xref: /netbsd-src/sys/dev/vnd.c (revision 8b0f9554ff8762542c4defc4f70e1eb76fb508fa)
1 /*	$NetBSD: vnd.c,v 1.173 2007/12/12 03:54:27 smb Exp $	*/
2 
3 /*-
4  * Copyright (c) 1996, 1997, 1998 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  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the NetBSD
21  *	Foundation, Inc. and its contributors.
22  * 4. Neither the name of The NetBSD Foundation nor the names of its
23  *    contributors may be used to endorse or promote products derived
24  *    from this software without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
27  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
28  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
29  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
30  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36  * POSSIBILITY OF SUCH DAMAGE.
37  */
38 
39 /*
40  * Copyright (c) 1990, 1993
41  *	The Regents of the University of California.  All rights reserved.
42  *
43  * This code is derived from software contributed to Berkeley by
44  * the Systems Programming Group of the University of Utah Computer
45  * Science Department.
46  *
47  * Redistribution and use in source and binary forms, with or without
48  * modification, are permitted provided that the following conditions
49  * are met:
50  * 1. Redistributions of source code must retain the above copyright
51  *    notice, this list of conditions and the following disclaimer.
52  * 2. Redistributions in binary form must reproduce the above copyright
53  *    notice, this list of conditions and the following disclaimer in the
54  *    documentation and/or other materials provided with the distribution.
55  * 3. Neither the name of the University nor the names of its contributors
56  *    may be used to endorse or promote products derived from this software
57  *    without specific prior written permission.
58  *
59  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
60  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
61  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
62  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
63  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
64  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
65  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
66  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
67  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
68  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
69  * SUCH DAMAGE.
70  *
71  * from: Utah $Hdr: vn.c 1.13 94/04/02$
72  *
73  *	@(#)vn.c	8.9 (Berkeley) 5/14/95
74  */
75 
76 /*
77  * Copyright (c) 1988 University of Utah.
78  *
79  * This code is derived from software contributed to Berkeley by
80  * the Systems Programming Group of the University of Utah Computer
81  * Science Department.
82  *
83  * Redistribution and use in source and binary forms, with or without
84  * modification, are permitted provided that the following conditions
85  * are met:
86  * 1. Redistributions of source code must retain the above copyright
87  *    notice, this list of conditions and the following disclaimer.
88  * 2. Redistributions in binary form must reproduce the above copyright
89  *    notice, this list of conditions and the following disclaimer in the
90  *    documentation and/or other materials provided with the distribution.
91  * 3. All advertising materials mentioning features or use of this software
92  *    must display the following acknowledgement:
93  *	This product includes software developed by the University of
94  *	California, Berkeley and its contributors.
95  * 4. Neither the name of the University nor the names of its contributors
96  *    may be used to endorse or promote products derived from this software
97  *    without specific prior written permission.
98  *
99  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
100  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
101  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
102  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
103  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
104  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
105  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
106  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
107  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
108  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
109  * SUCH DAMAGE.
110  *
111  * from: Utah $Hdr: vn.c 1.13 94/04/02$
112  *
113  *	@(#)vn.c	8.9 (Berkeley) 5/14/95
114  */
115 
116 /*
117  * Vnode disk driver.
118  *
119  * Block/character interface to a vnode.  Allows one to treat a file
120  * as a disk (e.g. build a filesystem in it, mount it, etc.).
121  *
122  * NOTE 1: If the vnode supports the VOP_BMAP and VOP_STRATEGY operations,
123  * this uses them to avoid distorting the local buffer cache.  If those
124  * block-level operations are not available, this falls back to the regular
125  * read and write calls.  Using these may distort the cache in some cases
126  * but better have the driver working than preventing it to work on file
127  * systems where the block-level operations are not implemented for
128  * whatever reason.
129  *
130  * NOTE 2: There is a security issue involved with this driver.
131  * Once mounted all access to the contents of the "mapped" file via
132  * the special file is controlled by the permissions on the special
133  * file, the protection of the mapped file is ignored (effectively,
134  * by using root credentials in all transactions).
135  *
136  * NOTE 3: Doesn't interact with leases, should it?
137  */
138 
139 #include <sys/cdefs.h>
140 __KERNEL_RCSID(0, "$NetBSD: vnd.c,v 1.173 2007/12/12 03:54:27 smb Exp $");
141 
142 #if defined(_KERNEL_OPT)
143 #include "fs_nfs.h"
144 #include "opt_vnd.h"
145 #endif
146 
147 #include <sys/param.h>
148 #include <sys/systm.h>
149 #include <sys/namei.h>
150 #include <sys/proc.h>
151 #include <sys/kthread.h>
152 #include <sys/errno.h>
153 #include <sys/buf.h>
154 #include <sys/bufq.h>
155 #include <sys/malloc.h>
156 #include <sys/ioctl.h>
157 #include <sys/disklabel.h>
158 #include <sys/device.h>
159 #include <sys/disk.h>
160 #include <sys/stat.h>
161 #include <sys/mount.h>
162 #include <sys/vnode.h>
163 #include <sys/file.h>
164 #include <sys/uio.h>
165 #include <sys/conf.h>
166 #include <sys/kauth.h>
167 
168 #include <net/zlib.h>
169 
170 #include <miscfs/genfs/genfs.h>
171 #include <miscfs/specfs/specdev.h>
172 
173 #include <dev/vndvar.h>
174 
175 #if defined(VNDDEBUG) && !defined(DEBUG)
176 #define DEBUG
177 #endif
178 
179 #ifdef DEBUG
180 int dovndcluster = 1;
181 #define VDB_FOLLOW	0x01
182 #define VDB_INIT	0x02
183 #define VDB_IO		0x04
184 #define VDB_LABEL	0x08
185 int vnddebug = 0x00;
186 #endif
187 
188 #define vndunit(x)	DISKUNIT(x)
189 
190 struct vndxfer {
191 	struct buf vx_buf;
192 	struct vnd_softc *vx_vnd;
193 };
194 #define	VND_BUFTOXFER(bp)	((struct vndxfer *)(void *)bp)
195 
196 #define VND_GETXFER(vnd)	pool_get(&(vnd)->sc_vxpool, PR_WAITOK)
197 #define VND_PUTXFER(vnd, vx)	pool_put(&(vnd)->sc_vxpool, (vx))
198 
199 #define VNDLABELDEV(dev) \
200     (MAKEDISKDEV(major((dev)), vndunit((dev)), RAW_PART))
201 
202 /* called by main() at boot time (XXX: and the LKM driver) */
203 void	vndattach(int);
204 
205 static void	vndclear(struct vnd_softc *, int);
206 static int	vndsetcred(struct vnd_softc *, kauth_cred_t);
207 static void	vndthrottle(struct vnd_softc *, struct vnode *);
208 static void	vndiodone(struct buf *);
209 #if 0
210 static void	vndshutdown(void);
211 #endif
212 
213 static void	vndgetdefaultlabel(struct vnd_softc *, struct disklabel *);
214 static void	vndgetdisklabel(dev_t, struct vnd_softc *);
215 
216 static int	vndlock(struct vnd_softc *);
217 static void	vndunlock(struct vnd_softc *);
218 #ifdef VND_COMPRESSION
219 static void	compstrategy(struct buf *, off_t);
220 static void	*vnd_alloc(void *, u_int, u_int);
221 static void	vnd_free(void *, void *);
222 #endif /* VND_COMPRESSION */
223 
224 static void	vndthread(void *);
225 static bool	vnode_has_op(const struct vnode *, int);
226 static void	handle_with_rdwr(struct vnd_softc *, const struct buf *,
227 		    struct buf *);
228 static void	handle_with_strategy(struct vnd_softc *, const struct buf *,
229 		    struct buf *);
230 
231 static dev_type_open(vndopen);
232 static dev_type_close(vndclose);
233 static dev_type_read(vndread);
234 static dev_type_write(vndwrite);
235 static dev_type_ioctl(vndioctl);
236 static dev_type_strategy(vndstrategy);
237 static dev_type_dump(vnddump);
238 static dev_type_size(vndsize);
239 
240 const struct bdevsw vnd_bdevsw = {
241 	vndopen, vndclose, vndstrategy, vndioctl, vnddump, vndsize, D_DISK
242 };
243 
244 const struct cdevsw vnd_cdevsw = {
245 	vndopen, vndclose, vndread, vndwrite, vndioctl,
246 	nostop, notty, nopoll, nommap, nokqfilter, D_DISK
247 };
248 
249 static int	vnd_match(struct device *, struct cfdata *, void *);
250 static void	vnd_attach(struct device *, struct device *, void *);
251 static int	vnd_detach(struct device *, int);
252 
253 CFATTACH_DECL(vnd, sizeof(struct vnd_softc),
254     vnd_match, vnd_attach, vnd_detach, NULL);
255 extern struct cfdriver vnd_cd;
256 
257 static struct vnd_softc	*vnd_spawn(int);
258 int	vnd_destroy(struct device *);
259 
260 void
261 vndattach(int num)
262 {
263 	int error;
264 
265 	error = config_cfattach_attach(vnd_cd.cd_name, &vnd_ca);
266 	if (error)
267 		aprint_error("%s: unable to register cfattach\n",
268 		    vnd_cd.cd_name);
269 }
270 
271 static int
272 vnd_match(struct device *self, struct cfdata *cfdata,
273     void *aux)
274 {
275 	return 1;
276 }
277 
278 static void
279 vnd_attach(struct device *parent, struct device *self,
280     void *aux)
281 {
282 	struct vnd_softc *sc = (struct vnd_softc *)self;
283 
284 	sc->sc_comp_offsets = NULL;
285 	sc->sc_comp_buff = NULL;
286 	sc->sc_comp_decombuf = NULL;
287 	bufq_alloc(&sc->sc_tab, "disksort", BUFQ_SORT_RAWBLOCK);
288 	disk_init(&sc->sc_dkdev, self->dv_xname, NULL);
289 	if (!pmf_device_register(self, NULL, NULL))
290 		aprint_error_dev(self, "couldn't establish power handler\n");
291 }
292 
293 static int
294 vnd_detach(struct device *self, int flags)
295 {
296 	struct vnd_softc *sc = (struct vnd_softc *)self;
297 	if (sc->sc_flags & VNF_INITED)
298 		return EBUSY;
299 
300 	pmf_device_deregister(self);
301 	bufq_free(sc->sc_tab);
302 	disk_destroy(&sc->sc_dkdev);
303 
304 	return 0;
305 }
306 
307 static struct vnd_softc *
308 vnd_spawn(int unit)
309 {
310 	struct cfdata *cf;
311 
312 	cf = malloc(sizeof(*cf), M_DEVBUF, M_WAITOK);
313 	cf->cf_name = vnd_cd.cd_name;
314 	cf->cf_atname = vnd_cd.cd_name;
315 	cf->cf_unit = unit;
316 	cf->cf_fstate = FSTATE_STAR;
317 
318 	return (struct vnd_softc *)config_attach_pseudo(cf);
319 }
320 
321 int
322 vnd_destroy(struct device *dev)
323 {
324 	int error;
325 	struct cfdata *cf;
326 
327 	cf = device_cfdata(dev);
328 	error = config_detach(dev, DETACH_QUIET);
329 	if (error)
330 		return error;
331 	free(cf, M_DEVBUF);
332 	return 0;
333 }
334 
335 static int
336 vndopen(dev_t dev, int flags, int mode, struct lwp *l)
337 {
338 	int unit = vndunit(dev);
339 	struct vnd_softc *sc;
340 	int error = 0, part, pmask;
341 	struct disklabel *lp;
342 
343 #ifdef DEBUG
344 	if (vnddebug & VDB_FOLLOW)
345 		printf("vndopen(0x%x, 0x%x, 0x%x, %p)\n", dev, flags, mode, l);
346 #endif
347 	sc = device_lookup(&vnd_cd, unit);
348 	if (sc == NULL) {
349 		sc = vnd_spawn(unit);
350 		if (sc == NULL)
351 			return ENOMEM;
352 	}
353 
354 	if ((error = vndlock(sc)) != 0)
355 		return (error);
356 
357 	lp = sc->sc_dkdev.dk_label;
358 
359 	part = DISKPART(dev);
360 	pmask = (1 << part);
361 
362 	/*
363 	 * If we're initialized, check to see if there are any other
364 	 * open partitions.  If not, then it's safe to update the
365 	 * in-core disklabel.  Only read the disklabel if it is
366 	 * not already valid.
367 	 */
368 	if ((sc->sc_flags & (VNF_INITED|VNF_VLABEL)) == VNF_INITED &&
369 	    sc->sc_dkdev.dk_openmask == 0)
370 		vndgetdisklabel(dev, sc);
371 
372 	/* Check that the partitions exists. */
373 	if (part != RAW_PART) {
374 		if (((sc->sc_flags & VNF_INITED) == 0) ||
375 		    ((part >= lp->d_npartitions) ||
376 		     (lp->d_partitions[part].p_fstype == FS_UNUSED))) {
377 			error = ENXIO;
378 			goto done;
379 		}
380 	}
381 
382 	/* Prevent our unit from being unconfigured while open. */
383 	switch (mode) {
384 	case S_IFCHR:
385 		sc->sc_dkdev.dk_copenmask |= pmask;
386 		break;
387 
388 	case S_IFBLK:
389 		sc->sc_dkdev.dk_bopenmask |= pmask;
390 		break;
391 	}
392 	sc->sc_dkdev.dk_openmask =
393 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
394 
395  done:
396 	vndunlock(sc);
397 	return (error);
398 }
399 
400 static int
401 vndclose(dev_t dev, int flags, int mode, struct lwp *l)
402 {
403 	int unit = vndunit(dev);
404 	struct vnd_softc *sc;
405 	int error = 0, part;
406 
407 #ifdef DEBUG
408 	if (vnddebug & VDB_FOLLOW)
409 		printf("vndclose(0x%x, 0x%x, 0x%x, %p)\n", dev, flags, mode, l);
410 #endif
411 	sc = device_lookup(&vnd_cd, unit);
412 	if (sc == NULL)
413 		return ENXIO;
414 
415 	if ((error = vndlock(sc)) != 0)
416 		return (error);
417 
418 	part = DISKPART(dev);
419 
420 	/* ...that much closer to allowing unconfiguration... */
421 	switch (mode) {
422 	case S_IFCHR:
423 		sc->sc_dkdev.dk_copenmask &= ~(1 << part);
424 		break;
425 
426 	case S_IFBLK:
427 		sc->sc_dkdev.dk_bopenmask &= ~(1 << part);
428 		break;
429 	}
430 	sc->sc_dkdev.dk_openmask =
431 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
432 
433 	vndunlock(sc);
434 
435 	if ((sc->sc_flags & VNF_INITED) == 0) {
436 		if ((error = vnd_destroy((struct device *)sc)) != 0) {
437 			aprint_error("%s: unable to detach instance\n",
438 			    sc->sc_dev.dv_xname);
439 			return error;
440 		}
441 	}
442 
443 	return (0);
444 }
445 
446 /*
447  * Queue the request, and wakeup the kernel thread to handle it.
448  */
449 static void
450 vndstrategy(struct buf *bp)
451 {
452 	int unit = vndunit(bp->b_dev);
453 	struct vnd_softc *vnd =
454 	    (struct vnd_softc *)device_lookup(&vnd_cd, unit);
455 	struct disklabel *lp = vnd->sc_dkdev.dk_label;
456 	daddr_t blkno;
457 	int s = splbio();
458 
459 	if ((vnd->sc_flags & VNF_INITED) == 0) {
460 		bp->b_error = ENXIO;
461 		goto done;
462 	}
463 
464 	/*
465 	 * The transfer must be a whole number of blocks.
466 	 */
467 	if ((bp->b_bcount % lp->d_secsize) != 0) {
468 		bp->b_error = EINVAL;
469 		goto done;
470 	}
471 
472 	/*
473 	 * check if we're read-only.
474 	 */
475 	if ((vnd->sc_flags & VNF_READONLY) && !(bp->b_flags & B_READ)) {
476 		bp->b_error = EACCES;
477 		goto done;
478 	}
479 
480 	/* If it's a nil transfer, wake up the top half now. */
481 	if (bp->b_bcount == 0) {
482 		goto done;
483 	}
484 
485 	/*
486 	 * Do bounds checking and adjust transfer.  If there's an error,
487 	 * the bounds check will flag that for us.
488 	 */
489 	if (DISKPART(bp->b_dev) == RAW_PART) {
490 		if (bounds_check_with_mediasize(bp, DEV_BSIZE,
491 		    vnd->sc_size) <= 0)
492 			goto done;
493 	} else {
494 		if (bounds_check_with_label(&vnd->sc_dkdev,
495 		    bp, vnd->sc_flags & (VNF_WLABEL|VNF_LABELLING)) <= 0)
496 			goto done;
497 	}
498 
499 	/*
500 	 * Put the block number in terms of the logical blocksize
501 	 * of the "device".
502 	 */
503 
504 	blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
505 
506 	/*
507 	 * Translate the partition-relative block number to an absolute.
508 	 */
509 	if (DISKPART(bp->b_dev) != RAW_PART) {
510 		struct partition *pp;
511 
512 		pp = &vnd->sc_dkdev.dk_label->d_partitions[
513 		    DISKPART(bp->b_dev)];
514 		blkno += pp->p_offset;
515 	}
516 	bp->b_rawblkno = blkno;
517 
518 #ifdef DEBUG
519 	if (vnddebug & VDB_FOLLOW)
520 		printf("vndstrategy(%p): unit %d\n", bp, unit);
521 #endif
522 	BUFQ_PUT(vnd->sc_tab, bp);
523 	wakeup(&vnd->sc_tab);
524 	splx(s);
525 	return;
526 
527 done:
528 	bp->b_resid = bp->b_bcount;
529 	biodone(bp);
530 	splx(s);
531 }
532 
533 static void
534 vndthread(void *arg)
535 {
536 	struct vnd_softc *vnd = arg;
537 	bool usestrategy;
538 	int s;
539 
540 	/* Determine whether we can use VOP_BMAP and VOP_STRATEGY to
541 	 * directly access the backing vnode.  If we can, use these two
542 	 * operations to avoid messing with the local buffer cache.
543 	 * Otherwise fall back to regular VOP_READ/VOP_WRITE operations
544 	 * which are guaranteed to work with any file system. */
545 	usestrategy = vnode_has_op(vnd->sc_vp, VOFFSET(vop_bmap)) &&
546 	    vnode_has_op(vnd->sc_vp, VOFFSET(vop_strategy));
547 
548 #ifdef DEBUG
549 	if (vnddebug & VDB_INIT)
550 		printf("vndthread: vp %p, %s\n", vnd->sc_vp,
551 		    usestrategy ?
552 		    "using bmap/strategy operations" :
553 		    "using read/write operations");
554 #endif
555 
556 	s = splbio();
557 	vnd->sc_flags |= VNF_KTHREAD;
558 	wakeup(&vnd->sc_kthread);
559 
560 	/*
561 	 * Dequeue requests and serve them depending on the available
562 	 * vnode operations.
563 	 */
564 	while ((vnd->sc_flags & VNF_VUNCONF) == 0) {
565 		struct vndxfer *vnx;
566 		int flags;
567 		struct buf *obp;
568 		struct buf *bp;
569 
570 		obp = BUFQ_GET(vnd->sc_tab);
571 		if (obp == NULL) {
572 			tsleep(&vnd->sc_tab, PRIBIO, "vndbp", 0);
573 			continue;
574 		};
575 		splx(s);
576 		flags = obp->b_flags;
577 #ifdef DEBUG
578 		if (vnddebug & VDB_FOLLOW)
579 			printf("vndthread(%p\n", obp);
580 #endif
581 
582 		if (vnd->sc_vp->v_mount == NULL) {
583 			obp->b_error = ENXIO;
584 			goto done;
585 		}
586 #ifdef VND_COMPRESSION
587 		/* handle a compressed read */
588 		if ((flags & B_READ) != 0 && (vnd->sc_flags & VNF_COMP)) {
589 			off_t bn;
590 
591 			/* Convert to a byte offset within the file. */
592 			bn = obp->b_rawblkno *
593 			    vnd->sc_dkdev.dk_label->d_secsize;
594 
595 			compstrategy(obp, bn);
596 			goto done;
597 		}
598 #endif /* VND_COMPRESSION */
599 
600 		/*
601 		 * Allocate a header for this transfer and link it to the
602 		 * buffer
603 		 */
604 		s = splbio();
605 		vnx = VND_GETXFER(vnd);
606 		splx(s);
607 		vnx->vx_vnd = vnd;
608 
609 		s = splbio();
610 		while (vnd->sc_active >= vnd->sc_maxactive) {
611 			tsleep(&vnd->sc_tab, PRIBIO, "vndac", 0);
612 		}
613 		vnd->sc_active++;
614 		splx(s);
615 
616 		/* Instrumentation. */
617 		disk_busy(&vnd->sc_dkdev);
618 
619 		bp = &vnx->vx_buf;
620 		BUF_INIT(bp);
621 		bp->b_flags = (obp->b_flags & B_READ) | B_CALL;
622 		bp->b_iodone = vndiodone;
623 		bp->b_private = obp;
624 		bp->b_vp = vnd->sc_vp;
625 		bp->b_data = obp->b_data;
626 		bp->b_bcount = obp->b_bcount;
627 		BIO_COPYPRIO(bp, obp);
628 
629 		/* Handle the request using the appropriate operations. */
630 		if (usestrategy)
631 			handle_with_strategy(vnd, obp, bp);
632 		else
633 			handle_with_rdwr(vnd, obp, bp);
634 
635 		s = splbio();
636 		continue;
637 
638 done:
639 		biodone(obp);
640 		s = splbio();
641 	}
642 
643 	vnd->sc_flags &= (~VNF_KTHREAD | VNF_VUNCONF);
644 	wakeup(&vnd->sc_kthread);
645 	splx(s);
646 	kthread_exit(0);
647 }
648 
649 /*
650  * Checks if the given vnode supports the requested operation.
651  * The operation is specified the offset returned by VOFFSET.
652  *
653  * XXX The test below used to determine this is quite fragile
654  * because it relies on the file system to use genfs to specify
655  * unimplemented operations.  There might be another way to do
656  * it more cleanly.
657  */
658 static bool
659 vnode_has_op(const struct vnode *vp, int opoffset)
660 {
661 	int (*defaultp)(void *);
662 	int (*opp)(void *);
663 
664 	defaultp = vp->v_op[VOFFSET(vop_default)];
665 	opp = vp->v_op[opoffset];
666 
667 	return opp != defaultp && opp != genfs_eopnotsupp &&
668 	    opp != genfs_badop && opp != genfs_nullop;
669 }
670 
671 /*
672  * Handes the read/write request given in 'bp' using the vnode's VOP_READ
673  * and VOP_WRITE operations.
674  *
675  * 'obp' is a pointer to the original request fed to the vnd device.
676  */
677 static void
678 handle_with_rdwr(struct vnd_softc *vnd, const struct buf *obp, struct buf *bp)
679 {
680 	bool doread;
681 	off_t offset;
682 	size_t resid;
683 	struct vnode *vp;
684 
685 	doread = bp->b_flags & B_READ;
686 	offset = obp->b_rawblkno * vnd->sc_dkdev.dk_label->d_secsize;
687 	vp = vnd->sc_vp;
688 
689 #if defined(DEBUG)
690 	if (vnddebug & VDB_IO)
691 		printf("vnd (rdwr): vp %p, %s, rawblkno 0x%" PRIx64
692 		    ", secsize %d, offset %" PRIu64
693 		    ", bcount %d\n",
694 		    vp, doread ? "read" : "write", obp->b_rawblkno,
695 		    vnd->sc_dkdev.dk_label->d_secsize, offset,
696 		    bp->b_bcount);
697 #endif
698 
699 	/* Issue the read or write operation. */
700 	bp->b_error =
701 	    vn_rdwr(doread ? UIO_READ : UIO_WRITE,
702 	    vp, bp->b_data, bp->b_bcount, offset,
703 	    UIO_SYSSPACE, 0, vnd->sc_cred, &resid, NULL);
704 	bp->b_resid = resid;
705 
706 	/* We need to increase the number of outputs on the vnode if
707 	 * there was any write to it. */
708 	if (!doread)
709 		V_INCR_NUMOUTPUT(vp);
710 
711 	biodone(bp);
712 }
713 
714 /*
715  * Handes the read/write request given in 'bp' using the vnode's VOP_BMAP
716  * and VOP_STRATEGY operations.
717  *
718  * 'obp' is a pointer to the original request fed to the vnd device.
719  */
720 static void
721 handle_with_strategy(struct vnd_softc *vnd, const struct buf *obp,
722     struct buf *bp)
723 {
724 	int bsize, error, flags, skipped;
725 	size_t resid, sz;
726 	off_t bn, offset;
727 
728 	flags = obp->b_flags;
729 
730 	if (!(flags & B_READ)) {
731 		int s;
732 
733 		s = splbio();
734 		V_INCR_NUMOUTPUT(bp->b_vp);
735 		splx(s);
736 	}
737 
738 	/* convert to a byte offset within the file. */
739 	bn = obp->b_rawblkno * vnd->sc_dkdev.dk_label->d_secsize;
740 
741 	bsize = vnd->sc_vp->v_mount->mnt_stat.f_iosize;
742 	skipped = 0;
743 
744 	/*
745 	 * Break the request into bsize pieces and feed them
746 	 * sequentially using VOP_BMAP/VOP_STRATEGY.
747 	 * We do it this way to keep from flooding NFS servers if we
748 	 * are connected to an NFS file.  This places the burden on
749 	 * the client rather than the server.
750 	 */
751 	error = 0;
752 	bp->b_resid = bp->b_bcount;
753 	for (offset = 0, resid = bp->b_resid; resid;
754 	    resid -= sz, offset += sz) {
755 		struct buf *nbp;
756 		struct vnode *vp;
757 		daddr_t nbn;
758 		int off, nra;
759 
760 		nra = 0;
761 		vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY | LK_CANRECURSE);
762 		error = VOP_BMAP(vnd->sc_vp, bn / bsize, &vp, &nbn, &nra);
763 		VOP_UNLOCK(vnd->sc_vp, 0);
764 
765 		if (error == 0 && (long)nbn == -1)
766 			error = EIO;
767 
768 		/*
769 		 * If there was an error or a hole in the file...punt.
770 		 * Note that we may have to wait for any operations
771 		 * that we have already fired off before releasing
772 		 * the buffer.
773 		 *
774 		 * XXX we could deal with holes here but it would be
775 		 * a hassle (in the write case).
776 		 */
777 		if (error) {
778 			skipped += resid;
779 			break;
780 		}
781 
782 #ifdef DEBUG
783 		if (!dovndcluster)
784 			nra = 0;
785 #endif
786 
787 		off = bn % bsize;
788 		sz = MIN(((off_t)1 + nra) * bsize - off, resid);
789 #ifdef	DEBUG
790 		if (vnddebug & VDB_IO)
791 			printf("vndstrategy: vp %p/%p bn 0x%qx/0x%" PRIx64
792 			       " sz 0x%zx\n",
793 			    vnd->sc_vp, vp, (long long)bn, nbn, sz);
794 #endif
795 
796 		nbp = getiobuf();
797 		nestiobuf_setup(bp, nbp, offset, sz);
798 		nbp->b_blkno = nbn + btodb(off);
799 
800 #if 0 /* XXX #ifdef DEBUG */
801 		if (vnddebug & VDB_IO)
802 			printf("vndstart(%ld): bp %p vp %p blkno "
803 			    "0x%" PRIx64 " flags %x addr %p cnt 0x%x\n",
804 			    (long) (vnd-vnd_softc), &nbp->vb_buf,
805 			    nbp->vb_buf.b_vp, nbp->vb_buf.b_blkno,
806 			    nbp->vb_buf.b_flags, nbp->vb_buf.b_data,
807 			    nbp->vb_buf.b_bcount);
808 #endif
809 		VOP_STRATEGY(vp, nbp);
810 		bn += sz;
811 	}
812 	nestiobuf_done(bp, skipped, error);
813 }
814 
815 static void
816 vndiodone(struct buf *bp)
817 {
818 	struct vndxfer *vnx = VND_BUFTOXFER(bp);
819 	struct vnd_softc *vnd = vnx->vx_vnd;
820 	struct buf *obp = bp->b_private;
821 
822 	KASSERT(&vnx->vx_buf == bp);
823 	KASSERT(vnd->sc_active > 0);
824 #ifdef DEBUG
825 	if (vnddebug & VDB_IO) {
826 		printf("vndiodone1: bp %p iodone: error %d\n",
827 		    bp, bp->b_error);
828 	}
829 #endif
830 	disk_unbusy(&vnd->sc_dkdev, bp->b_bcount - bp->b_resid,
831 	    (bp->b_flags & B_READ));
832 	vnd->sc_active--;
833 	if (vnd->sc_active == 0) {
834 		wakeup(&vnd->sc_tab);
835 	}
836 	obp->b_error = bp->b_error;
837 	obp->b_resid = bp->b_resid;
838 	VND_PUTXFER(vnd, vnx);
839 	biodone(obp);
840 }
841 
842 /* ARGSUSED */
843 static int
844 vndread(dev_t dev, struct uio *uio, int flags)
845 {
846 	int unit = vndunit(dev);
847 	struct vnd_softc *sc;
848 
849 #ifdef DEBUG
850 	if (vnddebug & VDB_FOLLOW)
851 		printf("vndread(0x%x, %p)\n", dev, uio);
852 #endif
853 
854 	sc = device_lookup(&vnd_cd, unit);
855 	if (sc == NULL)
856 		return ENXIO;
857 
858 	if ((sc->sc_flags & VNF_INITED) == 0)
859 		return (ENXIO);
860 
861 	return (physio(vndstrategy, NULL, dev, B_READ, minphys, uio));
862 }
863 
864 /* ARGSUSED */
865 static int
866 vndwrite(dev_t dev, struct uio *uio, int flags)
867 {
868 	int unit = vndunit(dev);
869 	struct vnd_softc *sc;
870 
871 #ifdef DEBUG
872 	if (vnddebug & VDB_FOLLOW)
873 		printf("vndwrite(0x%x, %p)\n", dev, uio);
874 #endif
875 
876 	sc = device_lookup(&vnd_cd, unit);
877 	if (sc == NULL)
878 		return ENXIO;
879 
880 	if ((sc->sc_flags & VNF_INITED) == 0)
881 		return (ENXIO);
882 
883 	return (physio(vndstrategy, NULL, dev, B_WRITE, minphys, uio));
884 }
885 
886 static int
887 vnd_cget(struct lwp *l, int unit, int *un, struct vattr *va)
888 {
889 	struct vnd_softc *vnd;
890 
891 	if (*un == -1)
892 		*un = unit;
893 	if (*un < 0)
894 		return EINVAL;
895 
896 	vnd = device_lookup(&vnd_cd, *un);
897 	if (vnd == NULL)
898 		return (*un >= vnd_cd.cd_ndevs) ? ENXIO : -1;
899 
900 	if ((vnd->sc_flags & VNF_INITED) == 0)
901 		return -1;
902 
903 	return VOP_GETATTR(vnd->sc_vp, va, l->l_cred);
904 }
905 
906 /* ARGSUSED */
907 static int
908 vndioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l)
909 {
910 	int unit = vndunit(dev);
911 	struct vnd_softc *vnd;
912 	struct vnd_ioctl *vio;
913 	struct vattr vattr;
914 	struct nameidata nd;
915 	int error, part, pmask;
916 	size_t geomsize;
917 	int fflags;
918 #ifdef __HAVE_OLD_DISKLABEL
919 	struct disklabel newlabel;
920 #endif
921 
922 #ifdef DEBUG
923 	if (vnddebug & VDB_FOLLOW)
924 		printf("vndioctl(0x%x, 0x%lx, %p, 0x%x, %p): unit %d\n",
925 		    dev, cmd, data, flag, l->l_proc, unit);
926 #endif
927 	vnd = device_lookup(&vnd_cd, unit);
928 	if (vnd == NULL &&
929 #ifdef COMPAT_30
930 	    cmd != VNDIOOCGET &&
931 #endif
932 	    cmd != VNDIOCGET)
933 		return ENXIO;
934 	vio = (struct vnd_ioctl *)data;
935 
936 	/* Must be open for writes for these commands... */
937 	switch (cmd) {
938 	case VNDIOCSET:
939 	case VNDIOCCLR:
940 	case DIOCSDINFO:
941 	case DIOCWDINFO:
942 #ifdef __HAVE_OLD_DISKLABEL
943 	case ODIOCSDINFO:
944 	case ODIOCWDINFO:
945 #endif
946 	case DIOCKLABEL:
947 	case DIOCWLABEL:
948 		if ((flag & FWRITE) == 0)
949 			return (EBADF);
950 	}
951 
952 	/* Must be initialized for these... */
953 	switch (cmd) {
954 	case VNDIOCCLR:
955 	case DIOCGDINFO:
956 	case DIOCSDINFO:
957 	case DIOCWDINFO:
958 	case DIOCGPART:
959 	case DIOCKLABEL:
960 	case DIOCWLABEL:
961 	case DIOCGDEFLABEL:
962 #ifdef __HAVE_OLD_DISKLABEL
963 	case ODIOCGDINFO:
964 	case ODIOCSDINFO:
965 	case ODIOCWDINFO:
966 	case ODIOCGDEFLABEL:
967 #endif
968 		if ((vnd->sc_flags & VNF_INITED) == 0)
969 			return (ENXIO);
970 	}
971 
972 	switch (cmd) {
973 	case VNDIOCSET:
974 		if (vnd->sc_flags & VNF_INITED)
975 			return (EBUSY);
976 
977 		if ((error = vndlock(vnd)) != 0)
978 			return (error);
979 
980 		fflags = FREAD;
981 		if ((vio->vnd_flags & VNDIOF_READONLY) == 0)
982 			fflags |= FWRITE;
983 		NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, vio->vnd_file);
984 		if ((error = vn_open(&nd, fflags, 0)) != 0)
985 			goto unlock_and_exit;
986 		KASSERT(l);
987 		error = VOP_GETATTR(nd.ni_vp, &vattr, l->l_cred);
988 		if (!error && nd.ni_vp->v_type != VREG)
989 			error = EOPNOTSUPP;
990 		if (error) {
991 			VOP_UNLOCK(nd.ni_vp, 0);
992 			goto close_and_exit;
993 		}
994 
995 		/* If using a compressed file, initialize its info */
996 		/* (or abort with an error if kernel has no compression) */
997 		if (vio->vnd_flags & VNF_COMP) {
998 #ifdef VND_COMPRESSION
999 			struct vnd_comp_header *ch;
1000 			int i;
1001 			u_int32_t comp_size;
1002 			u_int32_t comp_maxsize;
1003 
1004 			/* allocate space for compresed file header */
1005 			ch = malloc(sizeof(struct vnd_comp_header),
1006 			M_TEMP, M_WAITOK);
1007 
1008 			/* read compressed file header */
1009 			error = vn_rdwr(UIO_READ, nd.ni_vp, (void *)ch,
1010 			  sizeof(struct vnd_comp_header), 0, UIO_SYSSPACE,
1011 			  IO_UNIT|IO_NODELOCKED, l->l_cred, NULL, NULL);
1012 			if(error) {
1013 				free(ch, M_TEMP);
1014 				VOP_UNLOCK(nd.ni_vp, 0);
1015 				goto close_and_exit;
1016 			}
1017 
1018 			/* save some header info */
1019 			vnd->sc_comp_blksz = ntohl(ch->block_size);
1020 			/* note last offset is the file byte size */
1021 			vnd->sc_comp_numoffs = ntohl(ch->num_blocks)+1;
1022 			free(ch, M_TEMP);
1023 			if (vnd->sc_comp_blksz == 0 ||
1024 			    vnd->sc_comp_blksz % DEV_BSIZE !=0) {
1025 				VOP_UNLOCK(nd.ni_vp, 0);
1026 				error = EINVAL;
1027 				goto close_and_exit;
1028 			}
1029 			if(sizeof(struct vnd_comp_header) +
1030 			  sizeof(u_int64_t) * vnd->sc_comp_numoffs >
1031 			  vattr.va_size) {
1032 				VOP_UNLOCK(nd.ni_vp, 0);
1033 				error = EINVAL;
1034 				goto close_and_exit;
1035 			}
1036 
1037 			/* set decompressed file size */
1038 			vattr.va_size =
1039 			    ((u_quad_t)vnd->sc_comp_numoffs - 1) *
1040 			     (u_quad_t)vnd->sc_comp_blksz;
1041 
1042 			/* allocate space for all the compressed offsets */
1043 			vnd->sc_comp_offsets =
1044 			malloc(sizeof(u_int64_t) * vnd->sc_comp_numoffs,
1045 			M_DEVBUF, M_WAITOK);
1046 
1047 			/* read in the offsets */
1048 			error = vn_rdwr(UIO_READ, nd.ni_vp,
1049 			  (void *)vnd->sc_comp_offsets,
1050 			  sizeof(u_int64_t) * vnd->sc_comp_numoffs,
1051 			  sizeof(struct vnd_comp_header), UIO_SYSSPACE,
1052 			  IO_UNIT|IO_NODELOCKED, l->l_cred, NULL, NULL);
1053 			if(error) {
1054 				VOP_UNLOCK(nd.ni_vp, 0);
1055 				goto close_and_exit;
1056 			}
1057 			/*
1058 			 * find largest block size (used for allocation limit).
1059 			 * Also convert offset to native byte order.
1060 			 */
1061 			comp_maxsize = 0;
1062 			for (i = 0; i < vnd->sc_comp_numoffs - 1; i++) {
1063 				vnd->sc_comp_offsets[i] =
1064 				  be64toh(vnd->sc_comp_offsets[i]);
1065 				comp_size = be64toh(vnd->sc_comp_offsets[i + 1])
1066 				  - vnd->sc_comp_offsets[i];
1067 				if (comp_size > comp_maxsize)
1068 					comp_maxsize = comp_size;
1069 			}
1070 			vnd->sc_comp_offsets[vnd->sc_comp_numoffs - 1] =
1071 			  be64toh(vnd->sc_comp_offsets[vnd->sc_comp_numoffs - 1]);
1072 
1073 			/* create compressed data buffer */
1074 			vnd->sc_comp_buff = malloc(comp_maxsize,
1075 			  M_DEVBUF, M_WAITOK);
1076 
1077 			/* create decompressed buffer */
1078 			vnd->sc_comp_decombuf = malloc(vnd->sc_comp_blksz,
1079 			  M_DEVBUF, M_WAITOK);
1080 			vnd->sc_comp_buffblk = -1;
1081 
1082 			/* Initialize decompress stream */
1083 			bzero(&vnd->sc_comp_stream, sizeof(z_stream));
1084 			vnd->sc_comp_stream.zalloc = vnd_alloc;
1085 			vnd->sc_comp_stream.zfree = vnd_free;
1086 			error = inflateInit2(&vnd->sc_comp_stream, MAX_WBITS);
1087 			if(error) {
1088 				if(vnd->sc_comp_stream.msg)
1089 					printf("vnd%d: compressed file, %s\n",
1090 					  unit, vnd->sc_comp_stream.msg);
1091 				VOP_UNLOCK(nd.ni_vp, 0);
1092 				error = EINVAL;
1093 				goto close_and_exit;
1094 			}
1095 
1096 			vnd->sc_flags |= VNF_COMP | VNF_READONLY;
1097 #else /* !VND_COMPRESSION */
1098 			VOP_UNLOCK(nd.ni_vp, 0);
1099 			error = EOPNOTSUPP;
1100 			goto close_and_exit;
1101 #endif /* VND_COMPRESSION */
1102 		}
1103 
1104 		VOP_UNLOCK(nd.ni_vp, 0);
1105 		vnd->sc_vp = nd.ni_vp;
1106 		vnd->sc_size = btodb(vattr.va_size);	/* note truncation */
1107 
1108 		/*
1109 		 * Use pseudo-geometry specified.  If none was provided,
1110 		 * use "standard" Adaptec fictitious geometry.
1111 		 */
1112 		if (vio->vnd_flags & VNDIOF_HASGEOM) {
1113 
1114 			memcpy(&vnd->sc_geom, &vio->vnd_geom,
1115 			    sizeof(vio->vnd_geom));
1116 
1117 			/*
1118 			 * Sanity-check the sector size.
1119 			 * XXX Don't allow secsize < DEV_BSIZE.	 Should
1120 			 * XXX we?
1121 			 */
1122 			if (vnd->sc_geom.vng_secsize < DEV_BSIZE ||
1123 			    (vnd->sc_geom.vng_secsize % DEV_BSIZE) != 0 ||
1124 			    vnd->sc_geom.vng_ncylinders == 0 ||
1125 			    (vnd->sc_geom.vng_ntracks *
1126 			     vnd->sc_geom.vng_nsectors) == 0) {
1127 				error = EINVAL;
1128 				goto close_and_exit;
1129 			}
1130 
1131 			/*
1132 			 * Compute the size (in DEV_BSIZE blocks) specified
1133 			 * by the geometry.
1134 			 */
1135 			geomsize = (vnd->sc_geom.vng_nsectors *
1136 			    vnd->sc_geom.vng_ntracks *
1137 			    vnd->sc_geom.vng_ncylinders) *
1138 			    (vnd->sc_geom.vng_secsize / DEV_BSIZE);
1139 
1140 			/*
1141 			 * Sanity-check the size against the specified
1142 			 * geometry.
1143 			 */
1144 			if (vnd->sc_size < geomsize) {
1145 				error = EINVAL;
1146 				goto close_and_exit;
1147 			}
1148 		} else if (vnd->sc_size >= (32 * 64)) {
1149 			/*
1150 			 * Size must be at least 2048 DEV_BSIZE blocks
1151 			 * (1M) in order to use this geometry.
1152 			 */
1153 			vnd->sc_geom.vng_secsize = DEV_BSIZE;
1154 			vnd->sc_geom.vng_nsectors = 32;
1155 			vnd->sc_geom.vng_ntracks = 64;
1156 			vnd->sc_geom.vng_ncylinders = vnd->sc_size / (64 * 32);
1157 		} else {
1158 			vnd->sc_geom.vng_secsize = DEV_BSIZE;
1159 			vnd->sc_geom.vng_nsectors = 1;
1160 			vnd->sc_geom.vng_ntracks = 1;
1161 			vnd->sc_geom.vng_ncylinders = vnd->sc_size;
1162 		}
1163 
1164 		if (vio->vnd_flags & VNDIOF_READONLY) {
1165 			vnd->sc_flags |= VNF_READONLY;
1166 		}
1167 
1168 		if ((error = vndsetcred(vnd, l->l_cred)) != 0)
1169 			goto close_and_exit;
1170 
1171 		vndthrottle(vnd, vnd->sc_vp);
1172 		vio->vnd_size = dbtob(vnd->sc_size);
1173 		vnd->sc_flags |= VNF_INITED;
1174 
1175 		/* create the kernel thread, wait for it to be up */
1176 		error = kthread_create(PRI_NONE, 0, NULL, vndthread, vnd,
1177 		    &vnd->sc_kthread, vnd->sc_dev.dv_xname);
1178 		if (error)
1179 			goto close_and_exit;
1180 		while ((vnd->sc_flags & VNF_KTHREAD) == 0) {
1181 			tsleep(&vnd->sc_kthread, PRIBIO, "vndthr", 0);
1182 		}
1183 #ifdef DEBUG
1184 		if (vnddebug & VDB_INIT)
1185 			printf("vndioctl: SET vp %p size 0x%lx %d/%d/%d/%d\n",
1186 			    vnd->sc_vp, (unsigned long) vnd->sc_size,
1187 			    vnd->sc_geom.vng_secsize,
1188 			    vnd->sc_geom.vng_nsectors,
1189 			    vnd->sc_geom.vng_ntracks,
1190 			    vnd->sc_geom.vng_ncylinders);
1191 #endif
1192 
1193 		/* Attach the disk. */
1194 		disk_attach(&vnd->sc_dkdev);
1195 
1196 		/* Initialize the xfer and buffer pools. */
1197 		pool_init(&vnd->sc_vxpool, sizeof(struct vndxfer), 0,
1198 		    0, 0, "vndxpl", NULL, IPL_BIO);
1199 
1200 		/* Try and read the disklabel. */
1201 		vndgetdisklabel(dev, vnd);
1202 
1203 		vndunlock(vnd);
1204 
1205 		break;
1206 
1207 close_and_exit:
1208 		(void) vn_close(nd.ni_vp, fflags, l->l_cred, l);
1209 unlock_and_exit:
1210 #ifdef VND_COMPRESSION
1211 		/* free any allocated memory (for compressed file) */
1212 		if(vnd->sc_comp_offsets) {
1213 			free(vnd->sc_comp_offsets, M_DEVBUF);
1214 			vnd->sc_comp_offsets = NULL;
1215 		}
1216 		if(vnd->sc_comp_buff) {
1217 			free(vnd->sc_comp_buff, M_DEVBUF);
1218 			vnd->sc_comp_buff = NULL;
1219 		}
1220 		if(vnd->sc_comp_decombuf) {
1221 			free(vnd->sc_comp_decombuf, M_DEVBUF);
1222 			vnd->sc_comp_decombuf = NULL;
1223 		}
1224 #endif /* VND_COMPRESSION */
1225 		vndunlock(vnd);
1226 		return (error);
1227 
1228 	case VNDIOCCLR:
1229 		if ((error = vndlock(vnd)) != 0)
1230 			return (error);
1231 
1232 		/*
1233 		 * Don't unconfigure if any other partitions are open
1234 		 * or if both the character and block flavors of this
1235 		 * partition are open.
1236 		 */
1237 		part = DISKPART(dev);
1238 		pmask = (1 << part);
1239 		if (((vnd->sc_dkdev.dk_openmask & ~pmask) ||
1240 		    ((vnd->sc_dkdev.dk_bopenmask & pmask) &&
1241 		    (vnd->sc_dkdev.dk_copenmask & pmask))) &&
1242 			!(vio->vnd_flags & VNDIOF_FORCE)) {
1243 			vndunlock(vnd);
1244 			return (EBUSY);
1245 		}
1246 
1247 		/*
1248 		 * XXX vndclear() might call vndclose() implicitely;
1249 		 * release lock to avoid recursion
1250 		 */
1251 		vndunlock(vnd);
1252 		vndclear(vnd, minor(dev));
1253 #ifdef DEBUG
1254 		if (vnddebug & VDB_INIT)
1255 			printf("vndioctl: CLRed\n");
1256 #endif
1257 
1258 		/* Destroy the xfer and buffer pools. */
1259 		pool_destroy(&vnd->sc_vxpool);
1260 
1261 		/* Detatch the disk. */
1262 		disk_detach(&vnd->sc_dkdev);
1263 		break;
1264 
1265 #ifdef COMPAT_30
1266 	case VNDIOOCGET: {
1267 		struct vnd_ouser *vnu;
1268 		struct vattr va;
1269 		vnu = (struct vnd_ouser *)data;
1270 		KASSERT(l);
1271 		switch (error = vnd_cget(l, unit, &vnu->vnu_unit, &va)) {
1272 		case 0:
1273 			vnu->vnu_dev = va.va_fsid;
1274 			vnu->vnu_ino = va.va_fileid;
1275 			break;
1276 		case -1:
1277 			/* unused is not an error */
1278 			vnu->vnu_dev = 0;
1279 			vnu->vnu_ino = 0;
1280 			break;
1281 		default:
1282 			return error;
1283 		}
1284 		break;
1285 	}
1286 #endif
1287 	case VNDIOCGET: {
1288 		struct vnd_user *vnu;
1289 		struct vattr va;
1290 		vnu = (struct vnd_user *)data;
1291 		KASSERT(l);
1292 		switch (error = vnd_cget(l, unit, &vnu->vnu_unit, &va)) {
1293 		case 0:
1294 			vnu->vnu_dev = va.va_fsid;
1295 			vnu->vnu_ino = va.va_fileid;
1296 			break;
1297 		case -1:
1298 			/* unused is not an error */
1299 			vnu->vnu_dev = 0;
1300 			vnu->vnu_ino = 0;
1301 			break;
1302 		default:
1303 			return error;
1304 		}
1305 		break;
1306 	}
1307 
1308 	case DIOCGDINFO:
1309 		*(struct disklabel *)data = *(vnd->sc_dkdev.dk_label);
1310 		break;
1311 
1312 #ifdef __HAVE_OLD_DISKLABEL
1313 	case ODIOCGDINFO:
1314 		newlabel = *(vnd->sc_dkdev.dk_label);
1315 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1316 			return ENOTTY;
1317 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1318 		break;
1319 #endif
1320 
1321 	case DIOCGPART:
1322 		((struct partinfo *)data)->disklab = vnd->sc_dkdev.dk_label;
1323 		((struct partinfo *)data)->part =
1324 		    &vnd->sc_dkdev.dk_label->d_partitions[DISKPART(dev)];
1325 		break;
1326 
1327 	case DIOCWDINFO:
1328 	case DIOCSDINFO:
1329 #ifdef __HAVE_OLD_DISKLABEL
1330 	case ODIOCWDINFO:
1331 	case ODIOCSDINFO:
1332 #endif
1333 	{
1334 		struct disklabel *lp;
1335 
1336 		if ((error = vndlock(vnd)) != 0)
1337 			return (error);
1338 
1339 		vnd->sc_flags |= VNF_LABELLING;
1340 
1341 #ifdef __HAVE_OLD_DISKLABEL
1342 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
1343 			memset(&newlabel, 0, sizeof newlabel);
1344 			memcpy(&newlabel, data, sizeof (struct olddisklabel));
1345 			lp = &newlabel;
1346 		} else
1347 #endif
1348 		lp = (struct disklabel *)data;
1349 
1350 		error = setdisklabel(vnd->sc_dkdev.dk_label,
1351 		    lp, 0, vnd->sc_dkdev.dk_cpulabel);
1352 		if (error == 0) {
1353 			if (cmd == DIOCWDINFO
1354 #ifdef __HAVE_OLD_DISKLABEL
1355 			    || cmd == ODIOCWDINFO
1356 #endif
1357 			   )
1358 				error = writedisklabel(VNDLABELDEV(dev),
1359 				    vndstrategy, vnd->sc_dkdev.dk_label,
1360 				    vnd->sc_dkdev.dk_cpulabel);
1361 		}
1362 
1363 		vnd->sc_flags &= ~VNF_LABELLING;
1364 
1365 		vndunlock(vnd);
1366 
1367 		if (error)
1368 			return (error);
1369 		break;
1370 	}
1371 
1372 	case DIOCKLABEL:
1373 		if (*(int *)data != 0)
1374 			vnd->sc_flags |= VNF_KLABEL;
1375 		else
1376 			vnd->sc_flags &= ~VNF_KLABEL;
1377 		break;
1378 
1379 	case DIOCWLABEL:
1380 		if (*(int *)data != 0)
1381 			vnd->sc_flags |= VNF_WLABEL;
1382 		else
1383 			vnd->sc_flags &= ~VNF_WLABEL;
1384 		break;
1385 
1386 	case DIOCGDEFLABEL:
1387 		vndgetdefaultlabel(vnd, (struct disklabel *)data);
1388 		break;
1389 
1390 #ifdef __HAVE_OLD_DISKLABEL
1391 	case ODIOCGDEFLABEL:
1392 		vndgetdefaultlabel(vnd, &newlabel);
1393 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1394 			return ENOTTY;
1395 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1396 		break;
1397 #endif
1398 
1399 	default:
1400 		return (ENOTTY);
1401 	}
1402 
1403 	return (0);
1404 }
1405 
1406 /*
1407  * Duplicate the current processes' credentials.  Since we are called only
1408  * as the result of a SET ioctl and only root can do that, any future access
1409  * to this "disk" is essentially as root.  Note that credentials may change
1410  * if some other uid can write directly to the mapped file (NFS).
1411  */
1412 static int
1413 vndsetcred(struct vnd_softc *vnd, kauth_cred_t cred)
1414 {
1415 	struct uio auio;
1416 	struct iovec aiov;
1417 	char *tmpbuf;
1418 	int error;
1419 
1420 	vnd->sc_cred = kauth_cred_dup(cred);
1421 	tmpbuf = malloc(DEV_BSIZE, M_TEMP, M_WAITOK);
1422 
1423 	/* XXX: Horrible kludge to establish credentials for NFS */
1424 	aiov.iov_base = tmpbuf;
1425 	aiov.iov_len = min(DEV_BSIZE, dbtob(vnd->sc_size));
1426 	auio.uio_iov = &aiov;
1427 	auio.uio_iovcnt = 1;
1428 	auio.uio_offset = 0;
1429 	auio.uio_rw = UIO_READ;
1430 	auio.uio_resid = aiov.iov_len;
1431 	UIO_SETUP_SYSSPACE(&auio);
1432 	vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1433 	error = VOP_READ(vnd->sc_vp, &auio, 0, vnd->sc_cred);
1434 	if (error == 0) {
1435 		/*
1436 		 * Because vnd does all IO directly through the vnode
1437 		 * we need to flush (at least) the buffer from the above
1438 		 * VOP_READ from the buffer cache to prevent cache
1439 		 * incoherencies.  Also, be careful to write dirty
1440 		 * buffers back to stable storage.
1441 		 */
1442 		error = vinvalbuf(vnd->sc_vp, V_SAVE, vnd->sc_cred,
1443 			    curlwp, 0, 0);
1444 	}
1445 	VOP_UNLOCK(vnd->sc_vp, 0);
1446 
1447 	free(tmpbuf, M_TEMP);
1448 	return (error);
1449 }
1450 
1451 /*
1452  * Set maxactive based on FS type
1453  */
1454 static void
1455 vndthrottle(struct vnd_softc *vnd, struct vnode *vp)
1456 {
1457 #ifdef NFS
1458 	extern int (**nfsv2_vnodeop_p)(void *);
1459 
1460 	if (vp->v_op == nfsv2_vnodeop_p)
1461 		vnd->sc_maxactive = 2;
1462 	else
1463 #endif
1464 		vnd->sc_maxactive = 8;
1465 
1466 	if (vnd->sc_maxactive < 1)
1467 		vnd->sc_maxactive = 1;
1468 }
1469 
1470 #if 0
1471 static void
1472 vndshutdown(void)
1473 {
1474 	struct vnd_softc *vnd;
1475 
1476 	for (vnd = &vnd_softc[0]; vnd < &vnd_softc[numvnd]; vnd++)
1477 		if (vnd->sc_flags & VNF_INITED)
1478 			vndclear(vnd);
1479 }
1480 #endif
1481 
1482 static void
1483 vndclear(struct vnd_softc *vnd, int myminor)
1484 {
1485 	struct vnode *vp = vnd->sc_vp;
1486 	struct lwp *l = curlwp;
1487 	int fflags = FREAD;
1488 	int bmaj, cmaj, i, mn;
1489 	int s;
1490 
1491 #ifdef DEBUG
1492 	if (vnddebug & VDB_FOLLOW)
1493 		printf("vndclear(%p): vp %p\n", vnd, vp);
1494 #endif
1495 	/* locate the major number */
1496 	bmaj = bdevsw_lookup_major(&vnd_bdevsw);
1497 	cmaj = cdevsw_lookup_major(&vnd_cdevsw);
1498 
1499 	/* Nuke the vnodes for any open instances */
1500 	for (i = 0; i < MAXPARTITIONS; i++) {
1501 		mn = DISKMINOR(device_unit(&vnd->sc_dev), i);
1502 		vdevgone(bmaj, mn, mn, VBLK);
1503 		if (mn != myminor) /* XXX avoid to kill own vnode */
1504 			vdevgone(cmaj, mn, mn, VCHR);
1505 	}
1506 
1507 	if ((vnd->sc_flags & VNF_READONLY) == 0)
1508 		fflags |= FWRITE;
1509 
1510 	s = splbio();
1511 	bufq_drain(vnd->sc_tab);
1512 	splx(s);
1513 
1514 	vnd->sc_flags |= VNF_VUNCONF;
1515 	wakeup(&vnd->sc_tab);
1516 	while (vnd->sc_flags & VNF_KTHREAD)
1517 		tsleep(&vnd->sc_kthread, PRIBIO, "vnthr", 0);
1518 
1519 #ifdef VND_COMPRESSION
1520 	/* free the compressed file buffers */
1521 	if(vnd->sc_flags & VNF_COMP) {
1522 		if(vnd->sc_comp_offsets) {
1523 			free(vnd->sc_comp_offsets, M_DEVBUF);
1524 			vnd->sc_comp_offsets = NULL;
1525 		}
1526 		if(vnd->sc_comp_buff) {
1527 			free(vnd->sc_comp_buff, M_DEVBUF);
1528 			vnd->sc_comp_buff = NULL;
1529 		}
1530 		if(vnd->sc_comp_decombuf) {
1531 			free(vnd->sc_comp_decombuf, M_DEVBUF);
1532 			vnd->sc_comp_decombuf = NULL;
1533 		}
1534 	}
1535 #endif /* VND_COMPRESSION */
1536 	vnd->sc_flags &=
1537 	    ~(VNF_INITED | VNF_READONLY | VNF_VLABEL
1538 	      | VNF_VUNCONF | VNF_COMP);
1539 	if (vp == (struct vnode *)0)
1540 		panic("vndclear: null vp");
1541 	(void) vn_close(vp, fflags, vnd->sc_cred, l);
1542 	kauth_cred_free(vnd->sc_cred);
1543 	vnd->sc_vp = (struct vnode *)0;
1544 	vnd->sc_cred = (kauth_cred_t)0;
1545 	vnd->sc_size = 0;
1546 }
1547 
1548 static int
1549 vndsize(dev_t dev)
1550 {
1551 	struct vnd_softc *sc;
1552 	struct disklabel *lp;
1553 	int part, unit, omask;
1554 	int size;
1555 
1556 	unit = vndunit(dev);
1557 	sc = (struct vnd_softc *)device_lookup(&vnd_cd, unit);
1558 	if (sc == NULL)
1559 		return -1;
1560 
1561 	if ((sc->sc_flags & VNF_INITED) == 0)
1562 		return (-1);
1563 
1564 	part = DISKPART(dev);
1565 	omask = sc->sc_dkdev.dk_openmask & (1 << part);
1566 	lp = sc->sc_dkdev.dk_label;
1567 
1568 	if (omask == 0 && vndopen(dev, 0, S_IFBLK, curlwp))	/* XXX */
1569 		return (-1);
1570 
1571 	if (lp->d_partitions[part].p_fstype != FS_SWAP)
1572 		size = -1;
1573 	else
1574 		size = lp->d_partitions[part].p_size *
1575 		    (lp->d_secsize / DEV_BSIZE);
1576 
1577 	if (omask == 0 && vndclose(dev, 0, S_IFBLK, curlwp))	/* XXX */
1578 		return (-1);
1579 
1580 	return (size);
1581 }
1582 
1583 static int
1584 vnddump(dev_t dev, daddr_t blkno, void *va,
1585     size_t size)
1586 {
1587 
1588 	/* Not implemented. */
1589 	return ENXIO;
1590 }
1591 
1592 static void
1593 vndgetdefaultlabel(struct vnd_softc *sc, struct disklabel *lp)
1594 {
1595 	struct vndgeom *vng = &sc->sc_geom;
1596 	struct partition *pp;
1597 
1598 	memset(lp, 0, sizeof(*lp));
1599 
1600 	lp->d_secperunit = sc->sc_size / (vng->vng_secsize / DEV_BSIZE);
1601 	lp->d_secsize = vng->vng_secsize;
1602 	lp->d_nsectors = vng->vng_nsectors;
1603 	lp->d_ntracks = vng->vng_ntracks;
1604 	lp->d_ncylinders = vng->vng_ncylinders;
1605 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1606 
1607 	strncpy(lp->d_typename, "vnd", sizeof(lp->d_typename));
1608 	lp->d_type = DTYPE_VND;
1609 	strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
1610 	lp->d_rpm = 3600;
1611 	lp->d_interleave = 1;
1612 	lp->d_flags = 0;
1613 
1614 	pp = &lp->d_partitions[RAW_PART];
1615 	pp->p_offset = 0;
1616 	pp->p_size = lp->d_secperunit;
1617 	pp->p_fstype = FS_UNUSED;
1618 	lp->d_npartitions = RAW_PART + 1;
1619 
1620 	lp->d_magic = DISKMAGIC;
1621 	lp->d_magic2 = DISKMAGIC;
1622 	lp->d_checksum = dkcksum(lp);
1623 }
1624 
1625 /*
1626  * Read the disklabel from a vnd.  If one is not present, create a fake one.
1627  */
1628 static void
1629 vndgetdisklabel(dev_t dev, struct vnd_softc *sc)
1630 {
1631 	const char *errstring;
1632 	struct disklabel *lp = sc->sc_dkdev.dk_label;
1633 	struct cpu_disklabel *clp = sc->sc_dkdev.dk_cpulabel;
1634 	int i;
1635 
1636 	memset(clp, 0, sizeof(*clp));
1637 
1638 	vndgetdefaultlabel(sc, lp);
1639 
1640 	/*
1641 	 * Call the generic disklabel extraction routine.
1642 	 */
1643 	errstring = readdisklabel(VNDLABELDEV(dev), vndstrategy, lp, clp);
1644 	if (errstring) {
1645 		/*
1646 		 * Lack of disklabel is common, but we print the warning
1647 		 * anyway, since it might contain other useful information.
1648 		 */
1649 		printf("%s: %s\n", sc->sc_dev.dv_xname, errstring);
1650 
1651 		/*
1652 		 * For historical reasons, if there's no disklabel
1653 		 * present, all partitions must be FS_BSDFFS and
1654 		 * occupy the entire disk.
1655 		 */
1656 		for (i = 0; i < MAXPARTITIONS; i++) {
1657 			/*
1658 			 * Don't wipe out port specific hack (such as
1659 			 * dos partition hack of i386 port).
1660 			 */
1661 			if (lp->d_partitions[i].p_size != 0)
1662 				continue;
1663 
1664 			lp->d_partitions[i].p_size = lp->d_secperunit;
1665 			lp->d_partitions[i].p_offset = 0;
1666 			lp->d_partitions[i].p_fstype = FS_BSDFFS;
1667 		}
1668 
1669 		strncpy(lp->d_packname, "default label",
1670 		    sizeof(lp->d_packname));
1671 
1672 		lp->d_npartitions = MAXPARTITIONS;
1673 		lp->d_checksum = dkcksum(lp);
1674 	}
1675 
1676 	/* In-core label now valid. */
1677 	sc->sc_flags |= VNF_VLABEL;
1678 }
1679 
1680 /*
1681  * Wait interruptibly for an exclusive lock.
1682  *
1683  * XXX
1684  * Several drivers do this; it should be abstracted and made MP-safe.
1685  */
1686 static int
1687 vndlock(struct vnd_softc *sc)
1688 {
1689 	int error;
1690 
1691 	while ((sc->sc_flags & VNF_LOCKED) != 0) {
1692 		sc->sc_flags |= VNF_WANTED;
1693 		if ((error = tsleep(sc, PRIBIO | PCATCH, "vndlck", 0)) != 0)
1694 			return (error);
1695 	}
1696 	sc->sc_flags |= VNF_LOCKED;
1697 	return (0);
1698 }
1699 
1700 /*
1701  * Unlock and wake up any waiters.
1702  */
1703 static void
1704 vndunlock(struct vnd_softc *sc)
1705 {
1706 
1707 	sc->sc_flags &= ~VNF_LOCKED;
1708 	if ((sc->sc_flags & VNF_WANTED) != 0) {
1709 		sc->sc_flags &= ~VNF_WANTED;
1710 		wakeup(sc);
1711 	}
1712 }
1713 
1714 #ifdef VND_COMPRESSION
1715 /* compressed file read */
1716 static void
1717 compstrategy(struct buf *bp, off_t bn)
1718 {
1719 	int error;
1720 	int unit = vndunit(bp->b_dev);
1721 	struct vnd_softc *vnd =
1722 	    (struct vnd_softc *)device_lookup(&vnd_cd, unit);
1723 	u_int32_t comp_block;
1724 	struct uio auio;
1725 	char *addr;
1726 	int s;
1727 
1728 	/* set up constants for data move */
1729 	auio.uio_rw = UIO_READ;
1730 	UIO_SETUP_SYSSPACE(&auio);
1731 
1732 	/* read, and transfer the data */
1733 	addr = bp->b_data;
1734 	bp->b_resid = bp->b_bcount;
1735 	s = splbio();
1736 	while (bp->b_resid > 0) {
1737 		unsigned length;
1738 		size_t length_in_buffer;
1739 		u_int32_t offset_in_buffer;
1740 		struct iovec aiov;
1741 
1742 		/* calculate the compressed block number */
1743 		comp_block = bn / (off_t)vnd->sc_comp_blksz;
1744 
1745 		/* check for good block number */
1746 		if (comp_block >= vnd->sc_comp_numoffs) {
1747 			bp->b_error = EINVAL;
1748 			splx(s);
1749 			return;
1750 		}
1751 
1752 		/* read in the compressed block, if not in buffer */
1753 		if (comp_block != vnd->sc_comp_buffblk) {
1754 			length = vnd->sc_comp_offsets[comp_block + 1] -
1755 			    vnd->sc_comp_offsets[comp_block];
1756 			vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1757 			error = vn_rdwr(UIO_READ, vnd->sc_vp, vnd->sc_comp_buff,
1758 			    length, vnd->sc_comp_offsets[comp_block],
1759 			    UIO_SYSSPACE, IO_UNIT, vnd->sc_cred, NULL, NULL);
1760 			if (error) {
1761 				bp->b_error = error;
1762 				VOP_UNLOCK(vnd->sc_vp, 0);
1763 				splx(s);
1764 				return;
1765 			}
1766 			/* uncompress the buffer */
1767 			vnd->sc_comp_stream.next_in = vnd->sc_comp_buff;
1768 			vnd->sc_comp_stream.avail_in = length;
1769 			vnd->sc_comp_stream.next_out = vnd->sc_comp_decombuf;
1770 			vnd->sc_comp_stream.avail_out = vnd->sc_comp_blksz;
1771 			inflateReset(&vnd->sc_comp_stream);
1772 			error = inflate(&vnd->sc_comp_stream, Z_FINISH);
1773 			if (error != Z_STREAM_END) {
1774 				if (vnd->sc_comp_stream.msg)
1775 					printf("%s: compressed file, %s\n",
1776 					    vnd->sc_dev.dv_xname,
1777 					    vnd->sc_comp_stream.msg);
1778 				bp->b_error = EBADMSG;
1779 				VOP_UNLOCK(vnd->sc_vp, 0);
1780 				splx(s);
1781 				return;
1782 			}
1783 			vnd->sc_comp_buffblk = comp_block;
1784 			VOP_UNLOCK(vnd->sc_vp, 0);
1785 		}
1786 
1787 		/* transfer the usable uncompressed data */
1788 		offset_in_buffer = bn % (off_t)vnd->sc_comp_blksz;
1789 		length_in_buffer = vnd->sc_comp_blksz - offset_in_buffer;
1790 		if (length_in_buffer > bp->b_resid)
1791 			length_in_buffer = bp->b_resid;
1792 		auio.uio_iov = &aiov;
1793 		auio.uio_iovcnt = 1;
1794 		aiov.iov_base = addr;
1795 		aiov.iov_len = length_in_buffer;
1796 		auio.uio_resid = aiov.iov_len;
1797 		auio.uio_offset = 0;
1798 		error = uiomove(vnd->sc_comp_decombuf + offset_in_buffer,
1799 		    length_in_buffer, &auio);
1800 		if (error) {
1801 			bp->b_error = error;
1802 			splx(s);
1803 			return;
1804 		}
1805 
1806 		bn += length_in_buffer;
1807 		addr += length_in_buffer;
1808 		bp->b_resid -= length_in_buffer;
1809 	}
1810 	splx(s);
1811 }
1812 
1813 /* compression memory allocation routines */
1814 static void *
1815 vnd_alloc(void *aux, u_int items, u_int siz)
1816 {
1817 	return malloc(items * siz, M_TEMP, M_NOWAIT);
1818 }
1819 
1820 static void
1821 vnd_free(void *aux, void *ptr)
1822 {
1823 	free(ptr, M_TEMP);
1824 }
1825 #endif /* VND_COMPRESSION */
1826