xref: /netbsd-src/sys/dev/vnd.c (revision 20e85ad185ab16980f1219a557c42e057edb42ea)
1 /*	$NetBSD: vnd.c,v 1.114 2005/03/31 18:02:16 bouyer 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 1.114 2005/03/31 18:02:16 bouyer 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/kthread.h>
147 #include <sys/errno.h>
148 #include <sys/buf.h>
149 #include <sys/bufq.h>
150 #include <sys/malloc.h>
151 #include <sys/ioctl.h>
152 #include <sys/disklabel.h>
153 #include <sys/device.h>
154 #include <sys/disk.h>
155 #include <sys/stat.h>
156 #include <sys/mount.h>
157 #include <sys/vnode.h>
158 #include <sys/file.h>
159 #include <sys/uio.h>
160 #include <sys/conf.h>
161 
162 #include <miscfs/specfs/specdev.h>
163 
164 #include <dev/vndvar.h>
165 
166 #if defined(VNDDEBUG) && !defined(DEBUG)
167 #define	DEBUG
168 #endif
169 
170 #ifdef DEBUG
171 int dovndcluster = 1;
172 #define	VDB_FOLLOW	0x01
173 #define	VDB_INIT	0x02
174 #define	VDB_IO		0x04
175 #define	VDB_LABEL	0x08
176 int vnddebug = 0x00;
177 #endif
178 
179 #define	vndunit(x)	DISKUNIT(x)
180 
181 struct vndxfer {
182 	struct buf	*vx_bp;		/* Pointer to parent buffer */
183 	int		vx_error;
184 	int		vx_pending;	/* # of pending aux buffers */
185 	int		vx_flags;
186 #define VX_BUSY		1
187 };
188 
189 struct vndbuf {
190 	struct buf	vb_buf;
191 	struct vndxfer	*vb_xfer;
192 };
193 
194 #define	VND_GETXFER(vnd)	pool_get(&(vnd)->sc_vxpool, PR_WAITOK)
195 #define	VND_PUTXFER(vnd, vx)	pool_put(&(vnd)->sc_vxpool, (vx))
196 
197 #define	VND_GETBUF(vnd)		pool_get(&(vnd)->sc_vbpool, PR_WAITOK)
198 #define	VND_PUTBUF(vnd, vb)	pool_put(&(vnd)->sc_vbpool, (vb))
199 
200 struct vnd_softc *vnd_softc;
201 int numvnd = 0;
202 
203 #define	VNDLABELDEV(dev) \
204 	(MAKEDISKDEV(major((dev)), vndunit((dev)), RAW_PART))
205 
206 /* called by main() at boot time (XXX: and the LKM driver) */
207 void	vndattach(int);
208 int	vnddetach(void);
209 
210 static void	vndclear(struct vnd_softc *, int);
211 static int	vndsetcred(struct vnd_softc *, struct ucred *);
212 static void	vndthrottle(struct vnd_softc *, struct vnode *);
213 static void	vndiodone(struct buf *);
214 #if 0
215 static void	vndshutdown(void);
216 #endif
217 
218 static void	vndgetdefaultlabel(struct vnd_softc *, struct disklabel *);
219 static void	vndgetdisklabel(dev_t);
220 
221 static int	vndlock(struct vnd_softc *);
222 static void	vndunlock(struct vnd_softc *);
223 
224 void vndthread(void *);
225 
226 static dev_type_open(vndopen);
227 static dev_type_close(vndclose);
228 static dev_type_read(vndread);
229 static dev_type_write(vndwrite);
230 static dev_type_ioctl(vndioctl);
231 static dev_type_strategy(vndstrategy);
232 static dev_type_dump(vnddump);
233 static dev_type_size(vndsize);
234 
235 const struct bdevsw vnd_bdevsw = {
236 	vndopen, vndclose, vndstrategy, vndioctl, vnddump, vndsize, D_DISK
237 };
238 
239 const struct cdevsw vnd_cdevsw = {
240 	vndopen, vndclose, vndread, vndwrite, vndioctl,
241 	nostop, notty, nopoll, nommap, nokqfilter, D_DISK
242 };
243 
244 static int vndattached;
245 
246 void
247 vndattach(int num)
248 {
249 	int i;
250 	char *mem;
251 
252 	if (vndattached)
253 		return;
254 	vndattached = 1;
255 	if (num <= 0)
256 		return;
257 	i = num * sizeof(struct vnd_softc);
258 	mem = malloc(i, M_DEVBUF, M_NOWAIT|M_ZERO);
259 	if (mem == NULL) {
260 		printf("WARNING: no memory for vnode disks\n");
261 		return;
262 	}
263 	vnd_softc = (struct vnd_softc *)mem;
264 	numvnd = num;
265 
266 	for (i = 0; i < numvnd; i++) {
267 		vnd_softc[i].sc_unit = i;
268 		bufq_alloc(&vnd_softc[i].sc_tab,
269 		    BUFQ_DISKSORT|BUFQ_SORT_RAWBLOCK);
270 	}
271 }
272 
273 int
274 vnddetach(void)
275 {
276 	int i;
277 
278 	/* First check we aren't in use. */
279 	for (i = 0; i < numvnd; i++)
280 		if (vnd_softc[i].sc_flags & VNF_INITED)
281 			return (EBUSY);
282 
283 	for (i = 0; i < numvnd; i++)
284 		bufq_free(&vnd_softc[i].sc_tab);
285 
286 	free(vnd_softc, M_DEVBUF);
287 	vndattached = 0;
288 
289 	return (0);
290 }
291 
292 static int
293 vndopen(dev_t dev, int flags, int mode, struct proc *p)
294 {
295 	int unit = vndunit(dev);
296 	struct vnd_softc *sc;
297 	int error = 0, part, pmask;
298 	struct disklabel *lp;
299 
300 #ifdef DEBUG
301 	if (vnddebug & VDB_FOLLOW)
302 		printf("vndopen(0x%x, 0x%x, 0x%x, %p)\n", dev, flags, mode, p);
303 #endif
304 	if (unit >= numvnd)
305 		return (ENXIO);
306 	sc = &vnd_softc[unit];
307 
308 	if ((error = vndlock(sc)) != 0)
309 		return (error);
310 
311 	lp = sc->sc_dkdev.dk_label;
312 
313 	part = DISKPART(dev);
314 	pmask = (1 << part);
315 
316 	/*
317 	 * If we're initialized, check to see if there are any other
318 	 * open partitions.  If not, then it's safe to update the
319 	 * in-core disklabel.  Only read the disklabel if it is
320 	 * not realdy valid.
321 	 */
322 	if ((sc->sc_flags & (VNF_INITED|VNF_VLABEL)) == VNF_INITED &&
323 	    sc->sc_dkdev.dk_openmask == 0)
324 		vndgetdisklabel(dev);
325 
326 	/* Check that the partitions exists. */
327 	if (part != RAW_PART) {
328 		if (((sc->sc_flags & VNF_INITED) == 0) ||
329 		    ((part >= lp->d_npartitions) ||
330 		     (lp->d_partitions[part].p_fstype == FS_UNUSED))) {
331 			error = ENXIO;
332 			goto done;
333 		}
334 	}
335 
336 	/* Prevent our unit from being unconfigured while open. */
337 	switch (mode) {
338 	case S_IFCHR:
339 		sc->sc_dkdev.dk_copenmask |= pmask;
340 		break;
341 
342 	case S_IFBLK:
343 		sc->sc_dkdev.dk_bopenmask |= pmask;
344 		break;
345 	}
346 	sc->sc_dkdev.dk_openmask =
347 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
348 
349  done:
350 	vndunlock(sc);
351 	return (error);
352 }
353 
354 static int
355 vndclose(dev_t dev, int flags, int mode, struct proc *p)
356 {
357 	int unit = vndunit(dev);
358 	struct vnd_softc *sc;
359 	int error = 0, part;
360 
361 #ifdef DEBUG
362 	if (vnddebug & VDB_FOLLOW)
363 		printf("vndclose(0x%x, 0x%x, 0x%x, %p)\n", dev, flags, mode, p);
364 #endif
365 
366 	if (unit >= numvnd)
367 		return (ENXIO);
368 	sc = &vnd_softc[unit];
369 
370 	if ((error = vndlock(sc)) != 0)
371 		return (error);
372 
373 	part = DISKPART(dev);
374 
375 	/* ...that much closer to allowing unconfiguration... */
376 	switch (mode) {
377 	case S_IFCHR:
378 		sc->sc_dkdev.dk_copenmask &= ~(1 << part);
379 		break;
380 
381 	case S_IFBLK:
382 		sc->sc_dkdev.dk_bopenmask &= ~(1 << part);
383 		break;
384 	}
385 	sc->sc_dkdev.dk_openmask =
386 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
387 
388 	if (sc->sc_dkdev.dk_openmask == 0) {
389 		if ((sc->sc_flags & VNF_KLABEL) == 0)
390 			sc->sc_flags &= ~VNF_VLABEL;
391 	}
392 
393 	vndunlock(sc);
394 	return (0);
395 }
396 
397 /*
398  * Qeue the request, and wakeup the kernel thread to handle it.
399  */
400 static void
401 vndstrategy(struct buf *bp)
402 {
403 	int unit = vndunit(bp->b_dev);
404 	struct vnd_softc *vnd = &vnd_softc[unit];
405 	struct disklabel *lp = vnd->sc_dkdev.dk_label;
406 	int s = splbio();
407 
408 	bp->b_resid = bp->b_bcount;
409 
410 	if ((vnd->sc_flags & VNF_INITED) == 0) {
411 		bp->b_error = ENXIO;
412 		bp->b_flags |= B_ERROR;
413 		goto done;
414 	}
415 
416 	/*
417 	 * The transfer must be a whole number of blocks.
418 	 */
419 	if ((bp->b_bcount % lp->d_secsize) != 0) {
420 		bp->b_error = EINVAL;
421 		bp->b_flags |= B_ERROR;
422 		goto done;
423 	}
424 
425 	/*
426 	 * check if we're read-only.
427 	 */
428 	if ((vnd->sc_flags & VNF_READONLY) && !(bp->b_flags & B_READ)) {
429 		bp->b_error = EACCES;
430 		bp->b_flags |= B_ERROR;
431 		goto done;
432 	}
433 
434 	/*
435 	 * Do bounds checking and adjust transfer.  If there's an error,
436 	 * the bounds check will flag that for us.
437 	 */
438 	if (DISKPART(bp->b_dev) != RAW_PART) {
439 		if (bounds_check_with_label(&vnd->sc_dkdev,
440 		    bp, vnd->sc_flags & (VNF_WLABEL|VNF_LABELLING)) <= 0)
441 			goto done;
442 	}
443 
444 	/* If it's a nil transfer, wake up the top half now. */
445 	if (bp->b_bcount == 0)
446 		goto done;
447 #ifdef DEBUG
448 	if (vnddebug & VDB_FOLLOW)
449 		printf("vndstrategy(%p): unit %d\n", bp, unit);
450 #endif
451 	BUFQ_PUT(&vnd->sc_tab, bp);
452 	wakeup(&vnd->sc_tab);
453 	splx(s);
454 	return;
455 done:
456 	biodone(bp);
457 	splx(s);
458 }
459 
460 void
461 vndthread(void *arg)
462 {
463 	struct vnd_softc *vnd = arg;
464 	struct buf *bp;
465 	struct vndxfer *vnx;
466 	struct mount *mp;
467 	int s, bsize, resid;
468 	off_t bn;
469 	caddr_t addr;
470 	int sz, flags, error;
471 	struct disklabel *lp;
472 	struct partition *pp;
473 
474 	s = splbio();
475 	vnd->sc_flags |= VNF_KTHREAD;
476 	wakeup(&vnd->sc_kthread);
477 
478 	/*
479 	 * Dequeue requests, break them into bsize pieces and submit using
480 	 * VOP_BMAP/VOP_STRATEGY.
481 	 */
482 	while ((vnd->sc_flags & VNF_VUNCONF) == 0) {
483 		bp = BUFQ_GET(&vnd->sc_tab);
484 		if (bp == NULL) {
485 			tsleep(&vnd->sc_tab, PRIBIO, "vndbp", 0);
486 			continue;
487 		};
488 		splx(s);
489 
490 #ifdef DEBUG
491 		if (vnddebug & VDB_FOLLOW)
492 			printf("vndthread(%p\n", bp);
493 #endif
494 		lp = vnd->sc_dkdev.dk_label;
495 		bp->b_resid = bp->b_bcount;
496 
497 		/*
498 		 * Put the block number in terms of the logical blocksize
499 		 * of the "device".
500 		 */
501 		bn = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
502 
503 		/*
504 		 * Translate the partition-relative block number to an absolute.
505 		 */
506 		if (DISKPART(bp->b_dev) != RAW_PART) {
507 			pp = &vnd->sc_dkdev.dk_label->d_partitions[
508 			    DISKPART(bp->b_dev)];
509 			bn += pp->p_offset;
510 		}
511 
512 		/* ...and convert to a byte offset within the file. */
513 		bn *= lp->d_secsize;
514 
515 		if (vnd->sc_vp->v_mount == NULL) {
516 			bp->b_error = ENXIO;
517 			bp->b_flags |= B_ERROR;
518 			goto done;
519 		}
520  		bsize = vnd->sc_vp->v_mount->mnt_stat.f_iosize;
521 		addr = bp->b_data;
522 		flags = (bp->b_flags & (B_READ|B_ASYNC)) | B_CALL;
523 
524 		/*
525 		 * Allocate a header for this transfer and link it to the
526 		 * buffer
527 		 */
528 		s = splbio();
529 		vnx = VND_GETXFER(vnd);
530 		splx(s);
531 		vnx->vx_flags = VX_BUSY;
532 		vnx->vx_error = 0;
533 		vnx->vx_pending = 0;
534 		vnx->vx_bp = bp;
535 
536 		if ((flags & B_READ) == 0)
537 			vn_start_write(vnd->sc_vp, &mp, V_WAIT);
538 
539 		/*
540 		 * Feed requests sequentially.
541 		 * We do it this way to keep from flooding NFS servers if we
542 		 * are connected to an NFS file.  This places the burden on
543 		 * the client rather than the server.
544 		 */
545 		for (resid = bp->b_resid; resid; resid -= sz) {
546 			struct vndbuf *nbp;
547 			struct vnode *vp;
548 			daddr_t nbn;
549 			int off, nra;
550 
551 			nra = 0;
552 			vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY | LK_CANRECURSE);
553 			error = VOP_BMAP(vnd->sc_vp, bn / bsize, &vp, &nbn, &nra);
554 			VOP_UNLOCK(vnd->sc_vp, 0);
555 
556 			if (error == 0 && (long)nbn == -1)
557 				error = EIO;
558 
559 			/*
560 			 * If there was an error or a hole in the file...punt.
561 			 * Note that we may have to wait for any operations
562 			 * that we have already fired off before releasing
563 			 * the buffer.
564 			 *
565 			 * XXX we could deal with holes here but it would be
566 			 * a hassle (in the write case).
567 			 */
568 			if (error) {
569 				s = splbio();
570 				vnx->vx_error = error;
571 				goto out;
572 			}
573 
574 #ifdef DEBUG
575 			if (!dovndcluster)
576 				nra = 0;
577 #endif
578 
579 			if ((off = bn % bsize) != 0)
580 				sz = bsize - off;
581 			else
582 				sz = (1 + nra) * bsize;
583 			if (resid < sz)
584 				sz = resid;
585 #ifdef 	DEBUG
586 			if (vnddebug & VDB_IO)
587 				printf("vndstrategy: vp %p/%p bn 0x%qx/0x%" PRIx64
588 				       " sz 0x%x\n",
589 				    vnd->sc_vp, vp, (long long)bn, nbn, sz);
590 #endif
591 
592 			s = splbio();
593 			while (vnd->sc_active >= vnd->sc_maxactive) {
594 				tsleep(&vnd->sc_tab, PRIBIO, "vndac", 0);
595 			}
596 			vnd->sc_active++;
597 			nbp = VND_GETBUF(vnd);
598 			splx(s);
599 			BUF_INIT(&nbp->vb_buf);
600 			nbp->vb_buf.b_flags = flags;
601 			nbp->vb_buf.b_bcount = sz;
602 			nbp->vb_buf.b_bufsize = round_page((ulong)addr + sz)
603 			    - trunc_page((ulong) addr);
604 			nbp->vb_buf.b_error = 0;
605 			nbp->vb_buf.b_data = addr;
606 			nbp->vb_buf.b_blkno = nbp->vb_buf.b_rawblkno = nbn + btodb(off);
607 			nbp->vb_buf.b_proc = bp->b_proc;
608 			nbp->vb_buf.b_iodone = vndiodone;
609 			nbp->vb_buf.b_vp = vp;
610 
611 			nbp->vb_xfer = vnx;
612 
613 			BIO_COPYPRIO(&nbp->vb_buf, bp);
614 
615 			/*
616 			 * Just sort by block number
617 			 */
618 			s = splbio();
619 			if (vnx->vx_error != 0) {
620 				VND_PUTBUF(vnd, nbp);
621 				goto out;
622 			}
623 			vnx->vx_pending++;
624 #ifdef DEBUG
625 			if (vnddebug & VDB_IO)
626 				printf("vndstart(%ld): bp %p vp %p blkno "
627 				    "0x%" PRIx64 " flags %x addr %p cnt 0x%x\n",
628 				    (long) (vnd-vnd_softc), &nbp->vb_buf,
629 				    nbp->vb_buf.b_vp, nbp->vb_buf.b_blkno,
630 				    nbp->vb_buf.b_flags, nbp->vb_buf.b_data,
631 				    nbp->vb_buf.b_bcount);
632 #endif
633 
634 			/* Instrumentation. */
635 			disk_busy(&vnd->sc_dkdev);
636 
637 			if ((nbp->vb_buf.b_flags & B_READ) == 0)
638 				vp->v_numoutput++;
639 			VOP_STRATEGY(vp, &nbp->vb_buf);
640 
641 			splx(s);
642 			bn += sz;
643 			addr += sz;
644 		}
645 
646 		s = splbio();
647 
648 out: /* Arrive here at splbio */
649 		if ((flags & B_READ) == 0)
650 			vn_finished_write(mp, 0);
651 		vnx->vx_flags &= ~VX_BUSY;
652 		if (vnx->vx_pending == 0) {
653 			if (vnx->vx_error != 0) {
654 				bp->b_error = vnx->vx_error;
655 				bp->b_flags |= B_ERROR;
656 			}
657 			VND_PUTXFER(vnd, vnx);
658 			biodone(bp);
659 		}
660 		continue;
661 done:
662 		biodone(bp);
663 		s = splbio();
664 	}
665 
666 	vnd->sc_flags &= (~VNF_KTHREAD | VNF_VUNCONF);
667 	wakeup(&vnd->sc_kthread);
668 	splx(s);
669 	kthread_exit(0);
670 }
671 
672 
673 static void
674 vndiodone(struct buf *bp)
675 {
676 	struct vndbuf *vbp = (struct vndbuf *) bp;
677 	struct vndxfer *vnx = (struct vndxfer *)vbp->vb_xfer;
678 	struct buf *pbp = vnx->vx_bp;
679 	struct vnd_softc *vnd = &vnd_softc[vndunit(pbp->b_dev)];
680 	int s, resid;
681 
682 	s = splbio();
683 #ifdef DEBUG
684 	if (vnddebug & VDB_IO)
685 		printf("vndiodone(%ld): vbp %p vp %p blkno 0x%" PRIx64
686 		       " addr %p cnt 0x%x\n",
687 		    (long) (vnd-vnd_softc), vbp, vbp->vb_buf.b_vp,
688 		    vbp->vb_buf.b_blkno, vbp->vb_buf.b_data,
689 		    vbp->vb_buf.b_bcount);
690 #endif
691 
692 	resid = vbp->vb_buf.b_bcount - vbp->vb_buf.b_resid;
693 	pbp->b_resid -= resid;
694 	disk_unbusy(&vnd->sc_dkdev, resid, (pbp->b_flags & B_READ));
695 	vnx->vx_pending--;
696 
697 	if (vbp->vb_buf.b_error) {
698 #ifdef DEBUG
699 		if (vnddebug & VDB_IO)
700 			printf("vndiodone: vbp %p error %d\n", vbp,
701 			    vbp->vb_buf.b_error);
702 #endif
703 		vnx->vx_error = vbp->vb_buf.b_error;
704 	}
705 
706 	VND_PUTBUF(vnd, vbp);
707 
708 	/*
709 	 * Wrap up this transaction if it has run to completion or, in
710 	 * case of an error, when all auxiliary buffers have returned.
711 	 */
712 	if (vnx->vx_error != 0) {
713 		pbp->b_flags |= B_ERROR;
714 		pbp->b_error = vnx->vx_error;
715 		if ((vnx->vx_flags & VX_BUSY) == 0 && vnx->vx_pending == 0) {
716 
717 #ifdef DEBUG
718 			if (vnddebug & VDB_IO)
719 				printf("vndiodone: pbp %p iodone: error %d\n",
720 					pbp, vnx->vx_error);
721 #endif
722 			VND_PUTXFER(vnd, vnx);
723 			biodone(pbp);
724 		}
725 	} else if (pbp->b_resid == 0) {
726 
727 #ifdef DIAGNOSTIC
728 		if (vnx->vx_pending != 0)
729 			panic("vndiodone: vnx pending: %d", vnx->vx_pending);
730 #endif
731 
732 		if ((vnx->vx_flags & VX_BUSY) == 0) {
733 #ifdef DEBUG
734 			if (vnddebug & VDB_IO)
735 				printf("vndiodone: pbp %p iodone\n", pbp);
736 #endif
737 			VND_PUTXFER(vnd, vnx);
738 			biodone(pbp);
739 		}
740 	}
741 
742 	vnd->sc_active--;
743 	wakeup(&vnd->sc_tab);
744 	splx(s);
745 }
746 
747 /* ARGSUSED */
748 static int
749 vndread(dev_t dev, struct uio *uio, int flags)
750 {
751 	int unit = vndunit(dev);
752 	struct vnd_softc *sc;
753 
754 #ifdef DEBUG
755 	if (vnddebug & VDB_FOLLOW)
756 		printf("vndread(0x%x, %p)\n", dev, uio);
757 #endif
758 
759 	if (unit >= numvnd)
760 		return (ENXIO);
761 	sc = &vnd_softc[unit];
762 
763 	if ((sc->sc_flags & VNF_INITED) == 0)
764 		return (ENXIO);
765 
766 	return (physio(vndstrategy, NULL, dev, B_READ, minphys, uio));
767 }
768 
769 /* ARGSUSED */
770 static int
771 vndwrite(dev_t dev, struct uio *uio, int flags)
772 {
773 	int unit = vndunit(dev);
774 	struct vnd_softc *sc;
775 
776 #ifdef DEBUG
777 	if (vnddebug & VDB_FOLLOW)
778 		printf("vndwrite(0x%x, %p)\n", dev, uio);
779 #endif
780 
781 	if (unit >= numvnd)
782 		return (ENXIO);
783 	sc = &vnd_softc[unit];
784 
785 	if ((sc->sc_flags & VNF_INITED) == 0)
786 		return (ENXIO);
787 
788 	return (physio(vndstrategy, NULL, dev, B_WRITE, minphys, uio));
789 }
790 
791 /* ARGSUSED */
792 static int
793 vndioctl(dev_t dev, u_long cmd, caddr_t data, int flag, struct proc *p)
794 {
795 	int unit = vndunit(dev);
796 	struct vnd_softc *vnd;
797 	struct vnd_ioctl *vio;
798 	struct vattr vattr;
799 	struct nameidata nd;
800 	int error, part, pmask;
801 	size_t geomsize;
802 	int fflags;
803 #ifdef __HAVE_OLD_DISKLABEL
804 	struct disklabel newlabel;
805 #endif
806 
807 #ifdef DEBUG
808 	if (vnddebug & VDB_FOLLOW)
809 		printf("vndioctl(0x%x, 0x%lx, %p, 0x%x, %p): unit %d\n",
810 		    dev, cmd, data, flag, p, unit);
811 #endif
812 	if (unit >= numvnd)
813 		return (ENXIO);
814 
815 	vnd = &vnd_softc[unit];
816 	vio = (struct vnd_ioctl *)data;
817 
818 	/* Must be open for writes for these commands... */
819 	switch (cmd) {
820 	case VNDIOCSET:
821 	case VNDIOCCLR:
822 	case DIOCSDINFO:
823 	case DIOCWDINFO:
824 #ifdef __HAVE_OLD_DISKLABEL
825 	case ODIOCSDINFO:
826 	case ODIOCWDINFO:
827 #endif
828 	case DIOCKLABEL:
829 	case DIOCWLABEL:
830 		if ((flag & FWRITE) == 0)
831 			return (EBADF);
832 	}
833 
834 	/* Must be initialized for these... */
835 	switch (cmd) {
836 	case VNDIOCCLR:
837 	case DIOCGDINFO:
838 	case DIOCSDINFO:
839 	case DIOCWDINFO:
840 	case DIOCGPART:
841 	case DIOCKLABEL:
842 	case DIOCWLABEL:
843 	case DIOCGDEFLABEL:
844 #ifdef __HAVE_OLD_DISKLABEL
845 	case ODIOCGDINFO:
846 	case ODIOCSDINFO:
847 	case ODIOCWDINFO:
848 	case ODIOCGDEFLABEL:
849 #endif
850 		if ((vnd->sc_flags & VNF_INITED) == 0)
851 			return (ENXIO);
852 	}
853 
854 	switch (cmd) {
855 	case VNDIOCSET:
856 		if (vnd->sc_flags & VNF_INITED)
857 			return (EBUSY);
858 
859 		if ((error = vndlock(vnd)) != 0)
860 			return (error);
861 
862 		fflags = FREAD;
863 		if ((vio->vnd_flags & VNDIOF_READONLY) == 0)
864 			fflags |= FWRITE;
865 		NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, vio->vnd_file, p);
866 		if ((error = vn_open(&nd, fflags, 0)) != 0)
867 			goto unlock_and_exit;
868 		error = VOP_GETATTR(nd.ni_vp, &vattr, p->p_ucred, p);
869 		VOP_UNLOCK(nd.ni_vp, 0);
870 		if (!error && nd.ni_vp->v_type != VREG)
871 			error = EOPNOTSUPP;
872 		if (error)
873 			goto close_and_exit;
874 		vnd->sc_vp = nd.ni_vp;
875 		vnd->sc_size = btodb(vattr.va_size);	/* note truncation */
876 
877 		/*
878 		 * Use pseudo-geometry specified.  If none was provided,
879 		 * use "standard" Adaptec fictitious geometry.
880 		 */
881 		if (vio->vnd_flags & VNDIOF_HASGEOM) {
882 
883 			memcpy(&vnd->sc_geom, &vio->vnd_geom,
884 			    sizeof(vio->vnd_geom));
885 
886 			/*
887 			 * Sanity-check the sector size.
888 			 * XXX Don't allow secsize < DEV_BSIZE.  Should
889 			 * XXX we?
890 			 */
891 			if (vnd->sc_geom.vng_secsize < DEV_BSIZE ||
892 			    (vnd->sc_geom.vng_secsize % DEV_BSIZE) != 0 ||
893 			    vnd->sc_geom.vng_ncylinders == 0 ||
894 			    (vnd->sc_geom.vng_ntracks *
895 			     vnd->sc_geom.vng_nsectors) == 0) {
896 				error = EINVAL;
897 				goto close_and_exit;
898 			}
899 
900 			/*
901 			 * Compute the size (in DEV_BSIZE blocks) specified
902 			 * by the geometry.
903 			 */
904 			geomsize = (vnd->sc_geom.vng_nsectors *
905 			    vnd->sc_geom.vng_ntracks *
906 			    vnd->sc_geom.vng_ncylinders) *
907 			    (vnd->sc_geom.vng_secsize / DEV_BSIZE);
908 
909 			/*
910 			 * Sanity-check the size against the specified
911 			 * geometry.
912 			 */
913 			if (vnd->sc_size < geomsize) {
914 				error = EINVAL;
915 				goto close_and_exit;
916 			}
917 		} else {
918 			/*
919 			 * Size must be at least 2048 DEV_BSIZE blocks
920 			 * (1M) in order to use this geometry.
921 			 */
922 			if (vnd->sc_size < (32 * 64)) {
923 				error = EINVAL;
924 				goto close_and_exit;
925 			}
926 
927 			vnd->sc_geom.vng_secsize = DEV_BSIZE;
928 			vnd->sc_geom.vng_nsectors = 32;
929 			vnd->sc_geom.vng_ntracks = 64;
930 			vnd->sc_geom.vng_ncylinders = vnd->sc_size / (64 * 32);
931 		}
932 
933 		if (vio->vnd_flags & VNDIOF_READONLY) {
934 			vnd->sc_flags |= VNF_READONLY;
935 		}
936 
937 		if ((error = vndsetcred(vnd, p->p_ucred)) != 0)
938 			goto close_and_exit;
939 
940 		memset(vnd->sc_xname, 0, sizeof(vnd->sc_xname)); /* XXX */
941 		snprintf(vnd->sc_xname, sizeof(vnd->sc_xname), "vnd%d", unit);
942 
943 
944 		vndthrottle(vnd, vnd->sc_vp);
945 		vio->vnd_size = dbtob(vnd->sc_size);
946 		vnd->sc_flags |= VNF_INITED;
947 
948 		/* create the kernel thread, wait for it to be up */
949 		error = kthread_create1(vndthread, vnd, &vnd->sc_kthread,
950 		    vnd->sc_xname);
951 		if (error)
952 			goto close_and_exit;
953 		while ((vnd->sc_flags & VNF_KTHREAD) == 0) {
954 			tsleep(&vnd->sc_kthread, PRIBIO, "vndthr", 0);
955 		}
956 #ifdef DEBUG
957 		if (vnddebug & VDB_INIT)
958 			printf("vndioctl: SET vp %p size 0x%lx %d/%d/%d/%d\n",
959 			    vnd->sc_vp, (unsigned long) vnd->sc_size,
960 			    vnd->sc_geom.vng_secsize,
961 			    vnd->sc_geom.vng_nsectors,
962 			    vnd->sc_geom.vng_ntracks,
963 			    vnd->sc_geom.vng_ncylinders);
964 #endif
965 
966 		/* Attach the disk. */
967 		vnd->sc_dkdev.dk_name = vnd->sc_xname;
968 		disk_attach(&vnd->sc_dkdev);
969 
970 		/* Initialize the xfer and buffer pools. */
971 		pool_init(&vnd->sc_vxpool, sizeof(struct vndxfer), 0,
972 		    0, 0, "vndxpl", NULL);
973 		pool_init(&vnd->sc_vbpool, sizeof(struct vndbuf), 0,
974 		    0, 0, "vndbpl", NULL);
975 
976 		/* Try and read the disklabel. */
977 		vndgetdisklabel(dev);
978 
979 		vndunlock(vnd);
980 
981 		break;
982 
983 close_and_exit:
984 		(void) vn_close(nd.ni_vp, fflags, p->p_ucred, p);
985 unlock_and_exit:
986 		vndunlock(vnd);
987 		return (error);
988 
989 	case VNDIOCCLR:
990 		if ((error = vndlock(vnd)) != 0)
991 			return (error);
992 
993 		/*
994 		 * Don't unconfigure if any other partitions are open
995 		 * or if both the character and block flavors of this
996 		 * partition are open.
997 		 */
998 		part = DISKPART(dev);
999 		pmask = (1 << part);
1000 		if (((vnd->sc_dkdev.dk_openmask & ~pmask) ||
1001 		    ((vnd->sc_dkdev.dk_bopenmask & pmask) &&
1002 		    (vnd->sc_dkdev.dk_copenmask & pmask))) &&
1003 			!(vio->vnd_flags & VNDIOF_FORCE)) {
1004 			vndunlock(vnd);
1005 			return (EBUSY);
1006 		}
1007 
1008 		/*
1009 		 * XXX vndclear() might call vndclose() implicitely;
1010 		 * release lock to avoid recursion
1011 		 */
1012 		vndunlock(vnd);
1013 		vndclear(vnd, minor(dev));
1014 #ifdef DEBUG
1015 		if (vnddebug & VDB_INIT)
1016 			printf("vndioctl: CLRed\n");
1017 #endif
1018 
1019 		/* Destroy the xfer and buffer pools. */
1020 		pool_destroy(&vnd->sc_vxpool);
1021 		pool_destroy(&vnd->sc_vbpool);
1022 
1023 		/* Detatch the disk. */
1024 		disk_detach(&vnd->sc_dkdev);
1025 
1026 		break;
1027 
1028 	case VNDIOCGET: {
1029 		struct vnd_user *vnu;
1030 		struct vattr va;
1031 
1032 		vnu = (struct vnd_user *)data;
1033 
1034 		if (vnu->vnu_unit == -1)
1035 			vnu->vnu_unit = unit;
1036 		if (vnu->vnu_unit >= numvnd)
1037 			return (ENXIO);
1038 		if (vnu->vnu_unit < 0)
1039 			return (EINVAL);
1040 
1041 		vnd = &vnd_softc[vnu->vnu_unit];
1042 
1043 		if (vnd->sc_flags & VNF_INITED) {
1044 			error = VOP_GETATTR(vnd->sc_vp, &va, p->p_ucred, p);
1045 			if (error)
1046 				return (error);
1047 			vnu->vnu_dev = va.va_fsid;
1048 			vnu->vnu_ino = va.va_fileid;
1049 		}
1050 		else {
1051 			/* unused is not an error */
1052 			vnu->vnu_dev = 0;
1053 			vnu->vnu_ino = 0;
1054 		}
1055 
1056 		break;
1057 	}
1058 
1059 	case DIOCGDINFO:
1060 		*(struct disklabel *)data = *(vnd->sc_dkdev.dk_label);
1061 		break;
1062 
1063 #ifdef __HAVE_OLD_DISKLABEL
1064 	case ODIOCGDINFO:
1065 		newlabel = *(vnd->sc_dkdev.dk_label);
1066 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1067 			return ENOTTY;
1068 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1069 		break;
1070 #endif
1071 
1072 	case DIOCGPART:
1073 		((struct partinfo *)data)->disklab = vnd->sc_dkdev.dk_label;
1074 		((struct partinfo *)data)->part =
1075 		    &vnd->sc_dkdev.dk_label->d_partitions[DISKPART(dev)];
1076 		break;
1077 
1078 	case DIOCWDINFO:
1079 	case DIOCSDINFO:
1080 #ifdef __HAVE_OLD_DISKLABEL
1081 	case ODIOCWDINFO:
1082 	case ODIOCSDINFO:
1083 #endif
1084 	{
1085 		struct disklabel *lp;
1086 
1087 		if ((error = vndlock(vnd)) != 0)
1088 			return (error);
1089 
1090 		vnd->sc_flags |= VNF_LABELLING;
1091 
1092 #ifdef __HAVE_OLD_DISKLABEL
1093 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
1094 			memset(&newlabel, 0, sizeof newlabel);
1095 			memcpy(&newlabel, data, sizeof (struct olddisklabel));
1096 			lp = &newlabel;
1097 		} else
1098 #endif
1099 		lp = (struct disklabel *)data;
1100 
1101 		error = setdisklabel(vnd->sc_dkdev.dk_label,
1102 		    lp, 0, vnd->sc_dkdev.dk_cpulabel);
1103 		if (error == 0) {
1104 			if (cmd == DIOCWDINFO
1105 #ifdef __HAVE_OLD_DISKLABEL
1106 			    || cmd == ODIOCWDINFO
1107 #endif
1108 			   )
1109 				error = writedisklabel(VNDLABELDEV(dev),
1110 				    vndstrategy, vnd->sc_dkdev.dk_label,
1111 				    vnd->sc_dkdev.dk_cpulabel);
1112 		}
1113 
1114 		vnd->sc_flags &= ~VNF_LABELLING;
1115 
1116 		vndunlock(vnd);
1117 
1118 		if (error)
1119 			return (error);
1120 		break;
1121 	}
1122 
1123 	case DIOCKLABEL:
1124 		if (*(int *)data != 0)
1125 			vnd->sc_flags |= VNF_KLABEL;
1126 		else
1127 			vnd->sc_flags &= ~VNF_KLABEL;
1128 		break;
1129 
1130 	case DIOCWLABEL:
1131 		if (*(int *)data != 0)
1132 			vnd->sc_flags |= VNF_WLABEL;
1133 		else
1134 			vnd->sc_flags &= ~VNF_WLABEL;
1135 		break;
1136 
1137 	case DIOCGDEFLABEL:
1138 		vndgetdefaultlabel(vnd, (struct disklabel *)data);
1139 		break;
1140 
1141 #ifdef __HAVE_OLD_DISKLABEL
1142 	case ODIOCGDEFLABEL:
1143 		vndgetdefaultlabel(vnd, &newlabel);
1144 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1145 			return ENOTTY;
1146 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1147 		break;
1148 #endif
1149 
1150 	default:
1151 		return (ENOTTY);
1152 	}
1153 
1154 	return (0);
1155 }
1156 
1157 /*
1158  * Duplicate the current processes' credentials.  Since we are called only
1159  * as the result of a SET ioctl and only root can do that, any future access
1160  * to this "disk" is essentially as root.  Note that credentials may change
1161  * if some other uid can write directly to the mapped file (NFS).
1162  */
1163 static int
1164 vndsetcred(struct vnd_softc *vnd, struct ucred *cred)
1165 {
1166 	struct uio auio;
1167 	struct iovec aiov;
1168 	char *tmpbuf;
1169 	int error;
1170 
1171 	vnd->sc_cred = crdup(cred);
1172 	tmpbuf = malloc(DEV_BSIZE, M_TEMP, M_WAITOK);
1173 
1174 	/* XXX: Horrible kludge to establish credentials for NFS */
1175 	aiov.iov_base = tmpbuf;
1176 	aiov.iov_len = min(DEV_BSIZE, dbtob(vnd->sc_size));
1177 	auio.uio_iov = &aiov;
1178 	auio.uio_iovcnt = 1;
1179 	auio.uio_offset = 0;
1180 	auio.uio_rw = UIO_READ;
1181 	auio.uio_segflg = UIO_SYSSPACE;
1182 	auio.uio_resid = aiov.iov_len;
1183 	vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1184 	error = VOP_READ(vnd->sc_vp, &auio, 0, vnd->sc_cred);
1185 	if (error == 0) {
1186 		/*
1187 		 * Because vnd does all IO directly through the vnode
1188 		 * we need to flush (at least) the buffer from the above
1189 		 * VOP_READ from the buffer cache to prevent cache
1190 		 * incoherencies.  Also, be careful to write dirty
1191 		 * buffers back to stable storage.
1192 		 */
1193 		error = vinvalbuf(vnd->sc_vp, V_SAVE, vnd->sc_cred,
1194 			    curproc, 0, 0);
1195 	}
1196 	VOP_UNLOCK(vnd->sc_vp, 0);
1197 
1198 	free(tmpbuf, M_TEMP);
1199 	return (error);
1200 }
1201 
1202 /*
1203  * Set maxactive based on FS type
1204  */
1205 static void
1206 vndthrottle(struct vnd_softc *vnd, struct vnode *vp)
1207 {
1208 #ifdef NFS
1209 	extern int (**nfsv2_vnodeop_p)(void *);
1210 
1211 	if (vp->v_op == nfsv2_vnodeop_p)
1212 		vnd->sc_maxactive = 2;
1213 	else
1214 #endif
1215 		vnd->sc_maxactive = 8;
1216 
1217 	if (vnd->sc_maxactive < 1)
1218 		vnd->sc_maxactive = 1;
1219 }
1220 
1221 #if 0
1222 static void
1223 vndshutdown(void)
1224 {
1225 	struct vnd_softc *vnd;
1226 
1227 	for (vnd = &vnd_softc[0]; vnd < &vnd_softc[numvnd]; vnd++)
1228 		if (vnd->sc_flags & VNF_INITED)
1229 			vndclear(vnd);
1230 }
1231 #endif
1232 
1233 static void
1234 vndclear(struct vnd_softc *vnd, int myminor)
1235 {
1236 	struct vnode *vp = vnd->sc_vp;
1237 	struct proc *p = curproc;		/* XXX */
1238 	int fflags = FREAD;
1239 	int bmaj, cmaj, i, mn;
1240 	int s;
1241 
1242 #ifdef DEBUG
1243 	if (vnddebug & VDB_FOLLOW)
1244 		printf("vndclear(%p): vp %p\n", vnd, vp);
1245 #endif
1246 	/* locate the major number */
1247 	bmaj = bdevsw_lookup_major(&vnd_bdevsw);
1248 	cmaj = cdevsw_lookup_major(&vnd_cdevsw);
1249 
1250 	/* Nuke the vnodes for any open instances */
1251 	for (i = 0; i < MAXPARTITIONS; i++) {
1252 		mn = DISKMINOR(vnd->sc_unit, i);
1253 		vdevgone(bmaj, mn, mn, VBLK);
1254 		if (mn != myminor) /* XXX avoid to kill own vnode */
1255 			vdevgone(cmaj, mn, mn, VCHR);
1256 	}
1257 
1258 	if ((vnd->sc_flags & VNF_READONLY) == 0)
1259 		fflags |= FWRITE;
1260 
1261 	s = splbio();
1262 	bufq_drain(&vnd->sc_tab);
1263 	splx(s);
1264 
1265 	vnd->sc_flags |= VNF_VUNCONF;
1266 	wakeup(&vnd->sc_tab);
1267 	while (vnd->sc_flags & VNF_KTHREAD)
1268 		tsleep(&vnd->sc_kthread, PRIBIO, "vnthr", 0);
1269 
1270 	vnd->sc_flags &=
1271 	    ~(VNF_INITED | VNF_READONLY | VNF_VLABEL | VNF_VUNCONF);
1272 	if (vp == (struct vnode *)0)
1273 		panic("vndclear: null vp");
1274 	(void) vn_close(vp, fflags, vnd->sc_cred, p);
1275 	crfree(vnd->sc_cred);
1276 	vnd->sc_vp = (struct vnode *)0;
1277 	vnd->sc_cred = (struct ucred *)0;
1278 	vnd->sc_size = 0;
1279 }
1280 
1281 static int
1282 vndsize(dev_t dev)
1283 {
1284 	struct vnd_softc *sc;
1285 	struct disklabel *lp;
1286 	int part, unit, omask;
1287 	int size;
1288 
1289 	unit = vndunit(dev);
1290 	if (unit >= numvnd)
1291 		return (-1);
1292 	sc = &vnd_softc[unit];
1293 
1294 	if ((sc->sc_flags & VNF_INITED) == 0)
1295 		return (-1);
1296 
1297 	part = DISKPART(dev);
1298 	omask = sc->sc_dkdev.dk_openmask & (1 << part);
1299 	lp = sc->sc_dkdev.dk_label;
1300 
1301 	if (omask == 0 && vndopen(dev, 0, S_IFBLK, curproc))
1302 		return (-1);
1303 
1304 	if (lp->d_partitions[part].p_fstype != FS_SWAP)
1305 		size = -1;
1306 	else
1307 		size = lp->d_partitions[part].p_size *
1308 		    (lp->d_secsize / DEV_BSIZE);
1309 
1310 	if (omask == 0 && vndclose(dev, 0, S_IFBLK, curproc))
1311 		return (-1);
1312 
1313 	return (size);
1314 }
1315 
1316 static int
1317 vnddump(dev_t dev, daddr_t blkno, caddr_t va, size_t size)
1318 {
1319 
1320 	/* Not implemented. */
1321 	return ENXIO;
1322 }
1323 
1324 static void
1325 vndgetdefaultlabel(struct vnd_softc *sc, struct disklabel *lp)
1326 {
1327 	struct vndgeom *vng = &sc->sc_geom;
1328 	struct partition *pp;
1329 
1330 	memset(lp, 0, sizeof(*lp));
1331 
1332 	lp->d_secperunit = sc->sc_size / (vng->vng_secsize / DEV_BSIZE);
1333 	lp->d_secsize = vng->vng_secsize;
1334 	lp->d_nsectors = vng->vng_nsectors;
1335 	lp->d_ntracks = vng->vng_ntracks;
1336 	lp->d_ncylinders = vng->vng_ncylinders;
1337 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1338 
1339 	strncpy(lp->d_typename, "vnd", sizeof(lp->d_typename));
1340 	lp->d_type = DTYPE_VND;
1341 	strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
1342 	lp->d_rpm = 3600;
1343 	lp->d_interleave = 1;
1344 	lp->d_flags = 0;
1345 
1346 	pp = &lp->d_partitions[RAW_PART];
1347 	pp->p_offset = 0;
1348 	pp->p_size = lp->d_secperunit;
1349 	pp->p_fstype = FS_UNUSED;
1350 	lp->d_npartitions = RAW_PART + 1;
1351 
1352 	lp->d_magic = DISKMAGIC;
1353 	lp->d_magic2 = DISKMAGIC;
1354 	lp->d_checksum = dkcksum(lp);
1355 }
1356 
1357 /*
1358  * Read the disklabel from a vnd.  If one is not present, create a fake one.
1359  */
1360 static void
1361 vndgetdisklabel(dev_t dev)
1362 {
1363 	struct vnd_softc *sc = &vnd_softc[vndunit(dev)];
1364 	const char *errstring;
1365 	struct disklabel *lp = sc->sc_dkdev.dk_label;
1366 	struct cpu_disklabel *clp = sc->sc_dkdev.dk_cpulabel;
1367 	int i;
1368 
1369 	memset(clp, 0, sizeof(*clp));
1370 
1371 	vndgetdefaultlabel(sc, lp);
1372 
1373 	/*
1374 	 * Call the generic disklabel extraction routine.
1375 	 */
1376 	errstring = readdisklabel(VNDLABELDEV(dev), vndstrategy, lp, clp);
1377 	if (errstring) {
1378 		/*
1379 		 * Lack of disklabel is common, but we print the warning
1380 		 * anyway, since it might contain other useful information.
1381 		 */
1382 		printf("%s: %s\n", sc->sc_xname, errstring);
1383 
1384 		/*
1385 		 * For historical reasons, if there's no disklabel
1386 		 * present, all partitions must be FS_BSDFFS and
1387 		 * occupy the entire disk.
1388 		 */
1389 		for (i = 0; i < MAXPARTITIONS; i++) {
1390 			/*
1391 			 * Don't wipe out port specific hack (such as
1392 			 * dos partition hack of i386 port).
1393 			 */
1394 			if (lp->d_partitions[i].p_size != 0)
1395 				continue;
1396 
1397 			lp->d_partitions[i].p_size = lp->d_secperunit;
1398 			lp->d_partitions[i].p_offset = 0;
1399 			lp->d_partitions[i].p_fstype = FS_BSDFFS;
1400 		}
1401 
1402 		strncpy(lp->d_packname, "default label",
1403 		    sizeof(lp->d_packname));
1404 
1405 		lp->d_npartitions = MAXPARTITIONS;
1406 		lp->d_checksum = dkcksum(lp);
1407 	}
1408 
1409 	/* In-core label now valid. */
1410 	sc->sc_flags |= VNF_VLABEL;
1411 }
1412 
1413 /*
1414  * Wait interruptibly for an exclusive lock.
1415  *
1416  * XXX
1417  * Several drivers do this; it should be abstracted and made MP-safe.
1418  */
1419 static int
1420 vndlock(struct vnd_softc *sc)
1421 {
1422 	int error;
1423 
1424 	while ((sc->sc_flags & VNF_LOCKED) != 0) {
1425 		sc->sc_flags |= VNF_WANTED;
1426 		if ((error = tsleep(sc, PRIBIO | PCATCH, "vndlck", 0)) != 0)
1427 			return (error);
1428 	}
1429 	sc->sc_flags |= VNF_LOCKED;
1430 	return (0);
1431 }
1432 
1433 /*
1434  * Unlock and wake up any waiters.
1435  */
1436 static void
1437 vndunlock(struct vnd_softc *sc)
1438 {
1439 
1440 	sc->sc_flags &= ~VNF_LOCKED;
1441 	if ((sc->sc_flags & VNF_WANTED) != 0) {
1442 		sc->sc_flags &= ~VNF_WANTED;
1443 		wakeup(sc);
1444 	}
1445 }
1446