xref: /netbsd-src/sys/dev/vnd.c (revision 23c8222edbfb0f0932d88a8351d3a0cf817dfb9e)
1 /*	$NetBSD: vnd.c,v 1.111 2004/10/28 07:07:39 yamt 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 1.111 2004/10/28 07:07:39 yamt 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/bufq.h>
149 #include <sys/malloc.h>
150 #include <sys/ioctl.h>
151 #include <sys/disklabel.h>
152 #include <sys/device.h>
153 #include <sys/disk.h>
154 #include <sys/stat.h>
155 #include <sys/mount.h>
156 #include <sys/vnode.h>
157 #include <sys/file.h>
158 #include <sys/uio.h>
159 #include <sys/conf.h>
160 
161 #include <miscfs/specfs/specdev.h>
162 
163 #include <dev/vndvar.h>
164 
165 #if defined(VNDDEBUG) && !defined(DEBUG)
166 #define	DEBUG
167 #endif
168 
169 #ifdef DEBUG
170 int dovndcluster = 1;
171 #define	VDB_FOLLOW	0x01
172 #define	VDB_INIT	0x02
173 #define	VDB_IO		0x04
174 #define	VDB_LABEL	0x08
175 int vnddebug = 0x00;
176 #endif
177 
178 #define	vndunit(x)	DISKUNIT(x)
179 
180 struct vndxfer {
181 	struct buf	*vx_bp;		/* Pointer to parent buffer */
182 	int		vx_error;
183 	int		vx_pending;	/* # of pending aux buffers */
184 	int		vx_flags;
185 #define VX_BUSY		1
186 };
187 
188 struct vndbuf {
189 	struct buf	vb_buf;
190 	struct vndxfer	*vb_xfer;
191 };
192 
193 #define	VND_GETXFER(vnd)	pool_get(&(vnd)->sc_vxpool, PR_NOWAIT)
194 #define	VND_PUTXFER(vnd, vx)	pool_put(&(vnd)->sc_vxpool, (vx))
195 
196 #define	VND_GETBUF(vnd)		pool_get(&(vnd)->sc_vbpool, PR_NOWAIT)
197 #define	VND_PUTBUF(vnd, vb)	pool_put(&(vnd)->sc_vbpool, (vb))
198 
199 struct vnd_softc *vnd_softc;
200 int numvnd = 0;
201 
202 #define	VNDLABELDEV(dev) \
203 	(MAKEDISKDEV(major((dev)), vndunit((dev)), RAW_PART))
204 
205 /* called by main() at boot time (XXX: and the LKM driver) */
206 void	vndattach(int);
207 int	vnddetach(void);
208 
209 static void	vndclear(struct vnd_softc *, int);
210 static void	vndstart(struct vnd_softc *);
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 static dev_type_open(vndopen);
225 static dev_type_close(vndclose);
226 static dev_type_read(vndread);
227 static dev_type_write(vndwrite);
228 static dev_type_ioctl(vndioctl);
229 static dev_type_strategy(vndstrategy);
230 static dev_type_dump(vnddump);
231 static dev_type_size(vndsize);
232 
233 const struct bdevsw vnd_bdevsw = {
234 	vndopen, vndclose, vndstrategy, vndioctl, vnddump, vndsize, D_DISK
235 };
236 
237 const struct cdevsw vnd_cdevsw = {
238 	vndopen, vndclose, vndread, vndwrite, vndioctl,
239 	nostop, notty, nopoll, nommap, nokqfilter, D_DISK
240 };
241 
242 static int vndattached;
243 
244 void
245 vndattach(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(void)
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 static int
291 vndopen(dev_t dev, int flags, int mode, struct proc *p)
292 {
293 	int unit = vndunit(dev);
294 	struct vnd_softc *sc;
295 	int error = 0, part, pmask;
296 	struct disklabel *lp;
297 
298 #ifdef DEBUG
299 	if (vnddebug & VDB_FOLLOW)
300 		printf("vndopen(0x%x, 0x%x, 0x%x, %p)\n", dev, flags, mode, p);
301 #endif
302 	if (unit >= numvnd)
303 		return (ENXIO);
304 	sc = &vnd_softc[unit];
305 
306 	if ((error = vndlock(sc)) != 0)
307 		return (error);
308 
309 	lp = sc->sc_dkdev.dk_label;
310 
311 	part = DISKPART(dev);
312 	pmask = (1 << part);
313 
314 	/*
315 	 * If we're initialized, check to see if there are any other
316 	 * open partitions.  If not, then it's safe to update the
317 	 * in-core disklabel.  Only read the disklabel if it is
318 	 * not realdy valid.
319 	 */
320 	if ((sc->sc_flags & (VNF_INITED|VNF_VLABEL)) == VNF_INITED &&
321 	    sc->sc_dkdev.dk_openmask == 0)
322 		vndgetdisklabel(dev);
323 
324 	/* Check that the partitions exists. */
325 	if (part != RAW_PART) {
326 		if (((sc->sc_flags & VNF_INITED) == 0) ||
327 		    ((part >= lp->d_npartitions) ||
328 		     (lp->d_partitions[part].p_fstype == FS_UNUSED))) {
329 			error = ENXIO;
330 			goto done;
331 		}
332 	}
333 
334 	/* Prevent our unit from being unconfigured while open. */
335 	switch (mode) {
336 	case S_IFCHR:
337 		sc->sc_dkdev.dk_copenmask |= pmask;
338 		break;
339 
340 	case S_IFBLK:
341 		sc->sc_dkdev.dk_bopenmask |= pmask;
342 		break;
343 	}
344 	sc->sc_dkdev.dk_openmask =
345 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
346 
347  done:
348 	vndunlock(sc);
349 	return (error);
350 }
351 
352 static int
353 vndclose(dev_t dev, int flags, int mode, struct proc *p)
354 {
355 	int unit = vndunit(dev);
356 	struct vnd_softc *sc;
357 	int error = 0, part;
358 
359 #ifdef DEBUG
360 	if (vnddebug & VDB_FOLLOW)
361 		printf("vndclose(0x%x, 0x%x, 0x%x, %p)\n", dev, flags, mode, p);
362 #endif
363 
364 	if (unit >= numvnd)
365 		return (ENXIO);
366 	sc = &vnd_softc[unit];
367 
368 	if ((error = vndlock(sc)) != 0)
369 		return (error);
370 
371 	part = DISKPART(dev);
372 
373 	/* ...that much closer to allowing unconfiguration... */
374 	switch (mode) {
375 	case S_IFCHR:
376 		sc->sc_dkdev.dk_copenmask &= ~(1 << part);
377 		break;
378 
379 	case S_IFBLK:
380 		sc->sc_dkdev.dk_bopenmask &= ~(1 << part);
381 		break;
382 	}
383 	sc->sc_dkdev.dk_openmask =
384 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
385 
386 	if (sc->sc_dkdev.dk_openmask == 0) {
387 		if ((sc->sc_flags & VNF_KLABEL) == 0)
388 			sc->sc_flags &= ~VNF_VLABEL;
389 	}
390 
391 	vndunlock(sc);
392 	return (0);
393 }
394 
395 /*
396  * Break the request into bsize pieces and submit using VOP_BMAP/VOP_STRATEGY.
397  */
398 static void
399 vndstrategy(struct buf *bp)
400 {
401 	int unit = vndunit(bp->b_dev);
402 	struct vnd_softc *vnd = &vnd_softc[unit];
403 	struct vndxfer *vnx;
404 	struct mount *mp;
405 	int s, bsize, resid;
406 	off_t bn;
407 	caddr_t addr;
408 	int sz, flags, error, wlabel;
409 	struct disklabel *lp;
410 	struct partition *pp;
411 
412 #ifdef DEBUG
413 	if (vnddebug & VDB_FOLLOW)
414 		printf("vndstrategy(%p): unit %d\n", bp, unit);
415 #endif
416 	if ((vnd->sc_flags & VNF_INITED) == 0) {
417 		bp->b_error = ENXIO;
418 		bp->b_flags |= B_ERROR;
419 		goto done;
420 	}
421 
422 	/* If it's a nil transfer, wake up the top half now. */
423 	if (bp->b_bcount == 0)
424 		goto done;
425 
426 	lp = vnd->sc_dkdev.dk_label;
427 
428 	/*
429 	 * The transfer must be a whole number of blocks.
430 	 */
431 	if ((bp->b_bcount % lp->d_secsize) != 0) {
432 		bp->b_error = EINVAL;
433 		bp->b_flags |= B_ERROR;
434 		goto done;
435 	}
436 
437 	/*
438 	 * Do bounds checking and adjust transfer.  If there's an error,
439 	 * the bounds check will flag that for us.
440 	 */
441 	wlabel = vnd->sc_flags & (VNF_WLABEL|VNF_LABELLING);
442 	if (DISKPART(bp->b_dev) != RAW_PART)
443 		if (bounds_check_with_label(&vnd->sc_dkdev, bp, wlabel) <= 0)
444 			goto done;
445 
446 	/*
447 	 * check if we're read-only.
448 	 */
449 	if ((vnd->sc_flags & VNF_READONLY) && !(bp->b_flags & B_READ)) {
450 		bp->b_error = EACCES;
451 		bp->b_flags |= B_ERROR;
452 		goto done;
453 	}
454 
455 	bp->b_resid = bp->b_bcount;
456 
457 	/*
458 	 * Put the block number in terms of the logical blocksize
459 	 * of the "device".
460 	 */
461 	bn = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
462 
463 	/*
464 	 * Translate the partition-relative block number to an absolute.
465 	 */
466 	if (DISKPART(bp->b_dev) != RAW_PART) {
467 		pp = &vnd->sc_dkdev.dk_label->d_partitions[DISKPART(bp->b_dev)];
468 		bn += pp->p_offset;
469 	}
470 
471 	/* ...and convert to a byte offset within the file. */
472 	bn *= lp->d_secsize;
473 
474 	if (vnd->sc_vp->v_mount == NULL) {
475 		bp->b_error = ENXIO;
476 		bp->b_flags |= B_ERROR;
477 		goto done;
478 	}
479  	bsize = vnd->sc_vp->v_mount->mnt_stat.f_iosize;
480 	addr = bp->b_data;
481 	flags = (bp->b_flags & (B_READ|B_ASYNC)) | B_CALL;
482 
483 	/* Allocate a header for this transfer and link it to the buffer */
484 	s = splbio();
485 	vnx = VND_GETXFER(vnd);
486 	splx(s);
487 	vnx->vx_flags = VX_BUSY;
488 	vnx->vx_error = 0;
489 	vnx->vx_pending = 0;
490 	vnx->vx_bp = bp;
491 
492 	if ((flags & B_READ) == 0)
493 		vn_start_write(vnd->sc_vp, &mp, V_WAIT);
494 
495 	for (resid = bp->b_resid; resid; resid -= sz) {
496 		struct vndbuf *nbp;
497 		struct vnode *vp;
498 		daddr_t nbn;
499 		int off, nra;
500 
501 		nra = 0;
502 		vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY | LK_CANRECURSE);
503 		error = VOP_BMAP(vnd->sc_vp, bn / bsize, &vp, &nbn, &nra);
504 		VOP_UNLOCK(vnd->sc_vp, 0);
505 
506 		if (error == 0 && (long)nbn == -1)
507 			error = EIO;
508 
509 		/*
510 		 * If there was an error or a hole in the file...punt.
511 		 * Note that we may have to wait for any operations
512 		 * that we have already fired off before releasing
513 		 * the buffer.
514 		 *
515 		 * XXX we could deal with holes here but it would be
516 		 * a hassle (in the write case).
517 		 */
518 		if (error) {
519 			s = splbio();
520 			vnx->vx_error = error;
521 			goto out;
522 		}
523 
524 #ifdef DEBUG
525 		if (!dovndcluster)
526 			nra = 0;
527 #endif
528 
529 		if ((off = bn % bsize) != 0)
530 			sz = bsize - off;
531 		else
532 			sz = (1 + nra) * bsize;
533 		if (resid < sz)
534 			sz = resid;
535 #ifdef DEBUG
536 		if (vnddebug & VDB_IO)
537 			printf("vndstrategy: vp %p/%p bn 0x%qx/0x%" PRIx64
538 			       " sz 0x%x\n",
539 			    vnd->sc_vp, vp, (long long)bn, nbn, sz);
540 #endif
541 
542 		s = splbio();
543 		nbp = VND_GETBUF(vnd);
544 		splx(s);
545 		BUF_INIT(&nbp->vb_buf);
546 		nbp->vb_buf.b_flags = flags;
547 		nbp->vb_buf.b_bcount = sz;
548 		nbp->vb_buf.b_bufsize = round_page((ulong)addr + sz)
549 		    - trunc_page((ulong) addr);
550 		nbp->vb_buf.b_error = 0;
551 		nbp->vb_buf.b_data = addr;
552 		nbp->vb_buf.b_blkno = nbp->vb_buf.b_rawblkno = nbn + btodb(off);
553 		nbp->vb_buf.b_proc = bp->b_proc;
554 		nbp->vb_buf.b_iodone = vndiodone;
555 		nbp->vb_buf.b_vp = vp;
556 
557 		nbp->vb_xfer = vnx;
558 
559 		BIO_COPYPRIO(&nbp->vb_buf, bp);
560 
561 		/*
562 		 * Just sort by block number
563 		 */
564 		s = splbio();
565 		if (vnx->vx_error != 0) {
566 			VND_PUTBUF(vnd, nbp);
567 			goto out;
568 		}
569 		vnx->vx_pending++;
570 
571 		BUFQ_PUT(&vnd->sc_tab, &nbp->vb_buf);
572 		vndstart(vnd);
573 		splx(s);
574 		bn += sz;
575 		addr += sz;
576 	}
577 
578 	s = splbio();
579 
580 out: /* Arrive here at splbio */
581 	if ((flags & B_READ) == 0)
582 		vn_finished_write(mp, 0);
583 	vnx->vx_flags &= ~VX_BUSY;
584 	if (vnx->vx_pending == 0) {
585 		if (vnx->vx_error != 0) {
586 			bp->b_error = vnx->vx_error;
587 			bp->b_flags |= B_ERROR;
588 		}
589 		VND_PUTXFER(vnd, vnx);
590 		biodone(bp);
591 	}
592 	splx(s);
593 	return;
594 
595  done:
596 	biodone(bp);
597 }
598 
599 /*
600  * Feed requests sequentially.
601  * We do it this way to keep from flooding NFS servers if we are connected
602  * to an NFS file.  This places the burden on the client rather than the
603  * server.
604  */
605 static void
606 vndstart(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 %x addr %p cnt 0x%x\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->b_vp, bp);
639 	}
640 	vnd->sc_flags &= ~VNF_BUSY;
641 }
642 
643 static void
644 vndiodone(struct buf *bp)
645 {
646 	struct vndbuf *vbp = (struct vndbuf *) bp;
647 	struct vndxfer *vnx = (struct vndxfer *)vbp->vb_xfer;
648 	struct buf *pbp = vnx->vx_bp;
649 	struct vnd_softc *vnd = &vnd_softc[vndunit(pbp->b_dev)];
650 	int s, resid;
651 
652 	s = splbio();
653 #ifdef DEBUG
654 	if (vnddebug & VDB_IO)
655 		printf("vndiodone(%ld): vbp %p vp %p blkno 0x%" PRIx64
656 		       " addr %p cnt 0x%x\n",
657 		    (long) (vnd-vnd_softc), vbp, vbp->vb_buf.b_vp,
658 		    vbp->vb_buf.b_blkno, vbp->vb_buf.b_data,
659 		    vbp->vb_buf.b_bcount);
660 #endif
661 
662 	resid = vbp->vb_buf.b_bcount - vbp->vb_buf.b_resid;
663 	pbp->b_resid -= resid;
664 	disk_unbusy(&vnd->sc_dkdev, resid, (pbp->b_flags & B_READ));
665 	vnx->vx_pending--;
666 
667 	if (vbp->vb_buf.b_error) {
668 #ifdef DEBUG
669 		if (vnddebug & VDB_IO)
670 			printf("vndiodone: vbp %p error %d\n", vbp,
671 			    vbp->vb_buf.b_error);
672 #endif
673 		vnx->vx_error = vbp->vb_buf.b_error;
674 	}
675 
676 	VND_PUTBUF(vnd, vbp);
677 
678 	/*
679 	 * Wrap up this transaction if it has run to completion or, in
680 	 * case of an error, when all auxiliary buffers have returned.
681 	 */
682 	if (vnx->vx_error != 0) {
683 		pbp->b_flags |= B_ERROR;
684 		pbp->b_error = vnx->vx_error;
685 		if ((vnx->vx_flags & VX_BUSY) == 0 && vnx->vx_pending == 0) {
686 
687 #ifdef DEBUG
688 			if (vnddebug & VDB_IO)
689 				printf("vndiodone: pbp %p iodone: error %d\n",
690 					pbp, vnx->vx_error);
691 #endif
692 			VND_PUTXFER(vnd, vnx);
693 			biodone(pbp);
694 		}
695 	} else if (pbp->b_resid == 0) {
696 
697 #ifdef DIAGNOSTIC
698 		if (vnx->vx_pending != 0)
699 			panic("vndiodone: vnx pending: %d", vnx->vx_pending);
700 #endif
701 
702 		if ((vnx->vx_flags & VX_BUSY) == 0) {
703 #ifdef DEBUG
704 			if (vnddebug & VDB_IO)
705 				printf("vndiodone: pbp %p iodone\n", pbp);
706 #endif
707 			VND_PUTXFER(vnd, vnx);
708 			biodone(pbp);
709 		}
710 	}
711 
712 	vnd->sc_active--;
713 	vndstart(vnd);
714 	splx(s);
715 }
716 
717 /* ARGSUSED */
718 static int
719 vndread(dev_t dev, struct uio *uio, int flags)
720 {
721 	int unit = vndunit(dev);
722 	struct vnd_softc *sc;
723 
724 #ifdef DEBUG
725 	if (vnddebug & VDB_FOLLOW)
726 		printf("vndread(0x%x, %p)\n", dev, uio);
727 #endif
728 
729 	if (unit >= numvnd)
730 		return (ENXIO);
731 	sc = &vnd_softc[unit];
732 
733 	if ((sc->sc_flags & VNF_INITED) == 0)
734 		return (ENXIO);
735 
736 	return (physio(vndstrategy, NULL, dev, B_READ, minphys, uio));
737 }
738 
739 /* ARGSUSED */
740 static int
741 vndwrite(dev_t dev, struct uio *uio, int flags)
742 {
743 	int unit = vndunit(dev);
744 	struct vnd_softc *sc;
745 
746 #ifdef DEBUG
747 	if (vnddebug & VDB_FOLLOW)
748 		printf("vndwrite(0x%x, %p)\n", dev, uio);
749 #endif
750 
751 	if (unit >= numvnd)
752 		return (ENXIO);
753 	sc = &vnd_softc[unit];
754 
755 	if ((sc->sc_flags & VNF_INITED) == 0)
756 		return (ENXIO);
757 
758 	return (physio(vndstrategy, NULL, dev, B_WRITE, minphys, uio));
759 }
760 
761 /* ARGSUSED */
762 static int
763 vndioctl(dev_t dev, u_long cmd, caddr_t data, int flag, struct proc *p)
764 {
765 	int unit = vndunit(dev);
766 	struct vnd_softc *vnd;
767 	struct vnd_ioctl *vio;
768 	struct vattr vattr;
769 	struct nameidata nd;
770 	int error, part, pmask;
771 	size_t geomsize;
772 	int fflags;
773 #ifdef __HAVE_OLD_DISKLABEL
774 	struct disklabel newlabel;
775 #endif
776 
777 #ifdef DEBUG
778 	if (vnddebug & VDB_FOLLOW)
779 		printf("vndioctl(0x%x, 0x%lx, %p, 0x%x, %p): unit %d\n",
780 		    dev, cmd, data, flag, p, unit);
781 #endif
782 	if (unit >= numvnd)
783 		return (ENXIO);
784 
785 	vnd = &vnd_softc[unit];
786 	vio = (struct vnd_ioctl *)data;
787 
788 	/* Must be open for writes for these commands... */
789 	switch (cmd) {
790 	case VNDIOCSET:
791 	case VNDIOCCLR:
792 	case DIOCSDINFO:
793 	case DIOCWDINFO:
794 #ifdef __HAVE_OLD_DISKLABEL
795 	case ODIOCSDINFO:
796 	case ODIOCWDINFO:
797 #endif
798 	case DIOCKLABEL:
799 	case DIOCWLABEL:
800 		if ((flag & FWRITE) == 0)
801 			return (EBADF);
802 	}
803 
804 	/* Must be initialized for these... */
805 	switch (cmd) {
806 	case VNDIOCCLR:
807 	case DIOCGDINFO:
808 	case DIOCSDINFO:
809 	case DIOCWDINFO:
810 	case DIOCGPART:
811 	case DIOCKLABEL:
812 	case DIOCWLABEL:
813 	case DIOCGDEFLABEL:
814 #ifdef __HAVE_OLD_DISKLABEL
815 	case ODIOCGDINFO:
816 	case ODIOCSDINFO:
817 	case ODIOCWDINFO:
818 	case ODIOCGDEFLABEL:
819 #endif
820 		if ((vnd->sc_flags & VNF_INITED) == 0)
821 			return (ENXIO);
822 	}
823 
824 	switch (cmd) {
825 	case VNDIOCSET:
826 		if (vnd->sc_flags & VNF_INITED)
827 			return (EBUSY);
828 
829 		if ((error = vndlock(vnd)) != 0)
830 			return (error);
831 
832 		fflags = FREAD;
833 		if ((vio->vnd_flags & VNDIOF_READONLY) == 0)
834 			fflags |= FWRITE;
835 		NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, vio->vnd_file, p);
836 		if ((error = vn_open(&nd, fflags, 0)) != 0)
837 			goto unlock_and_exit;
838 		error = VOP_GETATTR(nd.ni_vp, &vattr, p->p_ucred, p);
839 		VOP_UNLOCK(nd.ni_vp, 0);
840 		if (!error && nd.ni_vp->v_type != VREG)
841 			error = EOPNOTSUPP;
842 		if (error)
843 			goto close_and_exit;
844 		vnd->sc_vp = nd.ni_vp;
845 		vnd->sc_size = btodb(vattr.va_size);	/* note truncation */
846 
847 		/*
848 		 * Use pseudo-geometry specified.  If none was provided,
849 		 * use "standard" Adaptec fictitious geometry.
850 		 */
851 		if (vio->vnd_flags & VNDIOF_HASGEOM) {
852 
853 			memcpy(&vnd->sc_geom, &vio->vnd_geom,
854 			    sizeof(vio->vnd_geom));
855 
856 			/*
857 			 * Sanity-check the sector size.
858 			 * XXX Don't allow secsize < DEV_BSIZE.  Should
859 			 * XXX we?
860 			 */
861 			if (vnd->sc_geom.vng_secsize < DEV_BSIZE ||
862 			    (vnd->sc_geom.vng_secsize % DEV_BSIZE) != 0 ||
863 			    vnd->sc_geom.vng_ncylinders == 0 ||
864 			    (vnd->sc_geom.vng_ntracks *
865 			     vnd->sc_geom.vng_nsectors) == 0) {
866 				error = EINVAL;
867 				goto close_and_exit;
868 			}
869 
870 			/*
871 			 * Compute the size (in DEV_BSIZE blocks) specified
872 			 * by the geometry.
873 			 */
874 			geomsize = (vnd->sc_geom.vng_nsectors *
875 			    vnd->sc_geom.vng_ntracks *
876 			    vnd->sc_geom.vng_ncylinders) *
877 			    (vnd->sc_geom.vng_secsize / DEV_BSIZE);
878 
879 			/*
880 			 * Sanity-check the size against the specified
881 			 * geometry.
882 			 */
883 			if (vnd->sc_size < geomsize) {
884 				error = EINVAL;
885 				goto close_and_exit;
886 			}
887 		} else {
888 			/*
889 			 * Size must be at least 2048 DEV_BSIZE blocks
890 			 * (1M) in order to use this geometry.
891 			 */
892 			if (vnd->sc_size < (32 * 64)) {
893 				error = EINVAL;
894 				goto close_and_exit;
895 			}
896 
897 			vnd->sc_geom.vng_secsize = DEV_BSIZE;
898 			vnd->sc_geom.vng_nsectors = 32;
899 			vnd->sc_geom.vng_ntracks = 64;
900 			vnd->sc_geom.vng_ncylinders = vnd->sc_size / (64 * 32);
901 		}
902 
903 		if (vio->vnd_flags & VNDIOF_READONLY) {
904 			vnd->sc_flags |= VNF_READONLY;
905 		}
906 
907 		if ((error = vndsetcred(vnd, p->p_ucred)) != 0)
908 			goto close_and_exit;
909 		vndthrottle(vnd, vnd->sc_vp);
910 		vio->vnd_size = dbtob(vnd->sc_size);
911 		vnd->sc_flags |= VNF_INITED;
912 #ifdef DEBUG
913 		if (vnddebug & VDB_INIT)
914 			printf("vndioctl: SET vp %p size 0x%lx %d/%d/%d/%d\n",
915 			    vnd->sc_vp, (unsigned long) vnd->sc_size,
916 			    vnd->sc_geom.vng_secsize,
917 			    vnd->sc_geom.vng_nsectors,
918 			    vnd->sc_geom.vng_ntracks,
919 			    vnd->sc_geom.vng_ncylinders);
920 #endif
921 
922 		/* Attach the disk. */
923 		memset(vnd->sc_xname, 0, sizeof(vnd->sc_xname)); /* XXX */
924 		snprintf(vnd->sc_xname, sizeof(vnd->sc_xname), "vnd%d", unit);
925 		vnd->sc_dkdev.dk_name = vnd->sc_xname;
926 		disk_attach(&vnd->sc_dkdev);
927 
928 		/* Initialize the xfer and buffer pools. */
929 		pool_init(&vnd->sc_vxpool, sizeof(struct vndxfer), 0,
930 		    0, 0, "vndxpl", NULL);
931 		pool_init(&vnd->sc_vbpool, sizeof(struct vndbuf), 0,
932 		    0, 0, "vndbpl", NULL);
933 
934 		/* Try and read the disklabel. */
935 		vndgetdisklabel(dev);
936 
937 		vndunlock(vnd);
938 
939 		break;
940 
941 close_and_exit:
942 		(void) vn_close(nd.ni_vp, fflags, p->p_ucred, p);
943 unlock_and_exit:
944 		vndunlock(vnd);
945 		return (error);
946 
947 	case VNDIOCCLR:
948 		if ((error = vndlock(vnd)) != 0)
949 			return (error);
950 
951 		/*
952 		 * Don't unconfigure if any other partitions are open
953 		 * or if both the character and block flavors of this
954 		 * partition are open.
955 		 */
956 		part = DISKPART(dev);
957 		pmask = (1 << part);
958 		if (((vnd->sc_dkdev.dk_openmask & ~pmask) ||
959 		    ((vnd->sc_dkdev.dk_bopenmask & pmask) &&
960 		    (vnd->sc_dkdev.dk_copenmask & pmask))) &&
961 			!(vio->vnd_flags & VNDIOF_FORCE)) {
962 			vndunlock(vnd);
963 			return (EBUSY);
964 		}
965 
966 		/*
967 		 * XXX vndclear() might call vndclose() implicitely;
968 		 * release lock to avoid recursion
969 		 */
970 		vndunlock(vnd);
971 		vndclear(vnd, minor(dev));
972 #ifdef DEBUG
973 		if (vnddebug & VDB_INIT)
974 			printf("vndioctl: CLRed\n");
975 #endif
976 
977 		/* Destroy the xfer and buffer pools. */
978 		pool_destroy(&vnd->sc_vxpool);
979 		pool_destroy(&vnd->sc_vbpool);
980 
981 		/* Detatch the disk. */
982 		disk_detach(&vnd->sc_dkdev);
983 
984 		break;
985 
986 	case VNDIOCGET: {
987 		struct vnd_user *vnu;
988 		struct vattr va;
989 
990 		vnu = (struct vnd_user *)data;
991 
992 		if (vnu->vnu_unit == -1)
993 			vnu->vnu_unit = unit;
994 		if (vnu->vnu_unit >= numvnd)
995 			return (ENXIO);
996 		if (vnu->vnu_unit < 0)
997 			return (EINVAL);
998 
999 		vnd = &vnd_softc[vnu->vnu_unit];
1000 
1001 		if (vnd->sc_flags & VNF_INITED) {
1002 			error = VOP_GETATTR(vnd->sc_vp, &va, p->p_ucred, p);
1003 			if (error)
1004 				return (error);
1005 			vnu->vnu_dev = va.va_fsid;
1006 			vnu->vnu_ino = va.va_fileid;
1007 		}
1008 		else {
1009 			/* unused is not an error */
1010 			vnu->vnu_dev = 0;
1011 			vnu->vnu_ino = 0;
1012 		}
1013 
1014 		break;
1015 	}
1016 
1017 	case DIOCGDINFO:
1018 		*(struct disklabel *)data = *(vnd->sc_dkdev.dk_label);
1019 		break;
1020 
1021 #ifdef __HAVE_OLD_DISKLABEL
1022 	case ODIOCGDINFO:
1023 		newlabel = *(vnd->sc_dkdev.dk_label);
1024 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1025 			return ENOTTY;
1026 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1027 		break;
1028 #endif
1029 
1030 	case DIOCGPART:
1031 		((struct partinfo *)data)->disklab = vnd->sc_dkdev.dk_label;
1032 		((struct partinfo *)data)->part =
1033 		    &vnd->sc_dkdev.dk_label->d_partitions[DISKPART(dev)];
1034 		break;
1035 
1036 	case DIOCWDINFO:
1037 	case DIOCSDINFO:
1038 #ifdef __HAVE_OLD_DISKLABEL
1039 	case ODIOCWDINFO:
1040 	case ODIOCSDINFO:
1041 #endif
1042 	{
1043 		struct disklabel *lp;
1044 
1045 		if ((error = vndlock(vnd)) != 0)
1046 			return (error);
1047 
1048 		vnd->sc_flags |= VNF_LABELLING;
1049 
1050 #ifdef __HAVE_OLD_DISKLABEL
1051 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
1052 			memset(&newlabel, 0, sizeof newlabel);
1053 			memcpy(&newlabel, data, sizeof (struct olddisklabel));
1054 			lp = &newlabel;
1055 		} else
1056 #endif
1057 		lp = (struct disklabel *)data;
1058 
1059 		error = setdisklabel(vnd->sc_dkdev.dk_label,
1060 		    lp, 0, vnd->sc_dkdev.dk_cpulabel);
1061 		if (error == 0) {
1062 			if (cmd == DIOCWDINFO
1063 #ifdef __HAVE_OLD_DISKLABEL
1064 			    || cmd == ODIOCWDINFO
1065 #endif
1066 			   )
1067 				error = writedisklabel(VNDLABELDEV(dev),
1068 				    vndstrategy, vnd->sc_dkdev.dk_label,
1069 				    vnd->sc_dkdev.dk_cpulabel);
1070 		}
1071 
1072 		vnd->sc_flags &= ~VNF_LABELLING;
1073 
1074 		vndunlock(vnd);
1075 
1076 		if (error)
1077 			return (error);
1078 		break;
1079 	}
1080 
1081 	case DIOCKLABEL:
1082 		if (*(int *)data != 0)
1083 			vnd->sc_flags |= VNF_KLABEL;
1084 		else
1085 			vnd->sc_flags &= ~VNF_KLABEL;
1086 		break;
1087 
1088 	case DIOCWLABEL:
1089 		if (*(int *)data != 0)
1090 			vnd->sc_flags |= VNF_WLABEL;
1091 		else
1092 			vnd->sc_flags &= ~VNF_WLABEL;
1093 		break;
1094 
1095 	case DIOCGDEFLABEL:
1096 		vndgetdefaultlabel(vnd, (struct disklabel *)data);
1097 		break;
1098 
1099 #ifdef __HAVE_OLD_DISKLABEL
1100 	case ODIOCGDEFLABEL:
1101 		vndgetdefaultlabel(vnd, &newlabel);
1102 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1103 			return ENOTTY;
1104 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1105 		break;
1106 #endif
1107 
1108 	default:
1109 		return (ENOTTY);
1110 	}
1111 
1112 	return (0);
1113 }
1114 
1115 /*
1116  * Duplicate the current processes' credentials.  Since we are called only
1117  * as the result of a SET ioctl and only root can do that, any future access
1118  * to this "disk" is essentially as root.  Note that credentials may change
1119  * if some other uid can write directly to the mapped file (NFS).
1120  */
1121 static int
1122 vndsetcred(struct vnd_softc *vnd, struct ucred *cred)
1123 {
1124 	struct uio auio;
1125 	struct iovec aiov;
1126 	char *tmpbuf;
1127 	int error;
1128 
1129 	vnd->sc_cred = crdup(cred);
1130 	tmpbuf = malloc(DEV_BSIZE, M_TEMP, M_WAITOK);
1131 
1132 	/* XXX: Horrible kludge to establish credentials for NFS */
1133 	aiov.iov_base = tmpbuf;
1134 	aiov.iov_len = min(DEV_BSIZE, dbtob(vnd->sc_size));
1135 	auio.uio_iov = &aiov;
1136 	auio.uio_iovcnt = 1;
1137 	auio.uio_offset = 0;
1138 	auio.uio_rw = UIO_READ;
1139 	auio.uio_segflg = UIO_SYSSPACE;
1140 	auio.uio_resid = aiov.iov_len;
1141 	vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1142 	error = VOP_READ(vnd->sc_vp, &auio, 0, vnd->sc_cred);
1143 	if (error == 0) {
1144 		/*
1145 		 * Because vnd does all IO directly through the vnode
1146 		 * we need to flush (at least) the buffer from the above
1147 		 * VOP_READ from the buffer cache to prevent cache
1148 		 * incoherencies.  Also, be careful to write dirty
1149 		 * buffers back to stable storage.
1150 		 */
1151 		error = vinvalbuf(vnd->sc_vp, V_SAVE, vnd->sc_cred,
1152 			    curproc, 0, 0);
1153 	}
1154 	VOP_UNLOCK(vnd->sc_vp, 0);
1155 
1156 	free(tmpbuf, M_TEMP);
1157 	return (error);
1158 }
1159 
1160 /*
1161  * Set maxactive based on FS type
1162  */
1163 static void
1164 vndthrottle(struct vnd_softc *vnd, struct vnode *vp)
1165 {
1166 #ifdef NFS
1167 	extern int (**nfsv2_vnodeop_p)(void *);
1168 
1169 	if (vp->v_op == nfsv2_vnodeop_p)
1170 		vnd->sc_maxactive = 2;
1171 	else
1172 #endif
1173 		vnd->sc_maxactive = 8;
1174 
1175 	if (vnd->sc_maxactive < 1)
1176 		vnd->sc_maxactive = 1;
1177 }
1178 
1179 #if 0
1180 static void
1181 vndshutdown(void)
1182 {
1183 	struct vnd_softc *vnd;
1184 
1185 	for (vnd = &vnd_softc[0]; vnd < &vnd_softc[numvnd]; vnd++)
1186 		if (vnd->sc_flags & VNF_INITED)
1187 			vndclear(vnd);
1188 }
1189 #endif
1190 
1191 static void
1192 vndclear(struct vnd_softc *vnd, int myminor)
1193 {
1194 	struct vnode *vp = vnd->sc_vp;
1195 	struct proc *p = curproc;		/* XXX */
1196 	int fflags = FREAD;
1197 	int bmaj, cmaj, i, mn;
1198 
1199 #ifdef DEBUG
1200 	if (vnddebug & VDB_FOLLOW)
1201 		printf("vndclear(%p): vp %p\n", vnd, vp);
1202 #endif
1203 	/* locate the major number */
1204 	bmaj = bdevsw_lookup_major(&vnd_bdevsw);
1205 	cmaj = cdevsw_lookup_major(&vnd_cdevsw);
1206 
1207 	/* Nuke the vnodes for any open instances */
1208 	for (i = 0; i < MAXPARTITIONS; i++) {
1209 		mn = DISKMINOR(vnd->sc_unit, i);
1210 		vdevgone(bmaj, mn, mn, VBLK);
1211 		if (mn != myminor) /* XXX avoid to kill own vnode */
1212 			vdevgone(cmaj, mn, mn, VCHR);
1213 	}
1214 
1215 	if ((vnd->sc_flags & VNF_READONLY) == 0)
1216 		fflags |= FWRITE;
1217 	vnd->sc_flags &= ~(VNF_INITED | VNF_READONLY | VNF_VLABEL);
1218 	if (vp == (struct vnode *)0)
1219 		panic("vndioctl: null vp");
1220 	(void) vn_close(vp, fflags, vnd->sc_cred, p);
1221 	crfree(vnd->sc_cred);
1222 	vnd->sc_vp = (struct vnode *)0;
1223 	vnd->sc_cred = (struct ucred *)0;
1224 	vnd->sc_size = 0;
1225 }
1226 
1227 static int
1228 vndsize(dev_t dev)
1229 {
1230 	struct vnd_softc *sc;
1231 	struct disklabel *lp;
1232 	int part, unit, omask;
1233 	int size;
1234 
1235 	unit = vndunit(dev);
1236 	if (unit >= numvnd)
1237 		return (-1);
1238 	sc = &vnd_softc[unit];
1239 
1240 	if ((sc->sc_flags & VNF_INITED) == 0)
1241 		return (-1);
1242 
1243 	part = DISKPART(dev);
1244 	omask = sc->sc_dkdev.dk_openmask & (1 << part);
1245 	lp = sc->sc_dkdev.dk_label;
1246 
1247 	if (omask == 0 && vndopen(dev, 0, S_IFBLK, curproc))
1248 		return (-1);
1249 
1250 	if (lp->d_partitions[part].p_fstype != FS_SWAP)
1251 		size = -1;
1252 	else
1253 		size = lp->d_partitions[part].p_size *
1254 		    (lp->d_secsize / DEV_BSIZE);
1255 
1256 	if (omask == 0 && vndclose(dev, 0, S_IFBLK, curproc))
1257 		return (-1);
1258 
1259 	return (size);
1260 }
1261 
1262 static int
1263 vnddump(dev_t dev, daddr_t blkno, caddr_t va, size_t size)
1264 {
1265 
1266 	/* Not implemented. */
1267 	return ENXIO;
1268 }
1269 
1270 static void
1271 vndgetdefaultlabel(struct vnd_softc *sc, struct disklabel *lp)
1272 {
1273 	struct vndgeom *vng = &sc->sc_geom;
1274 	struct partition *pp;
1275 
1276 	memset(lp, 0, sizeof(*lp));
1277 
1278 	lp->d_secperunit = sc->sc_size / (vng->vng_secsize / DEV_BSIZE);
1279 	lp->d_secsize = vng->vng_secsize;
1280 	lp->d_nsectors = vng->vng_nsectors;
1281 	lp->d_ntracks = vng->vng_ntracks;
1282 	lp->d_ncylinders = vng->vng_ncylinders;
1283 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1284 
1285 	strncpy(lp->d_typename, "vnd", sizeof(lp->d_typename));
1286 	lp->d_type = DTYPE_VND;
1287 	strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
1288 	lp->d_rpm = 3600;
1289 	lp->d_interleave = 1;
1290 	lp->d_flags = 0;
1291 
1292 	pp = &lp->d_partitions[RAW_PART];
1293 	pp->p_offset = 0;
1294 	pp->p_size = lp->d_secperunit;
1295 	pp->p_fstype = FS_UNUSED;
1296 	lp->d_npartitions = RAW_PART + 1;
1297 
1298 	lp->d_magic = DISKMAGIC;
1299 	lp->d_magic2 = DISKMAGIC;
1300 	lp->d_checksum = dkcksum(lp);
1301 }
1302 
1303 /*
1304  * Read the disklabel from a vnd.  If one is not present, create a fake one.
1305  */
1306 static void
1307 vndgetdisklabel(dev_t dev)
1308 {
1309 	struct vnd_softc *sc = &vnd_softc[vndunit(dev)];
1310 	const char *errstring;
1311 	struct disklabel *lp = sc->sc_dkdev.dk_label;
1312 	struct cpu_disklabel *clp = sc->sc_dkdev.dk_cpulabel;
1313 	int i;
1314 
1315 	memset(clp, 0, sizeof(*clp));
1316 
1317 	vndgetdefaultlabel(sc, lp);
1318 
1319 	/*
1320 	 * Call the generic disklabel extraction routine.
1321 	 */
1322 	errstring = readdisklabel(VNDLABELDEV(dev), vndstrategy, lp, clp);
1323 	if (errstring) {
1324 		/*
1325 		 * Lack of disklabel is common, but we print the warning
1326 		 * anyway, since it might contain other useful information.
1327 		 */
1328 		printf("%s: %s\n", sc->sc_xname, errstring);
1329 
1330 		/*
1331 		 * For historical reasons, if there's no disklabel
1332 		 * present, all partitions must be FS_BSDFFS and
1333 		 * occupy the entire disk.
1334 		 */
1335 		for (i = 0; i < MAXPARTITIONS; i++) {
1336 			/*
1337 			 * Don't wipe out port specific hack (such as
1338 			 * dos partition hack of i386 port).
1339 			 */
1340 			if (lp->d_partitions[i].p_size != 0)
1341 				continue;
1342 
1343 			lp->d_partitions[i].p_size = lp->d_secperunit;
1344 			lp->d_partitions[i].p_offset = 0;
1345 			lp->d_partitions[i].p_fstype = FS_BSDFFS;
1346 		}
1347 
1348 		strncpy(lp->d_packname, "default label",
1349 		    sizeof(lp->d_packname));
1350 
1351 		lp->d_npartitions = MAXPARTITIONS;
1352 		lp->d_checksum = dkcksum(lp);
1353 	}
1354 
1355 	/* In-core label now valid. */
1356 	sc->sc_flags |= VNF_VLABEL;
1357 }
1358 
1359 /*
1360  * Wait interruptibly for an exclusive lock.
1361  *
1362  * XXX
1363  * Several drivers do this; it should be abstracted and made MP-safe.
1364  */
1365 static int
1366 vndlock(struct vnd_softc *sc)
1367 {
1368 	int error;
1369 
1370 	while ((sc->sc_flags & VNF_LOCKED) != 0) {
1371 		sc->sc_flags |= VNF_WANTED;
1372 		if ((error = tsleep(sc, PRIBIO | PCATCH, "vndlck", 0)) != 0)
1373 			return (error);
1374 	}
1375 	sc->sc_flags |= VNF_LOCKED;
1376 	return (0);
1377 }
1378 
1379 /*
1380  * Unlock and wake up any waiters.
1381  */
1382 static void
1383 vndunlock(struct vnd_softc *sc)
1384 {
1385 
1386 	sc->sc_flags &= ~VNF_LOCKED;
1387 	if ((sc->sc_flags & VNF_WANTED) != 0) {
1388 		sc->sc_flags &= ~VNF_WANTED;
1389 		wakeup(sc);
1390 	}
1391 }
1392