xref: /netbsd-src/sys/dev/vnd.c (revision ba65fde2d7fefa7d39838fa5fa855e62bd606b5e)
1 /*	$NetBSD: vnd.c,v 1.221 2012/06/09 06:20:45 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 1.221 2012/06/09 06:20:45 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 #include <prop/proplib.h>
131 
132 #if defined(VNDDEBUG) && !defined(DEBUG)
133 #define DEBUG
134 #endif
135 
136 #ifdef DEBUG
137 int dovndcluster = 1;
138 #define VDB_FOLLOW	0x01
139 #define VDB_INIT	0x02
140 #define VDB_IO		0x04
141 #define VDB_LABEL	0x08
142 int vnddebug = 0x00;
143 #endif
144 
145 #define vndunit(x)	DISKUNIT(x)
146 
147 struct vndxfer {
148 	struct buf vx_buf;
149 	struct vnd_softc *vx_vnd;
150 };
151 #define	VND_BUFTOXFER(bp)	((struct vndxfer *)(void *)bp)
152 
153 #define VND_GETXFER(vnd)	pool_get(&(vnd)->sc_vxpool, PR_WAITOK)
154 #define VND_PUTXFER(vnd, vx)	pool_put(&(vnd)->sc_vxpool, (vx))
155 
156 #define VNDLABELDEV(dev) \
157     (MAKEDISKDEV(major((dev)), vndunit((dev)), RAW_PART))
158 
159 #define	VND_MAXPENDING(vnd)	((vnd)->sc_maxactive * 4)
160 
161 /* called by main() at boot time */
162 void	vndattach(int);
163 
164 static void	vndclear(struct vnd_softc *, int);
165 static int	vnddoclear(struct vnd_softc *, int, int, bool);
166 static int	vndsetcred(struct vnd_softc *, kauth_cred_t);
167 static void	vndthrottle(struct vnd_softc *, struct vnode *);
168 static void	vndiodone(struct buf *);
169 #if 0
170 static void	vndshutdown(void);
171 #endif
172 
173 static void	vndgetdefaultlabel(struct vnd_softc *, struct disklabel *);
174 static void	vndgetdisklabel(dev_t, struct vnd_softc *);
175 
176 static int	vndlock(struct vnd_softc *);
177 static void	vndunlock(struct vnd_softc *);
178 #ifdef VND_COMPRESSION
179 static void	compstrategy(struct buf *, off_t);
180 static void	*vnd_alloc(void *, u_int, u_int);
181 static void	vnd_free(void *, void *);
182 #endif /* VND_COMPRESSION */
183 
184 static void	vndthread(void *);
185 static bool	vnode_has_op(const struct vnode *, int);
186 static void	handle_with_rdwr(struct vnd_softc *, const struct buf *,
187 		    struct buf *);
188 static void	handle_with_strategy(struct vnd_softc *, const struct buf *,
189 		    struct buf *);
190 static void	vnd_set_properties(struct vnd_softc *);
191 
192 static dev_type_open(vndopen);
193 static dev_type_close(vndclose);
194 static dev_type_read(vndread);
195 static dev_type_write(vndwrite);
196 static dev_type_ioctl(vndioctl);
197 static dev_type_strategy(vndstrategy);
198 static dev_type_dump(vnddump);
199 static dev_type_size(vndsize);
200 
201 const struct bdevsw vnd_bdevsw = {
202 	vndopen, vndclose, vndstrategy, vndioctl, vnddump, vndsize, D_DISK
203 };
204 
205 const struct cdevsw vnd_cdevsw = {
206 	vndopen, vndclose, vndread, vndwrite, vndioctl,
207 	nostop, notty, nopoll, nommap, nokqfilter, D_DISK
208 };
209 
210 static int	vnd_match(device_t, cfdata_t, void *);
211 static void	vnd_attach(device_t, device_t, void *);
212 static int	vnd_detach(device_t, int);
213 
214 CFATTACH_DECL3_NEW(vnd, sizeof(struct vnd_softc),
215     vnd_match, vnd_attach, vnd_detach, NULL, NULL, NULL, DVF_DETACH_SHUTDOWN);
216 extern struct cfdriver vnd_cd;
217 
218 static struct vnd_softc	*vnd_spawn(int);
219 int	vnd_destroy(device_t);
220 
221 static struct	dkdriver vnddkdriver = { vndstrategy, minphys };
222 
223 void
224 vndattach(int num)
225 {
226 	int error;
227 
228 	error = config_cfattach_attach(vnd_cd.cd_name, &vnd_ca);
229 	if (error)
230 		aprint_error("%s: unable to register cfattach\n",
231 		    vnd_cd.cd_name);
232 }
233 
234 static int
235 vnd_match(device_t self, cfdata_t cfdata, void *aux)
236 {
237 
238 	return 1;
239 }
240 
241 static void
242 vnd_attach(device_t parent, device_t self, void *aux)
243 {
244 	struct vnd_softc *sc = device_private(self);
245 
246 	sc->sc_dev = self;
247 	sc->sc_comp_offsets = NULL;
248 	sc->sc_comp_buff = NULL;
249 	sc->sc_comp_decombuf = NULL;
250 	bufq_alloc(&sc->sc_tab, "disksort", BUFQ_SORT_RAWBLOCK);
251 	disk_init(&sc->sc_dkdev, device_xname(self), &vnddkdriver);
252 	if (!pmf_device_register(self, NULL, NULL))
253 		aprint_error_dev(self, "couldn't establish power handler\n");
254 }
255 
256 static int
257 vnd_detach(device_t self, int flags)
258 {
259 	int error;
260 	struct vnd_softc *sc = device_private(self);
261 
262 	if (sc->sc_flags & VNF_INITED) {
263 		error = vnddoclear(sc, 0, -1, (flags & DETACH_FORCE) != 0);
264 		if (error != 0)
265 			return error;
266 	}
267 
268 	pmf_device_deregister(self);
269 	bufq_free(sc->sc_tab);
270 	disk_destroy(&sc->sc_dkdev);
271 
272 	return 0;
273 }
274 
275 static struct vnd_softc *
276 vnd_spawn(int unit)
277 {
278 	cfdata_t cf;
279 
280 	cf = malloc(sizeof(*cf), M_DEVBUF, M_WAITOK);
281 	cf->cf_name = vnd_cd.cd_name;
282 	cf->cf_atname = vnd_cd.cd_name;
283 	cf->cf_unit = unit;
284 	cf->cf_fstate = FSTATE_STAR;
285 
286 	return device_private(config_attach_pseudo(cf));
287 }
288 
289 int
290 vnd_destroy(device_t dev)
291 {
292 	int error;
293 	cfdata_t cf;
294 
295 	cf = device_cfdata(dev);
296 	error = config_detach(dev, DETACH_QUIET);
297 	if (error)
298 		return error;
299 	free(cf, M_DEVBUF);
300 	return 0;
301 }
302 
303 static int
304 vndopen(dev_t dev, int flags, int mode, struct lwp *l)
305 {
306 	int unit = vndunit(dev);
307 	struct vnd_softc *sc;
308 	int error = 0, part, pmask;
309 	struct disklabel *lp;
310 
311 #ifdef DEBUG
312 	if (vnddebug & VDB_FOLLOW)
313 		printf("vndopen(0x%"PRIx64", 0x%x, 0x%x, %p)\n", dev, flags, mode, l);
314 #endif
315 	sc = device_lookup_private(&vnd_cd, unit);
316 	if (sc == NULL) {
317 		sc = vnd_spawn(unit);
318 		if (sc == NULL)
319 			return ENOMEM;
320 	}
321 
322 	if ((error = vndlock(sc)) != 0)
323 		return error;
324 
325 	if ((sc->sc_flags & VNF_CLEARING) != 0) {
326 		error = ENXIO;
327 		goto done;
328 	}
329 
330 	lp = sc->sc_dkdev.dk_label;
331 
332 	part = DISKPART(dev);
333 	pmask = (1 << part);
334 
335 	if (sc->sc_dkdev.dk_nwedges != 0 && part != RAW_PART) {
336 		error = EBUSY;
337 		goto done;
338 	}
339 
340 	if (sc->sc_flags & VNF_INITED) {
341 		if ((sc->sc_dkdev.dk_openmask & ~(1<<RAW_PART)) != 0) {
342 			/*
343 			 * If any non-raw partition is open, but the disk
344 			 * has been invalidated, disallow further opens.
345 			 */
346 			if ((sc->sc_flags & VNF_VLABEL) == 0) {
347 				error = EIO;
348 				goto done;
349 			}
350 		} else {
351 			/*
352 			 * Load the partition info if not already loaded.
353 			 */
354 			if ((sc->sc_flags & VNF_VLABEL) == 0) {
355 				sc->sc_flags |= VNF_VLABEL;
356 				vndgetdisklabel(dev, sc);
357 			}
358 		}
359 	}
360 
361 	/* Check that the partitions exists. */
362 	if (part != RAW_PART) {
363 		if (((sc->sc_flags & VNF_INITED) == 0) ||
364 		    ((part >= lp->d_npartitions) ||
365 		     (lp->d_partitions[part].p_fstype == FS_UNUSED))) {
366 			error = ENXIO;
367 			goto done;
368 		}
369 	}
370 
371 	/* Prevent our unit from being unconfigured while open. */
372 	switch (mode) {
373 	case S_IFCHR:
374 		sc->sc_dkdev.dk_copenmask |= pmask;
375 		break;
376 
377 	case S_IFBLK:
378 		sc->sc_dkdev.dk_bopenmask |= pmask;
379 		break;
380 	}
381 	sc->sc_dkdev.dk_openmask =
382 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
383 
384  done:
385 	vndunlock(sc);
386 	return error;
387 }
388 
389 static int
390 vndclose(dev_t dev, int flags, int mode, struct lwp *l)
391 {
392 	int unit = vndunit(dev);
393 	struct vnd_softc *sc;
394 	int error = 0, part;
395 
396 #ifdef DEBUG
397 	if (vnddebug & VDB_FOLLOW)
398 		printf("vndclose(0x%"PRIx64", 0x%x, 0x%x, %p)\n", dev, flags, mode, l);
399 #endif
400 	sc = device_lookup_private(&vnd_cd, unit);
401 	if (sc == NULL)
402 		return ENXIO;
403 
404 	if ((error = vndlock(sc)) != 0)
405 		return error;
406 
407 	part = DISKPART(dev);
408 
409 	/* ...that much closer to allowing unconfiguration... */
410 	switch (mode) {
411 	case S_IFCHR:
412 		sc->sc_dkdev.dk_copenmask &= ~(1 << part);
413 		break;
414 
415 	case S_IFBLK:
416 		sc->sc_dkdev.dk_bopenmask &= ~(1 << part);
417 		break;
418 	}
419 	sc->sc_dkdev.dk_openmask =
420 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
421 
422 	vndunlock(sc);
423 
424 	if ((sc->sc_flags & VNF_INITED) == 0) {
425 		if ((error = vnd_destroy(sc->sc_dev)) != 0) {
426 			aprint_error_dev(sc->sc_dev,
427 			    "unable to detach instance\n");
428 			return error;
429 		}
430 	}
431 
432 	return 0;
433 }
434 
435 /*
436  * Queue the request, and wakeup the kernel thread to handle it.
437  */
438 static void
439 vndstrategy(struct buf *bp)
440 {
441 	int unit = vndunit(bp->b_dev);
442 	struct vnd_softc *vnd =
443 	    device_lookup_private(&vnd_cd, unit);
444 	struct disklabel *lp;
445 	daddr_t blkno;
446 	int s = splbio();
447 
448 	if (vnd == NULL) {
449 		bp->b_error = ENXIO;
450 		goto done;
451 	}
452 	lp = vnd->sc_dkdev.dk_label;
453 
454 	if ((vnd->sc_flags & VNF_INITED) == 0) {
455 		bp->b_error = ENXIO;
456 		goto done;
457 	}
458 
459 	/*
460 	 * The transfer must be a whole number of blocks.
461 	 */
462 	if ((bp->b_bcount % lp->d_secsize) != 0) {
463 		bp->b_error = EINVAL;
464 		goto done;
465 	}
466 
467 	/*
468 	 * check if we're read-only.
469 	 */
470 	if ((vnd->sc_flags & VNF_READONLY) && !(bp->b_flags & B_READ)) {
471 		bp->b_error = EACCES;
472 		goto done;
473 	}
474 
475 	/* If it's a nil transfer, wake up the top half now. */
476 	if (bp->b_bcount == 0) {
477 		goto done;
478 	}
479 
480 	/*
481 	 * Do bounds checking and adjust transfer.  If there's an error,
482 	 * the bounds check will flag that for us.
483 	 */
484 	if (DISKPART(bp->b_dev) == RAW_PART) {
485 		if (bounds_check_with_mediasize(bp, DEV_BSIZE,
486 		    vnd->sc_size) <= 0)
487 			goto done;
488 	} else {
489 		if (bounds_check_with_label(&vnd->sc_dkdev,
490 		    bp, vnd->sc_flags & (VNF_WLABEL|VNF_LABELLING)) <= 0)
491 			goto done;
492 	}
493 
494 	/*
495 	 * Put the block number in terms of the logical blocksize
496 	 * of the "device".
497 	 */
498 
499 	blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
500 
501 	/*
502 	 * Translate the partition-relative block number to an absolute.
503 	 */
504 	if (DISKPART(bp->b_dev) != RAW_PART) {
505 		struct partition *pp;
506 
507 		pp = &vnd->sc_dkdev.dk_label->d_partitions[
508 		    DISKPART(bp->b_dev)];
509 		blkno += pp->p_offset;
510 	}
511 	bp->b_rawblkno = blkno;
512 
513 #ifdef DEBUG
514 	if (vnddebug & VDB_FOLLOW)
515 		printf("vndstrategy(%p): unit %d\n", bp, unit);
516 #endif
517 	if ((vnd->sc_flags & VNF_USE_VN_RDWR)) {
518 		KASSERT(vnd->sc_pending >= 0 &&
519 		    vnd->sc_pending <= VND_MAXPENDING(vnd));
520 		while (vnd->sc_pending == VND_MAXPENDING(vnd))
521 			tsleep(&vnd->sc_pending, PRIBIO, "vndpc", 0);
522 		vnd->sc_pending++;
523 	}
524 	bufq_put(vnd->sc_tab, bp);
525 	wakeup(&vnd->sc_tab);
526 	splx(s);
527 	return;
528 
529 done:
530 	bp->b_resid = bp->b_bcount;
531 	biodone(bp);
532 	splx(s);
533 }
534 
535 static bool
536 vnode_has_strategy(struct vnd_softc *vnd)
537 {
538 	return vnode_has_op(vnd->sc_vp, VOFFSET(vop_bmap)) &&
539 	    vnode_has_op(vnd->sc_vp, VOFFSET(vop_strategy));
540 }
541 
542 /* XXX this function needs a reliable check to detect
543  * sparse files. Otherwise, bmap/strategy may be used
544  * and fail on non-allocated blocks. VOP_READ/VOP_WRITE
545  * works on sparse files.
546  */
547 #if notyet
548 static bool
549 vnode_strategy_probe(struct vnd_softc *vnd)
550 {
551 	int error;
552 	daddr_t nbn;
553 
554 	if (!vnode_has_strategy(vnd))
555 		return false;
556 
557 	/* Convert the first logical block number to its
558 	 * physical block number.
559 	 */
560 	error = 0;
561 	vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
562 	error = VOP_BMAP(vnd->sc_vp, 0, NULL, &nbn, NULL);
563 	VOP_UNLOCK(vnd->sc_vp);
564 
565 	/* Test if that worked. */
566 	if (error == 0 && (long)nbn == -1)
567 		return false;
568 
569 	return true;
570 }
571 #endif
572 
573 static void
574 vndthread(void *arg)
575 {
576 	struct vnd_softc *vnd = arg;
577 	int s;
578 
579 	/* Determine whether we can *use* VOP_BMAP and VOP_STRATEGY to
580 	 * directly access the backing vnode.  If we can, use these two
581 	 * operations to avoid messing with the local buffer cache.
582 	 * Otherwise fall back to regular VOP_READ/VOP_WRITE operations
583 	 * which are guaranteed to work with any file system. */
584 	if ((vnd->sc_flags & VNF_USE_VN_RDWR) == 0 &&
585 	    ! vnode_has_strategy(vnd))
586 		vnd->sc_flags |= VNF_USE_VN_RDWR;
587 
588 #ifdef DEBUG
589 	if (vnddebug & VDB_INIT)
590 		printf("vndthread: vp %p, %s\n", vnd->sc_vp,
591 		    (vnd->sc_flags & VNF_USE_VN_RDWR) == 0 ?
592 		    "using bmap/strategy operations" :
593 		    "using read/write operations");
594 #endif
595 
596 	s = splbio();
597 	vnd->sc_flags |= VNF_KTHREAD;
598 	wakeup(&vnd->sc_kthread);
599 
600 	/*
601 	 * Dequeue requests and serve them depending on the available
602 	 * vnode operations.
603 	 */
604 	while ((vnd->sc_flags & VNF_VUNCONF) == 0) {
605 		struct vndxfer *vnx;
606 		int flags;
607 		struct buf *obp;
608 		struct buf *bp;
609 
610 		obp = bufq_get(vnd->sc_tab);
611 		if (obp == NULL) {
612 			tsleep(&vnd->sc_tab, PRIBIO, "vndbp", 0);
613 			continue;
614 		};
615 		if ((vnd->sc_flags & VNF_USE_VN_RDWR)) {
616 			KASSERT(vnd->sc_pending > 0 &&
617 			    vnd->sc_pending <= VND_MAXPENDING(vnd));
618 			if (vnd->sc_pending-- == VND_MAXPENDING(vnd))
619 				wakeup(&vnd->sc_pending);
620 		}
621 		splx(s);
622 		flags = obp->b_flags;
623 #ifdef DEBUG
624 		if (vnddebug & VDB_FOLLOW)
625 			printf("vndthread(%p)\n", obp);
626 #endif
627 
628 		if (vnd->sc_vp->v_mount == NULL) {
629 			obp->b_error = ENXIO;
630 			goto done;
631 		}
632 #ifdef VND_COMPRESSION
633 		/* handle a compressed read */
634 		if ((flags & B_READ) != 0 && (vnd->sc_flags & VNF_COMP)) {
635 			off_t bn;
636 
637 			/* Convert to a byte offset within the file. */
638 			bn = obp->b_rawblkno *
639 			    vnd->sc_dkdev.dk_label->d_secsize;
640 
641 			compstrategy(obp, bn);
642 			goto done;
643 		}
644 #endif /* VND_COMPRESSION */
645 
646 		/*
647 		 * Allocate a header for this transfer and link it to the
648 		 * buffer
649 		 */
650 		s = splbio();
651 		vnx = VND_GETXFER(vnd);
652 		splx(s);
653 		vnx->vx_vnd = vnd;
654 
655 		s = splbio();
656 		while (vnd->sc_active >= vnd->sc_maxactive) {
657 			tsleep(&vnd->sc_tab, PRIBIO, "vndac", 0);
658 		}
659 		vnd->sc_active++;
660 		splx(s);
661 
662 		/* Instrumentation. */
663 		disk_busy(&vnd->sc_dkdev);
664 
665 		bp = &vnx->vx_buf;
666 		buf_init(bp);
667 		bp->b_flags = (obp->b_flags & B_READ);
668 		bp->b_oflags = obp->b_oflags;
669 		bp->b_cflags = obp->b_cflags;
670 		bp->b_iodone = vndiodone;
671 		bp->b_private = obp;
672 		bp->b_vp = vnd->sc_vp;
673 		bp->b_objlock = bp->b_vp->v_interlock;
674 		bp->b_data = obp->b_data;
675 		bp->b_bcount = obp->b_bcount;
676 		BIO_COPYPRIO(bp, obp);
677 
678 		/* Handle the request using the appropriate operations. */
679 		if ((vnd->sc_flags & VNF_USE_VN_RDWR) == 0)
680 			handle_with_strategy(vnd, obp, bp);
681 		else
682 			handle_with_rdwr(vnd, obp, bp);
683 
684 		s = splbio();
685 		continue;
686 
687 done:
688 		biodone(obp);
689 		s = splbio();
690 	}
691 
692 	vnd->sc_flags &= (~VNF_KTHREAD | VNF_VUNCONF);
693 	wakeup(&vnd->sc_kthread);
694 	splx(s);
695 	kthread_exit(0);
696 }
697 
698 /*
699  * Checks if the given vnode supports the requested operation.
700  * The operation is specified the offset returned by VOFFSET.
701  *
702  * XXX The test below used to determine this is quite fragile
703  * because it relies on the file system to use genfs to specify
704  * unimplemented operations.  There might be another way to do
705  * it more cleanly.
706  */
707 static bool
708 vnode_has_op(const struct vnode *vp, int opoffset)
709 {
710 	int (*defaultp)(void *);
711 	int (*opp)(void *);
712 
713 	defaultp = vp->v_op[VOFFSET(vop_default)];
714 	opp = vp->v_op[opoffset];
715 
716 	return opp != defaultp && opp != genfs_eopnotsupp &&
717 	    opp != genfs_badop && opp != genfs_nullop;
718 }
719 
720 /*
721  * Handes the read/write request given in 'bp' using the vnode's VOP_READ
722  * and VOP_WRITE operations.
723  *
724  * 'obp' is a pointer to the original request fed to the vnd device.
725  */
726 static void
727 handle_with_rdwr(struct vnd_softc *vnd, const struct buf *obp, struct buf *bp)
728 {
729 	bool doread;
730 	off_t offset;
731 	size_t len, resid;
732 	struct vnode *vp;
733 
734 	doread = bp->b_flags & B_READ;
735 	offset = obp->b_rawblkno * vnd->sc_dkdev.dk_label->d_secsize;
736 	len = bp->b_bcount;
737 	vp = vnd->sc_vp;
738 
739 #if defined(DEBUG)
740 	if (vnddebug & VDB_IO)
741 		printf("vnd (rdwr): vp %p, %s, rawblkno 0x%" PRIx64
742 		    ", secsize %d, offset %" PRIu64
743 		    ", bcount %d\n",
744 		    vp, doread ? "read" : "write", obp->b_rawblkno,
745 		    vnd->sc_dkdev.dk_label->d_secsize, offset,
746 		    bp->b_bcount);
747 #endif
748 
749 	/* Issue the read or write operation. */
750 	bp->b_error =
751 	    vn_rdwr(doread ? UIO_READ : UIO_WRITE,
752 	    vp, bp->b_data, len, offset, UIO_SYSSPACE,
753 	    IO_ADV_ENCODE(POSIX_FADV_NOREUSE), vnd->sc_cred, &resid, NULL);
754 	bp->b_resid = resid;
755 
756 	mutex_enter(vp->v_interlock);
757 	(void) VOP_PUTPAGES(vp, 0, 0,
758 	    PGO_ALLPAGES | PGO_CLEANIT | PGO_FREE | PGO_SYNCIO);
759 
760 	/* We need to increase the number of outputs on the vnode if
761 	 * there was any write to it. */
762 	if (!doread) {
763 		mutex_enter(vp->v_interlock);
764 		vp->v_numoutput++;
765 		mutex_exit(vp->v_interlock);
766 	}
767 
768 	biodone(bp);
769 }
770 
771 /*
772  * Handes the read/write request given in 'bp' using the vnode's VOP_BMAP
773  * and VOP_STRATEGY operations.
774  *
775  * 'obp' is a pointer to the original request fed to the vnd device.
776  */
777 static void
778 handle_with_strategy(struct vnd_softc *vnd, const struct buf *obp,
779     struct buf *bp)
780 {
781 	int bsize, error, flags, skipped;
782 	size_t resid, sz;
783 	off_t bn, offset;
784 	struct vnode *vp;
785 
786 	flags = obp->b_flags;
787 
788 	if (!(flags & B_READ)) {
789 		vp = bp->b_vp;
790 		mutex_enter(vp->v_interlock);
791 		vp->v_numoutput++;
792 		mutex_exit(vp->v_interlock);
793 	}
794 
795 	/* convert to a byte offset within the file. */
796 	bn = obp->b_rawblkno * vnd->sc_dkdev.dk_label->d_secsize;
797 
798 	bsize = vnd->sc_vp->v_mount->mnt_stat.f_iosize;
799 	skipped = 0;
800 
801 	/*
802 	 * Break the request into bsize pieces and feed them
803 	 * sequentially using VOP_BMAP/VOP_STRATEGY.
804 	 * We do it this way to keep from flooding NFS servers if we
805 	 * are connected to an NFS file.  This places the burden on
806 	 * the client rather than the server.
807 	 */
808 	error = 0;
809 	bp->b_resid = bp->b_bcount;
810 	for (offset = 0, resid = bp->b_resid; resid;
811 	    resid -= sz, offset += sz) {
812 		struct buf *nbp;
813 		daddr_t nbn;
814 		int off, nra;
815 
816 		nra = 0;
817 		vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
818 		error = VOP_BMAP(vnd->sc_vp, bn / bsize, &vp, &nbn, &nra);
819 		VOP_UNLOCK(vnd->sc_vp);
820 
821 		if (error == 0 && (long)nbn == -1)
822 			error = EIO;
823 
824 		/*
825 		 * If there was an error or a hole in the file...punt.
826 		 * Note that we may have to wait for any operations
827 		 * that we have already fired off before releasing
828 		 * the buffer.
829 		 *
830 		 * XXX we could deal with holes here but it would be
831 		 * a hassle (in the write case).
832 		 */
833 		if (error) {
834 			skipped += resid;
835 			break;
836 		}
837 
838 #ifdef DEBUG
839 		if (!dovndcluster)
840 			nra = 0;
841 #endif
842 
843 		off = bn % bsize;
844 		sz = MIN(((off_t)1 + nra) * bsize - off, resid);
845 #ifdef	DEBUG
846 		if (vnddebug & VDB_IO)
847 			printf("vndstrategy: vp %p/%p bn 0x%qx/0x%" PRIx64
848 			    " sz 0x%zx\n", vnd->sc_vp, vp, (long long)bn,
849 			    nbn, sz);
850 #endif
851 
852 		nbp = getiobuf(vp, true);
853 		nestiobuf_setup(bp, nbp, offset, sz);
854 		nbp->b_blkno = nbn + btodb(off);
855 
856 #if 0 /* XXX #ifdef DEBUG */
857 		if (vnddebug & VDB_IO)
858 			printf("vndstart(%ld): bp %p vp %p blkno "
859 			    "0x%" PRIx64 " flags %x addr %p cnt 0x%x\n",
860 			    (long) (vnd-vnd_softc), &nbp->vb_buf,
861 			    nbp->vb_buf.b_vp, nbp->vb_buf.b_blkno,
862 			    nbp->vb_buf.b_flags, nbp->vb_buf.b_data,
863 			    nbp->vb_buf.b_bcount);
864 #endif
865 		VOP_STRATEGY(vp, nbp);
866 		bn += sz;
867 	}
868 	nestiobuf_done(bp, skipped, error);
869 }
870 
871 static void
872 vndiodone(struct buf *bp)
873 {
874 	struct vndxfer *vnx = VND_BUFTOXFER(bp);
875 	struct vnd_softc *vnd = vnx->vx_vnd;
876 	struct buf *obp = bp->b_private;
877 	int s = splbio();
878 
879 	KASSERT(&vnx->vx_buf == bp);
880 	KASSERT(vnd->sc_active > 0);
881 #ifdef DEBUG
882 	if (vnddebug & VDB_IO) {
883 		printf("vndiodone1: bp %p iodone: error %d\n",
884 		    bp, bp->b_error);
885 	}
886 #endif
887 	disk_unbusy(&vnd->sc_dkdev, bp->b_bcount - bp->b_resid,
888 	    (bp->b_flags & B_READ));
889 	vnd->sc_active--;
890 	if (vnd->sc_active == 0) {
891 		wakeup(&vnd->sc_tab);
892 	}
893 	splx(s);
894 	obp->b_error = bp->b_error;
895 	obp->b_resid = bp->b_resid;
896 	buf_destroy(bp);
897 	VND_PUTXFER(vnd, vnx);
898 	biodone(obp);
899 }
900 
901 /* ARGSUSED */
902 static int
903 vndread(dev_t dev, struct uio *uio, int flags)
904 {
905 	int unit = vndunit(dev);
906 	struct vnd_softc *sc;
907 
908 #ifdef DEBUG
909 	if (vnddebug & VDB_FOLLOW)
910 		printf("vndread(0x%"PRIx64", %p)\n", dev, uio);
911 #endif
912 
913 	sc = device_lookup_private(&vnd_cd, unit);
914 	if (sc == NULL)
915 		return ENXIO;
916 
917 	if ((sc->sc_flags & VNF_INITED) == 0)
918 		return ENXIO;
919 
920 	return physio(vndstrategy, NULL, dev, B_READ, minphys, uio);
921 }
922 
923 /* ARGSUSED */
924 static int
925 vndwrite(dev_t dev, struct uio *uio, int flags)
926 {
927 	int unit = vndunit(dev);
928 	struct vnd_softc *sc;
929 
930 #ifdef DEBUG
931 	if (vnddebug & VDB_FOLLOW)
932 		printf("vndwrite(0x%"PRIx64", %p)\n", dev, uio);
933 #endif
934 
935 	sc = device_lookup_private(&vnd_cd, unit);
936 	if (sc == NULL)
937 		return ENXIO;
938 
939 	if ((sc->sc_flags & VNF_INITED) == 0)
940 		return ENXIO;
941 
942 	return physio(vndstrategy, NULL, dev, B_WRITE, minphys, uio);
943 }
944 
945 static int
946 vnd_cget(struct lwp *l, int unit, int *un, struct vattr *va)
947 {
948 	int error;
949 	struct vnd_softc *vnd;
950 
951 	if (*un == -1)
952 		*un = unit;
953 	if (*un < 0)
954 		return EINVAL;
955 
956 	vnd = device_lookup_private(&vnd_cd, *un);
957 	if (vnd == NULL)
958 		return (*un >= vnd_cd.cd_ndevs) ? ENXIO : -1;
959 
960 	if ((vnd->sc_flags & VNF_INITED) == 0)
961 		return -1;
962 
963 	vn_lock(vnd->sc_vp, LK_SHARED | LK_RETRY);
964 	error = VOP_GETATTR(vnd->sc_vp, va, l->l_cred);
965 	VOP_UNLOCK(vnd->sc_vp);
966 	return error;
967 }
968 
969 static int
970 vnddoclear(struct vnd_softc *vnd, int pmask, int minor, bool force)
971 {
972 	int error;
973 
974 	if ((error = vndlock(vnd)) != 0)
975 		return error;
976 
977 	/*
978 	 * Don't unconfigure if any other partitions are open
979 	 * or if both the character and block flavors of this
980 	 * partition are open.
981 	 */
982 	if (DK_BUSY(vnd, pmask) && !force) {
983 		vndunlock(vnd);
984 		return EBUSY;
985 	}
986 
987 	/* Delete all of our wedges */
988 	dkwedge_delall(&vnd->sc_dkdev);
989 
990 	/*
991 	 * XXX vndclear() might call vndclose() implicitly;
992 	 * release lock to avoid recursion
993 	 *
994 	 * Set VNF_CLEARING to prevent vndopen() from
995 	 * sneaking in after we vndunlock().
996 	 */
997 	vnd->sc_flags |= VNF_CLEARING;
998 	vndunlock(vnd);
999 	vndclear(vnd, minor);
1000 #ifdef DEBUG
1001 	if (vnddebug & VDB_INIT)
1002 		printf("vndioctl: CLRed\n");
1003 #endif
1004 
1005 	/* Destroy the xfer and buffer pools. */
1006 	pool_destroy(&vnd->sc_vxpool);
1007 
1008 	/* Detach the disk. */
1009 	disk_detach(&vnd->sc_dkdev);
1010 
1011 	return 0;
1012 }
1013 
1014 /* ARGSUSED */
1015 static int
1016 vndioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l)
1017 {
1018 	bool force;
1019 	int unit = vndunit(dev);
1020 	struct vnd_softc *vnd;
1021 	struct vnd_ioctl *vio;
1022 	struct vattr vattr;
1023 	struct pathbuf *pb;
1024 	struct nameidata nd;
1025 	int error, part, pmask;
1026 	size_t geomsize;
1027 	int fflags;
1028 #ifdef __HAVE_OLD_DISKLABEL
1029 	struct disklabel newlabel;
1030 #endif
1031 	struct dkwedge_info *dkw;
1032 	struct dkwedge_list *dkwl;
1033 
1034 #ifdef DEBUG
1035 	if (vnddebug & VDB_FOLLOW)
1036 		printf("vndioctl(0x%"PRIx64", 0x%lx, %p, 0x%x, %p): unit %d\n",
1037 		    dev, cmd, data, flag, l->l_proc, unit);
1038 #endif
1039 	vnd = device_lookup_private(&vnd_cd, unit);
1040 	if (vnd == NULL &&
1041 #ifdef COMPAT_30
1042 	    cmd != VNDIOCGET30 &&
1043 #endif
1044 #ifdef COMPAT_50
1045 	    cmd != VNDIOCGET50 &&
1046 #endif
1047 	    cmd != VNDIOCGET)
1048 		return ENXIO;
1049 	vio = (struct vnd_ioctl *)data;
1050 
1051 	/* Must be open for writes for these commands... */
1052 	switch (cmd) {
1053 	case VNDIOCSET:
1054 	case VNDIOCCLR:
1055 #ifdef COMPAT_50
1056 	case VNDIOCSET50:
1057 	case VNDIOCCLR50:
1058 #endif
1059 	case DIOCSDINFO:
1060 	case DIOCWDINFO:
1061 #ifdef __HAVE_OLD_DISKLABEL
1062 	case ODIOCSDINFO:
1063 	case ODIOCWDINFO:
1064 #endif
1065 	case DIOCKLABEL:
1066 	case DIOCWLABEL:
1067 		if ((flag & FWRITE) == 0)
1068 			return EBADF;
1069 	}
1070 
1071 	/* Must be initialized for these... */
1072 	switch (cmd) {
1073 	case VNDIOCCLR:
1074 #ifdef VNDIOCCLR50
1075 	case VNDIOCCLR50:
1076 #endif
1077 	case DIOCGDINFO:
1078 	case DIOCSDINFO:
1079 	case DIOCWDINFO:
1080 	case DIOCGPART:
1081 	case DIOCKLABEL:
1082 	case DIOCWLABEL:
1083 	case DIOCGDEFLABEL:
1084 	case DIOCCACHESYNC:
1085 #ifdef __HAVE_OLD_DISKLABEL
1086 	case ODIOCGDINFO:
1087 	case ODIOCSDINFO:
1088 	case ODIOCWDINFO:
1089 	case ODIOCGDEFLABEL:
1090 #endif
1091 		if ((vnd->sc_flags & VNF_INITED) == 0)
1092 			return ENXIO;
1093 	}
1094 
1095 	switch (cmd) {
1096 #ifdef VNDIOCSET50
1097 	case VNDIOCSET50:
1098 #endif
1099 	case VNDIOCSET:
1100 		if (vnd->sc_flags & VNF_INITED)
1101 			return EBUSY;
1102 
1103 		if ((error = vndlock(vnd)) != 0)
1104 			return error;
1105 
1106 		fflags = FREAD;
1107 		if ((vio->vnd_flags & VNDIOF_READONLY) == 0)
1108 			fflags |= FWRITE;
1109 		error = pathbuf_copyin(vio->vnd_file, &pb);
1110 		if (error) {
1111 			goto unlock_and_exit;
1112 		}
1113 		NDINIT(&nd, LOOKUP, FOLLOW, pb);
1114 		if ((error = vn_open(&nd, fflags, 0)) != 0) {
1115 			pathbuf_destroy(pb);
1116 			goto unlock_and_exit;
1117 		}
1118 		KASSERT(l);
1119 		error = VOP_GETATTR(nd.ni_vp, &vattr, l->l_cred);
1120 		if (!error && nd.ni_vp->v_type != VREG)
1121 			error = EOPNOTSUPP;
1122 		if (!error && vattr.va_bytes < vattr.va_size)
1123 			/* File is definitely sparse, use vn_rdwr() */
1124 			vnd->sc_flags |= VNF_USE_VN_RDWR;
1125 		if (error) {
1126 			VOP_UNLOCK(nd.ni_vp);
1127 			goto close_and_exit;
1128 		}
1129 
1130 		/* If using a compressed file, initialize its info */
1131 		/* (or abort with an error if kernel has no compression) */
1132 		if (vio->vnd_flags & VNF_COMP) {
1133 #ifdef VND_COMPRESSION
1134 			struct vnd_comp_header *ch;
1135 			int i;
1136 			u_int32_t comp_size;
1137 			u_int32_t comp_maxsize;
1138 
1139 			/* allocate space for compresed file header */
1140 			ch = malloc(sizeof(struct vnd_comp_header),
1141 			M_TEMP, M_WAITOK);
1142 
1143 			/* read compressed file header */
1144 			error = vn_rdwr(UIO_READ, nd.ni_vp, (void *)ch,
1145 			  sizeof(struct vnd_comp_header), 0, UIO_SYSSPACE,
1146 			  IO_UNIT|IO_NODELOCKED, l->l_cred, NULL, NULL);
1147 			if (error) {
1148 				free(ch, M_TEMP);
1149 				VOP_UNLOCK(nd.ni_vp);
1150 				goto close_and_exit;
1151 			}
1152 
1153 			/* save some header info */
1154 			vnd->sc_comp_blksz = ntohl(ch->block_size);
1155 			/* note last offset is the file byte size */
1156 			vnd->sc_comp_numoffs = ntohl(ch->num_blocks)+1;
1157 			free(ch, M_TEMP);
1158 			if (vnd->sc_comp_blksz == 0 ||
1159 			    vnd->sc_comp_blksz % DEV_BSIZE !=0) {
1160 				VOP_UNLOCK(nd.ni_vp);
1161 				error = EINVAL;
1162 				goto close_and_exit;
1163 			}
1164 			if (sizeof(struct vnd_comp_header) +
1165 			  sizeof(u_int64_t) * vnd->sc_comp_numoffs >
1166 			  vattr.va_size) {
1167 				VOP_UNLOCK(nd.ni_vp);
1168 				error = EINVAL;
1169 				goto close_and_exit;
1170 			}
1171 
1172 			/* set decompressed file size */
1173 			vattr.va_size =
1174 			    ((u_quad_t)vnd->sc_comp_numoffs - 1) *
1175 			     (u_quad_t)vnd->sc_comp_blksz;
1176 
1177 			/* allocate space for all the compressed offsets */
1178 			vnd->sc_comp_offsets =
1179 			malloc(sizeof(u_int64_t) * vnd->sc_comp_numoffs,
1180 			M_DEVBUF, M_WAITOK);
1181 
1182 			/* read in the offsets */
1183 			error = vn_rdwr(UIO_READ, nd.ni_vp,
1184 			  (void *)vnd->sc_comp_offsets,
1185 			  sizeof(u_int64_t) * vnd->sc_comp_numoffs,
1186 			  sizeof(struct vnd_comp_header), UIO_SYSSPACE,
1187 			  IO_UNIT|IO_NODELOCKED, l->l_cred, NULL, NULL);
1188 			if (error) {
1189 				VOP_UNLOCK(nd.ni_vp);
1190 				goto close_and_exit;
1191 			}
1192 			/*
1193 			 * find largest block size (used for allocation limit).
1194 			 * Also convert offset to native byte order.
1195 			 */
1196 			comp_maxsize = 0;
1197 			for (i = 0; i < vnd->sc_comp_numoffs - 1; i++) {
1198 				vnd->sc_comp_offsets[i] =
1199 				  be64toh(vnd->sc_comp_offsets[i]);
1200 				comp_size = be64toh(vnd->sc_comp_offsets[i + 1])
1201 				  - vnd->sc_comp_offsets[i];
1202 				if (comp_size > comp_maxsize)
1203 					comp_maxsize = comp_size;
1204 			}
1205 			vnd->sc_comp_offsets[vnd->sc_comp_numoffs - 1] =
1206 			  be64toh(vnd->sc_comp_offsets[vnd->sc_comp_numoffs - 1]);
1207 
1208 			/* create compressed data buffer */
1209 			vnd->sc_comp_buff = malloc(comp_maxsize,
1210 			  M_DEVBUF, M_WAITOK);
1211 
1212 			/* create decompressed buffer */
1213 			vnd->sc_comp_decombuf = malloc(vnd->sc_comp_blksz,
1214 			  M_DEVBUF, M_WAITOK);
1215 			vnd->sc_comp_buffblk = -1;
1216 
1217 			/* Initialize decompress stream */
1218 			memset(&vnd->sc_comp_stream, 0, sizeof(z_stream));
1219 			vnd->sc_comp_stream.zalloc = vnd_alloc;
1220 			vnd->sc_comp_stream.zfree = vnd_free;
1221 			error = inflateInit2(&vnd->sc_comp_stream, MAX_WBITS);
1222 			if (error) {
1223 				if (vnd->sc_comp_stream.msg)
1224 					printf("vnd%d: compressed file, %s\n",
1225 					  unit, vnd->sc_comp_stream.msg);
1226 				VOP_UNLOCK(nd.ni_vp);
1227 				error = EINVAL;
1228 				goto close_and_exit;
1229 			}
1230 
1231 			vnd->sc_flags |= VNF_COMP | VNF_READONLY;
1232 #else /* !VND_COMPRESSION */
1233 			VOP_UNLOCK(nd.ni_vp);
1234 			error = EOPNOTSUPP;
1235 			goto close_and_exit;
1236 #endif /* VND_COMPRESSION */
1237 		}
1238 
1239 		VOP_UNLOCK(nd.ni_vp);
1240 		vnd->sc_vp = nd.ni_vp;
1241 		vnd->sc_size = btodb(vattr.va_size);	/* note truncation */
1242 
1243 		/*
1244 		 * Use pseudo-geometry specified.  If none was provided,
1245 		 * use "standard" Adaptec fictitious geometry.
1246 		 */
1247 		if (vio->vnd_flags & VNDIOF_HASGEOM) {
1248 
1249 			memcpy(&vnd->sc_geom, &vio->vnd_geom,
1250 			    sizeof(vio->vnd_geom));
1251 
1252 			/*
1253 			 * Sanity-check the sector size.
1254 			 * XXX Don't allow secsize < DEV_BSIZE.	 Should
1255 			 * XXX we?
1256 			 */
1257 			if (vnd->sc_geom.vng_secsize < DEV_BSIZE ||
1258 			    (vnd->sc_geom.vng_secsize % DEV_BSIZE) != 0 ||
1259 			    vnd->sc_geom.vng_ncylinders == 0 ||
1260 			    (vnd->sc_geom.vng_ntracks *
1261 			     vnd->sc_geom.vng_nsectors) == 0) {
1262 				error = EINVAL;
1263 				goto close_and_exit;
1264 			}
1265 
1266 			/*
1267 			 * Compute the size (in DEV_BSIZE blocks) specified
1268 			 * by the geometry.
1269 			 */
1270 			geomsize = (vnd->sc_geom.vng_nsectors *
1271 			    vnd->sc_geom.vng_ntracks *
1272 			    vnd->sc_geom.vng_ncylinders) *
1273 			    (vnd->sc_geom.vng_secsize / DEV_BSIZE);
1274 
1275 			/*
1276 			 * Sanity-check the size against the specified
1277 			 * geometry.
1278 			 */
1279 			if (vnd->sc_size < geomsize) {
1280 				error = EINVAL;
1281 				goto close_and_exit;
1282 			}
1283 		} else if (vnd->sc_size >= (32 * 64)) {
1284 			/*
1285 			 * Size must be at least 2048 DEV_BSIZE blocks
1286 			 * (1M) in order to use this geometry.
1287 			 */
1288 			vnd->sc_geom.vng_secsize = DEV_BSIZE;
1289 			vnd->sc_geom.vng_nsectors = 32;
1290 			vnd->sc_geom.vng_ntracks = 64;
1291 			vnd->sc_geom.vng_ncylinders = vnd->sc_size / (64 * 32);
1292 		} else {
1293 			vnd->sc_geom.vng_secsize = DEV_BSIZE;
1294 			vnd->sc_geom.vng_nsectors = 1;
1295 			vnd->sc_geom.vng_ntracks = 1;
1296 			vnd->sc_geom.vng_ncylinders = vnd->sc_size;
1297 		}
1298 
1299 		vnd_set_properties(vnd);
1300 
1301 		if (vio->vnd_flags & VNDIOF_READONLY) {
1302 			vnd->sc_flags |= VNF_READONLY;
1303 		}
1304 
1305 		if ((error = vndsetcred(vnd, l->l_cred)) != 0)
1306 			goto close_and_exit;
1307 
1308 		vndthrottle(vnd, vnd->sc_vp);
1309 		vio->vnd_osize = dbtob(vnd->sc_size);
1310 #ifdef VNDIOCSET50
1311 		if (cmd != VNDIOCSET50)
1312 #endif
1313 			vio->vnd_size = dbtob(vnd->sc_size);
1314 		vnd->sc_flags |= VNF_INITED;
1315 
1316 		/* create the kernel thread, wait for it to be up */
1317 		error = kthread_create(PRI_NONE, 0, NULL, vndthread, vnd,
1318 		    &vnd->sc_kthread, "%s", device_xname(vnd->sc_dev));
1319 		if (error)
1320 			goto close_and_exit;
1321 		while ((vnd->sc_flags & VNF_KTHREAD) == 0) {
1322 			tsleep(&vnd->sc_kthread, PRIBIO, "vndthr", 0);
1323 		}
1324 #ifdef DEBUG
1325 		if (vnddebug & VDB_INIT)
1326 			printf("vndioctl: SET vp %p size 0x%lx %d/%d/%d/%d\n",
1327 			    vnd->sc_vp, (unsigned long) vnd->sc_size,
1328 			    vnd->sc_geom.vng_secsize,
1329 			    vnd->sc_geom.vng_nsectors,
1330 			    vnd->sc_geom.vng_ntracks,
1331 			    vnd->sc_geom.vng_ncylinders);
1332 #endif
1333 
1334 		/* Attach the disk. */
1335 		disk_attach(&vnd->sc_dkdev);
1336 		disk_blocksize(&vnd->sc_dkdev, vnd->sc_geom.vng_secsize);
1337 
1338 		/* Initialize the xfer and buffer pools. */
1339 		pool_init(&vnd->sc_vxpool, sizeof(struct vndxfer), 0,
1340 		    0, 0, "vndxpl", NULL, IPL_BIO);
1341 
1342 		vndunlock(vnd);
1343 
1344 		pathbuf_destroy(pb);
1345 
1346 		/* Discover wedges on this disk */
1347 		dkwedge_discover(&vnd->sc_dkdev);
1348 
1349 		break;
1350 
1351 close_and_exit:
1352 		(void) vn_close(nd.ni_vp, fflags, l->l_cred);
1353 		pathbuf_destroy(pb);
1354 unlock_and_exit:
1355 #ifdef VND_COMPRESSION
1356 		/* free any allocated memory (for compressed file) */
1357 		if (vnd->sc_comp_offsets) {
1358 			free(vnd->sc_comp_offsets, M_DEVBUF);
1359 			vnd->sc_comp_offsets = NULL;
1360 		}
1361 		if (vnd->sc_comp_buff) {
1362 			free(vnd->sc_comp_buff, M_DEVBUF);
1363 			vnd->sc_comp_buff = NULL;
1364 		}
1365 		if (vnd->sc_comp_decombuf) {
1366 			free(vnd->sc_comp_decombuf, M_DEVBUF);
1367 			vnd->sc_comp_decombuf = NULL;
1368 		}
1369 #endif /* VND_COMPRESSION */
1370 		vndunlock(vnd);
1371 		return error;
1372 
1373 #ifdef VNDIOCCLR50
1374 	case VNDIOCCLR50:
1375 #endif
1376 	case VNDIOCCLR:
1377 		part = DISKPART(dev);
1378 		pmask = (1 << part);
1379 		force = (vio->vnd_flags & VNDIOF_FORCE) != 0;
1380 
1381 		if ((error = vnddoclear(vnd, pmask, minor(dev), force)) != 0)
1382 			return error;
1383 
1384 		break;
1385 
1386 #ifdef COMPAT_30
1387 	case VNDIOCGET30: {
1388 		struct vnd_user30 *vnu;
1389 		struct vattr va;
1390 		vnu = (struct vnd_user30 *)data;
1391 		KASSERT(l);
1392 		switch (error = vnd_cget(l, unit, &vnu->vnu_unit, &va)) {
1393 		case 0:
1394 			vnu->vnu_dev = va.va_fsid;
1395 			vnu->vnu_ino = va.va_fileid;
1396 			break;
1397 		case -1:
1398 			/* unused is not an error */
1399 			vnu->vnu_dev = 0;
1400 			vnu->vnu_ino = 0;
1401 			break;
1402 		default:
1403 			return error;
1404 		}
1405 		break;
1406 	}
1407 #endif
1408 
1409 #ifdef COMPAT_50
1410 	case VNDIOCGET50: {
1411 		struct vnd_user50 *vnu;
1412 		struct vattr va;
1413 		vnu = (struct vnd_user50 *)data;
1414 		KASSERT(l);
1415 		switch (error = vnd_cget(l, unit, &vnu->vnu_unit, &va)) {
1416 		case 0:
1417 			vnu->vnu_dev = va.va_fsid;
1418 			vnu->vnu_ino = va.va_fileid;
1419 			break;
1420 		case -1:
1421 			/* unused is not an error */
1422 			vnu->vnu_dev = 0;
1423 			vnu->vnu_ino = 0;
1424 			break;
1425 		default:
1426 			return error;
1427 		}
1428 		break;
1429 	}
1430 #endif
1431 
1432 	case VNDIOCGET: {
1433 		struct vnd_user *vnu;
1434 		struct vattr va;
1435 		vnu = (struct vnd_user *)data;
1436 		KASSERT(l);
1437 		switch (error = vnd_cget(l, unit, &vnu->vnu_unit, &va)) {
1438 		case 0:
1439 			vnu->vnu_dev = va.va_fsid;
1440 			vnu->vnu_ino = va.va_fileid;
1441 			break;
1442 		case -1:
1443 			/* unused is not an error */
1444 			vnu->vnu_dev = 0;
1445 			vnu->vnu_ino = 0;
1446 			break;
1447 		default:
1448 			return error;
1449 		}
1450 		break;
1451 	}
1452 
1453 	case DIOCGDINFO:
1454 		*(struct disklabel *)data = *(vnd->sc_dkdev.dk_label);
1455 		break;
1456 
1457 #ifdef __HAVE_OLD_DISKLABEL
1458 	case ODIOCGDINFO:
1459 		newlabel = *(vnd->sc_dkdev.dk_label);
1460 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1461 			return ENOTTY;
1462 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1463 		break;
1464 #endif
1465 
1466 	case DIOCGPART:
1467 		((struct partinfo *)data)->disklab = vnd->sc_dkdev.dk_label;
1468 		((struct partinfo *)data)->part =
1469 		    &vnd->sc_dkdev.dk_label->d_partitions[DISKPART(dev)];
1470 		break;
1471 
1472 	case DIOCWDINFO:
1473 	case DIOCSDINFO:
1474 #ifdef __HAVE_OLD_DISKLABEL
1475 	case ODIOCWDINFO:
1476 	case ODIOCSDINFO:
1477 #endif
1478 	{
1479 		struct disklabel *lp;
1480 
1481 		if ((error = vndlock(vnd)) != 0)
1482 			return error;
1483 
1484 		vnd->sc_flags |= VNF_LABELLING;
1485 
1486 #ifdef __HAVE_OLD_DISKLABEL
1487 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
1488 			memset(&newlabel, 0, sizeof newlabel);
1489 			memcpy(&newlabel, data, sizeof (struct olddisklabel));
1490 			lp = &newlabel;
1491 		} else
1492 #endif
1493 		lp = (struct disklabel *)data;
1494 
1495 		error = setdisklabel(vnd->sc_dkdev.dk_label,
1496 		    lp, 0, vnd->sc_dkdev.dk_cpulabel);
1497 		if (error == 0) {
1498 			if (cmd == DIOCWDINFO
1499 #ifdef __HAVE_OLD_DISKLABEL
1500 			    || cmd == ODIOCWDINFO
1501 #endif
1502 			   )
1503 				error = writedisklabel(VNDLABELDEV(dev),
1504 				    vndstrategy, vnd->sc_dkdev.dk_label,
1505 				    vnd->sc_dkdev.dk_cpulabel);
1506 		}
1507 
1508 		vnd->sc_flags &= ~VNF_LABELLING;
1509 
1510 		vndunlock(vnd);
1511 
1512 		if (error)
1513 			return error;
1514 		break;
1515 	}
1516 
1517 	case DIOCKLABEL:
1518 		if (*(int *)data != 0)
1519 			vnd->sc_flags |= VNF_KLABEL;
1520 		else
1521 			vnd->sc_flags &= ~VNF_KLABEL;
1522 		break;
1523 
1524 	case DIOCWLABEL:
1525 		if (*(int *)data != 0)
1526 			vnd->sc_flags |= VNF_WLABEL;
1527 		else
1528 			vnd->sc_flags &= ~VNF_WLABEL;
1529 		break;
1530 
1531 	case DIOCGDEFLABEL:
1532 		vndgetdefaultlabel(vnd, (struct disklabel *)data);
1533 		break;
1534 
1535 #ifdef __HAVE_OLD_DISKLABEL
1536 	case ODIOCGDEFLABEL:
1537 		vndgetdefaultlabel(vnd, &newlabel);
1538 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1539 			return ENOTTY;
1540 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1541 		break;
1542 #endif
1543 
1544 	case DIOCCACHESYNC:
1545 		vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1546 		error = VOP_FSYNC(vnd->sc_vp, vnd->sc_cred,
1547 		    FSYNC_WAIT | FSYNC_DATAONLY | FSYNC_CACHE, 0, 0);
1548 		VOP_UNLOCK(vnd->sc_vp);
1549 		return error;
1550 
1551 	case DIOCAWEDGE:
1552 		dkw = (void *) data;
1553 
1554 		if ((flag & FWRITE) == 0)
1555 			return EBADF;
1556 
1557 		/* If the ioctl happens here, the parent is us. */
1558 		strlcpy(dkw->dkw_parent, device_xname(vnd->sc_dev),
1559 		    sizeof(dkw->dkw_parent));
1560 		return dkwedge_add(dkw);
1561 
1562 	case DIOCDWEDGE:
1563 		dkw = (void *) data;
1564 
1565 		if ((flag & FWRITE) == 0)
1566 			return EBADF;
1567 
1568 		/* If the ioctl happens here, the parent is us. */
1569 		strlcpy(dkw->dkw_parent, device_xname(vnd->sc_dev),
1570 		    sizeof(dkw->dkw_parent));
1571 		return dkwedge_del(dkw);
1572 
1573 	case DIOCLWEDGES:
1574 		dkwl = (void *) data;
1575 
1576 		return dkwedge_list(&vnd->sc_dkdev, dkwl, l);
1577 
1578 	default:
1579 		return ENOTTY;
1580 	}
1581 
1582 	return 0;
1583 }
1584 
1585 /*
1586  * Duplicate the current processes' credentials.  Since we are called only
1587  * as the result of a SET ioctl and only root can do that, any future access
1588  * to this "disk" is essentially as root.  Note that credentials may change
1589  * if some other uid can write directly to the mapped file (NFS).
1590  */
1591 static int
1592 vndsetcred(struct vnd_softc *vnd, kauth_cred_t cred)
1593 {
1594 	struct uio auio;
1595 	struct iovec aiov;
1596 	char *tmpbuf;
1597 	int error;
1598 
1599 	vnd->sc_cred = kauth_cred_dup(cred);
1600 	tmpbuf = malloc(DEV_BSIZE, M_TEMP, M_WAITOK);
1601 
1602 	/* XXX: Horrible kludge to establish credentials for NFS */
1603 	aiov.iov_base = tmpbuf;
1604 	aiov.iov_len = min(DEV_BSIZE, dbtob(vnd->sc_size));
1605 	auio.uio_iov = &aiov;
1606 	auio.uio_iovcnt = 1;
1607 	auio.uio_offset = 0;
1608 	auio.uio_rw = UIO_READ;
1609 	auio.uio_resid = aiov.iov_len;
1610 	UIO_SETUP_SYSSPACE(&auio);
1611 	vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1612 	error = VOP_READ(vnd->sc_vp, &auio, 0, vnd->sc_cred);
1613 	if (error == 0) {
1614 		/*
1615 		 * Because vnd does all IO directly through the vnode
1616 		 * we need to flush (at least) the buffer from the above
1617 		 * VOP_READ from the buffer cache to prevent cache
1618 		 * incoherencies.  Also, be careful to write dirty
1619 		 * buffers back to stable storage.
1620 		 */
1621 		error = vinvalbuf(vnd->sc_vp, V_SAVE, vnd->sc_cred,
1622 			    curlwp, 0, 0);
1623 	}
1624 	VOP_UNLOCK(vnd->sc_vp);
1625 
1626 	free(tmpbuf, M_TEMP);
1627 	return error;
1628 }
1629 
1630 /*
1631  * Set maxactive based on FS type
1632  */
1633 static void
1634 vndthrottle(struct vnd_softc *vnd, struct vnode *vp)
1635 {
1636 
1637 	if (vp->v_tag == VT_NFS)
1638 		vnd->sc_maxactive = 2;
1639 	else
1640 		vnd->sc_maxactive = 8;
1641 
1642 	if (vnd->sc_maxactive < 1)
1643 		vnd->sc_maxactive = 1;
1644 }
1645 
1646 #if 0
1647 static void
1648 vndshutdown(void)
1649 {
1650 	struct vnd_softc *vnd;
1651 
1652 	for (vnd = &vnd_softc[0]; vnd < &vnd_softc[numvnd]; vnd++)
1653 		if (vnd->sc_flags & VNF_INITED)
1654 			vndclear(vnd);
1655 }
1656 #endif
1657 
1658 static void
1659 vndclear(struct vnd_softc *vnd, int myminor)
1660 {
1661 	struct vnode *vp = vnd->sc_vp;
1662 	int fflags = FREAD;
1663 	int bmaj, cmaj, i, mn;
1664 	int s;
1665 
1666 #ifdef DEBUG
1667 	if (vnddebug & VDB_FOLLOW)
1668 		printf("vndclear(%p): vp %p\n", vnd, vp);
1669 #endif
1670 	/* locate the major number */
1671 	bmaj = bdevsw_lookup_major(&vnd_bdevsw);
1672 	cmaj = cdevsw_lookup_major(&vnd_cdevsw);
1673 
1674 	/* Nuke the vnodes for any open instances */
1675 	for (i = 0; i < MAXPARTITIONS; i++) {
1676 		mn = DISKMINOR(device_unit(vnd->sc_dev), i);
1677 		vdevgone(bmaj, mn, mn, VBLK);
1678 		if (mn != myminor) /* XXX avoid to kill own vnode */
1679 			vdevgone(cmaj, mn, mn, VCHR);
1680 	}
1681 
1682 	if ((vnd->sc_flags & VNF_READONLY) == 0)
1683 		fflags |= FWRITE;
1684 
1685 	s = splbio();
1686 	bufq_drain(vnd->sc_tab);
1687 	splx(s);
1688 
1689 	vnd->sc_flags |= VNF_VUNCONF;
1690 	wakeup(&vnd->sc_tab);
1691 	while (vnd->sc_flags & VNF_KTHREAD)
1692 		tsleep(&vnd->sc_kthread, PRIBIO, "vnthr", 0);
1693 
1694 #ifdef VND_COMPRESSION
1695 	/* free the compressed file buffers */
1696 	if (vnd->sc_flags & VNF_COMP) {
1697 		if (vnd->sc_comp_offsets) {
1698 			free(vnd->sc_comp_offsets, M_DEVBUF);
1699 			vnd->sc_comp_offsets = NULL;
1700 		}
1701 		if (vnd->sc_comp_buff) {
1702 			free(vnd->sc_comp_buff, M_DEVBUF);
1703 			vnd->sc_comp_buff = NULL;
1704 		}
1705 		if (vnd->sc_comp_decombuf) {
1706 			free(vnd->sc_comp_decombuf, M_DEVBUF);
1707 			vnd->sc_comp_decombuf = NULL;
1708 		}
1709 	}
1710 #endif /* VND_COMPRESSION */
1711 	vnd->sc_flags &=
1712 	    ~(VNF_INITED | VNF_READONLY | VNF_VLABEL
1713 	      | VNF_VUNCONF | VNF_COMP | VNF_CLEARING);
1714 	if (vp == NULL)
1715 		panic("vndclear: null vp");
1716 	(void) vn_close(vp, fflags, vnd->sc_cred);
1717 	kauth_cred_free(vnd->sc_cred);
1718 	vnd->sc_vp = NULL;
1719 	vnd->sc_cred = NULL;
1720 	vnd->sc_size = 0;
1721 }
1722 
1723 static int
1724 vndsize(dev_t dev)
1725 {
1726 	struct vnd_softc *sc;
1727 	struct disklabel *lp;
1728 	int part, unit, omask;
1729 	int size;
1730 
1731 	unit = vndunit(dev);
1732 	sc = device_lookup_private(&vnd_cd, unit);
1733 	if (sc == NULL)
1734 		return -1;
1735 
1736 	if ((sc->sc_flags & VNF_INITED) == 0)
1737 		return -1;
1738 
1739 	part = DISKPART(dev);
1740 	omask = sc->sc_dkdev.dk_openmask & (1 << part);
1741 	lp = sc->sc_dkdev.dk_label;
1742 
1743 	if (omask == 0 && vndopen(dev, 0, S_IFBLK, curlwp))	/* XXX */
1744 		return -1;
1745 
1746 	if (lp->d_partitions[part].p_fstype != FS_SWAP)
1747 		size = -1;
1748 	else
1749 		size = lp->d_partitions[part].p_size *
1750 		    (lp->d_secsize / DEV_BSIZE);
1751 
1752 	if (omask == 0 && vndclose(dev, 0, S_IFBLK, curlwp))	/* XXX */
1753 		return -1;
1754 
1755 	return size;
1756 }
1757 
1758 static int
1759 vnddump(dev_t dev, daddr_t blkno, void *va,
1760     size_t size)
1761 {
1762 
1763 	/* Not implemented. */
1764 	return ENXIO;
1765 }
1766 
1767 static void
1768 vndgetdefaultlabel(struct vnd_softc *sc, struct disklabel *lp)
1769 {
1770 	struct vndgeom *vng = &sc->sc_geom;
1771 	struct partition *pp;
1772 
1773 	memset(lp, 0, sizeof(*lp));
1774 
1775 	lp->d_secperunit = sc->sc_size / (vng->vng_secsize / DEV_BSIZE);
1776 	lp->d_secsize = vng->vng_secsize;
1777 	lp->d_nsectors = vng->vng_nsectors;
1778 	lp->d_ntracks = vng->vng_ntracks;
1779 	lp->d_ncylinders = vng->vng_ncylinders;
1780 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1781 
1782 	strncpy(lp->d_typename, "vnd", sizeof(lp->d_typename));
1783 	lp->d_type = DTYPE_VND;
1784 	strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
1785 	lp->d_rpm = 3600;
1786 	lp->d_interleave = 1;
1787 	lp->d_flags = 0;
1788 
1789 	pp = &lp->d_partitions[RAW_PART];
1790 	pp->p_offset = 0;
1791 	pp->p_size = lp->d_secperunit;
1792 	pp->p_fstype = FS_UNUSED;
1793 	lp->d_npartitions = RAW_PART + 1;
1794 
1795 	lp->d_magic = DISKMAGIC;
1796 	lp->d_magic2 = DISKMAGIC;
1797 	lp->d_checksum = dkcksum(lp);
1798 }
1799 
1800 /*
1801  * Read the disklabel from a vnd.  If one is not present, create a fake one.
1802  */
1803 static void
1804 vndgetdisklabel(dev_t dev, struct vnd_softc *sc)
1805 {
1806 	const char *errstring;
1807 	struct disklabel *lp = sc->sc_dkdev.dk_label;
1808 	struct cpu_disklabel *clp = sc->sc_dkdev.dk_cpulabel;
1809 	int i;
1810 
1811 	memset(clp, 0, sizeof(*clp));
1812 
1813 	vndgetdefaultlabel(sc, lp);
1814 
1815 	/*
1816 	 * Call the generic disklabel extraction routine.
1817 	 */
1818 	errstring = readdisklabel(VNDLABELDEV(dev), vndstrategy, lp, clp);
1819 	if (errstring) {
1820 		/*
1821 		 * Lack of disklabel is common, but we print the warning
1822 		 * anyway, since it might contain other useful information.
1823 		 */
1824 		aprint_normal_dev(sc->sc_dev, "%s\n", errstring);
1825 
1826 		/*
1827 		 * For historical reasons, if there's no disklabel
1828 		 * present, all partitions must be FS_BSDFFS and
1829 		 * occupy the entire disk.
1830 		 */
1831 		for (i = 0; i < MAXPARTITIONS; i++) {
1832 			/*
1833 			 * Don't wipe out port specific hack (such as
1834 			 * dos partition hack of i386 port).
1835 			 */
1836 			if (lp->d_partitions[i].p_size != 0)
1837 				continue;
1838 
1839 			lp->d_partitions[i].p_size = lp->d_secperunit;
1840 			lp->d_partitions[i].p_offset = 0;
1841 			lp->d_partitions[i].p_fstype = FS_BSDFFS;
1842 		}
1843 
1844 		strncpy(lp->d_packname, "default label",
1845 		    sizeof(lp->d_packname));
1846 
1847 		lp->d_npartitions = MAXPARTITIONS;
1848 		lp->d_checksum = dkcksum(lp);
1849 	}
1850 }
1851 
1852 /*
1853  * Wait interruptibly for an exclusive lock.
1854  *
1855  * XXX
1856  * Several drivers do this; it should be abstracted and made MP-safe.
1857  */
1858 static int
1859 vndlock(struct vnd_softc *sc)
1860 {
1861 	int error;
1862 
1863 	while ((sc->sc_flags & VNF_LOCKED) != 0) {
1864 		sc->sc_flags |= VNF_WANTED;
1865 		if ((error = tsleep(sc, PRIBIO | PCATCH, "vndlck", 0)) != 0)
1866 			return error;
1867 	}
1868 	sc->sc_flags |= VNF_LOCKED;
1869 	return 0;
1870 }
1871 
1872 /*
1873  * Unlock and wake up any waiters.
1874  */
1875 static void
1876 vndunlock(struct vnd_softc *sc)
1877 {
1878 
1879 	sc->sc_flags &= ~VNF_LOCKED;
1880 	if ((sc->sc_flags & VNF_WANTED) != 0) {
1881 		sc->sc_flags &= ~VNF_WANTED;
1882 		wakeup(sc);
1883 	}
1884 }
1885 
1886 #ifdef VND_COMPRESSION
1887 /* compressed file read */
1888 static void
1889 compstrategy(struct buf *bp, off_t bn)
1890 {
1891 	int error;
1892 	int unit = vndunit(bp->b_dev);
1893 	struct vnd_softc *vnd =
1894 	    device_lookup_private(&vnd_cd, unit);
1895 	u_int32_t comp_block;
1896 	struct uio auio;
1897 	char *addr;
1898 	int s;
1899 
1900 	/* set up constants for data move */
1901 	auio.uio_rw = UIO_READ;
1902 	UIO_SETUP_SYSSPACE(&auio);
1903 
1904 	/* read, and transfer the data */
1905 	addr = bp->b_data;
1906 	bp->b_resid = bp->b_bcount;
1907 	s = splbio();
1908 	while (bp->b_resid > 0) {
1909 		unsigned length;
1910 		size_t length_in_buffer;
1911 		u_int32_t offset_in_buffer;
1912 		struct iovec aiov;
1913 
1914 		/* calculate the compressed block number */
1915 		comp_block = bn / (off_t)vnd->sc_comp_blksz;
1916 
1917 		/* check for good block number */
1918 		if (comp_block >= vnd->sc_comp_numoffs) {
1919 			bp->b_error = EINVAL;
1920 			splx(s);
1921 			return;
1922 		}
1923 
1924 		/* read in the compressed block, if not in buffer */
1925 		if (comp_block != vnd->sc_comp_buffblk) {
1926 			length = vnd->sc_comp_offsets[comp_block + 1] -
1927 			    vnd->sc_comp_offsets[comp_block];
1928 			vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1929 			error = vn_rdwr(UIO_READ, vnd->sc_vp, vnd->sc_comp_buff,
1930 			    length, vnd->sc_comp_offsets[comp_block],
1931 			    UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, vnd->sc_cred,
1932 			    NULL, NULL);
1933 			if (error) {
1934 				bp->b_error = error;
1935 				VOP_UNLOCK(vnd->sc_vp);
1936 				splx(s);
1937 				return;
1938 			}
1939 			/* uncompress the buffer */
1940 			vnd->sc_comp_stream.next_in = vnd->sc_comp_buff;
1941 			vnd->sc_comp_stream.avail_in = length;
1942 			vnd->sc_comp_stream.next_out = vnd->sc_comp_decombuf;
1943 			vnd->sc_comp_stream.avail_out = vnd->sc_comp_blksz;
1944 			inflateReset(&vnd->sc_comp_stream);
1945 			error = inflate(&vnd->sc_comp_stream, Z_FINISH);
1946 			if (error != Z_STREAM_END) {
1947 				if (vnd->sc_comp_stream.msg)
1948 					aprint_normal_dev(vnd->sc_dev,
1949 					    "compressed file, %s\n",
1950 					    vnd->sc_comp_stream.msg);
1951 				bp->b_error = EBADMSG;
1952 				VOP_UNLOCK(vnd->sc_vp);
1953 				splx(s);
1954 				return;
1955 			}
1956 			vnd->sc_comp_buffblk = comp_block;
1957 			VOP_UNLOCK(vnd->sc_vp);
1958 		}
1959 
1960 		/* transfer the usable uncompressed data */
1961 		offset_in_buffer = bn % (off_t)vnd->sc_comp_blksz;
1962 		length_in_buffer = vnd->sc_comp_blksz - offset_in_buffer;
1963 		if (length_in_buffer > bp->b_resid)
1964 			length_in_buffer = bp->b_resid;
1965 		auio.uio_iov = &aiov;
1966 		auio.uio_iovcnt = 1;
1967 		aiov.iov_base = addr;
1968 		aiov.iov_len = length_in_buffer;
1969 		auio.uio_resid = aiov.iov_len;
1970 		auio.uio_offset = 0;
1971 		error = uiomove(vnd->sc_comp_decombuf + offset_in_buffer,
1972 		    length_in_buffer, &auio);
1973 		if (error) {
1974 			bp->b_error = error;
1975 			splx(s);
1976 			return;
1977 		}
1978 
1979 		bn += length_in_buffer;
1980 		addr += length_in_buffer;
1981 		bp->b_resid -= length_in_buffer;
1982 	}
1983 	splx(s);
1984 }
1985 
1986 /* compression memory allocation routines */
1987 static void *
1988 vnd_alloc(void *aux, u_int items, u_int siz)
1989 {
1990 	return malloc(items * siz, M_TEMP, M_NOWAIT);
1991 }
1992 
1993 static void
1994 vnd_free(void *aux, void *ptr)
1995 {
1996 	free(ptr, M_TEMP);
1997 }
1998 #endif /* VND_COMPRESSION */
1999 
2000 static void
2001 vnd_set_properties(struct vnd_softc *vnd)
2002 {
2003 	prop_dictionary_t disk_info, odisk_info, geom;
2004 
2005 	disk_info = prop_dictionary_create();
2006 
2007 	geom = prop_dictionary_create();
2008 
2009 	prop_dictionary_set_uint64(geom, "sectors-per-unit",
2010 	    vnd->sc_geom.vng_nsectors * vnd->sc_geom.vng_ntracks *
2011 	    vnd->sc_geom.vng_ncylinders);
2012 
2013 	prop_dictionary_set_uint32(geom, "sector-size",
2014 	    vnd->sc_geom.vng_secsize);
2015 
2016 	prop_dictionary_set_uint16(geom, "sectors-per-track",
2017 	    vnd->sc_geom.vng_nsectors);
2018 
2019 	prop_dictionary_set_uint16(geom, "tracks-per-cylinder",
2020 	    vnd->sc_geom.vng_ntracks);
2021 
2022 	prop_dictionary_set_uint64(geom, "cylinders-per-unit",
2023 	    vnd->sc_geom.vng_ncylinders);
2024 
2025 	prop_dictionary_set(disk_info, "geometry", geom);
2026 	prop_object_release(geom);
2027 
2028 	prop_dictionary_set(device_properties(vnd->sc_dev),
2029 	    "disk-info", disk_info);
2030 
2031 	/*
2032 	 * Don't release disk_info here; we keep a reference to it.
2033 	 * disk_detach() will release it when we go away.
2034 	 */
2035 
2036 	odisk_info = vnd->sc_dkdev.dk_info;
2037 	vnd->sc_dkdev.dk_info = disk_info;
2038 	if (odisk_info)
2039 		prop_object_release(odisk_info);
2040 }
2041 
2042 #ifdef _MODULE
2043 
2044 #include <sys/module.h>
2045 
2046 MODULE(MODULE_CLASS_DRIVER, vnd, NULL);
2047 CFDRIVER_DECL(vnd, DV_DISK, NULL);
2048 
2049 static int
2050 vnd_modcmd(modcmd_t cmd, void *arg)
2051 {
2052 	int bmajor = -1, cmajor = -1,  error = 0;
2053 
2054 	switch (cmd) {
2055 	case MODULE_CMD_INIT:
2056 		error = config_cfdriver_attach(&vnd_cd);
2057 		if (error)
2058 			break;
2059 
2060 		error = config_cfattach_attach(vnd_cd.cd_name, &vnd_ca);
2061 	        if (error) {
2062 			config_cfdriver_detach(&vnd_cd);
2063 			aprint_error("%s: unable to register cfattach\n",
2064 			    vnd_cd.cd_name);
2065 			break;
2066 		}
2067 
2068 		error = devsw_attach("vnd", &vnd_bdevsw, &bmajor,
2069 		    &vnd_cdevsw, &cmajor);
2070 		if (error) {
2071 			config_cfattach_detach(vnd_cd.cd_name, &vnd_ca);
2072 			config_cfdriver_detach(&vnd_cd);
2073 			break;
2074 		}
2075 
2076 		break;
2077 
2078 	case MODULE_CMD_FINI:
2079 		error = config_cfattach_detach(vnd_cd.cd_name, &vnd_ca);
2080 		if (error)
2081 			break;
2082 		config_cfdriver_detach(&vnd_cd);
2083 		devsw_detach(&vnd_bdevsw, &vnd_cdevsw);
2084 		break;
2085 
2086 	case MODULE_CMD_STAT:
2087 		return ENOTTY;
2088 
2089 	default:
2090 		return ENOTTY;
2091 	}
2092 
2093 	return error;
2094 }
2095 
2096 #endif
2097