xref: /netbsd-src/sys/dev/vnd.c (revision 10ad5ffa714ce1a679dcc9dd8159648df2d67b5a)
1 /*	$NetBSD: vnd.c,v 1.203 2009/07/07 19:51:22 dyoung 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 1.203 2009/07/07 19:51:22 dyoung 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	vnddoclear(struct vnd_softc *, int, int, bool);
202 static int	vndsetcred(struct vnd_softc *, kauth_cred_t);
203 static void	vndthrottle(struct vnd_softc *, struct vnode *);
204 static void	vndiodone(struct buf *);
205 #if 0
206 static void	vndshutdown(void);
207 #endif
208 
209 static void	vndgetdefaultlabel(struct vnd_softc *, struct disklabel *);
210 static void	vndgetdisklabel(dev_t, struct vnd_softc *);
211 
212 static int	vndlock(struct vnd_softc *);
213 static void	vndunlock(struct vnd_softc *);
214 #ifdef VND_COMPRESSION
215 static void	compstrategy(struct buf *, off_t);
216 static void	*vnd_alloc(void *, u_int, u_int);
217 static void	vnd_free(void *, void *);
218 #endif /* VND_COMPRESSION */
219 
220 static void	vndthread(void *);
221 static bool	vnode_has_op(const struct vnode *, int);
222 static void	handle_with_rdwr(struct vnd_softc *, const struct buf *,
223 		    struct buf *);
224 static void	handle_with_strategy(struct vnd_softc *, const struct buf *,
225 		    struct buf *);
226 static void	vnd_set_properties(struct vnd_softc *);
227 
228 static dev_type_open(vndopen);
229 static dev_type_close(vndclose);
230 static dev_type_read(vndread);
231 static dev_type_write(vndwrite);
232 static dev_type_ioctl(vndioctl);
233 static dev_type_strategy(vndstrategy);
234 static dev_type_dump(vnddump);
235 static dev_type_size(vndsize);
236 
237 const struct bdevsw vnd_bdevsw = {
238 	vndopen, vndclose, vndstrategy, vndioctl, vnddump, vndsize, D_DISK
239 };
240 
241 const struct cdevsw vnd_cdevsw = {
242 	vndopen, vndclose, vndread, vndwrite, vndioctl,
243 	nostop, notty, nopoll, nommap, nokqfilter, D_DISK
244 };
245 
246 static int	vnd_match(device_t, cfdata_t, void *);
247 static void	vnd_attach(device_t, device_t, void *);
248 static int	vnd_detach(device_t, int);
249 
250 CFATTACH_DECL3_NEW(vnd, sizeof(struct vnd_softc),
251     vnd_match, vnd_attach, vnd_detach, NULL, NULL, NULL, DVF_DETACH_SHUTDOWN);
252 extern struct cfdriver vnd_cd;
253 
254 static struct vnd_softc	*vnd_spawn(int);
255 int	vnd_destroy(device_t);
256 
257 void
258 vndattach(int num)
259 {
260 	int error;
261 
262 	error = config_cfattach_attach(vnd_cd.cd_name, &vnd_ca);
263 	if (error)
264 		aprint_error("%s: unable to register cfattach\n",
265 		    vnd_cd.cd_name);
266 }
267 
268 static int
269 vnd_match(device_t self, cfdata_t cfdata, void *aux)
270 {
271 
272 	return 1;
273 }
274 
275 static void
276 vnd_attach(device_t parent, device_t self, void *aux)
277 {
278 	struct vnd_softc *sc = device_private(self);
279 
280 	sc->sc_dev = self;
281 	sc->sc_comp_offsets = NULL;
282 	sc->sc_comp_buff = NULL;
283 	sc->sc_comp_decombuf = NULL;
284 	bufq_alloc(&sc->sc_tab, "disksort", BUFQ_SORT_RAWBLOCK);
285 	disk_init(&sc->sc_dkdev, device_xname(self), NULL);
286 	if (!pmf_device_register(self, NULL, NULL))
287 		aprint_error_dev(self, "couldn't establish power handler\n");
288 }
289 
290 static int
291 vnd_detach(device_t self, int flags)
292 {
293 	int error;
294 	struct vnd_softc *sc = device_private(self);
295 
296 	if (sc->sc_flags & VNF_INITED) {
297 		error = vnddoclear(sc, 0, -1, (flags & DETACH_FORCE) != 0);
298 		if (error != 0)
299 			return error;
300 	}
301 
302 	pmf_device_deregister(self);
303 	bufq_free(sc->sc_tab);
304 	disk_destroy(&sc->sc_dkdev);
305 
306 	return 0;
307 }
308 
309 static struct vnd_softc *
310 vnd_spawn(int unit)
311 {
312 	cfdata_t cf;
313 
314 	cf = malloc(sizeof(*cf), M_DEVBUF, M_WAITOK);
315 	cf->cf_name = vnd_cd.cd_name;
316 	cf->cf_atname = vnd_cd.cd_name;
317 	cf->cf_unit = unit;
318 	cf->cf_fstate = FSTATE_STAR;
319 
320 	return device_private(config_attach_pseudo(cf));
321 }
322 
323 int
324 vnd_destroy(device_t dev)
325 {
326 	int error;
327 	cfdata_t cf;
328 
329 	cf = device_cfdata(dev);
330 	error = config_detach(dev, DETACH_QUIET);
331 	if (error)
332 		return error;
333 	free(cf, M_DEVBUF);
334 	return 0;
335 }
336 
337 static int
338 vndopen(dev_t dev, int flags, int mode, struct lwp *l)
339 {
340 	int unit = vndunit(dev);
341 	struct vnd_softc *sc;
342 	int error = 0, part, pmask;
343 	struct disklabel *lp;
344 
345 #ifdef DEBUG
346 	if (vnddebug & VDB_FOLLOW)
347 		printf("vndopen(0x%"PRIx64", 0x%x, 0x%x, %p)\n", dev, flags, mode, l);
348 #endif
349 	sc = device_lookup_private(&vnd_cd, unit);
350 	if (sc == NULL) {
351 		sc = vnd_spawn(unit);
352 		if (sc == NULL)
353 			return ENOMEM;
354 	}
355 
356 	if ((error = vndlock(sc)) != 0)
357 		return error;
358 
359 	if ((sc->sc_flags & VNF_CLEARING) != 0) {
360 		error = ENXIO;
361 		goto done;
362 	}
363 
364 	lp = sc->sc_dkdev.dk_label;
365 
366 	part = DISKPART(dev);
367 	pmask = (1 << part);
368 
369 	/*
370 	 * If we're initialized, check to see if there are any other
371 	 * open partitions.  If not, then it's safe to update the
372 	 * in-core disklabel.  Only read the disklabel if it is
373 	 * not already valid.
374 	 */
375 	if ((sc->sc_flags & (VNF_INITED|VNF_VLABEL)) == VNF_INITED &&
376 	    sc->sc_dkdev.dk_openmask == 0)
377 		vndgetdisklabel(dev, sc);
378 
379 	/* Check that the partitions exists. */
380 	if (part != RAW_PART) {
381 		if (((sc->sc_flags & VNF_INITED) == 0) ||
382 		    ((part >= lp->d_npartitions) ||
383 		     (lp->d_partitions[part].p_fstype == FS_UNUSED))) {
384 			error = ENXIO;
385 			goto done;
386 		}
387 	}
388 
389 	/* Prevent our unit from being unconfigured while open. */
390 	switch (mode) {
391 	case S_IFCHR:
392 		sc->sc_dkdev.dk_copenmask |= pmask;
393 		break;
394 
395 	case S_IFBLK:
396 		sc->sc_dkdev.dk_bopenmask |= pmask;
397 		break;
398 	}
399 	sc->sc_dkdev.dk_openmask =
400 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
401 
402  done:
403 	vndunlock(sc);
404 	return error;
405 }
406 
407 static int
408 vndclose(dev_t dev, int flags, int mode, struct lwp *l)
409 {
410 	int unit = vndunit(dev);
411 	struct vnd_softc *sc;
412 	int error = 0, part;
413 
414 #ifdef DEBUG
415 	if (vnddebug & VDB_FOLLOW)
416 		printf("vndclose(0x%"PRIx64", 0x%x, 0x%x, %p)\n", dev, flags, mode, l);
417 #endif
418 	sc = device_lookup_private(&vnd_cd, unit);
419 	if (sc == NULL)
420 		return ENXIO;
421 
422 	if ((error = vndlock(sc)) != 0)
423 		return error;
424 
425 	part = DISKPART(dev);
426 
427 	/* ...that much closer to allowing unconfiguration... */
428 	switch (mode) {
429 	case S_IFCHR:
430 		sc->sc_dkdev.dk_copenmask &= ~(1 << part);
431 		break;
432 
433 	case S_IFBLK:
434 		sc->sc_dkdev.dk_bopenmask &= ~(1 << part);
435 		break;
436 	}
437 	sc->sc_dkdev.dk_openmask =
438 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
439 
440 	vndunlock(sc);
441 
442 	if ((sc->sc_flags & VNF_INITED) == 0) {
443 		if ((error = vnd_destroy(sc->sc_dev)) != 0) {
444 			aprint_error_dev(sc->sc_dev,
445 			    "unable to detach instance\n");
446 			return error;
447 		}
448 	}
449 
450 	return 0;
451 }
452 
453 /*
454  * Queue the request, and wakeup the kernel thread to handle it.
455  */
456 static void
457 vndstrategy(struct buf *bp)
458 {
459 	int unit = vndunit(bp->b_dev);
460 	struct vnd_softc *vnd =
461 	    device_lookup_private(&vnd_cd, unit);
462 	struct disklabel *lp;
463 	daddr_t blkno;
464 	int s = splbio();
465 
466 	if (vnd == NULL) {
467 		bp->b_error = ENXIO;
468 		goto done;
469 	}
470 	lp = vnd->sc_dkdev.dk_label;
471 
472 	if ((vnd->sc_flags & VNF_INITED) == 0) {
473 		bp->b_error = ENXIO;
474 		goto done;
475 	}
476 
477 	/*
478 	 * The transfer must be a whole number of blocks.
479 	 */
480 	if ((bp->b_bcount % lp->d_secsize) != 0) {
481 		bp->b_error = EINVAL;
482 		goto done;
483 	}
484 
485 	/*
486 	 * check if we're read-only.
487 	 */
488 	if ((vnd->sc_flags & VNF_READONLY) && !(bp->b_flags & B_READ)) {
489 		bp->b_error = EACCES;
490 		goto done;
491 	}
492 
493 	/* If it's a nil transfer, wake up the top half now. */
494 	if (bp->b_bcount == 0) {
495 		goto done;
496 	}
497 
498 	/*
499 	 * Do bounds checking and adjust transfer.  If there's an error,
500 	 * the bounds check will flag that for us.
501 	 */
502 	if (DISKPART(bp->b_dev) == RAW_PART) {
503 		if (bounds_check_with_mediasize(bp, DEV_BSIZE,
504 		    vnd->sc_size) <= 0)
505 			goto done;
506 	} else {
507 		if (bounds_check_with_label(&vnd->sc_dkdev,
508 		    bp, vnd->sc_flags & (VNF_WLABEL|VNF_LABELLING)) <= 0)
509 			goto done;
510 	}
511 
512 	/*
513 	 * Put the block number in terms of the logical blocksize
514 	 * of the "device".
515 	 */
516 
517 	blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
518 
519 	/*
520 	 * Translate the partition-relative block number to an absolute.
521 	 */
522 	if (DISKPART(bp->b_dev) != RAW_PART) {
523 		struct partition *pp;
524 
525 		pp = &vnd->sc_dkdev.dk_label->d_partitions[
526 		    DISKPART(bp->b_dev)];
527 		blkno += pp->p_offset;
528 	}
529 	bp->b_rawblkno = blkno;
530 
531 #ifdef DEBUG
532 	if (vnddebug & VDB_FOLLOW)
533 		printf("vndstrategy(%p): unit %d\n", bp, unit);
534 #endif
535 	bufq_put(vnd->sc_tab, bp);
536 	wakeup(&vnd->sc_tab);
537 	splx(s);
538 	return;
539 
540 done:
541 	bp->b_resid = bp->b_bcount;
542 	biodone(bp);
543 	splx(s);
544 }
545 
546 static bool
547 vnode_has_strategy(struct vnd_softc *vnd)
548 {
549 	return vnode_has_op(vnd->sc_vp, VOFFSET(vop_bmap)) &&
550 	    vnode_has_op(vnd->sc_vp, VOFFSET(vop_strategy));
551 }
552 
553 /* XXX this function needs a reliable check to detect
554  * sparse files. Otherwise, bmap/strategy may be used
555  * and fail on non-allocated blocks. VOP_READ/VOP_WRITE
556  * works on sparse files.
557  */
558 #if notyet
559 static bool
560 vnode_strategy_probe(struct vnd_softc *vnd)
561 {
562 	int error;
563 	daddr_t nbn;
564 
565 	if (!vnode_has_strategy(vnd))
566 		return false;
567 
568 	/* Convert the first logical block number to its
569 	 * physical block number.
570 	 */
571 	error = 0;
572 	vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
573 	error = VOP_BMAP(vnd->sc_vp, 0, NULL, &nbn, NULL);
574 	VOP_UNLOCK(vnd->sc_vp, 0);
575 
576 	/* Test if that worked. */
577 	if (error == 0 && (long)nbn == -1)
578 		return false;
579 
580 	return true;
581 }
582 #endif
583 
584 static void
585 vndthread(void *arg)
586 {
587 	struct vnd_softc *vnd = arg;
588 	bool usestrategy;
589 	int s;
590 
591 	/* Determine whether we can *use* VOP_BMAP and VOP_STRATEGY to
592 	 * directly access the backing vnode.  If we can, use these two
593 	 * operations to avoid messing with the local buffer cache.
594 	 * Otherwise fall back to regular VOP_READ/VOP_WRITE operations
595 	 * which are guaranteed to work with any file system. */
596 	usestrategy = vnode_has_strategy(vnd);
597 
598 #ifdef DEBUG
599 	if (vnddebug & VDB_INIT)
600 		printf("vndthread: vp %p, %s\n", vnd->sc_vp,
601 		    usestrategy ?
602 		    "using bmap/strategy operations" :
603 		    "using read/write operations");
604 #endif
605 
606 	s = splbio();
607 	vnd->sc_flags |= VNF_KTHREAD;
608 	wakeup(&vnd->sc_kthread);
609 
610 	/*
611 	 * Dequeue requests and serve them depending on the available
612 	 * vnode operations.
613 	 */
614 	while ((vnd->sc_flags & VNF_VUNCONF) == 0) {
615 		struct vndxfer *vnx;
616 		int flags;
617 		struct buf *obp;
618 		struct buf *bp;
619 
620 		obp = bufq_get(vnd->sc_tab);
621 		if (obp == NULL) {
622 			tsleep(&vnd->sc_tab, PRIBIO, "vndbp", 0);
623 			continue;
624 		};
625 		splx(s);
626 		flags = obp->b_flags;
627 #ifdef DEBUG
628 		if (vnddebug & VDB_FOLLOW)
629 			printf("vndthread(%p)\n", obp);
630 #endif
631 
632 		if (vnd->sc_vp->v_mount == NULL) {
633 			obp->b_error = ENXIO;
634 			goto done;
635 		}
636 #ifdef VND_COMPRESSION
637 		/* handle a compressed read */
638 		if ((flags & B_READ) != 0 && (vnd->sc_flags & VNF_COMP)) {
639 			off_t bn;
640 
641 			/* Convert to a byte offset within the file. */
642 			bn = obp->b_rawblkno *
643 			    vnd->sc_dkdev.dk_label->d_secsize;
644 
645 			compstrategy(obp, bn);
646 			goto done;
647 		}
648 #endif /* VND_COMPRESSION */
649 
650 		/*
651 		 * Allocate a header for this transfer and link it to the
652 		 * buffer
653 		 */
654 		s = splbio();
655 		vnx = VND_GETXFER(vnd);
656 		splx(s);
657 		vnx->vx_vnd = vnd;
658 
659 		s = splbio();
660 		while (vnd->sc_active >= vnd->sc_maxactive) {
661 			tsleep(&vnd->sc_tab, PRIBIO, "vndac", 0);
662 		}
663 		vnd->sc_active++;
664 		splx(s);
665 
666 		/* Instrumentation. */
667 		disk_busy(&vnd->sc_dkdev);
668 
669 		bp = &vnx->vx_buf;
670 		buf_init(bp);
671 		bp->b_flags = (obp->b_flags & B_READ);
672 		bp->b_oflags = obp->b_oflags;
673 		bp->b_cflags = obp->b_cflags;
674 		bp->b_iodone = vndiodone;
675 		bp->b_private = obp;
676 		bp->b_vp = vnd->sc_vp;
677 		bp->b_objlock = &bp->b_vp->v_interlock;
678 		bp->b_data = obp->b_data;
679 		bp->b_bcount = obp->b_bcount;
680 		BIO_COPYPRIO(bp, obp);
681 
682 		/* Handle the request using the appropriate operations. */
683 		if (usestrategy)
684 			handle_with_strategy(vnd, obp, bp);
685 		else
686 			handle_with_rdwr(vnd, obp, bp);
687 
688 		s = splbio();
689 		continue;
690 
691 done:
692 		biodone(obp);
693 		s = splbio();
694 	}
695 
696 	vnd->sc_flags &= (~VNF_KTHREAD | VNF_VUNCONF);
697 	wakeup(&vnd->sc_kthread);
698 	splx(s);
699 	kthread_exit(0);
700 }
701 
702 /*
703  * Checks if the given vnode supports the requested operation.
704  * The operation is specified the offset returned by VOFFSET.
705  *
706  * XXX The test below used to determine this is quite fragile
707  * because it relies on the file system to use genfs to specify
708  * unimplemented operations.  There might be another way to do
709  * it more cleanly.
710  */
711 static bool
712 vnode_has_op(const struct vnode *vp, int opoffset)
713 {
714 	int (*defaultp)(void *);
715 	int (*opp)(void *);
716 
717 	defaultp = vp->v_op[VOFFSET(vop_default)];
718 	opp = vp->v_op[opoffset];
719 
720 	return opp != defaultp && opp != genfs_eopnotsupp &&
721 	    opp != genfs_badop && opp != genfs_nullop;
722 }
723 
724 /*
725  * Handes the read/write request given in 'bp' using the vnode's VOP_READ
726  * and VOP_WRITE operations.
727  *
728  * 'obp' is a pointer to the original request fed to the vnd device.
729  */
730 static void
731 handle_with_rdwr(struct vnd_softc *vnd, const struct buf *obp, struct buf *bp)
732 {
733 	bool doread;
734 	off_t offset;
735 	size_t resid;
736 	struct vnode *vp;
737 
738 	doread = bp->b_flags & B_READ;
739 	offset = obp->b_rawblkno * vnd->sc_dkdev.dk_label->d_secsize;
740 	vp = vnd->sc_vp;
741 
742 #if defined(DEBUG)
743 	if (vnddebug & VDB_IO)
744 		printf("vnd (rdwr): vp %p, %s, rawblkno 0x%" PRIx64
745 		    ", secsize %d, offset %" PRIu64
746 		    ", bcount %d\n",
747 		    vp, doread ? "read" : "write", obp->b_rawblkno,
748 		    vnd->sc_dkdev.dk_label->d_secsize, offset,
749 		    bp->b_bcount);
750 #endif
751 
752 	/* Issue the read or write operation. */
753 	bp->b_error =
754 	    vn_rdwr(doread ? UIO_READ : UIO_WRITE,
755 	    vp, bp->b_data, bp->b_bcount, offset,
756 	    UIO_SYSSPACE, 0, vnd->sc_cred, &resid, NULL);
757 	bp->b_resid = resid;
758 
759 	/* We need to increase the number of outputs on the vnode if
760 	 * there was any write to it. */
761 	if (!doread) {
762 		mutex_enter(&vp->v_interlock);
763 		vp->v_numoutput++;
764 		mutex_exit(&vp->v_interlock);
765 	}
766 
767 	biodone(bp);
768 }
769 
770 /*
771  * Handes the read/write request given in 'bp' using the vnode's VOP_BMAP
772  * and VOP_STRATEGY operations.
773  *
774  * 'obp' is a pointer to the original request fed to the vnd device.
775  */
776 static void
777 handle_with_strategy(struct vnd_softc *vnd, const struct buf *obp,
778     struct buf *bp)
779 {
780 	int bsize, error, flags, skipped;
781 	size_t resid, sz;
782 	off_t bn, offset;
783 	struct vnode *vp;
784 
785 	flags = obp->b_flags;
786 
787 	if (!(flags & B_READ)) {
788 		vp = bp->b_vp;
789 		mutex_enter(&vp->v_interlock);
790 		vp->v_numoutput++;
791 		mutex_exit(&vp->v_interlock);
792 	}
793 
794 	/* convert to a byte offset within the file. */
795 	bn = obp->b_rawblkno * vnd->sc_dkdev.dk_label->d_secsize;
796 
797 	bsize = vnd->sc_vp->v_mount->mnt_stat.f_iosize;
798 	skipped = 0;
799 
800 	/*
801 	 * Break the request into bsize pieces and feed them
802 	 * sequentially using VOP_BMAP/VOP_STRATEGY.
803 	 * We do it this way to keep from flooding NFS servers if we
804 	 * are connected to an NFS file.  This places the burden on
805 	 * the client rather than the server.
806 	 */
807 	error = 0;
808 	bp->b_resid = bp->b_bcount;
809 	for (offset = 0, resid = bp->b_resid; resid;
810 	    resid -= sz, offset += sz) {
811 		struct buf *nbp;
812 		daddr_t nbn;
813 		int off, nra;
814 
815 		nra = 0;
816 		vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
817 		error = VOP_BMAP(vnd->sc_vp, bn / bsize, &vp, &nbn, &nra);
818 		VOP_UNLOCK(vnd->sc_vp, 0);
819 
820 		if (error == 0 && (long)nbn == -1)
821 			error = EIO;
822 
823 		/*
824 		 * If there was an error or a hole in the file...punt.
825 		 * Note that we may have to wait for any operations
826 		 * that we have already fired off before releasing
827 		 * the buffer.
828 		 *
829 		 * XXX we could deal with holes here but it would be
830 		 * a hassle (in the write case).
831 		 */
832 		if (error) {
833 			skipped += resid;
834 			break;
835 		}
836 
837 #ifdef DEBUG
838 		if (!dovndcluster)
839 			nra = 0;
840 #endif
841 
842 		off = bn % bsize;
843 		sz = MIN(((off_t)1 + nra) * bsize - off, resid);
844 #ifdef	DEBUG
845 		if (vnddebug & VDB_IO)
846 			printf("vndstrategy: vp %p/%p bn 0x%qx/0x%" PRIx64
847 			    " sz 0x%zx\n", vnd->sc_vp, vp, (long long)bn,
848 			    nbn, sz);
849 #endif
850 
851 		nbp = getiobuf(vp, true);
852 		nestiobuf_setup(bp, nbp, offset, sz);
853 		nbp->b_blkno = nbn + btodb(off);
854 
855 #if 0 /* XXX #ifdef DEBUG */
856 		if (vnddebug & VDB_IO)
857 			printf("vndstart(%ld): bp %p vp %p blkno "
858 			    "0x%" PRIx64 " flags %x addr %p cnt 0x%x\n",
859 			    (long) (vnd-vnd_softc), &nbp->vb_buf,
860 			    nbp->vb_buf.b_vp, nbp->vb_buf.b_blkno,
861 			    nbp->vb_buf.b_flags, nbp->vb_buf.b_data,
862 			    nbp->vb_buf.b_bcount);
863 #endif
864 		VOP_STRATEGY(vp, nbp);
865 		bn += sz;
866 	}
867 	nestiobuf_done(bp, skipped, error);
868 }
869 
870 static void
871 vndiodone(struct buf *bp)
872 {
873 	struct vndxfer *vnx = VND_BUFTOXFER(bp);
874 	struct vnd_softc *vnd = vnx->vx_vnd;
875 	struct buf *obp = bp->b_private;
876 
877 	KASSERT(&vnx->vx_buf == bp);
878 	KASSERT(vnd->sc_active > 0);
879 #ifdef DEBUG
880 	if (vnddebug & VDB_IO) {
881 		printf("vndiodone1: bp %p iodone: error %d\n",
882 		    bp, bp->b_error);
883 	}
884 #endif
885 	disk_unbusy(&vnd->sc_dkdev, bp->b_bcount - bp->b_resid,
886 	    (bp->b_flags & B_READ));
887 	vnd->sc_active--;
888 	if (vnd->sc_active == 0) {
889 		wakeup(&vnd->sc_tab);
890 	}
891 	obp->b_error = bp->b_error;
892 	obp->b_resid = bp->b_resid;
893 	buf_destroy(bp);
894 	VND_PUTXFER(vnd, vnx);
895 	biodone(obp);
896 }
897 
898 /* ARGSUSED */
899 static int
900 vndread(dev_t dev, struct uio *uio, int flags)
901 {
902 	int unit = vndunit(dev);
903 	struct vnd_softc *sc;
904 
905 #ifdef DEBUG
906 	if (vnddebug & VDB_FOLLOW)
907 		printf("vndread(0x%"PRIx64", %p)\n", dev, uio);
908 #endif
909 
910 	sc = device_lookup_private(&vnd_cd, unit);
911 	if (sc == NULL)
912 		return ENXIO;
913 
914 	if ((sc->sc_flags & VNF_INITED) == 0)
915 		return ENXIO;
916 
917 	return physio(vndstrategy, NULL, dev, B_READ, minphys, uio);
918 }
919 
920 /* ARGSUSED */
921 static int
922 vndwrite(dev_t dev, struct uio *uio, int flags)
923 {
924 	int unit = vndunit(dev);
925 	struct vnd_softc *sc;
926 
927 #ifdef DEBUG
928 	if (vnddebug & VDB_FOLLOW)
929 		printf("vndwrite(0x%"PRIx64", %p)\n", dev, uio);
930 #endif
931 
932 	sc = device_lookup_private(&vnd_cd, unit);
933 	if (sc == NULL)
934 		return ENXIO;
935 
936 	if ((sc->sc_flags & VNF_INITED) == 0)
937 		return ENXIO;
938 
939 	return physio(vndstrategy, NULL, dev, B_WRITE, minphys, uio);
940 }
941 
942 static int
943 vnd_cget(struct lwp *l, int unit, int *un, struct vattr *va)
944 {
945 	struct vnd_softc *vnd;
946 
947 	if (*un == -1)
948 		*un = unit;
949 	if (*un < 0)
950 		return EINVAL;
951 
952 	vnd = device_lookup_private(&vnd_cd, *un);
953 	if (vnd == NULL)
954 		return (*un >= vnd_cd.cd_ndevs) ? ENXIO : -1;
955 
956 	if ((vnd->sc_flags & VNF_INITED) == 0)
957 		return -1;
958 
959 	return VOP_GETATTR(vnd->sc_vp, va, l->l_cred);
960 }
961 
962 static int
963 vnddoclear(struct vnd_softc *vnd, int pmask, int minor, bool force)
964 {
965 	int error;
966 
967 	if ((error = vndlock(vnd)) != 0)
968 		return error;
969 
970 	/*
971 	 * Don't unconfigure if any other partitions are open
972 	 * or if both the character and block flavors of this
973 	 * partition are open.
974 	 */
975 	if (((vnd->sc_dkdev.dk_openmask & ~pmask) ||
976 	    ((vnd->sc_dkdev.dk_bopenmask & pmask) &&
977 	    (vnd->sc_dkdev.dk_copenmask & pmask))) && !force) {
978 		vndunlock(vnd);
979 		return EBUSY;
980 	}
981 
982 	/*
983 	 * XXX vndclear() might call vndclose() implicitly;
984 	 * release lock to avoid recursion
985 	 *
986 	 * Set VNF_CLEARING to prevent vndopen() from
987 	 * sneaking in after we vndunlock().
988 	 */
989 	vnd->sc_flags |= VNF_CLEARING;
990 	vndunlock(vnd);
991 	vndclear(vnd, minor);
992 #ifdef DEBUG
993 	if (vnddebug & VDB_INIT)
994 		printf("vndioctl: CLRed\n");
995 #endif
996 
997 	/* Destroy the xfer and buffer pools. */
998 	pool_destroy(&vnd->sc_vxpool);
999 
1000 	/* Detach the disk. */
1001 	disk_detach(&vnd->sc_dkdev);
1002 
1003 	return 0;
1004 }
1005 
1006 /* ARGSUSED */
1007 static int
1008 vndioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l)
1009 {
1010 	bool force;
1011 	int unit = vndunit(dev);
1012 	struct vnd_softc *vnd;
1013 	struct vnd_ioctl *vio;
1014 	struct vattr vattr;
1015 	struct nameidata nd;
1016 	int error, part, pmask;
1017 	size_t geomsize;
1018 	int fflags;
1019 #ifdef __HAVE_OLD_DISKLABEL
1020 	struct disklabel newlabel;
1021 #endif
1022 
1023 #ifdef DEBUG
1024 	if (vnddebug & VDB_FOLLOW)
1025 		printf("vndioctl(0x%"PRIx64", 0x%lx, %p, 0x%x, %p): unit %d\n",
1026 		    dev, cmd, data, flag, l->l_proc, unit);
1027 #endif
1028 	vnd = device_lookup_private(&vnd_cd, unit);
1029 	if (vnd == NULL &&
1030 #ifdef COMPAT_30
1031 	    cmd != VNDIOOCGET &&
1032 #endif
1033 	    cmd != VNDIOCGET)
1034 		return ENXIO;
1035 	vio = (struct vnd_ioctl *)data;
1036 
1037 	/* Must be open for writes for these commands... */
1038 	switch (cmd) {
1039 	case VNDIOCSET:
1040 	case VNDIOCCLR:
1041 	case DIOCSDINFO:
1042 	case DIOCWDINFO:
1043 #ifdef __HAVE_OLD_DISKLABEL
1044 	case ODIOCSDINFO:
1045 	case ODIOCWDINFO:
1046 #endif
1047 	case DIOCKLABEL:
1048 	case DIOCWLABEL:
1049 		if ((flag & FWRITE) == 0)
1050 			return EBADF;
1051 	}
1052 
1053 	/* Must be initialized for these... */
1054 	switch (cmd) {
1055 	case VNDIOCCLR:
1056 	case DIOCGDINFO:
1057 	case DIOCSDINFO:
1058 	case DIOCWDINFO:
1059 	case DIOCGPART:
1060 	case DIOCKLABEL:
1061 	case DIOCWLABEL:
1062 	case DIOCGDEFLABEL:
1063 	case DIOCCACHESYNC:
1064 #ifdef __HAVE_OLD_DISKLABEL
1065 	case ODIOCGDINFO:
1066 	case ODIOCSDINFO:
1067 	case ODIOCWDINFO:
1068 	case ODIOCGDEFLABEL:
1069 #endif
1070 		if ((vnd->sc_flags & VNF_INITED) == 0)
1071 			return ENXIO;
1072 	}
1073 
1074 	switch (cmd) {
1075 	case VNDIOCSET:
1076 		if (vnd->sc_flags & VNF_INITED)
1077 			return EBUSY;
1078 
1079 		if ((error = vndlock(vnd)) != 0)
1080 			return error;
1081 
1082 		fflags = FREAD;
1083 		if ((vio->vnd_flags & VNDIOF_READONLY) == 0)
1084 			fflags |= FWRITE;
1085 		NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, vio->vnd_file);
1086 		if ((error = vn_open(&nd, fflags, 0)) != 0)
1087 			goto unlock_and_exit;
1088 		KASSERT(l);
1089 		error = VOP_GETATTR(nd.ni_vp, &vattr, l->l_cred);
1090 		if (!error && nd.ni_vp->v_type != VREG)
1091 			error = EOPNOTSUPP;
1092 		if (error) {
1093 			VOP_UNLOCK(nd.ni_vp, 0);
1094 			goto close_and_exit;
1095 		}
1096 
1097 		/* If using a compressed file, initialize its info */
1098 		/* (or abort with an error if kernel has no compression) */
1099 		if (vio->vnd_flags & VNF_COMP) {
1100 #ifdef VND_COMPRESSION
1101 			struct vnd_comp_header *ch;
1102 			int i;
1103 			u_int32_t comp_size;
1104 			u_int32_t comp_maxsize;
1105 
1106 			/* allocate space for compresed file header */
1107 			ch = malloc(sizeof(struct vnd_comp_header),
1108 			M_TEMP, M_WAITOK);
1109 
1110 			/* read compressed file header */
1111 			error = vn_rdwr(UIO_READ, nd.ni_vp, (void *)ch,
1112 			  sizeof(struct vnd_comp_header), 0, UIO_SYSSPACE,
1113 			  IO_UNIT|IO_NODELOCKED, l->l_cred, NULL, NULL);
1114 			if (error) {
1115 				free(ch, M_TEMP);
1116 				VOP_UNLOCK(nd.ni_vp, 0);
1117 				goto close_and_exit;
1118 			}
1119 
1120 			/* save some header info */
1121 			vnd->sc_comp_blksz = ntohl(ch->block_size);
1122 			/* note last offset is the file byte size */
1123 			vnd->sc_comp_numoffs = ntohl(ch->num_blocks)+1;
1124 			free(ch, M_TEMP);
1125 			if (vnd->sc_comp_blksz == 0 ||
1126 			    vnd->sc_comp_blksz % DEV_BSIZE !=0) {
1127 				VOP_UNLOCK(nd.ni_vp, 0);
1128 				error = EINVAL;
1129 				goto close_and_exit;
1130 			}
1131 			if (sizeof(struct vnd_comp_header) +
1132 			  sizeof(u_int64_t) * vnd->sc_comp_numoffs >
1133 			  vattr.va_size) {
1134 				VOP_UNLOCK(nd.ni_vp, 0);
1135 				error = EINVAL;
1136 				goto close_and_exit;
1137 			}
1138 
1139 			/* set decompressed file size */
1140 			vattr.va_size =
1141 			    ((u_quad_t)vnd->sc_comp_numoffs - 1) *
1142 			     (u_quad_t)vnd->sc_comp_blksz;
1143 
1144 			/* allocate space for all the compressed offsets */
1145 			vnd->sc_comp_offsets =
1146 			malloc(sizeof(u_int64_t) * vnd->sc_comp_numoffs,
1147 			M_DEVBUF, M_WAITOK);
1148 
1149 			/* read in the offsets */
1150 			error = vn_rdwr(UIO_READ, nd.ni_vp,
1151 			  (void *)vnd->sc_comp_offsets,
1152 			  sizeof(u_int64_t) * vnd->sc_comp_numoffs,
1153 			  sizeof(struct vnd_comp_header), UIO_SYSSPACE,
1154 			  IO_UNIT|IO_NODELOCKED, l->l_cred, NULL, NULL);
1155 			if (error) {
1156 				VOP_UNLOCK(nd.ni_vp, 0);
1157 				goto close_and_exit;
1158 			}
1159 			/*
1160 			 * find largest block size (used for allocation limit).
1161 			 * Also convert offset to native byte order.
1162 			 */
1163 			comp_maxsize = 0;
1164 			for (i = 0; i < vnd->sc_comp_numoffs - 1; i++) {
1165 				vnd->sc_comp_offsets[i] =
1166 				  be64toh(vnd->sc_comp_offsets[i]);
1167 				comp_size = be64toh(vnd->sc_comp_offsets[i + 1])
1168 				  - vnd->sc_comp_offsets[i];
1169 				if (comp_size > comp_maxsize)
1170 					comp_maxsize = comp_size;
1171 			}
1172 			vnd->sc_comp_offsets[vnd->sc_comp_numoffs - 1] =
1173 			  be64toh(vnd->sc_comp_offsets[vnd->sc_comp_numoffs - 1]);
1174 
1175 			/* create compressed data buffer */
1176 			vnd->sc_comp_buff = malloc(comp_maxsize,
1177 			  M_DEVBUF, M_WAITOK);
1178 
1179 			/* create decompressed buffer */
1180 			vnd->sc_comp_decombuf = malloc(vnd->sc_comp_blksz,
1181 			  M_DEVBUF, M_WAITOK);
1182 			vnd->sc_comp_buffblk = -1;
1183 
1184 			/* Initialize decompress stream */
1185 			memset(&vnd->sc_comp_stream, 0, sizeof(z_stream));
1186 			vnd->sc_comp_stream.zalloc = vnd_alloc;
1187 			vnd->sc_comp_stream.zfree = vnd_free;
1188 			error = inflateInit2(&vnd->sc_comp_stream, MAX_WBITS);
1189 			if (error) {
1190 				if (vnd->sc_comp_stream.msg)
1191 					printf("vnd%d: compressed file, %s\n",
1192 					  unit, vnd->sc_comp_stream.msg);
1193 				VOP_UNLOCK(nd.ni_vp, 0);
1194 				error = EINVAL;
1195 				goto close_and_exit;
1196 			}
1197 
1198 			vnd->sc_flags |= VNF_COMP | VNF_READONLY;
1199 #else /* !VND_COMPRESSION */
1200 			VOP_UNLOCK(nd.ni_vp, 0);
1201 			error = EOPNOTSUPP;
1202 			goto close_and_exit;
1203 #endif /* VND_COMPRESSION */
1204 		}
1205 
1206 		VOP_UNLOCK(nd.ni_vp, 0);
1207 		vnd->sc_vp = nd.ni_vp;
1208 		vnd->sc_size = btodb(vattr.va_size);	/* note truncation */
1209 
1210 		/*
1211 		 * Use pseudo-geometry specified.  If none was provided,
1212 		 * use "standard" Adaptec fictitious geometry.
1213 		 */
1214 		if (vio->vnd_flags & VNDIOF_HASGEOM) {
1215 
1216 			memcpy(&vnd->sc_geom, &vio->vnd_geom,
1217 			    sizeof(vio->vnd_geom));
1218 
1219 			/*
1220 			 * Sanity-check the sector size.
1221 			 * XXX Don't allow secsize < DEV_BSIZE.	 Should
1222 			 * XXX we?
1223 			 */
1224 			if (vnd->sc_geom.vng_secsize < DEV_BSIZE ||
1225 			    (vnd->sc_geom.vng_secsize % DEV_BSIZE) != 0 ||
1226 			    vnd->sc_geom.vng_ncylinders == 0 ||
1227 			    (vnd->sc_geom.vng_ntracks *
1228 			     vnd->sc_geom.vng_nsectors) == 0) {
1229 				error = EINVAL;
1230 				goto close_and_exit;
1231 			}
1232 
1233 			/*
1234 			 * Compute the size (in DEV_BSIZE blocks) specified
1235 			 * by the geometry.
1236 			 */
1237 			geomsize = (vnd->sc_geom.vng_nsectors *
1238 			    vnd->sc_geom.vng_ntracks *
1239 			    vnd->sc_geom.vng_ncylinders) *
1240 			    (vnd->sc_geom.vng_secsize / DEV_BSIZE);
1241 
1242 			/*
1243 			 * Sanity-check the size against the specified
1244 			 * geometry.
1245 			 */
1246 			if (vnd->sc_size < geomsize) {
1247 				error = EINVAL;
1248 				goto close_and_exit;
1249 			}
1250 		} else if (vnd->sc_size >= (32 * 64)) {
1251 			/*
1252 			 * Size must be at least 2048 DEV_BSIZE blocks
1253 			 * (1M) in order to use this geometry.
1254 			 */
1255 			vnd->sc_geom.vng_secsize = DEV_BSIZE;
1256 			vnd->sc_geom.vng_nsectors = 32;
1257 			vnd->sc_geom.vng_ntracks = 64;
1258 			vnd->sc_geom.vng_ncylinders = vnd->sc_size / (64 * 32);
1259 		} else {
1260 			vnd->sc_geom.vng_secsize = DEV_BSIZE;
1261 			vnd->sc_geom.vng_nsectors = 1;
1262 			vnd->sc_geom.vng_ntracks = 1;
1263 			vnd->sc_geom.vng_ncylinders = vnd->sc_size;
1264 		}
1265 
1266 		vnd_set_properties(vnd);
1267 
1268 		if (vio->vnd_flags & VNDIOF_READONLY) {
1269 			vnd->sc_flags |= VNF_READONLY;
1270 		}
1271 
1272 		if ((error = vndsetcred(vnd, l->l_cred)) != 0)
1273 			goto close_and_exit;
1274 
1275 		vndthrottle(vnd, vnd->sc_vp);
1276 		vio->vnd_size = dbtob(vnd->sc_size);
1277 		vnd->sc_flags |= VNF_INITED;
1278 
1279 		/* create the kernel thread, wait for it to be up */
1280 		error = kthread_create(PRI_NONE, 0, NULL, vndthread, vnd,
1281 		    &vnd->sc_kthread, device_xname(vnd->sc_dev));
1282 		if (error)
1283 			goto close_and_exit;
1284 		while ((vnd->sc_flags & VNF_KTHREAD) == 0) {
1285 			tsleep(&vnd->sc_kthread, PRIBIO, "vndthr", 0);
1286 		}
1287 #ifdef DEBUG
1288 		if (vnddebug & VDB_INIT)
1289 			printf("vndioctl: SET vp %p size 0x%lx %d/%d/%d/%d\n",
1290 			    vnd->sc_vp, (unsigned long) vnd->sc_size,
1291 			    vnd->sc_geom.vng_secsize,
1292 			    vnd->sc_geom.vng_nsectors,
1293 			    vnd->sc_geom.vng_ntracks,
1294 			    vnd->sc_geom.vng_ncylinders);
1295 #endif
1296 
1297 		/* Attach the disk. */
1298 		disk_attach(&vnd->sc_dkdev);
1299 
1300 		/* Initialize the xfer and buffer pools. */
1301 		pool_init(&vnd->sc_vxpool, sizeof(struct vndxfer), 0,
1302 		    0, 0, "vndxpl", NULL, IPL_BIO);
1303 
1304 		/* Try and read the disklabel. */
1305 		vndgetdisklabel(dev, vnd);
1306 
1307 		vndunlock(vnd);
1308 
1309 		break;
1310 
1311 close_and_exit:
1312 		(void) vn_close(nd.ni_vp, fflags, l->l_cred);
1313 unlock_and_exit:
1314 #ifdef VND_COMPRESSION
1315 		/* free any allocated memory (for compressed file) */
1316 		if (vnd->sc_comp_offsets) {
1317 			free(vnd->sc_comp_offsets, M_DEVBUF);
1318 			vnd->sc_comp_offsets = NULL;
1319 		}
1320 		if (vnd->sc_comp_buff) {
1321 			free(vnd->sc_comp_buff, M_DEVBUF);
1322 			vnd->sc_comp_buff = NULL;
1323 		}
1324 		if (vnd->sc_comp_decombuf) {
1325 			free(vnd->sc_comp_decombuf, M_DEVBUF);
1326 			vnd->sc_comp_decombuf = NULL;
1327 		}
1328 #endif /* VND_COMPRESSION */
1329 		vndunlock(vnd);
1330 		return error;
1331 
1332 	case VNDIOCCLR:
1333 		part = DISKPART(dev);
1334 		pmask = (1 << part);
1335 		force = (vio->vnd_flags & VNDIOF_FORCE) != 0;
1336 
1337 		if ((error = vnddoclear(vnd, pmask, minor(dev), force)) != 0)
1338 			return error;
1339 
1340 		break;
1341 
1342 #ifdef COMPAT_30
1343 	case VNDIOOCGET: {
1344 		struct vnd_ouser *vnu;
1345 		struct vattr va;
1346 		vnu = (struct vnd_ouser *)data;
1347 		KASSERT(l);
1348 		switch (error = vnd_cget(l, unit, &vnu->vnu_unit, &va)) {
1349 		case 0:
1350 			vnu->vnu_dev = va.va_fsid;
1351 			vnu->vnu_ino = va.va_fileid;
1352 			break;
1353 		case -1:
1354 			/* unused is not an error */
1355 			vnu->vnu_dev = 0;
1356 			vnu->vnu_ino = 0;
1357 			break;
1358 		default:
1359 			return error;
1360 		}
1361 		break;
1362 	}
1363 #endif
1364 	case VNDIOCGET: {
1365 		struct vnd_user *vnu;
1366 		struct vattr va;
1367 		vnu = (struct vnd_user *)data;
1368 		KASSERT(l);
1369 		switch (error = vnd_cget(l, unit, &vnu->vnu_unit, &va)) {
1370 		case 0:
1371 			vnu->vnu_dev = va.va_fsid;
1372 			vnu->vnu_ino = va.va_fileid;
1373 			break;
1374 		case -1:
1375 			/* unused is not an error */
1376 			vnu->vnu_dev = 0;
1377 			vnu->vnu_ino = 0;
1378 			break;
1379 		default:
1380 			return error;
1381 		}
1382 		break;
1383 	}
1384 
1385 	case DIOCGDINFO:
1386 		*(struct disklabel *)data = *(vnd->sc_dkdev.dk_label);
1387 		break;
1388 
1389 #ifdef __HAVE_OLD_DISKLABEL
1390 	case ODIOCGDINFO:
1391 		newlabel = *(vnd->sc_dkdev.dk_label);
1392 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1393 			return ENOTTY;
1394 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1395 		break;
1396 #endif
1397 
1398 	case DIOCGPART:
1399 		((struct partinfo *)data)->disklab = vnd->sc_dkdev.dk_label;
1400 		((struct partinfo *)data)->part =
1401 		    &vnd->sc_dkdev.dk_label->d_partitions[DISKPART(dev)];
1402 		break;
1403 
1404 	case DIOCWDINFO:
1405 	case DIOCSDINFO:
1406 #ifdef __HAVE_OLD_DISKLABEL
1407 	case ODIOCWDINFO:
1408 	case ODIOCSDINFO:
1409 #endif
1410 	{
1411 		struct disklabel *lp;
1412 
1413 		if ((error = vndlock(vnd)) != 0)
1414 			return error;
1415 
1416 		vnd->sc_flags |= VNF_LABELLING;
1417 
1418 #ifdef __HAVE_OLD_DISKLABEL
1419 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
1420 			memset(&newlabel, 0, sizeof newlabel);
1421 			memcpy(&newlabel, data, sizeof (struct olddisklabel));
1422 			lp = &newlabel;
1423 		} else
1424 #endif
1425 		lp = (struct disklabel *)data;
1426 
1427 		error = setdisklabel(vnd->sc_dkdev.dk_label,
1428 		    lp, 0, vnd->sc_dkdev.dk_cpulabel);
1429 		if (error == 0) {
1430 			if (cmd == DIOCWDINFO
1431 #ifdef __HAVE_OLD_DISKLABEL
1432 			    || cmd == ODIOCWDINFO
1433 #endif
1434 			   )
1435 				error = writedisklabel(VNDLABELDEV(dev),
1436 				    vndstrategy, vnd->sc_dkdev.dk_label,
1437 				    vnd->sc_dkdev.dk_cpulabel);
1438 		}
1439 
1440 		vnd->sc_flags &= ~VNF_LABELLING;
1441 
1442 		vndunlock(vnd);
1443 
1444 		if (error)
1445 			return error;
1446 		break;
1447 	}
1448 
1449 	case DIOCKLABEL:
1450 		if (*(int *)data != 0)
1451 			vnd->sc_flags |= VNF_KLABEL;
1452 		else
1453 			vnd->sc_flags &= ~VNF_KLABEL;
1454 		break;
1455 
1456 	case DIOCWLABEL:
1457 		if (*(int *)data != 0)
1458 			vnd->sc_flags |= VNF_WLABEL;
1459 		else
1460 			vnd->sc_flags &= ~VNF_WLABEL;
1461 		break;
1462 
1463 	case DIOCGDEFLABEL:
1464 		vndgetdefaultlabel(vnd, (struct disklabel *)data);
1465 		break;
1466 
1467 #ifdef __HAVE_OLD_DISKLABEL
1468 	case ODIOCGDEFLABEL:
1469 		vndgetdefaultlabel(vnd, &newlabel);
1470 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1471 			return ENOTTY;
1472 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1473 		break;
1474 #endif
1475 
1476 	case DIOCCACHESYNC:
1477 		vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1478 		error = VOP_FSYNC(vnd->sc_vp, vnd->sc_cred,
1479 		    FSYNC_WAIT | FSYNC_DATAONLY | FSYNC_CACHE, 0, 0);
1480 		VOP_UNLOCK(vnd->sc_vp, 0);
1481 		return error;
1482 
1483 	default:
1484 		return ENOTTY;
1485 	}
1486 
1487 	return 0;
1488 }
1489 
1490 /*
1491  * Duplicate the current processes' credentials.  Since we are called only
1492  * as the result of a SET ioctl and only root can do that, any future access
1493  * to this "disk" is essentially as root.  Note that credentials may change
1494  * if some other uid can write directly to the mapped file (NFS).
1495  */
1496 static int
1497 vndsetcred(struct vnd_softc *vnd, kauth_cred_t cred)
1498 {
1499 	struct uio auio;
1500 	struct iovec aiov;
1501 	char *tmpbuf;
1502 	int error;
1503 
1504 	vnd->sc_cred = kauth_cred_dup(cred);
1505 	tmpbuf = malloc(DEV_BSIZE, M_TEMP, M_WAITOK);
1506 
1507 	/* XXX: Horrible kludge to establish credentials for NFS */
1508 	aiov.iov_base = tmpbuf;
1509 	aiov.iov_len = min(DEV_BSIZE, dbtob(vnd->sc_size));
1510 	auio.uio_iov = &aiov;
1511 	auio.uio_iovcnt = 1;
1512 	auio.uio_offset = 0;
1513 	auio.uio_rw = UIO_READ;
1514 	auio.uio_resid = aiov.iov_len;
1515 	UIO_SETUP_SYSSPACE(&auio);
1516 	vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1517 	error = VOP_READ(vnd->sc_vp, &auio, 0, vnd->sc_cred);
1518 	if (error == 0) {
1519 		/*
1520 		 * Because vnd does all IO directly through the vnode
1521 		 * we need to flush (at least) the buffer from the above
1522 		 * VOP_READ from the buffer cache to prevent cache
1523 		 * incoherencies.  Also, be careful to write dirty
1524 		 * buffers back to stable storage.
1525 		 */
1526 		error = vinvalbuf(vnd->sc_vp, V_SAVE, vnd->sc_cred,
1527 			    curlwp, 0, 0);
1528 	}
1529 	VOP_UNLOCK(vnd->sc_vp, 0);
1530 
1531 	free(tmpbuf, M_TEMP);
1532 	return error;
1533 }
1534 
1535 /*
1536  * Set maxactive based on FS type
1537  */
1538 static void
1539 vndthrottle(struct vnd_softc *vnd, struct vnode *vp)
1540 {
1541 #ifdef NFS
1542 	extern int (**nfsv2_vnodeop_p)(void *);
1543 
1544 	if (vp->v_op == nfsv2_vnodeop_p)
1545 		vnd->sc_maxactive = 2;
1546 	else
1547 #endif
1548 		vnd->sc_maxactive = 8;
1549 
1550 	if (vnd->sc_maxactive < 1)
1551 		vnd->sc_maxactive = 1;
1552 }
1553 
1554 #if 0
1555 static void
1556 vndshutdown(void)
1557 {
1558 	struct vnd_softc *vnd;
1559 
1560 	for (vnd = &vnd_softc[0]; vnd < &vnd_softc[numvnd]; vnd++)
1561 		if (vnd->sc_flags & VNF_INITED)
1562 			vndclear(vnd);
1563 }
1564 #endif
1565 
1566 static void
1567 vndclear(struct vnd_softc *vnd, int myminor)
1568 {
1569 	struct vnode *vp = vnd->sc_vp;
1570 	int fflags = FREAD;
1571 	int bmaj, cmaj, i, mn;
1572 	int s;
1573 
1574 #ifdef DEBUG
1575 	if (vnddebug & VDB_FOLLOW)
1576 		printf("vndclear(%p): vp %p\n", vnd, vp);
1577 #endif
1578 	/* locate the major number */
1579 	bmaj = bdevsw_lookup_major(&vnd_bdevsw);
1580 	cmaj = cdevsw_lookup_major(&vnd_cdevsw);
1581 
1582 	/* Nuke the vnodes for any open instances */
1583 	for (i = 0; i < MAXPARTITIONS; i++) {
1584 		mn = DISKMINOR(device_unit(vnd->sc_dev), i);
1585 		vdevgone(bmaj, mn, mn, VBLK);
1586 		if (mn != myminor) /* XXX avoid to kill own vnode */
1587 			vdevgone(cmaj, mn, mn, VCHR);
1588 	}
1589 
1590 	if ((vnd->sc_flags & VNF_READONLY) == 0)
1591 		fflags |= FWRITE;
1592 
1593 	s = splbio();
1594 	bufq_drain(vnd->sc_tab);
1595 	splx(s);
1596 
1597 	vnd->sc_flags |= VNF_VUNCONF;
1598 	wakeup(&vnd->sc_tab);
1599 	while (vnd->sc_flags & VNF_KTHREAD)
1600 		tsleep(&vnd->sc_kthread, PRIBIO, "vnthr", 0);
1601 
1602 #ifdef VND_COMPRESSION
1603 	/* free the compressed file buffers */
1604 	if (vnd->sc_flags & VNF_COMP) {
1605 		if (vnd->sc_comp_offsets) {
1606 			free(vnd->sc_comp_offsets, M_DEVBUF);
1607 			vnd->sc_comp_offsets = NULL;
1608 		}
1609 		if (vnd->sc_comp_buff) {
1610 			free(vnd->sc_comp_buff, M_DEVBUF);
1611 			vnd->sc_comp_buff = NULL;
1612 		}
1613 		if (vnd->sc_comp_decombuf) {
1614 			free(vnd->sc_comp_decombuf, M_DEVBUF);
1615 			vnd->sc_comp_decombuf = NULL;
1616 		}
1617 	}
1618 #endif /* VND_COMPRESSION */
1619 	vnd->sc_flags &=
1620 	    ~(VNF_INITED | VNF_READONLY | VNF_VLABEL
1621 	      | VNF_VUNCONF | VNF_COMP | VNF_CLEARING);
1622 	if (vp == NULL)
1623 		panic("vndclear: null vp");
1624 	(void) vn_close(vp, fflags, vnd->sc_cred);
1625 	kauth_cred_free(vnd->sc_cred);
1626 	vnd->sc_vp = NULL;
1627 	vnd->sc_cred = NULL;
1628 	vnd->sc_size = 0;
1629 }
1630 
1631 static int
1632 vndsize(dev_t dev)
1633 {
1634 	struct vnd_softc *sc;
1635 	struct disklabel *lp;
1636 	int part, unit, omask;
1637 	int size;
1638 
1639 	unit = vndunit(dev);
1640 	sc = device_lookup_private(&vnd_cd, unit);
1641 	if (sc == NULL)
1642 		return -1;
1643 
1644 	if ((sc->sc_flags & VNF_INITED) == 0)
1645 		return -1;
1646 
1647 	part = DISKPART(dev);
1648 	omask = sc->sc_dkdev.dk_openmask & (1 << part);
1649 	lp = sc->sc_dkdev.dk_label;
1650 
1651 	if (omask == 0 && vndopen(dev, 0, S_IFBLK, curlwp))	/* XXX */
1652 		return -1;
1653 
1654 	if (lp->d_partitions[part].p_fstype != FS_SWAP)
1655 		size = -1;
1656 	else
1657 		size = lp->d_partitions[part].p_size *
1658 		    (lp->d_secsize / DEV_BSIZE);
1659 
1660 	if (omask == 0 && vndclose(dev, 0, S_IFBLK, curlwp))	/* XXX */
1661 		return -1;
1662 
1663 	return size;
1664 }
1665 
1666 static int
1667 vnddump(dev_t dev, daddr_t blkno, void *va,
1668     size_t size)
1669 {
1670 
1671 	/* Not implemented. */
1672 	return ENXIO;
1673 }
1674 
1675 static void
1676 vndgetdefaultlabel(struct vnd_softc *sc, struct disklabel *lp)
1677 {
1678 	struct vndgeom *vng = &sc->sc_geom;
1679 	struct partition *pp;
1680 
1681 	memset(lp, 0, sizeof(*lp));
1682 
1683 	lp->d_secperunit = sc->sc_size / (vng->vng_secsize / DEV_BSIZE);
1684 	lp->d_secsize = vng->vng_secsize;
1685 	lp->d_nsectors = vng->vng_nsectors;
1686 	lp->d_ntracks = vng->vng_ntracks;
1687 	lp->d_ncylinders = vng->vng_ncylinders;
1688 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1689 
1690 	strncpy(lp->d_typename, "vnd", sizeof(lp->d_typename));
1691 	lp->d_type = DTYPE_VND;
1692 	strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
1693 	lp->d_rpm = 3600;
1694 	lp->d_interleave = 1;
1695 	lp->d_flags = 0;
1696 
1697 	pp = &lp->d_partitions[RAW_PART];
1698 	pp->p_offset = 0;
1699 	pp->p_size = lp->d_secperunit;
1700 	pp->p_fstype = FS_UNUSED;
1701 	lp->d_npartitions = RAW_PART + 1;
1702 
1703 	lp->d_magic = DISKMAGIC;
1704 	lp->d_magic2 = DISKMAGIC;
1705 	lp->d_checksum = dkcksum(lp);
1706 }
1707 
1708 /*
1709  * Read the disklabel from a vnd.  If one is not present, create a fake one.
1710  */
1711 static void
1712 vndgetdisklabel(dev_t dev, struct vnd_softc *sc)
1713 {
1714 	const char *errstring;
1715 	struct disklabel *lp = sc->sc_dkdev.dk_label;
1716 	struct cpu_disklabel *clp = sc->sc_dkdev.dk_cpulabel;
1717 	int i;
1718 
1719 	memset(clp, 0, sizeof(*clp));
1720 
1721 	vndgetdefaultlabel(sc, lp);
1722 
1723 	/*
1724 	 * Call the generic disklabel extraction routine.
1725 	 */
1726 	errstring = readdisklabel(VNDLABELDEV(dev), vndstrategy, lp, clp);
1727 	if (errstring) {
1728 		/*
1729 		 * Lack of disklabel is common, but we print the warning
1730 		 * anyway, since it might contain other useful information.
1731 		 */
1732 		aprint_normal_dev(sc->sc_dev, "%s\n", errstring);
1733 
1734 		/*
1735 		 * For historical reasons, if there's no disklabel
1736 		 * present, all partitions must be FS_BSDFFS and
1737 		 * occupy the entire disk.
1738 		 */
1739 		for (i = 0; i < MAXPARTITIONS; i++) {
1740 			/*
1741 			 * Don't wipe out port specific hack (such as
1742 			 * dos partition hack of i386 port).
1743 			 */
1744 			if (lp->d_partitions[i].p_size != 0)
1745 				continue;
1746 
1747 			lp->d_partitions[i].p_size = lp->d_secperunit;
1748 			lp->d_partitions[i].p_offset = 0;
1749 			lp->d_partitions[i].p_fstype = FS_BSDFFS;
1750 		}
1751 
1752 		strncpy(lp->d_packname, "default label",
1753 		    sizeof(lp->d_packname));
1754 
1755 		lp->d_npartitions = MAXPARTITIONS;
1756 		lp->d_checksum = dkcksum(lp);
1757 	}
1758 
1759 	/* In-core label now valid. */
1760 	sc->sc_flags |= VNF_VLABEL;
1761 }
1762 
1763 /*
1764  * Wait interruptibly for an exclusive lock.
1765  *
1766  * XXX
1767  * Several drivers do this; it should be abstracted and made MP-safe.
1768  */
1769 static int
1770 vndlock(struct vnd_softc *sc)
1771 {
1772 	int error;
1773 
1774 	while ((sc->sc_flags & VNF_LOCKED) != 0) {
1775 		sc->sc_flags |= VNF_WANTED;
1776 		if ((error = tsleep(sc, PRIBIO | PCATCH, "vndlck", 0)) != 0)
1777 			return error;
1778 	}
1779 	sc->sc_flags |= VNF_LOCKED;
1780 	return 0;
1781 }
1782 
1783 /*
1784  * Unlock and wake up any waiters.
1785  */
1786 static void
1787 vndunlock(struct vnd_softc *sc)
1788 {
1789 
1790 	sc->sc_flags &= ~VNF_LOCKED;
1791 	if ((sc->sc_flags & VNF_WANTED) != 0) {
1792 		sc->sc_flags &= ~VNF_WANTED;
1793 		wakeup(sc);
1794 	}
1795 }
1796 
1797 #ifdef VND_COMPRESSION
1798 /* compressed file read */
1799 static void
1800 compstrategy(struct buf *bp, off_t bn)
1801 {
1802 	int error;
1803 	int unit = vndunit(bp->b_dev);
1804 	struct vnd_softc *vnd =
1805 	    device_lookup_private(&vnd_cd, unit);
1806 	u_int32_t comp_block;
1807 	struct uio auio;
1808 	char *addr;
1809 	int s;
1810 
1811 	/* set up constants for data move */
1812 	auio.uio_rw = UIO_READ;
1813 	UIO_SETUP_SYSSPACE(&auio);
1814 
1815 	/* read, and transfer the data */
1816 	addr = bp->b_data;
1817 	bp->b_resid = bp->b_bcount;
1818 	s = splbio();
1819 	while (bp->b_resid > 0) {
1820 		unsigned length;
1821 		size_t length_in_buffer;
1822 		u_int32_t offset_in_buffer;
1823 		struct iovec aiov;
1824 
1825 		/* calculate the compressed block number */
1826 		comp_block = bn / (off_t)vnd->sc_comp_blksz;
1827 
1828 		/* check for good block number */
1829 		if (comp_block >= vnd->sc_comp_numoffs) {
1830 			bp->b_error = EINVAL;
1831 			splx(s);
1832 			return;
1833 		}
1834 
1835 		/* read in the compressed block, if not in buffer */
1836 		if (comp_block != vnd->sc_comp_buffblk) {
1837 			length = vnd->sc_comp_offsets[comp_block + 1] -
1838 			    vnd->sc_comp_offsets[comp_block];
1839 			vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1840 			error = vn_rdwr(UIO_READ, vnd->sc_vp, vnd->sc_comp_buff,
1841 			    length, vnd->sc_comp_offsets[comp_block],
1842 			    UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, vnd->sc_cred,
1843 			    NULL, NULL);
1844 			if (error) {
1845 				bp->b_error = error;
1846 				VOP_UNLOCK(vnd->sc_vp, 0);
1847 				splx(s);
1848 				return;
1849 			}
1850 			/* uncompress the buffer */
1851 			vnd->sc_comp_stream.next_in = vnd->sc_comp_buff;
1852 			vnd->sc_comp_stream.avail_in = length;
1853 			vnd->sc_comp_stream.next_out = vnd->sc_comp_decombuf;
1854 			vnd->sc_comp_stream.avail_out = vnd->sc_comp_blksz;
1855 			inflateReset(&vnd->sc_comp_stream);
1856 			error = inflate(&vnd->sc_comp_stream, Z_FINISH);
1857 			if (error != Z_STREAM_END) {
1858 				if (vnd->sc_comp_stream.msg)
1859 					aprint_normal_dev(vnd->sc_dev,
1860 					    "compressed file, %s\n",
1861 					    vnd->sc_comp_stream.msg);
1862 				bp->b_error = EBADMSG;
1863 				VOP_UNLOCK(vnd->sc_vp, 0);
1864 				splx(s);
1865 				return;
1866 			}
1867 			vnd->sc_comp_buffblk = comp_block;
1868 			VOP_UNLOCK(vnd->sc_vp, 0);
1869 		}
1870 
1871 		/* transfer the usable uncompressed data */
1872 		offset_in_buffer = bn % (off_t)vnd->sc_comp_blksz;
1873 		length_in_buffer = vnd->sc_comp_blksz - offset_in_buffer;
1874 		if (length_in_buffer > bp->b_resid)
1875 			length_in_buffer = bp->b_resid;
1876 		auio.uio_iov = &aiov;
1877 		auio.uio_iovcnt = 1;
1878 		aiov.iov_base = addr;
1879 		aiov.iov_len = length_in_buffer;
1880 		auio.uio_resid = aiov.iov_len;
1881 		auio.uio_offset = 0;
1882 		error = uiomove(vnd->sc_comp_decombuf + offset_in_buffer,
1883 		    length_in_buffer, &auio);
1884 		if (error) {
1885 			bp->b_error = error;
1886 			splx(s);
1887 			return;
1888 		}
1889 
1890 		bn += length_in_buffer;
1891 		addr += length_in_buffer;
1892 		bp->b_resid -= length_in_buffer;
1893 	}
1894 	splx(s);
1895 }
1896 
1897 /* compression memory allocation routines */
1898 static void *
1899 vnd_alloc(void *aux, u_int items, u_int siz)
1900 {
1901 	return malloc(items * siz, M_TEMP, M_NOWAIT);
1902 }
1903 
1904 static void
1905 vnd_free(void *aux, void *ptr)
1906 {
1907 	free(ptr, M_TEMP);
1908 }
1909 #endif /* VND_COMPRESSION */
1910 
1911 static void
1912 vnd_set_properties(struct vnd_softc *vnd)
1913 {
1914 	prop_dictionary_t disk_info, odisk_info, geom;
1915 
1916 	disk_info = prop_dictionary_create();
1917 
1918 	geom = prop_dictionary_create();
1919 
1920 	prop_dictionary_set_uint64(geom, "sectors-per-unit",
1921 	    vnd->sc_geom.vng_nsectors * vnd->sc_geom.vng_ntracks *
1922 	    vnd->sc_geom.vng_ncylinders);
1923 
1924 	prop_dictionary_set_uint32(geom, "sector-size",
1925 	    vnd->sc_geom.vng_secsize);
1926 
1927 	prop_dictionary_set_uint16(geom, "sectors-per-track",
1928 	    vnd->sc_geom.vng_nsectors);
1929 
1930 	prop_dictionary_set_uint16(geom, "tracks-per-cylinder",
1931 	    vnd->sc_geom.vng_ntracks);
1932 
1933 	prop_dictionary_set_uint64(geom, "cylinders-per-unit",
1934 	    vnd->sc_geom.vng_ncylinders);
1935 
1936 	prop_dictionary_set(disk_info, "geometry", geom);
1937 	prop_object_release(geom);
1938 
1939 	prop_dictionary_set(device_properties(vnd->sc_dev),
1940 	    "disk-info", disk_info);
1941 
1942 	/*
1943 	 * Don't release disk_info here; we keep a reference to it.
1944 	 * disk_detach() will release it when we go away.
1945 	 */
1946 
1947 	odisk_info = vnd->sc_dkdev.dk_info;
1948 	vnd->sc_dkdev.dk_info = disk_info;
1949 	if (odisk_info)
1950 		prop_object_release(odisk_info);
1951 }
1952 
1953 #ifdef _MODULE
1954 
1955 #include <sys/module.h>
1956 
1957 MODULE(MODULE_CLASS_DRIVER, vnd, NULL);
1958 CFDRIVER_DECL(vnd, DV_DISK, NULL);
1959 
1960 static int
1961 vnd_modcmd(modcmd_t cmd, void *arg)
1962 {
1963 	int bmajor = -1, cmajor = -1,  error = 0;
1964 
1965 	switch (cmd) {
1966 	case MODULE_CMD_INIT:
1967 		error = config_cfdriver_attach(&vnd_cd);
1968 		if (error)
1969 			break;
1970 
1971 		error = config_cfattach_attach(vnd_cd.cd_name, &vnd_ca);
1972 	        if (error) {
1973 			config_cfdriver_detach(&vnd_cd);
1974 			aprint_error("%s: unable to register cfattach\n",
1975 			    vnd_cd.cd_name);
1976 			break;
1977 		}
1978 
1979 		error = devsw_attach("vnd", &vnd_bdevsw, &bmajor,
1980 		    &vnd_cdevsw, &cmajor);
1981 		if (error) {
1982 			config_cfattach_detach(vnd_cd.cd_name, &vnd_ca);
1983 			config_cfdriver_detach(&vnd_cd);
1984 			break;
1985 		}
1986 
1987 		break;
1988 
1989 	case MODULE_CMD_FINI:
1990 		error = config_cfattach_detach(vnd_cd.cd_name, &vnd_ca);
1991 		if (error)
1992 			break;
1993 		config_cfdriver_detach(&vnd_cd);
1994 		devsw_detach(&vnd_bdevsw, &vnd_cdevsw);
1995 		break;
1996 
1997 	case MODULE_CMD_STAT:
1998 		return ENOTTY;
1999 
2000 	default:
2001 		return ENOTTY;
2002 	}
2003 
2004 	return error;
2005 }
2006 
2007 #endif
2008