xref: /netbsd-src/sys/dev/vnd.c (revision 7fa608457b817eca6e0977b37f758ae064f3c99c)
1 /*	$NetBSD: vnd.c,v 1.170 2007/10/08 16:41:11 ad 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 1.170 2007/10/08 16:41:11 ad 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 }
290 
291 static int
292 vnd_detach(struct device *self, int flags)
293 {
294 	struct vnd_softc *sc = (struct vnd_softc *)self;
295 	if (sc->sc_flags & VNF_INITED)
296 		return EBUSY;
297 
298 	bufq_free(sc->sc_tab);
299 	disk_destroy(&sc->sc_dkdev);
300 
301 	return 0;
302 }
303 
304 static struct vnd_softc *
305 vnd_spawn(int unit)
306 {
307 	struct cfdata *cf;
308 
309 	cf = malloc(sizeof(*cf), M_DEVBUF, M_WAITOK);
310 	cf->cf_name = vnd_cd.cd_name;
311 	cf->cf_atname = vnd_cd.cd_name;
312 	cf->cf_unit = unit;
313 	cf->cf_fstate = FSTATE_STAR;
314 
315 	return (struct vnd_softc *)config_attach_pseudo(cf);
316 }
317 
318 int
319 vnd_destroy(struct device *dev)
320 {
321 	int error;
322 	struct cfdata *cf;
323 
324 	cf = device_cfdata(dev);
325 	error = config_detach(dev, DETACH_QUIET);
326 	if (error)
327 		return error;
328 	free(cf, M_DEVBUF);
329 	return 0;
330 }
331 
332 static int
333 vndopen(dev_t dev, int flags, int mode, struct lwp *l)
334 {
335 	int unit = vndunit(dev);
336 	struct vnd_softc *sc;
337 	int error = 0, part, pmask;
338 	struct disklabel *lp;
339 
340 #ifdef DEBUG
341 	if (vnddebug & VDB_FOLLOW)
342 		printf("vndopen(0x%x, 0x%x, 0x%x, %p)\n", dev, flags, mode, l);
343 #endif
344 	sc = device_lookup(&vnd_cd, unit);
345 	if (sc == NULL) {
346 		sc = vnd_spawn(unit);
347 		if (sc == NULL)
348 			return ENOMEM;
349 	}
350 
351 	if ((error = vndlock(sc)) != 0)
352 		return (error);
353 
354 	lp = sc->sc_dkdev.dk_label;
355 
356 	part = DISKPART(dev);
357 	pmask = (1 << part);
358 
359 	/*
360 	 * If we're initialized, check to see if there are any other
361 	 * open partitions.  If not, then it's safe to update the
362 	 * in-core disklabel.  Only read the disklabel if it is
363 	 * not already valid.
364 	 */
365 	if ((sc->sc_flags & (VNF_INITED|VNF_VLABEL)) == VNF_INITED &&
366 	    sc->sc_dkdev.dk_openmask == 0)
367 		vndgetdisklabel(dev, sc);
368 
369 	/* Check that the partitions exists. */
370 	if (part != RAW_PART) {
371 		if (((sc->sc_flags & VNF_INITED) == 0) ||
372 		    ((part >= lp->d_npartitions) ||
373 		     (lp->d_partitions[part].p_fstype == FS_UNUSED))) {
374 			error = ENXIO;
375 			goto done;
376 		}
377 	}
378 
379 	/* Prevent our unit from being unconfigured while open. */
380 	switch (mode) {
381 	case S_IFCHR:
382 		sc->sc_dkdev.dk_copenmask |= pmask;
383 		break;
384 
385 	case S_IFBLK:
386 		sc->sc_dkdev.dk_bopenmask |= pmask;
387 		break;
388 	}
389 	sc->sc_dkdev.dk_openmask =
390 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
391 
392  done:
393 	vndunlock(sc);
394 	return (error);
395 }
396 
397 static int
398 vndclose(dev_t dev, int flags, int mode, struct lwp *l)
399 {
400 	int unit = vndunit(dev);
401 	struct vnd_softc *sc;
402 	int error = 0, part;
403 
404 #ifdef DEBUG
405 	if (vnddebug & VDB_FOLLOW)
406 		printf("vndclose(0x%x, 0x%x, 0x%x, %p)\n", dev, flags, mode, l);
407 #endif
408 	sc = device_lookup(&vnd_cd, unit);
409 	if (sc == NULL)
410 		return ENXIO;
411 
412 	if ((error = vndlock(sc)) != 0)
413 		return (error);
414 
415 	part = DISKPART(dev);
416 
417 	/* ...that much closer to allowing unconfiguration... */
418 	switch (mode) {
419 	case S_IFCHR:
420 		sc->sc_dkdev.dk_copenmask &= ~(1 << part);
421 		break;
422 
423 	case S_IFBLK:
424 		sc->sc_dkdev.dk_bopenmask &= ~(1 << part);
425 		break;
426 	}
427 	sc->sc_dkdev.dk_openmask =
428 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
429 
430 	vndunlock(sc);
431 
432 	if ((sc->sc_flags & VNF_INITED) == 0) {
433 		if ((error = vnd_destroy((struct device *)sc)) != 0) {
434 			aprint_error("%s: unable to detach instance\n",
435 			    sc->sc_dev.dv_xname);
436 			return error;
437 		}
438 	}
439 
440 	return (0);
441 }
442 
443 /*
444  * Queue the request, and wakeup the kernel thread to handle it.
445  */
446 static void
447 vndstrategy(struct buf *bp)
448 {
449 	int unit = vndunit(bp->b_dev);
450 	struct vnd_softc *vnd =
451 	    (struct vnd_softc *)device_lookup(&vnd_cd, unit);
452 	struct disklabel *lp = vnd->sc_dkdev.dk_label;
453 	daddr_t blkno;
454 	int s = splbio();
455 
456 	if ((vnd->sc_flags & VNF_INITED) == 0) {
457 		bp->b_error = ENXIO;
458 		goto done;
459 	}
460 
461 	/*
462 	 * The transfer must be a whole number of blocks.
463 	 */
464 	if ((bp->b_bcount % lp->d_secsize) != 0) {
465 		bp->b_error = EINVAL;
466 		goto done;
467 	}
468 
469 	/*
470 	 * check if we're read-only.
471 	 */
472 	if ((vnd->sc_flags & VNF_READONLY) && !(bp->b_flags & B_READ)) {
473 		bp->b_error = EACCES;
474 		goto done;
475 	}
476 
477 	/* If it's a nil transfer, wake up the top half now. */
478 	if (bp->b_bcount == 0) {
479 		goto done;
480 	}
481 
482 	/*
483 	 * Do bounds checking and adjust transfer.  If there's an error,
484 	 * the bounds check will flag that for us.
485 	 */
486 	if (DISKPART(bp->b_dev) == RAW_PART) {
487 		if (bounds_check_with_mediasize(bp, DEV_BSIZE,
488 		    vnd->sc_size) <= 0)
489 			goto done;
490 	} else {
491 		if (bounds_check_with_label(&vnd->sc_dkdev,
492 		    bp, vnd->sc_flags & (VNF_WLABEL|VNF_LABELLING)) <= 0)
493 			goto done;
494 	}
495 
496 	/*
497 	 * Put the block number in terms of the logical blocksize
498 	 * of the "device".
499 	 */
500 
501 	blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
502 
503 	/*
504 	 * Translate the partition-relative block number to an absolute.
505 	 */
506 	if (DISKPART(bp->b_dev) != RAW_PART) {
507 		struct partition *pp;
508 
509 		pp = &vnd->sc_dkdev.dk_label->d_partitions[
510 		    DISKPART(bp->b_dev)];
511 		blkno += pp->p_offset;
512 	}
513 	bp->b_rawblkno = blkno;
514 
515 #ifdef DEBUG
516 	if (vnddebug & VDB_FOLLOW)
517 		printf("vndstrategy(%p): unit %d\n", bp, unit);
518 #endif
519 	BUFQ_PUT(vnd->sc_tab, bp);
520 	wakeup(&vnd->sc_tab);
521 	splx(s);
522 	return;
523 
524 done:
525 	bp->b_resid = bp->b_bcount;
526 	biodone(bp);
527 	splx(s);
528 }
529 
530 static void
531 vndthread(void *arg)
532 {
533 	struct vnd_softc *vnd = arg;
534 	bool usestrategy;
535 	int s;
536 
537 	/* Determine whether we can use VOP_BMAP and VOP_STRATEGY to
538 	 * directly access the backing vnode.  If we can, use these two
539 	 * operations to avoid messing with the local buffer cache.
540 	 * Otherwise fall back to regular VOP_READ/VOP_WRITE operations
541 	 * which are guaranteed to work with any file system. */
542 	usestrategy = vnode_has_op(vnd->sc_vp, VOFFSET(vop_bmap)) &&
543 	    vnode_has_op(vnd->sc_vp, VOFFSET(vop_strategy));
544 
545 #ifdef DEBUG
546 	if (vnddebug & VDB_INIT)
547 		printf("vndthread: vp %p, %s\n", vnd->sc_vp,
548 		    usestrategy ?
549 		    "using bmap/strategy operations" :
550 		    "using read/write operations");
551 #endif
552 
553 	s = splbio();
554 	vnd->sc_flags |= VNF_KTHREAD;
555 	wakeup(&vnd->sc_kthread);
556 
557 	/*
558 	 * Dequeue requests and serve them depending on the available
559 	 * vnode operations.
560 	 */
561 	while ((vnd->sc_flags & VNF_VUNCONF) == 0) {
562 		struct vndxfer *vnx;
563 		int flags;
564 		struct buf *obp;
565 		struct buf *bp;
566 
567 		obp = BUFQ_GET(vnd->sc_tab);
568 		if (obp == NULL) {
569 			tsleep(&vnd->sc_tab, PRIBIO, "vndbp", 0);
570 			continue;
571 		};
572 		splx(s);
573 		flags = obp->b_flags;
574 #ifdef DEBUG
575 		if (vnddebug & VDB_FOLLOW)
576 			printf("vndthread(%p\n", obp);
577 #endif
578 
579 		if (vnd->sc_vp->v_mount == NULL) {
580 			obp->b_error = ENXIO;
581 			goto done;
582 		}
583 #ifdef VND_COMPRESSION
584 		/* handle a compressed read */
585 		if ((flags & B_READ) != 0 && (vnd->sc_flags & VNF_COMP)) {
586 			off_t bn;
587 
588 			/* Convert to a byte offset within the file. */
589 			bn = obp->b_rawblkno *
590 			    vnd->sc_dkdev.dk_label->d_secsize;
591 
592 			compstrategy(obp, bn);
593 			goto done;
594 		}
595 #endif /* VND_COMPRESSION */
596 
597 		/*
598 		 * Allocate a header for this transfer and link it to the
599 		 * buffer
600 		 */
601 		s = splbio();
602 		vnx = VND_GETXFER(vnd);
603 		splx(s);
604 		vnx->vx_vnd = vnd;
605 
606 		s = splbio();
607 		while (vnd->sc_active >= vnd->sc_maxactive) {
608 			tsleep(&vnd->sc_tab, PRIBIO, "vndac", 0);
609 		}
610 		vnd->sc_active++;
611 		splx(s);
612 
613 		/* Instrumentation. */
614 		disk_busy(&vnd->sc_dkdev);
615 
616 		bp = &vnx->vx_buf;
617 		BUF_INIT(bp);
618 		bp->b_flags = (obp->b_flags & B_READ) | B_CALL;
619 		bp->b_iodone = vndiodone;
620 		bp->b_private = obp;
621 		bp->b_vp = vnd->sc_vp;
622 		bp->b_data = obp->b_data;
623 		bp->b_bcount = obp->b_bcount;
624 		BIO_COPYPRIO(bp, obp);
625 
626 		/* Handle the request using the appropriate operations. */
627 		if (usestrategy)
628 			handle_with_strategy(vnd, obp, bp);
629 		else
630 			handle_with_rdwr(vnd, obp, bp);
631 
632 		s = splbio();
633 		continue;
634 
635 done:
636 		biodone(obp);
637 		s = splbio();
638 	}
639 
640 	vnd->sc_flags &= (~VNF_KTHREAD | VNF_VUNCONF);
641 	wakeup(&vnd->sc_kthread);
642 	splx(s);
643 	kthread_exit(0);
644 }
645 
646 /*
647  * Checks if the given vnode supports the requested operation.
648  * The operation is specified the offset returned by VOFFSET.
649  *
650  * XXX The test below used to determine this is quite fragile
651  * because it relies on the file system to use genfs to specify
652  * unimplemented operations.  There might be another way to do
653  * it more cleanly.
654  */
655 static bool
656 vnode_has_op(const struct vnode *vp, int opoffset)
657 {
658 	int (*defaultp)(void *);
659 	int (*opp)(void *);
660 
661 	defaultp = vp->v_op[VOFFSET(vop_default)];
662 	opp = vp->v_op[opoffset];
663 
664 	return opp != defaultp && opp != genfs_eopnotsupp &&
665 	    opp != genfs_badop && opp != genfs_nullop;
666 }
667 
668 /*
669  * Handes the read/write request given in 'bp' using the vnode's VOP_READ
670  * and VOP_WRITE operations.
671  *
672  * 'obp' is a pointer to the original request fed to the vnd device.
673  */
674 static void
675 handle_with_rdwr(struct vnd_softc *vnd, const struct buf *obp, struct buf *bp)
676 {
677 	bool doread;
678 	off_t offset;
679 	size_t resid;
680 	struct vnode *vp;
681 
682 	doread = bp->b_flags & B_READ;
683 	offset = obp->b_rawblkno * vnd->sc_dkdev.dk_label->d_secsize;
684 	vp = vnd->sc_vp;
685 
686 #if defined(DEBUG)
687 	if (vnddebug & VDB_IO)
688 		printf("vnd (rdwr): vp %p, %s, rawblkno 0x%" PRIx64
689 		    ", secsize %d, offset %" PRIu64
690 		    ", bcount %d\n",
691 		    vp, doread ? "read" : "write", obp->b_rawblkno,
692 		    vnd->sc_dkdev.dk_label->d_secsize, offset,
693 		    bp->b_bcount);
694 #endif
695 
696 	/* Issue the read or write operation. */
697 	bp->b_error =
698 	    vn_rdwr(doread ? UIO_READ : UIO_WRITE,
699 	    vp, bp->b_data, bp->b_bcount, offset,
700 	    UIO_SYSSPACE, 0, vnd->sc_cred, &resid, NULL);
701 	bp->b_resid = resid;
702 
703 	/* We need to increase the number of outputs on the vnode if
704 	 * there was any write to it. */
705 	if (!doread)
706 		V_INCR_NUMOUTPUT(vp);
707 
708 	biodone(bp);
709 }
710 
711 /*
712  * Handes the read/write request given in 'bp' using the vnode's VOP_BMAP
713  * and VOP_STRATEGY operations.
714  *
715  * 'obp' is a pointer to the original request fed to the vnd device.
716  */
717 static void
718 handle_with_strategy(struct vnd_softc *vnd, const struct buf *obp,
719     struct buf *bp)
720 {
721 	int bsize, error, flags, skipped;
722 	size_t resid, sz;
723 	off_t bn, offset;
724 
725 	flags = obp->b_flags;
726 
727 	if (!(flags & B_READ)) {
728 		int s;
729 
730 		s = splbio();
731 		V_INCR_NUMOUTPUT(bp->b_vp);
732 		splx(s);
733 	}
734 
735 	/* convert to a byte offset within the file. */
736 	bn = obp->b_rawblkno * vnd->sc_dkdev.dk_label->d_secsize;
737 
738 	bsize = vnd->sc_vp->v_mount->mnt_stat.f_iosize;
739 	skipped = 0;
740 
741 	/*
742 	 * Break the request into bsize pieces and feed them
743 	 * sequentially using VOP_BMAP/VOP_STRATEGY.
744 	 * We do it this way to keep from flooding NFS servers if we
745 	 * are connected to an NFS file.  This places the burden on
746 	 * the client rather than the server.
747 	 */
748 	error = 0;
749 	bp->b_resid = bp->b_bcount;
750 	for (offset = 0, resid = bp->b_resid; resid;
751 	    resid -= sz, offset += sz) {
752 		struct buf *nbp;
753 		struct vnode *vp;
754 		daddr_t nbn;
755 		int off, nra;
756 
757 		nra = 0;
758 		vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY | LK_CANRECURSE);
759 		error = VOP_BMAP(vnd->sc_vp, bn / bsize, &vp, &nbn, &nra);
760 		VOP_UNLOCK(vnd->sc_vp, 0);
761 
762 		if (error == 0 && (long)nbn == -1)
763 			error = EIO;
764 
765 		/*
766 		 * If there was an error or a hole in the file...punt.
767 		 * Note that we may have to wait for any operations
768 		 * that we have already fired off before releasing
769 		 * the buffer.
770 		 *
771 		 * XXX we could deal with holes here but it would be
772 		 * a hassle (in the write case).
773 		 */
774 		if (error) {
775 			skipped += resid;
776 			break;
777 		}
778 
779 #ifdef DEBUG
780 		if (!dovndcluster)
781 			nra = 0;
782 #endif
783 
784 		off = bn % bsize;
785 		sz = MIN(((off_t)1 + nra) * bsize - off, resid);
786 #ifdef	DEBUG
787 		if (vnddebug & VDB_IO)
788 			printf("vndstrategy: vp %p/%p bn 0x%qx/0x%" PRIx64
789 			       " sz 0x%zx\n",
790 			    vnd->sc_vp, vp, (long long)bn, nbn, sz);
791 #endif
792 
793 		nbp = getiobuf();
794 		nestiobuf_setup(bp, nbp, offset, sz);
795 		nbp->b_blkno = nbn + btodb(off);
796 
797 #if 0 /* XXX #ifdef DEBUG */
798 		if (vnddebug & VDB_IO)
799 			printf("vndstart(%ld): bp %p vp %p blkno "
800 			    "0x%" PRIx64 " flags %x addr %p cnt 0x%x\n",
801 			    (long) (vnd-vnd_softc), &nbp->vb_buf,
802 			    nbp->vb_buf.b_vp, nbp->vb_buf.b_blkno,
803 			    nbp->vb_buf.b_flags, nbp->vb_buf.b_data,
804 			    nbp->vb_buf.b_bcount);
805 #endif
806 		VOP_STRATEGY(vp, nbp);
807 		bn += sz;
808 	}
809 	nestiobuf_done(bp, skipped, error);
810 }
811 
812 static void
813 vndiodone(struct buf *bp)
814 {
815 	struct vndxfer *vnx = VND_BUFTOXFER(bp);
816 	struct vnd_softc *vnd = vnx->vx_vnd;
817 	struct buf *obp = bp->b_private;
818 
819 	KASSERT(&vnx->vx_buf == bp);
820 	KASSERT(vnd->sc_active > 0);
821 #ifdef DEBUG
822 	if (vnddebug & VDB_IO) {
823 		printf("vndiodone1: bp %p iodone: error %d\n",
824 		    bp, bp->b_error);
825 	}
826 #endif
827 	disk_unbusy(&vnd->sc_dkdev, bp->b_bcount - bp->b_resid,
828 	    (bp->b_flags & B_READ));
829 	vnd->sc_active--;
830 	if (vnd->sc_active == 0) {
831 		wakeup(&vnd->sc_tab);
832 	}
833 	obp->b_error = bp->b_error;
834 	obp->b_resid = bp->b_resid;
835 	VND_PUTXFER(vnd, vnx);
836 	biodone(obp);
837 }
838 
839 /* ARGSUSED */
840 static int
841 vndread(dev_t dev, struct uio *uio, int flags)
842 {
843 	int unit = vndunit(dev);
844 	struct vnd_softc *sc;
845 
846 #ifdef DEBUG
847 	if (vnddebug & VDB_FOLLOW)
848 		printf("vndread(0x%x, %p)\n", dev, uio);
849 #endif
850 
851 	sc = device_lookup(&vnd_cd, unit);
852 	if (sc == NULL)
853 		return ENXIO;
854 
855 	if ((sc->sc_flags & VNF_INITED) == 0)
856 		return (ENXIO);
857 
858 	return (physio(vndstrategy, NULL, dev, B_READ, minphys, uio));
859 }
860 
861 /* ARGSUSED */
862 static int
863 vndwrite(dev_t dev, struct uio *uio, int flags)
864 {
865 	int unit = vndunit(dev);
866 	struct vnd_softc *sc;
867 
868 #ifdef DEBUG
869 	if (vnddebug & VDB_FOLLOW)
870 		printf("vndwrite(0x%x, %p)\n", dev, uio);
871 #endif
872 
873 	sc = device_lookup(&vnd_cd, unit);
874 	if (sc == NULL)
875 		return ENXIO;
876 
877 	if ((sc->sc_flags & VNF_INITED) == 0)
878 		return (ENXIO);
879 
880 	return (physio(vndstrategy, NULL, dev, B_WRITE, minphys, uio));
881 }
882 
883 static int
884 vnd_cget(struct lwp *l, int unit, int *un, struct vattr *va)
885 {
886 	struct vnd_softc *vnd;
887 
888 	if (*un == -1)
889 		*un = unit;
890 	if (*un < 0)
891 		return EINVAL;
892 
893 	vnd = device_lookup(&vnd_cd, *un);
894 	if (vnd == NULL)
895 		return (*un >= vnd_cd.cd_ndevs) ? ENXIO : -1;
896 
897 	if ((vnd->sc_flags & VNF_INITED) == 0)
898 		return -1;
899 
900 	return VOP_GETATTR(vnd->sc_vp, va, l->l_cred, l);
901 }
902 
903 /* ARGSUSED */
904 static int
905 vndioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l)
906 {
907 	int unit = vndunit(dev);
908 	struct vnd_softc *vnd;
909 	struct vnd_ioctl *vio;
910 	struct vattr vattr;
911 	struct nameidata nd;
912 	int error, part, pmask;
913 	size_t geomsize;
914 	int fflags;
915 #ifdef __HAVE_OLD_DISKLABEL
916 	struct disklabel newlabel;
917 #endif
918 
919 #ifdef DEBUG
920 	if (vnddebug & VDB_FOLLOW)
921 		printf("vndioctl(0x%x, 0x%lx, %p, 0x%x, %p): unit %d\n",
922 		    dev, cmd, data, flag, l->l_proc, unit);
923 #endif
924 	vnd = device_lookup(&vnd_cd, unit);
925 	if (vnd == NULL &&
926 #ifdef COMPAT_30
927 	    cmd != VNDIOOCGET &&
928 #endif
929 	    cmd != VNDIOCGET)
930 		return ENXIO;
931 	vio = (struct vnd_ioctl *)data;
932 
933 	/* Must be open for writes for these commands... */
934 	switch (cmd) {
935 	case VNDIOCSET:
936 	case VNDIOCCLR:
937 	case DIOCSDINFO:
938 	case DIOCWDINFO:
939 #ifdef __HAVE_OLD_DISKLABEL
940 	case ODIOCSDINFO:
941 	case ODIOCWDINFO:
942 #endif
943 	case DIOCKLABEL:
944 	case DIOCWLABEL:
945 		if ((flag & FWRITE) == 0)
946 			return (EBADF);
947 	}
948 
949 	/* Must be initialized for these... */
950 	switch (cmd) {
951 	case VNDIOCCLR:
952 	case DIOCGDINFO:
953 	case DIOCSDINFO:
954 	case DIOCWDINFO:
955 	case DIOCGPART:
956 	case DIOCKLABEL:
957 	case DIOCWLABEL:
958 	case DIOCGDEFLABEL:
959 #ifdef __HAVE_OLD_DISKLABEL
960 	case ODIOCGDINFO:
961 	case ODIOCSDINFO:
962 	case ODIOCWDINFO:
963 	case ODIOCGDEFLABEL:
964 #endif
965 		if ((vnd->sc_flags & VNF_INITED) == 0)
966 			return (ENXIO);
967 	}
968 
969 	switch (cmd) {
970 	case VNDIOCSET:
971 		if (vnd->sc_flags & VNF_INITED)
972 			return (EBUSY);
973 
974 		if ((error = vndlock(vnd)) != 0)
975 			return (error);
976 
977 		fflags = FREAD;
978 		if ((vio->vnd_flags & VNDIOF_READONLY) == 0)
979 			fflags |= FWRITE;
980 		NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, vio->vnd_file, l);
981 		if ((error = vn_open(&nd, fflags, 0)) != 0)
982 			goto unlock_and_exit;
983 		KASSERT(l);
984 		error = VOP_GETATTR(nd.ni_vp, &vattr, l->l_cred, l);
985 		if (!error && nd.ni_vp->v_type != VREG)
986 			error = EOPNOTSUPP;
987 		if (error) {
988 			VOP_UNLOCK(nd.ni_vp, 0);
989 			goto close_and_exit;
990 		}
991 
992 		/* If using a compressed file, initialize its info */
993 		/* (or abort with an error if kernel has no compression) */
994 		if (vio->vnd_flags & VNF_COMP) {
995 #ifdef VND_COMPRESSION
996 			struct vnd_comp_header *ch;
997 			int i;
998 			u_int32_t comp_size;
999 			u_int32_t comp_maxsize;
1000 
1001 			/* allocate space for compresed file header */
1002 			ch = malloc(sizeof(struct vnd_comp_header),
1003 			M_TEMP, M_WAITOK);
1004 
1005 			/* read compressed file header */
1006 			error = vn_rdwr(UIO_READ, nd.ni_vp, (void *)ch,
1007 			  sizeof(struct vnd_comp_header), 0, UIO_SYSSPACE,
1008 			  IO_UNIT|IO_NODELOCKED, l->l_cred, NULL, NULL);
1009 			if(error) {
1010 				free(ch, M_TEMP);
1011 				VOP_UNLOCK(nd.ni_vp, 0);
1012 				goto close_and_exit;
1013 			}
1014 
1015 			/* save some header info */
1016 			vnd->sc_comp_blksz = ntohl(ch->block_size);
1017 			/* note last offset is the file byte size */
1018 			vnd->sc_comp_numoffs = ntohl(ch->num_blocks)+1;
1019 			free(ch, M_TEMP);
1020 			if (vnd->sc_comp_blksz == 0 ||
1021 			    vnd->sc_comp_blksz % DEV_BSIZE !=0) {
1022 				VOP_UNLOCK(nd.ni_vp, 0);
1023 				error = EINVAL;
1024 				goto close_and_exit;
1025 			}
1026 			if(sizeof(struct vnd_comp_header) +
1027 			  sizeof(u_int64_t) * vnd->sc_comp_numoffs >
1028 			  vattr.va_size) {
1029 				VOP_UNLOCK(nd.ni_vp, 0);
1030 				error = EINVAL;
1031 				goto close_and_exit;
1032 			}
1033 
1034 			/* set decompressed file size */
1035 			vattr.va_size =
1036 			    ((u_quad_t)vnd->sc_comp_numoffs - 1) *
1037 			     (u_quad_t)vnd->sc_comp_blksz;
1038 
1039 			/* allocate space for all the compressed offsets */
1040 			vnd->sc_comp_offsets =
1041 			malloc(sizeof(u_int64_t) * vnd->sc_comp_numoffs,
1042 			M_DEVBUF, M_WAITOK);
1043 
1044 			/* read in the offsets */
1045 			error = vn_rdwr(UIO_READ, nd.ni_vp,
1046 			  (void *)vnd->sc_comp_offsets,
1047 			  sizeof(u_int64_t) * vnd->sc_comp_numoffs,
1048 			  sizeof(struct vnd_comp_header), UIO_SYSSPACE,
1049 			  IO_UNIT|IO_NODELOCKED, l->l_cred, NULL, NULL);
1050 			if(error) {
1051 				VOP_UNLOCK(nd.ni_vp, 0);
1052 				goto close_and_exit;
1053 			}
1054 			/*
1055 			 * find largest block size (used for allocation limit).
1056 			 * Also convert offset to native byte order.
1057 			 */
1058 			comp_maxsize = 0;
1059 			for (i = 0; i < vnd->sc_comp_numoffs - 1; i++) {
1060 				vnd->sc_comp_offsets[i] =
1061 				  be64toh(vnd->sc_comp_offsets[i]);
1062 				comp_size = be64toh(vnd->sc_comp_offsets[i + 1])
1063 				  - vnd->sc_comp_offsets[i];
1064 				if (comp_size > comp_maxsize)
1065 					comp_maxsize = comp_size;
1066 			}
1067 			vnd->sc_comp_offsets[vnd->sc_comp_numoffs - 1] =
1068 			  be64toh(vnd->sc_comp_offsets[vnd->sc_comp_numoffs - 1]);
1069 
1070 			/* create compressed data buffer */
1071 			vnd->sc_comp_buff = malloc(comp_maxsize,
1072 			  M_DEVBUF, M_WAITOK);
1073 
1074 			/* create decompressed buffer */
1075 			vnd->sc_comp_decombuf = malloc(vnd->sc_comp_blksz,
1076 			  M_DEVBUF, M_WAITOK);
1077 			vnd->sc_comp_buffblk = -1;
1078 
1079 			/* Initialize decompress stream */
1080 			bzero(&vnd->sc_comp_stream, sizeof(z_stream));
1081 			vnd->sc_comp_stream.zalloc = vnd_alloc;
1082 			vnd->sc_comp_stream.zfree = vnd_free;
1083 			error = inflateInit2(&vnd->sc_comp_stream, MAX_WBITS);
1084 			if(error) {
1085 				if(vnd->sc_comp_stream.msg)
1086 					printf("vnd%d: compressed file, %s\n",
1087 					  unit, vnd->sc_comp_stream.msg);
1088 				VOP_UNLOCK(nd.ni_vp, 0);
1089 				error = EINVAL;
1090 				goto close_and_exit;
1091 			}
1092 
1093 			vnd->sc_flags |= VNF_COMP | VNF_READONLY;
1094 #else /* !VND_COMPRESSION */
1095 			VOP_UNLOCK(nd.ni_vp, 0);
1096 			error = EOPNOTSUPP;
1097 			goto close_and_exit;
1098 #endif /* VND_COMPRESSION */
1099 		}
1100 
1101 		VOP_UNLOCK(nd.ni_vp, 0);
1102 		vnd->sc_vp = nd.ni_vp;
1103 		vnd->sc_size = btodb(vattr.va_size);	/* note truncation */
1104 
1105 		/*
1106 		 * Use pseudo-geometry specified.  If none was provided,
1107 		 * use "standard" Adaptec fictitious geometry.
1108 		 */
1109 		if (vio->vnd_flags & VNDIOF_HASGEOM) {
1110 
1111 			memcpy(&vnd->sc_geom, &vio->vnd_geom,
1112 			    sizeof(vio->vnd_geom));
1113 
1114 			/*
1115 			 * Sanity-check the sector size.
1116 			 * XXX Don't allow secsize < DEV_BSIZE.	 Should
1117 			 * XXX we?
1118 			 */
1119 			if (vnd->sc_geom.vng_secsize < DEV_BSIZE ||
1120 			    (vnd->sc_geom.vng_secsize % DEV_BSIZE) != 0 ||
1121 			    vnd->sc_geom.vng_ncylinders == 0 ||
1122 			    (vnd->sc_geom.vng_ntracks *
1123 			     vnd->sc_geom.vng_nsectors) == 0) {
1124 				error = EINVAL;
1125 				goto close_and_exit;
1126 			}
1127 
1128 			/*
1129 			 * Compute the size (in DEV_BSIZE blocks) specified
1130 			 * by the geometry.
1131 			 */
1132 			geomsize = (vnd->sc_geom.vng_nsectors *
1133 			    vnd->sc_geom.vng_ntracks *
1134 			    vnd->sc_geom.vng_ncylinders) *
1135 			    (vnd->sc_geom.vng_secsize / DEV_BSIZE);
1136 
1137 			/*
1138 			 * Sanity-check the size against the specified
1139 			 * geometry.
1140 			 */
1141 			if (vnd->sc_size < geomsize) {
1142 				error = EINVAL;
1143 				goto close_and_exit;
1144 			}
1145 		} else if (vnd->sc_size >= (32 * 64)) {
1146 			/*
1147 			 * Size must be at least 2048 DEV_BSIZE blocks
1148 			 * (1M) in order to use this geometry.
1149 			 */
1150 			vnd->sc_geom.vng_secsize = DEV_BSIZE;
1151 			vnd->sc_geom.vng_nsectors = 32;
1152 			vnd->sc_geom.vng_ntracks = 64;
1153 			vnd->sc_geom.vng_ncylinders = vnd->sc_size / (64 * 32);
1154 		} else {
1155 			vnd->sc_geom.vng_secsize = DEV_BSIZE;
1156 			vnd->sc_geom.vng_nsectors = 1;
1157 			vnd->sc_geom.vng_ntracks = 1;
1158 			vnd->sc_geom.vng_ncylinders = vnd->sc_size;
1159 		}
1160 
1161 		if (vio->vnd_flags & VNDIOF_READONLY) {
1162 			vnd->sc_flags |= VNF_READONLY;
1163 		}
1164 
1165 		if ((error = vndsetcred(vnd, l->l_cred)) != 0)
1166 			goto close_and_exit;
1167 
1168 		vndthrottle(vnd, vnd->sc_vp);
1169 		vio->vnd_size = dbtob(vnd->sc_size);
1170 		vnd->sc_flags |= VNF_INITED;
1171 
1172 		/* create the kernel thread, wait for it to be up */
1173 		error = kthread_create(PRI_NONE, 0, NULL, vndthread, vnd,
1174 		    &vnd->sc_kthread, vnd->sc_dev.dv_xname);
1175 		if (error)
1176 			goto close_and_exit;
1177 		while ((vnd->sc_flags & VNF_KTHREAD) == 0) {
1178 			tsleep(&vnd->sc_kthread, PRIBIO, "vndthr", 0);
1179 		}
1180 #ifdef DEBUG
1181 		if (vnddebug & VDB_INIT)
1182 			printf("vndioctl: SET vp %p size 0x%lx %d/%d/%d/%d\n",
1183 			    vnd->sc_vp, (unsigned long) vnd->sc_size,
1184 			    vnd->sc_geom.vng_secsize,
1185 			    vnd->sc_geom.vng_nsectors,
1186 			    vnd->sc_geom.vng_ntracks,
1187 			    vnd->sc_geom.vng_ncylinders);
1188 #endif
1189 
1190 		/* Attach the disk. */
1191 		disk_attach(&vnd->sc_dkdev);
1192 
1193 		/* Initialize the xfer and buffer pools. */
1194 		pool_init(&vnd->sc_vxpool, sizeof(struct vndxfer), 0,
1195 		    0, 0, "vndxpl", NULL, IPL_BIO);
1196 
1197 		/* Try and read the disklabel. */
1198 		vndgetdisklabel(dev, vnd);
1199 
1200 		vndunlock(vnd);
1201 
1202 		break;
1203 
1204 close_and_exit:
1205 		(void) vn_close(nd.ni_vp, fflags, l->l_cred, l);
1206 unlock_and_exit:
1207 #ifdef VND_COMPRESSION
1208 		/* free any allocated memory (for compressed file) */
1209 		if(vnd->sc_comp_offsets) {
1210 			free(vnd->sc_comp_offsets, M_DEVBUF);
1211 			vnd->sc_comp_offsets = NULL;
1212 		}
1213 		if(vnd->sc_comp_buff) {
1214 			free(vnd->sc_comp_buff, M_DEVBUF);
1215 			vnd->sc_comp_buff = NULL;
1216 		}
1217 		if(vnd->sc_comp_decombuf) {
1218 			free(vnd->sc_comp_decombuf, M_DEVBUF);
1219 			vnd->sc_comp_decombuf = NULL;
1220 		}
1221 #endif /* VND_COMPRESSION */
1222 		vndunlock(vnd);
1223 		return (error);
1224 
1225 	case VNDIOCCLR:
1226 		if ((error = vndlock(vnd)) != 0)
1227 			return (error);
1228 
1229 		/*
1230 		 * Don't unconfigure if any other partitions are open
1231 		 * or if both the character and block flavors of this
1232 		 * partition are open.
1233 		 */
1234 		part = DISKPART(dev);
1235 		pmask = (1 << part);
1236 		if (((vnd->sc_dkdev.dk_openmask & ~pmask) ||
1237 		    ((vnd->sc_dkdev.dk_bopenmask & pmask) &&
1238 		    (vnd->sc_dkdev.dk_copenmask & pmask))) &&
1239 			!(vio->vnd_flags & VNDIOF_FORCE)) {
1240 			vndunlock(vnd);
1241 			return (EBUSY);
1242 		}
1243 
1244 		/*
1245 		 * XXX vndclear() might call vndclose() implicitely;
1246 		 * release lock to avoid recursion
1247 		 */
1248 		vndunlock(vnd);
1249 		vndclear(vnd, minor(dev));
1250 #ifdef DEBUG
1251 		if (vnddebug & VDB_INIT)
1252 			printf("vndioctl: CLRed\n");
1253 #endif
1254 
1255 		/* Destroy the xfer and buffer pools. */
1256 		pool_destroy(&vnd->sc_vxpool);
1257 
1258 		/* Detatch the disk. */
1259 		disk_detach(&vnd->sc_dkdev);
1260 		break;
1261 
1262 #ifdef COMPAT_30
1263 	case VNDIOOCGET: {
1264 		struct vnd_ouser *vnu;
1265 		struct vattr va;
1266 		vnu = (struct vnd_ouser *)data;
1267 		KASSERT(l);
1268 		switch (error = vnd_cget(l, unit, &vnu->vnu_unit, &va)) {
1269 		case 0:
1270 			vnu->vnu_dev = va.va_fsid;
1271 			vnu->vnu_ino = va.va_fileid;
1272 			break;
1273 		case -1:
1274 			/* unused is not an error */
1275 			vnu->vnu_dev = 0;
1276 			vnu->vnu_ino = 0;
1277 			break;
1278 		default:
1279 			return error;
1280 		}
1281 		break;
1282 	}
1283 #endif
1284 	case VNDIOCGET: {
1285 		struct vnd_user *vnu;
1286 		struct vattr va;
1287 		vnu = (struct vnd_user *)data;
1288 		KASSERT(l);
1289 		switch (error = vnd_cget(l, unit, &vnu->vnu_unit, &va)) {
1290 		case 0:
1291 			vnu->vnu_dev = va.va_fsid;
1292 			vnu->vnu_ino = va.va_fileid;
1293 			break;
1294 		case -1:
1295 			/* unused is not an error */
1296 			vnu->vnu_dev = 0;
1297 			vnu->vnu_ino = 0;
1298 			break;
1299 		default:
1300 			return error;
1301 		}
1302 		break;
1303 	}
1304 
1305 	case DIOCGDINFO:
1306 		*(struct disklabel *)data = *(vnd->sc_dkdev.dk_label);
1307 		break;
1308 
1309 #ifdef __HAVE_OLD_DISKLABEL
1310 	case ODIOCGDINFO:
1311 		newlabel = *(vnd->sc_dkdev.dk_label);
1312 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1313 			return ENOTTY;
1314 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1315 		break;
1316 #endif
1317 
1318 	case DIOCGPART:
1319 		((struct partinfo *)data)->disklab = vnd->sc_dkdev.dk_label;
1320 		((struct partinfo *)data)->part =
1321 		    &vnd->sc_dkdev.dk_label->d_partitions[DISKPART(dev)];
1322 		break;
1323 
1324 	case DIOCWDINFO:
1325 	case DIOCSDINFO:
1326 #ifdef __HAVE_OLD_DISKLABEL
1327 	case ODIOCWDINFO:
1328 	case ODIOCSDINFO:
1329 #endif
1330 	{
1331 		struct disklabel *lp;
1332 
1333 		if ((error = vndlock(vnd)) != 0)
1334 			return (error);
1335 
1336 		vnd->sc_flags |= VNF_LABELLING;
1337 
1338 #ifdef __HAVE_OLD_DISKLABEL
1339 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
1340 			memset(&newlabel, 0, sizeof newlabel);
1341 			memcpy(&newlabel, data, sizeof (struct olddisklabel));
1342 			lp = &newlabel;
1343 		} else
1344 #endif
1345 		lp = (struct disklabel *)data;
1346 
1347 		error = setdisklabel(vnd->sc_dkdev.dk_label,
1348 		    lp, 0, vnd->sc_dkdev.dk_cpulabel);
1349 		if (error == 0) {
1350 			if (cmd == DIOCWDINFO
1351 #ifdef __HAVE_OLD_DISKLABEL
1352 			    || cmd == ODIOCWDINFO
1353 #endif
1354 			   )
1355 				error = writedisklabel(VNDLABELDEV(dev),
1356 				    vndstrategy, vnd->sc_dkdev.dk_label,
1357 				    vnd->sc_dkdev.dk_cpulabel);
1358 		}
1359 
1360 		vnd->sc_flags &= ~VNF_LABELLING;
1361 
1362 		vndunlock(vnd);
1363 
1364 		if (error)
1365 			return (error);
1366 		break;
1367 	}
1368 
1369 	case DIOCKLABEL:
1370 		if (*(int *)data != 0)
1371 			vnd->sc_flags |= VNF_KLABEL;
1372 		else
1373 			vnd->sc_flags &= ~VNF_KLABEL;
1374 		break;
1375 
1376 	case DIOCWLABEL:
1377 		if (*(int *)data != 0)
1378 			vnd->sc_flags |= VNF_WLABEL;
1379 		else
1380 			vnd->sc_flags &= ~VNF_WLABEL;
1381 		break;
1382 
1383 	case DIOCGDEFLABEL:
1384 		vndgetdefaultlabel(vnd, (struct disklabel *)data);
1385 		break;
1386 
1387 #ifdef __HAVE_OLD_DISKLABEL
1388 	case ODIOCGDEFLABEL:
1389 		vndgetdefaultlabel(vnd, &newlabel);
1390 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1391 			return ENOTTY;
1392 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1393 		break;
1394 #endif
1395 
1396 	default:
1397 		return (ENOTTY);
1398 	}
1399 
1400 	return (0);
1401 }
1402 
1403 /*
1404  * Duplicate the current processes' credentials.  Since we are called only
1405  * as the result of a SET ioctl and only root can do that, any future access
1406  * to this "disk" is essentially as root.  Note that credentials may change
1407  * if some other uid can write directly to the mapped file (NFS).
1408  */
1409 static int
1410 vndsetcred(struct vnd_softc *vnd, kauth_cred_t cred)
1411 {
1412 	struct uio auio;
1413 	struct iovec aiov;
1414 	char *tmpbuf;
1415 	int error;
1416 
1417 	vnd->sc_cred = kauth_cred_dup(cred);
1418 	tmpbuf = malloc(DEV_BSIZE, M_TEMP, M_WAITOK);
1419 
1420 	/* XXX: Horrible kludge to establish credentials for NFS */
1421 	aiov.iov_base = tmpbuf;
1422 	aiov.iov_len = min(DEV_BSIZE, dbtob(vnd->sc_size));
1423 	auio.uio_iov = &aiov;
1424 	auio.uio_iovcnt = 1;
1425 	auio.uio_offset = 0;
1426 	auio.uio_rw = UIO_READ;
1427 	auio.uio_resid = aiov.iov_len;
1428 	UIO_SETUP_SYSSPACE(&auio);
1429 	vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1430 	error = VOP_READ(vnd->sc_vp, &auio, 0, vnd->sc_cred);
1431 	if (error == 0) {
1432 		/*
1433 		 * Because vnd does all IO directly through the vnode
1434 		 * we need to flush (at least) the buffer from the above
1435 		 * VOP_READ from the buffer cache to prevent cache
1436 		 * incoherencies.  Also, be careful to write dirty
1437 		 * buffers back to stable storage.
1438 		 */
1439 		error = vinvalbuf(vnd->sc_vp, V_SAVE, vnd->sc_cred,
1440 			    curlwp, 0, 0);
1441 	}
1442 	VOP_UNLOCK(vnd->sc_vp, 0);
1443 
1444 	free(tmpbuf, M_TEMP);
1445 	return (error);
1446 }
1447 
1448 /*
1449  * Set maxactive based on FS type
1450  */
1451 static void
1452 vndthrottle(struct vnd_softc *vnd, struct vnode *vp)
1453 {
1454 #ifdef NFS
1455 	extern int (**nfsv2_vnodeop_p)(void *);
1456 
1457 	if (vp->v_op == nfsv2_vnodeop_p)
1458 		vnd->sc_maxactive = 2;
1459 	else
1460 #endif
1461 		vnd->sc_maxactive = 8;
1462 
1463 	if (vnd->sc_maxactive < 1)
1464 		vnd->sc_maxactive = 1;
1465 }
1466 
1467 #if 0
1468 static void
1469 vndshutdown(void)
1470 {
1471 	struct vnd_softc *vnd;
1472 
1473 	for (vnd = &vnd_softc[0]; vnd < &vnd_softc[numvnd]; vnd++)
1474 		if (vnd->sc_flags & VNF_INITED)
1475 			vndclear(vnd);
1476 }
1477 #endif
1478 
1479 static void
1480 vndclear(struct vnd_softc *vnd, int myminor)
1481 {
1482 	struct vnode *vp = vnd->sc_vp;
1483 	struct lwp *l = curlwp;
1484 	int fflags = FREAD;
1485 	int bmaj, cmaj, i, mn;
1486 	int s;
1487 
1488 #ifdef DEBUG
1489 	if (vnddebug & VDB_FOLLOW)
1490 		printf("vndclear(%p): vp %p\n", vnd, vp);
1491 #endif
1492 	/* locate the major number */
1493 	bmaj = bdevsw_lookup_major(&vnd_bdevsw);
1494 	cmaj = cdevsw_lookup_major(&vnd_cdevsw);
1495 
1496 	/* Nuke the vnodes for any open instances */
1497 	for (i = 0; i < MAXPARTITIONS; i++) {
1498 		mn = DISKMINOR(device_unit(&vnd->sc_dev), i);
1499 		vdevgone(bmaj, mn, mn, VBLK);
1500 		if (mn != myminor) /* XXX avoid to kill own vnode */
1501 			vdevgone(cmaj, mn, mn, VCHR);
1502 	}
1503 
1504 	if ((vnd->sc_flags & VNF_READONLY) == 0)
1505 		fflags |= FWRITE;
1506 
1507 	s = splbio();
1508 	bufq_drain(vnd->sc_tab);
1509 	splx(s);
1510 
1511 	vnd->sc_flags |= VNF_VUNCONF;
1512 	wakeup(&vnd->sc_tab);
1513 	while (vnd->sc_flags & VNF_KTHREAD)
1514 		tsleep(&vnd->sc_kthread, PRIBIO, "vnthr", 0);
1515 
1516 #ifdef VND_COMPRESSION
1517 	/* free the compressed file buffers */
1518 	if(vnd->sc_flags & VNF_COMP) {
1519 		if(vnd->sc_comp_offsets) {
1520 			free(vnd->sc_comp_offsets, M_DEVBUF);
1521 			vnd->sc_comp_offsets = NULL;
1522 		}
1523 		if(vnd->sc_comp_buff) {
1524 			free(vnd->sc_comp_buff, M_DEVBUF);
1525 			vnd->sc_comp_buff = NULL;
1526 		}
1527 		if(vnd->sc_comp_decombuf) {
1528 			free(vnd->sc_comp_decombuf, M_DEVBUF);
1529 			vnd->sc_comp_decombuf = NULL;
1530 		}
1531 	}
1532 #endif /* VND_COMPRESSION */
1533 	vnd->sc_flags &=
1534 	    ~(VNF_INITED | VNF_READONLY | VNF_VLABEL
1535 	      | VNF_VUNCONF | VNF_COMP);
1536 	if (vp == (struct vnode *)0)
1537 		panic("vndclear: null vp");
1538 	(void) vn_close(vp, fflags, vnd->sc_cred, l);
1539 	kauth_cred_free(vnd->sc_cred);
1540 	vnd->sc_vp = (struct vnode *)0;
1541 	vnd->sc_cred = (kauth_cred_t)0;
1542 	vnd->sc_size = 0;
1543 }
1544 
1545 static int
1546 vndsize(dev_t dev)
1547 {
1548 	struct vnd_softc *sc;
1549 	struct disklabel *lp;
1550 	int part, unit, omask;
1551 	int size;
1552 
1553 	unit = vndunit(dev);
1554 	sc = (struct vnd_softc *)device_lookup(&vnd_cd, unit);
1555 	if (sc == NULL)
1556 		return -1;
1557 
1558 	if ((sc->sc_flags & VNF_INITED) == 0)
1559 		return (-1);
1560 
1561 	part = DISKPART(dev);
1562 	omask = sc->sc_dkdev.dk_openmask & (1 << part);
1563 	lp = sc->sc_dkdev.dk_label;
1564 
1565 	if (omask == 0 && vndopen(dev, 0, S_IFBLK, curlwp))	/* XXX */
1566 		return (-1);
1567 
1568 	if (lp->d_partitions[part].p_fstype != FS_SWAP)
1569 		size = -1;
1570 	else
1571 		size = lp->d_partitions[part].p_size *
1572 		    (lp->d_secsize / DEV_BSIZE);
1573 
1574 	if (omask == 0 && vndclose(dev, 0, S_IFBLK, curlwp))	/* XXX */
1575 		return (-1);
1576 
1577 	return (size);
1578 }
1579 
1580 static int
1581 vnddump(dev_t dev, daddr_t blkno, void *va,
1582     size_t size)
1583 {
1584 
1585 	/* Not implemented. */
1586 	return ENXIO;
1587 }
1588 
1589 static void
1590 vndgetdefaultlabel(struct vnd_softc *sc, struct disklabel *lp)
1591 {
1592 	struct vndgeom *vng = &sc->sc_geom;
1593 	struct partition *pp;
1594 
1595 	memset(lp, 0, sizeof(*lp));
1596 
1597 	lp->d_secperunit = sc->sc_size / (vng->vng_secsize / DEV_BSIZE);
1598 	lp->d_secsize = vng->vng_secsize;
1599 	lp->d_nsectors = vng->vng_nsectors;
1600 	lp->d_ntracks = vng->vng_ntracks;
1601 	lp->d_ncylinders = vng->vng_ncylinders;
1602 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1603 
1604 	strncpy(lp->d_typename, "vnd", sizeof(lp->d_typename));
1605 	lp->d_type = DTYPE_VND;
1606 	strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
1607 	lp->d_rpm = 3600;
1608 	lp->d_interleave = 1;
1609 	lp->d_flags = 0;
1610 
1611 	pp = &lp->d_partitions[RAW_PART];
1612 	pp->p_offset = 0;
1613 	pp->p_size = lp->d_secperunit;
1614 	pp->p_fstype = FS_UNUSED;
1615 	lp->d_npartitions = RAW_PART + 1;
1616 
1617 	lp->d_magic = DISKMAGIC;
1618 	lp->d_magic2 = DISKMAGIC;
1619 	lp->d_checksum = dkcksum(lp);
1620 }
1621 
1622 /*
1623  * Read the disklabel from a vnd.  If one is not present, create a fake one.
1624  */
1625 static void
1626 vndgetdisklabel(dev_t dev, struct vnd_softc *sc)
1627 {
1628 	const char *errstring;
1629 	struct disklabel *lp = sc->sc_dkdev.dk_label;
1630 	struct cpu_disklabel *clp = sc->sc_dkdev.dk_cpulabel;
1631 	int i;
1632 
1633 	memset(clp, 0, sizeof(*clp));
1634 
1635 	vndgetdefaultlabel(sc, lp);
1636 
1637 	/*
1638 	 * Call the generic disklabel extraction routine.
1639 	 */
1640 	errstring = readdisklabel(VNDLABELDEV(dev), vndstrategy, lp, clp);
1641 	if (errstring) {
1642 		/*
1643 		 * Lack of disklabel is common, but we print the warning
1644 		 * anyway, since it might contain other useful information.
1645 		 */
1646 		printf("%s: %s\n", sc->sc_dev.dv_xname, errstring);
1647 
1648 		/*
1649 		 * For historical reasons, if there's no disklabel
1650 		 * present, all partitions must be FS_BSDFFS and
1651 		 * occupy the entire disk.
1652 		 */
1653 		for (i = 0; i < MAXPARTITIONS; i++) {
1654 			/*
1655 			 * Don't wipe out port specific hack (such as
1656 			 * dos partition hack of i386 port).
1657 			 */
1658 			if (lp->d_partitions[i].p_size != 0)
1659 				continue;
1660 
1661 			lp->d_partitions[i].p_size = lp->d_secperunit;
1662 			lp->d_partitions[i].p_offset = 0;
1663 			lp->d_partitions[i].p_fstype = FS_BSDFFS;
1664 		}
1665 
1666 		strncpy(lp->d_packname, "default label",
1667 		    sizeof(lp->d_packname));
1668 
1669 		lp->d_npartitions = MAXPARTITIONS;
1670 		lp->d_checksum = dkcksum(lp);
1671 	}
1672 
1673 	/* In-core label now valid. */
1674 	sc->sc_flags |= VNF_VLABEL;
1675 }
1676 
1677 /*
1678  * Wait interruptibly for an exclusive lock.
1679  *
1680  * XXX
1681  * Several drivers do this; it should be abstracted and made MP-safe.
1682  */
1683 static int
1684 vndlock(struct vnd_softc *sc)
1685 {
1686 	int error;
1687 
1688 	while ((sc->sc_flags & VNF_LOCKED) != 0) {
1689 		sc->sc_flags |= VNF_WANTED;
1690 		if ((error = tsleep(sc, PRIBIO | PCATCH, "vndlck", 0)) != 0)
1691 			return (error);
1692 	}
1693 	sc->sc_flags |= VNF_LOCKED;
1694 	return (0);
1695 }
1696 
1697 /*
1698  * Unlock and wake up any waiters.
1699  */
1700 static void
1701 vndunlock(struct vnd_softc *sc)
1702 {
1703 
1704 	sc->sc_flags &= ~VNF_LOCKED;
1705 	if ((sc->sc_flags & VNF_WANTED) != 0) {
1706 		sc->sc_flags &= ~VNF_WANTED;
1707 		wakeup(sc);
1708 	}
1709 }
1710 
1711 #ifdef VND_COMPRESSION
1712 /* compressed file read */
1713 static void
1714 compstrategy(struct buf *bp, off_t bn)
1715 {
1716 	int error;
1717 	int unit = vndunit(bp->b_dev);
1718 	struct vnd_softc *vnd =
1719 	    (struct vnd_softc *)device_lookup(&vnd_cd, unit);
1720 	u_int32_t comp_block;
1721 	struct uio auio;
1722 	char *addr;
1723 	int s;
1724 
1725 	/* set up constants for data move */
1726 	auio.uio_rw = UIO_READ;
1727 	UIO_SETUP_SYSSPACE(&auio);
1728 
1729 	/* read, and transfer the data */
1730 	addr = bp->b_data;
1731 	bp->b_resid = bp->b_bcount;
1732 	s = splbio();
1733 	while (bp->b_resid > 0) {
1734 		unsigned length;
1735 		size_t length_in_buffer;
1736 		u_int32_t offset_in_buffer;
1737 		struct iovec aiov;
1738 
1739 		/* calculate the compressed block number */
1740 		comp_block = bn / (off_t)vnd->sc_comp_blksz;
1741 
1742 		/* check for good block number */
1743 		if (comp_block >= vnd->sc_comp_numoffs) {
1744 			bp->b_error = EINVAL;
1745 			splx(s);
1746 			return;
1747 		}
1748 
1749 		/* read in the compressed block, if not in buffer */
1750 		if (comp_block != vnd->sc_comp_buffblk) {
1751 			length = vnd->sc_comp_offsets[comp_block + 1] -
1752 			    vnd->sc_comp_offsets[comp_block];
1753 			vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1754 			error = vn_rdwr(UIO_READ, vnd->sc_vp, vnd->sc_comp_buff,
1755 			    length, vnd->sc_comp_offsets[comp_block],
1756 			    UIO_SYSSPACE, IO_UNIT, vnd->sc_cred, NULL, NULL);
1757 			if (error) {
1758 				bp->b_error = error;
1759 				VOP_UNLOCK(vnd->sc_vp, 0);
1760 				splx(s);
1761 				return;
1762 			}
1763 			/* uncompress the buffer */
1764 			vnd->sc_comp_stream.next_in = vnd->sc_comp_buff;
1765 			vnd->sc_comp_stream.avail_in = length;
1766 			vnd->sc_comp_stream.next_out = vnd->sc_comp_decombuf;
1767 			vnd->sc_comp_stream.avail_out = vnd->sc_comp_blksz;
1768 			inflateReset(&vnd->sc_comp_stream);
1769 			error = inflate(&vnd->sc_comp_stream, Z_FINISH);
1770 			if (error != Z_STREAM_END) {
1771 				if (vnd->sc_comp_stream.msg)
1772 					printf("%s: compressed file, %s\n",
1773 					    vnd->sc_dev.dv_xname,
1774 					    vnd->sc_comp_stream.msg);
1775 				bp->b_error = EBADMSG;
1776 				VOP_UNLOCK(vnd->sc_vp, 0);
1777 				splx(s);
1778 				return;
1779 			}
1780 			vnd->sc_comp_buffblk = comp_block;
1781 			VOP_UNLOCK(vnd->sc_vp, 0);
1782 		}
1783 
1784 		/* transfer the usable uncompressed data */
1785 		offset_in_buffer = bn % (off_t)vnd->sc_comp_blksz;
1786 		length_in_buffer = vnd->sc_comp_blksz - offset_in_buffer;
1787 		if (length_in_buffer > bp->b_resid)
1788 			length_in_buffer = bp->b_resid;
1789 		auio.uio_iov = &aiov;
1790 		auio.uio_iovcnt = 1;
1791 		aiov.iov_base = addr;
1792 		aiov.iov_len = length_in_buffer;
1793 		auio.uio_resid = aiov.iov_len;
1794 		auio.uio_offset = 0;
1795 		error = uiomove(vnd->sc_comp_decombuf + offset_in_buffer,
1796 		    length_in_buffer, &auio);
1797 		if (error) {
1798 			bp->b_error = error;
1799 			splx(s);
1800 			return;
1801 		}
1802 
1803 		bn += length_in_buffer;
1804 		addr += length_in_buffer;
1805 		bp->b_resid -= length_in_buffer;
1806 	}
1807 	splx(s);
1808 }
1809 
1810 /* compression memory allocation routines */
1811 static void *
1812 vnd_alloc(void *aux, u_int items, u_int siz)
1813 {
1814 	return malloc(items * siz, M_TEMP, M_NOWAIT);
1815 }
1816 
1817 static void
1818 vnd_free(void *aux, void *ptr)
1819 {
1820 	free(ptr, M_TEMP);
1821 }
1822 #endif /* VND_COMPRESSION */
1823