xref: /netbsd-src/sys/dev/vnd.c (revision daf6c4152fcddc27c445489775ed1f66ab4ea9a9)
1 /*	$NetBSD: vnd.c,v 1.215 2011/02/08 20:20:26 rmind 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 1.215 2011/02/08 20:20:26 rmind 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 #include <prop/proplib.h>
131 
132 #if defined(VNDDEBUG) && !defined(DEBUG)
133 #define DEBUG
134 #endif
135 
136 #ifdef DEBUG
137 int dovndcluster = 1;
138 #define VDB_FOLLOW	0x01
139 #define VDB_INIT	0x02
140 #define VDB_IO		0x04
141 #define VDB_LABEL	0x08
142 int vnddebug = 0x00;
143 #endif
144 
145 #define vndunit(x)	DISKUNIT(x)
146 
147 struct vndxfer {
148 	struct buf vx_buf;
149 	struct vnd_softc *vx_vnd;
150 };
151 #define	VND_BUFTOXFER(bp)	((struct vndxfer *)(void *)bp)
152 
153 #define VND_GETXFER(vnd)	pool_get(&(vnd)->sc_vxpool, PR_WAITOK)
154 #define VND_PUTXFER(vnd, vx)	pool_put(&(vnd)->sc_vxpool, (vx))
155 
156 #define VNDLABELDEV(dev) \
157     (MAKEDISKDEV(major((dev)), vndunit((dev)), RAW_PART))
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_properties(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 	/*
334 	 * If we're initialized, check to see if there are any other
335 	 * open partitions.  If not, then it's safe to update the
336 	 * in-core disklabel.  Only read the disklabel if it is
337 	 * not already valid.
338 	 */
339 	if ((sc->sc_flags & (VNF_INITED|VNF_VLABEL)) == VNF_INITED &&
340 	    sc->sc_dkdev.dk_openmask == 0)
341 		vndgetdisklabel(dev, sc);
342 
343 	/* Check that the partitions exists. */
344 	if (part != RAW_PART) {
345 		if (((sc->sc_flags & VNF_INITED) == 0) ||
346 		    ((part >= lp->d_npartitions) ||
347 		     (lp->d_partitions[part].p_fstype == FS_UNUSED))) {
348 			error = ENXIO;
349 			goto done;
350 		}
351 	}
352 
353 	/* Prevent our unit from being unconfigured while open. */
354 	switch (mode) {
355 	case S_IFCHR:
356 		sc->sc_dkdev.dk_copenmask |= pmask;
357 		break;
358 
359 	case S_IFBLK:
360 		sc->sc_dkdev.dk_bopenmask |= pmask;
361 		break;
362 	}
363 	sc->sc_dkdev.dk_openmask =
364 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
365 
366  done:
367 	vndunlock(sc);
368 	return error;
369 }
370 
371 static int
372 vndclose(dev_t dev, int flags, int mode, struct lwp *l)
373 {
374 	int unit = vndunit(dev);
375 	struct vnd_softc *sc;
376 	int error = 0, part;
377 
378 #ifdef DEBUG
379 	if (vnddebug & VDB_FOLLOW)
380 		printf("vndclose(0x%"PRIx64", 0x%x, 0x%x, %p)\n", dev, flags, mode, l);
381 #endif
382 	sc = device_lookup_private(&vnd_cd, unit);
383 	if (sc == NULL)
384 		return ENXIO;
385 
386 	if ((error = vndlock(sc)) != 0)
387 		return error;
388 
389 	part = DISKPART(dev);
390 
391 	/* ...that much closer to allowing unconfiguration... */
392 	switch (mode) {
393 	case S_IFCHR:
394 		sc->sc_dkdev.dk_copenmask &= ~(1 << part);
395 		break;
396 
397 	case S_IFBLK:
398 		sc->sc_dkdev.dk_bopenmask &= ~(1 << part);
399 		break;
400 	}
401 	sc->sc_dkdev.dk_openmask =
402 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
403 
404 	vndunlock(sc);
405 
406 	if ((sc->sc_flags & VNF_INITED) == 0) {
407 		if ((error = vnd_destroy(sc->sc_dev)) != 0) {
408 			aprint_error_dev(sc->sc_dev,
409 			    "unable to detach instance\n");
410 			return error;
411 		}
412 	}
413 
414 	return 0;
415 }
416 
417 /*
418  * Queue the request, and wakeup the kernel thread to handle it.
419  */
420 static void
421 vndstrategy(struct buf *bp)
422 {
423 	int unit = vndunit(bp->b_dev);
424 	struct vnd_softc *vnd =
425 	    device_lookup_private(&vnd_cd, unit);
426 	struct disklabel *lp;
427 	daddr_t blkno;
428 	int s = splbio();
429 
430 	if (vnd == NULL) {
431 		bp->b_error = ENXIO;
432 		goto done;
433 	}
434 	lp = vnd->sc_dkdev.dk_label;
435 
436 	if ((vnd->sc_flags & VNF_INITED) == 0) {
437 		bp->b_error = ENXIO;
438 		goto done;
439 	}
440 
441 	/*
442 	 * The transfer must be a whole number of blocks.
443 	 */
444 	if ((bp->b_bcount % lp->d_secsize) != 0) {
445 		bp->b_error = EINVAL;
446 		goto done;
447 	}
448 
449 	/*
450 	 * check if we're read-only.
451 	 */
452 	if ((vnd->sc_flags & VNF_READONLY) && !(bp->b_flags & B_READ)) {
453 		bp->b_error = EACCES;
454 		goto done;
455 	}
456 
457 	/* If it's a nil transfer, wake up the top half now. */
458 	if (bp->b_bcount == 0) {
459 		goto done;
460 	}
461 
462 	/*
463 	 * Do bounds checking and adjust transfer.  If there's an error,
464 	 * the bounds check will flag that for us.
465 	 */
466 	if (DISKPART(bp->b_dev) == RAW_PART) {
467 		if (bounds_check_with_mediasize(bp, DEV_BSIZE,
468 		    vnd->sc_size) <= 0)
469 			goto done;
470 	} else {
471 		if (bounds_check_with_label(&vnd->sc_dkdev,
472 		    bp, vnd->sc_flags & (VNF_WLABEL|VNF_LABELLING)) <= 0)
473 			goto done;
474 	}
475 
476 	/*
477 	 * Put the block number in terms of the logical blocksize
478 	 * of the "device".
479 	 */
480 
481 	blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
482 
483 	/*
484 	 * Translate the partition-relative block number to an absolute.
485 	 */
486 	if (DISKPART(bp->b_dev) != RAW_PART) {
487 		struct partition *pp;
488 
489 		pp = &vnd->sc_dkdev.dk_label->d_partitions[
490 		    DISKPART(bp->b_dev)];
491 		blkno += pp->p_offset;
492 	}
493 	bp->b_rawblkno = blkno;
494 
495 #ifdef DEBUG
496 	if (vnddebug & VDB_FOLLOW)
497 		printf("vndstrategy(%p): unit %d\n", bp, unit);
498 #endif
499 	bufq_put(vnd->sc_tab, bp);
500 	wakeup(&vnd->sc_tab);
501 	splx(s);
502 	return;
503 
504 done:
505 	bp->b_resid = bp->b_bcount;
506 	biodone(bp);
507 	splx(s);
508 }
509 
510 static bool
511 vnode_has_strategy(struct vnd_softc *vnd)
512 {
513 	return vnode_has_op(vnd->sc_vp, VOFFSET(vop_bmap)) &&
514 	    vnode_has_op(vnd->sc_vp, VOFFSET(vop_strategy));
515 }
516 
517 /* XXX this function needs a reliable check to detect
518  * sparse files. Otherwise, bmap/strategy may be used
519  * and fail on non-allocated blocks. VOP_READ/VOP_WRITE
520  * works on sparse files.
521  */
522 #if notyet
523 static bool
524 vnode_strategy_probe(struct vnd_softc *vnd)
525 {
526 	int error;
527 	daddr_t nbn;
528 
529 	if (!vnode_has_strategy(vnd))
530 		return false;
531 
532 	/* Convert the first logical block number to its
533 	 * physical block number.
534 	 */
535 	error = 0;
536 	vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
537 	error = VOP_BMAP(vnd->sc_vp, 0, NULL, &nbn, NULL);
538 	VOP_UNLOCK(vnd->sc_vp);
539 
540 	/* Test if that worked. */
541 	if (error == 0 && (long)nbn == -1)
542 		return false;
543 
544 	return true;
545 }
546 #endif
547 
548 static void
549 vndthread(void *arg)
550 {
551 	struct vnd_softc *vnd = arg;
552 	bool usestrategy;
553 	int s;
554 
555 	/* Determine whether we can *use* VOP_BMAP and VOP_STRATEGY to
556 	 * directly access the backing vnode.  If we can, use these two
557 	 * operations to avoid messing with the local buffer cache.
558 	 * Otherwise fall back to regular VOP_READ/VOP_WRITE operations
559 	 * which are guaranteed to work with any file system. */
560 	usestrategy = vnode_has_strategy(vnd);
561 
562 #ifdef DEBUG
563 	if (vnddebug & VDB_INIT)
564 		printf("vndthread: vp %p, %s\n", vnd->sc_vp,
565 		    usestrategy ?
566 		    "using bmap/strategy operations" :
567 		    "using read/write operations");
568 #endif
569 
570 	s = splbio();
571 	vnd->sc_flags |= VNF_KTHREAD;
572 	wakeup(&vnd->sc_kthread);
573 
574 	/*
575 	 * Dequeue requests and serve them depending on the available
576 	 * vnode operations.
577 	 */
578 	while ((vnd->sc_flags & VNF_VUNCONF) == 0) {
579 		struct vndxfer *vnx;
580 		int flags;
581 		struct buf *obp;
582 		struct buf *bp;
583 
584 		obp = bufq_get(vnd->sc_tab);
585 		if (obp == NULL) {
586 			tsleep(&vnd->sc_tab, PRIBIO, "vndbp", 0);
587 			continue;
588 		};
589 		splx(s);
590 		flags = obp->b_flags;
591 #ifdef DEBUG
592 		if (vnddebug & VDB_FOLLOW)
593 			printf("vndthread(%p)\n", obp);
594 #endif
595 
596 		if (vnd->sc_vp->v_mount == NULL) {
597 			obp->b_error = ENXIO;
598 			goto done;
599 		}
600 #ifdef VND_COMPRESSION
601 		/* handle a compressed read */
602 		if ((flags & B_READ) != 0 && (vnd->sc_flags & VNF_COMP)) {
603 			off_t bn;
604 
605 			/* Convert to a byte offset within the file. */
606 			bn = obp->b_rawblkno *
607 			    vnd->sc_dkdev.dk_label->d_secsize;
608 
609 			compstrategy(obp, bn);
610 			goto done;
611 		}
612 #endif /* VND_COMPRESSION */
613 
614 		/*
615 		 * Allocate a header for this transfer and link it to the
616 		 * buffer
617 		 */
618 		s = splbio();
619 		vnx = VND_GETXFER(vnd);
620 		splx(s);
621 		vnx->vx_vnd = vnd;
622 
623 		s = splbio();
624 		while (vnd->sc_active >= vnd->sc_maxactive) {
625 			tsleep(&vnd->sc_tab, PRIBIO, "vndac", 0);
626 		}
627 		vnd->sc_active++;
628 		splx(s);
629 
630 		/* Instrumentation. */
631 		disk_busy(&vnd->sc_dkdev);
632 
633 		bp = &vnx->vx_buf;
634 		buf_init(bp);
635 		bp->b_flags = (obp->b_flags & B_READ);
636 		bp->b_oflags = obp->b_oflags;
637 		bp->b_cflags = obp->b_cflags;
638 		bp->b_iodone = vndiodone;
639 		bp->b_private = obp;
640 		bp->b_vp = vnd->sc_vp;
641 		bp->b_objlock = &bp->b_vp->v_interlock;
642 		bp->b_data = obp->b_data;
643 		bp->b_bcount = obp->b_bcount;
644 		BIO_COPYPRIO(bp, obp);
645 
646 		/* Handle the request using the appropriate operations. */
647 		if (usestrategy)
648 			handle_with_strategy(vnd, obp, bp);
649 		else
650 			handle_with_rdwr(vnd, obp, bp);
651 
652 		s = splbio();
653 		continue;
654 
655 done:
656 		biodone(obp);
657 		s = splbio();
658 	}
659 
660 	vnd->sc_flags &= (~VNF_KTHREAD | VNF_VUNCONF);
661 	wakeup(&vnd->sc_kthread);
662 	splx(s);
663 	kthread_exit(0);
664 }
665 
666 /*
667  * Checks if the given vnode supports the requested operation.
668  * The operation is specified the offset returned by VOFFSET.
669  *
670  * XXX The test below used to determine this is quite fragile
671  * because it relies on the file system to use genfs to specify
672  * unimplemented operations.  There might be another way to do
673  * it more cleanly.
674  */
675 static bool
676 vnode_has_op(const struct vnode *vp, int opoffset)
677 {
678 	int (*defaultp)(void *);
679 	int (*opp)(void *);
680 
681 	defaultp = vp->v_op[VOFFSET(vop_default)];
682 	opp = vp->v_op[opoffset];
683 
684 	return opp != defaultp && opp != genfs_eopnotsupp &&
685 	    opp != genfs_badop && opp != genfs_nullop;
686 }
687 
688 /*
689  * Handes the read/write request given in 'bp' using the vnode's VOP_READ
690  * and VOP_WRITE operations.
691  *
692  * 'obp' is a pointer to the original request fed to the vnd device.
693  */
694 static void
695 handle_with_rdwr(struct vnd_softc *vnd, const struct buf *obp, struct buf *bp)
696 {
697 	bool doread;
698 	off_t offset;
699 	size_t resid;
700 	struct vnode *vp;
701 
702 	doread = bp->b_flags & B_READ;
703 	offset = obp->b_rawblkno * vnd->sc_dkdev.dk_label->d_secsize;
704 	vp = vnd->sc_vp;
705 
706 #if defined(DEBUG)
707 	if (vnddebug & VDB_IO)
708 		printf("vnd (rdwr): vp %p, %s, rawblkno 0x%" PRIx64
709 		    ", secsize %d, offset %" PRIu64
710 		    ", bcount %d\n",
711 		    vp, doread ? "read" : "write", obp->b_rawblkno,
712 		    vnd->sc_dkdev.dk_label->d_secsize, offset,
713 		    bp->b_bcount);
714 #endif
715 
716 	/* Issue the read or write operation. */
717 	bp->b_error =
718 	    vn_rdwr(doread ? UIO_READ : UIO_WRITE,
719 	    vp, bp->b_data, bp->b_bcount, offset,
720 	    UIO_SYSSPACE, 0, vnd->sc_cred, &resid, NULL);
721 	bp->b_resid = resid;
722 
723 	/* We need to increase the number of outputs on the vnode if
724 	 * there was any write to it. */
725 	if (!doread) {
726 		mutex_enter(&vp->v_interlock);
727 		vp->v_numoutput++;
728 		mutex_exit(&vp->v_interlock);
729 	}
730 
731 	biodone(bp);
732 }
733 
734 /*
735  * Handes the read/write request given in 'bp' using the vnode's VOP_BMAP
736  * and VOP_STRATEGY operations.
737  *
738  * 'obp' is a pointer to the original request fed to the vnd device.
739  */
740 static void
741 handle_with_strategy(struct vnd_softc *vnd, const struct buf *obp,
742     struct buf *bp)
743 {
744 	int bsize, error, flags, skipped;
745 	size_t resid, sz;
746 	off_t bn, offset;
747 	struct vnode *vp;
748 
749 	flags = obp->b_flags;
750 
751 	if (!(flags & B_READ)) {
752 		vp = bp->b_vp;
753 		mutex_enter(&vp->v_interlock);
754 		vp->v_numoutput++;
755 		mutex_exit(&vp->v_interlock);
756 	}
757 
758 	/* convert to a byte offset within the file. */
759 	bn = obp->b_rawblkno * vnd->sc_dkdev.dk_label->d_secsize;
760 
761 	bsize = vnd->sc_vp->v_mount->mnt_stat.f_iosize;
762 	skipped = 0;
763 
764 	/*
765 	 * Break the request into bsize pieces and feed them
766 	 * sequentially using VOP_BMAP/VOP_STRATEGY.
767 	 * We do it this way to keep from flooding NFS servers if we
768 	 * are connected to an NFS file.  This places the burden on
769 	 * the client rather than the server.
770 	 */
771 	error = 0;
772 	bp->b_resid = bp->b_bcount;
773 	for (offset = 0, resid = bp->b_resid; resid;
774 	    resid -= sz, offset += sz) {
775 		struct buf *nbp;
776 		daddr_t nbn;
777 		int off, nra;
778 
779 		nra = 0;
780 		vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
781 		error = VOP_BMAP(vnd->sc_vp, bn / bsize, &vp, &nbn, &nra);
782 		VOP_UNLOCK(vnd->sc_vp);
783 
784 		if (error == 0 && (long)nbn == -1)
785 			error = EIO;
786 
787 		/*
788 		 * If there was an error or a hole in the file...punt.
789 		 * Note that we may have to wait for any operations
790 		 * that we have already fired off before releasing
791 		 * the buffer.
792 		 *
793 		 * XXX we could deal with holes here but it would be
794 		 * a hassle (in the write case).
795 		 */
796 		if (error) {
797 			skipped += resid;
798 			break;
799 		}
800 
801 #ifdef DEBUG
802 		if (!dovndcluster)
803 			nra = 0;
804 #endif
805 
806 		off = bn % bsize;
807 		sz = MIN(((off_t)1 + nra) * bsize - off, resid);
808 #ifdef	DEBUG
809 		if (vnddebug & VDB_IO)
810 			printf("vndstrategy: vp %p/%p bn 0x%qx/0x%" PRIx64
811 			    " sz 0x%zx\n", vnd->sc_vp, vp, (long long)bn,
812 			    nbn, sz);
813 #endif
814 
815 		nbp = getiobuf(vp, true);
816 		nestiobuf_setup(bp, nbp, offset, sz);
817 		nbp->b_blkno = nbn + btodb(off);
818 
819 #if 0 /* XXX #ifdef DEBUG */
820 		if (vnddebug & VDB_IO)
821 			printf("vndstart(%ld): bp %p vp %p blkno "
822 			    "0x%" PRIx64 " flags %x addr %p cnt 0x%x\n",
823 			    (long) (vnd-vnd_softc), &nbp->vb_buf,
824 			    nbp->vb_buf.b_vp, nbp->vb_buf.b_blkno,
825 			    nbp->vb_buf.b_flags, nbp->vb_buf.b_data,
826 			    nbp->vb_buf.b_bcount);
827 #endif
828 		VOP_STRATEGY(vp, nbp);
829 		bn += sz;
830 	}
831 	nestiobuf_done(bp, skipped, error);
832 }
833 
834 static void
835 vndiodone(struct buf *bp)
836 {
837 	struct vndxfer *vnx = VND_BUFTOXFER(bp);
838 	struct vnd_softc *vnd = vnx->vx_vnd;
839 	struct buf *obp = bp->b_private;
840 	int s = splbio();
841 
842 	KASSERT(&vnx->vx_buf == bp);
843 	KASSERT(vnd->sc_active > 0);
844 #ifdef DEBUG
845 	if (vnddebug & VDB_IO) {
846 		printf("vndiodone1: bp %p iodone: error %d\n",
847 		    bp, bp->b_error);
848 	}
849 #endif
850 	disk_unbusy(&vnd->sc_dkdev, bp->b_bcount - bp->b_resid,
851 	    (bp->b_flags & B_READ));
852 	vnd->sc_active--;
853 	if (vnd->sc_active == 0) {
854 		wakeup(&vnd->sc_tab);
855 	}
856 	splx(s);
857 	obp->b_error = bp->b_error;
858 	obp->b_resid = bp->b_resid;
859 	buf_destroy(bp);
860 	VND_PUTXFER(vnd, vnx);
861 	biodone(obp);
862 }
863 
864 /* ARGSUSED */
865 static int
866 vndread(dev_t dev, struct uio *uio, int flags)
867 {
868 	int unit = vndunit(dev);
869 	struct vnd_softc *sc;
870 
871 #ifdef DEBUG
872 	if (vnddebug & VDB_FOLLOW)
873 		printf("vndread(0x%"PRIx64", %p)\n", dev, uio);
874 #endif
875 
876 	sc = device_lookup_private(&vnd_cd, unit);
877 	if (sc == NULL)
878 		return ENXIO;
879 
880 	if ((sc->sc_flags & VNF_INITED) == 0)
881 		return ENXIO;
882 
883 	return physio(vndstrategy, NULL, dev, B_READ, minphys, uio);
884 }
885 
886 /* ARGSUSED */
887 static int
888 vndwrite(dev_t dev, struct uio *uio, int flags)
889 {
890 	int unit = vndunit(dev);
891 	struct vnd_softc *sc;
892 
893 #ifdef DEBUG
894 	if (vnddebug & VDB_FOLLOW)
895 		printf("vndwrite(0x%"PRIx64", %p)\n", dev, uio);
896 #endif
897 
898 	sc = device_lookup_private(&vnd_cd, unit);
899 	if (sc == NULL)
900 		return ENXIO;
901 
902 	if ((sc->sc_flags & VNF_INITED) == 0)
903 		return ENXIO;
904 
905 	return physio(vndstrategy, NULL, dev, B_WRITE, minphys, uio);
906 }
907 
908 static int
909 vnd_cget(struct lwp *l, int unit, int *un, struct vattr *va)
910 {
911 	struct vnd_softc *vnd;
912 
913 	if (*un == -1)
914 		*un = unit;
915 	if (*un < 0)
916 		return EINVAL;
917 
918 	vnd = device_lookup_private(&vnd_cd, *un);
919 	if (vnd == NULL)
920 		return (*un >= vnd_cd.cd_ndevs) ? ENXIO : -1;
921 
922 	if ((vnd->sc_flags & VNF_INITED) == 0)
923 		return -1;
924 
925 	return VOP_GETATTR(vnd->sc_vp, va, l->l_cred);
926 }
927 
928 static int
929 vnddoclear(struct vnd_softc *vnd, int pmask, int minor, bool force)
930 {
931 	int error;
932 
933 	if ((error = vndlock(vnd)) != 0)
934 		return error;
935 
936 	/*
937 	 * Don't unconfigure if any other partitions are open
938 	 * or if both the character and block flavors of this
939 	 * partition are open.
940 	 */
941 	if (DK_BUSY(vnd, pmask) && !force) {
942 		vndunlock(vnd);
943 		return EBUSY;
944 	}
945 
946 	/*
947 	 * XXX vndclear() might call vndclose() implicitly;
948 	 * release lock to avoid recursion
949 	 *
950 	 * Set VNF_CLEARING to prevent vndopen() from
951 	 * sneaking in after we vndunlock().
952 	 */
953 	vnd->sc_flags |= VNF_CLEARING;
954 	vndunlock(vnd);
955 	vndclear(vnd, minor);
956 #ifdef DEBUG
957 	if (vnddebug & VDB_INIT)
958 		printf("vndioctl: CLRed\n");
959 #endif
960 
961 	/* Destroy the xfer and buffer pools. */
962 	pool_destroy(&vnd->sc_vxpool);
963 
964 	/* Detach the disk. */
965 	disk_detach(&vnd->sc_dkdev);
966 
967 	return 0;
968 }
969 
970 /* ARGSUSED */
971 static int
972 vndioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l)
973 {
974 	bool force;
975 	int unit = vndunit(dev);
976 	struct vnd_softc *vnd;
977 	struct vnd_ioctl *vio;
978 	struct vattr vattr;
979 	struct pathbuf *pb;
980 	struct nameidata nd;
981 	int error, part, pmask;
982 	size_t geomsize;
983 	int fflags;
984 #ifdef __HAVE_OLD_DISKLABEL
985 	struct disklabel newlabel;
986 #endif
987 	struct dkwedge_info *dkw;
988 	struct dkwedge_list *dkwl;
989 
990 #ifdef DEBUG
991 	if (vnddebug & VDB_FOLLOW)
992 		printf("vndioctl(0x%"PRIx64", 0x%lx, %p, 0x%x, %p): unit %d\n",
993 		    dev, cmd, data, flag, l->l_proc, unit);
994 #endif
995 	vnd = device_lookup_private(&vnd_cd, unit);
996 	if (vnd == NULL &&
997 #ifdef COMPAT_30
998 	    cmd != VNDIOCGET30 &&
999 #endif
1000 #ifdef COMPAT_50
1001 	    cmd != VNDIOCGET50 &&
1002 #endif
1003 	    cmd != VNDIOCGET)
1004 		return ENXIO;
1005 	vio = (struct vnd_ioctl *)data;
1006 
1007 	/* Must be open for writes for these commands... */
1008 	switch (cmd) {
1009 	case VNDIOCSET:
1010 	case VNDIOCCLR:
1011 #ifdef COMPAT_50
1012 	case VNDIOCSET50:
1013 	case VNDIOCCLR50:
1014 #endif
1015 	case DIOCSDINFO:
1016 	case DIOCWDINFO:
1017 #ifdef __HAVE_OLD_DISKLABEL
1018 	case ODIOCSDINFO:
1019 	case ODIOCWDINFO:
1020 #endif
1021 	case DIOCKLABEL:
1022 	case DIOCWLABEL:
1023 		if ((flag & FWRITE) == 0)
1024 			return EBADF;
1025 	}
1026 
1027 	/* Must be initialized for these... */
1028 	switch (cmd) {
1029 	case VNDIOCCLR:
1030 #ifdef VNDIOCCLR50
1031 	case VNDIOCCLR50:
1032 #endif
1033 	case DIOCGDINFO:
1034 	case DIOCSDINFO:
1035 	case DIOCWDINFO:
1036 	case DIOCGPART:
1037 	case DIOCKLABEL:
1038 	case DIOCWLABEL:
1039 	case DIOCGDEFLABEL:
1040 	case DIOCCACHESYNC:
1041 #ifdef __HAVE_OLD_DISKLABEL
1042 	case ODIOCGDINFO:
1043 	case ODIOCSDINFO:
1044 	case ODIOCWDINFO:
1045 	case ODIOCGDEFLABEL:
1046 #endif
1047 		if ((vnd->sc_flags & VNF_INITED) == 0)
1048 			return ENXIO;
1049 	}
1050 
1051 	switch (cmd) {
1052 #ifdef VNDIOCSET50
1053 	case VNDIOCSET50:
1054 #endif
1055 	case VNDIOCSET:
1056 		if (vnd->sc_flags & VNF_INITED)
1057 			return EBUSY;
1058 
1059 		if ((error = vndlock(vnd)) != 0)
1060 			return error;
1061 
1062 		fflags = FREAD;
1063 		if ((vio->vnd_flags & VNDIOF_READONLY) == 0)
1064 			fflags |= FWRITE;
1065 		error = pathbuf_copyin(vio->vnd_file, &pb);
1066 		if (error) {
1067 			goto unlock_and_exit;
1068 		}
1069 		NDINIT(&nd, LOOKUP, FOLLOW, pb);
1070 		if ((error = vn_open(&nd, fflags, 0)) != 0) {
1071 			pathbuf_destroy(pb);
1072 			goto unlock_and_exit;
1073 		}
1074 		KASSERT(l);
1075 		error = VOP_GETATTR(nd.ni_vp, &vattr, l->l_cred);
1076 		if (!error && nd.ni_vp->v_type != VREG)
1077 			error = EOPNOTSUPP;
1078 		if (!error && vattr.va_bytes < vattr.va_size)
1079 			/* File is definitely sparse, reject here */
1080 			error = EINVAL;
1081 		if (error) {
1082 			VOP_UNLOCK(nd.ni_vp);
1083 			goto close_and_exit;
1084 		}
1085 
1086 		/* If using a compressed file, initialize its info */
1087 		/* (or abort with an error if kernel has no compression) */
1088 		if (vio->vnd_flags & VNF_COMP) {
1089 #ifdef VND_COMPRESSION
1090 			struct vnd_comp_header *ch;
1091 			int i;
1092 			u_int32_t comp_size;
1093 			u_int32_t comp_maxsize;
1094 
1095 			/* allocate space for compresed file header */
1096 			ch = malloc(sizeof(struct vnd_comp_header),
1097 			M_TEMP, M_WAITOK);
1098 
1099 			/* read compressed file header */
1100 			error = vn_rdwr(UIO_READ, nd.ni_vp, (void *)ch,
1101 			  sizeof(struct vnd_comp_header), 0, UIO_SYSSPACE,
1102 			  IO_UNIT|IO_NODELOCKED, l->l_cred, NULL, NULL);
1103 			if (error) {
1104 				free(ch, M_TEMP);
1105 				VOP_UNLOCK(nd.ni_vp);
1106 				goto close_and_exit;
1107 			}
1108 
1109 			/* save some header info */
1110 			vnd->sc_comp_blksz = ntohl(ch->block_size);
1111 			/* note last offset is the file byte size */
1112 			vnd->sc_comp_numoffs = ntohl(ch->num_blocks)+1;
1113 			free(ch, M_TEMP);
1114 			if (vnd->sc_comp_blksz == 0 ||
1115 			    vnd->sc_comp_blksz % DEV_BSIZE !=0) {
1116 				VOP_UNLOCK(nd.ni_vp);
1117 				error = EINVAL;
1118 				goto close_and_exit;
1119 			}
1120 			if (sizeof(struct vnd_comp_header) +
1121 			  sizeof(u_int64_t) * vnd->sc_comp_numoffs >
1122 			  vattr.va_size) {
1123 				VOP_UNLOCK(nd.ni_vp);
1124 				error = EINVAL;
1125 				goto close_and_exit;
1126 			}
1127 
1128 			/* set decompressed file size */
1129 			vattr.va_size =
1130 			    ((u_quad_t)vnd->sc_comp_numoffs - 1) *
1131 			     (u_quad_t)vnd->sc_comp_blksz;
1132 
1133 			/* allocate space for all the compressed offsets */
1134 			vnd->sc_comp_offsets =
1135 			malloc(sizeof(u_int64_t) * vnd->sc_comp_numoffs,
1136 			M_DEVBUF, M_WAITOK);
1137 
1138 			/* read in the offsets */
1139 			error = vn_rdwr(UIO_READ, nd.ni_vp,
1140 			  (void *)vnd->sc_comp_offsets,
1141 			  sizeof(u_int64_t) * vnd->sc_comp_numoffs,
1142 			  sizeof(struct vnd_comp_header), UIO_SYSSPACE,
1143 			  IO_UNIT|IO_NODELOCKED, l->l_cred, NULL, NULL);
1144 			if (error) {
1145 				VOP_UNLOCK(nd.ni_vp);
1146 				goto close_and_exit;
1147 			}
1148 			/*
1149 			 * find largest block size (used for allocation limit).
1150 			 * Also convert offset to native byte order.
1151 			 */
1152 			comp_maxsize = 0;
1153 			for (i = 0; i < vnd->sc_comp_numoffs - 1; i++) {
1154 				vnd->sc_comp_offsets[i] =
1155 				  be64toh(vnd->sc_comp_offsets[i]);
1156 				comp_size = be64toh(vnd->sc_comp_offsets[i + 1])
1157 				  - vnd->sc_comp_offsets[i];
1158 				if (comp_size > comp_maxsize)
1159 					comp_maxsize = comp_size;
1160 			}
1161 			vnd->sc_comp_offsets[vnd->sc_comp_numoffs - 1] =
1162 			  be64toh(vnd->sc_comp_offsets[vnd->sc_comp_numoffs - 1]);
1163 
1164 			/* create compressed data buffer */
1165 			vnd->sc_comp_buff = malloc(comp_maxsize,
1166 			  M_DEVBUF, M_WAITOK);
1167 
1168 			/* create decompressed buffer */
1169 			vnd->sc_comp_decombuf = malloc(vnd->sc_comp_blksz,
1170 			  M_DEVBUF, M_WAITOK);
1171 			vnd->sc_comp_buffblk = -1;
1172 
1173 			/* Initialize decompress stream */
1174 			memset(&vnd->sc_comp_stream, 0, sizeof(z_stream));
1175 			vnd->sc_comp_stream.zalloc = vnd_alloc;
1176 			vnd->sc_comp_stream.zfree = vnd_free;
1177 			error = inflateInit2(&vnd->sc_comp_stream, MAX_WBITS);
1178 			if (error) {
1179 				if (vnd->sc_comp_stream.msg)
1180 					printf("vnd%d: compressed file, %s\n",
1181 					  unit, vnd->sc_comp_stream.msg);
1182 				VOP_UNLOCK(nd.ni_vp);
1183 				error = EINVAL;
1184 				goto close_and_exit;
1185 			}
1186 
1187 			vnd->sc_flags |= VNF_COMP | VNF_READONLY;
1188 #else /* !VND_COMPRESSION */
1189 			VOP_UNLOCK(nd.ni_vp);
1190 			error = EOPNOTSUPP;
1191 			goto close_and_exit;
1192 #endif /* VND_COMPRESSION */
1193 		}
1194 
1195 		VOP_UNLOCK(nd.ni_vp);
1196 		vnd->sc_vp = nd.ni_vp;
1197 		vnd->sc_size = btodb(vattr.va_size);	/* note truncation */
1198 
1199 		/*
1200 		 * Use pseudo-geometry specified.  If none was provided,
1201 		 * use "standard" Adaptec fictitious geometry.
1202 		 */
1203 		if (vio->vnd_flags & VNDIOF_HASGEOM) {
1204 
1205 			memcpy(&vnd->sc_geom, &vio->vnd_geom,
1206 			    sizeof(vio->vnd_geom));
1207 
1208 			/*
1209 			 * Sanity-check the sector size.
1210 			 * XXX Don't allow secsize < DEV_BSIZE.	 Should
1211 			 * XXX we?
1212 			 */
1213 			if (vnd->sc_geom.vng_secsize < DEV_BSIZE ||
1214 			    (vnd->sc_geom.vng_secsize % DEV_BSIZE) != 0 ||
1215 			    vnd->sc_geom.vng_ncylinders == 0 ||
1216 			    (vnd->sc_geom.vng_ntracks *
1217 			     vnd->sc_geom.vng_nsectors) == 0) {
1218 				error = EINVAL;
1219 				goto close_and_exit;
1220 			}
1221 
1222 			/*
1223 			 * Compute the size (in DEV_BSIZE blocks) specified
1224 			 * by the geometry.
1225 			 */
1226 			geomsize = (vnd->sc_geom.vng_nsectors *
1227 			    vnd->sc_geom.vng_ntracks *
1228 			    vnd->sc_geom.vng_ncylinders) *
1229 			    (vnd->sc_geom.vng_secsize / DEV_BSIZE);
1230 
1231 			/*
1232 			 * Sanity-check the size against the specified
1233 			 * geometry.
1234 			 */
1235 			if (vnd->sc_size < geomsize) {
1236 				error = EINVAL;
1237 				goto close_and_exit;
1238 			}
1239 		} else if (vnd->sc_size >= (32 * 64)) {
1240 			/*
1241 			 * Size must be at least 2048 DEV_BSIZE blocks
1242 			 * (1M) in order to use this geometry.
1243 			 */
1244 			vnd->sc_geom.vng_secsize = DEV_BSIZE;
1245 			vnd->sc_geom.vng_nsectors = 32;
1246 			vnd->sc_geom.vng_ntracks = 64;
1247 			vnd->sc_geom.vng_ncylinders = vnd->sc_size / (64 * 32);
1248 		} else {
1249 			vnd->sc_geom.vng_secsize = DEV_BSIZE;
1250 			vnd->sc_geom.vng_nsectors = 1;
1251 			vnd->sc_geom.vng_ntracks = 1;
1252 			vnd->sc_geom.vng_ncylinders = vnd->sc_size;
1253 		}
1254 
1255 		vnd_set_properties(vnd);
1256 
1257 		if (vio->vnd_flags & VNDIOF_READONLY) {
1258 			vnd->sc_flags |= VNF_READONLY;
1259 		}
1260 
1261 		if ((error = vndsetcred(vnd, l->l_cred)) != 0)
1262 			goto close_and_exit;
1263 
1264 		vndthrottle(vnd, vnd->sc_vp);
1265 		vio->vnd_osize = dbtob(vnd->sc_size);
1266 #ifdef VNDIOCSET50
1267 		if (cmd != VNDIOCSET50)
1268 #endif
1269 			vio->vnd_size = dbtob(vnd->sc_size);
1270 		vnd->sc_flags |= VNF_INITED;
1271 
1272 		/* create the kernel thread, wait for it to be up */
1273 		error = kthread_create(PRI_NONE, 0, NULL, vndthread, vnd,
1274 		    &vnd->sc_kthread, device_xname(vnd->sc_dev));
1275 		if (error)
1276 			goto close_and_exit;
1277 		while ((vnd->sc_flags & VNF_KTHREAD) == 0) {
1278 			tsleep(&vnd->sc_kthread, PRIBIO, "vndthr", 0);
1279 		}
1280 #ifdef DEBUG
1281 		if (vnddebug & VDB_INIT)
1282 			printf("vndioctl: SET vp %p size 0x%lx %d/%d/%d/%d\n",
1283 			    vnd->sc_vp, (unsigned long) vnd->sc_size,
1284 			    vnd->sc_geom.vng_secsize,
1285 			    vnd->sc_geom.vng_nsectors,
1286 			    vnd->sc_geom.vng_ntracks,
1287 			    vnd->sc_geom.vng_ncylinders);
1288 #endif
1289 
1290 		/* Attach the disk. */
1291 		disk_attach(&vnd->sc_dkdev);
1292 		disk_blocksize(&vnd->sc_dkdev, vnd->sc_geom.vng_secsize);
1293 
1294 		/* Initialize the xfer and buffer pools. */
1295 		pool_init(&vnd->sc_vxpool, sizeof(struct vndxfer), 0,
1296 		    0, 0, "vndxpl", NULL, IPL_BIO);
1297 
1298 		/* Try and read the disklabel. */
1299 		vndgetdisklabel(dev, vnd);
1300 
1301 		vndunlock(vnd);
1302 
1303 		pathbuf_destroy(pb);
1304 		break;
1305 
1306 close_and_exit:
1307 		(void) vn_close(nd.ni_vp, fflags, l->l_cred);
1308 		pathbuf_destroy(pb);
1309 unlock_and_exit:
1310 #ifdef VND_COMPRESSION
1311 		/* free any allocated memory (for compressed file) */
1312 		if (vnd->sc_comp_offsets) {
1313 			free(vnd->sc_comp_offsets, M_DEVBUF);
1314 			vnd->sc_comp_offsets = NULL;
1315 		}
1316 		if (vnd->sc_comp_buff) {
1317 			free(vnd->sc_comp_buff, M_DEVBUF);
1318 			vnd->sc_comp_buff = NULL;
1319 		}
1320 		if (vnd->sc_comp_decombuf) {
1321 			free(vnd->sc_comp_decombuf, M_DEVBUF);
1322 			vnd->sc_comp_decombuf = NULL;
1323 		}
1324 #endif /* VND_COMPRESSION */
1325 		vndunlock(vnd);
1326 		return error;
1327 
1328 #ifdef VNDIOCCLR50
1329 	case VNDIOCCLR50:
1330 #endif
1331 	case VNDIOCCLR:
1332 		part = DISKPART(dev);
1333 		pmask = (1 << part);
1334 		force = (vio->vnd_flags & VNDIOF_FORCE) != 0;
1335 
1336 		if ((error = vnddoclear(vnd, pmask, minor(dev), force)) != 0)
1337 			return error;
1338 
1339 		break;
1340 
1341 #ifdef COMPAT_30
1342 	case VNDIOCGET30: {
1343 		struct vnd_user30 *vnu;
1344 		struct vattr va;
1345 		vnu = (struct vnd_user30 *)data;
1346 		KASSERT(l);
1347 		switch (error = vnd_cget(l, unit, &vnu->vnu_unit, &va)) {
1348 		case 0:
1349 			vnu->vnu_dev = va.va_fsid;
1350 			vnu->vnu_ino = va.va_fileid;
1351 			break;
1352 		case -1:
1353 			/* unused is not an error */
1354 			vnu->vnu_dev = 0;
1355 			vnu->vnu_ino = 0;
1356 			break;
1357 		default:
1358 			return error;
1359 		}
1360 		break;
1361 	}
1362 #endif
1363 
1364 #ifdef COMPAT_50
1365 	case VNDIOCGET50: {
1366 		struct vnd_user50 *vnu;
1367 		struct vattr va;
1368 		vnu = (struct vnd_user50 *)data;
1369 		KASSERT(l);
1370 		switch (error = vnd_cget(l, unit, &vnu->vnu_unit, &va)) {
1371 		case 0:
1372 			vnu->vnu_dev = va.va_fsid;
1373 			vnu->vnu_ino = va.va_fileid;
1374 			break;
1375 		case -1:
1376 			/* unused is not an error */
1377 			vnu->vnu_dev = 0;
1378 			vnu->vnu_ino = 0;
1379 			break;
1380 		default:
1381 			return error;
1382 		}
1383 		break;
1384 	}
1385 #endif
1386 
1387 	case VNDIOCGET: {
1388 		struct vnd_user *vnu;
1389 		struct vattr va;
1390 		vnu = (struct vnd_user *)data;
1391 		KASSERT(l);
1392 		switch (error = vnd_cget(l, unit, &vnu->vnu_unit, &va)) {
1393 		case 0:
1394 			vnu->vnu_dev = va.va_fsid;
1395 			vnu->vnu_ino = va.va_fileid;
1396 			break;
1397 		case -1:
1398 			/* unused is not an error */
1399 			vnu->vnu_dev = 0;
1400 			vnu->vnu_ino = 0;
1401 			break;
1402 		default:
1403 			return error;
1404 		}
1405 		break;
1406 	}
1407 
1408 	case DIOCGDINFO:
1409 		*(struct disklabel *)data = *(vnd->sc_dkdev.dk_label);
1410 		break;
1411 
1412 #ifdef __HAVE_OLD_DISKLABEL
1413 	case ODIOCGDINFO:
1414 		newlabel = *(vnd->sc_dkdev.dk_label);
1415 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1416 			return ENOTTY;
1417 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1418 		break;
1419 #endif
1420 
1421 	case DIOCGPART:
1422 		((struct partinfo *)data)->disklab = vnd->sc_dkdev.dk_label;
1423 		((struct partinfo *)data)->part =
1424 		    &vnd->sc_dkdev.dk_label->d_partitions[DISKPART(dev)];
1425 		break;
1426 
1427 	case DIOCWDINFO:
1428 	case DIOCSDINFO:
1429 #ifdef __HAVE_OLD_DISKLABEL
1430 	case ODIOCWDINFO:
1431 	case ODIOCSDINFO:
1432 #endif
1433 	{
1434 		struct disklabel *lp;
1435 
1436 		if ((error = vndlock(vnd)) != 0)
1437 			return error;
1438 
1439 		vnd->sc_flags |= VNF_LABELLING;
1440 
1441 #ifdef __HAVE_OLD_DISKLABEL
1442 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
1443 			memset(&newlabel, 0, sizeof newlabel);
1444 			memcpy(&newlabel, data, sizeof (struct olddisklabel));
1445 			lp = &newlabel;
1446 		} else
1447 #endif
1448 		lp = (struct disklabel *)data;
1449 
1450 		error = setdisklabel(vnd->sc_dkdev.dk_label,
1451 		    lp, 0, vnd->sc_dkdev.dk_cpulabel);
1452 		if (error == 0) {
1453 			if (cmd == DIOCWDINFO
1454 #ifdef __HAVE_OLD_DISKLABEL
1455 			    || cmd == ODIOCWDINFO
1456 #endif
1457 			   )
1458 				error = writedisklabel(VNDLABELDEV(dev),
1459 				    vndstrategy, vnd->sc_dkdev.dk_label,
1460 				    vnd->sc_dkdev.dk_cpulabel);
1461 		}
1462 
1463 		vnd->sc_flags &= ~VNF_LABELLING;
1464 
1465 		vndunlock(vnd);
1466 
1467 		if (error)
1468 			return error;
1469 		break;
1470 	}
1471 
1472 	case DIOCKLABEL:
1473 		if (*(int *)data != 0)
1474 			vnd->sc_flags |= VNF_KLABEL;
1475 		else
1476 			vnd->sc_flags &= ~VNF_KLABEL;
1477 		break;
1478 
1479 	case DIOCWLABEL:
1480 		if (*(int *)data != 0)
1481 			vnd->sc_flags |= VNF_WLABEL;
1482 		else
1483 			vnd->sc_flags &= ~VNF_WLABEL;
1484 		break;
1485 
1486 	case DIOCGDEFLABEL:
1487 		vndgetdefaultlabel(vnd, (struct disklabel *)data);
1488 		break;
1489 
1490 #ifdef __HAVE_OLD_DISKLABEL
1491 	case ODIOCGDEFLABEL:
1492 		vndgetdefaultlabel(vnd, &newlabel);
1493 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1494 			return ENOTTY;
1495 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1496 		break;
1497 #endif
1498 
1499 	case DIOCCACHESYNC:
1500 		vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1501 		error = VOP_FSYNC(vnd->sc_vp, vnd->sc_cred,
1502 		    FSYNC_WAIT | FSYNC_DATAONLY | FSYNC_CACHE, 0, 0);
1503 		VOP_UNLOCK(vnd->sc_vp);
1504 		return error;
1505 
1506 	case DIOCAWEDGE:
1507 		dkw = (void *) data;
1508 
1509 		if ((flag & FWRITE) == 0)
1510 			return EBADF;
1511 
1512 		/* If the ioctl happens here, the parent is us. */
1513 		strlcpy(dkw->dkw_parent, device_xname(vnd->sc_dev),
1514 		    sizeof(dkw->dkw_parent));
1515 		return dkwedge_add(dkw);
1516 
1517 	case DIOCDWEDGE:
1518 		dkw = (void *) data;
1519 
1520 		if ((flag & FWRITE) == 0)
1521 			return EBADF;
1522 
1523 		/* If the ioctl happens here, the parent is us. */
1524 		strlcpy(dkw->dkw_parent, device_xname(vnd->sc_dev),
1525 		    sizeof(dkw->dkw_parent));
1526 		return dkwedge_del(dkw);
1527 
1528 	case DIOCLWEDGES:
1529 		dkwl = (void *) data;
1530 
1531 		return dkwedge_list(&vnd->sc_dkdev, dkwl, l);
1532 
1533 	default:
1534 		return ENOTTY;
1535 	}
1536 
1537 	return 0;
1538 }
1539 
1540 /*
1541  * Duplicate the current processes' credentials.  Since we are called only
1542  * as the result of a SET ioctl and only root can do that, any future access
1543  * to this "disk" is essentially as root.  Note that credentials may change
1544  * if some other uid can write directly to the mapped file (NFS).
1545  */
1546 static int
1547 vndsetcred(struct vnd_softc *vnd, kauth_cred_t cred)
1548 {
1549 	struct uio auio;
1550 	struct iovec aiov;
1551 	char *tmpbuf;
1552 	int error;
1553 
1554 	vnd->sc_cred = kauth_cred_dup(cred);
1555 	tmpbuf = malloc(DEV_BSIZE, M_TEMP, M_WAITOK);
1556 
1557 	/* XXX: Horrible kludge to establish credentials for NFS */
1558 	aiov.iov_base = tmpbuf;
1559 	aiov.iov_len = min(DEV_BSIZE, dbtob(vnd->sc_size));
1560 	auio.uio_iov = &aiov;
1561 	auio.uio_iovcnt = 1;
1562 	auio.uio_offset = 0;
1563 	auio.uio_rw = UIO_READ;
1564 	auio.uio_resid = aiov.iov_len;
1565 	UIO_SETUP_SYSSPACE(&auio);
1566 	vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1567 	error = VOP_READ(vnd->sc_vp, &auio, 0, vnd->sc_cred);
1568 	if (error == 0) {
1569 		/*
1570 		 * Because vnd does all IO directly through the vnode
1571 		 * we need to flush (at least) the buffer from the above
1572 		 * VOP_READ from the buffer cache to prevent cache
1573 		 * incoherencies.  Also, be careful to write dirty
1574 		 * buffers back to stable storage.
1575 		 */
1576 		error = vinvalbuf(vnd->sc_vp, V_SAVE, vnd->sc_cred,
1577 			    curlwp, 0, 0);
1578 	}
1579 	VOP_UNLOCK(vnd->sc_vp);
1580 
1581 	free(tmpbuf, M_TEMP);
1582 	return error;
1583 }
1584 
1585 /*
1586  * Set maxactive based on FS type
1587  */
1588 static void
1589 vndthrottle(struct vnd_softc *vnd, struct vnode *vp)
1590 {
1591 
1592 	if (vp->v_tag == VT_NFS)
1593 		vnd->sc_maxactive = 2;
1594 	else
1595 		vnd->sc_maxactive = 8;
1596 
1597 	if (vnd->sc_maxactive < 1)
1598 		vnd->sc_maxactive = 1;
1599 }
1600 
1601 #if 0
1602 static void
1603 vndshutdown(void)
1604 {
1605 	struct vnd_softc *vnd;
1606 
1607 	for (vnd = &vnd_softc[0]; vnd < &vnd_softc[numvnd]; vnd++)
1608 		if (vnd->sc_flags & VNF_INITED)
1609 			vndclear(vnd);
1610 }
1611 #endif
1612 
1613 static void
1614 vndclear(struct vnd_softc *vnd, int myminor)
1615 {
1616 	struct vnode *vp = vnd->sc_vp;
1617 	int fflags = FREAD;
1618 	int bmaj, cmaj, i, mn;
1619 	int s;
1620 
1621 #ifdef DEBUG
1622 	if (vnddebug & VDB_FOLLOW)
1623 		printf("vndclear(%p): vp %p\n", vnd, vp);
1624 #endif
1625 	/* locate the major number */
1626 	bmaj = bdevsw_lookup_major(&vnd_bdevsw);
1627 	cmaj = cdevsw_lookup_major(&vnd_cdevsw);
1628 
1629 	/* Nuke the vnodes for any open instances */
1630 	for (i = 0; i < MAXPARTITIONS; i++) {
1631 		mn = DISKMINOR(device_unit(vnd->sc_dev), i);
1632 		vdevgone(bmaj, mn, mn, VBLK);
1633 		if (mn != myminor) /* XXX avoid to kill own vnode */
1634 			vdevgone(cmaj, mn, mn, VCHR);
1635 	}
1636 
1637 	if ((vnd->sc_flags & VNF_READONLY) == 0)
1638 		fflags |= FWRITE;
1639 
1640 	s = splbio();
1641 	bufq_drain(vnd->sc_tab);
1642 	splx(s);
1643 
1644 	vnd->sc_flags |= VNF_VUNCONF;
1645 	wakeup(&vnd->sc_tab);
1646 	while (vnd->sc_flags & VNF_KTHREAD)
1647 		tsleep(&vnd->sc_kthread, PRIBIO, "vnthr", 0);
1648 
1649 #ifdef VND_COMPRESSION
1650 	/* free the compressed file buffers */
1651 	if (vnd->sc_flags & VNF_COMP) {
1652 		if (vnd->sc_comp_offsets) {
1653 			free(vnd->sc_comp_offsets, M_DEVBUF);
1654 			vnd->sc_comp_offsets = NULL;
1655 		}
1656 		if (vnd->sc_comp_buff) {
1657 			free(vnd->sc_comp_buff, M_DEVBUF);
1658 			vnd->sc_comp_buff = NULL;
1659 		}
1660 		if (vnd->sc_comp_decombuf) {
1661 			free(vnd->sc_comp_decombuf, M_DEVBUF);
1662 			vnd->sc_comp_decombuf = NULL;
1663 		}
1664 	}
1665 #endif /* VND_COMPRESSION */
1666 	vnd->sc_flags &=
1667 	    ~(VNF_INITED | VNF_READONLY | VNF_VLABEL
1668 	      | VNF_VUNCONF | VNF_COMP | VNF_CLEARING);
1669 	if (vp == NULL)
1670 		panic("vndclear: null vp");
1671 	(void) vn_close(vp, fflags, vnd->sc_cred);
1672 	kauth_cred_free(vnd->sc_cred);
1673 	vnd->sc_vp = NULL;
1674 	vnd->sc_cred = NULL;
1675 	vnd->sc_size = 0;
1676 }
1677 
1678 static int
1679 vndsize(dev_t dev)
1680 {
1681 	struct vnd_softc *sc;
1682 	struct disklabel *lp;
1683 	int part, unit, omask;
1684 	int size;
1685 
1686 	unit = vndunit(dev);
1687 	sc = device_lookup_private(&vnd_cd, unit);
1688 	if (sc == NULL)
1689 		return -1;
1690 
1691 	if ((sc->sc_flags & VNF_INITED) == 0)
1692 		return -1;
1693 
1694 	part = DISKPART(dev);
1695 	omask = sc->sc_dkdev.dk_openmask & (1 << part);
1696 	lp = sc->sc_dkdev.dk_label;
1697 
1698 	if (omask == 0 && vndopen(dev, 0, S_IFBLK, curlwp))	/* XXX */
1699 		return -1;
1700 
1701 	if (lp->d_partitions[part].p_fstype != FS_SWAP)
1702 		size = -1;
1703 	else
1704 		size = lp->d_partitions[part].p_size *
1705 		    (lp->d_secsize / DEV_BSIZE);
1706 
1707 	if (omask == 0 && vndclose(dev, 0, S_IFBLK, curlwp))	/* XXX */
1708 		return -1;
1709 
1710 	return size;
1711 }
1712 
1713 static int
1714 vnddump(dev_t dev, daddr_t blkno, void *va,
1715     size_t size)
1716 {
1717 
1718 	/* Not implemented. */
1719 	return ENXIO;
1720 }
1721 
1722 static void
1723 vndgetdefaultlabel(struct vnd_softc *sc, struct disklabel *lp)
1724 {
1725 	struct vndgeom *vng = &sc->sc_geom;
1726 	struct partition *pp;
1727 
1728 	memset(lp, 0, sizeof(*lp));
1729 
1730 	lp->d_secperunit = sc->sc_size / (vng->vng_secsize / DEV_BSIZE);
1731 	lp->d_secsize = vng->vng_secsize;
1732 	lp->d_nsectors = vng->vng_nsectors;
1733 	lp->d_ntracks = vng->vng_ntracks;
1734 	lp->d_ncylinders = vng->vng_ncylinders;
1735 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1736 
1737 	strncpy(lp->d_typename, "vnd", sizeof(lp->d_typename));
1738 	lp->d_type = DTYPE_VND;
1739 	strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
1740 	lp->d_rpm = 3600;
1741 	lp->d_interleave = 1;
1742 	lp->d_flags = 0;
1743 
1744 	pp = &lp->d_partitions[RAW_PART];
1745 	pp->p_offset = 0;
1746 	pp->p_size = lp->d_secperunit;
1747 	pp->p_fstype = FS_UNUSED;
1748 	lp->d_npartitions = RAW_PART + 1;
1749 
1750 	lp->d_magic = DISKMAGIC;
1751 	lp->d_magic2 = DISKMAGIC;
1752 	lp->d_checksum = dkcksum(lp);
1753 }
1754 
1755 /*
1756  * Read the disklabel from a vnd.  If one is not present, create a fake one.
1757  */
1758 static void
1759 vndgetdisklabel(dev_t dev, struct vnd_softc *sc)
1760 {
1761 	const char *errstring;
1762 	struct disklabel *lp = sc->sc_dkdev.dk_label;
1763 	struct cpu_disklabel *clp = sc->sc_dkdev.dk_cpulabel;
1764 	int i;
1765 
1766 	memset(clp, 0, sizeof(*clp));
1767 
1768 	vndgetdefaultlabel(sc, lp);
1769 
1770 	/*
1771 	 * Call the generic disklabel extraction routine.
1772 	 */
1773 	errstring = readdisklabel(VNDLABELDEV(dev), vndstrategy, lp, clp);
1774 	if (errstring) {
1775 		/*
1776 		 * Lack of disklabel is common, but we print the warning
1777 		 * anyway, since it might contain other useful information.
1778 		 */
1779 		aprint_normal_dev(sc->sc_dev, "%s\n", errstring);
1780 
1781 		/*
1782 		 * For historical reasons, if there's no disklabel
1783 		 * present, all partitions must be FS_BSDFFS and
1784 		 * occupy the entire disk.
1785 		 */
1786 		for (i = 0; i < MAXPARTITIONS; i++) {
1787 			/*
1788 			 * Don't wipe out port specific hack (such as
1789 			 * dos partition hack of i386 port).
1790 			 */
1791 			if (lp->d_partitions[i].p_size != 0)
1792 				continue;
1793 
1794 			lp->d_partitions[i].p_size = lp->d_secperunit;
1795 			lp->d_partitions[i].p_offset = 0;
1796 			lp->d_partitions[i].p_fstype = FS_BSDFFS;
1797 		}
1798 
1799 		strncpy(lp->d_packname, "default label",
1800 		    sizeof(lp->d_packname));
1801 
1802 		lp->d_npartitions = MAXPARTITIONS;
1803 		lp->d_checksum = dkcksum(lp);
1804 	}
1805 
1806 	/* In-core label now valid. */
1807 	sc->sc_flags |= VNF_VLABEL;
1808 }
1809 
1810 /*
1811  * Wait interruptibly for an exclusive lock.
1812  *
1813  * XXX
1814  * Several drivers do this; it should be abstracted and made MP-safe.
1815  */
1816 static int
1817 vndlock(struct vnd_softc *sc)
1818 {
1819 	int error;
1820 
1821 	while ((sc->sc_flags & VNF_LOCKED) != 0) {
1822 		sc->sc_flags |= VNF_WANTED;
1823 		if ((error = tsleep(sc, PRIBIO | PCATCH, "vndlck", 0)) != 0)
1824 			return error;
1825 	}
1826 	sc->sc_flags |= VNF_LOCKED;
1827 	return 0;
1828 }
1829 
1830 /*
1831  * Unlock and wake up any waiters.
1832  */
1833 static void
1834 vndunlock(struct vnd_softc *sc)
1835 {
1836 
1837 	sc->sc_flags &= ~VNF_LOCKED;
1838 	if ((sc->sc_flags & VNF_WANTED) != 0) {
1839 		sc->sc_flags &= ~VNF_WANTED;
1840 		wakeup(sc);
1841 	}
1842 }
1843 
1844 #ifdef VND_COMPRESSION
1845 /* compressed file read */
1846 static void
1847 compstrategy(struct buf *bp, off_t bn)
1848 {
1849 	int error;
1850 	int unit = vndunit(bp->b_dev);
1851 	struct vnd_softc *vnd =
1852 	    device_lookup_private(&vnd_cd, unit);
1853 	u_int32_t comp_block;
1854 	struct uio auio;
1855 	char *addr;
1856 	int s;
1857 
1858 	/* set up constants for data move */
1859 	auio.uio_rw = UIO_READ;
1860 	UIO_SETUP_SYSSPACE(&auio);
1861 
1862 	/* read, and transfer the data */
1863 	addr = bp->b_data;
1864 	bp->b_resid = bp->b_bcount;
1865 	s = splbio();
1866 	while (bp->b_resid > 0) {
1867 		unsigned length;
1868 		size_t length_in_buffer;
1869 		u_int32_t offset_in_buffer;
1870 		struct iovec aiov;
1871 
1872 		/* calculate the compressed block number */
1873 		comp_block = bn / (off_t)vnd->sc_comp_blksz;
1874 
1875 		/* check for good block number */
1876 		if (comp_block >= vnd->sc_comp_numoffs) {
1877 			bp->b_error = EINVAL;
1878 			splx(s);
1879 			return;
1880 		}
1881 
1882 		/* read in the compressed block, if not in buffer */
1883 		if (comp_block != vnd->sc_comp_buffblk) {
1884 			length = vnd->sc_comp_offsets[comp_block + 1] -
1885 			    vnd->sc_comp_offsets[comp_block];
1886 			vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1887 			error = vn_rdwr(UIO_READ, vnd->sc_vp, vnd->sc_comp_buff,
1888 			    length, vnd->sc_comp_offsets[comp_block],
1889 			    UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, vnd->sc_cred,
1890 			    NULL, NULL);
1891 			if (error) {
1892 				bp->b_error = error;
1893 				VOP_UNLOCK(vnd->sc_vp);
1894 				splx(s);
1895 				return;
1896 			}
1897 			/* uncompress the buffer */
1898 			vnd->sc_comp_stream.next_in = vnd->sc_comp_buff;
1899 			vnd->sc_comp_stream.avail_in = length;
1900 			vnd->sc_comp_stream.next_out = vnd->sc_comp_decombuf;
1901 			vnd->sc_comp_stream.avail_out = vnd->sc_comp_blksz;
1902 			inflateReset(&vnd->sc_comp_stream);
1903 			error = inflate(&vnd->sc_comp_stream, Z_FINISH);
1904 			if (error != Z_STREAM_END) {
1905 				if (vnd->sc_comp_stream.msg)
1906 					aprint_normal_dev(vnd->sc_dev,
1907 					    "compressed file, %s\n",
1908 					    vnd->sc_comp_stream.msg);
1909 				bp->b_error = EBADMSG;
1910 				VOP_UNLOCK(vnd->sc_vp);
1911 				splx(s);
1912 				return;
1913 			}
1914 			vnd->sc_comp_buffblk = comp_block;
1915 			VOP_UNLOCK(vnd->sc_vp);
1916 		}
1917 
1918 		/* transfer the usable uncompressed data */
1919 		offset_in_buffer = bn % (off_t)vnd->sc_comp_blksz;
1920 		length_in_buffer = vnd->sc_comp_blksz - offset_in_buffer;
1921 		if (length_in_buffer > bp->b_resid)
1922 			length_in_buffer = bp->b_resid;
1923 		auio.uio_iov = &aiov;
1924 		auio.uio_iovcnt = 1;
1925 		aiov.iov_base = addr;
1926 		aiov.iov_len = length_in_buffer;
1927 		auio.uio_resid = aiov.iov_len;
1928 		auio.uio_offset = 0;
1929 		error = uiomove(vnd->sc_comp_decombuf + offset_in_buffer,
1930 		    length_in_buffer, &auio);
1931 		if (error) {
1932 			bp->b_error = error;
1933 			splx(s);
1934 			return;
1935 		}
1936 
1937 		bn += length_in_buffer;
1938 		addr += length_in_buffer;
1939 		bp->b_resid -= length_in_buffer;
1940 	}
1941 	splx(s);
1942 }
1943 
1944 /* compression memory allocation routines */
1945 static void *
1946 vnd_alloc(void *aux, u_int items, u_int siz)
1947 {
1948 	return malloc(items * siz, M_TEMP, M_NOWAIT);
1949 }
1950 
1951 static void
1952 vnd_free(void *aux, void *ptr)
1953 {
1954 	free(ptr, M_TEMP);
1955 }
1956 #endif /* VND_COMPRESSION */
1957 
1958 static void
1959 vnd_set_properties(struct vnd_softc *vnd)
1960 {
1961 	prop_dictionary_t disk_info, odisk_info, geom;
1962 
1963 	disk_info = prop_dictionary_create();
1964 
1965 	geom = prop_dictionary_create();
1966 
1967 	prop_dictionary_set_uint64(geom, "sectors-per-unit",
1968 	    vnd->sc_geom.vng_nsectors * vnd->sc_geom.vng_ntracks *
1969 	    vnd->sc_geom.vng_ncylinders);
1970 
1971 	prop_dictionary_set_uint32(geom, "sector-size",
1972 	    vnd->sc_geom.vng_secsize);
1973 
1974 	prop_dictionary_set_uint16(geom, "sectors-per-track",
1975 	    vnd->sc_geom.vng_nsectors);
1976 
1977 	prop_dictionary_set_uint16(geom, "tracks-per-cylinder",
1978 	    vnd->sc_geom.vng_ntracks);
1979 
1980 	prop_dictionary_set_uint64(geom, "cylinders-per-unit",
1981 	    vnd->sc_geom.vng_ncylinders);
1982 
1983 	prop_dictionary_set(disk_info, "geometry", geom);
1984 	prop_object_release(geom);
1985 
1986 	prop_dictionary_set(device_properties(vnd->sc_dev),
1987 	    "disk-info", disk_info);
1988 
1989 	/*
1990 	 * Don't release disk_info here; we keep a reference to it.
1991 	 * disk_detach() will release it when we go away.
1992 	 */
1993 
1994 	odisk_info = vnd->sc_dkdev.dk_info;
1995 	vnd->sc_dkdev.dk_info = disk_info;
1996 	if (odisk_info)
1997 		prop_object_release(odisk_info);
1998 }
1999 
2000 #ifdef _MODULE
2001 
2002 #include <sys/module.h>
2003 
2004 MODULE(MODULE_CLASS_DRIVER, vnd, NULL);
2005 CFDRIVER_DECL(vnd, DV_DISK, NULL);
2006 
2007 static int
2008 vnd_modcmd(modcmd_t cmd, void *arg)
2009 {
2010 	int bmajor = -1, cmajor = -1,  error = 0;
2011 
2012 	switch (cmd) {
2013 	case MODULE_CMD_INIT:
2014 		error = config_cfdriver_attach(&vnd_cd);
2015 		if (error)
2016 			break;
2017 
2018 		error = config_cfattach_attach(vnd_cd.cd_name, &vnd_ca);
2019 	        if (error) {
2020 			config_cfdriver_detach(&vnd_cd);
2021 			aprint_error("%s: unable to register cfattach\n",
2022 			    vnd_cd.cd_name);
2023 			break;
2024 		}
2025 
2026 		error = devsw_attach("vnd", &vnd_bdevsw, &bmajor,
2027 		    &vnd_cdevsw, &cmajor);
2028 		if (error) {
2029 			config_cfattach_detach(vnd_cd.cd_name, &vnd_ca);
2030 			config_cfdriver_detach(&vnd_cd);
2031 			break;
2032 		}
2033 
2034 		break;
2035 
2036 	case MODULE_CMD_FINI:
2037 		error = config_cfattach_detach(vnd_cd.cd_name, &vnd_ca);
2038 		if (error)
2039 			break;
2040 		config_cfdriver_detach(&vnd_cd);
2041 		devsw_detach(&vnd_bdevsw, &vnd_cdevsw);
2042 		break;
2043 
2044 	case MODULE_CMD_STAT:
2045 		return ENOTTY;
2046 
2047 	default:
2048 		return ENOTTY;
2049 	}
2050 
2051 	return error;
2052 }
2053 
2054 #endif
2055