Home | History | Annotate | Line # | Download | only in dev
vnd.c revision 1.36
      1 /*	$NetBSD: vnd.c,v 1.36 1997/05/25 19:37:36 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 	register struct vnd_softc *vnd = &vnd_softc[unit];
    268 	register struct vndbuf *nbp;
    269 	struct vndxfer *vnx;
    270 	register int bn, bsize, resid;
    271 	register 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 		if (error == 0 && (long)nbn == -1)
    316 			error = EIO;
    317 
    318 		/*
    319 		 * If there was an error or a hole in the file...punt.
    320 		 * Note that we may have to wait for any operations
    321 		 * that we have already fired off before releasing
    322 		 * the buffer.
    323 		 *
    324 		 * XXX we could deal with holes here but it would be
    325 		 * a hassle (in the write case).
    326 		 */
    327 		if (error) {
    328 			vnx->vx_error = error;
    329 			if (vnx->vx_pending == 0) {
    330 				bp->b_error = error;
    331 				bp->b_flags |= B_ERROR;
    332 				putvndxfer(vnx);
    333 				biodone(bp);
    334 			}
    335 			return;
    336 		}
    337 
    338 #ifdef DEBUG
    339 		if (!dovndcluster)
    340 			nra = 0;
    341 #endif
    342 
    343 		if ((off = bn % bsize) != 0)
    344 			sz = bsize - off;
    345 		else
    346 			sz = (1 + nra) * bsize;
    347 		if (resid < sz)
    348 			sz = resid;
    349 #ifdef DEBUG
    350 		if (vnddebug & VDB_IO)
    351 			printf("vndstrategy: vp %p/%p bn %x/%x sz %x\n",
    352 			    vnd->sc_vp, vp, bn, nbn, sz);
    353 #endif
    354 
    355 		nbp = getvndbuf();
    356 		nbp->vb_buf.b_flags = flags;
    357 		nbp->vb_buf.b_bcount = sz;
    358 		nbp->vb_buf.b_bufsize = bp->b_bufsize;
    359 		nbp->vb_buf.b_error = 0;
    360 		if (vp->v_type == VBLK || vp->v_type == VCHR)
    361 			nbp->vb_buf.b_dev = vp->v_rdev;
    362 		else
    363 			nbp->vb_buf.b_dev = NODEV;
    364 		nbp->vb_buf.b_data = addr;
    365 		nbp->vb_buf.b_blkno = nbn + btodb(off);
    366 		nbp->vb_buf.b_proc = bp->b_proc;
    367 		nbp->vb_buf.b_iodone = vndiodone;
    368 		nbp->vb_buf.b_vp = vp;
    369 		nbp->vb_buf.b_rcred = vnd->sc_cred;	/* XXX crdup? */
    370 		nbp->vb_buf.b_wcred = vnd->sc_cred;	/* XXX crdup? */
    371 		if (bp->b_dirtyend == 0) {
    372 			nbp->vb_buf.b_dirtyoff = 0;
    373 			nbp->vb_buf.b_dirtyend = sz;
    374 		} else {
    375 			nbp->vb_buf.b_dirtyoff =
    376 			    max(0, bp->b_dirtyoff - (bp->b_bcount - resid));
    377 			nbp->vb_buf.b_dirtyend =
    378 			    min(sz,
    379 				max(0, bp->b_dirtyend - (bp->b_bcount-resid)));
    380 		}
    381 		if (bp->b_validend == 0) {
    382 			nbp->vb_buf.b_validoff = 0;
    383 			nbp->vb_buf.b_validend = sz;
    384 		} else {
    385 			nbp->vb_buf.b_validoff =
    386 			    max(0, bp->b_validoff - (bp->b_bcount - resid));
    387 			nbp->vb_buf.b_validend =
    388 			    min(sz,
    389 				max(0, bp->b_validend - (bp->b_bcount-resid)));
    390 		}
    391 
    392 		nbp->vb_xfer = vnx;
    393 		vnx->vx_pending++;
    394 
    395 		/*
    396 		 * Just sort by block number
    397 		 */
    398 		nbp->vb_buf.b_cylin = nbp->vb_buf.b_blkno;
    399 		s = splbio();
    400 		disksort(&vnd->sc_tab, &nbp->vb_buf);
    401 		if (vnd->sc_tab.b_active < vnd->sc_maxactive) {
    402 			vnd->sc_tab.b_active++;
    403 			vndstart(vnd);
    404 		}
    405 		splx(s);
    406 		bn += sz;
    407 		addr += sz;
    408 	}
    409 }
    410 
    411 /*
    412  * Feed requests sequentially.
    413  * We do it this way to keep from flooding NFS servers if we are connected
    414  * to an NFS file.  This places the burden on the client rather than the
    415  * server.
    416  */
    417 void
    418 vndstart(vnd)
    419 	register struct vnd_softc *vnd;
    420 {
    421 	register struct buf *bp;
    422 
    423 	/*
    424 	 * Dequeue now since lower level strategy routine might
    425 	 * queue using same links
    426 	 */
    427 	bp = vnd->sc_tab.b_actf;
    428 	vnd->sc_tab.b_actf = bp->b_actf;
    429 #ifdef DEBUG
    430 	if (vnddebug & VDB_IO)
    431 		printf("vndstart(%ld): bp %p vp %p blkno %x addr %p cnt %lx\n",
    432 		    (long) (vnd-vnd_softc), bp, bp->b_vp, bp->b_blkno,
    433 		    bp->b_data, bp->b_bcount);
    434 #endif
    435 
    436 	/* Instrumentation. */
    437 	disk_busy(&vnd->sc_dkdev);
    438 
    439 	if ((bp->b_flags & B_READ) == 0)
    440 		bp->b_vp->v_numoutput++;
    441 	VOP_STRATEGY(bp);
    442 }
    443 
    444 void
    445 vndiodone(bp)
    446 	struct buf *bp;
    447 {
    448 	register struct vndbuf *vbp = (struct vndbuf *) bp;
    449 	register struct vndxfer *vnx = (struct vndxfer *)vbp->vb_xfer;
    450 	register struct buf *pbp = vnx->vx_bp;
    451 	register struct vnd_softc *vnd = &vnd_softc[vndunit(pbp->b_dev)];
    452 	int s, resid;
    453 
    454 	s = splbio();
    455 #ifdef DEBUG
    456 	if (vnddebug & VDB_IO)
    457 		printf("vndiodone(%ld): vbp %p vp %p blkno %x addr %p cnt %lx\n",
    458 		    (long) (vnd-vnd_softc), vbp, vbp->vb_buf.b_vp,
    459 		    vbp->vb_buf.b_blkno, vbp->vb_buf.b_data,
    460 		    vbp->vb_buf.b_bcount);
    461 #endif
    462 
    463 	resid = vbp->vb_buf.b_bcount - vbp->vb_buf.b_resid;
    464 	pbp->b_resid -= resid;
    465 	disk_unbusy(&vnd->sc_dkdev, resid);
    466 	vnx->vx_pending--;
    467 
    468 	if (vbp->vb_buf.b_error) {
    469 #ifdef DEBUG
    470 		if (vnddebug & VDB_IO)
    471 			printf("vndiodone: vbp %p error %d\n", vbp,
    472 			    vbp->vb_buf.b_error);
    473 #endif
    474 		vnx->vx_error = vbp->vb_buf.b_error;
    475 	}
    476 	putvndbuf(vbp);
    477 
    478 	/*
    479 	 * Wrap up this transaction if it has run to completion or, in
    480 	 * case of an error, when all auxiliary buffers have returned.
    481 	 */
    482 	if (pbp->b_resid == 0 || (vnx->vx_error && vnx->vx_pending == 0)) {
    483 
    484 		if (vnx->vx_error != 0) {
    485 			pbp->b_flags |= B_ERROR;
    486 			pbp->b_error = vnx->vx_error;
    487 		}
    488 		putvndxfer(vnx);
    489 #ifdef DEBUG
    490 		if (vnddebug & VDB_IO)
    491 			printf("vndiodone: pbp %p iodone\n", pbp);
    492 #endif
    493 		biodone(pbp);
    494 	}
    495 
    496 	if (vnd->sc_tab.b_actf)
    497 		vndstart(vnd);
    498 	else
    499 		vnd->sc_tab.b_active--;
    500 	splx(s);
    501 }
    502 
    503 /* ARGSUSED */
    504 int
    505 vndread(dev, uio, flags)
    506 	dev_t dev;
    507 	struct uio *uio;
    508 	int flags;
    509 {
    510 	int unit = vndunit(dev);
    511 	struct vnd_softc *sc;
    512 
    513 #ifdef DEBUG
    514 	if (vnddebug & VDB_FOLLOW)
    515 		printf("vndread(%x, %p)\n", dev, uio);
    516 #endif
    517 
    518 	if (unit >= numvnd)
    519 		return (ENXIO);
    520 	sc = &vnd_softc[unit];
    521 
    522 	if ((sc->sc_flags & VNF_INITED) == 0)
    523 		return (ENXIO);
    524 
    525 	return (physio(vndstrategy, NULL, dev, B_READ, minphys, uio));
    526 }
    527 
    528 /* ARGSUSED */
    529 int
    530 vndwrite(dev, uio, flags)
    531 	dev_t dev;
    532 	struct uio *uio;
    533 	int flags;
    534 {
    535 	int unit = vndunit(dev);
    536 	struct vnd_softc *sc;
    537 
    538 #ifdef DEBUG
    539 	if (vnddebug & VDB_FOLLOW)
    540 		printf("vndwrite(%x, %p)\n", dev, uio);
    541 #endif
    542 
    543 	if (unit >= numvnd)
    544 		return (ENXIO);
    545 	sc = &vnd_softc[unit];
    546 
    547 	if ((sc->sc_flags & VNF_INITED) == 0)
    548 		return (ENXIO);
    549 
    550 	return (physio(vndstrategy, NULL, dev, B_WRITE, minphys, uio));
    551 }
    552 
    553 /* ARGSUSED */
    554 int
    555 vndioctl(dev, cmd, data, flag, p)
    556 	dev_t dev;
    557 	u_long cmd;
    558 	caddr_t data;
    559 	int flag;
    560 	struct proc *p;
    561 {
    562 	int unit = vndunit(dev);
    563 	register struct vnd_softc *vnd;
    564 	struct vnd_ioctl *vio;
    565 	struct vattr vattr;
    566 	struct nameidata nd;
    567 	int error, part, pmask;
    568 
    569 #ifdef DEBUG
    570 	if (vnddebug & VDB_FOLLOW)
    571 		printf("vndioctl(%x, %lx, %p, %x, %p): unit %d\n",
    572 		    dev, cmd, data, flag, p, unit);
    573 #endif
    574 	error = suser(p->p_ucred, &p->p_acflag);
    575 	if (error)
    576 		return (error);
    577 	if (unit >= numvnd)
    578 		return (ENXIO);
    579 
    580 	vnd = &vnd_softc[unit];
    581 	vio = (struct vnd_ioctl *)data;
    582 	switch (cmd) {
    583 
    584 	case VNDIOCSET:
    585 		if (vnd->sc_flags & VNF_INITED)
    586 			return (EBUSY);
    587 
    588 		if ((error = vndlock(vnd)) != 0)
    589 			return (error);
    590 
    591 		/*
    592 		 * Always open for read and write.
    593 		 * This is probably bogus, but it lets vn_open()
    594 		 * weed out directories, sockets, etc. so we don't
    595 		 * have to worry about them.
    596 		 */
    597 		NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, vio->vnd_file, p);
    598 		if ((error = vn_open(&nd, FREAD|FWRITE, 0)) != 0) {
    599 			vndunlock(vnd);
    600 			return(error);
    601 		}
    602 		error = VOP_GETATTR(nd.ni_vp, &vattr, p->p_ucred, p);
    603 		if (error) {
    604 			VOP_UNLOCK(nd.ni_vp);
    605 			(void) vn_close(nd.ni_vp, FREAD|FWRITE, p->p_ucred, p);
    606 			vndunlock(vnd);
    607 			return(error);
    608 		}
    609 		VOP_UNLOCK(nd.ni_vp);
    610 		vnd->sc_vp = nd.ni_vp;
    611 		vnd->sc_size = btodb(vattr.va_size);	/* note truncation */
    612 		if ((error = vndsetcred(vnd, p->p_ucred)) != 0) {
    613 			(void) vn_close(nd.ni_vp, FREAD|FWRITE, p->p_ucred, p);
    614 			vndunlock(vnd);
    615 			return(error);
    616 		}
    617 		vndthrottle(vnd, vnd->sc_vp);
    618 		vio->vnd_size = dbtob(vnd->sc_size);
    619 		vnd->sc_flags |= VNF_INITED;
    620 #ifdef DEBUG
    621 		if (vnddebug & VDB_INIT)
    622 			printf("vndioctl: SET vp %p size %lx\n",
    623 			    vnd->sc_vp, (unsigned long) vnd->sc_size);
    624 #endif
    625 
    626 		/* Attach the disk. */
    627 		bzero(vnd->sc_xname, sizeof(vnd->sc_xname));	/* XXX */
    628 		sprintf(vnd->sc_xname, "vnd%d", unit);		/* XXX */
    629 		vnd->sc_dkdev.dk_name = vnd->sc_xname;
    630 		disk_attach(&vnd->sc_dkdev);
    631 
    632 		vndunlock(vnd);
    633 
    634 		break;
    635 
    636 	case VNDIOCCLR:
    637 		if ((vnd->sc_flags & VNF_INITED) == 0)
    638 			return (ENXIO);
    639 
    640 		if ((error = vndlock(vnd)) != 0)
    641 			return (error);
    642 
    643 		/*
    644 		 * Don't unconfigure if any other partitions are open
    645 		 * or if both the character and block flavors of this
    646 		 * partition are open.
    647 		 */
    648 		part = DISKPART(dev);
    649 		pmask = (1 << part);
    650 		if ((vnd->sc_dkdev.dk_openmask & ~pmask) ||
    651 		    ((vnd->sc_dkdev.dk_bopenmask & pmask) &&
    652 		    (vnd->sc_dkdev.dk_copenmask & pmask))) {
    653 			vndunlock(vnd);
    654 			return (EBUSY);
    655 		}
    656 
    657 		vndclear(vnd);
    658 #ifdef DEBUG
    659 		if (vnddebug & VDB_INIT)
    660 			printf("vndioctl: CLRed\n");
    661 #endif
    662 
    663 		/* Detatch the disk. */
    664 		disk_detach(&vnd->sc_dkdev);
    665 
    666 		vndunlock(vnd);
    667 
    668 		break;
    669 
    670 	/*
    671 	 * XXX Should support disklabels.
    672 	 */
    673 
    674 	default:
    675 		return(ENOTTY);
    676 	}
    677 
    678 	return (0);
    679 }
    680 
    681 /*
    682  * Duplicate the current processes' credentials.  Since we are called only
    683  * as the result of a SET ioctl and only root can do that, any future access
    684  * to this "disk" is essentially as root.  Note that credentials may change
    685  * if some other uid can write directly to the mapped file (NFS).
    686  */
    687 int
    688 vndsetcred(vnd, cred)
    689 	register struct vnd_softc *vnd;
    690 	struct ucred *cred;
    691 {
    692 	struct uio auio;
    693 	struct iovec aiov;
    694 	char *tmpbuf;
    695 	int error;
    696 
    697 	vnd->sc_cred = crdup(cred);
    698 	tmpbuf = malloc(DEV_BSIZE, M_TEMP, M_WAITOK);
    699 
    700 	/* XXX: Horrible kludge to establish credentials for NFS */
    701 	aiov.iov_base = tmpbuf;
    702 	aiov.iov_len = min(DEV_BSIZE, dbtob(vnd->sc_size));
    703 	auio.uio_iov = &aiov;
    704 	auio.uio_iovcnt = 1;
    705 	auio.uio_offset = 0;
    706 	auio.uio_rw = UIO_READ;
    707 	auio.uio_segflg = UIO_SYSSPACE;
    708 	auio.uio_resid = aiov.iov_len;
    709 	VOP_LOCK(vnd->sc_vp);
    710 	error = VOP_READ(vnd->sc_vp, &auio, 0, vnd->sc_cred);
    711 	VOP_UNLOCK(vnd->sc_vp);
    712 
    713 	free(tmpbuf, M_TEMP);
    714 	return (error);
    715 }
    716 
    717 /*
    718  * Set maxactive based on FS type
    719  */
    720 void
    721 vndthrottle(vnd, vp)
    722 	register struct vnd_softc *vnd;
    723 	struct vnode *vp;
    724 {
    725 #ifdef NFS
    726 	extern int (**nfsv2_vnodeop_p) __P((void *));
    727 
    728 	if (vp->v_op == nfsv2_vnodeop_p)
    729 		vnd->sc_maxactive = 2;
    730 	else
    731 #endif
    732 		vnd->sc_maxactive = 8;
    733 
    734 	if (vnd->sc_maxactive < 1)
    735 		vnd->sc_maxactive = 1;
    736 }
    737 
    738 void
    739 vndshutdown()
    740 {
    741 	register struct vnd_softc *vnd;
    742 
    743 	for (vnd = &vnd_softc[0]; vnd < &vnd_softc[numvnd]; vnd++)
    744 		if (vnd->sc_flags & VNF_INITED)
    745 			vndclear(vnd);
    746 }
    747 
    748 void
    749 vndclear(vnd)
    750 	register struct vnd_softc *vnd;
    751 {
    752 	register struct vnode *vp = vnd->sc_vp;
    753 	struct proc *p = curproc;		/* XXX */
    754 
    755 #ifdef DEBUG
    756 	if (vnddebug & VDB_FOLLOW)
    757 		printf("vndclear(%p): vp %p\n", vnd, vp);
    758 #endif
    759 	vnd->sc_flags &= ~VNF_INITED;
    760 	if (vp == (struct vnode *)0)
    761 		panic("vndioctl: null vp");
    762 	(void) vn_close(vp, FREAD|FWRITE, vnd->sc_cred, p);
    763 	crfree(vnd->sc_cred);
    764 	vnd->sc_vp = (struct vnode *)0;
    765 	vnd->sc_cred = (struct ucred *)0;
    766 	vnd->sc_size = 0;
    767 }
    768 
    769 int
    770 vndsize(dev)
    771 	dev_t dev;
    772 {
    773 	int unit = vndunit(dev);
    774 	register struct vnd_softc *vnd = &vnd_softc[unit];
    775 
    776 	if (unit >= numvnd || (vnd->sc_flags & VNF_INITED) == 0)
    777 		return(-1);
    778 	return(vnd->sc_size);
    779 }
    780 
    781 int
    782 vnddump(dev, blkno, va, size)
    783 	dev_t dev;
    784 	daddr_t blkno;
    785 	caddr_t va;
    786 	size_t size;
    787 {
    788 
    789 	/* Not implemented. */
    790 	return ENXIO;
    791 }
    792 
    793 /*
    794  * Wait interruptibly for an exclusive lock.
    795  *
    796  * XXX
    797  * Several drivers do this; it should be abstracted and made MP-safe.
    798  */
    799 static int
    800 vndlock(sc)
    801 	struct vnd_softc *sc;
    802 {
    803 	int error;
    804 
    805 	while ((sc->sc_flags & VNF_LOCKED) != 0) {
    806 		sc->sc_flags |= VNF_WANTED;
    807 		if ((error = tsleep(sc, PRIBIO | PCATCH, "vndlck", 0)) != 0)
    808 			return (error);
    809 	}
    810 	sc->sc_flags |= VNF_LOCKED;
    811 	return (0);
    812 }
    813 
    814 /*
    815  * Unlock and wake up any waiters.
    816  */
    817 static void
    818 vndunlock(sc)
    819 	struct vnd_softc *sc;
    820 {
    821 
    822 	sc->sc_flags &= ~VNF_LOCKED;
    823 	if ((sc->sc_flags & VNF_WANTED) != 0) {
    824 		sc->sc_flags &= ~VNF_WANTED;
    825 		wakeup(sc);
    826 	}
    827 }
    828