xref: /netbsd-src/sys/dev/vnd.c (revision 6d322f2f4598f0d8a138f10ea648ec4fabe41f8b)
1 /*	$NetBSD: vnd.c,v 1.227 2013/12/29 08:09:44 pgoyette Exp $	*/
2 
3 /*-
4  * Copyright (c) 1996, 1997, 1998, 2008 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  *
19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29  * POSSIBILITY OF SUCH DAMAGE.
30  */
31 
32 /*
33  * Copyright (c) 1988 University of Utah.
34  * Copyright (c) 1990, 1993
35  *	The Regents of the University of California.  All rights reserved.
36  *
37  * This code is derived from software contributed to Berkeley by
38  * the Systems Programming Group of the University of Utah Computer
39  * Science Department.
40  *
41  * Redistribution and use in source and binary forms, with or without
42  * modification, are permitted provided that the following conditions
43  * are met:
44  * 1. Redistributions of source code must retain the above copyright
45  *    notice, this list of conditions and the following disclaimer.
46  * 2. Redistributions in binary form must reproduce the above copyright
47  *    notice, this list of conditions and the following disclaimer in the
48  *    documentation and/or other materials provided with the distribution.
49  * 3. Neither the name of the University nor the names of its contributors
50  *    may be used to endorse or promote products derived from this software
51  *    without specific prior written permission.
52  *
53  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
54  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
55  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
56  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
57  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
58  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
59  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
60  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
61  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
62  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
63  * SUCH DAMAGE.
64  *
65  * from: Utah $Hdr: vn.c 1.13 94/04/02$
66  *
67  *	@(#)vn.c	8.9 (Berkeley) 5/14/95
68  */
69 
70 /*
71  * Vnode disk driver.
72  *
73  * Block/character interface to a vnode.  Allows one to treat a file
74  * as a disk (e.g. build a filesystem in it, mount it, etc.).
75  *
76  * NOTE 1: If the vnode supports the VOP_BMAP and VOP_STRATEGY operations,
77  * this uses them to avoid distorting the local buffer cache.  If those
78  * block-level operations are not available, this falls back to the regular
79  * read and write calls.  Using these may distort the cache in some cases
80  * but better have the driver working than preventing it to work on file
81  * systems where the block-level operations are not implemented for
82  * whatever reason.
83  *
84  * NOTE 2: There is a security issue involved with this driver.
85  * Once mounted all access to the contents of the "mapped" file via
86  * the special file is controlled by the permissions on the special
87  * file, the protection of the mapped file is ignored (effectively,
88  * by using root credentials in all transactions).
89  *
90  * NOTE 3: Doesn't interact with leases, should it?
91  */
92 
93 #include <sys/cdefs.h>
94 __KERNEL_RCSID(0, "$NetBSD: vnd.c,v 1.227 2013/12/29 08:09:44 pgoyette Exp $");
95 
96 #if defined(_KERNEL_OPT)
97 #include "opt_vnd.h"
98 #include "opt_compat_netbsd.h"
99 #endif
100 
101 #include <sys/param.h>
102 #include <sys/systm.h>
103 #include <sys/namei.h>
104 #include <sys/proc.h>
105 #include <sys/kthread.h>
106 #include <sys/errno.h>
107 #include <sys/buf.h>
108 #include <sys/bufq.h>
109 #include <sys/malloc.h>
110 #include <sys/ioctl.h>
111 #include <sys/disklabel.h>
112 #include <sys/device.h>
113 #include <sys/disk.h>
114 #include <sys/stat.h>
115 #include <sys/mount.h>
116 #include <sys/vnode.h>
117 #include <sys/file.h>
118 #include <sys/uio.h>
119 #include <sys/conf.h>
120 #include <sys/kauth.h>
121 
122 #include <net/zlib.h>
123 
124 #include <miscfs/genfs/genfs.h>
125 #include <miscfs/specfs/specdev.h>
126 
127 #include <dev/dkvar.h>
128 #include <dev/vndvar.h>
129 
130 #if defined(VNDDEBUG) && !defined(DEBUG)
131 #define DEBUG
132 #endif
133 
134 #ifdef DEBUG
135 int dovndcluster = 1;
136 #define VDB_FOLLOW	0x01
137 #define VDB_INIT	0x02
138 #define VDB_IO		0x04
139 #define VDB_LABEL	0x08
140 int vnddebug = 0x00;
141 #endif
142 
143 #define vndunit(x)	DISKUNIT(x)
144 
145 struct vndxfer {
146 	struct buf vx_buf;
147 	struct vnd_softc *vx_vnd;
148 };
149 #define	VND_BUFTOXFER(bp)	((struct vndxfer *)(void *)bp)
150 
151 #define VND_GETXFER(vnd)	pool_get(&(vnd)->sc_vxpool, PR_WAITOK)
152 #define VND_PUTXFER(vnd, vx)	pool_put(&(vnd)->sc_vxpool, (vx))
153 
154 #define VNDLABELDEV(dev) \
155     (MAKEDISKDEV(major((dev)), vndunit((dev)), RAW_PART))
156 
157 #define	VND_MAXPENDING(vnd)	((vnd)->sc_maxactive * 4)
158 
159 /* called by main() at boot time */
160 void	vndattach(int);
161 
162 static void	vndclear(struct vnd_softc *, int);
163 static int	vnddoclear(struct vnd_softc *, int, int, bool);
164 static int	vndsetcred(struct vnd_softc *, kauth_cred_t);
165 static void	vndthrottle(struct vnd_softc *, struct vnode *);
166 static void	vndiodone(struct buf *);
167 #if 0
168 static void	vndshutdown(void);
169 #endif
170 
171 static void	vndgetdefaultlabel(struct vnd_softc *, struct disklabel *);
172 static void	vndgetdisklabel(dev_t, struct vnd_softc *);
173 
174 static int	vndlock(struct vnd_softc *);
175 static void	vndunlock(struct vnd_softc *);
176 #ifdef VND_COMPRESSION
177 static void	compstrategy(struct buf *, off_t);
178 static void	*vnd_alloc(void *, u_int, u_int);
179 static void	vnd_free(void *, void *);
180 #endif /* VND_COMPRESSION */
181 
182 static void	vndthread(void *);
183 static bool	vnode_has_op(const struct vnode *, int);
184 static void	handle_with_rdwr(struct vnd_softc *, const struct buf *,
185 		    struct buf *);
186 static void	handle_with_strategy(struct vnd_softc *, const struct buf *,
187 		    struct buf *);
188 static void	vnd_set_geometry(struct vnd_softc *);
189 
190 static dev_type_open(vndopen);
191 static dev_type_close(vndclose);
192 static dev_type_read(vndread);
193 static dev_type_write(vndwrite);
194 static dev_type_ioctl(vndioctl);
195 static dev_type_strategy(vndstrategy);
196 static dev_type_dump(vnddump);
197 static dev_type_size(vndsize);
198 
199 const struct bdevsw vnd_bdevsw = {
200 	vndopen, vndclose, vndstrategy, vndioctl, vnddump, vndsize, D_DISK
201 };
202 
203 const struct cdevsw vnd_cdevsw = {
204 	vndopen, vndclose, vndread, vndwrite, vndioctl,
205 	nostop, notty, nopoll, nommap, nokqfilter, D_DISK
206 };
207 
208 static int	vnd_match(device_t, cfdata_t, void *);
209 static void	vnd_attach(device_t, device_t, void *);
210 static int	vnd_detach(device_t, int);
211 
212 CFATTACH_DECL3_NEW(vnd, sizeof(struct vnd_softc),
213     vnd_match, vnd_attach, vnd_detach, NULL, NULL, NULL, DVF_DETACH_SHUTDOWN);
214 extern struct cfdriver vnd_cd;
215 
216 static struct vnd_softc	*vnd_spawn(int);
217 int	vnd_destroy(device_t);
218 
219 static struct	dkdriver vnddkdriver = { vndstrategy, minphys };
220 
221 void
222 vndattach(int num)
223 {
224 	int error;
225 
226 	error = config_cfattach_attach(vnd_cd.cd_name, &vnd_ca);
227 	if (error)
228 		aprint_error("%s: unable to register cfattach\n",
229 		    vnd_cd.cd_name);
230 }
231 
232 static int
233 vnd_match(device_t self, cfdata_t cfdata, void *aux)
234 {
235 
236 	return 1;
237 }
238 
239 static void
240 vnd_attach(device_t parent, device_t self, void *aux)
241 {
242 	struct vnd_softc *sc = device_private(self);
243 
244 	sc->sc_dev = self;
245 	sc->sc_comp_offsets = NULL;
246 	sc->sc_comp_buff = NULL;
247 	sc->sc_comp_decombuf = NULL;
248 	bufq_alloc(&sc->sc_tab, "disksort", BUFQ_SORT_RAWBLOCK);
249 	disk_init(&sc->sc_dkdev, device_xname(self), &vnddkdriver);
250 	if (!pmf_device_register(self, NULL, NULL))
251 		aprint_error_dev(self, "couldn't establish power handler\n");
252 }
253 
254 static int
255 vnd_detach(device_t self, int flags)
256 {
257 	int error;
258 	struct vnd_softc *sc = device_private(self);
259 
260 	if (sc->sc_flags & VNF_INITED) {
261 		error = vnddoclear(sc, 0, -1, (flags & DETACH_FORCE) != 0);
262 		if (error != 0)
263 			return error;
264 	}
265 
266 	pmf_device_deregister(self);
267 	bufq_free(sc->sc_tab);
268 	disk_destroy(&sc->sc_dkdev);
269 
270 	return 0;
271 }
272 
273 static struct vnd_softc *
274 vnd_spawn(int unit)
275 {
276 	cfdata_t cf;
277 
278 	cf = malloc(sizeof(*cf), M_DEVBUF, M_WAITOK);
279 	cf->cf_name = vnd_cd.cd_name;
280 	cf->cf_atname = vnd_cd.cd_name;
281 	cf->cf_unit = unit;
282 	cf->cf_fstate = FSTATE_STAR;
283 
284 	return device_private(config_attach_pseudo(cf));
285 }
286 
287 int
288 vnd_destroy(device_t dev)
289 {
290 	int error;
291 	cfdata_t cf;
292 
293 	cf = device_cfdata(dev);
294 	error = config_detach(dev, DETACH_QUIET);
295 	if (error)
296 		return error;
297 	free(cf, M_DEVBUF);
298 	return 0;
299 }
300 
301 static int
302 vndopen(dev_t dev, int flags, int mode, struct lwp *l)
303 {
304 	int unit = vndunit(dev);
305 	struct vnd_softc *sc;
306 	int error = 0, part, pmask;
307 	struct disklabel *lp;
308 
309 #ifdef DEBUG
310 	if (vnddebug & VDB_FOLLOW)
311 		printf("vndopen(0x%"PRIx64", 0x%x, 0x%x, %p)\n", dev, flags, mode, l);
312 #endif
313 	sc = device_lookup_private(&vnd_cd, unit);
314 	if (sc == NULL) {
315 		sc = vnd_spawn(unit);
316 		if (sc == NULL)
317 			return ENOMEM;
318 	}
319 
320 	if ((error = vndlock(sc)) != 0)
321 		return error;
322 
323 	if ((sc->sc_flags & VNF_CLEARING) != 0) {
324 		error = ENXIO;
325 		goto done;
326 	}
327 
328 	lp = sc->sc_dkdev.dk_label;
329 
330 	part = DISKPART(dev);
331 	pmask = (1 << part);
332 
333 	if (sc->sc_dkdev.dk_nwedges != 0 && part != RAW_PART) {
334 		error = EBUSY;
335 		goto done;
336 	}
337 
338 	if (sc->sc_flags & VNF_INITED) {
339 		if ((sc->sc_dkdev.dk_openmask & ~(1<<RAW_PART)) != 0) {
340 			/*
341 			 * If any non-raw partition is open, but the disk
342 			 * has been invalidated, disallow further opens.
343 			 */
344 			if ((sc->sc_flags & VNF_VLABEL) == 0) {
345 				error = EIO;
346 				goto done;
347 			}
348 		} else {
349 			/*
350 			 * Load the partition info if not already loaded.
351 			 */
352 			if ((sc->sc_flags & VNF_VLABEL) == 0) {
353 				sc->sc_flags |= VNF_VLABEL;
354 				vndgetdisklabel(dev, sc);
355 			}
356 		}
357 	}
358 
359 	/* Check that the partitions exists. */
360 	if (part != RAW_PART) {
361 		if (((sc->sc_flags & VNF_INITED) == 0) ||
362 		    ((part >= lp->d_npartitions) ||
363 		     (lp->d_partitions[part].p_fstype == FS_UNUSED))) {
364 			error = ENXIO;
365 			goto done;
366 		}
367 	}
368 
369 	/* Prevent our unit from being unconfigured while open. */
370 	switch (mode) {
371 	case S_IFCHR:
372 		sc->sc_dkdev.dk_copenmask |= pmask;
373 		break;
374 
375 	case S_IFBLK:
376 		sc->sc_dkdev.dk_bopenmask |= pmask;
377 		break;
378 	}
379 	sc->sc_dkdev.dk_openmask =
380 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
381 
382  done:
383 	vndunlock(sc);
384 	return error;
385 }
386 
387 static int
388 vndclose(dev_t dev, int flags, int mode, struct lwp *l)
389 {
390 	int unit = vndunit(dev);
391 	struct vnd_softc *sc;
392 	int error = 0, part;
393 
394 #ifdef DEBUG
395 	if (vnddebug & VDB_FOLLOW)
396 		printf("vndclose(0x%"PRIx64", 0x%x, 0x%x, %p)\n", dev, flags, mode, l);
397 #endif
398 	sc = device_lookup_private(&vnd_cd, unit);
399 	if (sc == NULL)
400 		return ENXIO;
401 
402 	if ((error = vndlock(sc)) != 0)
403 		return error;
404 
405 	part = DISKPART(dev);
406 
407 	/* ...that much closer to allowing unconfiguration... */
408 	switch (mode) {
409 	case S_IFCHR:
410 		sc->sc_dkdev.dk_copenmask &= ~(1 << part);
411 		break;
412 
413 	case S_IFBLK:
414 		sc->sc_dkdev.dk_bopenmask &= ~(1 << part);
415 		break;
416 	}
417 	sc->sc_dkdev.dk_openmask =
418 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
419 
420 	vndunlock(sc);
421 
422 	if ((sc->sc_flags & VNF_INITED) == 0) {
423 		if ((error = vnd_destroy(sc->sc_dev)) != 0) {
424 			aprint_error_dev(sc->sc_dev,
425 			    "unable to detach instance\n");
426 			return error;
427 		}
428 	}
429 
430 	return 0;
431 }
432 
433 /*
434  * Queue the request, and wakeup the kernel thread to handle it.
435  */
436 static void
437 vndstrategy(struct buf *bp)
438 {
439 	int unit = vndunit(bp->b_dev);
440 	struct vnd_softc *vnd =
441 	    device_lookup_private(&vnd_cd, unit);
442 	struct disklabel *lp;
443 	daddr_t blkno;
444 	int s = splbio();
445 
446 	if (vnd == NULL) {
447 		bp->b_error = ENXIO;
448 		goto done;
449 	}
450 	lp = vnd->sc_dkdev.dk_label;
451 
452 	if ((vnd->sc_flags & VNF_INITED) == 0) {
453 		bp->b_error = ENXIO;
454 		goto done;
455 	}
456 
457 	/*
458 	 * The transfer must be a whole number of blocks.
459 	 */
460 	if ((bp->b_bcount % lp->d_secsize) != 0) {
461 		bp->b_error = EINVAL;
462 		goto done;
463 	}
464 
465 	/*
466 	 * check if we're read-only.
467 	 */
468 	if ((vnd->sc_flags & VNF_READONLY) && !(bp->b_flags & B_READ)) {
469 		bp->b_error = EACCES;
470 		goto done;
471 	}
472 
473 	/* If it's a nil transfer, wake up the top half now. */
474 	if (bp->b_bcount == 0) {
475 		goto done;
476 	}
477 
478 	/*
479 	 * Do bounds checking and adjust transfer.  If there's an error,
480 	 * the bounds check will flag that for us.
481 	 */
482 	if (DISKPART(bp->b_dev) == RAW_PART) {
483 		if (bounds_check_with_mediasize(bp, DEV_BSIZE,
484 		    vnd->sc_size) <= 0)
485 			goto done;
486 	} else {
487 		if (bounds_check_with_label(&vnd->sc_dkdev,
488 		    bp, vnd->sc_flags & (VNF_WLABEL|VNF_LABELLING)) <= 0)
489 			goto done;
490 	}
491 
492 	/*
493 	 * Put the block number in terms of the logical blocksize
494 	 * of the "device".
495 	 */
496 
497 	blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
498 
499 	/*
500 	 * Translate the partition-relative block number to an absolute.
501 	 */
502 	if (DISKPART(bp->b_dev) != RAW_PART) {
503 		struct partition *pp;
504 
505 		pp = &vnd->sc_dkdev.dk_label->d_partitions[
506 		    DISKPART(bp->b_dev)];
507 		blkno += pp->p_offset;
508 	}
509 	bp->b_rawblkno = blkno;
510 
511 #ifdef DEBUG
512 	if (vnddebug & VDB_FOLLOW)
513 		printf("vndstrategy(%p): unit %d\n", bp, unit);
514 #endif
515 	if ((vnd->sc_flags & VNF_USE_VN_RDWR)) {
516 		KASSERT(vnd->sc_pending >= 0 &&
517 		    vnd->sc_pending <= VND_MAXPENDING(vnd));
518 		while (vnd->sc_pending == VND_MAXPENDING(vnd))
519 			tsleep(&vnd->sc_pending, PRIBIO, "vndpc", 0);
520 		vnd->sc_pending++;
521 	}
522 	bufq_put(vnd->sc_tab, bp);
523 	wakeup(&vnd->sc_tab);
524 	splx(s);
525 	return;
526 
527 done:
528 	bp->b_resid = bp->b_bcount;
529 	biodone(bp);
530 	splx(s);
531 }
532 
533 static bool
534 vnode_has_strategy(struct vnd_softc *vnd)
535 {
536 	return vnode_has_op(vnd->sc_vp, VOFFSET(vop_bmap)) &&
537 	    vnode_has_op(vnd->sc_vp, VOFFSET(vop_strategy));
538 }
539 
540 /* XXX this function needs a reliable check to detect
541  * sparse files. Otherwise, bmap/strategy may be used
542  * and fail on non-allocated blocks. VOP_READ/VOP_WRITE
543  * works on sparse files.
544  */
545 #if notyet
546 static bool
547 vnode_strategy_probe(struct vnd_softc *vnd)
548 {
549 	int error;
550 	daddr_t nbn;
551 
552 	if (!vnode_has_strategy(vnd))
553 		return false;
554 
555 	/* Convert the first logical block number to its
556 	 * physical block number.
557 	 */
558 	error = 0;
559 	vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
560 	error = VOP_BMAP(vnd->sc_vp, 0, NULL, &nbn, NULL);
561 	VOP_UNLOCK(vnd->sc_vp);
562 
563 	/* Test if that worked. */
564 	if (error == 0 && (long)nbn == -1)
565 		return false;
566 
567 	return true;
568 }
569 #endif
570 
571 static void
572 vndthread(void *arg)
573 {
574 	struct vnd_softc *vnd = arg;
575 	int s;
576 
577 	/* Determine whether we can *use* VOP_BMAP and VOP_STRATEGY to
578 	 * directly access the backing vnode.  If we can, use these two
579 	 * operations to avoid messing with the local buffer cache.
580 	 * Otherwise fall back to regular VOP_READ/VOP_WRITE operations
581 	 * which are guaranteed to work with any file system. */
582 	if ((vnd->sc_flags & VNF_USE_VN_RDWR) == 0 &&
583 	    ! vnode_has_strategy(vnd))
584 		vnd->sc_flags |= VNF_USE_VN_RDWR;
585 
586 #ifdef DEBUG
587 	if (vnddebug & VDB_INIT)
588 		printf("vndthread: vp %p, %s\n", vnd->sc_vp,
589 		    (vnd->sc_flags & VNF_USE_VN_RDWR) == 0 ?
590 		    "using bmap/strategy operations" :
591 		    "using read/write operations");
592 #endif
593 
594 	s = splbio();
595 	vnd->sc_flags |= VNF_KTHREAD;
596 	wakeup(&vnd->sc_kthread);
597 
598 	/*
599 	 * Dequeue requests and serve them depending on the available
600 	 * vnode operations.
601 	 */
602 	while ((vnd->sc_flags & VNF_VUNCONF) == 0) {
603 		struct vndxfer *vnx;
604 		struct buf *obp;
605 		struct buf *bp;
606 
607 		obp = bufq_get(vnd->sc_tab);
608 		if (obp == NULL) {
609 			tsleep(&vnd->sc_tab, PRIBIO, "vndbp", 0);
610 			continue;
611 		};
612 		if ((vnd->sc_flags & VNF_USE_VN_RDWR)) {
613 			KASSERT(vnd->sc_pending > 0 &&
614 			    vnd->sc_pending <= VND_MAXPENDING(vnd));
615 			if (vnd->sc_pending-- == VND_MAXPENDING(vnd))
616 				wakeup(&vnd->sc_pending);
617 		}
618 		splx(s);
619 #ifdef DEBUG
620 		if (vnddebug & VDB_FOLLOW)
621 			printf("vndthread(%p)\n", obp);
622 #endif
623 
624 		if (vnd->sc_vp->v_mount == NULL) {
625 			obp->b_error = ENXIO;
626 			goto done;
627 		}
628 #ifdef VND_COMPRESSION
629 		/* handle a compressed read */
630 		if ((obp->b_flags & B_READ) != 0 && (vnd->sc_flags & VNF_COMP)) {
631 			off_t bn;
632 
633 			/* Convert to a byte offset within the file. */
634 			bn = obp->b_rawblkno *
635 			    vnd->sc_dkdev.dk_label->d_secsize;
636 
637 			compstrategy(obp, bn);
638 			goto done;
639 		}
640 #endif /* VND_COMPRESSION */
641 
642 		/*
643 		 * Allocate a header for this transfer and link it to the
644 		 * buffer
645 		 */
646 		s = splbio();
647 		vnx = VND_GETXFER(vnd);
648 		splx(s);
649 		vnx->vx_vnd = vnd;
650 
651 		s = splbio();
652 		while (vnd->sc_active >= vnd->sc_maxactive) {
653 			tsleep(&vnd->sc_tab, PRIBIO, "vndac", 0);
654 		}
655 		vnd->sc_active++;
656 		splx(s);
657 
658 		/* Instrumentation. */
659 		disk_busy(&vnd->sc_dkdev);
660 
661 		bp = &vnx->vx_buf;
662 		buf_init(bp);
663 		bp->b_flags = (obp->b_flags & B_READ);
664 		bp->b_oflags = obp->b_oflags;
665 		bp->b_cflags = obp->b_cflags;
666 		bp->b_iodone = vndiodone;
667 		bp->b_private = obp;
668 		bp->b_vp = vnd->sc_vp;
669 		bp->b_objlock = bp->b_vp->v_interlock;
670 		bp->b_data = obp->b_data;
671 		bp->b_bcount = obp->b_bcount;
672 		BIO_COPYPRIO(bp, obp);
673 
674 		/* Handle the request using the appropriate operations. */
675 		if ((vnd->sc_flags & VNF_USE_VN_RDWR) == 0)
676 			handle_with_strategy(vnd, obp, bp);
677 		else
678 			handle_with_rdwr(vnd, obp, bp);
679 
680 		s = splbio();
681 		continue;
682 
683 done:
684 		biodone(obp);
685 		s = splbio();
686 	}
687 
688 	vnd->sc_flags &= (~VNF_KTHREAD | VNF_VUNCONF);
689 	wakeup(&vnd->sc_kthread);
690 	splx(s);
691 	kthread_exit(0);
692 }
693 
694 /*
695  * Checks if the given vnode supports the requested operation.
696  * The operation is specified the offset returned by VOFFSET.
697  *
698  * XXX The test below used to determine this is quite fragile
699  * because it relies on the file system to use genfs to specify
700  * unimplemented operations.  There might be another way to do
701  * it more cleanly.
702  */
703 static bool
704 vnode_has_op(const struct vnode *vp, int opoffset)
705 {
706 	int (*defaultp)(void *);
707 	int (*opp)(void *);
708 
709 	defaultp = vp->v_op[VOFFSET(vop_default)];
710 	opp = vp->v_op[opoffset];
711 
712 	return opp != defaultp && opp != genfs_eopnotsupp &&
713 	    opp != genfs_badop && opp != genfs_nullop;
714 }
715 
716 /*
717  * Handes the read/write request given in 'bp' using the vnode's VOP_READ
718  * and VOP_WRITE operations.
719  *
720  * 'obp' is a pointer to the original request fed to the vnd device.
721  */
722 static void
723 handle_with_rdwr(struct vnd_softc *vnd, const struct buf *obp, struct buf *bp)
724 {
725 	bool doread;
726 	off_t offset;
727 	size_t len, resid;
728 	struct vnode *vp;
729 
730 	doread = bp->b_flags & B_READ;
731 	offset = obp->b_rawblkno * vnd->sc_dkdev.dk_label->d_secsize;
732 	len = bp->b_bcount;
733 	vp = vnd->sc_vp;
734 
735 #if defined(DEBUG)
736 	if (vnddebug & VDB_IO)
737 		printf("vnd (rdwr): vp %p, %s, rawblkno 0x%" PRIx64
738 		    ", secsize %d, offset %" PRIu64
739 		    ", bcount %d\n",
740 		    vp, doread ? "read" : "write", obp->b_rawblkno,
741 		    vnd->sc_dkdev.dk_label->d_secsize, offset,
742 		    bp->b_bcount);
743 #endif
744 
745 	/* Issue the read or write operation. */
746 	bp->b_error =
747 	    vn_rdwr(doread ? UIO_READ : UIO_WRITE,
748 	    vp, bp->b_data, len, offset, UIO_SYSSPACE,
749 	    IO_ADV_ENCODE(POSIX_FADV_NOREUSE), vnd->sc_cred, &resid, NULL);
750 	bp->b_resid = resid;
751 
752 	mutex_enter(vp->v_interlock);
753 	(void) VOP_PUTPAGES(vp, 0, 0,
754 	    PGO_ALLPAGES | PGO_CLEANIT | PGO_FREE | PGO_SYNCIO);
755 
756 	/* We need to increase the number of outputs on the vnode if
757 	 * there was any write to it. */
758 	if (!doread) {
759 		mutex_enter(vp->v_interlock);
760 		vp->v_numoutput++;
761 		mutex_exit(vp->v_interlock);
762 	}
763 
764 	biodone(bp);
765 }
766 
767 /*
768  * Handes the read/write request given in 'bp' using the vnode's VOP_BMAP
769  * and VOP_STRATEGY operations.
770  *
771  * 'obp' is a pointer to the original request fed to the vnd device.
772  */
773 static void
774 handle_with_strategy(struct vnd_softc *vnd, const struct buf *obp,
775     struct buf *bp)
776 {
777 	int bsize, error, flags, skipped;
778 	size_t resid, sz;
779 	off_t bn, offset;
780 	struct vnode *vp;
781 
782 	flags = obp->b_flags;
783 
784 	if (!(flags & B_READ)) {
785 		vp = bp->b_vp;
786 		mutex_enter(vp->v_interlock);
787 		vp->v_numoutput++;
788 		mutex_exit(vp->v_interlock);
789 	}
790 
791 	/* convert to a byte offset within the file. */
792 	bn = obp->b_rawblkno * vnd->sc_dkdev.dk_label->d_secsize;
793 
794 	bsize = vnd->sc_vp->v_mount->mnt_stat.f_iosize;
795 	skipped = 0;
796 
797 	/*
798 	 * Break the request into bsize pieces and feed them
799 	 * sequentially using VOP_BMAP/VOP_STRATEGY.
800 	 * We do it this way to keep from flooding NFS servers if we
801 	 * are connected to an NFS file.  This places the burden on
802 	 * the client rather than the server.
803 	 */
804 	error = 0;
805 	bp->b_resid = bp->b_bcount;
806 	for (offset = 0, resid = bp->b_resid; resid;
807 	    resid -= sz, offset += sz) {
808 		struct buf *nbp;
809 		daddr_t nbn;
810 		int off, nra;
811 
812 		nra = 0;
813 		vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
814 		error = VOP_BMAP(vnd->sc_vp, bn / bsize, &vp, &nbn, &nra);
815 		VOP_UNLOCK(vnd->sc_vp);
816 
817 		if (error == 0 && (long)nbn == -1)
818 			error = EIO;
819 
820 		/*
821 		 * If there was an error or a hole in the file...punt.
822 		 * Note that we may have to wait for any operations
823 		 * that we have already fired off before releasing
824 		 * the buffer.
825 		 *
826 		 * XXX we could deal with holes here but it would be
827 		 * a hassle (in the write case).
828 		 */
829 		if (error) {
830 			skipped += resid;
831 			break;
832 		}
833 
834 #ifdef DEBUG
835 		if (!dovndcluster)
836 			nra = 0;
837 #endif
838 
839 		off = bn % bsize;
840 		sz = MIN(((off_t)1 + nra) * bsize - off, resid);
841 #ifdef	DEBUG
842 		if (vnddebug & VDB_IO)
843 			printf("vndstrategy: vp %p/%p bn 0x%qx/0x%" PRIx64
844 			    " sz 0x%zx\n", vnd->sc_vp, vp, (long long)bn,
845 			    nbn, sz);
846 #endif
847 
848 		nbp = getiobuf(vp, true);
849 		nestiobuf_setup(bp, nbp, offset, sz);
850 		nbp->b_blkno = nbn + btodb(off);
851 
852 #if 0 /* XXX #ifdef DEBUG */
853 		if (vnddebug & VDB_IO)
854 			printf("vndstart(%ld): bp %p vp %p blkno "
855 			    "0x%" PRIx64 " flags %x addr %p cnt 0x%x\n",
856 			    (long) (vnd-vnd_softc), &nbp->vb_buf,
857 			    nbp->vb_buf.b_vp, nbp->vb_buf.b_blkno,
858 			    nbp->vb_buf.b_flags, nbp->vb_buf.b_data,
859 			    nbp->vb_buf.b_bcount);
860 #endif
861 		VOP_STRATEGY(vp, nbp);
862 		bn += sz;
863 	}
864 	nestiobuf_done(bp, skipped, error);
865 }
866 
867 static void
868 vndiodone(struct buf *bp)
869 {
870 	struct vndxfer *vnx = VND_BUFTOXFER(bp);
871 	struct vnd_softc *vnd = vnx->vx_vnd;
872 	struct buf *obp = bp->b_private;
873 	int s = splbio();
874 
875 	KASSERT(&vnx->vx_buf == bp);
876 	KASSERT(vnd->sc_active > 0);
877 #ifdef DEBUG
878 	if (vnddebug & VDB_IO) {
879 		printf("vndiodone1: bp %p iodone: error %d\n",
880 		    bp, bp->b_error);
881 	}
882 #endif
883 	disk_unbusy(&vnd->sc_dkdev, bp->b_bcount - bp->b_resid,
884 	    (bp->b_flags & B_READ));
885 	vnd->sc_active--;
886 	if (vnd->sc_active == 0) {
887 		wakeup(&vnd->sc_tab);
888 	}
889 	splx(s);
890 	obp->b_error = bp->b_error;
891 	obp->b_resid = bp->b_resid;
892 	buf_destroy(bp);
893 	VND_PUTXFER(vnd, vnx);
894 	biodone(obp);
895 }
896 
897 /* ARGSUSED */
898 static int
899 vndread(dev_t dev, struct uio *uio, int flags)
900 {
901 	int unit = vndunit(dev);
902 	struct vnd_softc *sc;
903 
904 #ifdef DEBUG
905 	if (vnddebug & VDB_FOLLOW)
906 		printf("vndread(0x%"PRIx64", %p)\n", dev, uio);
907 #endif
908 
909 	sc = device_lookup_private(&vnd_cd, unit);
910 	if (sc == NULL)
911 		return ENXIO;
912 
913 	if ((sc->sc_flags & VNF_INITED) == 0)
914 		return ENXIO;
915 
916 	return physio(vndstrategy, NULL, dev, B_READ, minphys, uio);
917 }
918 
919 /* ARGSUSED */
920 static int
921 vndwrite(dev_t dev, struct uio *uio, int flags)
922 {
923 	int unit = vndunit(dev);
924 	struct vnd_softc *sc;
925 
926 #ifdef DEBUG
927 	if (vnddebug & VDB_FOLLOW)
928 		printf("vndwrite(0x%"PRIx64", %p)\n", dev, uio);
929 #endif
930 
931 	sc = device_lookup_private(&vnd_cd, unit);
932 	if (sc == NULL)
933 		return ENXIO;
934 
935 	if ((sc->sc_flags & VNF_INITED) == 0)
936 		return ENXIO;
937 
938 	return physio(vndstrategy, NULL, dev, B_WRITE, minphys, uio);
939 }
940 
941 static int
942 vnd_cget(struct lwp *l, int unit, int *un, struct vattr *va)
943 {
944 	int error;
945 	struct vnd_softc *vnd;
946 
947 	if (*un == -1)
948 		*un = unit;
949 	if (*un < 0)
950 		return EINVAL;
951 
952 	vnd = device_lookup_private(&vnd_cd, *un);
953 	if (vnd == NULL)
954 		return -1;
955 
956 	if ((vnd->sc_flags & VNF_INITED) == 0)
957 		return -1;
958 
959 	vn_lock(vnd->sc_vp, LK_SHARED | LK_RETRY);
960 	error = VOP_GETATTR(vnd->sc_vp, va, l->l_cred);
961 	VOP_UNLOCK(vnd->sc_vp);
962 	return error;
963 }
964 
965 static int
966 vnddoclear(struct vnd_softc *vnd, int pmask, int minor, bool force)
967 {
968 	int error;
969 
970 	if ((error = vndlock(vnd)) != 0)
971 		return error;
972 
973 	/*
974 	 * Don't unconfigure if any other partitions are open
975 	 * or if both the character and block flavors of this
976 	 * partition are open.
977 	 */
978 	if (DK_BUSY(vnd, pmask) && !force) {
979 		vndunlock(vnd);
980 		return EBUSY;
981 	}
982 
983 	/* Delete all of our wedges */
984 	dkwedge_delall(&vnd->sc_dkdev);
985 
986 	/*
987 	 * XXX vndclear() might call vndclose() implicitly;
988 	 * release lock to avoid recursion
989 	 *
990 	 * Set VNF_CLEARING to prevent vndopen() from
991 	 * sneaking in after we vndunlock().
992 	 */
993 	vnd->sc_flags |= VNF_CLEARING;
994 	vndunlock(vnd);
995 	vndclear(vnd, minor);
996 #ifdef DEBUG
997 	if (vnddebug & VDB_INIT)
998 		printf("vndioctl: CLRed\n");
999 #endif
1000 
1001 	/* Destroy the xfer and buffer pools. */
1002 	pool_destroy(&vnd->sc_vxpool);
1003 
1004 	/* Detach the disk. */
1005 	disk_detach(&vnd->sc_dkdev);
1006 
1007 	return 0;
1008 }
1009 
1010 /* ARGSUSED */
1011 static int
1012 vndioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l)
1013 {
1014 	bool force;
1015 	int unit = vndunit(dev);
1016 	struct vnd_softc *vnd;
1017 	struct vnd_ioctl *vio;
1018 	struct vattr vattr;
1019 	struct pathbuf *pb;
1020 	struct nameidata nd;
1021 	int error, part, pmask;
1022 	uint64_t geomsize;
1023 	int fflags;
1024 #ifdef __HAVE_OLD_DISKLABEL
1025 	struct disklabel newlabel;
1026 #endif
1027 	struct dkwedge_info *dkw;
1028 	struct dkwedge_list *dkwl;
1029 
1030 #ifdef DEBUG
1031 	if (vnddebug & VDB_FOLLOW)
1032 		printf("vndioctl(0x%"PRIx64", 0x%lx, %p, 0x%x, %p): unit %d\n",
1033 		    dev, cmd, data, flag, l->l_proc, unit);
1034 #endif
1035 	vnd = device_lookup_private(&vnd_cd, unit);
1036 	if (vnd == NULL &&
1037 #ifdef COMPAT_30
1038 	    cmd != VNDIOCGET30 &&
1039 #endif
1040 #ifdef COMPAT_50
1041 	    cmd != VNDIOCGET50 &&
1042 #endif
1043 	    cmd != VNDIOCGET)
1044 		return ENXIO;
1045 	vio = (struct vnd_ioctl *)data;
1046 
1047 	/* Must be open for writes for these commands... */
1048 	switch (cmd) {
1049 	case VNDIOCSET:
1050 	case VNDIOCCLR:
1051 #ifdef COMPAT_50
1052 	case VNDIOCSET50:
1053 	case VNDIOCCLR50:
1054 #endif
1055 	case DIOCSDINFO:
1056 	case DIOCWDINFO:
1057 #ifdef __HAVE_OLD_DISKLABEL
1058 	case ODIOCSDINFO:
1059 	case ODIOCWDINFO:
1060 #endif
1061 	case DIOCKLABEL:
1062 	case DIOCWLABEL:
1063 		if ((flag & FWRITE) == 0)
1064 			return EBADF;
1065 	}
1066 
1067 	/* Must be initialized for these... */
1068 	switch (cmd) {
1069 	case VNDIOCCLR:
1070 #ifdef VNDIOCCLR50
1071 	case VNDIOCCLR50:
1072 #endif
1073 	case DIOCGDINFO:
1074 	case DIOCSDINFO:
1075 	case DIOCWDINFO:
1076 	case DIOCGPART:
1077 	case DIOCKLABEL:
1078 	case DIOCWLABEL:
1079 	case DIOCGDEFLABEL:
1080 	case DIOCCACHESYNC:
1081 #ifdef __HAVE_OLD_DISKLABEL
1082 	case ODIOCGDINFO:
1083 	case ODIOCSDINFO:
1084 	case ODIOCWDINFO:
1085 	case ODIOCGDEFLABEL:
1086 #endif
1087 		if ((vnd->sc_flags & VNF_INITED) == 0)
1088 			return ENXIO;
1089 	}
1090 
1091 	switch (cmd) {
1092 #ifdef VNDIOCSET50
1093 	case VNDIOCSET50:
1094 #endif
1095 	case VNDIOCSET:
1096 		if (vnd->sc_flags & VNF_INITED)
1097 			return EBUSY;
1098 
1099 		if ((error = vndlock(vnd)) != 0)
1100 			return error;
1101 
1102 		fflags = FREAD;
1103 		if ((vio->vnd_flags & VNDIOF_READONLY) == 0)
1104 			fflags |= FWRITE;
1105 		error = pathbuf_copyin(vio->vnd_file, &pb);
1106 		if (error) {
1107 			goto unlock_and_exit;
1108 		}
1109 		NDINIT(&nd, LOOKUP, FOLLOW, pb);
1110 		if ((error = vn_open(&nd, fflags, 0)) != 0) {
1111 			pathbuf_destroy(pb);
1112 			goto unlock_and_exit;
1113 		}
1114 		KASSERT(l);
1115 		error = VOP_GETATTR(nd.ni_vp, &vattr, l->l_cred);
1116 		if (!error && nd.ni_vp->v_type != VREG)
1117 			error = EOPNOTSUPP;
1118 		if (!error && vattr.va_bytes < vattr.va_size)
1119 			/* File is definitely sparse, use vn_rdwr() */
1120 			vnd->sc_flags |= VNF_USE_VN_RDWR;
1121 		if (error) {
1122 			VOP_UNLOCK(nd.ni_vp);
1123 			goto close_and_exit;
1124 		}
1125 
1126 		/* If using a compressed file, initialize its info */
1127 		/* (or abort with an error if kernel has no compression) */
1128 		if (vio->vnd_flags & VNF_COMP) {
1129 #ifdef VND_COMPRESSION
1130 			struct vnd_comp_header *ch;
1131 			int i;
1132 			u_int32_t comp_size;
1133 			u_int32_t comp_maxsize;
1134 
1135 			/* allocate space for compresed file header */
1136 			ch = malloc(sizeof(struct vnd_comp_header),
1137 			M_TEMP, M_WAITOK);
1138 
1139 			/* read compressed file header */
1140 			error = vn_rdwr(UIO_READ, nd.ni_vp, (void *)ch,
1141 			  sizeof(struct vnd_comp_header), 0, UIO_SYSSPACE,
1142 			  IO_UNIT|IO_NODELOCKED, l->l_cred, NULL, NULL);
1143 			if (error) {
1144 				free(ch, M_TEMP);
1145 				VOP_UNLOCK(nd.ni_vp);
1146 				goto close_and_exit;
1147 			}
1148 
1149 			/* save some header info */
1150 			vnd->sc_comp_blksz = ntohl(ch->block_size);
1151 			/* note last offset is the file byte size */
1152 			vnd->sc_comp_numoffs = ntohl(ch->num_blocks)+1;
1153 			free(ch, M_TEMP);
1154 			if (vnd->sc_comp_blksz == 0 ||
1155 			    vnd->sc_comp_blksz % DEV_BSIZE !=0) {
1156 				VOP_UNLOCK(nd.ni_vp);
1157 				error = EINVAL;
1158 				goto close_and_exit;
1159 			}
1160 			if (sizeof(struct vnd_comp_header) +
1161 			  sizeof(u_int64_t) * vnd->sc_comp_numoffs >
1162 			  vattr.va_size) {
1163 				VOP_UNLOCK(nd.ni_vp);
1164 				error = EINVAL;
1165 				goto close_and_exit;
1166 			}
1167 
1168 			/* set decompressed file size */
1169 			vattr.va_size =
1170 			    ((u_quad_t)vnd->sc_comp_numoffs - 1) *
1171 			     (u_quad_t)vnd->sc_comp_blksz;
1172 
1173 			/* allocate space for all the compressed offsets */
1174 			vnd->sc_comp_offsets =
1175 			malloc(sizeof(u_int64_t) * vnd->sc_comp_numoffs,
1176 			M_DEVBUF, M_WAITOK);
1177 
1178 			/* read in the offsets */
1179 			error = vn_rdwr(UIO_READ, nd.ni_vp,
1180 			  (void *)vnd->sc_comp_offsets,
1181 			  sizeof(u_int64_t) * vnd->sc_comp_numoffs,
1182 			  sizeof(struct vnd_comp_header), UIO_SYSSPACE,
1183 			  IO_UNIT|IO_NODELOCKED, l->l_cred, NULL, NULL);
1184 			if (error) {
1185 				VOP_UNLOCK(nd.ni_vp);
1186 				goto close_and_exit;
1187 			}
1188 			/*
1189 			 * find largest block size (used for allocation limit).
1190 			 * Also convert offset to native byte order.
1191 			 */
1192 			comp_maxsize = 0;
1193 			for (i = 0; i < vnd->sc_comp_numoffs - 1; i++) {
1194 				vnd->sc_comp_offsets[i] =
1195 				  be64toh(vnd->sc_comp_offsets[i]);
1196 				comp_size = be64toh(vnd->sc_comp_offsets[i + 1])
1197 				  - vnd->sc_comp_offsets[i];
1198 				if (comp_size > comp_maxsize)
1199 					comp_maxsize = comp_size;
1200 			}
1201 			vnd->sc_comp_offsets[vnd->sc_comp_numoffs - 1] =
1202 			  be64toh(vnd->sc_comp_offsets[vnd->sc_comp_numoffs - 1]);
1203 
1204 			/* create compressed data buffer */
1205 			vnd->sc_comp_buff = malloc(comp_maxsize,
1206 			  M_DEVBUF, M_WAITOK);
1207 
1208 			/* create decompressed buffer */
1209 			vnd->sc_comp_decombuf = malloc(vnd->sc_comp_blksz,
1210 			  M_DEVBUF, M_WAITOK);
1211 			vnd->sc_comp_buffblk = -1;
1212 
1213 			/* Initialize decompress stream */
1214 			memset(&vnd->sc_comp_stream, 0, sizeof(z_stream));
1215 			vnd->sc_comp_stream.zalloc = vnd_alloc;
1216 			vnd->sc_comp_stream.zfree = vnd_free;
1217 			error = inflateInit2(&vnd->sc_comp_stream, MAX_WBITS);
1218 			if (error) {
1219 				if (vnd->sc_comp_stream.msg)
1220 					printf("vnd%d: compressed file, %s\n",
1221 					  unit, vnd->sc_comp_stream.msg);
1222 				VOP_UNLOCK(nd.ni_vp);
1223 				error = EINVAL;
1224 				goto close_and_exit;
1225 			}
1226 
1227 			vnd->sc_flags |= VNF_COMP | VNF_READONLY;
1228 #else /* !VND_COMPRESSION */
1229 			VOP_UNLOCK(nd.ni_vp);
1230 			error = EOPNOTSUPP;
1231 			goto close_and_exit;
1232 #endif /* VND_COMPRESSION */
1233 		}
1234 
1235 		VOP_UNLOCK(nd.ni_vp);
1236 		vnd->sc_vp = nd.ni_vp;
1237 		vnd->sc_size = btodb(vattr.va_size);	/* note truncation */
1238 
1239 		/*
1240 		 * Use pseudo-geometry specified.  If none was provided,
1241 		 * use "standard" Adaptec fictitious geometry.
1242 		 */
1243 		if (vio->vnd_flags & VNDIOF_HASGEOM) {
1244 
1245 			memcpy(&vnd->sc_geom, &vio->vnd_geom,
1246 			    sizeof(vio->vnd_geom));
1247 
1248 			/*
1249 			 * Sanity-check the sector size.
1250 			 * XXX Don't allow secsize < DEV_BSIZE.	 Should
1251 			 * XXX we?
1252 			 */
1253 			if (vnd->sc_geom.vng_secsize < DEV_BSIZE ||
1254 			    (vnd->sc_geom.vng_secsize % DEV_BSIZE) != 0 ||
1255 			    vnd->sc_geom.vng_ncylinders == 0 ||
1256 			    (vnd->sc_geom.vng_ntracks *
1257 			     vnd->sc_geom.vng_nsectors) == 0) {
1258 				error = EINVAL;
1259 				goto close_and_exit;
1260 			}
1261 
1262 			/*
1263 			 * Compute the size (in DEV_BSIZE blocks) specified
1264 			 * by the geometry.
1265 			 */
1266 			geomsize = (vnd->sc_geom.vng_nsectors *
1267 			    vnd->sc_geom.vng_ntracks *
1268 			    vnd->sc_geom.vng_ncylinders) *
1269 			    (vnd->sc_geom.vng_secsize / DEV_BSIZE);
1270 
1271 			/*
1272 			 * Sanity-check the size against the specified
1273 			 * geometry.
1274 			 */
1275 			if (vnd->sc_size < geomsize) {
1276 				error = EINVAL;
1277 				goto close_and_exit;
1278 			}
1279 		} else if (vnd->sc_size >= (32 * 64)) {
1280 			/*
1281 			 * Size must be at least 2048 DEV_BSIZE blocks
1282 			 * (1M) in order to use this geometry.
1283 			 */
1284 			vnd->sc_geom.vng_secsize = DEV_BSIZE;
1285 			vnd->sc_geom.vng_nsectors = 32;
1286 			vnd->sc_geom.vng_ntracks = 64;
1287 			vnd->sc_geom.vng_ncylinders = vnd->sc_size / (64 * 32);
1288 		} else {
1289 			vnd->sc_geom.vng_secsize = DEV_BSIZE;
1290 			vnd->sc_geom.vng_nsectors = 1;
1291 			vnd->sc_geom.vng_ntracks = 1;
1292 			vnd->sc_geom.vng_ncylinders = vnd->sc_size;
1293 		}
1294 
1295 		vnd_set_geometry(vnd);
1296 
1297 		if (vio->vnd_flags & VNDIOF_READONLY) {
1298 			vnd->sc_flags |= VNF_READONLY;
1299 		}
1300 
1301 		if ((error = vndsetcred(vnd, l->l_cred)) != 0)
1302 			goto close_and_exit;
1303 
1304 		vndthrottle(vnd, vnd->sc_vp);
1305 		vio->vnd_osize = dbtob(vnd->sc_size);
1306 #ifdef VNDIOCSET50
1307 		if (cmd != VNDIOCSET50)
1308 #endif
1309 			vio->vnd_size = dbtob(vnd->sc_size);
1310 		vnd->sc_flags |= VNF_INITED;
1311 
1312 		/* create the kernel thread, wait for it to be up */
1313 		error = kthread_create(PRI_NONE, 0, NULL, vndthread, vnd,
1314 		    &vnd->sc_kthread, "%s", device_xname(vnd->sc_dev));
1315 		if (error)
1316 			goto close_and_exit;
1317 		while ((vnd->sc_flags & VNF_KTHREAD) == 0) {
1318 			tsleep(&vnd->sc_kthread, PRIBIO, "vndthr", 0);
1319 		}
1320 #ifdef DEBUG
1321 		if (vnddebug & VDB_INIT)
1322 			printf("vndioctl: SET vp %p size 0x%lx %d/%d/%d/%d\n",
1323 			    vnd->sc_vp, (unsigned long) vnd->sc_size,
1324 			    vnd->sc_geom.vng_secsize,
1325 			    vnd->sc_geom.vng_nsectors,
1326 			    vnd->sc_geom.vng_ntracks,
1327 			    vnd->sc_geom.vng_ncylinders);
1328 #endif
1329 
1330 		/* Attach the disk. */
1331 		disk_attach(&vnd->sc_dkdev);
1332 		disk_blocksize(&vnd->sc_dkdev, vnd->sc_geom.vng_secsize);
1333 
1334 		/* Initialize the xfer and buffer pools. */
1335 		pool_init(&vnd->sc_vxpool, sizeof(struct vndxfer), 0,
1336 		    0, 0, "vndxpl", NULL, IPL_BIO);
1337 
1338 		vndunlock(vnd);
1339 
1340 		pathbuf_destroy(pb);
1341 
1342 		/* Discover wedges on this disk */
1343 		dkwedge_discover(&vnd->sc_dkdev);
1344 
1345 		break;
1346 
1347 close_and_exit:
1348 		(void) vn_close(nd.ni_vp, fflags, l->l_cred);
1349 		pathbuf_destroy(pb);
1350 unlock_and_exit:
1351 #ifdef VND_COMPRESSION
1352 		/* free any allocated memory (for compressed file) */
1353 		if (vnd->sc_comp_offsets) {
1354 			free(vnd->sc_comp_offsets, M_DEVBUF);
1355 			vnd->sc_comp_offsets = NULL;
1356 		}
1357 		if (vnd->sc_comp_buff) {
1358 			free(vnd->sc_comp_buff, M_DEVBUF);
1359 			vnd->sc_comp_buff = NULL;
1360 		}
1361 		if (vnd->sc_comp_decombuf) {
1362 			free(vnd->sc_comp_decombuf, M_DEVBUF);
1363 			vnd->sc_comp_decombuf = NULL;
1364 		}
1365 #endif /* VND_COMPRESSION */
1366 		vndunlock(vnd);
1367 		return error;
1368 
1369 #ifdef VNDIOCCLR50
1370 	case VNDIOCCLR50:
1371 #endif
1372 	case VNDIOCCLR:
1373 		part = DISKPART(dev);
1374 		pmask = (1 << part);
1375 		force = (vio->vnd_flags & VNDIOF_FORCE) != 0;
1376 
1377 		if ((error = vnddoclear(vnd, pmask, minor(dev), force)) != 0)
1378 			return error;
1379 
1380 		break;
1381 
1382 #ifdef COMPAT_30
1383 	case VNDIOCGET30: {
1384 		struct vnd_user30 *vnu;
1385 		struct vattr va;
1386 		vnu = (struct vnd_user30 *)data;
1387 		KASSERT(l);
1388 		switch (error = vnd_cget(l, unit, &vnu->vnu_unit, &va)) {
1389 		case 0:
1390 			vnu->vnu_dev = va.va_fsid;
1391 			vnu->vnu_ino = va.va_fileid;
1392 			break;
1393 		case -1:
1394 			/* unused is not an error */
1395 			vnu->vnu_dev = 0;
1396 			vnu->vnu_ino = 0;
1397 			break;
1398 		default:
1399 			return error;
1400 		}
1401 		break;
1402 	}
1403 #endif
1404 
1405 #ifdef COMPAT_50
1406 	case VNDIOCGET50: {
1407 		struct vnd_user50 *vnu;
1408 		struct vattr va;
1409 		vnu = (struct vnd_user50 *)data;
1410 		KASSERT(l);
1411 		switch (error = vnd_cget(l, unit, &vnu->vnu_unit, &va)) {
1412 		case 0:
1413 			vnu->vnu_dev = va.va_fsid;
1414 			vnu->vnu_ino = va.va_fileid;
1415 			break;
1416 		case -1:
1417 			/* unused is not an error */
1418 			vnu->vnu_dev = 0;
1419 			vnu->vnu_ino = 0;
1420 			break;
1421 		default:
1422 			return error;
1423 		}
1424 		break;
1425 	}
1426 #endif
1427 
1428 	case VNDIOCGET: {
1429 		struct vnd_user *vnu;
1430 		struct vattr va;
1431 		vnu = (struct vnd_user *)data;
1432 		KASSERT(l);
1433 		switch (error = vnd_cget(l, unit, &vnu->vnu_unit, &va)) {
1434 		case 0:
1435 			vnu->vnu_dev = va.va_fsid;
1436 			vnu->vnu_ino = va.va_fileid;
1437 			break;
1438 		case -1:
1439 			/* unused is not an error */
1440 			vnu->vnu_dev = 0;
1441 			vnu->vnu_ino = 0;
1442 			break;
1443 		default:
1444 			return error;
1445 		}
1446 		break;
1447 	}
1448 
1449 	case DIOCGDINFO:
1450 		*(struct disklabel *)data = *(vnd->sc_dkdev.dk_label);
1451 		break;
1452 
1453 #ifdef __HAVE_OLD_DISKLABEL
1454 	case ODIOCGDINFO:
1455 		newlabel = *(vnd->sc_dkdev.dk_label);
1456 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1457 			return ENOTTY;
1458 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1459 		break;
1460 #endif
1461 
1462 	case DIOCGPART:
1463 		((struct partinfo *)data)->disklab = vnd->sc_dkdev.dk_label;
1464 		((struct partinfo *)data)->part =
1465 		    &vnd->sc_dkdev.dk_label->d_partitions[DISKPART(dev)];
1466 		break;
1467 
1468 	case DIOCWDINFO:
1469 	case DIOCSDINFO:
1470 #ifdef __HAVE_OLD_DISKLABEL
1471 	case ODIOCWDINFO:
1472 	case ODIOCSDINFO:
1473 #endif
1474 	{
1475 		struct disklabel *lp;
1476 
1477 		if ((error = vndlock(vnd)) != 0)
1478 			return error;
1479 
1480 		vnd->sc_flags |= VNF_LABELLING;
1481 
1482 #ifdef __HAVE_OLD_DISKLABEL
1483 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
1484 			memset(&newlabel, 0, sizeof newlabel);
1485 			memcpy(&newlabel, data, sizeof (struct olddisklabel));
1486 			lp = &newlabel;
1487 		} else
1488 #endif
1489 		lp = (struct disklabel *)data;
1490 
1491 		error = setdisklabel(vnd->sc_dkdev.dk_label,
1492 		    lp, 0, vnd->sc_dkdev.dk_cpulabel);
1493 		if (error == 0) {
1494 			if (cmd == DIOCWDINFO
1495 #ifdef __HAVE_OLD_DISKLABEL
1496 			    || cmd == ODIOCWDINFO
1497 #endif
1498 			   )
1499 				error = writedisklabel(VNDLABELDEV(dev),
1500 				    vndstrategy, vnd->sc_dkdev.dk_label,
1501 				    vnd->sc_dkdev.dk_cpulabel);
1502 		}
1503 
1504 		vnd->sc_flags &= ~VNF_LABELLING;
1505 
1506 		vndunlock(vnd);
1507 
1508 		if (error)
1509 			return error;
1510 		break;
1511 	}
1512 
1513 	case DIOCKLABEL:
1514 		if (*(int *)data != 0)
1515 			vnd->sc_flags |= VNF_KLABEL;
1516 		else
1517 			vnd->sc_flags &= ~VNF_KLABEL;
1518 		break;
1519 
1520 	case DIOCWLABEL:
1521 		if (*(int *)data != 0)
1522 			vnd->sc_flags |= VNF_WLABEL;
1523 		else
1524 			vnd->sc_flags &= ~VNF_WLABEL;
1525 		break;
1526 
1527 	case DIOCGDEFLABEL:
1528 		vndgetdefaultlabel(vnd, (struct disklabel *)data);
1529 		break;
1530 
1531 #ifdef __HAVE_OLD_DISKLABEL
1532 	case ODIOCGDEFLABEL:
1533 		vndgetdefaultlabel(vnd, &newlabel);
1534 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1535 			return ENOTTY;
1536 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1537 		break;
1538 #endif
1539 
1540 	case DIOCCACHESYNC:
1541 		vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1542 		error = VOP_FSYNC(vnd->sc_vp, vnd->sc_cred,
1543 		    FSYNC_WAIT | FSYNC_DATAONLY | FSYNC_CACHE, 0, 0);
1544 		VOP_UNLOCK(vnd->sc_vp);
1545 		return error;
1546 
1547 	case DIOCAWEDGE:
1548 		dkw = (void *) data;
1549 
1550 		if ((flag & FWRITE) == 0)
1551 			return EBADF;
1552 
1553 		/* If the ioctl happens here, the parent is us. */
1554 		strlcpy(dkw->dkw_parent, device_xname(vnd->sc_dev),
1555 		    sizeof(dkw->dkw_parent));
1556 		return dkwedge_add(dkw);
1557 
1558 	case DIOCDWEDGE:
1559 		dkw = (void *) data;
1560 
1561 		if ((flag & FWRITE) == 0)
1562 			return EBADF;
1563 
1564 		/* If the ioctl happens here, the parent is us. */
1565 		strlcpy(dkw->dkw_parent, device_xname(vnd->sc_dev),
1566 		    sizeof(dkw->dkw_parent));
1567 		return dkwedge_del(dkw);
1568 
1569 	case DIOCLWEDGES:
1570 		dkwl = (void *) data;
1571 
1572 		return dkwedge_list(&vnd->sc_dkdev, dkwl, l);
1573 
1574 	default:
1575 		return ENOTTY;
1576 	}
1577 
1578 	return 0;
1579 }
1580 
1581 /*
1582  * Duplicate the current processes' credentials.  Since we are called only
1583  * as the result of a SET ioctl and only root can do that, any future access
1584  * to this "disk" is essentially as root.  Note that credentials may change
1585  * if some other uid can write directly to the mapped file (NFS).
1586  */
1587 static int
1588 vndsetcred(struct vnd_softc *vnd, kauth_cred_t cred)
1589 {
1590 	struct uio auio;
1591 	struct iovec aiov;
1592 	char *tmpbuf;
1593 	int error;
1594 
1595 	vnd->sc_cred = kauth_cred_dup(cred);
1596 	tmpbuf = malloc(DEV_BSIZE, M_TEMP, M_WAITOK);
1597 
1598 	/* XXX: Horrible kludge to establish credentials for NFS */
1599 	aiov.iov_base = tmpbuf;
1600 	aiov.iov_len = min(DEV_BSIZE, dbtob(vnd->sc_size));
1601 	auio.uio_iov = &aiov;
1602 	auio.uio_iovcnt = 1;
1603 	auio.uio_offset = 0;
1604 	auio.uio_rw = UIO_READ;
1605 	auio.uio_resid = aiov.iov_len;
1606 	UIO_SETUP_SYSSPACE(&auio);
1607 	vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1608 	error = VOP_READ(vnd->sc_vp, &auio, 0, vnd->sc_cred);
1609 	if (error == 0) {
1610 		/*
1611 		 * Because vnd does all IO directly through the vnode
1612 		 * we need to flush (at least) the buffer from the above
1613 		 * VOP_READ from the buffer cache to prevent cache
1614 		 * incoherencies.  Also, be careful to write dirty
1615 		 * buffers back to stable storage.
1616 		 */
1617 		error = vinvalbuf(vnd->sc_vp, V_SAVE, vnd->sc_cred,
1618 			    curlwp, 0, 0);
1619 	}
1620 	VOP_UNLOCK(vnd->sc_vp);
1621 
1622 	free(tmpbuf, M_TEMP);
1623 	return error;
1624 }
1625 
1626 /*
1627  * Set maxactive based on FS type
1628  */
1629 static void
1630 vndthrottle(struct vnd_softc *vnd, struct vnode *vp)
1631 {
1632 
1633 	if (vp->v_tag == VT_NFS)
1634 		vnd->sc_maxactive = 2;
1635 	else
1636 		vnd->sc_maxactive = 8;
1637 
1638 	if (vnd->sc_maxactive < 1)
1639 		vnd->sc_maxactive = 1;
1640 }
1641 
1642 #if 0
1643 static void
1644 vndshutdown(void)
1645 {
1646 	struct vnd_softc *vnd;
1647 
1648 	for (vnd = &vnd_softc[0]; vnd < &vnd_softc[numvnd]; vnd++)
1649 		if (vnd->sc_flags & VNF_INITED)
1650 			vndclear(vnd);
1651 }
1652 #endif
1653 
1654 static void
1655 vndclear(struct vnd_softc *vnd, int myminor)
1656 {
1657 	struct vnode *vp = vnd->sc_vp;
1658 	int fflags = FREAD;
1659 	int bmaj, cmaj, i, mn;
1660 	int s;
1661 
1662 #ifdef DEBUG
1663 	if (vnddebug & VDB_FOLLOW)
1664 		printf("vndclear(%p): vp %p\n", vnd, vp);
1665 #endif
1666 	/* locate the major number */
1667 	bmaj = bdevsw_lookup_major(&vnd_bdevsw);
1668 	cmaj = cdevsw_lookup_major(&vnd_cdevsw);
1669 
1670 	/* Nuke the vnodes for any open instances */
1671 	for (i = 0; i < MAXPARTITIONS; i++) {
1672 		mn = DISKMINOR(device_unit(vnd->sc_dev), i);
1673 		vdevgone(bmaj, mn, mn, VBLK);
1674 		if (mn != myminor) /* XXX avoid to kill own vnode */
1675 			vdevgone(cmaj, mn, mn, VCHR);
1676 	}
1677 
1678 	if ((vnd->sc_flags & VNF_READONLY) == 0)
1679 		fflags |= FWRITE;
1680 
1681 	s = splbio();
1682 	bufq_drain(vnd->sc_tab);
1683 	splx(s);
1684 
1685 	vnd->sc_flags |= VNF_VUNCONF;
1686 	wakeup(&vnd->sc_tab);
1687 	while (vnd->sc_flags & VNF_KTHREAD)
1688 		tsleep(&vnd->sc_kthread, PRIBIO, "vnthr", 0);
1689 
1690 #ifdef VND_COMPRESSION
1691 	/* free the compressed file buffers */
1692 	if (vnd->sc_flags & VNF_COMP) {
1693 		if (vnd->sc_comp_offsets) {
1694 			free(vnd->sc_comp_offsets, M_DEVBUF);
1695 			vnd->sc_comp_offsets = NULL;
1696 		}
1697 		if (vnd->sc_comp_buff) {
1698 			free(vnd->sc_comp_buff, M_DEVBUF);
1699 			vnd->sc_comp_buff = NULL;
1700 		}
1701 		if (vnd->sc_comp_decombuf) {
1702 			free(vnd->sc_comp_decombuf, M_DEVBUF);
1703 			vnd->sc_comp_decombuf = NULL;
1704 		}
1705 	}
1706 #endif /* VND_COMPRESSION */
1707 	vnd->sc_flags &=
1708 	    ~(VNF_INITED | VNF_READONLY | VNF_VLABEL
1709 	      | VNF_VUNCONF | VNF_COMP | VNF_CLEARING);
1710 	if (vp == NULL)
1711 		panic("vndclear: null vp");
1712 	(void) vn_close(vp, fflags, vnd->sc_cred);
1713 	kauth_cred_free(vnd->sc_cred);
1714 	vnd->sc_vp = NULL;
1715 	vnd->sc_cred = NULL;
1716 	vnd->sc_size = 0;
1717 }
1718 
1719 static int
1720 vndsize(dev_t dev)
1721 {
1722 	struct vnd_softc *sc;
1723 	struct disklabel *lp;
1724 	int part, unit, omask;
1725 	int size;
1726 
1727 	unit = vndunit(dev);
1728 	sc = device_lookup_private(&vnd_cd, unit);
1729 	if (sc == NULL)
1730 		return -1;
1731 
1732 	if ((sc->sc_flags & VNF_INITED) == 0)
1733 		return -1;
1734 
1735 	part = DISKPART(dev);
1736 	omask = sc->sc_dkdev.dk_openmask & (1 << part);
1737 	lp = sc->sc_dkdev.dk_label;
1738 
1739 	if (omask == 0 && vndopen(dev, 0, S_IFBLK, curlwp))	/* XXX */
1740 		return -1;
1741 
1742 	if (lp->d_partitions[part].p_fstype != FS_SWAP)
1743 		size = -1;
1744 	else
1745 		size = lp->d_partitions[part].p_size *
1746 		    (lp->d_secsize / DEV_BSIZE);
1747 
1748 	if (omask == 0 && vndclose(dev, 0, S_IFBLK, curlwp))	/* XXX */
1749 		return -1;
1750 
1751 	return size;
1752 }
1753 
1754 static int
1755 vnddump(dev_t dev, daddr_t blkno, void *va,
1756     size_t size)
1757 {
1758 
1759 	/* Not implemented. */
1760 	return ENXIO;
1761 }
1762 
1763 static void
1764 vndgetdefaultlabel(struct vnd_softc *sc, struct disklabel *lp)
1765 {
1766 	struct vndgeom *vng = &sc->sc_geom;
1767 	struct partition *pp;
1768 
1769 	memset(lp, 0, sizeof(*lp));
1770 
1771 	lp->d_secperunit = sc->sc_size / (vng->vng_secsize / DEV_BSIZE);
1772 	lp->d_secsize = vng->vng_secsize;
1773 	lp->d_nsectors = vng->vng_nsectors;
1774 	lp->d_ntracks = vng->vng_ntracks;
1775 	lp->d_ncylinders = vng->vng_ncylinders;
1776 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1777 
1778 	strncpy(lp->d_typename, "vnd", sizeof(lp->d_typename));
1779 	lp->d_type = DTYPE_VND;
1780 	strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
1781 	lp->d_rpm = 3600;
1782 	lp->d_interleave = 1;
1783 	lp->d_flags = 0;
1784 
1785 	pp = &lp->d_partitions[RAW_PART];
1786 	pp->p_offset = 0;
1787 	pp->p_size = lp->d_secperunit;
1788 	pp->p_fstype = FS_UNUSED;
1789 	lp->d_npartitions = RAW_PART + 1;
1790 
1791 	lp->d_magic = DISKMAGIC;
1792 	lp->d_magic2 = DISKMAGIC;
1793 	lp->d_checksum = dkcksum(lp);
1794 }
1795 
1796 /*
1797  * Read the disklabel from a vnd.  If one is not present, create a fake one.
1798  */
1799 static void
1800 vndgetdisklabel(dev_t dev, struct vnd_softc *sc)
1801 {
1802 	const char *errstring;
1803 	struct disklabel *lp = sc->sc_dkdev.dk_label;
1804 	struct cpu_disklabel *clp = sc->sc_dkdev.dk_cpulabel;
1805 	int i;
1806 
1807 	memset(clp, 0, sizeof(*clp));
1808 
1809 	vndgetdefaultlabel(sc, lp);
1810 
1811 	/*
1812 	 * Call the generic disklabel extraction routine.
1813 	 */
1814 	errstring = readdisklabel(VNDLABELDEV(dev), vndstrategy, lp, clp);
1815 	if (errstring) {
1816 		/*
1817 		 * Lack of disklabel is common, but we print the warning
1818 		 * anyway, since it might contain other useful information.
1819 		 */
1820 		aprint_normal_dev(sc->sc_dev, "%s\n", errstring);
1821 
1822 		/*
1823 		 * For historical reasons, if there's no disklabel
1824 		 * present, all partitions must be FS_BSDFFS and
1825 		 * occupy the entire disk.
1826 		 */
1827 		for (i = 0; i < MAXPARTITIONS; i++) {
1828 			/*
1829 			 * Don't wipe out port specific hack (such as
1830 			 * dos partition hack of i386 port).
1831 			 */
1832 			if (lp->d_partitions[i].p_size != 0)
1833 				continue;
1834 
1835 			lp->d_partitions[i].p_size = lp->d_secperunit;
1836 			lp->d_partitions[i].p_offset = 0;
1837 			lp->d_partitions[i].p_fstype = FS_BSDFFS;
1838 		}
1839 
1840 		strncpy(lp->d_packname, "default label",
1841 		    sizeof(lp->d_packname));
1842 
1843 		lp->d_npartitions = MAXPARTITIONS;
1844 		lp->d_checksum = dkcksum(lp);
1845 	}
1846 }
1847 
1848 /*
1849  * Wait interruptibly for an exclusive lock.
1850  *
1851  * XXX
1852  * Several drivers do this; it should be abstracted and made MP-safe.
1853  */
1854 static int
1855 vndlock(struct vnd_softc *sc)
1856 {
1857 	int error;
1858 
1859 	while ((sc->sc_flags & VNF_LOCKED) != 0) {
1860 		sc->sc_flags |= VNF_WANTED;
1861 		if ((error = tsleep(sc, PRIBIO | PCATCH, "vndlck", 0)) != 0)
1862 			return error;
1863 	}
1864 	sc->sc_flags |= VNF_LOCKED;
1865 	return 0;
1866 }
1867 
1868 /*
1869  * Unlock and wake up any waiters.
1870  */
1871 static void
1872 vndunlock(struct vnd_softc *sc)
1873 {
1874 
1875 	sc->sc_flags &= ~VNF_LOCKED;
1876 	if ((sc->sc_flags & VNF_WANTED) != 0) {
1877 		sc->sc_flags &= ~VNF_WANTED;
1878 		wakeup(sc);
1879 	}
1880 }
1881 
1882 #ifdef VND_COMPRESSION
1883 /* compressed file read */
1884 static void
1885 compstrategy(struct buf *bp, off_t bn)
1886 {
1887 	int error;
1888 	int unit = vndunit(bp->b_dev);
1889 	struct vnd_softc *vnd =
1890 	    device_lookup_private(&vnd_cd, unit);
1891 	u_int32_t comp_block;
1892 	struct uio auio;
1893 	char *addr;
1894 	int s;
1895 
1896 	/* set up constants for data move */
1897 	auio.uio_rw = UIO_READ;
1898 	UIO_SETUP_SYSSPACE(&auio);
1899 
1900 	/* read, and transfer the data */
1901 	addr = bp->b_data;
1902 	bp->b_resid = bp->b_bcount;
1903 	s = splbio();
1904 	while (bp->b_resid > 0) {
1905 		unsigned length;
1906 		size_t length_in_buffer;
1907 		u_int32_t offset_in_buffer;
1908 		struct iovec aiov;
1909 
1910 		/* calculate the compressed block number */
1911 		comp_block = bn / (off_t)vnd->sc_comp_blksz;
1912 
1913 		/* check for good block number */
1914 		if (comp_block >= vnd->sc_comp_numoffs) {
1915 			bp->b_error = EINVAL;
1916 			splx(s);
1917 			return;
1918 		}
1919 
1920 		/* read in the compressed block, if not in buffer */
1921 		if (comp_block != vnd->sc_comp_buffblk) {
1922 			length = vnd->sc_comp_offsets[comp_block + 1] -
1923 			    vnd->sc_comp_offsets[comp_block];
1924 			vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1925 			error = vn_rdwr(UIO_READ, vnd->sc_vp, vnd->sc_comp_buff,
1926 			    length, vnd->sc_comp_offsets[comp_block],
1927 			    UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, vnd->sc_cred,
1928 			    NULL, NULL);
1929 			if (error) {
1930 				bp->b_error = error;
1931 				VOP_UNLOCK(vnd->sc_vp);
1932 				splx(s);
1933 				return;
1934 			}
1935 			/* uncompress the buffer */
1936 			vnd->sc_comp_stream.next_in = vnd->sc_comp_buff;
1937 			vnd->sc_comp_stream.avail_in = length;
1938 			vnd->sc_comp_stream.next_out = vnd->sc_comp_decombuf;
1939 			vnd->sc_comp_stream.avail_out = vnd->sc_comp_blksz;
1940 			inflateReset(&vnd->sc_comp_stream);
1941 			error = inflate(&vnd->sc_comp_stream, Z_FINISH);
1942 			if (error != Z_STREAM_END) {
1943 				if (vnd->sc_comp_stream.msg)
1944 					aprint_normal_dev(vnd->sc_dev,
1945 					    "compressed file, %s\n",
1946 					    vnd->sc_comp_stream.msg);
1947 				bp->b_error = EBADMSG;
1948 				VOP_UNLOCK(vnd->sc_vp);
1949 				splx(s);
1950 				return;
1951 			}
1952 			vnd->sc_comp_buffblk = comp_block;
1953 			VOP_UNLOCK(vnd->sc_vp);
1954 		}
1955 
1956 		/* transfer the usable uncompressed data */
1957 		offset_in_buffer = bn % (off_t)vnd->sc_comp_blksz;
1958 		length_in_buffer = vnd->sc_comp_blksz - offset_in_buffer;
1959 		if (length_in_buffer > bp->b_resid)
1960 			length_in_buffer = bp->b_resid;
1961 		auio.uio_iov = &aiov;
1962 		auio.uio_iovcnt = 1;
1963 		aiov.iov_base = addr;
1964 		aiov.iov_len = length_in_buffer;
1965 		auio.uio_resid = aiov.iov_len;
1966 		auio.uio_offset = 0;
1967 		error = uiomove(vnd->sc_comp_decombuf + offset_in_buffer,
1968 		    length_in_buffer, &auio);
1969 		if (error) {
1970 			bp->b_error = error;
1971 			splx(s);
1972 			return;
1973 		}
1974 
1975 		bn += length_in_buffer;
1976 		addr += length_in_buffer;
1977 		bp->b_resid -= length_in_buffer;
1978 	}
1979 	splx(s);
1980 }
1981 
1982 /* compression memory allocation routines */
1983 static void *
1984 vnd_alloc(void *aux, u_int items, u_int siz)
1985 {
1986 	return malloc(items * siz, M_TEMP, M_NOWAIT);
1987 }
1988 
1989 static void
1990 vnd_free(void *aux, void *ptr)
1991 {
1992 	free(ptr, M_TEMP);
1993 }
1994 #endif /* VND_COMPRESSION */
1995 
1996 static void
1997 vnd_set_geometry(struct vnd_softc *vnd)
1998 {
1999 	struct disk_geom *dg = &vnd->sc_dkdev.dk_geom;
2000 
2001 	memset(dg, 0, sizeof(*dg));
2002 
2003 	dg->dg_secperunit = (int64_t)vnd->sc_geom.vng_nsectors *
2004 	    vnd->sc_geom.vng_ntracks * vnd->sc_geom.vng_ncylinders;
2005 	dg->dg_secsize = vnd->sc_geom.vng_secsize;
2006 	dg->dg_nsectors = vnd->sc_geom.vng_nsectors;
2007 	dg->dg_ntracks = vnd->sc_geom.vng_ntracks;
2008 	dg->dg_ncylinders = vnd->sc_geom.vng_ncylinders;
2009 
2010 	disk_set_info(vnd->sc_dev, &vnd->sc_dkdev, NULL);
2011 }
2012 
2013 #ifdef _MODULE
2014 
2015 #include <sys/module.h>
2016 
2017 MODULE(MODULE_CLASS_DRIVER, vnd, "zlib");
2018 CFDRIVER_DECL(vnd, DV_DISK, NULL);
2019 
2020 static int
2021 vnd_modcmd(modcmd_t cmd, void *arg)
2022 {
2023 	int bmajor = -1, cmajor = -1,  error = 0;
2024 
2025 	switch (cmd) {
2026 	case MODULE_CMD_INIT:
2027 		error = config_cfdriver_attach(&vnd_cd);
2028 		if (error)
2029 			break;
2030 
2031 		error = config_cfattach_attach(vnd_cd.cd_name, &vnd_ca);
2032 	        if (error) {
2033 			config_cfdriver_detach(&vnd_cd);
2034 			aprint_error("%s: unable to register cfattach\n",
2035 			    vnd_cd.cd_name);
2036 			break;
2037 		}
2038 
2039 		error = devsw_attach("vnd", &vnd_bdevsw, &bmajor,
2040 		    &vnd_cdevsw, &cmajor);
2041 		if (error) {
2042 			config_cfattach_detach(vnd_cd.cd_name, &vnd_ca);
2043 			config_cfdriver_detach(&vnd_cd);
2044 			break;
2045 		}
2046 
2047 		break;
2048 
2049 	case MODULE_CMD_FINI:
2050 		error = config_cfattach_detach(vnd_cd.cd_name, &vnd_ca);
2051 		if (error)
2052 			break;
2053 		config_cfdriver_detach(&vnd_cd);
2054 		devsw_detach(&vnd_bdevsw, &vnd_cdevsw);
2055 		break;
2056 
2057 	case MODULE_CMD_STAT:
2058 		return ENOTTY;
2059 
2060 	default:
2061 		return ENOTTY;
2062 	}
2063 
2064 	return error;
2065 }
2066 
2067 #endif
2068