xref: /netbsd-src/sys/dev/ccd.c (revision de1dfb1250df962f1ff3a011772cf58e605aed11)
1 /*	$NetBSD: ccd.c,v 1.98 2004/08/23 05:38:15 thorpej 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 1.98 2004/08/23 05:38:15 thorpej 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 static 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(int);
186 
187 /* called by biodone() at interrupt time */
188 static void	ccdiodone(struct buf *);
189 
190 static void	ccdstart(struct ccd_softc *);
191 static void	ccdinterleave(struct ccd_softc *);
192 static void	ccdintr(struct ccd_softc *, struct buf *);
193 static int	ccdinit(struct ccd_softc *, char **, struct vnode **,
194 		    struct proc *);
195 static int	ccdlookup(char *, struct proc *p, struct vnode **);
196 static struct ccdbuf *ccdbuffer(struct ccd_softc *, struct buf *,
197 		    daddr_t, caddr_t, long);
198 static void	ccdgetdefaultlabel(struct ccd_softc *, struct disklabel *);
199 static void	ccdgetdisklabel(dev_t);
200 static void	ccdmakedisklabel(struct ccd_softc *);
201 
202 static dev_type_open(ccdopen);
203 static dev_type_close(ccdclose);
204 static dev_type_read(ccdread);
205 static dev_type_write(ccdwrite);
206 static dev_type_ioctl(ccdioctl);
207 static dev_type_strategy(ccdstrategy);
208 static dev_type_dump(ccddump);
209 static 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(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(int num)
234 {
235 	struct ccd_softc *cs;
236 	int i;
237 
238 	if (num <= 0) {
239 #ifdef DIAGNOSTIC
240 		panic("ccdattach: count <= 0");
241 #endif
242 		return;
243 	}
244 
245 	ccd_softc = (struct ccd_softc *)malloc(num * sizeof(struct ccd_softc),
246 	    M_DEVBUF, M_NOWAIT|M_ZERO);
247 	if (ccd_softc == NULL) {
248 		printf("WARNING: no memory for concatenated disks\n");
249 		return;
250 	}
251 	numccd = num;
252 
253 	/* Initialize the component buffer pool. */
254 	pool_init(&ccd_cbufpool, sizeof(struct ccdbuf), 0,
255 	    0, 0, "ccdpl", NULL);
256 
257 	/* Initialize per-softc structures. */
258 	for (i = 0; i < num; i++) {
259 		cs = &ccd_softc[i];
260 		snprintf(cs->sc_xname, sizeof(cs->sc_xname), "ccd%d", i);
261 		cs->sc_dkdev.dk_name = cs->sc_xname;	/* XXX */
262 		lockinit(&cs->sc_lock, PRIBIO, "ccdlk", 0, 0);
263 	}
264 }
265 
266 static int
267 ccdinit(struct ccd_softc *cs, char **cpaths, struct vnode **vpp,
268     struct proc *p)
269 {
270 	struct ccdcinfo *ci = NULL;
271 	size_t size;
272 	int ix;
273 	struct vattr va;
274 	size_t minsize;
275 	int maxsecsize;
276 	struct partinfo dpart;
277 	struct ccdgeom *ccg = &cs->sc_geom;
278 	char tmppath[MAXPATHLEN];
279 	int error, path_alloced;
280 
281 #ifdef DEBUG
282 	if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
283 		printf("%s: ccdinit\n", cs->sc_xname);
284 #endif
285 
286 	/* Allocate space for the component info. */
287 	cs->sc_cinfo = malloc(cs->sc_nccdisks * sizeof(struct ccdcinfo),
288 	    M_DEVBUF, M_WAITOK);
289 
290 	cs->sc_size = 0;
291 
292 	/*
293 	 * Verify that each component piece exists and record
294 	 * relevant information about it.
295 	 */
296 	maxsecsize = 0;
297 	minsize = 0;
298 	for (ix = 0, path_alloced = 0; ix < cs->sc_nccdisks; ix++) {
299 		ci = &cs->sc_cinfo[ix];
300 		ci->ci_vp = vpp[ix];
301 
302 		/*
303 		 * Copy in the pathname of the component.
304 		 */
305 		memset(tmppath, 0, sizeof(tmppath));	/* sanity */
306 		error = copyinstr(cpaths[ix], tmppath,
307 		    MAXPATHLEN, &ci->ci_pathlen);
308 		if (error) {
309 #ifdef DEBUG
310 			if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
311 				printf("%s: can't copy path, error = %d\n",
312 				    cs->sc_xname, error);
313 #endif
314 			goto out;
315 		}
316 		ci->ci_path = malloc(ci->ci_pathlen, M_DEVBUF, M_WAITOK);
317 		memcpy(ci->ci_path, tmppath, ci->ci_pathlen);
318 		path_alloced++;
319 
320 		/*
321 		 * XXX: Cache the component's dev_t.
322 		 */
323 		if ((error = VOP_GETATTR(vpp[ix], &va, p->p_ucred, p)) != 0) {
324 #ifdef DEBUG
325 			if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
326 				printf("%s: %s: getattr failed %s = %d\n",
327 				    cs->sc_xname, ci->ci_path,
328 				    "error", error);
329 #endif
330 			goto out;
331 		}
332 		ci->ci_dev = va.va_rdev;
333 
334 		/*
335 		 * Get partition information for the component.
336 		 */
337 		error = VOP_IOCTL(vpp[ix], DIOCGPART, &dpart,
338 		    FREAD, p->p_ucred, p);
339 		if (error) {
340 #ifdef DEBUG
341 			if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
342 				 printf("%s: %s: ioctl failed, error = %d\n",
343 				     cs->sc_xname, ci->ci_path, error);
344 #endif
345 			goto out;
346 		}
347 
348 /*
349  * This diagnostic test is disabled (for now?) since not all port supports
350  * on-disk BSD disklabel.
351  */
352 #if 0 /* def DIAGNOSTIC */
353 		/* Check fstype field of component. */
354 		if (dpart.part->p_fstype != FS_CCD)
355 			printf("%s: WARNING: %s: fstype %d != FS_CCD\n",
356 			    cs->sc_xname, ci->ci_path, dpart.part->p_fstype);
357 #endif
358 
359 		/*
360 		 * Calculate the size, truncating to an interleave
361 		 * boundary if necessary.
362 		 */
363 		maxsecsize =
364 		    ((dpart.disklab->d_secsize > maxsecsize) ?
365 		    dpart.disklab->d_secsize : maxsecsize);
366 		size = dpart.part->p_size;
367 		if (cs->sc_ileave > 1)
368 			size -= size % cs->sc_ileave;
369 
370 		if (size == 0) {
371 #ifdef DEBUG
372 			if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
373 				printf("%s: %s: size == 0\n",
374 				    cs->sc_xname, ci->ci_path);
375 #endif
376 			error = ENODEV;
377 			goto out;
378 		}
379 
380 		if (minsize == 0 || size < minsize)
381 			minsize = size;
382 		ci->ci_size = size;
383 		cs->sc_size += size;
384 	}
385 
386 	/*
387 	 * Don't allow the interleave to be smaller than
388 	 * the biggest component sector.
389 	 */
390 	if ((cs->sc_ileave > 0) &&
391 	    (cs->sc_ileave < (maxsecsize / DEV_BSIZE))) {
392 #ifdef DEBUG
393 		if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
394 			printf("%s: interleave must be at least %d\n",
395 			    cs->sc_xname, (maxsecsize / DEV_BSIZE));
396 #endif
397 		error = EINVAL;
398 		goto out;
399 	}
400 
401 	/*
402 	 * If uniform interleave is desired set all sizes to that of
403 	 * the smallest component.
404 	 */
405 	if (cs->sc_flags & CCDF_UNIFORM) {
406 		for (ci = cs->sc_cinfo;
407 		     ci < &cs->sc_cinfo[cs->sc_nccdisks]; ci++)
408 			ci->ci_size = minsize;
409 
410 		cs->sc_size = cs->sc_nccdisks * minsize;
411 	}
412 
413 	/*
414 	 * Construct the interleave table.
415 	 */
416 	ccdinterleave(cs);
417 
418 	/*
419 	 * Create pseudo-geometry based on 1MB cylinders.  It's
420 	 * pretty close.
421 	 */
422 	ccg->ccg_secsize = DEV_BSIZE;
423 	ccg->ccg_ntracks = 1;
424 	ccg->ccg_nsectors = 1024 * (1024 / ccg->ccg_secsize);
425 	ccg->ccg_ncylinders = cs->sc_size / ccg->ccg_nsectors;
426 
427 	cs->sc_flags |= CCDF_INITED;
428 
429 	return (0);
430 
431  out:
432 	for (ix = 0; ix < path_alloced; ix++)
433 		free(cs->sc_cinfo[ix].ci_path, M_DEVBUF);
434 	free(cs->sc_cinfo, M_DEVBUF);
435 	return (error);
436 }
437 
438 static void
439 ccdinterleave(struct ccd_softc *cs)
440 {
441 	struct ccdcinfo *ci, *smallci;
442 	struct ccdiinfo *ii;
443 	daddr_t bn, lbn;
444 	int ix;
445 	u_long size;
446 
447 #ifdef DEBUG
448 	if (ccddebug & CCDB_INIT)
449 		printf("ccdinterleave(%p): ileave %d\n", cs, cs->sc_ileave);
450 #endif
451 	/*
452 	 * Allocate an interleave table.
453 	 * Chances are this is too big, but we don't care.
454 	 */
455 	size = (cs->sc_nccdisks + 1) * sizeof(struct ccdiinfo);
456 	cs->sc_itable = (struct ccdiinfo *)malloc(size, M_DEVBUF,
457 	    M_WAITOK|M_ZERO);
458 
459 	/*
460 	 * Trivial case: no interleave (actually interleave of disk size).
461 	 * Each table entry represents a single component in its entirety.
462 	 */
463 	if (cs->sc_ileave == 0) {
464 		bn = 0;
465 		ii = cs->sc_itable;
466 
467 		for (ix = 0; ix < cs->sc_nccdisks; ix++) {
468 			/* Allocate space for ii_index. */
469 			ii->ii_index = malloc(sizeof(int), M_DEVBUF, M_WAITOK);
470 			ii->ii_ndisk = 1;
471 			ii->ii_startblk = bn;
472 			ii->ii_startoff = 0;
473 			ii->ii_index[0] = ix;
474 			bn += cs->sc_cinfo[ix].ci_size;
475 			ii++;
476 		}
477 		ii->ii_ndisk = 0;
478 #ifdef DEBUG
479 		if (ccddebug & CCDB_INIT)
480 			printiinfo(cs->sc_itable);
481 #endif
482 		return;
483 	}
484 
485 	/*
486 	 * The following isn't fast or pretty; it doesn't have to be.
487 	 */
488 	size = 0;
489 	bn = lbn = 0;
490 	for (ii = cs->sc_itable; ; ii++) {
491 		/* Allocate space for ii_index. */
492 		ii->ii_index = malloc((sizeof(int) * cs->sc_nccdisks),
493 		    M_DEVBUF, M_WAITOK);
494 
495 		/*
496 		 * Locate the smallest of the remaining components
497 		 */
498 		smallci = NULL;
499 		for (ci = cs->sc_cinfo;
500 		     ci < &cs->sc_cinfo[cs->sc_nccdisks]; ci++)
501 			if (ci->ci_size > size &&
502 			    (smallci == NULL ||
503 			     ci->ci_size < smallci->ci_size))
504 				smallci = ci;
505 
506 		/*
507 		 * Nobody left, all done
508 		 */
509 		if (smallci == NULL) {
510 			ii->ii_ndisk = 0;
511 			break;
512 		}
513 
514 		/*
515 		 * Record starting logical block and component offset
516 		 */
517 		ii->ii_startblk = bn / cs->sc_ileave;
518 		ii->ii_startoff = lbn;
519 
520 		/*
521 		 * Determine how many disks take part in this interleave
522 		 * and record their indices.
523 		 */
524 		ix = 0;
525 		for (ci = cs->sc_cinfo;
526 		     ci < &cs->sc_cinfo[cs->sc_nccdisks]; ci++)
527 			if (ci->ci_size >= smallci->ci_size)
528 				ii->ii_index[ix++] = ci - cs->sc_cinfo;
529 		ii->ii_ndisk = ix;
530 		bn += ix * (smallci->ci_size - size);
531 		lbn = smallci->ci_size / cs->sc_ileave;
532 		size = smallci->ci_size;
533 	}
534 #ifdef DEBUG
535 	if (ccddebug & CCDB_INIT)
536 		printiinfo(cs->sc_itable);
537 #endif
538 }
539 
540 /* ARGSUSED */
541 static int
542 ccdopen(dev_t dev, int flags, int fmt, struct proc *p)
543 {
544 	int unit = ccdunit(dev);
545 	struct ccd_softc *cs;
546 	struct disklabel *lp;
547 	int error = 0, part, pmask;
548 
549 #ifdef DEBUG
550 	if (ccddebug & CCDB_FOLLOW)
551 		printf("ccdopen(0x%x, 0x%x)\n", dev, flags);
552 #endif
553 	if (unit >= numccd)
554 		return (ENXIO);
555 	cs = &ccd_softc[unit];
556 
557 	if ((error = lockmgr(&cs->sc_lock, LK_EXCLUSIVE, NULL)) != 0)
558 		return (error);
559 
560 	lp = cs->sc_dkdev.dk_label;
561 
562 	part = DISKPART(dev);
563 	pmask = (1 << part);
564 
565 	/*
566 	 * If we're initialized, check to see if there are any other
567 	 * open partitions.  If not, then it's safe to update
568 	 * the in-core disklabel.  Only read the disklabel if it is
569 	 * not already valid.
570 	 */
571 	if ((cs->sc_flags & (CCDF_INITED|CCDF_VLABEL)) == CCDF_INITED &&
572 	    cs->sc_dkdev.dk_openmask == 0)
573 		ccdgetdisklabel(dev);
574 
575 	/* Check that the partition exists. */
576 	if (part != RAW_PART) {
577 		if (((cs->sc_flags & CCDF_INITED) == 0) ||
578 		    ((part >= lp->d_npartitions) ||
579 		     (lp->d_partitions[part].p_fstype == FS_UNUSED))) {
580 			error = ENXIO;
581 			goto done;
582 		}
583 	}
584 
585 	/* Prevent our unit from being unconfigured while open. */
586 	switch (fmt) {
587 	case S_IFCHR:
588 		cs->sc_dkdev.dk_copenmask |= pmask;
589 		break;
590 
591 	case S_IFBLK:
592 		cs->sc_dkdev.dk_bopenmask |= pmask;
593 		break;
594 	}
595 	cs->sc_dkdev.dk_openmask =
596 	    cs->sc_dkdev.dk_copenmask | cs->sc_dkdev.dk_bopenmask;
597 
598  done:
599 	(void) lockmgr(&cs->sc_lock, LK_RELEASE, NULL);
600 	return (error);
601 }
602 
603 /* ARGSUSED */
604 static int
605 ccdclose(dev_t dev, int flags, int fmt, struct proc *p)
606 {
607 	int unit = ccdunit(dev);
608 	struct ccd_softc *cs;
609 	int error = 0, part;
610 
611 #ifdef DEBUG
612 	if (ccddebug & CCDB_FOLLOW)
613 		printf("ccdclose(0x%x, 0x%x)\n", dev, flags);
614 #endif
615 
616 	if (unit >= numccd)
617 		return (ENXIO);
618 	cs = &ccd_softc[unit];
619 
620 	if ((error = lockmgr(&cs->sc_lock, LK_EXCLUSIVE, NULL)) != 0)
621 		return (error);
622 
623 	part = DISKPART(dev);
624 
625 	/* ...that much closer to allowing unconfiguration... */
626 	switch (fmt) {
627 	case S_IFCHR:
628 		cs->sc_dkdev.dk_copenmask &= ~(1 << part);
629 		break;
630 
631 	case S_IFBLK:
632 		cs->sc_dkdev.dk_bopenmask &= ~(1 << part);
633 		break;
634 	}
635 	cs->sc_dkdev.dk_openmask =
636 	    cs->sc_dkdev.dk_copenmask | cs->sc_dkdev.dk_bopenmask;
637 
638 	if (cs->sc_dkdev.dk_openmask == 0) {
639 		if ((cs->sc_flags & CCDF_KLABEL) == 0)
640 			cs->sc_flags &= ~CCDF_VLABEL;
641 	}
642 
643 	(void) lockmgr(&cs->sc_lock, LK_RELEASE, NULL);
644 	return (0);
645 }
646 
647 static void
648 ccdstrategy(struct buf *bp)
649 {
650 	int unit = ccdunit(bp->b_dev);
651 	struct ccd_softc *cs = &ccd_softc[unit];
652 	daddr_t blkno;
653 	int s;
654 	int wlabel;
655 	struct disklabel *lp;
656 
657 #ifdef DEBUG
658 	if (ccddebug & CCDB_FOLLOW)
659 		printf("ccdstrategy(%p): unit %d\n", bp, unit);
660 #endif
661 	if ((cs->sc_flags & CCDF_INITED) == 0) {
662 #ifdef DEBUG
663 		if (ccddebug & CCDB_FOLLOW)
664 			printf("ccdstrategy: unit %d: not inited\n", unit);
665 #endif
666 		bp->b_error = ENXIO;
667 		bp->b_flags |= B_ERROR;
668 		goto done;
669 	}
670 
671 	/* If it's a nil transfer, wake up the top half now. */
672 	if (bp->b_bcount == 0)
673 		goto done;
674 
675 	lp = cs->sc_dkdev.dk_label;
676 
677 	/*
678 	 * Do bounds checking and adjust transfer.  If there's an
679 	 * error, the bounds check will flag that for us.  Convert
680 	 * the partition relative block number to an absolute.
681 	 */
682 	blkno = bp->b_blkno;
683 	wlabel = cs->sc_flags & (CCDF_WLABEL|CCDF_LABELLING);
684 	if (DISKPART(bp->b_dev) != RAW_PART) {
685 		if (bounds_check_with_label(&cs->sc_dkdev, bp, wlabel) <= 0)
686 			goto done;
687 		blkno += lp->d_partitions[DISKPART(bp->b_dev)].p_offset;
688 	}
689 	bp->b_rawblkno = blkno;
690 
691 	/* Place it in the queue and start I/O on the unit. */
692 	s = splbio();
693 	BUFQ_PUT(&cs->sc_bufq, bp);
694 	ccdstart(cs);
695 	splx(s);
696 	return;
697 
698  done:
699 	bp->b_resid = bp->b_bcount;
700 	biodone(bp);
701 }
702 
703 static void
704 ccdstart(struct ccd_softc *cs)
705 {
706 	long bcount, rcount;
707 	struct buf *bp;
708 	struct ccdbuf *cbp;
709 	caddr_t addr;
710 	daddr_t bn;
711 	SIMPLEQ_HEAD(, ccdbuf) cbufq;
712 
713 #ifdef DEBUG
714 	if (ccddebug & CCDB_FOLLOW)
715 		printf("ccdstart(%p)\n", cs);
716 #endif
717 
718 	/* See if there is work for us to do. */
719 	while ((bp = BUFQ_PEEK(&cs->sc_bufq)) != NULL) {
720 		/* Instrumentation. */
721 		disk_busy(&cs->sc_dkdev);
722 
723 		bp->b_resid = bp->b_bcount;
724 		bn = bp->b_rawblkno;
725 
726 		/* Allocate the component buffers. */
727 		SIMPLEQ_INIT(&cbufq);
728 		addr = bp->b_data;
729 		for (bcount = bp->b_bcount; bcount > 0; bcount -= rcount) {
730 			cbp = ccdbuffer(cs, bp, bn, addr, bcount);
731 			if (cbp == NULL) {
732 				/*
733 				 * Can't allocate a component buffer; just
734 				 * defer the job until later.
735 				 *
736 				 * XXX We might consider a watchdog timer
737 				 * XXX to make sure we are kicked into action,
738 				 * XXX or consider a low-water mark for our
739 				 * XXX component buffer pool.
740 				 */
741 				while ((cbp = SIMPLEQ_FIRST(&cbufq)) != NULL) {
742 					SIMPLEQ_REMOVE_HEAD(&cbufq, cb_q);
743 					CCD_PUTBUF(cbp);
744 				}
745 				disk_unbusy(&cs->sc_dkdev, 0, 0);
746 				return;
747 			}
748 			SIMPLEQ_INSERT_TAIL(&cbufq, cbp, cb_q);
749 			rcount = cbp->cb_buf.b_bcount;
750 			bn += btodb(rcount);
751 			addr += rcount;
752 		}
753 
754 		/* Transfer all set up, remove job from the queue. */
755 		(void) BUFQ_GET(&cs->sc_bufq);
756 
757 		/* Now fire off the requests. */
758 		while ((cbp = SIMPLEQ_FIRST(&cbufq)) != NULL) {
759 			SIMPLEQ_REMOVE_HEAD(&cbufq, cb_q);
760 			if ((cbp->cb_buf.b_flags & B_READ) == 0)
761 				cbp->cb_buf.b_vp->v_numoutput++;
762 			DEV_STRATEGY(&cbp->cb_buf);
763 		}
764 	}
765 }
766 
767 /*
768  * Build a component buffer header.
769  */
770 static struct ccdbuf *
771 ccdbuffer(struct ccd_softc *cs, struct buf *bp, daddr_t bn, caddr_t addr,
772     long bcount)
773 {
774 	struct ccdcinfo *ci;
775 	struct ccdbuf *cbp;
776 	daddr_t cbn, cboff;
777 	u_int64_t cbc;
778 	int ccdisk;
779 
780 #ifdef DEBUG
781 	if (ccddebug & CCDB_IO)
782 		printf("ccdbuffer(%p, %p, %" PRId64 ", %p, %ld)\n",
783 		       cs, bp, bn, addr, bcount);
784 #endif
785 	/*
786 	 * Determine which component bn falls in.
787 	 */
788 	cbn = bn;
789 	cboff = 0;
790 
791 	/*
792 	 * Serially concatenated
793 	 */
794 	if (cs->sc_ileave == 0) {
795 		daddr_t sblk;
796 
797 		sblk = 0;
798 		for (ccdisk = 0, ci = &cs->sc_cinfo[ccdisk];
799 		    cbn >= sblk + ci->ci_size;
800 		    ccdisk++, ci = &cs->sc_cinfo[ccdisk])
801 			sblk += ci->ci_size;
802 		cbn -= sblk;
803 	}
804 	/*
805 	 * Interleaved
806 	 */
807 	else {
808 		struct ccdiinfo *ii;
809 		int off;
810 
811 		cboff = cbn % cs->sc_ileave;
812 		cbn /= cs->sc_ileave;
813 		for (ii = cs->sc_itable; ii->ii_ndisk; ii++)
814 			if (ii->ii_startblk > cbn)
815 				break;
816 		ii--;
817 		off = cbn - ii->ii_startblk;
818 		if (ii->ii_ndisk == 1) {
819 			ccdisk = ii->ii_index[0];
820 			cbn = ii->ii_startoff + off;
821 		} else {
822 			ccdisk = ii->ii_index[off % ii->ii_ndisk];
823 			cbn = ii->ii_startoff + off / ii->ii_ndisk;
824 		}
825 		cbn *= cs->sc_ileave;
826 		ci = &cs->sc_cinfo[ccdisk];
827 	}
828 
829 	/*
830 	 * Fill in the component buf structure.
831 	 */
832 	cbp = CCD_GETBUF();
833 	if (cbp == NULL)
834 		return (NULL);
835 	BUF_INIT(&cbp->cb_buf);
836 	cbp->cb_buf.b_flags = bp->b_flags | B_CALL;
837 	cbp->cb_buf.b_iodone = ccdiodone;
838 	cbp->cb_buf.b_proc = bp->b_proc;
839 	cbp->cb_buf.b_dev = ci->ci_dev;
840 	cbp->cb_buf.b_blkno = cbn + cboff;
841 	cbp->cb_buf.b_data = addr;
842 	cbp->cb_buf.b_vp = ci->ci_vp;
843 	if (cs->sc_ileave == 0)
844 		cbc = dbtob((u_int64_t)(ci->ci_size - cbn));
845 	else
846 		cbc = dbtob((u_int64_t)(cs->sc_ileave - cboff));
847 	cbp->cb_buf.b_bcount = cbc < bcount ? cbc : bcount;
848 
849 	/*
850 	 * context for ccdiodone
851 	 */
852 	cbp->cb_obp = bp;
853 	cbp->cb_sc = cs;
854 	cbp->cb_comp = ccdisk;
855 
856 	BIO_COPYPRIO(&cbp->cb_buf, bp);
857 
858 #ifdef DEBUG
859 	if (ccddebug & CCDB_IO)
860 		printf(" dev 0x%x(u%lu): cbp %p bn %" PRId64 " addr %p"
861 		       " bcnt %ld\n",
862 		    ci->ci_dev, (unsigned long) (ci-cs->sc_cinfo), cbp,
863 		    cbp->cb_buf.b_blkno, cbp->cb_buf.b_data,
864 		    cbp->cb_buf.b_bcount);
865 #endif
866 
867 	return (cbp);
868 }
869 
870 static void
871 ccdintr(struct ccd_softc *cs, struct buf *bp)
872 {
873 
874 #ifdef DEBUG
875 	if (ccddebug & CCDB_FOLLOW)
876 		printf("ccdintr(%p, %p)\n", cs, bp);
877 #endif
878 	/*
879 	 * Request is done for better or worse, wakeup the top half.
880 	 */
881 	if (bp->b_flags & B_ERROR)
882 		bp->b_resid = bp->b_bcount;
883 	disk_unbusy(&cs->sc_dkdev, (bp->b_bcount - bp->b_resid),
884 	    (bp->b_flags & B_READ));
885 	biodone(bp);
886 }
887 
888 /*
889  * Called at interrupt time.
890  * Mark the component as done and if all components are done,
891  * take a ccd interrupt.
892  */
893 static void
894 ccdiodone(struct buf *vbp)
895 {
896 	struct ccdbuf *cbp = (struct ccdbuf *) vbp;
897 	struct buf *bp = cbp->cb_obp;
898 	struct ccd_softc *cs = cbp->cb_sc;
899 	int count, s;
900 
901 	s = splbio();
902 #ifdef DEBUG
903 	if (ccddebug & CCDB_FOLLOW)
904 		printf("ccdiodone(%p)\n", cbp);
905 	if (ccddebug & CCDB_IO) {
906 		printf("ccdiodone: bp %p bcount %ld resid %ld\n",
907 		       bp, bp->b_bcount, bp->b_resid);
908 		printf(" dev 0x%x(u%d), cbp %p bn %" PRId64 " addr %p"
909 		       " bcnt %ld\n",
910 		       cbp->cb_buf.b_dev, cbp->cb_comp, cbp,
911 		       cbp->cb_buf.b_blkno, cbp->cb_buf.b_data,
912 		       cbp->cb_buf.b_bcount);
913 	}
914 #endif
915 
916 	if (cbp->cb_buf.b_flags & B_ERROR) {
917 		bp->b_flags |= B_ERROR;
918 		bp->b_error = cbp->cb_buf.b_error ?
919 		    cbp->cb_buf.b_error : EIO;
920 
921 		printf("%s: error %d on component %d\n",
922 		       cs->sc_xname, bp->b_error, cbp->cb_comp);
923 	}
924 	count = cbp->cb_buf.b_bcount;
925 	CCD_PUTBUF(cbp);
926 
927 	/*
928 	 * If all done, "interrupt".
929 	 */
930 	bp->b_resid -= count;
931 	if (bp->b_resid < 0)
932 		panic("ccdiodone: count");
933 	if (bp->b_resid == 0)
934 		ccdintr(cs, bp);
935 	splx(s);
936 }
937 
938 /* ARGSUSED */
939 static int
940 ccdread(dev_t dev, struct uio *uio, int flags)
941 {
942 	int unit = ccdunit(dev);
943 	struct ccd_softc *cs;
944 
945 #ifdef DEBUG
946 	if (ccddebug & CCDB_FOLLOW)
947 		printf("ccdread(0x%x, %p)\n", dev, uio);
948 #endif
949 	if (unit >= numccd)
950 		return (ENXIO);
951 	cs = &ccd_softc[unit];
952 
953 	if ((cs->sc_flags & CCDF_INITED) == 0)
954 		return (ENXIO);
955 
956 	return (physio(ccdstrategy, NULL, dev, B_READ, minphys, uio));
957 }
958 
959 /* ARGSUSED */
960 static int
961 ccdwrite(dev_t dev, struct uio *uio, int flags)
962 {
963 	int unit = ccdunit(dev);
964 	struct ccd_softc *cs;
965 
966 #ifdef DEBUG
967 	if (ccddebug & CCDB_FOLLOW)
968 		printf("ccdwrite(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 	return (physio(ccdstrategy, NULL, dev, B_WRITE, minphys, uio));
978 }
979 
980 static int
981 ccdioctl(dev_t dev, u_long cmd, caddr_t data, int flag, struct proc *p)
982 {
983 	int unit = ccdunit(dev);
984 	int s, i, j, lookedup = 0, error;
985 	int part, pmask;
986 	struct buf *bp;
987 	struct ccd_softc *cs;
988 	struct ccd_ioctl *ccio = (struct ccd_ioctl *)data;
989 	char **cpp;
990 	struct vnode **vpp;
991 #ifdef __HAVE_OLD_DISKLABEL
992 	struct disklabel newlabel;
993 #endif
994 
995 	if (unit >= numccd)
996 		return (ENXIO);
997 	cs = &ccd_softc[unit];
998 
999 	/* Must be open for writes for these commands... */
1000 	switch (cmd) {
1001 	case CCDIOCSET:
1002 	case CCDIOCCLR:
1003 	case DIOCSDINFO:
1004 	case DIOCWDINFO:
1005 #ifdef __HAVE_OLD_DISKLABEL
1006 	case ODIOCSDINFO:
1007 	case ODIOCWDINFO:
1008 #endif
1009 	case DIOCKLABEL:
1010 	case DIOCWLABEL:
1011 		if ((flag & FWRITE) == 0)
1012 			return (EBADF);
1013 	}
1014 
1015 	if ((error = lockmgr(&cs->sc_lock, LK_EXCLUSIVE, NULL)) != 0)
1016 		return (error);
1017 
1018 	/* Must be initialized for these... */
1019 	switch (cmd) {
1020 	case CCDIOCCLR:
1021 	case DIOCGDINFO:
1022 	case DIOCSDINFO:
1023 	case DIOCWDINFO:
1024 	case DIOCGPART:
1025 	case DIOCWLABEL:
1026 	case DIOCKLABEL:
1027 	case DIOCGDEFLABEL:
1028 #ifdef __HAVE_OLD_DISKLABEL
1029 	case ODIOCGDINFO:
1030 	case ODIOCSDINFO:
1031 	case ODIOCWDINFO:
1032 	case ODIOCGDEFLABEL:
1033 #endif
1034 		if ((cs->sc_flags & CCDF_INITED) == 0) {
1035 			error = ENXIO;
1036 			goto out;
1037 		}
1038 	}
1039 
1040 	switch (cmd) {
1041 	case CCDIOCSET:
1042 		if (cs->sc_flags & CCDF_INITED) {
1043 			error = EBUSY;
1044 			goto out;
1045 		}
1046 
1047 		/* Validate the flags. */
1048 		if ((ccio->ccio_flags & CCDF_USERMASK) != ccio->ccio_flags) {
1049 			error = EINVAL;
1050 			goto out;
1051 		}
1052 
1053 		if (ccio->ccio_ndisks > CCD_MAXNDISKS) {
1054 			error = EINVAL;
1055 			goto out;
1056 		}
1057 
1058 		/* Fill in some important bits. */
1059 		cs->sc_ileave = ccio->ccio_ileave;
1060 		cs->sc_nccdisks = ccio->ccio_ndisks;
1061 		cs->sc_flags = ccio->ccio_flags & CCDF_USERMASK;
1062 
1063 		/*
1064 		 * Allocate space for and copy in the array of
1065 		 * componet pathnames and device numbers.
1066 		 */
1067 		cpp = malloc(ccio->ccio_ndisks * sizeof(char *),
1068 		    M_DEVBUF, M_WAITOK);
1069 		vpp = malloc(ccio->ccio_ndisks * sizeof(struct vnode *),
1070 		    M_DEVBUF, M_WAITOK);
1071 
1072 		error = copyin(ccio->ccio_disks, cpp,
1073 		    ccio->ccio_ndisks * sizeof(char **));
1074 		if (error) {
1075 			free(vpp, M_DEVBUF);
1076 			free(cpp, M_DEVBUF);
1077 			goto out;
1078 		}
1079 
1080 #ifdef DEBUG
1081 		if (ccddebug & CCDB_INIT)
1082 			for (i = 0; i < ccio->ccio_ndisks; ++i)
1083 				printf("ccdioctl: component %d: 0x%p\n",
1084 				    i, cpp[i]);
1085 #endif
1086 
1087 		for (i = 0; i < ccio->ccio_ndisks; ++i) {
1088 #ifdef DEBUG
1089 			if (ccddebug & CCDB_INIT)
1090 				printf("ccdioctl: lookedup = %d\n", lookedup);
1091 #endif
1092 			if ((error = ccdlookup(cpp[i], p, &vpp[i])) != 0) {
1093 				for (j = 0; j < lookedup; ++j)
1094 					(void)vn_close(vpp[j], FREAD|FWRITE,
1095 					    p->p_ucred, p);
1096 				free(vpp, M_DEVBUF);
1097 				free(cpp, M_DEVBUF);
1098 				goto out;
1099 			}
1100 			++lookedup;
1101 		}
1102 
1103 		/*
1104 		 * Initialize the ccd.  Fills in the softc for us.
1105 		 */
1106 		if ((error = ccdinit(cs, cpp, vpp, p)) != 0) {
1107 			for (j = 0; j < lookedup; ++j)
1108 				(void)vn_close(vpp[j], FREAD|FWRITE,
1109 				    p->p_ucred, p);
1110 			free(vpp, M_DEVBUF);
1111 			free(cpp, M_DEVBUF);
1112 			goto out;
1113 		}
1114 
1115 		/* We can free the temporary variables now. */
1116 		free(vpp, M_DEVBUF);
1117 		free(cpp, M_DEVBUF);
1118 
1119 		/*
1120 		 * The ccd has been successfully initialized, so
1121 		 * we can place it into the array.  Don't try to
1122 		 * read the disklabel until the disk has been attached,
1123 		 * because space for the disklabel is allocated
1124 		 * in disk_attach();
1125 		 */
1126 		ccio->ccio_unit = unit;
1127 		ccio->ccio_size = cs->sc_size;
1128 
1129 		bufq_alloc(&cs->sc_bufq, BUFQ_FCFS);
1130 
1131 		/* Attach the disk. */
1132 		disk_attach(&cs->sc_dkdev);
1133 
1134 		/* Try and read the disklabel. */
1135 		ccdgetdisklabel(dev);
1136 		break;
1137 
1138 	case CCDIOCCLR:
1139 		/*
1140 		 * Don't unconfigure if any other partitions are open
1141 		 * or if both the character and block flavors of this
1142 		 * partition are open.
1143 		 */
1144 		part = DISKPART(dev);
1145 		pmask = (1 << part);
1146 		if ((cs->sc_dkdev.dk_openmask & ~pmask) ||
1147 		    ((cs->sc_dkdev.dk_bopenmask & pmask) &&
1148 		    (cs->sc_dkdev.dk_copenmask & pmask))) {
1149 			error = EBUSY;
1150 			goto out;
1151 		}
1152 
1153 		/* Kill off any queued buffers. */
1154 		s = splbio();
1155 		while ((bp = BUFQ_GET(&cs->sc_bufq)) != NULL) {
1156 			bp->b_error = EIO;
1157 			bp->b_flags |= B_ERROR;
1158 			bp->b_resid = bp->b_bcount;
1159 			biodone(bp);
1160 		}
1161 		splx(s);
1162 
1163 		bufq_free(&cs->sc_bufq);
1164 
1165 		/*
1166 		 * Free ccd_softc information and clear entry.
1167 		 */
1168 
1169 		/* Close the components and free their pathnames. */
1170 		for (i = 0; i < cs->sc_nccdisks; ++i) {
1171 			/*
1172 			 * XXX: this close could potentially fail and
1173 			 * cause Bad Things.  Maybe we need to force
1174 			 * the close to happen?
1175 			 */
1176 #ifdef DEBUG
1177 			if (ccddebug & CCDB_VNODE)
1178 				vprint("CCDIOCCLR: vnode info",
1179 				    cs->sc_cinfo[i].ci_vp);
1180 #endif
1181 			(void)vn_close(cs->sc_cinfo[i].ci_vp, FREAD|FWRITE,
1182 			    p->p_ucred, p);
1183 			free(cs->sc_cinfo[i].ci_path, M_DEVBUF);
1184 		}
1185 
1186 		/* Free interleave index. */
1187 		for (i = 0; cs->sc_itable[i].ii_ndisk; ++i)
1188 			free(cs->sc_itable[i].ii_index, M_DEVBUF);
1189 
1190 		/* Free component info and interleave table. */
1191 		free(cs->sc_cinfo, M_DEVBUF);
1192 		free(cs->sc_itable, M_DEVBUF);
1193 		cs->sc_flags &= ~(CCDF_INITED|CCDF_VLABEL);
1194 
1195 		/* Detatch the disk. */
1196 		disk_detach(&cs->sc_dkdev);
1197 		break;
1198 
1199 	case DIOCGDINFO:
1200 		*(struct disklabel *)data = *(cs->sc_dkdev.dk_label);
1201 		break;
1202 #ifdef __HAVE_OLD_DISKLABEL
1203 	case ODIOCGDINFO:
1204 		newlabel = *(cs->sc_dkdev.dk_label);
1205 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1206 			return ENOTTY;
1207 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1208 		break;
1209 #endif
1210 
1211 	case DIOCGPART:
1212 		((struct partinfo *)data)->disklab = cs->sc_dkdev.dk_label;
1213 		((struct partinfo *)data)->part =
1214 		    &cs->sc_dkdev.dk_label->d_partitions[DISKPART(dev)];
1215 		break;
1216 
1217 	case DIOCWDINFO:
1218 	case DIOCSDINFO:
1219 #ifdef __HAVE_OLD_DISKLABEL
1220 	case ODIOCWDINFO:
1221 	case ODIOCSDINFO:
1222 #endif
1223 	{
1224 		struct disklabel *lp;
1225 #ifdef __HAVE_OLD_DISKLABEL
1226 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
1227 			memset(&newlabel, 0, sizeof newlabel);
1228 			memcpy(&newlabel, data, sizeof (struct olddisklabel));
1229 			lp = &newlabel;
1230 		} else
1231 #endif
1232 		lp = (struct disklabel *)data;
1233 
1234 		cs->sc_flags |= CCDF_LABELLING;
1235 
1236 		error = setdisklabel(cs->sc_dkdev.dk_label,
1237 		    lp, 0, cs->sc_dkdev.dk_cpulabel);
1238 		if (error == 0) {
1239 			if (cmd == DIOCWDINFO
1240 #ifdef __HAVE_OLD_DISKLABEL
1241 			    || cmd == ODIOCWDINFO
1242 #endif
1243 			   )
1244 				error = writedisklabel(CCDLABELDEV(dev),
1245 				    ccdstrategy, cs->sc_dkdev.dk_label,
1246 				    cs->sc_dkdev.dk_cpulabel);
1247 		}
1248 
1249 		cs->sc_flags &= ~CCDF_LABELLING;
1250 		break;
1251 	}
1252 
1253 	case DIOCKLABEL:
1254 		if (*(int *)data != 0)
1255 			cs->sc_flags |= CCDF_KLABEL;
1256 		else
1257 			cs->sc_flags &= ~CCDF_KLABEL;
1258 		break;
1259 
1260 	case DIOCWLABEL:
1261 		if (*(int *)data != 0)
1262 			cs->sc_flags |= CCDF_WLABEL;
1263 		else
1264 			cs->sc_flags &= ~CCDF_WLABEL;
1265 		break;
1266 
1267 	case DIOCGDEFLABEL:
1268 		ccdgetdefaultlabel(cs, (struct disklabel *)data);
1269 		break;
1270 
1271 #ifdef __HAVE_OLD_DISKLABEL
1272 	case ODIOCGDEFLABEL:
1273 		ccdgetdefaultlabel(cs, &newlabel);
1274 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1275 			return ENOTTY;
1276 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1277 		break;
1278 #endif
1279 
1280 	default:
1281 		error = ENOTTY;
1282 	}
1283 
1284  out:
1285 	(void) lockmgr(&cs->sc_lock, LK_RELEASE, NULL);
1286 	return (error);
1287 }
1288 
1289 static int
1290 ccdsize(dev_t dev)
1291 {
1292 	struct ccd_softc *cs;
1293 	struct disklabel *lp;
1294 	int part, unit, omask, size;
1295 
1296 	unit = ccdunit(dev);
1297 	if (unit >= numccd)
1298 		return (-1);
1299 	cs = &ccd_softc[unit];
1300 
1301 	if ((cs->sc_flags & CCDF_INITED) == 0)
1302 		return (-1);
1303 
1304 	part = DISKPART(dev);
1305 	omask = cs->sc_dkdev.dk_openmask & (1 << part);
1306 	lp = cs->sc_dkdev.dk_label;
1307 
1308 	if (omask == 0 && ccdopen(dev, 0, S_IFBLK, curproc))
1309 		return (-1);
1310 
1311 	if (lp->d_partitions[part].p_fstype != FS_SWAP)
1312 		size = -1;
1313 	else
1314 		size = lp->d_partitions[part].p_size *
1315 		    (lp->d_secsize / DEV_BSIZE);
1316 
1317 	if (omask == 0 && ccdclose(dev, 0, S_IFBLK, curproc))
1318 		return (-1);
1319 
1320 	return (size);
1321 }
1322 
1323 static int
1324 ccddump(dev_t dev, daddr_t blkno, caddr_t va, size_t size)
1325 {
1326 
1327 	/* Not implemented. */
1328 	return ENXIO;
1329 }
1330 
1331 /*
1332  * Lookup the provided name in the filesystem.  If the file exists,
1333  * is a valid block device, and isn't being used by anyone else,
1334  * set *vpp to the file's vnode.
1335  */
1336 static int
1337 ccdlookup(char *path, struct proc *p, struct vnode **vpp /* result */)
1338 {
1339 	struct nameidata nd;
1340 	struct vnode *vp;
1341 	struct vattr va;
1342 	int error;
1343 
1344 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, path, p);
1345 	if ((error = vn_open(&nd, FREAD|FWRITE, 0)) != 0) {
1346 #ifdef DEBUG
1347 		if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
1348 			printf("ccdlookup: vn_open error = %d\n", error);
1349 #endif
1350 		return (error);
1351 	}
1352 	vp = nd.ni_vp;
1353 
1354 	if (vp->v_usecount > 1) {
1355 		VOP_UNLOCK(vp, 0);
1356 		(void)vn_close(vp, FREAD|FWRITE, p->p_ucred, p);
1357 		return (EBUSY);
1358 	}
1359 
1360 	if ((error = VOP_GETATTR(vp, &va, p->p_ucred, p)) != 0) {
1361 #ifdef DEBUG
1362 		if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
1363 			printf("ccdlookup: getattr error = %d\n", error);
1364 #endif
1365 		VOP_UNLOCK(vp, 0);
1366 		(void)vn_close(vp, FREAD|FWRITE, p->p_ucred, p);
1367 		return (error);
1368 	}
1369 
1370 	/* XXX: eventually we should handle VREG, too. */
1371 	if (va.va_type != VBLK) {
1372 		VOP_UNLOCK(vp, 0);
1373 		(void)vn_close(vp, FREAD|FWRITE, p->p_ucred, p);
1374 		return (ENOTBLK);
1375 	}
1376 
1377 #ifdef DEBUG
1378 	if (ccddebug & CCDB_VNODE)
1379 		vprint("ccdlookup: vnode info", vp);
1380 #endif
1381 
1382 	VOP_UNLOCK(vp, 0);
1383 	*vpp = vp;
1384 	return (0);
1385 }
1386 
1387 static void
1388 ccdgetdefaultlabel(struct ccd_softc *cs, struct disklabel *lp)
1389 {
1390 	struct ccdgeom *ccg = &cs->sc_geom;
1391 
1392 	memset(lp, 0, sizeof(*lp));
1393 
1394 	lp->d_secperunit = cs->sc_size;
1395 	lp->d_secsize = ccg->ccg_secsize;
1396 	lp->d_nsectors = ccg->ccg_nsectors;
1397 	lp->d_ntracks = ccg->ccg_ntracks;
1398 	lp->d_ncylinders = ccg->ccg_ncylinders;
1399 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1400 
1401 	strncpy(lp->d_typename, "ccd", sizeof(lp->d_typename));
1402 	lp->d_type = DTYPE_CCD;
1403 	strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
1404 	lp->d_rpm = 3600;
1405 	lp->d_interleave = 1;
1406 	lp->d_flags = 0;
1407 
1408 	lp->d_partitions[RAW_PART].p_offset = 0;
1409 	lp->d_partitions[RAW_PART].p_size = cs->sc_size;
1410 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
1411 	lp->d_npartitions = RAW_PART + 1;
1412 
1413 	lp->d_magic = DISKMAGIC;
1414 	lp->d_magic2 = DISKMAGIC;
1415 	lp->d_checksum = dkcksum(cs->sc_dkdev.dk_label);
1416 }
1417 
1418 /*
1419  * Read the disklabel from the ccd.  If one is not present, fake one
1420  * up.
1421  */
1422 static void
1423 ccdgetdisklabel(dev_t dev)
1424 {
1425 	int unit = ccdunit(dev);
1426 	struct ccd_softc *cs = &ccd_softc[unit];
1427 	const char *errstring;
1428 	struct disklabel *lp = cs->sc_dkdev.dk_label;
1429 	struct cpu_disklabel *clp = cs->sc_dkdev.dk_cpulabel;
1430 
1431 	memset(clp, 0, sizeof(*clp));
1432 
1433 	ccdgetdefaultlabel(cs, lp);
1434 
1435 	/*
1436 	 * Call the generic disklabel extraction routine.
1437 	 */
1438 	if ((cs->sc_flags & CCDF_NOLABEL) != 0)
1439 		errstring = "CCDF_NOLABEL set; ignoring on-disk label";
1440 	else
1441 		errstring = readdisklabel(CCDLABELDEV(dev), ccdstrategy,
1442 		    cs->sc_dkdev.dk_label, cs->sc_dkdev.dk_cpulabel);
1443 	if (errstring)
1444 		ccdmakedisklabel(cs);
1445 	else {
1446 		int i;
1447 		struct partition *pp;
1448 
1449 		/*
1450 		 * Sanity check whether the found disklabel is valid.
1451 		 *
1452 		 * This is necessary since total size of ccd may vary
1453 		 * when an interleave is changed even though exactly
1454 		 * same componets are used, and old disklabel may used
1455 		 * if that is found.
1456 		 */
1457 		if (lp->d_secperunit != cs->sc_size)
1458 			printf("WARNING: %s: "
1459 			    "total sector size in disklabel (%d) != "
1460 			    "the size of ccd (%lu)\n", cs->sc_xname,
1461 			    lp->d_secperunit, (u_long)cs->sc_size);
1462 		for (i = 0; i < lp->d_npartitions; i++) {
1463 			pp = &lp->d_partitions[i];
1464 			if (pp->p_offset + pp->p_size > cs->sc_size)
1465 				printf("WARNING: %s: end of partition `%c' "
1466 				    "exceeds the size of ccd (%lu)\n",
1467 				    cs->sc_xname, 'a' + i, (u_long)cs->sc_size);
1468 		}
1469 	}
1470 
1471 #ifdef DEBUG
1472 	/* It's actually extremely common to have unlabeled ccds. */
1473 	if (ccddebug & CCDB_LABEL)
1474 		if (errstring != NULL)
1475 			printf("%s: %s\n", cs->sc_xname, errstring);
1476 #endif
1477 
1478 	/* In-core label now valid. */
1479 	cs->sc_flags |= CCDF_VLABEL;
1480 }
1481 
1482 /*
1483  * Take care of things one might want to take care of in the event
1484  * that a disklabel isn't present.
1485  */
1486 static void
1487 ccdmakedisklabel(struct ccd_softc *cs)
1488 {
1489 	struct disklabel *lp = cs->sc_dkdev.dk_label;
1490 
1491 	/*
1492 	 * For historical reasons, if there's no disklabel present
1493 	 * the raw partition must be marked FS_BSDFFS.
1494 	 */
1495 	lp->d_partitions[RAW_PART].p_fstype = FS_BSDFFS;
1496 
1497 	strncpy(lp->d_packname, "default label", sizeof(lp->d_packname));
1498 
1499 	lp->d_checksum = dkcksum(lp);
1500 }
1501 
1502 #ifdef DEBUG
1503 static void
1504 printiinfo(struct ccdiinfo *ii)
1505 {
1506 	int ix, i;
1507 
1508 	for (ix = 0; ii->ii_ndisk; ix++, ii++) {
1509 		printf(" itab[%d]: #dk %d sblk %" PRId64 " soff %" PRId64,
1510 		    ix, ii->ii_ndisk, ii->ii_startblk, ii->ii_startoff);
1511 		for (i = 0; i < ii->ii_ndisk; i++)
1512 			printf(" %d", ii->ii_index[i]);
1513 		printf("\n");
1514 	}
1515 }
1516 #endif
1517