xref: /netbsd-src/sys/dev/vnd.c (revision b7b7574d3bf8eeb51a1fa3977b59142ec6434a55)
1 /*	$NetBSD: vnd.c,v 1.229 2014/03/22 16:08:51 prlw1 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 1.229 2014/03/22 16:08:51 prlw1 Exp $");
95 
96 #if defined(_KERNEL_OPT)
97 #include "opt_vnd.h"
98 #include "opt_compat_netbsd.h"
99 #endif
100 
101 #include <sys/param.h>
102 #include <sys/systm.h>
103 #include <sys/namei.h>
104 #include <sys/proc.h>
105 #include <sys/kthread.h>
106 #include <sys/errno.h>
107 #include <sys/buf.h>
108 #include <sys/bufq.h>
109 #include <sys/malloc.h>
110 #include <sys/ioctl.h>
111 #include <sys/disklabel.h>
112 #include <sys/device.h>
113 #include <sys/disk.h>
114 #include <sys/stat.h>
115 #include <sys/mount.h>
116 #include <sys/vnode.h>
117 #include <sys/file.h>
118 #include <sys/uio.h>
119 #include <sys/conf.h>
120 #include <sys/kauth.h>
121 
122 #include <net/zlib.h>
123 
124 #include <miscfs/genfs/genfs.h>
125 #include <miscfs/specfs/specdev.h>
126 
127 #include <dev/dkvar.h>
128 #include <dev/vndvar.h>
129 
130 #if defined(VNDDEBUG) && !defined(DEBUG)
131 #define DEBUG
132 #endif
133 
134 #ifdef DEBUG
135 int dovndcluster = 1;
136 #define VDB_FOLLOW	0x01
137 #define VDB_INIT	0x02
138 #define VDB_IO		0x04
139 #define VDB_LABEL	0x08
140 int vnddebug = 0x00;
141 #endif
142 
143 #define vndunit(x)	DISKUNIT(x)
144 
145 struct vndxfer {
146 	struct buf vx_buf;
147 	struct vnd_softc *vx_vnd;
148 };
149 #define	VND_BUFTOXFER(bp)	((struct vndxfer *)(void *)bp)
150 
151 #define VND_GETXFER(vnd)	pool_get(&(vnd)->sc_vxpool, PR_WAITOK)
152 #define VND_PUTXFER(vnd, vx)	pool_put(&(vnd)->sc_vxpool, (vx))
153 
154 #define VNDLABELDEV(dev) \
155     (MAKEDISKDEV(major((dev)), vndunit((dev)), RAW_PART))
156 
157 #define	VND_MAXPENDING(vnd)	((vnd)->sc_maxactive * 4)
158 
159 /* called by main() at boot time */
160 void	vndattach(int);
161 
162 static void	vndclear(struct vnd_softc *, int);
163 static int	vnddoclear(struct vnd_softc *, int, int, bool);
164 static int	vndsetcred(struct vnd_softc *, kauth_cred_t);
165 static void	vndthrottle(struct vnd_softc *, struct vnode *);
166 static void	vndiodone(struct buf *);
167 #if 0
168 static void	vndshutdown(void);
169 #endif
170 
171 static void	vndgetdefaultlabel(struct vnd_softc *, struct disklabel *);
172 static void	vndgetdisklabel(dev_t, struct vnd_softc *);
173 
174 static int	vndlock(struct vnd_softc *);
175 static void	vndunlock(struct vnd_softc *);
176 #ifdef VND_COMPRESSION
177 static void	compstrategy(struct buf *, off_t);
178 static void	*vnd_alloc(void *, u_int, u_int);
179 static void	vnd_free(void *, void *);
180 #endif /* VND_COMPRESSION */
181 
182 static void	vndthread(void *);
183 static bool	vnode_has_op(const struct vnode *, int);
184 static void	handle_with_rdwr(struct vnd_softc *, const struct buf *,
185 		    struct buf *);
186 static void	handle_with_strategy(struct vnd_softc *, const struct buf *,
187 		    struct buf *);
188 static void	vnd_set_geometry(struct vnd_softc *);
189 
190 static dev_type_open(vndopen);
191 static dev_type_close(vndclose);
192 static dev_type_read(vndread);
193 static dev_type_write(vndwrite);
194 static dev_type_ioctl(vndioctl);
195 static dev_type_strategy(vndstrategy);
196 static dev_type_dump(vnddump);
197 static dev_type_size(vndsize);
198 
199 const struct bdevsw vnd_bdevsw = {
200 	.d_open = vndopen,
201 	.d_close = vndclose,
202 	.d_strategy = vndstrategy,
203 	.d_ioctl = vndioctl,
204 	.d_dump = vnddump,
205 	.d_psize = vndsize,
206 	.d_flag = D_DISK
207 };
208 
209 const struct cdevsw vnd_cdevsw = {
210 	.d_open = vndopen,
211 	.d_close = vndclose,
212 	.d_read = vndread,
213 	.d_write = vndwrite,
214 	.d_ioctl = vndioctl,
215 	.d_stop = nostop,
216 	.d_tty = notty,
217 	.d_poll = nopoll,
218 	.d_mmap = nommap,
219 	.d_kqfilter = nokqfilter,
220 	.d_flag = D_DISK
221 };
222 
223 static int	vnd_match(device_t, cfdata_t, void *);
224 static void	vnd_attach(device_t, device_t, void *);
225 static int	vnd_detach(device_t, int);
226 
227 CFATTACH_DECL3_NEW(vnd, sizeof(struct vnd_softc),
228     vnd_match, vnd_attach, vnd_detach, NULL, NULL, NULL, DVF_DETACH_SHUTDOWN);
229 extern struct cfdriver vnd_cd;
230 
231 static struct vnd_softc	*vnd_spawn(int);
232 int	vnd_destroy(device_t);
233 
234 static struct	dkdriver vnddkdriver = { vndstrategy, minphys };
235 
236 void
237 vndattach(int num)
238 {
239 	int error;
240 
241 	error = config_cfattach_attach(vnd_cd.cd_name, &vnd_ca);
242 	if (error)
243 		aprint_error("%s: unable to register cfattach\n",
244 		    vnd_cd.cd_name);
245 }
246 
247 static int
248 vnd_match(device_t self, cfdata_t cfdata, void *aux)
249 {
250 
251 	return 1;
252 }
253 
254 static void
255 vnd_attach(device_t parent, device_t self, void *aux)
256 {
257 	struct vnd_softc *sc = device_private(self);
258 
259 	sc->sc_dev = self;
260 	sc->sc_comp_offsets = NULL;
261 	sc->sc_comp_buff = NULL;
262 	sc->sc_comp_decombuf = NULL;
263 	bufq_alloc(&sc->sc_tab, "disksort", BUFQ_SORT_RAWBLOCK);
264 	disk_init(&sc->sc_dkdev, device_xname(self), &vnddkdriver);
265 	if (!pmf_device_register(self, NULL, NULL))
266 		aprint_error_dev(self, "couldn't establish power handler\n");
267 }
268 
269 static int
270 vnd_detach(device_t self, int flags)
271 {
272 	int error;
273 	struct vnd_softc *sc = device_private(self);
274 
275 	if (sc->sc_flags & VNF_INITED) {
276 		error = vnddoclear(sc, 0, -1, (flags & DETACH_FORCE) != 0);
277 		if (error != 0)
278 			return error;
279 	}
280 
281 	pmf_device_deregister(self);
282 	bufq_free(sc->sc_tab);
283 	disk_destroy(&sc->sc_dkdev);
284 
285 	return 0;
286 }
287 
288 static struct vnd_softc *
289 vnd_spawn(int unit)
290 {
291 	cfdata_t cf;
292 
293 	cf = malloc(sizeof(*cf), M_DEVBUF, M_WAITOK);
294 	cf->cf_name = vnd_cd.cd_name;
295 	cf->cf_atname = vnd_cd.cd_name;
296 	cf->cf_unit = unit;
297 	cf->cf_fstate = FSTATE_STAR;
298 
299 	return device_private(config_attach_pseudo(cf));
300 }
301 
302 int
303 vnd_destroy(device_t dev)
304 {
305 	int error;
306 	cfdata_t cf;
307 
308 	cf = device_cfdata(dev);
309 	error = config_detach(dev, DETACH_QUIET);
310 	if (error)
311 		return error;
312 	free(cf, M_DEVBUF);
313 	return 0;
314 }
315 
316 static int
317 vndopen(dev_t dev, int flags, int mode, struct lwp *l)
318 {
319 	int unit = vndunit(dev);
320 	struct vnd_softc *sc;
321 	int error = 0, part, pmask;
322 	struct disklabel *lp;
323 
324 #ifdef DEBUG
325 	if (vnddebug & VDB_FOLLOW)
326 		printf("vndopen(0x%"PRIx64", 0x%x, 0x%x, %p)\n", dev, flags, mode, l);
327 #endif
328 	sc = device_lookup_private(&vnd_cd, unit);
329 	if (sc == NULL) {
330 		sc = vnd_spawn(unit);
331 		if (sc == NULL)
332 			return ENOMEM;
333 	}
334 
335 	if ((error = vndlock(sc)) != 0)
336 		return error;
337 
338 	if ((sc->sc_flags & VNF_CLEARING) != 0) {
339 		error = ENXIO;
340 		goto done;
341 	}
342 
343 	lp = sc->sc_dkdev.dk_label;
344 
345 	part = DISKPART(dev);
346 	pmask = (1 << part);
347 
348 	if (sc->sc_dkdev.dk_nwedges != 0 && part != RAW_PART) {
349 		error = EBUSY;
350 		goto done;
351 	}
352 
353 	if (sc->sc_flags & VNF_INITED) {
354 		if ((sc->sc_dkdev.dk_openmask & ~(1<<RAW_PART)) != 0) {
355 			/*
356 			 * If any non-raw partition is open, but the disk
357 			 * has been invalidated, disallow further opens.
358 			 */
359 			if ((sc->sc_flags & VNF_VLABEL) == 0) {
360 				error = EIO;
361 				goto done;
362 			}
363 		} else {
364 			/*
365 			 * Load the partition info if not already loaded.
366 			 */
367 			if ((sc->sc_flags & VNF_VLABEL) == 0) {
368 				sc->sc_flags |= VNF_VLABEL;
369 				vndgetdisklabel(dev, sc);
370 			}
371 		}
372 	}
373 
374 	/* Check that the partitions exists. */
375 	if (part != RAW_PART) {
376 		if (((sc->sc_flags & VNF_INITED) == 0) ||
377 		    ((part >= lp->d_npartitions) ||
378 		     (lp->d_partitions[part].p_fstype == FS_UNUSED))) {
379 			error = ENXIO;
380 			goto done;
381 		}
382 	}
383 
384 	/* Prevent our unit from being unconfigured while open. */
385 	switch (mode) {
386 	case S_IFCHR:
387 		sc->sc_dkdev.dk_copenmask |= pmask;
388 		break;
389 
390 	case S_IFBLK:
391 		sc->sc_dkdev.dk_bopenmask |= pmask;
392 		break;
393 	}
394 	sc->sc_dkdev.dk_openmask =
395 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
396 
397  done:
398 	vndunlock(sc);
399 	return error;
400 }
401 
402 static int
403 vndclose(dev_t dev, int flags, int mode, struct lwp *l)
404 {
405 	int unit = vndunit(dev);
406 	struct vnd_softc *sc;
407 	int error = 0, part;
408 
409 #ifdef DEBUG
410 	if (vnddebug & VDB_FOLLOW)
411 		printf("vndclose(0x%"PRIx64", 0x%x, 0x%x, %p)\n", dev, flags, mode, l);
412 #endif
413 	sc = device_lookup_private(&vnd_cd, unit);
414 	if (sc == NULL)
415 		return ENXIO;
416 
417 	if ((error = vndlock(sc)) != 0)
418 		return error;
419 
420 	part = DISKPART(dev);
421 
422 	/* ...that much closer to allowing unconfiguration... */
423 	switch (mode) {
424 	case S_IFCHR:
425 		sc->sc_dkdev.dk_copenmask &= ~(1 << part);
426 		break;
427 
428 	case S_IFBLK:
429 		sc->sc_dkdev.dk_bopenmask &= ~(1 << part);
430 		break;
431 	}
432 	sc->sc_dkdev.dk_openmask =
433 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
434 
435 	vndunlock(sc);
436 
437 	if ((sc->sc_flags & VNF_INITED) == 0) {
438 		if ((error = vnd_destroy(sc->sc_dev)) != 0) {
439 			aprint_error_dev(sc->sc_dev,
440 			    "unable to detach instance\n");
441 			return error;
442 		}
443 	}
444 
445 	return 0;
446 }
447 
448 /*
449  * Queue the request, and wakeup the kernel thread to handle it.
450  */
451 static void
452 vndstrategy(struct buf *bp)
453 {
454 	int unit = vndunit(bp->b_dev);
455 	struct vnd_softc *vnd =
456 	    device_lookup_private(&vnd_cd, unit);
457 	struct disklabel *lp;
458 	daddr_t blkno;
459 	int s = splbio();
460 
461 	if (vnd == NULL) {
462 		bp->b_error = ENXIO;
463 		goto done;
464 	}
465 	lp = vnd->sc_dkdev.dk_label;
466 
467 	if ((vnd->sc_flags & VNF_INITED) == 0) {
468 		bp->b_error = ENXIO;
469 		goto done;
470 	}
471 
472 	/*
473 	 * The transfer must be a whole number of blocks.
474 	 */
475 	if ((bp->b_bcount % lp->d_secsize) != 0) {
476 		bp->b_error = EINVAL;
477 		goto done;
478 	}
479 
480 	/*
481 	 * check if we're read-only.
482 	 */
483 	if ((vnd->sc_flags & VNF_READONLY) && !(bp->b_flags & B_READ)) {
484 		bp->b_error = EACCES;
485 		goto done;
486 	}
487 
488 	/* If it's a nil transfer, wake up the top half now. */
489 	if (bp->b_bcount == 0) {
490 		goto done;
491 	}
492 
493 	/*
494 	 * Do bounds checking and adjust transfer.  If there's an error,
495 	 * the bounds check will flag that for us.
496 	 */
497 	if (DISKPART(bp->b_dev) == RAW_PART) {
498 		if (bounds_check_with_mediasize(bp, DEV_BSIZE,
499 		    vnd->sc_size) <= 0)
500 			goto done;
501 	} else {
502 		if (bounds_check_with_label(&vnd->sc_dkdev,
503 		    bp, vnd->sc_flags & (VNF_WLABEL|VNF_LABELLING)) <= 0)
504 			goto done;
505 	}
506 
507 	/*
508 	 * Put the block number in terms of the logical blocksize
509 	 * of the "device".
510 	 */
511 
512 	blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
513 
514 	/*
515 	 * Translate the partition-relative block number to an absolute.
516 	 */
517 	if (DISKPART(bp->b_dev) != RAW_PART) {
518 		struct partition *pp;
519 
520 		pp = &vnd->sc_dkdev.dk_label->d_partitions[
521 		    DISKPART(bp->b_dev)];
522 		blkno += pp->p_offset;
523 	}
524 	bp->b_rawblkno = blkno;
525 
526 #ifdef DEBUG
527 	if (vnddebug & VDB_FOLLOW)
528 		printf("vndstrategy(%p): unit %d\n", bp, unit);
529 #endif
530 	if ((vnd->sc_flags & VNF_USE_VN_RDWR)) {
531 		KASSERT(vnd->sc_pending >= 0 &&
532 		    vnd->sc_pending <= VND_MAXPENDING(vnd));
533 		while (vnd->sc_pending == VND_MAXPENDING(vnd))
534 			tsleep(&vnd->sc_pending, PRIBIO, "vndpc", 0);
535 		vnd->sc_pending++;
536 	}
537 	bufq_put(vnd->sc_tab, bp);
538 	wakeup(&vnd->sc_tab);
539 	splx(s);
540 	return;
541 
542 done:
543 	bp->b_resid = bp->b_bcount;
544 	biodone(bp);
545 	splx(s);
546 }
547 
548 static bool
549 vnode_has_strategy(struct vnd_softc *vnd)
550 {
551 	return vnode_has_op(vnd->sc_vp, VOFFSET(vop_bmap)) &&
552 	    vnode_has_op(vnd->sc_vp, VOFFSET(vop_strategy));
553 }
554 
555 /* XXX this function needs a reliable check to detect
556  * sparse files. Otherwise, bmap/strategy may be used
557  * and fail on non-allocated blocks. VOP_READ/VOP_WRITE
558  * works on sparse files.
559  */
560 #if notyet
561 static bool
562 vnode_strategy_probe(struct vnd_softc *vnd)
563 {
564 	int error;
565 	daddr_t nbn;
566 
567 	if (!vnode_has_strategy(vnd))
568 		return false;
569 
570 	/* Convert the first logical block number to its
571 	 * physical block number.
572 	 */
573 	error = 0;
574 	vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
575 	error = VOP_BMAP(vnd->sc_vp, 0, NULL, &nbn, NULL);
576 	VOP_UNLOCK(vnd->sc_vp);
577 
578 	/* Test if that worked. */
579 	if (error == 0 && (long)nbn == -1)
580 		return false;
581 
582 	return true;
583 }
584 #endif
585 
586 static void
587 vndthread(void *arg)
588 {
589 	struct vnd_softc *vnd = arg;
590 	int s;
591 
592 	/* Determine whether we can *use* VOP_BMAP and VOP_STRATEGY to
593 	 * directly access the backing vnode.  If we can, use these two
594 	 * operations to avoid messing with the local buffer cache.
595 	 * Otherwise fall back to regular VOP_READ/VOP_WRITE operations
596 	 * which are guaranteed to work with any file system. */
597 	if ((vnd->sc_flags & VNF_USE_VN_RDWR) == 0 &&
598 	    ! vnode_has_strategy(vnd))
599 		vnd->sc_flags |= VNF_USE_VN_RDWR;
600 
601 #ifdef DEBUG
602 	if (vnddebug & VDB_INIT)
603 		printf("vndthread: vp %p, %s\n", vnd->sc_vp,
604 		    (vnd->sc_flags & VNF_USE_VN_RDWR) == 0 ?
605 		    "using bmap/strategy operations" :
606 		    "using read/write operations");
607 #endif
608 
609 	s = splbio();
610 	vnd->sc_flags |= VNF_KTHREAD;
611 	wakeup(&vnd->sc_kthread);
612 
613 	/*
614 	 * Dequeue requests and serve them depending on the available
615 	 * vnode operations.
616 	 */
617 	while ((vnd->sc_flags & VNF_VUNCONF) == 0) {
618 		struct vndxfer *vnx;
619 		struct buf *obp;
620 		struct buf *bp;
621 
622 		obp = bufq_get(vnd->sc_tab);
623 		if (obp == NULL) {
624 			tsleep(&vnd->sc_tab, PRIBIO, "vndbp", 0);
625 			continue;
626 		};
627 		if ((vnd->sc_flags & VNF_USE_VN_RDWR)) {
628 			KASSERT(vnd->sc_pending > 0 &&
629 			    vnd->sc_pending <= VND_MAXPENDING(vnd));
630 			if (vnd->sc_pending-- == VND_MAXPENDING(vnd))
631 				wakeup(&vnd->sc_pending);
632 		}
633 		splx(s);
634 #ifdef DEBUG
635 		if (vnddebug & VDB_FOLLOW)
636 			printf("vndthread(%p)\n", obp);
637 #endif
638 
639 		if (vnd->sc_vp->v_mount == NULL) {
640 			obp->b_error = ENXIO;
641 			goto done;
642 		}
643 #ifdef VND_COMPRESSION
644 		/* handle a compressed read */
645 		if ((obp->b_flags & B_READ) != 0 && (vnd->sc_flags & VNF_COMP)) {
646 			off_t bn;
647 
648 			/* Convert to a byte offset within the file. */
649 			bn = obp->b_rawblkno *
650 			    vnd->sc_dkdev.dk_label->d_secsize;
651 
652 			compstrategy(obp, bn);
653 			goto done;
654 		}
655 #endif /* VND_COMPRESSION */
656 
657 		/*
658 		 * Allocate a header for this transfer and link it to the
659 		 * buffer
660 		 */
661 		s = splbio();
662 		vnx = VND_GETXFER(vnd);
663 		splx(s);
664 		vnx->vx_vnd = vnd;
665 
666 		s = splbio();
667 		while (vnd->sc_active >= vnd->sc_maxactive) {
668 			tsleep(&vnd->sc_tab, PRIBIO, "vndac", 0);
669 		}
670 		vnd->sc_active++;
671 		splx(s);
672 
673 		/* Instrumentation. */
674 		disk_busy(&vnd->sc_dkdev);
675 
676 		bp = &vnx->vx_buf;
677 		buf_init(bp);
678 		bp->b_flags = (obp->b_flags & B_READ);
679 		bp->b_oflags = obp->b_oflags;
680 		bp->b_cflags = obp->b_cflags;
681 		bp->b_iodone = vndiodone;
682 		bp->b_private = obp;
683 		bp->b_vp = vnd->sc_vp;
684 		bp->b_objlock = bp->b_vp->v_interlock;
685 		bp->b_data = obp->b_data;
686 		bp->b_bcount = obp->b_bcount;
687 		BIO_COPYPRIO(bp, obp);
688 
689 		/* Handle the request using the appropriate operations. */
690 		if ((vnd->sc_flags & VNF_USE_VN_RDWR) == 0)
691 			handle_with_strategy(vnd, obp, bp);
692 		else
693 			handle_with_rdwr(vnd, obp, bp);
694 
695 		s = splbio();
696 		continue;
697 
698 done:
699 		biodone(obp);
700 		s = splbio();
701 	}
702 
703 	vnd->sc_flags &= (~VNF_KTHREAD | VNF_VUNCONF);
704 	wakeup(&vnd->sc_kthread);
705 	splx(s);
706 	kthread_exit(0);
707 }
708 
709 /*
710  * Checks if the given vnode supports the requested operation.
711  * The operation is specified the offset returned by VOFFSET.
712  *
713  * XXX The test below used to determine this is quite fragile
714  * because it relies on the file system to use genfs to specify
715  * unimplemented operations.  There might be another way to do
716  * it more cleanly.
717  */
718 static bool
719 vnode_has_op(const struct vnode *vp, int opoffset)
720 {
721 	int (*defaultp)(void *);
722 	int (*opp)(void *);
723 
724 	defaultp = vp->v_op[VOFFSET(vop_default)];
725 	opp = vp->v_op[opoffset];
726 
727 	return opp != defaultp && opp != genfs_eopnotsupp &&
728 	    opp != genfs_badop && opp != genfs_nullop;
729 }
730 
731 /*
732  * Handes the read/write request given in 'bp' using the vnode's VOP_READ
733  * and VOP_WRITE operations.
734  *
735  * 'obp' is a pointer to the original request fed to the vnd device.
736  */
737 static void
738 handle_with_rdwr(struct vnd_softc *vnd, const struct buf *obp, struct buf *bp)
739 {
740 	bool doread;
741 	off_t offset;
742 	size_t len, resid;
743 	struct vnode *vp;
744 
745 	doread = bp->b_flags & B_READ;
746 	offset = obp->b_rawblkno * vnd->sc_dkdev.dk_label->d_secsize;
747 	len = bp->b_bcount;
748 	vp = vnd->sc_vp;
749 
750 #if defined(DEBUG)
751 	if (vnddebug & VDB_IO)
752 		printf("vnd (rdwr): vp %p, %s, rawblkno 0x%" PRIx64
753 		    ", secsize %d, offset %" PRIu64
754 		    ", bcount %d\n",
755 		    vp, doread ? "read" : "write", obp->b_rawblkno,
756 		    vnd->sc_dkdev.dk_label->d_secsize, offset,
757 		    bp->b_bcount);
758 #endif
759 
760 	/* Issue the read or write operation. */
761 	bp->b_error =
762 	    vn_rdwr(doread ? UIO_READ : UIO_WRITE,
763 	    vp, bp->b_data, len, offset, UIO_SYSSPACE,
764 	    IO_ADV_ENCODE(POSIX_FADV_NOREUSE), vnd->sc_cred, &resid, NULL);
765 	bp->b_resid = resid;
766 
767 	mutex_enter(vp->v_interlock);
768 	(void) VOP_PUTPAGES(vp, 0, 0,
769 	    PGO_ALLPAGES | PGO_CLEANIT | PGO_FREE | PGO_SYNCIO);
770 
771 	/* We need to increase the number of outputs on the vnode if
772 	 * there was any write to it. */
773 	if (!doread) {
774 		mutex_enter(vp->v_interlock);
775 		vp->v_numoutput++;
776 		mutex_exit(vp->v_interlock);
777 	}
778 
779 	biodone(bp);
780 }
781 
782 /*
783  * Handes the read/write request given in 'bp' using the vnode's VOP_BMAP
784  * and VOP_STRATEGY operations.
785  *
786  * 'obp' is a pointer to the original request fed to the vnd device.
787  */
788 static void
789 handle_with_strategy(struct vnd_softc *vnd, const struct buf *obp,
790     struct buf *bp)
791 {
792 	int bsize, error, flags, skipped;
793 	size_t resid, sz;
794 	off_t bn, offset;
795 	struct vnode *vp;
796 
797 	flags = obp->b_flags;
798 
799 	if (!(flags & B_READ)) {
800 		vp = bp->b_vp;
801 		mutex_enter(vp->v_interlock);
802 		vp->v_numoutput++;
803 		mutex_exit(vp->v_interlock);
804 	}
805 
806 	/* convert to a byte offset within the file. */
807 	bn = obp->b_rawblkno * vnd->sc_dkdev.dk_label->d_secsize;
808 
809 	bsize = vnd->sc_vp->v_mount->mnt_stat.f_iosize;
810 	skipped = 0;
811 
812 	/*
813 	 * Break the request into bsize pieces and feed them
814 	 * sequentially using VOP_BMAP/VOP_STRATEGY.
815 	 * We do it this way to keep from flooding NFS servers if we
816 	 * are connected to an NFS file.  This places the burden on
817 	 * the client rather than the server.
818 	 */
819 	error = 0;
820 	bp->b_resid = bp->b_bcount;
821 	for (offset = 0, resid = bp->b_resid; resid;
822 	    resid -= sz, offset += sz) {
823 		struct buf *nbp;
824 		daddr_t nbn;
825 		int off, nra;
826 
827 		nra = 0;
828 		vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
829 		error = VOP_BMAP(vnd->sc_vp, bn / bsize, &vp, &nbn, &nra);
830 		VOP_UNLOCK(vnd->sc_vp);
831 
832 		if (error == 0 && (long)nbn == -1)
833 			error = EIO;
834 
835 		/*
836 		 * If there was an error or a hole in the file...punt.
837 		 * Note that we may have to wait for any operations
838 		 * that we have already fired off before releasing
839 		 * the buffer.
840 		 *
841 		 * XXX we could deal with holes here but it would be
842 		 * a hassle (in the write case).
843 		 */
844 		if (error) {
845 			skipped += resid;
846 			break;
847 		}
848 
849 #ifdef DEBUG
850 		if (!dovndcluster)
851 			nra = 0;
852 #endif
853 
854 		off = bn % bsize;
855 		sz = MIN(((off_t)1 + nra) * bsize - off, resid);
856 #ifdef	DEBUG
857 		if (vnddebug & VDB_IO)
858 			printf("vndstrategy: vp %p/%p bn 0x%qx/0x%" PRIx64
859 			    " sz 0x%zx\n", vnd->sc_vp, vp, (long long)bn,
860 			    nbn, sz);
861 #endif
862 
863 		nbp = getiobuf(vp, true);
864 		nestiobuf_setup(bp, nbp, offset, sz);
865 		nbp->b_blkno = nbn + btodb(off);
866 
867 #if 0 /* XXX #ifdef DEBUG */
868 		if (vnddebug & VDB_IO)
869 			printf("vndstart(%ld): bp %p vp %p blkno "
870 			    "0x%" PRIx64 " flags %x addr %p cnt 0x%x\n",
871 			    (long) (vnd-vnd_softc), &nbp->vb_buf,
872 			    nbp->vb_buf.b_vp, nbp->vb_buf.b_blkno,
873 			    nbp->vb_buf.b_flags, nbp->vb_buf.b_data,
874 			    nbp->vb_buf.b_bcount);
875 #endif
876 		VOP_STRATEGY(vp, nbp);
877 		bn += sz;
878 	}
879 	nestiobuf_done(bp, skipped, error);
880 }
881 
882 static void
883 vndiodone(struct buf *bp)
884 {
885 	struct vndxfer *vnx = VND_BUFTOXFER(bp);
886 	struct vnd_softc *vnd = vnx->vx_vnd;
887 	struct buf *obp = bp->b_private;
888 	int s = splbio();
889 
890 	KASSERT(&vnx->vx_buf == bp);
891 	KASSERT(vnd->sc_active > 0);
892 #ifdef DEBUG
893 	if (vnddebug & VDB_IO) {
894 		printf("vndiodone1: bp %p iodone: error %d\n",
895 		    bp, bp->b_error);
896 	}
897 #endif
898 	disk_unbusy(&vnd->sc_dkdev, bp->b_bcount - bp->b_resid,
899 	    (bp->b_flags & B_READ));
900 	vnd->sc_active--;
901 	if (vnd->sc_active == 0) {
902 		wakeup(&vnd->sc_tab);
903 	}
904 	splx(s);
905 	obp->b_error = bp->b_error;
906 	obp->b_resid = bp->b_resid;
907 	buf_destroy(bp);
908 	VND_PUTXFER(vnd, vnx);
909 	biodone(obp);
910 }
911 
912 /* ARGSUSED */
913 static int
914 vndread(dev_t dev, struct uio *uio, int flags)
915 {
916 	int unit = vndunit(dev);
917 	struct vnd_softc *sc;
918 
919 #ifdef DEBUG
920 	if (vnddebug & VDB_FOLLOW)
921 		printf("vndread(0x%"PRIx64", %p)\n", dev, uio);
922 #endif
923 
924 	sc = device_lookup_private(&vnd_cd, unit);
925 	if (sc == NULL)
926 		return ENXIO;
927 
928 	if ((sc->sc_flags & VNF_INITED) == 0)
929 		return ENXIO;
930 
931 	return physio(vndstrategy, NULL, dev, B_READ, minphys, uio);
932 }
933 
934 /* ARGSUSED */
935 static int
936 vndwrite(dev_t dev, struct uio *uio, int flags)
937 {
938 	int unit = vndunit(dev);
939 	struct vnd_softc *sc;
940 
941 #ifdef DEBUG
942 	if (vnddebug & VDB_FOLLOW)
943 		printf("vndwrite(0x%"PRIx64", %p)\n", dev, uio);
944 #endif
945 
946 	sc = device_lookup_private(&vnd_cd, unit);
947 	if (sc == NULL)
948 		return ENXIO;
949 
950 	if ((sc->sc_flags & VNF_INITED) == 0)
951 		return ENXIO;
952 
953 	return physio(vndstrategy, NULL, dev, B_WRITE, minphys, uio);
954 }
955 
956 static int
957 vnd_cget(struct lwp *l, int unit, int *un, struct vattr *va)
958 {
959 	int error;
960 	struct vnd_softc *vnd;
961 
962 	if (*un == -1)
963 		*un = unit;
964 	if (*un < 0)
965 		return EINVAL;
966 
967 	vnd = device_lookup_private(&vnd_cd, *un);
968 	if (vnd == NULL)
969 		return -1;
970 
971 	if ((vnd->sc_flags & VNF_INITED) == 0)
972 		return -1;
973 
974 	vn_lock(vnd->sc_vp, LK_SHARED | LK_RETRY);
975 	error = VOP_GETATTR(vnd->sc_vp, va, l->l_cred);
976 	VOP_UNLOCK(vnd->sc_vp);
977 	return error;
978 }
979 
980 static int
981 vnddoclear(struct vnd_softc *vnd, int pmask, int minor, bool force)
982 {
983 	int error;
984 
985 	if ((error = vndlock(vnd)) != 0)
986 		return error;
987 
988 	/*
989 	 * Don't unconfigure if any other partitions are open
990 	 * or if both the character and block flavors of this
991 	 * partition are open.
992 	 */
993 	if (DK_BUSY(vnd, pmask) && !force) {
994 		vndunlock(vnd);
995 		return EBUSY;
996 	}
997 
998 	/* Delete all of our wedges */
999 	dkwedge_delall(&vnd->sc_dkdev);
1000 
1001 	/*
1002 	 * XXX vndclear() might call vndclose() implicitly;
1003 	 * release lock to avoid recursion
1004 	 *
1005 	 * Set VNF_CLEARING to prevent vndopen() from
1006 	 * sneaking in after we vndunlock().
1007 	 */
1008 	vnd->sc_flags |= VNF_CLEARING;
1009 	vndunlock(vnd);
1010 	vndclear(vnd, minor);
1011 #ifdef DEBUG
1012 	if (vnddebug & VDB_INIT)
1013 		printf("vndioctl: CLRed\n");
1014 #endif
1015 
1016 	/* Destroy the xfer and buffer pools. */
1017 	pool_destroy(&vnd->sc_vxpool);
1018 
1019 	/* Detach the disk. */
1020 	disk_detach(&vnd->sc_dkdev);
1021 
1022 	return 0;
1023 }
1024 
1025 /* ARGSUSED */
1026 static int
1027 vndioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l)
1028 {
1029 	bool force;
1030 	int unit = vndunit(dev);
1031 	struct vnd_softc *vnd;
1032 	struct vnd_ioctl *vio;
1033 	struct vattr vattr;
1034 	struct pathbuf *pb;
1035 	struct nameidata nd;
1036 	int error, part, pmask;
1037 	uint64_t geomsize;
1038 	int fflags;
1039 #ifdef __HAVE_OLD_DISKLABEL
1040 	struct disklabel newlabel;
1041 #endif
1042 	struct dkwedge_info *dkw;
1043 	struct dkwedge_list *dkwl;
1044 
1045 #ifdef DEBUG
1046 	if (vnddebug & VDB_FOLLOW)
1047 		printf("vndioctl(0x%"PRIx64", 0x%lx, %p, 0x%x, %p): unit %d\n",
1048 		    dev, cmd, data, flag, l->l_proc, unit);
1049 #endif
1050 	vnd = device_lookup_private(&vnd_cd, unit);
1051 	if (vnd == NULL &&
1052 #ifdef COMPAT_30
1053 	    cmd != VNDIOCGET30 &&
1054 #endif
1055 #ifdef COMPAT_50
1056 	    cmd != VNDIOCGET50 &&
1057 #endif
1058 	    cmd != VNDIOCGET)
1059 		return ENXIO;
1060 	vio = (struct vnd_ioctl *)data;
1061 
1062 	error = disk_ioctl(&vnd->sc_dkdev, cmd, data, flag, l);
1063 	if (error != EPASSTHROUGH)
1064 		return (error);
1065 
1066 	/* Must be open for writes for these commands... */
1067 	switch (cmd) {
1068 	case VNDIOCSET:
1069 	case VNDIOCCLR:
1070 #ifdef COMPAT_50
1071 	case VNDIOCSET50:
1072 	case VNDIOCCLR50:
1073 #endif
1074 	case DIOCSDINFO:
1075 	case DIOCWDINFO:
1076 #ifdef __HAVE_OLD_DISKLABEL
1077 	case ODIOCSDINFO:
1078 	case ODIOCWDINFO:
1079 #endif
1080 	case DIOCKLABEL:
1081 	case DIOCWLABEL:
1082 		if ((flag & FWRITE) == 0)
1083 			return EBADF;
1084 	}
1085 
1086 	/* Must be initialized for these... */
1087 	switch (cmd) {
1088 	case VNDIOCCLR:
1089 #ifdef VNDIOCCLR50
1090 	case VNDIOCCLR50:
1091 #endif
1092 	case DIOCGDINFO:
1093 	case DIOCSDINFO:
1094 	case DIOCWDINFO:
1095 	case DIOCGPART:
1096 	case DIOCKLABEL:
1097 	case DIOCWLABEL:
1098 	case DIOCGDEFLABEL:
1099 	case DIOCCACHESYNC:
1100 #ifdef __HAVE_OLD_DISKLABEL
1101 	case ODIOCGDINFO:
1102 	case ODIOCSDINFO:
1103 	case ODIOCWDINFO:
1104 	case ODIOCGDEFLABEL:
1105 #endif
1106 		if ((vnd->sc_flags & VNF_INITED) == 0)
1107 			return ENXIO;
1108 	}
1109 
1110 	switch (cmd) {
1111 #ifdef VNDIOCSET50
1112 	case VNDIOCSET50:
1113 #endif
1114 	case VNDIOCSET:
1115 		if (vnd->sc_flags & VNF_INITED)
1116 			return EBUSY;
1117 
1118 		if ((error = vndlock(vnd)) != 0)
1119 			return error;
1120 
1121 		fflags = FREAD;
1122 		if ((vio->vnd_flags & VNDIOF_READONLY) == 0)
1123 			fflags |= FWRITE;
1124 		error = pathbuf_copyin(vio->vnd_file, &pb);
1125 		if (error) {
1126 			goto unlock_and_exit;
1127 		}
1128 		NDINIT(&nd, LOOKUP, FOLLOW, pb);
1129 		if ((error = vn_open(&nd, fflags, 0)) != 0) {
1130 			pathbuf_destroy(pb);
1131 			goto unlock_and_exit;
1132 		}
1133 		KASSERT(l);
1134 		error = VOP_GETATTR(nd.ni_vp, &vattr, l->l_cred);
1135 		if (!error && nd.ni_vp->v_type != VREG)
1136 			error = EOPNOTSUPP;
1137 		if (!error && vattr.va_bytes < vattr.va_size)
1138 			/* File is definitely sparse, use vn_rdwr() */
1139 			vnd->sc_flags |= VNF_USE_VN_RDWR;
1140 		if (error) {
1141 			VOP_UNLOCK(nd.ni_vp);
1142 			goto close_and_exit;
1143 		}
1144 
1145 		/* If using a compressed file, initialize its info */
1146 		/* (or abort with an error if kernel has no compression) */
1147 		if (vio->vnd_flags & VNF_COMP) {
1148 #ifdef VND_COMPRESSION
1149 			struct vnd_comp_header *ch;
1150 			int i;
1151 			u_int32_t comp_size;
1152 			u_int32_t comp_maxsize;
1153 
1154 			/* allocate space for compresed file header */
1155 			ch = malloc(sizeof(struct vnd_comp_header),
1156 			M_TEMP, M_WAITOK);
1157 
1158 			/* read compressed file header */
1159 			error = vn_rdwr(UIO_READ, nd.ni_vp, (void *)ch,
1160 			  sizeof(struct vnd_comp_header), 0, UIO_SYSSPACE,
1161 			  IO_UNIT|IO_NODELOCKED, l->l_cred, NULL, NULL);
1162 			if (error) {
1163 				free(ch, M_TEMP);
1164 				VOP_UNLOCK(nd.ni_vp);
1165 				goto close_and_exit;
1166 			}
1167 
1168 			/* save some header info */
1169 			vnd->sc_comp_blksz = ntohl(ch->block_size);
1170 			/* note last offset is the file byte size */
1171 			vnd->sc_comp_numoffs = ntohl(ch->num_blocks)+1;
1172 			free(ch, M_TEMP);
1173 			if (vnd->sc_comp_blksz == 0 ||
1174 			    vnd->sc_comp_blksz % DEV_BSIZE !=0) {
1175 				VOP_UNLOCK(nd.ni_vp);
1176 				error = EINVAL;
1177 				goto close_and_exit;
1178 			}
1179 			if (sizeof(struct vnd_comp_header) +
1180 			  sizeof(u_int64_t) * vnd->sc_comp_numoffs >
1181 			  vattr.va_size) {
1182 				VOP_UNLOCK(nd.ni_vp);
1183 				error = EINVAL;
1184 				goto close_and_exit;
1185 			}
1186 
1187 			/* set decompressed file size */
1188 			vattr.va_size =
1189 			    ((u_quad_t)vnd->sc_comp_numoffs - 1) *
1190 			     (u_quad_t)vnd->sc_comp_blksz;
1191 
1192 			/* allocate space for all the compressed offsets */
1193 			vnd->sc_comp_offsets =
1194 			malloc(sizeof(u_int64_t) * vnd->sc_comp_numoffs,
1195 			M_DEVBUF, M_WAITOK);
1196 
1197 			/* read in the offsets */
1198 			error = vn_rdwr(UIO_READ, nd.ni_vp,
1199 			  (void *)vnd->sc_comp_offsets,
1200 			  sizeof(u_int64_t) * vnd->sc_comp_numoffs,
1201 			  sizeof(struct vnd_comp_header), UIO_SYSSPACE,
1202 			  IO_UNIT|IO_NODELOCKED, l->l_cred, NULL, NULL);
1203 			if (error) {
1204 				VOP_UNLOCK(nd.ni_vp);
1205 				goto close_and_exit;
1206 			}
1207 			/*
1208 			 * find largest block size (used for allocation limit).
1209 			 * Also convert offset to native byte order.
1210 			 */
1211 			comp_maxsize = 0;
1212 			for (i = 0; i < vnd->sc_comp_numoffs - 1; i++) {
1213 				vnd->sc_comp_offsets[i] =
1214 				  be64toh(vnd->sc_comp_offsets[i]);
1215 				comp_size = be64toh(vnd->sc_comp_offsets[i + 1])
1216 				  - vnd->sc_comp_offsets[i];
1217 				if (comp_size > comp_maxsize)
1218 					comp_maxsize = comp_size;
1219 			}
1220 			vnd->sc_comp_offsets[vnd->sc_comp_numoffs - 1] =
1221 			  be64toh(vnd->sc_comp_offsets[vnd->sc_comp_numoffs - 1]);
1222 
1223 			/* create compressed data buffer */
1224 			vnd->sc_comp_buff = malloc(comp_maxsize,
1225 			  M_DEVBUF, M_WAITOK);
1226 
1227 			/* create decompressed buffer */
1228 			vnd->sc_comp_decombuf = malloc(vnd->sc_comp_blksz,
1229 			  M_DEVBUF, M_WAITOK);
1230 			vnd->sc_comp_buffblk = -1;
1231 
1232 			/* Initialize decompress stream */
1233 			memset(&vnd->sc_comp_stream, 0, sizeof(z_stream));
1234 			vnd->sc_comp_stream.zalloc = vnd_alloc;
1235 			vnd->sc_comp_stream.zfree = vnd_free;
1236 			error = inflateInit2(&vnd->sc_comp_stream, MAX_WBITS);
1237 			if (error) {
1238 				if (vnd->sc_comp_stream.msg)
1239 					printf("vnd%d: compressed file, %s\n",
1240 					  unit, vnd->sc_comp_stream.msg);
1241 				VOP_UNLOCK(nd.ni_vp);
1242 				error = EINVAL;
1243 				goto close_and_exit;
1244 			}
1245 
1246 			vnd->sc_flags |= VNF_COMP | VNF_READONLY;
1247 #else /* !VND_COMPRESSION */
1248 			VOP_UNLOCK(nd.ni_vp);
1249 			error = EOPNOTSUPP;
1250 			goto close_and_exit;
1251 #endif /* VND_COMPRESSION */
1252 		}
1253 
1254 		VOP_UNLOCK(nd.ni_vp);
1255 		vnd->sc_vp = nd.ni_vp;
1256 		vnd->sc_size = btodb(vattr.va_size);	/* note truncation */
1257 
1258 		/*
1259 		 * Use pseudo-geometry specified.  If none was provided,
1260 		 * use "standard" Adaptec fictitious geometry.
1261 		 */
1262 		if (vio->vnd_flags & VNDIOF_HASGEOM) {
1263 
1264 			memcpy(&vnd->sc_geom, &vio->vnd_geom,
1265 			    sizeof(vio->vnd_geom));
1266 
1267 			/*
1268 			 * Sanity-check the sector size.
1269 			 * XXX Don't allow secsize < DEV_BSIZE.	 Should
1270 			 * XXX we?
1271 			 */
1272 			if (vnd->sc_geom.vng_secsize < DEV_BSIZE ||
1273 			    (vnd->sc_geom.vng_secsize % DEV_BSIZE) != 0 ||
1274 			    vnd->sc_geom.vng_ncylinders == 0 ||
1275 			    (vnd->sc_geom.vng_ntracks *
1276 			     vnd->sc_geom.vng_nsectors) == 0) {
1277 				error = EINVAL;
1278 				goto close_and_exit;
1279 			}
1280 
1281 			/*
1282 			 * Compute the size (in DEV_BSIZE blocks) specified
1283 			 * by the geometry.
1284 			 */
1285 			geomsize = (vnd->sc_geom.vng_nsectors *
1286 			    vnd->sc_geom.vng_ntracks *
1287 			    vnd->sc_geom.vng_ncylinders) *
1288 			    (vnd->sc_geom.vng_secsize / DEV_BSIZE);
1289 
1290 			/*
1291 			 * Sanity-check the size against the specified
1292 			 * geometry.
1293 			 */
1294 			if (vnd->sc_size < geomsize) {
1295 				error = EINVAL;
1296 				goto close_and_exit;
1297 			}
1298 		} else if (vnd->sc_size >= (32 * 64)) {
1299 			/*
1300 			 * Size must be at least 2048 DEV_BSIZE blocks
1301 			 * (1M) in order to use this geometry.
1302 			 */
1303 			vnd->sc_geom.vng_secsize = DEV_BSIZE;
1304 			vnd->sc_geom.vng_nsectors = 32;
1305 			vnd->sc_geom.vng_ntracks = 64;
1306 			vnd->sc_geom.vng_ncylinders = vnd->sc_size / (64 * 32);
1307 		} else {
1308 			vnd->sc_geom.vng_secsize = DEV_BSIZE;
1309 			vnd->sc_geom.vng_nsectors = 1;
1310 			vnd->sc_geom.vng_ntracks = 1;
1311 			vnd->sc_geom.vng_ncylinders = vnd->sc_size;
1312 		}
1313 
1314 		vnd_set_geometry(vnd);
1315 
1316 		if (vio->vnd_flags & VNDIOF_READONLY) {
1317 			vnd->sc_flags |= VNF_READONLY;
1318 		}
1319 
1320 		if ((error = vndsetcred(vnd, l->l_cred)) != 0)
1321 			goto close_and_exit;
1322 
1323 		vndthrottle(vnd, vnd->sc_vp);
1324 		vio->vnd_osize = dbtob(vnd->sc_size);
1325 #ifdef VNDIOCSET50
1326 		if (cmd != VNDIOCSET50)
1327 #endif
1328 			vio->vnd_size = dbtob(vnd->sc_size);
1329 		vnd->sc_flags |= VNF_INITED;
1330 
1331 		/* create the kernel thread, wait for it to be up */
1332 		error = kthread_create(PRI_NONE, 0, NULL, vndthread, vnd,
1333 		    &vnd->sc_kthread, "%s", device_xname(vnd->sc_dev));
1334 		if (error)
1335 			goto close_and_exit;
1336 		while ((vnd->sc_flags & VNF_KTHREAD) == 0) {
1337 			tsleep(&vnd->sc_kthread, PRIBIO, "vndthr", 0);
1338 		}
1339 #ifdef DEBUG
1340 		if (vnddebug & VDB_INIT)
1341 			printf("vndioctl: SET vp %p size 0x%lx %d/%d/%d/%d\n",
1342 			    vnd->sc_vp, (unsigned long) vnd->sc_size,
1343 			    vnd->sc_geom.vng_secsize,
1344 			    vnd->sc_geom.vng_nsectors,
1345 			    vnd->sc_geom.vng_ntracks,
1346 			    vnd->sc_geom.vng_ncylinders);
1347 #endif
1348 
1349 		/* Attach the disk. */
1350 		disk_attach(&vnd->sc_dkdev);
1351 		disk_blocksize(&vnd->sc_dkdev, vnd->sc_geom.vng_secsize);
1352 
1353 		/* Initialize the xfer and buffer pools. */
1354 		pool_init(&vnd->sc_vxpool, sizeof(struct vndxfer), 0,
1355 		    0, 0, "vndxpl", NULL, IPL_BIO);
1356 
1357 		vndunlock(vnd);
1358 
1359 		pathbuf_destroy(pb);
1360 
1361 		/* Discover wedges on this disk */
1362 		dkwedge_discover(&vnd->sc_dkdev);
1363 
1364 		break;
1365 
1366 close_and_exit:
1367 		(void) vn_close(nd.ni_vp, fflags, l->l_cred);
1368 		pathbuf_destroy(pb);
1369 unlock_and_exit:
1370 #ifdef VND_COMPRESSION
1371 		/* free any allocated memory (for compressed file) */
1372 		if (vnd->sc_comp_offsets) {
1373 			free(vnd->sc_comp_offsets, M_DEVBUF);
1374 			vnd->sc_comp_offsets = NULL;
1375 		}
1376 		if (vnd->sc_comp_buff) {
1377 			free(vnd->sc_comp_buff, M_DEVBUF);
1378 			vnd->sc_comp_buff = NULL;
1379 		}
1380 		if (vnd->sc_comp_decombuf) {
1381 			free(vnd->sc_comp_decombuf, M_DEVBUF);
1382 			vnd->sc_comp_decombuf = NULL;
1383 		}
1384 #endif /* VND_COMPRESSION */
1385 		vndunlock(vnd);
1386 		return error;
1387 
1388 #ifdef VNDIOCCLR50
1389 	case VNDIOCCLR50:
1390 #endif
1391 	case VNDIOCCLR:
1392 		part = DISKPART(dev);
1393 		pmask = (1 << part);
1394 		force = (vio->vnd_flags & VNDIOF_FORCE) != 0;
1395 
1396 		if ((error = vnddoclear(vnd, pmask, minor(dev), force)) != 0)
1397 			return error;
1398 
1399 		break;
1400 
1401 #ifdef COMPAT_30
1402 	case VNDIOCGET30: {
1403 		struct vnd_user30 *vnu;
1404 		struct vattr va;
1405 		vnu = (struct vnd_user30 *)data;
1406 		KASSERT(l);
1407 		switch (error = vnd_cget(l, unit, &vnu->vnu_unit, &va)) {
1408 		case 0:
1409 			vnu->vnu_dev = va.va_fsid;
1410 			vnu->vnu_ino = va.va_fileid;
1411 			break;
1412 		case -1:
1413 			/* unused is not an error */
1414 			vnu->vnu_dev = 0;
1415 			vnu->vnu_ino = 0;
1416 			break;
1417 		default:
1418 			return error;
1419 		}
1420 		break;
1421 	}
1422 #endif
1423 
1424 #ifdef COMPAT_50
1425 	case VNDIOCGET50: {
1426 		struct vnd_user50 *vnu;
1427 		struct vattr va;
1428 		vnu = (struct vnd_user50 *)data;
1429 		KASSERT(l);
1430 		switch (error = vnd_cget(l, unit, &vnu->vnu_unit, &va)) {
1431 		case 0:
1432 			vnu->vnu_dev = va.va_fsid;
1433 			vnu->vnu_ino = va.va_fileid;
1434 			break;
1435 		case -1:
1436 			/* unused is not an error */
1437 			vnu->vnu_dev = 0;
1438 			vnu->vnu_ino = 0;
1439 			break;
1440 		default:
1441 			return error;
1442 		}
1443 		break;
1444 	}
1445 #endif
1446 
1447 	case VNDIOCGET: {
1448 		struct vnd_user *vnu;
1449 		struct vattr va;
1450 		vnu = (struct vnd_user *)data;
1451 		KASSERT(l);
1452 		switch (error = vnd_cget(l, unit, &vnu->vnu_unit, &va)) {
1453 		case 0:
1454 			vnu->vnu_dev = va.va_fsid;
1455 			vnu->vnu_ino = va.va_fileid;
1456 			break;
1457 		case -1:
1458 			/* unused is not an error */
1459 			vnu->vnu_dev = 0;
1460 			vnu->vnu_ino = 0;
1461 			break;
1462 		default:
1463 			return error;
1464 		}
1465 		break;
1466 	}
1467 
1468 	case DIOCGDINFO:
1469 		*(struct disklabel *)data = *(vnd->sc_dkdev.dk_label);
1470 		break;
1471 
1472 #ifdef __HAVE_OLD_DISKLABEL
1473 	case ODIOCGDINFO:
1474 		newlabel = *(vnd->sc_dkdev.dk_label);
1475 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1476 			return ENOTTY;
1477 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1478 		break;
1479 #endif
1480 
1481 	case DIOCGPART:
1482 		((struct partinfo *)data)->disklab = vnd->sc_dkdev.dk_label;
1483 		((struct partinfo *)data)->part =
1484 		    &vnd->sc_dkdev.dk_label->d_partitions[DISKPART(dev)];
1485 		break;
1486 
1487 	case DIOCWDINFO:
1488 	case DIOCSDINFO:
1489 #ifdef __HAVE_OLD_DISKLABEL
1490 	case ODIOCWDINFO:
1491 	case ODIOCSDINFO:
1492 #endif
1493 	{
1494 		struct disklabel *lp;
1495 
1496 		if ((error = vndlock(vnd)) != 0)
1497 			return error;
1498 
1499 		vnd->sc_flags |= VNF_LABELLING;
1500 
1501 #ifdef __HAVE_OLD_DISKLABEL
1502 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
1503 			memset(&newlabel, 0, sizeof newlabel);
1504 			memcpy(&newlabel, data, sizeof (struct olddisklabel));
1505 			lp = &newlabel;
1506 		} else
1507 #endif
1508 		lp = (struct disklabel *)data;
1509 
1510 		error = setdisklabel(vnd->sc_dkdev.dk_label,
1511 		    lp, 0, vnd->sc_dkdev.dk_cpulabel);
1512 		if (error == 0) {
1513 			if (cmd == DIOCWDINFO
1514 #ifdef __HAVE_OLD_DISKLABEL
1515 			    || cmd == ODIOCWDINFO
1516 #endif
1517 			   )
1518 				error = writedisklabel(VNDLABELDEV(dev),
1519 				    vndstrategy, vnd->sc_dkdev.dk_label,
1520 				    vnd->sc_dkdev.dk_cpulabel);
1521 		}
1522 
1523 		vnd->sc_flags &= ~VNF_LABELLING;
1524 
1525 		vndunlock(vnd);
1526 
1527 		if (error)
1528 			return error;
1529 		break;
1530 	}
1531 
1532 	case DIOCKLABEL:
1533 		if (*(int *)data != 0)
1534 			vnd->sc_flags |= VNF_KLABEL;
1535 		else
1536 			vnd->sc_flags &= ~VNF_KLABEL;
1537 		break;
1538 
1539 	case DIOCWLABEL:
1540 		if (*(int *)data != 0)
1541 			vnd->sc_flags |= VNF_WLABEL;
1542 		else
1543 			vnd->sc_flags &= ~VNF_WLABEL;
1544 		break;
1545 
1546 	case DIOCGDEFLABEL:
1547 		vndgetdefaultlabel(vnd, (struct disklabel *)data);
1548 		break;
1549 
1550 #ifdef __HAVE_OLD_DISKLABEL
1551 	case ODIOCGDEFLABEL:
1552 		vndgetdefaultlabel(vnd, &newlabel);
1553 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1554 			return ENOTTY;
1555 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1556 		break;
1557 #endif
1558 
1559 	case DIOCCACHESYNC:
1560 		vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1561 		error = VOP_FSYNC(vnd->sc_vp, vnd->sc_cred,
1562 		    FSYNC_WAIT | FSYNC_DATAONLY | FSYNC_CACHE, 0, 0);
1563 		VOP_UNLOCK(vnd->sc_vp);
1564 		return error;
1565 
1566 	case DIOCAWEDGE:
1567 		dkw = (void *) data;
1568 
1569 		if ((flag & FWRITE) == 0)
1570 			return EBADF;
1571 
1572 		/* If the ioctl happens here, the parent is us. */
1573 		strlcpy(dkw->dkw_parent, device_xname(vnd->sc_dev),
1574 		    sizeof(dkw->dkw_parent));
1575 		return dkwedge_add(dkw);
1576 
1577 	case DIOCDWEDGE:
1578 		dkw = (void *) data;
1579 
1580 		if ((flag & FWRITE) == 0)
1581 			return EBADF;
1582 
1583 		/* If the ioctl happens here, the parent is us. */
1584 		strlcpy(dkw->dkw_parent, device_xname(vnd->sc_dev),
1585 		    sizeof(dkw->dkw_parent));
1586 		return dkwedge_del(dkw);
1587 
1588 	case DIOCLWEDGES:
1589 		dkwl = (void *) data;
1590 
1591 		return dkwedge_list(&vnd->sc_dkdev, dkwl, l);
1592 
1593 	default:
1594 		return ENOTTY;
1595 	}
1596 
1597 	return 0;
1598 }
1599 
1600 /*
1601  * Duplicate the current processes' credentials.  Since we are called only
1602  * as the result of a SET ioctl and only root can do that, any future access
1603  * to this "disk" is essentially as root.  Note that credentials may change
1604  * if some other uid can write directly to the mapped file (NFS).
1605  */
1606 static int
1607 vndsetcred(struct vnd_softc *vnd, kauth_cred_t cred)
1608 {
1609 	struct uio auio;
1610 	struct iovec aiov;
1611 	char *tmpbuf;
1612 	int error;
1613 
1614 	vnd->sc_cred = kauth_cred_dup(cred);
1615 	tmpbuf = malloc(DEV_BSIZE, M_TEMP, M_WAITOK);
1616 
1617 	/* XXX: Horrible kludge to establish credentials for NFS */
1618 	aiov.iov_base = tmpbuf;
1619 	aiov.iov_len = min(DEV_BSIZE, dbtob(vnd->sc_size));
1620 	auio.uio_iov = &aiov;
1621 	auio.uio_iovcnt = 1;
1622 	auio.uio_offset = 0;
1623 	auio.uio_rw = UIO_READ;
1624 	auio.uio_resid = aiov.iov_len;
1625 	UIO_SETUP_SYSSPACE(&auio);
1626 	vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1627 	error = VOP_READ(vnd->sc_vp, &auio, 0, vnd->sc_cred);
1628 	if (error == 0) {
1629 		/*
1630 		 * Because vnd does all IO directly through the vnode
1631 		 * we need to flush (at least) the buffer from the above
1632 		 * VOP_READ from the buffer cache to prevent cache
1633 		 * incoherencies.  Also, be careful to write dirty
1634 		 * buffers back to stable storage.
1635 		 */
1636 		error = vinvalbuf(vnd->sc_vp, V_SAVE, vnd->sc_cred,
1637 			    curlwp, 0, 0);
1638 	}
1639 	VOP_UNLOCK(vnd->sc_vp);
1640 
1641 	free(tmpbuf, M_TEMP);
1642 	return error;
1643 }
1644 
1645 /*
1646  * Set maxactive based on FS type
1647  */
1648 static void
1649 vndthrottle(struct vnd_softc *vnd, struct vnode *vp)
1650 {
1651 
1652 	if (vp->v_tag == VT_NFS)
1653 		vnd->sc_maxactive = 2;
1654 	else
1655 		vnd->sc_maxactive = 8;
1656 
1657 	if (vnd->sc_maxactive < 1)
1658 		vnd->sc_maxactive = 1;
1659 }
1660 
1661 #if 0
1662 static void
1663 vndshutdown(void)
1664 {
1665 	struct vnd_softc *vnd;
1666 
1667 	for (vnd = &vnd_softc[0]; vnd < &vnd_softc[numvnd]; vnd++)
1668 		if (vnd->sc_flags & VNF_INITED)
1669 			vndclear(vnd);
1670 }
1671 #endif
1672 
1673 static void
1674 vndclear(struct vnd_softc *vnd, int myminor)
1675 {
1676 	struct vnode *vp = vnd->sc_vp;
1677 	int fflags = FREAD;
1678 	int bmaj, cmaj, i, mn;
1679 	int s;
1680 
1681 #ifdef DEBUG
1682 	if (vnddebug & VDB_FOLLOW)
1683 		printf("vndclear(%p): vp %p\n", vnd, vp);
1684 #endif
1685 	/* locate the major number */
1686 	bmaj = bdevsw_lookup_major(&vnd_bdevsw);
1687 	cmaj = cdevsw_lookup_major(&vnd_cdevsw);
1688 
1689 	/* Nuke the vnodes for any open instances */
1690 	for (i = 0; i < MAXPARTITIONS; i++) {
1691 		mn = DISKMINOR(device_unit(vnd->sc_dev), i);
1692 		vdevgone(bmaj, mn, mn, VBLK);
1693 		if (mn != myminor) /* XXX avoid to kill own vnode */
1694 			vdevgone(cmaj, mn, mn, VCHR);
1695 	}
1696 
1697 	if ((vnd->sc_flags & VNF_READONLY) == 0)
1698 		fflags |= FWRITE;
1699 
1700 	s = splbio();
1701 	bufq_drain(vnd->sc_tab);
1702 	splx(s);
1703 
1704 	vnd->sc_flags |= VNF_VUNCONF;
1705 	wakeup(&vnd->sc_tab);
1706 	while (vnd->sc_flags & VNF_KTHREAD)
1707 		tsleep(&vnd->sc_kthread, PRIBIO, "vnthr", 0);
1708 
1709 #ifdef VND_COMPRESSION
1710 	/* free the compressed file buffers */
1711 	if (vnd->sc_flags & VNF_COMP) {
1712 		if (vnd->sc_comp_offsets) {
1713 			free(vnd->sc_comp_offsets, M_DEVBUF);
1714 			vnd->sc_comp_offsets = NULL;
1715 		}
1716 		if (vnd->sc_comp_buff) {
1717 			free(vnd->sc_comp_buff, M_DEVBUF);
1718 			vnd->sc_comp_buff = NULL;
1719 		}
1720 		if (vnd->sc_comp_decombuf) {
1721 			free(vnd->sc_comp_decombuf, M_DEVBUF);
1722 			vnd->sc_comp_decombuf = NULL;
1723 		}
1724 	}
1725 #endif /* VND_COMPRESSION */
1726 	vnd->sc_flags &=
1727 	    ~(VNF_INITED | VNF_READONLY | VNF_VLABEL
1728 	      | VNF_VUNCONF | VNF_COMP | VNF_CLEARING);
1729 	if (vp == NULL)
1730 		panic("vndclear: null vp");
1731 	(void) vn_close(vp, fflags, vnd->sc_cred);
1732 	kauth_cred_free(vnd->sc_cred);
1733 	vnd->sc_vp = NULL;
1734 	vnd->sc_cred = NULL;
1735 	vnd->sc_size = 0;
1736 }
1737 
1738 static int
1739 vndsize(dev_t dev)
1740 {
1741 	struct vnd_softc *sc;
1742 	struct disklabel *lp;
1743 	int part, unit, omask;
1744 	int size;
1745 
1746 	unit = vndunit(dev);
1747 	sc = device_lookup_private(&vnd_cd, unit);
1748 	if (sc == NULL)
1749 		return -1;
1750 
1751 	if ((sc->sc_flags & VNF_INITED) == 0)
1752 		return -1;
1753 
1754 	part = DISKPART(dev);
1755 	omask = sc->sc_dkdev.dk_openmask & (1 << part);
1756 	lp = sc->sc_dkdev.dk_label;
1757 
1758 	if (omask == 0 && vndopen(dev, 0, S_IFBLK, curlwp))	/* XXX */
1759 		return -1;
1760 
1761 	if (lp->d_partitions[part].p_fstype != FS_SWAP)
1762 		size = -1;
1763 	else
1764 		size = lp->d_partitions[part].p_size *
1765 		    (lp->d_secsize / DEV_BSIZE);
1766 
1767 	if (omask == 0 && vndclose(dev, 0, S_IFBLK, curlwp))	/* XXX */
1768 		return -1;
1769 
1770 	return size;
1771 }
1772 
1773 static int
1774 vnddump(dev_t dev, daddr_t blkno, void *va,
1775     size_t size)
1776 {
1777 
1778 	/* Not implemented. */
1779 	return ENXIO;
1780 }
1781 
1782 static void
1783 vndgetdefaultlabel(struct vnd_softc *sc, struct disklabel *lp)
1784 {
1785 	struct vndgeom *vng = &sc->sc_geom;
1786 	struct partition *pp;
1787 
1788 	memset(lp, 0, sizeof(*lp));
1789 
1790 	lp->d_secperunit = sc->sc_size / (vng->vng_secsize / DEV_BSIZE);
1791 	lp->d_secsize = vng->vng_secsize;
1792 	lp->d_nsectors = vng->vng_nsectors;
1793 	lp->d_ntracks = vng->vng_ntracks;
1794 	lp->d_ncylinders = vng->vng_ncylinders;
1795 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1796 
1797 	strncpy(lp->d_typename, "vnd", sizeof(lp->d_typename));
1798 	lp->d_type = DTYPE_VND;
1799 	strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
1800 	lp->d_rpm = 3600;
1801 	lp->d_interleave = 1;
1802 	lp->d_flags = 0;
1803 
1804 	pp = &lp->d_partitions[RAW_PART];
1805 	pp->p_offset = 0;
1806 	pp->p_size = lp->d_secperunit;
1807 	pp->p_fstype = FS_UNUSED;
1808 	lp->d_npartitions = RAW_PART + 1;
1809 
1810 	lp->d_magic = DISKMAGIC;
1811 	lp->d_magic2 = DISKMAGIC;
1812 	lp->d_checksum = dkcksum(lp);
1813 }
1814 
1815 /*
1816  * Read the disklabel from a vnd.  If one is not present, create a fake one.
1817  */
1818 static void
1819 vndgetdisklabel(dev_t dev, struct vnd_softc *sc)
1820 {
1821 	const char *errstring;
1822 	struct disklabel *lp = sc->sc_dkdev.dk_label;
1823 	struct cpu_disklabel *clp = sc->sc_dkdev.dk_cpulabel;
1824 	int i;
1825 
1826 	memset(clp, 0, sizeof(*clp));
1827 
1828 	vndgetdefaultlabel(sc, lp);
1829 
1830 	/*
1831 	 * Call the generic disklabel extraction routine.
1832 	 */
1833 	errstring = readdisklabel(VNDLABELDEV(dev), vndstrategy, lp, clp);
1834 	if (errstring) {
1835 		/*
1836 		 * Lack of disklabel is common, but we print the warning
1837 		 * anyway, since it might contain other useful information.
1838 		 */
1839 		aprint_normal_dev(sc->sc_dev, "%s\n", errstring);
1840 
1841 		/*
1842 		 * For historical reasons, if there's no disklabel
1843 		 * present, all partitions must be FS_BSDFFS and
1844 		 * occupy the entire disk.
1845 		 */
1846 		for (i = 0; i < MAXPARTITIONS; i++) {
1847 			/*
1848 			 * Don't wipe out port specific hack (such as
1849 			 * dos partition hack of i386 port).
1850 			 */
1851 			if (lp->d_partitions[i].p_size != 0)
1852 				continue;
1853 
1854 			lp->d_partitions[i].p_size = lp->d_secperunit;
1855 			lp->d_partitions[i].p_offset = 0;
1856 			lp->d_partitions[i].p_fstype = FS_BSDFFS;
1857 		}
1858 
1859 		strncpy(lp->d_packname, "default label",
1860 		    sizeof(lp->d_packname));
1861 
1862 		lp->d_npartitions = MAXPARTITIONS;
1863 		lp->d_checksum = dkcksum(lp);
1864 	}
1865 }
1866 
1867 /*
1868  * Wait interruptibly for an exclusive lock.
1869  *
1870  * XXX
1871  * Several drivers do this; it should be abstracted and made MP-safe.
1872  */
1873 static int
1874 vndlock(struct vnd_softc *sc)
1875 {
1876 	int error;
1877 
1878 	while ((sc->sc_flags & VNF_LOCKED) != 0) {
1879 		sc->sc_flags |= VNF_WANTED;
1880 		if ((error = tsleep(sc, PRIBIO | PCATCH, "vndlck", 0)) != 0)
1881 			return error;
1882 	}
1883 	sc->sc_flags |= VNF_LOCKED;
1884 	return 0;
1885 }
1886 
1887 /*
1888  * Unlock and wake up any waiters.
1889  */
1890 static void
1891 vndunlock(struct vnd_softc *sc)
1892 {
1893 
1894 	sc->sc_flags &= ~VNF_LOCKED;
1895 	if ((sc->sc_flags & VNF_WANTED) != 0) {
1896 		sc->sc_flags &= ~VNF_WANTED;
1897 		wakeup(sc);
1898 	}
1899 }
1900 
1901 #ifdef VND_COMPRESSION
1902 /* compressed file read */
1903 static void
1904 compstrategy(struct buf *bp, off_t bn)
1905 {
1906 	int error;
1907 	int unit = vndunit(bp->b_dev);
1908 	struct vnd_softc *vnd =
1909 	    device_lookup_private(&vnd_cd, unit);
1910 	u_int32_t comp_block;
1911 	struct uio auio;
1912 	char *addr;
1913 	int s;
1914 
1915 	/* set up constants for data move */
1916 	auio.uio_rw = UIO_READ;
1917 	UIO_SETUP_SYSSPACE(&auio);
1918 
1919 	/* read, and transfer the data */
1920 	addr = bp->b_data;
1921 	bp->b_resid = bp->b_bcount;
1922 	s = splbio();
1923 	while (bp->b_resid > 0) {
1924 		unsigned length;
1925 		size_t length_in_buffer;
1926 		u_int32_t offset_in_buffer;
1927 		struct iovec aiov;
1928 
1929 		/* calculate the compressed block number */
1930 		comp_block = bn / (off_t)vnd->sc_comp_blksz;
1931 
1932 		/* check for good block number */
1933 		if (comp_block >= vnd->sc_comp_numoffs) {
1934 			bp->b_error = EINVAL;
1935 			splx(s);
1936 			return;
1937 		}
1938 
1939 		/* read in the compressed block, if not in buffer */
1940 		if (comp_block != vnd->sc_comp_buffblk) {
1941 			length = vnd->sc_comp_offsets[comp_block + 1] -
1942 			    vnd->sc_comp_offsets[comp_block];
1943 			vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1944 			error = vn_rdwr(UIO_READ, vnd->sc_vp, vnd->sc_comp_buff,
1945 			    length, vnd->sc_comp_offsets[comp_block],
1946 			    UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, vnd->sc_cred,
1947 			    NULL, NULL);
1948 			if (error) {
1949 				bp->b_error = error;
1950 				VOP_UNLOCK(vnd->sc_vp);
1951 				splx(s);
1952 				return;
1953 			}
1954 			/* uncompress the buffer */
1955 			vnd->sc_comp_stream.next_in = vnd->sc_comp_buff;
1956 			vnd->sc_comp_stream.avail_in = length;
1957 			vnd->sc_comp_stream.next_out = vnd->sc_comp_decombuf;
1958 			vnd->sc_comp_stream.avail_out = vnd->sc_comp_blksz;
1959 			inflateReset(&vnd->sc_comp_stream);
1960 			error = inflate(&vnd->sc_comp_stream, Z_FINISH);
1961 			if (error != Z_STREAM_END) {
1962 				if (vnd->sc_comp_stream.msg)
1963 					aprint_normal_dev(vnd->sc_dev,
1964 					    "compressed file, %s\n",
1965 					    vnd->sc_comp_stream.msg);
1966 				bp->b_error = EBADMSG;
1967 				VOP_UNLOCK(vnd->sc_vp);
1968 				splx(s);
1969 				return;
1970 			}
1971 			vnd->sc_comp_buffblk = comp_block;
1972 			VOP_UNLOCK(vnd->sc_vp);
1973 		}
1974 
1975 		/* transfer the usable uncompressed data */
1976 		offset_in_buffer = bn % (off_t)vnd->sc_comp_blksz;
1977 		length_in_buffer = vnd->sc_comp_blksz - offset_in_buffer;
1978 		if (length_in_buffer > bp->b_resid)
1979 			length_in_buffer = bp->b_resid;
1980 		auio.uio_iov = &aiov;
1981 		auio.uio_iovcnt = 1;
1982 		aiov.iov_base = addr;
1983 		aiov.iov_len = length_in_buffer;
1984 		auio.uio_resid = aiov.iov_len;
1985 		auio.uio_offset = 0;
1986 		error = uiomove(vnd->sc_comp_decombuf + offset_in_buffer,
1987 		    length_in_buffer, &auio);
1988 		if (error) {
1989 			bp->b_error = error;
1990 			splx(s);
1991 			return;
1992 		}
1993 
1994 		bn += length_in_buffer;
1995 		addr += length_in_buffer;
1996 		bp->b_resid -= length_in_buffer;
1997 	}
1998 	splx(s);
1999 }
2000 
2001 /* compression memory allocation routines */
2002 static void *
2003 vnd_alloc(void *aux, u_int items, u_int siz)
2004 {
2005 	return malloc(items * siz, M_TEMP, M_NOWAIT);
2006 }
2007 
2008 static void
2009 vnd_free(void *aux, void *ptr)
2010 {
2011 	free(ptr, M_TEMP);
2012 }
2013 #endif /* VND_COMPRESSION */
2014 
2015 static void
2016 vnd_set_geometry(struct vnd_softc *vnd)
2017 {
2018 	struct disk_geom *dg = &vnd->sc_dkdev.dk_geom;
2019 
2020 	memset(dg, 0, sizeof(*dg));
2021 
2022 	dg->dg_secperunit = (int64_t)vnd->sc_geom.vng_nsectors *
2023 	    vnd->sc_geom.vng_ntracks * vnd->sc_geom.vng_ncylinders;
2024 	dg->dg_secsize = vnd->sc_geom.vng_secsize;
2025 	dg->dg_nsectors = vnd->sc_geom.vng_nsectors;
2026 	dg->dg_ntracks = vnd->sc_geom.vng_ntracks;
2027 	dg->dg_ncylinders = vnd->sc_geom.vng_ncylinders;
2028 
2029 #ifdef DEBUG
2030 	if (vnddebug & VDB_LABEL) {
2031 		printf("dg->dg_secperunit: %" PRId64 "\n", dg->dg_secperunit);
2032 		printf("dg->dg_ncylinders: %u\n", dg->dg_ncylinders);
2033 	}
2034 #endif
2035 	disk_set_info(vnd->sc_dev, &vnd->sc_dkdev, NULL);
2036 }
2037 
2038 #ifdef _MODULE
2039 
2040 #include <sys/module.h>
2041 
2042 MODULE(MODULE_CLASS_DRIVER, vnd, "zlib");
2043 CFDRIVER_DECL(vnd, DV_DISK, NULL);
2044 
2045 static int
2046 vnd_modcmd(modcmd_t cmd, void *arg)
2047 {
2048 	int bmajor = -1, cmajor = -1,  error = 0;
2049 
2050 	switch (cmd) {
2051 	case MODULE_CMD_INIT:
2052 		error = config_cfdriver_attach(&vnd_cd);
2053 		if (error)
2054 			break;
2055 
2056 		error = config_cfattach_attach(vnd_cd.cd_name, &vnd_ca);
2057 	        if (error) {
2058 			config_cfdriver_detach(&vnd_cd);
2059 			aprint_error("%s: unable to register cfattach\n",
2060 			    vnd_cd.cd_name);
2061 			break;
2062 		}
2063 
2064 		error = devsw_attach("vnd", &vnd_bdevsw, &bmajor,
2065 		    &vnd_cdevsw, &cmajor);
2066 		if (error) {
2067 			config_cfattach_detach(vnd_cd.cd_name, &vnd_ca);
2068 			config_cfdriver_detach(&vnd_cd);
2069 			break;
2070 		}
2071 
2072 		break;
2073 
2074 	case MODULE_CMD_FINI:
2075 		error = config_cfattach_detach(vnd_cd.cd_name, &vnd_ca);
2076 		if (error)
2077 			break;
2078 		config_cfdriver_detach(&vnd_cd);
2079 		devsw_detach(&vnd_bdevsw, &vnd_cdevsw);
2080 		break;
2081 
2082 	case MODULE_CMD_STAT:
2083 		return ENOTTY;
2084 
2085 	default:
2086 		return ENOTTY;
2087 	}
2088 
2089 	return error;
2090 }
2091 
2092 #endif
2093