xref: /netbsd-src/sys/dev/ccd.c (revision 267197ec1eebfcb9810ea27a89625b6ddf68e3e7)
1 /*	$NetBSD: ccd.c,v 1.127 2008/02/02 10:40:50 hannken Exp $	*/
2 
3 /*-
4  * Copyright (c) 1996, 1997, 1998, 1999, 2007 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Jason R. Thorpe.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *        This product includes software developed by the NetBSD
21  *        Foundation, Inc. and its contributors.
22  * 4. Neither the name of The NetBSD Foundation nor the names of its
23  *    contributors may be used to endorse or promote products derived
24  *    from this software without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
27  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
28  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
29  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
30  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36  * POSSIBILITY OF SUCH DAMAGE.
37  */
38 
39 /*
40  * Copyright (c) 1990, 1993
41  *	The Regents of the University of California.  All rights reserved.
42  *
43  * This code is derived from software contributed to Berkeley by
44  * the Systems Programming Group of the University of Utah Computer
45  * Science Department.
46  *
47  * Redistribution and use in source and binary forms, with or without
48  * modification, are permitted provided that the following conditions
49  * are met:
50  * 1. Redistributions of source code must retain the above copyright
51  *    notice, this list of conditions and the following disclaimer.
52  * 2. Redistributions in binary form must reproduce the above copyright
53  *    notice, this list of conditions and the following disclaimer in the
54  *    documentation and/or other materials provided with the distribution.
55  * 3. Neither the name of the University nor the names of its contributors
56  *    may be used to endorse or promote products derived from this software
57  *    without specific prior written permission.
58  *
59  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
60  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
61  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
62  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
63  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
64  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
65  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
66  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
67  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
68  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
69  * SUCH DAMAGE.
70  *
71  * from: Utah $Hdr: cd.c 1.6 90/11/28$
72  *
73  *	@(#)cd.c	8.2 (Berkeley) 11/16/93
74  */
75 
76 /*
77  * Copyright (c) 1988 University of Utah.
78  *
79  * This code is derived from software contributed to Berkeley by
80  * the Systems Programming Group of the University of Utah Computer
81  * Science Department.
82  *
83  * Redistribution and use in source and binary forms, with or without
84  * modification, are permitted provided that the following conditions
85  * are met:
86  * 1. Redistributions of source code must retain the above copyright
87  *    notice, this list of conditions and the following disclaimer.
88  * 2. Redistributions in binary form must reproduce the above copyright
89  *    notice, this list of conditions and the following disclaimer in the
90  *    documentation and/or other materials provided with the distribution.
91  * 3. All advertising materials mentioning features or use of this software
92  *    must display the following acknowledgement:
93  *	This product includes software developed by the University of
94  *	California, Berkeley and its contributors.
95  * 4. Neither the name of the University nor the names of its contributors
96  *    may be used to endorse or promote products derived from this software
97  *    without specific prior written permission.
98  *
99  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
100  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
101  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
102  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
103  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
104  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
105  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
106  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
107  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
108  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
109  * SUCH DAMAGE.
110  *
111  * from: Utah $Hdr: cd.c 1.6 90/11/28$
112  *
113  *	@(#)cd.c	8.2 (Berkeley) 11/16/93
114  */
115 
116 /*
117  * "Concatenated" disk driver.
118  *
119  * Dynamic configuration and disklabel support by:
120  *	Jason R. Thorpe <thorpej@nas.nasa.gov>
121  *	Numerical Aerodynamic Simulation Facility
122  *	Mail Stop 258-6
123  *	NASA Ames Research Center
124  *	Moffett Field, CA 94035
125  */
126 
127 #include <sys/cdefs.h>
128 __KERNEL_RCSID(0, "$NetBSD: ccd.c,v 1.127 2008/02/02 10:40:50 hannken Exp $");
129 
130 #include <sys/param.h>
131 #include <sys/systm.h>
132 #include <sys/proc.h>
133 #include <sys/errno.h>
134 #include <sys/buf.h>
135 #include <sys/bufq.h>
136 #include <sys/malloc.h>
137 #include <sys/pool.h>
138 #include <sys/namei.h>
139 #include <sys/stat.h>
140 #include <sys/ioctl.h>
141 #include <sys/disklabel.h>
142 #include <sys/device.h>
143 #include <sys/disk.h>
144 #include <sys/syslog.h>
145 #include <sys/fcntl.h>
146 #include <sys/vnode.h>
147 #include <sys/conf.h>
148 #include <sys/mutex.h>
149 #include <sys/queue.h>
150 #include <sys/kauth.h>
151 
152 #include <dev/ccdvar.h>
153 #include <dev/dkvar.h>
154 
155 #if defined(CCDDEBUG) && !defined(DEBUG)
156 #define DEBUG
157 #endif
158 
159 #ifdef DEBUG
160 #define CCDB_FOLLOW	0x01
161 #define CCDB_INIT	0x02
162 #define CCDB_IO		0x04
163 #define CCDB_LABEL	0x08
164 #define CCDB_VNODE	0x10
165 int ccddebug = 0x00;
166 #endif
167 
168 #define	ccdunit(x)	DISKUNIT(x)
169 
170 struct ccdbuf {
171 	struct buf	cb_buf;		/* new I/O buf */
172 	struct buf	*cb_obp;	/* ptr. to original I/O buf */
173 	struct ccd_softc *cb_sc;	/* pointer to ccd softc */
174 	int		cb_comp;	/* target component */
175 	SIMPLEQ_ENTRY(ccdbuf) cb_q;	/* fifo of component buffers */
176 };
177 
178 /* component buffer pool */
179 static struct pool ccd_cbufpool;
180 
181 #define	CCD_GETBUF()		pool_get(&ccd_cbufpool, PR_NOWAIT)
182 #define	CCD_PUTBUF(cbp)		pool_put(&ccd_cbufpool, cbp)
183 
184 #define CCDLABELDEV(dev)	\
185 	(MAKEDISKDEV(major((dev)), ccdunit((dev)), RAW_PART))
186 
187 /* called by main() at boot time */
188 void	ccdattach(int);
189 
190 /* called by biodone() at interrupt time */
191 static void	ccdiodone(struct buf *);
192 
193 static void	ccdstart(struct ccd_softc *);
194 static void	ccdinterleave(struct ccd_softc *);
195 static void	ccdintr(struct ccd_softc *, struct buf *);
196 static int	ccdinit(struct ccd_softc *, char **, struct vnode **,
197 		    struct lwp *);
198 static struct ccdbuf *ccdbuffer(struct ccd_softc *, struct buf *,
199 		    daddr_t, void *, long);
200 static void	ccdgetdefaultlabel(struct ccd_softc *, struct disklabel *);
201 static void	ccdgetdisklabel(dev_t);
202 static void	ccdmakedisklabel(struct ccd_softc *);
203 
204 static dev_type_open(ccdopen);
205 static dev_type_close(ccdclose);
206 static dev_type_read(ccdread);
207 static dev_type_write(ccdwrite);
208 static dev_type_ioctl(ccdioctl);
209 static dev_type_strategy(ccdstrategy);
210 static dev_type_dump(ccddump);
211 static dev_type_size(ccdsize);
212 
213 const struct bdevsw ccd_bdevsw = {
214 	ccdopen, ccdclose, ccdstrategy, ccdioctl, ccddump, ccdsize, D_DISK
215 };
216 
217 const struct cdevsw ccd_cdevsw = {
218 	ccdopen, ccdclose, ccdread, ccdwrite, ccdioctl,
219 	nostop, notty, nopoll, nommap, nokqfilter, D_DISK
220 };
221 
222 #ifdef DEBUG
223 static	void printiinfo(struct ccdiinfo *);
224 #endif
225 
226 /* Publically visible for the benefit of libkvm and ccdconfig(8). */
227 struct ccd_softc 	*ccd_softc;
228 const int		ccd_softc_elemsize = sizeof(struct ccd_softc);
229 int			numccd = 0;
230 
231 /*
232  * Called by main() during pseudo-device attachment.  All we need
233  * to do is allocate enough space for devices to be configured later.
234  */
235 void
236 ccdattach(int num)
237 {
238 	struct ccd_softc *cs;
239 	int i;
240 
241 	if (num <= 0) {
242 #ifdef DIAGNOSTIC
243 		panic("ccdattach: count <= 0");
244 #endif
245 		return;
246 	}
247 
248 	ccd_softc = (struct ccd_softc *)malloc(num * ccd_softc_elemsize,
249 	    M_DEVBUF, M_NOWAIT|M_ZERO);
250 	if (ccd_softc == NULL) {
251 		printf("WARNING: no memory for concatenated disks\n");
252 		return;
253 	}
254 	numccd = num;
255 
256 	/* Initialize the component buffer pool. */
257 	pool_init(&ccd_cbufpool, sizeof(struct ccdbuf), 0,
258 	    0, 0, "ccdpl", NULL, IPL_BIO);
259 
260 	/* Initialize per-softc structures. */
261 	for (i = 0; i < num; i++) {
262 		cs = &ccd_softc[i];
263 		snprintf(cs->sc_xname, sizeof(cs->sc_xname), "ccd%d", i);
264 		mutex_init(&cs->sc_lock, MUTEX_DEFAULT, IPL_NONE);
265 		disk_init(&cs->sc_dkdev, cs->sc_xname, NULL); /* XXX */
266 	}
267 }
268 
269 static int
270 ccdinit(struct ccd_softc *cs, char **cpaths, struct vnode **vpp,
271     struct lwp *l)
272 {
273 	struct ccdcinfo *ci = NULL;
274 	size_t size;
275 	int ix;
276 	struct vattr va;
277 	size_t minsize;
278 	int maxsecsize;
279 	struct partinfo dpart;
280 	struct ccdgeom *ccg = &cs->sc_geom;
281 	char *tmppath;
282 	int error, path_alloced;
283 
284 #ifdef DEBUG
285 	if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
286 		printf("%s: ccdinit\n", cs->sc_xname);
287 #endif
288 
289 	/* Allocate space for the component info. */
290 	cs->sc_cinfo = malloc(cs->sc_nccdisks * sizeof(struct ccdcinfo),
291 	    M_DEVBUF, M_WAITOK);
292 
293 	tmppath = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
294 
295 	cs->sc_size = 0;
296 
297 	/*
298 	 * Verify that each component piece exists and record
299 	 * relevant information about it.
300 	 */
301 	maxsecsize = 0;
302 	minsize = 0;
303 	for (ix = 0, path_alloced = 0; ix < cs->sc_nccdisks; ix++) {
304 		ci = &cs->sc_cinfo[ix];
305 		ci->ci_vp = vpp[ix];
306 
307 		/*
308 		 * Copy in the pathname of the component.
309 		 */
310 		memset(tmppath, 0, sizeof(tmppath));	/* sanity */
311 		error = copyinstr(cpaths[ix], tmppath,
312 		    MAXPATHLEN, &ci->ci_pathlen);
313 		if (error) {
314 #ifdef DEBUG
315 			if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
316 				printf("%s: can't copy path, error = %d\n",
317 				    cs->sc_xname, error);
318 #endif
319 			goto out;
320 		}
321 		ci->ci_path = malloc(ci->ci_pathlen, M_DEVBUF, M_WAITOK);
322 		memcpy(ci->ci_path, tmppath, ci->ci_pathlen);
323 		path_alloced++;
324 
325 		/*
326 		 * XXX: Cache the component's dev_t.
327 		 */
328 		if ((error = VOP_GETATTR(vpp[ix], &va, l->l_cred)) != 0) {
329 #ifdef DEBUG
330 			if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
331 				printf("%s: %s: getattr failed %s = %d\n",
332 				    cs->sc_xname, ci->ci_path,
333 				    "error", error);
334 #endif
335 			goto out;
336 		}
337 		ci->ci_dev = va.va_rdev;
338 
339 		/*
340 		 * Get partition information for the component.
341 		 */
342 		error = VOP_IOCTL(vpp[ix], DIOCGPART, &dpart,
343 		    FREAD, l->l_cred);
344 		if (error) {
345 #ifdef DEBUG
346 			if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
347 				 printf("%s: %s: ioctl failed, error = %d\n",
348 				     cs->sc_xname, ci->ci_path, error);
349 #endif
350 			goto out;
351 		}
352 
353 /*
354  * This diagnostic test is disabled (for now?) since not all port supports
355  * on-disk BSD disklabel.
356  */
357 #if 0 /* def DIAGNOSTIC */
358 		/* Check fstype field of component. */
359 		if (dpart.part->p_fstype != FS_CCD)
360 			printf("%s: WARNING: %s: fstype %d != FS_CCD\n",
361 			    cs->sc_xname, ci->ci_path, dpart.part->p_fstype);
362 #endif
363 
364 		/*
365 		 * Calculate the size, truncating to an interleave
366 		 * boundary if necessary.
367 		 */
368 		maxsecsize =
369 		    ((dpart.disklab->d_secsize > maxsecsize) ?
370 		    dpart.disklab->d_secsize : maxsecsize);
371 		size = dpart.part->p_size;
372 		if (cs->sc_ileave > 1)
373 			size -= size % cs->sc_ileave;
374 
375 		if (size == 0) {
376 #ifdef DEBUG
377 			if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
378 				printf("%s: %s: size == 0\n",
379 				    cs->sc_xname, ci->ci_path);
380 #endif
381 			error = ENODEV;
382 			goto out;
383 		}
384 
385 		if (minsize == 0 || size < minsize)
386 			minsize = size;
387 		ci->ci_size = size;
388 		cs->sc_size += size;
389 	}
390 
391 	/*
392 	 * Don't allow the interleave to be smaller than
393 	 * the biggest component sector.
394 	 */
395 	if ((cs->sc_ileave > 0) &&
396 	    (cs->sc_ileave < (maxsecsize / DEV_BSIZE))) {
397 #ifdef DEBUG
398 		if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
399 			printf("%s: interleave must be at least %d\n",
400 			    cs->sc_xname, (maxsecsize / DEV_BSIZE));
401 #endif
402 		error = EINVAL;
403 		goto out;
404 	}
405 
406 	/*
407 	 * If uniform interleave is desired set all sizes to that of
408 	 * the smallest component.
409 	 */
410 	if (cs->sc_flags & CCDF_UNIFORM) {
411 		for (ci = cs->sc_cinfo;
412 		     ci < &cs->sc_cinfo[cs->sc_nccdisks]; ci++)
413 			ci->ci_size = minsize;
414 
415 		cs->sc_size = cs->sc_nccdisks * minsize;
416 	}
417 
418 	/*
419 	 * Construct the interleave table.
420 	 */
421 	ccdinterleave(cs);
422 
423 	/*
424 	 * Create pseudo-geometry based on 1MB cylinders.  It's
425 	 * pretty close.
426 	 */
427 	ccg->ccg_secsize = DEV_BSIZE;
428 	ccg->ccg_ntracks = 1;
429 	ccg->ccg_nsectors = 1024 * (1024 / ccg->ccg_secsize);
430 	ccg->ccg_ncylinders = cs->sc_size / ccg->ccg_nsectors;
431 
432 	cs->sc_flags |= CCDF_INITED;
433 
434 	free(tmppath, M_TEMP);
435 
436 	return (0);
437 
438  out:
439 	for (ix = 0; ix < path_alloced; ix++)
440 		free(cs->sc_cinfo[ix].ci_path, M_DEVBUF);
441 	free(cs->sc_cinfo, M_DEVBUF);
442 	free(tmppath, M_TEMP);
443 	return (error);
444 }
445 
446 static void
447 ccdinterleave(struct ccd_softc *cs)
448 {
449 	struct ccdcinfo *ci, *smallci;
450 	struct ccdiinfo *ii;
451 	daddr_t bn, lbn;
452 	int ix;
453 	u_long size;
454 
455 #ifdef DEBUG
456 	if (ccddebug & CCDB_INIT)
457 		printf("ccdinterleave(%p): ileave %d\n", cs, cs->sc_ileave);
458 #endif
459 	/*
460 	 * Allocate an interleave table.
461 	 * Chances are this is too big, but we don't care.
462 	 */
463 	size = (cs->sc_nccdisks + 1) * sizeof(struct ccdiinfo);
464 	cs->sc_itable = (struct ccdiinfo *)malloc(size, M_DEVBUF,
465 	    M_WAITOK|M_ZERO);
466 
467 	/*
468 	 * Trivial case: no interleave (actually interleave of disk size).
469 	 * Each table entry represents a single component in its entirety.
470 	 */
471 	if (cs->sc_ileave == 0) {
472 		bn = 0;
473 		ii = cs->sc_itable;
474 
475 		for (ix = 0; ix < cs->sc_nccdisks; ix++) {
476 			/* Allocate space for ii_index. */
477 			ii->ii_index = malloc(sizeof(int), M_DEVBUF, M_WAITOK);
478 			ii->ii_ndisk = 1;
479 			ii->ii_startblk = bn;
480 			ii->ii_startoff = 0;
481 			ii->ii_index[0] = ix;
482 			bn += cs->sc_cinfo[ix].ci_size;
483 			ii++;
484 		}
485 		ii->ii_ndisk = 0;
486 #ifdef DEBUG
487 		if (ccddebug & CCDB_INIT)
488 			printiinfo(cs->sc_itable);
489 #endif
490 		return;
491 	}
492 
493 	/*
494 	 * The following isn't fast or pretty; it doesn't have to be.
495 	 */
496 	size = 0;
497 	bn = lbn = 0;
498 	for (ii = cs->sc_itable; ; ii++) {
499 		/* Allocate space for ii_index. */
500 		ii->ii_index = malloc((sizeof(int) * cs->sc_nccdisks),
501 		    M_DEVBUF, M_WAITOK);
502 
503 		/*
504 		 * Locate the smallest of the remaining components
505 		 */
506 		smallci = NULL;
507 		for (ci = cs->sc_cinfo;
508 		     ci < &cs->sc_cinfo[cs->sc_nccdisks]; ci++)
509 			if (ci->ci_size > size &&
510 			    (smallci == NULL ||
511 			     ci->ci_size < smallci->ci_size))
512 				smallci = ci;
513 
514 		/*
515 		 * Nobody left, all done
516 		 */
517 		if (smallci == NULL) {
518 			ii->ii_ndisk = 0;
519 			break;
520 		}
521 
522 		/*
523 		 * Record starting logical block and component offset
524 		 */
525 		ii->ii_startblk = bn / cs->sc_ileave;
526 		ii->ii_startoff = lbn;
527 
528 		/*
529 		 * Determine how many disks take part in this interleave
530 		 * and record their indices.
531 		 */
532 		ix = 0;
533 		for (ci = cs->sc_cinfo;
534 		     ci < &cs->sc_cinfo[cs->sc_nccdisks]; ci++)
535 			if (ci->ci_size >= smallci->ci_size)
536 				ii->ii_index[ix++] = ci - cs->sc_cinfo;
537 		ii->ii_ndisk = ix;
538 		bn += ix * (smallci->ci_size - size);
539 		lbn = smallci->ci_size / cs->sc_ileave;
540 		size = smallci->ci_size;
541 	}
542 #ifdef DEBUG
543 	if (ccddebug & CCDB_INIT)
544 		printiinfo(cs->sc_itable);
545 #endif
546 }
547 
548 /* ARGSUSED */
549 static int
550 ccdopen(dev_t dev, int flags, int fmt, struct lwp *l)
551 {
552 	int unit = ccdunit(dev);
553 	struct ccd_softc *cs;
554 	struct disklabel *lp;
555 	int error = 0, part, pmask;
556 
557 #ifdef DEBUG
558 	if (ccddebug & CCDB_FOLLOW)
559 		printf("ccdopen(0x%x, 0x%x)\n", dev, flags);
560 #endif
561 	if (unit >= numccd)
562 		return (ENXIO);
563 	cs = &ccd_softc[unit];
564 
565 	mutex_enter(&cs->sc_lock);
566 
567 	lp = cs->sc_dkdev.dk_label;
568 
569 	part = DISKPART(dev);
570 	pmask = (1 << part);
571 
572 	/*
573 	 * If we're initialized, check to see if there are any other
574 	 * open partitions.  If not, then it's safe to update
575 	 * the in-core disklabel.  Only read the disklabel if it is
576 	 * not already valid.
577 	 */
578 	if ((cs->sc_flags & (CCDF_INITED|CCDF_VLABEL)) == CCDF_INITED &&
579 	    cs->sc_dkdev.dk_openmask == 0)
580 		ccdgetdisklabel(dev);
581 
582 	/* Check that the partition exists. */
583 	if (part != RAW_PART) {
584 		if (((cs->sc_flags & CCDF_INITED) == 0) ||
585 		    ((part >= lp->d_npartitions) ||
586 		     (lp->d_partitions[part].p_fstype == FS_UNUSED))) {
587 			error = ENXIO;
588 			goto done;
589 		}
590 	}
591 
592 	/* Prevent our unit from being unconfigured while open. */
593 	switch (fmt) {
594 	case S_IFCHR:
595 		cs->sc_dkdev.dk_copenmask |= pmask;
596 		break;
597 
598 	case S_IFBLK:
599 		cs->sc_dkdev.dk_bopenmask |= pmask;
600 		break;
601 	}
602 	cs->sc_dkdev.dk_openmask =
603 	    cs->sc_dkdev.dk_copenmask | cs->sc_dkdev.dk_bopenmask;
604 
605  done:
606 	mutex_exit(&cs->sc_lock);
607 	return (error);
608 }
609 
610 /* ARGSUSED */
611 static int
612 ccdclose(dev_t dev, int flags, int fmt, struct lwp *l)
613 {
614 	int unit = ccdunit(dev);
615 	struct ccd_softc *cs;
616 	int part;
617 
618 #ifdef DEBUG
619 	if (ccddebug & CCDB_FOLLOW)
620 		printf("ccdclose(0x%x, 0x%x)\n", dev, flags);
621 #endif
622 
623 	if (unit >= numccd)
624 		return (ENXIO);
625 	cs = &ccd_softc[unit];
626 
627 	mutex_enter(&cs->sc_lock);
628 
629 	part = DISKPART(dev);
630 
631 	/* ...that much closer to allowing unconfiguration... */
632 	switch (fmt) {
633 	case S_IFCHR:
634 		cs->sc_dkdev.dk_copenmask &= ~(1 << part);
635 		break;
636 
637 	case S_IFBLK:
638 		cs->sc_dkdev.dk_bopenmask &= ~(1 << part);
639 		break;
640 	}
641 	cs->sc_dkdev.dk_openmask =
642 	    cs->sc_dkdev.dk_copenmask | cs->sc_dkdev.dk_bopenmask;
643 
644 	if (cs->sc_dkdev.dk_openmask == 0) {
645 		if ((cs->sc_flags & CCDF_KLABEL) == 0)
646 			cs->sc_flags &= ~CCDF_VLABEL;
647 	}
648 
649 	mutex_exit(&cs->sc_lock);
650 	return (0);
651 }
652 
653 static void
654 ccdstrategy(struct buf *bp)
655 {
656 	int unit = ccdunit(bp->b_dev);
657 	struct ccd_softc *cs = &ccd_softc[unit];
658 	daddr_t blkno;
659 	int s;
660 	int wlabel;
661 	struct disklabel *lp;
662 
663 #ifdef DEBUG
664 	if (ccddebug & CCDB_FOLLOW)
665 		printf("ccdstrategy(%p): unit %d\n", bp, unit);
666 #endif
667 	if ((cs->sc_flags & CCDF_INITED) == 0) {
668 #ifdef DEBUG
669 		if (ccddebug & CCDB_FOLLOW)
670 			printf("ccdstrategy: unit %d: not inited\n", unit);
671 #endif
672 		bp->b_error = ENXIO;
673 		goto done;
674 	}
675 
676 	/* If it's a nil transfer, wake up the top half now. */
677 	if (bp->b_bcount == 0)
678 		goto done;
679 
680 	lp = cs->sc_dkdev.dk_label;
681 
682 	/*
683 	 * Do bounds checking and adjust transfer.  If there's an
684 	 * error, the bounds check will flag that for us.  Convert
685 	 * the partition relative block number to an absolute.
686 	 */
687 	blkno = bp->b_blkno;
688 	wlabel = cs->sc_flags & (CCDF_WLABEL|CCDF_LABELLING);
689 	if (DISKPART(bp->b_dev) != RAW_PART) {
690 		if (bounds_check_with_label(&cs->sc_dkdev, bp, wlabel) <= 0)
691 			goto done;
692 		blkno += lp->d_partitions[DISKPART(bp->b_dev)].p_offset;
693 	}
694 	bp->b_rawblkno = blkno;
695 
696 	/* Place it in the queue and start I/O on the unit. */
697 	s = splbio();
698 	BUFQ_PUT(cs->sc_bufq, bp);
699 	ccdstart(cs);
700 	splx(s);
701 	return;
702 
703  done:
704 	bp->b_resid = bp->b_bcount;
705 	biodone(bp);
706 }
707 
708 static void
709 ccdstart(struct ccd_softc *cs)
710 {
711 	long bcount, rcount;
712 	struct buf *bp;
713 	struct ccdbuf *cbp;
714 	char *addr;
715 	daddr_t bn;
716 	SIMPLEQ_HEAD(, ccdbuf) cbufq;
717 
718 #ifdef DEBUG
719 	if (ccddebug & CCDB_FOLLOW)
720 		printf("ccdstart(%p)\n", cs);
721 #endif
722 
723 	/* See if there is work for us to do. */
724 	while ((bp = BUFQ_PEEK(cs->sc_bufq)) != NULL) {
725 		/* Instrumentation. */
726 		disk_busy(&cs->sc_dkdev);
727 
728 		bp->b_resid = bp->b_bcount;
729 		bn = bp->b_rawblkno;
730 
731 		/* Allocate the component buffers. */
732 		SIMPLEQ_INIT(&cbufq);
733 		addr = bp->b_data;
734 		for (bcount = bp->b_bcount; bcount > 0; bcount -= rcount) {
735 			cbp = ccdbuffer(cs, bp, bn, addr, bcount);
736 			if (cbp == NULL) {
737 				/*
738 				 * Can't allocate a component buffer; just
739 				 * defer the job until later.
740 				 *
741 				 * XXX We might consider a watchdog timer
742 				 * XXX to make sure we are kicked into action,
743 				 * XXX or consider a low-water mark for our
744 				 * XXX component buffer pool.
745 				 */
746 				while ((cbp = SIMPLEQ_FIRST(&cbufq)) != NULL) {
747 					SIMPLEQ_REMOVE_HEAD(&cbufq, cb_q);
748 					CCD_PUTBUF(cbp);
749 				}
750 				disk_unbusy(&cs->sc_dkdev, 0, 0);
751 				return;
752 			}
753 			SIMPLEQ_INSERT_TAIL(&cbufq, cbp, cb_q);
754 			rcount = cbp->cb_buf.b_bcount;
755 			bn += btodb(rcount);
756 			addr += rcount;
757 		}
758 
759 		/* Transfer all set up, remove job from the queue. */
760 		(void) BUFQ_GET(cs->sc_bufq);
761 
762 		/* Now fire off the requests. */
763 		while ((cbp = SIMPLEQ_FIRST(&cbufq)) != NULL) {
764 			SIMPLEQ_REMOVE_HEAD(&cbufq, cb_q);
765 			if ((cbp->cb_buf.b_flags & B_READ) == 0)
766 				cbp->cb_buf.b_vp->v_numoutput++;
767 			bdev_strategy(&cbp->cb_buf);
768 		}
769 	}
770 }
771 
772 /*
773  * Build a component buffer header.
774  */
775 static struct ccdbuf *
776 ccdbuffer(struct ccd_softc *cs, struct buf *bp, daddr_t bn, void *addr,
777     long bcount)
778 {
779 	struct ccdcinfo *ci;
780 	struct ccdbuf *cbp;
781 	daddr_t cbn, cboff;
782 	u_int64_t cbc;
783 	int ccdisk;
784 
785 #ifdef DEBUG
786 	if (ccddebug & CCDB_IO)
787 		printf("ccdbuffer(%p, %p, %" PRId64 ", %p, %ld)\n",
788 		       cs, bp, bn, addr, bcount);
789 #endif
790 	/*
791 	 * Determine which component bn falls in.
792 	 */
793 	cbn = bn;
794 	cboff = 0;
795 
796 	/*
797 	 * Serially concatenated
798 	 */
799 	if (cs->sc_ileave == 0) {
800 		daddr_t sblk;
801 
802 		sblk = 0;
803 		for (ccdisk = 0, ci = &cs->sc_cinfo[ccdisk];
804 		    cbn >= sblk + ci->ci_size;
805 		    ccdisk++, ci = &cs->sc_cinfo[ccdisk])
806 			sblk += ci->ci_size;
807 		cbn -= sblk;
808 	}
809 	/*
810 	 * Interleaved
811 	 */
812 	else {
813 		struct ccdiinfo *ii;
814 		int off;
815 
816 		cboff = cbn % cs->sc_ileave;
817 		cbn /= cs->sc_ileave;
818 		for (ii = cs->sc_itable; ii->ii_ndisk; ii++)
819 			if (ii->ii_startblk > cbn)
820 				break;
821 		ii--;
822 		off = cbn - ii->ii_startblk;
823 		if (ii->ii_ndisk == 1) {
824 			ccdisk = ii->ii_index[0];
825 			cbn = ii->ii_startoff + off;
826 		} else {
827 			ccdisk = ii->ii_index[off % ii->ii_ndisk];
828 			cbn = ii->ii_startoff + off / ii->ii_ndisk;
829 		}
830 		cbn *= cs->sc_ileave;
831 		ci = &cs->sc_cinfo[ccdisk];
832 	}
833 
834 	/*
835 	 * Fill in the component buf structure.
836 	 */
837 	cbp = CCD_GETBUF();
838 	if (cbp == NULL)
839 		return (NULL);
840 	buf_init(&cbp->cb_buf);
841 	cbp->cb_buf.b_flags = bp->b_flags;
842 	cbp->cb_buf.b_oflags = bp->b_oflags;
843 	cbp->cb_buf.b_cflags = bp->b_cflags;
844 	cbp->cb_buf.b_iodone = ccdiodone;
845 	cbp->cb_buf.b_proc = bp->b_proc;
846 	cbp->cb_buf.b_dev = ci->ci_dev;
847 	cbp->cb_buf.b_blkno = cbn + cboff;
848 	cbp->cb_buf.b_data = addr;
849 	cbp->cb_buf.b_vp = ci->ci_vp;
850 	cbp->cb_buf.b_objlock = &ci->ci_vp->v_interlock;
851 	if (cs->sc_ileave == 0)
852 		cbc = dbtob((u_int64_t)(ci->ci_size - cbn));
853 	else
854 		cbc = dbtob((u_int64_t)(cs->sc_ileave - cboff));
855 	cbp->cb_buf.b_bcount = cbc < bcount ? cbc : bcount;
856 
857 	/*
858 	 * context for ccdiodone
859 	 */
860 	cbp->cb_obp = bp;
861 	cbp->cb_sc = cs;
862 	cbp->cb_comp = ccdisk;
863 
864 	BIO_COPYPRIO(&cbp->cb_buf, bp);
865 
866 #ifdef DEBUG
867 	if (ccddebug & CCDB_IO)
868 		printf(" dev 0x%x(u%lu): cbp %p bn %" PRId64 " addr %p"
869 		       " bcnt %d\n",
870 		    ci->ci_dev, (unsigned long) (ci-cs->sc_cinfo), cbp,
871 		    cbp->cb_buf.b_blkno, cbp->cb_buf.b_data,
872 		    cbp->cb_buf.b_bcount);
873 #endif
874 
875 	return (cbp);
876 }
877 
878 static void
879 ccdintr(struct ccd_softc *cs, struct buf *bp)
880 {
881 
882 #ifdef DEBUG
883 	if (ccddebug & CCDB_FOLLOW)
884 		printf("ccdintr(%p, %p)\n", cs, bp);
885 #endif
886 	/*
887 	 * Request is done for better or worse, wakeup the top half.
888 	 */
889 	if (bp->b_error != 0)
890 		bp->b_resid = bp->b_bcount;
891 	disk_unbusy(&cs->sc_dkdev, (bp->b_bcount - bp->b_resid),
892 	    (bp->b_flags & B_READ));
893 	biodone(bp);
894 }
895 
896 /*
897  * Called at interrupt time.
898  * Mark the component as done and if all components are done,
899  * take a ccd interrupt.
900  */
901 static void
902 ccdiodone(struct buf *vbp)
903 {
904 	struct ccdbuf *cbp = (struct ccdbuf *) vbp;
905 	struct buf *bp = cbp->cb_obp;
906 	struct ccd_softc *cs = cbp->cb_sc;
907 	int count, s;
908 
909 	s = splbio();
910 #ifdef DEBUG
911 	if (ccddebug & CCDB_FOLLOW)
912 		printf("ccdiodone(%p)\n", cbp);
913 	if (ccddebug & CCDB_IO) {
914 		printf("ccdiodone: bp %p bcount %d resid %d\n",
915 		       bp, bp->b_bcount, bp->b_resid);
916 		printf(" dev 0x%x(u%d), cbp %p bn %" PRId64 " addr %p"
917 		       " bcnt %d\n",
918 		       cbp->cb_buf.b_dev, cbp->cb_comp, cbp,
919 		       cbp->cb_buf.b_blkno, cbp->cb_buf.b_data,
920 		       cbp->cb_buf.b_bcount);
921 	}
922 #endif
923 
924 	if (cbp->cb_buf.b_error != 0) {
925 		bp->b_error = cbp->cb_buf.b_error;
926 		printf("%s: error %d on component %d\n",
927 		       cs->sc_xname, bp->b_error, cbp->cb_comp);
928 	}
929 	count = cbp->cb_buf.b_bcount;
930 	buf_destroy(&cbp->cb_buf);
931 	CCD_PUTBUF(cbp);
932 
933 	/*
934 	 * If all done, "interrupt".
935 	 */
936 	bp->b_resid -= count;
937 	if (bp->b_resid < 0)
938 		panic("ccdiodone: count");
939 	if (bp->b_resid == 0)
940 		ccdintr(cs, bp);
941 	splx(s);
942 }
943 
944 /* ARGSUSED */
945 static int
946 ccdread(dev_t dev, struct uio *uio, int flags)
947 {
948 	int unit = ccdunit(dev);
949 	struct ccd_softc *cs;
950 
951 #ifdef DEBUG
952 	if (ccddebug & CCDB_FOLLOW)
953 		printf("ccdread(0x%x, %p)\n", dev, uio);
954 #endif
955 	if (unit >= numccd)
956 		return (ENXIO);
957 	cs = &ccd_softc[unit];
958 
959 	if ((cs->sc_flags & CCDF_INITED) == 0)
960 		return (ENXIO);
961 
962 	return (physio(ccdstrategy, NULL, dev, B_READ, minphys, uio));
963 }
964 
965 /* ARGSUSED */
966 static int
967 ccdwrite(dev_t dev, struct uio *uio, int flags)
968 {
969 	int unit = ccdunit(dev);
970 	struct ccd_softc *cs;
971 
972 #ifdef DEBUG
973 	if (ccddebug & CCDB_FOLLOW)
974 		printf("ccdwrite(0x%x, %p)\n", dev, uio);
975 #endif
976 	if (unit >= numccd)
977 		return (ENXIO);
978 	cs = &ccd_softc[unit];
979 
980 	if ((cs->sc_flags & CCDF_INITED) == 0)
981 		return (ENXIO);
982 
983 	return (physio(ccdstrategy, NULL, dev, B_WRITE, minphys, uio));
984 }
985 
986 static int
987 ccdioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l)
988 {
989 	int unit = ccdunit(dev);
990 	int s, i, j, lookedup = 0, error = 0;
991 	int part, pmask;
992 	struct ccd_softc *cs;
993 	struct ccd_ioctl *ccio = (struct ccd_ioctl *)data;
994 	kauth_cred_t uc;
995 	char **cpp;
996 	struct vnode **vpp;
997 #ifdef __HAVE_OLD_DISKLABEL
998 	struct disklabel newlabel;
999 #endif
1000 
1001 	if (unit >= numccd)
1002 		return (ENXIO);
1003 	cs = &ccd_softc[unit];
1004 
1005 	uc = (l != NULL) ? l->l_cred : NOCRED;
1006 
1007 	/* Must be open for writes for these commands... */
1008 	switch (cmd) {
1009 	case CCDIOCSET:
1010 	case CCDIOCCLR:
1011 	case DIOCSDINFO:
1012 	case DIOCWDINFO:
1013 #ifdef __HAVE_OLD_DISKLABEL
1014 	case ODIOCSDINFO:
1015 	case ODIOCWDINFO:
1016 #endif
1017 	case DIOCKLABEL:
1018 	case DIOCWLABEL:
1019 		if ((flag & FWRITE) == 0)
1020 			return (EBADF);
1021 	}
1022 
1023 	mutex_enter(&cs->sc_lock);
1024 
1025 	/* Must be initialized for these... */
1026 	switch (cmd) {
1027 	case CCDIOCCLR:
1028 	case DIOCGDINFO:
1029 	case DIOCCACHESYNC:
1030 	case DIOCSDINFO:
1031 	case DIOCWDINFO:
1032 	case DIOCGPART:
1033 	case DIOCWLABEL:
1034 	case DIOCKLABEL:
1035 	case DIOCGDEFLABEL:
1036 #ifdef __HAVE_OLD_DISKLABEL
1037 	case ODIOCGDINFO:
1038 	case ODIOCSDINFO:
1039 	case ODIOCWDINFO:
1040 	case ODIOCGDEFLABEL:
1041 #endif
1042 		if ((cs->sc_flags & CCDF_INITED) == 0) {
1043 			error = ENXIO;
1044 			goto out;
1045 		}
1046 	}
1047 
1048 	switch (cmd) {
1049 	case CCDIOCSET:
1050 		if (cs->sc_flags & CCDF_INITED) {
1051 			error = EBUSY;
1052 			goto out;
1053 		}
1054 
1055 		/* Validate the flags. */
1056 		if ((ccio->ccio_flags & CCDF_USERMASK) != ccio->ccio_flags) {
1057 			error = EINVAL;
1058 			goto out;
1059 		}
1060 
1061 		if (ccio->ccio_ndisks > CCD_MAXNDISKS) {
1062 			error = EINVAL;
1063 			goto out;
1064 		}
1065 
1066 		/* Fill in some important bits. */
1067 		cs->sc_ileave = ccio->ccio_ileave;
1068 		cs->sc_nccdisks = ccio->ccio_ndisks;
1069 		cs->sc_flags = ccio->ccio_flags & CCDF_USERMASK;
1070 
1071 		/*
1072 		 * Allocate space for and copy in the array of
1073 		 * componet pathnames and device numbers.
1074 		 */
1075 		cpp = malloc(ccio->ccio_ndisks * sizeof(char *),
1076 		    M_DEVBUF, M_WAITOK);
1077 		vpp = malloc(ccio->ccio_ndisks * sizeof(struct vnode *),
1078 		    M_DEVBUF, M_WAITOK);
1079 
1080 		error = copyin(ccio->ccio_disks, cpp,
1081 		    ccio->ccio_ndisks * sizeof(char **));
1082 		if (error) {
1083 			free(vpp, M_DEVBUF);
1084 			free(cpp, M_DEVBUF);
1085 			goto out;
1086 		}
1087 
1088 #ifdef DEBUG
1089 		if (ccddebug & CCDB_INIT)
1090 			for (i = 0; i < ccio->ccio_ndisks; ++i)
1091 				printf("ccdioctl: component %d: %p\n",
1092 				    i, cpp[i]);
1093 #endif
1094 
1095 		for (i = 0; i < ccio->ccio_ndisks; ++i) {
1096 #ifdef DEBUG
1097 			if (ccddebug & CCDB_INIT)
1098 				printf("ccdioctl: lookedup = %d\n", lookedup);
1099 #endif
1100 			if ((error = dk_lookup(cpp[i], l, &vpp[i],
1101 			    UIO_USERSPACE)) != 0) {
1102 				for (j = 0; j < lookedup; ++j)
1103 					(void)vn_close(vpp[j], FREAD|FWRITE,
1104 					    uc, l);
1105 				free(vpp, M_DEVBUF);
1106 				free(cpp, M_DEVBUF);
1107 				goto out;
1108 			}
1109 			++lookedup;
1110 		}
1111 
1112 		/*
1113 		 * Initialize the ccd.  Fills in the softc for us.
1114 		 */
1115 		if ((error = ccdinit(cs, cpp, vpp, l)) != 0) {
1116 			for (j = 0; j < lookedup; ++j)
1117 				(void)vn_close(vpp[j], FREAD|FWRITE,
1118 				    uc, l);
1119 			free(vpp, M_DEVBUF);
1120 			free(cpp, M_DEVBUF);
1121 			goto out;
1122 		}
1123 
1124 		/* We can free the temporary variables now. */
1125 		free(vpp, M_DEVBUF);
1126 		free(cpp, M_DEVBUF);
1127 
1128 		/*
1129 		 * The ccd has been successfully initialized, so
1130 		 * we can place it into the array.  Don't try to
1131 		 * read the disklabel until the disk has been attached,
1132 		 * because space for the disklabel is allocated
1133 		 * in disk_attach();
1134 		 */
1135 		ccio->ccio_unit = unit;
1136 		ccio->ccio_size = cs->sc_size;
1137 
1138 		bufq_alloc(&cs->sc_bufq, "fcfs", 0);
1139 
1140 		/* Attach the disk. */
1141 		disk_attach(&cs->sc_dkdev);
1142 
1143 		/* Try and read the disklabel. */
1144 		ccdgetdisklabel(dev);
1145 		break;
1146 
1147 	case CCDIOCCLR:
1148 		/*
1149 		 * Don't unconfigure if any other partitions are open
1150 		 * or if both the character and block flavors of this
1151 		 * partition are open.
1152 		 */
1153 		part = DISKPART(dev);
1154 		pmask = (1 << part);
1155 		if ((cs->sc_dkdev.dk_openmask & ~pmask) ||
1156 		    ((cs->sc_dkdev.dk_bopenmask & pmask) &&
1157 		    (cs->sc_dkdev.dk_copenmask & pmask))) {
1158 			error = EBUSY;
1159 			goto out;
1160 		}
1161 
1162 		/* Kill off any queued buffers. */
1163 		s = splbio();
1164 		bufq_drain(cs->sc_bufq);
1165 		splx(s);
1166 
1167 		bufq_free(cs->sc_bufq);
1168 
1169 		/*
1170 		 * Free ccd_softc information and clear entry.
1171 		 */
1172 
1173 		/* Close the components and free their pathnames. */
1174 		for (i = 0; i < cs->sc_nccdisks; ++i) {
1175 			/*
1176 			 * XXX: this close could potentially fail and
1177 			 * cause Bad Things.  Maybe we need to force
1178 			 * the close to happen?
1179 			 */
1180 #ifdef DEBUG
1181 			if (ccddebug & CCDB_VNODE)
1182 				vprint("CCDIOCCLR: vnode info",
1183 				    cs->sc_cinfo[i].ci_vp);
1184 #endif
1185 			(void)vn_close(cs->sc_cinfo[i].ci_vp, FREAD|FWRITE,
1186 			    uc, l);
1187 			free(cs->sc_cinfo[i].ci_path, M_DEVBUF);
1188 		}
1189 
1190 		/* Free interleave index. */
1191 		for (i = 0; cs->sc_itable[i].ii_ndisk; ++i)
1192 			free(cs->sc_itable[i].ii_index, M_DEVBUF);
1193 
1194 		/* Free component info and interleave table. */
1195 		free(cs->sc_cinfo, M_DEVBUF);
1196 		free(cs->sc_itable, M_DEVBUF);
1197 		cs->sc_flags &= ~(CCDF_INITED|CCDF_VLABEL);
1198 
1199 		/* Detatch the disk. */
1200 		disk_detach(&cs->sc_dkdev);
1201 		break;
1202 
1203 	case DIOCGDINFO:
1204 		*(struct disklabel *)data = *(cs->sc_dkdev.dk_label);
1205 		break;
1206 #ifdef __HAVE_OLD_DISKLABEL
1207 	case ODIOCGDINFO:
1208 		newlabel = *(cs->sc_dkdev.dk_label);
1209 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1210 			return ENOTTY;
1211 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1212 		break;
1213 #endif
1214 
1215 	case DIOCGPART:
1216 		((struct partinfo *)data)->disklab = cs->sc_dkdev.dk_label;
1217 		((struct partinfo *)data)->part =
1218 		    &cs->sc_dkdev.dk_label->d_partitions[DISKPART(dev)];
1219 		break;
1220 
1221 	case DIOCCACHESYNC:
1222 		/*
1223 		 * XXX Do we really need to care about having a writable
1224 		 * file descriptor here?
1225 		 */
1226 		if ((flag & FWRITE) == 0)
1227 			return (EBADF);
1228 
1229 		/*
1230 		 * We pass this call down to all components and report
1231 		 * the first error we encounter.
1232 		 */
1233 		for (error = 0, i = 0; i < cs->sc_nccdisks; i++) {
1234 			j = VOP_IOCTL(cs->sc_cinfo[i].ci_vp, cmd, data,
1235 				      flag, uc);
1236 			if (j != 0 && error == 0)
1237 				error = j;
1238 		}
1239 		break;
1240 
1241 	case DIOCWDINFO:
1242 	case DIOCSDINFO:
1243 #ifdef __HAVE_OLD_DISKLABEL
1244 	case ODIOCWDINFO:
1245 	case ODIOCSDINFO:
1246 #endif
1247 	{
1248 		struct disklabel *lp;
1249 #ifdef __HAVE_OLD_DISKLABEL
1250 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
1251 			memset(&newlabel, 0, sizeof newlabel);
1252 			memcpy(&newlabel, data, sizeof (struct olddisklabel));
1253 			lp = &newlabel;
1254 		} else
1255 #endif
1256 		lp = (struct disklabel *)data;
1257 
1258 		cs->sc_flags |= CCDF_LABELLING;
1259 
1260 		error = setdisklabel(cs->sc_dkdev.dk_label,
1261 		    lp, 0, cs->sc_dkdev.dk_cpulabel);
1262 		if (error == 0) {
1263 			if (cmd == DIOCWDINFO
1264 #ifdef __HAVE_OLD_DISKLABEL
1265 			    || cmd == ODIOCWDINFO
1266 #endif
1267 			   )
1268 				error = writedisklabel(CCDLABELDEV(dev),
1269 				    ccdstrategy, cs->sc_dkdev.dk_label,
1270 				    cs->sc_dkdev.dk_cpulabel);
1271 		}
1272 
1273 		cs->sc_flags &= ~CCDF_LABELLING;
1274 		break;
1275 	}
1276 
1277 	case DIOCKLABEL:
1278 		if (*(int *)data != 0)
1279 			cs->sc_flags |= CCDF_KLABEL;
1280 		else
1281 			cs->sc_flags &= ~CCDF_KLABEL;
1282 		break;
1283 
1284 	case DIOCWLABEL:
1285 		if (*(int *)data != 0)
1286 			cs->sc_flags |= CCDF_WLABEL;
1287 		else
1288 			cs->sc_flags &= ~CCDF_WLABEL;
1289 		break;
1290 
1291 	case DIOCGDEFLABEL:
1292 		ccdgetdefaultlabel(cs, (struct disklabel *)data);
1293 		break;
1294 
1295 #ifdef __HAVE_OLD_DISKLABEL
1296 	case ODIOCGDEFLABEL:
1297 		ccdgetdefaultlabel(cs, &newlabel);
1298 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1299 			return ENOTTY;
1300 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1301 		break;
1302 #endif
1303 
1304 	default:
1305 		error = ENOTTY;
1306 	}
1307 
1308  out:
1309 	mutex_exit(&cs->sc_lock);
1310 	return (error);
1311 }
1312 
1313 static int
1314 ccdsize(dev_t dev)
1315 {
1316 	struct ccd_softc *cs;
1317 	struct disklabel *lp;
1318 	int part, unit, omask, size;
1319 
1320 	unit = ccdunit(dev);
1321 	if (unit >= numccd)
1322 		return (-1);
1323 	cs = &ccd_softc[unit];
1324 
1325 	if ((cs->sc_flags & CCDF_INITED) == 0)
1326 		return (-1);
1327 
1328 	part = DISKPART(dev);
1329 	omask = cs->sc_dkdev.dk_openmask & (1 << part);
1330 	lp = cs->sc_dkdev.dk_label;
1331 
1332 	if (omask == 0 && ccdopen(dev, 0, S_IFBLK, curlwp))
1333 		return (-1);
1334 
1335 	if (lp->d_partitions[part].p_fstype != FS_SWAP)
1336 		size = -1;
1337 	else
1338 		size = lp->d_partitions[part].p_size *
1339 		    (lp->d_secsize / DEV_BSIZE);
1340 
1341 	if (omask == 0 && ccdclose(dev, 0, S_IFBLK, curlwp))
1342 		return (-1);
1343 
1344 	return (size);
1345 }
1346 
1347 static int
1348 ccddump(dev_t dev, daddr_t blkno, void *va,
1349     size_t size)
1350 {
1351 
1352 	/* Not implemented. */
1353 	return ENXIO;
1354 }
1355 
1356 static void
1357 ccdgetdefaultlabel(struct ccd_softc *cs, struct disklabel *lp)
1358 {
1359 	struct ccdgeom *ccg = &cs->sc_geom;
1360 
1361 	memset(lp, 0, sizeof(*lp));
1362 
1363 	lp->d_secperunit = cs->sc_size;
1364 	lp->d_secsize = ccg->ccg_secsize;
1365 	lp->d_nsectors = ccg->ccg_nsectors;
1366 	lp->d_ntracks = ccg->ccg_ntracks;
1367 	lp->d_ncylinders = ccg->ccg_ncylinders;
1368 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1369 
1370 	strncpy(lp->d_typename, "ccd", sizeof(lp->d_typename));
1371 	lp->d_type = DTYPE_CCD;
1372 	strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
1373 	lp->d_rpm = 3600;
1374 	lp->d_interleave = 1;
1375 	lp->d_flags = 0;
1376 
1377 	lp->d_partitions[RAW_PART].p_offset = 0;
1378 	lp->d_partitions[RAW_PART].p_size = cs->sc_size;
1379 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
1380 	lp->d_npartitions = RAW_PART + 1;
1381 
1382 	lp->d_magic = DISKMAGIC;
1383 	lp->d_magic2 = DISKMAGIC;
1384 	lp->d_checksum = dkcksum(cs->sc_dkdev.dk_label);
1385 }
1386 
1387 /*
1388  * Read the disklabel from the ccd.  If one is not present, fake one
1389  * up.
1390  */
1391 static void
1392 ccdgetdisklabel(dev_t dev)
1393 {
1394 	int unit = ccdunit(dev);
1395 	struct ccd_softc *cs = &ccd_softc[unit];
1396 	const char *errstring;
1397 	struct disklabel *lp = cs->sc_dkdev.dk_label;
1398 	struct cpu_disklabel *clp = cs->sc_dkdev.dk_cpulabel;
1399 
1400 	memset(clp, 0, sizeof(*clp));
1401 
1402 	ccdgetdefaultlabel(cs, lp);
1403 
1404 	/*
1405 	 * Call the generic disklabel extraction routine.
1406 	 */
1407 	if ((cs->sc_flags & CCDF_NOLABEL) != 0)
1408 		errstring = "CCDF_NOLABEL set; ignoring on-disk label";
1409 	else
1410 		errstring = readdisklabel(CCDLABELDEV(dev), ccdstrategy,
1411 		    cs->sc_dkdev.dk_label, cs->sc_dkdev.dk_cpulabel);
1412 	if (errstring)
1413 		ccdmakedisklabel(cs);
1414 	else {
1415 		int i;
1416 		struct partition *pp;
1417 
1418 		/*
1419 		 * Sanity check whether the found disklabel is valid.
1420 		 *
1421 		 * This is necessary since total size of ccd may vary
1422 		 * when an interleave is changed even though exactly
1423 		 * same componets are used, and old disklabel may used
1424 		 * if that is found.
1425 		 */
1426 		if (lp->d_secperunit != cs->sc_size)
1427 			printf("WARNING: %s: "
1428 			    "total sector size in disklabel (%d) != "
1429 			    "the size of ccd (%lu)\n", cs->sc_xname,
1430 			    lp->d_secperunit, (u_long)cs->sc_size);
1431 		for (i = 0; i < lp->d_npartitions; i++) {
1432 			pp = &lp->d_partitions[i];
1433 			if (pp->p_offset + pp->p_size > cs->sc_size)
1434 				printf("WARNING: %s: end of partition `%c' "
1435 				    "exceeds the size of ccd (%lu)\n",
1436 				    cs->sc_xname, 'a' + i, (u_long)cs->sc_size);
1437 		}
1438 	}
1439 
1440 #ifdef DEBUG
1441 	/* It's actually extremely common to have unlabeled ccds. */
1442 	if (ccddebug & CCDB_LABEL)
1443 		if (errstring != NULL)
1444 			printf("%s: %s\n", cs->sc_xname, errstring);
1445 #endif
1446 
1447 	/* In-core label now valid. */
1448 	cs->sc_flags |= CCDF_VLABEL;
1449 }
1450 
1451 /*
1452  * Take care of things one might want to take care of in the event
1453  * that a disklabel isn't present.
1454  */
1455 static void
1456 ccdmakedisklabel(struct ccd_softc *cs)
1457 {
1458 	struct disklabel *lp = cs->sc_dkdev.dk_label;
1459 
1460 	/*
1461 	 * For historical reasons, if there's no disklabel present
1462 	 * the raw partition must be marked FS_BSDFFS.
1463 	 */
1464 	lp->d_partitions[RAW_PART].p_fstype = FS_BSDFFS;
1465 
1466 	strncpy(lp->d_packname, "default label", sizeof(lp->d_packname));
1467 
1468 	lp->d_checksum = dkcksum(lp);
1469 }
1470 
1471 #ifdef DEBUG
1472 static void
1473 printiinfo(struct ccdiinfo *ii)
1474 {
1475 	int ix, i;
1476 
1477 	for (ix = 0; ii->ii_ndisk; ix++, ii++) {
1478 		printf(" itab[%d]: #dk %d sblk %" PRId64 " soff %" PRId64,
1479 		    ix, ii->ii_ndisk, ii->ii_startblk, ii->ii_startoff);
1480 		for (i = 0; i < ii->ii_ndisk; i++)
1481 			printf(" %d", ii->ii_index[i]);
1482 		printf("\n");
1483 	}
1484 }
1485 #endif
1486