Home | History | Annotate | Line # | Download | only in dev
ccd.c revision 1.158
      1 /*	$NetBSD: ccd.c,v 1.158 2014/12/30 20:29:42 christos Exp $	*/
      2 
      3 /*-
      4  * Copyright (c) 1996, 1997, 1998, 1999, 2007, 2009 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, and by Andrew Doran.
      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 $Hdr: cd.c 1.6 90/11/28$
     66  *
     67  *	@(#)cd.c	8.2 (Berkeley) 11/16/93
     68  */
     69 
     70 /*
     71  * "Concatenated" disk driver.
     72  *
     73  * Notes on concurrency:
     74  *
     75  * => sc_dvlock serializes access to the device nodes, excluding block I/O.
     76  *
     77  * => sc_iolock serializes access to (sc_flags & CCDF_INITED), disk stats,
     78  *    sc_stop, sc_bufq and b_resid from master buffers.
     79  *
     80  * => a combination of CCDF_INITED, sc_inflight, and sc_iolock is used to
     81  *    serialize I/O and configuration changes.
     82  *
     83  * => the in-core disk label does not change while the device is open.
     84  *
     85  * On memory consumption: ccd fans out I/O requests and so needs to
     86  * allocate memory.  If the system is desperately low on memory, we
     87  * single thread I/O.
     88  */
     89 
     90 #include <sys/cdefs.h>
     91 __KERNEL_RCSID(0, "$NetBSD: ccd.c,v 1.158 2014/12/30 20:29:42 christos Exp $");
     92 
     93 #if defined(_KERNEL_OPT)
     94 #include "opt_compat_netbsd.h"
     95 #endif
     96 
     97 #include <sys/param.h>
     98 #include <sys/systm.h>
     99 #include <sys/kernel.h>
    100 #include <sys/proc.h>
    101 #include <sys/errno.h>
    102 #include <sys/buf.h>
    103 #include <sys/kmem.h>
    104 #include <sys/pool.h>
    105 #include <sys/module.h>
    106 #include <sys/namei.h>
    107 #include <sys/stat.h>
    108 #include <sys/ioctl.h>
    109 #include <sys/disklabel.h>
    110 #include <sys/device.h>
    111 #include <sys/disk.h>
    112 #include <sys/syslog.h>
    113 #include <sys/fcntl.h>
    114 #include <sys/vnode.h>
    115 #include <sys/conf.h>
    116 #include <sys/mutex.h>
    117 #include <sys/queue.h>
    118 #include <sys/kauth.h>
    119 #include <sys/kthread.h>
    120 #include <sys/bufq.h>
    121 #include <sys/sysctl.h>
    122 
    123 #include <uvm/uvm_extern.h>
    124 
    125 #include <dev/ccdvar.h>
    126 #include <dev/dkvar.h>
    127 
    128 #include <miscfs/specfs/specdev.h> /* for v_rdev */
    129 
    130 #if defined(CCDDEBUG) && !defined(DEBUG)
    131 #define DEBUG
    132 #endif
    133 
    134 #ifdef DEBUG
    135 #define CCDB_FOLLOW	0x01
    136 #define CCDB_INIT	0x02
    137 #define CCDB_IO		0x04
    138 #define CCDB_LABEL	0x08
    139 #define CCDB_VNODE	0x10
    140 int ccddebug = 0x00;
    141 #endif
    142 
    143 #define	ccdunit(x)	DISKUNIT(x)
    144 
    145 struct ccdbuf {
    146 	struct buf	cb_buf;		/* new I/O buf */
    147 	struct buf	*cb_obp;	/* ptr. to original I/O buf */
    148 	struct ccd_softc *cb_sc;	/* pointer to ccd softc */
    149 	int		cb_comp;	/* target component */
    150 	SIMPLEQ_ENTRY(ccdbuf) cb_q;	/* fifo of component buffers */
    151 };
    152 
    153 /* component buffer pool */
    154 static pool_cache_t ccd_cache;
    155 
    156 #define	CCD_GETBUF()		pool_cache_get(ccd_cache, PR_WAITOK)
    157 #define	CCD_PUTBUF(cbp)		pool_cache_put(ccd_cache, cbp)
    158 
    159 #define CCDLABELDEV(dev)	\
    160 	(MAKEDISKDEV(major((dev)), ccdunit((dev)), RAW_PART))
    161 
    162 /* called by main() at boot time */
    163 void	ccdattach(int);
    164 
    165 /* called by biodone() at interrupt time */
    166 static void	ccdiodone(struct buf *);
    167 
    168 static void	ccdinterleave(struct ccd_softc *);
    169 static int	ccdinit(struct ccd_softc *, char **, struct vnode **,
    170 		    struct lwp *);
    171 static struct ccdbuf *ccdbuffer(struct ccd_softc *, struct buf *,
    172 		    daddr_t, void *, long);
    173 static void	ccdgetdefaultlabel(struct ccd_softc *, struct disklabel *);
    174 static void	ccdgetdisklabel(dev_t);
    175 static void	ccdmakedisklabel(struct ccd_softc *);
    176 static void	ccdstart(struct ccd_softc *);
    177 static void	ccdthread(void *);
    178 
    179 static dev_type_open(ccdopen);
    180 static dev_type_close(ccdclose);
    181 static dev_type_read(ccdread);
    182 static dev_type_write(ccdwrite);
    183 static dev_type_ioctl(ccdioctl);
    184 static dev_type_strategy(ccdstrategy);
    185 static dev_type_size(ccdsize);
    186 
    187 const struct bdevsw ccd_bdevsw = {
    188 	.d_open = ccdopen,
    189 	.d_close = ccdclose,
    190 	.d_strategy = ccdstrategy,
    191 	.d_ioctl = ccdioctl,
    192 	.d_dump = nodump,
    193 	.d_psize = ccdsize,
    194 	.d_discard = nodiscard,
    195 	.d_flag = D_DISK | D_MPSAFE
    196 };
    197 
    198 const struct cdevsw ccd_cdevsw = {
    199 	.d_open = ccdopen,
    200 	.d_close = ccdclose,
    201 	.d_read = ccdread,
    202 	.d_write = ccdwrite,
    203 	.d_ioctl = ccdioctl,
    204 	.d_stop = nostop,
    205 	.d_tty = notty,
    206 	.d_poll = nopoll,
    207 	.d_mmap = nommap,
    208 	.d_kqfilter = nokqfilter,
    209 	.d_discard = nodiscard,
    210 	.d_flag = D_DISK | D_MPSAFE
    211 };
    212 
    213 #ifdef DEBUG
    214 static	void printiinfo(struct ccdiinfo *);
    215 #endif
    216 
    217 static LIST_HEAD(, ccd_softc) ccds = LIST_HEAD_INITIALIZER(ccds);
    218 static kmutex_t ccd_lock;
    219 static size_t ccd_nactive = 0;
    220 
    221 static struct ccd_softc *
    222 ccdcreate(int unit) {
    223 	struct ccd_softc *sc = kmem_zalloc(sizeof(*sc), KM_SLEEP);
    224 	if (sc == NULL) {
    225 #ifdef DIAGNOSTIC
    226 		printf("%s: out of memory\n", __func__);
    227 #endif
    228 		return NULL;
    229 	}
    230 	/* Initialize per-softc structures. */
    231 	snprintf(sc->sc_xname, sizeof(sc->sc_xname), "ccd%d", unit);
    232 	mutex_init(&sc->sc_dvlock, MUTEX_DEFAULT, IPL_NONE);
    233 	sc->sc_iolock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
    234 	cv_init(&sc->sc_stop, "ccdstop");
    235 	cv_init(&sc->sc_push, "ccdthr");
    236 	disk_init(&sc->sc_dkdev, sc->sc_xname, NULL); /* XXX */
    237 	return sc;
    238 }
    239 
    240 static void
    241 ccddestroy(struct ccd_softc *sc) {
    242 	mutex_obj_free(sc->sc_iolock);
    243 	mutex_exit(&sc->sc_dvlock);
    244 	mutex_destroy(&sc->sc_dvlock);
    245 	cv_destroy(&sc->sc_stop);
    246 	cv_destroy(&sc->sc_push);
    247 	disk_destroy(&sc->sc_dkdev);
    248 	kmem_free(sc, sizeof(*sc));
    249 }
    250 
    251 static struct ccd_softc *
    252 ccdget(int unit, int make) {
    253 	struct ccd_softc *sc;
    254 	if (unit < 0) {
    255 #ifdef DIAGNOSTIC
    256 		panic("%s: unit %d!", __func__, unit);
    257 #endif
    258 		return NULL;
    259 	}
    260 	mutex_enter(&ccd_lock);
    261 	LIST_FOREACH(sc, &ccds, sc_link) {
    262 		if (sc->sc_unit == unit) {
    263 			mutex_exit(&ccd_lock);
    264 			return sc;
    265 		}
    266 	}
    267 	mutex_exit(&ccd_lock);
    268 	if (!make)
    269 		return NULL;
    270 	if ((sc = ccdcreate(unit)) == NULL)
    271 		return NULL;
    272 	mutex_enter(&ccd_lock);
    273 	LIST_INSERT_HEAD(&ccds, sc, sc_link);
    274 	ccd_nactive++;
    275 	mutex_exit(&ccd_lock);
    276 	return sc;
    277 }
    278 
    279 static void
    280 ccdput(struct ccd_softc *sc) {
    281 	mutex_enter(&ccd_lock);
    282 	LIST_REMOVE(sc, sc_link);
    283 	ccd_nactive--;
    284 	mutex_exit(&ccd_lock);
    285 	ccddestroy(sc);
    286 }
    287 
    288 /*
    289  * Called by main() during pseudo-device attachment.  All we need
    290  * to do is allocate enough space for devices to be configured later.
    291  */
    292 void
    293 ccdattach(int num)
    294 {
    295 	mutex_init(&ccd_lock, MUTEX_DEFAULT, IPL_NONE);
    296 
    297 	/* Initialize the component buffer pool. */
    298 	ccd_cache = pool_cache_init(sizeof(struct ccdbuf), 0,
    299 	    0, 0, "ccdbuf", NULL, IPL_BIO, NULL, NULL, NULL);
    300 }
    301 
    302 static int
    303 ccdinit(struct ccd_softc *cs, char **cpaths, struct vnode **vpp,
    304     struct lwp *l)
    305 {
    306 	struct ccdcinfo *ci = NULL;
    307 	int ix;
    308 	struct ccdgeom *ccg = &cs->sc_geom;
    309 	char *tmppath;
    310 	int error, path_alloced;
    311 	uint64_t psize, minsize;
    312 	unsigned secsize, maxsecsize;
    313 	struct disk_geom *dg;
    314 
    315 #ifdef DEBUG
    316 	if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
    317 		printf("%s: ccdinit\n", cs->sc_xname);
    318 #endif
    319 
    320 	/* Allocate space for the component info. */
    321 	cs->sc_cinfo = kmem_alloc(cs->sc_nccdisks * sizeof(*cs->sc_cinfo),
    322 	    KM_SLEEP);
    323 	tmppath = kmem_alloc(MAXPATHLEN, KM_SLEEP);
    324 
    325 	cs->sc_size = 0;
    326 
    327 	/*
    328 	 * Verify that each component piece exists and record
    329 	 * relevant information about it.
    330 	 */
    331 	maxsecsize = 0;
    332 	minsize = 0;
    333 	for (ix = 0, path_alloced = 0; ix < cs->sc_nccdisks; ix++) {
    334 		ci = &cs->sc_cinfo[ix];
    335 		ci->ci_vp = vpp[ix];
    336 
    337 		/*
    338 		 * Copy in the pathname of the component.
    339 		 */
    340 		memset(tmppath, 0, MAXPATHLEN);	/* sanity */
    341 		error = copyinstr(cpaths[ix], tmppath,
    342 		    MAXPATHLEN, &ci->ci_pathlen);
    343 		if (ci->ci_pathlen == 0)
    344 			error = EINVAL;
    345 		if (error) {
    346 #ifdef DEBUG
    347 			if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
    348 				printf("%s: can't copy path, error = %d\n",
    349 				    cs->sc_xname, error);
    350 #endif
    351 			goto out;
    352 		}
    353 		ci->ci_path = kmem_alloc(ci->ci_pathlen, KM_SLEEP);
    354 		memcpy(ci->ci_path, tmppath, ci->ci_pathlen);
    355 		path_alloced++;
    356 
    357 		/*
    358 		 * XXX: Cache the component's dev_t.
    359 		 */
    360 		ci->ci_dev = vpp[ix]->v_rdev;
    361 
    362 		/*
    363 		 * Get partition information for the component.
    364 		 */
    365 		error = getdisksize(vpp[ix], &psize, &secsize);
    366 		if (error) {
    367 #ifdef DEBUG
    368 			if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
    369 				 printf("%s: %s: disksize failed, error = %d\n",
    370 				     cs->sc_xname, ci->ci_path, error);
    371 #endif
    372 			goto out;
    373 		}
    374 
    375 		/*
    376 		 * Calculate the size, truncating to an interleave
    377 		 * boundary if necessary.
    378 		 */
    379 		maxsecsize = secsize > maxsecsize ? secsize : maxsecsize;
    380 		if (cs->sc_ileave > 1)
    381 			psize -= psize % cs->sc_ileave;
    382 
    383 		if (psize == 0) {
    384 #ifdef DEBUG
    385 			if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
    386 				printf("%s: %s: size == 0\n",
    387 				    cs->sc_xname, ci->ci_path);
    388 #endif
    389 			error = ENODEV;
    390 			goto out;
    391 		}
    392 
    393 		if (minsize == 0 || psize < minsize)
    394 			minsize = psize;
    395 		ci->ci_size = psize;
    396 		cs->sc_size += psize;
    397 	}
    398 
    399 	/*
    400 	 * Don't allow the interleave to be smaller than
    401 	 * the biggest component sector.
    402 	 */
    403 	if ((cs->sc_ileave > 0) &&
    404 	    (cs->sc_ileave < (maxsecsize / DEV_BSIZE))) {
    405 #ifdef DEBUG
    406 		if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
    407 			printf("%s: interleave must be at least %d\n",
    408 			    cs->sc_xname, (maxsecsize / DEV_BSIZE));
    409 #endif
    410 		error = EINVAL;
    411 		goto out;
    412 	}
    413 
    414 	/*
    415 	 * If uniform interleave is desired set all sizes to that of
    416 	 * the smallest component.
    417 	 */
    418 	if (cs->sc_flags & CCDF_UNIFORM) {
    419 		for (ci = cs->sc_cinfo;
    420 		     ci < &cs->sc_cinfo[cs->sc_nccdisks]; ci++)
    421 			ci->ci_size = minsize;
    422 
    423 		cs->sc_size = cs->sc_nccdisks * minsize;
    424 	}
    425 
    426 	/*
    427 	 * Construct the interleave table.
    428 	 */
    429 	ccdinterleave(cs);
    430 
    431 	/*
    432 	 * Create pseudo-geometry based on 1MB cylinders.  It's
    433 	 * pretty close.
    434 	 */
    435 	ccg->ccg_secsize = DEV_BSIZE;
    436 	ccg->ccg_ntracks = 1;
    437 	ccg->ccg_nsectors = 1024 * (1024 / ccg->ccg_secsize);
    438 	ccg->ccg_ncylinders = cs->sc_size / ccg->ccg_nsectors;
    439 
    440         dg = &cs->sc_dkdev.dk_geom;
    441         memset(dg, 0, sizeof(*dg));
    442 	dg->dg_secperunit = cs->sc_size;
    443 	dg->dg_secsize = ccg->ccg_secsize;
    444 	dg->dg_nsectors = ccg->ccg_nsectors;
    445 	dg->dg_ntracks = ccg->ccg_ntracks;
    446 	dg->dg_ncylinders = ccg->ccg_ncylinders;
    447 
    448 	if (cs->sc_ileave > 0)
    449 	        aprint_normal("%s: Interleaving %d component%s "
    450 	            "(%d block interleave)\n", cs->sc_xname,
    451         	    cs->sc_nccdisks, (cs->sc_nccdisks != 0 ? "s" : ""),
    452         	    cs->sc_ileave);
    453 	else
    454 	        aprint_normal("%s: Concatenating %d component%s\n",
    455 	            cs->sc_xname,
    456         	    cs->sc_nccdisks, (cs->sc_nccdisks != 0 ? "s" : ""));
    457 	for (ix = 0; ix < cs->sc_nccdisks; ix++) {
    458 		ci = &cs->sc_cinfo[ix];
    459 		aprint_normal("%s: %s (%ju blocks)\n", cs->sc_xname,
    460 		    ci->ci_path, (uintmax_t)ci->ci_size);
    461 	}
    462 	aprint_normal("%s: total %ju blocks\n", cs->sc_xname, cs->sc_size);
    463 
    464 	/*
    465 	 * Create thread to handle deferred I/O.
    466 	 */
    467 	cs->sc_zap = false;
    468 	error = kthread_create(PRI_BIO, KTHREAD_MPSAFE, NULL, ccdthread,
    469 	    cs, &cs->sc_thread, "%s", cs->sc_xname);
    470 	if (error) {
    471 		printf("ccdinit: can't create thread: %d\n", error);
    472 		goto out;
    473 	}
    474 
    475 	/*
    476 	 * Only now that everything is set up can we enable the device.
    477 	 */
    478 	mutex_enter(cs->sc_iolock);
    479 	cs->sc_flags |= CCDF_INITED;
    480 	mutex_exit(cs->sc_iolock);
    481 	kmem_free(tmppath, MAXPATHLEN);
    482 	return (0);
    483 
    484  out:
    485 	for (ix = 0; ix < path_alloced; ix++) {
    486 		kmem_free(cs->sc_cinfo[ix].ci_path,
    487 		    cs->sc_cinfo[ix].ci_pathlen);
    488 	}
    489 	kmem_free(cs->sc_cinfo, cs->sc_nccdisks * sizeof(struct ccdcinfo));
    490 	kmem_free(tmppath, MAXPATHLEN);
    491 	return (error);
    492 }
    493 
    494 static void
    495 ccdinterleave(struct ccd_softc *cs)
    496 {
    497 	struct ccdcinfo *ci, *smallci;
    498 	struct ccdiinfo *ii;
    499 	daddr_t bn, lbn;
    500 	int ix;
    501 	u_long size;
    502 
    503 #ifdef DEBUG
    504 	if (ccddebug & CCDB_INIT)
    505 		printf("ccdinterleave(%p): ileave %d\n", cs, cs->sc_ileave);
    506 #endif
    507 	/*
    508 	 * Allocate an interleave table.
    509 	 * Chances are this is too big, but we don't care.
    510 	 */
    511 	size = (cs->sc_nccdisks + 1) * sizeof(struct ccdiinfo);
    512 	cs->sc_itable = kmem_zalloc(size, KM_SLEEP);
    513 
    514 	/*
    515 	 * Trivial case: no interleave (actually interleave of disk size).
    516 	 * Each table entry represents a single component in its entirety.
    517 	 */
    518 	if (cs->sc_ileave == 0) {
    519 		bn = 0;
    520 		ii = cs->sc_itable;
    521 
    522 		for (ix = 0; ix < cs->sc_nccdisks; ix++) {
    523 			/* Allocate space for ii_index. */
    524 			ii->ii_indexsz = sizeof(int);
    525 			ii->ii_index = kmem_alloc(ii->ii_indexsz, KM_SLEEP);
    526 			ii->ii_ndisk = 1;
    527 			ii->ii_startblk = bn;
    528 			ii->ii_startoff = 0;
    529 			ii->ii_index[0] = ix;
    530 			bn += cs->sc_cinfo[ix].ci_size;
    531 			ii++;
    532 		}
    533 		ii->ii_ndisk = 0;
    534 #ifdef DEBUG
    535 		if (ccddebug & CCDB_INIT)
    536 			printiinfo(cs->sc_itable);
    537 #endif
    538 		return;
    539 	}
    540 
    541 	/*
    542 	 * The following isn't fast or pretty; it doesn't have to be.
    543 	 */
    544 	size = 0;
    545 	bn = lbn = 0;
    546 	for (ii = cs->sc_itable; ; ii++) {
    547 		/* Allocate space for ii_index. */
    548 		ii->ii_indexsz = sizeof(int) * cs->sc_nccdisks;
    549 		ii->ii_index = kmem_alloc(ii->ii_indexsz, KM_SLEEP);
    550 
    551 		/*
    552 		 * Locate the smallest of the remaining components
    553 		 */
    554 		smallci = NULL;
    555 		for (ci = cs->sc_cinfo;
    556 		     ci < &cs->sc_cinfo[cs->sc_nccdisks]; ci++)
    557 			if (ci->ci_size > size &&
    558 			    (smallci == NULL ||
    559 			     ci->ci_size < smallci->ci_size))
    560 				smallci = ci;
    561 
    562 		/*
    563 		 * Nobody left, all done
    564 		 */
    565 		if (smallci == NULL) {
    566 			ii->ii_ndisk = 0;
    567 			break;
    568 		}
    569 
    570 		/*
    571 		 * Record starting logical block and component offset
    572 		 */
    573 		ii->ii_startblk = bn / cs->sc_ileave;
    574 		ii->ii_startoff = lbn;
    575 
    576 		/*
    577 		 * Determine how many disks take part in this interleave
    578 		 * and record their indices.
    579 		 */
    580 		ix = 0;
    581 		for (ci = cs->sc_cinfo;
    582 		     ci < &cs->sc_cinfo[cs->sc_nccdisks]; ci++)
    583 			if (ci->ci_size >= smallci->ci_size)
    584 				ii->ii_index[ix++] = ci - cs->sc_cinfo;
    585 		ii->ii_ndisk = ix;
    586 		bn += ix * (smallci->ci_size - size);
    587 		lbn = smallci->ci_size / cs->sc_ileave;
    588 		size = smallci->ci_size;
    589 	}
    590 #ifdef DEBUG
    591 	if (ccddebug & CCDB_INIT)
    592 		printiinfo(cs->sc_itable);
    593 #endif
    594 }
    595 
    596 /* ARGSUSED */
    597 static int
    598 ccdopen(dev_t dev, int flags, int fmt, struct lwp *l)
    599 {
    600 	int unit = ccdunit(dev);
    601 	struct ccd_softc *cs;
    602 	struct disklabel *lp;
    603 	int error = 0, part, pmask;
    604 
    605 #ifdef DEBUG
    606 	if (ccddebug & CCDB_FOLLOW)
    607 		printf("ccdopen(0x%"PRIx64", 0x%x)\n", dev, flags);
    608 #endif
    609 	if ((cs = ccdget(unit, 1)) == NULL)
    610 		return ENXIO;
    611 
    612 	mutex_enter(&cs->sc_dvlock);
    613 
    614 	lp = cs->sc_dkdev.dk_label;
    615 
    616 	part = DISKPART(dev);
    617 	pmask = (1 << part);
    618 
    619 	/*
    620 	 * If we're initialized, check to see if there are any other
    621 	 * open partitions.  If not, then it's safe to update
    622 	 * the in-core disklabel.  Only read the disklabel if it is
    623 	 * not already valid.
    624 	 */
    625 	if ((cs->sc_flags & (CCDF_INITED|CCDF_VLABEL)) == CCDF_INITED &&
    626 	    cs->sc_dkdev.dk_openmask == 0)
    627 		ccdgetdisklabel(dev);
    628 
    629 	/* Check that the partition exists. */
    630 	if (part != RAW_PART) {
    631 		if (((cs->sc_flags & CCDF_INITED) == 0) ||
    632 		    ((part >= lp->d_npartitions) ||
    633 		     (lp->d_partitions[part].p_fstype == FS_UNUSED))) {
    634 			error = ENXIO;
    635 			goto done;
    636 		}
    637 	}
    638 
    639 	/* Prevent our unit from being unconfigured while open. */
    640 	switch (fmt) {
    641 	case S_IFCHR:
    642 		cs->sc_dkdev.dk_copenmask |= pmask;
    643 		break;
    644 
    645 	case S_IFBLK:
    646 		cs->sc_dkdev.dk_bopenmask |= pmask;
    647 		break;
    648 	}
    649 	cs->sc_dkdev.dk_openmask =
    650 	    cs->sc_dkdev.dk_copenmask | cs->sc_dkdev.dk_bopenmask;
    651 
    652  done:
    653 	mutex_exit(&cs->sc_dvlock);
    654 	return (error);
    655 }
    656 
    657 /* ARGSUSED */
    658 static int
    659 ccdclose(dev_t dev, int flags, int fmt, struct lwp *l)
    660 {
    661 	int unit = ccdunit(dev);
    662 	struct ccd_softc *cs;
    663 	int part;
    664 
    665 #ifdef DEBUG
    666 	if (ccddebug & CCDB_FOLLOW)
    667 		printf("ccdclose(0x%"PRIx64", 0x%x)\n", dev, flags);
    668 #endif
    669 
    670 	if ((cs = ccdget(unit, 0)) == NULL)
    671 		return ENXIO;
    672 
    673 	mutex_enter(&cs->sc_dvlock);
    674 
    675 	part = DISKPART(dev);
    676 
    677 	/* ...that much closer to allowing unconfiguration... */
    678 	switch (fmt) {
    679 	case S_IFCHR:
    680 		cs->sc_dkdev.dk_copenmask &= ~(1 << part);
    681 		break;
    682 
    683 	case S_IFBLK:
    684 		cs->sc_dkdev.dk_bopenmask &= ~(1 << part);
    685 		break;
    686 	}
    687 	cs->sc_dkdev.dk_openmask =
    688 	    cs->sc_dkdev.dk_copenmask | cs->sc_dkdev.dk_bopenmask;
    689 
    690 	if (cs->sc_dkdev.dk_openmask == 0) {
    691 		if ((cs->sc_flags & CCDF_KLABEL) == 0)
    692 			cs->sc_flags &= ~CCDF_VLABEL;
    693 	}
    694 
    695 	mutex_exit(&cs->sc_dvlock);
    696 	return (0);
    697 }
    698 
    699 static bool
    700 ccdbackoff(struct ccd_softc *cs)
    701 {
    702 
    703 	/* XXX Arbitrary, should be a uvm call. */
    704 	return uvmexp.free < (uvmexp.freemin >> 1) &&
    705 	    disk_isbusy(&cs->sc_dkdev);
    706 }
    707 
    708 static void
    709 ccdthread(void *cookie)
    710 {
    711 	struct ccd_softc *cs;
    712 
    713 	cs = cookie;
    714 
    715 #ifdef DEBUG
    716  	if (ccddebug & CCDB_FOLLOW)
    717  		printf("ccdthread: hello\n");
    718 #endif
    719 
    720 	mutex_enter(cs->sc_iolock);
    721 	while (__predict_true(!cs->sc_zap)) {
    722 		if (bufq_peek(cs->sc_bufq) == NULL) {
    723 			/* Nothing to do. */
    724 			cv_wait(&cs->sc_push, cs->sc_iolock);
    725 			continue;
    726 		}
    727 		if (ccdbackoff(cs)) {
    728 			/* Wait for memory to become available. */
    729 			(void)cv_timedwait(&cs->sc_push, cs->sc_iolock, 1);
    730 			continue;
    731 		}
    732 #ifdef DEBUG
    733  		if (ccddebug & CCDB_FOLLOW)
    734  			printf("ccdthread: dispatching I/O\n");
    735 #endif
    736 		ccdstart(cs);
    737 		mutex_enter(cs->sc_iolock);
    738 	}
    739 	cs->sc_thread = NULL;
    740 	mutex_exit(cs->sc_iolock);
    741 #ifdef DEBUG
    742  	if (ccddebug & CCDB_FOLLOW)
    743  		printf("ccdthread: goodbye\n");
    744 #endif
    745 	kthread_exit(0);
    746 }
    747 
    748 static void
    749 ccdstrategy(struct buf *bp)
    750 {
    751 	int unit = ccdunit(bp->b_dev);
    752 	struct ccd_softc *cs;
    753 	if ((cs = ccdget(unit, 0)) == NULL)
    754 		return;
    755 
    756 	/* Must be open or reading label. */
    757 	KASSERT(cs->sc_dkdev.dk_openmask != 0 ||
    758 	    (cs->sc_flags & CCDF_RLABEL) != 0);
    759 
    760 	mutex_enter(cs->sc_iolock);
    761 	/* Synchronize with device init/uninit. */
    762 	if (__predict_false((cs->sc_flags & CCDF_INITED) == 0)) {
    763 		mutex_exit(cs->sc_iolock);
    764 #ifdef DEBUG
    765  		if (ccddebug & CCDB_FOLLOW)
    766  			printf("ccdstrategy: unit %d: not inited\n", unit);
    767 #endif
    768  		bp->b_error = ENXIO;
    769  		bp->b_resid = bp->b_bcount;
    770  		biodone(bp);
    771 		return;
    772 	}
    773 
    774 	/* Defer to thread if system is low on memory. */
    775 	bufq_put(cs->sc_bufq, bp);
    776 	if (__predict_false(ccdbackoff(cs))) {
    777 		mutex_exit(cs->sc_iolock);
    778 #ifdef DEBUG
    779  		if (ccddebug & CCDB_FOLLOW)
    780  			printf("ccdstrategy: holding off on I/O\n");
    781 #endif
    782 		return;
    783 	}
    784 	ccdstart(cs);
    785 }
    786 
    787 static void
    788 ccdstart(struct ccd_softc *cs)
    789 {
    790 	daddr_t blkno;
    791 	int wlabel;
    792 	struct disklabel *lp;
    793 	long bcount, rcount;
    794 	struct ccdbuf *cbp;
    795 	char *addr;
    796 	daddr_t bn;
    797 	vnode_t *vp;
    798 	buf_t *bp;
    799 
    800 	KASSERT(mutex_owned(cs->sc_iolock));
    801 
    802 	disk_busy(&cs->sc_dkdev);
    803 	bp = bufq_get(cs->sc_bufq);
    804 	KASSERT(bp != NULL);
    805 
    806 #ifdef DEBUG
    807 	if (ccddebug & CCDB_FOLLOW)
    808 		printf("ccdstart(%s, %p)\n", cs->sc_xname, bp);
    809 #endif
    810 
    811 	/* If it's a nil transfer, wake up the top half now. */
    812 	if (bp->b_bcount == 0)
    813 		goto done;
    814 
    815 	lp = cs->sc_dkdev.dk_label;
    816 
    817 	/*
    818 	 * Do bounds checking and adjust transfer.  If there's an
    819 	 * error, the bounds check will flag that for us.  Convert
    820 	 * the partition relative block number to an absolute.
    821 	 */
    822 	blkno = bp->b_blkno;
    823 	wlabel = cs->sc_flags & (CCDF_WLABEL|CCDF_LABELLING);
    824 	if (DISKPART(bp->b_dev) != RAW_PART) {
    825 		if (bounds_check_with_label(&cs->sc_dkdev, bp, wlabel) <= 0)
    826 			goto done;
    827 		blkno += lp->d_partitions[DISKPART(bp->b_dev)].p_offset;
    828 	}
    829 	mutex_exit(cs->sc_iolock);
    830 	bp->b_rawblkno = blkno;
    831 
    832 	/* Allocate the component buffers and start I/O! */
    833 	bp->b_resid = bp->b_bcount;
    834 	bn = bp->b_rawblkno;
    835 	addr = bp->b_data;
    836 	for (bcount = bp->b_bcount; bcount > 0; bcount -= rcount) {
    837 		cbp = ccdbuffer(cs, bp, bn, addr, bcount);
    838 		rcount = cbp->cb_buf.b_bcount;
    839 		bn += btodb(rcount);
    840 		addr += rcount;
    841 		vp = cbp->cb_buf.b_vp;
    842 		if ((cbp->cb_buf.b_flags & B_READ) == 0) {
    843 			mutex_enter(vp->v_interlock);
    844 			vp->v_numoutput++;
    845 			mutex_exit(vp->v_interlock);
    846 		}
    847 		(void)VOP_STRATEGY(vp, &cbp->cb_buf);
    848 	}
    849 	return;
    850 
    851  done:
    852 	disk_unbusy(&cs->sc_dkdev, 0, 0);
    853 	cv_broadcast(&cs->sc_stop);
    854 	cv_broadcast(&cs->sc_push);
    855 	mutex_exit(cs->sc_iolock);
    856 	bp->b_resid = bp->b_bcount;
    857 	biodone(bp);
    858 }
    859 
    860 /*
    861  * Build a component buffer header.
    862  */
    863 static struct ccdbuf *
    864 ccdbuffer(struct ccd_softc *cs, struct buf *bp, daddr_t bn, void *addr,
    865     long bcount)
    866 {
    867 	struct ccdcinfo *ci;
    868 	struct ccdbuf *cbp;
    869 	daddr_t cbn, cboff;
    870 	u_int64_t cbc;
    871 	int ccdisk;
    872 
    873 #ifdef DEBUG
    874 	if (ccddebug & CCDB_IO)
    875 		printf("ccdbuffer(%p, %p, %" PRId64 ", %p, %ld)\n",
    876 		       cs, bp, bn, addr, bcount);
    877 #endif
    878 	/*
    879 	 * Determine which component bn falls in.
    880 	 */
    881 	cbn = bn;
    882 	cboff = 0;
    883 
    884 	/*
    885 	 * Serially concatenated
    886 	 */
    887 	if (cs->sc_ileave == 0) {
    888 		daddr_t sblk;
    889 
    890 		sblk = 0;
    891 		for (ccdisk = 0, ci = &cs->sc_cinfo[ccdisk];
    892 		    cbn >= sblk + ci->ci_size;
    893 		    ccdisk++, ci = &cs->sc_cinfo[ccdisk])
    894 			sblk += ci->ci_size;
    895 		cbn -= sblk;
    896 	}
    897 	/*
    898 	 * Interleaved
    899 	 */
    900 	else {
    901 		struct ccdiinfo *ii;
    902 		int off;
    903 
    904 		cboff = cbn % cs->sc_ileave;
    905 		cbn /= cs->sc_ileave;
    906 		for (ii = cs->sc_itable; ii->ii_ndisk; ii++)
    907 			if (ii->ii_startblk > cbn)
    908 				break;
    909 		ii--;
    910 		off = cbn - ii->ii_startblk;
    911 		if (ii->ii_ndisk == 1) {
    912 			ccdisk = ii->ii_index[0];
    913 			cbn = ii->ii_startoff + off;
    914 		} else {
    915 			ccdisk = ii->ii_index[off % ii->ii_ndisk];
    916 			cbn = ii->ii_startoff + off / ii->ii_ndisk;
    917 		}
    918 		cbn *= cs->sc_ileave;
    919 		ci = &cs->sc_cinfo[ccdisk];
    920 	}
    921 
    922 	/*
    923 	 * Fill in the component buf structure.
    924 	 */
    925 	cbp = CCD_GETBUF();
    926 	KASSERT(cbp != NULL);
    927 	buf_init(&cbp->cb_buf);
    928 	cbp->cb_buf.b_flags = bp->b_flags;
    929 	cbp->cb_buf.b_oflags = bp->b_oflags;
    930 	cbp->cb_buf.b_cflags = bp->b_cflags;
    931 	cbp->cb_buf.b_iodone = ccdiodone;
    932 	cbp->cb_buf.b_proc = bp->b_proc;
    933 	cbp->cb_buf.b_dev = ci->ci_dev;
    934 	cbp->cb_buf.b_blkno = cbn + cboff;
    935 	cbp->cb_buf.b_data = addr;
    936 	cbp->cb_buf.b_vp = ci->ci_vp;
    937 	cbp->cb_buf.b_objlock = ci->ci_vp->v_interlock;
    938 	if (cs->sc_ileave == 0)
    939 		cbc = dbtob((u_int64_t)(ci->ci_size - cbn));
    940 	else
    941 		cbc = dbtob((u_int64_t)(cs->sc_ileave - cboff));
    942 	cbp->cb_buf.b_bcount = cbc < bcount ? cbc : bcount;
    943 
    944 	/*
    945 	 * context for ccdiodone
    946 	 */
    947 	cbp->cb_obp = bp;
    948 	cbp->cb_sc = cs;
    949 	cbp->cb_comp = ccdisk;
    950 
    951 	BIO_COPYPRIO(&cbp->cb_buf, bp);
    952 
    953 #ifdef DEBUG
    954 	if (ccddebug & CCDB_IO)
    955 		printf(" dev 0x%"PRIx64"(u%lu): cbp %p bn %" PRId64 " addr %p"
    956 		       " bcnt %d\n",
    957 		    ci->ci_dev, (unsigned long) (ci-cs->sc_cinfo), cbp,
    958 		    cbp->cb_buf.b_blkno, cbp->cb_buf.b_data,
    959 		    cbp->cb_buf.b_bcount);
    960 #endif
    961 
    962 	return (cbp);
    963 }
    964 
    965 /*
    966  * Called at interrupt time.
    967  * Mark the component as done and if all components are done,
    968  * take a ccd interrupt.
    969  */
    970 static void
    971 ccdiodone(struct buf *vbp)
    972 {
    973 	struct ccdbuf *cbp = (struct ccdbuf *) vbp;
    974 	struct buf *bp = cbp->cb_obp;
    975 	struct ccd_softc *cs = cbp->cb_sc;
    976 	int count;
    977 
    978 #ifdef DEBUG
    979 	if (ccddebug & CCDB_FOLLOW)
    980 		printf("ccdiodone(%p)\n", cbp);
    981 	if (ccddebug & CCDB_IO) {
    982 		printf("ccdiodone: bp %p bcount %d resid %d\n",
    983 		       bp, bp->b_bcount, bp->b_resid);
    984 		printf(" dev 0x%"PRIx64"(u%d), cbp %p bn %" PRId64 " addr %p"
    985 		       " bcnt %d\n",
    986 		       cbp->cb_buf.b_dev, cbp->cb_comp, cbp,
    987 		       cbp->cb_buf.b_blkno, cbp->cb_buf.b_data,
    988 		       cbp->cb_buf.b_bcount);
    989 	}
    990 #endif
    991 
    992 	if (cbp->cb_buf.b_error != 0) {
    993 		bp->b_error = cbp->cb_buf.b_error;
    994 		printf("%s: error %d on component %d\n",
    995 		       cs->sc_xname, bp->b_error, cbp->cb_comp);
    996 	}
    997 	count = cbp->cb_buf.b_bcount;
    998 	buf_destroy(&cbp->cb_buf);
    999 	CCD_PUTBUF(cbp);
   1000 
   1001 	/*
   1002 	 * If all done, "interrupt".
   1003 	 */
   1004 	mutex_enter(cs->sc_iolock);
   1005 	bp->b_resid -= count;
   1006 	if (bp->b_resid < 0)
   1007 		panic("ccdiodone: count");
   1008 	if (bp->b_resid == 0) {
   1009 		/*
   1010 		 * Request is done for better or worse, wakeup the top half.
   1011 		 */
   1012 		if (bp->b_error != 0)
   1013 			bp->b_resid = bp->b_bcount;
   1014 		disk_unbusy(&cs->sc_dkdev, (bp->b_bcount - bp->b_resid),
   1015 		    (bp->b_flags & B_READ));
   1016 		if (!disk_isbusy(&cs->sc_dkdev)) {
   1017 			if (bufq_peek(cs->sc_bufq) != NULL) {
   1018 				cv_broadcast(&cs->sc_push);
   1019 			}
   1020 			cv_broadcast(&cs->sc_stop);
   1021 		}
   1022 		mutex_exit(cs->sc_iolock);
   1023 		biodone(bp);
   1024 	} else
   1025 		mutex_exit(cs->sc_iolock);
   1026 }
   1027 
   1028 /* ARGSUSED */
   1029 static int
   1030 ccdread(dev_t dev, struct uio *uio, int flags)
   1031 {
   1032 	int unit = ccdunit(dev);
   1033 	struct ccd_softc *cs;
   1034 
   1035 #ifdef DEBUG
   1036 	if (ccddebug & CCDB_FOLLOW)
   1037 		printf("ccdread(0x%"PRIx64", %p)\n", dev, uio);
   1038 #endif
   1039 	if ((cs = ccdget(unit, 0)) == NULL)
   1040 		return 0;
   1041 
   1042 	/* Unlocked advisory check, ccdstrategy check is synchronous. */
   1043 	if ((cs->sc_flags & CCDF_INITED) == 0)
   1044 		return (ENXIO);
   1045 
   1046 	return (physio(ccdstrategy, NULL, dev, B_READ, minphys, uio));
   1047 }
   1048 
   1049 /* ARGSUSED */
   1050 static int
   1051 ccdwrite(dev_t dev, struct uio *uio, int flags)
   1052 {
   1053 	int unit = ccdunit(dev);
   1054 	struct ccd_softc *cs;
   1055 
   1056 #ifdef DEBUG
   1057 	if (ccddebug & CCDB_FOLLOW)
   1058 		printf("ccdwrite(0x%"PRIx64", %p)\n", dev, uio);
   1059 #endif
   1060 	if ((cs = ccdget(unit, 0)) == NULL)
   1061 		return ENOENT;
   1062 
   1063 	/* Unlocked advisory check, ccdstrategy check is synchronous. */
   1064 	if ((cs->sc_flags & CCDF_INITED) == 0)
   1065 		return (ENXIO);
   1066 
   1067 	return (physio(ccdstrategy, NULL, dev, B_WRITE, minphys, uio));
   1068 }
   1069 
   1070 static int
   1071 ccdioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l)
   1072 {
   1073 	int unit = ccdunit(dev);
   1074 	int i, j, lookedup = 0, error = 0;
   1075 	int part, pmask, make;
   1076 	struct ccd_softc *cs;
   1077 	struct ccd_ioctl *ccio = (struct ccd_ioctl *)data;
   1078 	kauth_cred_t uc;
   1079 	char **cpp;
   1080 	struct pathbuf *pb;
   1081 	struct vnode **vpp;
   1082 #ifdef __HAVE_OLD_DISKLABEL
   1083 	struct disklabel newlabel;
   1084 #endif
   1085 
   1086 	switch (cmd) {
   1087 #if defined(COMPAT_60) && !defined(_LP64)
   1088 	case CCDIOCSET_60:
   1089 #endif
   1090 	case CCDIOCSET:
   1091 		make = 1;
   1092 		break;
   1093 	default:
   1094 		make = 0;
   1095 		break;
   1096 	}
   1097 
   1098 	if ((cs = ccdget(unit, make)) == NULL)
   1099 		return ENOENT;
   1100 	uc = kauth_cred_get();
   1101 
   1102 /*
   1103  * Compat code must not be called if on a platform where
   1104  * sizeof (size_t) == sizeof (uint64_t) as CCDIOCSET will
   1105  * be the same as CCDIOCSET_60
   1106  */
   1107 #if defined(COMPAT_60) && !defined(_LP64)
   1108 	switch (cmd) {
   1109 	case CCDIOCSET_60: {
   1110 		struct ccd_ioctl ccionew;
   1111        		struct ccd_ioctl_60 *ccio60 =
   1112        		    (struct ccd_ioctl_60 *)data;
   1113 		ccionew.ccio_disks = ccio->ccio_disks;
   1114 		ccionew.ccio_ndisks = ccio->ccio_ndisks;
   1115 		ccionew.ccio_ileave = ccio->ccio_ileave;
   1116 		ccionew.ccio_flags = ccio->ccio_flags;
   1117 		ccionew.ccio_unit = ccio->ccio_unit;
   1118 		error = ccdioctl(dev, CCDIOCSET, &ccionew, flag, l);
   1119 		if (!error) {
   1120 			/* Copy data back, adjust types if necessary */
   1121 			ccio60->ccio_disks = ccionew.ccio_disks;
   1122 			ccio60->ccio_ndisks = ccionew.ccio_ndisks;
   1123 			ccio60->ccio_ileave = ccionew.ccio_ileave;
   1124 			ccio60->ccio_flags = ccionew.ccio_flags;
   1125 			ccio60->ccio_unit = ccionew.ccio_unit;
   1126 			ccio60->ccio_size = (size_t)ccionew.ccio_size;
   1127 		}
   1128 		return error;
   1129 		}
   1130 		break;
   1131 
   1132 	case CCDIOCCLR_60:
   1133 		/*
   1134 		 * ccio_size member not used, so existing struct OK
   1135 		 * drop through to existing non-compat version
   1136 		 */
   1137 		cmd = CCDIOCCLR;
   1138 		break;
   1139 	}
   1140 #endif /* COMPAT_60 && !_LP64*/
   1141 
   1142 	/* Must be open for writes for these commands... */
   1143 	switch (cmd) {
   1144 	case CCDIOCSET:
   1145 	case CCDIOCCLR:
   1146 	case DIOCSDINFO:
   1147 	case DIOCWDINFO:
   1148 	case DIOCCACHESYNC:
   1149 	case DIOCAWEDGE:
   1150 	case DIOCDWEDGE:
   1151 	case DIOCMWEDGES:
   1152 #ifdef __HAVE_OLD_DISKLABEL
   1153 	case ODIOCSDINFO:
   1154 	case ODIOCWDINFO:
   1155 #endif
   1156 	case DIOCKLABEL:
   1157 	case DIOCWLABEL:
   1158 		if ((flag & FWRITE) == 0)
   1159 			return (EBADF);
   1160 	}
   1161 
   1162 	mutex_enter(&cs->sc_dvlock);
   1163 
   1164 	/* Must be initialized for these... */
   1165 	switch (cmd) {
   1166 	case CCDIOCCLR:
   1167 	case DIOCGDINFO:
   1168 	case DIOCCACHESYNC:
   1169 	case DIOCAWEDGE:
   1170 	case DIOCDWEDGE:
   1171 	case DIOCLWEDGES:
   1172 	case DIOCMWEDGES:
   1173 	case DIOCSDINFO:
   1174 	case DIOCWDINFO:
   1175 	case DIOCGPART:
   1176 	case DIOCWLABEL:
   1177 	case DIOCKLABEL:
   1178 	case DIOCGDEFLABEL:
   1179 #ifdef __HAVE_OLD_DISKLABEL
   1180 	case ODIOCGDINFO:
   1181 	case ODIOCSDINFO:
   1182 	case ODIOCWDINFO:
   1183 	case ODIOCGDEFLABEL:
   1184 #endif
   1185 		if ((cs->sc_flags & CCDF_INITED) == 0) {
   1186 			error = ENXIO;
   1187 			goto out;
   1188 		}
   1189 	}
   1190 
   1191 	error = disk_ioctl(&cs->sc_dkdev, cmd, data, flag, l);
   1192 	if (error != EPASSTHROUGH)
   1193 		goto out;
   1194 
   1195 	error = 0;
   1196 	switch (cmd) {
   1197 	case CCDIOCSET:
   1198 		if (cs->sc_flags & CCDF_INITED) {
   1199 			error = EBUSY;
   1200 			goto out;
   1201 		}
   1202 
   1203 		/* Validate the flags. */
   1204 		if ((ccio->ccio_flags & CCDF_USERMASK) != ccio->ccio_flags) {
   1205 			error = EINVAL;
   1206 			goto out;
   1207 		}
   1208 
   1209 		if (ccio->ccio_ndisks > CCD_MAXNDISKS ||
   1210 		    ccio->ccio_ndisks == 0) {
   1211 			error = EINVAL;
   1212 			goto out;
   1213 		}
   1214 
   1215 		/* Fill in some important bits. */
   1216 		cs->sc_ileave = ccio->ccio_ileave;
   1217 		cs->sc_nccdisks = ccio->ccio_ndisks;
   1218 		cs->sc_flags = ccio->ccio_flags & CCDF_USERMASK;
   1219 
   1220 		/*
   1221 		 * Allocate space for and copy in the array of
   1222 		 * component pathnames and device numbers.
   1223 		 */
   1224 		cpp = kmem_alloc(ccio->ccio_ndisks * sizeof(*cpp), KM_SLEEP);
   1225 		vpp = kmem_alloc(ccio->ccio_ndisks * sizeof(*vpp), KM_SLEEP);
   1226 		error = copyin(ccio->ccio_disks, cpp,
   1227 		    ccio->ccio_ndisks * sizeof(*cpp));
   1228 		if (error) {
   1229 			kmem_free(vpp, ccio->ccio_ndisks * sizeof(*vpp));
   1230 			kmem_free(cpp, ccio->ccio_ndisks * sizeof(*cpp));
   1231 			goto out;
   1232 		}
   1233 
   1234 #ifdef DEBUG
   1235 		if (ccddebug & CCDB_INIT)
   1236 			for (i = 0; i < ccio->ccio_ndisks; ++i)
   1237 				printf("ccdioctl: component %d: %p\n",
   1238 				    i, cpp[i]);
   1239 #endif
   1240 
   1241 		for (i = 0; i < ccio->ccio_ndisks; ++i) {
   1242 #ifdef DEBUG
   1243 			if (ccddebug & CCDB_INIT)
   1244 				printf("ccdioctl: lookedup = %d\n", lookedup);
   1245 #endif
   1246 			error = pathbuf_copyin(cpp[i], &pb);
   1247 			if (error == 0) {
   1248 				error = dk_lookup(pb, l, &vpp[i]);
   1249 			}
   1250 			pathbuf_destroy(pb);
   1251 			if (error != 0) {
   1252 				for (j = 0; j < lookedup; ++j)
   1253 					(void)vn_close(vpp[j], FREAD|FWRITE,
   1254 					    uc);
   1255 				kmem_free(vpp, ccio->ccio_ndisks *
   1256 				    sizeof(*vpp));
   1257 				kmem_free(cpp, ccio->ccio_ndisks *
   1258 				    sizeof(*cpp));
   1259 				goto out;
   1260 			}
   1261 			++lookedup;
   1262 		}
   1263 
   1264 		/* Attach the disk. */
   1265 		disk_attach(&cs->sc_dkdev);
   1266 		bufq_alloc(&cs->sc_bufq, "fcfs", 0);
   1267 
   1268 		/*
   1269 		 * Initialize the ccd.  Fills in the softc for us.
   1270 		 */
   1271 		if ((error = ccdinit(cs, cpp, vpp, l)) != 0) {
   1272 			for (j = 0; j < lookedup; ++j)
   1273 				(void)vn_close(vpp[j], FREAD|FWRITE,
   1274 				    uc);
   1275 			kmem_free(vpp, ccio->ccio_ndisks * sizeof(*vpp));
   1276 			kmem_free(cpp, ccio->ccio_ndisks * sizeof(*cpp));
   1277 			disk_detach(&cs->sc_dkdev);
   1278 			bufq_free(cs->sc_bufq);
   1279 			goto out;
   1280 		}
   1281 
   1282 		/* We can free the temporary variables now. */
   1283 		kmem_free(vpp, ccio->ccio_ndisks * sizeof(*vpp));
   1284 		kmem_free(cpp, ccio->ccio_ndisks * sizeof(*cpp));
   1285 
   1286 		/*
   1287 		 * The ccd has been successfully initialized, so
   1288 		 * we can place it into the array.  Don't try to
   1289 		 * read the disklabel until the disk has been attached,
   1290 		 * because space for the disklabel is allocated
   1291 		 * in disk_attach();
   1292 		 */
   1293 		ccio->ccio_unit = unit;
   1294 		ccio->ccio_size = cs->sc_size;
   1295 
   1296 		/* Try and read the disklabel. */
   1297 		ccdgetdisklabel(dev);
   1298 		disk_set_info(NULL, &cs->sc_dkdev, NULL);
   1299 
   1300 		/* discover wedges */
   1301 		mutex_exit(&cs->sc_dvlock);
   1302 		dkwedge_discover(&cs->sc_dkdev);
   1303 		return 0;
   1304 
   1305 	case CCDIOCCLR:
   1306 		/*
   1307 		 * Don't unconfigure if any other partitions are open
   1308 		 * or if both the character and block flavors of this
   1309 		 * partition are open.
   1310 		 */
   1311 		part = DISKPART(dev);
   1312 		pmask = (1 << part);
   1313 		if ((cs->sc_dkdev.dk_openmask & ~pmask) ||
   1314 		    ((cs->sc_dkdev.dk_bopenmask & pmask) &&
   1315 		    (cs->sc_dkdev.dk_copenmask & pmask))) {
   1316 			error = EBUSY;
   1317 			goto out;
   1318 		}
   1319 
   1320 		/* Delete all of our wedges. */
   1321 		dkwedge_delall(&cs->sc_dkdev);
   1322 
   1323 		/* Stop new I/O, wait for in-flight I/O to complete. */
   1324 		mutex_enter(cs->sc_iolock);
   1325 		cs->sc_flags &= ~(CCDF_INITED|CCDF_VLABEL);
   1326 		cs->sc_zap = true;
   1327 		while (disk_isbusy(&cs->sc_dkdev) ||
   1328 		    bufq_peek(cs->sc_bufq) != NULL ||
   1329 		    cs->sc_thread != NULL) {
   1330 			cv_broadcast(&cs->sc_push);
   1331 			(void)cv_timedwait(&cs->sc_stop, cs->sc_iolock, hz);
   1332 		}
   1333 		mutex_exit(cs->sc_iolock);
   1334 
   1335 		/*
   1336 		 * Free ccd_softc information and clear entry.
   1337 		 */
   1338 
   1339 		/* Close the components and free their pathnames. */
   1340 		for (i = 0; i < cs->sc_nccdisks; ++i) {
   1341 			/*
   1342 			 * XXX: this close could potentially fail and
   1343 			 * cause Bad Things.  Maybe we need to force
   1344 			 * the close to happen?
   1345 			 */
   1346 #ifdef DEBUG
   1347 			if (ccddebug & CCDB_VNODE)
   1348 				vprint("CCDIOCCLR: vnode info",
   1349 				    cs->sc_cinfo[i].ci_vp);
   1350 #endif
   1351 			(void)vn_close(cs->sc_cinfo[i].ci_vp, FREAD|FWRITE,
   1352 			    uc);
   1353 			kmem_free(cs->sc_cinfo[i].ci_path,
   1354 			    cs->sc_cinfo[i].ci_pathlen);
   1355 		}
   1356 
   1357 		/* Free interleave index. */
   1358 		for (i = 0; cs->sc_itable[i].ii_ndisk; ++i) {
   1359 			kmem_free(cs->sc_itable[i].ii_index,
   1360 			    cs->sc_itable[i].ii_indexsz);
   1361 		}
   1362 
   1363 		/* Free component info and interleave table. */
   1364 		kmem_free(cs->sc_cinfo, cs->sc_nccdisks *
   1365 		    sizeof(struct ccdcinfo));
   1366 		kmem_free(cs->sc_itable, (cs->sc_nccdisks + 1) *
   1367 		    sizeof(struct ccdiinfo));
   1368 
   1369 		aprint_normal("%s: detached\n", cs->sc_xname);
   1370 
   1371 		/* Detach the disk. */
   1372 		disk_detach(&cs->sc_dkdev);
   1373 		bufq_free(cs->sc_bufq);
   1374 		ccdput(cs);
   1375 		/* Don't break, otherwise cs is read again. */
   1376 		return 0;
   1377 
   1378 	case DIOCGDINFO:
   1379 		*(struct disklabel *)data = *(cs->sc_dkdev.dk_label);
   1380 		break;
   1381 
   1382 #ifdef __HAVE_OLD_DISKLABEL
   1383 	case ODIOCGDINFO:
   1384 		newlabel = *(cs->sc_dkdev.dk_label);
   1385 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
   1386 			return ENOTTY;
   1387 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
   1388 		break;
   1389 #endif
   1390 
   1391 	case DIOCGPART:
   1392 		((struct partinfo *)data)->disklab = cs->sc_dkdev.dk_label;
   1393 		((struct partinfo *)data)->part =
   1394 		    &cs->sc_dkdev.dk_label->d_partitions[DISKPART(dev)];
   1395 		break;
   1396 
   1397 	case DIOCCACHESYNC:
   1398 		/*
   1399 		 * We pass this call down to all components and report
   1400 		 * the first error we encounter.
   1401 		 */
   1402 		for (error = 0, i = 0; i < cs->sc_nccdisks; i++) {
   1403 			j = VOP_IOCTL(cs->sc_cinfo[i].ci_vp, cmd, data,
   1404 				      flag, uc);
   1405 			if (j != 0 && error == 0)
   1406 				error = j;
   1407 		}
   1408 		break;
   1409 
   1410 	case DIOCAWEDGE:
   1411 	    {
   1412 	    	struct dkwedge_info *dkw = (void *) data;
   1413 
   1414 		/* If the ioctl happens here, the parent is us. */
   1415 		strlcpy(dkw->dkw_parent, cs->sc_xname,
   1416 			sizeof(dkw->dkw_parent));
   1417 		error = dkwedge_add(dkw);
   1418 		break;
   1419 	    }
   1420 
   1421 	case DIOCDWEDGE:
   1422 	    {
   1423 	    	struct dkwedge_info *dkw = (void *) data;
   1424 
   1425 		/* If the ioctl happens here, the parent is us. */
   1426 		strlcpy(dkw->dkw_parent, cs->sc_xname,
   1427 			sizeof(dkw->dkw_parent));
   1428 		error = dkwedge_del(dkw);
   1429 		break;
   1430 	    }
   1431 
   1432 	case DIOCLWEDGES:
   1433 	    {
   1434 	    	struct dkwedge_list *dkwl = (void *) data;
   1435 
   1436 		error = dkwedge_list(&cs->sc_dkdev, dkwl, l);
   1437 		break;
   1438 	    }
   1439 
   1440 	case DIOCMWEDGES:
   1441 	    {
   1442 		dkwedge_discover(&cs->sc_dkdev);
   1443 		break;
   1444 	    }
   1445 
   1446 	case DIOCWDINFO:
   1447 	case DIOCSDINFO:
   1448 #ifdef __HAVE_OLD_DISKLABEL
   1449 	case ODIOCWDINFO:
   1450 	case ODIOCSDINFO:
   1451 #endif
   1452 	{
   1453 		struct disklabel *lp;
   1454 #ifdef __HAVE_OLD_DISKLABEL
   1455 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
   1456 			memset(&newlabel, 0, sizeof newlabel);
   1457 			memcpy(&newlabel, data, sizeof (struct olddisklabel));
   1458 			lp = &newlabel;
   1459 		} else
   1460 #endif
   1461 		lp = (struct disklabel *)data;
   1462 
   1463 		cs->sc_flags |= CCDF_LABELLING;
   1464 
   1465 		error = setdisklabel(cs->sc_dkdev.dk_label,
   1466 		    lp, 0, cs->sc_dkdev.dk_cpulabel);
   1467 		if (error == 0) {
   1468 			if (cmd == DIOCWDINFO
   1469 #ifdef __HAVE_OLD_DISKLABEL
   1470 			    || cmd == ODIOCWDINFO
   1471 #endif
   1472 			   )
   1473 				error = writedisklabel(CCDLABELDEV(dev),
   1474 				    ccdstrategy, cs->sc_dkdev.dk_label,
   1475 				    cs->sc_dkdev.dk_cpulabel);
   1476 		}
   1477 
   1478 		cs->sc_flags &= ~CCDF_LABELLING;
   1479 		break;
   1480 	}
   1481 
   1482 	case DIOCKLABEL:
   1483 		if (*(int *)data != 0)
   1484 			cs->sc_flags |= CCDF_KLABEL;
   1485 		else
   1486 			cs->sc_flags &= ~CCDF_KLABEL;
   1487 		break;
   1488 
   1489 	case DIOCWLABEL:
   1490 		if (*(int *)data != 0)
   1491 			cs->sc_flags |= CCDF_WLABEL;
   1492 		else
   1493 			cs->sc_flags &= ~CCDF_WLABEL;
   1494 		break;
   1495 
   1496 	case DIOCGDEFLABEL:
   1497 		ccdgetdefaultlabel(cs, (struct disklabel *)data);
   1498 		break;
   1499 
   1500 #ifdef __HAVE_OLD_DISKLABEL
   1501 	case ODIOCGDEFLABEL:
   1502 		ccdgetdefaultlabel(cs, &newlabel);
   1503 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
   1504 			return ENOTTY;
   1505 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
   1506 		break;
   1507 #endif
   1508 
   1509 	default:
   1510 		error = ENOTTY;
   1511 	}
   1512 
   1513  out:
   1514 	mutex_exit(&cs->sc_dvlock);
   1515 	return (error);
   1516 }
   1517 
   1518 static int
   1519 ccdsize(dev_t dev)
   1520 {
   1521 	struct ccd_softc *cs;
   1522 	struct disklabel *lp;
   1523 	int part, unit, omask, size;
   1524 
   1525 	unit = ccdunit(dev);
   1526 	if ((cs = ccdget(unit, 0)) == NULL)
   1527 		return -1;
   1528 
   1529 	if ((cs->sc_flags & CCDF_INITED) == 0)
   1530 		return (-1);
   1531 
   1532 	part = DISKPART(dev);
   1533 	omask = cs->sc_dkdev.dk_openmask & (1 << part);
   1534 	lp = cs->sc_dkdev.dk_label;
   1535 
   1536 	if (omask == 0 && ccdopen(dev, 0, S_IFBLK, curlwp))
   1537 		return (-1);
   1538 
   1539 	if (lp->d_partitions[part].p_fstype != FS_SWAP)
   1540 		size = -1;
   1541 	else
   1542 		size = lp->d_partitions[part].p_size *
   1543 		    (lp->d_secsize / DEV_BSIZE);
   1544 
   1545 	if (omask == 0 && ccdclose(dev, 0, S_IFBLK, curlwp))
   1546 		return (-1);
   1547 
   1548 	return (size);
   1549 }
   1550 
   1551 static void
   1552 ccdgetdefaultlabel(struct ccd_softc *cs, struct disklabel *lp)
   1553 {
   1554 	struct ccdgeom *ccg = &cs->sc_geom;
   1555 
   1556 	memset(lp, 0, sizeof(*lp));
   1557 
   1558 	if (cs->sc_size > UINT32_MAX)
   1559 		lp->d_secperunit = UINT32_MAX;
   1560 	else
   1561 		lp->d_secperunit = cs->sc_size;
   1562 	lp->d_secsize = ccg->ccg_secsize;
   1563 	lp->d_nsectors = ccg->ccg_nsectors;
   1564 	lp->d_ntracks = ccg->ccg_ntracks;
   1565 	lp->d_ncylinders = ccg->ccg_ncylinders;
   1566 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
   1567 
   1568 	strncpy(lp->d_typename, "ccd", sizeof(lp->d_typename));
   1569 	lp->d_type = DTYPE_CCD;
   1570 	strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
   1571 	lp->d_rpm = 3600;
   1572 	lp->d_interleave = 1;
   1573 	lp->d_flags = 0;
   1574 
   1575 	lp->d_partitions[RAW_PART].p_offset = 0;
   1576 	lp->d_partitions[RAW_PART].p_size = lp->d_secperunit;
   1577 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
   1578 	lp->d_npartitions = RAW_PART + 1;
   1579 
   1580 	lp->d_magic = DISKMAGIC;
   1581 	lp->d_magic2 = DISKMAGIC;
   1582 	lp->d_checksum = dkcksum(cs->sc_dkdev.dk_label);
   1583 }
   1584 
   1585 /*
   1586  * Read the disklabel from the ccd.  If one is not present, fake one
   1587  * up.
   1588  */
   1589 static void
   1590 ccdgetdisklabel(dev_t dev)
   1591 {
   1592 	int unit = ccdunit(dev);
   1593 	struct ccd_softc *cs;
   1594 	const char *errstring;
   1595 	struct disklabel *lp;
   1596 	struct cpu_disklabel *clp;
   1597 
   1598 	if ((cs = ccdget(unit, 0)) == NULL)
   1599 		return;
   1600 	lp = cs->sc_dkdev.dk_label;
   1601 	clp = cs->sc_dkdev.dk_cpulabel;
   1602 	KASSERT(mutex_owned(&cs->sc_dvlock));
   1603 
   1604 	memset(clp, 0, sizeof(*clp));
   1605 
   1606 	ccdgetdefaultlabel(cs, lp);
   1607 
   1608 	/*
   1609 	 * Call the generic disklabel extraction routine.
   1610 	 */
   1611 	cs->sc_flags |= CCDF_RLABEL;
   1612 	if ((cs->sc_flags & CCDF_NOLABEL) != 0)
   1613 		errstring = "CCDF_NOLABEL set; ignoring on-disk label";
   1614 	else
   1615 		errstring = readdisklabel(CCDLABELDEV(dev), ccdstrategy,
   1616 		    cs->sc_dkdev.dk_label, cs->sc_dkdev.dk_cpulabel);
   1617 	if (errstring)
   1618 		ccdmakedisklabel(cs);
   1619 	else {
   1620 		int i;
   1621 		struct partition *pp;
   1622 
   1623 		/*
   1624 		 * Sanity check whether the found disklabel is valid.
   1625 		 *
   1626 		 * This is necessary since total size of ccd may vary
   1627 		 * when an interleave is changed even though exactly
   1628 		 * same componets are used, and old disklabel may used
   1629 		 * if that is found.
   1630 		 */
   1631 		if (lp->d_secperunit < UINT32_MAX ?
   1632 			lp->d_secperunit != cs->sc_size :
   1633 			lp->d_secperunit > cs->sc_size)
   1634 			printf("WARNING: %s: "
   1635 			    "total sector size in disklabel (%ju) != "
   1636 			    "the size of ccd (%ju)\n", cs->sc_xname,
   1637 			    (uintmax_t)lp->d_secperunit,
   1638 			    (uintmax_t)cs->sc_size);
   1639 		for (i = 0; i < lp->d_npartitions; i++) {
   1640 			pp = &lp->d_partitions[i];
   1641 			if (pp->p_offset + pp->p_size > cs->sc_size)
   1642 				printf("WARNING: %s: end of partition `%c' "
   1643 				    "exceeds the size of ccd (%ju)\n",
   1644 				    cs->sc_xname, 'a' + i, (uintmax_t)cs->sc_size);
   1645 		}
   1646 	}
   1647 
   1648 #ifdef DEBUG
   1649 	/* It's actually extremely common to have unlabeled ccds. */
   1650 	if (ccddebug & CCDB_LABEL)
   1651 		if (errstring != NULL)
   1652 			printf("%s: %s\n", cs->sc_xname, errstring);
   1653 #endif
   1654 
   1655 	/* In-core label now valid. */
   1656 	cs->sc_flags = (cs->sc_flags | CCDF_VLABEL) & ~CCDF_RLABEL;
   1657 }
   1658 
   1659 /*
   1660  * Take care of things one might want to take care of in the event
   1661  * that a disklabel isn't present.
   1662  */
   1663 static void
   1664 ccdmakedisklabel(struct ccd_softc *cs)
   1665 {
   1666 	struct disklabel *lp = cs->sc_dkdev.dk_label;
   1667 
   1668 	/*
   1669 	 * For historical reasons, if there's no disklabel present
   1670 	 * the raw partition must be marked FS_BSDFFS.
   1671 	 */
   1672 	lp->d_partitions[RAW_PART].p_fstype = FS_BSDFFS;
   1673 
   1674 	strncpy(lp->d_packname, "default label", sizeof(lp->d_packname));
   1675 
   1676 	lp->d_checksum = dkcksum(lp);
   1677 }
   1678 
   1679 #ifdef DEBUG
   1680 static void
   1681 printiinfo(struct ccdiinfo *ii)
   1682 {
   1683 	int ix, i;
   1684 
   1685 	for (ix = 0; ii->ii_ndisk; ix++, ii++) {
   1686 		printf(" itab[%d]: #dk %d sblk %" PRId64 " soff %" PRId64,
   1687 		    ix, ii->ii_ndisk, ii->ii_startblk, ii->ii_startoff);
   1688 		for (i = 0; i < ii->ii_ndisk; i++)
   1689 			printf(" %d", ii->ii_index[i]);
   1690 		printf("\n");
   1691 	}
   1692 }
   1693 #endif
   1694 
   1695 MODULE(MODULE_CLASS_DRIVER, ccd, "dk_subr");
   1696 
   1697 static int
   1698 ccd_modcmd(modcmd_t cmd, void *arg)
   1699 {
   1700 	int error = 0;
   1701 #ifdef _MODULE
   1702 	int bmajor = -1, cmajor = -1;
   1703 #endif
   1704 
   1705 
   1706 	switch (cmd) {
   1707 	case MODULE_CMD_INIT:
   1708 #ifdef _MODULE
   1709 		ccdattach(0);
   1710 
   1711 		error = devsw_attach("ccd", &ccd_bdevsw, &bmajor,
   1712 		    &ccd_cdevsw, &cmajor);
   1713 #endif
   1714 		break;
   1715 
   1716 	case MODULE_CMD_FINI:
   1717 #ifdef _MODULE
   1718 		mutex_enter(&ccd_lock);
   1719 		if (ccd_nactive)
   1720 			error = EBUSY;
   1721 		else
   1722 			error = devsw_detach(&ccd_bdevsw, &ccd_cdevsw);
   1723 		mutex_exit(&ccd_lock);
   1724 #endif
   1725 		break;
   1726 
   1727 	case MODULE_CMD_STAT:
   1728 		return ENOTTY;
   1729 
   1730 	default:
   1731 		return ENOTTY;
   1732 	}
   1733 
   1734 	return error;
   1735 }
   1736 
   1737 static int
   1738 ccd_units_sysctl(SYSCTLFN_ARGS)
   1739 {
   1740 	struct sysctlnode node;
   1741 	struct ccd_softc *sc;
   1742 	int error, i, nccd, *units;
   1743 	size_t size;
   1744 
   1745 	nccd = 0;
   1746 	mutex_enter(&ccd_lock);
   1747 	LIST_FOREACH(sc, &ccds, sc_link)
   1748 		nccd++;
   1749 	mutex_exit(&ccd_lock);
   1750 
   1751 	if (nccd != 0) {
   1752 		size = nccd * sizeof(*units);
   1753 		units = kmem_zalloc(size, KM_SLEEP);
   1754 		if (units == NULL)
   1755 			return ENOMEM;
   1756 
   1757 		i = 0;
   1758 		mutex_enter(&ccd_lock);
   1759 		LIST_FOREACH(sc, &ccds, sc_link) {
   1760 			if (i >= nccd)
   1761 				break;
   1762 			units[i] = sc->sc_unit;
   1763 		}
   1764 		mutex_exit(&ccd_lock);
   1765 	} else {
   1766 		units = NULL;
   1767 		size = 0;
   1768 	}
   1769 
   1770 	node = *rnode;
   1771 	node.sysctl_data = units;
   1772 	node.sysctl_size = size;
   1773 
   1774 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
   1775 	if (units)
   1776 		kmem_free(units, size);
   1777 	return error;
   1778 }
   1779 
   1780 static int
   1781 ccd_info_sysctl(SYSCTLFN_ARGS)
   1782 {
   1783 	struct sysctlnode node;
   1784 	struct ccddiskinfo ccd;
   1785 	struct ccd_softc *sc;
   1786 	int unit;
   1787 
   1788 	if (newp == NULL || newlen != sizeof(int))
   1789 		return EINVAL;
   1790 
   1791 	unit = *(const int *)newp;
   1792 	newp = NULL;
   1793 	newlen = 0;
   1794 	ccd.ccd_ndisks = ~0;
   1795 	mutex_enter(&ccd_lock);
   1796 	LIST_FOREACH(sc, &ccds, sc_link) {
   1797 		if (sc->sc_unit == unit) {
   1798 			ccd.ccd_ileave = sc->sc_ileave;
   1799 			ccd.ccd_size = sc->sc_size;
   1800 			ccd.ccd_ndisks = sc->sc_nccdisks;
   1801 			ccd.ccd_flags = sc->sc_flags;
   1802 			break;
   1803 		}
   1804 	}
   1805 	mutex_exit(&ccd_lock);
   1806 
   1807 	if (ccd.ccd_ndisks == ~0)
   1808 		return ENOENT;
   1809 
   1810 	node = *rnode;
   1811 	node.sysctl_data = &ccd;
   1812 	node.sysctl_size = sizeof(ccd);
   1813 
   1814 	return sysctl_lookup(SYSCTLFN_CALL(&node));
   1815 }
   1816 
   1817 static int
   1818 ccd_components_sysctl(SYSCTLFN_ARGS)
   1819 {
   1820 	struct sysctlnode node;
   1821 	int error, unit;
   1822 	size_t size;
   1823 	char *names, *p, *ep;
   1824 	struct ccd_softc *sc;
   1825 
   1826 	if (newp == NULL || newlen != sizeof(int))
   1827 		return EINVAL;
   1828 
   1829 	size = 0;
   1830 	unit = *(const int *)newp;
   1831 	newp = NULL;
   1832 	newlen = 0;
   1833 	mutex_enter(&ccd_lock);
   1834 	LIST_FOREACH(sc, &ccds, sc_link)
   1835 		if (sc->sc_unit == unit) {
   1836 			for (size_t i = 0; i < sc->sc_nccdisks; i++)
   1837 				size += strlen(sc->sc_cinfo[i].ci_path) + 1;
   1838 			break;
   1839 		}
   1840 	mutex_exit(&ccd_lock);
   1841 
   1842 	if (size == 0)
   1843 		return ENOENT;
   1844 	names = kmem_zalloc(size, KM_SLEEP);
   1845 	if (names == NULL)
   1846 		return ENOMEM;
   1847 
   1848 	p = names;
   1849 	ep = names + size;
   1850 	mutex_enter(&ccd_lock);
   1851 	LIST_FOREACH(sc, &ccds, sc_link)
   1852 		if (sc->sc_unit == unit) {
   1853 			for (size_t i = 0; i < sc->sc_nccdisks; i++) {
   1854 				char *d = sc->sc_cinfo[i].ci_path;
   1855 				while (p < ep && (*p++ = *d++) != '\0')
   1856 					continue;
   1857 			}
   1858 			break;
   1859 		}
   1860 	mutex_exit(&ccd_lock);
   1861 
   1862 	node = *rnode;
   1863 	node.sysctl_data = names;
   1864 	node.sysctl_size = ep - names;
   1865 
   1866 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
   1867 	kmem_free(names, size);
   1868 	return error;
   1869 }
   1870 
   1871 SYSCTL_SETUP(sysctl_kern_ccd_setup, "sysctl kern.ccd subtree setup")
   1872 {
   1873 	const struct sysctlnode *node = NULL;
   1874 
   1875 	sysctl_createv(clog, 0, NULL, &node,
   1876 	    CTLFLAG_PERMANENT,
   1877 	    CTLTYPE_NODE, "ccd",
   1878 	    SYSCTL_DESCR("ConCatenated Disk state"),
   1879 	    NULL, 0, NULL, 0,
   1880 	    CTL_KERN, CTL_CREATE, CTL_EOL);
   1881 
   1882 	if (node == NULL)
   1883 		return;
   1884 
   1885 	sysctl_createv(clog, 0, &node, NULL,
   1886 	    CTLFLAG_PERMANENT | CTLFLAG_READONLY,
   1887 	    CTLTYPE_STRUCT, "units",
   1888 	    SYSCTL_DESCR("List of ccd unit numbers"),
   1889 	    ccd_units_sysctl, 0, NULL, 0,
   1890 	    CTL_CREATE, CTL_EOL);
   1891 	sysctl_createv(clog, 0, &node, NULL,
   1892 	    CTLFLAG_PERMANENT | CTLFLAG_READWRITE,
   1893 	    CTLTYPE_STRUCT, "info",
   1894 	    SYSCTL_DESCR("Information about a CCD unit"),
   1895 	    ccd_info_sysctl, 0, NULL, 0,
   1896 	    CTL_CREATE, CTL_EOL);
   1897 	sysctl_createv(clog, 0, &node, NULL,
   1898 	    CTLFLAG_PERMANENT | CTLFLAG_READWRITE,
   1899 	    CTLTYPE_STRUCT, "components",
   1900 	    SYSCTL_DESCR("Information about CCD components"),
   1901 	    ccd_components_sysctl, 0, NULL, 0,
   1902 	    CTL_CREATE, CTL_EOL);
   1903 }
   1904