xref: /netbsd-src/sys/dev/vnd.c (revision 3816d47b2c42fcd6e549e3407f842a5b1a1d23ad)
1 /*	$NetBSD: vnd.c,v 1.205 2009/12/06 16:33:18 dsl 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 1.205 2009/12/06 16:33:18 dsl 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/dkvar.h>
167 #include <dev/vndvar.h>
168 
169 #include <prop/proplib.h>
170 
171 #if defined(VNDDEBUG) && !defined(DEBUG)
172 #define DEBUG
173 #endif
174 
175 #ifdef DEBUG
176 int dovndcluster = 1;
177 #define VDB_FOLLOW	0x01
178 #define VDB_INIT	0x02
179 #define VDB_IO		0x04
180 #define VDB_LABEL	0x08
181 int vnddebug = 0x00;
182 #endif
183 
184 #define vndunit(x)	DISKUNIT(x)
185 
186 struct vndxfer {
187 	struct buf vx_buf;
188 	struct vnd_softc *vx_vnd;
189 };
190 #define	VND_BUFTOXFER(bp)	((struct vndxfer *)(void *)bp)
191 
192 #define VND_GETXFER(vnd)	pool_get(&(vnd)->sc_vxpool, PR_WAITOK)
193 #define VND_PUTXFER(vnd, vx)	pool_put(&(vnd)->sc_vxpool, (vx))
194 
195 #define VNDLABELDEV(dev) \
196     (MAKEDISKDEV(major((dev)), vndunit((dev)), RAW_PART))
197 
198 /* called by main() at boot time */
199 void	vndattach(int);
200 
201 static void	vndclear(struct vnd_softc *, int);
202 static int	vnddoclear(struct vnd_softc *, int, int, bool);
203 static int	vndsetcred(struct vnd_softc *, kauth_cred_t);
204 static void	vndthrottle(struct vnd_softc *, struct vnode *);
205 static void	vndiodone(struct buf *);
206 #if 0
207 static void	vndshutdown(void);
208 #endif
209 
210 static void	vndgetdefaultlabel(struct vnd_softc *, struct disklabel *);
211 static void	vndgetdisklabel(dev_t, struct vnd_softc *);
212 
213 static int	vndlock(struct vnd_softc *);
214 static void	vndunlock(struct vnd_softc *);
215 #ifdef VND_COMPRESSION
216 static void	compstrategy(struct buf *, off_t);
217 static void	*vnd_alloc(void *, u_int, u_int);
218 static void	vnd_free(void *, void *);
219 #endif /* VND_COMPRESSION */
220 
221 static void	vndthread(void *);
222 static bool	vnode_has_op(const struct vnode *, int);
223 static void	handle_with_rdwr(struct vnd_softc *, const struct buf *,
224 		    struct buf *);
225 static void	handle_with_strategy(struct vnd_softc *, const struct buf *,
226 		    struct buf *);
227 static void	vnd_set_properties(struct vnd_softc *);
228 
229 static dev_type_open(vndopen);
230 static dev_type_close(vndclose);
231 static dev_type_read(vndread);
232 static dev_type_write(vndwrite);
233 static dev_type_ioctl(vndioctl);
234 static dev_type_strategy(vndstrategy);
235 static dev_type_dump(vnddump);
236 static dev_type_size(vndsize);
237 
238 const struct bdevsw vnd_bdevsw = {
239 	vndopen, vndclose, vndstrategy, vndioctl, vnddump, vndsize, D_DISK
240 };
241 
242 const struct cdevsw vnd_cdevsw = {
243 	vndopen, vndclose, vndread, vndwrite, vndioctl,
244 	nostop, notty, nopoll, nommap, nokqfilter, D_DISK
245 };
246 
247 static int	vnd_match(device_t, cfdata_t, void *);
248 static void	vnd_attach(device_t, device_t, void *);
249 static int	vnd_detach(device_t, int);
250 
251 CFATTACH_DECL3_NEW(vnd, sizeof(struct vnd_softc),
252     vnd_match, vnd_attach, vnd_detach, NULL, NULL, NULL, DVF_DETACH_SHUTDOWN);
253 extern struct cfdriver vnd_cd;
254 
255 static struct vnd_softc	*vnd_spawn(int);
256 int	vnd_destroy(device_t);
257 
258 void
259 vndattach(int num)
260 {
261 	int error;
262 
263 	error = config_cfattach_attach(vnd_cd.cd_name, &vnd_ca);
264 	if (error)
265 		aprint_error("%s: unable to register cfattach\n",
266 		    vnd_cd.cd_name);
267 }
268 
269 static int
270 vnd_match(device_t self, cfdata_t cfdata, void *aux)
271 {
272 
273 	return 1;
274 }
275 
276 static void
277 vnd_attach(device_t parent, device_t self, void *aux)
278 {
279 	struct vnd_softc *sc = device_private(self);
280 
281 	sc->sc_dev = self;
282 	sc->sc_comp_offsets = NULL;
283 	sc->sc_comp_buff = NULL;
284 	sc->sc_comp_decombuf = NULL;
285 	bufq_alloc(&sc->sc_tab, "disksort", BUFQ_SORT_RAWBLOCK);
286 	disk_init(&sc->sc_dkdev, device_xname(self), NULL);
287 	if (!pmf_device_register(self, NULL, NULL))
288 		aprint_error_dev(self, "couldn't establish power handler\n");
289 }
290 
291 static int
292 vnd_detach(device_t self, int flags)
293 {
294 	int error;
295 	struct vnd_softc *sc = device_private(self);
296 
297 	if (sc->sc_flags & VNF_INITED) {
298 		error = vnddoclear(sc, 0, -1, (flags & DETACH_FORCE) != 0);
299 		if (error != 0)
300 			return error;
301 	}
302 
303 	pmf_device_deregister(self);
304 	bufq_free(sc->sc_tab);
305 	disk_destroy(&sc->sc_dkdev);
306 
307 	return 0;
308 }
309 
310 static struct vnd_softc *
311 vnd_spawn(int unit)
312 {
313 	cfdata_t cf;
314 
315 	cf = malloc(sizeof(*cf), M_DEVBUF, M_WAITOK);
316 	cf->cf_name = vnd_cd.cd_name;
317 	cf->cf_atname = vnd_cd.cd_name;
318 	cf->cf_unit = unit;
319 	cf->cf_fstate = FSTATE_STAR;
320 
321 	return device_private(config_attach_pseudo(cf));
322 }
323 
324 int
325 vnd_destroy(device_t dev)
326 {
327 	int error;
328 	cfdata_t cf;
329 
330 	cf = device_cfdata(dev);
331 	error = config_detach(dev, DETACH_QUIET);
332 	if (error)
333 		return error;
334 	free(cf, M_DEVBUF);
335 	return 0;
336 }
337 
338 static int
339 vndopen(dev_t dev, int flags, int mode, struct lwp *l)
340 {
341 	int unit = vndunit(dev);
342 	struct vnd_softc *sc;
343 	int error = 0, part, pmask;
344 	struct disklabel *lp;
345 
346 #ifdef DEBUG
347 	if (vnddebug & VDB_FOLLOW)
348 		printf("vndopen(0x%"PRIx64", 0x%x, 0x%x, %p)\n", dev, flags, mode, l);
349 #endif
350 	sc = device_lookup_private(&vnd_cd, unit);
351 	if (sc == NULL) {
352 		sc = vnd_spawn(unit);
353 		if (sc == NULL)
354 			return ENOMEM;
355 	}
356 
357 	if ((error = vndlock(sc)) != 0)
358 		return error;
359 
360 	if ((sc->sc_flags & VNF_CLEARING) != 0) {
361 		error = ENXIO;
362 		goto done;
363 	}
364 
365 	lp = sc->sc_dkdev.dk_label;
366 
367 	part = DISKPART(dev);
368 	pmask = (1 << part);
369 
370 	/*
371 	 * If we're initialized, check to see if there are any other
372 	 * open partitions.  If not, then it's safe to update the
373 	 * in-core disklabel.  Only read the disklabel if it is
374 	 * not already valid.
375 	 */
376 	if ((sc->sc_flags & (VNF_INITED|VNF_VLABEL)) == VNF_INITED &&
377 	    sc->sc_dkdev.dk_openmask == 0)
378 		vndgetdisklabel(dev, sc);
379 
380 	/* Check that the partitions exists. */
381 	if (part != RAW_PART) {
382 		if (((sc->sc_flags & VNF_INITED) == 0) ||
383 		    ((part >= lp->d_npartitions) ||
384 		     (lp->d_partitions[part].p_fstype == FS_UNUSED))) {
385 			error = ENXIO;
386 			goto done;
387 		}
388 	}
389 
390 	/* Prevent our unit from being unconfigured while open. */
391 	switch (mode) {
392 	case S_IFCHR:
393 		sc->sc_dkdev.dk_copenmask |= pmask;
394 		break;
395 
396 	case S_IFBLK:
397 		sc->sc_dkdev.dk_bopenmask |= pmask;
398 		break;
399 	}
400 	sc->sc_dkdev.dk_openmask =
401 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
402 
403  done:
404 	vndunlock(sc);
405 	return error;
406 }
407 
408 static int
409 vndclose(dev_t dev, int flags, int mode, struct lwp *l)
410 {
411 	int unit = vndunit(dev);
412 	struct vnd_softc *sc;
413 	int error = 0, part;
414 
415 #ifdef DEBUG
416 	if (vnddebug & VDB_FOLLOW)
417 		printf("vndclose(0x%"PRIx64", 0x%x, 0x%x, %p)\n", dev, flags, mode, l);
418 #endif
419 	sc = device_lookup_private(&vnd_cd, unit);
420 	if (sc == NULL)
421 		return ENXIO;
422 
423 	if ((error = vndlock(sc)) != 0)
424 		return error;
425 
426 	part = DISKPART(dev);
427 
428 	/* ...that much closer to allowing unconfiguration... */
429 	switch (mode) {
430 	case S_IFCHR:
431 		sc->sc_dkdev.dk_copenmask &= ~(1 << part);
432 		break;
433 
434 	case S_IFBLK:
435 		sc->sc_dkdev.dk_bopenmask &= ~(1 << part);
436 		break;
437 	}
438 	sc->sc_dkdev.dk_openmask =
439 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
440 
441 	vndunlock(sc);
442 
443 	if ((sc->sc_flags & VNF_INITED) == 0) {
444 		if ((error = vnd_destroy(sc->sc_dev)) != 0) {
445 			aprint_error_dev(sc->sc_dev,
446 			    "unable to detach instance\n");
447 			return error;
448 		}
449 	}
450 
451 	return 0;
452 }
453 
454 /*
455  * Queue the request, and wakeup the kernel thread to handle it.
456  */
457 static void
458 vndstrategy(struct buf *bp)
459 {
460 	int unit = vndunit(bp->b_dev);
461 	struct vnd_softc *vnd =
462 	    device_lookup_private(&vnd_cd, unit);
463 	struct disklabel *lp;
464 	daddr_t blkno;
465 	int s = splbio();
466 
467 	if (vnd == NULL) {
468 		bp->b_error = ENXIO;
469 		goto done;
470 	}
471 	lp = vnd->sc_dkdev.dk_label;
472 
473 	if ((vnd->sc_flags & VNF_INITED) == 0) {
474 		bp->b_error = ENXIO;
475 		goto done;
476 	}
477 
478 	/*
479 	 * The transfer must be a whole number of blocks.
480 	 */
481 	if ((bp->b_bcount % lp->d_secsize) != 0) {
482 		bp->b_error = EINVAL;
483 		goto done;
484 	}
485 
486 	/*
487 	 * check if we're read-only.
488 	 */
489 	if ((vnd->sc_flags & VNF_READONLY) && !(bp->b_flags & B_READ)) {
490 		bp->b_error = EACCES;
491 		goto done;
492 	}
493 
494 	/* If it's a nil transfer, wake up the top half now. */
495 	if (bp->b_bcount == 0) {
496 		goto done;
497 	}
498 
499 	/*
500 	 * Do bounds checking and adjust transfer.  If there's an error,
501 	 * the bounds check will flag that for us.
502 	 */
503 	if (DISKPART(bp->b_dev) == RAW_PART) {
504 		if (bounds_check_with_mediasize(bp, DEV_BSIZE,
505 		    vnd->sc_size) <= 0)
506 			goto done;
507 	} else {
508 		if (bounds_check_with_label(&vnd->sc_dkdev,
509 		    bp, vnd->sc_flags & (VNF_WLABEL|VNF_LABELLING)) <= 0)
510 			goto done;
511 	}
512 
513 	/*
514 	 * Put the block number in terms of the logical blocksize
515 	 * of the "device".
516 	 */
517 
518 	blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
519 
520 	/*
521 	 * Translate the partition-relative block number to an absolute.
522 	 */
523 	if (DISKPART(bp->b_dev) != RAW_PART) {
524 		struct partition *pp;
525 
526 		pp = &vnd->sc_dkdev.dk_label->d_partitions[
527 		    DISKPART(bp->b_dev)];
528 		blkno += pp->p_offset;
529 	}
530 	bp->b_rawblkno = blkno;
531 
532 #ifdef DEBUG
533 	if (vnddebug & VDB_FOLLOW)
534 		printf("vndstrategy(%p): unit %d\n", bp, unit);
535 #endif
536 	bufq_put(vnd->sc_tab, bp);
537 	wakeup(&vnd->sc_tab);
538 	splx(s);
539 	return;
540 
541 done:
542 	bp->b_resid = bp->b_bcount;
543 	biodone(bp);
544 	splx(s);
545 }
546 
547 static bool
548 vnode_has_strategy(struct vnd_softc *vnd)
549 {
550 	return vnode_has_op(vnd->sc_vp, VOFFSET(vop_bmap)) &&
551 	    vnode_has_op(vnd->sc_vp, VOFFSET(vop_strategy));
552 }
553 
554 /* XXX this function needs a reliable check to detect
555  * sparse files. Otherwise, bmap/strategy may be used
556  * and fail on non-allocated blocks. VOP_READ/VOP_WRITE
557  * works on sparse files.
558  */
559 #if notyet
560 static bool
561 vnode_strategy_probe(struct vnd_softc *vnd)
562 {
563 	int error;
564 	daddr_t nbn;
565 
566 	if (!vnode_has_strategy(vnd))
567 		return false;
568 
569 	/* Convert the first logical block number to its
570 	 * physical block number.
571 	 */
572 	error = 0;
573 	vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
574 	error = VOP_BMAP(vnd->sc_vp, 0, NULL, &nbn, NULL);
575 	VOP_UNLOCK(vnd->sc_vp, 0);
576 
577 	/* Test if that worked. */
578 	if (error == 0 && (long)nbn == -1)
579 		return false;
580 
581 	return true;
582 }
583 #endif
584 
585 static void
586 vndthread(void *arg)
587 {
588 	struct vnd_softc *vnd = arg;
589 	bool usestrategy;
590 	int s;
591 
592 	/* Determine whether we can *use* VOP_BMAP and VOP_STRATEGY to
593 	 * directly access the backing vnode.  If we can, use these two
594 	 * operations to avoid messing with the local buffer cache.
595 	 * Otherwise fall back to regular VOP_READ/VOP_WRITE operations
596 	 * which are guaranteed to work with any file system. */
597 	usestrategy = vnode_has_strategy(vnd);
598 
599 #ifdef DEBUG
600 	if (vnddebug & VDB_INIT)
601 		printf("vndthread: vp %p, %s\n", vnd->sc_vp,
602 		    usestrategy ?
603 		    "using bmap/strategy operations" :
604 		    "using read/write operations");
605 #endif
606 
607 	s = splbio();
608 	vnd->sc_flags |= VNF_KTHREAD;
609 	wakeup(&vnd->sc_kthread);
610 
611 	/*
612 	 * Dequeue requests and serve them depending on the available
613 	 * vnode operations.
614 	 */
615 	while ((vnd->sc_flags & VNF_VUNCONF) == 0) {
616 		struct vndxfer *vnx;
617 		int flags;
618 		struct buf *obp;
619 		struct buf *bp;
620 
621 		obp = bufq_get(vnd->sc_tab);
622 		if (obp == NULL) {
623 			tsleep(&vnd->sc_tab, PRIBIO, "vndbp", 0);
624 			continue;
625 		};
626 		splx(s);
627 		flags = obp->b_flags;
628 #ifdef DEBUG
629 		if (vnddebug & VDB_FOLLOW)
630 			printf("vndthread(%p)\n", obp);
631 #endif
632 
633 		if (vnd->sc_vp->v_mount == NULL) {
634 			obp->b_error = ENXIO;
635 			goto done;
636 		}
637 #ifdef VND_COMPRESSION
638 		/* handle a compressed read */
639 		if ((flags & B_READ) != 0 && (vnd->sc_flags & VNF_COMP)) {
640 			off_t bn;
641 
642 			/* Convert to a byte offset within the file. */
643 			bn = obp->b_rawblkno *
644 			    vnd->sc_dkdev.dk_label->d_secsize;
645 
646 			compstrategy(obp, bn);
647 			goto done;
648 		}
649 #endif /* VND_COMPRESSION */
650 
651 		/*
652 		 * Allocate a header for this transfer and link it to the
653 		 * buffer
654 		 */
655 		s = splbio();
656 		vnx = VND_GETXFER(vnd);
657 		splx(s);
658 		vnx->vx_vnd = vnd;
659 
660 		s = splbio();
661 		while (vnd->sc_active >= vnd->sc_maxactive) {
662 			tsleep(&vnd->sc_tab, PRIBIO, "vndac", 0);
663 		}
664 		vnd->sc_active++;
665 		splx(s);
666 
667 		/* Instrumentation. */
668 		disk_busy(&vnd->sc_dkdev);
669 
670 		bp = &vnx->vx_buf;
671 		buf_init(bp);
672 		bp->b_flags = (obp->b_flags & B_READ);
673 		bp->b_oflags = obp->b_oflags;
674 		bp->b_cflags = obp->b_cflags;
675 		bp->b_iodone = vndiodone;
676 		bp->b_private = obp;
677 		bp->b_vp = vnd->sc_vp;
678 		bp->b_objlock = &bp->b_vp->v_interlock;
679 		bp->b_data = obp->b_data;
680 		bp->b_bcount = obp->b_bcount;
681 		BIO_COPYPRIO(bp, obp);
682 
683 		/* Handle the request using the appropriate operations. */
684 		if (usestrategy)
685 			handle_with_strategy(vnd, obp, bp);
686 		else
687 			handle_with_rdwr(vnd, obp, bp);
688 
689 		s = splbio();
690 		continue;
691 
692 done:
693 		biodone(obp);
694 		s = splbio();
695 	}
696 
697 	vnd->sc_flags &= (~VNF_KTHREAD | VNF_VUNCONF);
698 	wakeup(&vnd->sc_kthread);
699 	splx(s);
700 	kthread_exit(0);
701 }
702 
703 /*
704  * Checks if the given vnode supports the requested operation.
705  * The operation is specified the offset returned by VOFFSET.
706  *
707  * XXX The test below used to determine this is quite fragile
708  * because it relies on the file system to use genfs to specify
709  * unimplemented operations.  There might be another way to do
710  * it more cleanly.
711  */
712 static bool
713 vnode_has_op(const struct vnode *vp, int opoffset)
714 {
715 	int (*defaultp)(void *);
716 	int (*opp)(void *);
717 
718 	defaultp = vp->v_op[VOFFSET(vop_default)];
719 	opp = vp->v_op[opoffset];
720 
721 	return opp != defaultp && opp != genfs_eopnotsupp &&
722 	    opp != genfs_badop && opp != genfs_nullop;
723 }
724 
725 /*
726  * Handes the read/write request given in 'bp' using the vnode's VOP_READ
727  * and VOP_WRITE operations.
728  *
729  * 'obp' is a pointer to the original request fed to the vnd device.
730  */
731 static void
732 handle_with_rdwr(struct vnd_softc *vnd, const struct buf *obp, struct buf *bp)
733 {
734 	bool doread;
735 	off_t offset;
736 	size_t resid;
737 	struct vnode *vp;
738 
739 	doread = bp->b_flags & B_READ;
740 	offset = obp->b_rawblkno * vnd->sc_dkdev.dk_label->d_secsize;
741 	vp = vnd->sc_vp;
742 
743 #if defined(DEBUG)
744 	if (vnddebug & VDB_IO)
745 		printf("vnd (rdwr): vp %p, %s, rawblkno 0x%" PRIx64
746 		    ", secsize %d, offset %" PRIu64
747 		    ", bcount %d\n",
748 		    vp, doread ? "read" : "write", obp->b_rawblkno,
749 		    vnd->sc_dkdev.dk_label->d_secsize, offset,
750 		    bp->b_bcount);
751 #endif
752 
753 	/* Issue the read or write operation. */
754 	bp->b_error =
755 	    vn_rdwr(doread ? UIO_READ : UIO_WRITE,
756 	    vp, bp->b_data, bp->b_bcount, offset,
757 	    UIO_SYSSPACE, 0, vnd->sc_cred, &resid, NULL);
758 	bp->b_resid = resid;
759 
760 	/* We need to increase the number of outputs on the vnode if
761 	 * there was any write to it. */
762 	if (!doread) {
763 		mutex_enter(&vp->v_interlock);
764 		vp->v_numoutput++;
765 		mutex_exit(&vp->v_interlock);
766 	}
767 
768 	biodone(bp);
769 }
770 
771 /*
772  * Handes the read/write request given in 'bp' using the vnode's VOP_BMAP
773  * and VOP_STRATEGY operations.
774  *
775  * 'obp' is a pointer to the original request fed to the vnd device.
776  */
777 static void
778 handle_with_strategy(struct vnd_softc *vnd, const struct buf *obp,
779     struct buf *bp)
780 {
781 	int bsize, error, flags, skipped;
782 	size_t resid, sz;
783 	off_t bn, offset;
784 	struct vnode *vp;
785 
786 	flags = obp->b_flags;
787 
788 	if (!(flags & B_READ)) {
789 		vp = bp->b_vp;
790 		mutex_enter(&vp->v_interlock);
791 		vp->v_numoutput++;
792 		mutex_exit(&vp->v_interlock);
793 	}
794 
795 	/* convert to a byte offset within the file. */
796 	bn = obp->b_rawblkno * vnd->sc_dkdev.dk_label->d_secsize;
797 
798 	bsize = vnd->sc_vp->v_mount->mnt_stat.f_iosize;
799 	skipped = 0;
800 
801 	/*
802 	 * Break the request into bsize pieces and feed them
803 	 * sequentially using VOP_BMAP/VOP_STRATEGY.
804 	 * We do it this way to keep from flooding NFS servers if we
805 	 * are connected to an NFS file.  This places the burden on
806 	 * the client rather than the server.
807 	 */
808 	error = 0;
809 	bp->b_resid = bp->b_bcount;
810 	for (offset = 0, resid = bp->b_resid; resid;
811 	    resid -= sz, offset += sz) {
812 		struct buf *nbp;
813 		daddr_t nbn;
814 		int off, nra;
815 
816 		nra = 0;
817 		vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
818 		error = VOP_BMAP(vnd->sc_vp, bn / bsize, &vp, &nbn, &nra);
819 		VOP_UNLOCK(vnd->sc_vp, 0);
820 
821 		if (error == 0 && (long)nbn == -1)
822 			error = EIO;
823 
824 		/*
825 		 * If there was an error or a hole in the file...punt.
826 		 * Note that we may have to wait for any operations
827 		 * that we have already fired off before releasing
828 		 * the buffer.
829 		 *
830 		 * XXX we could deal with holes here but it would be
831 		 * a hassle (in the write case).
832 		 */
833 		if (error) {
834 			skipped += resid;
835 			break;
836 		}
837 
838 #ifdef DEBUG
839 		if (!dovndcluster)
840 			nra = 0;
841 #endif
842 
843 		off = bn % bsize;
844 		sz = MIN(((off_t)1 + nra) * bsize - off, resid);
845 #ifdef	DEBUG
846 		if (vnddebug & VDB_IO)
847 			printf("vndstrategy: vp %p/%p bn 0x%qx/0x%" PRIx64
848 			    " sz 0x%zx\n", vnd->sc_vp, vp, (long long)bn,
849 			    nbn, sz);
850 #endif
851 
852 		nbp = getiobuf(vp, true);
853 		nestiobuf_setup(bp, nbp, offset, sz);
854 		nbp->b_blkno = nbn + btodb(off);
855 
856 #if 0 /* XXX #ifdef DEBUG */
857 		if (vnddebug & VDB_IO)
858 			printf("vndstart(%ld): bp %p vp %p blkno "
859 			    "0x%" PRIx64 " flags %x addr %p cnt 0x%x\n",
860 			    (long) (vnd-vnd_softc), &nbp->vb_buf,
861 			    nbp->vb_buf.b_vp, nbp->vb_buf.b_blkno,
862 			    nbp->vb_buf.b_flags, nbp->vb_buf.b_data,
863 			    nbp->vb_buf.b_bcount);
864 #endif
865 		VOP_STRATEGY(vp, nbp);
866 		bn += sz;
867 	}
868 	nestiobuf_done(bp, skipped, error);
869 }
870 
871 static void
872 vndiodone(struct buf *bp)
873 {
874 	struct vndxfer *vnx = VND_BUFTOXFER(bp);
875 	struct vnd_softc *vnd = vnx->vx_vnd;
876 	struct buf *obp = bp->b_private;
877 
878 	KASSERT(&vnx->vx_buf == bp);
879 	KASSERT(vnd->sc_active > 0);
880 #ifdef DEBUG
881 	if (vnddebug & VDB_IO) {
882 		printf("vndiodone1: bp %p iodone: error %d\n",
883 		    bp, bp->b_error);
884 	}
885 #endif
886 	disk_unbusy(&vnd->sc_dkdev, bp->b_bcount - bp->b_resid,
887 	    (bp->b_flags & B_READ));
888 	vnd->sc_active--;
889 	if (vnd->sc_active == 0) {
890 		wakeup(&vnd->sc_tab);
891 	}
892 	obp->b_error = bp->b_error;
893 	obp->b_resid = bp->b_resid;
894 	buf_destroy(bp);
895 	VND_PUTXFER(vnd, vnx);
896 	biodone(obp);
897 }
898 
899 /* ARGSUSED */
900 static int
901 vndread(dev_t dev, struct uio *uio, int flags)
902 {
903 	int unit = vndunit(dev);
904 	struct vnd_softc *sc;
905 
906 #ifdef DEBUG
907 	if (vnddebug & VDB_FOLLOW)
908 		printf("vndread(0x%"PRIx64", %p)\n", dev, uio);
909 #endif
910 
911 	sc = device_lookup_private(&vnd_cd, unit);
912 	if (sc == NULL)
913 		return ENXIO;
914 
915 	if ((sc->sc_flags & VNF_INITED) == 0)
916 		return ENXIO;
917 
918 	return physio(vndstrategy, NULL, dev, B_READ, minphys, uio);
919 }
920 
921 /* ARGSUSED */
922 static int
923 vndwrite(dev_t dev, struct uio *uio, int flags)
924 {
925 	int unit = vndunit(dev);
926 	struct vnd_softc *sc;
927 
928 #ifdef DEBUG
929 	if (vnddebug & VDB_FOLLOW)
930 		printf("vndwrite(0x%"PRIx64", %p)\n", dev, uio);
931 #endif
932 
933 	sc = device_lookup_private(&vnd_cd, unit);
934 	if (sc == NULL)
935 		return ENXIO;
936 
937 	if ((sc->sc_flags & VNF_INITED) == 0)
938 		return ENXIO;
939 
940 	return physio(vndstrategy, NULL, dev, B_WRITE, minphys, uio);
941 }
942 
943 static int
944 vnd_cget(struct lwp *l, int unit, int *un, struct vattr *va)
945 {
946 	struct vnd_softc *vnd;
947 
948 	if (*un == -1)
949 		*un = unit;
950 	if (*un < 0)
951 		return EINVAL;
952 
953 	vnd = device_lookup_private(&vnd_cd, *un);
954 	if (vnd == NULL)
955 		return (*un >= vnd_cd.cd_ndevs) ? ENXIO : -1;
956 
957 	if ((vnd->sc_flags & VNF_INITED) == 0)
958 		return -1;
959 
960 	return VOP_GETATTR(vnd->sc_vp, va, l->l_cred);
961 }
962 
963 static int
964 vnddoclear(struct vnd_softc *vnd, int pmask, int minor, bool force)
965 {
966 	int error;
967 
968 	if ((error = vndlock(vnd)) != 0)
969 		return error;
970 
971 	/*
972 	 * Don't unconfigure if any other partitions are open
973 	 * or if both the character and block flavors of this
974 	 * partition are open.
975 	 */
976 	if (DK_BUSY(vnd, pmask) && !force) {
977 		vndunlock(vnd);
978 		return EBUSY;
979 	}
980 
981 	/*
982 	 * XXX vndclear() might call vndclose() implicitly;
983 	 * release lock to avoid recursion
984 	 *
985 	 * Set VNF_CLEARING to prevent vndopen() from
986 	 * sneaking in after we vndunlock().
987 	 */
988 	vnd->sc_flags |= VNF_CLEARING;
989 	vndunlock(vnd);
990 	vndclear(vnd, minor);
991 #ifdef DEBUG
992 	if (vnddebug & VDB_INIT)
993 		printf("vndioctl: CLRed\n");
994 #endif
995 
996 	/* Destroy the xfer and buffer pools. */
997 	pool_destroy(&vnd->sc_vxpool);
998 
999 	/* Detach the disk. */
1000 	disk_detach(&vnd->sc_dkdev);
1001 
1002 	return 0;
1003 }
1004 
1005 /* ARGSUSED */
1006 static int
1007 vndioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l)
1008 {
1009 	bool force;
1010 	int unit = vndunit(dev);
1011 	struct vnd_softc *vnd;
1012 	struct vnd_ioctl *vio;
1013 	struct vattr vattr;
1014 	struct nameidata nd;
1015 	int error, part, pmask;
1016 	size_t geomsize;
1017 	int fflags;
1018 #ifdef __HAVE_OLD_DISKLABEL
1019 	struct disklabel newlabel;
1020 #endif
1021 
1022 #ifdef DEBUG
1023 	if (vnddebug & VDB_FOLLOW)
1024 		printf("vndioctl(0x%"PRIx64", 0x%lx, %p, 0x%x, %p): unit %d\n",
1025 		    dev, cmd, data, flag, l->l_proc, unit);
1026 #endif
1027 	vnd = device_lookup_private(&vnd_cd, unit);
1028 	if (vnd == NULL &&
1029 #ifdef COMPAT_30
1030 	    cmd != VNDIOOCGET &&
1031 #endif
1032 	    cmd != VNDIOCGET)
1033 		return ENXIO;
1034 	vio = (struct vnd_ioctl *)data;
1035 
1036 	/* Must be open for writes for these commands... */
1037 	switch (cmd) {
1038 	case VNDIOCSET:
1039 	case VNDIOCCLR:
1040 #ifdef VNDIOOCSET
1041 	case VNDIOOCSET:
1042 	case VNDIOOCCLR:
1043 #endif
1044 	case DIOCSDINFO:
1045 	case DIOCWDINFO:
1046 #ifdef __HAVE_OLD_DISKLABEL
1047 	case ODIOCSDINFO:
1048 	case ODIOCWDINFO:
1049 #endif
1050 	case DIOCKLABEL:
1051 	case DIOCWLABEL:
1052 		if ((flag & FWRITE) == 0)
1053 			return EBADF;
1054 	}
1055 
1056 	/* Must be initialized for these... */
1057 	switch (cmd) {
1058 	case VNDIOCCLR:
1059 #ifdef VNDIOOCCLR
1060 	case VNDIOOCCLR:
1061 #endif
1062 	case DIOCGDINFO:
1063 	case DIOCSDINFO:
1064 	case DIOCWDINFO:
1065 	case DIOCGPART:
1066 	case DIOCKLABEL:
1067 	case DIOCWLABEL:
1068 	case DIOCGDEFLABEL:
1069 	case DIOCCACHESYNC:
1070 #ifdef __HAVE_OLD_DISKLABEL
1071 	case ODIOCGDINFO:
1072 	case ODIOCSDINFO:
1073 	case ODIOCWDINFO:
1074 	case ODIOCGDEFLABEL:
1075 #endif
1076 		if ((vnd->sc_flags & VNF_INITED) == 0)
1077 			return ENXIO;
1078 	}
1079 
1080 	switch (cmd) {
1081 #ifdef VNDIOOCSET
1082 	case VNDIOOCSET:
1083 #endif
1084 	case VNDIOCSET:
1085 		if (vnd->sc_flags & VNF_INITED)
1086 			return EBUSY;
1087 
1088 		if ((error = vndlock(vnd)) != 0)
1089 			return error;
1090 
1091 		fflags = FREAD;
1092 		if ((vio->vnd_flags & VNDIOF_READONLY) == 0)
1093 			fflags |= FWRITE;
1094 		NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, vio->vnd_file);
1095 		if ((error = vn_open(&nd, fflags, 0)) != 0)
1096 			goto unlock_and_exit;
1097 		KASSERT(l);
1098 		error = VOP_GETATTR(nd.ni_vp, &vattr, l->l_cred);
1099 		if (!error && nd.ni_vp->v_type != VREG)
1100 			error = EOPNOTSUPP;
1101 		if (!error && vattr.va_bytes < vattr.va_size)
1102 			/* File is definitely sparse, reject here */
1103 			error = EINVAL;
1104 		if (error) {
1105 			VOP_UNLOCK(nd.ni_vp, 0);
1106 			goto close_and_exit;
1107 		}
1108 
1109 		/* If using a compressed file, initialize its info */
1110 		/* (or abort with an error if kernel has no compression) */
1111 		if (vio->vnd_flags & VNF_COMP) {
1112 #ifdef VND_COMPRESSION
1113 			struct vnd_comp_header *ch;
1114 			int i;
1115 			u_int32_t comp_size;
1116 			u_int32_t comp_maxsize;
1117 
1118 			/* allocate space for compresed file header */
1119 			ch = malloc(sizeof(struct vnd_comp_header),
1120 			M_TEMP, M_WAITOK);
1121 
1122 			/* read compressed file header */
1123 			error = vn_rdwr(UIO_READ, nd.ni_vp, (void *)ch,
1124 			  sizeof(struct vnd_comp_header), 0, UIO_SYSSPACE,
1125 			  IO_UNIT|IO_NODELOCKED, l->l_cred, NULL, NULL);
1126 			if (error) {
1127 				free(ch, M_TEMP);
1128 				VOP_UNLOCK(nd.ni_vp, 0);
1129 				goto close_and_exit;
1130 			}
1131 
1132 			/* save some header info */
1133 			vnd->sc_comp_blksz = ntohl(ch->block_size);
1134 			/* note last offset is the file byte size */
1135 			vnd->sc_comp_numoffs = ntohl(ch->num_blocks)+1;
1136 			free(ch, M_TEMP);
1137 			if (vnd->sc_comp_blksz == 0 ||
1138 			    vnd->sc_comp_blksz % DEV_BSIZE !=0) {
1139 				VOP_UNLOCK(nd.ni_vp, 0);
1140 				error = EINVAL;
1141 				goto close_and_exit;
1142 			}
1143 			if (sizeof(struct vnd_comp_header) +
1144 			  sizeof(u_int64_t) * vnd->sc_comp_numoffs >
1145 			  vattr.va_size) {
1146 				VOP_UNLOCK(nd.ni_vp, 0);
1147 				error = EINVAL;
1148 				goto close_and_exit;
1149 			}
1150 
1151 			/* set decompressed file size */
1152 			vattr.va_size =
1153 			    ((u_quad_t)vnd->sc_comp_numoffs - 1) *
1154 			     (u_quad_t)vnd->sc_comp_blksz;
1155 
1156 			/* allocate space for all the compressed offsets */
1157 			vnd->sc_comp_offsets =
1158 			malloc(sizeof(u_int64_t) * vnd->sc_comp_numoffs,
1159 			M_DEVBUF, M_WAITOK);
1160 
1161 			/* read in the offsets */
1162 			error = vn_rdwr(UIO_READ, nd.ni_vp,
1163 			  (void *)vnd->sc_comp_offsets,
1164 			  sizeof(u_int64_t) * vnd->sc_comp_numoffs,
1165 			  sizeof(struct vnd_comp_header), UIO_SYSSPACE,
1166 			  IO_UNIT|IO_NODELOCKED, l->l_cred, NULL, NULL);
1167 			if (error) {
1168 				VOP_UNLOCK(nd.ni_vp, 0);
1169 				goto close_and_exit;
1170 			}
1171 			/*
1172 			 * find largest block size (used for allocation limit).
1173 			 * Also convert offset to native byte order.
1174 			 */
1175 			comp_maxsize = 0;
1176 			for (i = 0; i < vnd->sc_comp_numoffs - 1; i++) {
1177 				vnd->sc_comp_offsets[i] =
1178 				  be64toh(vnd->sc_comp_offsets[i]);
1179 				comp_size = be64toh(vnd->sc_comp_offsets[i + 1])
1180 				  - vnd->sc_comp_offsets[i];
1181 				if (comp_size > comp_maxsize)
1182 					comp_maxsize = comp_size;
1183 			}
1184 			vnd->sc_comp_offsets[vnd->sc_comp_numoffs - 1] =
1185 			  be64toh(vnd->sc_comp_offsets[vnd->sc_comp_numoffs - 1]);
1186 
1187 			/* create compressed data buffer */
1188 			vnd->sc_comp_buff = malloc(comp_maxsize,
1189 			  M_DEVBUF, M_WAITOK);
1190 
1191 			/* create decompressed buffer */
1192 			vnd->sc_comp_decombuf = malloc(vnd->sc_comp_blksz,
1193 			  M_DEVBUF, M_WAITOK);
1194 			vnd->sc_comp_buffblk = -1;
1195 
1196 			/* Initialize decompress stream */
1197 			memset(&vnd->sc_comp_stream, 0, sizeof(z_stream));
1198 			vnd->sc_comp_stream.zalloc = vnd_alloc;
1199 			vnd->sc_comp_stream.zfree = vnd_free;
1200 			error = inflateInit2(&vnd->sc_comp_stream, MAX_WBITS);
1201 			if (error) {
1202 				if (vnd->sc_comp_stream.msg)
1203 					printf("vnd%d: compressed file, %s\n",
1204 					  unit, vnd->sc_comp_stream.msg);
1205 				VOP_UNLOCK(nd.ni_vp, 0);
1206 				error = EINVAL;
1207 				goto close_and_exit;
1208 			}
1209 
1210 			vnd->sc_flags |= VNF_COMP | VNF_READONLY;
1211 #else /* !VND_COMPRESSION */
1212 			VOP_UNLOCK(nd.ni_vp, 0);
1213 			error = EOPNOTSUPP;
1214 			goto close_and_exit;
1215 #endif /* VND_COMPRESSION */
1216 		}
1217 
1218 		VOP_UNLOCK(nd.ni_vp, 0);
1219 		vnd->sc_vp = nd.ni_vp;
1220 		vnd->sc_size = btodb(vattr.va_size);	/* note truncation */
1221 
1222 		/*
1223 		 * Use pseudo-geometry specified.  If none was provided,
1224 		 * use "standard" Adaptec fictitious geometry.
1225 		 */
1226 		if (vio->vnd_flags & VNDIOF_HASGEOM) {
1227 
1228 			memcpy(&vnd->sc_geom, &vio->vnd_geom,
1229 			    sizeof(vio->vnd_geom));
1230 
1231 			/*
1232 			 * Sanity-check the sector size.
1233 			 * XXX Don't allow secsize < DEV_BSIZE.	 Should
1234 			 * XXX we?
1235 			 */
1236 			if (vnd->sc_geom.vng_secsize < DEV_BSIZE ||
1237 			    (vnd->sc_geom.vng_secsize % DEV_BSIZE) != 0 ||
1238 			    vnd->sc_geom.vng_ncylinders == 0 ||
1239 			    (vnd->sc_geom.vng_ntracks *
1240 			     vnd->sc_geom.vng_nsectors) == 0) {
1241 				error = EINVAL;
1242 				goto close_and_exit;
1243 			}
1244 
1245 			/*
1246 			 * Compute the size (in DEV_BSIZE blocks) specified
1247 			 * by the geometry.
1248 			 */
1249 			geomsize = (vnd->sc_geom.vng_nsectors *
1250 			    vnd->sc_geom.vng_ntracks *
1251 			    vnd->sc_geom.vng_ncylinders) *
1252 			    (vnd->sc_geom.vng_secsize / DEV_BSIZE);
1253 
1254 			/*
1255 			 * Sanity-check the size against the specified
1256 			 * geometry.
1257 			 */
1258 			if (vnd->sc_size < geomsize) {
1259 				error = EINVAL;
1260 				goto close_and_exit;
1261 			}
1262 		} else if (vnd->sc_size >= (32 * 64)) {
1263 			/*
1264 			 * Size must be at least 2048 DEV_BSIZE blocks
1265 			 * (1M) in order to use this geometry.
1266 			 */
1267 			vnd->sc_geom.vng_secsize = DEV_BSIZE;
1268 			vnd->sc_geom.vng_nsectors = 32;
1269 			vnd->sc_geom.vng_ntracks = 64;
1270 			vnd->sc_geom.vng_ncylinders = vnd->sc_size / (64 * 32);
1271 		} else {
1272 			vnd->sc_geom.vng_secsize = DEV_BSIZE;
1273 			vnd->sc_geom.vng_nsectors = 1;
1274 			vnd->sc_geom.vng_ntracks = 1;
1275 			vnd->sc_geom.vng_ncylinders = vnd->sc_size;
1276 		}
1277 
1278 		vnd_set_properties(vnd);
1279 
1280 		if (vio->vnd_flags & VNDIOF_READONLY) {
1281 			vnd->sc_flags |= VNF_READONLY;
1282 		}
1283 
1284 		if ((error = vndsetcred(vnd, l->l_cred)) != 0)
1285 			goto close_and_exit;
1286 
1287 		vndthrottle(vnd, vnd->sc_vp);
1288 		vio->vnd_osize = dbtob(vnd->sc_size);
1289 #ifdef VNDIOOCSET
1290 		if (cmd != VNDIOOCSET)
1291 #endif
1292 			vio->vnd_size = dbtob(vnd->sc_size);
1293 		vnd->sc_flags |= VNF_INITED;
1294 
1295 		/* create the kernel thread, wait for it to be up */
1296 		error = kthread_create(PRI_NONE, 0, NULL, vndthread, vnd,
1297 		    &vnd->sc_kthread, device_xname(vnd->sc_dev));
1298 		if (error)
1299 			goto close_and_exit;
1300 		while ((vnd->sc_flags & VNF_KTHREAD) == 0) {
1301 			tsleep(&vnd->sc_kthread, PRIBIO, "vndthr", 0);
1302 		}
1303 #ifdef DEBUG
1304 		if (vnddebug & VDB_INIT)
1305 			printf("vndioctl: SET vp %p size 0x%lx %d/%d/%d/%d\n",
1306 			    vnd->sc_vp, (unsigned long) vnd->sc_size,
1307 			    vnd->sc_geom.vng_secsize,
1308 			    vnd->sc_geom.vng_nsectors,
1309 			    vnd->sc_geom.vng_ntracks,
1310 			    vnd->sc_geom.vng_ncylinders);
1311 #endif
1312 
1313 		/* Attach the disk. */
1314 		disk_attach(&vnd->sc_dkdev);
1315 
1316 		/* Initialize the xfer and buffer pools. */
1317 		pool_init(&vnd->sc_vxpool, sizeof(struct vndxfer), 0,
1318 		    0, 0, "vndxpl", NULL, IPL_BIO);
1319 
1320 		/* Try and read the disklabel. */
1321 		vndgetdisklabel(dev, vnd);
1322 
1323 		vndunlock(vnd);
1324 
1325 		break;
1326 
1327 close_and_exit:
1328 		(void) vn_close(nd.ni_vp, fflags, l->l_cred);
1329 unlock_and_exit:
1330 #ifdef VND_COMPRESSION
1331 		/* free any allocated memory (for compressed file) */
1332 		if (vnd->sc_comp_offsets) {
1333 			free(vnd->sc_comp_offsets, M_DEVBUF);
1334 			vnd->sc_comp_offsets = NULL;
1335 		}
1336 		if (vnd->sc_comp_buff) {
1337 			free(vnd->sc_comp_buff, M_DEVBUF);
1338 			vnd->sc_comp_buff = NULL;
1339 		}
1340 		if (vnd->sc_comp_decombuf) {
1341 			free(vnd->sc_comp_decombuf, M_DEVBUF);
1342 			vnd->sc_comp_decombuf = NULL;
1343 		}
1344 #endif /* VND_COMPRESSION */
1345 		vndunlock(vnd);
1346 		return error;
1347 
1348 #ifdef VNDIOOCCLR
1349 	case VNDIOOCCLR:
1350 #endif
1351 	case VNDIOCCLR:
1352 		part = DISKPART(dev);
1353 		pmask = (1 << part);
1354 		force = (vio->vnd_flags & VNDIOF_FORCE) != 0;
1355 
1356 		if ((error = vnddoclear(vnd, pmask, minor(dev), force)) != 0)
1357 			return error;
1358 
1359 		break;
1360 
1361 #ifdef COMPAT_30
1362 	case VNDIOOCGET: {
1363 		struct vnd_ouser *vnu;
1364 		struct vattr va;
1365 		vnu = (struct vnd_ouser *)data;
1366 		KASSERT(l);
1367 		switch (error = vnd_cget(l, unit, &vnu->vnu_unit, &va)) {
1368 		case 0:
1369 			vnu->vnu_dev = va.va_fsid;
1370 			vnu->vnu_ino = va.va_fileid;
1371 			break;
1372 		case -1:
1373 			/* unused is not an error */
1374 			vnu->vnu_dev = 0;
1375 			vnu->vnu_ino = 0;
1376 			break;
1377 		default:
1378 			return error;
1379 		}
1380 		break;
1381 	}
1382 #endif
1383 	case VNDIOCGET: {
1384 		struct vnd_user *vnu;
1385 		struct vattr va;
1386 		vnu = (struct vnd_user *)data;
1387 		KASSERT(l);
1388 		switch (error = vnd_cget(l, unit, &vnu->vnu_unit, &va)) {
1389 		case 0:
1390 			vnu->vnu_dev = va.va_fsid;
1391 			vnu->vnu_ino = va.va_fileid;
1392 			break;
1393 		case -1:
1394 			/* unused is not an error */
1395 			vnu->vnu_dev = 0;
1396 			vnu->vnu_ino = 0;
1397 			break;
1398 		default:
1399 			return error;
1400 		}
1401 		break;
1402 	}
1403 
1404 	case DIOCGDINFO:
1405 		*(struct disklabel *)data = *(vnd->sc_dkdev.dk_label);
1406 		break;
1407 
1408 #ifdef __HAVE_OLD_DISKLABEL
1409 	case ODIOCGDINFO:
1410 		newlabel = *(vnd->sc_dkdev.dk_label);
1411 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1412 			return ENOTTY;
1413 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1414 		break;
1415 #endif
1416 
1417 	case DIOCGPART:
1418 		((struct partinfo *)data)->disklab = vnd->sc_dkdev.dk_label;
1419 		((struct partinfo *)data)->part =
1420 		    &vnd->sc_dkdev.dk_label->d_partitions[DISKPART(dev)];
1421 		break;
1422 
1423 	case DIOCWDINFO:
1424 	case DIOCSDINFO:
1425 #ifdef __HAVE_OLD_DISKLABEL
1426 	case ODIOCWDINFO:
1427 	case ODIOCSDINFO:
1428 #endif
1429 	{
1430 		struct disklabel *lp;
1431 
1432 		if ((error = vndlock(vnd)) != 0)
1433 			return error;
1434 
1435 		vnd->sc_flags |= VNF_LABELLING;
1436 
1437 #ifdef __HAVE_OLD_DISKLABEL
1438 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
1439 			memset(&newlabel, 0, sizeof newlabel);
1440 			memcpy(&newlabel, data, sizeof (struct olddisklabel));
1441 			lp = &newlabel;
1442 		} else
1443 #endif
1444 		lp = (struct disklabel *)data;
1445 
1446 		error = setdisklabel(vnd->sc_dkdev.dk_label,
1447 		    lp, 0, vnd->sc_dkdev.dk_cpulabel);
1448 		if (error == 0) {
1449 			if (cmd == DIOCWDINFO
1450 #ifdef __HAVE_OLD_DISKLABEL
1451 			    || cmd == ODIOCWDINFO
1452 #endif
1453 			   )
1454 				error = writedisklabel(VNDLABELDEV(dev),
1455 				    vndstrategy, vnd->sc_dkdev.dk_label,
1456 				    vnd->sc_dkdev.dk_cpulabel);
1457 		}
1458 
1459 		vnd->sc_flags &= ~VNF_LABELLING;
1460 
1461 		vndunlock(vnd);
1462 
1463 		if (error)
1464 			return error;
1465 		break;
1466 	}
1467 
1468 	case DIOCKLABEL:
1469 		if (*(int *)data != 0)
1470 			vnd->sc_flags |= VNF_KLABEL;
1471 		else
1472 			vnd->sc_flags &= ~VNF_KLABEL;
1473 		break;
1474 
1475 	case DIOCWLABEL:
1476 		if (*(int *)data != 0)
1477 			vnd->sc_flags |= VNF_WLABEL;
1478 		else
1479 			vnd->sc_flags &= ~VNF_WLABEL;
1480 		break;
1481 
1482 	case DIOCGDEFLABEL:
1483 		vndgetdefaultlabel(vnd, (struct disklabel *)data);
1484 		break;
1485 
1486 #ifdef __HAVE_OLD_DISKLABEL
1487 	case ODIOCGDEFLABEL:
1488 		vndgetdefaultlabel(vnd, &newlabel);
1489 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1490 			return ENOTTY;
1491 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1492 		break;
1493 #endif
1494 
1495 	case DIOCCACHESYNC:
1496 		vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1497 		error = VOP_FSYNC(vnd->sc_vp, vnd->sc_cred,
1498 		    FSYNC_WAIT | FSYNC_DATAONLY | FSYNC_CACHE, 0, 0);
1499 		VOP_UNLOCK(vnd->sc_vp, 0);
1500 		return error;
1501 
1502 	default:
1503 		return ENOTTY;
1504 	}
1505 
1506 	return 0;
1507 }
1508 
1509 /*
1510  * Duplicate the current processes' credentials.  Since we are called only
1511  * as the result of a SET ioctl and only root can do that, any future access
1512  * to this "disk" is essentially as root.  Note that credentials may change
1513  * if some other uid can write directly to the mapped file (NFS).
1514  */
1515 static int
1516 vndsetcred(struct vnd_softc *vnd, kauth_cred_t cred)
1517 {
1518 	struct uio auio;
1519 	struct iovec aiov;
1520 	char *tmpbuf;
1521 	int error;
1522 
1523 	vnd->sc_cred = kauth_cred_dup(cred);
1524 	tmpbuf = malloc(DEV_BSIZE, M_TEMP, M_WAITOK);
1525 
1526 	/* XXX: Horrible kludge to establish credentials for NFS */
1527 	aiov.iov_base = tmpbuf;
1528 	aiov.iov_len = min(DEV_BSIZE, dbtob(vnd->sc_size));
1529 	auio.uio_iov = &aiov;
1530 	auio.uio_iovcnt = 1;
1531 	auio.uio_offset = 0;
1532 	auio.uio_rw = UIO_READ;
1533 	auio.uio_resid = aiov.iov_len;
1534 	UIO_SETUP_SYSSPACE(&auio);
1535 	vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1536 	error = VOP_READ(vnd->sc_vp, &auio, 0, vnd->sc_cred);
1537 	if (error == 0) {
1538 		/*
1539 		 * Because vnd does all IO directly through the vnode
1540 		 * we need to flush (at least) the buffer from the above
1541 		 * VOP_READ from the buffer cache to prevent cache
1542 		 * incoherencies.  Also, be careful to write dirty
1543 		 * buffers back to stable storage.
1544 		 */
1545 		error = vinvalbuf(vnd->sc_vp, V_SAVE, vnd->sc_cred,
1546 			    curlwp, 0, 0);
1547 	}
1548 	VOP_UNLOCK(vnd->sc_vp, 0);
1549 
1550 	free(tmpbuf, M_TEMP);
1551 	return error;
1552 }
1553 
1554 /*
1555  * Set maxactive based on FS type
1556  */
1557 static void
1558 vndthrottle(struct vnd_softc *vnd, struct vnode *vp)
1559 {
1560 #ifdef NFS
1561 	extern int (**nfsv2_vnodeop_p)(void *);
1562 
1563 	if (vp->v_op == nfsv2_vnodeop_p)
1564 		vnd->sc_maxactive = 2;
1565 	else
1566 #endif
1567 		vnd->sc_maxactive = 8;
1568 
1569 	if (vnd->sc_maxactive < 1)
1570 		vnd->sc_maxactive = 1;
1571 }
1572 
1573 #if 0
1574 static void
1575 vndshutdown(void)
1576 {
1577 	struct vnd_softc *vnd;
1578 
1579 	for (vnd = &vnd_softc[0]; vnd < &vnd_softc[numvnd]; vnd++)
1580 		if (vnd->sc_flags & VNF_INITED)
1581 			vndclear(vnd);
1582 }
1583 #endif
1584 
1585 static void
1586 vndclear(struct vnd_softc *vnd, int myminor)
1587 {
1588 	struct vnode *vp = vnd->sc_vp;
1589 	int fflags = FREAD;
1590 	int bmaj, cmaj, i, mn;
1591 	int s;
1592 
1593 #ifdef DEBUG
1594 	if (vnddebug & VDB_FOLLOW)
1595 		printf("vndclear(%p): vp %p\n", vnd, vp);
1596 #endif
1597 	/* locate the major number */
1598 	bmaj = bdevsw_lookup_major(&vnd_bdevsw);
1599 	cmaj = cdevsw_lookup_major(&vnd_cdevsw);
1600 
1601 	/* Nuke the vnodes for any open instances */
1602 	for (i = 0; i < MAXPARTITIONS; i++) {
1603 		mn = DISKMINOR(device_unit(vnd->sc_dev), i);
1604 		vdevgone(bmaj, mn, mn, VBLK);
1605 		if (mn != myminor) /* XXX avoid to kill own vnode */
1606 			vdevgone(cmaj, mn, mn, VCHR);
1607 	}
1608 
1609 	if ((vnd->sc_flags & VNF_READONLY) == 0)
1610 		fflags |= FWRITE;
1611 
1612 	s = splbio();
1613 	bufq_drain(vnd->sc_tab);
1614 	splx(s);
1615 
1616 	vnd->sc_flags |= VNF_VUNCONF;
1617 	wakeup(&vnd->sc_tab);
1618 	while (vnd->sc_flags & VNF_KTHREAD)
1619 		tsleep(&vnd->sc_kthread, PRIBIO, "vnthr", 0);
1620 
1621 #ifdef VND_COMPRESSION
1622 	/* free the compressed file buffers */
1623 	if (vnd->sc_flags & VNF_COMP) {
1624 		if (vnd->sc_comp_offsets) {
1625 			free(vnd->sc_comp_offsets, M_DEVBUF);
1626 			vnd->sc_comp_offsets = NULL;
1627 		}
1628 		if (vnd->sc_comp_buff) {
1629 			free(vnd->sc_comp_buff, M_DEVBUF);
1630 			vnd->sc_comp_buff = NULL;
1631 		}
1632 		if (vnd->sc_comp_decombuf) {
1633 			free(vnd->sc_comp_decombuf, M_DEVBUF);
1634 			vnd->sc_comp_decombuf = NULL;
1635 		}
1636 	}
1637 #endif /* VND_COMPRESSION */
1638 	vnd->sc_flags &=
1639 	    ~(VNF_INITED | VNF_READONLY | VNF_VLABEL
1640 	      | VNF_VUNCONF | VNF_COMP | VNF_CLEARING);
1641 	if (vp == NULL)
1642 		panic("vndclear: null vp");
1643 	(void) vn_close(vp, fflags, vnd->sc_cred);
1644 	kauth_cred_free(vnd->sc_cred);
1645 	vnd->sc_vp = NULL;
1646 	vnd->sc_cred = NULL;
1647 	vnd->sc_size = 0;
1648 }
1649 
1650 static int
1651 vndsize(dev_t dev)
1652 {
1653 	struct vnd_softc *sc;
1654 	struct disklabel *lp;
1655 	int part, unit, omask;
1656 	int size;
1657 
1658 	unit = vndunit(dev);
1659 	sc = device_lookup_private(&vnd_cd, unit);
1660 	if (sc == NULL)
1661 		return -1;
1662 
1663 	if ((sc->sc_flags & VNF_INITED) == 0)
1664 		return -1;
1665 
1666 	part = DISKPART(dev);
1667 	omask = sc->sc_dkdev.dk_openmask & (1 << part);
1668 	lp = sc->sc_dkdev.dk_label;
1669 
1670 	if (omask == 0 && vndopen(dev, 0, S_IFBLK, curlwp))	/* XXX */
1671 		return -1;
1672 
1673 	if (lp->d_partitions[part].p_fstype != FS_SWAP)
1674 		size = -1;
1675 	else
1676 		size = lp->d_partitions[part].p_size *
1677 		    (lp->d_secsize / DEV_BSIZE);
1678 
1679 	if (omask == 0 && vndclose(dev, 0, S_IFBLK, curlwp))	/* XXX */
1680 		return -1;
1681 
1682 	return size;
1683 }
1684 
1685 static int
1686 vnddump(dev_t dev, daddr_t blkno, void *va,
1687     size_t size)
1688 {
1689 
1690 	/* Not implemented. */
1691 	return ENXIO;
1692 }
1693 
1694 static void
1695 vndgetdefaultlabel(struct vnd_softc *sc, struct disklabel *lp)
1696 {
1697 	struct vndgeom *vng = &sc->sc_geom;
1698 	struct partition *pp;
1699 
1700 	memset(lp, 0, sizeof(*lp));
1701 
1702 	lp->d_secperunit = sc->sc_size / (vng->vng_secsize / DEV_BSIZE);
1703 	lp->d_secsize = vng->vng_secsize;
1704 	lp->d_nsectors = vng->vng_nsectors;
1705 	lp->d_ntracks = vng->vng_ntracks;
1706 	lp->d_ncylinders = vng->vng_ncylinders;
1707 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1708 
1709 	strncpy(lp->d_typename, "vnd", sizeof(lp->d_typename));
1710 	lp->d_type = DTYPE_VND;
1711 	strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
1712 	lp->d_rpm = 3600;
1713 	lp->d_interleave = 1;
1714 	lp->d_flags = 0;
1715 
1716 	pp = &lp->d_partitions[RAW_PART];
1717 	pp->p_offset = 0;
1718 	pp->p_size = lp->d_secperunit;
1719 	pp->p_fstype = FS_UNUSED;
1720 	lp->d_npartitions = RAW_PART + 1;
1721 
1722 	lp->d_magic = DISKMAGIC;
1723 	lp->d_magic2 = DISKMAGIC;
1724 	lp->d_checksum = dkcksum(lp);
1725 }
1726 
1727 /*
1728  * Read the disklabel from a vnd.  If one is not present, create a fake one.
1729  */
1730 static void
1731 vndgetdisklabel(dev_t dev, struct vnd_softc *sc)
1732 {
1733 	const char *errstring;
1734 	struct disklabel *lp = sc->sc_dkdev.dk_label;
1735 	struct cpu_disklabel *clp = sc->sc_dkdev.dk_cpulabel;
1736 	int i;
1737 
1738 	memset(clp, 0, sizeof(*clp));
1739 
1740 	vndgetdefaultlabel(sc, lp);
1741 
1742 	/*
1743 	 * Call the generic disklabel extraction routine.
1744 	 */
1745 	errstring = readdisklabel(VNDLABELDEV(dev), vndstrategy, lp, clp);
1746 	if (errstring) {
1747 		/*
1748 		 * Lack of disklabel is common, but we print the warning
1749 		 * anyway, since it might contain other useful information.
1750 		 */
1751 		aprint_normal_dev(sc->sc_dev, "%s\n", errstring);
1752 
1753 		/*
1754 		 * For historical reasons, if there's no disklabel
1755 		 * present, all partitions must be FS_BSDFFS and
1756 		 * occupy the entire disk.
1757 		 */
1758 		for (i = 0; i < MAXPARTITIONS; i++) {
1759 			/*
1760 			 * Don't wipe out port specific hack (such as
1761 			 * dos partition hack of i386 port).
1762 			 */
1763 			if (lp->d_partitions[i].p_size != 0)
1764 				continue;
1765 
1766 			lp->d_partitions[i].p_size = lp->d_secperunit;
1767 			lp->d_partitions[i].p_offset = 0;
1768 			lp->d_partitions[i].p_fstype = FS_BSDFFS;
1769 		}
1770 
1771 		strncpy(lp->d_packname, "default label",
1772 		    sizeof(lp->d_packname));
1773 
1774 		lp->d_npartitions = MAXPARTITIONS;
1775 		lp->d_checksum = dkcksum(lp);
1776 	}
1777 
1778 	/* In-core label now valid. */
1779 	sc->sc_flags |= VNF_VLABEL;
1780 }
1781 
1782 /*
1783  * Wait interruptibly for an exclusive lock.
1784  *
1785  * XXX
1786  * Several drivers do this; it should be abstracted and made MP-safe.
1787  */
1788 static int
1789 vndlock(struct vnd_softc *sc)
1790 {
1791 	int error;
1792 
1793 	while ((sc->sc_flags & VNF_LOCKED) != 0) {
1794 		sc->sc_flags |= VNF_WANTED;
1795 		if ((error = tsleep(sc, PRIBIO | PCATCH, "vndlck", 0)) != 0)
1796 			return error;
1797 	}
1798 	sc->sc_flags |= VNF_LOCKED;
1799 	return 0;
1800 }
1801 
1802 /*
1803  * Unlock and wake up any waiters.
1804  */
1805 static void
1806 vndunlock(struct vnd_softc *sc)
1807 {
1808 
1809 	sc->sc_flags &= ~VNF_LOCKED;
1810 	if ((sc->sc_flags & VNF_WANTED) != 0) {
1811 		sc->sc_flags &= ~VNF_WANTED;
1812 		wakeup(sc);
1813 	}
1814 }
1815 
1816 #ifdef VND_COMPRESSION
1817 /* compressed file read */
1818 static void
1819 compstrategy(struct buf *bp, off_t bn)
1820 {
1821 	int error;
1822 	int unit = vndunit(bp->b_dev);
1823 	struct vnd_softc *vnd =
1824 	    device_lookup_private(&vnd_cd, unit);
1825 	u_int32_t comp_block;
1826 	struct uio auio;
1827 	char *addr;
1828 	int s;
1829 
1830 	/* set up constants for data move */
1831 	auio.uio_rw = UIO_READ;
1832 	UIO_SETUP_SYSSPACE(&auio);
1833 
1834 	/* read, and transfer the data */
1835 	addr = bp->b_data;
1836 	bp->b_resid = bp->b_bcount;
1837 	s = splbio();
1838 	while (bp->b_resid > 0) {
1839 		unsigned length;
1840 		size_t length_in_buffer;
1841 		u_int32_t offset_in_buffer;
1842 		struct iovec aiov;
1843 
1844 		/* calculate the compressed block number */
1845 		comp_block = bn / (off_t)vnd->sc_comp_blksz;
1846 
1847 		/* check for good block number */
1848 		if (comp_block >= vnd->sc_comp_numoffs) {
1849 			bp->b_error = EINVAL;
1850 			splx(s);
1851 			return;
1852 		}
1853 
1854 		/* read in the compressed block, if not in buffer */
1855 		if (comp_block != vnd->sc_comp_buffblk) {
1856 			length = vnd->sc_comp_offsets[comp_block + 1] -
1857 			    vnd->sc_comp_offsets[comp_block];
1858 			vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1859 			error = vn_rdwr(UIO_READ, vnd->sc_vp, vnd->sc_comp_buff,
1860 			    length, vnd->sc_comp_offsets[comp_block],
1861 			    UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, vnd->sc_cred,
1862 			    NULL, NULL);
1863 			if (error) {
1864 				bp->b_error = error;
1865 				VOP_UNLOCK(vnd->sc_vp, 0);
1866 				splx(s);
1867 				return;
1868 			}
1869 			/* uncompress the buffer */
1870 			vnd->sc_comp_stream.next_in = vnd->sc_comp_buff;
1871 			vnd->sc_comp_stream.avail_in = length;
1872 			vnd->sc_comp_stream.next_out = vnd->sc_comp_decombuf;
1873 			vnd->sc_comp_stream.avail_out = vnd->sc_comp_blksz;
1874 			inflateReset(&vnd->sc_comp_stream);
1875 			error = inflate(&vnd->sc_comp_stream, Z_FINISH);
1876 			if (error != Z_STREAM_END) {
1877 				if (vnd->sc_comp_stream.msg)
1878 					aprint_normal_dev(vnd->sc_dev,
1879 					    "compressed file, %s\n",
1880 					    vnd->sc_comp_stream.msg);
1881 				bp->b_error = EBADMSG;
1882 				VOP_UNLOCK(vnd->sc_vp, 0);
1883 				splx(s);
1884 				return;
1885 			}
1886 			vnd->sc_comp_buffblk = comp_block;
1887 			VOP_UNLOCK(vnd->sc_vp, 0);
1888 		}
1889 
1890 		/* transfer the usable uncompressed data */
1891 		offset_in_buffer = bn % (off_t)vnd->sc_comp_blksz;
1892 		length_in_buffer = vnd->sc_comp_blksz - offset_in_buffer;
1893 		if (length_in_buffer > bp->b_resid)
1894 			length_in_buffer = bp->b_resid;
1895 		auio.uio_iov = &aiov;
1896 		auio.uio_iovcnt = 1;
1897 		aiov.iov_base = addr;
1898 		aiov.iov_len = length_in_buffer;
1899 		auio.uio_resid = aiov.iov_len;
1900 		auio.uio_offset = 0;
1901 		error = uiomove(vnd->sc_comp_decombuf + offset_in_buffer,
1902 		    length_in_buffer, &auio);
1903 		if (error) {
1904 			bp->b_error = error;
1905 			splx(s);
1906 			return;
1907 		}
1908 
1909 		bn += length_in_buffer;
1910 		addr += length_in_buffer;
1911 		bp->b_resid -= length_in_buffer;
1912 	}
1913 	splx(s);
1914 }
1915 
1916 /* compression memory allocation routines */
1917 static void *
1918 vnd_alloc(void *aux, u_int items, u_int siz)
1919 {
1920 	return malloc(items * siz, M_TEMP, M_NOWAIT);
1921 }
1922 
1923 static void
1924 vnd_free(void *aux, void *ptr)
1925 {
1926 	free(ptr, M_TEMP);
1927 }
1928 #endif /* VND_COMPRESSION */
1929 
1930 static void
1931 vnd_set_properties(struct vnd_softc *vnd)
1932 {
1933 	prop_dictionary_t disk_info, odisk_info, geom;
1934 
1935 	disk_info = prop_dictionary_create();
1936 
1937 	geom = prop_dictionary_create();
1938 
1939 	prop_dictionary_set_uint64(geom, "sectors-per-unit",
1940 	    vnd->sc_geom.vng_nsectors * vnd->sc_geom.vng_ntracks *
1941 	    vnd->sc_geom.vng_ncylinders);
1942 
1943 	prop_dictionary_set_uint32(geom, "sector-size",
1944 	    vnd->sc_geom.vng_secsize);
1945 
1946 	prop_dictionary_set_uint16(geom, "sectors-per-track",
1947 	    vnd->sc_geom.vng_nsectors);
1948 
1949 	prop_dictionary_set_uint16(geom, "tracks-per-cylinder",
1950 	    vnd->sc_geom.vng_ntracks);
1951 
1952 	prop_dictionary_set_uint64(geom, "cylinders-per-unit",
1953 	    vnd->sc_geom.vng_ncylinders);
1954 
1955 	prop_dictionary_set(disk_info, "geometry", geom);
1956 	prop_object_release(geom);
1957 
1958 	prop_dictionary_set(device_properties(vnd->sc_dev),
1959 	    "disk-info", disk_info);
1960 
1961 	/*
1962 	 * Don't release disk_info here; we keep a reference to it.
1963 	 * disk_detach() will release it when we go away.
1964 	 */
1965 
1966 	odisk_info = vnd->sc_dkdev.dk_info;
1967 	vnd->sc_dkdev.dk_info = disk_info;
1968 	if (odisk_info)
1969 		prop_object_release(odisk_info);
1970 }
1971 
1972 #ifdef _MODULE
1973 
1974 #include <sys/module.h>
1975 
1976 MODULE(MODULE_CLASS_DRIVER, vnd, NULL);
1977 CFDRIVER_DECL(vnd, DV_DISK, NULL);
1978 
1979 static int
1980 vnd_modcmd(modcmd_t cmd, void *arg)
1981 {
1982 	int bmajor = -1, cmajor = -1,  error = 0;
1983 
1984 	switch (cmd) {
1985 	case MODULE_CMD_INIT:
1986 		error = config_cfdriver_attach(&vnd_cd);
1987 		if (error)
1988 			break;
1989 
1990 		error = config_cfattach_attach(vnd_cd.cd_name, &vnd_ca);
1991 	        if (error) {
1992 			config_cfdriver_detach(&vnd_cd);
1993 			aprint_error("%s: unable to register cfattach\n",
1994 			    vnd_cd.cd_name);
1995 			break;
1996 		}
1997 
1998 		error = devsw_attach("vnd", &vnd_bdevsw, &bmajor,
1999 		    &vnd_cdevsw, &cmajor);
2000 		if (error) {
2001 			config_cfattach_detach(vnd_cd.cd_name, &vnd_ca);
2002 			config_cfdriver_detach(&vnd_cd);
2003 			break;
2004 		}
2005 
2006 		break;
2007 
2008 	case MODULE_CMD_FINI:
2009 		error = config_cfattach_detach(vnd_cd.cd_name, &vnd_ca);
2010 		if (error)
2011 			break;
2012 		config_cfdriver_detach(&vnd_cd);
2013 		devsw_detach(&vnd_bdevsw, &vnd_cdevsw);
2014 		break;
2015 
2016 	case MODULE_CMD_STAT:
2017 		return ENOTTY;
2018 
2019 	default:
2020 		return ENOTTY;
2021 	}
2022 
2023 	return error;
2024 }
2025 
2026 #endif
2027