xref: /netbsd-src/sys/dev/ccd.c (revision a5847cc334d9a7029f6352b847e9e8d71a0f9e0c)
1 /*	$NetBSD: ccd.c,v 1.142 2011/10/14 09:23:29 hannken 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) 1988 University of Utah.
34  * Copyright (c) 1990, 1993
35  *	The Regents of the University of California.  All rights reserved.
36  *
37  * This code is derived from software contributed to Berkeley by
38  * the Systems Programming Group of the University of Utah Computer
39  * Science Department.
40  *
41  * Redistribution and use in source and binary forms, with or without
42  * modification, are permitted provided that the following conditions
43  * are met:
44  * 1. Redistributions of source code must retain the above copyright
45  *    notice, this list of conditions and the following disclaimer.
46  * 2. Redistributions in binary form must reproduce the above copyright
47  *    notice, this list of conditions and the following disclaimer in the
48  *    documentation and/or other materials provided with the distribution.
49  * 3. Neither the name of the University nor the names of its contributors
50  *    may be used to endorse or promote products derived from this software
51  *    without specific prior written permission.
52  *
53  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
54  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
55  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
56  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
57  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
58  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
59  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
60  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
61  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
62  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
63  * SUCH DAMAGE.
64  *
65  * from: Utah $Hdr: cd.c 1.6 90/11/28$
66  *
67  *	@(#)cd.c	8.2 (Berkeley) 11/16/93
68  */
69 
70 /*
71  * "Concatenated" disk driver.
72  *
73  * Notes on concurrency:
74  *
75  * => sc_dvlock serializes access to the device nodes, excluding block I/O.
76  *
77  * => sc_iolock serializes access to (sc_flags & CCDF_INITED), disk stats,
78  *    sc_stop, sc_bufq and b_resid from master buffers.
79  *
80  * => a combination of CCDF_INITED, sc_inflight, and sc_iolock is used to
81  *    serialize I/O and configuration changes.
82  *
83  * => the in-core disk label does not change while the device is open.
84  *
85  * On memory consumption: ccd fans out I/O requests and so needs to
86  * allocate memory.  If the system is desperately low on memory, we
87  * single thread I/O.
88  */
89 
90 #include <sys/cdefs.h>
91 __KERNEL_RCSID(0, "$NetBSD: ccd.c,v 1.142 2011/10/14 09:23:29 hannken Exp $");
92 
93 #include <sys/param.h>
94 #include <sys/systm.h>
95 #include <sys/kernel.h>
96 #include <sys/proc.h>
97 #include <sys/errno.h>
98 #include <sys/buf.h>
99 #include <sys/kmem.h>
100 #include <sys/pool.h>
101 #include <sys/module.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/mutex.h>
113 #include <sys/queue.h>
114 #include <sys/kauth.h>
115 #include <sys/kthread.h>
116 #include <sys/bufq.h>
117 
118 #include <uvm/uvm_extern.h>
119 
120 #include <dev/ccdvar.h>
121 #include <dev/dkvar.h>
122 
123 #if defined(CCDDEBUG) && !defined(DEBUG)
124 #define DEBUG
125 #endif
126 
127 #ifdef DEBUG
128 #define CCDB_FOLLOW	0x01
129 #define CCDB_INIT	0x02
130 #define CCDB_IO		0x04
131 #define CCDB_LABEL	0x08
132 #define CCDB_VNODE	0x10
133 int ccddebug = 0x00;
134 #endif
135 
136 #define	ccdunit(x)	DISKUNIT(x)
137 
138 struct ccdbuf {
139 	struct buf	cb_buf;		/* new I/O buf */
140 	struct buf	*cb_obp;	/* ptr. to original I/O buf */
141 	struct ccd_softc *cb_sc;	/* pointer to ccd softc */
142 	int		cb_comp;	/* target component */
143 	SIMPLEQ_ENTRY(ccdbuf) cb_q;	/* fifo of component buffers */
144 };
145 
146 /* component buffer pool */
147 static pool_cache_t ccd_cache;
148 
149 #define	CCD_GETBUF()		pool_cache_get(ccd_cache, PR_WAITOK)
150 #define	CCD_PUTBUF(cbp)		pool_cache_put(ccd_cache, cbp)
151 
152 #define CCDLABELDEV(dev)	\
153 	(MAKEDISKDEV(major((dev)), ccdunit((dev)), RAW_PART))
154 
155 /* called by main() at boot time */
156 void	ccdattach(int);
157 
158 /* called by biodone() at interrupt time */
159 static void	ccdiodone(struct buf *);
160 
161 static void	ccdinterleave(struct ccd_softc *);
162 static int	ccdinit(struct ccd_softc *, char **, struct vnode **,
163 		    struct lwp *);
164 static struct ccdbuf *ccdbuffer(struct ccd_softc *, struct buf *,
165 		    daddr_t, void *, long);
166 static void	ccdgetdefaultlabel(struct ccd_softc *, struct disklabel *);
167 static void	ccdgetdisklabel(dev_t);
168 static void	ccdmakedisklabel(struct ccd_softc *);
169 static void	ccdstart(struct ccd_softc *);
170 static void	ccdthread(void *);
171 
172 static dev_type_open(ccdopen);
173 static dev_type_close(ccdclose);
174 static dev_type_read(ccdread);
175 static dev_type_write(ccdwrite);
176 static dev_type_ioctl(ccdioctl);
177 static dev_type_strategy(ccdstrategy);
178 static dev_type_size(ccdsize);
179 
180 const struct bdevsw ccd_bdevsw = {
181 	.d_open = ccdopen,
182 	.d_close = ccdclose,
183 	.d_strategy = ccdstrategy,
184 	.d_ioctl = ccdioctl,
185 	.d_dump = nodump,
186 	.d_psize = ccdsize,
187 	.d_flag = D_DISK | D_MPSAFE
188 };
189 
190 const struct cdevsw ccd_cdevsw = {
191 	.d_open = ccdopen,
192 	.d_close = ccdclose,
193 	.d_read = ccdread,
194 	.d_write = ccdwrite,
195 	.d_ioctl = ccdioctl,
196 	.d_stop = nostop,
197 	.d_tty = notty,
198 	.d_poll = nopoll,
199 	.d_mmap = nommap,
200 	.d_kqfilter = nokqfilter,
201 	.d_flag = D_DISK | D_MPSAFE
202 };
203 
204 #ifdef DEBUG
205 static	void printiinfo(struct ccdiinfo *);
206 #endif
207 
208 /* Publically visible for the benefit of libkvm and ccdconfig(8). */
209 struct ccd_softc 	*ccd_softc;
210 const int		ccd_softc_elemsize = sizeof(struct ccd_softc);
211 int			numccd = 0;
212 
213 /*
214  * Called by main() during pseudo-device attachment.  All we need
215  * to do is allocate enough space for devices to be configured later.
216  */
217 void
218 ccdattach(int num)
219 {
220 	struct ccd_softc *cs;
221 	int i;
222 
223 	if (num <= 0) {
224 #ifdef DIAGNOSTIC
225 		panic("ccdattach: count <= 0");
226 #endif
227 		return;
228 	}
229 
230 	ccd_softc = kmem_zalloc(num * ccd_softc_elemsize, KM_SLEEP);
231 	if (ccd_softc == NULL) {
232 		printf("WARNING: no memory for concatenated disks\n");
233 		return;
234 	}
235 	numccd = num;
236 
237 	/* Initialize the component buffer pool. */
238 	ccd_cache = pool_cache_init(sizeof(struct ccdbuf), 0,
239 	    0, 0, "ccdbuf", NULL, IPL_BIO, NULL, NULL, NULL);
240 
241 	/* Initialize per-softc structures. */
242 	for (i = 0; i < num; i++) {
243 		cs = &ccd_softc[i];
244 		snprintf(cs->sc_xname, sizeof(cs->sc_xname), "ccd%d", i);
245 		mutex_init(&cs->sc_dvlock, MUTEX_DEFAULT, IPL_NONE);
246 		cs->sc_iolock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
247 		cv_init(&cs->sc_stop, "ccdstop");
248 		cv_init(&cs->sc_push, "ccdthr");
249 		disk_init(&cs->sc_dkdev, cs->sc_xname, NULL); /* XXX */
250 	}
251 }
252 
253 static int
254 ccdinit(struct ccd_softc *cs, char **cpaths, struct vnode **vpp,
255     struct lwp *l)
256 {
257 	struct ccdcinfo *ci = NULL;
258 	size_t size;
259 	int ix;
260 	struct vattr va;
261 	size_t minsize;
262 	int maxsecsize;
263 	struct partinfo dpart;
264 	struct ccdgeom *ccg = &cs->sc_geom;
265 	char *tmppath;
266 	int error, path_alloced;
267 
268 #ifdef DEBUG
269 	if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
270 		printf("%s: ccdinit\n", cs->sc_xname);
271 #endif
272 
273 	/* Allocate space for the component info. */
274 	cs->sc_cinfo = kmem_alloc(cs->sc_nccdisks * sizeof(*cs->sc_cinfo),
275 	    KM_SLEEP);
276 	tmppath = kmem_alloc(MAXPATHLEN, KM_SLEEP);
277 
278 	cs->sc_size = 0;
279 
280 	/*
281 	 * Verify that each component piece exists and record
282 	 * relevant information about it.
283 	 */
284 	maxsecsize = 0;
285 	minsize = 0;
286 	for (ix = 0, path_alloced = 0; ix < cs->sc_nccdisks; ix++) {
287 		ci = &cs->sc_cinfo[ix];
288 		ci->ci_vp = vpp[ix];
289 
290 		/*
291 		 * Copy in the pathname of the component.
292 		 */
293 		memset(tmppath, 0, MAXPATHLEN);	/* sanity */
294 		error = copyinstr(cpaths[ix], tmppath,
295 		    MAXPATHLEN, &ci->ci_pathlen);
296 		if (ci->ci_pathlen == 0)
297 			error = EINVAL;
298 		if (error) {
299 #ifdef DEBUG
300 			if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
301 				printf("%s: can't copy path, error = %d\n",
302 				    cs->sc_xname, error);
303 #endif
304 			goto out;
305 		}
306 		ci->ci_path = kmem_alloc(ci->ci_pathlen, KM_SLEEP);
307 		memcpy(ci->ci_path, tmppath, ci->ci_pathlen);
308 		path_alloced++;
309 
310 		/*
311 		 * XXX: Cache the component's dev_t.
312 		 */
313 		vn_lock(vpp[ix], LK_SHARED | LK_RETRY);
314 		error = VOP_GETATTR(vpp[ix], &va, l->l_cred);
315 		VOP_UNLOCK(vpp[ix]);
316 		if (error != 0) {
317 #ifdef DEBUG
318 			if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
319 				printf("%s: %s: getattr failed %s = %d\n",
320 				    cs->sc_xname, ci->ci_path,
321 				    "error", error);
322 #endif
323 			goto out;
324 		}
325 		ci->ci_dev = va.va_rdev;
326 
327 		/*
328 		 * Get partition information for the component.
329 		 */
330 		error = VOP_IOCTL(vpp[ix], DIOCGPART, &dpart,
331 		    FREAD, l->l_cred);
332 		if (error) {
333 #ifdef DEBUG
334 			if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
335 				 printf("%s: %s: ioctl failed, error = %d\n",
336 				     cs->sc_xname, ci->ci_path, error);
337 #endif
338 			goto out;
339 		}
340 
341 /*
342  * This diagnostic test is disabled (for now?) since not all port supports
343  * on-disk BSD disklabel.
344  */
345 #if 0 /* def DIAGNOSTIC */
346 		/* Check fstype field of component. */
347 		if (dpart.part->p_fstype != FS_CCD)
348 			printf("%s: WARNING: %s: fstype %d != FS_CCD\n",
349 			    cs->sc_xname, ci->ci_path, dpart.part->p_fstype);
350 #endif
351 
352 		/*
353 		 * Calculate the size, truncating to an interleave
354 		 * boundary if necessary.
355 		 */
356 		maxsecsize =
357 		    ((dpart.disklab->d_secsize > maxsecsize) ?
358 		    dpart.disklab->d_secsize : maxsecsize);
359 		size = dpart.part->p_size;
360 		if (cs->sc_ileave > 1)
361 			size -= size % cs->sc_ileave;
362 
363 		if (size == 0) {
364 #ifdef DEBUG
365 			if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
366 				printf("%s: %s: size == 0\n",
367 				    cs->sc_xname, ci->ci_path);
368 #endif
369 			error = ENODEV;
370 			goto out;
371 		}
372 
373 		if (minsize == 0 || size < minsize)
374 			minsize = size;
375 		ci->ci_size = size;
376 		cs->sc_size += size;
377 	}
378 
379 	/*
380 	 * Don't allow the interleave to be smaller than
381 	 * the biggest component sector.
382 	 */
383 	if ((cs->sc_ileave > 0) &&
384 	    (cs->sc_ileave < (maxsecsize / DEV_BSIZE))) {
385 #ifdef DEBUG
386 		if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
387 			printf("%s: interleave must be at least %d\n",
388 			    cs->sc_xname, (maxsecsize / DEV_BSIZE));
389 #endif
390 		error = EINVAL;
391 		goto out;
392 	}
393 
394 	/*
395 	 * If uniform interleave is desired set all sizes to that of
396 	 * the smallest component.
397 	 */
398 	if (cs->sc_flags & CCDF_UNIFORM) {
399 		for (ci = cs->sc_cinfo;
400 		     ci < &cs->sc_cinfo[cs->sc_nccdisks]; ci++)
401 			ci->ci_size = minsize;
402 
403 		cs->sc_size = cs->sc_nccdisks * minsize;
404 	}
405 
406 	/*
407 	 * Construct the interleave table.
408 	 */
409 	ccdinterleave(cs);
410 
411 	/*
412 	 * Create pseudo-geometry based on 1MB cylinders.  It's
413 	 * pretty close.
414 	 */
415 	ccg->ccg_secsize = DEV_BSIZE;
416 	ccg->ccg_ntracks = 1;
417 	ccg->ccg_nsectors = 1024 * (1024 / ccg->ccg_secsize);
418 	ccg->ccg_ncylinders = cs->sc_size / ccg->ccg_nsectors;
419 
420 	/*
421 	 * Create thread to handle deferred I/O.
422 	 */
423 	cs->sc_zap = false;
424 	error = kthread_create(PRI_BIO, KTHREAD_MPSAFE, NULL, ccdthread,
425 	    cs, &cs->sc_thread, "%s", cs->sc_xname);
426 	if (error) {
427 		printf("ccdinit: can't create thread: %d\n", error);
428 		goto out;
429 	}
430 
431 	/*
432 	 * Only now that everything is set up can we enable the device.
433 	 */
434 	mutex_enter(cs->sc_iolock);
435 	cs->sc_flags |= CCDF_INITED;
436 	mutex_exit(cs->sc_iolock);
437 	kmem_free(tmppath, MAXPATHLEN);
438 	return (0);
439 
440  out:
441 	for (ix = 0; ix < path_alloced; ix++) {
442 		kmem_free(cs->sc_cinfo[ix].ci_path,
443 		    cs->sc_cinfo[ix].ci_pathlen);
444 	}
445 	kmem_free(cs->sc_cinfo, cs->sc_nccdisks * sizeof(struct ccdcinfo));
446 	kmem_free(tmppath, MAXPATHLEN);
447 	return (error);
448 }
449 
450 static void
451 ccdinterleave(struct ccd_softc *cs)
452 {
453 	struct ccdcinfo *ci, *smallci;
454 	struct ccdiinfo *ii;
455 	daddr_t bn, lbn;
456 	int ix;
457 	u_long size;
458 
459 #ifdef DEBUG
460 	if (ccddebug & CCDB_INIT)
461 		printf("ccdinterleave(%p): ileave %d\n", cs, cs->sc_ileave);
462 #endif
463 	/*
464 	 * Allocate an interleave table.
465 	 * Chances are this is too big, but we don't care.
466 	 */
467 	size = (cs->sc_nccdisks + 1) * sizeof(struct ccdiinfo);
468 	cs->sc_itable = kmem_zalloc(size, KM_SLEEP);
469 
470 	/*
471 	 * Trivial case: no interleave (actually interleave of disk size).
472 	 * Each table entry represents a single component in its entirety.
473 	 */
474 	if (cs->sc_ileave == 0) {
475 		bn = 0;
476 		ii = cs->sc_itable;
477 
478 		for (ix = 0; ix < cs->sc_nccdisks; ix++) {
479 			/* Allocate space for ii_index. */
480 			ii->ii_indexsz = sizeof(int);
481 			ii->ii_index = kmem_alloc(ii->ii_indexsz, KM_SLEEP);
482 			ii->ii_ndisk = 1;
483 			ii->ii_startblk = bn;
484 			ii->ii_startoff = 0;
485 			ii->ii_index[0] = ix;
486 			bn += cs->sc_cinfo[ix].ci_size;
487 			ii++;
488 		}
489 		ii->ii_ndisk = 0;
490 #ifdef DEBUG
491 		if (ccddebug & CCDB_INIT)
492 			printiinfo(cs->sc_itable);
493 #endif
494 		return;
495 	}
496 
497 	/*
498 	 * The following isn't fast or pretty; it doesn't have to be.
499 	 */
500 	size = 0;
501 	bn = lbn = 0;
502 	for (ii = cs->sc_itable; ; ii++) {
503 		/* Allocate space for ii_index. */
504 		ii->ii_indexsz = sizeof(int) * cs->sc_nccdisks;
505 		ii->ii_index = kmem_alloc(ii->ii_indexsz, KM_SLEEP);
506 
507 		/*
508 		 * Locate the smallest of the remaining components
509 		 */
510 		smallci = NULL;
511 		for (ci = cs->sc_cinfo;
512 		     ci < &cs->sc_cinfo[cs->sc_nccdisks]; ci++)
513 			if (ci->ci_size > size &&
514 			    (smallci == NULL ||
515 			     ci->ci_size < smallci->ci_size))
516 				smallci = ci;
517 
518 		/*
519 		 * Nobody left, all done
520 		 */
521 		if (smallci == NULL) {
522 			ii->ii_ndisk = 0;
523 			break;
524 		}
525 
526 		/*
527 		 * Record starting logical block and component offset
528 		 */
529 		ii->ii_startblk = bn / cs->sc_ileave;
530 		ii->ii_startoff = lbn;
531 
532 		/*
533 		 * Determine how many disks take part in this interleave
534 		 * and record their indices.
535 		 */
536 		ix = 0;
537 		for (ci = cs->sc_cinfo;
538 		     ci < &cs->sc_cinfo[cs->sc_nccdisks]; ci++)
539 			if (ci->ci_size >= smallci->ci_size)
540 				ii->ii_index[ix++] = ci - cs->sc_cinfo;
541 		ii->ii_ndisk = ix;
542 		bn += ix * (smallci->ci_size - size);
543 		lbn = smallci->ci_size / cs->sc_ileave;
544 		size = smallci->ci_size;
545 	}
546 #ifdef DEBUG
547 	if (ccddebug & CCDB_INIT)
548 		printiinfo(cs->sc_itable);
549 #endif
550 }
551 
552 /* ARGSUSED */
553 static int
554 ccdopen(dev_t dev, int flags, int fmt, struct lwp *l)
555 {
556 	int unit = ccdunit(dev);
557 	struct ccd_softc *cs;
558 	struct disklabel *lp;
559 	int error = 0, part, pmask;
560 
561 #ifdef DEBUG
562 	if (ccddebug & CCDB_FOLLOW)
563 		printf("ccdopen(0x%"PRIx64", 0x%x)\n", dev, flags);
564 #endif
565 	if (unit >= numccd)
566 		return (ENXIO);
567 	cs = &ccd_softc[unit];
568 
569 	mutex_enter(&cs->sc_dvlock);
570 
571 	lp = cs->sc_dkdev.dk_label;
572 
573 	part = DISKPART(dev);
574 	pmask = (1 << part);
575 
576 	/*
577 	 * If we're initialized, check to see if there are any other
578 	 * open partitions.  If not, then it's safe to update
579 	 * the in-core disklabel.  Only read the disklabel if it is
580 	 * not already valid.
581 	 */
582 	if ((cs->sc_flags & (CCDF_INITED|CCDF_VLABEL)) == CCDF_INITED &&
583 	    cs->sc_dkdev.dk_openmask == 0)
584 		ccdgetdisklabel(dev);
585 
586 	/* Check that the partition exists. */
587 	if (part != RAW_PART) {
588 		if (((cs->sc_flags & CCDF_INITED) == 0) ||
589 		    ((part >= lp->d_npartitions) ||
590 		     (lp->d_partitions[part].p_fstype == FS_UNUSED))) {
591 			error = ENXIO;
592 			goto done;
593 		}
594 	}
595 
596 	/* Prevent our unit from being unconfigured while open. */
597 	switch (fmt) {
598 	case S_IFCHR:
599 		cs->sc_dkdev.dk_copenmask |= pmask;
600 		break;
601 
602 	case S_IFBLK:
603 		cs->sc_dkdev.dk_bopenmask |= pmask;
604 		break;
605 	}
606 	cs->sc_dkdev.dk_openmask =
607 	    cs->sc_dkdev.dk_copenmask | cs->sc_dkdev.dk_bopenmask;
608 
609  done:
610 	mutex_exit(&cs->sc_dvlock);
611 	return (error);
612 }
613 
614 /* ARGSUSED */
615 static int
616 ccdclose(dev_t dev, int flags, int fmt, struct lwp *l)
617 {
618 	int unit = ccdunit(dev);
619 	struct ccd_softc *cs;
620 	int part;
621 
622 #ifdef DEBUG
623 	if (ccddebug & CCDB_FOLLOW)
624 		printf("ccdclose(0x%"PRIx64", 0x%x)\n", dev, flags);
625 #endif
626 
627 	if (unit >= numccd)
628 		return (ENXIO);
629 	cs = &ccd_softc[unit];
630 
631 	mutex_enter(&cs->sc_dvlock);
632 
633 	part = DISKPART(dev);
634 
635 	/* ...that much closer to allowing unconfiguration... */
636 	switch (fmt) {
637 	case S_IFCHR:
638 		cs->sc_dkdev.dk_copenmask &= ~(1 << part);
639 		break;
640 
641 	case S_IFBLK:
642 		cs->sc_dkdev.dk_bopenmask &= ~(1 << part);
643 		break;
644 	}
645 	cs->sc_dkdev.dk_openmask =
646 	    cs->sc_dkdev.dk_copenmask | cs->sc_dkdev.dk_bopenmask;
647 
648 	if (cs->sc_dkdev.dk_openmask == 0) {
649 		if ((cs->sc_flags & CCDF_KLABEL) == 0)
650 			cs->sc_flags &= ~CCDF_VLABEL;
651 	}
652 
653 	mutex_exit(&cs->sc_dvlock);
654 	return (0);
655 }
656 
657 static bool
658 ccdbackoff(struct ccd_softc *cs)
659 {
660 
661 	/* XXX Arbitrary, should be a uvm call. */
662 	return uvmexp.free < (uvmexp.freemin >> 1) &&
663 	    disk_isbusy(&cs->sc_dkdev);
664 }
665 
666 static void
667 ccdthread(void *cookie)
668 {
669 	struct ccd_softc *cs;
670 
671 	cs = cookie;
672 
673 #ifdef DEBUG
674  	if (ccddebug & CCDB_FOLLOW)
675  		printf("ccdthread: hello\n");
676 #endif
677 
678 	mutex_enter(cs->sc_iolock);
679 	while (__predict_true(!cs->sc_zap)) {
680 		if (bufq_peek(cs->sc_bufq) == NULL) {
681 			/* Nothing to do. */
682 			cv_wait(&cs->sc_push, cs->sc_iolock);
683 			continue;
684 		}
685 		if (ccdbackoff(cs)) {
686 			/* Wait for memory to become available. */
687 			(void)cv_timedwait(&cs->sc_push, cs->sc_iolock, 1);
688 			continue;
689 		}
690 #ifdef DEBUG
691  		if (ccddebug & CCDB_FOLLOW)
692  			printf("ccdthread: dispatching I/O\n");
693 #endif
694 		ccdstart(cs);
695 		mutex_enter(cs->sc_iolock);
696 	}
697 	cs->sc_thread = NULL;
698 	mutex_exit(cs->sc_iolock);
699 #ifdef DEBUG
700  	if (ccddebug & CCDB_FOLLOW)
701  		printf("ccdthread: goodbye\n");
702 #endif
703 	kthread_exit(0);
704 }
705 
706 static void
707 ccdstrategy(struct buf *bp)
708 {
709 	int unit = ccdunit(bp->b_dev);
710 	struct ccd_softc *cs = &ccd_softc[unit];
711 
712 	/* Must be open or reading label. */
713 	KASSERT(cs->sc_dkdev.dk_openmask != 0 ||
714 	    (cs->sc_flags & CCDF_RLABEL) != 0);
715 
716 	mutex_enter(cs->sc_iolock);
717 	/* Synchronize with device init/uninit. */
718 	if (__predict_false((cs->sc_flags & CCDF_INITED) == 0)) {
719 		mutex_exit(cs->sc_iolock);
720 #ifdef DEBUG
721  		if (ccddebug & CCDB_FOLLOW)
722  			printf("ccdstrategy: unit %d: not inited\n", unit);
723 #endif
724  		bp->b_error = ENXIO;
725  		bp->b_resid = bp->b_bcount;
726  		biodone(bp);
727 		return;
728 	}
729 
730 	/* Defer to thread if system is low on memory. */
731 	bufq_put(cs->sc_bufq, bp);
732 	if (__predict_false(ccdbackoff(cs))) {
733 		mutex_exit(cs->sc_iolock);
734 #ifdef DEBUG
735  		if (ccddebug & CCDB_FOLLOW)
736  			printf("ccdstrategy: holding off on I/O\n");
737 #endif
738 		return;
739 	}
740 	ccdstart(cs);
741 }
742 
743 static void
744 ccdstart(struct ccd_softc *cs)
745 {
746 	daddr_t blkno;
747 	int wlabel;
748 	struct disklabel *lp;
749 	long bcount, rcount;
750 	struct ccdbuf *cbp;
751 	char *addr;
752 	daddr_t bn;
753 	vnode_t *vp;
754 	buf_t *bp;
755 
756 	KASSERT(mutex_owned(cs->sc_iolock));
757 
758 	disk_busy(&cs->sc_dkdev);
759 	bp = bufq_get(cs->sc_bufq);
760 	KASSERT(bp != NULL);
761 
762 #ifdef DEBUG
763 	if (ccddebug & CCDB_FOLLOW)
764 		printf("ccdstart(%s, %p)\n", cs->sc_xname, bp);
765 #endif
766 
767 	/* If it's a nil transfer, wake up the top half now. */
768 	if (bp->b_bcount == 0)
769 		goto done;
770 
771 	lp = cs->sc_dkdev.dk_label;
772 
773 	/*
774 	 * Do bounds checking and adjust transfer.  If there's an
775 	 * error, the bounds check will flag that for us.  Convert
776 	 * the partition relative block number to an absolute.
777 	 */
778 	blkno = bp->b_blkno;
779 	wlabel = cs->sc_flags & (CCDF_WLABEL|CCDF_LABELLING);
780 	if (DISKPART(bp->b_dev) != RAW_PART) {
781 		if (bounds_check_with_label(&cs->sc_dkdev, bp, wlabel) <= 0)
782 			goto done;
783 		blkno += lp->d_partitions[DISKPART(bp->b_dev)].p_offset;
784 	}
785 	mutex_exit(cs->sc_iolock);
786 	bp->b_rawblkno = blkno;
787 
788 	/* Allocate the component buffers and start I/O! */
789 	bp->b_resid = bp->b_bcount;
790 	bn = bp->b_rawblkno;
791 	addr = bp->b_data;
792 	for (bcount = bp->b_bcount; bcount > 0; bcount -= rcount) {
793 		cbp = ccdbuffer(cs, bp, bn, addr, bcount);
794 		rcount = cbp->cb_buf.b_bcount;
795 		bn += btodb(rcount);
796 		addr += rcount;
797 		vp = cbp->cb_buf.b_vp;
798 		if ((cbp->cb_buf.b_flags & B_READ) == 0) {
799 			mutex_enter(vp->v_interlock);
800 			vp->v_numoutput++;
801 			mutex_exit(vp->v_interlock);
802 		}
803 		(void)VOP_STRATEGY(vp, &cbp->cb_buf);
804 	}
805 	return;
806 
807  done:
808 	disk_unbusy(&cs->sc_dkdev, 0, 0);
809 	cv_broadcast(&cs->sc_stop);
810 	cv_broadcast(&cs->sc_push);
811 	mutex_exit(cs->sc_iolock);
812 	bp->b_resid = bp->b_bcount;
813 	biodone(bp);
814 }
815 
816 /*
817  * Build a component buffer header.
818  */
819 static struct ccdbuf *
820 ccdbuffer(struct ccd_softc *cs, struct buf *bp, daddr_t bn, void *addr,
821     long bcount)
822 {
823 	struct ccdcinfo *ci;
824 	struct ccdbuf *cbp;
825 	daddr_t cbn, cboff;
826 	u_int64_t cbc;
827 	int ccdisk;
828 
829 #ifdef DEBUG
830 	if (ccddebug & CCDB_IO)
831 		printf("ccdbuffer(%p, %p, %" PRId64 ", %p, %ld)\n",
832 		       cs, bp, bn, addr, bcount);
833 #endif
834 	/*
835 	 * Determine which component bn falls in.
836 	 */
837 	cbn = bn;
838 	cboff = 0;
839 
840 	/*
841 	 * Serially concatenated
842 	 */
843 	if (cs->sc_ileave == 0) {
844 		daddr_t sblk;
845 
846 		sblk = 0;
847 		for (ccdisk = 0, ci = &cs->sc_cinfo[ccdisk];
848 		    cbn >= sblk + ci->ci_size;
849 		    ccdisk++, ci = &cs->sc_cinfo[ccdisk])
850 			sblk += ci->ci_size;
851 		cbn -= sblk;
852 	}
853 	/*
854 	 * Interleaved
855 	 */
856 	else {
857 		struct ccdiinfo *ii;
858 		int off;
859 
860 		cboff = cbn % cs->sc_ileave;
861 		cbn /= cs->sc_ileave;
862 		for (ii = cs->sc_itable; ii->ii_ndisk; ii++)
863 			if (ii->ii_startblk > cbn)
864 				break;
865 		ii--;
866 		off = cbn - ii->ii_startblk;
867 		if (ii->ii_ndisk == 1) {
868 			ccdisk = ii->ii_index[0];
869 			cbn = ii->ii_startoff + off;
870 		} else {
871 			ccdisk = ii->ii_index[off % ii->ii_ndisk];
872 			cbn = ii->ii_startoff + off / ii->ii_ndisk;
873 		}
874 		cbn *= cs->sc_ileave;
875 		ci = &cs->sc_cinfo[ccdisk];
876 	}
877 
878 	/*
879 	 * Fill in the component buf structure.
880 	 */
881 	cbp = CCD_GETBUF();
882 	KASSERT(cbp != NULL);
883 	buf_init(&cbp->cb_buf);
884 	cbp->cb_buf.b_flags = bp->b_flags;
885 	cbp->cb_buf.b_oflags = bp->b_oflags;
886 	cbp->cb_buf.b_cflags = bp->b_cflags;
887 	cbp->cb_buf.b_iodone = ccdiodone;
888 	cbp->cb_buf.b_proc = bp->b_proc;
889 	cbp->cb_buf.b_dev = ci->ci_dev;
890 	cbp->cb_buf.b_blkno = cbn + cboff;
891 	cbp->cb_buf.b_data = addr;
892 	cbp->cb_buf.b_vp = ci->ci_vp;
893 	cbp->cb_buf.b_objlock = ci->ci_vp->v_interlock;
894 	if (cs->sc_ileave == 0)
895 		cbc = dbtob((u_int64_t)(ci->ci_size - cbn));
896 	else
897 		cbc = dbtob((u_int64_t)(cs->sc_ileave - cboff));
898 	cbp->cb_buf.b_bcount = cbc < bcount ? cbc : bcount;
899 
900 	/*
901 	 * context for ccdiodone
902 	 */
903 	cbp->cb_obp = bp;
904 	cbp->cb_sc = cs;
905 	cbp->cb_comp = ccdisk;
906 
907 	BIO_COPYPRIO(&cbp->cb_buf, bp);
908 
909 #ifdef DEBUG
910 	if (ccddebug & CCDB_IO)
911 		printf(" dev 0x%"PRIx64"(u%lu): cbp %p bn %" PRId64 " addr %p"
912 		       " bcnt %d\n",
913 		    ci->ci_dev, (unsigned long) (ci-cs->sc_cinfo), cbp,
914 		    cbp->cb_buf.b_blkno, cbp->cb_buf.b_data,
915 		    cbp->cb_buf.b_bcount);
916 #endif
917 
918 	return (cbp);
919 }
920 
921 /*
922  * Called at interrupt time.
923  * Mark the component as done and if all components are done,
924  * take a ccd interrupt.
925  */
926 static void
927 ccdiodone(struct buf *vbp)
928 {
929 	struct ccdbuf *cbp = (struct ccdbuf *) vbp;
930 	struct buf *bp = cbp->cb_obp;
931 	struct ccd_softc *cs = cbp->cb_sc;
932 	int count;
933 
934 #ifdef DEBUG
935 	if (ccddebug & CCDB_FOLLOW)
936 		printf("ccdiodone(%p)\n", cbp);
937 	if (ccddebug & CCDB_IO) {
938 		printf("ccdiodone: bp %p bcount %d resid %d\n",
939 		       bp, bp->b_bcount, bp->b_resid);
940 		printf(" dev 0x%"PRIx64"(u%d), cbp %p bn %" PRId64 " addr %p"
941 		       " bcnt %d\n",
942 		       cbp->cb_buf.b_dev, cbp->cb_comp, cbp,
943 		       cbp->cb_buf.b_blkno, cbp->cb_buf.b_data,
944 		       cbp->cb_buf.b_bcount);
945 	}
946 #endif
947 
948 	if (cbp->cb_buf.b_error != 0) {
949 		bp->b_error = cbp->cb_buf.b_error;
950 		printf("%s: error %d on component %d\n",
951 		       cs->sc_xname, bp->b_error, cbp->cb_comp);
952 	}
953 	count = cbp->cb_buf.b_bcount;
954 	buf_destroy(&cbp->cb_buf);
955 	CCD_PUTBUF(cbp);
956 
957 	/*
958 	 * If all done, "interrupt".
959 	 */
960 	mutex_enter(cs->sc_iolock);
961 	bp->b_resid -= count;
962 	if (bp->b_resid < 0)
963 		panic("ccdiodone: count");
964 	if (bp->b_resid == 0) {
965 		/*
966 		 * Request is done for better or worse, wakeup the top half.
967 		 */
968 		if (bp->b_error != 0)
969 			bp->b_resid = bp->b_bcount;
970 		disk_unbusy(&cs->sc_dkdev, (bp->b_bcount - bp->b_resid),
971 		    (bp->b_flags & B_READ));
972 		if (!disk_isbusy(&cs->sc_dkdev)) {
973 			if (bufq_peek(cs->sc_bufq) != NULL) {
974 				cv_broadcast(&cs->sc_push);
975 			}
976 			cv_broadcast(&cs->sc_stop);
977 		}
978 		mutex_exit(cs->sc_iolock);
979 		biodone(bp);
980 	} else
981 		mutex_exit(cs->sc_iolock);
982 }
983 
984 /* ARGSUSED */
985 static int
986 ccdread(dev_t dev, struct uio *uio, int flags)
987 {
988 	int unit = ccdunit(dev);
989 	struct ccd_softc *cs;
990 
991 #ifdef DEBUG
992 	if (ccddebug & CCDB_FOLLOW)
993 		printf("ccdread(0x%"PRIx64", %p)\n", dev, uio);
994 #endif
995 	if (unit >= numccd)
996 		return (ENXIO);
997 	cs = &ccd_softc[unit];
998 
999 	/* Unlocked advisory check, ccdstrategy check is synchronous. */
1000 	if ((cs->sc_flags & CCDF_INITED) == 0)
1001 		return (ENXIO);
1002 
1003 	return (physio(ccdstrategy, NULL, dev, B_READ, minphys, uio));
1004 }
1005 
1006 /* ARGSUSED */
1007 static int
1008 ccdwrite(dev_t dev, struct uio *uio, int flags)
1009 {
1010 	int unit = ccdunit(dev);
1011 	struct ccd_softc *cs;
1012 
1013 #ifdef DEBUG
1014 	if (ccddebug & CCDB_FOLLOW)
1015 		printf("ccdwrite(0x%"PRIx64", %p)\n", dev, uio);
1016 #endif
1017 	if (unit >= numccd)
1018 		return (ENXIO);
1019 	cs = &ccd_softc[unit];
1020 
1021 	/* Unlocked advisory check, ccdstrategy check is synchronous. */
1022 	if ((cs->sc_flags & CCDF_INITED) == 0)
1023 		return (ENXIO);
1024 
1025 	return (physio(ccdstrategy, NULL, dev, B_WRITE, minphys, uio));
1026 }
1027 
1028 static int
1029 ccdioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l)
1030 {
1031 	int unit = ccdunit(dev);
1032 	int i, j, lookedup = 0, error = 0;
1033 	int part, pmask;
1034 	struct ccd_softc *cs;
1035 	struct ccd_ioctl *ccio = (struct ccd_ioctl *)data;
1036 	kauth_cred_t uc;
1037 	char **cpp;
1038 	struct pathbuf *pb;
1039 	struct vnode **vpp;
1040 #ifdef __HAVE_OLD_DISKLABEL
1041 	struct disklabel newlabel;
1042 #endif
1043 
1044 	if (unit >= numccd)
1045 		return (ENXIO);
1046 	cs = &ccd_softc[unit];
1047 	uc = kauth_cred_get();
1048 
1049 	/* Must be open for writes for these commands... */
1050 	switch (cmd) {
1051 	case CCDIOCSET:
1052 	case CCDIOCCLR:
1053 	case DIOCSDINFO:
1054 	case DIOCWDINFO:
1055 #ifdef __HAVE_OLD_DISKLABEL
1056 	case ODIOCSDINFO:
1057 	case ODIOCWDINFO:
1058 #endif
1059 	case DIOCKLABEL:
1060 	case DIOCWLABEL:
1061 		if ((flag & FWRITE) == 0)
1062 			return (EBADF);
1063 	}
1064 
1065 	mutex_enter(&cs->sc_dvlock);
1066 
1067 	/* Must be initialized for these... */
1068 	switch (cmd) {
1069 	case CCDIOCCLR:
1070 	case DIOCGDINFO:
1071 	case DIOCCACHESYNC:
1072 	case DIOCSDINFO:
1073 	case DIOCWDINFO:
1074 	case DIOCGPART:
1075 	case DIOCWLABEL:
1076 	case DIOCKLABEL:
1077 	case DIOCGDEFLABEL:
1078 #ifdef __HAVE_OLD_DISKLABEL
1079 	case ODIOCGDINFO:
1080 	case ODIOCSDINFO:
1081 	case ODIOCWDINFO:
1082 	case ODIOCGDEFLABEL:
1083 #endif
1084 		if ((cs->sc_flags & CCDF_INITED) == 0) {
1085 			error = ENXIO;
1086 			goto out;
1087 		}
1088 	}
1089 
1090 	switch (cmd) {
1091 	case CCDIOCSET:
1092 		if (cs->sc_flags & CCDF_INITED) {
1093 			error = EBUSY;
1094 			goto out;
1095 		}
1096 
1097 		/* Validate the flags. */
1098 		if ((ccio->ccio_flags & CCDF_USERMASK) != ccio->ccio_flags) {
1099 			error = EINVAL;
1100 			goto out;
1101 		}
1102 
1103 		if (ccio->ccio_ndisks > CCD_MAXNDISKS ||
1104 		    ccio->ccio_ndisks == 0) {
1105 			error = EINVAL;
1106 			goto out;
1107 		}
1108 
1109 		/* Fill in some important bits. */
1110 		cs->sc_ileave = ccio->ccio_ileave;
1111 		cs->sc_nccdisks = ccio->ccio_ndisks;
1112 		cs->sc_flags = ccio->ccio_flags & CCDF_USERMASK;
1113 
1114 		/*
1115 		 * Allocate space for and copy in the array of
1116 		 * componet pathnames and device numbers.
1117 		 */
1118 		cpp = kmem_alloc(ccio->ccio_ndisks * sizeof(*cpp), KM_SLEEP);
1119 		vpp = kmem_alloc(ccio->ccio_ndisks * sizeof(*vpp), KM_SLEEP);
1120 		error = copyin(ccio->ccio_disks, cpp,
1121 		    ccio->ccio_ndisks * sizeof(*cpp));
1122 		if (error) {
1123 			kmem_free(vpp, ccio->ccio_ndisks * sizeof(*vpp));
1124 			kmem_free(cpp, ccio->ccio_ndisks * sizeof(*cpp));
1125 			goto out;
1126 		}
1127 
1128 #ifdef DEBUG
1129 		if (ccddebug & CCDB_INIT)
1130 			for (i = 0; i < ccio->ccio_ndisks; ++i)
1131 				printf("ccdioctl: component %d: %p\n",
1132 				    i, cpp[i]);
1133 #endif
1134 
1135 		for (i = 0; i < ccio->ccio_ndisks; ++i) {
1136 #ifdef DEBUG
1137 			if (ccddebug & CCDB_INIT)
1138 				printf("ccdioctl: lookedup = %d\n", lookedup);
1139 #endif
1140 			error = pathbuf_copyin(cpp[i], &pb);
1141 			if (error == 0) {
1142 				error = dk_lookup(pb, l, &vpp[i]);
1143 			}
1144 			pathbuf_destroy(pb);
1145 			if (error != 0) {
1146 				for (j = 0; j < lookedup; ++j)
1147 					(void)vn_close(vpp[j], FREAD|FWRITE,
1148 					    uc);
1149 				kmem_free(vpp, ccio->ccio_ndisks *
1150 				    sizeof(*vpp));
1151 				kmem_free(cpp, ccio->ccio_ndisks *
1152 				    sizeof(*cpp));
1153 				goto out;
1154 			}
1155 			++lookedup;
1156 		}
1157 
1158 		/* Attach the disk. */
1159 		disk_attach(&cs->sc_dkdev);
1160 		bufq_alloc(&cs->sc_bufq, "fcfs", 0);
1161 
1162 		/*
1163 		 * Initialize the ccd.  Fills in the softc for us.
1164 		 */
1165 		if ((error = ccdinit(cs, cpp, vpp, l)) != 0) {
1166 			for (j = 0; j < lookedup; ++j)
1167 				(void)vn_close(vpp[j], FREAD|FWRITE,
1168 				    uc);
1169 			kmem_free(vpp, ccio->ccio_ndisks * sizeof(*vpp));
1170 			kmem_free(cpp, ccio->ccio_ndisks * sizeof(*cpp));
1171 			disk_detach(&cs->sc_dkdev);
1172 			bufq_free(cs->sc_bufq);
1173 			goto out;
1174 		}
1175 
1176 		/* We can free the temporary variables now. */
1177 		kmem_free(vpp, ccio->ccio_ndisks * sizeof(*vpp));
1178 		kmem_free(cpp, ccio->ccio_ndisks * sizeof(*cpp));
1179 
1180 		/*
1181 		 * The ccd has been successfully initialized, so
1182 		 * we can place it into the array.  Don't try to
1183 		 * read the disklabel until the disk has been attached,
1184 		 * because space for the disklabel is allocated
1185 		 * in disk_attach();
1186 		 */
1187 		ccio->ccio_unit = unit;
1188 		ccio->ccio_size = cs->sc_size;
1189 
1190 		/* Try and read the disklabel. */
1191 		ccdgetdisklabel(dev);
1192 		break;
1193 
1194 	case CCDIOCCLR:
1195 		/*
1196 		 * Don't unconfigure if any other partitions are open
1197 		 * or if both the character and block flavors of this
1198 		 * partition are open.
1199 		 */
1200 		part = DISKPART(dev);
1201 		pmask = (1 << part);
1202 		if ((cs->sc_dkdev.dk_openmask & ~pmask) ||
1203 		    ((cs->sc_dkdev.dk_bopenmask & pmask) &&
1204 		    (cs->sc_dkdev.dk_copenmask & pmask))) {
1205 			error = EBUSY;
1206 			goto out;
1207 		}
1208 
1209 		/* Stop new I/O, wait for in-flight I/O to complete. */
1210 		mutex_enter(cs->sc_iolock);
1211 		cs->sc_flags &= ~(CCDF_INITED|CCDF_VLABEL);
1212 		cs->sc_zap = true;
1213 		while (disk_isbusy(&cs->sc_dkdev) ||
1214 		    bufq_peek(cs->sc_bufq) != NULL ||
1215 		    cs->sc_thread != NULL) {
1216 			cv_broadcast(&cs->sc_push);
1217 			(void)cv_timedwait(&cs->sc_stop, cs->sc_iolock, hz);
1218 		}
1219 		mutex_exit(cs->sc_iolock);
1220 
1221 		/*
1222 		 * Free ccd_softc information and clear entry.
1223 		 */
1224 
1225 		/* Close the components and free their pathnames. */
1226 		for (i = 0; i < cs->sc_nccdisks; ++i) {
1227 			/*
1228 			 * XXX: this close could potentially fail and
1229 			 * cause Bad Things.  Maybe we need to force
1230 			 * the close to happen?
1231 			 */
1232 #ifdef DEBUG
1233 			if (ccddebug & CCDB_VNODE)
1234 				vprint("CCDIOCCLR: vnode info",
1235 				    cs->sc_cinfo[i].ci_vp);
1236 #endif
1237 			(void)vn_close(cs->sc_cinfo[i].ci_vp, FREAD|FWRITE,
1238 			    uc);
1239 			kmem_free(cs->sc_cinfo[i].ci_path,
1240 			    cs->sc_cinfo[i].ci_pathlen);
1241 		}
1242 
1243 		/* Free interleave index. */
1244 		for (i = 0; cs->sc_itable[i].ii_ndisk; ++i) {
1245 			kmem_free(cs->sc_itable[i].ii_index,
1246 			    cs->sc_itable[i].ii_indexsz);
1247 		}
1248 
1249 		/* Free component info and interleave table. */
1250 		kmem_free(cs->sc_cinfo, cs->sc_nccdisks *
1251 		    sizeof(struct ccdcinfo));
1252 		kmem_free(cs->sc_itable, (cs->sc_nccdisks + 1) *
1253 		    sizeof(struct ccdiinfo));
1254 
1255 		/* Detatch the disk. */
1256 		disk_detach(&cs->sc_dkdev);
1257 		bufq_free(cs->sc_bufq);
1258 		break;
1259 
1260 	case DIOCGDINFO:
1261 		*(struct disklabel *)data = *(cs->sc_dkdev.dk_label);
1262 		break;
1263 
1264 #ifdef __HAVE_OLD_DISKLABEL
1265 	case ODIOCGDINFO:
1266 		newlabel = *(cs->sc_dkdev.dk_label);
1267 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1268 			return ENOTTY;
1269 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1270 		break;
1271 #endif
1272 
1273 	case DIOCGPART:
1274 		((struct partinfo *)data)->disklab = cs->sc_dkdev.dk_label;
1275 		((struct partinfo *)data)->part =
1276 		    &cs->sc_dkdev.dk_label->d_partitions[DISKPART(dev)];
1277 		break;
1278 
1279 	case DIOCCACHESYNC:
1280 		/*
1281 		 * XXX Do we really need to care about having a writable
1282 		 * file descriptor here?
1283 		 */
1284 		if ((flag & FWRITE) == 0)
1285 			return (EBADF);
1286 
1287 		/*
1288 		 * We pass this call down to all components and report
1289 		 * the first error we encounter.
1290 		 */
1291 		for (error = 0, i = 0; i < cs->sc_nccdisks; i++) {
1292 			j = VOP_IOCTL(cs->sc_cinfo[i].ci_vp, cmd, data,
1293 				      flag, uc);
1294 			if (j != 0 && error == 0)
1295 				error = j;
1296 		}
1297 		break;
1298 
1299 	case DIOCWDINFO:
1300 	case DIOCSDINFO:
1301 #ifdef __HAVE_OLD_DISKLABEL
1302 	case ODIOCWDINFO:
1303 	case ODIOCSDINFO:
1304 #endif
1305 	{
1306 		struct disklabel *lp;
1307 #ifdef __HAVE_OLD_DISKLABEL
1308 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
1309 			memset(&newlabel, 0, sizeof newlabel);
1310 			memcpy(&newlabel, data, sizeof (struct olddisklabel));
1311 			lp = &newlabel;
1312 		} else
1313 #endif
1314 		lp = (struct disklabel *)data;
1315 
1316 		cs->sc_flags |= CCDF_LABELLING;
1317 
1318 		error = setdisklabel(cs->sc_dkdev.dk_label,
1319 		    lp, 0, cs->sc_dkdev.dk_cpulabel);
1320 		if (error == 0) {
1321 			if (cmd == DIOCWDINFO
1322 #ifdef __HAVE_OLD_DISKLABEL
1323 			    || cmd == ODIOCWDINFO
1324 #endif
1325 			   )
1326 				error = writedisklabel(CCDLABELDEV(dev),
1327 				    ccdstrategy, cs->sc_dkdev.dk_label,
1328 				    cs->sc_dkdev.dk_cpulabel);
1329 		}
1330 
1331 		cs->sc_flags &= ~CCDF_LABELLING;
1332 		break;
1333 	}
1334 
1335 	case DIOCKLABEL:
1336 		if (*(int *)data != 0)
1337 			cs->sc_flags |= CCDF_KLABEL;
1338 		else
1339 			cs->sc_flags &= ~CCDF_KLABEL;
1340 		break;
1341 
1342 	case DIOCWLABEL:
1343 		if (*(int *)data != 0)
1344 			cs->sc_flags |= CCDF_WLABEL;
1345 		else
1346 			cs->sc_flags &= ~CCDF_WLABEL;
1347 		break;
1348 
1349 	case DIOCGDEFLABEL:
1350 		ccdgetdefaultlabel(cs, (struct disklabel *)data);
1351 		break;
1352 
1353 #ifdef __HAVE_OLD_DISKLABEL
1354 	case ODIOCGDEFLABEL:
1355 		ccdgetdefaultlabel(cs, &newlabel);
1356 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1357 			return ENOTTY;
1358 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1359 		break;
1360 #endif
1361 
1362 	default:
1363 		error = ENOTTY;
1364 	}
1365 
1366  out:
1367 	mutex_exit(&cs->sc_dvlock);
1368 	return (error);
1369 }
1370 
1371 static int
1372 ccdsize(dev_t dev)
1373 {
1374 	struct ccd_softc *cs;
1375 	struct disklabel *lp;
1376 	int part, unit, omask, size;
1377 
1378 	unit = ccdunit(dev);
1379 	if (unit >= numccd)
1380 		return (-1);
1381 	cs = &ccd_softc[unit];
1382 
1383 	if ((cs->sc_flags & CCDF_INITED) == 0)
1384 		return (-1);
1385 
1386 	part = DISKPART(dev);
1387 	omask = cs->sc_dkdev.dk_openmask & (1 << part);
1388 	lp = cs->sc_dkdev.dk_label;
1389 
1390 	if (omask == 0 && ccdopen(dev, 0, S_IFBLK, curlwp))
1391 		return (-1);
1392 
1393 	if (lp->d_partitions[part].p_fstype != FS_SWAP)
1394 		size = -1;
1395 	else
1396 		size = lp->d_partitions[part].p_size *
1397 		    (lp->d_secsize / DEV_BSIZE);
1398 
1399 	if (omask == 0 && ccdclose(dev, 0, S_IFBLK, curlwp))
1400 		return (-1);
1401 
1402 	return (size);
1403 }
1404 
1405 static void
1406 ccdgetdefaultlabel(struct ccd_softc *cs, struct disklabel *lp)
1407 {
1408 	struct ccdgeom *ccg = &cs->sc_geom;
1409 
1410 	memset(lp, 0, sizeof(*lp));
1411 
1412 	lp->d_secperunit = cs->sc_size;
1413 	lp->d_secsize = ccg->ccg_secsize;
1414 	lp->d_nsectors = ccg->ccg_nsectors;
1415 	lp->d_ntracks = ccg->ccg_ntracks;
1416 	lp->d_ncylinders = ccg->ccg_ncylinders;
1417 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1418 
1419 	strncpy(lp->d_typename, "ccd", sizeof(lp->d_typename));
1420 	lp->d_type = DTYPE_CCD;
1421 	strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
1422 	lp->d_rpm = 3600;
1423 	lp->d_interleave = 1;
1424 	lp->d_flags = 0;
1425 
1426 	lp->d_partitions[RAW_PART].p_offset = 0;
1427 	lp->d_partitions[RAW_PART].p_size = cs->sc_size;
1428 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
1429 	lp->d_npartitions = RAW_PART + 1;
1430 
1431 	lp->d_magic = DISKMAGIC;
1432 	lp->d_magic2 = DISKMAGIC;
1433 	lp->d_checksum = dkcksum(cs->sc_dkdev.dk_label);
1434 }
1435 
1436 /*
1437  * Read the disklabel from the ccd.  If one is not present, fake one
1438  * up.
1439  */
1440 static void
1441 ccdgetdisklabel(dev_t dev)
1442 {
1443 	int unit = ccdunit(dev);
1444 	struct ccd_softc *cs = &ccd_softc[unit];
1445 	const char *errstring;
1446 	struct disklabel *lp = cs->sc_dkdev.dk_label;
1447 	struct cpu_disklabel *clp = cs->sc_dkdev.dk_cpulabel;
1448 
1449 	KASSERT(mutex_owned(&cs->sc_dvlock));
1450 
1451 	memset(clp, 0, sizeof(*clp));
1452 
1453 	ccdgetdefaultlabel(cs, lp);
1454 
1455 	/*
1456 	 * Call the generic disklabel extraction routine.
1457 	 */
1458 	cs->sc_flags |= CCDF_RLABEL;
1459 	if ((cs->sc_flags & CCDF_NOLABEL) != 0)
1460 		errstring = "CCDF_NOLABEL set; ignoring on-disk label";
1461 	else
1462 		errstring = readdisklabel(CCDLABELDEV(dev), ccdstrategy,
1463 		    cs->sc_dkdev.dk_label, cs->sc_dkdev.dk_cpulabel);
1464 	if (errstring)
1465 		ccdmakedisklabel(cs);
1466 	else {
1467 		int i;
1468 		struct partition *pp;
1469 
1470 		/*
1471 		 * Sanity check whether the found disklabel is valid.
1472 		 *
1473 		 * This is necessary since total size of ccd may vary
1474 		 * when an interleave is changed even though exactly
1475 		 * same componets are used, and old disklabel may used
1476 		 * if that is found.
1477 		 */
1478 		if (lp->d_secperunit != cs->sc_size)
1479 			printf("WARNING: %s: "
1480 			    "total sector size in disklabel (%d) != "
1481 			    "the size of ccd (%lu)\n", cs->sc_xname,
1482 			    lp->d_secperunit, (u_long)cs->sc_size);
1483 		for (i = 0; i < lp->d_npartitions; i++) {
1484 			pp = &lp->d_partitions[i];
1485 			if (pp->p_offset + pp->p_size > cs->sc_size)
1486 				printf("WARNING: %s: end of partition `%c' "
1487 				    "exceeds the size of ccd (%lu)\n",
1488 				    cs->sc_xname, 'a' + i, (u_long)cs->sc_size);
1489 		}
1490 	}
1491 
1492 #ifdef DEBUG
1493 	/* It's actually extremely common to have unlabeled ccds. */
1494 	if (ccddebug & CCDB_LABEL)
1495 		if (errstring != NULL)
1496 			printf("%s: %s\n", cs->sc_xname, errstring);
1497 #endif
1498 
1499 	/* In-core label now valid. */
1500 	cs->sc_flags = (cs->sc_flags | CCDF_VLABEL) & ~CCDF_RLABEL;
1501 }
1502 
1503 /*
1504  * Take care of things one might want to take care of in the event
1505  * that a disklabel isn't present.
1506  */
1507 static void
1508 ccdmakedisklabel(struct ccd_softc *cs)
1509 {
1510 	struct disklabel *lp = cs->sc_dkdev.dk_label;
1511 
1512 	/*
1513 	 * For historical reasons, if there's no disklabel present
1514 	 * the raw partition must be marked FS_BSDFFS.
1515 	 */
1516 	lp->d_partitions[RAW_PART].p_fstype = FS_BSDFFS;
1517 
1518 	strncpy(lp->d_packname, "default label", sizeof(lp->d_packname));
1519 
1520 	lp->d_checksum = dkcksum(lp);
1521 }
1522 
1523 #ifdef DEBUG
1524 static void
1525 printiinfo(struct ccdiinfo *ii)
1526 {
1527 	int ix, i;
1528 
1529 	for (ix = 0; ii->ii_ndisk; ix++, ii++) {
1530 		printf(" itab[%d]: #dk %d sblk %" PRId64 " soff %" PRId64,
1531 		    ix, ii->ii_ndisk, ii->ii_startblk, ii->ii_startoff);
1532 		for (i = 0; i < ii->ii_ndisk; i++)
1533 			printf(" %d", ii->ii_index[i]);
1534 		printf("\n");
1535 	}
1536 }
1537 #endif
1538 
1539 MODULE(MODULE_CLASS_DRIVER, ccd, NULL);
1540 
1541 static int
1542 ccd_modcmd(modcmd_t cmd, void *arg)
1543 {
1544 	int bmajor, cmajor, error = 0;
1545 
1546 	bmajor = cmajor = -1;
1547 
1548 	switch (cmd) {
1549 	case MODULE_CMD_INIT:
1550 #ifdef _MODULE
1551 		ccdattach(4);
1552 
1553 		return devsw_attach("ccd", &ccd_bdevsw, &bmajor,
1554 		    &ccd_cdevsw, &cmajor);
1555 #endif
1556 		break;
1557 
1558 	case MODULE_CMD_FINI:
1559 #ifdef _MODULE
1560 		return devsw_detach(&ccd_bdevsw, &ccd_cdevsw);
1561 #endif
1562 		break;
1563 
1564 	case MODULE_CMD_STAT:
1565 		return ENOTTY;
1566 
1567 	default:
1568 		return ENOTTY;
1569 	}
1570 
1571 	return error;
1572 }
1573