xref: /netbsd-src/sys/dev/ccd.c (revision ce2c90c7c172d95d2402a5b3d96d8f8e6d138a21)
1 /*	$NetBSD: ccd.c,v 1.115 2006/10/12 01:30:50 christos Exp $	*/
2 
3 /*-
4  * Copyright (c) 1996, 1997, 1998, 1999 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Jason R. Thorpe.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *        This product includes software developed by the NetBSD
21  *        Foundation, Inc. and its contributors.
22  * 4. Neither the name of The NetBSD Foundation nor the names of its
23  *    contributors may be used to endorse or promote products derived
24  *    from this software without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
27  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
28  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
29  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
30  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36  * POSSIBILITY OF SUCH DAMAGE.
37  */
38 
39 /*
40  * Copyright (c) 1990, 1993
41  *	The Regents of the University of California.  All rights reserved.
42  *
43  * This code is derived from software contributed to Berkeley by
44  * the Systems Programming Group of the University of Utah Computer
45  * Science Department.
46  *
47  * Redistribution and use in source and binary forms, with or without
48  * modification, are permitted provided that the following conditions
49  * are met:
50  * 1. Redistributions of source code must retain the above copyright
51  *    notice, this list of conditions and the following disclaimer.
52  * 2. Redistributions in binary form must reproduce the above copyright
53  *    notice, this list of conditions and the following disclaimer in the
54  *    documentation and/or other materials provided with the distribution.
55  * 3. Neither the name of the University nor the names of its contributors
56  *    may be used to endorse or promote products derived from this software
57  *    without specific prior written permission.
58  *
59  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
60  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
61  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
62  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
63  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
64  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
65  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
66  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
67  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
68  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
69  * SUCH DAMAGE.
70  *
71  * from: Utah $Hdr: cd.c 1.6 90/11/28$
72  *
73  *	@(#)cd.c	8.2 (Berkeley) 11/16/93
74  */
75 
76 /*
77  * Copyright (c) 1988 University of Utah.
78  *
79  * This code is derived from software contributed to Berkeley by
80  * the Systems Programming Group of the University of Utah Computer
81  * Science Department.
82  *
83  * Redistribution and use in source and binary forms, with or without
84  * modification, are permitted provided that the following conditions
85  * are met:
86  * 1. Redistributions of source code must retain the above copyright
87  *    notice, this list of conditions and the following disclaimer.
88  * 2. Redistributions in binary form must reproduce the above copyright
89  *    notice, this list of conditions and the following disclaimer in the
90  *    documentation and/or other materials provided with the distribution.
91  * 3. All advertising materials mentioning features or use of this software
92  *    must display the following acknowledgement:
93  *	This product includes software developed by the University of
94  *	California, Berkeley and its contributors.
95  * 4. Neither the name of the University nor the names of its contributors
96  *    may be used to endorse or promote products derived from this software
97  *    without specific prior written permission.
98  *
99  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
100  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
101  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
102  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
103  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
104  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
105  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
106  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
107  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
108  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
109  * SUCH DAMAGE.
110  *
111  * from: Utah $Hdr: cd.c 1.6 90/11/28$
112  *
113  *	@(#)cd.c	8.2 (Berkeley) 11/16/93
114  */
115 
116 /*
117  * "Concatenated" disk driver.
118  *
119  * Dynamic configuration and disklabel support by:
120  *	Jason R. Thorpe <thorpej@nas.nasa.gov>
121  *	Numerical Aerodynamic Simulation Facility
122  *	Mail Stop 258-6
123  *	NASA Ames Research Center
124  *	Moffett Field, CA 94035
125  */
126 
127 #include <sys/cdefs.h>
128 __KERNEL_RCSID(0, "$NetBSD: ccd.c,v 1.115 2006/10/12 01:30:50 christos Exp $");
129 
130 #include <sys/param.h>
131 #include <sys/systm.h>
132 #include <sys/proc.h>
133 #include <sys/errno.h>
134 #include <sys/buf.h>
135 #include <sys/bufq.h>
136 #include <sys/malloc.h>
137 #include <sys/pool.h>
138 #include <sys/namei.h>
139 #include <sys/stat.h>
140 #include <sys/ioctl.h>
141 #include <sys/disklabel.h>
142 #include <sys/device.h>
143 #include <sys/disk.h>
144 #include <sys/syslog.h>
145 #include <sys/fcntl.h>
146 #include <sys/vnode.h>
147 #include <sys/conf.h>
148 #include <sys/lock.h>
149 #include <sys/queue.h>
150 #include <sys/kauth.h>
151 
152 #include <dev/ccdvar.h>
153 #include <dev/dkvar.h>
154 
155 #if defined(CCDDEBUG) && !defined(DEBUG)
156 #define DEBUG
157 #endif
158 
159 #ifdef DEBUG
160 #define CCDB_FOLLOW	0x01
161 #define CCDB_INIT	0x02
162 #define CCDB_IO		0x04
163 #define CCDB_LABEL	0x08
164 #define CCDB_VNODE	0x10
165 int ccddebug = 0x00;
166 #endif
167 
168 #define	ccdunit(x)	DISKUNIT(x)
169 
170 struct ccdbuf {
171 	struct buf	cb_buf;		/* new I/O buf */
172 	struct buf	*cb_obp;	/* ptr. to original I/O buf */
173 	struct ccd_softc *cb_sc;	/* pointer to ccd softc */
174 	int		cb_comp;	/* target component */
175 	SIMPLEQ_ENTRY(ccdbuf) cb_q;	/* fifo of component buffers */
176 };
177 
178 /* component buffer pool */
179 static struct pool ccd_cbufpool;
180 
181 #define	CCD_GETBUF()		pool_get(&ccd_cbufpool, PR_NOWAIT)
182 #define	CCD_PUTBUF(cbp)		pool_put(&ccd_cbufpool, cbp)
183 
184 #define CCDLABELDEV(dev)	\
185 	(MAKEDISKDEV(major((dev)), ccdunit((dev)), RAW_PART))
186 
187 /* called by main() at boot time */
188 void	ccdattach(int);
189 
190 /* called by biodone() at interrupt time */
191 static void	ccdiodone(struct buf *);
192 
193 static void	ccdstart(struct ccd_softc *);
194 static void	ccdinterleave(struct ccd_softc *);
195 static void	ccdintr(struct ccd_softc *, struct buf *);
196 static int	ccdinit(struct ccd_softc *, char **, struct vnode **,
197 		    struct lwp *);
198 static struct ccdbuf *ccdbuffer(struct ccd_softc *, struct buf *,
199 		    daddr_t, caddr_t, long);
200 static void	ccdgetdefaultlabel(struct ccd_softc *, struct disklabel *);
201 static void	ccdgetdisklabel(dev_t);
202 static void	ccdmakedisklabel(struct ccd_softc *);
203 
204 static dev_type_open(ccdopen);
205 static dev_type_close(ccdclose);
206 static dev_type_read(ccdread);
207 static dev_type_write(ccdwrite);
208 static dev_type_ioctl(ccdioctl);
209 static dev_type_strategy(ccdstrategy);
210 static dev_type_dump(ccddump);
211 static dev_type_size(ccdsize);
212 
213 const struct bdevsw ccd_bdevsw = {
214 	ccdopen, ccdclose, ccdstrategy, ccdioctl, ccddump, ccdsize, D_DISK
215 };
216 
217 const struct cdevsw ccd_cdevsw = {
218 	ccdopen, ccdclose, ccdread, ccdwrite, ccdioctl,
219 	nostop, notty, nopoll, nommap, nokqfilter, D_DISK
220 };
221 
222 #ifdef DEBUG
223 static	void printiinfo(struct ccdiinfo *);
224 #endif
225 
226 /* Publically visible for the benefit of libkvm and ccdconfig(8). */
227 struct ccd_softc 	*ccd_softc;
228 const int		ccd_softc_elemsize = sizeof(struct ccd_softc);
229 int			numccd = 0;
230 
231 /*
232  * Called by main() during pseudo-device attachment.  All we need
233  * to do is allocate enough space for devices to be configured later.
234  */
235 void
236 ccdattach(int num)
237 {
238 	struct ccd_softc *cs;
239 	int i;
240 
241 	if (num <= 0) {
242 #ifdef DIAGNOSTIC
243 		panic("ccdattach: count <= 0");
244 #endif
245 		return;
246 	}
247 
248 	ccd_softc = (struct ccd_softc *)malloc(num * ccd_softc_elemsize,
249 	    M_DEVBUF, M_NOWAIT|M_ZERO);
250 	if (ccd_softc == NULL) {
251 		printf("WARNING: no memory for concatenated disks\n");
252 		return;
253 	}
254 	numccd = num;
255 
256 	/* Initialize the component buffer pool. */
257 	pool_init(&ccd_cbufpool, sizeof(struct ccdbuf), 0,
258 	    0, 0, "ccdpl", NULL);
259 
260 	/* Initialize per-softc structures. */
261 	for (i = 0; i < num; i++) {
262 		cs = &ccd_softc[i];
263 		snprintf(cs->sc_xname, sizeof(cs->sc_xname), "ccd%d", i);
264 		cs->sc_dkdev.dk_name = cs->sc_xname;	/* XXX */
265 		lockinit(&cs->sc_lock, PRIBIO, "ccdlk", 0, 0);
266 		pseudo_disk_init(&cs->sc_dkdev);
267 	}
268 }
269 
270 static int
271 ccdinit(struct ccd_softc *cs, char **cpaths, struct vnode **vpp,
272     struct lwp *l)
273 {
274 	struct ccdcinfo *ci = NULL;
275 	size_t size;
276 	int ix;
277 	struct vattr va;
278 	size_t minsize;
279 	int maxsecsize;
280 	struct partinfo dpart;
281 	struct ccdgeom *ccg = &cs->sc_geom;
282 	char *tmppath;
283 	int error, path_alloced;
284 
285 #ifdef DEBUG
286 	if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
287 		printf("%s: ccdinit\n", cs->sc_xname);
288 #endif
289 
290 	/* Allocate space for the component info. */
291 	cs->sc_cinfo = malloc(cs->sc_nccdisks * sizeof(struct ccdcinfo),
292 	    M_DEVBUF, M_WAITOK);
293 
294 	tmppath = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
295 
296 	cs->sc_size = 0;
297 
298 	/*
299 	 * Verify that each component piece exists and record
300 	 * relevant information about it.
301 	 */
302 	maxsecsize = 0;
303 	minsize = 0;
304 	for (ix = 0, path_alloced = 0; ix < cs->sc_nccdisks; ix++) {
305 		ci = &cs->sc_cinfo[ix];
306 		ci->ci_vp = vpp[ix];
307 
308 		/*
309 		 * Copy in the pathname of the component.
310 		 */
311 		memset(tmppath, 0, sizeof(tmppath));	/* sanity */
312 		error = copyinstr(cpaths[ix], tmppath,
313 		    MAXPATHLEN, &ci->ci_pathlen);
314 		if (error) {
315 #ifdef DEBUG
316 			if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
317 				printf("%s: can't copy path, error = %d\n",
318 				    cs->sc_xname, error);
319 #endif
320 			goto out;
321 		}
322 		ci->ci_path = malloc(ci->ci_pathlen, M_DEVBUF, M_WAITOK);
323 		memcpy(ci->ci_path, tmppath, ci->ci_pathlen);
324 		path_alloced++;
325 
326 		/*
327 		 * XXX: Cache the component's dev_t.
328 		 */
329 		if ((error = VOP_GETATTR(vpp[ix], &va, l->l_cred, l)) != 0) {
330 #ifdef DEBUG
331 			if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
332 				printf("%s: %s: getattr failed %s = %d\n",
333 				    cs->sc_xname, ci->ci_path,
334 				    "error", error);
335 #endif
336 			goto out;
337 		}
338 		ci->ci_dev = va.va_rdev;
339 
340 		/*
341 		 * Get partition information for the component.
342 		 */
343 		error = VOP_IOCTL(vpp[ix], DIOCGPART, &dpart,
344 		    FREAD, l->l_cred, l);
345 		if (error) {
346 #ifdef DEBUG
347 			if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
348 				 printf("%s: %s: ioctl failed, error = %d\n",
349 				     cs->sc_xname, ci->ci_path, error);
350 #endif
351 			goto out;
352 		}
353 
354 /*
355  * This diagnostic test is disabled (for now?) since not all port supports
356  * on-disk BSD disklabel.
357  */
358 #if 0 /* def DIAGNOSTIC */
359 		/* Check fstype field of component. */
360 		if (dpart.part->p_fstype != FS_CCD)
361 			printf("%s: WARNING: %s: fstype %d != FS_CCD\n",
362 			    cs->sc_xname, ci->ci_path, dpart.part->p_fstype);
363 #endif
364 
365 		/*
366 		 * Calculate the size, truncating to an interleave
367 		 * boundary if necessary.
368 		 */
369 		maxsecsize =
370 		    ((dpart.disklab->d_secsize > maxsecsize) ?
371 		    dpart.disklab->d_secsize : maxsecsize);
372 		size = dpart.part->p_size;
373 		if (cs->sc_ileave > 1)
374 			size -= size % cs->sc_ileave;
375 
376 		if (size == 0) {
377 #ifdef DEBUG
378 			if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
379 				printf("%s: %s: size == 0\n",
380 				    cs->sc_xname, ci->ci_path);
381 #endif
382 			error = ENODEV;
383 			goto out;
384 		}
385 
386 		if (minsize == 0 || size < minsize)
387 			minsize = size;
388 		ci->ci_size = size;
389 		cs->sc_size += size;
390 	}
391 
392 	/*
393 	 * Don't allow the interleave to be smaller than
394 	 * the biggest component sector.
395 	 */
396 	if ((cs->sc_ileave > 0) &&
397 	    (cs->sc_ileave < (maxsecsize / DEV_BSIZE))) {
398 #ifdef DEBUG
399 		if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
400 			printf("%s: interleave must be at least %d\n",
401 			    cs->sc_xname, (maxsecsize / DEV_BSIZE));
402 #endif
403 		error = EINVAL;
404 		goto out;
405 	}
406 
407 	/*
408 	 * If uniform interleave is desired set all sizes to that of
409 	 * the smallest component.
410 	 */
411 	if (cs->sc_flags & CCDF_UNIFORM) {
412 		for (ci = cs->sc_cinfo;
413 		     ci < &cs->sc_cinfo[cs->sc_nccdisks]; ci++)
414 			ci->ci_size = minsize;
415 
416 		cs->sc_size = cs->sc_nccdisks * minsize;
417 	}
418 
419 	/*
420 	 * Construct the interleave table.
421 	 */
422 	ccdinterleave(cs);
423 
424 	/*
425 	 * Create pseudo-geometry based on 1MB cylinders.  It's
426 	 * pretty close.
427 	 */
428 	ccg->ccg_secsize = DEV_BSIZE;
429 	ccg->ccg_ntracks = 1;
430 	ccg->ccg_nsectors = 1024 * (1024 / ccg->ccg_secsize);
431 	ccg->ccg_ncylinders = cs->sc_size / ccg->ccg_nsectors;
432 
433 	cs->sc_flags |= CCDF_INITED;
434 
435 	free(tmppath, M_TEMP);
436 
437 	return (0);
438 
439  out:
440 	for (ix = 0; ix < path_alloced; ix++)
441 		free(cs->sc_cinfo[ix].ci_path, M_DEVBUF);
442 	free(cs->sc_cinfo, M_DEVBUF);
443 	free(tmppath, M_TEMP);
444 	return (error);
445 }
446 
447 static void
448 ccdinterleave(struct ccd_softc *cs)
449 {
450 	struct ccdcinfo *ci, *smallci;
451 	struct ccdiinfo *ii;
452 	daddr_t bn, lbn;
453 	int ix;
454 	u_long size;
455 
456 #ifdef DEBUG
457 	if (ccddebug & CCDB_INIT)
458 		printf("ccdinterleave(%p): ileave %d\n", cs, cs->sc_ileave);
459 #endif
460 	/*
461 	 * Allocate an interleave table.
462 	 * Chances are this is too big, but we don't care.
463 	 */
464 	size = (cs->sc_nccdisks + 1) * sizeof(struct ccdiinfo);
465 	cs->sc_itable = (struct ccdiinfo *)malloc(size, M_DEVBUF,
466 	    M_WAITOK|M_ZERO);
467 
468 	/*
469 	 * Trivial case: no interleave (actually interleave of disk size).
470 	 * Each table entry represents a single component in its entirety.
471 	 */
472 	if (cs->sc_ileave == 0) {
473 		bn = 0;
474 		ii = cs->sc_itable;
475 
476 		for (ix = 0; ix < cs->sc_nccdisks; ix++) {
477 			/* Allocate space for ii_index. */
478 			ii->ii_index = malloc(sizeof(int), M_DEVBUF, M_WAITOK);
479 			ii->ii_ndisk = 1;
480 			ii->ii_startblk = bn;
481 			ii->ii_startoff = 0;
482 			ii->ii_index[0] = ix;
483 			bn += cs->sc_cinfo[ix].ci_size;
484 			ii++;
485 		}
486 		ii->ii_ndisk = 0;
487 #ifdef DEBUG
488 		if (ccddebug & CCDB_INIT)
489 			printiinfo(cs->sc_itable);
490 #endif
491 		return;
492 	}
493 
494 	/*
495 	 * The following isn't fast or pretty; it doesn't have to be.
496 	 */
497 	size = 0;
498 	bn = lbn = 0;
499 	for (ii = cs->sc_itable; ; ii++) {
500 		/* Allocate space for ii_index. */
501 		ii->ii_index = malloc((sizeof(int) * cs->sc_nccdisks),
502 		    M_DEVBUF, M_WAITOK);
503 
504 		/*
505 		 * Locate the smallest of the remaining components
506 		 */
507 		smallci = NULL;
508 		for (ci = cs->sc_cinfo;
509 		     ci < &cs->sc_cinfo[cs->sc_nccdisks]; ci++)
510 			if (ci->ci_size > size &&
511 			    (smallci == NULL ||
512 			     ci->ci_size < smallci->ci_size))
513 				smallci = ci;
514 
515 		/*
516 		 * Nobody left, all done
517 		 */
518 		if (smallci == NULL) {
519 			ii->ii_ndisk = 0;
520 			break;
521 		}
522 
523 		/*
524 		 * Record starting logical block and component offset
525 		 */
526 		ii->ii_startblk = bn / cs->sc_ileave;
527 		ii->ii_startoff = lbn;
528 
529 		/*
530 		 * Determine how many disks take part in this interleave
531 		 * and record their indices.
532 		 */
533 		ix = 0;
534 		for (ci = cs->sc_cinfo;
535 		     ci < &cs->sc_cinfo[cs->sc_nccdisks]; ci++)
536 			if (ci->ci_size >= smallci->ci_size)
537 				ii->ii_index[ix++] = ci - cs->sc_cinfo;
538 		ii->ii_ndisk = ix;
539 		bn += ix * (smallci->ci_size - size);
540 		lbn = smallci->ci_size / cs->sc_ileave;
541 		size = smallci->ci_size;
542 	}
543 #ifdef DEBUG
544 	if (ccddebug & CCDB_INIT)
545 		printiinfo(cs->sc_itable);
546 #endif
547 }
548 
549 /* ARGSUSED */
550 static int
551 ccdopen(dev_t dev, int flags __unused, int fmt, struct lwp *l __unused)
552 {
553 	int unit = ccdunit(dev);
554 	struct ccd_softc *cs;
555 	struct disklabel *lp;
556 	int error = 0, part, pmask;
557 
558 #ifdef DEBUG
559 	if (ccddebug & CCDB_FOLLOW)
560 		printf("ccdopen(0x%x, 0x%x)\n", dev, flags);
561 #endif
562 	if (unit >= numccd)
563 		return (ENXIO);
564 	cs = &ccd_softc[unit];
565 
566 	if ((error = lockmgr(&cs->sc_lock, LK_EXCLUSIVE, NULL)) != 0)
567 		return (error);
568 
569 	lp = cs->sc_dkdev.dk_label;
570 
571 	part = DISKPART(dev);
572 	pmask = (1 << part);
573 
574 	/*
575 	 * If we're initialized, check to see if there are any other
576 	 * open partitions.  If not, then it's safe to update
577 	 * the in-core disklabel.  Only read the disklabel if it is
578 	 * not already valid.
579 	 */
580 	if ((cs->sc_flags & (CCDF_INITED|CCDF_VLABEL)) == CCDF_INITED &&
581 	    cs->sc_dkdev.dk_openmask == 0)
582 		ccdgetdisklabel(dev);
583 
584 	/* Check that the partition exists. */
585 	if (part != RAW_PART) {
586 		if (((cs->sc_flags & CCDF_INITED) == 0) ||
587 		    ((part >= lp->d_npartitions) ||
588 		     (lp->d_partitions[part].p_fstype == FS_UNUSED))) {
589 			error = ENXIO;
590 			goto done;
591 		}
592 	}
593 
594 	/* Prevent our unit from being unconfigured while open. */
595 	switch (fmt) {
596 	case S_IFCHR:
597 		cs->sc_dkdev.dk_copenmask |= pmask;
598 		break;
599 
600 	case S_IFBLK:
601 		cs->sc_dkdev.dk_bopenmask |= pmask;
602 		break;
603 	}
604 	cs->sc_dkdev.dk_openmask =
605 	    cs->sc_dkdev.dk_copenmask | cs->sc_dkdev.dk_bopenmask;
606 
607  done:
608 	(void) lockmgr(&cs->sc_lock, LK_RELEASE, NULL);
609 	return (error);
610 }
611 
612 /* ARGSUSED */
613 static int
614 ccdclose(dev_t dev, int flags __unused, int fmt, struct lwp *l __unused)
615 {
616 	int unit = ccdunit(dev);
617 	struct ccd_softc *cs;
618 	int error = 0, part;
619 
620 #ifdef DEBUG
621 	if (ccddebug & CCDB_FOLLOW)
622 		printf("ccdclose(0x%x, 0x%x)\n", dev, flags);
623 #endif
624 
625 	if (unit >= numccd)
626 		return (ENXIO);
627 	cs = &ccd_softc[unit];
628 
629 	if ((error = lockmgr(&cs->sc_lock, LK_EXCLUSIVE, NULL)) != 0)
630 		return (error);
631 
632 	part = DISKPART(dev);
633 
634 	/* ...that much closer to allowing unconfiguration... */
635 	switch (fmt) {
636 	case S_IFCHR:
637 		cs->sc_dkdev.dk_copenmask &= ~(1 << part);
638 		break;
639 
640 	case S_IFBLK:
641 		cs->sc_dkdev.dk_bopenmask &= ~(1 << part);
642 		break;
643 	}
644 	cs->sc_dkdev.dk_openmask =
645 	    cs->sc_dkdev.dk_copenmask | cs->sc_dkdev.dk_bopenmask;
646 
647 	if (cs->sc_dkdev.dk_openmask == 0) {
648 		if ((cs->sc_flags & CCDF_KLABEL) == 0)
649 			cs->sc_flags &= ~CCDF_VLABEL;
650 	}
651 
652 	(void) lockmgr(&cs->sc_lock, LK_RELEASE, NULL);
653 	return (0);
654 }
655 
656 static void
657 ccdstrategy(struct buf *bp)
658 {
659 	int unit = ccdunit(bp->b_dev);
660 	struct ccd_softc *cs = &ccd_softc[unit];
661 	daddr_t blkno;
662 	int s;
663 	int wlabel;
664 	struct disklabel *lp;
665 
666 #ifdef DEBUG
667 	if (ccddebug & CCDB_FOLLOW)
668 		printf("ccdstrategy(%p): unit %d\n", bp, unit);
669 #endif
670 	if ((cs->sc_flags & CCDF_INITED) == 0) {
671 #ifdef DEBUG
672 		if (ccddebug & CCDB_FOLLOW)
673 			printf("ccdstrategy: unit %d: not inited\n", unit);
674 #endif
675 		bp->b_error = ENXIO;
676 		bp->b_flags |= B_ERROR;
677 		goto done;
678 	}
679 
680 	/* If it's a nil transfer, wake up the top half now. */
681 	if (bp->b_bcount == 0)
682 		goto done;
683 
684 	lp = cs->sc_dkdev.dk_label;
685 
686 	/*
687 	 * Do bounds checking and adjust transfer.  If there's an
688 	 * error, the bounds check will flag that for us.  Convert
689 	 * the partition relative block number to an absolute.
690 	 */
691 	blkno = bp->b_blkno;
692 	wlabel = cs->sc_flags & (CCDF_WLABEL|CCDF_LABELLING);
693 	if (DISKPART(bp->b_dev) != RAW_PART) {
694 		if (bounds_check_with_label(&cs->sc_dkdev, bp, wlabel) <= 0)
695 			goto done;
696 		blkno += lp->d_partitions[DISKPART(bp->b_dev)].p_offset;
697 	}
698 	bp->b_rawblkno = blkno;
699 
700 	/* Place it in the queue and start I/O on the unit. */
701 	s = splbio();
702 	BUFQ_PUT(cs->sc_bufq, bp);
703 	ccdstart(cs);
704 	splx(s);
705 	return;
706 
707  done:
708 	bp->b_resid = bp->b_bcount;
709 	biodone(bp);
710 }
711 
712 static void
713 ccdstart(struct ccd_softc *cs)
714 {
715 	long bcount, rcount;
716 	struct buf *bp;
717 	struct ccdbuf *cbp;
718 	caddr_t addr;
719 	daddr_t bn;
720 	SIMPLEQ_HEAD(, ccdbuf) cbufq;
721 
722 #ifdef DEBUG
723 	if (ccddebug & CCDB_FOLLOW)
724 		printf("ccdstart(%p)\n", cs);
725 #endif
726 
727 	/* See if there is work for us to do. */
728 	while ((bp = BUFQ_PEEK(cs->sc_bufq)) != NULL) {
729 		/* Instrumentation. */
730 		disk_busy(&cs->sc_dkdev);
731 
732 		bp->b_resid = bp->b_bcount;
733 		bn = bp->b_rawblkno;
734 
735 		/* Allocate the component buffers. */
736 		SIMPLEQ_INIT(&cbufq);
737 		addr = bp->b_data;
738 		for (bcount = bp->b_bcount; bcount > 0; bcount -= rcount) {
739 			cbp = ccdbuffer(cs, bp, bn, addr, bcount);
740 			if (cbp == NULL) {
741 				/*
742 				 * Can't allocate a component buffer; just
743 				 * defer the job until later.
744 				 *
745 				 * XXX We might consider a watchdog timer
746 				 * XXX to make sure we are kicked into action,
747 				 * XXX or consider a low-water mark for our
748 				 * XXX component buffer pool.
749 				 */
750 				while ((cbp = SIMPLEQ_FIRST(&cbufq)) != NULL) {
751 					SIMPLEQ_REMOVE_HEAD(&cbufq, cb_q);
752 					CCD_PUTBUF(cbp);
753 				}
754 				disk_unbusy(&cs->sc_dkdev, 0, 0);
755 				return;
756 			}
757 			SIMPLEQ_INSERT_TAIL(&cbufq, cbp, cb_q);
758 			rcount = cbp->cb_buf.b_bcount;
759 			bn += btodb(rcount);
760 			addr += rcount;
761 		}
762 
763 		/* Transfer all set up, remove job from the queue. */
764 		(void) BUFQ_GET(cs->sc_bufq);
765 
766 		/* Now fire off the requests. */
767 		while ((cbp = SIMPLEQ_FIRST(&cbufq)) != NULL) {
768 			SIMPLEQ_REMOVE_HEAD(&cbufq, cb_q);
769 			if ((cbp->cb_buf.b_flags & B_READ) == 0)
770 				cbp->cb_buf.b_vp->v_numoutput++;
771 			DEV_STRATEGY(&cbp->cb_buf);
772 		}
773 	}
774 }
775 
776 /*
777  * Build a component buffer header.
778  */
779 static struct ccdbuf *
780 ccdbuffer(struct ccd_softc *cs, struct buf *bp, daddr_t bn, caddr_t addr,
781     long bcount)
782 {
783 	struct ccdcinfo *ci;
784 	struct ccdbuf *cbp;
785 	daddr_t cbn, cboff;
786 	u_int64_t cbc;
787 	int ccdisk;
788 
789 #ifdef DEBUG
790 	if (ccddebug & CCDB_IO)
791 		printf("ccdbuffer(%p, %p, %" PRId64 ", %p, %ld)\n",
792 		       cs, bp, bn, addr, bcount);
793 #endif
794 	/*
795 	 * Determine which component bn falls in.
796 	 */
797 	cbn = bn;
798 	cboff = 0;
799 
800 	/*
801 	 * Serially concatenated
802 	 */
803 	if (cs->sc_ileave == 0) {
804 		daddr_t sblk;
805 
806 		sblk = 0;
807 		for (ccdisk = 0, ci = &cs->sc_cinfo[ccdisk];
808 		    cbn >= sblk + ci->ci_size;
809 		    ccdisk++, ci = &cs->sc_cinfo[ccdisk])
810 			sblk += ci->ci_size;
811 		cbn -= sblk;
812 	}
813 	/*
814 	 * Interleaved
815 	 */
816 	else {
817 		struct ccdiinfo *ii;
818 		int off;
819 
820 		cboff = cbn % cs->sc_ileave;
821 		cbn /= cs->sc_ileave;
822 		for (ii = cs->sc_itable; ii->ii_ndisk; ii++)
823 			if (ii->ii_startblk > cbn)
824 				break;
825 		ii--;
826 		off = cbn - ii->ii_startblk;
827 		if (ii->ii_ndisk == 1) {
828 			ccdisk = ii->ii_index[0];
829 			cbn = ii->ii_startoff + off;
830 		} else {
831 			ccdisk = ii->ii_index[off % ii->ii_ndisk];
832 			cbn = ii->ii_startoff + off / ii->ii_ndisk;
833 		}
834 		cbn *= cs->sc_ileave;
835 		ci = &cs->sc_cinfo[ccdisk];
836 	}
837 
838 	/*
839 	 * Fill in the component buf structure.
840 	 */
841 	cbp = CCD_GETBUF();
842 	if (cbp == NULL)
843 		return (NULL);
844 	BUF_INIT(&cbp->cb_buf);
845 	cbp->cb_buf.b_flags = bp->b_flags | B_CALL;
846 	cbp->cb_buf.b_iodone = ccdiodone;
847 	cbp->cb_buf.b_proc = bp->b_proc;
848 	cbp->cb_buf.b_dev = ci->ci_dev;
849 	cbp->cb_buf.b_blkno = cbn + cboff;
850 	cbp->cb_buf.b_data = addr;
851 	cbp->cb_buf.b_vp = ci->ci_vp;
852 	if (cs->sc_ileave == 0)
853 		cbc = dbtob((u_int64_t)(ci->ci_size - cbn));
854 	else
855 		cbc = dbtob((u_int64_t)(cs->sc_ileave - cboff));
856 	cbp->cb_buf.b_bcount = cbc < bcount ? cbc : bcount;
857 
858 	/*
859 	 * context for ccdiodone
860 	 */
861 	cbp->cb_obp = bp;
862 	cbp->cb_sc = cs;
863 	cbp->cb_comp = ccdisk;
864 
865 	BIO_COPYPRIO(&cbp->cb_buf, bp);
866 
867 #ifdef DEBUG
868 	if (ccddebug & CCDB_IO)
869 		printf(" dev 0x%x(u%lu): cbp %p bn %" PRId64 " addr %p"
870 		       " bcnt %d\n",
871 		    ci->ci_dev, (unsigned long) (ci-cs->sc_cinfo), cbp,
872 		    cbp->cb_buf.b_blkno, cbp->cb_buf.b_data,
873 		    cbp->cb_buf.b_bcount);
874 #endif
875 
876 	return (cbp);
877 }
878 
879 static void
880 ccdintr(struct ccd_softc *cs, struct buf *bp)
881 {
882 
883 #ifdef DEBUG
884 	if (ccddebug & CCDB_FOLLOW)
885 		printf("ccdintr(%p, %p)\n", cs, bp);
886 #endif
887 	/*
888 	 * Request is done for better or worse, wakeup the top half.
889 	 */
890 	if (bp->b_flags & B_ERROR)
891 		bp->b_resid = bp->b_bcount;
892 	disk_unbusy(&cs->sc_dkdev, (bp->b_bcount - bp->b_resid),
893 	    (bp->b_flags & B_READ));
894 	biodone(bp);
895 }
896 
897 /*
898  * Called at interrupt time.
899  * Mark the component as done and if all components are done,
900  * take a ccd interrupt.
901  */
902 static void
903 ccdiodone(struct buf *vbp)
904 {
905 	struct ccdbuf *cbp = (struct ccdbuf *) vbp;
906 	struct buf *bp = cbp->cb_obp;
907 	struct ccd_softc *cs = cbp->cb_sc;
908 	int count, s;
909 
910 	s = splbio();
911 #ifdef DEBUG
912 	if (ccddebug & CCDB_FOLLOW)
913 		printf("ccdiodone(%p)\n", cbp);
914 	if (ccddebug & CCDB_IO) {
915 		printf("ccdiodone: bp %p bcount %d resid %d\n",
916 		       bp, bp->b_bcount, bp->b_resid);
917 		printf(" dev 0x%x(u%d), cbp %p bn %" PRId64 " addr %p"
918 		       " bcnt %d\n",
919 		       cbp->cb_buf.b_dev, cbp->cb_comp, cbp,
920 		       cbp->cb_buf.b_blkno, cbp->cb_buf.b_data,
921 		       cbp->cb_buf.b_bcount);
922 	}
923 #endif
924 
925 	if (cbp->cb_buf.b_flags & B_ERROR) {
926 		bp->b_flags |= B_ERROR;
927 		bp->b_error = cbp->cb_buf.b_error ?
928 		    cbp->cb_buf.b_error : EIO;
929 
930 		printf("%s: error %d on component %d\n",
931 		       cs->sc_xname, bp->b_error, cbp->cb_comp);
932 	}
933 	count = cbp->cb_buf.b_bcount;
934 	CCD_PUTBUF(cbp);
935 
936 	/*
937 	 * If all done, "interrupt".
938 	 */
939 	bp->b_resid -= count;
940 	if (bp->b_resid < 0)
941 		panic("ccdiodone: count");
942 	if (bp->b_resid == 0)
943 		ccdintr(cs, bp);
944 	splx(s);
945 }
946 
947 /* ARGSUSED */
948 static int
949 ccdread(dev_t dev, struct uio *uio, int flags __unused)
950 {
951 	int unit = ccdunit(dev);
952 	struct ccd_softc *cs;
953 
954 #ifdef DEBUG
955 	if (ccddebug & CCDB_FOLLOW)
956 		printf("ccdread(0x%x, %p)\n", dev, uio);
957 #endif
958 	if (unit >= numccd)
959 		return (ENXIO);
960 	cs = &ccd_softc[unit];
961 
962 	if ((cs->sc_flags & CCDF_INITED) == 0)
963 		return (ENXIO);
964 
965 	return (physio(ccdstrategy, NULL, dev, B_READ, minphys, uio));
966 }
967 
968 /* ARGSUSED */
969 static int
970 ccdwrite(dev_t dev, struct uio *uio, int flags __unused)
971 {
972 	int unit = ccdunit(dev);
973 	struct ccd_softc *cs;
974 
975 #ifdef DEBUG
976 	if (ccddebug & CCDB_FOLLOW)
977 		printf("ccdwrite(0x%x, %p)\n", dev, uio);
978 #endif
979 	if (unit >= numccd)
980 		return (ENXIO);
981 	cs = &ccd_softc[unit];
982 
983 	if ((cs->sc_flags & CCDF_INITED) == 0)
984 		return (ENXIO);
985 
986 	return (physio(ccdstrategy, NULL, dev, B_WRITE, minphys, uio));
987 }
988 
989 static int
990 ccdioctl(dev_t dev, u_long cmd, caddr_t data, int flag, struct lwp *l)
991 {
992 	int unit = ccdunit(dev);
993 	int s, i, j, lookedup = 0, error;
994 	int part, pmask;
995 	struct ccd_softc *cs;
996 	struct ccd_ioctl *ccio = (struct ccd_ioctl *)data;
997 	kauth_cred_t uc;
998 	char **cpp;
999 	struct vnode **vpp;
1000 #ifdef __HAVE_OLD_DISKLABEL
1001 	struct disklabel newlabel;
1002 #endif
1003 
1004 	if (unit >= numccd)
1005 		return (ENXIO);
1006 	cs = &ccd_softc[unit];
1007 
1008 	uc = (l != NULL) ? l->l_cred : NOCRED;
1009 
1010 	/* Must be open for writes for these commands... */
1011 	switch (cmd) {
1012 	case CCDIOCSET:
1013 	case CCDIOCCLR:
1014 	case DIOCSDINFO:
1015 	case DIOCWDINFO:
1016 #ifdef __HAVE_OLD_DISKLABEL
1017 	case ODIOCSDINFO:
1018 	case ODIOCWDINFO:
1019 #endif
1020 	case DIOCKLABEL:
1021 	case DIOCWLABEL:
1022 		if ((flag & FWRITE) == 0)
1023 			return (EBADF);
1024 	}
1025 
1026 	if ((error = lockmgr(&cs->sc_lock, LK_EXCLUSIVE, NULL)) != 0)
1027 		return (error);
1028 
1029 	/* Must be initialized for these... */
1030 	switch (cmd) {
1031 	case CCDIOCCLR:
1032 	case DIOCGDINFO:
1033 	case DIOCCACHESYNC:
1034 	case DIOCSDINFO:
1035 	case DIOCWDINFO:
1036 	case DIOCGPART:
1037 	case DIOCWLABEL:
1038 	case DIOCKLABEL:
1039 	case DIOCGDEFLABEL:
1040 #ifdef __HAVE_OLD_DISKLABEL
1041 	case ODIOCGDINFO:
1042 	case ODIOCSDINFO:
1043 	case ODIOCWDINFO:
1044 	case ODIOCGDEFLABEL:
1045 #endif
1046 		if ((cs->sc_flags & CCDF_INITED) == 0) {
1047 			error = ENXIO;
1048 			goto out;
1049 		}
1050 	}
1051 
1052 	switch (cmd) {
1053 	case CCDIOCSET:
1054 		if (cs->sc_flags & CCDF_INITED) {
1055 			error = EBUSY;
1056 			goto out;
1057 		}
1058 
1059 		/* Validate the flags. */
1060 		if ((ccio->ccio_flags & CCDF_USERMASK) != ccio->ccio_flags) {
1061 			error = EINVAL;
1062 			goto out;
1063 		}
1064 
1065 		if (ccio->ccio_ndisks > CCD_MAXNDISKS) {
1066 			error = EINVAL;
1067 			goto out;
1068 		}
1069 
1070 		/* Fill in some important bits. */
1071 		cs->sc_ileave = ccio->ccio_ileave;
1072 		cs->sc_nccdisks = ccio->ccio_ndisks;
1073 		cs->sc_flags = ccio->ccio_flags & CCDF_USERMASK;
1074 
1075 		/*
1076 		 * Allocate space for and copy in the array of
1077 		 * componet pathnames and device numbers.
1078 		 */
1079 		cpp = malloc(ccio->ccio_ndisks * sizeof(char *),
1080 		    M_DEVBUF, M_WAITOK);
1081 		vpp = malloc(ccio->ccio_ndisks * sizeof(struct vnode *),
1082 		    M_DEVBUF, M_WAITOK);
1083 
1084 		error = copyin(ccio->ccio_disks, cpp,
1085 		    ccio->ccio_ndisks * sizeof(char **));
1086 		if (error) {
1087 			free(vpp, M_DEVBUF);
1088 			free(cpp, M_DEVBUF);
1089 			goto out;
1090 		}
1091 
1092 #ifdef DEBUG
1093 		if (ccddebug & CCDB_INIT)
1094 			for (i = 0; i < ccio->ccio_ndisks; ++i)
1095 				printf("ccdioctl: component %d: %p\n",
1096 				    i, cpp[i]);
1097 #endif
1098 
1099 		for (i = 0; i < ccio->ccio_ndisks; ++i) {
1100 #ifdef DEBUG
1101 			if (ccddebug & CCDB_INIT)
1102 				printf("ccdioctl: lookedup = %d\n", lookedup);
1103 #endif
1104 			if ((error = dk_lookup(cpp[i], l, &vpp[i])) != 0) {
1105 				for (j = 0; j < lookedup; ++j)
1106 					(void)vn_close(vpp[j], FREAD|FWRITE,
1107 					    uc, l);
1108 				free(vpp, M_DEVBUF);
1109 				free(cpp, M_DEVBUF);
1110 				goto out;
1111 			}
1112 			++lookedup;
1113 		}
1114 
1115 		/*
1116 		 * Initialize the ccd.  Fills in the softc for us.
1117 		 */
1118 		if ((error = ccdinit(cs, cpp, vpp, l)) != 0) {
1119 			for (j = 0; j < lookedup; ++j)
1120 				(void)vn_close(vpp[j], FREAD|FWRITE,
1121 				    uc, l);
1122 			free(vpp, M_DEVBUF);
1123 			free(cpp, M_DEVBUF);
1124 			goto out;
1125 		}
1126 
1127 		/* We can free the temporary variables now. */
1128 		free(vpp, M_DEVBUF);
1129 		free(cpp, M_DEVBUF);
1130 
1131 		/*
1132 		 * The ccd has been successfully initialized, so
1133 		 * we can place it into the array.  Don't try to
1134 		 * read the disklabel until the disk has been attached,
1135 		 * because space for the disklabel is allocated
1136 		 * in disk_attach();
1137 		 */
1138 		ccio->ccio_unit = unit;
1139 		ccio->ccio_size = cs->sc_size;
1140 
1141 		bufq_alloc(&cs->sc_bufq, "fcfs", 0);
1142 
1143 		/* Attach the disk. */
1144 		pseudo_disk_attach(&cs->sc_dkdev);
1145 
1146 		/* Try and read the disklabel. */
1147 		ccdgetdisklabel(dev);
1148 		break;
1149 
1150 	case CCDIOCCLR:
1151 		/*
1152 		 * Don't unconfigure if any other partitions are open
1153 		 * or if both the character and block flavors of this
1154 		 * partition are open.
1155 		 */
1156 		part = DISKPART(dev);
1157 		pmask = (1 << part);
1158 		if ((cs->sc_dkdev.dk_openmask & ~pmask) ||
1159 		    ((cs->sc_dkdev.dk_bopenmask & pmask) &&
1160 		    (cs->sc_dkdev.dk_copenmask & pmask))) {
1161 			error = EBUSY;
1162 			goto out;
1163 		}
1164 
1165 		/* Kill off any queued buffers. */
1166 		s = splbio();
1167 		bufq_drain(cs->sc_bufq);
1168 		splx(s);
1169 
1170 		bufq_free(cs->sc_bufq);
1171 
1172 		/*
1173 		 * Free ccd_softc information and clear entry.
1174 		 */
1175 
1176 		/* Close the components and free their pathnames. */
1177 		for (i = 0; i < cs->sc_nccdisks; ++i) {
1178 			/*
1179 			 * XXX: this close could potentially fail and
1180 			 * cause Bad Things.  Maybe we need to force
1181 			 * the close to happen?
1182 			 */
1183 #ifdef DEBUG
1184 			if (ccddebug & CCDB_VNODE)
1185 				vprint("CCDIOCCLR: vnode info",
1186 				    cs->sc_cinfo[i].ci_vp);
1187 #endif
1188 			(void)vn_close(cs->sc_cinfo[i].ci_vp, FREAD|FWRITE,
1189 			    uc, l);
1190 			free(cs->sc_cinfo[i].ci_path, M_DEVBUF);
1191 		}
1192 
1193 		/* Free interleave index. */
1194 		for (i = 0; cs->sc_itable[i].ii_ndisk; ++i)
1195 			free(cs->sc_itable[i].ii_index, M_DEVBUF);
1196 
1197 		/* Free component info and interleave table. */
1198 		free(cs->sc_cinfo, M_DEVBUF);
1199 		free(cs->sc_itable, M_DEVBUF);
1200 		cs->sc_flags &= ~(CCDF_INITED|CCDF_VLABEL);
1201 
1202 		/* Detatch the disk. */
1203 		pseudo_disk_detach(&cs->sc_dkdev);
1204 		break;
1205 
1206 	case DIOCGDINFO:
1207 		*(struct disklabel *)data = *(cs->sc_dkdev.dk_label);
1208 		break;
1209 #ifdef __HAVE_OLD_DISKLABEL
1210 	case ODIOCGDINFO:
1211 		newlabel = *(cs->sc_dkdev.dk_label);
1212 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1213 			return ENOTTY;
1214 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1215 		break;
1216 #endif
1217 
1218 	case DIOCGPART:
1219 		((struct partinfo *)data)->disklab = cs->sc_dkdev.dk_label;
1220 		((struct partinfo *)data)->part =
1221 		    &cs->sc_dkdev.dk_label->d_partitions[DISKPART(dev)];
1222 		break;
1223 
1224 	case DIOCCACHESYNC:
1225 		/*
1226 		 * XXX Do we really need to care about having a writable
1227 		 * file descriptor here?
1228 		 */
1229 		if ((flag & FWRITE) == 0)
1230 			return (EBADF);
1231 
1232 		/*
1233 		 * We pass this call down to all components and report
1234 		 * the first error we encounter.
1235 		 */
1236 		for (error = 0, i = 0; i < cs->sc_nccdisks; i++) {
1237 			j = VOP_IOCTL(cs->sc_cinfo[i].ci_vp, cmd, data,
1238 				      flag, uc, l);
1239 			if (j != 0 && error == 0)
1240 				error = j;
1241 		}
1242 		break;
1243 
1244 	case DIOCWDINFO:
1245 	case DIOCSDINFO:
1246 #ifdef __HAVE_OLD_DISKLABEL
1247 	case ODIOCWDINFO:
1248 	case ODIOCSDINFO:
1249 #endif
1250 	{
1251 		struct disklabel *lp;
1252 #ifdef __HAVE_OLD_DISKLABEL
1253 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
1254 			memset(&newlabel, 0, sizeof newlabel);
1255 			memcpy(&newlabel, data, sizeof (struct olddisklabel));
1256 			lp = &newlabel;
1257 		} else
1258 #endif
1259 		lp = (struct disklabel *)data;
1260 
1261 		cs->sc_flags |= CCDF_LABELLING;
1262 
1263 		error = setdisklabel(cs->sc_dkdev.dk_label,
1264 		    lp, 0, cs->sc_dkdev.dk_cpulabel);
1265 		if (error == 0) {
1266 			if (cmd == DIOCWDINFO
1267 #ifdef __HAVE_OLD_DISKLABEL
1268 			    || cmd == ODIOCWDINFO
1269 #endif
1270 			   )
1271 				error = writedisklabel(CCDLABELDEV(dev),
1272 				    ccdstrategy, cs->sc_dkdev.dk_label,
1273 				    cs->sc_dkdev.dk_cpulabel);
1274 		}
1275 
1276 		cs->sc_flags &= ~CCDF_LABELLING;
1277 		break;
1278 	}
1279 
1280 	case DIOCKLABEL:
1281 		if (*(int *)data != 0)
1282 			cs->sc_flags |= CCDF_KLABEL;
1283 		else
1284 			cs->sc_flags &= ~CCDF_KLABEL;
1285 		break;
1286 
1287 	case DIOCWLABEL:
1288 		if (*(int *)data != 0)
1289 			cs->sc_flags |= CCDF_WLABEL;
1290 		else
1291 			cs->sc_flags &= ~CCDF_WLABEL;
1292 		break;
1293 
1294 	case DIOCGDEFLABEL:
1295 		ccdgetdefaultlabel(cs, (struct disklabel *)data);
1296 		break;
1297 
1298 #ifdef __HAVE_OLD_DISKLABEL
1299 	case ODIOCGDEFLABEL:
1300 		ccdgetdefaultlabel(cs, &newlabel);
1301 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1302 			return ENOTTY;
1303 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1304 		break;
1305 #endif
1306 
1307 	default:
1308 		error = ENOTTY;
1309 	}
1310 
1311  out:
1312 	(void) lockmgr(&cs->sc_lock, LK_RELEASE, NULL);
1313 	return (error);
1314 }
1315 
1316 static int
1317 ccdsize(dev_t dev)
1318 {
1319 	struct ccd_softc *cs;
1320 	struct disklabel *lp;
1321 	int part, unit, omask, size;
1322 
1323 	unit = ccdunit(dev);
1324 	if (unit >= numccd)
1325 		return (-1);
1326 	cs = &ccd_softc[unit];
1327 
1328 	if ((cs->sc_flags & CCDF_INITED) == 0)
1329 		return (-1);
1330 
1331 	part = DISKPART(dev);
1332 	omask = cs->sc_dkdev.dk_openmask & (1 << part);
1333 	lp = cs->sc_dkdev.dk_label;
1334 
1335 	if (omask == 0 && ccdopen(dev, 0, S_IFBLK, curlwp))
1336 		return (-1);
1337 
1338 	if (lp->d_partitions[part].p_fstype != FS_SWAP)
1339 		size = -1;
1340 	else
1341 		size = lp->d_partitions[part].p_size *
1342 		    (lp->d_secsize / DEV_BSIZE);
1343 
1344 	if (omask == 0 && ccdclose(dev, 0, S_IFBLK, curlwp))
1345 		return (-1);
1346 
1347 	return (size);
1348 }
1349 
1350 static int
1351 ccddump(dev_t dev __unused, daddr_t blkno __unused, caddr_t va __unused,
1352     size_t size __unused)
1353 {
1354 
1355 	/* Not implemented. */
1356 	return ENXIO;
1357 }
1358 
1359 static void
1360 ccdgetdefaultlabel(struct ccd_softc *cs, struct disklabel *lp)
1361 {
1362 	struct ccdgeom *ccg = &cs->sc_geom;
1363 
1364 	memset(lp, 0, sizeof(*lp));
1365 
1366 	lp->d_secperunit = cs->sc_size;
1367 	lp->d_secsize = ccg->ccg_secsize;
1368 	lp->d_nsectors = ccg->ccg_nsectors;
1369 	lp->d_ntracks = ccg->ccg_ntracks;
1370 	lp->d_ncylinders = ccg->ccg_ncylinders;
1371 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1372 
1373 	strncpy(lp->d_typename, "ccd", sizeof(lp->d_typename));
1374 	lp->d_type = DTYPE_CCD;
1375 	strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
1376 	lp->d_rpm = 3600;
1377 	lp->d_interleave = 1;
1378 	lp->d_flags = 0;
1379 
1380 	lp->d_partitions[RAW_PART].p_offset = 0;
1381 	lp->d_partitions[RAW_PART].p_size = cs->sc_size;
1382 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
1383 	lp->d_npartitions = RAW_PART + 1;
1384 
1385 	lp->d_magic = DISKMAGIC;
1386 	lp->d_magic2 = DISKMAGIC;
1387 	lp->d_checksum = dkcksum(cs->sc_dkdev.dk_label);
1388 }
1389 
1390 /*
1391  * Read the disklabel from the ccd.  If one is not present, fake one
1392  * up.
1393  */
1394 static void
1395 ccdgetdisklabel(dev_t dev)
1396 {
1397 	int unit = ccdunit(dev);
1398 	struct ccd_softc *cs = &ccd_softc[unit];
1399 	const char *errstring;
1400 	struct disklabel *lp = cs->sc_dkdev.dk_label;
1401 	struct cpu_disklabel *clp = cs->sc_dkdev.dk_cpulabel;
1402 
1403 	memset(clp, 0, sizeof(*clp));
1404 
1405 	ccdgetdefaultlabel(cs, lp);
1406 
1407 	/*
1408 	 * Call the generic disklabel extraction routine.
1409 	 */
1410 	if ((cs->sc_flags & CCDF_NOLABEL) != 0)
1411 		errstring = "CCDF_NOLABEL set; ignoring on-disk label";
1412 	else
1413 		errstring = readdisklabel(CCDLABELDEV(dev), ccdstrategy,
1414 		    cs->sc_dkdev.dk_label, cs->sc_dkdev.dk_cpulabel);
1415 	if (errstring)
1416 		ccdmakedisklabel(cs);
1417 	else {
1418 		int i;
1419 		struct partition *pp;
1420 
1421 		/*
1422 		 * Sanity check whether the found disklabel is valid.
1423 		 *
1424 		 * This is necessary since total size of ccd may vary
1425 		 * when an interleave is changed even though exactly
1426 		 * same componets are used, and old disklabel may used
1427 		 * if that is found.
1428 		 */
1429 		if (lp->d_secperunit != cs->sc_size)
1430 			printf("WARNING: %s: "
1431 			    "total sector size in disklabel (%d) != "
1432 			    "the size of ccd (%lu)\n", cs->sc_xname,
1433 			    lp->d_secperunit, (u_long)cs->sc_size);
1434 		for (i = 0; i < lp->d_npartitions; i++) {
1435 			pp = &lp->d_partitions[i];
1436 			if (pp->p_offset + pp->p_size > cs->sc_size)
1437 				printf("WARNING: %s: end of partition `%c' "
1438 				    "exceeds the size of ccd (%lu)\n",
1439 				    cs->sc_xname, 'a' + i, (u_long)cs->sc_size);
1440 		}
1441 	}
1442 
1443 #ifdef DEBUG
1444 	/* It's actually extremely common to have unlabeled ccds. */
1445 	if (ccddebug & CCDB_LABEL)
1446 		if (errstring != NULL)
1447 			printf("%s: %s\n", cs->sc_xname, errstring);
1448 #endif
1449 
1450 	/* In-core label now valid. */
1451 	cs->sc_flags |= CCDF_VLABEL;
1452 }
1453 
1454 /*
1455  * Take care of things one might want to take care of in the event
1456  * that a disklabel isn't present.
1457  */
1458 static void
1459 ccdmakedisklabel(struct ccd_softc *cs)
1460 {
1461 	struct disklabel *lp = cs->sc_dkdev.dk_label;
1462 
1463 	/*
1464 	 * For historical reasons, if there's no disklabel present
1465 	 * the raw partition must be marked FS_BSDFFS.
1466 	 */
1467 	lp->d_partitions[RAW_PART].p_fstype = FS_BSDFFS;
1468 
1469 	strncpy(lp->d_packname, "default label", sizeof(lp->d_packname));
1470 
1471 	lp->d_checksum = dkcksum(lp);
1472 }
1473 
1474 #ifdef DEBUG
1475 static void
1476 printiinfo(struct ccdiinfo *ii)
1477 {
1478 	int ix, i;
1479 
1480 	for (ix = 0; ii->ii_ndisk; ix++, ii++) {
1481 		printf(" itab[%d]: #dk %d sblk %" PRId64 " soff %" PRId64,
1482 		    ix, ii->ii_ndisk, ii->ii_startblk, ii->ii_startoff);
1483 		for (i = 0; i < ii->ii_ndisk; i++)
1484 			printf(" %d", ii->ii_index[i]);
1485 		printf("\n");
1486 	}
1487 }
1488 #endif
1489