xref: /netbsd-src/sys/dev/ccd.c (revision b1c86f5f087524e68db12794ee9c3e3da1ab17a0)
1 /*	$NetBSD: ccd.c,v 1.134 2009/06/05 19:21:02 haad Exp $	*/
2 
3 /*-
4  * Copyright (c) 1996, 1997, 1998, 1999, 2007, 2009 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, and by Andrew Doran.
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  *
19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29  * POSSIBILITY OF SUCH DAMAGE.
30  */
31 
32 /*
33  * Copyright (c) 1990, 1993
34  *	The Regents of the University of California.  All rights reserved.
35  *
36  * This code is derived from software contributed to Berkeley by
37  * the Systems Programming Group of the University of Utah Computer
38  * Science Department.
39  *
40  * Redistribution and use in source and binary forms, with or without
41  * modification, are permitted provided that the following conditions
42  * are met:
43  * 1. Redistributions of source code must retain the above copyright
44  *    notice, this list of conditions and the following disclaimer.
45  * 2. Redistributions in binary form must reproduce the above copyright
46  *    notice, this list of conditions and the following disclaimer in the
47  *    documentation and/or other materials provided with the distribution.
48  * 3. Neither the name of the University nor the names of its contributors
49  *    may be used to endorse or promote products derived from this software
50  *    without specific prior written permission.
51  *
52  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
53  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
54  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
55  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
56  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
57  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
58  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
59  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
60  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
61  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
62  * SUCH DAMAGE.
63  *
64  * from: Utah $Hdr: cd.c 1.6 90/11/28$
65  *
66  *	@(#)cd.c	8.2 (Berkeley) 11/16/93
67  */
68 
69 /*
70  * Copyright (c) 1988 University of Utah.
71  *
72  * This code is derived from software contributed to Berkeley by
73  * the Systems Programming Group of the University of Utah Computer
74  * Science Department.
75  *
76  * Redistribution and use in source and binary forms, with or without
77  * modification, are permitted provided that the following conditions
78  * are met:
79  * 1. Redistributions of source code must retain the above copyright
80  *    notice, this list of conditions and the following disclaimer.
81  * 2. Redistributions in binary form must reproduce the above copyright
82  *    notice, this list of conditions and the following disclaimer in the
83  *    documentation and/or other materials provided with the distribution.
84  * 3. All advertising materials mentioning features or use of this software
85  *    must display the following acknowledgement:
86  *	This product includes software developed by the University of
87  *	California, Berkeley and its contributors.
88  * 4. Neither the name of the University nor the names of its contributors
89  *    may be used to endorse or promote products derived from this software
90  *    without specific prior written permission.
91  *
92  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
93  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
94  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
95  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
96  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
97  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
98  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
99  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
100  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
101  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
102  * SUCH DAMAGE.
103  *
104  * from: Utah $Hdr: cd.c 1.6 90/11/28$
105  *
106  *	@(#)cd.c	8.2 (Berkeley) 11/16/93
107  */
108 
109 /*
110  * "Concatenated" disk driver.
111  *
112  * Notes on concurrency:
113  *
114  * => sc_dvlock serializes access to the device nodes, excluding block I/O.
115  *
116  * => sc_iolock serializes access to (sc_flags & CCDF_INITED), disk stats,
117  *    sc_stop, sc_bufq and b_resid from master buffers.
118  *
119  * => a combination of CCDF_INITED, sc_inflight, and sc_iolock is used to
120  *    serialize I/O and configuration changes.
121  *
122  * => the in-core disk label does not change while the device is open.
123  *
124  * On memory consumption: ccd fans out I/O requests and so needs to
125  * allocate memory.  If the system is desperately low on memory, we
126  * single thread I/O.
127  */
128 
129 #include <sys/cdefs.h>
130 __KERNEL_RCSID(0, "$NetBSD: ccd.c,v 1.134 2009/06/05 19:21:02 haad Exp $");
131 
132 #include <sys/param.h>
133 #include <sys/systm.h>
134 #include <sys/kernel.h>
135 #include <sys/proc.h>
136 #include <sys/errno.h>
137 #include <sys/buf.h>
138 #include <sys/kmem.h>
139 #include <sys/pool.h>
140 #include <sys/namei.h>
141 #include <sys/stat.h>
142 #include <sys/ioctl.h>
143 #include <sys/disklabel.h>
144 #include <sys/device.h>
145 #include <sys/disk.h>
146 #include <sys/syslog.h>
147 #include <sys/fcntl.h>
148 #include <sys/vnode.h>
149 #include <sys/conf.h>
150 #include <sys/mutex.h>
151 #include <sys/queue.h>
152 #include <sys/kauth.h>
153 #include <sys/kthread.h>
154 #include <sys/bufq.h>
155 
156 #include <dev/ccdvar.h>
157 #include <dev/dkvar.h>
158 
159 #if defined(CCDDEBUG) && !defined(DEBUG)
160 #define DEBUG
161 #endif
162 
163 #ifdef DEBUG
164 #define CCDB_FOLLOW	0x01
165 #define CCDB_INIT	0x02
166 #define CCDB_IO		0x04
167 #define CCDB_LABEL	0x08
168 #define CCDB_VNODE	0x10
169 int ccddebug = 0x00;
170 #endif
171 
172 #define	ccdunit(x)	DISKUNIT(x)
173 
174 struct ccdbuf {
175 	struct buf	cb_buf;		/* new I/O buf */
176 	struct buf	*cb_obp;	/* ptr. to original I/O buf */
177 	struct ccd_softc *cb_sc;	/* pointer to ccd softc */
178 	int		cb_comp;	/* target component */
179 	SIMPLEQ_ENTRY(ccdbuf) cb_q;	/* fifo of component buffers */
180 };
181 
182 /* component buffer pool */
183 static pool_cache_t ccd_cache;
184 
185 #define	CCD_GETBUF()		pool_cache_get(ccd_cache, PR_WAITOK)
186 #define	CCD_PUTBUF(cbp)		pool_cache_put(ccd_cache, cbp)
187 
188 #define CCDLABELDEV(dev)	\
189 	(MAKEDISKDEV(major((dev)), ccdunit((dev)), RAW_PART))
190 
191 /* called by main() at boot time */
192 void	ccdattach(int);
193 
194 /* called by biodone() at interrupt time */
195 static void	ccdiodone(struct buf *);
196 
197 static void	ccdinterleave(struct ccd_softc *);
198 static int	ccdinit(struct ccd_softc *, char **, struct vnode **,
199 		    struct lwp *);
200 static struct ccdbuf *ccdbuffer(struct ccd_softc *, struct buf *,
201 		    daddr_t, void *, long);
202 static void	ccdgetdefaultlabel(struct ccd_softc *, struct disklabel *);
203 static void	ccdgetdisklabel(dev_t);
204 static void	ccdmakedisklabel(struct ccd_softc *);
205 static void	ccdstart(struct ccd_softc *);
206 static void	ccdthread(void *);
207 
208 static dev_type_open(ccdopen);
209 static dev_type_close(ccdclose);
210 static dev_type_read(ccdread);
211 static dev_type_write(ccdwrite);
212 static dev_type_ioctl(ccdioctl);
213 static dev_type_strategy(ccdstrategy);
214 static dev_type_size(ccdsize);
215 
216 const struct bdevsw ccd_bdevsw = {
217 	.d_open = ccdopen,
218 	.d_close = ccdclose,
219 	.d_strategy = ccdstrategy,
220 	.d_ioctl = ccdioctl,
221 	.d_dump = nodump,
222 	.d_psize = ccdsize,
223 	.d_flag = D_DISK | D_MPSAFE
224 };
225 
226 const struct cdevsw ccd_cdevsw = {
227 	.d_open = ccdopen,
228 	.d_close = ccdclose,
229 	.d_read = ccdread,
230 	.d_write = ccdwrite,
231 	.d_ioctl = ccdioctl,
232 	.d_stop = nostop,
233 	.d_tty = notty,
234 	.d_poll = nopoll,
235 	.d_mmap = nommap,
236 	.d_kqfilter = nokqfilter,
237 	.d_flag = D_DISK | D_MPSAFE
238 };
239 
240 #ifdef DEBUG
241 static	void printiinfo(struct ccdiinfo *);
242 #endif
243 
244 /* Publically visible for the benefit of libkvm and ccdconfig(8). */
245 struct ccd_softc 	*ccd_softc;
246 const int		ccd_softc_elemsize = sizeof(struct ccd_softc);
247 int			numccd = 0;
248 
249 /*
250  * Called by main() during pseudo-device attachment.  All we need
251  * to do is allocate enough space for devices to be configured later.
252  */
253 void
254 ccdattach(int num)
255 {
256 	struct ccd_softc *cs;
257 	int i;
258 
259 	if (num <= 0) {
260 #ifdef DIAGNOSTIC
261 		panic("ccdattach: count <= 0");
262 #endif
263 		return;
264 	}
265 
266 	ccd_softc = kmem_zalloc(num * ccd_softc_elemsize, KM_SLEEP);
267 	if (ccd_softc == NULL) {
268 		printf("WARNING: no memory for concatenated disks\n");
269 		return;
270 	}
271 	numccd = num;
272 
273 	/* Initialize the component buffer pool. */
274 	ccd_cache = pool_cache_init(sizeof(struct ccdbuf), 0,
275 	    0, 0, "ccdbuf", NULL, IPL_BIO, NULL, NULL, NULL);
276 
277 	/* Initialize per-softc structures. */
278 	for (i = 0; i < num; i++) {
279 		cs = &ccd_softc[i];
280 		snprintf(cs->sc_xname, sizeof(cs->sc_xname), "ccd%d", i);
281 		mutex_init(&cs->sc_dvlock, MUTEX_DEFAULT, IPL_NONE);
282 		cs->sc_iolock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
283 		cv_init(&cs->sc_stop, "ccdstop");
284 		cv_init(&cs->sc_push, "ccdthr");
285 		disk_init(&cs->sc_dkdev, cs->sc_xname, NULL); /* XXX */
286 	}
287 }
288 
289 static int
290 ccdinit(struct ccd_softc *cs, char **cpaths, struct vnode **vpp,
291     struct lwp *l)
292 {
293 	struct ccdcinfo *ci = NULL;
294 	size_t size;
295 	int ix;
296 	struct vattr va;
297 	size_t minsize;
298 	int maxsecsize;
299 	struct partinfo dpart;
300 	struct ccdgeom *ccg = &cs->sc_geom;
301 	char *tmppath;
302 	int error, path_alloced;
303 
304 #ifdef DEBUG
305 	if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
306 		printf("%s: ccdinit\n", cs->sc_xname);
307 #endif
308 
309 	/* Allocate space for the component info. */
310 	cs->sc_cinfo = kmem_alloc(cs->sc_nccdisks * sizeof(*cs->sc_cinfo),
311 	    KM_SLEEP);
312 	tmppath = kmem_alloc(MAXPATHLEN, KM_SLEEP);
313 
314 	cs->sc_size = 0;
315 
316 	/*
317 	 * Verify that each component piece exists and record
318 	 * relevant information about it.
319 	 */
320 	maxsecsize = 0;
321 	minsize = 0;
322 	for (ix = 0, path_alloced = 0; ix < cs->sc_nccdisks; ix++) {
323 		ci = &cs->sc_cinfo[ix];
324 		ci->ci_vp = vpp[ix];
325 
326 		/*
327 		 * Copy in the pathname of the component.
328 		 */
329 		memset(tmppath, 0, sizeof(tmppath));	/* sanity */
330 		error = copyinstr(cpaths[ix], tmppath,
331 		    MAXPATHLEN, &ci->ci_pathlen);
332 		if (ci->ci_pathlen == 0)
333 			error = EINVAL;
334 		if (error) {
335 #ifdef DEBUG
336 			if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
337 				printf("%s: can't copy path, error = %d\n",
338 				    cs->sc_xname, error);
339 #endif
340 			goto out;
341 		}
342 		ci->ci_path = kmem_alloc(ci->ci_pathlen, KM_SLEEP);
343 		memcpy(ci->ci_path, tmppath, ci->ci_pathlen);
344 		path_alloced++;
345 
346 		/*
347 		 * XXX: Cache the component's dev_t.
348 		 */
349 		if ((error = VOP_GETATTR(vpp[ix], &va, l->l_cred)) != 0) {
350 #ifdef DEBUG
351 			if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
352 				printf("%s: %s: getattr failed %s = %d\n",
353 				    cs->sc_xname, ci->ci_path,
354 				    "error", error);
355 #endif
356 			goto out;
357 		}
358 		ci->ci_dev = va.va_rdev;
359 
360 		/*
361 		 * Get partition information for the component.
362 		 */
363 		error = VOP_IOCTL(vpp[ix], DIOCGPART, &dpart,
364 		    FREAD, l->l_cred);
365 		if (error) {
366 #ifdef DEBUG
367 			if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
368 				 printf("%s: %s: ioctl failed, error = %d\n",
369 				     cs->sc_xname, ci->ci_path, error);
370 #endif
371 			goto out;
372 		}
373 
374 /*
375  * This diagnostic test is disabled (for now?) since not all port supports
376  * on-disk BSD disklabel.
377  */
378 #if 0 /* def DIAGNOSTIC */
379 		/* Check fstype field of component. */
380 		if (dpart.part->p_fstype != FS_CCD)
381 			printf("%s: WARNING: %s: fstype %d != FS_CCD\n",
382 			    cs->sc_xname, ci->ci_path, dpart.part->p_fstype);
383 #endif
384 
385 		/*
386 		 * Calculate the size, truncating to an interleave
387 		 * boundary if necessary.
388 		 */
389 		maxsecsize =
390 		    ((dpart.disklab->d_secsize > maxsecsize) ?
391 		    dpart.disklab->d_secsize : maxsecsize);
392 		size = dpart.part->p_size;
393 		if (cs->sc_ileave > 1)
394 			size -= size % cs->sc_ileave;
395 
396 		if (size == 0) {
397 #ifdef DEBUG
398 			if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
399 				printf("%s: %s: size == 0\n",
400 				    cs->sc_xname, ci->ci_path);
401 #endif
402 			error = ENODEV;
403 			goto out;
404 		}
405 
406 		if (minsize == 0 || size < minsize)
407 			minsize = size;
408 		ci->ci_size = size;
409 		cs->sc_size += size;
410 	}
411 
412 	/*
413 	 * Don't allow the interleave to be smaller than
414 	 * the biggest component sector.
415 	 */
416 	if ((cs->sc_ileave > 0) &&
417 	    (cs->sc_ileave < (maxsecsize / DEV_BSIZE))) {
418 #ifdef DEBUG
419 		if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
420 			printf("%s: interleave must be at least %d\n",
421 			    cs->sc_xname, (maxsecsize / DEV_BSIZE));
422 #endif
423 		error = EINVAL;
424 		goto out;
425 	}
426 
427 	/*
428 	 * If uniform interleave is desired set all sizes to that of
429 	 * the smallest component.
430 	 */
431 	if (cs->sc_flags & CCDF_UNIFORM) {
432 		for (ci = cs->sc_cinfo;
433 		     ci < &cs->sc_cinfo[cs->sc_nccdisks]; ci++)
434 			ci->ci_size = minsize;
435 
436 		cs->sc_size = cs->sc_nccdisks * minsize;
437 	}
438 
439 	/*
440 	 * Construct the interleave table.
441 	 */
442 	ccdinterleave(cs);
443 
444 	/*
445 	 * Create pseudo-geometry based on 1MB cylinders.  It's
446 	 * pretty close.
447 	 */
448 	ccg->ccg_secsize = DEV_BSIZE;
449 	ccg->ccg_ntracks = 1;
450 	ccg->ccg_nsectors = 1024 * (1024 / ccg->ccg_secsize);
451 	ccg->ccg_ncylinders = cs->sc_size / ccg->ccg_nsectors;
452 
453 	/*
454 	 * Create thread to handle deferred I/O.
455 	 */
456 	cs->sc_zap = false;
457 	error = kthread_create(PRI_BIO, KTHREAD_MPSAFE, NULL, ccdthread,
458 	    cs, &cs->sc_thread, "%s", cs->sc_xname);
459 	if (error) {
460 		printf("ccdinit: can't create thread: %d\n", error);
461 		goto out;
462 	}
463 
464 	/*
465 	 * Only now that everything is set up can we enable the device.
466 	 */
467 	mutex_enter(cs->sc_iolock);
468 	cs->sc_flags |= CCDF_INITED;
469 	mutex_exit(cs->sc_iolock);
470 	kmem_free(tmppath, MAXPATHLEN);
471 	return (0);
472 
473  out:
474 	for (ix = 0; ix < path_alloced; ix++) {
475 		kmem_free(cs->sc_cinfo[ix].ci_path,
476 		    cs->sc_cinfo[ix].ci_pathlen);
477 	}
478 	kmem_free(cs->sc_cinfo, cs->sc_nccdisks * sizeof(struct ccdcinfo));
479 	kmem_free(tmppath, MAXPATHLEN);
480 	return (error);
481 }
482 
483 static void
484 ccdinterleave(struct ccd_softc *cs)
485 {
486 	struct ccdcinfo *ci, *smallci;
487 	struct ccdiinfo *ii;
488 	daddr_t bn, lbn;
489 	int ix;
490 	u_long size;
491 
492 #ifdef DEBUG
493 	if (ccddebug & CCDB_INIT)
494 		printf("ccdinterleave(%p): ileave %d\n", cs, cs->sc_ileave);
495 #endif
496 	/*
497 	 * Allocate an interleave table.
498 	 * Chances are this is too big, but we don't care.
499 	 */
500 	size = (cs->sc_nccdisks + 1) * sizeof(struct ccdiinfo);
501 	cs->sc_itable = kmem_zalloc(size, KM_SLEEP);
502 
503 	/*
504 	 * Trivial case: no interleave (actually interleave of disk size).
505 	 * Each table entry represents a single component in its entirety.
506 	 */
507 	if (cs->sc_ileave == 0) {
508 		bn = 0;
509 		ii = cs->sc_itable;
510 
511 		for (ix = 0; ix < cs->sc_nccdisks; ix++) {
512 			/* Allocate space for ii_index. */
513 			ii->ii_indexsz = sizeof(int);
514 			ii->ii_index = kmem_alloc(ii->ii_indexsz, KM_SLEEP);
515 			ii->ii_ndisk = 1;
516 			ii->ii_startblk = bn;
517 			ii->ii_startoff = 0;
518 			ii->ii_index[0] = ix;
519 			bn += cs->sc_cinfo[ix].ci_size;
520 			ii++;
521 		}
522 		ii->ii_ndisk = 0;
523 #ifdef DEBUG
524 		if (ccddebug & CCDB_INIT)
525 			printiinfo(cs->sc_itable);
526 #endif
527 		return;
528 	}
529 
530 	/*
531 	 * The following isn't fast or pretty; it doesn't have to be.
532 	 */
533 	size = 0;
534 	bn = lbn = 0;
535 	for (ii = cs->sc_itable; ; ii++) {
536 		/* Allocate space for ii_index. */
537 		ii->ii_indexsz = sizeof(int) * cs->sc_nccdisks;
538 		ii->ii_index = kmem_alloc(ii->ii_indexsz, KM_SLEEP);
539 
540 		/*
541 		 * Locate the smallest of the remaining components
542 		 */
543 		smallci = NULL;
544 		for (ci = cs->sc_cinfo;
545 		     ci < &cs->sc_cinfo[cs->sc_nccdisks]; ci++)
546 			if (ci->ci_size > size &&
547 			    (smallci == NULL ||
548 			     ci->ci_size < smallci->ci_size))
549 				smallci = ci;
550 
551 		/*
552 		 * Nobody left, all done
553 		 */
554 		if (smallci == NULL) {
555 			ii->ii_ndisk = 0;
556 			break;
557 		}
558 
559 		/*
560 		 * Record starting logical block and component offset
561 		 */
562 		ii->ii_startblk = bn / cs->sc_ileave;
563 		ii->ii_startoff = lbn;
564 
565 		/*
566 		 * Determine how many disks take part in this interleave
567 		 * and record their indices.
568 		 */
569 		ix = 0;
570 		for (ci = cs->sc_cinfo;
571 		     ci < &cs->sc_cinfo[cs->sc_nccdisks]; ci++)
572 			if (ci->ci_size >= smallci->ci_size)
573 				ii->ii_index[ix++] = ci - cs->sc_cinfo;
574 		ii->ii_ndisk = ix;
575 		bn += ix * (smallci->ci_size - size);
576 		lbn = smallci->ci_size / cs->sc_ileave;
577 		size = smallci->ci_size;
578 	}
579 #ifdef DEBUG
580 	if (ccddebug & CCDB_INIT)
581 		printiinfo(cs->sc_itable);
582 #endif
583 }
584 
585 /* ARGSUSED */
586 static int
587 ccdopen(dev_t dev, int flags, int fmt, struct lwp *l)
588 {
589 	int unit = ccdunit(dev);
590 	struct ccd_softc *cs;
591 	struct disklabel *lp;
592 	int error = 0, part, pmask;
593 
594 #ifdef DEBUG
595 	if (ccddebug & CCDB_FOLLOW)
596 		printf("ccdopen(0x%"PRIx64", 0x%x)\n", dev, flags);
597 #endif
598 	if (unit >= numccd)
599 		return (ENXIO);
600 	cs = &ccd_softc[unit];
601 
602 	mutex_enter(&cs->sc_dvlock);
603 
604 	lp = cs->sc_dkdev.dk_label;
605 
606 	part = DISKPART(dev);
607 	pmask = (1 << part);
608 
609 	/*
610 	 * If we're initialized, check to see if there are any other
611 	 * open partitions.  If not, then it's safe to update
612 	 * the in-core disklabel.  Only read the disklabel if it is
613 	 * not already valid.
614 	 */
615 	if ((cs->sc_flags & (CCDF_INITED|CCDF_VLABEL)) == CCDF_INITED &&
616 	    cs->sc_dkdev.dk_openmask == 0)
617 		ccdgetdisklabel(dev);
618 
619 	/* Check that the partition exists. */
620 	if (part != RAW_PART) {
621 		if (((cs->sc_flags & CCDF_INITED) == 0) ||
622 		    ((part >= lp->d_npartitions) ||
623 		     (lp->d_partitions[part].p_fstype == FS_UNUSED))) {
624 			error = ENXIO;
625 			goto done;
626 		}
627 	}
628 
629 	/* Prevent our unit from being unconfigured while open. */
630 	switch (fmt) {
631 	case S_IFCHR:
632 		cs->sc_dkdev.dk_copenmask |= pmask;
633 		break;
634 
635 	case S_IFBLK:
636 		cs->sc_dkdev.dk_bopenmask |= pmask;
637 		break;
638 	}
639 	cs->sc_dkdev.dk_openmask =
640 	    cs->sc_dkdev.dk_copenmask | cs->sc_dkdev.dk_bopenmask;
641 
642  done:
643 	mutex_exit(&cs->sc_dvlock);
644 	return (error);
645 }
646 
647 /* ARGSUSED */
648 static int
649 ccdclose(dev_t dev, int flags, int fmt, struct lwp *l)
650 {
651 	int unit = ccdunit(dev);
652 	struct ccd_softc *cs;
653 	int part;
654 
655 #ifdef DEBUG
656 	if (ccddebug & CCDB_FOLLOW)
657 		printf("ccdclose(0x%"PRIx64", 0x%x)\n", dev, flags);
658 #endif
659 
660 	if (unit >= numccd)
661 		return (ENXIO);
662 	cs = &ccd_softc[unit];
663 
664 	mutex_enter(&cs->sc_dvlock);
665 
666 	part = DISKPART(dev);
667 
668 	/* ...that much closer to allowing unconfiguration... */
669 	switch (fmt) {
670 	case S_IFCHR:
671 		cs->sc_dkdev.dk_copenmask &= ~(1 << part);
672 		break;
673 
674 	case S_IFBLK:
675 		cs->sc_dkdev.dk_bopenmask &= ~(1 << part);
676 		break;
677 	}
678 	cs->sc_dkdev.dk_openmask =
679 	    cs->sc_dkdev.dk_copenmask | cs->sc_dkdev.dk_bopenmask;
680 
681 	if (cs->sc_dkdev.dk_openmask == 0) {
682 		if ((cs->sc_flags & CCDF_KLABEL) == 0)
683 			cs->sc_flags &= ~CCDF_VLABEL;
684 	}
685 
686 	mutex_exit(&cs->sc_dvlock);
687 	return (0);
688 }
689 
690 static bool
691 ccdbackoff(struct ccd_softc *cs)
692 {
693 
694 	/* XXX Arbitrary, should be a uvm call. */
695 	return uvmexp.free < (uvmexp.freemin >> 1) &&
696 	    disk_isbusy(&cs->sc_dkdev);
697 }
698 
699 static void
700 ccdthread(void *cookie)
701 {
702 	struct ccd_softc *cs;
703 
704 	cs = cookie;
705 
706 #ifdef DEBUG
707  	if (ccddebug & CCDB_FOLLOW)
708  		printf("ccdthread: hello\n");
709 #endif
710 
711 	mutex_enter(cs->sc_iolock);
712 	while (__predict_true(!cs->sc_zap)) {
713 		if (bufq_peek(cs->sc_bufq) == NULL) {
714 			/* Nothing to do. */
715 			cv_wait(&cs->sc_push, cs->sc_iolock);
716 			continue;
717 		}
718 		if (ccdbackoff(cs)) {
719 			/* Wait for memory to become available. */
720 			(void)cv_timedwait(&cs->sc_push, cs->sc_iolock, 1);
721 			continue;
722 		}
723 #ifdef DEBUG
724  		if (ccddebug & CCDB_FOLLOW)
725  			printf("ccdthread: dispatching I/O\n");
726 #endif
727 		ccdstart(cs);
728 		mutex_enter(cs->sc_iolock);
729 	}
730 	cs->sc_thread = NULL;
731 	mutex_exit(cs->sc_iolock);
732 #ifdef DEBUG
733  	if (ccddebug & CCDB_FOLLOW)
734  		printf("ccdthread: goodbye\n");
735 #endif
736 	kthread_exit(0);
737 }
738 
739 static void
740 ccdstrategy(struct buf *bp)
741 {
742 	int unit = ccdunit(bp->b_dev);
743 	struct ccd_softc *cs = &ccd_softc[unit];
744 
745 	/* Must be open or reading label. */
746 	KASSERT(cs->sc_dkdev.dk_openmask != 0 ||
747 	    (cs->sc_flags & CCDF_RLABEL) != 0);
748 
749 	mutex_enter(cs->sc_iolock);
750 	/* Synchronize with device init/uninit. */
751 	if (__predict_false((cs->sc_flags & CCDF_INITED) == 0)) {
752 		mutex_exit(cs->sc_iolock);
753 #ifdef DEBUG
754  		if (ccddebug & CCDB_FOLLOW)
755  			printf("ccdstrategy: unit %d: not inited\n", unit);
756 #endif
757  		bp->b_error = ENXIO;
758  		bp->b_resid = bp->b_bcount;
759  		biodone(bp);
760 		return;
761 	}
762 
763 	/* Defer to thread if system is low on memory. */
764 	bufq_put(cs->sc_bufq, bp);
765 	if (__predict_false(ccdbackoff(cs))) {
766 		mutex_exit(cs->sc_iolock);
767 #ifdef DEBUG
768  		if (ccddebug & CCDB_FOLLOW)
769  			printf("ccdstrategy: holding off on I/O\n");
770 #endif
771 		return;
772 	}
773 	ccdstart(cs);
774 }
775 
776 static void
777 ccdstart(struct ccd_softc *cs)
778 {
779 	daddr_t blkno;
780 	int wlabel;
781 	struct disklabel *lp;
782 	long bcount, rcount;
783 	struct ccdbuf *cbp;
784 	char *addr;
785 	daddr_t bn;
786 	vnode_t *vp;
787 	buf_t *bp;
788 
789 	KASSERT(mutex_owned(cs->sc_iolock));
790 
791 	disk_busy(&cs->sc_dkdev);
792 	bp = bufq_get(cs->sc_bufq);
793 	KASSERT(bp != NULL);
794 
795 #ifdef DEBUG
796 	if (ccddebug & CCDB_FOLLOW)
797 		printf("ccdstart(%s, %p)\n", cs->sc_xname, bp);
798 #endif
799 
800 	/* If it's a nil transfer, wake up the top half now. */
801 	if (bp->b_bcount == 0)
802 		goto done;
803 
804 	lp = cs->sc_dkdev.dk_label;
805 
806 	/*
807 	 * Do bounds checking and adjust transfer.  If there's an
808 	 * error, the bounds check will flag that for us.  Convert
809 	 * the partition relative block number to an absolute.
810 	 */
811 	blkno = bp->b_blkno;
812 	wlabel = cs->sc_flags & (CCDF_WLABEL|CCDF_LABELLING);
813 	if (DISKPART(bp->b_dev) != RAW_PART) {
814 		if (bounds_check_with_label(&cs->sc_dkdev, bp, wlabel) <= 0)
815 			goto done;
816 		blkno += lp->d_partitions[DISKPART(bp->b_dev)].p_offset;
817 	}
818 	mutex_exit(cs->sc_iolock);
819 	bp->b_rawblkno = blkno;
820 
821 	/* Allocate the component buffers and start I/O! */
822 	bp->b_resid = bp->b_bcount;
823 	bn = bp->b_rawblkno;
824 	addr = bp->b_data;
825 	for (bcount = bp->b_bcount; bcount > 0; bcount -= rcount) {
826 		cbp = ccdbuffer(cs, bp, bn, addr, bcount);
827 		rcount = cbp->cb_buf.b_bcount;
828 		bn += btodb(rcount);
829 		addr += rcount;
830 		vp = cbp->cb_buf.b_vp;
831 		if ((cbp->cb_buf.b_flags & B_READ) == 0) {
832 			mutex_enter(&vp->v_interlock);
833 			vp->v_numoutput++;
834 			mutex_exit(&vp->v_interlock);
835 		}
836 		(void)VOP_STRATEGY(vp, &cbp->cb_buf);
837 	}
838 	return;
839 
840  done:
841 	disk_unbusy(&cs->sc_dkdev, 0, 0);
842 	cv_broadcast(&cs->sc_stop);
843 	cv_broadcast(&cs->sc_push);
844 	mutex_exit(cs->sc_iolock);
845 	bp->b_resid = bp->b_bcount;
846 	biodone(bp);
847 }
848 
849 /*
850  * Build a component buffer header.
851  */
852 static struct ccdbuf *
853 ccdbuffer(struct ccd_softc *cs, struct buf *bp, daddr_t bn, void *addr,
854     long bcount)
855 {
856 	struct ccdcinfo *ci;
857 	struct ccdbuf *cbp;
858 	daddr_t cbn, cboff;
859 	u_int64_t cbc;
860 	int ccdisk;
861 
862 #ifdef DEBUG
863 	if (ccddebug & CCDB_IO)
864 		printf("ccdbuffer(%p, %p, %" PRId64 ", %p, %ld)\n",
865 		       cs, bp, bn, addr, bcount);
866 #endif
867 	/*
868 	 * Determine which component bn falls in.
869 	 */
870 	cbn = bn;
871 	cboff = 0;
872 
873 	/*
874 	 * Serially concatenated
875 	 */
876 	if (cs->sc_ileave == 0) {
877 		daddr_t sblk;
878 
879 		sblk = 0;
880 		for (ccdisk = 0, ci = &cs->sc_cinfo[ccdisk];
881 		    cbn >= sblk + ci->ci_size;
882 		    ccdisk++, ci = &cs->sc_cinfo[ccdisk])
883 			sblk += ci->ci_size;
884 		cbn -= sblk;
885 	}
886 	/*
887 	 * Interleaved
888 	 */
889 	else {
890 		struct ccdiinfo *ii;
891 		int off;
892 
893 		cboff = cbn % cs->sc_ileave;
894 		cbn /= cs->sc_ileave;
895 		for (ii = cs->sc_itable; ii->ii_ndisk; ii++)
896 			if (ii->ii_startblk > cbn)
897 				break;
898 		ii--;
899 		off = cbn - ii->ii_startblk;
900 		if (ii->ii_ndisk == 1) {
901 			ccdisk = ii->ii_index[0];
902 			cbn = ii->ii_startoff + off;
903 		} else {
904 			ccdisk = ii->ii_index[off % ii->ii_ndisk];
905 			cbn = ii->ii_startoff + off / ii->ii_ndisk;
906 		}
907 		cbn *= cs->sc_ileave;
908 		ci = &cs->sc_cinfo[ccdisk];
909 	}
910 
911 	/*
912 	 * Fill in the component buf structure.
913 	 */
914 	cbp = CCD_GETBUF();
915 	KASSERT(cbp != NULL);
916 	buf_init(&cbp->cb_buf);
917 	cbp->cb_buf.b_flags = bp->b_flags;
918 	cbp->cb_buf.b_oflags = bp->b_oflags;
919 	cbp->cb_buf.b_cflags = bp->b_cflags;
920 	cbp->cb_buf.b_iodone = ccdiodone;
921 	cbp->cb_buf.b_proc = bp->b_proc;
922 	cbp->cb_buf.b_dev = ci->ci_dev;
923 	cbp->cb_buf.b_blkno = cbn + cboff;
924 	cbp->cb_buf.b_data = addr;
925 	cbp->cb_buf.b_vp = ci->ci_vp;
926 	cbp->cb_buf.b_objlock = &ci->ci_vp->v_interlock;
927 	if (cs->sc_ileave == 0)
928 		cbc = dbtob((u_int64_t)(ci->ci_size - cbn));
929 	else
930 		cbc = dbtob((u_int64_t)(cs->sc_ileave - cboff));
931 	cbp->cb_buf.b_bcount = cbc < bcount ? cbc : bcount;
932 
933 	/*
934 	 * context for ccdiodone
935 	 */
936 	cbp->cb_obp = bp;
937 	cbp->cb_sc = cs;
938 	cbp->cb_comp = ccdisk;
939 
940 	BIO_COPYPRIO(&cbp->cb_buf, bp);
941 
942 #ifdef DEBUG
943 	if (ccddebug & CCDB_IO)
944 		printf(" dev 0x%"PRIx64"(u%lu): cbp %p bn %" PRId64 " addr %p"
945 		       " bcnt %d\n",
946 		    ci->ci_dev, (unsigned long) (ci-cs->sc_cinfo), cbp,
947 		    cbp->cb_buf.b_blkno, cbp->cb_buf.b_data,
948 		    cbp->cb_buf.b_bcount);
949 #endif
950 
951 	return (cbp);
952 }
953 
954 /*
955  * Called at interrupt time.
956  * Mark the component as done and if all components are done,
957  * take a ccd interrupt.
958  */
959 static void
960 ccdiodone(struct buf *vbp)
961 {
962 	struct ccdbuf *cbp = (struct ccdbuf *) vbp;
963 	struct buf *bp = cbp->cb_obp;
964 	struct ccd_softc *cs = cbp->cb_sc;
965 	int count;
966 
967 #ifdef DEBUG
968 	if (ccddebug & CCDB_FOLLOW)
969 		printf("ccdiodone(%p)\n", cbp);
970 	if (ccddebug & CCDB_IO) {
971 		printf("ccdiodone: bp %p bcount %d resid %d\n",
972 		       bp, bp->b_bcount, bp->b_resid);
973 		printf(" dev 0x%"PRIx64"(u%d), cbp %p bn %" PRId64 " addr %p"
974 		       " bcnt %d\n",
975 		       cbp->cb_buf.b_dev, cbp->cb_comp, cbp,
976 		       cbp->cb_buf.b_blkno, cbp->cb_buf.b_data,
977 		       cbp->cb_buf.b_bcount);
978 	}
979 #endif
980 
981 	if (cbp->cb_buf.b_error != 0) {
982 		bp->b_error = cbp->cb_buf.b_error;
983 		printf("%s: error %d on component %d\n",
984 		       cs->sc_xname, bp->b_error, cbp->cb_comp);
985 	}
986 	count = cbp->cb_buf.b_bcount;
987 	buf_destroy(&cbp->cb_buf);
988 	CCD_PUTBUF(cbp);
989 
990 	/*
991 	 * If all done, "interrupt".
992 	 */
993 	mutex_enter(cs->sc_iolock);
994 	bp->b_resid -= count;
995 	if (bp->b_resid < 0)
996 		panic("ccdiodone: count");
997 	if (bp->b_resid == 0) {
998 		/*
999 		 * Request is done for better or worse, wakeup the top half.
1000 		 */
1001 		if (bp->b_error != 0)
1002 			bp->b_resid = bp->b_bcount;
1003 		disk_unbusy(&cs->sc_dkdev, (bp->b_bcount - bp->b_resid),
1004 		    (bp->b_flags & B_READ));
1005 		if (!disk_isbusy(&cs->sc_dkdev)) {
1006 			if (bufq_peek(cs->sc_bufq) != NULL) {
1007 				cv_broadcast(&cs->sc_push);
1008 			}
1009 			cv_broadcast(&cs->sc_stop);
1010 		}
1011 		mutex_exit(cs->sc_iolock);
1012 		biodone(bp);
1013 	} else
1014 		mutex_exit(cs->sc_iolock);
1015 }
1016 
1017 /* ARGSUSED */
1018 static int
1019 ccdread(dev_t dev, struct uio *uio, int flags)
1020 {
1021 	int unit = ccdunit(dev);
1022 	struct ccd_softc *cs;
1023 
1024 #ifdef DEBUG
1025 	if (ccddebug & CCDB_FOLLOW)
1026 		printf("ccdread(0x%"PRIx64", %p)\n", dev, uio);
1027 #endif
1028 	if (unit >= numccd)
1029 		return (ENXIO);
1030 	cs = &ccd_softc[unit];
1031 
1032 	/* Unlocked advisory check, ccdstrategy check is synchronous. */
1033 	if ((cs->sc_flags & CCDF_INITED) == 0)
1034 		return (ENXIO);
1035 
1036 	return (physio(ccdstrategy, NULL, dev, B_READ, minphys, uio));
1037 }
1038 
1039 /* ARGSUSED */
1040 static int
1041 ccdwrite(dev_t dev, struct uio *uio, int flags)
1042 {
1043 	int unit = ccdunit(dev);
1044 	struct ccd_softc *cs;
1045 
1046 #ifdef DEBUG
1047 	if (ccddebug & CCDB_FOLLOW)
1048 		printf("ccdwrite(0x%"PRIx64", %p)\n", dev, uio);
1049 #endif
1050 	if (unit >= numccd)
1051 		return (ENXIO);
1052 	cs = &ccd_softc[unit];
1053 
1054 	/* Unlocked advisory check, ccdstrategy check is synchronous. */
1055 	if ((cs->sc_flags & CCDF_INITED) == 0)
1056 		return (ENXIO);
1057 
1058 	return (physio(ccdstrategy, NULL, dev, B_WRITE, minphys, uio));
1059 }
1060 
1061 static int
1062 ccdioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l)
1063 {
1064 	int unit = ccdunit(dev);
1065 	int i, j, lookedup = 0, error = 0;
1066 	int part, pmask;
1067 	struct ccd_softc *cs;
1068 	struct ccd_ioctl *ccio = (struct ccd_ioctl *)data;
1069 	kauth_cred_t uc;
1070 	char **cpp;
1071 	struct vnode **vpp;
1072 #ifdef __HAVE_OLD_DISKLABEL
1073 	struct disklabel newlabel;
1074 #endif
1075 
1076 	if (unit >= numccd)
1077 		return (ENXIO);
1078 	cs = &ccd_softc[unit];
1079 	uc = kauth_cred_get();
1080 
1081 	/* Must be open for writes for these commands... */
1082 	switch (cmd) {
1083 	case CCDIOCSET:
1084 	case CCDIOCCLR:
1085 	case DIOCSDINFO:
1086 	case DIOCWDINFO:
1087 #ifdef __HAVE_OLD_DISKLABEL
1088 	case ODIOCSDINFO:
1089 	case ODIOCWDINFO:
1090 #endif
1091 	case DIOCKLABEL:
1092 	case DIOCWLABEL:
1093 		if ((flag & FWRITE) == 0)
1094 			return (EBADF);
1095 	}
1096 
1097 	mutex_enter(&cs->sc_dvlock);
1098 
1099 	/* Must be initialized for these... */
1100 	switch (cmd) {
1101 	case CCDIOCCLR:
1102 	case DIOCGDINFO:
1103 	case DIOCCACHESYNC:
1104 	case DIOCSDINFO:
1105 	case DIOCWDINFO:
1106 	case DIOCGPART:
1107 	case DIOCWLABEL:
1108 	case DIOCKLABEL:
1109 	case DIOCGDEFLABEL:
1110 #ifdef __HAVE_OLD_DISKLABEL
1111 	case ODIOCGDINFO:
1112 	case ODIOCSDINFO:
1113 	case ODIOCWDINFO:
1114 	case ODIOCGDEFLABEL:
1115 #endif
1116 		if ((cs->sc_flags & CCDF_INITED) == 0) {
1117 			error = ENXIO;
1118 			goto out;
1119 		}
1120 	}
1121 
1122 	switch (cmd) {
1123 	case CCDIOCSET:
1124 		if (cs->sc_flags & CCDF_INITED) {
1125 			error = EBUSY;
1126 			goto out;
1127 		}
1128 
1129 		/* Validate the flags. */
1130 		if ((ccio->ccio_flags & CCDF_USERMASK) != ccio->ccio_flags) {
1131 			error = EINVAL;
1132 			goto out;
1133 		}
1134 
1135 		if (ccio->ccio_ndisks > CCD_MAXNDISKS ||
1136 		    ccio->ccio_ndisks == 0) {
1137 			error = EINVAL;
1138 			goto out;
1139 		}
1140 
1141 		/* Fill in some important bits. */
1142 		cs->sc_ileave = ccio->ccio_ileave;
1143 		cs->sc_nccdisks = ccio->ccio_ndisks;
1144 		cs->sc_flags = ccio->ccio_flags & CCDF_USERMASK;
1145 
1146 		/*
1147 		 * Allocate space for and copy in the array of
1148 		 * componet pathnames and device numbers.
1149 		 */
1150 		cpp = kmem_alloc(ccio->ccio_ndisks * sizeof(*cpp), KM_SLEEP);
1151 		vpp = kmem_alloc(ccio->ccio_ndisks * sizeof(*vpp), KM_SLEEP);
1152 		error = copyin(ccio->ccio_disks, cpp,
1153 		    ccio->ccio_ndisks * sizeof(*cpp));
1154 		if (error) {
1155 			kmem_free(vpp, ccio->ccio_ndisks * sizeof(*vpp));
1156 			kmem_free(cpp, ccio->ccio_ndisks * sizeof(*cpp));
1157 			goto out;
1158 		}
1159 
1160 #ifdef DEBUG
1161 		if (ccddebug & CCDB_INIT)
1162 			for (i = 0; i < ccio->ccio_ndisks; ++i)
1163 				printf("ccdioctl: component %d: %p\n",
1164 				    i, cpp[i]);
1165 #endif
1166 
1167 		for (i = 0; i < ccio->ccio_ndisks; ++i) {
1168 #ifdef DEBUG
1169 			if (ccddebug & CCDB_INIT)
1170 				printf("ccdioctl: lookedup = %d\n", lookedup);
1171 #endif
1172 			if ((error = dk_lookup(cpp[i], l, &vpp[i],
1173 			    UIO_USERSPACE)) != 0) {
1174 				for (j = 0; j < lookedup; ++j)
1175 					(void)vn_close(vpp[j], FREAD|FWRITE,
1176 					    uc);
1177 				kmem_free(vpp, ccio->ccio_ndisks *
1178 				    sizeof(*vpp));
1179 				kmem_free(cpp, ccio->ccio_ndisks *
1180 				    sizeof(*cpp));
1181 				goto out;
1182 			}
1183 			++lookedup;
1184 		}
1185 
1186 		/* Attach the disk. */
1187 		disk_attach(&cs->sc_dkdev);
1188 		bufq_alloc(&cs->sc_bufq, "fcfs", 0);
1189 
1190 		/*
1191 		 * Initialize the ccd.  Fills in the softc for us.
1192 		 */
1193 		if ((error = ccdinit(cs, cpp, vpp, l)) != 0) {
1194 			for (j = 0; j < lookedup; ++j)
1195 				(void)vn_close(vpp[j], FREAD|FWRITE,
1196 				    uc);
1197 			kmem_free(vpp, ccio->ccio_ndisks * sizeof(*vpp));
1198 			kmem_free(cpp, ccio->ccio_ndisks * sizeof(*cpp));
1199 			disk_detach(&cs->sc_dkdev);
1200 			bufq_free(cs->sc_bufq);
1201 			goto out;
1202 		}
1203 
1204 		/* We can free the temporary variables now. */
1205 		kmem_free(vpp, ccio->ccio_ndisks * sizeof(*vpp));
1206 		kmem_free(cpp, ccio->ccio_ndisks * sizeof(*cpp));
1207 
1208 		/*
1209 		 * The ccd has been successfully initialized, so
1210 		 * we can place it into the array.  Don't try to
1211 		 * read the disklabel until the disk has been attached,
1212 		 * because space for the disklabel is allocated
1213 		 * in disk_attach();
1214 		 */
1215 		ccio->ccio_unit = unit;
1216 		ccio->ccio_size = cs->sc_size;
1217 
1218 		/* Try and read the disklabel. */
1219 		ccdgetdisklabel(dev);
1220 		break;
1221 
1222 	case CCDIOCCLR:
1223 		/*
1224 		 * Don't unconfigure if any other partitions are open
1225 		 * or if both the character and block flavors of this
1226 		 * partition are open.
1227 		 */
1228 		part = DISKPART(dev);
1229 		pmask = (1 << part);
1230 		if ((cs->sc_dkdev.dk_openmask & ~pmask) ||
1231 		    ((cs->sc_dkdev.dk_bopenmask & pmask) &&
1232 		    (cs->sc_dkdev.dk_copenmask & pmask))) {
1233 			error = EBUSY;
1234 			goto out;
1235 		}
1236 
1237 		/* Stop new I/O, wait for in-flight I/O to complete. */
1238 		mutex_enter(cs->sc_iolock);
1239 		cs->sc_flags &= ~(CCDF_INITED|CCDF_VLABEL);
1240 		cs->sc_zap = true;
1241 		while (disk_isbusy(&cs->sc_dkdev) ||
1242 		    bufq_peek(cs->sc_bufq) != NULL ||
1243 		    cs->sc_thread != NULL) {
1244 			cv_broadcast(&cs->sc_push);
1245 			(void)cv_timedwait(&cs->sc_stop, cs->sc_iolock, hz);
1246 		}
1247 		mutex_exit(cs->sc_iolock);
1248 
1249 		/*
1250 		 * Free ccd_softc information and clear entry.
1251 		 */
1252 
1253 		/* Close the components and free their pathnames. */
1254 		for (i = 0; i < cs->sc_nccdisks; ++i) {
1255 			/*
1256 			 * XXX: this close could potentially fail and
1257 			 * cause Bad Things.  Maybe we need to force
1258 			 * the close to happen?
1259 			 */
1260 #ifdef DEBUG
1261 			if (ccddebug & CCDB_VNODE)
1262 				vprint("CCDIOCCLR: vnode info",
1263 				    cs->sc_cinfo[i].ci_vp);
1264 #endif
1265 			(void)vn_close(cs->sc_cinfo[i].ci_vp, FREAD|FWRITE,
1266 			    uc);
1267 			kmem_free(cs->sc_cinfo[i].ci_path,
1268 			    cs->sc_cinfo[i].ci_pathlen);
1269 		}
1270 
1271 		/* Free interleave index. */
1272 		for (i = 0; cs->sc_itable[i].ii_ndisk; ++i) {
1273 			kmem_free(cs->sc_itable[i].ii_index,
1274 			    cs->sc_itable[i].ii_indexsz);
1275 		}
1276 
1277 		/* Free component info and interleave table. */
1278 		kmem_free(cs->sc_cinfo, cs->sc_nccdisks *
1279 		    sizeof(struct ccdcinfo));
1280 		kmem_free(cs->sc_itable, (cs->sc_nccdisks + 1) *
1281 		    sizeof(struct ccdiinfo));
1282 
1283 		/* Detatch the disk. */
1284 		disk_detach(&cs->sc_dkdev);
1285 		bufq_free(cs->sc_bufq);
1286 		break;
1287 
1288 	case DIOCGDINFO:
1289 		*(struct disklabel *)data = *(cs->sc_dkdev.dk_label);
1290 		break;
1291 
1292 #ifdef __HAVE_OLD_DISKLABEL
1293 	case ODIOCGDINFO:
1294 		newlabel = *(cs->sc_dkdev.dk_label);
1295 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1296 			return ENOTTY;
1297 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1298 		break;
1299 #endif
1300 
1301 	case DIOCGPART:
1302 		((struct partinfo *)data)->disklab = cs->sc_dkdev.dk_label;
1303 		((struct partinfo *)data)->part =
1304 		    &cs->sc_dkdev.dk_label->d_partitions[DISKPART(dev)];
1305 		break;
1306 
1307 	case DIOCCACHESYNC:
1308 		/*
1309 		 * XXX Do we really need to care about having a writable
1310 		 * file descriptor here?
1311 		 */
1312 		if ((flag & FWRITE) == 0)
1313 			return (EBADF);
1314 
1315 		/*
1316 		 * We pass this call down to all components and report
1317 		 * the first error we encounter.
1318 		 */
1319 		for (error = 0, i = 0; i < cs->sc_nccdisks; i++) {
1320 			j = VOP_IOCTL(cs->sc_cinfo[i].ci_vp, cmd, data,
1321 				      flag, uc);
1322 			if (j != 0 && error == 0)
1323 				error = j;
1324 		}
1325 		break;
1326 
1327 	case DIOCWDINFO:
1328 	case DIOCSDINFO:
1329 #ifdef __HAVE_OLD_DISKLABEL
1330 	case ODIOCWDINFO:
1331 	case ODIOCSDINFO:
1332 #endif
1333 	{
1334 		struct disklabel *lp;
1335 #ifdef __HAVE_OLD_DISKLABEL
1336 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
1337 			memset(&newlabel, 0, sizeof newlabel);
1338 			memcpy(&newlabel, data, sizeof (struct olddisklabel));
1339 			lp = &newlabel;
1340 		} else
1341 #endif
1342 		lp = (struct disklabel *)data;
1343 
1344 		cs->sc_flags |= CCDF_LABELLING;
1345 
1346 		error = setdisklabel(cs->sc_dkdev.dk_label,
1347 		    lp, 0, cs->sc_dkdev.dk_cpulabel);
1348 		if (error == 0) {
1349 			if (cmd == DIOCWDINFO
1350 #ifdef __HAVE_OLD_DISKLABEL
1351 			    || cmd == ODIOCWDINFO
1352 #endif
1353 			   )
1354 				error = writedisklabel(CCDLABELDEV(dev),
1355 				    ccdstrategy, cs->sc_dkdev.dk_label,
1356 				    cs->sc_dkdev.dk_cpulabel);
1357 		}
1358 
1359 		cs->sc_flags &= ~CCDF_LABELLING;
1360 		break;
1361 	}
1362 
1363 	case DIOCKLABEL:
1364 		if (*(int *)data != 0)
1365 			cs->sc_flags |= CCDF_KLABEL;
1366 		else
1367 			cs->sc_flags &= ~CCDF_KLABEL;
1368 		break;
1369 
1370 	case DIOCWLABEL:
1371 		if (*(int *)data != 0)
1372 			cs->sc_flags |= CCDF_WLABEL;
1373 		else
1374 			cs->sc_flags &= ~CCDF_WLABEL;
1375 		break;
1376 
1377 	case DIOCGDEFLABEL:
1378 		ccdgetdefaultlabel(cs, (struct disklabel *)data);
1379 		break;
1380 
1381 #ifdef __HAVE_OLD_DISKLABEL
1382 	case ODIOCGDEFLABEL:
1383 		ccdgetdefaultlabel(cs, &newlabel);
1384 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1385 			return ENOTTY;
1386 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1387 		break;
1388 #endif
1389 
1390 	default:
1391 		error = ENOTTY;
1392 	}
1393 
1394  out:
1395 	mutex_exit(&cs->sc_dvlock);
1396 	return (error);
1397 }
1398 
1399 static int
1400 ccdsize(dev_t dev)
1401 {
1402 	struct ccd_softc *cs;
1403 	struct disklabel *lp;
1404 	int part, unit, omask, size;
1405 
1406 	unit = ccdunit(dev);
1407 	if (unit >= numccd)
1408 		return (-1);
1409 	cs = &ccd_softc[unit];
1410 
1411 	if ((cs->sc_flags & CCDF_INITED) == 0)
1412 		return (-1);
1413 
1414 	part = DISKPART(dev);
1415 	omask = cs->sc_dkdev.dk_openmask & (1 << part);
1416 	lp = cs->sc_dkdev.dk_label;
1417 
1418 	if (omask == 0 && ccdopen(dev, 0, S_IFBLK, curlwp))
1419 		return (-1);
1420 
1421 	if (lp->d_partitions[part].p_fstype != FS_SWAP)
1422 		size = -1;
1423 	else
1424 		size = lp->d_partitions[part].p_size *
1425 		    (lp->d_secsize / DEV_BSIZE);
1426 
1427 	if (omask == 0 && ccdclose(dev, 0, S_IFBLK, curlwp))
1428 		return (-1);
1429 
1430 	return (size);
1431 }
1432 
1433 static void
1434 ccdgetdefaultlabel(struct ccd_softc *cs, struct disklabel *lp)
1435 {
1436 	struct ccdgeom *ccg = &cs->sc_geom;
1437 
1438 	memset(lp, 0, sizeof(*lp));
1439 
1440 	lp->d_secperunit = cs->sc_size;
1441 	lp->d_secsize = ccg->ccg_secsize;
1442 	lp->d_nsectors = ccg->ccg_nsectors;
1443 	lp->d_ntracks = ccg->ccg_ntracks;
1444 	lp->d_ncylinders = ccg->ccg_ncylinders;
1445 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1446 
1447 	strncpy(lp->d_typename, "ccd", sizeof(lp->d_typename));
1448 	lp->d_type = DTYPE_CCD;
1449 	strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
1450 	lp->d_rpm = 3600;
1451 	lp->d_interleave = 1;
1452 	lp->d_flags = 0;
1453 
1454 	lp->d_partitions[RAW_PART].p_offset = 0;
1455 	lp->d_partitions[RAW_PART].p_size = cs->sc_size;
1456 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
1457 	lp->d_npartitions = RAW_PART + 1;
1458 
1459 	lp->d_magic = DISKMAGIC;
1460 	lp->d_magic2 = DISKMAGIC;
1461 	lp->d_checksum = dkcksum(cs->sc_dkdev.dk_label);
1462 }
1463 
1464 /*
1465  * Read the disklabel from the ccd.  If one is not present, fake one
1466  * up.
1467  */
1468 static void
1469 ccdgetdisklabel(dev_t dev)
1470 {
1471 	int unit = ccdunit(dev);
1472 	struct ccd_softc *cs = &ccd_softc[unit];
1473 	const char *errstring;
1474 	struct disklabel *lp = cs->sc_dkdev.dk_label;
1475 	struct cpu_disklabel *clp = cs->sc_dkdev.dk_cpulabel;
1476 
1477 	KASSERT(mutex_owned(&cs->sc_dvlock));
1478 
1479 	memset(clp, 0, sizeof(*clp));
1480 
1481 	ccdgetdefaultlabel(cs, lp);
1482 
1483 	/*
1484 	 * Call the generic disklabel extraction routine.
1485 	 */
1486 	cs->sc_flags |= CCDF_RLABEL;
1487 	if ((cs->sc_flags & CCDF_NOLABEL) != 0)
1488 		errstring = "CCDF_NOLABEL set; ignoring on-disk label";
1489 	else
1490 		errstring = readdisklabel(CCDLABELDEV(dev), ccdstrategy,
1491 		    cs->sc_dkdev.dk_label, cs->sc_dkdev.dk_cpulabel);
1492 	if (errstring)
1493 		ccdmakedisklabel(cs);
1494 	else {
1495 		int i;
1496 		struct partition *pp;
1497 
1498 		/*
1499 		 * Sanity check whether the found disklabel is valid.
1500 		 *
1501 		 * This is necessary since total size of ccd may vary
1502 		 * when an interleave is changed even though exactly
1503 		 * same componets are used, and old disklabel may used
1504 		 * if that is found.
1505 		 */
1506 		if (lp->d_secperunit != cs->sc_size)
1507 			printf("WARNING: %s: "
1508 			    "total sector size in disklabel (%d) != "
1509 			    "the size of ccd (%lu)\n", cs->sc_xname,
1510 			    lp->d_secperunit, (u_long)cs->sc_size);
1511 		for (i = 0; i < lp->d_npartitions; i++) {
1512 			pp = &lp->d_partitions[i];
1513 			if (pp->p_offset + pp->p_size > cs->sc_size)
1514 				printf("WARNING: %s: end of partition `%c' "
1515 				    "exceeds the size of ccd (%lu)\n",
1516 				    cs->sc_xname, 'a' + i, (u_long)cs->sc_size);
1517 		}
1518 	}
1519 
1520 #ifdef DEBUG
1521 	/* It's actually extremely common to have unlabeled ccds. */
1522 	if (ccddebug & CCDB_LABEL)
1523 		if (errstring != NULL)
1524 			printf("%s: %s\n", cs->sc_xname, errstring);
1525 #endif
1526 
1527 	/* In-core label now valid. */
1528 	cs->sc_flags = (cs->sc_flags | CCDF_VLABEL) & ~CCDF_RLABEL;
1529 }
1530 
1531 /*
1532  * Take care of things one might want to take care of in the event
1533  * that a disklabel isn't present.
1534  */
1535 static void
1536 ccdmakedisklabel(struct ccd_softc *cs)
1537 {
1538 	struct disklabel *lp = cs->sc_dkdev.dk_label;
1539 
1540 	/*
1541 	 * For historical reasons, if there's no disklabel present
1542 	 * the raw partition must be marked FS_BSDFFS.
1543 	 */
1544 	lp->d_partitions[RAW_PART].p_fstype = FS_BSDFFS;
1545 
1546 	strncpy(lp->d_packname, "default label", sizeof(lp->d_packname));
1547 
1548 	lp->d_checksum = dkcksum(lp);
1549 }
1550 
1551 #ifdef DEBUG
1552 static void
1553 printiinfo(struct ccdiinfo *ii)
1554 {
1555 	int ix, i;
1556 
1557 	for (ix = 0; ii->ii_ndisk; ix++, ii++) {
1558 		printf(" itab[%d]: #dk %d sblk %" PRId64 " soff %" PRId64,
1559 		    ix, ii->ii_ndisk, ii->ii_startblk, ii->ii_startoff);
1560 		for (i = 0; i < ii->ii_ndisk; i++)
1561 			printf(" %d", ii->ii_index[i]);
1562 		printf("\n");
1563 	}
1564 }
1565 #endif
1566 
1567 #ifdef _MODULE
1568 
1569 #include <sys/module.h>
1570 
1571 MODULE(MODULE_CLASS_DRIVER, ccd, NULL);
1572 
1573 static int
1574 ccd_modcmd(modcmd_t cmd, void *arg)
1575 {
1576 	int bmajor = -1, cmajor = -1,  error = 0;
1577 
1578 	switch (cmd) {
1579 	case MODULE_CMD_INIT:
1580 		ccdattach(4);
1581 
1582 		return devsw_attach("ccd", &ccd_bdevsw, &bmajor,
1583 		    &ccd_cdevsw, &cmajor);
1584 		break;
1585 
1586 	case MODULE_CMD_FINI:
1587 		return devsw_detach(&ccd_bdevsw, &ccd_cdevsw);
1588 		break;
1589 
1590 	case MODULE_CMD_STAT:
1591 		return ENOTTY;
1592 
1593 	default:
1594 		return ENOTTY;
1595 	}
1596 
1597 	return error;
1598 }
1599 
1600 #endif
1601