xref: /netbsd-src/sys/dev/ccd.c (revision fdecd6a253f999ae92b139670d9e15cc9df4497c)
1 /*	$NetBSD: ccd.c,v 1.42 1997/06/26 16:28:34 kleink Exp $	*/
2 
3 /*-
4  * Copyright (c) 1996, 1997 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 REGENTS OR CONTRIBUTORS BE
30  * 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  * Mirroring support based on code written by Satoshi Asami
92  * and Nisha Talagala.
93  */
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/namei.h>
102 #include <sys/stat.h>
103 #include <sys/ioctl.h>
104 #include <sys/disklabel.h>
105 #include <sys/device.h>
106 #include <sys/disk.h>
107 #include <sys/syslog.h>
108 #include <sys/fcntl.h>
109 #include <sys/vnode.h>
110 #include <sys/conf.h>
111 
112 #include <dev/ccdvar.h>
113 
114 #if defined(CCDDEBUG) && !defined(DEBUG)
115 #define DEBUG
116 #endif
117 
118 #ifdef DEBUG
119 #define CCDB_FOLLOW	0x01
120 #define CCDB_INIT	0x02
121 #define CCDB_IO		0x04
122 #define CCDB_LABEL	0x08
123 #define CCDB_VNODE	0x10
124 int ccddebug = 0x00;
125 #endif
126 
127 #define	ccdunit(x)	DISKUNIT(x)
128 
129 struct ccdbuf {
130 	struct buf	cb_buf;		/* new I/O buf */
131 	struct buf	*cb_obp;	/* ptr. to original I/O buf */
132 	int		cb_unit;	/* target unit */
133 	int		cb_comp;	/* target component */
134 	int		cb_flags;	/* misc. flags */
135 	LIST_ENTRY(ccdbuf) cb_list;	/* entry on freelist */
136 };
137 
138 /* cb_flags */
139 #define CBF_MIRROR	0x01		/* we're for a mirror component */
140 
141 /*
142  * Number of freelist buffers per component.  Overridable in kernel
143  * config file and patchable.
144  */
145 #ifndef CCDNBUF
146 #define	CCDNBUF		8
147 #endif
148 int	ccdnbuf = CCDNBUF;
149 
150 /*
151  * XXX Is it OK to wait here?
152  * XXX maybe set up a timeout when we hit some lowater?
153  * XXX    --thorpej
154  */
155 #define	CCDGETBUF(cs, cbp)	do {					\
156 		(cs)->sc_ngetbuf++;					\
157 		if (((cbp) = (cs)->sc_freelist.lh_first) != NULL) {	\
158 			LIST_REMOVE((cbp), cb_list);			\
159 			(cs)->sc_freecount--;				\
160 		} else {						\
161 			(cs)->sc_nmisses++;				\
162 			MALLOC((cbp), struct ccdbuf *,			\
163 			    sizeof(struct ccdbuf), M_DEVBUF, M_WAITOK);	\
164 		}							\
165 	} while (0)
166 
167 #define	CCDPUTBUF(cs, cbp)	do {					\
168 		if ((cs)->sc_freecount == (cs)->sc_hiwat) {		\
169 			FREE((cbp), M_DEVBUF);				\
170 		} else {						\
171 			LIST_INSERT_HEAD(&(cs)->sc_freelist, (cbp), cb_list); \
172 			(cs)->sc_freecount++;				\
173 		}							\
174 	} while (0)
175 
176 #define CCDLABELDEV(dev)	\
177 	(MAKEDISKDEV(major((dev)), ccdunit((dev)), RAW_PART))
178 
179 /* called by main() at boot time */
180 void	ccdattach __P((int));
181 
182 /* called by biodone() at interrupt time */
183 void	ccdiodone __P((struct buf *));
184 int	ccdsize __P((dev_t));
185 
186 static	void ccdstart __P((struct ccd_softc *, struct buf *));
187 static	void ccdinterleave __P((struct ccd_softc *, int));
188 static	void ccdintr __P((struct ccd_softc *, struct buf *));
189 static	int ccdinit __P((struct ccddevice *, char **, struct proc *));
190 static	int ccdlookup __P((char *, struct proc *p, struct vnode **));
191 static	void ccdbuffer __P((struct ccd_softc *, struct buf *,
192 		daddr_t, caddr_t, long, struct ccdbuf **));
193 static	void ccdgetdisklabel __P((dev_t));
194 static	void ccdmakedisklabel __P((struct ccd_softc *));
195 static	int ccdlock __P((struct ccd_softc *));
196 static	void ccdunlock __P((struct ccd_softc *));
197 
198 #ifdef DEBUG
199 static	void printiinfo __P((struct ccdiinfo *));
200 #endif
201 
202 /* Non-private for the benefit of libkvm. */
203 struct	ccd_softc *ccd_softc;
204 struct	ccddevice *ccddevs;
205 int	numccd = 0;
206 
207 /*
208  * Called by main() during pseudo-device attachment.  All we need
209  * to do is allocate enough space for devices to be configured later.
210  */
211 void
212 ccdattach(num)
213 	int num;
214 {
215 	if (num <= 0) {
216 #ifdef DIAGNOSTIC
217 		panic("ccdattach: count <= 0");
218 #endif
219 		return;
220 	}
221 
222 	ccd_softc = (struct ccd_softc *)malloc(num * sizeof(struct ccd_softc),
223 	    M_DEVBUF, M_NOWAIT);
224 	ccddevs = (struct ccddevice *)malloc(num * sizeof(struct ccddevice),
225 	    M_DEVBUF, M_NOWAIT);
226 	if ((ccd_softc == NULL) || (ccddevs == NULL)) {
227 		printf("WARNING: no memory for concatenated disks\n");
228 		if (ccd_softc != NULL)
229 			free(ccd_softc, M_DEVBUF);
230 		if (ccddevs != NULL)
231 			free(ccddevs, M_DEVBUF);
232 		return;
233 	}
234 	numccd = num;
235 	bzero(ccd_softc, num * sizeof(struct ccd_softc));
236 	bzero(ccddevs, num * sizeof(struct ccddevice));
237 }
238 
239 static int
240 ccdinit(ccd, cpaths, p)
241 	struct ccddevice *ccd;
242 	char **cpaths;
243 	struct proc *p;
244 {
245 	register struct ccd_softc *cs = &ccd_softc[ccd->ccd_unit];
246 	register struct ccdcinfo *ci = NULL;
247 	register size_t size;
248 	register int ix;
249 	struct vnode *vp;
250 	struct vattr va;
251 	size_t minsize;
252 	int maxsecsize;
253 	struct partinfo dpart;
254 	struct ccdgeom *ccg = &cs->sc_geom;
255 	char tmppath[MAXPATHLEN];
256 	struct ccdbuf *cbp;
257 	int error;
258 
259 #ifdef DEBUG
260 	if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
261 		printf("ccdinit: unit %d\n", ccd->ccd_unit);
262 #endif
263 
264 	cs->sc_size = 0;
265 	cs->sc_ileave = ccd->ccd_interleave;
266 	cs->sc_nccdisks = ccd->ccd_ndev;
267 	sprintf(cs->sc_xname, "ccd%d", ccd->ccd_unit);	/* XXX */
268 
269 	/* Allocate space for the component info. */
270 	cs->sc_cinfo = malloc(cs->sc_nccdisks * sizeof(struct ccdcinfo),
271 	    M_DEVBUF, M_WAITOK);
272 
273 	/*
274 	 * Verify that each component piece exists and record
275 	 * relevant information about it.
276 	 */
277 	maxsecsize = 0;
278 	minsize = 0;
279 	for (ix = 0; ix < cs->sc_nccdisks; ix++) {
280 		vp = ccd->ccd_vpp[ix];
281 		ci = &cs->sc_cinfo[ix];
282 		ci->ci_vp = vp;
283 
284 		/*
285 		 * Copy in the pathname of the component.
286 		 */
287 		bzero(tmppath, sizeof(tmppath));	/* sanity */
288 		error = copyinstr(cpaths[ix], tmppath,
289 		    MAXPATHLEN, &ci->ci_pathlen);
290 		if (error) {
291 #ifdef DEBUG
292 			if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
293 				printf("%s: can't copy path, error = %d\n",
294 				    cs->sc_xname, error);
295 #endif
296 			free(cs->sc_cinfo, M_DEVBUF);
297 			return (error);
298 		}
299 		ci->ci_path = malloc(ci->ci_pathlen, M_DEVBUF, M_WAITOK);
300 		bcopy(tmppath, ci->ci_path, ci->ci_pathlen);
301 
302 		/*
303 		 * XXX: Cache the component's dev_t.
304 		 */
305 		if ((error = VOP_GETATTR(vp, &va, p->p_ucred, p)) != 0) {
306 #ifdef DEBUG
307 			if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
308 				printf("%s: %s: getattr failed %s = %d\n",
309 				    cs->sc_xname, ci->ci_path,
310 				    "error", error);
311 #endif
312 			free(ci->ci_path, M_DEVBUF);
313 			free(cs->sc_cinfo, M_DEVBUF);
314 			return (error);
315 		}
316 		ci->ci_dev = va.va_rdev;
317 
318 		/*
319 		 * Get partition information for the component.
320 		 */
321 		error = VOP_IOCTL(vp, DIOCGPART, (caddr_t)&dpart,
322 		    FREAD, p->p_ucred, p);
323 		if (error) {
324 #ifdef DEBUG
325 			if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
326 				 printf("%s: %s: ioctl failed, error = %d\n",
327 				     cs->sc_xname, ci->ci_path, error);
328 #endif
329 			free(ci->ci_path, M_DEVBUF);
330 			free(cs->sc_cinfo, M_DEVBUF);
331 			return (error);
332 		}
333 		if (dpart.part->p_fstype == FS_BSDFFS) {
334 			maxsecsize =
335 			    ((dpart.disklab->d_secsize > maxsecsize) ?
336 			    dpart.disklab->d_secsize : maxsecsize);
337 			size = dpart.part->p_size;
338 		} else {
339 #ifdef DEBUG
340 			if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
341 				printf("%s: %s: incorrect partition type\n",
342 				    cs->sc_xname, ci->ci_path);
343 #endif
344 			free(ci->ci_path, M_DEVBUF);
345 			free(cs->sc_cinfo, M_DEVBUF);
346 			return (EFTYPE);
347 		}
348 
349 		/*
350 		 * Calculate the size, truncating to an interleave
351 		 * boundary if necessary.
352 		 */
353 		if (cs->sc_ileave > 1)
354 			size -= size % cs->sc_ileave;
355 
356 		if (size == 0) {
357 #ifdef DEBUG
358 			if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
359 				printf("%s: %s: size == 0\n",
360 				    cs->sc_xname, ci->ci_path);
361 #endif
362 			free(ci->ci_path, M_DEVBUF);
363 			free(cs->sc_cinfo, M_DEVBUF);
364 			return (ENODEV);
365 		}
366 
367 		if (minsize == 0 || size < minsize)
368 			minsize = size;
369 		ci->ci_size = size;
370 		cs->sc_size += size;
371 	}
372 
373 	/*
374 	 * Don't allow the interleave to be smaller than
375 	 * the biggest component sector.
376 	 */
377 	if ((cs->sc_ileave > 0) &&
378 	    (cs->sc_ileave < (maxsecsize / DEV_BSIZE))) {
379 #ifdef DEBUG
380 		if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
381 			printf("%s: interleave must be at least %d\n",
382 			    cs->sc_xname, (maxsecsize / DEV_BSIZE));
383 #endif
384 		free(ci->ci_path, M_DEVBUF);
385 		free(cs->sc_cinfo, M_DEVBUF);
386 		return (EINVAL);
387 	}
388 
389 	/*
390 	 * Mirroring support requires uniform interleave and
391 	 * and even number of components.
392 	 */
393 	if (ccd->ccd_flags & CCDF_MIRROR) {
394 		ccd->ccd_flags |= CCDF_UNIFORM;
395 		if (cs->sc_ileave == 0) {
396 #ifdef DEBUG
397 			if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
398 			printf("%s: mirroring requires interleave\n",
399 			    cs->sc_xname);
400 #endif
401 			free(ci->ci_path, M_DEVBUF);
402 			free(cs->sc_cinfo, M_DEVBUF);
403 			return (EINVAL);
404 		}
405 		if (cs->sc_nccdisks % 2) {
406 #ifdef DEBUG
407 			if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
408 			printf("%s: mirroring requires even # of components\n",
409 			    cs->sc_xname);
410 #endif
411 			free(ci->ci_path, M_DEVBUF);
412 			free(cs->sc_cinfo, M_DEVBUF);
413 			return (EINVAL);
414 		}
415 	}
416 
417 	/*
418 	 * If uniform interleave is desired set all sizes to that of
419 	 * the smallest component.
420 	 */
421 	if (ccd->ccd_flags & CCDF_UNIFORM) {
422 		for (ci = cs->sc_cinfo;
423 		     ci < &cs->sc_cinfo[cs->sc_nccdisks]; ci++)
424 			ci->ci_size = minsize;
425 
426 		if (ccd->ccd_flags & CCDF_MIRROR)
427 			cs->sc_size = (cs->sc_nccdisks / 2) * minsize;
428 		else
429 			cs->sc_size = cs->sc_nccdisks * minsize;
430 	}
431 
432 	/*
433 	 * Construct the interleave table.
434 	 */
435 	ccdinterleave(cs, ccd->ccd_unit);
436 
437 	/*
438 	 * Create pseudo-geometry based on 1MB cylinders.  It's
439 	 * pretty close.
440 	 */
441 	ccg->ccg_secsize = DEV_BSIZE;
442 	ccg->ccg_ntracks = 1;
443 	ccg->ccg_nsectors = 1024 * (1024 / ccg->ccg_secsize);
444 	ccg->ccg_ncylinders = cs->sc_size / ccg->ccg_nsectors;
445 
446 	/*
447 	 * Allocate the component buffer header freelist.  We allocate
448 	 * ccdnbuf buffers per component.
449 	 */
450 	LIST_INIT(&cs->sc_freelist);
451 	cs->sc_hiwat = cs->sc_nccdisks * ccdnbuf;
452 	cs->sc_freecount = cs->sc_hiwat;
453 	for (ix = 0; ix < cs->sc_hiwat; ix++) {
454 		MALLOC(cbp, struct ccdbuf *, sizeof(struct ccdbuf),
455 		    M_DEVBUF, M_WAITOK);
456 		LIST_INSERT_HEAD(&cs->sc_freelist, cbp, cb_list);
457 	}
458 
459 	/* Reset statistics. */
460 	cs->sc_nmisses = 0;
461 	cs->sc_ngetbuf = 0;
462 
463 	cs->sc_flags |= CCDF_INITED;
464 	cs->sc_cflags = ccd->ccd_flags;	/* So we can find out later... */
465 	cs->sc_unit = ccd->ccd_unit;
466 
467 	return (0);
468 }
469 
470 static void
471 ccdinterleave(cs, unit)
472 	register struct ccd_softc *cs;
473 	int unit;
474 {
475 	register struct ccdcinfo *ci, *smallci;
476 	register struct ccdiinfo *ii;
477 	register daddr_t bn, lbn;
478 	register int ix;
479 	u_long size;
480 
481 #ifdef DEBUG
482 	if (ccddebug & CCDB_INIT)
483 		printf("ccdinterleave(%p): ileave %d\n", cs, cs->sc_ileave);
484 #endif
485 	/*
486 	 * Allocate an interleave table.
487 	 * Chances are this is too big, but we don't care.
488 	 */
489 	size = (cs->sc_nccdisks + 1) * sizeof(struct ccdiinfo);
490 	cs->sc_itable = (struct ccdiinfo *)malloc(size, M_DEVBUF, M_WAITOK);
491 	bzero((caddr_t)cs->sc_itable, size);
492 
493 	/*
494 	 * Trivial case: no interleave (actually interleave of disk size).
495 	 * Each table entry represents a single component in its entirety.
496 	 */
497 	if (cs->sc_ileave == 0) {
498 		bn = 0;
499 		ii = cs->sc_itable;
500 
501 		for (ix = 0; ix < cs->sc_nccdisks; ix++) {
502 			/* Allocate space for ii_index. */
503 			ii->ii_index = malloc(sizeof(int), M_DEVBUF, M_WAITOK);
504 			ii->ii_ndisk = 1;
505 			ii->ii_startblk = bn;
506 			ii->ii_startoff = 0;
507 			ii->ii_index[0] = ix;
508 			bn += cs->sc_cinfo[ix].ci_size;
509 			ii++;
510 		}
511 		ii->ii_ndisk = 0;
512 #ifdef DEBUG
513 		if (ccddebug & CCDB_INIT)
514 			printiinfo(cs->sc_itable);
515 #endif
516 		return;
517 	}
518 
519 	/*
520 	 * The following isn't fast or pretty; it doesn't have to be.
521 	 */
522 	size = 0;
523 	bn = lbn = 0;
524 	for (ii = cs->sc_itable; ; ii++) {
525 		/* Allocate space for ii_index. */
526 		ii->ii_index = malloc((sizeof(int) * cs->sc_nccdisks),
527 		    M_DEVBUF, M_WAITOK);
528 
529 		/*
530 		 * Locate the smallest of the remaining components
531 		 */
532 		smallci = NULL;
533 		for (ci = cs->sc_cinfo;
534 		     ci < &cs->sc_cinfo[cs->sc_nccdisks]; ci++)
535 			if (ci->ci_size > size &&
536 			    (smallci == NULL ||
537 			     ci->ci_size < smallci->ci_size))
538 				smallci = ci;
539 
540 		/*
541 		 * Nobody left, all done
542 		 */
543 		if (smallci == NULL) {
544 			ii->ii_ndisk = 0;
545 			break;
546 		}
547 
548 		/*
549 		 * Record starting logical block and component offset
550 		 */
551 		ii->ii_startblk = bn / cs->sc_ileave;
552 		ii->ii_startoff = lbn;
553 
554 		/*
555 		 * Determine how many disks take part in this interleave
556 		 * and record their indices.
557 		 */
558 		ix = 0;
559 		for (ci = cs->sc_cinfo;
560 		     ci < &cs->sc_cinfo[cs->sc_nccdisks]; ci++)
561 			if (ci->ci_size >= smallci->ci_size)
562 				ii->ii_index[ix++] = ci - cs->sc_cinfo;
563 		ii->ii_ndisk = ix;
564 		bn += ix * (smallci->ci_size - size);
565 		lbn = smallci->ci_size / cs->sc_ileave;
566 		size = smallci->ci_size;
567 	}
568 #ifdef DEBUG
569 	if (ccddebug & CCDB_INIT)
570 		printiinfo(cs->sc_itable);
571 #endif
572 }
573 
574 /* ARGSUSED */
575 int
576 ccdopen(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 	struct disklabel *lp;
584 	int error = 0, part, pmask;
585 
586 #ifdef DEBUG
587 	if (ccddebug & CCDB_FOLLOW)
588 		printf("ccdopen(%x, %x)\n", dev, flags);
589 #endif
590 	if (unit >= numccd)
591 		return (ENXIO);
592 	cs = &ccd_softc[unit];
593 
594 	if ((error = ccdlock(cs)) != 0)
595 		return (error);
596 
597 	lp = cs->sc_dkdev.dk_label;
598 
599 	part = DISKPART(dev);
600 	pmask = (1 << part);
601 
602 	/*
603 	 * If we're initialized, check to see if there are any other
604 	 * open partitions.  If not, then it's safe to update
605 	 * the in-core disklabel.
606 	 */
607 	if ((cs->sc_flags & CCDF_INITED) && (cs->sc_dkdev.dk_openmask == 0))
608 		ccdgetdisklabel(dev);
609 
610 	/* Check that the partition exists. */
611 	if (part != RAW_PART) {
612 		if (((cs->sc_flags & CCDF_INITED) == 0) ||
613 		    ((part >= lp->d_npartitions) ||
614 		     (lp->d_partitions[part].p_fstype == FS_UNUSED))) {
615 			error = ENXIO;
616 			goto done;
617 		}
618 	}
619 
620 	/* Prevent our unit from being unconfigured while open. */
621 	switch (fmt) {
622 	case S_IFCHR:
623 		cs->sc_dkdev.dk_copenmask |= pmask;
624 		break;
625 
626 	case S_IFBLK:
627 		cs->sc_dkdev.dk_bopenmask |= pmask;
628 		break;
629 	}
630 	cs->sc_dkdev.dk_openmask =
631 	    cs->sc_dkdev.dk_copenmask | cs->sc_dkdev.dk_bopenmask;
632 
633  done:
634 	ccdunlock(cs);
635 	return (error);
636 }
637 
638 /* ARGSUSED */
639 int
640 ccdclose(dev, flags, fmt, p)
641 	dev_t dev;
642 	int flags, fmt;
643 	struct proc *p;
644 {
645 	int unit = ccdunit(dev);
646 	struct ccd_softc *cs;
647 	int error = 0, part;
648 
649 #ifdef DEBUG
650 	if (ccddebug & CCDB_FOLLOW)
651 		printf("ccdclose(%x, %x)\n", dev, flags);
652 #endif
653 
654 	if (unit >= numccd)
655 		return (ENXIO);
656 	cs = &ccd_softc[unit];
657 
658 	if ((error = ccdlock(cs)) != 0)
659 		return (error);
660 
661 	part = DISKPART(dev);
662 
663 	/* ...that much closer to allowing unconfiguration... */
664 	switch (fmt) {
665 	case S_IFCHR:
666 		cs->sc_dkdev.dk_copenmask &= ~(1 << part);
667 		break;
668 
669 	case S_IFBLK:
670 		cs->sc_dkdev.dk_bopenmask &= ~(1 << part);
671 		break;
672 	}
673 	cs->sc_dkdev.dk_openmask =
674 	    cs->sc_dkdev.dk_copenmask | cs->sc_dkdev.dk_bopenmask;
675 
676 	ccdunlock(cs);
677 	return (0);
678 }
679 
680 void
681 ccdstrategy(bp)
682 	register struct buf *bp;
683 {
684 	register int unit = ccdunit(bp->b_dev);
685 	register struct ccd_softc *cs = &ccd_softc[unit];
686 	register int s;
687 	int wlabel;
688 	struct disklabel *lp;
689 
690 #ifdef DEBUG
691 	if (ccddebug & CCDB_FOLLOW)
692 		printf("ccdstrategy(%p): unit %d\n", bp, unit);
693 #endif
694 	if ((cs->sc_flags & CCDF_INITED) == 0) {
695 		bp->b_error = ENXIO;
696 		bp->b_flags |= B_ERROR;
697 		goto done;
698 	}
699 
700 	/* If it's a nil transfer, wake up the top half now. */
701 	if (bp->b_bcount == 0)
702 		goto done;
703 
704 	lp = cs->sc_dkdev.dk_label;
705 
706 	/*
707 	 * Do bounds checking and adjust transfer.  If there's an
708 	 * error, the bounds check will flag that for us.
709 	 */
710 	wlabel = cs->sc_flags & (CCDF_WLABEL|CCDF_LABELLING);
711 	if (DISKPART(bp->b_dev) != RAW_PART)
712 		if (bounds_check_with_label(bp, lp, wlabel) <= 0)
713 			goto done;
714 
715 	bp->b_resid = bp->b_bcount;
716 
717 	/*
718 	 * "Start" the unit.
719 	 */
720 	s = splbio();
721 	ccdstart(cs, bp);
722 	splx(s);
723 	return;
724 done:
725 	biodone(bp);
726 }
727 
728 static void
729 ccdstart(cs, bp)
730 	register struct ccd_softc *cs;
731 	register struct buf *bp;
732 {
733 	register long bcount, rcount;
734 	struct ccdbuf *cbp[4];
735 	caddr_t addr;
736 	daddr_t bn;
737 	struct partition *pp;
738 
739 #ifdef DEBUG
740 	if (ccddebug & CCDB_FOLLOW)
741 		printf("ccdstart(%p, %p)\n", cs, bp);
742 #endif
743 
744 	/* Instrumentation. */
745 	disk_busy(&cs->sc_dkdev);
746 
747 	/*
748 	 * Translate the partition-relative block number to an absolute.
749 	 */
750 	bn = bp->b_blkno;
751 	if (DISKPART(bp->b_dev) != RAW_PART) {
752 		pp = &cs->sc_dkdev.dk_label->d_partitions[DISKPART(bp->b_dev)];
753 		bn += pp->p_offset;
754 	}
755 
756 	/*
757 	 * Allocate component buffers and fire off the requests
758 	 */
759 	addr = bp->b_data;
760 	for (bcount = bp->b_bcount; bcount > 0; bcount -= rcount) {
761 		ccdbuffer(cs, bp, bn, addr, bcount, cbp);
762 		rcount = cbp[0]->cb_buf.b_bcount;
763 		if ((cbp[0]->cb_buf.b_flags & B_READ) == 0)
764 			cbp[0]->cb_buf.b_vp->v_numoutput++;
765 		VOP_STRATEGY(&cbp[0]->cb_buf);
766 
767 		/*
768 		 * Mirror requires additional write.
769 		 */
770 		if ((cs->sc_cflags & CCDF_MIRROR) &&
771 		    ((cbp[0]->cb_buf.b_flags & B_READ) == 0)) {
772 			cbp[1]->cb_buf.b_vp->v_numoutput++;
773 			VOP_STRATEGY(&cbp[1]->cb_buf);
774 		}
775 
776 		bn += btodb(rcount);
777 		addr += rcount;
778 	}
779 }
780 
781 /*
782  * Build a component buffer header.
783  */
784 static void
785 ccdbuffer(cs, bp, bn, addr, bcount, cbpp)
786 	register struct ccd_softc *cs;
787 	struct buf *bp;
788 	daddr_t bn;
789 	caddr_t addr;
790 	long bcount;
791 	struct ccdbuf **cbpp;
792 {
793 	register struct ccdcinfo *ci, *ci2 = NULL;
794 	register struct ccdbuf *cbp;
795 	register daddr_t cbn, cboff;
796 	int ccdisk;
797 
798 #ifdef DEBUG
799 	if (ccddebug & CCDB_IO)
800 		printf("ccdbuffer(%p, %p, %d, %p, %ld)\n",
801 		       cs, bp, bn, addr, bcount);
802 #endif
803 	/*
804 	 * Determine which component bn falls in.
805 	 */
806 	cbn = bn;
807 	cboff = 0;
808 
809 	/*
810 	 * Serially concatenated
811 	 */
812 	if (cs->sc_ileave == 0) {
813 		register daddr_t sblk;
814 
815 		sblk = 0;
816 		for (ccdisk = 0, ci = &cs->sc_cinfo[ccdisk];
817 		    cbn >= sblk + ci->ci_size;
818 		    ccdisk++, ci = &cs->sc_cinfo[ccdisk])
819 			sblk += ci->ci_size;
820 		cbn -= sblk;
821 	}
822 	/*
823 	 * Interleaved
824 	 */
825 	else {
826 		register struct ccdiinfo *ii;
827 		int off;
828 
829 		cboff = cbn % cs->sc_ileave;
830 		cbn /= cs->sc_ileave;
831 		for (ii = cs->sc_itable; ii->ii_ndisk; ii++)
832 			if (ii->ii_startblk > cbn)
833 				break;
834 		ii--;
835 		off = cbn - ii->ii_startblk;
836 		if (ii->ii_ndisk == 1) {
837 			ccdisk = ii->ii_index[0];
838 			cbn = ii->ii_startoff + off;
839 		} else {
840 			if (cs->sc_cflags & CCDF_MIRROR) {
841 				ccdisk =
842 				    ii->ii_index[off % (ii->ii_ndisk / 2)];
843 				cbn = ii->ii_startoff +
844 				    (off / (ii->ii_ndisk / 2));
845 				/* Mirrored data */
846 				ci2 =
847 				    &cs->sc_cinfo[ccdisk + (ii->ii_ndisk / 2)];
848 			} else {
849 				/* Normal case. */
850 				ccdisk = ii->ii_index[off % ii->ii_ndisk];
851 				cbn = ii->ii_startoff + off / ii->ii_ndisk;
852 			}
853 		}
854 		cbn *= cs->sc_ileave;
855 		ci = &cs->sc_cinfo[ccdisk];
856 	}
857 
858 	/*
859 	 * Fill in the component buf structure.
860 	 */
861 	CCDGETBUF(cs, cbp);
862 	cbp->cb_flags = 0;
863 	cbp->cb_buf.b_flags = bp->b_flags | B_CALL;
864 	cbp->cb_buf.b_iodone = ccdiodone;
865 	cbp->cb_buf.b_proc = bp->b_proc;
866 	cbp->cb_buf.b_dev = ci->ci_dev;		/* XXX */
867 	cbp->cb_buf.b_blkno = cbn + cboff;
868 	cbp->cb_buf.b_data = addr;
869 	cbp->cb_buf.b_vp = ci->ci_vp;
870 	if (cs->sc_ileave == 0)
871 		cbp->cb_buf.b_bcount = dbtob(ci->ci_size - cbn);
872 	else
873 		cbp->cb_buf.b_bcount = dbtob(cs->sc_ileave - cboff);
874 	if (cbp->cb_buf.b_bcount > bcount)
875 		cbp->cb_buf.b_bcount = bcount;
876 
877 	/*
878 	 * context for ccdiodone
879 	 */
880 	cbp->cb_obp = bp;
881 	cbp->cb_unit = cs->sc_unit;
882 	cbp->cb_comp = ccdisk;
883 
884 	/* First buffer is dealt with. */
885 	cbpp[0] = cbp;
886 
887 #ifdef DEBUG
888 	if (ccddebug & CCDB_IO)
889 		printf(" dev %x(u%d): cbp %p bn %d addr %p bcnt %ld\n",
890 		    ci->ci_dev, ci-cs->sc_cinfo, cbp, cbp->cb_buf.b_blkno,
891 		    cbp->cb_buf.b_data, cbp->cb_buf.b_bcount);
892 #endif
893 
894 	/*
895 	 * Mirrors have an additional write operation that is nearly
896 	 * identical to the first.
897 	 */
898 	if ((cs->sc_cflags & CCDF_MIRROR) &&
899 	    ((cbp->cb_buf.b_flags & B_READ) == 0)) {
900 		CCDGETBUF(cs, cbp);
901 		*cbp = *cbpp[0];
902 		cbp->cb_flags = CBF_MIRROR;
903 		cbp->cb_buf.b_dev = ci2->ci_dev;	/* XXX */
904 		cbp->cb_buf.b_vp = ci2->ci_vp;
905 		cbp->cb_comp = ci2 - cs->sc_cinfo;
906 		cbpp[1] = cbp;
907 	}
908 }
909 
910 static void
911 ccdintr(cs, bp)
912 	register struct ccd_softc *cs;
913 	register struct buf *bp;
914 {
915 
916 #ifdef DEBUG
917 	if (ccddebug & CCDB_FOLLOW)
918 		printf("ccdintr(%p, %p)\n", cs, bp);
919 #endif
920 	/*
921 	 * Request is done for better or worse, wakeup the top half.
922 	 */
923 	if (bp->b_flags & B_ERROR)
924 		bp->b_resid = bp->b_bcount;
925 	disk_unbusy(&cs->sc_dkdev, (bp->b_bcount - bp->b_resid));
926 	biodone(bp);
927 }
928 
929 /*
930  * Called at interrupt time.
931  * Mark the component as done and if all components are done,
932  * take a ccd interrupt.
933  */
934 void
935 ccdiodone(vbp)
936 	struct buf *vbp;
937 {
938 	struct ccdbuf *cbp = (struct ccdbuf *) vbp;
939 	register struct buf *bp = cbp->cb_obp;
940 	register int unit = cbp->cb_unit;
941 	struct ccd_softc *cs = &ccd_softc[unit];
942 	int count, cbflags, s;
943 	char *comptype;
944 
945 	s = splbio();
946 #ifdef DEBUG
947 	if (ccddebug & CCDB_FOLLOW)
948 		printf("ccdiodone(%p)\n", cbp);
949 	if (ccddebug & CCDB_IO) {
950 		if (cbp->cb_flags & CBF_MIRROR)
951 			printf("ccdiodone: mirror component\n");
952 		else
953 			printf("ccdiodone: bp %p bcount %ld resid %ld\n",
954 			       bp, bp->b_bcount, bp->b_resid);
955 		printf(" dev %x(u%d), cbp %p bn %d addr %p bcnt %ld\n",
956 		       cbp->cb_buf.b_dev, cbp->cb_comp, cbp,
957 		       cbp->cb_buf.b_blkno, cbp->cb_buf.b_data,
958 		       cbp->cb_buf.b_bcount);
959 	}
960 #endif
961 
962 	if (cbp->cb_buf.b_flags & B_ERROR) {
963 		if (cbp->cb_flags & CBF_MIRROR)
964 			comptype = " (mirror)";
965 		else {
966 			bp->b_flags |= B_ERROR;
967 			bp->b_error = cbp->cb_buf.b_error ?
968 			    cbp->cb_buf.b_error : EIO;
969 			comptype = "";
970 		}
971 
972 		printf("%s: error %d on component %d%s\n",
973 		       cs->sc_xname, bp->b_error, cbp->cb_comp, comptype);
974 	}
975 	count = cbp->cb_buf.b_bcount;
976 	cbflags = cbp->cb_flags;
977 	CCDPUTBUF(cs, cbp);
978 
979 	/*
980 	 * If all done, "interrupt".
981 	 *
982 	 * Note that mirror component buffers aren't counted against
983 	 * the original I/O buffer.
984 	 */
985 	if ((cbflags & CBF_MIRROR) == 0) {
986 		bp->b_resid -= count;
987 		if (bp->b_resid < 0)
988 			panic("ccdiodone: count");
989 		if (bp->b_resid == 0)
990 			ccdintr(&ccd_softc[unit], bp);
991 	}
992 	splx(s);
993 }
994 
995 /* ARGSUSED */
996 int
997 ccdread(dev, uio, flags)
998 	dev_t dev;
999 	struct uio *uio;
1000 	int flags;
1001 {
1002 	int unit = ccdunit(dev);
1003 	struct ccd_softc *cs;
1004 
1005 #ifdef DEBUG
1006 	if (ccddebug & CCDB_FOLLOW)
1007 		printf("ccdread(%x, %p)\n", dev, uio);
1008 #endif
1009 	if (unit >= numccd)
1010 		return (ENXIO);
1011 	cs = &ccd_softc[unit];
1012 
1013 	if ((cs->sc_flags & CCDF_INITED) == 0)
1014 		return (ENXIO);
1015 
1016 	/*
1017 	 * XXX: It's not clear that using minphys() is completely safe,
1018 	 * in particular, for raw I/O.  Underlying devices might have some
1019 	 * non-obvious limits, because of the copy to user-space.
1020 	 */
1021 	return (physio(ccdstrategy, NULL, dev, B_READ, minphys, uio));
1022 }
1023 
1024 /* ARGSUSED */
1025 int
1026 ccdwrite(dev, uio, flags)
1027 	dev_t dev;
1028 	struct uio *uio;
1029 	int flags;
1030 {
1031 	int unit = ccdunit(dev);
1032 	struct ccd_softc *cs;
1033 
1034 #ifdef DEBUG
1035 	if (ccddebug & CCDB_FOLLOW)
1036 		printf("ccdwrite(%x, %p)\n", dev, uio);
1037 #endif
1038 	if (unit >= numccd)
1039 		return (ENXIO);
1040 	cs = &ccd_softc[unit];
1041 
1042 	if ((cs->sc_flags & CCDF_INITED) == 0)
1043 		return (ENXIO);
1044 
1045 	/*
1046 	 * XXX: It's not clear that using minphys() is completely safe,
1047 	 * in particular, for raw I/O.  Underlying devices might have some
1048 	 * non-obvious limits, because of the copy to user-space.
1049 	 */
1050 	return (physio(ccdstrategy, NULL, dev, B_WRITE, minphys, uio));
1051 }
1052 
1053 int
1054 ccdioctl(dev, cmd, data, flag, p)
1055 	dev_t dev;
1056 	u_long cmd;
1057 	caddr_t data;
1058 	int flag;
1059 	struct proc *p;
1060 {
1061 	int unit = ccdunit(dev);
1062 	int i, j, lookedup = 0, error = 0;
1063 	int part, pmask;
1064 	struct ccd_softc *cs;
1065 	struct ccd_ioctl *ccio = (struct ccd_ioctl *)data;
1066 	struct ccddevice ccd;
1067 	struct ccdbuf *cbp;
1068 	char **cpp;
1069 	struct vnode **vpp;
1070 
1071 	if (unit >= numccd)
1072 		return (ENXIO);
1073 	cs = &ccd_softc[unit];
1074 
1075 	bzero(&ccd, sizeof(ccd));
1076 
1077 	/* Must be open for writes for these commands... */
1078 	switch (cmd) {
1079 	case CCDIOCSET:
1080 	case CCDIOCCLR:
1081 	case DIOCSDINFO:
1082 	case DIOCWDINFO:
1083 	case DIOCWLABEL:
1084 		if ((flag & FWRITE) == 0)
1085 			return (EBADF);
1086 	}
1087 
1088 	/* Must be initialized for these... */
1089 	switch (cmd) {
1090 	case CCDIOCCLR:
1091 	case DIOCGDINFO:
1092 	case DIOCSDINFO:
1093 	case DIOCWDINFO:
1094 	case DIOCGPART:
1095 	case DIOCWLABEL:
1096 		if ((cs->sc_flags & CCDF_INITED) == 0)
1097 			return (ENXIO);
1098 	}
1099 
1100 	switch (cmd) {
1101 	case CCDIOCSET:
1102 		if (cs->sc_flags & CCDF_INITED)
1103 			return (EBUSY);
1104 
1105 		if ((error = ccdlock(cs)) != 0)
1106 			return (error);
1107 
1108 		/* Fill in some important bits. */
1109 		ccd.ccd_unit = unit;
1110 		ccd.ccd_interleave = ccio->ccio_ileave;
1111 		ccd.ccd_flags = ccio->ccio_flags & CCDF_USERMASK;
1112 
1113 		/*
1114 		 * Allocate space for and copy in the array of
1115 		 * componet pathnames and device numbers.
1116 		 */
1117 		cpp = malloc(ccio->ccio_ndisks * sizeof(char *),
1118 		    M_DEVBUF, M_WAITOK);
1119 		vpp = malloc(ccio->ccio_ndisks * sizeof(struct vnode *),
1120 		    M_DEVBUF, M_WAITOK);
1121 
1122 		error = copyin((caddr_t)ccio->ccio_disks, (caddr_t)cpp,
1123 		    ccio->ccio_ndisks * sizeof(char **));
1124 		if (error) {
1125 			free(vpp, M_DEVBUF);
1126 			free(cpp, M_DEVBUF);
1127 			ccdunlock(cs);
1128 			return (error);
1129 		}
1130 
1131 #ifdef DEBUG
1132 		if (ccddebug & CCDB_INIT)
1133 			for (i = 0; i < ccio->ccio_ndisks; ++i)
1134 				printf("ccdioctl: component %d: 0x%p\n",
1135 				    i, cpp[i]);
1136 #endif
1137 
1138 		for (i = 0; i < ccio->ccio_ndisks; ++i) {
1139 #ifdef DEBUG
1140 			if (ccddebug & CCDB_INIT)
1141 				printf("ccdioctl: lookedup = %d\n", lookedup);
1142 #endif
1143 			if ((error = ccdlookup(cpp[i], p, &vpp[i])) != 0) {
1144 				for (j = 0; j < lookedup; ++j)
1145 					(void)vn_close(vpp[j], FREAD|FWRITE,
1146 					    p->p_ucred, p);
1147 				free(vpp, M_DEVBUF);
1148 				free(cpp, M_DEVBUF);
1149 				ccdunlock(cs);
1150 				return (error);
1151 			}
1152 			++lookedup;
1153 		}
1154 		ccd.ccd_cpp = cpp;
1155 		ccd.ccd_vpp = vpp;
1156 		ccd.ccd_ndev = ccio->ccio_ndisks;
1157 
1158 		/*
1159 		 * Initialize the ccd.  Fills in the softc for us.
1160 		 */
1161 		if ((error = ccdinit(&ccd, cpp, p)) != 0) {
1162 			for (j = 0; j < lookedup; ++j)
1163 				(void)vn_close(vpp[j], FREAD|FWRITE,
1164 				    p->p_ucred, p);
1165 			bzero(&ccd_softc[unit], sizeof(struct ccd_softc));
1166 			free(vpp, M_DEVBUF);
1167 			free(cpp, M_DEVBUF);
1168 			ccdunlock(cs);
1169 			return (error);
1170 		}
1171 
1172 		/*
1173 		 * The ccd has been successfully initialized, so
1174 		 * we can place it into the array.  Don't try to
1175 		 * read the disklabel until the disk has been attached,
1176 		 * because space for the disklabel is allocated
1177 		 * in disk_attach();
1178 		 */
1179 		bcopy(&ccd, &ccddevs[unit], sizeof(ccd));
1180 		ccio->ccio_unit = unit;
1181 		ccio->ccio_size = cs->sc_size;
1182 
1183 		/* Attach the disk. */
1184 		cs->sc_dkdev.dk_name = cs->sc_xname;
1185 		disk_attach(&cs->sc_dkdev);
1186 
1187 		/* Try and read the disklabel. */
1188 		ccdgetdisklabel(dev);
1189 
1190 		ccdunlock(cs);
1191 
1192 		break;
1193 
1194 	case CCDIOCCLR:
1195 		if ((error = ccdlock(cs)) != 0)
1196 			return (error);
1197 
1198 		/*
1199 		 * Don't unconfigure if any other partitions are open
1200 		 * or if both the character and block flavors of this
1201 		 * partition are open.
1202 		 */
1203 		part = DISKPART(dev);
1204 		pmask = (1 << part);
1205 		if ((cs->sc_dkdev.dk_openmask & ~pmask) ||
1206 		    ((cs->sc_dkdev.dk_bopenmask & pmask) &&
1207 		    (cs->sc_dkdev.dk_copenmask & pmask))) {
1208 			ccdunlock(cs);
1209 			return (EBUSY);
1210 		}
1211 
1212 		/*
1213 		 * Free ccd_softc information and clear entry.
1214 		 */
1215 
1216 		/* Close the components and free their pathnames. */
1217 		for (i = 0; i < cs->sc_nccdisks; ++i) {
1218 			/*
1219 			 * XXX: this close could potentially fail and
1220 			 * cause Bad Things.  Maybe we need to force
1221 			 * the close to happen?
1222 			 */
1223 #ifdef DEBUG
1224 			if (ccddebug & CCDB_VNODE)
1225 				vprint("CCDIOCCLR: vnode info",
1226 				    cs->sc_cinfo[i].ci_vp);
1227 #endif
1228 			(void)vn_close(cs->sc_cinfo[i].ci_vp, FREAD|FWRITE,
1229 			    p->p_ucred, p);
1230 			free(cs->sc_cinfo[i].ci_path, M_DEVBUF);
1231 		}
1232 
1233 		/* Free component buffer freelist. */
1234 		while ((cbp = cs->sc_freelist.lh_first) != NULL) {
1235 			LIST_REMOVE(cbp, cb_list);
1236 			FREE(cbp, M_DEVBUF);
1237 		}
1238 
1239 		/* Free interleave index. */
1240 		for (i = 0; cs->sc_itable[i].ii_ndisk; ++i)
1241 			free(cs->sc_itable[i].ii_index, M_DEVBUF);
1242 
1243 		/* Free component info and interleave table. */
1244 		free(cs->sc_cinfo, M_DEVBUF);
1245 		free(cs->sc_itable, M_DEVBUF);
1246 		cs->sc_flags &= ~CCDF_INITED;
1247 
1248 		/*
1249 		 * Free ccddevice information and clear entry.
1250 		 */
1251 		free(ccddevs[unit].ccd_cpp, M_DEVBUF);
1252 		free(ccddevs[unit].ccd_vpp, M_DEVBUF);
1253 		bcopy(&ccd, &ccddevs[unit], sizeof(ccd));
1254 
1255 		/* Detatch the disk. */
1256 		disk_detach(&cs->sc_dkdev);
1257 
1258 		ccdunlock(cs);
1259 
1260 		break;
1261 
1262 	case DIOCGDINFO:
1263 		*(struct disklabel *)data = *(cs->sc_dkdev.dk_label);
1264 		break;
1265 
1266 	case DIOCGPART:
1267 		((struct partinfo *)data)->disklab = cs->sc_dkdev.dk_label;
1268 		((struct partinfo *)data)->part =
1269 		    &cs->sc_dkdev.dk_label->d_partitions[DISKPART(dev)];
1270 		break;
1271 
1272 	case DIOCWDINFO:
1273 	case DIOCSDINFO:
1274 		if ((error = ccdlock(cs)) != 0)
1275 			return (error);
1276 
1277 		cs->sc_flags |= CCDF_LABELLING;
1278 
1279 		error = setdisklabel(cs->sc_dkdev.dk_label,
1280 		    (struct disklabel *)data, 0, cs->sc_dkdev.dk_cpulabel);
1281 		if (error == 0) {
1282 			if (cmd == DIOCWDINFO)
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 
1290 		ccdunlock(cs);
1291 
1292 		if (error)
1293 			return (error);
1294 		break;
1295 
1296 	case DIOCWLABEL:
1297 		if (*(int *)data != 0)
1298 			cs->sc_flags |= CCDF_WLABEL;
1299 		else
1300 			cs->sc_flags &= ~CCDF_WLABEL;
1301 		break;
1302 
1303 	default:
1304 		return (ENOTTY);
1305 	}
1306 
1307 	return (0);
1308 }
1309 
1310 int
1311 ccdsize(dev)
1312 	dev_t dev;
1313 {
1314 	struct ccd_softc *cs;
1315 	struct disklabel *lp;
1316 	int part, unit, omask, size;
1317 
1318 	unit = ccdunit(dev);
1319 	if (unit >= numccd)
1320 		return (-1);
1321 	cs = &ccd_softc[unit];
1322 
1323 	if ((cs->sc_flags & CCDF_INITED) == 0)
1324 		return (-1);
1325 
1326 	part = DISKPART(dev);
1327 	omask = cs->sc_dkdev.dk_openmask & (1 << part);
1328 	lp = cs->sc_dkdev.dk_label;
1329 
1330 	if (omask == 0 && ccdopen(dev, 0, S_IFBLK, curproc))
1331 		return (-1);
1332 
1333 	if (lp->d_partitions[part].p_fstype != FS_SWAP)
1334 		size = -1;
1335 	else
1336 		size = lp->d_partitions[part].p_size *
1337 		    (lp->d_secsize / DEV_BSIZE);
1338 
1339 	if (omask == 0 && ccdclose(dev, 0, S_IFBLK, curproc))
1340 		return (-1);
1341 
1342 	return (size);
1343 }
1344 
1345 int
1346 ccddump(dev, blkno, va, size)
1347 	dev_t dev;
1348 	daddr_t blkno;
1349 	caddr_t va;
1350 	size_t size;
1351 {
1352 
1353 	/* Not implemented. */
1354 	return ENXIO;
1355 }
1356 
1357 /*
1358  * Lookup the provided name in the filesystem.  If the file exists,
1359  * is a valid block device, and isn't being used by anyone else,
1360  * set *vpp to the file's vnode.
1361  */
1362 static int
1363 ccdlookup(path, p, vpp)
1364 	char *path;
1365 	struct proc *p;
1366 	struct vnode **vpp;	/* result */
1367 {
1368 	struct nameidata nd;
1369 	struct vnode *vp;
1370 	struct vattr va;
1371 	int error;
1372 
1373 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, path, p);
1374 	if ((error = vn_open(&nd, FREAD|FWRITE, 0)) != 0) {
1375 #ifdef DEBUG
1376 		if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
1377 			printf("ccdlookup: vn_open error = %d\n", error);
1378 #endif
1379 		return (error);
1380 	}
1381 	vp = nd.ni_vp;
1382 
1383 	if (vp->v_usecount > 1) {
1384 		VOP_UNLOCK(vp);
1385 		(void)vn_close(vp, FREAD|FWRITE, p->p_ucred, p);
1386 		return (EBUSY);
1387 	}
1388 
1389 	if ((error = VOP_GETATTR(vp, &va, p->p_ucred, p)) != 0) {
1390 #ifdef DEBUG
1391 		if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
1392 			printf("ccdlookup: getattr error = %d\n", error);
1393 #endif
1394 		VOP_UNLOCK(vp);
1395 		(void)vn_close(vp, FREAD|FWRITE, p->p_ucred, p);
1396 		return (error);
1397 	}
1398 
1399 	/* XXX: eventually we should handle VREG, too. */
1400 	if (va.va_type != VBLK) {
1401 		VOP_UNLOCK(vp);
1402 		(void)vn_close(vp, FREAD|FWRITE, p->p_ucred, p);
1403 		return (ENOTBLK);
1404 	}
1405 
1406 #ifdef DEBUG
1407 	if (ccddebug & CCDB_VNODE)
1408 		vprint("ccdlookup: vnode info", vp);
1409 #endif
1410 
1411 	VOP_UNLOCK(vp);
1412 	*vpp = vp;
1413 	return (0);
1414 }
1415 
1416 /*
1417  * Read the disklabel from the ccd.  If one is not present, fake one
1418  * up.
1419  */
1420 static void
1421 ccdgetdisklabel(dev)
1422 	dev_t dev;
1423 {
1424 	int unit = ccdunit(dev);
1425 	struct ccd_softc *cs = &ccd_softc[unit];
1426 	char *errstring;
1427 	struct disklabel *lp = cs->sc_dkdev.dk_label;
1428 	struct cpu_disklabel *clp = cs->sc_dkdev.dk_cpulabel;
1429 	struct ccdgeom *ccg = &cs->sc_geom;
1430 
1431 	bzero(lp, sizeof(*lp));
1432 	bzero(clp, sizeof(*clp));
1433 
1434 	lp->d_secperunit = cs->sc_size;
1435 	lp->d_secsize = ccg->ccg_secsize;
1436 	lp->d_nsectors = ccg->ccg_nsectors;
1437 	lp->d_ntracks = ccg->ccg_ntracks;
1438 	lp->d_ncylinders = ccg->ccg_ncylinders;
1439 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1440 
1441 	strncpy(lp->d_typename, "ccd", sizeof(lp->d_typename));
1442 	lp->d_type = DTYPE_CCD;
1443 	strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
1444 	lp->d_rpm = 3600;
1445 	lp->d_interleave = 1;
1446 	lp->d_flags = 0;
1447 
1448 	lp->d_partitions[RAW_PART].p_offset = 0;
1449 	lp->d_partitions[RAW_PART].p_size = cs->sc_size;
1450 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
1451 	lp->d_npartitions = RAW_PART + 1;
1452 
1453 	lp->d_magic = DISKMAGIC;
1454 	lp->d_magic2 = DISKMAGIC;
1455 	lp->d_checksum = dkcksum(cs->sc_dkdev.dk_label);
1456 
1457 	/*
1458 	 * Call the generic disklabel extraction routine.
1459 	 */
1460 	errstring = readdisklabel(CCDLABELDEV(dev), ccdstrategy,
1461 	    cs->sc_dkdev.dk_label, cs->sc_dkdev.dk_cpulabel);
1462 	if (errstring)
1463 		ccdmakedisklabel(cs);
1464 
1465 #ifdef DEBUG
1466 	/* It's actually extremely common to have unlabeled ccds. */
1467 	if (ccddebug & CCDB_LABEL)
1468 		if (errstring != NULL)
1469 			printf("%s: %s\n", cs->sc_xname, errstring);
1470 #endif
1471 }
1472 
1473 /*
1474  * Take care of things one might want to take care of in the event
1475  * that a disklabel isn't present.
1476  */
1477 static void
1478 ccdmakedisklabel(cs)
1479 	struct ccd_softc *cs;
1480 {
1481 	struct disklabel *lp = cs->sc_dkdev.dk_label;
1482 
1483 	/*
1484 	 * For historical reasons, if there's no disklabel present
1485 	 * the raw partition must be marked FS_BSDFFS.
1486 	 */
1487 	lp->d_partitions[RAW_PART].p_fstype = FS_BSDFFS;
1488 
1489 	strncpy(lp->d_packname, "default label", sizeof(lp->d_packname));
1490 }
1491 
1492 /*
1493  * Wait interruptibly for an exclusive lock.
1494  *
1495  * XXX
1496  * Several drivers do this; it should be abstracted and made MP-safe.
1497  */
1498 static int
1499 ccdlock(cs)
1500 	struct ccd_softc *cs;
1501 {
1502 	int error;
1503 
1504 	while ((cs->sc_flags & CCDF_LOCKED) != 0) {
1505 		cs->sc_flags |= CCDF_WANTED;
1506 		if ((error = tsleep(cs, PRIBIO | PCATCH, "ccdlck", 0)) != 0)
1507 			return (error);
1508 	}
1509 	cs->sc_flags |= CCDF_LOCKED;
1510 	return (0);
1511 }
1512 
1513 /*
1514  * Unlock and wake up any waiters.
1515  */
1516 static void
1517 ccdunlock(cs)
1518 	struct ccd_softc *cs;
1519 {
1520 
1521 	cs->sc_flags &= ~CCDF_LOCKED;
1522 	if ((cs->sc_flags & CCDF_WANTED) != 0) {
1523 		cs->sc_flags &= ~CCDF_WANTED;
1524 		wakeup(cs);
1525 	}
1526 }
1527 
1528 #ifdef DEBUG
1529 static void
1530 printiinfo(ii)
1531 	struct ccdiinfo *ii;
1532 {
1533 	register int ix, i;
1534 
1535 	for (ix = 0; ii->ii_ndisk; ix++, ii++) {
1536 		printf(" itab[%d]: #dk %d sblk %d soff %d",
1537 		    ix, ii->ii_ndisk, ii->ii_startblk, ii->ii_startoff);
1538 		for (i = 0; i < ii->ii_ndisk; i++)
1539 			printf(" %d", ii->ii_index[i]);
1540 		printf("\n");
1541 	}
1542 }
1543 #endif
1544