xref: /netbsd-src/sys/dev/vnd.c (revision f62466b674e882a8af7dc06836281f43462e8111)
1 /*	$NetBSD: vnd.c,v 1.290 2024/08/15 21:08:20 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 1.290 2024/08/15 21:08:20 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 <uvm/uvm.h>
127 
128 #include <net/zlib.h>
129 
130 #include <miscfs/genfs/genfs.h>
131 #include <miscfs/specfs/specdev.h>
132 
133 #include <dev/dkvar.h>
134 #include <dev/vndvar.h>
135 
136 #include "ioconf.h"
137 
138 #if defined(VNDDEBUG) && !defined(DEBUG)
139 #define DEBUG
140 #endif
141 
142 #ifdef DEBUG
143 int dovndcluster = 1;
144 #define VDB_FOLLOW	0x01
145 #define VDB_INIT	0x02
146 #define VDB_IO		0x04
147 #define VDB_LABEL	0x08
148 int vnddebug = 0;
149 #endif
150 
151 #define vndunit(x)	DISKUNIT(x)
152 
153 struct vndxfer {
154 	struct buf vx_buf;
155 	struct vnd_softc *vx_vnd;
156 };
157 #define	VND_BUFTOXFER(bp)	((struct vndxfer *)(void *)bp)
158 
159 #define VND_GETXFER(vnd)	pool_get(&(vnd)->sc_vxpool, PR_WAITOK)
160 #define VND_PUTXFER(vnd, vx)	pool_put(&(vnd)->sc_vxpool, (vx))
161 
162 #define VNDLABELDEV(dev) \
163     (MAKEDISKDEV(major((dev)), vndunit((dev)), RAW_PART))
164 
165 #define	VND_MAXPENDING(vnd)	((vnd)->sc_maxactive * 4)
166 #define	VND_MAXPAGES(vnd)	(1024 * 1024 / PAGE_SIZE)
167 
168 
169 static void	vndclear(struct vnd_softc *, int);
170 static int	vnddoclear(struct vnd_softc *, int, int, bool);
171 static int	vndsetcred(struct vnd_softc *, kauth_cred_t);
172 static void	vndthrottle(struct vnd_softc *, struct vnode *);
173 static void	vndiodone(struct buf *);
174 #if 0
175 static void	vndshutdown(void);
176 #endif
177 
178 static void	vndgetdefaultlabel(struct vnd_softc *, struct disklabel *);
179 static void	vndgetdisklabel(dev_t, struct vnd_softc *);
180 
181 static int	vndlock(struct vnd_softc *);
182 static void	vndunlock(struct vnd_softc *);
183 #ifdef VND_COMPRESSION
184 static void	compstrategy(struct buf *, off_t);
185 static void	*vnd_alloc(void *, u_int, u_int);
186 static void	vnd_free(void *, void *);
187 #endif /* VND_COMPRESSION */
188 
189 static void	vndthread(void *);
190 static bool	vnode_has_op(const struct vnode *, int);
191 static void	handle_with_rdwr(struct vnd_softc *, const struct buf *,
192 		    struct buf *);
193 static void	handle_with_strategy(struct vnd_softc *, const struct buf *,
194 		    struct buf *);
195 static void	vnd_set_geometry(struct vnd_softc *);
196 
197 static dev_type_open(vndopen);
198 static dev_type_close(vndclose);
199 static dev_type_read(vndread);
200 static dev_type_write(vndwrite);
201 static dev_type_ioctl(vndioctl);
202 static dev_type_strategy(vndstrategy);
203 static dev_type_dump(vnddump);
204 static dev_type_size(vndsize);
205 
206 const struct bdevsw vnd_bdevsw = {
207 	.d_open = vndopen,
208 	.d_close = vndclose,
209 	.d_strategy = vndstrategy,
210 	.d_ioctl = vndioctl,
211 	.d_dump = vnddump,
212 	.d_psize = vndsize,
213 	.d_discard = nodiscard,
214 	.d_flag = D_DISK
215 };
216 
217 const struct cdevsw vnd_cdevsw = {
218 	.d_open = vndopen,
219 	.d_close = vndclose,
220 	.d_read = vndread,
221 	.d_write = vndwrite,
222 	.d_ioctl = vndioctl,
223 	.d_stop = nostop,
224 	.d_tty = notty,
225 	.d_poll = nopoll,
226 	.d_mmap = nommap,
227 	.d_kqfilter = nokqfilter,
228 	.d_discard = nodiscard,
229 	.d_flag = D_DISK
230 };
231 
232 static int	vnd_match(device_t, cfdata_t, void *);
233 static void	vnd_attach(device_t, device_t, void *);
234 static int	vnd_detach(device_t, int);
235 
236 CFATTACH_DECL3_NEW(vnd, sizeof(struct vnd_softc),
237     vnd_match, vnd_attach, vnd_detach, NULL, NULL, NULL, DVF_DETACH_SHUTDOWN);
238 
239 static struct vnd_softc	*vnd_spawn(int);
240 static int	vnd_destroy(device_t);
241 
242 static const struct	dkdriver vnddkdriver = {
243 	.d_strategy = vndstrategy,
244 	.d_minphys = minphys
245 };
246 
247 void
248 vndattach(int num)
249 {
250 	int error;
251 
252 	error = config_cfattach_attach(vnd_cd.cd_name, &vnd_ca);
253 	if (error)
254 		aprint_error("%s: unable to register cfattach, error = %d\n",
255 		    vnd_cd.cd_name, error);
256 }
257 
258 static int
259 vnd_match(device_t self, cfdata_t cfdata, void *aux)
260 {
261 
262 	return 1;
263 }
264 
265 static void
266 vnd_attach(device_t parent, device_t self, void *aux)
267 {
268 	struct vnd_softc *sc = device_private(self);
269 
270 	sc->sc_dev = self;
271 	sc->sc_comp_offsets = NULL;
272 	sc->sc_comp_buff = NULL;
273 	sc->sc_comp_decombuf = NULL;
274 	bufq_alloc(&sc->sc_tab, "disksort", BUFQ_SORT_RAWBLOCK);
275 	disk_init(&sc->sc_dkdev, device_xname(self), &vnddkdriver);
276 	if (!pmf_device_register(self, NULL, NULL))
277 		aprint_error_dev(self, "couldn't establish power handler\n");
278 }
279 
280 static int
281 vnd_detach(device_t self, int flags)
282 {
283 	int error;
284 	struct vnd_softc *sc = device_private(self);
285 
286 	if (sc->sc_flags & VNF_INITED) {
287 		error = vnddoclear(sc, 0, -1, (flags & DETACH_FORCE) != 0);
288 		if (error != 0)
289 			return error;
290 	}
291 
292 	pmf_device_deregister(self);
293 	bufq_free(sc->sc_tab);
294 	disk_destroy(&sc->sc_dkdev);
295 
296 	return 0;
297 }
298 
299 static struct vnd_softc *
300 vnd_spawn(int unit)
301 {
302 	cfdata_t cf;
303 
304 	cf = malloc(sizeof(*cf), M_DEVBUF, M_WAITOK);
305 	cf->cf_name = vnd_cd.cd_name;
306 	cf->cf_atname = vnd_cd.cd_name;
307 	cf->cf_unit = unit;
308 	cf->cf_fstate = FSTATE_STAR;
309 
310 	return device_private(config_attach_pseudo(cf));
311 }
312 
313 static int
314 vnd_destroy(device_t dev)
315 {
316 	int error;
317 	cfdata_t cf;
318 
319 	cf = device_cfdata(dev);
320 	error = config_detach(dev, DETACH_QUIET);
321 	if (error)
322 		return error;
323 	free(cf, M_DEVBUF);
324 	return 0;
325 }
326 
327 static int
328 vndopen(dev_t dev, int flags, int mode, struct lwp *l)
329 {
330 	int unit = vndunit(dev);
331 	struct vnd_softc *sc;
332 	int error = 0, part, pmask;
333 	struct disklabel *lp;
334 
335 #ifdef DEBUG
336 	if (vnddebug & VDB_FOLLOW)
337 		printf("vndopen(0x%"PRIx64", 0x%x, 0x%x, %p)\n", dev, flags, mode, l);
338 #endif
339 	sc = device_lookup_private(&vnd_cd, unit);
340 	if (sc == NULL) {
341 		sc = vnd_spawn(unit);
342 		if (sc == NULL)
343 			return ENOMEM;
344 
345 		/* compatibility, keep disklabel after close */
346 		sc->sc_flags = VNF_KLABEL;
347 	}
348 
349 	if ((error = vndlock(sc)) != 0)
350 		return error;
351 
352 	mutex_enter(&sc->sc_dkdev.dk_openlock);
353 
354 	if ((sc->sc_flags & VNF_CLEARING) != 0) {
355 		error = ENXIO;
356 		goto done;
357 	}
358 
359 	lp = sc->sc_dkdev.dk_label;
360 
361 	part = DISKPART(dev);
362 	pmask = (1 << part);
363 
364 	if (sc->sc_dkdev.dk_nwedges != 0 && part != RAW_PART) {
365 		error = EBUSY;
366 		goto done;
367 	}
368 
369 	if (sc->sc_flags & VNF_INITED) {
370 		if ((sc->sc_dkdev.dk_openmask & ~(1<<RAW_PART)) != 0) {
371 			/*
372 			 * If any non-raw partition is open, but the disk
373 			 * has been invalidated, disallow further opens.
374 			 */
375 			if ((sc->sc_flags & VNF_VLABEL) == 0) {
376 				error = EIO;
377 				goto done;
378 			}
379 		} else {
380 			/*
381 			 * Load the partition info if not already loaded.
382 			 */
383 			if ((sc->sc_flags & VNF_VLABEL) == 0) {
384 				sc->sc_flags |= VNF_VLABEL;
385 				vndgetdisklabel(dev, sc);
386 			}
387 		}
388 	}
389 
390 	/* Check that the partitions exists. */
391 	if (part != RAW_PART) {
392 		if (((sc->sc_flags & VNF_INITED) == 0) ||
393 		    ((part >= lp->d_npartitions) ||
394 		     (lp->d_partitions[part].p_fstype == FS_UNUSED))) {
395 			error = ENXIO;
396 			goto done;
397 		}
398 	}
399 
400 	/* Prevent our unit from being unconfigured while open. */
401 	switch (mode) {
402 	case S_IFCHR:
403 		sc->sc_dkdev.dk_copenmask |= pmask;
404 		break;
405 
406 	case S_IFBLK:
407 		sc->sc_dkdev.dk_bopenmask |= pmask;
408 		break;
409 	}
410 	sc->sc_dkdev.dk_openmask =
411 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
412 
413  done:
414 	mutex_exit(&sc->sc_dkdev.dk_openlock);
415 	vndunlock(sc);
416 	return error;
417 }
418 
419 static int
420 vndclose(dev_t dev, int flags, int mode, struct lwp *l)
421 {
422 	int unit = vndunit(dev);
423 	struct vnd_softc *sc;
424 	int error = 0, part;
425 
426 #ifdef DEBUG
427 	if (vnddebug & VDB_FOLLOW)
428 		printf("vndclose(0x%"PRIx64", 0x%x, 0x%x, %p)\n", dev, flags, mode, l);
429 #endif
430 	sc = device_lookup_private(&vnd_cd, unit);
431 	if (sc == NULL)
432 		return ENXIO;
433 
434 	if ((error = vndlock(sc)) != 0)
435 		return error;
436 
437 	mutex_enter(&sc->sc_dkdev.dk_openlock);
438 
439 	part = DISKPART(dev);
440 
441 	/* ...that much closer to allowing unconfiguration... */
442 	switch (mode) {
443 	case S_IFCHR:
444 		sc->sc_dkdev.dk_copenmask &= ~(1 << part);
445 		break;
446 
447 	case S_IFBLK:
448 		sc->sc_dkdev.dk_bopenmask &= ~(1 << part);
449 		break;
450 	}
451 	sc->sc_dkdev.dk_openmask =
452 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
453 
454 	/* are we last opener ? */
455 	if (sc->sc_dkdev.dk_openmask == 0) {
456 		if ((sc->sc_flags & VNF_KLABEL) == 0)
457 			sc->sc_flags &= ~VNF_VLABEL;
458 	}
459 
460 	mutex_exit(&sc->sc_dkdev.dk_openlock);
461 
462 	vndunlock(sc);
463 
464 	if ((sc->sc_flags & VNF_INITED) == 0) {
465 		if ((error = vnd_destroy(sc->sc_dev)) != 0) {
466 			aprint_error_dev(sc->sc_dev,
467 			    "unable to detach instance\n");
468 			return error;
469 		}
470 	}
471 
472 	return 0;
473 }
474 
475 /*
476  * Queue the request, and wakeup the kernel thread to handle it.
477  */
478 static void
479 vndstrategy(struct buf *bp)
480 {
481 	int unit = vndunit(bp->b_dev);
482 	struct vnd_softc *vnd =
483 	    device_lookup_private(&vnd_cd, unit);
484 	struct disklabel *lp;
485 	daddr_t blkno;
486 	int s = splbio();
487 
488 	if (vnd == NULL) {
489 		bp->b_error = ENXIO;
490 		goto done;
491 	}
492 	lp = vnd->sc_dkdev.dk_label;
493 
494 	if ((vnd->sc_flags & VNF_INITED) == 0) {
495 		bp->b_error = ENXIO;
496 		goto done;
497 	}
498 
499 	/*
500 	 * The transfer must be a whole number of blocks.
501 	 */
502 	if ((bp->b_bcount % lp->d_secsize) != 0) {
503 		bp->b_error = EINVAL;
504 		goto done;
505 	}
506 
507 	/*
508 	 * check if we're read-only.
509 	 */
510 	if ((vnd->sc_flags & VNF_READONLY) && !(bp->b_flags & B_READ)) {
511 		bp->b_error = EACCES;
512 		goto done;
513 	}
514 
515 	/* If it's a nil transfer, wake up the top half now. */
516 	if (bp->b_bcount == 0) {
517 		goto done;
518 	}
519 
520 	/*
521 	 * Do bounds checking and adjust transfer.  If there's an error,
522 	 * the bounds check will flag that for us.
523 	 */
524 	if (DISKPART(bp->b_dev) == RAW_PART) {
525 		if (bounds_check_with_mediasize(bp, DEV_BSIZE,
526 		    vnd->sc_size) <= 0)
527 			goto done;
528 	} else {
529 		if (bounds_check_with_label(&vnd->sc_dkdev,
530 		    bp, vnd->sc_flags & (VNF_WLABEL|VNF_LABELLING)) <= 0)
531 			goto done;
532 	}
533 
534 	/*
535 	 * Put the block number in terms of the logical blocksize
536 	 * of the "device".
537 	 */
538 
539 	blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
540 
541 	/*
542 	 * Translate the partition-relative block number to an absolute.
543 	 */
544 	if (DISKPART(bp->b_dev) != RAW_PART) {
545 		struct partition *pp;
546 
547 		pp = &vnd->sc_dkdev.dk_label->d_partitions[
548 		    DISKPART(bp->b_dev)];
549 		blkno += pp->p_offset;
550 	}
551 	bp->b_rawblkno = blkno;
552 
553 #ifdef DEBUG
554 	if (vnddebug & VDB_FOLLOW)
555 		printf("vndstrategy(%p): unit %d\n", bp, unit);
556 #endif
557 	if ((vnd->sc_flags & VNF_USE_VN_RDWR)) {
558 		/*
559 		 * Limit the number of pending requests to not exhaust
560 		 * resources needed for I/O but always allow the worker
561 		 * thread to add requests, as a wedge on vnd queues
562 		 * requests with biodone() -> dkstart() -> vndstrategy().
563 		 */
564 		if (curlwp != vnd->sc_kthread && curlwp != uvm.pagedaemon_lwp) {
565 			while (vnd->sc_pending >= VND_MAXPENDING(vnd))
566 				tsleep(&vnd->sc_pending, PRIBIO, "vndpc", 0);
567 		}
568 		vnd->sc_pending++;
569 		KASSERT(vnd->sc_pending > 0);
570 	}
571 	bufq_put(vnd->sc_tab, bp);
572 	wakeup(&vnd->sc_tab);
573 	splx(s);
574 	return;
575 
576 done:
577 	bp->b_resid = bp->b_bcount;
578 	biodone(bp);
579 	splx(s);
580 }
581 
582 static bool
583 vnode_has_strategy(struct vnd_softc *vnd)
584 {
585 	return vnode_has_op(vnd->sc_vp, VOFFSET(vop_bmap)) &&
586 	    vnode_has_op(vnd->sc_vp, VOFFSET(vop_strategy));
587 }
588 
589 /* Verify that I/O requests cannot be smaller than the
590  * smallest I/O size supported by the backend.
591  */
592 static bool
593 vnode_has_large_blocks(struct vnd_softc *vnd)
594 {
595 	u_int32_t vnd_secsize, iosize;
596 
597 	iosize = vnd->sc_iosize;
598 	vnd_secsize = vnd->sc_geom.vng_secsize;
599 
600 	return vnd_secsize % iosize != 0;
601 }
602 
603 /* XXX this function needs a reliable check to detect
604  * sparse files. Otherwise, bmap/strategy may be used
605  * and fail on non-allocated blocks. VOP_READ/VOP_WRITE
606  * works on sparse files.
607  */
608 #if notyet
609 static bool
610 vnode_strategy_probe(struct vnd_softc *vnd)
611 {
612 	int error;
613 	daddr_t nbn;
614 
615 	if (!vnode_has_strategy(vnd))
616 		return false;
617 
618 	if (vnode_has_large_blocks(vnd))
619 		return false;
620 
621 	/* Convert the first logical block number to its
622 	 * physical block number.
623 	 */
624 	error = 0;
625 	vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
626 	error = VOP_BMAP(vnd->sc_vp, 0, NULL, &nbn, NULL);
627 	VOP_UNLOCK(vnd->sc_vp);
628 
629 	/* Test if that worked. */
630 	if (error == 0 && (long)nbn == -1)
631 		return false;
632 
633 	return true;
634 }
635 #endif
636 
637 static void
638 vndthread(void *arg)
639 {
640 	struct vnd_softc *vnd = arg;
641 	int s;
642 
643 	/* Determine whether we can *use* VOP_BMAP and VOP_STRATEGY to
644 	 * directly access the backing vnode.  If we can, use these two
645 	 * operations to avoid messing with the local buffer cache.
646 	 * Otherwise fall back to regular VOP_READ/VOP_WRITE operations
647 	 * which are guaranteed to work with any file system. */
648 	if ((vnd->sc_flags & VNF_USE_VN_RDWR) == 0 &&
649 	    ! vnode_has_strategy(vnd))
650 		vnd->sc_flags |= VNF_USE_VN_RDWR;
651 
652 	/* VOP_STRATEGY can only be used if the backing vnode allows
653 	 * to access blocks as small as defined by the vnd geometry.
654 	 */
655 	if ((vnd->sc_flags & VNF_USE_VN_RDWR) == 0 &&
656 	    vnode_has_large_blocks(vnd))
657 		vnd->sc_flags |= VNF_USE_VN_RDWR;
658 
659 #ifdef DEBUG
660 	if (vnddebug & VDB_INIT)
661 		printf("vndthread: vp %p, %s\n", vnd->sc_vp,
662 		    (vnd->sc_flags & VNF_USE_VN_RDWR) == 0 ?
663 		    "using bmap/strategy operations" :
664 		    "using read/write operations");
665 #endif
666 
667 	s = splbio();
668 	vnd->sc_flags |= VNF_KTHREAD;
669 	wakeup(&vnd->sc_kthread);
670 
671 	/*
672 	 * Dequeue requests and serve them depending on the available
673 	 * vnode operations.
674 	 */
675 	while ((vnd->sc_flags & VNF_VUNCONF) == 0) {
676 		struct vndxfer *vnx;
677 		struct buf *obp;
678 		struct buf *bp;
679 
680 		obp = bufq_get(vnd->sc_tab);
681 		if (obp == NULL) {
682 			tsleep(&vnd->sc_tab, PRIBIO, "vndbp", 0);
683 			continue;
684 		};
685 		if ((vnd->sc_flags & VNF_USE_VN_RDWR)) {
686 			KASSERT(vnd->sc_pending > 0);
687 			if (vnd->sc_pending-- == VND_MAXPENDING(vnd))
688 				wakeup(&vnd->sc_pending);
689 		}
690 		splx(s);
691 #ifdef DEBUG
692 		if (vnddebug & VDB_FOLLOW)
693 			printf("vndthread(%p)\n", obp);
694 #endif
695 
696 		if (vnd->sc_vp->v_mount == NULL) {
697 			obp->b_error = ENXIO;
698 			goto done;
699 		}
700 #ifdef VND_COMPRESSION
701 		/* handle a compressed read */
702 		if ((obp->b_flags & B_READ) != 0 && (vnd->sc_flags & VNF_COMP)) {
703 			off_t bn;
704 
705 			/* Convert to a byte offset within the file. */
706 			bn = obp->b_rawblkno *
707 			    vnd->sc_dkdev.dk_label->d_secsize;
708 
709 			compstrategy(obp, bn);
710 			goto done;
711 		}
712 #endif /* VND_COMPRESSION */
713 
714 		/*
715 		 * Allocate a header for this transfer and link it to the
716 		 * buffer
717 		 */
718 		s = splbio();
719 		vnx = VND_GETXFER(vnd);
720 		splx(s);
721 		vnx->vx_vnd = vnd;
722 
723 		s = splbio();
724 		while (vnd->sc_active >= vnd->sc_maxactive) {
725 			tsleep(&vnd->sc_tab, PRIBIO, "vndac", 0);
726 		}
727 		vnd->sc_active++;
728 		splx(s);
729 
730 		/* Instrumentation. */
731 		disk_busy(&vnd->sc_dkdev);
732 
733 		bp = &vnx->vx_buf;
734 		buf_init(bp);
735 		bp->b_flags = (obp->b_flags & (B_READ | B_PHYS | B_RAW));
736 		bp->b_oflags = obp->b_oflags;
737 		bp->b_cflags = obp->b_cflags;
738 		bp->b_iodone = vndiodone;
739 		bp->b_private = obp;
740 		bp->b_vp = vnd->sc_vp;
741 		bp->b_objlock = bp->b_vp->v_interlock;
742 		bp->b_data = obp->b_data;
743 		bp->b_bcount = obp->b_bcount;
744 		BIO_COPYPRIO(bp, obp);
745 
746 		/* Make sure the request succeeds while suspending this fs. */
747 		fstrans_start_lazy(vnd->sc_vp->v_mount);
748 
749 		/* Handle the request using the appropriate operations. */
750 		if ((vnd->sc_flags & VNF_USE_VN_RDWR) == 0)
751 			handle_with_strategy(vnd, obp, bp);
752 		else
753 			handle_with_rdwr(vnd, obp, bp);
754 
755 		fstrans_done(vnd->sc_vp->v_mount);
756 
757 		s = splbio();
758 		continue;
759 
760 done:
761 		biodone(obp);
762 		s = splbio();
763 	}
764 
765 	vnd->sc_flags &= (~VNF_KTHREAD | VNF_VUNCONF);
766 	wakeup(&vnd->sc_kthread);
767 	splx(s);
768 	kthread_exit(0);
769 }
770 
771 /*
772  * Checks if the given vnode supports the requested operation.
773  * The operation is specified the offset returned by VOFFSET.
774  *
775  * XXX The test below used to determine this is quite fragile
776  * because it relies on the file system to use genfs to specify
777  * unimplemented operations.  There might be another way to do
778  * it more cleanly.
779  */
780 static bool
781 vnode_has_op(const struct vnode *vp, int opoffset)
782 {
783 	int (*defaultp)(void *);
784 	int (*opp)(void *);
785 
786 	defaultp = vp->v_op[VOFFSET(vop_default)];
787 	opp = vp->v_op[opoffset];
788 
789 	return opp != defaultp && opp != genfs_eopnotsupp &&
790 	    opp != genfs_badop && opp != genfs_nullop;
791 }
792 
793 /*
794  * Handles the read/write request given in 'bp' using the vnode's VOP_READ
795  * and VOP_WRITE operations.
796  *
797  * 'obp' is a pointer to the original request fed to the vnd device.
798  */
799 static void
800 handle_with_rdwr(struct vnd_softc *vnd, const struct buf *obp, struct buf *bp)
801 {
802 	bool doread;
803 	off_t offset;
804 	size_t len, resid;
805 	struct vnode *vp;
806 	int npages;
807 
808 	doread = bp->b_flags & B_READ;
809 	offset = obp->b_rawblkno * vnd->sc_dkdev.dk_label->d_secsize;
810 	len = bp->b_bcount;
811 	vp = vnd->sc_vp;
812 
813 #if defined(DEBUG)
814 	if (vnddebug & VDB_IO)
815 		printf("vnd (rdwr): vp %p, %s, rawblkno 0x%" PRIx64
816 		    ", secsize %d, offset %" PRIu64
817 		    ", bcount %d\n",
818 		    vp, doread ? "read" : "write", obp->b_rawblkno,
819 		    vnd->sc_dkdev.dk_label->d_secsize, offset,
820 		    bp->b_bcount);
821 #endif
822 
823 	/* Issue the read or write operation. */
824 	bp->b_error =
825 	    vn_rdwr(doread ? UIO_READ : UIO_WRITE,
826 	    vp, bp->b_data, len, offset, UIO_SYSSPACE,
827 	    IO_ADV_ENCODE(POSIX_FADV_NOREUSE) | IO_DIRECT,
828 		vnd->sc_cred, &resid, NULL);
829 	bp->b_resid = resid;
830 
831 	/*
832 	 * Avoid caching too many pages, the vnd user
833 	 * is usually a filesystem and caches itself.
834 	 * We need some amount of caching to not hinder
835 	 * read-ahead and write-behind operations.
836 	 */
837 	npages = atomic_load_relaxed(&vp->v_uobj.uo_npages);
838 	if (npages > VND_MAXPAGES(vnd)) {
839 		rw_enter(vp->v_uobj.vmobjlock, RW_WRITER);
840 		(void) VOP_PUTPAGES(vp, 0, 0,
841 		    PGO_ALLPAGES | PGO_CLEANIT | PGO_FREE);
842 	}
843 
844 	/* We need to increase the number of outputs on the vnode if
845 	 * there was any write to it. */
846 	if (!doread) {
847 		mutex_enter(vp->v_interlock);
848 		vp->v_numoutput++;
849 		mutex_exit(vp->v_interlock);
850 	}
851 
852 	biodone(bp);
853 }
854 
855 /*
856  * Handes the read/write request given in 'bp' using the vnode's VOP_BMAP
857  * and VOP_STRATEGY operations.
858  *
859  * 'obp' is a pointer to the original request fed to the vnd device.
860  */
861 static void
862 handle_with_strategy(struct vnd_softc *vnd, const struct buf *obp,
863     struct buf *bp)
864 {
865 	int bsize, error, flags, skipped;
866 	size_t resid, sz;
867 	off_t bn, offset;
868 	struct vnode *vp;
869 	struct buf *nbp = NULL;
870 
871 	flags = obp->b_flags;
872 
873 
874 	/* convert to a byte offset within the file. */
875 	bn = obp->b_rawblkno * vnd->sc_dkdev.dk_label->d_secsize;
876 
877 	bsize = vnd->sc_vp->v_mount->mnt_stat.f_iosize;
878 	/* use default if the filesystem didn't specify a block size */
879 	if (bsize <= 0)
880 		bsize = BLKDEV_IOSIZE;
881 	skipped = 0;
882 
883 	/*
884 	 * Break the request into bsize pieces and feed them
885 	 * sequentially using VOP_BMAP/VOP_STRATEGY.
886 	 * We do it this way to keep from flooding NFS servers if we
887 	 * are connected to an NFS file.  This places the burden on
888 	 * the client rather than the server.
889 	 */
890 	error = 0;
891 	bp->b_resid = bp->b_bcount;
892 	for (offset = 0, resid = bp->b_resid; /* true */;
893 	    resid -= sz, offset += sz) {
894 		daddr_t nbn;
895 		int off, nra;
896 
897 		nra = 0;
898 		vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
899 		error = VOP_BMAP(vnd->sc_vp, bn / bsize, &vp, &nbn, &nra);
900 		VOP_UNLOCK(vnd->sc_vp);
901 
902 		if (error == 0 && (long)nbn == -1)
903 			error = EIO;
904 
905 		/*
906 		 * If there was an error or a hole in the file...punt.
907 		 * Note that we may have to wait for any operations
908 		 * that we have already fired off before releasing
909 		 * the buffer.
910 		 *
911 		 * XXX we could deal with holes here but it would be
912 		 * a hassle (in the write case).
913 		 */
914 		if (error) {
915 			skipped += resid;
916 			break;
917 		}
918 
919 #ifdef DEBUG
920 		if (!dovndcluster)
921 			nra = 0;
922 #endif
923 
924 		off = bn % bsize;
925 		sz = MIN(((off_t)1 + nra) * bsize - off, resid);
926 #ifdef	DEBUG
927 		if (vnddebug & VDB_IO)
928 			printf("vndstrategy: vp %p/%p bn 0x%qx/0x%" PRIx64
929 			    " sz 0x%zx\n", vnd->sc_vp, vp, (long long)bn,
930 			    nbn, sz);
931 #endif
932 
933 		nbp = getiobuf(vp, true);
934 		nestiobuf_setup(bp, nbp, offset, sz);
935 		nbp->b_blkno = nbn + btodb(off);
936 
937 #if 0 /* XXX #ifdef DEBUG */
938 		if (vnddebug & VDB_IO)
939 			printf("vndstart(%ld): bp %p vp %p blkno "
940 			    "0x%" PRIx64 " flags %x addr %p cnt 0x%x\n",
941 			    (long) (vnd-vnd_softc), &nbp->vb_buf,
942 			    nbp->vb_buf.b_vp, nbp->vb_buf.b_blkno,
943 			    nbp->vb_buf.b_flags, nbp->vb_buf.b_data,
944 			    nbp->vb_buf.b_bcount);
945 #endif
946 		if (resid == sz) {
947 			break;
948 		}
949 		VOP_STRATEGY(vp, nbp);
950 		bn += sz;
951 	}
952 	if (!(flags & B_READ)) {
953 		struct vnode *w_vp;
954 		/*
955 		 * this is the last nested buf, account for
956 		 * the parent buf write too.
957 		 * This has to be done last, so that
958 		 * fsync won't wait for this write which
959 		 * has no chance to complete before all nested bufs
960 		 * have been queued. But it has to be done
961 		 * before the last VOP_STRATEGY()
962 		 * or the call to nestiobuf_done().
963 		 */
964 		w_vp = bp->b_vp;
965 		mutex_enter(w_vp->v_interlock);
966 		w_vp->v_numoutput++;
967 		mutex_exit(w_vp->v_interlock);
968 	}
969 	KASSERT(skipped != 0 || nbp != NULL);
970 	if (skipped)
971 		nestiobuf_done(bp, skipped, error);
972 	else
973 		VOP_STRATEGY(vp, nbp);
974 }
975 
976 static void
977 vndiodone(struct buf *bp)
978 {
979 	struct vndxfer *vnx = VND_BUFTOXFER(bp);
980 	struct vnd_softc *vnd = vnx->vx_vnd;
981 	struct buf *obp = bp->b_private;
982 	int s = splbio();
983 
984 	KERNEL_LOCK(1, NULL);		/* XXXSMP */
985 	KASSERT(&vnx->vx_buf == bp);
986 	KASSERT(vnd->sc_active > 0);
987 #ifdef DEBUG
988 	if (vnddebug & VDB_IO) {
989 		printf("vndiodone1: bp %p iodone: error %d\n",
990 		    bp, bp->b_error);
991 	}
992 #endif
993 	disk_unbusy(&vnd->sc_dkdev, bp->b_bcount - bp->b_resid,
994 	    (bp->b_flags & B_READ));
995 	vnd->sc_active--;
996 	if (vnd->sc_active == 0) {
997 		wakeup(&vnd->sc_tab);
998 	}
999 	KERNEL_UNLOCK_ONE(NULL);	/* XXXSMP */
1000 	splx(s);
1001 	obp->b_error = bp->b_error;
1002 	obp->b_resid = bp->b_resid;
1003 	buf_destroy(bp);
1004 	VND_PUTXFER(vnd, vnx);
1005 	biodone(obp);
1006 }
1007 
1008 /* ARGSUSED */
1009 static int
1010 vndread(dev_t dev, struct uio *uio, int flags)
1011 {
1012 	int unit = vndunit(dev);
1013 	struct vnd_softc *sc;
1014 
1015 #ifdef DEBUG
1016 	if (vnddebug & VDB_FOLLOW)
1017 		printf("vndread(0x%"PRIx64", %p)\n", dev, uio);
1018 #endif
1019 
1020 	sc = device_lookup_private(&vnd_cd, unit);
1021 	if (sc == NULL)
1022 		return ENXIO;
1023 
1024 	if ((sc->sc_flags & VNF_INITED) == 0)
1025 		return ENXIO;
1026 
1027 	return physio(vndstrategy, NULL, dev, B_READ, minphys, uio);
1028 }
1029 
1030 /* ARGSUSED */
1031 static int
1032 vndwrite(dev_t dev, struct uio *uio, int flags)
1033 {
1034 	int unit = vndunit(dev);
1035 	struct vnd_softc *sc;
1036 
1037 #ifdef DEBUG
1038 	if (vnddebug & VDB_FOLLOW)
1039 		printf("vndwrite(0x%"PRIx64", %p)\n", dev, uio);
1040 #endif
1041 
1042 	sc = device_lookup_private(&vnd_cd, unit);
1043 	if (sc == NULL)
1044 		return ENXIO;
1045 
1046 	if ((sc->sc_flags & VNF_INITED) == 0)
1047 		return ENXIO;
1048 
1049 	return physio(vndstrategy, NULL, dev, B_WRITE, minphys, uio);
1050 }
1051 
1052 static int
1053 vnd_cget(struct lwp *l, int unit, int *un, struct vattr *va)
1054 {
1055 	int error;
1056 	struct vnd_softc *vnd;
1057 
1058 	if (*un == -1)
1059 		*un = unit;
1060 	if (*un < 0)
1061 		return EINVAL;
1062 
1063 	vnd = device_lookup_private(&vnd_cd, *un);
1064 	if (vnd == NULL)
1065 		return -1;
1066 
1067 	if ((vnd->sc_flags & VNF_INITED) == 0)
1068 		return -1;
1069 
1070 	vn_lock(vnd->sc_vp, LK_SHARED | LK_RETRY);
1071 	error = VOP_GETATTR(vnd->sc_vp, va, l->l_cred);
1072 	VOP_UNLOCK(vnd->sc_vp);
1073 	return error;
1074 }
1075 
1076 static int
1077 vnddoclear(struct vnd_softc *vnd, int pmask, int minor, bool force)
1078 {
1079 	int error;
1080 
1081 	if ((error = vndlock(vnd)) != 0)
1082 		return error;
1083 
1084 	/*
1085 	 * Don't unconfigure if any other partitions are open
1086 	 * or if both the character and block flavors of this
1087 	 * partition are open.
1088 	 */
1089 	if (DK_BUSY(vnd, pmask) && !force) {
1090 		vndunlock(vnd);
1091 		return EBUSY;
1092 	}
1093 
1094 	/* Delete all of our wedges */
1095 	dkwedge_delall(&vnd->sc_dkdev);
1096 
1097 	/*
1098 	 * XXX vndclear() might call vndclose() implicitly;
1099 	 * release lock to avoid recursion
1100 	 *
1101 	 * Set VNF_CLEARING to prevent vndopen() from
1102 	 * sneaking in after we vndunlock().
1103 	 */
1104 	vnd->sc_flags |= VNF_CLEARING;
1105 	vndunlock(vnd);
1106 	vndclear(vnd, minor);
1107 #ifdef DEBUG
1108 	if (vnddebug & VDB_INIT)
1109 		printf("%s: CLRed\n", __func__);
1110 #endif
1111 
1112 	/* Destroy the xfer and buffer pools. */
1113 	pool_destroy(&vnd->sc_vxpool);
1114 
1115 	/* Detach the disk. */
1116 	disk_detach(&vnd->sc_dkdev);
1117 
1118 	return 0;
1119 }
1120 
1121 static int
1122 vndioctl_get(struct lwp *l, void *data, int unit, struct vattr *va)
1123 {
1124 	int error;
1125 
1126 	KASSERT(l);
1127 
1128 	/* the first member is always int vnd_unit in all the versions */
1129 	if (*(int *)data >= vnd_cd.cd_ndevs)
1130 		return ENXIO;
1131 
1132 	switch (error = vnd_cget(l, unit, (int *)data, va)) {
1133 	case -1:
1134 		/* unused is not an error */
1135 		memset(va, 0, sizeof(*va));
1136 		/*FALLTHROUGH*/
1137 	case 0:
1138 		return 0;
1139 	default:
1140 		return error;
1141 	}
1142 }
1143 
1144 /* ARGSUSED */
1145 static int
1146 vndioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l)
1147 {
1148 	bool force;
1149 	int unit = vndunit(dev);
1150 	struct vnd_softc *vnd;
1151 	struct vnd_ioctl *vio;
1152 	struct vattr vattr;
1153 	struct pathbuf *pb;
1154 	struct vnode *vp;
1155 	int error, part, pmask;
1156 	uint64_t geomsize;
1157 	int fflags;
1158 #ifdef __HAVE_OLD_DISKLABEL
1159 	struct disklabel newlabel;
1160 #endif
1161 
1162 #ifdef DEBUG
1163 	if (vnddebug & VDB_FOLLOW)
1164 		printf("vndioctl(0x%"PRIx64", 0x%lx, %p, 0x%x, %p): unit %d\n",
1165 		    dev, cmd, data, flag, l->l_proc, unit);
1166 #endif
1167 	/* Do the get's first; they don't need initialization or verification */
1168 	switch (cmd) {
1169 	case VNDIOCGET:
1170 		if ((error = vndioctl_get(l, data, unit, &vattr)) != 0)
1171 			return error;
1172 
1173 		struct vnd_user *vnu = data;
1174 		vnu->vnu_dev = vattr.va_fsid;
1175 		vnu->vnu_ino = vattr.va_fileid;
1176 		return 0;
1177 
1178 	default:
1179 		/* First check for COMPAT_50 hook */
1180 		MODULE_HOOK_CALL(compat_vndioctl_50_hook,
1181 		    (cmd, l, data, unit, &vattr, vndioctl_get),
1182 		    enosys(), error);
1183 
1184 		/*
1185 		 * If not present, then COMPAT_30 hook also not
1186 		 * present, so just continue with checks for the
1187 		 * "write" commands
1188 		 */
1189 		if (error == ENOSYS) {
1190 			error = 0;
1191 			break;
1192 		}
1193 
1194 		/* If not already handled, try the COMPAT_30 hook */
1195 		if (error == EPASSTHROUGH)
1196 			MODULE_HOOK_CALL(compat_vndioctl_30_hook,
1197 			    (cmd, l, data, unit, &vattr, vndioctl_get),
1198 			    enosys(), error);
1199 
1200 		/* If no COMPAT_30 module, or not handled, check writes */
1201 		if (error == ENOSYS || error == EPASSTHROUGH) {
1202 			error = 0;
1203 			break;
1204 		}
1205 		return error;
1206 	}
1207 
1208 	vnd = device_lookup_private(&vnd_cd, unit);
1209 	if (vnd == NULL)
1210 		return ENXIO;
1211 	vio = (struct vnd_ioctl *)data;
1212 
1213 	/* Must be open for writes for these commands... */
1214 	switch (cmd) {
1215 	case VNDIOCSET50:
1216 	case VNDIOCCLR50:
1217 		if (!compat_vndioctl_50_hook.hooked)
1218 			return EINVAL;
1219 		/* FALLTHROUGH */
1220 	case VNDIOCSET:
1221 	case VNDIOCCLR:
1222 	case DIOCSDINFO:
1223 	case DIOCWDINFO:
1224 #ifdef __HAVE_OLD_DISKLABEL
1225 	case ODIOCSDINFO:
1226 	case ODIOCWDINFO:
1227 #endif
1228 	case DIOCKLABEL:
1229 	case DIOCWLABEL:
1230 	case DIOCCACHESYNC:
1231 		if ((flag & FWRITE) == 0)
1232 			return EBADF;
1233 	}
1234 
1235 	switch (cmd) {
1236 	case VNDIOCSET50:
1237 	case VNDIOCSET:
1238 		/* Must not be initialized */
1239 		if (vnd->sc_flags & VNF_INITED)
1240 			return EBUSY;
1241 		break;
1242 	default:
1243 		/* Must be initialized */
1244 		if ((vnd->sc_flags & VNF_INITED) == 0)
1245 			return ENXIO;
1246 		break;
1247 	}
1248 
1249 	error = disk_ioctl(&vnd->sc_dkdev, dev, cmd, data, flag, l);
1250 	if (error != EPASSTHROUGH)
1251 		return error;
1252 
1253 	switch (cmd) {
1254 	case VNDIOCSET50:
1255 	case VNDIOCSET:
1256 		if ((error = vndlock(vnd)) != 0)
1257 			return error;
1258 
1259 		fflags = FREAD;
1260 		if ((vio->vnd_flags & VNDIOF_READONLY) == 0)
1261 			fflags |= FWRITE;
1262 		if ((vio->vnd_flags & VNDIOF_FILEIO) != 0)
1263 			vnd->sc_flags |= VNF_USE_VN_RDWR;
1264 		error = pathbuf_copyin(vio->vnd_file, &pb);
1265 		if (error) {
1266 			goto unlock_and_exit;
1267 		}
1268 		error = vn_open(NULL, pb, 0, fflags, 0, &vp, NULL, NULL);
1269 		if (error != 0) {
1270 			pathbuf_destroy(pb);
1271 			goto unlock_and_exit;
1272 		}
1273 		KASSERT(l);
1274 		error = VOP_GETATTR(vp, &vattr, l->l_cred);
1275 		if (!error && vp->v_type != VREG)
1276 			error = EOPNOTSUPP;
1277 		if (!error && vattr.va_bytes < vattr.va_size)
1278 			/* File is definitely sparse, use vn_rdwr() */
1279 			vnd->sc_flags |= VNF_USE_VN_RDWR;
1280 		if (error) {
1281 			VOP_UNLOCK(vp);
1282 			goto close_and_exit;
1283 		}
1284 
1285 		/* If using a compressed file, initialize its info */
1286 		/* (or abort with an error if kernel has no compression) */
1287 		if (vio->vnd_flags & VNDIOF_COMP) {
1288 #ifdef VND_COMPRESSION
1289 			struct vnd_comp_header *ch;
1290 			int i;
1291 			uint32_t comp_size;
1292 			uint32_t comp_maxsize;
1293 
1294 			/* allocate space for compressed file header */
1295 			ch = malloc(sizeof(struct vnd_comp_header),
1296 			    M_TEMP, M_WAITOK);
1297 
1298 			/* read compressed file header */
1299 			error = vn_rdwr(UIO_READ, vp, (void *)ch,
1300 			    sizeof(struct vnd_comp_header), 0, UIO_SYSSPACE,
1301 			    IO_UNIT|IO_NODELOCKED, l->l_cred, NULL, NULL);
1302 			if (error) {
1303 				free(ch, M_TEMP);
1304 				VOP_UNLOCK(vp);
1305 				goto close_and_exit;
1306 			}
1307 
1308 			if (be32toh(ch->block_size) == 0 ||
1309 			    be32toh(ch->num_blocks) > UINT32_MAX - 1) {
1310 				free(ch, M_TEMP);
1311 				VOP_UNLOCK(vp);
1312 				goto close_and_exit;
1313 			}
1314 
1315 			/* save some header info */
1316 			vnd->sc_comp_blksz = be32toh(ch->block_size);
1317 			/* note last offset is the file byte size */
1318 			vnd->sc_comp_numoffs = be32toh(ch->num_blocks) + 1;
1319 			free(ch, M_TEMP);
1320 			if (!DK_DEV_BSIZE_OK(vnd->sc_comp_blksz)) {
1321 				VOP_UNLOCK(vp);
1322 				error = EINVAL;
1323 				goto close_and_exit;
1324 			}
1325 			KASSERT(0 < vnd->sc_comp_blksz);
1326 			KASSERT(0 < vnd->sc_comp_numoffs);
1327 			/*
1328 			 * @#^@!$& gcc -Wtype-limits refuses to let me
1329 			 * write SIZE_MAX/sizeof(uint64_t) < numoffs,
1330 			 * because the range of the type on amd64 makes
1331 			 * the comparisons always false.
1332 			 */
1333 #if SIZE_MAX <= UINT32_MAX*(64/CHAR_BIT)
1334 			if (SIZE_MAX/sizeof(uint64_t) < vnd->sc_comp_numoffs) {
1335 				VOP_UNLOCK(vp);
1336 				error = EINVAL;
1337 				goto close_and_exit;
1338 			}
1339 #endif
1340 			if ((vattr.va_size < sizeof(struct vnd_comp_header)) ||
1341 			    (vattr.va_size - sizeof(struct vnd_comp_header) <
1342 				sizeof(uint64_t)*vnd->sc_comp_numoffs) ||
1343 			    (UQUAD_MAX/vnd->sc_comp_blksz <
1344 				vnd->sc_comp_numoffs - 1)) {
1345 				VOP_UNLOCK(vp);
1346 				error = EINVAL;
1347 				goto close_and_exit;
1348 			}
1349 
1350 			/* set decompressed file size */
1351 			KASSERT(vnd->sc_comp_numoffs - 1 <=
1352 			    UQUAD_MAX/vnd->sc_comp_blksz);
1353 			vattr.va_size =
1354 			    ((u_quad_t)vnd->sc_comp_numoffs - 1) *
1355 			     (u_quad_t)vnd->sc_comp_blksz;
1356 
1357 			/* allocate space for all the compressed offsets */
1358 			__CTASSERT(UINT32_MAX <= UQUAD_MAX/sizeof(uint64_t));
1359 			vnd->sc_comp_offsets =
1360 			    malloc(sizeof(uint64_t) * vnd->sc_comp_numoffs,
1361 				M_DEVBUF, M_WAITOK);
1362 
1363 			/* read in the offsets */
1364 			error = vn_rdwr(UIO_READ, vp,
1365 			    (void *)vnd->sc_comp_offsets,
1366 			    sizeof(uint64_t) * vnd->sc_comp_numoffs,
1367 			    sizeof(struct vnd_comp_header), UIO_SYSSPACE,
1368 			  IO_UNIT|IO_NODELOCKED, l->l_cred, NULL, NULL);
1369 			if (error) {
1370 				VOP_UNLOCK(vp);
1371 				goto close_and_exit;
1372 			}
1373 			/*
1374 			 * find largest block size (used for allocation limit).
1375 			 * Also convert offset to native byte order.
1376 			 */
1377 			comp_maxsize = 0;
1378 			for (i = 0; i < vnd->sc_comp_numoffs - 1; i++) {
1379 				vnd->sc_comp_offsets[i] =
1380 				    be64toh(vnd->sc_comp_offsets[i]);
1381 				comp_size =
1382 				    be64toh(vnd->sc_comp_offsets[i + 1])
1383 				    - vnd->sc_comp_offsets[i];
1384 				if (comp_size > comp_maxsize)
1385 					comp_maxsize = comp_size;
1386 			}
1387 			vnd->sc_comp_offsets[vnd->sc_comp_numoffs - 1] =
1388 			    be64toh(vnd->sc_comp_offsets[vnd->sc_comp_numoffs
1389 				    - 1]);
1390 
1391 			/* create compressed data buffer */
1392 			vnd->sc_comp_buff = malloc(comp_maxsize,
1393 			    M_DEVBUF, M_WAITOK);
1394 
1395 			/* create decompressed buffer */
1396 			vnd->sc_comp_decombuf = malloc(vnd->sc_comp_blksz,
1397 			    M_DEVBUF, M_WAITOK);
1398 			vnd->sc_comp_buffblk = -1;
1399 
1400 			/* Initialize decompress stream */
1401 			memset(&vnd->sc_comp_stream, 0, sizeof(z_stream));
1402 			vnd->sc_comp_stream.zalloc = vnd_alloc;
1403 			vnd->sc_comp_stream.zfree = vnd_free;
1404 			error = inflateInit2(&vnd->sc_comp_stream, MAX_WBITS);
1405 			if (error) {
1406 				if (vnd->sc_comp_stream.msg)
1407 					printf("vnd%d: compressed file, %s\n",
1408 					    unit, vnd->sc_comp_stream.msg);
1409 				VOP_UNLOCK(vp);
1410 				error = EINVAL;
1411 				goto close_and_exit;
1412 			}
1413 
1414 			vnd->sc_flags |= VNF_COMP | VNF_READONLY;
1415 #else /* !VND_COMPRESSION */
1416 			VOP_UNLOCK(vp);
1417 			error = EOPNOTSUPP;
1418 			goto close_and_exit;
1419 #endif /* VND_COMPRESSION */
1420 		}
1421 
1422 		VOP_UNLOCK(vp);
1423 		vnd->sc_vp = vp;
1424 		vnd->sc_size = btodb(vattr.va_size);	/* note truncation */
1425 
1426 		/* get smallest I/O size for underlying device, fall back to
1427 		 * fundamental I/O size of underlying filesystem
1428 		 */
1429 		error = bdev_ioctl(vattr.va_fsid, DIOCGSECTORSIZE, &vnd->sc_iosize, FKIOCTL, l);
1430 		if (error)
1431 			vnd->sc_iosize = vnd->sc_vp->v_mount->mnt_stat.f_frsize;
1432 
1433 		/* Default I/O size to DEV_BSIZE */
1434 		if (vnd->sc_iosize == 0)
1435 			vnd->sc_iosize = DEV_BSIZE;
1436 
1437 		/*
1438 		 * Use pseudo-geometry specified.  If none was provided,
1439 		 * use "standard" Adaptec fictitious geometry.
1440 		 */
1441 		if (vio->vnd_flags & VNDIOF_HASGEOM) {
1442 
1443 			memcpy(&vnd->sc_geom, &vio->vnd_geom,
1444 			    sizeof(vio->vnd_geom));
1445 
1446 			/*
1447 			 * Sanity-check the sector size.
1448 			 */
1449 			if (!DK_DEV_BSIZE_OK(vnd->sc_geom.vng_secsize) ||
1450 			    vnd->sc_geom.vng_ntracks == 0 ||
1451 			    vnd->sc_geom.vng_nsectors == 0) {
1452 				error = EINVAL;
1453 				goto close_and_exit;
1454 			}
1455 
1456 			/*
1457 			 * Compute missing cylinder count from size
1458 			 */
1459 			if (vnd->sc_geom.vng_ncylinders == 0)
1460 				vnd->sc_geom.vng_ncylinders = vnd->sc_size / (
1461 					(vnd->sc_geom.vng_secsize / DEV_BSIZE) *
1462 					vnd->sc_geom.vng_ntracks *
1463 					vnd->sc_geom.vng_nsectors);
1464 
1465 			/*
1466 			 * Compute the size (in DEV_BSIZE blocks) specified
1467 			 * by the geometry.
1468 			 */
1469 			geomsize = (int64_t)vnd->sc_geom.vng_nsectors *
1470 			    vnd->sc_geom.vng_ntracks *
1471 			    vnd->sc_geom.vng_ncylinders *
1472 			    (vnd->sc_geom.vng_secsize / DEV_BSIZE);
1473 
1474 			/*
1475 			 * Sanity-check the size against the specified
1476 			 * geometry.
1477 			 */
1478 			if (vnd->sc_size < geomsize) {
1479 				error = EINVAL;
1480 				goto close_and_exit;
1481 			}
1482 		} else if (vnd->sc_size >= (32 * 64)) {
1483 			/*
1484 			 * Size must be at least 2048 DEV_BSIZE blocks
1485 			 * (1M) in order to use this geometry.
1486 			 */
1487 			vnd->sc_geom.vng_secsize = DEV_BSIZE;
1488 			vnd->sc_geom.vng_nsectors = 32;
1489 			vnd->sc_geom.vng_ntracks = 64;
1490 			vnd->sc_geom.vng_ncylinders = vnd->sc_size / (64 * 32);
1491 		} else {
1492 			vnd->sc_geom.vng_secsize = DEV_BSIZE;
1493 			vnd->sc_geom.vng_nsectors = 1;
1494 			vnd->sc_geom.vng_ntracks = 1;
1495 			vnd->sc_geom.vng_ncylinders = vnd->sc_size;
1496 		}
1497 
1498 		vnd_set_geometry(vnd);
1499 
1500 		if (vio->vnd_flags & VNDIOF_READONLY) {
1501 			vnd->sc_flags |= VNF_READONLY;
1502 		}
1503 
1504 		if ((error = vndsetcred(vnd, l->l_cred)) != 0)
1505 			goto close_and_exit;
1506 
1507 		vndthrottle(vnd, vnd->sc_vp);
1508 		vio->vnd_osize = dbtob(vnd->sc_size);
1509 		if (cmd != VNDIOCSET50)
1510 			vio->vnd_size = dbtob(vnd->sc_size);
1511 		vnd->sc_flags |= VNF_INITED;
1512 
1513 		/* create the kernel thread, wait for it to be up */
1514 		error = kthread_create(PRI_NONE, 0, NULL, vndthread, vnd,
1515 		    &vnd->sc_kthread, "%s", device_xname(vnd->sc_dev));
1516 		if (error)
1517 			goto close_and_exit;
1518 		while ((vnd->sc_flags & VNF_KTHREAD) == 0) {
1519 			tsleep(&vnd->sc_kthread, PRIBIO, "vndthr", 0);
1520 		}
1521 #ifdef DEBUG
1522 		if (vnddebug & VDB_INIT)
1523 			printf("vndioctl: SET vp %p size 0x%lx %d/%d/%d/%d\n",
1524 			    vnd->sc_vp, (unsigned long) vnd->sc_size,
1525 			    vnd->sc_geom.vng_secsize,
1526 			    vnd->sc_geom.vng_nsectors,
1527 			    vnd->sc_geom.vng_ntracks,
1528 			    vnd->sc_geom.vng_ncylinders);
1529 #endif
1530 
1531 		/* Attach the disk. */
1532 		disk_attach(&vnd->sc_dkdev);
1533 
1534 		/* Initialize the xfer and buffer pools. */
1535 		pool_init(&vnd->sc_vxpool, sizeof(struct vndxfer), 0,
1536 		    0, 0, "vndxpl", NULL, IPL_BIO);
1537 
1538 		vndunlock(vnd);
1539 
1540 		pathbuf_destroy(pb);
1541 
1542 		/* Discover wedges on this disk */
1543 		dkwedge_discover(&vnd->sc_dkdev);
1544 
1545 		break;
1546 
1547 close_and_exit:
1548 		(void) vn_close(vp, fflags, l->l_cred);
1549 		pathbuf_destroy(pb);
1550 unlock_and_exit:
1551 #ifdef VND_COMPRESSION
1552 		/* free any allocated memory (for compressed file) */
1553 		if (vnd->sc_comp_offsets) {
1554 			free(vnd->sc_comp_offsets, M_DEVBUF);
1555 			vnd->sc_comp_offsets = NULL;
1556 		}
1557 		if (vnd->sc_comp_buff) {
1558 			free(vnd->sc_comp_buff, M_DEVBUF);
1559 			vnd->sc_comp_buff = NULL;
1560 		}
1561 		if (vnd->sc_comp_decombuf) {
1562 			free(vnd->sc_comp_decombuf, M_DEVBUF);
1563 			vnd->sc_comp_decombuf = NULL;
1564 		}
1565 #endif /* VND_COMPRESSION */
1566 		vndunlock(vnd);
1567 		return error;
1568 
1569 	case VNDIOCCLR50:
1570 	case VNDIOCCLR:
1571 		part = DISKPART(dev);
1572 		pmask = (1 << part);
1573 		force = (vio->vnd_flags & VNDIOF_FORCE) != 0;
1574 
1575 		if ((error = vnddoclear(vnd, pmask, minor(dev), force)) != 0)
1576 			return error;
1577 
1578 		break;
1579 
1580 
1581 	case DIOCWDINFO:
1582 	case DIOCSDINFO:
1583 #ifdef __HAVE_OLD_DISKLABEL
1584 	case ODIOCWDINFO:
1585 	case ODIOCSDINFO:
1586 #endif
1587 	{
1588 		struct disklabel *lp;
1589 
1590 		if ((error = vndlock(vnd)) != 0)
1591 			return error;
1592 
1593 		vnd->sc_flags |= VNF_LABELLING;
1594 
1595 #ifdef __HAVE_OLD_DISKLABEL
1596 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
1597 			memset(&newlabel, 0, sizeof newlabel);
1598 			memcpy(&newlabel, data, sizeof (struct olddisklabel));
1599 			lp = &newlabel;
1600 		} else
1601 #endif
1602 		lp = (struct disklabel *)data;
1603 
1604 		error = setdisklabel(vnd->sc_dkdev.dk_label,
1605 		    lp, 0, vnd->sc_dkdev.dk_cpulabel);
1606 		if (error == 0) {
1607 			if (cmd == DIOCWDINFO
1608 #ifdef __HAVE_OLD_DISKLABEL
1609 			    || cmd == ODIOCWDINFO
1610 #endif
1611 			   )
1612 				error = writedisklabel(VNDLABELDEV(dev),
1613 				    vndstrategy, vnd->sc_dkdev.dk_label,
1614 				    vnd->sc_dkdev.dk_cpulabel);
1615 		}
1616 
1617 		vnd->sc_flags &= ~VNF_LABELLING;
1618 
1619 		vndunlock(vnd);
1620 
1621 		if (error)
1622 			return error;
1623 		break;
1624 	}
1625 
1626 	case DIOCKLABEL:
1627 		if (*(int *)data != 0)
1628 			vnd->sc_flags |= VNF_KLABEL;
1629 		else
1630 			vnd->sc_flags &= ~VNF_KLABEL;
1631 		break;
1632 
1633 	case DIOCWLABEL:
1634 		if (*(int *)data != 0)
1635 			vnd->sc_flags |= VNF_WLABEL;
1636 		else
1637 			vnd->sc_flags &= ~VNF_WLABEL;
1638 		break;
1639 
1640 	case DIOCGDEFLABEL:
1641 		vndgetdefaultlabel(vnd, (struct disklabel *)data);
1642 		break;
1643 
1644 #ifdef __HAVE_OLD_DISKLABEL
1645 	case ODIOCGDEFLABEL:
1646 		vndgetdefaultlabel(vnd, &newlabel);
1647 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1648 			return ENOTTY;
1649 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1650 		break;
1651 #endif
1652 
1653 	case DIOCGSTRATEGY:
1654 	    {
1655 		struct disk_strategy *dks = (void *)data;
1656 
1657 		/* No lock needed, never changed */
1658 		strlcpy(dks->dks_name,
1659 		    bufq_getstrategyname(vnd->sc_tab),
1660 		    sizeof(dks->dks_name));
1661 		dks->dks_paramlen = 0;
1662 		break;
1663 	    }
1664 	case DIOCGCACHE:
1665 	    {
1666 		int *bits = (int *)data;
1667 		*bits |= DKCACHE_READ | DKCACHE_WRITE;
1668 		break;
1669 	    }
1670 	case DIOCCACHESYNC:
1671 		vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1672 		error = VOP_FSYNC(vnd->sc_vp, vnd->sc_cred,
1673 		    FSYNC_WAIT | FSYNC_DATAONLY | FSYNC_CACHE, 0, 0);
1674 		VOP_UNLOCK(vnd->sc_vp);
1675 		return error;
1676 
1677 	default:
1678 		return ENOTTY;
1679 	}
1680 
1681 	return 0;
1682 }
1683 
1684 /*
1685  * Duplicate the current processes' credentials.  Since we are called only
1686  * as the result of a SET ioctl and only root can do that, any future access
1687  * to this "disk" is essentially as root.  Note that credentials may change
1688  * if some other uid can write directly to the mapped file (NFS).
1689  */
1690 static int
1691 vndsetcred(struct vnd_softc *vnd, kauth_cred_t cred)
1692 {
1693 	struct uio auio;
1694 	struct iovec aiov;
1695 	char *tmpbuf;
1696 	int error;
1697 
1698 	vnd->sc_cred = kauth_cred_dup(cred);
1699 	tmpbuf = malloc(DEV_BSIZE, M_TEMP, M_WAITOK);
1700 
1701 	/* XXX: Horrible kludge to establish credentials for NFS */
1702 	aiov.iov_base = tmpbuf;
1703 	aiov.iov_len = uimin(DEV_BSIZE, dbtob(vnd->sc_size));
1704 	auio.uio_iov = &aiov;
1705 	auio.uio_iovcnt = 1;
1706 	auio.uio_offset = 0;
1707 	auio.uio_rw = UIO_READ;
1708 	auio.uio_resid = aiov.iov_len;
1709 	UIO_SETUP_SYSSPACE(&auio);
1710 	vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1711 	error = VOP_READ(vnd->sc_vp, &auio, 0, vnd->sc_cred);
1712 	if (error == 0) {
1713 		/*
1714 		 * Because vnd does all IO directly through the vnode
1715 		 * we need to flush (at least) the buffer from the above
1716 		 * VOP_READ from the buffer cache to prevent cache
1717 		 * incoherencies.  Also, be careful to write dirty
1718 		 * buffers back to stable storage.
1719 		 */
1720 		error = vinvalbuf(vnd->sc_vp, V_SAVE, vnd->sc_cred,
1721 			    curlwp, 0, 0);
1722 	}
1723 	VOP_UNLOCK(vnd->sc_vp);
1724 
1725 	free(tmpbuf, M_TEMP);
1726 	return error;
1727 }
1728 
1729 /*
1730  * Set maxactive based on FS type
1731  */
1732 static void
1733 vndthrottle(struct vnd_softc *vnd, struct vnode *vp)
1734 {
1735 
1736 	if (vp->v_tag == VT_NFS)
1737 		vnd->sc_maxactive = 2;
1738 	else
1739 		vnd->sc_maxactive = 8;
1740 
1741 	if (vnd->sc_maxactive < 1)
1742 		vnd->sc_maxactive = 1;
1743 }
1744 
1745 #if 0
1746 static void
1747 vndshutdown(void)
1748 {
1749 	struct vnd_softc *vnd;
1750 
1751 	for (vnd = &vnd_softc[0]; vnd < &vnd_softc[numvnd]; vnd++)
1752 		if (vnd->sc_flags & VNF_INITED)
1753 			vndclear(vnd);
1754 }
1755 #endif
1756 
1757 static void
1758 vndclear(struct vnd_softc *vnd, int myminor)
1759 {
1760 	struct vnode *vp = vnd->sc_vp;
1761 	int fflags = FREAD;
1762 	int bmaj, cmaj, i, mn;
1763 	int s;
1764 
1765 #ifdef DEBUG
1766 	if (vnddebug & VDB_FOLLOW)
1767 		printf("vndclear(%p): vp %p\n", vnd, vp);
1768 #endif
1769 	/* locate the major number */
1770 	bmaj = bdevsw_lookup_major(&vnd_bdevsw);
1771 	cmaj = cdevsw_lookup_major(&vnd_cdevsw);
1772 
1773 	/* Nuke the vnodes for any open instances */
1774 	for (i = 0; i < MAXPARTITIONS; i++) {
1775 		mn = DISKMINOR(device_unit(vnd->sc_dev), i);
1776 		if (mn != myminor) { /* XXX avoid to kill own vnode */
1777 			vdevgone(bmaj, mn, mn, VBLK);
1778 			vdevgone(cmaj, mn, mn, VCHR);
1779 		}
1780 	}
1781 
1782 	if ((vnd->sc_flags & VNF_READONLY) == 0)
1783 		fflags |= FWRITE;
1784 
1785 	s = splbio();
1786 	bufq_drain(vnd->sc_tab);
1787 	splx(s);
1788 
1789 	vnd->sc_flags |= VNF_VUNCONF;
1790 	wakeup(&vnd->sc_tab);
1791 	while (vnd->sc_flags & VNF_KTHREAD)
1792 		tsleep(&vnd->sc_kthread, PRIBIO, "vnthr", 0);
1793 
1794 #ifdef VND_COMPRESSION
1795 	/* free the compressed file buffers */
1796 	if (vnd->sc_flags & VNF_COMP) {
1797 		if (vnd->sc_comp_offsets) {
1798 			free(vnd->sc_comp_offsets, M_DEVBUF);
1799 			vnd->sc_comp_offsets = NULL;
1800 		}
1801 		if (vnd->sc_comp_buff) {
1802 			free(vnd->sc_comp_buff, M_DEVBUF);
1803 			vnd->sc_comp_buff = NULL;
1804 		}
1805 		if (vnd->sc_comp_decombuf) {
1806 			free(vnd->sc_comp_decombuf, M_DEVBUF);
1807 			vnd->sc_comp_decombuf = NULL;
1808 		}
1809 	}
1810 #endif /* VND_COMPRESSION */
1811 	vnd->sc_flags &=
1812 	    ~(VNF_INITED | VNF_READONLY | VNF_KLABEL | VNF_VLABEL
1813 	      | VNF_VUNCONF | VNF_COMP | VNF_CLEARING);
1814 	if (vp == NULL)
1815 		panic("vndclear: null vp");
1816 	(void) vn_close(vp, fflags, vnd->sc_cred);
1817 	kauth_cred_free(vnd->sc_cred);
1818 	vnd->sc_vp = NULL;
1819 	vnd->sc_cred = NULL;
1820 	vnd->sc_size = 0;
1821 }
1822 
1823 static int
1824 vndsize(dev_t dev)
1825 {
1826 	struct vnd_softc *sc;
1827 	struct disklabel *lp;
1828 	int part, unit, omask;
1829 	int size;
1830 
1831 	unit = vndunit(dev);
1832 	sc = device_lookup_private(&vnd_cd, unit);
1833 	if (sc == NULL)
1834 		return -1;
1835 
1836 	if ((sc->sc_flags & VNF_INITED) == 0)
1837 		return -1;
1838 
1839 	part = DISKPART(dev);
1840 	omask = sc->sc_dkdev.dk_openmask & (1 << part);
1841 	lp = sc->sc_dkdev.dk_label;
1842 
1843 	if (omask == 0 && vndopen(dev, 0, S_IFBLK, curlwp))	/* XXX */
1844 		return -1;
1845 
1846 	if (lp->d_partitions[part].p_fstype != FS_SWAP)
1847 		size = -1;
1848 	else
1849 		size = lp->d_partitions[part].p_size *
1850 		    (lp->d_secsize / DEV_BSIZE);
1851 
1852 	if (omask == 0 && vndclose(dev, 0, S_IFBLK, curlwp))	/* XXX */
1853 		return -1;
1854 
1855 	return size;
1856 }
1857 
1858 static int
1859 vnddump(dev_t dev, daddr_t blkno, void *va,
1860     size_t size)
1861 {
1862 
1863 	/* Not implemented. */
1864 	return ENXIO;
1865 }
1866 
1867 static void
1868 vndgetdefaultlabel(struct vnd_softc *sc, struct disklabel *lp)
1869 {
1870 	struct vndgeom *vng = &sc->sc_geom;
1871 	struct partition *pp;
1872 	unsigned spb;
1873 
1874 	memset(lp, 0, sizeof(*lp));
1875 
1876 	spb = vng->vng_secsize / DEV_BSIZE;
1877 	if (sc->sc_size / spb > UINT32_MAX)
1878 		lp->d_secperunit = UINT32_MAX;
1879 	else
1880 		lp->d_secperunit = sc->sc_size / spb;
1881 	lp->d_secsize = vng->vng_secsize;
1882 	lp->d_nsectors = vng->vng_nsectors;
1883 	lp->d_ntracks = vng->vng_ntracks;
1884 	lp->d_ncylinders = vng->vng_ncylinders;
1885 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1886 
1887 	strncpy(lp->d_typename, "vnd", sizeof(lp->d_typename));
1888 	lp->d_type = DKTYPE_VND;
1889 	strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
1890 	lp->d_rpm = 3600;
1891 	lp->d_interleave = 1;
1892 	lp->d_flags = 0;
1893 
1894 	pp = &lp->d_partitions[RAW_PART];
1895 	pp->p_offset = 0;
1896 	pp->p_size = lp->d_secperunit;
1897 	pp->p_fstype = FS_UNUSED;
1898 	lp->d_npartitions = RAW_PART + 1;
1899 
1900 	lp->d_magic = DISKMAGIC;
1901 	lp->d_magic2 = DISKMAGIC;
1902 	lp->d_checksum = dkcksum(lp);
1903 }
1904 
1905 /*
1906  * Read the disklabel from a vnd.  If one is not present, create a fake one.
1907  */
1908 static void
1909 vndgetdisklabel(dev_t dev, struct vnd_softc *sc)
1910 {
1911 	const char *errstring;
1912 	struct disklabel *lp = sc->sc_dkdev.dk_label;
1913 	struct cpu_disklabel *clp = sc->sc_dkdev.dk_cpulabel;
1914 	int i;
1915 
1916 	memset(clp, 0, sizeof(*clp));
1917 
1918 	vndgetdefaultlabel(sc, lp);
1919 
1920 	/*
1921 	 * Call the generic disklabel extraction routine.
1922 	 */
1923 	errstring = readdisklabel(VNDLABELDEV(dev), vndstrategy, lp, clp);
1924 	if (errstring) {
1925 		/*
1926 		 * Lack of disklabel is common, but we print the warning
1927 		 * anyway, since it might contain other useful information.
1928 		 */
1929 		aprint_normal_dev(sc->sc_dev, "%s\n", errstring);
1930 
1931 		/*
1932 		 * For historical reasons, if there's no disklabel
1933 		 * present, all partitions must be FS_BSDFFS and
1934 		 * occupy the entire disk.
1935 		 */
1936 		for (i = 0; i < MAXPARTITIONS; i++) {
1937 			/*
1938 			 * Don't wipe out port specific hack (such as
1939 			 * dos partition hack of i386 port).
1940 			 */
1941 			if (lp->d_partitions[i].p_size != 0)
1942 				continue;
1943 
1944 			lp->d_partitions[i].p_size = lp->d_secperunit;
1945 			lp->d_partitions[i].p_offset = 0;
1946 			lp->d_partitions[i].p_fstype = FS_BSDFFS;
1947 		}
1948 
1949 		strncpy(lp->d_packname, "default label",
1950 		    sizeof(lp->d_packname));
1951 
1952 		lp->d_npartitions = MAXPARTITIONS;
1953 		lp->d_checksum = dkcksum(lp);
1954 	}
1955 }
1956 
1957 /*
1958  * Wait interruptibly for an exclusive lock.
1959  *
1960  * XXX
1961  * Several drivers do this; it should be abstracted and made MP-safe.
1962  */
1963 static int
1964 vndlock(struct vnd_softc *sc)
1965 {
1966 	int error;
1967 
1968 	while ((sc->sc_flags & VNF_LOCKED) != 0) {
1969 		sc->sc_flags |= VNF_WANTED;
1970 		if ((error = tsleep(sc, PRIBIO | PCATCH, "vndlck", 0)) != 0)
1971 			return error;
1972 	}
1973 	sc->sc_flags |= VNF_LOCKED;
1974 	return 0;
1975 }
1976 
1977 /*
1978  * Unlock and wake up any waiters.
1979  */
1980 static void
1981 vndunlock(struct vnd_softc *sc)
1982 {
1983 
1984 	sc->sc_flags &= ~VNF_LOCKED;
1985 	if ((sc->sc_flags & VNF_WANTED) != 0) {
1986 		sc->sc_flags &= ~VNF_WANTED;
1987 		wakeup(sc);
1988 	}
1989 }
1990 
1991 #ifdef VND_COMPRESSION
1992 /* compressed file read */
1993 static void
1994 compstrategy(struct buf *bp, off_t bn)
1995 {
1996 	int error;
1997 	int unit = vndunit(bp->b_dev);
1998 	struct vnd_softc *vnd =
1999 	    device_lookup_private(&vnd_cd, unit);
2000 	u_int32_t comp_block;
2001 	struct uio auio;
2002 	char *addr;
2003 	int s;
2004 
2005 	/* set up constants for data move */
2006 	auio.uio_rw = UIO_READ;
2007 	UIO_SETUP_SYSSPACE(&auio);
2008 
2009 	/* read, and transfer the data */
2010 	addr = bp->b_data;
2011 	bp->b_resid = bp->b_bcount;
2012 	s = splbio();
2013 	while (bp->b_resid > 0) {
2014 		unsigned length;
2015 		size_t length_in_buffer;
2016 		u_int32_t offset_in_buffer;
2017 		struct iovec aiov;
2018 
2019 		/* calculate the compressed block number */
2020 		comp_block = bn / (off_t)vnd->sc_comp_blksz;
2021 
2022 		/* check for good block number */
2023 		if (comp_block >= vnd->sc_comp_numoffs) {
2024 			bp->b_error = EINVAL;
2025 			splx(s);
2026 			return;
2027 		}
2028 
2029 		/* read in the compressed block, if not in buffer */
2030 		if (comp_block != vnd->sc_comp_buffblk) {
2031 			length = vnd->sc_comp_offsets[comp_block + 1] -
2032 			    vnd->sc_comp_offsets[comp_block];
2033 			vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
2034 			error = vn_rdwr(UIO_READ, vnd->sc_vp, vnd->sc_comp_buff,
2035 			    length, vnd->sc_comp_offsets[comp_block],
2036 			    UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, vnd->sc_cred,
2037 			    NULL, NULL);
2038 			if (error) {
2039 				bp->b_error = error;
2040 				VOP_UNLOCK(vnd->sc_vp);
2041 				splx(s);
2042 				return;
2043 			}
2044 			/* uncompress the buffer */
2045 			vnd->sc_comp_stream.next_in = vnd->sc_comp_buff;
2046 			vnd->sc_comp_stream.avail_in = length;
2047 			vnd->sc_comp_stream.next_out = vnd->sc_comp_decombuf;
2048 			vnd->sc_comp_stream.avail_out = vnd->sc_comp_blksz;
2049 			inflateReset(&vnd->sc_comp_stream);
2050 			error = inflate(&vnd->sc_comp_stream, Z_FINISH);
2051 			if (error != Z_STREAM_END) {
2052 				if (vnd->sc_comp_stream.msg)
2053 					aprint_normal_dev(vnd->sc_dev,
2054 					    "compressed file, %s\n",
2055 					    vnd->sc_comp_stream.msg);
2056 				bp->b_error = EBADMSG;
2057 				VOP_UNLOCK(vnd->sc_vp);
2058 				splx(s);
2059 				return;
2060 			}
2061 			vnd->sc_comp_buffblk = comp_block;
2062 			VOP_UNLOCK(vnd->sc_vp);
2063 		}
2064 
2065 		/* transfer the usable uncompressed data */
2066 		offset_in_buffer = bn % (off_t)vnd->sc_comp_blksz;
2067 		length_in_buffer = vnd->sc_comp_blksz - offset_in_buffer;
2068 		if (length_in_buffer > bp->b_resid)
2069 			length_in_buffer = bp->b_resid;
2070 		auio.uio_iov = &aiov;
2071 		auio.uio_iovcnt = 1;
2072 		aiov.iov_base = addr;
2073 		aiov.iov_len = length_in_buffer;
2074 		auio.uio_resid = aiov.iov_len;
2075 		auio.uio_offset = 0;
2076 		error = uiomove(vnd->sc_comp_decombuf + offset_in_buffer,
2077 		    length_in_buffer, &auio);
2078 		if (error) {
2079 			bp->b_error = error;
2080 			splx(s);
2081 			return;
2082 		}
2083 
2084 		bn += length_in_buffer;
2085 		addr += length_in_buffer;
2086 		bp->b_resid -= length_in_buffer;
2087 	}
2088 	splx(s);
2089 }
2090 
2091 /* compression memory allocation routines */
2092 static void *
2093 vnd_alloc(void *aux, u_int items, u_int siz)
2094 {
2095 	return malloc(items * siz, M_TEMP, M_NOWAIT);
2096 }
2097 
2098 static void
2099 vnd_free(void *aux, void *ptr)
2100 {
2101 	free(ptr, M_TEMP);
2102 }
2103 #endif /* VND_COMPRESSION */
2104 
2105 static void
2106 vnd_set_geometry(struct vnd_softc *vnd)
2107 {
2108 	struct disk_geom *dg = &vnd->sc_dkdev.dk_geom;
2109 	unsigned spb;
2110 
2111 	memset(dg, 0, sizeof(*dg));
2112 
2113 	spb = vnd->sc_geom.vng_secsize / DEV_BSIZE;
2114 	dg->dg_secperunit = vnd->sc_size / spb;
2115 	dg->dg_secsize = vnd->sc_geom.vng_secsize;
2116 	dg->dg_nsectors = vnd->sc_geom.vng_nsectors;
2117 	dg->dg_ntracks = vnd->sc_geom.vng_ntracks;
2118 	dg->dg_ncylinders = vnd->sc_geom.vng_ncylinders;
2119 
2120 #ifdef DEBUG
2121 	if (vnddebug & VDB_LABEL) {
2122 		printf("dg->dg_secperunit: %" PRId64 "\n", dg->dg_secperunit);
2123 		printf("dg->dg_ncylinders: %u\n", dg->dg_ncylinders);
2124 	}
2125 #endif
2126 	disk_set_info(vnd->sc_dev, &vnd->sc_dkdev, NULL);
2127 }
2128 
2129 #ifdef VND_COMPRESSION
2130 #define VND_DEPENDS "zlib"
2131 #else
2132 #define VND_DEPENDS NULL
2133 #endif
2134 
2135 MODULE(MODULE_CLASS_DRIVER, vnd, VND_DEPENDS);
2136 
2137 #ifdef _MODULE
2138 int vnd_bmajor = -1, vnd_cmajor = -1;
2139 
2140 CFDRIVER_DECL(vnd, DV_DISK, NULL);
2141 #endif
2142 
2143 static int
2144 vnd_modcmd(modcmd_t cmd, void *arg)
2145 {
2146 	int error = 0;
2147 
2148 	switch (cmd) {
2149 	case MODULE_CMD_INIT:
2150 #ifdef _MODULE
2151                 /*
2152                  * Attach the {b,c}devsw's
2153                  */
2154 		error = devsw_attach("vnd", &vnd_bdevsw, &vnd_bmajor,
2155 		    &vnd_cdevsw, &vnd_cmajor);
2156 		if (error) {
2157 #ifdef DIAGNOSTIC
2158                         aprint_error("%s: unable to attach %s devsw, "
2159                             "error %d", __func__, vnd_cd.cd_name, error);
2160 #endif
2161 			break;
2162 		}
2163 
2164 		error = config_cfdriver_attach(&vnd_cd);
2165 		if (error) {
2166 			devsw_detach(&vnd_bdevsw, &vnd_cdevsw);
2167 			break;
2168 		}
2169 
2170 		error = config_cfattach_attach(vnd_cd.cd_name, &vnd_ca);
2171 	        if (error) {
2172 			config_cfdriver_detach(&vnd_cd);
2173 			devsw_detach(&vnd_bdevsw, &vnd_cdevsw);
2174 #ifdef DIAGNOSTIC
2175 			aprint_error("%s: unable to register cfattach for \n"
2176 			    "%s, error %d", __func__, vnd_cd.cd_name, error);
2177 #endif
2178 			break;
2179 		}
2180 #endif
2181 		break;
2182 
2183 	case MODULE_CMD_FINI:
2184 #ifdef _MODULE
2185                 /*
2186                  * Remove device from autoconf database
2187                  */
2188 		error = config_cfattach_detach(vnd_cd.cd_name, &vnd_ca);
2189                 if (error) {
2190 #ifdef DIAGNOSTIC
2191                         aprint_error("%s: failed to detach %s cfattach, "
2192                             "error %d\n", __func__, vnd_cd.cd_name, error);
2193 #endif
2194                         break;
2195                 }
2196                 error = config_cfdriver_detach(&vnd_cd);
2197                 if (error) {
2198                         (void)config_cfattach_attach(vnd_cd.cd_name, &vnd_ca);
2199 #ifdef DIAGNOSTIC
2200                         aprint_error("%s: failed to detach %s cfdriver, "
2201                             "error %d\n", __func__, vnd_cd.cd_name, error);
2202                         break;
2203 #endif
2204                 }
2205                 /*
2206                  * Remove {b,c}devsw's
2207                  */
2208 		devsw_detach(&vnd_bdevsw, &vnd_cdevsw);
2209 
2210 #endif
2211 		break;
2212 
2213 	case MODULE_CMD_STAT:
2214 		return ENOTTY;
2215 
2216 	default:
2217 		return ENOTTY;
2218 	}
2219 
2220 	return error;
2221 }
2222