf;
    443 	vnd->sc_tab.b_actf = bp->b_actf;
    444 #ifdef DEBUG
    445 	if (vnddebug & VDB_IO)
    446 		printf("vndstart(%ld): bp %p vp %p blkno %x addr %p cnt %lx\n",
    447 		    (long) (vnd-vnd_softc), bp, bp->b_vp, bp->b_blkno,
    448 		    bp->b_data, bp->b_bcount);
    449 #endif
    450 
    451 	/* Instrumentation. */
    452 	disk_busy(&vnd->sc_dkdev);
    453 
    454 	if ((bp->b_flags & B_READ) == 0)
    455 		bp->b_vp->v_numoutput++;
    456 	VOP_STRATEGY(bp);
    457 }
    458 
    459 void
    460 vndiodone(bp)
    461 	struct buf *bp;
    462 {
    463 	register struct vndbuf *vbp = (struct vndbuf *) bp;
    464 	register struct vndxfer *vnx = (struct vndxfer *)vbp->vb_xfer;
    465 	register struct buf *pbp = vnx->vx_bp;
    466 	register struct vnd_softc *vnd = &vnd_softc[vndunit(pbp->b_dev)];
    467 	int s, resid;
    468 
    469 	s = splbio();
    470 #ifdef DEBUG
    471 	if (vnddebug & VDB_IO)
    472 		printf("vndiodone(%ld): vbp %p vp %p blkno %x addr %p cnt %lx\n",
    473 		    (long) (vnd-vnd_softc), vbp, vbp->vb_buf.b_vp,
    474 		    vbp->vb_buf.b_blkno, vbp->vb_buf.b_data,
    475 		    vbp->vb_buf.b_bcount);
    476 #endif
    477 
    478 	resid = vbp->vb_buf.b_bcount - vbp->vb_buf.b_resid;
    479 	pbp->b_resid -= resid;
    480 	disk_unbusy(&vnd->sc_dkdev, resid);
    481 	vnx->vx_pending--;
    482 
    483 	if (vbp->vb_buf.b_error) {
    484 #ifdef DEBUG
    485 		if (vnddebug & VDB_IO)
    486 			printf("vndiodone: vbp %p error %d\n", vbp,
    487 			    vbp->vb_buf.b_error);
    488 #endif
    489 		vnx->vx_error = vbp->vb_buf.b_error;
    490 	}
    491 	putvndbuf(vbp);
    492 
    493 	/*
    494 	 * Wrap up this transaction if it has run to completion or, in
    495 	 * case of an error, when all auxiliary buffers have returned.
    496 	 */
    497 	if (pbp->b_resid == 0 || (vnx->vx_error && vnx->vx_pending == 0)) {
    498 
    499 		if (vnx->vx_error != 0) {
    500 			pbp->b_flags |= B_ERROR;
    501 			pbp->b_error = vnx->vx_error;
    502 		}
    503 		putvndxfer(vnx);
    504 #ifdef DEBUG
    505 		if (vnddebug & VDB_IO)
    506 			printf("vndiodone: pbp %p iodone\n", pbp);
    507 #endif
    508 		biodone(pbp);
    509 	}
    510 
    511 	if (vnd->sc_tab.b_actf)
    512 		vndstart(vnd);
    513 	else
    514 		vnd->sc_tab.b_active--;
    515 	splx(s);
    516 }
    517 
    518 /* ARGSUSED */
    519 int
    520 vndread(dev, uio, flags)
    521 	dev_t dev;
    522 	struct uio *uio;
    523 	int flags;
    524 {
    525 	int unit = vndunit(dev);
    526 	struct vnd_softc *sc;
    527 
    528 #ifdef DEBUG
    529 	if (vnddebug & VDB_FOLLOW)
    530 		printf("vndread(%x, %p)\n", dev, uio);
    531 #endif
    532 
    533 	if (unit >= numvnd)
    534 		return (ENXIO);
    535 	sc = &vnd_softc[unit];
    536 
    537 	if ((sc->sc_flags & VNF_INITED) == 0)
    538 		return (ENXIO);
    539 
    540 	return (physio(vndstrategy, NULL, dev, B_READ, minphys, uio));
    541 }
    542 
    543 /* ARGSUSED */
    544 int
    545 vndwrite(dev, uio, flags)
    546 	dev_t dev;
    547 	struct uio *uio;
    548 	int flags;
    549 {
    550 	int unit = vndunit(dev);
    551 	struct vnd_softc *sc;
    552 
    553 #ifdef DEBUG
    554 	if (vnddebug & VDB_FOLLOW)
    555 		printf("vndwrite(%x, %p)\n", dev, uio);
    556 #endif
    557 
    558 	if (unit >= numvnd)
    559 		return (ENXIO);
    560 	sc = &vnd_softc[unit];
    561 
    562 	if ((sc->sc_flags & VNF_INITED) == 0)
    563 		return (ENXIO);
    564 
    565 	return (physio(vndstrategy, NULL, dev, B_WRITE, minphys, uio));
    566 }
    567 
    568 /* ARGSUSED */
    569 int
    570 vndioctl(dev, cmd, data, flag, p)
    571 	dev_t dev;
    572 	u_long cmd;
    573 	caddr_t data;
    574 	int flag;
    575 	struct proc *p;
    576 {
    577 	int unit = vndunit(dev);
    578 	register struct vnd_softc *vnd;
    579 	struct vnd_ioctl *vio;
    580 	struct vattr vattr;
    581 	struct nameidata nd;
    582 	int error, part, pmask;
    583 
    584 #ifdef DEBUG
    585 	if (vnddebug & VDB_FOLLOW)
    586 		printf("vndioctl(%x, %lx, %p, %x, %p): unit %d\n",
    587 		    dev, cmd, data, flag, p, unit);
    588 #endif
    589 	error = suser(p->p_ucred, &p->p_acflag);
    590 	if (error)
    591 		return (error);
    592 	if (unit >= numvnd)
    593 		return (ENXIO);
    594 
    595 	vnd = &vnd_softc[unit];
    596 	vio = (struct vnd_ioctl *)data;
    597 	switch (cmd) {
    598 
    599 	case VNDIOCSET:
    600 		if (vnd->sc_flags & VNF_INITED)
    601 			return (EBUSY);
    602 
    603 		if ((error = vndlock(vnd)) != 0)
    604 			return (error);
    605 
    606 		/*
    607 		 * Always open for read and write.
    608 		 * This is probably bogus, but it lets vn_open()
    609 		 * weed out directories, sockets, etc. so we don't
    610 		 * have to worry about them.
    611 		 */
    612 		NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, vio->vnd_file, p);
    613 		if ((error = vn_open(&nd, FREAD|FWRITE, 0)) != 0) {
    614 			vndunlock(vnd);
    615 			return(error);
    616 		}
    617 		error = VOP_GETATTR(nd.ni_vp, &vattr, p->p_ucred, p);
    618 		if (error) {
    619 			VOP_UNLOCK(nd.ni_vp);
    620 			(void) vn_close(nd.ni_vp, FREAD|FWRITE, p->p_ucred, p);
    621 			vndunlock(vnd);
    622 			return(error);
    623 		}
    624 		VOP_UNLOCK(nd.ni_vp);
    625 		vnd->sc_vp = nd.ni_vp;
    626 		vnd->sc_size = btodb(vattr.va_size);	/* note truncation */
    627 		if ((error = vndsetcred(vnd, p->p_ucred)) != 0) {
    628 			(void) vn_close(nd.ni_vp, FREAD|FWRITE, p->p_ucred, p);
    629 			vndunlock(vnd);
    630 			return(error);
    631 		}
    632 		vndthrottle(vnd, vnd->sc_vp);
    633 		vio->vnd_size = dbtob(vnd->sc_size);
    634 		vnd->sc_flags |= VNF_INITED;
    635 #ifdef DEBUG
    636 		if (vnddebug & VDB_INIT)
    637 			printf("vndioctl: SET vp %p size %lx\n",
    638 			    vnd->sc_vp, (unsigned long) vnd->sc_size);
    639 #endif
    640 
    641 		/* Attach the disk. */
    642 		bzero(vnd->sc_xname, sizeof(vnd->sc_xname));	/* XXX */
    643 		sprintf(vnd->sc_xname, "vnd%d", unit);		/* XXX */
    644 		vnd->sc_dkdev.dk_name = vnd->sc_xname;
    645 		disk_attach(&vnd->sc_dkdev);
    646 
    647 		vndunlock(vnd);
    648 
    649 		break;
    650 
    651 	case VNDIOCCLR:
    652 		if ((vnd->sc_flags & VNF_INITED) == 0)
    653 			return (ENXIO);
    654 
    655 		if ((error = vndlock(vnd)) != 0)
    656 			return (error);
    657 
    658 		/*
    659 		 * Don't unconfigure if any other partitions are open
    660 		 * or if both the character and block flavors of this
    661 		 * partition are open.
    662 		 */
    663 		part = DISKPART(dev);
    664 		pmask = (1 << part);
    665 		if ((vnd->sc_dkdev.dk_openmask & ~pmask) ||
    666 		    ((vnd->sc_dkdev.dk_bopenmask & pmask) &&
    667 		    (vnd->sc_dkdev.dk_copenmask & pmask))) {
    668 			vndunlock(vnd);
    669 			return (EBUSY);
    670 		}
    671 
    672 		vndclear(vnd);
    673 #ifdef DEBUG
    674 		if (vnddebug & VDB_INIT)
    675 			printf("vndioctl: CLRed\n");
    676 #endif
    677 
    678 		/* Detatch the disk. */
    679 		disk_detach(&vnd->sc_dkdev);
    680 
    681 		vndunlock(vnd);
    682 
    683 		break;
    684 
    685 	/*
    686 	 * XXX Should support disklabels.
    687 	 */
    688 
    689 	default:
    690 		return(ENOTTY);
    691 	}
    692 
    693 	return (0);
    694 }
    695 
    696 /*
    697  * Duplicate the current processes' credentials.  Since we are called only
    698  * as the result of a SET ioctl and only root can do that, any future access
    699  * to this "disk" is essentially as root.  Note that credentials may change
    700  * if some other uid can write directly to the mapped file (NFS).
    701  */
    702 int
    703 vndsetcred(vnd, cred)
    704 	register struct vnd_softc *vnd;
    705 	struct ucred *cred;
    706 {
    707 	struct uio auio;
    708 	struct iovec aiov;
    709 	char *tmpbuf;
    710 	int error;
    711 
    712 	vnd->sc_cred = crdup(cred);
    713 	tmpbuf = malloc(DEV_BSIZE, M_TEMP, M_WAITOK);
    714 
    715 	/* XXX: Horrible kludge to establish credentials for NFS */
    716 	aiov.iov_base = tmpbuf;
    717 	aiov.iov_len = min(DEV_BSIZE, dbtob(vnd->sc_size));
    718 	auio.uio_iov = &aiov;
    719 	auio.uio_iovcnt = 1;
    720 	auio.uio_offset = 0;
    721 	auio.uio_rw = UIO_READ;
    722 	auio.uio_segflg = UIO_SYSSPACE;
    723 	auio.uio_resid = aiov.iov_len;
    724 	VOP_LOCK(vnd->sc_vp);
    725 	error = VOP_READ(vnd->sc_vp, &auio, 0, vnd->sc_cred);
    726 	VOP_UNLOCK(vnd->sc_vp);
    727 
    728 	free(tmpbuf, M_TEMP);
    729 	return (error);
    730 }
    731 
    732 /*
    733  * Set maxactive based on FS type
    734  */
    735 void
    736 vndthrottle(vnd, vp)
    737 	register struct vnd_softc *vnd;
    738 	struct vnode *vp;
    739 {
    740 #ifdef NFS
    741 	extern int (**nfsv2_vnodeop_p) __P((void *));
    742 
    743 	if (vp->v_op == nfsv2_vnodeop_p)
    744 		vnd->sc_maxactive = 2;
    745 	else
    746 #endif
    747 		vnd->sc_maxactive = 8;
    748 
    749 	if (vnd->sc_maxactive < 1)
    750 		vnd->sc_maxactive = 1;
    751 }
    752 
    753 void
    754 vndshutdown()
    755 {
    756 	register struct vnd_softc *vnd;
    757 
    758 	for (vnd = &vnd_softc[0]; vnd < &vnd_softc[numvnd]; vnd++)
    759 		if (vnd->sc_flags & VNF_INITED)
    760 			vndclear(vnd);
    761 }
    762 
    763 void
    764 vndclear(vnd)
    765 	register struct vnd_softc *vnd;
    766 {
    767 	register struct vnode *vp = vnd->sc_vp;
    768 	struct proc *p = curproc;		/* XXX */
    769 
    770 #ifdef DEBUG
    771 	if (vnddebug & VDB_FOLLOW)
    772 		printf("vndclear(%p): vp %p\n", vnd, vp);
    773 #endif
    774 	vnd->sc_flags &= ~VNF_INITED;
    775 	if (vp == (struct vnode *)0)
    776 		panic("vndioctl: null vp");
    777 	(void) vn_close(vp, FREAD|FWRITE, vnd->sc_cred, p);
    778 	crfree(vnd->sc_cred);
    779 	vnd->sc_vp = (struct vnode *)0;
    780 	vnd->sc_cred = (struct ucred *)0;
    781 	vnd->sc_size = 0;
    782 }
    783 
    784 int
    785 vndsize(dev)
    786 	dev_t dev;
    787 {
    788 	int unit = vndunit(dev);
    789 	register struct vnd_softc *vnd = &vnd_softc[unit];
    790 
    791 	if (unit >= numvnd || (vnd->sc_flags & VNF_INITED) == 0)
    792 		return(-1);
    793 	return(vnd->sc_size);
    794 }
    795 
    796 int
    797 vnddump(dev, blkno, va, size)
    798 	dev_t dev;
    799 	daddr_t blkno;
    800 	caddr_t va;
    801 	size_t size;
    802 {
    803 
    804 	/* Not implemented. */
    805 	return ENXIO;
    806 }
    807 
    808 /*
    809  * Wait interruptibly for an exclusive lock.
    810  *
    811  * XXX
    812  * Several drivers do this; it should be abstracted and made MP-safe.
    813  */
    814 static int
    815 vndlock(sc)
    816 	struct vnd_softc *sc;
    817 {
    818 	int error;
    819 
    820 	while ((sc->sc_flags & VNF_LOCKED) != 0) {
    821 		sc->sc_flags |= VNF_WANTED;
    822 		if ((error = tsleep(sc, PRIBIO | PCATCH, "vndlck", 0)) != 0)
    823 			return (error);
    824 	}
    825 	sc->sc_flags |= VNF_LOCKED;
    826 	return (0);
    827 }
    828 
    829 /*
    830  * Unlock and wake up any waiters.
    831  */
    832 static void
    833 vndunlock(sc)
    834 	struct vnd_softc *sc;
    835 {
    836 
    837 	sc->sc_flags &= ~VNF_LOCKED;
    838 	if ((sc->sc_flags & VNF_WANTED) != 0) {
    839 		sc->sc_flags &= ~VNF_WANTED;
    840 		wakeup(sc);
    841 	}
    842 }
    843