xref: /netbsd-src/sys/dev/vnd.c (revision d48f14661dda8638fee055ba15d35bdfb29b9fa8)
1 /*	$NetBSD: vnd.c,v 1.147 2006/05/14 21:42:26 elad 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: This uses the VOP_BMAP/VOP_STRATEGY interface to the vnode
123  * instead of a simple VOP_RDWR.  We do this to avoid distorting the
124  * local buffer cache.
125  *
126  * NOTE 2: There is a security issue involved with this driver.
127  * Once mounted all access to the contents of the "mapped" file via
128  * the special file is controlled by the permissions on the special
129  * file, the protection of the mapped file is ignored (effectively,
130  * by using root credentials in all transactions).
131  *
132  * NOTE 3: Doesn't interact with leases, should it?
133  */
134 
135 #include <sys/cdefs.h>
136 __KERNEL_RCSID(0, "$NetBSD: vnd.c,v 1.147 2006/05/14 21:42:26 elad Exp $");
137 
138 #if defined(_KERNEL_OPT)
139 #include "fs_nfs.h"
140 #include "opt_vnd.h"
141 #endif
142 
143 #include <sys/param.h>
144 #include <sys/systm.h>
145 #include <sys/namei.h>
146 #include <sys/proc.h>
147 #include <sys/kthread.h>
148 #include <sys/errno.h>
149 #include <sys/buf.h>
150 #include <sys/bufq.h>
151 #include <sys/malloc.h>
152 #include <sys/ioctl.h>
153 #include <sys/disklabel.h>
154 #include <sys/device.h>
155 #include <sys/disk.h>
156 #include <sys/stat.h>
157 #include <sys/mount.h>
158 #include <sys/vnode.h>
159 #include <sys/file.h>
160 #include <sys/uio.h>
161 #include <sys/conf.h>
162 #include <sys/kauth.h>
163 
164 #include <net/zlib.h>
165 
166 #include <miscfs/specfs/specdev.h>
167 
168 #include <dev/vndvar.h>
169 
170 #if defined(VNDDEBUG) && !defined(DEBUG)
171 #define DEBUG
172 #endif
173 
174 #ifdef DEBUG
175 int dovndcluster = 1;
176 #define VDB_FOLLOW	0x01
177 #define VDB_INIT	0x02
178 #define VDB_IO		0x04
179 #define VDB_LABEL	0x08
180 int vnddebug = 0x00;
181 #endif
182 
183 #define vndunit(x)	DISKUNIT(x)
184 
185 struct vndxfer {
186 	struct buf vx_buf;
187 	struct vnd_softc *vx_vnd;
188 };
189 #define	VND_BUFTOXFER(bp)	((struct vndxfer *)(void *)bp)
190 
191 #define VND_GETXFER(vnd)	pool_get(&(vnd)->sc_vxpool, PR_WAITOK)
192 #define VND_PUTXFER(vnd, vx)	pool_put(&(vnd)->sc_vxpool, (vx))
193 
194 #define VNDLABELDEV(dev) \
195     (MAKEDISKDEV(major((dev)), vndunit((dev)), RAW_PART))
196 
197 /* called by main() at boot time (XXX: and the LKM driver) */
198 void	vndattach(int);
199 
200 static void	vndclear(struct vnd_softc *, int);
201 static int	vndsetcred(struct vnd_softc *, kauth_cred_t);
202 static void	vndthrottle(struct vnd_softc *, struct vnode *);
203 static void	vndiodone(struct buf *);
204 #if 0
205 static void	vndshutdown(void);
206 #endif
207 
208 static void	vndgetdefaultlabel(struct vnd_softc *, struct disklabel *);
209 static void	vndgetdisklabel(dev_t, struct vnd_softc *);
210 
211 static int	vndlock(struct vnd_softc *);
212 static void	vndunlock(struct vnd_softc *);
213 #ifdef VND_COMPRESSION
214 static void	compstrategy(struct buf *, off_t);
215 static void	*vnd_alloc(void *, u_int, u_int);
216 static void	vnd_free(void *, void *);
217 #endif /* VND_COMPRESSION */
218 
219 static void vndthread(void *);
220 
221 static dev_type_open(vndopen);
222 static dev_type_close(vndclose);
223 static dev_type_read(vndread);
224 static dev_type_write(vndwrite);
225 static dev_type_ioctl(vndioctl);
226 static dev_type_strategy(vndstrategy);
227 static dev_type_dump(vnddump);
228 static dev_type_size(vndsize);
229 
230 const struct bdevsw vnd_bdevsw = {
231 	vndopen, vndclose, vndstrategy, vndioctl, vnddump, vndsize, D_DISK
232 };
233 
234 const struct cdevsw vnd_cdevsw = {
235 	vndopen, vndclose, vndread, vndwrite, vndioctl,
236 	nostop, notty, nopoll, nommap, nokqfilter, D_DISK
237 };
238 
239 static int	vnd_match(struct device *, struct cfdata *, void *);
240 static void	vnd_attach(struct device *, struct device *, void *);
241 static int	vnd_detach(struct device *, int);
242 
243 CFATTACH_DECL(vnd, sizeof(struct vnd_softc),
244     vnd_match, vnd_attach, vnd_detach, NULL);
245 extern struct cfdriver vnd_cd;
246 
247 static struct vnd_softc	*vnd_spawn(int);
248 int	vnd_destroy(struct device *);
249 
250 void
251 vndattach(int num)
252 {
253 	int error;
254 
255 	error = config_cfattach_attach(vnd_cd.cd_name, &vnd_ca);
256 	if (error)
257 		aprint_error("%s: unable to register cfattach\n",
258 		    vnd_cd.cd_name);
259 }
260 
261 static int
262 vnd_match(struct device *self, struct cfdata *cfdata, void *aux)
263 {
264 	return 1;
265 }
266 
267 static void
268 vnd_attach(struct device *parent, struct device *self, void *aux)
269 {
270 	struct vnd_softc *sc = (struct vnd_softc *)self;
271 
272 	sc->sc_comp_offsets = NULL;
273 	sc->sc_comp_buff = NULL;
274 	sc->sc_comp_decombuf = NULL;
275 	bufq_alloc(&sc->sc_tab, "disksort", BUFQ_SORT_RAWBLOCK);
276 	pseudo_disk_init(&sc->sc_dkdev);
277 }
278 
279 static int
280 vnd_detach(struct device *self, int flags)
281 {
282 	struct vnd_softc *sc = (struct vnd_softc *)self;
283 	if (sc->sc_flags & VNF_INITED)
284 		return EBUSY;
285 
286 	bufq_free(sc->sc_tab);
287 
288 	return 0;
289 }
290 
291 static struct vnd_softc *
292 vnd_spawn(int unit)
293 {
294 	struct cfdata *cf;
295 
296 	cf = malloc(sizeof(*cf), M_DEVBUF, M_WAITOK);
297 	cf->cf_name = vnd_cd.cd_name;
298 	cf->cf_atname = vnd_cd.cd_name;
299 	cf->cf_unit = unit;
300 	cf->cf_fstate = FSTATE_STAR;
301 
302 	return (struct vnd_softc *)config_attach_pseudo(cf);
303 }
304 
305 int
306 vnd_destroy(struct device *dev)
307 {
308 	int error;
309 	struct cfdata *cf;
310 
311 	cf = device_cfdata(dev);
312 	error = config_detach(dev, DETACH_QUIET);
313 	if (error)
314 		return error;
315 	free(cf, M_DEVBUF);
316 	return 0;
317 }
318 
319 static int
320 vndopen(dev_t dev, int flags, int mode, struct lwp *l)
321 {
322 	int unit = vndunit(dev);
323 	struct vnd_softc *sc;
324 	int error = 0, part, pmask;
325 	struct disklabel *lp;
326 
327 #ifdef DEBUG
328 	if (vnddebug & VDB_FOLLOW)
329 		printf("vndopen(0x%x, 0x%x, 0x%x, %p)\n", dev, flags, mode, l);
330 #endif
331 	sc = device_lookup(&vnd_cd, unit);
332 	if (sc == NULL) {
333 		sc = vnd_spawn(unit);
334 		if (sc == NULL)
335 			return ENOMEM;
336 	}
337 
338 	if ((error = vndlock(sc)) != 0)
339 		return (error);
340 
341 	lp = sc->sc_dkdev.dk_label;
342 
343 	part = DISKPART(dev);
344 	pmask = (1 << part);
345 
346 	/*
347 	 * If we're initialized, check to see if there are any other
348 	 * open partitions.  If not, then it's safe to update the
349 	 * in-core disklabel.  Only read the disklabel if it is
350 	 * not already valid.
351 	 */
352 	if ((sc->sc_flags & (VNF_INITED|VNF_VLABEL)) == VNF_INITED &&
353 	    sc->sc_dkdev.dk_openmask == 0)
354 		vndgetdisklabel(dev, sc);
355 
356 	/* Check that the partitions exists. */
357 	if (part != RAW_PART) {
358 		if (((sc->sc_flags & VNF_INITED) == 0) ||
359 		    ((part >= lp->d_npartitions) ||
360 		     (lp->d_partitions[part].p_fstype == FS_UNUSED))) {
361 			error = ENXIO;
362 			goto done;
363 		}
364 	}
365 
366 	/* Prevent our unit from being unconfigured while open. */
367 	switch (mode) {
368 	case S_IFCHR:
369 		sc->sc_dkdev.dk_copenmask |= pmask;
370 		break;
371 
372 	case S_IFBLK:
373 		sc->sc_dkdev.dk_bopenmask |= pmask;
374 		break;
375 	}
376 	sc->sc_dkdev.dk_openmask =
377 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
378 
379  done:
380 	vndunlock(sc);
381 	return (error);
382 }
383 
384 static int
385 vndclose(dev_t dev, int flags, int mode, struct lwp *l)
386 {
387 	int unit = vndunit(dev);
388 	struct vnd_softc *sc;
389 	int error = 0, part;
390 
391 #ifdef DEBUG
392 	if (vnddebug & VDB_FOLLOW)
393 		printf("vndclose(0x%x, 0x%x, 0x%x, %p)\n", dev, flags, mode, l);
394 #endif
395 	sc = device_lookup(&vnd_cd, unit);
396 	if (sc == NULL)
397 		return ENXIO;
398 
399 	if ((error = vndlock(sc)) != 0)
400 		return (error);
401 
402 	part = DISKPART(dev);
403 
404 	/* ...that much closer to allowing unconfiguration... */
405 	switch (mode) {
406 	case S_IFCHR:
407 		sc->sc_dkdev.dk_copenmask &= ~(1 << part);
408 		break;
409 
410 	case S_IFBLK:
411 		sc->sc_dkdev.dk_bopenmask &= ~(1 << part);
412 		break;
413 	}
414 	sc->sc_dkdev.dk_openmask =
415 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
416 
417 	vndunlock(sc);
418 
419 	if ((sc->sc_flags & VNF_INITED) == 0) {
420 		if ((error = vnd_destroy((struct device *)sc)) != 0) {
421 			aprint_error("%s: unable to detach instance\n",
422 			    sc->sc_dev.dv_xname);
423 			return error;
424 		}
425 	}
426 
427 	return (0);
428 }
429 
430 /*
431  * Queue the request, and wakeup the kernel thread to handle it.
432  */
433 static void
434 vndstrategy(struct buf *bp)
435 {
436 	int unit = vndunit(bp->b_dev);
437 	struct vnd_softc *vnd =
438 	    (struct vnd_softc *)device_lookup(&vnd_cd, unit);
439 	struct disklabel *lp = vnd->sc_dkdev.dk_label;
440 	daddr_t blkno;
441 	int s = splbio();
442 
443 	bp->b_resid = bp->b_bcount;
444 
445 	if ((vnd->sc_flags & VNF_INITED) == 0) {
446 		bp->b_error = ENXIO;
447 		bp->b_flags |= B_ERROR;
448 		goto done;
449 	}
450 
451 	/*
452 	 * The transfer must be a whole number of blocks.
453 	 */
454 	if ((bp->b_bcount % lp->d_secsize) != 0) {
455 		bp->b_error = EINVAL;
456 		bp->b_flags |= B_ERROR;
457 		goto done;
458 	}
459 
460 	/*
461 	 * check if we're read-only.
462 	 */
463 	if ((vnd->sc_flags & VNF_READONLY) && !(bp->b_flags & B_READ)) {
464 		bp->b_error = EACCES;
465 		bp->b_flags |= B_ERROR;
466 		goto done;
467 	}
468 
469 	/*
470 	 * Do bounds checking and adjust transfer.  If there's an error,
471 	 * the bounds check will flag that for us.
472 	 */
473 	if (DISKPART(bp->b_dev) == RAW_PART) {
474 		if (bounds_check_with_mediasize(bp, DEV_BSIZE,
475 		    vnd->sc_size) <= 0)
476 			goto done;
477 	} else {
478 		if (bounds_check_with_label(&vnd->sc_dkdev,
479 		    bp, vnd->sc_flags & (VNF_WLABEL|VNF_LABELLING)) <= 0)
480 			goto done;
481 	}
482 
483 	/* If it's a nil transfer, wake up the top half now. */
484 	if (bp->b_bcount == 0)
485 		goto done;
486 
487 	/*
488 	 * Put the block number in terms of the logical blocksize
489 	 * of the "device".
490 	 */
491 
492 	blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
493 
494 	/*
495 	 * Translate the partition-relative block number to an absolute.
496 	 */
497 	if (DISKPART(bp->b_dev) != RAW_PART) {
498 		struct partition *pp;
499 
500 		pp = &vnd->sc_dkdev.dk_label->d_partitions[
501 		    DISKPART(bp->b_dev)];
502 		blkno += pp->p_offset;
503 	}
504 	bp->b_rawblkno = blkno;
505 
506 #ifdef DEBUG
507 	if (vnddebug & VDB_FOLLOW)
508 		printf("vndstrategy(%p): unit %d\n", bp, unit);
509 #endif
510 	BUFQ_PUT(vnd->sc_tab, bp);
511 	wakeup(&vnd->sc_tab);
512 	splx(s);
513 	return;
514 done:
515 	biodone(bp);
516 	splx(s);
517 }
518 
519 static void
520 vndthread(void *arg)
521 {
522 	struct vnd_softc *vnd = arg;
523 	struct mount *mp;
524 	int s, bsize;
525 	int sz, error;
526 	struct disklabel *lp;
527 
528 	s = splbio();
529 	vnd->sc_flags |= VNF_KTHREAD;
530 	wakeup(&vnd->sc_kthread);
531 
532 	/*
533 	 * Dequeue requests, break them into bsize pieces and submit using
534 	 * VOP_BMAP/VOP_STRATEGY.
535 	 */
536 	while ((vnd->sc_flags & VNF_VUNCONF) == 0) {
537 		struct vndxfer *vnx;
538 		off_t offset;
539 		int resid;
540 		int skipped = 0;
541 		off_t bn;
542 		int flags;
543 		struct buf *obp;
544 		struct buf *bp;
545 
546 		obp = BUFQ_GET(vnd->sc_tab);
547 		if (obp == NULL) {
548 			tsleep(&vnd->sc_tab, PRIBIO, "vndbp", 0);
549 			continue;
550 		};
551 		splx(s);
552 		flags = obp->b_flags;
553 #ifdef DEBUG
554 		if (vnddebug & VDB_FOLLOW)
555 			printf("vndthread(%p\n", obp);
556 #endif
557 		lp = vnd->sc_dkdev.dk_label;
558 
559 		/* convert to a byte offset within the file. */
560 		bn = obp->b_rawblkno * lp->d_secsize;
561 
562 		if (vnd->sc_vp->v_mount == NULL) {
563 			obp->b_error = ENXIO;
564 			obp->b_flags |= B_ERROR;
565 			goto done;
566 		}
567 #ifdef VND_COMPRESSION
568 		/* handle a compressed read */
569 		if ((flags & B_READ) != 0 && (vnd->sc_flags & VNF_COMP)) {
570 			compstrategy(obp, bn);
571 			goto done;
572 		}
573 #endif /* VND_COMPRESSION */
574 
575 		bsize = vnd->sc_vp->v_mount->mnt_stat.f_iosize;
576 
577 		/*
578 		 * Allocate a header for this transfer and link it to the
579 		 * buffer
580 		 */
581 		s = splbio();
582 		vnx = VND_GETXFER(vnd);
583 		splx(s);
584 		vnx->vx_vnd = vnd;
585 
586 		bp = &vnx->vx_buf;
587 		BUF_INIT(bp);
588 		bp->b_flags = (obp->b_flags & B_READ) | B_CALL;
589 		bp->b_iodone = vndiodone;
590 		bp->b_private = obp;
591 		bp->b_vp = vnd->sc_vp;
592 		bp->b_data = obp->b_data;
593 		bp->b_bcount = bp->b_resid = obp->b_bcount;
594 		BIO_COPYPRIO(bp, obp);
595 
596 		s = splbio();
597 		while (vnd->sc_active >= vnd->sc_maxactive) {
598 			tsleep(&vnd->sc_tab, PRIBIO, "vndac", 0);
599 		}
600 		vnd->sc_active++;
601 		splx(s);
602 
603 		if ((flags & B_READ) == 0) {
604 			s = splbio();
605 			V_INCR_NUMOUTPUT(bp->b_vp);
606 			splx(s);
607 
608 			vn_start_write(vnd->sc_vp, &mp, V_WAIT);
609 		}
610 
611 		/* Instrumentation. */
612 		disk_busy(&vnd->sc_dkdev);
613 
614 		/*
615 		 * Feed requests sequentially.
616 		 * We do it this way to keep from flooding NFS servers if we
617 		 * are connected to an NFS file.  This places the burden on
618 		 * the client rather than the server.
619 		 */
620 		error = 0;
621 		for (offset = 0, resid = bp->b_resid; resid;
622 		    resid -= sz, offset += sz) {
623 			struct buf *nbp;
624 			struct vnode *vp;
625 			daddr_t nbn;
626 			int off, nra;
627 
628 			nra = 0;
629 			vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY | LK_CANRECURSE);
630 			error = VOP_BMAP(vnd->sc_vp, bn / bsize, &vp, &nbn, &nra);
631 			VOP_UNLOCK(vnd->sc_vp, 0);
632 
633 			if (error == 0 && (long)nbn == -1)
634 				error = EIO;
635 
636 			/*
637 			 * If there was an error or a hole in the file...punt.
638 			 * Note that we may have to wait for any operations
639 			 * that we have already fired off before releasing
640 			 * the buffer.
641 			 *
642 			 * XXX we could deal with holes here but it would be
643 			 * a hassle (in the write case).
644 			 */
645 			if (error) {
646 				skipped += resid;
647 				break;
648 			}
649 
650 #ifdef DEBUG
651 			if (!dovndcluster)
652 				nra = 0;
653 #endif
654 
655 			off = bn % bsize;
656 			sz = MIN(((off_t)1 + nra) * bsize - off, resid);
657 #ifdef	DEBUG
658 			if (vnddebug & VDB_IO)
659 				printf("vndstrategy: vp %p/%p bn 0x%qx/0x%" PRIx64
660 				       " sz 0x%x\n",
661 				    vnd->sc_vp, vp, (long long)bn, nbn, sz);
662 #endif
663 
664 			nbp = getiobuf();
665 			nestiobuf_setup(bp, nbp, offset, sz);
666 			nbp->b_blkno = nbn + btodb(off);
667 
668 #if 0 /* XXX #ifdef DEBUG */
669 			if (vnddebug & VDB_IO)
670 				printf("vndstart(%ld): bp %p vp %p blkno "
671 				    "0x%" PRIx64 " flags %x addr %p cnt 0x%x\n",
672 				    (long) (vnd-vnd_softc), &nbp->vb_buf,
673 				    nbp->vb_buf.b_vp, nbp->vb_buf.b_blkno,
674 				    nbp->vb_buf.b_flags, nbp->vb_buf.b_data,
675 				    nbp->vb_buf.b_bcount);
676 #endif
677 			VOP_STRATEGY(vp, nbp);
678 			bn += sz;
679 		}
680 		nestiobuf_done(bp, skipped, error);
681 
682 		if ((flags & B_READ) == 0)
683 			vn_finished_write(mp, 0);
684 
685 		s = splbio();
686 		continue;
687 done:
688 		biodone(obp);
689 		s = splbio();
690 	}
691 
692 	vnd->sc_flags &= (~VNF_KTHREAD | VNF_VUNCONF);
693 	wakeup(&vnd->sc_kthread);
694 	splx(s);
695 	kthread_exit(0);
696 }
697 
698 static void
699 vndiodone(struct buf *bp)
700 {
701 	struct vndxfer *vnx = VND_BUFTOXFER(bp);
702 	struct vnd_softc *vnd = vnx->vx_vnd;
703 	struct buf *obp = bp->b_private;
704 
705 	KASSERT(&vnx->vx_buf == bp);
706 	KASSERT(vnd->sc_active > 0);
707 #ifdef DEBUG
708 	if (vnddebug & VDB_IO) {
709 		printf("vndiodone1: bp %p iodone: error %d\n",
710 		    bp, (bp->b_flags & B_ERROR) != 0 ? bp->b_error : 0);
711 	}
712 #endif
713 	disk_unbusy(&vnd->sc_dkdev, bp->b_bcount - bp->b_resid,
714 	    (bp->b_flags & B_READ));
715 	vnd->sc_active--;
716 	if (vnd->sc_active == 0) {
717 		wakeup(&vnd->sc_tab);
718 	}
719 	obp->b_flags |= bp->b_flags & B_ERROR;
720 	obp->b_error = bp->b_error;
721 	obp->b_resid = bp->b_resid;
722 	VND_PUTXFER(vnd, vnx);
723 	biodone(obp);
724 }
725 
726 /* ARGSUSED */
727 static int
728 vndread(dev_t dev, struct uio *uio, int flags)
729 {
730 	int unit = vndunit(dev);
731 	struct vnd_softc *sc;
732 
733 #ifdef DEBUG
734 	if (vnddebug & VDB_FOLLOW)
735 		printf("vndread(0x%x, %p)\n", dev, uio);
736 #endif
737 
738 	sc = device_lookup(&vnd_cd, unit);
739 	if (sc == NULL)
740 		return ENXIO;
741 
742 	if ((sc->sc_flags & VNF_INITED) == 0)
743 		return (ENXIO);
744 
745 	return (physio(vndstrategy, NULL, dev, B_READ, minphys, uio));
746 }
747 
748 /* ARGSUSED */
749 static int
750 vndwrite(dev_t dev, struct uio *uio, int flags)
751 {
752 	int unit = vndunit(dev);
753 	struct vnd_softc *sc;
754 
755 #ifdef DEBUG
756 	if (vnddebug & VDB_FOLLOW)
757 		printf("vndwrite(0x%x, %p)\n", dev, uio);
758 #endif
759 
760 	sc = device_lookup(&vnd_cd, unit);
761 	if (sc == NULL)
762 		return ENXIO;
763 
764 	if ((sc->sc_flags & VNF_INITED) == 0)
765 		return (ENXIO);
766 
767 	return (physio(vndstrategy, NULL, dev, B_WRITE, minphys, uio));
768 }
769 
770 static int
771 vnd_cget(struct lwp *l, int unit, int *un, struct vattr *va)
772 {
773 	struct vnd_softc *vnd;
774 
775 	if (*un == -1)
776 		*un = unit;
777 	if (*un < 0)
778 		return EINVAL;
779 
780 	vnd = device_lookup(&vnd_cd, *un);
781 	if (vnd == NULL)
782 		return (*un >= vnd_cd.cd_ndevs) ? ENXIO : -1;
783 
784 	if ((vnd->sc_flags & VNF_INITED) == 0)
785 		return -1;
786 
787 	return VOP_GETATTR(vnd->sc_vp, va, l->l_proc->p_cred, l);
788 }
789 
790 /* ARGSUSED */
791 static int
792 vndioctl(dev_t dev, u_long cmd, caddr_t data, int flag, struct lwp *l)
793 {
794 	int unit = vndunit(dev);
795 	struct vnd_softc *vnd;
796 	struct vnd_ioctl *vio;
797 	struct vattr vattr;
798 	struct nameidata nd;
799 	int error, part, pmask;
800 	size_t geomsize;
801 	struct proc *p =  (l != NULL) ? l->l_proc : NULL;
802 	int fflags;
803 #ifdef __HAVE_OLD_DISKLABEL
804 	struct disklabel newlabel;
805 #endif
806 
807 #ifdef DEBUG
808 	if (vnddebug & VDB_FOLLOW)
809 		printf("vndioctl(0x%x, 0x%lx, %p, 0x%x, %p): unit %d\n",
810 		    dev, cmd, data, flag, p, unit);
811 #endif
812 	vnd = device_lookup(&vnd_cd, unit);
813 	if (vnd == NULL &&
814 #ifdef COMPAT_30
815 	    cmd != VNDIOOCGET &&
816 #endif
817 	    cmd != VNDIOCGET)
818 		return ENXIO;
819 	vio = (struct vnd_ioctl *)data;
820 
821 	/* Must be open for writes for these commands... */
822 	switch (cmd) {
823 	case VNDIOCSET:
824 	case VNDIOCCLR:
825 	case DIOCSDINFO:
826 	case DIOCWDINFO:
827 #ifdef __HAVE_OLD_DISKLABEL
828 	case ODIOCSDINFO:
829 	case ODIOCWDINFO:
830 #endif
831 	case DIOCKLABEL:
832 	case DIOCWLABEL:
833 		if ((flag & FWRITE) == 0)
834 			return (EBADF);
835 	}
836 
837 	/* Must be initialized for these... */
838 	switch (cmd) {
839 	case VNDIOCCLR:
840 	case DIOCGDINFO:
841 	case DIOCSDINFO:
842 	case DIOCWDINFO:
843 	case DIOCGPART:
844 	case DIOCKLABEL:
845 	case DIOCWLABEL:
846 	case DIOCGDEFLABEL:
847 #ifdef __HAVE_OLD_DISKLABEL
848 	case ODIOCGDINFO:
849 	case ODIOCSDINFO:
850 	case ODIOCWDINFO:
851 	case ODIOCGDEFLABEL:
852 #endif
853 		if ((vnd->sc_flags & VNF_INITED) == 0)
854 			return (ENXIO);
855 	}
856 
857 	switch (cmd) {
858 	case VNDIOCSET:
859 		if (vnd->sc_flags & VNF_INITED)
860 			return (EBUSY);
861 
862 		if ((error = vndlock(vnd)) != 0)
863 			return (error);
864 
865 		fflags = FREAD;
866 		if ((vio->vnd_flags & VNDIOF_READONLY) == 0)
867 			fflags |= FWRITE;
868 		NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, vio->vnd_file, l);
869 		if ((error = vn_open(&nd, fflags, 0)) != 0)
870 			goto unlock_and_exit;
871 		KASSERT(l);
872 		error = VOP_GETATTR(nd.ni_vp, &vattr, l->l_proc->p_cred, l);
873 		if (!error && nd.ni_vp->v_type != VREG)
874 			error = EOPNOTSUPP;
875 		if (error) {
876 			VOP_UNLOCK(nd.ni_vp, 0);
877 			goto close_and_exit;
878 		}
879 
880 		/* If using a compressed file, initialize its info */
881 		/* (or abort with an error if kernel has no compression) */
882 		if (vio->vnd_flags & VNF_COMP) {
883 #ifdef VND_COMPRESSION
884 			struct vnd_comp_header *ch;
885 			int i;
886 			u_int32_t comp_size;
887 			u_int32_t comp_maxsize;
888 
889 			/* allocate space for compresed file header */
890 			ch = malloc(sizeof(struct vnd_comp_header),
891 			M_TEMP, M_WAITOK);
892 
893 			/* read compressed file header */
894 			error = vn_rdwr(UIO_READ, nd.ni_vp, (caddr_t)ch,
895 			  sizeof(struct vnd_comp_header), 0, UIO_SYSSPACE,
896 			  IO_UNIT|IO_NODELOCKED, p->p_cred, NULL, NULL);
897 			if(error) {
898 				free(ch, M_TEMP);
899 				VOP_UNLOCK(nd.ni_vp, 0);
900 				goto close_and_exit;
901 			}
902 
903 			/* save some header info */
904 			vnd->sc_comp_blksz = ntohl(ch->block_size);
905 			/* note last offset is the file byte size */
906 			vnd->sc_comp_numoffs = ntohl(ch->num_blocks)+1;
907 			free(ch, M_TEMP);
908 			if(vnd->sc_comp_blksz % DEV_BSIZE !=0) {
909 				VOP_UNLOCK(nd.ni_vp, 0);
910 				error = EINVAL;
911 				goto close_and_exit;
912 			}
913 			if(sizeof(struct vnd_comp_header) +
914 			  sizeof(u_int64_t) * vnd->sc_comp_numoffs >
915 			  vattr.va_size) {
916 				VOP_UNLOCK(nd.ni_vp, 0);
917 				error = EINVAL;
918 				goto close_and_exit;
919 			}
920 
921 			/* set decompressed file size */
922 			vattr.va_size =
923 			  (vnd->sc_comp_numoffs - 1) * vnd->sc_comp_blksz;
924 
925 			/* allocate space for all the compressed offsets */
926 			vnd->sc_comp_offsets =
927 			malloc(sizeof(u_int64_t) * vnd->sc_comp_numoffs,
928 			M_DEVBUF, M_WAITOK);
929 
930 			/* read in the offsets */
931 			error = vn_rdwr(UIO_READ, nd.ni_vp,
932 			  (caddr_t)vnd->sc_comp_offsets,
933 			  sizeof(u_int64_t) * vnd->sc_comp_numoffs,
934 			  sizeof(struct vnd_comp_header), UIO_SYSSPACE,
935 			  IO_UNIT|IO_NODELOCKED, p->p_cred, NULL, NULL);
936 			if(error) {
937 				VOP_UNLOCK(nd.ni_vp, 0);
938 				goto close_and_exit;
939 			}
940 			/*
941 			 * find largest block size (used for allocation limit).
942 			 * Also convert offset to native byte order.
943 			 */
944 			comp_maxsize = 0;
945 			for (i = 0; i < vnd->sc_comp_numoffs - 1; i++) {
946 				vnd->sc_comp_offsets[i] =
947 				  be64toh(vnd->sc_comp_offsets[i]);
948 				comp_size = be64toh(vnd->sc_comp_offsets[i + 1])
949 				  - vnd->sc_comp_offsets[i];
950 				if (comp_size > comp_maxsize)
951 					comp_maxsize = comp_size;
952 			}
953 			vnd->sc_comp_offsets[vnd->sc_comp_numoffs - 1] =
954 			  be64toh(vnd->sc_comp_offsets[vnd->sc_comp_numoffs - 1]);
955 
956 			/* create compressed data buffer */
957 			vnd->sc_comp_buff = malloc(comp_maxsize,
958 			  M_DEVBUF, M_WAITOK);
959 
960 			/* create decompressed buffer */
961 			vnd->sc_comp_decombuf = malloc(vnd->sc_comp_blksz,
962 			  M_DEVBUF, M_WAITOK);
963 			vnd->sc_comp_buffblk = -1;
964 
965 			/* Initialize decompress stream */
966 			bzero(&vnd->sc_comp_stream, sizeof(z_stream));
967 			vnd->sc_comp_stream.zalloc = vnd_alloc;
968 			vnd->sc_comp_stream.zfree = vnd_free;
969 			error = inflateInit2(&vnd->sc_comp_stream, MAX_WBITS);
970 			if(error) {
971 				if(vnd->sc_comp_stream.msg)
972 					printf("vnd%d: compressed file, %s\n",
973 					  unit, vnd->sc_comp_stream.msg);
974 				VOP_UNLOCK(nd.ni_vp, 0);
975 				error = EINVAL;
976 				goto close_and_exit;
977 			}
978 
979 			vnd->sc_flags |= VNF_COMP | VNF_READONLY;
980 #else /* !VND_COMPRESSION */
981 			VOP_UNLOCK(nd.ni_vp, 0);
982 			error = EOPNOTSUPP;
983 			goto close_and_exit;
984 #endif /* VND_COMPRESSION */
985 		}
986 
987 		VOP_UNLOCK(nd.ni_vp, 0);
988 		vnd->sc_vp = nd.ni_vp;
989 		vnd->sc_size = btodb(vattr.va_size);	/* note truncation */
990 
991 		/*
992 		 * Use pseudo-geometry specified.  If none was provided,
993 		 * use "standard" Adaptec fictitious geometry.
994 		 */
995 		if (vio->vnd_flags & VNDIOF_HASGEOM) {
996 
997 			memcpy(&vnd->sc_geom, &vio->vnd_geom,
998 			    sizeof(vio->vnd_geom));
999 
1000 			/*
1001 			 * Sanity-check the sector size.
1002 			 * XXX Don't allow secsize < DEV_BSIZE.	 Should
1003 			 * XXX we?
1004 			 */
1005 			if (vnd->sc_geom.vng_secsize < DEV_BSIZE ||
1006 			    (vnd->sc_geom.vng_secsize % DEV_BSIZE) != 0 ||
1007 			    vnd->sc_geom.vng_ncylinders == 0 ||
1008 			    (vnd->sc_geom.vng_ntracks *
1009 			     vnd->sc_geom.vng_nsectors) == 0) {
1010 				error = EINVAL;
1011 				goto close_and_exit;
1012 			}
1013 
1014 			/*
1015 			 * Compute the size (in DEV_BSIZE blocks) specified
1016 			 * by the geometry.
1017 			 */
1018 			geomsize = (vnd->sc_geom.vng_nsectors *
1019 			    vnd->sc_geom.vng_ntracks *
1020 			    vnd->sc_geom.vng_ncylinders) *
1021 			    (vnd->sc_geom.vng_secsize / DEV_BSIZE);
1022 
1023 			/*
1024 			 * Sanity-check the size against the specified
1025 			 * geometry.
1026 			 */
1027 			if (vnd->sc_size < geomsize) {
1028 				error = EINVAL;
1029 				goto close_and_exit;
1030 			}
1031 		} else if (vnd->sc_size >= (32 * 64)) {
1032 			/*
1033 			 * Size must be at least 2048 DEV_BSIZE blocks
1034 			 * (1M) in order to use this geometry.
1035 			 */
1036 			vnd->sc_geom.vng_secsize = DEV_BSIZE;
1037 			vnd->sc_geom.vng_nsectors = 32;
1038 			vnd->sc_geom.vng_ntracks = 64;
1039 			vnd->sc_geom.vng_ncylinders = vnd->sc_size / (64 * 32);
1040 		} else {
1041 			vnd->sc_geom.vng_secsize = DEV_BSIZE;
1042 			vnd->sc_geom.vng_nsectors = 1;
1043 			vnd->sc_geom.vng_ntracks = 1;
1044 			vnd->sc_geom.vng_ncylinders = vnd->sc_size;
1045 		}
1046 
1047 		if (vio->vnd_flags & VNDIOF_READONLY) {
1048 			vnd->sc_flags |= VNF_READONLY;
1049 		}
1050 
1051 		if ((error = vndsetcred(vnd, p->p_cred)) != 0)
1052 			goto close_and_exit;
1053 
1054 		vndthrottle(vnd, vnd->sc_vp);
1055 		vio->vnd_size = dbtob(vnd->sc_size);
1056 		vnd->sc_flags |= VNF_INITED;
1057 
1058 		/* create the kernel thread, wait for it to be up */
1059 		error = kthread_create1(vndthread, vnd, &vnd->sc_kthread,
1060 		    vnd->sc_dev.dv_xname);
1061 		if (error)
1062 			goto close_and_exit;
1063 		while ((vnd->sc_flags & VNF_KTHREAD) == 0) {
1064 			tsleep(&vnd->sc_kthread, PRIBIO, "vndthr", 0);
1065 		}
1066 #ifdef DEBUG
1067 		if (vnddebug & VDB_INIT)
1068 			printf("vndioctl: SET vp %p size 0x%lx %d/%d/%d/%d\n",
1069 			    vnd->sc_vp, (unsigned long) vnd->sc_size,
1070 			    vnd->sc_geom.vng_secsize,
1071 			    vnd->sc_geom.vng_nsectors,
1072 			    vnd->sc_geom.vng_ntracks,
1073 			    vnd->sc_geom.vng_ncylinders);
1074 #endif
1075 
1076 		/* Attach the disk. */
1077 		vnd->sc_dkdev.dk_name = vnd->sc_dev.dv_xname;
1078 		pseudo_disk_attach(&vnd->sc_dkdev);
1079 
1080 		/* Initialize the xfer and buffer pools. */
1081 		pool_init(&vnd->sc_vxpool, sizeof(struct vndxfer), 0,
1082 		    0, 0, "vndxpl", NULL);
1083 
1084 		/* Try and read the disklabel. */
1085 		vndgetdisklabel(dev, vnd);
1086 
1087 		vndunlock(vnd);
1088 
1089 		break;
1090 
1091 close_and_exit:
1092 		(void) vn_close(nd.ni_vp, fflags, p->p_cred, l);
1093 unlock_and_exit:
1094 #ifdef VND_COMPRESSION
1095 		/* free any allocated memory (for compressed file) */
1096 		if(vnd->sc_comp_offsets) {
1097 			free(vnd->sc_comp_offsets, M_DEVBUF);
1098 			vnd->sc_comp_offsets = NULL;
1099 		}
1100 		if(vnd->sc_comp_buff) {
1101 			free(vnd->sc_comp_buff, M_DEVBUF);
1102 			vnd->sc_comp_buff = NULL;
1103 		}
1104 		if(vnd->sc_comp_decombuf) {
1105 			free(vnd->sc_comp_decombuf, M_DEVBUF);
1106 			vnd->sc_comp_decombuf = NULL;
1107 		}
1108 #endif /* VND_COMPRESSION */
1109 		vndunlock(vnd);
1110 		return (error);
1111 
1112 	case VNDIOCCLR:
1113 		if ((error = vndlock(vnd)) != 0)
1114 			return (error);
1115 
1116 		/*
1117 		 * Don't unconfigure if any other partitions are open
1118 		 * or if both the character and block flavors of this
1119 		 * partition are open.
1120 		 */
1121 		part = DISKPART(dev);
1122 		pmask = (1 << part);
1123 		if (((vnd->sc_dkdev.dk_openmask & ~pmask) ||
1124 		    ((vnd->sc_dkdev.dk_bopenmask & pmask) &&
1125 		    (vnd->sc_dkdev.dk_copenmask & pmask))) &&
1126 			!(vio->vnd_flags & VNDIOF_FORCE)) {
1127 			vndunlock(vnd);
1128 			return (EBUSY);
1129 		}
1130 
1131 		/*
1132 		 * XXX vndclear() might call vndclose() implicitely;
1133 		 * release lock to avoid recursion
1134 		 */
1135 		vndunlock(vnd);
1136 		vndclear(vnd, minor(dev));
1137 #ifdef DEBUG
1138 		if (vnddebug & VDB_INIT)
1139 			printf("vndioctl: CLRed\n");
1140 #endif
1141 
1142 		/* Destroy the xfer and buffer pools. */
1143 		pool_destroy(&vnd->sc_vxpool);
1144 
1145 		/* Detatch the disk. */
1146 		pseudo_disk_detach(&vnd->sc_dkdev);
1147 		break;
1148 
1149 #ifdef COMPAT_30
1150 	case VNDIOOCGET: {
1151 		struct vnd_ouser *vnu;
1152 		struct vattr va;
1153 		vnu = (struct vnd_ouser *)data;
1154 		KASSERT(l);
1155 		switch (error = vnd_cget(l, unit, &vnu->vnu_unit, &va)) {
1156 		case 0:
1157 			vnu->vnu_dev = va.va_fsid;
1158 			vnu->vnu_ino = va.va_fileid;
1159 			break;
1160 		case -1:
1161 			/* unused is not an error */
1162 			vnu->vnu_dev = 0;
1163 			vnu->vnu_ino = 0;
1164 			break;
1165 		default:
1166 			return error;
1167 		}
1168 		break;
1169 	}
1170 #endif
1171 	case VNDIOCGET: {
1172 		struct vnd_user *vnu;
1173 		struct vattr va;
1174 		vnu = (struct vnd_user *)data;
1175 		KASSERT(l);
1176 		switch (error = vnd_cget(l, unit, &vnu->vnu_unit, &va)) {
1177 		case 0:
1178 			vnu->vnu_dev = va.va_fsid;
1179 			vnu->vnu_ino = va.va_fileid;
1180 			break;
1181 		case -1:
1182 			/* unused is not an error */
1183 			vnu->vnu_dev = 0;
1184 			vnu->vnu_ino = 0;
1185 			break;
1186 		default:
1187 			return error;
1188 		}
1189 		break;
1190 	}
1191 
1192 	case DIOCGDINFO:
1193 		*(struct disklabel *)data = *(vnd->sc_dkdev.dk_label);
1194 		break;
1195 
1196 #ifdef __HAVE_OLD_DISKLABEL
1197 	case ODIOCGDINFO:
1198 		newlabel = *(vnd->sc_dkdev.dk_label);
1199 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1200 			return ENOTTY;
1201 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1202 		break;
1203 #endif
1204 
1205 	case DIOCGPART:
1206 		((struct partinfo *)data)->disklab = vnd->sc_dkdev.dk_label;
1207 		((struct partinfo *)data)->part =
1208 		    &vnd->sc_dkdev.dk_label->d_partitions[DISKPART(dev)];
1209 		break;
1210 
1211 	case DIOCWDINFO:
1212 	case DIOCSDINFO:
1213 #ifdef __HAVE_OLD_DISKLABEL
1214 	case ODIOCWDINFO:
1215 	case ODIOCSDINFO:
1216 #endif
1217 	{
1218 		struct disklabel *lp;
1219 
1220 		if ((error = vndlock(vnd)) != 0)
1221 			return (error);
1222 
1223 		vnd->sc_flags |= VNF_LABELLING;
1224 
1225 #ifdef __HAVE_OLD_DISKLABEL
1226 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
1227 			memset(&newlabel, 0, sizeof newlabel);
1228 			memcpy(&newlabel, data, sizeof (struct olddisklabel));
1229 			lp = &newlabel;
1230 		} else
1231 #endif
1232 		lp = (struct disklabel *)data;
1233 
1234 		error = setdisklabel(vnd->sc_dkdev.dk_label,
1235 		    lp, 0, vnd->sc_dkdev.dk_cpulabel);
1236 		if (error == 0) {
1237 			if (cmd == DIOCWDINFO
1238 #ifdef __HAVE_OLD_DISKLABEL
1239 			    || cmd == ODIOCWDINFO
1240 #endif
1241 			   )
1242 				error = writedisklabel(VNDLABELDEV(dev),
1243 				    vndstrategy, vnd->sc_dkdev.dk_label,
1244 				    vnd->sc_dkdev.dk_cpulabel);
1245 		}
1246 
1247 		vnd->sc_flags &= ~VNF_LABELLING;
1248 
1249 		vndunlock(vnd);
1250 
1251 		if (error)
1252 			return (error);
1253 		break;
1254 	}
1255 
1256 	case DIOCKLABEL:
1257 		if (*(int *)data != 0)
1258 			vnd->sc_flags |= VNF_KLABEL;
1259 		else
1260 			vnd->sc_flags &= ~VNF_KLABEL;
1261 		break;
1262 
1263 	case DIOCWLABEL:
1264 		if (*(int *)data != 0)
1265 			vnd->sc_flags |= VNF_WLABEL;
1266 		else
1267 			vnd->sc_flags &= ~VNF_WLABEL;
1268 		break;
1269 
1270 	case DIOCGDEFLABEL:
1271 		vndgetdefaultlabel(vnd, (struct disklabel *)data);
1272 		break;
1273 
1274 #ifdef __HAVE_OLD_DISKLABEL
1275 	case ODIOCGDEFLABEL:
1276 		vndgetdefaultlabel(vnd, &newlabel);
1277 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1278 			return ENOTTY;
1279 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1280 		break;
1281 #endif
1282 
1283 	default:
1284 		return (ENOTTY);
1285 	}
1286 
1287 	return (0);
1288 }
1289 
1290 /*
1291  * Duplicate the current processes' credentials.  Since we are called only
1292  * as the result of a SET ioctl and only root can do that, any future access
1293  * to this "disk" is essentially as root.  Note that credentials may change
1294  * if some other uid can write directly to the mapped file (NFS).
1295  */
1296 static int
1297 vndsetcred(struct vnd_softc *vnd, kauth_cred_t cred)
1298 {
1299 	struct uio auio;
1300 	struct iovec aiov;
1301 	char *tmpbuf;
1302 	int error;
1303 
1304 	vnd->sc_cred = kauth_cred_dup(cred);
1305 	tmpbuf = malloc(DEV_BSIZE, M_TEMP, M_WAITOK);
1306 
1307 	/* XXX: Horrible kludge to establish credentials for NFS */
1308 	aiov.iov_base = tmpbuf;
1309 	aiov.iov_len = min(DEV_BSIZE, dbtob(vnd->sc_size));
1310 	auio.uio_iov = &aiov;
1311 	auio.uio_iovcnt = 1;
1312 	auio.uio_offset = 0;
1313 	auio.uio_rw = UIO_READ;
1314 	auio.uio_resid = aiov.iov_len;
1315 	UIO_SETUP_SYSSPACE(&auio);
1316 	vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1317 	error = VOP_READ(vnd->sc_vp, &auio, 0, vnd->sc_cred);
1318 	if (error == 0) {
1319 		/*
1320 		 * Because vnd does all IO directly through the vnode
1321 		 * we need to flush (at least) the buffer from the above
1322 		 * VOP_READ from the buffer cache to prevent cache
1323 		 * incoherencies.  Also, be careful to write dirty
1324 		 * buffers back to stable storage.
1325 		 */
1326 		error = vinvalbuf(vnd->sc_vp, V_SAVE, vnd->sc_cred,
1327 			    curlwp, 0, 0);
1328 	}
1329 	VOP_UNLOCK(vnd->sc_vp, 0);
1330 
1331 	free(tmpbuf, M_TEMP);
1332 	return (error);
1333 }
1334 
1335 /*
1336  * Set maxactive based on FS type
1337  */
1338 static void
1339 vndthrottle(struct vnd_softc *vnd, struct vnode *vp)
1340 {
1341 #ifdef NFS
1342 	extern int (**nfsv2_vnodeop_p)(void *);
1343 
1344 	if (vp->v_op == nfsv2_vnodeop_p)
1345 		vnd->sc_maxactive = 2;
1346 	else
1347 #endif
1348 		vnd->sc_maxactive = 8;
1349 
1350 	if (vnd->sc_maxactive < 1)
1351 		vnd->sc_maxactive = 1;
1352 }
1353 
1354 #if 0
1355 static void
1356 vndshutdown(void)
1357 {
1358 	struct vnd_softc *vnd;
1359 
1360 	for (vnd = &vnd_softc[0]; vnd < &vnd_softc[numvnd]; vnd++)
1361 		if (vnd->sc_flags & VNF_INITED)
1362 			vndclear(vnd);
1363 }
1364 #endif
1365 
1366 static void
1367 vndclear(struct vnd_softc *vnd, int myminor)
1368 {
1369 	struct vnode *vp = vnd->sc_vp;
1370 	struct lwp *l = curlwp;
1371 	int fflags = FREAD;
1372 	int bmaj, cmaj, i, mn;
1373 	int s;
1374 
1375 #ifdef DEBUG
1376 	if (vnddebug & VDB_FOLLOW)
1377 		printf("vndclear(%p): vp %p\n", vnd, vp);
1378 #endif
1379 	/* locate the major number */
1380 	bmaj = bdevsw_lookup_major(&vnd_bdevsw);
1381 	cmaj = cdevsw_lookup_major(&vnd_cdevsw);
1382 
1383 	/* Nuke the vnodes for any open instances */
1384 	for (i = 0; i < MAXPARTITIONS; i++) {
1385 		mn = DISKMINOR(device_unit(&vnd->sc_dev), i);
1386 		vdevgone(bmaj, mn, mn, VBLK);
1387 		if (mn != myminor) /* XXX avoid to kill own vnode */
1388 			vdevgone(cmaj, mn, mn, VCHR);
1389 	}
1390 
1391 	if ((vnd->sc_flags & VNF_READONLY) == 0)
1392 		fflags |= FWRITE;
1393 
1394 	s = splbio();
1395 	bufq_drain(vnd->sc_tab);
1396 	splx(s);
1397 
1398 	vnd->sc_flags |= VNF_VUNCONF;
1399 	wakeup(&vnd->sc_tab);
1400 	while (vnd->sc_flags & VNF_KTHREAD)
1401 		tsleep(&vnd->sc_kthread, PRIBIO, "vnthr", 0);
1402 
1403 #ifdef VND_COMPRESSION
1404 	/* free the compressed file buffers */
1405 	if(vnd->sc_flags & VNF_COMP) {
1406 		if(vnd->sc_comp_offsets) {
1407 			free(vnd->sc_comp_offsets, M_DEVBUF);
1408 			vnd->sc_comp_offsets = NULL;
1409 		}
1410 		if(vnd->sc_comp_buff) {
1411 			free(vnd->sc_comp_buff, M_DEVBUF);
1412 			vnd->sc_comp_buff = NULL;
1413 		}
1414 		if(vnd->sc_comp_decombuf) {
1415 			free(vnd->sc_comp_decombuf, M_DEVBUF);
1416 			vnd->sc_comp_decombuf = NULL;
1417 		}
1418 	}
1419 #endif /* VND_COMPRESSION */
1420 	vnd->sc_flags &=
1421 	    ~(VNF_INITED | VNF_READONLY | VNF_VLABEL
1422 	      | VNF_VUNCONF | VNF_COMP);
1423 	if (vp == (struct vnode *)0)
1424 		panic("vndclear: null vp");
1425 	(void) vn_close(vp, fflags, vnd->sc_cred, l);
1426 	kauth_cred_free(vnd->sc_cred);
1427 	vnd->sc_vp = (struct vnode *)0;
1428 	vnd->sc_cred = (kauth_cred_t)0;
1429 	vnd->sc_size = 0;
1430 }
1431 
1432 static int
1433 vndsize(dev_t dev)
1434 {
1435 	struct vnd_softc *sc;
1436 	struct disklabel *lp;
1437 	int part, unit, omask;
1438 	int size;
1439 
1440 	unit = vndunit(dev);
1441 	sc = (struct vnd_softc *)device_lookup(&vnd_cd, unit);
1442 	if (sc == NULL)
1443 		return -1;
1444 
1445 	if ((sc->sc_flags & VNF_INITED) == 0)
1446 		return (-1);
1447 
1448 	part = DISKPART(dev);
1449 	omask = sc->sc_dkdev.dk_openmask & (1 << part);
1450 	lp = sc->sc_dkdev.dk_label;
1451 
1452 	if (omask == 0 && vndopen(dev, 0, S_IFBLK, curlwp))	/* XXX */
1453 		return (-1);
1454 
1455 	if (lp->d_partitions[part].p_fstype != FS_SWAP)
1456 		size = -1;
1457 	else
1458 		size = lp->d_partitions[part].p_size *
1459 		    (lp->d_secsize / DEV_BSIZE);
1460 
1461 	if (omask == 0 && vndclose(dev, 0, S_IFBLK, curlwp))	/* XXX */
1462 		return (-1);
1463 
1464 	return (size);
1465 }
1466 
1467 static int
1468 vnddump(dev_t dev, daddr_t blkno, caddr_t va, size_t size)
1469 {
1470 
1471 	/* Not implemented. */
1472 	return ENXIO;
1473 }
1474 
1475 static void
1476 vndgetdefaultlabel(struct vnd_softc *sc, struct disklabel *lp)
1477 {
1478 	struct vndgeom *vng = &sc->sc_geom;
1479 	struct partition *pp;
1480 
1481 	memset(lp, 0, sizeof(*lp));
1482 
1483 	lp->d_secperunit = sc->sc_size / (vng->vng_secsize / DEV_BSIZE);
1484 	lp->d_secsize = vng->vng_secsize;
1485 	lp->d_nsectors = vng->vng_nsectors;
1486 	lp->d_ntracks = vng->vng_ntracks;
1487 	lp->d_ncylinders = vng->vng_ncylinders;
1488 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1489 
1490 	strncpy(lp->d_typename, "vnd", sizeof(lp->d_typename));
1491 	lp->d_type = DTYPE_VND;
1492 	strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
1493 	lp->d_rpm = 3600;
1494 	lp->d_interleave = 1;
1495 	lp->d_flags = 0;
1496 
1497 	pp = &lp->d_partitions[RAW_PART];
1498 	pp->p_offset = 0;
1499 	pp->p_size = lp->d_secperunit;
1500 	pp->p_fstype = FS_UNUSED;
1501 	lp->d_npartitions = RAW_PART + 1;
1502 
1503 	lp->d_magic = DISKMAGIC;
1504 	lp->d_magic2 = DISKMAGIC;
1505 	lp->d_checksum = dkcksum(lp);
1506 }
1507 
1508 /*
1509  * Read the disklabel from a vnd.  If one is not present, create a fake one.
1510  */
1511 static void
1512 vndgetdisklabel(dev_t dev, struct vnd_softc *sc)
1513 {
1514 	const char *errstring;
1515 	struct disklabel *lp = sc->sc_dkdev.dk_label;
1516 	struct cpu_disklabel *clp = sc->sc_dkdev.dk_cpulabel;
1517 	int i;
1518 
1519 	memset(clp, 0, sizeof(*clp));
1520 
1521 	vndgetdefaultlabel(sc, lp);
1522 
1523 	/*
1524 	 * Call the generic disklabel extraction routine.
1525 	 */
1526 	errstring = readdisklabel(VNDLABELDEV(dev), vndstrategy, lp, clp);
1527 	if (errstring) {
1528 		/*
1529 		 * Lack of disklabel is common, but we print the warning
1530 		 * anyway, since it might contain other useful information.
1531 		 */
1532 		printf("%s: %s\n", sc->sc_dev.dv_xname, errstring);
1533 
1534 		/*
1535 		 * For historical reasons, if there's no disklabel
1536 		 * present, all partitions must be FS_BSDFFS and
1537 		 * occupy the entire disk.
1538 		 */
1539 		for (i = 0; i < MAXPARTITIONS; i++) {
1540 			/*
1541 			 * Don't wipe out port specific hack (such as
1542 			 * dos partition hack of i386 port).
1543 			 */
1544 			if (lp->d_partitions[i].p_size != 0)
1545 				continue;
1546 
1547 			lp->d_partitions[i].p_size = lp->d_secperunit;
1548 			lp->d_partitions[i].p_offset = 0;
1549 			lp->d_partitions[i].p_fstype = FS_BSDFFS;
1550 		}
1551 
1552 		strncpy(lp->d_packname, "default label",
1553 		    sizeof(lp->d_packname));
1554 
1555 		lp->d_npartitions = MAXPARTITIONS;
1556 		lp->d_checksum = dkcksum(lp);
1557 	}
1558 
1559 	/* In-core label now valid. */
1560 	sc->sc_flags |= VNF_VLABEL;
1561 }
1562 
1563 /*
1564  * Wait interruptibly for an exclusive lock.
1565  *
1566  * XXX
1567  * Several drivers do this; it should be abstracted and made MP-safe.
1568  */
1569 static int
1570 vndlock(struct vnd_softc *sc)
1571 {
1572 	int error;
1573 
1574 	while ((sc->sc_flags & VNF_LOCKED) != 0) {
1575 		sc->sc_flags |= VNF_WANTED;
1576 		if ((error = tsleep(sc, PRIBIO | PCATCH, "vndlck", 0)) != 0)
1577 			return (error);
1578 	}
1579 	sc->sc_flags |= VNF_LOCKED;
1580 	return (0);
1581 }
1582 
1583 /*
1584  * Unlock and wake up any waiters.
1585  */
1586 static void
1587 vndunlock(struct vnd_softc *sc)
1588 {
1589 
1590 	sc->sc_flags &= ~VNF_LOCKED;
1591 	if ((sc->sc_flags & VNF_WANTED) != 0) {
1592 		sc->sc_flags &= ~VNF_WANTED;
1593 		wakeup(sc);
1594 	}
1595 }
1596 
1597 #ifdef VND_COMPRESSION
1598 /* compressed file read */
1599 static void
1600 compstrategy(struct buf *bp, off_t bn)
1601 {
1602 	int error;
1603 	int unit = vndunit(bp->b_dev);
1604 	struct vnd_softc *vnd =
1605 	    (struct vnd_softc *)device_lookup(&vnd_cd, unit);
1606 	u_int32_t comp_block;
1607 	struct uio auio;
1608 	caddr_t addr;
1609 	int s;
1610 
1611 	/* set up constants for data move */
1612 	auio.uio_rw = UIO_READ;
1613 	UIO_SETUP_SYSSPACE(&auio);
1614 
1615 	/* read, and transfer the data */
1616 	addr = bp->b_data;
1617 	s = splbio();
1618 	while (bp->b_resid > 0) {
1619 		unsigned length;
1620 		size_t length_in_buffer;
1621 		u_int32_t offset_in_buffer;
1622 		struct iovec aiov;
1623 
1624 		/* calculate the compressed block number */
1625 		comp_block = bn / (off_t)vnd->sc_comp_blksz;
1626 
1627 		/* check for good block number */
1628 		if (comp_block >= vnd->sc_comp_numoffs) {
1629 			bp->b_error = EINVAL;
1630 			bp->b_flags |= B_ERROR;
1631 			splx(s);
1632 			return;
1633 		}
1634 
1635 		/* read in the compressed block, if not in buffer */
1636 		if (comp_block != vnd->sc_comp_buffblk) {
1637 			length = vnd->sc_comp_offsets[comp_block + 1] -
1638 			    vnd->sc_comp_offsets[comp_block];
1639 			vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1640 			error = vn_rdwr(UIO_READ, vnd->sc_vp, vnd->sc_comp_buff,
1641 			    length, vnd->sc_comp_offsets[comp_block],
1642 			    UIO_SYSSPACE, IO_UNIT, vnd->sc_cred, NULL, NULL);
1643 			if (error) {
1644 				bp->b_error = error;
1645 				bp->b_flags |= B_ERROR;
1646 				VOP_UNLOCK(vnd->sc_vp, 0);
1647 				splx(s);
1648 				return;
1649 			}
1650 			/* uncompress the buffer */
1651 			vnd->sc_comp_stream.next_in = vnd->sc_comp_buff;
1652 			vnd->sc_comp_stream.avail_in = length;
1653 			vnd->sc_comp_stream.next_out = vnd->sc_comp_decombuf;
1654 			vnd->sc_comp_stream.avail_out = vnd->sc_comp_blksz;
1655 			inflateReset(&vnd->sc_comp_stream);
1656 			error = inflate(&vnd->sc_comp_stream, Z_FINISH);
1657 			if (error != Z_STREAM_END) {
1658 				if (vnd->sc_comp_stream.msg)
1659 					printf("%s: compressed file, %s\n",
1660 					    vnd->sc_dev.dv_xname,
1661 					    vnd->sc_comp_stream.msg);
1662 				bp->b_error = EBADMSG;
1663 				bp->b_flags |= B_ERROR;
1664 				VOP_UNLOCK(vnd->sc_vp, 0);
1665 				splx(s);
1666 				return;
1667 			}
1668 			vnd->sc_comp_buffblk = comp_block;
1669 			VOP_UNLOCK(vnd->sc_vp, 0);
1670 		}
1671 
1672 		/* transfer the usable uncompressed data */
1673 		offset_in_buffer = bn % (off_t)vnd->sc_comp_blksz;
1674 		length_in_buffer = vnd->sc_comp_blksz - offset_in_buffer;
1675 		if (length_in_buffer > bp->b_resid)
1676 			length_in_buffer = bp->b_resid;
1677 		auio.uio_iov = &aiov;
1678 		auio.uio_iovcnt = 1;
1679 		aiov.iov_base = addr;
1680 		aiov.iov_len = length_in_buffer;
1681 		auio.uio_resid = aiov.iov_len;
1682 		auio.uio_offset = 0;
1683 		error = uiomove(vnd->sc_comp_decombuf + offset_in_buffer,
1684 		    length_in_buffer, &auio);
1685 		if (error) {
1686 			bp->b_error = error;
1687 			bp->b_flags |= B_ERROR;
1688 			splx(s);
1689 			return;
1690 		}
1691 
1692 		bn += length_in_buffer;
1693 		addr += length_in_buffer;
1694 		bp->b_resid -= length_in_buffer;
1695 	}
1696 	splx(s);
1697 }
1698 
1699 /* compression memory allocation routines */
1700 static void *
1701 vnd_alloc(void *aux, u_int items, u_int siz)
1702 {
1703 	return malloc(items * siz, M_TEMP, M_NOWAIT);
1704 }
1705 
1706 static void
1707 vnd_free(void *aux, void *ptr)
1708 {
1709 	free(ptr, M_TEMP);
1710 }
1711 #endif /* VND_COMPRESSION */
1712