xref: /netbsd-src/sys/dev/vnd.c (revision aaf4ece63a859a04e37cf3a7229b5fab0157cc06)
1 /*	$NetBSD: vnd.c,v 1.129 2006/01/11 00:50:29 yamt 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 1.129 2006/01/11 00:50:29 yamt 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 <net/zlib.h>
163 
164 #include <miscfs/specfs/specdev.h>
165 
166 #include <dev/vndvar.h>
167 
168 #if defined(VNDDEBUG) && !defined(DEBUG)
169 #define DEBUG
170 #endif
171 
172 #ifdef DEBUG
173 int dovndcluster = 1;
174 #define VDB_FOLLOW	0x01
175 #define VDB_INIT	0x02
176 #define VDB_IO		0x04
177 #define VDB_LABEL	0x08
178 int vnddebug = 0x00;
179 #endif
180 
181 #define vndunit(x)	DISKUNIT(x)
182 
183 struct vndxfer {
184 	struct buf vx_buf;
185 	struct vnd_softc *vx_vnd;
186 };
187 #define	VND_BUFTOXFER(bp)	((struct vndxfer *)(void *)bp)
188 
189 #define VND_GETXFER(vnd)	pool_get(&(vnd)->sc_vxpool, PR_WAITOK)
190 #define VND_PUTXFER(vnd, vx)	pool_put(&(vnd)->sc_vxpool, (vx))
191 
192 struct vnd_softc *vnd_softc;
193 int numvnd = 0;
194 
195 #define VNDLABELDEV(dev) \
196     (MAKEDISKDEV(major((dev)), vndunit((dev)), RAW_PART))
197 
198 /* called by main() at boot time (XXX: and the LKM driver) */
199 void	vndattach(int);
200 int	vnddetach(void);
201 
202 static void	vndclear(struct vnd_softc *, int);
203 static int	vndsetcred(struct vnd_softc *, struct ucred *);
204 static void	vndthrottle(struct vnd_softc *, struct vnode *);
205 static void	vndiodone(struct buf *);
206 #if 0
207 static void	vndshutdown(void);
208 #endif
209 
210 static void	vndgetdefaultlabel(struct vnd_softc *, struct disklabel *);
211 static void	vndgetdisklabel(dev_t);
212 
213 static int	vndlock(struct vnd_softc *);
214 static void	vndunlock(struct vnd_softc *);
215 #ifdef VND_COMPRESSION
216 static void	compstrategy(struct buf *, off_t);
217 static void	*vnd_alloc(void *, u_int, u_int);
218 static void	vnd_free(void *, void *);
219 #endif /* VND_COMPRESSION */
220 
221 static void vndthread(void *);
222 
223 static dev_type_open(vndopen);
224 static dev_type_close(vndclose);
225 static dev_type_read(vndread);
226 static dev_type_write(vndwrite);
227 static dev_type_ioctl(vndioctl);
228 static dev_type_strategy(vndstrategy);
229 static dev_type_dump(vnddump);
230 static dev_type_size(vndsize);
231 
232 const struct bdevsw vnd_bdevsw = {
233 	vndopen, vndclose, vndstrategy, vndioctl, vnddump, vndsize, D_DISK
234 };
235 
236 const struct cdevsw vnd_cdevsw = {
237 	vndopen, vndclose, vndread, vndwrite, vndioctl,
238 	nostop, notty, nopoll, nommap, nokqfilter, D_DISK
239 };
240 
241 static int vndattached;
242 
243 void
244 vndattach(int num)
245 {
246 	int i;
247 	char *mem;
248 
249 	if (vndattached)
250 		return;
251 	vndattached = 1;
252 	if (num <= 0)
253 		return;
254 	i = num * sizeof(struct vnd_softc);
255 	mem = malloc(i, M_DEVBUF, M_NOWAIT|M_ZERO);
256 	if (mem == NULL) {
257 		printf("WARNING: no memory for vnode disks\n");
258 		return;
259 	}
260 	vnd_softc = (struct vnd_softc *)mem;
261 	numvnd = num;
262 
263 	for (i = 0; i < numvnd; i++) {
264 		vnd_softc[i].sc_unit = i;
265 		vnd_softc[i].sc_comp_offsets = NULL;
266 		vnd_softc[i].sc_comp_buff = NULL;
267 		vnd_softc[i].sc_comp_decombuf = NULL;
268 		bufq_alloc(&vnd_softc[i].sc_tab,
269 		    "disksort", BUFQ_SORT_RAWBLOCK);
270 		pseudo_disk_init(&vnd_softc[i].sc_dkdev);
271 	}
272 }
273 
274 int
275 vnddetach(void)
276 {
277 	int i;
278 
279 	/* First check we aren't in use. */
280 	for (i = 0; i < numvnd; i++)
281 		if (vnd_softc[i].sc_flags & VNF_INITED)
282 			return (EBUSY);
283 
284 	for (i = 0; i < numvnd; i++)
285 		bufq_free(vnd_softc[i].sc_tab);
286 
287 	free(vnd_softc, M_DEVBUF);
288 	vndattached = 0;
289 
290 	return (0);
291 }
292 
293 static int
294 vndopen(dev_t dev, int flags, int mode, struct lwp *l)
295 {
296 	int unit = vndunit(dev);
297 	struct vnd_softc *sc;
298 	int error = 0, part, pmask;
299 	struct disklabel *lp;
300 
301 #ifdef DEBUG
302 	if (vnddebug & VDB_FOLLOW)
303 		printf("vndopen(0x%x, 0x%x, 0x%x, %p)\n", dev, flags, mode, l);
304 #endif
305 	if (unit >= numvnd)
306 		return (ENXIO);
307 	sc = &vnd_softc[unit];
308 
309 	if ((error = vndlock(sc)) != 0)
310 		return (error);
311 
312 	lp = sc->sc_dkdev.dk_label;
313 
314 	part = DISKPART(dev);
315 	pmask = (1 << part);
316 
317 	/*
318 	 * If we're initialized, check to see if there are any other
319 	 * open partitions.  If not, then it's safe to update the
320 	 * in-core disklabel.  Only read the disklabel if it is
321 	 * not already valid.
322 	 */
323 	if ((sc->sc_flags & (VNF_INITED|VNF_VLABEL)) == VNF_INITED &&
324 	    sc->sc_dkdev.dk_openmask == 0)
325 		vndgetdisklabel(dev);
326 
327 	/* Check that the partitions exists. */
328 	if (part != RAW_PART) {
329 		if (((sc->sc_flags & VNF_INITED) == 0) ||
330 		    ((part >= lp->d_npartitions) ||
331 		     (lp->d_partitions[part].p_fstype == FS_UNUSED))) {
332 			error = ENXIO;
333 			goto done;
334 		}
335 	}
336 
337 	/* Prevent our unit from being unconfigured while open. */
338 	switch (mode) {
339 	case S_IFCHR:
340 		sc->sc_dkdev.dk_copenmask |= pmask;
341 		break;
342 
343 	case S_IFBLK:
344 		sc->sc_dkdev.dk_bopenmask |= pmask;
345 		break;
346 	}
347 	sc->sc_dkdev.dk_openmask =
348 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
349 
350  done:
351 	vndunlock(sc);
352 	return (error);
353 }
354 
355 static int
356 vndclose(dev_t dev, int flags, int mode, struct lwp *l)
357 {
358 	int unit = vndunit(dev);
359 	struct vnd_softc *sc;
360 	int error = 0, part;
361 
362 #ifdef DEBUG
363 	if (vnddebug & VDB_FOLLOW)
364 		printf("vndclose(0x%x, 0x%x, 0x%x, %p)\n", dev, flags, mode, l);
365 #endif
366 
367 	if (unit >= numvnd)
368 		return (ENXIO);
369 	sc = &vnd_softc[unit];
370 
371 	if ((error = vndlock(sc)) != 0)
372 		return (error);
373 
374 	part = DISKPART(dev);
375 
376 	/* ...that much closer to allowing unconfiguration... */
377 	switch (mode) {
378 	case S_IFCHR:
379 		sc->sc_dkdev.dk_copenmask &= ~(1 << part);
380 		break;
381 
382 	case S_IFBLK:
383 		sc->sc_dkdev.dk_bopenmask &= ~(1 << part);
384 		break;
385 	}
386 	sc->sc_dkdev.dk_openmask =
387 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
388 
389 	vndunlock(sc);
390 	return (0);
391 }
392 
393 /*
394  * Qeue the request, and wakeup the kernel thread to handle it.
395  */
396 static void
397 vndstrategy(struct buf *bp)
398 {
399 	int unit = vndunit(bp->b_dev);
400 	struct vnd_softc *vnd = &vnd_softc[unit];
401 	struct disklabel *lp = vnd->sc_dkdev.dk_label;
402 	daddr_t blkno;
403 	int s = splbio();
404 
405 	bp->b_resid = bp->b_bcount;
406 
407 	if ((vnd->sc_flags & VNF_INITED) == 0) {
408 		bp->b_error = ENXIO;
409 		bp->b_flags |= B_ERROR;
410 		goto done;
411 	}
412 
413 	/*
414 	 * The transfer must be a whole number of blocks.
415 	 */
416 	if ((bp->b_bcount % lp->d_secsize) != 0) {
417 		bp->b_error = EINVAL;
418 		bp->b_flags |= B_ERROR;
419 		goto done;
420 	}
421 
422 	/*
423 	 * check if we're read-only.
424 	 */
425 	if ((vnd->sc_flags & VNF_READONLY) && !(bp->b_flags & B_READ)) {
426 		bp->b_error = EACCES;
427 		bp->b_flags |= B_ERROR;
428 		goto done;
429 	}
430 
431 	/*
432 	 * Do bounds checking and adjust transfer.  If there's an error,
433 	 * the bounds check will flag that for us.
434 	 */
435 	if (DISKPART(bp->b_dev) != RAW_PART) {
436 		if (bounds_check_with_label(&vnd->sc_dkdev,
437 		    bp, vnd->sc_flags & (VNF_WLABEL|VNF_LABELLING)) <= 0)
438 			goto done;
439 	}
440 
441 	/* If it's a nil transfer, wake up the top half now. */
442 	if (bp->b_bcount == 0)
443 		goto done;
444 
445 	/*
446 	 * Put the block number in terms of the logical blocksize
447 	 * of the "device".
448 	 */
449 
450 	blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
451 
452 	/*
453 	 * Translate the partition-relative block number to an absolute.
454 	 */
455 	if (DISKPART(bp->b_dev) != RAW_PART) {
456 		struct partition *pp;
457 
458 		pp = &vnd->sc_dkdev.dk_label->d_partitions[
459 		    DISKPART(bp->b_dev)];
460 		blkno += pp->p_offset;
461 	}
462 	bp->b_rawblkno = blkno;
463 
464 #ifdef DEBUG
465 	if (vnddebug & VDB_FOLLOW)
466 		printf("vndstrategy(%p): unit %d\n", bp, unit);
467 #endif
468 	BUFQ_PUT(vnd->sc_tab, bp);
469 	wakeup(&vnd->sc_tab);
470 	splx(s);
471 	return;
472 done:
473 	biodone(bp);
474 	splx(s);
475 }
476 
477 static void
478 vndthread(void *arg)
479 {
480 	struct vnd_softc *vnd = arg;
481 	struct mount *mp;
482 	int s, bsize;
483 	int sz, error;
484 	struct disklabel *lp;
485 
486 	s = splbio();
487 	vnd->sc_flags |= VNF_KTHREAD;
488 	wakeup(&vnd->sc_kthread);
489 
490 	/*
491 	 * Dequeue requests, break them into bsize pieces and submit using
492 	 * VOP_BMAP/VOP_STRATEGY.
493 	 */
494 	while ((vnd->sc_flags & VNF_VUNCONF) == 0) {
495 		struct vndxfer *vnx;
496 		off_t offset;
497 		int resid;
498 		int skipped = 0;
499 		off_t bn;
500 		int flags;
501 		struct buf *obp;
502 		struct buf *bp;
503 
504 		obp = BUFQ_GET(vnd->sc_tab);
505 		if (obp == NULL) {
506 			tsleep(&vnd->sc_tab, PRIBIO, "vndbp", 0);
507 			continue;
508 		};
509 		splx(s);
510 		flags = obp->b_flags;
511 #ifdef DEBUG
512 		if (vnddebug & VDB_FOLLOW)
513 			printf("vndthread(%p\n", obp);
514 #endif
515 		lp = vnd->sc_dkdev.dk_label;
516 
517 		/* convert to a byte offset within the file. */
518 		bn = obp->b_rawblkno * lp->d_secsize;
519 
520 		if (vnd->sc_vp->v_mount == NULL) {
521 			obp->b_error = ENXIO;
522 			obp->b_flags |= B_ERROR;
523 			goto done;
524 		}
525 #ifdef VND_COMPRESSION
526 		/* handle a compressed read */
527 		if ((flags & B_READ) != 0 && (vnd->sc_flags & VNF_COMP)) {
528 			compstrategy(obp, bn);
529 			goto done;
530 		}
531 #endif /* VND_COMPRESSION */
532 
533 		bsize = vnd->sc_vp->v_mount->mnt_stat.f_iosize;
534 
535 		/*
536 		 * Allocate a header for this transfer and link it to the
537 		 * buffer
538 		 */
539 		s = splbio();
540 		vnx = VND_GETXFER(vnd);
541 		splx(s);
542 		vnx->vx_vnd = vnd;
543 
544 		bp = &vnx->vx_buf;
545 		BUF_INIT(bp);
546 		bp->b_flags = (obp->b_flags & B_READ) | B_CALL;
547 		bp->b_iodone = vndiodone;
548 		bp->b_private = obp;
549 		bp->b_vp = NULL;
550 		bp->b_data = obp->b_data;
551 		bp->b_bcount = bp->b_resid = obp->b_bcount;
552 		BIO_COPYPRIO(bp, obp);
553 
554 		s = splbio();
555 		while (vnd->sc_active >= vnd->sc_maxactive) {
556 			tsleep(&vnd->sc_tab, PRIBIO, "vndac", 0);
557 		}
558 		vnd->sc_active++;
559 		splx(s);
560 
561 		if ((flags & B_READ) == 0)
562 			vn_start_write(vnd->sc_vp, &mp, V_WAIT);
563 
564 		/* Instrumentation. */
565 		disk_busy(&vnd->sc_dkdev);
566 
567 		/*
568 		 * Feed requests sequentially.
569 		 * We do it this way to keep from flooding NFS servers if we
570 		 * are connected to an NFS file.  This places the burden on
571 		 * the client rather than the server.
572 		 */
573 		error = 0;
574 		for (offset = 0, resid = bp->b_resid; resid;
575 		    resid -= sz, offset += sz) {
576 			struct buf *nbp;
577 			struct vnode *vp;
578 			daddr_t nbn;
579 			int off, nra;
580 
581 			nra = 0;
582 			vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY | LK_CANRECURSE);
583 			error = VOP_BMAP(vnd->sc_vp, bn / bsize, &vp, &nbn, &nra);
584 			VOP_UNLOCK(vnd->sc_vp, 0);
585 
586 			if (error == 0 && (long)nbn == -1)
587 				error = EIO;
588 
589 			/*
590 			 * If there was an error or a hole in the file...punt.
591 			 * Note that we may have to wait for any operations
592 			 * that we have already fired off before releasing
593 			 * the buffer.
594 			 *
595 			 * XXX we could deal with holes here but it would be
596 			 * a hassle (in the write case).
597 			 */
598 			if (error) {
599 				skipped += resid;
600 				break;
601 			}
602 
603 #ifdef DEBUG
604 			if (!dovndcluster)
605 				nra = 0;
606 #endif
607 
608 			if ((off = bn % bsize) != 0)
609 				sz = bsize - off;
610 			else
611 				sz = (1 + nra) * bsize;
612 			if (resid < sz)
613 				sz = resid;
614 #ifdef	DEBUG
615 			if (vnddebug & VDB_IO)
616 				printf("vndstrategy: vp %p/%p bn 0x%qx/0x%" PRIx64
617 				       " sz 0x%x\n",
618 				    vnd->sc_vp, vp, (long long)bn, nbn, sz);
619 #endif
620 
621 			nbp = getiobuf();
622 			nestiobuf_setup(bp, nbp, offset, sz);
623 			nbp->b_blkno = nbn + btodb(off);
624 
625 #if 0 /* XXX #ifdef DEBUG */
626 			if (vnddebug & VDB_IO)
627 				printf("vndstart(%ld): bp %p vp %p blkno "
628 				    "0x%" PRIx64 " flags %x addr %p cnt 0x%x\n",
629 				    (long) (vnd-vnd_softc), &nbp->vb_buf,
630 				    nbp->vb_buf.b_vp, nbp->vb_buf.b_blkno,
631 				    nbp->vb_buf.b_flags, nbp->vb_buf.b_data,
632 				    nbp->vb_buf.b_bcount);
633 #endif
634 			VOP_STRATEGY(vp, nbp);
635 			bn += sz;
636 		}
637 		nestiobuf_done(bp, skipped, error);
638 
639 		if ((flags & B_READ) == 0)
640 			vn_finished_write(mp, 0);
641 
642 		s = splbio();
643 		continue;
644 done:
645 		biodone(obp);
646 		s = splbio();
647 	}
648 
649 	vnd->sc_flags &= (~VNF_KTHREAD | VNF_VUNCONF);
650 	wakeup(&vnd->sc_kthread);
651 	splx(s);
652 	kthread_exit(0);
653 }
654 
655 static void
656 vndiodone(struct buf *bp)
657 {
658 	struct vndxfer *vnx = VND_BUFTOXFER(bp);
659 	struct vnd_softc *vnd = vnx->vx_vnd;
660 	struct buf *obp = bp->b_private;
661 
662 	KASSERT(&vnx->vx_buf == bp);
663 	KASSERT(vnd->sc_active > 0);
664 #ifdef DEBUG
665 	if (vnddebug & VDB_IO) {
666 		printf("vndiodone1: bp %p iodone: error %d\n",
667 		    bp, (bp->b_flags & B_ERROR) != 0 ? bp->b_error : 0);
668 	}
669 #endif
670 	disk_unbusy(&vnd->sc_dkdev, bp->b_bcount - bp->b_resid,
671 	    (bp->b_flags & B_READ));
672 	vnd->sc_active--;
673 	if (vnd->sc_active == 0) {
674 		wakeup(&vnd->sc_tab);
675 	}
676 	obp->b_flags |= bp->b_flags & B_ERROR;
677 	obp->b_error = bp->b_error;
678 	obp->b_resid = bp->b_resid;
679 	VND_PUTXFER(vnd, vnx);
680 	biodone(obp);
681 }
682 
683 /* ARGSUSED */
684 static int
685 vndread(dev_t dev, struct uio *uio, int flags)
686 {
687 	int unit = vndunit(dev);
688 	struct vnd_softc *sc;
689 
690 #ifdef DEBUG
691 	if (vnddebug & VDB_FOLLOW)
692 		printf("vndread(0x%x, %p)\n", dev, uio);
693 #endif
694 
695 	if (unit >= numvnd)
696 		return (ENXIO);
697 	sc = &vnd_softc[unit];
698 
699 	if ((sc->sc_flags & VNF_INITED) == 0)
700 		return (ENXIO);
701 
702 	return (physio(vndstrategy, NULL, dev, B_READ, minphys, uio));
703 }
704 
705 /* ARGSUSED */
706 static int
707 vndwrite(dev_t dev, struct uio *uio, int flags)
708 {
709 	int unit = vndunit(dev);
710 	struct vnd_softc *sc;
711 
712 #ifdef DEBUG
713 	if (vnddebug & VDB_FOLLOW)
714 		printf("vndwrite(0x%x, %p)\n", dev, uio);
715 #endif
716 
717 	if (unit >= numvnd)
718 		return (ENXIO);
719 	sc = &vnd_softc[unit];
720 
721 	if ((sc->sc_flags & VNF_INITED) == 0)
722 		return (ENXIO);
723 
724 	return (physio(vndstrategy, NULL, dev, B_WRITE, minphys, uio));
725 }
726 
727 static int
728 vnd_cget(struct lwp *l, int unit, int *un, struct vattr *va)
729 {
730 	struct vnd_softc *vnd;
731 
732 	if (*un == -1)
733 		*un = unit;
734 	if (*un >= numvnd)
735 		return ENXIO;
736 	if (*un < 0)
737 		return EINVAL;
738 
739 	vnd = &vnd_softc[*un];
740 
741 	if ((vnd->sc_flags & VNF_INITED) == 0)
742 		return -1;
743 
744 	return VOP_GETATTR(vnd->sc_vp, va, l->l_proc->p_ucred, l);
745 }
746 
747 /* ARGSUSED */
748 static int
749 vndioctl(dev_t dev, u_long cmd, caddr_t data, int flag, struct lwp *l)
750 {
751 	int unit = vndunit(dev);
752 	struct vnd_softc *vnd;
753 	struct vnd_ioctl *vio;
754 	struct vattr vattr;
755 	struct nameidata nd;
756 	int error, part, pmask;
757 	size_t geomsize;
758 	struct proc *p =  (l != NULL) ? l->l_proc : NULL;
759 	int fflags;
760 #ifdef __HAVE_OLD_DISKLABEL
761 	struct disklabel newlabel;
762 #endif
763 
764 #ifdef DEBUG
765 	if (vnddebug & VDB_FOLLOW)
766 		printf("vndioctl(0x%x, 0x%lx, %p, 0x%x, %p): unit %d\n",
767 		    dev, cmd, data, flag, p, unit);
768 #endif
769 	if (unit >= numvnd)
770 		return (ENXIO);
771 
772 	vnd = &vnd_softc[unit];
773 	vio = (struct vnd_ioctl *)data;
774 
775 	/* Must be open for writes for these commands... */
776 	switch (cmd) {
777 	case VNDIOCSET:
778 	case VNDIOCCLR:
779 	case DIOCSDINFO:
780 	case DIOCWDINFO:
781 #ifdef __HAVE_OLD_DISKLABEL
782 	case ODIOCSDINFO:
783 	case ODIOCWDINFO:
784 #endif
785 	case DIOCKLABEL:
786 	case DIOCWLABEL:
787 		if ((flag & FWRITE) == 0)
788 			return (EBADF);
789 	}
790 
791 	/* Must be initialized for these... */
792 	switch (cmd) {
793 	case VNDIOCCLR:
794 	case DIOCGDINFO:
795 	case DIOCSDINFO:
796 	case DIOCWDINFO:
797 	case DIOCGPART:
798 	case DIOCKLABEL:
799 	case DIOCWLABEL:
800 	case DIOCGDEFLABEL:
801 #ifdef __HAVE_OLD_DISKLABEL
802 	case ODIOCGDINFO:
803 	case ODIOCSDINFO:
804 	case ODIOCWDINFO:
805 	case ODIOCGDEFLABEL:
806 #endif
807 		if ((vnd->sc_flags & VNF_INITED) == 0)
808 			return (ENXIO);
809 	}
810 
811 	switch (cmd) {
812 	case VNDIOCSET:
813 		if (vnd->sc_flags & VNF_INITED)
814 			return (EBUSY);
815 
816 		if ((error = vndlock(vnd)) != 0)
817 			return (error);
818 
819 		fflags = FREAD;
820 		if ((vio->vnd_flags & VNDIOF_READONLY) == 0)
821 			fflags |= FWRITE;
822 		NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, vio->vnd_file, l);
823 		if ((error = vn_open(&nd, fflags, 0)) != 0)
824 			goto unlock_and_exit;
825 		error = VOP_GETATTR(nd.ni_vp, &vattr, l->l_proc->p_ucred, l);
826 		if (!error && nd.ni_vp->v_type != VREG)
827 			error = EOPNOTSUPP;
828 		if (error) {
829 			VOP_UNLOCK(nd.ni_vp, 0);
830 			goto close_and_exit;
831 		}
832 
833 		/* If using a compressed file, initialize its info */
834 		/* (or abort with an error if kernel has no compression) */
835 		if (vio->vnd_flags & VNF_COMP) {
836 #ifdef VND_COMPRESSION
837 			struct vnd_comp_header *ch;
838 			int i;
839 			u_int32_t comp_size;
840 			u_int32_t comp_maxsize;
841 
842 			/* allocate space for compresed file header */
843 			ch = malloc(sizeof(struct vnd_comp_header),
844 			M_TEMP, M_WAITOK);
845 
846 			/* read compressed file header */
847 			error = vn_rdwr(UIO_READ, nd.ni_vp, (caddr_t)ch,
848 			  sizeof(struct vnd_comp_header), 0, UIO_SYSSPACE,
849 			  IO_UNIT|IO_NODELOCKED, p->p_ucred, NULL, NULL);
850 			if(error) {
851 				free(ch, M_TEMP);
852 				VOP_UNLOCK(nd.ni_vp, 0);
853 				goto close_and_exit;
854 			}
855 
856 			/* save some header info */
857 			vnd->sc_comp_blksz = ntohl(ch->block_size);
858 			/* note last offset is the file byte size */
859 			vnd->sc_comp_numoffs = ntohl(ch->num_blocks)+1;
860 			free(ch, M_TEMP);
861 			if(vnd->sc_comp_blksz % DEV_BSIZE !=0) {
862 				VOP_UNLOCK(nd.ni_vp, 0);
863 				error = EINVAL;
864 				goto close_and_exit;
865 			}
866 			if(sizeof(struct vnd_comp_header) +
867 			  sizeof(u_int64_t) * vnd->sc_comp_numoffs >
868 			  vattr.va_size) {
869 				VOP_UNLOCK(nd.ni_vp, 0);
870 				error = EINVAL;
871 				goto close_and_exit;
872 			}
873 
874 			/* set decompressed file size */
875 			vattr.va_size =
876 			  (vnd->sc_comp_numoffs - 1) * vnd->sc_comp_blksz;
877 
878 			/* allocate space for all the compressed offsets */
879 			vnd->sc_comp_offsets =
880 			malloc(sizeof(u_int64_t) * vnd->sc_comp_numoffs,
881 			M_DEVBUF, M_WAITOK);
882 
883 			/* read in the offsets */
884 			error = vn_rdwr(UIO_READ, nd.ni_vp,
885 			  (caddr_t)vnd->sc_comp_offsets,
886 			  sizeof(u_int64_t) * vnd->sc_comp_numoffs,
887 			  sizeof(struct vnd_comp_header), UIO_SYSSPACE,
888 			  IO_UNIT|IO_NODELOCKED, p->p_ucred, NULL, NULL);
889 			if(error) {
890 				VOP_UNLOCK(nd.ni_vp, 0);
891 				goto close_and_exit;
892 			}
893 			/*
894 			 * find largest block size (used for allocation limit).
895 			 * Also convert offset to native byte order.
896 			 */
897 			comp_maxsize = 0;
898 			for (i = 0; i < vnd->sc_comp_numoffs - 1; i++) {
899 				vnd->sc_comp_offsets[i] =
900 				  be64toh(vnd->sc_comp_offsets[i]);
901 				comp_size = be64toh(vnd->sc_comp_offsets[i + 1])
902 				  - vnd->sc_comp_offsets[i];
903 				if (comp_size > comp_maxsize)
904 					comp_maxsize = comp_size;
905 			}
906 			vnd->sc_comp_offsets[vnd->sc_comp_numoffs - 1] =
907 			  be64toh(vnd->sc_comp_offsets[vnd->sc_comp_numoffs - 1]);
908 
909 			/* create compressed data buffer */
910 			vnd->sc_comp_buff = malloc(comp_maxsize,
911 			  M_DEVBUF, M_WAITOK);
912 
913 			/* create decompressed buffer */
914 			vnd->sc_comp_decombuf = malloc(vnd->sc_comp_blksz,
915 			  M_DEVBUF, M_WAITOK);
916 			vnd->sc_comp_buffblk = -1;
917 
918 			/* Initialize decompress stream */
919 			bzero(&vnd->sc_comp_stream, sizeof(z_stream));
920 			vnd->sc_comp_stream.zalloc = vnd_alloc;
921 			vnd->sc_comp_stream.zfree = vnd_free;
922 			error = inflateInit2(&vnd->sc_comp_stream, MAX_WBITS);
923 			if(error) {
924 				if(vnd->sc_comp_stream.msg)
925 					printf("vnd%d: compressed file, %s\n",
926 					  unit, vnd->sc_comp_stream.msg);
927 				VOP_UNLOCK(nd.ni_vp, 0);
928 				error = EINVAL;
929 				goto close_and_exit;
930 			}
931 
932 			vnd->sc_flags |= VNF_COMP | VNF_READONLY;
933 #else /* !VND_COMPRESSION */
934 			VOP_UNLOCK(nd.ni_vp, 0);
935 			error = EOPNOTSUPP;
936 			goto close_and_exit;
937 #endif /* VND_COMPRESSION */
938 		}
939 
940 		VOP_UNLOCK(nd.ni_vp, 0);
941 		vnd->sc_vp = nd.ni_vp;
942 		vnd->sc_size = btodb(vattr.va_size);	/* note truncation */
943 
944 		/*
945 		 * Use pseudo-geometry specified.  If none was provided,
946 		 * use "standard" Adaptec fictitious geometry.
947 		 */
948 		if (vio->vnd_flags & VNDIOF_HASGEOM) {
949 
950 			memcpy(&vnd->sc_geom, &vio->vnd_geom,
951 			    sizeof(vio->vnd_geom));
952 
953 			/*
954 			 * Sanity-check the sector size.
955 			 * XXX Don't allow secsize < DEV_BSIZE.	 Should
956 			 * XXX we?
957 			 */
958 			if (vnd->sc_geom.vng_secsize < DEV_BSIZE ||
959 			    (vnd->sc_geom.vng_secsize % DEV_BSIZE) != 0 ||
960 			    vnd->sc_geom.vng_ncylinders == 0 ||
961 			    (vnd->sc_geom.vng_ntracks *
962 			     vnd->sc_geom.vng_nsectors) == 0) {
963 				error = EINVAL;
964 				goto close_and_exit;
965 			}
966 
967 			/*
968 			 * Compute the size (in DEV_BSIZE blocks) specified
969 			 * by the geometry.
970 			 */
971 			geomsize = (vnd->sc_geom.vng_nsectors *
972 			    vnd->sc_geom.vng_ntracks *
973 			    vnd->sc_geom.vng_ncylinders) *
974 			    (vnd->sc_geom.vng_secsize / DEV_BSIZE);
975 
976 			/*
977 			 * Sanity-check the size against the specified
978 			 * geometry.
979 			 */
980 			if (vnd->sc_size < geomsize) {
981 				error = EINVAL;
982 				goto close_and_exit;
983 			}
984 		} else if (vnd->sc_size >= (32 * 64)) {
985 			/*
986 			 * Size must be at least 2048 DEV_BSIZE blocks
987 			 * (1M) in order to use this geometry.
988 			 */
989 			vnd->sc_geom.vng_secsize = DEV_BSIZE;
990 			vnd->sc_geom.vng_nsectors = 32;
991 			vnd->sc_geom.vng_ntracks = 64;
992 			vnd->sc_geom.vng_ncylinders = vnd->sc_size / (64 * 32);
993 		} else {
994 			vnd->sc_geom.vng_secsize = DEV_BSIZE;
995 			vnd->sc_geom.vng_nsectors = 1;
996 			vnd->sc_geom.vng_ntracks = 1;
997 			vnd->sc_geom.vng_ncylinders = vnd->sc_size;
998 		}
999 
1000 		if (vio->vnd_flags & VNDIOF_READONLY) {
1001 			vnd->sc_flags |= VNF_READONLY;
1002 		}
1003 
1004 		if ((error = vndsetcred(vnd, p->p_ucred)) != 0)
1005 			goto close_and_exit;
1006 
1007 		memset(vnd->sc_xname, 0, sizeof(vnd->sc_xname)); /* XXX */
1008 		snprintf(vnd->sc_xname, sizeof(vnd->sc_xname), "vnd%d", unit);
1009 
1010 
1011 		vndthrottle(vnd, vnd->sc_vp);
1012 		vio->vnd_size = dbtob(vnd->sc_size);
1013 		vnd->sc_flags |= VNF_INITED;
1014 
1015 		/* create the kernel thread, wait for it to be up */
1016 		error = kthread_create1(vndthread, vnd, &vnd->sc_kthread,
1017 		    vnd->sc_xname);
1018 		if (error)
1019 			goto close_and_exit;
1020 		while ((vnd->sc_flags & VNF_KTHREAD) == 0) {
1021 			tsleep(&vnd->sc_kthread, PRIBIO, "vndthr", 0);
1022 		}
1023 #ifdef DEBUG
1024 		if (vnddebug & VDB_INIT)
1025 			printf("vndioctl: SET vp %p size 0x%lx %d/%d/%d/%d\n",
1026 			    vnd->sc_vp, (unsigned long) vnd->sc_size,
1027 			    vnd->sc_geom.vng_secsize,
1028 			    vnd->sc_geom.vng_nsectors,
1029 			    vnd->sc_geom.vng_ntracks,
1030 			    vnd->sc_geom.vng_ncylinders);
1031 #endif
1032 
1033 		/* Attach the disk. */
1034 		vnd->sc_dkdev.dk_name = vnd->sc_xname;
1035 		pseudo_disk_attach(&vnd->sc_dkdev);
1036 
1037 		/* Initialize the xfer and buffer pools. */
1038 		pool_init(&vnd->sc_vxpool, sizeof(struct vndxfer), 0,
1039 		    0, 0, "vndxpl", NULL);
1040 
1041 		/* Try and read the disklabel. */
1042 		vndgetdisklabel(dev);
1043 
1044 		vndunlock(vnd);
1045 
1046 		break;
1047 
1048 close_and_exit:
1049 		(void) vn_close(nd.ni_vp, fflags, p->p_ucred, l);
1050 unlock_and_exit:
1051 #ifdef VND_COMPRESSION
1052 		/* free any allocated memory (for compressed file) */
1053 		if(vnd->sc_comp_offsets) {
1054 			free(vnd->sc_comp_offsets, M_DEVBUF);
1055 			vnd->sc_comp_offsets = NULL;
1056 		}
1057 		if(vnd->sc_comp_buff) {
1058 			free(vnd->sc_comp_buff, M_DEVBUF);
1059 			vnd->sc_comp_buff = NULL;
1060 		}
1061 		if(vnd->sc_comp_decombuf) {
1062 			free(vnd->sc_comp_decombuf, M_DEVBUF);
1063 			vnd->sc_comp_decombuf = NULL;
1064 		}
1065 #endif /* VND_COMPRESSION */
1066 		vndunlock(vnd);
1067 		return (error);
1068 
1069 	case VNDIOCCLR:
1070 		if ((error = vndlock(vnd)) != 0)
1071 			return (error);
1072 
1073 		/*
1074 		 * Don't unconfigure if any other partitions are open
1075 		 * or if both the character and block flavors of this
1076 		 * partition are open.
1077 		 */
1078 		part = DISKPART(dev);
1079 		pmask = (1 << part);
1080 		if (((vnd->sc_dkdev.dk_openmask & ~pmask) ||
1081 		    ((vnd->sc_dkdev.dk_bopenmask & pmask) &&
1082 		    (vnd->sc_dkdev.dk_copenmask & pmask))) &&
1083 			!(vio->vnd_flags & VNDIOF_FORCE)) {
1084 			vndunlock(vnd);
1085 			return (EBUSY);
1086 		}
1087 
1088 		/*
1089 		 * XXX vndclear() might call vndclose() implicitely;
1090 		 * release lock to avoid recursion
1091 		 */
1092 		vndunlock(vnd);
1093 		vndclear(vnd, minor(dev));
1094 #ifdef DEBUG
1095 		if (vnddebug & VDB_INIT)
1096 			printf("vndioctl: CLRed\n");
1097 #endif
1098 
1099 		/* Destroy the xfer and buffer pools. */
1100 		pool_destroy(&vnd->sc_vxpool);
1101 
1102 		/* Detatch the disk. */
1103 		pseudo_disk_detach(&vnd->sc_dkdev);
1104 
1105 		break;
1106 
1107 #ifdef COMPAT_30
1108 	case VNDIOOCGET: {
1109 		struct vnd_ouser *vnu;
1110 		struct vattr va;
1111 		vnu = (struct vnd_ouser *)data;
1112 		switch (error = vnd_cget(l, unit, &vnu->vnu_unit, &va)) {
1113 		case 0:
1114 			vnu->vnu_dev = va.va_fsid;
1115 			vnu->vnu_ino = va.va_fileid;
1116 			break;
1117 		case -1:
1118 			/* unused is not an error */
1119 			vnu->vnu_dev = 0;
1120 			vnu->vnu_ino = 0;
1121 			break;
1122 		default:
1123 			return error;
1124 		}
1125 		break;
1126 	}
1127 #endif
1128 	case VNDIOCGET: {
1129 		struct vnd_user *vnu;
1130 		struct vattr va;
1131 		vnu = (struct vnd_user *)data;
1132 		switch (error = vnd_cget(l, unit, &vnu->vnu_unit, &va)) {
1133 		case 0:
1134 			vnu->vnu_dev = va.va_fsid;
1135 			vnu->vnu_ino = va.va_fileid;
1136 			break;
1137 		case -1:
1138 			/* unused is not an error */
1139 			vnu->vnu_dev = 0;
1140 			vnu->vnu_ino = 0;
1141 			break;
1142 		default:
1143 			return error;
1144 		}
1145 		break;
1146 	}
1147 
1148 	case DIOCGDINFO:
1149 		*(struct disklabel *)data = *(vnd->sc_dkdev.dk_label);
1150 		break;
1151 
1152 #ifdef __HAVE_OLD_DISKLABEL
1153 	case ODIOCGDINFO:
1154 		newlabel = *(vnd->sc_dkdev.dk_label);
1155 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1156 			return ENOTTY;
1157 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1158 		break;
1159 #endif
1160 
1161 	case DIOCGPART:
1162 		((struct partinfo *)data)->disklab = vnd->sc_dkdev.dk_label;
1163 		((struct partinfo *)data)->part =
1164 		    &vnd->sc_dkdev.dk_label->d_partitions[DISKPART(dev)];
1165 		break;
1166 
1167 	case DIOCWDINFO:
1168 	case DIOCSDINFO:
1169 #ifdef __HAVE_OLD_DISKLABEL
1170 	case ODIOCWDINFO:
1171 	case ODIOCSDINFO:
1172 #endif
1173 	{
1174 		struct disklabel *lp;
1175 
1176 		if ((error = vndlock(vnd)) != 0)
1177 			return (error);
1178 
1179 		vnd->sc_flags |= VNF_LABELLING;
1180 
1181 #ifdef __HAVE_OLD_DISKLABEL
1182 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
1183 			memset(&newlabel, 0, sizeof newlabel);
1184 			memcpy(&newlabel, data, sizeof (struct olddisklabel));
1185 			lp = &newlabel;
1186 		} else
1187 #endif
1188 		lp = (struct disklabel *)data;
1189 
1190 		error = setdisklabel(vnd->sc_dkdev.dk_label,
1191 		    lp, 0, vnd->sc_dkdev.dk_cpulabel);
1192 		if (error == 0) {
1193 			if (cmd == DIOCWDINFO
1194 #ifdef __HAVE_OLD_DISKLABEL
1195 			    || cmd == ODIOCWDINFO
1196 #endif
1197 			   )
1198 				error = writedisklabel(VNDLABELDEV(dev),
1199 				    vndstrategy, vnd->sc_dkdev.dk_label,
1200 				    vnd->sc_dkdev.dk_cpulabel);
1201 		}
1202 
1203 		vnd->sc_flags &= ~VNF_LABELLING;
1204 
1205 		vndunlock(vnd);
1206 
1207 		if (error)
1208 			return (error);
1209 		break;
1210 	}
1211 
1212 	case DIOCKLABEL:
1213 		if (*(int *)data != 0)
1214 			vnd->sc_flags |= VNF_KLABEL;
1215 		else
1216 			vnd->sc_flags &= ~VNF_KLABEL;
1217 		break;
1218 
1219 	case DIOCWLABEL:
1220 		if (*(int *)data != 0)
1221 			vnd->sc_flags |= VNF_WLABEL;
1222 		else
1223 			vnd->sc_flags &= ~VNF_WLABEL;
1224 		break;
1225 
1226 	case DIOCGDEFLABEL:
1227 		vndgetdefaultlabel(vnd, (struct disklabel *)data);
1228 		break;
1229 
1230 #ifdef __HAVE_OLD_DISKLABEL
1231 	case ODIOCGDEFLABEL:
1232 		vndgetdefaultlabel(vnd, &newlabel);
1233 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1234 			return ENOTTY;
1235 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1236 		break;
1237 #endif
1238 
1239 	default:
1240 		return (ENOTTY);
1241 	}
1242 
1243 	return (0);
1244 }
1245 
1246 /*
1247  * Duplicate the current processes' credentials.  Since we are called only
1248  * as the result of a SET ioctl and only root can do that, any future access
1249  * to this "disk" is essentially as root.  Note that credentials may change
1250  * if some other uid can write directly to the mapped file (NFS).
1251  */
1252 static int
1253 vndsetcred(struct vnd_softc *vnd, struct ucred *cred)
1254 {
1255 	struct uio auio;
1256 	struct iovec aiov;
1257 	char *tmpbuf;
1258 	int error;
1259 
1260 	vnd->sc_cred = crdup(cred);
1261 	tmpbuf = malloc(DEV_BSIZE, M_TEMP, M_WAITOK);
1262 
1263 	/* XXX: Horrible kludge to establish credentials for NFS */
1264 	aiov.iov_base = tmpbuf;
1265 	aiov.iov_len = min(DEV_BSIZE, dbtob(vnd->sc_size));
1266 	auio.uio_iov = &aiov;
1267 	auio.uio_iovcnt = 1;
1268 	auio.uio_offset = 0;
1269 	auio.uio_rw = UIO_READ;
1270 	auio.uio_segflg = UIO_SYSSPACE;
1271 	auio.uio_resid = aiov.iov_len;
1272 	vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1273 	error = VOP_READ(vnd->sc_vp, &auio, 0, vnd->sc_cred);
1274 	if (error == 0) {
1275 		/*
1276 		 * Because vnd does all IO directly through the vnode
1277 		 * we need to flush (at least) the buffer from the above
1278 		 * VOP_READ from the buffer cache to prevent cache
1279 		 * incoherencies.  Also, be careful to write dirty
1280 		 * buffers back to stable storage.
1281 		 */
1282 		error = vinvalbuf(vnd->sc_vp, V_SAVE, vnd->sc_cred,
1283 			    curlwp, 0, 0);
1284 	}
1285 	VOP_UNLOCK(vnd->sc_vp, 0);
1286 
1287 	free(tmpbuf, M_TEMP);
1288 	return (error);
1289 }
1290 
1291 /*
1292  * Set maxactive based on FS type
1293  */
1294 static void
1295 vndthrottle(struct vnd_softc *vnd, struct vnode *vp)
1296 {
1297 #ifdef NFS
1298 	extern int (**nfsv2_vnodeop_p)(void *);
1299 
1300 	if (vp->v_op == nfsv2_vnodeop_p)
1301 		vnd->sc_maxactive = 2;
1302 	else
1303 #endif
1304 		vnd->sc_maxactive = 8;
1305 
1306 	if (vnd->sc_maxactive < 1)
1307 		vnd->sc_maxactive = 1;
1308 }
1309 
1310 #if 0
1311 static void
1312 vndshutdown(void)
1313 {
1314 	struct vnd_softc *vnd;
1315 
1316 	for (vnd = &vnd_softc[0]; vnd < &vnd_softc[numvnd]; vnd++)
1317 		if (vnd->sc_flags & VNF_INITED)
1318 			vndclear(vnd);
1319 }
1320 #endif
1321 
1322 static void
1323 vndclear(struct vnd_softc *vnd, int myminor)
1324 {
1325 	struct vnode *vp = vnd->sc_vp;
1326 	struct lwp *l = curlwp;
1327 	int fflags = FREAD;
1328 	int bmaj, cmaj, i, mn;
1329 	int s;
1330 
1331 #ifdef DEBUG
1332 	if (vnddebug & VDB_FOLLOW)
1333 		printf("vndclear(%p): vp %p\n", vnd, vp);
1334 #endif
1335 	/* locate the major number */
1336 	bmaj = bdevsw_lookup_major(&vnd_bdevsw);
1337 	cmaj = cdevsw_lookup_major(&vnd_cdevsw);
1338 
1339 	/* Nuke the vnodes for any open instances */
1340 	for (i = 0; i < MAXPARTITIONS; i++) {
1341 		mn = DISKMINOR(vnd->sc_unit, i);
1342 		vdevgone(bmaj, mn, mn, VBLK);
1343 		if (mn != myminor) /* XXX avoid to kill own vnode */
1344 			vdevgone(cmaj, mn, mn, VCHR);
1345 	}
1346 
1347 	if ((vnd->sc_flags & VNF_READONLY) == 0)
1348 		fflags |= FWRITE;
1349 
1350 	s = splbio();
1351 	bufq_drain(vnd->sc_tab);
1352 	splx(s);
1353 
1354 	vnd->sc_flags |= VNF_VUNCONF;
1355 	wakeup(&vnd->sc_tab);
1356 	while (vnd->sc_flags & VNF_KTHREAD)
1357 		tsleep(&vnd->sc_kthread, PRIBIO, "vnthr", 0);
1358 
1359 #ifdef VND_COMPRESSION
1360 	/* free the compressed file buffers */
1361 	if(vnd->sc_flags & VNF_COMP) {
1362 		if(vnd->sc_comp_offsets) {
1363 			free(vnd->sc_comp_offsets, M_DEVBUF);
1364 			vnd->sc_comp_offsets = NULL;
1365 		}
1366 		if(vnd->sc_comp_buff) {
1367 			free(vnd->sc_comp_buff, M_DEVBUF);
1368 			vnd->sc_comp_buff = NULL;
1369 		}
1370 		if(vnd->sc_comp_decombuf) {
1371 			free(vnd->sc_comp_decombuf, M_DEVBUF);
1372 			vnd->sc_comp_decombuf = NULL;
1373 		}
1374 	}
1375 #endif /* VND_COMPRESSION */
1376 	vnd->sc_flags &=
1377 	    ~(VNF_INITED | VNF_READONLY | VNF_VLABEL
1378 	      | VNF_VUNCONF | VNF_COMP);
1379 	if (vp == (struct vnode *)0)
1380 		panic("vndclear: null vp");
1381 	(void) vn_close(vp, fflags, vnd->sc_cred, l);
1382 	crfree(vnd->sc_cred);
1383 	vnd->sc_vp = (struct vnode *)0;
1384 	vnd->sc_cred = (struct ucred *)0;
1385 	vnd->sc_size = 0;
1386 }
1387 
1388 static int
1389 vndsize(dev_t dev)
1390 {
1391 	struct vnd_softc *sc;
1392 	struct disklabel *lp;
1393 	int part, unit, omask;
1394 	int size;
1395 
1396 	unit = vndunit(dev);
1397 	if (unit >= numvnd)
1398 		return (-1);
1399 	sc = &vnd_softc[unit];
1400 
1401 	if ((sc->sc_flags & VNF_INITED) == 0)
1402 		return (-1);
1403 
1404 	part = DISKPART(dev);
1405 	omask = sc->sc_dkdev.dk_openmask & (1 << part);
1406 	lp = sc->sc_dkdev.dk_label;
1407 
1408 	if (omask == 0 && vndopen(dev, 0, S_IFBLK, curlwp))	/* XXX */
1409 		return (-1);
1410 
1411 	if (lp->d_partitions[part].p_fstype != FS_SWAP)
1412 		size = -1;
1413 	else
1414 		size = lp->d_partitions[part].p_size *
1415 		    (lp->d_secsize / DEV_BSIZE);
1416 
1417 	if (omask == 0 && vndclose(dev, 0, S_IFBLK, curlwp))	/* XXX */
1418 		return (-1);
1419 
1420 	return (size);
1421 }
1422 
1423 static int
1424 vnddump(dev_t dev, daddr_t blkno, caddr_t va, size_t size)
1425 {
1426 
1427 	/* Not implemented. */
1428 	return ENXIO;
1429 }
1430 
1431 static void
1432 vndgetdefaultlabel(struct vnd_softc *sc, struct disklabel *lp)
1433 {
1434 	struct vndgeom *vng = &sc->sc_geom;
1435 	struct partition *pp;
1436 
1437 	memset(lp, 0, sizeof(*lp));
1438 
1439 	lp->d_secperunit = sc->sc_size / (vng->vng_secsize / DEV_BSIZE);
1440 	lp->d_secsize = vng->vng_secsize;
1441 	lp->d_nsectors = vng->vng_nsectors;
1442 	lp->d_ntracks = vng->vng_ntracks;
1443 	lp->d_ncylinders = vng->vng_ncylinders;
1444 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1445 
1446 	strncpy(lp->d_typename, "vnd", sizeof(lp->d_typename));
1447 	lp->d_type = DTYPE_VND;
1448 	strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
1449 	lp->d_rpm = 3600;
1450 	lp->d_interleave = 1;
1451 	lp->d_flags = 0;
1452 
1453 	pp = &lp->d_partitions[RAW_PART];
1454 	pp->p_offset = 0;
1455 	pp->p_size = lp->d_secperunit;
1456 	pp->p_fstype = FS_UNUSED;
1457 	lp->d_npartitions = RAW_PART + 1;
1458 
1459 	lp->d_magic = DISKMAGIC;
1460 	lp->d_magic2 = DISKMAGIC;
1461 	lp->d_checksum = dkcksum(lp);
1462 }
1463 
1464 /*
1465  * Read the disklabel from a vnd.  If one is not present, create a fake one.
1466  */
1467 static void
1468 vndgetdisklabel(dev_t dev)
1469 {
1470 	struct vnd_softc *sc = &vnd_softc[vndunit(dev)];
1471 	const char *errstring;
1472 	struct disklabel *lp = sc->sc_dkdev.dk_label;
1473 	struct cpu_disklabel *clp = sc->sc_dkdev.dk_cpulabel;
1474 	int i;
1475 
1476 	memset(clp, 0, sizeof(*clp));
1477 
1478 	vndgetdefaultlabel(sc, lp);
1479 
1480 	/*
1481 	 * Call the generic disklabel extraction routine.
1482 	 */
1483 	errstring = readdisklabel(VNDLABELDEV(dev), vndstrategy, lp, clp);
1484 	if (errstring) {
1485 		/*
1486 		 * Lack of disklabel is common, but we print the warning
1487 		 * anyway, since it might contain other useful information.
1488 		 */
1489 		printf("%s: %s\n", sc->sc_xname, errstring);
1490 
1491 		/*
1492 		 * For historical reasons, if there's no disklabel
1493 		 * present, all partitions must be FS_BSDFFS and
1494 		 * occupy the entire disk.
1495 		 */
1496 		for (i = 0; i < MAXPARTITIONS; i++) {
1497 			/*
1498 			 * Don't wipe out port specific hack (such as
1499 			 * dos partition hack of i386 port).
1500 			 */
1501 			if (lp->d_partitions[i].p_size != 0)
1502 				continue;
1503 
1504 			lp->d_partitions[i].p_size = lp->d_secperunit;
1505 			lp->d_partitions[i].p_offset = 0;
1506 			lp->d_partitions[i].p_fstype = FS_BSDFFS;
1507 		}
1508 
1509 		strncpy(lp->d_packname, "default label",
1510 		    sizeof(lp->d_packname));
1511 
1512 		lp->d_npartitions = MAXPARTITIONS;
1513 		lp->d_checksum = dkcksum(lp);
1514 	}
1515 
1516 	/* In-core label now valid. */
1517 	sc->sc_flags |= VNF_VLABEL;
1518 }
1519 
1520 /*
1521  * Wait interruptibly for an exclusive lock.
1522  *
1523  * XXX
1524  * Several drivers do this; it should be abstracted and made MP-safe.
1525  */
1526 static int
1527 vndlock(struct vnd_softc *sc)
1528 {
1529 	int error;
1530 
1531 	while ((sc->sc_flags & VNF_LOCKED) != 0) {
1532 		sc->sc_flags |= VNF_WANTED;
1533 		if ((error = tsleep(sc, PRIBIO | PCATCH, "vndlck", 0)) != 0)
1534 			return (error);
1535 	}
1536 	sc->sc_flags |= VNF_LOCKED;
1537 	return (0);
1538 }
1539 
1540 /*
1541  * Unlock and wake up any waiters.
1542  */
1543 static void
1544 vndunlock(struct vnd_softc *sc)
1545 {
1546 
1547 	sc->sc_flags &= ~VNF_LOCKED;
1548 	if ((sc->sc_flags & VNF_WANTED) != 0) {
1549 		sc->sc_flags &= ~VNF_WANTED;
1550 		wakeup(sc);
1551 	}
1552 }
1553 
1554 #ifdef VND_COMPRESSION
1555 /* compressed file read */
1556 static void
1557 compstrategy(struct buf *bp, off_t bn)
1558 {
1559 	int error;
1560 	int unit = vndunit(bp->b_dev);
1561 	struct vnd_softc *vnd = &vnd_softc[unit];
1562 	u_int32_t comp_block;
1563 	struct uio auio;
1564 	caddr_t addr;
1565 	int s;
1566 
1567 	/* set up constants for data move */
1568 	auio.uio_rw = UIO_READ;
1569 	auio.uio_segflg = bp->b_flags & B_PHYS ? UIO_USERSPACE : UIO_SYSSPACE;
1570 	auio.uio_lwp = LIST_FIRST(&bp->b_proc->p_lwps);
1571 
1572 	/* read, and transfer the data */
1573 	addr = bp->b_data;
1574 	s = splbio();
1575 	while (bp->b_resid > 0) {
1576 		unsigned length;
1577 		size_t length_in_buffer;
1578 		u_int32_t offset_in_buffer;
1579 		struct iovec aiov;
1580 
1581 		/* calculate the compressed block number */
1582 		comp_block = bn / (off_t)vnd->sc_comp_blksz;
1583 
1584 		/* check for good block number */
1585 		if (comp_block >= vnd->sc_comp_numoffs) {
1586 			bp->b_error = EINVAL;
1587 			bp->b_flags |= B_ERROR;
1588 			splx(s);
1589 			return;
1590 		}
1591 
1592 		/* read in the compressed block, if not in buffer */
1593 		if (comp_block != vnd->sc_comp_buffblk) {
1594 			length = vnd->sc_comp_offsets[comp_block + 1] -
1595 			    vnd->sc_comp_offsets[comp_block];
1596 			vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1597 			error = vn_rdwr(UIO_READ, vnd->sc_vp, vnd->sc_comp_buff,
1598 			    length, vnd->sc_comp_offsets[comp_block],
1599 			    UIO_SYSSPACE, IO_UNIT, vnd->sc_cred, NULL, NULL);
1600 			if (error) {
1601 				bp->b_error = error;
1602 				bp->b_flags |= B_ERROR;
1603 				VOP_UNLOCK(vnd->sc_vp, 0);
1604 				splx(s);
1605 				return;
1606 			}
1607 			/* uncompress the buffer */
1608 			vnd->sc_comp_stream.next_in = vnd->sc_comp_buff;
1609 			vnd->sc_comp_stream.avail_in = length;
1610 			vnd->sc_comp_stream.next_out = vnd->sc_comp_decombuf;
1611 			vnd->sc_comp_stream.avail_out = vnd->sc_comp_blksz;
1612 			inflateReset(&vnd->sc_comp_stream);
1613 			error = inflate(&vnd->sc_comp_stream, Z_FINISH);
1614 			if (error != Z_STREAM_END) {
1615 				if (vnd->sc_comp_stream.msg)
1616 					printf("%s: compressed file, %s\n",
1617 					    vnd->sc_xname,
1618 					    vnd->sc_comp_stream.msg);
1619 				bp->b_error = EBADMSG;
1620 				bp->b_flags |= B_ERROR;
1621 				VOP_UNLOCK(vnd->sc_vp, 0);
1622 				splx(s);
1623 				return;
1624 			}
1625 			vnd->sc_comp_buffblk = comp_block;
1626 			VOP_UNLOCK(vnd->sc_vp, 0);
1627 		}
1628 
1629 		/* transfer the usable uncompressed data */
1630 		offset_in_buffer = bn % (off_t)vnd->sc_comp_blksz;
1631 		length_in_buffer = vnd->sc_comp_blksz - offset_in_buffer;
1632 		if (length_in_buffer > bp->b_resid)
1633 			length_in_buffer = bp->b_resid;
1634 		auio.uio_iov = &aiov;
1635 		auio.uio_iovcnt = 1;
1636 		aiov.iov_base = addr;
1637 		aiov.iov_len = length_in_buffer;
1638 		auio.uio_resid = aiov.iov_len;
1639 		auio.uio_offset = 0;
1640 		error = uiomove(vnd->sc_comp_decombuf + offset_in_buffer,
1641 		    length_in_buffer, &auio);
1642 		if (error) {
1643 			bp->b_error = error;
1644 			bp->b_flags |= B_ERROR;
1645 			splx(s);
1646 			return;
1647 		}
1648 
1649 		bn += length_in_buffer;
1650 		addr += length_in_buffer;
1651 		bp->b_resid -= length_in_buffer;
1652 	}
1653 	splx(s);
1654 }
1655 
1656 /* compression memory allocation routines */
1657 static void *
1658 vnd_alloc(void *aux, u_int items, u_int siz)
1659 {
1660 	return malloc(items * siz, M_TEMP, M_NOWAIT);
1661 }
1662 
1663 static void
1664 vnd_free(void *aux, void *ptr)
1665 {
1666 	free(ptr, M_TEMP);
1667 }
1668 #endif /* VND_COMPRESSION */
1669