xref: /netbsd-src/sys/dev/vnd.c (revision c2f76ff004a2cb67efe5b12d97bd3ef7fe89e18d)
1 /*	$NetBSD: vnd.c,v 1.214 2010/11/19 06:44:39 dholland 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 1.214 2010/11/19 06:44:39 dholland Exp $");
134 
135 #if defined(_KERNEL_OPT)
136 #include "opt_vnd.h"
137 #include "opt_compat_netbsd.h"
138 #endif
139 
140 #include <sys/param.h>
141 #include <sys/systm.h>
142 #include <sys/namei.h>
143 #include <sys/proc.h>
144 #include <sys/kthread.h>
145 #include <sys/errno.h>
146 #include <sys/buf.h>
147 #include <sys/bufq.h>
148 #include <sys/malloc.h>
149 #include <sys/ioctl.h>
150 #include <sys/disklabel.h>
151 #include <sys/device.h>
152 #include <sys/disk.h>
153 #include <sys/stat.h>
154 #include <sys/mount.h>
155 #include <sys/vnode.h>
156 #include <sys/file.h>
157 #include <sys/uio.h>
158 #include <sys/conf.h>
159 #include <sys/kauth.h>
160 
161 #include <net/zlib.h>
162 
163 #include <miscfs/genfs/genfs.h>
164 #include <miscfs/specfs/specdev.h>
165 
166 #include <dev/dkvar.h>
167 #include <dev/vndvar.h>
168 
169 #include <prop/proplib.h>
170 
171 #if defined(VNDDEBUG) && !defined(DEBUG)
172 #define DEBUG
173 #endif
174 
175 #ifdef DEBUG
176 int dovndcluster = 1;
177 #define VDB_FOLLOW	0x01
178 #define VDB_INIT	0x02
179 #define VDB_IO		0x04
180 #define VDB_LABEL	0x08
181 int vnddebug = 0x00;
182 #endif
183 
184 #define vndunit(x)	DISKUNIT(x)
185 
186 struct vndxfer {
187 	struct buf vx_buf;
188 	struct vnd_softc *vx_vnd;
189 };
190 #define	VND_BUFTOXFER(bp)	((struct vndxfer *)(void *)bp)
191 
192 #define VND_GETXFER(vnd)	pool_get(&(vnd)->sc_vxpool, PR_WAITOK)
193 #define VND_PUTXFER(vnd, vx)	pool_put(&(vnd)->sc_vxpool, (vx))
194 
195 #define VNDLABELDEV(dev) \
196     (MAKEDISKDEV(major((dev)), vndunit((dev)), RAW_PART))
197 
198 /* called by main() at boot time */
199 void	vndattach(int);
200 
201 static void	vndclear(struct vnd_softc *, int);
202 static int	vnddoclear(struct vnd_softc *, int, int, bool);
203 static int	vndsetcred(struct vnd_softc *, kauth_cred_t);
204 static void	vndthrottle(struct vnd_softc *, struct vnode *);
205 static void	vndiodone(struct buf *);
206 #if 0
207 static void	vndshutdown(void);
208 #endif
209 
210 static void	vndgetdefaultlabel(struct vnd_softc *, struct disklabel *);
211 static void	vndgetdisklabel(dev_t, struct vnd_softc *);
212 
213 static int	vndlock(struct vnd_softc *);
214 static void	vndunlock(struct vnd_softc *);
215 #ifdef VND_COMPRESSION
216 static void	compstrategy(struct buf *, off_t);
217 static void	*vnd_alloc(void *, u_int, u_int);
218 static void	vnd_free(void *, void *);
219 #endif /* VND_COMPRESSION */
220 
221 static void	vndthread(void *);
222 static bool	vnode_has_op(const struct vnode *, int);
223 static void	handle_with_rdwr(struct vnd_softc *, const struct buf *,
224 		    struct buf *);
225 static void	handle_with_strategy(struct vnd_softc *, const struct buf *,
226 		    struct buf *);
227 static void	vnd_set_properties(struct vnd_softc *);
228 
229 static dev_type_open(vndopen);
230 static dev_type_close(vndclose);
231 static dev_type_read(vndread);
232 static dev_type_write(vndwrite);
233 static dev_type_ioctl(vndioctl);
234 static dev_type_strategy(vndstrategy);
235 static dev_type_dump(vnddump);
236 static dev_type_size(vndsize);
237 
238 const struct bdevsw vnd_bdevsw = {
239 	vndopen, vndclose, vndstrategy, vndioctl, vnddump, vndsize, D_DISK
240 };
241 
242 const struct cdevsw vnd_cdevsw = {
243 	vndopen, vndclose, vndread, vndwrite, vndioctl,
244 	nostop, notty, nopoll, nommap, nokqfilter, D_DISK
245 };
246 
247 static int	vnd_match(device_t, cfdata_t, void *);
248 static void	vnd_attach(device_t, device_t, void *);
249 static int	vnd_detach(device_t, int);
250 
251 CFATTACH_DECL3_NEW(vnd, sizeof(struct vnd_softc),
252     vnd_match, vnd_attach, vnd_detach, NULL, NULL, NULL, DVF_DETACH_SHUTDOWN);
253 extern struct cfdriver vnd_cd;
254 
255 static struct vnd_softc	*vnd_spawn(int);
256 int	vnd_destroy(device_t);
257 
258 static struct	dkdriver vnddkdriver = { vndstrategy, minphys };
259 
260 void
261 vndattach(int num)
262 {
263 	int error;
264 
265 	error = config_cfattach_attach(vnd_cd.cd_name, &vnd_ca);
266 	if (error)
267 		aprint_error("%s: unable to register cfattach\n",
268 		    vnd_cd.cd_name);
269 }
270 
271 static int
272 vnd_match(device_t self, cfdata_t cfdata, void *aux)
273 {
274 
275 	return 1;
276 }
277 
278 static void
279 vnd_attach(device_t parent, device_t self, void *aux)
280 {
281 	struct vnd_softc *sc = device_private(self);
282 
283 	sc->sc_dev = self;
284 	sc->sc_comp_offsets = NULL;
285 	sc->sc_comp_buff = NULL;
286 	sc->sc_comp_decombuf = NULL;
287 	bufq_alloc(&sc->sc_tab, "disksort", BUFQ_SORT_RAWBLOCK);
288 	disk_init(&sc->sc_dkdev, device_xname(self), &vnddkdriver);
289 	if (!pmf_device_register(self, NULL, NULL))
290 		aprint_error_dev(self, "couldn't establish power handler\n");
291 }
292 
293 static int
294 vnd_detach(device_t self, int flags)
295 {
296 	int error;
297 	struct vnd_softc *sc = device_private(self);
298 
299 	if (sc->sc_flags & VNF_INITED) {
300 		error = vnddoclear(sc, 0, -1, (flags & DETACH_FORCE) != 0);
301 		if (error != 0)
302 			return error;
303 	}
304 
305 	pmf_device_deregister(self);
306 	bufq_free(sc->sc_tab);
307 	disk_destroy(&sc->sc_dkdev);
308 
309 	return 0;
310 }
311 
312 static struct vnd_softc *
313 vnd_spawn(int unit)
314 {
315 	cfdata_t cf;
316 
317 	cf = malloc(sizeof(*cf), M_DEVBUF, M_WAITOK);
318 	cf->cf_name = vnd_cd.cd_name;
319 	cf->cf_atname = vnd_cd.cd_name;
320 	cf->cf_unit = unit;
321 	cf->cf_fstate = FSTATE_STAR;
322 
323 	return device_private(config_attach_pseudo(cf));
324 }
325 
326 int
327 vnd_destroy(device_t dev)
328 {
329 	int error;
330 	cfdata_t cf;
331 
332 	cf = device_cfdata(dev);
333 	error = config_detach(dev, DETACH_QUIET);
334 	if (error)
335 		return error;
336 	free(cf, M_DEVBUF);
337 	return 0;
338 }
339 
340 static int
341 vndopen(dev_t dev, int flags, int mode, struct lwp *l)
342 {
343 	int unit = vndunit(dev);
344 	struct vnd_softc *sc;
345 	int error = 0, part, pmask;
346 	struct disklabel *lp;
347 
348 #ifdef DEBUG
349 	if (vnddebug & VDB_FOLLOW)
350 		printf("vndopen(0x%"PRIx64", 0x%x, 0x%x, %p)\n", dev, flags, mode, l);
351 #endif
352 	sc = device_lookup_private(&vnd_cd, unit);
353 	if (sc == NULL) {
354 		sc = vnd_spawn(unit);
355 		if (sc == NULL)
356 			return ENOMEM;
357 	}
358 
359 	if ((error = vndlock(sc)) != 0)
360 		return error;
361 
362 	if ((sc->sc_flags & VNF_CLEARING) != 0) {
363 		error = ENXIO;
364 		goto done;
365 	}
366 
367 	lp = sc->sc_dkdev.dk_label;
368 
369 	part = DISKPART(dev);
370 	pmask = (1 << part);
371 
372 	/*
373 	 * If we're initialized, check to see if there are any other
374 	 * open partitions.  If not, then it's safe to update the
375 	 * in-core disklabel.  Only read the disklabel if it is
376 	 * not already valid.
377 	 */
378 	if ((sc->sc_flags & (VNF_INITED|VNF_VLABEL)) == VNF_INITED &&
379 	    sc->sc_dkdev.dk_openmask == 0)
380 		vndgetdisklabel(dev, sc);
381 
382 	/* Check that the partitions exists. */
383 	if (part != RAW_PART) {
384 		if (((sc->sc_flags & VNF_INITED) == 0) ||
385 		    ((part >= lp->d_npartitions) ||
386 		     (lp->d_partitions[part].p_fstype == FS_UNUSED))) {
387 			error = ENXIO;
388 			goto done;
389 		}
390 	}
391 
392 	/* Prevent our unit from being unconfigured while open. */
393 	switch (mode) {
394 	case S_IFCHR:
395 		sc->sc_dkdev.dk_copenmask |= pmask;
396 		break;
397 
398 	case S_IFBLK:
399 		sc->sc_dkdev.dk_bopenmask |= pmask;
400 		break;
401 	}
402 	sc->sc_dkdev.dk_openmask =
403 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
404 
405  done:
406 	vndunlock(sc);
407 	return error;
408 }
409 
410 static int
411 vndclose(dev_t dev, int flags, int mode, struct lwp *l)
412 {
413 	int unit = vndunit(dev);
414 	struct vnd_softc *sc;
415 	int error = 0, part;
416 
417 #ifdef DEBUG
418 	if (vnddebug & VDB_FOLLOW)
419 		printf("vndclose(0x%"PRIx64", 0x%x, 0x%x, %p)\n", dev, flags, mode, l);
420 #endif
421 	sc = device_lookup_private(&vnd_cd, unit);
422 	if (sc == NULL)
423 		return ENXIO;
424 
425 	if ((error = vndlock(sc)) != 0)
426 		return error;
427 
428 	part = DISKPART(dev);
429 
430 	/* ...that much closer to allowing unconfiguration... */
431 	switch (mode) {
432 	case S_IFCHR:
433 		sc->sc_dkdev.dk_copenmask &= ~(1 << part);
434 		break;
435 
436 	case S_IFBLK:
437 		sc->sc_dkdev.dk_bopenmask &= ~(1 << part);
438 		break;
439 	}
440 	sc->sc_dkdev.dk_openmask =
441 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
442 
443 	vndunlock(sc);
444 
445 	if ((sc->sc_flags & VNF_INITED) == 0) {
446 		if ((error = vnd_destroy(sc->sc_dev)) != 0) {
447 			aprint_error_dev(sc->sc_dev,
448 			    "unable to detach instance\n");
449 			return error;
450 		}
451 	}
452 
453 	return 0;
454 }
455 
456 /*
457  * Queue the request, and wakeup the kernel thread to handle it.
458  */
459 static void
460 vndstrategy(struct buf *bp)
461 {
462 	int unit = vndunit(bp->b_dev);
463 	struct vnd_softc *vnd =
464 	    device_lookup_private(&vnd_cd, unit);
465 	struct disklabel *lp;
466 	daddr_t blkno;
467 	int s = splbio();
468 
469 	if (vnd == NULL) {
470 		bp->b_error = ENXIO;
471 		goto done;
472 	}
473 	lp = vnd->sc_dkdev.dk_label;
474 
475 	if ((vnd->sc_flags & VNF_INITED) == 0) {
476 		bp->b_error = ENXIO;
477 		goto done;
478 	}
479 
480 	/*
481 	 * The transfer must be a whole number of blocks.
482 	 */
483 	if ((bp->b_bcount % lp->d_secsize) != 0) {
484 		bp->b_error = EINVAL;
485 		goto done;
486 	}
487 
488 	/*
489 	 * check if we're read-only.
490 	 */
491 	if ((vnd->sc_flags & VNF_READONLY) && !(bp->b_flags & B_READ)) {
492 		bp->b_error = EACCES;
493 		goto done;
494 	}
495 
496 	/* If it's a nil transfer, wake up the top half now. */
497 	if (bp->b_bcount == 0) {
498 		goto done;
499 	}
500 
501 	/*
502 	 * Do bounds checking and adjust transfer.  If there's an error,
503 	 * the bounds check will flag that for us.
504 	 */
505 	if (DISKPART(bp->b_dev) == RAW_PART) {
506 		if (bounds_check_with_mediasize(bp, DEV_BSIZE,
507 		    vnd->sc_size) <= 0)
508 			goto done;
509 	} else {
510 		if (bounds_check_with_label(&vnd->sc_dkdev,
511 		    bp, vnd->sc_flags & (VNF_WLABEL|VNF_LABELLING)) <= 0)
512 			goto done;
513 	}
514 
515 	/*
516 	 * Put the block number in terms of the logical blocksize
517 	 * of the "device".
518 	 */
519 
520 	blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
521 
522 	/*
523 	 * Translate the partition-relative block number to an absolute.
524 	 */
525 	if (DISKPART(bp->b_dev) != RAW_PART) {
526 		struct partition *pp;
527 
528 		pp = &vnd->sc_dkdev.dk_label->d_partitions[
529 		    DISKPART(bp->b_dev)];
530 		blkno += pp->p_offset;
531 	}
532 	bp->b_rawblkno = blkno;
533 
534 #ifdef DEBUG
535 	if (vnddebug & VDB_FOLLOW)
536 		printf("vndstrategy(%p): unit %d\n", bp, unit);
537 #endif
538 	bufq_put(vnd->sc_tab, bp);
539 	wakeup(&vnd->sc_tab);
540 	splx(s);
541 	return;
542 
543 done:
544 	bp->b_resid = bp->b_bcount;
545 	biodone(bp);
546 	splx(s);
547 }
548 
549 static bool
550 vnode_has_strategy(struct vnd_softc *vnd)
551 {
552 	return vnode_has_op(vnd->sc_vp, VOFFSET(vop_bmap)) &&
553 	    vnode_has_op(vnd->sc_vp, VOFFSET(vop_strategy));
554 }
555 
556 /* XXX this function needs a reliable check to detect
557  * sparse files. Otherwise, bmap/strategy may be used
558  * and fail on non-allocated blocks. VOP_READ/VOP_WRITE
559  * works on sparse files.
560  */
561 #if notyet
562 static bool
563 vnode_strategy_probe(struct vnd_softc *vnd)
564 {
565 	int error;
566 	daddr_t nbn;
567 
568 	if (!vnode_has_strategy(vnd))
569 		return false;
570 
571 	/* Convert the first logical block number to its
572 	 * physical block number.
573 	 */
574 	error = 0;
575 	vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
576 	error = VOP_BMAP(vnd->sc_vp, 0, NULL, &nbn, NULL);
577 	VOP_UNLOCK(vnd->sc_vp);
578 
579 	/* Test if that worked. */
580 	if (error == 0 && (long)nbn == -1)
581 		return false;
582 
583 	return true;
584 }
585 #endif
586 
587 static void
588 vndthread(void *arg)
589 {
590 	struct vnd_softc *vnd = arg;
591 	bool usestrategy;
592 	int s;
593 
594 	/* Determine whether we can *use* VOP_BMAP and VOP_STRATEGY to
595 	 * directly access the backing vnode.  If we can, use these two
596 	 * operations to avoid messing with the local buffer cache.
597 	 * Otherwise fall back to regular VOP_READ/VOP_WRITE operations
598 	 * which are guaranteed to work with any file system. */
599 	usestrategy = vnode_has_strategy(vnd);
600 
601 #ifdef DEBUG
602 	if (vnddebug & VDB_INIT)
603 		printf("vndthread: vp %p, %s\n", vnd->sc_vp,
604 		    usestrategy ?
605 		    "using bmap/strategy operations" :
606 		    "using read/write operations");
607 #endif
608 
609 	s = splbio();
610 	vnd->sc_flags |= VNF_KTHREAD;
611 	wakeup(&vnd->sc_kthread);
612 
613 	/*
614 	 * Dequeue requests and serve them depending on the available
615 	 * vnode operations.
616 	 */
617 	while ((vnd->sc_flags & VNF_VUNCONF) == 0) {
618 		struct vndxfer *vnx;
619 		int flags;
620 		struct buf *obp;
621 		struct buf *bp;
622 
623 		obp = bufq_get(vnd->sc_tab);
624 		if (obp == NULL) {
625 			tsleep(&vnd->sc_tab, PRIBIO, "vndbp", 0);
626 			continue;
627 		};
628 		splx(s);
629 		flags = obp->b_flags;
630 #ifdef DEBUG
631 		if (vnddebug & VDB_FOLLOW)
632 			printf("vndthread(%p)\n", obp);
633 #endif
634 
635 		if (vnd->sc_vp->v_mount == NULL) {
636 			obp->b_error = ENXIO;
637 			goto done;
638 		}
639 #ifdef VND_COMPRESSION
640 		/* handle a compressed read */
641 		if ((flags & B_READ) != 0 && (vnd->sc_flags & VNF_COMP)) {
642 			off_t bn;
643 
644 			/* Convert to a byte offset within the file. */
645 			bn = obp->b_rawblkno *
646 			    vnd->sc_dkdev.dk_label->d_secsize;
647 
648 			compstrategy(obp, bn);
649 			goto done;
650 		}
651 #endif /* VND_COMPRESSION */
652 
653 		/*
654 		 * Allocate a header for this transfer and link it to the
655 		 * buffer
656 		 */
657 		s = splbio();
658 		vnx = VND_GETXFER(vnd);
659 		splx(s);
660 		vnx->vx_vnd = vnd;
661 
662 		s = splbio();
663 		while (vnd->sc_active >= vnd->sc_maxactive) {
664 			tsleep(&vnd->sc_tab, PRIBIO, "vndac", 0);
665 		}
666 		vnd->sc_active++;
667 		splx(s);
668 
669 		/* Instrumentation. */
670 		disk_busy(&vnd->sc_dkdev);
671 
672 		bp = &vnx->vx_buf;
673 		buf_init(bp);
674 		bp->b_flags = (obp->b_flags & B_READ);
675 		bp->b_oflags = obp->b_oflags;
676 		bp->b_cflags = obp->b_cflags;
677 		bp->b_iodone = vndiodone;
678 		bp->b_private = obp;
679 		bp->b_vp = vnd->sc_vp;
680 		bp->b_objlock = &bp->b_vp->v_interlock;
681 		bp->b_data = obp->b_data;
682 		bp->b_bcount = obp->b_bcount;
683 		BIO_COPYPRIO(bp, obp);
684 
685 		/* Handle the request using the appropriate operations. */
686 		if (usestrategy)
687 			handle_with_strategy(vnd, obp, bp);
688 		else
689 			handle_with_rdwr(vnd, obp, bp);
690 
691 		s = splbio();
692 		continue;
693 
694 done:
695 		biodone(obp);
696 		s = splbio();
697 	}
698 
699 	vnd->sc_flags &= (~VNF_KTHREAD | VNF_VUNCONF);
700 	wakeup(&vnd->sc_kthread);
701 	splx(s);
702 	kthread_exit(0);
703 }
704 
705 /*
706  * Checks if the given vnode supports the requested operation.
707  * The operation is specified the offset returned by VOFFSET.
708  *
709  * XXX The test below used to determine this is quite fragile
710  * because it relies on the file system to use genfs to specify
711  * unimplemented operations.  There might be another way to do
712  * it more cleanly.
713  */
714 static bool
715 vnode_has_op(const struct vnode *vp, int opoffset)
716 {
717 	int (*defaultp)(void *);
718 	int (*opp)(void *);
719 
720 	defaultp = vp->v_op[VOFFSET(vop_default)];
721 	opp = vp->v_op[opoffset];
722 
723 	return opp != defaultp && opp != genfs_eopnotsupp &&
724 	    opp != genfs_badop && opp != genfs_nullop;
725 }
726 
727 /*
728  * Handes the read/write request given in 'bp' using the vnode's VOP_READ
729  * and VOP_WRITE operations.
730  *
731  * 'obp' is a pointer to the original request fed to the vnd device.
732  */
733 static void
734 handle_with_rdwr(struct vnd_softc *vnd, const struct buf *obp, struct buf *bp)
735 {
736 	bool doread;
737 	off_t offset;
738 	size_t resid;
739 	struct vnode *vp;
740 
741 	doread = bp->b_flags & B_READ;
742 	offset = obp->b_rawblkno * vnd->sc_dkdev.dk_label->d_secsize;
743 	vp = vnd->sc_vp;
744 
745 #if defined(DEBUG)
746 	if (vnddebug & VDB_IO)
747 		printf("vnd (rdwr): vp %p, %s, rawblkno 0x%" PRIx64
748 		    ", secsize %d, offset %" PRIu64
749 		    ", bcount %d\n",
750 		    vp, doread ? "read" : "write", obp->b_rawblkno,
751 		    vnd->sc_dkdev.dk_label->d_secsize, offset,
752 		    bp->b_bcount);
753 #endif
754 
755 	/* Issue the read or write operation. */
756 	bp->b_error =
757 	    vn_rdwr(doread ? UIO_READ : UIO_WRITE,
758 	    vp, bp->b_data, bp->b_bcount, offset,
759 	    UIO_SYSSPACE, 0, vnd->sc_cred, &resid, NULL);
760 	bp->b_resid = resid;
761 
762 	/* We need to increase the number of outputs on the vnode if
763 	 * there was any write to it. */
764 	if (!doread) {
765 		mutex_enter(&vp->v_interlock);
766 		vp->v_numoutput++;
767 		mutex_exit(&vp->v_interlock);
768 	}
769 
770 	biodone(bp);
771 }
772 
773 /*
774  * Handes the read/write request given in 'bp' using the vnode's VOP_BMAP
775  * and VOP_STRATEGY operations.
776  *
777  * 'obp' is a pointer to the original request fed to the vnd device.
778  */
779 static void
780 handle_with_strategy(struct vnd_softc *vnd, const struct buf *obp,
781     struct buf *bp)
782 {
783 	int bsize, error, flags, skipped;
784 	size_t resid, sz;
785 	off_t bn, offset;
786 	struct vnode *vp;
787 
788 	flags = obp->b_flags;
789 
790 	if (!(flags & B_READ)) {
791 		vp = bp->b_vp;
792 		mutex_enter(&vp->v_interlock);
793 		vp->v_numoutput++;
794 		mutex_exit(&vp->v_interlock);
795 	}
796 
797 	/* convert to a byte offset within the file. */
798 	bn = obp->b_rawblkno * vnd->sc_dkdev.dk_label->d_secsize;
799 
800 	bsize = vnd->sc_vp->v_mount->mnt_stat.f_iosize;
801 	skipped = 0;
802 
803 	/*
804 	 * Break the request into bsize pieces and feed them
805 	 * sequentially using VOP_BMAP/VOP_STRATEGY.
806 	 * We do it this way to keep from flooding NFS servers if we
807 	 * are connected to an NFS file.  This places the burden on
808 	 * the client rather than the server.
809 	 */
810 	error = 0;
811 	bp->b_resid = bp->b_bcount;
812 	for (offset = 0, resid = bp->b_resid; resid;
813 	    resid -= sz, offset += sz) {
814 		struct buf *nbp;
815 		daddr_t nbn;
816 		int off, nra;
817 
818 		nra = 0;
819 		vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
820 		error = VOP_BMAP(vnd->sc_vp, bn / bsize, &vp, &nbn, &nra);
821 		VOP_UNLOCK(vnd->sc_vp);
822 
823 		if (error == 0 && (long)nbn == -1)
824 			error = EIO;
825 
826 		/*
827 		 * If there was an error or a hole in the file...punt.
828 		 * Note that we may have to wait for any operations
829 		 * that we have already fired off before releasing
830 		 * the buffer.
831 		 *
832 		 * XXX we could deal with holes here but it would be
833 		 * a hassle (in the write case).
834 		 */
835 		if (error) {
836 			skipped += resid;
837 			break;
838 		}
839 
840 #ifdef DEBUG
841 		if (!dovndcluster)
842 			nra = 0;
843 #endif
844 
845 		off = bn % bsize;
846 		sz = MIN(((off_t)1 + nra) * bsize - off, resid);
847 #ifdef	DEBUG
848 		if (vnddebug & VDB_IO)
849 			printf("vndstrategy: vp %p/%p bn 0x%qx/0x%" PRIx64
850 			    " sz 0x%zx\n", vnd->sc_vp, vp, (long long)bn,
851 			    nbn, sz);
852 #endif
853 
854 		nbp = getiobuf(vp, true);
855 		nestiobuf_setup(bp, nbp, offset, sz);
856 		nbp->b_blkno = nbn + btodb(off);
857 
858 #if 0 /* XXX #ifdef DEBUG */
859 		if (vnddebug & VDB_IO)
860 			printf("vndstart(%ld): bp %p vp %p blkno "
861 			    "0x%" PRIx64 " flags %x addr %p cnt 0x%x\n",
862 			    (long) (vnd-vnd_softc), &nbp->vb_buf,
863 			    nbp->vb_buf.b_vp, nbp->vb_buf.b_blkno,
864 			    nbp->vb_buf.b_flags, nbp->vb_buf.b_data,
865 			    nbp->vb_buf.b_bcount);
866 #endif
867 		VOP_STRATEGY(vp, nbp);
868 		bn += sz;
869 	}
870 	nestiobuf_done(bp, skipped, error);
871 }
872 
873 static void
874 vndiodone(struct buf *bp)
875 {
876 	struct vndxfer *vnx = VND_BUFTOXFER(bp);
877 	struct vnd_softc *vnd = vnx->vx_vnd;
878 	struct buf *obp = bp->b_private;
879 	int s = splbio();
880 
881 	KASSERT(&vnx->vx_buf == bp);
882 	KASSERT(vnd->sc_active > 0);
883 #ifdef DEBUG
884 	if (vnddebug & VDB_IO) {
885 		printf("vndiodone1: bp %p iodone: error %d\n",
886 		    bp, bp->b_error);
887 	}
888 #endif
889 	disk_unbusy(&vnd->sc_dkdev, bp->b_bcount - bp->b_resid,
890 	    (bp->b_flags & B_READ));
891 	vnd->sc_active--;
892 	if (vnd->sc_active == 0) {
893 		wakeup(&vnd->sc_tab);
894 	}
895 	splx(s);
896 	obp->b_error = bp->b_error;
897 	obp->b_resid = bp->b_resid;
898 	buf_destroy(bp);
899 	VND_PUTXFER(vnd, vnx);
900 	biodone(obp);
901 }
902 
903 /* ARGSUSED */
904 static int
905 vndread(dev_t dev, struct uio *uio, int flags)
906 {
907 	int unit = vndunit(dev);
908 	struct vnd_softc *sc;
909 
910 #ifdef DEBUG
911 	if (vnddebug & VDB_FOLLOW)
912 		printf("vndread(0x%"PRIx64", %p)\n", dev, uio);
913 #endif
914 
915 	sc = device_lookup_private(&vnd_cd, unit);
916 	if (sc == NULL)
917 		return ENXIO;
918 
919 	if ((sc->sc_flags & VNF_INITED) == 0)
920 		return ENXIO;
921 
922 	return physio(vndstrategy, NULL, dev, B_READ, minphys, uio);
923 }
924 
925 /* ARGSUSED */
926 static int
927 vndwrite(dev_t dev, struct uio *uio, int flags)
928 {
929 	int unit = vndunit(dev);
930 	struct vnd_softc *sc;
931 
932 #ifdef DEBUG
933 	if (vnddebug & VDB_FOLLOW)
934 		printf("vndwrite(0x%"PRIx64", %p)\n", dev, uio);
935 #endif
936 
937 	sc = device_lookup_private(&vnd_cd, unit);
938 	if (sc == NULL)
939 		return ENXIO;
940 
941 	if ((sc->sc_flags & VNF_INITED) == 0)
942 		return ENXIO;
943 
944 	return physio(vndstrategy, NULL, dev, B_WRITE, minphys, uio);
945 }
946 
947 static int
948 vnd_cget(struct lwp *l, int unit, int *un, struct vattr *va)
949 {
950 	struct vnd_softc *vnd;
951 
952 	if (*un == -1)
953 		*un = unit;
954 	if (*un < 0)
955 		return EINVAL;
956 
957 	vnd = device_lookup_private(&vnd_cd, *un);
958 	if (vnd == NULL)
959 		return (*un >= vnd_cd.cd_ndevs) ? ENXIO : -1;
960 
961 	if ((vnd->sc_flags & VNF_INITED) == 0)
962 		return -1;
963 
964 	return VOP_GETATTR(vnd->sc_vp, va, l->l_cred);
965 }
966 
967 static int
968 vnddoclear(struct vnd_softc *vnd, int pmask, int minor, bool force)
969 {
970 	int error;
971 
972 	if ((error = vndlock(vnd)) != 0)
973 		return error;
974 
975 	/*
976 	 * Don't unconfigure if any other partitions are open
977 	 * or if both the character and block flavors of this
978 	 * partition are open.
979 	 */
980 	if (DK_BUSY(vnd, pmask) && !force) {
981 		vndunlock(vnd);
982 		return EBUSY;
983 	}
984 
985 	/*
986 	 * XXX vndclear() might call vndclose() implicitly;
987 	 * release lock to avoid recursion
988 	 *
989 	 * Set VNF_CLEARING to prevent vndopen() from
990 	 * sneaking in after we vndunlock().
991 	 */
992 	vnd->sc_flags |= VNF_CLEARING;
993 	vndunlock(vnd);
994 	vndclear(vnd, minor);
995 #ifdef DEBUG
996 	if (vnddebug & VDB_INIT)
997 		printf("vndioctl: CLRed\n");
998 #endif
999 
1000 	/* Destroy the xfer and buffer pools. */
1001 	pool_destroy(&vnd->sc_vxpool);
1002 
1003 	/* Detach the disk. */
1004 	disk_detach(&vnd->sc_dkdev);
1005 
1006 	return 0;
1007 }
1008 
1009 /* ARGSUSED */
1010 static int
1011 vndioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l)
1012 {
1013 	bool force;
1014 	int unit = vndunit(dev);
1015 	struct vnd_softc *vnd;
1016 	struct vnd_ioctl *vio;
1017 	struct vattr vattr;
1018 	struct pathbuf *pb;
1019 	struct nameidata nd;
1020 	int error, part, pmask;
1021 	size_t geomsize;
1022 	int fflags;
1023 #ifdef __HAVE_OLD_DISKLABEL
1024 	struct disklabel newlabel;
1025 #endif
1026 	struct dkwedge_info *dkw;
1027 	struct dkwedge_list *dkwl;
1028 
1029 #ifdef DEBUG
1030 	if (vnddebug & VDB_FOLLOW)
1031 		printf("vndioctl(0x%"PRIx64", 0x%lx, %p, 0x%x, %p): unit %d\n",
1032 		    dev, cmd, data, flag, l->l_proc, unit);
1033 #endif
1034 	vnd = device_lookup_private(&vnd_cd, unit);
1035 	if (vnd == NULL &&
1036 #ifdef COMPAT_30
1037 	    cmd != VNDIOCGET30 &&
1038 #endif
1039 #ifdef COMPAT_50
1040 	    cmd != VNDIOCGET50 &&
1041 #endif
1042 	    cmd != VNDIOCGET)
1043 		return ENXIO;
1044 	vio = (struct vnd_ioctl *)data;
1045 
1046 	/* Must be open for writes for these commands... */
1047 	switch (cmd) {
1048 	case VNDIOCSET:
1049 	case VNDIOCCLR:
1050 #ifdef COMPAT_50
1051 	case VNDIOCSET50:
1052 	case VNDIOCCLR50:
1053 #endif
1054 	case DIOCSDINFO:
1055 	case DIOCWDINFO:
1056 #ifdef __HAVE_OLD_DISKLABEL
1057 	case ODIOCSDINFO:
1058 	case ODIOCWDINFO:
1059 #endif
1060 	case DIOCKLABEL:
1061 	case DIOCWLABEL:
1062 		if ((flag & FWRITE) == 0)
1063 			return EBADF;
1064 	}
1065 
1066 	/* Must be initialized for these... */
1067 	switch (cmd) {
1068 	case VNDIOCCLR:
1069 #ifdef VNDIOCCLR50
1070 	case VNDIOCCLR50:
1071 #endif
1072 	case DIOCGDINFO:
1073 	case DIOCSDINFO:
1074 	case DIOCWDINFO:
1075 	case DIOCGPART:
1076 	case DIOCKLABEL:
1077 	case DIOCWLABEL:
1078 	case DIOCGDEFLABEL:
1079 	case DIOCCACHESYNC:
1080 #ifdef __HAVE_OLD_DISKLABEL
1081 	case ODIOCGDINFO:
1082 	case ODIOCSDINFO:
1083 	case ODIOCWDINFO:
1084 	case ODIOCGDEFLABEL:
1085 #endif
1086 		if ((vnd->sc_flags & VNF_INITED) == 0)
1087 			return ENXIO;
1088 	}
1089 
1090 	switch (cmd) {
1091 #ifdef VNDIOCSET50
1092 	case VNDIOCSET50:
1093 #endif
1094 	case VNDIOCSET:
1095 		if (vnd->sc_flags & VNF_INITED)
1096 			return EBUSY;
1097 
1098 		if ((error = vndlock(vnd)) != 0)
1099 			return error;
1100 
1101 		fflags = FREAD;
1102 		if ((vio->vnd_flags & VNDIOF_READONLY) == 0)
1103 			fflags |= FWRITE;
1104 		error = pathbuf_copyin(vio->vnd_file, &pb);
1105 		if (error) {
1106 			goto unlock_and_exit;
1107 		}
1108 		NDINIT(&nd, LOOKUP, FOLLOW, pb);
1109 		if ((error = vn_open(&nd, fflags, 0)) != 0) {
1110 			pathbuf_destroy(pb);
1111 			goto unlock_and_exit;
1112 		}
1113 		KASSERT(l);
1114 		error = VOP_GETATTR(nd.ni_vp, &vattr, l->l_cred);
1115 		if (!error && nd.ni_vp->v_type != VREG)
1116 			error = EOPNOTSUPP;
1117 		if (!error && vattr.va_bytes < vattr.va_size)
1118 			/* File is definitely sparse, reject here */
1119 			error = EINVAL;
1120 		if (error) {
1121 			VOP_UNLOCK(nd.ni_vp);
1122 			goto close_and_exit;
1123 		}
1124 
1125 		/* If using a compressed file, initialize its info */
1126 		/* (or abort with an error if kernel has no compression) */
1127 		if (vio->vnd_flags & VNF_COMP) {
1128 #ifdef VND_COMPRESSION
1129 			struct vnd_comp_header *ch;
1130 			int i;
1131 			u_int32_t comp_size;
1132 			u_int32_t comp_maxsize;
1133 
1134 			/* allocate space for compresed file header */
1135 			ch = malloc(sizeof(struct vnd_comp_header),
1136 			M_TEMP, M_WAITOK);
1137 
1138 			/* read compressed file header */
1139 			error = vn_rdwr(UIO_READ, nd.ni_vp, (void *)ch,
1140 			  sizeof(struct vnd_comp_header), 0, UIO_SYSSPACE,
1141 			  IO_UNIT|IO_NODELOCKED, l->l_cred, NULL, NULL);
1142 			if (error) {
1143 				free(ch, M_TEMP);
1144 				VOP_UNLOCK(nd.ni_vp);
1145 				goto close_and_exit;
1146 			}
1147 
1148 			/* save some header info */
1149 			vnd->sc_comp_blksz = ntohl(ch->block_size);
1150 			/* note last offset is the file byte size */
1151 			vnd->sc_comp_numoffs = ntohl(ch->num_blocks)+1;
1152 			free(ch, M_TEMP);
1153 			if (vnd->sc_comp_blksz == 0 ||
1154 			    vnd->sc_comp_blksz % DEV_BSIZE !=0) {
1155 				VOP_UNLOCK(nd.ni_vp);
1156 				error = EINVAL;
1157 				goto close_and_exit;
1158 			}
1159 			if (sizeof(struct vnd_comp_header) +
1160 			  sizeof(u_int64_t) * vnd->sc_comp_numoffs >
1161 			  vattr.va_size) {
1162 				VOP_UNLOCK(nd.ni_vp);
1163 				error = EINVAL;
1164 				goto close_and_exit;
1165 			}
1166 
1167 			/* set decompressed file size */
1168 			vattr.va_size =
1169 			    ((u_quad_t)vnd->sc_comp_numoffs - 1) *
1170 			     (u_quad_t)vnd->sc_comp_blksz;
1171 
1172 			/* allocate space for all the compressed offsets */
1173 			vnd->sc_comp_offsets =
1174 			malloc(sizeof(u_int64_t) * vnd->sc_comp_numoffs,
1175 			M_DEVBUF, M_WAITOK);
1176 
1177 			/* read in the offsets */
1178 			error = vn_rdwr(UIO_READ, nd.ni_vp,
1179 			  (void *)vnd->sc_comp_offsets,
1180 			  sizeof(u_int64_t) * vnd->sc_comp_numoffs,
1181 			  sizeof(struct vnd_comp_header), UIO_SYSSPACE,
1182 			  IO_UNIT|IO_NODELOCKED, l->l_cred, NULL, NULL);
1183 			if (error) {
1184 				VOP_UNLOCK(nd.ni_vp);
1185 				goto close_and_exit;
1186 			}
1187 			/*
1188 			 * find largest block size (used for allocation limit).
1189 			 * Also convert offset to native byte order.
1190 			 */
1191 			comp_maxsize = 0;
1192 			for (i = 0; i < vnd->sc_comp_numoffs - 1; i++) {
1193 				vnd->sc_comp_offsets[i] =
1194 				  be64toh(vnd->sc_comp_offsets[i]);
1195 				comp_size = be64toh(vnd->sc_comp_offsets[i + 1])
1196 				  - vnd->sc_comp_offsets[i];
1197 				if (comp_size > comp_maxsize)
1198 					comp_maxsize = comp_size;
1199 			}
1200 			vnd->sc_comp_offsets[vnd->sc_comp_numoffs - 1] =
1201 			  be64toh(vnd->sc_comp_offsets[vnd->sc_comp_numoffs - 1]);
1202 
1203 			/* create compressed data buffer */
1204 			vnd->sc_comp_buff = malloc(comp_maxsize,
1205 			  M_DEVBUF, M_WAITOK);
1206 
1207 			/* create decompressed buffer */
1208 			vnd->sc_comp_decombuf = malloc(vnd->sc_comp_blksz,
1209 			  M_DEVBUF, M_WAITOK);
1210 			vnd->sc_comp_buffblk = -1;
1211 
1212 			/* Initialize decompress stream */
1213 			memset(&vnd->sc_comp_stream, 0, sizeof(z_stream));
1214 			vnd->sc_comp_stream.zalloc = vnd_alloc;
1215 			vnd->sc_comp_stream.zfree = vnd_free;
1216 			error = inflateInit2(&vnd->sc_comp_stream, MAX_WBITS);
1217 			if (error) {
1218 				if (vnd->sc_comp_stream.msg)
1219 					printf("vnd%d: compressed file, %s\n",
1220 					  unit, vnd->sc_comp_stream.msg);
1221 				VOP_UNLOCK(nd.ni_vp);
1222 				error = EINVAL;
1223 				goto close_and_exit;
1224 			}
1225 
1226 			vnd->sc_flags |= VNF_COMP | VNF_READONLY;
1227 #else /* !VND_COMPRESSION */
1228 			VOP_UNLOCK(nd.ni_vp);
1229 			error = EOPNOTSUPP;
1230 			goto close_and_exit;
1231 #endif /* VND_COMPRESSION */
1232 		}
1233 
1234 		VOP_UNLOCK(nd.ni_vp);
1235 		vnd->sc_vp = nd.ni_vp;
1236 		vnd->sc_size = btodb(vattr.va_size);	/* note truncation */
1237 
1238 		/*
1239 		 * Use pseudo-geometry specified.  If none was provided,
1240 		 * use "standard" Adaptec fictitious geometry.
1241 		 */
1242 		if (vio->vnd_flags & VNDIOF_HASGEOM) {
1243 
1244 			memcpy(&vnd->sc_geom, &vio->vnd_geom,
1245 			    sizeof(vio->vnd_geom));
1246 
1247 			/*
1248 			 * Sanity-check the sector size.
1249 			 * XXX Don't allow secsize < DEV_BSIZE.	 Should
1250 			 * XXX we?
1251 			 */
1252 			if (vnd->sc_geom.vng_secsize < DEV_BSIZE ||
1253 			    (vnd->sc_geom.vng_secsize % DEV_BSIZE) != 0 ||
1254 			    vnd->sc_geom.vng_ncylinders == 0 ||
1255 			    (vnd->sc_geom.vng_ntracks *
1256 			     vnd->sc_geom.vng_nsectors) == 0) {
1257 				error = EINVAL;
1258 				goto close_and_exit;
1259 			}
1260 
1261 			/*
1262 			 * Compute the size (in DEV_BSIZE blocks) specified
1263 			 * by the geometry.
1264 			 */
1265 			geomsize = (vnd->sc_geom.vng_nsectors *
1266 			    vnd->sc_geom.vng_ntracks *
1267 			    vnd->sc_geom.vng_ncylinders) *
1268 			    (vnd->sc_geom.vng_secsize / DEV_BSIZE);
1269 
1270 			/*
1271 			 * Sanity-check the size against the specified
1272 			 * geometry.
1273 			 */
1274 			if (vnd->sc_size < geomsize) {
1275 				error = EINVAL;
1276 				goto close_and_exit;
1277 			}
1278 		} else if (vnd->sc_size >= (32 * 64)) {
1279 			/*
1280 			 * Size must be at least 2048 DEV_BSIZE blocks
1281 			 * (1M) in order to use this geometry.
1282 			 */
1283 			vnd->sc_geom.vng_secsize = DEV_BSIZE;
1284 			vnd->sc_geom.vng_nsectors = 32;
1285 			vnd->sc_geom.vng_ntracks = 64;
1286 			vnd->sc_geom.vng_ncylinders = vnd->sc_size / (64 * 32);
1287 		} else {
1288 			vnd->sc_geom.vng_secsize = DEV_BSIZE;
1289 			vnd->sc_geom.vng_nsectors = 1;
1290 			vnd->sc_geom.vng_ntracks = 1;
1291 			vnd->sc_geom.vng_ncylinders = vnd->sc_size;
1292 		}
1293 
1294 		vnd_set_properties(vnd);
1295 
1296 		if (vio->vnd_flags & VNDIOF_READONLY) {
1297 			vnd->sc_flags |= VNF_READONLY;
1298 		}
1299 
1300 		if ((error = vndsetcred(vnd, l->l_cred)) != 0)
1301 			goto close_and_exit;
1302 
1303 		vndthrottle(vnd, vnd->sc_vp);
1304 		vio->vnd_osize = dbtob(vnd->sc_size);
1305 #ifdef VNDIOCSET50
1306 		if (cmd != VNDIOCSET50)
1307 #endif
1308 			vio->vnd_size = dbtob(vnd->sc_size);
1309 		vnd->sc_flags |= VNF_INITED;
1310 
1311 		/* create the kernel thread, wait for it to be up */
1312 		error = kthread_create(PRI_NONE, 0, NULL, vndthread, vnd,
1313 		    &vnd->sc_kthread, device_xname(vnd->sc_dev));
1314 		if (error)
1315 			goto close_and_exit;
1316 		while ((vnd->sc_flags & VNF_KTHREAD) == 0) {
1317 			tsleep(&vnd->sc_kthread, PRIBIO, "vndthr", 0);
1318 		}
1319 #ifdef DEBUG
1320 		if (vnddebug & VDB_INIT)
1321 			printf("vndioctl: SET vp %p size 0x%lx %d/%d/%d/%d\n",
1322 			    vnd->sc_vp, (unsigned long) vnd->sc_size,
1323 			    vnd->sc_geom.vng_secsize,
1324 			    vnd->sc_geom.vng_nsectors,
1325 			    vnd->sc_geom.vng_ntracks,
1326 			    vnd->sc_geom.vng_ncylinders);
1327 #endif
1328 
1329 		/* Attach the disk. */
1330 		disk_attach(&vnd->sc_dkdev);
1331 		disk_blocksize(&vnd->sc_dkdev, vnd->sc_geom.vng_secsize);
1332 
1333 		/* Initialize the xfer and buffer pools. */
1334 		pool_init(&vnd->sc_vxpool, sizeof(struct vndxfer), 0,
1335 		    0, 0, "vndxpl", NULL, IPL_BIO);
1336 
1337 		/* Try and read the disklabel. */
1338 		vndgetdisklabel(dev, vnd);
1339 
1340 		vndunlock(vnd);
1341 
1342 		pathbuf_destroy(pb);
1343 		break;
1344 
1345 close_and_exit:
1346 		(void) vn_close(nd.ni_vp, fflags, l->l_cred);
1347 		pathbuf_destroy(pb);
1348 unlock_and_exit:
1349 #ifdef VND_COMPRESSION
1350 		/* free any allocated memory (for compressed file) */
1351 		if (vnd->sc_comp_offsets) {
1352 			free(vnd->sc_comp_offsets, M_DEVBUF);
1353 			vnd->sc_comp_offsets = NULL;
1354 		}
1355 		if (vnd->sc_comp_buff) {
1356 			free(vnd->sc_comp_buff, M_DEVBUF);
1357 			vnd->sc_comp_buff = NULL;
1358 		}
1359 		if (vnd->sc_comp_decombuf) {
1360 			free(vnd->sc_comp_decombuf, M_DEVBUF);
1361 			vnd->sc_comp_decombuf = NULL;
1362 		}
1363 #endif /* VND_COMPRESSION */
1364 		vndunlock(vnd);
1365 		return error;
1366 
1367 #ifdef VNDIOCCLR50
1368 	case VNDIOCCLR50:
1369 #endif
1370 	case VNDIOCCLR:
1371 		part = DISKPART(dev);
1372 		pmask = (1 << part);
1373 		force = (vio->vnd_flags & VNDIOF_FORCE) != 0;
1374 
1375 		if ((error = vnddoclear(vnd, pmask, minor(dev), force)) != 0)
1376 			return error;
1377 
1378 		break;
1379 
1380 #ifdef COMPAT_30
1381 	case VNDIOCGET30: {
1382 		struct vnd_user30 *vnu;
1383 		struct vattr va;
1384 		vnu = (struct vnd_user30 *)data;
1385 		KASSERT(l);
1386 		switch (error = vnd_cget(l, unit, &vnu->vnu_unit, &va)) {
1387 		case 0:
1388 			vnu->vnu_dev = va.va_fsid;
1389 			vnu->vnu_ino = va.va_fileid;
1390 			break;
1391 		case -1:
1392 			/* unused is not an error */
1393 			vnu->vnu_dev = 0;
1394 			vnu->vnu_ino = 0;
1395 			break;
1396 		default:
1397 			return error;
1398 		}
1399 		break;
1400 	}
1401 #endif
1402 
1403 #ifdef COMPAT_50
1404 	case VNDIOCGET50: {
1405 		struct vnd_user50 *vnu;
1406 		struct vattr va;
1407 		vnu = (struct vnd_user50 *)data;
1408 		KASSERT(l);
1409 		switch (error = vnd_cget(l, unit, &vnu->vnu_unit, &va)) {
1410 		case 0:
1411 			vnu->vnu_dev = va.va_fsid;
1412 			vnu->vnu_ino = va.va_fileid;
1413 			break;
1414 		case -1:
1415 			/* unused is not an error */
1416 			vnu->vnu_dev = 0;
1417 			vnu->vnu_ino = 0;
1418 			break;
1419 		default:
1420 			return error;
1421 		}
1422 		break;
1423 	}
1424 #endif
1425 
1426 	case VNDIOCGET: {
1427 		struct vnd_user *vnu;
1428 		struct vattr va;
1429 		vnu = (struct vnd_user *)data;
1430 		KASSERT(l);
1431 		switch (error = vnd_cget(l, unit, &vnu->vnu_unit, &va)) {
1432 		case 0:
1433 			vnu->vnu_dev = va.va_fsid;
1434 			vnu->vnu_ino = va.va_fileid;
1435 			break;
1436 		case -1:
1437 			/* unused is not an error */
1438 			vnu->vnu_dev = 0;
1439 			vnu->vnu_ino = 0;
1440 			break;
1441 		default:
1442 			return error;
1443 		}
1444 		break;
1445 	}
1446 
1447 	case DIOCGDINFO:
1448 		*(struct disklabel *)data = *(vnd->sc_dkdev.dk_label);
1449 		break;
1450 
1451 #ifdef __HAVE_OLD_DISKLABEL
1452 	case ODIOCGDINFO:
1453 		newlabel = *(vnd->sc_dkdev.dk_label);
1454 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1455 			return ENOTTY;
1456 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1457 		break;
1458 #endif
1459 
1460 	case DIOCGPART:
1461 		((struct partinfo *)data)->disklab = vnd->sc_dkdev.dk_label;
1462 		((struct partinfo *)data)->part =
1463 		    &vnd->sc_dkdev.dk_label->d_partitions[DISKPART(dev)];
1464 		break;
1465 
1466 	case DIOCWDINFO:
1467 	case DIOCSDINFO:
1468 #ifdef __HAVE_OLD_DISKLABEL
1469 	case ODIOCWDINFO:
1470 	case ODIOCSDINFO:
1471 #endif
1472 	{
1473 		struct disklabel *lp;
1474 
1475 		if ((error = vndlock(vnd)) != 0)
1476 			return error;
1477 
1478 		vnd->sc_flags |= VNF_LABELLING;
1479 
1480 #ifdef __HAVE_OLD_DISKLABEL
1481 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
1482 			memset(&newlabel, 0, sizeof newlabel);
1483 			memcpy(&newlabel, data, sizeof (struct olddisklabel));
1484 			lp = &newlabel;
1485 		} else
1486 #endif
1487 		lp = (struct disklabel *)data;
1488 
1489 		error = setdisklabel(vnd->sc_dkdev.dk_label,
1490 		    lp, 0, vnd->sc_dkdev.dk_cpulabel);
1491 		if (error == 0) {
1492 			if (cmd == DIOCWDINFO
1493 #ifdef __HAVE_OLD_DISKLABEL
1494 			    || cmd == ODIOCWDINFO
1495 #endif
1496 			   )
1497 				error = writedisklabel(VNDLABELDEV(dev),
1498 				    vndstrategy, vnd->sc_dkdev.dk_label,
1499 				    vnd->sc_dkdev.dk_cpulabel);
1500 		}
1501 
1502 		vnd->sc_flags &= ~VNF_LABELLING;
1503 
1504 		vndunlock(vnd);
1505 
1506 		if (error)
1507 			return error;
1508 		break;
1509 	}
1510 
1511 	case DIOCKLABEL:
1512 		if (*(int *)data != 0)
1513 			vnd->sc_flags |= VNF_KLABEL;
1514 		else
1515 			vnd->sc_flags &= ~VNF_KLABEL;
1516 		break;
1517 
1518 	case DIOCWLABEL:
1519 		if (*(int *)data != 0)
1520 			vnd->sc_flags |= VNF_WLABEL;
1521 		else
1522 			vnd->sc_flags &= ~VNF_WLABEL;
1523 		break;
1524 
1525 	case DIOCGDEFLABEL:
1526 		vndgetdefaultlabel(vnd, (struct disklabel *)data);
1527 		break;
1528 
1529 #ifdef __HAVE_OLD_DISKLABEL
1530 	case ODIOCGDEFLABEL:
1531 		vndgetdefaultlabel(vnd, &newlabel);
1532 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1533 			return ENOTTY;
1534 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1535 		break;
1536 #endif
1537 
1538 	case DIOCCACHESYNC:
1539 		vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1540 		error = VOP_FSYNC(vnd->sc_vp, vnd->sc_cred,
1541 		    FSYNC_WAIT | FSYNC_DATAONLY | FSYNC_CACHE, 0, 0);
1542 		VOP_UNLOCK(vnd->sc_vp);
1543 		return error;
1544 
1545 	case DIOCAWEDGE:
1546 		dkw = (void *) data;
1547 
1548 		if ((flag & FWRITE) == 0)
1549 			return EBADF;
1550 
1551 		/* If the ioctl happens here, the parent is us. */
1552 		strlcpy(dkw->dkw_parent, device_xname(vnd->sc_dev),
1553 		    sizeof(dkw->dkw_parent));
1554 		return dkwedge_add(dkw);
1555 
1556 	case DIOCDWEDGE:
1557 		dkw = (void *) data;
1558 
1559 		if ((flag & FWRITE) == 0)
1560 			return EBADF;
1561 
1562 		/* If the ioctl happens here, the parent is us. */
1563 		strlcpy(dkw->dkw_parent, device_xname(vnd->sc_dev),
1564 		    sizeof(dkw->dkw_parent));
1565 		return dkwedge_del(dkw);
1566 
1567 	case DIOCLWEDGES:
1568 		dkwl = (void *) data;
1569 
1570 		return dkwedge_list(&vnd->sc_dkdev, dkwl, l);
1571 
1572 	default:
1573 		return ENOTTY;
1574 	}
1575 
1576 	return 0;
1577 }
1578 
1579 /*
1580  * Duplicate the current processes' credentials.  Since we are called only
1581  * as the result of a SET ioctl and only root can do that, any future access
1582  * to this "disk" is essentially as root.  Note that credentials may change
1583  * if some other uid can write directly to the mapped file (NFS).
1584  */
1585 static int
1586 vndsetcred(struct vnd_softc *vnd, kauth_cred_t cred)
1587 {
1588 	struct uio auio;
1589 	struct iovec aiov;
1590 	char *tmpbuf;
1591 	int error;
1592 
1593 	vnd->sc_cred = kauth_cred_dup(cred);
1594 	tmpbuf = malloc(DEV_BSIZE, M_TEMP, M_WAITOK);
1595 
1596 	/* XXX: Horrible kludge to establish credentials for NFS */
1597 	aiov.iov_base = tmpbuf;
1598 	aiov.iov_len = min(DEV_BSIZE, dbtob(vnd->sc_size));
1599 	auio.uio_iov = &aiov;
1600 	auio.uio_iovcnt = 1;
1601 	auio.uio_offset = 0;
1602 	auio.uio_rw = UIO_READ;
1603 	auio.uio_resid = aiov.iov_len;
1604 	UIO_SETUP_SYSSPACE(&auio);
1605 	vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1606 	error = VOP_READ(vnd->sc_vp, &auio, 0, vnd->sc_cred);
1607 	if (error == 0) {
1608 		/*
1609 		 * Because vnd does all IO directly through the vnode
1610 		 * we need to flush (at least) the buffer from the above
1611 		 * VOP_READ from the buffer cache to prevent cache
1612 		 * incoherencies.  Also, be careful to write dirty
1613 		 * buffers back to stable storage.
1614 		 */
1615 		error = vinvalbuf(vnd->sc_vp, V_SAVE, vnd->sc_cred,
1616 			    curlwp, 0, 0);
1617 	}
1618 	VOP_UNLOCK(vnd->sc_vp);
1619 
1620 	free(tmpbuf, M_TEMP);
1621 	return error;
1622 }
1623 
1624 /*
1625  * Set maxactive based on FS type
1626  */
1627 static void
1628 vndthrottle(struct vnd_softc *vnd, struct vnode *vp)
1629 {
1630 
1631 	if (vp->v_tag == VT_NFS)
1632 		vnd->sc_maxactive = 2;
1633 	else
1634 		vnd->sc_maxactive = 8;
1635 
1636 	if (vnd->sc_maxactive < 1)
1637 		vnd->sc_maxactive = 1;
1638 }
1639 
1640 #if 0
1641 static void
1642 vndshutdown(void)
1643 {
1644 	struct vnd_softc *vnd;
1645 
1646 	for (vnd = &vnd_softc[0]; vnd < &vnd_softc[numvnd]; vnd++)
1647 		if (vnd->sc_flags & VNF_INITED)
1648 			vndclear(vnd);
1649 }
1650 #endif
1651 
1652 static void
1653 vndclear(struct vnd_softc *vnd, int myminor)
1654 {
1655 	struct vnode *vp = vnd->sc_vp;
1656 	int fflags = FREAD;
1657 	int bmaj, cmaj, i, mn;
1658 	int s;
1659 
1660 #ifdef DEBUG
1661 	if (vnddebug & VDB_FOLLOW)
1662 		printf("vndclear(%p): vp %p\n", vnd, vp);
1663 #endif
1664 	/* locate the major number */
1665 	bmaj = bdevsw_lookup_major(&vnd_bdevsw);
1666 	cmaj = cdevsw_lookup_major(&vnd_cdevsw);
1667 
1668 	/* Nuke the vnodes for any open instances */
1669 	for (i = 0; i < MAXPARTITIONS; i++) {
1670 		mn = DISKMINOR(device_unit(vnd->sc_dev), i);
1671 		vdevgone(bmaj, mn, mn, VBLK);
1672 		if (mn != myminor) /* XXX avoid to kill own vnode */
1673 			vdevgone(cmaj, mn, mn, VCHR);
1674 	}
1675 
1676 	if ((vnd->sc_flags & VNF_READONLY) == 0)
1677 		fflags |= FWRITE;
1678 
1679 	s = splbio();
1680 	bufq_drain(vnd->sc_tab);
1681 	splx(s);
1682 
1683 	vnd->sc_flags |= VNF_VUNCONF;
1684 	wakeup(&vnd->sc_tab);
1685 	while (vnd->sc_flags & VNF_KTHREAD)
1686 		tsleep(&vnd->sc_kthread, PRIBIO, "vnthr", 0);
1687 
1688 #ifdef VND_COMPRESSION
1689 	/* free the compressed file buffers */
1690 	if (vnd->sc_flags & VNF_COMP) {
1691 		if (vnd->sc_comp_offsets) {
1692 			free(vnd->sc_comp_offsets, M_DEVBUF);
1693 			vnd->sc_comp_offsets = NULL;
1694 		}
1695 		if (vnd->sc_comp_buff) {
1696 			free(vnd->sc_comp_buff, M_DEVBUF);
1697 			vnd->sc_comp_buff = NULL;
1698 		}
1699 		if (vnd->sc_comp_decombuf) {
1700 			free(vnd->sc_comp_decombuf, M_DEVBUF);
1701 			vnd->sc_comp_decombuf = NULL;
1702 		}
1703 	}
1704 #endif /* VND_COMPRESSION */
1705 	vnd->sc_flags &=
1706 	    ~(VNF_INITED | VNF_READONLY | VNF_VLABEL
1707 	      | VNF_VUNCONF | VNF_COMP | VNF_CLEARING);
1708 	if (vp == NULL)
1709 		panic("vndclear: null vp");
1710 	(void) vn_close(vp, fflags, vnd->sc_cred);
1711 	kauth_cred_free(vnd->sc_cred);
1712 	vnd->sc_vp = NULL;
1713 	vnd->sc_cred = NULL;
1714 	vnd->sc_size = 0;
1715 }
1716 
1717 static int
1718 vndsize(dev_t dev)
1719 {
1720 	struct vnd_softc *sc;
1721 	struct disklabel *lp;
1722 	int part, unit, omask;
1723 	int size;
1724 
1725 	unit = vndunit(dev);
1726 	sc = device_lookup_private(&vnd_cd, unit);
1727 	if (sc == NULL)
1728 		return -1;
1729 
1730 	if ((sc->sc_flags & VNF_INITED) == 0)
1731 		return -1;
1732 
1733 	part = DISKPART(dev);
1734 	omask = sc->sc_dkdev.dk_openmask & (1 << part);
1735 	lp = sc->sc_dkdev.dk_label;
1736 
1737 	if (omask == 0 && vndopen(dev, 0, S_IFBLK, curlwp))	/* XXX */
1738 		return -1;
1739 
1740 	if (lp->d_partitions[part].p_fstype != FS_SWAP)
1741 		size = -1;
1742 	else
1743 		size = lp->d_partitions[part].p_size *
1744 		    (lp->d_secsize / DEV_BSIZE);
1745 
1746 	if (omask == 0 && vndclose(dev, 0, S_IFBLK, curlwp))	/* XXX */
1747 		return -1;
1748 
1749 	return size;
1750 }
1751 
1752 static int
1753 vnddump(dev_t dev, daddr_t blkno, void *va,
1754     size_t size)
1755 {
1756 
1757 	/* Not implemented. */
1758 	return ENXIO;
1759 }
1760 
1761 static void
1762 vndgetdefaultlabel(struct vnd_softc *sc, struct disklabel *lp)
1763 {
1764 	struct vndgeom *vng = &sc->sc_geom;
1765 	struct partition *pp;
1766 
1767 	memset(lp, 0, sizeof(*lp));
1768 
1769 	lp->d_secperunit = sc->sc_size / (vng->vng_secsize / DEV_BSIZE);
1770 	lp->d_secsize = vng->vng_secsize;
1771 	lp->d_nsectors = vng->vng_nsectors;
1772 	lp->d_ntracks = vng->vng_ntracks;
1773 	lp->d_ncylinders = vng->vng_ncylinders;
1774 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1775 
1776 	strncpy(lp->d_typename, "vnd", sizeof(lp->d_typename));
1777 	lp->d_type = DTYPE_VND;
1778 	strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
1779 	lp->d_rpm = 3600;
1780 	lp->d_interleave = 1;
1781 	lp->d_flags = 0;
1782 
1783 	pp = &lp->d_partitions[RAW_PART];
1784 	pp->p_offset = 0;
1785 	pp->p_size = lp->d_secperunit;
1786 	pp->p_fstype = FS_UNUSED;
1787 	lp->d_npartitions = RAW_PART + 1;
1788 
1789 	lp->d_magic = DISKMAGIC;
1790 	lp->d_magic2 = DISKMAGIC;
1791 	lp->d_checksum = dkcksum(lp);
1792 }
1793 
1794 /*
1795  * Read the disklabel from a vnd.  If one is not present, create a fake one.
1796  */
1797 static void
1798 vndgetdisklabel(dev_t dev, struct vnd_softc *sc)
1799 {
1800 	const char *errstring;
1801 	struct disklabel *lp = sc->sc_dkdev.dk_label;
1802 	struct cpu_disklabel *clp = sc->sc_dkdev.dk_cpulabel;
1803 	int i;
1804 
1805 	memset(clp, 0, sizeof(*clp));
1806 
1807 	vndgetdefaultlabel(sc, lp);
1808 
1809 	/*
1810 	 * Call the generic disklabel extraction routine.
1811 	 */
1812 	errstring = readdisklabel(VNDLABELDEV(dev), vndstrategy, lp, clp);
1813 	if (errstring) {
1814 		/*
1815 		 * Lack of disklabel is common, but we print the warning
1816 		 * anyway, since it might contain other useful information.
1817 		 */
1818 		aprint_normal_dev(sc->sc_dev, "%s\n", errstring);
1819 
1820 		/*
1821 		 * For historical reasons, if there's no disklabel
1822 		 * present, all partitions must be FS_BSDFFS and
1823 		 * occupy the entire disk.
1824 		 */
1825 		for (i = 0; i < MAXPARTITIONS; i++) {
1826 			/*
1827 			 * Don't wipe out port specific hack (such as
1828 			 * dos partition hack of i386 port).
1829 			 */
1830 			if (lp->d_partitions[i].p_size != 0)
1831 				continue;
1832 
1833 			lp->d_partitions[i].p_size = lp->d_secperunit;
1834 			lp->d_partitions[i].p_offset = 0;
1835 			lp->d_partitions[i].p_fstype = FS_BSDFFS;
1836 		}
1837 
1838 		strncpy(lp->d_packname, "default label",
1839 		    sizeof(lp->d_packname));
1840 
1841 		lp->d_npartitions = MAXPARTITIONS;
1842 		lp->d_checksum = dkcksum(lp);
1843 	}
1844 
1845 	/* In-core label now valid. */
1846 	sc->sc_flags |= VNF_VLABEL;
1847 }
1848 
1849 /*
1850  * Wait interruptibly for an exclusive lock.
1851  *
1852  * XXX
1853  * Several drivers do this; it should be abstracted and made MP-safe.
1854  */
1855 static int
1856 vndlock(struct vnd_softc *sc)
1857 {
1858 	int error;
1859 
1860 	while ((sc->sc_flags & VNF_LOCKED) != 0) {
1861 		sc->sc_flags |= VNF_WANTED;
1862 		if ((error = tsleep(sc, PRIBIO | PCATCH, "vndlck", 0)) != 0)
1863 			return error;
1864 	}
1865 	sc->sc_flags |= VNF_LOCKED;
1866 	return 0;
1867 }
1868 
1869 /*
1870  * Unlock and wake up any waiters.
1871  */
1872 static void
1873 vndunlock(struct vnd_softc *sc)
1874 {
1875 
1876 	sc->sc_flags &= ~VNF_LOCKED;
1877 	if ((sc->sc_flags & VNF_WANTED) != 0) {
1878 		sc->sc_flags &= ~VNF_WANTED;
1879 		wakeup(sc);
1880 	}
1881 }
1882 
1883 #ifdef VND_COMPRESSION
1884 /* compressed file read */
1885 static void
1886 compstrategy(struct buf *bp, off_t bn)
1887 {
1888 	int error;
1889 	int unit = vndunit(bp->b_dev);
1890 	struct vnd_softc *vnd =
1891 	    device_lookup_private(&vnd_cd, unit);
1892 	u_int32_t comp_block;
1893 	struct uio auio;
1894 	char *addr;
1895 	int s;
1896 
1897 	/* set up constants for data move */
1898 	auio.uio_rw = UIO_READ;
1899 	UIO_SETUP_SYSSPACE(&auio);
1900 
1901 	/* read, and transfer the data */
1902 	addr = bp->b_data;
1903 	bp->b_resid = bp->b_bcount;
1904 	s = splbio();
1905 	while (bp->b_resid > 0) {
1906 		unsigned length;
1907 		size_t length_in_buffer;
1908 		u_int32_t offset_in_buffer;
1909 		struct iovec aiov;
1910 
1911 		/* calculate the compressed block number */
1912 		comp_block = bn / (off_t)vnd->sc_comp_blksz;
1913 
1914 		/* check for good block number */
1915 		if (comp_block >= vnd->sc_comp_numoffs) {
1916 			bp->b_error = EINVAL;
1917 			splx(s);
1918 			return;
1919 		}
1920 
1921 		/* read in the compressed block, if not in buffer */
1922 		if (comp_block != vnd->sc_comp_buffblk) {
1923 			length = vnd->sc_comp_offsets[comp_block + 1] -
1924 			    vnd->sc_comp_offsets[comp_block];
1925 			vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1926 			error = vn_rdwr(UIO_READ, vnd->sc_vp, vnd->sc_comp_buff,
1927 			    length, vnd->sc_comp_offsets[comp_block],
1928 			    UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, vnd->sc_cred,
1929 			    NULL, NULL);
1930 			if (error) {
1931 				bp->b_error = error;
1932 				VOP_UNLOCK(vnd->sc_vp);
1933 				splx(s);
1934 				return;
1935 			}
1936 			/* uncompress the buffer */
1937 			vnd->sc_comp_stream.next_in = vnd->sc_comp_buff;
1938 			vnd->sc_comp_stream.avail_in = length;
1939 			vnd->sc_comp_stream.next_out = vnd->sc_comp_decombuf;
1940 			vnd->sc_comp_stream.avail_out = vnd->sc_comp_blksz;
1941 			inflateReset(&vnd->sc_comp_stream);
1942 			error = inflate(&vnd->sc_comp_stream, Z_FINISH);
1943 			if (error != Z_STREAM_END) {
1944 				if (vnd->sc_comp_stream.msg)
1945 					aprint_normal_dev(vnd->sc_dev,
1946 					    "compressed file, %s\n",
1947 					    vnd->sc_comp_stream.msg);
1948 				bp->b_error = EBADMSG;
1949 				VOP_UNLOCK(vnd->sc_vp);
1950 				splx(s);
1951 				return;
1952 			}
1953 			vnd->sc_comp_buffblk = comp_block;
1954 			VOP_UNLOCK(vnd->sc_vp);
1955 		}
1956 
1957 		/* transfer the usable uncompressed data */
1958 		offset_in_buffer = bn % (off_t)vnd->sc_comp_blksz;
1959 		length_in_buffer = vnd->sc_comp_blksz - offset_in_buffer;
1960 		if (length_in_buffer > bp->b_resid)
1961 			length_in_buffer = bp->b_resid;
1962 		auio.uio_iov = &aiov;
1963 		auio.uio_iovcnt = 1;
1964 		aiov.iov_base = addr;
1965 		aiov.iov_len = length_in_buffer;
1966 		auio.uio_resid = aiov.iov_len;
1967 		auio.uio_offset = 0;
1968 		error = uiomove(vnd->sc_comp_decombuf + offset_in_buffer,
1969 		    length_in_buffer, &auio);
1970 		if (error) {
1971 			bp->b_error = error;
1972 			splx(s);
1973 			return;
1974 		}
1975 
1976 		bn += length_in_buffer;
1977 		addr += length_in_buffer;
1978 		bp->b_resid -= length_in_buffer;
1979 	}
1980 	splx(s);
1981 }
1982 
1983 /* compression memory allocation routines */
1984 static void *
1985 vnd_alloc(void *aux, u_int items, u_int siz)
1986 {
1987 	return malloc(items * siz, M_TEMP, M_NOWAIT);
1988 }
1989 
1990 static void
1991 vnd_free(void *aux, void *ptr)
1992 {
1993 	free(ptr, M_TEMP);
1994 }
1995 #endif /* VND_COMPRESSION */
1996 
1997 static void
1998 vnd_set_properties(struct vnd_softc *vnd)
1999 {
2000 	prop_dictionary_t disk_info, odisk_info, geom;
2001 
2002 	disk_info = prop_dictionary_create();
2003 
2004 	geom = prop_dictionary_create();
2005 
2006 	prop_dictionary_set_uint64(geom, "sectors-per-unit",
2007 	    vnd->sc_geom.vng_nsectors * vnd->sc_geom.vng_ntracks *
2008 	    vnd->sc_geom.vng_ncylinders);
2009 
2010 	prop_dictionary_set_uint32(geom, "sector-size",
2011 	    vnd->sc_geom.vng_secsize);
2012 
2013 	prop_dictionary_set_uint16(geom, "sectors-per-track",
2014 	    vnd->sc_geom.vng_nsectors);
2015 
2016 	prop_dictionary_set_uint16(geom, "tracks-per-cylinder",
2017 	    vnd->sc_geom.vng_ntracks);
2018 
2019 	prop_dictionary_set_uint64(geom, "cylinders-per-unit",
2020 	    vnd->sc_geom.vng_ncylinders);
2021 
2022 	prop_dictionary_set(disk_info, "geometry", geom);
2023 	prop_object_release(geom);
2024 
2025 	prop_dictionary_set(device_properties(vnd->sc_dev),
2026 	    "disk-info", disk_info);
2027 
2028 	/*
2029 	 * Don't release disk_info here; we keep a reference to it.
2030 	 * disk_detach() will release it when we go away.
2031 	 */
2032 
2033 	odisk_info = vnd->sc_dkdev.dk_info;
2034 	vnd->sc_dkdev.dk_info = disk_info;
2035 	if (odisk_info)
2036 		prop_object_release(odisk_info);
2037 }
2038 
2039 #ifdef _MODULE
2040 
2041 #include <sys/module.h>
2042 
2043 MODULE(MODULE_CLASS_DRIVER, vnd, NULL);
2044 CFDRIVER_DECL(vnd, DV_DISK, NULL);
2045 
2046 static int
2047 vnd_modcmd(modcmd_t cmd, void *arg)
2048 {
2049 	int bmajor = -1, cmajor = -1,  error = 0;
2050 
2051 	switch (cmd) {
2052 	case MODULE_CMD_INIT:
2053 		error = config_cfdriver_attach(&vnd_cd);
2054 		if (error)
2055 			break;
2056 
2057 		error = config_cfattach_attach(vnd_cd.cd_name, &vnd_ca);
2058 	        if (error) {
2059 			config_cfdriver_detach(&vnd_cd);
2060 			aprint_error("%s: unable to register cfattach\n",
2061 			    vnd_cd.cd_name);
2062 			break;
2063 		}
2064 
2065 		error = devsw_attach("vnd", &vnd_bdevsw, &bmajor,
2066 		    &vnd_cdevsw, &cmajor);
2067 		if (error) {
2068 			config_cfattach_detach(vnd_cd.cd_name, &vnd_ca);
2069 			config_cfdriver_detach(&vnd_cd);
2070 			break;
2071 		}
2072 
2073 		break;
2074 
2075 	case MODULE_CMD_FINI:
2076 		error = config_cfattach_detach(vnd_cd.cd_name, &vnd_ca);
2077 		if (error)
2078 			break;
2079 		config_cfdriver_detach(&vnd_cd);
2080 		devsw_detach(&vnd_bdevsw, &vnd_cdevsw);
2081 		break;
2082 
2083 	case MODULE_CMD_STAT:
2084 		return ENOTTY;
2085 
2086 	default:
2087 		return ENOTTY;
2088 	}
2089 
2090 	return error;
2091 }
2092 
2093 #endif
2094