xref: /netbsd-src/sys/dev/vnd.c (revision 946379e7b37692fc43f68eb0d1c10daa0a7f3b6c)
1 /*	$NetBSD: vnd.c,v 1.256 2015/12/08 20:36:14 christos 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 1.256 2015/12/08 20:36:14 christos 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 "ioconf.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 = 0;
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 
162 static void	vndclear(struct vnd_softc *, int);
163 static int	vnddoclear(struct vnd_softc *, int, int, bool);
164 static int	vndsetcred(struct vnd_softc *, kauth_cred_t);
165 static void	vndthrottle(struct vnd_softc *, struct vnode *);
166 static void	vndiodone(struct buf *);
167 #if 0
168 static void	vndshutdown(void);
169 #endif
170 
171 static void	vndgetdefaultlabel(struct vnd_softc *, struct disklabel *);
172 static void	vndgetdisklabel(dev_t, struct vnd_softc *);
173 
174 static int	vndlock(struct vnd_softc *);
175 static void	vndunlock(struct vnd_softc *);
176 #ifdef VND_COMPRESSION
177 static void	compstrategy(struct buf *, off_t);
178 static void	*vnd_alloc(void *, u_int, u_int);
179 static void	vnd_free(void *, void *);
180 #endif /* VND_COMPRESSION */
181 
182 static void	vndthread(void *);
183 static bool	vnode_has_op(const struct vnode *, int);
184 static void	handle_with_rdwr(struct vnd_softc *, const struct buf *,
185 		    struct buf *);
186 static void	handle_with_strategy(struct vnd_softc *, const struct buf *,
187 		    struct buf *);
188 static void	vnd_set_geometry(struct vnd_softc *);
189 
190 static dev_type_open(vndopen);
191 static dev_type_close(vndclose);
192 static dev_type_read(vndread);
193 static dev_type_write(vndwrite);
194 static dev_type_ioctl(vndioctl);
195 static dev_type_strategy(vndstrategy);
196 static dev_type_dump(vnddump);
197 static dev_type_size(vndsize);
198 
199 const struct bdevsw vnd_bdevsw = {
200 	.d_open = vndopen,
201 	.d_close = vndclose,
202 	.d_strategy = vndstrategy,
203 	.d_ioctl = vndioctl,
204 	.d_dump = vnddump,
205 	.d_psize = vndsize,
206 	.d_discard = nodiscard,
207 	.d_flag = D_DISK
208 };
209 
210 const struct cdevsw vnd_cdevsw = {
211 	.d_open = vndopen,
212 	.d_close = vndclose,
213 	.d_read = vndread,
214 	.d_write = vndwrite,
215 	.d_ioctl = vndioctl,
216 	.d_stop = nostop,
217 	.d_tty = notty,
218 	.d_poll = nopoll,
219 	.d_mmap = nommap,
220 	.d_kqfilter = nokqfilter,
221 	.d_discard = nodiscard,
222 	.d_flag = D_DISK
223 };
224 
225 static int	vnd_match(device_t, cfdata_t, void *);
226 static void	vnd_attach(device_t, device_t, void *);
227 static int	vnd_detach(device_t, int);
228 
229 CFATTACH_DECL3_NEW(vnd, sizeof(struct vnd_softc),
230     vnd_match, vnd_attach, vnd_detach, NULL, NULL, NULL, DVF_DETACH_SHUTDOWN);
231 extern struct cfdriver vnd_cd;
232 
233 static struct vnd_softc	*vnd_spawn(int);
234 int	vnd_destroy(device_t);
235 
236 static struct	dkdriver vnddkdriver = {
237 	.d_strategy = vndstrategy,
238 	.d_minphys = minphys
239 };
240 
241 void
242 vndattach(int num)
243 {
244 	int error;
245 
246 	error = config_cfattach_attach(vnd_cd.cd_name, &vnd_ca);
247 	if (error)
248 		aprint_error("%s: unable to register cfattach, error = %d\n",
249 		    vnd_cd.cd_name, error);
250 }
251 
252 static int
253 vnd_match(device_t self, cfdata_t cfdata, void *aux)
254 {
255 
256 	return 1;
257 }
258 
259 static void
260 vnd_attach(device_t parent, device_t self, void *aux)
261 {
262 	struct vnd_softc *sc = device_private(self);
263 
264 	sc->sc_dev = self;
265 	sc->sc_comp_offsets = NULL;
266 	sc->sc_comp_buff = NULL;
267 	sc->sc_comp_decombuf = NULL;
268 	bufq_alloc(&sc->sc_tab, "disksort", BUFQ_SORT_RAWBLOCK);
269 	disk_init(&sc->sc_dkdev, device_xname(self), &vnddkdriver);
270 	if (!pmf_device_register(self, NULL, NULL))
271 		aprint_error_dev(self, "couldn't establish power handler\n");
272 }
273 
274 static int
275 vnd_detach(device_t self, int flags)
276 {
277 	int error;
278 	struct vnd_softc *sc = device_private(self);
279 
280 	if (sc->sc_flags & VNF_INITED) {
281 		error = vnddoclear(sc, 0, -1, (flags & DETACH_FORCE) != 0);
282 		if (error != 0)
283 			return error;
284 	}
285 
286 	pmf_device_deregister(self);
287 	bufq_free(sc->sc_tab);
288 	disk_destroy(&sc->sc_dkdev);
289 
290 	return 0;
291 }
292 
293 static struct vnd_softc *
294 vnd_spawn(int unit)
295 {
296 	cfdata_t cf;
297 
298 	cf = malloc(sizeof(*cf), M_DEVBUF, M_WAITOK);
299 	cf->cf_name = vnd_cd.cd_name;
300 	cf->cf_atname = vnd_cd.cd_name;
301 	cf->cf_unit = unit;
302 	cf->cf_fstate = FSTATE_STAR;
303 
304 	return device_private(config_attach_pseudo(cf));
305 }
306 
307 int
308 vnd_destroy(device_t dev)
309 {
310 	int error;
311 	cfdata_t cf;
312 
313 	cf = device_cfdata(dev);
314 	error = config_detach(dev, DETACH_QUIET);
315 	if (error)
316 		return error;
317 	free(cf, M_DEVBUF);
318 	return 0;
319 }
320 
321 static int
322 vndopen(dev_t dev, int flags, int mode, struct lwp *l)
323 {
324 	int unit = vndunit(dev);
325 	struct vnd_softc *sc;
326 	int error = 0, part, pmask;
327 	struct disklabel *lp;
328 
329 #ifdef DEBUG
330 	if (vnddebug & VDB_FOLLOW)
331 		printf("vndopen(0x%"PRIx64", 0x%x, 0x%x, %p)\n", dev, flags, mode, l);
332 #endif
333 	sc = device_lookup_private(&vnd_cd, unit);
334 	if (sc == NULL) {
335 		sc = vnd_spawn(unit);
336 		if (sc == NULL)
337 			return ENOMEM;
338 
339 		/* compatibility, keep disklabel after close */
340 		sc->sc_flags = VNF_KLABEL;
341 	}
342 
343 	if ((error = vndlock(sc)) != 0)
344 		return error;
345 
346 	mutex_enter(&sc->sc_dkdev.dk_openlock);
347 
348 	if ((sc->sc_flags & VNF_CLEARING) != 0) {
349 		error = ENXIO;
350 		goto done;
351 	}
352 
353 	lp = sc->sc_dkdev.dk_label;
354 
355 	part = DISKPART(dev);
356 	pmask = (1 << part);
357 
358 	if (sc->sc_dkdev.dk_nwedges != 0 && part != RAW_PART) {
359 		error = EBUSY;
360 		goto done;
361 	}
362 
363 	if (sc->sc_flags & VNF_INITED) {
364 		if ((sc->sc_dkdev.dk_openmask & ~(1<<RAW_PART)) != 0) {
365 			/*
366 			 * If any non-raw partition is open, but the disk
367 			 * has been invalidated, disallow further opens.
368 			 */
369 			if ((sc->sc_flags & VNF_VLABEL) == 0) {
370 				error = EIO;
371 				goto done;
372 			}
373 		} else {
374 			/*
375 			 * Load the partition info if not already loaded.
376 			 */
377 			if ((sc->sc_flags & VNF_VLABEL) == 0) {
378 				sc->sc_flags |= VNF_VLABEL;
379 				vndgetdisklabel(dev, sc);
380 			}
381 		}
382 	}
383 
384 	/* Check that the partitions exists. */
385 	if (part != RAW_PART) {
386 		if (((sc->sc_flags & VNF_INITED) == 0) ||
387 		    ((part >= lp->d_npartitions) ||
388 		     (lp->d_partitions[part].p_fstype == FS_UNUSED))) {
389 			error = ENXIO;
390 			goto done;
391 		}
392 	}
393 
394 	/* Prevent our unit from being unconfigured while open. */
395 	switch (mode) {
396 	case S_IFCHR:
397 		sc->sc_dkdev.dk_copenmask |= pmask;
398 		break;
399 
400 	case S_IFBLK:
401 		sc->sc_dkdev.dk_bopenmask |= pmask;
402 		break;
403 	}
404 	sc->sc_dkdev.dk_openmask =
405 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
406 
407  done:
408 	mutex_exit(&sc->sc_dkdev.dk_openlock);
409 	vndunlock(sc);
410 	return error;
411 }
412 
413 static int
414 vndclose(dev_t dev, int flags, int mode, struct lwp *l)
415 {
416 	int unit = vndunit(dev);
417 	struct vnd_softc *sc;
418 	int error = 0, part;
419 
420 #ifdef DEBUG
421 	if (vnddebug & VDB_FOLLOW)
422 		printf("vndclose(0x%"PRIx64", 0x%x, 0x%x, %p)\n", dev, flags, mode, l);
423 #endif
424 	sc = device_lookup_private(&vnd_cd, unit);
425 	if (sc == NULL)
426 		return ENXIO;
427 
428 	if ((error = vndlock(sc)) != 0)
429 		return error;
430 
431 	mutex_enter(&sc->sc_dkdev.dk_openlock);
432 
433 	part = DISKPART(dev);
434 
435 	/* ...that much closer to allowing unconfiguration... */
436 	switch (mode) {
437 	case S_IFCHR:
438 		sc->sc_dkdev.dk_copenmask &= ~(1 << part);
439 		break;
440 
441 	case S_IFBLK:
442 		sc->sc_dkdev.dk_bopenmask &= ~(1 << part);
443 		break;
444 	}
445 	sc->sc_dkdev.dk_openmask =
446 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
447 
448 	/* are we last opener ? */
449 	if (sc->sc_dkdev.dk_openmask == 0) {
450 		if ((sc->sc_flags & VNF_KLABEL) == 0)
451 			sc->sc_flags &= ~VNF_VLABEL;
452 	}
453 
454 	mutex_exit(&sc->sc_dkdev.dk_openlock);
455 
456 	vndunlock(sc);
457 
458 	if ((sc->sc_flags & VNF_INITED) == 0) {
459 		if ((error = vnd_destroy(sc->sc_dev)) != 0) {
460 			aprint_error_dev(sc->sc_dev,
461 			    "unable to detach instance\n");
462 			return error;
463 		}
464 	}
465 
466 	return 0;
467 }
468 
469 /*
470  * Queue the request, and wakeup the kernel thread to handle it.
471  */
472 static void
473 vndstrategy(struct buf *bp)
474 {
475 	int unit = vndunit(bp->b_dev);
476 	struct vnd_softc *vnd =
477 	    device_lookup_private(&vnd_cd, unit);
478 	struct disklabel *lp;
479 	daddr_t blkno;
480 	int s = splbio();
481 
482 	if (vnd == NULL) {
483 		bp->b_error = ENXIO;
484 		goto done;
485 	}
486 	lp = vnd->sc_dkdev.dk_label;
487 
488 	if ((vnd->sc_flags & VNF_INITED) == 0) {
489 		bp->b_error = ENXIO;
490 		goto done;
491 	}
492 
493 	/*
494 	 * The transfer must be a whole number of blocks.
495 	 */
496 	if ((bp->b_bcount % lp->d_secsize) != 0) {
497 		bp->b_error = EINVAL;
498 		goto done;
499 	}
500 
501 	/*
502 	 * check if we're read-only.
503 	 */
504 	if ((vnd->sc_flags & VNF_READONLY) && !(bp->b_flags & B_READ)) {
505 		bp->b_error = EACCES;
506 		goto done;
507 	}
508 
509 	/* If it's a nil transfer, wake up the top half now. */
510 	if (bp->b_bcount == 0) {
511 		goto done;
512 	}
513 
514 	/*
515 	 * Do bounds checking and adjust transfer.  If there's an error,
516 	 * the bounds check will flag that for us.
517 	 */
518 	if (DISKPART(bp->b_dev) == RAW_PART) {
519 		if (bounds_check_with_mediasize(bp, DEV_BSIZE,
520 		    vnd->sc_size) <= 0)
521 			goto done;
522 	} else {
523 		if (bounds_check_with_label(&vnd->sc_dkdev,
524 		    bp, vnd->sc_flags & (VNF_WLABEL|VNF_LABELLING)) <= 0)
525 			goto done;
526 	}
527 
528 	/*
529 	 * Put the block number in terms of the logical blocksize
530 	 * of the "device".
531 	 */
532 
533 	blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
534 
535 	/*
536 	 * Translate the partition-relative block number to an absolute.
537 	 */
538 	if (DISKPART(bp->b_dev) != RAW_PART) {
539 		struct partition *pp;
540 
541 		pp = &vnd->sc_dkdev.dk_label->d_partitions[
542 		    DISKPART(bp->b_dev)];
543 		blkno += pp->p_offset;
544 	}
545 	bp->b_rawblkno = blkno;
546 
547 #ifdef DEBUG
548 	if (vnddebug & VDB_FOLLOW)
549 		printf("vndstrategy(%p): unit %d\n", bp, unit);
550 #endif
551 	if ((vnd->sc_flags & VNF_USE_VN_RDWR)) {
552 		KASSERT(vnd->sc_pending >= 0 &&
553 		    vnd->sc_pending <= VND_MAXPENDING(vnd));
554 		while (vnd->sc_pending == VND_MAXPENDING(vnd))
555 			tsleep(&vnd->sc_pending, PRIBIO, "vndpc", 0);
556 		vnd->sc_pending++;
557 	}
558 	bufq_put(vnd->sc_tab, bp);
559 	wakeup(&vnd->sc_tab);
560 	splx(s);
561 	return;
562 
563 done:
564 	bp->b_resid = bp->b_bcount;
565 	biodone(bp);
566 	splx(s);
567 }
568 
569 static bool
570 vnode_has_strategy(struct vnd_softc *vnd)
571 {
572 	return vnode_has_op(vnd->sc_vp, VOFFSET(vop_bmap)) &&
573 	    vnode_has_op(vnd->sc_vp, VOFFSET(vop_strategy));
574 }
575 
576 static bool
577 vnode_has_large_blocks(struct vnd_softc *vnd)
578 {
579 	u_int32_t vnd_secsize, mnt_secsize;
580 	uint64_t numsec;
581 	unsigned secsize;
582 
583 	if (getdisksize(vnd->sc_vp, &numsec, &secsize))
584 		return true;
585 
586 	vnd_secsize = vnd->sc_geom.vng_secsize;
587 	mnt_secsize = secsize;
588 
589 	return vnd_secsize % mnt_secsize != 0;
590 }
591 
592 /* XXX this function needs a reliable check to detect
593  * sparse files. Otherwise, bmap/strategy may be used
594  * and fail on non-allocated blocks. VOP_READ/VOP_WRITE
595  * works on sparse files.
596  */
597 #if notyet
598 static bool
599 vnode_strategy_probe(struct vnd_softc *vnd)
600 {
601 	int error;
602 	daddr_t nbn;
603 
604 	if (!vnode_has_strategy(vnd))
605 		return false;
606 
607 	if (vnode_has_large_blocks(vnd))
608 		return false;
609 
610 	/* Convert the first logical block number to its
611 	 * physical block number.
612 	 */
613 	error = 0;
614 	vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
615 	error = VOP_BMAP(vnd->sc_vp, 0, NULL, &nbn, NULL);
616 	VOP_UNLOCK(vnd->sc_vp);
617 
618 	/* Test if that worked. */
619 	if (error == 0 && (long)nbn == -1)
620 		return false;
621 
622 	return true;
623 }
624 #endif
625 
626 static void
627 vndthread(void *arg)
628 {
629 	struct vnd_softc *vnd = arg;
630 	int s;
631 
632 	/* Determine whether we can *use* VOP_BMAP and VOP_STRATEGY to
633 	 * directly access the backing vnode.  If we can, use these two
634 	 * operations to avoid messing with the local buffer cache.
635 	 * Otherwise fall back to regular VOP_READ/VOP_WRITE operations
636 	 * which are guaranteed to work with any file system. */
637 	if ((vnd->sc_flags & VNF_USE_VN_RDWR) == 0 &&
638 	    ! vnode_has_strategy(vnd))
639 		vnd->sc_flags |= VNF_USE_VN_RDWR;
640 
641 	/* VOP_STRATEGY can only be used if the backing vnode allows
642 	 * to access blocks as small as defined by the vnd geometry.
643 	 */
644 	if ((vnd->sc_flags & VNF_USE_VN_RDWR) == 0 &&
645 	    vnode_has_large_blocks(vnd))
646 		vnd->sc_flags |= VNF_USE_VN_RDWR;
647 
648 #ifdef DEBUG
649 	if (vnddebug & VDB_INIT)
650 		printf("vndthread: vp %p, %s\n", vnd->sc_vp,
651 		    (vnd->sc_flags & VNF_USE_VN_RDWR) == 0 ?
652 		    "using bmap/strategy operations" :
653 		    "using read/write operations");
654 #endif
655 
656 	s = splbio();
657 	vnd->sc_flags |= VNF_KTHREAD;
658 	wakeup(&vnd->sc_kthread);
659 
660 	/*
661 	 * Dequeue requests and serve them depending on the available
662 	 * vnode operations.
663 	 */
664 	while ((vnd->sc_flags & VNF_VUNCONF) == 0) {
665 		struct vndxfer *vnx;
666 		struct buf *obp;
667 		struct buf *bp;
668 
669 		obp = bufq_get(vnd->sc_tab);
670 		if (obp == NULL) {
671 			tsleep(&vnd->sc_tab, PRIBIO, "vndbp", 0);
672 			continue;
673 		};
674 		if ((vnd->sc_flags & VNF_USE_VN_RDWR)) {
675 			KASSERT(vnd->sc_pending > 0 &&
676 			    vnd->sc_pending <= VND_MAXPENDING(vnd));
677 			if (vnd->sc_pending-- == VND_MAXPENDING(vnd))
678 				wakeup(&vnd->sc_pending);
679 		}
680 		splx(s);
681 #ifdef DEBUG
682 		if (vnddebug & VDB_FOLLOW)
683 			printf("vndthread(%p)\n", obp);
684 #endif
685 
686 		if (vnd->sc_vp->v_mount == NULL) {
687 			obp->b_error = ENXIO;
688 			goto done;
689 		}
690 #ifdef VND_COMPRESSION
691 		/* handle a compressed read */
692 		if ((obp->b_flags & B_READ) != 0 && (vnd->sc_flags & VNF_COMP)) {
693 			off_t bn;
694 
695 			/* Convert to a byte offset within the file. */
696 			bn = obp->b_rawblkno *
697 			    vnd->sc_dkdev.dk_label->d_secsize;
698 
699 			compstrategy(obp, bn);
700 			goto done;
701 		}
702 #endif /* VND_COMPRESSION */
703 
704 		/*
705 		 * Allocate a header for this transfer and link it to the
706 		 * buffer
707 		 */
708 		s = splbio();
709 		vnx = VND_GETXFER(vnd);
710 		splx(s);
711 		vnx->vx_vnd = vnd;
712 
713 		s = splbio();
714 		while (vnd->sc_active >= vnd->sc_maxactive) {
715 			tsleep(&vnd->sc_tab, PRIBIO, "vndac", 0);
716 		}
717 		vnd->sc_active++;
718 		splx(s);
719 
720 		/* Instrumentation. */
721 		disk_busy(&vnd->sc_dkdev);
722 
723 		bp = &vnx->vx_buf;
724 		buf_init(bp);
725 		bp->b_flags = (obp->b_flags & B_READ);
726 		bp->b_oflags = obp->b_oflags;
727 		bp->b_cflags = obp->b_cflags;
728 		bp->b_iodone = vndiodone;
729 		bp->b_private = obp;
730 		bp->b_vp = vnd->sc_vp;
731 		bp->b_objlock = bp->b_vp->v_interlock;
732 		bp->b_data = obp->b_data;
733 		bp->b_bcount = obp->b_bcount;
734 		BIO_COPYPRIO(bp, obp);
735 
736 		/* Handle the request using the appropriate operations. */
737 		if ((vnd->sc_flags & VNF_USE_VN_RDWR) == 0)
738 			handle_with_strategy(vnd, obp, bp);
739 		else
740 			handle_with_rdwr(vnd, obp, bp);
741 
742 		s = splbio();
743 		continue;
744 
745 done:
746 		biodone(obp);
747 		s = splbio();
748 	}
749 
750 	vnd->sc_flags &= (~VNF_KTHREAD | VNF_VUNCONF);
751 	wakeup(&vnd->sc_kthread);
752 	splx(s);
753 	kthread_exit(0);
754 }
755 
756 /*
757  * Checks if the given vnode supports the requested operation.
758  * The operation is specified the offset returned by VOFFSET.
759  *
760  * XXX The test below used to determine this is quite fragile
761  * because it relies on the file system to use genfs to specify
762  * unimplemented operations.  There might be another way to do
763  * it more cleanly.
764  */
765 static bool
766 vnode_has_op(const struct vnode *vp, int opoffset)
767 {
768 	int (*defaultp)(void *);
769 	int (*opp)(void *);
770 
771 	defaultp = vp->v_op[VOFFSET(vop_default)];
772 	opp = vp->v_op[opoffset];
773 
774 	return opp != defaultp && opp != genfs_eopnotsupp &&
775 	    opp != genfs_badop && opp != genfs_nullop;
776 }
777 
778 /*
779  * Handles the read/write request given in 'bp' using the vnode's VOP_READ
780  * and VOP_WRITE operations.
781  *
782  * 'obp' is a pointer to the original request fed to the vnd device.
783  */
784 static void
785 handle_with_rdwr(struct vnd_softc *vnd, const struct buf *obp, struct buf *bp)
786 {
787 	bool doread;
788 	off_t offset;
789 	size_t len, resid;
790 	struct vnode *vp;
791 
792 	doread = bp->b_flags & B_READ;
793 	offset = obp->b_rawblkno * vnd->sc_dkdev.dk_label->d_secsize;
794 	len = bp->b_bcount;
795 	vp = vnd->sc_vp;
796 
797 #if defined(DEBUG)
798 	if (vnddebug & VDB_IO)
799 		printf("vnd (rdwr): vp %p, %s, rawblkno 0x%" PRIx64
800 		    ", secsize %d, offset %" PRIu64
801 		    ", bcount %d\n",
802 		    vp, doread ? "read" : "write", obp->b_rawblkno,
803 		    vnd->sc_dkdev.dk_label->d_secsize, offset,
804 		    bp->b_bcount);
805 #endif
806 
807 	/* Issue the read or write operation. */
808 	bp->b_error =
809 	    vn_rdwr(doread ? UIO_READ : UIO_WRITE,
810 	    vp, bp->b_data, len, offset, UIO_SYSSPACE,
811 	    IO_ADV_ENCODE(POSIX_FADV_NOREUSE), vnd->sc_cred, &resid, NULL);
812 	bp->b_resid = resid;
813 
814 	mutex_enter(vp->v_interlock);
815 	(void) VOP_PUTPAGES(vp, 0, 0,
816 	    PGO_ALLPAGES | PGO_CLEANIT | PGO_FREE | PGO_SYNCIO);
817 
818 	/* We need to increase the number of outputs on the vnode if
819 	 * there was any write to it. */
820 	if (!doread) {
821 		mutex_enter(vp->v_interlock);
822 		vp->v_numoutput++;
823 		mutex_exit(vp->v_interlock);
824 	}
825 
826 	biodone(bp);
827 }
828 
829 /*
830  * Handes the read/write request given in 'bp' using the vnode's VOP_BMAP
831  * and VOP_STRATEGY operations.
832  *
833  * 'obp' is a pointer to the original request fed to the vnd device.
834  */
835 static void
836 handle_with_strategy(struct vnd_softc *vnd, const struct buf *obp,
837     struct buf *bp)
838 {
839 	int bsize, error, flags, skipped;
840 	size_t resid, sz;
841 	off_t bn, offset;
842 	struct vnode *vp;
843 	struct buf *nbp = NULL;
844 
845 	flags = obp->b_flags;
846 
847 
848 	/* convert to a byte offset within the file. */
849 	bn = obp->b_rawblkno * vnd->sc_dkdev.dk_label->d_secsize;
850 
851 	bsize = vnd->sc_vp->v_mount->mnt_stat.f_iosize;
852 	skipped = 0;
853 
854 	/*
855 	 * Break the request into bsize pieces and feed them
856 	 * sequentially using VOP_BMAP/VOP_STRATEGY.
857 	 * We do it this way to keep from flooding NFS servers if we
858 	 * are connected to an NFS file.  This places the burden on
859 	 * the client rather than the server.
860 	 */
861 	error = 0;
862 	bp->b_resid = bp->b_bcount;
863 	for (offset = 0, resid = bp->b_resid; /* true */;
864 	    resid -= sz, offset += sz) {
865 		daddr_t nbn;
866 		int off, nra;
867 
868 		nra = 0;
869 		vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
870 		error = VOP_BMAP(vnd->sc_vp, bn / bsize, &vp, &nbn, &nra);
871 		VOP_UNLOCK(vnd->sc_vp);
872 
873 		if (error == 0 && (long)nbn == -1)
874 			error = EIO;
875 
876 		/*
877 		 * If there was an error or a hole in the file...punt.
878 		 * Note that we may have to wait for any operations
879 		 * that we have already fired off before releasing
880 		 * the buffer.
881 		 *
882 		 * XXX we could deal with holes here but it would be
883 		 * a hassle (in the write case).
884 		 */
885 		if (error) {
886 			skipped += resid;
887 			break;
888 		}
889 
890 #ifdef DEBUG
891 		if (!dovndcluster)
892 			nra = 0;
893 #endif
894 
895 		off = bn % bsize;
896 		sz = MIN(((off_t)1 + nra) * bsize - off, resid);
897 #ifdef	DEBUG
898 		if (vnddebug & VDB_IO)
899 			printf("vndstrategy: vp %p/%p bn 0x%qx/0x%" PRIx64
900 			    " sz 0x%zx\n", vnd->sc_vp, vp, (long long)bn,
901 			    nbn, sz);
902 #endif
903 
904 		nbp = getiobuf(vp, true);
905 		nestiobuf_setup(bp, nbp, offset, sz);
906 		nbp->b_blkno = nbn + btodb(off);
907 
908 #if 0 /* XXX #ifdef DEBUG */
909 		if (vnddebug & VDB_IO)
910 			printf("vndstart(%ld): bp %p vp %p blkno "
911 			    "0x%" PRIx64 " flags %x addr %p cnt 0x%x\n",
912 			    (long) (vnd-vnd_softc), &nbp->vb_buf,
913 			    nbp->vb_buf.b_vp, nbp->vb_buf.b_blkno,
914 			    nbp->vb_buf.b_flags, nbp->vb_buf.b_data,
915 			    nbp->vb_buf.b_bcount);
916 #endif
917 		if (resid == sz) {
918 			break;
919 		}
920 		VOP_STRATEGY(vp, nbp);
921 		bn += sz;
922 	}
923 	if (!(flags & B_READ)) {
924 		struct vnode *w_vp;
925 		/*
926 		 * this is the last nested buf, account for
927 		 * the parent buf write too.
928 		 * This has to be done last, so that
929 		 * fsync won't wait for this write which
930 		 * has no chance to complete before all nested bufs
931 		 * have been queued. But it has to be done
932 		 * before the last VOP_STRATEGY()
933 		 * or the call to nestiobuf_done().
934 		 */
935 		w_vp = bp->b_vp;
936 		mutex_enter(w_vp->v_interlock);
937 		w_vp->v_numoutput++;
938 		mutex_exit(w_vp->v_interlock);
939 	}
940 	KASSERT(skipped != 0 || nbp != NULL);
941 	if (skipped)
942 		nestiobuf_done(bp, skipped, error);
943 	else
944 		VOP_STRATEGY(vp, nbp);
945 }
946 
947 static void
948 vndiodone(struct buf *bp)
949 {
950 	struct vndxfer *vnx = VND_BUFTOXFER(bp);
951 	struct vnd_softc *vnd = vnx->vx_vnd;
952 	struct buf *obp = bp->b_private;
953 	int s = splbio();
954 
955 	KASSERT(&vnx->vx_buf == bp);
956 	KASSERT(vnd->sc_active > 0);
957 #ifdef DEBUG
958 	if (vnddebug & VDB_IO) {
959 		printf("vndiodone1: bp %p iodone: error %d\n",
960 		    bp, bp->b_error);
961 	}
962 #endif
963 	disk_unbusy(&vnd->sc_dkdev, bp->b_bcount - bp->b_resid,
964 	    (bp->b_flags & B_READ));
965 	vnd->sc_active--;
966 	if (vnd->sc_active == 0) {
967 		wakeup(&vnd->sc_tab);
968 	}
969 	splx(s);
970 	obp->b_error = bp->b_error;
971 	obp->b_resid = bp->b_resid;
972 	buf_destroy(bp);
973 	VND_PUTXFER(vnd, vnx);
974 	biodone(obp);
975 }
976 
977 /* ARGSUSED */
978 static int
979 vndread(dev_t dev, struct uio *uio, int flags)
980 {
981 	int unit = vndunit(dev);
982 	struct vnd_softc *sc;
983 
984 #ifdef DEBUG
985 	if (vnddebug & VDB_FOLLOW)
986 		printf("vndread(0x%"PRIx64", %p)\n", dev, uio);
987 #endif
988 
989 	sc = device_lookup_private(&vnd_cd, unit);
990 	if (sc == NULL)
991 		return ENXIO;
992 
993 	if ((sc->sc_flags & VNF_INITED) == 0)
994 		return ENXIO;
995 
996 	return physio(vndstrategy, NULL, dev, B_READ, minphys, uio);
997 }
998 
999 /* ARGSUSED */
1000 static int
1001 vndwrite(dev_t dev, struct uio *uio, int flags)
1002 {
1003 	int unit = vndunit(dev);
1004 	struct vnd_softc *sc;
1005 
1006 #ifdef DEBUG
1007 	if (vnddebug & VDB_FOLLOW)
1008 		printf("vndwrite(0x%"PRIx64", %p)\n", dev, uio);
1009 #endif
1010 
1011 	sc = device_lookup_private(&vnd_cd, unit);
1012 	if (sc == NULL)
1013 		return ENXIO;
1014 
1015 	if ((sc->sc_flags & VNF_INITED) == 0)
1016 		return ENXIO;
1017 
1018 	return physio(vndstrategy, NULL, dev, B_WRITE, minphys, uio);
1019 }
1020 
1021 static int
1022 vnd_cget(struct lwp *l, int unit, int *un, struct vattr *va)
1023 {
1024 	int error;
1025 	struct vnd_softc *vnd;
1026 
1027 	if (*un == -1)
1028 		*un = unit;
1029 	if (*un < 0)
1030 		return EINVAL;
1031 
1032 	vnd = device_lookup_private(&vnd_cd, *un);
1033 	if (vnd == NULL)
1034 		return -1;
1035 
1036 	if ((vnd->sc_flags & VNF_INITED) == 0)
1037 		return -1;
1038 
1039 	vn_lock(vnd->sc_vp, LK_SHARED | LK_RETRY);
1040 	error = VOP_GETATTR(vnd->sc_vp, va, l->l_cred);
1041 	VOP_UNLOCK(vnd->sc_vp);
1042 	return error;
1043 }
1044 
1045 static int
1046 vnddoclear(struct vnd_softc *vnd, int pmask, int minor, bool force)
1047 {
1048 	int error;
1049 
1050 	if ((error = vndlock(vnd)) != 0)
1051 		return error;
1052 
1053 	/*
1054 	 * Don't unconfigure if any other partitions are open
1055 	 * or if both the character and block flavors of this
1056 	 * partition are open.
1057 	 */
1058 	if (DK_BUSY(vnd, pmask) && !force) {
1059 		vndunlock(vnd);
1060 		return EBUSY;
1061 	}
1062 
1063 	/* Delete all of our wedges */
1064 	dkwedge_delall(&vnd->sc_dkdev);
1065 
1066 	/*
1067 	 * XXX vndclear() might call vndclose() implicitly;
1068 	 * release lock to avoid recursion
1069 	 *
1070 	 * Set VNF_CLEARING to prevent vndopen() from
1071 	 * sneaking in after we vndunlock().
1072 	 */
1073 	vnd->sc_flags |= VNF_CLEARING;
1074 	vndunlock(vnd);
1075 	vndclear(vnd, minor);
1076 #ifdef DEBUG
1077 	if (vnddebug & VDB_INIT)
1078 		printf("%s: CLRed\n", __func__);
1079 #endif
1080 
1081 	/* Destroy the xfer and buffer pools. */
1082 	pool_destroy(&vnd->sc_vxpool);
1083 
1084 	/* Detach the disk. */
1085 	disk_detach(&vnd->sc_dkdev);
1086 
1087 	return 0;
1088 }
1089 
1090 static int
1091 vndioctl_get(struct lwp *l, void *data, int unit, struct vattr *va)
1092 {
1093 	int error;
1094 
1095 	KASSERT(l);
1096 
1097 	/* the first member is always int vnd_unit in all the versions */
1098 	if (*(int *)data >= vnd_cd.cd_ndevs)
1099 		return ENXIO;
1100 
1101 	switch (error = vnd_cget(l, unit, (int *)data, va)) {
1102 	case -1:
1103 		/* unused is not an error */
1104 		memset(va, 0, sizeof(*va));
1105 		/*FALLTHROUGH*/
1106 	case 0:
1107 		return 0;
1108 	default:
1109 		return error;
1110 	}
1111 }
1112 
1113 /* ARGSUSED */
1114 static int
1115 vndioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l)
1116 {
1117 	bool force;
1118 	int unit = vndunit(dev);
1119 	struct vnd_softc *vnd;
1120 	struct vnd_ioctl *vio;
1121 	struct vattr vattr;
1122 	struct pathbuf *pb;
1123 	struct nameidata nd;
1124 	int error, part, pmask;
1125 	uint64_t geomsize;
1126 	int fflags;
1127 #ifdef __HAVE_OLD_DISKLABEL
1128 	struct disklabel newlabel;
1129 #endif
1130 
1131 #ifdef DEBUG
1132 	if (vnddebug & VDB_FOLLOW)
1133 		printf("vndioctl(0x%"PRIx64", 0x%lx, %p, 0x%x, %p): unit %d\n",
1134 		    dev, cmd, data, flag, l->l_proc, unit);
1135 #endif
1136 	/* Do the get's first; they don't need initialization or verification */
1137 	switch (cmd) {
1138 #ifdef COMPAT_30
1139 	case VNDIOCGET30: {
1140 		if ((error = vndioctl_get(l, data, unit, &vattr)) != 0)
1141 			return error;
1142 
1143 		struct vnd_user30 *vnu = data;
1144 		vnu->vnu_dev = vattr.va_fsid;
1145 		vnu->vnu_ino = vattr.va_fileid;
1146 		return 0;
1147 	}
1148 #endif
1149 #ifdef COMPAT_50
1150 	case VNDIOCGET50: {
1151 		if ((error = vndioctl_get(l, data, unit, &vattr)) != 0)
1152 			return error;
1153 
1154 		struct vnd_user50 *vnu = data;
1155 		vnu->vnu_dev = vattr.va_fsid;
1156 		vnu->vnu_ino = vattr.va_fileid;
1157 		return 0;
1158 	}
1159 #endif
1160 
1161 	case VNDIOCGET: {
1162 		if ((error = vndioctl_get(l, data, unit, &vattr)) != 0)
1163 			return error;
1164 
1165 		struct vnd_user *vnu = data;
1166 		vnu->vnu_dev = vattr.va_fsid;
1167 		vnu->vnu_ino = vattr.va_fileid;
1168 		return 0;
1169 	}
1170 	default:
1171 		break;
1172 	}
1173 
1174 	vnd = device_lookup_private(&vnd_cd, unit);
1175 	if (vnd == NULL)
1176 		return ENXIO;
1177 	vio = (struct vnd_ioctl *)data;
1178 
1179 	/* Must be open for writes for these commands... */
1180 	switch (cmd) {
1181 	case VNDIOCSET:
1182 	case VNDIOCCLR:
1183 #ifdef COMPAT_50
1184 	case VNDIOCSET50:
1185 	case VNDIOCCLR50:
1186 #endif
1187 	case DIOCSDINFO:
1188 	case DIOCWDINFO:
1189 #ifdef __HAVE_OLD_DISKLABEL
1190 	case ODIOCSDINFO:
1191 	case ODIOCWDINFO:
1192 #endif
1193 	case DIOCKLABEL:
1194 	case DIOCWLABEL:
1195 		if ((flag & FWRITE) == 0)
1196 			return EBADF;
1197 	}
1198 
1199 	/* Must be initialized for these... */
1200 	switch (cmd) {
1201 	case VNDIOCCLR:
1202 #ifdef VNDIOCCLR50
1203 	case VNDIOCCLR50:
1204 #endif
1205 	case DIOCGDINFO:
1206 	case DIOCSDINFO:
1207 	case DIOCWDINFO:
1208 	case DIOCGPARTINFO:
1209 	case DIOCKLABEL:
1210 	case DIOCWLABEL:
1211 	case DIOCGDEFLABEL:
1212 	case DIOCCACHESYNC:
1213 #ifdef __HAVE_OLD_DISKLABEL
1214 	case ODIOCGDINFO:
1215 	case ODIOCSDINFO:
1216 	case ODIOCWDINFO:
1217 	case ODIOCGDEFLABEL:
1218 #endif
1219 		if ((vnd->sc_flags & VNF_INITED) == 0)
1220 			return ENXIO;
1221 	}
1222 
1223 	error = disk_ioctl(&vnd->sc_dkdev, dev, cmd, data, flag, l);
1224 	if (error != EPASSTHROUGH)
1225 		return error;
1226 
1227 
1228 	switch (cmd) {
1229 #ifdef VNDIOCSET50
1230 	case VNDIOCSET50:
1231 #endif
1232 	case VNDIOCSET:
1233 		if (vnd->sc_flags & VNF_INITED)
1234 			return EBUSY;
1235 
1236 		if ((error = vndlock(vnd)) != 0)
1237 			return error;
1238 
1239 		fflags = FREAD;
1240 		if ((vio->vnd_flags & VNDIOF_READONLY) == 0)
1241 			fflags |= FWRITE;
1242 		error = pathbuf_copyin(vio->vnd_file, &pb);
1243 		if (error) {
1244 			goto unlock_and_exit;
1245 		}
1246 		NDINIT(&nd, LOOKUP, FOLLOW, pb);
1247 		if ((error = vn_open(&nd, fflags, 0)) != 0) {
1248 			pathbuf_destroy(pb);
1249 			goto unlock_and_exit;
1250 		}
1251 		KASSERT(l);
1252 		error = VOP_GETATTR(nd.ni_vp, &vattr, l->l_cred);
1253 		if (!error && nd.ni_vp->v_type != VREG)
1254 			error = EOPNOTSUPP;
1255 		if (!error && vattr.va_bytes < vattr.va_size)
1256 			/* File is definitely sparse, use vn_rdwr() */
1257 			vnd->sc_flags |= VNF_USE_VN_RDWR;
1258 		if (error) {
1259 			VOP_UNLOCK(nd.ni_vp);
1260 			goto close_and_exit;
1261 		}
1262 
1263 		/* If using a compressed file, initialize its info */
1264 		/* (or abort with an error if kernel has no compression) */
1265 		if (vio->vnd_flags & VNF_COMP) {
1266 #ifdef VND_COMPRESSION
1267 			struct vnd_comp_header *ch;
1268 			int i;
1269 			u_int32_t comp_size;
1270 			u_int32_t comp_maxsize;
1271 
1272 			/* allocate space for compresed file header */
1273 			ch = malloc(sizeof(struct vnd_comp_header),
1274 			M_TEMP, M_WAITOK);
1275 
1276 			/* read compressed file header */
1277 			error = vn_rdwr(UIO_READ, nd.ni_vp, (void *)ch,
1278 			  sizeof(struct vnd_comp_header), 0, UIO_SYSSPACE,
1279 			  IO_UNIT|IO_NODELOCKED, l->l_cred, NULL, NULL);
1280 			if (error) {
1281 				free(ch, M_TEMP);
1282 				VOP_UNLOCK(nd.ni_vp);
1283 				goto close_and_exit;
1284 			}
1285 
1286 			/* save some header info */
1287 			vnd->sc_comp_blksz = ntohl(ch->block_size);
1288 			/* note last offset is the file byte size */
1289 			vnd->sc_comp_numoffs = ntohl(ch->num_blocks)+1;
1290 			free(ch, M_TEMP);
1291 			if (!DK_DEV_BSIZE_OK(vnd->sc_comp_blksz)) {
1292 				VOP_UNLOCK(nd.ni_vp);
1293 				error = EINVAL;
1294 				goto close_and_exit;
1295 			}
1296 			if (sizeof(struct vnd_comp_header) +
1297 			  sizeof(u_int64_t) * vnd->sc_comp_numoffs >
1298 			  vattr.va_size) {
1299 				VOP_UNLOCK(nd.ni_vp);
1300 				error = EINVAL;
1301 				goto close_and_exit;
1302 			}
1303 
1304 			/* set decompressed file size */
1305 			vattr.va_size =
1306 			    ((u_quad_t)vnd->sc_comp_numoffs - 1) *
1307 			     (u_quad_t)vnd->sc_comp_blksz;
1308 
1309 			/* allocate space for all the compressed offsets */
1310 			vnd->sc_comp_offsets =
1311 			malloc(sizeof(u_int64_t) * vnd->sc_comp_numoffs,
1312 			M_DEVBUF, M_WAITOK);
1313 
1314 			/* read in the offsets */
1315 			error = vn_rdwr(UIO_READ, nd.ni_vp,
1316 			  (void *)vnd->sc_comp_offsets,
1317 			  sizeof(u_int64_t) * vnd->sc_comp_numoffs,
1318 			  sizeof(struct vnd_comp_header), UIO_SYSSPACE,
1319 			  IO_UNIT|IO_NODELOCKED, l->l_cred, NULL, NULL);
1320 			if (error) {
1321 				VOP_UNLOCK(nd.ni_vp);
1322 				goto close_and_exit;
1323 			}
1324 			/*
1325 			 * find largest block size (used for allocation limit).
1326 			 * Also convert offset to native byte order.
1327 			 */
1328 			comp_maxsize = 0;
1329 			for (i = 0; i < vnd->sc_comp_numoffs - 1; i++) {
1330 				vnd->sc_comp_offsets[i] =
1331 				  be64toh(vnd->sc_comp_offsets[i]);
1332 				comp_size = be64toh(vnd->sc_comp_offsets[i + 1])
1333 				  - vnd->sc_comp_offsets[i];
1334 				if (comp_size > comp_maxsize)
1335 					comp_maxsize = comp_size;
1336 			}
1337 			vnd->sc_comp_offsets[vnd->sc_comp_numoffs - 1] =
1338 			  be64toh(vnd->sc_comp_offsets[vnd->sc_comp_numoffs - 1]);
1339 
1340 			/* create compressed data buffer */
1341 			vnd->sc_comp_buff = malloc(comp_maxsize,
1342 			  M_DEVBUF, M_WAITOK);
1343 
1344 			/* create decompressed buffer */
1345 			vnd->sc_comp_decombuf = malloc(vnd->sc_comp_blksz,
1346 			  M_DEVBUF, M_WAITOK);
1347 			vnd->sc_comp_buffblk = -1;
1348 
1349 			/* Initialize decompress stream */
1350 			memset(&vnd->sc_comp_stream, 0, sizeof(z_stream));
1351 			vnd->sc_comp_stream.zalloc = vnd_alloc;
1352 			vnd->sc_comp_stream.zfree = vnd_free;
1353 			error = inflateInit2(&vnd->sc_comp_stream, MAX_WBITS);
1354 			if (error) {
1355 				if (vnd->sc_comp_stream.msg)
1356 					printf("vnd%d: compressed file, %s\n",
1357 					  unit, vnd->sc_comp_stream.msg);
1358 				VOP_UNLOCK(nd.ni_vp);
1359 				error = EINVAL;
1360 				goto close_and_exit;
1361 			}
1362 
1363 			vnd->sc_flags |= VNF_COMP | VNF_READONLY;
1364 #else /* !VND_COMPRESSION */
1365 			VOP_UNLOCK(nd.ni_vp);
1366 			error = EOPNOTSUPP;
1367 			goto close_and_exit;
1368 #endif /* VND_COMPRESSION */
1369 		}
1370 
1371 		VOP_UNLOCK(nd.ni_vp);
1372 		vnd->sc_vp = nd.ni_vp;
1373 		vnd->sc_size = btodb(vattr.va_size);	/* note truncation */
1374 
1375 		/*
1376 		 * Use pseudo-geometry specified.  If none was provided,
1377 		 * use "standard" Adaptec fictitious geometry.
1378 		 */
1379 		if (vio->vnd_flags & VNDIOF_HASGEOM) {
1380 
1381 			memcpy(&vnd->sc_geom, &vio->vnd_geom,
1382 			    sizeof(vio->vnd_geom));
1383 
1384 			/*
1385 			 * Sanity-check the sector size.
1386 			 */
1387 			if (!DK_DEV_BSIZE_OK(vnd->sc_geom.vng_secsize) ||
1388 			    vnd->sc_geom.vng_ncylinders == 0 ||
1389 			    vnd->sc_geom.vng_ntracks == 0 ||
1390 			    vnd->sc_geom.vng_nsectors == 0) {
1391 				error = EINVAL;
1392 				goto close_and_exit;
1393 			}
1394 
1395 			/*
1396 			 * Compute the size (in DEV_BSIZE blocks) specified
1397 			 * by the geometry.
1398 			 */
1399 			geomsize = (int64_t)vnd->sc_geom.vng_nsectors *
1400 			    vnd->sc_geom.vng_ntracks *
1401 			    vnd->sc_geom.vng_ncylinders *
1402 			    (vnd->sc_geom.vng_secsize / DEV_BSIZE);
1403 
1404 			/*
1405 			 * Sanity-check the size against the specified
1406 			 * geometry.
1407 			 */
1408 			if (vnd->sc_size < geomsize) {
1409 				error = EINVAL;
1410 				goto close_and_exit;
1411 			}
1412 		} else if (vnd->sc_size >= (32 * 64)) {
1413 			/*
1414 			 * Size must be at least 2048 DEV_BSIZE blocks
1415 			 * (1M) in order to use this geometry.
1416 			 */
1417 			vnd->sc_geom.vng_secsize = DEV_BSIZE;
1418 			vnd->sc_geom.vng_nsectors = 32;
1419 			vnd->sc_geom.vng_ntracks = 64;
1420 			vnd->sc_geom.vng_ncylinders = vnd->sc_size / (64 * 32);
1421 		} else {
1422 			vnd->sc_geom.vng_secsize = DEV_BSIZE;
1423 			vnd->sc_geom.vng_nsectors = 1;
1424 			vnd->sc_geom.vng_ntracks = 1;
1425 			vnd->sc_geom.vng_ncylinders = vnd->sc_size;
1426 		}
1427 
1428 		vnd_set_geometry(vnd);
1429 
1430 		if (vio->vnd_flags & VNDIOF_READONLY) {
1431 			vnd->sc_flags |= VNF_READONLY;
1432 		}
1433 
1434 		if ((error = vndsetcred(vnd, l->l_cred)) != 0)
1435 			goto close_and_exit;
1436 
1437 		vndthrottle(vnd, vnd->sc_vp);
1438 		vio->vnd_osize = dbtob(vnd->sc_size);
1439 #ifdef VNDIOCSET50
1440 		if (cmd != VNDIOCSET50)
1441 #endif
1442 			vio->vnd_size = dbtob(vnd->sc_size);
1443 		vnd->sc_flags |= VNF_INITED;
1444 
1445 		/* create the kernel thread, wait for it to be up */
1446 		error = kthread_create(PRI_NONE, 0, NULL, vndthread, vnd,
1447 		    &vnd->sc_kthread, "%s", device_xname(vnd->sc_dev));
1448 		if (error)
1449 			goto close_and_exit;
1450 		while ((vnd->sc_flags & VNF_KTHREAD) == 0) {
1451 			tsleep(&vnd->sc_kthread, PRIBIO, "vndthr", 0);
1452 		}
1453 #ifdef DEBUG
1454 		if (vnddebug & VDB_INIT)
1455 			printf("vndioctl: SET vp %p size 0x%lx %d/%d/%d/%d\n",
1456 			    vnd->sc_vp, (unsigned long) vnd->sc_size,
1457 			    vnd->sc_geom.vng_secsize,
1458 			    vnd->sc_geom.vng_nsectors,
1459 			    vnd->sc_geom.vng_ntracks,
1460 			    vnd->sc_geom.vng_ncylinders);
1461 #endif
1462 
1463 		/* Attach the disk. */
1464 		disk_attach(&vnd->sc_dkdev);
1465 
1466 		/* Initialize the xfer and buffer pools. */
1467 		pool_init(&vnd->sc_vxpool, sizeof(struct vndxfer), 0,
1468 		    0, 0, "vndxpl", NULL, IPL_BIO);
1469 
1470 		vndunlock(vnd);
1471 
1472 		pathbuf_destroy(pb);
1473 
1474 		/* Discover wedges on this disk */
1475 		dkwedge_discover(&vnd->sc_dkdev);
1476 
1477 		break;
1478 
1479 close_and_exit:
1480 		(void) vn_close(nd.ni_vp, fflags, l->l_cred);
1481 		pathbuf_destroy(pb);
1482 unlock_and_exit:
1483 #ifdef VND_COMPRESSION
1484 		/* free any allocated memory (for compressed file) */
1485 		if (vnd->sc_comp_offsets) {
1486 			free(vnd->sc_comp_offsets, M_DEVBUF);
1487 			vnd->sc_comp_offsets = NULL;
1488 		}
1489 		if (vnd->sc_comp_buff) {
1490 			free(vnd->sc_comp_buff, M_DEVBUF);
1491 			vnd->sc_comp_buff = NULL;
1492 		}
1493 		if (vnd->sc_comp_decombuf) {
1494 			free(vnd->sc_comp_decombuf, M_DEVBUF);
1495 			vnd->sc_comp_decombuf = NULL;
1496 		}
1497 #endif /* VND_COMPRESSION */
1498 		vndunlock(vnd);
1499 		return error;
1500 
1501 #ifdef VNDIOCCLR50
1502 	case VNDIOCCLR50:
1503 #endif
1504 	case VNDIOCCLR:
1505 		part = DISKPART(dev);
1506 		pmask = (1 << part);
1507 		force = (vio->vnd_flags & VNDIOF_FORCE) != 0;
1508 
1509 		if ((error = vnddoclear(vnd, pmask, minor(dev), force)) != 0)
1510 			return error;
1511 
1512 		break;
1513 
1514 
1515 	case DIOCWDINFO:
1516 	case DIOCSDINFO:
1517 #ifdef __HAVE_OLD_DISKLABEL
1518 	case ODIOCWDINFO:
1519 	case ODIOCSDINFO:
1520 #endif
1521 	{
1522 		struct disklabel *lp;
1523 
1524 		if ((error = vndlock(vnd)) != 0)
1525 			return error;
1526 
1527 		vnd->sc_flags |= VNF_LABELLING;
1528 
1529 #ifdef __HAVE_OLD_DISKLABEL
1530 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
1531 			memset(&newlabel, 0, sizeof newlabel);
1532 			memcpy(&newlabel, data, sizeof (struct olddisklabel));
1533 			lp = &newlabel;
1534 		} else
1535 #endif
1536 		lp = (struct disklabel *)data;
1537 
1538 		error = setdisklabel(vnd->sc_dkdev.dk_label,
1539 		    lp, 0, vnd->sc_dkdev.dk_cpulabel);
1540 		if (error == 0) {
1541 			if (cmd == DIOCWDINFO
1542 #ifdef __HAVE_OLD_DISKLABEL
1543 			    || cmd == ODIOCWDINFO
1544 #endif
1545 			   )
1546 				error = writedisklabel(VNDLABELDEV(dev),
1547 				    vndstrategy, vnd->sc_dkdev.dk_label,
1548 				    vnd->sc_dkdev.dk_cpulabel);
1549 		}
1550 
1551 		vnd->sc_flags &= ~VNF_LABELLING;
1552 
1553 		vndunlock(vnd);
1554 
1555 		if (error)
1556 			return error;
1557 		break;
1558 	}
1559 
1560 	case DIOCKLABEL:
1561 		if (*(int *)data != 0)
1562 			vnd->sc_flags |= VNF_KLABEL;
1563 		else
1564 			vnd->sc_flags &= ~VNF_KLABEL;
1565 		break;
1566 
1567 	case DIOCWLABEL:
1568 		if (*(int *)data != 0)
1569 			vnd->sc_flags |= VNF_WLABEL;
1570 		else
1571 			vnd->sc_flags &= ~VNF_WLABEL;
1572 		break;
1573 
1574 	case DIOCGDEFLABEL:
1575 		vndgetdefaultlabel(vnd, (struct disklabel *)data);
1576 		break;
1577 
1578 #ifdef __HAVE_OLD_DISKLABEL
1579 	case ODIOCGDEFLABEL:
1580 		vndgetdefaultlabel(vnd, &newlabel);
1581 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1582 			return ENOTTY;
1583 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1584 		break;
1585 #endif
1586 
1587 	case DIOCCACHESYNC:
1588 		vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1589 		error = VOP_FSYNC(vnd->sc_vp, vnd->sc_cred,
1590 		    FSYNC_WAIT | FSYNC_DATAONLY | FSYNC_CACHE, 0, 0);
1591 		VOP_UNLOCK(vnd->sc_vp);
1592 		return error;
1593 
1594 	default:
1595 		return ENOTTY;
1596 	}
1597 
1598 	return 0;
1599 }
1600 
1601 /*
1602  * Duplicate the current processes' credentials.  Since we are called only
1603  * as the result of a SET ioctl and only root can do that, any future access
1604  * to this "disk" is essentially as root.  Note that credentials may change
1605  * if some other uid can write directly to the mapped file (NFS).
1606  */
1607 static int
1608 vndsetcred(struct vnd_softc *vnd, kauth_cred_t cred)
1609 {
1610 	struct uio auio;
1611 	struct iovec aiov;
1612 	char *tmpbuf;
1613 	int error;
1614 
1615 	vnd->sc_cred = kauth_cred_dup(cred);
1616 	tmpbuf = malloc(DEV_BSIZE, M_TEMP, M_WAITOK);
1617 
1618 	/* XXX: Horrible kludge to establish credentials for NFS */
1619 	aiov.iov_base = tmpbuf;
1620 	aiov.iov_len = min(DEV_BSIZE, dbtob(vnd->sc_size));
1621 	auio.uio_iov = &aiov;
1622 	auio.uio_iovcnt = 1;
1623 	auio.uio_offset = 0;
1624 	auio.uio_rw = UIO_READ;
1625 	auio.uio_resid = aiov.iov_len;
1626 	UIO_SETUP_SYSSPACE(&auio);
1627 	vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1628 	error = VOP_READ(vnd->sc_vp, &auio, 0, vnd->sc_cred);
1629 	if (error == 0) {
1630 		/*
1631 		 * Because vnd does all IO directly through the vnode
1632 		 * we need to flush (at least) the buffer from the above
1633 		 * VOP_READ from the buffer cache to prevent cache
1634 		 * incoherencies.  Also, be careful to write dirty
1635 		 * buffers back to stable storage.
1636 		 */
1637 		error = vinvalbuf(vnd->sc_vp, V_SAVE, vnd->sc_cred,
1638 			    curlwp, 0, 0);
1639 	}
1640 	VOP_UNLOCK(vnd->sc_vp);
1641 
1642 	free(tmpbuf, M_TEMP);
1643 	return error;
1644 }
1645 
1646 /*
1647  * Set maxactive based on FS type
1648  */
1649 static void
1650 vndthrottle(struct vnd_softc *vnd, struct vnode *vp)
1651 {
1652 
1653 	if (vp->v_tag == VT_NFS)
1654 		vnd->sc_maxactive = 2;
1655 	else
1656 		vnd->sc_maxactive = 8;
1657 
1658 	if (vnd->sc_maxactive < 1)
1659 		vnd->sc_maxactive = 1;
1660 }
1661 
1662 #if 0
1663 static void
1664 vndshutdown(void)
1665 {
1666 	struct vnd_softc *vnd;
1667 
1668 	for (vnd = &vnd_softc[0]; vnd < &vnd_softc[numvnd]; vnd++)
1669 		if (vnd->sc_flags & VNF_INITED)
1670 			vndclear(vnd);
1671 }
1672 #endif
1673 
1674 static void
1675 vndclear(struct vnd_softc *vnd, int myminor)
1676 {
1677 	struct vnode *vp = vnd->sc_vp;
1678 	int fflags = FREAD;
1679 	int bmaj, cmaj, i, mn;
1680 	int s;
1681 
1682 #ifdef DEBUG
1683 	if (vnddebug & VDB_FOLLOW)
1684 		printf("vndclear(%p): vp %p\n", vnd, vp);
1685 #endif
1686 	/* locate the major number */
1687 	bmaj = bdevsw_lookup_major(&vnd_bdevsw);
1688 	cmaj = cdevsw_lookup_major(&vnd_cdevsw);
1689 
1690 	/* Nuke the vnodes for any open instances */
1691 	for (i = 0; i < MAXPARTITIONS; i++) {
1692 		mn = DISKMINOR(device_unit(vnd->sc_dev), i);
1693 		vdevgone(bmaj, mn, mn, VBLK);
1694 		if (mn != myminor) /* XXX avoid to kill own vnode */
1695 			vdevgone(cmaj, mn, mn, VCHR);
1696 	}
1697 
1698 	if ((vnd->sc_flags & VNF_READONLY) == 0)
1699 		fflags |= FWRITE;
1700 
1701 	s = splbio();
1702 	bufq_drain(vnd->sc_tab);
1703 	splx(s);
1704 
1705 	vnd->sc_flags |= VNF_VUNCONF;
1706 	wakeup(&vnd->sc_tab);
1707 	while (vnd->sc_flags & VNF_KTHREAD)
1708 		tsleep(&vnd->sc_kthread, PRIBIO, "vnthr", 0);
1709 
1710 #ifdef VND_COMPRESSION
1711 	/* free the compressed file buffers */
1712 	if (vnd->sc_flags & VNF_COMP) {
1713 		if (vnd->sc_comp_offsets) {
1714 			free(vnd->sc_comp_offsets, M_DEVBUF);
1715 			vnd->sc_comp_offsets = NULL;
1716 		}
1717 		if (vnd->sc_comp_buff) {
1718 			free(vnd->sc_comp_buff, M_DEVBUF);
1719 			vnd->sc_comp_buff = NULL;
1720 		}
1721 		if (vnd->sc_comp_decombuf) {
1722 			free(vnd->sc_comp_decombuf, M_DEVBUF);
1723 			vnd->sc_comp_decombuf = NULL;
1724 		}
1725 	}
1726 #endif /* VND_COMPRESSION */
1727 	vnd->sc_flags &=
1728 	    ~(VNF_INITED | VNF_READONLY | VNF_KLABEL | VNF_VLABEL
1729 	      | VNF_VUNCONF | VNF_COMP | VNF_CLEARING);
1730 	if (vp == NULL)
1731 		panic("vndclear: null vp");
1732 	(void) vn_close(vp, fflags, vnd->sc_cred);
1733 	kauth_cred_free(vnd->sc_cred);
1734 	vnd->sc_vp = NULL;
1735 	vnd->sc_cred = NULL;
1736 	vnd->sc_size = 0;
1737 }
1738 
1739 static int
1740 vndsize(dev_t dev)
1741 {
1742 	struct vnd_softc *sc;
1743 	struct disklabel *lp;
1744 	int part, unit, omask;
1745 	int size;
1746 
1747 	unit = vndunit(dev);
1748 	sc = device_lookup_private(&vnd_cd, unit);
1749 	if (sc == NULL)
1750 		return -1;
1751 
1752 	if ((sc->sc_flags & VNF_INITED) == 0)
1753 		return -1;
1754 
1755 	part = DISKPART(dev);
1756 	omask = sc->sc_dkdev.dk_openmask & (1 << part);
1757 	lp = sc->sc_dkdev.dk_label;
1758 
1759 	if (omask == 0 && vndopen(dev, 0, S_IFBLK, curlwp))	/* XXX */
1760 		return -1;
1761 
1762 	if (lp->d_partitions[part].p_fstype != FS_SWAP)
1763 		size = -1;
1764 	else
1765 		size = lp->d_partitions[part].p_size *
1766 		    (lp->d_secsize / DEV_BSIZE);
1767 
1768 	if (omask == 0 && vndclose(dev, 0, S_IFBLK, curlwp))	/* XXX */
1769 		return -1;
1770 
1771 	return size;
1772 }
1773 
1774 static int
1775 vnddump(dev_t dev, daddr_t blkno, void *va,
1776     size_t size)
1777 {
1778 
1779 	/* Not implemented. */
1780 	return ENXIO;
1781 }
1782 
1783 static void
1784 vndgetdefaultlabel(struct vnd_softc *sc, struct disklabel *lp)
1785 {
1786 	struct vndgeom *vng = &sc->sc_geom;
1787 	struct partition *pp;
1788 	unsigned spb;
1789 
1790 	memset(lp, 0, sizeof(*lp));
1791 
1792 	spb = vng->vng_secsize / DEV_BSIZE;
1793 	if (sc->sc_size / spb > UINT32_MAX)
1794 		lp->d_secperunit = UINT32_MAX;
1795 	else
1796 		lp->d_secperunit = sc->sc_size / spb;
1797 	lp->d_secsize = vng->vng_secsize;
1798 	lp->d_nsectors = vng->vng_nsectors;
1799 	lp->d_ntracks = vng->vng_ntracks;
1800 	lp->d_ncylinders = vng->vng_ncylinders;
1801 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1802 
1803 	strncpy(lp->d_typename, "vnd", sizeof(lp->d_typename));
1804 	lp->d_type = DKTYPE_VND;
1805 	strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
1806 	lp->d_rpm = 3600;
1807 	lp->d_interleave = 1;
1808 	lp->d_flags = 0;
1809 
1810 	pp = &lp->d_partitions[RAW_PART];
1811 	pp->p_offset = 0;
1812 	pp->p_size = lp->d_secperunit;
1813 	pp->p_fstype = FS_UNUSED;
1814 	lp->d_npartitions = RAW_PART + 1;
1815 
1816 	lp->d_magic = DISKMAGIC;
1817 	lp->d_magic2 = DISKMAGIC;
1818 	lp->d_checksum = dkcksum(lp);
1819 }
1820 
1821 /*
1822  * Read the disklabel from a vnd.  If one is not present, create a fake one.
1823  */
1824 static void
1825 vndgetdisklabel(dev_t dev, struct vnd_softc *sc)
1826 {
1827 	const char *errstring;
1828 	struct disklabel *lp = sc->sc_dkdev.dk_label;
1829 	struct cpu_disklabel *clp = sc->sc_dkdev.dk_cpulabel;
1830 	int i;
1831 
1832 	memset(clp, 0, sizeof(*clp));
1833 
1834 	vndgetdefaultlabel(sc, lp);
1835 
1836 	/*
1837 	 * Call the generic disklabel extraction routine.
1838 	 */
1839 	errstring = readdisklabel(VNDLABELDEV(dev), vndstrategy, lp, clp);
1840 	if (errstring) {
1841 		/*
1842 		 * Lack of disklabel is common, but we print the warning
1843 		 * anyway, since it might contain other useful information.
1844 		 */
1845 		aprint_normal_dev(sc->sc_dev, "%s\n", errstring);
1846 
1847 		/*
1848 		 * For historical reasons, if there's no disklabel
1849 		 * present, all partitions must be FS_BSDFFS and
1850 		 * occupy the entire disk.
1851 		 */
1852 		for (i = 0; i < MAXPARTITIONS; i++) {
1853 			/*
1854 			 * Don't wipe out port specific hack (such as
1855 			 * dos partition hack of i386 port).
1856 			 */
1857 			if (lp->d_partitions[i].p_size != 0)
1858 				continue;
1859 
1860 			lp->d_partitions[i].p_size = lp->d_secperunit;
1861 			lp->d_partitions[i].p_offset = 0;
1862 			lp->d_partitions[i].p_fstype = FS_BSDFFS;
1863 		}
1864 
1865 		strncpy(lp->d_packname, "default label",
1866 		    sizeof(lp->d_packname));
1867 
1868 		lp->d_npartitions = MAXPARTITIONS;
1869 		lp->d_checksum = dkcksum(lp);
1870 	}
1871 }
1872 
1873 /*
1874  * Wait interruptibly for an exclusive lock.
1875  *
1876  * XXX
1877  * Several drivers do this; it should be abstracted and made MP-safe.
1878  */
1879 static int
1880 vndlock(struct vnd_softc *sc)
1881 {
1882 	int error;
1883 
1884 	while ((sc->sc_flags & VNF_LOCKED) != 0) {
1885 		sc->sc_flags |= VNF_WANTED;
1886 		if ((error = tsleep(sc, PRIBIO | PCATCH, "vndlck", 0)) != 0)
1887 			return error;
1888 	}
1889 	sc->sc_flags |= VNF_LOCKED;
1890 	return 0;
1891 }
1892 
1893 /*
1894  * Unlock and wake up any waiters.
1895  */
1896 static void
1897 vndunlock(struct vnd_softc *sc)
1898 {
1899 
1900 	sc->sc_flags &= ~VNF_LOCKED;
1901 	if ((sc->sc_flags & VNF_WANTED) != 0) {
1902 		sc->sc_flags &= ~VNF_WANTED;
1903 		wakeup(sc);
1904 	}
1905 }
1906 
1907 #ifdef VND_COMPRESSION
1908 /* compressed file read */
1909 static void
1910 compstrategy(struct buf *bp, off_t bn)
1911 {
1912 	int error;
1913 	int unit = vndunit(bp->b_dev);
1914 	struct vnd_softc *vnd =
1915 	    device_lookup_private(&vnd_cd, unit);
1916 	u_int32_t comp_block;
1917 	struct uio auio;
1918 	char *addr;
1919 	int s;
1920 
1921 	/* set up constants for data move */
1922 	auio.uio_rw = UIO_READ;
1923 	UIO_SETUP_SYSSPACE(&auio);
1924 
1925 	/* read, and transfer the data */
1926 	addr = bp->b_data;
1927 	bp->b_resid = bp->b_bcount;
1928 	s = splbio();
1929 	while (bp->b_resid > 0) {
1930 		unsigned length;
1931 		size_t length_in_buffer;
1932 		u_int32_t offset_in_buffer;
1933 		struct iovec aiov;
1934 
1935 		/* calculate the compressed block number */
1936 		comp_block = bn / (off_t)vnd->sc_comp_blksz;
1937 
1938 		/* check for good block number */
1939 		if (comp_block >= vnd->sc_comp_numoffs) {
1940 			bp->b_error = EINVAL;
1941 			splx(s);
1942 			return;
1943 		}
1944 
1945 		/* read in the compressed block, if not in buffer */
1946 		if (comp_block != vnd->sc_comp_buffblk) {
1947 			length = vnd->sc_comp_offsets[comp_block + 1] -
1948 			    vnd->sc_comp_offsets[comp_block];
1949 			vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1950 			error = vn_rdwr(UIO_READ, vnd->sc_vp, vnd->sc_comp_buff,
1951 			    length, vnd->sc_comp_offsets[comp_block],
1952 			    UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, vnd->sc_cred,
1953 			    NULL, NULL);
1954 			if (error) {
1955 				bp->b_error = error;
1956 				VOP_UNLOCK(vnd->sc_vp);
1957 				splx(s);
1958 				return;
1959 			}
1960 			/* uncompress the buffer */
1961 			vnd->sc_comp_stream.next_in = vnd->sc_comp_buff;
1962 			vnd->sc_comp_stream.avail_in = length;
1963 			vnd->sc_comp_stream.next_out = vnd->sc_comp_decombuf;
1964 			vnd->sc_comp_stream.avail_out = vnd->sc_comp_blksz;
1965 			inflateReset(&vnd->sc_comp_stream);
1966 			error = inflate(&vnd->sc_comp_stream, Z_FINISH);
1967 			if (error != Z_STREAM_END) {
1968 				if (vnd->sc_comp_stream.msg)
1969 					aprint_normal_dev(vnd->sc_dev,
1970 					    "compressed file, %s\n",
1971 					    vnd->sc_comp_stream.msg);
1972 				bp->b_error = EBADMSG;
1973 				VOP_UNLOCK(vnd->sc_vp);
1974 				splx(s);
1975 				return;
1976 			}
1977 			vnd->sc_comp_buffblk = comp_block;
1978 			VOP_UNLOCK(vnd->sc_vp);
1979 		}
1980 
1981 		/* transfer the usable uncompressed data */
1982 		offset_in_buffer = bn % (off_t)vnd->sc_comp_blksz;
1983 		length_in_buffer = vnd->sc_comp_blksz - offset_in_buffer;
1984 		if (length_in_buffer > bp->b_resid)
1985 			length_in_buffer = bp->b_resid;
1986 		auio.uio_iov = &aiov;
1987 		auio.uio_iovcnt = 1;
1988 		aiov.iov_base = addr;
1989 		aiov.iov_len = length_in_buffer;
1990 		auio.uio_resid = aiov.iov_len;
1991 		auio.uio_offset = 0;
1992 		error = uiomove(vnd->sc_comp_decombuf + offset_in_buffer,
1993 		    length_in_buffer, &auio);
1994 		if (error) {
1995 			bp->b_error = error;
1996 			splx(s);
1997 			return;
1998 		}
1999 
2000 		bn += length_in_buffer;
2001 		addr += length_in_buffer;
2002 		bp->b_resid -= length_in_buffer;
2003 	}
2004 	splx(s);
2005 }
2006 
2007 /* compression memory allocation routines */
2008 static void *
2009 vnd_alloc(void *aux, u_int items, u_int siz)
2010 {
2011 	return malloc(items * siz, M_TEMP, M_NOWAIT);
2012 }
2013 
2014 static void
2015 vnd_free(void *aux, void *ptr)
2016 {
2017 	free(ptr, M_TEMP);
2018 }
2019 #endif /* VND_COMPRESSION */
2020 
2021 static void
2022 vnd_set_geometry(struct vnd_softc *vnd)
2023 {
2024 	struct disk_geom *dg = &vnd->sc_dkdev.dk_geom;
2025 
2026 	memset(dg, 0, sizeof(*dg));
2027 
2028 	dg->dg_secperunit = (int64_t)vnd->sc_geom.vng_nsectors *
2029 	    vnd->sc_geom.vng_ntracks * vnd->sc_geom.vng_ncylinders;
2030 	dg->dg_secsize = vnd->sc_geom.vng_secsize;
2031 	dg->dg_nsectors = vnd->sc_geom.vng_nsectors;
2032 	dg->dg_ntracks = vnd->sc_geom.vng_ntracks;
2033 	dg->dg_ncylinders = vnd->sc_geom.vng_ncylinders;
2034 
2035 #ifdef DEBUG
2036 	if (vnddebug & VDB_LABEL) {
2037 		printf("dg->dg_secperunit: %" PRId64 "\n", dg->dg_secperunit);
2038 		printf("dg->dg_ncylinders: %u\n", dg->dg_ncylinders);
2039 	}
2040 #endif
2041 	disk_set_info(vnd->sc_dev, &vnd->sc_dkdev, NULL);
2042 }
2043 
2044 #ifdef _MODULE
2045 
2046 #include <sys/module.h>
2047 
2048 #ifdef VND_COMPRESSION
2049 #define VND_DEPENDS "zlib"
2050 #else
2051 #define VND_DEPENDS NULL
2052 #endif
2053 
2054 MODULE(MODULE_CLASS_DRIVER, vnd, VND_DEPENDS);
2055 CFDRIVER_DECL(vnd, DV_DISK, NULL);
2056 
2057 static int
2058 vnd_modcmd(modcmd_t cmd, void *arg)
2059 {
2060 	int bmajor = -1, cmajor = -1,  error = 0;
2061 
2062 	switch (cmd) {
2063 	case MODULE_CMD_INIT:
2064 		error = config_cfdriver_attach(&vnd_cd);
2065 		if (error)
2066 			break;
2067 
2068 		error = config_cfattach_attach(vnd_cd.cd_name, &vnd_ca);
2069 	        if (error) {
2070 			config_cfdriver_detach(&vnd_cd);
2071 			aprint_error("%s: unable to register cfattach\n",
2072 			    vnd_cd.cd_name);
2073 			break;
2074 		}
2075 
2076 		error = devsw_attach("vnd", &vnd_bdevsw, &bmajor,
2077 		    &vnd_cdevsw, &cmajor);
2078 		if (error) {
2079 			config_cfattach_detach(vnd_cd.cd_name, &vnd_ca);
2080 			config_cfdriver_detach(&vnd_cd);
2081 			break;
2082 		}
2083 
2084 		break;
2085 
2086 	case MODULE_CMD_FINI:
2087 		error = config_cfattach_detach(vnd_cd.cd_name, &vnd_ca);
2088 		if (error)
2089 			break;
2090 		config_cfdriver_detach(&vnd_cd);
2091 		devsw_detach(&vnd_bdevsw, &vnd_cdevsw);
2092 		break;
2093 
2094 	case MODULE_CMD_STAT:
2095 		return ENOTTY;
2096 
2097 	default:
2098 		return ENOTTY;
2099 	}
2100 
2101 	return error;
2102 }
2103 
2104 #endif
2105