xref: /netbsd-src/sys/dev/vnd.c (revision 4e6df137e8e14049b5a701d249962c480449c141)
1 /*	$NetBSD: vnd.c,v 1.208 2010/03/02 21:32:29 pooka 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) 1990, 1993
34  *	The Regents of the University of California.  All rights reserved.
35  *
36  * This code is derived from software contributed to Berkeley by
37  * the Systems Programming Group of the University of Utah Computer
38  * Science Department.
39  *
40  * Redistribution and use in source and binary forms, with or without
41  * modification, are permitted provided that the following conditions
42  * are met:
43  * 1. Redistributions of source code must retain the above copyright
44  *    notice, this list of conditions and the following disclaimer.
45  * 2. Redistributions in binary form must reproduce the above copyright
46  *    notice, this list of conditions and the following disclaimer in the
47  *    documentation and/or other materials provided with the distribution.
48  * 3. Neither the name of the University nor the names of its contributors
49  *    may be used to endorse or promote products derived from this software
50  *    without specific prior written permission.
51  *
52  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
53  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
54  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
55  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
56  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
57  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
58  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
59  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
60  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
61  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
62  * SUCH DAMAGE.
63  *
64  * from: Utah $Hdr: vn.c 1.13 94/04/02$
65  *
66  *	@(#)vn.c	8.9 (Berkeley) 5/14/95
67  */
68 
69 /*
70  * Copyright (c) 1988 University of Utah.
71  *
72  * This code is derived from software contributed to Berkeley by
73  * the Systems Programming Group of the University of Utah Computer
74  * Science Department.
75  *
76  * Redistribution and use in source and binary forms, with or without
77  * modification, are permitted provided that the following conditions
78  * are met:
79  * 1. Redistributions of source code must retain the above copyright
80  *    notice, this list of conditions and the following disclaimer.
81  * 2. Redistributions in binary form must reproduce the above copyright
82  *    notice, this list of conditions and the following disclaimer in the
83  *    documentation and/or other materials provided with the distribution.
84  * 3. All advertising materials mentioning features or use of this software
85  *    must display the following acknowledgement:
86  *	This product includes software developed by the University of
87  *	California, Berkeley and its contributors.
88  * 4. Neither the name of the University nor the names of its contributors
89  *    may be used to endorse or promote products derived from this software
90  *    without specific prior written permission.
91  *
92  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
93  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
94  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
95  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
96  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
97  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
98  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
99  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
100  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
101  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
102  * SUCH DAMAGE.
103  *
104  * from: Utah $Hdr: vn.c 1.13 94/04/02$
105  *
106  *	@(#)vn.c	8.9 (Berkeley) 5/14/95
107  */
108 
109 /*
110  * Vnode disk driver.
111  *
112  * Block/character interface to a vnode.  Allows one to treat a file
113  * as a disk (e.g. build a filesystem in it, mount it, etc.).
114  *
115  * NOTE 1: If the vnode supports the VOP_BMAP and VOP_STRATEGY operations,
116  * this uses them to avoid distorting the local buffer cache.  If those
117  * block-level operations are not available, this falls back to the regular
118  * read and write calls.  Using these may distort the cache in some cases
119  * but better have the driver working than preventing it to work on file
120  * systems where the block-level operations are not implemented for
121  * whatever reason.
122  *
123  * NOTE 2: There is a security issue involved with this driver.
124  * Once mounted all access to the contents of the "mapped" file via
125  * the special file is controlled by the permissions on the special
126  * file, the protection of the mapped file is ignored (effectively,
127  * by using root credentials in all transactions).
128  *
129  * NOTE 3: Doesn't interact with leases, should it?
130  */
131 
132 #include <sys/cdefs.h>
133 __KERNEL_RCSID(0, "$NetBSD: vnd.c,v 1.208 2010/03/02 21:32:29 pooka Exp $");
134 
135 #if defined(_KERNEL_OPT)
136 #include "opt_vnd.h"
137 #endif
138 
139 #include <sys/param.h>
140 #include <sys/systm.h>
141 #include <sys/namei.h>
142 #include <sys/proc.h>
143 #include <sys/kthread.h>
144 #include <sys/errno.h>
145 #include <sys/buf.h>
146 #include <sys/bufq.h>
147 #include <sys/malloc.h>
148 #include <sys/ioctl.h>
149 #include <sys/disklabel.h>
150 #include <sys/device.h>
151 #include <sys/disk.h>
152 #include <sys/stat.h>
153 #include <sys/mount.h>
154 #include <sys/vnode.h>
155 #include <sys/file.h>
156 #include <sys/uio.h>
157 #include <sys/conf.h>
158 #include <sys/kauth.h>
159 
160 #include <net/zlib.h>
161 
162 #include <miscfs/genfs/genfs.h>
163 #include <miscfs/specfs/specdev.h>
164 
165 #include <dev/dkvar.h>
166 #include <dev/vndvar.h>
167 
168 #include <prop/proplib.h>
169 
170 #if defined(VNDDEBUG) && !defined(DEBUG)
171 #define DEBUG
172 #endif
173 
174 #ifdef DEBUG
175 int dovndcluster = 1;
176 #define VDB_FOLLOW	0x01
177 #define VDB_INIT	0x02
178 #define VDB_IO		0x04
179 #define VDB_LABEL	0x08
180 int vnddebug = 0x00;
181 #endif
182 
183 #define vndunit(x)	DISKUNIT(x)
184 
185 struct vndxfer {
186 	struct buf vx_buf;
187 	struct vnd_softc *vx_vnd;
188 };
189 #define	VND_BUFTOXFER(bp)	((struct vndxfer *)(void *)bp)
190 
191 #define VND_GETXFER(vnd)	pool_get(&(vnd)->sc_vxpool, PR_WAITOK)
192 #define VND_PUTXFER(vnd, vx)	pool_put(&(vnd)->sc_vxpool, (vx))
193 
194 #define VNDLABELDEV(dev) \
195     (MAKEDISKDEV(major((dev)), vndunit((dev)), RAW_PART))
196 
197 /* called by main() at boot time */
198 void	vndattach(int);
199 
200 static void	vndclear(struct vnd_softc *, int);
201 static int	vnddoclear(struct vnd_softc *, int, int, bool);
202 static int	vndsetcred(struct vnd_softc *, kauth_cred_t);
203 static void	vndthrottle(struct vnd_softc *, struct vnode *);
204 static void	vndiodone(struct buf *);
205 #if 0
206 static void	vndshutdown(void);
207 #endif
208 
209 static void	vndgetdefaultlabel(struct vnd_softc *, struct disklabel *);
210 static void	vndgetdisklabel(dev_t, struct vnd_softc *);
211 
212 static int	vndlock(struct vnd_softc *);
213 static void	vndunlock(struct vnd_softc *);
214 #ifdef VND_COMPRESSION
215 static void	compstrategy(struct buf *, off_t);
216 static void	*vnd_alloc(void *, u_int, u_int);
217 static void	vnd_free(void *, void *);
218 #endif /* VND_COMPRESSION */
219 
220 static void	vndthread(void *);
221 static bool	vnode_has_op(const struct vnode *, int);
222 static void	handle_with_rdwr(struct vnd_softc *, const struct buf *,
223 		    struct buf *);
224 static void	handle_with_strategy(struct vnd_softc *, const struct buf *,
225 		    struct buf *);
226 static void	vnd_set_properties(struct vnd_softc *);
227 
228 static dev_type_open(vndopen);
229 static dev_type_close(vndclose);
230 static dev_type_read(vndread);
231 static dev_type_write(vndwrite);
232 static dev_type_ioctl(vndioctl);
233 static dev_type_strategy(vndstrategy);
234 static dev_type_dump(vnddump);
235 static dev_type_size(vndsize);
236 
237 const struct bdevsw vnd_bdevsw = {
238 	vndopen, vndclose, vndstrategy, vndioctl, vnddump, vndsize, D_DISK
239 };
240 
241 const struct cdevsw vnd_cdevsw = {
242 	vndopen, vndclose, vndread, vndwrite, vndioctl,
243 	nostop, notty, nopoll, nommap, nokqfilter, D_DISK
244 };
245 
246 static int	vnd_match(device_t, cfdata_t, void *);
247 static void	vnd_attach(device_t, device_t, void *);
248 static int	vnd_detach(device_t, int);
249 
250 CFATTACH_DECL3_NEW(vnd, sizeof(struct vnd_softc),
251     vnd_match, vnd_attach, vnd_detach, NULL, NULL, NULL, DVF_DETACH_SHUTDOWN);
252 extern struct cfdriver vnd_cd;
253 
254 static struct vnd_softc	*vnd_spawn(int);
255 int	vnd_destroy(device_t);
256 
257 void
258 vndattach(int num)
259 {
260 	int error;
261 
262 	error = config_cfattach_attach(vnd_cd.cd_name, &vnd_ca);
263 	if (error)
264 		aprint_error("%s: unable to register cfattach\n",
265 		    vnd_cd.cd_name);
266 }
267 
268 static int
269 vnd_match(device_t self, cfdata_t cfdata, void *aux)
270 {
271 
272 	return 1;
273 }
274 
275 static void
276 vnd_attach(device_t parent, device_t self, void *aux)
277 {
278 	struct vnd_softc *sc = device_private(self);
279 
280 	sc->sc_dev = self;
281 	sc->sc_comp_offsets = NULL;
282 	sc->sc_comp_buff = NULL;
283 	sc->sc_comp_decombuf = NULL;
284 	bufq_alloc(&sc->sc_tab, "disksort", BUFQ_SORT_RAWBLOCK);
285 	disk_init(&sc->sc_dkdev, device_xname(self), NULL);
286 	if (!pmf_device_register(self, NULL, NULL))
287 		aprint_error_dev(self, "couldn't establish power handler\n");
288 }
289 
290 static int
291 vnd_detach(device_t self, int flags)
292 {
293 	int error;
294 	struct vnd_softc *sc = device_private(self);
295 
296 	if (sc->sc_flags & VNF_INITED) {
297 		error = vnddoclear(sc, 0, -1, (flags & DETACH_FORCE) != 0);
298 		if (error != 0)
299 			return error;
300 	}
301 
302 	pmf_device_deregister(self);
303 	bufq_free(sc->sc_tab);
304 	disk_destroy(&sc->sc_dkdev);
305 
306 	return 0;
307 }
308 
309 static struct vnd_softc *
310 vnd_spawn(int unit)
311 {
312 	cfdata_t cf;
313 
314 	cf = malloc(sizeof(*cf), M_DEVBUF, M_WAITOK);
315 	cf->cf_name = vnd_cd.cd_name;
316 	cf->cf_atname = vnd_cd.cd_name;
317 	cf->cf_unit = unit;
318 	cf->cf_fstate = FSTATE_STAR;
319 
320 	return device_private(config_attach_pseudo(cf));
321 }
322 
323 int
324 vnd_destroy(device_t dev)
325 {
326 	int error;
327 	cfdata_t cf;
328 
329 	cf = device_cfdata(dev);
330 	error = config_detach(dev, DETACH_QUIET);
331 	if (error)
332 		return error;
333 	free(cf, M_DEVBUF);
334 	return 0;
335 }
336 
337 static int
338 vndopen(dev_t dev, int flags, int mode, struct lwp *l)
339 {
340 	int unit = vndunit(dev);
341 	struct vnd_softc *sc;
342 	int error = 0, part, pmask;
343 	struct disklabel *lp;
344 
345 #ifdef DEBUG
346 	if (vnddebug & VDB_FOLLOW)
347 		printf("vndopen(0x%"PRIx64", 0x%x, 0x%x, %p)\n", dev, flags, mode, l);
348 #endif
349 	sc = device_lookup_private(&vnd_cd, unit);
350 	if (sc == NULL) {
351 		sc = vnd_spawn(unit);
352 		if (sc == NULL)
353 			return ENOMEM;
354 	}
355 
356 	if ((error = vndlock(sc)) != 0)
357 		return error;
358 
359 	if ((sc->sc_flags & VNF_CLEARING) != 0) {
360 		error = ENXIO;
361 		goto done;
362 	}
363 
364 	lp = sc->sc_dkdev.dk_label;
365 
366 	part = DISKPART(dev);
367 	pmask = (1 << part);
368 
369 	/*
370 	 * If we're initialized, check to see if there are any other
371 	 * open partitions.  If not, then it's safe to update the
372 	 * in-core disklabel.  Only read the disklabel if it is
373 	 * not already valid.
374 	 */
375 	if ((sc->sc_flags & (VNF_INITED|VNF_VLABEL)) == VNF_INITED &&
376 	    sc->sc_dkdev.dk_openmask == 0)
377 		vndgetdisklabel(dev, sc);
378 
379 	/* Check that the partitions exists. */
380 	if (part != RAW_PART) {
381 		if (((sc->sc_flags & VNF_INITED) == 0) ||
382 		    ((part >= lp->d_npartitions) ||
383 		     (lp->d_partitions[part].p_fstype == FS_UNUSED))) {
384 			error = ENXIO;
385 			goto done;
386 		}
387 	}
388 
389 	/* Prevent our unit from being unconfigured while open. */
390 	switch (mode) {
391 	case S_IFCHR:
392 		sc->sc_dkdev.dk_copenmask |= pmask;
393 		break;
394 
395 	case S_IFBLK:
396 		sc->sc_dkdev.dk_bopenmask |= pmask;
397 		break;
398 	}
399 	sc->sc_dkdev.dk_openmask =
400 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
401 
402  done:
403 	vndunlock(sc);
404 	return error;
405 }
406 
407 static int
408 vndclose(dev_t dev, int flags, int mode, struct lwp *l)
409 {
410 	int unit = vndunit(dev);
411 	struct vnd_softc *sc;
412 	int error = 0, part;
413 
414 #ifdef DEBUG
415 	if (vnddebug & VDB_FOLLOW)
416 		printf("vndclose(0x%"PRIx64", 0x%x, 0x%x, %p)\n", dev, flags, mode, l);
417 #endif
418 	sc = device_lookup_private(&vnd_cd, unit);
419 	if (sc == NULL)
420 		return ENXIO;
421 
422 	if ((error = vndlock(sc)) != 0)
423 		return error;
424 
425 	part = DISKPART(dev);
426 
427 	/* ...that much closer to allowing unconfiguration... */
428 	switch (mode) {
429 	case S_IFCHR:
430 		sc->sc_dkdev.dk_copenmask &= ~(1 << part);
431 		break;
432 
433 	case S_IFBLK:
434 		sc->sc_dkdev.dk_bopenmask &= ~(1 << part);
435 		break;
436 	}
437 	sc->sc_dkdev.dk_openmask =
438 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
439 
440 	vndunlock(sc);
441 
442 	if ((sc->sc_flags & VNF_INITED) == 0) {
443 		if ((error = vnd_destroy(sc->sc_dev)) != 0) {
444 			aprint_error_dev(sc->sc_dev,
445 			    "unable to detach instance\n");
446 			return error;
447 		}
448 	}
449 
450 	return 0;
451 }
452 
453 /*
454  * Queue the request, and wakeup the kernel thread to handle it.
455  */
456 static void
457 vndstrategy(struct buf *bp)
458 {
459 	int unit = vndunit(bp->b_dev);
460 	struct vnd_softc *vnd =
461 	    device_lookup_private(&vnd_cd, unit);
462 	struct disklabel *lp;
463 	daddr_t blkno;
464 	int s = splbio();
465 
466 	if (vnd == NULL) {
467 		bp->b_error = ENXIO;
468 		goto done;
469 	}
470 	lp = vnd->sc_dkdev.dk_label;
471 
472 	if ((vnd->sc_flags & VNF_INITED) == 0) {
473 		bp->b_error = ENXIO;
474 		goto done;
475 	}
476 
477 	/*
478 	 * The transfer must be a whole number of blocks.
479 	 */
480 	if ((bp->b_bcount % lp->d_secsize) != 0) {
481 		bp->b_error = EINVAL;
482 		goto done;
483 	}
484 
485 	/*
486 	 * check if we're read-only.
487 	 */
488 	if ((vnd->sc_flags & VNF_READONLY) && !(bp->b_flags & B_READ)) {
489 		bp->b_error = EACCES;
490 		goto done;
491 	}
492 
493 	/* If it's a nil transfer, wake up the top half now. */
494 	if (bp->b_bcount == 0) {
495 		goto done;
496 	}
497 
498 	/*
499 	 * Do bounds checking and adjust transfer.  If there's an error,
500 	 * the bounds check will flag that for us.
501 	 */
502 	if (DISKPART(bp->b_dev) == RAW_PART) {
503 		if (bounds_check_with_mediasize(bp, DEV_BSIZE,
504 		    vnd->sc_size) <= 0)
505 			goto done;
506 	} else {
507 		if (bounds_check_with_label(&vnd->sc_dkdev,
508 		    bp, vnd->sc_flags & (VNF_WLABEL|VNF_LABELLING)) <= 0)
509 			goto done;
510 	}
511 
512 	/*
513 	 * Put the block number in terms of the logical blocksize
514 	 * of the "device".
515 	 */
516 
517 	blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
518 
519 	/*
520 	 * Translate the partition-relative block number to an absolute.
521 	 */
522 	if (DISKPART(bp->b_dev) != RAW_PART) {
523 		struct partition *pp;
524 
525 		pp = &vnd->sc_dkdev.dk_label->d_partitions[
526 		    DISKPART(bp->b_dev)];
527 		blkno += pp->p_offset;
528 	}
529 	bp->b_rawblkno = blkno;
530 
531 #ifdef DEBUG
532 	if (vnddebug & VDB_FOLLOW)
533 		printf("vndstrategy(%p): unit %d\n", bp, unit);
534 #endif
535 	bufq_put(vnd->sc_tab, bp);
536 	wakeup(&vnd->sc_tab);
537 	splx(s);
538 	return;
539 
540 done:
541 	bp->b_resid = bp->b_bcount;
542 	biodone(bp);
543 	splx(s);
544 }
545 
546 static bool
547 vnode_has_strategy(struct vnd_softc *vnd)
548 {
549 	return vnode_has_op(vnd->sc_vp, VOFFSET(vop_bmap)) &&
550 	    vnode_has_op(vnd->sc_vp, VOFFSET(vop_strategy));
551 }
552 
553 /* XXX this function needs a reliable check to detect
554  * sparse files. Otherwise, bmap/strategy may be used
555  * and fail on non-allocated blocks. VOP_READ/VOP_WRITE
556  * works on sparse files.
557  */
558 #if notyet
559 static bool
560 vnode_strategy_probe(struct vnd_softc *vnd)
561 {
562 	int error;
563 	daddr_t nbn;
564 
565 	if (!vnode_has_strategy(vnd))
566 		return false;
567 
568 	/* Convert the first logical block number to its
569 	 * physical block number.
570 	 */
571 	error = 0;
572 	vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
573 	error = VOP_BMAP(vnd->sc_vp, 0, NULL, &nbn, NULL);
574 	VOP_UNLOCK(vnd->sc_vp, 0);
575 
576 	/* Test if that worked. */
577 	if (error == 0 && (long)nbn == -1)
578 		return false;
579 
580 	return true;
581 }
582 #endif
583 
584 static void
585 vndthread(void *arg)
586 {
587 	struct vnd_softc *vnd = arg;
588 	bool usestrategy;
589 	int s;
590 
591 	/* Determine whether we can *use* VOP_BMAP and VOP_STRATEGY to
592 	 * directly access the backing vnode.  If we can, use these two
593 	 * operations to avoid messing with the local buffer cache.
594 	 * Otherwise fall back to regular VOP_READ/VOP_WRITE operations
595 	 * which are guaranteed to work with any file system. */
596 	usestrategy = vnode_has_strategy(vnd);
597 
598 #ifdef DEBUG
599 	if (vnddebug & VDB_INIT)
600 		printf("vndthread: vp %p, %s\n", vnd->sc_vp,
601 		    usestrategy ?
602 		    "using bmap/strategy operations" :
603 		    "using read/write operations");
604 #endif
605 
606 	s = splbio();
607 	vnd->sc_flags |= VNF_KTHREAD;
608 	wakeup(&vnd->sc_kthread);
609 
610 	/*
611 	 * Dequeue requests and serve them depending on the available
612 	 * vnode operations.
613 	 */
614 	while ((vnd->sc_flags & VNF_VUNCONF) == 0) {
615 		struct vndxfer *vnx;
616 		int flags;
617 		struct buf *obp;
618 		struct buf *bp;
619 
620 		obp = bufq_get(vnd->sc_tab);
621 		if (obp == NULL) {
622 			tsleep(&vnd->sc_tab, PRIBIO, "vndbp", 0);
623 			continue;
624 		};
625 		splx(s);
626 		flags = obp->b_flags;
627 #ifdef DEBUG
628 		if (vnddebug & VDB_FOLLOW)
629 			printf("vndthread(%p)\n", obp);
630 #endif
631 
632 		if (vnd->sc_vp->v_mount == NULL) {
633 			obp->b_error = ENXIO;
634 			goto done;
635 		}
636 #ifdef VND_COMPRESSION
637 		/* handle a compressed read */
638 		if ((flags & B_READ) != 0 && (vnd->sc_flags & VNF_COMP)) {
639 			off_t bn;
640 
641 			/* Convert to a byte offset within the file. */
642 			bn = obp->b_rawblkno *
643 			    vnd->sc_dkdev.dk_label->d_secsize;
644 
645 			compstrategy(obp, bn);
646 			goto done;
647 		}
648 #endif /* VND_COMPRESSION */
649 
650 		/*
651 		 * Allocate a header for this transfer and link it to the
652 		 * buffer
653 		 */
654 		s = splbio();
655 		vnx = VND_GETXFER(vnd);
656 		splx(s);
657 		vnx->vx_vnd = vnd;
658 
659 		s = splbio();
660 		while (vnd->sc_active >= vnd->sc_maxactive) {
661 			tsleep(&vnd->sc_tab, PRIBIO, "vndac", 0);
662 		}
663 		vnd->sc_active++;
664 		splx(s);
665 
666 		/* Instrumentation. */
667 		disk_busy(&vnd->sc_dkdev);
668 
669 		bp = &vnx->vx_buf;
670 		buf_init(bp);
671 		bp->b_flags = (obp->b_flags & B_READ);
672 		bp->b_oflags = obp->b_oflags;
673 		bp->b_cflags = obp->b_cflags;
674 		bp->b_iodone = vndiodone;
675 		bp->b_private = obp;
676 		bp->b_vp = vnd->sc_vp;
677 		bp->b_objlock = &bp->b_vp->v_interlock;
678 		bp->b_data = obp->b_data;
679 		bp->b_bcount = obp->b_bcount;
680 		BIO_COPYPRIO(bp, obp);
681 
682 		/* Handle the request using the appropriate operations. */
683 		if (usestrategy)
684 			handle_with_strategy(vnd, obp, bp);
685 		else
686 			handle_with_rdwr(vnd, obp, bp);
687 
688 		s = splbio();
689 		continue;
690 
691 done:
692 		biodone(obp);
693 		s = splbio();
694 	}
695 
696 	vnd->sc_flags &= (~VNF_KTHREAD | VNF_VUNCONF);
697 	wakeup(&vnd->sc_kthread);
698 	splx(s);
699 	kthread_exit(0);
700 }
701 
702 /*
703  * Checks if the given vnode supports the requested operation.
704  * The operation is specified the offset returned by VOFFSET.
705  *
706  * XXX The test below used to determine this is quite fragile
707  * because it relies on the file system to use genfs to specify
708  * unimplemented operations.  There might be another way to do
709  * it more cleanly.
710  */
711 static bool
712 vnode_has_op(const struct vnode *vp, int opoffset)
713 {
714 	int (*defaultp)(void *);
715 	int (*opp)(void *);
716 
717 	defaultp = vp->v_op[VOFFSET(vop_default)];
718 	opp = vp->v_op[opoffset];
719 
720 	return opp != defaultp && opp != genfs_eopnotsupp &&
721 	    opp != genfs_badop && opp != genfs_nullop;
722 }
723 
724 /*
725  * Handes the read/write request given in 'bp' using the vnode's VOP_READ
726  * and VOP_WRITE operations.
727  *
728  * 'obp' is a pointer to the original request fed to the vnd device.
729  */
730 static void
731 handle_with_rdwr(struct vnd_softc *vnd, const struct buf *obp, struct buf *bp)
732 {
733 	bool doread;
734 	off_t offset;
735 	size_t resid;
736 	struct vnode *vp;
737 
738 	doread = bp->b_flags & B_READ;
739 	offset = obp->b_rawblkno * vnd->sc_dkdev.dk_label->d_secsize;
740 	vp = vnd->sc_vp;
741 
742 #if defined(DEBUG)
743 	if (vnddebug & VDB_IO)
744 		printf("vnd (rdwr): vp %p, %s, rawblkno 0x%" PRIx64
745 		    ", secsize %d, offset %" PRIu64
746 		    ", bcount %d\n",
747 		    vp, doread ? "read" : "write", obp->b_rawblkno,
748 		    vnd->sc_dkdev.dk_label->d_secsize, offset,
749 		    bp->b_bcount);
750 #endif
751 
752 	/* Issue the read or write operation. */
753 	bp->b_error =
754 	    vn_rdwr(doread ? UIO_READ : UIO_WRITE,
755 	    vp, bp->b_data, bp->b_bcount, offset,
756 	    UIO_SYSSPACE, 0, vnd->sc_cred, &resid, NULL);
757 	bp->b_resid = resid;
758 
759 	/* We need to increase the number of outputs on the vnode if
760 	 * there was any write to it. */
761 	if (!doread) {
762 		mutex_enter(&vp->v_interlock);
763 		vp->v_numoutput++;
764 		mutex_exit(&vp->v_interlock);
765 	}
766 
767 	biodone(bp);
768 }
769 
770 /*
771  * Handes the read/write request given in 'bp' using the vnode's VOP_BMAP
772  * and VOP_STRATEGY operations.
773  *
774  * 'obp' is a pointer to the original request fed to the vnd device.
775  */
776 static void
777 handle_with_strategy(struct vnd_softc *vnd, const struct buf *obp,
778     struct buf *bp)
779 {
780 	int bsize, error, flags, skipped;
781 	size_t resid, sz;
782 	off_t bn, offset;
783 	struct vnode *vp;
784 
785 	flags = obp->b_flags;
786 
787 	if (!(flags & B_READ)) {
788 		vp = bp->b_vp;
789 		mutex_enter(&vp->v_interlock);
790 		vp->v_numoutput++;
791 		mutex_exit(&vp->v_interlock);
792 	}
793 
794 	/* convert to a byte offset within the file. */
795 	bn = obp->b_rawblkno * vnd->sc_dkdev.dk_label->d_secsize;
796 
797 	bsize = vnd->sc_vp->v_mount->mnt_stat.f_iosize;
798 	skipped = 0;
799 
800 	/*
801 	 * Break the request into bsize pieces and feed them
802 	 * sequentially using VOP_BMAP/VOP_STRATEGY.
803 	 * We do it this way to keep from flooding NFS servers if we
804 	 * are connected to an NFS file.  This places the burden on
805 	 * the client rather than the server.
806 	 */
807 	error = 0;
808 	bp->b_resid = bp->b_bcount;
809 	for (offset = 0, resid = bp->b_resid; resid;
810 	    resid -= sz, offset += sz) {
811 		struct buf *nbp;
812 		daddr_t nbn;
813 		int off, nra;
814 
815 		nra = 0;
816 		vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
817 		error = VOP_BMAP(vnd->sc_vp, bn / bsize, &vp, &nbn, &nra);
818 		VOP_UNLOCK(vnd->sc_vp, 0);
819 
820 		if (error == 0 && (long)nbn == -1)
821 			error = EIO;
822 
823 		/*
824 		 * If there was an error or a hole in the file...punt.
825 		 * Note that we may have to wait for any operations
826 		 * that we have already fired off before releasing
827 		 * the buffer.
828 		 *
829 		 * XXX we could deal with holes here but it would be
830 		 * a hassle (in the write case).
831 		 */
832 		if (error) {
833 			skipped += resid;
834 			break;
835 		}
836 
837 #ifdef DEBUG
838 		if (!dovndcluster)
839 			nra = 0;
840 #endif
841 
842 		off = bn % bsize;
843 		sz = MIN(((off_t)1 + nra) * bsize - off, resid);
844 #ifdef	DEBUG
845 		if (vnddebug & VDB_IO)
846 			printf("vndstrategy: vp %p/%p bn 0x%qx/0x%" PRIx64
847 			    " sz 0x%zx\n", vnd->sc_vp, vp, (long long)bn,
848 			    nbn, sz);
849 #endif
850 
851 		nbp = getiobuf(vp, true);
852 		nestiobuf_setup(bp, nbp, offset, sz);
853 		nbp->b_blkno = nbn + btodb(off);
854 
855 #if 0 /* XXX #ifdef DEBUG */
856 		if (vnddebug & VDB_IO)
857 			printf("vndstart(%ld): bp %p vp %p blkno "
858 			    "0x%" PRIx64 " flags %x addr %p cnt 0x%x\n",
859 			    (long) (vnd-vnd_softc), &nbp->vb_buf,
860 			    nbp->vb_buf.b_vp, nbp->vb_buf.b_blkno,
861 			    nbp->vb_buf.b_flags, nbp->vb_buf.b_data,
862 			    nbp->vb_buf.b_bcount);
863 #endif
864 		VOP_STRATEGY(vp, nbp);
865 		bn += sz;
866 	}
867 	nestiobuf_done(bp, skipped, error);
868 }
869 
870 static void
871 vndiodone(struct buf *bp)
872 {
873 	struct vndxfer *vnx = VND_BUFTOXFER(bp);
874 	struct vnd_softc *vnd = vnx->vx_vnd;
875 	struct buf *obp = bp->b_private;
876 	int s = splbio();
877 
878 	KASSERT(&vnx->vx_buf == bp);
879 	KASSERT(vnd->sc_active > 0);
880 #ifdef DEBUG
881 	if (vnddebug & VDB_IO) {
882 		printf("vndiodone1: bp %p iodone: error %d\n",
883 		    bp, bp->b_error);
884 	}
885 #endif
886 	disk_unbusy(&vnd->sc_dkdev, bp->b_bcount - bp->b_resid,
887 	    (bp->b_flags & B_READ));
888 	vnd->sc_active--;
889 	if (vnd->sc_active == 0) {
890 		wakeup(&vnd->sc_tab);
891 	}
892 	splx(s);
893 	obp->b_error = bp->b_error;
894 	obp->b_resid = bp->b_resid;
895 	buf_destroy(bp);
896 	VND_PUTXFER(vnd, vnx);
897 	biodone(obp);
898 }
899 
900 /* ARGSUSED */
901 static int
902 vndread(dev_t dev, struct uio *uio, int flags)
903 {
904 	int unit = vndunit(dev);
905 	struct vnd_softc *sc;
906 
907 #ifdef DEBUG
908 	if (vnddebug & VDB_FOLLOW)
909 		printf("vndread(0x%"PRIx64", %p)\n", dev, uio);
910 #endif
911 
912 	sc = device_lookup_private(&vnd_cd, unit);
913 	if (sc == NULL)
914 		return ENXIO;
915 
916 	if ((sc->sc_flags & VNF_INITED) == 0)
917 		return ENXIO;
918 
919 	return physio(vndstrategy, NULL, dev, B_READ, minphys, uio);
920 }
921 
922 /* ARGSUSED */
923 static int
924 vndwrite(dev_t dev, struct uio *uio, int flags)
925 {
926 	int unit = vndunit(dev);
927 	struct vnd_softc *sc;
928 
929 #ifdef DEBUG
930 	if (vnddebug & VDB_FOLLOW)
931 		printf("vndwrite(0x%"PRIx64", %p)\n", dev, uio);
932 #endif
933 
934 	sc = device_lookup_private(&vnd_cd, unit);
935 	if (sc == NULL)
936 		return ENXIO;
937 
938 	if ((sc->sc_flags & VNF_INITED) == 0)
939 		return ENXIO;
940 
941 	return physio(vndstrategy, NULL, dev, B_WRITE, minphys, uio);
942 }
943 
944 static int
945 vnd_cget(struct lwp *l, int unit, int *un, struct vattr *va)
946 {
947 	struct vnd_softc *vnd;
948 
949 	if (*un == -1)
950 		*un = unit;
951 	if (*un < 0)
952 		return EINVAL;
953 
954 	vnd = device_lookup_private(&vnd_cd, *un);
955 	if (vnd == NULL)
956 		return (*un >= vnd_cd.cd_ndevs) ? ENXIO : -1;
957 
958 	if ((vnd->sc_flags & VNF_INITED) == 0)
959 		return -1;
960 
961 	return VOP_GETATTR(vnd->sc_vp, va, l->l_cred);
962 }
963 
964 static int
965 vnddoclear(struct vnd_softc *vnd, int pmask, int minor, bool force)
966 {
967 	int error;
968 
969 	if ((error = vndlock(vnd)) != 0)
970 		return error;
971 
972 	/*
973 	 * Don't unconfigure if any other partitions are open
974 	 * or if both the character and block flavors of this
975 	 * partition are open.
976 	 */
977 	if (DK_BUSY(vnd, pmask) && !force) {
978 		vndunlock(vnd);
979 		return EBUSY;
980 	}
981 
982 	/*
983 	 * XXX vndclear() might call vndclose() implicitly;
984 	 * release lock to avoid recursion
985 	 *
986 	 * Set VNF_CLEARING to prevent vndopen() from
987 	 * sneaking in after we vndunlock().
988 	 */
989 	vnd->sc_flags |= VNF_CLEARING;
990 	vndunlock(vnd);
991 	vndclear(vnd, minor);
992 #ifdef DEBUG
993 	if (vnddebug & VDB_INIT)
994 		printf("vndioctl: CLRed\n");
995 #endif
996 
997 	/* Destroy the xfer and buffer pools. */
998 	pool_destroy(&vnd->sc_vxpool);
999 
1000 	/* Detach the disk. */
1001 	disk_detach(&vnd->sc_dkdev);
1002 
1003 	return 0;
1004 }
1005 
1006 /* ARGSUSED */
1007 static int
1008 vndioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l)
1009 {
1010 	bool force;
1011 	int unit = vndunit(dev);
1012 	struct vnd_softc *vnd;
1013 	struct vnd_ioctl *vio;
1014 	struct vattr vattr;
1015 	struct nameidata nd;
1016 	int error, part, pmask;
1017 	size_t geomsize;
1018 	int fflags;
1019 #ifdef __HAVE_OLD_DISKLABEL
1020 	struct disklabel newlabel;
1021 #endif
1022 
1023 #ifdef DEBUG
1024 	if (vnddebug & VDB_FOLLOW)
1025 		printf("vndioctl(0x%"PRIx64", 0x%lx, %p, 0x%x, %p): unit %d\n",
1026 		    dev, cmd, data, flag, l->l_proc, unit);
1027 #endif
1028 	vnd = device_lookup_private(&vnd_cd, unit);
1029 	if (vnd == NULL &&
1030 #ifdef COMPAT_30
1031 	    cmd != VNDIOOCGET &&
1032 #endif
1033 	    cmd != VNDIOCGET)
1034 		return ENXIO;
1035 	vio = (struct vnd_ioctl *)data;
1036 
1037 	/* Must be open for writes for these commands... */
1038 	switch (cmd) {
1039 	case VNDIOCSET:
1040 	case VNDIOCCLR:
1041 #ifdef VNDIOOCSET
1042 	case VNDIOOCSET:
1043 	case VNDIOOCCLR:
1044 #endif
1045 	case DIOCSDINFO:
1046 	case DIOCWDINFO:
1047 #ifdef __HAVE_OLD_DISKLABEL
1048 	case ODIOCSDINFO:
1049 	case ODIOCWDINFO:
1050 #endif
1051 	case DIOCKLABEL:
1052 	case DIOCWLABEL:
1053 		if ((flag & FWRITE) == 0)
1054 			return EBADF;
1055 	}
1056 
1057 	/* Must be initialized for these... */
1058 	switch (cmd) {
1059 	case VNDIOCCLR:
1060 #ifdef VNDIOOCCLR
1061 	case VNDIOOCCLR:
1062 #endif
1063 	case DIOCGDINFO:
1064 	case DIOCSDINFO:
1065 	case DIOCWDINFO:
1066 	case DIOCGPART:
1067 	case DIOCKLABEL:
1068 	case DIOCWLABEL:
1069 	case DIOCGDEFLABEL:
1070 	case DIOCCACHESYNC:
1071 #ifdef __HAVE_OLD_DISKLABEL
1072 	case ODIOCGDINFO:
1073 	case ODIOCSDINFO:
1074 	case ODIOCWDINFO:
1075 	case ODIOCGDEFLABEL:
1076 #endif
1077 		if ((vnd->sc_flags & VNF_INITED) == 0)
1078 			return ENXIO;
1079 	}
1080 
1081 	switch (cmd) {
1082 #ifdef VNDIOOCSET
1083 	case VNDIOOCSET:
1084 #endif
1085 	case VNDIOCSET:
1086 		if (vnd->sc_flags & VNF_INITED)
1087 			return EBUSY;
1088 
1089 		if ((error = vndlock(vnd)) != 0)
1090 			return error;
1091 
1092 		fflags = FREAD;
1093 		if ((vio->vnd_flags & VNDIOF_READONLY) == 0)
1094 			fflags |= FWRITE;
1095 		NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, vio->vnd_file);
1096 		if ((error = vn_open(&nd, fflags, 0)) != 0)
1097 			goto unlock_and_exit;
1098 		KASSERT(l);
1099 		error = VOP_GETATTR(nd.ni_vp, &vattr, l->l_cred);
1100 		if (!error && nd.ni_vp->v_type != VREG)
1101 			error = EOPNOTSUPP;
1102 		if (!error && vattr.va_bytes < vattr.va_size)
1103 			/* File is definitely sparse, reject here */
1104 			error = EINVAL;
1105 		if (error) {
1106 			VOP_UNLOCK(nd.ni_vp, 0);
1107 			goto close_and_exit;
1108 		}
1109 
1110 		/* If using a compressed file, initialize its info */
1111 		/* (or abort with an error if kernel has no compression) */
1112 		if (vio->vnd_flags & VNF_COMP) {
1113 #ifdef VND_COMPRESSION
1114 			struct vnd_comp_header *ch;
1115 			int i;
1116 			u_int32_t comp_size;
1117 			u_int32_t comp_maxsize;
1118 
1119 			/* allocate space for compresed file header */
1120 			ch = malloc(sizeof(struct vnd_comp_header),
1121 			M_TEMP, M_WAITOK);
1122 
1123 			/* read compressed file header */
1124 			error = vn_rdwr(UIO_READ, nd.ni_vp, (void *)ch,
1125 			  sizeof(struct vnd_comp_header), 0, UIO_SYSSPACE,
1126 			  IO_UNIT|IO_NODELOCKED, l->l_cred, NULL, NULL);
1127 			if (error) {
1128 				free(ch, M_TEMP);
1129 				VOP_UNLOCK(nd.ni_vp, 0);
1130 				goto close_and_exit;
1131 			}
1132 
1133 			/* save some header info */
1134 			vnd->sc_comp_blksz = ntohl(ch->block_size);
1135 			/* note last offset is the file byte size */
1136 			vnd->sc_comp_numoffs = ntohl(ch->num_blocks)+1;
1137 			free(ch, M_TEMP);
1138 			if (vnd->sc_comp_blksz == 0 ||
1139 			    vnd->sc_comp_blksz % DEV_BSIZE !=0) {
1140 				VOP_UNLOCK(nd.ni_vp, 0);
1141 				error = EINVAL;
1142 				goto close_and_exit;
1143 			}
1144 			if (sizeof(struct vnd_comp_header) +
1145 			  sizeof(u_int64_t) * vnd->sc_comp_numoffs >
1146 			  vattr.va_size) {
1147 				VOP_UNLOCK(nd.ni_vp, 0);
1148 				error = EINVAL;
1149 				goto close_and_exit;
1150 			}
1151 
1152 			/* set decompressed file size */
1153 			vattr.va_size =
1154 			    ((u_quad_t)vnd->sc_comp_numoffs - 1) *
1155 			     (u_quad_t)vnd->sc_comp_blksz;
1156 
1157 			/* allocate space for all the compressed offsets */
1158 			vnd->sc_comp_offsets =
1159 			malloc(sizeof(u_int64_t) * vnd->sc_comp_numoffs,
1160 			M_DEVBUF, M_WAITOK);
1161 
1162 			/* read in the offsets */
1163 			error = vn_rdwr(UIO_READ, nd.ni_vp,
1164 			  (void *)vnd->sc_comp_offsets,
1165 			  sizeof(u_int64_t) * vnd->sc_comp_numoffs,
1166 			  sizeof(struct vnd_comp_header), UIO_SYSSPACE,
1167 			  IO_UNIT|IO_NODELOCKED, l->l_cred, NULL, NULL);
1168 			if (error) {
1169 				VOP_UNLOCK(nd.ni_vp, 0);
1170 				goto close_and_exit;
1171 			}
1172 			/*
1173 			 * find largest block size (used for allocation limit).
1174 			 * Also convert offset to native byte order.
1175 			 */
1176 			comp_maxsize = 0;
1177 			for (i = 0; i < vnd->sc_comp_numoffs - 1; i++) {
1178 				vnd->sc_comp_offsets[i] =
1179 				  be64toh(vnd->sc_comp_offsets[i]);
1180 				comp_size = be64toh(vnd->sc_comp_offsets[i + 1])
1181 				  - vnd->sc_comp_offsets[i];
1182 				if (comp_size > comp_maxsize)
1183 					comp_maxsize = comp_size;
1184 			}
1185 			vnd->sc_comp_offsets[vnd->sc_comp_numoffs - 1] =
1186 			  be64toh(vnd->sc_comp_offsets[vnd->sc_comp_numoffs - 1]);
1187 
1188 			/* create compressed data buffer */
1189 			vnd->sc_comp_buff = malloc(comp_maxsize,
1190 			  M_DEVBUF, M_WAITOK);
1191 
1192 			/* create decompressed buffer */
1193 			vnd->sc_comp_decombuf = malloc(vnd->sc_comp_blksz,
1194 			  M_DEVBUF, M_WAITOK);
1195 			vnd->sc_comp_buffblk = -1;
1196 
1197 			/* Initialize decompress stream */
1198 			memset(&vnd->sc_comp_stream, 0, sizeof(z_stream));
1199 			vnd->sc_comp_stream.zalloc = vnd_alloc;
1200 			vnd->sc_comp_stream.zfree = vnd_free;
1201 			error = inflateInit2(&vnd->sc_comp_stream, MAX_WBITS);
1202 			if (error) {
1203 				if (vnd->sc_comp_stream.msg)
1204 					printf("vnd%d: compressed file, %s\n",
1205 					  unit, vnd->sc_comp_stream.msg);
1206 				VOP_UNLOCK(nd.ni_vp, 0);
1207 				error = EINVAL;
1208 				goto close_and_exit;
1209 			}
1210 
1211 			vnd->sc_flags |= VNF_COMP | VNF_READONLY;
1212 #else /* !VND_COMPRESSION */
1213 			VOP_UNLOCK(nd.ni_vp, 0);
1214 			error = EOPNOTSUPP;
1215 			goto close_and_exit;
1216 #endif /* VND_COMPRESSION */
1217 		}
1218 
1219 		VOP_UNLOCK(nd.ni_vp, 0);
1220 		vnd->sc_vp = nd.ni_vp;
1221 		vnd->sc_size = btodb(vattr.va_size);	/* note truncation */
1222 
1223 		/*
1224 		 * Use pseudo-geometry specified.  If none was provided,
1225 		 * use "standard" Adaptec fictitious geometry.
1226 		 */
1227 		if (vio->vnd_flags & VNDIOF_HASGEOM) {
1228 
1229 			memcpy(&vnd->sc_geom, &vio->vnd_geom,
1230 			    sizeof(vio->vnd_geom));
1231 
1232 			/*
1233 			 * Sanity-check the sector size.
1234 			 * XXX Don't allow secsize < DEV_BSIZE.	 Should
1235 			 * XXX we?
1236 			 */
1237 			if (vnd->sc_geom.vng_secsize < DEV_BSIZE ||
1238 			    (vnd->sc_geom.vng_secsize % DEV_BSIZE) != 0 ||
1239 			    vnd->sc_geom.vng_ncylinders == 0 ||
1240 			    (vnd->sc_geom.vng_ntracks *
1241 			     vnd->sc_geom.vng_nsectors) == 0) {
1242 				error = EINVAL;
1243 				goto close_and_exit;
1244 			}
1245 
1246 			/*
1247 			 * Compute the size (in DEV_BSIZE blocks) specified
1248 			 * by the geometry.
1249 			 */
1250 			geomsize = (vnd->sc_geom.vng_nsectors *
1251 			    vnd->sc_geom.vng_ntracks *
1252 			    vnd->sc_geom.vng_ncylinders) *
1253 			    (vnd->sc_geom.vng_secsize / DEV_BSIZE);
1254 
1255 			/*
1256 			 * Sanity-check the size against the specified
1257 			 * geometry.
1258 			 */
1259 			if (vnd->sc_size < geomsize) {
1260 				error = EINVAL;
1261 				goto close_and_exit;
1262 			}
1263 		} else if (vnd->sc_size >= (32 * 64)) {
1264 			/*
1265 			 * Size must be at least 2048 DEV_BSIZE blocks
1266 			 * (1M) in order to use this geometry.
1267 			 */
1268 			vnd->sc_geom.vng_secsize = DEV_BSIZE;
1269 			vnd->sc_geom.vng_nsectors = 32;
1270 			vnd->sc_geom.vng_ntracks = 64;
1271 			vnd->sc_geom.vng_ncylinders = vnd->sc_size / (64 * 32);
1272 		} else {
1273 			vnd->sc_geom.vng_secsize = DEV_BSIZE;
1274 			vnd->sc_geom.vng_nsectors = 1;
1275 			vnd->sc_geom.vng_ntracks = 1;
1276 			vnd->sc_geom.vng_ncylinders = vnd->sc_size;
1277 		}
1278 
1279 		vnd_set_properties(vnd);
1280 
1281 		if (vio->vnd_flags & VNDIOF_READONLY) {
1282 			vnd->sc_flags |= VNF_READONLY;
1283 		}
1284 
1285 		if ((error = vndsetcred(vnd, l->l_cred)) != 0)
1286 			goto close_and_exit;
1287 
1288 		vndthrottle(vnd, vnd->sc_vp);
1289 		vio->vnd_osize = dbtob(vnd->sc_size);
1290 #ifdef VNDIOOCSET
1291 		if (cmd != VNDIOOCSET)
1292 #endif
1293 			vio->vnd_size = dbtob(vnd->sc_size);
1294 		vnd->sc_flags |= VNF_INITED;
1295 
1296 		/* create the kernel thread, wait for it to be up */
1297 		error = kthread_create(PRI_NONE, 0, NULL, vndthread, vnd,
1298 		    &vnd->sc_kthread, device_xname(vnd->sc_dev));
1299 		if (error)
1300 			goto close_and_exit;
1301 		while ((vnd->sc_flags & VNF_KTHREAD) == 0) {
1302 			tsleep(&vnd->sc_kthread, PRIBIO, "vndthr", 0);
1303 		}
1304 #ifdef DEBUG
1305 		if (vnddebug & VDB_INIT)
1306 			printf("vndioctl: SET vp %p size 0x%lx %d/%d/%d/%d\n",
1307 			    vnd->sc_vp, (unsigned long) vnd->sc_size,
1308 			    vnd->sc_geom.vng_secsize,
1309 			    vnd->sc_geom.vng_nsectors,
1310 			    vnd->sc_geom.vng_ntracks,
1311 			    vnd->sc_geom.vng_ncylinders);
1312 #endif
1313 
1314 		/* Attach the disk. */
1315 		disk_attach(&vnd->sc_dkdev);
1316 		disk_blocksize(&vnd->sc_dkdev, vnd->sc_geom.vng_secsize);
1317 
1318 		/* Initialize the xfer and buffer pools. */
1319 		pool_init(&vnd->sc_vxpool, sizeof(struct vndxfer), 0,
1320 		    0, 0, "vndxpl", NULL, IPL_BIO);
1321 
1322 		/* Try and read the disklabel. */
1323 		vndgetdisklabel(dev, vnd);
1324 
1325 		vndunlock(vnd);
1326 
1327 		break;
1328 
1329 close_and_exit:
1330 		(void) vn_close(nd.ni_vp, fflags, l->l_cred);
1331 unlock_and_exit:
1332 #ifdef VND_COMPRESSION
1333 		/* free any allocated memory (for compressed file) */
1334 		if (vnd->sc_comp_offsets) {
1335 			free(vnd->sc_comp_offsets, M_DEVBUF);
1336 			vnd->sc_comp_offsets = NULL;
1337 		}
1338 		if (vnd->sc_comp_buff) {
1339 			free(vnd->sc_comp_buff, M_DEVBUF);
1340 			vnd->sc_comp_buff = NULL;
1341 		}
1342 		if (vnd->sc_comp_decombuf) {
1343 			free(vnd->sc_comp_decombuf, M_DEVBUF);
1344 			vnd->sc_comp_decombuf = NULL;
1345 		}
1346 #endif /* VND_COMPRESSION */
1347 		vndunlock(vnd);
1348 		return error;
1349 
1350 #ifdef VNDIOOCCLR
1351 	case VNDIOOCCLR:
1352 #endif
1353 	case VNDIOCCLR:
1354 		part = DISKPART(dev);
1355 		pmask = (1 << part);
1356 		force = (vio->vnd_flags & VNDIOF_FORCE) != 0;
1357 
1358 		if ((error = vnddoclear(vnd, pmask, minor(dev), force)) != 0)
1359 			return error;
1360 
1361 		break;
1362 
1363 #ifdef COMPAT_30
1364 	case VNDIOOCGET: {
1365 		struct vnd_ouser *vnu;
1366 		struct vattr va;
1367 		vnu = (struct vnd_ouser *)data;
1368 		KASSERT(l);
1369 		switch (error = vnd_cget(l, unit, &vnu->vnu_unit, &va)) {
1370 		case 0:
1371 			vnu->vnu_dev = va.va_fsid;
1372 			vnu->vnu_ino = va.va_fileid;
1373 			break;
1374 		case -1:
1375 			/* unused is not an error */
1376 			vnu->vnu_dev = 0;
1377 			vnu->vnu_ino = 0;
1378 			break;
1379 		default:
1380 			return error;
1381 		}
1382 		break;
1383 	}
1384 #endif
1385 	case VNDIOCGET: {
1386 		struct vnd_user *vnu;
1387 		struct vattr va;
1388 		vnu = (struct vnd_user *)data;
1389 		KASSERT(l);
1390 		switch (error = vnd_cget(l, unit, &vnu->vnu_unit, &va)) {
1391 		case 0:
1392 			vnu->vnu_dev = va.va_fsid;
1393 			vnu->vnu_ino = va.va_fileid;
1394 			break;
1395 		case -1:
1396 			/* unused is not an error */
1397 			vnu->vnu_dev = 0;
1398 			vnu->vnu_ino = 0;
1399 			break;
1400 		default:
1401 			return error;
1402 		}
1403 		break;
1404 	}
1405 
1406 	case DIOCGDINFO:
1407 		*(struct disklabel *)data = *(vnd->sc_dkdev.dk_label);
1408 		break;
1409 
1410 #ifdef __HAVE_OLD_DISKLABEL
1411 	case ODIOCGDINFO:
1412 		newlabel = *(vnd->sc_dkdev.dk_label);
1413 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1414 			return ENOTTY;
1415 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1416 		break;
1417 #endif
1418 
1419 	case DIOCGPART:
1420 		((struct partinfo *)data)->disklab = vnd->sc_dkdev.dk_label;
1421 		((struct partinfo *)data)->part =
1422 		    &vnd->sc_dkdev.dk_label->d_partitions[DISKPART(dev)];
1423 		break;
1424 
1425 	case DIOCWDINFO:
1426 	case DIOCSDINFO:
1427 #ifdef __HAVE_OLD_DISKLABEL
1428 	case ODIOCWDINFO:
1429 	case ODIOCSDINFO:
1430 #endif
1431 	{
1432 		struct disklabel *lp;
1433 
1434 		if ((error = vndlock(vnd)) != 0)
1435 			return error;
1436 
1437 		vnd->sc_flags |= VNF_LABELLING;
1438 
1439 #ifdef __HAVE_OLD_DISKLABEL
1440 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
1441 			memset(&newlabel, 0, sizeof newlabel);
1442 			memcpy(&newlabel, data, sizeof (struct olddisklabel));
1443 			lp = &newlabel;
1444 		} else
1445 #endif
1446 		lp = (struct disklabel *)data;
1447 
1448 		error = setdisklabel(vnd->sc_dkdev.dk_label,
1449 		    lp, 0, vnd->sc_dkdev.dk_cpulabel);
1450 		if (error == 0) {
1451 			if (cmd == DIOCWDINFO
1452 #ifdef __HAVE_OLD_DISKLABEL
1453 			    || cmd == ODIOCWDINFO
1454 #endif
1455 			   )
1456 				error = writedisklabel(VNDLABELDEV(dev),
1457 				    vndstrategy, vnd->sc_dkdev.dk_label,
1458 				    vnd->sc_dkdev.dk_cpulabel);
1459 		}
1460 
1461 		vnd->sc_flags &= ~VNF_LABELLING;
1462 
1463 		vndunlock(vnd);
1464 
1465 		if (error)
1466 			return error;
1467 		break;
1468 	}
1469 
1470 	case DIOCKLABEL:
1471 		if (*(int *)data != 0)
1472 			vnd->sc_flags |= VNF_KLABEL;
1473 		else
1474 			vnd->sc_flags &= ~VNF_KLABEL;
1475 		break;
1476 
1477 	case DIOCWLABEL:
1478 		if (*(int *)data != 0)
1479 			vnd->sc_flags |= VNF_WLABEL;
1480 		else
1481 			vnd->sc_flags &= ~VNF_WLABEL;
1482 		break;
1483 
1484 	case DIOCGDEFLABEL:
1485 		vndgetdefaultlabel(vnd, (struct disklabel *)data);
1486 		break;
1487 
1488 #ifdef __HAVE_OLD_DISKLABEL
1489 	case ODIOCGDEFLABEL:
1490 		vndgetdefaultlabel(vnd, &newlabel);
1491 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1492 			return ENOTTY;
1493 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1494 		break;
1495 #endif
1496 
1497 	case DIOCCACHESYNC:
1498 		vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1499 		error = VOP_FSYNC(vnd->sc_vp, vnd->sc_cred,
1500 		    FSYNC_WAIT | FSYNC_DATAONLY | FSYNC_CACHE, 0, 0);
1501 		VOP_UNLOCK(vnd->sc_vp, 0);
1502 		return error;
1503 
1504 	default:
1505 		return ENOTTY;
1506 	}
1507 
1508 	return 0;
1509 }
1510 
1511 /*
1512  * Duplicate the current processes' credentials.  Since we are called only
1513  * as the result of a SET ioctl and only root can do that, any future access
1514  * to this "disk" is essentially as root.  Note that credentials may change
1515  * if some other uid can write directly to the mapped file (NFS).
1516  */
1517 static int
1518 vndsetcred(struct vnd_softc *vnd, kauth_cred_t cred)
1519 {
1520 	struct uio auio;
1521 	struct iovec aiov;
1522 	char *tmpbuf;
1523 	int error;
1524 
1525 	vnd->sc_cred = kauth_cred_dup(cred);
1526 	tmpbuf = malloc(DEV_BSIZE, M_TEMP, M_WAITOK);
1527 
1528 	/* XXX: Horrible kludge to establish credentials for NFS */
1529 	aiov.iov_base = tmpbuf;
1530 	aiov.iov_len = min(DEV_BSIZE, dbtob(vnd->sc_size));
1531 	auio.uio_iov = &aiov;
1532 	auio.uio_iovcnt = 1;
1533 	auio.uio_offset = 0;
1534 	auio.uio_rw = UIO_READ;
1535 	auio.uio_resid = aiov.iov_len;
1536 	UIO_SETUP_SYSSPACE(&auio);
1537 	vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1538 	error = VOP_READ(vnd->sc_vp, &auio, 0, vnd->sc_cred);
1539 	if (error == 0) {
1540 		/*
1541 		 * Because vnd does all IO directly through the vnode
1542 		 * we need to flush (at least) the buffer from the above
1543 		 * VOP_READ from the buffer cache to prevent cache
1544 		 * incoherencies.  Also, be careful to write dirty
1545 		 * buffers back to stable storage.
1546 		 */
1547 		error = vinvalbuf(vnd->sc_vp, V_SAVE, vnd->sc_cred,
1548 			    curlwp, 0, 0);
1549 	}
1550 	VOP_UNLOCK(vnd->sc_vp, 0);
1551 
1552 	free(tmpbuf, M_TEMP);
1553 	return error;
1554 }
1555 
1556 /*
1557  * Set maxactive based on FS type
1558  */
1559 static void
1560 vndthrottle(struct vnd_softc *vnd, struct vnode *vp)
1561 {
1562 
1563 	if (vp->v_tag == VT_NFS)
1564 		vnd->sc_maxactive = 2;
1565 	else
1566 		vnd->sc_maxactive = 8;
1567 
1568 	if (vnd->sc_maxactive < 1)
1569 		vnd->sc_maxactive = 1;
1570 }
1571 
1572 #if 0
1573 static void
1574 vndshutdown(void)
1575 {
1576 	struct vnd_softc *vnd;
1577 
1578 	for (vnd = &vnd_softc[0]; vnd < &vnd_softc[numvnd]; vnd++)
1579 		if (vnd->sc_flags & VNF_INITED)
1580 			vndclear(vnd);
1581 }
1582 #endif
1583 
1584 static void
1585 vndclear(struct vnd_softc *vnd, int myminor)
1586 {
1587 	struct vnode *vp = vnd->sc_vp;
1588 	int fflags = FREAD;
1589 	int bmaj, cmaj, i, mn;
1590 	int s;
1591 
1592 #ifdef DEBUG
1593 	if (vnddebug & VDB_FOLLOW)
1594 		printf("vndclear(%p): vp %p\n", vnd, vp);
1595 #endif
1596 	/* locate the major number */
1597 	bmaj = bdevsw_lookup_major(&vnd_bdevsw);
1598 	cmaj = cdevsw_lookup_major(&vnd_cdevsw);
1599 
1600 	/* Nuke the vnodes for any open instances */
1601 	for (i = 0; i < MAXPARTITIONS; i++) {
1602 		mn = DISKMINOR(device_unit(vnd->sc_dev), i);
1603 		vdevgone(bmaj, mn, mn, VBLK);
1604 		if (mn != myminor) /* XXX avoid to kill own vnode */
1605 			vdevgone(cmaj, mn, mn, VCHR);
1606 	}
1607 
1608 	if ((vnd->sc_flags & VNF_READONLY) == 0)
1609 		fflags |= FWRITE;
1610 
1611 	s = splbio();
1612 	bufq_drain(vnd->sc_tab);
1613 	splx(s);
1614 
1615 	vnd->sc_flags |= VNF_VUNCONF;
1616 	wakeup(&vnd->sc_tab);
1617 	while (vnd->sc_flags & VNF_KTHREAD)
1618 		tsleep(&vnd->sc_kthread, PRIBIO, "vnthr", 0);
1619 
1620 #ifdef VND_COMPRESSION
1621 	/* free the compressed file buffers */
1622 	if (vnd->sc_flags & VNF_COMP) {
1623 		if (vnd->sc_comp_offsets) {
1624 			free(vnd->sc_comp_offsets, M_DEVBUF);
1625 			vnd->sc_comp_offsets = NULL;
1626 		}
1627 		if (vnd->sc_comp_buff) {
1628 			free(vnd->sc_comp_buff, M_DEVBUF);
1629 			vnd->sc_comp_buff = NULL;
1630 		}
1631 		if (vnd->sc_comp_decombuf) {
1632 			free(vnd->sc_comp_decombuf, M_DEVBUF);
1633 			vnd->sc_comp_decombuf = NULL;
1634 		}
1635 	}
1636 #endif /* VND_COMPRESSION */
1637 	vnd->sc_flags &=
1638 	    ~(VNF_INITED | VNF_READONLY | VNF_VLABEL
1639 	      | VNF_VUNCONF | VNF_COMP | VNF_CLEARING);
1640 	if (vp == NULL)
1641 		panic("vndclear: null vp");
1642 	(void) vn_close(vp, fflags, vnd->sc_cred);
1643 	kauth_cred_free(vnd->sc_cred);
1644 	vnd->sc_vp = NULL;
1645 	vnd->sc_cred = NULL;
1646 	vnd->sc_size = 0;
1647 }
1648 
1649 static int
1650 vndsize(dev_t dev)
1651 {
1652 	struct vnd_softc *sc;
1653 	struct disklabel *lp;
1654 	int part, unit, omask;
1655 	int size;
1656 
1657 	unit = vndunit(dev);
1658 	sc = device_lookup_private(&vnd_cd, unit);
1659 	if (sc == NULL)
1660 		return -1;
1661 
1662 	if ((sc->sc_flags & VNF_INITED) == 0)
1663 		return -1;
1664 
1665 	part = DISKPART(dev);
1666 	omask = sc->sc_dkdev.dk_openmask & (1 << part);
1667 	lp = sc->sc_dkdev.dk_label;
1668 
1669 	if (omask == 0 && vndopen(dev, 0, S_IFBLK, curlwp))	/* XXX */
1670 		return -1;
1671 
1672 	if (lp->d_partitions[part].p_fstype != FS_SWAP)
1673 		size = -1;
1674 	else
1675 		size = lp->d_partitions[part].p_size *
1676 		    (lp->d_secsize / DEV_BSIZE);
1677 
1678 	if (omask == 0 && vndclose(dev, 0, S_IFBLK, curlwp))	/* XXX */
1679 		return -1;
1680 
1681 	return size;
1682 }
1683 
1684 static int
1685 vnddump(dev_t dev, daddr_t blkno, void *va,
1686     size_t size)
1687 {
1688 
1689 	/* Not implemented. */
1690 	return ENXIO;
1691 }
1692 
1693 static void
1694 vndgetdefaultlabel(struct vnd_softc *sc, struct disklabel *lp)
1695 {
1696 	struct vndgeom *vng = &sc->sc_geom;
1697 	struct partition *pp;
1698 
1699 	memset(lp, 0, sizeof(*lp));
1700 
1701 	lp->d_secperunit = sc->sc_size / (vng->vng_secsize / DEV_BSIZE);
1702 	lp->d_secsize = vng->vng_secsize;
1703 	lp->d_nsectors = vng->vng_nsectors;
1704 	lp->d_ntracks = vng->vng_ntracks;
1705 	lp->d_ncylinders = vng->vng_ncylinders;
1706 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1707 
1708 	strncpy(lp->d_typename, "vnd", sizeof(lp->d_typename));
1709 	lp->d_type = DTYPE_VND;
1710 	strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
1711 	lp->d_rpm = 3600;
1712 	lp->d_interleave = 1;
1713 	lp->d_flags = 0;
1714 
1715 	pp = &lp->d_partitions[RAW_PART];
1716 	pp->p_offset = 0;
1717 	pp->p_size = lp->d_secperunit;
1718 	pp->p_fstype = FS_UNUSED;
1719 	lp->d_npartitions = RAW_PART + 1;
1720 
1721 	lp->d_magic = DISKMAGIC;
1722 	lp->d_magic2 = DISKMAGIC;
1723 	lp->d_checksum = dkcksum(lp);
1724 }
1725 
1726 /*
1727  * Read the disklabel from a vnd.  If one is not present, create a fake one.
1728  */
1729 static void
1730 vndgetdisklabel(dev_t dev, struct vnd_softc *sc)
1731 {
1732 	const char *errstring;
1733 	struct disklabel *lp = sc->sc_dkdev.dk_label;
1734 	struct cpu_disklabel *clp = sc->sc_dkdev.dk_cpulabel;
1735 	int i;
1736 
1737 	memset(clp, 0, sizeof(*clp));
1738 
1739 	vndgetdefaultlabel(sc, lp);
1740 
1741 	/*
1742 	 * Call the generic disklabel extraction routine.
1743 	 */
1744 	errstring = readdisklabel(VNDLABELDEV(dev), vndstrategy, lp, clp);
1745 	if (errstring) {
1746 		/*
1747 		 * Lack of disklabel is common, but we print the warning
1748 		 * anyway, since it might contain other useful information.
1749 		 */
1750 		aprint_normal_dev(sc->sc_dev, "%s\n", errstring);
1751 
1752 		/*
1753 		 * For historical reasons, if there's no disklabel
1754 		 * present, all partitions must be FS_BSDFFS and
1755 		 * occupy the entire disk.
1756 		 */
1757 		for (i = 0; i < MAXPARTITIONS; i++) {
1758 			/*
1759 			 * Don't wipe out port specific hack (such as
1760 			 * dos partition hack of i386 port).
1761 			 */
1762 			if (lp->d_partitions[i].p_size != 0)
1763 				continue;
1764 
1765 			lp->d_partitions[i].p_size = lp->d_secperunit;
1766 			lp->d_partitions[i].p_offset = 0;
1767 			lp->d_partitions[i].p_fstype = FS_BSDFFS;
1768 		}
1769 
1770 		strncpy(lp->d_packname, "default label",
1771 		    sizeof(lp->d_packname));
1772 
1773 		lp->d_npartitions = MAXPARTITIONS;
1774 		lp->d_checksum = dkcksum(lp);
1775 	}
1776 
1777 	/* In-core label now valid. */
1778 	sc->sc_flags |= VNF_VLABEL;
1779 }
1780 
1781 /*
1782  * Wait interruptibly for an exclusive lock.
1783  *
1784  * XXX
1785  * Several drivers do this; it should be abstracted and made MP-safe.
1786  */
1787 static int
1788 vndlock(struct vnd_softc *sc)
1789 {
1790 	int error;
1791 
1792 	while ((sc->sc_flags & VNF_LOCKED) != 0) {
1793 		sc->sc_flags |= VNF_WANTED;
1794 		if ((error = tsleep(sc, PRIBIO | PCATCH, "vndlck", 0)) != 0)
1795 			return error;
1796 	}
1797 	sc->sc_flags |= VNF_LOCKED;
1798 	return 0;
1799 }
1800 
1801 /*
1802  * Unlock and wake up any waiters.
1803  */
1804 static void
1805 vndunlock(struct vnd_softc *sc)
1806 {
1807 
1808 	sc->sc_flags &= ~VNF_LOCKED;
1809 	if ((sc->sc_flags & VNF_WANTED) != 0) {
1810 		sc->sc_flags &= ~VNF_WANTED;
1811 		wakeup(sc);
1812 	}
1813 }
1814 
1815 #ifdef VND_COMPRESSION
1816 /* compressed file read */
1817 static void
1818 compstrategy(struct buf *bp, off_t bn)
1819 {
1820 	int error;
1821 	int unit = vndunit(bp->b_dev);
1822 	struct vnd_softc *vnd =
1823 	    device_lookup_private(&vnd_cd, unit);
1824 	u_int32_t comp_block;
1825 	struct uio auio;
1826 	char *addr;
1827 	int s;
1828 
1829 	/* set up constants for data move */
1830 	auio.uio_rw = UIO_READ;
1831 	UIO_SETUP_SYSSPACE(&auio);
1832 
1833 	/* read, and transfer the data */
1834 	addr = bp->b_data;
1835 	bp->b_resid = bp->b_bcount;
1836 	s = splbio();
1837 	while (bp->b_resid > 0) {
1838 		unsigned length;
1839 		size_t length_in_buffer;
1840 		u_int32_t offset_in_buffer;
1841 		struct iovec aiov;
1842 
1843 		/* calculate the compressed block number */
1844 		comp_block = bn / (off_t)vnd->sc_comp_blksz;
1845 
1846 		/* check for good block number */
1847 		if (comp_block >= vnd->sc_comp_numoffs) {
1848 			bp->b_error = EINVAL;
1849 			splx(s);
1850 			return;
1851 		}
1852 
1853 		/* read in the compressed block, if not in buffer */
1854 		if (comp_block != vnd->sc_comp_buffblk) {
1855 			length = vnd->sc_comp_offsets[comp_block + 1] -
1856 			    vnd->sc_comp_offsets[comp_block];
1857 			vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1858 			error = vn_rdwr(UIO_READ, vnd->sc_vp, vnd->sc_comp_buff,
1859 			    length, vnd->sc_comp_offsets[comp_block],
1860 			    UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, vnd->sc_cred,
1861 			    NULL, NULL);
1862 			if (error) {
1863 				bp->b_error = error;
1864 				VOP_UNLOCK(vnd->sc_vp, 0);
1865 				splx(s);
1866 				return;
1867 			}
1868 			/* uncompress the buffer */
1869 			vnd->sc_comp_stream.next_in = vnd->sc_comp_buff;
1870 			vnd->sc_comp_stream.avail_in = length;
1871 			vnd->sc_comp_stream.next_out = vnd->sc_comp_decombuf;
1872 			vnd->sc_comp_stream.avail_out = vnd->sc_comp_blksz;
1873 			inflateReset(&vnd->sc_comp_stream);
1874 			error = inflate(&vnd->sc_comp_stream, Z_FINISH);
1875 			if (error != Z_STREAM_END) {
1876 				if (vnd->sc_comp_stream.msg)
1877 					aprint_normal_dev(vnd->sc_dev,
1878 					    "compressed file, %s\n",
1879 					    vnd->sc_comp_stream.msg);
1880 				bp->b_error = EBADMSG;
1881 				VOP_UNLOCK(vnd->sc_vp, 0);
1882 				splx(s);
1883 				return;
1884 			}
1885 			vnd->sc_comp_buffblk = comp_block;
1886 			VOP_UNLOCK(vnd->sc_vp, 0);
1887 		}
1888 
1889 		/* transfer the usable uncompressed data */
1890 		offset_in_buffer = bn % (off_t)vnd->sc_comp_blksz;
1891 		length_in_buffer = vnd->sc_comp_blksz - offset_in_buffer;
1892 		if (length_in_buffer > bp->b_resid)
1893 			length_in_buffer = bp->b_resid;
1894 		auio.uio_iov = &aiov;
1895 		auio.uio_iovcnt = 1;
1896 		aiov.iov_base = addr;
1897 		aiov.iov_len = length_in_buffer;
1898 		auio.uio_resid = aiov.iov_len;
1899 		auio.uio_offset = 0;
1900 		error = uiomove(vnd->sc_comp_decombuf + offset_in_buffer,
1901 		    length_in_buffer, &auio);
1902 		if (error) {
1903 			bp->b_error = error;
1904 			splx(s);
1905 			return;
1906 		}
1907 
1908 		bn += length_in_buffer;
1909 		addr += length_in_buffer;
1910 		bp->b_resid -= length_in_buffer;
1911 	}
1912 	splx(s);
1913 }
1914 
1915 /* compression memory allocation routines */
1916 static void *
1917 vnd_alloc(void *aux, u_int items, u_int siz)
1918 {
1919 	return malloc(items * siz, M_TEMP, M_NOWAIT);
1920 }
1921 
1922 static void
1923 vnd_free(void *aux, void *ptr)
1924 {
1925 	free(ptr, M_TEMP);
1926 }
1927 #endif /* VND_COMPRESSION */
1928 
1929 static void
1930 vnd_set_properties(struct vnd_softc *vnd)
1931 {
1932 	prop_dictionary_t disk_info, odisk_info, geom;
1933 
1934 	disk_info = prop_dictionary_create();
1935 
1936 	geom = prop_dictionary_create();
1937 
1938 	prop_dictionary_set_uint64(geom, "sectors-per-unit",
1939 	    vnd->sc_geom.vng_nsectors * vnd->sc_geom.vng_ntracks *
1940 	    vnd->sc_geom.vng_ncylinders);
1941 
1942 	prop_dictionary_set_uint32(geom, "sector-size",
1943 	    vnd->sc_geom.vng_secsize);
1944 
1945 	prop_dictionary_set_uint16(geom, "sectors-per-track",
1946 	    vnd->sc_geom.vng_nsectors);
1947 
1948 	prop_dictionary_set_uint16(geom, "tracks-per-cylinder",
1949 	    vnd->sc_geom.vng_ntracks);
1950 
1951 	prop_dictionary_set_uint64(geom, "cylinders-per-unit",
1952 	    vnd->sc_geom.vng_ncylinders);
1953 
1954 	prop_dictionary_set(disk_info, "geometry", geom);
1955 	prop_object_release(geom);
1956 
1957 	prop_dictionary_set(device_properties(vnd->sc_dev),
1958 	    "disk-info", disk_info);
1959 
1960 	/*
1961 	 * Don't release disk_info here; we keep a reference to it.
1962 	 * disk_detach() will release it when we go away.
1963 	 */
1964 
1965 	odisk_info = vnd->sc_dkdev.dk_info;
1966 	vnd->sc_dkdev.dk_info = disk_info;
1967 	if (odisk_info)
1968 		prop_object_release(odisk_info);
1969 }
1970 
1971 #ifdef _MODULE
1972 
1973 #include <sys/module.h>
1974 
1975 MODULE(MODULE_CLASS_DRIVER, vnd, NULL);
1976 CFDRIVER_DECL(vnd, DV_DISK, NULL);
1977 
1978 static int
1979 vnd_modcmd(modcmd_t cmd, void *arg)
1980 {
1981 	int bmajor = -1, cmajor = -1,  error = 0;
1982 
1983 	switch (cmd) {
1984 	case MODULE_CMD_INIT:
1985 		error = config_cfdriver_attach(&vnd_cd);
1986 		if (error)
1987 			break;
1988 
1989 		error = config_cfattach_attach(vnd_cd.cd_name, &vnd_ca);
1990 	        if (error) {
1991 			config_cfdriver_detach(&vnd_cd);
1992 			aprint_error("%s: unable to register cfattach\n",
1993 			    vnd_cd.cd_name);
1994 			break;
1995 		}
1996 
1997 		error = devsw_attach("vnd", &vnd_bdevsw, &bmajor,
1998 		    &vnd_cdevsw, &cmajor);
1999 		if (error) {
2000 			config_cfattach_detach(vnd_cd.cd_name, &vnd_ca);
2001 			config_cfdriver_detach(&vnd_cd);
2002 			break;
2003 		}
2004 
2005 		break;
2006 
2007 	case MODULE_CMD_FINI:
2008 		error = config_cfattach_detach(vnd_cd.cd_name, &vnd_ca);
2009 		if (error)
2010 			break;
2011 		config_cfdriver_detach(&vnd_cd);
2012 		devsw_detach(&vnd_bdevsw, &vnd_cdevsw);
2013 		break;
2014 
2015 	case MODULE_CMD_STAT:
2016 		return ENOTTY;
2017 
2018 	default:
2019 		return ENOTTY;
2020 	}
2021 
2022 	return error;
2023 }
2024 
2025 #endif
2026