xref: /netbsd/sys/dev/ccd.c (revision f9540ab9)
1 /*	$NetBSD: ccd.c,v 1.173 2017/12/19 03:24:09 pgoyette 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 1.173 2017/12/19 03:24:09 pgoyette 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 #include "ioconf.h"
131 
132 #if defined(CCDDEBUG) && !defined(DEBUG)
133 #define DEBUG
134 #endif
135 
136 #ifdef DEBUG
137 #define CCDB_FOLLOW	0x01
138 #define CCDB_INIT	0x02
139 #define CCDB_IO		0x04
140 #define CCDB_LABEL	0x08
141 #define CCDB_VNODE	0x10
142 int ccddebug = 0x00;
143 #endif
144 
145 #define	ccdunit(x)	DISKUNIT(x)
146 
147 struct ccdbuf {
148 	struct buf	cb_buf;		/* new I/O buf */
149 	struct buf	*cb_obp;	/* ptr. to original I/O buf */
150 	struct ccd_softc *cb_sc;	/* pointer to ccd softc */
151 	int		cb_comp;	/* target component */
152 	SIMPLEQ_ENTRY(ccdbuf) cb_q;	/* fifo of component buffers */
153 };
154 
155 /* component buffer pool */
156 static pool_cache_t ccd_cache;
157 
158 #define	CCD_GETBUF()		pool_cache_get(ccd_cache, PR_WAITOK)
159 #define	CCD_PUTBUF(cbp)		pool_cache_put(ccd_cache, cbp)
160 
161 #define CCDLABELDEV(dev)	\
162 	(MAKEDISKDEV(major((dev)), ccdunit((dev)), RAW_PART))
163 
164 /* called by main() at boot time */
165 void	ccddetach(void);
166 
167 /* called by biodone() at interrupt time */
168 static void	ccdiodone(struct buf *);
169 
170 static void	ccdinterleave(struct ccd_softc *);
171 static int	ccdinit(struct ccd_softc *, char **, struct vnode **,
172 		    struct lwp *);
173 static struct ccdbuf *ccdbuffer(struct ccd_softc *, struct buf *,
174 		    daddr_t, void *, long);
175 static void	ccdgetdefaultlabel(struct ccd_softc *, struct disklabel *);
176 static void	ccdgetdisklabel(dev_t);
177 static void	ccdmakedisklabel(struct ccd_softc *);
178 static void	ccdstart(struct ccd_softc *);
179 static void	ccdthread(void *);
180 
181 static dev_type_open(ccdopen);
182 static dev_type_close(ccdclose);
183 static dev_type_read(ccdread);
184 static dev_type_write(ccdwrite);
185 static dev_type_ioctl(ccdioctl);
186 static dev_type_strategy(ccdstrategy);
187 static dev_type_size(ccdsize);
188 
189 const struct bdevsw ccd_bdevsw = {
190 	.d_open = ccdopen,
191 	.d_close = ccdclose,
192 	.d_strategy = ccdstrategy,
193 	.d_ioctl = ccdioctl,
194 	.d_dump = nodump,
195 	.d_psize = ccdsize,
196 	.d_discard = nodiscard,
197 	.d_flag = D_DISK | D_MPSAFE
198 };
199 
200 const struct cdevsw ccd_cdevsw = {
201 	.d_open = ccdopen,
202 	.d_close = ccdclose,
203 	.d_read = ccdread,
204 	.d_write = ccdwrite,
205 	.d_ioctl = ccdioctl,
206 	.d_stop = nostop,
207 	.d_tty = notty,
208 	.d_poll = nopoll,
209 	.d_mmap = nommap,
210 	.d_kqfilter = nokqfilter,
211 	.d_discard = nodiscard,
212 	.d_flag = D_DISK | D_MPSAFE
213 };
214 
215 #ifdef DEBUG
216 static	void printiinfo(struct ccdiinfo *);
217 #endif
218 
219 static LIST_HEAD(, ccd_softc) ccds = LIST_HEAD_INITIALIZER(ccds);
220 static kmutex_t ccd_lock;
221 
222 #ifdef _MODULE
223 static struct sysctllog *ccd_clog;
224 #endif
225 
226 SYSCTL_SETUP_PROTO(sysctl_kern_ccd_setup);
227 
228 static struct ccd_softc *
229 ccdcreate(int unit) {
230 	struct ccd_softc *sc = kmem_zalloc(sizeof(*sc), KM_SLEEP);
231 
232 	/* Initialize per-softc structures. */
233 	snprintf(sc->sc_xname, sizeof(sc->sc_xname), "ccd%d", unit);
234 	sc->sc_unit = unit;
235 	mutex_init(&sc->sc_dvlock, MUTEX_DEFAULT, IPL_NONE);
236 	sc->sc_iolock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
237 	cv_init(&sc->sc_stop, "ccdstop");
238 	cv_init(&sc->sc_push, "ccdthr");
239 	disk_init(&sc->sc_dkdev, sc->sc_xname, NULL); /* XXX */
240 	return sc;
241 }
242 
243 static void
244 ccddestroy(struct ccd_softc *sc) {
245 	mutex_obj_free(sc->sc_iolock);
246 	mutex_exit(&sc->sc_dvlock);
247 	mutex_destroy(&sc->sc_dvlock);
248 	cv_destroy(&sc->sc_stop);
249 	cv_destroy(&sc->sc_push);
250 	disk_destroy(&sc->sc_dkdev);
251 	kmem_free(sc, sizeof(*sc));
252 }
253 
254 static struct ccd_softc *
255 ccdget(int unit, int make) {
256 	struct ccd_softc *sc;
257 	if (unit < 0) {
258 #ifdef DIAGNOSTIC
259 		panic("%s: unit %d!", __func__, unit);
260 #endif
261 		return NULL;
262 	}
263 	mutex_enter(&ccd_lock);
264 	LIST_FOREACH(sc, &ccds, sc_link) {
265 		if (sc->sc_unit == unit) {
266 			mutex_exit(&ccd_lock);
267 			return sc;
268 		}
269 	}
270 	mutex_exit(&ccd_lock);
271 	if (!make)
272 		return NULL;
273 	if ((sc = ccdcreate(unit)) == NULL)
274 		return NULL;
275 	mutex_enter(&ccd_lock);
276 	LIST_INSERT_HEAD(&ccds, sc, sc_link);
277 	mutex_exit(&ccd_lock);
278 	return sc;
279 }
280 
281 static void
282 ccdput(struct ccd_softc *sc) {
283 	mutex_enter(&ccd_lock);
284 	LIST_REMOVE(sc, sc_link);
285 	mutex_exit(&ccd_lock);
286 	ccddestroy(sc);
287 }
288 
289 /*
290  * Called by main() during pseudo-device attachment.  All we need
291  * to do is allocate enough space for devices to be configured later.
292  */
293 void
294 ccdattach(int num)
295 {
296 	mutex_init(&ccd_lock, MUTEX_DEFAULT, IPL_NONE);
297 
298 	/* Initialize the component buffer pool. */
299 	ccd_cache = pool_cache_init(sizeof(struct ccdbuf), 0,
300 	    0, 0, "ccdbuf", NULL, IPL_BIO, NULL, NULL, NULL);
301 }
302 
303 void
304 ccddetach(void)
305 {
306 	pool_cache_destroy(ccd_cache);
307 	mutex_destroy(&ccd_lock);
308 }
309 
310 static int
311 ccdinit(struct ccd_softc *cs, char **cpaths, struct vnode **vpp,
312     struct lwp *l)
313 {
314 	struct ccdcinfo *ci = NULL;
315 	int ix;
316 	struct ccdgeom *ccg = &cs->sc_geom;
317 	char *tmppath;
318 	int error, path_alloced;
319 	uint64_t psize, minsize;
320 	unsigned secsize, maxsecsize;
321 	struct disk_geom *dg;
322 
323 #ifdef DEBUG
324 	if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
325 		printf("%s: ccdinit\n", cs->sc_xname);
326 #endif
327 
328 	/* Allocate space for the component info. */
329 	cs->sc_cinfo = kmem_alloc(cs->sc_nccdisks * sizeof(*cs->sc_cinfo),
330 	    KM_SLEEP);
331 	tmppath = kmem_alloc(MAXPATHLEN, KM_SLEEP);
332 
333 	cs->sc_size = 0;
334 
335 	/*
336 	 * Verify that each component piece exists and record
337 	 * relevant information about it.
338 	 */
339 	maxsecsize = 0;
340 	minsize = 0;
341 	for (ix = 0, path_alloced = 0; ix < cs->sc_nccdisks; ix++) {
342 		ci = &cs->sc_cinfo[ix];
343 		ci->ci_vp = vpp[ix];
344 
345 		/*
346 		 * Copy in the pathname of the component.
347 		 */
348 		memset(tmppath, 0, MAXPATHLEN);	/* sanity */
349 		error = copyinstr(cpaths[ix], tmppath,
350 		    MAXPATHLEN, &ci->ci_pathlen);
351 		if (ci->ci_pathlen == 0)
352 			error = EINVAL;
353 		if (error) {
354 #ifdef DEBUG
355 			if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
356 				printf("%s: can't copy path, error = %d\n",
357 				    cs->sc_xname, error);
358 #endif
359 			goto out;
360 		}
361 		ci->ci_path = kmem_alloc(ci->ci_pathlen, KM_SLEEP);
362 		memcpy(ci->ci_path, tmppath, ci->ci_pathlen);
363 		path_alloced++;
364 
365 		/*
366 		 * XXX: Cache the component's dev_t.
367 		 */
368 		ci->ci_dev = vpp[ix]->v_rdev;
369 
370 		/*
371 		 * Get partition information for the component.
372 		 */
373 		error = getdisksize(vpp[ix], &psize, &secsize);
374 		if (error) {
375 #ifdef DEBUG
376 			if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
377 				 printf("%s: %s: disksize failed, error = %d\n",
378 				     cs->sc_xname, ci->ci_path, error);
379 #endif
380 			goto out;
381 		}
382 
383 		/*
384 		 * Calculate the size, truncating to an interleave
385 		 * boundary if necessary.
386 		 */
387 		maxsecsize = secsize > maxsecsize ? secsize : maxsecsize;
388 		if (cs->sc_ileave > 1)
389 			psize -= psize % cs->sc_ileave;
390 
391 		if (psize == 0) {
392 #ifdef DEBUG
393 			if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
394 				printf("%s: %s: size == 0\n",
395 				    cs->sc_xname, ci->ci_path);
396 #endif
397 			error = ENODEV;
398 			goto out;
399 		}
400 
401 		if (minsize == 0 || psize < minsize)
402 			minsize = psize;
403 		ci->ci_size = psize;
404 		cs->sc_size += psize;
405 	}
406 
407 	/*
408 	 * Don't allow the interleave to be smaller than
409 	 * the biggest component sector.
410 	 */
411 	if ((cs->sc_ileave > 0) &&
412 	    (cs->sc_ileave < (maxsecsize / DEV_BSIZE))) {
413 #ifdef DEBUG
414 		if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
415 			printf("%s: interleave must be at least %d\n",
416 			    cs->sc_xname, (maxsecsize / DEV_BSIZE));
417 #endif
418 		error = EINVAL;
419 		goto out;
420 	}
421 
422 	/*
423 	 * If uniform interleave is desired set all sizes to that of
424 	 * the smallest component.
425 	 */
426 	if (cs->sc_flags & CCDF_UNIFORM) {
427 		for (ci = cs->sc_cinfo;
428 		     ci < &cs->sc_cinfo[cs->sc_nccdisks]; ci++)
429 			ci->ci_size = minsize;
430 
431 		cs->sc_size = cs->sc_nccdisks * minsize;
432 	}
433 
434 	/*
435 	 * Construct the interleave table.
436 	 */
437 	ccdinterleave(cs);
438 
439 	/*
440 	 * Create pseudo-geometry based on 1MB cylinders.  It's
441 	 * pretty close.
442 	 */
443 	ccg->ccg_secsize = DEV_BSIZE;
444 	ccg->ccg_ntracks = 1;
445 	ccg->ccg_nsectors = 1024 * (1024 / ccg->ccg_secsize);
446 	ccg->ccg_ncylinders = cs->sc_size / ccg->ccg_nsectors;
447 
448         dg = &cs->sc_dkdev.dk_geom;
449         memset(dg, 0, sizeof(*dg));
450 	dg->dg_secperunit = cs->sc_size;
451 	dg->dg_secsize = ccg->ccg_secsize;
452 	dg->dg_nsectors = ccg->ccg_nsectors;
453 	dg->dg_ntracks = ccg->ccg_ntracks;
454 	dg->dg_ncylinders = ccg->ccg_ncylinders;
455 
456 	if (cs->sc_ileave > 0)
457 	        aprint_normal("%s: Interleaving %d component%s "
458 	            "(%d block interleave)\n", cs->sc_xname,
459         	    cs->sc_nccdisks, (cs->sc_nccdisks != 0 ? "s" : ""),
460         	    cs->sc_ileave);
461 	else
462 	        aprint_normal("%s: Concatenating %d component%s\n",
463 	            cs->sc_xname,
464         	    cs->sc_nccdisks, (cs->sc_nccdisks != 0 ? "s" : ""));
465 	for (ix = 0; ix < cs->sc_nccdisks; ix++) {
466 		ci = &cs->sc_cinfo[ix];
467 		aprint_normal("%s: %s (%ju blocks)\n", cs->sc_xname,
468 		    ci->ci_path, (uintmax_t)ci->ci_size);
469 	}
470 	aprint_normal("%s: total %ju blocks\n", cs->sc_xname, cs->sc_size);
471 
472 	/*
473 	 * Create thread to handle deferred I/O.
474 	 */
475 	cs->sc_zap = false;
476 	error = kthread_create(PRI_BIO, KTHREAD_MPSAFE, NULL, ccdthread,
477 	    cs, &cs->sc_thread, "%s", cs->sc_xname);
478 	if (error) {
479 		printf("ccdinit: can't create thread: %d\n", error);
480 		goto out;
481 	}
482 
483 	/*
484 	 * Only now that everything is set up can we enable the device.
485 	 */
486 	mutex_enter(cs->sc_iolock);
487 	cs->sc_flags |= CCDF_INITED;
488 	mutex_exit(cs->sc_iolock);
489 	kmem_free(tmppath, MAXPATHLEN);
490 	return (0);
491 
492  out:
493 	for (ix = 0; ix < path_alloced; ix++) {
494 		kmem_free(cs->sc_cinfo[ix].ci_path,
495 		    cs->sc_cinfo[ix].ci_pathlen);
496 	}
497 	kmem_free(cs->sc_cinfo, cs->sc_nccdisks * sizeof(struct ccdcinfo));
498 	kmem_free(tmppath, MAXPATHLEN);
499 	return (error);
500 }
501 
502 static void
503 ccdinterleave(struct ccd_softc *cs)
504 {
505 	struct ccdcinfo *ci, *smallci;
506 	struct ccdiinfo *ii;
507 	daddr_t bn, lbn;
508 	int ix;
509 	u_long size;
510 
511 #ifdef DEBUG
512 	if (ccddebug & CCDB_INIT)
513 		printf("ccdinterleave(%p): ileave %d\n", cs, cs->sc_ileave);
514 #endif
515 	/*
516 	 * Allocate an interleave table.
517 	 * Chances are this is too big, but we don't care.
518 	 */
519 	size = (cs->sc_nccdisks + 1) * sizeof(struct ccdiinfo);
520 	cs->sc_itable = kmem_zalloc(size, KM_SLEEP);
521 
522 	/*
523 	 * Trivial case: no interleave (actually interleave of disk size).
524 	 * Each table entry represents a single component in its entirety.
525 	 */
526 	if (cs->sc_ileave == 0) {
527 		bn = 0;
528 		ii = cs->sc_itable;
529 
530 		for (ix = 0; ix < cs->sc_nccdisks; ix++) {
531 			/* Allocate space for ii_index. */
532 			ii->ii_indexsz = sizeof(int);
533 			ii->ii_index = kmem_alloc(ii->ii_indexsz, KM_SLEEP);
534 			ii->ii_ndisk = 1;
535 			ii->ii_startblk = bn;
536 			ii->ii_startoff = 0;
537 			ii->ii_index[0] = ix;
538 			bn += cs->sc_cinfo[ix].ci_size;
539 			ii++;
540 		}
541 		ii->ii_ndisk = 0;
542 #ifdef DEBUG
543 		if (ccddebug & CCDB_INIT)
544 			printiinfo(cs->sc_itable);
545 #endif
546 		return;
547 	}
548 
549 	/*
550 	 * The following isn't fast or pretty; it doesn't have to be.
551 	 */
552 	size = 0;
553 	bn = lbn = 0;
554 	for (ii = cs->sc_itable; ; ii++) {
555 		/* Allocate space for ii_index. */
556 		ii->ii_indexsz = sizeof(int) * cs->sc_nccdisks;
557 		ii->ii_index = kmem_alloc(ii->ii_indexsz, KM_SLEEP);
558 
559 		/*
560 		 * Locate the smallest of the remaining components
561 		 */
562 		smallci = NULL;
563 		for (ci = cs->sc_cinfo;
564 		     ci < &cs->sc_cinfo[cs->sc_nccdisks]; ci++)
565 			if (ci->ci_size > size &&
566 			    (smallci == NULL ||
567 			     ci->ci_size < smallci->ci_size))
568 				smallci = ci;
569 
570 		/*
571 		 * Nobody left, all done
572 		 */
573 		if (smallci == NULL) {
574 			ii->ii_ndisk = 0;
575 			break;
576 		}
577 
578 		/*
579 		 * Record starting logical block and component offset
580 		 */
581 		ii->ii_startblk = bn / cs->sc_ileave;
582 		ii->ii_startoff = lbn;
583 
584 		/*
585 		 * Determine how many disks take part in this interleave
586 		 * and record their indices.
587 		 */
588 		ix = 0;
589 		for (ci = cs->sc_cinfo;
590 		     ci < &cs->sc_cinfo[cs->sc_nccdisks]; ci++)
591 			if (ci->ci_size >= smallci->ci_size)
592 				ii->ii_index[ix++] = ci - cs->sc_cinfo;
593 		ii->ii_ndisk = ix;
594 		bn += ix * (smallci->ci_size - size);
595 		lbn = smallci->ci_size / cs->sc_ileave;
596 		size = smallci->ci_size;
597 	}
598 #ifdef DEBUG
599 	if (ccddebug & CCDB_INIT)
600 		printiinfo(cs->sc_itable);
601 #endif
602 }
603 
604 /* ARGSUSED */
605 static int
606 ccdopen(dev_t dev, int flags, int fmt, struct lwp *l)
607 {
608 	int unit = ccdunit(dev);
609 	struct ccd_softc *cs;
610 	struct disklabel *lp;
611 	int error = 0, part, pmask;
612 
613 #ifdef DEBUG
614 	if (ccddebug & CCDB_FOLLOW)
615 		printf("ccdopen(0x%"PRIx64", 0x%x)\n", dev, flags);
616 #endif
617 	if ((cs = ccdget(unit, 1)) == NULL)
618 		return ENXIO;
619 
620 	mutex_enter(&cs->sc_dvlock);
621 
622 	lp = cs->sc_dkdev.dk_label;
623 
624 	part = DISKPART(dev);
625 	pmask = (1 << part);
626 
627 	/*
628 	 * If we're initialized, check to see if there are any other
629 	 * open partitions.  If not, then it's safe to update
630 	 * the in-core disklabel.  Only read the disklabel if it is
631 	 * not already valid.
632 	 */
633 	if ((cs->sc_flags & (CCDF_INITED|CCDF_VLABEL)) == CCDF_INITED &&
634 	    cs->sc_dkdev.dk_openmask == 0)
635 		ccdgetdisklabel(dev);
636 
637 	/* Check that the partition exists. */
638 	if (part != RAW_PART) {
639 		if (((cs->sc_flags & CCDF_INITED) == 0) ||
640 		    ((part >= lp->d_npartitions) ||
641 		     (lp->d_partitions[part].p_fstype == FS_UNUSED))) {
642 			error = ENXIO;
643 			goto done;
644 		}
645 	}
646 
647 	/* Prevent our unit from being unconfigured while open. */
648 	switch (fmt) {
649 	case S_IFCHR:
650 		cs->sc_dkdev.dk_copenmask |= pmask;
651 		break;
652 
653 	case S_IFBLK:
654 		cs->sc_dkdev.dk_bopenmask |= pmask;
655 		break;
656 	}
657 	cs->sc_dkdev.dk_openmask =
658 	    cs->sc_dkdev.dk_copenmask | cs->sc_dkdev.dk_bopenmask;
659 
660  done:
661 	mutex_exit(&cs->sc_dvlock);
662 	return (error);
663 }
664 
665 /* ARGSUSED */
666 static int
667 ccdclose(dev_t dev, int flags, int fmt, struct lwp *l)
668 {
669 	int unit = ccdunit(dev);
670 	struct ccd_softc *cs;
671 	int part;
672 
673 #ifdef DEBUG
674 	if (ccddebug & CCDB_FOLLOW)
675 		printf("ccdclose(0x%"PRIx64", 0x%x)\n", dev, flags);
676 #endif
677 
678 	if ((cs = ccdget(unit, 0)) == NULL)
679 		return ENXIO;
680 
681 	mutex_enter(&cs->sc_dvlock);
682 
683 	part = DISKPART(dev);
684 
685 	/* ...that much closer to allowing unconfiguration... */
686 	switch (fmt) {
687 	case S_IFCHR:
688 		cs->sc_dkdev.dk_copenmask &= ~(1 << part);
689 		break;
690 
691 	case S_IFBLK:
692 		cs->sc_dkdev.dk_bopenmask &= ~(1 << part);
693 		break;
694 	}
695 	cs->sc_dkdev.dk_openmask =
696 	    cs->sc_dkdev.dk_copenmask | cs->sc_dkdev.dk_bopenmask;
697 
698 	if (cs->sc_dkdev.dk_openmask == 0) {
699 		if ((cs->sc_flags & CCDF_KLABEL) == 0)
700 			cs->sc_flags &= ~CCDF_VLABEL;
701 	}
702 
703 	mutex_exit(&cs->sc_dvlock);
704 	return (0);
705 }
706 
707 static bool
708 ccdbackoff(struct ccd_softc *cs)
709 {
710 
711 	/* XXX Arbitrary, should be a uvm call. */
712 	return uvmexp.free < (uvmexp.freemin >> 1) &&
713 	    disk_isbusy(&cs->sc_dkdev);
714 }
715 
716 static void
717 ccdthread(void *cookie)
718 {
719 	struct ccd_softc *cs;
720 
721 	cs = cookie;
722 
723 #ifdef DEBUG
724  	if (ccddebug & CCDB_FOLLOW)
725  		printf("ccdthread: hello\n");
726 #endif
727 
728 	mutex_enter(cs->sc_iolock);
729 	while (__predict_true(!cs->sc_zap)) {
730 		if (bufq_peek(cs->sc_bufq) == NULL) {
731 			/* Nothing to do. */
732 			cv_wait(&cs->sc_push, cs->sc_iolock);
733 			continue;
734 		}
735 		if (ccdbackoff(cs)) {
736 			/* Wait for memory to become available. */
737 			(void)cv_timedwait(&cs->sc_push, cs->sc_iolock, 1);
738 			continue;
739 		}
740 #ifdef DEBUG
741  		if (ccddebug & CCDB_FOLLOW)
742  			printf("ccdthread: dispatching I/O\n");
743 #endif
744 		ccdstart(cs);
745 		mutex_enter(cs->sc_iolock);
746 	}
747 	cs->sc_thread = NULL;
748 	mutex_exit(cs->sc_iolock);
749 #ifdef DEBUG
750  	if (ccddebug & CCDB_FOLLOW)
751  		printf("ccdthread: goodbye\n");
752 #endif
753 	kthread_exit(0);
754 }
755 
756 static void
757 ccdstrategy(struct buf *bp)
758 {
759 	int unit = ccdunit(bp->b_dev);
760 	struct ccd_softc *cs;
761 	if ((cs = ccdget(unit, 0)) == NULL)
762 		return;
763 
764 	/* Must be open or reading label. */
765 	KASSERT(cs->sc_dkdev.dk_openmask != 0 ||
766 	    (cs->sc_flags & CCDF_RLABEL) != 0);
767 
768 	mutex_enter(cs->sc_iolock);
769 	/* Synchronize with device init/uninit. */
770 	if (__predict_false((cs->sc_flags & CCDF_INITED) == 0)) {
771 		mutex_exit(cs->sc_iolock);
772 #ifdef DEBUG
773  		if (ccddebug & CCDB_FOLLOW)
774  			printf("ccdstrategy: unit %d: not inited\n", unit);
775 #endif
776  		bp->b_error = ENXIO;
777  		bp->b_resid = bp->b_bcount;
778  		biodone(bp);
779 		return;
780 	}
781 
782 	/* Defer to thread if system is low on memory. */
783 	bufq_put(cs->sc_bufq, bp);
784 	if (__predict_false(ccdbackoff(cs))) {
785 		mutex_exit(cs->sc_iolock);
786 #ifdef DEBUG
787  		if (ccddebug & CCDB_FOLLOW)
788  			printf("ccdstrategy: holding off on I/O\n");
789 #endif
790 		return;
791 	}
792 	ccdstart(cs);
793 }
794 
795 static void
796 ccdstart(struct ccd_softc *cs)
797 {
798 	daddr_t blkno;
799 	int wlabel;
800 	struct disklabel *lp;
801 	long bcount, rcount;
802 	struct ccdbuf *cbp;
803 	char *addr;
804 	daddr_t bn;
805 	vnode_t *vp;
806 	buf_t *bp;
807 
808 	KASSERT(mutex_owned(cs->sc_iolock));
809 
810 	bp = bufq_get(cs->sc_bufq);
811 	KASSERT(bp != NULL);
812 
813 	disk_busy(&cs->sc_dkdev);
814 
815 #ifdef DEBUG
816 	if (ccddebug & CCDB_FOLLOW)
817 		printf("ccdstart(%s, %p)\n", cs->sc_xname, bp);
818 #endif
819 
820 	/* If it's a nil transfer, wake up the top half now. */
821 	if (bp->b_bcount == 0)
822 		goto done;
823 
824 	lp = cs->sc_dkdev.dk_label;
825 
826 	/*
827 	 * Do bounds checking and adjust transfer.  If there's an
828 	 * error, the bounds check will flag that for us.  Convert
829 	 * the partition relative block number to an absolute.
830 	 */
831 	blkno = bp->b_blkno;
832 	wlabel = cs->sc_flags & (CCDF_WLABEL|CCDF_LABELLING);
833 	if (DISKPART(bp->b_dev) != RAW_PART) {
834 		if (bounds_check_with_label(&cs->sc_dkdev, bp, wlabel) <= 0)
835 			goto done;
836 		blkno += lp->d_partitions[DISKPART(bp->b_dev)].p_offset;
837 	}
838 	mutex_exit(cs->sc_iolock);
839 	bp->b_rawblkno = blkno;
840 
841 	/* Allocate the component buffers and start I/O! */
842 	bp->b_resid = bp->b_bcount;
843 	bn = bp->b_rawblkno;
844 	addr = bp->b_data;
845 	for (bcount = bp->b_bcount; bcount > 0; bcount -= rcount) {
846 		cbp = ccdbuffer(cs, bp, bn, addr, bcount);
847 		rcount = cbp->cb_buf.b_bcount;
848 		bn += btodb(rcount);
849 		addr += rcount;
850 		vp = cbp->cb_buf.b_vp;
851 		if ((cbp->cb_buf.b_flags & B_READ) == 0) {
852 			mutex_enter(vp->v_interlock);
853 			vp->v_numoutput++;
854 			mutex_exit(vp->v_interlock);
855 		}
856 		(void)VOP_STRATEGY(vp, &cbp->cb_buf);
857 	}
858 	return;
859 
860  done:
861 	disk_unbusy(&cs->sc_dkdev, 0, 0);
862 	cv_broadcast(&cs->sc_stop);
863 	cv_broadcast(&cs->sc_push);
864 	mutex_exit(cs->sc_iolock);
865 	bp->b_resid = bp->b_bcount;
866 	biodone(bp);
867 }
868 
869 /*
870  * Build a component buffer header.
871  */
872 static struct ccdbuf *
873 ccdbuffer(struct ccd_softc *cs, struct buf *bp, daddr_t bn, void *addr,
874     long bcount)
875 {
876 	struct ccdcinfo *ci;
877 	struct ccdbuf *cbp;
878 	daddr_t cbn, cboff;
879 	u_int64_t cbc;
880 	int ccdisk;
881 
882 #ifdef DEBUG
883 	if (ccddebug & CCDB_IO)
884 		printf("ccdbuffer(%p, %p, %" PRId64 ", %p, %ld)\n",
885 		       cs, bp, bn, addr, bcount);
886 #endif
887 	/*
888 	 * Determine which component bn falls in.
889 	 */
890 	cbn = bn;
891 	cboff = 0;
892 
893 	/*
894 	 * Serially concatenated
895 	 */
896 	if (cs->sc_ileave == 0) {
897 		daddr_t sblk;
898 
899 		sblk = 0;
900 		for (ccdisk = 0, ci = &cs->sc_cinfo[ccdisk];
901 		    cbn >= sblk + ci->ci_size;
902 		    ccdisk++, ci = &cs->sc_cinfo[ccdisk])
903 			sblk += ci->ci_size;
904 		cbn -= sblk;
905 	}
906 	/*
907 	 * Interleaved
908 	 */
909 	else {
910 		struct ccdiinfo *ii;
911 		int off;
912 
913 		cboff = cbn % cs->sc_ileave;
914 		cbn /= cs->sc_ileave;
915 		for (ii = cs->sc_itable; ii->ii_ndisk; ii++)
916 			if (ii->ii_startblk > cbn)
917 				break;
918 		ii--;
919 		off = cbn - ii->ii_startblk;
920 		if (ii->ii_ndisk == 1) {
921 			ccdisk = ii->ii_index[0];
922 			cbn = ii->ii_startoff + off;
923 		} else {
924 			ccdisk = ii->ii_index[off % ii->ii_ndisk];
925 			cbn = ii->ii_startoff + off / ii->ii_ndisk;
926 		}
927 		cbn *= cs->sc_ileave;
928 		ci = &cs->sc_cinfo[ccdisk];
929 	}
930 
931 	/*
932 	 * Fill in the component buf structure.
933 	 */
934 	cbp = CCD_GETBUF();
935 	KASSERT(cbp != NULL);
936 	buf_init(&cbp->cb_buf);
937 	cbp->cb_buf.b_flags = bp->b_flags;
938 	cbp->cb_buf.b_oflags = bp->b_oflags;
939 	cbp->cb_buf.b_cflags = bp->b_cflags;
940 	cbp->cb_buf.b_iodone = ccdiodone;
941 	cbp->cb_buf.b_proc = bp->b_proc;
942 	cbp->cb_buf.b_dev = ci->ci_dev;
943 	cbp->cb_buf.b_blkno = cbn + cboff;
944 	cbp->cb_buf.b_data = addr;
945 	cbp->cb_buf.b_vp = ci->ci_vp;
946 	cbp->cb_buf.b_objlock = ci->ci_vp->v_interlock;
947 	if (cs->sc_ileave == 0)
948 		cbc = dbtob((u_int64_t)(ci->ci_size - cbn));
949 	else
950 		cbc = dbtob((u_int64_t)(cs->sc_ileave - cboff));
951 	cbp->cb_buf.b_bcount = cbc < bcount ? cbc : bcount;
952 
953 	/*
954 	 * context for ccdiodone
955 	 */
956 	cbp->cb_obp = bp;
957 	cbp->cb_sc = cs;
958 	cbp->cb_comp = ccdisk;
959 
960 	BIO_COPYPRIO(&cbp->cb_buf, bp);
961 
962 #ifdef DEBUG
963 	if (ccddebug & CCDB_IO)
964 		printf(" dev 0x%"PRIx64"(u%lu): cbp %p bn %" PRId64 " addr %p"
965 		       " bcnt %d\n",
966 		    ci->ci_dev, (unsigned long) (ci-cs->sc_cinfo), cbp,
967 		    cbp->cb_buf.b_blkno, cbp->cb_buf.b_data,
968 		    cbp->cb_buf.b_bcount);
969 #endif
970 
971 	return (cbp);
972 }
973 
974 /*
975  * Called at interrupt time.
976  * Mark the component as done and if all components are done,
977  * take a ccd interrupt.
978  */
979 static void
980 ccdiodone(struct buf *vbp)
981 {
982 	struct ccdbuf *cbp = (struct ccdbuf *) vbp;
983 	struct buf *bp = cbp->cb_obp;
984 	struct ccd_softc *cs = cbp->cb_sc;
985 	int count;
986 
987 #ifdef DEBUG
988 	if (ccddebug & CCDB_FOLLOW)
989 		printf("ccdiodone(%p)\n", cbp);
990 	if (ccddebug & CCDB_IO) {
991 		printf("ccdiodone: bp %p bcount %d resid %d\n",
992 		       bp, bp->b_bcount, bp->b_resid);
993 		printf(" dev 0x%"PRIx64"(u%d), cbp %p bn %" PRId64 " addr %p"
994 		       " bcnt %d\n",
995 		       cbp->cb_buf.b_dev, cbp->cb_comp, cbp,
996 		       cbp->cb_buf.b_blkno, cbp->cb_buf.b_data,
997 		       cbp->cb_buf.b_bcount);
998 	}
999 #endif
1000 
1001 	if (cbp->cb_buf.b_error != 0) {
1002 		bp->b_error = cbp->cb_buf.b_error;
1003 		printf("%s: error %d on component %d\n",
1004 		       cs->sc_xname, bp->b_error, cbp->cb_comp);
1005 	}
1006 	count = cbp->cb_buf.b_bcount;
1007 	buf_destroy(&cbp->cb_buf);
1008 	CCD_PUTBUF(cbp);
1009 
1010 	/*
1011 	 * If all done, "interrupt".
1012 	 */
1013 	mutex_enter(cs->sc_iolock);
1014 	bp->b_resid -= count;
1015 	if (bp->b_resid < 0)
1016 		panic("ccdiodone: count");
1017 	if (bp->b_resid == 0) {
1018 		/*
1019 		 * Request is done for better or worse, wakeup the top half.
1020 		 */
1021 		if (bp->b_error != 0)
1022 			bp->b_resid = bp->b_bcount;
1023 		disk_unbusy(&cs->sc_dkdev, (bp->b_bcount - bp->b_resid),
1024 		    (bp->b_flags & B_READ));
1025 		if (!disk_isbusy(&cs->sc_dkdev)) {
1026 			if (bufq_peek(cs->sc_bufq) != NULL) {
1027 				cv_broadcast(&cs->sc_push);
1028 			}
1029 			cv_broadcast(&cs->sc_stop);
1030 		}
1031 		mutex_exit(cs->sc_iolock);
1032 		biodone(bp);
1033 	} else
1034 		mutex_exit(cs->sc_iolock);
1035 }
1036 
1037 /* ARGSUSED */
1038 static int
1039 ccdread(dev_t dev, struct uio *uio, int flags)
1040 {
1041 	int unit = ccdunit(dev);
1042 	struct ccd_softc *cs;
1043 
1044 #ifdef DEBUG
1045 	if (ccddebug & CCDB_FOLLOW)
1046 		printf("ccdread(0x%"PRIx64", %p)\n", dev, uio);
1047 #endif
1048 	if ((cs = ccdget(unit, 0)) == NULL)
1049 		return 0;
1050 
1051 	/* Unlocked advisory check, ccdstrategy check is synchronous. */
1052 	if ((cs->sc_flags & CCDF_INITED) == 0)
1053 		return (ENXIO);
1054 
1055 	return (physio(ccdstrategy, NULL, dev, B_READ, minphys, uio));
1056 }
1057 
1058 /* ARGSUSED */
1059 static int
1060 ccdwrite(dev_t dev, struct uio *uio, int flags)
1061 {
1062 	int unit = ccdunit(dev);
1063 	struct ccd_softc *cs;
1064 
1065 #ifdef DEBUG
1066 	if (ccddebug & CCDB_FOLLOW)
1067 		printf("ccdwrite(0x%"PRIx64", %p)\n", dev, uio);
1068 #endif
1069 	if ((cs = ccdget(unit, 0)) == NULL)
1070 		return ENOENT;
1071 
1072 	/* Unlocked advisory check, ccdstrategy check is synchronous. */
1073 	if ((cs->sc_flags & CCDF_INITED) == 0)
1074 		return (ENXIO);
1075 
1076 	return (physio(ccdstrategy, NULL, dev, B_WRITE, minphys, uio));
1077 }
1078 
1079 static int
1080 ccdioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l)
1081 {
1082 	int unit = ccdunit(dev);
1083 	int i, j, lookedup = 0, error = 0;
1084 	int part, pmask, make;
1085 	struct ccd_softc *cs;
1086 	struct ccd_ioctl *ccio = (struct ccd_ioctl *)data;
1087 	kauth_cred_t uc;
1088 	char **cpp;
1089 	struct pathbuf *pb;
1090 	struct vnode **vpp;
1091 #ifdef __HAVE_OLD_DISKLABEL
1092 	struct disklabel newlabel;
1093 #endif
1094 
1095 	switch (cmd) {
1096 #if defined(COMPAT_60) && !defined(_LP64)
1097 	case CCDIOCSET_60:
1098 #endif
1099 	case CCDIOCSET:
1100 		make = 1;
1101 		break;
1102 	default:
1103 		make = 0;
1104 		break;
1105 	}
1106 
1107 	if ((cs = ccdget(unit, make)) == NULL)
1108 		return ENOENT;
1109 	uc = kauth_cred_get();
1110 
1111 /*
1112  * Compat code must not be called if on a platform where
1113  * sizeof (size_t) == sizeof (uint64_t) as CCDIOCSET will
1114  * be the same as CCDIOCSET_60
1115  */
1116 #if defined(COMPAT_60) && !defined(_LP64)
1117 	switch (cmd) {
1118 	case CCDIOCSET_60: {
1119 		struct ccd_ioctl ccionew;
1120        		struct ccd_ioctl_60 *ccio60 =
1121        		    (struct ccd_ioctl_60 *)data;
1122 		ccionew.ccio_disks = ccio->ccio_disks;
1123 		ccionew.ccio_ndisks = ccio->ccio_ndisks;
1124 		ccionew.ccio_ileave = ccio->ccio_ileave;
1125 		ccionew.ccio_flags = ccio->ccio_flags;
1126 		ccionew.ccio_unit = ccio->ccio_unit;
1127 		error = ccdioctl(dev, CCDIOCSET, &ccionew, flag, l);
1128 		if (!error) {
1129 			/* Copy data back, adjust types if necessary */
1130 			ccio60->ccio_disks = ccionew.ccio_disks;
1131 			ccio60->ccio_ndisks = ccionew.ccio_ndisks;
1132 			ccio60->ccio_ileave = ccionew.ccio_ileave;
1133 			ccio60->ccio_flags = ccionew.ccio_flags;
1134 			ccio60->ccio_unit = ccionew.ccio_unit;
1135 			ccio60->ccio_size = (size_t)ccionew.ccio_size;
1136 		}
1137 		return error;
1138 		}
1139 		break;
1140 
1141 	case CCDIOCCLR_60:
1142 		/*
1143 		 * ccio_size member not used, so existing struct OK
1144 		 * drop through to existing non-compat version
1145 		 */
1146 		cmd = CCDIOCCLR;
1147 		break;
1148 	}
1149 #endif /* COMPAT_60 && !_LP64*/
1150 
1151 	/* Must be open for writes for these commands... */
1152 	switch (cmd) {
1153 	case CCDIOCSET:
1154 	case CCDIOCCLR:
1155 	case DIOCSDINFO:
1156 	case DIOCWDINFO:
1157 	case DIOCCACHESYNC:
1158 	case DIOCAWEDGE:
1159 	case DIOCDWEDGE:
1160 	case DIOCMWEDGES:
1161 #ifdef __HAVE_OLD_DISKLABEL
1162 	case ODIOCSDINFO:
1163 	case ODIOCWDINFO:
1164 #endif
1165 	case DIOCKLABEL:
1166 	case DIOCWLABEL:
1167 		if ((flag & FWRITE) == 0)
1168 			return (EBADF);
1169 	}
1170 
1171 	mutex_enter(&cs->sc_dvlock);
1172 
1173 	/* Must be initialized for these... */
1174 	switch (cmd) {
1175 	case CCDIOCCLR:
1176 	case DIOCGDINFO:
1177 	case DIOCGSTRATEGY:
1178 	case DIOCGCACHE:
1179 	case DIOCCACHESYNC:
1180 	case DIOCAWEDGE:
1181 	case DIOCDWEDGE:
1182 	case DIOCLWEDGES:
1183 	case DIOCMWEDGES:
1184 	case DIOCSDINFO:
1185 	case DIOCWDINFO:
1186 	case DIOCGPARTINFO:
1187 	case DIOCWLABEL:
1188 	case DIOCKLABEL:
1189 	case DIOCGDEFLABEL:
1190 #ifdef __HAVE_OLD_DISKLABEL
1191 	case ODIOCGDINFO:
1192 	case ODIOCSDINFO:
1193 	case ODIOCWDINFO:
1194 	case ODIOCGDEFLABEL:
1195 #endif
1196 		if ((cs->sc_flags & CCDF_INITED) == 0) {
1197 			error = ENXIO;
1198 			goto out;
1199 		}
1200 	}
1201 
1202 	error = disk_ioctl(&cs->sc_dkdev, dev, cmd, data, flag, l);
1203 	if (error != EPASSTHROUGH)
1204 		goto out;
1205 
1206 	error = 0;
1207 	switch (cmd) {
1208 	case CCDIOCSET:
1209 		if (cs->sc_flags & CCDF_INITED) {
1210 			error = EBUSY;
1211 			goto out;
1212 		}
1213 
1214 		/* Validate the flags. */
1215 		if ((ccio->ccio_flags & CCDF_USERMASK) != ccio->ccio_flags) {
1216 			error = EINVAL;
1217 			goto out;
1218 		}
1219 
1220 		if (ccio->ccio_ndisks > CCD_MAXNDISKS ||
1221 		    ccio->ccio_ndisks == 0) {
1222 			error = EINVAL;
1223 			goto out;
1224 		}
1225 
1226 		/* Fill in some important bits. */
1227 		cs->sc_ileave = ccio->ccio_ileave;
1228 		cs->sc_nccdisks = ccio->ccio_ndisks;
1229 		cs->sc_flags = ccio->ccio_flags & CCDF_USERMASK;
1230 
1231 		/*
1232 		 * Allocate space for and copy in the array of
1233 		 * component pathnames and device numbers.
1234 		 */
1235 		cpp = kmem_alloc(ccio->ccio_ndisks * sizeof(*cpp), KM_SLEEP);
1236 		vpp = kmem_alloc(ccio->ccio_ndisks * sizeof(*vpp), KM_SLEEP);
1237 		error = copyin(ccio->ccio_disks, cpp,
1238 		    ccio->ccio_ndisks * sizeof(*cpp));
1239 		if (error) {
1240 			kmem_free(vpp, ccio->ccio_ndisks * sizeof(*vpp));
1241 			kmem_free(cpp, ccio->ccio_ndisks * sizeof(*cpp));
1242 			goto out;
1243 		}
1244 
1245 #ifdef DEBUG
1246 		if (ccddebug & CCDB_INIT)
1247 			for (i = 0; i < ccio->ccio_ndisks; ++i)
1248 				printf("ccdioctl: component %d: %p\n",
1249 				    i, cpp[i]);
1250 #endif
1251 
1252 		for (i = 0; i < ccio->ccio_ndisks; ++i) {
1253 #ifdef DEBUG
1254 			if (ccddebug & CCDB_INIT)
1255 				printf("ccdioctl: lookedup = %d\n", lookedup);
1256 #endif
1257 			error = pathbuf_copyin(cpp[i], &pb);
1258 			if (error == 0) {
1259 				error = dk_lookup(pb, l, &vpp[i]);
1260 			}
1261 			pathbuf_destroy(pb);
1262 			if (error != 0) {
1263 				for (j = 0; j < lookedup; ++j)
1264 					(void)vn_close(vpp[j], FREAD|FWRITE,
1265 					    uc);
1266 				kmem_free(vpp, ccio->ccio_ndisks *
1267 				    sizeof(*vpp));
1268 				kmem_free(cpp, ccio->ccio_ndisks *
1269 				    sizeof(*cpp));
1270 				goto out;
1271 			}
1272 			++lookedup;
1273 		}
1274 
1275 		/* Attach the disk. */
1276 		disk_attach(&cs->sc_dkdev);
1277 		bufq_alloc(&cs->sc_bufq, "fcfs", 0);
1278 
1279 		/*
1280 		 * Initialize the ccd.  Fills in the softc for us.
1281 		 */
1282 		if ((error = ccdinit(cs, cpp, vpp, l)) != 0) {
1283 			for (j = 0; j < lookedup; ++j)
1284 				(void)vn_close(vpp[j], FREAD|FWRITE,
1285 				    uc);
1286 			kmem_free(vpp, ccio->ccio_ndisks * sizeof(*vpp));
1287 			kmem_free(cpp, ccio->ccio_ndisks * sizeof(*cpp));
1288 			disk_detach(&cs->sc_dkdev);
1289 			mutex_exit(&cs->sc_dvlock);
1290 			bufq_free(cs->sc_bufq);
1291 			return error;
1292 		}
1293 
1294 		/* We can free the temporary variables now. */
1295 		kmem_free(vpp, ccio->ccio_ndisks * sizeof(*vpp));
1296 		kmem_free(cpp, ccio->ccio_ndisks * sizeof(*cpp));
1297 
1298 		/*
1299 		 * The ccd has been successfully initialized, so
1300 		 * we can place it into the array.  Don't try to
1301 		 * read the disklabel until the disk has been attached,
1302 		 * because space for the disklabel is allocated
1303 		 * in disk_attach();
1304 		 */
1305 		ccio->ccio_unit = unit;
1306 		ccio->ccio_size = cs->sc_size;
1307 
1308 		/* Try and read the disklabel. */
1309 		ccdgetdisklabel(dev);
1310 		disk_set_info(NULL, &cs->sc_dkdev, NULL);
1311 
1312 		/* discover wedges */
1313 		mutex_exit(&cs->sc_dvlock);
1314 		dkwedge_discover(&cs->sc_dkdev);
1315 		return 0;
1316 
1317 	case CCDIOCCLR:
1318 		/*
1319 		 * Don't unconfigure if any other partitions are open
1320 		 * or if both the character and block flavors of this
1321 		 * partition are open.
1322 		 */
1323 		part = DISKPART(dev);
1324 		pmask = (1 << part);
1325 		if ((cs->sc_dkdev.dk_openmask & ~pmask) ||
1326 		    ((cs->sc_dkdev.dk_bopenmask & pmask) &&
1327 		    (cs->sc_dkdev.dk_copenmask & pmask))) {
1328 			error = EBUSY;
1329 			goto out;
1330 		}
1331 
1332 		/* Delete all of our wedges. */
1333 		dkwedge_delall(&cs->sc_dkdev);
1334 
1335 		/* Stop new I/O, wait for in-flight I/O to complete. */
1336 		mutex_enter(cs->sc_iolock);
1337 		cs->sc_flags &= ~(CCDF_INITED|CCDF_VLABEL);
1338 		cs->sc_zap = true;
1339 		while (disk_isbusy(&cs->sc_dkdev) ||
1340 		    bufq_peek(cs->sc_bufq) != NULL ||
1341 		    cs->sc_thread != NULL) {
1342 			cv_broadcast(&cs->sc_push);
1343 			(void)cv_timedwait(&cs->sc_stop, cs->sc_iolock, hz);
1344 		}
1345 		mutex_exit(cs->sc_iolock);
1346 
1347 		/*
1348 		 * Free ccd_softc information and clear entry.
1349 		 */
1350 
1351 		/* Close the components and free their pathnames. */
1352 		for (i = 0; i < cs->sc_nccdisks; ++i) {
1353 			/*
1354 			 * XXX: this close could potentially fail and
1355 			 * cause Bad Things.  Maybe we need to force
1356 			 * the close to happen?
1357 			 */
1358 #ifdef DEBUG
1359 			if (ccddebug & CCDB_VNODE)
1360 				vprint("CCDIOCCLR: vnode info",
1361 				    cs->sc_cinfo[i].ci_vp);
1362 #endif
1363 			(void)vn_close(cs->sc_cinfo[i].ci_vp, FREAD|FWRITE,
1364 			    uc);
1365 			kmem_free(cs->sc_cinfo[i].ci_path,
1366 			    cs->sc_cinfo[i].ci_pathlen);
1367 		}
1368 
1369 		/* Free interleave index. */
1370 		for (i = 0; cs->sc_itable[i].ii_ndisk; ++i) {
1371 			kmem_free(cs->sc_itable[i].ii_index,
1372 			    cs->sc_itable[i].ii_indexsz);
1373 		}
1374 
1375 		/* Free component info and interleave table. */
1376 		kmem_free(cs->sc_cinfo, cs->sc_nccdisks *
1377 		    sizeof(struct ccdcinfo));
1378 		kmem_free(cs->sc_itable, (cs->sc_nccdisks + 1) *
1379 		    sizeof(struct ccdiinfo));
1380 
1381 		aprint_normal("%s: detached\n", cs->sc_xname);
1382 
1383 		/* Detach the disk. */
1384 		disk_detach(&cs->sc_dkdev);
1385 		bufq_free(cs->sc_bufq);
1386 		ccdput(cs);
1387 		/* Don't break, otherwise cs is read again. */
1388 		return 0;
1389 
1390 	case DIOCGSTRATEGY:
1391 	    {
1392 		struct disk_strategy *dks = (void *)data;
1393 
1394 		mutex_enter(cs->sc_iolock);
1395 		if (cs->sc_bufq != NULL)
1396 			strlcpy(dks->dks_name,
1397 			    bufq_getstrategyname(cs->sc_bufq),
1398 			    sizeof(dks->dks_name));
1399 		else
1400 			error = EINVAL;
1401 		mutex_exit(cs->sc_iolock);
1402 		dks->dks_paramlen = 0;
1403 		break;
1404 	    }
1405 
1406 	case DIOCGCACHE:
1407 	    {
1408 		int dkcache = 0;
1409 
1410 		/*
1411 		 * We pass this call down to all components and report
1412 		 * intersection of the flags returned by the components.
1413 		 * If any errors out, we return error. CCD components
1414 		 * can not change unless the device is unconfigured, so
1415 		 * device feature flags will remain static. RCE/WCE can change
1416 		 * of course, if set directly on underlying device.
1417 		 */
1418 		for (error = 0, i = 0; i < cs->sc_nccdisks; i++) {
1419 			error = VOP_IOCTL(cs->sc_cinfo[i].ci_vp, cmd, &j,
1420 				      flag, uc);
1421 			if (error)
1422 				break;
1423 
1424 			if (i == 0)
1425 				dkcache = j;
1426 			else
1427 				dkcache = DKCACHE_COMBINE(dkcache, j);
1428 		}
1429 
1430 		*((int *)data) = dkcache;
1431 		break;
1432 	    }
1433 
1434 	case DIOCCACHESYNC:
1435 		/*
1436 		 * We pass this call down to all components and report
1437 		 * the first error we encounter.
1438 		 */
1439 		for (error = 0, i = 0; i < cs->sc_nccdisks; i++) {
1440 			j = VOP_IOCTL(cs->sc_cinfo[i].ci_vp, cmd, data,
1441 				      flag, uc);
1442 			if (j != 0 && error == 0)
1443 				error = j;
1444 		}
1445 		break;
1446 
1447 	case DIOCWDINFO:
1448 	case DIOCSDINFO:
1449 #ifdef __HAVE_OLD_DISKLABEL
1450 	case ODIOCWDINFO:
1451 	case ODIOCSDINFO:
1452 #endif
1453 	{
1454 		struct disklabel *lp;
1455 #ifdef __HAVE_OLD_DISKLABEL
1456 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
1457 			memset(&newlabel, 0, sizeof newlabel);
1458 			memcpy(&newlabel, data, sizeof (struct olddisklabel));
1459 			lp = &newlabel;
1460 		} else
1461 #endif
1462 		lp = (struct disklabel *)data;
1463 
1464 		cs->sc_flags |= CCDF_LABELLING;
1465 
1466 		error = setdisklabel(cs->sc_dkdev.dk_label,
1467 		    lp, 0, cs->sc_dkdev.dk_cpulabel);
1468 		if (error == 0) {
1469 			if (cmd == DIOCWDINFO
1470 #ifdef __HAVE_OLD_DISKLABEL
1471 			    || cmd == ODIOCWDINFO
1472 #endif
1473 			   )
1474 				error = writedisklabel(CCDLABELDEV(dev),
1475 				    ccdstrategy, cs->sc_dkdev.dk_label,
1476 				    cs->sc_dkdev.dk_cpulabel);
1477 		}
1478 
1479 		cs->sc_flags &= ~CCDF_LABELLING;
1480 		break;
1481 	}
1482 
1483 	case DIOCKLABEL:
1484 		if (*(int *)data != 0)
1485 			cs->sc_flags |= CCDF_KLABEL;
1486 		else
1487 			cs->sc_flags &= ~CCDF_KLABEL;
1488 		break;
1489 
1490 	case DIOCWLABEL:
1491 		if (*(int *)data != 0)
1492 			cs->sc_flags |= CCDF_WLABEL;
1493 		else
1494 			cs->sc_flags &= ~CCDF_WLABEL;
1495 		break;
1496 
1497 	case DIOCGDEFLABEL:
1498 		ccdgetdefaultlabel(cs, (struct disklabel *)data);
1499 		break;
1500 
1501 #ifdef __HAVE_OLD_DISKLABEL
1502 	case ODIOCGDEFLABEL:
1503 		ccdgetdefaultlabel(cs, &newlabel);
1504 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1505 			return ENOTTY;
1506 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1507 		break;
1508 #endif
1509 
1510 	default:
1511 		error = ENOTTY;
1512 	}
1513 
1514  out:
1515 	mutex_exit(&cs->sc_dvlock);
1516 	return (error);
1517 }
1518 
1519 static int
1520 ccdsize(dev_t dev)
1521 {
1522 	struct ccd_softc *cs;
1523 	struct disklabel *lp;
1524 	int part, unit, omask, size;
1525 
1526 	unit = ccdunit(dev);
1527 	if ((cs = ccdget(unit, 0)) == NULL)
1528 		return -1;
1529 
1530 	if ((cs->sc_flags & CCDF_INITED) == 0)
1531 		return (-1);
1532 
1533 	part = DISKPART(dev);
1534 	omask = cs->sc_dkdev.dk_openmask & (1 << part);
1535 	lp = cs->sc_dkdev.dk_label;
1536 
1537 	if (omask == 0 && ccdopen(dev, 0, S_IFBLK, curlwp))
1538 		return (-1);
1539 
1540 	if (lp->d_partitions[part].p_fstype != FS_SWAP)
1541 		size = -1;
1542 	else
1543 		size = lp->d_partitions[part].p_size *
1544 		    (lp->d_secsize / DEV_BSIZE);
1545 
1546 	if (omask == 0 && ccdclose(dev, 0, S_IFBLK, curlwp))
1547 		return (-1);
1548 
1549 	return (size);
1550 }
1551 
1552 static void
1553 ccdgetdefaultlabel(struct ccd_softc *cs, struct disklabel *lp)
1554 {
1555 	struct ccdgeom *ccg = &cs->sc_geom;
1556 
1557 	memset(lp, 0, sizeof(*lp));
1558 
1559 	if (cs->sc_size > UINT32_MAX)
1560 		lp->d_secperunit = UINT32_MAX;
1561 	else
1562 		lp->d_secperunit = cs->sc_size;
1563 	lp->d_secsize = ccg->ccg_secsize;
1564 	lp->d_nsectors = ccg->ccg_nsectors;
1565 	lp->d_ntracks = ccg->ccg_ntracks;
1566 	lp->d_ncylinders = ccg->ccg_ncylinders;
1567 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1568 
1569 	strncpy(lp->d_typename, "ccd", sizeof(lp->d_typename));
1570 	lp->d_type = DKTYPE_CCD;
1571 	strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
1572 	lp->d_rpm = 3600;
1573 	lp->d_interleave = 1;
1574 	lp->d_flags = 0;
1575 
1576 	lp->d_partitions[RAW_PART].p_offset = 0;
1577 	lp->d_partitions[RAW_PART].p_size = lp->d_secperunit;
1578 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
1579 	lp->d_npartitions = RAW_PART + 1;
1580 
1581 	lp->d_magic = DISKMAGIC;
1582 	lp->d_magic2 = DISKMAGIC;
1583 	lp->d_checksum = dkcksum(cs->sc_dkdev.dk_label);
1584 }
1585 
1586 /*
1587  * Read the disklabel from the ccd.  If one is not present, fake one
1588  * up.
1589  */
1590 static void
1591 ccdgetdisklabel(dev_t dev)
1592 {
1593 	int unit = ccdunit(dev);
1594 	struct ccd_softc *cs;
1595 	const char *errstring;
1596 	struct disklabel *lp;
1597 	struct cpu_disklabel *clp;
1598 
1599 	if ((cs = ccdget(unit, 0)) == NULL)
1600 		return;
1601 	lp = cs->sc_dkdev.dk_label;
1602 	clp = cs->sc_dkdev.dk_cpulabel;
1603 	KASSERT(mutex_owned(&cs->sc_dvlock));
1604 
1605 	memset(clp, 0, sizeof(*clp));
1606 
1607 	ccdgetdefaultlabel(cs, lp);
1608 
1609 	/*
1610 	 * Call the generic disklabel extraction routine.
1611 	 */
1612 	cs->sc_flags |= CCDF_RLABEL;
1613 	if ((cs->sc_flags & CCDF_NOLABEL) != 0)
1614 		errstring = "CCDF_NOLABEL set; ignoring on-disk label";
1615 	else
1616 		errstring = readdisklabel(CCDLABELDEV(dev), ccdstrategy,
1617 		    cs->sc_dkdev.dk_label, cs->sc_dkdev.dk_cpulabel);
1618 	if (errstring)
1619 		ccdmakedisklabel(cs);
1620 	else {
1621 		int i;
1622 		struct partition *pp;
1623 
1624 		/*
1625 		 * Sanity check whether the found disklabel is valid.
1626 		 *
1627 		 * This is necessary since total size of ccd may vary
1628 		 * when an interleave is changed even though exactly
1629 		 * same componets are used, and old disklabel may used
1630 		 * if that is found.
1631 		 */
1632 		if (lp->d_secperunit < UINT32_MAX ?
1633 			lp->d_secperunit != cs->sc_size :
1634 			lp->d_secperunit > cs->sc_size)
1635 			printf("WARNING: %s: "
1636 			    "total sector size in disklabel (%ju) != "
1637 			    "the size of ccd (%ju)\n", cs->sc_xname,
1638 			    (uintmax_t)lp->d_secperunit,
1639 			    (uintmax_t)cs->sc_size);
1640 		for (i = 0; i < lp->d_npartitions; i++) {
1641 			pp = &lp->d_partitions[i];
1642 			if (pp->p_offset + pp->p_size > cs->sc_size)
1643 				printf("WARNING: %s: end of partition `%c' "
1644 				    "exceeds the size of ccd (%ju)\n",
1645 				    cs->sc_xname, 'a' + i, (uintmax_t)cs->sc_size);
1646 		}
1647 	}
1648 
1649 #ifdef DEBUG
1650 	/* It's actually extremely common to have unlabeled ccds. */
1651 	if (ccddebug & CCDB_LABEL)
1652 		if (errstring != NULL)
1653 			printf("%s: %s\n", cs->sc_xname, errstring);
1654 #endif
1655 
1656 	/* In-core label now valid. */
1657 	cs->sc_flags = (cs->sc_flags | CCDF_VLABEL) & ~CCDF_RLABEL;
1658 }
1659 
1660 /*
1661  * Take care of things one might want to take care of in the event
1662  * that a disklabel isn't present.
1663  */
1664 static void
1665 ccdmakedisklabel(struct ccd_softc *cs)
1666 {
1667 	struct disklabel *lp = cs->sc_dkdev.dk_label;
1668 
1669 	/*
1670 	 * For historical reasons, if there's no disklabel present
1671 	 * the raw partition must be marked FS_BSDFFS.
1672 	 */
1673 	lp->d_partitions[RAW_PART].p_fstype = FS_BSDFFS;
1674 
1675 	strncpy(lp->d_packname, "default label", sizeof(lp->d_packname));
1676 
1677 	lp->d_checksum = dkcksum(lp);
1678 }
1679 
1680 #ifdef DEBUG
1681 static void
1682 printiinfo(struct ccdiinfo *ii)
1683 {
1684 	int ix, i;
1685 
1686 	for (ix = 0; ii->ii_ndisk; ix++, ii++) {
1687 		printf(" itab[%d]: #dk %d sblk %" PRId64 " soff %" PRId64,
1688 		    ix, ii->ii_ndisk, ii->ii_startblk, ii->ii_startoff);
1689 		for (i = 0; i < ii->ii_ndisk; i++)
1690 			printf(" %d", ii->ii_index[i]);
1691 		printf("\n");
1692 	}
1693 }
1694 #endif
1695 
1696 MODULE(MODULE_CLASS_DRIVER, ccd, "dk_subr");
1697 
1698 static int
1699 ccd_modcmd(modcmd_t cmd, void *arg)
1700 {
1701 	int error = 0;
1702 #ifdef _MODULE
1703 	int bmajor = -1, cmajor = -1;
1704 #endif
1705 
1706 
1707 	switch (cmd) {
1708 	case MODULE_CMD_INIT:
1709 #ifdef _MODULE
1710 		ccdattach(0);
1711 
1712 		error = devsw_attach("ccd", &ccd_bdevsw, &bmajor,
1713 		    &ccd_cdevsw, &cmajor);
1714 		sysctl_kern_ccd_setup(&ccd_clog);
1715 #endif
1716 		break;
1717 
1718 	case MODULE_CMD_FINI:
1719 #ifdef _MODULE
1720 		mutex_enter(&ccd_lock);
1721 		if (!LIST_EMPTY(ccds)) {
1722 			mutex_exit(&ccd_lock);
1723 			error = EBUSY;
1724 		} else {
1725 			mutex_exit(&ccd_lock);
1726 			error = devsw_detach(&ccd_bdevsw, &ccd_cdevsw);
1727 			ccddetach();
1728 		}
1729 		sysctl_teardown(&ccd_clog);
1730 #endif
1731 		break;
1732 
1733 	case MODULE_CMD_STAT:
1734 		return ENOTTY;
1735 
1736 	default:
1737 		return ENOTTY;
1738 	}
1739 
1740 	return error;
1741 }
1742 
1743 static int
1744 ccd_units_sysctl(SYSCTLFN_ARGS)
1745 {
1746 	struct sysctlnode node;
1747 	struct ccd_softc *sc;
1748 	int error, i, nccd, *units;
1749 	size_t size;
1750 
1751 	nccd = 0;
1752 	mutex_enter(&ccd_lock);
1753 	LIST_FOREACH(sc, &ccds, sc_link)
1754 		nccd++;
1755 	mutex_exit(&ccd_lock);
1756 
1757 	if (nccd != 0) {
1758 		size = nccd * sizeof(*units);
1759 		units = kmem_zalloc(size, KM_SLEEP);
1760 		i = 0;
1761 		mutex_enter(&ccd_lock);
1762 		LIST_FOREACH(sc, &ccds, sc_link) {
1763 			if (i >= nccd)
1764 				break;
1765 			units[i] = sc->sc_unit;
1766 		}
1767 		mutex_exit(&ccd_lock);
1768 	} else {
1769 		units = NULL;
1770 		size = 0;
1771 	}
1772 
1773 	node = *rnode;
1774 	node.sysctl_data = units;
1775 	node.sysctl_size = size;
1776 
1777 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
1778 	if (units)
1779 		kmem_free(units, size);
1780 	return error;
1781 }
1782 
1783 static int
1784 ccd_info_sysctl(SYSCTLFN_ARGS)
1785 {
1786 	struct sysctlnode node;
1787 	struct ccddiskinfo ccd;
1788 	struct ccd_softc *sc;
1789 	int unit;
1790 
1791 	if (newp == NULL || newlen != sizeof(int))
1792 		return EINVAL;
1793 
1794 	unit = *(const int *)newp;
1795 	newp = NULL;
1796 	newlen = 0;
1797 	ccd.ccd_ndisks = ~0;
1798 	mutex_enter(&ccd_lock);
1799 	LIST_FOREACH(sc, &ccds, sc_link) {
1800 		if (sc->sc_unit == unit) {
1801 			ccd.ccd_ileave = sc->sc_ileave;
1802 			ccd.ccd_size = sc->sc_size;
1803 			ccd.ccd_ndisks = sc->sc_nccdisks;
1804 			ccd.ccd_flags = sc->sc_flags;
1805 			break;
1806 		}
1807 	}
1808 	mutex_exit(&ccd_lock);
1809 
1810 	if (ccd.ccd_ndisks == ~0)
1811 		return ENOENT;
1812 
1813 	node = *rnode;
1814 	node.sysctl_data = &ccd;
1815 	node.sysctl_size = sizeof(ccd);
1816 
1817 	return sysctl_lookup(SYSCTLFN_CALL(&node));
1818 }
1819 
1820 static int
1821 ccd_components_sysctl(SYSCTLFN_ARGS)
1822 {
1823 	struct sysctlnode node;
1824 	int error, unit;
1825 	size_t size;
1826 	char *names, *p, *ep;
1827 	struct ccd_softc *sc;
1828 
1829 	if (newp == NULL || newlen != sizeof(int))
1830 		return EINVAL;
1831 
1832 	size = 0;
1833 	unit = *(const int *)newp;
1834 	newp = NULL;
1835 	newlen = 0;
1836 	mutex_enter(&ccd_lock);
1837 	LIST_FOREACH(sc, &ccds, sc_link)
1838 		if (sc->sc_unit == unit) {
1839 			for (size_t i = 0; i < sc->sc_nccdisks; i++)
1840 				size += strlen(sc->sc_cinfo[i].ci_path) + 1;
1841 			break;
1842 		}
1843 	mutex_exit(&ccd_lock);
1844 
1845 	if (size == 0)
1846 		return ENOENT;
1847 	names = kmem_zalloc(size, KM_SLEEP);
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