xref: /netbsd-src/sys/dev/vnd.c (revision d909946ca08dceb44d7d0f22ec9488679695d976)
1 /*	$NetBSD: vnd.c,v 1.258 2016/08/05 08:21:24 pgoyette Exp $	*/
2 
3 /*-
4  * Copyright (c) 1996, 1997, 1998, 2008 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  *
19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29  * POSSIBILITY OF SUCH DAMAGE.
30  */
31 
32 /*
33  * Copyright (c) 1988 University of Utah.
34  * Copyright (c) 1990, 1993
35  *	The Regents of the University of California.  All rights reserved.
36  *
37  * This code is derived from software contributed to Berkeley by
38  * the Systems Programming Group of the University of Utah Computer
39  * Science Department.
40  *
41  * Redistribution and use in source and binary forms, with or without
42  * modification, are permitted provided that the following conditions
43  * are met:
44  * 1. Redistributions of source code must retain the above copyright
45  *    notice, this list of conditions and the following disclaimer.
46  * 2. Redistributions in binary form must reproduce the above copyright
47  *    notice, this list of conditions and the following disclaimer in the
48  *    documentation and/or other materials provided with the distribution.
49  * 3. Neither the name of the University nor the names of its contributors
50  *    may be used to endorse or promote products derived from this software
51  *    without specific prior written permission.
52  *
53  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
54  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
55  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
56  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
57  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
58  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
59  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
60  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
61  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
62  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
63  * SUCH DAMAGE.
64  *
65  * from: Utah $Hdr: vn.c 1.13 94/04/02$
66  *
67  *	@(#)vn.c	8.9 (Berkeley) 5/14/95
68  */
69 
70 /*
71  * Vnode disk driver.
72  *
73  * Block/character interface to a vnode.  Allows one to treat a file
74  * as a disk (e.g. build a filesystem in it, mount it, etc.).
75  *
76  * NOTE 1: If the vnode supports the VOP_BMAP and VOP_STRATEGY operations,
77  * this uses them to avoid distorting the local buffer cache.  If those
78  * block-level operations are not available, this falls back to the regular
79  * read and write calls.  Using these may distort the cache in some cases
80  * but better have the driver working than preventing it to work on file
81  * systems where the block-level operations are not implemented for
82  * whatever reason.
83  *
84  * NOTE 2: There is a security issue involved with this driver.
85  * Once mounted all access to the contents of the "mapped" file via
86  * the special file is controlled by the permissions on the special
87  * file, the protection of the mapped file is ignored (effectively,
88  * by using root credentials in all transactions).
89  *
90  * NOTE 3: Doesn't interact with leases, should it?
91  */
92 
93 #include <sys/cdefs.h>
94 __KERNEL_RCSID(0, "$NetBSD: vnd.c,v 1.258 2016/08/05 08:21:24 pgoyette Exp $");
95 
96 #if defined(_KERNEL_OPT)
97 #include "opt_vnd.h"
98 #include "opt_compat_netbsd.h"
99 #endif
100 
101 #include <sys/param.h>
102 #include <sys/systm.h>
103 #include <sys/namei.h>
104 #include <sys/proc.h>
105 #include <sys/kthread.h>
106 #include <sys/errno.h>
107 #include <sys/buf.h>
108 #include <sys/bufq.h>
109 #include <sys/malloc.h>
110 #include <sys/ioctl.h>
111 #include <sys/disklabel.h>
112 #include <sys/device.h>
113 #include <sys/disk.h>
114 #include <sys/stat.h>
115 #include <sys/mount.h>
116 #include <sys/vnode.h>
117 #include <sys/file.h>
118 #include <sys/uio.h>
119 #include <sys/conf.h>
120 #include <sys/kauth.h>
121 #include <sys/module.h>
122 
123 #include <net/zlib.h>
124 
125 #include <miscfs/genfs/genfs.h>
126 #include <miscfs/specfs/specdev.h>
127 
128 #include <dev/dkvar.h>
129 #include <dev/vndvar.h>
130 
131 #include "ioconf.h"
132 
133 #if defined(VNDDEBUG) && !defined(DEBUG)
134 #define DEBUG
135 #endif
136 
137 #ifdef DEBUG
138 int dovndcluster = 1;
139 #define VDB_FOLLOW	0x01
140 #define VDB_INIT	0x02
141 #define VDB_IO		0x04
142 #define VDB_LABEL	0x08
143 int vnddebug = 0;
144 #endif
145 
146 #define vndunit(x)	DISKUNIT(x)
147 
148 struct vndxfer {
149 	struct buf vx_buf;
150 	struct vnd_softc *vx_vnd;
151 };
152 #define	VND_BUFTOXFER(bp)	((struct vndxfer *)(void *)bp)
153 
154 #define VND_GETXFER(vnd)	pool_get(&(vnd)->sc_vxpool, PR_WAITOK)
155 #define VND_PUTXFER(vnd, vx)	pool_put(&(vnd)->sc_vxpool, (vx))
156 
157 #define VNDLABELDEV(dev) \
158     (MAKEDISKDEV(major((dev)), vndunit((dev)), RAW_PART))
159 
160 #define	VND_MAXPENDING(vnd)	((vnd)->sc_maxactive * 4)
161 
162 
163 static void	vndclear(struct vnd_softc *, int);
164 static int	vnddoclear(struct vnd_softc *, int, int, bool);
165 static int	vndsetcred(struct vnd_softc *, kauth_cred_t);
166 static void	vndthrottle(struct vnd_softc *, struct vnode *);
167 static void	vndiodone(struct buf *);
168 #if 0
169 static void	vndshutdown(void);
170 #endif
171 
172 static void	vndgetdefaultlabel(struct vnd_softc *, struct disklabel *);
173 static void	vndgetdisklabel(dev_t, struct vnd_softc *);
174 
175 static int	vndlock(struct vnd_softc *);
176 static void	vndunlock(struct vnd_softc *);
177 #ifdef VND_COMPRESSION
178 static void	compstrategy(struct buf *, off_t);
179 static void	*vnd_alloc(void *, u_int, u_int);
180 static void	vnd_free(void *, void *);
181 #endif /* VND_COMPRESSION */
182 
183 static void	vndthread(void *);
184 static bool	vnode_has_op(const struct vnode *, int);
185 static void	handle_with_rdwr(struct vnd_softc *, const struct buf *,
186 		    struct buf *);
187 static void	handle_with_strategy(struct vnd_softc *, const struct buf *,
188 		    struct buf *);
189 static void	vnd_set_geometry(struct vnd_softc *);
190 
191 static dev_type_open(vndopen);
192 static dev_type_close(vndclose);
193 static dev_type_read(vndread);
194 static dev_type_write(vndwrite);
195 static dev_type_ioctl(vndioctl);
196 static dev_type_strategy(vndstrategy);
197 static dev_type_dump(vnddump);
198 static dev_type_size(vndsize);
199 
200 const struct bdevsw vnd_bdevsw = {
201 	.d_open = vndopen,
202 	.d_close = vndclose,
203 	.d_strategy = vndstrategy,
204 	.d_ioctl = vndioctl,
205 	.d_dump = vnddump,
206 	.d_psize = vndsize,
207 	.d_discard = nodiscard,
208 	.d_flag = D_DISK
209 };
210 
211 const struct cdevsw vnd_cdevsw = {
212 	.d_open = vndopen,
213 	.d_close = vndclose,
214 	.d_read = vndread,
215 	.d_write = vndwrite,
216 	.d_ioctl = vndioctl,
217 	.d_stop = nostop,
218 	.d_tty = notty,
219 	.d_poll = nopoll,
220 	.d_mmap = nommap,
221 	.d_kqfilter = nokqfilter,
222 	.d_discard = nodiscard,
223 	.d_flag = D_DISK
224 };
225 
226 static int	vnd_match(device_t, cfdata_t, void *);
227 static void	vnd_attach(device_t, device_t, void *);
228 static int	vnd_detach(device_t, int);
229 
230 CFATTACH_DECL3_NEW(vnd, sizeof(struct vnd_softc),
231     vnd_match, vnd_attach, vnd_detach, NULL, NULL, NULL, DVF_DETACH_SHUTDOWN);
232 extern struct cfdriver vnd_cd;
233 
234 static struct vnd_softc	*vnd_spawn(int);
235 int	vnd_destroy(device_t);
236 
237 static struct	dkdriver vnddkdriver = {
238 	.d_strategy = vndstrategy,
239 	.d_minphys = minphys
240 };
241 
242 void
243 vndattach(int num)
244 {
245 	int error;
246 
247 	error = config_cfattach_attach(vnd_cd.cd_name, &vnd_ca);
248 	if (error)
249 		aprint_error("%s: unable to register cfattach, error = %d\n",
250 		    vnd_cd.cd_name, error);
251 }
252 
253 static int
254 vnd_match(device_t self, cfdata_t cfdata, void *aux)
255 {
256 
257 	return 1;
258 }
259 
260 static void
261 vnd_attach(device_t parent, device_t self, void *aux)
262 {
263 	struct vnd_softc *sc = device_private(self);
264 
265 	sc->sc_dev = self;
266 	sc->sc_comp_offsets = NULL;
267 	sc->sc_comp_buff = NULL;
268 	sc->sc_comp_decombuf = NULL;
269 	bufq_alloc(&sc->sc_tab, "disksort", BUFQ_SORT_RAWBLOCK);
270 	disk_init(&sc->sc_dkdev, device_xname(self), &vnddkdriver);
271 	if (!pmf_device_register(self, NULL, NULL))
272 		aprint_error_dev(self, "couldn't establish power handler\n");
273 }
274 
275 static int
276 vnd_detach(device_t self, int flags)
277 {
278 	int error;
279 	struct vnd_softc *sc = device_private(self);
280 
281 	if (sc->sc_flags & VNF_INITED) {
282 		error = vnddoclear(sc, 0, -1, (flags & DETACH_FORCE) != 0);
283 		if (error != 0)
284 			return error;
285 	}
286 
287 	pmf_device_deregister(self);
288 	bufq_free(sc->sc_tab);
289 	disk_destroy(&sc->sc_dkdev);
290 
291 	return 0;
292 }
293 
294 static struct vnd_softc *
295 vnd_spawn(int unit)
296 {
297 	cfdata_t cf;
298 
299 	cf = malloc(sizeof(*cf), M_DEVBUF, M_WAITOK);
300 	cf->cf_name = vnd_cd.cd_name;
301 	cf->cf_atname = vnd_cd.cd_name;
302 	cf->cf_unit = unit;
303 	cf->cf_fstate = FSTATE_STAR;
304 
305 	return device_private(config_attach_pseudo(cf));
306 }
307 
308 int
309 vnd_destroy(device_t dev)
310 {
311 	int error;
312 	cfdata_t cf;
313 
314 	cf = device_cfdata(dev);
315 	error = config_detach(dev, DETACH_QUIET);
316 	if (error)
317 		return error;
318 	free(cf, M_DEVBUF);
319 	return 0;
320 }
321 
322 static int
323 vndopen(dev_t dev, int flags, int mode, struct lwp *l)
324 {
325 	int unit = vndunit(dev);
326 	struct vnd_softc *sc;
327 	int error = 0, part, pmask;
328 	struct disklabel *lp;
329 
330 #ifdef DEBUG
331 	if (vnddebug & VDB_FOLLOW)
332 		printf("vndopen(0x%"PRIx64", 0x%x, 0x%x, %p)\n", dev, flags, mode, l);
333 #endif
334 	sc = device_lookup_private(&vnd_cd, unit);
335 	if (sc == NULL) {
336 		sc = vnd_spawn(unit);
337 		if (sc == NULL)
338 			return ENOMEM;
339 
340 		/* compatibility, keep disklabel after close */
341 		sc->sc_flags = VNF_KLABEL;
342 	}
343 
344 	if ((error = vndlock(sc)) != 0)
345 		return error;
346 
347 	mutex_enter(&sc->sc_dkdev.dk_openlock);
348 
349 	if ((sc->sc_flags & VNF_CLEARING) != 0) {
350 		error = ENXIO;
351 		goto done;
352 	}
353 
354 	lp = sc->sc_dkdev.dk_label;
355 
356 	part = DISKPART(dev);
357 	pmask = (1 << part);
358 
359 	if (sc->sc_dkdev.dk_nwedges != 0 && part != RAW_PART) {
360 		error = EBUSY;
361 		goto done;
362 	}
363 
364 	if (sc->sc_flags & VNF_INITED) {
365 		if ((sc->sc_dkdev.dk_openmask & ~(1<<RAW_PART)) != 0) {
366 			/*
367 			 * If any non-raw partition is open, but the disk
368 			 * has been invalidated, disallow further opens.
369 			 */
370 			if ((sc->sc_flags & VNF_VLABEL) == 0) {
371 				error = EIO;
372 				goto done;
373 			}
374 		} else {
375 			/*
376 			 * Load the partition info if not already loaded.
377 			 */
378 			if ((sc->sc_flags & VNF_VLABEL) == 0) {
379 				sc->sc_flags |= VNF_VLABEL;
380 				vndgetdisklabel(dev, sc);
381 			}
382 		}
383 	}
384 
385 	/* Check that the partitions exists. */
386 	if (part != RAW_PART) {
387 		if (((sc->sc_flags & VNF_INITED) == 0) ||
388 		    ((part >= lp->d_npartitions) ||
389 		     (lp->d_partitions[part].p_fstype == FS_UNUSED))) {
390 			error = ENXIO;
391 			goto done;
392 		}
393 	}
394 
395 	/* Prevent our unit from being unconfigured while open. */
396 	switch (mode) {
397 	case S_IFCHR:
398 		sc->sc_dkdev.dk_copenmask |= pmask;
399 		break;
400 
401 	case S_IFBLK:
402 		sc->sc_dkdev.dk_bopenmask |= pmask;
403 		break;
404 	}
405 	sc->sc_dkdev.dk_openmask =
406 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
407 
408  done:
409 	mutex_exit(&sc->sc_dkdev.dk_openlock);
410 	vndunlock(sc);
411 	return error;
412 }
413 
414 static int
415 vndclose(dev_t dev, int flags, int mode, struct lwp *l)
416 {
417 	int unit = vndunit(dev);
418 	struct vnd_softc *sc;
419 	int error = 0, part;
420 
421 #ifdef DEBUG
422 	if (vnddebug & VDB_FOLLOW)
423 		printf("vndclose(0x%"PRIx64", 0x%x, 0x%x, %p)\n", dev, flags, mode, l);
424 #endif
425 	sc = device_lookup_private(&vnd_cd, unit);
426 	if (sc == NULL)
427 		return ENXIO;
428 
429 	if ((error = vndlock(sc)) != 0)
430 		return error;
431 
432 	mutex_enter(&sc->sc_dkdev.dk_openlock);
433 
434 	part = DISKPART(dev);
435 
436 	/* ...that much closer to allowing unconfiguration... */
437 	switch (mode) {
438 	case S_IFCHR:
439 		sc->sc_dkdev.dk_copenmask &= ~(1 << part);
440 		break;
441 
442 	case S_IFBLK:
443 		sc->sc_dkdev.dk_bopenmask &= ~(1 << part);
444 		break;
445 	}
446 	sc->sc_dkdev.dk_openmask =
447 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
448 
449 	/* are we last opener ? */
450 	if (sc->sc_dkdev.dk_openmask == 0) {
451 		if ((sc->sc_flags & VNF_KLABEL) == 0)
452 			sc->sc_flags &= ~VNF_VLABEL;
453 	}
454 
455 	mutex_exit(&sc->sc_dkdev.dk_openlock);
456 
457 	vndunlock(sc);
458 
459 	if ((sc->sc_flags & VNF_INITED) == 0) {
460 		if ((error = vnd_destroy(sc->sc_dev)) != 0) {
461 			aprint_error_dev(sc->sc_dev,
462 			    "unable to detach instance\n");
463 			return error;
464 		}
465 	}
466 
467 	return 0;
468 }
469 
470 /*
471  * Queue the request, and wakeup the kernel thread to handle it.
472  */
473 static void
474 vndstrategy(struct buf *bp)
475 {
476 	int unit = vndunit(bp->b_dev);
477 	struct vnd_softc *vnd =
478 	    device_lookup_private(&vnd_cd, unit);
479 	struct disklabel *lp;
480 	daddr_t blkno;
481 	int s = splbio();
482 
483 	if (vnd == NULL) {
484 		bp->b_error = ENXIO;
485 		goto done;
486 	}
487 	lp = vnd->sc_dkdev.dk_label;
488 
489 	if ((vnd->sc_flags & VNF_INITED) == 0) {
490 		bp->b_error = ENXIO;
491 		goto done;
492 	}
493 
494 	/*
495 	 * The transfer must be a whole number of blocks.
496 	 */
497 	if ((bp->b_bcount % lp->d_secsize) != 0) {
498 		bp->b_error = EINVAL;
499 		goto done;
500 	}
501 
502 	/*
503 	 * check if we're read-only.
504 	 */
505 	if ((vnd->sc_flags & VNF_READONLY) && !(bp->b_flags & B_READ)) {
506 		bp->b_error = EACCES;
507 		goto done;
508 	}
509 
510 	/* If it's a nil transfer, wake up the top half now. */
511 	if (bp->b_bcount == 0) {
512 		goto done;
513 	}
514 
515 	/*
516 	 * Do bounds checking and adjust transfer.  If there's an error,
517 	 * the bounds check will flag that for us.
518 	 */
519 	if (DISKPART(bp->b_dev) == RAW_PART) {
520 		if (bounds_check_with_mediasize(bp, DEV_BSIZE,
521 		    vnd->sc_size) <= 0)
522 			goto done;
523 	} else {
524 		if (bounds_check_with_label(&vnd->sc_dkdev,
525 		    bp, vnd->sc_flags & (VNF_WLABEL|VNF_LABELLING)) <= 0)
526 			goto done;
527 	}
528 
529 	/*
530 	 * Put the block number in terms of the logical blocksize
531 	 * of the "device".
532 	 */
533 
534 	blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
535 
536 	/*
537 	 * Translate the partition-relative block number to an absolute.
538 	 */
539 	if (DISKPART(bp->b_dev) != RAW_PART) {
540 		struct partition *pp;
541 
542 		pp = &vnd->sc_dkdev.dk_label->d_partitions[
543 		    DISKPART(bp->b_dev)];
544 		blkno += pp->p_offset;
545 	}
546 	bp->b_rawblkno = blkno;
547 
548 #ifdef DEBUG
549 	if (vnddebug & VDB_FOLLOW)
550 		printf("vndstrategy(%p): unit %d\n", bp, unit);
551 #endif
552 	if ((vnd->sc_flags & VNF_USE_VN_RDWR)) {
553 		KASSERT(vnd->sc_pending >= 0 &&
554 		    vnd->sc_pending <= VND_MAXPENDING(vnd));
555 		while (vnd->sc_pending == VND_MAXPENDING(vnd))
556 			tsleep(&vnd->sc_pending, PRIBIO, "vndpc", 0);
557 		vnd->sc_pending++;
558 	}
559 	bufq_put(vnd->sc_tab, bp);
560 	wakeup(&vnd->sc_tab);
561 	splx(s);
562 	return;
563 
564 done:
565 	bp->b_resid = bp->b_bcount;
566 	biodone(bp);
567 	splx(s);
568 }
569 
570 static bool
571 vnode_has_strategy(struct vnd_softc *vnd)
572 {
573 	return vnode_has_op(vnd->sc_vp, VOFFSET(vop_bmap)) &&
574 	    vnode_has_op(vnd->sc_vp, VOFFSET(vop_strategy));
575 }
576 
577 static bool
578 vnode_has_large_blocks(struct vnd_softc *vnd)
579 {
580 	u_int32_t vnd_secsize, mnt_secsize;
581 	uint64_t numsec;
582 	unsigned secsize;
583 
584 	if (getdisksize(vnd->sc_vp, &numsec, &secsize))
585 		return true;
586 
587 	vnd_secsize = vnd->sc_geom.vng_secsize;
588 	mnt_secsize = secsize;
589 
590 	return vnd_secsize % mnt_secsize != 0;
591 }
592 
593 /* XXX this function needs a reliable check to detect
594  * sparse files. Otherwise, bmap/strategy may be used
595  * and fail on non-allocated blocks. VOP_READ/VOP_WRITE
596  * works on sparse files.
597  */
598 #if notyet
599 static bool
600 vnode_strategy_probe(struct vnd_softc *vnd)
601 {
602 	int error;
603 	daddr_t nbn;
604 
605 	if (!vnode_has_strategy(vnd))
606 		return false;
607 
608 	if (vnode_has_large_blocks(vnd))
609 		return false;
610 
611 	/* Convert the first logical block number to its
612 	 * physical block number.
613 	 */
614 	error = 0;
615 	vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
616 	error = VOP_BMAP(vnd->sc_vp, 0, NULL, &nbn, NULL);
617 	VOP_UNLOCK(vnd->sc_vp);
618 
619 	/* Test if that worked. */
620 	if (error == 0 && (long)nbn == -1)
621 		return false;
622 
623 	return true;
624 }
625 #endif
626 
627 static void
628 vndthread(void *arg)
629 {
630 	struct vnd_softc *vnd = arg;
631 	int s;
632 
633 	/* Determine whether we can *use* VOP_BMAP and VOP_STRATEGY to
634 	 * directly access the backing vnode.  If we can, use these two
635 	 * operations to avoid messing with the local buffer cache.
636 	 * Otherwise fall back to regular VOP_READ/VOP_WRITE operations
637 	 * which are guaranteed to work with any file system. */
638 	if ((vnd->sc_flags & VNF_USE_VN_RDWR) == 0 &&
639 	    ! vnode_has_strategy(vnd))
640 		vnd->sc_flags |= VNF_USE_VN_RDWR;
641 
642 	/* VOP_STRATEGY can only be used if the backing vnode allows
643 	 * to access blocks as small as defined by the vnd geometry.
644 	 */
645 	if ((vnd->sc_flags & VNF_USE_VN_RDWR) == 0 &&
646 	    vnode_has_large_blocks(vnd))
647 		vnd->sc_flags |= VNF_USE_VN_RDWR;
648 
649 #ifdef DEBUG
650 	if (vnddebug & VDB_INIT)
651 		printf("vndthread: vp %p, %s\n", vnd->sc_vp,
652 		    (vnd->sc_flags & VNF_USE_VN_RDWR) == 0 ?
653 		    "using bmap/strategy operations" :
654 		    "using read/write operations");
655 #endif
656 
657 	s = splbio();
658 	vnd->sc_flags |= VNF_KTHREAD;
659 	wakeup(&vnd->sc_kthread);
660 
661 	/*
662 	 * Dequeue requests and serve them depending on the available
663 	 * vnode operations.
664 	 */
665 	while ((vnd->sc_flags & VNF_VUNCONF) == 0) {
666 		struct vndxfer *vnx;
667 		struct buf *obp;
668 		struct buf *bp;
669 
670 		obp = bufq_get(vnd->sc_tab);
671 		if (obp == NULL) {
672 			tsleep(&vnd->sc_tab, PRIBIO, "vndbp", 0);
673 			continue;
674 		};
675 		if ((vnd->sc_flags & VNF_USE_VN_RDWR)) {
676 			KASSERT(vnd->sc_pending > 0 &&
677 			    vnd->sc_pending <= VND_MAXPENDING(vnd));
678 			if (vnd->sc_pending-- == VND_MAXPENDING(vnd))
679 				wakeup(&vnd->sc_pending);
680 		}
681 		splx(s);
682 #ifdef DEBUG
683 		if (vnddebug & VDB_FOLLOW)
684 			printf("vndthread(%p)\n", obp);
685 #endif
686 
687 		if (vnd->sc_vp->v_mount == NULL) {
688 			obp->b_error = ENXIO;
689 			goto done;
690 		}
691 #ifdef VND_COMPRESSION
692 		/* handle a compressed read */
693 		if ((obp->b_flags & B_READ) != 0 && (vnd->sc_flags & VNF_COMP)) {
694 			off_t bn;
695 
696 			/* Convert to a byte offset within the file. */
697 			bn = obp->b_rawblkno *
698 			    vnd->sc_dkdev.dk_label->d_secsize;
699 
700 			compstrategy(obp, bn);
701 			goto done;
702 		}
703 #endif /* VND_COMPRESSION */
704 
705 		/*
706 		 * Allocate a header for this transfer and link it to the
707 		 * buffer
708 		 */
709 		s = splbio();
710 		vnx = VND_GETXFER(vnd);
711 		splx(s);
712 		vnx->vx_vnd = vnd;
713 
714 		s = splbio();
715 		while (vnd->sc_active >= vnd->sc_maxactive) {
716 			tsleep(&vnd->sc_tab, PRIBIO, "vndac", 0);
717 		}
718 		vnd->sc_active++;
719 		splx(s);
720 
721 		/* Instrumentation. */
722 		disk_busy(&vnd->sc_dkdev);
723 
724 		bp = &vnx->vx_buf;
725 		buf_init(bp);
726 		bp->b_flags = (obp->b_flags & B_READ);
727 		bp->b_oflags = obp->b_oflags;
728 		bp->b_cflags = obp->b_cflags;
729 		bp->b_iodone = vndiodone;
730 		bp->b_private = obp;
731 		bp->b_vp = vnd->sc_vp;
732 		bp->b_objlock = bp->b_vp->v_interlock;
733 		bp->b_data = obp->b_data;
734 		bp->b_bcount = obp->b_bcount;
735 		BIO_COPYPRIO(bp, obp);
736 
737 		/* Handle the request using the appropriate operations. */
738 		if ((vnd->sc_flags & VNF_USE_VN_RDWR) == 0)
739 			handle_with_strategy(vnd, obp, bp);
740 		else
741 			handle_with_rdwr(vnd, obp, bp);
742 
743 		s = splbio();
744 		continue;
745 
746 done:
747 		biodone(obp);
748 		s = splbio();
749 	}
750 
751 	vnd->sc_flags &= (~VNF_KTHREAD | VNF_VUNCONF);
752 	wakeup(&vnd->sc_kthread);
753 	splx(s);
754 	kthread_exit(0);
755 }
756 
757 /*
758  * Checks if the given vnode supports the requested operation.
759  * The operation is specified the offset returned by VOFFSET.
760  *
761  * XXX The test below used to determine this is quite fragile
762  * because it relies on the file system to use genfs to specify
763  * unimplemented operations.  There might be another way to do
764  * it more cleanly.
765  */
766 static bool
767 vnode_has_op(const struct vnode *vp, int opoffset)
768 {
769 	int (*defaultp)(void *);
770 	int (*opp)(void *);
771 
772 	defaultp = vp->v_op[VOFFSET(vop_default)];
773 	opp = vp->v_op[opoffset];
774 
775 	return opp != defaultp && opp != genfs_eopnotsupp &&
776 	    opp != genfs_badop && opp != genfs_nullop;
777 }
778 
779 /*
780  * Handles the read/write request given in 'bp' using the vnode's VOP_READ
781  * and VOP_WRITE operations.
782  *
783  * 'obp' is a pointer to the original request fed to the vnd device.
784  */
785 static void
786 handle_with_rdwr(struct vnd_softc *vnd, const struct buf *obp, struct buf *bp)
787 {
788 	bool doread;
789 	off_t offset;
790 	size_t len, resid;
791 	struct vnode *vp;
792 
793 	doread = bp->b_flags & B_READ;
794 	offset = obp->b_rawblkno * vnd->sc_dkdev.dk_label->d_secsize;
795 	len = bp->b_bcount;
796 	vp = vnd->sc_vp;
797 
798 #if defined(DEBUG)
799 	if (vnddebug & VDB_IO)
800 		printf("vnd (rdwr): vp %p, %s, rawblkno 0x%" PRIx64
801 		    ", secsize %d, offset %" PRIu64
802 		    ", bcount %d\n",
803 		    vp, doread ? "read" : "write", obp->b_rawblkno,
804 		    vnd->sc_dkdev.dk_label->d_secsize, offset,
805 		    bp->b_bcount);
806 #endif
807 
808 	/* Issue the read or write operation. */
809 	bp->b_error =
810 	    vn_rdwr(doread ? UIO_READ : UIO_WRITE,
811 	    vp, bp->b_data, len, offset, UIO_SYSSPACE,
812 	    IO_ADV_ENCODE(POSIX_FADV_NOREUSE), vnd->sc_cred, &resid, NULL);
813 	bp->b_resid = resid;
814 
815 	mutex_enter(vp->v_interlock);
816 	(void) VOP_PUTPAGES(vp, 0, 0,
817 	    PGO_ALLPAGES | PGO_CLEANIT | PGO_FREE | PGO_SYNCIO);
818 
819 	/* We need to increase the number of outputs on the vnode if
820 	 * there was any write to it. */
821 	if (!doread) {
822 		mutex_enter(vp->v_interlock);
823 		vp->v_numoutput++;
824 		mutex_exit(vp->v_interlock);
825 	}
826 
827 	biodone(bp);
828 }
829 
830 /*
831  * Handes the read/write request given in 'bp' using the vnode's VOP_BMAP
832  * and VOP_STRATEGY operations.
833  *
834  * 'obp' is a pointer to the original request fed to the vnd device.
835  */
836 static void
837 handle_with_strategy(struct vnd_softc *vnd, const struct buf *obp,
838     struct buf *bp)
839 {
840 	int bsize, error, flags, skipped;
841 	size_t resid, sz;
842 	off_t bn, offset;
843 	struct vnode *vp;
844 	struct buf *nbp = NULL;
845 
846 	flags = obp->b_flags;
847 
848 
849 	/* convert to a byte offset within the file. */
850 	bn = obp->b_rawblkno * vnd->sc_dkdev.dk_label->d_secsize;
851 
852 	bsize = vnd->sc_vp->v_mount->mnt_stat.f_iosize;
853 	skipped = 0;
854 
855 	/*
856 	 * Break the request into bsize pieces and feed them
857 	 * sequentially using VOP_BMAP/VOP_STRATEGY.
858 	 * We do it this way to keep from flooding NFS servers if we
859 	 * are connected to an NFS file.  This places the burden on
860 	 * the client rather than the server.
861 	 */
862 	error = 0;
863 	bp->b_resid = bp->b_bcount;
864 	for (offset = 0, resid = bp->b_resid; /* true */;
865 	    resid -= sz, offset += sz) {
866 		daddr_t nbn;
867 		int off, nra;
868 
869 		nra = 0;
870 		vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
871 		error = VOP_BMAP(vnd->sc_vp, bn / bsize, &vp, &nbn, &nra);
872 		VOP_UNLOCK(vnd->sc_vp);
873 
874 		if (error == 0 && (long)nbn == -1)
875 			error = EIO;
876 
877 		/*
878 		 * If there was an error or a hole in the file...punt.
879 		 * Note that we may have to wait for any operations
880 		 * that we have already fired off before releasing
881 		 * the buffer.
882 		 *
883 		 * XXX we could deal with holes here but it would be
884 		 * a hassle (in the write case).
885 		 */
886 		if (error) {
887 			skipped += resid;
888 			break;
889 		}
890 
891 #ifdef DEBUG
892 		if (!dovndcluster)
893 			nra = 0;
894 #endif
895 
896 		off = bn % bsize;
897 		sz = MIN(((off_t)1 + nra) * bsize - off, resid);
898 #ifdef	DEBUG
899 		if (vnddebug & VDB_IO)
900 			printf("vndstrategy: vp %p/%p bn 0x%qx/0x%" PRIx64
901 			    " sz 0x%zx\n", vnd->sc_vp, vp, (long long)bn,
902 			    nbn, sz);
903 #endif
904 
905 		nbp = getiobuf(vp, true);
906 		nestiobuf_setup(bp, nbp, offset, sz);
907 		nbp->b_blkno = nbn + btodb(off);
908 
909 #if 0 /* XXX #ifdef DEBUG */
910 		if (vnddebug & VDB_IO)
911 			printf("vndstart(%ld): bp %p vp %p blkno "
912 			    "0x%" PRIx64 " flags %x addr %p cnt 0x%x\n",
913 			    (long) (vnd-vnd_softc), &nbp->vb_buf,
914 			    nbp->vb_buf.b_vp, nbp->vb_buf.b_blkno,
915 			    nbp->vb_buf.b_flags, nbp->vb_buf.b_data,
916 			    nbp->vb_buf.b_bcount);
917 #endif
918 		if (resid == sz) {
919 			break;
920 		}
921 		VOP_STRATEGY(vp, nbp);
922 		bn += sz;
923 	}
924 	if (!(flags & B_READ)) {
925 		struct vnode *w_vp;
926 		/*
927 		 * this is the last nested buf, account for
928 		 * the parent buf write too.
929 		 * This has to be done last, so that
930 		 * fsync won't wait for this write which
931 		 * has no chance to complete before all nested bufs
932 		 * have been queued. But it has to be done
933 		 * before the last VOP_STRATEGY()
934 		 * or the call to nestiobuf_done().
935 		 */
936 		w_vp = bp->b_vp;
937 		mutex_enter(w_vp->v_interlock);
938 		w_vp->v_numoutput++;
939 		mutex_exit(w_vp->v_interlock);
940 	}
941 	KASSERT(skipped != 0 || nbp != NULL);
942 	if (skipped)
943 		nestiobuf_done(bp, skipped, error);
944 	else
945 		VOP_STRATEGY(vp, nbp);
946 }
947 
948 static void
949 vndiodone(struct buf *bp)
950 {
951 	struct vndxfer *vnx = VND_BUFTOXFER(bp);
952 	struct vnd_softc *vnd = vnx->vx_vnd;
953 	struct buf *obp = bp->b_private;
954 	int s = splbio();
955 
956 	KASSERT(&vnx->vx_buf == bp);
957 	KASSERT(vnd->sc_active > 0);
958 #ifdef DEBUG
959 	if (vnddebug & VDB_IO) {
960 		printf("vndiodone1: bp %p iodone: error %d\n",
961 		    bp, bp->b_error);
962 	}
963 #endif
964 	disk_unbusy(&vnd->sc_dkdev, bp->b_bcount - bp->b_resid,
965 	    (bp->b_flags & B_READ));
966 	vnd->sc_active--;
967 	if (vnd->sc_active == 0) {
968 		wakeup(&vnd->sc_tab);
969 	}
970 	splx(s);
971 	obp->b_error = bp->b_error;
972 	obp->b_resid = bp->b_resid;
973 	buf_destroy(bp);
974 	VND_PUTXFER(vnd, vnx);
975 	biodone(obp);
976 }
977 
978 /* ARGSUSED */
979 static int
980 vndread(dev_t dev, struct uio *uio, int flags)
981 {
982 	int unit = vndunit(dev);
983 	struct vnd_softc *sc;
984 
985 #ifdef DEBUG
986 	if (vnddebug & VDB_FOLLOW)
987 		printf("vndread(0x%"PRIx64", %p)\n", dev, uio);
988 #endif
989 
990 	sc = device_lookup_private(&vnd_cd, unit);
991 	if (sc == NULL)
992 		return ENXIO;
993 
994 	if ((sc->sc_flags & VNF_INITED) == 0)
995 		return ENXIO;
996 
997 	return physio(vndstrategy, NULL, dev, B_READ, minphys, uio);
998 }
999 
1000 /* ARGSUSED */
1001 static int
1002 vndwrite(dev_t dev, struct uio *uio, int flags)
1003 {
1004 	int unit = vndunit(dev);
1005 	struct vnd_softc *sc;
1006 
1007 #ifdef DEBUG
1008 	if (vnddebug & VDB_FOLLOW)
1009 		printf("vndwrite(0x%"PRIx64", %p)\n", dev, uio);
1010 #endif
1011 
1012 	sc = device_lookup_private(&vnd_cd, unit);
1013 	if (sc == NULL)
1014 		return ENXIO;
1015 
1016 	if ((sc->sc_flags & VNF_INITED) == 0)
1017 		return ENXIO;
1018 
1019 	return physio(vndstrategy, NULL, dev, B_WRITE, minphys, uio);
1020 }
1021 
1022 static int
1023 vnd_cget(struct lwp *l, int unit, int *un, struct vattr *va)
1024 {
1025 	int error;
1026 	struct vnd_softc *vnd;
1027 
1028 	if (*un == -1)
1029 		*un = unit;
1030 	if (*un < 0)
1031 		return EINVAL;
1032 
1033 	vnd = device_lookup_private(&vnd_cd, *un);
1034 	if (vnd == NULL)
1035 		return -1;
1036 
1037 	if ((vnd->sc_flags & VNF_INITED) == 0)
1038 		return -1;
1039 
1040 	vn_lock(vnd->sc_vp, LK_SHARED | LK_RETRY);
1041 	error = VOP_GETATTR(vnd->sc_vp, va, l->l_cred);
1042 	VOP_UNLOCK(vnd->sc_vp);
1043 	return error;
1044 }
1045 
1046 static int
1047 vnddoclear(struct vnd_softc *vnd, int pmask, int minor, bool force)
1048 {
1049 	int error;
1050 
1051 	if ((error = vndlock(vnd)) != 0)
1052 		return error;
1053 
1054 	/*
1055 	 * Don't unconfigure if any other partitions are open
1056 	 * or if both the character and block flavors of this
1057 	 * partition are open.
1058 	 */
1059 	if (DK_BUSY(vnd, pmask) && !force) {
1060 		vndunlock(vnd);
1061 		return EBUSY;
1062 	}
1063 
1064 	/* Delete all of our wedges */
1065 	dkwedge_delall(&vnd->sc_dkdev);
1066 
1067 	/*
1068 	 * XXX vndclear() might call vndclose() implicitly;
1069 	 * release lock to avoid recursion
1070 	 *
1071 	 * Set VNF_CLEARING to prevent vndopen() from
1072 	 * sneaking in after we vndunlock().
1073 	 */
1074 	vnd->sc_flags |= VNF_CLEARING;
1075 	vndunlock(vnd);
1076 	vndclear(vnd, minor);
1077 #ifdef DEBUG
1078 	if (vnddebug & VDB_INIT)
1079 		printf("%s: CLRed\n", __func__);
1080 #endif
1081 
1082 	/* Destroy the xfer and buffer pools. */
1083 	pool_destroy(&vnd->sc_vxpool);
1084 
1085 	/* Detach the disk. */
1086 	disk_detach(&vnd->sc_dkdev);
1087 
1088 	return 0;
1089 }
1090 
1091 static int
1092 vndioctl_get(struct lwp *l, void *data, int unit, struct vattr *va)
1093 {
1094 	int error;
1095 
1096 	KASSERT(l);
1097 
1098 	/* the first member is always int vnd_unit in all the versions */
1099 	if (*(int *)data >= vnd_cd.cd_ndevs)
1100 		return ENXIO;
1101 
1102 	switch (error = vnd_cget(l, unit, (int *)data, va)) {
1103 	case -1:
1104 		/* unused is not an error */
1105 		memset(va, 0, sizeof(*va));
1106 		/*FALLTHROUGH*/
1107 	case 0:
1108 		return 0;
1109 	default:
1110 		return error;
1111 	}
1112 }
1113 
1114 /* ARGSUSED */
1115 static int
1116 vndioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l)
1117 {
1118 	bool force;
1119 	int unit = vndunit(dev);
1120 	struct vnd_softc *vnd;
1121 	struct vnd_ioctl *vio;
1122 	struct vattr vattr;
1123 	struct pathbuf *pb;
1124 	struct nameidata nd;
1125 	int error, part, pmask;
1126 	uint64_t geomsize;
1127 	int fflags;
1128 #ifdef __HAVE_OLD_DISKLABEL
1129 	struct disklabel newlabel;
1130 #endif
1131 
1132 #ifdef DEBUG
1133 	if (vnddebug & VDB_FOLLOW)
1134 		printf("vndioctl(0x%"PRIx64", 0x%lx, %p, 0x%x, %p): unit %d\n",
1135 		    dev, cmd, data, flag, l->l_proc, unit);
1136 #endif
1137 	/* Do the get's first; they don't need initialization or verification */
1138 	switch (cmd) {
1139 #ifdef COMPAT_30
1140 	case VNDIOCGET30: {
1141 		if ((error = vndioctl_get(l, data, unit, &vattr)) != 0)
1142 			return error;
1143 
1144 		struct vnd_user30 *vnu = data;
1145 		vnu->vnu_dev = vattr.va_fsid;
1146 		vnu->vnu_ino = vattr.va_fileid;
1147 		return 0;
1148 	}
1149 #endif
1150 #ifdef COMPAT_50
1151 	case VNDIOCGET50: {
1152 		if ((error = vndioctl_get(l, data, unit, &vattr)) != 0)
1153 			return error;
1154 
1155 		struct vnd_user50 *vnu = data;
1156 		vnu->vnu_dev = vattr.va_fsid;
1157 		vnu->vnu_ino = vattr.va_fileid;
1158 		return 0;
1159 	}
1160 #endif
1161 
1162 	case VNDIOCGET: {
1163 		if ((error = vndioctl_get(l, data, unit, &vattr)) != 0)
1164 			return error;
1165 
1166 		struct vnd_user *vnu = data;
1167 		vnu->vnu_dev = vattr.va_fsid;
1168 		vnu->vnu_ino = vattr.va_fileid;
1169 		return 0;
1170 	}
1171 	default:
1172 		break;
1173 	}
1174 
1175 	vnd = device_lookup_private(&vnd_cd, unit);
1176 	if (vnd == NULL)
1177 		return ENXIO;
1178 	vio = (struct vnd_ioctl *)data;
1179 
1180 	/* Must be open for writes for these commands... */
1181 	switch (cmd) {
1182 	case VNDIOCSET:
1183 	case VNDIOCCLR:
1184 #ifdef COMPAT_50
1185 	case VNDIOCSET50:
1186 	case VNDIOCCLR50:
1187 #endif
1188 	case DIOCSDINFO:
1189 	case DIOCWDINFO:
1190 #ifdef __HAVE_OLD_DISKLABEL
1191 	case ODIOCSDINFO:
1192 	case ODIOCWDINFO:
1193 #endif
1194 	case DIOCKLABEL:
1195 	case DIOCWLABEL:
1196 		if ((flag & FWRITE) == 0)
1197 			return EBADF;
1198 	}
1199 
1200 	/* Must be initialized for these... */
1201 	switch (cmd) {
1202 	case VNDIOCCLR:
1203 #ifdef VNDIOCCLR50
1204 	case VNDIOCCLR50:
1205 #endif
1206 	case DIOCGDINFO:
1207 	case DIOCSDINFO:
1208 	case DIOCWDINFO:
1209 	case DIOCGPARTINFO:
1210 	case DIOCKLABEL:
1211 	case DIOCWLABEL:
1212 	case DIOCGDEFLABEL:
1213 	case DIOCCACHESYNC:
1214 #ifdef __HAVE_OLD_DISKLABEL
1215 	case ODIOCGDINFO:
1216 	case ODIOCSDINFO:
1217 	case ODIOCWDINFO:
1218 	case ODIOCGDEFLABEL:
1219 #endif
1220 		if ((vnd->sc_flags & VNF_INITED) == 0)
1221 			return ENXIO;
1222 	}
1223 
1224 	error = disk_ioctl(&vnd->sc_dkdev, dev, cmd, data, flag, l);
1225 	if (error != EPASSTHROUGH)
1226 		return error;
1227 
1228 
1229 	switch (cmd) {
1230 #ifdef VNDIOCSET50
1231 	case VNDIOCSET50:
1232 #endif
1233 	case VNDIOCSET:
1234 		if (vnd->sc_flags & VNF_INITED)
1235 			return EBUSY;
1236 
1237 		if ((error = vndlock(vnd)) != 0)
1238 			return error;
1239 
1240 		fflags = FREAD;
1241 		if ((vio->vnd_flags & VNDIOF_READONLY) == 0)
1242 			fflags |= FWRITE;
1243 		error = pathbuf_copyin(vio->vnd_file, &pb);
1244 		if (error) {
1245 			goto unlock_and_exit;
1246 		}
1247 		NDINIT(&nd, LOOKUP, FOLLOW, pb);
1248 		if ((error = vn_open(&nd, fflags, 0)) != 0) {
1249 			pathbuf_destroy(pb);
1250 			goto unlock_and_exit;
1251 		}
1252 		KASSERT(l);
1253 		error = VOP_GETATTR(nd.ni_vp, &vattr, l->l_cred);
1254 		if (!error && nd.ni_vp->v_type != VREG)
1255 			error = EOPNOTSUPP;
1256 		if (!error && vattr.va_bytes < vattr.va_size)
1257 			/* File is definitely sparse, use vn_rdwr() */
1258 			vnd->sc_flags |= VNF_USE_VN_RDWR;
1259 		if (error) {
1260 			VOP_UNLOCK(nd.ni_vp);
1261 			goto close_and_exit;
1262 		}
1263 
1264 		/* If using a compressed file, initialize its info */
1265 		/* (or abort with an error if kernel has no compression) */
1266 		if (vio->vnd_flags & VNF_COMP) {
1267 #ifdef VND_COMPRESSION
1268 			struct vnd_comp_header *ch;
1269 			int i;
1270 			u_int32_t comp_size;
1271 			u_int32_t comp_maxsize;
1272 
1273 			/* allocate space for compresed file header */
1274 			ch = malloc(sizeof(struct vnd_comp_header),
1275 			M_TEMP, M_WAITOK);
1276 
1277 			/* read compressed file header */
1278 			error = vn_rdwr(UIO_READ, nd.ni_vp, (void *)ch,
1279 			  sizeof(struct vnd_comp_header), 0, UIO_SYSSPACE,
1280 			  IO_UNIT|IO_NODELOCKED, l->l_cred, NULL, NULL);
1281 			if (error) {
1282 				free(ch, M_TEMP);
1283 				VOP_UNLOCK(nd.ni_vp);
1284 				goto close_and_exit;
1285 			}
1286 
1287 			/* save some header info */
1288 			vnd->sc_comp_blksz = ntohl(ch->block_size);
1289 			/* note last offset is the file byte size */
1290 			vnd->sc_comp_numoffs = ntohl(ch->num_blocks)+1;
1291 			free(ch, M_TEMP);
1292 			if (!DK_DEV_BSIZE_OK(vnd->sc_comp_blksz)) {
1293 				VOP_UNLOCK(nd.ni_vp);
1294 				error = EINVAL;
1295 				goto close_and_exit;
1296 			}
1297 			if (sizeof(struct vnd_comp_header) +
1298 			  sizeof(u_int64_t) * vnd->sc_comp_numoffs >
1299 			  vattr.va_size) {
1300 				VOP_UNLOCK(nd.ni_vp);
1301 				error = EINVAL;
1302 				goto close_and_exit;
1303 			}
1304 
1305 			/* set decompressed file size */
1306 			vattr.va_size =
1307 			    ((u_quad_t)vnd->sc_comp_numoffs - 1) *
1308 			     (u_quad_t)vnd->sc_comp_blksz;
1309 
1310 			/* allocate space for all the compressed offsets */
1311 			vnd->sc_comp_offsets =
1312 			malloc(sizeof(u_int64_t) * vnd->sc_comp_numoffs,
1313 			M_DEVBUF, M_WAITOK);
1314 
1315 			/* read in the offsets */
1316 			error = vn_rdwr(UIO_READ, nd.ni_vp,
1317 			  (void *)vnd->sc_comp_offsets,
1318 			  sizeof(u_int64_t) * vnd->sc_comp_numoffs,
1319 			  sizeof(struct vnd_comp_header), UIO_SYSSPACE,
1320 			  IO_UNIT|IO_NODELOCKED, l->l_cred, NULL, NULL);
1321 			if (error) {
1322 				VOP_UNLOCK(nd.ni_vp);
1323 				goto close_and_exit;
1324 			}
1325 			/*
1326 			 * find largest block size (used for allocation limit).
1327 			 * Also convert offset to native byte order.
1328 			 */
1329 			comp_maxsize = 0;
1330 			for (i = 0; i < vnd->sc_comp_numoffs - 1; i++) {
1331 				vnd->sc_comp_offsets[i] =
1332 				  be64toh(vnd->sc_comp_offsets[i]);
1333 				comp_size = be64toh(vnd->sc_comp_offsets[i + 1])
1334 				  - vnd->sc_comp_offsets[i];
1335 				if (comp_size > comp_maxsize)
1336 					comp_maxsize = comp_size;
1337 			}
1338 			vnd->sc_comp_offsets[vnd->sc_comp_numoffs - 1] =
1339 			  be64toh(vnd->sc_comp_offsets[vnd->sc_comp_numoffs - 1]);
1340 
1341 			/* create compressed data buffer */
1342 			vnd->sc_comp_buff = malloc(comp_maxsize,
1343 			  M_DEVBUF, M_WAITOK);
1344 
1345 			/* create decompressed buffer */
1346 			vnd->sc_comp_decombuf = malloc(vnd->sc_comp_blksz,
1347 			  M_DEVBUF, M_WAITOK);
1348 			vnd->sc_comp_buffblk = -1;
1349 
1350 			/* Initialize decompress stream */
1351 			memset(&vnd->sc_comp_stream, 0, sizeof(z_stream));
1352 			vnd->sc_comp_stream.zalloc = vnd_alloc;
1353 			vnd->sc_comp_stream.zfree = vnd_free;
1354 			error = inflateInit2(&vnd->sc_comp_stream, MAX_WBITS);
1355 			if (error) {
1356 				if (vnd->sc_comp_stream.msg)
1357 					printf("vnd%d: compressed file, %s\n",
1358 					  unit, vnd->sc_comp_stream.msg);
1359 				VOP_UNLOCK(nd.ni_vp);
1360 				error = EINVAL;
1361 				goto close_and_exit;
1362 			}
1363 
1364 			vnd->sc_flags |= VNF_COMP | VNF_READONLY;
1365 #else /* !VND_COMPRESSION */
1366 			VOP_UNLOCK(nd.ni_vp);
1367 			error = EOPNOTSUPP;
1368 			goto close_and_exit;
1369 #endif /* VND_COMPRESSION */
1370 		}
1371 
1372 		VOP_UNLOCK(nd.ni_vp);
1373 		vnd->sc_vp = nd.ni_vp;
1374 		vnd->sc_size = btodb(vattr.va_size);	/* note truncation */
1375 
1376 		/*
1377 		 * Use pseudo-geometry specified.  If none was provided,
1378 		 * use "standard" Adaptec fictitious geometry.
1379 		 */
1380 		if (vio->vnd_flags & VNDIOF_HASGEOM) {
1381 
1382 			memcpy(&vnd->sc_geom, &vio->vnd_geom,
1383 			    sizeof(vio->vnd_geom));
1384 
1385 			/*
1386 			 * Sanity-check the sector size.
1387 			 */
1388 			if (!DK_DEV_BSIZE_OK(vnd->sc_geom.vng_secsize) ||
1389 			    vnd->sc_geom.vng_ncylinders == 0 ||
1390 			    vnd->sc_geom.vng_ntracks == 0 ||
1391 			    vnd->sc_geom.vng_nsectors == 0) {
1392 				error = EINVAL;
1393 				goto close_and_exit;
1394 			}
1395 
1396 			/*
1397 			 * Compute the size (in DEV_BSIZE blocks) specified
1398 			 * by the geometry.
1399 			 */
1400 			geomsize = (int64_t)vnd->sc_geom.vng_nsectors *
1401 			    vnd->sc_geom.vng_ntracks *
1402 			    vnd->sc_geom.vng_ncylinders *
1403 			    (vnd->sc_geom.vng_secsize / DEV_BSIZE);
1404 
1405 			/*
1406 			 * Sanity-check the size against the specified
1407 			 * geometry.
1408 			 */
1409 			if (vnd->sc_size < geomsize) {
1410 				error = EINVAL;
1411 				goto close_and_exit;
1412 			}
1413 		} else if (vnd->sc_size >= (32 * 64)) {
1414 			/*
1415 			 * Size must be at least 2048 DEV_BSIZE blocks
1416 			 * (1M) in order to use this geometry.
1417 			 */
1418 			vnd->sc_geom.vng_secsize = DEV_BSIZE;
1419 			vnd->sc_geom.vng_nsectors = 32;
1420 			vnd->sc_geom.vng_ntracks = 64;
1421 			vnd->sc_geom.vng_ncylinders = vnd->sc_size / (64 * 32);
1422 		} else {
1423 			vnd->sc_geom.vng_secsize = DEV_BSIZE;
1424 			vnd->sc_geom.vng_nsectors = 1;
1425 			vnd->sc_geom.vng_ntracks = 1;
1426 			vnd->sc_geom.vng_ncylinders = vnd->sc_size;
1427 		}
1428 
1429 		vnd_set_geometry(vnd);
1430 
1431 		if (vio->vnd_flags & VNDIOF_READONLY) {
1432 			vnd->sc_flags |= VNF_READONLY;
1433 		}
1434 
1435 		if ((error = vndsetcred(vnd, l->l_cred)) != 0)
1436 			goto close_and_exit;
1437 
1438 		vndthrottle(vnd, vnd->sc_vp);
1439 		vio->vnd_osize = dbtob(vnd->sc_size);
1440 #ifdef VNDIOCSET50
1441 		if (cmd != VNDIOCSET50)
1442 #endif
1443 			vio->vnd_size = dbtob(vnd->sc_size);
1444 		vnd->sc_flags |= VNF_INITED;
1445 
1446 		/* create the kernel thread, wait for it to be up */
1447 		error = kthread_create(PRI_NONE, 0, NULL, vndthread, vnd,
1448 		    &vnd->sc_kthread, "%s", device_xname(vnd->sc_dev));
1449 		if (error)
1450 			goto close_and_exit;
1451 		while ((vnd->sc_flags & VNF_KTHREAD) == 0) {
1452 			tsleep(&vnd->sc_kthread, PRIBIO, "vndthr", 0);
1453 		}
1454 #ifdef DEBUG
1455 		if (vnddebug & VDB_INIT)
1456 			printf("vndioctl: SET vp %p size 0x%lx %d/%d/%d/%d\n",
1457 			    vnd->sc_vp, (unsigned long) vnd->sc_size,
1458 			    vnd->sc_geom.vng_secsize,
1459 			    vnd->sc_geom.vng_nsectors,
1460 			    vnd->sc_geom.vng_ntracks,
1461 			    vnd->sc_geom.vng_ncylinders);
1462 #endif
1463 
1464 		/* Attach the disk. */
1465 		disk_attach(&vnd->sc_dkdev);
1466 
1467 		/* Initialize the xfer and buffer pools. */
1468 		pool_init(&vnd->sc_vxpool, sizeof(struct vndxfer), 0,
1469 		    0, 0, "vndxpl", NULL, IPL_BIO);
1470 
1471 		vndunlock(vnd);
1472 
1473 		pathbuf_destroy(pb);
1474 
1475 		/* Discover wedges on this disk */
1476 		dkwedge_discover(&vnd->sc_dkdev);
1477 
1478 		break;
1479 
1480 close_and_exit:
1481 		(void) vn_close(nd.ni_vp, fflags, l->l_cred);
1482 		pathbuf_destroy(pb);
1483 unlock_and_exit:
1484 #ifdef VND_COMPRESSION
1485 		/* free any allocated memory (for compressed file) */
1486 		if (vnd->sc_comp_offsets) {
1487 			free(vnd->sc_comp_offsets, M_DEVBUF);
1488 			vnd->sc_comp_offsets = NULL;
1489 		}
1490 		if (vnd->sc_comp_buff) {
1491 			free(vnd->sc_comp_buff, M_DEVBUF);
1492 			vnd->sc_comp_buff = NULL;
1493 		}
1494 		if (vnd->sc_comp_decombuf) {
1495 			free(vnd->sc_comp_decombuf, M_DEVBUF);
1496 			vnd->sc_comp_decombuf = NULL;
1497 		}
1498 #endif /* VND_COMPRESSION */
1499 		vndunlock(vnd);
1500 		return error;
1501 
1502 #ifdef VNDIOCCLR50
1503 	case VNDIOCCLR50:
1504 #endif
1505 	case VNDIOCCLR:
1506 		part = DISKPART(dev);
1507 		pmask = (1 << part);
1508 		force = (vio->vnd_flags & VNDIOF_FORCE) != 0;
1509 
1510 		if ((error = vnddoclear(vnd, pmask, minor(dev), force)) != 0)
1511 			return error;
1512 
1513 		break;
1514 
1515 
1516 	case DIOCWDINFO:
1517 	case DIOCSDINFO:
1518 #ifdef __HAVE_OLD_DISKLABEL
1519 	case ODIOCWDINFO:
1520 	case ODIOCSDINFO:
1521 #endif
1522 	{
1523 		struct disklabel *lp;
1524 
1525 		if ((error = vndlock(vnd)) != 0)
1526 			return error;
1527 
1528 		vnd->sc_flags |= VNF_LABELLING;
1529 
1530 #ifdef __HAVE_OLD_DISKLABEL
1531 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
1532 			memset(&newlabel, 0, sizeof newlabel);
1533 			memcpy(&newlabel, data, sizeof (struct olddisklabel));
1534 			lp = &newlabel;
1535 		} else
1536 #endif
1537 		lp = (struct disklabel *)data;
1538 
1539 		error = setdisklabel(vnd->sc_dkdev.dk_label,
1540 		    lp, 0, vnd->sc_dkdev.dk_cpulabel);
1541 		if (error == 0) {
1542 			if (cmd == DIOCWDINFO
1543 #ifdef __HAVE_OLD_DISKLABEL
1544 			    || cmd == ODIOCWDINFO
1545 #endif
1546 			   )
1547 				error = writedisklabel(VNDLABELDEV(dev),
1548 				    vndstrategy, vnd->sc_dkdev.dk_label,
1549 				    vnd->sc_dkdev.dk_cpulabel);
1550 		}
1551 
1552 		vnd->sc_flags &= ~VNF_LABELLING;
1553 
1554 		vndunlock(vnd);
1555 
1556 		if (error)
1557 			return error;
1558 		break;
1559 	}
1560 
1561 	case DIOCKLABEL:
1562 		if (*(int *)data != 0)
1563 			vnd->sc_flags |= VNF_KLABEL;
1564 		else
1565 			vnd->sc_flags &= ~VNF_KLABEL;
1566 		break;
1567 
1568 	case DIOCWLABEL:
1569 		if (*(int *)data != 0)
1570 			vnd->sc_flags |= VNF_WLABEL;
1571 		else
1572 			vnd->sc_flags &= ~VNF_WLABEL;
1573 		break;
1574 
1575 	case DIOCGDEFLABEL:
1576 		vndgetdefaultlabel(vnd, (struct disklabel *)data);
1577 		break;
1578 
1579 #ifdef __HAVE_OLD_DISKLABEL
1580 	case ODIOCGDEFLABEL:
1581 		vndgetdefaultlabel(vnd, &newlabel);
1582 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1583 			return ENOTTY;
1584 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1585 		break;
1586 #endif
1587 
1588 	case DIOCCACHESYNC:
1589 		vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1590 		error = VOP_FSYNC(vnd->sc_vp, vnd->sc_cred,
1591 		    FSYNC_WAIT | FSYNC_DATAONLY | FSYNC_CACHE, 0, 0);
1592 		VOP_UNLOCK(vnd->sc_vp);
1593 		return error;
1594 
1595 	default:
1596 		return ENOTTY;
1597 	}
1598 
1599 	return 0;
1600 }
1601 
1602 /*
1603  * Duplicate the current processes' credentials.  Since we are called only
1604  * as the result of a SET ioctl and only root can do that, any future access
1605  * to this "disk" is essentially as root.  Note that credentials may change
1606  * if some other uid can write directly to the mapped file (NFS).
1607  */
1608 static int
1609 vndsetcred(struct vnd_softc *vnd, kauth_cred_t cred)
1610 {
1611 	struct uio auio;
1612 	struct iovec aiov;
1613 	char *tmpbuf;
1614 	int error;
1615 
1616 	vnd->sc_cred = kauth_cred_dup(cred);
1617 	tmpbuf = malloc(DEV_BSIZE, M_TEMP, M_WAITOK);
1618 
1619 	/* XXX: Horrible kludge to establish credentials for NFS */
1620 	aiov.iov_base = tmpbuf;
1621 	aiov.iov_len = min(DEV_BSIZE, dbtob(vnd->sc_size));
1622 	auio.uio_iov = &aiov;
1623 	auio.uio_iovcnt = 1;
1624 	auio.uio_offset = 0;
1625 	auio.uio_rw = UIO_READ;
1626 	auio.uio_resid = aiov.iov_len;
1627 	UIO_SETUP_SYSSPACE(&auio);
1628 	vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1629 	error = VOP_READ(vnd->sc_vp, &auio, 0, vnd->sc_cred);
1630 	if (error == 0) {
1631 		/*
1632 		 * Because vnd does all IO directly through the vnode
1633 		 * we need to flush (at least) the buffer from the above
1634 		 * VOP_READ from the buffer cache to prevent cache
1635 		 * incoherencies.  Also, be careful to write dirty
1636 		 * buffers back to stable storage.
1637 		 */
1638 		error = vinvalbuf(vnd->sc_vp, V_SAVE, vnd->sc_cred,
1639 			    curlwp, 0, 0);
1640 	}
1641 	VOP_UNLOCK(vnd->sc_vp);
1642 
1643 	free(tmpbuf, M_TEMP);
1644 	return error;
1645 }
1646 
1647 /*
1648  * Set maxactive based on FS type
1649  */
1650 static void
1651 vndthrottle(struct vnd_softc *vnd, struct vnode *vp)
1652 {
1653 
1654 	if (vp->v_tag == VT_NFS)
1655 		vnd->sc_maxactive = 2;
1656 	else
1657 		vnd->sc_maxactive = 8;
1658 
1659 	if (vnd->sc_maxactive < 1)
1660 		vnd->sc_maxactive = 1;
1661 }
1662 
1663 #if 0
1664 static void
1665 vndshutdown(void)
1666 {
1667 	struct vnd_softc *vnd;
1668 
1669 	for (vnd = &vnd_softc[0]; vnd < &vnd_softc[numvnd]; vnd++)
1670 		if (vnd->sc_flags & VNF_INITED)
1671 			vndclear(vnd);
1672 }
1673 #endif
1674 
1675 static void
1676 vndclear(struct vnd_softc *vnd, int myminor)
1677 {
1678 	struct vnode *vp = vnd->sc_vp;
1679 	int fflags = FREAD;
1680 	int bmaj, cmaj, i, mn;
1681 	int s;
1682 
1683 #ifdef DEBUG
1684 	if (vnddebug & VDB_FOLLOW)
1685 		printf("vndclear(%p): vp %p\n", vnd, vp);
1686 #endif
1687 	/* locate the major number */
1688 	bmaj = bdevsw_lookup_major(&vnd_bdevsw);
1689 	cmaj = cdevsw_lookup_major(&vnd_cdevsw);
1690 
1691 	/* Nuke the vnodes for any open instances */
1692 	for (i = 0; i < MAXPARTITIONS; i++) {
1693 		mn = DISKMINOR(device_unit(vnd->sc_dev), i);
1694 		vdevgone(bmaj, mn, mn, VBLK);
1695 		if (mn != myminor) /* XXX avoid to kill own vnode */
1696 			vdevgone(cmaj, mn, mn, VCHR);
1697 	}
1698 
1699 	if ((vnd->sc_flags & VNF_READONLY) == 0)
1700 		fflags |= FWRITE;
1701 
1702 	s = splbio();
1703 	bufq_drain(vnd->sc_tab);
1704 	splx(s);
1705 
1706 	vnd->sc_flags |= VNF_VUNCONF;
1707 	wakeup(&vnd->sc_tab);
1708 	while (vnd->sc_flags & VNF_KTHREAD)
1709 		tsleep(&vnd->sc_kthread, PRIBIO, "vnthr", 0);
1710 
1711 #ifdef VND_COMPRESSION
1712 	/* free the compressed file buffers */
1713 	if (vnd->sc_flags & VNF_COMP) {
1714 		if (vnd->sc_comp_offsets) {
1715 			free(vnd->sc_comp_offsets, M_DEVBUF);
1716 			vnd->sc_comp_offsets = NULL;
1717 		}
1718 		if (vnd->sc_comp_buff) {
1719 			free(vnd->sc_comp_buff, M_DEVBUF);
1720 			vnd->sc_comp_buff = NULL;
1721 		}
1722 		if (vnd->sc_comp_decombuf) {
1723 			free(vnd->sc_comp_decombuf, M_DEVBUF);
1724 			vnd->sc_comp_decombuf = NULL;
1725 		}
1726 	}
1727 #endif /* VND_COMPRESSION */
1728 	vnd->sc_flags &=
1729 	    ~(VNF_INITED | VNF_READONLY | VNF_KLABEL | VNF_VLABEL
1730 	      | VNF_VUNCONF | VNF_COMP | VNF_CLEARING);
1731 	if (vp == NULL)
1732 		panic("vndclear: null vp");
1733 	(void) vn_close(vp, fflags, vnd->sc_cred);
1734 	kauth_cred_free(vnd->sc_cred);
1735 	vnd->sc_vp = NULL;
1736 	vnd->sc_cred = NULL;
1737 	vnd->sc_size = 0;
1738 }
1739 
1740 static int
1741 vndsize(dev_t dev)
1742 {
1743 	struct vnd_softc *sc;
1744 	struct disklabel *lp;
1745 	int part, unit, omask;
1746 	int size;
1747 
1748 	unit = vndunit(dev);
1749 	sc = device_lookup_private(&vnd_cd, unit);
1750 	if (sc == NULL)
1751 		return -1;
1752 
1753 	if ((sc->sc_flags & VNF_INITED) == 0)
1754 		return -1;
1755 
1756 	part = DISKPART(dev);
1757 	omask = sc->sc_dkdev.dk_openmask & (1 << part);
1758 	lp = sc->sc_dkdev.dk_label;
1759 
1760 	if (omask == 0 && vndopen(dev, 0, S_IFBLK, curlwp))	/* XXX */
1761 		return -1;
1762 
1763 	if (lp->d_partitions[part].p_fstype != FS_SWAP)
1764 		size = -1;
1765 	else
1766 		size = lp->d_partitions[part].p_size *
1767 		    (lp->d_secsize / DEV_BSIZE);
1768 
1769 	if (omask == 0 && vndclose(dev, 0, S_IFBLK, curlwp))	/* XXX */
1770 		return -1;
1771 
1772 	return size;
1773 }
1774 
1775 static int
1776 vnddump(dev_t dev, daddr_t blkno, void *va,
1777     size_t size)
1778 {
1779 
1780 	/* Not implemented. */
1781 	return ENXIO;
1782 }
1783 
1784 static void
1785 vndgetdefaultlabel(struct vnd_softc *sc, struct disklabel *lp)
1786 {
1787 	struct vndgeom *vng = &sc->sc_geom;
1788 	struct partition *pp;
1789 	unsigned spb;
1790 
1791 	memset(lp, 0, sizeof(*lp));
1792 
1793 	spb = vng->vng_secsize / DEV_BSIZE;
1794 	if (sc->sc_size / spb > UINT32_MAX)
1795 		lp->d_secperunit = UINT32_MAX;
1796 	else
1797 		lp->d_secperunit = sc->sc_size / spb;
1798 	lp->d_secsize = vng->vng_secsize;
1799 	lp->d_nsectors = vng->vng_nsectors;
1800 	lp->d_ntracks = vng->vng_ntracks;
1801 	lp->d_ncylinders = vng->vng_ncylinders;
1802 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1803 
1804 	strncpy(lp->d_typename, "vnd", sizeof(lp->d_typename));
1805 	lp->d_type = DKTYPE_VND;
1806 	strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
1807 	lp->d_rpm = 3600;
1808 	lp->d_interleave = 1;
1809 	lp->d_flags = 0;
1810 
1811 	pp = &lp->d_partitions[RAW_PART];
1812 	pp->p_offset = 0;
1813 	pp->p_size = lp->d_secperunit;
1814 	pp->p_fstype = FS_UNUSED;
1815 	lp->d_npartitions = RAW_PART + 1;
1816 
1817 	lp->d_magic = DISKMAGIC;
1818 	lp->d_magic2 = DISKMAGIC;
1819 	lp->d_checksum = dkcksum(lp);
1820 }
1821 
1822 /*
1823  * Read the disklabel from a vnd.  If one is not present, create a fake one.
1824  */
1825 static void
1826 vndgetdisklabel(dev_t dev, struct vnd_softc *sc)
1827 {
1828 	const char *errstring;
1829 	struct disklabel *lp = sc->sc_dkdev.dk_label;
1830 	struct cpu_disklabel *clp = sc->sc_dkdev.dk_cpulabel;
1831 	int i;
1832 
1833 	memset(clp, 0, sizeof(*clp));
1834 
1835 	vndgetdefaultlabel(sc, lp);
1836 
1837 	/*
1838 	 * Call the generic disklabel extraction routine.
1839 	 */
1840 	errstring = readdisklabel(VNDLABELDEV(dev), vndstrategy, lp, clp);
1841 	if (errstring) {
1842 		/*
1843 		 * Lack of disklabel is common, but we print the warning
1844 		 * anyway, since it might contain other useful information.
1845 		 */
1846 		aprint_normal_dev(sc->sc_dev, "%s\n", errstring);
1847 
1848 		/*
1849 		 * For historical reasons, if there's no disklabel
1850 		 * present, all partitions must be FS_BSDFFS and
1851 		 * occupy the entire disk.
1852 		 */
1853 		for (i = 0; i < MAXPARTITIONS; i++) {
1854 			/*
1855 			 * Don't wipe out port specific hack (such as
1856 			 * dos partition hack of i386 port).
1857 			 */
1858 			if (lp->d_partitions[i].p_size != 0)
1859 				continue;
1860 
1861 			lp->d_partitions[i].p_size = lp->d_secperunit;
1862 			lp->d_partitions[i].p_offset = 0;
1863 			lp->d_partitions[i].p_fstype = FS_BSDFFS;
1864 		}
1865 
1866 		strncpy(lp->d_packname, "default label",
1867 		    sizeof(lp->d_packname));
1868 
1869 		lp->d_npartitions = MAXPARTITIONS;
1870 		lp->d_checksum = dkcksum(lp);
1871 	}
1872 }
1873 
1874 /*
1875  * Wait interruptibly for an exclusive lock.
1876  *
1877  * XXX
1878  * Several drivers do this; it should be abstracted and made MP-safe.
1879  */
1880 static int
1881 vndlock(struct vnd_softc *sc)
1882 {
1883 	int error;
1884 
1885 	while ((sc->sc_flags & VNF_LOCKED) != 0) {
1886 		sc->sc_flags |= VNF_WANTED;
1887 		if ((error = tsleep(sc, PRIBIO | PCATCH, "vndlck", 0)) != 0)
1888 			return error;
1889 	}
1890 	sc->sc_flags |= VNF_LOCKED;
1891 	return 0;
1892 }
1893 
1894 /*
1895  * Unlock and wake up any waiters.
1896  */
1897 static void
1898 vndunlock(struct vnd_softc *sc)
1899 {
1900 
1901 	sc->sc_flags &= ~VNF_LOCKED;
1902 	if ((sc->sc_flags & VNF_WANTED) != 0) {
1903 		sc->sc_flags &= ~VNF_WANTED;
1904 		wakeup(sc);
1905 	}
1906 }
1907 
1908 #ifdef VND_COMPRESSION
1909 /* compressed file read */
1910 static void
1911 compstrategy(struct buf *bp, off_t bn)
1912 {
1913 	int error;
1914 	int unit = vndunit(bp->b_dev);
1915 	struct vnd_softc *vnd =
1916 	    device_lookup_private(&vnd_cd, unit);
1917 	u_int32_t comp_block;
1918 	struct uio auio;
1919 	char *addr;
1920 	int s;
1921 
1922 	/* set up constants for data move */
1923 	auio.uio_rw = UIO_READ;
1924 	UIO_SETUP_SYSSPACE(&auio);
1925 
1926 	/* read, and transfer the data */
1927 	addr = bp->b_data;
1928 	bp->b_resid = bp->b_bcount;
1929 	s = splbio();
1930 	while (bp->b_resid > 0) {
1931 		unsigned length;
1932 		size_t length_in_buffer;
1933 		u_int32_t offset_in_buffer;
1934 		struct iovec aiov;
1935 
1936 		/* calculate the compressed block number */
1937 		comp_block = bn / (off_t)vnd->sc_comp_blksz;
1938 
1939 		/* check for good block number */
1940 		if (comp_block >= vnd->sc_comp_numoffs) {
1941 			bp->b_error = EINVAL;
1942 			splx(s);
1943 			return;
1944 		}
1945 
1946 		/* read in the compressed block, if not in buffer */
1947 		if (comp_block != vnd->sc_comp_buffblk) {
1948 			length = vnd->sc_comp_offsets[comp_block + 1] -
1949 			    vnd->sc_comp_offsets[comp_block];
1950 			vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1951 			error = vn_rdwr(UIO_READ, vnd->sc_vp, vnd->sc_comp_buff,
1952 			    length, vnd->sc_comp_offsets[comp_block],
1953 			    UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, vnd->sc_cred,
1954 			    NULL, NULL);
1955 			if (error) {
1956 				bp->b_error = error;
1957 				VOP_UNLOCK(vnd->sc_vp);
1958 				splx(s);
1959 				return;
1960 			}
1961 			/* uncompress the buffer */
1962 			vnd->sc_comp_stream.next_in = vnd->sc_comp_buff;
1963 			vnd->sc_comp_stream.avail_in = length;
1964 			vnd->sc_comp_stream.next_out = vnd->sc_comp_decombuf;
1965 			vnd->sc_comp_stream.avail_out = vnd->sc_comp_blksz;
1966 			inflateReset(&vnd->sc_comp_stream);
1967 			error = inflate(&vnd->sc_comp_stream, Z_FINISH);
1968 			if (error != Z_STREAM_END) {
1969 				if (vnd->sc_comp_stream.msg)
1970 					aprint_normal_dev(vnd->sc_dev,
1971 					    "compressed file, %s\n",
1972 					    vnd->sc_comp_stream.msg);
1973 				bp->b_error = EBADMSG;
1974 				VOP_UNLOCK(vnd->sc_vp);
1975 				splx(s);
1976 				return;
1977 			}
1978 			vnd->sc_comp_buffblk = comp_block;
1979 			VOP_UNLOCK(vnd->sc_vp);
1980 		}
1981 
1982 		/* transfer the usable uncompressed data */
1983 		offset_in_buffer = bn % (off_t)vnd->sc_comp_blksz;
1984 		length_in_buffer = vnd->sc_comp_blksz - offset_in_buffer;
1985 		if (length_in_buffer > bp->b_resid)
1986 			length_in_buffer = bp->b_resid;
1987 		auio.uio_iov = &aiov;
1988 		auio.uio_iovcnt = 1;
1989 		aiov.iov_base = addr;
1990 		aiov.iov_len = length_in_buffer;
1991 		auio.uio_resid = aiov.iov_len;
1992 		auio.uio_offset = 0;
1993 		error = uiomove(vnd->sc_comp_decombuf + offset_in_buffer,
1994 		    length_in_buffer, &auio);
1995 		if (error) {
1996 			bp->b_error = error;
1997 			splx(s);
1998 			return;
1999 		}
2000 
2001 		bn += length_in_buffer;
2002 		addr += length_in_buffer;
2003 		bp->b_resid -= length_in_buffer;
2004 	}
2005 	splx(s);
2006 }
2007 
2008 /* compression memory allocation routines */
2009 static void *
2010 vnd_alloc(void *aux, u_int items, u_int siz)
2011 {
2012 	return malloc(items * siz, M_TEMP, M_NOWAIT);
2013 }
2014 
2015 static void
2016 vnd_free(void *aux, void *ptr)
2017 {
2018 	free(ptr, M_TEMP);
2019 }
2020 #endif /* VND_COMPRESSION */
2021 
2022 static void
2023 vnd_set_geometry(struct vnd_softc *vnd)
2024 {
2025 	struct disk_geom *dg = &vnd->sc_dkdev.dk_geom;
2026 
2027 	memset(dg, 0, sizeof(*dg));
2028 
2029 	dg->dg_secperunit = (int64_t)vnd->sc_geom.vng_nsectors *
2030 	    vnd->sc_geom.vng_ntracks * vnd->sc_geom.vng_ncylinders;
2031 	dg->dg_secsize = vnd->sc_geom.vng_secsize;
2032 	dg->dg_nsectors = vnd->sc_geom.vng_nsectors;
2033 	dg->dg_ntracks = vnd->sc_geom.vng_ntracks;
2034 	dg->dg_ncylinders = vnd->sc_geom.vng_ncylinders;
2035 
2036 #ifdef DEBUG
2037 	if (vnddebug & VDB_LABEL) {
2038 		printf("dg->dg_secperunit: %" PRId64 "\n", dg->dg_secperunit);
2039 		printf("dg->dg_ncylinders: %u\n", dg->dg_ncylinders);
2040 	}
2041 #endif
2042 	disk_set_info(vnd->sc_dev, &vnd->sc_dkdev, NULL);
2043 }
2044 
2045 #ifdef VND_COMPRESSION
2046 #define VND_DEPENDS "zlib"
2047 #else
2048 #define VND_DEPENDS NULL
2049 #endif
2050 
2051 MODULE(MODULE_CLASS_DRIVER, vnd, VND_DEPENDS);
2052 
2053 #ifdef _MODULE
2054 int vnd_bmajor = -1, vnd_cmajor = -1;
2055 
2056 CFDRIVER_DECL(vnd, DV_DISK, NULL);
2057 #endif
2058 
2059 static int
2060 vnd_modcmd(modcmd_t cmd, void *arg)
2061 {
2062 	int error = 0;
2063 
2064 	switch (cmd) {
2065 	case MODULE_CMD_INIT:
2066 #ifdef _MODULE
2067 		error = config_cfdriver_attach(&vnd_cd);
2068 		if (error)
2069 			break;
2070 
2071 		error = config_cfattach_attach(vnd_cd.cd_name, &vnd_ca);
2072 	        if (error) {
2073 			config_cfdriver_detach(&vnd_cd);
2074 			aprint_error("%s: unable to register cfattach for \n"
2075 			    "%s, error %d", __func__, vnd_cd.cd_name, error);
2076 			break;
2077 		}
2078 
2079                 /*
2080                  * Attach the {b,c}devsw's
2081                  */
2082 		error = devsw_attach("vnd", &vnd_bdevsw, &vnd_bmajor,
2083 		    &vnd_cdevsw, &vnd_cmajor);
2084                 /*
2085                  * If devsw_attach fails, remove from autoconf database
2086                  */
2087 		if (error) {
2088 			config_cfattach_detach(vnd_cd.cd_name, &vnd_ca);
2089 			config_cfdriver_detach(&vnd_cd);
2090                         aprint_error("%s: unable to attach %s devsw, "
2091                             "error %d", __func__, vnd_cd.cd_name, error);
2092 			break;
2093 		}
2094 #endif
2095 		break;
2096 
2097 	case MODULE_CMD_FINI:
2098 #ifdef _MODULE
2099                 /*
2100                  * Remove {b,c}devsw's
2101                  */
2102 		devsw_detach(&vnd_bdevsw, &vnd_cdevsw);
2103 
2104                 /*
2105                  * Now remove device from autoconf database
2106                  */
2107 		error = config_cfattach_detach(vnd_cd.cd_name, &vnd_ca);
2108                 if (error) {
2109                         (void)devsw_attach("vnd", &vnd_bdevsw, &vnd_bmajor,
2110                             &vnd_cdevsw, &vnd_cmajor);
2111                         aprint_error("%s: failed to detach %s cfattach, "
2112                             "error %d\n", __func__, vnd_cd.cd_name, error);
2113                         break;
2114                 }
2115                 error = config_cfdriver_detach(&vnd_cd);
2116                 if (error) {
2117                         (void)config_cfattach_attach(vnd_cd.cd_name, &vnd_ca);
2118                         (void)devsw_attach("vnd", &vnd_bdevsw, &vnd_bmajor,
2119                             &vnd_cdevsw, &vnd_cmajor);
2120                         aprint_error("%s: failed to detach %s cfdriver, "
2121                             "error %d\n", __func__, vnd_cd.cd_name, error);
2122                         break;
2123                 }
2124 #endif
2125 		break;
2126 
2127 	case MODULE_CMD_STAT:
2128 		return ENOTTY;
2129 
2130 	default:
2131 		return ENOTTY;
2132 	}
2133 
2134 	return error;
2135 }
2136