xref: /netbsd-src/sys/dev/vnd.c (revision 627f7eb200a4419d89b531d55fccd2ee3ffdcde0)
1 /*	$NetBSD: vnd.c,v 1.278 2021/01/04 16:17:26 mlelstv Exp $	*/
2 
3 /*-
4  * Copyright (c) 1996, 1997, 1998, 2008, 2020 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 1.278 2021/01/04 16:17:26 mlelstv 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/fstrans.h>
118 #include <sys/file.h>
119 #include <sys/uio.h>
120 #include <sys/conf.h>
121 #include <sys/kauth.h>
122 #include <sys/module.h>
123 #include <sys/compat_stub.h>
124 #include <sys/atomic.h>
125 
126 #include <net/zlib.h>
127 
128 #include <miscfs/genfs/genfs.h>
129 #include <miscfs/specfs/specdev.h>
130 
131 #include <dev/dkvar.h>
132 #include <dev/vndvar.h>
133 
134 #include "ioconf.h"
135 
136 #if defined(VNDDEBUG) && !defined(DEBUG)
137 #define DEBUG
138 #endif
139 
140 #ifdef DEBUG
141 int dovndcluster = 1;
142 #define VDB_FOLLOW	0x01
143 #define VDB_INIT	0x02
144 #define VDB_IO		0x04
145 #define VDB_LABEL	0x08
146 int vnddebug = 0;
147 #endif
148 
149 #define vndunit(x)	DISKUNIT(x)
150 
151 struct vndxfer {
152 	struct buf vx_buf;
153 	struct vnd_softc *vx_vnd;
154 };
155 #define	VND_BUFTOXFER(bp)	((struct vndxfer *)(void *)bp)
156 
157 #define VND_GETXFER(vnd)	pool_get(&(vnd)->sc_vxpool, PR_WAITOK)
158 #define VND_PUTXFER(vnd, vx)	pool_put(&(vnd)->sc_vxpool, (vx))
159 
160 #define VNDLABELDEV(dev) \
161     (MAKEDISKDEV(major((dev)), vndunit((dev)), RAW_PART))
162 
163 #define	VND_MAXPENDING(vnd)	((vnd)->sc_maxactive * 4)
164 #define	VND_MAXPAGES(vnd)	(1024 * 1024 / PAGE_SIZE)
165 
166 
167 static void	vndclear(struct vnd_softc *, int);
168 static int	vnddoclear(struct vnd_softc *, int, int, bool);
169 static int	vndsetcred(struct vnd_softc *, kauth_cred_t);
170 static void	vndthrottle(struct vnd_softc *, struct vnode *);
171 static void	vndiodone(struct buf *);
172 #if 0
173 static void	vndshutdown(void);
174 #endif
175 
176 static void	vndgetdefaultlabel(struct vnd_softc *, struct disklabel *);
177 static void	vndgetdisklabel(dev_t, struct vnd_softc *);
178 
179 static int	vndlock(struct vnd_softc *);
180 static void	vndunlock(struct vnd_softc *);
181 #ifdef VND_COMPRESSION
182 static void	compstrategy(struct buf *, off_t);
183 static void	*vnd_alloc(void *, u_int, u_int);
184 static void	vnd_free(void *, void *);
185 #endif /* VND_COMPRESSION */
186 
187 static void	vndthread(void *);
188 static bool	vnode_has_op(const struct vnode *, int);
189 static void	handle_with_rdwr(struct vnd_softc *, const struct buf *,
190 		    struct buf *);
191 static void	handle_with_strategy(struct vnd_softc *, const struct buf *,
192 		    struct buf *);
193 static void	vnd_set_geometry(struct vnd_softc *);
194 
195 static dev_type_open(vndopen);
196 static dev_type_close(vndclose);
197 static dev_type_read(vndread);
198 static dev_type_write(vndwrite);
199 static dev_type_ioctl(vndioctl);
200 static dev_type_strategy(vndstrategy);
201 static dev_type_dump(vnddump);
202 static dev_type_size(vndsize);
203 
204 const struct bdevsw vnd_bdevsw = {
205 	.d_open = vndopen,
206 	.d_close = vndclose,
207 	.d_strategy = vndstrategy,
208 	.d_ioctl = vndioctl,
209 	.d_dump = vnddump,
210 	.d_psize = vndsize,
211 	.d_discard = nodiscard,
212 	.d_flag = D_DISK
213 };
214 
215 const struct cdevsw vnd_cdevsw = {
216 	.d_open = vndopen,
217 	.d_close = vndclose,
218 	.d_read = vndread,
219 	.d_write = vndwrite,
220 	.d_ioctl = vndioctl,
221 	.d_stop = nostop,
222 	.d_tty = notty,
223 	.d_poll = nopoll,
224 	.d_mmap = nommap,
225 	.d_kqfilter = nokqfilter,
226 	.d_discard = nodiscard,
227 	.d_flag = D_DISK
228 };
229 
230 static int	vnd_match(device_t, cfdata_t, void *);
231 static void	vnd_attach(device_t, device_t, void *);
232 static int	vnd_detach(device_t, int);
233 
234 CFATTACH_DECL3_NEW(vnd, sizeof(struct vnd_softc),
235     vnd_match, vnd_attach, vnd_detach, NULL, NULL, NULL, DVF_DETACH_SHUTDOWN);
236 
237 static struct vnd_softc	*vnd_spawn(int);
238 static int	vnd_destroy(device_t);
239 
240 static const struct	dkdriver vnddkdriver = {
241 	.d_strategy = vndstrategy,
242 	.d_minphys = minphys
243 };
244 
245 void
246 vndattach(int num)
247 {
248 	int error;
249 
250 	error = config_cfattach_attach(vnd_cd.cd_name, &vnd_ca);
251 	if (error)
252 		aprint_error("%s: unable to register cfattach, error = %d\n",
253 		    vnd_cd.cd_name, error);
254 }
255 
256 static int
257 vnd_match(device_t self, cfdata_t cfdata, void *aux)
258 {
259 
260 	return 1;
261 }
262 
263 static void
264 vnd_attach(device_t parent, device_t self, void *aux)
265 {
266 	struct vnd_softc *sc = device_private(self);
267 
268 	sc->sc_dev = self;
269 	sc->sc_comp_offsets = NULL;
270 	sc->sc_comp_buff = NULL;
271 	sc->sc_comp_decombuf = NULL;
272 	bufq_alloc(&sc->sc_tab, "disksort", BUFQ_SORT_RAWBLOCK);
273 	disk_init(&sc->sc_dkdev, device_xname(self), &vnddkdriver);
274 	if (!pmf_device_register(self, NULL, NULL))
275 		aprint_error_dev(self, "couldn't establish power handler\n");
276 }
277 
278 static int
279 vnd_detach(device_t self, int flags)
280 {
281 	int error;
282 	struct vnd_softc *sc = device_private(self);
283 
284 	if (sc->sc_flags & VNF_INITED) {
285 		error = vnddoclear(sc, 0, -1, (flags & DETACH_FORCE) != 0);
286 		if (error != 0)
287 			return error;
288 	}
289 
290 	pmf_device_deregister(self);
291 	bufq_free(sc->sc_tab);
292 	disk_destroy(&sc->sc_dkdev);
293 
294 	return 0;
295 }
296 
297 static struct vnd_softc *
298 vnd_spawn(int unit)
299 {
300 	cfdata_t cf;
301 
302 	cf = malloc(sizeof(*cf), M_DEVBUF, M_WAITOK);
303 	cf->cf_name = vnd_cd.cd_name;
304 	cf->cf_atname = vnd_cd.cd_name;
305 	cf->cf_unit = unit;
306 	cf->cf_fstate = FSTATE_STAR;
307 
308 	return device_private(config_attach_pseudo(cf));
309 }
310 
311 static int
312 vnd_destroy(device_t dev)
313 {
314 	int error;
315 	cfdata_t cf;
316 
317 	cf = device_cfdata(dev);
318 	error = config_detach(dev, DETACH_QUIET);
319 	if (error)
320 		return error;
321 	free(cf, M_DEVBUF);
322 	return 0;
323 }
324 
325 static int
326 vndopen(dev_t dev, int flags, int mode, struct lwp *l)
327 {
328 	int unit = vndunit(dev);
329 	struct vnd_softc *sc;
330 	int error = 0, part, pmask;
331 	struct disklabel *lp;
332 
333 #ifdef DEBUG
334 	if (vnddebug & VDB_FOLLOW)
335 		printf("vndopen(0x%"PRIx64", 0x%x, 0x%x, %p)\n", dev, flags, mode, l);
336 #endif
337 	sc = device_lookup_private(&vnd_cd, unit);
338 	if (sc == NULL) {
339 		sc = vnd_spawn(unit);
340 		if (sc == NULL)
341 			return ENOMEM;
342 
343 		/* compatibility, keep disklabel after close */
344 		sc->sc_flags = VNF_KLABEL;
345 	}
346 
347 	if ((error = vndlock(sc)) != 0)
348 		return error;
349 
350 	mutex_enter(&sc->sc_dkdev.dk_openlock);
351 
352 	if ((sc->sc_flags & VNF_CLEARING) != 0) {
353 		error = ENXIO;
354 		goto done;
355 	}
356 
357 	lp = sc->sc_dkdev.dk_label;
358 
359 	part = DISKPART(dev);
360 	pmask = (1 << part);
361 
362 	if (sc->sc_dkdev.dk_nwedges != 0 && part != RAW_PART) {
363 		error = EBUSY;
364 		goto done;
365 	}
366 
367 	if (sc->sc_flags & VNF_INITED) {
368 		if ((sc->sc_dkdev.dk_openmask & ~(1<<RAW_PART)) != 0) {
369 			/*
370 			 * If any non-raw partition is open, but the disk
371 			 * has been invalidated, disallow further opens.
372 			 */
373 			if ((sc->sc_flags & VNF_VLABEL) == 0) {
374 				error = EIO;
375 				goto done;
376 			}
377 		} else {
378 			/*
379 			 * Load the partition info if not already loaded.
380 			 */
381 			if ((sc->sc_flags & VNF_VLABEL) == 0) {
382 				sc->sc_flags |= VNF_VLABEL;
383 				vndgetdisklabel(dev, sc);
384 			}
385 		}
386 	}
387 
388 	/* Check that the partitions exists. */
389 	if (part != RAW_PART) {
390 		if (((sc->sc_flags & VNF_INITED) == 0) ||
391 		    ((part >= lp->d_npartitions) ||
392 		     (lp->d_partitions[part].p_fstype == FS_UNUSED))) {
393 			error = ENXIO;
394 			goto done;
395 		}
396 	}
397 
398 	/* Prevent our unit from being unconfigured while open. */
399 	switch (mode) {
400 	case S_IFCHR:
401 		sc->sc_dkdev.dk_copenmask |= pmask;
402 		break;
403 
404 	case S_IFBLK:
405 		sc->sc_dkdev.dk_bopenmask |= pmask;
406 		break;
407 	}
408 	sc->sc_dkdev.dk_openmask =
409 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
410 
411  done:
412 	mutex_exit(&sc->sc_dkdev.dk_openlock);
413 	vndunlock(sc);
414 	return error;
415 }
416 
417 static int
418 vndclose(dev_t dev, int flags, int mode, struct lwp *l)
419 {
420 	int unit = vndunit(dev);
421 	struct vnd_softc *sc;
422 	int error = 0, part;
423 
424 #ifdef DEBUG
425 	if (vnddebug & VDB_FOLLOW)
426 		printf("vndclose(0x%"PRIx64", 0x%x, 0x%x, %p)\n", dev, flags, mode, l);
427 #endif
428 	sc = device_lookup_private(&vnd_cd, unit);
429 	if (sc == NULL)
430 		return ENXIO;
431 
432 	if ((error = vndlock(sc)) != 0)
433 		return error;
434 
435 	mutex_enter(&sc->sc_dkdev.dk_openlock);
436 
437 	part = DISKPART(dev);
438 
439 	/* ...that much closer to allowing unconfiguration... */
440 	switch (mode) {
441 	case S_IFCHR:
442 		sc->sc_dkdev.dk_copenmask &= ~(1 << part);
443 		break;
444 
445 	case S_IFBLK:
446 		sc->sc_dkdev.dk_bopenmask &= ~(1 << part);
447 		break;
448 	}
449 	sc->sc_dkdev.dk_openmask =
450 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
451 
452 	/* are we last opener ? */
453 	if (sc->sc_dkdev.dk_openmask == 0) {
454 		if ((sc->sc_flags & VNF_KLABEL) == 0)
455 			sc->sc_flags &= ~VNF_VLABEL;
456 	}
457 
458 	mutex_exit(&sc->sc_dkdev.dk_openlock);
459 
460 	vndunlock(sc);
461 
462 	if ((sc->sc_flags & VNF_INITED) == 0) {
463 		if ((error = vnd_destroy(sc->sc_dev)) != 0) {
464 			aprint_error_dev(sc->sc_dev,
465 			    "unable to detach instance\n");
466 			return error;
467 		}
468 	}
469 
470 	return 0;
471 }
472 
473 /*
474  * Queue the request, and wakeup the kernel thread to handle it.
475  */
476 static void
477 vndstrategy(struct buf *bp)
478 {
479 	int unit = vndunit(bp->b_dev);
480 	struct vnd_softc *vnd =
481 	    device_lookup_private(&vnd_cd, unit);
482 	struct disklabel *lp;
483 	daddr_t blkno;
484 	int s = splbio();
485 
486 	if (vnd == NULL) {
487 		bp->b_error = ENXIO;
488 		goto done;
489 	}
490 	lp = vnd->sc_dkdev.dk_label;
491 
492 	if ((vnd->sc_flags & VNF_INITED) == 0) {
493 		bp->b_error = ENXIO;
494 		goto done;
495 	}
496 
497 	/*
498 	 * The transfer must be a whole number of blocks.
499 	 */
500 	if ((bp->b_bcount % lp->d_secsize) != 0) {
501 		bp->b_error = EINVAL;
502 		goto done;
503 	}
504 
505 	/*
506 	 * check if we're read-only.
507 	 */
508 	if ((vnd->sc_flags & VNF_READONLY) && !(bp->b_flags & B_READ)) {
509 		bp->b_error = EACCES;
510 		goto done;
511 	}
512 
513 	/* If it's a nil transfer, wake up the top half now. */
514 	if (bp->b_bcount == 0) {
515 		goto done;
516 	}
517 
518 	/*
519 	 * Do bounds checking and adjust transfer.  If there's an error,
520 	 * the bounds check will flag that for us.
521 	 */
522 	if (DISKPART(bp->b_dev) == RAW_PART) {
523 		if (bounds_check_with_mediasize(bp, DEV_BSIZE,
524 		    vnd->sc_size) <= 0)
525 			goto done;
526 	} else {
527 		if (bounds_check_with_label(&vnd->sc_dkdev,
528 		    bp, vnd->sc_flags & (VNF_WLABEL|VNF_LABELLING)) <= 0)
529 			goto done;
530 	}
531 
532 	/*
533 	 * Put the block number in terms of the logical blocksize
534 	 * of the "device".
535 	 */
536 
537 	blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
538 
539 	/*
540 	 * Translate the partition-relative block number to an absolute.
541 	 */
542 	if (DISKPART(bp->b_dev) != RAW_PART) {
543 		struct partition *pp;
544 
545 		pp = &vnd->sc_dkdev.dk_label->d_partitions[
546 		    DISKPART(bp->b_dev)];
547 		blkno += pp->p_offset;
548 	}
549 	bp->b_rawblkno = blkno;
550 
551 #ifdef DEBUG
552 	if (vnddebug & VDB_FOLLOW)
553 		printf("vndstrategy(%p): unit %d\n", bp, unit);
554 #endif
555 	if ((vnd->sc_flags & VNF_USE_VN_RDWR)) {
556 		KASSERT(vnd->sc_pending >= 0 &&
557 		    vnd->sc_pending <= VND_MAXPENDING(vnd));
558 		while (vnd->sc_pending == VND_MAXPENDING(vnd))
559 			tsleep(&vnd->sc_pending, PRIBIO, "vndpc", 0);
560 		vnd->sc_pending++;
561 	}
562 	bufq_put(vnd->sc_tab, bp);
563 	wakeup(&vnd->sc_tab);
564 	splx(s);
565 	return;
566 
567 done:
568 	bp->b_resid = bp->b_bcount;
569 	biodone(bp);
570 	splx(s);
571 }
572 
573 static bool
574 vnode_has_strategy(struct vnd_softc *vnd)
575 {
576 	return vnode_has_op(vnd->sc_vp, VOFFSET(vop_bmap)) &&
577 	    vnode_has_op(vnd->sc_vp, VOFFSET(vop_strategy));
578 }
579 
580 /* Verify that I/O requests cannot be smaller than the
581  * smallest I/O size supported by the backend.
582  */
583 static bool
584 vnode_has_large_blocks(struct vnd_softc *vnd)
585 {
586 	u_int32_t vnd_secsize, iosize;
587 
588 	iosize = vnd->sc_iosize;
589 	vnd_secsize = vnd->sc_geom.vng_secsize;
590 
591 	return vnd_secsize % iosize != 0;
592 }
593 
594 /* XXX this function needs a reliable check to detect
595  * sparse files. Otherwise, bmap/strategy may be used
596  * and fail on non-allocated blocks. VOP_READ/VOP_WRITE
597  * works on sparse files.
598  */
599 #if notyet
600 static bool
601 vnode_strategy_probe(struct vnd_softc *vnd)
602 {
603 	int error;
604 	daddr_t nbn;
605 
606 	if (!vnode_has_strategy(vnd))
607 		return false;
608 
609 	if (vnode_has_large_blocks(vnd))
610 		return false;
611 
612 	/* Convert the first logical block number to its
613 	 * physical block number.
614 	 */
615 	error = 0;
616 	vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
617 	error = VOP_BMAP(vnd->sc_vp, 0, NULL, &nbn, NULL);
618 	VOP_UNLOCK(vnd->sc_vp);
619 
620 	/* Test if that worked. */
621 	if (error == 0 && (long)nbn == -1)
622 		return false;
623 
624 	return true;
625 }
626 #endif
627 
628 static void
629 vndthread(void *arg)
630 {
631 	struct vnd_softc *vnd = arg;
632 	int s;
633 
634 	/* Determine whether we can *use* VOP_BMAP and VOP_STRATEGY to
635 	 * directly access the backing vnode.  If we can, use these two
636 	 * operations to avoid messing with the local buffer cache.
637 	 * Otherwise fall back to regular VOP_READ/VOP_WRITE operations
638 	 * which are guaranteed to work with any file system. */
639 	if ((vnd->sc_flags & VNF_USE_VN_RDWR) == 0 &&
640 	    ! vnode_has_strategy(vnd))
641 		vnd->sc_flags |= VNF_USE_VN_RDWR;
642 
643 	/* VOP_STRATEGY can only be used if the backing vnode allows
644 	 * to access blocks as small as defined by the vnd geometry.
645 	 */
646 	if ((vnd->sc_flags & VNF_USE_VN_RDWR) == 0 &&
647 	    vnode_has_large_blocks(vnd))
648 		vnd->sc_flags |= VNF_USE_VN_RDWR;
649 
650 #ifdef DEBUG
651 	if (vnddebug & VDB_INIT)
652 		printf("vndthread: vp %p, %s\n", vnd->sc_vp,
653 		    (vnd->sc_flags & VNF_USE_VN_RDWR) == 0 ?
654 		    "using bmap/strategy operations" :
655 		    "using read/write operations");
656 #endif
657 
658 	s = splbio();
659 	vnd->sc_flags |= VNF_KTHREAD;
660 	wakeup(&vnd->sc_kthread);
661 
662 	/*
663 	 * Dequeue requests and serve them depending on the available
664 	 * vnode operations.
665 	 */
666 	while ((vnd->sc_flags & VNF_VUNCONF) == 0) {
667 		struct vndxfer *vnx;
668 		struct buf *obp;
669 		struct buf *bp;
670 
671 		obp = bufq_get(vnd->sc_tab);
672 		if (obp == NULL) {
673 			tsleep(&vnd->sc_tab, PRIBIO, "vndbp", 0);
674 			continue;
675 		};
676 		if ((vnd->sc_flags & VNF_USE_VN_RDWR)) {
677 			KASSERT(vnd->sc_pending > 0 &&
678 			    vnd->sc_pending <= VND_MAXPENDING(vnd));
679 			if (vnd->sc_pending-- == VND_MAXPENDING(vnd))
680 				wakeup(&vnd->sc_pending);
681 		}
682 		splx(s);
683 #ifdef DEBUG
684 		if (vnddebug & VDB_FOLLOW)
685 			printf("vndthread(%p)\n", obp);
686 #endif
687 
688 		if (vnd->sc_vp->v_mount == NULL) {
689 			obp->b_error = ENXIO;
690 			goto done;
691 		}
692 #ifdef VND_COMPRESSION
693 		/* handle a compressed read */
694 		if ((obp->b_flags & B_READ) != 0 && (vnd->sc_flags & VNF_COMP)) {
695 			off_t bn;
696 
697 			/* Convert to a byte offset within the file. */
698 			bn = obp->b_rawblkno *
699 			    vnd->sc_dkdev.dk_label->d_secsize;
700 
701 			compstrategy(obp, bn);
702 			goto done;
703 		}
704 #endif /* VND_COMPRESSION */
705 
706 		/*
707 		 * Allocate a header for this transfer and link it to the
708 		 * buffer
709 		 */
710 		s = splbio();
711 		vnx = VND_GETXFER(vnd);
712 		splx(s);
713 		vnx->vx_vnd = vnd;
714 
715 		s = splbio();
716 		while (vnd->sc_active >= vnd->sc_maxactive) {
717 			tsleep(&vnd->sc_tab, PRIBIO, "vndac", 0);
718 		}
719 		vnd->sc_active++;
720 		splx(s);
721 
722 		/* Instrumentation. */
723 		disk_busy(&vnd->sc_dkdev);
724 
725 		bp = &vnx->vx_buf;
726 		buf_init(bp);
727 		bp->b_flags = (obp->b_flags & (B_READ | B_PHYS | B_RAW));
728 		bp->b_oflags = obp->b_oflags;
729 		bp->b_cflags = obp->b_cflags;
730 		bp->b_iodone = vndiodone;
731 		bp->b_private = obp;
732 		bp->b_vp = vnd->sc_vp;
733 		bp->b_objlock = bp->b_vp->v_interlock;
734 		bp->b_data = obp->b_data;
735 		bp->b_bcount = obp->b_bcount;
736 		BIO_COPYPRIO(bp, obp);
737 
738 		/* Make sure the request succeeds while suspending this fs. */
739 		fstrans_start_lazy(vnd->sc_vp->v_mount);
740 
741 		/* Handle the request using the appropriate operations. */
742 		if ((vnd->sc_flags & VNF_USE_VN_RDWR) == 0)
743 			handle_with_strategy(vnd, obp, bp);
744 		else
745 			handle_with_rdwr(vnd, obp, bp);
746 
747 		fstrans_done(vnd->sc_vp->v_mount);
748 
749 		s = splbio();
750 		continue;
751 
752 done:
753 		biodone(obp);
754 		s = splbio();
755 	}
756 
757 	vnd->sc_flags &= (~VNF_KTHREAD | VNF_VUNCONF);
758 	wakeup(&vnd->sc_kthread);
759 	splx(s);
760 	kthread_exit(0);
761 }
762 
763 /*
764  * Checks if the given vnode supports the requested operation.
765  * The operation is specified the offset returned by VOFFSET.
766  *
767  * XXX The test below used to determine this is quite fragile
768  * because it relies on the file system to use genfs to specify
769  * unimplemented operations.  There might be another way to do
770  * it more cleanly.
771  */
772 static bool
773 vnode_has_op(const struct vnode *vp, int opoffset)
774 {
775 	int (*defaultp)(void *);
776 	int (*opp)(void *);
777 
778 	defaultp = vp->v_op[VOFFSET(vop_default)];
779 	opp = vp->v_op[opoffset];
780 
781 	return opp != defaultp && opp != genfs_eopnotsupp &&
782 	    opp != genfs_badop && opp != genfs_nullop;
783 }
784 
785 /*
786  * Handles the read/write request given in 'bp' using the vnode's VOP_READ
787  * and VOP_WRITE operations.
788  *
789  * 'obp' is a pointer to the original request fed to the vnd device.
790  */
791 static void
792 handle_with_rdwr(struct vnd_softc *vnd, const struct buf *obp, struct buf *bp)
793 {
794 	bool doread;
795 	off_t offset;
796 	size_t len, resid;
797 	struct vnode *vp;
798 	int npages;
799 
800 	doread = bp->b_flags & B_READ;
801 	offset = obp->b_rawblkno * vnd->sc_dkdev.dk_label->d_secsize;
802 	len = bp->b_bcount;
803 	vp = vnd->sc_vp;
804 
805 #if defined(DEBUG)
806 	if (vnddebug & VDB_IO)
807 		printf("vnd (rdwr): vp %p, %s, rawblkno 0x%" PRIx64
808 		    ", secsize %d, offset %" PRIu64
809 		    ", bcount %d\n",
810 		    vp, doread ? "read" : "write", obp->b_rawblkno,
811 		    vnd->sc_dkdev.dk_label->d_secsize, offset,
812 		    bp->b_bcount);
813 #endif
814 
815 	/* Issue the read or write operation. */
816 	bp->b_error =
817 	    vn_rdwr(doread ? UIO_READ : UIO_WRITE,
818 	    vp, bp->b_data, len, offset, UIO_SYSSPACE,
819 	    IO_ADV_ENCODE(POSIX_FADV_NOREUSE) | IO_DIRECT,
820 		vnd->sc_cred, &resid, NULL);
821 	bp->b_resid = resid;
822 
823 	/*
824 	 * Avoid caching too many pages, the vnd user
825 	 * is usually a filesystem and caches itself.
826 	 * We need some amount of caching to not hinder
827 	 * read-ahead and write-behind operations.
828 	 */
829 	npages = atomic_load_relaxed(&vp->v_uobj.uo_npages);
830 	if (npages > VND_MAXPAGES(vnd)) {
831 		rw_enter(vp->v_uobj.vmobjlock, RW_WRITER);
832 		(void) VOP_PUTPAGES(vp, 0, 0,
833 		    PGO_ALLPAGES | PGO_CLEANIT | PGO_FREE);
834 	}
835 
836 	/* We need to increase the number of outputs on the vnode if
837 	 * there was any write to it. */
838 	if (!doread) {
839 		mutex_enter(vp->v_interlock);
840 		vp->v_numoutput++;
841 		mutex_exit(vp->v_interlock);
842 	}
843 
844 	biodone(bp);
845 }
846 
847 /*
848  * Handes the read/write request given in 'bp' using the vnode's VOP_BMAP
849  * and VOP_STRATEGY operations.
850  *
851  * 'obp' is a pointer to the original request fed to the vnd device.
852  */
853 static void
854 handle_with_strategy(struct vnd_softc *vnd, const struct buf *obp,
855     struct buf *bp)
856 {
857 	int bsize, error, flags, skipped;
858 	size_t resid, sz;
859 	off_t bn, offset;
860 	struct vnode *vp;
861 	struct buf *nbp = NULL;
862 
863 	flags = obp->b_flags;
864 
865 
866 	/* convert to a byte offset within the file. */
867 	bn = obp->b_rawblkno * vnd->sc_dkdev.dk_label->d_secsize;
868 
869 	bsize = vnd->sc_vp->v_mount->mnt_stat.f_iosize;
870 	skipped = 0;
871 
872 	/*
873 	 * Break the request into bsize pieces and feed them
874 	 * sequentially using VOP_BMAP/VOP_STRATEGY.
875 	 * We do it this way to keep from flooding NFS servers if we
876 	 * are connected to an NFS file.  This places the burden on
877 	 * the client rather than the server.
878 	 */
879 	error = 0;
880 	bp->b_resid = bp->b_bcount;
881 	for (offset = 0, resid = bp->b_resid; /* true */;
882 	    resid -= sz, offset += sz) {
883 		daddr_t nbn;
884 		int off, nra;
885 
886 		nra = 0;
887 		vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
888 		error = VOP_BMAP(vnd->sc_vp, bn / bsize, &vp, &nbn, &nra);
889 		VOP_UNLOCK(vnd->sc_vp);
890 
891 		if (error == 0 && (long)nbn == -1)
892 			error = EIO;
893 
894 		/*
895 		 * If there was an error or a hole in the file...punt.
896 		 * Note that we may have to wait for any operations
897 		 * that we have already fired off before releasing
898 		 * the buffer.
899 		 *
900 		 * XXX we could deal with holes here but it would be
901 		 * a hassle (in the write case).
902 		 */
903 		if (error) {
904 			skipped += resid;
905 			break;
906 		}
907 
908 #ifdef DEBUG
909 		if (!dovndcluster)
910 			nra = 0;
911 #endif
912 
913 		off = bn % bsize;
914 		sz = MIN(((off_t)1 + nra) * bsize - off, resid);
915 #ifdef	DEBUG
916 		if (vnddebug & VDB_IO)
917 			printf("vndstrategy: vp %p/%p bn 0x%qx/0x%" PRIx64
918 			    " sz 0x%zx\n", vnd->sc_vp, vp, (long long)bn,
919 			    nbn, sz);
920 #endif
921 
922 		nbp = getiobuf(vp, true);
923 		nestiobuf_setup(bp, nbp, offset, sz);
924 		nbp->b_blkno = nbn + btodb(off);
925 
926 #if 0 /* XXX #ifdef DEBUG */
927 		if (vnddebug & VDB_IO)
928 			printf("vndstart(%ld): bp %p vp %p blkno "
929 			    "0x%" PRIx64 " flags %x addr %p cnt 0x%x\n",
930 			    (long) (vnd-vnd_softc), &nbp->vb_buf,
931 			    nbp->vb_buf.b_vp, nbp->vb_buf.b_blkno,
932 			    nbp->vb_buf.b_flags, nbp->vb_buf.b_data,
933 			    nbp->vb_buf.b_bcount);
934 #endif
935 		if (resid == sz) {
936 			break;
937 		}
938 		VOP_STRATEGY(vp, nbp);
939 		bn += sz;
940 	}
941 	if (!(flags & B_READ)) {
942 		struct vnode *w_vp;
943 		/*
944 		 * this is the last nested buf, account for
945 		 * the parent buf write too.
946 		 * This has to be done last, so that
947 		 * fsync won't wait for this write which
948 		 * has no chance to complete before all nested bufs
949 		 * have been queued. But it has to be done
950 		 * before the last VOP_STRATEGY()
951 		 * or the call to nestiobuf_done().
952 		 */
953 		w_vp = bp->b_vp;
954 		mutex_enter(w_vp->v_interlock);
955 		w_vp->v_numoutput++;
956 		mutex_exit(w_vp->v_interlock);
957 	}
958 	KASSERT(skipped != 0 || nbp != NULL);
959 	if (skipped)
960 		nestiobuf_done(bp, skipped, error);
961 	else
962 		VOP_STRATEGY(vp, nbp);
963 }
964 
965 static void
966 vndiodone(struct buf *bp)
967 {
968 	struct vndxfer *vnx = VND_BUFTOXFER(bp);
969 	struct vnd_softc *vnd = vnx->vx_vnd;
970 	struct buf *obp = bp->b_private;
971 	int s = splbio();
972 
973 	KERNEL_LOCK(1, NULL);		/* XXXSMP */
974 	KASSERT(&vnx->vx_buf == bp);
975 	KASSERT(vnd->sc_active > 0);
976 #ifdef DEBUG
977 	if (vnddebug & VDB_IO) {
978 		printf("vndiodone1: bp %p iodone: error %d\n",
979 		    bp, bp->b_error);
980 	}
981 #endif
982 	disk_unbusy(&vnd->sc_dkdev, bp->b_bcount - bp->b_resid,
983 	    (bp->b_flags & B_READ));
984 	vnd->sc_active--;
985 	if (vnd->sc_active == 0) {
986 		wakeup(&vnd->sc_tab);
987 	}
988 	KERNEL_UNLOCK_ONE(NULL);	/* XXXSMP */
989 	splx(s);
990 	obp->b_error = bp->b_error;
991 	obp->b_resid = bp->b_resid;
992 	buf_destroy(bp);
993 	VND_PUTXFER(vnd, vnx);
994 	biodone(obp);
995 }
996 
997 /* ARGSUSED */
998 static int
999 vndread(dev_t dev, struct uio *uio, int flags)
1000 {
1001 	int unit = vndunit(dev);
1002 	struct vnd_softc *sc;
1003 
1004 #ifdef DEBUG
1005 	if (vnddebug & VDB_FOLLOW)
1006 		printf("vndread(0x%"PRIx64", %p)\n", dev, uio);
1007 #endif
1008 
1009 	sc = device_lookup_private(&vnd_cd, unit);
1010 	if (sc == NULL)
1011 		return ENXIO;
1012 
1013 	if ((sc->sc_flags & VNF_INITED) == 0)
1014 		return ENXIO;
1015 
1016 	return physio(vndstrategy, NULL, dev, B_READ, minphys, uio);
1017 }
1018 
1019 /* ARGSUSED */
1020 static int
1021 vndwrite(dev_t dev, struct uio *uio, int flags)
1022 {
1023 	int unit = vndunit(dev);
1024 	struct vnd_softc *sc;
1025 
1026 #ifdef DEBUG
1027 	if (vnddebug & VDB_FOLLOW)
1028 		printf("vndwrite(0x%"PRIx64", %p)\n", dev, uio);
1029 #endif
1030 
1031 	sc = device_lookup_private(&vnd_cd, unit);
1032 	if (sc == NULL)
1033 		return ENXIO;
1034 
1035 	if ((sc->sc_flags & VNF_INITED) == 0)
1036 		return ENXIO;
1037 
1038 	return physio(vndstrategy, NULL, dev, B_WRITE, minphys, uio);
1039 }
1040 
1041 static int
1042 vnd_cget(struct lwp *l, int unit, int *un, struct vattr *va)
1043 {
1044 	int error;
1045 	struct vnd_softc *vnd;
1046 
1047 	if (*un == -1)
1048 		*un = unit;
1049 	if (*un < 0)
1050 		return EINVAL;
1051 
1052 	vnd = device_lookup_private(&vnd_cd, *un);
1053 	if (vnd == NULL)
1054 		return -1;
1055 
1056 	if ((vnd->sc_flags & VNF_INITED) == 0)
1057 		return -1;
1058 
1059 	vn_lock(vnd->sc_vp, LK_SHARED | LK_RETRY);
1060 	error = VOP_GETATTR(vnd->sc_vp, va, l->l_cred);
1061 	VOP_UNLOCK(vnd->sc_vp);
1062 	return error;
1063 }
1064 
1065 static int
1066 vnddoclear(struct vnd_softc *vnd, int pmask, int minor, bool force)
1067 {
1068 	int error;
1069 
1070 	if ((error = vndlock(vnd)) != 0)
1071 		return error;
1072 
1073 	/*
1074 	 * Don't unconfigure if any other partitions are open
1075 	 * or if both the character and block flavors of this
1076 	 * partition are open.
1077 	 */
1078 	if (DK_BUSY(vnd, pmask) && !force) {
1079 		vndunlock(vnd);
1080 		return EBUSY;
1081 	}
1082 
1083 	/* Delete all of our wedges */
1084 	dkwedge_delall(&vnd->sc_dkdev);
1085 
1086 	/*
1087 	 * XXX vndclear() might call vndclose() implicitly;
1088 	 * release lock to avoid recursion
1089 	 *
1090 	 * Set VNF_CLEARING to prevent vndopen() from
1091 	 * sneaking in after we vndunlock().
1092 	 */
1093 	vnd->sc_flags |= VNF_CLEARING;
1094 	vndunlock(vnd);
1095 	vndclear(vnd, minor);
1096 #ifdef DEBUG
1097 	if (vnddebug & VDB_INIT)
1098 		printf("%s: CLRed\n", __func__);
1099 #endif
1100 
1101 	/* Destroy the xfer and buffer pools. */
1102 	pool_destroy(&vnd->sc_vxpool);
1103 
1104 	/* Detach the disk. */
1105 	disk_detach(&vnd->sc_dkdev);
1106 
1107 	return 0;
1108 }
1109 
1110 static int
1111 vndioctl_get(struct lwp *l, void *data, int unit, struct vattr *va)
1112 {
1113 	int error;
1114 
1115 	KASSERT(l);
1116 
1117 	/* the first member is always int vnd_unit in all the versions */
1118 	if (*(int *)data >= vnd_cd.cd_ndevs)
1119 		return ENXIO;
1120 
1121 	switch (error = vnd_cget(l, unit, (int *)data, va)) {
1122 	case -1:
1123 		/* unused is not an error */
1124 		memset(va, 0, sizeof(*va));
1125 		/*FALLTHROUGH*/
1126 	case 0:
1127 		return 0;
1128 	default:
1129 		return error;
1130 	}
1131 }
1132 
1133 /* ARGSUSED */
1134 static int
1135 vndioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l)
1136 {
1137 	bool force;
1138 	int unit = vndunit(dev);
1139 	struct vnd_softc *vnd;
1140 	struct vnd_ioctl *vio;
1141 	struct vattr vattr;
1142 	struct pathbuf *pb;
1143 	struct nameidata nd;
1144 	int error, part, pmask;
1145 	uint64_t geomsize;
1146 	int fflags;
1147 #ifdef __HAVE_OLD_DISKLABEL
1148 	struct disklabel newlabel;
1149 #endif
1150 
1151 #ifdef DEBUG
1152 	if (vnddebug & VDB_FOLLOW)
1153 		printf("vndioctl(0x%"PRIx64", 0x%lx, %p, 0x%x, %p): unit %d\n",
1154 		    dev, cmd, data, flag, l->l_proc, unit);
1155 #endif
1156 	/* Do the get's first; they don't need initialization or verification */
1157 	switch (cmd) {
1158 	case VNDIOCGET:
1159 		if ((error = vndioctl_get(l, data, unit, &vattr)) != 0)
1160 			return error;
1161 
1162 		struct vnd_user *vnu = data;
1163 		vnu->vnu_dev = vattr.va_fsid;
1164 		vnu->vnu_ino = vattr.va_fileid;
1165 		return 0;
1166 
1167 	default:
1168 		/* First check for COMPAT_50 hook */
1169 		MODULE_HOOK_CALL(compat_vndioctl_50_hook,
1170 		    (cmd, l, data, unit, &vattr, vndioctl_get),
1171 		    enosys(), error);
1172 
1173 		/*
1174 		 * If not present, then COMPAT_30 hook also not
1175 		 * present, so just continue with checks for the
1176 		 * "write" commands
1177 		 */
1178 		if (error == ENOSYS) {
1179 			error = 0;
1180 			break;
1181 		}
1182 
1183 		/* If not already handled, try the COMPAT_30 hook */
1184 		if (error == EPASSTHROUGH)
1185 			MODULE_HOOK_CALL(compat_vndioctl_30_hook,
1186 			    (cmd, l, data, unit, &vattr, vndioctl_get),
1187 			    enosys(), error);
1188 
1189 		/* If no COMPAT_30 module, or not handled, check writes */
1190 		if (error == ENOSYS || error == EPASSTHROUGH) {
1191 			error = 0;
1192 			break;
1193 		}
1194 		return error;
1195 	}
1196 
1197 	vnd = device_lookup_private(&vnd_cd, unit);
1198 	if (vnd == NULL)
1199 		return ENXIO;
1200 	vio = (struct vnd_ioctl *)data;
1201 
1202 	/* Must be open for writes for these commands... */
1203 	switch (cmd) {
1204 	case VNDIOCSET50:
1205 	case VNDIOCCLR50:
1206 		if (!compat_vndioctl_50_hook.hooked)
1207 			return EINVAL;
1208 		/* FALLTHROUGH */
1209 	case VNDIOCSET:
1210 	case VNDIOCCLR:
1211 	case DIOCSDINFO:
1212 	case DIOCWDINFO:
1213 #ifdef __HAVE_OLD_DISKLABEL
1214 	case ODIOCSDINFO:
1215 	case ODIOCWDINFO:
1216 #endif
1217 	case DIOCKLABEL:
1218 	case DIOCWLABEL:
1219 	case DIOCCACHESYNC:
1220 		if ((flag & FWRITE) == 0)
1221 			return EBADF;
1222 	}
1223 
1224 	/* Must be initialized for these... */
1225 	switch (cmd) {
1226 	case VNDIOCCLR:
1227 	case VNDIOCCLR50:
1228 	case DIOCGDINFO:
1229 	case DIOCSDINFO:
1230 	case DIOCWDINFO:
1231 	case DIOCGPARTINFO:
1232 	case DIOCKLABEL:
1233 	case DIOCWLABEL:
1234 	case DIOCGDEFLABEL:
1235 	case DIOCGCACHE:
1236 	case DIOCGSTRATEGY:
1237 	case DIOCCACHESYNC:
1238 #ifdef __HAVE_OLD_DISKLABEL
1239 	case ODIOCGDINFO:
1240 	case ODIOCSDINFO:
1241 	case ODIOCWDINFO:
1242 	case ODIOCGDEFLABEL:
1243 #endif
1244 		if ((vnd->sc_flags & VNF_INITED) == 0)
1245 			return ENXIO;
1246 	}
1247 
1248 	error = disk_ioctl(&vnd->sc_dkdev, dev, cmd, data, flag, l);
1249 	if (error != EPASSTHROUGH)
1250 		return error;
1251 
1252 
1253 	switch (cmd) {
1254 	case VNDIOCSET50:
1255 	case VNDIOCSET:
1256 		if (vnd->sc_flags & VNF_INITED)
1257 			return EBUSY;
1258 
1259 		if ((error = vndlock(vnd)) != 0)
1260 			return error;
1261 
1262 		fflags = FREAD;
1263 		if ((vio->vnd_flags & VNDIOF_READONLY) == 0)
1264 			fflags |= FWRITE;
1265 		if ((vio->vnd_flags & VNDIOF_FILEIO) != 0)
1266 			vnd->sc_flags |= VNF_USE_VN_RDWR;
1267 		error = pathbuf_copyin(vio->vnd_file, &pb);
1268 		if (error) {
1269 			goto unlock_and_exit;
1270 		}
1271 		NDINIT(&nd, LOOKUP, FOLLOW, pb);
1272 		if ((error = vn_open(&nd, fflags, 0)) != 0) {
1273 			pathbuf_destroy(pb);
1274 			goto unlock_and_exit;
1275 		}
1276 		KASSERT(l);
1277 		error = VOP_GETATTR(nd.ni_vp, &vattr, l->l_cred);
1278 		if (!error && nd.ni_vp->v_type != VREG)
1279 			error = EOPNOTSUPP;
1280 		if (!error && vattr.va_bytes < vattr.va_size)
1281 			/* File is definitely sparse, use vn_rdwr() */
1282 			vnd->sc_flags |= VNF_USE_VN_RDWR;
1283 		if (error) {
1284 			VOP_UNLOCK(nd.ni_vp);
1285 			goto close_and_exit;
1286 		}
1287 
1288 		/* If using a compressed file, initialize its info */
1289 		/* (or abort with an error if kernel has no compression) */
1290 		if (vio->vnd_flags & VNDIOF_COMP) {
1291 #ifdef VND_COMPRESSION
1292 			struct vnd_comp_header *ch;
1293 			int i;
1294 			uint32_t comp_size;
1295 			uint32_t comp_maxsize;
1296 
1297 			/* allocate space for compresed file header */
1298 			ch = malloc(sizeof(struct vnd_comp_header),
1299 			    M_TEMP, M_WAITOK);
1300 
1301 			/* read compressed file header */
1302 			error = vn_rdwr(UIO_READ, nd.ni_vp, (void *)ch,
1303 			    sizeof(struct vnd_comp_header), 0, UIO_SYSSPACE,
1304 			    IO_UNIT|IO_NODELOCKED, l->l_cred, NULL, NULL);
1305 			if (error) {
1306 				free(ch, M_TEMP);
1307 				VOP_UNLOCK(nd.ni_vp);
1308 				goto close_and_exit;
1309 			}
1310 
1311 			if (be32toh(ch->block_size) == 0 ||
1312 			    be32toh(ch->num_blocks) > UINT32_MAX - 1) {
1313 				free(ch, M_TEMP);
1314 				VOP_UNLOCK(nd.ni_vp);
1315 				goto close_and_exit;
1316 			}
1317 
1318 			/* save some header info */
1319 			vnd->sc_comp_blksz = be32toh(ch->block_size);
1320 			/* note last offset is the file byte size */
1321 			vnd->sc_comp_numoffs = be32toh(ch->num_blocks) + 1;
1322 			free(ch, M_TEMP);
1323 			if (!DK_DEV_BSIZE_OK(vnd->sc_comp_blksz)) {
1324 				VOP_UNLOCK(nd.ni_vp);
1325 				error = EINVAL;
1326 				goto close_and_exit;
1327 			}
1328 			KASSERT(0 < vnd->sc_comp_blksz);
1329 			KASSERT(0 < vnd->sc_comp_numoffs);
1330 			/*
1331 			 * @#^@!$& gcc -Wtype-limits refuses to let me
1332 			 * write SIZE_MAX/sizeof(uint64_t) < numoffs,
1333 			 * because the range of the type on amd64 makes
1334 			 * the comparisons always false.
1335 			 */
1336 #if SIZE_MAX <= UINT32_MAX*(64/CHAR_BIT)
1337 			if (SIZE_MAX/sizeof(uint64_t) < vnd->sc_comp_numoffs) {
1338 				VOP_UNLOCK(nd.ni_vp);
1339 				error = EINVAL;
1340 				goto close_and_exit;
1341 			}
1342 #endif
1343 			if ((vattr.va_size < sizeof(struct vnd_comp_header)) ||
1344 			    (vattr.va_size - sizeof(struct vnd_comp_header) <
1345 				sizeof(uint64_t)*vnd->sc_comp_numoffs) ||
1346 			    (UQUAD_MAX/vnd->sc_comp_blksz <
1347 				vnd->sc_comp_numoffs - 1)) {
1348 				VOP_UNLOCK(nd.ni_vp);
1349 				error = EINVAL;
1350 				goto close_and_exit;
1351 			}
1352 
1353 			/* set decompressed file size */
1354 			KASSERT(vnd->sc_comp_numoffs - 1 <=
1355 			    UQUAD_MAX/vnd->sc_comp_blksz);
1356 			vattr.va_size =
1357 			    ((u_quad_t)vnd->sc_comp_numoffs - 1) *
1358 			     (u_quad_t)vnd->sc_comp_blksz;
1359 
1360 			/* allocate space for all the compressed offsets */
1361 			__CTASSERT(UINT32_MAX <= UQUAD_MAX/sizeof(uint64_t));
1362 			vnd->sc_comp_offsets =
1363 			    malloc(sizeof(uint64_t) * vnd->sc_comp_numoffs,
1364 				M_DEVBUF, M_WAITOK);
1365 
1366 			/* read in the offsets */
1367 			error = vn_rdwr(UIO_READ, nd.ni_vp,
1368 			    (void *)vnd->sc_comp_offsets,
1369 			    sizeof(uint64_t) * vnd->sc_comp_numoffs,
1370 			    sizeof(struct vnd_comp_header), UIO_SYSSPACE,
1371 			  IO_UNIT|IO_NODELOCKED, l->l_cred, NULL, NULL);
1372 			if (error) {
1373 				VOP_UNLOCK(nd.ni_vp);
1374 				goto close_and_exit;
1375 			}
1376 			/*
1377 			 * find largest block size (used for allocation limit).
1378 			 * Also convert offset to native byte order.
1379 			 */
1380 			comp_maxsize = 0;
1381 			for (i = 0; i < vnd->sc_comp_numoffs - 1; i++) {
1382 				vnd->sc_comp_offsets[i] =
1383 				    be64toh(vnd->sc_comp_offsets[i]);
1384 				comp_size =
1385 				    be64toh(vnd->sc_comp_offsets[i + 1])
1386 				    - vnd->sc_comp_offsets[i];
1387 				if (comp_size > comp_maxsize)
1388 					comp_maxsize = comp_size;
1389 			}
1390 			vnd->sc_comp_offsets[vnd->sc_comp_numoffs - 1] =
1391 			    be64toh(vnd->sc_comp_offsets[vnd->sc_comp_numoffs
1392 				    - 1]);
1393 
1394 			/* create compressed data buffer */
1395 			vnd->sc_comp_buff = malloc(comp_maxsize,
1396 			    M_DEVBUF, M_WAITOK);
1397 
1398 			/* create decompressed buffer */
1399 			vnd->sc_comp_decombuf = malloc(vnd->sc_comp_blksz,
1400 			    M_DEVBUF, M_WAITOK);
1401 			vnd->sc_comp_buffblk = -1;
1402 
1403 			/* Initialize decompress stream */
1404 			memset(&vnd->sc_comp_stream, 0, sizeof(z_stream));
1405 			vnd->sc_comp_stream.zalloc = vnd_alloc;
1406 			vnd->sc_comp_stream.zfree = vnd_free;
1407 			error = inflateInit2(&vnd->sc_comp_stream, MAX_WBITS);
1408 			if (error) {
1409 				if (vnd->sc_comp_stream.msg)
1410 					printf("vnd%d: compressed file, %s\n",
1411 					    unit, vnd->sc_comp_stream.msg);
1412 				VOP_UNLOCK(nd.ni_vp);
1413 				error = EINVAL;
1414 				goto close_and_exit;
1415 			}
1416 
1417 			vnd->sc_flags |= VNF_COMP | VNF_READONLY;
1418 #else /* !VND_COMPRESSION */
1419 			VOP_UNLOCK(nd.ni_vp);
1420 			error = EOPNOTSUPP;
1421 			goto close_and_exit;
1422 #endif /* VND_COMPRESSION */
1423 		}
1424 
1425 		VOP_UNLOCK(nd.ni_vp);
1426 		vnd->sc_vp = nd.ni_vp;
1427 		vnd->sc_size = btodb(vattr.va_size);	/* note truncation */
1428 
1429 		/* get smallest I/O size for underlying device, fall back to
1430 		 * fundamental I/O size of underlying filesystem
1431 		 */
1432 		error = bdev_ioctl(vattr.va_fsid, DIOCGSECTORSIZE, &vnd->sc_iosize, FKIOCTL, l);
1433 		if (error)
1434 			vnd->sc_iosize = vnd->sc_vp->v_mount->mnt_stat.f_frsize;
1435 
1436 		/*
1437 		 * Use pseudo-geometry specified.  If none was provided,
1438 		 * use "standard" Adaptec fictitious geometry.
1439 		 */
1440 		if (vio->vnd_flags & VNDIOF_HASGEOM) {
1441 
1442 			memcpy(&vnd->sc_geom, &vio->vnd_geom,
1443 			    sizeof(vio->vnd_geom));
1444 
1445 			/*
1446 			 * Sanity-check the sector size.
1447 			 */
1448 			if (!DK_DEV_BSIZE_OK(vnd->sc_geom.vng_secsize) ||
1449 			    vnd->sc_geom.vng_ntracks == 0 ||
1450 			    vnd->sc_geom.vng_nsectors == 0) {
1451 				error = EINVAL;
1452 				goto close_and_exit;
1453 			}
1454 
1455 			/*
1456 			 * Compute missing cylinder count from size
1457 			 */
1458 			if (vnd->sc_geom.vng_ncylinders == 0)
1459 				vnd->sc_geom.vng_ncylinders = vnd->sc_size / (
1460 					(vnd->sc_geom.vng_secsize / DEV_BSIZE) *
1461 					vnd->sc_geom.vng_ntracks *
1462 					vnd->sc_geom.vng_nsectors);
1463 
1464 			/*
1465 			 * Compute the size (in DEV_BSIZE blocks) specified
1466 			 * by the geometry.
1467 			 */
1468 			geomsize = (int64_t)vnd->sc_geom.vng_nsectors *
1469 			    vnd->sc_geom.vng_ntracks *
1470 			    vnd->sc_geom.vng_ncylinders *
1471 			    (vnd->sc_geom.vng_secsize / DEV_BSIZE);
1472 
1473 			/*
1474 			 * Sanity-check the size against the specified
1475 			 * geometry.
1476 			 */
1477 			if (vnd->sc_size < geomsize) {
1478 				error = EINVAL;
1479 				goto close_and_exit;
1480 			}
1481 		} else if (vnd->sc_size >= (32 * 64)) {
1482 			/*
1483 			 * Size must be at least 2048 DEV_BSIZE blocks
1484 			 * (1M) in order to use this geometry.
1485 			 */
1486 			vnd->sc_geom.vng_secsize = DEV_BSIZE;
1487 			vnd->sc_geom.vng_nsectors = 32;
1488 			vnd->sc_geom.vng_ntracks = 64;
1489 			vnd->sc_geom.vng_ncylinders = vnd->sc_size / (64 * 32);
1490 		} else {
1491 			vnd->sc_geom.vng_secsize = DEV_BSIZE;
1492 			vnd->sc_geom.vng_nsectors = 1;
1493 			vnd->sc_geom.vng_ntracks = 1;
1494 			vnd->sc_geom.vng_ncylinders = vnd->sc_size;
1495 		}
1496 
1497 		vnd_set_geometry(vnd);
1498 
1499 		if (vio->vnd_flags & VNDIOF_READONLY) {
1500 			vnd->sc_flags |= VNF_READONLY;
1501 		}
1502 
1503 		if ((error = vndsetcred(vnd, l->l_cred)) != 0)
1504 			goto close_and_exit;
1505 
1506 		vndthrottle(vnd, vnd->sc_vp);
1507 		vio->vnd_osize = dbtob(vnd->sc_size);
1508 		if (cmd != VNDIOCSET50)
1509 			vio->vnd_size = dbtob(vnd->sc_size);
1510 		vnd->sc_flags |= VNF_INITED;
1511 
1512 		/* create the kernel thread, wait for it to be up */
1513 		error = kthread_create(PRI_NONE, 0, NULL, vndthread, vnd,
1514 		    &vnd->sc_kthread, "%s", device_xname(vnd->sc_dev));
1515 		if (error)
1516 			goto close_and_exit;
1517 		while ((vnd->sc_flags & VNF_KTHREAD) == 0) {
1518 			tsleep(&vnd->sc_kthread, PRIBIO, "vndthr", 0);
1519 		}
1520 #ifdef DEBUG
1521 		if (vnddebug & VDB_INIT)
1522 			printf("vndioctl: SET vp %p size 0x%lx %d/%d/%d/%d\n",
1523 			    vnd->sc_vp, (unsigned long) vnd->sc_size,
1524 			    vnd->sc_geom.vng_secsize,
1525 			    vnd->sc_geom.vng_nsectors,
1526 			    vnd->sc_geom.vng_ntracks,
1527 			    vnd->sc_geom.vng_ncylinders);
1528 #endif
1529 
1530 		/* Attach the disk. */
1531 		disk_attach(&vnd->sc_dkdev);
1532 
1533 		/* Initialize the xfer and buffer pools. */
1534 		pool_init(&vnd->sc_vxpool, sizeof(struct vndxfer), 0,
1535 		    0, 0, "vndxpl", NULL, IPL_BIO);
1536 
1537 		vndunlock(vnd);
1538 
1539 		pathbuf_destroy(pb);
1540 
1541 		/* Discover wedges on this disk */
1542 		dkwedge_discover(&vnd->sc_dkdev);
1543 
1544 		break;
1545 
1546 close_and_exit:
1547 		(void) vn_close(nd.ni_vp, fflags, l->l_cred);
1548 		pathbuf_destroy(pb);
1549 unlock_and_exit:
1550 #ifdef VND_COMPRESSION
1551 		/* free any allocated memory (for compressed file) */
1552 		if (vnd->sc_comp_offsets) {
1553 			free(vnd->sc_comp_offsets, M_DEVBUF);
1554 			vnd->sc_comp_offsets = NULL;
1555 		}
1556 		if (vnd->sc_comp_buff) {
1557 			free(vnd->sc_comp_buff, M_DEVBUF);
1558 			vnd->sc_comp_buff = NULL;
1559 		}
1560 		if (vnd->sc_comp_decombuf) {
1561 			free(vnd->sc_comp_decombuf, M_DEVBUF);
1562 			vnd->sc_comp_decombuf = NULL;
1563 		}
1564 #endif /* VND_COMPRESSION */
1565 		vndunlock(vnd);
1566 		return error;
1567 
1568 	case VNDIOCCLR50:
1569 	case VNDIOCCLR:
1570 		part = DISKPART(dev);
1571 		pmask = (1 << part);
1572 		force = (vio->vnd_flags & VNDIOF_FORCE) != 0;
1573 
1574 		if ((error = vnddoclear(vnd, pmask, minor(dev), force)) != 0)
1575 			return error;
1576 
1577 		break;
1578 
1579 
1580 	case DIOCWDINFO:
1581 	case DIOCSDINFO:
1582 #ifdef __HAVE_OLD_DISKLABEL
1583 	case ODIOCWDINFO:
1584 	case ODIOCSDINFO:
1585 #endif
1586 	{
1587 		struct disklabel *lp;
1588 
1589 		if ((error = vndlock(vnd)) != 0)
1590 			return error;
1591 
1592 		vnd->sc_flags |= VNF_LABELLING;
1593 
1594 #ifdef __HAVE_OLD_DISKLABEL
1595 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
1596 			memset(&newlabel, 0, sizeof newlabel);
1597 			memcpy(&newlabel, data, sizeof (struct olddisklabel));
1598 			lp = &newlabel;
1599 		} else
1600 #endif
1601 		lp = (struct disklabel *)data;
1602 
1603 		error = setdisklabel(vnd->sc_dkdev.dk_label,
1604 		    lp, 0, vnd->sc_dkdev.dk_cpulabel);
1605 		if (error == 0) {
1606 			if (cmd == DIOCWDINFO
1607 #ifdef __HAVE_OLD_DISKLABEL
1608 			    || cmd == ODIOCWDINFO
1609 #endif
1610 			   )
1611 				error = writedisklabel(VNDLABELDEV(dev),
1612 				    vndstrategy, vnd->sc_dkdev.dk_label,
1613 				    vnd->sc_dkdev.dk_cpulabel);
1614 		}
1615 
1616 		vnd->sc_flags &= ~VNF_LABELLING;
1617 
1618 		vndunlock(vnd);
1619 
1620 		if (error)
1621 			return error;
1622 		break;
1623 	}
1624 
1625 	case DIOCKLABEL:
1626 		if (*(int *)data != 0)
1627 			vnd->sc_flags |= VNF_KLABEL;
1628 		else
1629 			vnd->sc_flags &= ~VNF_KLABEL;
1630 		break;
1631 
1632 	case DIOCWLABEL:
1633 		if (*(int *)data != 0)
1634 			vnd->sc_flags |= VNF_WLABEL;
1635 		else
1636 			vnd->sc_flags &= ~VNF_WLABEL;
1637 		break;
1638 
1639 	case DIOCGDEFLABEL:
1640 		vndgetdefaultlabel(vnd, (struct disklabel *)data);
1641 		break;
1642 
1643 #ifdef __HAVE_OLD_DISKLABEL
1644 	case ODIOCGDEFLABEL:
1645 		vndgetdefaultlabel(vnd, &newlabel);
1646 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1647 			return ENOTTY;
1648 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1649 		break;
1650 #endif
1651 
1652 	case DIOCGSTRATEGY:
1653 	    {
1654 		struct disk_strategy *dks = (void *)data;
1655 
1656 		/* No lock needed, never changed */
1657 		strlcpy(dks->dks_name,
1658 		    bufq_getstrategyname(vnd->sc_tab),
1659 		    sizeof(dks->dks_name));
1660 		dks->dks_paramlen = 0;
1661 		break;
1662 	    }
1663 	case DIOCGCACHE:
1664 	    {
1665 		int *bits = (int *)data;
1666 		*bits |= DKCACHE_READ | DKCACHE_WRITE;
1667 		break;
1668 	    }
1669 	case DIOCCACHESYNC:
1670 		vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1671 		error = VOP_FSYNC(vnd->sc_vp, vnd->sc_cred,
1672 		    FSYNC_WAIT | FSYNC_DATAONLY | FSYNC_CACHE, 0, 0);
1673 		VOP_UNLOCK(vnd->sc_vp);
1674 		return error;
1675 
1676 	default:
1677 		return ENOTTY;
1678 	}
1679 
1680 	return 0;
1681 }
1682 
1683 /*
1684  * Duplicate the current processes' credentials.  Since we are called only
1685  * as the result of a SET ioctl and only root can do that, any future access
1686  * to this "disk" is essentially as root.  Note that credentials may change
1687  * if some other uid can write directly to the mapped file (NFS).
1688  */
1689 static int
1690 vndsetcred(struct vnd_softc *vnd, kauth_cred_t cred)
1691 {
1692 	struct uio auio;
1693 	struct iovec aiov;
1694 	char *tmpbuf;
1695 	int error;
1696 
1697 	vnd->sc_cred = kauth_cred_dup(cred);
1698 	tmpbuf = malloc(DEV_BSIZE, M_TEMP, M_WAITOK);
1699 
1700 	/* XXX: Horrible kludge to establish credentials for NFS */
1701 	aiov.iov_base = tmpbuf;
1702 	aiov.iov_len = uimin(DEV_BSIZE, dbtob(vnd->sc_size));
1703 	auio.uio_iov = &aiov;
1704 	auio.uio_iovcnt = 1;
1705 	auio.uio_offset = 0;
1706 	auio.uio_rw = UIO_READ;
1707 	auio.uio_resid = aiov.iov_len;
1708 	UIO_SETUP_SYSSPACE(&auio);
1709 	vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1710 	error = VOP_READ(vnd->sc_vp, &auio, 0, vnd->sc_cred);
1711 	if (error == 0) {
1712 		/*
1713 		 * Because vnd does all IO directly through the vnode
1714 		 * we need to flush (at least) the buffer from the above
1715 		 * VOP_READ from the buffer cache to prevent cache
1716 		 * incoherencies.  Also, be careful to write dirty
1717 		 * buffers back to stable storage.
1718 		 */
1719 		error = vinvalbuf(vnd->sc_vp, V_SAVE, vnd->sc_cred,
1720 			    curlwp, 0, 0);
1721 	}
1722 	VOP_UNLOCK(vnd->sc_vp);
1723 
1724 	free(tmpbuf, M_TEMP);
1725 	return error;
1726 }
1727 
1728 /*
1729  * Set maxactive based on FS type
1730  */
1731 static void
1732 vndthrottle(struct vnd_softc *vnd, struct vnode *vp)
1733 {
1734 
1735 	if (vp->v_tag == VT_NFS)
1736 		vnd->sc_maxactive = 2;
1737 	else
1738 		vnd->sc_maxactive = 8;
1739 
1740 	if (vnd->sc_maxactive < 1)
1741 		vnd->sc_maxactive = 1;
1742 }
1743 
1744 #if 0
1745 static void
1746 vndshutdown(void)
1747 {
1748 	struct vnd_softc *vnd;
1749 
1750 	for (vnd = &vnd_softc[0]; vnd < &vnd_softc[numvnd]; vnd++)
1751 		if (vnd->sc_flags & VNF_INITED)
1752 			vndclear(vnd);
1753 }
1754 #endif
1755 
1756 static void
1757 vndclear(struct vnd_softc *vnd, int myminor)
1758 {
1759 	struct vnode *vp = vnd->sc_vp;
1760 	int fflags = FREAD;
1761 	int bmaj, cmaj, i, mn;
1762 	int s;
1763 
1764 #ifdef DEBUG
1765 	if (vnddebug & VDB_FOLLOW)
1766 		printf("vndclear(%p): vp %p\n", vnd, vp);
1767 #endif
1768 	/* locate the major number */
1769 	bmaj = bdevsw_lookup_major(&vnd_bdevsw);
1770 	cmaj = cdevsw_lookup_major(&vnd_cdevsw);
1771 
1772 	/* Nuke the vnodes for any open instances */
1773 	for (i = 0; i < MAXPARTITIONS; i++) {
1774 		mn = DISKMINOR(device_unit(vnd->sc_dev), i);
1775 		vdevgone(bmaj, mn, mn, VBLK);
1776 		if (mn != myminor) /* XXX avoid to kill own vnode */
1777 			vdevgone(cmaj, mn, mn, VCHR);
1778 	}
1779 
1780 	if ((vnd->sc_flags & VNF_READONLY) == 0)
1781 		fflags |= FWRITE;
1782 
1783 	s = splbio();
1784 	bufq_drain(vnd->sc_tab);
1785 	splx(s);
1786 
1787 	vnd->sc_flags |= VNF_VUNCONF;
1788 	wakeup(&vnd->sc_tab);
1789 	while (vnd->sc_flags & VNF_KTHREAD)
1790 		tsleep(&vnd->sc_kthread, PRIBIO, "vnthr", 0);
1791 
1792 #ifdef VND_COMPRESSION
1793 	/* free the compressed file buffers */
1794 	if (vnd->sc_flags & VNF_COMP) {
1795 		if (vnd->sc_comp_offsets) {
1796 			free(vnd->sc_comp_offsets, M_DEVBUF);
1797 			vnd->sc_comp_offsets = NULL;
1798 		}
1799 		if (vnd->sc_comp_buff) {
1800 			free(vnd->sc_comp_buff, M_DEVBUF);
1801 			vnd->sc_comp_buff = NULL;
1802 		}
1803 		if (vnd->sc_comp_decombuf) {
1804 			free(vnd->sc_comp_decombuf, M_DEVBUF);
1805 			vnd->sc_comp_decombuf = NULL;
1806 		}
1807 	}
1808 #endif /* VND_COMPRESSION */
1809 	vnd->sc_flags &=
1810 	    ~(VNF_INITED | VNF_READONLY | VNF_KLABEL | VNF_VLABEL
1811 	      | VNF_VUNCONF | VNF_COMP | VNF_CLEARING);
1812 	if (vp == NULL)
1813 		panic("vndclear: null vp");
1814 	(void) vn_close(vp, fflags, vnd->sc_cred);
1815 	kauth_cred_free(vnd->sc_cred);
1816 	vnd->sc_vp = NULL;
1817 	vnd->sc_cred = NULL;
1818 	vnd->sc_size = 0;
1819 }
1820 
1821 static int
1822 vndsize(dev_t dev)
1823 {
1824 	struct vnd_softc *sc;
1825 	struct disklabel *lp;
1826 	int part, unit, omask;
1827 	int size;
1828 
1829 	unit = vndunit(dev);
1830 	sc = device_lookup_private(&vnd_cd, unit);
1831 	if (sc == NULL)
1832 		return -1;
1833 
1834 	if ((sc->sc_flags & VNF_INITED) == 0)
1835 		return -1;
1836 
1837 	part = DISKPART(dev);
1838 	omask = sc->sc_dkdev.dk_openmask & (1 << part);
1839 	lp = sc->sc_dkdev.dk_label;
1840 
1841 	if (omask == 0 && vndopen(dev, 0, S_IFBLK, curlwp))	/* XXX */
1842 		return -1;
1843 
1844 	if (lp->d_partitions[part].p_fstype != FS_SWAP)
1845 		size = -1;
1846 	else
1847 		size = lp->d_partitions[part].p_size *
1848 		    (lp->d_secsize / DEV_BSIZE);
1849 
1850 	if (omask == 0 && vndclose(dev, 0, S_IFBLK, curlwp))	/* XXX */
1851 		return -1;
1852 
1853 	return size;
1854 }
1855 
1856 static int
1857 vnddump(dev_t dev, daddr_t blkno, void *va,
1858     size_t size)
1859 {
1860 
1861 	/* Not implemented. */
1862 	return ENXIO;
1863 }
1864 
1865 static void
1866 vndgetdefaultlabel(struct vnd_softc *sc, struct disklabel *lp)
1867 {
1868 	struct vndgeom *vng = &sc->sc_geom;
1869 	struct partition *pp;
1870 	unsigned spb;
1871 
1872 	memset(lp, 0, sizeof(*lp));
1873 
1874 	spb = vng->vng_secsize / DEV_BSIZE;
1875 	if (sc->sc_size / spb > UINT32_MAX)
1876 		lp->d_secperunit = UINT32_MAX;
1877 	else
1878 		lp->d_secperunit = sc->sc_size / spb;
1879 	lp->d_secsize = vng->vng_secsize;
1880 	lp->d_nsectors = vng->vng_nsectors;
1881 	lp->d_ntracks = vng->vng_ntracks;
1882 	lp->d_ncylinders = vng->vng_ncylinders;
1883 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1884 
1885 	strncpy(lp->d_typename, "vnd", sizeof(lp->d_typename));
1886 	lp->d_type = DKTYPE_VND;
1887 	strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
1888 	lp->d_rpm = 3600;
1889 	lp->d_interleave = 1;
1890 	lp->d_flags = 0;
1891 
1892 	pp = &lp->d_partitions[RAW_PART];
1893 	pp->p_offset = 0;
1894 	pp->p_size = lp->d_secperunit;
1895 	pp->p_fstype = FS_UNUSED;
1896 	lp->d_npartitions = RAW_PART + 1;
1897 
1898 	lp->d_magic = DISKMAGIC;
1899 	lp->d_magic2 = DISKMAGIC;
1900 	lp->d_checksum = dkcksum(lp);
1901 }
1902 
1903 /*
1904  * Read the disklabel from a vnd.  If one is not present, create a fake one.
1905  */
1906 static void
1907 vndgetdisklabel(dev_t dev, struct vnd_softc *sc)
1908 {
1909 	const char *errstring;
1910 	struct disklabel *lp = sc->sc_dkdev.dk_label;
1911 	struct cpu_disklabel *clp = sc->sc_dkdev.dk_cpulabel;
1912 	int i;
1913 
1914 	memset(clp, 0, sizeof(*clp));
1915 
1916 	vndgetdefaultlabel(sc, lp);
1917 
1918 	/*
1919 	 * Call the generic disklabel extraction routine.
1920 	 */
1921 	errstring = readdisklabel(VNDLABELDEV(dev), vndstrategy, lp, clp);
1922 	if (errstring) {
1923 		/*
1924 		 * Lack of disklabel is common, but we print the warning
1925 		 * anyway, since it might contain other useful information.
1926 		 */
1927 		aprint_normal_dev(sc->sc_dev, "%s\n", errstring);
1928 
1929 		/*
1930 		 * For historical reasons, if there's no disklabel
1931 		 * present, all partitions must be FS_BSDFFS and
1932 		 * occupy the entire disk.
1933 		 */
1934 		for (i = 0; i < MAXPARTITIONS; i++) {
1935 			/*
1936 			 * Don't wipe out port specific hack (such as
1937 			 * dos partition hack of i386 port).
1938 			 */
1939 			if (lp->d_partitions[i].p_size != 0)
1940 				continue;
1941 
1942 			lp->d_partitions[i].p_size = lp->d_secperunit;
1943 			lp->d_partitions[i].p_offset = 0;
1944 			lp->d_partitions[i].p_fstype = FS_BSDFFS;
1945 		}
1946 
1947 		strncpy(lp->d_packname, "default label",
1948 		    sizeof(lp->d_packname));
1949 
1950 		lp->d_npartitions = MAXPARTITIONS;
1951 		lp->d_checksum = dkcksum(lp);
1952 	}
1953 }
1954 
1955 /*
1956  * Wait interruptibly for an exclusive lock.
1957  *
1958  * XXX
1959  * Several drivers do this; it should be abstracted and made MP-safe.
1960  */
1961 static int
1962 vndlock(struct vnd_softc *sc)
1963 {
1964 	int error;
1965 
1966 	while ((sc->sc_flags & VNF_LOCKED) != 0) {
1967 		sc->sc_flags |= VNF_WANTED;
1968 		if ((error = tsleep(sc, PRIBIO | PCATCH, "vndlck", 0)) != 0)
1969 			return error;
1970 	}
1971 	sc->sc_flags |= VNF_LOCKED;
1972 	return 0;
1973 }
1974 
1975 /*
1976  * Unlock and wake up any waiters.
1977  */
1978 static void
1979 vndunlock(struct vnd_softc *sc)
1980 {
1981 
1982 	sc->sc_flags &= ~VNF_LOCKED;
1983 	if ((sc->sc_flags & VNF_WANTED) != 0) {
1984 		sc->sc_flags &= ~VNF_WANTED;
1985 		wakeup(sc);
1986 	}
1987 }
1988 
1989 #ifdef VND_COMPRESSION
1990 /* compressed file read */
1991 static void
1992 compstrategy(struct buf *bp, off_t bn)
1993 {
1994 	int error;
1995 	int unit = vndunit(bp->b_dev);
1996 	struct vnd_softc *vnd =
1997 	    device_lookup_private(&vnd_cd, unit);
1998 	u_int32_t comp_block;
1999 	struct uio auio;
2000 	char *addr;
2001 	int s;
2002 
2003 	/* set up constants for data move */
2004 	auio.uio_rw = UIO_READ;
2005 	UIO_SETUP_SYSSPACE(&auio);
2006 
2007 	/* read, and transfer the data */
2008 	addr = bp->b_data;
2009 	bp->b_resid = bp->b_bcount;
2010 	s = splbio();
2011 	while (bp->b_resid > 0) {
2012 		unsigned length;
2013 		size_t length_in_buffer;
2014 		u_int32_t offset_in_buffer;
2015 		struct iovec aiov;
2016 
2017 		/* calculate the compressed block number */
2018 		comp_block = bn / (off_t)vnd->sc_comp_blksz;
2019 
2020 		/* check for good block number */
2021 		if (comp_block >= vnd->sc_comp_numoffs) {
2022 			bp->b_error = EINVAL;
2023 			splx(s);
2024 			return;
2025 		}
2026 
2027 		/* read in the compressed block, if not in buffer */
2028 		if (comp_block != vnd->sc_comp_buffblk) {
2029 			length = vnd->sc_comp_offsets[comp_block + 1] -
2030 			    vnd->sc_comp_offsets[comp_block];
2031 			vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
2032 			error = vn_rdwr(UIO_READ, vnd->sc_vp, vnd->sc_comp_buff,
2033 			    length, vnd->sc_comp_offsets[comp_block],
2034 			    UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, vnd->sc_cred,
2035 			    NULL, NULL);
2036 			if (error) {
2037 				bp->b_error = error;
2038 				VOP_UNLOCK(vnd->sc_vp);
2039 				splx(s);
2040 				return;
2041 			}
2042 			/* uncompress the buffer */
2043 			vnd->sc_comp_stream.next_in = vnd->sc_comp_buff;
2044 			vnd->sc_comp_stream.avail_in = length;
2045 			vnd->sc_comp_stream.next_out = vnd->sc_comp_decombuf;
2046 			vnd->sc_comp_stream.avail_out = vnd->sc_comp_blksz;
2047 			inflateReset(&vnd->sc_comp_stream);
2048 			error = inflate(&vnd->sc_comp_stream, Z_FINISH);
2049 			if (error != Z_STREAM_END) {
2050 				if (vnd->sc_comp_stream.msg)
2051 					aprint_normal_dev(vnd->sc_dev,
2052 					    "compressed file, %s\n",
2053 					    vnd->sc_comp_stream.msg);
2054 				bp->b_error = EBADMSG;
2055 				VOP_UNLOCK(vnd->sc_vp);
2056 				splx(s);
2057 				return;
2058 			}
2059 			vnd->sc_comp_buffblk = comp_block;
2060 			VOP_UNLOCK(vnd->sc_vp);
2061 		}
2062 
2063 		/* transfer the usable uncompressed data */
2064 		offset_in_buffer = bn % (off_t)vnd->sc_comp_blksz;
2065 		length_in_buffer = vnd->sc_comp_blksz - offset_in_buffer;
2066 		if (length_in_buffer > bp->b_resid)
2067 			length_in_buffer = bp->b_resid;
2068 		auio.uio_iov = &aiov;
2069 		auio.uio_iovcnt = 1;
2070 		aiov.iov_base = addr;
2071 		aiov.iov_len = length_in_buffer;
2072 		auio.uio_resid = aiov.iov_len;
2073 		auio.uio_offset = 0;
2074 		error = uiomove(vnd->sc_comp_decombuf + offset_in_buffer,
2075 		    length_in_buffer, &auio);
2076 		if (error) {
2077 			bp->b_error = error;
2078 			splx(s);
2079 			return;
2080 		}
2081 
2082 		bn += length_in_buffer;
2083 		addr += length_in_buffer;
2084 		bp->b_resid -= length_in_buffer;
2085 	}
2086 	splx(s);
2087 }
2088 
2089 /* compression memory allocation routines */
2090 static void *
2091 vnd_alloc(void *aux, u_int items, u_int siz)
2092 {
2093 	return malloc(items * siz, M_TEMP, M_NOWAIT);
2094 }
2095 
2096 static void
2097 vnd_free(void *aux, void *ptr)
2098 {
2099 	free(ptr, M_TEMP);
2100 }
2101 #endif /* VND_COMPRESSION */
2102 
2103 static void
2104 vnd_set_geometry(struct vnd_softc *vnd)
2105 {
2106 	struct disk_geom *dg = &vnd->sc_dkdev.dk_geom;
2107 
2108 	memset(dg, 0, sizeof(*dg));
2109 
2110 	dg->dg_secperunit = (int64_t)vnd->sc_geom.vng_nsectors *
2111 	    vnd->sc_geom.vng_ntracks * vnd->sc_geom.vng_ncylinders;
2112 	dg->dg_secsize = vnd->sc_geom.vng_secsize;
2113 	dg->dg_nsectors = vnd->sc_geom.vng_nsectors;
2114 	dg->dg_ntracks = vnd->sc_geom.vng_ntracks;
2115 	dg->dg_ncylinders = vnd->sc_geom.vng_ncylinders;
2116 
2117 #ifdef DEBUG
2118 	if (vnddebug & VDB_LABEL) {
2119 		printf("dg->dg_secperunit: %" PRId64 "\n", dg->dg_secperunit);
2120 		printf("dg->dg_ncylinders: %u\n", dg->dg_ncylinders);
2121 	}
2122 #endif
2123 	disk_set_info(vnd->sc_dev, &vnd->sc_dkdev, NULL);
2124 }
2125 
2126 #ifdef VND_COMPRESSION
2127 #define VND_DEPENDS "zlib"
2128 #else
2129 #define VND_DEPENDS NULL
2130 #endif
2131 
2132 MODULE(MODULE_CLASS_DRIVER, vnd, VND_DEPENDS);
2133 
2134 #ifdef _MODULE
2135 int vnd_bmajor = -1, vnd_cmajor = -1;
2136 
2137 CFDRIVER_DECL(vnd, DV_DISK, NULL);
2138 #endif
2139 
2140 static int
2141 vnd_modcmd(modcmd_t cmd, void *arg)
2142 {
2143 	int error = 0;
2144 
2145 	switch (cmd) {
2146 	case MODULE_CMD_INIT:
2147 #ifdef _MODULE
2148 		error = config_cfdriver_attach(&vnd_cd);
2149 		if (error)
2150 			break;
2151 
2152 		error = config_cfattach_attach(vnd_cd.cd_name, &vnd_ca);
2153 	        if (error) {
2154 			config_cfdriver_detach(&vnd_cd);
2155 #ifdef DIAGNOSTIC
2156 			aprint_error("%s: unable to register cfattach for \n"
2157 			    "%s, error %d", __func__, vnd_cd.cd_name, error);
2158 #endif
2159 			break;
2160 		}
2161 
2162                 /*
2163                  * Attach the {b,c}devsw's
2164                  */
2165 		error = devsw_attach("vnd", &vnd_bdevsw, &vnd_bmajor,
2166 		    &vnd_cdevsw, &vnd_cmajor);
2167                 /*
2168                  * If devsw_attach fails, remove from autoconf database
2169                  */
2170 		if (error) {
2171 			config_cfattach_detach(vnd_cd.cd_name, &vnd_ca);
2172 			config_cfdriver_detach(&vnd_cd);
2173 #ifdef DIAGNOSTIC
2174                         aprint_error("%s: unable to attach %s devsw, "
2175                             "error %d", __func__, vnd_cd.cd_name, error);
2176 #endif
2177 			break;
2178 		}
2179 #endif
2180 		break;
2181 
2182 	case MODULE_CMD_FINI:
2183 #ifdef _MODULE
2184                 /*
2185                  * Remove {b,c}devsw's
2186                  */
2187 		devsw_detach(&vnd_bdevsw, &vnd_cdevsw);
2188 
2189                 /*
2190                  * Now remove device from autoconf database
2191                  */
2192 		error = config_cfattach_detach(vnd_cd.cd_name, &vnd_ca);
2193                 if (error) {
2194                         (void)devsw_attach("vnd", &vnd_bdevsw, &vnd_bmajor,
2195                             &vnd_cdevsw, &vnd_cmajor);
2196 #ifdef DIAGNOSTIC
2197                         aprint_error("%s: failed to detach %s cfattach, "
2198                             "error %d\n", __func__, vnd_cd.cd_name, error);
2199 #endif
2200                         break;
2201                 }
2202                 error = config_cfdriver_detach(&vnd_cd);
2203                 if (error) {
2204                         (void)config_cfattach_attach(vnd_cd.cd_name, &vnd_ca);
2205                         (void)devsw_attach("vnd", &vnd_bdevsw, &vnd_bmajor,
2206                             &vnd_cdevsw, &vnd_cmajor);
2207 #ifdef DIAGNOSTIC
2208                         aprint_error("%s: failed to detach %s cfdriver, "
2209                             "error %d\n", __func__, vnd_cd.cd_name, error);
2210                         break;
2211 #endif
2212                 }
2213 #endif
2214 		break;
2215 
2216 	case MODULE_CMD_STAT:
2217 		return ENOTTY;
2218 
2219 	default:
2220 		return ENOTTY;
2221 	}
2222 
2223 	return error;
2224 }
2225