Home | History | Annotate | Line # | Download | only in dev
vnd.c revision 1.218
      1 /*	$NetBSD: vnd.c,v 1.218 2011/06/29 09:12:42 hannken 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 1.218 2011/06/29 09:12:42 hannken Exp $");
     95 
     96 #if defined(_KERNEL_OPT)
     97 #include "opt_vnd.h"
     98 #include "opt_compat_netbsd.h"
     99 #endif
    100 
    101 #include <sys/param.h>
    102 #include <sys/systm.h>
    103 #include <sys/namei.h>
    104 #include <sys/proc.h>
    105 #include <sys/kthread.h>
    106 #include <sys/errno.h>
    107 #include <sys/buf.h>
    108 #include <sys/bufq.h>
    109 #include <sys/malloc.h>
    110 #include <sys/ioctl.h>
    111 #include <sys/disklabel.h>
    112 #include <sys/device.h>
    113 #include <sys/disk.h>
    114 #include <sys/stat.h>
    115 #include <sys/mount.h>
    116 #include <sys/vnode.h>
    117 #include <sys/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 #include <prop/proplib.h>
    131 
    132 #if defined(VNDDEBUG) && !defined(DEBUG)
    133 #define DEBUG
    134 #endif
    135 
    136 #ifdef DEBUG
    137 int dovndcluster = 1;
    138 #define VDB_FOLLOW	0x01
    139 #define VDB_INIT	0x02
    140 #define VDB_IO		0x04
    141 #define VDB_LABEL	0x08
    142 int vnddebug = 0x00;
    143 #endif
    144 
    145 #define vndunit(x)	DISKUNIT(x)
    146 
    147 struct vndxfer {
    148 	struct buf vx_buf;
    149 	struct vnd_softc *vx_vnd;
    150 };
    151 #define	VND_BUFTOXFER(bp)	((struct vndxfer *)(void *)bp)
    152 
    153 #define VND_GETXFER(vnd)	pool_get(&(vnd)->sc_vxpool, PR_WAITOK)
    154 #define VND_PUTXFER(vnd, vx)	pool_put(&(vnd)->sc_vxpool, (vx))
    155 
    156 #define VNDLABELDEV(dev) \
    157     (MAKEDISKDEV(major((dev)), vndunit((dev)), RAW_PART))
    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_properties(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 	vndopen, vndclose, vndstrategy, vndioctl, vnddump, vndsize, D_DISK
    201 };
    202 
    203 const struct cdevsw vnd_cdevsw = {
    204 	vndopen, vndclose, vndread, vndwrite, vndioctl,
    205 	nostop, notty, nopoll, nommap, nokqfilter, D_DISK
    206 };
    207 
    208 static int	vnd_match(device_t, cfdata_t, void *);
    209 static void	vnd_attach(device_t, device_t, void *);
    210 static int	vnd_detach(device_t, int);
    211 
    212 CFATTACH_DECL3_NEW(vnd, sizeof(struct vnd_softc),
    213     vnd_match, vnd_attach, vnd_detach, NULL, NULL, NULL, DVF_DETACH_SHUTDOWN);
    214 extern struct cfdriver vnd_cd;
    215 
    216 static struct vnd_softc	*vnd_spawn(int);
    217 int	vnd_destroy(device_t);
    218 
    219 static struct	dkdriver vnddkdriver = { vndstrategy, minphys };
    220 
    221 void
    222 vndattach(int num)
    223 {
    224 	int error;
    225 
    226 	error = config_cfattach_attach(vnd_cd.cd_name, &vnd_ca);
    227 	if (error)
    228 		aprint_error("%s: unable to register cfattach\n",
    229 		    vnd_cd.cd_name);
    230 }
    231 
    232 static int
    233 vnd_match(device_t self, cfdata_t cfdata, void *aux)
    234 {
    235 
    236 	return 1;
    237 }
    238 
    239 static void
    240 vnd_attach(device_t parent, device_t self, void *aux)
    241 {
    242 	struct vnd_softc *sc = device_private(self);
    243 
    244 	sc->sc_dev = self;
    245 	sc->sc_comp_offsets = NULL;
    246 	sc->sc_comp_buff = NULL;
    247 	sc->sc_comp_decombuf = NULL;
    248 	bufq_alloc(&sc->sc_tab, "disksort", BUFQ_SORT_RAWBLOCK);
    249 	disk_init(&sc->sc_dkdev, device_xname(self), &vnddkdriver);
    250 	if (!pmf_device_register(self, NULL, NULL))
    251 		aprint_error_dev(self, "couldn't establish power handler\n");
    252 }
    253 
    254 static int
    255 vnd_detach(device_t self, int flags)
    256 {
    257 	int error;
    258 	struct vnd_softc *sc = device_private(self);
    259 
    260 	if (sc->sc_flags & VNF_INITED) {
    261 		error = vnddoclear(sc, 0, -1, (flags & DETACH_FORCE) != 0);
    262 		if (error != 0)
    263 			return error;
    264 	}
    265 
    266 	pmf_device_deregister(self);
    267 	bufq_free(sc->sc_tab);
    268 	disk_destroy(&sc->sc_dkdev);
    269 
    270 	return 0;
    271 }
    272 
    273 static struct vnd_softc *
    274 vnd_spawn(int unit)
    275 {
    276 	cfdata_t cf;
    277 
    278 	cf = malloc(sizeof(*cf), M_DEVBUF, M_WAITOK);
    279 	cf->cf_name = vnd_cd.cd_name;
    280 	cf->cf_atname = vnd_cd.cd_name;
    281 	cf->cf_unit = unit;
    282 	cf->cf_fstate = FSTATE_STAR;
    283 
    284 	return device_private(config_attach_pseudo(cf));
    285 }
    286 
    287 int
    288 vnd_destroy(device_t dev)
    289 {
    290 	int error;
    291 	cfdata_t cf;
    292 
    293 	cf = device_cfdata(dev);
    294 	error = config_detach(dev, DETACH_QUIET);
    295 	if (error)
    296 		return error;
    297 	free(cf, M_DEVBUF);
    298 	return 0;
    299 }
    300 
    301 static int
    302 vndopen(dev_t dev, int flags, int mode, struct lwp *l)
    303 {
    304 	int unit = vndunit(dev);
    305 	struct vnd_softc *sc;
    306 	int error = 0, part, pmask;
    307 	struct disklabel *lp;
    308 
    309 #ifdef DEBUG
    310 	if (vnddebug & VDB_FOLLOW)
    311 		printf("vndopen(0x%"PRIx64", 0x%x, 0x%x, %p)\n", dev, flags, mode, l);
    312 #endif
    313 	sc = device_lookup_private(&vnd_cd, unit);
    314 	if (sc == NULL) {
    315 		sc = vnd_spawn(unit);
    316 		if (sc == NULL)
    317 			return ENOMEM;
    318 	}
    319 
    320 	if ((error = vndlock(sc)) != 0)
    321 		return error;
    322 
    323 	if ((sc->sc_flags & VNF_CLEARING) != 0) {
    324 		error = ENXIO;
    325 		goto done;
    326 	}
    327 
    328 	lp = sc->sc_dkdev.dk_label;
    329 
    330 	part = DISKPART(dev);
    331 	pmask = (1 << part);
    332 
    333 	/*
    334 	 * If we're initialized, check to see if there are any other
    335 	 * open partitions.  If not, then it's safe to update the
    336 	 * in-core disklabel.  Only read the disklabel if it is
    337 	 * not already valid.
    338 	 */
    339 	if ((sc->sc_flags & (VNF_INITED|VNF_VLABEL)) == VNF_INITED &&
    340 	    sc->sc_dkdev.dk_openmask == 0)
    341 		vndgetdisklabel(dev, sc);
    342 
    343 	/* Check that the partitions exists. */
    344 	if (part != RAW_PART) {
    345 		if (((sc->sc_flags & VNF_INITED) == 0) ||
    346 		    ((part >= lp->d_npartitions) ||
    347 		     (lp->d_partitions[part].p_fstype == FS_UNUSED))) {
    348 			error = ENXIO;
    349 			goto done;
    350 		}
    351 	}
    352 
    353 	/* Prevent our unit from being unconfigured while open. */
    354 	switch (mode) {
    355 	case S_IFCHR:
    356 		sc->sc_dkdev.dk_copenmask |= pmask;
    357 		break;
    358 
    359 	case S_IFBLK:
    360 		sc->sc_dkdev.dk_bopenmask |= pmask;
    361 		break;
    362 	}
    363 	sc->sc_dkdev.dk_openmask =
    364 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
    365 
    366  done:
    367 	vndunlock(sc);
    368 	return error;
    369 }
    370 
    371 static int
    372 vndclose(dev_t dev, int flags, int mode, struct lwp *l)
    373 {
    374 	int unit = vndunit(dev);
    375 	struct vnd_softc *sc;
    376 	int error = 0, part;
    377 
    378 #ifdef DEBUG
    379 	if (vnddebug & VDB_FOLLOW)
    380 		printf("vndclose(0x%"PRIx64", 0x%x, 0x%x, %p)\n", dev, flags, mode, l);
    381 #endif
    382 	sc = device_lookup_private(&vnd_cd, unit);
    383 	if (sc == NULL)
    384 		return ENXIO;
    385 
    386 	if ((error = vndlock(sc)) != 0)
    387 		return error;
    388 
    389 	part = DISKPART(dev);
    390 
    391 	/* ...that much closer to allowing unconfiguration... */
    392 	switch (mode) {
    393 	case S_IFCHR:
    394 		sc->sc_dkdev.dk_copenmask &= ~(1 << part);
    395 		break;
    396 
    397 	case S_IFBLK:
    398 		sc->sc_dkdev.dk_bopenmask &= ~(1 << part);
    399 		break;
    400 	}
    401 	sc->sc_dkdev.dk_openmask =
    402 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
    403 
    404 	vndunlock(sc);
    405 
    406 	if ((sc->sc_flags & VNF_INITED) == 0) {
    407 		if ((error = vnd_destroy(sc->sc_dev)) != 0) {
    408 			aprint_error_dev(sc->sc_dev,
    409 			    "unable to detach instance\n");
    410 			return error;
    411 		}
    412 	}
    413 
    414 	return 0;
    415 }
    416 
    417 /*
    418  * Queue the request, and wakeup the kernel thread to handle it.
    419  */
    420 static void
    421 vndstrategy(struct buf *bp)
    422 {
    423 	int unit = vndunit(bp->b_dev);
    424 	struct vnd_softc *vnd =
    425 	    device_lookup_private(&vnd_cd, unit);
    426 	struct disklabel *lp;
    427 	daddr_t blkno;
    428 	int s = splbio();
    429 
    430 	if (vnd == NULL) {
    431 		bp->b_error = ENXIO;
    432 		goto done;
    433 	}
    434 	lp = vnd->sc_dkdev.dk_label;
    435 
    436 	if ((vnd->sc_flags & VNF_INITED) == 0) {
    437 		bp->b_error = ENXIO;
    438 		goto done;
    439 	}
    440 
    441 	/*
    442 	 * The transfer must be a whole number of blocks.
    443 	 */
    444 	if ((bp->b_bcount % lp->d_secsize) != 0) {
    445 		bp->b_error = EINVAL;
    446 		goto done;
    447 	}
    448 
    449 	/*
    450 	 * check if we're read-only.
    451 	 */
    452 	if ((vnd->sc_flags & VNF_READONLY) && !(bp->b_flags & B_READ)) {
    453 		bp->b_error = EACCES;
    454 		goto done;
    455 	}
    456 
    457 	/* If it's a nil transfer, wake up the top half now. */
    458 	if (bp->b_bcount == 0) {
    459 		goto done;
    460 	}
    461 
    462 	/*
    463 	 * Do bounds checking and adjust transfer.  If there's an error,
    464 	 * the bounds check will flag that for us.
    465 	 */
    466 	if (DISKPART(bp->b_dev) == RAW_PART) {
    467 		if (bounds_check_with_mediasize(bp, DEV_BSIZE,
    468 		    vnd->sc_size) <= 0)
    469 			goto done;
    470 	} else {
    471 		if (bounds_check_with_label(&vnd->sc_dkdev,
    472 		    bp, vnd->sc_flags & (VNF_WLABEL|VNF_LABELLING)) <= 0)
    473 			goto done;
    474 	}
    475 
    476 	/*
    477 	 * Put the block number in terms of the logical blocksize
    478 	 * of the "device".
    479 	 */
    480 
    481 	blkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
    482 
    483 	/*
    484 	 * Translate the partition-relative block number to an absolute.
    485 	 */
    486 	if (DISKPART(bp->b_dev) != RAW_PART) {
    487 		struct partition *pp;
    488 
    489 		pp = &vnd->sc_dkdev.dk_label->d_partitions[
    490 		    DISKPART(bp->b_dev)];
    491 		blkno += pp->p_offset;
    492 	}
    493 	bp->b_rawblkno = blkno;
    494 
    495 #ifdef DEBUG
    496 	if (vnddebug & VDB_FOLLOW)
    497 		printf("vndstrategy(%p): unit %d\n", bp, unit);
    498 #endif
    499 	bufq_put(vnd->sc_tab, bp);
    500 	wakeup(&vnd->sc_tab);
    501 	splx(s);
    502 	return;
    503 
    504 done:
    505 	bp->b_resid = bp->b_bcount;
    506 	biodone(bp);
    507 	splx(s);
    508 }
    509 
    510 static bool
    511 vnode_has_strategy(struct vnd_softc *vnd)
    512 {
    513 	return vnode_has_op(vnd->sc_vp, VOFFSET(vop_bmap)) &&
    514 	    vnode_has_op(vnd->sc_vp, VOFFSET(vop_strategy));
    515 }
    516 
    517 /* XXX this function needs a reliable check to detect
    518  * sparse files. Otherwise, bmap/strategy may be used
    519  * and fail on non-allocated blocks. VOP_READ/VOP_WRITE
    520  * works on sparse files.
    521  */
    522 #if notyet
    523 static bool
    524 vnode_strategy_probe(struct vnd_softc *vnd)
    525 {
    526 	int error;
    527 	daddr_t nbn;
    528 
    529 	if (!vnode_has_strategy(vnd))
    530 		return false;
    531 
    532 	/* Convert the first logical block number to its
    533 	 * physical block number.
    534 	 */
    535 	error = 0;
    536 	vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
    537 	error = VOP_BMAP(vnd->sc_vp, 0, NULL, &nbn, NULL);
    538 	VOP_UNLOCK(vnd->sc_vp);
    539 
    540 	/* Test if that worked. */
    541 	if (error == 0 && (long)nbn == -1)
    542 		return false;
    543 
    544 	return true;
    545 }
    546 #endif
    547 
    548 static void
    549 vndthread(void *arg)
    550 {
    551 	struct vnd_softc *vnd = arg;
    552 	int s;
    553 
    554 	/* Determine whether we can *use* VOP_BMAP and VOP_STRATEGY to
    555 	 * directly access the backing vnode.  If we can, use these two
    556 	 * operations to avoid messing with the local buffer cache.
    557 	 * Otherwise fall back to regular VOP_READ/VOP_WRITE operations
    558 	 * which are guaranteed to work with any file system. */
    559 	if ((vnd->sc_flags & VNF_USE_VN_RDWR) == 0 &&
    560 	    ! vnode_has_strategy(vnd))
    561 		vnd->sc_flags |= VNF_USE_VN_RDWR;
    562 
    563 #ifdef DEBUG
    564 	if (vnddebug & VDB_INIT)
    565 		printf("vndthread: vp %p, %s\n", vnd->sc_vp,
    566 		    (vnd->sc_flags & VNF_USE_VN_RDWR) == 0 ?
    567 		    "using bmap/strategy operations" :
    568 		    "using read/write operations");
    569 #endif
    570 
    571 	s = splbio();
    572 	vnd->sc_flags |= VNF_KTHREAD;
    573 	wakeup(&vnd->sc_kthread);
    574 
    575 	/*
    576 	 * Dequeue requests and serve them depending on the available
    577 	 * vnode operations.
    578 	 */
    579 	while ((vnd->sc_flags & VNF_VUNCONF) == 0) {
    580 		struct vndxfer *vnx;
    581 		int flags;
    582 		struct buf *obp;
    583 		struct buf *bp;
    584 
    585 		obp = bufq_get(vnd->sc_tab);
    586 		if (obp == NULL) {
    587 			tsleep(&vnd->sc_tab, PRIBIO, "vndbp", 0);
    588 			continue;
    589 		};
    590 		splx(s);
    591 		flags = obp->b_flags;
    592 #ifdef DEBUG
    593 		if (vnddebug & VDB_FOLLOW)
    594 			printf("vndthread(%p)\n", obp);
    595 #endif
    596 
    597 		if (vnd->sc_vp->v_mount == NULL) {
    598 			obp->b_error = ENXIO;
    599 			goto done;
    600 		}
    601 #ifdef VND_COMPRESSION
    602 		/* handle a compressed read */
    603 		if ((flags & B_READ) != 0 && (vnd->sc_flags & VNF_COMP)) {
    604 			off_t bn;
    605 
    606 			/* Convert to a byte offset within the file. */
    607 			bn = obp->b_rawblkno *
    608 			    vnd->sc_dkdev.dk_label->d_secsize;
    609 
    610 			compstrategy(obp, bn);
    611 			goto done;
    612 		}
    613 #endif /* VND_COMPRESSION */
    614 
    615 		/*
    616 		 * Allocate a header for this transfer and link it to the
    617 		 * buffer
    618 		 */
    619 		s = splbio();
    620 		vnx = VND_GETXFER(vnd);
    621 		splx(s);
    622 		vnx->vx_vnd = vnd;
    623 
    624 		s = splbio();
    625 		while (vnd->sc_active >= vnd->sc_maxactive) {
    626 			tsleep(&vnd->sc_tab, PRIBIO, "vndac", 0);
    627 		}
    628 		vnd->sc_active++;
    629 		splx(s);
    630 
    631 		/* Instrumentation. */
    632 		disk_busy(&vnd->sc_dkdev);
    633 
    634 		bp = &vnx->vx_buf;
    635 		buf_init(bp);
    636 		bp->b_flags = (obp->b_flags & B_READ);
    637 		bp->b_oflags = obp->b_oflags;
    638 		bp->b_cflags = obp->b_cflags;
    639 		bp->b_iodone = vndiodone;
    640 		bp->b_private = obp;
    641 		bp->b_vp = vnd->sc_vp;
    642 		bp->b_objlock = bp->b_vp->v_interlock;
    643 		bp->b_data = obp->b_data;
    644 		bp->b_bcount = obp->b_bcount;
    645 		BIO_COPYPRIO(bp, obp);
    646 
    647 		/* Handle the request using the appropriate operations. */
    648 		if ((vnd->sc_flags & VNF_USE_VN_RDWR) == 0)
    649 			handle_with_strategy(vnd, obp, bp);
    650 		else
    651 			handle_with_rdwr(vnd, obp, bp);
    652 
    653 		s = splbio();
    654 		continue;
    655 
    656 done:
    657 		biodone(obp);
    658 		s = splbio();
    659 	}
    660 
    661 	vnd->sc_flags &= (~VNF_KTHREAD | VNF_VUNCONF);
    662 	wakeup(&vnd->sc_kthread);
    663 	splx(s);
    664 	kthread_exit(0);
    665 }
    666 
    667 /*
    668  * Checks if the given vnode supports the requested operation.
    669  * The operation is specified the offset returned by VOFFSET.
    670  *
    671  * XXX The test below used to determine this is quite fragile
    672  * because it relies on the file system to use genfs to specify
    673  * unimplemented operations.  There might be another way to do
    674  * it more cleanly.
    675  */
    676 static bool
    677 vnode_has_op(const struct vnode *vp, int opoffset)
    678 {
    679 	int (*defaultp)(void *);
    680 	int (*opp)(void *);
    681 
    682 	defaultp = vp->v_op[VOFFSET(vop_default)];
    683 	opp = vp->v_op[opoffset];
    684 
    685 	return opp != defaultp && opp != genfs_eopnotsupp &&
    686 	    opp != genfs_badop && opp != genfs_nullop;
    687 }
    688 
    689 /*
    690  * Handes the read/write request given in 'bp' using the vnode's VOP_READ
    691  * and VOP_WRITE operations.
    692  *
    693  * 'obp' is a pointer to the original request fed to the vnd device.
    694  */
    695 static void
    696 handle_with_rdwr(struct vnd_softc *vnd, const struct buf *obp, struct buf *bp)
    697 {
    698 	bool doread;
    699 	off_t offset;
    700 	size_t len, resid;
    701 	struct vnode *vp;
    702 
    703 	doread = bp->b_flags & B_READ;
    704 	offset = obp->b_rawblkno * vnd->sc_dkdev.dk_label->d_secsize;
    705 	len = bp->b_bcount;
    706 	vp = vnd->sc_vp;
    707 
    708 #if defined(DEBUG)
    709 	if (vnddebug & VDB_IO)
    710 		printf("vnd (rdwr): vp %p, %s, rawblkno 0x%" PRIx64
    711 		    ", secsize %d, offset %" PRIu64
    712 		    ", bcount %d\n",
    713 		    vp, doread ? "read" : "write", obp->b_rawblkno,
    714 		    vnd->sc_dkdev.dk_label->d_secsize, offset,
    715 		    bp->b_bcount);
    716 #endif
    717 
    718 	/* Issue the read or write operation. */
    719 	bp->b_error =
    720 	    vn_rdwr(doread ? UIO_READ : UIO_WRITE,
    721 	    vp, bp->b_data, len, offset, UIO_SYSSPACE,
    722 	    IO_ADV_ENCODE(POSIX_FADV_NOREUSE), vnd->sc_cred, &resid, NULL);
    723 	bp->b_resid = resid;
    724 
    725 	/* Keep mapped pages below threshold. */
    726 	mutex_enter(vp->v_interlock);
    727 	if (vp->v_uobj.uo_npages > 1024*1024 / PAGE_SIZE)
    728 		(void) VOP_PUTPAGES(vp, 0, 0,
    729 		    PGO_ALLPAGES | PGO_CLEANIT | PGO_FREE | PGO_SYNCIO);
    730 	else
    731 		mutex_exit(vp->v_interlock);
    732 
    733 	/* We need to increase the number of outputs on the vnode if
    734 	 * there was any write to it. */
    735 	if (!doread) {
    736 		mutex_enter(vp->v_interlock);
    737 		vp->v_numoutput++;
    738 		mutex_exit(vp->v_interlock);
    739 	}
    740 
    741 	biodone(bp);
    742 }
    743 
    744 /*
    745  * Handes the read/write request given in 'bp' using the vnode's VOP_BMAP
    746  * and VOP_STRATEGY operations.
    747  *
    748  * 'obp' is a pointer to the original request fed to the vnd device.
    749  */
    750 static void
    751 handle_with_strategy(struct vnd_softc *vnd, const struct buf *obp,
    752     struct buf *bp)
    753 {
    754 	int bsize, error, flags, skipped;
    755 	size_t resid, sz;
    756 	off_t bn, offset;
    757 	struct vnode *vp;
    758 
    759 	flags = obp->b_flags;
    760 
    761 	if (!(flags & B_READ)) {
    762 		vp = bp->b_vp;
    763 		mutex_enter(vp->v_interlock);
    764 		vp->v_numoutput++;
    765 		mutex_exit(vp->v_interlock);
    766 	}
    767 
    768 	/* convert to a byte offset within the file. */
    769 	bn = obp->b_rawblkno * vnd->sc_dkdev.dk_label->d_secsize;
    770 
    771 	bsize = vnd->sc_vp->v_mount->mnt_stat.f_iosize;
    772 	skipped = 0;
    773 
    774 	/*
    775 	 * Break the request into bsize pieces and feed them
    776 	 * sequentially using VOP_BMAP/VOP_STRATEGY.
    777 	 * We do it this way to keep from flooding NFS servers if we
    778 	 * are connected to an NFS file.  This places the burden on
    779 	 * the client rather than the server.
    780 	 */
    781 	error = 0;
    782 	bp->b_resid = bp->b_bcount;
    783 	for (offset = 0, resid = bp->b_resid; resid;
    784 	    resid -= sz, offset += sz) {
    785 		struct buf *nbp;
    786 		daddr_t nbn;
    787 		int off, nra;
    788 
    789 		nra = 0;
    790 		vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
    791 		error = VOP_BMAP(vnd->sc_vp, bn / bsize, &vp, &nbn, &nra);
    792 		VOP_UNLOCK(vnd->sc_vp);
    793 
    794 		if (error == 0 && (long)nbn == -1)
    795 			error = EIO;
    796 
    797 		/*
    798 		 * If there was an error or a hole in the file...punt.
    799 		 * Note that we may have to wait for any operations
    800 		 * that we have already fired off before releasing
    801 		 * the buffer.
    802 		 *
    803 		 * XXX we could deal with holes here but it would be
    804 		 * a hassle (in the write case).
    805 		 */
    806 		if (error) {
    807 			skipped += resid;
    808 			break;
    809 		}
    810 
    811 #ifdef DEBUG
    812 		if (!dovndcluster)
    813 			nra = 0;
    814 #endif
    815 
    816 		off = bn % bsize;
    817 		sz = MIN(((off_t)1 + nra) * bsize - off, resid);
    818 #ifdef	DEBUG
    819 		if (vnddebug & VDB_IO)
    820 			printf("vndstrategy: vp %p/%p bn 0x%qx/0x%" PRIx64
    821 			    " sz 0x%zx\n", vnd->sc_vp, vp, (long long)bn,
    822 			    nbn, sz);
    823 #endif
    824 
    825 		nbp = getiobuf(vp, true);
    826 		nestiobuf_setup(bp, nbp, offset, sz);
    827 		nbp->b_blkno = nbn + btodb(off);
    828 
    829 #if 0 /* XXX #ifdef DEBUG */
    830 		if (vnddebug & VDB_IO)
    831 			printf("vndstart(%ld): bp %p vp %p blkno "
    832 			    "0x%" PRIx64 " flags %x addr %p cnt 0x%x\n",
    833 			    (long) (vnd-vnd_softc), &nbp->vb_buf,
    834 			    nbp->vb_buf.b_vp, nbp->vb_buf.b_blkno,
    835 			    nbp->vb_buf.b_flags, nbp->vb_buf.b_data,
    836 			    nbp->vb_buf.b_bcount);
    837 #endif
    838 		VOP_STRATEGY(vp, nbp);
    839 		bn += sz;
    840 	}
    841 	nestiobuf_done(bp, skipped, error);
    842 }
    843 
    844 static void
    845 vndiodone(struct buf *bp)
    846 {
    847 	struct vndxfer *vnx = VND_BUFTOXFER(bp);
    848 	struct vnd_softc *vnd = vnx->vx_vnd;
    849 	struct buf *obp = bp->b_private;
    850 	int s = splbio();
    851 
    852 	KASSERT(&vnx->vx_buf == bp);
    853 	KASSERT(vnd->sc_active > 0);
    854 #ifdef DEBUG
    855 	if (vnddebug & VDB_IO) {
    856 		printf("vndiodone1: bp %p iodone: error %d\n",
    857 		    bp, bp->b_error);
    858 	}
    859 #endif
    860 	disk_unbusy(&vnd->sc_dkdev, bp->b_bcount - bp->b_resid,
    861 	    (bp->b_flags & B_READ));
    862 	vnd->sc_active--;
    863 	if (vnd->sc_active == 0) {
    864 		wakeup(&vnd->sc_tab);
    865 	}
    866 	splx(s);
    867 	obp->b_error = bp->b_error;
    868 	obp->b_resid = bp->b_resid;
    869 	buf_destroy(bp);
    870 	VND_PUTXFER(vnd, vnx);
    871 	biodone(obp);
    872 }
    873 
    874 /* ARGSUSED */
    875 static int
    876 vndread(dev_t dev, struct uio *uio, int flags)
    877 {
    878 	int unit = vndunit(dev);
    879 	struct vnd_softc *sc;
    880 
    881 #ifdef DEBUG
    882 	if (vnddebug & VDB_FOLLOW)
    883 		printf("vndread(0x%"PRIx64", %p)\n", dev, uio);
    884 #endif
    885 
    886 	sc = device_lookup_private(&vnd_cd, unit);
    887 	if (sc == NULL)
    888 		return ENXIO;
    889 
    890 	if ((sc->sc_flags & VNF_INITED) == 0)
    891 		return ENXIO;
    892 
    893 	return physio(vndstrategy, NULL, dev, B_READ, minphys, uio);
    894 }
    895 
    896 /* ARGSUSED */
    897 static int
    898 vndwrite(dev_t dev, struct uio *uio, int flags)
    899 {
    900 	int unit = vndunit(dev);
    901 	struct vnd_softc *sc;
    902 
    903 #ifdef DEBUG
    904 	if (vnddebug & VDB_FOLLOW)
    905 		printf("vndwrite(0x%"PRIx64", %p)\n", dev, uio);
    906 #endif
    907 
    908 	sc = device_lookup_private(&vnd_cd, unit);
    909 	if (sc == NULL)
    910 		return ENXIO;
    911 
    912 	if ((sc->sc_flags & VNF_INITED) == 0)
    913 		return ENXIO;
    914 
    915 	return physio(vndstrategy, NULL, dev, B_WRITE, minphys, uio);
    916 }
    917 
    918 static int
    919 vnd_cget(struct lwp *l, int unit, int *un, struct vattr *va)
    920 {
    921 	struct vnd_softc *vnd;
    922 
    923 	if (*un == -1)
    924 		*un = unit;
    925 	if (*un < 0)
    926 		return EINVAL;
    927 
    928 	vnd = device_lookup_private(&vnd_cd, *un);
    929 	if (vnd == NULL)
    930 		return (*un >= vnd_cd.cd_ndevs) ? ENXIO : -1;
    931 
    932 	if ((vnd->sc_flags & VNF_INITED) == 0)
    933 		return -1;
    934 
    935 	return VOP_GETATTR(vnd->sc_vp, va, l->l_cred);
    936 }
    937 
    938 static int
    939 vnddoclear(struct vnd_softc *vnd, int pmask, int minor, bool force)
    940 {
    941 	int error;
    942 
    943 	if ((error = vndlock(vnd)) != 0)
    944 		return error;
    945 
    946 	/*
    947 	 * Don't unconfigure if any other partitions are open
    948 	 * or if both the character and block flavors of this
    949 	 * partition are open.
    950 	 */
    951 	if (DK_BUSY(vnd, pmask) && !force) {
    952 		vndunlock(vnd);
    953 		return EBUSY;
    954 	}
    955 
    956 	/*
    957 	 * XXX vndclear() might call vndclose() implicitly;
    958 	 * release lock to avoid recursion
    959 	 *
    960 	 * Set VNF_CLEARING to prevent vndopen() from
    961 	 * sneaking in after we vndunlock().
    962 	 */
    963 	vnd->sc_flags |= VNF_CLEARING;
    964 	vndunlock(vnd);
    965 	vndclear(vnd, minor);
    966 #ifdef DEBUG
    967 	if (vnddebug & VDB_INIT)
    968 		printf("vndioctl: CLRed\n");
    969 #endif
    970 
    971 	/* Destroy the xfer and buffer pools. */
    972 	pool_destroy(&vnd->sc_vxpool);
    973 
    974 	/* Detach the disk. */
    975 	disk_detach(&vnd->sc_dkdev);
    976 
    977 	return 0;
    978 }
    979 
    980 /* ARGSUSED */
    981 static int
    982 vndioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l)
    983 {
    984 	bool force;
    985 	int unit = vndunit(dev);
    986 	struct vnd_softc *vnd;
    987 	struct vnd_ioctl *vio;
    988 	struct vattr vattr;
    989 	struct pathbuf *pb;
    990 	struct nameidata nd;
    991 	int error, part, pmask;
    992 	size_t geomsize;
    993 	int fflags;
    994 #ifdef __HAVE_OLD_DISKLABEL
    995 	struct disklabel newlabel;
    996 #endif
    997 	struct dkwedge_info *dkw;
    998 	struct dkwedge_list *dkwl;
    999 
   1000 #ifdef DEBUG
   1001 	if (vnddebug & VDB_FOLLOW)
   1002 		printf("vndioctl(0x%"PRIx64", 0x%lx, %p, 0x%x, %p): unit %d\n",
   1003 		    dev, cmd, data, flag, l->l_proc, unit);
   1004 #endif
   1005 	vnd = device_lookup_private(&vnd_cd, unit);
   1006 	if (vnd == NULL &&
   1007 #ifdef COMPAT_30
   1008 	    cmd != VNDIOCGET30 &&
   1009 #endif
   1010 #ifdef COMPAT_50
   1011 	    cmd != VNDIOCGET50 &&
   1012 #endif
   1013 	    cmd != VNDIOCGET)
   1014 		return ENXIO;
   1015 	vio = (struct vnd_ioctl *)data;
   1016 
   1017 	/* Must be open for writes for these commands... */
   1018 	switch (cmd) {
   1019 	case VNDIOCSET:
   1020 	case VNDIOCCLR:
   1021 #ifdef COMPAT_50
   1022 	case VNDIOCSET50:
   1023 	case VNDIOCCLR50:
   1024 #endif
   1025 	case DIOCSDINFO:
   1026 	case DIOCWDINFO:
   1027 #ifdef __HAVE_OLD_DISKLABEL
   1028 	case ODIOCSDINFO:
   1029 	case ODIOCWDINFO:
   1030 #endif
   1031 	case DIOCKLABEL:
   1032 	case DIOCWLABEL:
   1033 		if ((flag & FWRITE) == 0)
   1034 			return EBADF;
   1035 	}
   1036 
   1037 	/* Must be initialized for these... */
   1038 	switch (cmd) {
   1039 	case VNDIOCCLR:
   1040 #ifdef VNDIOCCLR50
   1041 	case VNDIOCCLR50:
   1042 #endif
   1043 	case DIOCGDINFO:
   1044 	case DIOCSDINFO:
   1045 	case DIOCWDINFO:
   1046 	case DIOCGPART:
   1047 	case DIOCKLABEL:
   1048 	case DIOCWLABEL:
   1049 	case DIOCGDEFLABEL:
   1050 	case DIOCCACHESYNC:
   1051 #ifdef __HAVE_OLD_DISKLABEL
   1052 	case ODIOCGDINFO:
   1053 	case ODIOCSDINFO:
   1054 	case ODIOCWDINFO:
   1055 	case ODIOCGDEFLABEL:
   1056 #endif
   1057 		if ((vnd->sc_flags & VNF_INITED) == 0)
   1058 			return ENXIO;
   1059 	}
   1060 
   1061 	switch (cmd) {
   1062 #ifdef VNDIOCSET50
   1063 	case VNDIOCSET50:
   1064 #endif
   1065 	case VNDIOCSET:
   1066 		if (vnd->sc_flags & VNF_INITED)
   1067 			return EBUSY;
   1068 
   1069 		if ((error = vndlock(vnd)) != 0)
   1070 			return error;
   1071 
   1072 		fflags = FREAD;
   1073 		if ((vio->vnd_flags & VNDIOF_READONLY) == 0)
   1074 			fflags |= FWRITE;
   1075 		error = pathbuf_copyin(vio->vnd_file, &pb);
   1076 		if (error) {
   1077 			goto unlock_and_exit;
   1078 		}
   1079 		NDINIT(&nd, LOOKUP, FOLLOW, pb);
   1080 		if ((error = vn_open(&nd, fflags, 0)) != 0) {
   1081 			pathbuf_destroy(pb);
   1082 			goto unlock_and_exit;
   1083 		}
   1084 		KASSERT(l);
   1085 		error = VOP_GETATTR(nd.ni_vp, &vattr, l->l_cred);
   1086 		if (!error && nd.ni_vp->v_type != VREG)
   1087 			error = EOPNOTSUPP;
   1088 		if (!error && vattr.va_bytes < vattr.va_size)
   1089 			/* File is definitely sparse, use vn_rdwr() */
   1090 			vnd->sc_flags |= VNF_USE_VN_RDWR;
   1091 		if (error) {
   1092 			VOP_UNLOCK(nd.ni_vp);
   1093 			goto close_and_exit;
   1094 		}
   1095 
   1096 		/* If using a compressed file, initialize its info */
   1097 		/* (or abort with an error if kernel has no compression) */
   1098 		if (vio->vnd_flags & VNF_COMP) {
   1099 #ifdef VND_COMPRESSION
   1100 			struct vnd_comp_header *ch;
   1101 			int i;
   1102 			u_int32_t comp_size;
   1103 			u_int32_t comp_maxsize;
   1104 
   1105 			/* allocate space for compresed file header */
   1106 			ch = malloc(sizeof(struct vnd_comp_header),
   1107 			M_TEMP, M_WAITOK);
   1108 
   1109 			/* read compressed file header */
   1110 			error = vn_rdwr(UIO_READ, nd.ni_vp, (void *)ch,
   1111 			  sizeof(struct vnd_comp_header), 0, UIO_SYSSPACE,
   1112 			  IO_UNIT|IO_NODELOCKED, l->l_cred, NULL, NULL);
   1113 			if (error) {
   1114 				free(ch, M_TEMP);
   1115 				VOP_UNLOCK(nd.ni_vp);
   1116 				goto close_and_exit;
   1117 			}
   1118 
   1119 			/* save some header info */
   1120 			vnd->sc_comp_blksz = ntohl(ch->block_size);
   1121 			/* note last offset is the file byte size */
   1122 			vnd->sc_comp_numoffs = ntohl(ch->num_blocks)+1;
   1123 			free(ch, M_TEMP);
   1124 			if (vnd->sc_comp_blksz == 0 ||
   1125 			    vnd->sc_comp_blksz % DEV_BSIZE !=0) {
   1126 				VOP_UNLOCK(nd.ni_vp);
   1127 				error = EINVAL;
   1128 				goto close_and_exit;
   1129 			}
   1130 			if (sizeof(struct vnd_comp_header) +
   1131 			  sizeof(u_int64_t) * vnd->sc_comp_numoffs >
   1132 			  vattr.va_size) {
   1133 				VOP_UNLOCK(nd.ni_vp);
   1134 				error = EINVAL;
   1135 				goto close_and_exit;
   1136 			}
   1137 
   1138 			/* set decompressed file size */
   1139 			vattr.va_size =
   1140 			    ((u_quad_t)vnd->sc_comp_numoffs - 1) *
   1141 			     (u_quad_t)vnd->sc_comp_blksz;
   1142 
   1143 			/* allocate space for all the compressed offsets */
   1144 			vnd->sc_comp_offsets =
   1145 			malloc(sizeof(u_int64_t) * vnd->sc_comp_numoffs,
   1146 			M_DEVBUF, M_WAITOK);
   1147 
   1148 			/* read in the offsets */
   1149 			error = vn_rdwr(UIO_READ, nd.ni_vp,
   1150 			  (void *)vnd->sc_comp_offsets,
   1151 			  sizeof(u_int64_t) * vnd->sc_comp_numoffs,
   1152 			  sizeof(struct vnd_comp_header), UIO_SYSSPACE,
   1153 			  IO_UNIT|IO_NODELOCKED, l->l_cred, NULL, NULL);
   1154 			if (error) {
   1155 				VOP_UNLOCK(nd.ni_vp);
   1156 				goto close_and_exit;
   1157 			}
   1158 			/*
   1159 			 * find largest block size (used for allocation limit).
   1160 			 * Also convert offset to native byte order.
   1161 			 */
   1162 			comp_maxsize = 0;
   1163 			for (i = 0; i < vnd->sc_comp_numoffs - 1; i++) {
   1164 				vnd->sc_comp_offsets[i] =
   1165 				  be64toh(vnd->sc_comp_offsets[i]);
   1166 				comp_size = be64toh(vnd->sc_comp_offsets[i + 1])
   1167 				  - vnd->sc_comp_offsets[i];
   1168 				if (comp_size > comp_maxsize)
   1169 					comp_maxsize = comp_size;
   1170 			}
   1171 			vnd->sc_comp_offsets[vnd->sc_comp_numoffs - 1] =
   1172 			  be64toh(vnd->sc_comp_offsets[vnd->sc_comp_numoffs - 1]);
   1173 
   1174 			/* create compressed data buffer */
   1175 			vnd->sc_comp_buff = malloc(comp_maxsize,
   1176 			  M_DEVBUF, M_WAITOK);
   1177 
   1178 			/* create decompressed buffer */
   1179 			vnd->sc_comp_decombuf = malloc(vnd->sc_comp_blksz,
   1180 			  M_DEVBUF, M_WAITOK);
   1181 			vnd->sc_comp_buffblk = -1;
   1182 
   1183 			/* Initialize decompress stream */
   1184 			memset(&vnd->sc_comp_stream, 0, sizeof(z_stream));
   1185 			vnd->sc_comp_stream.zalloc = vnd_alloc;
   1186 			vnd->sc_comp_stream.zfree = vnd_free;
   1187 			error = inflateInit2(&vnd->sc_comp_stream, MAX_WBITS);
   1188 			if (error) {
   1189 				if (vnd->sc_comp_stream.msg)
   1190 					printf("vnd%d: compressed file, %s\n",
   1191 					  unit, vnd->sc_comp_stream.msg);
   1192 				VOP_UNLOCK(nd.ni_vp);
   1193 				error = EINVAL;
   1194 				goto close_and_exit;
   1195 			}
   1196 
   1197 			vnd->sc_flags |= VNF_COMP | VNF_READONLY;
   1198 #else /* !VND_COMPRESSION */
   1199 			VOP_UNLOCK(nd.ni_vp);
   1200 			error = EOPNOTSUPP;
   1201 			goto close_and_exit;
   1202 #endif /* VND_COMPRESSION */
   1203 		}
   1204 
   1205 		VOP_UNLOCK(nd.ni_vp);
   1206 		vnd->sc_vp = nd.ni_vp;
   1207 		vnd->sc_size = btodb(vattr.va_size);	/* note truncation */
   1208 
   1209 		/*
   1210 		 * Use pseudo-geometry specified.  If none was provided,
   1211 		 * use "standard" Adaptec fictitious geometry.
   1212 		 */
   1213 		if (vio->vnd_flags & VNDIOF_HASGEOM) {
   1214 
   1215 			memcpy(&vnd->sc_geom, &vio->vnd_geom,
   1216 			    sizeof(vio->vnd_geom));
   1217 
   1218 			/*
   1219 			 * Sanity-check the sector size.
   1220 			 * XXX Don't allow secsize < DEV_BSIZE.	 Should
   1221 			 * XXX we?
   1222 			 */
   1223 			if (vnd->sc_geom.vng_secsize < DEV_BSIZE ||
   1224 			    (vnd->sc_geom.vng_secsize % DEV_BSIZE) != 0 ||
   1225 			    vnd->sc_geom.vng_ncylinders == 0 ||
   1226 			    (vnd->sc_geom.vng_ntracks *
   1227 			     vnd->sc_geom.vng_nsectors) == 0) {
   1228 				error = EINVAL;
   1229 				goto close_and_exit;
   1230 			}
   1231 
   1232 			/*
   1233 			 * Compute the size (in DEV_BSIZE blocks) specified
   1234 			 * by the geometry.
   1235 			 */
   1236 			geomsize = (vnd->sc_geom.vng_nsectors *
   1237 			    vnd->sc_geom.vng_ntracks *
   1238 			    vnd->sc_geom.vng_ncylinders) *
   1239 			    (vnd->sc_geom.vng_secsize / DEV_BSIZE);
   1240 
   1241 			/*
   1242 			 * Sanity-check the size against the specified
   1243 			 * geometry.
   1244 			 */
   1245 			if (vnd->sc_size < geomsize) {
   1246 				error = EINVAL;
   1247 				goto close_and_exit;
   1248 			}
   1249 		} else if (vnd->sc_size >= (32 * 64)) {
   1250 			/*
   1251 			 * Size must be at least 2048 DEV_BSIZE blocks
   1252 			 * (1M) in order to use this geometry.
   1253 			 */
   1254 			vnd->sc_geom.vng_secsize = DEV_BSIZE;
   1255 			vnd->sc_geom.vng_nsectors = 32;
   1256 			vnd->sc_geom.vng_ntracks = 64;
   1257 			vnd->sc_geom.vng_ncylinders = vnd->sc_size / (64 * 32);
   1258 		} else {
   1259 			vnd->sc_geom.vng_secsize = DEV_BSIZE;
   1260 			vnd->sc_geom.vng_nsectors = 1;
   1261 			vnd->sc_geom.vng_ntracks = 1;
   1262 			vnd->sc_geom.vng_ncylinders = vnd->sc_size;
   1263 		}
   1264 
   1265 		vnd_set_properties(vnd);
   1266 
   1267 		if (vio->vnd_flags & VNDIOF_READONLY) {
   1268 			vnd->sc_flags |= VNF_READONLY;
   1269 		}
   1270 
   1271 		if ((error = vndsetcred(vnd, l->l_cred)) != 0)
   1272 			goto close_and_exit;
   1273 
   1274 		vndthrottle(vnd, vnd->sc_vp);
   1275 		vio->vnd_osize = dbtob(vnd->sc_size);
   1276 #ifdef VNDIOCSET50
   1277 		if (cmd != VNDIOCSET50)
   1278 #endif
   1279 			vio->vnd_size = dbtob(vnd->sc_size);
   1280 		vnd->sc_flags |= VNF_INITED;
   1281 
   1282 		/* create the kernel thread, wait for it to be up */
   1283 		error = kthread_create(PRI_NONE, 0, NULL, vndthread, vnd,
   1284 		    &vnd->sc_kthread, "%s", device_xname(vnd->sc_dev));
   1285 		if (error)
   1286 			goto close_and_exit;
   1287 		while ((vnd->sc_flags & VNF_KTHREAD) == 0) {
   1288 			tsleep(&vnd->sc_kthread, PRIBIO, "vndthr", 0);
   1289 		}
   1290 #ifdef DEBUG
   1291 		if (vnddebug & VDB_INIT)
   1292 			printf("vndioctl: SET vp %p size 0x%lx %d/%d/%d/%d\n",
   1293 			    vnd->sc_vp, (unsigned long) vnd->sc_size,
   1294 			    vnd->sc_geom.vng_secsize,
   1295 			    vnd->sc_geom.vng_nsectors,
   1296 			    vnd->sc_geom.vng_ntracks,
   1297 			    vnd->sc_geom.vng_ncylinders);
   1298 #endif
   1299 
   1300 		/* Attach the disk. */
   1301 		disk_attach(&vnd->sc_dkdev);
   1302 		disk_blocksize(&vnd->sc_dkdev, vnd->sc_geom.vng_secsize);
   1303 
   1304 		/* Initialize the xfer and buffer pools. */
   1305 		pool_init(&vnd->sc_vxpool, sizeof(struct vndxfer), 0,
   1306 		    0, 0, "vndxpl", NULL, IPL_BIO);
   1307 
   1308 		/* Try and read the disklabel. */
   1309 		vndgetdisklabel(dev, vnd);
   1310 
   1311 		vndunlock(vnd);
   1312 
   1313 		pathbuf_destroy(pb);
   1314 		break;
   1315 
   1316 close_and_exit:
   1317 		(void) vn_close(nd.ni_vp, fflags, l->l_cred);
   1318 		pathbuf_destroy(pb);
   1319 unlock_and_exit:
   1320 #ifdef VND_COMPRESSION
   1321 		/* free any allocated memory (for compressed file) */
   1322 		if (vnd->sc_comp_offsets) {
   1323 			free(vnd->sc_comp_offsets, M_DEVBUF);
   1324 			vnd->sc_comp_offsets = NULL;
   1325 		}
   1326 		if (vnd->sc_comp_buff) {
   1327 			free(vnd->sc_comp_buff, M_DEVBUF);
   1328 			vnd->sc_comp_buff = NULL;
   1329 		}
   1330 		if (vnd->sc_comp_decombuf) {
   1331 			free(vnd->sc_comp_decombuf, M_DEVBUF);
   1332 			vnd->sc_comp_decombuf = NULL;
   1333 		}
   1334 #endif /* VND_COMPRESSION */
   1335 		vndunlock(vnd);
   1336 		return error;
   1337 
   1338 #ifdef VNDIOCCLR50
   1339 	case VNDIOCCLR50:
   1340 #endif
   1341 	case VNDIOCCLR:
   1342 		part = DISKPART(dev);
   1343 		pmask = (1 << part);
   1344 		force = (vio->vnd_flags & VNDIOF_FORCE) != 0;
   1345 
   1346 		if ((error = vnddoclear(vnd, pmask, minor(dev), force)) != 0)
   1347 			return error;
   1348 
   1349 		break;
   1350 
   1351 #ifdef COMPAT_30
   1352 	case VNDIOCGET30: {
   1353 		struct vnd_user30 *vnu;
   1354 		struct vattr va;
   1355 		vnu = (struct vnd_user30 *)data;
   1356 		KASSERT(l);
   1357 		switch (error = vnd_cget(l, unit, &vnu->vnu_unit, &va)) {
   1358 		case 0:
   1359 			vnu->vnu_dev = va.va_fsid;
   1360 			vnu->vnu_ino = va.va_fileid;
   1361 			break;
   1362 		case -1:
   1363 			/* unused is not an error */
   1364 			vnu->vnu_dev = 0;
   1365 			vnu->vnu_ino = 0;
   1366 			break;
   1367 		default:
   1368 			return error;
   1369 		}
   1370 		break;
   1371 	}
   1372 #endif
   1373 
   1374 #ifdef COMPAT_50
   1375 	case VNDIOCGET50: {
   1376 		struct vnd_user50 *vnu;
   1377 		struct vattr va;
   1378 		vnu = (struct vnd_user50 *)data;
   1379 		KASSERT(l);
   1380 		switch (error = vnd_cget(l, unit, &vnu->vnu_unit, &va)) {
   1381 		case 0:
   1382 			vnu->vnu_dev = va.va_fsid;
   1383 			vnu->vnu_ino = va.va_fileid;
   1384 			break;
   1385 		case -1:
   1386 			/* unused is not an error */
   1387 			vnu->vnu_dev = 0;
   1388 			vnu->vnu_ino = 0;
   1389 			break;
   1390 		default:
   1391 			return error;
   1392 		}
   1393 		break;
   1394 	}
   1395 #endif
   1396 
   1397 	case VNDIOCGET: {
   1398 		struct vnd_user *vnu;
   1399 		struct vattr va;
   1400 		vnu = (struct vnd_user *)data;
   1401 		KASSERT(l);
   1402 		switch (error = vnd_cget(l, unit, &vnu->vnu_unit, &va)) {
   1403 		case 0:
   1404 			vnu->vnu_dev = va.va_fsid;
   1405 			vnu->vnu_ino = va.va_fileid;
   1406 			break;
   1407 		case -1:
   1408 			/* unused is not an error */
   1409 			vnu->vnu_dev = 0;
   1410 			vnu->vnu_ino = 0;
   1411 			break;
   1412 		default:
   1413 			return error;
   1414 		}
   1415 		break;
   1416 	}
   1417 
   1418 	case DIOCGDINFO:
   1419 		*(struct disklabel *)data = *(vnd->sc_dkdev.dk_label);
   1420 		break;
   1421 
   1422 #ifdef __HAVE_OLD_DISKLABEL
   1423 	case ODIOCGDINFO:
   1424 		newlabel = *(vnd->sc_dkdev.dk_label);
   1425 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
   1426 			return ENOTTY;
   1427 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
   1428 		break;
   1429 #endif
   1430 
   1431 	case DIOCGPART:
   1432 		((struct partinfo *)data)->disklab = vnd->sc_dkdev.dk_label;
   1433 		((struct partinfo *)data)->part =
   1434 		    &vnd->sc_dkdev.dk_label->d_partitions[DISKPART(dev)];
   1435 		break;
   1436 
   1437 	case DIOCWDINFO:
   1438 	case DIOCSDINFO:
   1439 #ifdef __HAVE_OLD_DISKLABEL
   1440 	case ODIOCWDINFO:
   1441 	case ODIOCSDINFO:
   1442 #endif
   1443 	{
   1444 		struct disklabel *lp;
   1445 
   1446 		if ((error = vndlock(vnd)) != 0)
   1447 			return error;
   1448 
   1449 		vnd->sc_flags |= VNF_LABELLING;
   1450 
   1451 #ifdef __HAVE_OLD_DISKLABEL
   1452 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
   1453 			memset(&newlabel, 0, sizeof newlabel);
   1454 			memcpy(&newlabel, data, sizeof (struct olddisklabel));
   1455 			lp = &newlabel;
   1456 		} else
   1457 #endif
   1458 		lp = (struct disklabel *)data;
   1459 
   1460 		error = setdisklabel(vnd->sc_dkdev.dk_label,
   1461 		    lp, 0, vnd->sc_dkdev.dk_cpulabel);
   1462 		if (error == 0) {
   1463 			if (cmd == DIOCWDINFO
   1464 #ifdef __HAVE_OLD_DISKLABEL
   1465 			    || cmd == ODIOCWDINFO
   1466 #endif
   1467 			   )
   1468 				error = writedisklabel(VNDLABELDEV(dev),
   1469 				    vndstrategy, vnd->sc_dkdev.dk_label,
   1470 				    vnd->sc_dkdev.dk_cpulabel);
   1471 		}
   1472 
   1473 		vnd->sc_flags &= ~VNF_LABELLING;
   1474 
   1475 		vndunlock(vnd);
   1476 
   1477 		if (error)
   1478 			return error;
   1479 		break;
   1480 	}
   1481 
   1482 	case DIOCKLABEL:
   1483 		if (*(int *)data != 0)
   1484 			vnd->sc_flags |= VNF_KLABEL;
   1485 		else
   1486 			vnd->sc_flags &= ~VNF_KLABEL;
   1487 		break;
   1488 
   1489 	case DIOCWLABEL:
   1490 		if (*(int *)data != 0)
   1491 			vnd->sc_flags |= VNF_WLABEL;
   1492 		else
   1493 			vnd->sc_flags &= ~VNF_WLABEL;
   1494 		break;
   1495 
   1496 	case DIOCGDEFLABEL:
   1497 		vndgetdefaultlabel(vnd, (struct disklabel *)data);
   1498 		break;
   1499 
   1500 #ifdef __HAVE_OLD_DISKLABEL
   1501 	case ODIOCGDEFLABEL:
   1502 		vndgetdefaultlabel(vnd, &newlabel);
   1503 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
   1504 			return ENOTTY;
   1505 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
   1506 		break;
   1507 #endif
   1508 
   1509 	case DIOCCACHESYNC:
   1510 		vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
   1511 		error = VOP_FSYNC(vnd->sc_vp, vnd->sc_cred,
   1512 		    FSYNC_WAIT | FSYNC_DATAONLY | FSYNC_CACHE, 0, 0);
   1513 		VOP_UNLOCK(vnd->sc_vp);
   1514 		return error;
   1515 
   1516 	case DIOCAWEDGE:
   1517 		dkw = (void *) data;
   1518 
   1519 		if ((flag & FWRITE) == 0)
   1520 			return EBADF;
   1521 
   1522 		/* If the ioctl happens here, the parent is us. */
   1523 		strlcpy(dkw->dkw_parent, device_xname(vnd->sc_dev),
   1524 		    sizeof(dkw->dkw_parent));
   1525 		return dkwedge_add(dkw);
   1526 
   1527 	case DIOCDWEDGE:
   1528 		dkw = (void *) data;
   1529 
   1530 		if ((flag & FWRITE) == 0)
   1531 			return EBADF;
   1532 
   1533 		/* If the ioctl happens here, the parent is us. */
   1534 		strlcpy(dkw->dkw_parent, device_xname(vnd->sc_dev),
   1535 		    sizeof(dkw->dkw_parent));
   1536 		return dkwedge_del(dkw);
   1537 
   1538 	case DIOCLWEDGES:
   1539 		dkwl = (void *) data;
   1540 
   1541 		return dkwedge_list(&vnd->sc_dkdev, dkwl, l);
   1542 
   1543 	default:
   1544 		return ENOTTY;
   1545 	}
   1546 
   1547 	return 0;
   1548 }
   1549 
   1550 /*
   1551  * Duplicate the current processes' credentials.  Since we are called only
   1552  * as the result of a SET ioctl and only root can do that, any future access
   1553  * to this "disk" is essentially as root.  Note that credentials may change
   1554  * if some other uid can write directly to the mapped file (NFS).
   1555  */
   1556 static int
   1557 vndsetcred(struct vnd_softc *vnd, kauth_cred_t cred)
   1558 {
   1559 	struct uio auio;
   1560 	struct iovec aiov;
   1561 	char *tmpbuf;
   1562 	int error;
   1563 
   1564 	vnd->sc_cred = kauth_cred_dup(cred);
   1565 	tmpbuf = malloc(DEV_BSIZE, M_TEMP, M_WAITOK);
   1566 
   1567 	/* XXX: Horrible kludge to establish credentials for NFS */
   1568 	aiov.iov_base = tmpbuf;
   1569 	aiov.iov_len = min(DEV_BSIZE, dbtob(vnd->sc_size));
   1570 	auio.uio_iov = &aiov;
   1571 	auio.uio_iovcnt = 1;
   1572 	auio.uio_offset = 0;
   1573 	auio.uio_rw = UIO_READ;
   1574 	auio.uio_resid = aiov.iov_len;
   1575 	UIO_SETUP_SYSSPACE(&auio);
   1576 	vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
   1577 	error = VOP_READ(vnd->sc_vp, &auio, 0, vnd->sc_cred);
   1578 	if (error == 0) {
   1579 		/*
   1580 		 * Because vnd does all IO directly through the vnode
   1581 		 * we need to flush (at least) the buffer from the above
   1582 		 * VOP_READ from the buffer cache to prevent cache
   1583 		 * incoherencies.  Also, be careful to write dirty
   1584 		 * buffers back to stable storage.
   1585 		 */
   1586 		error = vinvalbuf(vnd->sc_vp, V_SAVE, vnd->sc_cred,
   1587 			    curlwp, 0, 0);
   1588 	}
   1589 	VOP_UNLOCK(vnd->sc_vp);
   1590 
   1591 	free(tmpbuf, M_TEMP);
   1592 	return error;
   1593 }
   1594 
   1595 /*
   1596  * Set maxactive based on FS type
   1597  */
   1598 static void
   1599 vndthrottle(struct vnd_softc *vnd, struct vnode *vp)
   1600 {
   1601 
   1602 	if (vp->v_tag == VT_NFS)
   1603 		vnd->sc_maxactive = 2;
   1604 	else
   1605 		vnd->sc_maxactive = 8;
   1606 
   1607 	if (vnd->sc_maxactive < 1)
   1608 		vnd->sc_maxactive = 1;
   1609 }
   1610 
   1611 #if 0
   1612 static void
   1613 vndshutdown(void)
   1614 {
   1615 	struct vnd_softc *vnd;
   1616 
   1617 	for (vnd = &vnd_softc[0]; vnd < &vnd_softc[numvnd]; vnd++)
   1618 		if (vnd->sc_flags & VNF_INITED)
   1619 			vndclear(vnd);
   1620 }
   1621 #endif
   1622 
   1623 static void
   1624 vndclear(struct vnd_softc *vnd, int myminor)
   1625 {
   1626 	struct vnode *vp = vnd->sc_vp;
   1627 	int fflags = FREAD;
   1628 	int bmaj, cmaj, i, mn;
   1629 	int s;
   1630 
   1631 #ifdef DEBUG
   1632 	if (vnddebug & VDB_FOLLOW)
   1633 		printf("vndclear(%p): vp %p\n", vnd, vp);
   1634 #endif
   1635 	/* locate the major number */
   1636 	bmaj = bdevsw_lookup_major(&vnd_bdevsw);
   1637 	cmaj = cdevsw_lookup_major(&vnd_cdevsw);
   1638 
   1639 	/* Nuke the vnodes for any open instances */
   1640 	for (i = 0; i < MAXPARTITIONS; i++) {
   1641 		mn = DISKMINOR(device_unit(vnd->sc_dev), i);
   1642 		vdevgone(bmaj, mn, mn, VBLK);
   1643 		if (mn != myminor) /* XXX avoid to kill own vnode */
   1644 			vdevgone(cmaj, mn, mn, VCHR);
   1645 	}
   1646 
   1647 	if ((vnd->sc_flags & VNF_READONLY) == 0)
   1648 		fflags |= FWRITE;
   1649 
   1650 	s = splbio();
   1651 	bufq_drain(vnd->sc_tab);
   1652 	splx(s);
   1653 
   1654 	vnd->sc_flags |= VNF_VUNCONF;
   1655 	wakeup(&vnd->sc_tab);
   1656 	while (vnd->sc_flags & VNF_KTHREAD)
   1657 		tsleep(&vnd->sc_kthread, PRIBIO, "vnthr", 0);
   1658 
   1659 #ifdef VND_COMPRESSION
   1660 	/* free the compressed file buffers */
   1661 	if (vnd->sc_flags & VNF_COMP) {
   1662 		if (vnd->sc_comp_offsets) {
   1663 			free(vnd->sc_comp_offsets, M_DEVBUF);
   1664 			vnd->sc_comp_offsets = NULL;
   1665 		}
   1666 		if (vnd->sc_comp_buff) {
   1667 			free(vnd->sc_comp_buff, M_DEVBUF);
   1668 			vnd->sc_comp_buff = NULL;
   1669 		}
   1670 		if (vnd->sc_comp_decombuf) {
   1671 			free(vnd->sc_comp_decombuf, M_DEVBUF);
   1672 			vnd->sc_comp_decombuf = NULL;
   1673 		}
   1674 	}
   1675 #endif /* VND_COMPRESSION */
   1676 	vnd->sc_flags &=
   1677 	    ~(VNF_INITED | VNF_READONLY | VNF_VLABEL
   1678 	      | VNF_VUNCONF | VNF_COMP | VNF_CLEARING);
   1679 	if (vp == NULL)
   1680 		panic("vndclear: null vp");
   1681 	(void) vn_close(vp, fflags, vnd->sc_cred);
   1682 	kauth_cred_free(vnd->sc_cred);
   1683 	vnd->sc_vp = NULL;
   1684 	vnd->sc_cred = NULL;
   1685 	vnd->sc_size = 0;
   1686 }
   1687 
   1688 static int
   1689 vndsize(dev_t dev)
   1690 {
   1691 	struct vnd_softc *sc;
   1692 	struct disklabel *lp;
   1693 	int part, unit, omask;
   1694 	int size;
   1695 
   1696 	unit = vndunit(dev);
   1697 	sc = device_lookup_private(&vnd_cd, unit);
   1698 	if (sc == NULL)
   1699 		return -1;
   1700 
   1701 	if ((sc->sc_flags & VNF_INITED) == 0)
   1702 		return -1;
   1703 
   1704 	part = DISKPART(dev);
   1705 	omask = sc->sc_dkdev.dk_openmask & (1 << part);
   1706 	lp = sc->sc_dkdev.dk_label;
   1707 
   1708 	if (omask == 0 && vndopen(dev, 0, S_IFBLK, curlwp))	/* XXX */
   1709 		return -1;
   1710 
   1711 	if (lp->d_partitions[part].p_fstype != FS_SWAP)
   1712 		size = -1;
   1713 	else
   1714 		size = lp->d_partitions[part].p_size *
   1715 		    (lp->d_secsize / DEV_BSIZE);
   1716 
   1717 	if (omask == 0 && vndclose(dev, 0, S_IFBLK, curlwp))	/* XXX */
   1718 		return -1;
   1719 
   1720 	return size;
   1721 }
   1722 
   1723 static int
   1724 vnddump(dev_t dev, daddr_t blkno, void *va,
   1725     size_t size)
   1726 {
   1727 
   1728 	/* Not implemented. */
   1729 	return ENXIO;
   1730 }
   1731 
   1732 static void
   1733 vndgetdefaultlabel(struct vnd_softc *sc, struct disklabel *lp)
   1734 {
   1735 	struct vndgeom *vng = &sc->sc_geom;
   1736 	struct partition *pp;
   1737 
   1738 	memset(lp, 0, sizeof(*lp));
   1739 
   1740 	lp->d_secperunit = sc->sc_size / (vng->vng_secsize / DEV_BSIZE);
   1741 	lp->d_secsize = vng->vng_secsize;
   1742 	lp->d_nsectors = vng->vng_nsectors;
   1743 	lp->d_ntracks = vng->vng_ntracks;
   1744 	lp->d_ncylinders = vng->vng_ncylinders;
   1745 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
   1746 
   1747 	strncpy(lp->d_typename, "vnd", sizeof(lp->d_typename));
   1748 	lp->d_type = DTYPE_VND;
   1749 	strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
   1750 	lp->d_rpm = 3600;
   1751 	lp->d_interleave = 1;
   1752 	lp->d_flags = 0;
   1753 
   1754 	pp = &lp->d_partitions[RAW_PART];
   1755 	pp->p_offset = 0;
   1756 	pp->p_size = lp->d_secperunit;
   1757 	pp->p_fstype = FS_UNUSED;
   1758 	lp->d_npartitions = RAW_PART + 1;
   1759 
   1760 	lp->d_magic = DISKMAGIC;
   1761 	lp->d_magic2 = DISKMAGIC;
   1762 	lp->d_checksum = dkcksum(lp);
   1763 }
   1764 
   1765 /*
   1766  * Read the disklabel from a vnd.  If one is not present, create a fake one.
   1767  */
   1768 static void
   1769 vndgetdisklabel(dev_t dev, struct vnd_softc *sc)
   1770 {
   1771 	const char *errstring;
   1772 	struct disklabel *lp = sc->sc_dkdev.dk_label;
   1773 	struct cpu_disklabel *clp = sc->sc_dkdev.dk_cpulabel;
   1774 	int i;
   1775 
   1776 	memset(clp, 0, sizeof(*clp));
   1777 
   1778 	vndgetdefaultlabel(sc, lp);
   1779 
   1780 	/*
   1781 	 * Call the generic disklabel extraction routine.
   1782 	 */
   1783 	errstring = readdisklabel(VNDLABELDEV(dev), vndstrategy, lp, clp);
   1784 	if (errstring) {
   1785 		/*
   1786 		 * Lack of disklabel is common, but we print the warning
   1787 		 * anyway, since it might contain other useful information.
   1788 		 */
   1789 		aprint_normal_dev(sc->sc_dev, "%s\n", errstring);
   1790 
   1791 		/*
   1792 		 * For historical reasons, if there's no disklabel
   1793 		 * present, all partitions must be FS_BSDFFS and
   1794 		 * occupy the entire disk.
   1795 		 */
   1796 		for (i = 0; i < MAXPARTITIONS; i++) {
   1797 			/*
   1798 			 * Don't wipe out port specific hack (such as
   1799 			 * dos partition hack of i386 port).
   1800 			 */
   1801 			if (lp->d_partitions[i].p_size != 0)
   1802 				continue;
   1803 
   1804 			lp->d_partitions[i].p_size = lp->d_secperunit;
   1805 			lp->d_partitions[i].p_offset = 0;
   1806 			lp->d_partitions[i].p_fstype = FS_BSDFFS;
   1807 		}
   1808 
   1809 		strncpy(lp->d_packname, "default label",
   1810 		    sizeof(lp->d_packname));
   1811 
   1812 		lp->d_npartitions = MAXPARTITIONS;
   1813 		lp->d_checksum = dkcksum(lp);
   1814 	}
   1815 
   1816 	/* In-core label now valid. */
   1817 	sc->sc_flags |= VNF_VLABEL;
   1818 }
   1819 
   1820 /*
   1821  * Wait interruptibly for an exclusive lock.
   1822  *
   1823  * XXX
   1824  * Several drivers do this; it should be abstracted and made MP-safe.
   1825  */
   1826 static int
   1827 vndlock(struct vnd_softc *sc)
   1828 {
   1829 	int error;
   1830 
   1831 	while ((sc->sc_flags & VNF_LOCKED) != 0) {
   1832 		sc->sc_flags |= VNF_WANTED;
   1833 		if ((error = tsleep(sc, PRIBIO | PCATCH, "vndlck", 0)) != 0)
   1834 			return error;
   1835 	}
   1836 	sc->sc_flags |= VNF_LOCKED;
   1837 	return 0;
   1838 }
   1839 
   1840 /*
   1841  * Unlock and wake up any waiters.
   1842  */
   1843 static void
   1844 vndunlock(struct vnd_softc *sc)
   1845 {
   1846 
   1847 	sc->sc_flags &= ~VNF_LOCKED;
   1848 	if ((sc->sc_flags & VNF_WANTED) != 0) {
   1849 		sc->sc_flags &= ~VNF_WANTED;
   1850 		wakeup(sc);
   1851 	}
   1852 }
   1853 
   1854 #ifdef VND_COMPRESSION
   1855 /* compressed file read */
   1856 static void
   1857 compstrategy(struct buf *bp, off_t bn)
   1858 {
   1859 	int error;
   1860 	int unit = vndunit(bp->b_dev);
   1861 	struct vnd_softc *vnd =
   1862 	    device_lookup_private(&vnd_cd, unit);
   1863 	u_int32_t comp_block;
   1864 	struct uio auio;
   1865 	char *addr;
   1866 	int s;
   1867 
   1868 	/* set up constants for data move */
   1869 	auio.uio_rw = UIO_READ;
   1870 	UIO_SETUP_SYSSPACE(&auio);
   1871 
   1872 	/* read, and transfer the data */
   1873 	addr = bp->b_data;
   1874 	bp->b_resid = bp->b_bcount;
   1875 	s = splbio();
   1876 	while (bp->b_resid > 0) {
   1877 		unsigned length;
   1878 		size_t length_in_buffer;
   1879 		u_int32_t offset_in_buffer;
   1880 		struct iovec aiov;
   1881 
   1882 		/* calculate the compressed block number */
   1883 		comp_block = bn / (off_t)vnd->sc_comp_blksz;
   1884 
   1885 		/* check for good block number */
   1886 		if (comp_block >= vnd->sc_comp_numoffs) {
   1887 			bp->b_error = EINVAL;
   1888 			splx(s);
   1889 			return;
   1890 		}
   1891 
   1892 		/* read in the compressed block, if not in buffer */
   1893 		if (comp_block != vnd->sc_comp_buffblk) {
   1894 			length = vnd->sc_comp_offsets[comp_block + 1] -
   1895 			    vnd->sc_comp_offsets[comp_block];
   1896 			vn_lock(vnd->sc_vp, LK_EXCLUSIVE | LK_RETRY);
   1897 			error = vn_rdwr(UIO_READ, vnd->sc_vp, vnd->sc_comp_buff,
   1898 			    length, vnd->sc_comp_offsets[comp_block],
   1899 			    UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, vnd->sc_cred,
   1900 			    NULL, NULL);
   1901 			if (error) {
   1902 				bp->b_error = error;
   1903 				VOP_UNLOCK(vnd->sc_vp);
   1904 				splx(s);
   1905 				return;
   1906 			}
   1907 			/* uncompress the buffer */
   1908 			vnd->sc_comp_stream.next_in = vnd->sc_comp_buff;
   1909 			vnd->sc_comp_stream.avail_in = length;
   1910 			vnd->sc_comp_stream.next_out = vnd->sc_comp_decombuf;
   1911 			vnd->sc_comp_stream.avail_out = vnd->sc_comp_blksz;
   1912 			inflateReset(&vnd->sc_comp_stream);
   1913 			error = inflate(&vnd->sc_comp_stream, Z_FINISH);
   1914 			if (error != Z_STREAM_END) {
   1915 				if (vnd->sc_comp_stream.msg)
   1916 					aprint_normal_dev(vnd->sc_dev,
   1917 					    "compressed file, %s\n",
   1918 					    vnd->sc_comp_stream.msg);
   1919 				bp->b_error = EBADMSG;
   1920 				VOP_UNLOCK(vnd->sc_vp);
   1921 				splx(s);
   1922 				return;
   1923 			}
   1924 			vnd->sc_comp_buffblk = comp_block;
   1925 			VOP_UNLOCK(vnd->sc_vp);
   1926 		}
   1927 
   1928 		/* transfer the usable uncompressed data */
   1929 		offset_in_buffer = bn % (off_t)vnd->sc_comp_blksz;
   1930 		length_in_buffer = vnd->sc_comp_blksz - offset_in_buffer;
   1931 		if (length_in_buffer > bp->b_resid)
   1932 			length_in_buffer = bp->b_resid;
   1933 		auio.uio_iov = &aiov;
   1934 		auio.uio_iovcnt = 1;
   1935 		aiov.iov_base = addr;
   1936 		aiov.iov_len = length_in_buffer;
   1937 		auio.uio_resid = aiov.iov_len;
   1938 		auio.uio_offset = 0;
   1939 		error = uiomove(vnd->sc_comp_decombuf + offset_in_buffer,
   1940 		    length_in_buffer, &auio);
   1941 		if (error) {
   1942 			bp->b_error = error;
   1943 			splx(s);
   1944 			return;
   1945 		}
   1946 
   1947 		bn += length_in_buffer;
   1948 		addr += length_in_buffer;
   1949 		bp->b_resid -= length_in_buffer;
   1950 	}
   1951 	splx(s);
   1952 }
   1953 
   1954 /* compression memory allocation routines */
   1955 static void *
   1956 vnd_alloc(void *aux, u_int items, u_int siz)
   1957 {
   1958 	return malloc(items * siz, M_TEMP, M_NOWAIT);
   1959 }
   1960 
   1961 static void
   1962 vnd_free(void *aux, void *ptr)
   1963 {
   1964 	free(ptr, M_TEMP);
   1965 }
   1966 #endif /* VND_COMPRESSION */
   1967 
   1968 static void
   1969 vnd_set_properties(struct vnd_softc *vnd)
   1970 {
   1971 	prop_dictionary_t disk_info, odisk_info, geom;
   1972 
   1973 	disk_info = prop_dictionary_create();
   1974 
   1975 	geom = prop_dictionary_create();
   1976 
   1977 	prop_dictionary_set_uint64(geom, "sectors-per-unit",
   1978 	    vnd->sc_geom.vng_nsectors * vnd->sc_geom.vng_ntracks *
   1979 	    vnd->sc_geom.vng_ncylinders);
   1980 
   1981 	prop_dictionary_set_uint32(geom, "sector-size",
   1982 	    vnd->sc_geom.vng_secsize);
   1983 
   1984 	prop_dictionary_set_uint16(geom, "sectors-per-track",
   1985 	    vnd->sc_geom.vng_nsectors);
   1986 
   1987 	prop_dictionary_set_uint16(geom, "tracks-per-cylinder",
   1988 	    vnd->sc_geom.vng_ntracks);
   1989 
   1990 	prop_dictionary_set_uint64(geom, "cylinders-per-unit",
   1991 	    vnd->sc_geom.vng_ncylinders);
   1992 
   1993 	prop_dictionary_set(disk_info, "geometry", geom);
   1994 	prop_object_release(geom);
   1995 
   1996 	prop_dictionary_set(device_properties(vnd->sc_dev),
   1997 	    "disk-info", disk_info);
   1998 
   1999 	/*
   2000 	 * Don't release disk_info here; we keep a reference to it.
   2001 	 * disk_detach() will release it when we go away.
   2002 	 */
   2003 
   2004 	odisk_info = vnd->sc_dkdev.dk_info;
   2005 	vnd->sc_dkdev.dk_info = disk_info;
   2006 	if (odisk_info)
   2007 		prop_object_release(odisk_info);
   2008 }
   2009 
   2010 #ifdef _MODULE
   2011 
   2012 #include <sys/module.h>
   2013 
   2014 MODULE(MODULE_CLASS_DRIVER, vnd, NULL);
   2015 CFDRIVER_DECL(vnd, DV_DISK, NULL);
   2016 
   2017 static int
   2018 vnd_modcmd(modcmd_t cmd, void *arg)
   2019 {
   2020 	int bmajor = -1, cmajor = -1,  error = 0;
   2021 
   2022 	switch (cmd) {
   2023 	case MODULE_CMD_INIT:
   2024 		error = config_cfdriver_attach(&vnd_cd);
   2025 		if (error)
   2026 			break;
   2027 
   2028 		error = config_cfattach_attach(vnd_cd.cd_name, &vnd_ca);
   2029 	        if (error) {
   2030 			config_cfdriver_detach(&vnd_cd);
   2031 			aprint_error("%s: unable to register cfattach\n",
   2032 			    vnd_cd.cd_name);
   2033 			break;
   2034 		}
   2035 
   2036 		error = devsw_attach("vnd", &vnd_bdevsw, &bmajor,
   2037 		    &vnd_cdevsw, &cmajor);
   2038 		if (error) {
   2039 			config_cfattach_detach(vnd_cd.cd_name, &vnd_ca);
   2040 			config_cfdriver_detach(&vnd_cd);
   2041 			break;
   2042 		}
   2043 
   2044 		break;
   2045 
   2046 	case MODULE_CMD_FINI:
   2047 		error = config_cfattach_detach(vnd_cd.cd_name, &vnd_ca);
   2048 		if (error)
   2049 			break;
   2050 		config_cfdriver_detach(&vnd_cd);
   2051 		devsw_detach(&vnd_bdevsw, &vnd_cdevsw);
   2052 		break;
   2053 
   2054 	case MODULE_CMD_STAT:
   2055 		return ENOTTY;
   2056 
   2057 	default:
   2058 		return ENOTTY;
   2059 	}
   2060 
   2061 	return error;
   2062 }
   2063 
   2064 #endif
   2065