xref: /netbsd-src/sys/dev/ccd.c (revision 1ffa7b76c40339c17a0fb2a09fac93f287cfc046)
1 /*	$NetBSD: ccd.c,v 1.85 2003/05/02 08:45:26 dsl 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) 1988 University of Utah.
41  * Copyright (c) 1990, 1993
42  *	The Regents of the University of California.  All rights reserved.
43  *
44  * This code is derived from software contributed to Berkeley by
45  * the Systems Programming Group of the University of Utah Computer
46  * Science Department.
47  *
48  * Redistribution and use in source and binary forms, with or without
49  * modification, are permitted provided that the following conditions
50  * are met:
51  * 1. Redistributions of source code must retain the above copyright
52  *    notice, this list of conditions and the following disclaimer.
53  * 2. Redistributions in binary form must reproduce the above copyright
54  *    notice, this list of conditions and the following disclaimer in the
55  *    documentation and/or other materials provided with the distribution.
56  * 3. All advertising materials mentioning features or use of this software
57  *    must display the following acknowledgement:
58  *	This product includes software developed by the University of
59  *	California, Berkeley and its contributors.
60  * 4. Neither the name of the University nor the names of its contributors
61  *    may be used to endorse or promote products derived from this software
62  *    without specific prior written permission.
63  *
64  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
65  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
66  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
67  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
68  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
69  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
70  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
71  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
72  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
73  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
74  * SUCH DAMAGE.
75  *
76  * from: Utah $Hdr: cd.c 1.6 90/11/28$
77  *
78  *	@(#)cd.c	8.2 (Berkeley) 11/16/93
79  */
80 
81 /*
82  * "Concatenated" disk driver.
83  *
84  * Dynamic configuration and disklabel support by:
85  *	Jason R. Thorpe <thorpej@nas.nasa.gov>
86  *	Numerical Aerodynamic Simulation Facility
87  *	Mail Stop 258-6
88  *	NASA Ames Research Center
89  *	Moffett Field, CA 94035
90  */
91 
92 #include <sys/cdefs.h>
93 __KERNEL_RCSID(0, "$NetBSD: ccd.c,v 1.85 2003/05/02 08:45:26 dsl Exp $");
94 
95 #include <sys/param.h>
96 #include <sys/systm.h>
97 #include <sys/proc.h>
98 #include <sys/errno.h>
99 #include <sys/buf.h>
100 #include <sys/malloc.h>
101 #include <sys/pool.h>
102 #include <sys/namei.h>
103 #include <sys/stat.h>
104 #include <sys/ioctl.h>
105 #include <sys/disklabel.h>
106 #include <sys/device.h>
107 #include <sys/disk.h>
108 #include <sys/syslog.h>
109 #include <sys/fcntl.h>
110 #include <sys/vnode.h>
111 #include <sys/conf.h>
112 #include <sys/lock.h>
113 #include <sys/queue.h>
114 
115 #include <dev/ccdvar.h>
116 
117 #if defined(CCDDEBUG) && !defined(DEBUG)
118 #define DEBUG
119 #endif
120 
121 #ifdef DEBUG
122 #define CCDB_FOLLOW	0x01
123 #define CCDB_INIT	0x02
124 #define CCDB_IO		0x04
125 #define CCDB_LABEL	0x08
126 #define CCDB_VNODE	0x10
127 int ccddebug = 0x00;
128 #endif
129 
130 #define	ccdunit(x)	DISKUNIT(x)
131 
132 struct ccdbuf {
133 	struct buf	cb_buf;		/* new I/O buf */
134 	struct buf	*cb_obp;	/* ptr. to original I/O buf */
135 	struct ccd_softc *cb_sc;	/* pointer to ccd softc */
136 	int		cb_comp;	/* target component */
137 	SIMPLEQ_ENTRY(ccdbuf) cb_q;	/* fifo of component buffers */
138 };
139 
140 /* component buffer pool */
141 struct pool ccd_cbufpool;
142 
143 #define	CCD_GETBUF()		pool_get(&ccd_cbufpool, PR_NOWAIT)
144 #define	CCD_PUTBUF(cbp)		pool_put(&ccd_cbufpool, cbp)
145 
146 #define CCDLABELDEV(dev)	\
147 	(MAKEDISKDEV(major((dev)), ccdunit((dev)), RAW_PART))
148 
149 /* called by main() at boot time */
150 void	ccdattach __P((int));
151 
152 /* called by biodone() at interrupt time */
153 void	ccdiodone __P((struct buf *));
154 
155 static	void ccdstart __P((struct ccd_softc *, struct buf *));
156 static	void ccdinterleave __P((struct ccd_softc *));
157 static	void ccdintr __P((struct ccd_softc *, struct buf *));
158 static	int ccdinit __P((struct ccd_softc *, char **, struct vnode **,
159 	    struct proc *));
160 static	int ccdlookup __P((char *, struct proc *p, struct vnode **));
161 static	struct ccdbuf *ccdbuffer __P((struct ccd_softc *, struct buf *,
162 		daddr_t, caddr_t, long));
163 static	void ccdgetdefaultlabel __P((struct ccd_softc *, struct disklabel *));
164 static	void ccdgetdisklabel __P((dev_t));
165 static	void ccdmakedisklabel __P((struct ccd_softc *));
166 
167 dev_type_open(ccdopen);
168 dev_type_close(ccdclose);
169 dev_type_read(ccdread);
170 dev_type_write(ccdwrite);
171 dev_type_ioctl(ccdioctl);
172 dev_type_strategy(ccdstrategy);
173 dev_type_dump(ccddump);
174 dev_type_size(ccdsize);
175 
176 const struct bdevsw ccd_bdevsw = {
177 	ccdopen, ccdclose, ccdstrategy, ccdioctl, ccddump, ccdsize, D_DISK
178 };
179 
180 const struct cdevsw ccd_cdevsw = {
181 	ccdopen, ccdclose, ccdread, ccdwrite, ccdioctl,
182 	nostop, notty, nopoll, nommap, nokqfilter, D_DISK
183 };
184 
185 #ifdef DEBUG
186 static	void printiinfo __P((struct ccdiinfo *));
187 #endif
188 
189 /* Non-private for the benefit of libkvm. */
190 struct	ccd_softc *ccd_softc;
191 int	numccd = 0;
192 
193 /*
194  * Called by main() during pseudo-device attachment.  All we need
195  * to do is allocate enough space for devices to be configured later.
196  */
197 void
198 ccdattach(num)
199 	int num;
200 {
201 	struct ccd_softc *cs;
202 	int i;
203 
204 	if (num <= 0) {
205 #ifdef DIAGNOSTIC
206 		panic("ccdattach: count <= 0");
207 #endif
208 		return;
209 	}
210 
211 	ccd_softc = (struct ccd_softc *)malloc(num * sizeof(struct ccd_softc),
212 	    M_DEVBUF, M_NOWAIT|M_ZERO);
213 	if (ccd_softc == NULL) {
214 		printf("WARNING: no memory for concatenated disks\n");
215 		return;
216 	}
217 	numccd = num;
218 
219 	/* Initialize the component buffer pool. */
220 	pool_init(&ccd_cbufpool, sizeof(struct ccdbuf), 0,
221 	    0, 0, "ccdpl", NULL);
222 
223 	/* Initialize per-softc structures. */
224 	for (i = 0; i < num; i++) {
225 		cs = &ccd_softc[i];
226 		sprintf(cs->sc_xname, "ccd%d", i);	/* XXX */
227 		cs->sc_dkdev.dk_name = cs->sc_xname;	/* XXX */
228 		lockinit(&cs->sc_lock, PRIBIO, "ccdlk", 0, 0);
229 	}
230 }
231 
232 static int
233 ccdinit(cs, cpaths, vpp, p)
234 	struct ccd_softc *cs;
235 	char **cpaths;
236 	struct vnode **vpp;
237 	struct proc *p;
238 {
239 	struct ccdcinfo *ci = NULL;
240 	size_t size;
241 	int ix;
242 	struct vattr va;
243 	size_t minsize;
244 	int maxsecsize;
245 	struct partinfo dpart;
246 	struct ccdgeom *ccg = &cs->sc_geom;
247 	char tmppath[MAXPATHLEN];
248 	int error, path_alloced;
249 
250 #ifdef DEBUG
251 	if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
252 		printf("%s: ccdinit\n", cs->sc_xname);
253 #endif
254 
255 	/* Allocate space for the component info. */
256 	cs->sc_cinfo = malloc(cs->sc_nccdisks * sizeof(struct ccdcinfo),
257 	    M_DEVBUF, M_WAITOK);
258 
259 	cs->sc_size = 0;
260 
261 	/*
262 	 * Verify that each component piece exists and record
263 	 * relevant information about it.
264 	 */
265 	maxsecsize = 0;
266 	minsize = 0;
267 	for (ix = 0, path_alloced = 0; ix < cs->sc_nccdisks; ix++) {
268 		ci = &cs->sc_cinfo[ix];
269 		ci->ci_vp = vpp[ix];
270 
271 		/*
272 		 * Copy in the pathname of the component.
273 		 */
274 		memset(tmppath, 0, sizeof(tmppath));	/* sanity */
275 		error = copyinstr(cpaths[ix], tmppath,
276 		    MAXPATHLEN, &ci->ci_pathlen);
277 		if (error) {
278 #ifdef DEBUG
279 			if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
280 				printf("%s: can't copy path, error = %d\n",
281 				    cs->sc_xname, error);
282 #endif
283 			goto out;
284 		}
285 		ci->ci_path = malloc(ci->ci_pathlen, M_DEVBUF, M_WAITOK);
286 		memcpy(ci->ci_path, tmppath, ci->ci_pathlen);
287 		path_alloced++;
288 
289 		/*
290 		 * XXX: Cache the component's dev_t.
291 		 */
292 		if ((error = VOP_GETATTR(vpp[ix], &va, p->p_ucred, p)) != 0) {
293 #ifdef DEBUG
294 			if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
295 				printf("%s: %s: getattr failed %s = %d\n",
296 				    cs->sc_xname, ci->ci_path,
297 				    "error", error);
298 #endif
299 			goto out;
300 		}
301 		ci->ci_dev = va.va_rdev;
302 
303 		/*
304 		 * Get partition information for the component.
305 		 */
306 		error = VOP_IOCTL(vpp[ix], DIOCGPART, &dpart,
307 		    FREAD, p->p_ucred, p);
308 		if (error) {
309 #ifdef DEBUG
310 			if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
311 				 printf("%s: %s: ioctl failed, error = %d\n",
312 				     cs->sc_xname, ci->ci_path, error);
313 #endif
314 			goto out;
315 		}
316 
317 /*
318  * This diagnostic test is disabled (for now?) since not all port supports
319  * on-disk BSD disklabel.
320  */
321 #if 0 /* def DIAGNOSTIC */
322 		/* Check fstype field of component. */
323 		if (dpart.part->p_fstype != FS_CCD)
324 			printf("%s: WARNING: %s: fstype %d != FS_CCD\n",
325 			    cs->sc_xname, ci->ci_path, dpart.part->p_fstype);
326 #endif
327 
328 		/*
329 		 * Calculate the size, truncating to an interleave
330 		 * boundary if necessary.
331 		 */
332 		maxsecsize =
333 		    ((dpart.disklab->d_secsize > maxsecsize) ?
334 		    dpart.disklab->d_secsize : maxsecsize);
335 		size = dpart.part->p_size;
336 		if (cs->sc_ileave > 1)
337 			size -= size % cs->sc_ileave;
338 
339 		if (size == 0) {
340 #ifdef DEBUG
341 			if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
342 				printf("%s: %s: size == 0\n",
343 				    cs->sc_xname, ci->ci_path);
344 #endif
345 			error = ENODEV;
346 			goto out;
347 		}
348 
349 		if (minsize == 0 || size < minsize)
350 			minsize = size;
351 		ci->ci_size = size;
352 		cs->sc_size += size;
353 	}
354 
355 	/*
356 	 * Don't allow the interleave to be smaller than
357 	 * the biggest component sector.
358 	 */
359 	if ((cs->sc_ileave > 0) &&
360 	    (cs->sc_ileave < (maxsecsize / DEV_BSIZE))) {
361 #ifdef DEBUG
362 		if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
363 			printf("%s: interleave must be at least %d\n",
364 			    cs->sc_xname, (maxsecsize / DEV_BSIZE));
365 #endif
366 		error = EINVAL;
367 		goto out;
368 	}
369 
370 	/*
371 	 * If uniform interleave is desired set all sizes to that of
372 	 * the smallest component.
373 	 */
374 	if (cs->sc_flags & CCDF_UNIFORM) {
375 		for (ci = cs->sc_cinfo;
376 		     ci < &cs->sc_cinfo[cs->sc_nccdisks]; ci++)
377 			ci->ci_size = minsize;
378 
379 		cs->sc_size = cs->sc_nccdisks * minsize;
380 	}
381 
382 	/*
383 	 * Construct the interleave table.
384 	 */
385 	ccdinterleave(cs);
386 
387 	/*
388 	 * Create pseudo-geometry based on 1MB cylinders.  It's
389 	 * pretty close.
390 	 */
391 	ccg->ccg_secsize = DEV_BSIZE;
392 	ccg->ccg_ntracks = 1;
393 	ccg->ccg_nsectors = 1024 * (1024 / ccg->ccg_secsize);
394 	ccg->ccg_ncylinders = cs->sc_size / ccg->ccg_nsectors;
395 
396 	cs->sc_flags |= CCDF_INITED;
397 
398 	return (0);
399 
400  out:
401 	for (ix = 0; ix < path_alloced; ix++)
402 		free(cs->sc_cinfo[ix].ci_path, M_DEVBUF);
403 	free(cs->sc_cinfo, M_DEVBUF);
404 	return (error);
405 }
406 
407 static void
408 ccdinterleave(cs)
409 	struct ccd_softc *cs;
410 {
411 	struct ccdcinfo *ci, *smallci;
412 	struct ccdiinfo *ii;
413 	daddr_t bn, lbn;
414 	int ix;
415 	u_long size;
416 
417 #ifdef DEBUG
418 	if (ccddebug & CCDB_INIT)
419 		printf("ccdinterleave(%p): ileave %d\n", cs, cs->sc_ileave);
420 #endif
421 	/*
422 	 * Allocate an interleave table.
423 	 * Chances are this is too big, but we don't care.
424 	 */
425 	size = (cs->sc_nccdisks + 1) * sizeof(struct ccdiinfo);
426 	cs->sc_itable = (struct ccdiinfo *)malloc(size, M_DEVBUF,
427 	    M_WAITOK|M_ZERO);
428 
429 	/*
430 	 * Trivial case: no interleave (actually interleave of disk size).
431 	 * Each table entry represents a single component in its entirety.
432 	 */
433 	if (cs->sc_ileave == 0) {
434 		bn = 0;
435 		ii = cs->sc_itable;
436 
437 		for (ix = 0; ix < cs->sc_nccdisks; ix++) {
438 			/* Allocate space for ii_index. */
439 			ii->ii_index = malloc(sizeof(int), M_DEVBUF, M_WAITOK);
440 			ii->ii_ndisk = 1;
441 			ii->ii_startblk = bn;
442 			ii->ii_startoff = 0;
443 			ii->ii_index[0] = ix;
444 			bn += cs->sc_cinfo[ix].ci_size;
445 			ii++;
446 		}
447 		ii->ii_ndisk = 0;
448 #ifdef DEBUG
449 		if (ccddebug & CCDB_INIT)
450 			printiinfo(cs->sc_itable);
451 #endif
452 		return;
453 	}
454 
455 	/*
456 	 * The following isn't fast or pretty; it doesn't have to be.
457 	 */
458 	size = 0;
459 	bn = lbn = 0;
460 	for (ii = cs->sc_itable; ; ii++) {
461 		/* Allocate space for ii_index. */
462 		ii->ii_index = malloc((sizeof(int) * cs->sc_nccdisks),
463 		    M_DEVBUF, M_WAITOK);
464 
465 		/*
466 		 * Locate the smallest of the remaining components
467 		 */
468 		smallci = NULL;
469 		for (ci = cs->sc_cinfo;
470 		     ci < &cs->sc_cinfo[cs->sc_nccdisks]; ci++)
471 			if (ci->ci_size > size &&
472 			    (smallci == NULL ||
473 			     ci->ci_size < smallci->ci_size))
474 				smallci = ci;
475 
476 		/*
477 		 * Nobody left, all done
478 		 */
479 		if (smallci == NULL) {
480 			ii->ii_ndisk = 0;
481 			break;
482 		}
483 
484 		/*
485 		 * Record starting logical block and component offset
486 		 */
487 		ii->ii_startblk = bn / cs->sc_ileave;
488 		ii->ii_startoff = lbn;
489 
490 		/*
491 		 * Determine how many disks take part in this interleave
492 		 * and record their indices.
493 		 */
494 		ix = 0;
495 		for (ci = cs->sc_cinfo;
496 		     ci < &cs->sc_cinfo[cs->sc_nccdisks]; ci++)
497 			if (ci->ci_size >= smallci->ci_size)
498 				ii->ii_index[ix++] = ci - cs->sc_cinfo;
499 		ii->ii_ndisk = ix;
500 		bn += ix * (smallci->ci_size - size);
501 		lbn = smallci->ci_size / cs->sc_ileave;
502 		size = smallci->ci_size;
503 	}
504 #ifdef DEBUG
505 	if (ccddebug & CCDB_INIT)
506 		printiinfo(cs->sc_itable);
507 #endif
508 }
509 
510 /* ARGSUSED */
511 int
512 ccdopen(dev, flags, fmt, p)
513 	dev_t dev;
514 	int flags, fmt;
515 	struct proc *p;
516 {
517 	int unit = ccdunit(dev);
518 	struct ccd_softc *cs;
519 	struct disklabel *lp;
520 	int error = 0, part, pmask;
521 
522 #ifdef DEBUG
523 	if (ccddebug & CCDB_FOLLOW)
524 		printf("ccdopen(0x%x, 0x%x)\n", dev, flags);
525 #endif
526 	if (unit >= numccd)
527 		return (ENXIO);
528 	cs = &ccd_softc[unit];
529 
530 	if ((error = lockmgr(&cs->sc_lock, LK_EXCLUSIVE, NULL)) != 0)
531 		return (error);
532 
533 	lp = cs->sc_dkdev.dk_label;
534 
535 	part = DISKPART(dev);
536 	pmask = (1 << part);
537 
538 	/*
539 	 * If we're initialized, check to see if there are any other
540 	 * open partitions.  If not, then it's safe to update
541 	 * the in-core disklabel.
542 	 */
543 	if ((cs->sc_flags & CCDF_INITED) && (cs->sc_dkdev.dk_openmask == 0))
544 		ccdgetdisklabel(dev);
545 
546 	/* Check that the partition exists. */
547 	if (part != RAW_PART) {
548 		if (((cs->sc_flags & CCDF_INITED) == 0) ||
549 		    ((part >= lp->d_npartitions) ||
550 		     (lp->d_partitions[part].p_fstype == FS_UNUSED))) {
551 			error = ENXIO;
552 			goto done;
553 		}
554 	}
555 
556 	/* Prevent our unit from being unconfigured while open. */
557 	switch (fmt) {
558 	case S_IFCHR:
559 		cs->sc_dkdev.dk_copenmask |= pmask;
560 		break;
561 
562 	case S_IFBLK:
563 		cs->sc_dkdev.dk_bopenmask |= pmask;
564 		break;
565 	}
566 	cs->sc_dkdev.dk_openmask =
567 	    cs->sc_dkdev.dk_copenmask | cs->sc_dkdev.dk_bopenmask;
568 
569  done:
570 	(void) lockmgr(&cs->sc_lock, LK_RELEASE, NULL);
571 	return (error);
572 }
573 
574 /* ARGSUSED */
575 int
576 ccdclose(dev, flags, fmt, p)
577 	dev_t dev;
578 	int flags, fmt;
579 	struct proc *p;
580 {
581 	int unit = ccdunit(dev);
582 	struct ccd_softc *cs;
583 	int error = 0, part;
584 
585 #ifdef DEBUG
586 	if (ccddebug & CCDB_FOLLOW)
587 		printf("ccdclose(0x%x, 0x%x)\n", dev, flags);
588 #endif
589 
590 	if (unit >= numccd)
591 		return (ENXIO);
592 	cs = &ccd_softc[unit];
593 
594 	if ((error = lockmgr(&cs->sc_lock, LK_EXCLUSIVE, NULL)) != 0)
595 		return (error);
596 
597 	part = DISKPART(dev);
598 
599 	/* ...that much closer to allowing unconfiguration... */
600 	switch (fmt) {
601 	case S_IFCHR:
602 		cs->sc_dkdev.dk_copenmask &= ~(1 << part);
603 		break;
604 
605 	case S_IFBLK:
606 		cs->sc_dkdev.dk_bopenmask &= ~(1 << part);
607 		break;
608 	}
609 	cs->sc_dkdev.dk_openmask =
610 	    cs->sc_dkdev.dk_copenmask | cs->sc_dkdev.dk_bopenmask;
611 
612 	(void) lockmgr(&cs->sc_lock, LK_RELEASE, NULL);
613 	return (0);
614 }
615 
616 void
617 ccdstrategy(bp)
618 	struct buf *bp;
619 {
620 	int unit = ccdunit(bp->b_dev);
621 	struct ccd_softc *cs = &ccd_softc[unit];
622 	int s;
623 	int wlabel;
624 	struct disklabel *lp;
625 
626 #ifdef DEBUG
627 	if (ccddebug & CCDB_FOLLOW)
628 		printf("ccdstrategy(%p): unit %d\n", bp, unit);
629 #endif
630 	if ((cs->sc_flags & CCDF_INITED) == 0) {
631 #ifdef DEBUG
632 		if (ccddebug & CCDB_FOLLOW)
633 			printf("ccdstrategy: unit %d: not inited\n", unit);
634 #endif
635 		bp->b_error = ENXIO;
636 		bp->b_flags |= B_ERROR;
637 		goto done;
638 	}
639 
640 	/* If it's a nil transfer, wake up the top half now. */
641 	if (bp->b_bcount == 0)
642 		goto done;
643 
644 	lp = cs->sc_dkdev.dk_label;
645 
646 	/*
647 	 * Do bounds checking and adjust transfer.  If there's an
648 	 * error, the bounds check will flag that for us.
649 	 */
650 	wlabel = cs->sc_flags & (CCDF_WLABEL|CCDF_LABELLING);
651 	if (DISKPART(bp->b_dev) != RAW_PART)
652 		if (bounds_check_with_label(bp, lp, wlabel) <= 0)
653 			goto done;
654 
655 	bp->b_resid = bp->b_bcount;
656 
657 	/*
658 	 * "Start" the unit.
659 	 */
660 	s = splbio();
661 	ccdstart(cs, bp);
662 	splx(s);
663 	return;
664 done:
665 	biodone(bp);
666 }
667 
668 static void
669 ccdstart(cs, bp)
670 	struct ccd_softc *cs;
671 	struct buf *bp;
672 {
673 	long bcount, rcount;
674 	struct ccdbuf *cbp;
675 	caddr_t addr;
676 	daddr_t bn;
677 	struct partition *pp;
678 	SIMPLEQ_HEAD(, ccdbuf) cbufq;
679 
680 #ifdef DEBUG
681 	if (ccddebug & CCDB_FOLLOW)
682 		printf("ccdstart(%p, %p)\n", cs, bp);
683 #endif
684 
685 	/* Instrumentation. */
686 	disk_busy(&cs->sc_dkdev);
687 
688 	/*
689 	 * Translate the partition-relative block number to an absolute.
690 	 */
691 	bn = bp->b_blkno;
692 	if (DISKPART(bp->b_dev) != RAW_PART) {
693 		pp = &cs->sc_dkdev.dk_label->d_partitions[DISKPART(bp->b_dev)];
694 		bn += pp->p_offset;
695 	}
696 
697 	/*
698 	 * Allocate the component buffers.
699 	 */
700 	SIMPLEQ_INIT(&cbufq);
701 	addr = bp->b_data;
702 	for (bcount = bp->b_bcount; bcount > 0; bcount -= rcount) {
703 		cbp = ccdbuffer(cs, bp, bn, addr, bcount);
704 		if (cbp == NULL) {
705 			/* Free the already allocated component buffers. */
706 			while ((cbp = SIMPLEQ_FIRST(&cbufq)) != NULL) {
707 				SIMPLEQ_REMOVE_HEAD(&cbufq, cb_q);
708 				CCD_PUTBUF(cbp);
709 			}
710 
711 			/* Notify the upper layer we are out of memory. */
712 			bp->b_error = ENOMEM;
713 			bp->b_flags |= B_ERROR;
714 			biodone(bp);
715 			disk_unbusy(&cs->sc_dkdev, 0, 0);
716 			return;
717 		}
718 		SIMPLEQ_INSERT_TAIL(&cbufq, cbp, cb_q);
719 		rcount = cbp->cb_buf.b_bcount;
720 		bn += btodb(rcount);
721 		addr += rcount;
722 	}
723 
724 	/* Now fire off the requests. */
725 	while ((cbp = SIMPLEQ_FIRST(&cbufq)) != NULL) {
726 		SIMPLEQ_REMOVE_HEAD(&cbufq, cb_q);
727 		if ((cbp->cb_buf.b_flags & B_READ) == 0)
728 			cbp->cb_buf.b_vp->v_numoutput++;
729 		VOP_STRATEGY(&cbp->cb_buf);
730 	}
731 }
732 
733 /*
734  * Build a component buffer header.
735  */
736 static struct ccdbuf *
737 ccdbuffer(cs, bp, bn, addr, bcount)
738 	struct ccd_softc *cs;
739 	struct buf *bp;
740 	daddr_t bn;
741 	caddr_t addr;
742 	long bcount;
743 {
744 	struct ccdcinfo *ci;
745 	struct ccdbuf *cbp;
746 	daddr_t cbn, cboff;
747 	u_int64_t cbc;
748 	int ccdisk;
749 
750 #ifdef DEBUG
751 	if (ccddebug & CCDB_IO)
752 		printf("ccdbuffer(%p, %p, %" PRId64 ", %p, %ld)\n",
753 		       cs, bp, bn, addr, bcount);
754 #endif
755 	/*
756 	 * Determine which component bn falls in.
757 	 */
758 	cbn = bn;
759 	cboff = 0;
760 
761 	/*
762 	 * Serially concatenated
763 	 */
764 	if (cs->sc_ileave == 0) {
765 		daddr_t sblk;
766 
767 		sblk = 0;
768 		for (ccdisk = 0, ci = &cs->sc_cinfo[ccdisk];
769 		    cbn >= sblk + ci->ci_size;
770 		    ccdisk++, ci = &cs->sc_cinfo[ccdisk])
771 			sblk += ci->ci_size;
772 		cbn -= sblk;
773 	}
774 	/*
775 	 * Interleaved
776 	 */
777 	else {
778 		struct ccdiinfo *ii;
779 		int off;
780 
781 		cboff = cbn % cs->sc_ileave;
782 		cbn /= cs->sc_ileave;
783 		for (ii = cs->sc_itable; ii->ii_ndisk; ii++)
784 			if (ii->ii_startblk > cbn)
785 				break;
786 		ii--;
787 		off = cbn - ii->ii_startblk;
788 		if (ii->ii_ndisk == 1) {
789 			ccdisk = ii->ii_index[0];
790 			cbn = ii->ii_startoff + off;
791 		} else {
792 			ccdisk = ii->ii_index[off % ii->ii_ndisk];
793 			cbn = ii->ii_startoff + off / ii->ii_ndisk;
794 		}
795 		cbn *= cs->sc_ileave;
796 		ci = &cs->sc_cinfo[ccdisk];
797 	}
798 
799 	/*
800 	 * Fill in the component buf structure.
801 	 */
802 	cbp = CCD_GETBUF();
803 	if (cbp == NULL)
804 		return (NULL);
805 	BUF_INIT(&cbp->cb_buf);
806 	cbp->cb_buf.b_flags = bp->b_flags | B_CALL;
807 	cbp->cb_buf.b_iodone = ccdiodone;
808 	cbp->cb_buf.b_proc = bp->b_proc;
809 	cbp->cb_buf.b_dev = ci->ci_dev;		/* XXX */
810 	cbp->cb_buf.b_blkno = cbn + cboff;
811 	cbp->cb_buf.b_data = addr;
812 	cbp->cb_buf.b_vp = ci->ci_vp;
813 	if (cs->sc_ileave == 0)
814 		cbc = dbtob((u_int64_t)(ci->ci_size - cbn));
815 	else
816 		cbc = dbtob((u_int64_t)(cs->sc_ileave - cboff));
817 	cbp->cb_buf.b_bcount = cbc < bcount ? cbc : bcount;
818 
819 	/*
820 	 * context for ccdiodone
821 	 */
822 	cbp->cb_obp = bp;
823 	cbp->cb_sc = cs;
824 	cbp->cb_comp = ccdisk;
825 
826 #ifdef DEBUG
827 	if (ccddebug & CCDB_IO)
828 		printf(" dev 0x%x(u%lu): cbp %p bn %" PRId64 " addr %p"
829 		       " bcnt %ld\n",
830 		    ci->ci_dev, (unsigned long) (ci-cs->sc_cinfo), cbp,
831 		    cbp->cb_buf.b_blkno, cbp->cb_buf.b_data,
832 		    cbp->cb_buf.b_bcount);
833 #endif
834 
835 	return (cbp);
836 }
837 
838 static void
839 ccdintr(cs, bp)
840 	struct ccd_softc *cs;
841 	struct buf *bp;
842 {
843 
844 #ifdef DEBUG
845 	if (ccddebug & CCDB_FOLLOW)
846 		printf("ccdintr(%p, %p)\n", cs, bp);
847 #endif
848 	/*
849 	 * Request is done for better or worse, wakeup the top half.
850 	 */
851 	if (bp->b_flags & B_ERROR)
852 		bp->b_resid = bp->b_bcount;
853 	disk_unbusy(&cs->sc_dkdev, (bp->b_bcount - bp->b_resid),
854 	    (bp->b_flags & B_READ));
855 	biodone(bp);
856 }
857 
858 /*
859  * Called at interrupt time.
860  * Mark the component as done and if all components are done,
861  * take a ccd interrupt.
862  */
863 void
864 ccdiodone(vbp)
865 	struct buf *vbp;
866 {
867 	struct ccdbuf *cbp = (struct ccdbuf *) vbp;
868 	struct buf *bp = cbp->cb_obp;
869 	struct ccd_softc *cs = cbp->cb_sc;
870 	int count, s;
871 
872 	s = splbio();
873 #ifdef DEBUG
874 	if (ccddebug & CCDB_FOLLOW)
875 		printf("ccdiodone(%p)\n", cbp);
876 	if (ccddebug & CCDB_IO) {
877 		printf("ccdiodone: bp %p bcount %ld resid %ld\n",
878 		       bp, bp->b_bcount, bp->b_resid);
879 		printf(" dev 0x%x(u%d), cbp %p bn %" PRId64 " addr %p"
880 		       " bcnt %ld\n",
881 		       cbp->cb_buf.b_dev, cbp->cb_comp, cbp,
882 		       cbp->cb_buf.b_blkno, cbp->cb_buf.b_data,
883 		       cbp->cb_buf.b_bcount);
884 	}
885 #endif
886 
887 	if (cbp->cb_buf.b_flags & B_ERROR) {
888 		bp->b_flags |= B_ERROR;
889 		bp->b_error = cbp->cb_buf.b_error ?
890 		    cbp->cb_buf.b_error : EIO;
891 
892 		printf("%s: error %d on component %d\n",
893 		       cs->sc_xname, bp->b_error, cbp->cb_comp);
894 	}
895 	count = cbp->cb_buf.b_bcount;
896 	CCD_PUTBUF(cbp);
897 
898 	/*
899 	 * If all done, "interrupt".
900 	 */
901 	bp->b_resid -= count;
902 	if (bp->b_resid < 0)
903 		panic("ccdiodone: count");
904 	if (bp->b_resid == 0)
905 		ccdintr(cs, bp);
906 	splx(s);
907 }
908 
909 /* ARGSUSED */
910 int
911 ccdread(dev, uio, flags)
912 	dev_t dev;
913 	struct uio *uio;
914 	int flags;
915 {
916 	int unit = ccdunit(dev);
917 	struct ccd_softc *cs;
918 
919 #ifdef DEBUG
920 	if (ccddebug & CCDB_FOLLOW)
921 		printf("ccdread(0x%x, %p)\n", dev, uio);
922 #endif
923 	if (unit >= numccd)
924 		return (ENXIO);
925 	cs = &ccd_softc[unit];
926 
927 	if ((cs->sc_flags & CCDF_INITED) == 0)
928 		return (ENXIO);
929 
930 	/*
931 	 * XXX: It's not clear that using minphys() is completely safe,
932 	 * in particular, for raw I/O.  Underlying devices might have some
933 	 * non-obvious limits, because of the copy to user-space.
934 	 */
935 	return (physio(ccdstrategy, NULL, dev, B_READ, minphys, uio));
936 }
937 
938 /* ARGSUSED */
939 int
940 ccdwrite(dev, uio, flags)
941 	dev_t dev;
942 	struct uio *uio;
943 	int flags;
944 {
945 	int unit = ccdunit(dev);
946 	struct ccd_softc *cs;
947 
948 #ifdef DEBUG
949 	if (ccddebug & CCDB_FOLLOW)
950 		printf("ccdwrite(0x%x, %p)\n", dev, uio);
951 #endif
952 	if (unit >= numccd)
953 		return (ENXIO);
954 	cs = &ccd_softc[unit];
955 
956 	if ((cs->sc_flags & CCDF_INITED) == 0)
957 		return (ENXIO);
958 
959 	/*
960 	 * XXX: It's not clear that using minphys() is completely safe,
961 	 * in particular, for raw I/O.  Underlying devices might have some
962 	 * non-obvious limits, because of the copy to user-space.
963 	 */
964 	return (physio(ccdstrategy, NULL, dev, B_WRITE, minphys, uio));
965 }
966 
967 int
968 ccdioctl(dev, cmd, data, flag, p)
969 	dev_t dev;
970 	u_long cmd;
971 	caddr_t data;
972 	int flag;
973 	struct proc *p;
974 {
975 	int unit = ccdunit(dev);
976 	int i, j, lookedup = 0, error;
977 	int part, pmask;
978 	struct ccd_softc *cs;
979 	struct ccd_ioctl *ccio = (struct ccd_ioctl *)data;
980 	char **cpp;
981 	struct vnode **vpp;
982 #ifdef __HAVE_OLD_DISKLABEL
983 	struct disklabel newlabel;
984 #endif
985 
986 	if (unit >= numccd)
987 		return (ENXIO);
988 	cs = &ccd_softc[unit];
989 
990 	/* Must be open for writes for these commands... */
991 	switch (cmd) {
992 	case CCDIOCSET:
993 	case CCDIOCCLR:
994 	case DIOCSDINFO:
995 	case DIOCWDINFO:
996 #ifdef __HAVE_OLD_DISKLABEL
997 	case ODIOCSDINFO:
998 	case ODIOCWDINFO:
999 #endif
1000 	case DIOCWLABEL:
1001 		if ((flag & FWRITE) == 0)
1002 			return (EBADF);
1003 	}
1004 
1005 	if ((error = lockmgr(&cs->sc_lock, LK_EXCLUSIVE, NULL)) != 0)
1006 		return (error);
1007 
1008 	/* Must be initialized for these... */
1009 	switch (cmd) {
1010 	case CCDIOCCLR:
1011 	case DIOCGDINFO:
1012 	case DIOCSDINFO:
1013 	case DIOCWDINFO:
1014 	case DIOCGPART:
1015 	case DIOCWLABEL:
1016 	case DIOCGDEFLABEL:
1017 #ifdef __HAVE_OLD_DISKLABEL
1018 	case ODIOCGDINFO:
1019 	case ODIOCSDINFO:
1020 	case ODIOCWDINFO:
1021 	case ODIOCGDEFLABEL:
1022 #endif
1023 		if ((cs->sc_flags & CCDF_INITED) == 0) {
1024 			error = ENXIO;
1025 			goto out;
1026 		}
1027 	}
1028 
1029 	switch (cmd) {
1030 	case CCDIOCSET:
1031 		if (cs->sc_flags & CCDF_INITED) {
1032 			error = EBUSY;
1033 			goto out;
1034 		}
1035 
1036 		/* Validate the flags. */
1037 		if ((ccio->ccio_flags & CCDF_USERMASK) != ccio->ccio_flags) {
1038 			error = EINVAL;
1039 			goto out;
1040 		}
1041 
1042 		if (ccio->ccio_ndisks > CCD_MAXNDISKS) {
1043 			error = EINVAL;
1044 			goto out;
1045 		}
1046 
1047 		/* Fill in some important bits. */
1048 		cs->sc_ileave = ccio->ccio_ileave;
1049 		cs->sc_nccdisks = ccio->ccio_ndisks;
1050 		cs->sc_flags = ccio->ccio_flags & CCDF_USERMASK;
1051 
1052 		/*
1053 		 * Allocate space for and copy in the array of
1054 		 * componet pathnames and device numbers.
1055 		 */
1056 		cpp = malloc(ccio->ccio_ndisks * sizeof(char *),
1057 		    M_DEVBUF, M_WAITOK);
1058 		vpp = malloc(ccio->ccio_ndisks * sizeof(struct vnode *),
1059 		    M_DEVBUF, M_WAITOK);
1060 
1061 		error = copyin(ccio->ccio_disks, cpp,
1062 		    ccio->ccio_ndisks * sizeof(char **));
1063 		if (error) {
1064 			free(vpp, M_DEVBUF);
1065 			free(cpp, M_DEVBUF);
1066 			goto out;
1067 		}
1068 
1069 #ifdef DEBUG
1070 		if (ccddebug & CCDB_INIT)
1071 			for (i = 0; i < ccio->ccio_ndisks; ++i)
1072 				printf("ccdioctl: component %d: 0x%p\n",
1073 				    i, cpp[i]);
1074 #endif
1075 
1076 		for (i = 0; i < ccio->ccio_ndisks; ++i) {
1077 #ifdef DEBUG
1078 			if (ccddebug & CCDB_INIT)
1079 				printf("ccdioctl: lookedup = %d\n", lookedup);
1080 #endif
1081 			if ((error = ccdlookup(cpp[i], p, &vpp[i])) != 0) {
1082 				for (j = 0; j < lookedup; ++j)
1083 					(void)vn_close(vpp[j], FREAD|FWRITE,
1084 					    p->p_ucred, p);
1085 				free(vpp, M_DEVBUF);
1086 				free(cpp, M_DEVBUF);
1087 				goto out;
1088 			}
1089 			++lookedup;
1090 		}
1091 
1092 		/*
1093 		 * Initialize the ccd.  Fills in the softc for us.
1094 		 */
1095 		if ((error = ccdinit(cs, cpp, vpp, p)) != 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 
1104 		/* We can free the temporary variables now. */
1105 		free(vpp, M_DEVBUF);
1106 		free(cpp, M_DEVBUF);
1107 
1108 		/*
1109 		 * The ccd has been successfully initialized, so
1110 		 * we can place it into the array.  Don't try to
1111 		 * read the disklabel until the disk has been attached,
1112 		 * because space for the disklabel is allocated
1113 		 * in disk_attach();
1114 		 */
1115 		ccio->ccio_unit = unit;
1116 		ccio->ccio_size = cs->sc_size;
1117 
1118 		/* Attach the disk. */
1119 		disk_attach(&cs->sc_dkdev);
1120 
1121 		/* Try and read the disklabel. */
1122 		ccdgetdisklabel(dev);
1123 		break;
1124 
1125 	case CCDIOCCLR:
1126 		/*
1127 		 * Don't unconfigure if any other partitions are open
1128 		 * or if both the character and block flavors of this
1129 		 * partition are open.
1130 		 */
1131 		part = DISKPART(dev);
1132 		pmask = (1 << part);
1133 		if ((cs->sc_dkdev.dk_openmask & ~pmask) ||
1134 		    ((cs->sc_dkdev.dk_bopenmask & pmask) &&
1135 		    (cs->sc_dkdev.dk_copenmask & pmask))) {
1136 			error = EBUSY;
1137 			goto out;
1138 		}
1139 
1140 		/*
1141 		 * Free ccd_softc information and clear entry.
1142 		 */
1143 
1144 		/* Close the components and free their pathnames. */
1145 		for (i = 0; i < cs->sc_nccdisks; ++i) {
1146 			/*
1147 			 * XXX: this close could potentially fail and
1148 			 * cause Bad Things.  Maybe we need to force
1149 			 * the close to happen?
1150 			 */
1151 #ifdef DEBUG
1152 			if (ccddebug & CCDB_VNODE)
1153 				vprint("CCDIOCCLR: vnode info",
1154 				    cs->sc_cinfo[i].ci_vp);
1155 #endif
1156 			(void)vn_close(cs->sc_cinfo[i].ci_vp, FREAD|FWRITE,
1157 			    p->p_ucred, p);
1158 			free(cs->sc_cinfo[i].ci_path, M_DEVBUF);
1159 		}
1160 
1161 		/* Free interleave index. */
1162 		for (i = 0; cs->sc_itable[i].ii_ndisk; ++i)
1163 			free(cs->sc_itable[i].ii_index, M_DEVBUF);
1164 
1165 		/* Free component info and interleave table. */
1166 		free(cs->sc_cinfo, M_DEVBUF);
1167 		free(cs->sc_itable, M_DEVBUF);
1168 		cs->sc_flags &= ~CCDF_INITED;
1169 
1170 		/* Detatch the disk. */
1171 		disk_detach(&cs->sc_dkdev);
1172 		break;
1173 
1174 	case DIOCGDINFO:
1175 		*(struct disklabel *)data = *(cs->sc_dkdev.dk_label);
1176 		break;
1177 #ifdef __HAVE_OLD_DISKLABEL
1178 	case ODIOCGDINFO:
1179 		newlabel = *(cs->sc_dkdev.dk_label);
1180 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1181 			return ENOTTY;
1182 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1183 		break;
1184 #endif
1185 
1186 	case DIOCGPART:
1187 		((struct partinfo *)data)->disklab = cs->sc_dkdev.dk_label;
1188 		((struct partinfo *)data)->part =
1189 		    &cs->sc_dkdev.dk_label->d_partitions[DISKPART(dev)];
1190 		break;
1191 
1192 	case DIOCWDINFO:
1193 	case DIOCSDINFO:
1194 #ifdef __HAVE_OLD_DISKLABEL
1195 	case ODIOCWDINFO:
1196 	case ODIOCSDINFO:
1197 #endif
1198 	{
1199 		struct disklabel *lp;
1200 #ifdef __HAVE_OLD_DISKLABEL
1201 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
1202 			memset(&newlabel, 0, sizeof newlabel);
1203 			memcpy(&newlabel, data, sizeof (struct olddisklabel));
1204 			lp = &newlabel;
1205 		} else
1206 #endif
1207 		lp = (struct disklabel *)data;
1208 
1209 		cs->sc_flags |= CCDF_LABELLING;
1210 
1211 		error = setdisklabel(cs->sc_dkdev.dk_label,
1212 		    lp, 0, cs->sc_dkdev.dk_cpulabel);
1213 		if (error == 0) {
1214 			if (cmd == DIOCWDINFO
1215 #ifdef __HAVE_OLD_DISKLABEL
1216 			    || cmd == ODIOCWDINFO
1217 #endif
1218 			   )
1219 				error = writedisklabel(CCDLABELDEV(dev),
1220 				    ccdstrategy, cs->sc_dkdev.dk_label,
1221 				    cs->sc_dkdev.dk_cpulabel);
1222 		}
1223 
1224 		cs->sc_flags &= ~CCDF_LABELLING;
1225 		break;
1226 	}
1227 
1228 	case DIOCWLABEL:
1229 		if (*(int *)data != 0)
1230 			cs->sc_flags |= CCDF_WLABEL;
1231 		else
1232 			cs->sc_flags &= ~CCDF_WLABEL;
1233 		break;
1234 
1235 	case DIOCGDEFLABEL:
1236 		ccdgetdefaultlabel(cs, (struct disklabel *)data);
1237 		break;
1238 
1239 #ifdef __HAVE_OLD_DISKLABEL
1240 	case ODIOCGDEFLABEL:
1241 		ccdgetdefaultlabel(cs, &newlabel);
1242 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1243 			return ENOTTY;
1244 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1245 		break;
1246 #endif
1247 
1248 	default:
1249 		error = ENOTTY;
1250 	}
1251 
1252  out:
1253 	(void) lockmgr(&cs->sc_lock, LK_RELEASE, NULL);
1254 	return (error);
1255 }
1256 
1257 int
1258 ccdsize(dev)
1259 	dev_t dev;
1260 {
1261 	struct ccd_softc *cs;
1262 	struct disklabel *lp;
1263 	int part, unit, omask, size;
1264 
1265 	unit = ccdunit(dev);
1266 	if (unit >= numccd)
1267 		return (-1);
1268 	cs = &ccd_softc[unit];
1269 
1270 	if ((cs->sc_flags & CCDF_INITED) == 0)
1271 		return (-1);
1272 
1273 	part = DISKPART(dev);
1274 	omask = cs->sc_dkdev.dk_openmask & (1 << part);
1275 	lp = cs->sc_dkdev.dk_label;
1276 
1277 	if (omask == 0 && ccdopen(dev, 0, S_IFBLK, curproc))
1278 		return (-1);
1279 
1280 	if (lp->d_partitions[part].p_fstype != FS_SWAP)
1281 		size = -1;
1282 	else
1283 		size = lp->d_partitions[part].p_size *
1284 		    (lp->d_secsize / DEV_BSIZE);
1285 
1286 	if (omask == 0 && ccdclose(dev, 0, S_IFBLK, curproc))
1287 		return (-1);
1288 
1289 	return (size);
1290 }
1291 
1292 int
1293 ccddump(dev, blkno, va, size)
1294 	dev_t dev;
1295 	daddr_t blkno;
1296 	caddr_t va;
1297 	size_t size;
1298 {
1299 
1300 	/* Not implemented. */
1301 	return ENXIO;
1302 }
1303 
1304 /*
1305  * Lookup the provided name in the filesystem.  If the file exists,
1306  * is a valid block device, and isn't being used by anyone else,
1307  * set *vpp to the file's vnode.
1308  */
1309 static int
1310 ccdlookup(path, p, vpp)
1311 	char *path;
1312 	struct proc *p;
1313 	struct vnode **vpp;	/* result */
1314 {
1315 	struct nameidata nd;
1316 	struct vnode *vp;
1317 	struct vattr va;
1318 	int error;
1319 
1320 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, path, p);
1321 	if ((error = vn_open(&nd, FREAD|FWRITE, 0)) != 0) {
1322 #ifdef DEBUG
1323 		if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
1324 			printf("ccdlookup: vn_open error = %d\n", error);
1325 #endif
1326 		return (error);
1327 	}
1328 	vp = nd.ni_vp;
1329 
1330 	if (vp->v_usecount > 1) {
1331 		VOP_UNLOCK(vp, 0);
1332 		(void)vn_close(vp, FREAD|FWRITE, p->p_ucred, p);
1333 		return (EBUSY);
1334 	}
1335 
1336 	if ((error = VOP_GETATTR(vp, &va, p->p_ucred, p)) != 0) {
1337 #ifdef DEBUG
1338 		if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
1339 			printf("ccdlookup: getattr error = %d\n", error);
1340 #endif
1341 		VOP_UNLOCK(vp, 0);
1342 		(void)vn_close(vp, FREAD|FWRITE, p->p_ucred, p);
1343 		return (error);
1344 	}
1345 
1346 	/* XXX: eventually we should handle VREG, too. */
1347 	if (va.va_type != VBLK) {
1348 		VOP_UNLOCK(vp, 0);
1349 		(void)vn_close(vp, FREAD|FWRITE, p->p_ucred, p);
1350 		return (ENOTBLK);
1351 	}
1352 
1353 #ifdef DEBUG
1354 	if (ccddebug & CCDB_VNODE)
1355 		vprint("ccdlookup: vnode info", vp);
1356 #endif
1357 
1358 	VOP_UNLOCK(vp, 0);
1359 	*vpp = vp;
1360 	return (0);
1361 }
1362 
1363 static void
1364 ccdgetdefaultlabel(cs, lp)
1365 	struct ccd_softc *cs;
1366 	struct disklabel *lp;
1367 {
1368 	struct ccdgeom *ccg = &cs->sc_geom;
1369 
1370 	memset(lp, 0, sizeof(*lp));
1371 
1372 	lp->d_secperunit = cs->sc_size;
1373 	lp->d_secsize = ccg->ccg_secsize;
1374 	lp->d_nsectors = ccg->ccg_nsectors;
1375 	lp->d_ntracks = ccg->ccg_ntracks;
1376 	lp->d_ncylinders = ccg->ccg_ncylinders;
1377 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1378 
1379 	strncpy(lp->d_typename, "ccd", sizeof(lp->d_typename));
1380 	lp->d_type = DTYPE_CCD;
1381 	strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
1382 	lp->d_rpm = 3600;
1383 	lp->d_interleave = 1;
1384 	lp->d_flags = 0;
1385 
1386 	lp->d_partitions[RAW_PART].p_offset = 0;
1387 	lp->d_partitions[RAW_PART].p_size = cs->sc_size;
1388 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
1389 	lp->d_npartitions = RAW_PART + 1;
1390 
1391 	lp->d_magic = DISKMAGIC;
1392 	lp->d_magic2 = DISKMAGIC;
1393 	lp->d_checksum = dkcksum(cs->sc_dkdev.dk_label);
1394 }
1395 
1396 /*
1397  * Read the disklabel from the ccd.  If one is not present, fake one
1398  * up.
1399  */
1400 static void
1401 ccdgetdisklabel(dev)
1402 	dev_t dev;
1403 {
1404 	int unit = ccdunit(dev);
1405 	struct ccd_softc *cs = &ccd_softc[unit];
1406 	const char *errstring;
1407 	struct disklabel *lp = cs->sc_dkdev.dk_label;
1408 	struct cpu_disklabel *clp = cs->sc_dkdev.dk_cpulabel;
1409 
1410 	memset(clp, 0, sizeof(*clp));
1411 
1412 	ccdgetdefaultlabel(cs, lp);
1413 
1414 	/*
1415 	 * Call the generic disklabel extraction routine.
1416 	 */
1417 	errstring = readdisklabel(CCDLABELDEV(dev), ccdstrategy,
1418 	    cs->sc_dkdev.dk_label, cs->sc_dkdev.dk_cpulabel);
1419 	if (errstring)
1420 		ccdmakedisklabel(cs);
1421 	else {
1422 		int i;
1423 		struct partition *pp;
1424 
1425 		/*
1426 		 * Sanity check whether the found disklabel is valid.
1427 		 *
1428 		 * This is necessary since total size of ccd may vary
1429 		 * when an interleave is changed even though exactly
1430 		 * same componets are used, and old disklabel may used
1431 		 * if that is found.
1432 		 */
1433 		if (lp->d_secperunit != cs->sc_size)
1434 			printf("WARNING: %s: "
1435 			    "total sector size in disklabel (%d) != "
1436 			    "the size of ccd (%lu)\n", cs->sc_xname,
1437 			    lp->d_secperunit, (u_long)cs->sc_size);
1438 		for (i = 0; i < lp->d_npartitions; i++) {
1439 			pp = &lp->d_partitions[i];
1440 			if (pp->p_offset + pp->p_size > cs->sc_size)
1441 				printf("WARNING: %s: end of partition `%c' "
1442 				    "exceeds the size of ccd (%lu)\n",
1443 				    cs->sc_xname, 'a' + i, (u_long)cs->sc_size);
1444 		}
1445 	}
1446 
1447 #ifdef DEBUG
1448 	/* It's actually extremely common to have unlabeled ccds. */
1449 	if (ccddebug & CCDB_LABEL)
1450 		if (errstring != NULL)
1451 			printf("%s: %s\n", cs->sc_xname, errstring);
1452 #endif
1453 }
1454 
1455 /*
1456  * Take care of things one might want to take care of in the event
1457  * that a disklabel isn't present.
1458  */
1459 static void
1460 ccdmakedisklabel(cs)
1461 	struct ccd_softc *cs;
1462 {
1463 	struct disklabel *lp = cs->sc_dkdev.dk_label;
1464 
1465 	/*
1466 	 * For historical reasons, if there's no disklabel present
1467 	 * the raw partition must be marked FS_BSDFFS.
1468 	 */
1469 	lp->d_partitions[RAW_PART].p_fstype = FS_BSDFFS;
1470 
1471 	strncpy(lp->d_packname, "default label", sizeof(lp->d_packname));
1472 
1473 	lp->d_checksum = dkcksum(lp);
1474 }
1475 
1476 #ifdef DEBUG
1477 static void
1478 printiinfo(ii)
1479 	struct ccdiinfo *ii;
1480 {
1481 	int ix, i;
1482 
1483 	for (ix = 0; ii->ii_ndisk; ix++, ii++) {
1484 		printf(" itab[%d]: #dk %d sblk %" PRId64 " soff %" PRId64,
1485 		    ix, ii->ii_ndisk, ii->ii_startblk, ii->ii_startoff);
1486 		for (i = 0; i < ii->ii_ndisk; i++)
1487 			printf(" %d", ii->ii_index[i]);
1488 		printf("\n");
1489 	}
1490 }
1491 #endif
1492