xref: /netbsd-src/sys/dev/ccd.c (revision 001c68bd94f75ce9270b69227c4199fbf34ee396)
1 /*	$NetBSD: ccd.c,v 1.90 2003/06/29 22:29:57 fvdl 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 1.90 2003/06/29 22:29:57 fvdl 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 *));
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.  Only read the disklabel if it is
542 	 * not already valid.
543 	 */
544 	if ((cs->sc_flags & (CCDF_INITED|CCDF_VLABEL)) == CCDF_INITED &&
545 	    cs->sc_dkdev.dk_openmask == 0)
546 		ccdgetdisklabel(dev);
547 
548 	/* Check that the partition exists. */
549 	if (part != RAW_PART) {
550 		if (((cs->sc_flags & CCDF_INITED) == 0) ||
551 		    ((part >= lp->d_npartitions) ||
552 		     (lp->d_partitions[part].p_fstype == FS_UNUSED))) {
553 			error = ENXIO;
554 			goto done;
555 		}
556 	}
557 
558 	/* Prevent our unit from being unconfigured while open. */
559 	switch (fmt) {
560 	case S_IFCHR:
561 		cs->sc_dkdev.dk_copenmask |= pmask;
562 		break;
563 
564 	case S_IFBLK:
565 		cs->sc_dkdev.dk_bopenmask |= pmask;
566 		break;
567 	}
568 	cs->sc_dkdev.dk_openmask =
569 	    cs->sc_dkdev.dk_copenmask | cs->sc_dkdev.dk_bopenmask;
570 
571  done:
572 	(void) lockmgr(&cs->sc_lock, LK_RELEASE, NULL);
573 	return (error);
574 }
575 
576 /* ARGSUSED */
577 int
578 ccdclose(dev, flags, fmt, p)
579 	dev_t dev;
580 	int flags, fmt;
581 	struct proc *p;
582 {
583 	int unit = ccdunit(dev);
584 	struct ccd_softc *cs;
585 	int error = 0, part;
586 
587 #ifdef DEBUG
588 	if (ccddebug & CCDB_FOLLOW)
589 		printf("ccdclose(0x%x, 0x%x)\n", dev, flags);
590 #endif
591 
592 	if (unit >= numccd)
593 		return (ENXIO);
594 	cs = &ccd_softc[unit];
595 
596 	if ((error = lockmgr(&cs->sc_lock, LK_EXCLUSIVE, NULL)) != 0)
597 		return (error);
598 
599 	part = DISKPART(dev);
600 
601 	/* ...that much closer to allowing unconfiguration... */
602 	switch (fmt) {
603 	case S_IFCHR:
604 		cs->sc_dkdev.dk_copenmask &= ~(1 << part);
605 		break;
606 
607 	case S_IFBLK:
608 		cs->sc_dkdev.dk_bopenmask &= ~(1 << part);
609 		break;
610 	}
611 	cs->sc_dkdev.dk_openmask =
612 	    cs->sc_dkdev.dk_copenmask | cs->sc_dkdev.dk_bopenmask;
613 
614 	if (cs->sc_dkdev.dk_openmask == 0) {
615 		if ((cs->sc_flags & CCDF_KLABEL) == 0)
616 			cs->sc_flags &= ~CCDF_VLABEL;
617 	}
618 
619 	(void) lockmgr(&cs->sc_lock, LK_RELEASE, NULL);
620 	return (0);
621 }
622 
623 void
624 ccdstrategy(bp)
625 	struct buf *bp;
626 {
627 	int unit = ccdunit(bp->b_dev);
628 	struct ccd_softc *cs = &ccd_softc[unit];
629 	daddr_t blkno;
630 	int s;
631 	int wlabel;
632 	struct disklabel *lp;
633 
634 #ifdef DEBUG
635 	if (ccddebug & CCDB_FOLLOW)
636 		printf("ccdstrategy(%p): unit %d\n", bp, unit);
637 #endif
638 	if ((cs->sc_flags & CCDF_INITED) == 0) {
639 #ifdef DEBUG
640 		if (ccddebug & CCDB_FOLLOW)
641 			printf("ccdstrategy: unit %d: not inited\n", unit);
642 #endif
643 		bp->b_error = ENXIO;
644 		bp->b_flags |= B_ERROR;
645 		goto done;
646 	}
647 
648 	/* If it's a nil transfer, wake up the top half now. */
649 	if (bp->b_bcount == 0)
650 		goto done;
651 
652 	lp = cs->sc_dkdev.dk_label;
653 
654 	/*
655 	 * Do bounds checking and adjust transfer.  If there's an
656 	 * error, the bounds check will flag that for us.  Convert
657 	 * the partition relative block number to an absolute.
658 	 */
659 	blkno = bp->b_blkno;
660 	wlabel = cs->sc_flags & (CCDF_WLABEL|CCDF_LABELLING);
661 	if (DISKPART(bp->b_dev) != RAW_PART) {
662 		if (bounds_check_with_label(&cs->sc_dkdev, bp, wlabel) <= 0)
663 			goto done;
664 		blkno += lp->d_partitions[DISKPART(bp->b_dev)].p_offset;
665 	}
666 	bp->b_rawblkno = blkno;
667 
668 	/* Place it in the queue and start I/O on the unit. */
669 	s = splbio();
670 	BUFQ_PUT(&cs->sc_bufq, bp);
671 	ccdstart(cs);
672 	splx(s);
673 	return;
674 
675  done:
676 	bp->b_resid = bp->b_bcount;
677 	biodone(bp);
678 }
679 
680 static void
681 ccdstart(cs)
682 	struct ccd_softc *cs;
683 {
684 	long bcount, rcount;
685 	struct buf *bp;
686 	struct ccdbuf *cbp;
687 	caddr_t addr;
688 	daddr_t bn;
689 	SIMPLEQ_HEAD(, ccdbuf) cbufq;
690 
691 #ifdef DEBUG
692 	if (ccddebug & CCDB_FOLLOW)
693 		printf("ccdstart(%p, %p)\n", cs, bp);
694 #endif
695 
696 	/* See if there is work for us to do. */
697 	while ((bp = BUFQ_PEEK(&cs->sc_bufq)) != NULL) {
698 		/* Instrumentation. */
699 		disk_busy(&cs->sc_dkdev);
700 
701 		bp->b_resid = bp->b_bcount;
702 		bn = bp->b_rawblkno;
703 
704 		/* Allocate the component buffers. */
705 		SIMPLEQ_INIT(&cbufq);
706 		addr = bp->b_data;
707 		for (bcount = bp->b_bcount; bcount > 0; bcount -= rcount) {
708 			cbp = ccdbuffer(cs, bp, bn, addr, bcount);
709 			if (cbp == NULL) {
710 				/*
711 				 * Can't allocate a component buffer; just
712 				 * defer the job until later.
713 				 *
714 				 * XXX We might consider a watchdog timer
715 				 * XXX to make sure we are kicked into action,
716 				 * XXX or consider a low-water mark for our
717 				 * XXX component buffer pool.
718 				 */
719 				while ((cbp = SIMPLEQ_FIRST(&cbufq)) != NULL) {
720 					SIMPLEQ_REMOVE_HEAD(&cbufq, cb_q);
721 					CCD_PUTBUF(cbp);
722 				}
723 				disk_unbusy(&cs->sc_dkdev, 0, 0);
724 				return;
725 			}
726 			SIMPLEQ_INSERT_TAIL(&cbufq, cbp, cb_q);
727 			rcount = cbp->cb_buf.b_bcount;
728 			bn += btodb(rcount);
729 			addr += rcount;
730 		}
731 
732 		/* Transfer all set up, remove job from the queue. */
733 		(void) BUFQ_GET(&cs->sc_bufq);
734 
735 		/* Now fire off the requests. */
736 		while ((cbp = SIMPLEQ_FIRST(&cbufq)) != NULL) {
737 			SIMPLEQ_REMOVE_HEAD(&cbufq, cb_q);
738 			if ((cbp->cb_buf.b_flags & B_READ) == 0)
739 				cbp->cb_buf.b_vp->v_numoutput++;
740 			VOP_STRATEGY(&cbp->cb_buf);
741 		}
742 	}
743 }
744 
745 /*
746  * Build a component buffer header.
747  */
748 static struct ccdbuf *
749 ccdbuffer(cs, bp, bn, addr, bcount)
750 	struct ccd_softc *cs;
751 	struct buf *bp;
752 	daddr_t bn;
753 	caddr_t addr;
754 	long bcount;
755 {
756 	struct ccdcinfo *ci;
757 	struct ccdbuf *cbp;
758 	daddr_t cbn, cboff;
759 	u_int64_t cbc;
760 	int ccdisk;
761 
762 #ifdef DEBUG
763 	if (ccddebug & CCDB_IO)
764 		printf("ccdbuffer(%p, %p, %" PRId64 ", %p, %ld)\n",
765 		       cs, bp, bn, addr, bcount);
766 #endif
767 	/*
768 	 * Determine which component bn falls in.
769 	 */
770 	cbn = bn;
771 	cboff = 0;
772 
773 	/*
774 	 * Serially concatenated
775 	 */
776 	if (cs->sc_ileave == 0) {
777 		daddr_t sblk;
778 
779 		sblk = 0;
780 		for (ccdisk = 0, ci = &cs->sc_cinfo[ccdisk];
781 		    cbn >= sblk + ci->ci_size;
782 		    ccdisk++, ci = &cs->sc_cinfo[ccdisk])
783 			sblk += ci->ci_size;
784 		cbn -= sblk;
785 	}
786 	/*
787 	 * Interleaved
788 	 */
789 	else {
790 		struct ccdiinfo *ii;
791 		int off;
792 
793 		cboff = cbn % cs->sc_ileave;
794 		cbn /= cs->sc_ileave;
795 		for (ii = cs->sc_itable; ii->ii_ndisk; ii++)
796 			if (ii->ii_startblk > cbn)
797 				break;
798 		ii--;
799 		off = cbn - ii->ii_startblk;
800 		if (ii->ii_ndisk == 1) {
801 			ccdisk = ii->ii_index[0];
802 			cbn = ii->ii_startoff + off;
803 		} else {
804 			ccdisk = ii->ii_index[off % ii->ii_ndisk];
805 			cbn = ii->ii_startoff + off / ii->ii_ndisk;
806 		}
807 		cbn *= cs->sc_ileave;
808 		ci = &cs->sc_cinfo[ccdisk];
809 	}
810 
811 	/*
812 	 * Fill in the component buf structure.
813 	 */
814 	cbp = CCD_GETBUF();
815 	if (cbp == NULL)
816 		return (NULL);
817 	BUF_INIT(&cbp->cb_buf);
818 	cbp->cb_buf.b_flags = bp->b_flags | B_CALL;
819 	cbp->cb_buf.b_iodone = ccdiodone;
820 	cbp->cb_buf.b_proc = bp->b_proc;
821 	cbp->cb_buf.b_dev = ci->ci_dev;		/* XXX */
822 	cbp->cb_buf.b_blkno = cbn + cboff;
823 	cbp->cb_buf.b_data = addr;
824 	cbp->cb_buf.b_vp = ci->ci_vp;
825 	if (cs->sc_ileave == 0)
826 		cbc = dbtob((u_int64_t)(ci->ci_size - cbn));
827 	else
828 		cbc = dbtob((u_int64_t)(cs->sc_ileave - cboff));
829 	cbp->cb_buf.b_bcount = cbc < bcount ? cbc : bcount;
830 
831 	/*
832 	 * context for ccdiodone
833 	 */
834 	cbp->cb_obp = bp;
835 	cbp->cb_sc = cs;
836 	cbp->cb_comp = ccdisk;
837 
838 #ifdef DEBUG
839 	if (ccddebug & CCDB_IO)
840 		printf(" dev 0x%x(u%lu): cbp %p bn %" PRId64 " addr %p"
841 		       " bcnt %ld\n",
842 		    ci->ci_dev, (unsigned long) (ci-cs->sc_cinfo), cbp,
843 		    cbp->cb_buf.b_blkno, cbp->cb_buf.b_data,
844 		    cbp->cb_buf.b_bcount);
845 #endif
846 
847 	return (cbp);
848 }
849 
850 static void
851 ccdintr(cs, bp)
852 	struct ccd_softc *cs;
853 	struct buf *bp;
854 {
855 
856 #ifdef DEBUG
857 	if (ccddebug & CCDB_FOLLOW)
858 		printf("ccdintr(%p, %p)\n", cs, bp);
859 #endif
860 	/*
861 	 * Request is done for better or worse, wakeup the top half.
862 	 */
863 	if (bp->b_flags & B_ERROR)
864 		bp->b_resid = bp->b_bcount;
865 	disk_unbusy(&cs->sc_dkdev, (bp->b_bcount - bp->b_resid),
866 	    (bp->b_flags & B_READ));
867 	biodone(bp);
868 }
869 
870 /*
871  * Called at interrupt time.
872  * Mark the component as done and if all components are done,
873  * take a ccd interrupt.
874  */
875 void
876 ccdiodone(vbp)
877 	struct buf *vbp;
878 {
879 	struct ccdbuf *cbp = (struct ccdbuf *) vbp;
880 	struct buf *bp = cbp->cb_obp;
881 	struct ccd_softc *cs = cbp->cb_sc;
882 	int count, s;
883 
884 	s = splbio();
885 #ifdef DEBUG
886 	if (ccddebug & CCDB_FOLLOW)
887 		printf("ccdiodone(%p)\n", cbp);
888 	if (ccddebug & CCDB_IO) {
889 		printf("ccdiodone: bp %p bcount %ld resid %ld\n",
890 		       bp, bp->b_bcount, bp->b_resid);
891 		printf(" dev 0x%x(u%d), cbp %p bn %" PRId64 " addr %p"
892 		       " bcnt %ld\n",
893 		       cbp->cb_buf.b_dev, cbp->cb_comp, cbp,
894 		       cbp->cb_buf.b_blkno, cbp->cb_buf.b_data,
895 		       cbp->cb_buf.b_bcount);
896 	}
897 #endif
898 
899 	if (cbp->cb_buf.b_flags & B_ERROR) {
900 		bp->b_flags |= B_ERROR;
901 		bp->b_error = cbp->cb_buf.b_error ?
902 		    cbp->cb_buf.b_error : EIO;
903 
904 		printf("%s: error %d on component %d\n",
905 		       cs->sc_xname, bp->b_error, cbp->cb_comp);
906 	}
907 	count = cbp->cb_buf.b_bcount;
908 	CCD_PUTBUF(cbp);
909 
910 	/*
911 	 * If all done, "interrupt".
912 	 */
913 	bp->b_resid -= count;
914 	if (bp->b_resid < 0)
915 		panic("ccdiodone: count");
916 	if (bp->b_resid == 0)
917 		ccdintr(cs, bp);
918 	splx(s);
919 }
920 
921 /* ARGSUSED */
922 int
923 ccdread(dev, uio, flags)
924 	dev_t dev;
925 	struct uio *uio;
926 	int flags;
927 {
928 	int unit = ccdunit(dev);
929 	struct ccd_softc *cs;
930 
931 #ifdef DEBUG
932 	if (ccddebug & CCDB_FOLLOW)
933 		printf("ccdread(0x%x, %p)\n", dev, uio);
934 #endif
935 	if (unit >= numccd)
936 		return (ENXIO);
937 	cs = &ccd_softc[unit];
938 
939 	if ((cs->sc_flags & CCDF_INITED) == 0)
940 		return (ENXIO);
941 
942 	/*
943 	 * XXX: It's not clear that using minphys() is completely safe,
944 	 * in particular, for raw I/O.  Underlying devices might have some
945 	 * non-obvious limits, because of the copy to user-space.
946 	 */
947 	return (physio(ccdstrategy, NULL, dev, B_READ, minphys, uio));
948 }
949 
950 /* ARGSUSED */
951 int
952 ccdwrite(dev, uio, flags)
953 	dev_t dev;
954 	struct uio *uio;
955 	int flags;
956 {
957 	int unit = ccdunit(dev);
958 	struct ccd_softc *cs;
959 
960 #ifdef DEBUG
961 	if (ccddebug & CCDB_FOLLOW)
962 		printf("ccdwrite(0x%x, %p)\n", dev, uio);
963 #endif
964 	if (unit >= numccd)
965 		return (ENXIO);
966 	cs = &ccd_softc[unit];
967 
968 	if ((cs->sc_flags & CCDF_INITED) == 0)
969 		return (ENXIO);
970 
971 	/*
972 	 * XXX: It's not clear that using minphys() is completely safe,
973 	 * in particular, for raw I/O.  Underlying devices might have some
974 	 * non-obvious limits, because of the copy to user-space.
975 	 */
976 	return (physio(ccdstrategy, NULL, dev, B_WRITE, minphys, uio));
977 }
978 
979 int
980 ccdioctl(dev, cmd, data, flag, p)
981 	dev_t dev;
982 	u_long cmd;
983 	caddr_t data;
984 	int flag;
985 	struct proc *p;
986 {
987 	int unit = ccdunit(dev);
988 	int s, i, j, lookedup = 0, error;
989 	int part, pmask;
990 	struct buf *bp;
991 	struct ccd_softc *cs;
992 	struct ccd_ioctl *ccio = (struct ccd_ioctl *)data;
993 	char **cpp;
994 	struct vnode **vpp;
995 #ifdef __HAVE_OLD_DISKLABEL
996 	struct disklabel newlabel;
997 #endif
998 
999 	if (unit >= numccd)
1000 		return (ENXIO);
1001 	cs = &ccd_softc[unit];
1002 
1003 	/* Must be open for writes for these commands... */
1004 	switch (cmd) {
1005 	case CCDIOCSET:
1006 	case CCDIOCCLR:
1007 	case DIOCSDINFO:
1008 	case DIOCWDINFO:
1009 #ifdef __HAVE_OLD_DISKLABEL
1010 	case ODIOCSDINFO:
1011 	case ODIOCWDINFO:
1012 #endif
1013 	case DIOCKLABEL:
1014 	case DIOCWLABEL:
1015 		if ((flag & FWRITE) == 0)
1016 			return (EBADF);
1017 	}
1018 
1019 	if ((error = lockmgr(&cs->sc_lock, LK_EXCLUSIVE, NULL)) != 0)
1020 		return (error);
1021 
1022 	/* Must be initialized for these... */
1023 	switch (cmd) {
1024 	case CCDIOCCLR:
1025 	case DIOCGDINFO:
1026 	case DIOCSDINFO:
1027 	case DIOCWDINFO:
1028 	case DIOCGPART:
1029 	case DIOCWLABEL:
1030 	case DIOCKLABEL:
1031 	case DIOCGDEFLABEL:
1032 #ifdef __HAVE_OLD_DISKLABEL
1033 	case ODIOCGDINFO:
1034 	case ODIOCSDINFO:
1035 	case ODIOCWDINFO:
1036 	case ODIOCGDEFLABEL:
1037 #endif
1038 		if ((cs->sc_flags & CCDF_INITED) == 0) {
1039 			error = ENXIO;
1040 			goto out;
1041 		}
1042 	}
1043 
1044 	switch (cmd) {
1045 	case CCDIOCSET:
1046 		if (cs->sc_flags & CCDF_INITED) {
1047 			error = EBUSY;
1048 			goto out;
1049 		}
1050 
1051 		/* Validate the flags. */
1052 		if ((ccio->ccio_flags & CCDF_USERMASK) != ccio->ccio_flags) {
1053 			error = EINVAL;
1054 			goto out;
1055 		}
1056 
1057 		if (ccio->ccio_ndisks > CCD_MAXNDISKS) {
1058 			error = EINVAL;
1059 			goto out;
1060 		}
1061 
1062 		/* Fill in some important bits. */
1063 		cs->sc_ileave = ccio->ccio_ileave;
1064 		cs->sc_nccdisks = ccio->ccio_ndisks;
1065 		cs->sc_flags = ccio->ccio_flags & CCDF_USERMASK;
1066 
1067 		/*
1068 		 * Allocate space for and copy in the array of
1069 		 * componet pathnames and device numbers.
1070 		 */
1071 		cpp = malloc(ccio->ccio_ndisks * sizeof(char *),
1072 		    M_DEVBUF, M_WAITOK);
1073 		vpp = malloc(ccio->ccio_ndisks * sizeof(struct vnode *),
1074 		    M_DEVBUF, M_WAITOK);
1075 
1076 		error = copyin(ccio->ccio_disks, cpp,
1077 		    ccio->ccio_ndisks * sizeof(char **));
1078 		if (error) {
1079 			free(vpp, M_DEVBUF);
1080 			free(cpp, M_DEVBUF);
1081 			goto out;
1082 		}
1083 
1084 #ifdef DEBUG
1085 		if (ccddebug & CCDB_INIT)
1086 			for (i = 0; i < ccio->ccio_ndisks; ++i)
1087 				printf("ccdioctl: component %d: 0x%p\n",
1088 				    i, cpp[i]);
1089 #endif
1090 
1091 		for (i = 0; i < ccio->ccio_ndisks; ++i) {
1092 #ifdef DEBUG
1093 			if (ccddebug & CCDB_INIT)
1094 				printf("ccdioctl: lookedup = %d\n", lookedup);
1095 #endif
1096 			if ((error = ccdlookup(cpp[i], p, &vpp[i])) != 0) {
1097 				for (j = 0; j < lookedup; ++j)
1098 					(void)vn_close(vpp[j], FREAD|FWRITE,
1099 					    p->p_ucred, p);
1100 				free(vpp, M_DEVBUF);
1101 				free(cpp, M_DEVBUF);
1102 				goto out;
1103 			}
1104 			++lookedup;
1105 		}
1106 
1107 		/*
1108 		 * Initialize the ccd.  Fills in the softc for us.
1109 		 */
1110 		if ((error = ccdinit(cs, cpp, vpp, p)) != 0) {
1111 			for (j = 0; j < lookedup; ++j)
1112 				(void)vn_close(vpp[j], FREAD|FWRITE,
1113 				    p->p_ucred, p);
1114 			free(vpp, M_DEVBUF);
1115 			free(cpp, M_DEVBUF);
1116 			goto out;
1117 		}
1118 
1119 		/* We can free the temporary variables now. */
1120 		free(vpp, M_DEVBUF);
1121 		free(cpp, M_DEVBUF);
1122 
1123 		/*
1124 		 * The ccd has been successfully initialized, so
1125 		 * we can place it into the array.  Don't try to
1126 		 * read the disklabel until the disk has been attached,
1127 		 * because space for the disklabel is allocated
1128 		 * in disk_attach();
1129 		 */
1130 		ccio->ccio_unit = unit;
1131 		ccio->ccio_size = cs->sc_size;
1132 
1133 		bufq_alloc(&cs->sc_bufq, BUFQ_FCFS);
1134 
1135 		/* Attach the disk. */
1136 		disk_attach(&cs->sc_dkdev);
1137 
1138 		/* Try and read the disklabel. */
1139 		ccdgetdisklabel(dev);
1140 		break;
1141 
1142 	case CCDIOCCLR:
1143 		/*
1144 		 * Don't unconfigure if any other partitions are open
1145 		 * or if both the character and block flavors of this
1146 		 * partition are open.
1147 		 */
1148 		part = DISKPART(dev);
1149 		pmask = (1 << part);
1150 		if ((cs->sc_dkdev.dk_openmask & ~pmask) ||
1151 		    ((cs->sc_dkdev.dk_bopenmask & pmask) &&
1152 		    (cs->sc_dkdev.dk_copenmask & pmask))) {
1153 			error = EBUSY;
1154 			goto out;
1155 		}
1156 
1157 		/* Kill off any queued buffers. */
1158 		s = splbio();
1159 		while ((bp = BUFQ_GET(&cs->sc_bufq)) != NULL) {
1160 			bp->b_error = EIO;
1161 			bp->b_flags |= B_ERROR;
1162 			bp->b_resid = bp->b_bcount;
1163 			biodone(bp);
1164 		}
1165 		splx(s);
1166 
1167 		bufq_free(&cs->sc_bufq);
1168 
1169 		/*
1170 		 * Free ccd_softc information and clear entry.
1171 		 */
1172 
1173 		/* Close the components and free their pathnames. */
1174 		for (i = 0; i < cs->sc_nccdisks; ++i) {
1175 			/*
1176 			 * XXX: this close could potentially fail and
1177 			 * cause Bad Things.  Maybe we need to force
1178 			 * the close to happen?
1179 			 */
1180 #ifdef DEBUG
1181 			if (ccddebug & CCDB_VNODE)
1182 				vprint("CCDIOCCLR: vnode info",
1183 				    cs->sc_cinfo[i].ci_vp);
1184 #endif
1185 			(void)vn_close(cs->sc_cinfo[i].ci_vp, FREAD|FWRITE,
1186 			    p->p_ucred, p);
1187 			free(cs->sc_cinfo[i].ci_path, M_DEVBUF);
1188 		}
1189 
1190 		/* Free interleave index. */
1191 		for (i = 0; cs->sc_itable[i].ii_ndisk; ++i)
1192 			free(cs->sc_itable[i].ii_index, M_DEVBUF);
1193 
1194 		/* Free component info and interleave table. */
1195 		free(cs->sc_cinfo, M_DEVBUF);
1196 		free(cs->sc_itable, M_DEVBUF);
1197 		cs->sc_flags &= ~(CCDF_INITED|CCDF_VLABEL);
1198 
1199 		/* Detatch the disk. */
1200 		disk_detach(&cs->sc_dkdev);
1201 		break;
1202 
1203 	case DIOCGDINFO:
1204 		*(struct disklabel *)data = *(cs->sc_dkdev.dk_label);
1205 		break;
1206 #ifdef __HAVE_OLD_DISKLABEL
1207 	case ODIOCGDINFO:
1208 		newlabel = *(cs->sc_dkdev.dk_label);
1209 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1210 			return ENOTTY;
1211 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1212 		break;
1213 #endif
1214 
1215 	case DIOCGPART:
1216 		((struct partinfo *)data)->disklab = cs->sc_dkdev.dk_label;
1217 		((struct partinfo *)data)->part =
1218 		    &cs->sc_dkdev.dk_label->d_partitions[DISKPART(dev)];
1219 		break;
1220 
1221 	case DIOCWDINFO:
1222 	case DIOCSDINFO:
1223 #ifdef __HAVE_OLD_DISKLABEL
1224 	case ODIOCWDINFO:
1225 	case ODIOCSDINFO:
1226 #endif
1227 	{
1228 		struct disklabel *lp;
1229 #ifdef __HAVE_OLD_DISKLABEL
1230 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
1231 			memset(&newlabel, 0, sizeof newlabel);
1232 			memcpy(&newlabel, data, sizeof (struct olddisklabel));
1233 			lp = &newlabel;
1234 		} else
1235 #endif
1236 		lp = (struct disklabel *)data;
1237 
1238 		cs->sc_flags |= CCDF_LABELLING;
1239 
1240 		error = setdisklabel(cs->sc_dkdev.dk_label,
1241 		    lp, 0, cs->sc_dkdev.dk_cpulabel);
1242 		if (error == 0) {
1243 			if (cmd == DIOCWDINFO
1244 #ifdef __HAVE_OLD_DISKLABEL
1245 			    || cmd == ODIOCWDINFO
1246 #endif
1247 			   )
1248 				error = writedisklabel(CCDLABELDEV(dev),
1249 				    ccdstrategy, cs->sc_dkdev.dk_label,
1250 				    cs->sc_dkdev.dk_cpulabel);
1251 		}
1252 
1253 		cs->sc_flags &= ~CCDF_LABELLING;
1254 		break;
1255 	}
1256 
1257 	case DIOCKLABEL:
1258 		if (*(int *)data != 0)
1259 			cs->sc_flags |= CCDF_KLABEL;
1260 		else
1261 			cs->sc_flags &= ~CCDF_KLABEL;
1262 		break;
1263 
1264 	case DIOCWLABEL:
1265 		if (*(int *)data != 0)
1266 			cs->sc_flags |= CCDF_WLABEL;
1267 		else
1268 			cs->sc_flags &= ~CCDF_WLABEL;
1269 		break;
1270 
1271 	case DIOCGDEFLABEL:
1272 		ccdgetdefaultlabel(cs, (struct disklabel *)data);
1273 		break;
1274 
1275 #ifdef __HAVE_OLD_DISKLABEL
1276 	case ODIOCGDEFLABEL:
1277 		ccdgetdefaultlabel(cs, &newlabel);
1278 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1279 			return ENOTTY;
1280 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1281 		break;
1282 #endif
1283 
1284 	default:
1285 		error = ENOTTY;
1286 	}
1287 
1288  out:
1289 	(void) lockmgr(&cs->sc_lock, LK_RELEASE, NULL);
1290 	return (error);
1291 }
1292 
1293 int
1294 ccdsize(dev)
1295 	dev_t dev;
1296 {
1297 	struct ccd_softc *cs;
1298 	struct disklabel *lp;
1299 	int part, unit, omask, size;
1300 
1301 	unit = ccdunit(dev);
1302 	if (unit >= numccd)
1303 		return (-1);
1304 	cs = &ccd_softc[unit];
1305 
1306 	if ((cs->sc_flags & CCDF_INITED) == 0)
1307 		return (-1);
1308 
1309 	part = DISKPART(dev);
1310 	omask = cs->sc_dkdev.dk_openmask & (1 << part);
1311 	lp = cs->sc_dkdev.dk_label;
1312 
1313 	if (omask == 0 && ccdopen(dev, 0, S_IFBLK, curproc))
1314 		return (-1);
1315 
1316 	if (lp->d_partitions[part].p_fstype != FS_SWAP)
1317 		size = -1;
1318 	else
1319 		size = lp->d_partitions[part].p_size *
1320 		    (lp->d_secsize / DEV_BSIZE);
1321 
1322 	if (omask == 0 && ccdclose(dev, 0, S_IFBLK, curproc))
1323 		return (-1);
1324 
1325 	return (size);
1326 }
1327 
1328 int
1329 ccddump(dev, blkno, va, size)
1330 	dev_t dev;
1331 	daddr_t blkno;
1332 	caddr_t va;
1333 	size_t size;
1334 {
1335 
1336 	/* Not implemented. */
1337 	return ENXIO;
1338 }
1339 
1340 /*
1341  * Lookup the provided name in the filesystem.  If the file exists,
1342  * is a valid block device, and isn't being used by anyone else,
1343  * set *vpp to the file's vnode.
1344  */
1345 static int
1346 ccdlookup(path, p, vpp)
1347 	char *path;
1348 	struct proc *p;
1349 	struct vnode **vpp;	/* result */
1350 {
1351 	struct nameidata nd;
1352 	struct vnode *vp;
1353 	struct vattr va;
1354 	int error;
1355 
1356 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, path, p);
1357 	if ((error = vn_open(&nd, FREAD|FWRITE, 0)) != 0) {
1358 #ifdef DEBUG
1359 		if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
1360 			printf("ccdlookup: vn_open error = %d\n", error);
1361 #endif
1362 		return (error);
1363 	}
1364 	vp = nd.ni_vp;
1365 
1366 	if (vp->v_usecount > 1) {
1367 		VOP_UNLOCK(vp, 0);
1368 		(void)vn_close(vp, FREAD|FWRITE, p->p_ucred, p);
1369 		return (EBUSY);
1370 	}
1371 
1372 	if ((error = VOP_GETATTR(vp, &va, p->p_ucred, p)) != 0) {
1373 #ifdef DEBUG
1374 		if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
1375 			printf("ccdlookup: getattr error = %d\n", error);
1376 #endif
1377 		VOP_UNLOCK(vp, 0);
1378 		(void)vn_close(vp, FREAD|FWRITE, p->p_ucred, p);
1379 		return (error);
1380 	}
1381 
1382 	/* XXX: eventually we should handle VREG, too. */
1383 	if (va.va_type != VBLK) {
1384 		VOP_UNLOCK(vp, 0);
1385 		(void)vn_close(vp, FREAD|FWRITE, p->p_ucred, p);
1386 		return (ENOTBLK);
1387 	}
1388 
1389 #ifdef DEBUG
1390 	if (ccddebug & CCDB_VNODE)
1391 		vprint("ccdlookup: vnode info", vp);
1392 #endif
1393 
1394 	VOP_UNLOCK(vp, 0);
1395 	*vpp = vp;
1396 	return (0);
1397 }
1398 
1399 static void
1400 ccdgetdefaultlabel(cs, lp)
1401 	struct ccd_softc *cs;
1402 	struct disklabel *lp;
1403 {
1404 	struct ccdgeom *ccg = &cs->sc_geom;
1405 
1406 	memset(lp, 0, sizeof(*lp));
1407 
1408 	lp->d_secperunit = cs->sc_size;
1409 	lp->d_secsize = ccg->ccg_secsize;
1410 	lp->d_nsectors = ccg->ccg_nsectors;
1411 	lp->d_ntracks = ccg->ccg_ntracks;
1412 	lp->d_ncylinders = ccg->ccg_ncylinders;
1413 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1414 
1415 	strncpy(lp->d_typename, "ccd", sizeof(lp->d_typename));
1416 	lp->d_type = DTYPE_CCD;
1417 	strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
1418 	lp->d_rpm = 3600;
1419 	lp->d_interleave = 1;
1420 	lp->d_flags = 0;
1421 
1422 	lp->d_partitions[RAW_PART].p_offset = 0;
1423 	lp->d_partitions[RAW_PART].p_size = cs->sc_size;
1424 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
1425 	lp->d_npartitions = RAW_PART + 1;
1426 
1427 	lp->d_magic = DISKMAGIC;
1428 	lp->d_magic2 = DISKMAGIC;
1429 	lp->d_checksum = dkcksum(cs->sc_dkdev.dk_label);
1430 }
1431 
1432 /*
1433  * Read the disklabel from the ccd.  If one is not present, fake one
1434  * up.
1435  */
1436 static void
1437 ccdgetdisklabel(dev)
1438 	dev_t dev;
1439 {
1440 	int unit = ccdunit(dev);
1441 	struct ccd_softc *cs = &ccd_softc[unit];
1442 	const char *errstring;
1443 	struct disklabel *lp = cs->sc_dkdev.dk_label;
1444 	struct cpu_disklabel *clp = cs->sc_dkdev.dk_cpulabel;
1445 
1446 	memset(clp, 0, sizeof(*clp));
1447 
1448 	ccdgetdefaultlabel(cs, lp);
1449 
1450 	/*
1451 	 * Call the generic disklabel extraction routine.
1452 	 */
1453 	errstring = readdisklabel(CCDLABELDEV(dev), ccdstrategy,
1454 	    cs->sc_dkdev.dk_label, cs->sc_dkdev.dk_cpulabel);
1455 	if (errstring)
1456 		ccdmakedisklabel(cs);
1457 	else {
1458 		int i;
1459 		struct partition *pp;
1460 
1461 		/*
1462 		 * Sanity check whether the found disklabel is valid.
1463 		 *
1464 		 * This is necessary since total size of ccd may vary
1465 		 * when an interleave is changed even though exactly
1466 		 * same componets are used, and old disklabel may used
1467 		 * if that is found.
1468 		 */
1469 		if (lp->d_secperunit != cs->sc_size)
1470 			printf("WARNING: %s: "
1471 			    "total sector size in disklabel (%d) != "
1472 			    "the size of ccd (%lu)\n", cs->sc_xname,
1473 			    lp->d_secperunit, (u_long)cs->sc_size);
1474 		for (i = 0; i < lp->d_npartitions; i++) {
1475 			pp = &lp->d_partitions[i];
1476 			if (pp->p_offset + pp->p_size > cs->sc_size)
1477 				printf("WARNING: %s: end of partition `%c' "
1478 				    "exceeds the size of ccd (%lu)\n",
1479 				    cs->sc_xname, 'a' + i, (u_long)cs->sc_size);
1480 		}
1481 	}
1482 
1483 #ifdef DEBUG
1484 	/* It's actually extremely common to have unlabeled ccds. */
1485 	if (ccddebug & CCDB_LABEL)
1486 		if (errstring != NULL)
1487 			printf("%s: %s\n", cs->sc_xname, errstring);
1488 #endif
1489 
1490 	/* In-core label now valid. */
1491 	cs->sc_flags |= CCDF_VLABEL;
1492 }
1493 
1494 /*
1495  * Take care of things one might want to take care of in the event
1496  * that a disklabel isn't present.
1497  */
1498 static void
1499 ccdmakedisklabel(cs)
1500 	struct ccd_softc *cs;
1501 {
1502 	struct disklabel *lp = cs->sc_dkdev.dk_label;
1503 
1504 	/*
1505 	 * For historical reasons, if there's no disklabel present
1506 	 * the raw partition must be marked FS_BSDFFS.
1507 	 */
1508 	lp->d_partitions[RAW_PART].p_fstype = FS_BSDFFS;
1509 
1510 	strncpy(lp->d_packname, "default label", sizeof(lp->d_packname));
1511 
1512 	lp->d_checksum = dkcksum(lp);
1513 }
1514 
1515 #ifdef DEBUG
1516 static void
1517 printiinfo(ii)
1518 	struct ccdiinfo *ii;
1519 {
1520 	int ix, i;
1521 
1522 	for (ix = 0; ii->ii_ndisk; ix++, ii++) {
1523 		printf(" itab[%d]: #dk %d sblk %" PRId64 " soff %" PRId64,
1524 		    ix, ii->ii_ndisk, ii->ii_startblk, ii->ii_startoff);
1525 		for (i = 0; i < ii->ii_ndisk; i++)
1526 			printf(" %d", ii->ii_index[i]);
1527 		printf("\n");
1528 	}
1529 }
1530 #endif
1531