xref: /netbsd-src/sys/dev/vnd.c (revision da9817918ec7e88db2912a2882967c7570a83f47)
1 /*	$NetBSD: vnd.c,v 1.201 2009/05/07 09:13:13 cegger Exp $	*/
2 
3 /*-
4  * Copyright (c) 1996, 1997, 1998, 2008 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Jason R. Thorpe.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29  * POSSIBILITY OF SUCH DAMAGE.
30  */
31 
32 /*
33  * Copyright (c) 1990, 1993
34  *	The Regents of the University of California.  All rights reserved.
35  *
36  * This code is derived from software contributed to Berkeley by
37  * the Systems Programming Group of the University of Utah Computer
38  * Science Department.
39  *
40  * Redistribution and use in source and binary forms, with or without
41  * modification, are permitted provided that the following conditions
42  * are met:
43  * 1. Redistributions of source code must retain the above copyright
44  *    notice, this list of conditions and the following disclaimer.
45  * 2. Redistributions in binary form must reproduce the above copyright
46  *    notice, this list of conditions and the following disclaimer in the
47  *    documentation and/or other materials provided with the distribution.
48  * 3. Neither the name of the University nor the names of its contributors
49  *    may be used to endorse or promote products derived from this software
50  *    without specific prior written permission.
51  *
52  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
53  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
54  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
55  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
56  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
57  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
58  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
59  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
60  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
61  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
62  * SUCH DAMAGE.
63  *
64  * from: Utah $Hdr: vn.c 1.13 94/04/02$
65  *
66  *	@(#)vn.c	8.9 (Berkeley) 5/14/95
67  */
68 
69 /*
70  * Copyright (c) 1988 University of Utah.
71  *
72  * This code is derived from software contributed to Berkeley by
73  * the Systems Programming Group of the University of Utah Computer
74  * Science Department.
75  *
76  * Redistribution and use in source and binary forms, with or without
77  * modification, are permitted provided that the following conditions
78  * are met:
79  * 1. Redistributions of source code must retain the above copyright
80  *    notice, this list of conditions and the following disclaimer.
81  * 2. Redistributions in binary form must reproduce the above copyright
82  *    notice, this list of conditions and the following disclaimer in the
83  *    documentation and/or other materials provided with the distribution.
84  * 3. All advertising materials mentioning features or use of this software
85  *    must display the following acknowledgement:
86  *	This product includes software developed by the University of
87  *	California, Berkeley and its contributors.
88  * 4. Neither the name of the University nor the names of its contributors
89  *    may be used to endorse or promote products derived from this software
90  *    without specific prior written permission.
91  *
92  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
93  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
94  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
95  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
96  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
97  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
98  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
99  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
100  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
101  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
102  * SUCH DAMAGE.
103  *
104  * from: Utah $Hdr: vn.c 1.13 94/04/02$
105  *
106  *	@(#)vn.c	8.9 (Berkeley) 5/14/95
107  */
108 
109 /*
110  * Vnode disk driver.
111  *
112  * Block/character interface to a vnode.  Allows one to treat a file
113  * as a disk (e.g. build a filesystem in it, mount it, etc.).
114  *
115  * NOTE 1: If the vnode supports the VOP_BMAP and VOP_STRATEGY operations,
116  * this uses them to avoid distorting the local buffer cache.  If those
117  * block-level operations are not available, this falls back to the regular
118  * read and write calls.  Using these may distort the cache in some cases
119  * but better have the driver working than preventing it to work on file
120  * systems where the block-level operations are not implemented for
121  * whatever reason.
122  *
123  * NOTE 2: There is a security issue involved with this driver.
124  * Once mounted all access to the contents of the "mapped" file via
125  * the special file is controlled by the permissions on the special
126  * file, the protection of the mapped file is ignored (effectively,
127  * by using root credentials in all transactions).
128  *
129  * NOTE 3: Doesn't interact with leases, should it?
130  */
131 
132 #include <sys/cdefs.h>
133 __KERNEL_RCSID(0, "$NetBSD: vnd.c,v 1.201 2009/05/07 09:13:13 cegger Exp $");
134 
135 #if defined(_KERNEL_OPT)
136 #include "fs_nfs.h"
137 #include "opt_vnd.h"
138 #endif
139 
140 #include <sys/param.h>
141 #include <sys/systm.h>
142 #include <sys/namei.h>
143 #include <sys/proc.h>
144 #include <sys/kthread.h>
145 #include <sys/errno.h>
146 #include <sys/buf.h>
147 #include <sys/bufq.h>
148 #include <sys/malloc.h>
149 #include <sys/ioctl.h>
150 #include <sys/disklabel.h>
151 #include <sys/device.h>
152 #include <sys/disk.h>
153 #include <sys/stat.h>
154 #include <sys/mount.h>
155 #include <sys/vnode.h>
156 #include <sys/file.h>
157 #include <sys/uio.h>
158 #include <sys/conf.h>
159 #include <sys/kauth.h>
160 
161 #include <net/zlib.h>
162 
163 #include <miscfs/genfs/genfs.h>
164 #include <miscfs/specfs/specdev.h>
165 
166 #include <dev/vndvar.h>
167 
168 #include <prop/proplib.h>
169 
170 #if defined(VNDDEBUG) && !defined(DEBUG)
171 #define DEBUG
172 #endif
173 
174 #ifdef DEBUG
175 int dovndcluster = 1;
176 #define VDB_FOLLOW	0x01
177 #define VDB_INIT	0x02
178 #define VDB_IO		0x04
179 #define VDB_LABEL	0x08
180 int vnddebug = 0x00;
181 #endif
182 
183 #define vndunit(x)	DISKUNIT(x)
184 
185 struct vndxfer {
186 	struct buf vx_buf;
187 	struct vnd_softc *vx_vnd;
188 };
189 #define	VND_BUFTOXFER(bp)	((struct vndxfer *)(void *)bp)
190 
191 #define VND_GETXFER(vnd)	pool_get(&(vnd)->sc_vxpool, PR_WAITOK)
192 #define VND_PUTXFER(vnd, vx)	pool_put(&(vnd)->sc_vxpool, (vx))
193 
194 #define VNDLABELDEV(dev) \
195     (MAKEDISKDEV(major((dev)), vndunit((dev)), RAW_PART))
196 
197 /* called by main() at boot time */
198 void	vndattach(int);
199 
200 static void	vndclear(struct vnd_softc *, int);
201 static int	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 		return ENXIO;
361 
362 	lp = sc->sc_dkdev.dk_label;
363 
364 	part = DISKPART(dev);
365 	pmask = (1 << part);
366 
367 	/*
368 	 * If we're initialized, check to see if there are any other
369 	 * open partitions.  If not, then it's safe to update the
370 	 * in-core disklabel.  Only read the disklabel if it is
371 	 * not already valid.
372 	 */
373 	if ((sc->sc_flags & (VNF_INITED|VNF_VLABEL)) == VNF_INITED &&
374 	    sc->sc_dkdev.dk_openmask == 0)
375 		vndgetdisklabel(dev, sc);
376 
377 	/* Check that the partitions exists. */
378 	if (part != RAW_PART) {
379 		if (((sc->sc_flags & VNF_INITED) == 0) ||
380 		    ((part >= lp->d_npartitions) ||
381 		     (lp->d_partitions[part].p_fstype == FS_UNUSED))) {
382 			error = ENXIO;
383 			goto done;
384 		}
385 	}
386 
387 	/* Prevent our unit from being unconfigured while open. */
388 	switch (mode) {
389 	case S_IFCHR:
390 		sc->sc_dkdev.dk_copenmask |= pmask;
391 		break;
392 
393 	case S_IFBLK:
394 		sc->sc_dkdev.dk_bopenmask |= pmask;
395 		break;
396 	}
397 	sc->sc_dkdev.dk_openmask =
398 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
399 
400  done:
401 	vndunlock(sc);
402 	return error;
403 }
404 
405 static int
406 vndclose(dev_t dev, int flags, int mode, struct lwp *l)
407 {
408 	int unit = vndunit(dev);
409 	struct vnd_softc *sc;
410 	int error = 0, part;
411 
412 #ifdef DEBUG
413 	if (vnddebug & VDB_FOLLOW)
414 		printf("vndclose(0x%"PRIx64", 0x%x, 0x%x, %p)\n", dev, flags, mode, l);
415 #endif
416 	sc = device_lookup_private(&vnd_cd, unit);
417 	if (sc == NULL)
418 		return ENXIO;
419 
420 	if ((error = vndlock(sc)) != 0)
421 		return error;
422 
423 	part = DISKPART(dev);
424 
425 	/* ...that much closer to allowing unconfiguration... */
426 	switch (mode) {
427 	case S_IFCHR:
428 		sc->sc_dkdev.dk_copenmask &= ~(1 << part);
429 		break;
430 
431 	case S_IFBLK:
432 		sc->sc_dkdev.dk_bopenmask &= ~(1 << part);
433 		break;
434 	}
435 	sc->sc_dkdev.dk_openmask =
436 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
437 
438 	vndunlock(sc);
439 
440 	if ((sc->sc_flags & VNF_INITED) == 0) {
441 		if ((error = vnd_destroy(sc->sc_dev)) != 0) {
442 			aprint_error_dev(sc->sc_dev,
443 			    "unable to detach instance\n");
444 			return error;
445 		}
446 	}
447 
448 	return 0;
449 }
450 
451 /*
452  * Queue the request, and wakeup the kernel thread to handle it.
453  */
454 static void
455 vndstrategy(struct buf *bp)
456 {
457 	int unit = vndunit(bp->b_dev);
458 	struct vnd_softc *vnd =
459 	    device_lookup_private(&vnd_cd, unit);
460 	struct disklabel *lp;
461 	daddr_t blkno;
462 	int s = splbio();
463 
464 	if (vnd == NULL) {
465 		bp->b_error = ENXIO;
466 		goto done;
467 	}
468 	lp = vnd->sc_dkdev.dk_label;
469 
470 	if ((vnd->sc_flags & VNF_INITED) == 0) {
471 		bp->b_error = ENXIO;
472 		goto done;
473 	}
474 
475 	/*
476 	 * The transfer must be a whole number of blocks.
477 	 */
478 	if ((bp->b_bcount % lp->d_secsize) != 0) {
479 		bp->b_error = EINVAL;
480 		goto done;
481 	}
482 
483 	/*
484 	 * check if we're read-only.
485 	 */
486 	if ((vnd->sc_flags & VNF_READONLY) && !(bp->b_flags & B_READ)) {
487 		bp->b_error = EACCES;
488 		goto done;
489 	}
490 
491 	/* If it's a nil transfer, wake up the top half now. */
492 	if (bp->b_bcount == 0) {
493 		goto done;
494 	}
495 
496 	/*
497 	 * Do bounds checking and adjust transfer.  If there's an error,
498 	 * the bounds check will flag that for us.
499 	 */
500 	if (DISKPART(bp->b_dev) == RAW_PART) {
501 		if (bounds_check_with_mediasize(bp, DEV_BSIZE,
502 		    vnd->sc_size) <= 0)
503 			goto done;
504 	} else {
505 		if (bounds_check_with_label(&vnd->sc_dkdev,
506 		    bp, vnd->sc_flags & (VNF_WLABEL|VNF_LABELLING)) <= 0)
507 			goto done;
508 	}
509 
510 	/*
511 	 * Put the block number in terms of the logical blocksize
512 	 * of the "device".
513 	 */
514 
515 	blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
516 
517 	/*
518 	 * Translate the partition-relative block number to an absolute.
519 	 */
520 	if (DISKPART(bp->b_dev) != RAW_PART) {
521 		struct partition *pp;
522 
523 		pp = &vnd->sc_dkdev.dk_label->d_partitions[
524 		    DISKPART(bp->b_dev)];
525 		blkno += pp->p_offset;
526 	}
527 	bp->b_rawblkno = blkno;
528 
529 #ifdef DEBUG
530 	if (vnddebug & VDB_FOLLOW)
531 		printf("vndstrategy(%p): unit %d\n", bp, unit);
532 #endif
533 	bufq_put(vnd->sc_tab, bp);
534 	wakeup(&vnd->sc_tab);
535 	splx(s);
536 	return;
537 
538 done:
539 	bp->b_resid = bp->b_bcount;
540 	biodone(bp);
541 	splx(s);
542 }
543 
544 static bool
545 vnode_has_strategy(struct vnd_softc *vnd)
546 {
547 	return vnode_has_op(vnd->sc_vp, VOFFSET(vop_bmap)) &&
548 	    vnode_has_op(vnd->sc_vp, VOFFSET(vop_strategy));
549 }
550 
551 /* XXX this function needs a reliable check to detect
552  * sparse files. Otherwise, bmap/strategy may be used
553  * and fail on non-allocated blocks. VOP_READ/VOP_WRITE
554  * works on sparse files.
555  */
556 #if notyet
557 static bool
558 vnode_strategy_probe(struct vnd_softc *vnd)
559 {
560 	int error;
561 	daddr_t nbn;
562 
563 	if (!vnode_has_strategy(vnd))
564 		return false;
565 
566 	/* Convert the first logical block number to its
567 	 * physical block number.
568 	 */
569 	error = 0;
570 	vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
571 	error = VOP_BMAP(vnd->sc_vp, 0, NULL, &nbn, NULL);
572 	VOP_UNLOCK(vnd->sc_vp, 0);
573 
574 	/* Test if that worked. */
575 	if (error == 0 && (long)nbn == -1)
576 		return false;
577 
578 	return true;
579 }
580 #endif
581 
582 static void
583 vndthread(void *arg)
584 {
585 	struct vnd_softc *vnd = arg;
586 	bool usestrategy;
587 	int s;
588 
589 	/* Determine whether we can *use* VOP_BMAP and VOP_STRATEGY to
590 	 * directly access the backing vnode.  If we can, use these two
591 	 * operations to avoid messing with the local buffer cache.
592 	 * Otherwise fall back to regular VOP_READ/VOP_WRITE operations
593 	 * which are guaranteed to work with any file system. */
594 	usestrategy = vnode_has_strategy(vnd);
595 
596 #ifdef DEBUG
597 	if (vnddebug & VDB_INIT)
598 		printf("vndthread: vp %p, %s\n", vnd->sc_vp,
599 		    usestrategy ?
600 		    "using bmap/strategy operations" :
601 		    "using read/write operations");
602 #endif
603 
604 	s = splbio();
605 	vnd->sc_flags |= VNF_KTHREAD;
606 	wakeup(&vnd->sc_kthread);
607 
608 	/*
609 	 * Dequeue requests and serve them depending on the available
610 	 * vnode operations.
611 	 */
612 	while ((vnd->sc_flags & VNF_VUNCONF) == 0) {
613 		struct vndxfer *vnx;
614 		int flags;
615 		struct buf *obp;
616 		struct buf *bp;
617 
618 		obp = bufq_get(vnd->sc_tab);
619 		if (obp == NULL) {
620 			tsleep(&vnd->sc_tab, PRIBIO, "vndbp", 0);
621 			continue;
622 		};
623 		splx(s);
624 		flags = obp->b_flags;
625 #ifdef DEBUG
626 		if (vnddebug & VDB_FOLLOW)
627 			printf("vndthread(%p)\n", obp);
628 #endif
629 
630 		if (vnd->sc_vp->v_mount == NULL) {
631 			obp->b_error = ENXIO;
632 			goto done;
633 		}
634 #ifdef VND_COMPRESSION
635 		/* handle a compressed read */
636 		if ((flags & B_READ) != 0 && (vnd->sc_flags & VNF_COMP)) {
637 			off_t bn;
638 
639 			/* Convert to a byte offset within the file. */
640 			bn = obp->b_rawblkno *
641 			    vnd->sc_dkdev.dk_label->d_secsize;
642 
643 			compstrategy(obp, bn);
644 			goto done;
645 		}
646 #endif /* VND_COMPRESSION */
647 
648 		/*
649 		 * Allocate a header for this transfer and link it to the
650 		 * buffer
651 		 */
652 		s = splbio();
653 		vnx = VND_GETXFER(vnd);
654 		splx(s);
655 		vnx->vx_vnd = vnd;
656 
657 		s = splbio();
658 		while (vnd->sc_active >= vnd->sc_maxactive) {
659 			tsleep(&vnd->sc_tab, PRIBIO, "vndac", 0);
660 		}
661 		vnd->sc_active++;
662 		splx(s);
663 
664 		/* Instrumentation. */
665 		disk_busy(&vnd->sc_dkdev);
666 
667 		bp = &vnx->vx_buf;
668 		buf_init(bp);
669 		bp->b_flags = (obp->b_flags & B_READ);
670 		bp->b_oflags = obp->b_oflags;
671 		bp->b_cflags = obp->b_cflags;
672 		bp->b_iodone = vndiodone;
673 		bp->b_private = obp;
674 		bp->b_vp = vnd->sc_vp;
675 		bp->b_objlock = &bp->b_vp->v_interlock;
676 		bp->b_data = obp->b_data;
677 		bp->b_bcount = obp->b_bcount;
678 		BIO_COPYPRIO(bp, obp);
679 
680 		/* Handle the request using the appropriate operations. */
681 		if (usestrategy)
682 			handle_with_strategy(vnd, obp, bp);
683 		else
684 			handle_with_rdwr(vnd, obp, bp);
685 
686 		s = splbio();
687 		continue;
688 
689 done:
690 		biodone(obp);
691 		s = splbio();
692 	}
693 
694 	vnd->sc_flags &= (~VNF_KTHREAD | VNF_VUNCONF);
695 	wakeup(&vnd->sc_kthread);
696 	splx(s);
697 	kthread_exit(0);
698 }
699 
700 /*
701  * Checks if the given vnode supports the requested operation.
702  * The operation is specified the offset returned by VOFFSET.
703  *
704  * XXX The test below used to determine this is quite fragile
705  * because it relies on the file system to use genfs to specify
706  * unimplemented operations.  There might be another way to do
707  * it more cleanly.
708  */
709 static bool
710 vnode_has_op(const struct vnode *vp, int opoffset)
711 {
712 	int (*defaultp)(void *);
713 	int (*opp)(void *);
714 
715 	defaultp = vp->v_op[VOFFSET(vop_default)];
716 	opp = vp->v_op[opoffset];
717 
718 	return opp != defaultp && opp != genfs_eopnotsupp &&
719 	    opp != genfs_badop && opp != genfs_nullop;
720 }
721 
722 /*
723  * Handes the read/write request given in 'bp' using the vnode's VOP_READ
724  * and VOP_WRITE operations.
725  *
726  * 'obp' is a pointer to the original request fed to the vnd device.
727  */
728 static void
729 handle_with_rdwr(struct vnd_softc *vnd, const struct buf *obp, struct buf *bp)
730 {
731 	bool doread;
732 	off_t offset;
733 	size_t resid;
734 	struct vnode *vp;
735 
736 	doread = bp->b_flags & B_READ;
737 	offset = obp->b_rawblkno * vnd->sc_dkdev.dk_label->d_secsize;
738 	vp = vnd->sc_vp;
739 
740 #if defined(DEBUG)
741 	if (vnddebug & VDB_IO)
742 		printf("vnd (rdwr): vp %p, %s, rawblkno 0x%" PRIx64
743 		    ", secsize %d, offset %" PRIu64
744 		    ", bcount %d\n",
745 		    vp, doread ? "read" : "write", obp->b_rawblkno,
746 		    vnd->sc_dkdev.dk_label->d_secsize, offset,
747 		    bp->b_bcount);
748 #endif
749 
750 	/* Issue the read or write operation. */
751 	bp->b_error =
752 	    vn_rdwr(doread ? UIO_READ : UIO_WRITE,
753 	    vp, bp->b_data, bp->b_bcount, offset,
754 	    UIO_SYSSPACE, 0, vnd->sc_cred, &resid, NULL);
755 	bp->b_resid = resid;
756 
757 	/* We need to increase the number of outputs on the vnode if
758 	 * there was any write to it. */
759 	if (!doread) {
760 		mutex_enter(&vp->v_interlock);
761 		vp->v_numoutput++;
762 		mutex_exit(&vp->v_interlock);
763 	}
764 
765 	biodone(bp);
766 }
767 
768 /*
769  * Handes the read/write request given in 'bp' using the vnode's VOP_BMAP
770  * and VOP_STRATEGY operations.
771  *
772  * 'obp' is a pointer to the original request fed to the vnd device.
773  */
774 static void
775 handle_with_strategy(struct vnd_softc *vnd, const struct buf *obp,
776     struct buf *bp)
777 {
778 	int bsize, error, flags, skipped;
779 	size_t resid, sz;
780 	off_t bn, offset;
781 	struct vnode *vp;
782 
783 	flags = obp->b_flags;
784 
785 	if (!(flags & B_READ)) {
786 		vp = bp->b_vp;
787 		mutex_enter(&vp->v_interlock);
788 		vp->v_numoutput++;
789 		mutex_exit(&vp->v_interlock);
790 	}
791 
792 	/* convert to a byte offset within the file. */
793 	bn = obp->b_rawblkno * vnd->sc_dkdev.dk_label->d_secsize;
794 
795 	bsize = vnd->sc_vp->v_mount->mnt_stat.f_iosize;
796 	skipped = 0;
797 
798 	/*
799 	 * Break the request into bsize pieces and feed them
800 	 * sequentially using VOP_BMAP/VOP_STRATEGY.
801 	 * We do it this way to keep from flooding NFS servers if we
802 	 * are connected to an NFS file.  This places the burden on
803 	 * the client rather than the server.
804 	 */
805 	error = 0;
806 	bp->b_resid = bp->b_bcount;
807 	for (offset = 0, resid = bp->b_resid; resid;
808 	    resid -= sz, offset += sz) {
809 		struct buf *nbp;
810 		daddr_t nbn;
811 		int off, nra;
812 
813 		nra = 0;
814 		vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
815 		error = VOP_BMAP(vnd->sc_vp, bn / bsize, &vp, &nbn, &nra);
816 		VOP_UNLOCK(vnd->sc_vp, 0);
817 
818 		if (error == 0 && (long)nbn == -1)
819 			error = EIO;
820 
821 		/*
822 		 * If there was an error or a hole in the file...punt.
823 		 * Note that we may have to wait for any operations
824 		 * that we have already fired off before releasing
825 		 * the buffer.
826 		 *
827 		 * XXX we could deal with holes here but it would be
828 		 * a hassle (in the write case).
829 		 */
830 		if (error) {
831 			skipped += resid;
832 			break;
833 		}
834 
835 #ifdef DEBUG
836 		if (!dovndcluster)
837 			nra = 0;
838 #endif
839 
840 		off = bn % bsize;
841 		sz = MIN(((off_t)1 + nra) * bsize - off, resid);
842 #ifdef	DEBUG
843 		if (vnddebug & VDB_IO)
844 			printf("vndstrategy: vp %p/%p bn 0x%qx/0x%" PRIx64
845 			    " sz 0x%zx\n", vnd->sc_vp, vp, (long long)bn,
846 			    nbn, sz);
847 #endif
848 
849 		nbp = getiobuf(vp, true);
850 		nestiobuf_setup(bp, nbp, offset, sz);
851 		nbp->b_blkno = nbn + btodb(off);
852 
853 #if 0 /* XXX #ifdef DEBUG */
854 		if (vnddebug & VDB_IO)
855 			printf("vndstart(%ld): bp %p vp %p blkno "
856 			    "0x%" PRIx64 " flags %x addr %p cnt 0x%x\n",
857 			    (long) (vnd-vnd_softc), &nbp->vb_buf,
858 			    nbp->vb_buf.b_vp, nbp->vb_buf.b_blkno,
859 			    nbp->vb_buf.b_flags, nbp->vb_buf.b_data,
860 			    nbp->vb_buf.b_bcount);
861 #endif
862 		VOP_STRATEGY(vp, nbp);
863 		bn += sz;
864 	}
865 	nestiobuf_done(bp, skipped, error);
866 }
867 
868 static void
869 vndiodone(struct buf *bp)
870 {
871 	struct vndxfer *vnx = VND_BUFTOXFER(bp);
872 	struct vnd_softc *vnd = vnx->vx_vnd;
873 	struct buf *obp = bp->b_private;
874 
875 	KASSERT(&vnx->vx_buf == bp);
876 	KASSERT(vnd->sc_active > 0);
877 #ifdef DEBUG
878 	if (vnddebug & VDB_IO) {
879 		printf("vndiodone1: bp %p iodone: error %d\n",
880 		    bp, bp->b_error);
881 	}
882 #endif
883 	disk_unbusy(&vnd->sc_dkdev, bp->b_bcount - bp->b_resid,
884 	    (bp->b_flags & B_READ));
885 	vnd->sc_active--;
886 	if (vnd->sc_active == 0) {
887 		wakeup(&vnd->sc_tab);
888 	}
889 	obp->b_error = bp->b_error;
890 	obp->b_resid = bp->b_resid;
891 	buf_destroy(bp);
892 	VND_PUTXFER(vnd, vnx);
893 	biodone(obp);
894 }
895 
896 /* ARGSUSED */
897 static int
898 vndread(dev_t dev, struct uio *uio, int flags)
899 {
900 	int unit = vndunit(dev);
901 	struct vnd_softc *sc;
902 
903 #ifdef DEBUG
904 	if (vnddebug & VDB_FOLLOW)
905 		printf("vndread(0x%"PRIx64", %p)\n", dev, uio);
906 #endif
907 
908 	sc = device_lookup_private(&vnd_cd, unit);
909 	if (sc == NULL)
910 		return ENXIO;
911 
912 	if ((sc->sc_flags & VNF_INITED) == 0)
913 		return ENXIO;
914 
915 	return physio(vndstrategy, NULL, dev, B_READ, minphys, uio);
916 }
917 
918 /* ARGSUSED */
919 static int
920 vndwrite(dev_t dev, struct uio *uio, int flags)
921 {
922 	int unit = vndunit(dev);
923 	struct vnd_softc *sc;
924 
925 #ifdef DEBUG
926 	if (vnddebug & VDB_FOLLOW)
927 		printf("vndwrite(0x%"PRIx64", %p)\n", dev, uio);
928 #endif
929 
930 	sc = device_lookup_private(&vnd_cd, unit);
931 	if (sc == NULL)
932 		return ENXIO;
933 
934 	if ((sc->sc_flags & VNF_INITED) == 0)
935 		return ENXIO;
936 
937 	return physio(vndstrategy, NULL, dev, B_WRITE, minphys, uio);
938 }
939 
940 static int
941 vnd_cget(struct lwp *l, int unit, int *un, struct vattr *va)
942 {
943 	struct vnd_softc *vnd;
944 
945 	if (*un == -1)
946 		*un = unit;
947 	if (*un < 0)
948 		return EINVAL;
949 
950 	vnd = device_lookup_private(&vnd_cd, *un);
951 	if (vnd == NULL)
952 		return (*un >= vnd_cd.cd_ndevs) ? ENXIO : -1;
953 
954 	if ((vnd->sc_flags & VNF_INITED) == 0)
955 		return -1;
956 
957 	return VOP_GETATTR(vnd->sc_vp, va, l->l_cred);
958 }
959 
960 static int
961 vnddoclear(struct vnd_softc *vnd, int pmask, int minor, bool force)
962 {
963 	int error;
964 
965 	if ((error = vndlock(vnd)) != 0)
966 		return error;
967 
968 	/*
969 	 * Don't unconfigure if any other partitions are open
970 	 * or if both the character and block flavors of this
971 	 * partition are open.
972 	 */
973 	if (((vnd->sc_dkdev.dk_openmask & ~pmask) ||
974 	    ((vnd->sc_dkdev.dk_bopenmask & pmask) &&
975 	    (vnd->sc_dkdev.dk_copenmask & pmask))) && !force) {
976 		vndunlock(vnd);
977 		return EBUSY;
978 	}
979 
980 	/*
981 	 * XXX vndclear() might call vndclose() implicitly;
982 	 * release lock to avoid recursion
983 	 *
984 	 * Set VNF_CLEARING to prevent vndopen() from
985 	 * sneaking in after we vndunlock().
986 	 */
987 	vnd->sc_flags |= VNF_CLEARING;
988 	vndunlock(vnd);
989 	vndclear(vnd, minor);
990 #ifdef DEBUG
991 	if (vnddebug & VDB_INIT)
992 		printf("vndioctl: CLRed\n");
993 #endif
994 
995 	/* Destroy the xfer and buffer pools. */
996 	pool_destroy(&vnd->sc_vxpool);
997 
998 	/* Detach the disk. */
999 	disk_detach(&vnd->sc_dkdev);
1000 
1001 	return 0;
1002 }
1003 
1004 /* ARGSUSED */
1005 static int
1006 vndioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l)
1007 {
1008 	bool force;
1009 	int unit = vndunit(dev);
1010 	struct vnd_softc *vnd;
1011 	struct vnd_ioctl *vio;
1012 	struct vattr vattr;
1013 	struct nameidata nd;
1014 	int error, part, pmask;
1015 	size_t geomsize;
1016 	int fflags;
1017 #ifdef __HAVE_OLD_DISKLABEL
1018 	struct disklabel newlabel;
1019 #endif
1020 
1021 #ifdef DEBUG
1022 	if (vnddebug & VDB_FOLLOW)
1023 		printf("vndioctl(0x%"PRIx64", 0x%lx, %p, 0x%x, %p): unit %d\n",
1024 		    dev, cmd, data, flag, l->l_proc, unit);
1025 #endif
1026 	vnd = device_lookup_private(&vnd_cd, unit);
1027 	if (vnd == NULL &&
1028 #ifdef COMPAT_30
1029 	    cmd != VNDIOOCGET &&
1030 #endif
1031 	    cmd != VNDIOCGET)
1032 		return ENXIO;
1033 	vio = (struct vnd_ioctl *)data;
1034 
1035 	/* Must be open for writes for these commands... */
1036 	switch (cmd) {
1037 	case VNDIOCSET:
1038 	case VNDIOCCLR:
1039 	case DIOCSDINFO:
1040 	case DIOCWDINFO:
1041 #ifdef __HAVE_OLD_DISKLABEL
1042 	case ODIOCSDINFO:
1043 	case ODIOCWDINFO:
1044 #endif
1045 	case DIOCKLABEL:
1046 	case DIOCWLABEL:
1047 		if ((flag & FWRITE) == 0)
1048 			return EBADF;
1049 	}
1050 
1051 	/* Must be initialized for these... */
1052 	switch (cmd) {
1053 	case VNDIOCCLR:
1054 	case DIOCGDINFO:
1055 	case DIOCSDINFO:
1056 	case DIOCWDINFO:
1057 	case DIOCGPART:
1058 	case DIOCKLABEL:
1059 	case DIOCWLABEL:
1060 	case DIOCGDEFLABEL:
1061 	case DIOCCACHESYNC:
1062 #ifdef __HAVE_OLD_DISKLABEL
1063 	case ODIOCGDINFO:
1064 	case ODIOCSDINFO:
1065 	case ODIOCWDINFO:
1066 	case ODIOCGDEFLABEL:
1067 #endif
1068 		if ((vnd->sc_flags & VNF_INITED) == 0)
1069 			return ENXIO;
1070 	}
1071 
1072 	switch (cmd) {
1073 	case VNDIOCSET:
1074 		if (vnd->sc_flags & VNF_INITED)
1075 			return EBUSY;
1076 
1077 		if ((error = vndlock(vnd)) != 0)
1078 			return error;
1079 
1080 		fflags = FREAD;
1081 		if ((vio->vnd_flags & VNDIOF_READONLY) == 0)
1082 			fflags |= FWRITE;
1083 		NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, vio->vnd_file);
1084 		if ((error = vn_open(&nd, fflags, 0)) != 0)
1085 			goto unlock_and_exit;
1086 		KASSERT(l);
1087 		error = VOP_GETATTR(nd.ni_vp, &vattr, l->l_cred);
1088 		if (!error && nd.ni_vp->v_type != VREG)
1089 			error = EOPNOTSUPP;
1090 		if (error) {
1091 			VOP_UNLOCK(nd.ni_vp, 0);
1092 			goto close_and_exit;
1093 		}
1094 
1095 		/* If using a compressed file, initialize its info */
1096 		/* (or abort with an error if kernel has no compression) */
1097 		if (vio->vnd_flags & VNF_COMP) {
1098 #ifdef VND_COMPRESSION
1099 			struct vnd_comp_header *ch;
1100 			int i;
1101 			u_int32_t comp_size;
1102 			u_int32_t comp_maxsize;
1103 
1104 			/* allocate space for compresed file header */
1105 			ch = malloc(sizeof(struct vnd_comp_header),
1106 			M_TEMP, M_WAITOK);
1107 
1108 			/* read compressed file header */
1109 			error = vn_rdwr(UIO_READ, nd.ni_vp, (void *)ch,
1110 			  sizeof(struct vnd_comp_header), 0, UIO_SYSSPACE,
1111 			  IO_UNIT|IO_NODELOCKED, l->l_cred, NULL, NULL);
1112 			if (error) {
1113 				free(ch, M_TEMP);
1114 				VOP_UNLOCK(nd.ni_vp, 0);
1115 				goto close_and_exit;
1116 			}
1117 
1118 			/* save some header info */
1119 			vnd->sc_comp_blksz = ntohl(ch->block_size);
1120 			/* note last offset is the file byte size */
1121 			vnd->sc_comp_numoffs = ntohl(ch->num_blocks)+1;
1122 			free(ch, M_TEMP);
1123 			if (vnd->sc_comp_blksz == 0 ||
1124 			    vnd->sc_comp_blksz % DEV_BSIZE !=0) {
1125 				VOP_UNLOCK(nd.ni_vp, 0);
1126 				error = EINVAL;
1127 				goto close_and_exit;
1128 			}
1129 			if (sizeof(struct vnd_comp_header) +
1130 			  sizeof(u_int64_t) * vnd->sc_comp_numoffs >
1131 			  vattr.va_size) {
1132 				VOP_UNLOCK(nd.ni_vp, 0);
1133 				error = EINVAL;
1134 				goto close_and_exit;
1135 			}
1136 
1137 			/* set decompressed file size */
1138 			vattr.va_size =
1139 			    ((u_quad_t)vnd->sc_comp_numoffs - 1) *
1140 			     (u_quad_t)vnd->sc_comp_blksz;
1141 
1142 			/* allocate space for all the compressed offsets */
1143 			vnd->sc_comp_offsets =
1144 			malloc(sizeof(u_int64_t) * vnd->sc_comp_numoffs,
1145 			M_DEVBUF, M_WAITOK);
1146 
1147 			/* read in the offsets */
1148 			error = vn_rdwr(UIO_READ, nd.ni_vp,
1149 			  (void *)vnd->sc_comp_offsets,
1150 			  sizeof(u_int64_t) * vnd->sc_comp_numoffs,
1151 			  sizeof(struct vnd_comp_header), UIO_SYSSPACE,
1152 			  IO_UNIT|IO_NODELOCKED, l->l_cred, NULL, NULL);
1153 			if (error) {
1154 				VOP_UNLOCK(nd.ni_vp, 0);
1155 				goto close_and_exit;
1156 			}
1157 			/*
1158 			 * find largest block size (used for allocation limit).
1159 			 * Also convert offset to native byte order.
1160 			 */
1161 			comp_maxsize = 0;
1162 			for (i = 0; i < vnd->sc_comp_numoffs - 1; i++) {
1163 				vnd->sc_comp_offsets[i] =
1164 				  be64toh(vnd->sc_comp_offsets[i]);
1165 				comp_size = be64toh(vnd->sc_comp_offsets[i + 1])
1166 				  - vnd->sc_comp_offsets[i];
1167 				if (comp_size > comp_maxsize)
1168 					comp_maxsize = comp_size;
1169 			}
1170 			vnd->sc_comp_offsets[vnd->sc_comp_numoffs - 1] =
1171 			  be64toh(vnd->sc_comp_offsets[vnd->sc_comp_numoffs - 1]);
1172 
1173 			/* create compressed data buffer */
1174 			vnd->sc_comp_buff = malloc(comp_maxsize,
1175 			  M_DEVBUF, M_WAITOK);
1176 
1177 			/* create decompressed buffer */
1178 			vnd->sc_comp_decombuf = malloc(vnd->sc_comp_blksz,
1179 			  M_DEVBUF, M_WAITOK);
1180 			vnd->sc_comp_buffblk = -1;
1181 
1182 			/* Initialize decompress stream */
1183 			memset(&vnd->sc_comp_stream, 0, sizeof(z_stream));
1184 			vnd->sc_comp_stream.zalloc = vnd_alloc;
1185 			vnd->sc_comp_stream.zfree = vnd_free;
1186 			error = inflateInit2(&vnd->sc_comp_stream, MAX_WBITS);
1187 			if (error) {
1188 				if (vnd->sc_comp_stream.msg)
1189 					printf("vnd%d: compressed file, %s\n",
1190 					  unit, vnd->sc_comp_stream.msg);
1191 				VOP_UNLOCK(nd.ni_vp, 0);
1192 				error = EINVAL;
1193 				goto close_and_exit;
1194 			}
1195 
1196 			vnd->sc_flags |= VNF_COMP | VNF_READONLY;
1197 #else /* !VND_COMPRESSION */
1198 			VOP_UNLOCK(nd.ni_vp, 0);
1199 			error = EOPNOTSUPP;
1200 			goto close_and_exit;
1201 #endif /* VND_COMPRESSION */
1202 		}
1203 
1204 		VOP_UNLOCK(nd.ni_vp, 0);
1205 		vnd->sc_vp = nd.ni_vp;
1206 		vnd->sc_size = btodb(vattr.va_size);	/* note truncation */
1207 
1208 		/*
1209 		 * Use pseudo-geometry specified.  If none was provided,
1210 		 * use "standard" Adaptec fictitious geometry.
1211 		 */
1212 		if (vio->vnd_flags & VNDIOF_HASGEOM) {
1213 
1214 			memcpy(&vnd->sc_geom, &vio->vnd_geom,
1215 			    sizeof(vio->vnd_geom));
1216 
1217 			/*
1218 			 * Sanity-check the sector size.
1219 			 * XXX Don't allow secsize < DEV_BSIZE.	 Should
1220 			 * XXX we?
1221 			 */
1222 			if (vnd->sc_geom.vng_secsize < DEV_BSIZE ||
1223 			    (vnd->sc_geom.vng_secsize % DEV_BSIZE) != 0 ||
1224 			    vnd->sc_geom.vng_ncylinders == 0 ||
1225 			    (vnd->sc_geom.vng_ntracks *
1226 			     vnd->sc_geom.vng_nsectors) == 0) {
1227 				error = EINVAL;
1228 				goto close_and_exit;
1229 			}
1230 
1231 			/*
1232 			 * Compute the size (in DEV_BSIZE blocks) specified
1233 			 * by the geometry.
1234 			 */
1235 			geomsize = (vnd->sc_geom.vng_nsectors *
1236 			    vnd->sc_geom.vng_ntracks *
1237 			    vnd->sc_geom.vng_ncylinders) *
1238 			    (vnd->sc_geom.vng_secsize / DEV_BSIZE);
1239 
1240 			/*
1241 			 * Sanity-check the size against the specified
1242 			 * geometry.
1243 			 */
1244 			if (vnd->sc_size < geomsize) {
1245 				error = EINVAL;
1246 				goto close_and_exit;
1247 			}
1248 		} else if (vnd->sc_size >= (32 * 64)) {
1249 			/*
1250 			 * Size must be at least 2048 DEV_BSIZE blocks
1251 			 * (1M) in order to use this geometry.
1252 			 */
1253 			vnd->sc_geom.vng_secsize = DEV_BSIZE;
1254 			vnd->sc_geom.vng_nsectors = 32;
1255 			vnd->sc_geom.vng_ntracks = 64;
1256 			vnd->sc_geom.vng_ncylinders = vnd->sc_size / (64 * 32);
1257 		} else {
1258 			vnd->sc_geom.vng_secsize = DEV_BSIZE;
1259 			vnd->sc_geom.vng_nsectors = 1;
1260 			vnd->sc_geom.vng_ntracks = 1;
1261 			vnd->sc_geom.vng_ncylinders = vnd->sc_size;
1262 		}
1263 
1264 		vnd_set_properties(vnd);
1265 
1266 		if (vio->vnd_flags & VNDIOF_READONLY) {
1267 			vnd->sc_flags |= VNF_READONLY;
1268 		}
1269 
1270 		if ((error = vndsetcred(vnd, l->l_cred)) != 0)
1271 			goto close_and_exit;
1272 
1273 		vndthrottle(vnd, vnd->sc_vp);
1274 		vio->vnd_size = dbtob(vnd->sc_size);
1275 		vnd->sc_flags |= VNF_INITED;
1276 
1277 		/* create the kernel thread, wait for it to be up */
1278 		error = kthread_create(PRI_NONE, 0, NULL, vndthread, vnd,
1279 		    &vnd->sc_kthread, device_xname(vnd->sc_dev));
1280 		if (error)
1281 			goto close_and_exit;
1282 		while ((vnd->sc_flags & VNF_KTHREAD) == 0) {
1283 			tsleep(&vnd->sc_kthread, PRIBIO, "vndthr", 0);
1284 		}
1285 #ifdef DEBUG
1286 		if (vnddebug & VDB_INIT)
1287 			printf("vndioctl: SET vp %p size 0x%lx %d/%d/%d/%d\n",
1288 			    vnd->sc_vp, (unsigned long) vnd->sc_size,
1289 			    vnd->sc_geom.vng_secsize,
1290 			    vnd->sc_geom.vng_nsectors,
1291 			    vnd->sc_geom.vng_ntracks,
1292 			    vnd->sc_geom.vng_ncylinders);
1293 #endif
1294 
1295 		/* Attach the disk. */
1296 		disk_attach(&vnd->sc_dkdev);
1297 
1298 		/* Initialize the xfer and buffer pools. */
1299 		pool_init(&vnd->sc_vxpool, sizeof(struct vndxfer), 0,
1300 		    0, 0, "vndxpl", NULL, IPL_BIO);
1301 
1302 		/* Try and read the disklabel. */
1303 		vndgetdisklabel(dev, vnd);
1304 
1305 		vndunlock(vnd);
1306 
1307 		break;
1308 
1309 close_and_exit:
1310 		(void) vn_close(nd.ni_vp, fflags, l->l_cred);
1311 unlock_and_exit:
1312 #ifdef VND_COMPRESSION
1313 		/* free any allocated memory (for compressed file) */
1314 		if (vnd->sc_comp_offsets) {
1315 			free(vnd->sc_comp_offsets, M_DEVBUF);
1316 			vnd->sc_comp_offsets = NULL;
1317 		}
1318 		if (vnd->sc_comp_buff) {
1319 			free(vnd->sc_comp_buff, M_DEVBUF);
1320 			vnd->sc_comp_buff = NULL;
1321 		}
1322 		if (vnd->sc_comp_decombuf) {
1323 			free(vnd->sc_comp_decombuf, M_DEVBUF);
1324 			vnd->sc_comp_decombuf = NULL;
1325 		}
1326 #endif /* VND_COMPRESSION */
1327 		vndunlock(vnd);
1328 		return error;
1329 
1330 	case VNDIOCCLR:
1331 		part = DISKPART(dev);
1332 		pmask = (1 << part);
1333 		force = (vio->vnd_flags & VNDIOF_FORCE) != 0;
1334 
1335 		if ((error = vnddoclear(vnd, pmask, minor(dev), force)) != 0)
1336 			return error;
1337 
1338 		break;
1339 
1340 #ifdef COMPAT_30
1341 	case VNDIOOCGET: {
1342 		struct vnd_ouser *vnu;
1343 		struct vattr va;
1344 		vnu = (struct vnd_ouser *)data;
1345 		KASSERT(l);
1346 		switch (error = vnd_cget(l, unit, &vnu->vnu_unit, &va)) {
1347 		case 0:
1348 			vnu->vnu_dev = va.va_fsid;
1349 			vnu->vnu_ino = va.va_fileid;
1350 			break;
1351 		case -1:
1352 			/* unused is not an error */
1353 			vnu->vnu_dev = 0;
1354 			vnu->vnu_ino = 0;
1355 			break;
1356 		default:
1357 			return error;
1358 		}
1359 		break;
1360 	}
1361 #endif
1362 	case VNDIOCGET: {
1363 		struct vnd_user *vnu;
1364 		struct vattr va;
1365 		vnu = (struct vnd_user *)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 
1383 	case DIOCGDINFO:
1384 		*(struct disklabel *)data = *(vnd->sc_dkdev.dk_label);
1385 		break;
1386 
1387 #ifdef __HAVE_OLD_DISKLABEL
1388 	case ODIOCGDINFO:
1389 		newlabel = *(vnd->sc_dkdev.dk_label);
1390 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1391 			return ENOTTY;
1392 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1393 		break;
1394 #endif
1395 
1396 	case DIOCGPART:
1397 		((struct partinfo *)data)->disklab = vnd->sc_dkdev.dk_label;
1398 		((struct partinfo *)data)->part =
1399 		    &vnd->sc_dkdev.dk_label->d_partitions[DISKPART(dev)];
1400 		break;
1401 
1402 	case DIOCWDINFO:
1403 	case DIOCSDINFO:
1404 #ifdef __HAVE_OLD_DISKLABEL
1405 	case ODIOCWDINFO:
1406 	case ODIOCSDINFO:
1407 #endif
1408 	{
1409 		struct disklabel *lp;
1410 
1411 		if ((error = vndlock(vnd)) != 0)
1412 			return error;
1413 
1414 		vnd->sc_flags |= VNF_LABELLING;
1415 
1416 #ifdef __HAVE_OLD_DISKLABEL
1417 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
1418 			memset(&newlabel, 0, sizeof newlabel);
1419 			memcpy(&newlabel, data, sizeof (struct olddisklabel));
1420 			lp = &newlabel;
1421 		} else
1422 #endif
1423 		lp = (struct disklabel *)data;
1424 
1425 		error = setdisklabel(vnd->sc_dkdev.dk_label,
1426 		    lp, 0, vnd->sc_dkdev.dk_cpulabel);
1427 		if (error == 0) {
1428 			if (cmd == DIOCWDINFO
1429 #ifdef __HAVE_OLD_DISKLABEL
1430 			    || cmd == ODIOCWDINFO
1431 #endif
1432 			   )
1433 				error = writedisklabel(VNDLABELDEV(dev),
1434 				    vndstrategy, vnd->sc_dkdev.dk_label,
1435 				    vnd->sc_dkdev.dk_cpulabel);
1436 		}
1437 
1438 		vnd->sc_flags &= ~VNF_LABELLING;
1439 
1440 		vndunlock(vnd);
1441 
1442 		if (error)
1443 			return error;
1444 		break;
1445 	}
1446 
1447 	case DIOCKLABEL:
1448 		if (*(int *)data != 0)
1449 			vnd->sc_flags |= VNF_KLABEL;
1450 		else
1451 			vnd->sc_flags &= ~VNF_KLABEL;
1452 		break;
1453 
1454 	case DIOCWLABEL:
1455 		if (*(int *)data != 0)
1456 			vnd->sc_flags |= VNF_WLABEL;
1457 		else
1458 			vnd->sc_flags &= ~VNF_WLABEL;
1459 		break;
1460 
1461 	case DIOCGDEFLABEL:
1462 		vndgetdefaultlabel(vnd, (struct disklabel *)data);
1463 		break;
1464 
1465 #ifdef __HAVE_OLD_DISKLABEL
1466 	case ODIOCGDEFLABEL:
1467 		vndgetdefaultlabel(vnd, &newlabel);
1468 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1469 			return ENOTTY;
1470 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1471 		break;
1472 #endif
1473 
1474 	case DIOCCACHESYNC:
1475 		vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1476 		error = VOP_FSYNC(vnd->sc_vp, vnd->sc_cred,
1477 		    FSYNC_WAIT | FSYNC_DATAONLY | FSYNC_CACHE, 0, 0);
1478 		VOP_UNLOCK(vnd->sc_vp, 0);
1479 		return error;
1480 
1481 	default:
1482 		return ENOTTY;
1483 	}
1484 
1485 	return 0;
1486 }
1487 
1488 /*
1489  * Duplicate the current processes' credentials.  Since we are called only
1490  * as the result of a SET ioctl and only root can do that, any future access
1491  * to this "disk" is essentially as root.  Note that credentials may change
1492  * if some other uid can write directly to the mapped file (NFS).
1493  */
1494 static int
1495 vndsetcred(struct vnd_softc *vnd, kauth_cred_t cred)
1496 {
1497 	struct uio auio;
1498 	struct iovec aiov;
1499 	char *tmpbuf;
1500 	int error;
1501 
1502 	vnd->sc_cred = kauth_cred_dup(cred);
1503 	tmpbuf = malloc(DEV_BSIZE, M_TEMP, M_WAITOK);
1504 
1505 	/* XXX: Horrible kludge to establish credentials for NFS */
1506 	aiov.iov_base = tmpbuf;
1507 	aiov.iov_len = min(DEV_BSIZE, dbtob(vnd->sc_size));
1508 	auio.uio_iov = &aiov;
1509 	auio.uio_iovcnt = 1;
1510 	auio.uio_offset = 0;
1511 	auio.uio_rw = UIO_READ;
1512 	auio.uio_resid = aiov.iov_len;
1513 	UIO_SETUP_SYSSPACE(&auio);
1514 	vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1515 	error = VOP_READ(vnd->sc_vp, &auio, 0, vnd->sc_cred);
1516 	if (error == 0) {
1517 		/*
1518 		 * Because vnd does all IO directly through the vnode
1519 		 * we need to flush (at least) the buffer from the above
1520 		 * VOP_READ from the buffer cache to prevent cache
1521 		 * incoherencies.  Also, be careful to write dirty
1522 		 * buffers back to stable storage.
1523 		 */
1524 		error = vinvalbuf(vnd->sc_vp, V_SAVE, vnd->sc_cred,
1525 			    curlwp, 0, 0);
1526 	}
1527 	VOP_UNLOCK(vnd->sc_vp, 0);
1528 
1529 	free(tmpbuf, M_TEMP);
1530 	return error;
1531 }
1532 
1533 /*
1534  * Set maxactive based on FS type
1535  */
1536 static void
1537 vndthrottle(struct vnd_softc *vnd, struct vnode *vp)
1538 {
1539 #ifdef NFS
1540 	extern int (**nfsv2_vnodeop_p)(void *);
1541 
1542 	if (vp->v_op == nfsv2_vnodeop_p)
1543 		vnd->sc_maxactive = 2;
1544 	else
1545 #endif
1546 		vnd->sc_maxactive = 8;
1547 
1548 	if (vnd->sc_maxactive < 1)
1549 		vnd->sc_maxactive = 1;
1550 }
1551 
1552 #if 0
1553 static void
1554 vndshutdown(void)
1555 {
1556 	struct vnd_softc *vnd;
1557 
1558 	for (vnd = &vnd_softc[0]; vnd < &vnd_softc[numvnd]; vnd++)
1559 		if (vnd->sc_flags & VNF_INITED)
1560 			vndclear(vnd);
1561 }
1562 #endif
1563 
1564 static void
1565 vndclear(struct vnd_softc *vnd, int myminor)
1566 {
1567 	struct vnode *vp = vnd->sc_vp;
1568 	int fflags = FREAD;
1569 	int bmaj, cmaj, i, mn;
1570 	int s;
1571 
1572 #ifdef DEBUG
1573 	if (vnddebug & VDB_FOLLOW)
1574 		printf("vndclear(%p): vp %p\n", vnd, vp);
1575 #endif
1576 	/* locate the major number */
1577 	bmaj = bdevsw_lookup_major(&vnd_bdevsw);
1578 	cmaj = cdevsw_lookup_major(&vnd_cdevsw);
1579 
1580 	/* Nuke the vnodes for any open instances */
1581 	for (i = 0; i < MAXPARTITIONS; i++) {
1582 		mn = DISKMINOR(device_unit(vnd->sc_dev), i);
1583 		vdevgone(bmaj, mn, mn, VBLK);
1584 		if (mn != myminor) /* XXX avoid to kill own vnode */
1585 			vdevgone(cmaj, mn, mn, VCHR);
1586 	}
1587 
1588 	if ((vnd->sc_flags & VNF_READONLY) == 0)
1589 		fflags |= FWRITE;
1590 
1591 	s = splbio();
1592 	bufq_drain(vnd->sc_tab);
1593 	splx(s);
1594 
1595 	vnd->sc_flags |= VNF_VUNCONF;
1596 	wakeup(&vnd->sc_tab);
1597 	while (vnd->sc_flags & VNF_KTHREAD)
1598 		tsleep(&vnd->sc_kthread, PRIBIO, "vnthr", 0);
1599 
1600 #ifdef VND_COMPRESSION
1601 	/* free the compressed file buffers */
1602 	if (vnd->sc_flags & VNF_COMP) {
1603 		if (vnd->sc_comp_offsets) {
1604 			free(vnd->sc_comp_offsets, M_DEVBUF);
1605 			vnd->sc_comp_offsets = NULL;
1606 		}
1607 		if (vnd->sc_comp_buff) {
1608 			free(vnd->sc_comp_buff, M_DEVBUF);
1609 			vnd->sc_comp_buff = NULL;
1610 		}
1611 		if (vnd->sc_comp_decombuf) {
1612 			free(vnd->sc_comp_decombuf, M_DEVBUF);
1613 			vnd->sc_comp_decombuf = NULL;
1614 		}
1615 	}
1616 #endif /* VND_COMPRESSION */
1617 	vnd->sc_flags &=
1618 	    ~(VNF_INITED | VNF_READONLY | VNF_VLABEL
1619 	      | VNF_VUNCONF | VNF_COMP);
1620 	if (vp == NULL)
1621 		panic("vndclear: null vp");
1622 	(void) vn_close(vp, fflags, vnd->sc_cred);
1623 	kauth_cred_free(vnd->sc_cred);
1624 	vnd->sc_vp = NULL;
1625 	vnd->sc_cred = NULL;
1626 	vnd->sc_size = 0;
1627 }
1628 
1629 static int
1630 vndsize(dev_t dev)
1631 {
1632 	struct vnd_softc *sc;
1633 	struct disklabel *lp;
1634 	int part, unit, omask;
1635 	int size;
1636 
1637 	unit = vndunit(dev);
1638 	sc = device_lookup_private(&vnd_cd, unit);
1639 	if (sc == NULL)
1640 		return -1;
1641 
1642 	if ((sc->sc_flags & VNF_INITED) == 0)
1643 		return -1;
1644 
1645 	part = DISKPART(dev);
1646 	omask = sc->sc_dkdev.dk_openmask & (1 << part);
1647 	lp = sc->sc_dkdev.dk_label;
1648 
1649 	if (omask == 0 && vndopen(dev, 0, S_IFBLK, curlwp))	/* XXX */
1650 		return -1;
1651 
1652 	if (lp->d_partitions[part].p_fstype != FS_SWAP)
1653 		size = -1;
1654 	else
1655 		size = lp->d_partitions[part].p_size *
1656 		    (lp->d_secsize / DEV_BSIZE);
1657 
1658 	if (omask == 0 && vndclose(dev, 0, S_IFBLK, curlwp))	/* XXX */
1659 		return -1;
1660 
1661 	return size;
1662 }
1663 
1664 static int
1665 vnddump(dev_t dev, daddr_t blkno, void *va,
1666     size_t size)
1667 {
1668 
1669 	/* Not implemented. */
1670 	return ENXIO;
1671 }
1672 
1673 static void
1674 vndgetdefaultlabel(struct vnd_softc *sc, struct disklabel *lp)
1675 {
1676 	struct vndgeom *vng = &sc->sc_geom;
1677 	struct partition *pp;
1678 
1679 	memset(lp, 0, sizeof(*lp));
1680 
1681 	lp->d_secperunit = sc->sc_size / (vng->vng_secsize / DEV_BSIZE);
1682 	lp->d_secsize = vng->vng_secsize;
1683 	lp->d_nsectors = vng->vng_nsectors;
1684 	lp->d_ntracks = vng->vng_ntracks;
1685 	lp->d_ncylinders = vng->vng_ncylinders;
1686 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1687 
1688 	strncpy(lp->d_typename, "vnd", sizeof(lp->d_typename));
1689 	lp->d_type = DTYPE_VND;
1690 	strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
1691 	lp->d_rpm = 3600;
1692 	lp->d_interleave = 1;
1693 	lp->d_flags = 0;
1694 
1695 	pp = &lp->d_partitions[RAW_PART];
1696 	pp->p_offset = 0;
1697 	pp->p_size = lp->d_secperunit;
1698 	pp->p_fstype = FS_UNUSED;
1699 	lp->d_npartitions = RAW_PART + 1;
1700 
1701 	lp->d_magic = DISKMAGIC;
1702 	lp->d_magic2 = DISKMAGIC;
1703 	lp->d_checksum = dkcksum(lp);
1704 }
1705 
1706 /*
1707  * Read the disklabel from a vnd.  If one is not present, create a fake one.
1708  */
1709 static void
1710 vndgetdisklabel(dev_t dev, struct vnd_softc *sc)
1711 {
1712 	const char *errstring;
1713 	struct disklabel *lp = sc->sc_dkdev.dk_label;
1714 	struct cpu_disklabel *clp = sc->sc_dkdev.dk_cpulabel;
1715 	int i;
1716 
1717 	memset(clp, 0, sizeof(*clp));
1718 
1719 	vndgetdefaultlabel(sc, lp);
1720 
1721 	/*
1722 	 * Call the generic disklabel extraction routine.
1723 	 */
1724 	errstring = readdisklabel(VNDLABELDEV(dev), vndstrategy, lp, clp);
1725 	if (errstring) {
1726 		/*
1727 		 * Lack of disklabel is common, but we print the warning
1728 		 * anyway, since it might contain other useful information.
1729 		 */
1730 		aprint_normal_dev(sc->sc_dev, "%s\n", errstring);
1731 
1732 		/*
1733 		 * For historical reasons, if there's no disklabel
1734 		 * present, all partitions must be FS_BSDFFS and
1735 		 * occupy the entire disk.
1736 		 */
1737 		for (i = 0; i < MAXPARTITIONS; i++) {
1738 			/*
1739 			 * Don't wipe out port specific hack (such as
1740 			 * dos partition hack of i386 port).
1741 			 */
1742 			if (lp->d_partitions[i].p_size != 0)
1743 				continue;
1744 
1745 			lp->d_partitions[i].p_size = lp->d_secperunit;
1746 			lp->d_partitions[i].p_offset = 0;
1747 			lp->d_partitions[i].p_fstype = FS_BSDFFS;
1748 		}
1749 
1750 		strncpy(lp->d_packname, "default label",
1751 		    sizeof(lp->d_packname));
1752 
1753 		lp->d_npartitions = MAXPARTITIONS;
1754 		lp->d_checksum = dkcksum(lp);
1755 	}
1756 
1757 	/* In-core label now valid. */
1758 	sc->sc_flags |= VNF_VLABEL;
1759 }
1760 
1761 /*
1762  * Wait interruptibly for an exclusive lock.
1763  *
1764  * XXX
1765  * Several drivers do this; it should be abstracted and made MP-safe.
1766  */
1767 static int
1768 vndlock(struct vnd_softc *sc)
1769 {
1770 	int error;
1771 
1772 	while ((sc->sc_flags & VNF_LOCKED) != 0) {
1773 		sc->sc_flags |= VNF_WANTED;
1774 		if ((error = tsleep(sc, PRIBIO | PCATCH, "vndlck", 0)) != 0)
1775 			return error;
1776 	}
1777 	sc->sc_flags |= VNF_LOCKED;
1778 	return 0;
1779 }
1780 
1781 /*
1782  * Unlock and wake up any waiters.
1783  */
1784 static void
1785 vndunlock(struct vnd_softc *sc)
1786 {
1787 
1788 	sc->sc_flags &= ~VNF_LOCKED;
1789 	if ((sc->sc_flags & VNF_WANTED) != 0) {
1790 		sc->sc_flags &= ~VNF_WANTED;
1791 		wakeup(sc);
1792 	}
1793 }
1794 
1795 #ifdef VND_COMPRESSION
1796 /* compressed file read */
1797 static void
1798 compstrategy(struct buf *bp, off_t bn)
1799 {
1800 	int error;
1801 	int unit = vndunit(bp->b_dev);
1802 	struct vnd_softc *vnd =
1803 	    device_lookup_private(&vnd_cd, unit);
1804 	u_int32_t comp_block;
1805 	struct uio auio;
1806 	char *addr;
1807 	int s;
1808 
1809 	/* set up constants for data move */
1810 	auio.uio_rw = UIO_READ;
1811 	UIO_SETUP_SYSSPACE(&auio);
1812 
1813 	/* read, and transfer the data */
1814 	addr = bp->b_data;
1815 	bp->b_resid = bp->b_bcount;
1816 	s = splbio();
1817 	while (bp->b_resid > 0) {
1818 		unsigned length;
1819 		size_t length_in_buffer;
1820 		u_int32_t offset_in_buffer;
1821 		struct iovec aiov;
1822 
1823 		/* calculate the compressed block number */
1824 		comp_block = bn / (off_t)vnd->sc_comp_blksz;
1825 
1826 		/* check for good block number */
1827 		if (comp_block >= vnd->sc_comp_numoffs) {
1828 			bp->b_error = EINVAL;
1829 			splx(s);
1830 			return;
1831 		}
1832 
1833 		/* read in the compressed block, if not in buffer */
1834 		if (comp_block != vnd->sc_comp_buffblk) {
1835 			length = vnd->sc_comp_offsets[comp_block + 1] -
1836 			    vnd->sc_comp_offsets[comp_block];
1837 			vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1838 			error = vn_rdwr(UIO_READ, vnd->sc_vp, vnd->sc_comp_buff,
1839 			    length, vnd->sc_comp_offsets[comp_block],
1840 			    UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, vnd->sc_cred,
1841 			    NULL, NULL);
1842 			if (error) {
1843 				bp->b_error = error;
1844 				VOP_UNLOCK(vnd->sc_vp, 0);
1845 				splx(s);
1846 				return;
1847 			}
1848 			/* uncompress the buffer */
1849 			vnd->sc_comp_stream.next_in = vnd->sc_comp_buff;
1850 			vnd->sc_comp_stream.avail_in = length;
1851 			vnd->sc_comp_stream.next_out = vnd->sc_comp_decombuf;
1852 			vnd->sc_comp_stream.avail_out = vnd->sc_comp_blksz;
1853 			inflateReset(&vnd->sc_comp_stream);
1854 			error = inflate(&vnd->sc_comp_stream, Z_FINISH);
1855 			if (error != Z_STREAM_END) {
1856 				if (vnd->sc_comp_stream.msg)
1857 					aprint_normal_dev(vnd->sc_dev,
1858 					    "compressed file, %s\n",
1859 					    vnd->sc_comp_stream.msg);
1860 				bp->b_error = EBADMSG;
1861 				VOP_UNLOCK(vnd->sc_vp, 0);
1862 				splx(s);
1863 				return;
1864 			}
1865 			vnd->sc_comp_buffblk = comp_block;
1866 			VOP_UNLOCK(vnd->sc_vp, 0);
1867 		}
1868 
1869 		/* transfer the usable uncompressed data */
1870 		offset_in_buffer = bn % (off_t)vnd->sc_comp_blksz;
1871 		length_in_buffer = vnd->sc_comp_blksz - offset_in_buffer;
1872 		if (length_in_buffer > bp->b_resid)
1873 			length_in_buffer = bp->b_resid;
1874 		auio.uio_iov = &aiov;
1875 		auio.uio_iovcnt = 1;
1876 		aiov.iov_base = addr;
1877 		aiov.iov_len = length_in_buffer;
1878 		auio.uio_resid = aiov.iov_len;
1879 		auio.uio_offset = 0;
1880 		error = uiomove(vnd->sc_comp_decombuf + offset_in_buffer,
1881 		    length_in_buffer, &auio);
1882 		if (error) {
1883 			bp->b_error = error;
1884 			splx(s);
1885 			return;
1886 		}
1887 
1888 		bn += length_in_buffer;
1889 		addr += length_in_buffer;
1890 		bp->b_resid -= length_in_buffer;
1891 	}
1892 	splx(s);
1893 }
1894 
1895 /* compression memory allocation routines */
1896 static void *
1897 vnd_alloc(void *aux, u_int items, u_int siz)
1898 {
1899 	return malloc(items * siz, M_TEMP, M_NOWAIT);
1900 }
1901 
1902 static void
1903 vnd_free(void *aux, void *ptr)
1904 {
1905 	free(ptr, M_TEMP);
1906 }
1907 #endif /* VND_COMPRESSION */
1908 
1909 static void
1910 vnd_set_properties(struct vnd_softc *vnd)
1911 {
1912 	prop_dictionary_t disk_info, odisk_info, geom;
1913 
1914 	disk_info = prop_dictionary_create();
1915 
1916 	geom = prop_dictionary_create();
1917 
1918 	prop_dictionary_set_uint64(geom, "sectors-per-unit",
1919 	    vnd->sc_geom.vng_nsectors * vnd->sc_geom.vng_ntracks *
1920 	    vnd->sc_geom.vng_ncylinders);
1921 
1922 	prop_dictionary_set_uint32(geom, "sector-size",
1923 	    vnd->sc_geom.vng_secsize);
1924 
1925 	prop_dictionary_set_uint16(geom, "sectors-per-track",
1926 	    vnd->sc_geom.vng_nsectors);
1927 
1928 	prop_dictionary_set_uint16(geom, "tracks-per-cylinder",
1929 	    vnd->sc_geom.vng_ntracks);
1930 
1931 	prop_dictionary_set_uint64(geom, "cylinders-per-unit",
1932 	    vnd->sc_geom.vng_ncylinders);
1933 
1934 	prop_dictionary_set(disk_info, "geometry", geom);
1935 	prop_object_release(geom);
1936 
1937 	prop_dictionary_set(device_properties(vnd->sc_dev),
1938 	    "disk-info", disk_info);
1939 
1940 	/*
1941 	 * Don't release disk_info here; we keep a reference to it.
1942 	 * disk_detach() will release it when we go away.
1943 	 */
1944 
1945 	odisk_info = vnd->sc_dkdev.dk_info;
1946 	vnd->sc_dkdev.dk_info = disk_info;
1947 	if (odisk_info)
1948 		prop_object_release(odisk_info);
1949 }
1950 
1951 #ifdef _MODULE
1952 
1953 #include <sys/module.h>
1954 
1955 MODULE(MODULE_CLASS_DRIVER, vnd, NULL);
1956 CFDRIVER_DECL(vnd, DV_DISK, NULL);
1957 
1958 static int
1959 vnd_modcmd(modcmd_t cmd, void *arg)
1960 {
1961 	int bmajor = -1, cmajor = -1,  error = 0;
1962 
1963 	switch (cmd) {
1964 	case MODULE_CMD_INIT:
1965 		error = config_cfdriver_attach(&vnd_cd);
1966 		if (error)
1967 			break;
1968 
1969 		error = config_cfattach_attach(vnd_cd.cd_name, &vnd_ca);
1970 	        if (error) {
1971 			config_cfdriver_detach(&vnd_cd);
1972 			aprint_error("%s: unable to register cfattach\n",
1973 			    vnd_cd.cd_name);
1974 			break;
1975 		}
1976 
1977 		error = devsw_attach("vnd", &vnd_bdevsw, &bmajor,
1978 		    &vnd_cdevsw, &cmajor);
1979 		if (error) {
1980 			config_cfattach_detach(vnd_cd.cd_name, &vnd_ca);
1981 			config_cfdriver_detach(&vnd_cd);
1982 			break;
1983 		}
1984 
1985 		break;
1986 
1987 	case MODULE_CMD_FINI:
1988 		error = config_cfattach_detach(vnd_cd.cd_name, &vnd_ca);
1989 		if (error)
1990 			break;
1991 		config_cfdriver_detach(&vnd_cd);
1992 		devsw_detach(&vnd_bdevsw, &vnd_cdevsw);
1993 		break;
1994 
1995 	case MODULE_CMD_STAT:
1996 		return ENOTTY;
1997 
1998 	default:
1999 		return ENOTTY;
2000 	}
2001 
2002 	return error;
2003 }
2004 
2005 #endif
2006