xref: /netbsd-src/sys/dev/vnd.c (revision a6f3f22f245acb8ee3bbf6871d7dce989204fa97)
1 /*	$NetBSD: vnd.c,v 1.248 2015/08/20 14:40:17 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 1.248 2015/08/20 14:40:17 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 = 0x00;
143 #endif
144 
145 #define vndunit(x)	DISKUNIT(x)
146 
147 struct vndxfer {
148 	struct buf vx_buf;
149 	struct vnd_softc *vx_vnd;
150 };
151 #define	VND_BUFTOXFER(bp)	((struct vndxfer *)(void *)bp)
152 
153 #define VND_GETXFER(vnd)	pool_get(&(vnd)->sc_vxpool, PR_WAITOK)
154 #define VND_PUTXFER(vnd, vx)	pool_put(&(vnd)->sc_vxpool, (vx))
155 
156 #define VNDLABELDEV(dev) \
157     (MAKEDISKDEV(major((dev)), vndunit((dev)), RAW_PART))
158 
159 #define	VND_MAXPENDING(vnd)	((vnd)->sc_maxactive * 4)
160 
161 
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 /* XXX this function needs a reliable check to detect
577  * sparse files. Otherwise, bmap/strategy may be used
578  * and fail on non-allocated blocks. VOP_READ/VOP_WRITE
579  * works on sparse files.
580  */
581 #if notyet
582 static bool
583 vnode_strategy_probe(struct vnd_softc *vnd)
584 {
585 	int error;
586 	daddr_t nbn;
587 
588 	if (!vnode_has_strategy(vnd))
589 		return false;
590 
591 	/* Convert the first logical block number to its
592 	 * physical block number.
593 	 */
594 	error = 0;
595 	vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
596 	error = VOP_BMAP(vnd->sc_vp, 0, NULL, &nbn, NULL);
597 	VOP_UNLOCK(vnd->sc_vp);
598 
599 	/* Test if that worked. */
600 	if (error == 0 && (long)nbn == -1)
601 		return false;
602 
603 	return true;
604 }
605 #endif
606 
607 static void
608 vndthread(void *arg)
609 {
610 	struct vnd_softc *vnd = arg;
611 	int s;
612 
613 	/* Determine whether we can *use* VOP_BMAP and VOP_STRATEGY to
614 	 * directly access the backing vnode.  If we can, use these two
615 	 * operations to avoid messing with the local buffer cache.
616 	 * Otherwise fall back to regular VOP_READ/VOP_WRITE operations
617 	 * which are guaranteed to work with any file system. */
618 	if ((vnd->sc_flags & VNF_USE_VN_RDWR) == 0 &&
619 	    ! vnode_has_strategy(vnd))
620 		vnd->sc_flags |= VNF_USE_VN_RDWR;
621 
622 #ifdef DEBUG
623 	if (vnddebug & VDB_INIT)
624 		printf("vndthread: vp %p, %s\n", vnd->sc_vp,
625 		    (vnd->sc_flags & VNF_USE_VN_RDWR) == 0 ?
626 		    "using bmap/strategy operations" :
627 		    "using read/write operations");
628 #endif
629 
630 	s = splbio();
631 	vnd->sc_flags |= VNF_KTHREAD;
632 	wakeup(&vnd->sc_kthread);
633 
634 	/*
635 	 * Dequeue requests and serve them depending on the available
636 	 * vnode operations.
637 	 */
638 	while ((vnd->sc_flags & VNF_VUNCONF) == 0) {
639 		struct vndxfer *vnx;
640 		struct buf *obp;
641 		struct buf *bp;
642 
643 		obp = bufq_get(vnd->sc_tab);
644 		if (obp == NULL) {
645 			tsleep(&vnd->sc_tab, PRIBIO, "vndbp", 0);
646 			continue;
647 		};
648 		if ((vnd->sc_flags & VNF_USE_VN_RDWR)) {
649 			KASSERT(vnd->sc_pending > 0 &&
650 			    vnd->sc_pending <= VND_MAXPENDING(vnd));
651 			if (vnd->sc_pending-- == VND_MAXPENDING(vnd))
652 				wakeup(&vnd->sc_pending);
653 		}
654 		splx(s);
655 #ifdef DEBUG
656 		if (vnddebug & VDB_FOLLOW)
657 			printf("vndthread(%p)\n", obp);
658 #endif
659 
660 		if (vnd->sc_vp->v_mount == NULL) {
661 			obp->b_error = ENXIO;
662 			goto done;
663 		}
664 #ifdef VND_COMPRESSION
665 		/* handle a compressed read */
666 		if ((obp->b_flags & B_READ) != 0 && (vnd->sc_flags & VNF_COMP)) {
667 			off_t bn;
668 
669 			/* Convert to a byte offset within the file. */
670 			bn = obp->b_rawblkno *
671 			    vnd->sc_dkdev.dk_label->d_secsize;
672 
673 			compstrategy(obp, bn);
674 			goto done;
675 		}
676 #endif /* VND_COMPRESSION */
677 
678 		/*
679 		 * Allocate a header for this transfer and link it to the
680 		 * buffer
681 		 */
682 		s = splbio();
683 		vnx = VND_GETXFER(vnd);
684 		splx(s);
685 		vnx->vx_vnd = vnd;
686 
687 		s = splbio();
688 		while (vnd->sc_active >= vnd->sc_maxactive) {
689 			tsleep(&vnd->sc_tab, PRIBIO, "vndac", 0);
690 		}
691 		vnd->sc_active++;
692 		splx(s);
693 
694 		/* Instrumentation. */
695 		disk_busy(&vnd->sc_dkdev);
696 
697 		bp = &vnx->vx_buf;
698 		buf_init(bp);
699 		bp->b_flags = (obp->b_flags & B_READ);
700 		bp->b_oflags = obp->b_oflags;
701 		bp->b_cflags = obp->b_cflags;
702 		bp->b_iodone = vndiodone;
703 		bp->b_private = obp;
704 		bp->b_vp = vnd->sc_vp;
705 		bp->b_objlock = bp->b_vp->v_interlock;
706 		bp->b_data = obp->b_data;
707 		bp->b_bcount = obp->b_bcount;
708 		BIO_COPYPRIO(bp, obp);
709 
710 		/* Handle the request using the appropriate operations. */
711 		if ((vnd->sc_flags & VNF_USE_VN_RDWR) == 0)
712 			handle_with_strategy(vnd, obp, bp);
713 		else
714 			handle_with_rdwr(vnd, obp, bp);
715 
716 		s = splbio();
717 		continue;
718 
719 done:
720 		biodone(obp);
721 		s = splbio();
722 	}
723 
724 	vnd->sc_flags &= (~VNF_KTHREAD | VNF_VUNCONF);
725 	wakeup(&vnd->sc_kthread);
726 	splx(s);
727 	kthread_exit(0);
728 }
729 
730 /*
731  * Checks if the given vnode supports the requested operation.
732  * The operation is specified the offset returned by VOFFSET.
733  *
734  * XXX The test below used to determine this is quite fragile
735  * because it relies on the file system to use genfs to specify
736  * unimplemented operations.  There might be another way to do
737  * it more cleanly.
738  */
739 static bool
740 vnode_has_op(const struct vnode *vp, int opoffset)
741 {
742 	int (*defaultp)(void *);
743 	int (*opp)(void *);
744 
745 	defaultp = vp->v_op[VOFFSET(vop_default)];
746 	opp = vp->v_op[opoffset];
747 
748 	return opp != defaultp && opp != genfs_eopnotsupp &&
749 	    opp != genfs_badop && opp != genfs_nullop;
750 }
751 
752 /*
753  * Handles the read/write request given in 'bp' using the vnode's VOP_READ
754  * and VOP_WRITE operations.
755  *
756  * 'obp' is a pointer to the original request fed to the vnd device.
757  */
758 static void
759 handle_with_rdwr(struct vnd_softc *vnd, const struct buf *obp, struct buf *bp)
760 {
761 	bool doread;
762 	off_t offset;
763 	size_t len, resid;
764 	struct vnode *vp;
765 
766 	doread = bp->b_flags & B_READ;
767 	offset = obp->b_rawblkno * vnd->sc_dkdev.dk_label->d_secsize;
768 	len = bp->b_bcount;
769 	vp = vnd->sc_vp;
770 
771 #if defined(DEBUG)
772 	if (vnddebug & VDB_IO)
773 		printf("vnd (rdwr): vp %p, %s, rawblkno 0x%" PRIx64
774 		    ", secsize %d, offset %" PRIu64
775 		    ", bcount %d\n",
776 		    vp, doread ? "read" : "write", obp->b_rawblkno,
777 		    vnd->sc_dkdev.dk_label->d_secsize, offset,
778 		    bp->b_bcount);
779 #endif
780 
781 	/* Issue the read or write operation. */
782 	bp->b_error =
783 	    vn_rdwr(doread ? UIO_READ : UIO_WRITE,
784 	    vp, bp->b_data, len, offset, UIO_SYSSPACE,
785 	    IO_ADV_ENCODE(POSIX_FADV_NOREUSE), vnd->sc_cred, &resid, NULL);
786 	bp->b_resid = resid;
787 
788 	mutex_enter(vp->v_interlock);
789 	(void) VOP_PUTPAGES(vp, 0, 0,
790 	    PGO_ALLPAGES | PGO_CLEANIT | PGO_FREE | PGO_SYNCIO);
791 
792 	/* We need to increase the number of outputs on the vnode if
793 	 * there was any write to it. */
794 	if (!doread) {
795 		mutex_enter(vp->v_interlock);
796 		vp->v_numoutput++;
797 		mutex_exit(vp->v_interlock);
798 	}
799 
800 	biodone(bp);
801 }
802 
803 /*
804  * Handes the read/write request given in 'bp' using the vnode's VOP_BMAP
805  * and VOP_STRATEGY operations.
806  *
807  * 'obp' is a pointer to the original request fed to the vnd device.
808  */
809 static void
810 handle_with_strategy(struct vnd_softc *vnd, const struct buf *obp,
811     struct buf *bp)
812 {
813 	int bsize, error, flags, skipped;
814 	size_t resid, sz;
815 	off_t bn, offset;
816 	struct vnode *vp;
817 	struct buf *nbp = NULL;
818 
819 	flags = obp->b_flags;
820 
821 
822 	/* convert to a byte offset within the file. */
823 	bn = obp->b_rawblkno * vnd->sc_dkdev.dk_label->d_secsize;
824 
825 	bsize = vnd->sc_vp->v_mount->mnt_stat.f_iosize;
826 	skipped = 0;
827 
828 	/*
829 	 * Break the request into bsize pieces and feed them
830 	 * sequentially using VOP_BMAP/VOP_STRATEGY.
831 	 * We do it this way to keep from flooding NFS servers if we
832 	 * are connected to an NFS file.  This places the burden on
833 	 * the client rather than the server.
834 	 */
835 	error = 0;
836 	bp->b_resid = bp->b_bcount;
837 	for (offset = 0, resid = bp->b_resid; /* true */;
838 	    resid -= sz, offset += sz) {
839 		daddr_t nbn;
840 		int off, nra;
841 
842 		nra = 0;
843 		vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
844 		error = VOP_BMAP(vnd->sc_vp, bn / bsize, &vp, &nbn, &nra);
845 		VOP_UNLOCK(vnd->sc_vp);
846 
847 		if (error == 0 && (long)nbn == -1)
848 			error = EIO;
849 
850 		/*
851 		 * If there was an error or a hole in the file...punt.
852 		 * Note that we may have to wait for any operations
853 		 * that we have already fired off before releasing
854 		 * the buffer.
855 		 *
856 		 * XXX we could deal with holes here but it would be
857 		 * a hassle (in the write case).
858 		 */
859 		if (error) {
860 			skipped += resid;
861 			break;
862 		}
863 
864 #ifdef DEBUG
865 		if (!dovndcluster)
866 			nra = 0;
867 #endif
868 
869 		off = bn % bsize;
870 		sz = MIN(((off_t)1 + nra) * bsize - off, resid);
871 #ifdef	DEBUG
872 		if (vnddebug & VDB_IO)
873 			printf("vndstrategy: vp %p/%p bn 0x%qx/0x%" PRIx64
874 			    " sz 0x%zx\n", vnd->sc_vp, vp, (long long)bn,
875 			    nbn, sz);
876 #endif
877 
878 		nbp = getiobuf(vp, true);
879 		nestiobuf_setup(bp, nbp, offset, sz);
880 		nbp->b_blkno = nbn + btodb(off);
881 
882 #if 0 /* XXX #ifdef DEBUG */
883 		if (vnddebug & VDB_IO)
884 			printf("vndstart(%ld): bp %p vp %p blkno "
885 			    "0x%" PRIx64 " flags %x addr %p cnt 0x%x\n",
886 			    (long) (vnd-vnd_softc), &nbp->vb_buf,
887 			    nbp->vb_buf.b_vp, nbp->vb_buf.b_blkno,
888 			    nbp->vb_buf.b_flags, nbp->vb_buf.b_data,
889 			    nbp->vb_buf.b_bcount);
890 #endif
891 		if (resid == sz) {
892 			break;
893 		}
894 		VOP_STRATEGY(vp, nbp);
895 		bn += sz;
896 	}
897 	if (!(flags & B_READ)) {
898 		struct vnode *w_vp;
899 		/*
900 		 * this is the last nested buf, account for
901 		 * the parent buf write too.
902 		 * This has to be done last, so that
903 		 * fsync won't wait for this write which
904 		 * has no chance to complete before all nested bufs
905 		 * have been queued. But it has to be done
906 		 * before the last VOP_STRATEGY()
907 		 * or the call to nestiobuf_done().
908 		 */
909 		w_vp = bp->b_vp;
910 		mutex_enter(w_vp->v_interlock);
911 		w_vp->v_numoutput++;
912 		mutex_exit(w_vp->v_interlock);
913 	}
914 	KASSERT(skipped != 0 || nbp != NULL);
915 	if (skipped)
916 		nestiobuf_done(bp, skipped, error);
917 	else
918 		VOP_STRATEGY(vp, nbp);
919 }
920 
921 static void
922 vndiodone(struct buf *bp)
923 {
924 	struct vndxfer *vnx = VND_BUFTOXFER(bp);
925 	struct vnd_softc *vnd = vnx->vx_vnd;
926 	struct buf *obp = bp->b_private;
927 	int s = splbio();
928 
929 	KASSERT(&vnx->vx_buf == bp);
930 	KASSERT(vnd->sc_active > 0);
931 #ifdef DEBUG
932 	if (vnddebug & VDB_IO) {
933 		printf("vndiodone1: bp %p iodone: error %d\n",
934 		    bp, bp->b_error);
935 	}
936 #endif
937 	disk_unbusy(&vnd->sc_dkdev, bp->b_bcount - bp->b_resid,
938 	    (bp->b_flags & B_READ));
939 	vnd->sc_active--;
940 	if (vnd->sc_active == 0) {
941 		wakeup(&vnd->sc_tab);
942 	}
943 	splx(s);
944 	obp->b_error = bp->b_error;
945 	obp->b_resid = bp->b_resid;
946 	buf_destroy(bp);
947 	VND_PUTXFER(vnd, vnx);
948 	biodone(obp);
949 }
950 
951 /* ARGSUSED */
952 static int
953 vndread(dev_t dev, struct uio *uio, int flags)
954 {
955 	int unit = vndunit(dev);
956 	struct vnd_softc *sc;
957 
958 #ifdef DEBUG
959 	if (vnddebug & VDB_FOLLOW)
960 		printf("vndread(0x%"PRIx64", %p)\n", dev, uio);
961 #endif
962 
963 	sc = device_lookup_private(&vnd_cd, unit);
964 	if (sc == NULL)
965 		return ENXIO;
966 
967 	if ((sc->sc_flags & VNF_INITED) == 0)
968 		return ENXIO;
969 
970 	return physio(vndstrategy, NULL, dev, B_READ, minphys, uio);
971 }
972 
973 /* ARGSUSED */
974 static int
975 vndwrite(dev_t dev, struct uio *uio, int flags)
976 {
977 	int unit = vndunit(dev);
978 	struct vnd_softc *sc;
979 
980 #ifdef DEBUG
981 	if (vnddebug & VDB_FOLLOW)
982 		printf("vndwrite(0x%"PRIx64", %p)\n", dev, uio);
983 #endif
984 
985 	sc = device_lookup_private(&vnd_cd, unit);
986 	if (sc == NULL)
987 		return ENXIO;
988 
989 	if ((sc->sc_flags & VNF_INITED) == 0)
990 		return ENXIO;
991 
992 	return physio(vndstrategy, NULL, dev, B_WRITE, minphys, uio);
993 }
994 
995 static int
996 vnd_cget(struct lwp *l, int unit, int *un, struct vattr *va)
997 {
998 	int error;
999 	struct vnd_softc *vnd;
1000 
1001 	if (*un == -1)
1002 		*un = unit;
1003 	if (*un < 0)
1004 		return EINVAL;
1005 
1006 	vnd = device_lookup_private(&vnd_cd, *un);
1007 	if (vnd == NULL)
1008 		return -1;
1009 
1010 	if ((vnd->sc_flags & VNF_INITED) == 0)
1011 		return -1;
1012 
1013 	vn_lock(vnd->sc_vp, LK_SHARED | LK_RETRY);
1014 	error = VOP_GETATTR(vnd->sc_vp, va, l->l_cred);
1015 	VOP_UNLOCK(vnd->sc_vp);
1016 	return error;
1017 }
1018 
1019 static int
1020 vnddoclear(struct vnd_softc *vnd, int pmask, int minor, bool force)
1021 {
1022 	int error;
1023 
1024 	if ((error = vndlock(vnd)) != 0)
1025 		return error;
1026 
1027 	/*
1028 	 * Don't unconfigure if any other partitions are open
1029 	 * or if both the character and block flavors of this
1030 	 * partition are open.
1031 	 */
1032 	if (DK_BUSY(vnd, pmask) && !force) {
1033 		vndunlock(vnd);
1034 		return EBUSY;
1035 	}
1036 
1037 	/* Delete all of our wedges */
1038 	dkwedge_delall(&vnd->sc_dkdev);
1039 
1040 	/*
1041 	 * XXX vndclear() might call vndclose() implicitly;
1042 	 * release lock to avoid recursion
1043 	 *
1044 	 * Set VNF_CLEARING to prevent vndopen() from
1045 	 * sneaking in after we vndunlock().
1046 	 */
1047 	vnd->sc_flags |= VNF_CLEARING;
1048 	vndunlock(vnd);
1049 	vndclear(vnd, minor);
1050 #ifdef DEBUG
1051 	if (vnddebug & VDB_INIT)
1052 		printf("vndioctl: CLRed\n");
1053 #endif
1054 
1055 	/* Destroy the xfer and buffer pools. */
1056 	pool_destroy(&vnd->sc_vxpool);
1057 
1058 	/* Detach the disk. */
1059 	disk_detach(&vnd->sc_dkdev);
1060 
1061 	return 0;
1062 }
1063 
1064 /* ARGSUSED */
1065 static int
1066 vndioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l)
1067 {
1068 	bool force;
1069 	int unit = vndunit(dev);
1070 	struct vnd_softc *vnd;
1071 	struct vnd_ioctl *vio;
1072 	struct vattr vattr;
1073 	struct pathbuf *pb;
1074 	struct nameidata nd;
1075 	int error, part, pmask;
1076 	uint64_t geomsize;
1077 	int fflags;
1078 #ifdef __HAVE_OLD_DISKLABEL
1079 	struct disklabel newlabel;
1080 #endif
1081 
1082 #ifdef DEBUG
1083 	if (vnddebug & VDB_FOLLOW)
1084 		printf("vndioctl(0x%"PRIx64", 0x%lx, %p, 0x%x, %p): unit %d\n",
1085 		    dev, cmd, data, flag, l->l_proc, unit);
1086 #endif
1087 	vnd = device_lookup_private(&vnd_cd, unit);
1088 	if (vnd == NULL &&
1089 #ifdef COMPAT_30
1090 	    cmd != VNDIOCGET30 &&
1091 #endif
1092 #ifdef COMPAT_50
1093 	    cmd != VNDIOCGET50 &&
1094 #endif
1095 	    cmd != VNDIOCGET)
1096 		return ENXIO;
1097 	vio = (struct vnd_ioctl *)data;
1098 
1099 	/* Must be open for writes for these commands... */
1100 	switch (cmd) {
1101 	case VNDIOCSET:
1102 	case VNDIOCCLR:
1103 #ifdef COMPAT_50
1104 	case VNDIOCSET50:
1105 	case VNDIOCCLR50:
1106 #endif
1107 	case DIOCSDINFO:
1108 	case DIOCWDINFO:
1109 #ifdef __HAVE_OLD_DISKLABEL
1110 	case ODIOCSDINFO:
1111 	case ODIOCWDINFO:
1112 #endif
1113 	case DIOCKLABEL:
1114 	case DIOCWLABEL:
1115 		if ((flag & FWRITE) == 0)
1116 			return EBADF;
1117 	}
1118 
1119 	/* Must be initialized for these... */
1120 	switch (cmd) {
1121 	case VNDIOCCLR:
1122 #ifdef VNDIOCCLR50
1123 	case VNDIOCCLR50:
1124 #endif
1125 	case DIOCGDINFO:
1126 	case DIOCSDINFO:
1127 	case DIOCWDINFO:
1128 	case DIOCGPART:
1129 	case DIOCKLABEL:
1130 	case DIOCWLABEL:
1131 	case DIOCGDEFLABEL:
1132 	case DIOCCACHESYNC:
1133 #ifdef __HAVE_OLD_DISKLABEL
1134 	case ODIOCGDINFO:
1135 	case ODIOCSDINFO:
1136 	case ODIOCWDINFO:
1137 	case ODIOCGDEFLABEL:
1138 #endif
1139 		if ((vnd->sc_flags & VNF_INITED) == 0)
1140 			return ENXIO;
1141 	}
1142 
1143 	error = disk_ioctl(&vnd->sc_dkdev, dev, cmd, data, flag, l);
1144 	if (error != EPASSTHROUGH)
1145 		return error;
1146 
1147 
1148 	switch (cmd) {
1149 #ifdef VNDIOCSET50
1150 	case VNDIOCSET50:
1151 #endif
1152 	case VNDIOCSET:
1153 		if (vnd->sc_flags & VNF_INITED)
1154 			return EBUSY;
1155 
1156 		if ((error = vndlock(vnd)) != 0)
1157 			return error;
1158 
1159 		fflags = FREAD;
1160 		if ((vio->vnd_flags & VNDIOF_READONLY) == 0)
1161 			fflags |= FWRITE;
1162 		error = pathbuf_copyin(vio->vnd_file, &pb);
1163 		if (error) {
1164 			goto unlock_and_exit;
1165 		}
1166 		NDINIT(&nd, LOOKUP, FOLLOW, pb);
1167 		if ((error = vn_open(&nd, fflags, 0)) != 0) {
1168 			pathbuf_destroy(pb);
1169 			goto unlock_and_exit;
1170 		}
1171 		KASSERT(l);
1172 		error = VOP_GETATTR(nd.ni_vp, &vattr, l->l_cred);
1173 		if (!error && nd.ni_vp->v_type != VREG)
1174 			error = EOPNOTSUPP;
1175 		if (!error && vattr.va_bytes < vattr.va_size)
1176 			/* File is definitely sparse, use vn_rdwr() */
1177 			vnd->sc_flags |= VNF_USE_VN_RDWR;
1178 		if (error) {
1179 			VOP_UNLOCK(nd.ni_vp);
1180 			goto close_and_exit;
1181 		}
1182 
1183 		/* If using a compressed file, initialize its info */
1184 		/* (or abort with an error if kernel has no compression) */
1185 		if (vio->vnd_flags & VNF_COMP) {
1186 #ifdef VND_COMPRESSION
1187 			struct vnd_comp_header *ch;
1188 			int i;
1189 			u_int32_t comp_size;
1190 			u_int32_t comp_maxsize;
1191 
1192 			/* allocate space for compresed file header */
1193 			ch = malloc(sizeof(struct vnd_comp_header),
1194 			M_TEMP, M_WAITOK);
1195 
1196 			/* read compressed file header */
1197 			error = vn_rdwr(UIO_READ, nd.ni_vp, (void *)ch,
1198 			  sizeof(struct vnd_comp_header), 0, UIO_SYSSPACE,
1199 			  IO_UNIT|IO_NODELOCKED, l->l_cred, NULL, NULL);
1200 			if (error) {
1201 				free(ch, M_TEMP);
1202 				VOP_UNLOCK(nd.ni_vp);
1203 				goto close_and_exit;
1204 			}
1205 
1206 			/* save some header info */
1207 			vnd->sc_comp_blksz = ntohl(ch->block_size);
1208 			/* note last offset is the file byte size */
1209 			vnd->sc_comp_numoffs = ntohl(ch->num_blocks)+1;
1210 			free(ch, M_TEMP);
1211 			if (vnd->sc_comp_blksz == 0 ||
1212 			    vnd->sc_comp_blksz % DEV_BSIZE !=0) {
1213 				VOP_UNLOCK(nd.ni_vp);
1214 				error = EINVAL;
1215 				goto close_and_exit;
1216 			}
1217 			if (sizeof(struct vnd_comp_header) +
1218 			  sizeof(u_int64_t) * vnd->sc_comp_numoffs >
1219 			  vattr.va_size) {
1220 				VOP_UNLOCK(nd.ni_vp);
1221 				error = EINVAL;
1222 				goto close_and_exit;
1223 			}
1224 
1225 			/* set decompressed file size */
1226 			vattr.va_size =
1227 			    ((u_quad_t)vnd->sc_comp_numoffs - 1) *
1228 			     (u_quad_t)vnd->sc_comp_blksz;
1229 
1230 			/* allocate space for all the compressed offsets */
1231 			vnd->sc_comp_offsets =
1232 			malloc(sizeof(u_int64_t) * vnd->sc_comp_numoffs,
1233 			M_DEVBUF, M_WAITOK);
1234 
1235 			/* read in the offsets */
1236 			error = vn_rdwr(UIO_READ, nd.ni_vp,
1237 			  (void *)vnd->sc_comp_offsets,
1238 			  sizeof(u_int64_t) * vnd->sc_comp_numoffs,
1239 			  sizeof(struct vnd_comp_header), UIO_SYSSPACE,
1240 			  IO_UNIT|IO_NODELOCKED, l->l_cred, NULL, NULL);
1241 			if (error) {
1242 				VOP_UNLOCK(nd.ni_vp);
1243 				goto close_and_exit;
1244 			}
1245 			/*
1246 			 * find largest block size (used for allocation limit).
1247 			 * Also convert offset to native byte order.
1248 			 */
1249 			comp_maxsize = 0;
1250 			for (i = 0; i < vnd->sc_comp_numoffs - 1; i++) {
1251 				vnd->sc_comp_offsets[i] =
1252 				  be64toh(vnd->sc_comp_offsets[i]);
1253 				comp_size = be64toh(vnd->sc_comp_offsets[i + 1])
1254 				  - vnd->sc_comp_offsets[i];
1255 				if (comp_size > comp_maxsize)
1256 					comp_maxsize = comp_size;
1257 			}
1258 			vnd->sc_comp_offsets[vnd->sc_comp_numoffs - 1] =
1259 			  be64toh(vnd->sc_comp_offsets[vnd->sc_comp_numoffs - 1]);
1260 
1261 			/* create compressed data buffer */
1262 			vnd->sc_comp_buff = malloc(comp_maxsize,
1263 			  M_DEVBUF, M_WAITOK);
1264 
1265 			/* create decompressed buffer */
1266 			vnd->sc_comp_decombuf = malloc(vnd->sc_comp_blksz,
1267 			  M_DEVBUF, M_WAITOK);
1268 			vnd->sc_comp_buffblk = -1;
1269 
1270 			/* Initialize decompress stream */
1271 			memset(&vnd->sc_comp_stream, 0, sizeof(z_stream));
1272 			vnd->sc_comp_stream.zalloc = vnd_alloc;
1273 			vnd->sc_comp_stream.zfree = vnd_free;
1274 			error = inflateInit2(&vnd->sc_comp_stream, MAX_WBITS);
1275 			if (error) {
1276 				if (vnd->sc_comp_stream.msg)
1277 					printf("vnd%d: compressed file, %s\n",
1278 					  unit, vnd->sc_comp_stream.msg);
1279 				VOP_UNLOCK(nd.ni_vp);
1280 				error = EINVAL;
1281 				goto close_and_exit;
1282 			}
1283 
1284 			vnd->sc_flags |= VNF_COMP | VNF_READONLY;
1285 #else /* !VND_COMPRESSION */
1286 			VOP_UNLOCK(nd.ni_vp);
1287 			error = EOPNOTSUPP;
1288 			goto close_and_exit;
1289 #endif /* VND_COMPRESSION */
1290 		}
1291 
1292 		VOP_UNLOCK(nd.ni_vp);
1293 		vnd->sc_vp = nd.ni_vp;
1294 		vnd->sc_size = btodb(vattr.va_size);	/* note truncation */
1295 
1296 		/*
1297 		 * Use pseudo-geometry specified.  If none was provided,
1298 		 * use "standard" Adaptec fictitious geometry.
1299 		 */
1300 		if (vio->vnd_flags & VNDIOF_HASGEOM) {
1301 
1302 			memcpy(&vnd->sc_geom, &vio->vnd_geom,
1303 			    sizeof(vio->vnd_geom));
1304 
1305 			/*
1306 			 * Sanity-check the sector size.
1307 			 * XXX Don't allow secsize < DEV_BSIZE.	 Should
1308 			 * XXX we?
1309 			 */
1310 			if (vnd->sc_geom.vng_secsize < DEV_BSIZE ||
1311 			    (vnd->sc_geom.vng_secsize % DEV_BSIZE) != 0 ||
1312 			    vnd->sc_geom.vng_ncylinders == 0 ||
1313 			    (vnd->sc_geom.vng_ntracks *
1314 			     vnd->sc_geom.vng_nsectors) == 0) {
1315 				error = EINVAL;
1316 				goto close_and_exit;
1317 			}
1318 
1319 			/*
1320 			 * Compute the size (in DEV_BSIZE blocks) specified
1321 			 * by the geometry.
1322 			 */
1323 			geomsize = (int64_t)vnd->sc_geom.vng_nsectors *
1324 			    vnd->sc_geom.vng_ntracks *
1325 			    vnd->sc_geom.vng_ncylinders *
1326 			    (vnd->sc_geom.vng_secsize / DEV_BSIZE);
1327 
1328 			/*
1329 			 * Sanity-check the size against the specified
1330 			 * geometry.
1331 			 */
1332 			if (vnd->sc_size < geomsize) {
1333 				error = EINVAL;
1334 				goto close_and_exit;
1335 			}
1336 		} else if (vnd->sc_size >= (32 * 64)) {
1337 			/*
1338 			 * Size must be at least 2048 DEV_BSIZE blocks
1339 			 * (1M) in order to use this geometry.
1340 			 */
1341 			vnd->sc_geom.vng_secsize = DEV_BSIZE;
1342 			vnd->sc_geom.vng_nsectors = 32;
1343 			vnd->sc_geom.vng_ntracks = 64;
1344 			vnd->sc_geom.vng_ncylinders = vnd->sc_size / (64 * 32);
1345 		} else {
1346 			vnd->sc_geom.vng_secsize = DEV_BSIZE;
1347 			vnd->sc_geom.vng_nsectors = 1;
1348 			vnd->sc_geom.vng_ntracks = 1;
1349 			vnd->sc_geom.vng_ncylinders = vnd->sc_size;
1350 		}
1351 
1352 		vnd_set_geometry(vnd);
1353 
1354 		if (vio->vnd_flags & VNDIOF_READONLY) {
1355 			vnd->sc_flags |= VNF_READONLY;
1356 		}
1357 
1358 		if ((error = vndsetcred(vnd, l->l_cred)) != 0)
1359 			goto close_and_exit;
1360 
1361 		vndthrottle(vnd, vnd->sc_vp);
1362 		vio->vnd_osize = dbtob(vnd->sc_size);
1363 #ifdef VNDIOCSET50
1364 		if (cmd != VNDIOCSET50)
1365 #endif
1366 			vio->vnd_size = dbtob(vnd->sc_size);
1367 		vnd->sc_flags |= VNF_INITED;
1368 
1369 		/* create the kernel thread, wait for it to be up */
1370 		error = kthread_create(PRI_NONE, 0, NULL, vndthread, vnd,
1371 		    &vnd->sc_kthread, "%s", device_xname(vnd->sc_dev));
1372 		if (error)
1373 			goto close_and_exit;
1374 		while ((vnd->sc_flags & VNF_KTHREAD) == 0) {
1375 			tsleep(&vnd->sc_kthread, PRIBIO, "vndthr", 0);
1376 		}
1377 #ifdef DEBUG
1378 		if (vnddebug & VDB_INIT)
1379 			printf("vndioctl: SET vp %p size 0x%lx %d/%d/%d/%d\n",
1380 			    vnd->sc_vp, (unsigned long) vnd->sc_size,
1381 			    vnd->sc_geom.vng_secsize,
1382 			    vnd->sc_geom.vng_nsectors,
1383 			    vnd->sc_geom.vng_ntracks,
1384 			    vnd->sc_geom.vng_ncylinders);
1385 #endif
1386 
1387 		/* Attach the disk. */
1388 		disk_attach(&vnd->sc_dkdev);
1389 
1390 		/* Initialize the xfer and buffer pools. */
1391 		pool_init(&vnd->sc_vxpool, sizeof(struct vndxfer), 0,
1392 		    0, 0, "vndxpl", NULL, IPL_BIO);
1393 
1394 		vndunlock(vnd);
1395 
1396 		pathbuf_destroy(pb);
1397 
1398 		/* Discover wedges on this disk */
1399 		dkwedge_discover(&vnd->sc_dkdev);
1400 
1401 		break;
1402 
1403 close_and_exit:
1404 		(void) vn_close(nd.ni_vp, fflags, l->l_cred);
1405 		pathbuf_destroy(pb);
1406 unlock_and_exit:
1407 #ifdef VND_COMPRESSION
1408 		/* free any allocated memory (for compressed file) */
1409 		if (vnd->sc_comp_offsets) {
1410 			free(vnd->sc_comp_offsets, M_DEVBUF);
1411 			vnd->sc_comp_offsets = NULL;
1412 		}
1413 		if (vnd->sc_comp_buff) {
1414 			free(vnd->sc_comp_buff, M_DEVBUF);
1415 			vnd->sc_comp_buff = NULL;
1416 		}
1417 		if (vnd->sc_comp_decombuf) {
1418 			free(vnd->sc_comp_decombuf, M_DEVBUF);
1419 			vnd->sc_comp_decombuf = NULL;
1420 		}
1421 #endif /* VND_COMPRESSION */
1422 		vndunlock(vnd);
1423 		return error;
1424 
1425 #ifdef VNDIOCCLR50
1426 	case VNDIOCCLR50:
1427 #endif
1428 	case VNDIOCCLR:
1429 		part = DISKPART(dev);
1430 		pmask = (1 << part);
1431 		force = (vio->vnd_flags & VNDIOF_FORCE) != 0;
1432 
1433 		if ((error = vnddoclear(vnd, pmask, minor(dev), force)) != 0)
1434 			return error;
1435 
1436 		break;
1437 
1438 #ifdef COMPAT_30
1439 	case VNDIOCGET30: {
1440 		struct vnd_user30 *vnu;
1441 		struct vattr va;
1442 		vnu = (struct vnd_user30 *)data;
1443 		KASSERT(l);
1444 		switch (error = vnd_cget(l, unit, &vnu->vnu_unit, &va)) {
1445 		case 0:
1446 			vnu->vnu_dev = va.va_fsid;
1447 			vnu->vnu_ino = va.va_fileid;
1448 			break;
1449 		case -1:
1450 			/* unused is not an error */
1451 			vnu->vnu_dev = 0;
1452 			vnu->vnu_ino = 0;
1453 			break;
1454 		default:
1455 			return error;
1456 		}
1457 		break;
1458 	}
1459 #endif
1460 
1461 #ifdef COMPAT_50
1462 	case VNDIOCGET50: {
1463 		struct vnd_user50 *vnu;
1464 		struct vattr va;
1465 		vnu = (struct vnd_user50 *)data;
1466 		KASSERT(l);
1467 		switch (error = vnd_cget(l, unit, &vnu->vnu_unit, &va)) {
1468 		case 0:
1469 			vnu->vnu_dev = va.va_fsid;
1470 			vnu->vnu_ino = va.va_fileid;
1471 			break;
1472 		case -1:
1473 			/* unused is not an error */
1474 			vnu->vnu_dev = 0;
1475 			vnu->vnu_ino = 0;
1476 			break;
1477 		default:
1478 			return error;
1479 		}
1480 		break;
1481 	}
1482 #endif
1483 
1484 	case VNDIOCGET: {
1485 		struct vnd_user *vnu;
1486 		struct vattr va;
1487 		vnu = (struct vnd_user *)data;
1488 		KASSERT(l);
1489 		switch (error = vnd_cget(l, unit, &vnu->vnu_unit, &va)) {
1490 		case 0:
1491 			vnu->vnu_dev = va.va_fsid;
1492 			vnu->vnu_ino = va.va_fileid;
1493 			break;
1494 		case -1:
1495 			/* unused is not an error */
1496 			vnu->vnu_dev = 0;
1497 			vnu->vnu_ino = 0;
1498 			break;
1499 		default:
1500 			return error;
1501 		}
1502 		break;
1503 	}
1504 
1505 	case DIOCWDINFO:
1506 	case DIOCSDINFO:
1507 #ifdef __HAVE_OLD_DISKLABEL
1508 	case ODIOCWDINFO:
1509 	case ODIOCSDINFO:
1510 #endif
1511 	{
1512 		struct disklabel *lp;
1513 
1514 		if ((error = vndlock(vnd)) != 0)
1515 			return error;
1516 
1517 		vnd->sc_flags |= VNF_LABELLING;
1518 
1519 #ifdef __HAVE_OLD_DISKLABEL
1520 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
1521 			memset(&newlabel, 0, sizeof newlabel);
1522 			memcpy(&newlabel, data, sizeof (struct olddisklabel));
1523 			lp = &newlabel;
1524 		} else
1525 #endif
1526 		lp = (struct disklabel *)data;
1527 
1528 		error = setdisklabel(vnd->sc_dkdev.dk_label,
1529 		    lp, 0, vnd->sc_dkdev.dk_cpulabel);
1530 		if (error == 0) {
1531 			if (cmd == DIOCWDINFO
1532 #ifdef __HAVE_OLD_DISKLABEL
1533 			    || cmd == ODIOCWDINFO
1534 #endif
1535 			   )
1536 				error = writedisklabel(VNDLABELDEV(dev),
1537 				    vndstrategy, vnd->sc_dkdev.dk_label,
1538 				    vnd->sc_dkdev.dk_cpulabel);
1539 		}
1540 
1541 		vnd->sc_flags &= ~VNF_LABELLING;
1542 
1543 		vndunlock(vnd);
1544 
1545 		if (error)
1546 			return error;
1547 		break;
1548 	}
1549 
1550 	case DIOCKLABEL:
1551 		if (*(int *)data != 0)
1552 			vnd->sc_flags |= VNF_KLABEL;
1553 		else
1554 			vnd->sc_flags &= ~VNF_KLABEL;
1555 		break;
1556 
1557 	case DIOCWLABEL:
1558 		if (*(int *)data != 0)
1559 			vnd->sc_flags |= VNF_WLABEL;
1560 		else
1561 			vnd->sc_flags &= ~VNF_WLABEL;
1562 		break;
1563 
1564 	case DIOCGDEFLABEL:
1565 		vndgetdefaultlabel(vnd, (struct disklabel *)data);
1566 		break;
1567 
1568 #ifdef __HAVE_OLD_DISKLABEL
1569 	case ODIOCGDEFLABEL:
1570 		vndgetdefaultlabel(vnd, &newlabel);
1571 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1572 			return ENOTTY;
1573 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1574 		break;
1575 #endif
1576 
1577 	case DIOCCACHESYNC:
1578 		vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1579 		error = VOP_FSYNC(vnd->sc_vp, vnd->sc_cred,
1580 		    FSYNC_WAIT | FSYNC_DATAONLY | FSYNC_CACHE, 0, 0);
1581 		VOP_UNLOCK(vnd->sc_vp);
1582 		return error;
1583 
1584 	default:
1585 		return ENOTTY;
1586 	}
1587 
1588 	return 0;
1589 }
1590 
1591 /*
1592  * Duplicate the current processes' credentials.  Since we are called only
1593  * as the result of a SET ioctl and only root can do that, any future access
1594  * to this "disk" is essentially as root.  Note that credentials may change
1595  * if some other uid can write directly to the mapped file (NFS).
1596  */
1597 static int
1598 vndsetcred(struct vnd_softc *vnd, kauth_cred_t cred)
1599 {
1600 	struct uio auio;
1601 	struct iovec aiov;
1602 	char *tmpbuf;
1603 	int error;
1604 
1605 	vnd->sc_cred = kauth_cred_dup(cred);
1606 	tmpbuf = malloc(DEV_BSIZE, M_TEMP, M_WAITOK);
1607 
1608 	/* XXX: Horrible kludge to establish credentials for NFS */
1609 	aiov.iov_base = tmpbuf;
1610 	aiov.iov_len = min(DEV_BSIZE, dbtob(vnd->sc_size));
1611 	auio.uio_iov = &aiov;
1612 	auio.uio_iovcnt = 1;
1613 	auio.uio_offset = 0;
1614 	auio.uio_rw = UIO_READ;
1615 	auio.uio_resid = aiov.iov_len;
1616 	UIO_SETUP_SYSSPACE(&auio);
1617 	vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1618 	error = VOP_READ(vnd->sc_vp, &auio, 0, vnd->sc_cred);
1619 	if (error == 0) {
1620 		/*
1621 		 * Because vnd does all IO directly through the vnode
1622 		 * we need to flush (at least) the buffer from the above
1623 		 * VOP_READ from the buffer cache to prevent cache
1624 		 * incoherencies.  Also, be careful to write dirty
1625 		 * buffers back to stable storage.
1626 		 */
1627 		error = vinvalbuf(vnd->sc_vp, V_SAVE, vnd->sc_cred,
1628 			    curlwp, 0, 0);
1629 	}
1630 	VOP_UNLOCK(vnd->sc_vp);
1631 
1632 	free(tmpbuf, M_TEMP);
1633 	return error;
1634 }
1635 
1636 /*
1637  * Set maxactive based on FS type
1638  */
1639 static void
1640 vndthrottle(struct vnd_softc *vnd, struct vnode *vp)
1641 {
1642 
1643 	if (vp->v_tag == VT_NFS)
1644 		vnd->sc_maxactive = 2;
1645 	else
1646 		vnd->sc_maxactive = 8;
1647 
1648 	if (vnd->sc_maxactive < 1)
1649 		vnd->sc_maxactive = 1;
1650 }
1651 
1652 #if 0
1653 static void
1654 vndshutdown(void)
1655 {
1656 	struct vnd_softc *vnd;
1657 
1658 	for (vnd = &vnd_softc[0]; vnd < &vnd_softc[numvnd]; vnd++)
1659 		if (vnd->sc_flags & VNF_INITED)
1660 			vndclear(vnd);
1661 }
1662 #endif
1663 
1664 static void
1665 vndclear(struct vnd_softc *vnd, int myminor)
1666 {
1667 	struct vnode *vp = vnd->sc_vp;
1668 	int fflags = FREAD;
1669 	int bmaj, cmaj, i, mn;
1670 	int s;
1671 
1672 #ifdef DEBUG
1673 	if (vnddebug & VDB_FOLLOW)
1674 		printf("vndclear(%p): vp %p\n", vnd, vp);
1675 #endif
1676 	/* locate the major number */
1677 	bmaj = bdevsw_lookup_major(&vnd_bdevsw);
1678 	cmaj = cdevsw_lookup_major(&vnd_cdevsw);
1679 
1680 	/* Nuke the vnodes for any open instances */
1681 	for (i = 0; i < MAXPARTITIONS; i++) {
1682 		mn = DISKMINOR(device_unit(vnd->sc_dev), i);
1683 		vdevgone(bmaj, mn, mn, VBLK);
1684 		if (mn != myminor) /* XXX avoid to kill own vnode */
1685 			vdevgone(cmaj, mn, mn, VCHR);
1686 	}
1687 
1688 	if ((vnd->sc_flags & VNF_READONLY) == 0)
1689 		fflags |= FWRITE;
1690 
1691 	s = splbio();
1692 	bufq_drain(vnd->sc_tab);
1693 	splx(s);
1694 
1695 	vnd->sc_flags |= VNF_VUNCONF;
1696 	wakeup(&vnd->sc_tab);
1697 	while (vnd->sc_flags & VNF_KTHREAD)
1698 		tsleep(&vnd->sc_kthread, PRIBIO, "vnthr", 0);
1699 
1700 #ifdef VND_COMPRESSION
1701 	/* free the compressed file buffers */
1702 	if (vnd->sc_flags & VNF_COMP) {
1703 		if (vnd->sc_comp_offsets) {
1704 			free(vnd->sc_comp_offsets, M_DEVBUF);
1705 			vnd->sc_comp_offsets = NULL;
1706 		}
1707 		if (vnd->sc_comp_buff) {
1708 			free(vnd->sc_comp_buff, M_DEVBUF);
1709 			vnd->sc_comp_buff = NULL;
1710 		}
1711 		if (vnd->sc_comp_decombuf) {
1712 			free(vnd->sc_comp_decombuf, M_DEVBUF);
1713 			vnd->sc_comp_decombuf = NULL;
1714 		}
1715 	}
1716 #endif /* VND_COMPRESSION */
1717 	vnd->sc_flags &=
1718 	    ~(VNF_INITED | VNF_READONLY | VNF_KLABEL | VNF_VLABEL
1719 	      | VNF_VUNCONF | VNF_COMP | VNF_CLEARING);
1720 	if (vp == NULL)
1721 		panic("vndclear: null vp");
1722 	(void) vn_close(vp, fflags, vnd->sc_cred);
1723 	kauth_cred_free(vnd->sc_cred);
1724 	vnd->sc_vp = NULL;
1725 	vnd->sc_cred = NULL;
1726 	vnd->sc_size = 0;
1727 }
1728 
1729 static int
1730 vndsize(dev_t dev)
1731 {
1732 	struct vnd_softc *sc;
1733 	struct disklabel *lp;
1734 	int part, unit, omask;
1735 	int size;
1736 
1737 	unit = vndunit(dev);
1738 	sc = device_lookup_private(&vnd_cd, unit);
1739 	if (sc == NULL)
1740 		return -1;
1741 
1742 	if ((sc->sc_flags & VNF_INITED) == 0)
1743 		return -1;
1744 
1745 	part = DISKPART(dev);
1746 	omask = sc->sc_dkdev.dk_openmask & (1 << part);
1747 	lp = sc->sc_dkdev.dk_label;
1748 
1749 	if (omask == 0 && vndopen(dev, 0, S_IFBLK, curlwp))	/* XXX */
1750 		return -1;
1751 
1752 	if (lp->d_partitions[part].p_fstype != FS_SWAP)
1753 		size = -1;
1754 	else
1755 		size = lp->d_partitions[part].p_size *
1756 		    (lp->d_secsize / DEV_BSIZE);
1757 
1758 	if (omask == 0 && vndclose(dev, 0, S_IFBLK, curlwp))	/* XXX */
1759 		return -1;
1760 
1761 	return size;
1762 }
1763 
1764 static int
1765 vnddump(dev_t dev, daddr_t blkno, void *va,
1766     size_t size)
1767 {
1768 
1769 	/* Not implemented. */
1770 	return ENXIO;
1771 }
1772 
1773 static void
1774 vndgetdefaultlabel(struct vnd_softc *sc, struct disklabel *lp)
1775 {
1776 	struct vndgeom *vng = &sc->sc_geom;
1777 	struct partition *pp;
1778 	unsigned spb;
1779 
1780 	memset(lp, 0, sizeof(*lp));
1781 
1782 	spb = vng->vng_secsize / DEV_BSIZE;
1783 	if (sc->sc_size / spb > UINT32_MAX)
1784 		lp->d_secperunit = UINT32_MAX;
1785 	else
1786 		lp->d_secperunit = sc->sc_size / spb;
1787 	lp->d_secsize = vng->vng_secsize;
1788 	lp->d_nsectors = vng->vng_nsectors;
1789 	lp->d_ntracks = vng->vng_ntracks;
1790 	lp->d_ncylinders = vng->vng_ncylinders;
1791 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1792 
1793 	strncpy(lp->d_typename, "vnd", sizeof(lp->d_typename));
1794 	lp->d_type = DKTYPE_VND;
1795 	strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
1796 	lp->d_rpm = 3600;
1797 	lp->d_interleave = 1;
1798 	lp->d_flags = 0;
1799 
1800 	pp = &lp->d_partitions[RAW_PART];
1801 	pp->p_offset = 0;
1802 	pp->p_size = lp->d_secperunit;
1803 	pp->p_fstype = FS_UNUSED;
1804 	lp->d_npartitions = RAW_PART + 1;
1805 
1806 	lp->d_magic = DISKMAGIC;
1807 	lp->d_magic2 = DISKMAGIC;
1808 	lp->d_checksum = dkcksum(lp);
1809 }
1810 
1811 /*
1812  * Read the disklabel from a vnd.  If one is not present, create a fake one.
1813  */
1814 static void
1815 vndgetdisklabel(dev_t dev, struct vnd_softc *sc)
1816 {
1817 	const char *errstring;
1818 	struct disklabel *lp = sc->sc_dkdev.dk_label;
1819 	struct cpu_disklabel *clp = sc->sc_dkdev.dk_cpulabel;
1820 	int i;
1821 
1822 	memset(clp, 0, sizeof(*clp));
1823 
1824 	vndgetdefaultlabel(sc, lp);
1825 
1826 	/*
1827 	 * Call the generic disklabel extraction routine.
1828 	 */
1829 	errstring = readdisklabel(VNDLABELDEV(dev), vndstrategy, lp, clp);
1830 	if (errstring) {
1831 		/*
1832 		 * Lack of disklabel is common, but we print the warning
1833 		 * anyway, since it might contain other useful information.
1834 		 */
1835 		aprint_normal_dev(sc->sc_dev, "%s\n", errstring);
1836 
1837 		/*
1838 		 * For historical reasons, if there's no disklabel
1839 		 * present, all partitions must be FS_BSDFFS and
1840 		 * occupy the entire disk.
1841 		 */
1842 		for (i = 0; i < MAXPARTITIONS; i++) {
1843 			/*
1844 			 * Don't wipe out port specific hack (such as
1845 			 * dos partition hack of i386 port).
1846 			 */
1847 			if (lp->d_partitions[i].p_size != 0)
1848 				continue;
1849 
1850 			lp->d_partitions[i].p_size = lp->d_secperunit;
1851 			lp->d_partitions[i].p_offset = 0;
1852 			lp->d_partitions[i].p_fstype = FS_BSDFFS;
1853 		}
1854 
1855 		strncpy(lp->d_packname, "default label",
1856 		    sizeof(lp->d_packname));
1857 
1858 		lp->d_npartitions = MAXPARTITIONS;
1859 		lp->d_checksum = dkcksum(lp);
1860 	}
1861 }
1862 
1863 /*
1864  * Wait interruptibly for an exclusive lock.
1865  *
1866  * XXX
1867  * Several drivers do this; it should be abstracted and made MP-safe.
1868  */
1869 static int
1870 vndlock(struct vnd_softc *sc)
1871 {
1872 	int error;
1873 
1874 	while ((sc->sc_flags & VNF_LOCKED) != 0) {
1875 		sc->sc_flags |= VNF_WANTED;
1876 		if ((error = tsleep(sc, PRIBIO | PCATCH, "vndlck", 0)) != 0)
1877 			return error;
1878 	}
1879 	sc->sc_flags |= VNF_LOCKED;
1880 	return 0;
1881 }
1882 
1883 /*
1884  * Unlock and wake up any waiters.
1885  */
1886 static void
1887 vndunlock(struct vnd_softc *sc)
1888 {
1889 
1890 	sc->sc_flags &= ~VNF_LOCKED;
1891 	if ((sc->sc_flags & VNF_WANTED) != 0) {
1892 		sc->sc_flags &= ~VNF_WANTED;
1893 		wakeup(sc);
1894 	}
1895 }
1896 
1897 #ifdef VND_COMPRESSION
1898 /* compressed file read */
1899 static void
1900 compstrategy(struct buf *bp, off_t bn)
1901 {
1902 	int error;
1903 	int unit = vndunit(bp->b_dev);
1904 	struct vnd_softc *vnd =
1905 	    device_lookup_private(&vnd_cd, unit);
1906 	u_int32_t comp_block;
1907 	struct uio auio;
1908 	char *addr;
1909 	int s;
1910 
1911 	/* set up constants for data move */
1912 	auio.uio_rw = UIO_READ;
1913 	UIO_SETUP_SYSSPACE(&auio);
1914 
1915 	/* read, and transfer the data */
1916 	addr = bp->b_data;
1917 	bp->b_resid = bp->b_bcount;
1918 	s = splbio();
1919 	while (bp->b_resid > 0) {
1920 		unsigned length;
1921 		size_t length_in_buffer;
1922 		u_int32_t offset_in_buffer;
1923 		struct iovec aiov;
1924 
1925 		/* calculate the compressed block number */
1926 		comp_block = bn / (off_t)vnd->sc_comp_blksz;
1927 
1928 		/* check for good block number */
1929 		if (comp_block >= vnd->sc_comp_numoffs) {
1930 			bp->b_error = EINVAL;
1931 			splx(s);
1932 			return;
1933 		}
1934 
1935 		/* read in the compressed block, if not in buffer */
1936 		if (comp_block != vnd->sc_comp_buffblk) {
1937 			length = vnd->sc_comp_offsets[comp_block + 1] -
1938 			    vnd->sc_comp_offsets[comp_block];
1939 			vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1940 			error = vn_rdwr(UIO_READ, vnd->sc_vp, vnd->sc_comp_buff,
1941 			    length, vnd->sc_comp_offsets[comp_block],
1942 			    UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, vnd->sc_cred,
1943 			    NULL, NULL);
1944 			if (error) {
1945 				bp->b_error = error;
1946 				VOP_UNLOCK(vnd->sc_vp);
1947 				splx(s);
1948 				return;
1949 			}
1950 			/* uncompress the buffer */
1951 			vnd->sc_comp_stream.next_in = vnd->sc_comp_buff;
1952 			vnd->sc_comp_stream.avail_in = length;
1953 			vnd->sc_comp_stream.next_out = vnd->sc_comp_decombuf;
1954 			vnd->sc_comp_stream.avail_out = vnd->sc_comp_blksz;
1955 			inflateReset(&vnd->sc_comp_stream);
1956 			error = inflate(&vnd->sc_comp_stream, Z_FINISH);
1957 			if (error != Z_STREAM_END) {
1958 				if (vnd->sc_comp_stream.msg)
1959 					aprint_normal_dev(vnd->sc_dev,
1960 					    "compressed file, %s\n",
1961 					    vnd->sc_comp_stream.msg);
1962 				bp->b_error = EBADMSG;
1963 				VOP_UNLOCK(vnd->sc_vp);
1964 				splx(s);
1965 				return;
1966 			}
1967 			vnd->sc_comp_buffblk = comp_block;
1968 			VOP_UNLOCK(vnd->sc_vp);
1969 		}
1970 
1971 		/* transfer the usable uncompressed data */
1972 		offset_in_buffer = bn % (off_t)vnd->sc_comp_blksz;
1973 		length_in_buffer = vnd->sc_comp_blksz - offset_in_buffer;
1974 		if (length_in_buffer > bp->b_resid)
1975 			length_in_buffer = bp->b_resid;
1976 		auio.uio_iov = &aiov;
1977 		auio.uio_iovcnt = 1;
1978 		aiov.iov_base = addr;
1979 		aiov.iov_len = length_in_buffer;
1980 		auio.uio_resid = aiov.iov_len;
1981 		auio.uio_offset = 0;
1982 		error = uiomove(vnd->sc_comp_decombuf + offset_in_buffer,
1983 		    length_in_buffer, &auio);
1984 		if (error) {
1985 			bp->b_error = error;
1986 			splx(s);
1987 			return;
1988 		}
1989 
1990 		bn += length_in_buffer;
1991 		addr += length_in_buffer;
1992 		bp->b_resid -= length_in_buffer;
1993 	}
1994 	splx(s);
1995 }
1996 
1997 /* compression memory allocation routines */
1998 static void *
1999 vnd_alloc(void *aux, u_int items, u_int siz)
2000 {
2001 	return malloc(items * siz, M_TEMP, M_NOWAIT);
2002 }
2003 
2004 static void
2005 vnd_free(void *aux, void *ptr)
2006 {
2007 	free(ptr, M_TEMP);
2008 }
2009 #endif /* VND_COMPRESSION */
2010 
2011 static void
2012 vnd_set_geometry(struct vnd_softc *vnd)
2013 {
2014 	struct disk_geom *dg = &vnd->sc_dkdev.dk_geom;
2015 
2016 	memset(dg, 0, sizeof(*dg));
2017 
2018 	dg->dg_secperunit = (int64_t)vnd->sc_geom.vng_nsectors *
2019 	    vnd->sc_geom.vng_ntracks * vnd->sc_geom.vng_ncylinders;
2020 	dg->dg_secsize = vnd->sc_geom.vng_secsize;
2021 	dg->dg_nsectors = vnd->sc_geom.vng_nsectors;
2022 	dg->dg_ntracks = vnd->sc_geom.vng_ntracks;
2023 	dg->dg_ncylinders = vnd->sc_geom.vng_ncylinders;
2024 
2025 #ifdef DEBUG
2026 	if (vnddebug & VDB_LABEL) {
2027 		printf("dg->dg_secperunit: %" PRId64 "\n", dg->dg_secperunit);
2028 		printf("dg->dg_ncylinders: %u\n", dg->dg_ncylinders);
2029 	}
2030 #endif
2031 	disk_set_info(vnd->sc_dev, &vnd->sc_dkdev, NULL);
2032 }
2033 
2034 #ifdef _MODULE
2035 
2036 #include <sys/module.h>
2037 
2038 #ifdef VND_COMPRESSION
2039 #define VND_DEPENDS "zlib"
2040 #else
2041 #define VND_DEPENDS NULL
2042 #endif
2043 
2044 MODULE(MODULE_CLASS_DRIVER, vnd, VND_DEPENDS);
2045 CFDRIVER_DECL(vnd, DV_DISK, NULL);
2046 
2047 static int
2048 vnd_modcmd(modcmd_t cmd, void *arg)
2049 {
2050 	int bmajor = -1, cmajor = -1,  error = 0;
2051 
2052 	switch (cmd) {
2053 	case MODULE_CMD_INIT:
2054 		error = config_cfdriver_attach(&vnd_cd);
2055 		if (error)
2056 			break;
2057 
2058 		error = config_cfattach_attach(vnd_cd.cd_name, &vnd_ca);
2059 	        if (error) {
2060 			config_cfdriver_detach(&vnd_cd);
2061 			aprint_error("%s: unable to register cfattach\n",
2062 			    vnd_cd.cd_name);
2063 			break;
2064 		}
2065 
2066 		error = devsw_attach("vnd", &vnd_bdevsw, &bmajor,
2067 		    &vnd_cdevsw, &cmajor);
2068 		if (error) {
2069 			config_cfattach_detach(vnd_cd.cd_name, &vnd_ca);
2070 			config_cfdriver_detach(&vnd_cd);
2071 			break;
2072 		}
2073 
2074 		break;
2075 
2076 	case MODULE_CMD_FINI:
2077 		error = config_cfattach_detach(vnd_cd.cd_name, &vnd_ca);
2078 		if (error)
2079 			break;
2080 		config_cfdriver_detach(&vnd_cd);
2081 		devsw_detach(&vnd_bdevsw, &vnd_cdevsw);
2082 		break;
2083 
2084 	case MODULE_CMD_STAT:
2085 		return ENOTTY;
2086 
2087 	default:
2088 		return ENOTTY;
2089 	}
2090 
2091 	return error;
2092 }
2093 
2094 #endif
2095