Home | History | Annotate | Line # | Download | only in dev
vnd.c revision 1.249
      1 /*	$NetBSD: vnd.c,v 1.249 2015/11/09 16:52:09 christos Exp $	*/
      2 
      3 /*-
      4  * Copyright (c) 1996, 1997, 1998, 2008 The NetBSD Foundation, Inc.
      5  * All rights reserved.
      6  *
      7  * This code is derived from software contributed to The NetBSD Foundation
      8  * by Jason R. Thorpe.
      9  *
     10  * Redistribution and use in source and binary forms, with or without
     11  * modification, are permitted provided that the following conditions
     12  * are met:
     13  * 1. Redistributions of source code must retain the above copyright
     14  *    notice, this list of conditions and the following disclaimer.
     15  * 2. Redistributions in binary form must reproduce the above copyright
     16  *    notice, this list of conditions and the following disclaimer in the
     17  *    documentation and/or other materials provided with the distribution.
     18  *
     19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     29  * POSSIBILITY OF SUCH DAMAGE.
     30  */
     31 
     32 /*
     33  * Copyright (c) 1988 University of Utah.
     34  * Copyright (c) 1990, 1993
     35  *	The Regents of the University of California.  All rights reserved.
     36  *
     37  * This code is derived from software contributed to Berkeley by
     38  * the Systems Programming Group of the University of Utah Computer
     39  * Science Department.
     40  *
     41  * Redistribution and use in source and binary forms, with or without
     42  * modification, are permitted provided that the following conditions
     43  * are met:
     44  * 1. Redistributions of source code must retain the above copyright
     45  *    notice, this list of conditions and the following disclaimer.
     46  * 2. Redistributions in binary form must reproduce the above copyright
     47  *    notice, this list of conditions and the following disclaimer in the
     48  *    documentation and/or other materials provided with the distribution.
     49  * 3. Neither the name of the University nor the names of its contributors
     50  *    may be used to endorse or promote products derived from this software
     51  *    without specific prior written permission.
     52  *
     53  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     54  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     55  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     56  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     57  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     58  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     59  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     60  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     61  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     62  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     63  * SUCH DAMAGE.
     64  *
     65  * from: Utah $Hdr: vn.c 1.13 94/04/02$
     66  *
     67  *	@(#)vn.c	8.9 (Berkeley) 5/14/95
     68  */
     69 
     70 /*
     71  * Vnode disk driver.
     72  *
     73  * Block/character interface to a vnode.  Allows one to treat a file
     74  * as a disk (e.g. build a filesystem in it, mount it, etc.).
     75  *
     76  * NOTE 1: If the vnode supports the VOP_BMAP and VOP_STRATEGY operations,
     77  * this uses them to avoid distorting the local buffer cache.  If those
     78  * block-level operations are not available, this falls back to the regular
     79  * read and write calls.  Using these may distort the cache in some cases
     80  * but better have the driver working than preventing it to work on file
     81  * systems where the block-level operations are not implemented for
     82  * whatever reason.
     83  *
     84  * NOTE 2: There is a security issue involved with this driver.
     85  * Once mounted all access to the contents of the "mapped" file via
     86  * the special file is controlled by the permissions on the special
     87  * file, the protection of the mapped file is ignored (effectively,
     88  * by using root credentials in all transactions).
     89  *
     90  * NOTE 3: Doesn't interact with leases, should it?
     91  */
     92 
     93 #include <sys/cdefs.h>
     94 __KERNEL_RCSID(0, "$NetBSD: vnd.c,v 1.249 2015/11/09 16:52:09 christos Exp $");
     95 
     96 #if defined(_KERNEL_OPT)
     97 #include "opt_vnd.h"
     98 #include "opt_compat_netbsd.h"
     99 #endif
    100 
    101 #define DEBUG
    102 #include <sys/param.h>
    103 #include <sys/systm.h>
    104 #include <sys/namei.h>
    105 #include <sys/proc.h>
    106 #include <sys/kthread.h>
    107 #include <sys/errno.h>
    108 #include <sys/buf.h>
    109 #include <sys/bufq.h>
    110 #include <sys/malloc.h>
    111 #include <sys/ioctl.h>
    112 #include <sys/disklabel.h>
    113 #include <sys/device.h>
    114 #include <sys/disk.h>
    115 #include <sys/stat.h>
    116 #include <sys/mount.h>
    117 #include <sys/vnode.h>
    118 #include <sys/file.h>
    119 #include <sys/uio.h>
    120 #include <sys/conf.h>
    121 #include <sys/kauth.h>
    122 
    123 #include <net/zlib.h>
    124 
    125 #include <miscfs/genfs/genfs.h>
    126 #include <miscfs/specfs/specdev.h>
    127 
    128 #include <dev/dkvar.h>
    129 #include <dev/vndvar.h>
    130 
    131 #include "ioconf.h"
    132 
    133 #if defined(VNDDEBUG) && !defined(DEBUG)
    134 #define DEBUG
    135 #endif
    136 
    137 #ifdef DEBUG
    138 int dovndcluster = 1;
    139 #define VDB_FOLLOW	0x01
    140 #define VDB_INIT	0x02
    141 #define VDB_IO		0x04
    142 #define VDB_LABEL	0x08
    143 int vnddebug = 0xff;
    144 #endif
    145 
    146 #define vndunit(x)	DISKUNIT(x)
    147 
    148 struct vndxfer {
    149 	struct buf vx_buf;
    150 	struct vnd_softc *vx_vnd;
    151 };
    152 #define	VND_BUFTOXFER(bp)	((struct vndxfer *)(void *)bp)
    153 
    154 #define VND_GETXFER(vnd)	pool_get(&(vnd)->sc_vxpool, PR_WAITOK)
    155 #define VND_PUTXFER(vnd, vx)	pool_put(&(vnd)->sc_vxpool, (vx))
    156 
    157 #define VNDLABELDEV(dev) \
    158     (MAKEDISKDEV(major((dev)), vndunit((dev)), RAW_PART))
    159 
    160 #define	VND_MAXPENDING(vnd)	((vnd)->sc_maxactive * 4)
    161 
    162 
    163 static void	vndclear(struct vnd_softc *, int);
    164 static int	vnddoclear(struct vnd_softc *, int, int, bool);
    165 static int	vndsetcred(struct vnd_softc *, kauth_cred_t);
    166 static void	vndthrottle(struct vnd_softc *, struct vnode *);
    167 static void	vndiodone(struct buf *);
    168 #if 0
    169 static void	vndshutdown(void);
    170 #endif
    171 
    172 static void	vndgetdefaultlabel(struct vnd_softc *, struct disklabel *);
    173 static void	vndgetdisklabel(dev_t, struct vnd_softc *);
    174 
    175 static int	vndlock(struct vnd_softc *);
    176 static void	vndunlock(struct vnd_softc *);
    177 #ifdef VND_COMPRESSION
    178 static void	compstrategy(struct buf *, off_t);
    179 static void	*vnd_alloc(void *, u_int, u_int);
    180 static void	vnd_free(void *, void *);
    181 #endif /* VND_COMPRESSION */
    182 
    183 static void	vndthread(void *);
    184 static bool	vnode_has_op(const struct vnode *, int);
    185 static void	handle_with_rdwr(struct vnd_softc *, const struct buf *,
    186 		    struct buf *);
    187 static void	handle_with_strategy(struct vnd_softc *, const struct buf *,
    188 		    struct buf *);
    189 static void	vnd_set_geometry(struct vnd_softc *);
    190 
    191 static dev_type_open(vndopen);
    192 static dev_type_close(vndclose);
    193 static dev_type_read(vndread);
    194 static dev_type_write(vndwrite);
    195 static dev_type_ioctl(vndioctl);
    196 static dev_type_strategy(vndstrategy);
    197 static dev_type_dump(vnddump);
    198 static dev_type_size(vndsize);
    199 
    200 const struct bdevsw vnd_bdevsw = {
    201 	.d_open = vndopen,
    202 	.d_close = vndclose,
    203 	.d_strategy = vndstrategy,
    204 	.d_ioctl = vndioctl,
    205 	.d_dump = vnddump,
    206 	.d_psize = vndsize,
    207 	.d_discard = nodiscard,
    208 	.d_flag = D_DISK
    209 };
    210 
    211 const struct cdevsw vnd_cdevsw = {
    212 	.d_open = vndopen,
    213 	.d_close = vndclose,
    214 	.d_read = vndread,
    215 	.d_write = vndwrite,
    216 	.d_ioctl = vndioctl,
    217 	.d_stop = nostop,
    218 	.d_tty = notty,
    219 	.d_poll = nopoll,
    220 	.d_mmap = nommap,
    221 	.d_kqfilter = nokqfilter,
    222 	.d_discard = nodiscard,
    223 	.d_flag = D_DISK
    224 };
    225 
    226 static int	vnd_match(device_t, cfdata_t, void *);
    227 static void	vnd_attach(device_t, device_t, void *);
    228 static int	vnd_detach(device_t, int);
    229 
    230 CFATTACH_DECL3_NEW(vnd, sizeof(struct vnd_softc),
    231     vnd_match, vnd_attach, vnd_detach, NULL, NULL, NULL, DVF_DETACH_SHUTDOWN);
    232 extern struct cfdriver vnd_cd;
    233 
    234 static struct vnd_softc	*vnd_spawn(int);
    235 int	vnd_destroy(device_t);
    236 
    237 static struct	dkdriver vnddkdriver = {
    238 	.d_strategy = vndstrategy,
    239 	.d_minphys = minphys
    240 };
    241 
    242 void
    243 vndattach(int num)
    244 {
    245 	int error;
    246 
    247 	error = config_cfattach_attach(vnd_cd.cd_name, &vnd_ca);
    248 	if (error)
    249 		aprint_error("%s: unable to register cfattach, error = %d\n",
    250 		    vnd_cd.cd_name, error);
    251 }
    252 
    253 static int
    254 vnd_match(device_t self, cfdata_t cfdata, void *aux)
    255 {
    256 
    257 	return 1;
    258 }
    259 
    260 static void
    261 vnd_attach(device_t parent, device_t self, void *aux)
    262 {
    263 	struct vnd_softc *sc = device_private(self);
    264 
    265 	sc->sc_dev = self;
    266 	sc->sc_comp_offsets = NULL;
    267 	sc->sc_comp_buff = NULL;
    268 	sc->sc_comp_decombuf = NULL;
    269 	bufq_alloc(&sc->sc_tab, "disksort", BUFQ_SORT_RAWBLOCK);
    270 	disk_init(&sc->sc_dkdev, device_xname(self), &vnddkdriver);
    271 	if (!pmf_device_register(self, NULL, NULL))
    272 		aprint_error_dev(self, "couldn't establish power handler\n");
    273 }
    274 
    275 static int
    276 vnd_detach(device_t self, int flags)
    277 {
    278 	int error;
    279 	struct vnd_softc *sc = device_private(self);
    280 
    281 	if (sc->sc_flags & VNF_INITED) {
    282 		error = vnddoclear(sc, 0, -1, (flags & DETACH_FORCE) != 0);
    283 		if (error != 0)
    284 			return error;
    285 	}
    286 
    287 	pmf_device_deregister(self);
    288 	bufq_free(sc->sc_tab);
    289 	disk_destroy(&sc->sc_dkdev);
    290 
    291 	return 0;
    292 }
    293 
    294 static struct vnd_softc *
    295 vnd_spawn(int unit)
    296 {
    297 	cfdata_t cf;
    298 
    299 	cf = malloc(sizeof(*cf), M_DEVBUF, M_WAITOK);
    300 	cf->cf_name = vnd_cd.cd_name;
    301 	cf->cf_atname = vnd_cd.cd_name;
    302 	cf->cf_unit = unit;
    303 	cf->cf_fstate = FSTATE_STAR;
    304 
    305 	return device_private(config_attach_pseudo(cf));
    306 }
    307 
    308 int
    309 vnd_destroy(device_t dev)
    310 {
    311 	int error;
    312 	cfdata_t cf;
    313 
    314 	cf = device_cfdata(dev);
    315 	error = config_detach(dev, DETACH_QUIET);
    316 	if (error)
    317 		return error;
    318 	free(cf, M_DEVBUF);
    319 	return 0;
    320 }
    321 
    322 static int
    323 vndopen(dev_t dev, int flags, int mode, struct lwp *l)
    324 {
    325 	int unit = vndunit(dev);
    326 	struct vnd_softc *sc;
    327 	int error = 0, part, pmask;
    328 	struct disklabel *lp;
    329 
    330 #ifdef DEBUG
    331 	if (vnddebug & VDB_FOLLOW)
    332 		printf("vndopen(0x%"PRIx64", 0x%x, 0x%x, %p)\n", dev, flags, mode, l);
    333 #endif
    334 	sc = device_lookup_private(&vnd_cd, unit);
    335 	if (sc == NULL) {
    336 		sc = vnd_spawn(unit);
    337 		if (sc == NULL)
    338 			return ENOMEM;
    339 
    340 		/* compatibility, keep disklabel after close */
    341 		sc->sc_flags = VNF_KLABEL;
    342 	}
    343 
    344 	if ((error = vndlock(sc)) != 0)
    345 		return error;
    346 
    347 	mutex_enter(&sc->sc_dkdev.dk_openlock);
    348 
    349 	if ((sc->sc_flags & VNF_CLEARING) != 0) {
    350 		error = ENXIO;
    351 		goto done;
    352 	}
    353 
    354 	lp = sc->sc_dkdev.dk_label;
    355 
    356 	part = DISKPART(dev);
    357 	pmask = (1 << part);
    358 
    359 	if (sc->sc_dkdev.dk_nwedges != 0 && part != RAW_PART) {
    360 		error = EBUSY;
    361 		goto done;
    362 	}
    363 
    364 	if (sc->sc_flags & VNF_INITED) {
    365 		if ((sc->sc_dkdev.dk_openmask & ~(1<<RAW_PART)) != 0) {
    366 			/*
    367 			 * If any non-raw partition is open, but the disk
    368 			 * has been invalidated, disallow further opens.
    369 			 */
    370 			if ((sc->sc_flags & VNF_VLABEL) == 0) {
    371 				error = EIO;
    372 				goto done;
    373 			}
    374 		} else {
    375 			/*
    376 			 * Load the partition info if not already loaded.
    377 			 */
    378 			if ((sc->sc_flags & VNF_VLABEL) == 0) {
    379 				sc->sc_flags |= VNF_VLABEL;
    380 				vndgetdisklabel(dev, sc);
    381 			}
    382 		}
    383 	}
    384 
    385 	/* Check that the partitions exists. */
    386 	if (part != RAW_PART) {
    387 		if (((sc->sc_flags & VNF_INITED) == 0) ||
    388 		    ((part >= lp->d_npartitions) ||
    389 		     (lp->d_partitions[part].p_fstype == FS_UNUSED))) {
    390 			error = ENXIO;
    391 			goto done;
    392 		}
    393 	}
    394 
    395 	/* Prevent our unit from being unconfigured while open. */
    396 	switch (mode) {
    397 	case S_IFCHR:
    398 		sc->sc_dkdev.dk_copenmask |= pmask;
    399 		break;
    400 
    401 	case S_IFBLK:
    402 		sc->sc_dkdev.dk_bopenmask |= pmask;
    403 		break;
    404 	}
    405 	sc->sc_dkdev.dk_openmask =
    406 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
    407 
    408  done:
    409 	mutex_exit(&sc->sc_dkdev.dk_openlock);
    410 	vndunlock(sc);
    411 	return error;
    412 }
    413 
    414 static int
    415 vndclose(dev_t dev, int flags, int mode, struct lwp *l)
    416 {
    417 	int unit = vndunit(dev);
    418 	struct vnd_softc *sc;
    419 	int error = 0, part;
    420 
    421 #ifdef DEBUG
    422 	if (vnddebug & VDB_FOLLOW)
    423 		printf("vndclose(0x%"PRIx64", 0x%x, 0x%x, %p)\n", dev, flags, mode, l);
    424 #endif
    425 	sc = device_lookup_private(&vnd_cd, unit);
    426 	if (sc == NULL)
    427 		return ENXIO;
    428 
    429 	if ((error = vndlock(sc)) != 0)
    430 		return error;
    431 
    432 	mutex_enter(&sc->sc_dkdev.dk_openlock);
    433 
    434 	part = DISKPART(dev);
    435 
    436 	/* ...that much closer to allowing unconfiguration... */
    437 	switch (mode) {
    438 	case S_IFCHR:
    439 		sc->sc_dkdev.dk_copenmask &= ~(1 << part);
    440 		break;
    441 
    442 	case S_IFBLK:
    443 		sc->sc_dkdev.dk_bopenmask &= ~(1 << part);
    444 		break;
    445 	}
    446 	sc->sc_dkdev.dk_openmask =
    447 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
    448 
    449 	/* are we last opener ? */
    450 	if (sc->sc_dkdev.dk_openmask == 0) {
    451 		if ((sc->sc_flags & VNF_KLABEL) == 0)
    452 			sc->sc_flags &= ~VNF_VLABEL;
    453 	}
    454 
    455 	mutex_exit(&sc->sc_dkdev.dk_openlock);
    456 
    457 	vndunlock(sc);
    458 
    459 	if ((sc->sc_flags & VNF_INITED) == 0) {
    460 		if ((error = vnd_destroy(sc->sc_dev)) != 0) {
    461 			aprint_error_dev(sc->sc_dev,
    462 			    "unable to detach instance\n");
    463 			return error;
    464 		}
    465 	}
    466 
    467 	return 0;
    468 }
    469 
    470 /*
    471  * Queue the request, and wakeup the kernel thread to handle it.
    472  */
    473 static void
    474 vndstrategy(struct buf *bp)
    475 {
    476 	int unit = vndunit(bp->b_dev);
    477 	struct vnd_softc *vnd =
    478 	    device_lookup_private(&vnd_cd, unit);
    479 	struct disklabel *lp;
    480 	daddr_t blkno;
    481 	int s = splbio();
    482 
    483 	if (vnd == NULL) {
    484 		bp->b_error = ENXIO;
    485 		goto done;
    486 	}
    487 	lp = vnd->sc_dkdev.dk_label;
    488 
    489 	if ((vnd->sc_flags & VNF_INITED) == 0) {
    490 		bp->b_error = ENXIO;
    491 		goto done;
    492 	}
    493 
    494 	/*
    495 	 * The transfer must be a whole number of blocks.
    496 	 */
    497 	if ((bp->b_bcount % lp->d_secsize) != 0) {
    498 		bp->b_error = EINVAL;
    499 		goto done;
    500 	}
    501 
    502 	/*
    503 	 * check if we're read-only.
    504 	 */
    505 	if ((vnd->sc_flags & VNF_READONLY) && !(bp->b_flags & B_READ)) {
    506 		bp->b_error = EACCES;
    507 		goto done;
    508 	}
    509 
    510 	/* If it's a nil transfer, wake up the top half now. */
    511 	if (bp->b_bcount == 0) {
    512 		goto done;
    513 	}
    514 
    515 	/*
    516 	 * Do bounds checking and adjust transfer.  If there's an error,
    517 	 * the bounds check will flag that for us.
    518 	 */
    519 	if (DISKPART(bp->b_dev) == RAW_PART) {
    520 		if (bounds_check_with_mediasize(bp, DEV_BSIZE,
    521 		    vnd->sc_size) <= 0)
    522 			goto done;
    523 	} else {
    524 		if (bounds_check_with_label(&vnd->sc_dkdev,
    525 		    bp, vnd->sc_flags & (VNF_WLABEL|VNF_LABELLING)) <= 0)
    526 			goto done;
    527 	}
    528 
    529 	/*
    530 	 * Put the block number in terms of the logical blocksize
    531 	 * of the "device".
    532 	 */
    533 
    534 	blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
    535 
    536 	/*
    537 	 * Translate the partition-relative block number to an absolute.
    538 	 */
    539 	if (DISKPART(bp->b_dev) != RAW_PART) {
    540 		struct partition *pp;
    541 
    542 		pp = &vnd->sc_dkdev.dk_label->d_partitions[
    543 		    DISKPART(bp->b_dev)];
    544 		blkno += pp->p_offset;
    545 	}
    546 	bp->b_rawblkno = blkno;
    547 
    548 #ifdef DEBUG
    549 	if (vnddebug & VDB_FOLLOW)
    550 		printf("vndstrategy(%p): unit %d\n", bp, unit);
    551 #endif
    552 	if ((vnd->sc_flags & VNF_USE_VN_RDWR)) {
    553 		KASSERT(vnd->sc_pending >= 0 &&
    554 		    vnd->sc_pending <= VND_MAXPENDING(vnd));
    555 		while (vnd->sc_pending == VND_MAXPENDING(vnd))
    556 			tsleep(&vnd->sc_pending, PRIBIO, "vndpc", 0);
    557 		vnd->sc_pending++;
    558 	}
    559 	bufq_put(vnd->sc_tab, bp);
    560 	wakeup(&vnd->sc_tab);
    561 	splx(s);
    562 	return;
    563 
    564 done:
    565 	bp->b_resid = bp->b_bcount;
    566 	biodone(bp);
    567 	splx(s);
    568 }
    569 
    570 static bool
    571 vnode_has_strategy(struct vnd_softc *vnd)
    572 {
    573 	return vnode_has_op(vnd->sc_vp, VOFFSET(vop_bmap)) &&
    574 	    vnode_has_op(vnd->sc_vp, VOFFSET(vop_strategy));
    575 }
    576 
    577 /* XXX this function needs a reliable check to detect
    578  * sparse files. Otherwise, bmap/strategy may be used
    579  * and fail on non-allocated blocks. VOP_READ/VOP_WRITE
    580  * works on sparse files.
    581  */
    582 #if notyet
    583 static bool
    584 vnode_strategy_probe(struct vnd_softc *vnd)
    585 {
    586 	int error;
    587 	daddr_t nbn;
    588 
    589 	if (!vnode_has_strategy(vnd))
    590 		return false;
    591 
    592 	/* Convert the first logical block number to its
    593 	 * physical block number.
    594 	 */
    595 	error = 0;
    596 	vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
    597 	error = VOP_BMAP(vnd->sc_vp, 0, NULL, &nbn, NULL);
    598 	VOP_UNLOCK(vnd->sc_vp);
    599 
    600 	/* Test if that worked. */
    601 	if (error == 0 && (long)nbn == -1)
    602 		return false;
    603 
    604 	return true;
    605 }
    606 #endif
    607 
    608 static void
    609 vndthread(void *arg)
    610 {
    611 	struct vnd_softc *vnd = arg;
    612 	int s;
    613 
    614 	/* Determine whether we can *use* VOP_BMAP and VOP_STRATEGY to
    615 	 * directly access the backing vnode.  If we can, use these two
    616 	 * operations to avoid messing with the local buffer cache.
    617 	 * Otherwise fall back to regular VOP_READ/VOP_WRITE operations
    618 	 * which are guaranteed to work with any file system. */
    619 	if ((vnd->sc_flags & VNF_USE_VN_RDWR) == 0 &&
    620 	    ! vnode_has_strategy(vnd))
    621 		vnd->sc_flags |= VNF_USE_VN_RDWR;
    622 
    623 #ifdef DEBUG
    624 	if (vnddebug & VDB_INIT)
    625 		printf("vndthread: vp %p, %s\n", vnd->sc_vp,
    626 		    (vnd->sc_flags & VNF_USE_VN_RDWR) == 0 ?
    627 		    "using bmap/strategy operations" :
    628 		    "using read/write operations");
    629 #endif
    630 
    631 	s = splbio();
    632 	vnd->sc_flags |= VNF_KTHREAD;
    633 	wakeup(&vnd->sc_kthread);
    634 
    635 	/*
    636 	 * Dequeue requests and serve them depending on the available
    637 	 * vnode operations.
    638 	 */
    639 	while ((vnd->sc_flags & VNF_VUNCONF) == 0) {
    640 		struct vndxfer *vnx;
    641 		struct buf *obp;
    642 		struct buf *bp;
    643 
    644 		obp = bufq_get(vnd->sc_tab);
    645 		if (obp == NULL) {
    646 			tsleep(&vnd->sc_tab, PRIBIO, "vndbp", 0);
    647 			continue;
    648 		};
    649 		if ((vnd->sc_flags & VNF_USE_VN_RDWR)) {
    650 			KASSERT(vnd->sc_pending > 0 &&
    651 			    vnd->sc_pending <= VND_MAXPENDING(vnd));
    652 			if (vnd->sc_pending-- == VND_MAXPENDING(vnd))
    653 				wakeup(&vnd->sc_pending);
    654 		}
    655 		splx(s);
    656 #ifdef DEBUG
    657 		if (vnddebug & VDB_FOLLOW)
    658 			printf("vndthread(%p)\n", obp);
    659 #endif
    660 
    661 		if (vnd->sc_vp->v_mount == NULL) {
    662 			obp->b_error = ENXIO;
    663 			goto done;
    664 		}
    665 #ifdef VND_COMPRESSION
    666 		/* handle a compressed read */
    667 		if ((obp->b_flags & B_READ) != 0 && (vnd->sc_flags & VNF_COMP)) {
    668 			off_t bn;
    669 
    670 			/* Convert to a byte offset within the file. */
    671 			bn = obp->b_rawblkno *
    672 			    vnd->sc_dkdev.dk_label->d_secsize;
    673 
    674 			compstrategy(obp, bn);
    675 			goto done;
    676 		}
    677 #endif /* VND_COMPRESSION */
    678 
    679 		/*
    680 		 * Allocate a header for this transfer and link it to the
    681 		 * buffer
    682 		 */
    683 		s = splbio();
    684 		vnx = VND_GETXFER(vnd);
    685 		splx(s);
    686 		vnx->vx_vnd = vnd;
    687 
    688 		s = splbio();
    689 		while (vnd->sc_active >= vnd->sc_maxactive) {
    690 			tsleep(&vnd->sc_tab, PRIBIO, "vndac", 0);
    691 		}
    692 		vnd->sc_active++;
    693 		splx(s);
    694 
    695 		/* Instrumentation. */
    696 		disk_busy(&vnd->sc_dkdev);
    697 
    698 		bp = &vnx->vx_buf;
    699 		buf_init(bp);
    700 		bp->b_flags = (obp->b_flags & B_READ);
    701 		bp->b_oflags = obp->b_oflags;
    702 		bp->b_cflags = obp->b_cflags;
    703 		bp->b_iodone = vndiodone;
    704 		bp->b_private = obp;
    705 		bp->b_vp = vnd->sc_vp;
    706 		bp->b_objlock = bp->b_vp->v_interlock;
    707 		bp->b_data = obp->b_data;
    708 		bp->b_bcount = obp->b_bcount;
    709 		BIO_COPYPRIO(bp, obp);
    710 
    711 		/* Handle the request using the appropriate operations. */
    712 		if ((vnd->sc_flags & VNF_USE_VN_RDWR) == 0)
    713 			handle_with_strategy(vnd, obp, bp);
    714 		else
    715 			handle_with_rdwr(vnd, obp, bp);
    716 
    717 		s = splbio();
    718 		continue;
    719 
    720 done:
    721 		biodone(obp);
    722 		s = splbio();
    723 	}
    724 
    725 	vnd->sc_flags &= (~VNF_KTHREAD | VNF_VUNCONF);
    726 	wakeup(&vnd->sc_kthread);
    727 	splx(s);
    728 	kthread_exit(0);
    729 }
    730 
    731 /*
    732  * Checks if the given vnode supports the requested operation.
    733  * The operation is specified the offset returned by VOFFSET.
    734  *
    735  * XXX The test below used to determine this is quite fragile
    736  * because it relies on the file system to use genfs to specify
    737  * unimplemented operations.  There might be another way to do
    738  * it more cleanly.
    739  */
    740 static bool
    741 vnode_has_op(const struct vnode *vp, int opoffset)
    742 {
    743 	int (*defaultp)(void *);
    744 	int (*opp)(void *);
    745 
    746 	defaultp = vp->v_op[VOFFSET(vop_default)];
    747 	opp = vp->v_op[opoffset];
    748 
    749 	return opp != defaultp && opp != genfs_eopnotsupp &&
    750 	    opp != genfs_badop && opp != genfs_nullop;
    751 }
    752 
    753 /*
    754  * Handles the read/write request given in 'bp' using the vnode's VOP_READ
    755  * and VOP_WRITE operations.
    756  *
    757  * 'obp' is a pointer to the original request fed to the vnd device.
    758  */
    759 static void
    760 handle_with_rdwr(struct vnd_softc *vnd, const struct buf *obp, struct buf *bp)
    761 {
    762 	bool doread;
    763 	off_t offset;
    764 	size_t len, resid;
    765 	struct vnode *vp;
    766 
    767 	doread = bp->b_flags & B_READ;
    768 	offset = obp->b_rawblkno * vnd->sc_dkdev.dk_label->d_secsize;
    769 	len = bp->b_bcount;
    770 	vp = vnd->sc_vp;
    771 
    772 #if defined(DEBUG)
    773 	if (vnddebug & VDB_IO)
    774 		printf("vnd (rdwr): vp %p, %s, rawblkno 0x%" PRIx64
    775 		    ", secsize %d, offset %" PRIu64
    776 		    ", bcount %d\n",
    777 		    vp, doread ? "read" : "write", obp->b_rawblkno,
    778 		    vnd->sc_dkdev.dk_label->d_secsize, offset,
    779 		    bp->b_bcount);
    780 #endif
    781 
    782 	/* Issue the read or write operation. */
    783 	bp->b_error =
    784 	    vn_rdwr(doread ? UIO_READ : UIO_WRITE,
    785 	    vp, bp->b_data, len, offset, UIO_SYSSPACE,
    786 	    IO_ADV_ENCODE(POSIX_FADV_NOREUSE), vnd->sc_cred, &resid, NULL);
    787 	bp->b_resid = resid;
    788 
    789 	mutex_enter(vp->v_interlock);
    790 	(void) VOP_PUTPAGES(vp, 0, 0,
    791 	    PGO_ALLPAGES | PGO_CLEANIT | PGO_FREE | PGO_SYNCIO);
    792 
    793 	/* We need to increase the number of outputs on the vnode if
    794 	 * there was any write to it. */
    795 	if (!doread) {
    796 		mutex_enter(vp->v_interlock);
    797 		vp->v_numoutput++;
    798 		mutex_exit(vp->v_interlock);
    799 	}
    800 
    801 	biodone(bp);
    802 }
    803 
    804 /*
    805  * Handes the read/write request given in 'bp' using the vnode's VOP_BMAP
    806  * and VOP_STRATEGY operations.
    807  *
    808  * 'obp' is a pointer to the original request fed to the vnd device.
    809  */
    810 static void
    811 handle_with_strategy(struct vnd_softc *vnd, const struct buf *obp,
    812     struct buf *bp)
    813 {
    814 	int bsize, error, flags, skipped;
    815 	size_t resid, sz;
    816 	off_t bn, offset;
    817 	struct vnode *vp;
    818 	struct buf *nbp = NULL;
    819 
    820 	flags = obp->b_flags;
    821 
    822 
    823 	/* convert to a byte offset within the file. */
    824 	bn = obp->b_rawblkno * vnd->sc_dkdev.dk_label->d_secsize;
    825 
    826 	bsize = vnd->sc_vp->v_mount->mnt_stat.f_iosize;
    827 	skipped = 0;
    828 
    829 	/*
    830 	 * Break the request into bsize pieces and feed them
    831 	 * sequentially using VOP_BMAP/VOP_STRATEGY.
    832 	 * We do it this way to keep from flooding NFS servers if we
    833 	 * are connected to an NFS file.  This places the burden on
    834 	 * the client rather than the server.
    835 	 */
    836 	error = 0;
    837 	bp->b_resid = bp->b_bcount;
    838 	for (offset = 0, resid = bp->b_resid; /* true */;
    839 	    resid -= sz, offset += sz) {
    840 		daddr_t nbn;
    841 		int off, nra;
    842 
    843 		nra = 0;
    844 		vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
    845 		error = VOP_BMAP(vnd->sc_vp, bn / bsize, &vp, &nbn, &nra);
    846 		VOP_UNLOCK(vnd->sc_vp);
    847 
    848 		if (error == 0 && (long)nbn == -1)
    849 			error = EIO;
    850 
    851 		/*
    852 		 * If there was an error or a hole in the file...punt.
    853 		 * Note that we may have to wait for any operations
    854 		 * that we have already fired off before releasing
    855 		 * the buffer.
    856 		 *
    857 		 * XXX we could deal with holes here but it would be
    858 		 * a hassle (in the write case).
    859 		 */
    860 		if (error) {
    861 			skipped += resid;
    862 			break;
    863 		}
    864 
    865 #ifdef DEBUG
    866 		if (!dovndcluster)
    867 			nra = 0;
    868 #endif
    869 
    870 		off = bn % bsize;
    871 		sz = MIN(((off_t)1 + nra) * bsize - off, resid);
    872 #ifdef	DEBUG
    873 		if (vnddebug & VDB_IO)
    874 			printf("vndstrategy: vp %p/%p bn 0x%qx/0x%" PRIx64
    875 			    " sz 0x%zx\n", vnd->sc_vp, vp, (long long)bn,
    876 			    nbn, sz);
    877 #endif
    878 
    879 		nbp = getiobuf(vp, true);
    880 		nestiobuf_setup(bp, nbp, offset, sz);
    881 		nbp->b_blkno = nbn + btodb(off);
    882 
    883 #if 0 /* XXX #ifdef DEBUG */
    884 		if (vnddebug & VDB_IO)
    885 			printf("vndstart(%ld): bp %p vp %p blkno "
    886 			    "0x%" PRIx64 " flags %x addr %p cnt 0x%x\n",
    887 			    (long) (vnd-vnd_softc), &nbp->vb_buf,
    888 			    nbp->vb_buf.b_vp, nbp->vb_buf.b_blkno,
    889 			    nbp->vb_buf.b_flags, nbp->vb_buf.b_data,
    890 			    nbp->vb_buf.b_bcount);
    891 #endif
    892 		if (resid == sz) {
    893 			break;
    894 		}
    895 		VOP_STRATEGY(vp, nbp);
    896 		bn += sz;
    897 	}
    898 	if (!(flags & B_READ)) {
    899 		struct vnode *w_vp;
    900 		/*
    901 		 * this is the last nested buf, account for
    902 		 * the parent buf write too.
    903 		 * This has to be done last, so that
    904 		 * fsync won't wait for this write which
    905 		 * has no chance to complete before all nested bufs
    906 		 * have been queued. But it has to be done
    907 		 * before the last VOP_STRATEGY()
    908 		 * or the call to nestiobuf_done().
    909 		 */
    910 		w_vp = bp->b_vp;
    911 		mutex_enter(w_vp->v_interlock);
    912 		w_vp->v_numoutput++;
    913 		mutex_exit(w_vp->v_interlock);
    914 	}
    915 	KASSERT(skipped != 0 || nbp != NULL);
    916 	if (skipped)
    917 		nestiobuf_done(bp, skipped, error);
    918 	else
    919 		VOP_STRATEGY(vp, nbp);
    920 }
    921 
    922 static void
    923 vndiodone(struct buf *bp)
    924 {
    925 	struct vndxfer *vnx = VND_BUFTOXFER(bp);
    926 	struct vnd_softc *vnd = vnx->vx_vnd;
    927 	struct buf *obp = bp->b_private;
    928 	int s = splbio();
    929 
    930 	KASSERT(&vnx->vx_buf == bp);
    931 	KASSERT(vnd->sc_active > 0);
    932 #ifdef DEBUG
    933 	if (vnddebug & VDB_IO) {
    934 		printf("vndiodone1: bp %p iodone: error %d\n",
    935 		    bp, bp->b_error);
    936 	}
    937 #endif
    938 	disk_unbusy(&vnd->sc_dkdev, bp->b_bcount - bp->b_resid,
    939 	    (bp->b_flags & B_READ));
    940 	vnd->sc_active--;
    941 	if (vnd->sc_active == 0) {
    942 		wakeup(&vnd->sc_tab);
    943 	}
    944 	splx(s);
    945 	obp->b_error = bp->b_error;
    946 	obp->b_resid = bp->b_resid;
    947 	buf_destroy(bp);
    948 	VND_PUTXFER(vnd, vnx);
    949 	biodone(obp);
    950 }
    951 
    952 /* ARGSUSED */
    953 static int
    954 vndread(dev_t dev, struct uio *uio, int flags)
    955 {
    956 	int unit = vndunit(dev);
    957 	struct vnd_softc *sc;
    958 
    959 #ifdef DEBUG
    960 	if (vnddebug & VDB_FOLLOW)
    961 		printf("vndread(0x%"PRIx64", %p)\n", dev, uio);
    962 #endif
    963 
    964 	sc = device_lookup_private(&vnd_cd, unit);
    965 	if (sc == NULL)
    966 		return ENXIO;
    967 
    968 	if ((sc->sc_flags & VNF_INITED) == 0)
    969 		return ENXIO;
    970 
    971 	return physio(vndstrategy, NULL, dev, B_READ, minphys, uio);
    972 }
    973 
    974 /* ARGSUSED */
    975 static int
    976 vndwrite(dev_t dev, struct uio *uio, int flags)
    977 {
    978 	int unit = vndunit(dev);
    979 	struct vnd_softc *sc;
    980 
    981 #ifdef DEBUG
    982 	if (vnddebug & VDB_FOLLOW)
    983 		printf("vndwrite(0x%"PRIx64", %p)\n", dev, uio);
    984 #endif
    985 
    986 	sc = device_lookup_private(&vnd_cd, unit);
    987 	if (sc == NULL)
    988 		return ENXIO;
    989 
    990 	if ((sc->sc_flags & VNF_INITED) == 0)
    991 		return ENXIO;
    992 
    993 	return physio(vndstrategy, NULL, dev, B_WRITE, minphys, uio);
    994 }
    995 
    996 static int
    997 vnd_cget(struct lwp *l, int unit, int *un, struct vattr *va)
    998 {
    999 	int error;
   1000 	struct vnd_softc *vnd;
   1001 
   1002 	if (*un == -1)
   1003 		*un = unit;
   1004 	if (*un < 0)
   1005 		return EINVAL;
   1006 
   1007 	vnd = device_lookup_private(&vnd_cd, *un);
   1008 	if (vnd == NULL)
   1009 		return -1;
   1010 
   1011 	if ((vnd->sc_flags & VNF_INITED) == 0)
   1012 		return -1;
   1013 
   1014 	vn_lock(vnd->sc_vp, LK_SHARED | LK_RETRY);
   1015 	error = VOP_GETATTR(vnd->sc_vp, va, l->l_cred);
   1016 	VOP_UNLOCK(vnd->sc_vp);
   1017 	return error;
   1018 }
   1019 
   1020 static int
   1021 vnddoclear(struct vnd_softc *vnd, int pmask, int minor, bool force)
   1022 {
   1023 	int error;
   1024 
   1025 	if ((error = vndlock(vnd)) != 0)
   1026 		return error;
   1027 
   1028 	/*
   1029 	 * Don't unconfigure if any other partitions are open
   1030 	 * or if both the character and block flavors of this
   1031 	 * partition are open.
   1032 	 */
   1033 	if (DK_BUSY(vnd, pmask) && !force) {
   1034 		vndunlock(vnd);
   1035 		return EBUSY;
   1036 	}
   1037 
   1038 	/* Delete all of our wedges */
   1039 	dkwedge_delall(&vnd->sc_dkdev);
   1040 
   1041 	/*
   1042 	 * XXX vndclear() might call vndclose() implicitly;
   1043 	 * release lock to avoid recursion
   1044 	 *
   1045 	 * Set VNF_CLEARING to prevent vndopen() from
   1046 	 * sneaking in after we vndunlock().
   1047 	 */
   1048 	vnd->sc_flags |= VNF_CLEARING;
   1049 	vndunlock(vnd);
   1050 	vndclear(vnd, minor);
   1051 #ifdef DEBUG
   1052 	if (vnddebug & VDB_INIT)
   1053 		printf("%s: CLRed\n", __func__);
   1054 #endif
   1055 
   1056 	/* Destroy the xfer and buffer pools. */
   1057 	pool_destroy(&vnd->sc_vxpool);
   1058 
   1059 	/* Detach the disk. */
   1060 	disk_detach(&vnd->sc_dkdev);
   1061 
   1062 	return 0;
   1063 }
   1064 
   1065 static int
   1066 vndioctl_get(struct lwp *l, void *data, int unit, struct vattr *va)
   1067 {
   1068 	int error;
   1069 
   1070 	KASSERT(l);
   1071 #ifdef notyet
   1072 	/* Current userland code does not handle this */
   1073 	if (*(int *)data >= vnd_cd.cd_ndevs)
   1074 		return ENXIO;
   1075 #endif
   1076 
   1077 	switch (error = vnd_cget(l, unit, (int *)data, va)) {
   1078 	case -1:
   1079 		/* unused is not an error */
   1080 		memset(&va, 0, sizeof(va));
   1081 		/*FALLTHROUGH*/
   1082 	case 0:
   1083 		return 0;
   1084 	default:
   1085 		return error;
   1086 	}
   1087 }
   1088 
   1089 /* ARGSUSED */
   1090 static int
   1091 vndioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l)
   1092 {
   1093 	bool force;
   1094 	int unit = vndunit(dev);
   1095 	struct vnd_softc *vnd;
   1096 	struct vnd_ioctl *vio;
   1097 	struct vattr vattr;
   1098 	struct pathbuf *pb;
   1099 	struct nameidata nd;
   1100 	int error, part, pmask;
   1101 	uint64_t geomsize;
   1102 	int fflags;
   1103 #ifdef __HAVE_OLD_DISKLABEL
   1104 	struct disklabel newlabel;
   1105 #endif
   1106 
   1107 #ifdef DEBUG
   1108 	if (vnddebug & VDB_FOLLOW)
   1109 		printf("vndioctl(0x%"PRIx64", 0x%lx, %p, 0x%x, %p): unit %d\n",
   1110 		    dev, cmd, data, flag, l->l_proc, unit);
   1111 #endif
   1112 	/* Do the get's first; they don't need initialization or verification */
   1113 	switch (cmd) {
   1114 #ifdef COMPAT_30
   1115 	case VNDIOCGET30: {
   1116 		if ((error = vndioctl_get(l, data, unit, &vattr)) != 0)
   1117 			return error;
   1118 
   1119 		struct vnd_user30 *vnu = data;
   1120 		vnu->vnu_dev = vattr.va_fsid;
   1121 		vnu->vnu_ino = vattr.va_fileid;
   1122 		return 0;
   1123 	}
   1124 #endif
   1125 #ifdef COMPAT_50
   1126 	case VNDIOCGET50: {
   1127 		if ((error = vndioctl_get(l, data, unit, &vattr)) != 0)
   1128 			return error;
   1129 
   1130 		struct vnd_user50 *vnu = data;
   1131 		vnu->vnu_dev = vattr.va_fsid;
   1132 		vnu->vnu_ino = vattr.va_fileid;
   1133 		return 0;
   1134 	}
   1135 #endif
   1136 
   1137 	case VNDIOCGET: {
   1138 		if ((error = vndioctl_get(l, data, unit, &vattr)) != 0)
   1139 			return error;
   1140 
   1141 		struct vnd_user *vnu = data;
   1142 		vnu->vnu_dev = vattr.va_fsid;
   1143 		vnu->vnu_ino = vattr.va_fileid;
   1144 		return 0;
   1145 	}
   1146 	default:
   1147 		break;
   1148 	}
   1149 
   1150 	vnd = device_lookup_private(&vnd_cd, unit);
   1151 	if (vnd == NULL)
   1152 		return ENXIO;
   1153 	vio = (struct vnd_ioctl *)data;
   1154 
   1155 	/* Must be open for writes for these commands... */
   1156 	switch (cmd) {
   1157 	case VNDIOCSET:
   1158 	case VNDIOCCLR:
   1159 #ifdef COMPAT_50
   1160 	case VNDIOCSET50:
   1161 	case VNDIOCCLR50:
   1162 #endif
   1163 	case DIOCSDINFO:
   1164 	case DIOCWDINFO:
   1165 #ifdef __HAVE_OLD_DISKLABEL
   1166 	case ODIOCSDINFO:
   1167 	case ODIOCWDINFO:
   1168 #endif
   1169 	case DIOCKLABEL:
   1170 	case DIOCWLABEL:
   1171 		if ((flag & FWRITE) == 0)
   1172 			return EBADF;
   1173 	}
   1174 
   1175 	/* Must be initialized for these... */
   1176 	switch (cmd) {
   1177 	case VNDIOCCLR:
   1178 #ifdef VNDIOCCLR50
   1179 	case VNDIOCCLR50:
   1180 #endif
   1181 	case DIOCGDINFO:
   1182 	case DIOCSDINFO:
   1183 	case DIOCWDINFO:
   1184 	case DIOCGPART:
   1185 	case DIOCKLABEL:
   1186 	case DIOCWLABEL:
   1187 	case DIOCGDEFLABEL:
   1188 	case DIOCCACHESYNC:
   1189 #ifdef __HAVE_OLD_DISKLABEL
   1190 	case ODIOCGDINFO:
   1191 	case ODIOCSDINFO:
   1192 	case ODIOCWDINFO:
   1193 	case ODIOCGDEFLABEL:
   1194 #endif
   1195 		if ((vnd->sc_flags & VNF_INITED) == 0)
   1196 			return ENXIO;
   1197 	}
   1198 
   1199 	error = disk_ioctl(&vnd->sc_dkdev, dev, cmd, data, flag, l);
   1200 	if (error != EPASSTHROUGH)
   1201 		return error;
   1202 
   1203 
   1204 	switch (cmd) {
   1205 #ifdef VNDIOCSET50
   1206 	case VNDIOCSET50:
   1207 #endif
   1208 	case VNDIOCSET:
   1209 		if (vnd->sc_flags & VNF_INITED)
   1210 			return EBUSY;
   1211 
   1212 		if ((error = vndlock(vnd)) != 0)
   1213 			return error;
   1214 
   1215 		fflags = FREAD;
   1216 		if ((vio->vnd_flags & VNDIOF_READONLY) == 0)
   1217 			fflags |= FWRITE;
   1218 		error = pathbuf_copyin(vio->vnd_file, &pb);
   1219 		if (error) {
   1220 			goto unlock_and_exit;
   1221 		}
   1222 		NDINIT(&nd, LOOKUP, FOLLOW, pb);
   1223 		if ((error = vn_open(&nd, fflags, 0)) != 0) {
   1224 			pathbuf_destroy(pb);
   1225 			goto unlock_and_exit;
   1226 		}
   1227 		KASSERT(l);
   1228 		error = VOP_GETATTR(nd.ni_vp, &vattr, l->l_cred);
   1229 		if (!error && nd.ni_vp->v_type != VREG)
   1230 			error = EOPNOTSUPP;
   1231 		if (!error && vattr.va_bytes < vattr.va_size)
   1232 			/* File is definitely sparse, use vn_rdwr() */
   1233 			vnd->sc_flags |= VNF_USE_VN_RDWR;
   1234 		if (error) {
   1235 			VOP_UNLOCK(nd.ni_vp);
   1236 			goto close_and_exit;
   1237 		}
   1238 
   1239 		/* If using a compressed file, initialize its info */
   1240 		/* (or abort with an error if kernel has no compression) */
   1241 		if (vio->vnd_flags & VNF_COMP) {
   1242 #ifdef VND_COMPRESSION
   1243 			struct vnd_comp_header *ch;
   1244 			int i;
   1245 			u_int32_t comp_size;
   1246 			u_int32_t comp_maxsize;
   1247 
   1248 			/* allocate space for compresed file header */
   1249 			ch = malloc(sizeof(struct vnd_comp_header),
   1250 			M_TEMP, M_WAITOK);
   1251 
   1252 			/* read compressed file header */
   1253 			error = vn_rdwr(UIO_READ, nd.ni_vp, (void *)ch,
   1254 			  sizeof(struct vnd_comp_header), 0, UIO_SYSSPACE,
   1255 			  IO_UNIT|IO_NODELOCKED, l->l_cred, NULL, NULL);
   1256 			if (error) {
   1257 				free(ch, M_TEMP);
   1258 				VOP_UNLOCK(nd.ni_vp);
   1259 				goto close_and_exit;
   1260 			}
   1261 
   1262 			/* save some header info */
   1263 			vnd->sc_comp_blksz = ntohl(ch->block_size);
   1264 			/* note last offset is the file byte size */
   1265 			vnd->sc_comp_numoffs = ntohl(ch->num_blocks)+1;
   1266 			free(ch, M_TEMP);
   1267 			if (vnd->sc_comp_blksz == 0 ||
   1268 			    vnd->sc_comp_blksz % DEV_BSIZE !=0) {
   1269 				VOP_UNLOCK(nd.ni_vp);
   1270 				error = EINVAL;
   1271 				goto close_and_exit;
   1272 			}
   1273 			if (sizeof(struct vnd_comp_header) +
   1274 			  sizeof(u_int64_t) * vnd->sc_comp_numoffs >
   1275 			  vattr.va_size) {
   1276 				VOP_UNLOCK(nd.ni_vp);
   1277 				error = EINVAL;
   1278 				goto close_and_exit;
   1279 			}
   1280 
   1281 			/* set decompressed file size */
   1282 			vattr.va_size =
   1283 			    ((u_quad_t)vnd->sc_comp_numoffs - 1) *
   1284 			     (u_quad_t)vnd->sc_comp_blksz;
   1285 
   1286 			/* allocate space for all the compressed offsets */
   1287 			vnd->sc_comp_offsets =
   1288 			malloc(sizeof(u_int64_t) * vnd->sc_comp_numoffs,
   1289 			M_DEVBUF, M_WAITOK);
   1290 
   1291 			/* read in the offsets */
   1292 			error = vn_rdwr(UIO_READ, nd.ni_vp,
   1293 			  (void *)vnd->sc_comp_offsets,
   1294 			  sizeof(u_int64_t) * vnd->sc_comp_numoffs,
   1295 			  sizeof(struct vnd_comp_header), UIO_SYSSPACE,
   1296 			  IO_UNIT|IO_NODELOCKED, l->l_cred, NULL, NULL);
   1297 			if (error) {
   1298 				VOP_UNLOCK(nd.ni_vp);
   1299 				goto close_and_exit;
   1300 			}
   1301 			/*
   1302 			 * find largest block size (used for allocation limit).
   1303 			 * Also convert offset to native byte order.
   1304 			 */
   1305 			comp_maxsize = 0;
   1306 			for (i = 0; i < vnd->sc_comp_numoffs - 1; i++) {
   1307 				vnd->sc_comp_offsets[i] =
   1308 				  be64toh(vnd->sc_comp_offsets[i]);
   1309 				comp_size = be64toh(vnd->sc_comp_offsets[i + 1])
   1310 				  - vnd->sc_comp_offsets[i];
   1311 				if (comp_size > comp_maxsize)
   1312 					comp_maxsize = comp_size;
   1313 			}
   1314 			vnd->sc_comp_offsets[vnd->sc_comp_numoffs - 1] =
   1315 			  be64toh(vnd->sc_comp_offsets[vnd->sc_comp_numoffs - 1]);
   1316 
   1317 			/* create compressed data buffer */
   1318 			vnd->sc_comp_buff = malloc(comp_maxsize,
   1319 			  M_DEVBUF, M_WAITOK);
   1320 
   1321 			/* create decompressed buffer */
   1322 			vnd->sc_comp_decombuf = malloc(vnd->sc_comp_blksz,
   1323 			  M_DEVBUF, M_WAITOK);
   1324 			vnd->sc_comp_buffblk = -1;
   1325 
   1326 			/* Initialize decompress stream */
   1327 			memset(&vnd->sc_comp_stream, 0, sizeof(z_stream));
   1328 			vnd->sc_comp_stream.zalloc = vnd_alloc;
   1329 			vnd->sc_comp_stream.zfree = vnd_free;
   1330 			error = inflateInit2(&vnd->sc_comp_stream, MAX_WBITS);
   1331 			if (error) {
   1332 				if (vnd->sc_comp_stream.msg)
   1333 					printf("vnd%d: compressed file, %s\n",
   1334 					  unit, vnd->sc_comp_stream.msg);
   1335 				VOP_UNLOCK(nd.ni_vp);
   1336 				error = EINVAL;
   1337 				goto close_and_exit;
   1338 			}
   1339 
   1340 			vnd->sc_flags |= VNF_COMP | VNF_READONLY;
   1341 #else /* !VND_COMPRESSION */
   1342 			VOP_UNLOCK(nd.ni_vp);
   1343 			error = EOPNOTSUPP;
   1344 			goto close_and_exit;
   1345 #endif /* VND_COMPRESSION */
   1346 		}
   1347 
   1348 		VOP_UNLOCK(nd.ni_vp);
   1349 		vnd->sc_vp = nd.ni_vp;
   1350 		vnd->sc_size = btodb(vattr.va_size);	/* note truncation */
   1351 
   1352 		/*
   1353 		 * Use pseudo-geometry specified.  If none was provided,
   1354 		 * use "standard" Adaptec fictitious geometry.
   1355 		 */
   1356 		if (vio->vnd_flags & VNDIOF_HASGEOM) {
   1357 
   1358 			memcpy(&vnd->sc_geom, &vio->vnd_geom,
   1359 			    sizeof(vio->vnd_geom));
   1360 
   1361 			/*
   1362 			 * Sanity-check the sector size.
   1363 			 * XXX Don't allow secsize < DEV_BSIZE.	 Should
   1364 			 * XXX we?
   1365 			 */
   1366 			if (vnd->sc_geom.vng_secsize < DEV_BSIZE ||
   1367 			    (vnd->sc_geom.vng_secsize % DEV_BSIZE) != 0 ||
   1368 			    vnd->sc_geom.vng_ncylinders == 0 ||
   1369 			    (vnd->sc_geom.vng_ntracks *
   1370 			     vnd->sc_geom.vng_nsectors) == 0) {
   1371 				error = EINVAL;
   1372 				goto close_and_exit;
   1373 			}
   1374 
   1375 			/*
   1376 			 * Compute the size (in DEV_BSIZE blocks) specified
   1377 			 * by the geometry.
   1378 			 */
   1379 			geomsize = (int64_t)vnd->sc_geom.vng_nsectors *
   1380 			    vnd->sc_geom.vng_ntracks *
   1381 			    vnd->sc_geom.vng_ncylinders *
   1382 			    (vnd->sc_geom.vng_secsize / DEV_BSIZE);
   1383 
   1384 			/*
   1385 			 * Sanity-check the size against the specified
   1386 			 * geometry.
   1387 			 */
   1388 			if (vnd->sc_size < geomsize) {
   1389 				error = EINVAL;
   1390 				goto close_and_exit;
   1391 			}
   1392 		} else if (vnd->sc_size >= (32 * 64)) {
   1393 			/*
   1394 			 * Size must be at least 2048 DEV_BSIZE blocks
   1395 			 * (1M) in order to use this geometry.
   1396 			 */
   1397 			vnd->sc_geom.vng_secsize = DEV_BSIZE;
   1398 			vnd->sc_geom.vng_nsectors = 32;
   1399 			vnd->sc_geom.vng_ntracks = 64;
   1400 			vnd->sc_geom.vng_ncylinders = vnd->sc_size / (64 * 32);
   1401 		} else {
   1402 			vnd->sc_geom.vng_secsize = DEV_BSIZE;
   1403 			vnd->sc_geom.vng_nsectors = 1;
   1404 			vnd->sc_geom.vng_ntracks = 1;
   1405 			vnd->sc_geom.vng_ncylinders = vnd->sc_size;
   1406 		}
   1407 
   1408 		vnd_set_geometry(vnd);
   1409 
   1410 		if (vio->vnd_flags & VNDIOF_READONLY) {
   1411 			vnd->sc_flags |= VNF_READONLY;
   1412 		}
   1413 
   1414 		if ((error = vndsetcred(vnd, l->l_cred)) != 0)
   1415 			goto close_and_exit;
   1416 
   1417 		vndthrottle(vnd, vnd->sc_vp);
   1418 		vio->vnd_osize = dbtob(vnd->sc_size);
   1419 #ifdef VNDIOCSET50
   1420 		if (cmd != VNDIOCSET50)
   1421 #endif
   1422 			vio->vnd_size = dbtob(vnd->sc_size);
   1423 		vnd->sc_flags |= VNF_INITED;
   1424 
   1425 		/* create the kernel thread, wait for it to be up */
   1426 		error = kthread_create(PRI_NONE, 0, NULL, vndthread, vnd,
   1427 		    &vnd->sc_kthread, "%s", device_xname(vnd->sc_dev));
   1428 		if (error)
   1429 			goto close_and_exit;
   1430 		while ((vnd->sc_flags & VNF_KTHREAD) == 0) {
   1431 			tsleep(&vnd->sc_kthread, PRIBIO, "vndthr", 0);
   1432 		}
   1433 #ifdef DEBUG
   1434 		if (vnddebug & VDB_INIT)
   1435 			printf("vndioctl: SET vp %p size 0x%lx %d/%d/%d/%d\n",
   1436 			    vnd->sc_vp, (unsigned long) vnd->sc_size,
   1437 			    vnd->sc_geom.vng_secsize,
   1438 			    vnd->sc_geom.vng_nsectors,
   1439 			    vnd->sc_geom.vng_ntracks,
   1440 			    vnd->sc_geom.vng_ncylinders);
   1441 #endif
   1442 
   1443 		/* Attach the disk. */
   1444 		disk_attach(&vnd->sc_dkdev);
   1445 
   1446 		/* Initialize the xfer and buffer pools. */
   1447 		pool_init(&vnd->sc_vxpool, sizeof(struct vndxfer), 0,
   1448 		    0, 0, "vndxpl", NULL, IPL_BIO);
   1449 
   1450 		vndunlock(vnd);
   1451 
   1452 		pathbuf_destroy(pb);
   1453 
   1454 		/* Discover wedges on this disk */
   1455 		dkwedge_discover(&vnd->sc_dkdev);
   1456 
   1457 		break;
   1458 
   1459 close_and_exit:
   1460 		(void) vn_close(nd.ni_vp, fflags, l->l_cred);
   1461 		pathbuf_destroy(pb);
   1462 unlock_and_exit:
   1463 #ifdef VND_COMPRESSION
   1464 		/* free any allocated memory (for compressed file) */
   1465 		if (vnd->sc_comp_offsets) {
   1466 			free(vnd->sc_comp_offsets, M_DEVBUF);
   1467 			vnd->sc_comp_offsets = NULL;
   1468 		}
   1469 		if (vnd->sc_comp_buff) {
   1470 			free(vnd->sc_comp_buff, M_DEVBUF);
   1471 			vnd->sc_comp_buff = NULL;
   1472 		}
   1473 		if (vnd->sc_comp_decombuf) {
   1474 			free(vnd->sc_comp_decombuf, M_DEVBUF);
   1475 			vnd->sc_comp_decombuf = NULL;
   1476 		}
   1477 #endif /* VND_COMPRESSION */
   1478 		vndunlock(vnd);
   1479 		return error;
   1480 
   1481 #ifdef VNDIOCCLR50
   1482 	case VNDIOCCLR50:
   1483 #endif
   1484 	case VNDIOCCLR:
   1485 		part = DISKPART(dev);
   1486 		pmask = (1 << part);
   1487 		force = (vio->vnd_flags & VNDIOF_FORCE) != 0;
   1488 
   1489 		if ((error = vnddoclear(vnd, pmask, minor(dev), force)) != 0)
   1490 			return error;
   1491 
   1492 		break;
   1493 
   1494 
   1495 	case DIOCWDINFO:
   1496 	case DIOCSDINFO:
   1497 #ifdef __HAVE_OLD_DISKLABEL
   1498 	case ODIOCWDINFO:
   1499 	case ODIOCSDINFO:
   1500 #endif
   1501 	{
   1502 		struct disklabel *lp;
   1503 
   1504 		if ((error = vndlock(vnd)) != 0)
   1505 			return error;
   1506 
   1507 		vnd->sc_flags |= VNF_LABELLING;
   1508 
   1509 #ifdef __HAVE_OLD_DISKLABEL
   1510 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
   1511 			memset(&newlabel, 0, sizeof newlabel);
   1512 			memcpy(&newlabel, data, sizeof (struct olddisklabel));
   1513 			lp = &newlabel;
   1514 		} else
   1515 #endif
   1516 		lp = (struct disklabel *)data;
   1517 
   1518 		error = setdisklabel(vnd->sc_dkdev.dk_label,
   1519 		    lp, 0, vnd->sc_dkdev.dk_cpulabel);
   1520 		if (error == 0) {
   1521 			if (cmd == DIOCWDINFO
   1522 #ifdef __HAVE_OLD_DISKLABEL
   1523 			    || cmd == ODIOCWDINFO
   1524 #endif
   1525 			   )
   1526 				error = writedisklabel(VNDLABELDEV(dev),
   1527 				    vndstrategy, vnd->sc_dkdev.dk_label,
   1528 				    vnd->sc_dkdev.dk_cpulabel);
   1529 		}
   1530 
   1531 		vnd->sc_flags &= ~VNF_LABELLING;
   1532 
   1533 		vndunlock(vnd);
   1534 
   1535 		if (error)
   1536 			return error;
   1537 		break;
   1538 	}
   1539 
   1540 	case DIOCKLABEL:
   1541 		if (*(int *)data != 0)
   1542 			vnd->sc_flags |= VNF_KLABEL;
   1543 		else
   1544 			vnd->sc_flags &= ~VNF_KLABEL;
   1545 		break;
   1546 
   1547 	case DIOCWLABEL:
   1548 		if (*(int *)data != 0)
   1549 			vnd->sc_flags |= VNF_WLABEL;
   1550 		else
   1551 			vnd->sc_flags &= ~VNF_WLABEL;
   1552 		break;
   1553 
   1554 	case DIOCGDEFLABEL:
   1555 		vndgetdefaultlabel(vnd, (struct disklabel *)data);
   1556 		break;
   1557 
   1558 #ifdef __HAVE_OLD_DISKLABEL
   1559 	case ODIOCGDEFLABEL:
   1560 		vndgetdefaultlabel(vnd, &newlabel);
   1561 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
   1562 			return ENOTTY;
   1563 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
   1564 		break;
   1565 #endif
   1566 
   1567 	case DIOCCACHESYNC:
   1568 		vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
   1569 		error = VOP_FSYNC(vnd->sc_vp, vnd->sc_cred,
   1570 		    FSYNC_WAIT | FSYNC_DATAONLY | FSYNC_CACHE, 0, 0);
   1571 		VOP_UNLOCK(vnd->sc_vp);
   1572 		return error;
   1573 
   1574 	default:
   1575 		return ENOTTY;
   1576 	}
   1577 
   1578 	return 0;
   1579 }
   1580 
   1581 /*
   1582  * Duplicate the current processes' credentials.  Since we are called only
   1583  * as the result of a SET ioctl and only root can do that, any future access
   1584  * to this "disk" is essentially as root.  Note that credentials may change
   1585  * if some other uid can write directly to the mapped file (NFS).
   1586  */
   1587 static int
   1588 vndsetcred(struct vnd_softc *vnd, kauth_cred_t cred)
   1589 {
   1590 	struct uio auio;
   1591 	struct iovec aiov;
   1592 	char *tmpbuf;
   1593 	int error;
   1594 
   1595 	vnd->sc_cred = kauth_cred_dup(cred);
   1596 	tmpbuf = malloc(DEV_BSIZE, M_TEMP, M_WAITOK);
   1597 
   1598 	/* XXX: Horrible kludge to establish credentials for NFS */
   1599 	aiov.iov_base = tmpbuf;
   1600 	aiov.iov_len = min(DEV_BSIZE, dbtob(vnd->sc_size));
   1601 	auio.uio_iov = &aiov;
   1602 	auio.uio_iovcnt = 1;
   1603 	auio.uio_offset = 0;
   1604 	auio.uio_rw = UIO_READ;
   1605 	auio.uio_resid = aiov.iov_len;
   1606 	UIO_SETUP_SYSSPACE(&auio);
   1607 	vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
   1608 	error = VOP_READ(vnd->sc_vp, &auio, 0, vnd->sc_cred);
   1609 	if (error == 0) {
   1610 		/*
   1611 		 * Because vnd does all IO directly through the vnode
   1612 		 * we need to flush (at least) the buffer from the above
   1613 		 * VOP_READ from the buffer cache to prevent cache
   1614 		 * incoherencies.  Also, be careful to write dirty
   1615 		 * buffers back to stable storage.
   1616 		 */
   1617 		error = vinvalbuf(vnd->sc_vp, V_SAVE, vnd->sc_cred,
   1618 			    curlwp, 0, 0);
   1619 	}
   1620 	VOP_UNLOCK(vnd->sc_vp);
   1621 
   1622 	free(tmpbuf, M_TEMP);
   1623 	return error;
   1624 }
   1625 
   1626 /*
   1627  * Set maxactive based on FS type
   1628  */
   1629 static void
   1630 vndthrottle(struct vnd_softc *vnd, struct vnode *vp)
   1631 {
   1632 
   1633 	if (vp->v_tag == VT_NFS)
   1634 		vnd->sc_maxactive = 2;
   1635 	else
   1636 		vnd->sc_maxactive = 8;
   1637 
   1638 	if (vnd->sc_maxactive < 1)
   1639 		vnd->sc_maxactive = 1;
   1640 }
   1641 
   1642 #if 0
   1643 static void
   1644 vndshutdown(void)
   1645 {
   1646 	struct vnd_softc *vnd;
   1647 
   1648 	for (vnd = &vnd_softc[0]; vnd < &vnd_softc[numvnd]; vnd++)
   1649 		if (vnd->sc_flags & VNF_INITED)
   1650 			vndclear(vnd);
   1651 }
   1652 #endif
   1653 
   1654 static void
   1655 vndclear(struct vnd_softc *vnd, int myminor)
   1656 {
   1657 	struct vnode *vp = vnd->sc_vp;
   1658 	int fflags = FREAD;
   1659 	int bmaj, cmaj, i, mn;
   1660 	int s;
   1661 
   1662 #ifdef DEBUG
   1663 	if (vnddebug & VDB_FOLLOW)
   1664 		printf("vndclear(%p): vp %p\n", vnd, vp);
   1665 #endif
   1666 	/* locate the major number */
   1667 	bmaj = bdevsw_lookup_major(&vnd_bdevsw);
   1668 	cmaj = cdevsw_lookup_major(&vnd_cdevsw);
   1669 
   1670 	/* Nuke the vnodes for any open instances */
   1671 	for (i = 0; i < MAXPARTITIONS; i++) {
   1672 		mn = DISKMINOR(device_unit(vnd->sc_dev), i);
   1673 		vdevgone(bmaj, mn, mn, VBLK);
   1674 		if (mn != myminor) /* XXX avoid to kill own vnode */
   1675 			vdevgone(cmaj, mn, mn, VCHR);
   1676 	}
   1677 
   1678 	if ((vnd->sc_flags & VNF_READONLY) == 0)
   1679 		fflags |= FWRITE;
   1680 
   1681 	s = splbio();
   1682 	bufq_drain(vnd->sc_tab);
   1683 	splx(s);
   1684 
   1685 	vnd->sc_flags |= VNF_VUNCONF;
   1686 	wakeup(&vnd->sc_tab);
   1687 	while (vnd->sc_flags & VNF_KTHREAD)
   1688 		tsleep(&vnd->sc_kthread, PRIBIO, "vnthr", 0);
   1689 
   1690 #ifdef VND_COMPRESSION
   1691 	/* free the compressed file buffers */
   1692 	if (vnd->sc_flags & VNF_COMP) {
   1693 		if (vnd->sc_comp_offsets) {
   1694 			free(vnd->sc_comp_offsets, M_DEVBUF);
   1695 			vnd->sc_comp_offsets = NULL;
   1696 		}
   1697 		if (vnd->sc_comp_buff) {
   1698 			free(vnd->sc_comp_buff, M_DEVBUF);
   1699 			vnd->sc_comp_buff = NULL;
   1700 		}
   1701 		if (vnd->sc_comp_decombuf) {
   1702 			free(vnd->sc_comp_decombuf, M_DEVBUF);
   1703 			vnd->sc_comp_decombuf = NULL;
   1704 		}
   1705 	}
   1706 #endif /* VND_COMPRESSION */
   1707 	vnd->sc_flags &=
   1708 	    ~(VNF_INITED | VNF_READONLY | VNF_KLABEL | VNF_VLABEL
   1709 	      | VNF_VUNCONF | VNF_COMP | VNF_CLEARING);
   1710 	if (vp == NULL)
   1711 		panic("vndclear: null vp");
   1712 	(void) vn_close(vp, fflags, vnd->sc_cred);
   1713 	kauth_cred_free(vnd->sc_cred);
   1714 	vnd->sc_vp = NULL;
   1715 	vnd->sc_cred = NULL;
   1716 	vnd->sc_size = 0;
   1717 }
   1718 
   1719 static int
   1720 vndsize(dev_t dev)
   1721 {
   1722 	struct vnd_softc *sc;
   1723 	struct disklabel *lp;
   1724 	int part, unit, omask;
   1725 	int size;
   1726 
   1727 	unit = vndunit(dev);
   1728 	sc = device_lookup_private(&vnd_cd, unit);
   1729 	if (sc == NULL)
   1730 		return -1;
   1731 
   1732 	if ((sc->sc_flags & VNF_INITED) == 0)
   1733 		return -1;
   1734 
   1735 	part = DISKPART(dev);
   1736 	omask = sc->sc_dkdev.dk_openmask & (1 << part);
   1737 	lp = sc->sc_dkdev.dk_label;
   1738 
   1739 	if (omask == 0 && vndopen(dev, 0, S_IFBLK, curlwp))	/* XXX */
   1740 		return -1;
   1741 
   1742 	if (lp->d_partitions[part].p_fstype != FS_SWAP)
   1743 		size = -1;
   1744 	else
   1745 		size = lp->d_partitions[part].p_size *
   1746 		    (lp->d_secsize / DEV_BSIZE);
   1747 
   1748 	if (omask == 0 && vndclose(dev, 0, S_IFBLK, curlwp))	/* XXX */
   1749 		return -1;
   1750 
   1751 	return size;
   1752 }
   1753 
   1754 static int
   1755 vnddump(dev_t dev, daddr_t blkno, void *va,
   1756     size_t size)
   1757 {
   1758 
   1759 	/* Not implemented. */
   1760 	return ENXIO;
   1761 }
   1762 
   1763 static void
   1764 vndgetdefaultlabel(struct vnd_softc *sc, struct disklabel *lp)
   1765 {
   1766 	struct vndgeom *vng = &sc->sc_geom;
   1767 	struct partition *pp;
   1768 	unsigned spb;
   1769 
   1770 	memset(lp, 0, sizeof(*lp));
   1771 
   1772 	spb = vng->vng_secsize / DEV_BSIZE;
   1773 	if (sc->sc_size / spb > UINT32_MAX)
   1774 		lp->d_secperunit = UINT32_MAX;
   1775 	else
   1776 		lp->d_secperunit = sc->sc_size / spb;
   1777 	lp->d_secsize = vng->vng_secsize;
   1778 	lp->d_nsectors = vng->vng_nsectors;
   1779 	lp->d_ntracks = vng->vng_ntracks;
   1780 	lp->d_ncylinders = vng->vng_ncylinders;
   1781 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
   1782 
   1783 	strncpy(lp->d_typename, "vnd", sizeof(lp->d_typename));
   1784 	lp->d_type = DKTYPE_VND;
   1785 	strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
   1786 	lp->d_rpm = 3600;
   1787 	lp->d_interleave = 1;
   1788 	lp->d_flags = 0;
   1789 
   1790 	pp = &lp->d_partitions[RAW_PART];
   1791 	pp->p_offset = 0;
   1792 	pp->p_size = lp->d_secperunit;
   1793 	pp->p_fstype = FS_UNUSED;
   1794 	lp->d_npartitions = RAW_PART + 1;
   1795 
   1796 	lp->d_magic = DISKMAGIC;
   1797 	lp->d_magic2 = DISKMAGIC;
   1798 	lp->d_checksum = dkcksum(lp);
   1799 }
   1800 
   1801 /*
   1802  * Read the disklabel from a vnd.  If one is not present, create a fake one.
   1803  */
   1804 static void
   1805 vndgetdisklabel(dev_t dev, struct vnd_softc *sc)
   1806 {
   1807 	const char *errstring;
   1808 	struct disklabel *lp = sc->sc_dkdev.dk_label;
   1809 	struct cpu_disklabel *clp = sc->sc_dkdev.dk_cpulabel;
   1810 	int i;
   1811 
   1812 	memset(clp, 0, sizeof(*clp));
   1813 
   1814 	vndgetdefaultlabel(sc, lp);
   1815 
   1816 	/*
   1817 	 * Call the generic disklabel extraction routine.
   1818 	 */
   1819 	errstring = readdisklabel(VNDLABELDEV(dev), vndstrategy, lp, clp);
   1820 	if (errstring) {
   1821 		/*
   1822 		 * Lack of disklabel is common, but we print the warning
   1823 		 * anyway, since it might contain other useful information.
   1824 		 */
   1825 		aprint_normal_dev(sc->sc_dev, "%s\n", errstring);
   1826 
   1827 		/*
   1828 		 * For historical reasons, if there's no disklabel
   1829 		 * present, all partitions must be FS_BSDFFS and
   1830 		 * occupy the entire disk.
   1831 		 */
   1832 		for (i = 0; i < MAXPARTITIONS; i++) {
   1833 			/*
   1834 			 * Don't wipe out port specific hack (such as
   1835 			 * dos partition hack of i386 port).
   1836 			 */
   1837 			if (lp->d_partitions[i].p_size != 0)
   1838 				continue;
   1839 
   1840 			lp->d_partitions[i].p_size = lp->d_secperunit;
   1841 			lp->d_partitions[i].p_offset = 0;
   1842 			lp->d_partitions[i].p_fstype = FS_BSDFFS;
   1843 		}
   1844 
   1845 		strncpy(lp->d_packname, "default label",
   1846 		    sizeof(lp->d_packname));
   1847 
   1848 		lp->d_npartitions = MAXPARTITIONS;
   1849 		lp->d_checksum = dkcksum(lp);
   1850 	}
   1851 }
   1852 
   1853 /*
   1854  * Wait interruptibly for an exclusive lock.
   1855  *
   1856  * XXX
   1857  * Several drivers do this; it should be abstracted and made MP-safe.
   1858  */
   1859 static int
   1860 vndlock(struct vnd_softc *sc)
   1861 {
   1862 	int error;
   1863 
   1864 	while ((sc->sc_flags & VNF_LOCKED) != 0) {
   1865 		sc->sc_flags |= VNF_WANTED;
   1866 		if ((error = tsleep(sc, PRIBIO | PCATCH, "vndlck", 0)) != 0)
   1867 			return error;
   1868 	}
   1869 	sc->sc_flags |= VNF_LOCKED;
   1870 	return 0;
   1871 }
   1872 
   1873 /*
   1874  * Unlock and wake up any waiters.
   1875  */
   1876 static void
   1877 vndunlock(struct vnd_softc *sc)
   1878 {
   1879 
   1880 	sc->sc_flags &= ~VNF_LOCKED;
   1881 	if ((sc->sc_flags & VNF_WANTED) != 0) {
   1882 		sc->sc_flags &= ~VNF_WANTED;
   1883 		wakeup(sc);
   1884 	}
   1885 }
   1886 
   1887 #ifdef VND_COMPRESSION
   1888 /* compressed file read */
   1889 static void
   1890 compstrategy(struct buf *bp, off_t bn)
   1891 {
   1892 	int error;
   1893 	int unit = vndunit(bp->b_dev);
   1894 	struct vnd_softc *vnd =
   1895 	    device_lookup_private(&vnd_cd, unit);
   1896 	u_int32_t comp_block;
   1897 	struct uio auio;
   1898 	char *addr;
   1899 	int s;
   1900 
   1901 	/* set up constants for data move */
   1902 	auio.uio_rw = UIO_READ;
   1903 	UIO_SETUP_SYSSPACE(&auio);
   1904 
   1905 	/* read, and transfer the data */
   1906 	addr = bp->b_data;
   1907 	bp->b_resid = bp->b_bcount;
   1908 	s = splbio();
   1909 	while (bp->b_resid > 0) {
   1910 		unsigned length;
   1911 		size_t length_in_buffer;
   1912 		u_int32_t offset_in_buffer;
   1913 		struct iovec aiov;
   1914 
   1915 		/* calculate the compressed block number */
   1916 		comp_block = bn / (off_t)vnd->sc_comp_blksz;
   1917 
   1918 		/* check for good block number */
   1919 		if (comp_block >= vnd->sc_comp_numoffs) {
   1920 			bp->b_error = EINVAL;
   1921 			splx(s);
   1922 			return;
   1923 		}
   1924 
   1925 		/* read in the compressed block, if not in buffer */
   1926 		if (comp_block != vnd->sc_comp_buffblk) {
   1927 			length = vnd->sc_comp_offsets[comp_block + 1] -
   1928 			    vnd->sc_comp_offsets[comp_block];
   1929 			vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
   1930 			error = vn_rdwr(UIO_READ, vnd->sc_vp, vnd->sc_comp_buff,
   1931 			    length, vnd->sc_comp_offsets[comp_block],
   1932 			    UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, vnd->sc_cred,
   1933 			    NULL, NULL);
   1934 			if (error) {
   1935 				bp->b_error = error;
   1936 				VOP_UNLOCK(vnd->sc_vp);
   1937 				splx(s);
   1938 				return;
   1939 			}
   1940 			/* uncompress the buffer */
   1941 			vnd->sc_comp_stream.next_in = vnd->sc_comp_buff;
   1942 			vnd->sc_comp_stream.avail_in = length;
   1943 			vnd->sc_comp_stream.next_out = vnd->sc_comp_decombuf;
   1944 			vnd->sc_comp_stream.avail_out = vnd->sc_comp_blksz;
   1945 			inflateReset(&vnd->sc_comp_stream);
   1946 			error = inflate(&vnd->sc_comp_stream, Z_FINISH);
   1947 			if (error != Z_STREAM_END) {
   1948 				if (vnd->sc_comp_stream.msg)
   1949 					aprint_normal_dev(vnd->sc_dev,
   1950 					    "compressed file, %s\n",
   1951 					    vnd->sc_comp_stream.msg);
   1952 				bp->b_error = EBADMSG;
   1953 				VOP_UNLOCK(vnd->sc_vp);
   1954 				splx(s);
   1955 				return;
   1956 			}
   1957 			vnd->sc_comp_buffblk = comp_block;
   1958 			VOP_UNLOCK(vnd->sc_vp);
   1959 		}
   1960 
   1961 		/* transfer the usable uncompressed data */
   1962 		offset_in_buffer = bn % (off_t)vnd->sc_comp_blksz;
   1963 		length_in_buffer = vnd->sc_comp_blksz - offset_in_buffer;
   1964 		if (length_in_buffer > bp->b_resid)
   1965 			length_in_buffer = bp->b_resid;
   1966 		auio.uio_iov = &aiov;
   1967 		auio.uio_iovcnt = 1;
   1968 		aiov.iov_base = addr;
   1969 		aiov.iov_len = length_in_buffer;
   1970 		auio.uio_resid = aiov.iov_len;
   1971 		auio.uio_offset = 0;
   1972 		error = uiomove(vnd->sc_comp_decombuf + offset_in_buffer,
   1973 		    length_in_buffer, &auio);
   1974 		if (error) {
   1975 			bp->b_error = error;
   1976 			splx(s);
   1977 			return;
   1978 		}
   1979 
   1980 		bn += length_in_buffer;
   1981 		addr += length_in_buffer;
   1982 		bp->b_resid -= length_in_buffer;
   1983 	}
   1984 	splx(s);
   1985 }
   1986 
   1987 /* compression memory allocation routines */
   1988 static void *
   1989 vnd_alloc(void *aux, u_int items, u_int siz)
   1990 {
   1991 	return malloc(items * siz, M_TEMP, M_NOWAIT);
   1992 }
   1993 
   1994 static void
   1995 vnd_free(void *aux, void *ptr)
   1996 {
   1997 	free(ptr, M_TEMP);
   1998 }
   1999 #endif /* VND_COMPRESSION */
   2000 
   2001 static void
   2002 vnd_set_geometry(struct vnd_softc *vnd)
   2003 {
   2004 	struct disk_geom *dg = &vnd->sc_dkdev.dk_geom;
   2005 
   2006 	memset(dg, 0, sizeof(*dg));
   2007 
   2008 	dg->dg_secperunit = (int64_t)vnd->sc_geom.vng_nsectors *
   2009 	    vnd->sc_geom.vng_ntracks * vnd->sc_geom.vng_ncylinders;
   2010 	dg->dg_secsize = vnd->sc_geom.vng_secsize;
   2011 	dg->dg_nsectors = vnd->sc_geom.vng_nsectors;
   2012 	dg->dg_ntracks = vnd->sc_geom.vng_ntracks;
   2013 	dg->dg_ncylinders = vnd->sc_geom.vng_ncylinders;
   2014 
   2015 #ifdef DEBUG
   2016 	if (vnddebug & VDB_LABEL) {
   2017 		printf("dg->dg_secperunit: %" PRId64 "\n", dg->dg_secperunit);
   2018 		printf("dg->dg_ncylinders: %u\n", dg->dg_ncylinders);
   2019 	}
   2020 #endif
   2021 	disk_set_info(vnd->sc_dev, &vnd->sc_dkdev, NULL);
   2022 }
   2023 
   2024 #ifdef _MODULE
   2025 
   2026 #include <sys/module.h>
   2027 
   2028 #ifdef VND_COMPRESSION
   2029 #define VND_DEPENDS "zlib"
   2030 #else
   2031 #define VND_DEPENDS NULL
   2032 #endif
   2033 
   2034 MODULE(MODULE_CLASS_DRIVER, vnd, VND_DEPENDS);
   2035 CFDRIVER_DECL(vnd, DV_DISK, NULL);
   2036 
   2037 static int
   2038 vnd_modcmd(modcmd_t cmd, void *arg)
   2039 {
   2040 	int bmajor = -1, cmajor = -1,  error = 0;
   2041 
   2042 	switch (cmd) {
   2043 	case MODULE_CMD_INIT:
   2044 		error = config_cfdriver_attach(&vnd_cd);
   2045 		if (error)
   2046 			break;
   2047 
   2048 		error = config_cfattach_attach(vnd_cd.cd_name, &vnd_ca);
   2049 	        if (error) {
   2050 			config_cfdriver_detach(&vnd_cd);
   2051 			aprint_error("%s: unable to register cfattach\n",
   2052 			    vnd_cd.cd_name);
   2053 			break;
   2054 		}
   2055 
   2056 		error = devsw_attach("vnd", &vnd_bdevsw, &bmajor,
   2057 		    &vnd_cdevsw, &cmajor);
   2058 		if (error) {
   2059 			config_cfattach_detach(vnd_cd.cd_name, &vnd_ca);
   2060 			config_cfdriver_detach(&vnd_cd);
   2061 			break;
   2062 		}
   2063 
   2064 		break;
   2065 
   2066 	case MODULE_CMD_FINI:
   2067 		error = config_cfattach_detach(vnd_cd.cd_name, &vnd_ca);
   2068 		if (error)
   2069 			break;
   2070 		config_cfdriver_detach(&vnd_cd);
   2071 		devsw_detach(&vnd_bdevsw, &vnd_cdevsw);
   2072 		break;
   2073 
   2074 	case MODULE_CMD_STAT:
   2075 		return ENOTTY;
   2076 
   2077 	default:
   2078 		return ENOTTY;
   2079 	}
   2080 
   2081 	return error;
   2082 }
   2083 
   2084 #endif
   2085