xref: /netbsd-src/sys/dev/vnd.c (revision 27527e67bbdf8d9ec84fd58803048ed6d181ece2)
1 /*	$NetBSD: vnd.c,v 1.140 2006/02/04 13:40:38 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 1.140 2006/02/04 13:40:38 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 #define VNDLABELDEV(dev) \
193     (MAKEDISKDEV(major((dev)), vndunit((dev)), RAW_PART))
194 
195 /* called by main() at boot time (XXX: and the LKM driver) */
196 void	vndattach(int);
197 
198 static void	vndclear(struct vnd_softc *, int);
199 static int	vndsetcred(struct vnd_softc *, struct ucred *);
200 static void	vndthrottle(struct vnd_softc *, struct vnode *);
201 static void	vndiodone(struct buf *);
202 #if 0
203 static void	vndshutdown(void);
204 #endif
205 
206 static void	vndgetdefaultlabel(struct vnd_softc *, struct disklabel *);
207 static void	vndgetdisklabel(dev_t, struct vnd_softc *);
208 
209 static int	vndlock(struct vnd_softc *);
210 static void	vndunlock(struct vnd_softc *);
211 #ifdef VND_COMPRESSION
212 static void	compstrategy(struct buf *, off_t);
213 static void	*vnd_alloc(void *, u_int, u_int);
214 static void	vnd_free(void *, void *);
215 #endif /* VND_COMPRESSION */
216 
217 static void vndthread(void *);
218 
219 static dev_type_open(vndopen);
220 static dev_type_close(vndclose);
221 static dev_type_read(vndread);
222 static dev_type_write(vndwrite);
223 static dev_type_ioctl(vndioctl);
224 static dev_type_strategy(vndstrategy);
225 static dev_type_dump(vnddump);
226 static dev_type_size(vndsize);
227 
228 const struct bdevsw vnd_bdevsw = {
229 	vndopen, vndclose, vndstrategy, vndioctl, vnddump, vndsize, D_DISK
230 };
231 
232 const struct cdevsw vnd_cdevsw = {
233 	vndopen, vndclose, vndread, vndwrite, vndioctl,
234 	nostop, notty, nopoll, nommap, nokqfilter, D_DISK
235 };
236 
237 static int	vnd_match(struct device *, struct cfdata *, void *);
238 static void	vnd_attach(struct device *, struct device *, void *);
239 static int	vnd_detach(struct device *, int);
240 
241 CFATTACH_DECL(vnd, sizeof(struct vnd_softc),
242     vnd_match, vnd_attach, vnd_detach, NULL);
243 extern struct cfdriver vnd_cd;
244 
245 static struct vnd_softc	*vnd_spawn(int);
246 int	vnd_destroy(struct device *);
247 
248 void
249 vndattach(int num)
250 {
251 	int error;
252 
253 	error = config_cfattach_attach(vnd_cd.cd_name, &vnd_ca);
254 	if (error)
255 		aprint_error("%s: unable to register cfattach\n",
256 		    vnd_cd.cd_name);
257 }
258 
259 static int
260 vnd_match(struct device *self, struct cfdata *cfdata, void *aux)
261 {
262 	return 1;
263 }
264 
265 static void
266 vnd_attach(struct device *parent, struct device *self, void *aux)
267 {
268 	struct vnd_softc *sc = (struct vnd_softc *)self;
269 
270 	sc->sc_comp_offsets = NULL;
271 	sc->sc_comp_buff = NULL;
272 	sc->sc_comp_decombuf = NULL;
273 	bufq_alloc(&sc->sc_tab, "disksort", BUFQ_SORT_RAWBLOCK);
274 	pseudo_disk_init(&sc->sc_dkdev);
275 
276 	aprint_normal("%s: vnode disk driver\n", self->dv_xname);
277 }
278 
279 static int
280 vnd_detach(struct device *self, int flags)
281 {
282 	struct vnd_softc *sc = (struct vnd_softc *)self;
283 	if (sc->sc_flags & VNF_INITED)
284 		return EBUSY;
285 
286 	bufq_free(sc->sc_tab);
287 
288 	return 0;
289 }
290 
291 static struct vnd_softc *
292 vnd_spawn(int unit)
293 {
294 	struct cfdata *cf;
295 
296 	cf = malloc(sizeof(*cf), M_DEVBUF, M_WAITOK);
297 	cf->cf_name = vnd_cd.cd_name;
298 	cf->cf_atname = vnd_cd.cd_name;
299 	cf->cf_unit = unit;
300 	cf->cf_fstate = FSTATE_STAR;
301 
302 	return (struct vnd_softc *)config_attach_pseudo(cf);
303 }
304 
305 int
306 vnd_destroy(struct device *dev)
307 {
308 	int error;
309 	struct cfdata *cf;
310 
311 	cf = dev->dv_cfdata;
312 	error = config_detach(dev, 0);
313 	if (error)
314 		return error;
315 	free(cf, M_DEVBUF);
316 	return 0;
317 }
318 
319 static int
320 vndopen(dev_t dev, int flags, int mode, struct lwp *l)
321 {
322 	int unit = vndunit(dev);
323 	struct vnd_softc *sc;
324 	int error = 0, part, pmask;
325 	struct disklabel *lp;
326 
327 #ifdef DEBUG
328 	if (vnddebug & VDB_FOLLOW)
329 		printf("vndopen(0x%x, 0x%x, 0x%x, %p)\n", dev, flags, mode, l);
330 #endif
331 	sc = device_lookup(&vnd_cd, unit);
332 	if (sc == NULL) {
333 		sc = vnd_spawn(unit);
334 		if (sc == NULL)
335 			return ENOMEM;
336 	}
337 
338 	if ((error = vndlock(sc)) != 0)
339 		return (error);
340 
341 	lp = sc->sc_dkdev.dk_label;
342 
343 	part = DISKPART(dev);
344 	pmask = (1 << part);
345 
346 	/*
347 	 * If we're initialized, check to see if there are any other
348 	 * open partitions.  If not, then it's safe to update the
349 	 * in-core disklabel.  Only read the disklabel if it is
350 	 * not already valid.
351 	 */
352 	if ((sc->sc_flags & (VNF_INITED|VNF_VLABEL)) == VNF_INITED &&
353 	    sc->sc_dkdev.dk_openmask == 0)
354 		vndgetdisklabel(dev, sc);
355 
356 	/* Check that the partitions exists. */
357 	if (part != RAW_PART) {
358 		if (((sc->sc_flags & VNF_INITED) == 0) ||
359 		    ((part >= lp->d_npartitions) ||
360 		     (lp->d_partitions[part].p_fstype == FS_UNUSED))) {
361 			error = ENXIO;
362 			goto done;
363 		}
364 	}
365 
366 	/* Prevent our unit from being unconfigured while open. */
367 	switch (mode) {
368 	case S_IFCHR:
369 		sc->sc_dkdev.dk_copenmask |= pmask;
370 		break;
371 
372 	case S_IFBLK:
373 		sc->sc_dkdev.dk_bopenmask |= pmask;
374 		break;
375 	}
376 	sc->sc_dkdev.dk_openmask =
377 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
378 
379  done:
380 	vndunlock(sc);
381 	return (error);
382 }
383 
384 static int
385 vndclose(dev_t dev, int flags, int mode, struct lwp *l)
386 {
387 	int unit = vndunit(dev);
388 	struct vnd_softc *sc;
389 	int error = 0, part;
390 
391 #ifdef DEBUG
392 	if (vnddebug & VDB_FOLLOW)
393 		printf("vndclose(0x%x, 0x%x, 0x%x, %p)\n", dev, flags, mode, l);
394 #endif
395 	sc = device_lookup(&vnd_cd, unit);
396 	if (sc == NULL)
397 		return ENXIO;
398 
399 	if ((error = vndlock(sc)) != 0)
400 		return (error);
401 
402 	part = DISKPART(dev);
403 
404 	/* ...that much closer to allowing unconfiguration... */
405 	switch (mode) {
406 	case S_IFCHR:
407 		sc->sc_dkdev.dk_copenmask &= ~(1 << part);
408 		break;
409 
410 	case S_IFBLK:
411 		sc->sc_dkdev.dk_bopenmask &= ~(1 << part);
412 		break;
413 	}
414 	sc->sc_dkdev.dk_openmask =
415 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
416 
417 	vndunlock(sc);
418 
419 	if ((sc->sc_flags & VNF_INITED) == 0) {
420 		if ((error = vnd_destroy((struct device *)sc)) != 0) {
421 			aprint_error("%s: unable to detach instance\n",
422 			    sc->sc_dev.dv_xname);
423 			return error;
424 		}
425 	}
426 
427 	return (0);
428 }
429 
430 /*
431  * Queue the request, and wakeup the kernel thread to handle it.
432  */
433 static void
434 vndstrategy(struct buf *bp)
435 {
436 	int unit = vndunit(bp->b_dev);
437 	struct vnd_softc *vnd =
438 	    (struct vnd_softc *)device_lookup(&vnd_cd, unit);
439 	struct disklabel *lp = vnd->sc_dkdev.dk_label;
440 	daddr_t blkno;
441 	int s = splbio();
442 
443 	bp->b_resid = bp->b_bcount;
444 
445 	if ((vnd->sc_flags & VNF_INITED) == 0) {
446 		bp->b_error = ENXIO;
447 		bp->b_flags |= B_ERROR;
448 		goto done;
449 	}
450 
451 	/*
452 	 * The transfer must be a whole number of blocks.
453 	 */
454 	if ((bp->b_bcount % lp->d_secsize) != 0) {
455 		bp->b_error = EINVAL;
456 		bp->b_flags |= B_ERROR;
457 		goto done;
458 	}
459 
460 	/*
461 	 * check if we're read-only.
462 	 */
463 	if ((vnd->sc_flags & VNF_READONLY) && !(bp->b_flags & B_READ)) {
464 		bp->b_error = EACCES;
465 		bp->b_flags |= B_ERROR;
466 		goto done;
467 	}
468 
469 	/*
470 	 * Do bounds checking and adjust transfer.  If there's an error,
471 	 * the bounds check will flag that for us.
472 	 */
473 	if (DISKPART(bp->b_dev) == RAW_PART) {
474 		if (bounds_check_with_mediasize(bp, DEV_BSIZE,
475 		    vnd->sc_size) <= 0)
476 			goto done;
477 	} else {
478 		if (bounds_check_with_label(&vnd->sc_dkdev,
479 		    bp, vnd->sc_flags & (VNF_WLABEL|VNF_LABELLING)) <= 0)
480 			goto done;
481 	}
482 
483 	/* If it's a nil transfer, wake up the top half now. */
484 	if (bp->b_bcount == 0)
485 		goto done;
486 
487 	/*
488 	 * Put the block number in terms of the logical blocksize
489 	 * of the "device".
490 	 */
491 
492 	blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
493 
494 	/*
495 	 * Translate the partition-relative block number to an absolute.
496 	 */
497 	if (DISKPART(bp->b_dev) != RAW_PART) {
498 		struct partition *pp;
499 
500 		pp = &vnd->sc_dkdev.dk_label->d_partitions[
501 		    DISKPART(bp->b_dev)];
502 		blkno += pp->p_offset;
503 	}
504 	bp->b_rawblkno = blkno;
505 
506 #ifdef DEBUG
507 	if (vnddebug & VDB_FOLLOW)
508 		printf("vndstrategy(%p): unit %d\n", bp, unit);
509 #endif
510 	BUFQ_PUT(vnd->sc_tab, bp);
511 	wakeup(&vnd->sc_tab);
512 	splx(s);
513 	return;
514 done:
515 	biodone(bp);
516 	splx(s);
517 }
518 
519 static void
520 vndthread(void *arg)
521 {
522 	struct vnd_softc *vnd = arg;
523 	struct mount *mp;
524 	int s, bsize;
525 	int sz, error;
526 	struct disklabel *lp;
527 
528 	s = splbio();
529 	vnd->sc_flags |= VNF_KTHREAD;
530 	wakeup(&vnd->sc_kthread);
531 
532 	/*
533 	 * Dequeue requests, break them into bsize pieces and submit using
534 	 * VOP_BMAP/VOP_STRATEGY.
535 	 */
536 	while ((vnd->sc_flags & VNF_VUNCONF) == 0) {
537 		struct vndxfer *vnx;
538 		off_t offset;
539 		int resid;
540 		int skipped = 0;
541 		off_t bn;
542 		int flags;
543 		struct buf *obp;
544 		struct buf *bp;
545 
546 		obp = BUFQ_GET(vnd->sc_tab);
547 		if (obp == NULL) {
548 			tsleep(&vnd->sc_tab, PRIBIO, "vndbp", 0);
549 			continue;
550 		};
551 		splx(s);
552 		flags = obp->b_flags;
553 #ifdef DEBUG
554 		if (vnddebug & VDB_FOLLOW)
555 			printf("vndthread(%p\n", obp);
556 #endif
557 		lp = vnd->sc_dkdev.dk_label;
558 
559 		/* convert to a byte offset within the file. */
560 		bn = obp->b_rawblkno * lp->d_secsize;
561 
562 		if (vnd->sc_vp->v_mount == NULL) {
563 			obp->b_error = ENXIO;
564 			obp->b_flags |= B_ERROR;
565 			goto done;
566 		}
567 #ifdef VND_COMPRESSION
568 		/* handle a compressed read */
569 		if ((flags & B_READ) != 0 && (vnd->sc_flags & VNF_COMP)) {
570 			compstrategy(obp, bn);
571 			goto done;
572 		}
573 #endif /* VND_COMPRESSION */
574 
575 		bsize = vnd->sc_vp->v_mount->mnt_stat.f_iosize;
576 
577 		/*
578 		 * Allocate a header for this transfer and link it to the
579 		 * buffer
580 		 */
581 		s = splbio();
582 		vnx = VND_GETXFER(vnd);
583 		splx(s);
584 		vnx->vx_vnd = vnd;
585 
586 		bp = &vnx->vx_buf;
587 		BUF_INIT(bp);
588 		bp->b_flags = (obp->b_flags & B_READ) | B_CALL;
589 		bp->b_iodone = vndiodone;
590 		bp->b_private = obp;
591 		bp->b_vp = vnd->sc_vp;
592 		bp->b_data = obp->b_data;
593 		bp->b_bcount = bp->b_resid = obp->b_bcount;
594 		BIO_COPYPRIO(bp, obp);
595 
596 		s = splbio();
597 		while (vnd->sc_active >= vnd->sc_maxactive) {
598 			tsleep(&vnd->sc_tab, PRIBIO, "vndac", 0);
599 		}
600 		vnd->sc_active++;
601 		splx(s);
602 
603 		if ((flags & B_READ) == 0) {
604 			s = splbio();
605 			V_INCR_NUMOUTPUT(bp->b_vp);
606 			splx(s);
607 
608 			vn_start_write(vnd->sc_vp, &mp, V_WAIT);
609 		}
610 
611 		/* Instrumentation. */
612 		disk_busy(&vnd->sc_dkdev);
613 
614 		/*
615 		 * Feed requests sequentially.
616 		 * We do it this way to keep from flooding NFS servers if we
617 		 * are connected to an NFS file.  This places the burden on
618 		 * the client rather than the server.
619 		 */
620 		error = 0;
621 		for (offset = 0, resid = bp->b_resid; resid;
622 		    resid -= sz, offset += sz) {
623 			struct buf *nbp;
624 			struct vnode *vp;
625 			daddr_t nbn;
626 			int off, nra;
627 
628 			nra = 0;
629 			vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY | LK_CANRECURSE);
630 			error = VOP_BMAP(vnd->sc_vp, bn / bsize, &vp, &nbn, &nra);
631 			VOP_UNLOCK(vnd->sc_vp, 0);
632 
633 			if (error == 0 && (long)nbn == -1)
634 				error = EIO;
635 
636 			/*
637 			 * If there was an error or a hole in the file...punt.
638 			 * Note that we may have to wait for any operations
639 			 * that we have already fired off before releasing
640 			 * the buffer.
641 			 *
642 			 * XXX we could deal with holes here but it would be
643 			 * a hassle (in the write case).
644 			 */
645 			if (error) {
646 				skipped += resid;
647 				break;
648 			}
649 
650 #ifdef DEBUG
651 			if (!dovndcluster)
652 				nra = 0;
653 #endif
654 
655 			off = bn % bsize;
656 			sz = MIN(((off_t)1 + nra) * bsize - off, resid);
657 #ifdef	DEBUG
658 			if (vnddebug & VDB_IO)
659 				printf("vndstrategy: vp %p/%p bn 0x%qx/0x%" PRIx64
660 				       " sz 0x%x\n",
661 				    vnd->sc_vp, vp, (long long)bn, nbn, sz);
662 #endif
663 
664 			nbp = getiobuf();
665 			nestiobuf_setup(bp, nbp, offset, sz);
666 			nbp->b_blkno = nbn + btodb(off);
667 
668 #if 0 /* XXX #ifdef DEBUG */
669 			if (vnddebug & VDB_IO)
670 				printf("vndstart(%ld): bp %p vp %p blkno "
671 				    "0x%" PRIx64 " flags %x addr %p cnt 0x%x\n",
672 				    (long) (vnd-vnd_softc), &nbp->vb_buf,
673 				    nbp->vb_buf.b_vp, nbp->vb_buf.b_blkno,
674 				    nbp->vb_buf.b_flags, nbp->vb_buf.b_data,
675 				    nbp->vb_buf.b_bcount);
676 #endif
677 			VOP_STRATEGY(vp, nbp);
678 			bn += sz;
679 		}
680 		nestiobuf_done(bp, skipped, error);
681 
682 		if ((flags & B_READ) == 0)
683 			vn_finished_write(mp, 0);
684 
685 		s = splbio();
686 		continue;
687 done:
688 		biodone(obp);
689 		s = splbio();
690 	}
691 
692 	vnd->sc_flags &= (~VNF_KTHREAD | VNF_VUNCONF);
693 	wakeup(&vnd->sc_kthread);
694 	splx(s);
695 	kthread_exit(0);
696 }
697 
698 static void
699 vndiodone(struct buf *bp)
700 {
701 	struct vndxfer *vnx = VND_BUFTOXFER(bp);
702 	struct vnd_softc *vnd = vnx->vx_vnd;
703 	struct buf *obp = bp->b_private;
704 
705 	KASSERT(&vnx->vx_buf == bp);
706 	KASSERT(vnd->sc_active > 0);
707 #ifdef DEBUG
708 	if (vnddebug & VDB_IO) {
709 		printf("vndiodone1: bp %p iodone: error %d\n",
710 		    bp, (bp->b_flags & B_ERROR) != 0 ? bp->b_error : 0);
711 	}
712 #endif
713 	disk_unbusy(&vnd->sc_dkdev, bp->b_bcount - bp->b_resid,
714 	    (bp->b_flags & B_READ));
715 	vnd->sc_active--;
716 	if (vnd->sc_active == 0) {
717 		wakeup(&vnd->sc_tab);
718 	}
719 	obp->b_flags |= bp->b_flags & B_ERROR;
720 	obp->b_error = bp->b_error;
721 	obp->b_resid = bp->b_resid;
722 	VND_PUTXFER(vnd, vnx);
723 	biodone(obp);
724 }
725 
726 /* ARGSUSED */
727 static int
728 vndread(dev_t dev, struct uio *uio, int flags)
729 {
730 	int unit = vndunit(dev);
731 	struct vnd_softc *sc;
732 
733 #ifdef DEBUG
734 	if (vnddebug & VDB_FOLLOW)
735 		printf("vndread(0x%x, %p)\n", dev, uio);
736 #endif
737 
738 	sc = device_lookup(&vnd_cd, unit);
739 	if (sc == NULL)
740 		return ENXIO;
741 
742 	if ((sc->sc_flags & VNF_INITED) == 0)
743 		return (ENXIO);
744 
745 	return (physio(vndstrategy, NULL, dev, B_READ, minphys, uio));
746 }
747 
748 /* ARGSUSED */
749 static int
750 vndwrite(dev_t dev, struct uio *uio, int flags)
751 {
752 	int unit = vndunit(dev);
753 	struct vnd_softc *sc;
754 
755 #ifdef DEBUG
756 	if (vnddebug & VDB_FOLLOW)
757 		printf("vndwrite(0x%x, %p)\n", dev, uio);
758 #endif
759 
760 	sc = device_lookup(&vnd_cd, unit);
761 	if (sc == NULL)
762 		return ENXIO;
763 
764 	if ((sc->sc_flags & VNF_INITED) == 0)
765 		return (ENXIO);
766 
767 	return (physio(vndstrategy, NULL, dev, B_WRITE, minphys, uio));
768 }
769 
770 static int
771 vnd_cget(struct lwp *l, int unit, int *un, struct vattr *va)
772 {
773 	struct vnd_softc *vnd;
774 
775 	if (*un == -1)
776 		*un = unit;
777 	if (*un < 0)
778 		return EINVAL;
779 
780 	vnd = device_lookup(&vnd_cd, *un);
781 	if (vnd == NULL)
782 		return (*un >= vnd_cd.cd_ndevs) ? ENXIO : -1;
783 
784 	if ((vnd->sc_flags & VNF_INITED) == 0)
785 		return -1;
786 
787 	return VOP_GETATTR(vnd->sc_vp, va, l->l_proc->p_ucred, l);
788 }
789 
790 /* ARGSUSED */
791 static int
792 vndioctl(dev_t dev, u_long cmd, caddr_t data, int flag, struct lwp *l)
793 {
794 	int unit = vndunit(dev);
795 	struct vnd_softc *vnd;
796 	struct vnd_ioctl *vio;
797 	struct vattr vattr;
798 	struct nameidata nd;
799 	int error, part, pmask;
800 	size_t geomsize;
801 	struct proc *p =  (l != NULL) ? l->l_proc : NULL;
802 	int fflags;
803 #ifdef __HAVE_OLD_DISKLABEL
804 	struct disklabel newlabel;
805 #endif
806 
807 #ifdef DEBUG
808 	if (vnddebug & VDB_FOLLOW)
809 		printf("vndioctl(0x%x, 0x%lx, %p, 0x%x, %p): unit %d\n",
810 		    dev, cmd, data, flag, p, unit);
811 #endif
812 	vnd = device_lookup(&vnd_cd, unit);
813 	if (vnd == NULL &&
814 #ifdef COMPAT_30
815 	    cmd != VNDIOOCGET &&
816 #endif
817 	    cmd != VNDIOCGET)
818 		return ENXIO;
819 	vio = (struct vnd_ioctl *)data;
820 
821 	/* Must be open for writes for these commands... */
822 	switch (cmd) {
823 	case VNDIOCSET:
824 	case VNDIOCCLR:
825 	case DIOCSDINFO:
826 	case DIOCWDINFO:
827 #ifdef __HAVE_OLD_DISKLABEL
828 	case ODIOCSDINFO:
829 	case ODIOCWDINFO:
830 #endif
831 	case DIOCKLABEL:
832 	case DIOCWLABEL:
833 		if ((flag & FWRITE) == 0)
834 			return (EBADF);
835 	}
836 
837 	/* Must be initialized for these... */
838 	switch (cmd) {
839 	case VNDIOCCLR:
840 	case DIOCGDINFO:
841 	case DIOCSDINFO:
842 	case DIOCWDINFO:
843 	case DIOCGPART:
844 	case DIOCKLABEL:
845 	case DIOCWLABEL:
846 	case DIOCGDEFLABEL:
847 #ifdef __HAVE_OLD_DISKLABEL
848 	case ODIOCGDINFO:
849 	case ODIOCSDINFO:
850 	case ODIOCWDINFO:
851 	case ODIOCGDEFLABEL:
852 #endif
853 		if ((vnd->sc_flags & VNF_INITED) == 0)
854 			return (ENXIO);
855 	}
856 
857 	switch (cmd) {
858 	case VNDIOCSET:
859 		if (vnd->sc_flags & VNF_INITED)
860 			return (EBUSY);
861 
862 		if ((error = vndlock(vnd)) != 0)
863 			return (error);
864 
865 		fflags = FREAD;
866 		if ((vio->vnd_flags & VNDIOF_READONLY) == 0)
867 			fflags |= FWRITE;
868 		NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, vio->vnd_file, l);
869 		if ((error = vn_open(&nd, fflags, 0)) != 0)
870 			goto unlock_and_exit;
871 		error = VOP_GETATTR(nd.ni_vp, &vattr, l->l_proc->p_ucred, l);
872 		if (!error && nd.ni_vp->v_type != VREG)
873 			error = EOPNOTSUPP;
874 		if (error) {
875 			VOP_UNLOCK(nd.ni_vp, 0);
876 			goto close_and_exit;
877 		}
878 
879 		/* If using a compressed file, initialize its info */
880 		/* (or abort with an error if kernel has no compression) */
881 		if (vio->vnd_flags & VNF_COMP) {
882 #ifdef VND_COMPRESSION
883 			struct vnd_comp_header *ch;
884 			int i;
885 			u_int32_t comp_size;
886 			u_int32_t comp_maxsize;
887 
888 			/* allocate space for compresed file header */
889 			ch = malloc(sizeof(struct vnd_comp_header),
890 			M_TEMP, M_WAITOK);
891 
892 			/* read compressed file header */
893 			error = vn_rdwr(UIO_READ, nd.ni_vp, (caddr_t)ch,
894 			  sizeof(struct vnd_comp_header), 0, UIO_SYSSPACE,
895 			  IO_UNIT|IO_NODELOCKED, p->p_ucred, NULL, NULL);
896 			if(error) {
897 				free(ch, M_TEMP);
898 				VOP_UNLOCK(nd.ni_vp, 0);
899 				goto close_and_exit;
900 			}
901 
902 			/* save some header info */
903 			vnd->sc_comp_blksz = ntohl(ch->block_size);
904 			/* note last offset is the file byte size */
905 			vnd->sc_comp_numoffs = ntohl(ch->num_blocks)+1;
906 			free(ch, M_TEMP);
907 			if(vnd->sc_comp_blksz % DEV_BSIZE !=0) {
908 				VOP_UNLOCK(nd.ni_vp, 0);
909 				error = EINVAL;
910 				goto close_and_exit;
911 			}
912 			if(sizeof(struct vnd_comp_header) +
913 			  sizeof(u_int64_t) * vnd->sc_comp_numoffs >
914 			  vattr.va_size) {
915 				VOP_UNLOCK(nd.ni_vp, 0);
916 				error = EINVAL;
917 				goto close_and_exit;
918 			}
919 
920 			/* set decompressed file size */
921 			vattr.va_size =
922 			  (vnd->sc_comp_numoffs - 1) * vnd->sc_comp_blksz;
923 
924 			/* allocate space for all the compressed offsets */
925 			vnd->sc_comp_offsets =
926 			malloc(sizeof(u_int64_t) * vnd->sc_comp_numoffs,
927 			M_DEVBUF, M_WAITOK);
928 
929 			/* read in the offsets */
930 			error = vn_rdwr(UIO_READ, nd.ni_vp,
931 			  (caddr_t)vnd->sc_comp_offsets,
932 			  sizeof(u_int64_t) * vnd->sc_comp_numoffs,
933 			  sizeof(struct vnd_comp_header), UIO_SYSSPACE,
934 			  IO_UNIT|IO_NODELOCKED, p->p_ucred, NULL, NULL);
935 			if(error) {
936 				VOP_UNLOCK(nd.ni_vp, 0);
937 				goto close_and_exit;
938 			}
939 			/*
940 			 * find largest block size (used for allocation limit).
941 			 * Also convert offset to native byte order.
942 			 */
943 			comp_maxsize = 0;
944 			for (i = 0; i < vnd->sc_comp_numoffs - 1; i++) {
945 				vnd->sc_comp_offsets[i] =
946 				  be64toh(vnd->sc_comp_offsets[i]);
947 				comp_size = be64toh(vnd->sc_comp_offsets[i + 1])
948 				  - vnd->sc_comp_offsets[i];
949 				if (comp_size > comp_maxsize)
950 					comp_maxsize = comp_size;
951 			}
952 			vnd->sc_comp_offsets[vnd->sc_comp_numoffs - 1] =
953 			  be64toh(vnd->sc_comp_offsets[vnd->sc_comp_numoffs - 1]);
954 
955 			/* create compressed data buffer */
956 			vnd->sc_comp_buff = malloc(comp_maxsize,
957 			  M_DEVBUF, M_WAITOK);
958 
959 			/* create decompressed buffer */
960 			vnd->sc_comp_decombuf = malloc(vnd->sc_comp_blksz,
961 			  M_DEVBUF, M_WAITOK);
962 			vnd->sc_comp_buffblk = -1;
963 
964 			/* Initialize decompress stream */
965 			bzero(&vnd->sc_comp_stream, sizeof(z_stream));
966 			vnd->sc_comp_stream.zalloc = vnd_alloc;
967 			vnd->sc_comp_stream.zfree = vnd_free;
968 			error = inflateInit2(&vnd->sc_comp_stream, MAX_WBITS);
969 			if(error) {
970 				if(vnd->sc_comp_stream.msg)
971 					printf("vnd%d: compressed file, %s\n",
972 					  unit, vnd->sc_comp_stream.msg);
973 				VOP_UNLOCK(nd.ni_vp, 0);
974 				error = EINVAL;
975 				goto close_and_exit;
976 			}
977 
978 			vnd->sc_flags |= VNF_COMP | VNF_READONLY;
979 #else /* !VND_COMPRESSION */
980 			VOP_UNLOCK(nd.ni_vp, 0);
981 			error = EOPNOTSUPP;
982 			goto close_and_exit;
983 #endif /* VND_COMPRESSION */
984 		}
985 
986 		VOP_UNLOCK(nd.ni_vp, 0);
987 		vnd->sc_vp = nd.ni_vp;
988 		vnd->sc_size = btodb(vattr.va_size);	/* note truncation */
989 
990 		/*
991 		 * Use pseudo-geometry specified.  If none was provided,
992 		 * use "standard" Adaptec fictitious geometry.
993 		 */
994 		if (vio->vnd_flags & VNDIOF_HASGEOM) {
995 
996 			memcpy(&vnd->sc_geom, &vio->vnd_geom,
997 			    sizeof(vio->vnd_geom));
998 
999 			/*
1000 			 * Sanity-check the sector size.
1001 			 * XXX Don't allow secsize < DEV_BSIZE.	 Should
1002 			 * XXX we?
1003 			 */
1004 			if (vnd->sc_geom.vng_secsize < DEV_BSIZE ||
1005 			    (vnd->sc_geom.vng_secsize % DEV_BSIZE) != 0 ||
1006 			    vnd->sc_geom.vng_ncylinders == 0 ||
1007 			    (vnd->sc_geom.vng_ntracks *
1008 			     vnd->sc_geom.vng_nsectors) == 0) {
1009 				error = EINVAL;
1010 				goto close_and_exit;
1011 			}
1012 
1013 			/*
1014 			 * Compute the size (in DEV_BSIZE blocks) specified
1015 			 * by the geometry.
1016 			 */
1017 			geomsize = (vnd->sc_geom.vng_nsectors *
1018 			    vnd->sc_geom.vng_ntracks *
1019 			    vnd->sc_geom.vng_ncylinders) *
1020 			    (vnd->sc_geom.vng_secsize / DEV_BSIZE);
1021 
1022 			/*
1023 			 * Sanity-check the size against the specified
1024 			 * geometry.
1025 			 */
1026 			if (vnd->sc_size < geomsize) {
1027 				error = EINVAL;
1028 				goto close_and_exit;
1029 			}
1030 		} else if (vnd->sc_size >= (32 * 64)) {
1031 			/*
1032 			 * Size must be at least 2048 DEV_BSIZE blocks
1033 			 * (1M) in order to use this geometry.
1034 			 */
1035 			vnd->sc_geom.vng_secsize = DEV_BSIZE;
1036 			vnd->sc_geom.vng_nsectors = 32;
1037 			vnd->sc_geom.vng_ntracks = 64;
1038 			vnd->sc_geom.vng_ncylinders = vnd->sc_size / (64 * 32);
1039 		} else {
1040 			vnd->sc_geom.vng_secsize = DEV_BSIZE;
1041 			vnd->sc_geom.vng_nsectors = 1;
1042 			vnd->sc_geom.vng_ntracks = 1;
1043 			vnd->sc_geom.vng_ncylinders = vnd->sc_size;
1044 		}
1045 
1046 		if (vio->vnd_flags & VNDIOF_READONLY) {
1047 			vnd->sc_flags |= VNF_READONLY;
1048 		}
1049 
1050 		if ((error = vndsetcred(vnd, p->p_ucred)) != 0)
1051 			goto close_and_exit;
1052 
1053 		vndthrottle(vnd, vnd->sc_vp);
1054 		vio->vnd_size = dbtob(vnd->sc_size);
1055 		vnd->sc_flags |= VNF_INITED;
1056 
1057 		/* create the kernel thread, wait for it to be up */
1058 		error = kthread_create1(vndthread, vnd, &vnd->sc_kthread,
1059 		    vnd->sc_dev.dv_xname);
1060 		if (error)
1061 			goto close_and_exit;
1062 		while ((vnd->sc_flags & VNF_KTHREAD) == 0) {
1063 			tsleep(&vnd->sc_kthread, PRIBIO, "vndthr", 0);
1064 		}
1065 #ifdef DEBUG
1066 		if (vnddebug & VDB_INIT)
1067 			printf("vndioctl: SET vp %p size 0x%lx %d/%d/%d/%d\n",
1068 			    vnd->sc_vp, (unsigned long) vnd->sc_size,
1069 			    vnd->sc_geom.vng_secsize,
1070 			    vnd->sc_geom.vng_nsectors,
1071 			    vnd->sc_geom.vng_ntracks,
1072 			    vnd->sc_geom.vng_ncylinders);
1073 #endif
1074 
1075 		/* Attach the disk. */
1076 		vnd->sc_dkdev.dk_name = vnd->sc_dev.dv_xname;
1077 		pseudo_disk_attach(&vnd->sc_dkdev);
1078 
1079 		/* Initialize the xfer and buffer pools. */
1080 		pool_init(&vnd->sc_vxpool, sizeof(struct vndxfer), 0,
1081 		    0, 0, "vndxpl", NULL);
1082 
1083 		/* Try and read the disklabel. */
1084 		vndgetdisklabel(dev, vnd);
1085 
1086 		vndunlock(vnd);
1087 
1088 		break;
1089 
1090 close_and_exit:
1091 		(void) vn_close(nd.ni_vp, fflags, p->p_ucred, l);
1092 unlock_and_exit:
1093 #ifdef VND_COMPRESSION
1094 		/* free any allocated memory (for compressed file) */
1095 		if(vnd->sc_comp_offsets) {
1096 			free(vnd->sc_comp_offsets, M_DEVBUF);
1097 			vnd->sc_comp_offsets = NULL;
1098 		}
1099 		if(vnd->sc_comp_buff) {
1100 			free(vnd->sc_comp_buff, M_DEVBUF);
1101 			vnd->sc_comp_buff = NULL;
1102 		}
1103 		if(vnd->sc_comp_decombuf) {
1104 			free(vnd->sc_comp_decombuf, M_DEVBUF);
1105 			vnd->sc_comp_decombuf = NULL;
1106 		}
1107 #endif /* VND_COMPRESSION */
1108 		vndunlock(vnd);
1109 		return (error);
1110 
1111 	case VNDIOCCLR:
1112 		if ((error = vndlock(vnd)) != 0)
1113 			return (error);
1114 
1115 		/*
1116 		 * Don't unconfigure if any other partitions are open
1117 		 * or if both the character and block flavors of this
1118 		 * partition are open.
1119 		 */
1120 		part = DISKPART(dev);
1121 		pmask = (1 << part);
1122 		if (((vnd->sc_dkdev.dk_openmask & ~pmask) ||
1123 		    ((vnd->sc_dkdev.dk_bopenmask & pmask) &&
1124 		    (vnd->sc_dkdev.dk_copenmask & pmask))) &&
1125 			!(vio->vnd_flags & VNDIOF_FORCE)) {
1126 			vndunlock(vnd);
1127 			return (EBUSY);
1128 		}
1129 
1130 		/*
1131 		 * XXX vndclear() might call vndclose() implicitely;
1132 		 * release lock to avoid recursion
1133 		 */
1134 		vndunlock(vnd);
1135 		vndclear(vnd, minor(dev));
1136 #ifdef DEBUG
1137 		if (vnddebug & VDB_INIT)
1138 			printf("vndioctl: CLRed\n");
1139 #endif
1140 
1141 		/* Destroy the xfer and buffer pools. */
1142 		pool_destroy(&vnd->sc_vxpool);
1143 
1144 		/* Detatch the disk. */
1145 		pseudo_disk_detach(&vnd->sc_dkdev);
1146 		break;
1147 
1148 #ifdef COMPAT_30
1149 	case VNDIOOCGET: {
1150 		struct vnd_ouser *vnu;
1151 		struct vattr va;
1152 		vnu = (struct vnd_ouser *)data;
1153 		switch (error = vnd_cget(l, unit, &vnu->vnu_unit, &va)) {
1154 		case 0:
1155 			vnu->vnu_dev = va.va_fsid;
1156 			vnu->vnu_ino = va.va_fileid;
1157 			break;
1158 		case -1:
1159 			/* unused is not an error */
1160 			vnu->vnu_dev = 0;
1161 			vnu->vnu_ino = 0;
1162 			break;
1163 		default:
1164 			return error;
1165 		}
1166 		break;
1167 	}
1168 #endif
1169 	case VNDIOCGET: {
1170 		struct vnd_user *vnu;
1171 		struct vattr va;
1172 		vnu = (struct vnd_user *)data;
1173 		switch (error = vnd_cget(l, unit, &vnu->vnu_unit, &va)) {
1174 		case 0:
1175 			vnu->vnu_dev = va.va_fsid;
1176 			vnu->vnu_ino = va.va_fileid;
1177 			break;
1178 		case -1:
1179 			/* unused is not an error */
1180 			vnu->vnu_dev = 0;
1181 			vnu->vnu_ino = 0;
1182 			break;
1183 		default:
1184 			return error;
1185 		}
1186 		break;
1187 	}
1188 
1189 	case DIOCGDINFO:
1190 		*(struct disklabel *)data = *(vnd->sc_dkdev.dk_label);
1191 		break;
1192 
1193 #ifdef __HAVE_OLD_DISKLABEL
1194 	case ODIOCGDINFO:
1195 		newlabel = *(vnd->sc_dkdev.dk_label);
1196 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1197 			return ENOTTY;
1198 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1199 		break;
1200 #endif
1201 
1202 	case DIOCGPART:
1203 		((struct partinfo *)data)->disklab = vnd->sc_dkdev.dk_label;
1204 		((struct partinfo *)data)->part =
1205 		    &vnd->sc_dkdev.dk_label->d_partitions[DISKPART(dev)];
1206 		break;
1207 
1208 	case DIOCWDINFO:
1209 	case DIOCSDINFO:
1210 #ifdef __HAVE_OLD_DISKLABEL
1211 	case ODIOCWDINFO:
1212 	case ODIOCSDINFO:
1213 #endif
1214 	{
1215 		struct disklabel *lp;
1216 
1217 		if ((error = vndlock(vnd)) != 0)
1218 			return (error);
1219 
1220 		vnd->sc_flags |= VNF_LABELLING;
1221 
1222 #ifdef __HAVE_OLD_DISKLABEL
1223 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
1224 			memset(&newlabel, 0, sizeof newlabel);
1225 			memcpy(&newlabel, data, sizeof (struct olddisklabel));
1226 			lp = &newlabel;
1227 		} else
1228 #endif
1229 		lp = (struct disklabel *)data;
1230 
1231 		error = setdisklabel(vnd->sc_dkdev.dk_label,
1232 		    lp, 0, vnd->sc_dkdev.dk_cpulabel);
1233 		if (error == 0) {
1234 			if (cmd == DIOCWDINFO
1235 #ifdef __HAVE_OLD_DISKLABEL
1236 			    || cmd == ODIOCWDINFO
1237 #endif
1238 			   )
1239 				error = writedisklabel(VNDLABELDEV(dev),
1240 				    vndstrategy, vnd->sc_dkdev.dk_label,
1241 				    vnd->sc_dkdev.dk_cpulabel);
1242 		}
1243 
1244 		vnd->sc_flags &= ~VNF_LABELLING;
1245 
1246 		vndunlock(vnd);
1247 
1248 		if (error)
1249 			return (error);
1250 		break;
1251 	}
1252 
1253 	case DIOCKLABEL:
1254 		if (*(int *)data != 0)
1255 			vnd->sc_flags |= VNF_KLABEL;
1256 		else
1257 			vnd->sc_flags &= ~VNF_KLABEL;
1258 		break;
1259 
1260 	case DIOCWLABEL:
1261 		if (*(int *)data != 0)
1262 			vnd->sc_flags |= VNF_WLABEL;
1263 		else
1264 			vnd->sc_flags &= ~VNF_WLABEL;
1265 		break;
1266 
1267 	case DIOCGDEFLABEL:
1268 		vndgetdefaultlabel(vnd, (struct disklabel *)data);
1269 		break;
1270 
1271 #ifdef __HAVE_OLD_DISKLABEL
1272 	case ODIOCGDEFLABEL:
1273 		vndgetdefaultlabel(vnd, &newlabel);
1274 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1275 			return ENOTTY;
1276 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1277 		break;
1278 #endif
1279 
1280 	default:
1281 		return (ENOTTY);
1282 	}
1283 
1284 	return (0);
1285 }
1286 
1287 /*
1288  * Duplicate the current processes' credentials.  Since we are called only
1289  * as the result of a SET ioctl and only root can do that, any future access
1290  * to this "disk" is essentially as root.  Note that credentials may change
1291  * if some other uid can write directly to the mapped file (NFS).
1292  */
1293 static int
1294 vndsetcred(struct vnd_softc *vnd, struct ucred *cred)
1295 {
1296 	struct uio auio;
1297 	struct iovec aiov;
1298 	char *tmpbuf;
1299 	int error;
1300 
1301 	vnd->sc_cred = crdup(cred);
1302 	tmpbuf = malloc(DEV_BSIZE, M_TEMP, M_WAITOK);
1303 
1304 	/* XXX: Horrible kludge to establish credentials for NFS */
1305 	aiov.iov_base = tmpbuf;
1306 	aiov.iov_len = min(DEV_BSIZE, dbtob(vnd->sc_size));
1307 	auio.uio_iov = &aiov;
1308 	auio.uio_iovcnt = 1;
1309 	auio.uio_offset = 0;
1310 	auio.uio_rw = UIO_READ;
1311 	auio.uio_segflg = UIO_SYSSPACE;
1312 	auio.uio_resid = aiov.iov_len;
1313 	vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1314 	error = VOP_READ(vnd->sc_vp, &auio, 0, vnd->sc_cred);
1315 	if (error == 0) {
1316 		/*
1317 		 * Because vnd does all IO directly through the vnode
1318 		 * we need to flush (at least) the buffer from the above
1319 		 * VOP_READ from the buffer cache to prevent cache
1320 		 * incoherencies.  Also, be careful to write dirty
1321 		 * buffers back to stable storage.
1322 		 */
1323 		error = vinvalbuf(vnd->sc_vp, V_SAVE, vnd->sc_cred,
1324 			    curlwp, 0, 0);
1325 	}
1326 	VOP_UNLOCK(vnd->sc_vp, 0);
1327 
1328 	free(tmpbuf, M_TEMP);
1329 	return (error);
1330 }
1331 
1332 /*
1333  * Set maxactive based on FS type
1334  */
1335 static void
1336 vndthrottle(struct vnd_softc *vnd, struct vnode *vp)
1337 {
1338 #ifdef NFS
1339 	extern int (**nfsv2_vnodeop_p)(void *);
1340 
1341 	if (vp->v_op == nfsv2_vnodeop_p)
1342 		vnd->sc_maxactive = 2;
1343 	else
1344 #endif
1345 		vnd->sc_maxactive = 8;
1346 
1347 	if (vnd->sc_maxactive < 1)
1348 		vnd->sc_maxactive = 1;
1349 }
1350 
1351 #if 0
1352 static void
1353 vndshutdown(void)
1354 {
1355 	struct vnd_softc *vnd;
1356 
1357 	for (vnd = &vnd_softc[0]; vnd < &vnd_softc[numvnd]; vnd++)
1358 		if (vnd->sc_flags & VNF_INITED)
1359 			vndclear(vnd);
1360 }
1361 #endif
1362 
1363 static void
1364 vndclear(struct vnd_softc *vnd, int myminor)
1365 {
1366 	struct vnode *vp = vnd->sc_vp;
1367 	struct lwp *l = curlwp;
1368 	int fflags = FREAD;
1369 	int bmaj, cmaj, i, mn;
1370 	int s;
1371 
1372 #ifdef DEBUG
1373 	if (vnddebug & VDB_FOLLOW)
1374 		printf("vndclear(%p): vp %p\n", vnd, vp);
1375 #endif
1376 	/* locate the major number */
1377 	bmaj = bdevsw_lookup_major(&vnd_bdevsw);
1378 	cmaj = cdevsw_lookup_major(&vnd_cdevsw);
1379 
1380 	/* Nuke the vnodes for any open instances */
1381 	for (i = 0; i < MAXPARTITIONS; i++) {
1382 		mn = DISKMINOR(vnd->sc_dev.dv_unit, i);
1383 		vdevgone(bmaj, mn, mn, VBLK);
1384 		if (mn != myminor) /* XXX avoid to kill own vnode */
1385 			vdevgone(cmaj, mn, mn, VCHR);
1386 	}
1387 
1388 	if ((vnd->sc_flags & VNF_READONLY) == 0)
1389 		fflags |= FWRITE;
1390 
1391 	s = splbio();
1392 	bufq_drain(vnd->sc_tab);
1393 	splx(s);
1394 
1395 	vnd->sc_flags |= VNF_VUNCONF;
1396 	wakeup(&vnd->sc_tab);
1397 	while (vnd->sc_flags & VNF_KTHREAD)
1398 		tsleep(&vnd->sc_kthread, PRIBIO, "vnthr", 0);
1399 
1400 #ifdef VND_COMPRESSION
1401 	/* free the compressed file buffers */
1402 	if(vnd->sc_flags & VNF_COMP) {
1403 		if(vnd->sc_comp_offsets) {
1404 			free(vnd->sc_comp_offsets, M_DEVBUF);
1405 			vnd->sc_comp_offsets = NULL;
1406 		}
1407 		if(vnd->sc_comp_buff) {
1408 			free(vnd->sc_comp_buff, M_DEVBUF);
1409 			vnd->sc_comp_buff = NULL;
1410 		}
1411 		if(vnd->sc_comp_decombuf) {
1412 			free(vnd->sc_comp_decombuf, M_DEVBUF);
1413 			vnd->sc_comp_decombuf = NULL;
1414 		}
1415 	}
1416 #endif /* VND_COMPRESSION */
1417 	vnd->sc_flags &=
1418 	    ~(VNF_INITED | VNF_READONLY | VNF_VLABEL
1419 	      | VNF_VUNCONF | VNF_COMP);
1420 	if (vp == (struct vnode *)0)
1421 		panic("vndclear: null vp");
1422 	(void) vn_close(vp, fflags, vnd->sc_cred, l);
1423 	crfree(vnd->sc_cred);
1424 	vnd->sc_vp = (struct vnode *)0;
1425 	vnd->sc_cred = (struct ucred *)0;
1426 	vnd->sc_size = 0;
1427 }
1428 
1429 static int
1430 vndsize(dev_t dev)
1431 {
1432 	struct vnd_softc *sc;
1433 	struct disklabel *lp;
1434 	int part, unit, omask;
1435 	int size;
1436 
1437 	unit = vndunit(dev);
1438 	sc = (struct vnd_softc *)device_lookup(&vnd_cd, unit);
1439 	if (sc == NULL)
1440 		return -1;
1441 
1442 	if ((sc->sc_flags & VNF_INITED) == 0)
1443 		return (-1);
1444 
1445 	part = DISKPART(dev);
1446 	omask = sc->sc_dkdev.dk_openmask & (1 << part);
1447 	lp = sc->sc_dkdev.dk_label;
1448 
1449 	if (omask == 0 && vndopen(dev, 0, S_IFBLK, curlwp))	/* XXX */
1450 		return (-1);
1451 
1452 	if (lp->d_partitions[part].p_fstype != FS_SWAP)
1453 		size = -1;
1454 	else
1455 		size = lp->d_partitions[part].p_size *
1456 		    (lp->d_secsize / DEV_BSIZE);
1457 
1458 	if (omask == 0 && vndclose(dev, 0, S_IFBLK, curlwp))	/* XXX */
1459 		return (-1);
1460 
1461 	return (size);
1462 }
1463 
1464 static int
1465 vnddump(dev_t dev, daddr_t blkno, caddr_t va, size_t size)
1466 {
1467 
1468 	/* Not implemented. */
1469 	return ENXIO;
1470 }
1471 
1472 static void
1473 vndgetdefaultlabel(struct vnd_softc *sc, struct disklabel *lp)
1474 {
1475 	struct vndgeom *vng = &sc->sc_geom;
1476 	struct partition *pp;
1477 
1478 	memset(lp, 0, sizeof(*lp));
1479 
1480 	lp->d_secperunit = sc->sc_size / (vng->vng_secsize / DEV_BSIZE);
1481 	lp->d_secsize = vng->vng_secsize;
1482 	lp->d_nsectors = vng->vng_nsectors;
1483 	lp->d_ntracks = vng->vng_ntracks;
1484 	lp->d_ncylinders = vng->vng_ncylinders;
1485 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1486 
1487 	strncpy(lp->d_typename, "vnd", sizeof(lp->d_typename));
1488 	lp->d_type = DTYPE_VND;
1489 	strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
1490 	lp->d_rpm = 3600;
1491 	lp->d_interleave = 1;
1492 	lp->d_flags = 0;
1493 
1494 	pp = &lp->d_partitions[RAW_PART];
1495 	pp->p_offset = 0;
1496 	pp->p_size = lp->d_secperunit;
1497 	pp->p_fstype = FS_UNUSED;
1498 	lp->d_npartitions = RAW_PART + 1;
1499 
1500 	lp->d_magic = DISKMAGIC;
1501 	lp->d_magic2 = DISKMAGIC;
1502 	lp->d_checksum = dkcksum(lp);
1503 }
1504 
1505 /*
1506  * Read the disklabel from a vnd.  If one is not present, create a fake one.
1507  */
1508 static void
1509 vndgetdisklabel(dev_t dev, struct vnd_softc *sc)
1510 {
1511 	const char *errstring;
1512 	struct disklabel *lp = sc->sc_dkdev.dk_label;
1513 	struct cpu_disklabel *clp = sc->sc_dkdev.dk_cpulabel;
1514 	int i;
1515 
1516 	memset(clp, 0, sizeof(*clp));
1517 
1518 	vndgetdefaultlabel(sc, lp);
1519 
1520 	/*
1521 	 * Call the generic disklabel extraction routine.
1522 	 */
1523 	errstring = readdisklabel(VNDLABELDEV(dev), vndstrategy, lp, clp);
1524 	if (errstring) {
1525 		/*
1526 		 * Lack of disklabel is common, but we print the warning
1527 		 * anyway, since it might contain other useful information.
1528 		 */
1529 		printf("%s: %s\n", sc->sc_dev.dv_xname, errstring);
1530 
1531 		/*
1532 		 * For historical reasons, if there's no disklabel
1533 		 * present, all partitions must be FS_BSDFFS and
1534 		 * occupy the entire disk.
1535 		 */
1536 		for (i = 0; i < MAXPARTITIONS; i++) {
1537 			/*
1538 			 * Don't wipe out port specific hack (such as
1539 			 * dos partition hack of i386 port).
1540 			 */
1541 			if (lp->d_partitions[i].p_size != 0)
1542 				continue;
1543 
1544 			lp->d_partitions[i].p_size = lp->d_secperunit;
1545 			lp->d_partitions[i].p_offset = 0;
1546 			lp->d_partitions[i].p_fstype = FS_BSDFFS;
1547 		}
1548 
1549 		strncpy(lp->d_packname, "default label",
1550 		    sizeof(lp->d_packname));
1551 
1552 		lp->d_npartitions = MAXPARTITIONS;
1553 		lp->d_checksum = dkcksum(lp);
1554 	}
1555 
1556 	/* In-core label now valid. */
1557 	sc->sc_flags |= VNF_VLABEL;
1558 }
1559 
1560 /*
1561  * Wait interruptibly for an exclusive lock.
1562  *
1563  * XXX
1564  * Several drivers do this; it should be abstracted and made MP-safe.
1565  */
1566 static int
1567 vndlock(struct vnd_softc *sc)
1568 {
1569 	int error;
1570 
1571 	while ((sc->sc_flags & VNF_LOCKED) != 0) {
1572 		sc->sc_flags |= VNF_WANTED;
1573 		if ((error = tsleep(sc, PRIBIO | PCATCH, "vndlck", 0)) != 0)
1574 			return (error);
1575 	}
1576 	sc->sc_flags |= VNF_LOCKED;
1577 	return (0);
1578 }
1579 
1580 /*
1581  * Unlock and wake up any waiters.
1582  */
1583 static void
1584 vndunlock(struct vnd_softc *sc)
1585 {
1586 
1587 	sc->sc_flags &= ~VNF_LOCKED;
1588 	if ((sc->sc_flags & VNF_WANTED) != 0) {
1589 		sc->sc_flags &= ~VNF_WANTED;
1590 		wakeup(sc);
1591 	}
1592 }
1593 
1594 #ifdef VND_COMPRESSION
1595 /* compressed file read */
1596 static void
1597 compstrategy(struct buf *bp, off_t bn)
1598 {
1599 	int error;
1600 	int unit = vndunit(bp->b_dev);
1601 	struct vnd_softc *vnd =
1602 	    (struct vnd_softc *)device_lookup(&vnd_cd, unit);
1603 	u_int32_t comp_block;
1604 	struct uio auio;
1605 	caddr_t addr;
1606 	int s;
1607 
1608 	/* set up constants for data move */
1609 	auio.uio_rw = UIO_READ;
1610 	auio.uio_segflg = UIO_SYSSPACE;
1611 
1612 	/* read, and transfer the data */
1613 	addr = bp->b_data;
1614 	s = splbio();
1615 	while (bp->b_resid > 0) {
1616 		unsigned length;
1617 		size_t length_in_buffer;
1618 		u_int32_t offset_in_buffer;
1619 		struct iovec aiov;
1620 
1621 		/* calculate the compressed block number */
1622 		comp_block = bn / (off_t)vnd->sc_comp_blksz;
1623 
1624 		/* check for good block number */
1625 		if (comp_block >= vnd->sc_comp_numoffs) {
1626 			bp->b_error = EINVAL;
1627 			bp->b_flags |= B_ERROR;
1628 			splx(s);
1629 			return;
1630 		}
1631 
1632 		/* read in the compressed block, if not in buffer */
1633 		if (comp_block != vnd->sc_comp_buffblk) {
1634 			length = vnd->sc_comp_offsets[comp_block + 1] -
1635 			    vnd->sc_comp_offsets[comp_block];
1636 			vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1637 			error = vn_rdwr(UIO_READ, vnd->sc_vp, vnd->sc_comp_buff,
1638 			    length, vnd->sc_comp_offsets[comp_block],
1639 			    UIO_SYSSPACE, IO_UNIT, vnd->sc_cred, NULL, NULL);
1640 			if (error) {
1641 				bp->b_error = error;
1642 				bp->b_flags |= B_ERROR;
1643 				VOP_UNLOCK(vnd->sc_vp, 0);
1644 				splx(s);
1645 				return;
1646 			}
1647 			/* uncompress the buffer */
1648 			vnd->sc_comp_stream.next_in = vnd->sc_comp_buff;
1649 			vnd->sc_comp_stream.avail_in = length;
1650 			vnd->sc_comp_stream.next_out = vnd->sc_comp_decombuf;
1651 			vnd->sc_comp_stream.avail_out = vnd->sc_comp_blksz;
1652 			inflateReset(&vnd->sc_comp_stream);
1653 			error = inflate(&vnd->sc_comp_stream, Z_FINISH);
1654 			if (error != Z_STREAM_END) {
1655 				if (vnd->sc_comp_stream.msg)
1656 					printf("%s: compressed file, %s\n",
1657 					    vnd->sc_dev.dv_xname,
1658 					    vnd->sc_comp_stream.msg);
1659 				bp->b_error = EBADMSG;
1660 				bp->b_flags |= B_ERROR;
1661 				VOP_UNLOCK(vnd->sc_vp, 0);
1662 				splx(s);
1663 				return;
1664 			}
1665 			vnd->sc_comp_buffblk = comp_block;
1666 			VOP_UNLOCK(vnd->sc_vp, 0);
1667 		}
1668 
1669 		/* transfer the usable uncompressed data */
1670 		offset_in_buffer = bn % (off_t)vnd->sc_comp_blksz;
1671 		length_in_buffer = vnd->sc_comp_blksz - offset_in_buffer;
1672 		if (length_in_buffer > bp->b_resid)
1673 			length_in_buffer = bp->b_resid;
1674 		auio.uio_iov = &aiov;
1675 		auio.uio_iovcnt = 1;
1676 		aiov.iov_base = addr;
1677 		aiov.iov_len = length_in_buffer;
1678 		auio.uio_resid = aiov.iov_len;
1679 		auio.uio_offset = 0;
1680 		error = uiomove(vnd->sc_comp_decombuf + offset_in_buffer,
1681 		    length_in_buffer, &auio);
1682 		if (error) {
1683 			bp->b_error = error;
1684 			bp->b_flags |= B_ERROR;
1685 			splx(s);
1686 			return;
1687 		}
1688 
1689 		bn += length_in_buffer;
1690 		addr += length_in_buffer;
1691 		bp->b_resid -= length_in_buffer;
1692 	}
1693 	splx(s);
1694 }
1695 
1696 /* compression memory allocation routines */
1697 static void *
1698 vnd_alloc(void *aux, u_int items, u_int siz)
1699 {
1700 	return malloc(items * siz, M_TEMP, M_NOWAIT);
1701 }
1702 
1703 static void
1704 vnd_free(void *aux, void *ptr)
1705 {
1706 	free(ptr, M_TEMP);
1707 }
1708 #endif /* VND_COMPRESSION */
1709