xref: /netbsd-src/sys/dev/ccd.c (revision da5f4674a3fc214be3572d358b66af40ab9401e7)
1 /*	$NetBSD: ccd.c,v 1.91 2003/08/07 16:30:50 agc 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 1.91 2003/08/07 16:30:50 agc 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/malloc.h>
136 #include <sys/pool.h>
137 #include <sys/namei.h>
138 #include <sys/stat.h>
139 #include <sys/ioctl.h>
140 #include <sys/disklabel.h>
141 #include <sys/device.h>
142 #include <sys/disk.h>
143 #include <sys/syslog.h>
144 #include <sys/fcntl.h>
145 #include <sys/vnode.h>
146 #include <sys/conf.h>
147 #include <sys/lock.h>
148 #include <sys/queue.h>
149 
150 #include <dev/ccdvar.h>
151 
152 #if defined(CCDDEBUG) && !defined(DEBUG)
153 #define DEBUG
154 #endif
155 
156 #ifdef DEBUG
157 #define CCDB_FOLLOW	0x01
158 #define CCDB_INIT	0x02
159 #define CCDB_IO		0x04
160 #define CCDB_LABEL	0x08
161 #define CCDB_VNODE	0x10
162 int ccddebug = 0x00;
163 #endif
164 
165 #define	ccdunit(x)	DISKUNIT(x)
166 
167 struct ccdbuf {
168 	struct buf	cb_buf;		/* new I/O buf */
169 	struct buf	*cb_obp;	/* ptr. to original I/O buf */
170 	struct ccd_softc *cb_sc;	/* pointer to ccd softc */
171 	int		cb_comp;	/* target component */
172 	SIMPLEQ_ENTRY(ccdbuf) cb_q;	/* fifo of component buffers */
173 };
174 
175 /* component buffer pool */
176 struct pool ccd_cbufpool;
177 
178 #define	CCD_GETBUF()		pool_get(&ccd_cbufpool, PR_NOWAIT)
179 #define	CCD_PUTBUF(cbp)		pool_put(&ccd_cbufpool, cbp)
180 
181 #define CCDLABELDEV(dev)	\
182 	(MAKEDISKDEV(major((dev)), ccdunit((dev)), RAW_PART))
183 
184 /* called by main() at boot time */
185 void	ccdattach __P((int));
186 
187 /* called by biodone() at interrupt time */
188 void	ccdiodone __P((struct buf *));
189 
190 static	void ccdstart __P((struct ccd_softc *));
191 static	void ccdinterleave __P((struct ccd_softc *));
192 static	void ccdintr __P((struct ccd_softc *, struct buf *));
193 static	int ccdinit __P((struct ccd_softc *, char **, struct vnode **,
194 	    struct proc *));
195 static	int ccdlookup __P((char *, struct proc *p, struct vnode **));
196 static	struct ccdbuf *ccdbuffer __P((struct ccd_softc *, struct buf *,
197 		daddr_t, caddr_t, long));
198 static	void ccdgetdefaultlabel __P((struct ccd_softc *, struct disklabel *));
199 static	void ccdgetdisklabel __P((dev_t));
200 static	void ccdmakedisklabel __P((struct ccd_softc *));
201 
202 dev_type_open(ccdopen);
203 dev_type_close(ccdclose);
204 dev_type_read(ccdread);
205 dev_type_write(ccdwrite);
206 dev_type_ioctl(ccdioctl);
207 dev_type_strategy(ccdstrategy);
208 dev_type_dump(ccddump);
209 dev_type_size(ccdsize);
210 
211 const struct bdevsw ccd_bdevsw = {
212 	ccdopen, ccdclose, ccdstrategy, ccdioctl, ccddump, ccdsize, D_DISK
213 };
214 
215 const struct cdevsw ccd_cdevsw = {
216 	ccdopen, ccdclose, ccdread, ccdwrite, ccdioctl,
217 	nostop, notty, nopoll, nommap, nokqfilter, D_DISK
218 };
219 
220 #ifdef DEBUG
221 static	void printiinfo __P((struct ccdiinfo *));
222 #endif
223 
224 /* Non-private for the benefit of libkvm. */
225 struct	ccd_softc *ccd_softc;
226 int	numccd = 0;
227 
228 /*
229  * Called by main() during pseudo-device attachment.  All we need
230  * to do is allocate enough space for devices to be configured later.
231  */
232 void
233 ccdattach(num)
234 	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 		sprintf(cs->sc_xname, "ccd%d", i);	/* XXX */
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(cs, cpaths, vpp, p)
269 	struct ccd_softc *cs;
270 	char **cpaths;
271 	struct vnode **vpp;
272 	struct proc *p;
273 {
274 	struct ccdcinfo *ci = NULL;
275 	size_t size;
276 	int ix;
277 	struct vattr va;
278 	size_t minsize;
279 	int maxsecsize;
280 	struct partinfo dpart;
281 	struct ccdgeom *ccg = &cs->sc_geom;
282 	char tmppath[MAXPATHLEN];
283 	int error, path_alloced;
284 
285 #ifdef DEBUG
286 	if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
287 		printf("%s: ccdinit\n", cs->sc_xname);
288 #endif
289 
290 	/* Allocate space for the component info. */
291 	cs->sc_cinfo = malloc(cs->sc_nccdisks * sizeof(struct ccdcinfo),
292 	    M_DEVBUF, M_WAITOK);
293 
294 	cs->sc_size = 0;
295 
296 	/*
297 	 * Verify that each component piece exists and record
298 	 * relevant information about it.
299 	 */
300 	maxsecsize = 0;
301 	minsize = 0;
302 	for (ix = 0, path_alloced = 0; ix < cs->sc_nccdisks; ix++) {
303 		ci = &cs->sc_cinfo[ix];
304 		ci->ci_vp = vpp[ix];
305 
306 		/*
307 		 * Copy in the pathname of the component.
308 		 */
309 		memset(tmppath, 0, sizeof(tmppath));	/* sanity */
310 		error = copyinstr(cpaths[ix], tmppath,
311 		    MAXPATHLEN, &ci->ci_pathlen);
312 		if (error) {
313 #ifdef DEBUG
314 			if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
315 				printf("%s: can't copy path, error = %d\n",
316 				    cs->sc_xname, error);
317 #endif
318 			goto out;
319 		}
320 		ci->ci_path = malloc(ci->ci_pathlen, M_DEVBUF, M_WAITOK);
321 		memcpy(ci->ci_path, tmppath, ci->ci_pathlen);
322 		path_alloced++;
323 
324 		/*
325 		 * XXX: Cache the component's dev_t.
326 		 */
327 		if ((error = VOP_GETATTR(vpp[ix], &va, p->p_ucred, p)) != 0) {
328 #ifdef DEBUG
329 			if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
330 				printf("%s: %s: getattr failed %s = %d\n",
331 				    cs->sc_xname, ci->ci_path,
332 				    "error", error);
333 #endif
334 			goto out;
335 		}
336 		ci->ci_dev = va.va_rdev;
337 
338 		/*
339 		 * Get partition information for the component.
340 		 */
341 		error = VOP_IOCTL(vpp[ix], DIOCGPART, &dpart,
342 		    FREAD, p->p_ucred, p);
343 		if (error) {
344 #ifdef DEBUG
345 			if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
346 				 printf("%s: %s: ioctl failed, error = %d\n",
347 				     cs->sc_xname, ci->ci_path, error);
348 #endif
349 			goto out;
350 		}
351 
352 /*
353  * This diagnostic test is disabled (for now?) since not all port supports
354  * on-disk BSD disklabel.
355  */
356 #if 0 /* def DIAGNOSTIC */
357 		/* Check fstype field of component. */
358 		if (dpart.part->p_fstype != FS_CCD)
359 			printf("%s: WARNING: %s: fstype %d != FS_CCD\n",
360 			    cs->sc_xname, ci->ci_path, dpart.part->p_fstype);
361 #endif
362 
363 		/*
364 		 * Calculate the size, truncating to an interleave
365 		 * boundary if necessary.
366 		 */
367 		maxsecsize =
368 		    ((dpart.disklab->d_secsize > maxsecsize) ?
369 		    dpart.disklab->d_secsize : maxsecsize);
370 		size = dpart.part->p_size;
371 		if (cs->sc_ileave > 1)
372 			size -= size % cs->sc_ileave;
373 
374 		if (size == 0) {
375 #ifdef DEBUG
376 			if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
377 				printf("%s: %s: size == 0\n",
378 				    cs->sc_xname, ci->ci_path);
379 #endif
380 			error = ENODEV;
381 			goto out;
382 		}
383 
384 		if (minsize == 0 || size < minsize)
385 			minsize = size;
386 		ci->ci_size = size;
387 		cs->sc_size += size;
388 	}
389 
390 	/*
391 	 * Don't allow the interleave to be smaller than
392 	 * the biggest component sector.
393 	 */
394 	if ((cs->sc_ileave > 0) &&
395 	    (cs->sc_ileave < (maxsecsize / DEV_BSIZE))) {
396 #ifdef DEBUG
397 		if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
398 			printf("%s: interleave must be at least %d\n",
399 			    cs->sc_xname, (maxsecsize / DEV_BSIZE));
400 #endif
401 		error = EINVAL;
402 		goto out;
403 	}
404 
405 	/*
406 	 * If uniform interleave is desired set all sizes to that of
407 	 * the smallest component.
408 	 */
409 	if (cs->sc_flags & CCDF_UNIFORM) {
410 		for (ci = cs->sc_cinfo;
411 		     ci < &cs->sc_cinfo[cs->sc_nccdisks]; ci++)
412 			ci->ci_size = minsize;
413 
414 		cs->sc_size = cs->sc_nccdisks * minsize;
415 	}
416 
417 	/*
418 	 * Construct the interleave table.
419 	 */
420 	ccdinterleave(cs);
421 
422 	/*
423 	 * Create pseudo-geometry based on 1MB cylinders.  It's
424 	 * pretty close.
425 	 */
426 	ccg->ccg_secsize = DEV_BSIZE;
427 	ccg->ccg_ntracks = 1;
428 	ccg->ccg_nsectors = 1024 * (1024 / ccg->ccg_secsize);
429 	ccg->ccg_ncylinders = cs->sc_size / ccg->ccg_nsectors;
430 
431 	cs->sc_flags |= CCDF_INITED;
432 
433 	return (0);
434 
435  out:
436 	for (ix = 0; ix < path_alloced; ix++)
437 		free(cs->sc_cinfo[ix].ci_path, M_DEVBUF);
438 	free(cs->sc_cinfo, M_DEVBUF);
439 	return (error);
440 }
441 
442 static void
443 ccdinterleave(cs)
444 	struct ccd_softc *cs;
445 {
446 	struct ccdcinfo *ci, *smallci;
447 	struct ccdiinfo *ii;
448 	daddr_t bn, lbn;
449 	int ix;
450 	u_long size;
451 
452 #ifdef DEBUG
453 	if (ccddebug & CCDB_INIT)
454 		printf("ccdinterleave(%p): ileave %d\n", cs, cs->sc_ileave);
455 #endif
456 	/*
457 	 * Allocate an interleave table.
458 	 * Chances are this is too big, but we don't care.
459 	 */
460 	size = (cs->sc_nccdisks + 1) * sizeof(struct ccdiinfo);
461 	cs->sc_itable = (struct ccdiinfo *)malloc(size, M_DEVBUF,
462 	    M_WAITOK|M_ZERO);
463 
464 	/*
465 	 * Trivial case: no interleave (actually interleave of disk size).
466 	 * Each table entry represents a single component in its entirety.
467 	 */
468 	if (cs->sc_ileave == 0) {
469 		bn = 0;
470 		ii = cs->sc_itable;
471 
472 		for (ix = 0; ix < cs->sc_nccdisks; ix++) {
473 			/* Allocate space for ii_index. */
474 			ii->ii_index = malloc(sizeof(int), M_DEVBUF, M_WAITOK);
475 			ii->ii_ndisk = 1;
476 			ii->ii_startblk = bn;
477 			ii->ii_startoff = 0;
478 			ii->ii_index[0] = ix;
479 			bn += cs->sc_cinfo[ix].ci_size;
480 			ii++;
481 		}
482 		ii->ii_ndisk = 0;
483 #ifdef DEBUG
484 		if (ccddebug & CCDB_INIT)
485 			printiinfo(cs->sc_itable);
486 #endif
487 		return;
488 	}
489 
490 	/*
491 	 * The following isn't fast or pretty; it doesn't have to be.
492 	 */
493 	size = 0;
494 	bn = lbn = 0;
495 	for (ii = cs->sc_itable; ; ii++) {
496 		/* Allocate space for ii_index. */
497 		ii->ii_index = malloc((sizeof(int) * cs->sc_nccdisks),
498 		    M_DEVBUF, M_WAITOK);
499 
500 		/*
501 		 * Locate the smallest of the remaining components
502 		 */
503 		smallci = NULL;
504 		for (ci = cs->sc_cinfo;
505 		     ci < &cs->sc_cinfo[cs->sc_nccdisks]; ci++)
506 			if (ci->ci_size > size &&
507 			    (smallci == NULL ||
508 			     ci->ci_size < smallci->ci_size))
509 				smallci = ci;
510 
511 		/*
512 		 * Nobody left, all done
513 		 */
514 		if (smallci == NULL) {
515 			ii->ii_ndisk = 0;
516 			break;
517 		}
518 
519 		/*
520 		 * Record starting logical block and component offset
521 		 */
522 		ii->ii_startblk = bn / cs->sc_ileave;
523 		ii->ii_startoff = lbn;
524 
525 		/*
526 		 * Determine how many disks take part in this interleave
527 		 * and record their indices.
528 		 */
529 		ix = 0;
530 		for (ci = cs->sc_cinfo;
531 		     ci < &cs->sc_cinfo[cs->sc_nccdisks]; ci++)
532 			if (ci->ci_size >= smallci->ci_size)
533 				ii->ii_index[ix++] = ci - cs->sc_cinfo;
534 		ii->ii_ndisk = ix;
535 		bn += ix * (smallci->ci_size - size);
536 		lbn = smallci->ci_size / cs->sc_ileave;
537 		size = smallci->ci_size;
538 	}
539 #ifdef DEBUG
540 	if (ccddebug & CCDB_INIT)
541 		printiinfo(cs->sc_itable);
542 #endif
543 }
544 
545 /* ARGSUSED */
546 int
547 ccdopen(dev, flags, fmt, p)
548 	dev_t dev;
549 	int flags, fmt;
550 	struct proc *p;
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 	if ((error = lockmgr(&cs->sc_lock, LK_EXCLUSIVE, NULL)) != 0)
566 		return (error);
567 
568 	lp = cs->sc_dkdev.dk_label;
569 
570 	part = DISKPART(dev);
571 	pmask = (1 << part);
572 
573 	/*
574 	 * If we're initialized, check to see if there are any other
575 	 * open partitions.  If not, then it's safe to update
576 	 * the in-core disklabel.  Only read the disklabel if it is
577 	 * not already valid.
578 	 */
579 	if ((cs->sc_flags & (CCDF_INITED|CCDF_VLABEL)) == CCDF_INITED &&
580 	    cs->sc_dkdev.dk_openmask == 0)
581 		ccdgetdisklabel(dev);
582 
583 	/* Check that the partition exists. */
584 	if (part != RAW_PART) {
585 		if (((cs->sc_flags & CCDF_INITED) == 0) ||
586 		    ((part >= lp->d_npartitions) ||
587 		     (lp->d_partitions[part].p_fstype == FS_UNUSED))) {
588 			error = ENXIO;
589 			goto done;
590 		}
591 	}
592 
593 	/* Prevent our unit from being unconfigured while open. */
594 	switch (fmt) {
595 	case S_IFCHR:
596 		cs->sc_dkdev.dk_copenmask |= pmask;
597 		break;
598 
599 	case S_IFBLK:
600 		cs->sc_dkdev.dk_bopenmask |= pmask;
601 		break;
602 	}
603 	cs->sc_dkdev.dk_openmask =
604 	    cs->sc_dkdev.dk_copenmask | cs->sc_dkdev.dk_bopenmask;
605 
606  done:
607 	(void) lockmgr(&cs->sc_lock, LK_RELEASE, NULL);
608 	return (error);
609 }
610 
611 /* ARGSUSED */
612 int
613 ccdclose(dev, flags, fmt, p)
614 	dev_t dev;
615 	int flags, fmt;
616 	struct proc *p;
617 {
618 	int unit = ccdunit(dev);
619 	struct ccd_softc *cs;
620 	int error = 0, part;
621 
622 #ifdef DEBUG
623 	if (ccddebug & CCDB_FOLLOW)
624 		printf("ccdclose(0x%x, 0x%x)\n", dev, flags);
625 #endif
626 
627 	if (unit >= numccd)
628 		return (ENXIO);
629 	cs = &ccd_softc[unit];
630 
631 	if ((error = lockmgr(&cs->sc_lock, LK_EXCLUSIVE, NULL)) != 0)
632 		return (error);
633 
634 	part = DISKPART(dev);
635 
636 	/* ...that much closer to allowing unconfiguration... */
637 	switch (fmt) {
638 	case S_IFCHR:
639 		cs->sc_dkdev.dk_copenmask &= ~(1 << part);
640 		break;
641 
642 	case S_IFBLK:
643 		cs->sc_dkdev.dk_bopenmask &= ~(1 << part);
644 		break;
645 	}
646 	cs->sc_dkdev.dk_openmask =
647 	    cs->sc_dkdev.dk_copenmask | cs->sc_dkdev.dk_bopenmask;
648 
649 	if (cs->sc_dkdev.dk_openmask == 0) {
650 		if ((cs->sc_flags & CCDF_KLABEL) == 0)
651 			cs->sc_flags &= ~CCDF_VLABEL;
652 	}
653 
654 	(void) lockmgr(&cs->sc_lock, LK_RELEASE, NULL);
655 	return (0);
656 }
657 
658 void
659 ccdstrategy(bp)
660 	struct buf *bp;
661 {
662 	int unit = ccdunit(bp->b_dev);
663 	struct ccd_softc *cs = &ccd_softc[unit];
664 	daddr_t blkno;
665 	int s;
666 	int wlabel;
667 	struct disklabel *lp;
668 
669 #ifdef DEBUG
670 	if (ccddebug & CCDB_FOLLOW)
671 		printf("ccdstrategy(%p): unit %d\n", bp, unit);
672 #endif
673 	if ((cs->sc_flags & CCDF_INITED) == 0) {
674 #ifdef DEBUG
675 		if (ccddebug & CCDB_FOLLOW)
676 			printf("ccdstrategy: unit %d: not inited\n", unit);
677 #endif
678 		bp->b_error = ENXIO;
679 		bp->b_flags |= B_ERROR;
680 		goto done;
681 	}
682 
683 	/* If it's a nil transfer, wake up the top half now. */
684 	if (bp->b_bcount == 0)
685 		goto done;
686 
687 	lp = cs->sc_dkdev.dk_label;
688 
689 	/*
690 	 * Do bounds checking and adjust transfer.  If there's an
691 	 * error, the bounds check will flag that for us.  Convert
692 	 * the partition relative block number to an absolute.
693 	 */
694 	blkno = bp->b_blkno;
695 	wlabel = cs->sc_flags & (CCDF_WLABEL|CCDF_LABELLING);
696 	if (DISKPART(bp->b_dev) != RAW_PART) {
697 		if (bounds_check_with_label(&cs->sc_dkdev, bp, wlabel) <= 0)
698 			goto done;
699 		blkno += lp->d_partitions[DISKPART(bp->b_dev)].p_offset;
700 	}
701 	bp->b_rawblkno = blkno;
702 
703 	/* Place it in the queue and start I/O on the unit. */
704 	s = splbio();
705 	BUFQ_PUT(&cs->sc_bufq, bp);
706 	ccdstart(cs);
707 	splx(s);
708 	return;
709 
710  done:
711 	bp->b_resid = bp->b_bcount;
712 	biodone(bp);
713 }
714 
715 static void
716 ccdstart(cs)
717 	struct ccd_softc *cs;
718 {
719 	long bcount, rcount;
720 	struct buf *bp;
721 	struct ccdbuf *cbp;
722 	caddr_t addr;
723 	daddr_t bn;
724 	SIMPLEQ_HEAD(, ccdbuf) cbufq;
725 
726 #ifdef DEBUG
727 	if (ccddebug & CCDB_FOLLOW)
728 		printf("ccdstart(%p, %p)\n", cs, bp);
729 #endif
730 
731 	/* See if there is work for us to do. */
732 	while ((bp = BUFQ_PEEK(&cs->sc_bufq)) != NULL) {
733 		/* Instrumentation. */
734 		disk_busy(&cs->sc_dkdev);
735 
736 		bp->b_resid = bp->b_bcount;
737 		bn = bp->b_rawblkno;
738 
739 		/* Allocate the component buffers. */
740 		SIMPLEQ_INIT(&cbufq);
741 		addr = bp->b_data;
742 		for (bcount = bp->b_bcount; bcount > 0; bcount -= rcount) {
743 			cbp = ccdbuffer(cs, bp, bn, addr, bcount);
744 			if (cbp == NULL) {
745 				/*
746 				 * Can't allocate a component buffer; just
747 				 * defer the job until later.
748 				 *
749 				 * XXX We might consider a watchdog timer
750 				 * XXX to make sure we are kicked into action,
751 				 * XXX or consider a low-water mark for our
752 				 * XXX component buffer pool.
753 				 */
754 				while ((cbp = SIMPLEQ_FIRST(&cbufq)) != NULL) {
755 					SIMPLEQ_REMOVE_HEAD(&cbufq, cb_q);
756 					CCD_PUTBUF(cbp);
757 				}
758 				disk_unbusy(&cs->sc_dkdev, 0, 0);
759 				return;
760 			}
761 			SIMPLEQ_INSERT_TAIL(&cbufq, cbp, cb_q);
762 			rcount = cbp->cb_buf.b_bcount;
763 			bn += btodb(rcount);
764 			addr += rcount;
765 		}
766 
767 		/* Transfer all set up, remove job from the queue. */
768 		(void) BUFQ_GET(&cs->sc_bufq);
769 
770 		/* Now fire off the requests. */
771 		while ((cbp = SIMPLEQ_FIRST(&cbufq)) != NULL) {
772 			SIMPLEQ_REMOVE_HEAD(&cbufq, cb_q);
773 			if ((cbp->cb_buf.b_flags & B_READ) == 0)
774 				cbp->cb_buf.b_vp->v_numoutput++;
775 			VOP_STRATEGY(&cbp->cb_buf);
776 		}
777 	}
778 }
779 
780 /*
781  * Build a component buffer header.
782  */
783 static struct ccdbuf *
784 ccdbuffer(cs, bp, bn, addr, bcount)
785 	struct ccd_softc *cs;
786 	struct buf *bp;
787 	daddr_t bn;
788 	caddr_t addr;
789 	long bcount;
790 {
791 	struct ccdcinfo *ci;
792 	struct ccdbuf *cbp;
793 	daddr_t cbn, cboff;
794 	u_int64_t cbc;
795 	int ccdisk;
796 
797 #ifdef DEBUG
798 	if (ccddebug & CCDB_IO)
799 		printf("ccdbuffer(%p, %p, %" PRId64 ", %p, %ld)\n",
800 		       cs, bp, bn, addr, bcount);
801 #endif
802 	/*
803 	 * Determine which component bn falls in.
804 	 */
805 	cbn = bn;
806 	cboff = 0;
807 
808 	/*
809 	 * Serially concatenated
810 	 */
811 	if (cs->sc_ileave == 0) {
812 		daddr_t sblk;
813 
814 		sblk = 0;
815 		for (ccdisk = 0, ci = &cs->sc_cinfo[ccdisk];
816 		    cbn >= sblk + ci->ci_size;
817 		    ccdisk++, ci = &cs->sc_cinfo[ccdisk])
818 			sblk += ci->ci_size;
819 		cbn -= sblk;
820 	}
821 	/*
822 	 * Interleaved
823 	 */
824 	else {
825 		struct ccdiinfo *ii;
826 		int off;
827 
828 		cboff = cbn % cs->sc_ileave;
829 		cbn /= cs->sc_ileave;
830 		for (ii = cs->sc_itable; ii->ii_ndisk; ii++)
831 			if (ii->ii_startblk > cbn)
832 				break;
833 		ii--;
834 		off = cbn - ii->ii_startblk;
835 		if (ii->ii_ndisk == 1) {
836 			ccdisk = ii->ii_index[0];
837 			cbn = ii->ii_startoff + off;
838 		} else {
839 			ccdisk = ii->ii_index[off % ii->ii_ndisk];
840 			cbn = ii->ii_startoff + off / ii->ii_ndisk;
841 		}
842 		cbn *= cs->sc_ileave;
843 		ci = &cs->sc_cinfo[ccdisk];
844 	}
845 
846 	/*
847 	 * Fill in the component buf structure.
848 	 */
849 	cbp = CCD_GETBUF();
850 	if (cbp == NULL)
851 		return (NULL);
852 	BUF_INIT(&cbp->cb_buf);
853 	cbp->cb_buf.b_flags = bp->b_flags | B_CALL;
854 	cbp->cb_buf.b_iodone = ccdiodone;
855 	cbp->cb_buf.b_proc = bp->b_proc;
856 	cbp->cb_buf.b_dev = ci->ci_dev;		/* XXX */
857 	cbp->cb_buf.b_blkno = cbn + cboff;
858 	cbp->cb_buf.b_data = addr;
859 	cbp->cb_buf.b_vp = ci->ci_vp;
860 	if (cs->sc_ileave == 0)
861 		cbc = dbtob((u_int64_t)(ci->ci_size - cbn));
862 	else
863 		cbc = dbtob((u_int64_t)(cs->sc_ileave - cboff));
864 	cbp->cb_buf.b_bcount = cbc < bcount ? cbc : bcount;
865 
866 	/*
867 	 * context for ccdiodone
868 	 */
869 	cbp->cb_obp = bp;
870 	cbp->cb_sc = cs;
871 	cbp->cb_comp = ccdisk;
872 
873 #ifdef DEBUG
874 	if (ccddebug & CCDB_IO)
875 		printf(" dev 0x%x(u%lu): cbp %p bn %" PRId64 " addr %p"
876 		       " bcnt %ld\n",
877 		    ci->ci_dev, (unsigned long) (ci-cs->sc_cinfo), cbp,
878 		    cbp->cb_buf.b_blkno, cbp->cb_buf.b_data,
879 		    cbp->cb_buf.b_bcount);
880 #endif
881 
882 	return (cbp);
883 }
884 
885 static void
886 ccdintr(cs, bp)
887 	struct ccd_softc *cs;
888 	struct buf *bp;
889 {
890 
891 #ifdef DEBUG
892 	if (ccddebug & CCDB_FOLLOW)
893 		printf("ccdintr(%p, %p)\n", cs, bp);
894 #endif
895 	/*
896 	 * Request is done for better or worse, wakeup the top half.
897 	 */
898 	if (bp->b_flags & B_ERROR)
899 		bp->b_resid = bp->b_bcount;
900 	disk_unbusy(&cs->sc_dkdev, (bp->b_bcount - bp->b_resid),
901 	    (bp->b_flags & B_READ));
902 	biodone(bp);
903 }
904 
905 /*
906  * Called at interrupt time.
907  * Mark the component as done and if all components are done,
908  * take a ccd interrupt.
909  */
910 void
911 ccdiodone(vbp)
912 	struct buf *vbp;
913 {
914 	struct ccdbuf *cbp = (struct ccdbuf *) vbp;
915 	struct buf *bp = cbp->cb_obp;
916 	struct ccd_softc *cs = cbp->cb_sc;
917 	int count, s;
918 
919 	s = splbio();
920 #ifdef DEBUG
921 	if (ccddebug & CCDB_FOLLOW)
922 		printf("ccdiodone(%p)\n", cbp);
923 	if (ccddebug & CCDB_IO) {
924 		printf("ccdiodone: bp %p bcount %ld resid %ld\n",
925 		       bp, bp->b_bcount, bp->b_resid);
926 		printf(" dev 0x%x(u%d), cbp %p bn %" PRId64 " addr %p"
927 		       " bcnt %ld\n",
928 		       cbp->cb_buf.b_dev, cbp->cb_comp, cbp,
929 		       cbp->cb_buf.b_blkno, cbp->cb_buf.b_data,
930 		       cbp->cb_buf.b_bcount);
931 	}
932 #endif
933 
934 	if (cbp->cb_buf.b_flags & B_ERROR) {
935 		bp->b_flags |= B_ERROR;
936 		bp->b_error = cbp->cb_buf.b_error ?
937 		    cbp->cb_buf.b_error : EIO;
938 
939 		printf("%s: error %d on component %d\n",
940 		       cs->sc_xname, bp->b_error, cbp->cb_comp);
941 	}
942 	count = cbp->cb_buf.b_bcount;
943 	CCD_PUTBUF(cbp);
944 
945 	/*
946 	 * If all done, "interrupt".
947 	 */
948 	bp->b_resid -= count;
949 	if (bp->b_resid < 0)
950 		panic("ccdiodone: count");
951 	if (bp->b_resid == 0)
952 		ccdintr(cs, bp);
953 	splx(s);
954 }
955 
956 /* ARGSUSED */
957 int
958 ccdread(dev, uio, flags)
959 	dev_t dev;
960 	struct uio *uio;
961 	int flags;
962 {
963 	int unit = ccdunit(dev);
964 	struct ccd_softc *cs;
965 
966 #ifdef DEBUG
967 	if (ccddebug & CCDB_FOLLOW)
968 		printf("ccdread(0x%x, %p)\n", dev, uio);
969 #endif
970 	if (unit >= numccd)
971 		return (ENXIO);
972 	cs = &ccd_softc[unit];
973 
974 	if ((cs->sc_flags & CCDF_INITED) == 0)
975 		return (ENXIO);
976 
977 	/*
978 	 * XXX: It's not clear that using minphys() is completely safe,
979 	 * in particular, for raw I/O.  Underlying devices might have some
980 	 * non-obvious limits, because of the copy to user-space.
981 	 */
982 	return (physio(ccdstrategy, NULL, dev, B_READ, minphys, uio));
983 }
984 
985 /* ARGSUSED */
986 int
987 ccdwrite(dev, uio, flags)
988 	dev_t dev;
989 	struct uio *uio;
990 	int flags;
991 {
992 	int unit = ccdunit(dev);
993 	struct ccd_softc *cs;
994 
995 #ifdef DEBUG
996 	if (ccddebug & CCDB_FOLLOW)
997 		printf("ccdwrite(0x%x, %p)\n", dev, uio);
998 #endif
999 	if (unit >= numccd)
1000 		return (ENXIO);
1001 	cs = &ccd_softc[unit];
1002 
1003 	if ((cs->sc_flags & CCDF_INITED) == 0)
1004 		return (ENXIO);
1005 
1006 	/*
1007 	 * XXX: It's not clear that using minphys() is completely safe,
1008 	 * in particular, for raw I/O.  Underlying devices might have some
1009 	 * non-obvious limits, because of the copy to user-space.
1010 	 */
1011 	return (physio(ccdstrategy, NULL, dev, B_WRITE, minphys, uio));
1012 }
1013 
1014 int
1015 ccdioctl(dev, cmd, data, flag, p)
1016 	dev_t dev;
1017 	u_long cmd;
1018 	caddr_t data;
1019 	int flag;
1020 	struct proc *p;
1021 {
1022 	int unit = ccdunit(dev);
1023 	int s, i, j, lookedup = 0, error;
1024 	int part, pmask;
1025 	struct buf *bp;
1026 	struct ccd_softc *cs;
1027 	struct ccd_ioctl *ccio = (struct ccd_ioctl *)data;
1028 	char **cpp;
1029 	struct vnode **vpp;
1030 #ifdef __HAVE_OLD_DISKLABEL
1031 	struct disklabel newlabel;
1032 #endif
1033 
1034 	if (unit >= numccd)
1035 		return (ENXIO);
1036 	cs = &ccd_softc[unit];
1037 
1038 	/* Must be open for writes for these commands... */
1039 	switch (cmd) {
1040 	case CCDIOCSET:
1041 	case CCDIOCCLR:
1042 	case DIOCSDINFO:
1043 	case DIOCWDINFO:
1044 #ifdef __HAVE_OLD_DISKLABEL
1045 	case ODIOCSDINFO:
1046 	case ODIOCWDINFO:
1047 #endif
1048 	case DIOCKLABEL:
1049 	case DIOCWLABEL:
1050 		if ((flag & FWRITE) == 0)
1051 			return (EBADF);
1052 	}
1053 
1054 	if ((error = lockmgr(&cs->sc_lock, LK_EXCLUSIVE, NULL)) != 0)
1055 		return (error);
1056 
1057 	/* Must be initialized for these... */
1058 	switch (cmd) {
1059 	case CCDIOCCLR:
1060 	case DIOCGDINFO:
1061 	case DIOCSDINFO:
1062 	case DIOCWDINFO:
1063 	case DIOCGPART:
1064 	case DIOCWLABEL:
1065 	case DIOCKLABEL:
1066 	case DIOCGDEFLABEL:
1067 #ifdef __HAVE_OLD_DISKLABEL
1068 	case ODIOCGDINFO:
1069 	case ODIOCSDINFO:
1070 	case ODIOCWDINFO:
1071 	case ODIOCGDEFLABEL:
1072 #endif
1073 		if ((cs->sc_flags & CCDF_INITED) == 0) {
1074 			error = ENXIO;
1075 			goto out;
1076 		}
1077 	}
1078 
1079 	switch (cmd) {
1080 	case CCDIOCSET:
1081 		if (cs->sc_flags & CCDF_INITED) {
1082 			error = EBUSY;
1083 			goto out;
1084 		}
1085 
1086 		/* Validate the flags. */
1087 		if ((ccio->ccio_flags & CCDF_USERMASK) != ccio->ccio_flags) {
1088 			error = EINVAL;
1089 			goto out;
1090 		}
1091 
1092 		if (ccio->ccio_ndisks > CCD_MAXNDISKS) {
1093 			error = EINVAL;
1094 			goto out;
1095 		}
1096 
1097 		/* Fill in some important bits. */
1098 		cs->sc_ileave = ccio->ccio_ileave;
1099 		cs->sc_nccdisks = ccio->ccio_ndisks;
1100 		cs->sc_flags = ccio->ccio_flags & CCDF_USERMASK;
1101 
1102 		/*
1103 		 * Allocate space for and copy in the array of
1104 		 * componet pathnames and device numbers.
1105 		 */
1106 		cpp = malloc(ccio->ccio_ndisks * sizeof(char *),
1107 		    M_DEVBUF, M_WAITOK);
1108 		vpp = malloc(ccio->ccio_ndisks * sizeof(struct vnode *),
1109 		    M_DEVBUF, M_WAITOK);
1110 
1111 		error = copyin(ccio->ccio_disks, cpp,
1112 		    ccio->ccio_ndisks * sizeof(char **));
1113 		if (error) {
1114 			free(vpp, M_DEVBUF);
1115 			free(cpp, M_DEVBUF);
1116 			goto out;
1117 		}
1118 
1119 #ifdef DEBUG
1120 		if (ccddebug & CCDB_INIT)
1121 			for (i = 0; i < ccio->ccio_ndisks; ++i)
1122 				printf("ccdioctl: component %d: 0x%p\n",
1123 				    i, cpp[i]);
1124 #endif
1125 
1126 		for (i = 0; i < ccio->ccio_ndisks; ++i) {
1127 #ifdef DEBUG
1128 			if (ccddebug & CCDB_INIT)
1129 				printf("ccdioctl: lookedup = %d\n", lookedup);
1130 #endif
1131 			if ((error = ccdlookup(cpp[i], p, &vpp[i])) != 0) {
1132 				for (j = 0; j < lookedup; ++j)
1133 					(void)vn_close(vpp[j], FREAD|FWRITE,
1134 					    p->p_ucred, p);
1135 				free(vpp, M_DEVBUF);
1136 				free(cpp, M_DEVBUF);
1137 				goto out;
1138 			}
1139 			++lookedup;
1140 		}
1141 
1142 		/*
1143 		 * Initialize the ccd.  Fills in the softc for us.
1144 		 */
1145 		if ((error = ccdinit(cs, cpp, vpp, p)) != 0) {
1146 			for (j = 0; j < lookedup; ++j)
1147 				(void)vn_close(vpp[j], FREAD|FWRITE,
1148 				    p->p_ucred, p);
1149 			free(vpp, M_DEVBUF);
1150 			free(cpp, M_DEVBUF);
1151 			goto out;
1152 		}
1153 
1154 		/* We can free the temporary variables now. */
1155 		free(vpp, M_DEVBUF);
1156 		free(cpp, M_DEVBUF);
1157 
1158 		/*
1159 		 * The ccd has been successfully initialized, so
1160 		 * we can place it into the array.  Don't try to
1161 		 * read the disklabel until the disk has been attached,
1162 		 * because space for the disklabel is allocated
1163 		 * in disk_attach();
1164 		 */
1165 		ccio->ccio_unit = unit;
1166 		ccio->ccio_size = cs->sc_size;
1167 
1168 		bufq_alloc(&cs->sc_bufq, BUFQ_FCFS);
1169 
1170 		/* Attach the disk. */
1171 		disk_attach(&cs->sc_dkdev);
1172 
1173 		/* Try and read the disklabel. */
1174 		ccdgetdisklabel(dev);
1175 		break;
1176 
1177 	case CCDIOCCLR:
1178 		/*
1179 		 * Don't unconfigure if any other partitions are open
1180 		 * or if both the character and block flavors of this
1181 		 * partition are open.
1182 		 */
1183 		part = DISKPART(dev);
1184 		pmask = (1 << part);
1185 		if ((cs->sc_dkdev.dk_openmask & ~pmask) ||
1186 		    ((cs->sc_dkdev.dk_bopenmask & pmask) &&
1187 		    (cs->sc_dkdev.dk_copenmask & pmask))) {
1188 			error = EBUSY;
1189 			goto out;
1190 		}
1191 
1192 		/* Kill off any queued buffers. */
1193 		s = splbio();
1194 		while ((bp = BUFQ_GET(&cs->sc_bufq)) != NULL) {
1195 			bp->b_error = EIO;
1196 			bp->b_flags |= B_ERROR;
1197 			bp->b_resid = bp->b_bcount;
1198 			biodone(bp);
1199 		}
1200 		splx(s);
1201 
1202 		bufq_free(&cs->sc_bufq);
1203 
1204 		/*
1205 		 * Free ccd_softc information and clear entry.
1206 		 */
1207 
1208 		/* Close the components and free their pathnames. */
1209 		for (i = 0; i < cs->sc_nccdisks; ++i) {
1210 			/*
1211 			 * XXX: this close could potentially fail and
1212 			 * cause Bad Things.  Maybe we need to force
1213 			 * the close to happen?
1214 			 */
1215 #ifdef DEBUG
1216 			if (ccddebug & CCDB_VNODE)
1217 				vprint("CCDIOCCLR: vnode info",
1218 				    cs->sc_cinfo[i].ci_vp);
1219 #endif
1220 			(void)vn_close(cs->sc_cinfo[i].ci_vp, FREAD|FWRITE,
1221 			    p->p_ucred, p);
1222 			free(cs->sc_cinfo[i].ci_path, M_DEVBUF);
1223 		}
1224 
1225 		/* Free interleave index. */
1226 		for (i = 0; cs->sc_itable[i].ii_ndisk; ++i)
1227 			free(cs->sc_itable[i].ii_index, M_DEVBUF);
1228 
1229 		/* Free component info and interleave table. */
1230 		free(cs->sc_cinfo, M_DEVBUF);
1231 		free(cs->sc_itable, M_DEVBUF);
1232 		cs->sc_flags &= ~(CCDF_INITED|CCDF_VLABEL);
1233 
1234 		/* Detatch the disk. */
1235 		disk_detach(&cs->sc_dkdev);
1236 		break;
1237 
1238 	case DIOCGDINFO:
1239 		*(struct disklabel *)data = *(cs->sc_dkdev.dk_label);
1240 		break;
1241 #ifdef __HAVE_OLD_DISKLABEL
1242 	case ODIOCGDINFO:
1243 		newlabel = *(cs->sc_dkdev.dk_label);
1244 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1245 			return ENOTTY;
1246 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1247 		break;
1248 #endif
1249 
1250 	case DIOCGPART:
1251 		((struct partinfo *)data)->disklab = cs->sc_dkdev.dk_label;
1252 		((struct partinfo *)data)->part =
1253 		    &cs->sc_dkdev.dk_label->d_partitions[DISKPART(dev)];
1254 		break;
1255 
1256 	case DIOCWDINFO:
1257 	case DIOCSDINFO:
1258 #ifdef __HAVE_OLD_DISKLABEL
1259 	case ODIOCWDINFO:
1260 	case ODIOCSDINFO:
1261 #endif
1262 	{
1263 		struct disklabel *lp;
1264 #ifdef __HAVE_OLD_DISKLABEL
1265 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
1266 			memset(&newlabel, 0, sizeof newlabel);
1267 			memcpy(&newlabel, data, sizeof (struct olddisklabel));
1268 			lp = &newlabel;
1269 		} else
1270 #endif
1271 		lp = (struct disklabel *)data;
1272 
1273 		cs->sc_flags |= CCDF_LABELLING;
1274 
1275 		error = setdisklabel(cs->sc_dkdev.dk_label,
1276 		    lp, 0, cs->sc_dkdev.dk_cpulabel);
1277 		if (error == 0) {
1278 			if (cmd == DIOCWDINFO
1279 #ifdef __HAVE_OLD_DISKLABEL
1280 			    || cmd == ODIOCWDINFO
1281 #endif
1282 			   )
1283 				error = writedisklabel(CCDLABELDEV(dev),
1284 				    ccdstrategy, cs->sc_dkdev.dk_label,
1285 				    cs->sc_dkdev.dk_cpulabel);
1286 		}
1287 
1288 		cs->sc_flags &= ~CCDF_LABELLING;
1289 		break;
1290 	}
1291 
1292 	case DIOCKLABEL:
1293 		if (*(int *)data != 0)
1294 			cs->sc_flags |= CCDF_KLABEL;
1295 		else
1296 			cs->sc_flags &= ~CCDF_KLABEL;
1297 		break;
1298 
1299 	case DIOCWLABEL:
1300 		if (*(int *)data != 0)
1301 			cs->sc_flags |= CCDF_WLABEL;
1302 		else
1303 			cs->sc_flags &= ~CCDF_WLABEL;
1304 		break;
1305 
1306 	case DIOCGDEFLABEL:
1307 		ccdgetdefaultlabel(cs, (struct disklabel *)data);
1308 		break;
1309 
1310 #ifdef __HAVE_OLD_DISKLABEL
1311 	case ODIOCGDEFLABEL:
1312 		ccdgetdefaultlabel(cs, &newlabel);
1313 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1314 			return ENOTTY;
1315 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1316 		break;
1317 #endif
1318 
1319 	default:
1320 		error = ENOTTY;
1321 	}
1322 
1323  out:
1324 	(void) lockmgr(&cs->sc_lock, LK_RELEASE, NULL);
1325 	return (error);
1326 }
1327 
1328 int
1329 ccdsize(dev)
1330 	dev_t dev;
1331 {
1332 	struct ccd_softc *cs;
1333 	struct disklabel *lp;
1334 	int part, unit, omask, size;
1335 
1336 	unit = ccdunit(dev);
1337 	if (unit >= numccd)
1338 		return (-1);
1339 	cs = &ccd_softc[unit];
1340 
1341 	if ((cs->sc_flags & CCDF_INITED) == 0)
1342 		return (-1);
1343 
1344 	part = DISKPART(dev);
1345 	omask = cs->sc_dkdev.dk_openmask & (1 << part);
1346 	lp = cs->sc_dkdev.dk_label;
1347 
1348 	if (omask == 0 && ccdopen(dev, 0, S_IFBLK, curproc))
1349 		return (-1);
1350 
1351 	if (lp->d_partitions[part].p_fstype != FS_SWAP)
1352 		size = -1;
1353 	else
1354 		size = lp->d_partitions[part].p_size *
1355 		    (lp->d_secsize / DEV_BSIZE);
1356 
1357 	if (omask == 0 && ccdclose(dev, 0, S_IFBLK, curproc))
1358 		return (-1);
1359 
1360 	return (size);
1361 }
1362 
1363 int
1364 ccddump(dev, blkno, va, size)
1365 	dev_t dev;
1366 	daddr_t blkno;
1367 	caddr_t va;
1368 	size_t size;
1369 {
1370 
1371 	/* Not implemented. */
1372 	return ENXIO;
1373 }
1374 
1375 /*
1376  * Lookup the provided name in the filesystem.  If the file exists,
1377  * is a valid block device, and isn't being used by anyone else,
1378  * set *vpp to the file's vnode.
1379  */
1380 static int
1381 ccdlookup(path, p, vpp)
1382 	char *path;
1383 	struct proc *p;
1384 	struct vnode **vpp;	/* result */
1385 {
1386 	struct nameidata nd;
1387 	struct vnode *vp;
1388 	struct vattr va;
1389 	int error;
1390 
1391 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, path, p);
1392 	if ((error = vn_open(&nd, FREAD|FWRITE, 0)) != 0) {
1393 #ifdef DEBUG
1394 		if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
1395 			printf("ccdlookup: vn_open error = %d\n", error);
1396 #endif
1397 		return (error);
1398 	}
1399 	vp = nd.ni_vp;
1400 
1401 	if (vp->v_usecount > 1) {
1402 		VOP_UNLOCK(vp, 0);
1403 		(void)vn_close(vp, FREAD|FWRITE, p->p_ucred, p);
1404 		return (EBUSY);
1405 	}
1406 
1407 	if ((error = VOP_GETATTR(vp, &va, p->p_ucred, p)) != 0) {
1408 #ifdef DEBUG
1409 		if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
1410 			printf("ccdlookup: getattr error = %d\n", error);
1411 #endif
1412 		VOP_UNLOCK(vp, 0);
1413 		(void)vn_close(vp, FREAD|FWRITE, p->p_ucred, p);
1414 		return (error);
1415 	}
1416 
1417 	/* XXX: eventually we should handle VREG, too. */
1418 	if (va.va_type != VBLK) {
1419 		VOP_UNLOCK(vp, 0);
1420 		(void)vn_close(vp, FREAD|FWRITE, p->p_ucred, p);
1421 		return (ENOTBLK);
1422 	}
1423 
1424 #ifdef DEBUG
1425 	if (ccddebug & CCDB_VNODE)
1426 		vprint("ccdlookup: vnode info", vp);
1427 #endif
1428 
1429 	VOP_UNLOCK(vp, 0);
1430 	*vpp = vp;
1431 	return (0);
1432 }
1433 
1434 static void
1435 ccdgetdefaultlabel(cs, lp)
1436 	struct ccd_softc *cs;
1437 	struct disklabel *lp;
1438 {
1439 	struct ccdgeom *ccg = &cs->sc_geom;
1440 
1441 	memset(lp, 0, sizeof(*lp));
1442 
1443 	lp->d_secperunit = cs->sc_size;
1444 	lp->d_secsize = ccg->ccg_secsize;
1445 	lp->d_nsectors = ccg->ccg_nsectors;
1446 	lp->d_ntracks = ccg->ccg_ntracks;
1447 	lp->d_ncylinders = ccg->ccg_ncylinders;
1448 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1449 
1450 	strncpy(lp->d_typename, "ccd", sizeof(lp->d_typename));
1451 	lp->d_type = DTYPE_CCD;
1452 	strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
1453 	lp->d_rpm = 3600;
1454 	lp->d_interleave = 1;
1455 	lp->d_flags = 0;
1456 
1457 	lp->d_partitions[RAW_PART].p_offset = 0;
1458 	lp->d_partitions[RAW_PART].p_size = cs->sc_size;
1459 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
1460 	lp->d_npartitions = RAW_PART + 1;
1461 
1462 	lp->d_magic = DISKMAGIC;
1463 	lp->d_magic2 = DISKMAGIC;
1464 	lp->d_checksum = dkcksum(cs->sc_dkdev.dk_label);
1465 }
1466 
1467 /*
1468  * Read the disklabel from the ccd.  If one is not present, fake one
1469  * up.
1470  */
1471 static void
1472 ccdgetdisklabel(dev)
1473 	dev_t dev;
1474 {
1475 	int unit = ccdunit(dev);
1476 	struct ccd_softc *cs = &ccd_softc[unit];
1477 	const char *errstring;
1478 	struct disklabel *lp = cs->sc_dkdev.dk_label;
1479 	struct cpu_disklabel *clp = cs->sc_dkdev.dk_cpulabel;
1480 
1481 	memset(clp, 0, sizeof(*clp));
1482 
1483 	ccdgetdefaultlabel(cs, lp);
1484 
1485 	/*
1486 	 * Call the generic disklabel extraction routine.
1487 	 */
1488 	errstring = readdisklabel(CCDLABELDEV(dev), ccdstrategy,
1489 	    cs->sc_dkdev.dk_label, cs->sc_dkdev.dk_cpulabel);
1490 	if (errstring)
1491 		ccdmakedisklabel(cs);
1492 	else {
1493 		int i;
1494 		struct partition *pp;
1495 
1496 		/*
1497 		 * Sanity check whether the found disklabel is valid.
1498 		 *
1499 		 * This is necessary since total size of ccd may vary
1500 		 * when an interleave is changed even though exactly
1501 		 * same componets are used, and old disklabel may used
1502 		 * if that is found.
1503 		 */
1504 		if (lp->d_secperunit != cs->sc_size)
1505 			printf("WARNING: %s: "
1506 			    "total sector size in disklabel (%d) != "
1507 			    "the size of ccd (%lu)\n", cs->sc_xname,
1508 			    lp->d_secperunit, (u_long)cs->sc_size);
1509 		for (i = 0; i < lp->d_npartitions; i++) {
1510 			pp = &lp->d_partitions[i];
1511 			if (pp->p_offset + pp->p_size > cs->sc_size)
1512 				printf("WARNING: %s: end of partition `%c' "
1513 				    "exceeds the size of ccd (%lu)\n",
1514 				    cs->sc_xname, 'a' + i, (u_long)cs->sc_size);
1515 		}
1516 	}
1517 
1518 #ifdef DEBUG
1519 	/* It's actually extremely common to have unlabeled ccds. */
1520 	if (ccddebug & CCDB_LABEL)
1521 		if (errstring != NULL)
1522 			printf("%s: %s\n", cs->sc_xname, errstring);
1523 #endif
1524 
1525 	/* In-core label now valid. */
1526 	cs->sc_flags |= CCDF_VLABEL;
1527 }
1528 
1529 /*
1530  * Take care of things one might want to take care of in the event
1531  * that a disklabel isn't present.
1532  */
1533 static void
1534 ccdmakedisklabel(cs)
1535 	struct ccd_softc *cs;
1536 {
1537 	struct disklabel *lp = cs->sc_dkdev.dk_label;
1538 
1539 	/*
1540 	 * For historical reasons, if there's no disklabel present
1541 	 * the raw partition must be marked FS_BSDFFS.
1542 	 */
1543 	lp->d_partitions[RAW_PART].p_fstype = FS_BSDFFS;
1544 
1545 	strncpy(lp->d_packname, "default label", sizeof(lp->d_packname));
1546 
1547 	lp->d_checksum = dkcksum(lp);
1548 }
1549 
1550 #ifdef DEBUG
1551 static void
1552 printiinfo(ii)
1553 	struct ccdiinfo *ii;
1554 {
1555 	int ix, i;
1556 
1557 	for (ix = 0; ii->ii_ndisk; ix++, ii++) {
1558 		printf(" itab[%d]: #dk %d sblk %" PRId64 " soff %" PRId64,
1559 		    ix, ii->ii_ndisk, ii->ii_startblk, ii->ii_startoff);
1560 		for (i = 0; i < ii->ii_ndisk; i++)
1561 			printf(" %d", ii->ii_index[i]);
1562 		printf("\n");
1563 	}
1564 }
1565 #endif
1566