xref: /netbsd-src/sys/dev/vnd.c (revision 3b01aba77a7a698587faaae455bbfe740923c1f5)
1 /*	$NetBSD: vnd.c,v 1.72 2001/07/07 17:04:02 thorpej Exp $	*/
2 
3 /*-
4  * Copyright (c) 1996, 1997, 1998 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Jason R. Thorpe.
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 NetBSD
21  *	Foundation, Inc. and its contributors.
22  * 4. Neither the name of The NetBSD Foundation nor the names of its
23  *    contributors may be used to endorse or promote products derived
24  *    from this software without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
27  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
28  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
29  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
30  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36  * POSSIBILITY OF SUCH DAMAGE.
37  */
38 
39 /*
40  * Copyright (c) 1988 University of Utah.
41  * Copyright (c) 1990, 1993
42  *	The Regents of the University of California.  All rights reserved.
43  *
44  * This code is derived from software contributed to Berkeley by
45  * the Systems Programming Group of the University of Utah Computer
46  * Science Department.
47  *
48  * Redistribution and use in source and binary forms, with or without
49  * modification, are permitted provided that the following conditions
50  * are met:
51  * 1. Redistributions of source code must retain the above copyright
52  *    notice, this list of conditions and the following disclaimer.
53  * 2. Redistributions in binary form must reproduce the above copyright
54  *    notice, this list of conditions and the following disclaimer in the
55  *    documentation and/or other materials provided with the distribution.
56  * 3. All advertising materials mentioning features or use of this software
57  *    must display the following acknowledgement:
58  *	This product includes software developed by the University of
59  *	California, Berkeley and its contributors.
60  * 4. Neither the name of the University nor the names of its contributors
61  *    may be used to endorse or promote products derived from this software
62  *    without specific prior written permission.
63  *
64  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
65  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
66  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
67  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
68  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
69  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
70  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
71  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
72  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
73  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
74  * SUCH DAMAGE.
75  *
76  * from: Utah $Hdr: vn.c 1.13 94/04/02$
77  *
78  *	@(#)vn.c	8.9 (Berkeley) 5/14/95
79  */
80 
81 /*
82  * Vnode disk driver.
83  *
84  * Block/character interface to a vnode.  Allows one to treat a file
85  * as a disk (e.g. build a filesystem in it, mount it, etc.).
86  *
87  * NOTE 1: This uses the VOP_BMAP/VOP_STRATEGY interface to the vnode
88  * instead of a simple VOP_RDWR.  We do this to avoid distorting the
89  * local buffer cache.
90  *
91  * NOTE 2: There is a security issue involved with this driver.
92  * Once mounted all access to the contents of the "mapped" file via
93  * the special file is controlled by the permissions on the special
94  * file, the protection of the mapped file is ignored (effectively,
95  * by using root credentials in all transactions).
96  *
97  * NOTE 3: Doesn't interact with leases, should it?
98  */
99 
100 #include "fs_nfs.h"
101 
102 #include <sys/param.h>
103 #include <sys/systm.h>
104 #include <sys/namei.h>
105 #include <sys/proc.h>
106 #include <sys/errno.h>
107 #include <sys/buf.h>
108 #include <sys/malloc.h>
109 #include <sys/ioctl.h>
110 #include <sys/disklabel.h>
111 #include <sys/device.h>
112 #include <sys/disk.h>
113 #include <sys/stat.h>
114 #include <sys/mount.h>
115 #include <sys/vnode.h>
116 #include <sys/file.h>
117 #include <sys/uio.h>
118 #include <sys/conf.h>
119 
120 #include <miscfs/specfs/specdev.h>
121 
122 #include <dev/vndvar.h>
123 
124 #if defined(VNDDEBUG) && !defined(DEBUG)
125 #define	DEBUG
126 #endif
127 
128 #ifdef DEBUG
129 int dovndcluster = 1;
130 #define	VDB_FOLLOW	0x01
131 #define	VDB_INIT	0x02
132 #define	VDB_IO		0x04
133 #define	VDB_LABEL	0x08
134 int vnddebug = 0x00;
135 #endif
136 
137 #define	vndunit(x)	DISKUNIT(x)
138 
139 struct vndxfer {
140 	struct buf	*vx_bp;		/* Pointer to parent buffer */
141 	int		vx_error;
142 	int		vx_pending;	/* # of pending aux buffers */
143 	int		vx_flags;
144 #define VX_BUSY		1
145 };
146 
147 struct vndbuf {
148 	struct buf	vb_buf;
149 	struct vndxfer	*vb_xfer;
150 };
151 
152 #define	VND_GETXFER(vnd)	pool_get(&(vnd)->sc_vxpool, PR_NOWAIT)
153 #define	VND_PUTXFER(vnd, vx)	pool_put(&(vnd)->sc_vxpool, (vx))
154 
155 #define	VND_GETBUF(vnd)		pool_get(&(vnd)->sc_vbpool, PR_NOWAIT)
156 #define	VND_PUTBUF(vnd, vb)	pool_put(&(vnd)->sc_vbpool, (vb))
157 
158 struct vnd_softc *vnd_softc;
159 int numvnd = 0;
160 
161 #define	VNDLABELDEV(dev) \
162 	(MAKEDISKDEV(major((dev)), vndunit((dev)), RAW_PART))
163 
164 /* called by main() at boot time */
165 void	vndattach __P((int));
166 
167 void	vndclear __P((struct vnd_softc *));
168 void	vndstart __P((struct vnd_softc *));
169 int	vndsetcred __P((struct vnd_softc *, struct ucred *));
170 void	vndthrottle __P((struct vnd_softc *, struct vnode *));
171 void	vndiodone __P((struct buf *));
172 void	vndshutdown __P((void));
173 
174 void	vndgetdefaultlabel __P((struct vnd_softc *, struct disklabel *));
175 void	vndgetdisklabel __P((dev_t));
176 
177 static	int vndlock __P((struct vnd_softc *));
178 static	void vndunlock __P((struct vnd_softc *));
179 
180 void
181 vndattach(num)
182 	int num;
183 {
184 	int i;
185 	char *mem;
186 
187 	if (num <= 0)
188 		return;
189 	i = num * sizeof(struct vnd_softc);
190 	mem = malloc(i, M_DEVBUF, M_NOWAIT);
191 	if (mem == NULL) {
192 		printf("WARNING: no memory for vnode disks\n");
193 		return;
194 	}
195 	memset(mem, 0, i);
196 	vnd_softc = (struct vnd_softc *)mem;
197 	numvnd = num;
198 
199 	for (i = 0; i < numvnd; i++)
200 		BUFQ_INIT(&vnd_softc[i].sc_tab);
201 }
202 
203 int
204 vndopen(dev, flags, mode, p)
205 	dev_t dev;
206 	int flags, mode;
207 	struct proc *p;
208 {
209 	int unit = vndunit(dev);
210 	struct vnd_softc *sc;
211 	int error = 0, part, pmask;
212 	struct disklabel *lp;
213 
214 #ifdef DEBUG
215 	if (vnddebug & VDB_FOLLOW)
216 		printf("vndopen(0x%x, 0x%x, 0x%x, %p)\n", dev, flags, mode, p);
217 #endif
218 	if (unit >= numvnd)
219 		return (ENXIO);
220 	sc = &vnd_softc[unit];
221 
222 	if ((error = vndlock(sc)) != 0)
223 		return (error);
224 
225 	lp = sc->sc_dkdev.dk_label;
226 
227 	part = DISKPART(dev);
228 	pmask = (1 << part);
229 
230 	/*
231 	 * If we're initialized, check to see if there are any other
232 	 * open partitions.  If not, then it's safe to update the
233 	 * in-core disklabel.
234 	 */
235 	if ((sc->sc_flags & VNF_INITED) && (sc->sc_dkdev.dk_openmask == 0))
236 		vndgetdisklabel(dev);
237 
238 	/* Check that the partitions exists. */
239 	if (part != RAW_PART) {
240 		if (((sc->sc_flags & VNF_INITED) == 0) ||
241 		    ((part >= lp->d_npartitions) ||
242 		     (lp->d_partitions[part].p_fstype == FS_UNUSED))) {
243 			error = ENXIO;
244 			goto done;
245 		}
246 	}
247 
248 	/* Prevent our unit from being unconfigured while open. */
249 	switch (mode) {
250 	case S_IFCHR:
251 		sc->sc_dkdev.dk_copenmask |= pmask;
252 		break;
253 
254 	case S_IFBLK:
255 		sc->sc_dkdev.dk_bopenmask |= pmask;
256 		break;
257 	}
258 	sc->sc_dkdev.dk_openmask =
259 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
260 
261  done:
262 	vndunlock(sc);
263 	return (error);
264 }
265 
266 int
267 vndclose(dev, flags, mode, p)
268 	dev_t dev;
269 	int flags, mode;
270 	struct proc *p;
271 {
272 	int unit = vndunit(dev);
273 	struct vnd_softc *sc;
274 	int error = 0, part;
275 
276 #ifdef DEBUG
277 	if (vnddebug & VDB_FOLLOW)
278 		printf("vndclose(0x%x, 0x%x, 0x%x, %p)\n", dev, flags, mode, p);
279 #endif
280 
281 	if (unit >= numvnd)
282 		return (ENXIO);
283 	sc = &vnd_softc[unit];
284 
285 	if ((error = vndlock(sc)) != 0)
286 		return (error);
287 
288 	part = DISKPART(dev);
289 
290 	/* ...that much closer to allowing unconfiguration... */
291 	switch (mode) {
292 	case S_IFCHR:
293 		sc->sc_dkdev.dk_copenmask &= ~(1 << part);
294 		break;
295 
296 	case S_IFBLK:
297 		sc->sc_dkdev.dk_bopenmask &= ~(1 << part);
298 		break;
299 	}
300 	sc->sc_dkdev.dk_openmask =
301 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
302 
303 	vndunlock(sc);
304 	return (0);
305 }
306 
307 /*
308  * Break the request into bsize pieces and submit using VOP_BMAP/VOP_STRATEGY.
309  */
310 void
311 vndstrategy(bp)
312 	struct buf *bp;
313 {
314 	int unit = vndunit(bp->b_dev);
315 	struct vnd_softc *vnd = &vnd_softc[unit];
316 	struct vndxfer *vnx;
317 	int s, bsize, resid;
318 	off_t bn;
319 	caddr_t addr;
320 	int sz, flags, error, wlabel;
321 	struct disklabel *lp;
322 	struct partition *pp;
323 
324 #ifdef DEBUG
325 	if (vnddebug & VDB_FOLLOW)
326 		printf("vndstrategy(%p): unit %d\n", bp, unit);
327 #endif
328 	if ((vnd->sc_flags & VNF_INITED) == 0) {
329 		bp->b_error = ENXIO;
330 		bp->b_flags |= B_ERROR;
331 		goto done;
332 	}
333 
334 	/* If it's a nil transfer, wake up the top half now. */
335 	if (bp->b_bcount == 0)
336 		goto done;
337 
338 	lp = vnd->sc_dkdev.dk_label;
339 
340 	/*
341 	 * The transfer must be a whole number of blocks.
342 	 */
343 	if ((bp->b_bcount % lp->d_secsize) != 0) {
344 		bp->b_error = EINVAL;
345 		bp->b_flags |= B_ERROR;
346 		goto done;
347 	}
348 
349 	/*
350 	 * Do bounds checking and adjust transfer.  If there's an error,
351 	 * the bounds check will flag that for us.
352 	 */
353 	wlabel = vnd->sc_flags & (VNF_WLABEL|VNF_LABELLING);
354 	if (DISKPART(bp->b_dev) != RAW_PART)
355 		if (bounds_check_with_label(bp, lp, wlabel) <= 0)
356 			goto done;
357 
358 	bp->b_resid = bp->b_bcount;
359 
360 	/*
361 	 * Put the block number in terms of the logical blocksize
362 	 * of the "device".
363 	 */
364 	bn = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
365 
366 	/*
367 	 * Translate the partition-relative block number to an absolute.
368 	 */
369 	if (DISKPART(bp->b_dev) != RAW_PART) {
370 		pp = &vnd->sc_dkdev.dk_label->d_partitions[DISKPART(bp->b_dev)];
371 		bn += pp->p_offset;
372 	}
373 
374 	/* ...and convert to a byte offset within the file. */
375 	bn *= lp->d_secsize;
376 
377  	bsize = vnd->sc_vp->v_mount->mnt_stat.f_iosize;
378 	addr = bp->b_data;
379 	flags = (bp->b_flags & (B_READ|B_ASYNC)) | B_CALL;
380 
381 	/* Allocate a header for this transfer and link it to the buffer */
382 	s = splbio();
383 	vnx = VND_GETXFER(vnd);
384 	splx(s);
385 	vnx->vx_flags = VX_BUSY;
386 	vnx->vx_error = 0;
387 	vnx->vx_pending = 0;
388 	vnx->vx_bp = bp;
389 
390 	for (resid = bp->b_resid; resid; resid -= sz) {
391 		struct vndbuf *nbp;
392 		struct vnode *vp;
393 		daddr_t nbn;
394 		int off, nra;
395 
396 		nra = 0;
397 		vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY | LK_CANRECURSE);
398 		error = VOP_BMAP(vnd->sc_vp, bn / bsize, &vp, &nbn, &nra);
399 		VOP_UNLOCK(vnd->sc_vp, 0);
400 
401 		if (error == 0 && (long)nbn == -1)
402 			error = EIO;
403 
404 		/*
405 		 * If there was an error or a hole in the file...punt.
406 		 * Note that we may have to wait for any operations
407 		 * that we have already fired off before releasing
408 		 * the buffer.
409 		 *
410 		 * XXX we could deal with holes here but it would be
411 		 * a hassle (in the write case).
412 		 */
413 		if (error) {
414 			s = splbio();
415 			vnx->vx_error = error;
416 			goto out;
417 		}
418 
419 #ifdef DEBUG
420 		if (!dovndcluster)
421 			nra = 0;
422 #endif
423 
424 		if ((off = bn % bsize) != 0)
425 			sz = bsize - off;
426 		else
427 			sz = (1 + nra) * bsize;
428 		if (resid < sz)
429 			sz = resid;
430 #ifdef DEBUG
431 		if (vnddebug & VDB_IO)
432 			printf("vndstrategy: vp %p/%p bn 0x%qx/0x%x sz 0x%x\n",
433 			    vnd->sc_vp, vp, (long long)bn, nbn, sz);
434 #endif
435 
436 		s = splbio();
437 		nbp = VND_GETBUF(vnd);
438 		splx(s);
439 		nbp->vb_buf.b_flags = flags;
440 		nbp->vb_buf.b_bcount = sz;
441 		nbp->vb_buf.b_bufsize = bp->b_bufsize;
442 		nbp->vb_buf.b_error = 0;
443 		nbp->vb_buf.b_data = addr;
444 		nbp->vb_buf.b_blkno = nbp->vb_buf.b_rawblkno = nbn + btodb(off);
445 		nbp->vb_buf.b_proc = bp->b_proc;
446 		nbp->vb_buf.b_iodone = vndiodone;
447 		nbp->vb_buf.b_vp = NULLVP;
448 		LIST_INIT(&nbp->vb_buf.b_dep);
449 
450 		nbp->vb_xfer = vnx;
451 
452 		/*
453 		 * Just sort by block number
454 		 */
455 		s = splbio();
456 		if (vnx->vx_error != 0) {
457 			VND_PUTBUF(vnd, nbp);
458 			goto out;
459 		}
460 		vnx->vx_pending++;
461 		bgetvp(vp, &nbp->vb_buf);
462 		disksort_blkno(&vnd->sc_tab, &nbp->vb_buf);
463 		vndstart(vnd);
464 		splx(s);
465 		bn += sz;
466 		addr += sz;
467 	}
468 
469 	s = splbio();
470 
471 out: /* Arrive here at splbio */
472 	vnx->vx_flags &= ~VX_BUSY;
473 	if (vnx->vx_pending == 0) {
474 		if (vnx->vx_error != 0) {
475 			bp->b_error = vnx->vx_error;
476 			bp->b_flags |= B_ERROR;
477 		}
478 		VND_PUTXFER(vnd, vnx);
479 		biodone(bp);
480 	}
481 	splx(s);
482 	return;
483 
484  done:
485 	biodone(bp);
486 }
487 
488 /*
489  * Feed requests sequentially.
490  * We do it this way to keep from flooding NFS servers if we are connected
491  * to an NFS file.  This places the burden on the client rather than the
492  * server.
493  */
494 void
495 vndstart(vnd)
496 	struct vnd_softc *vnd;
497 {
498 	struct buf	*bp;
499 
500 	/*
501 	 * Dequeue now since lower level strategy routine might
502 	 * queue using same links
503 	 */
504 
505 	if ((vnd->sc_flags & VNF_BUSY) != 0)
506 		return;
507 
508 	vnd->sc_flags |= VNF_BUSY;
509 
510 	while (vnd->sc_active < vnd->sc_maxactive) {
511 		bp = BUFQ_FIRST(&vnd->sc_tab);
512 		if (bp == NULL)
513 			break;
514 		BUFQ_REMOVE(&vnd->sc_tab, bp);
515 		vnd->sc_active++;
516 #ifdef DEBUG
517 		if (vnddebug & VDB_IO)
518 			printf("vndstart(%ld): bp %p vp %p blkno 0x%x"
519 				" flags %lx addr %p cnt 0x%lx\n",
520 			    (long) (vnd-vnd_softc), bp, bp->b_vp, bp->b_blkno,
521 			    bp->b_flags, bp->b_data, bp->b_bcount);
522 #endif
523 
524 		/* Instrumentation. */
525 		disk_busy(&vnd->sc_dkdev);
526 
527 		if ((bp->b_flags & B_READ) == 0)
528 			bp->b_vp->v_numoutput++;
529 		VOP_STRATEGY(bp);
530 	}
531 	vnd->sc_flags &= ~VNF_BUSY;
532 }
533 
534 void
535 vndiodone(bp)
536 	struct buf *bp;
537 {
538 	struct vndbuf *vbp = (struct vndbuf *) bp;
539 	struct vndxfer *vnx = (struct vndxfer *)vbp->vb_xfer;
540 	struct buf *pbp = vnx->vx_bp;
541 	struct vnd_softc *vnd = &vnd_softc[vndunit(pbp->b_dev)];
542 	int s, resid;
543 
544 	s = splbio();
545 #ifdef DEBUG
546 	if (vnddebug & VDB_IO)
547 		printf("vndiodone(%ld): vbp %p vp %p blkno 0x%x addr %p cnt 0x%lx\n",
548 		    (long) (vnd-vnd_softc), vbp, vbp->vb_buf.b_vp,
549 		    vbp->vb_buf.b_blkno, vbp->vb_buf.b_data,
550 		    vbp->vb_buf.b_bcount);
551 #endif
552 
553 	resid = vbp->vb_buf.b_bcount - vbp->vb_buf.b_resid;
554 	pbp->b_resid -= resid;
555 	disk_unbusy(&vnd->sc_dkdev, resid);
556 	vnx->vx_pending--;
557 
558 	if (vbp->vb_buf.b_error) {
559 #ifdef DEBUG
560 		if (vnddebug & VDB_IO)
561 			printf("vndiodone: vbp %p error %d\n", vbp,
562 			    vbp->vb_buf.b_error);
563 #endif
564 		vnx->vx_error = vbp->vb_buf.b_error;
565 	}
566 
567 	if (vbp->vb_buf.b_vp != NULLVP)
568 		brelvp(&vbp->vb_buf);
569 
570 	VND_PUTBUF(vnd, vbp);
571 
572 	/*
573 	 * Wrap up this transaction if it has run to completion or, in
574 	 * case of an error, when all auxiliary buffers have returned.
575 	 */
576 	if (vnx->vx_error != 0) {
577 		pbp->b_flags |= B_ERROR;
578 		pbp->b_error = vnx->vx_error;
579 		if ((vnx->vx_flags & VX_BUSY) == 0 && vnx->vx_pending == 0) {
580 
581 #ifdef DEBUG
582 			if (vnddebug & VDB_IO)
583 				printf("vndiodone: pbp %p iodone: error %d\n",
584 					pbp, vnx->vx_error);
585 #endif
586 			VND_PUTXFER(vnd, vnx);
587 			biodone(pbp);
588 		}
589 	} else if (pbp->b_resid == 0) {
590 
591 #ifdef DIAGNOSTIC
592 		if (vnx->vx_pending != 0)
593 			panic("vndiodone: vnx pending: %d", vnx->vx_pending);
594 #endif
595 
596 		if ((vnx->vx_flags & VX_BUSY) == 0) {
597 #ifdef DEBUG
598 			if (vnddebug & VDB_IO)
599 				printf("vndiodone: pbp %p iodone\n", pbp);
600 #endif
601 			VND_PUTXFER(vnd, vnx);
602 			biodone(pbp);
603 		}
604 	}
605 
606 	vnd->sc_active--;
607 	vndstart(vnd);
608 	splx(s);
609 }
610 
611 /* ARGSUSED */
612 int
613 vndread(dev, uio, flags)
614 	dev_t dev;
615 	struct uio *uio;
616 	int flags;
617 {
618 	int unit = vndunit(dev);
619 	struct vnd_softc *sc;
620 
621 #ifdef DEBUG
622 	if (vnddebug & VDB_FOLLOW)
623 		printf("vndread(0x%x, %p)\n", dev, uio);
624 #endif
625 
626 	if (unit >= numvnd)
627 		return (ENXIO);
628 	sc = &vnd_softc[unit];
629 
630 	if ((sc->sc_flags & VNF_INITED) == 0)
631 		return (ENXIO);
632 
633 	return (physio(vndstrategy, NULL, dev, B_READ, minphys, uio));
634 }
635 
636 /* ARGSUSED */
637 int
638 vndwrite(dev, uio, flags)
639 	dev_t dev;
640 	struct uio *uio;
641 	int flags;
642 {
643 	int unit = vndunit(dev);
644 	struct vnd_softc *sc;
645 
646 #ifdef DEBUG
647 	if (vnddebug & VDB_FOLLOW)
648 		printf("vndwrite(0x%x, %p)\n", dev, uio);
649 #endif
650 
651 	if (unit >= numvnd)
652 		return (ENXIO);
653 	sc = &vnd_softc[unit];
654 
655 	if ((sc->sc_flags & VNF_INITED) == 0)
656 		return (ENXIO);
657 
658 	return (physio(vndstrategy, NULL, dev, B_WRITE, minphys, uio));
659 }
660 
661 /* ARGSUSED */
662 int
663 vndioctl(dev, cmd, data, flag, p)
664 	dev_t dev;
665 	u_long cmd;
666 	caddr_t data;
667 	int flag;
668 	struct proc *p;
669 {
670 	int unit = vndunit(dev);
671 	struct vnd_softc *vnd;
672 	struct vnd_ioctl *vio;
673 	struct vattr vattr;
674 	struct nameidata nd;
675 	int error, part, pmask;
676 	size_t geomsize;
677 #ifdef __HAVE_OLD_DISKLABEL
678 	struct disklabel newlabel;
679 #endif
680 
681 #ifdef DEBUG
682 	if (vnddebug & VDB_FOLLOW)
683 		printf("vndioctl(0x%x, 0x%lx, %p, 0x%x, %p): unit %d\n",
684 		    dev, cmd, data, flag, p, unit);
685 #endif
686 	error = suser(p->p_ucred, &p->p_acflag);
687 	if (error)
688 		return (error);
689 	if (unit >= numvnd)
690 		return (ENXIO);
691 
692 	vnd = &vnd_softc[unit];
693 	vio = (struct vnd_ioctl *)data;
694 
695 	/* Must be open for writes for these commands... */
696 	switch (cmd) {
697 	case VNDIOCSET:
698 	case VNDIOCCLR:
699 	case DIOCSDINFO:
700 	case DIOCWDINFO:
701 #ifdef __HAVE_OLD_DISKLABEL
702 	case ODIOCSDINFO:
703 	case ODIOCWDINFO:
704 #endif
705 	case DIOCWLABEL:
706 		if ((flag & FWRITE) == 0)
707 			return (EBADF);
708 	}
709 
710 	/* Must be initialized for these... */
711 	switch (cmd) {
712 	case VNDIOCCLR:
713 	case DIOCGDINFO:
714 	case DIOCSDINFO:
715 	case DIOCWDINFO:
716 	case DIOCGPART:
717 	case DIOCWLABEL:
718 	case DIOCGDEFLABEL:
719 #ifdef __HAVE_OLD_DISKLABEL
720 	case ODIOCGDINFO:
721 	case ODIOCSDINFO:
722 	case ODIOCWDINFO:
723 	case ODIOCGDEFLABEL:
724 #endif
725 		if ((vnd->sc_flags & VNF_INITED) == 0)
726 			return (ENXIO);
727 	}
728 
729 	switch (cmd) {
730 	case VNDIOCSET:
731 		if (vnd->sc_flags & VNF_INITED)
732 			return (EBUSY);
733 
734 		if ((error = vndlock(vnd)) != 0)
735 			return (error);
736 
737 		/*
738 		 * Always open for read and write.
739 		 * This is probably bogus, but it lets vn_open()
740 		 * weed out directories, sockets, etc. so we don't
741 		 * have to worry about them.
742 		 */
743 		NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, vio->vnd_file, p);
744 		if ((error = vn_open(&nd, FREAD|FWRITE, 0)) != 0) {
745 			vndunlock(vnd);
746 			return(error);
747 		}
748 		error = VOP_GETATTR(nd.ni_vp, &vattr, p->p_ucred, p);
749 		if (error) {
750 			VOP_UNLOCK(nd.ni_vp, 0);
751 			(void) vn_close(nd.ni_vp, FREAD|FWRITE, p->p_ucred, p);
752 			vndunlock(vnd);
753 			return(error);
754 		}
755 		VOP_UNLOCK(nd.ni_vp, 0);
756 		vnd->sc_vp = nd.ni_vp;
757 		vnd->sc_size = btodb(vattr.va_size);	/* note truncation */
758 
759 		/*
760 		 * Use pseudo-geometry specified.  If none was provided,
761 		 * use "standard" Adaptec fictitious geometry.
762 		 */
763 		if (vio->vnd_flags & VNDIOF_HASGEOM) {
764 
765 			memcpy(&vnd->sc_geom, &vio->vnd_geom,
766 			    sizeof(vio->vnd_geom));
767 
768 			/*
769 			 * Sanity-check the sector size.
770 			 * XXX Don't allow secsize < DEV_BSIZE.  Should
771 			 * XXX we?
772 			 */
773 			if (vnd->sc_geom.vng_secsize < DEV_BSIZE ||
774 			    (vnd->sc_geom.vng_secsize % DEV_BSIZE) != 0) {
775 				(void) vn_close(nd.ni_vp, FREAD|FWRITE,
776 				    p->p_ucred, p);
777 				vndunlock(vnd);
778 				return (EINVAL);
779 			}
780 
781 			/*
782 			 * Compute the size (in DEV_BSIZE blocks) specified
783 			 * by the geometry.
784 			 */
785 			geomsize = (vnd->sc_geom.vng_nsectors *
786 			    vnd->sc_geom.vng_ntracks *
787 			    vnd->sc_geom.vng_ncylinders) *
788 			    (vnd->sc_geom.vng_secsize / DEV_BSIZE);
789 
790 			/*
791 			 * Sanity-check the size against the specified
792 			 * geometry.
793 			 */
794 			if (vnd->sc_size < geomsize) {
795 				(void) vn_close(nd.ni_vp, FREAD|FWRITE,
796 				    p->p_ucred, p);
797 				vndunlock(vnd);
798 				return (EINVAL);
799 			}
800 		} else {
801 			/*
802 			 * Size must be at least 2048 DEV_BSIZE blocks
803 			 * (1M) in order to use this geometry.
804 			 */
805 			if (vnd->sc_size < (32 * 64)) {
806 				vndunlock(vnd);
807 				return (EINVAL);
808 			}
809 
810 			vnd->sc_geom.vng_secsize = DEV_BSIZE;
811 			vnd->sc_geom.vng_nsectors = 32;
812 			vnd->sc_geom.vng_ntracks = 64;
813 			vnd->sc_geom.vng_ncylinders = vnd->sc_size / (64 * 32);
814 
815 			/*
816 			 * Compute the actual size allowed by this geometry.
817 			 */
818 			geomsize = 32 * 64 * vnd->sc_geom.vng_ncylinders;
819 		}
820 
821 		/*
822 		 * Truncate the size to that specified by
823 		 * the geometry.
824 		 * XXX Should we even bother with this?
825 		 */
826 		vnd->sc_size = geomsize;
827 
828 		if ((error = vndsetcred(vnd, p->p_ucred)) != 0) {
829 			(void) vn_close(nd.ni_vp, FREAD|FWRITE, p->p_ucred, p);
830 			vndunlock(vnd);
831 			return(error);
832 		}
833 		vndthrottle(vnd, vnd->sc_vp);
834 		vio->vnd_size = dbtob(vnd->sc_size);
835 		vnd->sc_flags |= VNF_INITED;
836 #ifdef DEBUG
837 		if (vnddebug & VDB_INIT)
838 			printf("vndioctl: SET vp %p size 0x%lx %d/%d/%d/%d\n",
839 			    vnd->sc_vp, (unsigned long) vnd->sc_size,
840 			    vnd->sc_geom.vng_secsize,
841 			    vnd->sc_geom.vng_nsectors,
842 			    vnd->sc_geom.vng_ntracks,
843 			    vnd->sc_geom.vng_ncylinders);
844 #endif
845 
846 		/* Attach the disk. */
847 		memset(vnd->sc_xname, 0, sizeof(vnd->sc_xname)); /* XXX */
848 		sprintf(vnd->sc_xname, "vnd%d", unit);		/* XXX */
849 		vnd->sc_dkdev.dk_name = vnd->sc_xname;
850 		disk_attach(&vnd->sc_dkdev);
851 
852 		/* Initialize the xfer and buffer pools. */
853 		pool_init(&vnd->sc_vxpool, sizeof(struct vndxfer), 0,
854 		    0, 0, "vndxpl", 0, NULL, NULL, M_DEVBUF);
855 		pool_init(&vnd->sc_vbpool, sizeof(struct vndbuf), 0,
856 		    0, 0, "vndbpl", 0, NULL, NULL, M_DEVBUF);
857 
858 		/* Try and read the disklabel. */
859 		vndgetdisklabel(dev);
860 
861 		vndunlock(vnd);
862 
863 		break;
864 
865 	case VNDIOCCLR:
866 		if ((error = vndlock(vnd)) != 0)
867 			return (error);
868 
869 		/*
870 		 * Don't unconfigure if any other partitions are open
871 		 * or if both the character and block flavors of this
872 		 * partition are open.
873 		 */
874 		part = DISKPART(dev);
875 		pmask = (1 << part);
876 		if ((vnd->sc_dkdev.dk_openmask & ~pmask) ||
877 		    ((vnd->sc_dkdev.dk_bopenmask & pmask) &&
878 		    (vnd->sc_dkdev.dk_copenmask & pmask))) {
879 			vndunlock(vnd);
880 			return (EBUSY);
881 		}
882 
883 		vndclear(vnd);
884 #ifdef DEBUG
885 		if (vnddebug & VDB_INIT)
886 			printf("vndioctl: CLRed\n");
887 #endif
888 
889 		/* Destroy the xfer and buffer pools. */
890 		pool_destroy(&vnd->sc_vxpool);
891 		pool_destroy(&vnd->sc_vbpool);
892 
893 		/* Detatch the disk. */
894 		disk_detach(&vnd->sc_dkdev);
895 
896 		vndunlock(vnd);
897 
898 		break;
899 
900 	case DIOCGDINFO:
901 		*(struct disklabel *)data = *(vnd->sc_dkdev.dk_label);
902 		break;
903 
904 #ifdef __HAVE_OLD_DISKLABEL
905 	case ODIOCGDINFO:
906 		newlabel = *(vnd->sc_dkdev.dk_label);
907 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
908 			return ENOTTY;
909 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
910 		break;
911 #endif
912 
913 	case DIOCGPART:
914 		((struct partinfo *)data)->disklab = vnd->sc_dkdev.dk_label;
915 		((struct partinfo *)data)->part =
916 		    &vnd->sc_dkdev.dk_label->d_partitions[DISKPART(dev)];
917 		break;
918 
919 	case DIOCWDINFO:
920 	case DIOCSDINFO:
921 #ifdef __HAVE_OLD_DISKLABEL
922 	case ODIOCWDINFO:
923 	case ODIOCSDINFO:
924 #endif
925 	{
926 		struct disklabel *lp;
927 
928 		if ((error = vndlock(vnd)) != 0)
929 			return (error);
930 
931 		vnd->sc_flags |= VNF_LABELLING;
932 
933 #ifdef __HAVE_OLD_DISKLABEL
934 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
935 			memset(&newlabel, 0, sizeof newlabel);
936 			memcpy(&newlabel, data, sizeof (struct olddisklabel));
937 			lp = &newlabel;
938 		} else
939 #endif
940 		lp = (struct disklabel *)data;
941 
942 		error = setdisklabel(vnd->sc_dkdev.dk_label,
943 		    lp, 0, vnd->sc_dkdev.dk_cpulabel);
944 		if (error == 0) {
945 			if (cmd == DIOCWDINFO
946 #ifdef __HAVE_OLD_DISKLABEL
947 			    || cmd == ODIOCWDINFO
948 #endif
949 			   )
950 				error = writedisklabel(VNDLABELDEV(dev),
951 				    vndstrategy, vnd->sc_dkdev.dk_label,
952 				    vnd->sc_dkdev.dk_cpulabel);
953 		}
954 
955 		vnd->sc_flags &= ~VNF_LABELLING;
956 
957 		vndunlock(vnd);
958 
959 		if (error)
960 			return (error);
961 		break;
962 	}
963 
964 	case DIOCWLABEL:
965 		if (*(int *)data != 0)
966 			vnd->sc_flags |= VNF_WLABEL;
967 		else
968 			vnd->sc_flags &= ~VNF_WLABEL;
969 		break;
970 
971 	case DIOCGDEFLABEL:
972 		vndgetdefaultlabel(vnd, (struct disklabel *)data);
973 		break;
974 
975 #ifdef __HAVE_OLD_DISKLABEL
976 	case ODIOCGDEFLABEL:
977 		vndgetdefaultlabel(vnd, &newlabel);
978 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
979 			return ENOTTY;
980 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
981 		break;
982 #endif
983 
984 	default:
985 		return (ENOTTY);
986 	}
987 
988 	return (0);
989 }
990 
991 /*
992  * Duplicate the current processes' credentials.  Since we are called only
993  * as the result of a SET ioctl and only root can do that, any future access
994  * to this "disk" is essentially as root.  Note that credentials may change
995  * if some other uid can write directly to the mapped file (NFS).
996  */
997 int
998 vndsetcred(vnd, cred)
999 	struct vnd_softc *vnd;
1000 	struct ucred *cred;
1001 {
1002 	struct uio auio;
1003 	struct iovec aiov;
1004 	char *tmpbuf;
1005 	int error;
1006 
1007 	vnd->sc_cred = crdup(cred);
1008 	tmpbuf = malloc(DEV_BSIZE, M_TEMP, M_WAITOK);
1009 
1010 	/* XXX: Horrible kludge to establish credentials for NFS */
1011 	aiov.iov_base = tmpbuf;
1012 	aiov.iov_len = min(DEV_BSIZE, dbtob(vnd->sc_size));
1013 	auio.uio_iov = &aiov;
1014 	auio.uio_iovcnt = 1;
1015 	auio.uio_offset = 0;
1016 	auio.uio_rw = UIO_READ;
1017 	auio.uio_segflg = UIO_SYSSPACE;
1018 	auio.uio_resid = aiov.iov_len;
1019 	vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1020 	error = VOP_READ(vnd->sc_vp, &auio, 0, vnd->sc_cred);
1021 	if (error == 0) {
1022 		/*
1023 		 * Because vnd does all IO directly through the vnode
1024 		 * we need to flush (at least) the buffer from the above
1025 		 * VOP_READ from the buffer cache to prevent cache
1026 		 * incoherencies.  Also, be careful to write dirty
1027 		 * buffers back to stable storage.
1028 		 */
1029 		error = vinvalbuf(vnd->sc_vp, V_SAVE, vnd->sc_cred,
1030 			    curproc, 0, 0);
1031 	}
1032 	VOP_UNLOCK(vnd->sc_vp, 0);
1033 
1034 	free(tmpbuf, M_TEMP);
1035 	return (error);
1036 }
1037 
1038 /*
1039  * Set maxactive based on FS type
1040  */
1041 void
1042 vndthrottle(vnd, vp)
1043 	struct vnd_softc *vnd;
1044 	struct vnode *vp;
1045 {
1046 #ifdef NFS
1047 	extern int (**nfsv2_vnodeop_p) __P((void *));
1048 
1049 	if (vp->v_op == nfsv2_vnodeop_p)
1050 		vnd->sc_maxactive = 2;
1051 	else
1052 #endif
1053 		vnd->sc_maxactive = 8;
1054 
1055 	if (vnd->sc_maxactive < 1)
1056 		vnd->sc_maxactive = 1;
1057 }
1058 
1059 void
1060 vndshutdown()
1061 {
1062 	struct vnd_softc *vnd;
1063 
1064 	for (vnd = &vnd_softc[0]; vnd < &vnd_softc[numvnd]; vnd++)
1065 		if (vnd->sc_flags & VNF_INITED)
1066 			vndclear(vnd);
1067 }
1068 
1069 void
1070 vndclear(vnd)
1071 	struct vnd_softc *vnd;
1072 {
1073 	struct vnode *vp = vnd->sc_vp;
1074 	struct proc *p = curproc;		/* XXX */
1075 
1076 #ifdef DEBUG
1077 	if (vnddebug & VDB_FOLLOW)
1078 		printf("vndclear(%p): vp %p\n", vnd, vp);
1079 #endif
1080 	vnd->sc_flags &= ~VNF_INITED;
1081 	if (vp == (struct vnode *)0)
1082 		panic("vndioctl: null vp");
1083 	(void) vn_close(vp, FREAD|FWRITE, vnd->sc_cred, p);
1084 	crfree(vnd->sc_cred);
1085 	vnd->sc_vp = (struct vnode *)0;
1086 	vnd->sc_cred = (struct ucred *)0;
1087 	vnd->sc_size = 0;
1088 }
1089 
1090 int
1091 vndsize(dev)
1092 	dev_t dev;
1093 {
1094 	struct vnd_softc *sc;
1095 	struct disklabel *lp;
1096 	int part, unit, omask;
1097 	int size;
1098 
1099 	unit = vndunit(dev);
1100 	if (unit >= numvnd)
1101 		return (-1);
1102 	sc = &vnd_softc[unit];
1103 
1104 	if ((sc->sc_flags & VNF_INITED) == 0)
1105 		return (-1);
1106 
1107 	part = DISKPART(dev);
1108 	omask = sc->sc_dkdev.dk_openmask & (1 << part);
1109 	lp = sc->sc_dkdev.dk_label;
1110 
1111 	if (omask == 0 && vndopen(dev, 0, S_IFBLK, curproc))
1112 		return (-1);
1113 
1114 	if (lp->d_partitions[part].p_fstype != FS_SWAP)
1115 		size = -1;
1116 	else
1117 		size = lp->d_partitions[part].p_size *
1118 		    (lp->d_secsize / DEV_BSIZE);
1119 
1120 	if (omask == 0 && vndclose(dev, 0, S_IFBLK, curproc))
1121 		return (-1);
1122 
1123 	return (size);
1124 }
1125 
1126 int
1127 vnddump(dev, blkno, va, size)
1128 	dev_t dev;
1129 	daddr_t blkno;
1130 	caddr_t va;
1131 	size_t size;
1132 {
1133 
1134 	/* Not implemented. */
1135 	return ENXIO;
1136 }
1137 
1138 void
1139 vndgetdefaultlabel(sc, lp)
1140 	struct vnd_softc *sc;
1141 	struct disklabel *lp;
1142 {
1143 	struct vndgeom *vng = &sc->sc_geom;
1144 	struct partition *pp;
1145 
1146 	memset(lp, 0, sizeof(*lp));
1147 
1148 	lp->d_secperunit = sc->sc_size / (vng->vng_secsize / DEV_BSIZE);
1149 	lp->d_secsize = vng->vng_secsize;
1150 	lp->d_nsectors = vng->vng_nsectors;
1151 	lp->d_ntracks = vng->vng_ntracks;
1152 	lp->d_ncylinders = vng->vng_ncylinders;
1153 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1154 
1155 	strncpy(lp->d_typename, "vnd", sizeof(lp->d_typename));
1156 	lp->d_type = DTYPE_VND;
1157 	strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
1158 	lp->d_rpm = 3600;
1159 	lp->d_interleave = 1;
1160 	lp->d_flags = 0;
1161 
1162 	pp = &lp->d_partitions[RAW_PART];
1163 	pp->p_offset = 0;
1164 	pp->p_size = lp->d_secperunit;
1165 	pp->p_fstype = FS_UNUSED;
1166 	lp->d_npartitions = RAW_PART + 1;
1167 
1168 	lp->d_magic = DISKMAGIC;
1169 	lp->d_magic2 = DISKMAGIC;
1170 	lp->d_checksum = dkcksum(lp);
1171 }
1172 
1173 /*
1174  * Read the disklabel from a vnd.  If one is not present, create a fake one.
1175  */
1176 void
1177 vndgetdisklabel(dev)
1178 	dev_t dev;
1179 {
1180 	struct vnd_softc *sc = &vnd_softc[vndunit(dev)];
1181 	char *errstring;
1182 	struct disklabel *lp = sc->sc_dkdev.dk_label;
1183 	struct cpu_disklabel *clp = sc->sc_dkdev.dk_cpulabel;
1184 	int i;
1185 
1186 	memset(clp, 0, sizeof(*clp));
1187 
1188 	vndgetdefaultlabel(sc, lp);
1189 
1190 	/*
1191 	 * Call the generic disklabel extraction routine.
1192 	 */
1193 	errstring = readdisklabel(VNDLABELDEV(dev), vndstrategy, lp, clp);
1194 	if (errstring) {
1195 		/*
1196 		 * Lack of disklabel is common, but we print the warning
1197 		 * anyway, since it might contain other useful information.
1198 		 */
1199 		printf("%s: %s\n", sc->sc_xname, errstring);
1200 
1201 		/*
1202 		 * For historical reasons, if there's no disklabel
1203 		 * present, all partitions must be FS_BSDFFS and
1204 		 * occupy the entire disk.
1205 		 */
1206 		for (i = 0; i < MAXPARTITIONS; i++) {
1207 			/*
1208 			 * Don't wipe out port specific hack (such as
1209 			 * dos partition hack of i386 port).
1210 			 */
1211 			if (lp->d_partitions[i].p_fstype != FS_UNUSED)
1212 				continue;
1213 
1214 			lp->d_partitions[i].p_size = lp->d_secperunit;
1215 			lp->d_partitions[i].p_offset = 0;
1216 			lp->d_partitions[i].p_fstype = FS_BSDFFS;
1217 		}
1218 
1219 		strncpy(lp->d_packname, "default label",
1220 		    sizeof(lp->d_packname));
1221 
1222 		lp->d_checksum = dkcksum(lp);
1223 	}
1224 }
1225 
1226 /*
1227  * Wait interruptibly for an exclusive lock.
1228  *
1229  * XXX
1230  * Several drivers do this; it should be abstracted and made MP-safe.
1231  */
1232 static int
1233 vndlock(sc)
1234 	struct vnd_softc *sc;
1235 {
1236 	int error;
1237 
1238 	while ((sc->sc_flags & VNF_LOCKED) != 0) {
1239 		sc->sc_flags |= VNF_WANTED;
1240 		if ((error = tsleep(sc, PRIBIO | PCATCH, "vndlck", 0)) != 0)
1241 			return (error);
1242 	}
1243 	sc->sc_flags |= VNF_LOCKED;
1244 	return (0);
1245 }
1246 
1247 /*
1248  * Unlock and wake up any waiters.
1249  */
1250 static void
1251 vndunlock(sc)
1252 	struct vnd_softc *sc;
1253 {
1254 
1255 	sc->sc_flags &= ~VNF_LOCKED;
1256 	if ((sc->sc_flags & VNF_WANTED) != 0) {
1257 		sc->sc_flags &= ~VNF_WANTED;
1258 		wakeup(sc);
1259 	}
1260 }
1261