xref: /netbsd-src/sys/dev/vnd.c (revision 7d62b00eb9ad855ffcd7da46b41e23feb5476fac)
1 /*	$NetBSD: vnd.c,v 1.288 2023/03/14 12:55:43 hannken 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 1.288 2023/03/14 12:55:43 hannken 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 		/*
557 		 * Limit the number of pending requests to not exhaust
558 		 * resources needed for I/O but always allow the worker
559 		 * thread to add requests, as a wedge on vnd queues
560 		 * requests with biodone() -> dkstart() -> vndstrategy().
561 		 */
562 		if (curlwp != vnd->sc_kthread) {
563 			while (vnd->sc_pending >= VND_MAXPENDING(vnd))
564 				tsleep(&vnd->sc_pending, PRIBIO, "vndpc", 0);
565 		}
566 		vnd->sc_pending++;
567 		KASSERT(vnd->sc_pending > 0);
568 	}
569 	bufq_put(vnd->sc_tab, bp);
570 	wakeup(&vnd->sc_tab);
571 	splx(s);
572 	return;
573 
574 done:
575 	bp->b_resid = bp->b_bcount;
576 	biodone(bp);
577 	splx(s);
578 }
579 
580 static bool
581 vnode_has_strategy(struct vnd_softc *vnd)
582 {
583 	return vnode_has_op(vnd->sc_vp, VOFFSET(vop_bmap)) &&
584 	    vnode_has_op(vnd->sc_vp, VOFFSET(vop_strategy));
585 }
586 
587 /* Verify that I/O requests cannot be smaller than the
588  * smallest I/O size supported by the backend.
589  */
590 static bool
591 vnode_has_large_blocks(struct vnd_softc *vnd)
592 {
593 	u_int32_t vnd_secsize, iosize;
594 
595 	iosize = vnd->sc_iosize;
596 	vnd_secsize = vnd->sc_geom.vng_secsize;
597 
598 	return vnd_secsize % iosize != 0;
599 }
600 
601 /* XXX this function needs a reliable check to detect
602  * sparse files. Otherwise, bmap/strategy may be used
603  * and fail on non-allocated blocks. VOP_READ/VOP_WRITE
604  * works on sparse files.
605  */
606 #if notyet
607 static bool
608 vnode_strategy_probe(struct vnd_softc *vnd)
609 {
610 	int error;
611 	daddr_t nbn;
612 
613 	if (!vnode_has_strategy(vnd))
614 		return false;
615 
616 	if (vnode_has_large_blocks(vnd))
617 		return false;
618 
619 	/* Convert the first logical block number to its
620 	 * physical block number.
621 	 */
622 	error = 0;
623 	vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
624 	error = VOP_BMAP(vnd->sc_vp, 0, NULL, &nbn, NULL);
625 	VOP_UNLOCK(vnd->sc_vp);
626 
627 	/* Test if that worked. */
628 	if (error == 0 && (long)nbn == -1)
629 		return false;
630 
631 	return true;
632 }
633 #endif
634 
635 static void
636 vndthread(void *arg)
637 {
638 	struct vnd_softc *vnd = arg;
639 	int s;
640 
641 	/* Determine whether we can *use* VOP_BMAP and VOP_STRATEGY to
642 	 * directly access the backing vnode.  If we can, use these two
643 	 * operations to avoid messing with the local buffer cache.
644 	 * Otherwise fall back to regular VOP_READ/VOP_WRITE operations
645 	 * which are guaranteed to work with any file system. */
646 	if ((vnd->sc_flags & VNF_USE_VN_RDWR) == 0 &&
647 	    ! vnode_has_strategy(vnd))
648 		vnd->sc_flags |= VNF_USE_VN_RDWR;
649 
650 	/* VOP_STRATEGY can only be used if the backing vnode allows
651 	 * to access blocks as small as defined by the vnd geometry.
652 	 */
653 	if ((vnd->sc_flags & VNF_USE_VN_RDWR) == 0 &&
654 	    vnode_has_large_blocks(vnd))
655 		vnd->sc_flags |= VNF_USE_VN_RDWR;
656 
657 #ifdef DEBUG
658 	if (vnddebug & VDB_INIT)
659 		printf("vndthread: vp %p, %s\n", vnd->sc_vp,
660 		    (vnd->sc_flags & VNF_USE_VN_RDWR) == 0 ?
661 		    "using bmap/strategy operations" :
662 		    "using read/write operations");
663 #endif
664 
665 	s = splbio();
666 	vnd->sc_flags |= VNF_KTHREAD;
667 	wakeup(&vnd->sc_kthread);
668 
669 	/*
670 	 * Dequeue requests and serve them depending on the available
671 	 * vnode operations.
672 	 */
673 	while ((vnd->sc_flags & VNF_VUNCONF) == 0) {
674 		struct vndxfer *vnx;
675 		struct buf *obp;
676 		struct buf *bp;
677 
678 		obp = bufq_get(vnd->sc_tab);
679 		if (obp == NULL) {
680 			tsleep(&vnd->sc_tab, PRIBIO, "vndbp", 0);
681 			continue;
682 		};
683 		if ((vnd->sc_flags & VNF_USE_VN_RDWR)) {
684 			KASSERT(vnd->sc_pending > 0);
685 			if (vnd->sc_pending-- == VND_MAXPENDING(vnd))
686 				wakeup(&vnd->sc_pending);
687 		}
688 		splx(s);
689 #ifdef DEBUG
690 		if (vnddebug & VDB_FOLLOW)
691 			printf("vndthread(%p)\n", obp);
692 #endif
693 
694 		if (vnd->sc_vp->v_mount == NULL) {
695 			obp->b_error = ENXIO;
696 			goto done;
697 		}
698 #ifdef VND_COMPRESSION
699 		/* handle a compressed read */
700 		if ((obp->b_flags & B_READ) != 0 && (vnd->sc_flags & VNF_COMP)) {
701 			off_t bn;
702 
703 			/* Convert to a byte offset within the file. */
704 			bn = obp->b_rawblkno *
705 			    vnd->sc_dkdev.dk_label->d_secsize;
706 
707 			compstrategy(obp, bn);
708 			goto done;
709 		}
710 #endif /* VND_COMPRESSION */
711 
712 		/*
713 		 * Allocate a header for this transfer and link it to the
714 		 * buffer
715 		 */
716 		s = splbio();
717 		vnx = VND_GETXFER(vnd);
718 		splx(s);
719 		vnx->vx_vnd = vnd;
720 
721 		s = splbio();
722 		while (vnd->sc_active >= vnd->sc_maxactive) {
723 			tsleep(&vnd->sc_tab, PRIBIO, "vndac", 0);
724 		}
725 		vnd->sc_active++;
726 		splx(s);
727 
728 		/* Instrumentation. */
729 		disk_busy(&vnd->sc_dkdev);
730 
731 		bp = &vnx->vx_buf;
732 		buf_init(bp);
733 		bp->b_flags = (obp->b_flags & (B_READ | B_PHYS | B_RAW));
734 		bp->b_oflags = obp->b_oflags;
735 		bp->b_cflags = obp->b_cflags;
736 		bp->b_iodone = vndiodone;
737 		bp->b_private = obp;
738 		bp->b_vp = vnd->sc_vp;
739 		bp->b_objlock = bp->b_vp->v_interlock;
740 		bp->b_data = obp->b_data;
741 		bp->b_bcount = obp->b_bcount;
742 		BIO_COPYPRIO(bp, obp);
743 
744 		/* Make sure the request succeeds while suspending this fs. */
745 		fstrans_start_lazy(vnd->sc_vp->v_mount);
746 
747 		/* Handle the request using the appropriate operations. */
748 		if ((vnd->sc_flags & VNF_USE_VN_RDWR) == 0)
749 			handle_with_strategy(vnd, obp, bp);
750 		else
751 			handle_with_rdwr(vnd, obp, bp);
752 
753 		fstrans_done(vnd->sc_vp->v_mount);
754 
755 		s = splbio();
756 		continue;
757 
758 done:
759 		biodone(obp);
760 		s = splbio();
761 	}
762 
763 	vnd->sc_flags &= (~VNF_KTHREAD | VNF_VUNCONF);
764 	wakeup(&vnd->sc_kthread);
765 	splx(s);
766 	kthread_exit(0);
767 }
768 
769 /*
770  * Checks if the given vnode supports the requested operation.
771  * The operation is specified the offset returned by VOFFSET.
772  *
773  * XXX The test below used to determine this is quite fragile
774  * because it relies on the file system to use genfs to specify
775  * unimplemented operations.  There might be another way to do
776  * it more cleanly.
777  */
778 static bool
779 vnode_has_op(const struct vnode *vp, int opoffset)
780 {
781 	int (*defaultp)(void *);
782 	int (*opp)(void *);
783 
784 	defaultp = vp->v_op[VOFFSET(vop_default)];
785 	opp = vp->v_op[opoffset];
786 
787 	return opp != defaultp && opp != genfs_eopnotsupp &&
788 	    opp != genfs_badop && opp != genfs_nullop;
789 }
790 
791 /*
792  * Handles the read/write request given in 'bp' using the vnode's VOP_READ
793  * and VOP_WRITE operations.
794  *
795  * 'obp' is a pointer to the original request fed to the vnd device.
796  */
797 static void
798 handle_with_rdwr(struct vnd_softc *vnd, const struct buf *obp, struct buf *bp)
799 {
800 	bool doread;
801 	off_t offset;
802 	size_t len, resid;
803 	struct vnode *vp;
804 	int npages;
805 
806 	doread = bp->b_flags & B_READ;
807 	offset = obp->b_rawblkno * vnd->sc_dkdev.dk_label->d_secsize;
808 	len = bp->b_bcount;
809 	vp = vnd->sc_vp;
810 
811 #if defined(DEBUG)
812 	if (vnddebug & VDB_IO)
813 		printf("vnd (rdwr): vp %p, %s, rawblkno 0x%" PRIx64
814 		    ", secsize %d, offset %" PRIu64
815 		    ", bcount %d\n",
816 		    vp, doread ? "read" : "write", obp->b_rawblkno,
817 		    vnd->sc_dkdev.dk_label->d_secsize, offset,
818 		    bp->b_bcount);
819 #endif
820 
821 	/* Issue the read or write operation. */
822 	bp->b_error =
823 	    vn_rdwr(doread ? UIO_READ : UIO_WRITE,
824 	    vp, bp->b_data, len, offset, UIO_SYSSPACE,
825 	    IO_ADV_ENCODE(POSIX_FADV_NOREUSE) | IO_DIRECT,
826 		vnd->sc_cred, &resid, NULL);
827 	bp->b_resid = resid;
828 
829 	/*
830 	 * Avoid caching too many pages, the vnd user
831 	 * is usually a filesystem and caches itself.
832 	 * We need some amount of caching to not hinder
833 	 * read-ahead and write-behind operations.
834 	 */
835 	npages = atomic_load_relaxed(&vp->v_uobj.uo_npages);
836 	if (npages > VND_MAXPAGES(vnd)) {
837 		rw_enter(vp->v_uobj.vmobjlock, RW_WRITER);
838 		(void) VOP_PUTPAGES(vp, 0, 0,
839 		    PGO_ALLPAGES | PGO_CLEANIT | PGO_FREE);
840 	}
841 
842 	/* We need to increase the number of outputs on the vnode if
843 	 * there was any write to it. */
844 	if (!doread) {
845 		mutex_enter(vp->v_interlock);
846 		vp->v_numoutput++;
847 		mutex_exit(vp->v_interlock);
848 	}
849 
850 	biodone(bp);
851 }
852 
853 /*
854  * Handes the read/write request given in 'bp' using the vnode's VOP_BMAP
855  * and VOP_STRATEGY operations.
856  *
857  * 'obp' is a pointer to the original request fed to the vnd device.
858  */
859 static void
860 handle_with_strategy(struct vnd_softc *vnd, const struct buf *obp,
861     struct buf *bp)
862 {
863 	int bsize, error, flags, skipped;
864 	size_t resid, sz;
865 	off_t bn, offset;
866 	struct vnode *vp;
867 	struct buf *nbp = NULL;
868 
869 	flags = obp->b_flags;
870 
871 
872 	/* convert to a byte offset within the file. */
873 	bn = obp->b_rawblkno * vnd->sc_dkdev.dk_label->d_secsize;
874 
875 	bsize = vnd->sc_vp->v_mount->mnt_stat.f_iosize;
876 	skipped = 0;
877 
878 	/*
879 	 * Break the request into bsize pieces and feed them
880 	 * sequentially using VOP_BMAP/VOP_STRATEGY.
881 	 * We do it this way to keep from flooding NFS servers if we
882 	 * are connected to an NFS file.  This places the burden on
883 	 * the client rather than the server.
884 	 */
885 	error = 0;
886 	bp->b_resid = bp->b_bcount;
887 	for (offset = 0, resid = bp->b_resid; /* true */;
888 	    resid -= sz, offset += sz) {
889 		daddr_t nbn;
890 		int off, nra;
891 
892 		nra = 0;
893 		vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
894 		error = VOP_BMAP(vnd->sc_vp, bn / bsize, &vp, &nbn, &nra);
895 		VOP_UNLOCK(vnd->sc_vp);
896 
897 		if (error == 0 && (long)nbn == -1)
898 			error = EIO;
899 
900 		/*
901 		 * If there was an error or a hole in the file...punt.
902 		 * Note that we may have to wait for any operations
903 		 * that we have already fired off before releasing
904 		 * the buffer.
905 		 *
906 		 * XXX we could deal with holes here but it would be
907 		 * a hassle (in the write case).
908 		 */
909 		if (error) {
910 			skipped += resid;
911 			break;
912 		}
913 
914 #ifdef DEBUG
915 		if (!dovndcluster)
916 			nra = 0;
917 #endif
918 
919 		off = bn % bsize;
920 		sz = MIN(((off_t)1 + nra) * bsize - off, resid);
921 #ifdef	DEBUG
922 		if (vnddebug & VDB_IO)
923 			printf("vndstrategy: vp %p/%p bn 0x%qx/0x%" PRIx64
924 			    " sz 0x%zx\n", vnd->sc_vp, vp, (long long)bn,
925 			    nbn, sz);
926 #endif
927 
928 		nbp = getiobuf(vp, true);
929 		nestiobuf_setup(bp, nbp, offset, sz);
930 		nbp->b_blkno = nbn + btodb(off);
931 
932 #if 0 /* XXX #ifdef DEBUG */
933 		if (vnddebug & VDB_IO)
934 			printf("vndstart(%ld): bp %p vp %p blkno "
935 			    "0x%" PRIx64 " flags %x addr %p cnt 0x%x\n",
936 			    (long) (vnd-vnd_softc), &nbp->vb_buf,
937 			    nbp->vb_buf.b_vp, nbp->vb_buf.b_blkno,
938 			    nbp->vb_buf.b_flags, nbp->vb_buf.b_data,
939 			    nbp->vb_buf.b_bcount);
940 #endif
941 		if (resid == sz) {
942 			break;
943 		}
944 		VOP_STRATEGY(vp, nbp);
945 		bn += sz;
946 	}
947 	if (!(flags & B_READ)) {
948 		struct vnode *w_vp;
949 		/*
950 		 * this is the last nested buf, account for
951 		 * the parent buf write too.
952 		 * This has to be done last, so that
953 		 * fsync won't wait for this write which
954 		 * has no chance to complete before all nested bufs
955 		 * have been queued. But it has to be done
956 		 * before the last VOP_STRATEGY()
957 		 * or the call to nestiobuf_done().
958 		 */
959 		w_vp = bp->b_vp;
960 		mutex_enter(w_vp->v_interlock);
961 		w_vp->v_numoutput++;
962 		mutex_exit(w_vp->v_interlock);
963 	}
964 	KASSERT(skipped != 0 || nbp != NULL);
965 	if (skipped)
966 		nestiobuf_done(bp, skipped, error);
967 	else
968 		VOP_STRATEGY(vp, nbp);
969 }
970 
971 static void
972 vndiodone(struct buf *bp)
973 {
974 	struct vndxfer *vnx = VND_BUFTOXFER(bp);
975 	struct vnd_softc *vnd = vnx->vx_vnd;
976 	struct buf *obp = bp->b_private;
977 	int s = splbio();
978 
979 	KERNEL_LOCK(1, NULL);		/* XXXSMP */
980 	KASSERT(&vnx->vx_buf == bp);
981 	KASSERT(vnd->sc_active > 0);
982 #ifdef DEBUG
983 	if (vnddebug & VDB_IO) {
984 		printf("vndiodone1: bp %p iodone: error %d\n",
985 		    bp, bp->b_error);
986 	}
987 #endif
988 	disk_unbusy(&vnd->sc_dkdev, bp->b_bcount - bp->b_resid,
989 	    (bp->b_flags & B_READ));
990 	vnd->sc_active--;
991 	if (vnd->sc_active == 0) {
992 		wakeup(&vnd->sc_tab);
993 	}
994 	KERNEL_UNLOCK_ONE(NULL);	/* XXXSMP */
995 	splx(s);
996 	obp->b_error = bp->b_error;
997 	obp->b_resid = bp->b_resid;
998 	buf_destroy(bp);
999 	VND_PUTXFER(vnd, vnx);
1000 	biodone(obp);
1001 }
1002 
1003 /* ARGSUSED */
1004 static int
1005 vndread(dev_t dev, struct uio *uio, int flags)
1006 {
1007 	int unit = vndunit(dev);
1008 	struct vnd_softc *sc;
1009 
1010 #ifdef DEBUG
1011 	if (vnddebug & VDB_FOLLOW)
1012 		printf("vndread(0x%"PRIx64", %p)\n", dev, uio);
1013 #endif
1014 
1015 	sc = device_lookup_private(&vnd_cd, unit);
1016 	if (sc == NULL)
1017 		return ENXIO;
1018 
1019 	if ((sc->sc_flags & VNF_INITED) == 0)
1020 		return ENXIO;
1021 
1022 	return physio(vndstrategy, NULL, dev, B_READ, minphys, uio);
1023 }
1024 
1025 /* ARGSUSED */
1026 static int
1027 vndwrite(dev_t dev, struct uio *uio, int flags)
1028 {
1029 	int unit = vndunit(dev);
1030 	struct vnd_softc *sc;
1031 
1032 #ifdef DEBUG
1033 	if (vnddebug & VDB_FOLLOW)
1034 		printf("vndwrite(0x%"PRIx64", %p)\n", dev, uio);
1035 #endif
1036 
1037 	sc = device_lookup_private(&vnd_cd, unit);
1038 	if (sc == NULL)
1039 		return ENXIO;
1040 
1041 	if ((sc->sc_flags & VNF_INITED) == 0)
1042 		return ENXIO;
1043 
1044 	return physio(vndstrategy, NULL, dev, B_WRITE, minphys, uio);
1045 }
1046 
1047 static int
1048 vnd_cget(struct lwp *l, int unit, int *un, struct vattr *va)
1049 {
1050 	int error;
1051 	struct vnd_softc *vnd;
1052 
1053 	if (*un == -1)
1054 		*un = unit;
1055 	if (*un < 0)
1056 		return EINVAL;
1057 
1058 	vnd = device_lookup_private(&vnd_cd, *un);
1059 	if (vnd == NULL)
1060 		return -1;
1061 
1062 	if ((vnd->sc_flags & VNF_INITED) == 0)
1063 		return -1;
1064 
1065 	vn_lock(vnd->sc_vp, LK_SHARED | LK_RETRY);
1066 	error = VOP_GETATTR(vnd->sc_vp, va, l->l_cred);
1067 	VOP_UNLOCK(vnd->sc_vp);
1068 	return error;
1069 }
1070 
1071 static int
1072 vnddoclear(struct vnd_softc *vnd, int pmask, int minor, bool force)
1073 {
1074 	int error;
1075 
1076 	if ((error = vndlock(vnd)) != 0)
1077 		return error;
1078 
1079 	/*
1080 	 * Don't unconfigure if any other partitions are open
1081 	 * or if both the character and block flavors of this
1082 	 * partition are open.
1083 	 */
1084 	if (DK_BUSY(vnd, pmask) && !force) {
1085 		vndunlock(vnd);
1086 		return EBUSY;
1087 	}
1088 
1089 	/* Delete all of our wedges */
1090 	dkwedge_delall(&vnd->sc_dkdev);
1091 
1092 	/*
1093 	 * XXX vndclear() might call vndclose() implicitly;
1094 	 * release lock to avoid recursion
1095 	 *
1096 	 * Set VNF_CLEARING to prevent vndopen() from
1097 	 * sneaking in after we vndunlock().
1098 	 */
1099 	vnd->sc_flags |= VNF_CLEARING;
1100 	vndunlock(vnd);
1101 	vndclear(vnd, minor);
1102 #ifdef DEBUG
1103 	if (vnddebug & VDB_INIT)
1104 		printf("%s: CLRed\n", __func__);
1105 #endif
1106 
1107 	/* Destroy the xfer and buffer pools. */
1108 	pool_destroy(&vnd->sc_vxpool);
1109 
1110 	/* Detach the disk. */
1111 	disk_detach(&vnd->sc_dkdev);
1112 
1113 	return 0;
1114 }
1115 
1116 static int
1117 vndioctl_get(struct lwp *l, void *data, int unit, struct vattr *va)
1118 {
1119 	int error;
1120 
1121 	KASSERT(l);
1122 
1123 	/* the first member is always int vnd_unit in all the versions */
1124 	if (*(int *)data >= vnd_cd.cd_ndevs)
1125 		return ENXIO;
1126 
1127 	switch (error = vnd_cget(l, unit, (int *)data, va)) {
1128 	case -1:
1129 		/* unused is not an error */
1130 		memset(va, 0, sizeof(*va));
1131 		/*FALLTHROUGH*/
1132 	case 0:
1133 		return 0;
1134 	default:
1135 		return error;
1136 	}
1137 }
1138 
1139 /* ARGSUSED */
1140 static int
1141 vndioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l)
1142 {
1143 	bool force;
1144 	int unit = vndunit(dev);
1145 	struct vnd_softc *vnd;
1146 	struct vnd_ioctl *vio;
1147 	struct vattr vattr;
1148 	struct pathbuf *pb;
1149 	struct vnode *vp;
1150 	int error, part, pmask;
1151 	uint64_t geomsize;
1152 	int fflags;
1153 #ifdef __HAVE_OLD_DISKLABEL
1154 	struct disklabel newlabel;
1155 #endif
1156 
1157 #ifdef DEBUG
1158 	if (vnddebug & VDB_FOLLOW)
1159 		printf("vndioctl(0x%"PRIx64", 0x%lx, %p, 0x%x, %p): unit %d\n",
1160 		    dev, cmd, data, flag, l->l_proc, unit);
1161 #endif
1162 	/* Do the get's first; they don't need initialization or verification */
1163 	switch (cmd) {
1164 	case VNDIOCGET:
1165 		if ((error = vndioctl_get(l, data, unit, &vattr)) != 0)
1166 			return error;
1167 
1168 		struct vnd_user *vnu = data;
1169 		vnu->vnu_dev = vattr.va_fsid;
1170 		vnu->vnu_ino = vattr.va_fileid;
1171 		return 0;
1172 
1173 	default:
1174 		/* First check for COMPAT_50 hook */
1175 		MODULE_HOOK_CALL(compat_vndioctl_50_hook,
1176 		    (cmd, l, data, unit, &vattr, vndioctl_get),
1177 		    enosys(), error);
1178 
1179 		/*
1180 		 * If not present, then COMPAT_30 hook also not
1181 		 * present, so just continue with checks for the
1182 		 * "write" commands
1183 		 */
1184 		if (error == ENOSYS) {
1185 			error = 0;
1186 			break;
1187 		}
1188 
1189 		/* If not already handled, try the COMPAT_30 hook */
1190 		if (error == EPASSTHROUGH)
1191 			MODULE_HOOK_CALL(compat_vndioctl_30_hook,
1192 			    (cmd, l, data, unit, &vattr, vndioctl_get),
1193 			    enosys(), error);
1194 
1195 		/* If no COMPAT_30 module, or not handled, check writes */
1196 		if (error == ENOSYS || error == EPASSTHROUGH) {
1197 			error = 0;
1198 			break;
1199 		}
1200 		return error;
1201 	}
1202 
1203 	vnd = device_lookup_private(&vnd_cd, unit);
1204 	if (vnd == NULL)
1205 		return ENXIO;
1206 	vio = (struct vnd_ioctl *)data;
1207 
1208 	/* Must be open for writes for these commands... */
1209 	switch (cmd) {
1210 	case VNDIOCSET50:
1211 	case VNDIOCCLR50:
1212 		if (!compat_vndioctl_50_hook.hooked)
1213 			return EINVAL;
1214 		/* FALLTHROUGH */
1215 	case VNDIOCSET:
1216 	case VNDIOCCLR:
1217 	case DIOCSDINFO:
1218 	case DIOCWDINFO:
1219 #ifdef __HAVE_OLD_DISKLABEL
1220 	case ODIOCSDINFO:
1221 	case ODIOCWDINFO:
1222 #endif
1223 	case DIOCKLABEL:
1224 	case DIOCWLABEL:
1225 	case DIOCCACHESYNC:
1226 		if ((flag & FWRITE) == 0)
1227 			return EBADF;
1228 	}
1229 
1230 	switch (cmd) {
1231 	case VNDIOCSET50:
1232 	case VNDIOCSET:
1233 		/* Must not be initialized */
1234 		if (vnd->sc_flags & VNF_INITED)
1235 			return EBUSY;
1236 		break;
1237 	default:
1238 		/* Must be initialized */
1239 		if ((vnd->sc_flags & VNF_INITED) == 0)
1240 			return ENXIO;
1241 		break;
1242 	}
1243 
1244 	error = disk_ioctl(&vnd->sc_dkdev, dev, cmd, data, flag, l);
1245 	if (error != EPASSTHROUGH)
1246 		return error;
1247 
1248 	switch (cmd) {
1249 	case VNDIOCSET50:
1250 	case VNDIOCSET:
1251 		if ((error = vndlock(vnd)) != 0)
1252 			return error;
1253 
1254 		fflags = FREAD;
1255 		if ((vio->vnd_flags & VNDIOF_READONLY) == 0)
1256 			fflags |= FWRITE;
1257 		if ((vio->vnd_flags & VNDIOF_FILEIO) != 0)
1258 			vnd->sc_flags |= VNF_USE_VN_RDWR;
1259 		error = pathbuf_copyin(vio->vnd_file, &pb);
1260 		if (error) {
1261 			goto unlock_and_exit;
1262 		}
1263 		error = vn_open(NULL, pb, 0, fflags, 0, &vp, NULL, NULL);
1264 		if (error != 0) {
1265 			pathbuf_destroy(pb);
1266 			goto unlock_and_exit;
1267 		}
1268 		KASSERT(l);
1269 		error = VOP_GETATTR(vp, &vattr, l->l_cred);
1270 		if (!error && vp->v_type != VREG)
1271 			error = EOPNOTSUPP;
1272 		if (!error && vattr.va_bytes < vattr.va_size)
1273 			/* File is definitely sparse, use vn_rdwr() */
1274 			vnd->sc_flags |= VNF_USE_VN_RDWR;
1275 		if (error) {
1276 			VOP_UNLOCK(vp);
1277 			goto close_and_exit;
1278 		}
1279 
1280 		/* If using a compressed file, initialize its info */
1281 		/* (or abort with an error if kernel has no compression) */
1282 		if (vio->vnd_flags & VNDIOF_COMP) {
1283 #ifdef VND_COMPRESSION
1284 			struct vnd_comp_header *ch;
1285 			int i;
1286 			uint32_t comp_size;
1287 			uint32_t comp_maxsize;
1288 
1289 			/* allocate space for compressed file header */
1290 			ch = malloc(sizeof(struct vnd_comp_header),
1291 			    M_TEMP, M_WAITOK);
1292 
1293 			/* read compressed file header */
1294 			error = vn_rdwr(UIO_READ, vp, (void *)ch,
1295 			    sizeof(struct vnd_comp_header), 0, UIO_SYSSPACE,
1296 			    IO_UNIT|IO_NODELOCKED, l->l_cred, NULL, NULL);
1297 			if (error) {
1298 				free(ch, M_TEMP);
1299 				VOP_UNLOCK(vp);
1300 				goto close_and_exit;
1301 			}
1302 
1303 			if (be32toh(ch->block_size) == 0 ||
1304 			    be32toh(ch->num_blocks) > UINT32_MAX - 1) {
1305 				free(ch, M_TEMP);
1306 				VOP_UNLOCK(vp);
1307 				goto close_and_exit;
1308 			}
1309 
1310 			/* save some header info */
1311 			vnd->sc_comp_blksz = be32toh(ch->block_size);
1312 			/* note last offset is the file byte size */
1313 			vnd->sc_comp_numoffs = be32toh(ch->num_blocks) + 1;
1314 			free(ch, M_TEMP);
1315 			if (!DK_DEV_BSIZE_OK(vnd->sc_comp_blksz)) {
1316 				VOP_UNLOCK(vp);
1317 				error = EINVAL;
1318 				goto close_and_exit;
1319 			}
1320 			KASSERT(0 < vnd->sc_comp_blksz);
1321 			KASSERT(0 < vnd->sc_comp_numoffs);
1322 			/*
1323 			 * @#^@!$& gcc -Wtype-limits refuses to let me
1324 			 * write SIZE_MAX/sizeof(uint64_t) < numoffs,
1325 			 * because the range of the type on amd64 makes
1326 			 * the comparisons always false.
1327 			 */
1328 #if SIZE_MAX <= UINT32_MAX*(64/CHAR_BIT)
1329 			if (SIZE_MAX/sizeof(uint64_t) < vnd->sc_comp_numoffs) {
1330 				VOP_UNLOCK(vp);
1331 				error = EINVAL;
1332 				goto close_and_exit;
1333 			}
1334 #endif
1335 			if ((vattr.va_size < sizeof(struct vnd_comp_header)) ||
1336 			    (vattr.va_size - sizeof(struct vnd_comp_header) <
1337 				sizeof(uint64_t)*vnd->sc_comp_numoffs) ||
1338 			    (UQUAD_MAX/vnd->sc_comp_blksz <
1339 				vnd->sc_comp_numoffs - 1)) {
1340 				VOP_UNLOCK(vp);
1341 				error = EINVAL;
1342 				goto close_and_exit;
1343 			}
1344 
1345 			/* set decompressed file size */
1346 			KASSERT(vnd->sc_comp_numoffs - 1 <=
1347 			    UQUAD_MAX/vnd->sc_comp_blksz);
1348 			vattr.va_size =
1349 			    ((u_quad_t)vnd->sc_comp_numoffs - 1) *
1350 			     (u_quad_t)vnd->sc_comp_blksz;
1351 
1352 			/* allocate space for all the compressed offsets */
1353 			__CTASSERT(UINT32_MAX <= UQUAD_MAX/sizeof(uint64_t));
1354 			vnd->sc_comp_offsets =
1355 			    malloc(sizeof(uint64_t) * vnd->sc_comp_numoffs,
1356 				M_DEVBUF, M_WAITOK);
1357 
1358 			/* read in the offsets */
1359 			error = vn_rdwr(UIO_READ, vp,
1360 			    (void *)vnd->sc_comp_offsets,
1361 			    sizeof(uint64_t) * vnd->sc_comp_numoffs,
1362 			    sizeof(struct vnd_comp_header), UIO_SYSSPACE,
1363 			  IO_UNIT|IO_NODELOCKED, l->l_cred, NULL, NULL);
1364 			if (error) {
1365 				VOP_UNLOCK(vp);
1366 				goto close_and_exit;
1367 			}
1368 			/*
1369 			 * find largest block size (used for allocation limit).
1370 			 * Also convert offset to native byte order.
1371 			 */
1372 			comp_maxsize = 0;
1373 			for (i = 0; i < vnd->sc_comp_numoffs - 1; i++) {
1374 				vnd->sc_comp_offsets[i] =
1375 				    be64toh(vnd->sc_comp_offsets[i]);
1376 				comp_size =
1377 				    be64toh(vnd->sc_comp_offsets[i + 1])
1378 				    - vnd->sc_comp_offsets[i];
1379 				if (comp_size > comp_maxsize)
1380 					comp_maxsize = comp_size;
1381 			}
1382 			vnd->sc_comp_offsets[vnd->sc_comp_numoffs - 1] =
1383 			    be64toh(vnd->sc_comp_offsets[vnd->sc_comp_numoffs
1384 				    - 1]);
1385 
1386 			/* create compressed data buffer */
1387 			vnd->sc_comp_buff = malloc(comp_maxsize,
1388 			    M_DEVBUF, M_WAITOK);
1389 
1390 			/* create decompressed buffer */
1391 			vnd->sc_comp_decombuf = malloc(vnd->sc_comp_blksz,
1392 			    M_DEVBUF, M_WAITOK);
1393 			vnd->sc_comp_buffblk = -1;
1394 
1395 			/* Initialize decompress stream */
1396 			memset(&vnd->sc_comp_stream, 0, sizeof(z_stream));
1397 			vnd->sc_comp_stream.zalloc = vnd_alloc;
1398 			vnd->sc_comp_stream.zfree = vnd_free;
1399 			error = inflateInit2(&vnd->sc_comp_stream, MAX_WBITS);
1400 			if (error) {
1401 				if (vnd->sc_comp_stream.msg)
1402 					printf("vnd%d: compressed file, %s\n",
1403 					    unit, vnd->sc_comp_stream.msg);
1404 				VOP_UNLOCK(vp);
1405 				error = EINVAL;
1406 				goto close_and_exit;
1407 			}
1408 
1409 			vnd->sc_flags |= VNF_COMP | VNF_READONLY;
1410 #else /* !VND_COMPRESSION */
1411 			VOP_UNLOCK(vp);
1412 			error = EOPNOTSUPP;
1413 			goto close_and_exit;
1414 #endif /* VND_COMPRESSION */
1415 		}
1416 
1417 		VOP_UNLOCK(vp);
1418 		vnd->sc_vp = vp;
1419 		vnd->sc_size = btodb(vattr.va_size);	/* note truncation */
1420 
1421 		/* get smallest I/O size for underlying device, fall back to
1422 		 * fundamental I/O size of underlying filesystem
1423 		 */
1424 		error = bdev_ioctl(vattr.va_fsid, DIOCGSECTORSIZE, &vnd->sc_iosize, FKIOCTL, l);
1425 		if (error)
1426 			vnd->sc_iosize = vnd->sc_vp->v_mount->mnt_stat.f_frsize;
1427 
1428 		/* Default I/O size to DEV_BSIZE */
1429 		if (vnd->sc_iosize == 0)
1430 			vnd->sc_iosize = DEV_BSIZE;
1431 
1432 		/*
1433 		 * Use pseudo-geometry specified.  If none was provided,
1434 		 * use "standard" Adaptec fictitious geometry.
1435 		 */
1436 		if (vio->vnd_flags & VNDIOF_HASGEOM) {
1437 
1438 			memcpy(&vnd->sc_geom, &vio->vnd_geom,
1439 			    sizeof(vio->vnd_geom));
1440 
1441 			/*
1442 			 * Sanity-check the sector size.
1443 			 */
1444 			if (!DK_DEV_BSIZE_OK(vnd->sc_geom.vng_secsize) ||
1445 			    vnd->sc_geom.vng_ntracks == 0 ||
1446 			    vnd->sc_geom.vng_nsectors == 0) {
1447 				error = EINVAL;
1448 				goto close_and_exit;
1449 			}
1450 
1451 			/*
1452 			 * Compute missing cylinder count from size
1453 			 */
1454 			if (vnd->sc_geom.vng_ncylinders == 0)
1455 				vnd->sc_geom.vng_ncylinders = vnd->sc_size / (
1456 					(vnd->sc_geom.vng_secsize / DEV_BSIZE) *
1457 					vnd->sc_geom.vng_ntracks *
1458 					vnd->sc_geom.vng_nsectors);
1459 
1460 			/*
1461 			 * Compute the size (in DEV_BSIZE blocks) specified
1462 			 * by the geometry.
1463 			 */
1464 			geomsize = (int64_t)vnd->sc_geom.vng_nsectors *
1465 			    vnd->sc_geom.vng_ntracks *
1466 			    vnd->sc_geom.vng_ncylinders *
1467 			    (vnd->sc_geom.vng_secsize / DEV_BSIZE);
1468 
1469 			/*
1470 			 * Sanity-check the size against the specified
1471 			 * geometry.
1472 			 */
1473 			if (vnd->sc_size < geomsize) {
1474 				error = EINVAL;
1475 				goto close_and_exit;
1476 			}
1477 		} else if (vnd->sc_size >= (32 * 64)) {
1478 			/*
1479 			 * Size must be at least 2048 DEV_BSIZE blocks
1480 			 * (1M) in order to use this geometry.
1481 			 */
1482 			vnd->sc_geom.vng_secsize = DEV_BSIZE;
1483 			vnd->sc_geom.vng_nsectors = 32;
1484 			vnd->sc_geom.vng_ntracks = 64;
1485 			vnd->sc_geom.vng_ncylinders = vnd->sc_size / (64 * 32);
1486 		} else {
1487 			vnd->sc_geom.vng_secsize = DEV_BSIZE;
1488 			vnd->sc_geom.vng_nsectors = 1;
1489 			vnd->sc_geom.vng_ntracks = 1;
1490 			vnd->sc_geom.vng_ncylinders = vnd->sc_size;
1491 		}
1492 
1493 		vnd_set_geometry(vnd);
1494 
1495 		if (vio->vnd_flags & VNDIOF_READONLY) {
1496 			vnd->sc_flags |= VNF_READONLY;
1497 		}
1498 
1499 		if ((error = vndsetcred(vnd, l->l_cred)) != 0)
1500 			goto close_and_exit;
1501 
1502 		vndthrottle(vnd, vnd->sc_vp);
1503 		vio->vnd_osize = dbtob(vnd->sc_size);
1504 		if (cmd != VNDIOCSET50)
1505 			vio->vnd_size = dbtob(vnd->sc_size);
1506 		vnd->sc_flags |= VNF_INITED;
1507 
1508 		/* create the kernel thread, wait for it to be up */
1509 		error = kthread_create(PRI_NONE, 0, NULL, vndthread, vnd,
1510 		    &vnd->sc_kthread, "%s", device_xname(vnd->sc_dev));
1511 		if (error)
1512 			goto close_and_exit;
1513 		while ((vnd->sc_flags & VNF_KTHREAD) == 0) {
1514 			tsleep(&vnd->sc_kthread, PRIBIO, "vndthr", 0);
1515 		}
1516 #ifdef DEBUG
1517 		if (vnddebug & VDB_INIT)
1518 			printf("vndioctl: SET vp %p size 0x%lx %d/%d/%d/%d\n",
1519 			    vnd->sc_vp, (unsigned long) vnd->sc_size,
1520 			    vnd->sc_geom.vng_secsize,
1521 			    vnd->sc_geom.vng_nsectors,
1522 			    vnd->sc_geom.vng_ntracks,
1523 			    vnd->sc_geom.vng_ncylinders);
1524 #endif
1525 
1526 		/* Attach the disk. */
1527 		disk_attach(&vnd->sc_dkdev);
1528 
1529 		/* Initialize the xfer and buffer pools. */
1530 		pool_init(&vnd->sc_vxpool, sizeof(struct vndxfer), 0,
1531 		    0, 0, "vndxpl", NULL, IPL_BIO);
1532 
1533 		vndunlock(vnd);
1534 
1535 		pathbuf_destroy(pb);
1536 
1537 		/* Discover wedges on this disk */
1538 		dkwedge_discover(&vnd->sc_dkdev);
1539 
1540 		break;
1541 
1542 close_and_exit:
1543 		(void) vn_close(vp, fflags, l->l_cred);
1544 		pathbuf_destroy(pb);
1545 unlock_and_exit:
1546 #ifdef VND_COMPRESSION
1547 		/* free any allocated memory (for compressed file) */
1548 		if (vnd->sc_comp_offsets) {
1549 			free(vnd->sc_comp_offsets, M_DEVBUF);
1550 			vnd->sc_comp_offsets = NULL;
1551 		}
1552 		if (vnd->sc_comp_buff) {
1553 			free(vnd->sc_comp_buff, M_DEVBUF);
1554 			vnd->sc_comp_buff = NULL;
1555 		}
1556 		if (vnd->sc_comp_decombuf) {
1557 			free(vnd->sc_comp_decombuf, M_DEVBUF);
1558 			vnd->sc_comp_decombuf = NULL;
1559 		}
1560 #endif /* VND_COMPRESSION */
1561 		vndunlock(vnd);
1562 		return error;
1563 
1564 	case VNDIOCCLR50:
1565 	case VNDIOCCLR:
1566 		part = DISKPART(dev);
1567 		pmask = (1 << part);
1568 		force = (vio->vnd_flags & VNDIOF_FORCE) != 0;
1569 
1570 		if ((error = vnddoclear(vnd, pmask, minor(dev), force)) != 0)
1571 			return error;
1572 
1573 		break;
1574 
1575 
1576 	case DIOCWDINFO:
1577 	case DIOCSDINFO:
1578 #ifdef __HAVE_OLD_DISKLABEL
1579 	case ODIOCWDINFO:
1580 	case ODIOCSDINFO:
1581 #endif
1582 	{
1583 		struct disklabel *lp;
1584 
1585 		if ((error = vndlock(vnd)) != 0)
1586 			return error;
1587 
1588 		vnd->sc_flags |= VNF_LABELLING;
1589 
1590 #ifdef __HAVE_OLD_DISKLABEL
1591 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
1592 			memset(&newlabel, 0, sizeof newlabel);
1593 			memcpy(&newlabel, data, sizeof (struct olddisklabel));
1594 			lp = &newlabel;
1595 		} else
1596 #endif
1597 		lp = (struct disklabel *)data;
1598 
1599 		error = setdisklabel(vnd->sc_dkdev.dk_label,
1600 		    lp, 0, vnd->sc_dkdev.dk_cpulabel);
1601 		if (error == 0) {
1602 			if (cmd == DIOCWDINFO
1603 #ifdef __HAVE_OLD_DISKLABEL
1604 			    || cmd == ODIOCWDINFO
1605 #endif
1606 			   )
1607 				error = writedisklabel(VNDLABELDEV(dev),
1608 				    vndstrategy, vnd->sc_dkdev.dk_label,
1609 				    vnd->sc_dkdev.dk_cpulabel);
1610 		}
1611 
1612 		vnd->sc_flags &= ~VNF_LABELLING;
1613 
1614 		vndunlock(vnd);
1615 
1616 		if (error)
1617 			return error;
1618 		break;
1619 	}
1620 
1621 	case DIOCKLABEL:
1622 		if (*(int *)data != 0)
1623 			vnd->sc_flags |= VNF_KLABEL;
1624 		else
1625 			vnd->sc_flags &= ~VNF_KLABEL;
1626 		break;
1627 
1628 	case DIOCWLABEL:
1629 		if (*(int *)data != 0)
1630 			vnd->sc_flags |= VNF_WLABEL;
1631 		else
1632 			vnd->sc_flags &= ~VNF_WLABEL;
1633 		break;
1634 
1635 	case DIOCGDEFLABEL:
1636 		vndgetdefaultlabel(vnd, (struct disklabel *)data);
1637 		break;
1638 
1639 #ifdef __HAVE_OLD_DISKLABEL
1640 	case ODIOCGDEFLABEL:
1641 		vndgetdefaultlabel(vnd, &newlabel);
1642 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1643 			return ENOTTY;
1644 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1645 		break;
1646 #endif
1647 
1648 	case DIOCGSTRATEGY:
1649 	    {
1650 		struct disk_strategy *dks = (void *)data;
1651 
1652 		/* No lock needed, never changed */
1653 		strlcpy(dks->dks_name,
1654 		    bufq_getstrategyname(vnd->sc_tab),
1655 		    sizeof(dks->dks_name));
1656 		dks->dks_paramlen = 0;
1657 		break;
1658 	    }
1659 	case DIOCGCACHE:
1660 	    {
1661 		int *bits = (int *)data;
1662 		*bits |= DKCACHE_READ | DKCACHE_WRITE;
1663 		break;
1664 	    }
1665 	case DIOCCACHESYNC:
1666 		vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1667 		error = VOP_FSYNC(vnd->sc_vp, vnd->sc_cred,
1668 		    FSYNC_WAIT | FSYNC_DATAONLY | FSYNC_CACHE, 0, 0);
1669 		VOP_UNLOCK(vnd->sc_vp);
1670 		return error;
1671 
1672 	default:
1673 		return ENOTTY;
1674 	}
1675 
1676 	return 0;
1677 }
1678 
1679 /*
1680  * Duplicate the current processes' credentials.  Since we are called only
1681  * as the result of a SET ioctl and only root can do that, any future access
1682  * to this "disk" is essentially as root.  Note that credentials may change
1683  * if some other uid can write directly to the mapped file (NFS).
1684  */
1685 static int
1686 vndsetcred(struct vnd_softc *vnd, kauth_cred_t cred)
1687 {
1688 	struct uio auio;
1689 	struct iovec aiov;
1690 	char *tmpbuf;
1691 	int error;
1692 
1693 	vnd->sc_cred = kauth_cred_dup(cred);
1694 	tmpbuf = malloc(DEV_BSIZE, M_TEMP, M_WAITOK);
1695 
1696 	/* XXX: Horrible kludge to establish credentials for NFS */
1697 	aiov.iov_base = tmpbuf;
1698 	aiov.iov_len = uimin(DEV_BSIZE, dbtob(vnd->sc_size));
1699 	auio.uio_iov = &aiov;
1700 	auio.uio_iovcnt = 1;
1701 	auio.uio_offset = 0;
1702 	auio.uio_rw = UIO_READ;
1703 	auio.uio_resid = aiov.iov_len;
1704 	UIO_SETUP_SYSSPACE(&auio);
1705 	vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1706 	error = VOP_READ(vnd->sc_vp, &auio, 0, vnd->sc_cred);
1707 	if (error == 0) {
1708 		/*
1709 		 * Because vnd does all IO directly through the vnode
1710 		 * we need to flush (at least) the buffer from the above
1711 		 * VOP_READ from the buffer cache to prevent cache
1712 		 * incoherencies.  Also, be careful to write dirty
1713 		 * buffers back to stable storage.
1714 		 */
1715 		error = vinvalbuf(vnd->sc_vp, V_SAVE, vnd->sc_cred,
1716 			    curlwp, 0, 0);
1717 	}
1718 	VOP_UNLOCK(vnd->sc_vp);
1719 
1720 	free(tmpbuf, M_TEMP);
1721 	return error;
1722 }
1723 
1724 /*
1725  * Set maxactive based on FS type
1726  */
1727 static void
1728 vndthrottle(struct vnd_softc *vnd, struct vnode *vp)
1729 {
1730 
1731 	if (vp->v_tag == VT_NFS)
1732 		vnd->sc_maxactive = 2;
1733 	else
1734 		vnd->sc_maxactive = 8;
1735 
1736 	if (vnd->sc_maxactive < 1)
1737 		vnd->sc_maxactive = 1;
1738 }
1739 
1740 #if 0
1741 static void
1742 vndshutdown(void)
1743 {
1744 	struct vnd_softc *vnd;
1745 
1746 	for (vnd = &vnd_softc[0]; vnd < &vnd_softc[numvnd]; vnd++)
1747 		if (vnd->sc_flags & VNF_INITED)
1748 			vndclear(vnd);
1749 }
1750 #endif
1751 
1752 static void
1753 vndclear(struct vnd_softc *vnd, int myminor)
1754 {
1755 	struct vnode *vp = vnd->sc_vp;
1756 	int fflags = FREAD;
1757 	int bmaj, cmaj, i, mn;
1758 	int s;
1759 
1760 #ifdef DEBUG
1761 	if (vnddebug & VDB_FOLLOW)
1762 		printf("vndclear(%p): vp %p\n", vnd, vp);
1763 #endif
1764 	/* locate the major number */
1765 	bmaj = bdevsw_lookup_major(&vnd_bdevsw);
1766 	cmaj = cdevsw_lookup_major(&vnd_cdevsw);
1767 
1768 	/* Nuke the vnodes for any open instances */
1769 	for (i = 0; i < MAXPARTITIONS; i++) {
1770 		mn = DISKMINOR(device_unit(vnd->sc_dev), i);
1771 		if (mn != myminor) { /* XXX avoid to kill own vnode */
1772 			vdevgone(bmaj, mn, mn, VBLK);
1773 			vdevgone(cmaj, mn, mn, VCHR);
1774 		}
1775 	}
1776 
1777 	if ((vnd->sc_flags & VNF_READONLY) == 0)
1778 		fflags |= FWRITE;
1779 
1780 	s = splbio();
1781 	bufq_drain(vnd->sc_tab);
1782 	splx(s);
1783 
1784 	vnd->sc_flags |= VNF_VUNCONF;
1785 	wakeup(&vnd->sc_tab);
1786 	while (vnd->sc_flags & VNF_KTHREAD)
1787 		tsleep(&vnd->sc_kthread, PRIBIO, "vnthr", 0);
1788 
1789 #ifdef VND_COMPRESSION
1790 	/* free the compressed file buffers */
1791 	if (vnd->sc_flags & VNF_COMP) {
1792 		if (vnd->sc_comp_offsets) {
1793 			free(vnd->sc_comp_offsets, M_DEVBUF);
1794 			vnd->sc_comp_offsets = NULL;
1795 		}
1796 		if (vnd->sc_comp_buff) {
1797 			free(vnd->sc_comp_buff, M_DEVBUF);
1798 			vnd->sc_comp_buff = NULL;
1799 		}
1800 		if (vnd->sc_comp_decombuf) {
1801 			free(vnd->sc_comp_decombuf, M_DEVBUF);
1802 			vnd->sc_comp_decombuf = NULL;
1803 		}
1804 	}
1805 #endif /* VND_COMPRESSION */
1806 	vnd->sc_flags &=
1807 	    ~(VNF_INITED | VNF_READONLY | VNF_KLABEL | VNF_VLABEL
1808 	      | VNF_VUNCONF | VNF_COMP | VNF_CLEARING);
1809 	if (vp == NULL)
1810 		panic("vndclear: null vp");
1811 	(void) vn_close(vp, fflags, vnd->sc_cred);
1812 	kauth_cred_free(vnd->sc_cred);
1813 	vnd->sc_vp = NULL;
1814 	vnd->sc_cred = NULL;
1815 	vnd->sc_size = 0;
1816 }
1817 
1818 static int
1819 vndsize(dev_t dev)
1820 {
1821 	struct vnd_softc *sc;
1822 	struct disklabel *lp;
1823 	int part, unit, omask;
1824 	int size;
1825 
1826 	unit = vndunit(dev);
1827 	sc = device_lookup_private(&vnd_cd, unit);
1828 	if (sc == NULL)
1829 		return -1;
1830 
1831 	if ((sc->sc_flags & VNF_INITED) == 0)
1832 		return -1;
1833 
1834 	part = DISKPART(dev);
1835 	omask = sc->sc_dkdev.dk_openmask & (1 << part);
1836 	lp = sc->sc_dkdev.dk_label;
1837 
1838 	if (omask == 0 && vndopen(dev, 0, S_IFBLK, curlwp))	/* XXX */
1839 		return -1;
1840 
1841 	if (lp->d_partitions[part].p_fstype != FS_SWAP)
1842 		size = -1;
1843 	else
1844 		size = lp->d_partitions[part].p_size *
1845 		    (lp->d_secsize / DEV_BSIZE);
1846 
1847 	if (omask == 0 && vndclose(dev, 0, S_IFBLK, curlwp))	/* XXX */
1848 		return -1;
1849 
1850 	return size;
1851 }
1852 
1853 static int
1854 vnddump(dev_t dev, daddr_t blkno, void *va,
1855     size_t size)
1856 {
1857 
1858 	/* Not implemented. */
1859 	return ENXIO;
1860 }
1861 
1862 static void
1863 vndgetdefaultlabel(struct vnd_softc *sc, struct disklabel *lp)
1864 {
1865 	struct vndgeom *vng = &sc->sc_geom;
1866 	struct partition *pp;
1867 	unsigned spb;
1868 
1869 	memset(lp, 0, sizeof(*lp));
1870 
1871 	spb = vng->vng_secsize / DEV_BSIZE;
1872 	if (sc->sc_size / spb > UINT32_MAX)
1873 		lp->d_secperunit = UINT32_MAX;
1874 	else
1875 		lp->d_secperunit = sc->sc_size / spb;
1876 	lp->d_secsize = vng->vng_secsize;
1877 	lp->d_nsectors = vng->vng_nsectors;
1878 	lp->d_ntracks = vng->vng_ntracks;
1879 	lp->d_ncylinders = vng->vng_ncylinders;
1880 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1881 
1882 	strncpy(lp->d_typename, "vnd", sizeof(lp->d_typename));
1883 	lp->d_type = DKTYPE_VND;
1884 	strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
1885 	lp->d_rpm = 3600;
1886 	lp->d_interleave = 1;
1887 	lp->d_flags = 0;
1888 
1889 	pp = &lp->d_partitions[RAW_PART];
1890 	pp->p_offset = 0;
1891 	pp->p_size = lp->d_secperunit;
1892 	pp->p_fstype = FS_UNUSED;
1893 	lp->d_npartitions = RAW_PART + 1;
1894 
1895 	lp->d_magic = DISKMAGIC;
1896 	lp->d_magic2 = DISKMAGIC;
1897 	lp->d_checksum = dkcksum(lp);
1898 }
1899 
1900 /*
1901  * Read the disklabel from a vnd.  If one is not present, create a fake one.
1902  */
1903 static void
1904 vndgetdisklabel(dev_t dev, struct vnd_softc *sc)
1905 {
1906 	const char *errstring;
1907 	struct disklabel *lp = sc->sc_dkdev.dk_label;
1908 	struct cpu_disklabel *clp = sc->sc_dkdev.dk_cpulabel;
1909 	int i;
1910 
1911 	memset(clp, 0, sizeof(*clp));
1912 
1913 	vndgetdefaultlabel(sc, lp);
1914 
1915 	/*
1916 	 * Call the generic disklabel extraction routine.
1917 	 */
1918 	errstring = readdisklabel(VNDLABELDEV(dev), vndstrategy, lp, clp);
1919 	if (errstring) {
1920 		/*
1921 		 * Lack of disklabel is common, but we print the warning
1922 		 * anyway, since it might contain other useful information.
1923 		 */
1924 		aprint_normal_dev(sc->sc_dev, "%s\n", errstring);
1925 
1926 		/*
1927 		 * For historical reasons, if there's no disklabel
1928 		 * present, all partitions must be FS_BSDFFS and
1929 		 * occupy the entire disk.
1930 		 */
1931 		for (i = 0; i < MAXPARTITIONS; i++) {
1932 			/*
1933 			 * Don't wipe out port specific hack (such as
1934 			 * dos partition hack of i386 port).
1935 			 */
1936 			if (lp->d_partitions[i].p_size != 0)
1937 				continue;
1938 
1939 			lp->d_partitions[i].p_size = lp->d_secperunit;
1940 			lp->d_partitions[i].p_offset = 0;
1941 			lp->d_partitions[i].p_fstype = FS_BSDFFS;
1942 		}
1943 
1944 		strncpy(lp->d_packname, "default label",
1945 		    sizeof(lp->d_packname));
1946 
1947 		lp->d_npartitions = MAXPARTITIONS;
1948 		lp->d_checksum = dkcksum(lp);
1949 	}
1950 }
1951 
1952 /*
1953  * Wait interruptibly for an exclusive lock.
1954  *
1955  * XXX
1956  * Several drivers do this; it should be abstracted and made MP-safe.
1957  */
1958 static int
1959 vndlock(struct vnd_softc *sc)
1960 {
1961 	int error;
1962 
1963 	while ((sc->sc_flags & VNF_LOCKED) != 0) {
1964 		sc->sc_flags |= VNF_WANTED;
1965 		if ((error = tsleep(sc, PRIBIO | PCATCH, "vndlck", 0)) != 0)
1966 			return error;
1967 	}
1968 	sc->sc_flags |= VNF_LOCKED;
1969 	return 0;
1970 }
1971 
1972 /*
1973  * Unlock and wake up any waiters.
1974  */
1975 static void
1976 vndunlock(struct vnd_softc *sc)
1977 {
1978 
1979 	sc->sc_flags &= ~VNF_LOCKED;
1980 	if ((sc->sc_flags & VNF_WANTED) != 0) {
1981 		sc->sc_flags &= ~VNF_WANTED;
1982 		wakeup(sc);
1983 	}
1984 }
1985 
1986 #ifdef VND_COMPRESSION
1987 /* compressed file read */
1988 static void
1989 compstrategy(struct buf *bp, off_t bn)
1990 {
1991 	int error;
1992 	int unit = vndunit(bp->b_dev);
1993 	struct vnd_softc *vnd =
1994 	    device_lookup_private(&vnd_cd, unit);
1995 	u_int32_t comp_block;
1996 	struct uio auio;
1997 	char *addr;
1998 	int s;
1999 
2000 	/* set up constants for data move */
2001 	auio.uio_rw = UIO_READ;
2002 	UIO_SETUP_SYSSPACE(&auio);
2003 
2004 	/* read, and transfer the data */
2005 	addr = bp->b_data;
2006 	bp->b_resid = bp->b_bcount;
2007 	s = splbio();
2008 	while (bp->b_resid > 0) {
2009 		unsigned length;
2010 		size_t length_in_buffer;
2011 		u_int32_t offset_in_buffer;
2012 		struct iovec aiov;
2013 
2014 		/* calculate the compressed block number */
2015 		comp_block = bn / (off_t)vnd->sc_comp_blksz;
2016 
2017 		/* check for good block number */
2018 		if (comp_block >= vnd->sc_comp_numoffs) {
2019 			bp->b_error = EINVAL;
2020 			splx(s);
2021 			return;
2022 		}
2023 
2024 		/* read in the compressed block, if not in buffer */
2025 		if (comp_block != vnd->sc_comp_buffblk) {
2026 			length = vnd->sc_comp_offsets[comp_block + 1] -
2027 			    vnd->sc_comp_offsets[comp_block];
2028 			vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
2029 			error = vn_rdwr(UIO_READ, vnd->sc_vp, vnd->sc_comp_buff,
2030 			    length, vnd->sc_comp_offsets[comp_block],
2031 			    UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, vnd->sc_cred,
2032 			    NULL, NULL);
2033 			if (error) {
2034 				bp->b_error = error;
2035 				VOP_UNLOCK(vnd->sc_vp);
2036 				splx(s);
2037 				return;
2038 			}
2039 			/* uncompress the buffer */
2040 			vnd->sc_comp_stream.next_in = vnd->sc_comp_buff;
2041 			vnd->sc_comp_stream.avail_in = length;
2042 			vnd->sc_comp_stream.next_out = vnd->sc_comp_decombuf;
2043 			vnd->sc_comp_stream.avail_out = vnd->sc_comp_blksz;
2044 			inflateReset(&vnd->sc_comp_stream);
2045 			error = inflate(&vnd->sc_comp_stream, Z_FINISH);
2046 			if (error != Z_STREAM_END) {
2047 				if (vnd->sc_comp_stream.msg)
2048 					aprint_normal_dev(vnd->sc_dev,
2049 					    "compressed file, %s\n",
2050 					    vnd->sc_comp_stream.msg);
2051 				bp->b_error = EBADMSG;
2052 				VOP_UNLOCK(vnd->sc_vp);
2053 				splx(s);
2054 				return;
2055 			}
2056 			vnd->sc_comp_buffblk = comp_block;
2057 			VOP_UNLOCK(vnd->sc_vp);
2058 		}
2059 
2060 		/* transfer the usable uncompressed data */
2061 		offset_in_buffer = bn % (off_t)vnd->sc_comp_blksz;
2062 		length_in_buffer = vnd->sc_comp_blksz - offset_in_buffer;
2063 		if (length_in_buffer > bp->b_resid)
2064 			length_in_buffer = bp->b_resid;
2065 		auio.uio_iov = &aiov;
2066 		auio.uio_iovcnt = 1;
2067 		aiov.iov_base = addr;
2068 		aiov.iov_len = length_in_buffer;
2069 		auio.uio_resid = aiov.iov_len;
2070 		auio.uio_offset = 0;
2071 		error = uiomove(vnd->sc_comp_decombuf + offset_in_buffer,
2072 		    length_in_buffer, &auio);
2073 		if (error) {
2074 			bp->b_error = error;
2075 			splx(s);
2076 			return;
2077 		}
2078 
2079 		bn += length_in_buffer;
2080 		addr += length_in_buffer;
2081 		bp->b_resid -= length_in_buffer;
2082 	}
2083 	splx(s);
2084 }
2085 
2086 /* compression memory allocation routines */
2087 static void *
2088 vnd_alloc(void *aux, u_int items, u_int siz)
2089 {
2090 	return malloc(items * siz, M_TEMP, M_NOWAIT);
2091 }
2092 
2093 static void
2094 vnd_free(void *aux, void *ptr)
2095 {
2096 	free(ptr, M_TEMP);
2097 }
2098 #endif /* VND_COMPRESSION */
2099 
2100 static void
2101 vnd_set_geometry(struct vnd_softc *vnd)
2102 {
2103 	struct disk_geom *dg = &vnd->sc_dkdev.dk_geom;
2104 	unsigned spb;
2105 
2106 	memset(dg, 0, sizeof(*dg));
2107 
2108 	spb = vnd->sc_geom.vng_secsize / DEV_BSIZE;
2109 	dg->dg_secperunit = vnd->sc_size / spb;
2110 	dg->dg_secsize = vnd->sc_geom.vng_secsize;
2111 	dg->dg_nsectors = vnd->sc_geom.vng_nsectors;
2112 	dg->dg_ntracks = vnd->sc_geom.vng_ntracks;
2113 	dg->dg_ncylinders = vnd->sc_geom.vng_ncylinders;
2114 
2115 #ifdef DEBUG
2116 	if (vnddebug & VDB_LABEL) {
2117 		printf("dg->dg_secperunit: %" PRId64 "\n", dg->dg_secperunit);
2118 		printf("dg->dg_ncylinders: %u\n", dg->dg_ncylinders);
2119 	}
2120 #endif
2121 	disk_set_info(vnd->sc_dev, &vnd->sc_dkdev, NULL);
2122 }
2123 
2124 #ifdef VND_COMPRESSION
2125 #define VND_DEPENDS "zlib"
2126 #else
2127 #define VND_DEPENDS NULL
2128 #endif
2129 
2130 MODULE(MODULE_CLASS_DRIVER, vnd, VND_DEPENDS);
2131 
2132 #ifdef _MODULE
2133 int vnd_bmajor = -1, vnd_cmajor = -1;
2134 
2135 CFDRIVER_DECL(vnd, DV_DISK, NULL);
2136 #endif
2137 
2138 static int
2139 vnd_modcmd(modcmd_t cmd, void *arg)
2140 {
2141 	int error = 0;
2142 
2143 	switch (cmd) {
2144 	case MODULE_CMD_INIT:
2145 #ifdef _MODULE
2146                 /*
2147                  * Attach the {b,c}devsw's
2148                  */
2149 		error = devsw_attach("vnd", &vnd_bdevsw, &vnd_bmajor,
2150 		    &vnd_cdevsw, &vnd_cmajor);
2151 		if (error) {
2152 #ifdef DIAGNOSTIC
2153                         aprint_error("%s: unable to attach %s devsw, "
2154                             "error %d", __func__, vnd_cd.cd_name, error);
2155 #endif
2156 			break;
2157 		}
2158 
2159 		error = config_cfdriver_attach(&vnd_cd);
2160 		if (error) {
2161 			devsw_detach(&vnd_bdevsw, &vnd_cdevsw);
2162 			break;
2163 		}
2164 
2165 		error = config_cfattach_attach(vnd_cd.cd_name, &vnd_ca);
2166 	        if (error) {
2167 			config_cfdriver_detach(&vnd_cd);
2168 			devsw_detach(&vnd_bdevsw, &vnd_cdevsw);
2169 #ifdef DIAGNOSTIC
2170 			aprint_error("%s: unable to register cfattach for \n"
2171 			    "%s, error %d", __func__, vnd_cd.cd_name, error);
2172 #endif
2173 			break;
2174 		}
2175 #endif
2176 		break;
2177 
2178 	case MODULE_CMD_FINI:
2179 #ifdef _MODULE
2180                 /*
2181                  * Remove device from autoconf database
2182                  */
2183 		error = config_cfattach_detach(vnd_cd.cd_name, &vnd_ca);
2184                 if (error) {
2185 #ifdef DIAGNOSTIC
2186                         aprint_error("%s: failed to detach %s cfattach, "
2187                             "error %d\n", __func__, vnd_cd.cd_name, error);
2188 #endif
2189                         break;
2190                 }
2191                 error = config_cfdriver_detach(&vnd_cd);
2192                 if (error) {
2193                         (void)config_cfattach_attach(vnd_cd.cd_name, &vnd_ca);
2194 #ifdef DIAGNOSTIC
2195                         aprint_error("%s: failed to detach %s cfdriver, "
2196                             "error %d\n", __func__, vnd_cd.cd_name, error);
2197                         break;
2198 #endif
2199                 }
2200                 /*
2201                  * Remove {b,c}devsw's
2202                  */
2203 		devsw_detach(&vnd_bdevsw, &vnd_cdevsw);
2204 
2205 #endif
2206 		break;
2207 
2208 	case MODULE_CMD_STAT:
2209 		return ENOTTY;
2210 
2211 	default:
2212 		return ENOTTY;
2213 	}
2214 
2215 	return error;
2216 }
2217