xref: /netbsd-src/sys/dev/vnd.c (revision da5f4674a3fc214be3572d358b66af40ab9401e7)
1 /*	$NetBSD: vnd.c,v 1.102 2003/08/07 16:30:52 agc 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) 1990, 1993
41  *	The Regents of the University of California.  All rights reserved.
42  *
43  * This code is derived from software contributed to Berkeley by
44  * the Systems Programming Group of the University of Utah Computer
45  * Science Department.
46  *
47  * Redistribution and use in source and binary forms, with or without
48  * modification, are permitted provided that the following conditions
49  * are met:
50  * 1. Redistributions of source code must retain the above copyright
51  *    notice, this list of conditions and the following disclaimer.
52  * 2. Redistributions in binary form must reproduce the above copyright
53  *    notice, this list of conditions and the following disclaimer in the
54  *    documentation and/or other materials provided with the distribution.
55  * 3. Neither the name of the University nor the names of its contributors
56  *    may be used to endorse or promote products derived from this software
57  *    without specific prior written permission.
58  *
59  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
60  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
61  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
62  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
63  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
64  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
65  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
66  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
67  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
68  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
69  * SUCH DAMAGE.
70  *
71  * from: Utah $Hdr: vn.c 1.13 94/04/02$
72  *
73  *	@(#)vn.c	8.9 (Berkeley) 5/14/95
74  */
75 
76 /*
77  * Copyright (c) 1988 University of Utah.
78  *
79  * This code is derived from software contributed to Berkeley by
80  * the Systems Programming Group of the University of Utah Computer
81  * Science Department.
82  *
83  * Redistribution and use in source and binary forms, with or without
84  * modification, are permitted provided that the following conditions
85  * are met:
86  * 1. Redistributions of source code must retain the above copyright
87  *    notice, this list of conditions and the following disclaimer.
88  * 2. Redistributions in binary form must reproduce the above copyright
89  *    notice, this list of conditions and the following disclaimer in the
90  *    documentation and/or other materials provided with the distribution.
91  * 3. All advertising materials mentioning features or use of this software
92  *    must display the following acknowledgement:
93  *	This product includes software developed by the University of
94  *	California, Berkeley and its contributors.
95  * 4. Neither the name of the University nor the names of its contributors
96  *    may be used to endorse or promote products derived from this software
97  *    without specific prior written permission.
98  *
99  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
100  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
101  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
102  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
103  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
104  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
105  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
106  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
107  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
108  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
109  * SUCH DAMAGE.
110  *
111  * from: Utah $Hdr: vn.c 1.13 94/04/02$
112  *
113  *	@(#)vn.c	8.9 (Berkeley) 5/14/95
114  */
115 
116 /*
117  * Vnode disk driver.
118  *
119  * Block/character interface to a vnode.  Allows one to treat a file
120  * as a disk (e.g. build a filesystem in it, mount it, etc.).
121  *
122  * NOTE 1: This uses the VOP_BMAP/VOP_STRATEGY interface to the vnode
123  * instead of a simple VOP_RDWR.  We do this to avoid distorting the
124  * local buffer cache.
125  *
126  * NOTE 2: There is a security issue involved with this driver.
127  * Once mounted all access to the contents of the "mapped" file via
128  * the special file is controlled by the permissions on the special
129  * file, the protection of the mapped file is ignored (effectively,
130  * by using root credentials in all transactions).
131  *
132  * NOTE 3: Doesn't interact with leases, should it?
133  */
134 
135 #include <sys/cdefs.h>
136 __KERNEL_RCSID(0, "$NetBSD: vnd.c,v 1.102 2003/08/07 16:30:52 agc Exp $");
137 
138 #if defined(_KERNEL_OPT)
139 #include "fs_nfs.h"
140 #endif
141 
142 #include <sys/param.h>
143 #include <sys/systm.h>
144 #include <sys/namei.h>
145 #include <sys/proc.h>
146 #include <sys/errno.h>
147 #include <sys/buf.h>
148 #include <sys/malloc.h>
149 #include <sys/ioctl.h>
150 #include <sys/disklabel.h>
151 #include <sys/device.h>
152 #include <sys/disk.h>
153 #include <sys/stat.h>
154 #include <sys/mount.h>
155 #include <sys/vnode.h>
156 #include <sys/file.h>
157 #include <sys/uio.h>
158 #include <sys/conf.h>
159 
160 #include <miscfs/specfs/specdev.h>
161 
162 #include <dev/vndvar.h>
163 
164 #if defined(VNDDEBUG) && !defined(DEBUG)
165 #define	DEBUG
166 #endif
167 
168 #ifdef DEBUG
169 int dovndcluster = 1;
170 #define	VDB_FOLLOW	0x01
171 #define	VDB_INIT	0x02
172 #define	VDB_IO		0x04
173 #define	VDB_LABEL	0x08
174 int vnddebug = 0x00;
175 #endif
176 
177 #define	vndunit(x)	DISKUNIT(x)
178 
179 struct vndxfer {
180 	struct buf	*vx_bp;		/* Pointer to parent buffer */
181 	int		vx_error;
182 	int		vx_pending;	/* # of pending aux buffers */
183 	int		vx_flags;
184 #define VX_BUSY		1
185 };
186 
187 struct vndbuf {
188 	struct buf	vb_buf;
189 	struct vndxfer	*vb_xfer;
190 };
191 
192 #define	VND_GETXFER(vnd)	pool_get(&(vnd)->sc_vxpool, PR_NOWAIT)
193 #define	VND_PUTXFER(vnd, vx)	pool_put(&(vnd)->sc_vxpool, (vx))
194 
195 #define	VND_GETBUF(vnd)		pool_get(&(vnd)->sc_vbpool, PR_NOWAIT)
196 #define	VND_PUTBUF(vnd, vb)	pool_put(&(vnd)->sc_vbpool, (vb))
197 
198 struct vnd_softc *vnd_softc;
199 int numvnd = 0;
200 
201 #define	VNDLABELDEV(dev) \
202 	(MAKEDISKDEV(major((dev)), vndunit((dev)), RAW_PART))
203 
204 /* called by main() at boot time (XXX: and the LKM driver) */
205 void	vndattach __P((int));
206 int	vnddetach __P((void));
207 
208 void	vndclear __P((struct vnd_softc *, int));
209 void	vndstart __P((struct vnd_softc *));
210 int	vndsetcred __P((struct vnd_softc *, struct ucred *));
211 void	vndthrottle __P((struct vnd_softc *, struct vnode *));
212 void	vndiodone __P((struct buf *));
213 #if 0
214 void	vndshutdown __P((void));
215 #endif
216 
217 void	vndgetdefaultlabel __P((struct vnd_softc *, struct disklabel *));
218 void	vndgetdisklabel __P((dev_t));
219 
220 static	int vndlock __P((struct vnd_softc *));
221 static	void vndunlock __P((struct vnd_softc *));
222 
223 dev_type_open(vndopen);
224 dev_type_close(vndclose);
225 dev_type_read(vndread);
226 dev_type_write(vndwrite);
227 dev_type_ioctl(vndioctl);
228 dev_type_strategy(vndstrategy);
229 dev_type_dump(vnddump);
230 dev_type_size(vndsize);
231 
232 const struct bdevsw vnd_bdevsw = {
233 	vndopen, vndclose, vndstrategy, vndioctl, vnddump, vndsize, D_DISK
234 };
235 
236 const struct cdevsw vnd_cdevsw = {
237 	vndopen, vndclose, vndread, vndwrite, vndioctl,
238 	nostop, notty, nopoll, nommap, nokqfilter, D_DISK
239 };
240 
241 int vndattached = 0;
242 
243 void
244 vndattach(num)
245 	int num;
246 {
247 	int i;
248 	char *mem;
249 
250 	if (vndattached)
251 		return;
252 	vndattached = 1;
253 	if (num <= 0)
254 		return;
255 	i = num * sizeof(struct vnd_softc);
256 	mem = malloc(i, M_DEVBUF, M_NOWAIT|M_ZERO);
257 	if (mem == NULL) {
258 		printf("WARNING: no memory for vnode disks\n");
259 		return;
260 	}
261 	vnd_softc = (struct vnd_softc *)mem;
262 	numvnd = num;
263 
264 	for (i = 0; i < numvnd; i++) {
265 		vnd_softc[i].sc_unit = i;
266 		bufq_alloc(&vnd_softc[i].sc_tab,
267 		    BUFQ_DISKSORT|BUFQ_SORT_RAWBLOCK);
268 	}
269 }
270 
271 int
272 vnddetach()
273 {
274 	int i;
275 
276 	/* First check we aren't in use. */
277 	for (i = 0; i < numvnd; i++)
278 		if (vnd_softc[i].sc_flags & VNF_INITED)
279 			return (EBUSY);
280 
281 	for (i = 0; i < numvnd; i++)
282 		bufq_free(&vnd_softc[i].sc_tab);
283 
284 	free(vnd_softc, M_DEVBUF);
285 	vndattached = 0;
286 
287 	return (0);
288 }
289 
290 int
291 vndopen(dev, flags, mode, p)
292 	dev_t dev;
293 	int flags, mode;
294 	struct proc *p;
295 {
296 	int unit = vndunit(dev);
297 	struct vnd_softc *sc;
298 	int error = 0, part, pmask;
299 	struct disklabel *lp;
300 
301 #ifdef DEBUG
302 	if (vnddebug & VDB_FOLLOW)
303 		printf("vndopen(0x%x, 0x%x, 0x%x, %p)\n", dev, flags, mode, p);
304 #endif
305 	if (unit >= numvnd)
306 		return (ENXIO);
307 	sc = &vnd_softc[unit];
308 
309 	if ((error = vndlock(sc)) != 0)
310 		return (error);
311 
312 	lp = sc->sc_dkdev.dk_label;
313 
314 	part = DISKPART(dev);
315 	pmask = (1 << part);
316 
317 	/*
318 	 * If we're initialized, check to see if there are any other
319 	 * open partitions.  If not, then it's safe to update the
320 	 * in-core disklabel.  Only read the disklabel if it is
321 	 * not realdy valid.
322 	 */
323 	if ((sc->sc_flags & (VNF_INITED|VNF_VLABEL)) == VNF_INITED &&
324 	    sc->sc_dkdev.dk_openmask == 0)
325 		vndgetdisklabel(dev);
326 
327 	/* Check that the partitions exists. */
328 	if (part != RAW_PART) {
329 		if (((sc->sc_flags & VNF_INITED) == 0) ||
330 		    ((part >= lp->d_npartitions) ||
331 		     (lp->d_partitions[part].p_fstype == FS_UNUSED))) {
332 			error = ENXIO;
333 			goto done;
334 		}
335 	}
336 
337 	/* Prevent our unit from being unconfigured while open. */
338 	switch (mode) {
339 	case S_IFCHR:
340 		sc->sc_dkdev.dk_copenmask |= pmask;
341 		break;
342 
343 	case S_IFBLK:
344 		sc->sc_dkdev.dk_bopenmask |= pmask;
345 		break;
346 	}
347 	sc->sc_dkdev.dk_openmask =
348 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
349 
350  done:
351 	vndunlock(sc);
352 	return (error);
353 }
354 
355 int
356 vndclose(dev, flags, mode, p)
357 	dev_t dev;
358 	int flags, mode;
359 	struct proc *p;
360 {
361 	int unit = vndunit(dev);
362 	struct vnd_softc *sc;
363 	int error = 0, part;
364 
365 #ifdef DEBUG
366 	if (vnddebug & VDB_FOLLOW)
367 		printf("vndclose(0x%x, 0x%x, 0x%x, %p)\n", dev, flags, mode, p);
368 #endif
369 
370 	if (unit >= numvnd)
371 		return (ENXIO);
372 	sc = &vnd_softc[unit];
373 
374 	if ((error = vndlock(sc)) != 0)
375 		return (error);
376 
377 	part = DISKPART(dev);
378 
379 	/* ...that much closer to allowing unconfiguration... */
380 	switch (mode) {
381 	case S_IFCHR:
382 		sc->sc_dkdev.dk_copenmask &= ~(1 << part);
383 		break;
384 
385 	case S_IFBLK:
386 		sc->sc_dkdev.dk_bopenmask &= ~(1 << part);
387 		break;
388 	}
389 	sc->sc_dkdev.dk_openmask =
390 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
391 
392 	if (sc->sc_dkdev.dk_openmask == 0) {
393 		if ((sc->sc_flags & VNF_KLABEL) == 0)
394 			sc->sc_flags &= ~VNF_VLABEL;
395 	}
396 
397 	vndunlock(sc);
398 	return (0);
399 }
400 
401 /*
402  * Break the request into bsize pieces and submit using VOP_BMAP/VOP_STRATEGY.
403  */
404 void
405 vndstrategy(bp)
406 	struct buf *bp;
407 {
408 	int unit = vndunit(bp->b_dev);
409 	struct vnd_softc *vnd = &vnd_softc[unit];
410 	struct vndxfer *vnx;
411 	int s, bsize, resid;
412 	off_t bn;
413 	caddr_t addr;
414 	int sz, flags, error, wlabel;
415 	struct disklabel *lp;
416 	struct partition *pp;
417 
418 #ifdef DEBUG
419 	if (vnddebug & VDB_FOLLOW)
420 		printf("vndstrategy(%p): unit %d\n", bp, unit);
421 #endif
422 	if ((vnd->sc_flags & VNF_INITED) == 0) {
423 		bp->b_error = ENXIO;
424 		bp->b_flags |= B_ERROR;
425 		goto done;
426 	}
427 
428 	/* If it's a nil transfer, wake up the top half now. */
429 	if (bp->b_bcount == 0)
430 		goto done;
431 
432 	lp = vnd->sc_dkdev.dk_label;
433 
434 	/*
435 	 * The transfer must be a whole number of blocks.
436 	 */
437 	if ((bp->b_bcount % lp->d_secsize) != 0) {
438 		bp->b_error = EINVAL;
439 		bp->b_flags |= B_ERROR;
440 		goto done;
441 	}
442 
443 	/*
444 	 * Do bounds checking and adjust transfer.  If there's an error,
445 	 * the bounds check will flag that for us.
446 	 */
447 	wlabel = vnd->sc_flags & (VNF_WLABEL|VNF_LABELLING);
448 	if (DISKPART(bp->b_dev) != RAW_PART)
449 		if (bounds_check_with_label(&vnd->sc_dkdev, bp, wlabel) <= 0)
450 			goto done;
451 
452 	/*
453 	 * check if we're read-only.
454 	 */
455 	if ((vnd->sc_flags & VNF_READONLY) && !(bp->b_flags & B_READ)) {
456 		bp->b_error = EACCES;
457 		bp->b_flags |= B_ERROR;
458 		goto done;
459 	}
460 
461 	bp->b_resid = bp->b_bcount;
462 
463 	/*
464 	 * Put the block number in terms of the logical blocksize
465 	 * of the "device".
466 	 */
467 	bn = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
468 
469 	/*
470 	 * Translate the partition-relative block number to an absolute.
471 	 */
472 	if (DISKPART(bp->b_dev) != RAW_PART) {
473 		pp = &vnd->sc_dkdev.dk_label->d_partitions[DISKPART(bp->b_dev)];
474 		bn += pp->p_offset;
475 	}
476 
477 	/* ...and convert to a byte offset within the file. */
478 	bn *= lp->d_secsize;
479 
480 	if (vnd->sc_vp->v_mount == NULL) {
481 		bp->b_error = ENXIO;
482 		bp->b_flags |= B_ERROR;
483 		goto done;
484 	}
485  	bsize = vnd->sc_vp->v_mount->mnt_stat.f_iosize;
486 	addr = bp->b_data;
487 	flags = (bp->b_flags & (B_READ|B_ASYNC)) | B_CALL;
488 
489 	/* Allocate a header for this transfer and link it to the buffer */
490 	s = splbio();
491 	vnx = VND_GETXFER(vnd);
492 	splx(s);
493 	vnx->vx_flags = VX_BUSY;
494 	vnx->vx_error = 0;
495 	vnx->vx_pending = 0;
496 	vnx->vx_bp = bp;
497 
498 	for (resid = bp->b_resid; resid; resid -= sz) {
499 		struct vndbuf *nbp;
500 		struct vnode *vp;
501 		daddr_t nbn;
502 		int off, nra;
503 
504 		nra = 0;
505 		vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY | LK_CANRECURSE);
506 		error = VOP_BMAP(vnd->sc_vp, bn / bsize, &vp, &nbn, &nra);
507 		VOP_UNLOCK(vnd->sc_vp, 0);
508 
509 		if (error == 0 && (long)nbn == -1)
510 			error = EIO;
511 
512 		/*
513 		 * If there was an error or a hole in the file...punt.
514 		 * Note that we may have to wait for any operations
515 		 * that we have already fired off before releasing
516 		 * the buffer.
517 		 *
518 		 * XXX we could deal with holes here but it would be
519 		 * a hassle (in the write case).
520 		 */
521 		if (error) {
522 			s = splbio();
523 			vnx->vx_error = error;
524 			goto out;
525 		}
526 
527 #ifdef DEBUG
528 		if (!dovndcluster)
529 			nra = 0;
530 #endif
531 
532 		if ((off = bn % bsize) != 0)
533 			sz = bsize - off;
534 		else
535 			sz = (1 + nra) * bsize;
536 		if (resid < sz)
537 			sz = resid;
538 #ifdef DEBUG
539 		if (vnddebug & VDB_IO)
540 			printf("vndstrategy: vp %p/%p bn 0x%qx/0x%" PRIx64
541 			       " sz 0x%x\n",
542 			    vnd->sc_vp, vp, (long long)bn, nbn, sz);
543 #endif
544 
545 		s = splbio();
546 		nbp = VND_GETBUF(vnd);
547 		splx(s);
548 		BUF_INIT(&nbp->vb_buf);
549 		nbp->vb_buf.b_flags = flags;
550 		nbp->vb_buf.b_bcount = sz;
551 		nbp->vb_buf.b_bufsize = round_page((ulong)addr + sz)
552 		    - trunc_page((ulong) addr);
553 		nbp->vb_buf.b_error = 0;
554 		nbp->vb_buf.b_data = addr;
555 		nbp->vb_buf.b_blkno = nbp->vb_buf.b_rawblkno = nbn + btodb(off);
556 		nbp->vb_buf.b_proc = bp->b_proc;
557 		nbp->vb_buf.b_iodone = vndiodone;
558 		nbp->vb_buf.b_vp = NULLVP;
559 
560 		nbp->vb_xfer = vnx;
561 
562 		/*
563 		 * Just sort by block number
564 		 */
565 		s = splbio();
566 		if (vnx->vx_error != 0) {
567 			VND_PUTBUF(vnd, nbp);
568 			goto out;
569 		}
570 		vnx->vx_pending++;
571 		bgetvp(vp, &nbp->vb_buf);
572 		BUFQ_PUT(&vnd->sc_tab, &nbp->vb_buf);
573 		vndstart(vnd);
574 		splx(s);
575 		bn += sz;
576 		addr += sz;
577 	}
578 
579 	s = splbio();
580 
581 out: /* Arrive here at splbio */
582 	vnx->vx_flags &= ~VX_BUSY;
583 	if (vnx->vx_pending == 0) {
584 		if (vnx->vx_error != 0) {
585 			bp->b_error = vnx->vx_error;
586 			bp->b_flags |= B_ERROR;
587 		}
588 		VND_PUTXFER(vnd, vnx);
589 		biodone(bp);
590 	}
591 	splx(s);
592 	return;
593 
594  done:
595 	biodone(bp);
596 }
597 
598 /*
599  * Feed requests sequentially.
600  * We do it this way to keep from flooding NFS servers if we are connected
601  * to an NFS file.  This places the burden on the client rather than the
602  * server.
603  */
604 void
605 vndstart(vnd)
606 	struct vnd_softc *vnd;
607 {
608 	struct buf	*bp;
609 
610 	/*
611 	 * Dequeue now since lower level strategy routine might
612 	 * queue using same links
613 	 */
614 
615 	if ((vnd->sc_flags & VNF_BUSY) != 0)
616 		return;
617 
618 	vnd->sc_flags |= VNF_BUSY;
619 
620 	while (vnd->sc_active < vnd->sc_maxactive) {
621 		bp = BUFQ_GET(&vnd->sc_tab);
622 		if (bp == NULL)
623 			break;
624 		vnd->sc_active++;
625 #ifdef DEBUG
626 		if (vnddebug & VDB_IO)
627 			printf("vndstart(%ld): bp %p vp %p blkno 0x%" PRIx64
628 				" flags %lx addr %p cnt 0x%lx\n",
629 			    (long) (vnd-vnd_softc), bp, bp->b_vp, bp->b_blkno,
630 			    bp->b_flags, bp->b_data, bp->b_bcount);
631 #endif
632 
633 		/* Instrumentation. */
634 		disk_busy(&vnd->sc_dkdev);
635 
636 		if ((bp->b_flags & B_READ) == 0)
637 			bp->b_vp->v_numoutput++;
638 		VOP_STRATEGY(bp);
639 	}
640 	vnd->sc_flags &= ~VNF_BUSY;
641 }
642 
643 void
644 vndiodone(bp)
645 	struct buf *bp;
646 {
647 	struct vndbuf *vbp = (struct vndbuf *) bp;
648 	struct vndxfer *vnx = (struct vndxfer *)vbp->vb_xfer;
649 	struct buf *pbp = vnx->vx_bp;
650 	struct vnd_softc *vnd = &vnd_softc[vndunit(pbp->b_dev)];
651 	int s, resid;
652 
653 	s = splbio();
654 #ifdef DEBUG
655 	if (vnddebug & VDB_IO)
656 		printf("vndiodone(%ld): vbp %p vp %p blkno 0x%" PRIx64
657 		       " addr %p cnt 0x%lx\n",
658 		    (long) (vnd-vnd_softc), vbp, vbp->vb_buf.b_vp,
659 		    vbp->vb_buf.b_blkno, vbp->vb_buf.b_data,
660 		    vbp->vb_buf.b_bcount);
661 #endif
662 
663 	resid = vbp->vb_buf.b_bcount - vbp->vb_buf.b_resid;
664 	pbp->b_resid -= resid;
665 	disk_unbusy(&vnd->sc_dkdev, resid, (pbp->b_flags & B_READ));
666 	vnx->vx_pending--;
667 
668 	if (vbp->vb_buf.b_error) {
669 #ifdef DEBUG
670 		if (vnddebug & VDB_IO)
671 			printf("vndiodone: vbp %p error %d\n", vbp,
672 			    vbp->vb_buf.b_error);
673 #endif
674 		vnx->vx_error = vbp->vb_buf.b_error;
675 	}
676 
677 	if (vbp->vb_buf.b_vp != NULLVP)
678 		brelvp(&vbp->vb_buf);
679 
680 	VND_PUTBUF(vnd, vbp);
681 
682 	/*
683 	 * Wrap up this transaction if it has run to completion or, in
684 	 * case of an error, when all auxiliary buffers have returned.
685 	 */
686 	if (vnx->vx_error != 0) {
687 		pbp->b_flags |= B_ERROR;
688 		pbp->b_error = vnx->vx_error;
689 		if ((vnx->vx_flags & VX_BUSY) == 0 && vnx->vx_pending == 0) {
690 
691 #ifdef DEBUG
692 			if (vnddebug & VDB_IO)
693 				printf("vndiodone: pbp %p iodone: error %d\n",
694 					pbp, vnx->vx_error);
695 #endif
696 			VND_PUTXFER(vnd, vnx);
697 			biodone(pbp);
698 		}
699 	} else if (pbp->b_resid == 0) {
700 
701 #ifdef DIAGNOSTIC
702 		if (vnx->vx_pending != 0)
703 			panic("vndiodone: vnx pending: %d", vnx->vx_pending);
704 #endif
705 
706 		if ((vnx->vx_flags & VX_BUSY) == 0) {
707 #ifdef DEBUG
708 			if (vnddebug & VDB_IO)
709 				printf("vndiodone: pbp %p iodone\n", pbp);
710 #endif
711 			VND_PUTXFER(vnd, vnx);
712 			biodone(pbp);
713 		}
714 	}
715 
716 	vnd->sc_active--;
717 	vndstart(vnd);
718 	splx(s);
719 }
720 
721 /* ARGSUSED */
722 int
723 vndread(dev, uio, flags)
724 	dev_t dev;
725 	struct uio *uio;
726 	int flags;
727 {
728 	int unit = vndunit(dev);
729 	struct vnd_softc *sc;
730 
731 #ifdef DEBUG
732 	if (vnddebug & VDB_FOLLOW)
733 		printf("vndread(0x%x, %p)\n", dev, uio);
734 #endif
735 
736 	if (unit >= numvnd)
737 		return (ENXIO);
738 	sc = &vnd_softc[unit];
739 
740 	if ((sc->sc_flags & VNF_INITED) == 0)
741 		return (ENXIO);
742 
743 	return (physio(vndstrategy, NULL, dev, B_READ, minphys, uio));
744 }
745 
746 /* ARGSUSED */
747 int
748 vndwrite(dev, uio, flags)
749 	dev_t dev;
750 	struct uio *uio;
751 	int flags;
752 {
753 	int unit = vndunit(dev);
754 	struct vnd_softc *sc;
755 
756 #ifdef DEBUG
757 	if (vnddebug & VDB_FOLLOW)
758 		printf("vndwrite(0x%x, %p)\n", dev, uio);
759 #endif
760 
761 	if (unit >= numvnd)
762 		return (ENXIO);
763 	sc = &vnd_softc[unit];
764 
765 	if ((sc->sc_flags & VNF_INITED) == 0)
766 		return (ENXIO);
767 
768 	return (physio(vndstrategy, NULL, dev, B_WRITE, minphys, uio));
769 }
770 
771 /* ARGSUSED */
772 int
773 vndioctl(dev, cmd, data, flag, p)
774 	dev_t dev;
775 	u_long cmd;
776 	caddr_t data;
777 	int flag;
778 	struct proc *p;
779 {
780 	int unit = vndunit(dev);
781 	struct vnd_softc *vnd;
782 	struct vnd_ioctl *vio;
783 	struct vattr vattr;
784 	struct nameidata nd;
785 	int error, part, pmask;
786 	size_t geomsize;
787 	int fflags;
788 #ifdef __HAVE_OLD_DISKLABEL
789 	struct disklabel newlabel;
790 #endif
791 
792 #ifdef DEBUG
793 	if (vnddebug & VDB_FOLLOW)
794 		printf("vndioctl(0x%x, 0x%lx, %p, 0x%x, %p): unit %d\n",
795 		    dev, cmd, data, flag, p, unit);
796 #endif
797 	if (unit >= numvnd)
798 		return (ENXIO);
799 
800 	vnd = &vnd_softc[unit];
801 	vio = (struct vnd_ioctl *)data;
802 
803 	/* Must be open for writes for these commands... */
804 	switch (cmd) {
805 	case VNDIOCSET:
806 	case VNDIOCCLR:
807 	case DIOCSDINFO:
808 	case DIOCWDINFO:
809 #ifdef __HAVE_OLD_DISKLABEL
810 	case ODIOCSDINFO:
811 	case ODIOCWDINFO:
812 #endif
813 	case DIOCKLABEL:
814 	case DIOCWLABEL:
815 		if ((flag & FWRITE) == 0)
816 			return (EBADF);
817 	}
818 
819 	/* Must be initialized for these... */
820 	switch (cmd) {
821 	case VNDIOCCLR:
822 	case DIOCGDINFO:
823 	case DIOCSDINFO:
824 	case DIOCWDINFO:
825 	case DIOCGPART:
826 	case DIOCKLABEL:
827 	case DIOCWLABEL:
828 	case DIOCGDEFLABEL:
829 #ifdef __HAVE_OLD_DISKLABEL
830 	case ODIOCGDINFO:
831 	case ODIOCSDINFO:
832 	case ODIOCWDINFO:
833 	case ODIOCGDEFLABEL:
834 #endif
835 		if ((vnd->sc_flags & VNF_INITED) == 0)
836 			return (ENXIO);
837 	}
838 
839 	switch (cmd) {
840 	case VNDIOCSET:
841 		if (vnd->sc_flags & VNF_INITED)
842 			return (EBUSY);
843 
844 		if ((error = vndlock(vnd)) != 0)
845 			return (error);
846 
847 		fflags = FREAD;
848 		if ((vio->vnd_flags & VNDIOF_READONLY) == 0)
849 			fflags |= FWRITE;
850 		NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, vio->vnd_file, p);
851 		if ((error = vn_open(&nd, fflags, 0)) != 0)
852 			goto unlock_and_exit;
853 		error = VOP_GETATTR(nd.ni_vp, &vattr, p->p_ucred, p);
854 		VOP_UNLOCK(nd.ni_vp, 0);
855 		if (!error && nd.ni_vp->v_type != VREG)
856 			error = EOPNOTSUPP;
857 		if (error)
858 			goto close_and_exit;
859 		vnd->sc_vp = nd.ni_vp;
860 		vnd->sc_size = btodb(vattr.va_size);	/* note truncation */
861 
862 		/*
863 		 * Use pseudo-geometry specified.  If none was provided,
864 		 * use "standard" Adaptec fictitious geometry.
865 		 */
866 		if (vio->vnd_flags & VNDIOF_HASGEOM) {
867 
868 			memcpy(&vnd->sc_geom, &vio->vnd_geom,
869 			    sizeof(vio->vnd_geom));
870 
871 			/*
872 			 * Sanity-check the sector size.
873 			 * XXX Don't allow secsize < DEV_BSIZE.  Should
874 			 * XXX we?
875 			 */
876 			if (vnd->sc_geom.vng_secsize < DEV_BSIZE ||
877 			    (vnd->sc_geom.vng_secsize % DEV_BSIZE) != 0) {
878 				error = EINVAL;
879 				goto close_and_exit;
880 			}
881 
882 			/*
883 			 * Compute the size (in DEV_BSIZE blocks) specified
884 			 * by the geometry.
885 			 */
886 			geomsize = (vnd->sc_geom.vng_nsectors *
887 			    vnd->sc_geom.vng_ntracks *
888 			    vnd->sc_geom.vng_ncylinders) *
889 			    (vnd->sc_geom.vng_secsize / DEV_BSIZE);
890 
891 			/*
892 			 * Sanity-check the size against the specified
893 			 * geometry.
894 			 */
895 			if (vnd->sc_size < geomsize) {
896 				error = EINVAL;
897 				goto close_and_exit;
898 			}
899 		} else {
900 			/*
901 			 * Size must be at least 2048 DEV_BSIZE blocks
902 			 * (1M) in order to use this geometry.
903 			 */
904 			if (vnd->sc_size < (32 * 64)) {
905 				error = EINVAL;
906 				goto close_and_exit;
907 			}
908 
909 			vnd->sc_geom.vng_secsize = DEV_BSIZE;
910 			vnd->sc_geom.vng_nsectors = 32;
911 			vnd->sc_geom.vng_ntracks = 64;
912 			vnd->sc_geom.vng_ncylinders = vnd->sc_size / (64 * 32);
913 		}
914 
915 		if (vio->vnd_flags & VNDIOF_READONLY) {
916 			vnd->sc_flags |= VNF_READONLY;
917 		}
918 
919 		if ((error = vndsetcred(vnd, p->p_ucred)) != 0)
920 			goto close_and_exit;
921 		vndthrottle(vnd, vnd->sc_vp);
922 		vio->vnd_size = dbtob(vnd->sc_size);
923 		vnd->sc_flags |= VNF_INITED;
924 #ifdef DEBUG
925 		if (vnddebug & VDB_INIT)
926 			printf("vndioctl: SET vp %p size 0x%lx %d/%d/%d/%d\n",
927 			    vnd->sc_vp, (unsigned long) vnd->sc_size,
928 			    vnd->sc_geom.vng_secsize,
929 			    vnd->sc_geom.vng_nsectors,
930 			    vnd->sc_geom.vng_ntracks,
931 			    vnd->sc_geom.vng_ncylinders);
932 #endif
933 
934 		/* Attach the disk. */
935 		memset(vnd->sc_xname, 0, sizeof(vnd->sc_xname)); /* XXX */
936 		sprintf(vnd->sc_xname, "vnd%d", unit);		/* XXX */
937 		vnd->sc_dkdev.dk_name = vnd->sc_xname;
938 		disk_attach(&vnd->sc_dkdev);
939 
940 		/* Initialize the xfer and buffer pools. */
941 		pool_init(&vnd->sc_vxpool, sizeof(struct vndxfer), 0,
942 		    0, 0, "vndxpl", NULL);
943 		pool_init(&vnd->sc_vbpool, sizeof(struct vndbuf), 0,
944 		    0, 0, "vndbpl", NULL);
945 
946 		/* Try and read the disklabel. */
947 		vndgetdisklabel(dev);
948 
949 		vndunlock(vnd);
950 
951 		break;
952 
953 close_and_exit:
954 		(void) vn_close(nd.ni_vp, fflags, p->p_ucred, p);
955 unlock_and_exit:
956 		vndunlock(vnd);
957 		return (error);
958 
959 	case VNDIOCCLR:
960 		if ((error = vndlock(vnd)) != 0)
961 			return (error);
962 
963 		/*
964 		 * Don't unconfigure if any other partitions are open
965 		 * or if both the character and block flavors of this
966 		 * partition are open.
967 		 */
968 		part = DISKPART(dev);
969 		pmask = (1 << part);
970 		if (((vnd->sc_dkdev.dk_openmask & ~pmask) ||
971 		    ((vnd->sc_dkdev.dk_bopenmask & pmask) &&
972 		    (vnd->sc_dkdev.dk_copenmask & pmask))) &&
973 			!(vio->vnd_flags & VNDIOF_FORCE)) {
974 			vndunlock(vnd);
975 			return (EBUSY);
976 		}
977 
978 		/*
979 		 * XXX vndclear() might call vndclose() implicitely;
980 		 * release lock to avoid recursion
981 		 */
982 		vndunlock(vnd);
983 		vndclear(vnd, minor(dev));
984 #ifdef DEBUG
985 		if (vnddebug & VDB_INIT)
986 			printf("vndioctl: CLRed\n");
987 #endif
988 
989 		/* Destroy the xfer and buffer pools. */
990 		pool_destroy(&vnd->sc_vxpool);
991 		pool_destroy(&vnd->sc_vbpool);
992 
993 		/* Detatch the disk. */
994 		disk_detach(&vnd->sc_dkdev);
995 
996 		break;
997 
998 	case VNDIOCGET: {
999 		struct vnd_user *vnu;
1000 		struct vattr va;
1001 
1002 		vnu = (struct vnd_user *)data;
1003 
1004 		if (vnu->vnu_unit == -1)
1005 			vnu->vnu_unit = unit;
1006 		if (vnu->vnu_unit >= numvnd)
1007 			return (ENXIO);
1008 		if (vnu->vnu_unit < 0)
1009 			return (EINVAL);
1010 
1011 		vnd = &vnd_softc[vnu->vnu_unit];
1012 
1013 		if (vnd->sc_flags & VNF_INITED) {
1014 			error = VOP_GETATTR(vnd->sc_vp, &va, p->p_ucred, p);
1015 			if (error)
1016 				return (error);
1017 			vnu->vnu_dev = va.va_fsid;
1018 			vnu->vnu_ino = va.va_fileid;
1019 		}
1020 		else {
1021 			/* unused is not an error */
1022 			vnu->vnu_dev = 0;
1023 			vnu->vnu_ino = 0;
1024 		}
1025 
1026 		break;
1027 	}
1028 
1029 	case DIOCGDINFO:
1030 		*(struct disklabel *)data = *(vnd->sc_dkdev.dk_label);
1031 		break;
1032 
1033 #ifdef __HAVE_OLD_DISKLABEL
1034 	case ODIOCGDINFO:
1035 		newlabel = *(vnd->sc_dkdev.dk_label);
1036 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1037 			return ENOTTY;
1038 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1039 		break;
1040 #endif
1041 
1042 	case DIOCGPART:
1043 		((struct partinfo *)data)->disklab = vnd->sc_dkdev.dk_label;
1044 		((struct partinfo *)data)->part =
1045 		    &vnd->sc_dkdev.dk_label->d_partitions[DISKPART(dev)];
1046 		break;
1047 
1048 	case DIOCWDINFO:
1049 	case DIOCSDINFO:
1050 #ifdef __HAVE_OLD_DISKLABEL
1051 	case ODIOCWDINFO:
1052 	case ODIOCSDINFO:
1053 #endif
1054 	{
1055 		struct disklabel *lp;
1056 
1057 		if ((error = vndlock(vnd)) != 0)
1058 			return (error);
1059 
1060 		vnd->sc_flags |= VNF_LABELLING;
1061 
1062 #ifdef __HAVE_OLD_DISKLABEL
1063 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
1064 			memset(&newlabel, 0, sizeof newlabel);
1065 			memcpy(&newlabel, data, sizeof (struct olddisklabel));
1066 			lp = &newlabel;
1067 		} else
1068 #endif
1069 		lp = (struct disklabel *)data;
1070 
1071 		error = setdisklabel(vnd->sc_dkdev.dk_label,
1072 		    lp, 0, vnd->sc_dkdev.dk_cpulabel);
1073 		if (error == 0) {
1074 			if (cmd == DIOCWDINFO
1075 #ifdef __HAVE_OLD_DISKLABEL
1076 			    || cmd == ODIOCWDINFO
1077 #endif
1078 			   )
1079 				error = writedisklabel(VNDLABELDEV(dev),
1080 				    vndstrategy, vnd->sc_dkdev.dk_label,
1081 				    vnd->sc_dkdev.dk_cpulabel);
1082 		}
1083 
1084 		vnd->sc_flags &= ~VNF_LABELLING;
1085 
1086 		vndunlock(vnd);
1087 
1088 		if (error)
1089 			return (error);
1090 		break;
1091 	}
1092 
1093 	case DIOCKLABEL:
1094 		if (*(int *)data != 0)
1095 			vnd->sc_flags |= VNF_KLABEL;
1096 		else
1097 			vnd->sc_flags &= ~VNF_KLABEL;
1098 		break;
1099 
1100 	case DIOCWLABEL:
1101 		if (*(int *)data != 0)
1102 			vnd->sc_flags |= VNF_WLABEL;
1103 		else
1104 			vnd->sc_flags &= ~VNF_WLABEL;
1105 		break;
1106 
1107 	case DIOCGDEFLABEL:
1108 		vndgetdefaultlabel(vnd, (struct disklabel *)data);
1109 		break;
1110 
1111 #ifdef __HAVE_OLD_DISKLABEL
1112 	case ODIOCGDEFLABEL:
1113 		vndgetdefaultlabel(vnd, &newlabel);
1114 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1115 			return ENOTTY;
1116 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1117 		break;
1118 #endif
1119 
1120 	default:
1121 		return (ENOTTY);
1122 	}
1123 
1124 	return (0);
1125 }
1126 
1127 /*
1128  * Duplicate the current processes' credentials.  Since we are called only
1129  * as the result of a SET ioctl and only root can do that, any future access
1130  * to this "disk" is essentially as root.  Note that credentials may change
1131  * if some other uid can write directly to the mapped file (NFS).
1132  */
1133 int
1134 vndsetcred(vnd, cred)
1135 	struct vnd_softc *vnd;
1136 	struct ucred *cred;
1137 {
1138 	struct uio auio;
1139 	struct iovec aiov;
1140 	char *tmpbuf;
1141 	int error;
1142 
1143 	vnd->sc_cred = crdup(cred);
1144 	tmpbuf = malloc(DEV_BSIZE, M_TEMP, M_WAITOK);
1145 
1146 	/* XXX: Horrible kludge to establish credentials for NFS */
1147 	aiov.iov_base = tmpbuf;
1148 	aiov.iov_len = min(DEV_BSIZE, dbtob(vnd->sc_size));
1149 	auio.uio_iov = &aiov;
1150 	auio.uio_iovcnt = 1;
1151 	auio.uio_offset = 0;
1152 	auio.uio_rw = UIO_READ;
1153 	auio.uio_segflg = UIO_SYSSPACE;
1154 	auio.uio_resid = aiov.iov_len;
1155 	vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1156 	error = VOP_READ(vnd->sc_vp, &auio, 0, vnd->sc_cred);
1157 	if (error == 0) {
1158 		/*
1159 		 * Because vnd does all IO directly through the vnode
1160 		 * we need to flush (at least) the buffer from the above
1161 		 * VOP_READ from the buffer cache to prevent cache
1162 		 * incoherencies.  Also, be careful to write dirty
1163 		 * buffers back to stable storage.
1164 		 */
1165 		error = vinvalbuf(vnd->sc_vp, V_SAVE, vnd->sc_cred,
1166 			    curproc, 0, 0);
1167 	}
1168 	VOP_UNLOCK(vnd->sc_vp, 0);
1169 
1170 	free(tmpbuf, M_TEMP);
1171 	return (error);
1172 }
1173 
1174 /*
1175  * Set maxactive based on FS type
1176  */
1177 void
1178 vndthrottle(vnd, vp)
1179 	struct vnd_softc *vnd;
1180 	struct vnode *vp;
1181 {
1182 #ifdef NFS
1183 	extern int (**nfsv2_vnodeop_p) __P((void *));
1184 
1185 	if (vp->v_op == nfsv2_vnodeop_p)
1186 		vnd->sc_maxactive = 2;
1187 	else
1188 #endif
1189 		vnd->sc_maxactive = 8;
1190 
1191 	if (vnd->sc_maxactive < 1)
1192 		vnd->sc_maxactive = 1;
1193 }
1194 
1195 #if 0
1196 void
1197 vndshutdown()
1198 {
1199 	struct vnd_softc *vnd;
1200 
1201 	for (vnd = &vnd_softc[0]; vnd < &vnd_softc[numvnd]; vnd++)
1202 		if (vnd->sc_flags & VNF_INITED)
1203 			vndclear(vnd);
1204 }
1205 #endif
1206 
1207 void
1208 vndclear(vnd, myminor)
1209 	struct vnd_softc *vnd;
1210 	int myminor;
1211 {
1212 	struct vnode *vp = vnd->sc_vp;
1213 	struct proc *p = curproc;		/* XXX */
1214 	int fflags = FREAD;
1215 	int bmaj, cmaj, i, mn;
1216 
1217 #ifdef DEBUG
1218 	if (vnddebug & VDB_FOLLOW)
1219 		printf("vndclear(%p): vp %p\n", vnd, vp);
1220 #endif
1221 	/* locate the major number */
1222 	bmaj = bdevsw_lookup_major(&vnd_bdevsw);
1223 	cmaj = cdevsw_lookup_major(&vnd_cdevsw);
1224 
1225 	/* Nuke the vnodes for any open instances */
1226 	for (i = 0; i < MAXPARTITIONS; i++) {
1227 		mn = DISKMINOR(vnd->sc_unit, i);
1228 		vdevgone(bmaj, mn, mn, VBLK);
1229 		if (mn != myminor) /* XXX avoid to kill own vnode */
1230 			vdevgone(cmaj, mn, mn, VCHR);
1231 	}
1232 
1233 	if ((vnd->sc_flags & VNF_READONLY) == 0)
1234 		fflags |= FWRITE;
1235 	vnd->sc_flags &= ~(VNF_INITED | VNF_READONLY | VNF_VLABEL);
1236 	if (vp == (struct vnode *)0)
1237 		panic("vndioctl: null vp");
1238 	(void) vn_close(vp, fflags, vnd->sc_cred, p);
1239 	crfree(vnd->sc_cred);
1240 	vnd->sc_vp = (struct vnode *)0;
1241 	vnd->sc_cred = (struct ucred *)0;
1242 	vnd->sc_size = 0;
1243 }
1244 
1245 int
1246 vndsize(dev)
1247 	dev_t dev;
1248 {
1249 	struct vnd_softc *sc;
1250 	struct disklabel *lp;
1251 	int part, unit, omask;
1252 	int size;
1253 
1254 	unit = vndunit(dev);
1255 	if (unit >= numvnd)
1256 		return (-1);
1257 	sc = &vnd_softc[unit];
1258 
1259 	if ((sc->sc_flags & VNF_INITED) == 0)
1260 		return (-1);
1261 
1262 	part = DISKPART(dev);
1263 	omask = sc->sc_dkdev.dk_openmask & (1 << part);
1264 	lp = sc->sc_dkdev.dk_label;
1265 
1266 	if (omask == 0 && vndopen(dev, 0, S_IFBLK, curproc))
1267 		return (-1);
1268 
1269 	if (lp->d_partitions[part].p_fstype != FS_SWAP)
1270 		size = -1;
1271 	else
1272 		size = lp->d_partitions[part].p_size *
1273 		    (lp->d_secsize / DEV_BSIZE);
1274 
1275 	if (omask == 0 && vndclose(dev, 0, S_IFBLK, curproc))
1276 		return (-1);
1277 
1278 	return (size);
1279 }
1280 
1281 int
1282 vnddump(dev, blkno, va, size)
1283 	dev_t dev;
1284 	daddr_t blkno;
1285 	caddr_t va;
1286 	size_t size;
1287 {
1288 
1289 	/* Not implemented. */
1290 	return ENXIO;
1291 }
1292 
1293 void
1294 vndgetdefaultlabel(sc, lp)
1295 	struct vnd_softc *sc;
1296 	struct disklabel *lp;
1297 {
1298 	struct vndgeom *vng = &sc->sc_geom;
1299 	struct partition *pp;
1300 
1301 	memset(lp, 0, sizeof(*lp));
1302 
1303 	lp->d_secperunit = sc->sc_size / (vng->vng_secsize / DEV_BSIZE);
1304 	lp->d_secsize = vng->vng_secsize;
1305 	lp->d_nsectors = vng->vng_nsectors;
1306 	lp->d_ntracks = vng->vng_ntracks;
1307 	lp->d_ncylinders = vng->vng_ncylinders;
1308 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1309 
1310 	strncpy(lp->d_typename, "vnd", sizeof(lp->d_typename));
1311 	lp->d_type = DTYPE_VND;
1312 	strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
1313 	lp->d_rpm = 3600;
1314 	lp->d_interleave = 1;
1315 	lp->d_flags = 0;
1316 
1317 	pp = &lp->d_partitions[RAW_PART];
1318 	pp->p_offset = 0;
1319 	pp->p_size = lp->d_secperunit;
1320 	pp->p_fstype = FS_UNUSED;
1321 	lp->d_npartitions = RAW_PART + 1;
1322 
1323 	lp->d_magic = DISKMAGIC;
1324 	lp->d_magic2 = DISKMAGIC;
1325 	lp->d_checksum = dkcksum(lp);
1326 }
1327 
1328 /*
1329  * Read the disklabel from a vnd.  If one is not present, create a fake one.
1330  */
1331 void
1332 vndgetdisklabel(dev)
1333 	dev_t dev;
1334 {
1335 	struct vnd_softc *sc = &vnd_softc[vndunit(dev)];
1336 	const char *errstring;
1337 	struct disklabel *lp = sc->sc_dkdev.dk_label;
1338 	struct cpu_disklabel *clp = sc->sc_dkdev.dk_cpulabel;
1339 	int i;
1340 
1341 	memset(clp, 0, sizeof(*clp));
1342 
1343 	vndgetdefaultlabel(sc, lp);
1344 
1345 	/*
1346 	 * Call the generic disklabel extraction routine.
1347 	 */
1348 	errstring = readdisklabel(VNDLABELDEV(dev), vndstrategy, lp, clp);
1349 	if (errstring) {
1350 		/*
1351 		 * Lack of disklabel is common, but we print the warning
1352 		 * anyway, since it might contain other useful information.
1353 		 */
1354 		printf("%s: %s\n", sc->sc_xname, errstring);
1355 
1356 		/*
1357 		 * For historical reasons, if there's no disklabel
1358 		 * present, all partitions must be FS_BSDFFS and
1359 		 * occupy the entire disk.
1360 		 */
1361 		for (i = 0; i < MAXPARTITIONS; i++) {
1362 			/*
1363 			 * Don't wipe out port specific hack (such as
1364 			 * dos partition hack of i386 port).
1365 			 */
1366 			if (lp->d_partitions[i].p_size != 0)
1367 				continue;
1368 
1369 			lp->d_partitions[i].p_size = lp->d_secperunit;
1370 			lp->d_partitions[i].p_offset = 0;
1371 			lp->d_partitions[i].p_fstype = FS_BSDFFS;
1372 		}
1373 
1374 		strncpy(lp->d_packname, "default label",
1375 		    sizeof(lp->d_packname));
1376 
1377 		lp->d_npartitions = MAXPARTITIONS;
1378 		lp->d_checksum = dkcksum(lp);
1379 	}
1380 
1381 	/* In-core label now valid. */
1382 	sc->sc_flags |= VNF_VLABEL;
1383 }
1384 
1385 /*
1386  * Wait interruptibly for an exclusive lock.
1387  *
1388  * XXX
1389  * Several drivers do this; it should be abstracted and made MP-safe.
1390  */
1391 static int
1392 vndlock(sc)
1393 	struct vnd_softc *sc;
1394 {
1395 	int error;
1396 
1397 	while ((sc->sc_flags & VNF_LOCKED) != 0) {
1398 		sc->sc_flags |= VNF_WANTED;
1399 		if ((error = tsleep(sc, PRIBIO | PCATCH, "vndlck", 0)) != 0)
1400 			return (error);
1401 	}
1402 	sc->sc_flags |= VNF_LOCKED;
1403 	return (0);
1404 }
1405 
1406 /*
1407  * Unlock and wake up any waiters.
1408  */
1409 static void
1410 vndunlock(sc)
1411 	struct vnd_softc *sc;
1412 {
1413 
1414 	sc->sc_flags &= ~VNF_LOCKED;
1415 	if ((sc->sc_flags & VNF_WANTED) != 0) {
1416 		sc->sc_flags &= ~VNF_WANTED;
1417 		wakeup(sc);
1418 	}
1419 }
1420