xref: /netbsd-src/sys/dev/ccd.c (revision b7b7574d3bf8eeb51a1fa3977b59142ec6434a55)
1 /*	$NetBSD: ccd.c,v 1.149 2014/06/14 07:39:00 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 1.149 2014/06/14 07:39:00 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 #include <sys/sysctl.h>
118 
119 #include <uvm/uvm_extern.h>
120 
121 #include <dev/ccdvar.h>
122 #include <dev/dkvar.h>
123 
124 #include <miscfs/specfs/specdev.h> /* for v_rdev */
125 
126 #if defined(CCDDEBUG) && !defined(DEBUG)
127 #define DEBUG
128 #endif
129 
130 #ifdef DEBUG
131 #define CCDB_FOLLOW	0x01
132 #define CCDB_INIT	0x02
133 #define CCDB_IO		0x04
134 #define CCDB_LABEL	0x08
135 #define CCDB_VNODE	0x10
136 int ccddebug = 0x00;
137 #endif
138 
139 #define	ccdunit(x)	DISKUNIT(x)
140 
141 struct ccdbuf {
142 	struct buf	cb_buf;		/* new I/O buf */
143 	struct buf	*cb_obp;	/* ptr. to original I/O buf */
144 	struct ccd_softc *cb_sc;	/* pointer to ccd softc */
145 	int		cb_comp;	/* target component */
146 	SIMPLEQ_ENTRY(ccdbuf) cb_q;	/* fifo of component buffers */
147 };
148 
149 /* component buffer pool */
150 static pool_cache_t ccd_cache;
151 
152 #define	CCD_GETBUF()		pool_cache_get(ccd_cache, PR_WAITOK)
153 #define	CCD_PUTBUF(cbp)		pool_cache_put(ccd_cache, cbp)
154 
155 #define CCDLABELDEV(dev)	\
156 	(MAKEDISKDEV(major((dev)), ccdunit((dev)), RAW_PART))
157 
158 /* called by main() at boot time */
159 void	ccdattach(int);
160 
161 /* called by biodone() at interrupt time */
162 static void	ccdiodone(struct buf *);
163 
164 static void	ccdinterleave(struct ccd_softc *);
165 static int	ccdinit(struct ccd_softc *, char **, struct vnode **,
166 		    struct lwp *);
167 static struct ccdbuf *ccdbuffer(struct ccd_softc *, struct buf *,
168 		    daddr_t, void *, long);
169 static void	ccdgetdefaultlabel(struct ccd_softc *, struct disklabel *);
170 static void	ccdgetdisklabel(dev_t);
171 static void	ccdmakedisklabel(struct ccd_softc *);
172 static void	ccdstart(struct ccd_softc *);
173 static void	ccdthread(void *);
174 
175 static dev_type_open(ccdopen);
176 static dev_type_close(ccdclose);
177 static dev_type_read(ccdread);
178 static dev_type_write(ccdwrite);
179 static dev_type_ioctl(ccdioctl);
180 static dev_type_strategy(ccdstrategy);
181 static dev_type_size(ccdsize);
182 
183 const struct bdevsw ccd_bdevsw = {
184 	.d_open = ccdopen,
185 	.d_close = ccdclose,
186 	.d_strategy = ccdstrategy,
187 	.d_ioctl = ccdioctl,
188 	.d_dump = nodump,
189 	.d_psize = ccdsize,
190 	.d_flag = D_DISK | D_MPSAFE
191 };
192 
193 const struct cdevsw ccd_cdevsw = {
194 	.d_open = ccdopen,
195 	.d_close = ccdclose,
196 	.d_read = ccdread,
197 	.d_write = ccdwrite,
198 	.d_ioctl = ccdioctl,
199 	.d_stop = nostop,
200 	.d_tty = notty,
201 	.d_poll = nopoll,
202 	.d_mmap = nommap,
203 	.d_kqfilter = nokqfilter,
204 	.d_flag = D_DISK | D_MPSAFE
205 };
206 
207 #ifdef DEBUG
208 static	void printiinfo(struct ccdiinfo *);
209 #endif
210 
211 static LIST_HEAD(, ccd_softc) ccds = LIST_HEAD_INITIALIZER(ccds);
212 static kmutex_t ccd_lock;
213 
214 static struct ccd_softc *
215 ccdcreate(int unit) {
216 	struct ccd_softc *sc = kmem_zalloc(sizeof(*sc), KM_SLEEP);
217 	if (sc == NULL) {
218 #ifdef DIAGNOSTIC
219 		printf("%s: out of memory\n", __func__);
220 #endif
221 		return NULL;
222 	}
223 	/* Initialize per-softc structures. */
224 	snprintf(sc->sc_xname, sizeof(sc->sc_xname), "ccd%d", unit);
225 	mutex_init(&sc->sc_dvlock, MUTEX_DEFAULT, IPL_NONE);
226 	sc->sc_iolock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
227 	cv_init(&sc->sc_stop, "ccdstop");
228 	cv_init(&sc->sc_push, "ccdthr");
229 	disk_init(&sc->sc_dkdev, sc->sc_xname, NULL); /* XXX */
230 	return sc;
231 }
232 
233 static void
234 ccddestroy(struct ccd_softc *sc) {
235 	mutex_obj_free(sc->sc_iolock);
236 	mutex_exit(&sc->sc_dvlock);
237 	mutex_destroy(&sc->sc_dvlock);
238 	cv_destroy(&sc->sc_stop);
239 	cv_destroy(&sc->sc_push);
240 	disk_destroy(&sc->sc_dkdev);
241 	kmem_free(sc, sizeof(*sc));
242 }
243 
244 static struct ccd_softc *
245 ccdget(int unit) {
246 	struct ccd_softc *sc;
247 	if (unit < 0) {
248 #ifdef DIAGNOSTIC
249 		panic("%s: unit %d!", __func__, unit);
250 #endif
251 		return NULL;
252 	}
253 	mutex_enter(&ccd_lock);
254 	LIST_FOREACH(sc, &ccds, sc_link) {
255 		if (sc->sc_unit == unit) {
256 			mutex_exit(&ccd_lock);
257 			return sc;
258 		}
259 	}
260 	mutex_exit(&ccd_lock);
261 	if ((sc = ccdcreate(unit)) == NULL)
262 		return NULL;
263 	mutex_enter(&ccd_lock);
264 	LIST_INSERT_HEAD(&ccds, sc, sc_link);
265 	mutex_exit(&ccd_lock);
266 	return sc;
267 }
268 
269 static void
270 ccdput(struct ccd_softc *sc) {
271 	mutex_enter(&ccd_lock);
272 	LIST_REMOVE(sc, sc_link);
273 	mutex_exit(&ccd_lock);
274 	ccddestroy(sc);
275 }
276 
277 /*
278  * Called by main() during pseudo-device attachment.  All we need
279  * to do is allocate enough space for devices to be configured later.
280  */
281 void
282 ccdattach(int num)
283 {
284 	mutex_init(&ccd_lock, MUTEX_DEFAULT, IPL_NONE);
285 
286 	/* Initialize the component buffer pool. */
287 	ccd_cache = pool_cache_init(sizeof(struct ccdbuf), 0,
288 	    0, 0, "ccdbuf", NULL, IPL_BIO, NULL, NULL, NULL);
289 }
290 
291 static int
292 ccdinit(struct ccd_softc *cs, char **cpaths, struct vnode **vpp,
293     struct lwp *l)
294 {
295 	struct ccdcinfo *ci = NULL;
296 	int ix;
297 	struct ccdgeom *ccg = &cs->sc_geom;
298 	char *tmppath;
299 	int error, path_alloced;
300 	uint64_t psize, minsize;
301 	unsigned secsize, maxsecsize;
302 
303 #ifdef DEBUG
304 	if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
305 		printf("%s: ccdinit\n", cs->sc_xname);
306 #endif
307 
308 	/* Allocate space for the component info. */
309 	cs->sc_cinfo = kmem_alloc(cs->sc_nccdisks * sizeof(*cs->sc_cinfo),
310 	    KM_SLEEP);
311 	tmppath = kmem_alloc(MAXPATHLEN, KM_SLEEP);
312 
313 	cs->sc_size = 0;
314 
315 	/*
316 	 * Verify that each component piece exists and record
317 	 * relevant information about it.
318 	 */
319 	maxsecsize = 0;
320 	minsize = 0;
321 	for (ix = 0, path_alloced = 0; ix < cs->sc_nccdisks; ix++) {
322 		ci = &cs->sc_cinfo[ix];
323 		ci->ci_vp = vpp[ix];
324 
325 		/*
326 		 * Copy in the pathname of the component.
327 		 */
328 		memset(tmppath, 0, MAXPATHLEN);	/* sanity */
329 		error = copyinstr(cpaths[ix], tmppath,
330 		    MAXPATHLEN, &ci->ci_pathlen);
331 		if (ci->ci_pathlen == 0)
332 			error = EINVAL;
333 		if (error) {
334 #ifdef DEBUG
335 			if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
336 				printf("%s: can't copy path, error = %d\n",
337 				    cs->sc_xname, error);
338 #endif
339 			goto out;
340 		}
341 		ci->ci_path = kmem_alloc(ci->ci_pathlen, KM_SLEEP);
342 		memcpy(ci->ci_path, tmppath, ci->ci_pathlen);
343 		path_alloced++;
344 
345 		/*
346 		 * XXX: Cache the component's dev_t.
347 		 */
348 		ci->ci_dev = vpp[ix]->v_rdev;
349 
350 		/*
351 		 * Get partition information for the component.
352 		 */
353 		error = getdisksize(vpp[ix], &psize, &secsize);
354 		if (error) {
355 #ifdef DEBUG
356 			if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
357 				 printf("%s: %s: disksize failed, error = %d\n",
358 				     cs->sc_xname, ci->ci_path, error);
359 #endif
360 			goto out;
361 		}
362 
363 		/*
364 		 * Calculate the size, truncating to an interleave
365 		 * boundary if necessary.
366 		 */
367 		maxsecsize = secsize > maxsecsize ? secsize : maxsecsize;
368 		if (cs->sc_ileave > 1)
369 			psize -= psize % cs->sc_ileave;
370 
371 		if (psize == 0) {
372 #ifdef DEBUG
373 			if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
374 				printf("%s: %s: size == 0\n",
375 				    cs->sc_xname, ci->ci_path);
376 #endif
377 			error = ENODEV;
378 			goto out;
379 		}
380 
381 		if (minsize == 0 || psize < minsize)
382 			minsize = psize;
383 		ci->ci_size = psize;
384 		cs->sc_size += psize;
385 	}
386 
387 	/*
388 	 * Don't allow the interleave to be smaller than
389 	 * the biggest component sector.
390 	 */
391 	if ((cs->sc_ileave > 0) &&
392 	    (cs->sc_ileave < (maxsecsize / DEV_BSIZE))) {
393 #ifdef DEBUG
394 		if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
395 			printf("%s: interleave must be at least %d\n",
396 			    cs->sc_xname, (maxsecsize / DEV_BSIZE));
397 #endif
398 		error = EINVAL;
399 		goto out;
400 	}
401 
402 	/*
403 	 * If uniform interleave is desired set all sizes to that of
404 	 * the smallest component.
405 	 */
406 	if (cs->sc_flags & CCDF_UNIFORM) {
407 		for (ci = cs->sc_cinfo;
408 		     ci < &cs->sc_cinfo[cs->sc_nccdisks]; ci++)
409 			ci->ci_size = minsize;
410 
411 		cs->sc_size = cs->sc_nccdisks * minsize;
412 	}
413 
414 	/*
415 	 * Construct the interleave table.
416 	 */
417 	ccdinterleave(cs);
418 
419 	/*
420 	 * Create pseudo-geometry based on 1MB cylinders.  It's
421 	 * pretty close.
422 	 */
423 	ccg->ccg_secsize = DEV_BSIZE;
424 	ccg->ccg_ntracks = 1;
425 	ccg->ccg_nsectors = 1024 * (1024 / ccg->ccg_secsize);
426 	ccg->ccg_ncylinders = cs->sc_size / ccg->ccg_nsectors;
427 
428 	/*
429 	 * Create thread to handle deferred I/O.
430 	 */
431 	cs->sc_zap = false;
432 	error = kthread_create(PRI_BIO, KTHREAD_MPSAFE, NULL, ccdthread,
433 	    cs, &cs->sc_thread, "%s", cs->sc_xname);
434 	if (error) {
435 		printf("ccdinit: can't create thread: %d\n", error);
436 		goto out;
437 	}
438 
439 	/*
440 	 * Only now that everything is set up can we enable the device.
441 	 */
442 	mutex_enter(cs->sc_iolock);
443 	cs->sc_flags |= CCDF_INITED;
444 	mutex_exit(cs->sc_iolock);
445 	kmem_free(tmppath, MAXPATHLEN);
446 	return (0);
447 
448  out:
449 	for (ix = 0; ix < path_alloced; ix++) {
450 		kmem_free(cs->sc_cinfo[ix].ci_path,
451 		    cs->sc_cinfo[ix].ci_pathlen);
452 	}
453 	kmem_free(cs->sc_cinfo, cs->sc_nccdisks * sizeof(struct ccdcinfo));
454 	kmem_free(tmppath, MAXPATHLEN);
455 	return (error);
456 }
457 
458 static void
459 ccdinterleave(struct ccd_softc *cs)
460 {
461 	struct ccdcinfo *ci, *smallci;
462 	struct ccdiinfo *ii;
463 	daddr_t bn, lbn;
464 	int ix;
465 	u_long size;
466 
467 #ifdef DEBUG
468 	if (ccddebug & CCDB_INIT)
469 		printf("ccdinterleave(%p): ileave %d\n", cs, cs->sc_ileave);
470 #endif
471 	/*
472 	 * Allocate an interleave table.
473 	 * Chances are this is too big, but we don't care.
474 	 */
475 	size = (cs->sc_nccdisks + 1) * sizeof(struct ccdiinfo);
476 	cs->sc_itable = kmem_zalloc(size, KM_SLEEP);
477 
478 	/*
479 	 * Trivial case: no interleave (actually interleave of disk size).
480 	 * Each table entry represents a single component in its entirety.
481 	 */
482 	if (cs->sc_ileave == 0) {
483 		bn = 0;
484 		ii = cs->sc_itable;
485 
486 		for (ix = 0; ix < cs->sc_nccdisks; ix++) {
487 			/* Allocate space for ii_index. */
488 			ii->ii_indexsz = sizeof(int);
489 			ii->ii_index = kmem_alloc(ii->ii_indexsz, KM_SLEEP);
490 			ii->ii_ndisk = 1;
491 			ii->ii_startblk = bn;
492 			ii->ii_startoff = 0;
493 			ii->ii_index[0] = ix;
494 			bn += cs->sc_cinfo[ix].ci_size;
495 			ii++;
496 		}
497 		ii->ii_ndisk = 0;
498 #ifdef DEBUG
499 		if (ccddebug & CCDB_INIT)
500 			printiinfo(cs->sc_itable);
501 #endif
502 		return;
503 	}
504 
505 	/*
506 	 * The following isn't fast or pretty; it doesn't have to be.
507 	 */
508 	size = 0;
509 	bn = lbn = 0;
510 	for (ii = cs->sc_itable; ; ii++) {
511 		/* Allocate space for ii_index. */
512 		ii->ii_indexsz = sizeof(int) * cs->sc_nccdisks;
513 		ii->ii_index = kmem_alloc(ii->ii_indexsz, KM_SLEEP);
514 
515 		/*
516 		 * Locate the smallest of the remaining components
517 		 */
518 		smallci = NULL;
519 		for (ci = cs->sc_cinfo;
520 		     ci < &cs->sc_cinfo[cs->sc_nccdisks]; ci++)
521 			if (ci->ci_size > size &&
522 			    (smallci == NULL ||
523 			     ci->ci_size < smallci->ci_size))
524 				smallci = ci;
525 
526 		/*
527 		 * Nobody left, all done
528 		 */
529 		if (smallci == NULL) {
530 			ii->ii_ndisk = 0;
531 			break;
532 		}
533 
534 		/*
535 		 * Record starting logical block and component offset
536 		 */
537 		ii->ii_startblk = bn / cs->sc_ileave;
538 		ii->ii_startoff = lbn;
539 
540 		/*
541 		 * Determine how many disks take part in this interleave
542 		 * and record their indices.
543 		 */
544 		ix = 0;
545 		for (ci = cs->sc_cinfo;
546 		     ci < &cs->sc_cinfo[cs->sc_nccdisks]; ci++)
547 			if (ci->ci_size >= smallci->ci_size)
548 				ii->ii_index[ix++] = ci - cs->sc_cinfo;
549 		ii->ii_ndisk = ix;
550 		bn += ix * (smallci->ci_size - size);
551 		lbn = smallci->ci_size / cs->sc_ileave;
552 		size = smallci->ci_size;
553 	}
554 #ifdef DEBUG
555 	if (ccddebug & CCDB_INIT)
556 		printiinfo(cs->sc_itable);
557 #endif
558 }
559 
560 /* ARGSUSED */
561 static int
562 ccdopen(dev_t dev, int flags, int fmt, struct lwp *l)
563 {
564 	int unit = ccdunit(dev);
565 	struct ccd_softc *cs;
566 	struct disklabel *lp;
567 	int error = 0, part, pmask;
568 
569 #ifdef DEBUG
570 	if (ccddebug & CCDB_FOLLOW)
571 		printf("ccdopen(0x%"PRIx64", 0x%x)\n", dev, flags);
572 #endif
573 	if ((cs = ccdget(unit)) == NULL)
574 		return ENXIO;
575 
576 	mutex_enter(&cs->sc_dvlock);
577 
578 	lp = cs->sc_dkdev.dk_label;
579 
580 	part = DISKPART(dev);
581 	pmask = (1 << part);
582 
583 	/*
584 	 * If we're initialized, check to see if there are any other
585 	 * open partitions.  If not, then it's safe to update
586 	 * the in-core disklabel.  Only read the disklabel if it is
587 	 * not already valid.
588 	 */
589 	if ((cs->sc_flags & (CCDF_INITED|CCDF_VLABEL)) == CCDF_INITED &&
590 	    cs->sc_dkdev.dk_openmask == 0)
591 		ccdgetdisklabel(dev);
592 
593 	/* Check that the partition exists. */
594 	if (part != RAW_PART) {
595 		if (((cs->sc_flags & CCDF_INITED) == 0) ||
596 		    ((part >= lp->d_npartitions) ||
597 		     (lp->d_partitions[part].p_fstype == FS_UNUSED))) {
598 			error = ENXIO;
599 			goto done;
600 		}
601 	}
602 
603 	/* Prevent our unit from being unconfigured while open. */
604 	switch (fmt) {
605 	case S_IFCHR:
606 		cs->sc_dkdev.dk_copenmask |= pmask;
607 		break;
608 
609 	case S_IFBLK:
610 		cs->sc_dkdev.dk_bopenmask |= pmask;
611 		break;
612 	}
613 	cs->sc_dkdev.dk_openmask =
614 	    cs->sc_dkdev.dk_copenmask | cs->sc_dkdev.dk_bopenmask;
615 
616  done:
617 	mutex_exit(&cs->sc_dvlock);
618 	return (error);
619 }
620 
621 /* ARGSUSED */
622 static int
623 ccdclose(dev_t dev, int flags, int fmt, struct lwp *l)
624 {
625 	int unit = ccdunit(dev);
626 	struct ccd_softc *cs;
627 	int part;
628 
629 #ifdef DEBUG
630 	if (ccddebug & CCDB_FOLLOW)
631 		printf("ccdclose(0x%"PRIx64", 0x%x)\n", dev, flags);
632 #endif
633 
634 	if ((cs = ccdget(unit)) == NULL)
635 		return ENXIO;
636 
637 	mutex_enter(&cs->sc_dvlock);
638 
639 	part = DISKPART(dev);
640 
641 	/* ...that much closer to allowing unconfiguration... */
642 	switch (fmt) {
643 	case S_IFCHR:
644 		cs->sc_dkdev.dk_copenmask &= ~(1 << part);
645 		break;
646 
647 	case S_IFBLK:
648 		cs->sc_dkdev.dk_bopenmask &= ~(1 << part);
649 		break;
650 	}
651 	cs->sc_dkdev.dk_openmask =
652 	    cs->sc_dkdev.dk_copenmask | cs->sc_dkdev.dk_bopenmask;
653 
654 	if (cs->sc_dkdev.dk_openmask == 0) {
655 		if ((cs->sc_flags & CCDF_KLABEL) == 0)
656 			cs->sc_flags &= ~CCDF_VLABEL;
657 	}
658 
659 	mutex_exit(&cs->sc_dvlock);
660 	return (0);
661 }
662 
663 static bool
664 ccdbackoff(struct ccd_softc *cs)
665 {
666 
667 	/* XXX Arbitrary, should be a uvm call. */
668 	return uvmexp.free < (uvmexp.freemin >> 1) &&
669 	    disk_isbusy(&cs->sc_dkdev);
670 }
671 
672 static void
673 ccdthread(void *cookie)
674 {
675 	struct ccd_softc *cs;
676 
677 	cs = cookie;
678 
679 #ifdef DEBUG
680  	if (ccddebug & CCDB_FOLLOW)
681  		printf("ccdthread: hello\n");
682 #endif
683 
684 	mutex_enter(cs->sc_iolock);
685 	while (__predict_true(!cs->sc_zap)) {
686 		if (bufq_peek(cs->sc_bufq) == NULL) {
687 			/* Nothing to do. */
688 			cv_wait(&cs->sc_push, cs->sc_iolock);
689 			continue;
690 		}
691 		if (ccdbackoff(cs)) {
692 			/* Wait for memory to become available. */
693 			(void)cv_timedwait(&cs->sc_push, cs->sc_iolock, 1);
694 			continue;
695 		}
696 #ifdef DEBUG
697  		if (ccddebug & CCDB_FOLLOW)
698  			printf("ccdthread: dispatching I/O\n");
699 #endif
700 		ccdstart(cs);
701 		mutex_enter(cs->sc_iolock);
702 	}
703 	cs->sc_thread = NULL;
704 	mutex_exit(cs->sc_iolock);
705 #ifdef DEBUG
706  	if (ccddebug & CCDB_FOLLOW)
707  		printf("ccdthread: goodbye\n");
708 #endif
709 	kthread_exit(0);
710 }
711 
712 static void
713 ccdstrategy(struct buf *bp)
714 {
715 	int unit = ccdunit(bp->b_dev);
716 	struct ccd_softc *cs;
717 	if ((cs = ccdget(unit)) == NULL)
718 		return;
719 
720 	/* Must be open or reading label. */
721 	KASSERT(cs->sc_dkdev.dk_openmask != 0 ||
722 	    (cs->sc_flags & CCDF_RLABEL) != 0);
723 
724 	mutex_enter(cs->sc_iolock);
725 	/* Synchronize with device init/uninit. */
726 	if (__predict_false((cs->sc_flags & CCDF_INITED) == 0)) {
727 		mutex_exit(cs->sc_iolock);
728 #ifdef DEBUG
729  		if (ccddebug & CCDB_FOLLOW)
730  			printf("ccdstrategy: unit %d: not inited\n", unit);
731 #endif
732  		bp->b_error = ENXIO;
733  		bp->b_resid = bp->b_bcount;
734  		biodone(bp);
735 		return;
736 	}
737 
738 	/* Defer to thread if system is low on memory. */
739 	bufq_put(cs->sc_bufq, bp);
740 	if (__predict_false(ccdbackoff(cs))) {
741 		mutex_exit(cs->sc_iolock);
742 #ifdef DEBUG
743  		if (ccddebug & CCDB_FOLLOW)
744  			printf("ccdstrategy: holding off on I/O\n");
745 #endif
746 		return;
747 	}
748 	ccdstart(cs);
749 }
750 
751 static void
752 ccdstart(struct ccd_softc *cs)
753 {
754 	daddr_t blkno;
755 	int wlabel;
756 	struct disklabel *lp;
757 	long bcount, rcount;
758 	struct ccdbuf *cbp;
759 	char *addr;
760 	daddr_t bn;
761 	vnode_t *vp;
762 	buf_t *bp;
763 
764 	KASSERT(mutex_owned(cs->sc_iolock));
765 
766 	disk_busy(&cs->sc_dkdev);
767 	bp = bufq_get(cs->sc_bufq);
768 	KASSERT(bp != NULL);
769 
770 #ifdef DEBUG
771 	if (ccddebug & CCDB_FOLLOW)
772 		printf("ccdstart(%s, %p)\n", cs->sc_xname, bp);
773 #endif
774 
775 	/* If it's a nil transfer, wake up the top half now. */
776 	if (bp->b_bcount == 0)
777 		goto done;
778 
779 	lp = cs->sc_dkdev.dk_label;
780 
781 	/*
782 	 * Do bounds checking and adjust transfer.  If there's an
783 	 * error, the bounds check will flag that for us.  Convert
784 	 * the partition relative block number to an absolute.
785 	 */
786 	blkno = bp->b_blkno;
787 	wlabel = cs->sc_flags & (CCDF_WLABEL|CCDF_LABELLING);
788 	if (DISKPART(bp->b_dev) != RAW_PART) {
789 		if (bounds_check_with_label(&cs->sc_dkdev, bp, wlabel) <= 0)
790 			goto done;
791 		blkno += lp->d_partitions[DISKPART(bp->b_dev)].p_offset;
792 	}
793 	mutex_exit(cs->sc_iolock);
794 	bp->b_rawblkno = blkno;
795 
796 	/* Allocate the component buffers and start I/O! */
797 	bp->b_resid = bp->b_bcount;
798 	bn = bp->b_rawblkno;
799 	addr = bp->b_data;
800 	for (bcount = bp->b_bcount; bcount > 0; bcount -= rcount) {
801 		cbp = ccdbuffer(cs, bp, bn, addr, bcount);
802 		rcount = cbp->cb_buf.b_bcount;
803 		bn += btodb(rcount);
804 		addr += rcount;
805 		vp = cbp->cb_buf.b_vp;
806 		if ((cbp->cb_buf.b_flags & B_READ) == 0) {
807 			mutex_enter(vp->v_interlock);
808 			vp->v_numoutput++;
809 			mutex_exit(vp->v_interlock);
810 		}
811 		(void)VOP_STRATEGY(vp, &cbp->cb_buf);
812 	}
813 	return;
814 
815  done:
816 	disk_unbusy(&cs->sc_dkdev, 0, 0);
817 	cv_broadcast(&cs->sc_stop);
818 	cv_broadcast(&cs->sc_push);
819 	mutex_exit(cs->sc_iolock);
820 	bp->b_resid = bp->b_bcount;
821 	biodone(bp);
822 }
823 
824 /*
825  * Build a component buffer header.
826  */
827 static struct ccdbuf *
828 ccdbuffer(struct ccd_softc *cs, struct buf *bp, daddr_t bn, void *addr,
829     long bcount)
830 {
831 	struct ccdcinfo *ci;
832 	struct ccdbuf *cbp;
833 	daddr_t cbn, cboff;
834 	u_int64_t cbc;
835 	int ccdisk;
836 
837 #ifdef DEBUG
838 	if (ccddebug & CCDB_IO)
839 		printf("ccdbuffer(%p, %p, %" PRId64 ", %p, %ld)\n",
840 		       cs, bp, bn, addr, bcount);
841 #endif
842 	/*
843 	 * Determine which component bn falls in.
844 	 */
845 	cbn = bn;
846 	cboff = 0;
847 
848 	/*
849 	 * Serially concatenated
850 	 */
851 	if (cs->sc_ileave == 0) {
852 		daddr_t sblk;
853 
854 		sblk = 0;
855 		for (ccdisk = 0, ci = &cs->sc_cinfo[ccdisk];
856 		    cbn >= sblk + ci->ci_size;
857 		    ccdisk++, ci = &cs->sc_cinfo[ccdisk])
858 			sblk += ci->ci_size;
859 		cbn -= sblk;
860 	}
861 	/*
862 	 * Interleaved
863 	 */
864 	else {
865 		struct ccdiinfo *ii;
866 		int off;
867 
868 		cboff = cbn % cs->sc_ileave;
869 		cbn /= cs->sc_ileave;
870 		for (ii = cs->sc_itable; ii->ii_ndisk; ii++)
871 			if (ii->ii_startblk > cbn)
872 				break;
873 		ii--;
874 		off = cbn - ii->ii_startblk;
875 		if (ii->ii_ndisk == 1) {
876 			ccdisk = ii->ii_index[0];
877 			cbn = ii->ii_startoff + off;
878 		} else {
879 			ccdisk = ii->ii_index[off % ii->ii_ndisk];
880 			cbn = ii->ii_startoff + off / ii->ii_ndisk;
881 		}
882 		cbn *= cs->sc_ileave;
883 		ci = &cs->sc_cinfo[ccdisk];
884 	}
885 
886 	/*
887 	 * Fill in the component buf structure.
888 	 */
889 	cbp = CCD_GETBUF();
890 	KASSERT(cbp != NULL);
891 	buf_init(&cbp->cb_buf);
892 	cbp->cb_buf.b_flags = bp->b_flags;
893 	cbp->cb_buf.b_oflags = bp->b_oflags;
894 	cbp->cb_buf.b_cflags = bp->b_cflags;
895 	cbp->cb_buf.b_iodone = ccdiodone;
896 	cbp->cb_buf.b_proc = bp->b_proc;
897 	cbp->cb_buf.b_dev = ci->ci_dev;
898 	cbp->cb_buf.b_blkno = cbn + cboff;
899 	cbp->cb_buf.b_data = addr;
900 	cbp->cb_buf.b_vp = ci->ci_vp;
901 	cbp->cb_buf.b_objlock = ci->ci_vp->v_interlock;
902 	if (cs->sc_ileave == 0)
903 		cbc = dbtob((u_int64_t)(ci->ci_size - cbn));
904 	else
905 		cbc = dbtob((u_int64_t)(cs->sc_ileave - cboff));
906 	cbp->cb_buf.b_bcount = cbc < bcount ? cbc : bcount;
907 
908 	/*
909 	 * context for ccdiodone
910 	 */
911 	cbp->cb_obp = bp;
912 	cbp->cb_sc = cs;
913 	cbp->cb_comp = ccdisk;
914 
915 	BIO_COPYPRIO(&cbp->cb_buf, bp);
916 
917 #ifdef DEBUG
918 	if (ccddebug & CCDB_IO)
919 		printf(" dev 0x%"PRIx64"(u%lu): cbp %p bn %" PRId64 " addr %p"
920 		       " bcnt %d\n",
921 		    ci->ci_dev, (unsigned long) (ci-cs->sc_cinfo), cbp,
922 		    cbp->cb_buf.b_blkno, cbp->cb_buf.b_data,
923 		    cbp->cb_buf.b_bcount);
924 #endif
925 
926 	return (cbp);
927 }
928 
929 /*
930  * Called at interrupt time.
931  * Mark the component as done and if all components are done,
932  * take a ccd interrupt.
933  */
934 static void
935 ccdiodone(struct buf *vbp)
936 {
937 	struct ccdbuf *cbp = (struct ccdbuf *) vbp;
938 	struct buf *bp = cbp->cb_obp;
939 	struct ccd_softc *cs = cbp->cb_sc;
940 	int count;
941 
942 #ifdef DEBUG
943 	if (ccddebug & CCDB_FOLLOW)
944 		printf("ccdiodone(%p)\n", cbp);
945 	if (ccddebug & CCDB_IO) {
946 		printf("ccdiodone: bp %p bcount %d resid %d\n",
947 		       bp, bp->b_bcount, bp->b_resid);
948 		printf(" dev 0x%"PRIx64"(u%d), cbp %p bn %" PRId64 " addr %p"
949 		       " bcnt %d\n",
950 		       cbp->cb_buf.b_dev, cbp->cb_comp, cbp,
951 		       cbp->cb_buf.b_blkno, cbp->cb_buf.b_data,
952 		       cbp->cb_buf.b_bcount);
953 	}
954 #endif
955 
956 	if (cbp->cb_buf.b_error != 0) {
957 		bp->b_error = cbp->cb_buf.b_error;
958 		printf("%s: error %d on component %d\n",
959 		       cs->sc_xname, bp->b_error, cbp->cb_comp);
960 	}
961 	count = cbp->cb_buf.b_bcount;
962 	buf_destroy(&cbp->cb_buf);
963 	CCD_PUTBUF(cbp);
964 
965 	/*
966 	 * If all done, "interrupt".
967 	 */
968 	mutex_enter(cs->sc_iolock);
969 	bp->b_resid -= count;
970 	if (bp->b_resid < 0)
971 		panic("ccdiodone: count");
972 	if (bp->b_resid == 0) {
973 		/*
974 		 * Request is done for better or worse, wakeup the top half.
975 		 */
976 		if (bp->b_error != 0)
977 			bp->b_resid = bp->b_bcount;
978 		disk_unbusy(&cs->sc_dkdev, (bp->b_bcount - bp->b_resid),
979 		    (bp->b_flags & B_READ));
980 		if (!disk_isbusy(&cs->sc_dkdev)) {
981 			if (bufq_peek(cs->sc_bufq) != NULL) {
982 				cv_broadcast(&cs->sc_push);
983 			}
984 			cv_broadcast(&cs->sc_stop);
985 		}
986 		mutex_exit(cs->sc_iolock);
987 		biodone(bp);
988 	} else
989 		mutex_exit(cs->sc_iolock);
990 }
991 
992 /* ARGSUSED */
993 static int
994 ccdread(dev_t dev, struct uio *uio, int flags)
995 {
996 	int unit = ccdunit(dev);
997 	struct ccd_softc *cs;
998 
999 #ifdef DEBUG
1000 	if (ccddebug & CCDB_FOLLOW)
1001 		printf("ccdread(0x%"PRIx64", %p)\n", dev, uio);
1002 #endif
1003 	if ((cs = ccdget(unit)) == NULL)
1004 		return 0;
1005 
1006 	/* Unlocked advisory check, ccdstrategy check is synchronous. */
1007 	if ((cs->sc_flags & CCDF_INITED) == 0)
1008 		return (ENXIO);
1009 
1010 	return (physio(ccdstrategy, NULL, dev, B_READ, minphys, uio));
1011 }
1012 
1013 /* ARGSUSED */
1014 static int
1015 ccdwrite(dev_t dev, struct uio *uio, int flags)
1016 {
1017 	int unit = ccdunit(dev);
1018 	struct ccd_softc *cs;
1019 
1020 #ifdef DEBUG
1021 	if (ccddebug & CCDB_FOLLOW)
1022 		printf("ccdwrite(0x%"PRIx64", %p)\n", dev, uio);
1023 #endif
1024 	if ((cs = ccdget(unit)) == NULL)
1025 		return ENOENT;
1026 
1027 	/* Unlocked advisory check, ccdstrategy check is synchronous. */
1028 	if ((cs->sc_flags & CCDF_INITED) == 0)
1029 		return (ENXIO);
1030 
1031 	return (physio(ccdstrategy, NULL, dev, B_WRITE, minphys, uio));
1032 }
1033 
1034 static int
1035 ccdioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l)
1036 {
1037 	int unit = ccdunit(dev);
1038 	int i, j, lookedup = 0, error = 0;
1039 	int part, pmask;
1040 	struct ccd_softc *cs;
1041 	struct ccd_ioctl *ccio = (struct ccd_ioctl *)data;
1042 	kauth_cred_t uc;
1043 	char **cpp;
1044 	struct pathbuf *pb;
1045 	struct vnode **vpp;
1046 #ifdef __HAVE_OLD_DISKLABEL
1047 	struct disklabel newlabel;
1048 #endif
1049 
1050 	if ((cs = ccdget(unit)) == NULL)
1051 		return ENOENT;
1052 	uc = kauth_cred_get();
1053 
1054 	/* Must be open for writes for these commands... */
1055 	switch (cmd) {
1056 	case CCDIOCSET:
1057 	case CCDIOCCLR:
1058 	case DIOCSDINFO:
1059 	case DIOCWDINFO:
1060 #ifdef __HAVE_OLD_DISKLABEL
1061 	case ODIOCSDINFO:
1062 	case ODIOCWDINFO:
1063 #endif
1064 	case DIOCKLABEL:
1065 	case DIOCWLABEL:
1066 		if ((flag & FWRITE) == 0)
1067 			return (EBADF);
1068 	}
1069 
1070 	mutex_enter(&cs->sc_dvlock);
1071 
1072 	/* Must be initialized for these... */
1073 	switch (cmd) {
1074 	case CCDIOCCLR:
1075 	case DIOCGDINFO:
1076 	case DIOCCACHESYNC:
1077 	case DIOCSDINFO:
1078 	case DIOCWDINFO:
1079 	case DIOCGPART:
1080 	case DIOCWLABEL:
1081 	case DIOCKLABEL:
1082 	case DIOCGDEFLABEL:
1083 #ifdef __HAVE_OLD_DISKLABEL
1084 	case ODIOCGDINFO:
1085 	case ODIOCSDINFO:
1086 	case ODIOCWDINFO:
1087 	case ODIOCGDEFLABEL:
1088 #endif
1089 		if ((cs->sc_flags & CCDF_INITED) == 0) {
1090 			error = ENXIO;
1091 			goto out;
1092 		}
1093 	}
1094 
1095 	switch (cmd) {
1096 	case CCDIOCSET:
1097 		if (cs->sc_flags & CCDF_INITED) {
1098 			error = EBUSY;
1099 			goto out;
1100 		}
1101 
1102 		/* Validate the flags. */
1103 		if ((ccio->ccio_flags & CCDF_USERMASK) != ccio->ccio_flags) {
1104 			error = EINVAL;
1105 			goto out;
1106 		}
1107 
1108 		if (ccio->ccio_ndisks > CCD_MAXNDISKS ||
1109 		    ccio->ccio_ndisks == 0) {
1110 			error = EINVAL;
1111 			goto out;
1112 		}
1113 
1114 		/* Fill in some important bits. */
1115 		cs->sc_ileave = ccio->ccio_ileave;
1116 		cs->sc_nccdisks = ccio->ccio_ndisks;
1117 		cs->sc_flags = ccio->ccio_flags & CCDF_USERMASK;
1118 
1119 		/*
1120 		 * Allocate space for and copy in the array of
1121 		 * componet pathnames and device numbers.
1122 		 */
1123 		cpp = kmem_alloc(ccio->ccio_ndisks * sizeof(*cpp), KM_SLEEP);
1124 		vpp = kmem_alloc(ccio->ccio_ndisks * sizeof(*vpp), KM_SLEEP);
1125 		error = copyin(ccio->ccio_disks, cpp,
1126 		    ccio->ccio_ndisks * sizeof(*cpp));
1127 		if (error) {
1128 			kmem_free(vpp, ccio->ccio_ndisks * sizeof(*vpp));
1129 			kmem_free(cpp, ccio->ccio_ndisks * sizeof(*cpp));
1130 			goto out;
1131 		}
1132 
1133 #ifdef DEBUG
1134 		if (ccddebug & CCDB_INIT)
1135 			for (i = 0; i < ccio->ccio_ndisks; ++i)
1136 				printf("ccdioctl: component %d: %p\n",
1137 				    i, cpp[i]);
1138 #endif
1139 
1140 		for (i = 0; i < ccio->ccio_ndisks; ++i) {
1141 #ifdef DEBUG
1142 			if (ccddebug & CCDB_INIT)
1143 				printf("ccdioctl: lookedup = %d\n", lookedup);
1144 #endif
1145 			error = pathbuf_copyin(cpp[i], &pb);
1146 			if (error == 0) {
1147 				error = dk_lookup(pb, l, &vpp[i]);
1148 			}
1149 			pathbuf_destroy(pb);
1150 			if (error != 0) {
1151 				for (j = 0; j < lookedup; ++j)
1152 					(void)vn_close(vpp[j], FREAD|FWRITE,
1153 					    uc);
1154 				kmem_free(vpp, ccio->ccio_ndisks *
1155 				    sizeof(*vpp));
1156 				kmem_free(cpp, ccio->ccio_ndisks *
1157 				    sizeof(*cpp));
1158 				goto out;
1159 			}
1160 			++lookedup;
1161 		}
1162 
1163 		/* Attach the disk. */
1164 		disk_attach(&cs->sc_dkdev);
1165 		bufq_alloc(&cs->sc_bufq, "fcfs", 0);
1166 
1167 		/*
1168 		 * Initialize the ccd.  Fills in the softc for us.
1169 		 */
1170 		if ((error = ccdinit(cs, cpp, vpp, l)) != 0) {
1171 			for (j = 0; j < lookedup; ++j)
1172 				(void)vn_close(vpp[j], FREAD|FWRITE,
1173 				    uc);
1174 			kmem_free(vpp, ccio->ccio_ndisks * sizeof(*vpp));
1175 			kmem_free(cpp, ccio->ccio_ndisks * sizeof(*cpp));
1176 			disk_detach(&cs->sc_dkdev);
1177 			bufq_free(cs->sc_bufq);
1178 			goto out;
1179 		}
1180 
1181 		/* We can free the temporary variables now. */
1182 		kmem_free(vpp, ccio->ccio_ndisks * sizeof(*vpp));
1183 		kmem_free(cpp, ccio->ccio_ndisks * sizeof(*cpp));
1184 
1185 		/*
1186 		 * The ccd has been successfully initialized, so
1187 		 * we can place it into the array.  Don't try to
1188 		 * read the disklabel until the disk has been attached,
1189 		 * because space for the disklabel is allocated
1190 		 * in disk_attach();
1191 		 */
1192 		ccio->ccio_unit = unit;
1193 		ccio->ccio_size = cs->sc_size;
1194 
1195 		/* Try and read the disklabel. */
1196 		ccdgetdisklabel(dev);
1197 		break;
1198 
1199 	case CCDIOCCLR:
1200 		/*
1201 		 * Don't unconfigure if any other partitions are open
1202 		 * or if both the character and block flavors of this
1203 		 * partition are open.
1204 		 */
1205 		part = DISKPART(dev);
1206 		pmask = (1 << part);
1207 		if ((cs->sc_dkdev.dk_openmask & ~pmask) ||
1208 		    ((cs->sc_dkdev.dk_bopenmask & pmask) &&
1209 		    (cs->sc_dkdev.dk_copenmask & pmask))) {
1210 			error = EBUSY;
1211 			goto out;
1212 		}
1213 
1214 		/* Stop new I/O, wait for in-flight I/O to complete. */
1215 		mutex_enter(cs->sc_iolock);
1216 		cs->sc_flags &= ~(CCDF_INITED|CCDF_VLABEL);
1217 		cs->sc_zap = true;
1218 		while (disk_isbusy(&cs->sc_dkdev) ||
1219 		    bufq_peek(cs->sc_bufq) != NULL ||
1220 		    cs->sc_thread != NULL) {
1221 			cv_broadcast(&cs->sc_push);
1222 			(void)cv_timedwait(&cs->sc_stop, cs->sc_iolock, hz);
1223 		}
1224 		mutex_exit(cs->sc_iolock);
1225 
1226 		/*
1227 		 * Free ccd_softc information and clear entry.
1228 		 */
1229 
1230 		/* Close the components and free their pathnames. */
1231 		for (i = 0; i < cs->sc_nccdisks; ++i) {
1232 			/*
1233 			 * XXX: this close could potentially fail and
1234 			 * cause Bad Things.  Maybe we need to force
1235 			 * the close to happen?
1236 			 */
1237 #ifdef DEBUG
1238 			if (ccddebug & CCDB_VNODE)
1239 				vprint("CCDIOCCLR: vnode info",
1240 				    cs->sc_cinfo[i].ci_vp);
1241 #endif
1242 			(void)vn_close(cs->sc_cinfo[i].ci_vp, FREAD|FWRITE,
1243 			    uc);
1244 			kmem_free(cs->sc_cinfo[i].ci_path,
1245 			    cs->sc_cinfo[i].ci_pathlen);
1246 		}
1247 
1248 		/* Free interleave index. */
1249 		for (i = 0; cs->sc_itable[i].ii_ndisk; ++i) {
1250 			kmem_free(cs->sc_itable[i].ii_index,
1251 			    cs->sc_itable[i].ii_indexsz);
1252 		}
1253 
1254 		/* Free component info and interleave table. */
1255 		kmem_free(cs->sc_cinfo, cs->sc_nccdisks *
1256 		    sizeof(struct ccdcinfo));
1257 		kmem_free(cs->sc_itable, (cs->sc_nccdisks + 1) *
1258 		    sizeof(struct ccdiinfo));
1259 
1260 		/* Detatch the disk. */
1261 		disk_detach(&cs->sc_dkdev);
1262 		bufq_free(cs->sc_bufq);
1263 		ccdput(cs);
1264 		/* Don't break, otherwise cs is read again. */
1265 		return 0;
1266 
1267 	case DIOCGDINFO:
1268 		*(struct disklabel *)data = *(cs->sc_dkdev.dk_label);
1269 		break;
1270 
1271 #ifdef __HAVE_OLD_DISKLABEL
1272 	case ODIOCGDINFO:
1273 		newlabel = *(cs->sc_dkdev.dk_label);
1274 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1275 			return ENOTTY;
1276 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1277 		break;
1278 #endif
1279 
1280 	case DIOCGPART:
1281 		((struct partinfo *)data)->disklab = cs->sc_dkdev.dk_label;
1282 		((struct partinfo *)data)->part =
1283 		    &cs->sc_dkdev.dk_label->d_partitions[DISKPART(dev)];
1284 		break;
1285 
1286 	case DIOCCACHESYNC:
1287 		/*
1288 		 * XXX Do we really need to care about having a writable
1289 		 * file descriptor here?
1290 		 */
1291 		if ((flag & FWRITE) == 0)
1292 			return (EBADF);
1293 
1294 		/*
1295 		 * We pass this call down to all components and report
1296 		 * the first error we encounter.
1297 		 */
1298 		for (error = 0, i = 0; i < cs->sc_nccdisks; i++) {
1299 			j = VOP_IOCTL(cs->sc_cinfo[i].ci_vp, cmd, data,
1300 				      flag, uc);
1301 			if (j != 0 && error == 0)
1302 				error = j;
1303 		}
1304 		break;
1305 
1306 	case DIOCWDINFO:
1307 	case DIOCSDINFO:
1308 #ifdef __HAVE_OLD_DISKLABEL
1309 	case ODIOCWDINFO:
1310 	case ODIOCSDINFO:
1311 #endif
1312 	{
1313 		struct disklabel *lp;
1314 #ifdef __HAVE_OLD_DISKLABEL
1315 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
1316 			memset(&newlabel, 0, sizeof newlabel);
1317 			memcpy(&newlabel, data, sizeof (struct olddisklabel));
1318 			lp = &newlabel;
1319 		} else
1320 #endif
1321 		lp = (struct disklabel *)data;
1322 
1323 		cs->sc_flags |= CCDF_LABELLING;
1324 
1325 		error = setdisklabel(cs->sc_dkdev.dk_label,
1326 		    lp, 0, cs->sc_dkdev.dk_cpulabel);
1327 		if (error == 0) {
1328 			if (cmd == DIOCWDINFO
1329 #ifdef __HAVE_OLD_DISKLABEL
1330 			    || cmd == ODIOCWDINFO
1331 #endif
1332 			   )
1333 				error = writedisklabel(CCDLABELDEV(dev),
1334 				    ccdstrategy, cs->sc_dkdev.dk_label,
1335 				    cs->sc_dkdev.dk_cpulabel);
1336 		}
1337 
1338 		cs->sc_flags &= ~CCDF_LABELLING;
1339 		break;
1340 	}
1341 
1342 	case DIOCKLABEL:
1343 		if (*(int *)data != 0)
1344 			cs->sc_flags |= CCDF_KLABEL;
1345 		else
1346 			cs->sc_flags &= ~CCDF_KLABEL;
1347 		break;
1348 
1349 	case DIOCWLABEL:
1350 		if (*(int *)data != 0)
1351 			cs->sc_flags |= CCDF_WLABEL;
1352 		else
1353 			cs->sc_flags &= ~CCDF_WLABEL;
1354 		break;
1355 
1356 	case DIOCGDEFLABEL:
1357 		ccdgetdefaultlabel(cs, (struct disklabel *)data);
1358 		break;
1359 
1360 #ifdef __HAVE_OLD_DISKLABEL
1361 	case ODIOCGDEFLABEL:
1362 		ccdgetdefaultlabel(cs, &newlabel);
1363 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1364 			return ENOTTY;
1365 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1366 		break;
1367 #endif
1368 
1369 	default:
1370 		error = ENOTTY;
1371 	}
1372 
1373  out:
1374 	mutex_exit(&cs->sc_dvlock);
1375 	return (error);
1376 }
1377 
1378 static int
1379 ccdsize(dev_t dev)
1380 {
1381 	struct ccd_softc *cs;
1382 	struct disklabel *lp;
1383 	int part, unit, omask, size;
1384 
1385 	unit = ccdunit(dev);
1386 	if ((cs = ccdget(unit)) == NULL)
1387 		return -1;
1388 
1389 	if ((cs->sc_flags & CCDF_INITED) == 0)
1390 		return (-1);
1391 
1392 	part = DISKPART(dev);
1393 	omask = cs->sc_dkdev.dk_openmask & (1 << part);
1394 	lp = cs->sc_dkdev.dk_label;
1395 
1396 	if (omask == 0 && ccdopen(dev, 0, S_IFBLK, curlwp))
1397 		return (-1);
1398 
1399 	if (lp->d_partitions[part].p_fstype != FS_SWAP)
1400 		size = -1;
1401 	else
1402 		size = lp->d_partitions[part].p_size *
1403 		    (lp->d_secsize / DEV_BSIZE);
1404 
1405 	if (omask == 0 && ccdclose(dev, 0, S_IFBLK, curlwp))
1406 		return (-1);
1407 
1408 	return (size);
1409 }
1410 
1411 static void
1412 ccdgetdefaultlabel(struct ccd_softc *cs, struct disklabel *lp)
1413 {
1414 	struct ccdgeom *ccg = &cs->sc_geom;
1415 
1416 	memset(lp, 0, sizeof(*lp));
1417 
1418 	lp->d_secperunit = cs->sc_size;
1419 	lp->d_secsize = ccg->ccg_secsize;
1420 	lp->d_nsectors = ccg->ccg_nsectors;
1421 	lp->d_ntracks = ccg->ccg_ntracks;
1422 	lp->d_ncylinders = ccg->ccg_ncylinders;
1423 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1424 
1425 	strncpy(lp->d_typename, "ccd", sizeof(lp->d_typename));
1426 	lp->d_type = DTYPE_CCD;
1427 	strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
1428 	lp->d_rpm = 3600;
1429 	lp->d_interleave = 1;
1430 	lp->d_flags = 0;
1431 
1432 	lp->d_partitions[RAW_PART].p_offset = 0;
1433 	lp->d_partitions[RAW_PART].p_size = cs->sc_size;
1434 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
1435 	lp->d_npartitions = RAW_PART + 1;
1436 
1437 	lp->d_magic = DISKMAGIC;
1438 	lp->d_magic2 = DISKMAGIC;
1439 	lp->d_checksum = dkcksum(cs->sc_dkdev.dk_label);
1440 }
1441 
1442 /*
1443  * Read the disklabel from the ccd.  If one is not present, fake one
1444  * up.
1445  */
1446 static void
1447 ccdgetdisklabel(dev_t dev)
1448 {
1449 	int unit = ccdunit(dev);
1450 	struct ccd_softc *cs;
1451 	const char *errstring;
1452 	struct disklabel *lp;
1453 	struct cpu_disklabel *clp;
1454 
1455 	if ((cs = ccdget(unit)) == NULL)
1456 		return;
1457 	lp = cs->sc_dkdev.dk_label;
1458 	clp = cs->sc_dkdev.dk_cpulabel;
1459 	KASSERT(mutex_owned(&cs->sc_dvlock));
1460 
1461 	memset(clp, 0, sizeof(*clp));
1462 
1463 	ccdgetdefaultlabel(cs, lp);
1464 
1465 	/*
1466 	 * Call the generic disklabel extraction routine.
1467 	 */
1468 	cs->sc_flags |= CCDF_RLABEL;
1469 	if ((cs->sc_flags & CCDF_NOLABEL) != 0)
1470 		errstring = "CCDF_NOLABEL set; ignoring on-disk label";
1471 	else
1472 		errstring = readdisklabel(CCDLABELDEV(dev), ccdstrategy,
1473 		    cs->sc_dkdev.dk_label, cs->sc_dkdev.dk_cpulabel);
1474 	if (errstring)
1475 		ccdmakedisklabel(cs);
1476 	else {
1477 		int i;
1478 		struct partition *pp;
1479 
1480 		/*
1481 		 * Sanity check whether the found disklabel is valid.
1482 		 *
1483 		 * This is necessary since total size of ccd may vary
1484 		 * when an interleave is changed even though exactly
1485 		 * same componets are used, and old disklabel may used
1486 		 * if that is found.
1487 		 */
1488 		if (lp->d_secperunit != cs->sc_size)
1489 			printf("WARNING: %s: "
1490 			    "total sector size in disklabel (%d) != "
1491 			    "the size of ccd (%lu)\n", cs->sc_xname,
1492 			    lp->d_secperunit, (u_long)cs->sc_size);
1493 		for (i = 0; i < lp->d_npartitions; i++) {
1494 			pp = &lp->d_partitions[i];
1495 			if (pp->p_offset + pp->p_size > cs->sc_size)
1496 				printf("WARNING: %s: end of partition `%c' "
1497 				    "exceeds the size of ccd (%lu)\n",
1498 				    cs->sc_xname, 'a' + i, (u_long)cs->sc_size);
1499 		}
1500 	}
1501 
1502 #ifdef DEBUG
1503 	/* It's actually extremely common to have unlabeled ccds. */
1504 	if (ccddebug & CCDB_LABEL)
1505 		if (errstring != NULL)
1506 			printf("%s: %s\n", cs->sc_xname, errstring);
1507 #endif
1508 
1509 	/* In-core label now valid. */
1510 	cs->sc_flags = (cs->sc_flags | CCDF_VLABEL) & ~CCDF_RLABEL;
1511 }
1512 
1513 /*
1514  * Take care of things one might want to take care of in the event
1515  * that a disklabel isn't present.
1516  */
1517 static void
1518 ccdmakedisklabel(struct ccd_softc *cs)
1519 {
1520 	struct disklabel *lp = cs->sc_dkdev.dk_label;
1521 
1522 	/*
1523 	 * For historical reasons, if there's no disklabel present
1524 	 * the raw partition must be marked FS_BSDFFS.
1525 	 */
1526 	lp->d_partitions[RAW_PART].p_fstype = FS_BSDFFS;
1527 
1528 	strncpy(lp->d_packname, "default label", sizeof(lp->d_packname));
1529 
1530 	lp->d_checksum = dkcksum(lp);
1531 }
1532 
1533 #ifdef DEBUG
1534 static void
1535 printiinfo(struct ccdiinfo *ii)
1536 {
1537 	int ix, i;
1538 
1539 	for (ix = 0; ii->ii_ndisk; ix++, ii++) {
1540 		printf(" itab[%d]: #dk %d sblk %" PRId64 " soff %" PRId64,
1541 		    ix, ii->ii_ndisk, ii->ii_startblk, ii->ii_startoff);
1542 		for (i = 0; i < ii->ii_ndisk; i++)
1543 			printf(" %d", ii->ii_index[i]);
1544 		printf("\n");
1545 	}
1546 }
1547 #endif
1548 
1549 MODULE(MODULE_CLASS_DRIVER, ccd, "dk_subr");
1550 
1551 static int
1552 ccd_modcmd(modcmd_t cmd, void *arg)
1553 {
1554 	int error = 0;
1555 #ifdef _MODULE
1556 	int bmajor = -1, cmajor = -1;
1557 #endif
1558 
1559 
1560 	switch (cmd) {
1561 	case MODULE_CMD_INIT:
1562 #ifdef _MODULE
1563 		ccdattach(4);
1564 
1565 		return devsw_attach("ccd", &ccd_bdevsw, &bmajor,
1566 		    &ccd_cdevsw, &cmajor);
1567 #endif
1568 		break;
1569 
1570 	case MODULE_CMD_FINI:
1571 #ifdef _MODULE
1572 		return devsw_detach(&ccd_bdevsw, &ccd_cdevsw);
1573 #endif
1574 		break;
1575 
1576 	case MODULE_CMD_STAT:
1577 		return ENOTTY;
1578 
1579 	default:
1580 		return ENOTTY;
1581 	}
1582 
1583 	return error;
1584 }
1585 
1586 static int
1587 ccd_units_sysctl(SYSCTLFN_ARGS)
1588 {
1589 	struct sysctlnode node;
1590 	struct ccd_softc *sc;
1591 	int error, i, nccd, *units;
1592 	size_t size;
1593 
1594 	nccd = 0;
1595 	mutex_enter(&ccd_lock);
1596 	LIST_FOREACH(sc, &ccds, sc_link)
1597 		nccd++;
1598 	mutex_exit(&ccd_lock);
1599 
1600 	if (nccd != 0) {
1601 		size = nccd * sizeof(*units);
1602 		units = kmem_zalloc(size, KM_SLEEP);
1603 		if (units == NULL)
1604 			return ENOMEM;
1605 
1606 		i = 0;
1607 		mutex_enter(&ccd_lock);
1608 		LIST_FOREACH(sc, &ccds, sc_link) {
1609 			if (i >= nccd)
1610 				break;
1611 			units[i] = sc->sc_unit;
1612 		}
1613 		mutex_exit(&ccd_lock);
1614 	} else {
1615 		units = NULL;
1616 		size = 0;
1617 	}
1618 
1619 	node = *rnode;
1620 	node.sysctl_data = units;
1621 	node.sysctl_size = size;
1622 
1623 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
1624 	if (units)
1625 		kmem_free(units, size);
1626 	return error;
1627 }
1628 
1629 static int
1630 ccd_info_sysctl(SYSCTLFN_ARGS)
1631 {
1632 	struct sysctlnode node;
1633 	struct ccddiskinfo ccd;
1634 	struct ccd_softc *sc;
1635 	int unit;
1636 
1637 	if (newp == NULL || newlen != sizeof(int))
1638 		return EINVAL;
1639 
1640 	unit = *(const int *)newp;
1641 	newp = NULL;
1642 	newlen = 0;
1643 	ccd.ccd_ndisks = ~0;
1644 	mutex_enter(&ccd_lock);
1645 	LIST_FOREACH(sc, &ccds, sc_link) {
1646 		if (sc->sc_unit == unit) {
1647 			ccd.ccd_ileave = sc->sc_ileave;
1648 			ccd.ccd_size = sc->sc_size;
1649 			ccd.ccd_ndisks = sc->sc_nccdisks;
1650 			ccd.ccd_flags = sc->sc_flags;
1651 			break;
1652 		}
1653 	}
1654 	mutex_exit(&ccd_lock);
1655 
1656 	if (ccd.ccd_ndisks == ~0)
1657 		return ENOENT;
1658 
1659 	node = *rnode;
1660 	node.sysctl_data = &ccd;
1661 	node.sysctl_size = sizeof(ccd);
1662 
1663 	return sysctl_lookup(SYSCTLFN_CALL(&node));
1664 }
1665 
1666 static int
1667 ccd_components_sysctl(SYSCTLFN_ARGS)
1668 {
1669 	struct sysctlnode node;
1670 	int error, unit;
1671 	size_t size;
1672 	char *names, *p, *ep;
1673 	struct ccd_softc *sc;
1674 
1675 	if (newp == NULL || newlen != sizeof(int))
1676 		return EINVAL;
1677 
1678 	size = 0;
1679 	unit = *(const int *)newp;
1680 	newp = NULL;
1681 	newlen = 0;
1682 	mutex_enter(&ccd_lock);
1683 	LIST_FOREACH(sc, &ccds, sc_link)
1684 		if (sc->sc_unit == unit) {
1685 			for (size_t i = 0; i < sc->sc_nccdisks; i++)
1686 				size += strlen(sc->sc_cinfo[i].ci_path) + 1;
1687 			break;
1688 		}
1689 	mutex_exit(&ccd_lock);
1690 
1691 	if (size == 0)
1692 		return ENOENT;
1693 	names = kmem_zalloc(size, KM_SLEEP);
1694 	if (names == NULL)
1695 		return ENOMEM;
1696 
1697 	p = names;
1698 	ep = names + size;
1699 	mutex_enter(&ccd_lock);
1700 	LIST_FOREACH(sc, &ccds, sc_link)
1701 		if (sc->sc_unit == unit) {
1702 			for (size_t i = 0; i < sc->sc_nccdisks; i++) {
1703 				char *d = sc->sc_cinfo[i].ci_path;
1704 				while (p < ep && (*p++ = *d++) != '\0')
1705 					continue;
1706 			}
1707 			break;
1708 		}
1709 	mutex_exit(&ccd_lock);
1710 
1711 	node = *rnode;
1712 	node.sysctl_data = names;
1713 	node.sysctl_size = ep - names;
1714 
1715 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
1716 	kmem_free(names, size);
1717 	return error;
1718 }
1719 
1720 SYSCTL_SETUP(sysctl_kern_ccd_setup, "sysctl kern.ccd subtree setup")
1721 {
1722 	const struct sysctlnode *node = NULL;
1723 
1724 	sysctl_createv(clog, 0, NULL, &node,
1725 	    CTLFLAG_PERMANENT,
1726 	    CTLTYPE_NODE, "ccd",
1727 	    SYSCTL_DESCR("ConCatenated Disk state"),
1728 	    NULL, 0, NULL, 0,
1729 	    CTL_KERN, CTL_CREATE, CTL_EOL);
1730 
1731 	if (node == NULL)
1732 		return;
1733 
1734 	sysctl_createv(clog, 0, &node, NULL,
1735 	    CTLFLAG_PERMANENT | CTLFLAG_READONLY,
1736 	    CTLTYPE_STRUCT, "units",
1737 	    SYSCTL_DESCR("List of ccd unit numbers"),
1738 	    ccd_units_sysctl, 0, NULL, 0,
1739 	    CTL_CREATE, CTL_EOL);
1740 	sysctl_createv(clog, 0, &node, NULL,
1741 	    CTLFLAG_PERMANENT | CTLFLAG_READWRITE,
1742 	    CTLTYPE_STRUCT, "info",
1743 	    SYSCTL_DESCR("Information about a CCD unit"),
1744 	    ccd_info_sysctl, 0, NULL, 0,
1745 	    CTL_CREATE, CTL_EOL);
1746 	sysctl_createv(clog, 0, &node, NULL,
1747 	    CTLFLAG_PERMANENT | CTLFLAG_READWRITE,
1748 	    CTLTYPE_STRUCT, "components",
1749 	    SYSCTL_DESCR("Information about CCD components"),
1750 	    ccd_components_sysctl, 0, NULL, 0,
1751 	    CTL_CREATE, CTL_EOL);
1752 }
1753