Home | History | Annotate | Line # | Download | only in dev
ccd.c revision 1.109
      1 /*	$NetBSD: ccd.c,v 1.109 2006/04/24 21:35:57 jld Exp $	*/
      2 
      3 /*-
      4  * Copyright (c) 1996, 1997, 1998, 1999 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  * 3. All advertising materials mentioning features or use of this software
     19  *    must display the following acknowledgement:
     20  *        This product includes software developed by the NetBSD
     21  *        Foundation, Inc. and its contributors.
     22  * 4. Neither the name of The NetBSD Foundation nor the names of its
     23  *    contributors may be used to endorse or promote products derived
     24  *    from this software without specific prior written permission.
     25  *
     26  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     27  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     28  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     29  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     30  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     31  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     32  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     33  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     34  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     35  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     36  * POSSIBILITY OF SUCH DAMAGE.
     37  */
     38 
     39 /*
     40  * Copyright (c) 1990, 1993
     41  *	The Regents of the University of California.  All rights reserved.
     42  *
     43  * This code is derived from software contributed to Berkeley by
     44  * the Systems Programming Group of the University of Utah Computer
     45  * Science Department.
     46  *
     47  * Redistribution and use in source and binary forms, with or without
     48  * modification, are permitted provided that the following conditions
     49  * are met:
     50  * 1. Redistributions of source code must retain the above copyright
     51  *    notice, this list of conditions and the following disclaimer.
     52  * 2. Redistributions in binary form must reproduce the above copyright
     53  *    notice, this list of conditions and the following disclaimer in the
     54  *    documentation and/or other materials provided with the distribution.
     55  * 3. Neither the name of the University nor the names of its contributors
     56  *    may be used to endorse or promote products derived from this software
     57  *    without specific prior written permission.
     58  *
     59  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     60  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     61  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     62  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     63  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     64  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     65  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     66  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     67  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     68  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     69  * SUCH DAMAGE.
     70  *
     71  * from: Utah $Hdr: cd.c 1.6 90/11/28$
     72  *
     73  *	@(#)cd.c	8.2 (Berkeley) 11/16/93
     74  */
     75 
     76 /*
     77  * Copyright (c) 1988 University of Utah.
     78  *
     79  * This code is derived from software contributed to Berkeley by
     80  * the Systems Programming Group of the University of Utah Computer
     81  * Science Department.
     82  *
     83  * Redistribution and use in source and binary forms, with or without
     84  * modification, are permitted provided that the following conditions
     85  * are met:
     86  * 1. Redistributions of source code must retain the above copyright
     87  *    notice, this list of conditions and the following disclaimer.
     88  * 2. Redistributions in binary form must reproduce the above copyright
     89  *    notice, this list of conditions and the following disclaimer in the
     90  *    documentation and/or other materials provided with the distribution.
     91  * 3. All advertising materials mentioning features or use of this software
     92  *    must display the following acknowledgement:
     93  *	This product includes software developed by the University of
     94  *	California, Berkeley and its contributors.
     95  * 4. Neither the name of the University nor the names of its contributors
     96  *    may be used to endorse or promote products derived from this software
     97  *    without specific prior written permission.
     98  *
     99  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
    100  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
    101  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
    102  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
    103  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
    104  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
    105  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
    106  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
    107  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
    108  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
    109  * SUCH DAMAGE.
    110  *
    111  * from: Utah $Hdr: cd.c 1.6 90/11/28$
    112  *
    113  *	@(#)cd.c	8.2 (Berkeley) 11/16/93
    114  */
    115 
    116 /*
    117  * "Concatenated" disk driver.
    118  *
    119  * Dynamic configuration and disklabel support by:
    120  *	Jason R. Thorpe <thorpej (at) nas.nasa.gov>
    121  *	Numerical Aerodynamic Simulation Facility
    122  *	Mail Stop 258-6
    123  *	NASA Ames Research Center
    124  *	Moffett Field, CA 94035
    125  */
    126 
    127 #include <sys/cdefs.h>
    128 __KERNEL_RCSID(0, "$NetBSD: ccd.c,v 1.109 2006/04/24 21:35:57 jld Exp $");
    129 
    130 #include <sys/param.h>
    131 #include <sys/systm.h>
    132 #include <sys/proc.h>
    133 #include <sys/errno.h>
    134 #include <sys/buf.h>
    135 #include <sys/bufq.h>
    136 #include <sys/malloc.h>
    137 #include <sys/pool.h>
    138 #include <sys/namei.h>
    139 #include <sys/stat.h>
    140 #include <sys/ioctl.h>
    141 #include <sys/disklabel.h>
    142 #include <sys/device.h>
    143 #include <sys/disk.h>
    144 #include <sys/syslog.h>
    145 #include <sys/fcntl.h>
    146 #include <sys/vnode.h>
    147 #include <sys/conf.h>
    148 #include <sys/lock.h>
    149 #include <sys/queue.h>
    150 
    151 #include <dev/ccdvar.h>
    152 
    153 #if defined(CCDDEBUG) && !defined(DEBUG)
    154 #define DEBUG
    155 #endif
    156 
    157 #ifdef DEBUG
    158 #define CCDB_FOLLOW	0x01
    159 #define CCDB_INIT	0x02
    160 #define CCDB_IO		0x04
    161 #define CCDB_LABEL	0x08
    162 #define CCDB_VNODE	0x10
    163 int ccddebug = 0x00;
    164 #endif
    165 
    166 #define	ccdunit(x)	DISKUNIT(x)
    167 
    168 struct ccdbuf {
    169 	struct buf	cb_buf;		/* new I/O buf */
    170 	struct buf	*cb_obp;	/* ptr. to original I/O buf */
    171 	struct ccd_softc *cb_sc;	/* pointer to ccd softc */
    172 	int		cb_comp;	/* target component */
    173 	SIMPLEQ_ENTRY(ccdbuf) cb_q;	/* fifo of component buffers */
    174 };
    175 
    176 /* component buffer pool */
    177 static struct pool ccd_cbufpool;
    178 
    179 #define	CCD_GETBUF()		pool_get(&ccd_cbufpool, PR_NOWAIT)
    180 #define	CCD_PUTBUF(cbp)		pool_put(&ccd_cbufpool, cbp)
    181 
    182 #define CCDLABELDEV(dev)	\
    183 	(MAKEDISKDEV(major((dev)), ccdunit((dev)), RAW_PART))
    184 
    185 /* called by main() at boot time */
    186 void	ccdattach(int);
    187 
    188 /* called by biodone() at interrupt time */
    189 static void	ccdiodone(struct buf *);
    190 
    191 static void	ccdstart(struct ccd_softc *);
    192 static void	ccdinterleave(struct ccd_softc *);
    193 static void	ccdintr(struct ccd_softc *, struct buf *);
    194 static int	ccdinit(struct ccd_softc *, char **, struct vnode **,
    195 		    struct lwp *);
    196 static int	ccdlookup(char *, struct lwp *, struct vnode **);
    197 static struct ccdbuf *ccdbuffer(struct ccd_softc *, struct buf *,
    198 		    daddr_t, caddr_t, long);
    199 static void	ccdgetdefaultlabel(struct ccd_softc *, struct disklabel *);
    200 static void	ccdgetdisklabel(dev_t);
    201 static void	ccdmakedisklabel(struct ccd_softc *);
    202 
    203 static dev_type_open(ccdopen);
    204 static dev_type_close(ccdclose);
    205 static dev_type_read(ccdread);
    206 static dev_type_write(ccdwrite);
    207 static dev_type_ioctl(ccdioctl);
    208 static dev_type_strategy(ccdstrategy);
    209 static dev_type_dump(ccddump);
    210 static dev_type_size(ccdsize);
    211 
    212 const struct bdevsw ccd_bdevsw = {
    213 	ccdopen, ccdclose, ccdstrategy, ccdioctl, ccddump, ccdsize, D_DISK
    214 };
    215 
    216 const struct cdevsw ccd_cdevsw = {
    217 	ccdopen, ccdclose, ccdread, ccdwrite, ccdioctl,
    218 	nostop, notty, nopoll, nommap, nokqfilter, D_DISK
    219 };
    220 
    221 #ifdef DEBUG
    222 static	void printiinfo(struct ccdiinfo *);
    223 #endif
    224 
    225 /* Publically visible for the benefit of libkvm and ccdconfig(8). */
    226 struct ccd_softc 	*ccd_softc;
    227 const int		ccd_softc_elemsize = sizeof(struct ccd_softc);
    228 int			numccd = 0;
    229 
    230 /*
    231  * Called by main() during pseudo-device attachment.  All we need
    232  * to do is allocate enough space for devices to be configured later.
    233  */
    234 void
    235 ccdattach(int num)
    236 {
    237 	struct ccd_softc *cs;
    238 	int i;
    239 
    240 	if (num <= 0) {
    241 #ifdef DIAGNOSTIC
    242 		panic("ccdattach: count <= 0");
    243 #endif
    244 		return;
    245 	}
    246 
    247 	ccd_softc = (struct ccd_softc *)malloc(num * ccd_softc_elemsize,
    248 	    M_DEVBUF, M_NOWAIT|M_ZERO);
    249 	if (ccd_softc == NULL) {
    250 		printf("WARNING: no memory for concatenated disks\n");
    251 		return;
    252 	}
    253 	numccd = num;
    254 
    255 	/* Initialize the component buffer pool. */
    256 	pool_init(&ccd_cbufpool, sizeof(struct ccdbuf), 0,
    257 	    0, 0, "ccdpl", NULL);
    258 
    259 	/* Initialize per-softc structures. */
    260 	for (i = 0; i < num; i++) {
    261 		cs = &ccd_softc[i];
    262 		snprintf(cs->sc_xname, sizeof(cs->sc_xname), "ccd%d", i);
    263 		cs->sc_dkdev.dk_name = cs->sc_xname;	/* XXX */
    264 		lockinit(&cs->sc_lock, PRIBIO, "ccdlk", 0, 0);
    265 		pseudo_disk_init(&cs->sc_dkdev);
    266 	}
    267 }
    268 
    269 static int
    270 ccdinit(struct ccd_softc *cs, char **cpaths, struct vnode **vpp,
    271     struct lwp *l)
    272 {
    273 	struct ccdcinfo *ci = NULL;
    274 	size_t size;
    275 	int ix;
    276 	struct vattr va;
    277 	size_t minsize;
    278 	int maxsecsize;
    279 	struct partinfo dpart;
    280 	struct ccdgeom *ccg = &cs->sc_geom;
    281 	char tmppath[MAXPATHLEN];
    282 	int error, path_alloced;
    283 
    284 #ifdef DEBUG
    285 	if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
    286 		printf("%s: ccdinit\n", cs->sc_xname);
    287 #endif
    288 
    289 	/* Allocate space for the component info. */
    290 	cs->sc_cinfo = malloc(cs->sc_nccdisks * sizeof(struct ccdcinfo),
    291 	    M_DEVBUF, M_WAITOK);
    292 
    293 	cs->sc_size = 0;
    294 
    295 	/*
    296 	 * Verify that each component piece exists and record
    297 	 * relevant information about it.
    298 	 */
    299 	maxsecsize = 0;
    300 	minsize = 0;
    301 	for (ix = 0, path_alloced = 0; ix < cs->sc_nccdisks; ix++) {
    302 		ci = &cs->sc_cinfo[ix];
    303 		ci->ci_vp = vpp[ix];
    304 
    305 		/*
    306 		 * Copy in the pathname of the component.
    307 		 */
    308 		memset(tmppath, 0, sizeof(tmppath));	/* sanity */
    309 		error = copyinstr(cpaths[ix], tmppath,
    310 		    MAXPATHLEN, &ci->ci_pathlen);
    311 		if (error) {
    312 #ifdef DEBUG
    313 			if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
    314 				printf("%s: can't copy path, error = %d\n",
    315 				    cs->sc_xname, error);
    316 #endif
    317 			goto out;
    318 		}
    319 		ci->ci_path = malloc(ci->ci_pathlen, M_DEVBUF, M_WAITOK);
    320 		memcpy(ci->ci_path, tmppath, ci->ci_pathlen);
    321 		path_alloced++;
    322 
    323 		/*
    324 		 * XXX: Cache the component's dev_t.
    325 		 */
    326 		if ((error = VOP_GETATTR(vpp[ix], &va, l->l_proc->p_ucred, l)) != 0) {
    327 #ifdef DEBUG
    328 			if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
    329 				printf("%s: %s: getattr failed %s = %d\n",
    330 				    cs->sc_xname, ci->ci_path,
    331 				    "error", error);
    332 #endif
    333 			goto out;
    334 		}
    335 		ci->ci_dev = va.va_rdev;
    336 
    337 		/*
    338 		 * Get partition information for the component.
    339 		 */
    340 		error = VOP_IOCTL(vpp[ix], DIOCGPART, &dpart,
    341 		    FREAD, l->l_proc->p_ucred, l);
    342 		if (error) {
    343 #ifdef DEBUG
    344 			if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
    345 				 printf("%s: %s: ioctl failed, error = %d\n",
    346 				     cs->sc_xname, ci->ci_path, error);
    347 #endif
    348 			goto out;
    349 		}
    350 
    351 /*
    352  * This diagnostic test is disabled (for now?) since not all port supports
    353  * on-disk BSD disklabel.
    354  */
    355 #if 0 /* def DIAGNOSTIC */
    356 		/* Check fstype field of component. */
    357 		if (dpart.part->p_fstype != FS_CCD)
    358 			printf("%s: WARNING: %s: fstype %d != FS_CCD\n",
    359 			    cs->sc_xname, ci->ci_path, dpart.part->p_fstype);
    360 #endif
    361 
    362 		/*
    363 		 * Calculate the size, truncating to an interleave
    364 		 * boundary if necessary.
    365 		 */
    366 		maxsecsize =
    367 		    ((dpart.disklab->d_secsize > maxsecsize) ?
    368 		    dpart.disklab->d_secsize : maxsecsize);
    369 		size = dpart.part->p_size;
    370 		if (cs->sc_ileave > 1)
    371 			size -= size % cs->sc_ileave;
    372 
    373 		if (size == 0) {
    374 #ifdef DEBUG
    375 			if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
    376 				printf("%s: %s: size == 0\n",
    377 				    cs->sc_xname, ci->ci_path);
    378 #endif
    379 			error = ENODEV;
    380 			goto out;
    381 		}
    382 
    383 		if (minsize == 0 || size < minsize)
    384 			minsize = size;
    385 		ci->ci_size = size;
    386 		cs->sc_size += size;
    387 	}
    388 
    389 	/*
    390 	 * Don't allow the interleave to be smaller than
    391 	 * the biggest component sector.
    392 	 */
    393 	if ((cs->sc_ileave > 0) &&
    394 	    (cs->sc_ileave < (maxsecsize / DEV_BSIZE))) {
    395 #ifdef DEBUG
    396 		if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
    397 			printf("%s: interleave must be at least %d\n",
    398 			    cs->sc_xname, (maxsecsize / DEV_BSIZE));
    399 #endif
    400 		error = EINVAL;
    401 		goto out;
    402 	}
    403 
    404 	/*
    405 	 * If uniform interleave is desired set all sizes to that of
    406 	 * the smallest component.
    407 	 */
    408 	if (cs->sc_flags & CCDF_UNIFORM) {
    409 		for (ci = cs->sc_cinfo;
    410 		     ci < &cs->sc_cinfo[cs->sc_nccdisks]; ci++)
    411 			ci->ci_size = minsize;
    412 
    413 		cs->sc_size = cs->sc_nccdisks * minsize;
    414 	}
    415 
    416 	/*
    417 	 * Construct the interleave table.
    418 	 */
    419 	ccdinterleave(cs);
    420 
    421 	/*
    422 	 * Create pseudo-geometry based on 1MB cylinders.  It's
    423 	 * pretty close.
    424 	 */
    425 	ccg->ccg_secsize = DEV_BSIZE;
    426 	ccg->ccg_ntracks = 1;
    427 	ccg->ccg_nsectors = 1024 * (1024 / ccg->ccg_secsize);
    428 	ccg->ccg_ncylinders = cs->sc_size / ccg->ccg_nsectors;
    429 
    430 	cs->sc_flags |= CCDF_INITED;
    431 
    432 	return (0);
    433 
    434  out:
    435 	for (ix = 0; ix < path_alloced; ix++)
    436 		free(cs->sc_cinfo[ix].ci_path, M_DEVBUF);
    437 	free(cs->sc_cinfo, M_DEVBUF);
    438 	return (error);
    439 }
    440 
    441 static void
    442 ccdinterleave(struct ccd_softc *cs)
    443 {
    444 	struct ccdcinfo *ci, *smallci;
    445 	struct ccdiinfo *ii;
    446 	daddr_t bn, lbn;
    447 	int ix;
    448 	u_long size;
    449 
    450 #ifdef DEBUG
    451 	if (ccddebug & CCDB_INIT)
    452 		printf("ccdinterleave(%p): ileave %d\n", cs, cs->sc_ileave);
    453 #endif
    454 	/*
    455 	 * Allocate an interleave table.
    456 	 * Chances are this is too big, but we don't care.
    457 	 */
    458 	size = (cs->sc_nccdisks + 1) * sizeof(struct ccdiinfo);
    459 	cs->sc_itable = (struct ccdiinfo *)malloc(size, M_DEVBUF,
    460 	    M_WAITOK|M_ZERO);
    461 
    462 	/*
    463 	 * Trivial case: no interleave (actually interleave of disk size).
    464 	 * Each table entry represents a single component in its entirety.
    465 	 */
    466 	if (cs->sc_ileave == 0) {
    467 		bn = 0;
    468 		ii = cs->sc_itable;
    469 
    470 		for (ix = 0; ix < cs->sc_nccdisks; ix++) {
    471 			/* Allocate space for ii_index. */
    472 			ii->ii_index = malloc(sizeof(int), M_DEVBUF, M_WAITOK);
    473 			ii->ii_ndisk = 1;
    474 			ii->ii_startblk = bn;
    475 			ii->ii_startoff = 0;
    476 			ii->ii_index[0] = ix;
    477 			bn += cs->sc_cinfo[ix].ci_size;
    478 			ii++;
    479 		}
    480 		ii->ii_ndisk = 0;
    481 #ifdef DEBUG
    482 		if (ccddebug & CCDB_INIT)
    483 			printiinfo(cs->sc_itable);
    484 #endif
    485 		return;
    486 	}
    487 
    488 	/*
    489 	 * The following isn't fast or pretty; it doesn't have to be.
    490 	 */
    491 	size = 0;
    492 	bn = lbn = 0;
    493 	for (ii = cs->sc_itable; ; ii++) {
    494 		/* Allocate space for ii_index. */
    495 		ii->ii_index = malloc((sizeof(int) * cs->sc_nccdisks),
    496 		    M_DEVBUF, M_WAITOK);
    497 
    498 		/*
    499 		 * Locate the smallest of the remaining components
    500 		 */
    501 		smallci = NULL;
    502 		for (ci = cs->sc_cinfo;
    503 		     ci < &cs->sc_cinfo[cs->sc_nccdisks]; ci++)
    504 			if (ci->ci_size > size &&
    505 			    (smallci == NULL ||
    506 			     ci->ci_size < smallci->ci_size))
    507 				smallci = ci;
    508 
    509 		/*
    510 		 * Nobody left, all done
    511 		 */
    512 		if (smallci == NULL) {
    513 			ii->ii_ndisk = 0;
    514 			break;
    515 		}
    516 
    517 		/*
    518 		 * Record starting logical block and component offset
    519 		 */
    520 		ii->ii_startblk = bn / cs->sc_ileave;
    521 		ii->ii_startoff = lbn;
    522 
    523 		/*
    524 		 * Determine how many disks take part in this interleave
    525 		 * and record their indices.
    526 		 */
    527 		ix = 0;
    528 		for (ci = cs->sc_cinfo;
    529 		     ci < &cs->sc_cinfo[cs->sc_nccdisks]; ci++)
    530 			if (ci->ci_size >= smallci->ci_size)
    531 				ii->ii_index[ix++] = ci - cs->sc_cinfo;
    532 		ii->ii_ndisk = ix;
    533 		bn += ix * (smallci->ci_size - size);
    534 		lbn = smallci->ci_size / cs->sc_ileave;
    535 		size = smallci->ci_size;
    536 	}
    537 #ifdef DEBUG
    538 	if (ccddebug & CCDB_INIT)
    539 		printiinfo(cs->sc_itable);
    540 #endif
    541 }
    542 
    543 /* ARGSUSED */
    544 static int
    545 ccdopen(dev_t dev, int flags, int fmt, struct lwp *l)
    546 {
    547 	int unit = ccdunit(dev);
    548 	struct ccd_softc *cs;
    549 	struct disklabel *lp;
    550 	int error = 0, part, pmask;
    551 
    552 #ifdef DEBUG
    553 	if (ccddebug & CCDB_FOLLOW)
    554 		printf("ccdopen(0x%x, 0x%x)\n", dev, flags);
    555 #endif
    556 	if (unit >= numccd)
    557 		return (ENXIO);
    558 	cs = &ccd_softc[unit];
    559 
    560 	if ((error = lockmgr(&cs->sc_lock, LK_EXCLUSIVE, NULL)) != 0)
    561 		return (error);
    562 
    563 	lp = cs->sc_dkdev.dk_label;
    564 
    565 	part = DISKPART(dev);
    566 	pmask = (1 << part);
    567 
    568 	/*
    569 	 * If we're initialized, check to see if there are any other
    570 	 * open partitions.  If not, then it's safe to update
    571 	 * the in-core disklabel.  Only read the disklabel if it is
    572 	 * not already valid.
    573 	 */
    574 	if ((cs->sc_flags & (CCDF_INITED|CCDF_VLABEL)) == CCDF_INITED &&
    575 	    cs->sc_dkdev.dk_openmask == 0)
    576 		ccdgetdisklabel(dev);
    577 
    578 	/* Check that the partition exists. */
    579 	if (part != RAW_PART) {
    580 		if (((cs->sc_flags & CCDF_INITED) == 0) ||
    581 		    ((part >= lp->d_npartitions) ||
    582 		     (lp->d_partitions[part].p_fstype == FS_UNUSED))) {
    583 			error = ENXIO;
    584 			goto done;
    585 		}
    586 	}
    587 
    588 	/* Prevent our unit from being unconfigured while open. */
    589 	switch (fmt) {
    590 	case S_IFCHR:
    591 		cs->sc_dkdev.dk_copenmask |= pmask;
    592 		break;
    593 
    594 	case S_IFBLK:
    595 		cs->sc_dkdev.dk_bopenmask |= pmask;
    596 		break;
    597 	}
    598 	cs->sc_dkdev.dk_openmask =
    599 	    cs->sc_dkdev.dk_copenmask | cs->sc_dkdev.dk_bopenmask;
    600 
    601  done:
    602 	(void) lockmgr(&cs->sc_lock, LK_RELEASE, NULL);
    603 	return (error);
    604 }
    605 
    606 /* ARGSUSED */
    607 static int
    608 ccdclose(dev_t dev, int flags, int fmt, struct lwp *l)
    609 {
    610 	int unit = ccdunit(dev);
    611 	struct ccd_softc *cs;
    612 	int error = 0, part;
    613 
    614 #ifdef DEBUG
    615 	if (ccddebug & CCDB_FOLLOW)
    616 		printf("ccdclose(0x%x, 0x%x)\n", dev, flags);
    617 #endif
    618 
    619 	if (unit >= numccd)
    620 		return (ENXIO);
    621 	cs = &ccd_softc[unit];
    622 
    623 	if ((error = lockmgr(&cs->sc_lock, LK_EXCLUSIVE, NULL)) != 0)
    624 		return (error);
    625 
    626 	part = DISKPART(dev);
    627 
    628 	/* ...that much closer to allowing unconfiguration... */
    629 	switch (fmt) {
    630 	case S_IFCHR:
    631 		cs->sc_dkdev.dk_copenmask &= ~(1 << part);
    632 		break;
    633 
    634 	case S_IFBLK:
    635 		cs->sc_dkdev.dk_bopenmask &= ~(1 << part);
    636 		break;
    637 	}
    638 	cs->sc_dkdev.dk_openmask =
    639 	    cs->sc_dkdev.dk_copenmask | cs->sc_dkdev.dk_bopenmask;
    640 
    641 	if (cs->sc_dkdev.dk_openmask == 0) {
    642 		if ((cs->sc_flags & CCDF_KLABEL) == 0)
    643 			cs->sc_flags &= ~CCDF_VLABEL;
    644 	}
    645 
    646 	(void) lockmgr(&cs->sc_lock, LK_RELEASE, NULL);
    647 	return (0);
    648 }
    649 
    650 static void
    651 ccdstrategy(struct buf *bp)
    652 {
    653 	int unit = ccdunit(bp->b_dev);
    654 	struct ccd_softc *cs = &ccd_softc[unit];
    655 	daddr_t blkno;
    656 	int s;
    657 	int wlabel;
    658 	struct disklabel *lp;
    659 
    660 #ifdef DEBUG
    661 	if (ccddebug & CCDB_FOLLOW)
    662 		printf("ccdstrategy(%p): unit %d\n", bp, unit);
    663 #endif
    664 	if ((cs->sc_flags & CCDF_INITED) == 0) {
    665 #ifdef DEBUG
    666 		if (ccddebug & CCDB_FOLLOW)
    667 			printf("ccdstrategy: unit %d: not inited\n", unit);
    668 #endif
    669 		bp->b_error = ENXIO;
    670 		bp->b_flags |= B_ERROR;
    671 		goto done;
    672 	}
    673 
    674 	/* If it's a nil transfer, wake up the top half now. */
    675 	if (bp->b_bcount == 0)
    676 		goto done;
    677 
    678 	lp = cs->sc_dkdev.dk_label;
    679 
    680 	/*
    681 	 * Do bounds checking and adjust transfer.  If there's an
    682 	 * error, the bounds check will flag that for us.  Convert
    683 	 * the partition relative block number to an absolute.
    684 	 */
    685 	blkno = bp->b_blkno;
    686 	wlabel = cs->sc_flags & (CCDF_WLABEL|CCDF_LABELLING);
    687 	if (DISKPART(bp->b_dev) != RAW_PART) {
    688 		if (bounds_check_with_label(&cs->sc_dkdev, bp, wlabel) <= 0)
    689 			goto done;
    690 		blkno += lp->d_partitions[DISKPART(bp->b_dev)].p_offset;
    691 	}
    692 	bp->b_rawblkno = blkno;
    693 
    694 	/* Place it in the queue and start I/O on the unit. */
    695 	s = splbio();
    696 	BUFQ_PUT(cs->sc_bufq, bp);
    697 	ccdstart(cs);
    698 	splx(s);
    699 	return;
    700 
    701  done:
    702 	bp->b_resid = bp->b_bcount;
    703 	biodone(bp);
    704 }
    705 
    706 static void
    707 ccdstart(struct ccd_softc *cs)
    708 {
    709 	long bcount, rcount;
    710 	struct buf *bp;
    711 	struct ccdbuf *cbp;
    712 	caddr_t addr;
    713 	daddr_t bn;
    714 	SIMPLEQ_HEAD(, ccdbuf) cbufq;
    715 
    716 #ifdef DEBUG
    717 	if (ccddebug & CCDB_FOLLOW)
    718 		printf("ccdstart(%p)\n", cs);
    719 #endif
    720 
    721 	/* See if there is work for us to do. */
    722 	while ((bp = BUFQ_PEEK(cs->sc_bufq)) != NULL) {
    723 		/* Instrumentation. */
    724 		disk_busy(&cs->sc_dkdev);
    725 
    726 		bp->b_resid = bp->b_bcount;
    727 		bn = bp->b_rawblkno;
    728 
    729 		/* Allocate the component buffers. */
    730 		SIMPLEQ_INIT(&cbufq);
    731 		addr = bp->b_data;
    732 		for (bcount = bp->b_bcount; bcount > 0; bcount -= rcount) {
    733 			cbp = ccdbuffer(cs, bp, bn, addr, bcount);
    734 			if (cbp == NULL) {
    735 				/*
    736 				 * Can't allocate a component buffer; just
    737 				 * defer the job until later.
    738 				 *
    739 				 * XXX We might consider a watchdog timer
    740 				 * XXX to make sure we are kicked into action,
    741 				 * XXX or consider a low-water mark for our
    742 				 * XXX component buffer pool.
    743 				 */
    744 				while ((cbp = SIMPLEQ_FIRST(&cbufq)) != NULL) {
    745 					SIMPLEQ_REMOVE_HEAD(&cbufq, cb_q);
    746 					CCD_PUTBUF(cbp);
    747 				}
    748 				disk_unbusy(&cs->sc_dkdev, 0, 0);
    749 				return;
    750 			}
    751 			SIMPLEQ_INSERT_TAIL(&cbufq, cbp, cb_q);
    752 			rcount = cbp->cb_buf.b_bcount;
    753 			bn += btodb(rcount);
    754 			addr += rcount;
    755 		}
    756 
    757 		/* Transfer all set up, remove job from the queue. */
    758 		(void) BUFQ_GET(cs->sc_bufq);
    759 
    760 		/* Now fire off the requests. */
    761 		while ((cbp = SIMPLEQ_FIRST(&cbufq)) != NULL) {
    762 			SIMPLEQ_REMOVE_HEAD(&cbufq, cb_q);
    763 			if ((cbp->cb_buf.b_flags & B_READ) == 0)
    764 				cbp->cb_buf.b_vp->v_numoutput++;
    765 			DEV_STRATEGY(&cbp->cb_buf);
    766 		}
    767 	}
    768 }
    769 
    770 /*
    771  * Build a component buffer header.
    772  */
    773 static struct ccdbuf *
    774 ccdbuffer(struct ccd_softc *cs, struct buf *bp, daddr_t bn, caddr_t addr,
    775     long bcount)
    776 {
    777 	struct ccdcinfo *ci;
    778 	struct ccdbuf *cbp;
    779 	daddr_t cbn, cboff;
    780 	u_int64_t cbc;
    781 	int ccdisk;
    782 
    783 #ifdef DEBUG
    784 	if (ccddebug & CCDB_IO)
    785 		printf("ccdbuffer(%p, %p, %" PRId64 ", %p, %ld)\n",
    786 		       cs, bp, bn, addr, bcount);
    787 #endif
    788 	/*
    789 	 * Determine which component bn falls in.
    790 	 */
    791 	cbn = bn;
    792 	cboff = 0;
    793 
    794 	/*
    795 	 * Serially concatenated
    796 	 */
    797 	if (cs->sc_ileave == 0) {
    798 		daddr_t sblk;
    799 
    800 		sblk = 0;
    801 		for (ccdisk = 0, ci = &cs->sc_cinfo[ccdisk];
    802 		    cbn >= sblk + ci->ci_size;
    803 		    ccdisk++, ci = &cs->sc_cinfo[ccdisk])
    804 			sblk += ci->ci_size;
    805 		cbn -= sblk;
    806 	}
    807 	/*
    808 	 * Interleaved
    809 	 */
    810 	else {
    811 		struct ccdiinfo *ii;
    812 		int off;
    813 
    814 		cboff = cbn % cs->sc_ileave;
    815 		cbn /= cs->sc_ileave;
    816 		for (ii = cs->sc_itable; ii->ii_ndisk; ii++)
    817 			if (ii->ii_startblk > cbn)
    818 				break;
    819 		ii--;
    820 		off = cbn - ii->ii_startblk;
    821 		if (ii->ii_ndisk == 1) {
    822 			ccdisk = ii->ii_index[0];
    823 			cbn = ii->ii_startoff + off;
    824 		} else {
    825 			ccdisk = ii->ii_index[off % ii->ii_ndisk];
    826 			cbn = ii->ii_startoff + off / ii->ii_ndisk;
    827 		}
    828 		cbn *= cs->sc_ileave;
    829 		ci = &cs->sc_cinfo[ccdisk];
    830 	}
    831 
    832 	/*
    833 	 * Fill in the component buf structure.
    834 	 */
    835 	cbp = CCD_GETBUF();
    836 	if (cbp == NULL)
    837 		return (NULL);
    838 	BUF_INIT(&cbp->cb_buf);
    839 	cbp->cb_buf.b_flags = bp->b_flags | B_CALL;
    840 	cbp->cb_buf.b_iodone = ccdiodone;
    841 	cbp->cb_buf.b_proc = bp->b_proc;
    842 	cbp->cb_buf.b_dev = ci->ci_dev;
    843 	cbp->cb_buf.b_blkno = cbn + cboff;
    844 	cbp->cb_buf.b_data = addr;
    845 	cbp->cb_buf.b_vp = ci->ci_vp;
    846 	if (cs->sc_ileave == 0)
    847 		cbc = dbtob((u_int64_t)(ci->ci_size - cbn));
    848 	else
    849 		cbc = dbtob((u_int64_t)(cs->sc_ileave - cboff));
    850 	cbp->cb_buf.b_bcount = cbc < bcount ? cbc : bcount;
    851 
    852 	/*
    853 	 * context for ccdiodone
    854 	 */
    855 	cbp->cb_obp = bp;
    856 	cbp->cb_sc = cs;
    857 	cbp->cb_comp = ccdisk;
    858 
    859 	BIO_COPYPRIO(&cbp->cb_buf, bp);
    860 
    861 #ifdef DEBUG
    862 	if (ccddebug & CCDB_IO)
    863 		printf(" dev 0x%x(u%lu): cbp %p bn %" PRId64 " addr %p"
    864 		       " bcnt %d\n",
    865 		    ci->ci_dev, (unsigned long) (ci-cs->sc_cinfo), cbp,
    866 		    cbp->cb_buf.b_blkno, cbp->cb_buf.b_data,
    867 		    cbp->cb_buf.b_bcount);
    868 #endif
    869 
    870 	return (cbp);
    871 }
    872 
    873 static void
    874 ccdintr(struct ccd_softc *cs, struct buf *bp)
    875 {
    876 
    877 #ifdef DEBUG
    878 	if (ccddebug & CCDB_FOLLOW)
    879 		printf("ccdintr(%p, %p)\n", cs, bp);
    880 #endif
    881 	/*
    882 	 * Request is done for better or worse, wakeup the top half.
    883 	 */
    884 	if (bp->b_flags & B_ERROR)
    885 		bp->b_resid = bp->b_bcount;
    886 	disk_unbusy(&cs->sc_dkdev, (bp->b_bcount - bp->b_resid),
    887 	    (bp->b_flags & B_READ));
    888 	biodone(bp);
    889 }
    890 
    891 /*
    892  * Called at interrupt time.
    893  * Mark the component as done and if all components are done,
    894  * take a ccd interrupt.
    895  */
    896 static void
    897 ccdiodone(struct buf *vbp)
    898 {
    899 	struct ccdbuf *cbp = (struct ccdbuf *) vbp;
    900 	struct buf *bp = cbp->cb_obp;
    901 	struct ccd_softc *cs = cbp->cb_sc;
    902 	int count, s;
    903 
    904 	s = splbio();
    905 #ifdef DEBUG
    906 	if (ccddebug & CCDB_FOLLOW)
    907 		printf("ccdiodone(%p)\n", cbp);
    908 	if (ccddebug & CCDB_IO) {
    909 		printf("ccdiodone: bp %p bcount %d resid %d\n",
    910 		       bp, bp->b_bcount, bp->b_resid);
    911 		printf(" dev 0x%x(u%d), cbp %p bn %" PRId64 " addr %p"
    912 		       " bcnt %d\n",
    913 		       cbp->cb_buf.b_dev, cbp->cb_comp, cbp,
    914 		       cbp->cb_buf.b_blkno, cbp->cb_buf.b_data,
    915 		       cbp->cb_buf.b_bcount);
    916 	}
    917 #endif
    918 
    919 	if (cbp->cb_buf.b_flags & B_ERROR) {
    920 		bp->b_flags |= B_ERROR;
    921 		bp->b_error = cbp->cb_buf.b_error ?
    922 		    cbp->cb_buf.b_error : EIO;
    923 
    924 		printf("%s: error %d on component %d\n",
    925 		       cs->sc_xname, bp->b_error, cbp->cb_comp);
    926 	}
    927 	count = cbp->cb_buf.b_bcount;
    928 	CCD_PUTBUF(cbp);
    929 
    930 	/*
    931 	 * If all done, "interrupt".
    932 	 */
    933 	bp->b_resid -= count;
    934 	if (bp->b_resid < 0)
    935 		panic("ccdiodone: count");
    936 	if (bp->b_resid == 0)
    937 		ccdintr(cs, bp);
    938 	splx(s);
    939 }
    940 
    941 /* ARGSUSED */
    942 static int
    943 ccdread(dev_t dev, struct uio *uio, int flags)
    944 {
    945 	int unit = ccdunit(dev);
    946 	struct ccd_softc *cs;
    947 
    948 #ifdef DEBUG
    949 	if (ccddebug & CCDB_FOLLOW)
    950 		printf("ccdread(0x%x, %p)\n", dev, uio);
    951 #endif
    952 	if (unit >= numccd)
    953 		return (ENXIO);
    954 	cs = &ccd_softc[unit];
    955 
    956 	if ((cs->sc_flags & CCDF_INITED) == 0)
    957 		return (ENXIO);
    958 
    959 	return (physio(ccdstrategy, NULL, dev, B_READ, minphys, uio));
    960 }
    961 
    962 /* ARGSUSED */
    963 static int
    964 ccdwrite(dev_t dev, struct uio *uio, int flags)
    965 {
    966 	int unit = ccdunit(dev);
    967 	struct ccd_softc *cs;
    968 
    969 #ifdef DEBUG
    970 	if (ccddebug & CCDB_FOLLOW)
    971 		printf("ccdwrite(0x%x, %p)\n", dev, uio);
    972 #endif
    973 	if (unit >= numccd)
    974 		return (ENXIO);
    975 	cs = &ccd_softc[unit];
    976 
    977 	if ((cs->sc_flags & CCDF_INITED) == 0)
    978 		return (ENXIO);
    979 
    980 	return (physio(ccdstrategy, NULL, dev, B_WRITE, minphys, uio));
    981 }
    982 
    983 static int
    984 ccdioctl(dev_t dev, u_long cmd, caddr_t data, int flag, struct lwp *l)
    985 {
    986 	int unit = ccdunit(dev);
    987 	int s, i, j, lookedup = 0, error;
    988 	int part, pmask;
    989 	struct ccd_softc *cs;
    990 	struct ccd_ioctl *ccio = (struct ccd_ioctl *)data;
    991 	struct ucred *uc;
    992 	char **cpp;
    993 	struct vnode **vpp;
    994 #ifdef __HAVE_OLD_DISKLABEL
    995 	struct disklabel newlabel;
    996 #endif
    997 
    998 	if (unit >= numccd)
    999 		return (ENXIO);
   1000 	cs = &ccd_softc[unit];
   1001 
   1002 	uc = (l != NULL) ? l->l_proc->p_ucred : NOCRED;
   1003 
   1004 	/* Must be open for writes for these commands... */
   1005 	switch (cmd) {
   1006 	case CCDIOCSET:
   1007 	case CCDIOCCLR:
   1008 	case DIOCSDINFO:
   1009 	case DIOCWDINFO:
   1010 #ifdef __HAVE_OLD_DISKLABEL
   1011 	case ODIOCSDINFO:
   1012 	case ODIOCWDINFO:
   1013 #endif
   1014 	case DIOCKLABEL:
   1015 	case DIOCWLABEL:
   1016 		if ((flag & FWRITE) == 0)
   1017 			return (EBADF);
   1018 	}
   1019 
   1020 	if ((error = lockmgr(&cs->sc_lock, LK_EXCLUSIVE, NULL)) != 0)
   1021 		return (error);
   1022 
   1023 	/* Must be initialized for these... */
   1024 	switch (cmd) {
   1025 	case CCDIOCCLR:
   1026 	case DIOCGDINFO:
   1027 	case DIOCCACHESYNC:
   1028 	case DIOCSDINFO:
   1029 	case DIOCWDINFO:
   1030 	case DIOCGPART:
   1031 	case DIOCWLABEL:
   1032 	case DIOCKLABEL:
   1033 	case DIOCGDEFLABEL:
   1034 #ifdef __HAVE_OLD_DISKLABEL
   1035 	case ODIOCGDINFO:
   1036 	case ODIOCSDINFO:
   1037 	case ODIOCWDINFO:
   1038 	case ODIOCGDEFLABEL:
   1039 #endif
   1040 		if ((cs->sc_flags & CCDF_INITED) == 0) {
   1041 			error = ENXIO;
   1042 			goto out;
   1043 		}
   1044 	}
   1045 
   1046 	switch (cmd) {
   1047 	case CCDIOCSET:
   1048 		if (cs->sc_flags & CCDF_INITED) {
   1049 			error = EBUSY;
   1050 			goto out;
   1051 		}
   1052 
   1053 		/* Validate the flags. */
   1054 		if ((ccio->ccio_flags & CCDF_USERMASK) != ccio->ccio_flags) {
   1055 			error = EINVAL;
   1056 			goto out;
   1057 		}
   1058 
   1059 		if (ccio->ccio_ndisks > CCD_MAXNDISKS) {
   1060 			error = EINVAL;
   1061 			goto out;
   1062 		}
   1063 
   1064 		/* Fill in some important bits. */
   1065 		cs->sc_ileave = ccio->ccio_ileave;
   1066 		cs->sc_nccdisks = ccio->ccio_ndisks;
   1067 		cs->sc_flags = ccio->ccio_flags & CCDF_USERMASK;
   1068 
   1069 		/*
   1070 		 * Allocate space for and copy in the array of
   1071 		 * componet pathnames and device numbers.
   1072 		 */
   1073 		cpp = malloc(ccio->ccio_ndisks * sizeof(char *),
   1074 		    M_DEVBUF, M_WAITOK);
   1075 		vpp = malloc(ccio->ccio_ndisks * sizeof(struct vnode *),
   1076 		    M_DEVBUF, M_WAITOK);
   1077 
   1078 		error = copyin(ccio->ccio_disks, cpp,
   1079 		    ccio->ccio_ndisks * sizeof(char **));
   1080 		if (error) {
   1081 			free(vpp, M_DEVBUF);
   1082 			free(cpp, M_DEVBUF);
   1083 			goto out;
   1084 		}
   1085 
   1086 #ifdef DEBUG
   1087 		if (ccddebug & CCDB_INIT)
   1088 			for (i = 0; i < ccio->ccio_ndisks; ++i)
   1089 				printf("ccdioctl: component %d: %p\n",
   1090 				    i, cpp[i]);
   1091 #endif
   1092 
   1093 		for (i = 0; i < ccio->ccio_ndisks; ++i) {
   1094 #ifdef DEBUG
   1095 			if (ccddebug & CCDB_INIT)
   1096 				printf("ccdioctl: lookedup = %d\n", lookedup);
   1097 #endif
   1098 			if ((error = ccdlookup(cpp[i], l, &vpp[i])) != 0) {
   1099 				for (j = 0; j < lookedup; ++j)
   1100 					(void)vn_close(vpp[j], FREAD|FWRITE,
   1101 					    uc, l);
   1102 				free(vpp, M_DEVBUF);
   1103 				free(cpp, M_DEVBUF);
   1104 				goto out;
   1105 			}
   1106 			++lookedup;
   1107 		}
   1108 
   1109 		/*
   1110 		 * Initialize the ccd.  Fills in the softc for us.
   1111 		 */
   1112 		if ((error = ccdinit(cs, cpp, vpp, l)) != 0) {
   1113 			for (j = 0; j < lookedup; ++j)
   1114 				(void)vn_close(vpp[j], FREAD|FWRITE,
   1115 				    uc, l);
   1116 			free(vpp, M_DEVBUF);
   1117 			free(cpp, M_DEVBUF);
   1118 			goto out;
   1119 		}
   1120 
   1121 		/* We can free the temporary variables now. */
   1122 		free(vpp, M_DEVBUF);
   1123 		free(cpp, M_DEVBUF);
   1124 
   1125 		/*
   1126 		 * The ccd has been successfully initialized, so
   1127 		 * we can place it into the array.  Don't try to
   1128 		 * read the disklabel until the disk has been attached,
   1129 		 * because space for the disklabel is allocated
   1130 		 * in disk_attach();
   1131 		 */
   1132 		ccio->ccio_unit = unit;
   1133 		ccio->ccio_size = cs->sc_size;
   1134 
   1135 		bufq_alloc(&cs->sc_bufq, "fcfs", 0);
   1136 
   1137 		/* Attach the disk. */
   1138 		pseudo_disk_attach(&cs->sc_dkdev);
   1139 
   1140 		/* Try and read the disklabel. */
   1141 		ccdgetdisklabel(dev);
   1142 		break;
   1143 
   1144 	case CCDIOCCLR:
   1145 		/*
   1146 		 * Don't unconfigure if any other partitions are open
   1147 		 * or if both the character and block flavors of this
   1148 		 * partition are open.
   1149 		 */
   1150 		part = DISKPART(dev);
   1151 		pmask = (1 << part);
   1152 		if ((cs->sc_dkdev.dk_openmask & ~pmask) ||
   1153 		    ((cs->sc_dkdev.dk_bopenmask & pmask) &&
   1154 		    (cs->sc_dkdev.dk_copenmask & pmask))) {
   1155 			error = EBUSY;
   1156 			goto out;
   1157 		}
   1158 
   1159 		/* Kill off any queued buffers. */
   1160 		s = splbio();
   1161 		bufq_drain(cs->sc_bufq);
   1162 		splx(s);
   1163 
   1164 		bufq_free(cs->sc_bufq);
   1165 
   1166 		/*
   1167 		 * Free ccd_softc information and clear entry.
   1168 		 */
   1169 
   1170 		/* Close the components and free their pathnames. */
   1171 		for (i = 0; i < cs->sc_nccdisks; ++i) {
   1172 			/*
   1173 			 * XXX: this close could potentially fail and
   1174 			 * cause Bad Things.  Maybe we need to force
   1175 			 * the close to happen?
   1176 			 */
   1177 #ifdef DEBUG
   1178 			if (ccddebug & CCDB_VNODE)
   1179 				vprint("CCDIOCCLR: vnode info",
   1180 				    cs->sc_cinfo[i].ci_vp);
   1181 #endif
   1182 			(void)vn_close(cs->sc_cinfo[i].ci_vp, FREAD|FWRITE,
   1183 			    uc, l);
   1184 			free(cs->sc_cinfo[i].ci_path, M_DEVBUF);
   1185 		}
   1186 
   1187 		/* Free interleave index. */
   1188 		for (i = 0; cs->sc_itable[i].ii_ndisk; ++i)
   1189 			free(cs->sc_itable[i].ii_index, M_DEVBUF);
   1190 
   1191 		/* Free component info and interleave table. */
   1192 		free(cs->sc_cinfo, M_DEVBUF);
   1193 		free(cs->sc_itable, M_DEVBUF);
   1194 		cs->sc_flags &= ~(CCDF_INITED|CCDF_VLABEL);
   1195 
   1196 		/* Detatch the disk. */
   1197 		pseudo_disk_detach(&cs->sc_dkdev);
   1198 		break;
   1199 
   1200 	case DIOCGDINFO:
   1201 		*(struct disklabel *)data = *(cs->sc_dkdev.dk_label);
   1202 		break;
   1203 #ifdef __HAVE_OLD_DISKLABEL
   1204 	case ODIOCGDINFO:
   1205 		newlabel = *(cs->sc_dkdev.dk_label);
   1206 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
   1207 			return ENOTTY;
   1208 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
   1209 		break;
   1210 #endif
   1211 
   1212 	case DIOCGPART:
   1213 		((struct partinfo *)data)->disklab = cs->sc_dkdev.dk_label;
   1214 		((struct partinfo *)data)->part =
   1215 		    &cs->sc_dkdev.dk_label->d_partitions[DISKPART(dev)];
   1216 		break;
   1217 
   1218 	case DIOCCACHESYNC:
   1219 		/*
   1220 		 * XXX Do we really need to care about having a writable
   1221 		 * file descriptor here?
   1222 		 */
   1223 		if ((flag & FWRITE) == 0)
   1224 			return (EBADF);
   1225 
   1226 		/*
   1227 		 * We pass this call down to all components and report
   1228 		 * the first error we encounter.
   1229 		 */
   1230 		for (error = 0, i = 0; i < cs->sc_nccdisks; i++) {
   1231 			j = VOP_IOCTL(cs->sc_cinfo[i].ci_vp, cmd, data,
   1232 				      flag, uc, l);
   1233 			if (j != 0 && error == 0)
   1234 				error = j;
   1235 		}
   1236 		break;
   1237 
   1238 	case DIOCWDINFO:
   1239 	case DIOCSDINFO:
   1240 #ifdef __HAVE_OLD_DISKLABEL
   1241 	case ODIOCWDINFO:
   1242 	case ODIOCSDINFO:
   1243 #endif
   1244 	{
   1245 		struct disklabel *lp;
   1246 #ifdef __HAVE_OLD_DISKLABEL
   1247 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
   1248 			memset(&newlabel, 0, sizeof newlabel);
   1249 			memcpy(&newlabel, data, sizeof (struct olddisklabel));
   1250 			lp = &newlabel;
   1251 		} else
   1252 #endif
   1253 		lp = (struct disklabel *)data;
   1254 
   1255 		cs->sc_flags |= CCDF_LABELLING;
   1256 
   1257 		error = setdisklabel(cs->sc_dkdev.dk_label,
   1258 		    lp, 0, cs->sc_dkdev.dk_cpulabel);
   1259 		if (error == 0) {
   1260 			if (cmd == DIOCWDINFO
   1261 #ifdef __HAVE_OLD_DISKLABEL
   1262 			    || cmd == ODIOCWDINFO
   1263 #endif
   1264 			   )
   1265 				error = writedisklabel(CCDLABELDEV(dev),
   1266 				    ccdstrategy, cs->sc_dkdev.dk_label,
   1267 				    cs->sc_dkdev.dk_cpulabel);
   1268 		}
   1269 
   1270 		cs->sc_flags &= ~CCDF_LABELLING;
   1271 		break;
   1272 	}
   1273 
   1274 	case DIOCKLABEL:
   1275 		if (*(int *)data != 0)
   1276 			cs->sc_flags |= CCDF_KLABEL;
   1277 		else
   1278 			cs->sc_flags &= ~CCDF_KLABEL;
   1279 		break;
   1280 
   1281 	case DIOCWLABEL:
   1282 		if (*(int *)data != 0)
   1283 			cs->sc_flags |= CCDF_WLABEL;
   1284 		else
   1285 			cs->sc_flags &= ~CCDF_WLABEL;
   1286 		break;
   1287 
   1288 	case DIOCGDEFLABEL:
   1289 		ccdgetdefaultlabel(cs, (struct disklabel *)data);
   1290 		break;
   1291 
   1292 #ifdef __HAVE_OLD_DISKLABEL
   1293 	case ODIOCGDEFLABEL:
   1294 		ccdgetdefaultlabel(cs, &newlabel);
   1295 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
   1296 			return ENOTTY;
   1297 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
   1298 		break;
   1299 #endif
   1300 
   1301 	default:
   1302 		error = ENOTTY;
   1303 	}
   1304 
   1305  out:
   1306 	(void) lockmgr(&cs->sc_lock, LK_RELEASE, NULL);
   1307 	return (error);
   1308 }
   1309 
   1310 static int
   1311 ccdsize(dev_t dev)
   1312 {
   1313 	struct ccd_softc *cs;
   1314 	struct disklabel *lp;
   1315 	int part, unit, omask, size;
   1316 
   1317 	unit = ccdunit(dev);
   1318 	if (unit >= numccd)
   1319 		return (-1);
   1320 	cs = &ccd_softc[unit];
   1321 
   1322 	if ((cs->sc_flags & CCDF_INITED) == 0)
   1323 		return (-1);
   1324 
   1325 	part = DISKPART(dev);
   1326 	omask = cs->sc_dkdev.dk_openmask & (1 << part);
   1327 	lp = cs->sc_dkdev.dk_label;
   1328 
   1329 	if (omask == 0 && ccdopen(dev, 0, S_IFBLK, curlwp))
   1330 		return (-1);
   1331 
   1332 	if (lp->d_partitions[part].p_fstype != FS_SWAP)
   1333 		size = -1;
   1334 	else
   1335 		size = lp->d_partitions[part].p_size *
   1336 		    (lp->d_secsize / DEV_BSIZE);
   1337 
   1338 	if (omask == 0 && ccdclose(dev, 0, S_IFBLK, curlwp))
   1339 		return (-1);
   1340 
   1341 	return (size);
   1342 }
   1343 
   1344 static int
   1345 ccddump(dev_t dev, daddr_t blkno, caddr_t va, size_t size)
   1346 {
   1347 
   1348 	/* Not implemented. */
   1349 	return ENXIO;
   1350 }
   1351 
   1352 /*
   1353  * Lookup the provided name in the filesystem.  If the file exists,
   1354  * is a valid block device, and isn't being used by anyone else,
   1355  * set *vpp to the file's vnode.
   1356  */
   1357 static int
   1358 ccdlookup(char *path, struct lwp *l, struct vnode **vpp /* result */)
   1359 {
   1360 	struct nameidata nd;
   1361 	struct vnode *vp;
   1362 	struct vattr va;
   1363 	struct proc *p;
   1364 	int error;
   1365 
   1366 	p = l->l_proc;
   1367 
   1368 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, path, l);
   1369 	if ((error = vn_open(&nd, FREAD|FWRITE, 0)) != 0) {
   1370 #ifdef DEBUG
   1371 		if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
   1372 			printf("ccdlookup: vn_open error = %d\n", error);
   1373 #endif
   1374 		return (error);
   1375 	}
   1376 	vp = nd.ni_vp;
   1377 
   1378 	if (vp->v_usecount > 1) {
   1379 		VOP_UNLOCK(vp, 0);
   1380 		(void)vn_close(vp, FREAD|FWRITE, p->p_ucred, l);
   1381 		return (EBUSY);
   1382 	}
   1383 
   1384 	if ((error = VOP_GETATTR(vp, &va, p->p_ucred, l)) != 0) {
   1385 #ifdef DEBUG
   1386 		if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
   1387 			printf("ccdlookup: getattr error = %d\n", error);
   1388 #endif
   1389 		VOP_UNLOCK(vp, 0);
   1390 		(void)vn_close(vp, FREAD|FWRITE, p->p_ucred, l);
   1391 		return (error);
   1392 	}
   1393 
   1394 	/* XXX: eventually we should handle VREG, too. */
   1395 	if (va.va_type != VBLK) {
   1396 		VOP_UNLOCK(vp, 0);
   1397 		(void)vn_close(vp, FREAD|FWRITE, p->p_ucred, l);
   1398 		return (ENOTBLK);
   1399 	}
   1400 
   1401 #ifdef DEBUG
   1402 	if (ccddebug & CCDB_VNODE)
   1403 		vprint("ccdlookup: vnode info", vp);
   1404 #endif
   1405 
   1406 	VOP_UNLOCK(vp, 0);
   1407 	*vpp = vp;
   1408 	return (0);
   1409 }
   1410 
   1411 static void
   1412 ccdgetdefaultlabel(struct ccd_softc *cs, struct disklabel *lp)
   1413 {
   1414 	struct ccdgeom *ccg = &cs->sc_geom;
   1415 
   1416 	memset(lp, 0, sizeof(*lp));
   1417 
   1418 	lp->d_secperunit = cs->sc_size;
   1419 	lp->d_secsize = ccg->ccg_secsize;
   1420 	lp->d_nsectors = ccg->ccg_nsectors;
   1421 	lp->d_ntracks = ccg->ccg_ntracks;
   1422 	lp->d_ncylinders = ccg->ccg_ncylinders;
   1423 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
   1424 
   1425 	strncpy(lp->d_typename, "ccd", sizeof(lp->d_typename));
   1426 	lp->d_type = DTYPE_CCD;
   1427 	strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
   1428 	lp->d_rpm = 3600;
   1429 	lp->d_interleave = 1;
   1430 	lp->d_flags = 0;
   1431 
   1432 	lp->d_partitions[RAW_PART].p_offset = 0;
   1433 	lp->d_partitions[RAW_PART].p_size = cs->sc_size;
   1434 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
   1435 	lp->d_npartitions = RAW_PART + 1;
   1436 
   1437 	lp->d_magic = DISKMAGIC;
   1438 	lp->d_magic2 = DISKMAGIC;
   1439 	lp->d_checksum = dkcksum(cs->sc_dkdev.dk_label);
   1440 }
   1441 
   1442 /*
   1443  * Read the disklabel from the ccd.  If one is not present, fake one
   1444  * up.
   1445  */
   1446 static void
   1447 ccdgetdisklabel(dev_t dev)
   1448 {
   1449 	int unit = ccdunit(dev);
   1450 	struct ccd_softc *cs = &ccd_softc[unit];
   1451 	const char *errstring;
   1452 	struct disklabel *lp = cs->sc_dkdev.dk_label;
   1453 	struct cpu_disklabel *clp = cs->sc_dkdev.dk_cpulabel;
   1454 
   1455 	memset(clp, 0, sizeof(*clp));
   1456 
   1457 	ccdgetdefaultlabel(cs, lp);
   1458 
   1459 	/*
   1460 	 * Call the generic disklabel extraction routine.
   1461 	 */
   1462 	if ((cs->sc_flags & CCDF_NOLABEL) != 0)
   1463 		errstring = "CCDF_NOLABEL set; ignoring on-disk label";
   1464 	else
   1465 		errstring = readdisklabel(CCDLABELDEV(dev), ccdstrategy,
   1466 		    cs->sc_dkdev.dk_label, cs->sc_dkdev.dk_cpulabel);
   1467 	if (errstring)
   1468 		ccdmakedisklabel(cs);
   1469 	else {
   1470 		int i;
   1471 		struct partition *pp;
   1472 
   1473 		/*
   1474 		 * Sanity check whether the found disklabel is valid.
   1475 		 *
   1476 		 * This is necessary since total size of ccd may vary
   1477 		 * when an interleave is changed even though exactly
   1478 		 * same componets are used, and old disklabel may used
   1479 		 * if that is found.
   1480 		 */
   1481 		if (lp->d_secperunit != cs->sc_size)
   1482 			printf("WARNING: %s: "
   1483 			    "total sector size in disklabel (%d) != "
   1484 			    "the size of ccd (%lu)\n", cs->sc_xname,
   1485 			    lp->d_secperunit, (u_long)cs->sc_size);
   1486 		for (i = 0; i < lp->d_npartitions; i++) {
   1487 			pp = &lp->d_partitions[i];
   1488 			if (pp->p_offset + pp->p_size > cs->sc_size)
   1489 				printf("WARNING: %s: end of partition `%c' "
   1490 				    "exceeds the size of ccd (%lu)\n",
   1491 				    cs->sc_xname, 'a' + i, (u_long)cs->sc_size);
   1492 		}
   1493 	}
   1494 
   1495 #ifdef DEBUG
   1496 	/* It's actually extremely common to have unlabeled ccds. */
   1497 	if (ccddebug & CCDB_LABEL)
   1498 		if (errstring != NULL)
   1499 			printf("%s: %s\n", cs->sc_xname, errstring);
   1500 #endif
   1501 
   1502 	/* In-core label now valid. */
   1503 	cs->sc_flags |= CCDF_VLABEL;
   1504 }
   1505 
   1506 /*
   1507  * Take care of things one might want to take care of in the event
   1508  * that a disklabel isn't present.
   1509  */
   1510 static void
   1511 ccdmakedisklabel(struct ccd_softc *cs)
   1512 {
   1513 	struct disklabel *lp = cs->sc_dkdev.dk_label;
   1514 
   1515 	/*
   1516 	 * For historical reasons, if there's no disklabel present
   1517 	 * the raw partition must be marked FS_BSDFFS.
   1518 	 */
   1519 	lp->d_partitions[RAW_PART].p_fstype = FS_BSDFFS;
   1520 
   1521 	strncpy(lp->d_packname, "default label", sizeof(lp->d_packname));
   1522 
   1523 	lp->d_checksum = dkcksum(lp);
   1524 }
   1525 
   1526 #ifdef DEBUG
   1527 static void
   1528 printiinfo(struct ccdiinfo *ii)
   1529 {
   1530 	int ix, i;
   1531 
   1532 	for (ix = 0; ii->ii_ndisk; ix++, ii++) {
   1533 		printf(" itab[%d]: #dk %d sblk %" PRId64 " soff %" PRId64,
   1534 		    ix, ii->ii_ndisk, ii->ii_startblk, ii->ii_startoff);
   1535 		for (i = 0; i < ii->ii_ndisk; i++)
   1536 			printf(" %d", ii->ii_index[i]);
   1537 		printf("\n");
   1538 	}
   1539 }
   1540 #endif
   1541