Home | History | Annotate | Line # | Download | only in dev
vnd.c revision 1.5
      1 /*
      2  * Copyright (c) 1988 University of Utah.
      3  * Copyright (c) 1990, 1993
      4  *	The Regents of the University of California.  All rights reserved.
      5  *
      6  * This code is derived from software contributed to Berkeley by
      7  * the Systems Programming Group of the University of Utah Computer
      8  * Science Department.
      9  *
     10  * Redistribution and use in source and binary forms, with or without
     11  * modification, are permitted provided that the following conditions
     12  * are met:
     13  * 1. Redistributions of source code must retain the above copyright
     14  *    notice, this list of conditions and the following disclaimer.
     15  * 2. Redistributions in binary form must reproduce the above copyright
     16  *    notice, this list of conditions and the following disclaimer in the
     17  *    documentation and/or other materials provided with the distribution.
     18  * 3. All advertising materials mentioning features or use of this software
     19  *    must display the following acknowledgement:
     20  *	This product includes software developed by the University of
     21  *	California, Berkeley and its contributors.
     22  * 4. Neither the name of the University nor the names of its contributors
     23  *    may be used to endorse or promote products derived from this software
     24  *    without specific prior written permission.
     25  *
     26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     36  * SUCH DAMAGE.
     37  *
     38  * from: Utah $Hdr: vn.c 1.13 94/04/02$
     39  *
     40  *	@(#)vn.c	8.6 (Berkeley) 4/1/94
     41  */
     42 
     43 /*
     44  * Vnode disk driver.
     45  *
     46  * Block/character interface to a vnode.  Allows one to treat a file
     47  * as a disk (e.g. build a filesystem in it, mount it, etc.).
     48  *
     49  * NOTE 1: This uses the VOP_BMAP/VOP_STRATEGY interface to the vnode
     50  * instead of a simple VOP_RDWR.  We do this to avoid distorting the
     51  * local buffer cache.
     52  *
     53  * NOTE 2: There is a security issue involved with this driver.
     54  * Once mounted all access to the contents of the "mapped" file via
     55  * the special file is controlled by the permissions on the special
     56  * file, the protection of the mapped file is ignored (effectively,
     57  * by using root credentials in all transactions).
     58  *
     59  * NOTE 3: Doesn't interact with leases, should it?
     60  */
     61 #include "vn.h"
     62 #if NVN > 0
     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/dkstat.h>
     70 #include <sys/buf.h>
     71 #include <sys/malloc.h>
     72 #include <sys/ioctl.h>
     73 #include <sys/mount.h>
     74 #include <sys/vnode.h>
     75 #include <sys/file.h>
     76 #include <sys/uio.h>
     77 
     78 #include <miscfs/specfs/specdev.h>
     79 
     80 #include <dev/vnioctl.h>
     81 
     82 #ifdef DEBUG
     83 int dovncluster = 1;
     84 int vndebug = 0x00;
     85 #define VDB_FOLLOW	0x01
     86 #define VDB_INIT	0x02
     87 #define VDB_IO		0x04
     88 #endif
     89 
     90 #define b_cylin	b_resid
     91 
     92 #define	vnunit(x)	((minor(x) >> 3) & 0x7)	/* for consistency */
     93 
     94 #define	getvnbuf()	\
     95 	((struct buf *)malloc(sizeof(struct buf), M_DEVBUF, M_WAITOK))
     96 #define putvnbuf(bp)	\
     97 	free((caddr_t)(bp), M_DEVBUF)
     98 
     99 struct vn_softc {
    100 	int		 sc_flags;	/* flags */
    101 	size_t		 sc_size;	/* size of vn */
    102 	struct vnode	*sc_vp;		/* vnode */
    103 	struct ucred	*sc_cred;	/* credentials */
    104 	int		 sc_maxactive;	/* max # of active requests */
    105 	struct buf	 sc_tab;	/* transfer queue */
    106 };
    107 
    108 /* sc_flags */
    109 #define	VNF_ALIVE	0x01
    110 #define VNF_INITED	0x02
    111 
    112 #if 0	/* if you need static allocation */
    113 struct vn_softc vn_softc[NVN];
    114 int numvnd = NVN;
    115 #else
    116 struct vn_softc *vn_softc;
    117 int numvnd;
    118 #endif
    119 
    120 void
    121 vnattach(num)
    122 	int num;
    123 {
    124 	char *mem;
    125 	register u_long size;
    126 
    127 	if (num <= 0)
    128 		return;
    129 	size = num * sizeof(struct vn_softc);
    130 	mem = malloc(size, M_DEVBUF, M_NOWAIT);
    131 	if (mem == NULL) {
    132 		printf("WARNING: no memory for vnode disks\n");
    133 		return;
    134 	}
    135 	bzero(mem, size);
    136 	vn_softc = (struct vn_softc *)mem;
    137 	numvnd = num;
    138 }
    139 
    140 int
    141 vnopen(dev, flags, mode, p)
    142 	dev_t dev;
    143 	int flags, mode;
    144 	struct proc *p;
    145 {
    146 	int unit = vnunit(dev);
    147 
    148 #ifdef DEBUG
    149 	if (vndebug & VDB_FOLLOW)
    150 		printf("vnopen(%x, %x, %x, %x)\n", dev, flags, mode, p);
    151 #endif
    152 	if (unit >= numvnd)
    153 		return(ENXIO);
    154 	return(0);
    155 }
    156 
    157 int
    158 vnclose(dev, flags, mode, p)
    159 	dev_t dev;
    160 	int flags, mode;
    161 	struct proc *p;
    162 {
    163 #ifdef DEBUG
    164 	if (vndebug & VDB_FOLLOW)
    165 		printf("vnclose(%x, %x, %x, %x)\n", dev, flags, mode, p);
    166 #endif
    167 	return 0;
    168 }
    169 
    170 /*
    171  * Break the request into bsize pieces and submit using VOP_BMAP/VOP_STRATEGY.
    172  * Note that this driver can only be used for swapping over NFS on the hp
    173  * since nfs_strategy on the vax cannot handle u-areas and page tables.
    174  */
    175 vnstrategy(bp)
    176 	register struct buf *bp;
    177 {
    178 	int unit = vnunit(bp->b_dev);
    179 	register struct vn_softc *vn = &vn_softc[unit];
    180 	register struct buf *nbp;
    181 	register int bn, bsize, resid;
    182 	register caddr_t addr;
    183 	int sz, flags, error;
    184 	extern void vniodone();
    185 
    186 #ifdef DEBUG
    187 	if (vndebug & VDB_FOLLOW)
    188 		printf("vnstrategy(%x): unit %d\n", bp, unit);
    189 #endif
    190 	if ((vn->sc_flags & VNF_INITED) == 0) {
    191 		bp->b_error = ENXIO;
    192 		bp->b_flags |= B_ERROR;
    193 		biodone(bp);
    194 		return;
    195 	}
    196 	bn = bp->b_blkno;
    197 	sz = howmany(bp->b_bcount, DEV_BSIZE);
    198 	bp->b_resid = bp->b_bcount;
    199 	if (bn < 0 || bn + sz > vn->sc_size) {
    200 		if (bn != vn->sc_size) {
    201 			bp->b_error = EINVAL;
    202 			bp->b_flags |= B_ERROR;
    203 		}
    204 		biodone(bp);
    205 		return;
    206 	}
    207 	bn = dbtob(bn);
    208  	bsize = vn->sc_vp->v_mount->mnt_stat.f_iosize;
    209 	addr = bp->b_data;
    210 	flags = bp->b_flags | B_CALL;
    211 	for (resid = bp->b_resid; resid; resid -= sz) {
    212 		struct vnode *vp;
    213 		daddr_t nbn;
    214 		int off, s, nra;
    215 
    216 		nra = 0;
    217 #if (BSD > 199103)
    218 		error = VOP_BMAP(vn->sc_vp, bn / bsize, &vp, &nbn, &nra);
    219 #else
    220 		error = VOP_BMAP(vn->sc_vp, bn / bsize, &vp, &nbn);
    221 #endif
    222 		if (error == 0 && (long)nbn == -1)
    223 			error = EIO;
    224 #ifdef DEBUG
    225 		if (!dovncluster)
    226 			nra = 0;
    227 #endif
    228 
    229 		if (off = bn % bsize)
    230 			sz = bsize - off;
    231 		else
    232 			sz = (1 + nra) * bsize;
    233 		if (resid < sz)
    234 			sz = resid;
    235 #ifdef DEBUG
    236 		if (vndebug & VDB_IO)
    237 			printf("vnstrategy: vp %x/%x bn %x/%x sz %x\n",
    238 			       vn->sc_vp, vp, bn, nbn, sz);
    239 #endif
    240 
    241 		nbp = getvnbuf();
    242 		nbp->b_flags = flags;
    243 		nbp->b_bcount = sz;
    244 		nbp->b_bufsize = bp->b_bufsize;
    245 		nbp->b_error = 0;
    246 		if (vp->v_type == VBLK || vp->v_type == VCHR)
    247 			nbp->b_dev = vp->v_rdev;
    248 		else
    249 			nbp->b_dev = NODEV;
    250 		nbp->b_data = addr;
    251 		nbp->b_blkno = nbn + btodb(off);
    252 		nbp->b_proc = bp->b_proc;
    253 		nbp->b_iodone = vniodone;
    254 		nbp->b_vp = vp;
    255 		nbp->b_pfcent = (int) bp;	/* XXX */
    256 		nbp->b_rcred = vn->sc_cred;	/* XXX crdup? */
    257 		nbp->b_wcred = vn->sc_cred;	/* XXX crdup? */
    258 		nbp->b_dirtyoff = bp->b_dirtyoff;
    259 		nbp->b_dirtyend = bp->b_dirtyend;
    260 		nbp->b_validoff = bp->b_validoff;
    261 		nbp->b_validend = bp->b_validend;
    262 		/*
    263 		 * If there was an error or a hole in the file...punt.
    264 		 * Note that we deal with this after the nbp allocation.
    265 		 * This ensures that we properly clean up any operations
    266 		 * that we have already fired off.
    267 		 *
    268 		 * XXX we could deal with holes here but it would be
    269 		 * a hassle (in the write case).
    270 		 */
    271 		if (error) {
    272 			nbp->b_error = error;
    273 			nbp->b_flags |= B_ERROR;
    274 			bp->b_resid -= (resid - sz);
    275 			biodone(nbp);
    276 			return;
    277 		}
    278 		/*
    279 		 * Just sort by block number
    280 		 */
    281 		nbp->b_cylin = nbp->b_blkno;
    282 		s = splbio();
    283 		disksort(&vn->sc_tab, nbp);
    284 		if (vn->sc_tab.b_active < vn->sc_maxactive) {
    285 			vn->sc_tab.b_active++;
    286 			vnstart(vn);
    287 		}
    288 		splx(s);
    289 		bn += sz;
    290 		addr += sz;
    291 	}
    292 }
    293 
    294 /*
    295  * Feed requests sequentially.
    296  * We do it this way to keep from flooding NFS servers if we are connected
    297  * to an NFS file.  This places the burden on the client rather than the
    298  * server.
    299  */
    300 vnstart(vn)
    301 	register struct vn_softc *vn;
    302 {
    303 	register struct buf *bp;
    304 
    305 	/*
    306 	 * Dequeue now since lower level strategy routine might
    307 	 * queue using same links
    308 	 */
    309 	bp = vn->sc_tab.b_actf;
    310 	vn->sc_tab.b_actf = bp->b_actf;
    311 #ifdef DEBUG
    312 	if (vndebug & VDB_IO)
    313 		printf("vnstart(%d): bp %x vp %x blkno %x addr %x cnt %x\n",
    314 		    vn-vn_softc, bp, bp->b_vp, bp->b_blkno, bp->b_data,
    315 		    bp->b_bcount);
    316 #endif
    317 	if ((bp->b_flags & B_READ) == 0)
    318 		bp->b_vp->v_numoutput++;
    319 	VOP_STRATEGY(bp);
    320 }
    321 
    322 void
    323 vniodone(bp)
    324 	register struct buf *bp;
    325 {
    326 	register struct buf *pbp = (struct buf *)bp->b_pfcent;	/* XXX */
    327 	register struct vn_softc *vn = &vn_softc[vnunit(pbp->b_dev)];
    328 	int s;
    329 
    330 	s = splbio();
    331 #ifdef DEBUG
    332 	if (vndebug & VDB_IO)
    333 		printf("vniodone(%d): bp %x vp %x blkno %x addr %x cnt %x\n",
    334 		    vn-vn_softc, bp, bp->b_vp, bp->b_blkno, bp->b_data,
    335 		    bp->b_bcount);
    336 #endif
    337 	if (bp->b_error) {
    338 #ifdef DEBUG
    339 		if (vndebug & VDB_IO)
    340 			printf("vniodone: bp %x error %d\n", bp, bp->b_error);
    341 #endif
    342 		pbp->b_flags |= B_ERROR;
    343 		pbp->b_error = biowait(bp);
    344 	}
    345 	pbp->b_resid -= bp->b_bcount;
    346 	putvnbuf(bp);
    347 	if (pbp->b_resid == 0) {
    348 #ifdef DEBUG
    349 		if (vndebug & VDB_IO)
    350 			printf("vniodone: pbp %x iodone\n", pbp);
    351 #endif
    352 		biodone(pbp);
    353 	}
    354 	if (vn->sc_tab.b_actf)
    355 		vnstart(vn);
    356 	else
    357 		vn->sc_tab.b_active--;
    358 	splx(s);
    359 }
    360 
    361 vnread(dev, uio, flags, p)
    362 	dev_t dev;
    363 	struct uio *uio;
    364 	int flags;
    365 	struct proc *p;
    366 {
    367 
    368 #ifdef DEBUG
    369 	if (vndebug & VDB_FOLLOW)
    370 		printf("vnread(%x, %x, %x, %x)\n", dev, uio, flags, p);
    371 #endif
    372 	return(physio(vnstrategy, NULL, dev, B_READ, minphys, uio));
    373 }
    374 
    375 vnwrite(dev, uio, flags, p)
    376 	dev_t dev;
    377 	struct uio *uio;
    378 	int flags;
    379 	struct proc *p;
    380 {
    381 
    382 #ifdef DEBUG
    383 	if (vndebug & VDB_FOLLOW)
    384 		printf("vnwrite(%x, %x, %x, %x)\n", dev, uio, flags, p);
    385 #endif
    386 	return(physio(vnstrategy, NULL, dev, B_WRITE, minphys, uio));
    387 }
    388 
    389 /* ARGSUSED */
    390 vnioctl(dev, cmd, data, flag, p)
    391 	dev_t dev;
    392 	u_long cmd;
    393 	caddr_t data;
    394 	int flag;
    395 	struct proc *p;
    396 {
    397 	int unit = vnunit(dev);
    398 	register struct vn_softc *vn;
    399 	struct vn_ioctl *vio;
    400 	struct vattr vattr;
    401 	struct nameidata nd;
    402 	int error;
    403 
    404 #ifdef DEBUG
    405 	if (vndebug & VDB_FOLLOW)
    406 		printf("vnioctl(%x, %x, %x, %x, %x): unit %d\n",
    407 		    dev, cmd, data, flag, p, unit);
    408 #endif
    409 	error = suser(p->p_ucred, &p->p_acflag);
    410 	if (error)
    411 		return (error);
    412 	if (unit >= numvnd)
    413 		return (ENXIO);
    414 
    415 	vn = &vn_softc[unit];
    416 	vio = (struct vn_ioctl *)data;
    417 	switch (cmd) {
    418 
    419 	case VNIOCSET:
    420 		if (vn->sc_flags & VNF_INITED)
    421 			return(EBUSY);
    422 		/*
    423 		 * Always open for read and write.
    424 		 * This is probably bogus, but it lets vn_open()
    425 		 * weed out directories, sockets, etc. so we don't
    426 		 * have to worry about them.
    427 		 */
    428 #if (BSD > 199103)
    429 		NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, vio->vn_file, p);
    430 		if (error = vn_open(&nd, FREAD|FWRITE, 0))
    431 			return(error);
    432 #else
    433 		nd.ni_nameiop = LOOKUP | FOLLOW;
    434 		nd.ni_segflg = UIO_USERSPACE;
    435 		nd.ni_dirp = vio->vn_file;
    436 		if (error = vn_open(&nd, p, FREAD|FWRITE, 0))
    437 			return(error);
    438 #endif
    439 		if (error = VOP_GETATTR(nd.ni_vp, &vattr, p->p_ucred, p)) {
    440 			VOP_UNLOCK(nd.ni_vp);
    441 			(void) vn_close(nd.ni_vp, FREAD|FWRITE, p->p_ucred, p);
    442 			return(error);
    443 		}
    444 		VOP_UNLOCK(nd.ni_vp);
    445 		vn->sc_vp = nd.ni_vp;
    446 		vn->sc_size = btodb(vattr.va_size);	/* note truncation */
    447 		if (error = vnsetcred(vn, p->p_ucred)) {
    448 			(void) vn_close(nd.ni_vp, FREAD|FWRITE, p->p_ucred, p);
    449 			return(error);
    450 		}
    451 		vnthrottle(vn, vn->sc_vp);
    452 		vio->vn_size = dbtob(vn->sc_size);
    453 		vn->sc_flags |= VNF_INITED;
    454 #ifdef DEBUG
    455 		if (vndebug & VDB_INIT)
    456 			printf("vnioctl: SET vp %x size %x\n",
    457 			    vn->sc_vp, vn->sc_size);
    458 #endif
    459 		break;
    460 
    461 	case VNIOCCLR:
    462 		if ((vn->sc_flags & VNF_INITED) == 0)
    463 			return(ENXIO);
    464 		vnclear(vn);
    465 #ifdef DEBUG
    466 		if (vndebug & VDB_INIT)
    467 			printf("vnioctl: CLRed\n");
    468 #endif
    469 		break;
    470 
    471 	default:
    472 		return(ENXIO);
    473 	}
    474 	return(0);
    475 }
    476 
    477 /*
    478  * Duplicate the current processes' credentials.  Since we are called only
    479  * as the result of a SET ioctl and only root can do that, any future access
    480  * to this "disk" is essentially as root.  Note that credentials may change
    481  * if some other uid can write directly to the mapped file (NFS).
    482  */
    483 vnsetcred(vn, cred)
    484 	register struct vn_softc *vn;
    485 	struct ucred *cred;
    486 {
    487 	struct uio auio;
    488 	struct iovec aiov;
    489 	char *tmpbuf;
    490 	int error;
    491 
    492 	vn->sc_cred = crdup(cred);
    493 	tmpbuf = malloc(DEV_BSIZE, M_TEMP, M_WAITOK);
    494 
    495 	/* XXX: Horrible kludge to establish credentials for NFS */
    496 	aiov.iov_base = tmpbuf;
    497 	aiov.iov_len = min(DEV_BSIZE, dbtob(vn->sc_size));
    498 	auio.uio_iov = &aiov;
    499 	auio.uio_iovcnt = 1;
    500 	auio.uio_offset = 0;
    501 	auio.uio_rw = UIO_READ;
    502 	auio.uio_segflg = UIO_SYSSPACE;
    503 	auio.uio_resid = aiov.iov_len;
    504 	error = VOP_READ(vn->sc_vp, &auio, 0, vn->sc_cred);
    505 
    506 	free(tmpbuf, M_TEMP);
    507 	return (error);
    508 }
    509 
    510 /*
    511  * Set maxactive based on FS type
    512  */
    513 vnthrottle(vn, vp)
    514 	register struct vn_softc *vn;
    515 	struct vnode *vp;
    516 {
    517 #ifdef NFSCLIENT
    518 #if (BSD > 199103)
    519 	extern int (**nfsv2_vnodeop_p)();
    520 
    521 	if (vp->v_op == nfsv2_vnodeop_p)
    522 #else
    523  	extern struct vnodeops nfsv2_vnodeops;
    524 
    525 	if (vp->v_op == &nfsv2_vnodeops)
    526 #endif
    527 		vn->sc_maxactive = 2;
    528 	else
    529 #endif
    530 		vn->sc_maxactive = 8;
    531 
    532 	if (vn->sc_maxactive < 1)
    533 		vn->sc_maxactive = 1;
    534 }
    535 
    536 vnshutdown()
    537 {
    538 	register struct vn_softc *vn;
    539 
    540 	for (vn = &vn_softc[0]; vn < &vn_softc[numvnd]; vn++)
    541 		if (vn->sc_flags & VNF_INITED)
    542 			vnclear(vn);
    543 }
    544 
    545 vnclear(vn)
    546 	register struct vn_softc *vn;
    547 {
    548 	register struct vnode *vp = vn->sc_vp;
    549 	struct proc *p = curproc;		/* XXX */
    550 
    551 #ifdef DEBUG
    552 	if (vndebug & VDB_FOLLOW)
    553 		printf("vnclear(%x): vp %x\n", vp);
    554 #endif
    555 	vn->sc_flags &= ~VNF_INITED;
    556 	if (vp == (struct vnode *)0)
    557 		panic("vnioctl: null vp");
    558 	(void) vn_close(vp, FREAD|FWRITE, vn->sc_cred, p);
    559 	crfree(vn->sc_cred);
    560 	vn->sc_vp = (struct vnode *)0;
    561 	vn->sc_cred = (struct ucred *)0;
    562 	vn->sc_size = 0;
    563 }
    564 
    565 vnsize(dev)
    566 	dev_t dev;
    567 {
    568 	int unit = vnunit(dev);
    569 	register struct vn_softc *vn = &vn_softc[unit];
    570 
    571 	if (unit >= numvnd || (vn->sc_flags & VNF_INITED) == 0)
    572 		return(-1);
    573 	return(vn->sc_size);
    574 }
    575 
    576 vndump(dev)
    577 {
    578 	return(ENXIO);
    579 }
    580 #endif
    581