xref: /netbsd-src/sys/dev/vnd.c (revision f89f6560d453f5e37386cc7938c072d2f528b9fa)
1 /*	$NetBSD: vnd.c,v 1.242 2015/04/06 12:02:08 mlelstv Exp $	*/
2 
3 /*-
4  * Copyright (c) 1996, 1997, 1998, 2008 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Jason R. Thorpe.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29  * POSSIBILITY OF SUCH DAMAGE.
30  */
31 
32 /*
33  * Copyright (c) 1988 University of Utah.
34  * Copyright (c) 1990, 1993
35  *	The Regents of the University of California.  All rights reserved.
36  *
37  * This code is derived from software contributed to Berkeley by
38  * the Systems Programming Group of the University of Utah Computer
39  * Science Department.
40  *
41  * Redistribution and use in source and binary forms, with or without
42  * modification, are permitted provided that the following conditions
43  * are met:
44  * 1. Redistributions of source code must retain the above copyright
45  *    notice, this list of conditions and the following disclaimer.
46  * 2. Redistributions in binary form must reproduce the above copyright
47  *    notice, this list of conditions and the following disclaimer in the
48  *    documentation and/or other materials provided with the distribution.
49  * 3. Neither the name of the University nor the names of its contributors
50  *    may be used to endorse or promote products derived from this software
51  *    without specific prior written permission.
52  *
53  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
54  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
55  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
56  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
57  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
58  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
59  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
60  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
61  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
62  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
63  * SUCH DAMAGE.
64  *
65  * from: Utah $Hdr: vn.c 1.13 94/04/02$
66  *
67  *	@(#)vn.c	8.9 (Berkeley) 5/14/95
68  */
69 
70 /*
71  * Vnode disk driver.
72  *
73  * Block/character interface to a vnode.  Allows one to treat a file
74  * as a disk (e.g. build a filesystem in it, mount it, etc.).
75  *
76  * NOTE 1: If the vnode supports the VOP_BMAP and VOP_STRATEGY operations,
77  * this uses them to avoid distorting the local buffer cache.  If those
78  * block-level operations are not available, this falls back to the regular
79  * read and write calls.  Using these may distort the cache in some cases
80  * but better have the driver working than preventing it to work on file
81  * systems where the block-level operations are not implemented for
82  * whatever reason.
83  *
84  * NOTE 2: There is a security issue involved with this driver.
85  * Once mounted all access to the contents of the "mapped" file via
86  * the special file is controlled by the permissions on the special
87  * file, the protection of the mapped file is ignored (effectively,
88  * by using root credentials in all transactions).
89  *
90  * NOTE 3: Doesn't interact with leases, should it?
91  */
92 
93 #include <sys/cdefs.h>
94 __KERNEL_RCSID(0, "$NetBSD: vnd.c,v 1.242 2015/04/06 12:02:08 mlelstv Exp $");
95 
96 #if defined(_KERNEL_OPT)
97 #include "opt_vnd.h"
98 #include "opt_compat_netbsd.h"
99 #endif
100 
101 #include <sys/param.h>
102 #include <sys/systm.h>
103 #include <sys/namei.h>
104 #include <sys/proc.h>
105 #include <sys/kthread.h>
106 #include <sys/errno.h>
107 #include <sys/buf.h>
108 #include <sys/bufq.h>
109 #include <sys/malloc.h>
110 #include <sys/ioctl.h>
111 #include <sys/disklabel.h>
112 #include <sys/device.h>
113 #include <sys/disk.h>
114 #include <sys/stat.h>
115 #include <sys/mount.h>
116 #include <sys/vnode.h>
117 #include <sys/file.h>
118 #include <sys/uio.h>
119 #include <sys/conf.h>
120 #include <sys/kauth.h>
121 
122 #include <net/zlib.h>
123 
124 #include <miscfs/genfs/genfs.h>
125 #include <miscfs/specfs/specdev.h>
126 
127 #include <dev/dkvar.h>
128 #include <dev/vndvar.h>
129 
130 #if defined(VNDDEBUG) && !defined(DEBUG)
131 #define DEBUG
132 #endif
133 
134 #ifdef DEBUG
135 int dovndcluster = 1;
136 #define VDB_FOLLOW	0x01
137 #define VDB_INIT	0x02
138 #define VDB_IO		0x04
139 #define VDB_LABEL	0x08
140 int vnddebug = 0x00;
141 #endif
142 
143 #define vndunit(x)	DISKUNIT(x)
144 
145 struct vndxfer {
146 	struct buf vx_buf;
147 	struct vnd_softc *vx_vnd;
148 };
149 #define	VND_BUFTOXFER(bp)	((struct vndxfer *)(void *)bp)
150 
151 #define VND_GETXFER(vnd)	pool_get(&(vnd)->sc_vxpool, PR_WAITOK)
152 #define VND_PUTXFER(vnd, vx)	pool_put(&(vnd)->sc_vxpool, (vx))
153 
154 #define VNDLABELDEV(dev) \
155     (MAKEDISKDEV(major((dev)), vndunit((dev)), RAW_PART))
156 
157 #define	VND_MAXPENDING(vnd)	((vnd)->sc_maxactive * 4)
158 
159 /* called by main() at boot time */
160 void	vndattach(int);
161 
162 static void	vndclear(struct vnd_softc *, int);
163 static int	vnddoclear(struct vnd_softc *, int, int, bool);
164 static int	vndsetcred(struct vnd_softc *, kauth_cred_t);
165 static void	vndthrottle(struct vnd_softc *, struct vnode *);
166 static void	vndiodone(struct buf *);
167 #if 0
168 static void	vndshutdown(void);
169 #endif
170 
171 static void	vndgetdefaultlabel(struct vnd_softc *, struct disklabel *);
172 static void	vndgetdisklabel(dev_t, struct vnd_softc *);
173 
174 static int	vndlock(struct vnd_softc *);
175 static void	vndunlock(struct vnd_softc *);
176 #ifdef VND_COMPRESSION
177 static void	compstrategy(struct buf *, off_t);
178 static void	*vnd_alloc(void *, u_int, u_int);
179 static void	vnd_free(void *, void *);
180 #endif /* VND_COMPRESSION */
181 
182 static void	vndthread(void *);
183 static bool	vnode_has_op(const struct vnode *, int);
184 static void	handle_with_rdwr(struct vnd_softc *, const struct buf *,
185 		    struct buf *);
186 static void	handle_with_strategy(struct vnd_softc *, const struct buf *,
187 		    struct buf *);
188 static void	vnd_set_geometry(struct vnd_softc *);
189 
190 static dev_type_open(vndopen);
191 static dev_type_close(vndclose);
192 static dev_type_read(vndread);
193 static dev_type_write(vndwrite);
194 static dev_type_ioctl(vndioctl);
195 static dev_type_strategy(vndstrategy);
196 static dev_type_dump(vnddump);
197 static dev_type_size(vndsize);
198 
199 const struct bdevsw vnd_bdevsw = {
200 	.d_open = vndopen,
201 	.d_close = vndclose,
202 	.d_strategy = vndstrategy,
203 	.d_ioctl = vndioctl,
204 	.d_dump = vnddump,
205 	.d_psize = vndsize,
206 	.d_discard = nodiscard,
207 	.d_flag = D_DISK
208 };
209 
210 const struct cdevsw vnd_cdevsw = {
211 	.d_open = vndopen,
212 	.d_close = vndclose,
213 	.d_read = vndread,
214 	.d_write = vndwrite,
215 	.d_ioctl = vndioctl,
216 	.d_stop = nostop,
217 	.d_tty = notty,
218 	.d_poll = nopoll,
219 	.d_mmap = nommap,
220 	.d_kqfilter = nokqfilter,
221 	.d_discard = nodiscard,
222 	.d_flag = D_DISK
223 };
224 
225 static int	vnd_match(device_t, cfdata_t, void *);
226 static void	vnd_attach(device_t, device_t, void *);
227 static int	vnd_detach(device_t, int);
228 
229 CFATTACH_DECL3_NEW(vnd, sizeof(struct vnd_softc),
230     vnd_match, vnd_attach, vnd_detach, NULL, NULL, NULL, DVF_DETACH_SHUTDOWN);
231 extern struct cfdriver vnd_cd;
232 
233 static struct vnd_softc	*vnd_spawn(int);
234 int	vnd_destroy(device_t);
235 
236 static struct	dkdriver vnddkdriver = { vndstrategy, minphys };
237 
238 void
239 vndattach(int num)
240 {
241 	int error;
242 
243 	error = config_cfattach_attach(vnd_cd.cd_name, &vnd_ca);
244 	if (error)
245 		aprint_error("%s: unable to register cfattach\n",
246 		    vnd_cd.cd_name);
247 }
248 
249 static int
250 vnd_match(device_t self, cfdata_t cfdata, void *aux)
251 {
252 
253 	return 1;
254 }
255 
256 static void
257 vnd_attach(device_t parent, device_t self, void *aux)
258 {
259 	struct vnd_softc *sc = device_private(self);
260 
261 	sc->sc_dev = self;
262 	sc->sc_comp_offsets = NULL;
263 	sc->sc_comp_buff = NULL;
264 	sc->sc_comp_decombuf = NULL;
265 	bufq_alloc(&sc->sc_tab, "disksort", BUFQ_SORT_RAWBLOCK);
266 	disk_init(&sc->sc_dkdev, device_xname(self), &vnddkdriver);
267 	if (!pmf_device_register(self, NULL, NULL))
268 		aprint_error_dev(self, "couldn't establish power handler\n");
269 }
270 
271 static int
272 vnd_detach(device_t self, int flags)
273 {
274 	int error;
275 	struct vnd_softc *sc = device_private(self);
276 
277 	if (sc->sc_flags & VNF_INITED) {
278 		error = vnddoclear(sc, 0, -1, (flags & DETACH_FORCE) != 0);
279 		if (error != 0)
280 			return error;
281 	}
282 
283 	pmf_device_deregister(self);
284 	bufq_free(sc->sc_tab);
285 	disk_destroy(&sc->sc_dkdev);
286 
287 	return 0;
288 }
289 
290 static struct vnd_softc *
291 vnd_spawn(int unit)
292 {
293 	cfdata_t cf;
294 
295 	cf = malloc(sizeof(*cf), M_DEVBUF, M_WAITOK);
296 	cf->cf_name = vnd_cd.cd_name;
297 	cf->cf_atname = vnd_cd.cd_name;
298 	cf->cf_unit = unit;
299 	cf->cf_fstate = FSTATE_STAR;
300 
301 	return device_private(config_attach_pseudo(cf));
302 }
303 
304 int
305 vnd_destroy(device_t dev)
306 {
307 	int error;
308 	cfdata_t cf;
309 
310 	cf = device_cfdata(dev);
311 	error = config_detach(dev, DETACH_QUIET);
312 	if (error)
313 		return error;
314 	free(cf, M_DEVBUF);
315 	return 0;
316 }
317 
318 static int
319 vndopen(dev_t dev, int flags, int mode, struct lwp *l)
320 {
321 	int unit = vndunit(dev);
322 	struct vnd_softc *sc;
323 	int error = 0, part, pmask;
324 	struct disklabel *lp;
325 
326 #ifdef DEBUG
327 	if (vnddebug & VDB_FOLLOW)
328 		printf("vndopen(0x%"PRIx64", 0x%x, 0x%x, %p)\n", dev, flags, mode, l);
329 #endif
330 	sc = device_lookup_private(&vnd_cd, unit);
331 	if (sc == NULL) {
332 		sc = vnd_spawn(unit);
333 		if (sc == NULL)
334 			return ENOMEM;
335 
336 		/* compatibility, keep disklabel after close */
337 		sc->sc_flags = VNF_KLABEL;
338 	}
339 
340 	if ((error = vndlock(sc)) != 0)
341 		return error;
342 
343 	if ((sc->sc_flags & VNF_CLEARING) != 0) {
344 		error = ENXIO;
345 		goto done;
346 	}
347 
348 	lp = sc->sc_dkdev.dk_label;
349 
350 	part = DISKPART(dev);
351 	pmask = (1 << part);
352 
353 	if (sc->sc_dkdev.dk_nwedges != 0 && part != RAW_PART) {
354 		error = EBUSY;
355 		goto done;
356 	}
357 
358 	if (sc->sc_flags & VNF_INITED) {
359 		if ((sc->sc_dkdev.dk_openmask & ~(1<<RAW_PART)) != 0) {
360 			/*
361 			 * If any non-raw partition is open, but the disk
362 			 * has been invalidated, disallow further opens.
363 			 */
364 			if ((sc->sc_flags & VNF_VLABEL) == 0) {
365 				error = EIO;
366 				goto done;
367 			}
368 		} else {
369 			/*
370 			 * Load the partition info if not already loaded.
371 			 */
372 			if ((sc->sc_flags & VNF_VLABEL) == 0) {
373 				sc->sc_flags |= VNF_VLABEL;
374 				vndgetdisklabel(dev, sc);
375 			}
376 		}
377 	}
378 
379 	/* Check that the partitions exists. */
380 	if (part != RAW_PART) {
381 		if (((sc->sc_flags & VNF_INITED) == 0) ||
382 		    ((part >= lp->d_npartitions) ||
383 		     (lp->d_partitions[part].p_fstype == FS_UNUSED))) {
384 			error = ENXIO;
385 			goto done;
386 		}
387 	}
388 
389 	/* Prevent our unit from being unconfigured while open. */
390 	switch (mode) {
391 	case S_IFCHR:
392 		sc->sc_dkdev.dk_copenmask |= pmask;
393 		break;
394 
395 	case S_IFBLK:
396 		sc->sc_dkdev.dk_bopenmask |= pmask;
397 		break;
398 	}
399 	sc->sc_dkdev.dk_openmask =
400 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
401 
402  done:
403 	vndunlock(sc);
404 	return error;
405 }
406 
407 static int
408 vndclose(dev_t dev, int flags, int mode, struct lwp *l)
409 {
410 	int unit = vndunit(dev);
411 	struct vnd_softc *sc;
412 	int error = 0, part;
413 
414 #ifdef DEBUG
415 	if (vnddebug & VDB_FOLLOW)
416 		printf("vndclose(0x%"PRIx64", 0x%x, 0x%x, %p)\n", dev, flags, mode, l);
417 #endif
418 	sc = device_lookup_private(&vnd_cd, unit);
419 	if (sc == NULL)
420 		return ENXIO;
421 
422 	if ((error = vndlock(sc)) != 0)
423 		return error;
424 
425 	part = DISKPART(dev);
426 
427 	/* ...that much closer to allowing unconfiguration... */
428 	switch (mode) {
429 	case S_IFCHR:
430 		sc->sc_dkdev.dk_copenmask &= ~(1 << part);
431 		break;
432 
433 	case S_IFBLK:
434 		sc->sc_dkdev.dk_bopenmask &= ~(1 << part);
435 		break;
436 	}
437 	sc->sc_dkdev.dk_openmask =
438 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
439 
440 	/* are we last opener ? */
441 	if (sc->sc_dkdev.dk_openmask == 0) {
442 		if ((sc->sc_flags & VNF_KLABEL) == 0)
443 			sc->sc_flags &= ~VNF_VLABEL;
444 	}
445 
446 	vndunlock(sc);
447 
448 	if ((sc->sc_flags & VNF_INITED) == 0) {
449 		if ((error = vnd_destroy(sc->sc_dev)) != 0) {
450 			aprint_error_dev(sc->sc_dev,
451 			    "unable to detach instance\n");
452 			return error;
453 		}
454 	}
455 
456 	return 0;
457 }
458 
459 /*
460  * Queue the request, and wakeup the kernel thread to handle it.
461  */
462 static void
463 vndstrategy(struct buf *bp)
464 {
465 	int unit = vndunit(bp->b_dev);
466 	struct vnd_softc *vnd =
467 	    device_lookup_private(&vnd_cd, unit);
468 	struct disklabel *lp;
469 	daddr_t blkno;
470 	int s = splbio();
471 
472 	if (vnd == NULL) {
473 		bp->b_error = ENXIO;
474 		goto done;
475 	}
476 	lp = vnd->sc_dkdev.dk_label;
477 
478 	if ((vnd->sc_flags & VNF_INITED) == 0) {
479 		bp->b_error = ENXIO;
480 		goto done;
481 	}
482 
483 	/*
484 	 * The transfer must be a whole number of blocks.
485 	 */
486 	if ((bp->b_bcount % lp->d_secsize) != 0) {
487 		bp->b_error = EINVAL;
488 		goto done;
489 	}
490 
491 	/*
492 	 * check if we're read-only.
493 	 */
494 	if ((vnd->sc_flags & VNF_READONLY) && !(bp->b_flags & B_READ)) {
495 		bp->b_error = EACCES;
496 		goto done;
497 	}
498 
499 	/* If it's a nil transfer, wake up the top half now. */
500 	if (bp->b_bcount == 0) {
501 		goto done;
502 	}
503 
504 	/*
505 	 * Do bounds checking and adjust transfer.  If there's an error,
506 	 * the bounds check will flag that for us.
507 	 */
508 	if (DISKPART(bp->b_dev) == RAW_PART) {
509 		if (bounds_check_with_mediasize(bp, DEV_BSIZE,
510 		    vnd->sc_size) <= 0)
511 			goto done;
512 	} else {
513 		if (bounds_check_with_label(&vnd->sc_dkdev,
514 		    bp, vnd->sc_flags & (VNF_WLABEL|VNF_LABELLING)) <= 0)
515 			goto done;
516 	}
517 
518 	/*
519 	 * Put the block number in terms of the logical blocksize
520 	 * of the "device".
521 	 */
522 
523 	blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
524 
525 	/*
526 	 * Translate the partition-relative block number to an absolute.
527 	 */
528 	if (DISKPART(bp->b_dev) != RAW_PART) {
529 		struct partition *pp;
530 
531 		pp = &vnd->sc_dkdev.dk_label->d_partitions[
532 		    DISKPART(bp->b_dev)];
533 		blkno += pp->p_offset;
534 	}
535 	bp->b_rawblkno = blkno;
536 
537 #ifdef DEBUG
538 	if (vnddebug & VDB_FOLLOW)
539 		printf("vndstrategy(%p): unit %d\n", bp, unit);
540 #endif
541 	if ((vnd->sc_flags & VNF_USE_VN_RDWR)) {
542 		KASSERT(vnd->sc_pending >= 0 &&
543 		    vnd->sc_pending <= VND_MAXPENDING(vnd));
544 		while (vnd->sc_pending == VND_MAXPENDING(vnd))
545 			tsleep(&vnd->sc_pending, PRIBIO, "vndpc", 0);
546 		vnd->sc_pending++;
547 	}
548 	bufq_put(vnd->sc_tab, bp);
549 	wakeup(&vnd->sc_tab);
550 	splx(s);
551 	return;
552 
553 done:
554 	bp->b_resid = bp->b_bcount;
555 	biodone(bp);
556 	splx(s);
557 }
558 
559 static bool
560 vnode_has_strategy(struct vnd_softc *vnd)
561 {
562 	return vnode_has_op(vnd->sc_vp, VOFFSET(vop_bmap)) &&
563 	    vnode_has_op(vnd->sc_vp, VOFFSET(vop_strategy));
564 }
565 
566 /* XXX this function needs a reliable check to detect
567  * sparse files. Otherwise, bmap/strategy may be used
568  * and fail on non-allocated blocks. VOP_READ/VOP_WRITE
569  * works on sparse files.
570  */
571 #if notyet
572 static bool
573 vnode_strategy_probe(struct vnd_softc *vnd)
574 {
575 	int error;
576 	daddr_t nbn;
577 
578 	if (!vnode_has_strategy(vnd))
579 		return false;
580 
581 	/* Convert the first logical block number to its
582 	 * physical block number.
583 	 */
584 	error = 0;
585 	vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
586 	error = VOP_BMAP(vnd->sc_vp, 0, NULL, &nbn, NULL);
587 	VOP_UNLOCK(vnd->sc_vp);
588 
589 	/* Test if that worked. */
590 	if (error == 0 && (long)nbn == -1)
591 		return false;
592 
593 	return true;
594 }
595 #endif
596 
597 static void
598 vndthread(void *arg)
599 {
600 	struct vnd_softc *vnd = arg;
601 	int s;
602 
603 	/* Determine whether we can *use* VOP_BMAP and VOP_STRATEGY to
604 	 * directly access the backing vnode.  If we can, use these two
605 	 * operations to avoid messing with the local buffer cache.
606 	 * Otherwise fall back to regular VOP_READ/VOP_WRITE operations
607 	 * which are guaranteed to work with any file system. */
608 	if ((vnd->sc_flags & VNF_USE_VN_RDWR) == 0 &&
609 	    ! vnode_has_strategy(vnd))
610 		vnd->sc_flags |= VNF_USE_VN_RDWR;
611 
612 #ifdef DEBUG
613 	if (vnddebug & VDB_INIT)
614 		printf("vndthread: vp %p, %s\n", vnd->sc_vp,
615 		    (vnd->sc_flags & VNF_USE_VN_RDWR) == 0 ?
616 		    "using bmap/strategy operations" :
617 		    "using read/write operations");
618 #endif
619 
620 	s = splbio();
621 	vnd->sc_flags |= VNF_KTHREAD;
622 	wakeup(&vnd->sc_kthread);
623 
624 	/*
625 	 * Dequeue requests and serve them depending on the available
626 	 * vnode operations.
627 	 */
628 	while ((vnd->sc_flags & VNF_VUNCONF) == 0) {
629 		struct vndxfer *vnx;
630 		struct buf *obp;
631 		struct buf *bp;
632 
633 		obp = bufq_get(vnd->sc_tab);
634 		if (obp == NULL) {
635 			tsleep(&vnd->sc_tab, PRIBIO, "vndbp", 0);
636 			continue;
637 		};
638 		if ((vnd->sc_flags & VNF_USE_VN_RDWR)) {
639 			KASSERT(vnd->sc_pending > 0 &&
640 			    vnd->sc_pending <= VND_MAXPENDING(vnd));
641 			if (vnd->sc_pending-- == VND_MAXPENDING(vnd))
642 				wakeup(&vnd->sc_pending);
643 		}
644 		splx(s);
645 #ifdef DEBUG
646 		if (vnddebug & VDB_FOLLOW)
647 			printf("vndthread(%p)\n", obp);
648 #endif
649 
650 		if (vnd->sc_vp->v_mount == NULL) {
651 			obp->b_error = ENXIO;
652 			goto done;
653 		}
654 #ifdef VND_COMPRESSION
655 		/* handle a compressed read */
656 		if ((obp->b_flags & B_READ) != 0 && (vnd->sc_flags & VNF_COMP)) {
657 			off_t bn;
658 
659 			/* Convert to a byte offset within the file. */
660 			bn = obp->b_rawblkno *
661 			    vnd->sc_dkdev.dk_label->d_secsize;
662 
663 			compstrategy(obp, bn);
664 			goto done;
665 		}
666 #endif /* VND_COMPRESSION */
667 
668 		/*
669 		 * Allocate a header for this transfer and link it to the
670 		 * buffer
671 		 */
672 		s = splbio();
673 		vnx = VND_GETXFER(vnd);
674 		splx(s);
675 		vnx->vx_vnd = vnd;
676 
677 		s = splbio();
678 		while (vnd->sc_active >= vnd->sc_maxactive) {
679 			tsleep(&vnd->sc_tab, PRIBIO, "vndac", 0);
680 		}
681 		vnd->sc_active++;
682 		splx(s);
683 
684 		/* Instrumentation. */
685 		disk_busy(&vnd->sc_dkdev);
686 
687 		bp = &vnx->vx_buf;
688 		buf_init(bp);
689 		bp->b_flags = (obp->b_flags & B_READ);
690 		bp->b_oflags = obp->b_oflags;
691 		bp->b_cflags = obp->b_cflags;
692 		bp->b_iodone = vndiodone;
693 		bp->b_private = obp;
694 		bp->b_vp = vnd->sc_vp;
695 		bp->b_objlock = bp->b_vp->v_interlock;
696 		bp->b_data = obp->b_data;
697 		bp->b_bcount = obp->b_bcount;
698 		BIO_COPYPRIO(bp, obp);
699 
700 		/* Handle the request using the appropriate operations. */
701 		if ((vnd->sc_flags & VNF_USE_VN_RDWR) == 0)
702 			handle_with_strategy(vnd, obp, bp);
703 		else
704 			handle_with_rdwr(vnd, obp, bp);
705 
706 		s = splbio();
707 		continue;
708 
709 done:
710 		biodone(obp);
711 		s = splbio();
712 	}
713 
714 	vnd->sc_flags &= (~VNF_KTHREAD | VNF_VUNCONF);
715 	wakeup(&vnd->sc_kthread);
716 	splx(s);
717 	kthread_exit(0);
718 }
719 
720 /*
721  * Checks if the given vnode supports the requested operation.
722  * The operation is specified the offset returned by VOFFSET.
723  *
724  * XXX The test below used to determine this is quite fragile
725  * because it relies on the file system to use genfs to specify
726  * unimplemented operations.  There might be another way to do
727  * it more cleanly.
728  */
729 static bool
730 vnode_has_op(const struct vnode *vp, int opoffset)
731 {
732 	int (*defaultp)(void *);
733 	int (*opp)(void *);
734 
735 	defaultp = vp->v_op[VOFFSET(vop_default)];
736 	opp = vp->v_op[opoffset];
737 
738 	return opp != defaultp && opp != genfs_eopnotsupp &&
739 	    opp != genfs_badop && opp != genfs_nullop;
740 }
741 
742 /*
743  * Handes the read/write request given in 'bp' using the vnode's VOP_READ
744  * and VOP_WRITE operations.
745  *
746  * 'obp' is a pointer to the original request fed to the vnd device.
747  */
748 static void
749 handle_with_rdwr(struct vnd_softc *vnd, const struct buf *obp, struct buf *bp)
750 {
751 	bool doread;
752 	off_t offset;
753 	size_t len, resid;
754 	struct vnode *vp;
755 
756 	doread = bp->b_flags & B_READ;
757 	offset = obp->b_rawblkno * vnd->sc_dkdev.dk_label->d_secsize;
758 	len = bp->b_bcount;
759 	vp = vnd->sc_vp;
760 
761 #if defined(DEBUG)
762 	if (vnddebug & VDB_IO)
763 		printf("vnd (rdwr): vp %p, %s, rawblkno 0x%" PRIx64
764 		    ", secsize %d, offset %" PRIu64
765 		    ", bcount %d\n",
766 		    vp, doread ? "read" : "write", obp->b_rawblkno,
767 		    vnd->sc_dkdev.dk_label->d_secsize, offset,
768 		    bp->b_bcount);
769 #endif
770 
771 	/* Issue the read or write operation. */
772 	bp->b_error =
773 	    vn_rdwr(doread ? UIO_READ : UIO_WRITE,
774 	    vp, bp->b_data, len, offset, UIO_SYSSPACE,
775 	    IO_ADV_ENCODE(POSIX_FADV_NOREUSE), vnd->sc_cred, &resid, NULL);
776 	bp->b_resid = resid;
777 
778 	mutex_enter(vp->v_interlock);
779 	(void) VOP_PUTPAGES(vp, 0, 0,
780 	    PGO_ALLPAGES | PGO_CLEANIT | PGO_FREE | PGO_SYNCIO);
781 
782 	/* We need to increase the number of outputs on the vnode if
783 	 * there was any write to it. */
784 	if (!doread) {
785 		mutex_enter(vp->v_interlock);
786 		vp->v_numoutput++;
787 		mutex_exit(vp->v_interlock);
788 	}
789 
790 	biodone(bp);
791 }
792 
793 /*
794  * Handes the read/write request given in 'bp' using the vnode's VOP_BMAP
795  * and VOP_STRATEGY operations.
796  *
797  * 'obp' is a pointer to the original request fed to the vnd device.
798  */
799 static void
800 handle_with_strategy(struct vnd_softc *vnd, const struct buf *obp,
801     struct buf *bp)
802 {
803 	int bsize, error, flags, skipped;
804 	size_t resid, sz;
805 	off_t bn, offset;
806 	struct vnode *vp;
807 	struct buf *nbp = NULL;
808 
809 	flags = obp->b_flags;
810 
811 
812 	/* convert to a byte offset within the file. */
813 	bn = obp->b_rawblkno * vnd->sc_dkdev.dk_label->d_secsize;
814 
815 	bsize = vnd->sc_vp->v_mount->mnt_stat.f_iosize;
816 	skipped = 0;
817 
818 	/*
819 	 * Break the request into bsize pieces and feed them
820 	 * sequentially using VOP_BMAP/VOP_STRATEGY.
821 	 * We do it this way to keep from flooding NFS servers if we
822 	 * are connected to an NFS file.  This places the burden on
823 	 * the client rather than the server.
824 	 */
825 	error = 0;
826 	bp->b_resid = bp->b_bcount;
827 	for (offset = 0, resid = bp->b_resid; /* true */;
828 	    resid -= sz, offset += sz) {
829 		daddr_t nbn;
830 		int off, nra;
831 
832 		nra = 0;
833 		vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
834 		error = VOP_BMAP(vnd->sc_vp, bn / bsize, &vp, &nbn, &nra);
835 		VOP_UNLOCK(vnd->sc_vp);
836 
837 		if (error == 0 && (long)nbn == -1)
838 			error = EIO;
839 
840 		/*
841 		 * If there was an error or a hole in the file...punt.
842 		 * Note that we may have to wait for any operations
843 		 * that we have already fired off before releasing
844 		 * the buffer.
845 		 *
846 		 * XXX we could deal with holes here but it would be
847 		 * a hassle (in the write case).
848 		 */
849 		if (error) {
850 			skipped += resid;
851 			break;
852 		}
853 
854 #ifdef DEBUG
855 		if (!dovndcluster)
856 			nra = 0;
857 #endif
858 
859 		off = bn % bsize;
860 		sz = MIN(((off_t)1 + nra) * bsize - off, resid);
861 #ifdef	DEBUG
862 		if (vnddebug & VDB_IO)
863 			printf("vndstrategy: vp %p/%p bn 0x%qx/0x%" PRIx64
864 			    " sz 0x%zx\n", vnd->sc_vp, vp, (long long)bn,
865 			    nbn, sz);
866 #endif
867 
868 		nbp = getiobuf(vp, true);
869 		nestiobuf_setup(bp, nbp, offset, sz);
870 		nbp->b_blkno = nbn + btodb(off);
871 
872 #if 0 /* XXX #ifdef DEBUG */
873 		if (vnddebug & VDB_IO)
874 			printf("vndstart(%ld): bp %p vp %p blkno "
875 			    "0x%" PRIx64 " flags %x addr %p cnt 0x%x\n",
876 			    (long) (vnd-vnd_softc), &nbp->vb_buf,
877 			    nbp->vb_buf.b_vp, nbp->vb_buf.b_blkno,
878 			    nbp->vb_buf.b_flags, nbp->vb_buf.b_data,
879 			    nbp->vb_buf.b_bcount);
880 #endif
881 		if (resid == sz) {
882 			break;
883 		}
884 		VOP_STRATEGY(vp, nbp);
885 		bn += sz;
886 	}
887 	if (!(flags & B_READ)) {
888 		struct vnode *w_vp;
889 		/*
890 		 * this is the last nested buf, account for
891 		 * the parent buf write too.
892 		 * This has to be done last, so that
893 		 * fsync won't wait for this write which
894 		 * has no chance to complete before all nested bufs
895 		 * have been queued. But it has to be done
896 		 * before the last VOP_STRATEGY()
897 		 * or the call to nestiobuf_done().
898 		 */
899 		w_vp = bp->b_vp;
900 		mutex_enter(w_vp->v_interlock);
901 		w_vp->v_numoutput++;
902 		mutex_exit(w_vp->v_interlock);
903 	}
904 	KASSERT(skipped != 0 || nbp != NULL);
905 	if (skipped)
906 		nestiobuf_done(bp, skipped, error);
907 	else
908 		VOP_STRATEGY(vp, nbp);
909 }
910 
911 static void
912 vndiodone(struct buf *bp)
913 {
914 	struct vndxfer *vnx = VND_BUFTOXFER(bp);
915 	struct vnd_softc *vnd = vnx->vx_vnd;
916 	struct buf *obp = bp->b_private;
917 	int s = splbio();
918 
919 	KASSERT(&vnx->vx_buf == bp);
920 	KASSERT(vnd->sc_active > 0);
921 #ifdef DEBUG
922 	if (vnddebug & VDB_IO) {
923 		printf("vndiodone1: bp %p iodone: error %d\n",
924 		    bp, bp->b_error);
925 	}
926 #endif
927 	disk_unbusy(&vnd->sc_dkdev, bp->b_bcount - bp->b_resid,
928 	    (bp->b_flags & B_READ));
929 	vnd->sc_active--;
930 	if (vnd->sc_active == 0) {
931 		wakeup(&vnd->sc_tab);
932 	}
933 	splx(s);
934 	obp->b_error = bp->b_error;
935 	obp->b_resid = bp->b_resid;
936 	buf_destroy(bp);
937 	VND_PUTXFER(vnd, vnx);
938 	biodone(obp);
939 }
940 
941 /* ARGSUSED */
942 static int
943 vndread(dev_t dev, struct uio *uio, int flags)
944 {
945 	int unit = vndunit(dev);
946 	struct vnd_softc *sc;
947 
948 #ifdef DEBUG
949 	if (vnddebug & VDB_FOLLOW)
950 		printf("vndread(0x%"PRIx64", %p)\n", dev, uio);
951 #endif
952 
953 	sc = device_lookup_private(&vnd_cd, unit);
954 	if (sc == NULL)
955 		return ENXIO;
956 
957 	if ((sc->sc_flags & VNF_INITED) == 0)
958 		return ENXIO;
959 
960 	return physio(vndstrategy, NULL, dev, B_READ, minphys, uio);
961 }
962 
963 /* ARGSUSED */
964 static int
965 vndwrite(dev_t dev, struct uio *uio, int flags)
966 {
967 	int unit = vndunit(dev);
968 	struct vnd_softc *sc;
969 
970 #ifdef DEBUG
971 	if (vnddebug & VDB_FOLLOW)
972 		printf("vndwrite(0x%"PRIx64", %p)\n", dev, uio);
973 #endif
974 
975 	sc = device_lookup_private(&vnd_cd, unit);
976 	if (sc == NULL)
977 		return ENXIO;
978 
979 	if ((sc->sc_flags & VNF_INITED) == 0)
980 		return ENXIO;
981 
982 	return physio(vndstrategy, NULL, dev, B_WRITE, minphys, uio);
983 }
984 
985 static int
986 vnd_cget(struct lwp *l, int unit, int *un, struct vattr *va)
987 {
988 	int error;
989 	struct vnd_softc *vnd;
990 
991 	if (*un == -1)
992 		*un = unit;
993 	if (*un < 0)
994 		return EINVAL;
995 
996 	vnd = device_lookup_private(&vnd_cd, *un);
997 	if (vnd == NULL)
998 		return -1;
999 
1000 	if ((vnd->sc_flags & VNF_INITED) == 0)
1001 		return -1;
1002 
1003 	vn_lock(vnd->sc_vp, LK_SHARED | LK_RETRY);
1004 	error = VOP_GETATTR(vnd->sc_vp, va, l->l_cred);
1005 	VOP_UNLOCK(vnd->sc_vp);
1006 	return error;
1007 }
1008 
1009 static int
1010 vnddoclear(struct vnd_softc *vnd, int pmask, int minor, bool force)
1011 {
1012 	int error;
1013 
1014 	if ((error = vndlock(vnd)) != 0)
1015 		return error;
1016 
1017 	/*
1018 	 * Don't unconfigure if any other partitions are open
1019 	 * or if both the character and block flavors of this
1020 	 * partition are open.
1021 	 */
1022 	if (DK_BUSY(vnd, pmask) && !force) {
1023 		vndunlock(vnd);
1024 		return EBUSY;
1025 	}
1026 
1027 	/* Delete all of our wedges */
1028 	dkwedge_delall(&vnd->sc_dkdev);
1029 
1030 	/*
1031 	 * XXX vndclear() might call vndclose() implicitly;
1032 	 * release lock to avoid recursion
1033 	 *
1034 	 * Set VNF_CLEARING to prevent vndopen() from
1035 	 * sneaking in after we vndunlock().
1036 	 */
1037 	vnd->sc_flags |= VNF_CLEARING;
1038 	vndunlock(vnd);
1039 	vndclear(vnd, minor);
1040 #ifdef DEBUG
1041 	if (vnddebug & VDB_INIT)
1042 		printf("vndioctl: CLRed\n");
1043 #endif
1044 
1045 	/* Destroy the xfer and buffer pools. */
1046 	pool_destroy(&vnd->sc_vxpool);
1047 
1048 	/* Detach the disk. */
1049 	disk_detach(&vnd->sc_dkdev);
1050 
1051 	return 0;
1052 }
1053 
1054 /* ARGSUSED */
1055 static int
1056 vndioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l)
1057 {
1058 	bool force;
1059 	int unit = vndunit(dev);
1060 	struct vnd_softc *vnd;
1061 	struct vnd_ioctl *vio;
1062 	struct vattr vattr;
1063 	struct pathbuf *pb;
1064 	struct nameidata nd;
1065 	int error, part, pmask;
1066 	uint64_t geomsize;
1067 	int fflags;
1068 #ifdef __HAVE_OLD_DISKLABEL
1069 	struct disklabel newlabel;
1070 #endif
1071 
1072 #ifdef DEBUG
1073 	if (vnddebug & VDB_FOLLOW)
1074 		printf("vndioctl(0x%"PRIx64", 0x%lx, %p, 0x%x, %p): unit %d\n",
1075 		    dev, cmd, data, flag, l->l_proc, unit);
1076 #endif
1077 	vnd = device_lookup_private(&vnd_cd, unit);
1078 	if (vnd == NULL &&
1079 #ifdef COMPAT_30
1080 	    cmd != VNDIOCGET30 &&
1081 #endif
1082 #ifdef COMPAT_50
1083 	    cmd != VNDIOCGET50 &&
1084 #endif
1085 	    cmd != VNDIOCGET)
1086 		return ENXIO;
1087 	vio = (struct vnd_ioctl *)data;
1088 
1089 	/* Must be open for writes for these commands... */
1090 	switch (cmd) {
1091 	case VNDIOCSET:
1092 	case VNDIOCCLR:
1093 #ifdef COMPAT_50
1094 	case VNDIOCSET50:
1095 	case VNDIOCCLR50:
1096 #endif
1097 	case DIOCSDINFO:
1098 	case DIOCWDINFO:
1099 #ifdef __HAVE_OLD_DISKLABEL
1100 	case ODIOCSDINFO:
1101 	case ODIOCWDINFO:
1102 #endif
1103 	case DIOCKLABEL:
1104 	case DIOCWLABEL:
1105 		if ((flag & FWRITE) == 0)
1106 			return EBADF;
1107 	}
1108 
1109 	/* Must be initialized for these... */
1110 	switch (cmd) {
1111 	case VNDIOCCLR:
1112 #ifdef VNDIOCCLR50
1113 	case VNDIOCCLR50:
1114 #endif
1115 	case DIOCGDINFO:
1116 	case DIOCSDINFO:
1117 	case DIOCWDINFO:
1118 	case DIOCGPART:
1119 	case DIOCKLABEL:
1120 	case DIOCWLABEL:
1121 	case DIOCGDEFLABEL:
1122 	case DIOCCACHESYNC:
1123 #ifdef __HAVE_OLD_DISKLABEL
1124 	case ODIOCGDINFO:
1125 	case ODIOCSDINFO:
1126 	case ODIOCWDINFO:
1127 	case ODIOCGDEFLABEL:
1128 #endif
1129 		if ((vnd->sc_flags & VNF_INITED) == 0)
1130 			return ENXIO;
1131 	}
1132 
1133 	error = disk_ioctl(&vnd->sc_dkdev, dev, cmd, data, flag, l);
1134 	if (error != EPASSTHROUGH)
1135 		return error;
1136 
1137 
1138 	switch (cmd) {
1139 #ifdef VNDIOCSET50
1140 	case VNDIOCSET50:
1141 #endif
1142 	case VNDIOCSET:
1143 		if (vnd->sc_flags & VNF_INITED)
1144 			return EBUSY;
1145 
1146 		if ((error = vndlock(vnd)) != 0)
1147 			return error;
1148 
1149 		fflags = FREAD;
1150 		if ((vio->vnd_flags & VNDIOF_READONLY) == 0)
1151 			fflags |= FWRITE;
1152 		error = pathbuf_copyin(vio->vnd_file, &pb);
1153 		if (error) {
1154 			goto unlock_and_exit;
1155 		}
1156 		NDINIT(&nd, LOOKUP, FOLLOW, pb);
1157 		if ((error = vn_open(&nd, fflags, 0)) != 0) {
1158 			pathbuf_destroy(pb);
1159 			goto unlock_and_exit;
1160 		}
1161 		KASSERT(l);
1162 		error = VOP_GETATTR(nd.ni_vp, &vattr, l->l_cred);
1163 		if (!error && nd.ni_vp->v_type != VREG)
1164 			error = EOPNOTSUPP;
1165 		if (!error && vattr.va_bytes < vattr.va_size)
1166 			/* File is definitely sparse, use vn_rdwr() */
1167 			vnd->sc_flags |= VNF_USE_VN_RDWR;
1168 		if (error) {
1169 			VOP_UNLOCK(nd.ni_vp);
1170 			goto close_and_exit;
1171 		}
1172 
1173 		/* If using a compressed file, initialize its info */
1174 		/* (or abort with an error if kernel has no compression) */
1175 		if (vio->vnd_flags & VNF_COMP) {
1176 #ifdef VND_COMPRESSION
1177 			struct vnd_comp_header *ch;
1178 			int i;
1179 			u_int32_t comp_size;
1180 			u_int32_t comp_maxsize;
1181 
1182 			/* allocate space for compresed file header */
1183 			ch = malloc(sizeof(struct vnd_comp_header),
1184 			M_TEMP, M_WAITOK);
1185 
1186 			/* read compressed file header */
1187 			error = vn_rdwr(UIO_READ, nd.ni_vp, (void *)ch,
1188 			  sizeof(struct vnd_comp_header), 0, UIO_SYSSPACE,
1189 			  IO_UNIT|IO_NODELOCKED, l->l_cred, NULL, NULL);
1190 			if (error) {
1191 				free(ch, M_TEMP);
1192 				VOP_UNLOCK(nd.ni_vp);
1193 				goto close_and_exit;
1194 			}
1195 
1196 			/* save some header info */
1197 			vnd->sc_comp_blksz = ntohl(ch->block_size);
1198 			/* note last offset is the file byte size */
1199 			vnd->sc_comp_numoffs = ntohl(ch->num_blocks)+1;
1200 			free(ch, M_TEMP);
1201 			if (vnd->sc_comp_blksz == 0 ||
1202 			    vnd->sc_comp_blksz % DEV_BSIZE !=0) {
1203 				VOP_UNLOCK(nd.ni_vp);
1204 				error = EINVAL;
1205 				goto close_and_exit;
1206 			}
1207 			if (sizeof(struct vnd_comp_header) +
1208 			  sizeof(u_int64_t) * vnd->sc_comp_numoffs >
1209 			  vattr.va_size) {
1210 				VOP_UNLOCK(nd.ni_vp);
1211 				error = EINVAL;
1212 				goto close_and_exit;
1213 			}
1214 
1215 			/* set decompressed file size */
1216 			vattr.va_size =
1217 			    ((u_quad_t)vnd->sc_comp_numoffs - 1) *
1218 			     (u_quad_t)vnd->sc_comp_blksz;
1219 
1220 			/* allocate space for all the compressed offsets */
1221 			vnd->sc_comp_offsets =
1222 			malloc(sizeof(u_int64_t) * vnd->sc_comp_numoffs,
1223 			M_DEVBUF, M_WAITOK);
1224 
1225 			/* read in the offsets */
1226 			error = vn_rdwr(UIO_READ, nd.ni_vp,
1227 			  (void *)vnd->sc_comp_offsets,
1228 			  sizeof(u_int64_t) * vnd->sc_comp_numoffs,
1229 			  sizeof(struct vnd_comp_header), UIO_SYSSPACE,
1230 			  IO_UNIT|IO_NODELOCKED, l->l_cred, NULL, NULL);
1231 			if (error) {
1232 				VOP_UNLOCK(nd.ni_vp);
1233 				goto close_and_exit;
1234 			}
1235 			/*
1236 			 * find largest block size (used for allocation limit).
1237 			 * Also convert offset to native byte order.
1238 			 */
1239 			comp_maxsize = 0;
1240 			for (i = 0; i < vnd->sc_comp_numoffs - 1; i++) {
1241 				vnd->sc_comp_offsets[i] =
1242 				  be64toh(vnd->sc_comp_offsets[i]);
1243 				comp_size = be64toh(vnd->sc_comp_offsets[i + 1])
1244 				  - vnd->sc_comp_offsets[i];
1245 				if (comp_size > comp_maxsize)
1246 					comp_maxsize = comp_size;
1247 			}
1248 			vnd->sc_comp_offsets[vnd->sc_comp_numoffs - 1] =
1249 			  be64toh(vnd->sc_comp_offsets[vnd->sc_comp_numoffs - 1]);
1250 
1251 			/* create compressed data buffer */
1252 			vnd->sc_comp_buff = malloc(comp_maxsize,
1253 			  M_DEVBUF, M_WAITOK);
1254 
1255 			/* create decompressed buffer */
1256 			vnd->sc_comp_decombuf = malloc(vnd->sc_comp_blksz,
1257 			  M_DEVBUF, M_WAITOK);
1258 			vnd->sc_comp_buffblk = -1;
1259 
1260 			/* Initialize decompress stream */
1261 			memset(&vnd->sc_comp_stream, 0, sizeof(z_stream));
1262 			vnd->sc_comp_stream.zalloc = vnd_alloc;
1263 			vnd->sc_comp_stream.zfree = vnd_free;
1264 			error = inflateInit2(&vnd->sc_comp_stream, MAX_WBITS);
1265 			if (error) {
1266 				if (vnd->sc_comp_stream.msg)
1267 					printf("vnd%d: compressed file, %s\n",
1268 					  unit, vnd->sc_comp_stream.msg);
1269 				VOP_UNLOCK(nd.ni_vp);
1270 				error = EINVAL;
1271 				goto close_and_exit;
1272 			}
1273 
1274 			vnd->sc_flags |= VNF_COMP | VNF_READONLY;
1275 #else /* !VND_COMPRESSION */
1276 			VOP_UNLOCK(nd.ni_vp);
1277 			error = EOPNOTSUPP;
1278 			goto close_and_exit;
1279 #endif /* VND_COMPRESSION */
1280 		}
1281 
1282 		VOP_UNLOCK(nd.ni_vp);
1283 		vnd->sc_vp = nd.ni_vp;
1284 		vnd->sc_size = btodb(vattr.va_size);	/* note truncation */
1285 
1286 		/*
1287 		 * Use pseudo-geometry specified.  If none was provided,
1288 		 * use "standard" Adaptec fictitious geometry.
1289 		 */
1290 		if (vio->vnd_flags & VNDIOF_HASGEOM) {
1291 
1292 			memcpy(&vnd->sc_geom, &vio->vnd_geom,
1293 			    sizeof(vio->vnd_geom));
1294 
1295 			/*
1296 			 * Sanity-check the sector size.
1297 			 * XXX Don't allow secsize < DEV_BSIZE.	 Should
1298 			 * XXX we?
1299 			 */
1300 			if (vnd->sc_geom.vng_secsize < DEV_BSIZE ||
1301 			    (vnd->sc_geom.vng_secsize % DEV_BSIZE) != 0 ||
1302 			    vnd->sc_geom.vng_ncylinders == 0 ||
1303 			    (vnd->sc_geom.vng_ntracks *
1304 			     vnd->sc_geom.vng_nsectors) == 0) {
1305 				error = EINVAL;
1306 				goto close_and_exit;
1307 			}
1308 
1309 			/*
1310 			 * Compute the size (in DEV_BSIZE blocks) specified
1311 			 * by the geometry.
1312 			 */
1313 			geomsize = (int64_t)vnd->sc_geom.vng_nsectors *
1314 			    vnd->sc_geom.vng_ntracks *
1315 			    vnd->sc_geom.vng_ncylinders *
1316 			    (vnd->sc_geom.vng_secsize / DEV_BSIZE);
1317 
1318 			/*
1319 			 * Sanity-check the size against the specified
1320 			 * geometry.
1321 			 */
1322 			if (vnd->sc_size < geomsize) {
1323 				error = EINVAL;
1324 				goto close_and_exit;
1325 			}
1326 		} else if (vnd->sc_size >= (32 * 64)) {
1327 			/*
1328 			 * Size must be at least 2048 DEV_BSIZE blocks
1329 			 * (1M) in order to use this geometry.
1330 			 */
1331 			vnd->sc_geom.vng_secsize = DEV_BSIZE;
1332 			vnd->sc_geom.vng_nsectors = 32;
1333 			vnd->sc_geom.vng_ntracks = 64;
1334 			vnd->sc_geom.vng_ncylinders = vnd->sc_size / (64 * 32);
1335 		} else {
1336 			vnd->sc_geom.vng_secsize = DEV_BSIZE;
1337 			vnd->sc_geom.vng_nsectors = 1;
1338 			vnd->sc_geom.vng_ntracks = 1;
1339 			vnd->sc_geom.vng_ncylinders = vnd->sc_size;
1340 		}
1341 
1342 		vnd_set_geometry(vnd);
1343 
1344 		if (vio->vnd_flags & VNDIOF_READONLY) {
1345 			vnd->sc_flags |= VNF_READONLY;
1346 		}
1347 
1348 		if ((error = vndsetcred(vnd, l->l_cred)) != 0)
1349 			goto close_and_exit;
1350 
1351 		vndthrottle(vnd, vnd->sc_vp);
1352 		vio->vnd_osize = dbtob(vnd->sc_size);
1353 #ifdef VNDIOCSET50
1354 		if (cmd != VNDIOCSET50)
1355 #endif
1356 			vio->vnd_size = dbtob(vnd->sc_size);
1357 		vnd->sc_flags |= VNF_INITED;
1358 
1359 		/* create the kernel thread, wait for it to be up */
1360 		error = kthread_create(PRI_NONE, 0, NULL, vndthread, vnd,
1361 		    &vnd->sc_kthread, "%s", device_xname(vnd->sc_dev));
1362 		if (error)
1363 			goto close_and_exit;
1364 		while ((vnd->sc_flags & VNF_KTHREAD) == 0) {
1365 			tsleep(&vnd->sc_kthread, PRIBIO, "vndthr", 0);
1366 		}
1367 #ifdef DEBUG
1368 		if (vnddebug & VDB_INIT)
1369 			printf("vndioctl: SET vp %p size 0x%lx %d/%d/%d/%d\n",
1370 			    vnd->sc_vp, (unsigned long) vnd->sc_size,
1371 			    vnd->sc_geom.vng_secsize,
1372 			    vnd->sc_geom.vng_nsectors,
1373 			    vnd->sc_geom.vng_ntracks,
1374 			    vnd->sc_geom.vng_ncylinders);
1375 #endif
1376 
1377 		/* Attach the disk. */
1378 		disk_attach(&vnd->sc_dkdev);
1379 
1380 		/* Initialize the xfer and buffer pools. */
1381 		pool_init(&vnd->sc_vxpool, sizeof(struct vndxfer), 0,
1382 		    0, 0, "vndxpl", NULL, IPL_BIO);
1383 
1384 		vndunlock(vnd);
1385 
1386 		pathbuf_destroy(pb);
1387 
1388 		/* Discover wedges on this disk */
1389 		dkwedge_discover(&vnd->sc_dkdev);
1390 
1391 		break;
1392 
1393 close_and_exit:
1394 		(void) vn_close(nd.ni_vp, fflags, l->l_cred);
1395 		pathbuf_destroy(pb);
1396 unlock_and_exit:
1397 #ifdef VND_COMPRESSION
1398 		/* free any allocated memory (for compressed file) */
1399 		if (vnd->sc_comp_offsets) {
1400 			free(vnd->sc_comp_offsets, M_DEVBUF);
1401 			vnd->sc_comp_offsets = NULL;
1402 		}
1403 		if (vnd->sc_comp_buff) {
1404 			free(vnd->sc_comp_buff, M_DEVBUF);
1405 			vnd->sc_comp_buff = NULL;
1406 		}
1407 		if (vnd->sc_comp_decombuf) {
1408 			free(vnd->sc_comp_decombuf, M_DEVBUF);
1409 			vnd->sc_comp_decombuf = NULL;
1410 		}
1411 #endif /* VND_COMPRESSION */
1412 		vndunlock(vnd);
1413 		return error;
1414 
1415 #ifdef VNDIOCCLR50
1416 	case VNDIOCCLR50:
1417 #endif
1418 	case VNDIOCCLR:
1419 		part = DISKPART(dev);
1420 		pmask = (1 << part);
1421 		force = (vio->vnd_flags & VNDIOF_FORCE) != 0;
1422 
1423 		if ((error = vnddoclear(vnd, pmask, minor(dev), force)) != 0)
1424 			return error;
1425 
1426 		break;
1427 
1428 #ifdef COMPAT_30
1429 	case VNDIOCGET30: {
1430 		struct vnd_user30 *vnu;
1431 		struct vattr va;
1432 		vnu = (struct vnd_user30 *)data;
1433 		KASSERT(l);
1434 		switch (error = vnd_cget(l, unit, &vnu->vnu_unit, &va)) {
1435 		case 0:
1436 			vnu->vnu_dev = va.va_fsid;
1437 			vnu->vnu_ino = va.va_fileid;
1438 			break;
1439 		case -1:
1440 			/* unused is not an error */
1441 			vnu->vnu_dev = 0;
1442 			vnu->vnu_ino = 0;
1443 			break;
1444 		default:
1445 			return error;
1446 		}
1447 		break;
1448 	}
1449 #endif
1450 
1451 #ifdef COMPAT_50
1452 	case VNDIOCGET50: {
1453 		struct vnd_user50 *vnu;
1454 		struct vattr va;
1455 		vnu = (struct vnd_user50 *)data;
1456 		KASSERT(l);
1457 		switch (error = vnd_cget(l, unit, &vnu->vnu_unit, &va)) {
1458 		case 0:
1459 			vnu->vnu_dev = va.va_fsid;
1460 			vnu->vnu_ino = va.va_fileid;
1461 			break;
1462 		case -1:
1463 			/* unused is not an error */
1464 			vnu->vnu_dev = 0;
1465 			vnu->vnu_ino = 0;
1466 			break;
1467 		default:
1468 			return error;
1469 		}
1470 		break;
1471 	}
1472 #endif
1473 
1474 	case VNDIOCGET: {
1475 		struct vnd_user *vnu;
1476 		struct vattr va;
1477 		vnu = (struct vnd_user *)data;
1478 		KASSERT(l);
1479 		switch (error = vnd_cget(l, unit, &vnu->vnu_unit, &va)) {
1480 		case 0:
1481 			vnu->vnu_dev = va.va_fsid;
1482 			vnu->vnu_ino = va.va_fileid;
1483 			break;
1484 		case -1:
1485 			/* unused is not an error */
1486 			vnu->vnu_dev = 0;
1487 			vnu->vnu_ino = 0;
1488 			break;
1489 		default:
1490 			return error;
1491 		}
1492 		break;
1493 	}
1494 
1495 	case DIOCWDINFO:
1496 	case DIOCSDINFO:
1497 #ifdef __HAVE_OLD_DISKLABEL
1498 	case ODIOCWDINFO:
1499 	case ODIOCSDINFO:
1500 #endif
1501 	{
1502 		struct disklabel *lp;
1503 
1504 		if ((error = vndlock(vnd)) != 0)
1505 			return error;
1506 
1507 		vnd->sc_flags |= VNF_LABELLING;
1508 
1509 #ifdef __HAVE_OLD_DISKLABEL
1510 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
1511 			memset(&newlabel, 0, sizeof newlabel);
1512 			memcpy(&newlabel, data, sizeof (struct olddisklabel));
1513 			lp = &newlabel;
1514 		} else
1515 #endif
1516 		lp = (struct disklabel *)data;
1517 
1518 		error = setdisklabel(vnd->sc_dkdev.dk_label,
1519 		    lp, 0, vnd->sc_dkdev.dk_cpulabel);
1520 		if (error == 0) {
1521 			if (cmd == DIOCWDINFO
1522 #ifdef __HAVE_OLD_DISKLABEL
1523 			    || cmd == ODIOCWDINFO
1524 #endif
1525 			   )
1526 				error = writedisklabel(VNDLABELDEV(dev),
1527 				    vndstrategy, vnd->sc_dkdev.dk_label,
1528 				    vnd->sc_dkdev.dk_cpulabel);
1529 		}
1530 
1531 		vnd->sc_flags &= ~VNF_LABELLING;
1532 
1533 		vndunlock(vnd);
1534 
1535 		if (error)
1536 			return error;
1537 		break;
1538 	}
1539 
1540 	case DIOCKLABEL:
1541 		if (*(int *)data != 0)
1542 			vnd->sc_flags |= VNF_KLABEL;
1543 		else
1544 			vnd->sc_flags &= ~VNF_KLABEL;
1545 		break;
1546 
1547 	case DIOCWLABEL:
1548 		if (*(int *)data != 0)
1549 			vnd->sc_flags |= VNF_WLABEL;
1550 		else
1551 			vnd->sc_flags &= ~VNF_WLABEL;
1552 		break;
1553 
1554 	case DIOCGDEFLABEL:
1555 		vndgetdefaultlabel(vnd, (struct disklabel *)data);
1556 		break;
1557 
1558 #ifdef __HAVE_OLD_DISKLABEL
1559 	case ODIOCGDEFLABEL:
1560 		vndgetdefaultlabel(vnd, &newlabel);
1561 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1562 			return ENOTTY;
1563 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1564 		break;
1565 #endif
1566 
1567 	case DIOCCACHESYNC:
1568 		vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1569 		error = VOP_FSYNC(vnd->sc_vp, vnd->sc_cred,
1570 		    FSYNC_WAIT | FSYNC_DATAONLY | FSYNC_CACHE, 0, 0);
1571 		VOP_UNLOCK(vnd->sc_vp);
1572 		return error;
1573 
1574 	default:
1575 		return ENOTTY;
1576 	}
1577 
1578 	return 0;
1579 }
1580 
1581 /*
1582  * Duplicate the current processes' credentials.  Since we are called only
1583  * as the result of a SET ioctl and only root can do that, any future access
1584  * to this "disk" is essentially as root.  Note that credentials may change
1585  * if some other uid can write directly to the mapped file (NFS).
1586  */
1587 static int
1588 vndsetcred(struct vnd_softc *vnd, kauth_cred_t cred)
1589 {
1590 	struct uio auio;
1591 	struct iovec aiov;
1592 	char *tmpbuf;
1593 	int error;
1594 
1595 	vnd->sc_cred = kauth_cred_dup(cred);
1596 	tmpbuf = malloc(DEV_BSIZE, M_TEMP, M_WAITOK);
1597 
1598 	/* XXX: Horrible kludge to establish credentials for NFS */
1599 	aiov.iov_base = tmpbuf;
1600 	aiov.iov_len = min(DEV_BSIZE, dbtob(vnd->sc_size));
1601 	auio.uio_iov = &aiov;
1602 	auio.uio_iovcnt = 1;
1603 	auio.uio_offset = 0;
1604 	auio.uio_rw = UIO_READ;
1605 	auio.uio_resid = aiov.iov_len;
1606 	UIO_SETUP_SYSSPACE(&auio);
1607 	vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1608 	error = VOP_READ(vnd->sc_vp, &auio, 0, vnd->sc_cred);
1609 	if (error == 0) {
1610 		/*
1611 		 * Because vnd does all IO directly through the vnode
1612 		 * we need to flush (at least) the buffer from the above
1613 		 * VOP_READ from the buffer cache to prevent cache
1614 		 * incoherencies.  Also, be careful to write dirty
1615 		 * buffers back to stable storage.
1616 		 */
1617 		error = vinvalbuf(vnd->sc_vp, V_SAVE, vnd->sc_cred,
1618 			    curlwp, 0, 0);
1619 	}
1620 	VOP_UNLOCK(vnd->sc_vp);
1621 
1622 	free(tmpbuf, M_TEMP);
1623 	return error;
1624 }
1625 
1626 /*
1627  * Set maxactive based on FS type
1628  */
1629 static void
1630 vndthrottle(struct vnd_softc *vnd, struct vnode *vp)
1631 {
1632 
1633 	if (vp->v_tag == VT_NFS)
1634 		vnd->sc_maxactive = 2;
1635 	else
1636 		vnd->sc_maxactive = 8;
1637 
1638 	if (vnd->sc_maxactive < 1)
1639 		vnd->sc_maxactive = 1;
1640 }
1641 
1642 #if 0
1643 static void
1644 vndshutdown(void)
1645 {
1646 	struct vnd_softc *vnd;
1647 
1648 	for (vnd = &vnd_softc[0]; vnd < &vnd_softc[numvnd]; vnd++)
1649 		if (vnd->sc_flags & VNF_INITED)
1650 			vndclear(vnd);
1651 }
1652 #endif
1653 
1654 static void
1655 vndclear(struct vnd_softc *vnd, int myminor)
1656 {
1657 	struct vnode *vp = vnd->sc_vp;
1658 	int fflags = FREAD;
1659 	int bmaj, cmaj, i, mn;
1660 	int s;
1661 
1662 #ifdef DEBUG
1663 	if (vnddebug & VDB_FOLLOW)
1664 		printf("vndclear(%p): vp %p\n", vnd, vp);
1665 #endif
1666 	/* locate the major number */
1667 	bmaj = bdevsw_lookup_major(&vnd_bdevsw);
1668 	cmaj = cdevsw_lookup_major(&vnd_cdevsw);
1669 
1670 	/* Nuke the vnodes for any open instances */
1671 	for (i = 0; i < MAXPARTITIONS; i++) {
1672 		mn = DISKMINOR(device_unit(vnd->sc_dev), i);
1673 		vdevgone(bmaj, mn, mn, VBLK);
1674 		if (mn != myminor) /* XXX avoid to kill own vnode */
1675 			vdevgone(cmaj, mn, mn, VCHR);
1676 	}
1677 
1678 	if ((vnd->sc_flags & VNF_READONLY) == 0)
1679 		fflags |= FWRITE;
1680 
1681 	s = splbio();
1682 	bufq_drain(vnd->sc_tab);
1683 	splx(s);
1684 
1685 	vnd->sc_flags |= VNF_VUNCONF;
1686 	wakeup(&vnd->sc_tab);
1687 	while (vnd->sc_flags & VNF_KTHREAD)
1688 		tsleep(&vnd->sc_kthread, PRIBIO, "vnthr", 0);
1689 
1690 #ifdef VND_COMPRESSION
1691 	/* free the compressed file buffers */
1692 	if (vnd->sc_flags & VNF_COMP) {
1693 		if (vnd->sc_comp_offsets) {
1694 			free(vnd->sc_comp_offsets, M_DEVBUF);
1695 			vnd->sc_comp_offsets = NULL;
1696 		}
1697 		if (vnd->sc_comp_buff) {
1698 			free(vnd->sc_comp_buff, M_DEVBUF);
1699 			vnd->sc_comp_buff = NULL;
1700 		}
1701 		if (vnd->sc_comp_decombuf) {
1702 			free(vnd->sc_comp_decombuf, M_DEVBUF);
1703 			vnd->sc_comp_decombuf = NULL;
1704 		}
1705 	}
1706 #endif /* VND_COMPRESSION */
1707 	vnd->sc_flags &=
1708 	    ~(VNF_INITED | VNF_READONLY | VNF_KLABEL | VNF_VLABEL
1709 	      | VNF_VUNCONF | VNF_COMP | VNF_CLEARING);
1710 	if (vp == NULL)
1711 		panic("vndclear: null vp");
1712 	(void) vn_close(vp, fflags, vnd->sc_cred);
1713 	kauth_cred_free(vnd->sc_cred);
1714 	vnd->sc_vp = NULL;
1715 	vnd->sc_cred = NULL;
1716 	vnd->sc_size = 0;
1717 }
1718 
1719 static int
1720 vndsize(dev_t dev)
1721 {
1722 	struct vnd_softc *sc;
1723 	struct disklabel *lp;
1724 	int part, unit, omask;
1725 	int size;
1726 
1727 	unit = vndunit(dev);
1728 	sc = device_lookup_private(&vnd_cd, unit);
1729 	if (sc == NULL)
1730 		return -1;
1731 
1732 	if ((sc->sc_flags & VNF_INITED) == 0)
1733 		return -1;
1734 
1735 	part = DISKPART(dev);
1736 	omask = sc->sc_dkdev.dk_openmask & (1 << part);
1737 	lp = sc->sc_dkdev.dk_label;
1738 
1739 	if (omask == 0 && vndopen(dev, 0, S_IFBLK, curlwp))	/* XXX */
1740 		return -1;
1741 
1742 	if (lp->d_partitions[part].p_fstype != FS_SWAP)
1743 		size = -1;
1744 	else
1745 		size = lp->d_partitions[part].p_size *
1746 		    (lp->d_secsize / DEV_BSIZE);
1747 
1748 	if (omask == 0 && vndclose(dev, 0, S_IFBLK, curlwp))	/* XXX */
1749 		return -1;
1750 
1751 	return size;
1752 }
1753 
1754 static int
1755 vnddump(dev_t dev, daddr_t blkno, void *va,
1756     size_t size)
1757 {
1758 
1759 	/* Not implemented. */
1760 	return ENXIO;
1761 }
1762 
1763 static void
1764 vndgetdefaultlabel(struct vnd_softc *sc, struct disklabel *lp)
1765 {
1766 	struct vndgeom *vng = &sc->sc_geom;
1767 	struct partition *pp;
1768 	unsigned spb;
1769 
1770 	memset(lp, 0, sizeof(*lp));
1771 
1772 	spb = vng->vng_secsize / DEV_BSIZE;
1773 	if (sc->sc_size / spb > UINT32_MAX)
1774 		lp->d_secperunit = UINT32_MAX;
1775 	else
1776 		lp->d_secperunit = sc->sc_size / spb;
1777 	lp->d_secsize = vng->vng_secsize;
1778 	lp->d_nsectors = vng->vng_nsectors;
1779 	lp->d_ntracks = vng->vng_ntracks;
1780 	lp->d_ncylinders = vng->vng_ncylinders;
1781 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1782 
1783 	strncpy(lp->d_typename, "vnd", sizeof(lp->d_typename));
1784 	lp->d_type = DKTYPE_VND;
1785 	strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
1786 	lp->d_rpm = 3600;
1787 	lp->d_interleave = 1;
1788 	lp->d_flags = 0;
1789 
1790 	pp = &lp->d_partitions[RAW_PART];
1791 	pp->p_offset = 0;
1792 	pp->p_size = lp->d_secperunit;
1793 	pp->p_fstype = FS_UNUSED;
1794 	lp->d_npartitions = RAW_PART + 1;
1795 
1796 	lp->d_magic = DISKMAGIC;
1797 	lp->d_magic2 = DISKMAGIC;
1798 	lp->d_checksum = dkcksum(lp);
1799 }
1800 
1801 /*
1802  * Read the disklabel from a vnd.  If one is not present, create a fake one.
1803  */
1804 static void
1805 vndgetdisklabel(dev_t dev, struct vnd_softc *sc)
1806 {
1807 	const char *errstring;
1808 	struct disklabel *lp = sc->sc_dkdev.dk_label;
1809 	struct cpu_disklabel *clp = sc->sc_dkdev.dk_cpulabel;
1810 	int i;
1811 
1812 	memset(clp, 0, sizeof(*clp));
1813 
1814 	vndgetdefaultlabel(sc, lp);
1815 
1816 	/*
1817 	 * Call the generic disklabel extraction routine.
1818 	 */
1819 	errstring = readdisklabel(VNDLABELDEV(dev), vndstrategy, lp, clp);
1820 	if (errstring) {
1821 		/*
1822 		 * Lack of disklabel is common, but we print the warning
1823 		 * anyway, since it might contain other useful information.
1824 		 */
1825 		aprint_normal_dev(sc->sc_dev, "%s\n", errstring);
1826 
1827 		/*
1828 		 * For historical reasons, if there's no disklabel
1829 		 * present, all partitions must be FS_BSDFFS and
1830 		 * occupy the entire disk.
1831 		 */
1832 		for (i = 0; i < MAXPARTITIONS; i++) {
1833 			/*
1834 			 * Don't wipe out port specific hack (such as
1835 			 * dos partition hack of i386 port).
1836 			 */
1837 			if (lp->d_partitions[i].p_size != 0)
1838 				continue;
1839 
1840 			lp->d_partitions[i].p_size = lp->d_secperunit;
1841 			lp->d_partitions[i].p_offset = 0;
1842 			lp->d_partitions[i].p_fstype = FS_BSDFFS;
1843 		}
1844 
1845 		strncpy(lp->d_packname, "default label",
1846 		    sizeof(lp->d_packname));
1847 
1848 		lp->d_npartitions = MAXPARTITIONS;
1849 		lp->d_checksum = dkcksum(lp);
1850 	}
1851 }
1852 
1853 /*
1854  * Wait interruptibly for an exclusive lock.
1855  *
1856  * XXX
1857  * Several drivers do this; it should be abstracted and made MP-safe.
1858  */
1859 static int
1860 vndlock(struct vnd_softc *sc)
1861 {
1862 	int error;
1863 
1864 	while ((sc->sc_flags & VNF_LOCKED) != 0) {
1865 		sc->sc_flags |= VNF_WANTED;
1866 		if ((error = tsleep(sc, PRIBIO | PCATCH, "vndlck", 0)) != 0)
1867 			return error;
1868 	}
1869 	sc->sc_flags |= VNF_LOCKED;
1870 	return 0;
1871 }
1872 
1873 /*
1874  * Unlock and wake up any waiters.
1875  */
1876 static void
1877 vndunlock(struct vnd_softc *sc)
1878 {
1879 
1880 	sc->sc_flags &= ~VNF_LOCKED;
1881 	if ((sc->sc_flags & VNF_WANTED) != 0) {
1882 		sc->sc_flags &= ~VNF_WANTED;
1883 		wakeup(sc);
1884 	}
1885 }
1886 
1887 #ifdef VND_COMPRESSION
1888 /* compressed file read */
1889 static void
1890 compstrategy(struct buf *bp, off_t bn)
1891 {
1892 	int error;
1893 	int unit = vndunit(bp->b_dev);
1894 	struct vnd_softc *vnd =
1895 	    device_lookup_private(&vnd_cd, unit);
1896 	u_int32_t comp_block;
1897 	struct uio auio;
1898 	char *addr;
1899 	int s;
1900 
1901 	/* set up constants for data move */
1902 	auio.uio_rw = UIO_READ;
1903 	UIO_SETUP_SYSSPACE(&auio);
1904 
1905 	/* read, and transfer the data */
1906 	addr = bp->b_data;
1907 	bp->b_resid = bp->b_bcount;
1908 	s = splbio();
1909 	while (bp->b_resid > 0) {
1910 		unsigned length;
1911 		size_t length_in_buffer;
1912 		u_int32_t offset_in_buffer;
1913 		struct iovec aiov;
1914 
1915 		/* calculate the compressed block number */
1916 		comp_block = bn / (off_t)vnd->sc_comp_blksz;
1917 
1918 		/* check for good block number */
1919 		if (comp_block >= vnd->sc_comp_numoffs) {
1920 			bp->b_error = EINVAL;
1921 			splx(s);
1922 			return;
1923 		}
1924 
1925 		/* read in the compressed block, if not in buffer */
1926 		if (comp_block != vnd->sc_comp_buffblk) {
1927 			length = vnd->sc_comp_offsets[comp_block + 1] -
1928 			    vnd->sc_comp_offsets[comp_block];
1929 			vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1930 			error = vn_rdwr(UIO_READ, vnd->sc_vp, vnd->sc_comp_buff,
1931 			    length, vnd->sc_comp_offsets[comp_block],
1932 			    UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, vnd->sc_cred,
1933 			    NULL, NULL);
1934 			if (error) {
1935 				bp->b_error = error;
1936 				VOP_UNLOCK(vnd->sc_vp);
1937 				splx(s);
1938 				return;
1939 			}
1940 			/* uncompress the buffer */
1941 			vnd->sc_comp_stream.next_in = vnd->sc_comp_buff;
1942 			vnd->sc_comp_stream.avail_in = length;
1943 			vnd->sc_comp_stream.next_out = vnd->sc_comp_decombuf;
1944 			vnd->sc_comp_stream.avail_out = vnd->sc_comp_blksz;
1945 			inflateReset(&vnd->sc_comp_stream);
1946 			error = inflate(&vnd->sc_comp_stream, Z_FINISH);
1947 			if (error != Z_STREAM_END) {
1948 				if (vnd->sc_comp_stream.msg)
1949 					aprint_normal_dev(vnd->sc_dev,
1950 					    "compressed file, %s\n",
1951 					    vnd->sc_comp_stream.msg);
1952 				bp->b_error = EBADMSG;
1953 				VOP_UNLOCK(vnd->sc_vp);
1954 				splx(s);
1955 				return;
1956 			}
1957 			vnd->sc_comp_buffblk = comp_block;
1958 			VOP_UNLOCK(vnd->sc_vp);
1959 		}
1960 
1961 		/* transfer the usable uncompressed data */
1962 		offset_in_buffer = bn % (off_t)vnd->sc_comp_blksz;
1963 		length_in_buffer = vnd->sc_comp_blksz - offset_in_buffer;
1964 		if (length_in_buffer > bp->b_resid)
1965 			length_in_buffer = bp->b_resid;
1966 		auio.uio_iov = &aiov;
1967 		auio.uio_iovcnt = 1;
1968 		aiov.iov_base = addr;
1969 		aiov.iov_len = length_in_buffer;
1970 		auio.uio_resid = aiov.iov_len;
1971 		auio.uio_offset = 0;
1972 		error = uiomove(vnd->sc_comp_decombuf + offset_in_buffer,
1973 		    length_in_buffer, &auio);
1974 		if (error) {
1975 			bp->b_error = error;
1976 			splx(s);
1977 			return;
1978 		}
1979 
1980 		bn += length_in_buffer;
1981 		addr += length_in_buffer;
1982 		bp->b_resid -= length_in_buffer;
1983 	}
1984 	splx(s);
1985 }
1986 
1987 /* compression memory allocation routines */
1988 static void *
1989 vnd_alloc(void *aux, u_int items, u_int siz)
1990 {
1991 	return malloc(items * siz, M_TEMP, M_NOWAIT);
1992 }
1993 
1994 static void
1995 vnd_free(void *aux, void *ptr)
1996 {
1997 	free(ptr, M_TEMP);
1998 }
1999 #endif /* VND_COMPRESSION */
2000 
2001 static void
2002 vnd_set_geometry(struct vnd_softc *vnd)
2003 {
2004 	struct disk_geom *dg = &vnd->sc_dkdev.dk_geom;
2005 
2006 	memset(dg, 0, sizeof(*dg));
2007 
2008 	dg->dg_secperunit = (int64_t)vnd->sc_geom.vng_nsectors *
2009 	    vnd->sc_geom.vng_ntracks * vnd->sc_geom.vng_ncylinders;
2010 	dg->dg_secsize = vnd->sc_geom.vng_secsize;
2011 	dg->dg_nsectors = vnd->sc_geom.vng_nsectors;
2012 	dg->dg_ntracks = vnd->sc_geom.vng_ntracks;
2013 	dg->dg_ncylinders = vnd->sc_geom.vng_ncylinders;
2014 
2015 #ifdef DEBUG
2016 	if (vnddebug & VDB_LABEL) {
2017 		printf("dg->dg_secperunit: %" PRId64 "\n", dg->dg_secperunit);
2018 		printf("dg->dg_ncylinders: %u\n", dg->dg_ncylinders);
2019 	}
2020 #endif
2021 	disk_set_info(vnd->sc_dev, &vnd->sc_dkdev, NULL);
2022 }
2023 
2024 #ifdef _MODULE
2025 
2026 #include <sys/module.h>
2027 
2028 #ifdef VND_COMPRESSION
2029 #define VND_DEPENDS "zlib"
2030 #else
2031 #define VND_DEPENDS NULL
2032 #endif
2033 
2034 MODULE(MODULE_CLASS_DRIVER, vnd, VND_DEPENDS);
2035 CFDRIVER_DECL(vnd, DV_DISK, NULL);
2036 
2037 static int
2038 vnd_modcmd(modcmd_t cmd, void *arg)
2039 {
2040 	int bmajor = -1, cmajor = -1,  error = 0;
2041 
2042 	switch (cmd) {
2043 	case MODULE_CMD_INIT:
2044 		error = config_cfdriver_attach(&vnd_cd);
2045 		if (error)
2046 			break;
2047 
2048 		error = config_cfattach_attach(vnd_cd.cd_name, &vnd_ca);
2049 	        if (error) {
2050 			config_cfdriver_detach(&vnd_cd);
2051 			aprint_error("%s: unable to register cfattach\n",
2052 			    vnd_cd.cd_name);
2053 			break;
2054 		}
2055 
2056 		error = devsw_attach("vnd", &vnd_bdevsw, &bmajor,
2057 		    &vnd_cdevsw, &cmajor);
2058 		if (error) {
2059 			config_cfattach_detach(vnd_cd.cd_name, &vnd_ca);
2060 			config_cfdriver_detach(&vnd_cd);
2061 			break;
2062 		}
2063 
2064 		break;
2065 
2066 	case MODULE_CMD_FINI:
2067 		error = config_cfattach_detach(vnd_cd.cd_name, &vnd_ca);
2068 		if (error)
2069 			break;
2070 		config_cfdriver_detach(&vnd_cd);
2071 		devsw_detach(&vnd_bdevsw, &vnd_cdevsw);
2072 		break;
2073 
2074 	case MODULE_CMD_STAT:
2075 		return ENOTTY;
2076 
2077 	default:
2078 		return ENOTTY;
2079 	}
2080 
2081 	return error;
2082 }
2083 
2084 #endif
2085