xref: /netbsd-src/sys/dev/vnd.c (revision 53b02e147d4ed531c0d2a5ca9b3e8026ba3e99b5)
1 /*	$NetBSD: vnd.c,v 1.283 2021/07/24 21:31:36 andvar 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 1.283 2021/07/24 21:31:36 andvar 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 ((flags & FWRITE) && (sc->sc_flags & VNF_READONLY)) {
368 		error = EROFS;
369 		goto done;
370 	}
371 
372 	if (sc->sc_flags & VNF_INITED) {
373 		if ((sc->sc_dkdev.dk_openmask & ~(1<<RAW_PART)) != 0) {
374 			/*
375 			 * If any non-raw partition is open, but the disk
376 			 * has been invalidated, disallow further opens.
377 			 */
378 			if ((sc->sc_flags & VNF_VLABEL) == 0) {
379 				error = EIO;
380 				goto done;
381 			}
382 		} else {
383 			/*
384 			 * Load the partition info if not already loaded.
385 			 */
386 			if ((sc->sc_flags & VNF_VLABEL) == 0) {
387 				sc->sc_flags |= VNF_VLABEL;
388 				vndgetdisklabel(dev, sc);
389 			}
390 		}
391 	}
392 
393 	/* Check that the partitions exists. */
394 	if (part != RAW_PART) {
395 		if (((sc->sc_flags & VNF_INITED) == 0) ||
396 		    ((part >= lp->d_npartitions) ||
397 		     (lp->d_partitions[part].p_fstype == FS_UNUSED))) {
398 			error = ENXIO;
399 			goto done;
400 		}
401 	}
402 
403 	/* Prevent our unit from being unconfigured while open. */
404 	switch (mode) {
405 	case S_IFCHR:
406 		sc->sc_dkdev.dk_copenmask |= pmask;
407 		break;
408 
409 	case S_IFBLK:
410 		sc->sc_dkdev.dk_bopenmask |= pmask;
411 		break;
412 	}
413 	sc->sc_dkdev.dk_openmask =
414 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
415 
416  done:
417 	mutex_exit(&sc->sc_dkdev.dk_openlock);
418 	vndunlock(sc);
419 	return error;
420 }
421 
422 static int
423 vndclose(dev_t dev, int flags, int mode, struct lwp *l)
424 {
425 	int unit = vndunit(dev);
426 	struct vnd_softc *sc;
427 	int error = 0, part;
428 
429 #ifdef DEBUG
430 	if (vnddebug & VDB_FOLLOW)
431 		printf("vndclose(0x%"PRIx64", 0x%x, 0x%x, %p)\n", dev, flags, mode, l);
432 #endif
433 	sc = device_lookup_private(&vnd_cd, unit);
434 	if (sc == NULL)
435 		return ENXIO;
436 
437 	if ((error = vndlock(sc)) != 0)
438 		return error;
439 
440 	mutex_enter(&sc->sc_dkdev.dk_openlock);
441 
442 	part = DISKPART(dev);
443 
444 	/* ...that much closer to allowing unconfiguration... */
445 	switch (mode) {
446 	case S_IFCHR:
447 		sc->sc_dkdev.dk_copenmask &= ~(1 << part);
448 		break;
449 
450 	case S_IFBLK:
451 		sc->sc_dkdev.dk_bopenmask &= ~(1 << part);
452 		break;
453 	}
454 	sc->sc_dkdev.dk_openmask =
455 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
456 
457 	/* are we last opener ? */
458 	if (sc->sc_dkdev.dk_openmask == 0) {
459 		if ((sc->sc_flags & VNF_KLABEL) == 0)
460 			sc->sc_flags &= ~VNF_VLABEL;
461 	}
462 
463 	mutex_exit(&sc->sc_dkdev.dk_openlock);
464 
465 	vndunlock(sc);
466 
467 	if ((sc->sc_flags & VNF_INITED) == 0) {
468 		if ((error = vnd_destroy(sc->sc_dev)) != 0) {
469 			aprint_error_dev(sc->sc_dev,
470 			    "unable to detach instance\n");
471 			return error;
472 		}
473 	}
474 
475 	return 0;
476 }
477 
478 /*
479  * Queue the request, and wakeup the kernel thread to handle it.
480  */
481 static void
482 vndstrategy(struct buf *bp)
483 {
484 	int unit = vndunit(bp->b_dev);
485 	struct vnd_softc *vnd =
486 	    device_lookup_private(&vnd_cd, unit);
487 	struct disklabel *lp;
488 	daddr_t blkno;
489 	int s = splbio();
490 
491 	if (vnd == NULL) {
492 		bp->b_error = ENXIO;
493 		goto done;
494 	}
495 	lp = vnd->sc_dkdev.dk_label;
496 
497 	if ((vnd->sc_flags & VNF_INITED) == 0) {
498 		bp->b_error = ENXIO;
499 		goto done;
500 	}
501 
502 	/*
503 	 * The transfer must be a whole number of blocks.
504 	 */
505 	if ((bp->b_bcount % lp->d_secsize) != 0) {
506 		bp->b_error = EINVAL;
507 		goto done;
508 	}
509 
510 	/*
511 	 * check if we're read-only.
512 	 */
513 	if ((vnd->sc_flags & VNF_READONLY) && !(bp->b_flags & B_READ)) {
514 		bp->b_error = EACCES;
515 		goto done;
516 	}
517 
518 	/* If it's a nil transfer, wake up the top half now. */
519 	if (bp->b_bcount == 0) {
520 		goto done;
521 	}
522 
523 	/*
524 	 * Do bounds checking and adjust transfer.  If there's an error,
525 	 * the bounds check will flag that for us.
526 	 */
527 	if (DISKPART(bp->b_dev) == RAW_PART) {
528 		if (bounds_check_with_mediasize(bp, DEV_BSIZE,
529 		    vnd->sc_size) <= 0)
530 			goto done;
531 	} else {
532 		if (bounds_check_with_label(&vnd->sc_dkdev,
533 		    bp, vnd->sc_flags & (VNF_WLABEL|VNF_LABELLING)) <= 0)
534 			goto done;
535 	}
536 
537 	/*
538 	 * Put the block number in terms of the logical blocksize
539 	 * of the "device".
540 	 */
541 
542 	blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
543 
544 	/*
545 	 * Translate the partition-relative block number to an absolute.
546 	 */
547 	if (DISKPART(bp->b_dev) != RAW_PART) {
548 		struct partition *pp;
549 
550 		pp = &vnd->sc_dkdev.dk_label->d_partitions[
551 		    DISKPART(bp->b_dev)];
552 		blkno += pp->p_offset;
553 	}
554 	bp->b_rawblkno = blkno;
555 
556 #ifdef DEBUG
557 	if (vnddebug & VDB_FOLLOW)
558 		printf("vndstrategy(%p): unit %d\n", bp, unit);
559 #endif
560 	if ((vnd->sc_flags & VNF_USE_VN_RDWR)) {
561 		KASSERT(vnd->sc_pending >= 0 &&
562 		    vnd->sc_pending <= VND_MAXPENDING(vnd));
563 		while (vnd->sc_pending == VND_MAXPENDING(vnd))
564 			tsleep(&vnd->sc_pending, PRIBIO, "vndpc", 0);
565 		vnd->sc_pending++;
566 	}
567 	bufq_put(vnd->sc_tab, bp);
568 	wakeup(&vnd->sc_tab);
569 	splx(s);
570 	return;
571 
572 done:
573 	bp->b_resid = bp->b_bcount;
574 	biodone(bp);
575 	splx(s);
576 }
577 
578 static bool
579 vnode_has_strategy(struct vnd_softc *vnd)
580 {
581 	return vnode_has_op(vnd->sc_vp, VOFFSET(vop_bmap)) &&
582 	    vnode_has_op(vnd->sc_vp, VOFFSET(vop_strategy));
583 }
584 
585 /* Verify that I/O requests cannot be smaller than the
586  * smallest I/O size supported by the backend.
587  */
588 static bool
589 vnode_has_large_blocks(struct vnd_softc *vnd)
590 {
591 	u_int32_t vnd_secsize, iosize;
592 
593 	iosize = vnd->sc_iosize;
594 	vnd_secsize = vnd->sc_geom.vng_secsize;
595 
596 	return vnd_secsize % iosize != 0;
597 }
598 
599 /* XXX this function needs a reliable check to detect
600  * sparse files. Otherwise, bmap/strategy may be used
601  * and fail on non-allocated blocks. VOP_READ/VOP_WRITE
602  * works on sparse files.
603  */
604 #if notyet
605 static bool
606 vnode_strategy_probe(struct vnd_softc *vnd)
607 {
608 	int error;
609 	daddr_t nbn;
610 
611 	if (!vnode_has_strategy(vnd))
612 		return false;
613 
614 	if (vnode_has_large_blocks(vnd))
615 		return false;
616 
617 	/* Convert the first logical block number to its
618 	 * physical block number.
619 	 */
620 	error = 0;
621 	vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
622 	error = VOP_BMAP(vnd->sc_vp, 0, NULL, &nbn, NULL);
623 	VOP_UNLOCK(vnd->sc_vp);
624 
625 	/* Test if that worked. */
626 	if (error == 0 && (long)nbn == -1)
627 		return false;
628 
629 	return true;
630 }
631 #endif
632 
633 static void
634 vndthread(void *arg)
635 {
636 	struct vnd_softc *vnd = arg;
637 	int s;
638 
639 	/* Determine whether we can *use* VOP_BMAP and VOP_STRATEGY to
640 	 * directly access the backing vnode.  If we can, use these two
641 	 * operations to avoid messing with the local buffer cache.
642 	 * Otherwise fall back to regular VOP_READ/VOP_WRITE operations
643 	 * which are guaranteed to work with any file system. */
644 	if ((vnd->sc_flags & VNF_USE_VN_RDWR) == 0 &&
645 	    ! vnode_has_strategy(vnd))
646 		vnd->sc_flags |= VNF_USE_VN_RDWR;
647 
648 	/* VOP_STRATEGY can only be used if the backing vnode allows
649 	 * to access blocks as small as defined by the vnd geometry.
650 	 */
651 	if ((vnd->sc_flags & VNF_USE_VN_RDWR) == 0 &&
652 	    vnode_has_large_blocks(vnd))
653 		vnd->sc_flags |= VNF_USE_VN_RDWR;
654 
655 #ifdef DEBUG
656 	if (vnddebug & VDB_INIT)
657 		printf("vndthread: vp %p, %s\n", vnd->sc_vp,
658 		    (vnd->sc_flags & VNF_USE_VN_RDWR) == 0 ?
659 		    "using bmap/strategy operations" :
660 		    "using read/write operations");
661 #endif
662 
663 	s = splbio();
664 	vnd->sc_flags |= VNF_KTHREAD;
665 	wakeup(&vnd->sc_kthread);
666 
667 	/*
668 	 * Dequeue requests and serve them depending on the available
669 	 * vnode operations.
670 	 */
671 	while ((vnd->sc_flags & VNF_VUNCONF) == 0) {
672 		struct vndxfer *vnx;
673 		struct buf *obp;
674 		struct buf *bp;
675 
676 		obp = bufq_get(vnd->sc_tab);
677 		if (obp == NULL) {
678 			tsleep(&vnd->sc_tab, PRIBIO, "vndbp", 0);
679 			continue;
680 		};
681 		if ((vnd->sc_flags & VNF_USE_VN_RDWR)) {
682 			KASSERT(vnd->sc_pending > 0 &&
683 			    vnd->sc_pending <= VND_MAXPENDING(vnd));
684 			if (vnd->sc_pending-- == VND_MAXPENDING(vnd))
685 				wakeup(&vnd->sc_pending);
686 		}
687 		splx(s);
688 #ifdef DEBUG
689 		if (vnddebug & VDB_FOLLOW)
690 			printf("vndthread(%p)\n", obp);
691 #endif
692 
693 		if (vnd->sc_vp->v_mount == NULL) {
694 			obp->b_error = ENXIO;
695 			goto done;
696 		}
697 #ifdef VND_COMPRESSION
698 		/* handle a compressed read */
699 		if ((obp->b_flags & B_READ) != 0 && (vnd->sc_flags & VNF_COMP)) {
700 			off_t bn;
701 
702 			/* Convert to a byte offset within the file. */
703 			bn = obp->b_rawblkno *
704 			    vnd->sc_dkdev.dk_label->d_secsize;
705 
706 			compstrategy(obp, bn);
707 			goto done;
708 		}
709 #endif /* VND_COMPRESSION */
710 
711 		/*
712 		 * Allocate a header for this transfer and link it to the
713 		 * buffer
714 		 */
715 		s = splbio();
716 		vnx = VND_GETXFER(vnd);
717 		splx(s);
718 		vnx->vx_vnd = vnd;
719 
720 		s = splbio();
721 		while (vnd->sc_active >= vnd->sc_maxactive) {
722 			tsleep(&vnd->sc_tab, PRIBIO, "vndac", 0);
723 		}
724 		vnd->sc_active++;
725 		splx(s);
726 
727 		/* Instrumentation. */
728 		disk_busy(&vnd->sc_dkdev);
729 
730 		bp = &vnx->vx_buf;
731 		buf_init(bp);
732 		bp->b_flags = (obp->b_flags & (B_READ | B_PHYS | B_RAW));
733 		bp->b_oflags = obp->b_oflags;
734 		bp->b_cflags = obp->b_cflags;
735 		bp->b_iodone = vndiodone;
736 		bp->b_private = obp;
737 		bp->b_vp = vnd->sc_vp;
738 		bp->b_objlock = bp->b_vp->v_interlock;
739 		bp->b_data = obp->b_data;
740 		bp->b_bcount = obp->b_bcount;
741 		BIO_COPYPRIO(bp, obp);
742 
743 		/* Make sure the request succeeds while suspending this fs. */
744 		fstrans_start_lazy(vnd->sc_vp->v_mount);
745 
746 		/* Handle the request using the appropriate operations. */
747 		if ((vnd->sc_flags & VNF_USE_VN_RDWR) == 0)
748 			handle_with_strategy(vnd, obp, bp);
749 		else
750 			handle_with_rdwr(vnd, obp, bp);
751 
752 		fstrans_done(vnd->sc_vp->v_mount);
753 
754 		s = splbio();
755 		continue;
756 
757 done:
758 		biodone(obp);
759 		s = splbio();
760 	}
761 
762 	vnd->sc_flags &= (~VNF_KTHREAD | VNF_VUNCONF);
763 	wakeup(&vnd->sc_kthread);
764 	splx(s);
765 	kthread_exit(0);
766 }
767 
768 /*
769  * Checks if the given vnode supports the requested operation.
770  * The operation is specified the offset returned by VOFFSET.
771  *
772  * XXX The test below used to determine this is quite fragile
773  * because it relies on the file system to use genfs to specify
774  * unimplemented operations.  There might be another way to do
775  * it more cleanly.
776  */
777 static bool
778 vnode_has_op(const struct vnode *vp, int opoffset)
779 {
780 	int (*defaultp)(void *);
781 	int (*opp)(void *);
782 
783 	defaultp = vp->v_op[VOFFSET(vop_default)];
784 	opp = vp->v_op[opoffset];
785 
786 	return opp != defaultp && opp != genfs_eopnotsupp &&
787 	    opp != genfs_badop && opp != genfs_nullop;
788 }
789 
790 /*
791  * Handles the read/write request given in 'bp' using the vnode's VOP_READ
792  * and VOP_WRITE operations.
793  *
794  * 'obp' is a pointer to the original request fed to the vnd device.
795  */
796 static void
797 handle_with_rdwr(struct vnd_softc *vnd, const struct buf *obp, struct buf *bp)
798 {
799 	bool doread;
800 	off_t offset;
801 	size_t len, resid;
802 	struct vnode *vp;
803 	int npages;
804 
805 	doread = bp->b_flags & B_READ;
806 	offset = obp->b_rawblkno * vnd->sc_dkdev.dk_label->d_secsize;
807 	len = bp->b_bcount;
808 	vp = vnd->sc_vp;
809 
810 #if defined(DEBUG)
811 	if (vnddebug & VDB_IO)
812 		printf("vnd (rdwr): vp %p, %s, rawblkno 0x%" PRIx64
813 		    ", secsize %d, offset %" PRIu64
814 		    ", bcount %d\n",
815 		    vp, doread ? "read" : "write", obp->b_rawblkno,
816 		    vnd->sc_dkdev.dk_label->d_secsize, offset,
817 		    bp->b_bcount);
818 #endif
819 
820 	/* Issue the read or write operation. */
821 	bp->b_error =
822 	    vn_rdwr(doread ? UIO_READ : UIO_WRITE,
823 	    vp, bp->b_data, len, offset, UIO_SYSSPACE,
824 	    IO_ADV_ENCODE(POSIX_FADV_NOREUSE) | IO_DIRECT,
825 		vnd->sc_cred, &resid, NULL);
826 	bp->b_resid = resid;
827 
828 	/*
829 	 * Avoid caching too many pages, the vnd user
830 	 * is usually a filesystem and caches itself.
831 	 * We need some amount of caching to not hinder
832 	 * read-ahead and write-behind operations.
833 	 */
834 	npages = atomic_load_relaxed(&vp->v_uobj.uo_npages);
835 	if (npages > VND_MAXPAGES(vnd)) {
836 		rw_enter(vp->v_uobj.vmobjlock, RW_WRITER);
837 		(void) VOP_PUTPAGES(vp, 0, 0,
838 		    PGO_ALLPAGES | PGO_CLEANIT | PGO_FREE);
839 	}
840 
841 	/* We need to increase the number of outputs on the vnode if
842 	 * there was any write to it. */
843 	if (!doread) {
844 		mutex_enter(vp->v_interlock);
845 		vp->v_numoutput++;
846 		mutex_exit(vp->v_interlock);
847 	}
848 
849 	biodone(bp);
850 }
851 
852 /*
853  * Handes the read/write request given in 'bp' using the vnode's VOP_BMAP
854  * and VOP_STRATEGY operations.
855  *
856  * 'obp' is a pointer to the original request fed to the vnd device.
857  */
858 static void
859 handle_with_strategy(struct vnd_softc *vnd, const struct buf *obp,
860     struct buf *bp)
861 {
862 	int bsize, error, flags, skipped;
863 	size_t resid, sz;
864 	off_t bn, offset;
865 	struct vnode *vp;
866 	struct buf *nbp = NULL;
867 
868 	flags = obp->b_flags;
869 
870 
871 	/* convert to a byte offset within the file. */
872 	bn = obp->b_rawblkno * vnd->sc_dkdev.dk_label->d_secsize;
873 
874 	bsize = vnd->sc_vp->v_mount->mnt_stat.f_iosize;
875 	skipped = 0;
876 
877 	/*
878 	 * Break the request into bsize pieces and feed them
879 	 * sequentially using VOP_BMAP/VOP_STRATEGY.
880 	 * We do it this way to keep from flooding NFS servers if we
881 	 * are connected to an NFS file.  This places the burden on
882 	 * the client rather than the server.
883 	 */
884 	error = 0;
885 	bp->b_resid = bp->b_bcount;
886 	for (offset = 0, resid = bp->b_resid; /* true */;
887 	    resid -= sz, offset += sz) {
888 		daddr_t nbn;
889 		int off, nra;
890 
891 		nra = 0;
892 		vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
893 		error = VOP_BMAP(vnd->sc_vp, bn / bsize, &vp, &nbn, &nra);
894 		VOP_UNLOCK(vnd->sc_vp);
895 
896 		if (error == 0 && (long)nbn == -1)
897 			error = EIO;
898 
899 		/*
900 		 * If there was an error or a hole in the file...punt.
901 		 * Note that we may have to wait for any operations
902 		 * that we have already fired off before releasing
903 		 * the buffer.
904 		 *
905 		 * XXX we could deal with holes here but it would be
906 		 * a hassle (in the write case).
907 		 */
908 		if (error) {
909 			skipped += resid;
910 			break;
911 		}
912 
913 #ifdef DEBUG
914 		if (!dovndcluster)
915 			nra = 0;
916 #endif
917 
918 		off = bn % bsize;
919 		sz = MIN(((off_t)1 + nra) * bsize - off, resid);
920 #ifdef	DEBUG
921 		if (vnddebug & VDB_IO)
922 			printf("vndstrategy: vp %p/%p bn 0x%qx/0x%" PRIx64
923 			    " sz 0x%zx\n", vnd->sc_vp, vp, (long long)bn,
924 			    nbn, sz);
925 #endif
926 
927 		nbp = getiobuf(vp, true);
928 		nestiobuf_setup(bp, nbp, offset, sz);
929 		nbp->b_blkno = nbn + btodb(off);
930 
931 #if 0 /* XXX #ifdef DEBUG */
932 		if (vnddebug & VDB_IO)
933 			printf("vndstart(%ld): bp %p vp %p blkno "
934 			    "0x%" PRIx64 " flags %x addr %p cnt 0x%x\n",
935 			    (long) (vnd-vnd_softc), &nbp->vb_buf,
936 			    nbp->vb_buf.b_vp, nbp->vb_buf.b_blkno,
937 			    nbp->vb_buf.b_flags, nbp->vb_buf.b_data,
938 			    nbp->vb_buf.b_bcount);
939 #endif
940 		if (resid == sz) {
941 			break;
942 		}
943 		VOP_STRATEGY(vp, nbp);
944 		bn += sz;
945 	}
946 	if (!(flags & B_READ)) {
947 		struct vnode *w_vp;
948 		/*
949 		 * this is the last nested buf, account for
950 		 * the parent buf write too.
951 		 * This has to be done last, so that
952 		 * fsync won't wait for this write which
953 		 * has no chance to complete before all nested bufs
954 		 * have been queued. But it has to be done
955 		 * before the last VOP_STRATEGY()
956 		 * or the call to nestiobuf_done().
957 		 */
958 		w_vp = bp->b_vp;
959 		mutex_enter(w_vp->v_interlock);
960 		w_vp->v_numoutput++;
961 		mutex_exit(w_vp->v_interlock);
962 	}
963 	KASSERT(skipped != 0 || nbp != NULL);
964 	if (skipped)
965 		nestiobuf_done(bp, skipped, error);
966 	else
967 		VOP_STRATEGY(vp, nbp);
968 }
969 
970 static void
971 vndiodone(struct buf *bp)
972 {
973 	struct vndxfer *vnx = VND_BUFTOXFER(bp);
974 	struct vnd_softc *vnd = vnx->vx_vnd;
975 	struct buf *obp = bp->b_private;
976 	int s = splbio();
977 
978 	KERNEL_LOCK(1, NULL);		/* XXXSMP */
979 	KASSERT(&vnx->vx_buf == bp);
980 	KASSERT(vnd->sc_active > 0);
981 #ifdef DEBUG
982 	if (vnddebug & VDB_IO) {
983 		printf("vndiodone1: bp %p iodone: error %d\n",
984 		    bp, bp->b_error);
985 	}
986 #endif
987 	disk_unbusy(&vnd->sc_dkdev, bp->b_bcount - bp->b_resid,
988 	    (bp->b_flags & B_READ));
989 	vnd->sc_active--;
990 	if (vnd->sc_active == 0) {
991 		wakeup(&vnd->sc_tab);
992 	}
993 	KERNEL_UNLOCK_ONE(NULL);	/* XXXSMP */
994 	splx(s);
995 	obp->b_error = bp->b_error;
996 	obp->b_resid = bp->b_resid;
997 	buf_destroy(bp);
998 	VND_PUTXFER(vnd, vnx);
999 	biodone(obp);
1000 }
1001 
1002 /* ARGSUSED */
1003 static int
1004 vndread(dev_t dev, struct uio *uio, int flags)
1005 {
1006 	int unit = vndunit(dev);
1007 	struct vnd_softc *sc;
1008 
1009 #ifdef DEBUG
1010 	if (vnddebug & VDB_FOLLOW)
1011 		printf("vndread(0x%"PRIx64", %p)\n", dev, uio);
1012 #endif
1013 
1014 	sc = device_lookup_private(&vnd_cd, unit);
1015 	if (sc == NULL)
1016 		return ENXIO;
1017 
1018 	if ((sc->sc_flags & VNF_INITED) == 0)
1019 		return ENXIO;
1020 
1021 	return physio(vndstrategy, NULL, dev, B_READ, minphys, uio);
1022 }
1023 
1024 /* ARGSUSED */
1025 static int
1026 vndwrite(dev_t dev, struct uio *uio, int flags)
1027 {
1028 	int unit = vndunit(dev);
1029 	struct vnd_softc *sc;
1030 
1031 #ifdef DEBUG
1032 	if (vnddebug & VDB_FOLLOW)
1033 		printf("vndwrite(0x%"PRIx64", %p)\n", dev, uio);
1034 #endif
1035 
1036 	sc = device_lookup_private(&vnd_cd, unit);
1037 	if (sc == NULL)
1038 		return ENXIO;
1039 
1040 	if ((sc->sc_flags & VNF_INITED) == 0)
1041 		return ENXIO;
1042 
1043 	return physio(vndstrategy, NULL, dev, B_WRITE, minphys, uio);
1044 }
1045 
1046 static int
1047 vnd_cget(struct lwp *l, int unit, int *un, struct vattr *va)
1048 {
1049 	int error;
1050 	struct vnd_softc *vnd;
1051 
1052 	if (*un == -1)
1053 		*un = unit;
1054 	if (*un < 0)
1055 		return EINVAL;
1056 
1057 	vnd = device_lookup_private(&vnd_cd, *un);
1058 	if (vnd == NULL)
1059 		return -1;
1060 
1061 	if ((vnd->sc_flags & VNF_INITED) == 0)
1062 		return -1;
1063 
1064 	vn_lock(vnd->sc_vp, LK_SHARED | LK_RETRY);
1065 	error = VOP_GETATTR(vnd->sc_vp, va, l->l_cred);
1066 	VOP_UNLOCK(vnd->sc_vp);
1067 	return error;
1068 }
1069 
1070 static int
1071 vnddoclear(struct vnd_softc *vnd, int pmask, int minor, bool force)
1072 {
1073 	int error;
1074 
1075 	if ((error = vndlock(vnd)) != 0)
1076 		return error;
1077 
1078 	/*
1079 	 * Don't unconfigure if any other partitions are open
1080 	 * or if both the character and block flavors of this
1081 	 * partition are open.
1082 	 */
1083 	if (DK_BUSY(vnd, pmask) && !force) {
1084 		vndunlock(vnd);
1085 		return EBUSY;
1086 	}
1087 
1088 	/* Delete all of our wedges */
1089 	dkwedge_delall(&vnd->sc_dkdev);
1090 
1091 	/*
1092 	 * XXX vndclear() might call vndclose() implicitly;
1093 	 * release lock to avoid recursion
1094 	 *
1095 	 * Set VNF_CLEARING to prevent vndopen() from
1096 	 * sneaking in after we vndunlock().
1097 	 */
1098 	vnd->sc_flags |= VNF_CLEARING;
1099 	vndunlock(vnd);
1100 	vndclear(vnd, minor);
1101 #ifdef DEBUG
1102 	if (vnddebug & VDB_INIT)
1103 		printf("%s: CLRed\n", __func__);
1104 #endif
1105 
1106 	/* Destroy the xfer and buffer pools. */
1107 	pool_destroy(&vnd->sc_vxpool);
1108 
1109 	/* Detach the disk. */
1110 	disk_detach(&vnd->sc_dkdev);
1111 
1112 	return 0;
1113 }
1114 
1115 static int
1116 vndioctl_get(struct lwp *l, void *data, int unit, struct vattr *va)
1117 {
1118 	int error;
1119 
1120 	KASSERT(l);
1121 
1122 	/* the first member is always int vnd_unit in all the versions */
1123 	if (*(int *)data >= vnd_cd.cd_ndevs)
1124 		return ENXIO;
1125 
1126 	switch (error = vnd_cget(l, unit, (int *)data, va)) {
1127 	case -1:
1128 		/* unused is not an error */
1129 		memset(va, 0, sizeof(*va));
1130 		/*FALLTHROUGH*/
1131 	case 0:
1132 		return 0;
1133 	default:
1134 		return error;
1135 	}
1136 }
1137 
1138 /* ARGSUSED */
1139 static int
1140 vndioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l)
1141 {
1142 	bool force;
1143 	int unit = vndunit(dev);
1144 	struct vnd_softc *vnd;
1145 	struct vnd_ioctl *vio;
1146 	struct vattr vattr;
1147 	struct pathbuf *pb;
1148 	struct vnode *vp;
1149 	int error, part, pmask;
1150 	uint64_t geomsize;
1151 	int fflags;
1152 #ifdef __HAVE_OLD_DISKLABEL
1153 	struct disklabel newlabel;
1154 #endif
1155 
1156 #ifdef DEBUG
1157 	if (vnddebug & VDB_FOLLOW)
1158 		printf("vndioctl(0x%"PRIx64", 0x%lx, %p, 0x%x, %p): unit %d\n",
1159 		    dev, cmd, data, flag, l->l_proc, unit);
1160 #endif
1161 	/* Do the get's first; they don't need initialization or verification */
1162 	switch (cmd) {
1163 	case VNDIOCGET:
1164 		if ((error = vndioctl_get(l, data, unit, &vattr)) != 0)
1165 			return error;
1166 
1167 		struct vnd_user *vnu = data;
1168 		vnu->vnu_dev = vattr.va_fsid;
1169 		vnu->vnu_ino = vattr.va_fileid;
1170 		return 0;
1171 
1172 	default:
1173 		/* First check for COMPAT_50 hook */
1174 		MODULE_HOOK_CALL(compat_vndioctl_50_hook,
1175 		    (cmd, l, data, unit, &vattr, vndioctl_get),
1176 		    enosys(), error);
1177 
1178 		/*
1179 		 * If not present, then COMPAT_30 hook also not
1180 		 * present, so just continue with checks for the
1181 		 * "write" commands
1182 		 */
1183 		if (error == ENOSYS) {
1184 			error = 0;
1185 			break;
1186 		}
1187 
1188 		/* If not already handled, try the COMPAT_30 hook */
1189 		if (error == EPASSTHROUGH)
1190 			MODULE_HOOK_CALL(compat_vndioctl_30_hook,
1191 			    (cmd, l, data, unit, &vattr, vndioctl_get),
1192 			    enosys(), error);
1193 
1194 		/* If no COMPAT_30 module, or not handled, check writes */
1195 		if (error == ENOSYS || error == EPASSTHROUGH) {
1196 			error = 0;
1197 			break;
1198 		}
1199 		return error;
1200 	}
1201 
1202 	vnd = device_lookup_private(&vnd_cd, unit);
1203 	if (vnd == NULL)
1204 		return ENXIO;
1205 	vio = (struct vnd_ioctl *)data;
1206 
1207 	/* Must be open for writes for these commands... */
1208 	switch (cmd) {
1209 	case VNDIOCSET50:
1210 	case VNDIOCCLR50:
1211 		if (!compat_vndioctl_50_hook.hooked)
1212 			return EINVAL;
1213 		/* FALLTHROUGH */
1214 	case VNDIOCSET:
1215 	case VNDIOCCLR:
1216 	case DIOCSDINFO:
1217 	case DIOCWDINFO:
1218 #ifdef __HAVE_OLD_DISKLABEL
1219 	case ODIOCSDINFO:
1220 	case ODIOCWDINFO:
1221 #endif
1222 	case DIOCKLABEL:
1223 	case DIOCWLABEL:
1224 	case DIOCCACHESYNC:
1225 		if ((flag & FWRITE) == 0)
1226 			return EBADF;
1227 	}
1228 
1229 	/* Must be initialized for these... */
1230 	switch (cmd) {
1231 	case VNDIOCCLR:
1232 	case VNDIOCCLR50:
1233 	case DIOCGDINFO:
1234 	case DIOCSDINFO:
1235 	case DIOCWDINFO:
1236 	case DIOCGPARTINFO:
1237 	case DIOCKLABEL:
1238 	case DIOCWLABEL:
1239 	case DIOCGDEFLABEL:
1240 	case DIOCGCACHE:
1241 	case DIOCGSTRATEGY:
1242 	case DIOCCACHESYNC:
1243 #ifdef __HAVE_OLD_DISKLABEL
1244 	case ODIOCGDINFO:
1245 	case ODIOCSDINFO:
1246 	case ODIOCWDINFO:
1247 	case ODIOCGDEFLABEL:
1248 #endif
1249 		if ((vnd->sc_flags & VNF_INITED) == 0)
1250 			return ENXIO;
1251 	}
1252 
1253 	error = disk_ioctl(&vnd->sc_dkdev, dev, cmd, data, flag, l);
1254 	if (error != EPASSTHROUGH)
1255 		return error;
1256 
1257 
1258 	switch (cmd) {
1259 	case VNDIOCSET50:
1260 	case VNDIOCSET:
1261 		if (vnd->sc_flags & VNF_INITED)
1262 			return EBUSY;
1263 
1264 		if ((error = vndlock(vnd)) != 0)
1265 			return error;
1266 
1267 		fflags = FREAD;
1268 		if ((vio->vnd_flags & VNDIOF_READONLY) == 0)
1269 			fflags |= FWRITE;
1270 		if ((vio->vnd_flags & VNDIOF_FILEIO) != 0)
1271 			vnd->sc_flags |= VNF_USE_VN_RDWR;
1272 		error = pathbuf_copyin(vio->vnd_file, &pb);
1273 		if (error) {
1274 			goto unlock_and_exit;
1275 		}
1276 		error = vn_open(NULL, pb, 0, fflags, 0, &vp, NULL, NULL);
1277 		if (error != 0) {
1278 			pathbuf_destroy(pb);
1279 			goto unlock_and_exit;
1280 		}
1281 		KASSERT(l);
1282 		error = VOP_GETATTR(vp, &vattr, l->l_cred);
1283 		if (!error && vp->v_type != VREG)
1284 			error = EOPNOTSUPP;
1285 		if (!error && vattr.va_bytes < vattr.va_size)
1286 			/* File is definitely sparse, use vn_rdwr() */
1287 			vnd->sc_flags |= VNF_USE_VN_RDWR;
1288 		if (error) {
1289 			VOP_UNLOCK(vp);
1290 			goto close_and_exit;
1291 		}
1292 
1293 		/* If using a compressed file, initialize its info */
1294 		/* (or abort with an error if kernel has no compression) */
1295 		if (vio->vnd_flags & VNDIOF_COMP) {
1296 #ifdef VND_COMPRESSION
1297 			struct vnd_comp_header *ch;
1298 			int i;
1299 			uint32_t comp_size;
1300 			uint32_t comp_maxsize;
1301 
1302 			/* allocate space for compressed file header */
1303 			ch = malloc(sizeof(struct vnd_comp_header),
1304 			    M_TEMP, M_WAITOK);
1305 
1306 			/* read compressed file header */
1307 			error = vn_rdwr(UIO_READ, vp, (void *)ch,
1308 			    sizeof(struct vnd_comp_header), 0, UIO_SYSSPACE,
1309 			    IO_UNIT|IO_NODELOCKED, l->l_cred, NULL, NULL);
1310 			if (error) {
1311 				free(ch, M_TEMP);
1312 				VOP_UNLOCK(vp);
1313 				goto close_and_exit;
1314 			}
1315 
1316 			if (be32toh(ch->block_size) == 0 ||
1317 			    be32toh(ch->num_blocks) > UINT32_MAX - 1) {
1318 				free(ch, M_TEMP);
1319 				VOP_UNLOCK(vp);
1320 				goto close_and_exit;
1321 			}
1322 
1323 			/* save some header info */
1324 			vnd->sc_comp_blksz = be32toh(ch->block_size);
1325 			/* note last offset is the file byte size */
1326 			vnd->sc_comp_numoffs = be32toh(ch->num_blocks) + 1;
1327 			free(ch, M_TEMP);
1328 			if (!DK_DEV_BSIZE_OK(vnd->sc_comp_blksz)) {
1329 				VOP_UNLOCK(vp);
1330 				error = EINVAL;
1331 				goto close_and_exit;
1332 			}
1333 			KASSERT(0 < vnd->sc_comp_blksz);
1334 			KASSERT(0 < vnd->sc_comp_numoffs);
1335 			/*
1336 			 * @#^@!$& gcc -Wtype-limits refuses to let me
1337 			 * write SIZE_MAX/sizeof(uint64_t) < numoffs,
1338 			 * because the range of the type on amd64 makes
1339 			 * the comparisons always false.
1340 			 */
1341 #if SIZE_MAX <= UINT32_MAX*(64/CHAR_BIT)
1342 			if (SIZE_MAX/sizeof(uint64_t) < vnd->sc_comp_numoffs) {
1343 				VOP_UNLOCK(vp);
1344 				error = EINVAL;
1345 				goto close_and_exit;
1346 			}
1347 #endif
1348 			if ((vattr.va_size < sizeof(struct vnd_comp_header)) ||
1349 			    (vattr.va_size - sizeof(struct vnd_comp_header) <
1350 				sizeof(uint64_t)*vnd->sc_comp_numoffs) ||
1351 			    (UQUAD_MAX/vnd->sc_comp_blksz <
1352 				vnd->sc_comp_numoffs - 1)) {
1353 				VOP_UNLOCK(vp);
1354 				error = EINVAL;
1355 				goto close_and_exit;
1356 			}
1357 
1358 			/* set decompressed file size */
1359 			KASSERT(vnd->sc_comp_numoffs - 1 <=
1360 			    UQUAD_MAX/vnd->sc_comp_blksz);
1361 			vattr.va_size =
1362 			    ((u_quad_t)vnd->sc_comp_numoffs - 1) *
1363 			     (u_quad_t)vnd->sc_comp_blksz;
1364 
1365 			/* allocate space for all the compressed offsets */
1366 			__CTASSERT(UINT32_MAX <= UQUAD_MAX/sizeof(uint64_t));
1367 			vnd->sc_comp_offsets =
1368 			    malloc(sizeof(uint64_t) * vnd->sc_comp_numoffs,
1369 				M_DEVBUF, M_WAITOK);
1370 
1371 			/* read in the offsets */
1372 			error = vn_rdwr(UIO_READ, vp,
1373 			    (void *)vnd->sc_comp_offsets,
1374 			    sizeof(uint64_t) * vnd->sc_comp_numoffs,
1375 			    sizeof(struct vnd_comp_header), UIO_SYSSPACE,
1376 			  IO_UNIT|IO_NODELOCKED, l->l_cred, NULL, NULL);
1377 			if (error) {
1378 				VOP_UNLOCK(vp);
1379 				goto close_and_exit;
1380 			}
1381 			/*
1382 			 * find largest block size (used for allocation limit).
1383 			 * Also convert offset to native byte order.
1384 			 */
1385 			comp_maxsize = 0;
1386 			for (i = 0; i < vnd->sc_comp_numoffs - 1; i++) {
1387 				vnd->sc_comp_offsets[i] =
1388 				    be64toh(vnd->sc_comp_offsets[i]);
1389 				comp_size =
1390 				    be64toh(vnd->sc_comp_offsets[i + 1])
1391 				    - vnd->sc_comp_offsets[i];
1392 				if (comp_size > comp_maxsize)
1393 					comp_maxsize = comp_size;
1394 			}
1395 			vnd->sc_comp_offsets[vnd->sc_comp_numoffs - 1] =
1396 			    be64toh(vnd->sc_comp_offsets[vnd->sc_comp_numoffs
1397 				    - 1]);
1398 
1399 			/* create compressed data buffer */
1400 			vnd->sc_comp_buff = malloc(comp_maxsize,
1401 			    M_DEVBUF, M_WAITOK);
1402 
1403 			/* create decompressed buffer */
1404 			vnd->sc_comp_decombuf = malloc(vnd->sc_comp_blksz,
1405 			    M_DEVBUF, M_WAITOK);
1406 			vnd->sc_comp_buffblk = -1;
1407 
1408 			/* Initialize decompress stream */
1409 			memset(&vnd->sc_comp_stream, 0, sizeof(z_stream));
1410 			vnd->sc_comp_stream.zalloc = vnd_alloc;
1411 			vnd->sc_comp_stream.zfree = vnd_free;
1412 			error = inflateInit2(&vnd->sc_comp_stream, MAX_WBITS);
1413 			if (error) {
1414 				if (vnd->sc_comp_stream.msg)
1415 					printf("vnd%d: compressed file, %s\n",
1416 					    unit, vnd->sc_comp_stream.msg);
1417 				VOP_UNLOCK(vp);
1418 				error = EINVAL;
1419 				goto close_and_exit;
1420 			}
1421 
1422 			vnd->sc_flags |= VNF_COMP | VNF_READONLY;
1423 #else /* !VND_COMPRESSION */
1424 			VOP_UNLOCK(vp);
1425 			error = EOPNOTSUPP;
1426 			goto close_and_exit;
1427 #endif /* VND_COMPRESSION */
1428 		}
1429 
1430 		VOP_UNLOCK(vp);
1431 		vnd->sc_vp = vp;
1432 		vnd->sc_size = btodb(vattr.va_size);	/* note truncation */
1433 
1434 		/* get smallest I/O size for underlying device, fall back to
1435 		 * fundamental I/O size of underlying filesystem
1436 		 */
1437 		error = bdev_ioctl(vattr.va_fsid, DIOCGSECTORSIZE, &vnd->sc_iosize, FKIOCTL, l);
1438 		if (error)
1439 			vnd->sc_iosize = vnd->sc_vp->v_mount->mnt_stat.f_frsize;
1440 
1441 		/* Default I/O size to DEV_BSIZE */
1442 		if (vnd->sc_iosize == 0)
1443 			vnd->sc_iosize = DEV_BSIZE;
1444 
1445 		/*
1446 		 * Use pseudo-geometry specified.  If none was provided,
1447 		 * use "standard" Adaptec fictitious geometry.
1448 		 */
1449 		if (vio->vnd_flags & VNDIOF_HASGEOM) {
1450 
1451 			memcpy(&vnd->sc_geom, &vio->vnd_geom,
1452 			    sizeof(vio->vnd_geom));
1453 
1454 			/*
1455 			 * Sanity-check the sector size.
1456 			 */
1457 			if (!DK_DEV_BSIZE_OK(vnd->sc_geom.vng_secsize) ||
1458 			    vnd->sc_geom.vng_ntracks == 0 ||
1459 			    vnd->sc_geom.vng_nsectors == 0) {
1460 				error = EINVAL;
1461 				goto close_and_exit;
1462 			}
1463 
1464 			/*
1465 			 * Compute missing cylinder count from size
1466 			 */
1467 			if (vnd->sc_geom.vng_ncylinders == 0)
1468 				vnd->sc_geom.vng_ncylinders = vnd->sc_size / (
1469 					(vnd->sc_geom.vng_secsize / DEV_BSIZE) *
1470 					vnd->sc_geom.vng_ntracks *
1471 					vnd->sc_geom.vng_nsectors);
1472 
1473 			/*
1474 			 * Compute the size (in DEV_BSIZE blocks) specified
1475 			 * by the geometry.
1476 			 */
1477 			geomsize = (int64_t)vnd->sc_geom.vng_nsectors *
1478 			    vnd->sc_geom.vng_ntracks *
1479 			    vnd->sc_geom.vng_ncylinders *
1480 			    (vnd->sc_geom.vng_secsize / DEV_BSIZE);
1481 
1482 			/*
1483 			 * Sanity-check the size against the specified
1484 			 * geometry.
1485 			 */
1486 			if (vnd->sc_size < geomsize) {
1487 				error = EINVAL;
1488 				goto close_and_exit;
1489 			}
1490 		} else if (vnd->sc_size >= (32 * 64)) {
1491 			/*
1492 			 * Size must be at least 2048 DEV_BSIZE blocks
1493 			 * (1M) in order to use this geometry.
1494 			 */
1495 			vnd->sc_geom.vng_secsize = DEV_BSIZE;
1496 			vnd->sc_geom.vng_nsectors = 32;
1497 			vnd->sc_geom.vng_ntracks = 64;
1498 			vnd->sc_geom.vng_ncylinders = vnd->sc_size / (64 * 32);
1499 		} else {
1500 			vnd->sc_geom.vng_secsize = DEV_BSIZE;
1501 			vnd->sc_geom.vng_nsectors = 1;
1502 			vnd->sc_geom.vng_ntracks = 1;
1503 			vnd->sc_geom.vng_ncylinders = vnd->sc_size;
1504 		}
1505 
1506 		vnd_set_geometry(vnd);
1507 
1508 		if (vio->vnd_flags & VNDIOF_READONLY) {
1509 			vnd->sc_flags |= VNF_READONLY;
1510 		}
1511 
1512 		if ((error = vndsetcred(vnd, l->l_cred)) != 0)
1513 			goto close_and_exit;
1514 
1515 		vndthrottle(vnd, vnd->sc_vp);
1516 		vio->vnd_osize = dbtob(vnd->sc_size);
1517 		if (cmd != VNDIOCSET50)
1518 			vio->vnd_size = dbtob(vnd->sc_size);
1519 		vnd->sc_flags |= VNF_INITED;
1520 
1521 		/* create the kernel thread, wait for it to be up */
1522 		error = kthread_create(PRI_NONE, 0, NULL, vndthread, vnd,
1523 		    &vnd->sc_kthread, "%s", device_xname(vnd->sc_dev));
1524 		if (error)
1525 			goto close_and_exit;
1526 		while ((vnd->sc_flags & VNF_KTHREAD) == 0) {
1527 			tsleep(&vnd->sc_kthread, PRIBIO, "vndthr", 0);
1528 		}
1529 #ifdef DEBUG
1530 		if (vnddebug & VDB_INIT)
1531 			printf("vndioctl: SET vp %p size 0x%lx %d/%d/%d/%d\n",
1532 			    vnd->sc_vp, (unsigned long) vnd->sc_size,
1533 			    vnd->sc_geom.vng_secsize,
1534 			    vnd->sc_geom.vng_nsectors,
1535 			    vnd->sc_geom.vng_ntracks,
1536 			    vnd->sc_geom.vng_ncylinders);
1537 #endif
1538 
1539 		/* Attach the disk. */
1540 		disk_attach(&vnd->sc_dkdev);
1541 
1542 		/* Initialize the xfer and buffer pools. */
1543 		pool_init(&vnd->sc_vxpool, sizeof(struct vndxfer), 0,
1544 		    0, 0, "vndxpl", NULL, IPL_BIO);
1545 
1546 		vndunlock(vnd);
1547 
1548 		pathbuf_destroy(pb);
1549 
1550 		/* Discover wedges on this disk */
1551 		dkwedge_discover(&vnd->sc_dkdev);
1552 
1553 		break;
1554 
1555 close_and_exit:
1556 		(void) vn_close(vp, fflags, l->l_cred);
1557 		pathbuf_destroy(pb);
1558 unlock_and_exit:
1559 #ifdef VND_COMPRESSION
1560 		/* free any allocated memory (for compressed file) */
1561 		if (vnd->sc_comp_offsets) {
1562 			free(vnd->sc_comp_offsets, M_DEVBUF);
1563 			vnd->sc_comp_offsets = NULL;
1564 		}
1565 		if (vnd->sc_comp_buff) {
1566 			free(vnd->sc_comp_buff, M_DEVBUF);
1567 			vnd->sc_comp_buff = NULL;
1568 		}
1569 		if (vnd->sc_comp_decombuf) {
1570 			free(vnd->sc_comp_decombuf, M_DEVBUF);
1571 			vnd->sc_comp_decombuf = NULL;
1572 		}
1573 #endif /* VND_COMPRESSION */
1574 		vndunlock(vnd);
1575 		return error;
1576 
1577 	case VNDIOCCLR50:
1578 	case VNDIOCCLR:
1579 		part = DISKPART(dev);
1580 		pmask = (1 << part);
1581 		force = (vio->vnd_flags & VNDIOF_FORCE) != 0;
1582 
1583 		if ((error = vnddoclear(vnd, pmask, minor(dev), force)) != 0)
1584 			return error;
1585 
1586 		break;
1587 
1588 
1589 	case DIOCWDINFO:
1590 	case DIOCSDINFO:
1591 #ifdef __HAVE_OLD_DISKLABEL
1592 	case ODIOCWDINFO:
1593 	case ODIOCSDINFO:
1594 #endif
1595 	{
1596 		struct disklabel *lp;
1597 
1598 		if ((error = vndlock(vnd)) != 0)
1599 			return error;
1600 
1601 		vnd->sc_flags |= VNF_LABELLING;
1602 
1603 #ifdef __HAVE_OLD_DISKLABEL
1604 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
1605 			memset(&newlabel, 0, sizeof newlabel);
1606 			memcpy(&newlabel, data, sizeof (struct olddisklabel));
1607 			lp = &newlabel;
1608 		} else
1609 #endif
1610 		lp = (struct disklabel *)data;
1611 
1612 		error = setdisklabel(vnd->sc_dkdev.dk_label,
1613 		    lp, 0, vnd->sc_dkdev.dk_cpulabel);
1614 		if (error == 0) {
1615 			if (cmd == DIOCWDINFO
1616 #ifdef __HAVE_OLD_DISKLABEL
1617 			    || cmd == ODIOCWDINFO
1618 #endif
1619 			   )
1620 				error = writedisklabel(VNDLABELDEV(dev),
1621 				    vndstrategy, vnd->sc_dkdev.dk_label,
1622 				    vnd->sc_dkdev.dk_cpulabel);
1623 		}
1624 
1625 		vnd->sc_flags &= ~VNF_LABELLING;
1626 
1627 		vndunlock(vnd);
1628 
1629 		if (error)
1630 			return error;
1631 		break;
1632 	}
1633 
1634 	case DIOCKLABEL:
1635 		if (*(int *)data != 0)
1636 			vnd->sc_flags |= VNF_KLABEL;
1637 		else
1638 			vnd->sc_flags &= ~VNF_KLABEL;
1639 		break;
1640 
1641 	case DIOCWLABEL:
1642 		if (*(int *)data != 0)
1643 			vnd->sc_flags |= VNF_WLABEL;
1644 		else
1645 			vnd->sc_flags &= ~VNF_WLABEL;
1646 		break;
1647 
1648 	case DIOCGDEFLABEL:
1649 		vndgetdefaultlabel(vnd, (struct disklabel *)data);
1650 		break;
1651 
1652 #ifdef __HAVE_OLD_DISKLABEL
1653 	case ODIOCGDEFLABEL:
1654 		vndgetdefaultlabel(vnd, &newlabel);
1655 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1656 			return ENOTTY;
1657 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1658 		break;
1659 #endif
1660 
1661 	case DIOCGSTRATEGY:
1662 	    {
1663 		struct disk_strategy *dks = (void *)data;
1664 
1665 		/* No lock needed, never changed */
1666 		strlcpy(dks->dks_name,
1667 		    bufq_getstrategyname(vnd->sc_tab),
1668 		    sizeof(dks->dks_name));
1669 		dks->dks_paramlen = 0;
1670 		break;
1671 	    }
1672 	case DIOCGCACHE:
1673 	    {
1674 		int *bits = (int *)data;
1675 		*bits |= DKCACHE_READ | DKCACHE_WRITE;
1676 		break;
1677 	    }
1678 	case DIOCCACHESYNC:
1679 		vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1680 		error = VOP_FSYNC(vnd->sc_vp, vnd->sc_cred,
1681 		    FSYNC_WAIT | FSYNC_DATAONLY | FSYNC_CACHE, 0, 0);
1682 		VOP_UNLOCK(vnd->sc_vp);
1683 		return error;
1684 
1685 	default:
1686 		return ENOTTY;
1687 	}
1688 
1689 	return 0;
1690 }
1691 
1692 /*
1693  * Duplicate the current processes' credentials.  Since we are called only
1694  * as the result of a SET ioctl and only root can do that, any future access
1695  * to this "disk" is essentially as root.  Note that credentials may change
1696  * if some other uid can write directly to the mapped file (NFS).
1697  */
1698 static int
1699 vndsetcred(struct vnd_softc *vnd, kauth_cred_t cred)
1700 {
1701 	struct uio auio;
1702 	struct iovec aiov;
1703 	char *tmpbuf;
1704 	int error;
1705 
1706 	vnd->sc_cred = kauth_cred_dup(cred);
1707 	tmpbuf = malloc(DEV_BSIZE, M_TEMP, M_WAITOK);
1708 
1709 	/* XXX: Horrible kludge to establish credentials for NFS */
1710 	aiov.iov_base = tmpbuf;
1711 	aiov.iov_len = uimin(DEV_BSIZE, dbtob(vnd->sc_size));
1712 	auio.uio_iov = &aiov;
1713 	auio.uio_iovcnt = 1;
1714 	auio.uio_offset = 0;
1715 	auio.uio_rw = UIO_READ;
1716 	auio.uio_resid = aiov.iov_len;
1717 	UIO_SETUP_SYSSPACE(&auio);
1718 	vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
1719 	error = VOP_READ(vnd->sc_vp, &auio, 0, vnd->sc_cred);
1720 	if (error == 0) {
1721 		/*
1722 		 * Because vnd does all IO directly through the vnode
1723 		 * we need to flush (at least) the buffer from the above
1724 		 * VOP_READ from the buffer cache to prevent cache
1725 		 * incoherencies.  Also, be careful to write dirty
1726 		 * buffers back to stable storage.
1727 		 */
1728 		error = vinvalbuf(vnd->sc_vp, V_SAVE, vnd->sc_cred,
1729 			    curlwp, 0, 0);
1730 	}
1731 	VOP_UNLOCK(vnd->sc_vp);
1732 
1733 	free(tmpbuf, M_TEMP);
1734 	return error;
1735 }
1736 
1737 /*
1738  * Set maxactive based on FS type
1739  */
1740 static void
1741 vndthrottle(struct vnd_softc *vnd, struct vnode *vp)
1742 {
1743 
1744 	if (vp->v_tag == VT_NFS)
1745 		vnd->sc_maxactive = 2;
1746 	else
1747 		vnd->sc_maxactive = 8;
1748 
1749 	if (vnd->sc_maxactive < 1)
1750 		vnd->sc_maxactive = 1;
1751 }
1752 
1753 #if 0
1754 static void
1755 vndshutdown(void)
1756 {
1757 	struct vnd_softc *vnd;
1758 
1759 	for (vnd = &vnd_softc[0]; vnd < &vnd_softc[numvnd]; vnd++)
1760 		if (vnd->sc_flags & VNF_INITED)
1761 			vndclear(vnd);
1762 }
1763 #endif
1764 
1765 static void
1766 vndclear(struct vnd_softc *vnd, int myminor)
1767 {
1768 	struct vnode *vp = vnd->sc_vp;
1769 	int fflags = FREAD;
1770 	int bmaj, cmaj, i, mn;
1771 	int s;
1772 
1773 #ifdef DEBUG
1774 	if (vnddebug & VDB_FOLLOW)
1775 		printf("vndclear(%p): vp %p\n", vnd, vp);
1776 #endif
1777 	/* locate the major number */
1778 	bmaj = bdevsw_lookup_major(&vnd_bdevsw);
1779 	cmaj = cdevsw_lookup_major(&vnd_cdevsw);
1780 
1781 	/* Nuke the vnodes for any open instances */
1782 	for (i = 0; i < MAXPARTITIONS; i++) {
1783 		mn = DISKMINOR(device_unit(vnd->sc_dev), i);
1784 		vdevgone(bmaj, mn, mn, VBLK);
1785 		if (mn != myminor) /* XXX avoid to kill own vnode */
1786 			vdevgone(cmaj, mn, mn, VCHR);
1787 	}
1788 
1789 	if ((vnd->sc_flags & VNF_READONLY) == 0)
1790 		fflags |= FWRITE;
1791 
1792 	s = splbio();
1793 	bufq_drain(vnd->sc_tab);
1794 	splx(s);
1795 
1796 	vnd->sc_flags |= VNF_VUNCONF;
1797 	wakeup(&vnd->sc_tab);
1798 	while (vnd->sc_flags & VNF_KTHREAD)
1799 		tsleep(&vnd->sc_kthread, PRIBIO, "vnthr", 0);
1800 
1801 #ifdef VND_COMPRESSION
1802 	/* free the compressed file buffers */
1803 	if (vnd->sc_flags & VNF_COMP) {
1804 		if (vnd->sc_comp_offsets) {
1805 			free(vnd->sc_comp_offsets, M_DEVBUF);
1806 			vnd->sc_comp_offsets = NULL;
1807 		}
1808 		if (vnd->sc_comp_buff) {
1809 			free(vnd->sc_comp_buff, M_DEVBUF);
1810 			vnd->sc_comp_buff = NULL;
1811 		}
1812 		if (vnd->sc_comp_decombuf) {
1813 			free(vnd->sc_comp_decombuf, M_DEVBUF);
1814 			vnd->sc_comp_decombuf = NULL;
1815 		}
1816 	}
1817 #endif /* VND_COMPRESSION */
1818 	vnd->sc_flags &=
1819 	    ~(VNF_INITED | VNF_READONLY | VNF_KLABEL | VNF_VLABEL
1820 	      | VNF_VUNCONF | VNF_COMP | VNF_CLEARING);
1821 	if (vp == NULL)
1822 		panic("vndclear: null vp");
1823 	(void) vn_close(vp, fflags, vnd->sc_cred);
1824 	kauth_cred_free(vnd->sc_cred);
1825 	vnd->sc_vp = NULL;
1826 	vnd->sc_cred = NULL;
1827 	vnd->sc_size = 0;
1828 }
1829 
1830 static int
1831 vndsize(dev_t dev)
1832 {
1833 	struct vnd_softc *sc;
1834 	struct disklabel *lp;
1835 	int part, unit, omask;
1836 	int size;
1837 
1838 	unit = vndunit(dev);
1839 	sc = device_lookup_private(&vnd_cd, unit);
1840 	if (sc == NULL)
1841 		return -1;
1842 
1843 	if ((sc->sc_flags & VNF_INITED) == 0)
1844 		return -1;
1845 
1846 	part = DISKPART(dev);
1847 	omask = sc->sc_dkdev.dk_openmask & (1 << part);
1848 	lp = sc->sc_dkdev.dk_label;
1849 
1850 	if (omask == 0 && vndopen(dev, 0, S_IFBLK, curlwp))	/* XXX */
1851 		return -1;
1852 
1853 	if (lp->d_partitions[part].p_fstype != FS_SWAP)
1854 		size = -1;
1855 	else
1856 		size = lp->d_partitions[part].p_size *
1857 		    (lp->d_secsize / DEV_BSIZE);
1858 
1859 	if (omask == 0 && vndclose(dev, 0, S_IFBLK, curlwp))	/* XXX */
1860 		return -1;
1861 
1862 	return size;
1863 }
1864 
1865 static int
1866 vnddump(dev_t dev, daddr_t blkno, void *va,
1867     size_t size)
1868 {
1869 
1870 	/* Not implemented. */
1871 	return ENXIO;
1872 }
1873 
1874 static void
1875 vndgetdefaultlabel(struct vnd_softc *sc, struct disklabel *lp)
1876 {
1877 	struct vndgeom *vng = &sc->sc_geom;
1878 	struct partition *pp;
1879 	unsigned spb;
1880 
1881 	memset(lp, 0, sizeof(*lp));
1882 
1883 	spb = vng->vng_secsize / DEV_BSIZE;
1884 	if (sc->sc_size / spb > UINT32_MAX)
1885 		lp->d_secperunit = UINT32_MAX;
1886 	else
1887 		lp->d_secperunit = sc->sc_size / spb;
1888 	lp->d_secsize = vng->vng_secsize;
1889 	lp->d_nsectors = vng->vng_nsectors;
1890 	lp->d_ntracks = vng->vng_ntracks;
1891 	lp->d_ncylinders = vng->vng_ncylinders;
1892 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1893 
1894 	strncpy(lp->d_typename, "vnd", sizeof(lp->d_typename));
1895 	lp->d_type = DKTYPE_VND;
1896 	strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
1897 	lp->d_rpm = 3600;
1898 	lp->d_interleave = 1;
1899 	lp->d_flags = 0;
1900 
1901 	pp = &lp->d_partitions[RAW_PART];
1902 	pp->p_offset = 0;
1903 	pp->p_size = lp->d_secperunit;
1904 	pp->p_fstype = FS_UNUSED;
1905 	lp->d_npartitions = RAW_PART + 1;
1906 
1907 	lp->d_magic = DISKMAGIC;
1908 	lp->d_magic2 = DISKMAGIC;
1909 	lp->d_checksum = dkcksum(lp);
1910 }
1911 
1912 /*
1913  * Read the disklabel from a vnd.  If one is not present, create a fake one.
1914  */
1915 static void
1916 vndgetdisklabel(dev_t dev, struct vnd_softc *sc)
1917 {
1918 	const char *errstring;
1919 	struct disklabel *lp = sc->sc_dkdev.dk_label;
1920 	struct cpu_disklabel *clp = sc->sc_dkdev.dk_cpulabel;
1921 	int i;
1922 
1923 	memset(clp, 0, sizeof(*clp));
1924 
1925 	vndgetdefaultlabel(sc, lp);
1926 
1927 	/*
1928 	 * Call the generic disklabel extraction routine.
1929 	 */
1930 	errstring = readdisklabel(VNDLABELDEV(dev), vndstrategy, lp, clp);
1931 	if (errstring) {
1932 		/*
1933 		 * Lack of disklabel is common, but we print the warning
1934 		 * anyway, since it might contain other useful information.
1935 		 */
1936 		aprint_normal_dev(sc->sc_dev, "%s\n", errstring);
1937 
1938 		/*
1939 		 * For historical reasons, if there's no disklabel
1940 		 * present, all partitions must be FS_BSDFFS and
1941 		 * occupy the entire disk.
1942 		 */
1943 		for (i = 0; i < MAXPARTITIONS; i++) {
1944 			/*
1945 			 * Don't wipe out port specific hack (such as
1946 			 * dos partition hack of i386 port).
1947 			 */
1948 			if (lp->d_partitions[i].p_size != 0)
1949 				continue;
1950 
1951 			lp->d_partitions[i].p_size = lp->d_secperunit;
1952 			lp->d_partitions[i].p_offset = 0;
1953 			lp->d_partitions[i].p_fstype = FS_BSDFFS;
1954 		}
1955 
1956 		strncpy(lp->d_packname, "default label",
1957 		    sizeof(lp->d_packname));
1958 
1959 		lp->d_npartitions = MAXPARTITIONS;
1960 		lp->d_checksum = dkcksum(lp);
1961 	}
1962 }
1963 
1964 /*
1965  * Wait interruptibly for an exclusive lock.
1966  *
1967  * XXX
1968  * Several drivers do this; it should be abstracted and made MP-safe.
1969  */
1970 static int
1971 vndlock(struct vnd_softc *sc)
1972 {
1973 	int error;
1974 
1975 	while ((sc->sc_flags & VNF_LOCKED) != 0) {
1976 		sc->sc_flags |= VNF_WANTED;
1977 		if ((error = tsleep(sc, PRIBIO | PCATCH, "vndlck", 0)) != 0)
1978 			return error;
1979 	}
1980 	sc->sc_flags |= VNF_LOCKED;
1981 	return 0;
1982 }
1983 
1984 /*
1985  * Unlock and wake up any waiters.
1986  */
1987 static void
1988 vndunlock(struct vnd_softc *sc)
1989 {
1990 
1991 	sc->sc_flags &= ~VNF_LOCKED;
1992 	if ((sc->sc_flags & VNF_WANTED) != 0) {
1993 		sc->sc_flags &= ~VNF_WANTED;
1994 		wakeup(sc);
1995 	}
1996 }
1997 
1998 #ifdef VND_COMPRESSION
1999 /* compressed file read */
2000 static void
2001 compstrategy(struct buf *bp, off_t bn)
2002 {
2003 	int error;
2004 	int unit = vndunit(bp->b_dev);
2005 	struct vnd_softc *vnd =
2006 	    device_lookup_private(&vnd_cd, unit);
2007 	u_int32_t comp_block;
2008 	struct uio auio;
2009 	char *addr;
2010 	int s;
2011 
2012 	/* set up constants for data move */
2013 	auio.uio_rw = UIO_READ;
2014 	UIO_SETUP_SYSSPACE(&auio);
2015 
2016 	/* read, and transfer the data */
2017 	addr = bp->b_data;
2018 	bp->b_resid = bp->b_bcount;
2019 	s = splbio();
2020 	while (bp->b_resid > 0) {
2021 		unsigned length;
2022 		size_t length_in_buffer;
2023 		u_int32_t offset_in_buffer;
2024 		struct iovec aiov;
2025 
2026 		/* calculate the compressed block number */
2027 		comp_block = bn / (off_t)vnd->sc_comp_blksz;
2028 
2029 		/* check for good block number */
2030 		if (comp_block >= vnd->sc_comp_numoffs) {
2031 			bp->b_error = EINVAL;
2032 			splx(s);
2033 			return;
2034 		}
2035 
2036 		/* read in the compressed block, if not in buffer */
2037 		if (comp_block != vnd->sc_comp_buffblk) {
2038 			length = vnd->sc_comp_offsets[comp_block + 1] -
2039 			    vnd->sc_comp_offsets[comp_block];
2040 			vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
2041 			error = vn_rdwr(UIO_READ, vnd->sc_vp, vnd->sc_comp_buff,
2042 			    length, vnd->sc_comp_offsets[comp_block],
2043 			    UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, vnd->sc_cred,
2044 			    NULL, NULL);
2045 			if (error) {
2046 				bp->b_error = error;
2047 				VOP_UNLOCK(vnd->sc_vp);
2048 				splx(s);
2049 				return;
2050 			}
2051 			/* uncompress the buffer */
2052 			vnd->sc_comp_stream.next_in = vnd->sc_comp_buff;
2053 			vnd->sc_comp_stream.avail_in = length;
2054 			vnd->sc_comp_stream.next_out = vnd->sc_comp_decombuf;
2055 			vnd->sc_comp_stream.avail_out = vnd->sc_comp_blksz;
2056 			inflateReset(&vnd->sc_comp_stream);
2057 			error = inflate(&vnd->sc_comp_stream, Z_FINISH);
2058 			if (error != Z_STREAM_END) {
2059 				if (vnd->sc_comp_stream.msg)
2060 					aprint_normal_dev(vnd->sc_dev,
2061 					    "compressed file, %s\n",
2062 					    vnd->sc_comp_stream.msg);
2063 				bp->b_error = EBADMSG;
2064 				VOP_UNLOCK(vnd->sc_vp);
2065 				splx(s);
2066 				return;
2067 			}
2068 			vnd->sc_comp_buffblk = comp_block;
2069 			VOP_UNLOCK(vnd->sc_vp);
2070 		}
2071 
2072 		/* transfer the usable uncompressed data */
2073 		offset_in_buffer = bn % (off_t)vnd->sc_comp_blksz;
2074 		length_in_buffer = vnd->sc_comp_blksz - offset_in_buffer;
2075 		if (length_in_buffer > bp->b_resid)
2076 			length_in_buffer = bp->b_resid;
2077 		auio.uio_iov = &aiov;
2078 		auio.uio_iovcnt = 1;
2079 		aiov.iov_base = addr;
2080 		aiov.iov_len = length_in_buffer;
2081 		auio.uio_resid = aiov.iov_len;
2082 		auio.uio_offset = 0;
2083 		error = uiomove(vnd->sc_comp_decombuf + offset_in_buffer,
2084 		    length_in_buffer, &auio);
2085 		if (error) {
2086 			bp->b_error = error;
2087 			splx(s);
2088 			return;
2089 		}
2090 
2091 		bn += length_in_buffer;
2092 		addr += length_in_buffer;
2093 		bp->b_resid -= length_in_buffer;
2094 	}
2095 	splx(s);
2096 }
2097 
2098 /* compression memory allocation routines */
2099 static void *
2100 vnd_alloc(void *aux, u_int items, u_int siz)
2101 {
2102 	return malloc(items * siz, M_TEMP, M_NOWAIT);
2103 }
2104 
2105 static void
2106 vnd_free(void *aux, void *ptr)
2107 {
2108 	free(ptr, M_TEMP);
2109 }
2110 #endif /* VND_COMPRESSION */
2111 
2112 static void
2113 vnd_set_geometry(struct vnd_softc *vnd)
2114 {
2115 	struct disk_geom *dg = &vnd->sc_dkdev.dk_geom;
2116 	unsigned spb;
2117 
2118 	memset(dg, 0, sizeof(*dg));
2119 
2120 	spb = vnd->sc_geom.vng_secsize / DEV_BSIZE;
2121 	dg->dg_secperunit = vnd->sc_size / spb;
2122 	dg->dg_secsize = vnd->sc_geom.vng_secsize;
2123 	dg->dg_nsectors = vnd->sc_geom.vng_nsectors;
2124 	dg->dg_ntracks = vnd->sc_geom.vng_ntracks;
2125 	dg->dg_ncylinders = vnd->sc_geom.vng_ncylinders;
2126 
2127 #ifdef DEBUG
2128 	if (vnddebug & VDB_LABEL) {
2129 		printf("dg->dg_secperunit: %" PRId64 "\n", dg->dg_secperunit);
2130 		printf("dg->dg_ncylinders: %u\n", dg->dg_ncylinders);
2131 	}
2132 #endif
2133 	disk_set_info(vnd->sc_dev, &vnd->sc_dkdev, NULL);
2134 }
2135 
2136 #ifdef VND_COMPRESSION
2137 #define VND_DEPENDS "zlib"
2138 #else
2139 #define VND_DEPENDS NULL
2140 #endif
2141 
2142 MODULE(MODULE_CLASS_DRIVER, vnd, VND_DEPENDS);
2143 
2144 #ifdef _MODULE
2145 int vnd_bmajor = -1, vnd_cmajor = -1;
2146 
2147 CFDRIVER_DECL(vnd, DV_DISK, NULL);
2148 #endif
2149 
2150 static int
2151 vnd_modcmd(modcmd_t cmd, void *arg)
2152 {
2153 	int error = 0;
2154 
2155 	switch (cmd) {
2156 	case MODULE_CMD_INIT:
2157 #ifdef _MODULE
2158 		error = config_cfdriver_attach(&vnd_cd);
2159 		if (error)
2160 			break;
2161 
2162 		error = config_cfattach_attach(vnd_cd.cd_name, &vnd_ca);
2163 	        if (error) {
2164 			config_cfdriver_detach(&vnd_cd);
2165 #ifdef DIAGNOSTIC
2166 			aprint_error("%s: unable to register cfattach for \n"
2167 			    "%s, error %d", __func__, vnd_cd.cd_name, error);
2168 #endif
2169 			break;
2170 		}
2171 
2172                 /*
2173                  * Attach the {b,c}devsw's
2174                  */
2175 		error = devsw_attach("vnd", &vnd_bdevsw, &vnd_bmajor,
2176 		    &vnd_cdevsw, &vnd_cmajor);
2177                 /*
2178                  * If devsw_attach fails, remove from autoconf database
2179                  */
2180 		if (error) {
2181 			config_cfattach_detach(vnd_cd.cd_name, &vnd_ca);
2182 			config_cfdriver_detach(&vnd_cd);
2183 #ifdef DIAGNOSTIC
2184                         aprint_error("%s: unable to attach %s devsw, "
2185                             "error %d", __func__, vnd_cd.cd_name, error);
2186 #endif
2187 			break;
2188 		}
2189 #endif
2190 		break;
2191 
2192 	case MODULE_CMD_FINI:
2193 #ifdef _MODULE
2194                 /*
2195                  * Remove {b,c}devsw's
2196                  */
2197 		devsw_detach(&vnd_bdevsw, &vnd_cdevsw);
2198 
2199                 /*
2200                  * Now remove device from autoconf database
2201                  */
2202 		error = config_cfattach_detach(vnd_cd.cd_name, &vnd_ca);
2203                 if (error) {
2204                         (void)devsw_attach("vnd", &vnd_bdevsw, &vnd_bmajor,
2205                             &vnd_cdevsw, &vnd_cmajor);
2206 #ifdef DIAGNOSTIC
2207                         aprint_error("%s: failed to detach %s cfattach, "
2208                             "error %d\n", __func__, vnd_cd.cd_name, error);
2209 #endif
2210                         break;
2211                 }
2212                 error = config_cfdriver_detach(&vnd_cd);
2213                 if (error) {
2214                         (void)config_cfattach_attach(vnd_cd.cd_name, &vnd_ca);
2215                         (void)devsw_attach("vnd", &vnd_bdevsw, &vnd_bmajor,
2216                             &vnd_cdevsw, &vnd_cmajor);
2217 #ifdef DIAGNOSTIC
2218                         aprint_error("%s: failed to detach %s cfdriver, "
2219                             "error %d\n", __func__, vnd_cd.cd_name, error);
2220                         break;
2221 #endif
2222                 }
2223 #endif
2224 		break;
2225 
2226 	case MODULE_CMD_STAT:
2227 		return ENOTTY;
2228 
2229 	default:
2230 		return ENOTTY;
2231 	}
2232 
2233 	return error;
2234 }
2235