Home | History | Annotate | Line # | Download | only in dev
vnd.c revision 1.238
      1 /*	$NetBSD: vnd.c,v 1.238 2014/12/31 19:52:05 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 1.238 2014/12/31 19:52:05 christos 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/file.h>
    118 #include <sys/uio.h>
    119 #include <sys/conf.h>
    120 #include <sys/kauth.h>
    121 
    122 #include <net/zlib.h>
    123 
    124 #include <miscfs/genfs/genfs.h>
    125 #include <miscfs/specfs/specdev.h>
    126 
    127 #include <dev/dkvar.h>
    128 #include <dev/vndvar.h>
    129 
    130 #if defined(VNDDEBUG) && !defined(DEBUG)
    131 #define DEBUG
    132 #endif
    133 
    134 #ifdef DEBUG
    135 int dovndcluster = 1;
    136 #define VDB_FOLLOW	0x01
    137 #define VDB_INIT	0x02
    138 #define VDB_IO		0x04
    139 #define VDB_LABEL	0x08
    140 int vnddebug = 0x00;
    141 #endif
    142 
    143 #define vndunit(x)	DISKUNIT(x)
    144 
    145 struct vndxfer {
    146 	struct buf vx_buf;
    147 	struct vnd_softc *vx_vnd;
    148 };
    149 #define	VND_BUFTOXFER(bp)	((struct vndxfer *)(void *)bp)
    150 
    151 #define VND_GETXFER(vnd)	pool_get(&(vnd)->sc_vxpool, PR_WAITOK)
    152 #define VND_PUTXFER(vnd, vx)	pool_put(&(vnd)->sc_vxpool, (vx))
    153 
    154 #define VNDLABELDEV(dev) \
    155     (MAKEDISKDEV(major((dev)), vndunit((dev)), RAW_PART))
    156 
    157 #define	VND_MAXPENDING(vnd)	((vnd)->sc_maxactive * 4)
    158 
    159 /* called by main() at boot time */
    160 void	vndattach(int);
    161 
    162 static void	vndclear(struct vnd_softc *, int);
    163 static int	vnddoclear(struct vnd_softc *, int, int, bool);
    164 static int	vndsetcred(struct vnd_softc *, kauth_cred_t);
    165 static void	vndthrottle(struct vnd_softc *, struct vnode *);
    166 static void	vndiodone(struct buf *);
    167 #if 0
    168 static void	vndshutdown(void);
    169 #endif
    170 
    171 static void	vndgetdefaultlabel(struct vnd_softc *, struct disklabel *);
    172 static void	vndgetdisklabel(dev_t, struct vnd_softc *);
    173 
    174 static int	vndlock(struct vnd_softc *);
    175 static void	vndunlock(struct vnd_softc *);
    176 #ifdef VND_COMPRESSION
    177 static void	compstrategy(struct buf *, off_t);
    178 static void	*vnd_alloc(void *, u_int, u_int);
    179 static void	vnd_free(void *, void *);
    180 #endif /* VND_COMPRESSION */
    181 
    182 static void	vndthread(void *);
    183 static bool	vnode_has_op(const struct vnode *, int);
    184 static void	handle_with_rdwr(struct vnd_softc *, const struct buf *,
    185 		    struct buf *);
    186 static void	handle_with_strategy(struct vnd_softc *, const struct buf *,
    187 		    struct buf *);
    188 static void	vnd_set_geometry(struct vnd_softc *);
    189 
    190 static dev_type_open(vndopen);
    191 static dev_type_close(vndclose);
    192 static dev_type_read(vndread);
    193 static dev_type_write(vndwrite);
    194 static dev_type_ioctl(vndioctl);
    195 static dev_type_strategy(vndstrategy);
    196 static dev_type_dump(vnddump);
    197 static dev_type_size(vndsize);
    198 
    199 const struct bdevsw vnd_bdevsw = {
    200 	.d_open = vndopen,
    201 	.d_close = vndclose,
    202 	.d_strategy = vndstrategy,
    203 	.d_ioctl = vndioctl,
    204 	.d_dump = vnddump,
    205 	.d_psize = vndsize,
    206 	.d_discard = nodiscard,
    207 	.d_flag = D_DISK
    208 };
    209 
    210 const struct cdevsw vnd_cdevsw = {
    211 	.d_open = vndopen,
    212 	.d_close = vndclose,
    213 	.d_read = vndread,
    214 	.d_write = vndwrite,
    215 	.d_ioctl = vndioctl,
    216 	.d_stop = nostop,
    217 	.d_tty = notty,
    218 	.d_poll = nopoll,
    219 	.d_mmap = nommap,
    220 	.d_kqfilter = nokqfilter,
    221 	.d_discard = nodiscard,
    222 	.d_flag = D_DISK
    223 };
    224 
    225 static int	vnd_match(device_t, cfdata_t, void *);
    226 static void	vnd_attach(device_t, device_t, void *);
    227 static int	vnd_detach(device_t, int);
    228 
    229 CFATTACH_DECL3_NEW(vnd, sizeof(struct vnd_softc),
    230     vnd_match, vnd_attach, vnd_detach, NULL, NULL, NULL, DVF_DETACH_SHUTDOWN);
    231 extern struct cfdriver vnd_cd;
    232 
    233 static struct vnd_softc	*vnd_spawn(int);
    234 int	vnd_destroy(device_t);
    235 
    236 static struct	dkdriver vnddkdriver = { vndstrategy, minphys };
    237 
    238 void
    239 vndattach(int num)
    240 {
    241 	int error;
    242 
    243 	error = config_cfattach_attach(vnd_cd.cd_name, &vnd_ca);
    244 	if (error)
    245 		aprint_error("%s: unable to register cfattach\n",
    246 		    vnd_cd.cd_name);
    247 }
    248 
    249 static int
    250 vnd_match(device_t self, cfdata_t cfdata, void *aux)
    251 {
    252 
    253 	return 1;
    254 }
    255 
    256 static void
    257 vnd_attach(device_t parent, device_t self, void *aux)
    258 {
    259 	struct vnd_softc *sc = device_private(self);
    260 
    261 	sc->sc_dev = self;
    262 	sc->sc_comp_offsets = NULL;
    263 	sc->sc_comp_buff = NULL;
    264 	sc->sc_comp_decombuf = NULL;
    265 	bufq_alloc(&sc->sc_tab, "disksort", BUFQ_SORT_RAWBLOCK);
    266 	disk_init(&sc->sc_dkdev, device_xname(self), &vnddkdriver);
    267 	if (!pmf_device_register(self, NULL, NULL))
    268 		aprint_error_dev(self, "couldn't establish power handler\n");
    269 }
    270 
    271 static int
    272 vnd_detach(device_t self, int flags)
    273 {
    274 	int error;
    275 	struct vnd_softc *sc = device_private(self);
    276 
    277 	if (sc->sc_flags & VNF_INITED) {
    278 		error = vnddoclear(sc, 0, -1, (flags & DETACH_FORCE) != 0);
    279 		if (error != 0)
    280 			return error;
    281 	}
    282 
    283 	pmf_device_deregister(self);
    284 	bufq_free(sc->sc_tab);
    285 	disk_destroy(&sc->sc_dkdev);
    286 
    287 	return 0;
    288 }
    289 
    290 static struct vnd_softc *
    291 vnd_spawn(int unit)
    292 {
    293 	cfdata_t cf;
    294 
    295 	cf = malloc(sizeof(*cf), M_DEVBUF, M_WAITOK);
    296 	cf->cf_name = vnd_cd.cd_name;
    297 	cf->cf_atname = vnd_cd.cd_name;
    298 	cf->cf_unit = unit;
    299 	cf->cf_fstate = FSTATE_STAR;
    300 
    301 	return device_private(config_attach_pseudo(cf));
    302 }
    303 
    304 int
    305 vnd_destroy(device_t dev)
    306 {
    307 	int error;
    308 	cfdata_t cf;
    309 
    310 	cf = device_cfdata(dev);
    311 	error = config_detach(dev, DETACH_QUIET);
    312 	if (error)
    313 		return error;
    314 	free(cf, M_DEVBUF);
    315 	return 0;
    316 }
    317 
    318 static int
    319 vndopen(dev_t dev, int flags, int mode, struct lwp *l)
    320 {
    321 	int unit = vndunit(dev);
    322 	struct vnd_softc *sc;
    323 	int error = 0, part, pmask;
    324 	struct disklabel *lp;
    325 
    326 #ifdef DEBUG
    327 	if (vnddebug & VDB_FOLLOW)
    328 		printf("vndopen(0x%"PRIx64", 0x%x, 0x%x, %p)\n", dev, flags, mode, l);
    329 #endif
    330 	sc = device_lookup_private(&vnd_cd, unit);
    331 	if (sc == NULL) {
    332 		sc = vnd_spawn(unit);
    333 		if (sc == NULL)
    334 			return ENOMEM;
    335 	}
    336 
    337 	if ((error = vndlock(sc)) != 0)
    338 		return error;
    339 
    340 	if ((sc->sc_flags & VNF_CLEARING) != 0) {
    341 		error = ENXIO;
    342 		goto done;
    343 	}
    344 
    345 	lp = sc->sc_dkdev.dk_label;
    346 
    347 	part = DISKPART(dev);
    348 	pmask = (1 << part);
    349 
    350 	if (sc->sc_dkdev.dk_nwedges != 0 && part != RAW_PART) {
    351 		error = EBUSY;
    352 		goto done;
    353 	}
    354 
    355 	if (sc->sc_flags & VNF_INITED) {
    356 		if ((sc->sc_dkdev.dk_openmask & ~(1<<RAW_PART)) != 0) {
    357 			/*
    358 			 * If any non-raw partition is open, but the disk
    359 			 * has been invalidated, disallow further opens.
    360 			 */
    361 			if ((sc->sc_flags & VNF_VLABEL) == 0) {
    362 				error = EIO;
    363 				goto done;
    364 			}
    365 		} else {
    366 			/*
    367 			 * Load the partition info if not already loaded.
    368 			 */
    369 			if ((sc->sc_flags & VNF_VLABEL) == 0) {
    370 				sc->sc_flags |= VNF_VLABEL;
    371 				vndgetdisklabel(dev, sc);
    372 			}
    373 		}
    374 	}
    375 
    376 	/* Check that the partitions exists. */
    377 	if (part != RAW_PART) {
    378 		if (((sc->sc_flags & VNF_INITED) == 0) ||
    379 		    ((part >= lp->d_npartitions) ||
    380 		     (lp->d_partitions[part].p_fstype == FS_UNUSED))) {
    381 			error = ENXIO;
    382 			goto done;
    383 		}
    384 	}
    385 
    386 	/* Prevent our unit from being unconfigured while open. */
    387 	switch (mode) {
    388 	case S_IFCHR:
    389 		sc->sc_dkdev.dk_copenmask |= pmask;
    390 		break;
    391 
    392 	case S_IFBLK:
    393 		sc->sc_dkdev.dk_bopenmask |= pmask;
    394 		break;
    395 	}
    396 	sc->sc_dkdev.dk_openmask =
    397 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
    398 
    399  done:
    400 	vndunlock(sc);
    401 	return error;
    402 }
    403 
    404 static int
    405 vndclose(dev_t dev, int flags, int mode, struct lwp *l)
    406 {
    407 	int unit = vndunit(dev);
    408 	struct vnd_softc *sc;
    409 	int error = 0, part;
    410 
    411 #ifdef DEBUG
    412 	if (vnddebug & VDB_FOLLOW)
    413 		printf("vndclose(0x%"PRIx64", 0x%x, 0x%x, %p)\n", dev, flags, mode, l);
    414 #endif
    415 	sc = device_lookup_private(&vnd_cd, unit);
    416 	if (sc == NULL)
    417 		return ENXIO;
    418 
    419 	if ((error = vndlock(sc)) != 0)
    420 		return error;
    421 
    422 	part = DISKPART(dev);
    423 
    424 	/* ...that much closer to allowing unconfiguration... */
    425 	switch (mode) {
    426 	case S_IFCHR:
    427 		sc->sc_dkdev.dk_copenmask &= ~(1 << part);
    428 		break;
    429 
    430 	case S_IFBLK:
    431 		sc->sc_dkdev.dk_bopenmask &= ~(1 << part);
    432 		break;
    433 	}
    434 	sc->sc_dkdev.dk_openmask =
    435 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
    436 
    437 	vndunlock(sc);
    438 
    439 	if ((sc->sc_flags & VNF_INITED) == 0) {
    440 		if ((error = vnd_destroy(sc->sc_dev)) != 0) {
    441 			aprint_error_dev(sc->sc_dev,
    442 			    "unable to detach instance\n");
    443 			return error;
    444 		}
    445 	}
    446 
    447 	return 0;
    448 }
    449 
    450 /*
    451  * Queue the request, and wakeup the kernel thread to handle it.
    452  */
    453 static void
    454 vndstrategy(struct buf *bp)
    455 {
    456 	int unit = vndunit(bp->b_dev);
    457 	struct vnd_softc *vnd =
    458 	    device_lookup_private(&vnd_cd, unit);
    459 	struct disklabel *lp;
    460 	daddr_t blkno;
    461 	int s = splbio();
    462 
    463 	if (vnd == NULL) {
    464 		bp->b_error = ENXIO;
    465 		goto done;
    466 	}
    467 	lp = vnd->sc_dkdev.dk_label;
    468 
    469 	if ((vnd->sc_flags & VNF_INITED) == 0) {
    470 		bp->b_error = ENXIO;
    471 		goto done;
    472 	}
    473 
    474 	/*
    475 	 * The transfer must be a whole number of blocks.
    476 	 */
    477 	if ((bp->b_bcount % lp->d_secsize) != 0) {
    478 		bp->b_error = EINVAL;
    479 		goto done;
    480 	}
    481 
    482 	/*
    483 	 * check if we're read-only.
    484 	 */
    485 	if ((vnd->sc_flags & VNF_READONLY) && !(bp->b_flags & B_READ)) {
    486 		bp->b_error = EACCES;
    487 		goto done;
    488 	}
    489 
    490 	/* If it's a nil transfer, wake up the top half now. */
    491 	if (bp->b_bcount == 0) {
    492 		goto done;
    493 	}
    494 
    495 	/*
    496 	 * Do bounds checking and adjust transfer.  If there's an error,
    497 	 * the bounds check will flag that for us.
    498 	 */
    499 	if (DISKPART(bp->b_dev) == RAW_PART) {
    500 		if (bounds_check_with_mediasize(bp, DEV_BSIZE,
    501 		    vnd->sc_size) <= 0)
    502 			goto done;
    503 	} else {
    504 		if (bounds_check_with_label(&vnd->sc_dkdev,
    505 		    bp, vnd->sc_flags & (VNF_WLABEL|VNF_LABELLING)) <= 0)
    506 			goto done;
    507 	}
    508 
    509 	/*
    510 	 * Put the block number in terms of the logical blocksize
    511 	 * of the "device".
    512 	 */
    513 
    514 	blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
    515 
    516 	/*
    517 	 * Translate the partition-relative block number to an absolute.
    518 	 */
    519 	if (DISKPART(bp->b_dev) != RAW_PART) {
    520 		struct partition *pp;
    521 
    522 		pp = &vnd->sc_dkdev.dk_label->d_partitions[
    523 		    DISKPART(bp->b_dev)];
    524 		blkno += pp->p_offset;
    525 	}
    526 	bp->b_rawblkno = blkno;
    527 
    528 #ifdef DEBUG
    529 	if (vnddebug & VDB_FOLLOW)
    530 		printf("vndstrategy(%p): unit %d\n", bp, unit);
    531 #endif
    532 	if ((vnd->sc_flags & VNF_USE_VN_RDWR)) {
    533 		KASSERT(vnd->sc_pending >= 0 &&
    534 		    vnd->sc_pending <= VND_MAXPENDING(vnd));
    535 		while (vnd->sc_pending == VND_MAXPENDING(vnd))
    536 			tsleep(&vnd->sc_pending, PRIBIO, "vndpc", 0);
    537 		vnd->sc_pending++;
    538 	}
    539 	bufq_put(vnd->sc_tab, bp);
    540 	wakeup(&vnd->sc_tab);
    541 	splx(s);
    542 	return;
    543 
    544 done:
    545 	bp->b_resid = bp->b_bcount;
    546 	biodone(bp);
    547 	splx(s);
    548 }
    549 
    550 static bool
    551 vnode_has_strategy(struct vnd_softc *vnd)
    552 {
    553 	return vnode_has_op(vnd->sc_vp, VOFFSET(vop_bmap)) &&
    554 	    vnode_has_op(vnd->sc_vp, VOFFSET(vop_strategy));
    555 }
    556 
    557 /* XXX this function needs a reliable check to detect
    558  * sparse files. Otherwise, bmap/strategy may be used
    559  * and fail on non-allocated blocks. VOP_READ/VOP_WRITE
    560  * works on sparse files.
    561  */
    562 #if notyet
    563 static bool
    564 vnode_strategy_probe(struct vnd_softc *vnd)
    565 {
    566 	int error;
    567 	daddr_t nbn;
    568 
    569 	if (!vnode_has_strategy(vnd))
    570 		return false;
    571 
    572 	/* Convert the first logical block number to its
    573 	 * physical block number.
    574 	 */
    575 	error = 0;
    576 	vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
    577 	error = VOP_BMAP(vnd->sc_vp, 0, NULL, &nbn, NULL);
    578 	VOP_UNLOCK(vnd->sc_vp);
    579 
    580 	/* Test if that worked. */
    581 	if (error == 0 && (long)nbn == -1)
    582 		return false;
    583 
    584 	return true;
    585 }
    586 #endif
    587 
    588 static void
    589 vndthread(void *arg)
    590 {
    591 	struct vnd_softc *vnd = arg;
    592 	int s;
    593 
    594 	/* Determine whether we can *use* VOP_BMAP and VOP_STRATEGY to
    595 	 * directly access the backing vnode.  If we can, use these two
    596 	 * operations to avoid messing with the local buffer cache.
    597 	 * Otherwise fall back to regular VOP_READ/VOP_WRITE operations
    598 	 * which are guaranteed to work with any file system. */
    599 	if ((vnd->sc_flags & VNF_USE_VN_RDWR) == 0 &&
    600 	    ! vnode_has_strategy(vnd))
    601 		vnd->sc_flags |= VNF_USE_VN_RDWR;
    602 
    603 #ifdef DEBUG
    604 	if (vnddebug & VDB_INIT)
    605 		printf("vndthread: vp %p, %s\n", vnd->sc_vp,
    606 		    (vnd->sc_flags & VNF_USE_VN_RDWR) == 0 ?
    607 		    "using bmap/strategy operations" :
    608 		    "using read/write operations");
    609 #endif
    610 
    611 	s = splbio();
    612 	vnd->sc_flags |= VNF_KTHREAD;
    613 	wakeup(&vnd->sc_kthread);
    614 
    615 	/*
    616 	 * Dequeue requests and serve them depending on the available
    617 	 * vnode operations.
    618 	 */
    619 	while ((vnd->sc_flags & VNF_VUNCONF) == 0) {
    620 		struct vndxfer *vnx;
    621 		struct buf *obp;
    622 		struct buf *bp;
    623 
    624 		obp = bufq_get(vnd->sc_tab);
    625 		if (obp == NULL) {
    626 			tsleep(&vnd->sc_tab, PRIBIO, "vndbp", 0);
    627 			continue;
    628 		};
    629 		if ((vnd->sc_flags & VNF_USE_VN_RDWR)) {
    630 			KASSERT(vnd->sc_pending > 0 &&
    631 			    vnd->sc_pending <= VND_MAXPENDING(vnd));
    632 			if (vnd->sc_pending-- == VND_MAXPENDING(vnd))
    633 				wakeup(&vnd->sc_pending);
    634 		}
    635 		splx(s);
    636 #ifdef DEBUG
    637 		if (vnddebug & VDB_FOLLOW)
    638 			printf("vndthread(%p)\n", obp);
    639 #endif
    640 
    641 		if (vnd->sc_vp->v_mount == NULL) {
    642 			obp->b_error = ENXIO;
    643 			goto done;
    644 		}
    645 #ifdef VND_COMPRESSION
    646 		/* handle a compressed read */
    647 		if ((obp->b_flags & B_READ) != 0 && (vnd->sc_flags & VNF_COMP)) {
    648 			off_t bn;
    649 
    650 			/* Convert to a byte offset within the file. */
    651 			bn = obp->b_rawblkno *
    652 			    vnd->sc_dkdev.dk_label->d_secsize;
    653 
    654 			compstrategy(obp, bn);
    655 			goto done;
    656 		}
    657 #endif /* VND_COMPRESSION */
    658 
    659 		/*
    660 		 * Allocate a header for this transfer and link it to the
    661 		 * buffer
    662 		 */
    663 		s = splbio();
    664 		vnx = VND_GETXFER(vnd);
    665 		splx(s);
    666 		vnx->vx_vnd = vnd;
    667 
    668 		s = splbio();
    669 		while (vnd->sc_active >= vnd->sc_maxactive) {
    670 			tsleep(&vnd->sc_tab, PRIBIO, "vndac", 0);
    671 		}
    672 		vnd->sc_active++;
    673 		splx(s);
    674 
    675 		/* Instrumentation. */
    676 		disk_busy(&vnd->sc_dkdev);
    677 
    678 		bp = &vnx->vx_buf;
    679 		buf_init(bp);
    680 		bp->b_flags = (obp->b_flags & B_READ);
    681 		bp->b_oflags = obp->b_oflags;
    682 		bp->b_cflags = obp->b_cflags;
    683 		bp->b_iodone = vndiodone;
    684 		bp->b_private = obp;
    685 		bp->b_vp = vnd->sc_vp;
    686 		bp->b_objlock = bp->b_vp->v_interlock;
    687 		bp->b_data = obp->b_data;
    688 		bp->b_bcount = obp->b_bcount;
    689 		BIO_COPYPRIO(bp, obp);
    690 
    691 		/* Handle the request using the appropriate operations. */
    692 		if ((vnd->sc_flags & VNF_USE_VN_RDWR) == 0)
    693 			handle_with_strategy(vnd, obp, bp);
    694 		else
    695 			handle_with_rdwr(vnd, obp, bp);
    696 
    697 		s = splbio();
    698 		continue;
    699 
    700 done:
    701 		biodone(obp);
    702 		s = splbio();
    703 	}
    704 
    705 	vnd->sc_flags &= (~VNF_KTHREAD | VNF_VUNCONF);
    706 	wakeup(&vnd->sc_kthread);
    707 	splx(s);
    708 	kthread_exit(0);
    709 }
    710 
    711 /*
    712  * Checks if the given vnode supports the requested operation.
    713  * The operation is specified the offset returned by VOFFSET.
    714  *
    715  * XXX The test below used to determine this is quite fragile
    716  * because it relies on the file system to use genfs to specify
    717  * unimplemented operations.  There might be another way to do
    718  * it more cleanly.
    719  */
    720 static bool
    721 vnode_has_op(const struct vnode *vp, int opoffset)
    722 {
    723 	int (*defaultp)(void *);
    724 	int (*opp)(void *);
    725 
    726 	defaultp = vp->v_op[VOFFSET(vop_default)];
    727 	opp = vp->v_op[opoffset];
    728 
    729 	return opp != defaultp && opp != genfs_eopnotsupp &&
    730 	    opp != genfs_badop && opp != genfs_nullop;
    731 }
    732 
    733 /*
    734  * Handes the read/write request given in 'bp' using the vnode's VOP_READ
    735  * and VOP_WRITE operations.
    736  *
    737  * 'obp' is a pointer to the original request fed to the vnd device.
    738  */
    739 static void
    740 handle_with_rdwr(struct vnd_softc *vnd, const struct buf *obp, struct buf *bp)
    741 {
    742 	bool doread;
    743 	off_t offset;
    744 	size_t len, resid;
    745 	struct vnode *vp;
    746 
    747 	doread = bp->b_flags & B_READ;
    748 	offset = obp->b_rawblkno * vnd->sc_dkdev.dk_label->d_secsize;
    749 	len = bp->b_bcount;
    750 	vp = vnd->sc_vp;
    751 
    752 #if defined(DEBUG)
    753 	if (vnddebug & VDB_IO)
    754 		printf("vnd (rdwr): vp %p, %s, rawblkno 0x%" PRIx64
    755 		    ", secsize %d, offset %" PRIu64
    756 		    ", bcount %d\n",
    757 		    vp, doread ? "read" : "write", obp->b_rawblkno,
    758 		    vnd->sc_dkdev.dk_label->d_secsize, offset,
    759 		    bp->b_bcount);
    760 #endif
    761 
    762 	/* Issue the read or write operation. */
    763 	bp->b_error =
    764 	    vn_rdwr(doread ? UIO_READ : UIO_WRITE,
    765 	    vp, bp->b_data, len, offset, UIO_SYSSPACE,
    766 	    IO_ADV_ENCODE(POSIX_FADV_NOREUSE), vnd->sc_cred, &resid, NULL);
    767 	bp->b_resid = resid;
    768 
    769 	mutex_enter(vp->v_interlock);
    770 	(void) VOP_PUTPAGES(vp, 0, 0,
    771 	    PGO_ALLPAGES | PGO_CLEANIT | PGO_FREE | PGO_SYNCIO);
    772 
    773 	/* We need to increase the number of outputs on the vnode if
    774 	 * there was any write to it. */
    775 	if (!doread) {
    776 		mutex_enter(vp->v_interlock);
    777 		vp->v_numoutput++;
    778 		mutex_exit(vp->v_interlock);
    779 	}
    780 
    781 	biodone(bp);
    782 }
    783 
    784 /*
    785  * Handes the read/write request given in 'bp' using the vnode's VOP_BMAP
    786  * and VOP_STRATEGY operations.
    787  *
    788  * 'obp' is a pointer to the original request fed to the vnd device.
    789  */
    790 static void
    791 handle_with_strategy(struct vnd_softc *vnd, const struct buf *obp,
    792     struct buf *bp)
    793 {
    794 	int bsize, error, flags, skipped;
    795 	size_t resid, sz;
    796 	off_t bn, offset;
    797 	struct vnode *vp;
    798 
    799 	flags = obp->b_flags;
    800 
    801 	if (!(flags & B_READ)) {
    802 		vp = bp->b_vp;
    803 		mutex_enter(vp->v_interlock);
    804 		vp->v_numoutput++;
    805 		mutex_exit(vp->v_interlock);
    806 	}
    807 
    808 	/* convert to a byte offset within the file. */
    809 	bn = obp->b_rawblkno * vnd->sc_dkdev.dk_label->d_secsize;
    810 
    811 	bsize = vnd->sc_vp->v_mount->mnt_stat.f_iosize;
    812 	skipped = 0;
    813 
    814 	/*
    815 	 * Break the request into bsize pieces and feed them
    816 	 * sequentially using VOP_BMAP/VOP_STRATEGY.
    817 	 * We do it this way to keep from flooding NFS servers if we
    818 	 * are connected to an NFS file.  This places the burden on
    819 	 * the client rather than the server.
    820 	 */
    821 	error = 0;
    822 	bp->b_resid = bp->b_bcount;
    823 	for (offset = 0, resid = bp->b_resid; resid;
    824 	    resid -= sz, offset += sz) {
    825 		struct buf *nbp;
    826 		daddr_t nbn;
    827 		int off, nra;
    828 
    829 		nra = 0;
    830 		vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
    831 		error = VOP_BMAP(vnd->sc_vp, bn / bsize, &vp, &nbn, &nra);
    832 		VOP_UNLOCK(vnd->sc_vp);
    833 
    834 		if (error == 0 && (long)nbn == -1)
    835 			error = EIO;
    836 
    837 		/*
    838 		 * If there was an error or a hole in the file...punt.
    839 		 * Note that we may have to wait for any operations
    840 		 * that we have already fired off before releasing
    841 		 * the buffer.
    842 		 *
    843 		 * XXX we could deal with holes here but it would be
    844 		 * a hassle (in the write case).
    845 		 */
    846 		if (error) {
    847 			skipped += resid;
    848 			break;
    849 		}
    850 
    851 #ifdef DEBUG
    852 		if (!dovndcluster)
    853 			nra = 0;
    854 #endif
    855 
    856 		off = bn % bsize;
    857 		sz = MIN(((off_t)1 + nra) * bsize - off, resid);
    858 #ifdef	DEBUG
    859 		if (vnddebug & VDB_IO)
    860 			printf("vndstrategy: vp %p/%p bn 0x%qx/0x%" PRIx64
    861 			    " sz 0x%zx\n", vnd->sc_vp, vp, (long long)bn,
    862 			    nbn, sz);
    863 #endif
    864 
    865 		nbp = getiobuf(vp, true);
    866 		nestiobuf_setup(bp, nbp, offset, sz);
    867 		nbp->b_blkno = nbn + btodb(off);
    868 
    869 #if 0 /* XXX #ifdef DEBUG */
    870 		if (vnddebug & VDB_IO)
    871 			printf("vndstart(%ld): bp %p vp %p blkno "
    872 			    "0x%" PRIx64 " flags %x addr %p cnt 0x%x\n",
    873 			    (long) (vnd-vnd_softc), &nbp->vb_buf,
    874 			    nbp->vb_buf.b_vp, nbp->vb_buf.b_blkno,
    875 			    nbp->vb_buf.b_flags, nbp->vb_buf.b_data,
    876 			    nbp->vb_buf.b_bcount);
    877 #endif
    878 		VOP_STRATEGY(vp, nbp);
    879 		bn += sz;
    880 	}
    881 	nestiobuf_done(bp, skipped, error);
    882 }
    883 
    884 static void
    885 vndiodone(struct buf *bp)
    886 {
    887 	struct vndxfer *vnx = VND_BUFTOXFER(bp);
    888 	struct vnd_softc *vnd = vnx->vx_vnd;
    889 	struct buf *obp = bp->b_private;
    890 	int s = splbio();
    891 
    892 	KASSERT(&vnx->vx_buf == bp);
    893 	KASSERT(vnd->sc_active > 0);
    894 #ifdef DEBUG
    895 	if (vnddebug & VDB_IO) {
    896 		printf("vndiodone1: bp %p iodone: error %d\n",
    897 		    bp, bp->b_error);
    898 	}
    899 #endif
    900 	disk_unbusy(&vnd->sc_dkdev, bp->b_bcount - bp->b_resid,
    901 	    (bp->b_flags & B_READ));
    902 	vnd->sc_active--;
    903 	if (vnd->sc_active == 0) {
    904 		wakeup(&vnd->sc_tab);
    905 	}
    906 	splx(s);
    907 	obp->b_error = bp->b_error;
    908 	obp->b_resid = bp->b_resid;
    909 	buf_destroy(bp);
    910 	VND_PUTXFER(vnd, vnx);
    911 	biodone(obp);
    912 }
    913 
    914 /* ARGSUSED */
    915 static int
    916 vndread(dev_t dev, struct uio *uio, int flags)
    917 {
    918 	int unit = vndunit(dev);
    919 	struct vnd_softc *sc;
    920 
    921 #ifdef DEBUG
    922 	if (vnddebug & VDB_FOLLOW)
    923 		printf("vndread(0x%"PRIx64", %p)\n", dev, uio);
    924 #endif
    925 
    926 	sc = device_lookup_private(&vnd_cd, unit);
    927 	if (sc == NULL)
    928 		return ENXIO;
    929 
    930 	if ((sc->sc_flags & VNF_INITED) == 0)
    931 		return ENXIO;
    932 
    933 	return physio(vndstrategy, NULL, dev, B_READ, minphys, uio);
    934 }
    935 
    936 /* ARGSUSED */
    937 static int
    938 vndwrite(dev_t dev, struct uio *uio, int flags)
    939 {
    940 	int unit = vndunit(dev);
    941 	struct vnd_softc *sc;
    942 
    943 #ifdef DEBUG
    944 	if (vnddebug & VDB_FOLLOW)
    945 		printf("vndwrite(0x%"PRIx64", %p)\n", dev, uio);
    946 #endif
    947 
    948 	sc = device_lookup_private(&vnd_cd, unit);
    949 	if (sc == NULL)
    950 		return ENXIO;
    951 
    952 	if ((sc->sc_flags & VNF_INITED) == 0)
    953 		return ENXIO;
    954 
    955 	return physio(vndstrategy, NULL, dev, B_WRITE, minphys, uio);
    956 }
    957 
    958 static int
    959 vnd_cget(struct lwp *l, int unit, int *un, struct vattr *va)
    960 {
    961 	int error;
    962 	struct vnd_softc *vnd;
    963 
    964 	if (*un == -1)
    965 		*un = unit;
    966 	if (*un < 0)
    967 		return EINVAL;
    968 
    969 	vnd = device_lookup_private(&vnd_cd, *un);
    970 	if (vnd == NULL)
    971 		return -1;
    972 
    973 	if ((vnd->sc_flags & VNF_INITED) == 0)
    974 		return -1;
    975 
    976 	vn_lock(vnd->sc_vp, LK_SHARED | LK_RETRY);
    977 	error = VOP_GETATTR(vnd->sc_vp, va, l->l_cred);
    978 	VOP_UNLOCK(vnd->sc_vp);
    979 	return error;
    980 }
    981 
    982 static int
    983 vnddoclear(struct vnd_softc *vnd, int pmask, int minor, bool force)
    984 {
    985 	int error;
    986 
    987 	if ((error = vndlock(vnd)) != 0)
    988 		return error;
    989 
    990 	/*
    991 	 * Don't unconfigure if any other partitions are open
    992 	 * or if both the character and block flavors of this
    993 	 * partition are open.
    994 	 */
    995 	if (DK_BUSY(vnd, pmask) && !force) {
    996 		vndunlock(vnd);
    997 		return EBUSY;
    998 	}
    999 
   1000 	/* Delete all of our wedges */
   1001 	dkwedge_delall(&vnd->sc_dkdev);
   1002 
   1003 	/*
   1004 	 * XXX vndclear() might call vndclose() implicitly;
   1005 	 * release lock to avoid recursion
   1006 	 *
   1007 	 * Set VNF_CLEARING to prevent vndopen() from
   1008 	 * sneaking in after we vndunlock().
   1009 	 */
   1010 	vnd->sc_flags |= VNF_CLEARING;
   1011 	vndunlock(vnd);
   1012 	vndclear(vnd, minor);
   1013 #ifdef DEBUG
   1014 	if (vnddebug & VDB_INIT)
   1015 		printf("vndioctl: CLRed\n");
   1016 #endif
   1017 
   1018 	/* Destroy the xfer and buffer pools. */
   1019 	pool_destroy(&vnd->sc_vxpool);
   1020 
   1021 	/* Detach the disk. */
   1022 	disk_detach(&vnd->sc_dkdev);
   1023 
   1024 	return 0;
   1025 }
   1026 
   1027 /* ARGSUSED */
   1028 static int
   1029 vndioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l)
   1030 {
   1031 	bool force;
   1032 	int unit = vndunit(dev);
   1033 	struct vnd_softc *vnd;
   1034 	struct vnd_ioctl *vio;
   1035 	struct vattr vattr;
   1036 	struct pathbuf *pb;
   1037 	struct nameidata nd;
   1038 	int error, part, pmask;
   1039 	uint64_t geomsize;
   1040 	int fflags;
   1041 #ifdef __HAVE_OLD_DISKLABEL
   1042 	struct disklabel newlabel;
   1043 #endif
   1044 
   1045 #ifdef DEBUG
   1046 	if (vnddebug & VDB_FOLLOW)
   1047 		printf("vndioctl(0x%"PRIx64", 0x%lx, %p, 0x%x, %p): unit %d\n",
   1048 		    dev, cmd, data, flag, l->l_proc, unit);
   1049 #endif
   1050 	vnd = device_lookup_private(&vnd_cd, unit);
   1051 	if (vnd == NULL &&
   1052 #ifdef COMPAT_30
   1053 	    cmd != VNDIOCGET30 &&
   1054 #endif
   1055 #ifdef COMPAT_50
   1056 	    cmd != VNDIOCGET50 &&
   1057 #endif
   1058 	    cmd != VNDIOCGET)
   1059 		return ENXIO;
   1060 	vio = (struct vnd_ioctl *)data;
   1061 
   1062 	/* Must be open for writes for these commands... */
   1063 	switch (cmd) {
   1064 	case VNDIOCSET:
   1065 	case VNDIOCCLR:
   1066 #ifdef COMPAT_50
   1067 	case VNDIOCSET50:
   1068 	case VNDIOCCLR50:
   1069 #endif
   1070 	case DIOCSDINFO:
   1071 	case DIOCWDINFO:
   1072 #ifdef __HAVE_OLD_DISKLABEL
   1073 	case ODIOCSDINFO:
   1074 	case ODIOCWDINFO:
   1075 #endif
   1076 	case DIOCKLABEL:
   1077 	case DIOCWLABEL:
   1078 		if ((flag & FWRITE) == 0)
   1079 			return EBADF;
   1080 	}
   1081 
   1082 	/* Must be initialized for these... */
   1083 	switch (cmd) {
   1084 	case VNDIOCCLR:
   1085 #ifdef VNDIOCCLR50
   1086 	case VNDIOCCLR50:
   1087 #endif
   1088 	case DIOCGDINFO:
   1089 	case DIOCSDINFO:
   1090 	case DIOCWDINFO:
   1091 	case DIOCGPART:
   1092 	case DIOCKLABEL:
   1093 	case DIOCWLABEL:
   1094 	case DIOCGDEFLABEL:
   1095 	case DIOCCACHESYNC:
   1096 #ifdef __HAVE_OLD_DISKLABEL
   1097 	case ODIOCGDINFO:
   1098 	case ODIOCSDINFO:
   1099 	case ODIOCWDINFO:
   1100 	case ODIOCGDEFLABEL:
   1101 #endif
   1102 		if ((vnd->sc_flags & VNF_INITED) == 0)
   1103 			return ENXIO;
   1104 	}
   1105 
   1106 	error = disk_ioctl(&vnd->sc_dkdev, dev, cmd, data, flag, l);
   1107 	if (error != EPASSTHROUGH)
   1108 		return error;
   1109 
   1110 
   1111 	switch (cmd) {
   1112 #ifdef VNDIOCSET50
   1113 	case VNDIOCSET50:
   1114 #endif
   1115 	case VNDIOCSET:
   1116 		if (vnd->sc_flags & VNF_INITED)
   1117 			return EBUSY;
   1118 
   1119 		if ((error = vndlock(vnd)) != 0)
   1120 			return error;
   1121 
   1122 		fflags = FREAD;
   1123 		if ((vio->vnd_flags & VNDIOF_READONLY) == 0)
   1124 			fflags |= FWRITE;
   1125 		error = pathbuf_copyin(vio->vnd_file, &pb);
   1126 		if (error) {
   1127 			goto unlock_and_exit;
   1128 		}
   1129 		NDINIT(&nd, LOOKUP, FOLLOW, pb);
   1130 		if ((error = vn_open(&nd, fflags, 0)) != 0) {
   1131 			pathbuf_destroy(pb);
   1132 			goto unlock_and_exit;
   1133 		}
   1134 		KASSERT(l);
   1135 		error = VOP_GETATTR(nd.ni_vp, &vattr, l->l_cred);
   1136 		if (!error && nd.ni_vp->v_type != VREG)
   1137 			error = EOPNOTSUPP;
   1138 		if (!error && vattr.va_bytes < vattr.va_size)
   1139 			/* File is definitely sparse, use vn_rdwr() */
   1140 			vnd->sc_flags |= VNF_USE_VN_RDWR;
   1141 		if (error) {
   1142 			VOP_UNLOCK(nd.ni_vp);
   1143 			goto close_and_exit;
   1144 		}
   1145 
   1146 		/* If using a compressed file, initialize its info */
   1147 		/* (or abort with an error if kernel has no compression) */
   1148 		if (vio->vnd_flags & VNF_COMP) {
   1149 #ifdef VND_COMPRESSION
   1150 			struct vnd_comp_header *ch;
   1151 			int i;
   1152 			u_int32_t comp_size;
   1153 			u_int32_t comp_maxsize;
   1154 
   1155 			/* allocate space for compresed file header */
   1156 			ch = malloc(sizeof(struct vnd_comp_header),
   1157 			M_TEMP, M_WAITOK);
   1158 
   1159 			/* read compressed file header */
   1160 			error = vn_rdwr(UIO_READ, nd.ni_vp, (void *)ch,
   1161 			  sizeof(struct vnd_comp_header), 0, UIO_SYSSPACE,
   1162 			  IO_UNIT|IO_NODELOCKED, l->l_cred, NULL, NULL);
   1163 			if (error) {
   1164 				free(ch, M_TEMP);
   1165 				VOP_UNLOCK(nd.ni_vp);
   1166 				goto close_and_exit;
   1167 			}
   1168 
   1169 			/* save some header info */
   1170 			vnd->sc_comp_blksz = ntohl(ch->block_size);
   1171 			/* note last offset is the file byte size */
   1172 			vnd->sc_comp_numoffs = ntohl(ch->num_blocks)+1;
   1173 			free(ch, M_TEMP);
   1174 			if (vnd->sc_comp_blksz == 0 ||
   1175 			    vnd->sc_comp_blksz % DEV_BSIZE !=0) {
   1176 				VOP_UNLOCK(nd.ni_vp);
   1177 				error = EINVAL;
   1178 				goto close_and_exit;
   1179 			}
   1180 			if (sizeof(struct vnd_comp_header) +
   1181 			  sizeof(u_int64_t) * vnd->sc_comp_numoffs >
   1182 			  vattr.va_size) {
   1183 				VOP_UNLOCK(nd.ni_vp);
   1184 				error = EINVAL;
   1185 				goto close_and_exit;
   1186 			}
   1187 
   1188 			/* set decompressed file size */
   1189 			vattr.va_size =
   1190 			    ((u_quad_t)vnd->sc_comp_numoffs - 1) *
   1191 			     (u_quad_t)vnd->sc_comp_blksz;
   1192 
   1193 			/* allocate space for all the compressed offsets */
   1194 			vnd->sc_comp_offsets =
   1195 			malloc(sizeof(u_int64_t) * vnd->sc_comp_numoffs,
   1196 			M_DEVBUF, M_WAITOK);
   1197 
   1198 			/* read in the offsets */
   1199 			error = vn_rdwr(UIO_READ, nd.ni_vp,
   1200 			  (void *)vnd->sc_comp_offsets,
   1201 			  sizeof(u_int64_t) * vnd->sc_comp_numoffs,
   1202 			  sizeof(struct vnd_comp_header), UIO_SYSSPACE,
   1203 			  IO_UNIT|IO_NODELOCKED, l->l_cred, NULL, NULL);
   1204 			if (error) {
   1205 				VOP_UNLOCK(nd.ni_vp);
   1206 				goto close_and_exit;
   1207 			}
   1208 			/*
   1209 			 * find largest block size (used for allocation limit).
   1210 			 * Also convert offset to native byte order.
   1211 			 */
   1212 			comp_maxsize = 0;
   1213 			for (i = 0; i < vnd->sc_comp_numoffs - 1; i++) {
   1214 				vnd->sc_comp_offsets[i] =
   1215 				  be64toh(vnd->sc_comp_offsets[i]);
   1216 				comp_size = be64toh(vnd->sc_comp_offsets[i + 1])
   1217 				  - vnd->sc_comp_offsets[i];
   1218 				if (comp_size > comp_maxsize)
   1219 					comp_maxsize = comp_size;
   1220 			}
   1221 			vnd->sc_comp_offsets[vnd->sc_comp_numoffs - 1] =
   1222 			  be64toh(vnd->sc_comp_offsets[vnd->sc_comp_numoffs - 1]);
   1223 
   1224 			/* create compressed data buffer */
   1225 			vnd->sc_comp_buff = malloc(comp_maxsize,
   1226 			  M_DEVBUF, M_WAITOK);
   1227 
   1228 			/* create decompressed buffer */
   1229 			vnd->sc_comp_decombuf = malloc(vnd->sc_comp_blksz,
   1230 			  M_DEVBUF, M_WAITOK);
   1231 			vnd->sc_comp_buffblk = -1;
   1232 
   1233 			/* Initialize decompress stream */
   1234 			memset(&vnd->sc_comp_stream, 0, sizeof(z_stream));
   1235 			vnd->sc_comp_stream.zalloc = vnd_alloc;
   1236 			vnd->sc_comp_stream.zfree = vnd_free;
   1237 			error = inflateInit2(&vnd->sc_comp_stream, MAX_WBITS);
   1238 			if (error) {
   1239 				if (vnd->sc_comp_stream.msg)
   1240 					printf("vnd%d: compressed file, %s\n",
   1241 					  unit, vnd->sc_comp_stream.msg);
   1242 				VOP_UNLOCK(nd.ni_vp);
   1243 				error = EINVAL;
   1244 				goto close_and_exit;
   1245 			}
   1246 
   1247 			vnd->sc_flags |= VNF_COMP | VNF_READONLY;
   1248 #else /* !VND_COMPRESSION */
   1249 			VOP_UNLOCK(nd.ni_vp);
   1250 			error = EOPNOTSUPP;
   1251 			goto close_and_exit;
   1252 #endif /* VND_COMPRESSION */
   1253 		}
   1254 
   1255 		VOP_UNLOCK(nd.ni_vp);
   1256 		vnd->sc_vp = nd.ni_vp;
   1257 		vnd->sc_size = btodb(vattr.va_size);	/* note truncation */
   1258 
   1259 		/*
   1260 		 * Use pseudo-geometry specified.  If none was provided,
   1261 		 * use "standard" Adaptec fictitious geometry.
   1262 		 */
   1263 		if (vio->vnd_flags & VNDIOF_HASGEOM) {
   1264 
   1265 			memcpy(&vnd->sc_geom, &vio->vnd_geom,
   1266 			    sizeof(vio->vnd_geom));
   1267 
   1268 			/*
   1269 			 * Sanity-check the sector size.
   1270 			 * XXX Don't allow secsize < DEV_BSIZE.	 Should
   1271 			 * XXX we?
   1272 			 */
   1273 			if (vnd->sc_geom.vng_secsize < DEV_BSIZE ||
   1274 			    (vnd->sc_geom.vng_secsize % DEV_BSIZE) != 0 ||
   1275 			    vnd->sc_geom.vng_ncylinders == 0 ||
   1276 			    (vnd->sc_geom.vng_ntracks *
   1277 			     vnd->sc_geom.vng_nsectors) == 0) {
   1278 				error = EINVAL;
   1279 				goto close_and_exit;
   1280 			}
   1281 
   1282 			/*
   1283 			 * Compute the size (in DEV_BSIZE blocks) specified
   1284 			 * by the geometry.
   1285 			 */
   1286 			geomsize = (int64_t)vnd->sc_geom.vng_nsectors *
   1287 			    vnd->sc_geom.vng_ntracks *
   1288 			    vnd->sc_geom.vng_ncylinders *
   1289 			    (vnd->sc_geom.vng_secsize / DEV_BSIZE);
   1290 
   1291 			/*
   1292 			 * Sanity-check the size against the specified
   1293 			 * geometry.
   1294 			 */
   1295 			if (vnd->sc_size < geomsize) {
   1296 				error = EINVAL;
   1297 				goto close_and_exit;
   1298 			}
   1299 		} else if (vnd->sc_size >= (32 * 64)) {
   1300 			/*
   1301 			 * Size must be at least 2048 DEV_BSIZE blocks
   1302 			 * (1M) in order to use this geometry.
   1303 			 */
   1304 			vnd->sc_geom.vng_secsize = DEV_BSIZE;
   1305 			vnd->sc_geom.vng_nsectors = 32;
   1306 			vnd->sc_geom.vng_ntracks = 64;
   1307 			vnd->sc_geom.vng_ncylinders = vnd->sc_size / (64 * 32);
   1308 		} else {
   1309 			vnd->sc_geom.vng_secsize = DEV_BSIZE;
   1310 			vnd->sc_geom.vng_nsectors = 1;
   1311 			vnd->sc_geom.vng_ntracks = 1;
   1312 			vnd->sc_geom.vng_ncylinders = vnd->sc_size;
   1313 		}
   1314 
   1315 		vnd_set_geometry(vnd);
   1316 
   1317 		if (vio->vnd_flags & VNDIOF_READONLY) {
   1318 			vnd->sc_flags |= VNF_READONLY;
   1319 		}
   1320 
   1321 		if ((error = vndsetcred(vnd, l->l_cred)) != 0)
   1322 			goto close_and_exit;
   1323 
   1324 		vndthrottle(vnd, vnd->sc_vp);
   1325 		vio->vnd_osize = dbtob(vnd->sc_size);
   1326 #ifdef VNDIOCSET50
   1327 		if (cmd != VNDIOCSET50)
   1328 #endif
   1329 			vio->vnd_size = dbtob(vnd->sc_size);
   1330 		vnd->sc_flags |= VNF_INITED;
   1331 
   1332 		/* create the kernel thread, wait for it to be up */
   1333 		error = kthread_create(PRI_NONE, 0, NULL, vndthread, vnd,
   1334 		    &vnd->sc_kthread, "%s", device_xname(vnd->sc_dev));
   1335 		if (error)
   1336 			goto close_and_exit;
   1337 		while ((vnd->sc_flags & VNF_KTHREAD) == 0) {
   1338 			tsleep(&vnd->sc_kthread, PRIBIO, "vndthr", 0);
   1339 		}
   1340 #ifdef DEBUG
   1341 		if (vnddebug & VDB_INIT)
   1342 			printf("vndioctl: SET vp %p size 0x%lx %d/%d/%d/%d\n",
   1343 			    vnd->sc_vp, (unsigned long) vnd->sc_size,
   1344 			    vnd->sc_geom.vng_secsize,
   1345 			    vnd->sc_geom.vng_nsectors,
   1346 			    vnd->sc_geom.vng_ntracks,
   1347 			    vnd->sc_geom.vng_ncylinders);
   1348 #endif
   1349 
   1350 		/* Attach the disk. */
   1351 		disk_attach(&vnd->sc_dkdev);
   1352 
   1353 		/* Initialize the xfer and buffer pools. */
   1354 		pool_init(&vnd->sc_vxpool, sizeof(struct vndxfer), 0,
   1355 		    0, 0, "vndxpl", NULL, IPL_BIO);
   1356 
   1357 		vndunlock(vnd);
   1358 
   1359 		pathbuf_destroy(pb);
   1360 
   1361 		/* Discover wedges on this disk */
   1362 		dkwedge_discover(&vnd->sc_dkdev);
   1363 
   1364 		break;
   1365 
   1366 close_and_exit:
   1367 		(void) vn_close(nd.ni_vp, fflags, l->l_cred);
   1368 		pathbuf_destroy(pb);
   1369 unlock_and_exit:
   1370 #ifdef VND_COMPRESSION
   1371 		/* free any allocated memory (for compressed file) */
   1372 		if (vnd->sc_comp_offsets) {
   1373 			free(vnd->sc_comp_offsets, M_DEVBUF);
   1374 			vnd->sc_comp_offsets = NULL;
   1375 		}
   1376 		if (vnd->sc_comp_buff) {
   1377 			free(vnd->sc_comp_buff, M_DEVBUF);
   1378 			vnd->sc_comp_buff = NULL;
   1379 		}
   1380 		if (vnd->sc_comp_decombuf) {
   1381 			free(vnd->sc_comp_decombuf, M_DEVBUF);
   1382 			vnd->sc_comp_decombuf = NULL;
   1383 		}
   1384 #endif /* VND_COMPRESSION */
   1385 		vndunlock(vnd);
   1386 		return error;
   1387 
   1388 #ifdef VNDIOCCLR50
   1389 	case VNDIOCCLR50:
   1390 #endif
   1391 	case VNDIOCCLR:
   1392 		part = DISKPART(dev);
   1393 		pmask = (1 << part);
   1394 		force = (vio->vnd_flags & VNDIOF_FORCE) != 0;
   1395 
   1396 		if ((error = vnddoclear(vnd, pmask, minor(dev), force)) != 0)
   1397 			return error;
   1398 
   1399 		break;
   1400 
   1401 #ifdef COMPAT_30
   1402 	case VNDIOCGET30: {
   1403 		struct vnd_user30 *vnu;
   1404 		struct vattr va;
   1405 		vnu = (struct vnd_user30 *)data;
   1406 		KASSERT(l);
   1407 		switch (error = vnd_cget(l, unit, &vnu->vnu_unit, &va)) {
   1408 		case 0:
   1409 			vnu->vnu_dev = va.va_fsid;
   1410 			vnu->vnu_ino = va.va_fileid;
   1411 			break;
   1412 		case -1:
   1413 			/* unused is not an error */
   1414 			vnu->vnu_dev = 0;
   1415 			vnu->vnu_ino = 0;
   1416 			break;
   1417 		default:
   1418 			return error;
   1419 		}
   1420 		break;
   1421 	}
   1422 #endif
   1423 
   1424 #ifdef COMPAT_50
   1425 	case VNDIOCGET50: {
   1426 		struct vnd_user50 *vnu;
   1427 		struct vattr va;
   1428 		vnu = (struct vnd_user50 *)data;
   1429 		KASSERT(l);
   1430 		switch (error = vnd_cget(l, unit, &vnu->vnu_unit, &va)) {
   1431 		case 0:
   1432 			vnu->vnu_dev = va.va_fsid;
   1433 			vnu->vnu_ino = va.va_fileid;
   1434 			break;
   1435 		case -1:
   1436 			/* unused is not an error */
   1437 			vnu->vnu_dev = 0;
   1438 			vnu->vnu_ino = 0;
   1439 			break;
   1440 		default:
   1441 			return error;
   1442 		}
   1443 		break;
   1444 	}
   1445 #endif
   1446 
   1447 	case VNDIOCGET: {
   1448 		struct vnd_user *vnu;
   1449 		struct vattr va;
   1450 		vnu = (struct vnd_user *)data;
   1451 		KASSERT(l);
   1452 		switch (error = vnd_cget(l, unit, &vnu->vnu_unit, &va)) {
   1453 		case 0:
   1454 			vnu->vnu_dev = va.va_fsid;
   1455 			vnu->vnu_ino = va.va_fileid;
   1456 			break;
   1457 		case -1:
   1458 			/* unused is not an error */
   1459 			vnu->vnu_dev = 0;
   1460 			vnu->vnu_ino = 0;
   1461 			break;
   1462 		default:
   1463 			return error;
   1464 		}
   1465 		break;
   1466 	}
   1467 
   1468 	case DIOCWDINFO:
   1469 	case DIOCSDINFO:
   1470 #ifdef __HAVE_OLD_DISKLABEL
   1471 	case ODIOCWDINFO:
   1472 	case ODIOCSDINFO:
   1473 #endif
   1474 	{
   1475 		struct disklabel *lp;
   1476 
   1477 		if ((error = vndlock(vnd)) != 0)
   1478 			return error;
   1479 
   1480 		vnd->sc_flags |= VNF_LABELLING;
   1481 
   1482 #ifdef __HAVE_OLD_DISKLABEL
   1483 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
   1484 			memset(&newlabel, 0, sizeof newlabel);
   1485 			memcpy(&newlabel, data, sizeof (struct olddisklabel));
   1486 			lp = &newlabel;
   1487 		} else
   1488 #endif
   1489 		lp = (struct disklabel *)data;
   1490 
   1491 		error = setdisklabel(vnd->sc_dkdev.dk_label,
   1492 		    lp, 0, vnd->sc_dkdev.dk_cpulabel);
   1493 		if (error == 0) {
   1494 			if (cmd == DIOCWDINFO
   1495 #ifdef __HAVE_OLD_DISKLABEL
   1496 			    || cmd == ODIOCWDINFO
   1497 #endif
   1498 			   )
   1499 				error = writedisklabel(VNDLABELDEV(dev),
   1500 				    vndstrategy, vnd->sc_dkdev.dk_label,
   1501 				    vnd->sc_dkdev.dk_cpulabel);
   1502 		}
   1503 
   1504 		vnd->sc_flags &= ~VNF_LABELLING;
   1505 
   1506 		vndunlock(vnd);
   1507 
   1508 		if (error)
   1509 			return error;
   1510 		break;
   1511 	}
   1512 
   1513 	case DIOCKLABEL:
   1514 		if (*(int *)data != 0)
   1515 			vnd->sc_flags |= VNF_KLABEL;
   1516 		else
   1517 			vnd->sc_flags &= ~VNF_KLABEL;
   1518 		break;
   1519 
   1520 	case DIOCWLABEL:
   1521 		if (*(int *)data != 0)
   1522 			vnd->sc_flags |= VNF_WLABEL;
   1523 		else
   1524 			vnd->sc_flags &= ~VNF_WLABEL;
   1525 		break;
   1526 
   1527 	case DIOCGDEFLABEL:
   1528 		vndgetdefaultlabel(vnd, (struct disklabel *)data);
   1529 		break;
   1530 
   1531 #ifdef __HAVE_OLD_DISKLABEL
   1532 	case ODIOCGDEFLABEL:
   1533 		vndgetdefaultlabel(vnd, &newlabel);
   1534 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
   1535 			return ENOTTY;
   1536 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
   1537 		break;
   1538 #endif
   1539 
   1540 	case DIOCCACHESYNC:
   1541 		vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
   1542 		error = VOP_FSYNC(vnd->sc_vp, vnd->sc_cred,
   1543 		    FSYNC_WAIT | FSYNC_DATAONLY | FSYNC_CACHE, 0, 0);
   1544 		VOP_UNLOCK(vnd->sc_vp);
   1545 		return error;
   1546 
   1547 	default:
   1548 		return ENOTTY;
   1549 	}
   1550 
   1551 	return 0;
   1552 }
   1553 
   1554 /*
   1555  * Duplicate the current processes' credentials.  Since we are called only
   1556  * as the result of a SET ioctl and only root can do that, any future access
   1557  * to this "disk" is essentially as root.  Note that credentials may change
   1558  * if some other uid can write directly to the mapped file (NFS).
   1559  */
   1560 static int
   1561 vndsetcred(struct vnd_softc *vnd, kauth_cred_t cred)
   1562 {
   1563 	struct uio auio;
   1564 	struct iovec aiov;
   1565 	char *tmpbuf;
   1566 	int error;
   1567 
   1568 	vnd->sc_cred = kauth_cred_dup(cred);
   1569 	tmpbuf = malloc(DEV_BSIZE, M_TEMP, M_WAITOK);
   1570 
   1571 	/* XXX: Horrible kludge to establish credentials for NFS */
   1572 	aiov.iov_base = tmpbuf;
   1573 	aiov.iov_len = min(DEV_BSIZE, dbtob(vnd->sc_size));
   1574 	auio.uio_iov = &aiov;
   1575 	auio.uio_iovcnt = 1;
   1576 	auio.uio_offset = 0;
   1577 	auio.uio_rw = UIO_READ;
   1578 	auio.uio_resid = aiov.iov_len;
   1579 	UIO_SETUP_SYSSPACE(&auio);
   1580 	vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
   1581 	error = VOP_READ(vnd->sc_vp, &auio, 0, vnd->sc_cred);
   1582 	if (error == 0) {
   1583 		/*
   1584 		 * Because vnd does all IO directly through the vnode
   1585 		 * we need to flush (at least) the buffer from the above
   1586 		 * VOP_READ from the buffer cache to prevent cache
   1587 		 * incoherencies.  Also, be careful to write dirty
   1588 		 * buffers back to stable storage.
   1589 		 */
   1590 		error = vinvalbuf(vnd->sc_vp, V_SAVE, vnd->sc_cred,
   1591 			    curlwp, 0, 0);
   1592 	}
   1593 	VOP_UNLOCK(vnd->sc_vp);
   1594 
   1595 	free(tmpbuf, M_TEMP);
   1596 	return error;
   1597 }
   1598 
   1599 /*
   1600  * Set maxactive based on FS type
   1601  */
   1602 static void
   1603 vndthrottle(struct vnd_softc *vnd, struct vnode *vp)
   1604 {
   1605 
   1606 	if (vp->v_tag == VT_NFS)
   1607 		vnd->sc_maxactive = 2;
   1608 	else
   1609 		vnd->sc_maxactive = 8;
   1610 
   1611 	if (vnd->sc_maxactive < 1)
   1612 		vnd->sc_maxactive = 1;
   1613 }
   1614 
   1615 #if 0
   1616 static void
   1617 vndshutdown(void)
   1618 {
   1619 	struct vnd_softc *vnd;
   1620 
   1621 	for (vnd = &vnd_softc[0]; vnd < &vnd_softc[numvnd]; vnd++)
   1622 		if (vnd->sc_flags & VNF_INITED)
   1623 			vndclear(vnd);
   1624 }
   1625 #endif
   1626 
   1627 static void
   1628 vndclear(struct vnd_softc *vnd, int myminor)
   1629 {
   1630 	struct vnode *vp = vnd->sc_vp;
   1631 	int fflags = FREAD;
   1632 	int bmaj, cmaj, i, mn;
   1633 	int s;
   1634 
   1635 #ifdef DEBUG
   1636 	if (vnddebug & VDB_FOLLOW)
   1637 		printf("vndclear(%p): vp %p\n", vnd, vp);
   1638 #endif
   1639 	/* locate the major number */
   1640 	bmaj = bdevsw_lookup_major(&vnd_bdevsw);
   1641 	cmaj = cdevsw_lookup_major(&vnd_cdevsw);
   1642 
   1643 	/* Nuke the vnodes for any open instances */
   1644 	for (i = 0; i < MAXPARTITIONS; i++) {
   1645 		mn = DISKMINOR(device_unit(vnd->sc_dev), i);
   1646 		vdevgone(bmaj, mn, mn, VBLK);
   1647 		if (mn != myminor) /* XXX avoid to kill own vnode */
   1648 			vdevgone(cmaj, mn, mn, VCHR);
   1649 	}
   1650 
   1651 	if ((vnd->sc_flags & VNF_READONLY) == 0)
   1652 		fflags |= FWRITE;
   1653 
   1654 	s = splbio();
   1655 	bufq_drain(vnd->sc_tab);
   1656 	splx(s);
   1657 
   1658 	vnd->sc_flags |= VNF_VUNCONF;
   1659 	wakeup(&vnd->sc_tab);
   1660 	while (vnd->sc_flags & VNF_KTHREAD)
   1661 		tsleep(&vnd->sc_kthread, PRIBIO, "vnthr", 0);
   1662 
   1663 #ifdef VND_COMPRESSION
   1664 	/* free the compressed file buffers */
   1665 	if (vnd->sc_flags & VNF_COMP) {
   1666 		if (vnd->sc_comp_offsets) {
   1667 			free(vnd->sc_comp_offsets, M_DEVBUF);
   1668 			vnd->sc_comp_offsets = NULL;
   1669 		}
   1670 		if (vnd->sc_comp_buff) {
   1671 			free(vnd->sc_comp_buff, M_DEVBUF);
   1672 			vnd->sc_comp_buff = NULL;
   1673 		}
   1674 		if (vnd->sc_comp_decombuf) {
   1675 			free(vnd->sc_comp_decombuf, M_DEVBUF);
   1676 			vnd->sc_comp_decombuf = NULL;
   1677 		}
   1678 	}
   1679 #endif /* VND_COMPRESSION */
   1680 	vnd->sc_flags &=
   1681 	    ~(VNF_INITED | VNF_READONLY | VNF_VLABEL
   1682 	      | VNF_VUNCONF | VNF_COMP | VNF_CLEARING);
   1683 	if (vp == NULL)
   1684 		panic("vndclear: null vp");
   1685 	(void) vn_close(vp, fflags, vnd->sc_cred);
   1686 	kauth_cred_free(vnd->sc_cred);
   1687 	vnd->sc_vp = NULL;
   1688 	vnd->sc_cred = NULL;
   1689 	vnd->sc_size = 0;
   1690 }
   1691 
   1692 static int
   1693 vndsize(dev_t dev)
   1694 {
   1695 	struct vnd_softc *sc;
   1696 	struct disklabel *lp;
   1697 	int part, unit, omask;
   1698 	int size;
   1699 
   1700 	unit = vndunit(dev);
   1701 	sc = device_lookup_private(&vnd_cd, unit);
   1702 	if (sc == NULL)
   1703 		return -1;
   1704 
   1705 	if ((sc->sc_flags & VNF_INITED) == 0)
   1706 		return -1;
   1707 
   1708 	part = DISKPART(dev);
   1709 	omask = sc->sc_dkdev.dk_openmask & (1 << part);
   1710 	lp = sc->sc_dkdev.dk_label;
   1711 
   1712 	if (omask == 0 && vndopen(dev, 0, S_IFBLK, curlwp))	/* XXX */
   1713 		return -1;
   1714 
   1715 	if (lp->d_partitions[part].p_fstype != FS_SWAP)
   1716 		size = -1;
   1717 	else
   1718 		size = lp->d_partitions[part].p_size *
   1719 		    (lp->d_secsize / DEV_BSIZE);
   1720 
   1721 	if (omask == 0 && vndclose(dev, 0, S_IFBLK, curlwp))	/* XXX */
   1722 		return -1;
   1723 
   1724 	return size;
   1725 }
   1726 
   1727 static int
   1728 vnddump(dev_t dev, daddr_t blkno, void *va,
   1729     size_t size)
   1730 {
   1731 
   1732 	/* Not implemented. */
   1733 	return ENXIO;
   1734 }
   1735 
   1736 static void
   1737 vndgetdefaultlabel(struct vnd_softc *sc, struct disklabel *lp)
   1738 {
   1739 	struct vndgeom *vng = &sc->sc_geom;
   1740 	struct partition *pp;
   1741 	unsigned spb;
   1742 
   1743 	memset(lp, 0, sizeof(*lp));
   1744 
   1745 	spb = vng->vng_secsize / DEV_BSIZE;
   1746 	if (sc->sc_size / spb > UINT32_MAX)
   1747 		lp->d_secperunit = UINT32_MAX;
   1748 	else
   1749 		lp->d_secperunit = sc->sc_size / spb;
   1750 	lp->d_secsize = vng->vng_secsize;
   1751 	lp->d_nsectors = vng->vng_nsectors;
   1752 	lp->d_ntracks = vng->vng_ntracks;
   1753 	lp->d_ncylinders = vng->vng_ncylinders;
   1754 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
   1755 
   1756 	strncpy(lp->d_typename, "vnd", sizeof(lp->d_typename));
   1757 	lp->d_type = DTYPE_VND;
   1758 	strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
   1759 	lp->d_rpm = 3600;
   1760 	lp->d_interleave = 1;
   1761 	lp->d_flags = 0;
   1762 
   1763 	pp = &lp->d_partitions[RAW_PART];
   1764 	pp->p_offset = 0;
   1765 	pp->p_size = lp->d_secperunit;
   1766 	pp->p_fstype = FS_UNUSED;
   1767 	lp->d_npartitions = RAW_PART + 1;
   1768 
   1769 	lp->d_magic = DISKMAGIC;
   1770 	lp->d_magic2 = DISKMAGIC;
   1771 	lp->d_checksum = dkcksum(lp);
   1772 }
   1773 
   1774 /*
   1775  * Read the disklabel from a vnd.  If one is not present, create a fake one.
   1776  */
   1777 static void
   1778 vndgetdisklabel(dev_t dev, struct vnd_softc *sc)
   1779 {
   1780 	const char *errstring;
   1781 	struct disklabel *lp = sc->sc_dkdev.dk_label;
   1782 	struct cpu_disklabel *clp = sc->sc_dkdev.dk_cpulabel;
   1783 	int i;
   1784 
   1785 	memset(clp, 0, sizeof(*clp));
   1786 
   1787 	vndgetdefaultlabel(sc, lp);
   1788 
   1789 	/*
   1790 	 * Call the generic disklabel extraction routine.
   1791 	 */
   1792 	errstring = readdisklabel(VNDLABELDEV(dev), vndstrategy, lp, clp);
   1793 	if (errstring) {
   1794 		/*
   1795 		 * Lack of disklabel is common, but we print the warning
   1796 		 * anyway, since it might contain other useful information.
   1797 		 */
   1798 		aprint_normal_dev(sc->sc_dev, "%s\n", errstring);
   1799 
   1800 		/*
   1801 		 * For historical reasons, if there's no disklabel
   1802 		 * present, all partitions must be FS_BSDFFS and
   1803 		 * occupy the entire disk.
   1804 		 */
   1805 		for (i = 0; i < MAXPARTITIONS; i++) {
   1806 			/*
   1807 			 * Don't wipe out port specific hack (such as
   1808 			 * dos partition hack of i386 port).
   1809 			 */
   1810 			if (lp->d_partitions[i].p_size != 0)
   1811 				continue;
   1812 
   1813 			lp->d_partitions[i].p_size = lp->d_secperunit;
   1814 			lp->d_partitions[i].p_offset = 0;
   1815 			lp->d_partitions[i].p_fstype = FS_BSDFFS;
   1816 		}
   1817 
   1818 		strncpy(lp->d_packname, "default label",
   1819 		    sizeof(lp->d_packname));
   1820 
   1821 		lp->d_npartitions = MAXPARTITIONS;
   1822 		lp->d_checksum = dkcksum(lp);
   1823 	}
   1824 }
   1825 
   1826 /*
   1827  * Wait interruptibly for an exclusive lock.
   1828  *
   1829  * XXX
   1830  * Several drivers do this; it should be abstracted and made MP-safe.
   1831  */
   1832 static int
   1833 vndlock(struct vnd_softc *sc)
   1834 {
   1835 	int error;
   1836 
   1837 	while ((sc->sc_flags & VNF_LOCKED) != 0) {
   1838 		sc->sc_flags |= VNF_WANTED;
   1839 		if ((error = tsleep(sc, PRIBIO | PCATCH, "vndlck", 0)) != 0)
   1840 			return error;
   1841 	}
   1842 	sc->sc_flags |= VNF_LOCKED;
   1843 	return 0;
   1844 }
   1845 
   1846 /*
   1847  * Unlock and wake up any waiters.
   1848  */
   1849 static void
   1850 vndunlock(struct vnd_softc *sc)
   1851 {
   1852 
   1853 	sc->sc_flags &= ~VNF_LOCKED;
   1854 	if ((sc->sc_flags & VNF_WANTED) != 0) {
   1855 		sc->sc_flags &= ~VNF_WANTED;
   1856 		wakeup(sc);
   1857 	}
   1858 }
   1859 
   1860 #ifdef VND_COMPRESSION
   1861 /* compressed file read */
   1862 static void
   1863 compstrategy(struct buf *bp, off_t bn)
   1864 {
   1865 	int error;
   1866 	int unit = vndunit(bp->b_dev);
   1867 	struct vnd_softc *vnd =
   1868 	    device_lookup_private(&vnd_cd, unit);
   1869 	u_int32_t comp_block;
   1870 	struct uio auio;
   1871 	char *addr;
   1872 	int s;
   1873 
   1874 	/* set up constants for data move */
   1875 	auio.uio_rw = UIO_READ;
   1876 	UIO_SETUP_SYSSPACE(&auio);
   1877 
   1878 	/* read, and transfer the data */
   1879 	addr = bp->b_data;
   1880 	bp->b_resid = bp->b_bcount;
   1881 	s = splbio();
   1882 	while (bp->b_resid > 0) {
   1883 		unsigned length;
   1884 		size_t length_in_buffer;
   1885 		u_int32_t offset_in_buffer;
   1886 		struct iovec aiov;
   1887 
   1888 		/* calculate the compressed block number */
   1889 		comp_block = bn / (off_t)vnd->sc_comp_blksz;
   1890 
   1891 		/* check for good block number */
   1892 		if (comp_block >= vnd->sc_comp_numoffs) {
   1893 			bp->b_error = EINVAL;
   1894 			splx(s);
   1895 			return;
   1896 		}
   1897 
   1898 		/* read in the compressed block, if not in buffer */
   1899 		if (comp_block != vnd->sc_comp_buffblk) {
   1900 			length = vnd->sc_comp_offsets[comp_block + 1] -
   1901 			    vnd->sc_comp_offsets[comp_block];
   1902 			vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
   1903 			error = vn_rdwr(UIO_READ, vnd->sc_vp, vnd->sc_comp_buff,
   1904 			    length, vnd->sc_comp_offsets[comp_block],
   1905 			    UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, vnd->sc_cred,
   1906 			    NULL, NULL);
   1907 			if (error) {
   1908 				bp->b_error = error;
   1909 				VOP_UNLOCK(vnd->sc_vp);
   1910 				splx(s);
   1911 				return;
   1912 			}
   1913 			/* uncompress the buffer */
   1914 			vnd->sc_comp_stream.next_in = vnd->sc_comp_buff;
   1915 			vnd->sc_comp_stream.avail_in = length;
   1916 			vnd->sc_comp_stream.next_out = vnd->sc_comp_decombuf;
   1917 			vnd->sc_comp_stream.avail_out = vnd->sc_comp_blksz;
   1918 			inflateReset(&vnd->sc_comp_stream);
   1919 			error = inflate(&vnd->sc_comp_stream, Z_FINISH);
   1920 			if (error != Z_STREAM_END) {
   1921 				if (vnd->sc_comp_stream.msg)
   1922 					aprint_normal_dev(vnd->sc_dev,
   1923 					    "compressed file, %s\n",
   1924 					    vnd->sc_comp_stream.msg);
   1925 				bp->b_error = EBADMSG;
   1926 				VOP_UNLOCK(vnd->sc_vp);
   1927 				splx(s);
   1928 				return;
   1929 			}
   1930 			vnd->sc_comp_buffblk = comp_block;
   1931 			VOP_UNLOCK(vnd->sc_vp);
   1932 		}
   1933 
   1934 		/* transfer the usable uncompressed data */
   1935 		offset_in_buffer = bn % (off_t)vnd->sc_comp_blksz;
   1936 		length_in_buffer = vnd->sc_comp_blksz - offset_in_buffer;
   1937 		if (length_in_buffer > bp->b_resid)
   1938 			length_in_buffer = bp->b_resid;
   1939 		auio.uio_iov = &aiov;
   1940 		auio.uio_iovcnt = 1;
   1941 		aiov.iov_base = addr;
   1942 		aiov.iov_len = length_in_buffer;
   1943 		auio.uio_resid = aiov.iov_len;
   1944 		auio.uio_offset = 0;
   1945 		error = uiomove(vnd->sc_comp_decombuf + offset_in_buffer,
   1946 		    length_in_buffer, &auio);
   1947 		if (error) {
   1948 			bp->b_error = error;
   1949 			splx(s);
   1950 			return;
   1951 		}
   1952 
   1953 		bn += length_in_buffer;
   1954 		addr += length_in_buffer;
   1955 		bp->b_resid -= length_in_buffer;
   1956 	}
   1957 	splx(s);
   1958 }
   1959 
   1960 /* compression memory allocation routines */
   1961 static void *
   1962 vnd_alloc(void *aux, u_int items, u_int siz)
   1963 {
   1964 	return malloc(items * siz, M_TEMP, M_NOWAIT);
   1965 }
   1966 
   1967 static void
   1968 vnd_free(void *aux, void *ptr)
   1969 {
   1970 	free(ptr, M_TEMP);
   1971 }
   1972 #endif /* VND_COMPRESSION */
   1973 
   1974 static void
   1975 vnd_set_geometry(struct vnd_softc *vnd)
   1976 {
   1977 	struct disk_geom *dg = &vnd->sc_dkdev.dk_geom;
   1978 
   1979 	memset(dg, 0, sizeof(*dg));
   1980 
   1981 	dg->dg_secperunit = (int64_t)vnd->sc_geom.vng_nsectors *
   1982 	    vnd->sc_geom.vng_ntracks * vnd->sc_geom.vng_ncylinders;
   1983 	dg->dg_secsize = vnd->sc_geom.vng_secsize;
   1984 	dg->dg_nsectors = vnd->sc_geom.vng_nsectors;
   1985 	dg->dg_ntracks = vnd->sc_geom.vng_ntracks;
   1986 	dg->dg_ncylinders = vnd->sc_geom.vng_ncylinders;
   1987 
   1988 #ifdef DEBUG
   1989 	if (vnddebug & VDB_LABEL) {
   1990 		printf("dg->dg_secperunit: %" PRId64 "\n", dg->dg_secperunit);
   1991 		printf("dg->dg_ncylinders: %u\n", dg->dg_ncylinders);
   1992 	}
   1993 #endif
   1994 	disk_set_info(vnd->sc_dev, &vnd->sc_dkdev, NULL);
   1995 }
   1996 
   1997 #ifdef _MODULE
   1998 
   1999 #include <sys/module.h>
   2000 
   2001 #ifdef VND_COMPRESSION
   2002 #define VND_DEPENDS "zlib"
   2003 #else
   2004 #define VND_DEPENDS NULL
   2005 #endif
   2006 
   2007 MODULE(MODULE_CLASS_DRIVER, vnd, VND_DEPENDS);
   2008 CFDRIVER_DECL(vnd, DV_DISK, NULL);
   2009 
   2010 static int
   2011 vnd_modcmd(modcmd_t cmd, void *arg)
   2012 {
   2013 	int bmajor = -1, cmajor = -1,  error = 0;
   2014 
   2015 	switch (cmd) {
   2016 	case MODULE_CMD_INIT:
   2017 		error = config_cfdriver_attach(&vnd_cd);
   2018 		if (error)
   2019 			break;
   2020 
   2021 		error = config_cfattach_attach(vnd_cd.cd_name, &vnd_ca);
   2022 	        if (error) {
   2023 			config_cfdriver_detach(&vnd_cd);
   2024 			aprint_error("%s: unable to register cfattach\n",
   2025 			    vnd_cd.cd_name);
   2026 			break;
   2027 		}
   2028 
   2029 		error = devsw_attach("vnd", &vnd_bdevsw, &bmajor,
   2030 		    &vnd_cdevsw, &cmajor);
   2031 		if (error) {
   2032 			config_cfattach_detach(vnd_cd.cd_name, &vnd_ca);
   2033 			config_cfdriver_detach(&vnd_cd);
   2034 			break;
   2035 		}
   2036 
   2037 		break;
   2038 
   2039 	case MODULE_CMD_FINI:
   2040 		error = config_cfattach_detach(vnd_cd.cd_name, &vnd_ca);
   2041 		if (error)
   2042 			break;
   2043 		config_cfdriver_detach(&vnd_cd);
   2044 		devsw_detach(&vnd_bdevsw, &vnd_cdevsw);
   2045 		break;
   2046 
   2047 	case MODULE_CMD_STAT:
   2048 		return ENOTTY;
   2049 
   2050 	default:
   2051 		return ENOTTY;
   2052 	}
   2053 
   2054 	return error;
   2055 }
   2056 
   2057 #endif
   2058