xref: /netbsd-src/sys/dev/vnd.c (revision 1ca5c1b28139779176bd5c13ad7c5f25c0bcd5f8)
1 /*	$NetBSD: vnd.c,v 1.77 2002/01/13 19:28:08 tsutsui 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 <sys/cdefs.h>
101 __KERNEL_RCSID(0, "$NetBSD: vnd.c,v 1.77 2002/01/13 19:28:08 tsutsui Exp $");
102 
103 #if defined(_KERNEL_OPT)
104 #include "fs_nfs.h"
105 #endif
106 
107 #include <sys/param.h>
108 #include <sys/systm.h>
109 #include <sys/namei.h>
110 #include <sys/proc.h>
111 #include <sys/errno.h>
112 #include <sys/buf.h>
113 #include <sys/malloc.h>
114 #include <sys/ioctl.h>
115 #include <sys/disklabel.h>
116 #include <sys/device.h>
117 #include <sys/disk.h>
118 #include <sys/stat.h>
119 #include <sys/mount.h>
120 #include <sys/vnode.h>
121 #include <sys/file.h>
122 #include <sys/uio.h>
123 #include <sys/conf.h>
124 
125 #include <miscfs/specfs/specdev.h>
126 
127 #include <dev/vndvar.h>
128 
129 #if defined(VNDDEBUG) && !defined(DEBUG)
130 #define	DEBUG
131 #endif
132 
133 #ifdef DEBUG
134 int dovndcluster = 1;
135 #define	VDB_FOLLOW	0x01
136 #define	VDB_INIT	0x02
137 #define	VDB_IO		0x04
138 #define	VDB_LABEL	0x08
139 int vnddebug = 0x00;
140 #endif
141 
142 #define	vndunit(x)	DISKUNIT(x)
143 
144 struct vndxfer {
145 	struct buf	*vx_bp;		/* Pointer to parent buffer */
146 	int		vx_error;
147 	int		vx_pending;	/* # of pending aux buffers */
148 	int		vx_flags;
149 #define VX_BUSY		1
150 };
151 
152 struct vndbuf {
153 	struct buf	vb_buf;
154 	struct vndxfer	*vb_xfer;
155 };
156 
157 #define	VND_GETXFER(vnd)	pool_get(&(vnd)->sc_vxpool, PR_NOWAIT)
158 #define	VND_PUTXFER(vnd, vx)	pool_put(&(vnd)->sc_vxpool, (vx))
159 
160 #define	VND_GETBUF(vnd)		pool_get(&(vnd)->sc_vbpool, PR_NOWAIT)
161 #define	VND_PUTBUF(vnd, vb)	pool_put(&(vnd)->sc_vbpool, (vb))
162 
163 struct vnd_softc *vnd_softc;
164 int numvnd = 0;
165 
166 #define	VNDLABELDEV(dev) \
167 	(MAKEDISKDEV(major((dev)), vndunit((dev)), RAW_PART))
168 
169 /* called by main() at boot time */
170 void	vndattach __P((int));
171 void	vnddetach __P((void));
172 
173 void	vndclear __P((struct vnd_softc *));
174 void	vndstart __P((struct vnd_softc *));
175 int	vndsetcred __P((struct vnd_softc *, struct ucred *));
176 void	vndthrottle __P((struct vnd_softc *, struct vnode *));
177 void	vndiodone __P((struct buf *));
178 void	vndshutdown __P((void));
179 
180 void	vndgetdefaultlabel __P((struct vnd_softc *, struct disklabel *));
181 void	vndgetdisklabel __P((dev_t));
182 
183 static	int vndlock __P((struct vnd_softc *));
184 static	void vndunlock __P((struct vnd_softc *));
185 
186 void
187 vndattach(num)
188 	int num;
189 {
190 	int i;
191 	char *mem;
192 
193 	if (num <= 0)
194 		return;
195 	i = num * sizeof(struct vnd_softc);
196 	mem = malloc(i, M_DEVBUF, M_NOWAIT|M_ZERO);
197 	if (mem == NULL) {
198 		printf("WARNING: no memory for vnode disks\n");
199 		return;
200 	}
201 	vnd_softc = (struct vnd_softc *)mem;
202 	numvnd = num;
203 
204 	for (i = 0; i < numvnd; i++)
205 		BUFQ_INIT(&vnd_softc[i].sc_tab);
206 }
207 
208 void
209 vnddetach()
210 {
211 
212 	free(vnd_softc, M_DEVBUF);
213 }
214 
215 int
216 vndopen(dev, flags, mode, p)
217 	dev_t dev;
218 	int flags, mode;
219 	struct proc *p;
220 {
221 	int unit = vndunit(dev);
222 	struct vnd_softc *sc;
223 	int error = 0, part, pmask;
224 	struct disklabel *lp;
225 
226 #ifdef DEBUG
227 	if (vnddebug & VDB_FOLLOW)
228 		printf("vndopen(0x%x, 0x%x, 0x%x, %p)\n", dev, flags, mode, p);
229 #endif
230 	if (unit >= numvnd)
231 		return (ENXIO);
232 	sc = &vnd_softc[unit];
233 
234 	if ((error = vndlock(sc)) != 0)
235 		return (error);
236 
237 	lp = sc->sc_dkdev.dk_label;
238 
239 	part = DISKPART(dev);
240 	pmask = (1 << part);
241 
242 	/*
243 	 * If we're initialized, check to see if there are any other
244 	 * open partitions.  If not, then it's safe to update the
245 	 * in-core disklabel.
246 	 */
247 	if ((sc->sc_flags & VNF_INITED) && (sc->sc_dkdev.dk_openmask == 0))
248 		vndgetdisklabel(dev);
249 
250 	/* Check that the partitions exists. */
251 	if (part != RAW_PART) {
252 		if (((sc->sc_flags & VNF_INITED) == 0) ||
253 		    ((part >= lp->d_npartitions) ||
254 		     (lp->d_partitions[part].p_fstype == FS_UNUSED))) {
255 			error = ENXIO;
256 			goto done;
257 		}
258 	}
259 
260 	/* Prevent our unit from being unconfigured while open. */
261 	switch (mode) {
262 	case S_IFCHR:
263 		sc->sc_dkdev.dk_copenmask |= pmask;
264 		break;
265 
266 	case S_IFBLK:
267 		sc->sc_dkdev.dk_bopenmask |= pmask;
268 		break;
269 	}
270 	sc->sc_dkdev.dk_openmask =
271 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
272 
273  done:
274 	vndunlock(sc);
275 	return (error);
276 }
277 
278 int
279 vndclose(dev, flags, mode, p)
280 	dev_t dev;
281 	int flags, mode;
282 	struct proc *p;
283 {
284 	int unit = vndunit(dev);
285 	struct vnd_softc *sc;
286 	int error = 0, part;
287 
288 #ifdef DEBUG
289 	if (vnddebug & VDB_FOLLOW)
290 		printf("vndclose(0x%x, 0x%x, 0x%x, %p)\n", dev, flags, mode, p);
291 #endif
292 
293 	if (unit >= numvnd)
294 		return (ENXIO);
295 	sc = &vnd_softc[unit];
296 
297 	if ((error = vndlock(sc)) != 0)
298 		return (error);
299 
300 	part = DISKPART(dev);
301 
302 	/* ...that much closer to allowing unconfiguration... */
303 	switch (mode) {
304 	case S_IFCHR:
305 		sc->sc_dkdev.dk_copenmask &= ~(1 << part);
306 		break;
307 
308 	case S_IFBLK:
309 		sc->sc_dkdev.dk_bopenmask &= ~(1 << part);
310 		break;
311 	}
312 	sc->sc_dkdev.dk_openmask =
313 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
314 
315 	vndunlock(sc);
316 	return (0);
317 }
318 
319 /*
320  * Break the request into bsize pieces and submit using VOP_BMAP/VOP_STRATEGY.
321  */
322 void
323 vndstrategy(bp)
324 	struct buf *bp;
325 {
326 	int unit = vndunit(bp->b_dev);
327 	struct vnd_softc *vnd = &vnd_softc[unit];
328 	struct vndxfer *vnx;
329 	int s, bsize, resid;
330 	off_t bn;
331 	caddr_t addr;
332 	int sz, flags, error, wlabel;
333 	struct disklabel *lp;
334 	struct partition *pp;
335 
336 #ifdef DEBUG
337 	if (vnddebug & VDB_FOLLOW)
338 		printf("vndstrategy(%p): unit %d\n", bp, unit);
339 #endif
340 	if ((vnd->sc_flags & VNF_INITED) == 0) {
341 		bp->b_error = ENXIO;
342 		bp->b_flags |= B_ERROR;
343 		goto done;
344 	}
345 
346 	/* If it's a nil transfer, wake up the top half now. */
347 	if (bp->b_bcount == 0)
348 		goto done;
349 
350 	lp = vnd->sc_dkdev.dk_label;
351 
352 	/*
353 	 * The transfer must be a whole number of blocks.
354 	 */
355 	if ((bp->b_bcount % lp->d_secsize) != 0) {
356 		bp->b_error = EINVAL;
357 		bp->b_flags |= B_ERROR;
358 		goto done;
359 	}
360 
361 	/*
362 	 * Do bounds checking and adjust transfer.  If there's an error,
363 	 * the bounds check will flag that for us.
364 	 */
365 	wlabel = vnd->sc_flags & (VNF_WLABEL|VNF_LABELLING);
366 	if (DISKPART(bp->b_dev) != RAW_PART)
367 		if (bounds_check_with_label(bp, lp, wlabel) <= 0)
368 			goto done;
369 
370 	bp->b_resid = bp->b_bcount;
371 
372 	/*
373 	 * Put the block number in terms of the logical blocksize
374 	 * of the "device".
375 	 */
376 	bn = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
377 
378 	/*
379 	 * Translate the partition-relative block number to an absolute.
380 	 */
381 	if (DISKPART(bp->b_dev) != RAW_PART) {
382 		pp = &vnd->sc_dkdev.dk_label->d_partitions[DISKPART(bp->b_dev)];
383 		bn += pp->p_offset;
384 	}
385 
386 	/* ...and convert to a byte offset within the file. */
387 	bn *= lp->d_secsize;
388 
389 	if (vnd->sc_vp->v_mount == NULL) {
390 		bp->b_error = ENXIO;
391 		bp->b_flags |= B_ERROR;
392 		goto done;
393 	}
394  	bsize = vnd->sc_vp->v_mount->mnt_stat.f_iosize;
395 	addr = bp->b_data;
396 	flags = (bp->b_flags & (B_READ|B_ASYNC)) | B_CALL;
397 
398 	/* Allocate a header for this transfer and link it to the buffer */
399 	s = splbio();
400 	vnx = VND_GETXFER(vnd);
401 	splx(s);
402 	vnx->vx_flags = VX_BUSY;
403 	vnx->vx_error = 0;
404 	vnx->vx_pending = 0;
405 	vnx->vx_bp = bp;
406 
407 	for (resid = bp->b_resid; resid; resid -= sz) {
408 		struct vndbuf *nbp;
409 		struct vnode *vp;
410 		daddr_t nbn;
411 		int off, nra;
412 
413 		nra = 0;
414 		vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY | LK_CANRECURSE);
415 		error = VOP_BMAP(vnd->sc_vp, bn / bsize, &vp, &nbn, &nra);
416 		VOP_UNLOCK(vnd->sc_vp, 0);
417 
418 		if (error == 0 && (long)nbn == -1)
419 			error = EIO;
420 
421 		/*
422 		 * If there was an error or a hole in the file...punt.
423 		 * Note that we may have to wait for any operations
424 		 * that we have already fired off before releasing
425 		 * the buffer.
426 		 *
427 		 * XXX we could deal with holes here but it would be
428 		 * a hassle (in the write case).
429 		 */
430 		if (error) {
431 			s = splbio();
432 			vnx->vx_error = error;
433 			goto out;
434 		}
435 
436 #ifdef DEBUG
437 		if (!dovndcluster)
438 			nra = 0;
439 #endif
440 
441 		if ((off = bn % bsize) != 0)
442 			sz = bsize - off;
443 		else
444 			sz = (1 + nra) * bsize;
445 		if (resid < sz)
446 			sz = resid;
447 #ifdef DEBUG
448 		if (vnddebug & VDB_IO)
449 			printf("vndstrategy: vp %p/%p bn 0x%qx/0x%x sz 0x%x\n",
450 			    vnd->sc_vp, vp, (long long)bn, nbn, sz);
451 #endif
452 
453 		s = splbio();
454 		nbp = VND_GETBUF(vnd);
455 		splx(s);
456 		nbp->vb_buf.b_flags = flags;
457 		nbp->vb_buf.b_bcount = sz;
458 		nbp->vb_buf.b_bufsize = bp->b_bufsize;
459 		nbp->vb_buf.b_error = 0;
460 		nbp->vb_buf.b_data = addr;
461 		nbp->vb_buf.b_blkno = nbp->vb_buf.b_rawblkno = nbn + btodb(off);
462 		nbp->vb_buf.b_proc = bp->b_proc;
463 		nbp->vb_buf.b_iodone = vndiodone;
464 		nbp->vb_buf.b_vp = NULLVP;
465 		LIST_INIT(&nbp->vb_buf.b_dep);
466 
467 		nbp->vb_xfer = vnx;
468 
469 		/*
470 		 * Just sort by block number
471 		 */
472 		s = splbio();
473 		if (vnx->vx_error != 0) {
474 			VND_PUTBUF(vnd, nbp);
475 			goto out;
476 		}
477 		vnx->vx_pending++;
478 		bgetvp(vp, &nbp->vb_buf);
479 		disksort_blkno(&vnd->sc_tab, &nbp->vb_buf);
480 		vndstart(vnd);
481 		splx(s);
482 		bn += sz;
483 		addr += sz;
484 	}
485 
486 	s = splbio();
487 
488 out: /* Arrive here at splbio */
489 	vnx->vx_flags &= ~VX_BUSY;
490 	if (vnx->vx_pending == 0) {
491 		if (vnx->vx_error != 0) {
492 			bp->b_error = vnx->vx_error;
493 			bp->b_flags |= B_ERROR;
494 		}
495 		VND_PUTXFER(vnd, vnx);
496 		biodone(bp);
497 	}
498 	splx(s);
499 	return;
500 
501  done:
502 	biodone(bp);
503 }
504 
505 /*
506  * Feed requests sequentially.
507  * We do it this way to keep from flooding NFS servers if we are connected
508  * to an NFS file.  This places the burden on the client rather than the
509  * server.
510  */
511 void
512 vndstart(vnd)
513 	struct vnd_softc *vnd;
514 {
515 	struct buf	*bp;
516 
517 	/*
518 	 * Dequeue now since lower level strategy routine might
519 	 * queue using same links
520 	 */
521 
522 	if ((vnd->sc_flags & VNF_BUSY) != 0)
523 		return;
524 
525 	vnd->sc_flags |= VNF_BUSY;
526 
527 	while (vnd->sc_active < vnd->sc_maxactive) {
528 		bp = BUFQ_FIRST(&vnd->sc_tab);
529 		if (bp == NULL)
530 			break;
531 		BUFQ_REMOVE(&vnd->sc_tab, bp);
532 		vnd->sc_active++;
533 #ifdef DEBUG
534 		if (vnddebug & VDB_IO)
535 			printf("vndstart(%ld): bp %p vp %p blkno 0x%x"
536 				" flags %lx addr %p cnt 0x%lx\n",
537 			    (long) (vnd-vnd_softc), bp, bp->b_vp, bp->b_blkno,
538 			    bp->b_flags, bp->b_data, bp->b_bcount);
539 #endif
540 
541 		/* Instrumentation. */
542 		disk_busy(&vnd->sc_dkdev);
543 
544 		if ((bp->b_flags & B_READ) == 0)
545 			bp->b_vp->v_numoutput++;
546 		VOP_STRATEGY(bp);
547 	}
548 	vnd->sc_flags &= ~VNF_BUSY;
549 }
550 
551 void
552 vndiodone(bp)
553 	struct buf *bp;
554 {
555 	struct vndbuf *vbp = (struct vndbuf *) bp;
556 	struct vndxfer *vnx = (struct vndxfer *)vbp->vb_xfer;
557 	struct buf *pbp = vnx->vx_bp;
558 	struct vnd_softc *vnd = &vnd_softc[vndunit(pbp->b_dev)];
559 	int s, resid;
560 
561 	s = splbio();
562 #ifdef DEBUG
563 	if (vnddebug & VDB_IO)
564 		printf("vndiodone(%ld): vbp %p vp %p blkno 0x%x addr %p cnt 0x%lx\n",
565 		    (long) (vnd-vnd_softc), vbp, vbp->vb_buf.b_vp,
566 		    vbp->vb_buf.b_blkno, vbp->vb_buf.b_data,
567 		    vbp->vb_buf.b_bcount);
568 #endif
569 
570 	resid = vbp->vb_buf.b_bcount - vbp->vb_buf.b_resid;
571 	pbp->b_resid -= resid;
572 	disk_unbusy(&vnd->sc_dkdev, resid);
573 	vnx->vx_pending--;
574 
575 	if (vbp->vb_buf.b_error) {
576 #ifdef DEBUG
577 		if (vnddebug & VDB_IO)
578 			printf("vndiodone: vbp %p error %d\n", vbp,
579 			    vbp->vb_buf.b_error);
580 #endif
581 		vnx->vx_error = vbp->vb_buf.b_error;
582 	}
583 
584 	if (vbp->vb_buf.b_vp != NULLVP)
585 		brelvp(&vbp->vb_buf);
586 
587 	VND_PUTBUF(vnd, vbp);
588 
589 	/*
590 	 * Wrap up this transaction if it has run to completion or, in
591 	 * case of an error, when all auxiliary buffers have returned.
592 	 */
593 	if (vnx->vx_error != 0) {
594 		pbp->b_flags |= B_ERROR;
595 		pbp->b_error = vnx->vx_error;
596 		if ((vnx->vx_flags & VX_BUSY) == 0 && vnx->vx_pending == 0) {
597 
598 #ifdef DEBUG
599 			if (vnddebug & VDB_IO)
600 				printf("vndiodone: pbp %p iodone: error %d\n",
601 					pbp, vnx->vx_error);
602 #endif
603 			VND_PUTXFER(vnd, vnx);
604 			biodone(pbp);
605 		}
606 	} else if (pbp->b_resid == 0) {
607 
608 #ifdef DIAGNOSTIC
609 		if (vnx->vx_pending != 0)
610 			panic("vndiodone: vnx pending: %d", vnx->vx_pending);
611 #endif
612 
613 		if ((vnx->vx_flags & VX_BUSY) == 0) {
614 #ifdef DEBUG
615 			if (vnddebug & VDB_IO)
616 				printf("vndiodone: pbp %p iodone\n", pbp);
617 #endif
618 			VND_PUTXFER(vnd, vnx);
619 			biodone(pbp);
620 		}
621 	}
622 
623 	vnd->sc_active--;
624 	vndstart(vnd);
625 	splx(s);
626 }
627 
628 /* ARGSUSED */
629 int
630 vndread(dev, uio, flags)
631 	dev_t dev;
632 	struct uio *uio;
633 	int flags;
634 {
635 	int unit = vndunit(dev);
636 	struct vnd_softc *sc;
637 
638 #ifdef DEBUG
639 	if (vnddebug & VDB_FOLLOW)
640 		printf("vndread(0x%x, %p)\n", dev, uio);
641 #endif
642 
643 	if (unit >= numvnd)
644 		return (ENXIO);
645 	sc = &vnd_softc[unit];
646 
647 	if ((sc->sc_flags & VNF_INITED) == 0)
648 		return (ENXIO);
649 
650 	return (physio(vndstrategy, NULL, dev, B_READ, minphys, uio));
651 }
652 
653 /* ARGSUSED */
654 int
655 vndwrite(dev, uio, flags)
656 	dev_t dev;
657 	struct uio *uio;
658 	int flags;
659 {
660 	int unit = vndunit(dev);
661 	struct vnd_softc *sc;
662 
663 #ifdef DEBUG
664 	if (vnddebug & VDB_FOLLOW)
665 		printf("vndwrite(0x%x, %p)\n", dev, uio);
666 #endif
667 
668 	if (unit >= numvnd)
669 		return (ENXIO);
670 	sc = &vnd_softc[unit];
671 
672 	if ((sc->sc_flags & VNF_INITED) == 0)
673 		return (ENXIO);
674 
675 	return (physio(vndstrategy, NULL, dev, B_WRITE, minphys, uio));
676 }
677 
678 /* ARGSUSED */
679 int
680 vndioctl(dev, cmd, data, flag, p)
681 	dev_t dev;
682 	u_long cmd;
683 	caddr_t data;
684 	int flag;
685 	struct proc *p;
686 {
687 	int unit = vndunit(dev);
688 	struct vnd_softc *vnd;
689 	struct vnd_ioctl *vio;
690 	struct vattr vattr;
691 	struct nameidata nd;
692 	int error, part, pmask;
693 	size_t geomsize;
694 #ifdef __HAVE_OLD_DISKLABEL
695 	struct disklabel newlabel;
696 #endif
697 
698 #ifdef DEBUG
699 	if (vnddebug & VDB_FOLLOW)
700 		printf("vndioctl(0x%x, 0x%lx, %p, 0x%x, %p): unit %d\n",
701 		    dev, cmd, data, flag, p, unit);
702 #endif
703 	error = suser(p->p_ucred, &p->p_acflag);
704 	if (error)
705 		return (error);
706 	if (unit >= numvnd)
707 		return (ENXIO);
708 
709 	vnd = &vnd_softc[unit];
710 	vio = (struct vnd_ioctl *)data;
711 
712 	/* Must be open for writes for these commands... */
713 	switch (cmd) {
714 	case VNDIOCSET:
715 	case VNDIOCCLR:
716 	case DIOCSDINFO:
717 	case DIOCWDINFO:
718 #ifdef __HAVE_OLD_DISKLABEL
719 	case ODIOCSDINFO:
720 	case ODIOCWDINFO:
721 #endif
722 	case DIOCWLABEL:
723 		if ((flag & FWRITE) == 0)
724 			return (EBADF);
725 	}
726 
727 	/* Must be initialized for these... */
728 	switch (cmd) {
729 	case VNDIOCCLR:
730 	case DIOCGDINFO:
731 	case DIOCSDINFO:
732 	case DIOCWDINFO:
733 	case DIOCGPART:
734 	case DIOCWLABEL:
735 	case DIOCGDEFLABEL:
736 #ifdef __HAVE_OLD_DISKLABEL
737 	case ODIOCGDINFO:
738 	case ODIOCSDINFO:
739 	case ODIOCWDINFO:
740 	case ODIOCGDEFLABEL:
741 #endif
742 		if ((vnd->sc_flags & VNF_INITED) == 0)
743 			return (ENXIO);
744 	}
745 
746 	switch (cmd) {
747 	case VNDIOCSET:
748 		if (vnd->sc_flags & VNF_INITED)
749 			return (EBUSY);
750 
751 		if ((error = vndlock(vnd)) != 0)
752 			return (error);
753 
754 		/*
755 		 * Always open for read and write.
756 		 * This is probably bogus, but it lets vn_open()
757 		 * weed out directories, sockets, etc. so we don't
758 		 * have to worry about them.
759 		 */
760 		NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, vio->vnd_file, p);
761 		if ((error = vn_open(&nd, FREAD|FWRITE, 0)) != 0) {
762 			vndunlock(vnd);
763 			return(error);
764 		}
765 		error = VOP_GETATTR(nd.ni_vp, &vattr, p->p_ucred, p);
766 		if (error) {
767 			VOP_UNLOCK(nd.ni_vp, 0);
768 			(void) vn_close(nd.ni_vp, FREAD|FWRITE, p->p_ucred, p);
769 			vndunlock(vnd);
770 			return(error);
771 		}
772 		VOP_UNLOCK(nd.ni_vp, 0);
773 		vnd->sc_vp = nd.ni_vp;
774 		vnd->sc_size = btodb(vattr.va_size);	/* note truncation */
775 
776 		/*
777 		 * Use pseudo-geometry specified.  If none was provided,
778 		 * use "standard" Adaptec fictitious geometry.
779 		 */
780 		if (vio->vnd_flags & VNDIOF_HASGEOM) {
781 
782 			memcpy(&vnd->sc_geom, &vio->vnd_geom,
783 			    sizeof(vio->vnd_geom));
784 
785 			/*
786 			 * Sanity-check the sector size.
787 			 * XXX Don't allow secsize < DEV_BSIZE.  Should
788 			 * XXX we?
789 			 */
790 			if (vnd->sc_geom.vng_secsize < DEV_BSIZE ||
791 			    (vnd->sc_geom.vng_secsize % DEV_BSIZE) != 0) {
792 				(void) vn_close(nd.ni_vp, FREAD|FWRITE,
793 				    p->p_ucred, p);
794 				vndunlock(vnd);
795 				return (EINVAL);
796 			}
797 
798 			/*
799 			 * Compute the size (in DEV_BSIZE blocks) specified
800 			 * by the geometry.
801 			 */
802 			geomsize = (vnd->sc_geom.vng_nsectors *
803 			    vnd->sc_geom.vng_ntracks *
804 			    vnd->sc_geom.vng_ncylinders) *
805 			    (vnd->sc_geom.vng_secsize / DEV_BSIZE);
806 
807 			/*
808 			 * Sanity-check the size against the specified
809 			 * geometry.
810 			 */
811 			if (vnd->sc_size < geomsize) {
812 				(void) vn_close(nd.ni_vp, FREAD|FWRITE,
813 				    p->p_ucred, p);
814 				vndunlock(vnd);
815 				return (EINVAL);
816 			}
817 		} else {
818 			/*
819 			 * Size must be at least 2048 DEV_BSIZE blocks
820 			 * (1M) in order to use this geometry.
821 			 */
822 			if (vnd->sc_size < (32 * 64)) {
823 				vndunlock(vnd);
824 				return (EINVAL);
825 			}
826 
827 			vnd->sc_geom.vng_secsize = DEV_BSIZE;
828 			vnd->sc_geom.vng_nsectors = 32;
829 			vnd->sc_geom.vng_ntracks = 64;
830 			vnd->sc_geom.vng_ncylinders = vnd->sc_size / (64 * 32);
831 
832 			/*
833 			 * Compute the actual size allowed by this geometry.
834 			 */
835 			geomsize = 32 * 64 * vnd->sc_geom.vng_ncylinders;
836 		}
837 
838 		/*
839 		 * Truncate the size to that specified by
840 		 * the geometry.
841 		 * XXX Should we even bother with this?
842 		 */
843 		vnd->sc_size = geomsize;
844 
845 		if ((error = vndsetcred(vnd, p->p_ucred)) != 0) {
846 			(void) vn_close(nd.ni_vp, FREAD|FWRITE, p->p_ucred, p);
847 			vndunlock(vnd);
848 			return(error);
849 		}
850 		vndthrottle(vnd, vnd->sc_vp);
851 		vio->vnd_size = dbtob(vnd->sc_size);
852 		vnd->sc_flags |= VNF_INITED;
853 #ifdef DEBUG
854 		if (vnddebug & VDB_INIT)
855 			printf("vndioctl: SET vp %p size 0x%lx %d/%d/%d/%d\n",
856 			    vnd->sc_vp, (unsigned long) vnd->sc_size,
857 			    vnd->sc_geom.vng_secsize,
858 			    vnd->sc_geom.vng_nsectors,
859 			    vnd->sc_geom.vng_ntracks,
860 			    vnd->sc_geom.vng_ncylinders);
861 #endif
862 
863 		/* Attach the disk. */
864 		memset(vnd->sc_xname, 0, sizeof(vnd->sc_xname)); /* XXX */
865 		sprintf(vnd->sc_xname, "vnd%d", unit);		/* XXX */
866 		vnd->sc_dkdev.dk_name = vnd->sc_xname;
867 		disk_attach(&vnd->sc_dkdev);
868 
869 		/* Initialize the xfer and buffer pools. */
870 		pool_init(&vnd->sc_vxpool, sizeof(struct vndxfer), 0,
871 		    0, 0, "vndxpl", 0, NULL, NULL, M_DEVBUF);
872 		pool_init(&vnd->sc_vbpool, sizeof(struct vndbuf), 0,
873 		    0, 0, "vndbpl", 0, NULL, NULL, M_DEVBUF);
874 
875 		/* Try and read the disklabel. */
876 		vndgetdisklabel(dev);
877 
878 		vndunlock(vnd);
879 
880 		break;
881 
882 	case VNDIOCCLR:
883 		if ((error = vndlock(vnd)) != 0)
884 			return (error);
885 
886 		/*
887 		 * Don't unconfigure if any other partitions are open
888 		 * or if both the character and block flavors of this
889 		 * partition are open.
890 		 */
891 		part = DISKPART(dev);
892 		pmask = (1 << part);
893 		if ((vnd->sc_dkdev.dk_openmask & ~pmask) ||
894 		    ((vnd->sc_dkdev.dk_bopenmask & pmask) &&
895 		    (vnd->sc_dkdev.dk_copenmask & pmask))) {
896 			vndunlock(vnd);
897 			return (EBUSY);
898 		}
899 
900 		vndclear(vnd);
901 #ifdef DEBUG
902 		if (vnddebug & VDB_INIT)
903 			printf("vndioctl: CLRed\n");
904 #endif
905 
906 		/* Destroy the xfer and buffer pools. */
907 		pool_destroy(&vnd->sc_vxpool);
908 		pool_destroy(&vnd->sc_vbpool);
909 
910 		/* Detatch the disk. */
911 		disk_detach(&vnd->sc_dkdev);
912 
913 		vndunlock(vnd);
914 
915 		break;
916 
917 	case DIOCGDINFO:
918 		*(struct disklabel *)data = *(vnd->sc_dkdev.dk_label);
919 		break;
920 
921 #ifdef __HAVE_OLD_DISKLABEL
922 	case ODIOCGDINFO:
923 		newlabel = *(vnd->sc_dkdev.dk_label);
924 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
925 			return ENOTTY;
926 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
927 		break;
928 #endif
929 
930 	case DIOCGPART:
931 		((struct partinfo *)data)->disklab = vnd->sc_dkdev.dk_label;
932 		((struct partinfo *)data)->part =
933 		    &vnd->sc_dkdev.dk_label->d_partitions[DISKPART(dev)];
934 		break;
935 
936 	case DIOCWDINFO:
937 	case DIOCSDINFO:
938 #ifdef __HAVE_OLD_DISKLABEL
939 	case ODIOCWDINFO:
940 	case ODIOCSDINFO:
941 #endif
942 	{
943 		struct disklabel *lp;
944 
945 		if ((error = vndlock(vnd)) != 0)
946 			return (error);
947 
948 		vnd->sc_flags |= VNF_LABELLING;
949 
950 #ifdef __HAVE_OLD_DISKLABEL
951 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
952 			memset(&newlabel, 0, sizeof newlabel);
953 			memcpy(&newlabel, data, sizeof (struct olddisklabel));
954 			lp = &newlabel;
955 		} else
956 #endif
957 		lp = (struct disklabel *)data;
958 
959 		error = setdisklabel(vnd->sc_dkdev.dk_label,
960 		    lp, 0, vnd->sc_dkdev.dk_cpulabel);
961 		if (error == 0) {
962 			if (cmd == DIOCWDINFO
963 #ifdef __HAVE_OLD_DISKLABEL
964 			    || cmd == ODIOCWDINFO
965 #endif
966 			   )
967 				error = writedisklabel(VNDLABELDEV(dev),
968 				    vndstrategy, vnd->sc_dkdev.dk_label,
969 				    vnd->sc_dkdev.dk_cpulabel);
970 		}
971 
972 		vnd->sc_flags &= ~VNF_LABELLING;
973 
974 		vndunlock(vnd);
975 
976 		if (error)
977 			return (error);
978 		break;
979 	}
980 
981 	case DIOCWLABEL:
982 		if (*(int *)data != 0)
983 			vnd->sc_flags |= VNF_WLABEL;
984 		else
985 			vnd->sc_flags &= ~VNF_WLABEL;
986 		break;
987 
988 	case DIOCGDEFLABEL:
989 		vndgetdefaultlabel(vnd, (struct disklabel *)data);
990 		break;
991 
992 #ifdef __HAVE_OLD_DISKLABEL
993 	case ODIOCGDEFLABEL:
994 		vndgetdefaultlabel(vnd, &newlabel);
995 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
996 			return ENOTTY;
997 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
998 		break;
999 #endif
1000 
1001 	default:
1002 		return (ENOTTY);
1003 	}
1004 
1005 	return (0);
1006 }
1007 
1008 /*
1009  * Duplicate the current processes' credentials.  Since we are called only
1010  * as the result of a SET ioctl and only root can do that, any future access
1011  * to this "disk" is essentially as root.  Note that credentials may change
1012  * if some other uid can write directly to the mapped file (NFS).
1013  */
1014 int
1015 vndsetcred(vnd, cred)
1016 	struct vnd_softc *vnd;
1017 	struct ucred *cred;
1018 {
1019 	struct uio auio;
1020 	struct iovec aiov;
1021 	char *tmpbuf;
1022 	int error;
1023 
1024 	vnd->sc_cred = crdup(cred);
1025 	tmpbuf = malloc(DEV_BSIZE, M_TEMP, M_WAITOK);
1026 
1027 	/* XXX: Horrible kludge to establish credentials for NFS */
1028 	aiov.iov_base = tmpbuf;
1029 	aiov.iov_len = min(DEV_BSIZE, dbtob(vnd->sc_size));
1030 	auio.uio_iov = &aiov;
1031 	auio.uio_iovcnt = 1;
1032 	auio.uio_offset = 0;
1033 	auio.uio_rw = UIO_READ;
1034 	auio.uio_segflg = UIO_SYSSPACE;
1035 	auio.uio_resid = aiov.iov_len;
1036 	vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1037 	error = VOP_READ(vnd->sc_vp, &auio, 0, vnd->sc_cred);
1038 	if (error == 0) {
1039 		/*
1040 		 * Because vnd does all IO directly through the vnode
1041 		 * we need to flush (at least) the buffer from the above
1042 		 * VOP_READ from the buffer cache to prevent cache
1043 		 * incoherencies.  Also, be careful to write dirty
1044 		 * buffers back to stable storage.
1045 		 */
1046 		error = vinvalbuf(vnd->sc_vp, V_SAVE, vnd->sc_cred,
1047 			    curproc, 0, 0);
1048 	}
1049 	VOP_UNLOCK(vnd->sc_vp, 0);
1050 
1051 	free(tmpbuf, M_TEMP);
1052 	return (error);
1053 }
1054 
1055 /*
1056  * Set maxactive based on FS type
1057  */
1058 void
1059 vndthrottle(vnd, vp)
1060 	struct vnd_softc *vnd;
1061 	struct vnode *vp;
1062 {
1063 #ifdef NFS
1064 	extern int (**nfsv2_vnodeop_p) __P((void *));
1065 
1066 	if (vp->v_op == nfsv2_vnodeop_p)
1067 		vnd->sc_maxactive = 2;
1068 	else
1069 #endif
1070 		vnd->sc_maxactive = 8;
1071 
1072 	if (vnd->sc_maxactive < 1)
1073 		vnd->sc_maxactive = 1;
1074 }
1075 
1076 void
1077 vndshutdown()
1078 {
1079 	struct vnd_softc *vnd;
1080 
1081 	for (vnd = &vnd_softc[0]; vnd < &vnd_softc[numvnd]; vnd++)
1082 		if (vnd->sc_flags & VNF_INITED)
1083 			vndclear(vnd);
1084 }
1085 
1086 void
1087 vndclear(vnd)
1088 	struct vnd_softc *vnd;
1089 {
1090 	struct vnode *vp = vnd->sc_vp;
1091 	struct proc *p = curproc;		/* XXX */
1092 
1093 #ifdef DEBUG
1094 	if (vnddebug & VDB_FOLLOW)
1095 		printf("vndclear(%p): vp %p\n", vnd, vp);
1096 #endif
1097 	vnd->sc_flags &= ~VNF_INITED;
1098 	if (vp == (struct vnode *)0)
1099 		panic("vndioctl: null vp");
1100 	(void) vn_close(vp, FREAD|FWRITE, vnd->sc_cred, p);
1101 	crfree(vnd->sc_cred);
1102 	vnd->sc_vp = (struct vnode *)0;
1103 	vnd->sc_cred = (struct ucred *)0;
1104 	vnd->sc_size = 0;
1105 }
1106 
1107 int
1108 vndsize(dev)
1109 	dev_t dev;
1110 {
1111 	struct vnd_softc *sc;
1112 	struct disklabel *lp;
1113 	int part, unit, omask;
1114 	int size;
1115 
1116 	unit = vndunit(dev);
1117 	if (unit >= numvnd)
1118 		return (-1);
1119 	sc = &vnd_softc[unit];
1120 
1121 	if ((sc->sc_flags & VNF_INITED) == 0)
1122 		return (-1);
1123 
1124 	part = DISKPART(dev);
1125 	omask = sc->sc_dkdev.dk_openmask & (1 << part);
1126 	lp = sc->sc_dkdev.dk_label;
1127 
1128 	if (omask == 0 && vndopen(dev, 0, S_IFBLK, curproc))
1129 		return (-1);
1130 
1131 	if (lp->d_partitions[part].p_fstype != FS_SWAP)
1132 		size = -1;
1133 	else
1134 		size = lp->d_partitions[part].p_size *
1135 		    (lp->d_secsize / DEV_BSIZE);
1136 
1137 	if (omask == 0 && vndclose(dev, 0, S_IFBLK, curproc))
1138 		return (-1);
1139 
1140 	return (size);
1141 }
1142 
1143 int
1144 vnddump(dev, blkno, va, size)
1145 	dev_t dev;
1146 	daddr_t blkno;
1147 	caddr_t va;
1148 	size_t size;
1149 {
1150 
1151 	/* Not implemented. */
1152 	return ENXIO;
1153 }
1154 
1155 void
1156 vndgetdefaultlabel(sc, lp)
1157 	struct vnd_softc *sc;
1158 	struct disklabel *lp;
1159 {
1160 	struct vndgeom *vng = &sc->sc_geom;
1161 	struct partition *pp;
1162 
1163 	memset(lp, 0, sizeof(*lp));
1164 
1165 	lp->d_secperunit = sc->sc_size / (vng->vng_secsize / DEV_BSIZE);
1166 	lp->d_secsize = vng->vng_secsize;
1167 	lp->d_nsectors = vng->vng_nsectors;
1168 	lp->d_ntracks = vng->vng_ntracks;
1169 	lp->d_ncylinders = vng->vng_ncylinders;
1170 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1171 
1172 	strncpy(lp->d_typename, "vnd", sizeof(lp->d_typename));
1173 	lp->d_type = DTYPE_VND;
1174 	strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
1175 	lp->d_rpm = 3600;
1176 	lp->d_interleave = 1;
1177 	lp->d_flags = 0;
1178 
1179 	pp = &lp->d_partitions[RAW_PART];
1180 	pp->p_offset = 0;
1181 	pp->p_size = lp->d_secperunit;
1182 	pp->p_fstype = FS_UNUSED;
1183 	lp->d_npartitions = RAW_PART + 1;
1184 
1185 	lp->d_magic = DISKMAGIC;
1186 	lp->d_magic2 = DISKMAGIC;
1187 	lp->d_checksum = dkcksum(lp);
1188 }
1189 
1190 /*
1191  * Read the disklabel from a vnd.  If one is not present, create a fake one.
1192  */
1193 void
1194 vndgetdisklabel(dev)
1195 	dev_t dev;
1196 {
1197 	struct vnd_softc *sc = &vnd_softc[vndunit(dev)];
1198 	char *errstring;
1199 	struct disklabel *lp = sc->sc_dkdev.dk_label;
1200 	struct cpu_disklabel *clp = sc->sc_dkdev.dk_cpulabel;
1201 	int i;
1202 
1203 	memset(clp, 0, sizeof(*clp));
1204 
1205 	vndgetdefaultlabel(sc, lp);
1206 
1207 	/*
1208 	 * Call the generic disklabel extraction routine.
1209 	 */
1210 	errstring = readdisklabel(VNDLABELDEV(dev), vndstrategy, lp, clp);
1211 	if (errstring) {
1212 		/*
1213 		 * Lack of disklabel is common, but we print the warning
1214 		 * anyway, since it might contain other useful information.
1215 		 */
1216 		printf("%s: %s\n", sc->sc_xname, errstring);
1217 
1218 		/*
1219 		 * For historical reasons, if there's no disklabel
1220 		 * present, all partitions must be FS_BSDFFS and
1221 		 * occupy the entire disk.
1222 		 */
1223 		for (i = 0; i < MAXPARTITIONS; i++) {
1224 			/*
1225 			 * Don't wipe out port specific hack (such as
1226 			 * dos partition hack of i386 port).
1227 			 */
1228 			if (lp->d_partitions[i].p_fstype != FS_UNUSED)
1229 				continue;
1230 
1231 			lp->d_partitions[i].p_size = lp->d_secperunit;
1232 			lp->d_partitions[i].p_offset = 0;
1233 			lp->d_partitions[i].p_fstype = FS_BSDFFS;
1234 		}
1235 
1236 		strncpy(lp->d_packname, "default label",
1237 		    sizeof(lp->d_packname));
1238 
1239 		lp->d_checksum = dkcksum(lp);
1240 	}
1241 }
1242 
1243 /*
1244  * Wait interruptibly for an exclusive lock.
1245  *
1246  * XXX
1247  * Several drivers do this; it should be abstracted and made MP-safe.
1248  */
1249 static int
1250 vndlock(sc)
1251 	struct vnd_softc *sc;
1252 {
1253 	int error;
1254 
1255 	while ((sc->sc_flags & VNF_LOCKED) != 0) {
1256 		sc->sc_flags |= VNF_WANTED;
1257 		if ((error = tsleep(sc, PRIBIO | PCATCH, "vndlck", 0)) != 0)
1258 			return (error);
1259 	}
1260 	sc->sc_flags |= VNF_LOCKED;
1261 	return (0);
1262 }
1263 
1264 /*
1265  * Unlock and wake up any waiters.
1266  */
1267 static void
1268 vndunlock(sc)
1269 	struct vnd_softc *sc;
1270 {
1271 
1272 	sc->sc_flags &= ~VNF_LOCKED;
1273 	if ((sc->sc_flags & VNF_WANTED) != 0) {
1274 		sc->sc_flags &= ~VNF_WANTED;
1275 		wakeup(sc);
1276 	}
1277 }
1278