xref: /netbsd-src/sys/dev/ccd.c (revision 23c8222edbfb0f0932d88a8351d3a0cf817dfb9e)
1 /*	$NetBSD: ccd.c,v 1.101 2004/10/28 07:07:39 yamt Exp $	*/
2 
3 /*-
4  * Copyright (c) 1996, 1997, 1998, 1999 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Jason R. Thorpe.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *        This product includes software developed by the NetBSD
21  *        Foundation, Inc. and its contributors.
22  * 4. Neither the name of The NetBSD Foundation nor the names of its
23  *    contributors may be used to endorse or promote products derived
24  *    from this software without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
27  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
28  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
29  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
30  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36  * POSSIBILITY OF SUCH DAMAGE.
37  */
38 
39 /*
40  * Copyright (c) 1990, 1993
41  *	The Regents of the University of California.  All rights reserved.
42  *
43  * This code is derived from software contributed to Berkeley by
44  * the Systems Programming Group of the University of Utah Computer
45  * Science Department.
46  *
47  * Redistribution and use in source and binary forms, with or without
48  * modification, are permitted provided that the following conditions
49  * are met:
50  * 1. Redistributions of source code must retain the above copyright
51  *    notice, this list of conditions and the following disclaimer.
52  * 2. Redistributions in binary form must reproduce the above copyright
53  *    notice, this list of conditions and the following disclaimer in the
54  *    documentation and/or other materials provided with the distribution.
55  * 3. Neither the name of the University nor the names of its contributors
56  *    may be used to endorse or promote products derived from this software
57  *    without specific prior written permission.
58  *
59  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
60  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
61  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
62  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
63  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
64  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
65  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
66  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
67  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
68  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
69  * SUCH DAMAGE.
70  *
71  * from: Utah $Hdr: cd.c 1.6 90/11/28$
72  *
73  *	@(#)cd.c	8.2 (Berkeley) 11/16/93
74  */
75 
76 /*
77  * Copyright (c) 1988 University of Utah.
78  *
79  * This code is derived from software contributed to Berkeley by
80  * the Systems Programming Group of the University of Utah Computer
81  * Science Department.
82  *
83  * Redistribution and use in source and binary forms, with or without
84  * modification, are permitted provided that the following conditions
85  * are met:
86  * 1. Redistributions of source code must retain the above copyright
87  *    notice, this list of conditions and the following disclaimer.
88  * 2. Redistributions in binary form must reproduce the above copyright
89  *    notice, this list of conditions and the following disclaimer in the
90  *    documentation and/or other materials provided with the distribution.
91  * 3. All advertising materials mentioning features or use of this software
92  *    must display the following acknowledgement:
93  *	This product includes software developed by the University of
94  *	California, Berkeley and its contributors.
95  * 4. Neither the name of the University nor the names of its contributors
96  *    may be used to endorse or promote products derived from this software
97  *    without specific prior written permission.
98  *
99  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
100  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
101  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
102  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
103  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
104  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
105  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
106  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
107  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
108  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
109  * SUCH DAMAGE.
110  *
111  * from: Utah $Hdr: cd.c 1.6 90/11/28$
112  *
113  *	@(#)cd.c	8.2 (Berkeley) 11/16/93
114  */
115 
116 /*
117  * "Concatenated" disk driver.
118  *
119  * Dynamic configuration and disklabel support by:
120  *	Jason R. Thorpe <thorpej@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 1.101 2004/10/28 07:07:39 yamt Exp $");
129 
130 #include <sys/param.h>
131 #include <sys/systm.h>
132 #include <sys/proc.h>
133 #include <sys/errno.h>
134 #include <sys/buf.h>
135 #include <sys/bufq.h>
136 #include <sys/malloc.h>
137 #include <sys/pool.h>
138 #include <sys/namei.h>
139 #include <sys/stat.h>
140 #include <sys/ioctl.h>
141 #include <sys/disklabel.h>
142 #include <sys/device.h>
143 #include <sys/disk.h>
144 #include <sys/syslog.h>
145 #include <sys/fcntl.h>
146 #include <sys/vnode.h>
147 #include <sys/conf.h>
148 #include <sys/lock.h>
149 #include <sys/queue.h>
150 
151 #include <dev/ccdvar.h>
152 
153 #if defined(CCDDEBUG) && !defined(DEBUG)
154 #define DEBUG
155 #endif
156 
157 #ifdef DEBUG
158 #define CCDB_FOLLOW	0x01
159 #define CCDB_INIT	0x02
160 #define CCDB_IO		0x04
161 #define CCDB_LABEL	0x08
162 #define CCDB_VNODE	0x10
163 int ccddebug = 0x00;
164 #endif
165 
166 #define	ccdunit(x)	DISKUNIT(x)
167 
168 struct ccdbuf {
169 	struct buf	cb_buf;		/* new I/O buf */
170 	struct buf	*cb_obp;	/* ptr. to original I/O buf */
171 	struct ccd_softc *cb_sc;	/* pointer to ccd softc */
172 	int		cb_comp;	/* target component */
173 	SIMPLEQ_ENTRY(ccdbuf) cb_q;	/* fifo of component buffers */
174 };
175 
176 /* component buffer pool */
177 static struct pool ccd_cbufpool;
178 
179 #define	CCD_GETBUF()		pool_get(&ccd_cbufpool, PR_NOWAIT)
180 #define	CCD_PUTBUF(cbp)		pool_put(&ccd_cbufpool, cbp)
181 
182 #define CCDLABELDEV(dev)	\
183 	(MAKEDISKDEV(major((dev)), ccdunit((dev)), RAW_PART))
184 
185 /* called by main() at boot time */
186 void	ccdattach(int);
187 
188 /* called by biodone() at interrupt time */
189 static void	ccdiodone(struct buf *);
190 
191 static void	ccdstart(struct ccd_softc *);
192 static void	ccdinterleave(struct ccd_softc *);
193 static void	ccdintr(struct ccd_softc *, struct buf *);
194 static int	ccdinit(struct ccd_softc *, char **, struct vnode **,
195 		    struct proc *);
196 static int	ccdlookup(char *, struct proc *p, struct vnode **);
197 static struct ccdbuf *ccdbuffer(struct ccd_softc *, struct buf *,
198 		    daddr_t, caddr_t, long);
199 static void	ccdgetdefaultlabel(struct ccd_softc *, struct disklabel *);
200 static void	ccdgetdisklabel(dev_t);
201 static void	ccdmakedisklabel(struct ccd_softc *);
202 
203 static dev_type_open(ccdopen);
204 static dev_type_close(ccdclose);
205 static dev_type_read(ccdread);
206 static dev_type_write(ccdwrite);
207 static dev_type_ioctl(ccdioctl);
208 static dev_type_strategy(ccdstrategy);
209 static dev_type_dump(ccddump);
210 static dev_type_size(ccdsize);
211 
212 const struct bdevsw ccd_bdevsw = {
213 	ccdopen, ccdclose, ccdstrategy, ccdioctl, ccddump, ccdsize, D_DISK
214 };
215 
216 const struct cdevsw ccd_cdevsw = {
217 	ccdopen, ccdclose, ccdread, ccdwrite, ccdioctl,
218 	nostop, notty, nopoll, nommap, nokqfilter, D_DISK
219 };
220 
221 #ifdef DEBUG
222 static	void printiinfo(struct ccdiinfo *);
223 #endif
224 
225 /* Non-private for the benefit of libkvm. */
226 struct	ccd_softc *ccd_softc;
227 int	numccd = 0;
228 
229 /*
230  * Called by main() during pseudo-device attachment.  All we need
231  * to do is allocate enough space for devices to be configured later.
232  */
233 void
234 ccdattach(int num)
235 {
236 	struct ccd_softc *cs;
237 	int i;
238 
239 	if (num <= 0) {
240 #ifdef DIAGNOSTIC
241 		panic("ccdattach: count <= 0");
242 #endif
243 		return;
244 	}
245 
246 	ccd_softc = (struct ccd_softc *)malloc(num * sizeof(struct ccd_softc),
247 	    M_DEVBUF, M_NOWAIT|M_ZERO);
248 	if (ccd_softc == NULL) {
249 		printf("WARNING: no memory for concatenated disks\n");
250 		return;
251 	}
252 	numccd = num;
253 
254 	/* Initialize the component buffer pool. */
255 	pool_init(&ccd_cbufpool, sizeof(struct ccdbuf), 0,
256 	    0, 0, "ccdpl", NULL);
257 
258 	/* Initialize per-softc structures. */
259 	for (i = 0; i < num; i++) {
260 		cs = &ccd_softc[i];
261 		snprintf(cs->sc_xname, sizeof(cs->sc_xname), "ccd%d", i);
262 		cs->sc_dkdev.dk_name = cs->sc_xname;	/* XXX */
263 		lockinit(&cs->sc_lock, PRIBIO, "ccdlk", 0, 0);
264 	}
265 }
266 
267 static int
268 ccdinit(struct ccd_softc *cs, char **cpaths, struct vnode **vpp,
269     struct proc *p)
270 {
271 	struct ccdcinfo *ci = NULL;
272 	size_t size;
273 	int ix;
274 	struct vattr va;
275 	size_t minsize;
276 	int maxsecsize;
277 	struct partinfo dpart;
278 	struct ccdgeom *ccg = &cs->sc_geom;
279 	char tmppath[MAXPATHLEN];
280 	int error, path_alloced;
281 
282 #ifdef DEBUG
283 	if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
284 		printf("%s: ccdinit\n", cs->sc_xname);
285 #endif
286 
287 	/* Allocate space for the component info. */
288 	cs->sc_cinfo = malloc(cs->sc_nccdisks * sizeof(struct ccdcinfo),
289 	    M_DEVBUF, M_WAITOK);
290 
291 	cs->sc_size = 0;
292 
293 	/*
294 	 * Verify that each component piece exists and record
295 	 * relevant information about it.
296 	 */
297 	maxsecsize = 0;
298 	minsize = 0;
299 	for (ix = 0, path_alloced = 0; ix < cs->sc_nccdisks; ix++) {
300 		ci = &cs->sc_cinfo[ix];
301 		ci->ci_vp = vpp[ix];
302 
303 		/*
304 		 * Copy in the pathname of the component.
305 		 */
306 		memset(tmppath, 0, sizeof(tmppath));	/* sanity */
307 		error = copyinstr(cpaths[ix], tmppath,
308 		    MAXPATHLEN, &ci->ci_pathlen);
309 		if (error) {
310 #ifdef DEBUG
311 			if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
312 				printf("%s: can't copy path, error = %d\n",
313 				    cs->sc_xname, error);
314 #endif
315 			goto out;
316 		}
317 		ci->ci_path = malloc(ci->ci_pathlen, M_DEVBUF, M_WAITOK);
318 		memcpy(ci->ci_path, tmppath, ci->ci_pathlen);
319 		path_alloced++;
320 
321 		/*
322 		 * XXX: Cache the component's dev_t.
323 		 */
324 		if ((error = VOP_GETATTR(vpp[ix], &va, p->p_ucred, p)) != 0) {
325 #ifdef DEBUG
326 			if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
327 				printf("%s: %s: getattr failed %s = %d\n",
328 				    cs->sc_xname, ci->ci_path,
329 				    "error", error);
330 #endif
331 			goto out;
332 		}
333 		ci->ci_dev = va.va_rdev;
334 
335 		/*
336 		 * Get partition information for the component.
337 		 */
338 		error = VOP_IOCTL(vpp[ix], DIOCGPART, &dpart,
339 		    FREAD, p->p_ucred, p);
340 		if (error) {
341 #ifdef DEBUG
342 			if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
343 				 printf("%s: %s: ioctl failed, error = %d\n",
344 				     cs->sc_xname, ci->ci_path, error);
345 #endif
346 			goto out;
347 		}
348 
349 /*
350  * This diagnostic test is disabled (for now?) since not all port supports
351  * on-disk BSD disklabel.
352  */
353 #if 0 /* def DIAGNOSTIC */
354 		/* Check fstype field of component. */
355 		if (dpart.part->p_fstype != FS_CCD)
356 			printf("%s: WARNING: %s: fstype %d != FS_CCD\n",
357 			    cs->sc_xname, ci->ci_path, dpart.part->p_fstype);
358 #endif
359 
360 		/*
361 		 * Calculate the size, truncating to an interleave
362 		 * boundary if necessary.
363 		 */
364 		maxsecsize =
365 		    ((dpart.disklab->d_secsize > maxsecsize) ?
366 		    dpart.disklab->d_secsize : maxsecsize);
367 		size = dpart.part->p_size;
368 		if (cs->sc_ileave > 1)
369 			size -= size % cs->sc_ileave;
370 
371 		if (size == 0) {
372 #ifdef DEBUG
373 			if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
374 				printf("%s: %s: size == 0\n",
375 				    cs->sc_xname, ci->ci_path);
376 #endif
377 			error = ENODEV;
378 			goto out;
379 		}
380 
381 		if (minsize == 0 || size < minsize)
382 			minsize = size;
383 		ci->ci_size = size;
384 		cs->sc_size += size;
385 	}
386 
387 	/*
388 	 * Don't allow the interleave to be smaller than
389 	 * the biggest component sector.
390 	 */
391 	if ((cs->sc_ileave > 0) &&
392 	    (cs->sc_ileave < (maxsecsize / DEV_BSIZE))) {
393 #ifdef DEBUG
394 		if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
395 			printf("%s: interleave must be at least %d\n",
396 			    cs->sc_xname, (maxsecsize / DEV_BSIZE));
397 #endif
398 		error = EINVAL;
399 		goto out;
400 	}
401 
402 	/*
403 	 * If uniform interleave is desired set all sizes to that of
404 	 * the smallest component.
405 	 */
406 	if (cs->sc_flags & CCDF_UNIFORM) {
407 		for (ci = cs->sc_cinfo;
408 		     ci < &cs->sc_cinfo[cs->sc_nccdisks]; ci++)
409 			ci->ci_size = minsize;
410 
411 		cs->sc_size = cs->sc_nccdisks * minsize;
412 	}
413 
414 	/*
415 	 * Construct the interleave table.
416 	 */
417 	ccdinterleave(cs);
418 
419 	/*
420 	 * Create pseudo-geometry based on 1MB cylinders.  It's
421 	 * pretty close.
422 	 */
423 	ccg->ccg_secsize = DEV_BSIZE;
424 	ccg->ccg_ntracks = 1;
425 	ccg->ccg_nsectors = 1024 * (1024 / ccg->ccg_secsize);
426 	ccg->ccg_ncylinders = cs->sc_size / ccg->ccg_nsectors;
427 
428 	cs->sc_flags |= CCDF_INITED;
429 
430 	return (0);
431 
432  out:
433 	for (ix = 0; ix < path_alloced; ix++)
434 		free(cs->sc_cinfo[ix].ci_path, M_DEVBUF);
435 	free(cs->sc_cinfo, M_DEVBUF);
436 	return (error);
437 }
438 
439 static void
440 ccdinterleave(struct ccd_softc *cs)
441 {
442 	struct ccdcinfo *ci, *smallci;
443 	struct ccdiinfo *ii;
444 	daddr_t bn, lbn;
445 	int ix;
446 	u_long size;
447 
448 #ifdef DEBUG
449 	if (ccddebug & CCDB_INIT)
450 		printf("ccdinterleave(%p): ileave %d\n", cs, cs->sc_ileave);
451 #endif
452 	/*
453 	 * Allocate an interleave table.
454 	 * Chances are this is too big, but we don't care.
455 	 */
456 	size = (cs->sc_nccdisks + 1) * sizeof(struct ccdiinfo);
457 	cs->sc_itable = (struct ccdiinfo *)malloc(size, M_DEVBUF,
458 	    M_WAITOK|M_ZERO);
459 
460 	/*
461 	 * Trivial case: no interleave (actually interleave of disk size).
462 	 * Each table entry represents a single component in its entirety.
463 	 */
464 	if (cs->sc_ileave == 0) {
465 		bn = 0;
466 		ii = cs->sc_itable;
467 
468 		for (ix = 0; ix < cs->sc_nccdisks; ix++) {
469 			/* Allocate space for ii_index. */
470 			ii->ii_index = malloc(sizeof(int), M_DEVBUF, M_WAITOK);
471 			ii->ii_ndisk = 1;
472 			ii->ii_startblk = bn;
473 			ii->ii_startoff = 0;
474 			ii->ii_index[0] = ix;
475 			bn += cs->sc_cinfo[ix].ci_size;
476 			ii++;
477 		}
478 		ii->ii_ndisk = 0;
479 #ifdef DEBUG
480 		if (ccddebug & CCDB_INIT)
481 			printiinfo(cs->sc_itable);
482 #endif
483 		return;
484 	}
485 
486 	/*
487 	 * The following isn't fast or pretty; it doesn't have to be.
488 	 */
489 	size = 0;
490 	bn = lbn = 0;
491 	for (ii = cs->sc_itable; ; ii++) {
492 		/* Allocate space for ii_index. */
493 		ii->ii_index = malloc((sizeof(int) * cs->sc_nccdisks),
494 		    M_DEVBUF, M_WAITOK);
495 
496 		/*
497 		 * Locate the smallest of the remaining components
498 		 */
499 		smallci = NULL;
500 		for (ci = cs->sc_cinfo;
501 		     ci < &cs->sc_cinfo[cs->sc_nccdisks]; ci++)
502 			if (ci->ci_size > size &&
503 			    (smallci == NULL ||
504 			     ci->ci_size < smallci->ci_size))
505 				smallci = ci;
506 
507 		/*
508 		 * Nobody left, all done
509 		 */
510 		if (smallci == NULL) {
511 			ii->ii_ndisk = 0;
512 			break;
513 		}
514 
515 		/*
516 		 * Record starting logical block and component offset
517 		 */
518 		ii->ii_startblk = bn / cs->sc_ileave;
519 		ii->ii_startoff = lbn;
520 
521 		/*
522 		 * Determine how many disks take part in this interleave
523 		 * and record their indices.
524 		 */
525 		ix = 0;
526 		for (ci = cs->sc_cinfo;
527 		     ci < &cs->sc_cinfo[cs->sc_nccdisks]; ci++)
528 			if (ci->ci_size >= smallci->ci_size)
529 				ii->ii_index[ix++] = ci - cs->sc_cinfo;
530 		ii->ii_ndisk = ix;
531 		bn += ix * (smallci->ci_size - size);
532 		lbn = smallci->ci_size / cs->sc_ileave;
533 		size = smallci->ci_size;
534 	}
535 #ifdef DEBUG
536 	if (ccddebug & CCDB_INIT)
537 		printiinfo(cs->sc_itable);
538 #endif
539 }
540 
541 /* ARGSUSED */
542 static int
543 ccdopen(dev_t dev, int flags, int fmt, struct proc *p)
544 {
545 	int unit = ccdunit(dev);
546 	struct ccd_softc *cs;
547 	struct disklabel *lp;
548 	int error = 0, part, pmask;
549 
550 #ifdef DEBUG
551 	if (ccddebug & CCDB_FOLLOW)
552 		printf("ccdopen(0x%x, 0x%x)\n", dev, flags);
553 #endif
554 	if (unit >= numccd)
555 		return (ENXIO);
556 	cs = &ccd_softc[unit];
557 
558 	if ((error = lockmgr(&cs->sc_lock, LK_EXCLUSIVE, NULL)) != 0)
559 		return (error);
560 
561 	lp = cs->sc_dkdev.dk_label;
562 
563 	part = DISKPART(dev);
564 	pmask = (1 << part);
565 
566 	/*
567 	 * If we're initialized, check to see if there are any other
568 	 * open partitions.  If not, then it's safe to update
569 	 * the in-core disklabel.  Only read the disklabel if it is
570 	 * not already valid.
571 	 */
572 	if ((cs->sc_flags & (CCDF_INITED|CCDF_VLABEL)) == CCDF_INITED &&
573 	    cs->sc_dkdev.dk_openmask == 0)
574 		ccdgetdisklabel(dev);
575 
576 	/* Check that the partition exists. */
577 	if (part != RAW_PART) {
578 		if (((cs->sc_flags & CCDF_INITED) == 0) ||
579 		    ((part >= lp->d_npartitions) ||
580 		     (lp->d_partitions[part].p_fstype == FS_UNUSED))) {
581 			error = ENXIO;
582 			goto done;
583 		}
584 	}
585 
586 	/* Prevent our unit from being unconfigured while open. */
587 	switch (fmt) {
588 	case S_IFCHR:
589 		cs->sc_dkdev.dk_copenmask |= pmask;
590 		break;
591 
592 	case S_IFBLK:
593 		cs->sc_dkdev.dk_bopenmask |= pmask;
594 		break;
595 	}
596 	cs->sc_dkdev.dk_openmask =
597 	    cs->sc_dkdev.dk_copenmask | cs->sc_dkdev.dk_bopenmask;
598 
599  done:
600 	(void) lockmgr(&cs->sc_lock, LK_RELEASE, NULL);
601 	return (error);
602 }
603 
604 /* ARGSUSED */
605 static int
606 ccdclose(dev_t dev, int flags, int fmt, struct proc *p)
607 {
608 	int unit = ccdunit(dev);
609 	struct ccd_softc *cs;
610 	int error = 0, part;
611 
612 #ifdef DEBUG
613 	if (ccddebug & CCDB_FOLLOW)
614 		printf("ccdclose(0x%x, 0x%x)\n", dev, flags);
615 #endif
616 
617 	if (unit >= numccd)
618 		return (ENXIO);
619 	cs = &ccd_softc[unit];
620 
621 	if ((error = lockmgr(&cs->sc_lock, LK_EXCLUSIVE, NULL)) != 0)
622 		return (error);
623 
624 	part = DISKPART(dev);
625 
626 	/* ...that much closer to allowing unconfiguration... */
627 	switch (fmt) {
628 	case S_IFCHR:
629 		cs->sc_dkdev.dk_copenmask &= ~(1 << part);
630 		break;
631 
632 	case S_IFBLK:
633 		cs->sc_dkdev.dk_bopenmask &= ~(1 << part);
634 		break;
635 	}
636 	cs->sc_dkdev.dk_openmask =
637 	    cs->sc_dkdev.dk_copenmask | cs->sc_dkdev.dk_bopenmask;
638 
639 	if (cs->sc_dkdev.dk_openmask == 0) {
640 		if ((cs->sc_flags & CCDF_KLABEL) == 0)
641 			cs->sc_flags &= ~CCDF_VLABEL;
642 	}
643 
644 	(void) lockmgr(&cs->sc_lock, LK_RELEASE, NULL);
645 	return (0);
646 }
647 
648 static void
649 ccdstrategy(struct buf *bp)
650 {
651 	int unit = ccdunit(bp->b_dev);
652 	struct ccd_softc *cs = &ccd_softc[unit];
653 	daddr_t blkno;
654 	int s;
655 	int wlabel;
656 	struct disklabel *lp;
657 
658 #ifdef DEBUG
659 	if (ccddebug & CCDB_FOLLOW)
660 		printf("ccdstrategy(%p): unit %d\n", bp, unit);
661 #endif
662 	if ((cs->sc_flags & CCDF_INITED) == 0) {
663 #ifdef DEBUG
664 		if (ccddebug & CCDB_FOLLOW)
665 			printf("ccdstrategy: unit %d: not inited\n", unit);
666 #endif
667 		bp->b_error = ENXIO;
668 		bp->b_flags |= B_ERROR;
669 		goto done;
670 	}
671 
672 	/* If it's a nil transfer, wake up the top half now. */
673 	if (bp->b_bcount == 0)
674 		goto done;
675 
676 	lp = cs->sc_dkdev.dk_label;
677 
678 	/*
679 	 * Do bounds checking and adjust transfer.  If there's an
680 	 * error, the bounds check will flag that for us.  Convert
681 	 * the partition relative block number to an absolute.
682 	 */
683 	blkno = bp->b_blkno;
684 	wlabel = cs->sc_flags & (CCDF_WLABEL|CCDF_LABELLING);
685 	if (DISKPART(bp->b_dev) != RAW_PART) {
686 		if (bounds_check_with_label(&cs->sc_dkdev, bp, wlabel) <= 0)
687 			goto done;
688 		blkno += lp->d_partitions[DISKPART(bp->b_dev)].p_offset;
689 	}
690 	bp->b_rawblkno = blkno;
691 
692 	/* Place it in the queue and start I/O on the unit. */
693 	s = splbio();
694 	BUFQ_PUT(&cs->sc_bufq, bp);
695 	ccdstart(cs);
696 	splx(s);
697 	return;
698 
699  done:
700 	bp->b_resid = bp->b_bcount;
701 	biodone(bp);
702 }
703 
704 static void
705 ccdstart(struct ccd_softc *cs)
706 {
707 	long bcount, rcount;
708 	struct buf *bp;
709 	struct ccdbuf *cbp;
710 	caddr_t addr;
711 	daddr_t bn;
712 	SIMPLEQ_HEAD(, ccdbuf) cbufq;
713 
714 #ifdef DEBUG
715 	if (ccddebug & CCDB_FOLLOW)
716 		printf("ccdstart(%p)\n", cs);
717 #endif
718 
719 	/* See if there is work for us to do. */
720 	while ((bp = BUFQ_PEEK(&cs->sc_bufq)) != NULL) {
721 		/* Instrumentation. */
722 		disk_busy(&cs->sc_dkdev);
723 
724 		bp->b_resid = bp->b_bcount;
725 		bn = bp->b_rawblkno;
726 
727 		/* Allocate the component buffers. */
728 		SIMPLEQ_INIT(&cbufq);
729 		addr = bp->b_data;
730 		for (bcount = bp->b_bcount; bcount > 0; bcount -= rcount) {
731 			cbp = ccdbuffer(cs, bp, bn, addr, bcount);
732 			if (cbp == NULL) {
733 				/*
734 				 * Can't allocate a component buffer; just
735 				 * defer the job until later.
736 				 *
737 				 * XXX We might consider a watchdog timer
738 				 * XXX to make sure we are kicked into action,
739 				 * XXX or consider a low-water mark for our
740 				 * XXX component buffer pool.
741 				 */
742 				while ((cbp = SIMPLEQ_FIRST(&cbufq)) != NULL) {
743 					SIMPLEQ_REMOVE_HEAD(&cbufq, cb_q);
744 					CCD_PUTBUF(cbp);
745 				}
746 				disk_unbusy(&cs->sc_dkdev, 0, 0);
747 				return;
748 			}
749 			SIMPLEQ_INSERT_TAIL(&cbufq, cbp, cb_q);
750 			rcount = cbp->cb_buf.b_bcount;
751 			bn += btodb(rcount);
752 			addr += rcount;
753 		}
754 
755 		/* Transfer all set up, remove job from the queue. */
756 		(void) BUFQ_GET(&cs->sc_bufq);
757 
758 		/* Now fire off the requests. */
759 		while ((cbp = SIMPLEQ_FIRST(&cbufq)) != NULL) {
760 			SIMPLEQ_REMOVE_HEAD(&cbufq, cb_q);
761 			if ((cbp->cb_buf.b_flags & B_READ) == 0)
762 				cbp->cb_buf.b_vp->v_numoutput++;
763 			DEV_STRATEGY(&cbp->cb_buf);
764 		}
765 	}
766 }
767 
768 /*
769  * Build a component buffer header.
770  */
771 static struct ccdbuf *
772 ccdbuffer(struct ccd_softc *cs, struct buf *bp, daddr_t bn, caddr_t addr,
773     long bcount)
774 {
775 	struct ccdcinfo *ci;
776 	struct ccdbuf *cbp;
777 	daddr_t cbn, cboff;
778 	u_int64_t cbc;
779 	int ccdisk;
780 
781 #ifdef DEBUG
782 	if (ccddebug & CCDB_IO)
783 		printf("ccdbuffer(%p, %p, %" PRId64 ", %p, %ld)\n",
784 		       cs, bp, bn, addr, bcount);
785 #endif
786 	/*
787 	 * Determine which component bn falls in.
788 	 */
789 	cbn = bn;
790 	cboff = 0;
791 
792 	/*
793 	 * Serially concatenated
794 	 */
795 	if (cs->sc_ileave == 0) {
796 		daddr_t sblk;
797 
798 		sblk = 0;
799 		for (ccdisk = 0, ci = &cs->sc_cinfo[ccdisk];
800 		    cbn >= sblk + ci->ci_size;
801 		    ccdisk++, ci = &cs->sc_cinfo[ccdisk])
802 			sblk += ci->ci_size;
803 		cbn -= sblk;
804 	}
805 	/*
806 	 * Interleaved
807 	 */
808 	else {
809 		struct ccdiinfo *ii;
810 		int off;
811 
812 		cboff = cbn % cs->sc_ileave;
813 		cbn /= cs->sc_ileave;
814 		for (ii = cs->sc_itable; ii->ii_ndisk; ii++)
815 			if (ii->ii_startblk > cbn)
816 				break;
817 		ii--;
818 		off = cbn - ii->ii_startblk;
819 		if (ii->ii_ndisk == 1) {
820 			ccdisk = ii->ii_index[0];
821 			cbn = ii->ii_startoff + off;
822 		} else {
823 			ccdisk = ii->ii_index[off % ii->ii_ndisk];
824 			cbn = ii->ii_startoff + off / ii->ii_ndisk;
825 		}
826 		cbn *= cs->sc_ileave;
827 		ci = &cs->sc_cinfo[ccdisk];
828 	}
829 
830 	/*
831 	 * Fill in the component buf structure.
832 	 */
833 	cbp = CCD_GETBUF();
834 	if (cbp == NULL)
835 		return (NULL);
836 	BUF_INIT(&cbp->cb_buf);
837 	cbp->cb_buf.b_flags = bp->b_flags | B_CALL;
838 	cbp->cb_buf.b_iodone = ccdiodone;
839 	cbp->cb_buf.b_proc = bp->b_proc;
840 	cbp->cb_buf.b_dev = ci->ci_dev;
841 	cbp->cb_buf.b_blkno = cbn + cboff;
842 	cbp->cb_buf.b_data = addr;
843 	cbp->cb_buf.b_vp = ci->ci_vp;
844 	if (cs->sc_ileave == 0)
845 		cbc = dbtob((u_int64_t)(ci->ci_size - cbn));
846 	else
847 		cbc = dbtob((u_int64_t)(cs->sc_ileave - cboff));
848 	cbp->cb_buf.b_bcount = cbc < bcount ? cbc : bcount;
849 
850 	/*
851 	 * context for ccdiodone
852 	 */
853 	cbp->cb_obp = bp;
854 	cbp->cb_sc = cs;
855 	cbp->cb_comp = ccdisk;
856 
857 	BIO_COPYPRIO(&cbp->cb_buf, bp);
858 
859 #ifdef DEBUG
860 	if (ccddebug & CCDB_IO)
861 		printf(" dev 0x%x(u%lu): cbp %p bn %" PRId64 " addr %p"
862 		       " bcnt %d\n",
863 		    ci->ci_dev, (unsigned long) (ci-cs->sc_cinfo), cbp,
864 		    cbp->cb_buf.b_blkno, cbp->cb_buf.b_data,
865 		    cbp->cb_buf.b_bcount);
866 #endif
867 
868 	return (cbp);
869 }
870 
871 static void
872 ccdintr(struct ccd_softc *cs, struct buf *bp)
873 {
874 
875 #ifdef DEBUG
876 	if (ccddebug & CCDB_FOLLOW)
877 		printf("ccdintr(%p, %p)\n", cs, bp);
878 #endif
879 	/*
880 	 * Request is done for better or worse, wakeup the top half.
881 	 */
882 	if (bp->b_flags & B_ERROR)
883 		bp->b_resid = bp->b_bcount;
884 	disk_unbusy(&cs->sc_dkdev, (bp->b_bcount - bp->b_resid),
885 	    (bp->b_flags & B_READ));
886 	biodone(bp);
887 }
888 
889 /*
890  * Called at interrupt time.
891  * Mark the component as done and if all components are done,
892  * take a ccd interrupt.
893  */
894 static void
895 ccdiodone(struct buf *vbp)
896 {
897 	struct ccdbuf *cbp = (struct ccdbuf *) vbp;
898 	struct buf *bp = cbp->cb_obp;
899 	struct ccd_softc *cs = cbp->cb_sc;
900 	int count, s;
901 
902 	s = splbio();
903 #ifdef DEBUG
904 	if (ccddebug & CCDB_FOLLOW)
905 		printf("ccdiodone(%p)\n", cbp);
906 	if (ccddebug & CCDB_IO) {
907 		printf("ccdiodone: bp %p bcount %d resid %d\n",
908 		       bp, bp->b_bcount, bp->b_resid);
909 		printf(" dev 0x%x(u%d), cbp %p bn %" PRId64 " addr %p"
910 		       " bcnt %d\n",
911 		       cbp->cb_buf.b_dev, cbp->cb_comp, cbp,
912 		       cbp->cb_buf.b_blkno, cbp->cb_buf.b_data,
913 		       cbp->cb_buf.b_bcount);
914 	}
915 #endif
916 
917 	if (cbp->cb_buf.b_flags & B_ERROR) {
918 		bp->b_flags |= B_ERROR;
919 		bp->b_error = cbp->cb_buf.b_error ?
920 		    cbp->cb_buf.b_error : EIO;
921 
922 		printf("%s: error %d on component %d\n",
923 		       cs->sc_xname, bp->b_error, cbp->cb_comp);
924 	}
925 	count = cbp->cb_buf.b_bcount;
926 	CCD_PUTBUF(cbp);
927 
928 	/*
929 	 * If all done, "interrupt".
930 	 */
931 	bp->b_resid -= count;
932 	if (bp->b_resid < 0)
933 		panic("ccdiodone: count");
934 	if (bp->b_resid == 0)
935 		ccdintr(cs, bp);
936 	splx(s);
937 }
938 
939 /* ARGSUSED */
940 static int
941 ccdread(dev_t dev, struct uio *uio, int flags)
942 {
943 	int unit = ccdunit(dev);
944 	struct ccd_softc *cs;
945 
946 #ifdef DEBUG
947 	if (ccddebug & CCDB_FOLLOW)
948 		printf("ccdread(0x%x, %p)\n", dev, uio);
949 #endif
950 	if (unit >= numccd)
951 		return (ENXIO);
952 	cs = &ccd_softc[unit];
953 
954 	if ((cs->sc_flags & CCDF_INITED) == 0)
955 		return (ENXIO);
956 
957 	return (physio(ccdstrategy, NULL, dev, B_READ, minphys, uio));
958 }
959 
960 /* ARGSUSED */
961 static int
962 ccdwrite(dev_t dev, struct uio *uio, int flags)
963 {
964 	int unit = ccdunit(dev);
965 	struct ccd_softc *cs;
966 
967 #ifdef DEBUG
968 	if (ccddebug & CCDB_FOLLOW)
969 		printf("ccdwrite(0x%x, %p)\n", dev, uio);
970 #endif
971 	if (unit >= numccd)
972 		return (ENXIO);
973 	cs = &ccd_softc[unit];
974 
975 	if ((cs->sc_flags & CCDF_INITED) == 0)
976 		return (ENXIO);
977 
978 	return (physio(ccdstrategy, NULL, dev, B_WRITE, minphys, uio));
979 }
980 
981 static int
982 ccdioctl(dev_t dev, u_long cmd, caddr_t data, int flag, struct proc *p)
983 {
984 	int unit = ccdunit(dev);
985 	int s, i, j, lookedup = 0, error;
986 	int part, pmask;
987 	struct buf *bp;
988 	struct ccd_softc *cs;
989 	struct ccd_ioctl *ccio = (struct ccd_ioctl *)data;
990 	struct ucred *uc;
991 	char **cpp;
992 	struct vnode **vpp;
993 #ifdef __HAVE_OLD_DISKLABEL
994 	struct disklabel newlabel;
995 #endif
996 
997 	if (unit >= numccd)
998 		return (ENXIO);
999 	cs = &ccd_softc[unit];
1000 
1001 	/* Must be open for writes for these commands... */
1002 	switch (cmd) {
1003 	case CCDIOCSET:
1004 	case CCDIOCCLR:
1005 	case DIOCSDINFO:
1006 	case DIOCWDINFO:
1007 #ifdef __HAVE_OLD_DISKLABEL
1008 	case ODIOCSDINFO:
1009 	case ODIOCWDINFO:
1010 #endif
1011 	case DIOCKLABEL:
1012 	case DIOCWLABEL:
1013 		if ((flag & FWRITE) == 0)
1014 			return (EBADF);
1015 	}
1016 
1017 	if ((error = lockmgr(&cs->sc_lock, LK_EXCLUSIVE, NULL)) != 0)
1018 		return (error);
1019 
1020 	/* Must be initialized for these... */
1021 	switch (cmd) {
1022 	case CCDIOCCLR:
1023 	case DIOCGDINFO:
1024 	case DIOCCACHESYNC:
1025 	case DIOCSDINFO:
1026 	case DIOCWDINFO:
1027 	case DIOCGPART:
1028 	case DIOCWLABEL:
1029 	case DIOCKLABEL:
1030 	case DIOCGDEFLABEL:
1031 #ifdef __HAVE_OLD_DISKLABEL
1032 	case ODIOCGDINFO:
1033 	case ODIOCSDINFO:
1034 	case ODIOCWDINFO:
1035 	case ODIOCGDEFLABEL:
1036 #endif
1037 		if ((cs->sc_flags & CCDF_INITED) == 0) {
1038 			error = ENXIO;
1039 			goto out;
1040 		}
1041 	}
1042 
1043 	switch (cmd) {
1044 	case CCDIOCSET:
1045 		if (cs->sc_flags & CCDF_INITED) {
1046 			error = EBUSY;
1047 			goto out;
1048 		}
1049 
1050 		/* Validate the flags. */
1051 		if ((ccio->ccio_flags & CCDF_USERMASK) != ccio->ccio_flags) {
1052 			error = EINVAL;
1053 			goto out;
1054 		}
1055 
1056 		if (ccio->ccio_ndisks > CCD_MAXNDISKS) {
1057 			error = EINVAL;
1058 			goto out;
1059 		}
1060 
1061 		/* Fill in some important bits. */
1062 		cs->sc_ileave = ccio->ccio_ileave;
1063 		cs->sc_nccdisks = ccio->ccio_ndisks;
1064 		cs->sc_flags = ccio->ccio_flags & CCDF_USERMASK;
1065 
1066 		/*
1067 		 * Allocate space for and copy in the array of
1068 		 * componet pathnames and device numbers.
1069 		 */
1070 		cpp = malloc(ccio->ccio_ndisks * sizeof(char *),
1071 		    M_DEVBUF, M_WAITOK);
1072 		vpp = malloc(ccio->ccio_ndisks * sizeof(struct vnode *),
1073 		    M_DEVBUF, M_WAITOK);
1074 
1075 		error = copyin(ccio->ccio_disks, cpp,
1076 		    ccio->ccio_ndisks * sizeof(char **));
1077 		if (error) {
1078 			free(vpp, M_DEVBUF);
1079 			free(cpp, M_DEVBUF);
1080 			goto out;
1081 		}
1082 
1083 #ifdef DEBUG
1084 		if (ccddebug & CCDB_INIT)
1085 			for (i = 0; i < ccio->ccio_ndisks; ++i)
1086 				printf("ccdioctl: component %d: 0x%p\n",
1087 				    i, cpp[i]);
1088 #endif
1089 
1090 		for (i = 0; i < ccio->ccio_ndisks; ++i) {
1091 #ifdef DEBUG
1092 			if (ccddebug & CCDB_INIT)
1093 				printf("ccdioctl: lookedup = %d\n", lookedup);
1094 #endif
1095 			if ((error = ccdlookup(cpp[i], p, &vpp[i])) != 0) {
1096 				for (j = 0; j < lookedup; ++j)
1097 					(void)vn_close(vpp[j], FREAD|FWRITE,
1098 					    p->p_ucred, p);
1099 				free(vpp, M_DEVBUF);
1100 				free(cpp, M_DEVBUF);
1101 				goto out;
1102 			}
1103 			++lookedup;
1104 		}
1105 
1106 		/*
1107 		 * Initialize the ccd.  Fills in the softc for us.
1108 		 */
1109 		if ((error = ccdinit(cs, cpp, vpp, p)) != 0) {
1110 			for (j = 0; j < lookedup; ++j)
1111 				(void)vn_close(vpp[j], FREAD|FWRITE,
1112 				    p->p_ucred, p);
1113 			free(vpp, M_DEVBUF);
1114 			free(cpp, M_DEVBUF);
1115 			goto out;
1116 		}
1117 
1118 		/* We can free the temporary variables now. */
1119 		free(vpp, M_DEVBUF);
1120 		free(cpp, M_DEVBUF);
1121 
1122 		/*
1123 		 * The ccd has been successfully initialized, so
1124 		 * we can place it into the array.  Don't try to
1125 		 * read the disklabel until the disk has been attached,
1126 		 * because space for the disklabel is allocated
1127 		 * in disk_attach();
1128 		 */
1129 		ccio->ccio_unit = unit;
1130 		ccio->ccio_size = cs->sc_size;
1131 
1132 		bufq_alloc(&cs->sc_bufq, BUFQ_FCFS);
1133 
1134 		/* Attach the disk. */
1135 		disk_attach(&cs->sc_dkdev);
1136 
1137 		/* Try and read the disklabel. */
1138 		ccdgetdisklabel(dev);
1139 		break;
1140 
1141 	case CCDIOCCLR:
1142 		/*
1143 		 * Don't unconfigure if any other partitions are open
1144 		 * or if both the character and block flavors of this
1145 		 * partition are open.
1146 		 */
1147 		part = DISKPART(dev);
1148 		pmask = (1 << part);
1149 		if ((cs->sc_dkdev.dk_openmask & ~pmask) ||
1150 		    ((cs->sc_dkdev.dk_bopenmask & pmask) &&
1151 		    (cs->sc_dkdev.dk_copenmask & pmask))) {
1152 			error = EBUSY;
1153 			goto out;
1154 		}
1155 
1156 		/* Kill off any queued buffers. */
1157 		s = splbio();
1158 		while ((bp = BUFQ_GET(&cs->sc_bufq)) != NULL) {
1159 			bp->b_error = EIO;
1160 			bp->b_flags |= B_ERROR;
1161 			bp->b_resid = bp->b_bcount;
1162 			biodone(bp);
1163 		}
1164 		splx(s);
1165 
1166 		bufq_free(&cs->sc_bufq);
1167 
1168 		/*
1169 		 * Free ccd_softc information and clear entry.
1170 		 */
1171 
1172 		/* Close the components and free their pathnames. */
1173 		for (i = 0; i < cs->sc_nccdisks; ++i) {
1174 			/*
1175 			 * XXX: this close could potentially fail and
1176 			 * cause Bad Things.  Maybe we need to force
1177 			 * the close to happen?
1178 			 */
1179 #ifdef DEBUG
1180 			if (ccddebug & CCDB_VNODE)
1181 				vprint("CCDIOCCLR: vnode info",
1182 				    cs->sc_cinfo[i].ci_vp);
1183 #endif
1184 			(void)vn_close(cs->sc_cinfo[i].ci_vp, FREAD|FWRITE,
1185 			    p->p_ucred, p);
1186 			free(cs->sc_cinfo[i].ci_path, M_DEVBUF);
1187 		}
1188 
1189 		/* Free interleave index. */
1190 		for (i = 0; cs->sc_itable[i].ii_ndisk; ++i)
1191 			free(cs->sc_itable[i].ii_index, M_DEVBUF);
1192 
1193 		/* Free component info and interleave table. */
1194 		free(cs->sc_cinfo, M_DEVBUF);
1195 		free(cs->sc_itable, M_DEVBUF);
1196 		cs->sc_flags &= ~(CCDF_INITED|CCDF_VLABEL);
1197 
1198 		/* Detatch the disk. */
1199 		disk_detach(&cs->sc_dkdev);
1200 		break;
1201 
1202 	case DIOCGDINFO:
1203 		*(struct disklabel *)data = *(cs->sc_dkdev.dk_label);
1204 		break;
1205 #ifdef __HAVE_OLD_DISKLABEL
1206 	case ODIOCGDINFO:
1207 		newlabel = *(cs->sc_dkdev.dk_label);
1208 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1209 			return ENOTTY;
1210 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1211 		break;
1212 #endif
1213 
1214 	case DIOCGPART:
1215 		((struct partinfo *)data)->disklab = cs->sc_dkdev.dk_label;
1216 		((struct partinfo *)data)->part =
1217 		    &cs->sc_dkdev.dk_label->d_partitions[DISKPART(dev)];
1218 		break;
1219 
1220 	case DIOCCACHESYNC:
1221 		/*
1222 		 * XXX Do we really need to care about having a writable
1223 		 * file descriptor here?
1224 		 */
1225 		if ((flag & FWRITE) == 0)
1226 			return (EBADF);
1227 
1228 		/*
1229 		 * We pass this call down to all components and report
1230 		 * the first error we encounter.
1231 		 */
1232 		uc = (p != NULL) ? p->p_ucred : NOCRED;
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, p);
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 	(void) lockmgr(&cs->sc_lock, LK_RELEASE, NULL);
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, curproc))
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, curproc))
1342 		return (-1);
1343 
1344 	return (size);
1345 }
1346 
1347 static int
1348 ccddump(dev_t dev, daddr_t blkno, caddr_t va, size_t size)
1349 {
1350 
1351 	/* Not implemented. */
1352 	return ENXIO;
1353 }
1354 
1355 /*
1356  * Lookup the provided name in the filesystem.  If the file exists,
1357  * is a valid block device, and isn't being used by anyone else,
1358  * set *vpp to the file's vnode.
1359  */
1360 static int
1361 ccdlookup(char *path, struct proc *p, struct vnode **vpp /* result */)
1362 {
1363 	struct nameidata nd;
1364 	struct vnode *vp;
1365 	struct vattr va;
1366 	int error;
1367 
1368 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, path, p);
1369 	if ((error = vn_open(&nd, FREAD|FWRITE, 0)) != 0) {
1370 #ifdef DEBUG
1371 		if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
1372 			printf("ccdlookup: vn_open error = %d\n", error);
1373 #endif
1374 		return (error);
1375 	}
1376 	vp = nd.ni_vp;
1377 
1378 	if (vp->v_usecount > 1) {
1379 		VOP_UNLOCK(vp, 0);
1380 		(void)vn_close(vp, FREAD|FWRITE, p->p_ucred, p);
1381 		return (EBUSY);
1382 	}
1383 
1384 	if ((error = VOP_GETATTR(vp, &va, p->p_ucred, p)) != 0) {
1385 #ifdef DEBUG
1386 		if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
1387 			printf("ccdlookup: getattr error = %d\n", error);
1388 #endif
1389 		VOP_UNLOCK(vp, 0);
1390 		(void)vn_close(vp, FREAD|FWRITE, p->p_ucred, p);
1391 		return (error);
1392 	}
1393 
1394 	/* XXX: eventually we should handle VREG, too. */
1395 	if (va.va_type != VBLK) {
1396 		VOP_UNLOCK(vp, 0);
1397 		(void)vn_close(vp, FREAD|FWRITE, p->p_ucred, p);
1398 		return (ENOTBLK);
1399 	}
1400 
1401 #ifdef DEBUG
1402 	if (ccddebug & CCDB_VNODE)
1403 		vprint("ccdlookup: vnode info", vp);
1404 #endif
1405 
1406 	VOP_UNLOCK(vp, 0);
1407 	*vpp = vp;
1408 	return (0);
1409 }
1410 
1411 static void
1412 ccdgetdefaultlabel(struct ccd_softc *cs, struct disklabel *lp)
1413 {
1414 	struct ccdgeom *ccg = &cs->sc_geom;
1415 
1416 	memset(lp, 0, sizeof(*lp));
1417 
1418 	lp->d_secperunit = cs->sc_size;
1419 	lp->d_secsize = ccg->ccg_secsize;
1420 	lp->d_nsectors = ccg->ccg_nsectors;
1421 	lp->d_ntracks = ccg->ccg_ntracks;
1422 	lp->d_ncylinders = ccg->ccg_ncylinders;
1423 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1424 
1425 	strncpy(lp->d_typename, "ccd", sizeof(lp->d_typename));
1426 	lp->d_type = DTYPE_CCD;
1427 	strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
1428 	lp->d_rpm = 3600;
1429 	lp->d_interleave = 1;
1430 	lp->d_flags = 0;
1431 
1432 	lp->d_partitions[RAW_PART].p_offset = 0;
1433 	lp->d_partitions[RAW_PART].p_size = cs->sc_size;
1434 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
1435 	lp->d_npartitions = RAW_PART + 1;
1436 
1437 	lp->d_magic = DISKMAGIC;
1438 	lp->d_magic2 = DISKMAGIC;
1439 	lp->d_checksum = dkcksum(cs->sc_dkdev.dk_label);
1440 }
1441 
1442 /*
1443  * Read the disklabel from the ccd.  If one is not present, fake one
1444  * up.
1445  */
1446 static void
1447 ccdgetdisklabel(dev_t dev)
1448 {
1449 	int unit = ccdunit(dev);
1450 	struct ccd_softc *cs = &ccd_softc[unit];
1451 	const char *errstring;
1452 	struct disklabel *lp = cs->sc_dkdev.dk_label;
1453 	struct cpu_disklabel *clp = cs->sc_dkdev.dk_cpulabel;
1454 
1455 	memset(clp, 0, sizeof(*clp));
1456 
1457 	ccdgetdefaultlabel(cs, lp);
1458 
1459 	/*
1460 	 * Call the generic disklabel extraction routine.
1461 	 */
1462 	if ((cs->sc_flags & CCDF_NOLABEL) != 0)
1463 		errstring = "CCDF_NOLABEL set; ignoring on-disk label";
1464 	else
1465 		errstring = readdisklabel(CCDLABELDEV(dev), ccdstrategy,
1466 		    cs->sc_dkdev.dk_label, cs->sc_dkdev.dk_cpulabel);
1467 	if (errstring)
1468 		ccdmakedisklabel(cs);
1469 	else {
1470 		int i;
1471 		struct partition *pp;
1472 
1473 		/*
1474 		 * Sanity check whether the found disklabel is valid.
1475 		 *
1476 		 * This is necessary since total size of ccd may vary
1477 		 * when an interleave is changed even though exactly
1478 		 * same componets are used, and old disklabel may used
1479 		 * if that is found.
1480 		 */
1481 		if (lp->d_secperunit != cs->sc_size)
1482 			printf("WARNING: %s: "
1483 			    "total sector size in disklabel (%d) != "
1484 			    "the size of ccd (%lu)\n", cs->sc_xname,
1485 			    lp->d_secperunit, (u_long)cs->sc_size);
1486 		for (i = 0; i < lp->d_npartitions; i++) {
1487 			pp = &lp->d_partitions[i];
1488 			if (pp->p_offset + pp->p_size > cs->sc_size)
1489 				printf("WARNING: %s: end of partition `%c' "
1490 				    "exceeds the size of ccd (%lu)\n",
1491 				    cs->sc_xname, 'a' + i, (u_long)cs->sc_size);
1492 		}
1493 	}
1494 
1495 #ifdef DEBUG
1496 	/* It's actually extremely common to have unlabeled ccds. */
1497 	if (ccddebug & CCDB_LABEL)
1498 		if (errstring != NULL)
1499 			printf("%s: %s\n", cs->sc_xname, errstring);
1500 #endif
1501 
1502 	/* In-core label now valid. */
1503 	cs->sc_flags |= CCDF_VLABEL;
1504 }
1505 
1506 /*
1507  * Take care of things one might want to take care of in the event
1508  * that a disklabel isn't present.
1509  */
1510 static void
1511 ccdmakedisklabel(struct ccd_softc *cs)
1512 {
1513 	struct disklabel *lp = cs->sc_dkdev.dk_label;
1514 
1515 	/*
1516 	 * For historical reasons, if there's no disklabel present
1517 	 * the raw partition must be marked FS_BSDFFS.
1518 	 */
1519 	lp->d_partitions[RAW_PART].p_fstype = FS_BSDFFS;
1520 
1521 	strncpy(lp->d_packname, "default label", sizeof(lp->d_packname));
1522 
1523 	lp->d_checksum = dkcksum(lp);
1524 }
1525 
1526 #ifdef DEBUG
1527 static void
1528 printiinfo(struct ccdiinfo *ii)
1529 {
1530 	int ix, i;
1531 
1532 	for (ix = 0; ii->ii_ndisk; ix++, ii++) {
1533 		printf(" itab[%d]: #dk %d sblk %" PRId64 " soff %" PRId64,
1534 		    ix, ii->ii_ndisk, ii->ii_startblk, ii->ii_startoff);
1535 		for (i = 0; i < ii->ii_ndisk; i++)
1536 			printf(" %d", ii->ii_index[i]);
1537 		printf("\n");
1538 	}
1539 }
1540 #endif
1541