Home | History | Annotate | Line # | Download | only in dev
vnd.c revision 1.16
      1 /*	$NetBSD: vnd.c,v 1.16 1994/12/24 14:05:51 cgd 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 #include "vn.h"
     64 #if NVN > 0
     65 
     66 #include <sys/param.h>
     67 #include <sys/systm.h>
     68 #include <sys/namei.h>
     69 #include <sys/proc.h>
     70 #include <sys/errno.h>
     71 #include <sys/dkstat.h>
     72 #include <sys/buf.h>
     73 #include <sys/malloc.h>
     74 #include <sys/ioctl.h>
     75 #include <sys/disklabel.h>
     76 #include <sys/mount.h>
     77 #include <sys/vnode.h>
     78 #include <sys/file.h>
     79 #include <sys/uio.h>
     80 
     81 #include <miscfs/specfs/specdev.h>
     82 
     83 #include <dev/vnioctl.h>
     84 
     85 #ifdef DEBUG
     86 int dovncluster = 1;
     87 int vndebug = 0x00;
     88 #define VDB_FOLLOW	0x01
     89 #define VDB_INIT	0x02
     90 #define VDB_IO		0x04
     91 #endif
     92 
     93 #define b_cylin	b_resid
     94 
     95 #define	vnunit(x)	((minor(x) >> 3) & 0x7)	/* for consistency */
     96 
     97 #define	getvnbuf()	\
     98 	((struct buf *)malloc(sizeof(struct buf), M_DEVBUF, M_WAITOK))
     99 #define putvnbuf(bp)	\
    100 	free((caddr_t)(bp), M_DEVBUF)
    101 
    102 struct vn_softc {
    103 	int		 sc_flags;	/* flags */
    104 	size_t		 sc_size;	/* size of vn */
    105 	struct vnode	*sc_vp;		/* vnode */
    106 	struct ucred	*sc_cred;	/* credentials */
    107 	int		 sc_maxactive;	/* max # of active requests */
    108 	struct buf	 sc_tab;	/* transfer queue */
    109 };
    110 
    111 /* sc_flags */
    112 #define	VNF_ALIVE	0x01
    113 #define VNF_INITED	0x02
    114 
    115 #if 0	/* if you need static allocation */
    116 struct vn_softc vn_softc[NVN];
    117 int numvnd = NVN;
    118 #else
    119 struct vn_softc *vn_softc;
    120 int numvnd;
    121 #endif
    122 
    123 void	vnclear __P((struct vn_softc *));
    124 void	vnstart __P((struct vn_softc *));
    125 int	vnsetcred __P((struct vn_softc *, struct ucred *));
    126 void	vnthrottle __P((struct vn_softc *, struct vnode *));
    127 
    128 void
    129 vnattach(num)
    130 	int num;
    131 {
    132 	char *mem;
    133 	register u_long size;
    134 
    135 	if (num <= 0)
    136 		return;
    137 	size = num * sizeof(struct vn_softc);
    138 	mem = malloc(size, M_DEVBUF, M_NOWAIT);
    139 	if (mem == NULL) {
    140 		printf("WARNING: no memory for vnode disks\n");
    141 		return;
    142 	}
    143 	bzero(mem, size);
    144 	vn_softc = (struct vn_softc *)mem;
    145 	numvnd = num;
    146 }
    147 
    148 int
    149 vnopen(dev, flags, mode, p)
    150 	dev_t dev;
    151 	int flags, mode;
    152 	struct proc *p;
    153 {
    154 	int unit = vnunit(dev);
    155 
    156 #ifdef DEBUG
    157 	if (vndebug & VDB_FOLLOW)
    158 		printf("vnopen(%x, %x, %x, %x)\n", dev, flags, mode, p);
    159 #endif
    160 	if (unit >= numvnd)
    161 		return(ENXIO);
    162 	return(0);
    163 }
    164 
    165 int
    166 vnclose(dev, flags, mode, p)
    167 	dev_t dev;
    168 	int flags, mode;
    169 	struct proc *p;
    170 {
    171 #ifdef DEBUG
    172 	if (vndebug & VDB_FOLLOW)
    173 		printf("vnclose(%x, %x, %x, %x)\n", dev, flags, mode, p);
    174 #endif
    175 	return 0;
    176 }
    177 
    178 /*
    179  * Break the request into bsize pieces and submit using VOP_BMAP/VOP_STRATEGY.
    180  * Note that this driver can only be used for swapping over NFS on the hp
    181  * since nfs_strategy on the vax cannot handle u-areas and page tables.
    182  */
    183 void
    184 vnstrategy(bp)
    185 	register struct buf *bp;
    186 {
    187 	int unit = vnunit(bp->b_dev);
    188 	register struct vn_softc *vn = &vn_softc[unit];
    189 	register struct buf *nbp;
    190 	register int bn, bsize, resid;
    191 	register caddr_t addr;
    192 	int sz, flags, error;
    193 	extern void vniodone();
    194 
    195 #ifdef DEBUG
    196 	if (vndebug & VDB_FOLLOW)
    197 		printf("vnstrategy(%x): unit %d\n", bp, unit);
    198 #endif
    199 	if ((vn->sc_flags & VNF_INITED) == 0) {
    200 		bp->b_error = ENXIO;
    201 		bp->b_flags |= B_ERROR;
    202 		biodone(bp);
    203 		return;
    204 	}
    205 	bn = bp->b_blkno;
    206 	sz = howmany(bp->b_bcount, DEV_BSIZE);
    207 	bp->b_resid = bp->b_bcount;
    208 	if (bn < 0 || bn + sz > vn->sc_size) {
    209 		if (bn != vn->sc_size) {
    210 			bp->b_error = EINVAL;
    211 			bp->b_flags |= B_ERROR;
    212 		}
    213 		biodone(bp);
    214 		return;
    215 	}
    216 	bn = dbtob(bn);
    217  	bsize = vn->sc_vp->v_mount->mnt_stat.f_iosize;
    218 	addr = bp->b_data;
    219 	flags = bp->b_flags | B_CALL;
    220 	for (resid = bp->b_resid; resid; resid -= sz) {
    221 		struct vnode *vp;
    222 		daddr_t nbn;
    223 		int off, s, nra;
    224 
    225 		nra = 0;
    226 		error = VOP_BMAP(vn->sc_vp, bn / bsize, &vp, &nbn, &nra);
    227 		if (error == 0 && (long)nbn == -1)
    228 			error = EIO;
    229 #ifdef DEBUG
    230 		if (!dovncluster)
    231 			nra = 0;
    232 #endif
    233 
    234 		if (off = bn % bsize)
    235 			sz = bsize - off;
    236 		else
    237 			sz = (1 + nra) * bsize;
    238 		if (resid < sz)
    239 			sz = resid;
    240 #ifdef DEBUG
    241 		if (vndebug & VDB_IO)
    242 			printf("vnstrategy: vp %x/%x bn %x/%x sz %x\n",
    243 			       vn->sc_vp, vp, bn, nbn, sz);
    244 #endif
    245 
    246 		nbp = getvnbuf();
    247 		nbp->b_flags = flags;
    248 		nbp->b_bcount = sz;
    249 		nbp->b_bufsize = bp->b_bufsize;
    250 		nbp->b_error = 0;
    251 		if (vp->v_type == VBLK || vp->v_type == VCHR)
    252 			nbp->b_dev = vp->v_rdev;
    253 		else
    254 			nbp->b_dev = NODEV;
    255 		nbp->b_data = addr;
    256 		nbp->b_blkno = nbn + btodb(off);
    257 		nbp->b_proc = bp->b_proc;
    258 		nbp->b_iodone = vniodone;
    259 		nbp->b_vp = vp;
    260 		nbp->b_pfcent = (int) bp;	/* XXX */
    261 		nbp->b_rcred = vn->sc_cred;	/* XXX crdup? */
    262 		nbp->b_wcred = vn->sc_cred;	/* XXX crdup? */
    263 		nbp->b_dirtyoff = bp->b_dirtyoff;
    264 		nbp->b_dirtyend = bp->b_dirtyend;
    265 		nbp->b_validoff = bp->b_validoff;
    266 		nbp->b_validend = bp->b_validend;
    267 		/*
    268 		 * If there was an error or a hole in the file...punt.
    269 		 * Note that we deal with this after the nbp allocation.
    270 		 * This ensures that we properly clean up any operations
    271 		 * that we have already fired off.
    272 		 *
    273 		 * XXX we could deal with holes here but it would be
    274 		 * a hassle (in the write case).
    275 		 */
    276 		if (error) {
    277 			nbp->b_error = error;
    278 			nbp->b_flags |= B_ERROR;
    279 			bp->b_resid -= (resid - sz);
    280 			biodone(nbp);
    281 			return;
    282 		}
    283 		/*
    284 		 * Just sort by block number
    285 		 */
    286 		nbp->b_cylin = nbp->b_blkno;
    287 		s = splbio();
    288 		disksort(&vn->sc_tab, nbp);
    289 		if (vn->sc_tab.b_active < vn->sc_maxactive) {
    290 			vn->sc_tab.b_active++;
    291 			vnstart(vn);
    292 		}
    293 		splx(s);
    294 		bn += sz;
    295 		addr += sz;
    296 	}
    297 }
    298 
    299 /*
    300  * Feed requests sequentially.
    301  * We do it this way to keep from flooding NFS servers if we are connected
    302  * to an NFS file.  This places the burden on the client rather than the
    303  * server.
    304  */
    305 void
    306 vnstart(vn)
    307 	register struct vn_softc *vn;
    308 {
    309 	register struct buf *bp;
    310 
    311 	/*
    312 	 * Dequeue now since lower level strategy routine might
    313 	 * queue using same links
    314 	 */
    315 	bp = vn->sc_tab.b_actf;
    316 	vn->sc_tab.b_actf = bp->b_actf;
    317 #ifdef DEBUG
    318 	if (vndebug & VDB_IO)
    319 		printf("vnstart(%d): bp %x vp %x blkno %x addr %x cnt %x\n",
    320 		    vn-vn_softc, bp, bp->b_vp, bp->b_blkno, bp->b_data,
    321 		    bp->b_bcount);
    322 #endif
    323 	if ((bp->b_flags & B_READ) == 0)
    324 		bp->b_vp->v_numoutput++;
    325 	VOP_STRATEGY(bp);
    326 }
    327 
    328 void
    329 vniodone(bp)
    330 	register struct buf *bp;
    331 {
    332 	register struct buf *pbp = (struct buf *)bp->b_pfcent;	/* XXX */
    333 	register struct vn_softc *vn = &vn_softc[vnunit(pbp->b_dev)];
    334 	int s;
    335 
    336 	s = splbio();
    337 #ifdef DEBUG
    338 	if (vndebug & VDB_IO)
    339 		printf("vniodone(%d): bp %x vp %x blkno %x addr %x cnt %x\n",
    340 		    vn-vn_softc, bp, bp->b_vp, bp->b_blkno, bp->b_data,
    341 		    bp->b_bcount);
    342 #endif
    343 	if (bp->b_error) {
    344 #ifdef DEBUG
    345 		if (vndebug & VDB_IO)
    346 			printf("vniodone: bp %x error %d\n", bp, bp->b_error);
    347 #endif
    348 		pbp->b_flags |= B_ERROR;
    349 		pbp->b_error = biowait(bp);
    350 	}
    351 	pbp->b_resid -= bp->b_bcount;
    352 	putvnbuf(bp);
    353 	if (pbp->b_resid == 0) {
    354 #ifdef DEBUG
    355 		if (vndebug & VDB_IO)
    356 			printf("vniodone: pbp %x iodone\n", pbp);
    357 #endif
    358 		biodone(pbp);
    359 	}
    360 	if (vn->sc_tab.b_actf)
    361 		vnstart(vn);
    362 	else
    363 		vn->sc_tab.b_active--;
    364 	splx(s);
    365 }
    366 
    367 /* ARGSUSED */
    368 int
    369 vnioctl(dev, cmd, data, flag, p)
    370 	dev_t dev;
    371 	u_long cmd;
    372 	caddr_t data;
    373 	int flag;
    374 	struct proc *p;
    375 {
    376 	int unit = vnunit(dev);
    377 	register struct vn_softc *vn;
    378 	struct vn_ioctl *vio;
    379 	struct vattr vattr;
    380 	struct nameidata nd;
    381 	int error;
    382 
    383 #ifdef DEBUG
    384 	if (vndebug & VDB_FOLLOW)
    385 		printf("vnioctl(%x, %lx, %x, %x, %x): unit %d\n",
    386 		    dev, cmd, data, flag, p, unit);
    387 #endif
    388 	error = suser(p->p_ucred, &p->p_acflag);
    389 	if (error)
    390 		return (error);
    391 	if (unit >= numvnd)
    392 		return (ENXIO);
    393 
    394 	vn = &vn_softc[unit];
    395 	vio = (struct vn_ioctl *)data;
    396 	switch (cmd) {
    397 
    398 	case VNIOCSET:
    399 		if (vn->sc_flags & VNF_INITED)
    400 			return(EBUSY);
    401 		/*
    402 		 * Always open for read and write.
    403 		 * This is probably bogus, but it lets vn_open()
    404 		 * weed out directories, sockets, etc. so we don't
    405 		 * have to worry about them.
    406 		 */
    407 		NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, vio->vn_file, p);
    408 		if (error = vn_open(&nd, FREAD|FWRITE, 0))
    409 			return(error);
    410 		if (error = VOP_GETATTR(nd.ni_vp, &vattr, p->p_ucred, p)) {
    411 			VOP_UNLOCK(nd.ni_vp);
    412 			(void) vn_close(nd.ni_vp, FREAD|FWRITE, p->p_ucred, p);
    413 			return(error);
    414 		}
    415 		VOP_UNLOCK(nd.ni_vp);
    416 		vn->sc_vp = nd.ni_vp;
    417 		vn->sc_size = btodb(vattr.va_size);	/* note truncation */
    418 		if (error = vnsetcred(vn, p->p_ucred)) {
    419 			(void) vn_close(nd.ni_vp, FREAD|FWRITE, p->p_ucred, p);
    420 			return(error);
    421 		}
    422 		vnthrottle(vn, vn->sc_vp);
    423 		vio->vn_size = dbtob(vn->sc_size);
    424 		vn->sc_flags |= VNF_INITED;
    425 #ifdef DEBUG
    426 		if (vndebug & VDB_INIT)
    427 			printf("vnioctl: SET vp %x size %x\n",
    428 			    vn->sc_vp, vn->sc_size);
    429 #endif
    430 		break;
    431 
    432 	case VNIOCCLR:
    433 		if ((vn->sc_flags & VNF_INITED) == 0)
    434 			return(ENXIO);
    435 		vnclear(vn);
    436 #ifdef DEBUG
    437 		if (vndebug & VDB_INIT)
    438 			printf("vnioctl: CLRed\n");
    439 #endif
    440 		break;
    441 
    442 	default:
    443 		return(ENOTTY);
    444 	}
    445 	return(0);
    446 }
    447 
    448 /*
    449  * Duplicate the current processes' credentials.  Since we are called only
    450  * as the result of a SET ioctl and only root can do that, any future access
    451  * to this "disk" is essentially as root.  Note that credentials may change
    452  * if some other uid can write directly to the mapped file (NFS).
    453  */
    454 int
    455 vnsetcred(vn, cred)
    456 	register struct vn_softc *vn;
    457 	struct ucred *cred;
    458 {
    459 	struct uio auio;
    460 	struct iovec aiov;
    461 	char *tmpbuf;
    462 	int error;
    463 
    464 	vn->sc_cred = crdup(cred);
    465 	tmpbuf = malloc(DEV_BSIZE, M_TEMP, M_WAITOK);
    466 
    467 	/* XXX: Horrible kludge to establish credentials for NFS */
    468 	aiov.iov_base = tmpbuf;
    469 	aiov.iov_len = min(DEV_BSIZE, dbtob(vn->sc_size));
    470 	auio.uio_iov = &aiov;
    471 	auio.uio_iovcnt = 1;
    472 	auio.uio_offset = 0;
    473 	auio.uio_rw = UIO_READ;
    474 	auio.uio_segflg = UIO_SYSSPACE;
    475 	auio.uio_resid = aiov.iov_len;
    476 	error = VOP_READ(vn->sc_vp, &auio, 0, vn->sc_cred);
    477 
    478 	free(tmpbuf, M_TEMP);
    479 	return (error);
    480 }
    481 
    482 /*
    483  * Set maxactive based on FS type
    484  */
    485 void
    486 vnthrottle(vn, vp)
    487 	register struct vn_softc *vn;
    488 	struct vnode *vp;
    489 {
    490 #ifdef NFSCLIENT
    491 	extern int (**nfsv2_vnodeop_p)();
    492 
    493 	if (vp->v_op == nfsv2_vnodeop_p)
    494 		vn->sc_maxactive = 2;
    495 	else
    496 #endif
    497 		vn->sc_maxactive = 8;
    498 
    499 	if (vn->sc_maxactive < 1)
    500 		vn->sc_maxactive = 1;
    501 }
    502 
    503 void
    504 vnshutdown()
    505 {
    506 	register struct vn_softc *vn;
    507 
    508 	for (vn = &vn_softc[0]; vn < &vn_softc[numvnd]; vn++)
    509 		if (vn->sc_flags & VNF_INITED)
    510 			vnclear(vn);
    511 }
    512 
    513 void
    514 vnclear(vn)
    515 	register struct vn_softc *vn;
    516 {
    517 	register struct vnode *vp = vn->sc_vp;
    518 	struct proc *p = curproc;		/* XXX */
    519 
    520 #ifdef DEBUG
    521 	if (vndebug & VDB_FOLLOW)
    522 		printf("vnclear(%x): vp %x\n", vp);
    523 #endif
    524 	vn->sc_flags &= ~VNF_INITED;
    525 	if (vp == (struct vnode *)0)
    526 		panic("vnioctl: null vp");
    527 	(void) vn_close(vp, FREAD|FWRITE, vn->sc_cred, p);
    528 	crfree(vn->sc_cred);
    529 	vn->sc_vp = (struct vnode *)0;
    530 	vn->sc_cred = (struct ucred *)0;
    531 	vn->sc_size = 0;
    532 }
    533 
    534 int
    535 vnsize(dev)
    536 	dev_t dev;
    537 {
    538 	int unit = vnunit(dev);
    539 	register struct vn_softc *vn = &vn_softc[unit];
    540 
    541 	if (unit >= numvnd || (vn->sc_flags & VNF_INITED) == 0)
    542 		return(-1);
    543 	return(vn->sc_size);
    544 }
    545 
    546 int
    547 vndump(dev)
    548 	dev_t dev;
    549 {
    550 
    551 	return(ENXIO);
    552 }
    553 #endif
    554