xref: /netbsd-src/sys/dev/vnd.c (revision b5677b36047b601b9addaaa494a58ceae82c2a6c)
1 /*	$NetBSD: vnd.c,v 1.196 2009/03/18 16:00:17 cegger 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 1.196 2009/03/18 16:00:17 cegger Exp $");
134 
135 #if defined(_KERNEL_OPT)
136 #include "fs_nfs.h"
137 #include "opt_vnd.h"
138 #endif
139 
140 #include <sys/param.h>
141 #include <sys/systm.h>
142 #include <sys/namei.h>
143 #include <sys/proc.h>
144 #include <sys/kthread.h>
145 #include <sys/errno.h>
146 #include <sys/buf.h>
147 #include <sys/bufq.h>
148 #include <sys/malloc.h>
149 #include <sys/ioctl.h>
150 #include <sys/disklabel.h>
151 #include <sys/device.h>
152 #include <sys/disk.h>
153 #include <sys/stat.h>
154 #include <sys/mount.h>
155 #include <sys/vnode.h>
156 #include <sys/file.h>
157 #include <sys/uio.h>
158 #include <sys/conf.h>
159 #include <sys/kauth.h>
160 
161 #include <net/zlib.h>
162 
163 #include <miscfs/genfs/genfs.h>
164 #include <miscfs/specfs/specdev.h>
165 
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	vndsetcred(struct vnd_softc *, kauth_cred_t);
202 static void	vndthrottle(struct vnd_softc *, struct vnode *);
203 static void	vndiodone(struct buf *);
204 #if 0
205 static void	vndshutdown(void);
206 #endif
207 
208 static void	vndgetdefaultlabel(struct vnd_softc *, struct disklabel *);
209 static void	vndgetdisklabel(dev_t, struct vnd_softc *);
210 
211 static int	vndlock(struct vnd_softc *);
212 static void	vndunlock(struct vnd_softc *);
213 #ifdef VND_COMPRESSION
214 static void	compstrategy(struct buf *, off_t);
215 static void	*vnd_alloc(void *, u_int, u_int);
216 static void	vnd_free(void *, void *);
217 #endif /* VND_COMPRESSION */
218 
219 static void	vndthread(void *);
220 static bool	vnode_has_op(const struct vnode *, int);
221 static void	handle_with_rdwr(struct vnd_softc *, const struct buf *,
222 		    struct buf *);
223 static void	handle_with_strategy(struct vnd_softc *, const struct buf *,
224 		    struct buf *);
225 static void	vnd_set_properties(struct vnd_softc *);
226 
227 static dev_type_open(vndopen);
228 static dev_type_close(vndclose);
229 static dev_type_read(vndread);
230 static dev_type_write(vndwrite);
231 static dev_type_ioctl(vndioctl);
232 static dev_type_strategy(vndstrategy);
233 static dev_type_dump(vnddump);
234 static dev_type_size(vndsize);
235 
236 const struct bdevsw vnd_bdevsw = {
237 	vndopen, vndclose, vndstrategy, vndioctl, vnddump, vndsize, D_DISK
238 };
239 
240 const struct cdevsw vnd_cdevsw = {
241 	vndopen, vndclose, vndread, vndwrite, vndioctl,
242 	nostop, notty, nopoll, nommap, nokqfilter, D_DISK
243 };
244 
245 static int	vnd_match(device_t, cfdata_t, void *);
246 static void	vnd_attach(device_t, device_t, void *);
247 static int	vnd_detach(device_t, int);
248 
249 CFATTACH_DECL_NEW(vnd, sizeof(struct vnd_softc),
250     vnd_match, vnd_attach, vnd_detach, NULL);
251 extern struct cfdriver vnd_cd;
252 
253 static struct vnd_softc	*vnd_spawn(int);
254 int	vnd_destroy(device_t);
255 
256 void
257 vndattach(int num)
258 {
259 	int error;
260 
261 	error = config_cfattach_attach(vnd_cd.cd_name, &vnd_ca);
262 	if (error)
263 		aprint_error("%s: unable to register cfattach\n",
264 		    vnd_cd.cd_name);
265 }
266 
267 static int
268 vnd_match(device_t self, cfdata_t cfdata, void *aux)
269 {
270 
271 	return 1;
272 }
273 
274 static void
275 vnd_attach(device_t parent, device_t self, void *aux)
276 {
277 	struct vnd_softc *sc = device_private(self);
278 
279 	sc->sc_dev = self;
280 	sc->sc_comp_offsets = NULL;
281 	sc->sc_comp_buff = NULL;
282 	sc->sc_comp_decombuf = NULL;
283 	bufq_alloc(&sc->sc_tab, "disksort", BUFQ_SORT_RAWBLOCK);
284 	disk_init(&sc->sc_dkdev, device_xname(self), NULL);
285 	if (!pmf_device_register(self, NULL, NULL))
286 		aprint_error_dev(self, "couldn't establish power handler\n");
287 }
288 
289 static int
290 vnd_detach(device_t self, int flags)
291 {
292 	struct vnd_softc *sc = device_private(self);
293 	if (sc->sc_flags & VNF_INITED)
294 		return EBUSY;
295 
296 	pmf_device_deregister(self);
297 	bufq_free(sc->sc_tab);
298 	disk_destroy(&sc->sc_dkdev);
299 
300 	return 0;
301 }
302 
303 static struct vnd_softc *
304 vnd_spawn(int unit)
305 {
306 	struct cfdata *cf;
307 
308 	cf = malloc(sizeof(*cf), M_DEVBUF, M_WAITOK);
309 	cf->cf_name = vnd_cd.cd_name;
310 	cf->cf_atname = vnd_cd.cd_name;
311 	cf->cf_unit = unit;
312 	cf->cf_fstate = FSTATE_STAR;
313 
314 	return device_private(config_attach_pseudo(cf));
315 }
316 
317 int
318 vnd_destroy(device_t dev)
319 {
320 	int error;
321 	cfdata_t cf;
322 
323 	cf = device_cfdata(dev);
324 	error = config_detach(dev, DETACH_QUIET);
325 	if (error)
326 		return error;
327 	free(cf, M_DEVBUF);
328 	return 0;
329 }
330 
331 static int
332 vndopen(dev_t dev, int flags, int mode, struct lwp *l)
333 {
334 	int unit = vndunit(dev);
335 	struct vnd_softc *sc;
336 	int error = 0, part, pmask;
337 	struct disklabel *lp;
338 
339 #ifdef DEBUG
340 	if (vnddebug & VDB_FOLLOW)
341 		printf("vndopen(0x%"PRIx64", 0x%x, 0x%x, %p)\n", dev, flags, mode, l);
342 #endif
343 	sc = device_lookup_private(&vnd_cd, unit);
344 	if (sc == NULL) {
345 		sc = vnd_spawn(unit);
346 		if (sc == NULL)
347 			return ENOMEM;
348 	}
349 
350 	if ((error = vndlock(sc)) != 0)
351 		return (error);
352 
353 	lp = sc->sc_dkdev.dk_label;
354 
355 	part = DISKPART(dev);
356 	pmask = (1 << part);
357 
358 	/*
359 	 * If we're initialized, check to see if there are any other
360 	 * open partitions.  If not, then it's safe to update the
361 	 * in-core disklabel.  Only read the disklabel if it is
362 	 * not already valid.
363 	 */
364 	if ((sc->sc_flags & (VNF_INITED|VNF_VLABEL)) == VNF_INITED &&
365 	    sc->sc_dkdev.dk_openmask == 0)
366 		vndgetdisklabel(dev, sc);
367 
368 	/* Check that the partitions exists. */
369 	if (part != RAW_PART) {
370 		if (((sc->sc_flags & VNF_INITED) == 0) ||
371 		    ((part >= lp->d_npartitions) ||
372 		     (lp->d_partitions[part].p_fstype == FS_UNUSED))) {
373 			error = ENXIO;
374 			goto done;
375 		}
376 	}
377 
378 	/* Prevent our unit from being unconfigured while open. */
379 	switch (mode) {
380 	case S_IFCHR:
381 		sc->sc_dkdev.dk_copenmask |= pmask;
382 		break;
383 
384 	case S_IFBLK:
385 		sc->sc_dkdev.dk_bopenmask |= pmask;
386 		break;
387 	}
388 	sc->sc_dkdev.dk_openmask =
389 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
390 
391  done:
392 	vndunlock(sc);
393 	return (error);
394 }
395 
396 static int
397 vndclose(dev_t dev, int flags, int mode, struct lwp *l)
398 {
399 	int unit = vndunit(dev);
400 	struct vnd_softc *sc;
401 	int error = 0, part;
402 
403 #ifdef DEBUG
404 	if (vnddebug & VDB_FOLLOW)
405 		printf("vndclose(0x%"PRIx64", 0x%x, 0x%x, %p)\n", dev, flags, mode, l);
406 #endif
407 	sc = device_lookup_private(&vnd_cd, unit);
408 	if (sc == NULL)
409 		return ENXIO;
410 
411 	if ((error = vndlock(sc)) != 0)
412 		return (error);
413 
414 	part = DISKPART(dev);
415 
416 	/* ...that much closer to allowing unconfiguration... */
417 	switch (mode) {
418 	case S_IFCHR:
419 		sc->sc_dkdev.dk_copenmask &= ~(1 << part);
420 		break;
421 
422 	case S_IFBLK:
423 		sc->sc_dkdev.dk_bopenmask &= ~(1 << part);
424 		break;
425 	}
426 	sc->sc_dkdev.dk_openmask =
427 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
428 
429 	vndunlock(sc);
430 
431 	if ((sc->sc_flags & VNF_INITED) == 0) {
432 		if ((error = vnd_destroy(sc->sc_dev)) != 0) {
433 			aprint_error_dev(sc->sc_dev,
434 			    "unable to detach instance\n");
435 			return error;
436 		}
437 	}
438 
439 	return (0);
440 }
441 
442 /*
443  * Queue the request, and wakeup the kernel thread to handle it.
444  */
445 static void
446 vndstrategy(struct buf *bp)
447 {
448 	int unit = vndunit(bp->b_dev);
449 	struct vnd_softc *vnd =
450 	    device_lookup_private(&vnd_cd, unit);
451 	struct disklabel *lp;
452 	daddr_t blkno;
453 	int s = splbio();
454 
455 	if (vnd == NULL) {
456 		bp->b_error = ENXIO;
457 		goto done;
458 	}
459 	lp = vnd->sc_dkdev.dk_label;
460 
461 	if ((vnd->sc_flags & VNF_INITED) == 0) {
462 		bp->b_error = ENXIO;
463 		goto done;
464 	}
465 
466 	/*
467 	 * The transfer must be a whole number of blocks.
468 	 */
469 	if ((bp->b_bcount % lp->d_secsize) != 0) {
470 		bp->b_error = EINVAL;
471 		goto done;
472 	}
473 
474 	/*
475 	 * check if we're read-only.
476 	 */
477 	if ((vnd->sc_flags & VNF_READONLY) && !(bp->b_flags & B_READ)) {
478 		bp->b_error = EACCES;
479 		goto done;
480 	}
481 
482 	/* If it's a nil transfer, wake up the top half now. */
483 	if (bp->b_bcount == 0) {
484 		goto done;
485 	}
486 
487 	/*
488 	 * Do bounds checking and adjust transfer.  If there's an error,
489 	 * the bounds check will flag that for us.
490 	 */
491 	if (DISKPART(bp->b_dev) == RAW_PART) {
492 		if (bounds_check_with_mediasize(bp, DEV_BSIZE,
493 		    vnd->sc_size) <= 0)
494 			goto done;
495 	} else {
496 		if (bounds_check_with_label(&vnd->sc_dkdev,
497 		    bp, vnd->sc_flags & (VNF_WLABEL|VNF_LABELLING)) <= 0)
498 			goto done;
499 	}
500 
501 	/*
502 	 * Put the block number in terms of the logical blocksize
503 	 * of the "device".
504 	 */
505 
506 	blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
507 
508 	/*
509 	 * Translate the partition-relative block number to an absolute.
510 	 */
511 	if (DISKPART(bp->b_dev) != RAW_PART) {
512 		struct partition *pp;
513 
514 		pp = &vnd->sc_dkdev.dk_label->d_partitions[
515 		    DISKPART(bp->b_dev)];
516 		blkno += pp->p_offset;
517 	}
518 	bp->b_rawblkno = blkno;
519 
520 #ifdef DEBUG
521 	if (vnddebug & VDB_FOLLOW)
522 		printf("vndstrategy(%p): unit %d\n", bp, unit);
523 #endif
524 	bufq_put(vnd->sc_tab, bp);
525 	wakeup(&vnd->sc_tab);
526 	splx(s);
527 	return;
528 
529 done:
530 	bp->b_resid = bp->b_bcount;
531 	biodone(bp);
532 	splx(s);
533 }
534 
535 static bool
536 vnode_has_strategy(struct vnd_softc *vnd)
537 {
538 	return vnode_has_op(vnd->sc_vp, VOFFSET(vop_bmap)) &&
539 	    vnode_has_op(vnd->sc_vp, VOFFSET(vop_strategy));
540 }
541 
542 /* XXX this function needs a reliable check to detect
543  * sparse files. Otherwise, bmap/strategy may be used
544  * and fail on non-allocated blocks. VOP_READ/VOP_WRITE
545  * works on sparse files.
546  */
547 #if notyet
548 static bool
549 vnode_strategy_probe(struct vnd_softc *vnd)
550 {
551 	int error;
552 	daddr_t nbn;
553 
554 	if (!vnode_has_strategy(vnd))
555 		return false;
556 
557 	/* Convert the first logical block number to its
558 	 * physical block number.
559 	 */
560 	error = 0;
561 	vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY | LK_CANRECURSE);
562 	error = VOP_BMAP(vnd->sc_vp, 0, NULL, &nbn, NULL);
563 	VOP_UNLOCK(vnd->sc_vp, 0);
564 
565 	/* Test if that worked. */
566 	if (error == 0 && (long)nbn == -1)
567 		return false;
568 
569 	return true;
570 }
571 #endif
572 
573 static void
574 vndthread(void *arg)
575 {
576 	struct vnd_softc *vnd = arg;
577 	bool usestrategy;
578 	int s;
579 
580 	/* Determine whether we can *use* VOP_BMAP and VOP_STRATEGY to
581 	 * directly access the backing vnode.  If we can, use these two
582 	 * operations to avoid messing with the local buffer cache.
583 	 * Otherwise fall back to regular VOP_READ/VOP_WRITE operations
584 	 * which are guaranteed to work with any file system. */
585 	usestrategy = vnode_has_strategy(vnd);
586 
587 #ifdef DEBUG
588 	if (vnddebug & VDB_INIT)
589 		printf("vndthread: vp %p, %s\n", vnd->sc_vp,
590 		    usestrategy ?
591 		    "using bmap/strategy operations" :
592 		    "using read/write operations");
593 #endif
594 
595 	s = splbio();
596 	vnd->sc_flags |= VNF_KTHREAD;
597 	wakeup(&vnd->sc_kthread);
598 
599 	/*
600 	 * Dequeue requests and serve them depending on the available
601 	 * vnode operations.
602 	 */
603 	while ((vnd->sc_flags & VNF_VUNCONF) == 0) {
604 		struct vndxfer *vnx;
605 		int flags;
606 		struct buf *obp;
607 		struct buf *bp;
608 
609 		obp = bufq_get(vnd->sc_tab);
610 		if (obp == NULL) {
611 			tsleep(&vnd->sc_tab, PRIBIO, "vndbp", 0);
612 			continue;
613 		};
614 		splx(s);
615 		flags = obp->b_flags;
616 #ifdef DEBUG
617 		if (vnddebug & VDB_FOLLOW)
618 			printf("vndthread(%p)\n", obp);
619 #endif
620 
621 		if (vnd->sc_vp->v_mount == NULL) {
622 			obp->b_error = ENXIO;
623 			goto done;
624 		}
625 #ifdef VND_COMPRESSION
626 		/* handle a compressed read */
627 		if ((flags & B_READ) != 0 && (vnd->sc_flags & VNF_COMP)) {
628 			off_t bn;
629 
630 			/* Convert to a byte offset within the file. */
631 			bn = obp->b_rawblkno *
632 			    vnd->sc_dkdev.dk_label->d_secsize;
633 
634 			compstrategy(obp, bn);
635 			goto done;
636 		}
637 #endif /* VND_COMPRESSION */
638 
639 		/*
640 		 * Allocate a header for this transfer and link it to the
641 		 * buffer
642 		 */
643 		s = splbio();
644 		vnx = VND_GETXFER(vnd);
645 		splx(s);
646 		vnx->vx_vnd = vnd;
647 
648 		s = splbio();
649 		while (vnd->sc_active >= vnd->sc_maxactive) {
650 			tsleep(&vnd->sc_tab, PRIBIO, "vndac", 0);
651 		}
652 		vnd->sc_active++;
653 		splx(s);
654 
655 		/* Instrumentation. */
656 		disk_busy(&vnd->sc_dkdev);
657 
658 		bp = &vnx->vx_buf;
659 		buf_init(bp);
660 		bp->b_flags = (obp->b_flags & B_READ);
661 		bp->b_oflags = obp->b_oflags;
662 		bp->b_cflags = obp->b_cflags;
663 		bp->b_iodone = vndiodone;
664 		bp->b_private = obp;
665 		bp->b_vp = vnd->sc_vp;
666 		bp->b_objlock = &bp->b_vp->v_interlock;
667 		bp->b_data = obp->b_data;
668 		bp->b_bcount = obp->b_bcount;
669 		BIO_COPYPRIO(bp, obp);
670 
671 		/* Handle the request using the appropriate operations. */
672 		if (usestrategy)
673 			handle_with_strategy(vnd, obp, bp);
674 		else
675 			handle_with_rdwr(vnd, obp, bp);
676 
677 		s = splbio();
678 		continue;
679 
680 done:
681 		biodone(obp);
682 		s = splbio();
683 	}
684 
685 	vnd->sc_flags &= (~VNF_KTHREAD | VNF_VUNCONF);
686 	wakeup(&vnd->sc_kthread);
687 	splx(s);
688 	kthread_exit(0);
689 }
690 
691 /*
692  * Checks if the given vnode supports the requested operation.
693  * The operation is specified the offset returned by VOFFSET.
694  *
695  * XXX The test below used to determine this is quite fragile
696  * because it relies on the file system to use genfs to specify
697  * unimplemented operations.  There might be another way to do
698  * it more cleanly.
699  */
700 static bool
701 vnode_has_op(const struct vnode *vp, int opoffset)
702 {
703 	int (*defaultp)(void *);
704 	int (*opp)(void *);
705 
706 	defaultp = vp->v_op[VOFFSET(vop_default)];
707 	opp = vp->v_op[opoffset];
708 
709 	return opp != defaultp && opp != genfs_eopnotsupp &&
710 	    opp != genfs_badop && opp != genfs_nullop;
711 }
712 
713 /*
714  * Handes the read/write request given in 'bp' using the vnode's VOP_READ
715  * and VOP_WRITE operations.
716  *
717  * 'obp' is a pointer to the original request fed to the vnd device.
718  */
719 static void
720 handle_with_rdwr(struct vnd_softc *vnd, const struct buf *obp, struct buf *bp)
721 {
722 	bool doread;
723 	off_t offset;
724 	size_t resid;
725 	struct vnode *vp;
726 
727 	doread = bp->b_flags & B_READ;
728 	offset = obp->b_rawblkno * vnd->sc_dkdev.dk_label->d_secsize;
729 	vp = vnd->sc_vp;
730 
731 #if defined(DEBUG)
732 	if (vnddebug & VDB_IO)
733 		printf("vnd (rdwr): vp %p, %s, rawblkno 0x%" PRIx64
734 		    ", secsize %d, offset %" PRIu64
735 		    ", bcount %d\n",
736 		    vp, doread ? "read" : "write", obp->b_rawblkno,
737 		    vnd->sc_dkdev.dk_label->d_secsize, offset,
738 		    bp->b_bcount);
739 #endif
740 
741 	/* Issue the read or write operation. */
742 	bp->b_error =
743 	    vn_rdwr(doread ? UIO_READ : UIO_WRITE,
744 	    vp, bp->b_data, bp->b_bcount, offset,
745 	    UIO_SYSSPACE, 0, vnd->sc_cred, &resid, NULL);
746 	bp->b_resid = resid;
747 
748 	/* We need to increase the number of outputs on the vnode if
749 	 * there was any write to it. */
750 	if (!doread) {
751 		mutex_enter(&vp->v_interlock);
752 		vp->v_numoutput++;
753 		mutex_exit(&vp->v_interlock);
754 	}
755 
756 	biodone(bp);
757 }
758 
759 /*
760  * Handes the read/write request given in 'bp' using the vnode's VOP_BMAP
761  * and VOP_STRATEGY operations.
762  *
763  * 'obp' is a pointer to the original request fed to the vnd device.
764  */
765 static void
766 handle_with_strategy(struct vnd_softc *vnd, const struct buf *obp,
767     struct buf *bp)
768 {
769 	int bsize, error, flags, skipped;
770 	size_t resid, sz;
771 	off_t bn, offset;
772 	struct vnode *vp;
773 
774 	flags = obp->b_flags;
775 
776 	if (!(flags & B_READ)) {
777 		vp = bp->b_vp;
778 		mutex_enter(&vp->v_interlock);
779 		vp->v_numoutput++;
780 		mutex_exit(&vp->v_interlock);
781 	}
782 
783 	/* convert to a byte offset within the file. */
784 	bn = obp->b_rawblkno * vnd->sc_dkdev.dk_label->d_secsize;
785 
786 	bsize = vnd->sc_vp->v_mount->mnt_stat.f_iosize;
787 	skipped = 0;
788 
789 	/*
790 	 * Break the request into bsize pieces and feed them
791 	 * sequentially using VOP_BMAP/VOP_STRATEGY.
792 	 * We do it this way to keep from flooding NFS servers if we
793 	 * are connected to an NFS file.  This places the burden on
794 	 * the client rather than the server.
795 	 */
796 	error = 0;
797 	bp->b_resid = bp->b_bcount;
798 	for (offset = 0, resid = bp->b_resid; resid;
799 	    resid -= sz, offset += sz) {
800 		struct buf *nbp;
801 		daddr_t nbn;
802 		int off, nra;
803 
804 		nra = 0;
805 		vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY | LK_CANRECURSE);
806 		error = VOP_BMAP(vnd->sc_vp, bn / bsize, &vp, &nbn, &nra);
807 		VOP_UNLOCK(vnd->sc_vp, 0);
808 
809 		if (error == 0 && (long)nbn == -1)
810 			error = EIO;
811 
812 		/*
813 		 * If there was an error or a hole in the file...punt.
814 		 * Note that we may have to wait for any operations
815 		 * that we have already fired off before releasing
816 		 * the buffer.
817 		 *
818 		 * XXX we could deal with holes here but it would be
819 		 * a hassle (in the write case).
820 		 */
821 		if (error) {
822 			skipped += resid;
823 			break;
824 		}
825 
826 #ifdef DEBUG
827 		if (!dovndcluster)
828 			nra = 0;
829 #endif
830 
831 		off = bn % bsize;
832 		sz = MIN(((off_t)1 + nra) * bsize - off, resid);
833 #ifdef	DEBUG
834 		if (vnddebug & VDB_IO)
835 			printf("vndstrategy: vp %p/%p bn 0x%qx/0x%" PRIx64
836 			    " sz 0x%zx\n", vnd->sc_vp, vp, (long long)bn,
837 			    nbn, sz);
838 #endif
839 
840 		nbp = getiobuf(vp, true);
841 		nestiobuf_setup(bp, nbp, offset, sz);
842 		nbp->b_blkno = nbn + btodb(off);
843 
844 #if 0 /* XXX #ifdef DEBUG */
845 		if (vnddebug & VDB_IO)
846 			printf("vndstart(%ld): bp %p vp %p blkno "
847 			    "0x%" PRIx64 " flags %x addr %p cnt 0x%x\n",
848 			    (long) (vnd-vnd_softc), &nbp->vb_buf,
849 			    nbp->vb_buf.b_vp, nbp->vb_buf.b_blkno,
850 			    nbp->vb_buf.b_flags, nbp->vb_buf.b_data,
851 			    nbp->vb_buf.b_bcount);
852 #endif
853 		VOP_STRATEGY(vp, nbp);
854 		bn += sz;
855 	}
856 	nestiobuf_done(bp, skipped, error);
857 }
858 
859 static void
860 vndiodone(struct buf *bp)
861 {
862 	struct vndxfer *vnx = VND_BUFTOXFER(bp);
863 	struct vnd_softc *vnd = vnx->vx_vnd;
864 	struct buf *obp = bp->b_private;
865 
866 	KASSERT(&vnx->vx_buf == bp);
867 	KASSERT(vnd->sc_active > 0);
868 #ifdef DEBUG
869 	if (vnddebug & VDB_IO) {
870 		printf("vndiodone1: bp %p iodone: error %d\n",
871 		    bp, bp->b_error);
872 	}
873 #endif
874 	disk_unbusy(&vnd->sc_dkdev, bp->b_bcount - bp->b_resid,
875 	    (bp->b_flags & B_READ));
876 	vnd->sc_active--;
877 	if (vnd->sc_active == 0) {
878 		wakeup(&vnd->sc_tab);
879 	}
880 	obp->b_error = bp->b_error;
881 	obp->b_resid = bp->b_resid;
882 	buf_destroy(bp);
883 	VND_PUTXFER(vnd, vnx);
884 	biodone(obp);
885 }
886 
887 /* ARGSUSED */
888 static int
889 vndread(dev_t dev, struct uio *uio, int flags)
890 {
891 	int unit = vndunit(dev);
892 	struct vnd_softc *sc;
893 
894 #ifdef DEBUG
895 	if (vnddebug & VDB_FOLLOW)
896 		printf("vndread(0x%"PRIx64", %p)\n", dev, uio);
897 #endif
898 
899 	sc = device_lookup_private(&vnd_cd, unit);
900 	if (sc == NULL)
901 		return ENXIO;
902 
903 	if ((sc->sc_flags & VNF_INITED) == 0)
904 		return (ENXIO);
905 
906 	return (physio(vndstrategy, NULL, dev, B_READ, minphys, uio));
907 }
908 
909 /* ARGSUSED */
910 static int
911 vndwrite(dev_t dev, struct uio *uio, int flags)
912 {
913 	int unit = vndunit(dev);
914 	struct vnd_softc *sc;
915 
916 #ifdef DEBUG
917 	if (vnddebug & VDB_FOLLOW)
918 		printf("vndwrite(0x%"PRIx64", %p)\n", dev, uio);
919 #endif
920 
921 	sc = device_lookup_private(&vnd_cd, unit);
922 	if (sc == NULL)
923 		return ENXIO;
924 
925 	if ((sc->sc_flags & VNF_INITED) == 0)
926 		return (ENXIO);
927 
928 	return (physio(vndstrategy, NULL, dev, B_WRITE, minphys, uio));
929 }
930 
931 static int
932 vnd_cget(struct lwp *l, int unit, int *un, struct vattr *va)
933 {
934 	struct vnd_softc *vnd;
935 
936 	if (*un == -1)
937 		*un = unit;
938 	if (*un < 0)
939 		return EINVAL;
940 
941 	vnd = device_lookup_private(&vnd_cd, *un);
942 	if (vnd == NULL)
943 		return (*un >= vnd_cd.cd_ndevs) ? ENXIO : -1;
944 
945 	if ((vnd->sc_flags & VNF_INITED) == 0)
946 		return -1;
947 
948 	return VOP_GETATTR(vnd->sc_vp, va, l->l_cred);
949 }
950 
951 /* ARGSUSED */
952 static int
953 vndioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l)
954 {
955 	int unit = vndunit(dev);
956 	struct vnd_softc *vnd;
957 	struct vnd_ioctl *vio;
958 	struct vattr vattr;
959 	struct nameidata nd;
960 	int error, part, pmask;
961 	size_t geomsize;
962 	int fflags;
963 #ifdef __HAVE_OLD_DISKLABEL
964 	struct disklabel newlabel;
965 #endif
966 
967 #ifdef DEBUG
968 	if (vnddebug & VDB_FOLLOW)
969 		printf("vndioctl(0x%"PRIx64", 0x%lx, %p, 0x%x, %p): unit %d\n",
970 		    dev, cmd, data, flag, l->l_proc, unit);
971 #endif
972 	vnd = device_lookup_private(&vnd_cd, unit);
973 	if (vnd == NULL &&
974 #ifdef COMPAT_30
975 	    cmd != VNDIOOCGET &&
976 #endif
977 	    cmd != VNDIOCGET)
978 		return ENXIO;
979 	vio = (struct vnd_ioctl *)data;
980 
981 	/* Must be open for writes for these commands... */
982 	switch (cmd) {
983 	case VNDIOCSET:
984 	case VNDIOCCLR:
985 	case DIOCSDINFO:
986 	case DIOCWDINFO:
987 #ifdef __HAVE_OLD_DISKLABEL
988 	case ODIOCSDINFO:
989 	case ODIOCWDINFO:
990 #endif
991 	case DIOCKLABEL:
992 	case DIOCWLABEL:
993 		if ((flag & FWRITE) == 0)
994 			return (EBADF);
995 	}
996 
997 	/* Must be initialized for these... */
998 	switch (cmd) {
999 	case VNDIOCCLR:
1000 	case DIOCGDINFO:
1001 	case DIOCSDINFO:
1002 	case DIOCWDINFO:
1003 	case DIOCGPART:
1004 	case DIOCKLABEL:
1005 	case DIOCWLABEL:
1006 	case DIOCGDEFLABEL:
1007 	case DIOCCACHESYNC:
1008 #ifdef __HAVE_OLD_DISKLABEL
1009 	case ODIOCGDINFO:
1010 	case ODIOCSDINFO:
1011 	case ODIOCWDINFO:
1012 	case ODIOCGDEFLABEL:
1013 #endif
1014 		if ((vnd->sc_flags & VNF_INITED) == 0)
1015 			return (ENXIO);
1016 	}
1017 
1018 	switch (cmd) {
1019 	case VNDIOCSET:
1020 		if (vnd->sc_flags & VNF_INITED)
1021 			return (EBUSY);
1022 
1023 		if ((error = vndlock(vnd)) != 0)
1024 			return (error);
1025 
1026 		fflags = FREAD;
1027 		if ((vio->vnd_flags & VNDIOF_READONLY) == 0)
1028 			fflags |= FWRITE;
1029 		NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, vio->vnd_file);
1030 		if ((error = vn_open(&nd, fflags, 0)) != 0)
1031 			goto unlock_and_exit;
1032 		KASSERT(l);
1033 		error = VOP_GETATTR(nd.ni_vp, &vattr, l->l_cred);
1034 		if (!error && nd.ni_vp->v_type != VREG)
1035 			error = EOPNOTSUPP;
1036 		if (error) {
1037 			VOP_UNLOCK(nd.ni_vp, 0);
1038 			goto close_and_exit;
1039 		}
1040 
1041 		/* If using a compressed file, initialize its info */
1042 		/* (or abort with an error if kernel has no compression) */
1043 		if (vio->vnd_flags & VNF_COMP) {
1044 #ifdef VND_COMPRESSION
1045 			struct vnd_comp_header *ch;
1046 			int i;
1047 			u_int32_t comp_size;
1048 			u_int32_t comp_maxsize;
1049 
1050 			/* allocate space for compresed file header */
1051 			ch = malloc(sizeof(struct vnd_comp_header),
1052 			M_TEMP, M_WAITOK);
1053 
1054 			/* read compressed file header */
1055 			error = vn_rdwr(UIO_READ, nd.ni_vp, (void *)ch,
1056 			  sizeof(struct vnd_comp_header), 0, UIO_SYSSPACE,
1057 			  IO_UNIT|IO_NODELOCKED, l->l_cred, NULL, NULL);
1058 			if(error) {
1059 				free(ch, M_TEMP);
1060 				VOP_UNLOCK(nd.ni_vp, 0);
1061 				goto close_and_exit;
1062 			}
1063 
1064 			/* save some header info */
1065 			vnd->sc_comp_blksz = ntohl(ch->block_size);
1066 			/* note last offset is the file byte size */
1067 			vnd->sc_comp_numoffs = ntohl(ch->num_blocks)+1;
1068 			free(ch, M_TEMP);
1069 			if (vnd->sc_comp_blksz == 0 ||
1070 			    vnd->sc_comp_blksz % DEV_BSIZE !=0) {
1071 				VOP_UNLOCK(nd.ni_vp, 0);
1072 				error = EINVAL;
1073 				goto close_and_exit;
1074 			}
1075 			if(sizeof(struct vnd_comp_header) +
1076 			  sizeof(u_int64_t) * vnd->sc_comp_numoffs >
1077 			  vattr.va_size) {
1078 				VOP_UNLOCK(nd.ni_vp, 0);
1079 				error = EINVAL;
1080 				goto close_and_exit;
1081 			}
1082 
1083 			/* set decompressed file size */
1084 			vattr.va_size =
1085 			    ((u_quad_t)vnd->sc_comp_numoffs - 1) *
1086 			     (u_quad_t)vnd->sc_comp_blksz;
1087 
1088 			/* allocate space for all the compressed offsets */
1089 			vnd->sc_comp_offsets =
1090 			malloc(sizeof(u_int64_t) * vnd->sc_comp_numoffs,
1091 			M_DEVBUF, M_WAITOK);
1092 
1093 			/* read in the offsets */
1094 			error = vn_rdwr(UIO_READ, nd.ni_vp,
1095 			  (void *)vnd->sc_comp_offsets,
1096 			  sizeof(u_int64_t) * vnd->sc_comp_numoffs,
1097 			  sizeof(struct vnd_comp_header), UIO_SYSSPACE,
1098 			  IO_UNIT|IO_NODELOCKED, l->l_cred, NULL, NULL);
1099 			if(error) {
1100 				VOP_UNLOCK(nd.ni_vp, 0);
1101 				goto close_and_exit;
1102 			}
1103 			/*
1104 			 * find largest block size (used for allocation limit).
1105 			 * Also convert offset to native byte order.
1106 			 */
1107 			comp_maxsize = 0;
1108 			for (i = 0; i < vnd->sc_comp_numoffs - 1; i++) {
1109 				vnd->sc_comp_offsets[i] =
1110 				  be64toh(vnd->sc_comp_offsets[i]);
1111 				comp_size = be64toh(vnd->sc_comp_offsets[i + 1])
1112 				  - vnd->sc_comp_offsets[i];
1113 				if (comp_size > comp_maxsize)
1114 					comp_maxsize = comp_size;
1115 			}
1116 			vnd->sc_comp_offsets[vnd->sc_comp_numoffs - 1] =
1117 			  be64toh(vnd->sc_comp_offsets[vnd->sc_comp_numoffs - 1]);
1118 
1119 			/* create compressed data buffer */
1120 			vnd->sc_comp_buff = malloc(comp_maxsize,
1121 			  M_DEVBUF, M_WAITOK);
1122 
1123 			/* create decompressed buffer */
1124 			vnd->sc_comp_decombuf = malloc(vnd->sc_comp_blksz,
1125 			  M_DEVBUF, M_WAITOK);
1126 			vnd->sc_comp_buffblk = -1;
1127 
1128 			/* Initialize decompress stream */
1129 			memset(&vnd->sc_comp_stream, 0, sizeof(z_stream));
1130 			vnd->sc_comp_stream.zalloc = vnd_alloc;
1131 			vnd->sc_comp_stream.zfree = vnd_free;
1132 			error = inflateInit2(&vnd->sc_comp_stream, MAX_WBITS);
1133 			if(error) {
1134 				if(vnd->sc_comp_stream.msg)
1135 					printf("vnd%d: compressed file, %s\n",
1136 					  unit, vnd->sc_comp_stream.msg);
1137 				VOP_UNLOCK(nd.ni_vp, 0);
1138 				error = EINVAL;
1139 				goto close_and_exit;
1140 			}
1141 
1142 			vnd->sc_flags |= VNF_COMP | VNF_READONLY;
1143 #else /* !VND_COMPRESSION */
1144 			VOP_UNLOCK(nd.ni_vp, 0);
1145 			error = EOPNOTSUPP;
1146 			goto close_and_exit;
1147 #endif /* VND_COMPRESSION */
1148 		}
1149 
1150 		VOP_UNLOCK(nd.ni_vp, 0);
1151 		vnd->sc_vp = nd.ni_vp;
1152 		vnd->sc_size = btodb(vattr.va_size);	/* note truncation */
1153 
1154 		/*
1155 		 * Use pseudo-geometry specified.  If none was provided,
1156 		 * use "standard" Adaptec fictitious geometry.
1157 		 */
1158 		if (vio->vnd_flags & VNDIOF_HASGEOM) {
1159 
1160 			memcpy(&vnd->sc_geom, &vio->vnd_geom,
1161 			    sizeof(vio->vnd_geom));
1162 
1163 			/*
1164 			 * Sanity-check the sector size.
1165 			 * XXX Don't allow secsize < DEV_BSIZE.	 Should
1166 			 * XXX we?
1167 			 */
1168 			if (vnd->sc_geom.vng_secsize < DEV_BSIZE ||
1169 			    (vnd->sc_geom.vng_secsize % DEV_BSIZE) != 0 ||
1170 			    vnd->sc_geom.vng_ncylinders == 0 ||
1171 			    (vnd->sc_geom.vng_ntracks *
1172 			     vnd->sc_geom.vng_nsectors) == 0) {
1173 				error = EINVAL;
1174 				goto close_and_exit;
1175 			}
1176 
1177 			/*
1178 			 * Compute the size (in DEV_BSIZE blocks) specified
1179 			 * by the geometry.
1180 			 */
1181 			geomsize = (vnd->sc_geom.vng_nsectors *
1182 			    vnd->sc_geom.vng_ntracks *
1183 			    vnd->sc_geom.vng_ncylinders) *
1184 			    (vnd->sc_geom.vng_secsize / DEV_BSIZE);
1185 
1186 			/*
1187 			 * Sanity-check the size against the specified
1188 			 * geometry.
1189 			 */
1190 			if (vnd->sc_size < geomsize) {
1191 				error = EINVAL;
1192 				goto close_and_exit;
1193 			}
1194 		} else if (vnd->sc_size >= (32 * 64)) {
1195 			/*
1196 			 * Size must be at least 2048 DEV_BSIZE blocks
1197 			 * (1M) in order to use this geometry.
1198 			 */
1199 			vnd->sc_geom.vng_secsize = DEV_BSIZE;
1200 			vnd->sc_geom.vng_nsectors = 32;
1201 			vnd->sc_geom.vng_ntracks = 64;
1202 			vnd->sc_geom.vng_ncylinders = vnd->sc_size / (64 * 32);
1203 		} else {
1204 			vnd->sc_geom.vng_secsize = DEV_BSIZE;
1205 			vnd->sc_geom.vng_nsectors = 1;
1206 			vnd->sc_geom.vng_ntracks = 1;
1207 			vnd->sc_geom.vng_ncylinders = vnd->sc_size;
1208 		}
1209 
1210 		vnd_set_properties(vnd);
1211 
1212 		if (vio->vnd_flags & VNDIOF_READONLY) {
1213 			vnd->sc_flags |= VNF_READONLY;
1214 		}
1215 
1216 		if ((error = vndsetcred(vnd, l->l_cred)) != 0)
1217 			goto close_and_exit;
1218 
1219 		vndthrottle(vnd, vnd->sc_vp);
1220 		vio->vnd_size = dbtob(vnd->sc_size);
1221 		vnd->sc_flags |= VNF_INITED;
1222 
1223 		/* create the kernel thread, wait for it to be up */
1224 		error = kthread_create(PRI_NONE, 0, NULL, vndthread, vnd,
1225 		    &vnd->sc_kthread, device_xname(vnd->sc_dev));
1226 		if (error)
1227 			goto close_and_exit;
1228 		while ((vnd->sc_flags & VNF_KTHREAD) == 0) {
1229 			tsleep(&vnd->sc_kthread, PRIBIO, "vndthr", 0);
1230 		}
1231 #ifdef DEBUG
1232 		if (vnddebug & VDB_INIT)
1233 			printf("vndioctl: SET vp %p size 0x%lx %d/%d/%d/%d\n",
1234 			    vnd->sc_vp, (unsigned long) vnd->sc_size,
1235 			    vnd->sc_geom.vng_secsize,
1236 			    vnd->sc_geom.vng_nsectors,
1237 			    vnd->sc_geom.vng_ntracks,
1238 			    vnd->sc_geom.vng_ncylinders);
1239 #endif
1240 
1241 		/* Attach the disk. */
1242 		disk_attach(&vnd->sc_dkdev);
1243 
1244 		/* Initialize the xfer and buffer pools. */
1245 		pool_init(&vnd->sc_vxpool, sizeof(struct vndxfer), 0,
1246 		    0, 0, "vndxpl", NULL, IPL_BIO);
1247 
1248 		/* Try and read the disklabel. */
1249 		vndgetdisklabel(dev, vnd);
1250 
1251 		vndunlock(vnd);
1252 
1253 		break;
1254 
1255 close_and_exit:
1256 		(void) vn_close(nd.ni_vp, fflags, l->l_cred);
1257 unlock_and_exit:
1258 #ifdef VND_COMPRESSION
1259 		/* free any allocated memory (for compressed file) */
1260 		if(vnd->sc_comp_offsets) {
1261 			free(vnd->sc_comp_offsets, M_DEVBUF);
1262 			vnd->sc_comp_offsets = NULL;
1263 		}
1264 		if(vnd->sc_comp_buff) {
1265 			free(vnd->sc_comp_buff, M_DEVBUF);
1266 			vnd->sc_comp_buff = NULL;
1267 		}
1268 		if(vnd->sc_comp_decombuf) {
1269 			free(vnd->sc_comp_decombuf, M_DEVBUF);
1270 			vnd->sc_comp_decombuf = NULL;
1271 		}
1272 #endif /* VND_COMPRESSION */
1273 		vndunlock(vnd);
1274 		return (error);
1275 
1276 	case VNDIOCCLR:
1277 		if ((error = vndlock(vnd)) != 0)
1278 			return (error);
1279 
1280 		/*
1281 		 * Don't unconfigure if any other partitions are open
1282 		 * or if both the character and block flavors of this
1283 		 * partition are open.
1284 		 */
1285 		part = DISKPART(dev);
1286 		pmask = (1 << part);
1287 		if (((vnd->sc_dkdev.dk_openmask & ~pmask) ||
1288 		    ((vnd->sc_dkdev.dk_bopenmask & pmask) &&
1289 		    (vnd->sc_dkdev.dk_copenmask & pmask))) &&
1290 			!(vio->vnd_flags & VNDIOF_FORCE)) {
1291 			vndunlock(vnd);
1292 			return (EBUSY);
1293 		}
1294 
1295 		/*
1296 		 * XXX vndclear() might call vndclose() implicitely;
1297 		 * release lock to avoid recursion
1298 		 */
1299 		vndunlock(vnd);
1300 		vndclear(vnd, minor(dev));
1301 #ifdef DEBUG
1302 		if (vnddebug & VDB_INIT)
1303 			printf("vndioctl: CLRed\n");
1304 #endif
1305 
1306 		/* Destroy the xfer and buffer pools. */
1307 		pool_destroy(&vnd->sc_vxpool);
1308 
1309 		/* Detatch the disk. */
1310 		disk_detach(&vnd->sc_dkdev);
1311 		break;
1312 
1313 #ifdef COMPAT_30
1314 	case VNDIOOCGET: {
1315 		struct vnd_ouser *vnu;
1316 		struct vattr va;
1317 		vnu = (struct vnd_ouser *)data;
1318 		KASSERT(l);
1319 		switch (error = vnd_cget(l, unit, &vnu->vnu_unit, &va)) {
1320 		case 0:
1321 			vnu->vnu_dev = va.va_fsid;
1322 			vnu->vnu_ino = va.va_fileid;
1323 			break;
1324 		case -1:
1325 			/* unused is not an error */
1326 			vnu->vnu_dev = 0;
1327 			vnu->vnu_ino = 0;
1328 			break;
1329 		default:
1330 			return error;
1331 		}
1332 		break;
1333 	}
1334 #endif
1335 	case VNDIOCGET: {
1336 		struct vnd_user *vnu;
1337 		struct vattr va;
1338 		vnu = (struct vnd_user *)data;
1339 		KASSERT(l);
1340 		switch (error = vnd_cget(l, unit, &vnu->vnu_unit, &va)) {
1341 		case 0:
1342 			vnu->vnu_dev = va.va_fsid;
1343 			vnu->vnu_ino = va.va_fileid;
1344 			break;
1345 		case -1:
1346 			/* unused is not an error */
1347 			vnu->vnu_dev = 0;
1348 			vnu->vnu_ino = 0;
1349 			break;
1350 		default:
1351 			return error;
1352 		}
1353 		break;
1354 	}
1355 
1356 	case DIOCGDINFO:
1357 		*(struct disklabel *)data = *(vnd->sc_dkdev.dk_label);
1358 		break;
1359 
1360 #ifdef __HAVE_OLD_DISKLABEL
1361 	case ODIOCGDINFO:
1362 		newlabel = *(vnd->sc_dkdev.dk_label);
1363 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1364 			return ENOTTY;
1365 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1366 		break;
1367 #endif
1368 
1369 	case DIOCGPART:
1370 		((struct partinfo *)data)->disklab = vnd->sc_dkdev.dk_label;
1371 		((struct partinfo *)data)->part =
1372 		    &vnd->sc_dkdev.dk_label->d_partitions[DISKPART(dev)];
1373 		break;
1374 
1375 	case DIOCWDINFO:
1376 	case DIOCSDINFO:
1377 #ifdef __HAVE_OLD_DISKLABEL
1378 	case ODIOCWDINFO:
1379 	case ODIOCSDINFO:
1380 #endif
1381 	{
1382 		struct disklabel *lp;
1383 
1384 		if ((error = vndlock(vnd)) != 0)
1385 			return (error);
1386 
1387 		vnd->sc_flags |= VNF_LABELLING;
1388 
1389 #ifdef __HAVE_OLD_DISKLABEL
1390 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
1391 			memset(&newlabel, 0, sizeof newlabel);
1392 			memcpy(&newlabel, data, sizeof (struct olddisklabel));
1393 			lp = &newlabel;
1394 		} else
1395 #endif
1396 		lp = (struct disklabel *)data;
1397 
1398 		error = setdisklabel(vnd->sc_dkdev.dk_label,
1399 		    lp, 0, vnd->sc_dkdev.dk_cpulabel);
1400 		if (error == 0) {
1401 			if (cmd == DIOCWDINFO
1402 #ifdef __HAVE_OLD_DISKLABEL
1403 			    || cmd == ODIOCWDINFO
1404 #endif
1405 			   )
1406 				error = writedisklabel(VNDLABELDEV(dev),
1407 				    vndstrategy, vnd->sc_dkdev.dk_label,
1408 				    vnd->sc_dkdev.dk_cpulabel);
1409 		}
1410 
1411 		vnd->sc_flags &= ~VNF_LABELLING;
1412 
1413 		vndunlock(vnd);
1414 
1415 		if (error)
1416 			return (error);
1417 		break;
1418 	}
1419 
1420 	case DIOCKLABEL:
1421 		if (*(int *)data != 0)
1422 			vnd->sc_flags |= VNF_KLABEL;
1423 		else
1424 			vnd->sc_flags &= ~VNF_KLABEL;
1425 		break;
1426 
1427 	case DIOCWLABEL:
1428 		if (*(int *)data != 0)
1429 			vnd->sc_flags |= VNF_WLABEL;
1430 		else
1431 			vnd->sc_flags &= ~VNF_WLABEL;
1432 		break;
1433 
1434 	case DIOCGDEFLABEL:
1435 		vndgetdefaultlabel(vnd, (struct disklabel *)data);
1436 		break;
1437 
1438 #ifdef __HAVE_OLD_DISKLABEL
1439 	case ODIOCGDEFLABEL:
1440 		vndgetdefaultlabel(vnd, &newlabel);
1441 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1442 			return ENOTTY;
1443 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1444 		break;
1445 #endif
1446 
1447 	case DIOCCACHESYNC:
1448 		vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1449 		error = VOP_FSYNC(vnd->sc_vp, vnd->sc_cred,
1450 		    FSYNC_WAIT | FSYNC_DATAONLY | FSYNC_CACHE, 0, 0);
1451 		VOP_UNLOCK(vnd->sc_vp, 0);
1452 		return error;
1453 
1454 	default:
1455 		return (ENOTTY);
1456 	}
1457 
1458 	return (0);
1459 }
1460 
1461 /*
1462  * Duplicate the current processes' credentials.  Since we are called only
1463  * as the result of a SET ioctl and only root can do that, any future access
1464  * to this "disk" is essentially as root.  Note that credentials may change
1465  * if some other uid can write directly to the mapped file (NFS).
1466  */
1467 static int
1468 vndsetcred(struct vnd_softc *vnd, kauth_cred_t cred)
1469 {
1470 	struct uio auio;
1471 	struct iovec aiov;
1472 	char *tmpbuf;
1473 	int error;
1474 
1475 	vnd->sc_cred = kauth_cred_dup(cred);
1476 	tmpbuf = malloc(DEV_BSIZE, M_TEMP, M_WAITOK);
1477 
1478 	/* XXX: Horrible kludge to establish credentials for NFS */
1479 	aiov.iov_base = tmpbuf;
1480 	aiov.iov_len = min(DEV_BSIZE, dbtob(vnd->sc_size));
1481 	auio.uio_iov = &aiov;
1482 	auio.uio_iovcnt = 1;
1483 	auio.uio_offset = 0;
1484 	auio.uio_rw = UIO_READ;
1485 	auio.uio_resid = aiov.iov_len;
1486 	UIO_SETUP_SYSSPACE(&auio);
1487 	vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1488 	error = VOP_READ(vnd->sc_vp, &auio, 0, vnd->sc_cred);
1489 	if (error == 0) {
1490 		/*
1491 		 * Because vnd does all IO directly through the vnode
1492 		 * we need to flush (at least) the buffer from the above
1493 		 * VOP_READ from the buffer cache to prevent cache
1494 		 * incoherencies.  Also, be careful to write dirty
1495 		 * buffers back to stable storage.
1496 		 */
1497 		error = vinvalbuf(vnd->sc_vp, V_SAVE, vnd->sc_cred,
1498 			    curlwp, 0, 0);
1499 	}
1500 	VOP_UNLOCK(vnd->sc_vp, 0);
1501 
1502 	free(tmpbuf, M_TEMP);
1503 	return (error);
1504 }
1505 
1506 /*
1507  * Set maxactive based on FS type
1508  */
1509 static void
1510 vndthrottle(struct vnd_softc *vnd, struct vnode *vp)
1511 {
1512 #ifdef NFS
1513 	extern int (**nfsv2_vnodeop_p)(void *);
1514 
1515 	if (vp->v_op == nfsv2_vnodeop_p)
1516 		vnd->sc_maxactive = 2;
1517 	else
1518 #endif
1519 		vnd->sc_maxactive = 8;
1520 
1521 	if (vnd->sc_maxactive < 1)
1522 		vnd->sc_maxactive = 1;
1523 }
1524 
1525 #if 0
1526 static void
1527 vndshutdown(void)
1528 {
1529 	struct vnd_softc *vnd;
1530 
1531 	for (vnd = &vnd_softc[0]; vnd < &vnd_softc[numvnd]; vnd++)
1532 		if (vnd->sc_flags & VNF_INITED)
1533 			vndclear(vnd);
1534 }
1535 #endif
1536 
1537 static void
1538 vndclear(struct vnd_softc *vnd, int myminor)
1539 {
1540 	struct vnode *vp = vnd->sc_vp;
1541 	int fflags = FREAD;
1542 	int bmaj, cmaj, i, mn;
1543 	int s;
1544 
1545 #ifdef DEBUG
1546 	if (vnddebug & VDB_FOLLOW)
1547 		printf("vndclear(%p): vp %p\n", vnd, vp);
1548 #endif
1549 	/* locate the major number */
1550 	bmaj = bdevsw_lookup_major(&vnd_bdevsw);
1551 	cmaj = cdevsw_lookup_major(&vnd_cdevsw);
1552 
1553 	/* Nuke the vnodes for any open instances */
1554 	for (i = 0; i < MAXPARTITIONS; i++) {
1555 		mn = DISKMINOR(device_unit(vnd->sc_dev), i);
1556 		vdevgone(bmaj, mn, mn, VBLK);
1557 		if (mn != myminor) /* XXX avoid to kill own vnode */
1558 			vdevgone(cmaj, mn, mn, VCHR);
1559 	}
1560 
1561 	if ((vnd->sc_flags & VNF_READONLY) == 0)
1562 		fflags |= FWRITE;
1563 
1564 	s = splbio();
1565 	bufq_drain(vnd->sc_tab);
1566 	splx(s);
1567 
1568 	vnd->sc_flags |= VNF_VUNCONF;
1569 	wakeup(&vnd->sc_tab);
1570 	while (vnd->sc_flags & VNF_KTHREAD)
1571 		tsleep(&vnd->sc_kthread, PRIBIO, "vnthr", 0);
1572 
1573 #ifdef VND_COMPRESSION
1574 	/* free the compressed file buffers */
1575 	if(vnd->sc_flags & VNF_COMP) {
1576 		if(vnd->sc_comp_offsets) {
1577 			free(vnd->sc_comp_offsets, M_DEVBUF);
1578 			vnd->sc_comp_offsets = NULL;
1579 		}
1580 		if(vnd->sc_comp_buff) {
1581 			free(vnd->sc_comp_buff, M_DEVBUF);
1582 			vnd->sc_comp_buff = NULL;
1583 		}
1584 		if(vnd->sc_comp_decombuf) {
1585 			free(vnd->sc_comp_decombuf, M_DEVBUF);
1586 			vnd->sc_comp_decombuf = NULL;
1587 		}
1588 	}
1589 #endif /* VND_COMPRESSION */
1590 	vnd->sc_flags &=
1591 	    ~(VNF_INITED | VNF_READONLY | VNF_VLABEL
1592 	      | VNF_VUNCONF | VNF_COMP);
1593 	if (vp == (struct vnode *)0)
1594 		panic("vndclear: null vp");
1595 	(void) vn_close(vp, fflags, vnd->sc_cred);
1596 	kauth_cred_free(vnd->sc_cred);
1597 	vnd->sc_vp = (struct vnode *)0;
1598 	vnd->sc_cred = (kauth_cred_t)0;
1599 	vnd->sc_size = 0;
1600 }
1601 
1602 static int
1603 vndsize(dev_t dev)
1604 {
1605 	struct vnd_softc *sc;
1606 	struct disklabel *lp;
1607 	int part, unit, omask;
1608 	int size;
1609 
1610 	unit = vndunit(dev);
1611 	sc = device_lookup_private(&vnd_cd, unit);
1612 	if (sc == NULL)
1613 		return -1;
1614 
1615 	if ((sc->sc_flags & VNF_INITED) == 0)
1616 		return (-1);
1617 
1618 	part = DISKPART(dev);
1619 	omask = sc->sc_dkdev.dk_openmask & (1 << part);
1620 	lp = sc->sc_dkdev.dk_label;
1621 
1622 	if (omask == 0 && vndopen(dev, 0, S_IFBLK, curlwp))	/* XXX */
1623 		return (-1);
1624 
1625 	if (lp->d_partitions[part].p_fstype != FS_SWAP)
1626 		size = -1;
1627 	else
1628 		size = lp->d_partitions[part].p_size *
1629 		    (lp->d_secsize / DEV_BSIZE);
1630 
1631 	if (omask == 0 && vndclose(dev, 0, S_IFBLK, curlwp))	/* XXX */
1632 		return (-1);
1633 
1634 	return (size);
1635 }
1636 
1637 static int
1638 vnddump(dev_t dev, daddr_t blkno, void *va,
1639     size_t size)
1640 {
1641 
1642 	/* Not implemented. */
1643 	return ENXIO;
1644 }
1645 
1646 static void
1647 vndgetdefaultlabel(struct vnd_softc *sc, struct disklabel *lp)
1648 {
1649 	struct vndgeom *vng = &sc->sc_geom;
1650 	struct partition *pp;
1651 
1652 	memset(lp, 0, sizeof(*lp));
1653 
1654 	lp->d_secperunit = sc->sc_size / (vng->vng_secsize / DEV_BSIZE);
1655 	lp->d_secsize = vng->vng_secsize;
1656 	lp->d_nsectors = vng->vng_nsectors;
1657 	lp->d_ntracks = vng->vng_ntracks;
1658 	lp->d_ncylinders = vng->vng_ncylinders;
1659 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1660 
1661 	strncpy(lp->d_typename, "vnd", sizeof(lp->d_typename));
1662 	lp->d_type = DTYPE_VND;
1663 	strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
1664 	lp->d_rpm = 3600;
1665 	lp->d_interleave = 1;
1666 	lp->d_flags = 0;
1667 
1668 	pp = &lp->d_partitions[RAW_PART];
1669 	pp->p_offset = 0;
1670 	pp->p_size = lp->d_secperunit;
1671 	pp->p_fstype = FS_UNUSED;
1672 	lp->d_npartitions = RAW_PART + 1;
1673 
1674 	lp->d_magic = DISKMAGIC;
1675 	lp->d_magic2 = DISKMAGIC;
1676 	lp->d_checksum = dkcksum(lp);
1677 }
1678 
1679 /*
1680  * Read the disklabel from a vnd.  If one is not present, create a fake one.
1681  */
1682 static void
1683 vndgetdisklabel(dev_t dev, struct vnd_softc *sc)
1684 {
1685 	const char *errstring;
1686 	struct disklabel *lp = sc->sc_dkdev.dk_label;
1687 	struct cpu_disklabel *clp = sc->sc_dkdev.dk_cpulabel;
1688 	int i;
1689 
1690 	memset(clp, 0, sizeof(*clp));
1691 
1692 	vndgetdefaultlabel(sc, lp);
1693 
1694 	/*
1695 	 * Call the generic disklabel extraction routine.
1696 	 */
1697 	errstring = readdisklabel(VNDLABELDEV(dev), vndstrategy, lp, clp);
1698 	if (errstring) {
1699 		/*
1700 		 * Lack of disklabel is common, but we print the warning
1701 		 * anyway, since it might contain other useful information.
1702 		 */
1703 		aprint_normal_dev(sc->sc_dev, "%s\n", errstring);
1704 
1705 		/*
1706 		 * For historical reasons, if there's no disklabel
1707 		 * present, all partitions must be FS_BSDFFS and
1708 		 * occupy the entire disk.
1709 		 */
1710 		for (i = 0; i < MAXPARTITIONS; i++) {
1711 			/*
1712 			 * Don't wipe out port specific hack (such as
1713 			 * dos partition hack of i386 port).
1714 			 */
1715 			if (lp->d_partitions[i].p_size != 0)
1716 				continue;
1717 
1718 			lp->d_partitions[i].p_size = lp->d_secperunit;
1719 			lp->d_partitions[i].p_offset = 0;
1720 			lp->d_partitions[i].p_fstype = FS_BSDFFS;
1721 		}
1722 
1723 		strncpy(lp->d_packname, "default label",
1724 		    sizeof(lp->d_packname));
1725 
1726 		lp->d_npartitions = MAXPARTITIONS;
1727 		lp->d_checksum = dkcksum(lp);
1728 	}
1729 
1730 	/* In-core label now valid. */
1731 	sc->sc_flags |= VNF_VLABEL;
1732 }
1733 
1734 /*
1735  * Wait interruptibly for an exclusive lock.
1736  *
1737  * XXX
1738  * Several drivers do this; it should be abstracted and made MP-safe.
1739  */
1740 static int
1741 vndlock(struct vnd_softc *sc)
1742 {
1743 	int error;
1744 
1745 	while ((sc->sc_flags & VNF_LOCKED) != 0) {
1746 		sc->sc_flags |= VNF_WANTED;
1747 		if ((error = tsleep(sc, PRIBIO | PCATCH, "vndlck", 0)) != 0)
1748 			return (error);
1749 	}
1750 	sc->sc_flags |= VNF_LOCKED;
1751 	return (0);
1752 }
1753 
1754 /*
1755  * Unlock and wake up any waiters.
1756  */
1757 static void
1758 vndunlock(struct vnd_softc *sc)
1759 {
1760 
1761 	sc->sc_flags &= ~VNF_LOCKED;
1762 	if ((sc->sc_flags & VNF_WANTED) != 0) {
1763 		sc->sc_flags &= ~VNF_WANTED;
1764 		wakeup(sc);
1765 	}
1766 }
1767 
1768 #ifdef VND_COMPRESSION
1769 /* compressed file read */
1770 static void
1771 compstrategy(struct buf *bp, off_t bn)
1772 {
1773 	int error;
1774 	int unit = vndunit(bp->b_dev);
1775 	struct vnd_softc *vnd =
1776 	    device_lookup_private(&vnd_cd, unit);
1777 	u_int32_t comp_block;
1778 	struct uio auio;
1779 	char *addr;
1780 	int s;
1781 
1782 	/* set up constants for data move */
1783 	auio.uio_rw = UIO_READ;
1784 	UIO_SETUP_SYSSPACE(&auio);
1785 
1786 	/* read, and transfer the data */
1787 	addr = bp->b_data;
1788 	bp->b_resid = bp->b_bcount;
1789 	s = splbio();
1790 	while (bp->b_resid > 0) {
1791 		unsigned length;
1792 		size_t length_in_buffer;
1793 		u_int32_t offset_in_buffer;
1794 		struct iovec aiov;
1795 
1796 		/* calculate the compressed block number */
1797 		comp_block = bn / (off_t)vnd->sc_comp_blksz;
1798 
1799 		/* check for good block number */
1800 		if (comp_block >= vnd->sc_comp_numoffs) {
1801 			bp->b_error = EINVAL;
1802 			splx(s);
1803 			return;
1804 		}
1805 
1806 		/* read in the compressed block, if not in buffer */
1807 		if (comp_block != vnd->sc_comp_buffblk) {
1808 			length = vnd->sc_comp_offsets[comp_block + 1] -
1809 			    vnd->sc_comp_offsets[comp_block];
1810 			vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1811 			error = vn_rdwr(UIO_READ, vnd->sc_vp, vnd->sc_comp_buff,
1812 			    length, vnd->sc_comp_offsets[comp_block],
1813 			    UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, vnd->sc_cred,
1814 			    NULL, NULL);
1815 			if (error) {
1816 				bp->b_error = error;
1817 				VOP_UNLOCK(vnd->sc_vp, 0);
1818 				splx(s);
1819 				return;
1820 			}
1821 			/* uncompress the buffer */
1822 			vnd->sc_comp_stream.next_in = vnd->sc_comp_buff;
1823 			vnd->sc_comp_stream.avail_in = length;
1824 			vnd->sc_comp_stream.next_out = vnd->sc_comp_decombuf;
1825 			vnd->sc_comp_stream.avail_out = vnd->sc_comp_blksz;
1826 			inflateReset(&vnd->sc_comp_stream);
1827 			error = inflate(&vnd->sc_comp_stream, Z_FINISH);
1828 			if (error != Z_STREAM_END) {
1829 				if (vnd->sc_comp_stream.msg)
1830 					aprint_normal_dev(vnd->sc_dev,
1831 					    "compressed file, %s\n",
1832 					    vnd->sc_comp_stream.msg);
1833 				bp->b_error = EBADMSG;
1834 				VOP_UNLOCK(vnd->sc_vp, 0);
1835 				splx(s);
1836 				return;
1837 			}
1838 			vnd->sc_comp_buffblk = comp_block;
1839 			VOP_UNLOCK(vnd->sc_vp, 0);
1840 		}
1841 
1842 		/* transfer the usable uncompressed data */
1843 		offset_in_buffer = bn % (off_t)vnd->sc_comp_blksz;
1844 		length_in_buffer = vnd->sc_comp_blksz - offset_in_buffer;
1845 		if (length_in_buffer > bp->b_resid)
1846 			length_in_buffer = bp->b_resid;
1847 		auio.uio_iov = &aiov;
1848 		auio.uio_iovcnt = 1;
1849 		aiov.iov_base = addr;
1850 		aiov.iov_len = length_in_buffer;
1851 		auio.uio_resid = aiov.iov_len;
1852 		auio.uio_offset = 0;
1853 		error = uiomove(vnd->sc_comp_decombuf + offset_in_buffer,
1854 		    length_in_buffer, &auio);
1855 		if (error) {
1856 			bp->b_error = error;
1857 			splx(s);
1858 			return;
1859 		}
1860 
1861 		bn += length_in_buffer;
1862 		addr += length_in_buffer;
1863 		bp->b_resid -= length_in_buffer;
1864 	}
1865 	splx(s);
1866 }
1867 
1868 /* compression memory allocation routines */
1869 static void *
1870 vnd_alloc(void *aux, u_int items, u_int siz)
1871 {
1872 	return malloc(items * siz, M_TEMP, M_NOWAIT);
1873 }
1874 
1875 static void
1876 vnd_free(void *aux, void *ptr)
1877 {
1878 	free(ptr, M_TEMP);
1879 }
1880 #endif /* VND_COMPRESSION */
1881 
1882 static void
1883 vnd_set_properties(struct vnd_softc *vnd)
1884 {
1885 	prop_dictionary_t disk_info, odisk_info, geom;
1886 
1887 	disk_info = prop_dictionary_create();
1888 
1889 	geom = prop_dictionary_create();
1890 
1891 	prop_dictionary_set_uint64(geom, "sectors-per-unit",
1892 	    vnd->sc_geom.vng_nsectors * vnd->sc_geom.vng_ntracks *
1893 	    vnd->sc_geom.vng_ncylinders);
1894 
1895 	prop_dictionary_set_uint32(geom, "sector-size",
1896 	    vnd->sc_geom.vng_secsize);
1897 
1898 	prop_dictionary_set_uint16(geom, "sectors-per-track",
1899 	    vnd->sc_geom.vng_nsectors);
1900 
1901 	prop_dictionary_set_uint16(geom, "tracks-per-cylinder",
1902 	    vnd->sc_geom.vng_ntracks);
1903 
1904 	prop_dictionary_set_uint64(geom, "cylinders-per-unit",
1905 	    vnd->sc_geom.vng_ncylinders);
1906 
1907 	prop_dictionary_set(disk_info, "geometry", geom);
1908 	prop_object_release(geom);
1909 
1910 	prop_dictionary_set(device_properties(vnd->sc_dev),
1911 	    "disk-info", disk_info);
1912 
1913 	/*
1914 	 * Don't release disk_info here; we keep a reference to it.
1915 	 * disk_detach() will release it when we go away.
1916 	 */
1917 
1918 	odisk_info = vnd->sc_dkdev.dk_info;
1919 	vnd->sc_dkdev.dk_info = disk_info;
1920 	if (odisk_info)
1921 		prop_object_release(odisk_info);
1922 }
1923 
1924 #ifdef _MODULE
1925 
1926 #include <sys/module.h>
1927 
1928 MODULE(MODULE_CLASS_DRIVER, vnd, NULL);
1929 CFDRIVER_DECL(vnd, DV_DISK, NULL);
1930 
1931 static int
1932 vnd_modcmd(modcmd_t cmd, void *arg)
1933 {
1934 	int bmajor = -1, cmajor = -1,  error = 0;
1935 
1936 	switch (cmd) {
1937 	case MODULE_CMD_INIT:
1938 		error = config_cfdriver_attach(&vnd_cd);
1939 		if (error)
1940 			break;
1941 
1942 		error = config_cfattach_attach(vnd_cd.cd_name, &vnd_ca);
1943 	        if (error) {
1944 			config_cfdriver_detach(&vnd_cd);
1945 			aprint_error("%s: unable to register cfattach\n",
1946 			    vnd_cd.cd_name);
1947 			break;
1948 		}
1949 
1950 		error = devsw_attach("vnd", &vnd_bdevsw, &bmajor,
1951 		    &vnd_cdevsw, &cmajor);
1952 		if (error) {
1953 			config_cfattach_detach(vnd_cd.cd_name, &vnd_ca);
1954 			config_cfdriver_detach(&vnd_cd);
1955 			break;
1956 		}
1957 
1958 		break;
1959 
1960 	case MODULE_CMD_FINI:
1961 		error = config_cfattach_detach(vnd_cd.cd_name, &vnd_ca);
1962 		if (error)
1963 			break;
1964 		config_cfdriver_detach(&vnd_cd);
1965 		devsw_detach(&vnd_bdevsw, &vnd_cdevsw);
1966 		break;
1967 
1968 	case MODULE_CMD_STAT:
1969 		return ENOTTY;
1970 
1971 	default:
1972 		return ENOTTY;
1973 	}
1974 
1975 	return error;
1976 }
1977 
1978 #endif
1979